From cc0180f0389e3db92dc1c078c933dd82d6639667 Mon Sep 17 00:00:00 2001
From: Stan Jenkins <mc.stanislaw@gmail.com>
Date: Tue, 19 May 2026 14:10:23 +0000
Subject: [PATCH] Documentation generated from Jenkins for docs-2-39

---
 docs/2_38/cmdstan-guide/bib.html              |    6 +-
 .../cmdstan-guide/command_line_options.html   |    6 +-
 .../compiling_stan_programs.html              |    6 +-
 docs/2_38/cmdstan-guide/diagnose_config.html  |    6 +-
 docs/2_38/cmdstan-guide/diagnose_utility.html |    6 +-
 .../cmdstan-guide/example_model_data.html     |    6 +-
 docs/2_38/cmdstan-guide/external_code.html    |    6 +-
 .../generate_quantities_config.html           |    6 +-
 docs/2_38/cmdstan-guide/index.html            |    2 +-
 docs/2_38/cmdstan-guide/installation.html     |    6 +-
 docs/2_38/cmdstan-guide/json_apdx.html        |    6 +-
 .../cmdstan-guide/laplace_sample_config.html  |    6 +-
 docs/2_38/cmdstan-guide/log_prob_config.html  |    6 +-
 docs/2_38/cmdstan-guide/mcmc_config.html      |    6 +-
 docs/2_38/cmdstan-guide/optimize_config.html  |    6 +-
 docs/2_38/cmdstan-guide/parallelization.html  |    6 +-
 .../2_38/cmdstan-guide/pathfinder_config.html |    6 +-
 docs/2_38/cmdstan-guide/print.html            |    6 +-
 docs/2_38/cmdstan-guide/rdump_apdx.html       |    6 +-
 docs/2_38/cmdstan-guide/stan_csv_apdx.html    |    6 +-
 docs/2_38/cmdstan-guide/stanc.html            |    6 +-
 docs/2_38/cmdstan-guide/stansummary.html      |    6 +-
 .../cmdstan-guide/variational_config.html     |    6 +-
 .../functions-reference/array_operations.html |    6 +-
 .../binary_distributions.html                 |    6 +-
 .../bounded_continuous_distributions.html     |    6 +-
 .../bounded_discrete_distributions.html       |    6 +-
 .../circular_distributions.html               |    6 +-
 .../complex-valued_basic_functions.html       |    6 +-
 .../complex_matrix_operations.html            |    6 +-
 .../compound_arithmetic_and_assignment.html   |    6 +-
 .../continuous_distributions_on_0_1.html      |    6 +-
 ...conventions_for_probability_functions.html |    6 +-
 .../correlation_matrix_distributions.html     |    6 +-
 .../covariance_matrix_distributions.html      |    6 +-
 .../deprecated_functions.html                 |    6 +-
 .../distributions_over_unbounded_vectors.html |    6 +-
 .../functions-reference/functions_index.html  |    6 +-
 .../hidden_markov_models.html                 |    6 +-
 .../higher-order_functions.html               |    6 +-
 docs/2_38/functions-reference/index.html      |    2 +-
 .../integer-valued_basic_functions.html       |    6 +-
 .../mathematical_functions.html               |    6 +-
 .../matrix_operations.html                    |    6 +-
 .../functions-reference/mixed_operations.html |    6 +-
 .../multivariate_discrete_distributions.html  |    6 +-
 .../positive_continuous_distributions.html    |    6 +-
 .../positive_lower-bounded_distributions.html |    6 +-
 .../real-valued_basic_functions.html          |    6 +-
 docs/2_38/functions-reference/references.html |    6 +-
 .../removed_functions.html                    |    6 +-
 .../simplex_distributions.html                |    6 +-
 .../sparse_matrix_operations.html             |    6 +-
 .../transform_functions.html                  |    6 +-
 .../unbounded_continuous_distributions.html   |    6 +-
 .../unbounded_discrete_distributions.html     |    6 +-
 .../functions-reference/void_functions.html   |    6 +-
 docs/2_38/reference-manual/analysis.html      |    6 +-
 docs/2_38/reference-manual/blocks.html        |    6 +-
 docs/2_38/reference-manual/comments.html      |    6 +-
 docs/2_38/reference-manual/deprecations.html  |    6 +-
 docs/2_38/reference-manual/diagnostics.html   |    6 +-
 docs/2_38/reference-manual/encoding.html      |    6 +-
 docs/2_38/reference-manual/execution.html     |    6 +-
 docs/2_38/reference-manual/expressions.html   |    6 +-
 docs/2_38/reference-manual/includes.html      |    6 +-
 docs/2_38/reference-manual/index.html         |    2 +-
 docs/2_38/reference-manual/laplace.html       |    6 +-
 docs/2_38/reference-manual/licenses.html      |    6 +-
 docs/2_38/reference-manual/mcmc.html          |    6 +-
 docs/2_38/reference-manual/optimization.html  |    6 +-
 docs/2_38/reference-manual/pathfinder.html    |    6 +-
 docs/2_38/reference-manual/references.html    |    6 +-
 docs/2_38/reference-manual/removals.html      |    6 +-
 .../reference-manual/reproducibility.html     |    6 +-
 docs/2_38/reference-manual/statements.html    |    6 +-
 docs/2_38/reference-manual/syntax.html        |    6 +-
 docs/2_38/reference-manual/transforms.html    |    6 +-
 docs/2_38/reference-manual/types.html         |    6 +-
 .../2_38/reference-manual/user-functions.html |    6 +-
 docs/2_38/reference-manual/variational.html   |    6 +-
 docs/2_38/reference-manual/whitespace.html    |    6 +-
 .../stan-users-guide/algebraic-equations.html |    6 +-
 docs/2_38/stan-users-guide/bootstrap.html     |    6 +-
 docs/2_38/stan-users-guide/clustering.html    |    6 +-
 .../stan-users-guide/complex-numbers.html     |    6 +-
 docs/2_38/stan-users-guide/copulas.html       |    6 +-
 .../stan-users-guide/cross-validation.html    |    6 +-
 .../stan-users-guide/custom-probability.html  |    6 +-
 docs/2_38/stan-users-guide/dae.html           |    6 +-
 .../stan-users-guide/decision-analysis.html   |    6 +-
 .../stan-users-guide/efficiency-tuning.html   |    6 +-
 .../stan-users-guide/finite-mixtures.html     |    6 +-
 .../2_38/stan-users-guide/floating-point.html |    6 +-
 .../2_38/stan-users-guide/for-bugs-users.html |    6 +-
 .../stan-users-guide/gaussian-processes.html  |    6 +-
 .../hyperspherical-models.html                |    6 +-
 docs/2_38/stan-users-guide/index.html         |    2 +-
 .../stan-users-guide/latent-discrete.html     |    6 +-
 .../stan-users-guide/matrices-arrays.html     |    6 +-
 .../stan-users-guide/measurement-error.html   |    6 +-
 docs/2_38/stan-users-guide/missing-data.html  |    6 +-
 .../2_38/stan-users-guide/multi-indexing.html |    6 +-
 docs/2_38/stan-users-guide/odes.html          |    6 +-
 .../one-dimensional-integrals.html            |    6 +-
 .../stan-users-guide/parallelization.html     |    6 +-
 .../posterior-prediction.html                 |    6 +-
 .../posterior-predictive-checks.html          |    6 +-
 .../stan-users-guide/poststratification.html  |    6 +-
 .../problematic-posteriors.html               |    6 +-
 .../proportionality-constants.html            |    6 +-
 docs/2_38/stan-users-guide/references.html    |    6 +-
 docs/2_38/stan-users-guide/regression.html    |    6 +-
 .../stan-users-guide/reparameterization.html  |    6 +-
 .../simulation-based-calibration.html         |    6 +-
 docs/2_38/stan-users-guide/sparse-ragged.html |    6 +-
 docs/2_38/stan-users-guide/style-guide.html   |    6 +-
 docs/2_38/stan-users-guide/survival.html      |    6 +-
 docs/2_38/stan-users-guide/time-series.html   |    6 +-
 .../truncation-censoring.html                 |    6 +-
 .../2_38/stan-users-guide/user-functions.html |    6 +-
 docs/2_38/stan-users-guide/using-stanc.html   |    6 +-
 docs/2_39/404.html                            |  871 ++
 docs/2_39/cmdstan-guide-2_39.pdf              |  Bin 0 -> 551581 bytes
 docs/2_39/cmdstan-guide/bib.html              |  821 ++
 .../cmdstan-guide/command_line_options.html   | 1151 ++
 .../compiling_stan_programs.html              | 1191 ++
 docs/2_39/cmdstan-guide/diagnose_config.html  | 1102 ++
 docs/2_39/cmdstan-guide/diagnose_utility.html | 1226 +++
 .../cmdstan-guide/example_model_data.html     | 1091 ++
 docs/2_39/cmdstan-guide/external_code.html    | 1197 ++
 .../generate_quantities_config.html           | 1159 ++
 docs/2_39/cmdstan-guide/img/logo_tm.png       |  Bin 0 -> 21065 bytes
 docs/2_39/cmdstan-guide/img/warmup-epochs.png |  Bin 0 -> 16881 bytes
 docs/2_39/cmdstan-guide/index.html            | 1077 ++
 docs/2_39/cmdstan-guide/installation.html     | 1357 +++
 docs/2_39/cmdstan-guide/json_apdx.html        | 1273 +++
 .../cmdstan-guide/laplace_sample_config.html  | 1134 ++
 docs/2_39/cmdstan-guide/log_prob_config.html  | 1121 ++
 docs/2_39/cmdstan-guide/mcmc_config.html      | 1597 +++
 docs/2_39/cmdstan-guide/optimize_config.html  | 1234 +++
 docs/2_39/cmdstan-guide/parallelization.html  | 1162 ++
 .../2_39/cmdstan-guide/pathfinder_config.html | 1269 +++
 docs/2_39/cmdstan-guide/print.html            | 1034 ++
 docs/2_39/cmdstan-guide/rdump_apdx.html       | 1290 +++
 docs/2_39/cmdstan-guide/stan_csv_apdx.html    | 1279 +++
 docs/2_39/cmdstan-guide/stanc.html            | 1064 ++
 docs/2_39/cmdstan-guide/stansummary.html      | 1212 ++
 .../cmdstan-guide/variational_config.html     | 1240 +++
 docs/2_39/functions-reference-2_39.pdf        |  Bin 0 -> 991980 bytes
 .../functions-reference/array_operations.html | 1507 +++
 .../binary_distributions.html                 | 1331 +++
 .../bounded_continuous_distributions.html     | 1197 ++
 .../bounded_discrete_distributions.html       | 1768 +++
 .../circular_distributions.html               | 1248 +++
 .../complex-valued_basic_functions.html       | 1544 +++
 .../complex_matrix_operations.html            | 2106 ++++
 .../compound_arithmetic_and_assignment.html   | 1177 ++
 .../continuous_distributions_on_0_1.html      | 1248 +++
 ...conventions_for_probability_functions.html | 1379 +++
 .../correlation_matrix_distributions.html     | 1308 +++
 .../covariance_matrix_distributions.html      | 1312 +++
 .../deprecated_functions.html                 | 1357 +++
 .../distributions_over_unbounded_vectors.html | 1624 +++
 .../functions-reference/embedded_laplace.html | 1517 +++
 .../functions-reference/functions_index.html  | 7656 +++++++++++++
 .../hidden_markov_models.html                 | 1178 ++
 .../higher-order_functions.html               | 1670 +++
 docs/2_39/functions-reference/img/logo_tm.png |  Bin 0 -> 21065 bytes
 docs/2_39/functions-reference/index.html      | 1142 ++
 .../integer-valued_basic_functions.html       | 1274 +++
 .../mathematical_functions.html               | 1178 ++
 .../matrix_operations.html                    | 2681 +++++
 .../functions-reference/mixed_operations.html | 1378 +++
 .../multivariate_discrete_distributions.html  | 1273 +++
 .../positive_continuous_distributions.html    | 1699 +++
 .../positive_lower-bounded_distributions.html | 1452 +++
 .../real-valued_basic_functions.html          | 2260 ++++
 docs/2_39/functions-reference/references.html |  821 ++
 .../removed_functions.html                    | 1209 ++
 .../simplex_distributions.html                | 1256 +++
 .../sparse_matrix_operations.html             | 1217 ++
 .../transform_functions.html                  | 1469 +++
 .../unbounded_continuous_distributions.html   | 1805 +++
 .../unbounded_discrete_distributions.html     | 1662 +++
 .../functions-reference/void_functions.html   | 1148 ++
 docs/2_39/img/logo_tm.png                     |  Bin 0 -> 21065 bytes
 docs/2_39/img/ulam-fermiac.png                |  Bin 0 -> 370448 bytes
 docs/2_39/index.html                          | 1129 ++
 docs/2_39/quarto-config/quarto_styles.css     |  299 +
 docs/2_39/reference-manual-2_39.pdf           |  Bin 0 -> 900313 bytes
 docs/2_39/reference-manual/analysis.html      | 1359 +++
 docs/2_39/reference-manual/blocks.html        | 1556 +++
 docs/2_39/reference-manual/comments.html      | 1091 ++
 docs/2_39/reference-manual/deprecations.html  | 1110 ++
 docs/2_39/reference-manual/diagnostics.html   | 1132 ++
 docs/2_39/reference-manual/encoding.html      | 1065 ++
 docs/2_39/reference-manual/execution.html     | 1263 +++
 docs/2_39/reference-manual/expressions.html   | 2293 ++++
 docs/2_39/reference-manual/img/logo_tm.png    |  Bin 0 -> 21065 bytes
 .../reference-manual/img/warmup-epochs.png    |  Bin 0 -> 16881 bytes
 docs/2_39/reference-manual/includes.html      | 1152 ++
 docs/2_39/reference-manual/index.html         | 1062 ++
 docs/2_39/reference-manual/laplace.html       | 1078 ++
 .../reference-manual/laplace_embedded.html    | 1226 +++
 docs/2_39/reference-manual/licenses.html      | 1101 ++
 docs/2_39/reference-manual/mcmc.html          | 1551 +++
 docs/2_39/reference-manual/optimization.html  | 1213 ++
 docs/2_39/reference-manual/pathfinder.html    | 1117 ++
 docs/2_39/reference-manual/references.html    |  821 ++
 docs/2_39/reference-manual/removals.html      | 1203 ++
 .../reference-manual/reproducibility.html     | 1094 ++
 docs/2_39/reference-manual/statements.html    | 1951 ++++
 docs/2_39/reference-manual/syntax.html        | 1480 +++
 docs/2_39/reference-manual/transforms.html    | 2232 ++++
 docs/2_39/reference-manual/types.html         | 2326 ++++
 .../2_39/reference-manual/user-functions.html | 1351 +++
 docs/2_39/reference-manual/variational.html   | 1127 ++
 docs/2_39/reference-manual/whitespace.html    | 1058 ++
 docs/2_39/search.json                         | 9494 ++++++++++++++++
 ...p-c777a6582eb78ab223b9896ab55f446e.min.css |   12 +
 ...k-2df8a8547a8386440a781d5ef2c5f5b1.min.css |   12 +
 .../site_libs/bootstrap/bootstrap-icons.css   | 2078 ++++
 .../site_libs/bootstrap/bootstrap-icons.woff  |  Bin 0 -> 176200 bytes
 .../2_39/site_libs/bootstrap/bootstrap.min.js |    7 +
 .../2_39/site_libs/clipboard/clipboard.min.js |    7 +
 .../glightbox/glightbox.min.css               |    1 +
 .../quarto-contrib/glightbox/glightbox.min.js |    1 +
 .../quarto-contrib/glightbox/lightbox.css     |   26 +
 docs/2_39/site_libs/quarto-html/anchor.min.js |    9 +
 docs/2_39/site_libs/quarto-html/popper.min.js |    6 +
 ...hting-05fe91a66cf75bbbb8c9664867fe5124.css |  191 +
 ...-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css |  185 +
 docs/2_39/site_libs/quarto-html/quarto.js     |  911 ++
 docs/2_39/site_libs/quarto-html/tippy.css     |    1 +
 .../site_libs/quarto-html/tippy.umd.min.js    |    2 +
 .../2_39/site_libs/quarto-nav/headroom.min.js |    7 +
 docs/2_39/site_libs/quarto-nav/quarto-nav.js  |  325 +
 .../quarto-search/autocomplete.umd.js         |    3 +
 docs/2_39/site_libs/quarto-search/fuse.min.js |    9 +
 .../site_libs/quarto-search/quarto-search.js  | 1290 +++
 docs/2_39/stan-users-guide-2_39.pdf           |  Bin 0 -> 3281908 bytes
 .../stan-users-guide/algebraic-equations.html | 1342 +++
 docs/2_39/stan-users-guide/bootstrap.html     | 1363 +++
 docs/2_39/stan-users-guide/clustering.html    | 1655 +++
 .../stan-users-guide/complex-numbers.html     | 1350 +++
 docs/2_39/stan-users-guide/copulas.html       | 1441 +++
 .../stan-users-guide/cross-validation.html    | 1520 +++
 .../stan-users-guide/custom-probability.html  | 1326 +++
 docs/2_39/stan-users-guide/dae.html           | 1358 +++
 .../stan-users-guide/decision-analysis.html   | 1372 +++
 .../stan-users-guide/efficiency-tuning.html   | 2128 ++++
 .../stan-users-guide/finite-mixtures.html     | 1659 +++
 .../2_39/stan-users-guide/floating-point.html | 1387 +++
 .../2_39/stan-users-guide/for-bugs-users.html | 1450 +++
 .../stan-users-guide/gaussian-processes.html  | 2113 ++++
 .../hyperspherical-models.html                | 1309 +++
 docs/2_39/stan-users-guide/img/Figure_DDM.pdf |  Bin 0 -> 100560 bytes
 docs/2_39/stan-users-guide/img/Figure_DDM.png |  Bin 0 -> 74686 bytes
 .../img/change-point-posterior.png            |  Bin 0 -> 37215 bytes
 docs/2_39/stan-users-guide/img/funnel-fit.png |  Bin 0 -> 77471 bytes
 docs/2_39/stan-users-guide/img/funnel.png     |  Bin 0 -> 99805 bytes
 docs/2_39/stan-users-guide/img/logo_tm.png    |  Bin 0 -> 21065 bytes
 .../img/non-identified-plus-prior.png         |  Bin 0 -> 96043 bytes
 .../stan-users-guide/img/non-identified.png   |  Bin 0 -> 64857 bytes
 .../img/one-param-identified.png              |  Bin 0 -> 61295 bytes
 .../2_39/stan-users-guide/img/ppc-nb-pois.jpg |  Bin 0 -> 195082 bytes
 .../stan-users-guide/img/ppc-pois-pois.jpg    |  Bin 0 -> 179965 bytes
 .../img/ppc-pvalue-nb-pois-mean.jpg           |  Bin 0 -> 125174 bytes
 .../img/ppc-pvalue-nb-pois-sd.jpg             |  Bin 0 -> 108780 bytes
 .../img/s-discrete-posterior.png              |  Bin 0 -> 27054 bytes
 .../img/sbc-ctr-8-schools-mu.png              |  Bin 0 -> 48757 bytes
 .../img/sbc-ctr-8-schools-tau.png             |  Bin 0 -> 47009 bytes
 .../img/sbc-ctr-8-schools-theta1.png          |  Bin 0 -> 50644 bytes
 .../img/sbc-normal-normal.png                 |  Bin 0 -> 286125 bytes
 .../img/sbc-student-t-normal.png              |  Bin 0 -> 295585 bytes
 .../img/sho-ode-trajectory.png                |  Bin 0 -> 322150 bytes
 docs/2_39/stan-users-guide/index.html         | 1175 ++
 .../stan-users-guide/latent-discrete.html     | 1933 ++++
 .../stan-users-guide/matrices-arrays.html     | 1434 +++
 .../stan-users-guide/measurement-error.html   | 1467 +++
 docs/2_39/stan-users-guide/missing-data.html  | 1415 +++
 .../2_39/stan-users-guide/multi-indexing.html | 1460 +++
 .../stan-users-guide/multiple-imputation.html | 1450 +++
 docs/2_39/stan-users-guide/odes.html          | 1543 +++
 .../one-dimensional-integrals.html            | 1414 +++
 .../stan-users-guide/parallelization.html     | 1636 +++
 .../posterior-prediction.html                 | 1476 +++
 .../posterior-predictive-checks.html          | 1594 +++
 .../stan-users-guide/poststratification.html  | 1511 +++
 .../problematic-posteriors.html               | 1619 +++
 .../proportionality-constants.html            | 1299 +++
 docs/2_39/stan-users-guide/references.html    |  821 ++
 docs/2_39/stan-users-guide/regression.html    | 2305 ++++
 .../stan-users-guide/reparameterization.html  | 1629 +++
 .../simulation-based-calibration.html         | 1619 +++
 docs/2_39/stan-users-guide/sparse-ragged.html | 1371 +++
 docs/2_39/stan-users-guide/style-guide.html   | 1467 +++
 docs/2_39/stan-users-guide/survival.html      | 1670 +++
 docs/2_39/stan-users-guide/time-series.html   | 1689 +++
 .../truncation-censoring.html                 | 1355 +++
 .../2_39/stan-users-guide/user-functions.html | 1642 +++
 docs/2_39/stan-users-guide/using-stanc.html   | 2048 ++++
 .../wiener_diffusion_model.html               | 1598 +++
 .../wiener_diffusion_model.pdf                |  Bin 0 -> 168306 bytes
 docs/404.html                                 |   29 +-
 docs/cmdstan-guide/bib.html                   |   29 +-
 docs/cmdstan-guide/command_line_options.html  |   35 +-
 .../compiling_stan_programs.html              |   31 +-
 docs/cmdstan-guide/diagnose_config.html       |   31 +-
 docs/cmdstan-guide/diagnose_utility.html      |   31 +-
 docs/cmdstan-guide/example_model_data.html    |   31 +-
 docs/cmdstan-guide/external_code.html         |   31 +-
 .../generate_quantities_config.html           |   31 +-
 docs/cmdstan-guide/index.html                 |   37 +-
 docs/cmdstan-guide/installation.html          |   31 +-
 docs/cmdstan-guide/json_apdx.html             |   31 +-
 docs/cmdstan-guide/laplace_sample_config.html |   31 +-
 docs/cmdstan-guide/log_prob_config.html       |   31 +-
 docs/cmdstan-guide/mcmc_config.html           |   31 +-
 docs/cmdstan-guide/optimize_config.html       |  104 +-
 docs/cmdstan-guide/parallelization.html       |   31 +-
 docs/cmdstan-guide/pathfinder_config.html     |   31 +-
 docs/cmdstan-guide/print.html                 |   31 +-
 docs/cmdstan-guide/rdump_apdx.html            |   31 +-
 docs/cmdstan-guide/stan_csv_apdx.html         |   31 +-
 docs/cmdstan-guide/stanc.html                 |   31 +-
 docs/cmdstan-guide/stansummary.html           |   31 +-
 docs/cmdstan-guide/variational_config.html    |   31 +-
 .../functions-reference/array_operations.html |   37 +-
 .../binary_distributions.html                 |   37 +-
 .../bounded_continuous_distributions.html     |   37 +-
 .../bounded_discrete_distributions.html       |   37 +-
 .../circular_distributions.html               |   37 +-
 .../complex-valued_basic_functions.html       |   37 +-
 .../complex_matrix_operations.html            |   37 +-
 .../compound_arithmetic_and_assignment.html   |   37 +-
 .../continuous_distributions_on_0_1.html      |   37 +-
 ...conventions_for_probability_functions.html |   37 +-
 .../correlation_matrix_distributions.html     |   37 +-
 .../covariance_matrix_distributions.html      |   37 +-
 .../deprecated_functions.html                 |   37 +-
 .../distributions_over_unbounded_vectors.html |   37 +-
 .../functions-reference/embedded_laplace.html | 1517 +++
 docs/functions-reference/functions_index.html |  266 +-
 .../hidden_markov_models.html                 |   43 +-
 .../higher-order_functions.html               |   37 +-
 docs/functions-reference/index.html           |   43 +-
 .../integer-valued_basic_functions.html       |   37 +-
 .../mathematical_functions.html               |   43 +-
 .../matrix_operations.html                    |   43 +-
 .../functions-reference/mixed_operations.html |   37 +-
 .../multivariate_discrete_distributions.html  |   37 +-
 .../positive_continuous_distributions.html    |   37 +-
 .../positive_lower-bounded_distributions.html |   37 +-
 .../real-valued_basic_functions.html          |   37 +-
 docs/functions-reference/references.html      |   29 +-
 .../removed_functions.html                    |   37 +-
 .../simplex_distributions.html                |   37 +-
 .../sparse_matrix_operations.html             |   37 +-
 .../transform_functions.html                  |   37 +-
 .../unbounded_continuous_distributions.html   |   37 +-
 .../unbounded_discrete_distributions.html     |   87 +-
 docs/functions-reference/void_functions.html  |   37 +-
 docs/index.html                               |   90 +-
 docs/reference-manual/analysis.html           |   37 +-
 docs/reference-manual/blocks.html             |   37 +-
 docs/reference-manual/comments.html           |   37 +-
 docs/reference-manual/deprecations.html       |   37 +-
 docs/reference-manual/diagnostics.html        |   43 +-
 docs/reference-manual/encoding.html           |   37 +-
 docs/reference-manual/execution.html          |   37 +-
 docs/reference-manual/expressions.html        |   46 +-
 docs/reference-manual/includes.html           |   37 +-
 docs/reference-manual/index.html              |   43 +-
 docs/reference-manual/laplace.html            |   43 +-
 docs/reference-manual/laplace_embedded.html   | 1226 +++
 docs/reference-manual/licenses.html           |   37 +-
 docs/reference-manual/mcmc.html               |   41 +-
 docs/reference-manual/optimization.html       |   37 +-
 docs/reference-manual/pathfinder.html         |   37 +-
 docs/reference-manual/references.html         |   29 +-
 docs/reference-manual/removals.html           |   37 +-
 docs/reference-manual/reproducibility.html    |   37 +-
 docs/reference-manual/statements.html         |   37 +-
 docs/reference-manual/syntax.html             |   57 +-
 docs/reference-manual/transforms.html         |   37 +-
 docs/reference-manual/types.html              |  129 +-
 docs/reference-manual/user-functions.html     |   37 +-
 docs/reference-manual/variational.html        |   37 +-
 docs/reference-manual/whitespace.html         |   37 +-
 docs/search.json                              | 9765 +++++++++--------
 .../stan-users-guide/algebraic-equations.html |   45 +-
 docs/stan-users-guide/bootstrap.html          |   51 +-
 docs/stan-users-guide/clustering.html         |   45 +-
 docs/stan-users-guide/complex-numbers.html    |   45 +-
 docs/stan-users-guide/copulas.html            |   45 +-
 docs/stan-users-guide/cross-validation.html   |   45 +-
 docs/stan-users-guide/custom-probability.html |   45 +-
 docs/stan-users-guide/dae.html                |   45 +-
 docs/stan-users-guide/decision-analysis.html  |   45 +-
 docs/stan-users-guide/efficiency-tuning.html  |   45 +-
 docs/stan-users-guide/finite-mixtures.html    |   45 +-
 docs/stan-users-guide/floating-point.html     |   51 +-
 docs/stan-users-guide/for-bugs-users.html     |   45 +-
 docs/stan-users-guide/gaussian-processes.html |  810 +-
 .../hyperspherical-models.html                |   45 +-
 docs/stan-users-guide/img/Figure_DDM.pdf      |  Bin 0 -> 100560 bytes
 docs/stan-users-guide/img/Figure_DDM.png      |  Bin 0 -> 74686 bytes
 docs/stan-users-guide/index.html              |   51 +-
 docs/stan-users-guide/latent-discrete.html    |   45 +-
 docs/stan-users-guide/matrices-arrays.html    |   45 +-
 docs/stan-users-guide/measurement-error.html  |   45 +-
 docs/stan-users-guide/missing-data.html       |   45 +-
 docs/stan-users-guide/multi-indexing.html     |   45 +-
 .../stan-users-guide/multiple-imputation.html | 1450 +++
 docs/stan-users-guide/odes.html               |   45 +-
 .../one-dimensional-integrals.html            |   45 +-
 docs/stan-users-guide/parallelization.html    |   45 +-
 .../posterior-prediction.html                 |   49 +-
 .../posterior-predictive-checks.html          |   49 +-
 docs/stan-users-guide/poststratification.html |   45 +-
 .../problematic-posteriors.html               |   45 +-
 .../proportionality-constants.html            |   45 +-
 docs/stan-users-guide/references.html         |   29 +-
 docs/stan-users-guide/regression.html         |  808 +-
 docs/stan-users-guide/reparameterization.html |   45 +-
 .../simulation-based-calibration.html         |  117 +-
 docs/stan-users-guide/sparse-ragged.html      |   45 +-
 docs/stan-users-guide/style-guide.html        |   45 +-
 docs/stan-users-guide/survival.html           |   51 +-
 docs/stan-users-guide/time-series.html        |   45 +-
 .../truncation-censoring.html                 |   45 +-
 docs/stan-users-guide/user-functions.html     |   45 +-
 docs/stan-users-guide/using-stanc.html        |   51 +-
 .../wiener_diffusion_model.html               | 1598 +++
 .../wiener_diffusion_model.pdf                |  Bin 0 -> 168306 bytes
 src/functions-reference/functions_index.qmd   |    5 +
 src/index.qmd                                 |    1 +
 439 files changed, 217856 insertions(+), 6003 deletions(-)
 create mode 100644 docs/2_39/404.html
 create mode 100644 docs/2_39/cmdstan-guide-2_39.pdf
 create mode 100644 docs/2_39/cmdstan-guide/bib.html
 create mode 100644 docs/2_39/cmdstan-guide/command_line_options.html
 create mode 100644 docs/2_39/cmdstan-guide/compiling_stan_programs.html
 create mode 100644 docs/2_39/cmdstan-guide/diagnose_config.html
 create mode 100644 docs/2_39/cmdstan-guide/diagnose_utility.html
 create mode 100644 docs/2_39/cmdstan-guide/example_model_data.html
 create mode 100644 docs/2_39/cmdstan-guide/external_code.html
 create mode 100644 docs/2_39/cmdstan-guide/generate_quantities_config.html
 create mode 100644 docs/2_39/cmdstan-guide/img/logo_tm.png
 create mode 100644 docs/2_39/cmdstan-guide/img/warmup-epochs.png
 create mode 100644 docs/2_39/cmdstan-guide/index.html
 create mode 100644 docs/2_39/cmdstan-guide/installation.html
 create mode 100644 docs/2_39/cmdstan-guide/json_apdx.html
 create mode 100644 docs/2_39/cmdstan-guide/laplace_sample_config.html
 create mode 100644 docs/2_39/cmdstan-guide/log_prob_config.html
 create mode 100644 docs/2_39/cmdstan-guide/mcmc_config.html
 create mode 100644 docs/2_39/cmdstan-guide/optimize_config.html
 create mode 100644 docs/2_39/cmdstan-guide/parallelization.html
 create mode 100644 docs/2_39/cmdstan-guide/pathfinder_config.html
 create mode 100644 docs/2_39/cmdstan-guide/print.html
 create mode 100644 docs/2_39/cmdstan-guide/rdump_apdx.html
 create mode 100644 docs/2_39/cmdstan-guide/stan_csv_apdx.html
 create mode 100644 docs/2_39/cmdstan-guide/stanc.html
 create mode 100644 docs/2_39/cmdstan-guide/stansummary.html
 create mode 100644 docs/2_39/cmdstan-guide/variational_config.html
 create mode 100644 docs/2_39/functions-reference-2_39.pdf
 create mode 100644 docs/2_39/functions-reference/array_operations.html
 create mode 100644 docs/2_39/functions-reference/binary_distributions.html
 create mode 100644 docs/2_39/functions-reference/bounded_continuous_distributions.html
 create mode 100644 docs/2_39/functions-reference/bounded_discrete_distributions.html
 create mode 100644 docs/2_39/functions-reference/circular_distributions.html
 create mode 100644 docs/2_39/functions-reference/complex-valued_basic_functions.html
 create mode 100644 docs/2_39/functions-reference/complex_matrix_operations.html
 create mode 100644 docs/2_39/functions-reference/compound_arithmetic_and_assignment.html
 create mode 100644 docs/2_39/functions-reference/continuous_distributions_on_0_1.html
 create mode 100644 docs/2_39/functions-reference/conventions_for_probability_functions.html
 create mode 100644 docs/2_39/functions-reference/correlation_matrix_distributions.html
 create mode 100644 docs/2_39/functions-reference/covariance_matrix_distributions.html
 create mode 100644 docs/2_39/functions-reference/deprecated_functions.html
 create mode 100644 docs/2_39/functions-reference/distributions_over_unbounded_vectors.html
 create mode 100644 docs/2_39/functions-reference/embedded_laplace.html
 create mode 100644 docs/2_39/functions-reference/functions_index.html
 create mode 100644 docs/2_39/functions-reference/hidden_markov_models.html
 create mode 100644 docs/2_39/functions-reference/higher-order_functions.html
 create mode 100644 docs/2_39/functions-reference/img/logo_tm.png
 create mode 100644 docs/2_39/functions-reference/index.html
 create mode 100644 docs/2_39/functions-reference/integer-valued_basic_functions.html
 create mode 100644 docs/2_39/functions-reference/mathematical_functions.html
 create mode 100644 docs/2_39/functions-reference/matrix_operations.html
 create mode 100644 docs/2_39/functions-reference/mixed_operations.html
 create mode 100644 docs/2_39/functions-reference/multivariate_discrete_distributions.html
 create mode 100644 docs/2_39/functions-reference/positive_continuous_distributions.html
 create mode 100644 docs/2_39/functions-reference/positive_lower-bounded_distributions.html
 create mode 100644 docs/2_39/functions-reference/real-valued_basic_functions.html
 create mode 100644 docs/2_39/functions-reference/references.html
 create mode 100644 docs/2_39/functions-reference/removed_functions.html
 create mode 100644 docs/2_39/functions-reference/simplex_distributions.html
 create mode 100644 docs/2_39/functions-reference/sparse_matrix_operations.html
 create mode 100644 docs/2_39/functions-reference/transform_functions.html
 create mode 100644 docs/2_39/functions-reference/unbounded_continuous_distributions.html
 create mode 100644 docs/2_39/functions-reference/unbounded_discrete_distributions.html
 create mode 100644 docs/2_39/functions-reference/void_functions.html
 create mode 100644 docs/2_39/img/logo_tm.png
 create mode 100644 docs/2_39/img/ulam-fermiac.png
 create mode 100644 docs/2_39/index.html
 create mode 100644 docs/2_39/quarto-config/quarto_styles.css
 create mode 100644 docs/2_39/reference-manual-2_39.pdf
 create mode 100644 docs/2_39/reference-manual/analysis.html
 create mode 100644 docs/2_39/reference-manual/blocks.html
 create mode 100644 docs/2_39/reference-manual/comments.html
 create mode 100644 docs/2_39/reference-manual/deprecations.html
 create mode 100644 docs/2_39/reference-manual/diagnostics.html
 create mode 100644 docs/2_39/reference-manual/encoding.html
 create mode 100644 docs/2_39/reference-manual/execution.html
 create mode 100644 docs/2_39/reference-manual/expressions.html
 create mode 100644 docs/2_39/reference-manual/img/logo_tm.png
 create mode 100644 docs/2_39/reference-manual/img/warmup-epochs.png
 create mode 100644 docs/2_39/reference-manual/includes.html
 create mode 100644 docs/2_39/reference-manual/index.html
 create mode 100644 docs/2_39/reference-manual/laplace.html
 create mode 100644 docs/2_39/reference-manual/laplace_embedded.html
 create mode 100644 docs/2_39/reference-manual/licenses.html
 create mode 100644 docs/2_39/reference-manual/mcmc.html
 create mode 100644 docs/2_39/reference-manual/optimization.html
 create mode 100644 docs/2_39/reference-manual/pathfinder.html
 create mode 100644 docs/2_39/reference-manual/references.html
 create mode 100644 docs/2_39/reference-manual/removals.html
 create mode 100644 docs/2_39/reference-manual/reproducibility.html
 create mode 100644 docs/2_39/reference-manual/statements.html
 create mode 100644 docs/2_39/reference-manual/syntax.html
 create mode 100644 docs/2_39/reference-manual/transforms.html
 create mode 100644 docs/2_39/reference-manual/types.html
 create mode 100644 docs/2_39/reference-manual/user-functions.html
 create mode 100644 docs/2_39/reference-manual/variational.html
 create mode 100644 docs/2_39/reference-manual/whitespace.html
 create mode 100644 docs/2_39/search.json
 create mode 100644 docs/2_39/site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css
 create mode 100644 docs/2_39/site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css
 create mode 100644 docs/2_39/site_libs/bootstrap/bootstrap-icons.css
 create mode 100644 docs/2_39/site_libs/bootstrap/bootstrap-icons.woff
 create mode 100644 docs/2_39/site_libs/bootstrap/bootstrap.min.js
 create mode 100644 docs/2_39/site_libs/clipboard/clipboard.min.js
 create mode 100644 docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.css
 create mode 100644 docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.js
 create mode 100644 docs/2_39/site_libs/quarto-contrib/glightbox/lightbox.css
 create mode 100644 docs/2_39/site_libs/quarto-html/anchor.min.js
 create mode 100644 docs/2_39/site_libs/quarto-html/popper.min.js
 create mode 100644 docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css
 create mode 100644 docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css
 create mode 100644 docs/2_39/site_libs/quarto-html/quarto.js
 create mode 100644 docs/2_39/site_libs/quarto-html/tippy.css
 create mode 100644 docs/2_39/site_libs/quarto-html/tippy.umd.min.js
 create mode 100644 docs/2_39/site_libs/quarto-nav/headroom.min.js
 create mode 100644 docs/2_39/site_libs/quarto-nav/quarto-nav.js
 create mode 100644 docs/2_39/site_libs/quarto-search/autocomplete.umd.js
 create mode 100644 docs/2_39/site_libs/quarto-search/fuse.min.js
 create mode 100644 docs/2_39/site_libs/quarto-search/quarto-search.js
 create mode 100644 docs/2_39/stan-users-guide-2_39.pdf
 create mode 100644 docs/2_39/stan-users-guide/algebraic-equations.html
 create mode 100644 docs/2_39/stan-users-guide/bootstrap.html
 create mode 100644 docs/2_39/stan-users-guide/clustering.html
 create mode 100644 docs/2_39/stan-users-guide/complex-numbers.html
 create mode 100644 docs/2_39/stan-users-guide/copulas.html
 create mode 100644 docs/2_39/stan-users-guide/cross-validation.html
 create mode 100644 docs/2_39/stan-users-guide/custom-probability.html
 create mode 100644 docs/2_39/stan-users-guide/dae.html
 create mode 100644 docs/2_39/stan-users-guide/decision-analysis.html
 create mode 100644 docs/2_39/stan-users-guide/efficiency-tuning.html
 create mode 100644 docs/2_39/stan-users-guide/finite-mixtures.html
 create mode 100644 docs/2_39/stan-users-guide/floating-point.html
 create mode 100644 docs/2_39/stan-users-guide/for-bugs-users.html
 create mode 100644 docs/2_39/stan-users-guide/gaussian-processes.html
 create mode 100644 docs/2_39/stan-users-guide/hyperspherical-models.html
 create mode 100644 docs/2_39/stan-users-guide/img/Figure_DDM.pdf
 create mode 100644 docs/2_39/stan-users-guide/img/Figure_DDM.png
 create mode 100644 docs/2_39/stan-users-guide/img/change-point-posterior.png
 create mode 100644 docs/2_39/stan-users-guide/img/funnel-fit.png
 create mode 100644 docs/2_39/stan-users-guide/img/funnel.png
 create mode 100644 docs/2_39/stan-users-guide/img/logo_tm.png
 create mode 100644 docs/2_39/stan-users-guide/img/non-identified-plus-prior.png
 create mode 100644 docs/2_39/stan-users-guide/img/non-identified.png
 create mode 100644 docs/2_39/stan-users-guide/img/one-param-identified.png
 create mode 100644 docs/2_39/stan-users-guide/img/ppc-nb-pois.jpg
 create mode 100644 docs/2_39/stan-users-guide/img/ppc-pois-pois.jpg
 create mode 100644 docs/2_39/stan-users-guide/img/ppc-pvalue-nb-pois-mean.jpg
 create mode 100644 docs/2_39/stan-users-guide/img/ppc-pvalue-nb-pois-sd.jpg
 create mode 100644 docs/2_39/stan-users-guide/img/s-discrete-posterior.png
 create mode 100644 docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-mu.png
 create mode 100644 docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-tau.png
 create mode 100644 docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-theta1.png
 create mode 100644 docs/2_39/stan-users-guide/img/sbc-normal-normal.png
 create mode 100644 docs/2_39/stan-users-guide/img/sbc-student-t-normal.png
 create mode 100644 docs/2_39/stan-users-guide/img/sho-ode-trajectory.png
 create mode 100644 docs/2_39/stan-users-guide/index.html
 create mode 100644 docs/2_39/stan-users-guide/latent-discrete.html
 create mode 100644 docs/2_39/stan-users-guide/matrices-arrays.html
 create mode 100644 docs/2_39/stan-users-guide/measurement-error.html
 create mode 100644 docs/2_39/stan-users-guide/missing-data.html
 create mode 100644 docs/2_39/stan-users-guide/multi-indexing.html
 create mode 100644 docs/2_39/stan-users-guide/multiple-imputation.html
 create mode 100644 docs/2_39/stan-users-guide/odes.html
 create mode 100644 docs/2_39/stan-users-guide/one-dimensional-integrals.html
 create mode 100644 docs/2_39/stan-users-guide/parallelization.html
 create mode 100644 docs/2_39/stan-users-guide/posterior-prediction.html
 create mode 100644 docs/2_39/stan-users-guide/posterior-predictive-checks.html
 create mode 100644 docs/2_39/stan-users-guide/poststratification.html
 create mode 100644 docs/2_39/stan-users-guide/problematic-posteriors.html
 create mode 100644 docs/2_39/stan-users-guide/proportionality-constants.html
 create mode 100644 docs/2_39/stan-users-guide/references.html
 create mode 100644 docs/2_39/stan-users-guide/regression.html
 create mode 100644 docs/2_39/stan-users-guide/reparameterization.html
 create mode 100644 docs/2_39/stan-users-guide/simulation-based-calibration.html
 create mode 100644 docs/2_39/stan-users-guide/sparse-ragged.html
 create mode 100644 docs/2_39/stan-users-guide/style-guide.html
 create mode 100644 docs/2_39/stan-users-guide/survival.html
 create mode 100644 docs/2_39/stan-users-guide/time-series.html
 create mode 100644 docs/2_39/stan-users-guide/truncation-censoring.html
 create mode 100644 docs/2_39/stan-users-guide/user-functions.html
 create mode 100644 docs/2_39/stan-users-guide/using-stanc.html
 create mode 100644 docs/2_39/stan-users-guide/wiener_diffusion_model.html
 create mode 100644 docs/2_39/stan-users-guide/wiener_diffusion_model.pdf
 create mode 100644 docs/functions-reference/embedded_laplace.html
 create mode 100644 docs/reference-manual/laplace_embedded.html
 create mode 100644 docs/stan-users-guide/img/Figure_DDM.pdf
 create mode 100644 docs/stan-users-guide/img/Figure_DDM.png
 create mode 100644 docs/stan-users-guide/multiple-imputation.html
 create mode 100644 docs/stan-users-guide/wiener_diffusion_model.html
 create mode 100644 docs/stan-users-guide/wiener_diffusion_model.pdf

diff --git a/docs/2_38/cmdstan-guide/bib.html b/docs/2_38/cmdstan-guide/bib.html
index cda8b7cf5..c17d804f1 100644
--- a/docs/2_38/cmdstan-guide/bib.html
+++ b/docs/2_38/cmdstan-guide/bib.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/bib.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -215,6 +215,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/bib.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/command_line_options.html b/docs/2_38/cmdstan-guide/command_line_options.html
index 81addd6f9..d9d9bc2de 100644
--- a/docs/2_38/cmdstan-guide/command_line_options.html
+++ b/docs/2_38/cmdstan-guide/command_line_options.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/command_line_options.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -453,6 +453,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Command-Line Interface Overview</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/command_line_options.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/compiling_stan_programs.html b/docs/2_38/cmdstan-guide/compiling_stan_programs.html
index d4b8592ce..3057aa25e 100644
--- a/docs/2_38/cmdstan-guide/compiling_stan_programs.html
+++ b/docs/2_38/cmdstan-guide/compiling_stan_programs.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/compiling_stan_programs.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -488,6 +488,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/compiling_stan_programs.html">Compiling a Stan Program</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/compiling_stan_programs.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/diagnose_config.html b/docs/2_38/cmdstan-guide/diagnose_config.html
index 7fb35fc53..d46f113ee 100644
--- a/docs/2_38/cmdstan-guide/diagnose_config.html
+++ b/docs/2_38/cmdstan-guide/diagnose_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/diagnose_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -444,6 +444,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/diagnose_config.html">Diagnosing HMC by Comparison of Gradients</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/diagnose_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/diagnose_utility.html b/docs/2_38/cmdstan-guide/diagnose_utility.html
index b23b61ea8..1b4c84a96 100644
--- a/docs/2_38/cmdstan-guide/diagnose_utility.html
+++ b/docs/2_38/cmdstan-guide/diagnose_utility.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/diagnose_utility.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -490,6 +490,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/diagnose_utility.html"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/diagnose_utility.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/example_model_data.html b/docs/2_38/cmdstan-guide/example_model_data.html
index 6e9039a47..de05b64dc 100644
--- a/docs/2_38/cmdstan-guide/example_model_data.html
+++ b/docs/2_38/cmdstan-guide/example_model_data.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/example_model_data.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -449,6 +449,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/example_model_data.html">Example Model and Data</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/example_model_data.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/external_code.html b/docs/2_38/cmdstan-guide/external_code.html
index 0fd42c59f..9df8c2332 100644
--- a/docs/2_38/cmdstan-guide/external_code.html
+++ b/docs/2_38/cmdstan-guide/external_code.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/external_code.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -452,6 +452,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/external_code.html">Using external C++ code</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/external_code.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/generate_quantities_config.html b/docs/2_38/cmdstan-guide/generate_quantities_config.html
index 9f631fb8f..5c7a2eebf 100644
--- a/docs/2_38/cmdstan-guide/generate_quantities_config.html
+++ b/docs/2_38/cmdstan-guide/generate_quantities_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/generate_quantities_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -453,6 +453,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/generate_quantities_config.html">Generating Quantities of Interest from a Fitted Model</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/generate_quantities_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/index.html b/docs/2_38/cmdstan-guide/index.html
index fe5f177f6..9529afc1c 100644
--- a/docs/2_38/cmdstan-guide/index.html
+++ b/docs/2_38/cmdstan-guide/index.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/index.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
diff --git a/docs/2_38/cmdstan-guide/installation.html b/docs/2_38/cmdstan-guide/installation.html
index 74fa8aa25..86f1f3d08 100644
--- a/docs/2_38/cmdstan-guide/installation.html
+++ b/docs/2_38/cmdstan-guide/installation.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/installation.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -462,6 +462,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">CmdStan Installation</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/installation.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/json_apdx.html b/docs/2_38/cmdstan-guide/json_apdx.html
index 5b2b06ab7..c9bfe4cf3 100644
--- a/docs/2_38/cmdstan-guide/json_apdx.html
+++ b/docs/2_38/cmdstan-guide/json_apdx.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/json_apdx.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -486,6 +486,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/json_apdx.html">JSON Format for CmdStan</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/json_apdx.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/laplace_sample_config.html b/docs/2_38/cmdstan-guide/laplace_sample_config.html
index 6d639198d..6b461ca4e 100644
--- a/docs/2_38/cmdstan-guide/laplace_sample_config.html
+++ b/docs/2_38/cmdstan-guide/laplace_sample_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/laplace_sample_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -450,6 +450,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/laplace_sample_config.html">Laplace sampling</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/laplace_sample_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/log_prob_config.html b/docs/2_38/cmdstan-guide/log_prob_config.html
index 01ea7fdcc..5112cfc84 100644
--- a/docs/2_38/cmdstan-guide/log_prob_config.html
+++ b/docs/2_38/cmdstan-guide/log_prob_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/log_prob_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -453,6 +453,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/log_prob_config.html">Extracting log probabilities and gradients for diagnostics</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/log_prob_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/mcmc_config.html b/docs/2_38/cmdstan-guide/mcmc_config.html
index dd4f21874..0981094f6 100644
--- a/docs/2_38/cmdstan-guide/mcmc_config.html
+++ b/docs/2_38/cmdstan-guide/mcmc_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/mcmc_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -498,6 +498,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/mcmc_config.html">MCMC Sampling using Hamiltonian Monte Carlo</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/mcmc_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/optimize_config.html b/docs/2_38/cmdstan-guide/optimize_config.html
index a16b09626..84ba9183e 100644
--- a/docs/2_38/cmdstan-guide/optimize_config.html
+++ b/docs/2_38/cmdstan-guide/optimize_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/optimize_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -450,6 +450,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/optimize_config.html">Optimization</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/optimize_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/parallelization.html b/docs/2_38/cmdstan-guide/parallelization.html
index ba2d09864..bf56ac662 100644
--- a/docs/2_38/cmdstan-guide/parallelization.html
+++ b/docs/2_38/cmdstan-guide/parallelization.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/parallelization.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -434,6 +434,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/parallelization.html">Parallelization</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/parallelization.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/pathfinder_config.html b/docs/2_38/cmdstan-guide/pathfinder_config.html
index 247e91e8b..de14d9417 100644
--- a/docs/2_38/cmdstan-guide/pathfinder_config.html
+++ b/docs/2_38/cmdstan-guide/pathfinder_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/pathfinder_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -470,6 +470,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/pathfinder_config.html">Pathfinder Method for Approximate Bayesian Inference</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/pathfinder_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/print.html b/docs/2_38/cmdstan-guide/print.html
index 1a33ecfea..1b1850ff2 100644
--- a/docs/2_38/cmdstan-guide/print.html
+++ b/docs/2_38/cmdstan-guide/print.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/print.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -415,6 +415,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/print.html"><code>print</code> (deprecated): MCMC Output Analysis</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/print.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/rdump_apdx.html b/docs/2_38/cmdstan-guide/rdump_apdx.html
index fd5461561..5e7fc4c20 100644
--- a/docs/2_38/cmdstan-guide/rdump_apdx.html
+++ b/docs/2_38/cmdstan-guide/rdump_apdx.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/rdump_apdx.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -465,6 +465,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/rdump_apdx.html">RDump Format for CmdStan</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/rdump_apdx.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/stan_csv_apdx.html b/docs/2_38/cmdstan-guide/stan_csv_apdx.html
index f764c5ff8..cd03fe40f 100644
--- a/docs/2_38/cmdstan-guide/stan_csv_apdx.html
+++ b/docs/2_38/cmdstan-guide/stan_csv_apdx.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/stan_csv_apdx.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -457,6 +457,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Stan CSV File Format</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/stan_csv_apdx.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/stanc.html b/docs/2_38/cmdstan-guide/stanc.html
index 7ed425e4a..a11a17e70 100644
--- a/docs/2_38/cmdstan-guide/stanc.html
+++ b/docs/2_38/cmdstan-guide/stanc.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/stanc.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -419,6 +419,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html"><code>stanc</code>: Translating Stan to C++</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/stanc.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/stansummary.html b/docs/2_38/cmdstan-guide/stansummary.html
index 4611817d4..1eb154019 100644
--- a/docs/2_38/cmdstan-guide/stansummary.html
+++ b/docs/2_38/cmdstan-guide/stansummary.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/stansummary.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -473,6 +473,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stansummary.html"><code>stansummary</code>: MCMC Output Analysis</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/stansummary.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/cmdstan-guide/variational_config.html b/docs/2_38/cmdstan-guide/variational_config.html
index e0ca7d9f0..d6b583d1f 100644
--- a/docs/2_38/cmdstan-guide/variational_config.html
+++ b/docs/2_38/cmdstan-guide/variational_config.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/cmdstan-guide/variational_config.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -469,6 +469,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/variational_config.html">Variational Inference using ADVI</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/cmdstan-guide/variational_config.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/array_operations.html b/docs/2_38/functions-reference/array_operations.html
index a61e53b3a..a489b29eb 100644
--- a/docs/2_38/functions-reference/array_operations.html
+++ b/docs/2_38/functions-reference/array_operations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/array_operations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -570,6 +570,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/array_operations.html">Array Operations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/array_operations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/binary_distributions.html b/docs/2_38/functions-reference/binary_distributions.html
index 6db376df1..341ebee29 100644
--- a/docs/2_38/functions-reference/binary_distributions.html
+++ b/docs/2_38/functions-reference/binary_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/binary_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -541,6 +541,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Binary Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/binary_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/bounded_continuous_distributions.html b/docs/2_38/functions-reference/bounded_continuous_distributions.html
index f69007d73..a1330a390 100644
--- a/docs/2_38/functions-reference/bounded_continuous_distributions.html
+++ b/docs/2_38/functions-reference/bounded_continuous_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/bounded_continuous_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -529,6 +529,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/bounded_continuous_distributions.html">Bounded Continuous Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/bounded_continuous_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/bounded_discrete_distributions.html b/docs/2_38/functions-reference/bounded_discrete_distributions.html
index 5b142af7e..47cedad56 100644
--- a/docs/2_38/functions-reference/bounded_discrete_distributions.html
+++ b/docs/2_38/functions-reference/bounded_discrete_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/bounded_discrete_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -594,6 +594,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/bounded_discrete_distributions.html">Bounded Discrete Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/bounded_discrete_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/circular_distributions.html b/docs/2_38/functions-reference/circular_distributions.html
index 7e81ff5cb..bb6b00d58 100644
--- a/docs/2_38/functions-reference/circular_distributions.html
+++ b/docs/2_38/functions-reference/circular_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/circular_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -564,6 +564,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/circular_distributions.html">Circular Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/circular_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/complex-valued_basic_functions.html b/docs/2_38/functions-reference/complex-valued_basic_functions.html
index 80143c22c..db600e3ba 100644
--- a/docs/2_38/functions-reference/complex-valued_basic_functions.html
+++ b/docs/2_38/functions-reference/complex-valued_basic_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/complex-valued_basic_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -577,6 +577,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/complex-valued_basic_functions.html">Complex-Valued Basic Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/complex-valued_basic_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/complex_matrix_operations.html b/docs/2_38/functions-reference/complex_matrix_operations.html
index 03bb5f7e1..864630ec4 100644
--- a/docs/2_38/functions-reference/complex_matrix_operations.html
+++ b/docs/2_38/functions-reference/complex_matrix_operations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/complex_matrix_operations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -611,6 +611,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/complex_matrix_operations.html">Complex Matrix Operations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/complex_matrix_operations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/compound_arithmetic_and_assignment.html b/docs/2_38/functions-reference/compound_arithmetic_and_assignment.html
index 8d34e2cbc..dfca90c6d 100644
--- a/docs/2_38/functions-reference/compound_arithmetic_and_assignment.html
+++ b/docs/2_38/functions-reference/compound_arithmetic_and_assignment.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/compound_arithmetic_and_assignment.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -500,6 +500,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/compound_arithmetic_and_assignment.html">Compound Arithmetic and Assignment</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/compound_arithmetic_and_assignment.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/continuous_distributions_on_0_1.html b/docs/2_38/functions-reference/continuous_distributions_on_0_1.html
index 252cce468..6d653285c 100644
--- a/docs/2_38/functions-reference/continuous_distributions_on_0_1.html
+++ b/docs/2_38/functions-reference/continuous_distributions_on_0_1.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/continuous_distributions_on_0_1.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -535,6 +535,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/continuous_distributions_on_0_1.html">Continuous Distributions on [0, 1]</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/continuous_distributions_on_0_1.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/conventions_for_probability_functions.html b/docs/2_38/functions-reference/conventions_for_probability_functions.html
index 838febad9..a74286267 100644
--- a/docs/2_38/functions-reference/conventions_for_probability_functions.html
+++ b/docs/2_38/functions-reference/conventions_for_probability_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/conventions_for_probability_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -574,6 +574,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/conventions_for_probability_functions.html">Conventions for Probability Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/conventions_for_probability_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/correlation_matrix_distributions.html b/docs/2_38/functions-reference/correlation_matrix_distributions.html
index b1b16ce4c..e74f96d7d 100644
--- a/docs/2_38/functions-reference/correlation_matrix_distributions.html
+++ b/docs/2_38/functions-reference/correlation_matrix_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/correlation_matrix_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -589,6 +589,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/correlation_matrix_distributions.html">Correlation Matrix Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/correlation_matrix_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/covariance_matrix_distributions.html b/docs/2_38/functions-reference/covariance_matrix_distributions.html
index 77e397def..67bfa00b5 100644
--- a/docs/2_38/functions-reference/covariance_matrix_distributions.html
+++ b/docs/2_38/functions-reference/covariance_matrix_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/covariance_matrix_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -545,6 +545,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/covariance_matrix_distributions.html">Covariance Matrix Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/covariance_matrix_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/deprecated_functions.html b/docs/2_38/functions-reference/deprecated_functions.html
index 2b5f328d0..24e6b568c 100644
--- a/docs/2_38/functions-reference/deprecated_functions.html
+++ b/docs/2_38/functions-reference/deprecated_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/deprecated_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -570,6 +570,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/deprecated_functions.html">Deprecated Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/deprecated_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/distributions_over_unbounded_vectors.html b/docs/2_38/functions-reference/distributions_over_unbounded_vectors.html
index dc0eee2d7..4ee9b089d 100644
--- a/docs/2_38/functions-reference/distributions_over_unbounded_vectors.html
+++ b/docs/2_38/functions-reference/distributions_over_unbounded_vectors.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/distributions_over_unbounded_vectors.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -590,6 +590,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/distributions_over_unbounded_vectors.html">Distributions over Unbounded Vectors</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/distributions_over_unbounded_vectors.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/functions_index.html b/docs/2_38/functions-reference/functions_index.html
index b89c87236..1a442781d 100644
--- a/docs/2_38/functions-reference/functions_index.html
+++ b/docs/2_38/functions-reference/functions_index.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/functions_index.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -514,6 +514,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Appendix</a></li><li class="breadcrumb-item"><a href="../functions-reference/functions_index.html">Alphabetical Index</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/functions_index.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/hidden_markov_models.html b/docs/2_38/functions-reference/hidden_markov_models.html
index 639e59fb4..50ef7d251 100644
--- a/docs/2_38/functions-reference/hidden_markov_models.html
+++ b/docs/2_38/functions-reference/hidden_markov_models.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/hidden_markov_models.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -524,6 +524,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Hidden Markov Models</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/hidden_markov_models.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/higher-order_functions.html b/docs/2_38/functions-reference/higher-order_functions.html
index d37d0bdb1..82220b73f 100644
--- a/docs/2_38/functions-reference/higher-order_functions.html
+++ b/docs/2_38/functions-reference/higher-order_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/higher-order_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -612,6 +612,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/higher-order_functions.html">Higher-Order Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/higher-order_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/index.html b/docs/2_38/functions-reference/index.html
index 273f5b5b9..e5b300e0b 100644
--- a/docs/2_38/functions-reference/index.html
+++ b/docs/2_38/functions-reference/index.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/index.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
diff --git a/docs/2_38/functions-reference/integer-valued_basic_functions.html b/docs/2_38/functions-reference/integer-valued_basic_functions.html
index 61f8975e8..73077daba 100644
--- a/docs/2_38/functions-reference/integer-valued_basic_functions.html
+++ b/docs/2_38/functions-reference/integer-valued_basic_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/integer-valued_basic_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -532,6 +532,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/integer-valued_basic_functions.html">Integer-Valued Basic Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/integer-valued_basic_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/mathematical_functions.html b/docs/2_38/functions-reference/mathematical_functions.html
index 47259bf12..cad19a557 100644
--- a/docs/2_38/functions-reference/mathematical_functions.html
+++ b/docs/2_38/functions-reference/mathematical_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/mathematical_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -527,6 +527,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Appendix</a></li><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Mathematical Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/mathematical_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/matrix_operations.html b/docs/2_38/functions-reference/matrix_operations.html
index c4aba8c04..c09abbf5f 100644
--- a/docs/2_38/functions-reference/matrix_operations.html
+++ b/docs/2_38/functions-reference/matrix_operations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/matrix_operations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -633,6 +633,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/matrix_operations.html">Matrix Operations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/matrix_operations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/mixed_operations.html b/docs/2_38/functions-reference/mixed_operations.html
index 8101005d9..e7bfb8235 100644
--- a/docs/2_38/functions-reference/mixed_operations.html
+++ b/docs/2_38/functions-reference/mixed_operations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/mixed_operations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -521,6 +521,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/mixed_operations.html">Mixed Operations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/mixed_operations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/multivariate_discrete_distributions.html b/docs/2_38/functions-reference/multivariate_discrete_distributions.html
index 348a6c101..417c70760 100644
--- a/docs/2_38/functions-reference/multivariate_discrete_distributions.html
+++ b/docs/2_38/functions-reference/multivariate_discrete_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/multivariate_discrete_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -541,6 +541,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/multivariate_discrete_distributions.html">Multivariate Discrete Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/multivariate_discrete_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/positive_continuous_distributions.html b/docs/2_38/functions-reference/positive_continuous_distributions.html
index c29d8e03e..7a68c0426 100644
--- a/docs/2_38/functions-reference/positive_continuous_distributions.html
+++ b/docs/2_38/functions-reference/positive_continuous_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/positive_continuous_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -589,6 +589,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/positive_continuous_distributions.html">Positive Continuous Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/positive_continuous_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/positive_lower-bounded_distributions.html b/docs/2_38/functions-reference/positive_lower-bounded_distributions.html
index c658d57df..137031d3a 100644
--- a/docs/2_38/functions-reference/positive_lower-bounded_distributions.html
+++ b/docs/2_38/functions-reference/positive_lower-bounded_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/positive_lower-bounded_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -565,6 +565,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/positive_lower-bounded_distributions.html">Positive Lower-Bounded Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/positive_lower-bounded_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/real-valued_basic_functions.html b/docs/2_38/functions-reference/real-valued_basic_functions.html
index b46001086..4f7a96719 100644
--- a/docs/2_38/functions-reference/real-valued_basic_functions.html
+++ b/docs/2_38/functions-reference/real-valued_basic_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/real-valued_basic_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -616,6 +616,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/real-valued_basic_functions.html">Real-Valued Basic Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/real-valued_basic_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/references.html b/docs/2_38/functions-reference/references.html
index 4458bfda0..a65d071b8 100644
--- a/docs/2_38/functions-reference/references.html
+++ b/docs/2_38/functions-reference/references.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/references.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -215,6 +215,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/references.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/removed_functions.html b/docs/2_38/functions-reference/removed_functions.html
index 041785972..26b8a6d22 100644
--- a/docs/2_38/functions-reference/removed_functions.html
+++ b/docs/2_38/functions-reference/removed_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/removed_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -528,6 +528,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/removed_functions.html">Removed Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/removed_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/simplex_distributions.html b/docs/2_38/functions-reference/simplex_distributions.html
index bd3e01b16..4a51774c9 100644
--- a/docs/2_38/functions-reference/simplex_distributions.html
+++ b/docs/2_38/functions-reference/simplex_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/simplex_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -564,6 +564,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/simplex_distributions.html">Simplex Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/simplex_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/sparse_matrix_operations.html b/docs/2_38/functions-reference/sparse_matrix_operations.html
index 25534a49a..9f68ccada 100644
--- a/docs/2_38/functions-reference/sparse_matrix_operations.html
+++ b/docs/2_38/functions-reference/sparse_matrix_operations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/sparse_matrix_operations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -533,6 +533,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/sparse_matrix_operations.html">Sparse Matrix Operations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/sparse_matrix_operations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/transform_functions.html b/docs/2_38/functions-reference/transform_functions.html
index 43ab4cbc9..e0ba6b791 100644
--- a/docs/2_38/functions-reference/transform_functions.html
+++ b/docs/2_38/functions-reference/transform_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/transform_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -548,6 +548,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/transform_functions.html">Variable Transformation Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/transform_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/unbounded_continuous_distributions.html b/docs/2_38/functions-reference/unbounded_continuous_distributions.html
index b08557e47..eb5a37ad9 100644
--- a/docs/2_38/functions-reference/unbounded_continuous_distributions.html
+++ b/docs/2_38/functions-reference/unbounded_continuous_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/unbounded_continuous_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -606,6 +606,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Unbounded Continuous Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/unbounded_continuous_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/unbounded_discrete_distributions.html b/docs/2_38/functions-reference/unbounded_discrete_distributions.html
index a8a99e58b..8d1121eb9 100644
--- a/docs/2_38/functions-reference/unbounded_discrete_distributions.html
+++ b/docs/2_38/functions-reference/unbounded_discrete_distributions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/unbounded_discrete_distributions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -590,6 +590,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/unbounded_discrete_distributions.html">Unbounded Discrete Distributions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/unbounded_discrete_distributions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/functions-reference/void_functions.html b/docs/2_38/functions-reference/void_functions.html
index 5752c5b39..af45c6e4f 100644
--- a/docs/2_38/functions-reference/void_functions.html
+++ b/docs/2_38/functions-reference/void_functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/functions-reference/void_functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -497,6 +497,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Void Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/functions-reference/void_functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/analysis.html b/docs/2_38/reference-manual/analysis.html
index 2831fdcf7..136b4103b 100644
--- a/docs/2_38/reference-manual/analysis.html
+++ b/docs/2_38/reference-manual/analysis.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/analysis.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -483,6 +483,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/analysis.html">Posterior Analysis</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/analysis.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/blocks.html b/docs/2_38/reference-manual/blocks.html
index cab4af7f5..d3767571e 100644
--- a/docs/2_38/reference-manual/blocks.html
+++ b/docs/2_38/reference-manual/blocks.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/blocks.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -536,6 +536,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/blocks.html">Program Blocks</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/blocks.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/comments.html b/docs/2_38/reference-manual/comments.html
index 9cb0493df..bd5915c10 100644
--- a/docs/2_38/reference-manual/comments.html
+++ b/docs/2_38/reference-manual/comments.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/comments.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -454,6 +454,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/comments.html">Comments</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/comments.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/deprecations.html b/docs/2_38/reference-manual/deprecations.html
index 9ba0faea4..1ebc12003 100644
--- a/docs/2_38/reference-manual/deprecations.html
+++ b/docs/2_38/reference-manual/deprecations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/deprecations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -455,6 +455,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/deprecations.html">Deprecated Features</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/deprecations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/diagnostics.html b/docs/2_38/reference-manual/diagnostics.html
index 3feadead2..ee4750e0b 100644
--- a/docs/2_38/reference-manual/diagnostics.html
+++ b/docs/2_38/reference-manual/diagnostics.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/diagnostics.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -454,6 +454,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/diagnostics.html">Diagnostic Mode</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/diagnostics.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/encoding.html b/docs/2_38/reference-manual/encoding.html
index 4c2d756d6..6f16c051b 100644
--- a/docs/2_38/reference-manual/encoding.html
+++ b/docs/2_38/reference-manual/encoding.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/encoding.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -423,6 +423,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Character Encoding</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/encoding.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/execution.html b/docs/2_38/reference-manual/execution.html
index 4f0598b44..6138cb4ba 100644
--- a/docs/2_38/reference-manual/execution.html
+++ b/docs/2_38/reference-manual/execution.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/execution.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -526,6 +526,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/execution.html">Program Execution</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/execution.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/expressions.html b/docs/2_38/reference-manual/expressions.html
index 273c729f1..77b4a0315 100644
--- a/docs/2_38/reference-manual/expressions.html
+++ b/docs/2_38/reference-manual/expressions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/expressions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -568,6 +568,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/expressions.html">Expressions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/expressions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/includes.html b/docs/2_38/reference-manual/includes.html
index 2b2fd3e95..509dba21d 100644
--- a/docs/2_38/reference-manual/includes.html
+++ b/docs/2_38/reference-manual/includes.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/includes.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -459,6 +459,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/includes.html">Includes</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/includes.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/index.html b/docs/2_38/reference-manual/index.html
index 3697e8e63..432c5c578 100644
--- a/docs/2_38/reference-manual/index.html
+++ b/docs/2_38/reference-manual/index.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/index.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
diff --git a/docs/2_38/reference-manual/laplace.html b/docs/2_38/reference-manual/laplace.html
index 533cc32b2..2a02354ea 100644
--- a/docs/2_38/reference-manual/laplace.html
+++ b/docs/2_38/reference-manual/laplace.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/laplace.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -445,6 +445,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace.html">Laplace Approximation</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/laplace.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/licenses.html b/docs/2_38/reference-manual/licenses.html
index e6ff6813a..d3fd619f2 100644
--- a/docs/2_38/reference-manual/licenses.html
+++ b/docs/2_38/reference-manual/licenses.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/licenses.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -423,6 +423,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Usage</a></li><li class="breadcrumb-item"><a href="../reference-manual/licenses.html">Licenses and Dependencies</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/licenses.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/mcmc.html b/docs/2_38/reference-manual/mcmc.html
index 768f46a75..6de90a2f0 100644
--- a/docs/2_38/reference-manual/mcmc.html
+++ b/docs/2_38/reference-manual/mcmc.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/mcmc.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -530,6 +530,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">MCMC Sampling</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/mcmc.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/optimization.html b/docs/2_38/reference-manual/optimization.html
index d0542e3ed..73f21e6ef 100644
--- a/docs/2_38/reference-manual/optimization.html
+++ b/docs/2_38/reference-manual/optimization.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/optimization.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -512,6 +512,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/optimization.html">Optimization</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/optimization.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/pathfinder.html b/docs/2_38/reference-manual/pathfinder.html
index 077f145f0..6d87c0785 100644
--- a/docs/2_38/reference-manual/pathfinder.html
+++ b/docs/2_38/reference-manual/pathfinder.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/pathfinder.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -469,6 +469,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/pathfinder.html">Pathfinder</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/pathfinder.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/references.html b/docs/2_38/reference-manual/references.html
index 1c0ed4dae..6a0b1651a 100644
--- a/docs/2_38/reference-manual/references.html
+++ b/docs/2_38/reference-manual/references.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/references.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -215,6 +215,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/references.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/removals.html b/docs/2_38/reference-manual/removals.html
index 82d0e1fc1..b35cacb95 100644
--- a/docs/2_38/reference-manual/removals.html
+++ b/docs/2_38/reference-manual/removals.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/removals.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -464,6 +464,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/removals.html">Removed Features</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/removals.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/reproducibility.html b/docs/2_38/reference-manual/reproducibility.html
index 980cc8215..f1bd3bd29 100644
--- a/docs/2_38/reference-manual/reproducibility.html
+++ b/docs/2_38/reference-manual/reproducibility.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/reproducibility.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -439,6 +439,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Usage</a></li><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Reproducibility</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/reproducibility.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/statements.html b/docs/2_38/reference-manual/statements.html
index b7678fe74..2a729ef5b 100644
--- a/docs/2_38/reference-manual/statements.html
+++ b/docs/2_38/reference-manual/statements.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/statements.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -536,6 +536,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/statements.html">Statements</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/statements.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/syntax.html b/docs/2_38/reference-manual/syntax.html
index 45bf930b2..eb1a253a8 100644
--- a/docs/2_38/reference-manual/syntax.html
+++ b/docs/2_38/reference-manual/syntax.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/syntax.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -472,6 +472,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/syntax.html">Language Syntax</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/syntax.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/transforms.html b/docs/2_38/reference-manual/transforms.html
index ec54cce49..e13178c82 100644
--- a/docs/2_38/reference-manual/transforms.html
+++ b/docs/2_38/reference-manual/transforms.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/transforms.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -588,6 +588,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/transforms.html">Constraint Transforms</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/transforms.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/types.html b/docs/2_38/reference-manual/types.html
index 9a11f4252..03f2e002a 100644
--- a/docs/2_38/reference-manual/types.html
+++ b/docs/2_38/reference-manual/types.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/types.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -574,6 +574,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/types.html">Data Types and Declarations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/types.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/user-functions.html b/docs/2_38/reference-manual/user-functions.html
index a9b5cb0f0..c2fa8d664 100644
--- a/docs/2_38/reference-manual/user-functions.html
+++ b/docs/2_38/reference-manual/user-functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/user-functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -491,6 +491,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/user-functions.html">User-Defined Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/user-functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/variational.html b/docs/2_38/reference-manual/variational.html
index e979ed016..a41df9856 100644
--- a/docs/2_38/reference-manual/variational.html
+++ b/docs/2_38/reference-manual/variational.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/variational.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -474,6 +474,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/variational.html">Variational Inference</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/variational.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/reference-manual/whitespace.html b/docs/2_38/reference-manual/whitespace.html
index f7df00591..c2969ea8a 100644
--- a/docs/2_38/reference-manual/whitespace.html
+++ b/docs/2_38/reference-manual/whitespace.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/reference-manual/whitespace.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -420,6 +420,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/whitespace.html">Whitespace</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/reference-manual/whitespace.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/algebraic-equations.html b/docs/2_38/stan-users-guide/algebraic-equations.html
index 85d905234..1ed515448 100644
--- a/docs/2_38/stan-users-guide/algebraic-equations.html
+++ b/docs/2_38/stan-users-guide/algebraic-equations.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/algebraic-equations.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -615,6 +615,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/algebraic-equations.html">Solving Algebraic Equations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/algebraic-equations.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/bootstrap.html b/docs/2_38/stan-users-guide/bootstrap.html
index 1140f65ac..299c32214 100644
--- a/docs/2_38/stan-users-guide/bootstrap.html
+++ b/docs/2_38/stan-users-guide/bootstrap.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/bootstrap.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -615,6 +615,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/bootstrap.html">The Bootstrap and Bagging</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/bootstrap.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/clustering.html b/docs/2_38/stan-users-guide/clustering.html
index bb3a43bfb..d7ba47c64 100644
--- a/docs/2_38/stan-users-guide/clustering.html
+++ b/docs/2_38/stan-users-guide/clustering.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/clustering.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -630,6 +630,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/clustering.html">Clustering Models</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/clustering.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/complex-numbers.html b/docs/2_38/stan-users-guide/complex-numbers.html
index 5ab464f1a..e2068d39e 100644
--- a/docs/2_38/stan-users-guide/complex-numbers.html
+++ b/docs/2_38/stan-users-guide/complex-numbers.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/complex-numbers.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -596,6 +596,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/complex-numbers.html">Complex Numbers</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/complex-numbers.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/copulas.html b/docs/2_38/stan-users-guide/copulas.html
index bbd72e525..e6e9b063a 100644
--- a/docs/2_38/stan-users-guide/copulas.html
+++ b/docs/2_38/stan-users-guide/copulas.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/copulas.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -609,6 +609,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/copulas.html">Copulas</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/copulas.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/cross-validation.html b/docs/2_38/stan-users-guide/cross-validation.html
index 78393efbe..e8fe2d025 100644
--- a/docs/2_38/stan-users-guide/cross-validation.html
+++ b/docs/2_38/stan-users-guide/cross-validation.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/cross-validation.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -620,6 +620,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/cross-validation.html">Held-Out Evaluation and Cross-Validation</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/cross-validation.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/custom-probability.html b/docs/2_38/stan-users-guide/custom-probability.html
index 8e724a256..0e791c57d 100644
--- a/docs/2_38/stan-users-guide/custom-probability.html
+++ b/docs/2_38/stan-users-guide/custom-probability.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/custom-probability.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -588,6 +588,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/custom-probability.html">Custom Probability Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/custom-probability.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/dae.html b/docs/2_38/stan-users-guide/dae.html
index 665b22ee7..40b27c2dd 100644
--- a/docs/2_38/stan-users-guide/dae.html
+++ b/docs/2_38/stan-users-guide/dae.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/dae.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -614,6 +614,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/dae.html">Differential-Algebraic Equations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/dae.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/decision-analysis.html b/docs/2_38/stan-users-guide/decision-analysis.html
index 918b21040..2d7598b1e 100644
--- a/docs/2_38/stan-users-guide/decision-analysis.html
+++ b/docs/2_38/stan-users-guide/decision-analysis.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/decision-analysis.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -611,6 +611,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/decision-analysis.html">Decision Analysis</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/decision-analysis.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/efficiency-tuning.html b/docs/2_38/stan-users-guide/efficiency-tuning.html
index df3e37a7e..263d19cdc 100644
--- a/docs/2_38/stan-users-guide/efficiency-tuning.html
+++ b/docs/2_38/stan-users-guide/efficiency-tuning.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -644,6 +644,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/efficiency-tuning.html">Efficiency Tuning</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/finite-mixtures.html b/docs/2_38/stan-users-guide/finite-mixtures.html
index ea76dd5bb..f70019097 100644
--- a/docs/2_38/stan-users-guide/finite-mixtures.html
+++ b/docs/2_38/stan-users-guide/finite-mixtures.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/finite-mixtures.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -626,6 +626,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/finite-mixtures.html">Finite Mixtures</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/finite-mixtures.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/floating-point.html b/docs/2_38/stan-users-guide/floating-point.html
index 4261cedd6..a3d47d045 100644
--- a/docs/2_38/stan-users-guide/floating-point.html
+++ b/docs/2_38/stan-users-guide/floating-point.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/floating-point.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -574,6 +574,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Floating Point Arithmetic</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/floating-point.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/for-bugs-users.html b/docs/2_38/stan-users-guide/for-bugs-users.html
index ff54338ba..e4799bc41 100644
--- a/docs/2_38/stan-users-guide/for-bugs-users.html
+++ b/docs/2_38/stan-users-guide/for-bugs-users.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/for-bugs-users.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -618,6 +618,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/for-bugs-users.html">Transitioning from BUGS</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/for-bugs-users.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/gaussian-processes.html b/docs/2_38/stan-users-guide/gaussian-processes.html
index 7b794b927..eec146a84 100644
--- a/docs/2_38/stan-users-guide/gaussian-processes.html
+++ b/docs/2_38/stan-users-guide/gaussian-processes.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/gaussian-processes.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -617,6 +617,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/gaussian-processes.html">Gaussian Processes</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/gaussian-processes.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/hyperspherical-models.html b/docs/2_38/stan-users-guide/hyperspherical-models.html
index e0d154ccf..42f788dbd 100644
--- a/docs/2_38/stan-users-guide/hyperspherical-models.html
+++ b/docs/2_38/stan-users-guide/hyperspherical-models.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/hyperspherical-models.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -610,6 +610,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/hyperspherical-models.html">Directions, Rotations, and Hyperspheres</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/hyperspherical-models.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/index.html b/docs/2_38/stan-users-guide/index.html
index 8f1c647cd..f70af5a3c 100644
--- a/docs/2_38/stan-users-guide/index.html
+++ b/docs/2_38/stan-users-guide/index.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/index.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
diff --git a/docs/2_38/stan-users-guide/latent-discrete.html b/docs/2_38/stan-users-guide/latent-discrete.html
index 148e5785d..9a0fcff74 100644
--- a/docs/2_38/stan-users-guide/latent-discrete.html
+++ b/docs/2_38/stan-users-guide/latent-discrete.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/latent-discrete.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -636,6 +636,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/latent-discrete.html">Latent Discrete Parameters</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/latent-discrete.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/matrices-arrays.html b/docs/2_38/stan-users-guide/matrices-arrays.html
index a68551950..bf685d9f3 100644
--- a/docs/2_38/stan-users-guide/matrices-arrays.html
+++ b/docs/2_38/stan-users-guide/matrices-arrays.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/matrices-arrays.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -602,6 +602,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/matrices-arrays.html">Matrices, Vectors, Arrays, and Tuples</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/matrices-arrays.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/measurement-error.html b/docs/2_38/stan-users-guide/measurement-error.html
index 063fd2045..260bd57e4 100644
--- a/docs/2_38/stan-users-guide/measurement-error.html
+++ b/docs/2_38/stan-users-guide/measurement-error.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/measurement-error.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -611,6 +611,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/measurement-error.html">Measurement Error and Meta-Analysis</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/measurement-error.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/missing-data.html b/docs/2_38/stan-users-guide/missing-data.html
index edfdaf45f..2764f483d 100644
--- a/docs/2_38/stan-users-guide/missing-data.html
+++ b/docs/2_38/stan-users-guide/missing-data.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/missing-data.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -607,6 +607,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/missing-data.html">Missing Data and Partially Known Parameters</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/missing-data.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/multi-indexing.html b/docs/2_38/stan-users-guide/multi-indexing.html
index 70ceb95af..385db967b 100644
--- a/docs/2_38/stan-users-guide/multi-indexing.html
+++ b/docs/2_38/stan-users-guide/multi-indexing.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/multi-indexing.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -603,6 +603,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multi-indexing.html">Multiple Indexing and Range Indexing</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/multi-indexing.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/odes.html b/docs/2_38/stan-users-guide/odes.html
index 6b50a7f03..26fa7bbe5 100644
--- a/docs/2_38/stan-users-guide/odes.html
+++ b/docs/2_38/stan-users-guide/odes.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/odes.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -622,6 +622,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/odes.html">Ordinary Differential Equations</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/odes.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/one-dimensional-integrals.html b/docs/2_38/stan-users-guide/one-dimensional-integrals.html
index 19a3910ee..0398aca3a 100644
--- a/docs/2_38/stan-users-guide/one-dimensional-integrals.html
+++ b/docs/2_38/stan-users-guide/one-dimensional-integrals.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/one-dimensional-integrals.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -612,6 +612,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/one-dimensional-integrals.html">Computing One Dimensional Integrals</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/one-dimensional-integrals.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/parallelization.html b/docs/2_38/stan-users-guide/parallelization.html
index 992015a2e..94f81f806 100644
--- a/docs/2_38/stan-users-guide/parallelization.html
+++ b/docs/2_38/stan-users-guide/parallelization.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/parallelization.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -616,6 +616,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/parallelization.html">Parallelization</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/parallelization.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/posterior-prediction.html b/docs/2_38/stan-users-guide/posterior-prediction.html
index 1c53ea2a6..52a391595 100644
--- a/docs/2_38/stan-users-guide/posterior-prediction.html
+++ b/docs/2_38/stan-users-guide/posterior-prediction.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/posterior-prediction.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -598,6 +598,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Predictive Sampling</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/posterior-prediction.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/posterior-predictive-checks.html b/docs/2_38/stan-users-guide/posterior-predictive-checks.html
index 36e0a391b..aaeb1764a 100644
--- a/docs/2_38/stan-users-guide/posterior-predictive-checks.html
+++ b/docs/2_38/stan-users-guide/posterior-predictive-checks.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/posterior-predictive-checks.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -624,6 +624,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-predictive-checks.html">Posterior and Prior Predictive Checks</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/posterior-predictive-checks.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/poststratification.html b/docs/2_38/stan-users-guide/poststratification.html
index f410ffbe1..e2aea641f 100644
--- a/docs/2_38/stan-users-guide/poststratification.html
+++ b/docs/2_38/stan-users-guide/poststratification.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/poststratification.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -620,6 +620,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/poststratification.html">Poststratification</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/poststratification.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/problematic-posteriors.html b/docs/2_38/stan-users-guide/problematic-posteriors.html
index 19076b1b9..40671ebab 100644
--- a/docs/2_38/stan-users-guide/problematic-posteriors.html
+++ b/docs/2_38/stan-users-guide/problematic-posteriors.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/problematic-posteriors.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -633,6 +633,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/problematic-posteriors.html">Problematic Posteriors</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/problematic-posteriors.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/proportionality-constants.html b/docs/2_38/stan-users-guide/proportionality-constants.html
index af4998be0..409044254 100644
--- a/docs/2_38/stan-users-guide/proportionality-constants.html
+++ b/docs/2_38/stan-users-guide/proportionality-constants.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/proportionality-constants.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -586,6 +586,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/proportionality-constants.html">Proportionality Constants</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/proportionality-constants.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/references.html b/docs/2_38/stan-users-guide/references.html
index 9c8f97f9b..e1059738e 100644
--- a/docs/2_38/stan-users-guide/references.html
+++ b/docs/2_38/stan-users-guide/references.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/references.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -215,6 +215,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/references.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/regression.html b/docs/2_38/stan-users-guide/regression.html
index 226edeb9d..ffc6748ac 100644
--- a/docs/2_38/stan-users-guide/regression.html
+++ b/docs/2_38/stan-users-guide/regression.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/regression.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -659,6 +659,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Regression Models</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/regression.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/reparameterization.html b/docs/2_38/stan-users-guide/reparameterization.html
index ad073e45d..a0dfcd3bb 100644
--- a/docs/2_38/stan-users-guide/reparameterization.html
+++ b/docs/2_38/stan-users-guide/reparameterization.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -618,6 +618,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/reparameterization.html">Reparameterization and Change of Variables</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/simulation-based-calibration.html b/docs/2_38/stan-users-guide/simulation-based-calibration.html
index 2f0477d4d..435be6b27 100644
--- a/docs/2_38/stan-users-guide/simulation-based-calibration.html
+++ b/docs/2_38/stan-users-guide/simulation-based-calibration.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/simulation-based-calibration.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -621,6 +621,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/simulation-based-calibration.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/sparse-ragged.html b/docs/2_38/stan-users-guide/sparse-ragged.html
index 70f1f99ea..567cb0970 100644
--- a/docs/2_38/stan-users-guide/sparse-ragged.html
+++ b/docs/2_38/stan-users-guide/sparse-ragged.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/sparse-ragged.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -604,6 +604,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/sparse-ragged.html">Sparse and Ragged Data Structures</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/sparse-ragged.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/style-guide.html b/docs/2_38/stan-users-guide/style-guide.html
index ca18559fc..572389e67 100644
--- a/docs/2_38/stan-users-guide/style-guide.html
+++ b/docs/2_38/stan-users-guide/style-guide.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/style-guide.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -608,6 +608,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/style-guide.html">Stan Program Style Guide</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/style-guide.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/survival.html b/docs/2_38/stan-users-guide/survival.html
index e7b04bf70..b205ad88c 100644
--- a/docs/2_38/stan-users-guide/survival.html
+++ b/docs/2_38/stan-users-guide/survival.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/survival.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -618,6 +618,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/survival.html">Survival Models</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/survival.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/time-series.html b/docs/2_38/stan-users-guide/time-series.html
index 5a6c93f76..cad04948c 100644
--- a/docs/2_38/stan-users-guide/time-series.html
+++ b/docs/2_38/stan-users-guide/time-series.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/time-series.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -625,6 +625,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/time-series.html">Time-Series Models</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/time-series.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/truncation-censoring.html b/docs/2_38/stan-users-guide/truncation-censoring.html
index 03906c6ae..c0499c9a8 100644
--- a/docs/2_38/stan-users-guide/truncation-censoring.html
+++ b/docs/2_38/stan-users-guide/truncation-censoring.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/truncation-censoring.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -593,6 +593,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/truncation-censoring.html">Truncated or Censored Data</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/truncation-censoring.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/user-functions.html b/docs/2_38/stan-users-guide/user-functions.html
index 769fbe52d..f17f58e15 100644
--- a/docs/2_38/stan-users-guide/user-functions.html
+++ b/docs/2_38/stan-users-guide/user-functions.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/user-functions.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -614,6 +614,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/user-functions.html">User-Defined Functions</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/user-functions.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_38/stan-users-guide/using-stanc.html b/docs/2_38/stan-users-guide/using-stanc.html
index a509cde5f..5ce26b061 100644
--- a/docs/2_38/stan-users-guide/using-stanc.html
+++ b/docs/2_38/stan-users-guide/using-stanc.html
@@ -1,6 +1,6 @@
 <!DOCTYPE html>
 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
-
+<link rel="canonical" href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html" />
 <meta charset="utf-8">
 <meta name="generator" content="quarto-1.6.39">
 
@@ -585,6 +585,10 @@ <h2 id="toc-title">On this page</h2>
 <!-- main -->
 <main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Using the Stan Compiler</a></li></ol></nav></header>
 
+<div>
+This is an old version, <a href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html">view current version</a>.
+</div>
+
 
 
 
diff --git a/docs/2_39/404.html b/docs/2_39/404.html
new file mode 100644
index 000000000..7f4b1223b
--- /dev/null
+++ b/docs/2_39/404.html
@@ -0,0 +1,871 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Page Not Found</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="/docs/site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="/docs/site_libs/quarto-nav/headroom.min.js"></script>
+<script src="/docs/site_libs/clipboard/clipboard.min.js"></script>
+<script src="/docs/site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="/docs/site_libs/quarto-search/fuse.min.js"></script>
+<script src="/docs/site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="/docs/">
+<link href="/docs/./img/logo_tm.png" rel="icon" type="image/png">
+<script src="/docs/site_libs/quarto-html/quarto.js"></script>
+<script src="/docs/site_libs/quarto-html/popper.min.js"></script>
+<script src="/docs/site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="/docs/site_libs/quarto-html/anchor.min.js"></script>
+<link href="/docs/site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="/docs/site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="/docs/site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="/docs/site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="/docs/site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="/docs/site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="/docs/site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="/docs/quarto-config/quarto_styles.css">
+<meta property="og:title" content="Page Not Found – Stan Docs">
+<meta property="og:description" content="">
+<meta property="og:image" content="/docs/img/ulam-fermiac.png">
+<meta property="og:site_name" content="Stan Docs">
+<meta property="og:image:alt" content="Ulam">
+</head>
+
+<body class="nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="/docs/img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="/docs/index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="/docs/stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="/docs/reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="/docs/functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="/docs/cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar zindex-bottom">
+        
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">Page Not Found</h1>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    
+  
+    
+  </div>
+  
+
+
+</header>
+
+
+<p><img style="float:right; margin: 0em 1em 0em 1em" src="/docs/img/ulam-fermiac.png" width="200" alt="Ulam"></p>
+<p>The page you requested cannot be found (perhaps it was moved or renamed).</p>
+<p>You may want to try searching to find the page’s new location.</p>
+<hr>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>        <span class="kw">reject</span>(<span class="st">"This page cannot be found"</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<hr>
+
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/404.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide-2_39.pdf b/docs/2_39/cmdstan-guide-2_39.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..0df8dd16a8729de3f96cc53b779507fc9f724c78
GIT binary patch
literal 551581
zcmcG#Q;;Y@v?bcMZQHhO8>emCwr%^gZQD9++qS!>=iYe}H{wpj%+ITz%*fc4Rh4_^
z%2gYr3L;{(jC8C}q;reIt5EC&3<P$DmQXxAQ1mh;wr0-e1Wb&a1pj$K(TiDFJDWHX
z(2H3cIGc!=7}*({K=JWGIXOF;7}!9$uZJk`$^$dPY&}qSCBU4lOJ*ypQK-}r0zscE
zH~^{=@t&gk@=Ml23<dJ=e60C?6SxBgrWE?!@{|v!xb^XBp~0VFu0CT#2x>4$z+3o>
z`zVHX4w{;AQ5WBob+PN-Jgu=HSH@k%PjhWeX+<jzvC%?G$mKT%&wOZ}_cJukd{}Vj
zKGJlob`&_F0gp!Fg~%8A9n(FS`Qq+KZR45}Wth8cCofmgCGC460|JM22u1I%5-a<c
z6>g()xhlq@_HG(gTrCIMRUPhkFwls*XkXnd^Z*pq$eux&*c$(@ul()(&lO_%FDs?u
zVQ)e}uOV+}X=3CIMX&5)==>iek~RirCQ$V17RJBp$jrh5MK57uVP@|9pM4<<XD0;{
zM`1e~dplbbTW10eD0*Q#Ydc3}djq52`9w@yEsRW*#D#umlr?a&B4A|teWud?RtzHp
z69>nCXy*T}m@6L(XBE{B*J>Bk?2q-AiIWv|8=DkCtl<$rW2g*a2uTP@0s#zQ12D)U
zm=YO6$ns*Pl!^%%FeEBw$w(E;8!A9S6p>^wL9m4Ja@>uNuW2_qDuVceQe~IY6U|vi
zNw?Z-*A?HL2e{b}+Ylgt|8m7MMNWY~(6(HvWp2wXVAFDAo3HzJH0O4y%eSs*vn|bS
zc#n>KN5QZ!A)3HS+7V%rRaOkA8nKlvE%Zwf!yyBJL?KBnk!gDpD=sH4i8ZukPf?O7
z-%Avyf*2<-W@Ix;$nM{tW)n431=iFMVaKRwO1rUXxh>%i)dB$U33pGAq<_YWHMrty
zs)!*$N&F3!mVW6IohyHHHlWmPjO)@CXV$`4afawc2e1d?C5)D}sUVc5rcr(vZ=(^^
z=w`;rk~DlY_Qllfj=MI`e?x=zl?AXz8fM#=5$mWAg}kvat`VhCyMZBgv4&}^PrWoJ
z$QV)kTn4ZwhhfteL*7!5Jx-b1wI(9nxQY0tVyX<c{@ZeGONb?84-T+Kv}J7C^AUXo
zZmNB#|G~>h5H&^DQpNu760NeZ8y6ioo+sT-PYhX8Pd43HZ>D7jncRFl(XpkyoGrs%
z8{jOXw;`7#4JpI#T~`99LBbJpOHJN2!@7;%|F7b_H%D^<T_ph_CxZ?9Ss778i18NR
z1$Q$FI}`$<47!YkZ1kIQI)5wTK%r9NvP`sDzBLF)%%<r@DHfL`c;gchx#jqVs*0xB
zhz~GGh*Be`4S!B=7_6Z|8Q;MpP(yP7(>S^<K_7k7VpmV5+?4239DAtMSchc=lcobU
z;0|%)a7S_W(qJDFsFIuv9kKRB!5=VhK%-7YAAdR;qo;>DK0rrHK5iD+H^5!e#SLq1
z>@$>8zo9-1Ud3fiCq*s*4jLegm<cVc5IX-=i~<ukCgT>vQpj5fvLwptd9F1917)IR
zDyuOtV$~(#G>A5sLCC?*C+7x0SCW+S6P<?kSgNc{`&K}+x$W)Tb&354eeQo%*w$7g
z>6Rd#VHF`;z2II6T>utl$p1j{x#8!$?L9z20vz{NWg7lQ+g66K1K6ejWy)FCkP>2{
z03We;786ZUTYwn?uMsyVj1*!}yc7b;Zoz9d$P_SVrK$Eo^~sT?&FePDlTJea+ISmU
z8h%$)fMvt7x$1g6nFPM2^2R4`+k?IxGStR1F+)5}v{`idM}!PoP!QfSyt+zaiWm)F
zA{J974Re&kpx{Za#UvT^v(m%pV_Ky~Bt^NUKswk8wsT(_PeV~cX6*b;&D@Ty^ZgF{
zyi~jZYhQ3F!0Ja?UOS;^@b|+dev2Qy-9Npa5r{^O`VA5Civl<xug}M%gGW7eFSZ}<
zS35Qafmni`#9b|7k%3)Z7`3FO4~Zp0&pYWd>t1TVZprlxMIIFJ`>{QbQ<y+qzAyei
zP79wmGl*3`-)PheA+b|v)Vc5gL5tSQdbUMrboK@xFCw7qRlR?NV+nMuD@TC?9e*Cn
z*cXJ!Io}_-Az*f)K*Ai8ET;LAK}_Dy<jaXhgom*6e8IcqHfnK0o*+v(%_V7qMaQhX
z`YsI%El5Y@=JRUbHa~BO-oCu1h;QSffz_;$QYS|gY1cG_v0~#Z(_Wjuyxu-s+JOG#
z2p~fSC@B1UR8HU|74h-^di-(J)j0N3zq`dwhJNi|27}H;P3jtY$)X<ky{TDj1+wa2
znnFu!Eqft7fo8<b_(Ea*7hQw!zgUwoK0yPcgMW@1F~tMEvecaSly7Gp8q+`9WCIb%
zb;>trAl9;yb5N0ah*}*bu`IZxD|E<|tG?(KfbS98K{M95+CaoY!}2(a{qX$W09i*k
znP{0BASLmy7&V006`DD-DbgbQEjz=gpPsPtRgRS?M3C<GI>CA0!sH$+Z@23*U4lPf
z(}b(^<YT-7rN_ib2n%IDl~MQ2Up1e87*h8bYtw3%NQgZ{?5!!jpY1RWzHV>5>IkT9
zaX5{%a5r$!toQPCPs{qXByS&X3YhJvkSv;|=zaSj<WG|2hN#~}iEof$8jhz8hIPYi
zMF?m93LK1S*5$<6)sG4E<1ZL;&RC$;;%4ZBkhimEAvssydIkfV^<UI&K%;Y5G0;*5
zK|JK&qQSYm;8{=qV1R6T^#e;@0yT;p7)`-n_4U~_JL6dJ3fOxB%ioQ`01qU1ji&k1
zi{&xnmo#5K&|LWQq@(q3w)VHUFiLL%e@svr{I-jLoRMyZpTW53pwZ5CnaH~u5*P6c
zPTwJoY3^RI?LqwnmW)R{8ST`+qaePp9Q6RtqiQsF0aw8UqhO=Nx`!}$J7<3uG1If^
zbOSC$K`T5rSi-^DO*P;;H*bU=zeGu>IEXLIkkhV0)*Qz<NiW-W;PnO@hL-u_?=B>=
zE(ov=Lld~vUfsTVV5AnCCgvcE*)%@q33dy=%(%q-2Ks7Qz!GdHDiz}AjkD6Dh7a&S
z0X!Q&*?h(Y96z(HbpqCnGea1OW~l&vI-n{Os%GlCb<tP{msA<d+(7!WEYI#1>QE$N
zOA4@?VncL0U)##yDE&8)g6)cRm{pcG3Z2IwbXLC$*o{C0atQ-e2PUH!lrKrhu~(P;
zD1gSi*)$X7^rdNGHW|%O8jSq;jlR&wBBP2^aW7dif)>XhJdK0^dF`Y1pV-|`1omfl
zB+6nmtQ$!U!}C|92$DbvzyLZ?F{Mgl+O{E#ZVX`CGM9jhkRYB@adpy!T<f~aX$i2&
zcBZA`abr-n1m-Xk_UMWHfJv9lvo%2~d(*~C4j(l)?dMYZjTvuPw={y*f6-#vKpKz&
zEbtyvMy<KYU}H5$rkuzf7QvZc@O7C+jGv(wre$aRuZ|^GXcmm81_gp%zdf{T^yYr3
zWfv5bV=}Hcap|RirXGF@G<;tVoR00_6==H(TEwYr(~=knFbP|ncW5cV*w<zNhi5pq
zY-+xB`$s^C$>BMvj+R1?*9>~<#1>?idYmAk!ZKn1vnSO9EAiR1J1f9rznC)M3taL<
zu(u|601veDK>j9b=!8~fc8-53Y8vumLAF3VVL(GUrPc`ds4E*05`Ygr5`-;@hfj0#
zN$A8_90g1u4-76urT$+Uel|sFWpV~d@fy969FR9CNRUP0O@b9!Vk7494|02K!Myvj
zi^oUJb;syat+P_D0u{*Qk!46nSg;n>d>R1Hc2W17Tz>g!`~`35Ee&BL=rwVh>&2Q{
z3M<q=e2q@|(j!5B)Hft~m>`{p%BaUVNiEwVZY&E&%FeEu7q3@W(jV2ps~F=hoirHh
zzhVpko)eX6r+e}iAu(IZt@+7YDhKDhIC;X!UN#V5ulskid*A{quUDv^_bfIX1J`A;
z<%a0W_Kv74SZ^?phB0PjmL~7soCh5amHpdqgQ_Fvl3U;`l_6ED(ngQl9WjVnTsw0^
zY51!uTVxyh`A@{JfOT1mxj1E_!M5eZ3(#v&iL-Qv6+T_8aHSJ2nFMU`@bCeTo!WxF
zp3J&xg0EMZqUYgxiD6XzrC3qcK><+Mfq>rEKJV>rp_4E73Z&`UL<>*`=5T)q5~XEt
zRw*rchJ3Gc&UWr*ID%?h(QQ-ias#Z90|6oLPplp%A1+V+*qL$<Or89RX7~2#mKP2z
z@iM`Jd9Az!3vkAI36uNBtM+b-o1yBRCL!OYqjSv7pU=_)b~46-v8ofLFEEtpy}<bF
zl&2)Re_!mu58No(a2*~;70l$@`+8{c3OK*`XidJFxm~KoRTTeTdFUvre)FIx5wAd?
zOkUT9^Gv#FkiD_S`_LY@IR2u{swt1Ut*6=I(d!*Dlp1H?AauFaIQ(hdmE|N3Tt9IU
z6J&zQ^aQM&kBc;dlFQ)r61d;RyZr<#EnBM4t_DF3U&yq(K(~FyTzm&nId!62LEclN
z#uh_sdzXvvUM5xwj$K2@?+yF#*kNvV^Wp3ToS{6h+-)$EYahbi9MTwR;#sk-1_B)Z
zFg$byibl^Fa5Pft(C}IRsd{)Xs!=oQ@Qr)*5#bKE@sY)hj^=3s<j}2S0zAQm(ZKES
z9ad&O{Ba2-PBxPkswJ<ZbCk2FZ<-*(4O%usKxxm`FP=@0BR74*$Eso0H2#pLs|H9u
z$J^V!aXUKj2=Y20M>&cOxvVznyy1^Ji~Ojt7L)>+M@`LvhD3rkeuse!W&9QX);w8-
zy_mH&z3QlT9+`Icrt==qT4|hleZ@`D*_|o#SI?3-%aVjUft-DY@}SC!V3u7*^J-4@
zEbheHH+31uG;b<hRb-@~f}WGEdbi4|E?*@!2S~Qk82Upd)9&VA@);KE&0dU<y*wcA
zA6h1HClTR;N%Ft#0{U6E*yCwRL2-H4MlfG3xDGJv^TSojRSn{6*FCo>@gL;pqF!tI
zge!9l^AjXGr6AQ5N1l#@vSG6ibvi`H><ak!L#5M~KK{1wsA~i7p36vJCn*uM+Z*#@
z^V1++JBoW2JPj?A!~(ghH(^2bSNUwymFsR@$pM7hpcx0>`nAS2qRLG<8ZB|(4myqv
zewWbX^;IbTM4!2wj3TNyp+M5325BfLNJP0QJD#DQa>Uqcyi<#civ;{|*0GDWz=rXd
z+SiY3y`pAXRMKj2wq<=G{KLe4%R&XJEhnI;NE=6LxJ&CpLJW=<kA-EOt5Z$NIJl?|
zKY`sKdimE4mZLEhWLx?%LSd&Z7)Z7;8M!Hr?gcZS=DZWC@=ZO&Ue}L>0Ok)E7~`oQ
zJif<?p%diR7%8h^P%zHDeMfe1ShX=3&N}bfm7svy+2*daB0RL#*MhWBw_md1?QXsp
z)u=39GKlSucSA(s<U~6TlhIt<-D<hJX(oHY-1sL_=9-&g74xB&oeg6h5xehK&|WH+
z+p@4<s;J6U-0L7CF)ns_%F}FP0We}q?7QYZ!GIAs$u(|<;Xo?rI%W!nnWg<4e0g!z
z*hR{#U7OYZ+Icb&7t;?m)8nU)V0OB<k{u&HnbFe?n~v?2N?oNr8>?!Kyy&j54(Gmm
zbT(P1(_X)us?%>MkPkSBll{l*^%UEIo>nPtEv6H;LkZByXV<(udz*2J&EpVo!tw<s
zuV+7?-(Z{VSF&tRIh1A^w{as78;)zRy{0naP?_YHKuYU%l)P|z5m;D>_0B``{fc(?
zhnEMXncI&GhDA@g)T^v+<5i?is+;;WPvYfQ)L+I}P~&6ZOw@w6CHdiTKcplR>A2Ln
z#jbm-da&2kDTp12ipz0Yee>&<oDGN2%SoWczPUYt39F+Yv;^@w6fTD>>g&en1O}Xi
z3}s9oveRW|pzPl5-32}YxHaf0+}FV#{r)zn(^0q(vPzUEODD`3blx75K4jCD$8}**
zLc!mHgM&RmYu%n_B5iu#((%`ghAbO2@SvfD9md<$K3)RSwUXY9d0swYBzXAKWemh0
zd2agMx&mBw;iqTPM|Ban_`TrykE1M`Sm1jJZ{qFB@$qrp-AFeVG0zz!a}(>7!;cy7
zEuC!1qO+yP%!ZIkjsS0#h<ei=zYW>gU0H{6j-~p-q{vk>`YF>!jlYPRyvU*Fa$8}s
znsvi3`hVW{r;+nu_d2|KxiCOwC>^Op=)du4qJV3~ZXJQEn1~wEgSHLv1qprIy*3n3
zkW9omc*wm4G&8NrVwDr~|LNR9w#Mf4%3(JR6zS)WNniK-^FF3a%Ycw!?;4@(tM8KX
zcmTLn-3q<Kzj<4{Bje%e&;8c7Gnf}C<#CnM&44`a^d~|YDbU2Lgg@Npjpzd8F=ZIO
z7YUHup8Ut^vPy7Vb2W_2V1~-I!X#dW=*~&Q%s8hHt-7XmmK;PB84m{YCoLS6zE{`Z
zK<v!HlFvbohv2SjcTTcFv+|0W0^<2PKCT<{#e2VHZv)$a?;j?kz3n-J7QG37yn=u#
z>6;H&NZ*LACEX*mCWmo5oL4-Fzz<EKxt-f7cpxdSLsxmzaX}h+U(?pXpm0%c%+Bk`
zu}>YDg*<+kq?QOuG1+X49)Qzx_7A7yyMJ+XLg(#jq=ae+97)J`hLRlAcRdw6*^~|E
z-C24w1iZ*7KuVC(-T)Kj&jCMfBiP7A4CV>>XHqE5f<de#P4qlxJW_Q<vLoT?_?@&t
z%by%0PlIJR3FMI33{f~OFe?&TGH=g+S{jurC+VTVjZQM!7Y4S0CW*w9^Hh@j+6?Xy
zV6NwtYDTwuL~30v^+;mm0(2Bk2=HCHlt+X`iWcB_cvzssPiLM@@gW+i6f86Jq_Lm=
zL(;s!$0dV2ZyGrryGt%53<3f6w)H_=ds5}_m@#WM9z+a~q2thbxALAYv}||~`ip<Z
z67c`Ve-m~!H@?;HDeI^3)Toa7Gxy5S*vF&OB#kzP4z{v?pNyR`Na9_9$%oQ<7)i6r
zB;xb$np@o^I=15s|5}JV2!Y_Phxb|e-z5xFlze;*9l9d3SaGxTU&{#Kz3+iKd&Ari
zD=~VAxZxN~h&5NtxSTN_o9GG?ebGsHB*P37g_oKcSVzWx3lNG-i|%k7dj9$nLL0yc
zxqU|(0^{3t7IzsCHBo5^V|K4U{)ILMv7RU}LHW@t+>FQwQ@#U0&X0iDNWqO?7|n=A
zeIEN5KxUKmpZ2tx8588GYs%hF4z;WPksgwwVK~%tLy3P3Md}u%KMlv3#-!bV=ARp9
z->LSYq7uEI&4$eO0T1KNYqOcM)sdj%9Gyo7lcu(px&<4k#E;DYokuKt0EOERIP!9k
z+KP{m%Zt@~0NCc<RW@bl`RQ_IL2F;*SJ>}^uln1xpX`YvBVps~JJJ&Ng9uxBSZ@gM
zRRI6lH$j!1XoR{i4c&;3A94qhs|8sr)|03d?sW=z!nw`w@7|{gx1sQ@q5_3nJHY0{
zU;1eMDOM%wlGTX21gWmFYA;1c22J~QQv78xq>^41_qOxxltAoWZpTrEP<BYBr>lL>
zuULtP5qR4dhMgETtPfg<9X;O2<vuC``{i#&jj*5%PCx$$EfoqgZsrqAIO)>nPUoPL
z3e1m?9+@3Qyw~s5JS*$<GQVPEP{D^G>F(f4Rg?YjDDU)m=-}0<hXxlZ{cWVX#$10f
zZ}Z5Om(hiH4n#igeU01WhLak!*ay8S8(Kr@L1wxoEKELN+C5ZOsFW|@nvL=|+w+BI
z#J};r;<NAmego7mLL(rukp1=db$6jW&C)|DviqipK-BrlDN1oMx-=H(ofUV|g!FrL
zIs(U9_n<uYNV<Q4$N+O@q7^DQXBrsYm=7_2(F_3xh~1mb7x`LMK3i=!8DArI-st-n
z9gNhB7qg}KW>;agAq*$J37fqltib5xQ-Jgn;D%E8;uoJ=LILI*z@_IIL+85xuFf|Y
zav!Ja4+#js?cGST$cNGoR<N+Y&QT+W^{R(yJKOp6={t<TM^A|5DEKppG9wf6pr!=2
z6jd@q{KeefsGo48q7P@qA|yq+UwoLPe#nc|0u9@BE+D{N8Bs+{z)kSPNrP~P_&t$+
zSX~E+YYkRBZ`K5d!f|EW7+gauA_gp`==vifB5Lg_y!jk9@<wO6{0Ks`I&6q-he+$4
z8gjXAz*AEXhD65Mlb7hHHn;Z>^uOJB3tpFJOm*NxnWJ>W!Bf==RVZv@SL^*09MyXB
zjl)r|F8STiZ9zK&iswN28HW9r3!IlpyPG7w{036Tyw~pY#L4%W7^24<2<dN?*GKl}
zNLmR*XL3gQf&wwp!mSYEkA!&o2(;i;3!1p&DB<B#R;+H~(K{At+E`s6k`jicg3S)t
z`QAfmQpsSY)W$#2^AhECpMayJ6{(WdPwwm%lH*|Wh|ZJ@GB=}whv`Ovr6=n9bYi`2
zjE<Zjweh@|<ve}$cfH%?3FM~T*5JF!+1|?a(Zdn;Ne{q7Yv@wn?9S5+)Y@Xl70`oZ
zf^TQGPfD$QA7LHDA%3^&c$XeRqK$(ot${kpmEa`6B@$scvxz&IXUb+$AW+l*Ee(y`
z;1PJI<7()A?}#nJJlXRvg%NheII5wbBf5SIfPmcxoD<xrH$wIb?)&PqVqV^Tfl_X5
zKLIz486jZp2+J2XP}+z$=^!7azoM2nyC)VVxBVOcv03kjcBj0Ka!pi^KnQqBBYtg(
znVkJ3qMw+*?JC1+KFI3$W3dl|76Pu(gZ}GJ^jUY%9{dHUpGP&I8@WEa<%tpE*VF{f
zYcy~!aCqnfGN7%KN~;$h)aLi!WBhg>B=<SX^I#Hq+&zvsNyM_8+*e?Jn@=#x2!t2c
z#oHnwa~$2f9rx&Ng=jXrivIyL5D@89)1fbxb5#+{F^!%lukC8j;Ex>#-|Gv5wDy`E
z`{^|r4tGYTJSv`%FQsduKh6s}82$3+OI!P&_0red_&eASXiW0kLDfho!a)lUwb?I!
zJ*;23ZA1!NHa=z#-qxj{(NH-8+ysEZ%s~k#t`R6E8z?%4HU9^Y?iCvrSc!>UW8b}q
zl_EYK9Sb)g7&FuZ_i+(gRr}zOqg&wow4{(NnTGg>*S}}iaj4+rpARBiHstn~D0H-w
zn4sBj0Yiam+2p6#aJssi^yUGTas;mP6E}2OcHDrWH!6ehb<ty_P}d06gXMvID0o%W
zSLn%?I)Vf|+IbFT;QFsiqUUi}m>DDzvIC1`a9;wBm|(C1dcP#G@CS>#5}llZ&0z0s
z7VFvp2&TE3qdQhkkT9(u02}_$01WdmOgQxd-{)&vw&H6$vYG{}!xJEs|4$LJlY=cy
zkt63ba8(_G8T5Ub#OKE2%R3yL337lsv9jy+`<|3`tDL_9FBlXTMqvw#8WyUKksM+y
z`eDk3m6@Gnr$a$AIlTT!Hj<8tSD8OWm;d-1hju(UX>{aoScXooxZ4XL6yFbyd~V!4
zM*R~Y2!ppGB)>L8NmMS65RTVBC(CjVTqE>YVb(0l({~Yo9OgY%lJxMD^1K*apmVM=
zC$N((1(U{?!*~vz*m1}?`rC65l)R7O1RyX#dm;cQYalEAkH0Wj_ZS9mIhTlE&v7Pr
zpYxt>3LvkdJ7)S8K=)7^kCLrX=TD6%0Cy$aK8&5HGT6BdKLgQ&c|4XltmAk)>Ch+Q
z*HK3no<CudAPEWqr9Ni5D3CHO)D19pq(Du&j?|dL%k&T186=J=c^SG=h&)1(#jq*T
zoG@=67xC923pifijEP2-Pn2!1`aYN`%6^|<tE~l`q+KccjMFh7MlPT5twDp6A)U(&
zqw)&_{6ScLRJO9v1z-M8$3zKuos%|j?2B4;vp<!0#PhQ`ZY-|zls`>FK_(8@c=y|e
z#}yOg)Ci{=44O&Ou5;r+YkE{Yah)5fe*g;7z-oCKkAF@m+vecLK6S+{r`Qz_3_pzP
zLhA%smgQK#JkL2W*EoLdyu?sRjzd}qWH*YpzAIt{$ZifUwUWr`)8rrapN*;yfPMjT
zWT@9)+ZkA7`=R*+aWo4~^<g`0Z0uq7<F5TvsLTkx(_b%xQQAQ9I@rZ)KgM*_M&nA<
z_C&njg0za9`xUrEc%a@!I)1zWStk_e9feQEYw~=isUb!M@7WEo3Z2DYOvw34dkrz}
ztkTcU;6Z}S#P7^VnntJ4GqsZoZCyTS=YVbmD?$h$a!f!H@BSh0<Imv05qCU$n|9H(
z2hnfrpe|OSjXKcBl5ACM{#}9Qc+WC>Y<Dug;7nJ#(NPV9@mGieVkLv{HWlpPVYRkb
zU~1<N{HBUDqW!Z<C8Q$s7`%Wq!FM1t_qWp}3TpTp^XEO!*1W`M0pnKXf*J%&J(x7W
zVMrnM7wS9@H0MKz|I`j1@Nd*691v&9wy|HJQpi1BFH%n&r|~ChQldQNYZq$b)2~}_
z=O?&J7ZQFKr3``|qi#^%1bRB4N+KU-!qkL+MEKX&Xpq4_=%;Jq1*l=U_m2h8$QG)=
z&@=(HHz6ORnz%gX@C!K^eJM{kms?(y_R<mQQ>g|N_)RUuf3K`bc5dHnzc-qvGyT^y
zIKFVWf-S`IbJJ@o9*mZ15EsvFyjgnl0@D^*F8W&H9{ZW&0Uj?-+7~PKH;^~kRUrf;
zK0b~r03KZKLDCeO(>W1>`M&jGz(Y&F-<<$FyO_9VQ{WVGzZq?c*z04Vw=hwqOFmfb
z;pTl;KO==q5Nb?!1U=PAH5~lmc~|}58*Yk*M02X_2f5bz<1n7Rr|<*r->}IkfFu2j
zQaIEgj4TC)<3L1v)!))Rh<3*ah_&adLI0(&RRPdAIIkc=%SrQQNkTT6ajQYQ6I-`*
z8d@_qZ=ZMh_sTk(9Xfz*qmuF0cQ0L0Yt?XMNOzpI;lFm`#H_eQJAV*6t{Q8@vP+`N
zr07I%X338PyUizpw#Fd;SraN?JWtredX#0V;g^11zu5nIeffsWY{`>h+iiWo1#XOp
zUkZ|=rK1kTKP9`=QxUtTIUKUkqj`DGPNP>kAXLjn2v%hJ!}M#hLZ9iOm-$8=W@?C)
z1<g76{NA)c8$WMYU#H9!HYPuI2PX#!id;xmEK+^&B9_w)OL1$;->F94wolG3$8Nvg
zwrM^rsPR02(>&6-Jwx{45l1aX+=7zdP>s(`$b$+>%j_$~0;AeBc2Jd`el)60T>dCq
zVe#-tzon~v&*BK*$z3Ieb>6saIt*!J1{qr6F8v}WP)@iP2rr?4@BnM#x-J`C)Yf6G
zozaTN7fGn&4*km6-R#=?;l3jlz+P(|W~Q)sr537^h>&j<2p=!x8%cG924CrpE~5CV
z-!smZSA5Nuc>H~ol|!&lM>^%qzEfN1s`7V~UgY3&FVC#&d&^=12CubLut|V-MLf$d
zJK#uw$zw`**k@AGH)!>M=$)J;A;b5Ri=vb*AMnJVU&#o=%tTQV5AH`Q3~ft`(%MmN
zVNn#(*btwhM?K3YWjM(f!b>--i$zTb-_{U5=hHtq0=4~M?=3v!m}Z-Q^I}*g7nGZ^
zQ67U-hIV)BGS&obo_&Pmx$Jcg@Y!Rpj@aV&&)}o1oGJc;b_#ZNrF{Pe^AVzR)e8Q3
z#P)2XC-XQ~DyWZX%25!H|BKg(?L8cObvUl2agZM&T@wdUPEB_ZTg`bhbY$*R(zY-Q
zbuJbmI*7H*;9k{ztljZG4G(jY0X5e@-xo7+-=UT*0)X1_w4>!?ASt4Rb|h%_84sby
z@vve_9{~uh?q7~Mzc^k2zhDD<xxqX3*e~~0CwTpvGh&*4Q@-u4hB(eC5a%uB+e1b(
z_{&zXugqAI4w;aOM{>YB(W)qT`tWmxfYN}CMyi&tjreN^ES?!df3;%6CB>9=R>%~%
z!X*H%E7T1_9>KGOdCZ9^QV;6qs6)}&iAg{7nKz%c=EA#&wP`c=21?8#6Z{d@4a;9`
zleqnFfhq4$Nl9Yt#O%#4>toZSB`eRyP#sFRAG(85cd`T!3#jr2t)I)O__}6n=UX2?
zPL#i;%`J5adp8Q+v8~!r#ufXr8(qXK3mMA%7b89?`GtQ&v_`0*OJNr0zL`y7hRoz+
zv>Di>HLc_B%2>IK->!q_lajp0Q9Jg*#q5n6ufPVc`RHL9v<aN2NL*1_>{(rAuRFSL
z!p+=e!TQYbr_9|7DtOucR<CD?N3GbwmAc-ouzp-1JOubSm4%F4#lY9QtyyexLO$Mi
zg1I>YWD)-gxp(=#UnU62@wWtG;h~lxht3GG9#MMz#C<#0x{cEqKby#0iNP!0^BKlT
z3LHE|fHrs2YT=n{048`9yjcN(vjNh=XInu`HSL=P=AFSCBv4Gly9gO4tf2eH@F>9E
z%@QPvf&U0c&mh$dvjT_EqD_z`V2y=#<Di<4iQv*<Q*5pU)ZuUd>gjhaqg4+Q(JR=)
ztSI!tFw-oW-72q<5mEt!G|Ttxrx^s<ysx8D*g(Fu`Q{UFNXm;D<Atc+hj4oR)kAXx
ziI*3<h!83dX1tHeAZ^iSEe0jPckHKO!pt=2*AMJK{SuiQg;7jWpg~5!*p^}!FLH!<
zucP8{M1Kzw5~-T!poPBx6{t-b2L+>!spB9d=9|Sy44NQ3CGD>xl?n^==RioB>_G^S
zhIX91!$K;w2=R%K(N*v#0~#}Nnv)03h^IiTCdN<%;WduW6tS8^NASTXq=P>Knv<2A
zA}vx$CrE`bMbQY%X9dz-gcv?uH;g%jmXNk8wu@ri^$e0@6p{{PO~*;D>LXyw1Jf_3
zT$MIc7)}Ktg#S$^ai#~;6|LJS^^c975nvS%3tw2c<NS-^-<k=u#-b$7$VMR^6EMNn
z9WE3GBll1FN9HexNz9$5Gm$B(i8lOVd^nCfKt8aP$*&Spfji?@7VVp2-zf<ZL9T`D
zJ2(pf<cr6JFn)t(;+3*(4+3$bz|A&VfOkRN60O7xVxS2OyC{K<>vc;aj?%yq{#d+%
zeI_8L+J=l-1Q8hdWuS&W;Gj>K2+liw`9ByKx4Y23zmv#As3*PhXe4oz1veRcyT0J<
z_5=jH!0WG>37qWI;f#GU0=VIwLL#;=YebWWQ-h6N*Au#L+@k(^6vJV`q8mSgKi&sX
z0lE!zQe7PJwCzI+Ad`%kC94A7&0feL+j{*FtS)i$zWH$P`uw`aKIirBgfzmUAWrQ;
z1>3F!ZZN7Xb5g@#$p^CF{~?CWLO<g!?t!FB_`5&=zU*Y6z?|NKwHAU7?*KV00gmE|
zRoh7MMa&h(k1-27nI{K;m}j4ZZm#qWj-XC#`0nO90)B)M?%}!s2ZQ)8C{T@>G|T=o
z(Ce6020)FOE()-$l+Nt?u*Lq^^7WU5x?~;P01f5?MuZ2b2?MuiS>-{>fQEkJ8f0Nn
zvU9MrG9MM^Kb7O}TL$9}436yu+xm{%Vh8XAu*VNrYq0>Ga)Sb!{MrCadXNxk5s!;x
zO7-modlmQ5&;POuw3UWfeFQ8K>N5f<%NlH29`}5H9th^80T?;a9y!qvyMFI`*a>+8
z1_tgp27P|t%kk*{8iNM7{+|>)c!rBL4#Y5UeE5-{YXvD$_>-v(2eOGB_5%*X?bjP&
z`|-8L?@AQ~_IB_Llx>t=U=g@JBGMu#f$3>lmnkz&zK1m#E<@jqQ{44b9oHR~0tK?U
zN5r)s@a=0|kk20o*w^{%uMjuUIJlRv#=|TVTGsa?>l;Y#&5ML#&c1zqd-3Vt`Gf49
zA637@KG)mm#jy&Hj(UsvOQU)CMl%)qdUD9wH_W~hj9^peXprKME*J&K=ua;ZVIT}~
zgz&xk?e!JPkWb;s0dB#1S+w<ANPCFoZf8?XCuY504(`qCCW!o7fsAg@kij;0`2WIt
zg81iq2mWKzT0%>bH+rB!)QXZ2l`>K_4ov+L51(m{R#fDtP4x%#|A0`=8o6u%DW^;l
z8O4%lQ%DL&XeJetKX{zlnN!46v;F}Fu=5I4q-v={OOY-^i7$V%7D6l_HDdNrbjL|E
zfn2C_;+^-L494MdWw&~=Gd#ctfB+_DihD1&DRR}$;P93q%TktMtD_@|eY4_=V4pL`
zMgE&435Hcdoa9WRBR$B9c5Ii<&*VD&sXgUKz!x{t7y$mUG7g4SLtMl>e!68d<wyIV
zH}fgKn+^D1vh@D5OA`76|7QZK<-cb;{zvNJ|Cbp0|7W_RxTAr`|ITn^W@2FYe@#8T
z+VZeZe!?BSyD=q%Boy+8C=fH|h&fU)$PdI2DMKj%HK@ag#Ca-IMXBR}b63E?E$AO0
zK$t|7<Phl##0f%bC)5`)BoZPJAVP>00FvyOG<|tP1=-rVe8TB9d*S4}>3wxNVKsPp
z`^w!B3`P(di{9FDo5E%`o1LD}YJ3yJZfroC&EoWM?roXRPf*-IWM)6it8VJl=mYUO
zl7GuOj+G5a4M!5AnV@cjlo1N7rm-6z<2FFg4=eRkwKQvR%HF#BzR5pTy{uC$%Z?VF
zX`nsG8h9{`P<B&is8PN4ZOj-~{HX32Q#m_dW{HxN&61sQVnB65o0en4KL;nJfytR=
zK(<Lb8wP?_Piwqk1ht%cT-X!tRy|`GjGO7OsWCwuOJSeu`wMHPtB+WDe?T^(a+Pvn
zh&Y|HvEg1}L7oA|^%Gg3P_0bQ7^A{VKf>UoqPdb0>S&C;AI0X|o@7C@BDOU<Jv8OS
z0DHV1pKrGG?!;RD{&3kJ$<bY*^{<v0_C#XkBzlE8Mdm=Yx0rt^S*SEf?&5M<c9SdM
ztY}jNw_eVTVe2rx6n#l<SVlc(Wi+kdI^c}Pz=n6avNV+$SB+I-qhj7FU2f7M-(Wec
zmTBwRNZY%yqVaGS)TV+g&zQ@u8I2qv?{^mU`|xAmx@RWh!_P8R<tp5X48iqn9cg+{
zs|*9Gs{+hr7}Mg*SqtH2ML=ea$tDId_xw!EnlDL2Gu(Y+u*<10MAIA^24Pnl0TInK
zh9b2O!$XxKR|a5qa-~|U8;)t1t%i3*1KF-2nC(&t9y`|F0hpaw?tvN4Syt0PxKOWv
zO`jj5(o+TZ7iPD5#-zssT?6B*WBt<dzb301H_gn*E={888rn+?$}Su{wcUv|j7iP>
zd@xnA)ahpv5vn<?-$2X;+DqHn#<T{d))^B?3?CViUh1mLGK~#P2Fa0Cb_@^4?mDc?
z{ecm>YfCd!_1ibbrPs+^b{5E%>Ejz^6Aomf>1SEBKo?MTOl+y#7mWhl3Dw*;<$~l2
zc4`A|UWQYs@!OVJsDtX$@aE+H4=Xz5n;P0HfeX5ntt;T9eUCIz$QMcqo@pZ?&3OQl
zH&ohA`Z3i&##^pV?yK)e(LJ&XQ|}lEUn&b?XI=wP1n<=1dtO6F1dlrsCbbZ^Gzj^Q
zq95k7Knx0NP#R_d^!E2|?xgyE2qegjYiTC{|J0F<Zz)FOSLf>{y={M>oCc;5M!B%;
z3LT!4nLSV+YYVlAkNmabTNPp#VY12wwk+T#^B7>jHaEl~OlU@2HV8j&8%i?b2hM2Y
zOK8b}@T(z0Cb4<g1YoH5Aq;2^f&Nnn4vwTY-~&f@=R=msiuDkLRDtRfFpHxvX_2e3
zVyPiHPIiy^Y@#4}VdGF^5}&w?s#w(vp1cSa<uDzVwAh2r96YQ|dlsaL7{$WWN76l9
zpj#^>>8VPnHwFWcm>nA8I31Lb4JFh~-Y1jVrB_2D1N-;iz-|pX0ZDCCe;gi2{3aKe
zlP(B;y9t<6Hw1pWF5JL69MAm)%IXvH>6o2|{)*%2SS=BO75FYixfQ{RN1fuoe*!hb
zzl}Vg#&2j%U=2r*Cl*h+1iemII>q(Xts4x}MD@eO1{$Me>tdD_mrm-Vm49W>UU41K
zuZGoLQOhPd&q}Yau#2vzCMmzL%fj7LwS+gFScP;4HEUkudc-KQ&#bmzKmT^jEa(L&
z{nlJjcLkVXPtjL*Mf$f<#Z`koZhBKWxxI=-*jp&#kkm^l_3Eq%g2-#}?8FO!$dmKx
zR1*r8XWg#ha0^!dqfX2o2wtniql+H^UYq;uoC5*>XL<mYp`xjO!3CYi{VAvHLel;A
zJiCnAc4;?t#gGV_$NlJjb3H~=$k$^EWkeuI?`pHgO)y9gzFqf2KsW_(MZ)~{!lm^B
z#r&3f>xv5<>r-prRTl=<w|d0JtD<TAssjw`bN$-&3+T7GaovJ~_4PS$-GYSm#dYnx
z4g%|ia&XV3!gcBNh5j9K>O2+-(gSt3`70H+S+kK}F!69fW`1i$e!ZqxLcdi*UC>_}
zW4j{rI~23=MEv{AjjMMwmRDA6b%)A-BCQuibGs`FICsTTyjwLfzrrneokk9SfqSvM
zs>p|q?lzHVI5HJttaC3SrFZ))rE_aO_WCP@<X2E{cT-k)PnGxjscU^4<OhP(E!svU
zQo3S&n+)~yN9q97bDS<ZUfSWv2?olT>{eoa*PG5xoyXjPYHOpJT;!FdZ-W($$lEet
zl5KJ9&kN5qF~i>cT}RET<xt7&MaMTN8BgY+(;lac`=QsvjB36;#;}&RPCZ(#Flf(<
zP!~Ar(e#~*4+7PM^pi{T{$HixcEhjDUKboXDe5P;+BKmz3Y%vBmFr*I!T)O4y7X5}
zWnR7!E!z<g(+v5(Im&^-G?9I{iOeEKU8d^`xo=yLP(%gPU<9wMf(_8|>JyU;0&Z7A
zwLd%whdk}T<GTOgBA_l25|60`Y}mbaTGNe#yOhe>BW|S+s@oSZe}jcNj%*EcY+Qmq
zk4$1VZegULX2Tk*_Oe%}0!zrA3DGZ6Hy|maEYQJT1d{Ze87a`YgP<<Uj<0&urjVp(
zZKA%8P5RR=8hnAlA!$<qq1K>Z8L&v17HX(t!B$-Sgs)h&2tvI}a%4$3gIWt)+^k7L
zO=`5#m8>+>vHuNA=|i5GqLfeC<GZhW2GBP290a0v4ntr6l&~L}B|GG*fkf>RnBL|o
z;Z%Z4fV?0WQMCv)TW5GqwS?+sA8wImtR~f`*;!x)rX}q*+@yBOWL&32b8H~2vZWA9
z9TTW?JCe}l3!+;mr)}I(x~oa8(Jg6P>5tx4Yuu=W6H5=9{w4QT({@T>??+5(K=0@e
zMH{J;#BhF})yPVD!HV&21x9%VPVlZ~sCwzijyCi@)9OrmBpYnBRf-sQ4{)a26B1>`
ze0lviv7Gq&2tvb)E!U+5B;#h)UZ2>ElHKdf7@J4kJWzL$7NBOfuZbmM_k?o?TPF0a
zgJ!1IfCEBhB>LphsdNBUr`AmH?ZWis3Rgg1K2OouDM#S)KDSwgpQM%H39J+=%Bo%&
zQPvVW67M;j$4q@uVp?oEE!%-Hvz@zGokWRmWaX8#E^5b!Us=gtuD40I5Ca~n#9?XO
zAj&YH0VN+1QtIfFGN3w1gk0YvxK}y+QflCnG@yx`boq)lqU?@+$OL*J0p(Us=ALr;
z&K%MUZY^f!!Y7#UH>m#v>av@Xi8KW><7uKBr5p*y8`?*aNVLZDhn4DL1`K6>oAEez
zsEQ)eo*3r!UE4i}NN+`|sDAxD3C+^BPOtW$m(a$Tb+NF`WxhBg6x&qc!ZV{uC5w4P
zzx$KGqSA9Ry6@cu$f5?ZBQMM$nv`}~=R`U#YCT>oEB}BokYQ9Rv%1@s{(@sAsVV!(
zNB4WY<6n{FLGzoOUlsW^G45mZj2xq*Oazf`a-F1nmDe@V*5jCF_upK60Z%H+Y`Pwp
zOXc%~8O!vVw;+Sy*=rKq)GpnPs-4n!xr>^;BM0%1_*gHEBJQL}yI#xsCVVyv{RTXf
z4|Dr$d$z0RunWD3iF>Nm0y@uR*`?L{>%uZyKZZ$fgZ}Bp5H4E>nZ=`4!P%>IO_RFT
zm0J9Zo9Ay`Fj&t9bz&@fp3>QywN&V<%yhlA%lgv_pJ{$`O6nTJG%v;93Zb0uz~NpO
z(tj1IKlkg=Z*qc*u<B7I^dHHYD)0+eje?7L&tkV10%O=pX4cK$#2sf+j4iYxUfU}V
zM(Y1I5ZBwhZ3M0I;GOPEvTB%_V{>ohn)G?kI>TB+eX0YTR`5TkLGau^N-rvMmb~nE
z`f8?0x>s)3d<?wHcez-O;ePUr_Byn8fUk6!Z9ngNguwpx&(#jzt5c)M!gcYu7grz5
z)~Iv^g4KhMV+_Zv*4zgIfAl!W27Sj~&os-;wBN-=Yet+fk8x;r`q+z9+vuY+;lq1`
zYHU>Gw5Gy5QgI>5k5ovmn_iQPs#t$iwgC@uUDy73-QQhN>rf+41eC^loJW-FH693g
zGSZfQ#b>@=Ao|#wDjTaCe$K5wDJ(PyvUixR^1Tc8j`GlVHvLvTD`pK@n{|r{(Sz7x
zi|cK19%V%AnYsULzJa?~hA>UUb4==d_VOIKOwqfVrsZ(#dYme`XD%U+MLE3QYB?X2
z9K^rK+yoGVAPh$MXa55pAtc%R-~X7HIsZ$6*#9b}<6!%LFQtn~omARjKnZ=}2}zq_
zP;}|FE}$&|rd<eC-3SR6-6yPPjuoPp*BkCY2xzOKnZ?PuNqfvA?zdFzKfi<sy&1ne
z-4YL>CIJwOR%_gdyp5CsO-EX)D)(J%gcSO_%udamVF-5dz3eCdF_SrCOWtdmrwgY>
zna^?jfl`-=1cyM#H@M=(yxDr(8dqG5Bmf!yoDdB_SDeKFRu~VBf0!8>Pgx)k9DI^Q
z8HliKHnQHL?asSXjaOx{2`>(`Y@BCR3qstN0RymWszi3a;Bv6f&HvD~q1CNrvGJ@c
zNIU>#FH$IxDg-A&GzMro2K%>26vOdt>k@~7=tlc$fq;VQ?B1TKkN_xhH$3x2w+K3N
z^H?hO51I%GLL?7Ta2~2Nw@iH1NAnN2)m7P+tOfn1g?<H5Ci*-fnRt>vpKj)xE*DED
z%h_XKqDSv`6E$=4X*S@HMe!sN@6e(1xuY;?P6le#my+Ae{D#TSMw#KdIt(FXez+)R
z1qJcYA(O^t@PU!Mqo@7WOXZtVt0+oxpR5EZ(B$!hAZHZ#@)H~F?d4=MzRo$fvNIQ;
z1jz@XQiZq(r+#qgfD#|d=)n7}crAOqs?Kwa=3Buv7QcI~&G^4?D1mSZ^FLxA`7p9d
zvT-AlQA=?#F07AHg0)dKBLY&nyM=9&HLwiX*iksiuReNpPi^xD=4Lr;AFUE6{t$Oi
zydX!98nZXkBhE2r$JJ|Sx<deE@HQ<v9E23)#SJm4oOB~FJ)xmQu!!u(wR1uWa$d?A
zrTZCXneVqQaBW^ma$Ro7awgG{E;vDsIr8VL(wwJd<2PQMlJD{m5r0tFWfl93JB%4<
z-tarEv=mc!ELh%uq~L2TGE_fq+DA%A2xzvjF;dWIC={4(NLkuH9a|f&7U@QHSvw@5
zKepCQPR+^5MdP3e?x;^Ok{caHwqJpBvaVVHdb_yj|6sdUZ9<G5c>(iN+GSq;EG?~^
zi4wLtw|Z0Q>e!~w;Z}DpHloXT4|s35Yy~wgR0bPPSxz;u%bdDh=I>PJOqZhDnC%OX
zWbUcW4&FfD&=qG#tOadN@QMrgtN7ww<{#L&E(c}$$iKAU$C=e@ccvLYN@MgX9IifM
z_za0NS!V&ohYq*eupJCup7N8><?#RaP0Pr@_}_OeBO~X3mRS<~-|H-6)F&M`*ig1U
zsN0g!ga8mg!mq8D<Z>eR<x1{K&x|pkQ|I=CkBTHep8!L`K;EQgMhaO*UvL1xPyXy1
zC+!}7Elr=FZ{ZS>BuF?;BrQ25-jf*N7^IR`y4=4Qv!Yr%`aImZaus3b@@%fnz73Ww
zU6n04zR#Wyc5<iUsgFTOIaoqGgElP_pF(lZ7^f-)kPwegc^@ZgYlQwSm-(B`1M_8j
zN;|Er{Bo6im{i)tkMd!cRiNau1L~+Oug{d2BttY6LSDv6V1F#gJb*CE<9GggW3_cr
zxJW;tq&BU^`K&vSmU6DT(X9p?UI0>*+(;lMWXMB-Hkh5>8|k^Fl3I;dQCrbGi>+GO
za+>wL(h`!0qKGv^GF*oQ9hqV*H}0){y54ZsTri)<SPHfKGrdfvLDEh+qwGZogP0Af
zTt%3G5z)$m)UP1-fE=@FWnD30oxY@UzMiwImcl(oO({6V^M}kiKV5-iU}L#s!A7{`
zmoZb^z||VZDSpi@Z!cMaX0oG)<3A!mCp<3FDb3yWpN`We7nRSaMT{Hvq-)2TC|*s$
zSzopmK*#x^q|kkoE{7;LGPq6IEOwbsf^lKm);MQ^o1iqwbA1EHxkeK2$M7t=%+1;1
z8n9QekOjhp6-ED>OvXb|ekag;_KkQrD)KRpZ7_MXD;GFNssGd$0aMmikilgt#<wf8
zwLDFCaw`8P>;aeJmfrS`0j04&{?xJh^l>lXl2Ld!^L7qYan0cHG^I|@E?fCw11gzD
zVFZrvA8^dz;NT~mcHQ?vUbj44YTNOLzuSc<nlKfVnPC+M4h0!{%x_8IdHvLW#X*Lx
zEkl816}t)zIqBH9IcU5{trukf+Wl=gI?o}L`yp%<zc(W%3@F(!8Uk8-4%5!<<D4I0
zPY26-4)#3_3ugMu7O__6UH_v;{XEh77$nWA*5@+Vq+DuCkd3gkji3O!5k`v1P$xVj
zI!Wf0#B24b;~zb`POjq_B-5=qh#1DG?w?>Dt-t%bVaIRr(h3g)D3Vy#cA<$ss-21#
z^m+hV$SK+Q29pJWXnj*a=BouXz_Sq8pOV68Y}*hy-W0i+wS>b_Prbq`dALzBY(}0L
z(*}l2_~!Et8bbKUC$@6F9C*Ih#~3nGX2f)PRo-E*b`)H^AH_XSxo!1IBZM#nXqW@w
zCpiYL`RQETC18^M&Qb|ugFcfWr>?eLqis%|O>>)_U$ccfv1jUG&@BpIEe;QZjd}Dn
zZk;V}HI)PGms$H^=YinY7iLQp?~suSmRyk;UO{6xaG-&~(d=aK>c`Dv9?Eh2)}hav
zytpTP%G(Q$?jw%*>C3HmagAA>FOzy?#wZDc2ri*>0fs_1Ps6LYrreEM(oj_!v&bz*
z{f*h<(L5Ll$O>V=!(5-2y)17k6;VQ$o&Pv}sRliqpx}~Lr)CcLaI%y<mHFhRD+@5M
zo_uSg?iY>Al8&Bk;$HHR9=I@p5(d`>Wzk9R<_KO!2ANoeNPJ2T9KRU??};}cS3%KG
zlK$P&-IXp)T_0(53GNFpMzpS<=|iz_F+*go-r{&X(p~Y}sjKq?My&4${2$<tjrsoo
ze+(Rq|F2Hi5_MUpbqR#-5A{=s6hodJe6J`?`_;&Ub=Gl{#xO>RLMlSV#MPJ|A1!kh
zi3$IR7ZO2PmtNn8cNb{!4G8fUcDKo_4NkKJK`|vvT6V;wgk2;9y22x_cdz76?r*cV
z(W`8Pf1IKr0xHbm4m?db)O?<PFAi+E3IXDA567H7pH105T6{~0wsq1T-0V%r7XLys
z6+)U4WCa4}-3^1WwnjB|R;k8YWXJg7*(wvMA|{XBdMt9Isv@J>sxtMmh;~@*T0Ewb
zDk4{=KZJW?7fLDq>JSz&a4r{<aHSN7^Zx^$L`dVxM2IcE)fjgdB4I6BbL-qFb0BCd
z4Ht4i59(wNT5XEaMKLNA_fP~IpblhM=Hs!8eT-nkVFV>b2(Qs~-#7m@e<{};h7~l?
z6`6jW^$-9ZL&y^cnVgfj1r!}pW|?lR#aOs1=vm@{KI#LqsG=jcfQM~pGM3Hj;qbw!
z_$q)hW&~20n3UYi5QIB+{LEdoNljC2uvA(yDRE%8IW2`SlVI9Ip44}_o+DuXQOy{S
zhqFAl1Z9z8pyP6pO82RS>f~b8$?Qz?N#Nf&A-Q77o7R!Gf$o&suX6mcyD#+6^okTx
zQ3w)D7%5unk4=IyU}`ge=RYj2O-1(U%yy1<FHGnc-p?T39v?^k*)v%0;_2mJpLX-o
zCLvW>YwYoo_w_e~JGSg%G{+fO!pmfcMR93EivmT7V?6vHl)YncXWh3g9NX^L`Ng(v
zqhs6X*tYGYW2<A^wr$(V+vlEh-?~-*r|Ld;f7-RbthK84oO7%-#~5p~Dwpa!!KYdd
zW=AAGzSvnpVx}gdz{NTYsyrnDZ`aS$+!vipL<by5@y1}^<aXVy4=My1bc>nFTCF(S
z!5&2IFs8`$T8KCg&cJ!kKeSzrB-}y|Y(v^6e$lU@FtO|HC?n3#Ma~4J5*cx6Rayex
zuxsL&v|&&hhznF9sXG8yEY)ux^q|ozc(&lM*-c^@W+181VfN|C;g0~$wa|fN8m<UD
z5!^r(dcw?v_m41Y69s%1Zap9RT86MJx~cHLt7RKDnvWZPeomLw8zVhwdvtUtC$w(E
z_EN0l3-8nW-XxM-uG!?_$XNKWHt7sP2}|f=YKhaiEa5l8w2xoYic}L!9YBEy!8f$~
zJg*T>(E;9+c)~bIT?CaK<U;rb!FNHw!vp#<k@*py2~9(G>9x6ROOR8;+rCT=P=hW{
zFs+3fr5-pV*TfBjVa935w}IFBq^mP&5A&&QC#|}pvB<n53Z-3IO{1OrIH^Qcymnz_
zM{A4YcAa9Yy)v!f5b;-#fT9;d-C*5S1K=ak(VKU2PZ`+}nW1TU0&Aa|fL-PT2Q?&o
znt*9{Bj~r3NX=;Hv16T|b=LUnDj?<N>4JW*{;oD?p~YJ#tL`0MUCTkQ&qErAod(ls
z{&MTnNAO-&{R!o+N2-C2-L;wjfDYRc^DNp?fAb2~J(3Bp-Vev=osYHnPHvssa5#zh
z3oIVyKQqyPnU4I26A2>&!@mY%*{O<71#C#e+xL`iL9LqcU?`<_m=seRlY{9)wz_M4
z05OMj>>Gm{(&tm^8IUF_f^<F&Bl?f8wZ*T-p{=F<t-na@?aV_E>J|(o%hB%hEO^5R
zS)9Xgi|g_0W?2yTHP^0d6P$Uh_Pd24_kQ^B&9+q++_$P*m;U$713Zn8fecHYXKx*R
zYo9=eE|dof_j#Zhi&*&1Q9S*0Z<K!2=k$9g)tq$!LdKZrgjf8Zn)#Xx#Q+ZJ2~bkO
zUYXhzy~TPU{|d7Pbz^Suiuy4VUO-HI7^8)191;mJ#31(NN{ZTmqEN7|NCuBz_|jMz
zmVc(>@<HD3IQJ4AdXxAJQ4Mg?TV>VOqyacG2bLnR-!x=sNjwt-M-a|&uRbM$$3G}D
zJ+?XSd%*jIqJHfa3MYvo`hT_5H1NC=ombKnyAcVe4U@!jyCa$05>FZtKjeZ6dTW^A
zDO|xZQR}LvnyOka98^2&^h=^6nUv|*o_ng6N-g!gZH@e%I;*VdG$5es5vJh-vMZqB
zfPAqbZx(e2a;rY;9{+_h@!&5hW@~LY)t*S&HcAXLNlvIH(BQ4ZOE=9`q1B!8={`PR
zji+;g6h7}Gz2Q6@w*E;z<iq^9XhPov!EyGAYSgCumwx1LPFg&JWRl+wLGl3h%Z=Xh
z72=$H+T`&nKy2+mf&)Al0K4|QXl_$hS){Re{r(kZ5}Y0MyHXaF3ykdT28j=fctzD)
zD$IwNvxaX_;j?6&eEYlm&Pm6IEdv*N^Ry?-La^j?=4Sj*9+H=z(pN%(WlJ{ozXtU5
z)v-B{>!XXUy6@2TLyjPpDbYtV=?$r`xSljO<R3hxW*P8;fmFnpYANe-sF3RNHMRm9
zOpC4@>10JUsq>ijigQlxoUA;*EoEzLB{mapwNl}#3fW6>s&_h4nitow^gzu{z0`Sq
zGANb9j3`J{zHqS(tIuN;=tgT?%g7v?0?oSK90rH>j^jv20aHxL0|Wb9N@|4X#xv1A
z8j8o45W4_(VphaZ1A5X@1Nz&&|BQ_!Bd+oeFgCPa?h#x!@9Wm+18T1_wpvcHh{m5w
zue1_?Rm2kep4H#X3?Qj4S#>%l-O1`EfS6XYYb*~bGx<7I<`*g+OSgpQFJo|)4}F_M
zTJkDh?R3~Gr{cC;=Mm~d^lThGKp<y(mW)L<*?$z>G$InwSm|Yfhq&8B%Zz!lVt|Ey
z%1wC=)g%0?EokqPp8tf~E@L^j?kYG_9F%5B3Dr42*Al8Vh$%B^RIMhW|NSOJP?aWD
zU2mZ$5;cGSjUg`;n@|PXt~Bq$PB6`9V&?1eHr@WX*nPjscm59v<o`$?|J|?t>&~P6
z%LayCj__|D`LCUvqtjo@9p=BLJOA70h>e}?-vXUZEiF5|W@I1!?toYuf|jh^IAIV$
zzm*VeP+Dwt9|`K0UPT0@#<G}v>G+R#pN2v&#q1&@8c+$%OZ=;gW`a<seO$+0QOr>?
zaYbzhm57R%bRCzhTLTA0d(O5fJ(QB)F$vU~2uXP0#NsVN0jN>?0rwI9SpC2x;B)qr
znEonoy@Vlo86%8z!2u4V_!5vqBt}ZZFY(xTM&?cYNckGta4jm4z0tttB>Q^cVjY5T
z9L)spi8qOdgf9T24*iZb!ogS&k=p06mblv=K*mww0_hc~lB<Hng=2D~;UGX)@gV-#
zIxsC7gJcELiPBNM89E7m?+ZmiB^C~p`yv(6;D3^Ll)#a9$k$;pYpv+Dd&)Z*T1J&d
ze<VdU)32I<cm4KLb08oD>rgH83p!TinbXB;F5RJpx193=GAkdb2GQ)TTm*x1GoN&0
zAukkSNdpJIkQ$hmZCd$dqAlTK*4w|5)SHvi@86>c-xA-WYf>wr5RTn&Dl&))0)Z1m
zW8DyAwsQ$h(U{<k7JUUKi^OQtOC7)C5k?cLY5`I@#Hx*0w&L5u3-}eNR!Ao7dGVzH
z9FxZ;5|myNq-WS4|NE5S#jk0XUj|$kDuL39OQCPR6l_GwrD~12ms9Z3UXLmJFd1Y6
z_AyY!lUO`%szMCboU|WmbI4mAZD3I$z$GHi@#T!VIjjBf^>lHRc6vi}pxom$rJ<sI
z`Rcsp_SV4taeTc%tIPX-|F)KgS7#fq=Gr!Ut)>6&&dsx}yOn3laIy;X%j0ABY%+0v
z`igi5L}lK<MUTM#9sa6c`j5f1o!qlso#Umk_rzP0k2|NH?q)}8m$%ik_SarY$?<r@
zzHbd(EPXD&!#J#~tCR1=&X(}k)%ADnVd72O@-zSE!`sa%@{B7T{(~>i-&V*`R8)iW
z+A_|~?A_*e*Z1;mz4yZ&zq-mb-|N%SFL$2~C(NCr)pRXgJm0Pc#+}nLPm1Tn1HpIv
zI?PA*+2_f}M`%a7iIyMumbHLJVOvO>&^d&2=E>SBNJ~iDQ1akRG+U(>$=&~cF-%0s
zOj;=%ASYv^tdn#X8G!ky&D!?$ZO-{VJbT*?Jsdqf3f-T48{2)nJ3bt|dk9!U-G{o+
z8#n+}3nzUtotM5br6Em}r@?xR_ADInZTUFPY<3;a<ahrnkX^0@0NkC})I+=cKzA>w
zm*xBX5Vqad@PsNym-Qh?-{C+>iQJcHiJ!pYBVucn>f+2Al8ORiVHA3gUnG&5TdvQ3
zU0OgyWqFSsHITFbvBTJo0YO|N3YQN{0TFFkD+JjNA^&jjMN{+=LqaJC-39}k42Qw4
z5`Z^M7ysZG1_7sfG7u44BY;o?r=hk83Q)f(;)@v>X|GYGo7F=54Z;A0SNAK~+^Ooy
zt^q0f4JuJm)0-?(YA_UyUGbtOo92v+kOyH^i(|aZBj08)3{?g~m|)S6K5ZH(bedTA
z9%}fD4OB5hoW}wg_ZO(wfZvR9kWpVp&o7vlopRkw_{0fNtoKZK4plv&%$NwcAW&SI
zX#2uVOxQ!1VOs;{3rHqhEfxt7Hq3D_Hlzy)9aPyC4V`3C@Ks}BavF?6`L3QCrdvy{
zxQ&+tF^(ra8VPG2N5UQd8g&OjX$ZjHRU0t6Y6XbIjdl7D6_xrAkBjDR%1}qmMWUA=
zV=bG(LfOfU?HVD*<)}ZTx$&{W^pqu&=@JCzMTf>FSxw2?3C%=B=vS-MbGq&$#@okE
z<3H-1C<4)Jxt$9HVx5{byJ=h_X?w>mX`RL^{|GLoXjpg>*<AmIZ*7{8TEASoW42NH
zBS|SMWuYXMsFaa3SC~Xz#6ns0-=8cKl`@mo3Kz(W*eDw%%Veb%d7Zi-O-N{rnECTF
zlmoDh2_TOsWEuFCDyB_im=oj!MRoh3hQ6vbYfl1Z<Kr)Tuwk`bu;%KzT~Xlqu<>r<
z;Q|5RTb3ikUriDyeHPR+L_bskdB#Bu!hCmM96??Te1`tXY}z+!!hFAk9DY|&;wA?t
zf!I+xtP9KNlMv?7sbP<V{a~OD^6R#J5929dQ$SV_@Avn@8d3z&PK~mc0U-3qYqdfo
zs<gJLqvgZF>&uiP7+0uk*sRy{8xM>@U5fpZDMf@LUtb_0#%WA|ubx-nB}apzNn+tE
z_k(QY#mWuwy6Dy>P7$Z93Nl&8K+c!@ahHz@Lw>RPls6Gpsw#stg7^w}KQ2%wDEk9g
z7UB!JmKEkZ3o4gfR^CxR$ACFTahp%O;-}6UX;?!(R%`b^KTcp@ouO1kJ>k>ZsKJN8
zv+%g(jIoy~&0u&fud)QjUoteqyJIY{1SxB}k3^D?k*>4ci;IO04j5Cfa&I(j5{KW;
z4E;OpjM6g7C_wM<8U%@#ZQiA9I8(kQ$gl;veZ}V*hD8&H1p6w0xtu%<a<QB(Ez5Y6
zX;1JFIp&8d>H(y;c+b=1<Y5RA!8<)hngb1rj&#jxQ;|jo+-uxs;~!#~W>KM=KgREG
zYS7i0Qkt$5!n<mfZeW}GR=;cehFmBs<P>J3CJ3u0!SRb1EJS>Gg&2uy#DEO6p@fWx
z%(%sp{{G1Yh6#;>)@q4cso^7SK7Dx{4+gBK(y*eOd~|3Uu0xor5tiG1*Pdq!)Z%FG
z2)<q~9Z$A?GqDS?8?lI+Ae}k_Z9J!Uh2x2I@@*XphedSs&hBO7(mJ|Vj!LPS_*3m0
zD&!Af>fQh&A(G#~=TVyH!hwdo!4`)^nwL1<YTjXeL8`Q2Q+I9df@*QV48|}PK}JHU
zBDU+dGjcQ+Cn7e?ChbUqs3!t(<tcm4)D&u6IE5@M1+s#~j<(}uZTYCU&I+J;KcZpO
zh!S!hu!+<Nu~AWu7sK8nUbC=~XM^H^4B{=0)(G?+I#`#Hj%{SK6-a~%sWP@LENLXT
zARRh_*WN(|OEqbx&oi){?kMk|M!v`g*2D%hEtBKb4GafdGRB?;F1imf167HZX!+5p
z6IR|RySomX`9X}o)ytQfp|v8+d(KG8*A=NRbIqBE#p+W`TUL~8bSt$R@tEeKXf4EM
z084~Gr&118Q4&8u(r45H7x<GJnNS>cVuPDKJpr=jB>n-KR*FK^nDFQmeR;E#lOyg1
z9H{e}wJ286h-b>UyJ>OveU2E7wtrDl$B~A~-$)!!sqBR2Ahut~&<YM}8V~*B3!1@d
zO==ZF50#{)tXp>IDtucV4MrzqT_r$SNq{hfd+y??W_kgQ{KF5TFs<7}<jMfbI3NS!
zX(@Ude0;GBg|I6II(=qNh;Q^Q7#&ObJtXLu>4x@=19<wQ75YV-@>zWJB@_?$DI0Oj
zqPYvc`1`!o>}{_2ec^s@<9=@oTd<Qoz{3gb@e1nf5_#dB<b)`<29fu*lal|T`eSDO
zj++_R!^r0(`!&D$D0C~6z{APs<nAPYw0pdOuEtl(?|m7(#aGYo9TVc{;85+QLG7Tb
z!wWw13Go8_3!vG6@j=W?Se1M$2%|}Ee6WPsEhY<PS_DBL_PPX#e)<jU1?oQa4-l92
z|MZ~z9}vsP!Nl}$L2R<RWa<JNQrC3N4g7N;MuLsIr0op|BHckRDW?DezSd?{XzEak
z*=UWtr0U&VI<TXK6|HJ=*IW4l4+jK~o9E!^``O#sVWtoSu~B@Sq`f!P9*e|Xz$h_n
z_4$s|%dQ%Hv*zaISC`NCoh%9@IQ1N$Q5t+7Ic(P-%5Zc}2rOJiBcp!`!$mjup=Dj1
z3HH1zLv!)Ry0kFA=AIe$V(q$_;!D&}Y+^M!9q$8ZOsDO+eDQ^@0R3S;=d9Oav(4(_
zyO=8s=Gj!~u7cX1o(B$VS3lsNYFbl;a3Mwv&HP=RYbzl55-1pEjB5sn#giwEJNJAs
z^Y$!7vjwxNs~7r&u<KoFd7}=;#hLZoa)*j_o4Tg0%|ORf{c^?zzOW%N9D6dFvUs8O
zMU)IkhnZb=ViluBO%4MsGzwaOIO?l8&&3QN<ztL5296utxFPWiXlru$heO(aE*g(z
zibc1KXGsYgDk`9D#hn<WhN42(G$vM4&GYPB0LNN&VjV6Xze@Qk*3w8S`nD;p#2w^!
z{`5Q!XZ<SG|Mo+5UCXLf-FpB=I(ti;jb>Z3K+rM6@v6v%k}rRg%KQM{2P=VpBM_~?
zs};aqW0Z@odp`Z8dpgbx!6)whKAJ_r%w)G>sRQRgilY5gE!;~@C~OBx#JxkvV)$-5
zGXz*GIpwyJ_z6SXHkWw^LDSsF*8fzj1u4Z1mAAnkkjEjdfv>p1wE(i`hMRH%d6AZX
zF;vV_KEww?aTMPP%Dr=Agi}%Auya*dfqIGUfQU7>3q_P%NcVifw&vON=BNflLJLy@
z{a5RpbRs~*h#?vC6RH^$3EN?<Ak7opF5G;8`1fh-lz(lha(R<x>nr{+P{tIeOQkio
zHDTbY#ij<$42ymg%k-g8;*aLsu<!s7m;Fi!saU`Tp;A7hQ%<u&p9N$v<^s`n@7>V~
zsI}bX;ZxHuV9<9QZ|0=_ez`CXFxv=GU>Vu<ZrFDYI(39%?I6mu%hS}FG^2fUJKX1w
z-f;dFXM9Ou0y}lJ0efxy3(+a!5n4JN?{H%GI!j<jnCc@9Byo^A%t0L+BC307z194#
zemk8Els(2N<r-UFRf+vMrQlCfeGQEtYaBBR=}X&I_Vkjfe{~%-{5!6uU80L<0EQ#*
zNXI@wexb9u&ztk5p3WVbNS}j{Y0pu2H<fJcja{f6Zfmc<J9;;zpsq!jUCR+l67Zpn
z8uycHsZA|i2V6r><K5t#ES0FQ*odX)*lZc|V~MarS=C2dMc#_N{sd#($137vag(N|
z9O<4s5YL|~*>)AF667?yWi`6EvmoSSXVe29VZ6W7Vpj@ym=M3eYIk+_Fw=6!QBVRc
zD0LVyej%j2X<c}CM0ErFS@aav(v7a4A)U^2_-?aGhM0}@?M00OwS!pL&(zotYw^7J
zS=+QZ8mxzWz~O5c9w>K(B2`S)kpe1JMMs($yun>RgYQSQcc!#DLxxRkP6PQ(>)cQt
zQ?wGv3Q?fSp!&#&EI+Nx#${D_pwU(85myRF;+WIP58A|%3o1DD2TX|c#rMQ$v`_~$
zLBKfN?4Gm5jqBwE<;R11KeKT5Xg$M;&4>$ByiUA9nhxj1o>otfq$Nt!!J7|Zn|$~X
zA+5KD2xEnB`b#|%0$ou;yQt~{khJP3&Vxp{az?p`>UHnamHAR><SA+ClXt0dtXhJL
z`QMj;^r#zvjDsa&`4?F(60mIGF!_JuX>p=qS{DCEOc!9~$7Lhyacofs?~N3PiBjg&
z#Xx*>V`5HiD!)Xw2cfHHKL$SNOl#hmoMOM&wa+h0qDe?1`h%W@CXL`+7yMU)U@1-$
z2RkhTj&uS?)Q)ahCAL}9;{1zM0HurBszU#CBMlFfl?@v*u#u~xI>flV%84#7PT&AW
zzk)cQM&^+~O+k;h#GDIdPxu{XjQI_lr|sTEtRlJwL&Y_+Cc3LZpE!R1VEVSFEW;v8
z!#7iPKG#pKi;9kPBPI+bWz0OAgpjV*xPRmZI>vBhKBSm^?p)s25Y)Ddb~2WDN)6mo
zih`}9bHyTyvcy&2g9cl4AkSQ@(~zL(DMvdtg8d8s&D1MtfA=l>A^IBQ^s5RHG4t#A
z$5%Lx+ga<^9X#&G2?4UP2Lh{dRuyfm6Y|$CjI;e}1K!6<sKiu%^VN08HD#LwCW@5w
zj0^3Se~;AjvHf?d*u~det@5yB|9eAmaK$oz)Er7Fy^XoyFh8ixdN;{ZtSJ@2d={DF
zN3Pkx-3D_xm-E|t;fZAUcZop3*}40yD?$Ft)r7Fj>kavM9Ho9l1UR)Ie6yt&TmX|v
zfZGz}fszVW;PYdxiVe_YD+s%j+D+C?xtBu0(|f9~mV)J_XeG?29^Dqq&6U(H)ALP&
zKTl*HY9?#9j@~D1uM{}OKa^Xx|FU2H-x`SjDYu-=|07WN7u<j%we8qU4kVxL?to?z
zrTnH|h%1Nrb*T~?YUfko#jQY8TtmDxDB-owYrBuYpcz>Y)nv$NxTlxQbh{xzLA<=3
z%S~_h=QkqR0Ae~U1Ho2U6pWdqK|qn92DqM3eFTb(60Q$N#1~+&d6EBc+jS&2cL|8X
zA?H|7q@II5jr%Pbsci+mJ8dqHy)Bi$3mghOU9g9-Z@{((SlGe8-fxd*$O1Wu_Qe~!
zMN^b?vOm_4!Xv+_uT<kDrXrpl^C#W!%=V!lAQKPb+rH+<otIRz5~|$C@IN0%taiI3
zyTz~FyZhV$R3-OSs@#{|_r(sw?b-Ys=AcN1^mn3BVMCmS+k0`ANS`rw!7=$jrBLIY
zt(-*44}`oq$o{5I2Kg;sext|zW00Ze0P6hg;C7SsvZ{qA5frsOY9WzPL6@<<pdebQ
zcnnDyXa@#a6=>SN-aO_$PHD0EY0I7M?2dB}ij^cRL6+a>Fn0#5WwveDZJIq4`+@m4
znYd~k8{EO3rLb9?G0<bYQ{f%R`6G)UYM#l+g>4gfX>Q7{ow!d`Pi^H{_9rY!Vj1Su
zxOi>V{`Hd<SRyM5HGZtIlK#H!f+?Z~q;&Rd*Ret9yZxn2#Yt;r$1bg##3cx$lyCvT
zuF@j=mrtzJS+7Z0^JMQ)CWe~7qD|2T;mEG0MmbHY5yX(T-962N*gak9!gZ|oki9bU
zK^OICDl;eMZJRbj!Ij2CI}@>6Pu;)0;4MK|CqIE4ct-O&W3<`4=CPEFSGJ5MVw_gX
zVjGmLmU!*7*ll|_>b{c|4XjITnAJ*ZY%8;-ZC}LL9Nm2CkfEX@qqy9X7li5A3uMC%
zPtGBuZY#eSiZ1G4i>urFo#@X$Yt2LQkCV;azprJ*UYmiJ4*pW;>5)av-4O+``qjjc
z@@uk2OW+IH_W&7J`eI?sNfiM>0nO4;o<!aaj#VaALFz6Pce4rSCrWISJFrL*&4vzo
z^NZQ+t-U*F6w7lmvGb9P^7&o3a@WI8yZvwT^W&mihriSc#aq#(aAH188F$#mj?6YL
zgls)qL1<=t_|9*9;1mM*J43BNFF?SG*jBf8J@7Vtp9H%dtguOPWpEPRUggCHu)zF%
znN;k=8?M&+&87AN%LeF0`*tRWN+hmwdA>;3*t{0@zQN)d-JWjcvD~=B#Pc@jd)at%
z>Q7(Nr_8Qm);=46>hOy`Nk|_ek!8;H4mW-EGE)ivDdai+GvparnExAK|BH}Uooq$u
zV?Y9%+0v^sQF3eGbFpcbAPs<lb`1le3^fPDed`TK3H;8m*n;Z?0So-ii>Tr3U9S%<
zN+X=pODaSUfeQ&oCxpMdq>X<)XF0x*wdToZH)Isnt5PsX*bbBdPIhIQCgt9h60I}d
z#t@DEJ=MO&Es-`Lg%^&`$z<M+anI}GyO*{I`HO|R_aD~B{tqeb|F`@97wcnIkod`<
zV1LZiBM1s)1OVUthcW)L2mT{%GIB8d>!bE;b&c3{HYA_vnwf&Afhf1r5v@!%3)nvZ
z6wLQHwBV)R%R;}bX;)*uz0EiZs~_<JAB2u>@W(G*`2{<4VF9r<zXzX(Co7r>n2>LR
zpsQ7DSHYPOR5BQ|$$E>Sx`MK*zP3KtDxtb6-3vuty6E_x-5WGv@q)YAbaH>HZl&4;
zHf;MSFpa2VL}E2-i_3^<)TRWOWVvi!Od1T%&&)u)&M0c6C0d(JE$|<w#)y3KU6(mq
z#Im$JawQdlI%m09H--fE=T3B5MAWwy^fvu?;+P@7?Bw*jV*pM<%xcnR7)38z*J;~Y
zkXtTJs|r;!ksHe1xH_@w>d*0BO=njQ&SwNzkKsq%*xKsQX=D`JyNv}kFa9KYFvN&-
zgP8>E2;$b!NP&<UViXatEY$AROl2qq_C{$z^A|hYx4?!xQJl->@7b@jiFW;BuI`?0
z-oE2**2rd`sg@1o^HU8}yva*3F2EE>n?J_S$_z5~_N1xE5utE8kvfMw@f)4Fu&1nR
zK~*OSS8kgp-PvQP(Y!N7$?1(E9T?X{2A}U9PC%eDZ>d$(&#<xJ5>p&==7F9M2O&PB
zNj}vMls8>pV!hgR*=U61x-lhV0E&Ftm{rAi4&zWx(`gy63vAbuL;)VARyWb<8sVJC
zk}#u(iVOxfq52U^mi<xj9F3L`MHURh9c5U%yfo9(M7uFEHMaiN3e~4dF#eUgPg{k<
z6AXF~`QAkR{$#H)fVJUNWddQY;PS`ijLV$E;!DDBgw(KX?8Ji=;M1e5jNj$^iH#Fy
zW8<x-=EtgLt&BP7y{@+ItxGIKniZNG?EpGWF=72ZNfnsonn}wHW#(dPr8qk^<kyNq
zpj09wExWvehZqJMgdBW&Uw?MAD-brRQQTO2LzT2?N=urk<hIfUu`vr$aY<Cd<a*lU
z+(VLb5_RWqRjhhp<Omclx+AwiYykX=G{^#@)va5@0(}iPS#yRQD4a<W8l~t6ES!TY
z4^nEFAt0Uz!{(Yyx6k{wmK7Usi8w+7{>gnFNRvNm0IP3Nlz4(WRrT<MaGYlagmVl4
z;aJ!%%Mprd;}Gd`3U;Yqfqbc-1r;EF8x`voa{NS%e^s_m7)^iwoA4BEPU4PrPH_~j
z;PBPn-@3zZ1PxX4LIE9gbuK|ZNhCO?9v3lJAo~`Cr2jL=_x^xQVULy7AAfvZ*arwt
zkUJ)8f`kU>qZdmJG`MC4YyO0W+c!KM6)BhK>jWYAbgX>{pPZ?O`K<{apHNA>g^*KI
zEePT-6CihE0o42ZuXkFr?U#*3WXKe)<9m!xf4_|G&%f02@dzjMH4hSaUqA#3eXVXE
zh9$lk%~KpWdvLHqejeUD+8kJy<dI=1Z*uUX)t+hygc4FlCJIJ9z7nli&^CN&AL=Mt
z?Z8E;jgCMi+c%+<n!Fj~4F_snMosf_=)IF)gYs5pux5d3p0cp9S2VF=`j55UOJ@+I
z!+6^fi-`R3Rm|Y&S>`KJ<{jx_Yu`Ve?r(CMpp|8><*pTtDB<$20No{cIVsVSC=`Nn
zz{aud!%|v^7#@~B?w)||ES<Xpk>L4}|FEzYr(wxd^Lu@-m+H<pj&WkBSl}F;16N{1
zH3CsTUVDDMw`42i{8>GvxQcd)w*<GzGQrq72}Io^c*+gcy>nRs){NQ<$XA5!6n>y%
z1E%om1VeFM;}<Txj1w`I1x_XBaq!(pa1y9u)OE5>Ub;#i8IPGEZ&8&q+(I)i7EbGm
zF8jk}rIcE|a|4gnMr#?Jb2S(iF*QLe{S?D1n92=8I^a*xA+MCySzMz)>eT+9PY&v^
zSr{)1aCXL;$wVYVC2(gersX5D)G(zKQp+ye8Jx5AV$1tJD_I5>cvS@S=%>eeSU+2g
zOzCKJKtfplklVrnm7%tY&(;fFFP0-kR<!Sz`zXx;M8etg5oHrLz(I+W&SB$KGdy0h
z3`GQJA+fE;#e;&}D{9vU>Q&SoSn1l@oG!~oLy&7r>W}!kn`VU0Gd}~(PGyGGNux(`
zP#_3HCViIu)Wj-?Yux^kpR;hd{p6xwFBP7)?C6Wak8lda93PFf8vi@Da-E}LL?&PR
zk#CuYP*yrLMJmUq*45PYPR`YzA=n5`tWhdFm9pJ-u-HZAm_g$2+kR2I?hWUYyE5PY
z`i{G$nGpKk7al0l3z3c<J^omV{e#CzgvxGArgVV>0bx>QXa05uA47H&2h}{S{ZWz~
zdtA~#U%JO2yMteImY<WWAPjU`Ny^@1$+H>ui39R_Bocn9<+*v(aykVQru=M80i*JN
zx=HaFD7x8N@`7q$$48#KS)tK4MERPAB#uQ#ll@Rcz?S8yo&K<4kLB*!#j^K`4nu7X
z{r*_|zCT*@&5T1H`iJt$`48oliGk^Vga-eDv8=xCxXyv%{rvO9pd_Najt~UQZMDkX
zK5N{VI<e4jaYBVmON#|uawA^?QE>Hk+YWRSi9Y2JjX+`pJNjzg8yn7SU4z5Z_14n4
z&WpN0W2jzzpkdJzW00B52t$fhtIA9-(CsUCu{+g{^EgnFM=l+WuvfPA>+$UVY%=v#
zsFY#Hx=X<UOZ)rj&6HK%*!~e2a?7{yDHQSNQ>VxGX+g3#asnNys8CAL48c^(e0WLJ
zxOLN7e@G3!pm7qoRp%(8sw*@tWq@r-UHcff4cCY5@n7fnA}8U%Nr>&`_ea^(<rM&Y
zr`7zSYSFTNSXkf`%WEAc?qHGs*So@R?Td+9ldE5fKCr@yaL0Q~_ESI1^09CMDm1~(
z+;OT@5iLJq5YWm>f6gLoo-ZZI+SHPzh5Z=5?$Aw)RpEJ1?9L9;_Zb=XTzV61Zm~Ch
z-fiPBcC4?wf!Vb5BCrwLjD2xgTH#4~Oi6-G)PTtxh+i*c51Jq@4kA2#XO*b%eLZ|;
z*RzEpoZb(B9py0RMp4wv4WaBMlIaLo0c3pQ!<l~K-0j%R8#aICTwhbkQfE@Y&i4oo
zPED+rFVMOqD9g3{EXI||dAdSR$I?f@Ho1HnZ4vz$={=Y)y0C!f%<{-JEU&YjYFk_O
z%wWriA_b&@jVnQ%g9T=rQErJ}c`6ykeNR@-IfMM0W0Ld?D$)v0WkEk4Q^y@KJ}4W2
z!$)ozMcfi={NRFR7Vfla-HR$fQ-sE)f*aBqZeyglvmQ;)CfmNaAXo~7MnU6cIxYhz
zKnzTZ>mvJOC9r;B_QkTc#zsXp-gPOu+PQRjsxtzOlYlthkg-r~lo>+|SJvHRk}lBd
zy8R@4cB4g_iy_U(DVq7$EbAqSi*tKNKWCm~YHDMzPXd<|$Bd5Ox&dbn*buUdZhZVb
zx-=od2!koKyux{V#|!ngrEr-YE93!-Ib=Zz;D(%e<rjBvW}FwxCgnhybbUiQ3|ZC}
zQXyI`Fh6L8tZ>>~$tQcQ2pm*6%9NYTmlkKRF?L$E3s!tMzsQ^?64&*8(}W$hH(mcN
zsVR&L!KMA@>Hc}ex6jwz(d9e;^f?7-VUu<JN45Wffr5Cd$lr8IeYuy?)ro1$_0ZRM
zT%qtdN~jbEM{3|71I71>QC|CJE4jz%k%Xlc^sj`Lbra!TJl+KF#3R<W^Xt{@%p#+F
z<*+C>+>HRduy`;D7aH#vO2mXFdQkmyy`Ilz@vBF2Y{4y_bAwKf^bRV52wjxJ47;FC
z;AA)}2IrX)=Pf6MH!9ZTh5^=EV8B6vX_9U_8(oBHXRl7`8_UoFnnF*hiO0Y{7%63x
zY#>@6ITee2X^>#-?Is)KzPaP&ml2+~Qo4xH;js7V^Su3YZr80u0L0T{2a0yzR^1)S
z#g*?wGDvDoCS)KG_bx=WI(e8pzcMd{@U)t%?79KYs8JgBVVs%uK7u!he>?k1!_W#u
zu1b^FC=;fd%Rs9tg5;>&hLV}ZgrK|yD-6jFn0y6oq3RAQz>S`=uD3raqj7!p_I<oP
za`mU#M|N>37P(+tKY0;TE#5JWvJ_HpM3`v2K&H@XFF6PX5199M!4C-QW{XT-pHAv_
zJ;de`7&<>`IG&9B8)c>}-eE@`*tEG{s3{z#AqLro7RB?fN4Z9W?gMiyWaK6L-k?~q
zW`eR%fBf+MGE_&=P$*1Yj)S+4?R}wt!Gf1<MtGX421t3ntQ^u=_36jpy+gqGV|O!#
zt_Jgkw$cCsaSFwD?-$&^9qA>Iy5E9098$6^!p1^R0vb$^qKQGwDA3?CLjbN-`68tg
z1NX4DtE8Tz0eYA)H@9u>r4{6Q@*gZxH$#p*FYO%lPnx;Cy({wj2^XBKx_Dz}l?bZt
z>mbjT%m=tymR`$)`3*_da&}$_f~PAZ8%iP)8c!Z2k=^=;$cbVgHIPmM(gd(G7B&1X
z7z<3UCc!fy<d#tTmpp+m4aLCn>kI|_1T1goP%?EKLEhy~3p>s+yJ`}t2U<XOPNn4u
zyoq6`;J9%wJQF^JmAb|ef}%`p#gLPXF|m9N3O{-e29>bmdP%fy!we`PD%~fjZla-N
zko;pDmLCqKEQjl!57!n^91g6Y+mQhHm4DfNNXKU#E3<PMrmm!*a=`>S1~QUS*%0zS
zB7R}Ud3fvpzDmc|eP2+_umP6osSBLrL-=yF_odU{gAI;_5-!!5yG=1uceSrbRsSnl
z`>pW>E-C8<8Gz8arl5$*ya*0|=ApGT;=|?v(MXyG#YPS>QDHNmvT19_aI07=rrR?4
zr<Iy)7Z&s#nZg>u*`dQY%%BtS)*ef4(dy3m$pwZ4lvBLnZ&_@013B#>b)_+J;l2M-
zU;y<@bRR9CY<W+zajnCyMh*AIJ-d216_HrK>sT7ynBIva5N&iCho{5dIw7<dSiiPy
zqPL-e;M9Qo>EP*z)%Hkrib7gu(t+6C(MabUj#$idA%&T3`@|?w{2o7F_$-V_b4trz
zMrQQZd1@a!BfSXZVc|6Z9aPeeES(R?F*xdSV1$`3O&DI|!xQs3Bj!Ez<rNC$=IXdV
zBl2Yd5OqUNJ-LpY()WdtedP+-`hEy9tvTZ@NQ@t15TDtP_e4m;W)5m@zxVlazTBFp
zyfCB1APGl%2MfIsM5b*eD1Y5uvsXlZgBt%XfBPr&@J|>1G5<dh{xzf%_nVNru4`r{
zRosJ!1VBiewX5%JO$aF0Gw-NMIFbWMqXuY1$A~%h|Mn5tvGW9?S-3N^z}R#@KiY7l
z{$d9%&+oIzyTwcHsn<%wWE9J$j!Y9va;L#L?WvT^ReqEyuuS?wsl?cBAKj@)p4Zdk
zT;IO!!(sXEDNLc_(e`b=PoaWmf>eU9FVnB-Lp{6gLk+LWB$`9?jWan~`_YPp2F86F
zbOtH2jm>H4p?k}wjsCH=T>AP-iL1%Fn)W5@8)dQzapLR8?)Lo-q)~XUajMnj%Y*m4
zOAm4Bb%X8z;g28#tq@_P3s)khUXektD95Cx(u8hSvF8w%j3$B@CtdQd_d1-oZ;i2X
zs>l~PGV0ys+s;$`wc>f4xyT`c9ChyxURhD2u(#TZc+M<A#`uM^&Rfs8sz;xBtxY<g
zn>~7ou-D&l9J%vj1x?(fdn;GJq{R`uH*N89l^>5(hYFDf$gM@l0IuuDF$3D%88TZg
zIKs_45yiIM4=nk#@gKJ7m0VWsyvXg`zoW$y&$9RgTVVj>oboQk&-We+v_5e5aZQh!
z?34RsSJF2le@L&Ii&+Uf=GBwAOORAkI%+OM4MbTs-$rG{Z8gG2XbT$L<C&^o9s-&g
zFta$fR8wfED^9DOmo;^of7K|1I)W<&P=7KmkXD;j5U9P|GP4vzZXUQN(^z=pS42M`
zQ3@tkRyx}Cl_|qt3f6SHxt43Ki{ht|<M_U>xeO1R=^()S3L8fym%#4D-Oeifb;L<3
zV<ExVZmCeIw{647p7V|~)W31*ezhL8DoTBud34@UAVX~2oQ2XUl4|PT`m@ErLu*Q=
zE@m>tlmkoz%TS!=Sf`FV>EXxG#*P_<-7t+les_9=V+f=@?sPXGKYcJQMLq^(Xk%lw
zx$w=nMR$;FP8$}~523vxk0l!xmxD{9&NRS|(jz~GHt3$R5ma<;y5hzg=q+m(=zlsR
z8dkv)i-nP(8cZ58OK_U;7USdmE)h6|+VLtTCKxyLfZVX3Q8K}`Cb#{*stgyk{?uu4
zr6XfZIXD2`GwD8r7F22UT5Xka)hnGX2zT|YJN|=1VSWY>a!4~0N7gCaHAX~mx%D!e
zZG=~6QxE(RW^7r-zp{~8%ndf2C1N$!o6W;k8Pf(&?ouAK4FKvQGp2v_O|B=N7Rq?c
z_n;)elo@5J^yHNs-et$lDKGEgg(Z@L=$pXk&=;aydnXa_$RO8(gTvL2!oj|(<b9~I
zvoo+|uf$3_SR#TvIa_KN8uFeNH_HMeRYs>$tN(pBq<1wU)9<njZyt|wJ4m*uBose`
z(AIlW3M2!EELrACXSa72Gh%?Lgb)c)C51{s(hK=2E%*hpJta0cDkd=3$&QqY2?Vo2
zocLorG63v_DDSL#J(GYQkQYa@FUo*51THUlp}?+fkCGewhQ+SW5RQu3dAqgtplp^#
zi)PkZ+H(SAUTW7tprEHqyNeb?$sq?3vS$JaCI8?J+|U-#rF{L>{B!CO-7Mw`!=jgI
zq<{XBLbi~{_#jCd&1YBe=|@UEvDqu9^iOr*24g~Rh%*%Z+Kme8<Wr|!KT!4lX_jfC
zZxu6&PLbPM`LR1Kbi@qQj;yN~bl>#_YW5e*gU-#L6NKR775DvO%p)>!==L|!Vqx+f
zeIwtdU_2M$$a3l&WjL4}hA0XSC}?&eZJb4$ewdil(Nt0$Abv83_>g<dB3W2qq#^GJ
zPQ>1L+FeYL-nc)l-cF;7fQvBW)A)hat8^791+{>wo5ee?%T#Z|mGqZ*7A9BNMPxwK
zO*AlA;lUV<5@hT6v8>RoVq86JrXsR^IB?qv@KnJNoEM-tNwU&mp=VumP{+vc4w-{M
zj8aWbZHTG|Mp3)>9QY&TYP<BQ^7E#u<t*mvI3s-b;vX<P7#>()-Mj4Mf;TN}3f3?4
zn#m=OMZdm0fAP6M0LPV3>-vCub<=6_`{fIL=v|tKjLk9SLIZq-iNc6eH(J%w$*4D%
zt#FT^7aa8u+PLSUY@P2{l+PxzM{BcpjZR$}73k3u%hNVq4MXxv4W&j#k9q^X*TnbE
zcIq9QIj%`~6(Z0W$XG6{=AVm8FoA#e8b?q7@KK)d4qf?qSNioo_x-IpJv4o&Fm9#i
zBlbIjimg3zA{E73A~7G8iQgt%bDnlnn|~JWB<u4$+;?sf%pgY6I6wdrT?@js;iEKX
zkzOGxmm$|}%x{MCv|zI%Zcz~gc9sVlB0IJqU_O-qJ%mYO8c}rY(;9sCdMcGN6?T7k
z)S$_J&Tr<Aa%djtsZLq#h5{+RE`dWkc}3JYBVK0eVL!{<w6eDidep_ml)wW|hmaF#
zEY!AJ*^vXvYPD$0w9(_<I@s}}d=-E@15$&0-VnlF;}Xx`XGA#LUl||5*(pwbxt7A5
zdH5Mf%u?k6$#?n`MJOlw1?cQdw>&h%js-30fCsfKPx=vhsd*>Okd%IW_btj+HQOxf
zm7U}>Av8iBoNG_HAy@~;;N(*zPldJO=0k?`4)2m6vac|N&j=UXv_bSgPR%4jMO|z2
z^yJyH&vX{xeMW6sc-kr+|Ndd>->?vRi={C|@*Q~9Pc1XjOc54qgWH_!;J~L)kRkS>
zc>vvhgEZ7jQ-2pclg4oP610aOQ@fPV`fMvakCtaz7ogOSVxihhhd6Huhgzr<3=uRx
zTe1=Sz<8eR1PLzducoRn_WrXxHL2$uPI>3t>>hmhxyjmttrl_6g8QSjC1rSj?oT;Z
zh=o^Ag8;bPNZzWK$DwYGo}+<U(oAXjsSm-*&+DMDkT}50DVzB!+ote$0!}LsRKh&T
zjRp==8g*S>JdeA~5c<NbUX-t}m)vu!J=seTbLG$O-#0>j1xJZCYR`HnXU?wnF`-5{
zY4eb|q6p#wYxC>!4_#`G#m0=~;-S^cF4N&a#Xyu(?AyBY7G4ygnhLVlk0#V6D<#()
z_LlQeK&YM1x%l0FEllKS@$jGB$o%lTo{OigbdHD;P#Rb9C<K4{hNyd(3~amzhQAS}
za*^HL#BiU*wgge9+~o4OdKMq95!yZPjA>zSpb&o_A31ySdFwtvHdNmhH~y40UZ!?L
zw$?^DY`=+>>fBsCV$|LHj%Zmp{xo?#>y8TpLj6y4#?0^!a!e)$_J1wM{9mmT4wS#G
z6NXxPra8m=H~^(O+{~z)Y5fSJ1WsD$s3Fn#FRgHd8ZzhM_w(ivs<njnOIw<c9h6wZ
zx~ZBGGgoH5_shlg!HV<e_zV-8d8KKL^Ypm}j=3oDTkE=vPxQijU)nGCPhBWRP10Xu
z#L&1K39jASA9XK}=R8S_h+vtabw|DJ^4~2w6N!M9*C<p1zWL|jSoSC1+ud=q@pp6z
z4LLCA(>Miknn@OX=k-jR(2qBa?^I!psyMHoHMZ{^?#u?<s1ZdLFds_f%>G_ol0R}Q
z4YRBWUu7m(%znt5DyG!2$eBfAyLhc;K*)vH6E^!=DEx8GX0NEF!U;s}TwOFt{Q2^z
zz*h0(-^4<__GdJcTeFMG)hQmL_+(wjf@^Pf?t5qlp+(u847{$muLwG_04lhlSq$BJ
zZH<uCv^zKO6NIhP0kAVg?CEc)Kl@;v&G>_w=ly)=0mVJOAaBYy%`R&^qU(J_($UQ0
zj!Zv3AiL4AOQfT!m2Abc4>$VUMK;TbCTRqvcraWAF|&SU=hkH2-{XzxT5*&_a@63)
z=w!V7ViaX;^@6f<IKzahU%3ae$;4!Y<i1?!H)HT=8}34k?duyWifZe|;nI?_G%hF*
zYrk_!3shV@Hxn9kV#eNWs(UFXVg&lvcW3F(Cy#vA4|+NMD;nzF!(p6C#kjUSWl7gy
zQY7dpH;jE={>ohtO~`kb6%DGtVw#>^2d_j|i3Q=sR4p^NW+=E^FggC4a7CbsHUk^H
zzBdWQjN;*09bG-(WT<+?i<>(52CZY^DpeW;v?*Hwn#ln<DpTlC{+kF=big;;w!edn
z>esSDxq5<4*rj(XRVKF-4bFXGE|t9>$F1aFiCq#5x?7&!tnGWf4Iv&?Z{YJKLThHO
zJ3LGqSLT3b!~xWI76MTecehbDtDWu<AF<L7|7qfv)34AQ&xcAqduJ6}INjfWjFpt`
zUY$VGZ-w;F!$jjWDLi!#qC=H~4`j?vlSCyIm*In*BOVoWqLq&e;M@d-%9w`s&BuJP
zHI~QgTYcm>^R7j5Yuy3fYJj!k>2Zi`XB8X56%{bt(VvLv+aD~0Gsn`IdUS`f_zg00
z?&__P%yVx>!3i|rc6itAZEkS^LNs&Ykj|iJLNz=0!zOX;4fFdOqLlMFjWK960euWx
ztyK5z#g`Mzp+X+D7vx&qw`uKIfr;N)!wx%Vzr?3Ncg1d-Y;aOh=_E{PA*I4?Iy9Ex
zWL(64Kd6hg#Le5NQ*$H9;G#2$0=qyHMY7a}6$sbsc-CcBSD8C-Kg4CmlRtx&ue-50
z?s(X_=$tzHSI!B`W*w-slt$$j1Fj3)<=se17(V!ZvHW2v<}_{0xl1u{y)bc)lt)#=
z3Q40x7+mw&5DpV;<v)tcsn@{z4bJ4=^k<aCC_LMK)hb7zYMWL6jNSNLoxhV6ohDqq
ziJoEgWG8Xl9Oc8}JVMtK@)54&%qy0ew`$gZr^ce%8shI&F!XT@mXn+df7IsYwzu}E
z%m9B625E)hY|cl*JB<@+p?GY-3Wl)Kd25O|jD%on$ciWx@^UonEXqGK#|ed?MpY8z
ze&9u!rxS9$ObfVYs24a9nb5?86J%_ua5TBSV?cu``&ql2qJ{^g1PJ72({gXLW<>&Z
z_IqD?KjG=2tz{=s6kJpkBHWpHUXKircza5SwDcp;TO79GRc_K)C<bS9+53Z`;ypqz
zZ+4x_+ISWMfR8CHu+SN1gNzcNp-$CP$VtM{lNMy6&%tq>Xk6`&&8T!S7)SRh3lam&
zMy(X;NxpQBKZm>@#}83WbD_)&*pZcw)T>A$2RsUAq8Bw<4qM53$ar_3h9~!)nRR;A
zWtEr&FEcH!6nSlwN%p`#w?VO*9~TW-sspu*0D`@-gSXbM?A$YY`RH%3E*8}G$0(&P
zRfD!UV8RCZy6iSt_K$18{Df){<-Uu%kmQ?S?}8#ZBd@=y_L5dJW3C@jWDB#P&5wuI
zvG7~c%z!v@1Ev#%uPZ1=ZSw0TaTK&fOIwRYOndKuL({wX^-mf@+aVd$;8fIcs8EF=
zH$lN@7a`4Qd9FLf$1uJ-kS-hnkgLoR%d7#a=sB*t1qgrqz#ytiI2t|Jtu9;$OU1<s
z)VqT+^4irDz9=T9)NM!u!btb2(rrC21_+~F4J*XSBxC2A466wPgJ}75RtlGPB#iLw
z^VYFK5epwpG~p?XcdEE^se-vel4knyIVh%j8=)1<_h7*!Gotz~0hB)R-3kIb?V(R(
zbM^MTd)RY7f=j?{(-?J5W_{<MzAlXBqmiFD7GZn*xr%<jP}1vLoavd(vAjSHzEH#%
z<XqW%Swwi*Gv;zV{1J!YR!Ae5Mjj{YVYJuH+7`Mi1gDL`#ZWS7D6F!k5Zw^6Euu8_
z?se+Dt-U3Lo!N;M-bjwV2E@t>E$nIX%rQbT{bkhn0fVWt#*qE>QcF{U36x+Ml$+1k
z!HBkzxiy+&&c&6sBWY(|m5eon&uhJ0N*-3Zhl5FDHTrug#W0~{-n?S$ww)otmb-i$
z8#zIE79iPSDU4^#5$EPRe{)t=pO#NU{TWi*Eg06tsn)@X=Y2rRVuHBLOGZWY-F^o4
zpk<C9(V%x9UvB0EyGl9t1o@?MVrm_GyM7ND!et-dbAEC%0uSHsA~8n<)hGXOtNl`1
zwc&2G>BaHgxA(4lmK#m;2Y=Q>N}hNO#p_p_+arH+*zf#*!WsWabNTD}`@f$5dk<;z
z=X%UviiY=g4Z{rP4~<z;(aiuh0h9wZu+`rVGogj%jVPUJ)Zqbu%j)+>JV(IlnkpN#
z!XJTLk-7feiF9Vn`NhSEYTaz!^&A^rrXe*-NfP|3rgl}NDj{TKs=mc0b?*l(WaG<F
z<i{p{JukPXr@{ASOw4umYuzeu{J!0D%oex%t4|j^*z42vX8B3i_oKbdx8=*&%v53y
zRf3S9dNkeUHDwY<v20F#*yx|o`HI*$2EIv8?lK81ecbX(J|bm5&3vV_N7pNDsms_?
z>#Y_PR1Y0Y9s+H1olu6Rimq?gsxKU7sXv;s+vs35cB)6-;C|+sOK_KkmM^>IxxAPx
zP@arc&K<Yx%9);|20z9fD_NBZ(&9`SMct65zlyUVz6G~c_E2CzRRG8oiL>gllJPoM
zl9NAniIN9K68QgERx)i{G>nKcAHcMZ14WO{x;#7<Q@`u7IjK+fkATskgxO+?_KBnh
zQ-&wTLw6U9WB~;Xuwawj5G9<Sli3gx1$YNGIsHm#NE5M+SxrctRQL`t=SE2KfTJ!t
zevYq|Gm3ihtKMJWYj-KpfVFNXD~OqiFQCG4+qo}n8NEIKSt1{k<`r_X$slG-{|$XY
zH*Q++UO6kQW3~zGjOeYA{Aqyikp%d5lxx;VU93;d9@pATbzb5Q{bV4FiWF_e)aw`z
z)yQVNAqy&jd(>u8FY?g{&Hu?3+%_9)R&?w{HgXaU9eWgZRBl@Wgrg#VJ_EY@-j*SH
zQ1#KhM0>t2dD&)l->g>l9ds*B@g=)Npy!?TO3IouiymkyO*{gG&-gyn1hJ{rua*cy
z@}*sf;OnHx=5`mZ#Ft_N22l%Zz3MWifrKnJsQu0;LoURohL|}5PUe6$-FTvQR8-f4
zre%aKUO>Y>xn)0(i)E{+CuW<okKeadrLiZu9l(<T*8qRxM+%p_N{N<nwex#Cqe67L
zz)vJ8I;+%VDGe^r)0^!G42d=<6#$KrR=$Jq(BgM@i|~R*Yas_EJqRd_1aD;i5CV#m
zV_sO%XYg`oMVmsF;%H^oEAj#tk?xOMe~lo^@{^1c28IsUG2Y|uY<fG%q&v1*W}x!O
z#-<};f0lYQadL)iuLtf^Q(a0;%Z2sFr0C6~G++Q(G50cu%|y?ms4X+@nm*s-HSXcO
zy6c3FMy-=!4ONfi-Y8q*iRMAdaV+Uiur*`hxjUJCi7elVo=e7HVB^Ccp9>=`M#M5Y
zzwrARC%85REnX2s!Tkb`wQjMoeNoiVe2sw6TtKPJ&zK2kht++X*BDz${r^$+j!l}i
zUC?IPwr$(CZQHi(tE?{Dwr$%+m)&J`O~0QeCgzTbXTI%Uu-Ca(W*$jJ%ESaF=Y`3e
z(KJWN*^&b-;$_3%=_C@64uujC|LuJ$*PHt0URQDHGs~ylL5Nr`xVB)m;K4iuS+|$o
zl^?G^vUAfEEJB43p{*)6rJ`80?6k;(4&18}g(ZV_>2TDPfl4r9<=f9Vt|nLkjQL!k
z?|!}(&B^AOVXae7YvX}fziyOB%fu^ICt#p6p+b;>!(X=h>aCw~d3~o&pHG*_AMr!y
zhQcTl#&oCd`PQ{Lj7TI%#TLE^b(AG~yY=nQ+Q;Sl8$FeQaOtVKgfSAOYPzXcNg@6(
zjBMNH#xtIf1_KIHsoY3UC>m*-SYAsF+FzWKi;c#a8ho;$<)A!jI**j<vPd=jK$TS^
z)_ptCY^_tTbmSK>1QW%!sw?=>*~tp0{ny%no=vpUxzz=0>B}#JA1M<O6iqO=YcSsl
zbRYKN(7j=kD%F)0$dhz-x4P@8MV9WItR2O9!{ehC%s^nyZS0O=`&^Y2Mc}itVjWLo
z?v}UFr{**q6>20r*U_&qXgiH^ZPOR}p|kFg(p~GdX9sJZ=68iImvBu5Hb&=*(U)t-
zL_Wj+AZV~MbNq+zj{E<Q?@l&ln*;9OY5$(yrO86-GI!e}n_^p<BXcYlFL^GEb~&A*
z>Gx%$!RtCEnUJ<i&-VCe2o(y@!y8b>JO`_Quj|9|pQ97TU_n`JoJp%Tz4<RI3<%_7
zVJ_V|3q!uavp~8)|91eANsFu)G%Q&|gKyowfxlkgU;Dm9;@5vi|Dh<~^}E2oEji=C
zp6&ob2ZxLA-EP`Py*<AU_8vwqYS*G@qaCROR^v1jXjEYyT?b(!?H%?i?ip}Jek9C^
zm2U0ORI-;Yz4oodq;}g=l9wlv>h|uU6<WRTsZKg~KsUObG+!K}`#!R8&z5M~VWI!x
z31}c$nO^&|%&%WU<<Yw!uOPh%TaSt4_bpq?lMu5_j8iDo%?4FPT_c*a9M`KOLXx7;
z1oJ3k#cUkuN6q0;nXx2pkq%j1-t2~AIlhGT`y`DJ(NfQ~oxO!>;qzU3*!E?gR~FSg
zCEYv9;JGz6e(CP5&XU^!?bX=`bS9t{N3zAnLuNEe!`U$P8z&Kkz1=qKyS%imMR!#?
z#K8_-SAb-Uy7p?tT$eJHk<!^KetdceWSnUHKOKyhdpB#^nk(p{{;KaCefehraCn-}
zec;_6A-?+@j0Sw97yE6O-5!kWozJ>!p7T=Z$(pP9yAqbqWi)?zY^$6%BI!WTLG~jl
z!RMED#$Qn_f&G|qG6wQs>r3v=EkvpPYMu$AIo2tD`*o=9qnj_HJDdqHHsx?(WcLiq
zYdAK0NY+fI*4CYaV)>%n6T=>&1f$I;l3jj#6GWH|O*blT|0~zFd5l^nN}Xk}T&Z=?
zZCy+-4~H9e)@UG-r(%Vg$wN+;Z=y7sM4lS3?XaAZ1*1zX+;&TspDH4A;*IwfiDq@-
zF58oa!B8*HmPt!ZOjY?X?tyu02#@V(X@e*+=0kH595hQV*zhUGbw5M#Uth_VurH^(
zQ?t~&CJO$phI%sX=+9RIn^QHiEMRCe`q!pAc_N#FH?*w2jW__4kKmplR(H#8^1F*+
z+Q)H`)x7{lZ2LkV@Oz5~zpZK&2Gp#lvHFT-It?D{Gsag~h1Ej=N4&3h2|4CliiH}q
zXL`K)WZTtgbZI1$=qMI&K^;PlI&WV-m3+kM<czD63*T0OA{hH;#&7P@S4cj|$Hypa
z9svcbdOSR74qME)ITQZL(UHNzFiPA!w*Lax91Q>2`m}Og*T`0}>I&>|3<KIP{X<~y
zW?8|!E4S=e!JV64Ta@s@Gse3ODv$LCItQo4XXLU(9je-y>5+WIPzWlykn5sW-Pj4i
z@TIeDdPvv$gpXK%Fyh3tH}WLprOtlRR`Hca&ngbyskF8oM+|NeP8Pfpc#a{f5ia|e
z6Lioev9Ad#?~r9*$u)DE8HGH^U9n9U$(8JCD&%if)GO77RMFB@N!Fbtf!kiIZ3D$P
z5=wWK<l+*5YA}JV;vjGO{Q(eY2nYe{2HI9JF%d@AaL^6|gI#nVUo$pOwp@Kc*t_c_
z4ruxmu&kh3n57`+=yzua^x!F<`}SgM*`Qgd=#Xi{wAj!$)ka`Fka<_1N-hK@6mjqL
zI85S|>b*i$vnRC5-L2EsREzjJB6qZgR4#f1hESR>yBYS2juUsp+dFi~ZA)k>^9rlN
z_g$q%Xkgprfe$Xf^U#kCnS7oK;rrLSE182s;c|bk&+F7GJE(izc2DIkF{FO9Eywp-
z3)C4IV$smJxIZ(1u`;F?hTA{T=-uHJa;aXv8c^bgqZ|t@Y?Lu#CBn;yH6pGEnK(>r
z_-gL0bsmXN)MGFhGa!$qn$7_&Cuz&aJv5Z4VmE>*yjYi%J>)DGR?Ox?Cn`q@7d(kZ
zW*nc`cy}6A6$-Oaga^<UF!>c{lA$974hOs}U{#|xD0S}SqFyuG116X^Hd%RjFRN`S
zDg<fJI7YcU7U{3OoBcO@Lp^f&0KL-8T58bNIR?aYl@l~<gLfGBv(ZjCAtj=)1St)u
zyszXTHGKJ7i2yv(sUzS4WRroo9#~ty``)b^v?_wMp|33k$6D$Hc&XVn5fI1Zu6>HS
zbu-^raO&YQCnj?2guRNIVyx6LCxVCkF?)08b;OT88?7ITER|43YR3r$BWFbWl1m*?
zJgrIn9fyxqP{_=za3!0Sb2!n9UAIX6)RKGBsO*TiP-(wuTEOyM4+Ny8fE(l~zABjG
z368nEhAQEQ0Tsyhg=ZN#IOjm+CXlbNA^xc*OI5U?5fxn!cW9<WZbXgpSFnOs+$=$H
zBhg$?txnLUyYYJ~#YD94^!zHC5*p?WYHlgx!EJ(CHO_=HlfHs^lq5|_YtXYDQ5>Xk
z$rc=SBLzanFJ0(SF#S~ul5<Vz4;DoAB(*0tFb8wlVdKO30y*4A=3y)3Gh#<YUQf33
zhqMkEFd*fAC!fCV^wt=pOeKF9%OwFGv=(hwni;?9C71TtBDL*CSp7A7hkSuAjo?{W
zyRP|exBwQjDq@0-Rqu2Ek{NxHm<AJep@TijjyLB5$?6{&q>V7F*TCQlUR4hV{6F?G
zmj4R)a&U6~uX|ammhAtr&i@~Eo}R_T^?ix9oy{tlev59QLkb{|&>~(RK9O8K?(lub
z`#Z78?2gAeXb2L1pV0C&8#@c)6&QX`|GTB9D_|N6mL++V4<S8UXhfLIRKy&O?(u-Z
zjEtmN9Bui2cKTN{c!E_&P9s}S@AvcU=Q<~$g<*QXUVoZNkGL2#I&@ADFB9uoOcyJt
z1Qld7#0<fh%8(fbohQ??1t{f+^@hEBbzDYRRc0gEQyjv$Y^@HZ$S?5$E0hu-Vhjin
zS4bw}s2K%?&PYH`F~Oh-G9?L*_H#jYTLmqjdU7n(h#DtC%4G|xdYU1~=yApfU$#TP
zs^PP_8ecJ4+%jwhnDnM<0SfY5VeUjj)hu>7MFFz(p)IGHw}D$jG0iN3vzD5Dxy%OS
zht4$%h4dnsv(`ncgYGJxMXUvJM@-PMV~sOO#|p4ztG;p<yrzS3-FA_3^%8qGPXD(6
z<V>C8RT!sLqi`sY#=nC49MW_5-fc|sohRs8SqT%RO4l(<JHk+j?^JG|cbR&1$6h3y
zy$&Ltxb_X6OAs8~dzoSfXdf<hSFxj5aF%E&Oei1jX)7y-+){(-jR9LrEB>o9B0_lh
zI3Ly-yWO3-x@nBjdA6c8V4>UB*SM)&ypvzQIvpsshfpA1F=E5N8;&(Tbr-x!-ZD~R
z$-N@w_0ykCm3N0xGO>BD2j1ZkGK)|}812;9HcEeP0g&pKRl|jG`K}edoaBsKb<`TY
z&K=wc1g)UQh~YSG_zL<tosc6188Gwo15Y+6;a}^%uyM^JULf!obi@bU=dkwqq2VSl
z5qq9ga7J>&ohkZbP>RcUx-W2>)RRqSZjdxmC~=~uBe6wFTGf~0kpb&70}HF8TvDba
zZ<Fa5C+29elw{#*Oxm1+!hfQ6AS#!5B&E{l7K*p(hp!<@UlywT{`HDi;n%L2@AJeD
z()<IA3%R?y+UO)PlgRt9Xx--|`P>}FV_-lo4K*wif-6n9&M{4f!?6txHc=nmx2Au~
z*E$Lc;FsHjVLR)KVV~bhd00#`q&W4iUe!}!;C>Y`2rH38prsM!U2IfA-bB#6?%v(x
z`0S%$@fI%Y`Sq%YXCFTmKXSKS^?Hzk&MGcUgB{s&6r2@03{d?;uFcT_T4!PEie1tn
z9j;D!CndmU%ffVjRk4&bv1!x!t>C)S4Q8f`o-NPia^rU*gjXa=kV;4+UxKJD6xe}Q
zr>=!s%#68P#SLdc2~Vq^nPG;Cb+Ag`8yL!cCU42H0Ux%F0n}PyF|xZ}$w!*ok#ia@
z4Axp2U&#->p_5NepyP<Nkdqo<**;wL&pyIU=YdwrhKq0>0dOzOa>iEqx=&0c#f*4A
zgq&Gj3xW_YMXRL-o_oSyy?II(sr!VxETD(1XGaj2AymB)3Jl<zjP->6<u11H{XGT)
zR>v$rB5;u*p|F}gJ$oG;nD6o@C6_}4sz<3fm%-c$?u(lLz;pcGyQb{sJo-_Mz9eOj
z^2PVctz0WFwyI;frOF7-f~J+SFTKrMmj{pW(Q>C%TEKq8YkMsjbu(FnO?RGQQBVD0
zQQ%?3dGfihjG+YVa=j7X{%PG~bp-1HHtMgtQJiz|J5xojY&HOePu<}JG3A9F$j~p|
zo`l;1xZ=aMtCySVt1CUha}>2$^U^~A#R2U4+H`$6W3{taloZ$~B`USF%9&NZMP`dr
zr!z(kfzXI9R)DQ7@%d#IsLsj+dq5+@?>%W@uh&CFw0m=?A*D#l#yANOmG@8@8O_sS
zxSBm7FT=Yhn1lefj#E(0_`Pv2IkRh*1)0{ijsuj|sZvRh`K*BUCOno<iZIv1y4HH!
z>JQ_CScrubQ5*~GmD$vJ3?=PnbIo&%_Dv(3-9tI4AF3dZERUE|{l(rp&2l&sITVae
zsM;J%J{v+)|M`LJgxpsS)p{Uhj~hfx+I))vXWYCk%gdgWGwG>8X#dfWs0}mBZDs~~
z5(J5Qor=|+f>oBMu4FI+!tNIn4YVbn^SASILTl5W0>b16>H}_q9q*mn9~h25G8eY%
zjy((FV2mbS$Fc4GO3jEmawsGNr}G6of!>s<PSKBva?)T779r{0HN&VCbX~{8JgDQG
zrY^A$4P)s^(rjpbOg>-xU^rHASk*kClqY?EL)l>j$G;FBx9NGPp7#AuW1AQq9NaY|
z#GtW;zC9Q4Up_7p|33b?G;G4Ill_nB#`<5=jf?Gno^Cf<vabJUy8VOAV%V4rfk_1!
zIaEDb@ibmFrd<%<vY|<9+fk{YOVj>+-5m`<k)D?Ey(&C@Fv0}G0QmcLCvtjcc%>j(
z3eL_m>_Pq|Gf6_RVKG$jV^0b;F(9IRbo(=UX!7y-^$YNt^SDP-#|8@)#bnD6n72E#
zzpJ26%pciYNt<@2XVBZ5tvEc8slObVu%7<&WF2w*bzHwr^*Pf#6)VZL0!x|YO#}0w
zw4QevlXxMuxQSSAsHlJ99e*WYvCUI2lM4N*KbN^cT&pv_-$Xhs;dwSQ%<GLBZvoX}
zf$~!S5OMWX7d>5h^*AtTi|2o)t9^5=rc%r4oWd9r`=-BF6J?`;UE7!YF_o!EI$v2;
zgg`dSF+f($#kgp*<9bR<Yy4A)s)gppjB!y>ZdQ?T-m;_6>Cx1`l>^1r?X)Eila+=z
z@jj=z>QE~hXHQz{=ju+I{Fw|c%$4fPn*HN9qUW#P5vxUX;<RttGi%x~$!+cO%Y1GB
ze%$Ec#ki^5Q7}io4-0+J#9d<RRP{F1y*byWTpq_shaU>7+gRKFB8biQsHei3(Q894
z1bG^=M^MM{xA93cRGnF>xMVI<>{Dj_lhZV;#|1u$%@{i2?bNRW72*YSv5|v1q}^~i
zU4~<KgWpWp9|#iQ5wlYP`WoWOH?f>avyx-XDHLi_PWtV(-Obvg@G1eG%4*BU{H9WK
zsZmnnY`SCii%LgF)XvHyUJ$b-97D-vTLOzK40X{~S9JW8O?6g!X6cTI!56||+wfQg
zyE{`TiSo3xObW?U*RqFuqo@~q%i9SV^CyFC9pf7q(du7T<P_Cab2WdRJ>$ry^esv>
zHu1Xgfe1Sa2|Y0HNSWQ*DSbbHZmsI12<ONiGv?n`;z*#eh{%nA&U}%urtl*d4H#a*
zkAn$d@(c6Dk>VZ&N;lhrnQZ>lP`vD=7>p5YyjN0T0uRt*xjMxI&lel_tt(bUNe8yv
zJxLo6kcDj<L592|BE;~Otytme#VwOV!vqNlMO@)-z_y_NvbLm|NME`^33|mllXQZd
z&?9YX(H#bTB5kf)ngN1aNXz}SrzxaJwQg-DJn~NkZDy8x&M5!=YeAcDQK}I^OcVR9
zk0oP15+vw4`FG80<<%?OI`Od%q0@}sb9w}BD8p(#(wu6n85N}xAd(t>`jTRML+av!
zFz%Ui6Ap<jF3i>8muFim*L85(2TB&5cA7WJw$Y7{PwJS!NGO6mjwe@hWN06A5cxw!
zJ1AM>gh$taSG>KN!@0#6Q|hZah?6#lS(<SnF%;Gc;-p$wsELZMRXbN)G(sp2LGUB4
z3xrPdt%iUttkGHqO0>8+4s_ykqg~UHiN|xgK*fOmZw;EV7th*%ls|J222O*)gy!vP
z($aZE5C~~m8_%XN4G9f7<ir}D;J3nn5G3t&Bq3m%<`QiX9U=r_P&wU~0)n?)8bu8;
z9K*xptrqYkk@`a2Rz4_Mxjy2?sbohnhP)T)YTws;Zh#piVcM*Y^!S0)CegSyO(Cvz
z1AWzebZs+8kJ&3^q`%?aOs^p_<l!QUd9i`V<?H+hkFJ{i<7LLC_lTMibJ><rL7}j@
z7zDX35wZs)Dr^4bg!h(NeN^u&UlKYsv$z2#yZv97IS)+d+)YIe@06N4=7NGkFYxu}
z-il?MwZz5X3QcubmT~Wk@EJ@D-HLCl6N;S55Ycj`b4duV!FVjO^C<7W8_aszYtSzx
z&)t~9u1W(@xEkxBVW<KS9S~MjL6EwUl@L%_YOoOj)u1+EIEf$-6IZY>A*QwB<0(5&
zIejZfQRC)s*UV$xAs}5SL;}1H*X?xqczw({V-Ekm61;IW6SiG><VelIq`D9_Xr!kM
z<ncoLExRgv(;S=hMXp#Yt?}N*$1oVJgH5{fDDcRERUqy|aEtAhDk)yF)|W>5Tx&8!
zFJ3P#Bh+Wah8G$|Du^qaO}LL<)3*;5061e@g=S@VJ(kd2!|)IRK(v^t&OVu!0iGXB
zod2Z)GG7A9r-=5dI%IL6F!J{55o0e*9KD=uodWvK=X`GN9MD(~#wL~|&SGa6bM38Z
zmmMT;s^7xc`AzUWyU?~{iIWBT%a%e()(>$XmRQp94s(}^pDt4%1c~e+sSjQ3H37+O
z#Qmk0o6!B(e2pDhe=>gg$*ofG&&v5r-%~nzz=T;rmm#HrS7~tP?BjUNg+Y~L{9mGn
z>qw|VqU(ao4!$rNn?UMLKiBzhqwU1-M29PS1c2XLc36Nlzj#4+=nuPE5kA{Dy9X)s
zwwn9jUdpQx-I7QRJV3EeIu!^3>4!FA1+`)eu-g%$KQOU&kAC$5NLQ%8rPIed$%}lq
z(Y+k*<+3c=Zl~cZs(C8mPU@UfnEB7U7;|LTx6FIs<rk6r4#SXvLgBsSC*6C!OC|39
z?EGH{%@wu(3W5JWw#@%8Rw_61|IJF}Wc{B5+@m_Wj<}uZ0qgZd3%}%p4-@%%FvdV8
zAvPHnz(dHuw;;T+WKDHj*+{J<_I~fb@}~AQ(ruf{qM|v8$Q+&A^S+8+iitqy)-0ec
z@EC)9Ou|;PZxQ9hSry+e7Rr<0kjJW!bOvFVqEX<QhjElpN+4_eA=o#<j#QK|&THqJ
zgvAG$KgwmBkNIjI3zWvfe@GeQVtIq^#=UyT=NRfhqm(eM71GMI92*ei7LHk8Fi31G
zz;8)9q<2FS;~P#Q0_4WYhvW<zMKb4&bIB#lDblq^w=;*J6o&fOn$QSkulA&Y5@p0I
zAgwW@Gr-`=Q7CLc6~s~=qA^~lIq58qcnyq~vkO2=4ovv8fRXTdg79eVXxqM94RM1Y
z05|Io7?XIaOyM~LtHSrWO9KNA!Zm{}!4FwAw5Bb0ijfIqL%CYC(5@c^On)r^<55r_
zf?Ba9g(+rvHjz}MQb6z_L?~kRLTN}s=zw8*JD~v^8h0TRGiws@5F2|F;i3d^Aca+6
z9YHZG=k-63TpmzBfS|pbMUY2j5F)ioaiI;P&?kb>in|fz6Y($`nL5!(fI^7Om^^*S
zA~6HYg$$6L+<?K!;GjZ-370GkZb^f}4-KL?j+-RXg+&mdO$N6gn7SNbS=Ssz^J3r|
zNpH}HTTz-dIQ`y)II3<y_enHN;|$Fx9PbteS^yOgL-HF<1_c(I7h!<bvKl=|kv<on
zJ!bmVpOcd-Kc%>@(Mn~=DZi~am5}`?5cqezr!z0raqGFpAW&mL@wxQN_g(jW+A>#A
zZVK_a1v6hib;jjNS|pM9Nm^&XCNw4WUnmX3(z57_YRYh1v6gB-CZ9(xr#<dp=0x-8
zET8fXHvtLPoag{Ow<WJsU)uH=iI`1**sOFa3e0){%eGf-IBMeo+Bh8v^~gMkxTU0&
zKw>0LF{+7$iG)T5k<}!#z*y|dD=aYuCPqHVEXaCQSr&XZq(H|0AFf5NzDnJ&zEq+3
zSeW1%;W{z-X%kpTtvC=Tf$aG$j4L^bUKS*D!@`RaKa*HwOw)h_4>1a>1!gT3Ps|+)
z8OKnhpaqZOyo9V9eM8~G%}+%n7Kaeg+$;>yR5(rm5{o6`o{T`^8wfhMKOI>>%oj8*
z?wYccIW#Pmkr+z3#8eCf3#>lQO1R5dY>e|Rs@F`!X((#o(`ABenTg~$+f%F}5k8(J
z6Dc0Nmu)yI)^QKsZN#;sj5rtywl+Ng3+dmMtdu+woFD$jDa5EpZa66H0a{tdMGGJY
z!IHHMB|RK9vIxD1oP;4JrkjB{?jbM$jEKV0H!g|;+C4{S3>oW$7Ow?QTC{o&VX$4>
zU{gWABe<d^>vCZYD%hvLIzqtd%oBqI6eN@xs-Os36c`k|9;8hS5Q8DcKNgmt5tNF;
z&SzG_3BP7@vCH6+PkN!Q`*PoJ>qpN|8ZkXYEc#mS{|SW(dDL9A!vHY0o9?yB=l}98
z@9ZnE;{D1!xwc-H46&ZAt513$<~>E!D)j4jC?~X4qMLek+>idF`rV3Y_#*6OsK1u-
z2h&cC(;qTBr~Fa5+oL&eO<&>ZQL}%&+55Tdn}&r8qp}|s@1duDEdv&)ZJ4D7pW}oC
zBxQ&l%%PtT>POUz^aAzY6@G09(<yF1HeqIe2w?Yp5g(ixcyiU;+!OH5)7<x^_4jeu
z@$0U<y3f!(kaqf{<@}>6>E@%!?>R~i4)p{jrwWQ`Bdk7Y${SKSsC@Q)3Yu(o%UbhK
zxykbCRb9u<jHNrbqs6nP_bMZ1{fqd-;xdUF(->)Cf(ogHzt{>FWq{Pei}-!_20pFz
z9-dT|=6o{S@vgtJs}H43z<Pu}uBSbZG+6tN$Nf6k_Yy&p9~sbvam)?Wm%%zFdpt)i
zz<kxNYIOKZAjS$n(aeA|xk?2)Up9*+;?rpTXQS>d<xBZ|ZA+RVDVcmE5=!F0Cn*dS
zZBO#mWvKkB>++2AxkaBr*j`}3M4&di?sn?YQJj^wR_&yyuLfh;LnB|P(>V&GBgov0
zL6wAiIm;g%j1r%i>IO#5;%1}#V|Kh99KD~8W+Uo&GL9oMx5UY>?!4%viqLxg;oTZV
zCAU4h?`yh6-(Pp0vHqjT#zMTm&f>XA$8ew3#6WNzQQ-3ZB{n4jo9Qo`<oR)-_a8<U
zMNdSmh}x$;ta)*d5#9t4=hypvVeXfP!vzdT_iN)}AkI5vE+L(j$dXdiU9pWRFhh8|
z?c8oCUQgsBX$O#tln^AH3PHx%)1W)&jsY7N=ksuwQ40-)Z3~S_Z{0Qd&Wew&D@7>X
z4mq=6Sa2!KD*+Kx08|7@6HK5On>+ViJ;xc5ODDXJ#j`xO{#?h8O-&H)b@DsGFk~g2
zp)xUv90#EYw@=ski`cgj7DJeV3hI8$eLONRB-m%3BfSyHF>+s2rncUI$QmU?0nU4Z
zxRxD}U9{{7Ff)}V>%XsS>6{)>^UfU3$DMA==3*QUYhP0WXg)$L$vJ6~N0#Mw$LCE_
zr@HI0>DZ-0;uuNXe;tYHdTyI-yrF2G5Y)GDDtxm<;J>Hd;}ZeH8bXuoa{Six#qq__
z6AJqPK$k>$uEDRzy6u%+F9;%V>)|#POwxVHg_nScdeG#bBnYBGASZN9?#j-~Ai0DN
zoZ+NgL4={AYzEedGkf1zjqkD4{-Y~J`|fj?nsz(O?9t<U{$UQM3b^XL*jNl3qy>n+
z%#)z`IMd(xP51K;q7o2Y<cY^-mct!j4Ap-xy+CTyqD({55H_1ndCA1^7iG_^)hMG^
z9F;Y93EDTw$}BV1TO2uR<=-z&t#YK5%gi9gt?!zjy(FmqdcE~p`ixWImN*$jr0{a6
z$^XTs`t&{T^pob-D7+m|&CTmx@c8M*dMHjjGZMS(t3HaPDDNP;DG=rAIBbWt8-`9t
zyPZ!=MoWlNt-Bnx(ePaV<?f%WewpCSYWs5zQ#)nDLF*t`SMVdgyEC(H-#oGzlWoIa
zKZohwS#R@IVj%_fGVa8V=Rl5Eh>dtPB&C-SsrSlw^sx&XX&Ka%--sjo&fsT-r+IyG
z^3fcb+xTU1@lyUJBY)Ls2WK4^RMIwnnNUXgbaqmxwDaNOqO8L}Pki><(mYIQJ}dI>
zyL*=V#~(4Ogr@^=0BIBow~ssA?v4j`ap;UjE@+3RsQ<@yWcOyWQH}$W6aP!#Mv4O0
zKRJnQOlG+L7-_-fBpS(s1|h0P^uSpxBO0pMprzV}zIxS0*^>7?4UqjBmyg{ixIs9a
zMs1e$%6|ZQ+(OZZz%9L;V9tQg@+;o*b}irmbkx4PSMFdudizD})V2=iQC5tUzW+`h
zDdJ0EP3G$PXH&FqW0$l;>_x^e_!d?0H2ib@oje=nBIDA3b;7+-r~PLXqSMMnz|kQ=
zE#REnmS(-P(P5qY0V0Gjq5%rKzT`}EGtn!2WHU(AfG`QjEqk>}|J8*%zizv(_ncMN
zrtg^c(}O)fJ$tq0(pI4YAkW%gxvt+zRW|Cg{Pbw9Ka(xk$)T>GTkM9+U}H$8sq9=w
zqnUw)QVHu~Ag|cV7ScM0&e3;VmQz3+rFu~^W2s-@8L8aeQ$A9dC9XT`0LX49HT!;U
zJAD##Ol1lCk3$t3`~TZX!NSVL{=dpFquRO&$8G2V`-X?@Rv4yPC1Uy<;=v^o3|U2=
ziMiqg$s{LQPg+{iZtj`dXNCQ}k$fd$lLm|nXjESVfQjKdJue2fT|WEs)7|PjLW7$<
zmN5=+YLz?(0)%X?-6%40HstI`mq*zVDVf_glG4R-#6bU_TZd`zxV_$=sNDX**LK3&
zzYq7vxArKJVqoNKa|duG$l2`m6}+qM-R<D}OcF4}9h<Gs+iiTL;s#(qL;sR5ZLul?
z%bGXKI0;`Rbq9qdRB$+5#_7USQWQ<IqM%PdOR0nqn_1s<m$zC3m0c?Q@vYf}@i4r%
zWD%5`dt7Sn>^Zw4aW}_WtEE?$haE*Mgd687s~hV1s{wewP2nR!VS(I1a(;tL2a&>R
zwUM)2O)MsYXRLPJ){d;%QDT$ML~SW`8ikRZ2Eap%QzptGUz96wIE04|7Ipbto1KYC
z6o;l2I^3JqOY-jV^7tOC=XxNBGsqUpO>-2gffa}={Na*Z=vSGt^;J?4^VMFj%6aa4
zdvmFIk-UDPY^_JI#KU_AV4^CdR#oSrzfcAvEEV~vfCg|8D@OQsz%dsudcGM8ti$pl
z77X*F@Cl#$?!KrCVtJ{FeFW9A$>Th~6c`bNHxYId3~pBP;Uo!S_2Q+iMu~vP=s292
zb2Zf9`UPZ<Se)JwP-H~WAl9f+uBKP*h6>!(KcG>DH7_6?Z%ooFoIbw13FoU(M#5-~
z-$A__nudS9-{ACr@S>`;NWq4#K;>a9LatC4_-i47jP*k9dW)JA5<5%$8PfM5jU*Gl
z%v^W9eE!l)vy#j-d?T56h#d1)Itv2Wz7CQ^Fy;X=cMP8S6`mYkol-@E25g4#xv(8B
zwX#zaA$k@<qITXvpOlHN2RFdnQV#M4zGT(cxxOo$Ecnt~2C7jdo5mArRZ3|Vy;+hN
zde7XrEb(qFdnA<d>wR~<4eh@C`gR~Ct{ZgE$H<TXId-@ZpT&0he5~h&vk<%8cN4Eu
zreop~UOIF;5#3#g<Bf<#;WxEth^%|LRiW=D_FXDBK`B-UdjO&B2@L^m9pBv?n`dmu
zUI9493R7{0+Y};eVEObE5?6*>V0uF^1^gPmd)61z)$X>@H7!D$cnjQu^s>gqAWW!f
zu1WDmz6-_Avg^aJ8)eSRdBcT#O+v9#ve|Sn8PB-a(A&AF<<2l{BiiiAiaHShIHWka
z>=b;1Z-tFJu9_YcN+>-;8%p5oyUc{O%o5v!2*=wIS+rlZ?g<a1-Hks-t_uEdEkp+J
zU_xY*Wq$jzzlt2>%(C`!?72$%7Iq~I^~o7fdX6~?NmF#8L1l(`1<eHbNH;<3t4yNZ
z<jM^JNr_rS^}0q>rBCT{pw{TzhY~u#z-NVHV_AW6tYKog^Q`ziQm3d`f;-_Tbvnss
zi?23};|8@Fis|+odJ+pvJ0s8?xK{LuRlDn^7o>jmdBhznT=#vB$jQt>n{=+3Fdv{=
z2yMR@5uF7F7uXEoQ=oVKFa0TuOtzkY@v)w&UbF+77|<ai5tj0&H?SW=%yv!;n^B0R
z^l6YCzyX^s)2*=Gr}ifB92DG!C_%7S3|F(hqe~$rbcVOz7+S|hMZ*PIs1~r?U4mA%
z@!ag#Y9|8pd>FBzEl3hg&p4rjO9@`IQXqqku|fwzL;{Z2!(v7Xb;Q8<w!f6~f2Vq1
zsY3{4?_l~{03!@*NqW86kFyz{8Yvg7jjmzt2CyC+j^)`gnJnt=zK7{%H8|fnc>tS!
zWYH@NTOB%7OoqjsoTNM#^}wcjqD=8lGJds5>&(4b6RTl$*HFBQgy*4F&6=#(sH4*0
zk9jE|-aZzU9D-A7&wb4rsIx)Z>m91?8wbGQXA@>u>=bglGkl3efIqYcK8Ew4vdgKa
z{0%PS(6wj+)jZ_Tuox5f(R{z#Rb|H;AJ6TzFZJ2$%x;H$cD!J6BH~`!^uJbLzZ;PM
zG4M|tz)kAU{BF{3_e8y8Z~L}73c1)swV!9Q+ZvVs8!8lo1og(Ldh%sVwHXLXUP_!Q
zEZq6@lO#PQLc~z^U0;Be6~A`1UjIw1docy)DRy9XVh)XF?Q!5l<X4*3FY*E9v%jq;
z`GuN~9Qlc{WdY(7X~mRc4OlV`7OWuMAK6z=f6R1%`(mAi-i#M0U3mMQHnD0TMlN#x
z1kiVz?4;_I?qprgVaRd%bE;8`Q>2e`(eICqi^;i6P-F5cy8K8BR6f(D0=LrP&BTtB
ztv_4Q7~*ZlPPSxt&EXQ-t{#lIk~><sYl8si>Nm8sQnbY#?*u@ICuWdWU=H6pz74sh
zT~zWh$Pu843V?~R+!E3)2sHeCR<De*(X1-VD9VcWceTQFchc$~xMpE0G3Edw-OMvA
z193v9)Jhkja-ZyFY?q5xUuDpcvOC3Onto7&3}Ue~AtCk7n$6q`n*__`#mL<j?wXpV
zJTChIvnB|$lH-)`v|1{d!^h;C`6qi*`TkIq>N|9BndMwg&TA!QamGRW+%K#s!hwYg
z%7W>^XbNze-tCnW=ud4S`UGe<YFkGQFh7F0g8|}yc{%J==q1vb%L4tLevOVmgt&lt
zYd>_wb;G#(C96uUan$QI^H`~D#o{dIr58p4Dy~YoO^_pb93lCNafGO#p_n%Lq=p@e
z#gBeLpi-r=wF3R|C|fCi5vF*NEH3j)PN?UmZdz?%HW)c??s+iXG7}Q2H^slEY0P>1
z-JA@9v6pB)C(yMp*KZdZhftNNtwf0%jl*}Qs3EZ<J`|8T;wK)Sh%W?;(_KFTEdgyh
z>e8HK#?WKsf{6=XF&auATm6?7ou!z2h3g+66l0m)D=7L@6>ZdV48V%zm0>;0J_Sa1
ziu0V2w9@Cdo!hS>+}e#IxDWlfq8cI?VjWR8t5wEtGc7ke=e#|vKS+8<-bm+!sQ<bl
zUJj36qowKXd2a~OT95In={-|ecSs^fS;sVjja(qyel8$)-5VL!6e^7Dc8AQ}Z|$yK
z<&ZfCS`1#oEPcvlp`erlIO!UQRcRmy$`cNr&lP>c2aQWorf3KWYFl)D>r9KvXsw%=
zN*;}CLH!7;-~IK^qiHBSLlM*Bb#y&x1cZF=ApWWCf&LdejMlFkDluL#<%d}lP9%@_
z8x!<osWs$0GE(yg*nW)d;lEyJ|J4@9&iy|_uxd5`Nr_x=eP^2IRP!sbRLUr69GTM1
z)||Z7ThhC?g}ajz81~JJ!=#ha&HVwwA*tk&3cee7Ly<s8U-!Ua{FWzPbDzcS&Be!_
zbWSMAVgbKr(iM=ni2+gLB9`gg`+8FpKQn)M0wk$#q!*)5PeY;UT_4Wg%QsJM%q}DI
zWGBsVB=US84in~^rDt{?alf40P1n}f&unt^{d<(#68Mf|@n89}mJ*Yg?sF>%QS-#S
zPCfS#A-Cq*t<Sx0`-}3FZZ*6tT1Ofyp6UixBcJwW2$pop9`)GjopPfsl&3o<oVYQ)
zpSk!(K?AuSP<<xgW%$5j&y*<tNF9wL$f691-p;@%oxlGI3S^Jp>c{#Ltr@OJc%CDA
zran~WYS}2;I+AIr;0LO&pWzz#1(!D6f;4+I$yN}&tx(aeF+jFr4@RexV~aE>O%G7D
z9+PV+Vlb-&4t@TD9>=&LTG{8HpMPE*01&hAQ4pJ+%kUrGVu%PMB!+~5P#UC;WT6Wz
zQ<Sk$1ei_eoDWC{g>p{w>9bo}EQW7709gcK(y^OF?~B?o4)nrW$@t|8QFN^BO-G3`
zG!hXd9oZS%QEPcITHs8K=$DWOWa37{qj(UrjHKh*b!Mkqs5k&RE1ct*k3>BO>v}sE
zgP`&RO5QPwfrSs&Rue}E+_8}84g|WorOn|N7d-Z?%Y1@q1Ik7zFlh|Y%+V6~_x06A
zB~H`(c$B2kz;O^|`>7~|J3PUN$48g-DQwy8E^XAy!L4<Gj=3BB9eabE^5O02YXLYi
zPM60LshH~>ue-4Sg2jCULXcu=#D7pmY1mXs0PG8!si6`rT*3m?B2AK!2>XrMku<uf
zy5xfMbB?zH0bM=7`FZ;cMPxA@Y2bR}a4I6l3KkxpCoUsMLMThO3n}~ZWl?Xytea*s
zHPY$&D$fp1Rf_A~rp$Qruv4BDkj_8>AKfG8O6HB?=rV~6Tn(Y`z$t8wXlGhsnIODa
zWUUWoX3XwO8tJsfVIbe6YBXky537nKD8tpPPHgoU)_<#~-;oP}@o8j^1yG><I;ooG
z@#n|+m8gK}t%m41??$J+k%t)6UV=56b4BeMGFsGYnPqmyTlH#^h5%J|s@fbKGR_m`
z%N~m7vQ)ygaGUodoQCs`^>h0MhIzR|Pao4Kn~Y7^QsWnU=I#%R)=-E_)=l=s)A$|v
zqVXtlW=Gz}h)mBD&W}Y#iy#jMizavgjO)5Yu4s4|Ph6t#MJ<|<bRg`y<+1a+I3b>T
zf*G_J!=!X<m?%hNW(V_NNs3$5*3tQdRgU=xbSPR!Br&qs%FZv3{@V)Jgi+$BL-Z0-
z^3x|F?*pWIsy3J=yXxK)kNW2c?E&lR4+~|pDwX}(YLOg#S{DLBxr89(^e~N9Ku-TK
zkqjF0!C{;{c^-8n$=uU+vS@xVZSD;+;xnL}<D*ryw!3}Ls(Pr9!Zp}u+NRp1w5{ab
z?QKdw5sQuHegTfJ1xoJOy&$|u$3g~}a_YVcDju0{?Z<_CJ*At(emyGNTi}X!w>1|1
znTwcN76Rz@nO>FlvBrO}V@~e4$T$9sQjR|giIC=}7lWK;`cGk3yZ%G5-Fliil?SO!
z2>xsvlDVyIblT*j=4b9}SvoH_-Wth%hNcFad>2aptC#jTVvk<eV(@(G8{gRc^3RQ<
zQSNJPf`22sA@3^?9W>Nvcwg6&6ciGQ$%QC0ll;wg?F=g#F4jgH%dc9rS?O|YxOv}~
zp!rtnW9^^VO&#|q88BYc9Kk6*LG13RkGOKMIvFCDJlxq<dsH4w`uf=m6jr0)X>)ya
zuH|fDN-$`<0Pt(bInJ{eq6>x9uuIa%7T(rZdG?UZjTMumHWJ_7?usOiH5+9ie@;s&
z+@>HeeaKEo>H84d%Ya!7g6NNnpiuv4<l}4b<4ieC^VKNO^zM05r-=DkkHs(=^(&!_
z?<FX;N5T)~739>;Z-+o2xbVZ(u6A9nn0%8)`v=U)MB3iDM$>ZXv+SyHouKu)54kZ~
z&aU}ox&71H9j$FZ)tcb7vefk<$G0&<i?z8rLTz;{b4*ChL^bpM(>y%+pig<6#dI(m
zn?b`Oi6=Mj2O>0ryUBmdqyKDXVPgHC+e>RTlwG&}q2m8Fv;3yc92+vSBg)-_JAs;Q
zCC7I1LG<K|HGw^jrt>D_O$`iWNw%BD@<AL2#P)7|d%SKdy3}mA)Qqm49aOz8+`ZO9
zXS{*#!?dMe>FzB+&Mh>DOnh9GA^JJl`u4d#HU;Q0INSuE?a-w&{yF@7y83#_KS<ac
zUa;qRz&q$5MHc<Im&vA;6h|k{O=5?L7sO6mYvXkby6Qaor^T6SI(+Dnadox&k9P*U
zXrCocl*cY+3_kU_(GGo>B=BPaIsTsR1^mWOn)Y2?*Q?W&@vTn;W|+DCUA->h*NXlv
zW}Gyv8JjvOT=z~aD@C`+mA-oyI_O1~^hH6h@)i**we=FElg6__uE!llCjql1Fu38y
zzm}T>kswwurggJZ=rKer-wZ}d@$-G-HD@d0V_sJHwyb#}f}TZ$RtV(=j#m-Hi1$5i
z#GPgv3uVHjxz>zG*y0>^I|Fgu1n0K8sNQI~6iC~5s4Rf4*CN&ncLKu@Z1XaDTd=|w
zlgw8LcH|*xt<iS|ft|guKIq2n`^KSPb9WK}Wp5np>M+#Vr|}kozXTouXMK7ChjWOQ
zNqk1dg0H??3j{waaO>&Ar3PIkgYCZZv^DiC1S>8j<-%HIM7Uesgqox;!d7N4JOr^5
zwi0(mv@w}T<5D1Ir<59!q|ly+q0?RxX<eZ!rVXG7$)P5*5yYS}pZ`ET_$YRjPg%?V
zZL&SC$LXFW*#}1<B$u2yGwsv2<(H`#m|w59f7R9Ad-wE|`_^wusY_m)Lc+8vwv)Ab
zWA%+ND5dGLok3a~s*Z+obW{Nvll(D^eT#wD80TgXfB$^qQH>)BEo!W=d59L!ZV~Fk
zqw9?id+87`qj@!6URw2<uinc&b=N{iw5Pxm>0AT-8IOhOr@qaCMzIYhPKZ8~9Pt@u
zyBKqxX9PnUP`BfrlI->CI25^<G#lcbqxtM?cG>m*O@CTKiHTnlhlozRa*a&YuN`Yl
z50LHii<lyB^q*R(dX$j#OEG)Ztuw>X!tk`Yp63gu^a;OpbDSvp+3iN@tqhbUtBp>T
znEFlAspU@>1J>sy3DXbsvYhIw22LbaA^4c|@M~}SR5ACRI>+<!gP&k0o7buLR)<%{
zL7EO+Fr@>x#HND6yDX7ek-ozda5s$<tpMY$It#qBT;_t9`cX&UjhlM6N)Oa&11aQV
znG3Ra9QZ1OE&ID$<4Y;MR_?q3;ln<aKdYmqx(4<Mw@Gwag4mzwwsX)VOp(8|t<-;N
zCWI&U)4~u32Qo8evW-j7t<oh*@pkh`6LR7)cK*O*;(oT;sm$=+xlxM2$U>bI7F?Fa
zS6Ntn27bB|SM{sCdiKtRRtEy-7Rpio*YD*&pS$e;b6)pGYuE9gb1rb+@Sk)FY?nny
zLhfN7>~0&0t`$-0?fOWhD?=xj&L<sj3HPrmk~Bd@e$;uJZP)+;6vO*yeD^MZZ!a36
z9T+}<r?~j+s=9eocNP=_b<PY;Zx>7iT)8NE#n~wh{4p`*@v7Q$IMztOobBi^TX6Zg
zY-YY(Z|VLQ&$O&vnR-MV;W7g|tED3;vQ-hfriHfz)oKF@ycQ9eau#@!nizS*E>2sw
zP9{4k{@#kY_Fc>+Mz^?ao^*yuJN?Wp>NTD1^2N@`tG<r0?Q%{tS&d`1+iv})(Qxe_
z9>*#l-E?q|^*OkWvOeL(qKYL8$7?96MV+TCP$^|GF*>CyxY!%#5W*xDD~HfR|M%t0
zaAFD;3#d?CsWGw5Q4=lvbrXU;(l>-OrDP?H&ln=VWW+WhR{QN69>>PzXV4t;Hhspe
zp3Z63YxB@Ethi9H7mXt&O9Wik!LPViR}X&TFu9jUz!ho0>jJ7#>(O73O80ctIwJ7N
ziFY+7<l7hrbNm5>0n8kw3n1Pn5+uh?nQC)^TF0^*kItFxmkN44BysW@OKF4(Jabx2
zvc)dK6Fxb&nZe5Aa_2YD`hlN>?B~<V>xAY(2jIUDZz*A9sgTpH<k>)u=UmVkN@A`U
zY~D_;FWaWA*T<m72XsTNhxG%}dB2xN5ypN4SB<gCA?HDX)mSb8XZv%$z5PA8|Nhz=
zL~An^8BReRB*`&U+ojS7XYBKuY^2XYo`hH>LOYq&%=oi(BP$Qq5C}c&&p%nJo*dsA
z;4Yw^8iBd*4^E^kLX!R#R3e{pI}Yf6d^ew-t0hr{sR=nIKl*{#m_%JP?%w?J)oF%2
zZ)kS|LLBy&9vnTl@mvuQ*kZYKwMZFqhyG=~oY$ggGemUBcl4Wp)F!3;1eHJ^@$(1!
zWfZP?-DsZWJ|RBsuAI^oyNt4YCKNYs{O>A~9KHm><HK#+$4^`WIFx{rbU$?Rq~=Up
z_nt1Wnvp-)FPI2e*L9Wl7|O;V=qdzOG{8?{;f#gVb{h>5K;P7I8I7WteYT6|PL#dG
z#VN}!L19BWvHY7M&-I0rP2b0ysMQEDVvtRS;Ao=#H{mM);&vU54+4xB`y=%UrflGC
zwvBZbOBIp!ONg4$HI?D2N)}Kx%Bk(Jr!KC!(mSC6Wns2nzj<Smx>G8j#4y#vpN0!c
z8x0cS^P!Jd9}1DerG&rmc$&d;xgvlZIb_Uc`IG5SMw*l0wp|-@0z=P^?FEPZMzwy*
zLM+$)r)Kx}^iV;1H%ooR{lz8BWe{hrxHe<pco}^;*)vU(>6)+ul6cYC*8tyAR@QBw
zvol=geWDCjKpEhJf%A*Ze>NQtY20icq-<EYdCd<|r%~8^Y>4Ad2@ED*$Z)s|Y4;z<
z(tyRM=k|i#h>k=#`E9x3Nd_N$cCBp*TtCJ}+kq9p>dMtjVCj{M2+k#iC`)wyGD-+H
z4HKjn`QB75xY{@w?8%K*npMo=fl53i4o{BR;`Mg>U5`mseK}_n?Lzk`?3Zc3j+@Gj
zC2U2awpHlXJ{38I&;=b|(SF-%4gm$k_j!Ggm`2~HTe``FWApX(I~Yh6Yo#<XMN2jQ
z9c1Gu>4buQivAy8g!Ln65WmqkydVw+TOePFYDozn5fjr_2+^Y=idHSD-3i0`t(jK?
z;zv4gD8|EYB<7_PbAQ}9kH*zWK1Vaj`E$kZT=~zvf_WAcdUiI@WgFN&U$<98;}5q0
zR_;F&71OKr`Q{!WN92`-NH1Zx$LmOvzJO2FXK5V_TBuOHTpDkGnJ|Qjs6Md7^>qO|
z5q(sf9z*D>(x0smv>=XWC%fQyBJpINL}Wdc#SKDcxX|kLgd~(YQ>m%}#4+)F2TTd`
z6+&7oSzno2re08tOy9xY4fI`ZSk1b1_>@2`%Qy3>spcYZ6Z$r<BB#PR_783p7w501
z;M-W@V~iQnUbb7cye`J{tC_HAY7j;ua~9FEtdS%MI1Qm)O;${ytxY5gOGL08&=NE!
zPMkobPRTyxZEUW*;Z3tSaGUG5K%h7I0ptG|FdYA(crkOa{?FFzG%an%V+pwbqj*VT
z<Ic|&x+>Exz;}rQAao0WKN@t=V%XwW&zZ=$B?JArMdNFfugUzEc7}-s1IR`-+iCkg
z-ti6n1P%K?@5xhWlbEQBoZNf$E4|RmumMeG?)r6o;4!s7U4PxT8b8{Q`BDeX*RQL)
zY#i5du^))W=Lt>sOs-!0x^EuGuD|ce)v9C78)Ij|L!6-{AJVicqeew)hNvu@VP4st
z_7kq+j)-$hI$&PDbxMHqEXuQnnT}TC#*$*vPdC2jI%}*i3;%gnD+wic5!Ba<<U%N`
z(ZXWV&|}Dwm4{f;K1dn|p&pN9&2XC3$N~OW@)>4>&+%rRc2mdASzih%1x`)n=nX<x
zENN7HWrpR0g=?Y8b~tdp`fA&h*T3EEaikY6MG!^zmfx5z`D+Pz^_XA0c~{;moU!W5
z*el4o|Fu7CaWp3|UPwPgO^SVl^a(4<gwsua*;5(EV98L@YuC=G6_Fh)N9aBkGa#4-
zsM?Bc!;u!ak#kY?;yBIuLl+&3wgYoJe&<T~68aawjxMq3+lx(^Q7mBh$zs#*5+Chc
zR&!Rga9%$x;qt-FhiBEOsPqrtoqWD-ba?X+_zV%Kl+l6LI;GBI6>u2?XEL+X)6`sD
z?gd6I@~0R>Ag0Sj-|vKSPpdS^z}Y8K{0<3^;HzQw-0IWjVV8QvaPQebP<0$2Aogza
zIY%gHrK56!_VUy{LMh1YHts4Z)bOk>TtC(kdeQ;oMHC~e1MxtWw3gnVW3P4IQhwd;
zS+I<i49RJEVq5=xu8KF(d-2$}6gpob)e~Z_Ococ%!P4@_zQjaEdc5xl><8p1!1}Jc
zJ34?;xGqS&!5u3&Z@e7UC_{SziOEPeT<9&w@5(1PJUF;8$1V)thpgpgjY)%aI2?N_
zr2IC>bz-he9UJ+^Kp13*2wEx<?_cCsmoLr75=f9)B%^^Ut?yhAP2O71$n*w*kLVf)
zak0B@F<-%!+?5Vp5>5%kER88i+{;5-+uNPlW1vZ7Vt^%1%!d=ojYvyi)@XNjXJISm
zgF9eqYUx(v82{!xqE^Pj(_n!)()rM=&x^oQg6{Ah*sHX1VEmfWsm;4M8Xz>$Mv6e&
z<uOA$^wcm9+irCXbcR(^mt`{}iK@wOQJv+_$))2;XHj!^{C*B^Wl4}dGMpi55dk1#
zjwWr{DSxJb_>D;(klXkpV5uxvBID#sKAF*<K}1KYMrZ}G=kPX;;LatYicGu)huutJ
zrO^-yV+k1BsUoC;`jmjR04OuYyXop!dqt1MLj${6FwbVpA24S}=aa?4vKTHtTiqD%
zw~{vo6&w)fEhlS{s3?6=LIswpYkCLAMUMtlZw()Z^J3W3g1nXj!1$N#nC9J7pJLf^
z_(>5mTN}J1P#LpOVGufE$$a7<!(uY!NPiAoqbKyRfE5TpxxH|C2Utc-s?)<(P{0H0
zim~|Gt_kS22cc~Kbv0o<n52_s0J(hZ4^fW>uVP#_@pxafNr~LzFgSo}^?GQdh<s`M
z74WG59(sI2gMx%gUTTNH#t#{D66!T_+&L+>T|`<$XrT)^@YUVj_9J%9?BXw<i_9%U
zhZ46U;uJonV7#@8o@@YiLu%@P*S2_3T0ME#6UU$yo#WzvXujRPRa!?U(G)zL?OVeP
zJ6>bf_`fLory$Xqu-z6ctg>wzt8Cl0ZQHhO+g@ecwr$%u)!jR~J7Rxv_J8K(yqS?#
z&&+)D8KWNja+!puJ9;pz*C4i6PQOH06&BAXi+p4`9iG_}5>B0X6kL}g2SBNiUC9&y
z$iSVkSPtb#Y2lI&mzr!PvsvI+IVIW9&0OmfWfHi+>u$Qx&LfVAUiE$B-Z*K2yoWF&
zzEp~u0YW~LA%LA0>%jX2f$Xf%jWir6pXiLi$S*^OSS7TNhz-}O<D1|H0tvRBEFdoE
z5M3)+CgfviqwXKVu2Fe}%y4fD|0zf{Eh3-(__CY=(5c#&N=v(IQ@CAv8^1!iHgfu+
zdqy?^XCj~Y_M7ON%f}bhy*S{QM+5vUdAxyw>jPX77kLWR#|~kd6d|dQmrUI?a}z%X
z72}f@HLVxq%o|H-1A{)KzPKK9xy#w;D6JhHWmPPjahzOo+OA@IfU*Cwe#D56PM8!+
zXR(a&j9YFVq2aLdK0k?lus}FTCv8`{b*KgUnJut~-2Sy}=|^H7pwZGJD}XbbCQId9
zLikh;7#U!&Ct)~$o)S}s;t`57Mcrl%r3;&2Y<H~$mgD6*1rn{9r_PQGH=QBlu1}_A
zpPTqj<wT`q7DkkyfQ$)z2B7Bkh=rqslp?bt&x+E|I`q1gIJVb5TWe^;oXjCJtXn`j
zHHpk9C|%ptdAnz?6?{AJ>ZV;YI4E}B<5DLQ7`Ce(oVoXOQ5^jPb;*5cX2evn3pJ<A
zKMvZmMzcKJ8wuVlnY6+V^J2yi>dBP1G|uT^rKKNEVAx3zC2)v3Z;7UCu})!L?Vb-f
z1R1H+|43g}y8r#Y2|e5Y{Ju#xy7uq(?v}epz&3FsjDSKOQPYYqrAa&F3Ao&!yAfx2
zv{6tg1zNnu5PHLQCth?d;b3s%vmeRbweyBE12obP)OOk@D@VIiW}go+{hldpI*<)U
zC;>4eOpgPKCzv|dEB)KyE`y$>*Faoqo(mKg_U!y*0&YQ-H8BhdbJz$mJSjx=X2+-X
z`{{iPYqNXs=)fs~WbcUc^WE;@``7xd={bb)#PD=6WH|;r9EKdQ!CCk{A7(0aj9423
zK}K}Ht2jtryeI2XT$x@2S*WR`UqnWf1k3}Y1`kEJD<~(z5D&6IQulQwqnb!QZB};U
zTz@Uq)GHf2q4}gtv{-J^-(Z*D@gSZghg#7vh_uiIJ{eMgkCrzkIN~2Xlugrpo^Gmo
zux>H4Gx^n*y{*A5l2FQ@AerE2B6H5YArx|vtnGLRej(-jzIiVppgd(aL*K%d(pHFB
zH#$&p%_STg7g|;$%6GWYM8SFMF`jViF@xZ5-6fL8Hvn)L4c1~U<$97-usi3kpe}-*
zTBa0A=<n}MJWoB^7i)g`d0IZ7Mn_NDxg@PzJ28)K|Fg?8Y^639D9Ks^fWcm7L8+Cj
zoi)x-#fp?3q=z*ftXQDqid(r(n3IDMmH`2UCL2Dr5M=a8-zUCNer|qt5QGHEB?^5G
z8eHB7H8tG$uD07d#V#42*Q&S#WXM&z?F!Ji(YzpuwL%Shaek|W)&xcDC;e1SqhxwM
zu20q#yJu#Ywd0rxmTi+#GS5x4bfu0PFK3sUCL<sv=&Y$xE7n>!q@0W%Xwf_rEn({v
zh_9vRohq5MX5)RiiOOf)<&qT4`zA$kliu-J2QYrUx{TA%e@lCQH2!$XdG`WBQb;X^
zD4~{9_(|oe-etCUKSXo{OB7VXAc*Hkk$-9&RN9UD13&?qY<(tWN#I?ZQmN2_))Cf*
z4#G_rlNaDX6=bYOp{vJrM?7LDb!n`HNU5ef{ev!IMGH4bA?+EZhz9IS06I9iylUmz
z=Q=_Y7Z)rZtG{K*+$cO*(;T+8gi3!eCISs9Ms0{l29dFyLQK;N2L#|!1Bx!ts#xMs
zKO@{JdLIuo+w0uAX8^bSG#<go$*ZS(_A$-qN-|Slom&A!ZyMz9E@HsRnmfycz&A4D
z#ACAe?(LiteSid=Bv|uSvj-9y)0Oq4vB2LS{c-V2i-$k3^TdAnV&mD4t)(->oar(@
zKaTI_5-U{FAi)SPmMtt#MBxZUL&m?!(R8+ySa2ypx7Muv+vRGmCXlXK8GN$$Y38|Y
zh+M~BV)r$^z(uWAFI}jWR}@V_T``&$ArQ8;LGE8J*0wq};G(_x_IH5*Hv~oL%Dt3o
z@I22q&(-S3ikVPvuQ9H__*U8-_N`^8TH26fhLkgEw=?$rnP06g$p}77%>j$pQsX{P
zw9vA3QaoYI-!ctZ-%`u^^tO^*d4o`G#(s-&*!IZdJZybn+F>NT3Xr>PCw(Ob4jPiJ
zX06MHEJ-J$tZbs4W1lUjbXLflHYf*5T2^&=5CeUqiF#%IQUun0&C6LII<!L6K_)=(
zibFJoT^k4VI(7D`s>9KLYHi+1)j6`XEK<-`)JSrnnMymw<C)LmlpdQaZipDIThvD}
z`VNO7QNK7PuAMHSWY4lw@!5GpJYXSr%Na?A^n>U5v`S!42Hbxxo6I!Xf08RU-wGR?
zJ4~8(S>(mQ*^NaL%|{Xk1J!*zEX>=xPNa!odb4$3VS5@2ObIcA3dK<34S*tWz$nV8
zL-?4b!zzkf|1W4$`Uu!O|9%D1f>{b*Dc39{LsKBc2o;O(yg2kWa2DQnrR>(k-W=M7
zALe{Cr{>Rjy+H5hzd;I!=9BTHYF21eF+W^r;XFk`h?8#Vsf&gx*6py;()>GOWMPD9
z-~|z7>vFoiE6`^UR?;O#?U>Tg%99xc@?B_xiL=onC(8;btl3#9bXT*dmk+rf&A0kP
z{l^X;a7W8ylCQEpnHq{V|EXr~Khn-ta^w0DXm^Q@h*4K#HVGUlK87*oA`M*JWAQKO
zC*(PjSYTrh+a^aIXVw!VE;zWxHobo=AtlIf@zR;!Yk5bY$OMg&2#8uia2{6ZZ;rms
z@Ik-1LdbPjZDh)4q(Wa^Z`NhJO9B{bc>>{{=O4OSZx76k^De0G442Zj+nkp;G1f;|
zkA(BMsmLO7+e9tCVwxlKdB)wSDGR%*Hh|f^|ETF}7Zm1+HX2)5HG$*sChY#WlN-o2
z?P5gc=Mq+4U@*xuvd&V+4i<+IWdC@p!AhK@!O7;OOeE4qYX9;cvSj&MnuWXiF<8Th
zCz}{BOzC(su+RUQ&cMGiRe<Q)JuLCGt%Zwna@imuLh<^2sHYUO0WtV{Ur!P$0<E(|
z>VfUF#;+5$;C!RW{~fpf6Fu`CZ1mzvbm0aKr@P)mRph(Lz>B3GV8ENfIM;)x*V8fZ
z-Ek-2POwEx`jrb!A%h@;&CA*(O7&iOw#GY4iy!r0Y5iaDL>Bh{<pusf!xM}DzcuDt
zR0~9cKfR@#ZNvv`Hi@mt$%^aKv&h8uQV5g~@AT>4FEn)k!U{6`q&(=6BS0TlnUI~)
zXl^HDwW(yZI|>qAQAYHUT}SH&^=QHL5dEW?HJvj&WCZC`g>1K}krT4DJGt6j9i3V_
zUR=3T>d3FQIlL`{N1;2k+qr-3T+Bya(C~a6Elnebj`l5{SMFC*Z3TI}7$C)Zk>W->
z#TdPhiR&0VlKw!K^%jb-WgudjY&(F78Z%}-kUtH5b*IatZ!{-4M@F_3S2>W_6*R}V
zt!h@J7yGD&qwWq7yrf<aYN#Z4Qe8wFn~iTU)Y=8ij7lfoz=Oje(t;4vuZWc;G8$lR
zsjW686)C%ag@GPt4TtE<VC*Fo5<;q1b}rgAzWdEZ87qfJr;Jbo?MincHTAP^)&}L|
z6u^q7)REtp82-EK{;i<?@Yno18sjDF0%C*ctrkE9`d*}RA@j%{LCBKrA|}ius72U*
zKa9mgnQs^6<I@Jz$M}q7V~AEk?aCAdmh>kJ#%R?}%CAvv8yB@<W&Rr}o?$#K+N1X5
z^J7hC4js9++87N071R0ozy%BR^NrR8X$qV$m_j0y;+y+<h-^vMEynD=Rb%`QsWRRD
zv164{qM!U5;nJLM-wLSckE!>ccVNf|i)SU_&!`ot16q7803>=51Djm>H@=a@(9dCo
z;LWRhkKjaS9TgU)8;K@i3HB16@T{0R8D{K3bpxrOn}wjdRnl!LK-=;LKJT!SxiOi<
z;wgO0`MnS(0^~*;>qo1!?9KSNA_8Z!;WaZaG+FWf!kg`(&2ThwHRWZX^3fXPhI1F@
zd1x?_*cNimre%}3OF?T}3AK#<tB0Qt)9Z1utdQq|!IXvtj81?X3^)dTdN)$1j@XgV
z{BEvcbU-4gX#K^@i<(J>Q5GlaxgBuNMq}IPk^bpWHy;UlN6dsU;o-HyL5>Y!<tb<5
z#;^yIg)_?e{l=^3j|!*E>41Ut!V?GD3@vtA?_g44UW{;U>#L@XO+>ZCgM3KoH1mSD
zVe?dBvN`CxoFQ+LkUv7)s3gQj*iiK`R?y>orIeExF*<2MgM{b06x3~tx?xSd-vu;M
zUW{Up&aPSgPx>2#n<-Rtgq|_LMzP)`NxkYFXu@x%@O6Z$v;TGrv<|LVqM?n^op8-W
z`O-~7yTgJJ^8KxuT}nLl5nd`gx1HWZ_!5D+hZpi5zSXaNEdbxR!aq&_(qy>;T%x<4
zo+wk@jn>)jh4DPDXe;jlofeyocM0NXEN<o>bF5+1Nn6Iv14>MbB?yN#$PT%>D+2E~
z-}s;RlUl7Xg=?3O77ZIu0=G;Oo(k28G03o)1~$6}B!KQZPGCGv_z-nin&Gr4P*$Gv
zQ&quet7zDIkY0hW-3?c#TW#~~w;$-Fnq<wg2{oc=Na|ZP{RtYt5E(#w5>Z8-g|Lj~
zRm07mcVgV!=NjqDz@VAkPG0RsWhb@)C__thosyq(N%p_ivd3pe{#wiccSf|xRf8zv
zGuLhscwevC5+IlFX{_X4*lMP3PQfGLu2UOL0l5zm8qY?0u`49!NDj=8df7nT8P!uV
z^|GwGWeZ*=5m^J1kWSZaX47C@Ny0sqbJkGAX!3y|<NKCSa|te$MqPkf5m*!p_ar%Q
z?nD{Zr~@!jFfgVS6gj?e=-8lq1xgzRi8koh^7(hS(vcBK5*1T(>Y1!1-JTL%<_wEg
zNffISURX0?_PKj~IEq5|q>vLAwIFy70%cTF4s<@N)MvDlRB~>n(e^J8j)8m2-&RoM
zU+CfC?-<F5_rd!QW1wAYS8B<i*}V<bIT1Gld<H4GRk;Yj$4%W5LEc6<kZ1JSq)s>)
zi|@g(ESiPl_I8>p;7fi_JG};7<~xa1kxDc6I^@q(QbCWq0mmqLc_j-4p7&M;h^e*r
zs7IG_$W|%~$Nf8?YMs<^byf7wb1GwH#sCZq!(-q4iKtqq=it!qXePgDQl@zvV-NHI
z)a-z3S<FSX{Sex?$_aJQgAp7iOQR>l&DB;TtU`r@HTRP=KK-UJP(hlW5d(O9gJtK2
zptp&KaI_J{Omr8P(vlV{_~8JEO)||EBY6{?ZFX1wWo%A$%1*9v#AOj>*(Pj$lWxum
zJQsGh2;*)^5i=;ym+E#$;(>F#hWi=<wn=sw$r$%_Waz$A>8AanR7|efdS#<xED~#w
zIj)y1R%B=G{*V`&y<z4*qK@sqbkQtq|I3H{e~!9nRcpl!Hu&%9+FMc6b?~m&CzE%7
zhhF;?AmFn`#w(b@)c`lK%c!G2t-8Kc)rW%way;rfk{y*>rn%Y+yF7V)YkenUeS7A_
z0a=F-vH@e-_zCzihysOdX-R!EaSL-ji9K+#7>1f1YQ$m*5VQKMI6ux#%<OHr(e6@$
z{2~Wz`FnRqE0qtlCupYRJpCh6MoA)jb0FQ?+PXQ~c;3vUuw@Tpyl}FPxO5`-Y{$dL
zt5`S>Nr&n&B_QEryKc9RljI-|cszO01?tq5ZI=}#BK}p9T9cLU`4u;JM`*<Y+m@YI
z_{33i<9w_zX)ticzx*k^27rg1Mhj;0u=~0=nI07!iB1WsE=!Fh7sg=3$Jwk#T*?>k
z-EF8nJKL4KMs=5^f_K>`)QV*-;ymJ+f6*86DT(0MCfcg96>8279IBWMr08%xu@1Nz
zCu-yq&c$0EM>G&2v-nV%9^X-w^WyF5#a}G~?&%I4Mw|xeE7htSwx`wFGQR_h7E$NQ
z;ssaJn}*~7a%qSI_U!{V`Sa<K@A`9-sS2zT<Q}H0K+>{sx$ER1;xTpjA&fDh5J<8(
z{6Q}1Kxn-<a429`;*j&OgT43zJYJa|?Ch;e(OA}D^3dBGZ02<D0n7uS-H}k!?$0+|
zOK~Cc4feJ;ILKcWBD@!YS!;^5F@aw#>VOHo^A5oiNsZJo$PB`ki*2mX!*LD$DB`Wy
zBoWhQpgSuqI7pY0uxd3*9%LG6qhFDx-{VK`slb~CfNFcU51`1pb@jPUBW)vBxOfu1
ztPUx0=tD`|1NU8!kz*ca3kp#vet;eTToIon#vTAc$6lkos*<cjeO4u}jy{W17^%BO
zl$=}wqTNJ&IoJXlAnr#{>>Y_O=N9N#?;PO?u{aVmhh0tHnaydG*gkL~1XNYMDnGdc
zp&FCYx6Gs|{AChc>^JB`)-dtZ!Ctzneh${?mBz)M2~7leXs|FFSBJ7D6<YGJujloE
zA>eaJt)FAOJRxGtunnmwTFZ#4YD=nW;zKQKZo5md4^lq(C^Q|5oFP8-{wXu~OE@4Q
zG6GhPH`44zA0mP{2p;Y;JRX_c1yU8oftc*>GF34r);~3?1;4dpb3Ldu`20V;xJOu|
z=z<*#Lm7wN&THRQ@{3bN`nQXB@d$vt=z%8Y8$ew3zWS=6)qZ+ZG)fy|<2ag3#TgsV
zDJ3D&X~sya+=-S=Gjr&1WCl)+=>j{N8=egQxK$~jBGe5s=;g-3Zl?6zw8_zm`RCV#
z1R|&AC0J|K&HNct#To(x7RZ{D^plM!YPnl`ppW8kecHPE*b!`kx=)NoM*6K}5B~%p
zQ~#LG*LiK8cc&LSA;pJn$b>e;pS89&`v;lImaVbq2eyltBcsLDSNwr1WrH@#&tZ*i
zrOX`hXd&p04(qFba<ioVI6dt=GyhnVw!rE!S7WWIHk4(kEZ<PA;B3oqJ$zz!Py|4j
zjIkPBPX+JmPIhZr^4@CfYD|9Wax6|O-`C$M1AkCy-a9~_A9{e!fqmaG9S2xwl&t3G
zJP%RbV8P$9(*<az-U372Ru})*%gVU5>(AUcQL;qpKqwMmi)$M?AAGr9Q8#ujotD;o
zuU^=l{5-qv!Uf@$0-r*b6Cv^R=Ob!aEQpoOGJI1MuxK%rT8sqp(V{I&l|>%Ek>fTn
z8<D>;<)K1ojuumKl4jK2BT|b$uy+`zneL?bpo-F@Rk$XzVzGnnpd4TQ_*JUX^VDN&
z`PDYRz|{u?6Cm4i%h=hp-#o0&FZ-YdBh{k)o*bROX)&f-s?e7H^(to#Q45eKT>JjY
z3csxHt^XPK_!OeJW;1FPY&lH3$Dk&hcB}P|szALH2TVrQ#JsQic=JFkX~OiNWXHLA
z=Zrl*TF!49$6aNRi%GTRia#hfSAKCuA86!_l@qlwfGIWv)M}x#wX=(-osF}N`}^YX
zHTT6yh94<`{{Zt8Q>t|r+X3xe<&CrRs<5rcFG6gP6SLxNNMegWjl&HU4$f2elh4A}
zT*m>T=8iRFZO@4DktTy#fkkR~$Emh*e&c`%*9ltRxtb-76~)AN?NfHorNSQ)X(#Ke
z2kWKVL9=5PHwPQR!HfzMP0{!aTBSLSWb#jm%+}6f+lPwS6_dUx;N<cQx!ThC81(n(
zi3q{t5u~YExAPks3T!x9xAe=;LMGd5kF{d%t>l_&GsS^t>&Ul=Jp=nI-hF}^*?(jl
z`+sKKZw2uG{rsOZE?af&mxYh)J)?SB%|Le?mYD+JK3XKsA-wPhsfAXYes{F7S|kPf
z@4v?j&oRmMghKnRUN}N?niHGQQ29F8g59sk(^JnTH~TjlLqKWpLWMt(%Do4CMHunw
zh}0>4sd8~^x{<msQ~|V(m=LHw3el93+NeKeHDy$cC->RA{2~o$LnejfX^tWc5gT6D
zO3ayz7{MJTA`5lhv?=XF@ivB`jl<W09uB^XR4qonYVy_gV!Y>_Eo8AKl%Lo#8_SZ~
z-PoCYzb#Ovw1p(P*c&Rt`_*3^==(3fw@tJx#S;KdxTSB|dFc!>GxFv{`$?GnG}-xk
zO8td7VQ&8%lIPLDi>Wn+p}(9BEc3y%X2S(&6AWF&@?=fcf^MYOMul<(j0ulN8;fZ=
zdFg&cuKb*$b`GvEj5qN1Kt|Qp2g?+Ll!PQbY4UJ>Gc?M+pZy_SoXi0U-s_e=GY0L|
zi%^>)ya2g59LU1LkF1V66mzv-e1$aqPnsT#`Z;KacC_wSE7e88z_e3;FJAD5(B4RK
znYzBhW50FVTwvd_?OOZP&1O&Qo1mgoc5@S&5V7PDfRB{;I9q=UTQT^@GN@3<R#b>x
zoHI~#0zL@s;HF2Pb-aj>U#{o|lgHZ0d-GL`;t~&A?tL1uRE%nelgBk5C5d<&dhJ@p
zmUZF7HSe?`0tuJQu%)VZGM10onANm7c6TM#qkw#qHlySK?$0p)A;NDF4-EU3hto~t
z<|_;OliH865IXj;iU@WN>(c4=Tf|Tio6n$SX`O`WGc1Y!AMEoLR_3lr2KxzF%VDev
zXU8%(9VN>{^tv+}O}4oKUxDW^u9A$K{ksuIT{>Bi$^6msFbX?|MJ|C0wQBW8%hmhQ
zf_=TlzRf3&UC3EYBiZUNMovbq<1IRP@A;~uokGdo>N<^Gq|;H10kE&+z;j=^YYU~A
z-8Dgg9P5<=(JdO2*Sm_EF#h$ug(84RIC95c8IGLPehfI8o4>5%U1z2RQ=*OGZc!3@
zr*v`1jNgH679nEHgpUhgl0ig`622}Ua~eOjs?WrAy46trw+2WGKLD5zewUMrR_oiN
z{ps-V{4|qW`Ze-Q@;9&O-`#!ndr6N70u>IEnPXGPdkR-zr7Kf}ougu$34HaQ{>F{V
z?<5JGGVNXfC@ewHC6`X2-#W>p5C!zWd;xKk^b{KHNG=W5bVvW74=9C(3?E!mn|+a#
z-)jGZ4m<s?^>P*h(ZMF0WY2&v1R)63u{#y?9f{!rJrgN*K*&nvKYVoG5r}YOFO`tq
zMX@wZOq4LjwMF+mp{02faHRobWmU(nEoWIq33E64lnjeC4|O??E0+^=LDh5E`z=Hv
zx=<vLQ9lH*&YDh8<qhrf5lAQ5Fn1WB%5zu+mvI<uCD62EXpS>43_+4h>k-YNLT(lt
zji*kpkOz20<V-%y>SSSfQwSJ0l)?Iio50@A^9&!#u@i(nJG2<DrrYb=N5(Le){6V?
zdt$?}%`e@hDV5}t{AW?D(fi%0HLjS+kNgcSmnRZ~*WAII#baq*<%LGL1bi$pT1>|T
z2v=^Mvn4*rf3uKqo1@PtJ~6^X9sxK-1dVu`sL9Vi2C#K?TW6hW+--f)nJvn(L#Va}
zYKyiveDH1{M4)ygB=n`=Xo+eDq{bf%@jj|&O`q-Znx)K1i85xG%~OU2m7K~v|F$Yz
z_JxvdAjMGOO^%=RpEQ8N71_cSTZJWC59^#?f_79~vm?h4!WDD+B#*36O|xQB{d#Qf
z0OxX!`6Wo=_fPqZvm3Ad|8dfZ8Zmr2rtll22JhXHL*4AHs~Fqr^p;qD+DDXua0R=)
z2f|0g`eUaBl-<2*Z}lfT+<1W8E^Dy<ZFk;P1ii!x56o&2<{0r44!I+ff{zLd^H}DH
z#yIrEzQLIO_vg>robZ*blTi(qbET(p5*Y4d?Tje}?qbrxOJP;GLV{n=Kh*A(WR?hw
zYdE_Ib_RNj@qUT2TJjAWR%cyIU*eEP4(!EPb`{FTdM}a6&1gVF!N*ygf>4?iNT%yX
zG;7>CVu)6jNQh9$O@r;e$n7}NFzNk>Y-7==PgL7#MssJW@<I7B<E7;-pOP-6?EYm)
zx6kCWgdWX@M>@m6x*kL}q0zfY->mNNKlCu8AaN8U0~k2fZXIrSwAGgVpZ|M$G11Zg
z@1sKoI;Q{e#q(k{b;ko1M4w#UKG8PB0RZsZG)i7KY*~|L(sYVgg)y6kqX3fi{g6;X
z1w&8YFPFkhc!~*8V@WtRvAUk_ZW}R?miQ<iuLqm&<HN4-LaoDanSwWIWU_eX@VH@-
z2jj;Nsf$A2nIHRm9SUX-Ny)!XxC5-8k|%i6-IKI5u`x}_j6^Xp)`X_IF%I0E9b6xe
zXZs5yzS%g%CM-X%Z>FD{SY!7uK~x0HmlI;b^(d&QD#iloTp26)$1BQn*f=BW)9=ds
ziS62`DAE%#!@`=i6BTOv<CE9pK0}zJOyhh?h@SG)qDLY|<-gTg4&IM9i|@Vr{mal=
zr&o$Jjyb~U7rSlx)?mZd?8+4sTMxb<Z_Ou?>%btf2(EvD*So;x!k~pe+XhGDC4v{j
zUf8BFAvIPucN=W9*O5cbt}6rYaOTFUw*^$F-fJvYzON9Lg;UnZ$Em#C;XW_5HZcur
z&E~rkPE7buR3VJdM7bU2te+7V5rA|YFwEGTtl?6)zjP~#)!NPnBLSHtm*eiX1_5C{
z5D-A)1|vJ(PUhjz3mhJ<j)9k?g{CRtl1pTvf7P|`OC@#_*Y3F^8c^vc3k!ij$mxCs
zE-%Ji)nH4-m}fZ@`NpV&I%rrD+0P*&e)dHT>7o5tH6!T-Y5mm~0mIgz`PZ(vW=_h?
zS(LAG;P~NZ_S!5%h*6kS%$c>XoI_ojk4(m5U|A_fgcC4;WOW$+(!i_KepC2r5E8>?
z4}L6I1|Yx9pHqS;3s5n0X0O3HJ53*|*Rl4$SUzI3fs+k1Zl#5(DZ;u|7K@K@ge(Xt
zIJ;9R=FO;5$2Z%1e;7>aDN{BerQ(Mfp#_ZBbTEcJl$%_W2<^ZpMcHE_Yg{vPN-{!r
z{i%!SE={Q0uC$r`RxuU^aP?V4PKilo33#_<?(Ie@kzX0c6r*%7a4L~y$Oc#TgWMV8
z8RC$p>A}Fy%S*D0FL0)8{FG$vDZUo~qk~@9O=TF@5d|*gy=sVxg>0%#$T?8y?Mgy{
zIKsUt<q^9dkd7bkJQipoY79ve{J(eKQe$b0%n^swDGi!eMWc|*#HwYhi)Guetz!%O
zdyAqoj`lT))^3B02cV5V6iIv+=Wd%`)*#?Iy?aDS>9ExXOjyItJTkOzAbY{xErJ^-
ze}q6?GaGKaA^H5=R_PF}Hq}sO7(No8WU2D4cu>?jo|ZtzhX8?N^(u%F>c>k&ZJVcB
zC!u+U9{DvpJ<;tF)%78VKYdCAbM2fg<T77jpKIRM)~)?^BPu>=O<4#rKd`A9^i^D-
z?hQ}G`=lV#0V#<m?ga^n1%rTPsq7GryPtLmUdFCaE!p5vKnifOV96xD<yy3BH>72y
zz<&MOylq1#wNAs#pO=bjRyRXP?9kzi5mqX5dEI#wb^i4ci=X*wXj2VlsK#^Ld;jEZ
z5k?R_P$WfXyjzedy#bncfG!)M<+;v9CiODY0-tKF*&D2}Er5)G(`<%dPjP$unY-#8
zthf`9;2>g79}wJj>#0q;B#mm?%{)#7r&l<3YjA;sAhozWWo?xI>1N#Lm=HG?3X>-|
zZ@@4&S6?)j<C-xiQ8OdDcFIo6-32R4F@uLxRufTmZaTROaQ@u3Rwitd)w2Cy82DMU
zNO!9KNxh?YhL)zjMSgiCpwlr>&$fuCy7uN}g|hNXz+M0k%@+(Uzlql?LPUFoZi$<*
z6XNk~)?DdTIa4}hv|_$q+!M)n;N*rJG|XFWJ!%jtWe3&h;LX2FYBx~U!|l)%k)4cY
zG_hfBZva3#6N(UW@VFBDW$(MJq~gdShDpA0Jw1y;lPPxqNX0%i)QfS$DYB1<rLJZ;
z4G~|KOaZ>CQpOjZEzbFxS-(&|=iK#0P~ru>b<FMUTN)4?_?`ddZ!@ctJ|o2`KWt~V
za8KtQM)AqRW~X{9E`Euy)hUfo2Sx#ZnwN{8+?l`lY}tVYkt0;fx+zEYI38Xa@;TAD
zJahcsa?QUR_M~Wecg1SaF?6Qy^A>H(;;xdsWs{rnf1M>oDrX_!+g<wj08TS(%`APR
zVJj(#AlXzxdh%U^<T#|NTUq+d)un-5!7}3v{qBcMOaqr_5o54H+(yiu20J&#Itz~$
zS)3eJ>a)<Fqo3$8>%>td)A@qP*%;afVcwtQ8zxyWv}PRh9sR&IvLFYs$6TQ5?#7EY
z_rgmP7shljk#7qOs&uT6>`(buFznOcfySsAn2t1UbzMD^Vb^rf0U>-&a4jA@zjo0|
zhQQ(ZXPD}}eF9tv7al~ckTX2{e%!og%|R5JwpW;aLLeaL+JJ{k8ZY$lOy3t*5@Ev2
zq9ATua3b95A5o=ALc7<D%V(!KYg46;JD}klQSjWTld(pJkMn$s7sZF?(R)H>(@A6#
z3Iw*8BS9!SDR7#Xo@OG5V(+n-y;|3+t_CC;5Io2CH{+3kR@NSB!WF@T4IR5{0TW&g
zjNkknsy(n2C75cga%<JdUc)pk_ct0q89ojoO2)8>xG~ZXyOiSTzO8L!%}7NqDfJus
z(cdE8+q2g~J+4p5YxgbX6o}08G5(!YzQFn`@pdAyoxI->E@}%9%Km<c%Nt}cXCG!G
z<qvg^>CouDHz8ar>&7vi6~;rl5|9ofJKA@agYrtk(TK0&)jRU~hCZ}87}@Q!QnFyf
zQy>(*IBttPv7x5Q7oYPU-mIXz6TR;PFXDIS<A3~6(Ek^tnVIc>41X+El}ub<gYW)r
zvdxwEV~RIhMOJXvK?_dNKp_ViSL(q@R85-MUn*J69PHRi{^e{Q<MJElR}lT{+|vEs
zE*&`G2U;uZqnneJ!#E5iZZy82*9NiA1TQXOMi^j>;01w%m?%uFfZ6T&bAC0!8tI$r
zo0UU$rzR{XZ~!gigvr(HDO1)%lqYmdA!XjXc}7h}9pC2w<*BSpO-Pea*EP>p-hL%<
zW3OX-i5}YV6>y~AwJcqT)zdRvZ@$#+9V0@Y0m819JiJVKTx-NY)`q;`S!jV!oD}kd
zZEtEtdy*xVA)Czl^qmjo{@ao;iLr4N**SkQTrp<PFVCvranPg@msx1@>?nOhV-&~d
z<2}^HCJP$ib^LkOep<Dr$V!>o>}Z?6c7F<V4G2_<!xkxZnyigF1O|-`gn+p_)Gl9c
z=NmhR6@_EN%&1sO4Qtisg^jM!P}H7&#siSx<~u{q8(c>fn$7~}I!Y*%ueDinQGGYG
ziu+1-UsYbRi|^OhL~9>Bi#D&dTTa^`LgUFj)Z~56|HEZ`P0*Vbb-Q@t3?~Mb$b+MR
z?NZUjVSS-!8;HWZ?Zio|RB6{tU;I5qGv+IOHG>}{b;-^R_JcvB0DgV7OCN5?v!4o~
z>lKb3?6`g96zt?U$jo+sdkeuA#ePF9m(KYl73h***#y}Cic1+Kmw}U)To+`tc`&zw
zbw><06%Y&~lWEgZ`BEJypAfvPNwvK!9f^ol)s@=q^;6DIfruTZ69^v0RO0R?|HimU
zgbFKFHz1E0h5%14Xh;}JPM2{+R7gPv@$i;M)kHbh%=EcRSxN&hG{uNqNQc8t{;6HP
zX7LFp4}lcKc?sa7T}R!y__qei+j`mOrP;F6>4>BRNPwmfQ3@#Yvuh}z2oY2qq7b`U
zN2iVJ2>#G!>D1X$ON|{)a-s2?*0<>VM<|?elj7CdjJ*u@+W&SpXXgkJRm2I-rS{^g
zxn`UbZ(q>ufSO1>=Ot=YIfr-l$K+NvMnaRCSlp^v9p_~{Kw3xLPPbL9wR`@;mW>(f
z7q;U{ep~C~rW6Nb?~2tm+lmjt;w`$W->H=>W?XSA8e3)!kc{mfXB2cq?=j2#4iEx1
z%3eqeIws_*onfJfC)>3r0Gc$t?^Ut3?0kj@5_gX9F|4$8^$#r4i!`s{kXIkrG|7lC
z-Tb+2e*0<?bJjI<;zln<Qg-&Bk>XLr`G{cnjsS)sf|*H{``~i%J+pl=M?|T(4!duk
z3T3g>2-QND2NyyRlt(7Uw!J{aLZySnA>0aW8%&D}y`s-$_lo#nVZl<?Bv-|+2Nv4o
z`);~ElI<`F9|(~*Bt`<<e>@jWz>h%K`#{Ws+=@fP8MKhql{qv5b)tKE$C5QIt$3h0
zxqNXO&Bse~0|GjEm6UZE<JI4}@0$Vr7*9*Q0c#|Byi6;&dzbN+I(dG@OE*j>NjiFE
z)rw<ZRzUwD_I=@<vG!@(s#90GrTMD&*OS<7(w|vqW2`rt|6vC$er-uax&tN;sx!@*
z+2X64rQO+!ls`Ll{(~$B-F}C>m=AVK|32Q9)&c;;cQ*gc9ssWBg18=u+shU2y+0O6
zLDy2!DqhdJlBU>DJNi=A`RR0QIuU$GU^mCQz#$|lD>^TT1Lf!KVn=*;GtU~{T-7Vw
z4Rw`kDaH=g(|&V4Ipb`X*f}iT5@;d?mqOof4boxy5Cl+RsfdoEpI1O!jC9vdrL<^3
z7xVJkaI%1ms>hrdV0V1>2<)=^PsGCwnOu_295hQ~*)#6Q0h~Q4dJJdzF!b*MvdWp%
zloC>9o<Ys|#W#X-;otnJj$3`(@(h&@RI_L8ymg4DCwjmR$1`_Fa6#QhxHtCcO?6vr
z9LXTVx5Qp{A{XuiVXqwF?SEwOdt6&vhtRo>CZUnT3#Ut^xRWB~u1S`7WE3B3u;T2+
z`#sY+({lT%=oY9OL~6&kOEjS~OK=slUoQ(I9t+!LT|dhXOT1UL94dJS$Qf31xkWX}
z6xJKB6Cl>m9H+q~#=Mant;w>^yc=N42eLMjY3(=l-auYISK5L{_e(X`S^MweG2MYS
z?+0tSvcL<KM`J&I934~t5kU<9bt23B`}Ti)BCBq@Ap-xq|Jv20z}L7Ms`SHC+7Q=n
zjoB+BUoRpjw;-wC%a5oeKd-%+yiOogW=DnOv3veDdCoK}Z@DjfxjqkFTRN;p))7dE
z3*ELl3!jBD5-Wr5T<iZhb!2}xd|o*GYgok^sVL({s0YV*c06*naCz;{bPV`Jf)O1U
z_byu)^s#VqHZZU?@O85wPyDs^^&z@>9nt<E`?H;<DDmf$5Roi@%lrMIdh6}@3=PV}
z&uvl+NN!jPzV9;PqhiinepT6{x=PMg$+);=+gg<Rnek}0S2Un^gAkI87dRu<<XO@C
zfsPM`EbgC0Rz|rnYG6HwnFdNWN1|(L#bYx~goiC!26wlm717N_MD!K;vupbF#rTxk
zkuKjI6tXszA&PTB6~>1Uu!ZQ)_K}L+#>OyZXrdw)ib+H2Wa4d3N&+!Gmc~g54Xb;+
zY9*tHevm_827!2S5pmLX6%z|ibg+?+bsjYz)WTxZS0!aO+W1p-_ATxxz>3A8dxp2l
z(!P<1I_+&qRI6oU<~b-Muu>cIhgNNDWp&T&Xe6K~IrKBG9WIhTCc!=2zq5nfo7`*M
zzXLkUY~c`41$+ApUsXKv<MOO_BCUj!^U29*CPfO7;>xOq$~KSsgJwXTIFjkF|DMvB
z)iu=hZ9t26E?F`OMM>&eT?^65M)CdV?1kJ#SZVYnuy`9{dLOcz`*ohyK#pmqVAAug
zYvKt7Z?2j8rr+7zKq*TDf&`jNiTcImfx{Tt0^@GkY)Um8v0MbBB1p7ECiJcYlA-9&
zEj4T=r~KM3jPa<QjF?;PinGD`ZrKAN#aHJ<Y6~xsEtYJ<W65mh|Dvm-yzvf;XOqC#
z+OM_`ywD0W14o638x<--lL<QE{tLUEC5gaIu6=ST)p^oGaOk&dve&PR^DDA1bH)D?
zp4&ep+T9SXZFr4E=HvslHGP$bmLB=jIZtHwiX!w&_MHtzcs9ByD7RKEo}@lqs40EI
z+VIFe7a%2uO;g;rcX_#X=jiC@8ZAxt13irF!%Z2s-TfqgTmYS8-_Dz@TBv3YE{=Xn
zt*B}Nqnbg%)T^aifJkMCj>~vOZG7(4nnTBYNk%|`5Hj*-67JTx*w;&M7)?D5lT0O`
z<i?L1<7#K@5a-x%`ZxXWm_cHWAobgDG!>mW#K50a5H70;p%28)#@QNYR{jUQw?4W3
zu3|gI+@8S1)d{X8-1#6*!H%^n=hVZ1{<FvEwX?EfMNywmK&=<pj`X-@T8%Gk_9rje
zFf32fWOhwbdzOLq6(;6iT7uZ{05FY=rj`H*@FZr{G<8{>vydOFt6wYzm}dLZ0DD*d
z1>Y0+J`vh3&!Tarbo@Dq7NM3YIrW*mS{bc-NtKuHhFYJ{k=(B<UBF_sOQF3dBCli6
zZ1LDBG<e=Z^EjDft;|=;T7p(j5|K=Fmw%EQR=7rr`kMPS;*NN)pygH?N41S-!hDaK
zUW9W=^TeuO_qxKcFI1G!g!F4iB(9eJQomOE#{V%ATR^DLx^QF7;M%3`2N-qy@26qj
zE;T;2gRz*SS))Lp<T|Luw+G1}`Yk2;brSzkOVtcrg)2uQAc8!TRxACBir^$g^zq~&
z(w{>H8mv4mEZ32Mxy>vmegscyLt+u258DSA*!CANhq$L|z}S4YL)3^a=IqpS7#c{8
zKMk_dr!HV)M5=iUh)}=hOCOOqy}h#9;WnXgnb?-41FZj$6XmX;<>w!4JQpJrsb@ef
zXB(6<3^oKLn}$+~qXz;ho{3BA{@`N3^Ty=Z?pNP=DSjli@suBOBpxW9G-HUI+AItF
zUC_>-vB(LE8)+Xu%WAmoj9R}tz{s`+E^^4kzOUfsaF&xlXa+Ar)TtPwbNGpn)oqD|
zFb{?GDwILqaHW9Qw%=e3u70m*$aYEAnn9du-|k3u1hE~E{Fm}a7t^R(zlPzg(te0)
zJW{DtMMGe{b(r+6vJc$f`?YL(vm#}0WwDxRvD?Jde`~7d8^m{%nB_krgz>)`6--S3
zp9I1GKM(bkr|Mw?=-~bKvP)YLLU30sJ*`!v(X*ml7-5Np_VvI5Z)m|Y*uEB0bM*iM
zy!kCz%gZuHdxHos@wfQE_$^`3Pv$ZK%0ngHFLMlwSLN55mzZ%-jpo2=O^Z79@S(;b
zB-G~4k{`I?A~$&6A6Abrey*Jvy$MD1-?GFHn3S-o&u`MV@Yag=VWvE&(f(_J|Fuo~
zU-&Eh|G@$o<-}C!<m^rudij9?49P+E{_7k6MIB&cr2n7h@bB4j>|ry)Zz(jt#mtW>
zHMDF%49JRQ+X!X^be~_R)4wleb!+gnA%^|_LeWTmD?!7?)+9#=rS_c%^#UIz1`y`^
zJEo@Fx6AkCTX7gT1^yQ}u!#qR04AS-kWSL)jqr&r;qa3CaWl;rkn0*H!u7eTlw1+H
z^-{9z%1LAX+J@`oD%bY&btmr{E5f_BcEc-7I4Xwvuxir>epHA4|KbjWV?YsNB(eqL
zP_5n#Y{~Qj;oYdIq42^~4EugnJ=n{AC>->h?O_@3`7j;%6$rYH%-5CQ+OnBr+p8B^
zKx!Az*`8~)_z>drQ%u))d{py)1b6q!AS1+sFEWc_>d8n9%-a(vp#Iu}on2GI;x?}k
zMFu_+Z{DII#?5`OskGt9&veiExewT8n^?bQ4P_nq%xg^(V7NiZluXJ_0)5;qHgSWq
ziHbt0_uW^n<&z7Czg5y}z^;rDgYit(r5E^<f_yF!-0?KE^YnfKEQzx5*_e#ixYFOl
zBBmRSFvJU9H70CROh*3v4tH1UlkDi!Ofz)ZQX}PWc<D3#P@rp2XXOJDQ3ukncFqr!
zQ`l@6q9Reu;}=8;J@cqlyPyKiZk=y><tJU>g5YAq;6Xcx-<j=No(H<HIc0zF3oCx8
zDApcxRyHb4Fny{Vl}0P8j$kP!04e~YCA<Gs)pWoJe*L&GtD7r)q;sirH1!Q?qJ}zS
zbFrXPp{Md<f((5=@L_F{vV;pTozB3CrfQ!_%|H}i6X>cX&@D;;*YoCqYuWii1r(Q3
zxc(Y;@?@Je^Y|=C&Y>Wu_7e-r{i@SiT3xmL)PZf|;|I2}BG<hYZBy<_wFCOJ(juU0
z{?$auB89OLCWN!g`hF*J%6KOEuI*A!gaIsW*JQ$S*Oaf%9}Zheb~P_=cS<%dkCBa-
zVV!8cO`T)*!E{_UOCs61eV$(rvEJll@|Les%}U$iBOr)R^SS6uedS4A(eIATxL(mJ
zDcE^SiV2tIc%16vuCl{Z-!@Cb(T__7MRz%i)x5>}`gDo|zJ3a+L9!t{mwo#}X1LHL
zzullh#1ki&QlV;SS0WChy=s=5Dn{eVnd3T{%Saz^I>T8WfOLZhNFW+d28QVX7bp*$
z0r-f!V$tD;JUGD|tjIr^yaf7SV)_=vRxrw_DtmNjyVYm{7_9PlL-cw*SpK0WvhNhh
z8_uskJl>w4>^t{RacAS4-U)VL*@5%@cri7tYUw4-i*iF+RcI<6XZ;(4NUF+q;%epY
z`$<xjS(D*c-aJGwleoujUt^6JR<jI;t>QWh+OeSm*-rlrxoH68wAd8lR^khIPzlC}
z4(`2z?e{Ph^fCyol@(+aREiKz!aL$k!yXjJgl%cnA@&veouYUmf3C1Tq>w1WCn2>}
z{L0Z4Bqa{SZ(G0GZFLW}SXz$urZi4~DFP_#$1!TKY4!!3ey}z(1?O&7^JNVK@!@1k
zhJqHeoCVj4H#+4gr05kmLY;X)YrS%5O9m5FDnzQ@OOT;JLo13##<FHJ!hZm#p_jGp
zEWa=9R{LDSGywJ7-nsgW%c+bmMZP<;ms`6p2|>llv~^4`1R>swm?d(8?}gHyry>Xc
z)HK!HlQN^)7T8~Axb^6<aXgQatH%czIxl537Cd7%P4<Z|FMnB-R`Q0+)Bx*g%CwH6
zlpb6BfX*$Eh0?!5s@;U~NSdexpYaZDe&y%bMM&d^oor;6Nw2S8ONk1}Tn1C%CPc0+
z`YZpfvvVl{?e^NNgL((&b1(ln6?LPMdz!#8r0xtL?5o&Od0{%xYkKW!Cj+fCpjhPn
zS3#z(%6;Z!zN<!s+<KI+g6H8!Y=C;5xfhfOwZOZJKe$=Anwnz{E~1NpMqQ1LI9K4V
zk<~L3r;dog!VQO?Rmu30wj+=P_~H4KzA<J38CwH>T1wop0q%eFe%~{Ut7E#3a|@U6
z&504)v)$^+4f19gW?ef{u=Z9>*BG_5-5o9zW{z?q25p8O*dOvn=NfAXp*ySs`5;3k
zursF5OZ}J^$6fBys?;Rc^cJJ90cvRHUU=O^!SPO_=5dLToi-aTmwa^pCLmx3T5S$3
z_bC|JgS(q#Wv@Rmb@pT`Z};W|k^DT!QN6-dkh)<_oPY&?`bVEA!4Jl0z65@G!%@x^
zOhy@;zU#g|+Lgyb_lboo?cA%n3rD=*sC-zCXLxL1?&vZ^2&r!#;nI^Jemj+oFC=(c
z5=);US7jXj0fij_B|AAW^-;`&F+G}NhYLf8aQUrRs2Y{G_3vGUn(5g7L<h>Ef|Ccn
zLo7J_IjJq2bS-S-?1MD$ZDq8tzyGr<@7F7EGd2`F#E){6a%4fEbQ_qZu+s0<5Dspb
zcZ5}R84u0Kr$8*Sv3={Kdn<z-rIi_*aG{oQ5}|$M=wPx){+;cf4?K#m>hS+PcNrP~
z|8w{M&d<Zj{QvXw{6E2yYpC09iXi^_dCc%>#|D@_cu)lZhC&kcvYYhT`5Hn9<gg_O
zCkeo~sEvKUe1<Tn9K|va%!&MtY+x%ZFL>I^n~Va?t<w>43k}E3qH@al4D~eT%F<lW
zqxFY5q5{?c+=$ZQ{OBWui)9+^r6h{emeBup<%t)G>nUJS*9V9LyLrS7AD{qk=+W0p
zwF5X*8O%#w0wd3h`WDoeM<tZO=OIg0_lXg63?&g`{xwU1jl|ZWkge^FcJT+SW$34q
z-1G-A5Kh;VX^fOohpWIw?f4a41UP;QtQIWlAr+DsF@>kWw<h)O1J<0PTp8mFTtTz|
z>Bz$Q&kC0mCQr_bAwI-l<kLYUU_g|c(dmnw?oN^iz`-}9F!j-Ts6nV1672XbrL=-l
z9{5WNzLNr_f|{qSx~@t=4B?|yicKR#RRZ=4&_<BX=744)q@fe}u?mO*SHor@IU*2-
zX}~Y#L06>#l0gIYJ3?k*`iG}5(Wj|ps7Vo4$4p{a%Rm#KQm--tCAH}%o6<n{`n%^-
zlg7J;M@J)1BB>DleL!9JBkfY$g>T-?E(|x_jszG~yMp;Dpd+9t>q8Z<NB{xMfHp!e
zBCEGR6+?+*gbLcy+tH(z#wNwbC@}3q9Fx!!NTQ?$n<|1v<pX2uLpQJN74RdE0^JwW
z<o_TDumqM32a*M?E%dv`*F`O6CMzu^8~f-<VV9h)sJNC%$UQg7Va5A{E$1cWBkm)u
zv&+x$c1P@<YKrrrM-+Nq46lXx8Zx)pD_9)1$pMi4l8OUQS{r$;=-}ePktaLdP*YZU
z$o2NMRr~nzCB5=@!<S^td2Z`5m8%j>oN<8KyuF=9(=d^g_MvfV!$@aG#W;I9k>TpT
zxg#{K*gSEIZQ!BfGZfF4tB=><gybWS0H|-f{xFvJ2MR9;Ls;d>`0Y*}U9{mC(t+zw
zY1KZai5)ns&7Zzk>sD3grTPWMl+FTlVihPIaq;^_-5kBBFtAE6t3R&!>~zGonH-!m
z`M@ci-fs~kHxC^#TSU5Nch(-+xx}<}@;c6c4g89aiaMLGGpTJuUs*pNHF;>~bI&<h
z$}+14Jrwpf8v8VU9Rz23n15Z0;Uyd8!nC8Q)?LKd){}~sHUZK0k>HRQtB(2=_!zia
zZ@nzrySk26lxHrEIm|CAuTiszBOj5zRG-Y#H=QEy3o1_B@3DNkD962?bDiQ_vUIG_
z=TM0(30L`s2E4}h_O^xjx=Rmt*{C7e0~D3`EOzOEPOa$t9aBEFdTKGEPCi)h*!B}9
z@K%YGunA;HUf~gdRl{f`*lFPX19v4ZY$+#MCHv>5r$2@Wt-sfkN!v!y5`D%=TJ*4u
z!DA-B(x9w6rb-0Au9*D_{#;B7P6t~6v~AVoA|G{mBGU<T1)JBPy=c$iop`v|)%{%y
zDm`*lGIQ5LT>y?|_T1E9PSLE4J$kmMTHUo}lKsGJoLSfvxedC-gG-r-rH?7B%9o1i
zjJy{npi(3-S)T63GxBMxxPZ+(+Tb^VcV)ZG(xi)JJE6%!$r`sMYfFr^uv^f?o7<_W
zwcsad$|}-q$wkk`wXD4kJ3XGX2io)({5wrCCzG%pauA!GiuQ32Y3SYeCLSE9T831n
zb&|rt5}U9n7uh#Q$HhDMw?TcQcq#7m*m=3&v2&c?J!NJ&-6qH7EK05!u9yWaR?de@
z;cdgt?9xq+sA9Alxz!v}mv`!|&J=w+ni={Ry|BKrWVT$GZA3@TR4%TdWLKaa$ZN-B
zb_+2v$La%>NKxVL^J1j#F2!%khC@eA)Q0NRx~s?RqbDf&g)&-Fs=jk9&Dn7crR3nP
zsk8zt#}e-bmBG*XF39~DsxKy2b{mh7Bm35A#;B&6`5=uutXLYC(b+`?m*ns|p?1#F
zzRF+l@cBSD=A_VQo!i2RFz_*eU<fq~v*Z=!qkz$<dT-<QvqBCSxE}e39~%uKU`x83
zA2($VbrMP58B7M)rgkM4UU_7k>Ux5TM^*<bp0R3Nk|S6%T3=yMNx+d~e}32j4)8RL
z9$GYG(>HVVpHyZe^+Us}E{D^=!hURX7VKtYNT{fgSHvSQtA|-XnH=w|6xSovsX-p|
z*gOlE5HGER1n2|L1ns(y{oIrG7P2`zGt!^MCnx02eb6T)G`oYjODo)tzYU#M_l+Im
z4MSOJooYoBk)R*jiaki(`v^4{u})JsY(!i;(GGBa=GJWLp9E|eXXGXlcU<!=(QzN$
zIHDiV6_iDZw@a<hcj(<FlI6%1L{>H+<Cg7u2%EFa{!ChFWHZ%r($ZWl2~LRzB&w@%
z0<p!JcuqXy-oavyOR8(<e&^j%k~R?+nVsn{gW8+G&^GT|JGZw%cQO}$OBeHlg$4(q
zAf8W~#S8u~%HFX@*EU+aEPIu0+qP}nwr$(CwaT_v*|u$)T|0S`?xdfjcYnHn!F^6#
z#~cG)#c?xW(@?{h4ao}c?-8rP_T8ST4V1wXW36evZ5k^(iOv~zB-#eOGj?s;{T_aH
z=><eJ+`s<PZuV$7$^H=LpU3*{1g|m+=fBLf-m^StbAl_{l?+uQD(VbYsiJ96r1k<Y
zF^u+hggTlvpBhRrJTzS0?bH_{UmRoOg&h^Wog!K)J1vbZdbAg%S-ZeOR3a<PhsETz
zA^aXOjQ#eN!yf!k?f!;f6?(qy4Vo?i7M)#LvlaN<p?eiJGM4_G>vv6s9Ai>#R*e!4
z%QFiYI#oWDq~a>=eZbt<&4r1W$paiHEPkfiSu%z)Xb%AcU6RY1a=bde?Ks(U9;8q;
zmSA-kCVUr_`xT=nOzBd7Y)XcvgCMTa2Vh63F4;A53jPiq>Yzyjs}iPP4;wo>@_pVq
z+N>5qnRiv0dA*-*SzR7?o}pu3S0}Q>DPeT1tp@NCIY+25+XC!DlAO_T4WaeN_}Rg?
z#qvm4%V!D7S1cONS-gyilI^kS4(mf@uG|1=A=DwP=3{@?3KgaPHBCN9L#K-9Cg`Sm
zsMbW~s=Q{bI?slXC<mC!7AjCWzH1Zkf0+uezBz`BLYtRrgXW|pj!w-fCxq1+DJ8P%
zB&6J0Q&KfJvf4$?n0Mf^`;^F0!*&F7)?lm!X`u{+^+k&nR1FBQFRCN6f1eqF7Y+4@
z1t*L!KMIT++}&VX64A^GyfkJu%uQaSDZaM}M)F#%89dLrk#K?F64G2J9g8JB529@U
z0QRh-KmETuIOg9?6Ft72p(PYI_y66YF*E+}aatN$b_cAezp1>vR!aU!@h(XlC4J7U
zwQ0jF4eOw_FxUEglXZmORJnxg{@at6dy@E&10x-Shd3@y?uA>rbG=S+RB!Lc)AiPp
zm?SDFCHrwy=TzV=Xc2ZvGR^gJ6dzKy-sj6To*G^wHMA=V_K^oH^<*wwZ}-p2)>JQD
zYI+Hz5SrBR!r{M@9X~(6kJckTzOC)m6uH*AvxAWQu@AV|$2VwfBkH&Nvtzz@t{X+N
z)lW1K@gYJn3c6ra{WD^9!+`rhEskZ!cr>Zp(y`KtrKg$q8(r9!-c>5$mr41^0H=;o
z)L;T#V`xoea<jL6TkxvuYZUu&^LrH>yp3zZiKJ$w;Dn;xY;@LGu~bIkMhqmPnKe1H
z;QZ*5yZWsHyvq?a*>D-Hm8&An3k->6lu;r?71*aaMOCp*J9dJndx2$ev27`K&tp`5
z?@Z`H^5(=YiI~1I9nu5wEHXpDZT@-gw<f@#g#pQMi`>ig<_a2{t%{yEgSy9NN-GOu
zH!Kq;=Z$C=mNS9hJ>^nG@1u0(whb#;YgmpY7>$iHQ$IYZDrF<yXYD$xs*GjK<@bqg
zKRcy3LL2C@ha$^IP-;3f9UA<Pq>^^$pXG2*<(4V@%3&K-PbH;4DjSckKER0S0aaF9
zx7b6@O0D!1W>-gby_&U}-^eJSu-DQQ`{|(95G!DQe>d(cTVP&;>tX&$Y>;=F4+0L<
z;7kE1`=M|AOGS?ar()!;t9<XVY;iR)QeDr{X^;glt{w8{bI|O1JQjc0<pd!DQZTJy
z>X_eV+;Me0fmC5akAvUkO)cf@taQ3Ad)Kwa2`Xsz`Y@<&<!IH~2<Tq0KgGN;a&=l@
zov<5c%(-cLK-9+es5~TW69ofQZZrQmeX7nqw5_GXy%+@>Z(kH*3ML>Ys-aq9xhklk
zT7=6FJPE*x4Bj)AbeQb-sA#>qU7f};X!veV1~L!^iSpR#A^bgM5>y!sYZ940roH{c
z%jM*KC8UinU>0T?fDo#B8ugc-+fBpP`OsNfX($<%*=5p!^McvLN4+yeg==%;OF=UT
z<pRCV$H*_|Ut9*f`1fc;+~uMCb{-U=qIr}s1MzVNDEU8!T5kvp#IONikLHz(i}|a&
z5i<hv#F24gCiK5$mbc7aN_feHu#yO1Byxw1AhM2uio-%<d$}K|fm^V6H%rL@{C2Tb
zSF=yHg_#XJ%t7bcpb+!P&0Ff=4ft`0vI)D?(ds!$)`RO2_4CC^e}!kc$#sh4%yYtA
z?Cv2m<+ukN4Jfc`tL7e03<ZtD&6w`u$|hc&bpSzY_3sKMB<k8#&_$Emv#WBIJ<v(x
z7Khy;U6=8`(7{lbqu1Szf9{Ckvw{1Z>nA}5Hv1Q|$`r4fQ+=BN?4e+I7BX9kN^vd$
zh%dtOF5P#UOB`+UPvRL{&pic`Qds?~D(w{g_R5s>%nxOC@V4e9pJS_nPS;<{a=%g<
znjFW_nNT-F0i#i@K^FOfMW4LL$1RTC*jFxUhlTxID33<;#?%7_FYx+(M@mpSYcSn#
z_Q_4O@LXeXjNBHU2Nhx2rU{*NeL&n3P0ICQu(08FM9|Fu4j~DHSK)x=h34vq@+PQ3
zWAottrJJ+VoUgCT8k-5|{g9T&5r8s^l1aOfhN24rO^b<XQVp-kp>#|SjtuA;MnKyd
znD-|N*)`Bc!ypIj_|}eFIyE8vVOh2$f_u`3yxKkx(>G)aEcHC<TRp}Y*;bhrNAM0H
z$T2(zS{cH{_OFw}?X*N!E+QKYFODF!p5k}utVsApAie6e#-WK+T#$wh-=O#ylEN>(
zygt||kkj@FF(TkaD_aI9>3z|7KIL<2%pjqn>SIvn=DfrtQc!j%Fcy8+X@=^E{3zxD
z$OAeI6!HSjNKdnvH{k89Rc@(+`LIh_Oult;YDE`RA2I+RXsq*CD+TCqf_VU}NU53o
zZ`xA3+^W49#oM0!!_yPh7XTibM@EPyiUQ*w?W?f91c93(9(#8TwO_}dE=$yi_Srbi
zFa5I8fpv%Nbk^jJ!-QlH$}ck(Gx@m{9O2UCTp=NkliPs7^T3||0}>S=<xzJk7IRy2
z)gvok+z@p%>pBtzgZxr>P!hkygv+T*YZ{vtHFu}~+HoKNEN7a%k`ip%j#qK-w{sE{
zvOE>7aT(k$CPNRA4R5*|JcFbLMiNv?EKhdIn-Er3QA12VA>9prc6aI*U3aL~+E2D9
zoyH-^V7$i0*kE)O`Mx5s^Hi{C2XTz#_ReE?46HFNpV?^!lGTeN=|a23`pz;E_}-!E
z6}&h_uLQb=hWxl&+OXXM5B<`aB!e#Nv{2OMxvkaqzNrlopL^S>d)^2oyxKOFkLLra
zk+%!ZPL$<8ya+rJo8=HIOdvP{8jitL<8f6V3tjNlaR9Up3XkA$y;!RB_4PF4|IgQ>
z01q3>z%bv=nfG}KG6zk<X%+CaK{$z)CARgsehK5bYVo#QnTZv}cs@W}z)%$7N#}4?
zBxCgEqnygI5&t3LZqM-YU+H|6R(UydQnD9Gl(8UEosDu`-84imKnu2jp4r=E1_c;!
zC%~dxj`XA%E0}s()U5&v{+ao~z$!X~(V2-|9Ruqa#r~$6)~MvhcU8ihItl?%Jx0*z
zD1In3J0}K4`0(g0be`4Zwc`4r3;~aAWdUDGHNX|)j_8^@>Ol8XKfr2b4Q0$&aHEJ_
z+T=#=?Q%%JI*LINu={$QNJ7_2OtMvHaCn<^oFo0+wmE6WhCL7tka>v2uhfgHliDq=
zI_^cq1J6@xw&wa!shO)yjlF###uXdKOKq{j*V_CjsLGP;eR1a)vTQ<yC0d8cf}ld$
zVbNN8=0|&_-M7NUmqJEUS-!R@)_$)$4rCh66^kdXO<Fqf)9|c<#+m9eB+RSt2{fwL
zeFL_|g9m&R!dMd%Cun-dQX~`h5!)FRJ)bE`)@#PE<R(D3U0<XsU*#jea72`|f;7_2
zJ`|fS*jK_;WTpFhwTzmsD^sPS=lxjp%CrS}fa?z_aP~<>0kxDG2tV1A2f!e^b1b_w
zgW&+*KruSQ9^M`6XOaRY!?s&B823(u4+tRSL%2e=Md|Id&?!-{N~V_Y^;xnY&6rU+
zj~F>=G6ZQ)O*#vb5ZF(8D{YV(Gw}*>SV{QLNpz`z)wzG{h!~SCEKdwOb+V})%q(Yt
zEYHQ}X7>kQ!+?U`fAw<zfqP+OW&K~@z~^Y#IBl>Ze&^~H*on=oLTpR0c2Bdr$&TcZ
zUJ+mWv^6Cakc=OP5V8=PjC?({MuCzCDjt^V3BwOgTpPZunGy7~Y26OR(dq7Vxxa}*
z)Q@V|mgsC&p^ei<w?tbtwrSAtl7>>ax*AZ+)$Hi(XzP7@f9+H@LR+}sKd1TWeMwL0
z%{XRg_w;&uxi~$213&5X{M@y+s@>R5f49Az9br|MdZACZNeY0phD<t4LSKhyLJWJr
z%Jjg?Vka;ZJPb{N4S)Y~QKeP|$|RN+8ZE#rtvi_#NUkhfsnzhC_AMN{hvhUFH$Nn@
zTve7AH=j)KWSpbarf^(Pry`BfxOI&2yHv=#ZWAS}GU^#o?KYY(DroukW)OymrjiX)
zFWi0QEmj^k95o8~kIu2x2HG7Bk!{t%<k+H;8=u2*XWJ=5fwV$Z!*WMzFhd&((h5rA
zK)YkXLS2=t#}`=T6D6{S=H0RCOn&S)K?MF}Sj}z@Hdf-Jo*5?|)c{w|7h8j!yIXh8
zlxH)IZxeE=O^QoYR#BQoIC3;?Env9255Pv7r%w$DmW4`)G}><FTI71u=INtp8-(C!
z+{-U={n%*rt&>?3I*$v}7KME57koBtR=pRP^U`dX>TG>Id~CE-PZYlGH6<Xo(yA67
ziSC?Abd7u<UM?m0+vkUU7gig$i@)sW(=2qK_{+ELZq-nQR@j_cb^czDV1}or<Si=7
zxupgai9*K|IIo~&{q-w3)H`c^OJl@P(j1)k+qn+CvbC4gC8yX5hX#}@fRvy8^1XF6
z{kagxNg_ps+8ynS0HoHqw$Wa+aavKMLS90z&8wHV`18j{7j}iT*;5$Dq>HHkRHdEP
zwHXg7d$diQyQD=%W7vjzwxj&x9z!p{U$Q;o1i{lj)ogA1K?dx+B6@5Rc;>*Txl3$K
zxdE_jkco}+R|eMx?(M>v7g1cS7Dh{-x{J;^JB1A!HAiSm;-A|eXR~5ZMDO(ocS_Tj
zc3<bMXH(H}8fYgrP4j0$>7c)^<nsW5FARq*tB#1wJHz)IfI&j%=lvG%oDgIB?@MZI
zI@y;`XQdkReoJxBpMqe&jg~F>&1K&`Ly3afj~u-58n-=Wa9Y5KKByblsYuZ@5xs<>
zJB1RPxhImS#G?r|02aZzO@H7Zw%ef)WoZhIv26v%FLwg=Fyn1yA(h3YyduJ6R1385
z3Bcl#m%`4C0>HdGo=*9{ru`SgPu;!bwAn%AtDQFIWaktu8y$k`{kDsUTxA~z9~FZS
z<^y%(wa2ix_T1R&TeN&Z8)wSIwRxg9D@X#>`hRaBN5W3q-P$`!MXf>aMf_%=j3Jh_
zJbrA>*81O?u`Fvl_UGhulgM7-KzUqcnr%Dj+s?ykfb=RMs*g*TI2Ja^TAu-NgdA|J
zcb~f-cJWIzAf}bUnrt&u@y8-32dR7>m0<E@6{JV0<C*Gs{J?lY;R*Wc%}pSmkyT0V
zo8<kwJZcX9`f-OXLmeC6VRK(g<BvaX6T(`1S;m=#(H+JhZCqjY5!d-Dc;Eq7X`s@6
zqx8w`DDIf9z%6#rh?Qk#$4g?snQOA3U!GDUBI0Aq#*tlRgL2wN9hv?7Eir>rn6J(A
zKIzV5k36cA!n9v9xpKLa>Tg(F?W`p0shu#Xzj#TM1rU=EM*^h%p&nzS&_jUwxBOyr
z13_5I`*XE?(iDhGEC#r?B&3^tZ4H6wiQ%LdkAa!h@{-{|*A4y8InP$$ZQdjx3OEZZ
zRnnb*e+7^SZ@@K~V;!g9FA2><e#r?rT>O?7V&6p*m~Gf<uui3AOwn@Zb`;QIHCDJv
z(I`#$aiUgg7bK?2AI2^dNQt{_Ps0<=;(S@q=>YIDfsh)(pl9XQO9OC5D0O7jTzt4k
zgnjk`1NWuuXtps>CB`_n&m1gYWN_xBzuy<^_yln!9+-6BEB5@7#f0ztw0D!S^1kLQ
zS+26K8EXts4r=f%kIuS(W^6WsI(d`#q6djvOu)RQK@HpZ5}Z>Zn&Ki|-Q48J4D+}U
z0?5Hcqgqz;MztRlC=L9XPuIAc0vlUp6-(uAvH%;Wm->75+kKqIqEqj>oD^9YJviCg
z#kgV=juWdy#W8TD!kFlDmGBtw&BHkq66Jv~IQT$E&I+9;lSVAtUWzBm0If&NgtTwV
zW2k3KrLTB?PA$yUnm-uyJ?@?5gG|=uOjNEbw!SL%+f_9By0nKxsQ8J4(lb=6m6)Qj
zbG*1bfWUbMgt+sN6SmDme&{KD=|oqLm0vngo|ki$$I8w#kmLI&2SqL7(DCP4!+ynk
zW%a96T|7;p@A5O33<$m$?7sNdHb|Hk7CZnkF2A!l(e9^PqZS1_Cd48%pggG7Ji!10
z-lh(ybAx)YiN8(intXd4`GGm{mX-W;V24F(JI%kP!1@IM>>%7!$p`l_x|vy$-42Y~
z9&^}ZnhdhE<{eJbPAFM+@Af}KI@pHejX*BH`@0!%>-Zokar)f-Tx^*!`m)K1GUN0S
z<8;|GgQ}1Lj-321j=_A4TD0DX!{N}G^yy)N5tu*UV<Yjrd@aGIo_LRXvE$I@Oo3Aj
z_m%nUO9-TwUw<li`kgI%r|9mag>7+al#bJtkf1Hb%H3x{-D6M>SyscL=xDlD^r5@K
z@)B0`!z%A~&tBCtUt1F@g*N7#aK^dJvyJ5i(qI5WKVIDqakw`r1=REF_@Rlr%3gdV
zc6GCvs!yj-(%tv!{$U}AfMW6DMv?Wd`Y^~_6^<~AN7hGY!K=oB6N19B>Y|AWBb!5P
zxl5}zPY-u-CiXcbF*T|EY<ZIbSd>b4q3~fRGgz50hnKfcAG0J7i62<>a^~;1>!Y5Z
zohcd;D>6~TlbUfi&xNvz<|`fUEi={n7uUi*|6<SNr`?IBC-?X()0K~@5T{n9SAWv=
z!@#eQmr_fXiEMQ^+TX-IOr1cs^^RF{2jRqy0QO07kABL+hyTJzI%32jffqFa!OY*4
ztC5qU2*Z1SEr7|ERHhPhD-y&l#`d;NgS$=09&|skZ!g<i+)J(X5>j2!Z*<aS0j*HQ
z9YYqt(-N5tW^aM@)LU~!Zg@&M45X|zxR`}Lp>0{b<>*cgKiTX8IMJh=;<k?{zWq_B
zq(&mm`4(xWkvAs^4w7-#o)7wtQ3#A<%UQjk2hZj6c|F2umiO@P(Ys>RvMgM~x#h>I
zRp^J*;fRci8DjM<*6(=5RCgR=yy~AXYE9QphY%_qneN$2UPST4c5}nAI>*ls4eb>i
zw41XDMqqraru!GEv>ss;bQm_EDL(5@EgCb_-8;u{U^01l4x-n07<lbr+Df*Hu#>w>
z>6~KzPbD5hWp}yVb4W6@m4CjZhzygdUAkfHy8(8WuE8;riK6i5gislnLv8V79z`_i
z7el}zD9E)vdBB(0<njRb<7f6}qNO%rxw?|6z>*1uD^}-0yj>L%w)t>Z9-g^L?ac%%
zB&_mm$50pk4v=_4<OPMYtagpa?C<f+=ArDayVdo|2Wh@C@E?8ae=4asnE%%)>ZM;y
z;XxZh&xhLWFP&&$R{;J{HL)VXh%9M*ve;CmGi(c6A0aWLKs*G_<MU<#0Fzi`Y`gsz
z88~-p`*?5>YTguQ{@9w;Yi+yDLrgd>k$ca)eo2@h$|wOTW86ehsAnXg_OtzYumx;n
zJ)BZTJi!RR=YIYD{jHbvL08F=#++*0oJtXM!|V6?`n{mH^7DCnwteIM3L*_hl`?p9
z6aV;T#!hPT`4IM@SCxwu85*+V2FDSHPSl>fNvT_O%%Vm-<5;Vg7@VCG>#Os7yZ<S&
zBI0ygalUtJuGzbtwcgZWcdOL;v8R9D^PtA_08W$~B#l`wE}nQyotTV3qSD0Se?FC@
zQ5_<Fu>4mOoJO#`9d+n{&pG#Qj{8h$4PP!|ap19(C%Ks-{$l9oS+^%Ltb^+1z|~A2
zbeMKZiCVHDo7f)IY}-zG-e@7k$Aj^?hMM+b`%(rPYZB32`?+oIZKr0%#_nl#>ZHLh
z&>f0)AC9ZHHem`R*S^lyQ|>g@RO?%mf^Dd}Bz%?YeS8fpMDbI&XWV&f=xLr@A(B4#
zqtc`KQ4s5e*U3}2V_X0A<6i%ha+S+@8*0^x8x%Oy8}lfrej0QU+{+h`Q**8+J#q?_
zwvQ3QG~lH=%4$U%DX|{8R>7{aA8$1H;nywqTBf_gYC_$N@qRG-(X6EZyiP%#U4-5H
zP@HmWy!&1KV_zRKWfqfYIxRRiicu5$JRO~qEnc^cago&JdFQsxp(W^3uyPi%qGKUy
zB?&TB#4-U9mgwrznGw6#{T-?(q!J6Xp@AZoyCU03w+SDGr`HcJs*;S5c|4x#Tqlf>
zOE?aSLh(fT{U}zR#OH^G9nMYkSK8Db-vZT-g>EJtCOmICm|G?UDTomrC3xaf8v^rJ
z)|4tg+r#dT-mzWjC(Pt3qrx(?Z1P-5?(=Nzg(Wx=*87Fhxk9G%N1WG^{e74RUqLt(
zF(9<-Opqdcd}P}Jb`#A$ZqkZ29G<YRy9@TkO|GMg0S!tVI`0fR{)gd_333}8f2P|C
z<&^=gzoOx_XH}?7nK^GLmrSBdgh{`h9+*6;shK<HM})fpnY-KO{0HW#g$|1tpSy1y
z?6TTj9q-=hBbDQui%mIDykuGg<X{DGX%%KUSUh}`q?AxsPJ`cZ*$hAjqp7~5>c#hj
zUfIW^DthOue4}=1?1mlAHTU`sAUuZ`n~L4KKfPl2+S;+T!`{s_#|yGA&`FT#m`0@a
zWS})5d!&Q?%+o6VbP>Qsm;gVv5oGqH5yaX;jwj=J2#L};vz_~#9?M**UrXpjqLlq}
zHoF}8BqqsL%HQH7i~AbWZ{l0xC3l{Wjj0x2OB}1p$xRV&)+fB2zj13kA=1%IW*Y+z
zjb%L73oS^g>3z*EcB*eXaKM^12rQQ>Q5;+1yi}KMH&+rbju%D$`yzB?`qPbm*L6i6
zDI3Q46v&Z7zg@|FFfTn$-u30qBoRJ(+)UIYu?|Cbt;vvayUJ0%=S2otR3%$~7KVWp
zKj1o0M4Vrp)<8^J)q_ZkV*&jXtawXsikrlgl}eT|HtF9}0IKx2pLyHS#n;}XeG7dT
zJ=dSQJ-#}DsS|q9n&=@{-#uhZ)6klaA)`de83qXnwVv_mm2SPZzP=lE*}g`n#x$pR
zC`e0<%zynSAl}Nn`*m?c;7#}+)?iBW?W?64rVB<u)S0pYW*VZw`zPes^r)7ve=#Vj
zw@2!092WXBLqehZK-{aKAEp2;QFoGf+Mf<r=((bjI65|IK|b7rHM2BocLgZw!=hUh
zuWd@GO<RLENma3fps-r}fpEIi^(74^kD2m~aRb3d)>U{hKzABNf%J^Z`ERX-LvXGl
zaImp6d}eLaQic`Lxe={R52$UC{%p%#$r8!Vq03%$Yaq`)UkMmP;X%y_?O;3W*}5EE
zPwq3s02Y!KA#&`U2Hix>>jGY1>(WR>-UJo(x-nQ!j5JVs!Am-Ynb`N~I|13+B-Otd
zr=3e+4nTT8jlo6B;3(3>ScxxB-+T;8G8`CdvuxBZG$v0isD&d~MIE<4e>UzCJ1Ty>
zku^=V5?Odzgb~Tha~k!2?1hcNnV8H{?RdOfqjB4FP}9qjO*Uuf3#EcShKZ=Wi5SF|
z(!ygi@%xXUW;UN$dyw4b|8l1F`8w(`$rBhYU*_7r_e(V<SpU8oYLkV&kUVp8Q7;A6
z3_P1OrL9Rb5n>F06Q3YBx^oERH3x3Z?o-Vh*#3eZyu1|4*oz+Mi^^&zPl+T0jfExD
zc{0x3OV9vf`Jpjx5=@Dqoi?@#AZYg)Ja<8smP2|mw>@1QPH8^1x82nHc4=TTJX~S&
zUTZBCqs8%QSgLK-yQ*V^dX1BFaCp$eb#+|BZso#;{p=zexK!Uu9-NgzPh);U0p4@)
zBQLDBL`wCEVAJ2{;0~cIkJ@Sgs3?ChhHu}X9jr|@9O$|*uTB~iC|+VO2LO7<H8$Nj
zGkcZ0KIsSpMjFzy-J{H-NZ5gQH5_MGy_q%&Mxy%JT#ePgrm1q6AWQgoRrJQ-VG&F{
z^gDPFw(M-b!T)R~;{I2e|9^(Xq)lwioXznW82|TWK15Z@Zj%-Mx6Gfa(&*0>lb<ID
z0kUb7k;fO;wUUv6Uq>!Or3_XJS7&{`!*VQYCh8mx?%A_{9?M`V1ODOF{c7p#Xg7)m
z1*Mp0|KlnLC<mNS985~Mn$mAXE-uQNAa2pt-TUL%8(}E$Pi`YDY$I+1d~?h9+kfy^
zUAgjQR=nf)25_?D>bJQ`;$5RBFX#a_!}XM{kwJA1Sgpn{)`(BWOvC8Km02U^AbL+I
zLUL$Y`z{+Nqhwc@JbzoBxgdBmrGo$-ogf&#dR~z4k2HOdk>K>U%V3C%Z1bV}VlH19
zDxhlFmvbpe>}fLGBlMmx#qMXrM8pi%%&foMEs=`Uf3LqXFmbbS-!JZ#7j~}af*!NU
z_6n7!ySHb4Udp=T;B?ML4ehctu{;qIBxp!)pBi@V$moBs*a5AZf;(lwEbdnl2wMt*
zDo=}udI4menV3!r@oe}gVukA&d%k7sn5qV&UAB4`{fz4bA=kmyR@+Io)jGe}k@dk!
zu`F2zp*`~iL*_#Nq`|Vo7Cko`!jNC=zv<2(XfKZ1v?tIytp>B0QAAtj&Ke}@TEe*`
zbr6zkva_V~#0k24TFzNKIVUwyI`3l>OEaD}4!}<5J1^=ocX-eqg(NDRO2Dl&0e0B9
zDS}}gW^7*%GR<V(qI`@8g8C|3fY%We$SGrv?JG~Vl=2iB0?Q`OF*0yoYqITpF~Z%v
z*<YE30il@A3Pmt;I}9436zdMYlwsnIlCS*ra}<Hw-FEGlE2*)6fT1uz;EymNq5YcK
zZG)No65DXJi$XndO6X7{byUPNlW+1x*sOM}K`~QWnHU%1-g=2jO|VNJ7e~-j>p;EC
zBQhJ+0+GhIvg@@*^JSsDv%J;C9XckR^<JUvLi^5yE|1+FXDTUyc>*|tc%p|k*Jv$z
z0t&bmLZx59eUY&DZj0oQ8cL126OXWWKW4J*{8ja%>g(kHd^skIRqK`%R^(TdjVsa_
z0<C$M_|xa`#wRE0uh-|kYawk;3hxWeV4X^#DoLa!8>x>f-!fjlQX^PUQi~|tkWJ^4
z5`L?ut)5MUu;&PF$0pv<H4U#~_~*L5ew8S#9)8bi^COstpM%gy66i!`_2!CPeBQpi
zQkGTzHJfzLR;f;24#)J0cv^fvEBZ=GH7|&_85Cb0Ykq5WKz_@ZyX!gZVEgf2zmCO=
z>){r%poK~$YCf?0VX@inqL{F@csr7c3Id73(D$#*aMTb0I@SS|VdzoyKzq7bO5Iw%
zq6)N)HK|99%R!5)XM!4!Ry^{zbhct*nJ@G9y60~Ey3V7JL3TnA4>FFyi+NayHKBd#
zYXCPs(#KyDtXhB(@_WnE3vjlq%P~j6Ri|wD=SfhY+AbESee7@iK!WEZh<cuLx38*c
z*7y1t-Z-WW296>stJ7bsPAA#nC%*}1dySc9%O;%akD{w8g6jvPPnG=NrP(jTZJ@mV
zyx<S*>*Q+0wUt{y|DA#yo0cHWaqN-|@{b1bGLMW9T$t=7;F|b?W}FCa@Uo1fHGrL3
zu(30qpUkLvXDp(F9W9w(0Aq2u&yISuO<vc5{hQgSMblVg6&Ba1H&C3-g|2RNQA4Uq
z8Q{Lb_{YEc${&m+{_RgKJGWnHJXt>h;eR}ptpAb9U|{**YgYb88h^kF({ochTYWD*
zc}s9UcXn;CP-;=OCbVw@w<arAZ?1(S;rt}-`r{)I=(#4DDI`ZI5C{V3jfa1J#dH4k
z8Jp|PXS=PVOSWl*uqO?+ZF=&1Te=@SN_@|n@v<9c{B&}>BJj^B*b!YJ&ci@_A0NFn
zt?ujj{MoWOw*B^I;a>f43D->LAT+=)Db-`{oKlX;i5Z)jk2ZQ2>Ns{mrBc<3_OD;c
zatJF_+DgDJVq!2iGngFA?=R#75CZ%#du|;Exw0Yiz2PiW$dV)FFY@Z4UDAkJtyV8(
z)~Q>u89XPT+29iI0wv1lE7`0<U~J?u%ktM~iu&f!tW-yGuhOhSJ}-w~ueA|0An=z-
zW%pcFO=J7GL-88ys*BBoH#n9Fk%VDAsl+C-$7%aSfbpS6=Rya@>eP_xOOT|yy*m(2
zNZ?>VLe&b;)bBHNU;1jk#++_4n&aPI@!Laoj<hsP2_)n`z#o+*-r6cth|+cN0dl+B
z3)nwja-Lrv-5NvnJ4+~$^vW!<(NlT6cGod;XylSxJ<L*ZHs6P^>t`K<3EO)w)MRG5
zC=$Zz%oF1M#y7P6ziRK}q99^w-3v@NlM0!K!hT4Qsq0<_a)!XBaeIB3-admO<m(1H
zw!{zCYJ2ChX>6`~CRlN5Av1GL7mEOeb==*zWf}VZK@YPg@`U-ymu>nU;9x#|O?CXN
zIe<e9IPO;`Lw;NX;)F!3->cmzP=vr2<^kF8LxU;ul5ZU~mGo?Tl+cbDR{B@$pk#i}
znO#~S3HCJD>q_EV8Q6u@ol422zcRL%Zxm2CZPn&s_%sfoY5!7-Z2L6YA@dW=JQkMU
zjLiDNu&kG*Sv<4nm9iD177Q9ARvrQ28T%y$Rw2~Gip@^KJwexWbKJ#zR5og4@Sww3
z#aKT$X&nAJZR5&1E3PKJe$*(deEi<rk`o|sF<gp>X7h%E62m*zlwYh|BXScX3p1e@
z&%My;wjcL(i&{;w2S_pWYIcKn^A4eH=c$d<xMYLcwlIs@xUjqZ6_A#Ti-QhGKa9Dv
zUUD+9J+j<5f8U5?2!+=umk$g)W5%XJ4pw;YtRcs){);NY!zZBXQ;>YaHEJd;)acG^
zG?I9RR41K>I4{3Bnp+#H6`wxdE@DUFA2=*5I;QPV?yoQ4YpS02wU_>l?HNYw@bobE
zt^m6v3HD>+GXM#_|3<;nCxo9CL>YU5+|`lY-c3NAC=5cj%)E7=nnnWOG?mgelhP+h
z;VN%Ad&s+2HRKT^I2r243F@5qR4X@_4u!x57%q<99{H7wnsbFl{9*&JcR^W|w7)ej
z6DfKCQ6i?(n(8{K8Knmf02=`#(o&`wo57-kCq*HRzk{jMhO*TR#zi(18n)XsJ6A(v
z1ks4J$zfu>ckOr62|W(|q;pUQX3JVb3|*22KKn-$$gW-CRaR@$=umvL4lon&QKkB$
z<878(-CmT3{c8PzU<Vh>TOdW0`j2EnWv3}~nL7M#luXusIH>XHPAvj1Hw5RfIO%IX
zUW>_cAI?P(?NXT3)6#H$ySK7SNx+|b5*Wb&af60Y#GRMM3QkocX8+Jq2JrdI=H=2Q
zHU*{*Jb01~nBIw%EJ;|GoRK+cNKVYg^mzCV_D%SMPz5aX%y|b{>bvhP^lwegb}l5W
z9qJVS!F}^7!vUlfeg3_Zp4`OtOtu<Y95=GJwKUrW5;YAye&318>&gZD!VPDyIk$y#
z8{M+L=&{0yvB42ldJ~$<jVbdqwDYgTAJo^$_MknUUUuy7dLY8_N@+eew1WvWMP%~$
zxwaWf$M$58K>n;FSA6_rfd(h#>s@UN@E6Iy(k6D&v+SgT!%@?iH~;}(d7CD_Vu>CK
zrpkz8ff@?1V&9p`O##Y%Fl@hEB6u0Sr@CYzrX3t`O|Z9majl1sx%5Iop&!m#tdAEE
ztuQ0BitaBJ4uf5g{eWfIHScA}X90zcAl29+EuVE8;7P_R>b2zNo%ExJ{3jlN$kC4n
z^NhSEy!zs`ya_aMV{3uSigYLWr4+xtSX0-8w<7nu%rW5OzxV{-w+juZ1du>(y}L{)
zd8{|zpuMP0ga1{5{-arEXZt^&lK&T1n>nguwj1>D-6vFwLV#H1=!y+3Ks?3fi7n<q
zgf4SJi0IMGi5ccW;wEP+X^3a5idCV1=?aafv)GvILI+h%4K5yc5AJtwgW(ND$nSA>
zG)w=-5kDa2zNaxU*pdDgp5K1;=I;7^ad|nhcDCut&8_9js~VLjw&Ul`#ha%`(~Fx+
zy{SVCM|24`uK;)0v6Ty4r6bkLUMCp*MSkkKQ4^Jbr{02;Pk)~xvxZvfwT^7V)9Mp#
zhPN1`AH8&r==F+J+%y%Uoj4sN+zKfmi%t$Bd!vshF6jhIR}>5-Ql-IhsBbNaq#5-c
z2l2LNI0`!+Z5j`gFS3w{G-0kV8_+)45f;X`f=(Nt9lbYzus*>4(%STE3kWdFb>^n*
ztS6z$P}ga-fSLjr7;HbL&8K)4a6hKL*2qwP0k#_9e%EaS4G$fvMX2c(Ghv>o2#s)I
z0MJCcW1ggqBwf79OJ$%k@Yih#{K4AB8F&Fhs)2s5lgvl8G7?xN7PMuPbG5i0GaWE*
zVYFilsgu;`iAiM-1x1l}dIs0!5(_MNI}e3qKs<_yfEa0~RlUn>NX$yaICfevji889
zpoUCQGA&#V+H{SxG9Ry)ath^Sa{gC&-#+h#>9=Y>oQ+A@*_>`TWWZdaGveeuLy%y%
z8iiSIDY7M-3qBL5wxKY2GilbA?5yE%*UM>xZ{r`QWz*4C<g`P24+7P344Z(25aN;i
zoOopmWBA7sLG6fSSaz9JnO<Yz`ykV?xeaVK4{`V+n^*8&wp;0G_o4y;GZCd~a<Kzo
z`Y547HyvzZ6(`x;9x$ULT>mgMffx^8Ls#UxsQ@xa5gPj1eA{yP%Rxt_l~ZCA5bFYR
zMqoJY)(3KXTQ3+H#|$Ucb};oN;8L~~+dPLg<$GNYu(^s&NI$A@`^KczydKIEIx8uS
z9oW<yues>3^oVnIq+$tnw;)t~7AKcEwvseK`(cF5i`icR*#i+vtGH7k6n#v#J~iag
z9t^IKj(CkcV6>>F+^Gc4Y<W_GJxCKZEv|*(Zvolgc-!!?IeIu$rkK{|Ma$Ltt3;BY
z7!!<UPHc+4>1qHbgLoV6o+#`YwL9;%cuzV3v<(8!@{<{*YC31|{Z=)u?$^%V4B2Kv
z!H;$vI=F8j!5Sdn|M=&!{a5eE$nw9wh0am4`5y!OZwiDU9~~c<Nh+d{{Lg9_wbVL{
zgeLR$djSp5aG_1LXOQmw{f*n9LQBbLKQ>5!Eb-0t9?uJ*lG&B~$-gnYsH7J(Kqo`Q
zFyB*ZRaj^o+~8oaO%38{wlvU2yr|x7J@?#Jw{d95C~kU#>1-eFuP*i<#;lL*j$bK|
zhxpb-;{}s;J5Tqww~NOE;Ldl?RwSNWEnTfCp*<QCog?3tcJB7>@KagV5J#k0nO+UA
z1q9imAsDqXJ3@g0HFIh^DANmQ#6CQMPD@3qjV5aE)G<LxjU0u@*_ZoXme?5`pQ#gE
z>gG-If>wr_hGvz(^X+{Qgu*R40^jrOf{5lQwO6iy|E3;-Q>Ows^Lk>%UCDBqvWiA)
z3{UFuF6dE{dWwrJ!p+X}Y&9?<C@((3ARLX#Nk$cB%5F8PMa|_FMS*huYPpF9@t>a4
zPTv$4-kXZ=3R&{-aQ4-wa6A!nw8w6t_XG8eQy7=8UH|JsYm;TFYaQli9C`$&Yb?(Y
zIYB*u?r8esqP6%}<NPD<EZ~@Ut7>ZN{mt(n!L{aJ;qPaTBA*gcv8O*0jKatQ3j8aw
zA{475j%xd+zbkU*xdaWD3_?#Tl{euQ{-H@BBXYuJ=>*JLUA$&VacZ2ag$Zm5M~{Rz
zGyLS_dS~j>%st@Ik?1q&b&4z&jy1o`0WK>U_kE7PU@Qb)0pGI424yuWL@Uh`6JzYI
zfsjE<A0rsjH^hl-w`<Ei8j^N(J@te3G?t14IZck4=97SF&N+=Elmc6HCXe(Pf*kE5
zxM)b{s(Q0<HSr*%3RK3136jnqucF%;E>HEzh+T$lhd|w`@?ln5B~}M?#wn(-ag&YV
zkPhQDZs}h!k?^2`U-Bd1i`yM+0XRK@I6Tx~pxb@`pduR>iW7k@4Q&=zzvM>VMu3^u
zE^rp_#5{Kr7+R|)T!k$?w7?%5Tepl2l!%H(vklEa8wq+gsT(T|zFr8k(TEWYT>;dw
z@(6(?OY;j@EK1t((w#{sl}8*IGk!_rdLMYqpa!lN%N*39mMxyrE7QYL%Fb<<q?)Ll
zehN&DKKu<4x#>=d_x!*QpPa;Zzer(NIv7%2<n_Y04h9q=&tLLA*}~`^$z`QCkTRZP
z9JqX-wfU3olE|8gh9u_HNF~mw8SmXHgz7qeswXPo5IJA5BOZl66N}|OIsG^JntKpP
z<2mRLH9mLv>{JaV%LM7qa!!Fe_ok{zYD%1zB-sLS^COvslF7XW$&X#z+@8Ld88`&U
z_4M(#4((6KIsHZNMg%X5AF0ZMVbwTT$9|nHi5AbFXXi62+bjN6u5Ov2P8&>|d6lbD
zX|VB_sR3XA?+ESg^lq+C-6o|h*P+Kn>W)uXITqls?PsDh7B{3}+`Y!a%58lq7#o&*
z0T~Rfu_^^IA}cn};A<n!^<Zh^C0u^&Q#f&zsv}JjJ2#`MDw5<o`Y8x=Jc%#{P>tnS
z8Ife%u~3F5CDo<XWOuBJ)Y!xz`FBzwi-Z=pmo3YZP!gKgxs{+in73pKUv5hCD2X@r
zONLK#D$<VvxW%)3nl-)(l8kDpSKQ4*V67}~4`$kGrtT?VYCtCU=PqQO4QyccO-h*$
zgpS&y`!iYvl}?P3c>yuH<LS5kL%!&zLH+-FGyfww$H30~zfR8m4?p7f&HUv@805n}
z0kQrQLM#FB8Euw{B&Dg7)yo{nCs|j`{}+F{W$Wh?mgX|JY})9{AmI?|<jX{ydu8hW
ze6e(~v=oIUI7%%bAzZX($z?Q04MP|lWZyDw>Ni6qJom&Hef#O^G8ObzDYqhpx&M~6
z(R;ZYH&e;SS9|$-{O{(b=11e}d7o6@Czn7p-GP0+&bVOPDI^CTAMm&leUE(4^JU&n
zPce3%xIvgN03LAoL9v2Vt~L!J+OOVEgVxo`VnokAd}q9x9Th567!M?j8d8Ay_=LT>
z)q_Cp62d1~jQeno(!>1W=a3&au<nhAb`I3`qPAVJ0fT;MXR1jlssG!<g?i;o_fUlk
z`JQ6_Eayo9C-!^*4DnxV00|dja*LZ>!ustzZ}qt<&p&}sIil7W>ldLVvMTn-O%hQU
zvOZ2xW`fiw%e+?B&99$dGSs`SscVZz-H%~1xfccFm9mhAi3g>!?$xFCRSAxAVREch
zH;zvy($JirwTK6~aVS`ZGeuDsBKF*9iHn1JEgGTBkwCQ;3G$@uT&r%m=r|1Tm4S9O
zGzIKN>2^Qy#Eo%dQpz?R6*}+fme=1wqv4`$ci9X&^o3FtSrsfT)9mG;`l{^ADVB(n
zdwDiX<ER(BoyR{8j2A7ol66)(-Pr1;qCmoZ=)hvW1&`c_BnpE~K=AVE#JDX@W7Fo4
zUgXyOOpH-N?8G2#&<Nl|Y9#VB0*uqyyV=t!)xriVk)CENu%c6*M64bOmEa#NNh%kl
zi^Au3yXs4>GRm&B4b*D8Xq}nMD6fqTfHu8UA0m;?c8VeF4n)z2vuU6(4DcJ_R#U&>
z!1*=6L^uY-&-gpA0#+N50rbA_Xa@DDv9U-1)9FYFMl|?PL$a%LF#t!SJ|Ugc<VCAz
zZiN!t88*`8sHYrQY@QBttZ+132eYg$ZdAcHZa=#5E_s&}EV0vF>p``QiVlzT2D}e~
z50^2IVRr*#{}~&dcWrsc6$tnd5-59nBAvQWL{V(T6uyanJd^B1hQjj5F(S+xV)|xG
z+a_=II6>!SGnHC!_ypIcuwdNLFUkkVB6knU7ec!nTkK6zMqBxpw%pVnN7QIY;jk}C
z3dNkI0CnqR6(xIL9+ppip9#^$$U_3mMGAp}f?j>1Cth>w0e@vvXHMm3LcqyEvN7xl
z;<B-o1`ANIXAiZ&OLTfN+ULZy+{hY3bb+{l4Vhm&xG2z#X0_+^-63*OqwC>8-GzZ;
z7^8xEpam%H_Z(qp5Ms@k59MhLr9Aq_K!{}Xj*Rq*XUvg;$&1Ixt*g;yUhsVSu*M+)
zkOjfG!FK^_mMQ7-0PK4bpX*@fM}2^0b$HJzif&*|FvDsBge3;L+z6HnjPf^8?p7F4
z{6#Wn(;U&mWqeLTb?ip8)kcqL7o>QKa8l%KT`hHXRh#XL$Lo<|!|Xyyi_kvPL4*Rz
z`rY#_5%FOKSn>6waUq8LK~i-lZXr0I8!?QKL%qdoRN{<IO&cU`^X<t(N&wGZfYo8l
z#|{XZc@Y?yGt8Q;EAvuvE<^4tF0tk$q;}y0(n77tfZOBn4Fd6>LkIow0b{k(VCMND
z?%JJS0nZTKc<~(q3ug)BAYZt=G9U(Y!EH!mX#QRmutNRIJj37Xtzyzx-4kH8fAE(w
zs}z6%f52K`JDn0mMej<LQNfEMyg{J5Tpg_bwgVS0D><CipbvA@i@R!zTUqfRFRMrS
zw7w;o)eL)Fw9;;K;Ex|1gH()X(=u~J7I=_rIx5J<$cX-naZ-l7d_Y(`X=$CPQM<|Z
zE~&c~(USM}()9gZNYNx3qO~L57KeVpMO#SWf>rxsa|nBVsBZ)NbrM!><V=swn&U$p
zo=?uWQ@1nLeBi~pU{vcfpfLE4pn0gmep}K6Jbzsom63eJm^i@Fd#l%-84^%?_4G1?
zYc5pKnK4#(64T*E7-`$zb6yA8K+d+exWe6Dt93lkKCaS<Do&f-yk2y#4+uHF7!5Yr
zbtxpVCZ(u`d|+KdUHxHTpy656RP8}xu{4<oHx711jwPw)agYI+DK;Ds=3x@A$0MB<
zV;0HE?eXk)Sy&opvMX|X7%%OW;OM1L8<&!*4+A*MgLnyW?S4A$7)#8Vb=lE+o(ru6
zsm9vZ!gLb}KezXnW;W$>BRTAWVUx_|qj;v5L5}5mi{#D#@g<P%ERe3=(Dn*df#;N{
zX#j!6VC*E8vw8w?Jf5gy=j3b~Rwq~!eiY2?J?^Uhcc#`R+8_aWvrLHya-_qO67_^X
zuMLHMxl}gkp$I%*?T{%k-va%|mk)cRwpP!<Bbw0zVoKM0<6!Yxcc=@kmowPxBtFSl
zJ9+eWk91JoUU)voW><ke<N!a|5Aw93eQBvA^(c)x-#*wSY<u-HM(Sy3V<@Jd<TrEi
zVe9h5=L0_bUvtZU8r=SW)Zp&_Q-fz};b+lk>P$Vti&tB(SS=Kd*+dQs;batMQ>`l#
zNYqAfzn*X$$u;ghTft{}J=phQ%-DCZnPXZIVtRa@p6-{@><|$!43X959FZVV#)J_v
zN6j-t_(lYEJ{vx-C)lFv$7oB##WBiny=r!JeR{pW{tX36VNyBP$43sct+1_(zPY<R
ze0{jJ?(CnB#1XwcpC501K<5ljK&KAl+3x6U?T2N-wN9*v>*{^oeg`q_q5NSDqEQy3
zC>S?4n;=AQ+|4W&Es)UDrbaGdAW;;T41!}4qhO#oEkcM2CLXUNX9z83j7jy9uiHG*
z_ZQ!%-dXLr+!P$F-ohL2>ZcQln%uxk!%J||bLKi+gW_Oq45OV1bWX#I-$L~}{5MQ}
zIq0IgYNNBph^?@7B!ks*ZL`5CMkq>-C#|NCXi1m^GPS)?LxGDA5=e5FwXTtw-nQYP
z6)=w11)c`XCq}5AS{!83hb2rdu|O=5gdYTk7qSqM(Dx+jX{n=@LsOxDKd4IsiDpiF
z?oZ$P8;AqWQPX({KSby2)m)J|h)-+{>lo(;TGPSt1vdfTSoPT!$cFF0r4^;v+n**0
zyTHtcX3XIUf~^6NGF=r3waV)tKw%ij$J=3k-tMeEh+(VWKA_Z2jD=&_bWqY4R)lcc
z(5Y*qQ1ww;L}8@yL<7kz6g=!JyyhaV_$=rYW^BE{Q8ZBiEwDdvXQ^wEqvewb-$$&@
z)%66!#Yf@i?w)ng`89ujnJKIuu?E5OdnRdo^QMg;Gio5!2<~a{gCJTyk`PzOrpKod
z!LOCH6{pb(ZldcmC6y|=tK}eyx@X2p!YqtaNsv4YmIr%~ggO;uCnh@)nMpd!7ZzZb
zuDeSm7al7^dZ$qu$VLE>9Aik$D>Ge#<cd3qb~~S2vs3*+``3>|W+k0izA!N>3=P0u
zCqh}QaPo;CoGtJUBwQYlL$McnFnt6s)VI|yG*f1>%qt*Xj|UhpK#F4Wv9tsIi?~9x
zCtr&en)20I7rOAjfqHFNJ<<EqoUIsRoCQ3a!(w|EA^DzU^BnaJK+NUllfMpva>;>$
zr(*zOmU}uY=X`sJfiuz3c$BZd&$UTa&Zu|~;QSji0zwgb$s`khd<aySeT9Vzby+I#
zDfKy%_z8TTY{s`f)l=iOB%Cf_<s<a4)~zE4J4$PVweV_yM(V!6lJ?!}Ox{Qae|jRt
zUI0S2P7AZa(mNq`Zf+2sR!^E~rmq2*mME9(=?@nk>V0WBkF2qm4513dI_&I#2(@rK
z?d#XO3Et?aZR{eZrMLK$*>R8V`19hQ^bJ4@Jr%Y^$!t$%Hes^`(D4%PKNb!kN=o5O
zgc?2Rd$BQ{_E~TUiPZZ<1ZN-X`RRE5X(5I=&Qz8Ofi`j`fWM4<k#>TX3(z(g5nvxh
zmvdz6O}l3bu=Aytn^I0Foz-rxP>=`jV!A(BL<UI%==%_3<X4z16ghGMgz5d}4htrT
zFzrk~asZD^e&AK%%7em_XvrCgtAZ{;teldOd3`5e^~DbAiEW+d$u`#q5{#c?We-^h
z+>)J0N!QY;a;0RXwo)TxRSx=9afPiL)>!&yhh{D$MF-_qRzUcW5dq7I$l#2fV4r{5
zT62tvvvU*af<H%Ew2!snU2_xPzO8`=r@w>7`?WbeynPga+jHFeqCZhV@y=5$Nhssp
zeCB;Q$*hD9^l2QwD?e+DIgE$44$fD}j|UO{#0(o{+k@2?a_KM|AhH9S_z%K?Kd#Z2
z=Mqlxem1i1e0O$!-=5XV=(0G*=DeZmW<>Uh!>?=P)g+2b>zS#t<qP+gW15pb%F}2(
z>8#YL1b3`rr?yTXmj-L%LV_<^9q1r3cjGO>$>TYK$+Q}aXT$<|`U9o0h$kVF?HM0b
z2W4FHzBW#f_;ZD<*?;EejNJW?3SC;jExz$2e09=Uko^5B(eOdgAD`Gf{gaS^iDG=^
z+xF|Ea9%rf@Yg<euC(t;MLI;nb`Q_}yB0n6*){>C>5vznje?Cz?3@1mhKfp&6g$9I
zH#sOUbTLGh$0*tzSUX@ngJESJ-c|z!iD~-<X5FQKw;I93F8>fYlth()AKc63zbUAR
zm1nxB(XI&e&}z)v#}C?kb?uEpm8a2AKQKU0Gp?$r0{_As09oUb5z|@VCHN~kq!B?4
z^u*Y78eqY)C>|(zqtgz~ssFhFrU(=BJ`!WasGy%8Ner~|YapVBV655uoRPo#N;!Y+
z7<S`m(i{%9Gpe)D;X1$zjNEPEqijOKcVi8<;{~ATQ$XbORkhBpyoUYV9RHeM89`s<
zKXiAKLP9DXgYyuzUobWRF1J@kQg}IG^-N<MXM44oa-^~99=CvfZ5FZv@VeeP;BdCU
z+->T;;P|cN;$1(TfPQf9{FI_;sxq59ALKSyCe-|Zo=A?4|6>kjXZsHt02}N7p*8-0
z!zfcUq?~>^iai%<x8f?*ZSMC8i@3-mS?nsE4z4O={jTlh1fqr+ad+|B8Xp%6{A}X-
z+?Z==1mOa}F}}WA#$!SQiGJE&SJ&rPUeg|mCMvY|hqig6+;fyO>GNYcNuTk@8RIo}
z&eQW7y*s{sZaU?U%io(mFE?>3d|O`^S6(DSj#G@J(5qVt69YBvhR(VJOr-jQh_F<R
zUHcKoCc`Qzn@@G0*K^uC;fZDc=6AA+a2lOA4>pxw!fJ4EZ>w{r>T@?7((RYH_doAC
zq|veuEsS7I&Hpd<-a06*@7ou|AtXTX;I6^lCAhnLaCdhI?(Pzt#@$_m2I=7L?$+3J
z-o5wLoA-WGbMMs5)Kty)`~|1`?B2^hYn`>$-rHh`YjgJ$h56Z9zr?#CYlRVRp_TRU
zqWsm0?%~5Y0D#!a5MW)>fF`=v`j%jMfA;)vxhYhRa|!XBIpC2k2h9qQe%L{MXSq8^
zl?VfAKA5?+{J^~Xy~$S(A3Q8ee=eGZe`&iQU)2#sxzu~(scS*RI%#V#Z3a~(g{+OY
zO7N}kAOc`g5+ZaT=-Q#$K6%=Zd$p9|RX}3?3I6Pdf?=`bWX<^S&zvUd+Uf4Mq2KzC
zkF$Rs2;@AGtF-FST<s1Ns<EARs}Iyr*9&(u3@O$JF03g?=?bTKgn2%gXVCSH^$BpJ
zrqO;Ch5%G@@DpZN`FXqFoun9;?WxklV{0}^yS)NY1nY_dd+5F*%KJHed@h=B((5mi
z5iocSdyzf}G*+cG_H1*s>DtML4&mIDHE~FL%=9064$R2!th8LBlFwq?f3yiB29;P=
z1<w<_0Rt?&-<=MR2@>*=`ao-T+<!W(b7$esY6q%Le#yS6CZEtsRQ%|ikp7&wbN+^d
zR6jh@y*{~q+f^0B;b)$>sg_R!+ec=O=HBjVE=bpfW0s6=4m;X5=8R*h9BgGG%L_D^
zm{z!}Ge9;M#mApegS4XgevDI@_I)_5U1!4O$OAGVV2^a&$iO|R&8I%zFfe~E+>TG5
z4>28+PtfSTwbKplur!{B68b=;@)~jhF3Y#e#o56czZsj6@dk<^;r(b_<}x%Wr}lPV
zH$TyoQhE&8)~uehSDi$8_AF-frjh%Pg1Vk(!k>z-z2i(%t-abJ!Ojdai=`c$FPscF
zMj4LR62qeA1BnykxBhWvy%dejnnO^|HQ{2xa>*?FZ?0m$NZ~ER)1w(|GV&ob;q(i)
z#6IMxg$(ai7&hv|7OD|{@$lqnzB#pnkIEJ`Oq`9C?|FJLz^^O_^jBhQtcrPxw{UZQ
zh-SSboLgM7C6rrx^e6Gn!p_OV)Y1nCp1s5qeb)|G8u+R|ZHUVxA=-==wv@x;#xcUo
zg2sDkAzF{Gf;N%haOMza^7;V?NqWQ;XtI12&RVg#){e^giM<1-yK>VERt&ZVLGmCg
zh9%dJ&^vf;+$0tq`N4I=&DNG;$`dp{6~`s$RtERwPgt<Z7J}j6=Zpqu4?l3SijxcR
z63b@3#UUqb1ZU%%GdEb4SSEv6ZV5xvYLrqP-8u}CbKWV3U2mlYde63YKCr}}Hz4GR
z{@9ezeFqh(#<>;c-6CxC9OHP%h$!c6PH(81l(Uwk?<NP{2b=CIM&HlA`4M0(m5=;*
zBWqY&*MmU7g!M#NtIJ;PC+>L|YjNT;OA1ncnNHagjtUc(;}#z#Xv0<H^eevc0CPPv
zX~p<nMR;=R{LiUNOGf1H6<Lh*WqWcSyn91Mb9w}qpY~_ycoDUjQ$J{-N^_B8Ua6os
z;~_$S;f``?)&tOfZH;ty1d!|J5$eE9(FRden5YutP#bV92z-ejrdx!^_wyZ?!@L|Y
z`?}ep-5jtHb3S>)#YTl{TP0;<vkLzwR7D3PybReIf!gKxv_7Cx=J<Z~B*2@7q}ZvS
zIc{ebm1yTTDnkd=Iwy(<WBd{FUAGR%vN|k>1Xhr{QuFCi5R$xt=y0g8XE6`~LSDe;
z;cP!r<Oa)88Ud<!GlQ1K(6{|n=yKXv&>EVJ+#h0}vi`Al|0;}?C;MrjVFOmGl87P+
zr{=oW52pNCwF@Avg6P9v?C^o2Cj7ydq!?!P68{(yN9Asu&D)!lP(90{>qAfnN`og;
zmhUd1x1wQqaKFiF?h|g%CBVGX1t|Zac6;|55#z@2b^{t>ZumaF0J`Fr6bdo<N`3~1
z#0uHS^!=UY?opS%HLn0cu@uPwSZ*xoW<=AYoxMtZ#NWNY?yb`dLu`?M*&<L=3EOd9
zh{<Xe<#q9fBCrUB7+R$%{_Lk^_<reHOX`gDkVsYVW^2w=wp>gXPE<h6nv-U9tD?pu
z%wl_{=fh}(WX9x12&M3qPV#zD%`@2f1UdEb@H9i+t8RHALY^ew_GlR03j+Yn(7gXZ
z-klv?MLYfT{0_5t3HbTw)QLzvC7N3iVm&obn+9!L$2Xhmg)DsR@*dNAA9aHX^(!_J
zymhCfyt!x=pfSv5G3Fp0h}KJv0-IYgNqhTMUFoWFe(#8(hEy|G1W04dGWzVwx%G5j
zBF3Z5z7HJM*!>I>9Wc@)O4^uO5o6*EPtXqCt1vo9DR7?c(|hVHoDo=bIC`?Rgsk})
z`K%Eo?3{7IE=a0|t2MULrW=B{_^!f;XZy^H#B+@wIE_BlR?&~-G|Bv1ztSF^gjVA_
z@su_wkDt!(VXHH;RofLNgRi6o%boddRQ4NMj724L=kw#T)@Sl--I3&}5MSNkzWtkI
zOd|%X!v1G09859rVQl&2KcsXvU%Ty44bz)-q+^r-xHzh(whLQ_>)=DyV3McQI;W=^
z)wi3cL0e}M3+7$f(U6e)$3HSYDZHN-a60J;W6^#e%g?$!51|#{!rx8a(;n8*R9k#I
z486$^ItZppHr6{?MbXcm$cX!9W9yo|N#BQ&Tn!>0M>PQ)J(;vKMB!fAZ)iY$Vy$Sz
zioMts5115QUd?<}8h8uCoBASSp9NwnG|pikoU86`lGrDET-y<gvF^o9t%xsbKIn2%
zEoV{6#lMU4^y~MtKrf=4<Ws-T_+X|hy(*}}gCCYv1#313B#Ec$Uu4g+e9!4^InQyh
zlUEyT3IHT#dz>vQC_0IJD)RkPE!f+R^U2t&oVUMtEcsm)&0<G966%D5Mmmi3Y_Jm_
zejzEW4YkK<lc<^Bn(BxvJ3I9`YkBBmVWy7y<0gIF6Y~J{Jo@BUF7f2gwI;GhnZyO*
zLZ=@SS|`YfXTw1#uDI+0)k!K1td@}<6C9<GUAZL}IV!_HNJ3q_K}cj{<1C2DqFf0t
zP3sdQQ?P4c&}vGF_07drhMqOxz>A1Zh2p0eu(iiQ+j+Dn0q@%W#Mnlp1O<p{(kjtN
zZs}AsfozE2K-a$&JBD@w%Lp2HQeOK%ry6!MJI^d)jLHp7cq;WIL;sPHDud6Dj|$(Y
z%%P^W_3bU2-5(5eRfkZ6Rr^48Sr0ywb4O2RkM2(P%dZf~KaDuat2{P{FsA2rr{0Py
z>T~Ljk94#PhYvHXFTLn4PM$o1F<c+t_(BYvMT<OIg0g*pCkB^u%WPTDRr&W}4t+KD
zJ=OC*&zf!hu7tM*5Y-y8HlN`OzSs4|Hc=oY34lx&pMP|a)T8J;2+gKjy4zeJ+3r~x
zdT^Lkufzt7e<n~V%A3fA2~kI|7MO=7-kxjy^dz39N-Wom{X^<(jKMJD&#x;It&u!k
z<VKzjz5P1)gj2<k4fs6!t-Etv!=vpi9^p6Hogc?}8|!+nL}3rNQU4-B=KdGebrv?}
z|8#^5)LgS);X?DD2Je@p5|BBL;=8_0=p@#1Rv?8P<NaLAj}bk=)jpOwotS9$4$9ez
zPo$=@tLxo%VB`=C9oU+PV-<{K<oMX_;d?&~TtX2XvQnxgO*#7R+M#$v8lf`0V^XKh
z=oh8x_jUoC7JMQ9kuDjLT&^2{`1R&`a)_})gJ>#^WlL#nDotvva6B!I(Q+5GeVyiR
zU!$*bv2*v^$Jz1ee$5k0qQW^r;-Y3<Uw3A3@3{i_)vCMm@$#HM(_uJgk~J812W|N2
z8=Y$Z>^C~J1Xfh+C?J-N&Kizla@1jorB$K(yybINu|Red-SVwY!^tX+XY;7ABcSqo
zsajFCV{Vw0j<3Cdy_|cQwDxnhoU*h-38IGJVe0AclO`>d2B#7|_w`Y4XN!tqHxu%V
z$i9ZU_?a4aGxssWPb3PO$nVZN?=u8mH$VEaG22=+^M|ysjy(crBnzgzSR?uzx-Ke@
zn3mnlHq?8b%X9GD&VTknhb~I7`L<l}qSfaH)aBJ|7_=qWHK##_7&?7EazVw8q)<-u
zoIHC{7G|r`YKN{NK@=0-8o_bt&&*tBu<_s4{UdKu+ET67bi45cT;G05#9PaBToM7U
z#eB?T;>MPRN~ETJrEiC6^5w^zWHVDX-O|REyU%N0$_dSj=N^wF*G2AAl@<%NO}8CZ
zGJQ=G_tiP$Sd@KPnzOLAb2V_E$011IkyzZx`$pC)Wg=z)S#D(N?Z51kLQy$dPc`X3
zJYqR@O-h&$XGr&mAa3!D@csEGz}jjpn=dq+Rnk+cEHC!zbtaw}MkP}5!@5<7q}Y58
zOH*S+AYIeDODX72US0jljw#$)p+sUb1@xhv_RR(%E1@p^Xs}mD%_Jl*+WekUkF{m!
zdGT$_=NlEET6>8S%N6Y8sz#<b)$)&ys}1=YY2l{8T~_v_dH?RJWAj9yM>nYv3m@4>
zz>p?8A~#Z`4w3PGuwV*nM#rfS(ou9|oPrgf)xBO*m0wGFi1?UUSl`kuJT701F+jFX
zwKgYHof08y{4pe4p!u*%R@~p(a?{nIDo52?x`jJ>&d<CoyMbH_pCjt$>(<@UHYb#F
z0|z@jg!KYub;|=D6^$HOC1{}Q+)D>ZZj-mv%)T_!n`wwd)jNAHDHHu;-|YO8gI<&A
zM*DRL%{$gRkX#skh!oi>JP2m>zQi=yT)q1<mqgzSyt6gJYsajd{;{6Ir7F6?)}M__
zOv1u1aX%x2XwvIPJVfx}PAfSlA!nac+r51Q8W)5E1)RRv*ms{>>z(PE;vFvnM}T)u
z^ItrmY#1&k+G{2!7)4UEG}Fz6XkfG!?-+_8enqG?`>){R$I2}rDNhmF=?QFSKbi2r
zR|T1*P^e4{!<JQ_><7Ahdx-M%v9J;PGw!YTAa9ikt&Iys?B>>3Zkc~Wo0A|vWy+3W
zak!x;%*v4U44Wv#7oWT7;Eysujxm1YQwJjtmliB}N=1`|Go4LAcNs&?cMy^VJeo`E
z-lT&irU{vt=5~m%iH)mlbY@^7xMPgDnd&aL<%clRONGn-CJB!t$u+FpCyYWA#pdP%
z6R*}h!p>sUdnwYuH6FLi;vk<L=47|hvX-Dt{Hq+sGNNe4Xl^9|68lt#jX|c|!&DNN
zWXS!lkm!u?)68<dp%df+8xru^xV#@uAbxhFQb{^KN`%>w9EOiKfi$^aR+W5aMNdjg
z0$t(T_=g^cd}bIX9&VoXZ=-p|mSM-0;-#1Y%&;D^qWcVuJ-NbTs4hr`Ul>9M?$j5m
zmt)<vG+R~nF>+OUsU6`zS2lY?Ep}-Hi+M{Tn-LeA!nWOAMKDp}{Kz09at=8~SKzAZ
z80w6QG+M8NTxfp4((!Kmf`_?Fll5lUmQP~x_NXwQWDK-gi>G*O9TO)WK+xn7RB~Y`
zAc%!9FK<2!@{Dm>PvwqhG&m7F91C&EEcd-FbI36XfsJ3um^x4%n8$)_`3gCVM(3hC
zpPJQtML*h5Ey|%|o=kr#;`tL^1)rbI_Y5EoEy2yFT7{DxKb0L1X*8x=U36+p<mCG_
zRBmg5;PCxQ896KX_*f{3(^bW91~06IuRVKFERG~uTZ63ZFvt)#<aeeBkV_5^8k82?
zbTy&v15<on1@Ae~%HIr1z%~4i3Rzcm=*7eMSu^sgJ@)RlEkHV#<z*iA)PpwvDxL!k
z3){1mX|uO^EX{RpHm8XpH%}v?A8S)6iVI7G@Yjywsl<UhtLuj!w!EH#HzxawqP3(`
zul)PhhI4Fx76x`kzv#jW90FYTT8=}@+pHjALq(>LT?X7ng1F83CPjolA`~Ip1=E)V
zJW108LKZL0VEw8i?`%!<X$}+n8u#XkJov@Fy$~M+x9%)~i-D1*TXwN1Bkb81i7FXO
zBoM#tB3jIE3kXPNgy$ck;{X_zJH-vovJP=kF%ayGgs~$njzdE}Ckzu5j@6a9Vc9S(
za<~ctYE2=<@Q%~my;>YAR<CV51Z3U5&@Qg3BZOZ++{7Zv_Ow0klTlp7Ee5uOIR9j;
z2e%NO1Fok8yLACOGsdPqHzjp6@S-h~k6o7f^Z2#+18}L8%;>Ntvw6}GiJtpC6pA^Y
z^3r-qBCycXlKmrP>~7tj=^1D?l?t#5B(*5T8Ok>+v6I4J?#l)2K-wr%jXSpG*1It@
zeB})E1OrM%#*Z~eD6>v`(p=@=>Y<4hH_P;Tx*#P?X}zvs=!L=}_{^BG1*}0A;#Qj?
z(xwWE%8+d;gIiXTNWT(vNwrASsKbds^{Yw*t7Grh%XUIoG1{a*UrPDEw6bu#q@vBt
z<E|&jAh{~oGr<Z#eoBxUcJ?3OS8%r6Te$h=wVMF#i{{pNE-cubJ5JCuCXSoIMfr|R
zd>*xa(G$Z}{`(-(RE&od86Sff8T6x7+dn6|53hBp&vf9jH)#+OnbHpblaGgZZvrI>
zSBgK%=J&pB-8ZdXISprzU4f6S<(uy?4vN}8-kY!A5n9h#1(7&6p%6t#H)%3W$ySS^
z83L9RP~k-eb|5?MT_F&D5WR#4Tq)H_;7@1|_L0i1eb_F_CPnsA85D-Y2zJi(klA!@
zfxyZGTsB1wHt=ghMDW55bLbI2CL{}iK8)l=A{q1@(wCNb?93V7(_4XxZ9OK+aMm5H
zLR&7Ou?wf)A!so(N^8?gW!n*NvGXV2d$e<2AvdGAUH;b?j{RSqM9$0!?wkD2p4z4<
z^S9gQ@9G*QxbNcR6w=Z&A_gmVGx-))`C&!W77@)Mhg;I~JSzqpBw_OdSCqmGsa%;P
z<^sq)I6R24!Wb_;53lFvd;G<0qf!j%2yS5kQN$78c13dVV#F22_#ZaS5#jw8e$P)2
zoG2Xz`u1<@ub^vzeVGb!Qd9*aW$QubLTBtY)9O6N3+HC0^=mE?CDI^z8L8Bd**i6I
zxMB77!#8H*iPZR!Nwg!{TV;pK?n;BP#*D7OCQjW*$y~g!f`A}z@(9g%(QK3aM(c?e
zD&V$meX@DX+@d3UFYA;DkZ?~ag)urg)JtR_N$2&t(%{b@$!S1YbQnOi@9hH>!SJj4
zJzi$WNS#s@q5hYRK8b^JgzfCaAA5(zmDY_H$aY66YT1`+;}3Q~XHM02uUmmeZ349m
zzJ&89>JufT8g)E$MF{foAo@v!aT=)i*gBJ9SQ@lqVYpF5yZGW{QMLzfy{42mb3sC6
zKu_Jr{4rgNWXD94&jw>3c5sGU*3#Ed+~KqcM~rwmywgV!<wnKTEkio^^6_hJ*}ISH
zs+P$Bmf6tV)~@0+imwMIvv>isX!eknP<=!DzrNp=<K5-o;@)#hjN<lHi5Sb^Y_Yx3
zA~34yxTUQeDxLF9qLbEhh3`Zo@C1%*f9f-Kv}!3xWS8bo+z*a{GLa#lY*mfRX~sd6
zZWkdjJl>z<a>U}0eo(uDf=?UUC->Rd7*6~sNg2qHy4G6rit}Z-Kyd&fLR-)_z^z~5
z`>!5EnebXq0fN2(8)*sgZN6}luQ$<*5SXm_tfBoz)YVL~1CBxvA<#@nTQ1O^HvF;u
zJL&BcY@`#i9h|mxiX9<TFw{SGpjSWqxY>=n*Y~7lwNAwXnwAOqL&<Yab8lHjq8ml?
zHOs=zb0lGo#($mQMZtUe(ci^t(OwS~T>Pe}F_KiB_q)JK30&U&N*_Snd>NqGG!XuI
zOEbuEWs%Hw{$AQZmQSN{ly%W;OTO%l4|V(xMb3Gn3qFtSL81aS>x53!XNJ*n5$Z2e
zD$N1$RE{2p$L$@&sC~aNOOWJ5N(}))p18wyIWe%M)~oQ@v>)E*XK6Z>C_{uLmSC6*
z7U~I1kw!lE27Jp;m6>;{v1$=8eu4DcjtlzStT+-uwo!267_zmh)v7!Wb-$>a!|ZQC
zDjlG?Fz4e58xSr3MNbD${TDveaA3xua^O!xy8E5woh=#N{Y@5`8r?UH9eE4KGK~)%
zM^)LI-&#V}Zn0MNi)}yI!W&Mpev=af66Xku3Xre-(hW07x>uL<6xF;p6;0dyQId52
zbzaj|PsR^|$Wp$3utgjXVKncfWL@5ro+C@H66(4}{Bk7ML%wB)T(x8LfhqTy`?$cz
z655gVT%3kMmrug$m_v&ldvs}v!K})m{ag7zS@8lw^!vCJM2_b7cOh4T@f%C~+c1A<
z7ABLga%wJ260s?~7Me%9E1~iCqWErU4k^VPe`T+X5Cw19=}<<`Oj!t9bSWElgoT4u
zQJH$ROVI^InMNRMBwQ^XMQ`$(go)DbhpA354ITAXYrbB>=EL(;6Jk&2TOaVrw^TvH
zH$ZPW^0$b}ER*D%E?p1{ZaVOB68jfM>nyzqKbz(gfi%`nljz;#s?l7~vE_QcS9>!)
z$VlO-p#VC!I<Vwpp5dlX5a;a&v$B*^ofdzW-05C57q}{kB@MKX9HTm>D>%JN(=dA7
zC{+4ClAtd3meja3#~}UXwH`NR;+}?ei!=EGT`rp9gbNXeezv5SQ~~=eKdBv=q!+4)
z+At`g<1*}h5mMh&N`fxo8W=m9s2O9C+1!tBt6$AxFkG8^ZR~h{K#oL<|G_?n6G)02
z$|jY+QzAUG+T~-Rd1>taFd~61<}C>OCkUCEQ#JZQWei7`{Ceos1t`|NHcK6JV&!`Z
zlFrMVuI0mWla`y~S-4Zcz;NJ`dK&xYx1`O6MviV{3exA-(=BXlvQ495;^P}1+cg=+
z;`{|ExJB8Y=Bf78{VBWc73!B#6J<qq`v!kfc2sOYe=(N=f$gVy?uD_v-SZ-KoO=92
znXooUjp?7u3-X<<CIaw8(D#_6dNtlwk>=Hx>nhY)<WbcVX2`v~%vs;u!c5QWnMSTY
z;J46ww}v*N6SESw9W@i&b*8>PpQD;H$k$u-Fkh?^rYg8f(UIbi1DyaOcfu*}*`|zH
z=A?JSy}nY3)zNt$<BbC$x)rhA_s`(ATR!f>t(sCjtQ3~63_;2}TO8I`G<*F{XOFt(
z@Xcb5_>7)KpU|NgP^${h{pb9x^QFQV6t)ZJLRUg=I?qX`V&(76%L>yu<<FUr_09o#
z_kzL&H!tzOx=%Ll|FK#2-wu7zRQPUXYV3l@sN(wF#nZu*m{H2s$lMf>QNz;Y?<#*b
zPHsd-NmEO63m0NmE)GP-Z<a33il$B?_O=f8cBXbN#9WAsBK9`+PAU#Y#^A?AP2DVw
zO_e3Sfrpeca<(RBXZo9XQU33Hlyh^j{ijznoUM6A=csMHUp0%+xD;YaKV1i71!~)*
ze3qoiSF<7gTuhw52}c{4_~kR=z6@I+LPRpL;%otR6m^tZk1fQ>w?)JsLlB5iB#KyM
zGB_y|i?hnoR}Vj^RzI1ikDuLUd+$$qcD)gExn~@#KVM$0yO+$pgO61Izwrv_JN+3(
z2qG%LWw)BIb+@S-!m@pIL#L9*zpt)pS@%*DD2{>W@_X^1(xk!_@bP#>j3wax(*SwK
z>2-Su>ghuz6ZhbNpjPB2qONY+$4x1mB=pz}Tyx*>-Ji(FBTXc5TQr(4QzDay&E|6a
zj@P~B_k6~AtOu0grp}uMJypJO`+{7+V*}o%D+I*cmrSvz;TkuJq>_$Fa+r*Q%%KCI
zQZ54vK!%#Wgq_EL3e{MSj*IT+&`&r{eX!UbwBNbk9(O8AT}~J4;0at5!9P$@)2;2>
zq19lf_O9FQ^OWx+9fnNMJ}vO9isEoITP$BI&t^K7T&_~8UbESDkbANeaL@a+n^f@*
zK_#1hd8FXK+yE&QDE%npcv97(j&eCG!+7V^l2L}1Q}!MDLpW`GS{ankmMG>ds#*cM
znhi0e66`ga8U%HA**Bg7X_taXBywVakbVIoL<qW?q=agOylPz_j40-)j4EWlLjmGZ
zIPFXXEk}Gn8I)%PEd++VDhYuS1&`7Y9NK3!Vn!v{fc^;DA~Jv|6<|>assM490sx`N
zr4-^KnsO+Li6<$f8rUg{34urjh>!heQvY!3AD#Ng`u>50f57t}H}j7}{wGTKe<7YC
zKvm=D%xl^3^EM>3{i;;_i)?ujK8p>&CYavb#Sm->P1b!&AgbyZNeT$SWG`Z}D3E?%
zXDrr-P!dx;8s8p@8i^wk-0c#N{!;h;_Uvw<YIz+Fx-y&-OWB8?n_4jaEt=BG*i{y!
zf}sS<bEsEA64HmN272%-NBT$%2&bK_(yS|}sHhNl*q_XL*w3;@UKFvtihF1C6vewU
zRC&M30}qqCU>g70s9%|;ui8fwSq4S$Pz{X#Rw|Jq=h>6&0x62ARGczxdUw=fvs54G
zlU?;LpF$Z|b_c%_x=}v+Vz_+S{;69K3TLkvTg@SImQ)EAZoWfZ1o13w8(QoyF**=R
z6lutjw5(<mWePnV=k3<SMCK=dd13Y1*G%%47aW}o!HfV=&m-QGuHv$73<|RU!%?2k
zJM4WyuTRtdgrg%%UVj;)?O$sPrWK_Gsb+jLidIP<giH$H68j~i8VGs!mzL+^N1>UX
z2bm)*A=O`Ef1~W9aHQLarW{esQfhEu&6L01gsW^J`0pl>61ymrr@0~p7E2;HfQLT*
z%h&t=2+;p|^#6=arNISb(KxKfa~IVe=eYmZq*NSSAbr<XDe#&~ufq`^*Kk)hyNlZN
zX1MZMEt5zVPd}{>Tdkn;Hs{Vm3@k&O|1QHG3?gc<{?n{Mh10gVNUM6jT#bYw;i>2O
zm1aAQwY#3co&Ie8qL#^55aU%_UMPJNY&m_xGFl-0zgPeM&2ZXR@BdJL#vTx`W*Gc3
zyW16tLiARy;0q8blunYEkAu%*_^ZEe-3-B$fI;6QG0g0OaN6I(7urZPpJK~q!AfVV
z`PZ0n*tFkw1eF@M;@&@%Ad!Fk_ZLCk4@A_WDZ2l>uKn|3`%enspJd5D8JvG|UH@bS
z|4CK;lhpnHZ2EglsKM_Q#BTlPQ>?lE|D3QN6PjCF)9Yg8KB;Wb_c|(;_um)$pEL2F
z*w)t8MAn&|c1O___^vX{bB|X4<w)VQP1V)ag|R%xf$)Tfga7I~=%Oj1OAY4l*J(X3
z3q5Z@;b|WNweHIh#tHJoV?FDf<>qCp3J#x+uKVQ*kel&vbotxUOq`z!xKaZY!Q=OG
zxtl(2XhW|=kNLv<J&L}$Ouvu7eRbw<$;NSe@K~^Sl4GS3B=~U}4g!{T3wp?|<4he~
zAohP8AaHE-K5sjSFA<NyZ5lx4^#Qk$t$1H_on+zr-p+uRnyvc3zk(|{_yRtUilj%L
zl#NfkUeZ^HP;YoJ0pH(i9KZ#W^>Xyn)}2@p?KvQj9b6*%fKG|;&<hEExKguj`$m-A
z7j$d8!3?gzfJuNx;d5AVpUh;p{%>Wb)(yXB?mnKGzttZNhiU$YTN|Ng-<KVtns>Za
z(=<IkJP3@kY!nR6bq`|+@O$ohGx&j>Sl;vYlWLtVuO`7!IB;bQquNsyhr?!ZB7+rN
z^f~Q$y90NBzc!bC1b>##k?yBN5B@`F<x@$IGS1&ywzx{ByQ2wIxgOiB(|kMJkl>q-
zLST8KQ_6Dvo6IS|yaGQH^{2+5$7~XJ8XElHf8lcEzwCvUc-Z=ZE>C^nso89!^;*W*
zx(Q_t3m)V$FxU5fNrs_qcY<{qTqE0#cc0AVf9|}Q>Hr;AjC&I9Pv;p3-Up)yENcCo
zTI2ci#=9RGfDc@=^3^-L@PB^>zX9=FL2?K>Dg7PrFfWRsy$n@N-}|g-bCz?%=LRf~
z70-j53K;{Rt081BU3$W&>%H;!2LcU!&*G%NgIcUj`@h|Yt9m}4wFJv()ab9ZHZ~MU
z9}sebKm3jAcNJwYzh`v#`)3y;rW*Jb#JYxo(%#$ig+aZ^FZKo9IZ5<11HY$4_t&?F
z9{&w*@8j~)YOsPPDF{4$HFz3}^SdqF8&B)G`=i5Ob9p<b!01^6F5`7{TbwO70b97h
zH9QpV-FV)M&W48^!TUwyk63hEd&w%@x5bRUcNGTwHxsP(ty`h%f_igf37p)~55Tpx
z*D%4Ct!(G1a>9kxHfP4}$L(W;zf*tCoV<7f7d3fyqFK9N_85u0PwS$vz^9u6HF5&~
z((@A(NlTUO(4Wq1EFMR=CW&4MR`{E08kG{5hnrzMd!Of(jp}N*UPyRdu)U}0I{m8D
zYB+vhVkKDh*ogsZl*ncMX<Y$v4z4?n5Ukj?=<P%az8)#yacZ^Lt$}U7?WDRx-QeQ*
zcp0pC^1Nq_A}IVWi?v2T4YP^#tIc2(yVHaR*DqbR;CpW`Pgy-XwFAf<*JEG7D!+Q1
zrOvrz6h>gz{j|Rc@?E<bOZkezciBsF!R~pz`+U)}A%y@YB8Pg6=-u;v$^2p2y5^W?
z<7J-{Pmre`3}XW*f}Nmk34!Ybut@Xz<dFGqe~UY}KJ91y)rI4YmL;>Sb*18tj*ho&
zLhn<hSMQw@&Q%8y@YWUXt5+!+`bQfH4}8(<Y1O)I*XVhVq6E&V40_9tUUy&O`0p1%
zQqNQF-E+^Sra|6*k6U3aYi=u!JHt6Hvr-zmjzePDdaC@Xpjq%;`bQ%|@J#c`L?dXO
z_b1A$Yk}%IRay-t2Mc*%oH@3ui2BQ%P0G?-FDu^`a4IRC&2qNW^G4reE9Ce@g2>m!
zK*<!Ws(m^0f}d2L`cV9S_mJ7+4?l~ikj3%Ucf0fj!S;#ScisIl7kpj<&4Mex?Z5b^
zv$>rQa@?zP19Sb`l<!34a(J4$`@w0qoUp&bcjVv)ta_bP-C^!e<*Yo<f?3|sZUF{>
zK9q`VuqDQLcW+02#WfIof4K=RyZ&om?_GtUw9ToW=e3JJ4-UHt68Af!-h|>8Wlkdm
z?(^8T;0f4wUWs3<th=q)-eCfaPhOwZa36H}?@wxax*sO-;Qc{IrMxI$OUvJ}@lss^
zV`|n*@9=@(`_uYf5sUxZ3+SZ5Z1VW&?ynVrRrjyjX>&h*WN6LrPCD+m{rBIi2LswX
z_GA`kO-&Z3-}3IN^R(@2t~F5Z_$$8SAnKF|n7~7>tS@SRs&p2BKsAG{<`ugW-w-gC
zS9agdt0<g3-J#eZS829782M7=J+B^c+>d4v4TJ7Z!cd6x4P3_QTW(!%Ja6`a3+04n
zQ4g!N7BdAPL~vz2{^m1yaGohj@f5IN8XZGEGIn0Rdu4DtoB1BqjS#NfguceoS}#=V
z)qNiVgO$fc*F&z;7)?aM7}z7uw0hp1f!d#fU_Tz8$Yya=4SCzOK5j>eVg8-5E_lYN
zbXv2m{iqW7S1H}}+KpugNSqviPOwdYkszJJcKNrv4Or9urW@aL_b`<9+<Tt&cJ}-1
zH_j}<Mg2S1zu8oyfzTMQ0C3?Q0>Y(HE}AwbOGEobjJhc%l6C=%hN(JsT6&(dMN%KJ
z8FZVh)L!pg{9g{P#~Hyw8p6_H<E|aP093*FS2dJDfrkiQ0n-Z3Ox8Gt9OQae-(9SD
zsXpxxy=%UEB}pYwy1KhRbW-c7YIXZ~8Z;G^T35G~bWgQ9?r_{gM$&?x|2@OO45-v@
z0<;_XhOF54bp5vxmcacFC)Nx)&2+7QA8vCp-P!jC;|-WO+y~hz2Ik<l3A|^Be`Acr
zWZ@WCh28=!znA8?0cx1Qn)0W|K1!+h>Z5VRfHxS?f@QKqQ`BQSpNzoSxsq;+Ux^gH
zYcDz*Aa&my<5+GpW1{^3Mh!vL0i)QF?~xAJTmmh(Uv8#rYM8)Ycr&euW9GZ+(_M|f
zo<UVbg*6zE`-)@AoWSS@Ry(kAaCTqCfC|8<^5Sd%3_7Wa!{e;(o^oB#YpFx2MO8!Q
zyA(Pg`OEU`yq8_7$@k2xtDTXBG=bOKxeITUlZBc*0beiQoN<PZ%ApJIVI1>$&J2EU
zx1@^KJw|^^{B=3bH4NS5aN55~V8OqHRO+@l-5$;W#j<Rhm!8cszSeZs{}OrZh4)zZ
zJOFB>vs<VByE6U_TEv^cXo2lzuvJzV3t;R5w}^n+lm4<BoFZ?Zy4)H_1B2Hy;>n*X
zZ?NTh`Rxs&D^y;%fuYR8$peUl*P~*3JYUZ5k9{7A`PFPXw+oD1YS^|7lN_(S*ZW~J
zk0~?;FWcC`ME#y`PX|;Axw`KNlYcR`?m?FuOykK{**t+Y&%c;VU~+d11be()uw&c^
zr^VLyFaj${mVNiqjg<ob-TY_**@=TA*f8wCzz>dkfY`y_9bnyZTXS6)CK`TPn)Yk<
zdAbHm+;6Kt#Q9+A^>>(n%e;~VaN*^ze_k|>JL`Tvze}m$?E!Pkz;iF9`|a>|-rH^I
zFsfY>P<;g7dE%k_p^MvjUv;Fx<7%7e{c+?@49MwqHH5tZ_FY(UohbulZ?9nS6ES}m
zbqzUI>U;y8+%3Fc#rglWOVIZ2uO#nxuXU8p1Moij{9wln4kI*tLyq4rz#$3cG?+{X
zCI?;r*E<o=?`L~>Kd`3<b4|~H$L(}+l5^vwvSI4jH?KmYM&PI@iBc|e{$`zH#ijw=
zZ{zoRV$cI7_w9LI&~|?1`R%E23@rS{n<-w-YH+|-3!XL~7<GYSU>JGbr4n3fJzV#`
zu-zNY!XIAfzHi#>5WF5G%X^wI{~Mt0rRos+T=sz>;R}z;;a~*%!BoyO*crlVy!Co_
zpVp6q+WSeu@V6&mARq$z%=gAp{`x<lhJo+HnmgxvM&w_t1!Ka~nZ5t>C|O*kP7A}B
zyFriNYhj!KC=2XSr+6>^;)KHy7{QbXG+%e9vz&J2R5^9Q5SK_HW7mFEa(-@i@L!Au
z|8nrZdI?E9zSGt_RQ|6mbN|-7X8+v^k&h3N@&D15ISD5t&;Q;e$Ii*|Z$0h*i4D8Y
z`XG_O6!+)9mD{7hWBgZ%|L6Pv|Kvb#Fl6VCWvqXt^Z#R;=)WKL#KO$R^`Bj35u>T)
z3|=YgH(m34WmXMK1i!P{iioe=B(a<;qrRyLg;hs|>4Q|X)VF~F+>4qGxOU-$#2ne9
zKrUP%A_Id+^$xX0(9@(s7N@<SWZl>R-_|dy2E7Oq%yIH)lg{PjiVY|ny;t{}@fDQl
zIvh-yZwbt;{WG1<9ycfZuWP`}5vBMa8Yf&4K`ZVp7Y=)G`}YgedEGD34gMPn196-!
zc>)H-EQ`8zbdR9Dxk>ER)>n|}4Pn@Y-MN5&6J2LS_`%J^3JP~#U9%Zoas|7>0G&d7
zk2etCYHcTEWr~eU`SIv30J>r5;zDWgeU?HV8K4qF=(12qh&2}3&j(Fdfa&ZNC7Ry%
zRuaf5*k?*NeMWXam#)0PMHh3v;{S>32&KKOd0KR3cJua--J!3*aPb679F=KJua0jq
zCdLlFUN*U!#gGPdie(UGTJ(!!n~5}#cTR1f`)Q#W>chLkMJMq>=J<S$J}V(2rBV#0
zWv!BJIouq-0VH+M?_BrH&-$2->AN-*DyU8gTjN<>&0%}GZQJBj=XX=`1m|Zi6};^E
zOXRpRFjWFmmX22lq#2IAG2d&J^?>IN)|VGp#tT}>3CmNmROso!ov{mk0p?rDqo`;s
z1=faJ>aO4;R!1(Cfl$k^7Df&3%aF6*np$6lB!<@`J{I;7|3bA9%0kBGkBxf<ERuy_
z6LKPV5y~{U`7|(nzlwIu2>Bq=-AAo1!-AYki;DqpYZ>@V6c@~U;LEfaYs;)!Qv3>_
z;GqH#xB$*tO04mpYaQB&AOfJLspQ`A)p%7!ioJJThf5{D*}cxNfE2T~>rf1Ck6G<g
z4JqN6CN|kbzWCibhHPFZ8dvWq7CFd<k`@p<s4QT(zC8WHLs;;>D{^EzMg0omgL`rA
zyFolZu#`S^^f_>!+Mxj2pM1e@A-WBGtbQv^!$9TEd;}k2Euw+RNR(1T&?Kbr>oeBg
zXL=YP6{iUR!7Z{Zf)H_BIj&O$yBBiX=g0L$K97q*O`054egf6F%yY%2HizCgK>2ZB
zW6Cv(>v0x?F8)<77F938FOdw?qz7%r-VB(BVA9U@BYLJG<;&BcGCMV10PaW^gCqwR
ztie3eOFx&sYX^|6+u5pF;xn<303=L}c1cj1BOS9x<U<AW;^`E>P&l=TrEE0-UJY2&
z-m6uQ<P5O@W%7V+$m0z0o6;cEIM>`LJJOSDs<SdfcNFo&7Z}YhwlQ%+FnXI*E}=|a
z$@`^OW9xi&_&(DdLK2!;*=dVN?Dr$_CuSpM)M;YLL2faluZy#<4%+E%4(FD~>m6T}
z3Ks6Eu#7oO?;=LcamsjgFJptI_rn`yh>BO3^G7(SQA!b<%=D0dF{GKS(J*lz*PVsc
zdEDw!x9?gEXIgzSzH+Mc^1Yur&6ab-C<d&pS8PWc$PS$#1nGB2XW%iu{F!g!R`!Yx
zl6duDlegXjNdd`}Xj*=ViTnS&bM|FOZ;g(pQ%IZ89W0B&9Ucx^)ktF)eEGcRAd)T&
zu|<j`_KhhF0nXIwa%Ne>gqR;;U)_eK@ncLK$wsw>sdJ6q0K{S0J)7p{@+@)n5qaz2
zoYrum(8r)}6mkKf5RRflsXsba>ZoJ{55=UsnD|X*N?`$*M>z#d`470*(p_rCa@dp6
zhvV$c(Ja+M&Bw2Pe?Z<GPdhs7pT3-Q8{>YkWPzI#l}aBzGQrJd%1W7>Vab#g3R1g%
zg+VfXAhbaCLG+rVE&kCqNKmFOJvTqi^>qYUIF~Cd&x*>AMG?*PYwEb7G~gMzi5Dsm
z_X^VB&IFJc+RW7w99);6kFf7#Uh0Et`tVzhfZMOcX+d65h|>vbyNN<S6zWXgU3inU
zgskod=D5cw!Zru~VrSA4;)4_t*}5?*)*hiz?Fj~gfkxk+vut01L`)%Gtab~w!p)%i
zjm?#ti&V-tQ=>bc!g@6Q-+S?&ezIiE)nS)?&qw=6i}nc`_tN{@yK*d`r&YeUkoXxz
z^z$Wk(N%-Qv^Jc5IC)rp(j%$q7rKD%x#Ye40T0jPkz6+K)Z!2GpM5-`4ez3P;XK*G
zFlvd7u~h?nS7hkQ5Mj?E?<8_P{Ts=~LixnPc7ITH6kC_o>Ow=XE{OJGCqc&tYn+EC
zUh>&@I$^P-(sPGX0v=_T9C3#?d-W~oNaIo}K;QiUDr)7(!S>9)ZfNod*dC^rQFC|m
z<N}AdUjxc@_r_&P%B;NQPRU?KwOjRdEA)Fl>*J53L7YY{iV^t2dKxQ9N2$<ELRXr<
z<A05G8dTQFz4?v4!$&=taBKx$fMc_u=`Nyi^t@}Iv{=)tb;)!?0q}BYOJKE^GdwE7
zaC74L@;w3u23erhO=unOlT<#X?l90^MGZMj+CPSscHHNR8`Z*QP3|ExlUwz&`=i@u
z*)9qdKlS8(`KQ%><(ZvN+Pgve4Tqz%n}Qmi7{9mt9GU~%hD}rFS4x<WFeE7Se;yZl
z`PD%fqV8%+(b%7swp-KYG9*20oP6jl?9cvXYUIj$7nVjsp63effDV%=nmDrN!KP&X
z>38#81^rv{dvNa-LC?Hb9|GVs6X0Ux{XTo204q;ap{{`b%otcfo#!a<e$uzS^m?_G
z)vKkJBlONNx>M&eo8G;;VaCz2Er_DUM0gqX?zG#j1Y>cIB#w|V)5FcZFBGuk987S>
z&hQMd*0b_<jq;y`ATeyMc<aA1j*2y_Ir5briYKDF0h!<T5W|;KBd`JT-o%{K^l`@%
zosYiKvohY{veNVcuk{L#ao<Bb?qPR7cADU<asllot*N(g$UBd=^76%OKSEDdzH49v
z*$*l|)~7Lp{BHK&HKKbe{zcFNKGBtl_&);m{|H)G+5gi=Q5WNkSFdom<<(p&V~1-c
zkkXWkVb2NA6{VL5?HFQB66f14V*$b>P&!%+1jG1>$d5swk|a#a$b;9C9iSs-ji1-u
zz&?=Ep1QaqdpGeHaKBc1SK2}9JA-=G>+ypmnlRy?Z^JW!Z+CMwJzXoEuX|g9S$RF4
z14-P;;ZoT}0rOCfR_C3O1R20(AB9%kDlO@H-R-JLDKc8z(uz+GzXmr8xPP+Tl+muW
z44q?1$t!0(TLXOawOsIN^&U{HDYwgbPxP*6Vl&P!^WmM`8hF`<rp~gUs|Are>KCom
zcBWf;UeGzkTp5~#0ic5N4*N8xJ@UN@_P>ibovtI8QCGc4i71#SMV`8ejthE7ii$+#
z+dW7VL+W`;DLdXVqC5=Bj8!UTebp}SrWLnK{t&DdnUE-tg-<BQM<o2Xu1WZHd8xG#
zkI2B<idS<ON%`SdApW#6DOY?&suQInA&sU5zTCGA5ytwZFw&J8c@t=NdJuYUZdUbS
zRFo-V85s%-CAZ{&im+x7{*Pnmo0gD;)S*^x%qf_m*^YZqe&}q*Ucs0!Bl~G`2(Em$
z^g#Rh?Ktg}B<>AnTJ$bk*U#_6ZAx`obmFVA46xpX8&5F{W{d^rb%0ywm>01vj878S
zG4S5DWtH}{_=!Q$Im3NPcdJ?w3)xXnzapr_tuX@qJbW~7vb}SlHNHu*g>c`arFYdO
zX}K<73khEk?Ec2OLw(Qyst@4M$p*UJHn$x|*%2vYad$2xRohe2P3w()n6?-ty$tBD
zL9Y#d!e4xq*&&Btactg>$G_ndnV0j9!}}a0nHx&%KW@*$l(b6%r6|@fYI0o^5ezk<
z{us}T-0lqN4Y7U-YaM0Uc+U6u$2Ysv(cTsI2WyS0uqa&62u%@18WB^jmMlGpmnT3s
zb?>90$?nZhYWrV;g|a{U1Ki<yqgc>I?6}!E3<GfRPFtI_`mW4taJHm<)yZ~{su9!j
z1YbaejJQJ?NgMYfuEX^Gv5`_DljLp05mCn{<%>8_K@uHsR_aPCr^%Wu4vT>`X}AyA
zKrAbrbs&+P$~Q~N!ZOCZJo+)`H_h+;d?*37AdDHjuf<GS2gt5)K*p&g2=In4Xo_$F
zvSSbqwDB(#HBww;bZ=S83gS{YbXwoOlc4^2hAoP>I`mq#U5n`7D{$wh6>CSNFZ~|E
z`6>0hsYAJAE*O3mksiempw<g;BZ|b;(8wnzKdW9yse_5CqDV!BkCZV!v`)Wy@^0V$
zV6=62;5;4;R~A;Hg%E!1>$^Zy9)K|>(#pt=B=<8T(h}j+HmPWxLn^d65Isb&i{Ovu
zxNdRK#hO37ECOM$eFMbkewpVCs}-gqq~<x|x}xojxQc>V0=ve7a+%i!?w=BTz;cci
z@i7Q3OLV_$gg(2F_c0XQY^pY?pQa!~uNt>H_)WG9C1~=s6&0;I^7F1=bB+ey;ZK$H
z&rpZ^Htpx|Dm;2K4u6<WAoRq|k3F`5w(jcIZ)G@M?(G`pr(0Jo)pF-6Su>QzGabR$
z8_k;Qx@MYJ1Y7u!J7}zYmB<OU;mW0|TmpqOj7%!0S9A07!c<Dh6z9$7Z@;(rB@x!3
zfSSgWC>q9B_!?=GP!hphsJC!C!$k271e<kB>9|KCap~0kC<=~$P6)Gx5$S9KojuZA
zcohAP@LTP?1ii{ZTq*UQV-X}2y9~zLssqI2?o{LC?y@Y?<XIFh`(KQO;1?2vIatzV
z<4H;<bfDdk*nH^IZG^2#P(ndp24E?RlweXFsXeMyYF~nI?4RWjLn^cI;a~FWasnHk
zyD-vC6}7%;Gr}#6QtU9*2F1Z|Qhq;A-CDh`fc%sk@2v51hsIxv|BHAb^wvx+c`npB
zH3X0Ka6|O`#iJt4`BebF(kw~kvOxH!x2c`D`J!u2g}1Mb`4l2+A1mfplpB(rVfz~z
zqVFOzw}*ru9;E{^$*ej=7}@6#O?Cifwu8{4_l{OFwO=#cGal%Yqz*YuMw6@K0zj}v
zUnro!dfKpGr{Qn0i)3Y>QJME8)Ks*jAE#u#!N~nd^c#AjaJK&>x)gsv)+#+Aj_mnw
zc_!SX!IoodL)sex4=jm`th<rBeUjk*nLm$b!7Mer^lz8j1Hf0~D!}*Bv1`<x7_1X4
zNx48k?=5GEPNqnH?Ai{Jf5$W3<vV|;3*}tri)Qov>(F*x1>POW_Ai3(D%n0VEQFfa
zdeiWn8a%m4E=^dy3Y6b87FT^ZJVBWnE}YPiW(|CM`V20w7kO9M#P(kD?rE!E*yI9K
z?5ZhlAi|Ds-gfqbmm)G^Hy_Kak@~gIdO1T%tNXtlE*3#har7I(J~wvx+Se|4^EJ5z
zY?Mq0DFmm4f?6;tBymk*N|AWw$YuE$2#K(f{1+j|EH(hvuM-Z6w1W9UX;E&Cuv~HL
zcROnEMT~NuGiy)-m6{-}0Xe>yi|{_vB%F(A^ge<crfJ<7khcbI?Z?nb^|Lokp8+iD
zsLExVOm^7Dz5x;FQri&0<Pjx+*MUyGty0DPjIa?|O-y5*ODgB?xbfUKDlP8FH5TlC
zA6d<9b5zx3G5@f$1gG<pQdXXF*BNI|kcV~6^sD8%3#4J!z*oRcUubHH8LrP+K^aan
z1y6du!hTR`jV1RF_`7MT65|9*RV!Bkh;lZoSdZ^f{C3ZChVH#1eGbk6ZJ2{i5oiL_
zhcX-e{bmcQ2S%qs5MLVlamATSExEGTUgvunFvL027MVZk&R-333vZi!pHlqHby$OB
zSAC2+mgws^E7U8__<%%Wqw4~FSj~R#IH~<TkN;yqg!j3s2gq^F_(KM2-kad%#I)`p
zAKf!Y3GdgJ3t`c25yi&_yyo=}6FUc;=1i8QjO{cth=PxCzni)lW&)nQrRl4uNPe(O
zmV1j1OB_7g`Q<1q%n)G5+UF+Cx$N`y`O`T={d6ChDL+iB*-rsAl!cBh1kF7EgyYm$
zI9ftB<H$q2G?|K+&?URRz+N|}pu(`)!{7BAN}nsFgCr|~ygeB*gao{A|DnfX+8=`;
zINZD`JL0VrVhKSdPWCNbO>S`QYz>^PUD&kJhmjL+94h~$sX+Nr(6ZEjwG%{iT*f)|
zF9I+YmVXg|v2p$9vqoAv_A9Mueit>DE3`uWiSq=Dgv2Z3%~6zEYz+5C4}T-D{c^ge
zbY%u`g_FdbS|)bqmV$6SY#_tFTj|k+eJg2W<AuZb=3WI&L0a96B}+Cfi7Tr}YS<%D
zz9j0`h7#%Y8~^Lh85PN(Vd|)+F2Z8}wCDTN_4Lt*y*W0^p3?9X!)&yK26*o@<LR?K
zV4gvb?t5(2MN8M?)AjNALIYn!iFrY|U3HJH{!TA&^6r)_@`AC)x66^XbF1hkus=Hm
z_Vy;?vOOv&q9|LIGhd_O2R|B4ll9pTE-UwQ;ws@G%~6ZUR~GHV;$r|76-~OTc!gb|
z^)Yj*P4A5qB)oZD3uF$h^pr<#T-<uuNJq$e^ND-V&URXIs%3vY7Xc#`Y|ng=yP``q
zo!*{NRafr3Dp$89R(pSLQ~hG;>eKn4GV<x5ih6I2psi8r6vMj%!+}Mps`$!qvzK`$
z0Y}=p%j9s=>XBLZaA0$vK?TQ(bY5P4G6v1~VBc0(acow(hPD}wjH!UW&4Qc;y<%X7
z_}WB9_%U}kTe@{%m$01P3i3#&*5Eq16wUx78mHX?KnxyV8aL$RVov9nG_3!RE|9wN
zw<gNsH40=;%Jte;{`2ZPU7ZRy1@q6wm)BcOiTT=N$_!Lexhyz^ZvIuSa(^UpOhmm4
z7jm!FKdcN5S@7!B$X0DEsW+Lu^|O+9wH#DV3|Vy5?Lz7Bb52FwZode1O<6e=$(gV8
z-R{lx$*Qz?HK|}f@<pR$bNnLzDkjfpf@>aJ0ApFG)Xs?*LIgVcd4;8w%_go}A02fy
zoZp+Xaot7&cN`9NnqIn7OhT8fd>qzH4Y-t%1_0&?2@%-mGQBsnL|t1l0_&WP6p15L
zP*lRYGo2K(Cwe3pFH359!$a|-KkN1r+-n3LM7^5w1=+{xa+rU^GJ@g_Fhr0c(ab2B
zmFuD=4{pL{F>}P<iHtVg5liY}ggJ;u=2JXXYza}Sz>SA{nQF;-P_0QPgd+XAH3RdI
zduS`f;dS&=#X()sv$5!C*>)jU05M<B1YQEk_oItE*~%n??&F>yx}RyE(*ef-!Gn0K
zPNsCvGw^i{;PzUh@hJg;GU-bH`AcJ&9$u$8hBWhGLNRQ0d1hOIgEXw!B>!rJq-$IW
z3v#}X;L;=t7)slwKS2DDTccIc!LJ>F=e(<5vV|L=XC6R+IC%o4m%uv2kdw%q%@1h(
zA`3{<ggOLX!Z1%OX;48p(S7O+94g^8VOt00XW6>v=ae+LRny|0Jyc|3BI>+XPs&dA
z!c@&P)@Ur(lmQ%VO6cH)B5r<!T&@CgmMun$IW#wka~_M%SOwzz$jDS=^Hd@&s7*U>
z(R?n^o_jaG<7&G&cX1K);`bL8J*JTO-RxXFS5;W6)JVE3&e{z0-8A{%70UA>8W1Nr
zX4kfocRL@IN#p31uf^>Xn)alj>kU`l$v3SR8VzzoMnJ1W(HBU;1~8f~r>cMPaAG|f
zJ;^UEub|4VNUeI1J7i<aDN{Lmc_iK4A;sgX`$0s5!>By}u0Iz@)|?;;>d7f2OUXF<
z-za;BAW@(-ICyN^wrzXnj&0kvZQHhO+vXkHwsKQHlB&GL%lhojrn~xUNVnbFWfN8X
z;jOarUQJuzeChZ2mVM^A!;JDun4E9A=r19V30+F|*|gpQy954v4{1j2MU(0dc}5I0
z;s9>!GkD(!@B0SJ2Ide#h>!_i3c9#rXte75Jg7q*cdt3$wq3zlUbvm(zjZqh!$ttK
zr<w$0+12*I5ld1ZLsD|pSrgsokeKhY@3ypD`V0URbGjNdYl1;~D&bw+Pi6U>299|d
zVtlKwKkBiSprCW9n;|t*DKGGMnYT4Ez^g1ITVfBKh!DbE{>ZxAjd8ilUb$BX9A9~z
z{m%D6%!~}`OdJ@XG6*=#>h#Dr$mql0-d94_eVEpmE<iC%n*}N*@T-m>k1jidT}u=N
zi@X_4+|;w?ror&+bWA}w4H&W8b-}v(fMMGoHo9{9w|kQz`fgtx-uegfg;SrG>5v9h
z<a+^y*)U0%%m9)VA0kCwO&&Y(k;oZ!0bw+e(6_;iOlS~r1SN@S0~t<T31R%AWapi|
zKb_`mLMf8s?V2V>HCmtvBLCb(xGx+Z^jy-%sz-2w6qlanUqQt;2nxY|Vln>n<2;bC
zl)tGU2g9xA>Ebw=C#~1P9Z?cuBBWW$zAl1{8$DS-02~pudScq)$Pm&fxcW);U$u_>
zL_*mjs;#S(C!r?TWFpebj;XT+m7VsXfsn-sTh{YqJHD0<i(RWni7wt(-h`%R5|R&l
z;paRZD{pAou!n&pTvKs(n8p@fiE8jC&*YNVmp}c!hK__#b0$8CKek@zH|N89u<<Nc
zA9k3+tZ|ob%yHuT1W@u6BzJ3XZU+}Z3z2!QsQPV~$;U+_MV9CYbS0bEAbHA2zi)W3
zP4PgAk=O_bXH93VD$n%sd9-X1<(k<?NVrlUMVQ@t+ICv%d)n8Wu~7XBH1a@hV$vAi
zgG?Hu?jH3|BSPNlevbNf-8l4LS9~2E!VQIC1e0H8bm;80p937w!;)~inO^iR4?DmI
z#_-~@aBlr}9<rOrwFBEcJyT$=0NJshfgl4Elwd!%09bHvs6pH+3aw3(YXgvuZh6$t
zr5g&hm(`jJcZExb#^7HNx}jvN0<jD%Ul27J>|13<x2Y)6VTR+oocR{nC;M5MmqQ-_
z!6B9@O}0cu@Se~XBYDTBpOM74B+mo7bhkl~->&d<P22+|$`tSBPPgvl+M^a%ZweHV
zwq<E+ZcB5!e7e5vK5I9fM#oN+{^BMDnf$keLyCWk%-oZilI!5>X_9Gcvn!}%ny@n?
z$Xv~Zojt=P#sU1HUoukgz5gN&-t%z2b-GX)14xT->N+9Ggn~ms+5_z|Y1)Jv)SOir
z<Qv@_EBkcjhS8)S9bOMpb4kn=xa1G<^BnW)6+OUz409&*0Uji@mRo}=-}%27(5egX
zZ5Sz&xt#%CvCzGeTeTXJ1M-+TFzTXJ&cRp;{caqf_9p_Uhk0e^{U~QJXOJGE#<KLf
zA{k=?YT@cC>bC^QCjqYHda;Zl1&{iZP~o&}dT?!`iBTa}qR}Lrap#s$fq(C+@PA;s
z0H9Nn5H|mgH_SCZwYX(3^M~Xw#?kvJ$rU6<Eg{UOcf_WHMqcs2H@yh*y*+Um!|fO`
z7^i_7U3-32Y+3w9`J8hH`$Wv|k-DFxFFQIakRkj~YVY>2e2+!+galUGc2+W-X*+Y$
z-yVMHMDxN$g3GSgncx-u8sQRjEtuUz;<^RSTz_-29@QgujV^e3XZODItM2>&{Rbk*
z%<vzGAS>g4O$1y1O;H@SBli9~A1YKPGP(!-rNZx?n7%1C)v3U|wqQRlB%L%2CBjW;
z9D9GZ$OlZ!H=b0Ilm!uF`3M+YgJ0Yjik|iLczwLs{)~%m(2!D`dA4rJf#g+EM5bD?
zZprv&TpNxbNm+FMIey*Rl56XusckQO-1YtKW^nBNczT_~w?Cop+wt!F?tB}&u)F1#
zV*IP1AQS=^YAZ~6mFcz|#n9Eb8?xL)Z7Y#zsok=P3VZXZom_&C8_70i(DfMfvCfC5
zg@cajXtq3!q+b_eGS^N<g&M$%^p~;GHg^`Hid2CLGAiwqEty1S-JJWUZ(CN{W=$h;
z9hDjCi9y(GIyXmJDG^|71rN=8Q*-D{pO$rZEvtLD<FtC9qte#FROc2yifMQ6$!Xzc
zHgoCGkdl1!E`5J(W#)SOVvQc(_UGfdlNaZanG5FVID#aa{2p&AVS2;oJJ}8Crl}Gc
z;cF1fAGTQzMnUmFEKHP%U-b+N2>((?Re7f_^Gk26+cK(}Mi0#%d^Ss|ESQ>+p$tKn
z1p~xs5Mc%~C;;{PiIf^kk|o=G_*A^6oYg>j*h`%nYAI^d!gfp}BxqKMf$r;(k=tMN
z!7-whG#T8Aq7aG0u0)bb@#tW=>wpPR44Q4OOqn$jqQ>asqP@?`EsST2#mpIp!E6wL
zi)a=OI&~U9){8?F#W3ktlCm!HVxSlNdVh%TQ}-&)%Nwxt^&L=cd5fTTYZ~%^EEX1L
ze;tUxKlt$g3e?={Qqp?QoTr`k5e)ncCHpgKd{AviPsEgym80^@cePT~<=D0V*qvf^
z(B%HMn%0v4mt4hp@<b@`&5`7V&#RmJ=+brbfM`Jk*iyXs*&q>{0Vx|IoZKDoamFEM
zmsS5w5*&=gfGNUaE#D6Kg7I=(F|?@o@w2-+?>ZiKPoNQjOkdOuhk^QQa_>8_uxLp8
z_AGXQm_vZ8E-zB%#tkI(427IxNFas<)3YE>>=J92XV_RyL&5ab<VOkRVUT$};bNI1
zX}UI@&_0K^yAPbb=E6t=Es9zOu%YqWDC2N_2WR<0^I)3HiW$cS(65$T(~v*jk8RZq
zsHD$H!1G0)0K#n&s^6S)gCW0uXNsG@6qq0H)6VWP;IT1k7CKB+k=PKDEZdEO1N>Gc
zd8B!1XKE=J!qzbH58YoasR334rqbB#^{i_YrC`8c@X>HeL<EcDuPKh-MaNv+El+qd
z?XLXeUiN<n#Be}xU0kw(rY;ay+tZ05#yytqmfVOS!Kvw6NC-{OO#4XJ6+(%*#eg@c
zKz8J!^esINR3h4bbP%m<sBQyHA%-)sh;cXv4Y>Ivp~ML6iJzN4W}1CoTeqZP%Yu>B
zup}z+d6fZQNRD91c-j~XdFM&>d&ZI{?0n0{82osVW!(>qr8mF|p>eaZ3~;U>g+4Ef
zpqPDlDV$U8=Y7%qG+oXn@pTa4i!=*(nU&~Xh&Bw$NDguu#I184y0rH-e0#`FKo~q`
zqESUQZ>jr5X4kuy^#yup^Rt<e7Jk-u16}6yl0kD!yRAmnA?A`u64c;;si9tvv$cis
zfOBy0O}1a4mf`$=8&$~j%|uW6@uM8?R97$U2)`d+PY~3I`Ko-GRjZJzdx3X<{PtPf
z$EXKIS#f~%4+!*yICygW^B%*jzXL0`NGhhQtZ$1}eR`*IBmHCRL@hXQ;5I6IxP55$
z20g&jFrpnwCurxJ;S%>+Xfn~(+JnL{X*l)un`qK1T+P0yxR}L|W{eO$i*zUB3mTId
zuPMBa=tOJr9O;Py(Z{t{!hBS718izB7c3zDc$hBDLM)0{n?pOmP4UXHy;kM9b%;+7
z5~nmF&+`+!1X03s0P3sCXi_#XMvpW$oQuSj$@8U?lr}VUz}9z7v~4t4+YmTWP+YC}
zo~v*r5jkWa8F)u3bI!Yha+yDbqnzQfNkUB|y2;l2W(|!1))>(EocB?$a`#Ct9b@}~
z!z-iEprUcsok4XP@C|{U;UPR2iN9}eubAwm%4KE%N>I)m^nkHv>q($tBv(X}GJZQQ
zZ-nDGUreVENR_)c&7+;~nP{!~$OTY-TVesH6Anu-!w;Fjz7|aZmSbX|=8u9bj{?`)
z94w3MBh&#$669xcuaLH|mPWTlk>GPXkb@ux+D{lM;&a``@zD%zG-LGc9V<`2bl^d4
z(vZ*CvLq03COogqDjGA{hQ||&+E6lUOj(RKO;Za0poQvd*yb2ZNY55vfwm}b3U}*!
z|0v=+#HYx_6rH@iK~SC@kp?gr8+RKubf}_J`A~*Si%PdPn>a_h4LQDo2h#@y{awf0
zw-Wrk_T?%+8-C6q(p@dSa7l7)JwQSFM!X&&E+*MONY3*G<}Xhq#a;jsxPV6G{6ng|
zc2RS(mdpT=fS4lVm=zEFQLDY5B!GUQT+XOJh&M9;>YeW5e&#<j?Q>{b4-OW{OLML`
z)GJsPn$m*heCh3YA!G!r;LIP(OB0|hST8U4PL!K{Znj}R#?oJQi}6wFl;|mbO4-lw
z=8^DuzAA@?%9dkLIRpg}4da+&*P+*0h0XtzKn0rsRVrXo6;&5TSSe-T%_7xh4l?;}
z6Lnpp6gofsZb5>CPKS*;T%P_B2+wTGJ2cFqnsfPi<!^DxGKI2e%3KJK?z+eKQc`_s
z=USY-Pd0#o#KX6Vas<fGOOxQQZpN*R*zkQikS;DTJzb6S?j~WnZ2)abKKCk-0f2~!
ze=!;}rA~%jY>y^AHf*;N04ddug;>;<5wC>7)t(LJL}K4y$u-R=Sk8eO*(XR*t5i5m
znvv<PqgD&Wuo)Db<Kh$Td_xz;P{A7kt=9j2D(O3I3s5PUIoh1HDT;RC7qcF~?TQ|Z
z(FrtdW8(xK_PX(b9ZMsyUiPdFZW({7?X25R4IFZ%9?;fKr!59Nz(nNYsK`>1QnFr%
zF(RK)YKITwNf<3Th`iw<jDr+r2muyThbaoc;O#LvX{5g80xQHsC&Y9^w@76jqVq;}
zwaRGu_1f$DP(bbktNFXK|A_HgA;Y0wucwB$tg<YMfe&olJ&n1%GBI+JJwn)~Y%sz=
zua~P#-mCh+Dy7x;BF<E>jX&Y^oIyks;rFQH5b5EoyxOPO-@LuMWU7X5z0w{LMIK%X
znvJ#EOJT0<6O3@AE0i{+O8R2=SMJv&%D9o2$hVr;QSp3lm>P1WJ`eCduOTD&@U+*<
z1DH$bV;!e$b)_1du6F&aE9ZD$4;^|fAfmPmTu+A&L}|AFF1VSn?oJKJ+vNmTG4_Y8
zU;33eE9L|2inh;#V}EZ*SjUh#8)r3arRx^}B_4SFKhebhbg*Uk*YNvacll~LZH=P(
zzSi~sAvY$jed%3qK98<kZ{SYNNmxH2ys;<Hr!;^8!T~5g`1;80*%J$hEG+`h-R3>L
zYCDD7KO-Dmn6vxo@O65RZIU3St%zRT3ZIa$k6=Jw;EM9qvhV%6{m@Ihg&-uRBSMUV
zEVt)byRGxp=<Cs`E$@epNTnvcKREd~*}VO$v;FmOKU9|YeX;At`~GurGBj3l5cQlm
z95t86r|t6q&Ys6#v+Mcraq)6!2E7jpmo5i@987t27|wA;A{KmU-UF=d(V^{7fp%SV
zFXE_&3!I_arK4!7S5JTrklrY&<nq_!u%fJ+f(Dg7v5<Rv4b3VAcQOAAdbF8ArXERE
z0cM`WX`V|vov3;oMcc*vHOy65eiNtX?%hn1q#`#`uiL%gAbf0&UR$HUd_?^!RxOxF
zEZwaTRCvyRzhCZPcDqT<sW{9Z!p_KSSu!^GY^4DC))EUbuo+?bp%y}+UVb5xX;>YA
zT7d`k-_IaM6RyJTt#bVm)8l)8u}s+NrEk*n(A{y`Rv&(!DDCjMGSU$RCLnDN6ifSW
zsfRULBsDr*^nFM4fh1;%;QSS1d>z#jhVbmZ1Fws_2rDnMyKXw=r36t#SVh*U13Qo&
zTT%X)KM~#W<!-@SX^66EDOX%#L72C#qKYAXDy;k-t#y<p!{H9{jHN^TJFr@{`L1%n
zd12Dcku2Ew;uT{tG`b2(-cN2hZbpX@DGcD{;02hLitD>_02h6)v+i|E1&0u{WN;sk
zFLORIJatt$jbczX&67ZC1C$bF(BZBII#CRvZJ~nW`eS-r8M*DOGJ=FA6y~EG$w4kb
zaonJ{`wgC60zW{i!BI4@h<#n6O~xUnTvgGz7Un+~j=+`TN;-X2)oH^6Wre=*J{b7-
zL$AFQwBBRs#6o{&s+q%>uSdUw**-TJ-E{=Fn#SRufh{3{vs`SMZ-!`mno(}E+c7Xc
z$m=jzQAx--lk6(9X8Uhu^Nh^~-XiDxJ3g7cCi~_m1q&ITIEn%UQCf>0#lS$fP^@)x
zU?NS$_qm3hnY0C?8|Yp-3iwjKSuNZ-T#B%8!s~hOvc4$xdUw<Zx;9Ngw<dfXYxm9&
z3^C=CawpaIHlG+PHYpW8Ih7ty`r@*r71NyCuMN>Sx>jDfwSw}>j$H?x=f2a(BlDQw
zfqkhF0GgLT!K0?$vx-2lK+rCMuO)%GQDI%*Ybjlou)-1Z3&Nf2U$<nszH89^!YmS4
z>;hEp>8YVxe(B1+`9K~hngPmMP=x@Fr~r8cxJA8(SLZYh@*2FVocFO~A6?y8pVq|o
zOZKe3-18s|mdztF-NRG==d}`qcr%UXWzxX{$kr~o#X6Y;mbh7;Rr{1*8J`+1J1Hci
z@@l6Un2g;S7>>QR+dP{^Gg?qmKuRPS(i52H5C*u`M*cDkU8BZRkFj#0z_?8sbwSz1
zF`FkyfL58R#(>P#;3S(j+Mw#ZyWG;UR+JYPgjFwoCFh<2YS$V;=NiFiP2l0O)S|xx
zKnh}`Jq7r3Jl9Fft|UTNP2MxD8|HrWV0y2AU6x!SPwv7XCZ>>gke4`oueSu}7(A)^
zxdyKb&J1`xL84nSgL9EO4R~&PCja&2_fx%ranP(YQSyxGTjPb5@?lX|M=~sQgoicR
zBSsSdIDlVdHB@8yzfmzzN-;<)ASu8W#0Zmf1>PL__JR(jM~9&RjMy+CC!&NZ)Nt3T
zrJ`U2pr+d<Qs7V4RLEq6O53NbA<>279Opg-m3=z4W8zstLkY6P4z*>SO5-PWTL-2~
zE{e-_bgU_OSY6;>zkRjYCoXpl`E~>EiX6w0&|LixoNc-vuE{vEU|!_QLlKL+|FdwI
z$g8U8@tTl3Q2bFpfbKIu_wGhm3d!5tRsn0|unJX*BnRJCv0<{x<i;+sjm!{7iMt^+
zJ4fHEe-q`9BBXm>7God%$k>=m&${G(`ypNS;QyS!v60yNuF3-ndxOp%OYO%k+0ISz
z@N#SG;78FDpJJPcN?MvCv;J#~XE5uO7mBQ1u!Q<KW7?sZ>QsUNEp!)Qs+ta_FaKiU
zAAoompgkf1>KZ?c%jpn)(Ra3ILf4H#j7w?YuH`^I9J0rx4d!A#TPB&A>C@G~ItB!%
z?+l(!1nRE=pFVF7uyas+n&!|~qq5$D?kNIP1xJ1;4QhrkJpHPk9Z=SpYW3o3+_Ih?
zhqPTmo8;IZBHnHsM~$~P+BhZ}=#d>5faG1+eAPL>tZ!o@4Myg5MTj^&M2nTn(wd4v
zW8Bo|cQ!5%Ti8>9kx<E=!dlO)awR>Y_hQ^ZDc!icx)LND+VlF+-{ecubAJ4%UtO-_
z?XjTPqhC~q6uTq7?`jaHSJ*N=eG4zO?<oVGG1ng}$vO7x*T;P9HOvkz?of^5W=QAb
z3uX>mzkF7j?ljP$GkHZA4+{6yHIa+rAbfVqLVLoyN?Q{DSotOX=@BM6kJM<|D-(`H
z%|`)&SESafj7jntQ`u4ZQHEkfZr+kNWo1I?tS+q74l=ImsaYm9tX0s-;J)^9S=ect
z+@+Cu9#2>tgQ`)d<Ckk((!}+8n`EXS==cPq%z4WX@DuN{&mR}|_{dltV|Y)gGdH0D
zcb3SZ$I4dKqs#Y^U*pd5@joyG7UusTBr!6w{<jQaMtwW(uocm-uWt}uh5*<qS0eI5
zVjgH~hfJa#e;_9ctQ{IoJFCl-`cEuHjT`on-)+3`Wa5=etpGk0zrBd#c*g#;ohi;7
zp4t7?C%d0cug52DiL+`vk;%FBJHz3*Gznz{GNV?l8GcbRig7Q31&+?Qx5wks^Geuz
zU<8Nep1s@m%>{2F{|q#mPH$g#Z*S+%{?N=<+Hss+-=DWTYs92<Gy4{9E=1S4taWJM
z#NiDZatBRg*H<EIz53`Q(Mah=%{g=n9C3J;AdxNwiFkE_g<HaqJc|-K*u*?T#4S0D
z^AY3U&pGT3q7Ics=NP6WZdaG8O4TVO{u!xM-x7sO+m(yX!8wLZor>>-s+i>zLRAZS
zcR@-6k5EmWP+FJg;!JUsn&ku$FP#sq+ixOkpL=f;n=fOnLkUToB;+`N$7cV}L(G=A
zoFh&|f)TXhH$Y=kNuW*3pk&X~Eu9LwtF-VJC^Nyyme$Hj7yUcxEzcS$l1r$J)y5Rw
ztcB8*m7?8yhW*+uR~5UI^R${00bR6VRFQu1#FAjp3=1ATdrg0t%*vZd#jQCWOKf*y
zU`MB1(a#vm6P75H+7*@eKxylC%rbo=yb7J|W>rpZqs|p6Ors+I2=k8nvhcdEb0WCp
zN8%~P@VdJcNd*${{igKU#N;F(ykW>brJK$a7s^X<>H0It)Ewr3z1gvCXhUNZi^*-|
z4`NCXupl+u9LQCWApE{hc{D1;L?%X4ks1CGMztz5lYulqK`#4ou)<PHJ$q5Lc6m2g
zSqk^{8ta^@{;$r*3w5_<<R1#Oj2aMrS9<O{Awj`>Akc)Z&MHk1)Kg*xP5MrO2L=8L
zC`mtxV3VTD(o+r-CNDv(_)|P-V+R5NfEIlQz@oqz0wHL^bYNn?{y>L8PJI2tc&ir&
zK_iY^sG-s5{6@O9$*J}vSPa3GO=|`9cIzj__kI>tsC4_%4$wEw@74feW0DXzJO>}(
zHZ)0RV(L#%hXB=*Szk0ZnhH3gp*1O03H9;@9I>HFuv2}ylj}+7*qj0X(Zg=eJ@Mdl
z*0`ozUq6Aj3j8J_Vo3i*iL=TrU$J4HItKLYdgbVh5~q_fTj&H<kAm=SAnNWtgaS&u
zySSKVMp+H{3)T%Mu|^(z<~&p%d&XnXiQ%J@gfm$n8~bWYc0uBIkoP}3X;7y%r(5Us
zbycgAXD9XhgjegrObYR=rTb$}QXFr40N}+H)cE$VIdoegaExkeC2rQdu)>dqnk~;y
zVjJEjl22s&J(B8q3YSWA@o1_zvJ?O%NhX~11okvaV2;Q6Sc_Y=*3WQ+%+C?1YM#yj
z94}rzvPgLOv<pL$h?Lv540KK7<$B!Gx5v=4%f6zY%)&>yAl`FP0qDomD;yd(NlQCJ
zW$Xe-f|>X3>mB{b6M#m)y60xugMU1pZf<~O^J(YY38OeAG4OYH%;JjXA(@y*^Z#Nw
zrMUDwCVD8uSUMDL?FWN_KNj*yj_15OVR<zFDU_;c8G?ZRcupw0&nC&3<M6bJ`MwL3
zEjY3~{2u8gPy89_U(>vGmM6LDbd?u74i4Dj5a@l>W_NTsy<Cn*t94o7)sy~O(S*qM
zL3D1#({6cquaC6VYRiH=?P1A^2>yj*jwe-V8e#yV0*QJz4Hl_O=}|42Cd^c-pm(yP
zx2ui+DEF~}G<kVQvrEwZY`nE|V`oomyXAXYO5uVR;=5AMZBGOkf}>_a6b!XY5JCq#
z*g>~K;@2^}>tgg){ehn1HdsK2WInW#FjK<NyRdb(KOhD&-L&`eM<7^vlofi05H%^U
zg5w)u5(7AqcmDM^GbEjXIG#wqWS1JhzO7T10jx2SU}(T#zG27eNi7Pua%#NVH^&Y~
zW{Io0&JJ4Mst|&?qeu4khUMK%Q4ng1xWTZ^Z<7eK!&SGglv+<ee(9dA_J?Qd7)^7J
z<VkdLJxN9uTrv_S_tQ7eJ*s8KVuY)nf{)DY#kfHkDZt3bfPw*!6?UxT580GNDUcvx
zGaFTReBVpE1_2YGigrTwew3h~f|wKKcv2BRgq--8#V{Ns^#TudS$VNHcSZ6<9Oe_~
z&6Wr72^I*g{<9Ub*8UmKSknt`&Wh}fT~6%7(q+8uru@<*SG2yB8?-?W!Zmnk2;=IP
z(fqOrk9P|<%UlGyZ*h`!hSRZ@ZUdEgGSS{cW(PjmUi<g*oM4f){6^P&@aOL>iEF^u
zQn2g~a{g$!+v}R45LUK7a^CM#)=y6LG&0|+e5LLV{jh1~<{5?9It*@|_V-kBdf495
z3$t0)2uFaJUNk&3S5Ae(;UJqJGTRnihgJc~JT(OstgjSK|FpleuIy$19xBo=wjddf
z&B$S)G*0tJVE9E$5&{(Mh6Iz<6Mwx0*n&_JW7hc%y7)5%4y;7-2_L0U(3Y$0zd81$
zfmk|Mn*9$~5}T7I;#Ac3EfwkZIaTT7jnY6sECK82+O^Xa!}&iX8|foICJCfIh&Fcc
z@%rcfpu#S2&)0!bJdUw8rzWg5M-+1Ztgz-AZt(}7if62PDE=Xs&=jxFkTEpsp)R*3
zt)pCn+xALte))P8$rdSk!xVNvv9@glduay9AWfIkKj>%4J(05Szrb7_DRpszRG;*M
zP&;fgN&X6N;6zV<a8Vp$WrYLm%o<)A{HpemlBZee-Vf-V?q0$_xn4KxX_d9=VRSWb
zdEMp?FWtSQWB%ZIFRyzwsQlsB=Tm%{2U*Q|ZR}b8k(*4CbX;K#m*3~F6zp1DB?K^3
zwP+$F@42>oKj9Z2JvRMMoa8_3tTVDQGW^$^MD3r;i4Em{X5)sS#sy~Xh!Pj;QSKI5
zX5_g;YHZ=JH7QcF%acfj(>Mm|fB6!_8cjHz{@xe}`^R&U!1p=5&lWT|UFx4c9G%`B
zJ;jV=1AqoXk!y2EbNoey@cIjBbXs&7F>t$2UuFkEn4=>?aDrSic*W_*?TN1|4K6&{
zKs~QN4(d8~dgv5CEq3^NTT7O9_iWIb(Tf{Xw*y;z8VY>SwU$)10D?BUmUIVHdRlbe
zWd*wrqBYIiFC1keLHYm;7+&kd@xA5m#6ELqsM7-?>66xf^!z+ssp1#zRJMF^?d|Dt
z;LWmyCR#B7+f=ANFmS<GR}_XT;cBNj5KGkNh7yJE_F^MLpw7iNNEE*bR$Uz*4t43|
zb1JWY)hl5j?S#Q^5Nn2BfCT|@05tk|dGF5R(eucNBSX(0O;1+Pu=H0D;o0OicIm2L
zXWtyFH~3b={eDN5o~*79T!`NugAs>9nXeEKRC%8u-xhF|8oYOFox_u1`nHgzO;+}L
zk&3E_=1=45vsVB=69}3#IZR#b17d@$g}B`=s|s#P%ZfAbp^V~*Xp#w~6dqC+wPr07
z_ph~(SaF%W12zY~{D=bXK&faZshO6jC!kLaRFuv8n@QL6qY|Y&%X4J~cc;MJv1sJU
z$~Q71F@(|(;sHafhyYF;g9wS!U<q7qLF#SSqlAN4rzC9N=AYJZj5Kf|hN9Vvhm1mP
zxi#|Wa|4iG#o>>z=F{gCs)uu)$|NX0VqOs2)Dzl4qPnH-h>K;CXimT)Drg8(Y^TI9
z)O95l&riTV>5kAQyvkG-a9xZ4{4{ev%S3fr2&F&i#pZF4V?!A4JoV%c93BD)W+-nt
zHq&L{(D<!7n1Oe#a8N1`jn(R9@3iAdVO$EChU(II(uzX&M>n2MraQ_6RMow8GAcub
zE7RLN6&I?AByNz<kA@x55AAkvNf-(@fU-lpOY{#Y)1HBYmDbie7noHA*$Crz7D=b8
zOrqgrRZ2tkzLXke&$!b>S_f(qJS&tCcMaS&TB<1|_ssjsp$riq==$T>M&Xu@@xs(g
zvXSi>W$YHIt8u`Az-pgsWqW!P`#&SZ*}#657R}NEAok0{{ftoyI4Ei<-?L2Xjf~3B
zj~Kx+CN7r8-wg7x>1L1f*{z^!A4E*(+iX)LbvAOeT(->hT`{hQU4{klbd|-SEZg+t
zFn@|U{?d{M)I$O7XjX}Y;sM@*O40u1!8;&NK=Dw#+EJpaL;0f>v;u3!U*hq7VIN<C
zQ2FtM65tk;3J8_ks@{2#r>Ol)eRM&G<u1Wwu~Dl(m;~~XaPXrP|4(lAYr19m1gVVF
z4L*sd$&s*4<?3F!%{ufvpV#eLl3A#@keDKpZTxm!yR^6-_5JfHl5WNn_8DHTj<Yi3
z`^Nk|_0*(;>fTm$Lkd_heYo0npP}j~4U6TdGGtK)m9s1Qeq9P<qm4}-C6Pj+6<FxF
zu?|jlq;UB)Gtew6wNzhOq{&>OEC|u5ooMJ4m$_w0(D^FS(F2b5D}lFm2NoOls<5gq
zI)6|ArA(kH1RZZ97tm3|5Qc<J&Fu{*VWWAaRi2e;2VWsZC|d7TAEd-;FUDiVVwUw~
z%GRfW4YE38cT#)1_>L<b^;t}_w9Z4&`N=J4l#|c+8sRGjv;szQz5c9B(-)y}4G((H
zQ}$r`4hoR#5Z%^@Hm>R!uevKWgIAYpa{QWTiMDOmy+JCcEp~YbAw<Em>Q4{^aEXsM
z{j~83Vp?NZ`hbCg?O;5X;<0)us&W=Hd=!y8eL)dc^0-k)V5w7Z+IeRdQ8|Y9Qh<4z
zZR?~#HQQ2yltk-Y(Jcre3fS;v8?j+G7v@=DsS|K@zEIkxp_C@`&`iuOA8skR{^;g{
z@uD!&`*K=e1D##2$4k%=(k9gu_=<Yo-A}U=u2rt#(e7C9BmQOdj=@AV^HQ)m3lT(f
zfdZ@Ln3ZL)7KHk7>)wGL{Y$PzKbX`9xTRJp_J6_#%fHE+|CLVqe`*B&4euy9+d$FF
z5->6_K+%g?I6D0s)8SzJ|2qJf7+C&$Skct7!)8bGU8^Igr@s#3Ac2Rc4^beH*|+uA
zu}&x!*dq8B6g9LWwrD1%lzmxy<t8S7Hov-Hl3D`Sz`2NbqTB6AfD^<-EqoC3h@>Jr
z=B1oUzpuvOHi=QAo9I?LI^y2(X*8toYdS~=+W|+V@ikLK#Gg|w!MMW}Ag<F_U3Ew#
zhiM!mp!!1sJgxYYKM=Yz3BY`H)el}|UEe4^3ee}pU?$u;32u*t93G^?b{q}p&IT#B
zN?}F|Q8OYi3<(!R6I7fXWw1wVPBF=+zzMJGO#&mLI^C&HATk`9U?NNtCuz<gAq%FS
zg8PKlL}Y{+<PR8RW-pN1Z%K4E4T;SJ(8q@<O2SQ!;8lN42+%O(*qvS&lpLRD7>}u%
zoUfPHz{7o)O;C*Jz@*wq!mq&y`eTf?k$;C1{HL3jdw+tG?E$MiFJ?Lr77flZ)op_2
z7%-P1d<K7Hc7p0l1OQa{aoQg=NXN7wK$TM+)1R&$=FVvxav0E_0LJ+A$fKVD*bE>#
z0B%+uNw{BMUkL*P7$c-X&0fBgf(QW`x;RimUl6<r7;c^1K3*jqk_-mppIXHfbPU85
zT3Z0z9abF~W6vPmT%T`^NQ*KuOogNjw6FXG3mZQN#A1)ZX=LWl1O^2=XoTWRF~qI!
z{h~_>3`yf2>QEmCz9^F_C@;w{8qbz+HcFm?B|4~u1ObkY9(3UKJSi+J#;Z0m*Y{O<
z#7k!}ux`kl)r05#d2Qsps@@~)qCvZQaUg`6W@y(&+J{2V=4N!yY)=cXuJom@c8Gqg
z*`qiTx2soqVDj!;!)NpQFw{5<Uv%jx&oi=mY}hk+Vfvo$u3K9v!%L?o!((3hqa#~d
zbX)m)k<vXUyo)nr)orS5L`x<l>?TKc_Yx*O4|}(i&?SmIY{oKBgO|`vO51M%>taGr
zhwJTsl_REgfpl8nizTH48<NC|0zWTrZ2<u&P2T=OjAdzcjOg6#azH}S@LtR<LTByR
z2BEMW;+uXzSNg^wjVgy01*O$v*%)(2v?12(rn@~48+lvlskl0md*X6Gcc5{P0xv&p
zwQmUXk|U!@L>Cw%R$q+cJS)?=7%K2==+rVAJJMGg)i#n86=nK+FyYQygEkU%P<A?Z
z^tdwP*`w{Fgz|WA--1W$tfGgWWZ6x1@gM_<IC^Z)r)e$5EKPH1*@!7uSK4&`5!9<O
zXQmhQy0lKdIx4~U%teGCUX&n18hl|sgs!h=KMC&6y=1Up9~O&b0e*E;)HEsx`15w=
zu#2ndOH={6BEkQ4O80YEff%#|sJS!2nc6&O28~fxN$5Nzz`+0$EQ<lrn)az><}5}b
zSh-MsBrlCcd!EB&fqZp(m;dh0=R0mQ2xyB8=bVMh6&Y>{Fr=p^7EE^}%rm+ATCxfJ
zPTb)AZ^t=m*PY4NNbXB>RYMbz1c2N|#xQ13-NP6FJ#HKTXmfY8%B?RZHR<)k?{NZ-
z>k#lP+Ge@ncWIzFmsI=L9l0w+6Q<0NVYCkvEa`GQdAfp$v}?DO7R$nG33KvMwUt*m
zAj43N+Pl(>Y<G%Tc(K+Z&P94&OqPoX2)t}z(Qmc+;a?su&U8(MS7&!RY)XMQGwrfb
z_uAbV3S6DHx%y^48tN%l?CY>+c%z5ct;mUwf$MaAKwyUrbK|eiYxFAQOlT7Vq6-%h
z<a<7^iGk=A${YZ+2bQgB{mY#vb5648`^k(m52!pJ%v#onvqR6wcz0%U(|)_AbZH~&
zNm$W{F~`pGt=OfE&dZGiMN=v>!jvR9C$C5wRJf1B5k93%ELV`%sl?QKstlfufg+q@
z<KdhE1B*KTd+v`P$rg_?%?a?L!;Z{Yu~D-g7j9F74*1L%=NY_9iw4!sgpuP#ha9C%
z<{0&v4%d{d85iLL?~eS<QCHrt)3aVc_SLI3-B|1S+zJep`OCSki(4f=X9F)S-sFI3
z-4L1?A_@wkVPJN$mpgi-s!$?K#t)o^0^HE65d~g&K@KoC)MoLw^AXc>W7;&<s-vd&
zD0A!&DaO`(K-RPPz;*6$>MM4UDDjsWiuTJNl;bNd(GG3(>@!POEzj}q@k2VdQ~FAW
z^f2CeHu$_oW+-_fr5o*e$As7W?xh5Mb)?NCF$H?%d*SCsMfrQRs>XYM*vYyT;*2ME
zu~w77CJ8BxSse-B<k<$<QM{p0{r+n^{Y@4@Tb7pHKcb2wNn3wTtJfSg)*o93B^fyN
zN=cVSjJQza=w7VCvT=A5XEyU!9N{+8MD$xC$g`DpcXlo)hqKQeSm{T#B1~G>t0Ns7
zWWZJmwe@sRlOA+9t5QvRXzN3Zj-B5q<xplhxSLzsD{R9q7cQ>tS$iqTvoOc)sy?nS
z0uKJr;lB>i-abtgA2T)VZdi3W5D^2kd=o6ft~sl>n@V*d5MF>HhQdoSlMqVTRxaRh
zQor8ldL_kjPioqnJSuz_c*T>Z4S2<fUuOpzA_zWIhvB~3BF`qdwWVdy3VV7r5<YS|
z#R|vX$Ib&0jU8R%W+nOU2%y8fdU34V)&7LPDr{|s|0*Q;nU0{vh<G2NHJWwP^G0nO
zuGxB5Zg-!%@dzrA)Bk4gvkDrT1<Gb0{nE2E@Y71~%AzM-xU!1mi^=e1$e^1bH}lW-
zRb1>>SEdd<e4F=Xr&Jk^8TfGYFB_cp4w*aAcm2Xiv+~^kpG1R|mGyrk8vkz$nvvuG
zB88)_5qCI>=)0>&xbBhfuV9Byl4WBIG)>ktfOWXOj+h9bD}07pfl0;#g2$YT@pD^o
zVX3Pb(>)Jd6<1kX`*C4@zcLhA_w#eI{W7!lDioe#K$d!M>7`!bjvB&vXwbY_T9p@+
zj@h~+!U%8s>+?7y=hpUm858YZUoaQv=GXNtaU^F&9F!Hu?&j9!hJWLy{@eKdbNwkl
zV<MjIP_l1M&b8cr3i~>l?YrB{?l{`5U@+{CQpcv)G(1dlD4gQfLHqD~yaHU|&>TbC
zL}#_hj0G`D3UNxkPdADBN8d*)u-J|MGA2?a-V}TbeH%5}8EL(+-+~kUvUS6hHj$uc
z#Cx9Tk3~!ES8XKQ(@JmcW^IE9tF^^z>v;H;<>tydRm~q|Y}8}ZMHZGo(Z%>u3!^+i
zXiTz!0SiDLhQ)xAiQY+g7o|mlnf>*zM@<l)v?5Io4So|?g>56TZu=KuF|;HmO~_eg
zpnicY4_4#kdJc`sho;l;w`^fX!lZ$zJdRT{o&*MiL4&xjgIUF}X_Ea1mb-{>;?3{Q
zOv!N+{)NZ1NA>jBKP*Z@bQu7zk(c^iY<u!dYe(jdnp$ZY>=uVU{~Ecb?B)IL^oCi`
ztK_lCL`51;?Gp<e5c+S0b;^PdTbpFW%R}Fda@2Bm|LC#=-AdQ@qsh8k7hs5l-$O9M
z5i(?K&Mtzui=*jE-yHk;{o7#}Cu9Z~jfWs`&3?G4B{(tKOcVEm0hl$M-^;iA-Y+IZ
z_?^+swF+Btel(8SkyAq8eKJvXIU*cj6&M<B5JHzEO-}@tntxmK<2w>{ZQefAn+=F;
zT91A|?rJ(W;RrgEO;Zat);?PiGbc03Q(ND9Fi~5z0NclRx(lU^Gbb<3!K=UWZXZ&Z
z`dKPevrn;(Lb3GZ4L9K6-+GZOpU31>w5P(M!|;!XtNXlJcVi}GY4}{Sd%v~b<;&@K
zN3v}IY?qX1O(2T{3>6OBmjFfheBTGPfP`zro$*ww$mccv#4*_K2l2}8)712yWyB|X
z<}6aVbGb0nMN2Gh(`Z^%uowM9?;<yhhfI89v6@~dt5Sb_(n4^MdCu@8sNBv%hCE;8
zMN-Mv$>Ihk4V(!Yxo1f^i3aZiEHrs3{9~1xXh*-L5g1@(lj?X^2xTju#J1DxDADXF
zO~`^_1O23~V81sL8ErMmC%|XCVw`3VC2K;&mfTR?ki%@mvCjMy+wV+PxW`xbvY~(1
zU4l!n!T3?|?5!a#;*)QeE0llYb;-B~wXo8@W>N(mszzlRkqqrp)%%xstCUCgk)LLb
zDqxVv@V1_+$5|C+NVUg!plKZCx~1J4YneWsZ{O}ZiCMl7uu-LTZgZ6l;)4Y40qz=n
z+-hLDfMr1D5A+d;K4u^mxp>>q*O4m@ia(AX^#1z!MP`$+bRV4@d|?2`#4qCUZ&i}c
zkqkjLy8${*b<mYM`*uJ31!$LdIWTC1&#$|#`ccS8RI`I3Hwp`!L1O#^@)zOnl85!M
zGm{Y%r!xefPfuASGT27ltRJ3JN012$k5{(bgfW=!oKTRPhSG|U!V$|8B%WlRAi~5H
z2*#`mTu_5&i27ryS_w)?J-^JdWKJ2hU!r1wX$~mSSLerY2FoZq`EJs$Sjj#%`8hV0
zw>bVR5s@!H@Ez><A_X;IMxZh8aWDD$nhB)f^NN&(lO`j-+$b+?#_3X(78a909HoAC
z#=%purqC{gLp?=WEnZw0GfuB=T$zuOAfFLsFwIknF_q-0b(n1^QEQ0W;M)a85%yW%
z&hL_!mw9crM-?1V^K(qx?Y+O>y8pzJRZQI*BvIs{t((c~=hgakzQdLNQNLiq@l49R
zt}iIR|AhC1*29+5XJ1IIR^YoWw3o9Aq)o@nl(m1hn;V##wAL*>-o0Od68krz#C@c?
zQ^YqOBEZXgpq?M@Z|jZq{m1?(Y>&Ic0kjBB(J`LW(7uWD=lY&dH2eDwcN>=0+tvDk
z*1k2q!GXpQS^ddg*Xn5u_AdU6V;sP^i^us?m-G3UOdRZ5ad0{pjpg!in$LFpeKE3q
zLF(YOMz?aD?52Up_x^Ey-^zaCck`_t%15|pgf=9g-nE|LTpTMOWpbkMk0??W`d8;C
zvEY4Z9B`lf(%Rx!j}=Zx>QJ3uzjU_<KmUmry&=?AMn3u0(TkJAg1P)4h*~mU#SQOu
zw;!D=UW5Ub<}@h1tF30&ms3R@XcQ(Qtp+`}<<7nNU7RM<r>-Zl(cyhD=RPEvcNkPM
zcp_f>ZJA{nyaQXy7+e<dnQ_(9Nk4!((eAo2swS(+VIN?VufQL+@{r%<J#LqeGSho3
zydr?S<{!UALdUO9j_a&@6PcDV{y1<M>NT3jy`g=PH<UO+APmiZ`aA48!}0k7I~^Pe
z;%mH@w>+elU{o2|(}i}>B2>x+m2JWVkWcZ~wdkyA?vK>(A4KJ7;eF*{%19vX$5-*4
z{in)YD+jvQZQI>1d`KYL)c*k3Z2!A5@c#ie8|QyLX+NU=$LW7g+JCB<XX2G(LIXnZ
zE!_{e=*|wilBs%1{)>|bldzyvBqrmKC;je#qUB3ywEkUtDjq?K(z&|S3WqZV=5lwr
zyL{U_pYn%SOce8F!H@&z!3q`>F-L$K-PkKaCmEVMWS7%p8s+Lz7jYllgJ?QEdcN<v
zk9r+Sz*nRv8dAVexV<{M0^v*>GrQf%&Fvdp37bmw_VIT8aK!-MaUVyzJ@|Th!{N`?
z$&S+_@NoS4xej65_t(s7gE0;M=r02{wKjt^5ahz3@2@qbi`owd*H)4Um!`p9C?|ZB
z`n(o9=2Se^R85&YN{yyUMnrJkvL?OM++~XCNizwNepjJM?cfPkL9sL5)z-}>evVM&
zPg~7k!6UH`3eUoJo-QMl+*?ccqpd2`KJFmOKx(3v^sO5(brRSGo+h`z=`hDBJR~z&
z7%!~w>fL%nkjYoTx+E66iAIcyz!!C2eBGxgT>g|qy{rncreb}E<ea-ZB_`#=NNAjE
z|7jTFfD$LGkJ(Q^R>=2QL-JE=nT|(d{lVI`7<|*=)}fd8^E+*xetanPeJQ6_^kUCi
z3+B=Jt4lj6ePZrya}d@%Ze~)F`F>#Dn@#}ztS6&F7v~B-YW+Fvo<D;b93K{+zNdab
z4tX&xLTsPKIeJc>*>?%EiB7LJ<GA7sCosA)x4Qz*RI%2j7$lEoy-4j((L^x%#FZ0>
z9@I_?S+fIAB)2j?r%YSxlz~h1*=c?9CJ25Oz+wu?!FzRi;cgd~;v1isUgG1RU<_~Z
zRFrH#`pM)%oYwNuDu-_dC<)~EN3R5X(Yl2J)R<0Zj__bK(zfS;CB*Z%R+o|XvLu^A
z=Z1;+*VppGoRBgqBYon^oa#Z7kLdfKJ;rGfk@TE*0=?)?Z^rM#uj~gx8v(%ld2??4
z84X!qqwYZrmD!jR7tDpuPk7MPMwnQ>eTPc9VWqAfx8-zY5l>Y%XIY3Gh0alBe9V<@
z&|*_Af&47;Q1<-7g}5q&N`$ycN48lg4Z#W9M+YyIYY;yN^c9*fOg}B8UXs$uVoJE~
z3HO;^Af#HGjZQ>|8>;d2@?VkQ^KgY2#RjURy+Goozz2nkwww1qUoj;(jV3?G@aQl(
z563XAO|~>7=dy(ZsOw+AGR^84zPFRY82{M)+C5D)n>DGuUQZX^L90ZT4Hl@pvEZ^V
zTQ86DI#T<48x5O8&_f;bE`>>VDIs$wUJDyqxgP1&Bq^P&uKRK1HHP<ofg#KgD8}!+
zuz5O`b|T&tqu3qI<Z*i|3HBBSj2cq5^D~}!SM9y9;L(c9*+R#XQ!k~%c*g=Y7$6gA
z06x{chG?%cSf=f<&W7hRuaNA139?EpF98d;Gclf?krBM*heHGgl8x2}U6^y7*)o41
z2fbEwV<W&=+Uc$Y{Wd3gc5#3}L!EuGzt$(8)3F@+CX47k7}i}<m>AWkZ@njUZ{Lj$
zxs2lEJ=L87dpAY5HKM(ZpY^BL&s=4I6gQW<hfk{pgw4?;XH%P{!E9$4J?moDd5TEE
z)jkAI&^}i4!(UCLwPEqTLvsa7s;;Wrc)GWEK~IV-kDsML=C)Z~VD@XYPh1&MhN}mE
zsxH2U-W;Dcds;mYvlY}K8NfOZZY0ZnctS{kL}?-=kv3fe>>-!t&^^3>N$2#ba5^Y1
z7Z&P0ose832^LOPT(JoXYM(o)jrnJ$3$<w*<_7IGW=X;~g;hmEkN%A%2u%&f2M5R)
zGLvoh)WatkUg%2Qs79C(doimeb-<b{O?hv^*Vaz}(fW3EvQzlQ<?tDMu3NSuZ|i)v
z*M{kUAxR(Y7^MkXfz$8D%sugdhCYx4N{Y^1WKJ$CM_s~X_tytam-v1QWvEPc>jp3<
zl}M^vz^jLAk=V<A7?m)XV>+rZ4!{?(otJxf9VQKmOn$HhjWD3y2zbwSZ14sighFo5
z*137}O-ya75U}}NH1x4M@!KA;fn6GKe!=H?-Sd0g*rYh2B!IZ9&kUg~qI9DGv-m3f
zDfav(Isy2!fgF^2?d4dhbN-5y0Uze7*l-1*{T^Qk1%5dD#q4}}-erd0r2}=C^g8IG
zB7ym5AQJGk-Dz6qGx9y%dFl9<W`_G~h5lwkyv76uw8w0SlDxnk5j|tEX#`0kZz0+V
z1Yx<Ii`rks?AjkDH@ySWIXq!r)A(YS*zFAAZ?bbPtuycNOaSde`^PRbz%pvQn5*5y
zZMz!-{|mByiUagN!QFqlpR%z3*EL<&Y8wB9ZiIiX@AWV=bwHiWzgdDP1Tr}+Awggb
z^Jo%4L92>Lhdqs9I&EkF{9HpVt|l6;NiQ`B`M7va_8hy@@myA6Ty?R%dN{a!#?hSF
zck6LEBa@JZ<WNRJL*p98vtI0E+0)$=r8g6HbGzNVyqsAcz9KsC{QP-8_789qZOk`a
zbGy9!{~mmNUQAyc?w9VT7%-+{DGCIEDr3&#FL}k6v6?<Xr_2wEgGYH5{_gbOyhBQQ
zL<>Q1(%H^NQ6;@kM^U<HDJ*x0Xc`yZ#3vtXR-3TtMRd1R;j1H9LT6KuMt{VPVcai)
z97lTVuvTuNJjygQCvt1FUBGje1>Z{Kr*ev(2)V{+=GH0GouW5VR`JcDP9&@6{duUU
znH;*xI0Lz1Kni_C<n;^R&eVO@*>)oa<Z$?z$-AI$w61&k3zJO&%D8O61u#fK0{P~v
z;NP4?4f)~C?@%{@n<=$WSE!VtGoegP=a}IGK15)z61G%1Vz*VngU?ZCPhabd>YJr;
z@e+i!fZ`Lgr9zHfG$y6!Dz`GNeqq|kcxo2hsBi3hmgFC1-n)Q6W}oNb5@X!9usMWL
zTp=5QV`jAg(aw%w5voRXYhs029SuE2-YCZlBFv_fVm~e947~I#M$0ZCG<`f?p#x#Y
zAQlG$M^LY1MUN3y?vEcqRHnh*_SB^&KI;YCdQ+<U$A}9?b3-P+*np8^6n|NS<YXF>
zP(JZrN^Y9NW#EG#;_r*%N+y)ucb%E&j*1l|8!HE!ZCajJh=`SkRflpFwo0vow}N1P
zI)O0ABL|#wGOYLuvMdsRI+oJQ2xn#P5-%bbpA>{7;jxmdf@zpC7$bP}Z8rW=kZUqJ
zNPxyG#xNX665t|>+20mO;|CiqVy~u+92(OMo>T;WUumQ)j+5VJSB9Ev2mV)a_7hS2
zncm)$<hOc5kUqM!7?c`Fy&&2oe3k`?bCX8ALJxM62ijm@tV9M=3^lu_24}4BK|%|l
zl!_Fv8E~m@lRP&L4Yj==F%?+2wMHh{)YL8nc8TbKF`Xn&ebs<t)BV*CdkKP9AVO)0
z<PW7F==;xipHm!Qde?r7;E+}ph$8Q8^v2~e$c`>1W|aWql@<wON7B1-28Ik5cHV~M
zg!xmY_voFCQOpQ}7@1Z<KcxC`9c$?xNTus)<IIxg9f#CKqkt?ei)~n?6{x<nhm*}z
z3_Es>%cB&q0bO{3>183->m>0KAj>5G>t?)-@qSNo7S$qQd%4>XZetjgNso=osRiV=
zG=8%M0}x+I|D|83ei+a`fYYZLT3<hukibqADusjKS$R?b8CtFE`H?|{OEF`pn_8cI
z^1*+iy!84!IxXD4-o$c-10}Re|Da06a;5v&N(GM1^lSQV$unGr;Cp`%Nrsg6X9&~B
zvX4Hr?UN1Q*!f%u*XgcMh$QoLo0@hwT{}7GM0wa}e-dFYFvMCd=-u5HJ=)kATnj@l
zano`}%T*V-5=iI_A12ilRJxFx5tbCz11!qzHcY`c4~Lg2x@8R@$%z*zE}PQ_J@6O)
zkY&Dj{_v_H6%TS0G5vQ><J@PL?QvHvt;<r)OGun&7gKCg9h<70hpyn(zRBaCOp!xM
zHqJ2is$X?;NAuD#oL3^JYRVOz%iwG308C)KQXuROpj+z`Ih74wuYMf8w&tHFoYK*A
z`IKvr_=+1M`a7GXO(ruvHHf8iU<S~suYEy7Jv~8~+q@gi%rycSOPXgZYl-O?Fv$Nw
z**i69)^<UIW!tu^x@=orwr$(CahGk|UAEa}+qUub7ZWij^BlbM1NPaDl`HegmDwC&
z3kzRiI+sy^i#7FiUdJSb+DNTpNx<Y)N5kS#aEElk#nxt_*19E<5Yri6(+y3gZN?JS
zxHm*4X7mx34aMH?;T^KLMVRcVs;ecf>xccRkNBQvQ3r~G4RVhqz+ci@<1Y}r0_k(s
zM)>O~Q2bGS;T?4Nxxk%nv%X$vy?eh1bpIy%%Ea-J1EH|{@IL^7gY~}zOV<CH|I}z~
z{WH`c`QPa8g~KNVrjt`A0<^G1Fq*ByDSl~T`dUN_wpeCW^_0dJU)TTmN)VX-jyfpq
zBLGP%JJ`D8_j8BKv7vCeA1&I!j3*F4{FIV08b2~=$bEGp0g<vGqKIBPETTMpeAu5I
zoy9ihqo;@@WI{6Zq}R*t)#UZ&$#E3olPYB)u5kBs^bAaPKS{IK>+X67b-&rUN@JK|
z(C^6%3v5eo9P{>ly8na`(CO*X;vGC1e%ny5i-nFaZ^E`I*drrR87XKmJ&wt^OAu`k
zP10kpl*Z9LlrLo&p?=!6=T55kh!LoF4~c&rJgv*azoDc#4KJQpM^T%if*fWUfr=mS
z9{LbKV~~p=91cL95F*WCM2#bgiU8`yKK0d8*~P&2YMGf($!K|OG%A_LIcwAI-XwWL
zE$I?E->K2_lXX`XH@<<BBz`%^coVMsr7l%u#oH|7RP5(Cki^U2#=XPKHcC=QouJcZ
zswt~^TG7#pV8j<JBIRN3*rbXr&McK>RAiu^k&s8+pcXsHfV-d}z2gB?i`VGur)s5J
z;l)!X;e#$pW}x_Q@|>`o|7z_Je-lGkX{{=)aT0+us?*5w(dKzq!*2>u6e@<|?(YuR
zX%Uh{5i2yF?js);Uy_HsC+0gym&6t#aoleL%19Hl+i$j1?08V=aWZ#>s2y#YSrJQl
zYNv6AJ^EZdWgR}UEy)t+qQGS^2d7s7XHH6nvpVltXSO6np~uHc7+t$wFU6&B%`w2p
zWFu1ys%nv2>M|w^ll!iWT{=FXJce8*Bq4#D$D5=PeiPeBwp~aH3c1k#{`&(k^<_a3
z3NMo{880s;`h$F^URMe-_61^Bl%@bT@)gL5hT(Gw13M;ruS(3+pXV|`^q2ilV0wGq
z@=KpR>>ZqC!cFXU(|SG-q`1=+%tS?C3>Fmg`DKk1-b>|lyf4a7?fU%wa!7`^XO{07
zI?fB)S_01~@M0vxngT?T``aIRVfrwiWn1xkc(RQgRHSXj7zw~lqZHDVcu>x8$&<8W
zkvc|pfYIhJpBwhIJ%Xexk!*yLzHU>kMVc+wJg2Sv5<hjY&7)s`Bs^Z0gr+{$4yNGy
z+)i2jsG!PFYJA;_Ds`K-Rw~}Z#2mRHfpv^f`C)$1iVMLS=g5=ZkZ5;X!y#nBtG$&@
zt85dM$W-Fkv{`p?^zqPJ75k}HHb1A!x~q4d?T+|b^-jItU(oX>ULIi*RRg(*u)0Hi
zqv2ON2PCPGb6h_|&aTfw+8!^|*tHgMp5j$@@{7<cd$lbfaZpv+ePEbF5W<Y9m5oCP
zjw_zr`JVC;OK7KUi1MH<BFKM0D4uk6$Yyl}uOj&CQqjuv!?Yo+;_n*$ux>s<8Lf!H
zxI)o;-Ya9#^2CG=Oh>cJJ|POZ&2ovanWQ8mNB|FHnji?Fk3wV+6j^2*t6OdqU2!kC
zHqpk8PS<iOOtNZLqH{qE+N-q4N>kdXH($p{JC&7>ZQ14{Zs=cdc~{Ih|LK-ex>6B2
zT<*94JlU%;G%rwLozD>QL@#ySG=AgQdk?{1Z^|lJRuN0_z>0?ihpZt#&7wy9Q@aC_
z-dc{%5!7pEX7FI2_s#qbwV&Ms2UQM{I6aUA3&Z(#cZDCfMngAA6TjHSl>;qcrp16!
zTXAR$g`I*vOoNO=t%U_Gxox|Z1ezs87?zZ>n3e6A$(nb(+h=!;_%rQ*fvQGKZZ*}z
zJN}Lwm}!9ikRuVOF1h!tw@ainY5A)cjYE4eK4eNWY!Tyf;7UmdHgoad;R->UBklAF
zJ#1VL@4H;ws4f2RiwLds^t3PpzwH4#rl<QPFw1Zn$78otC)G~Z^0QlS=|A$h2qFO?
zujsEoX{kE9jRq-}h*Z`@ZM5fyyZGI%k!_(dmfJPGv1@bhj&pFUKXlpRd6<m1U8K-t
zU1diua0Gz`YRszB)Gf7~BaP10zATh)<=)3NO%*4PQ{6_P31^#quT{94ybMBY+tr&2
z+**Qr>X^{i<jc<(+QD}8)gFtV29E$DP=$gC{(xl<Dl~Ksb6Kkr`ZBtb=H%1_2PV7&
zlOjJ9m>BsxTwfL!3ijRp3$R&r$W|krHkSPk$z&w58M`TTQ$m~yC%ZjJ%+)l=TdY!s
zX}2%^&Chv8;;m&oCJ9BgAkgMdnhC|OOO;iaUBcK}W;RKuyc(O1(BK__uQY&%U{@d6
zYzmKfH5cEFsLWECA!KVEMk!$S(o^<gSUPm#TYxBy5RsW6^rK;)9D8OU!yZerOT4|}
zb6Ijtz2HO9=UiQ9vU4rJC__XHKkUd+@O`Zh<Pi0QM=OjH+E`ZZy*YSbj~M@Y&%-M8
z!sRZn@)DB##WQ1IXC)62IEtqQO{m2PbXcT*jZKJz{T}?#{X71L5X>&c=jVp#vRM5F
z^Tv#!!ylR8+P$lcUz_(Gv8hq-=08A~^FJb6*jTy#A5-i9P(c!_BpbiLh}8X}{@O>(
zx7d=zNhryM_E#*wR^#e80F3IMLe3g(RPoYA6X^$^GKn5sC~d*#bvi5KgxDe;%B#sE
zg<)AeSt)s+^`10&g(?CGmQ+SK>>kGVktBQnC`xQ-!*5k*=d#s(GG#WR4dlV^#mgJ^
zdYbvEd-}?(ZR2`-TbDuL<Yrmj&|zj8ibCj8OMWrtUm{LMxDZ$wwJW{QqLAYuh&Zau
zAllKy-a{}J&s*-&RwXiGA@*S3#>?+a53f@=-n6dJ@DFT#c_89UiHIjOb&z><viNy|
ze7N~i)Dpg@V92eRkglcY-3jAkE=LfY@64KI*`oL=N0!Itn1y7<Y)+nEBpZB^6!03`
zSbW|F1RkWc4!>*A?5omxEnOS0UnFBtWC~dc&?zj$j4vgSsbW^n8x%TlQlg;lyd;Xz
zU&SM6iu{n6BTQ4eU+Mh?ZzCc0nn7UkvPer-<&B7H{u~8VnvymuL-HQqVHRx&P>t#2
z3bqs2aICU`NTL}#>tu#|Dk;7U;4w?nH(}Ryhy#`4t2iEZ<8(|>9_oiaU^J$<Wp^L^
zvjbse{9xsBcyD!zD6R>m0HI&$wUSsvS`!t#t`wr=xDGZ=+DNbJrD~FEkw$q!kD<%Y
z%k7nnAN$6j=OpfkiaQLgvc!50cfP@H&T3vq%`aecl1F_M)Qa5<`HO!QxxBbct->`G
zs-Lv@cdpv@yh5K7Fi`y0U-w_DA1e#X|LUc#QU4wPf2Hpe^;0$MSS)}SrX-?NQodLq
zLE|YoHAb@R1{|4Uiiv$|Z;uE<MfghACB~E3;3%AV*n=<L!Y<O>?jWq!ov-u5{?Z2k
zl8!RFM9QHFai2~y3Q%lHuz49IZz#v^zxBy(36@79j!N+TH?~soCqnUT*~a6EKQb7Q
zj6EBJE>0Ex#cCjw4X8E|0+^A}_E~_6+s8Sf-mxO?uy@86SvV5)+G$X=l_{#UXz#qY
z_$h}_QsTxELL@)UBc_iFD-ycYjTp0Zg+#`Lxl6S0?O)gK)RQclw<L|6rDP$w0}?pn
z|B4y1?mJLQtZ`JT4wkoaE{!37@}A=(NW|{UCf_g58u~b{o&i6mTu&#2$8JkCMjGXY
z_g?q5upul3(9{q(8dYsHeZJ3s$R30qFj2Dp`s>JOQPgZky)GqPAAB$<^;n2sqV2X8
z4)0!B>sae#{f$F40_A;-RpE|Dx7>lkW>>%BXqEjPpuZ(#;}p&#eiGg%h#^!DIFt$#
zTJ)l|s_SN~czLaP0tR<4N-0;V8To)WrXxeVUQQu0C1X<MPcz5}`$&q#Ef$KST2<bK
zrRG2;l<TfzXj;Yg><i?CRqf=()_#5#WE^iOH(*7C2@u23*5C~H&`aPvwvN}Kp1kRc
z6iGCD(?XX<xj>Fn`@K>~Yf1pxm7y%UVmvW$PS9H>)B^c0?i9Q~fX$+vnIja*G&xK{
z(>*cZ(vO8`25Wpbpd@Ec#nyEp!VS_w$|<hRbQ9HKhYQV-bi=W)f(M94dbj?(PAuEJ
zcsI@)3C}-|tZ%xM4tBOMte{iRJ(#B*v&`#pym_=k8{;WoabQjEWd^M}))`o%006nv
z%3gL%rLVQnA+PBl8P54za^fGTk|V4aDdvq_vvOm3a#lWrq)zJE3w%cgQJ-ji?*>>p
z+1y726cmH`=@C=NF%NKQC%VD$!jr+Hao(}c6}P^Dj^Cjpr6tpb9mc5}CAFY<=yBv!
zei69jWjzi)U7GZ&;qL+xhYlR<pIn2BgjjWB&q96ccfuRTOz@Hb7*!}O0qlsPfKEJ$
z$Vq%C9wwY``G1s}0+|AMGq!tsD9~nEG8Q*=Eu%`6rESHv=4Os#HJZD^SJS|PAO*q`
zHGQPtL=`BzpXCy1&(hiN<FwG_a*lW<B6>d*xe*2Gp6=u0j?=cv$5%$o*keVYgk?i+
zlAb}CxW*gw=P5*pWkK}Pke(S{Jk?O*j-?UQQ+2#&sjmlya``Y&0^1T%_1<g!_Y}w*
zvyLs#loa@$Li=;Ed#S=GwN8YO;w!3Chh7W6Y6?L-;@&}$z>y*N==j`cvo7eCaUII+
z{o9svOE812Kn&-p=na`S=`s4#hnf6T<qP&<My3vE{ziBuW*x3f?ruS=YxZ<cgmm5!
zy3QRd`J9(W*gbV9=4$<*0X8tiKu;Z!;KH!5b~<N|L*KJ?bvAB#cP<Y0aT21{WTW|_
zR}Ad4TN;ml{C3yR?Oo=Hlnj<OOHlaVsq!8wyv!*e1k0szAA1maNsd)OH&uw1a8gT3
z8juaEg@}6;`gna=lZ#tI{Z|=f|G2WKuvFr0z(X=xg*W9lrUJO4(JzTx=7&N-x;KnL
zs@|S@TNmlmYt*v+Z*E?V_!WvWIbYM3+UIcK;aS#_UW0of6T}%uH-!Ua$|>Kis1BZ}
zC2BNoi|1Er7he#a)0*0zYC1-2rf)jf?k=OyftwGt`?ZMz%+B8u7I!uhwdrq($Ft0!
z&19KB`%%#lr&xUs{yZ{IXH&)TSrNPX!rAE7<~<QLwaIU!O^l7<?6Uk^1Z+5;42y5h
z*QQA3p4Ku@FFU(ueg=YXp&Xd%4~L9&eQ9J)Ywjo11=VmQcHaS<CCPOv+gn|)m~Vf@
zHU4YTbNvU&XZ_zvevbNh+&|T~dsdxw_cv_41${U77<n}ONdi!yog0yYmGpiwW{K4F
z(1?#u6kRKwRdK^UYV>b(!f4ex$xIv6%&y)KOUIXET?#<vA~*wJm;zZ<sGKT^H<*1g
zNi4}n#+m8E`fdXsUdw))Ns9u#KuniAr+thHxn1(Oi%eBFe_p)Jik*Q<)_=&XWW<lN
zxrQRP8L*jE;BJ=WDH}!e`#1(!FQvTN+m_TcZkjT4L4Q%4Kns@S;hnWyt(<$J2b){K
z*@|-trzCUCL6v#NSy8vmA9}~3{v)`%b-bzQz^RcPk%O@4uhRzN$oUg9^XMzbBj|vl
zt+=ThGMxC?^apNtXlxsW+w%1JhoeJ%+N=&Th24V~XLKT>Rr@uLv1zXkBHwP5jh%`0
zVk{=s?Nhj&rF#)T)<@WFX-cz@QJ8DD82Y!UTog{`e2Zq|Y2UXIWPnGGP#^Fv#8Efi
zFX_RN7Lt~QMkEEjW#6Ked79v@t=ou!A}j6WpiD{APJD4&cl<c}Lr!DU%FLCf_I$Ap
z<+-}~dPtxQ@9?43-w3kA5#>nlpft9wh<eB9vL&PX%N44|W!{$YEMJUnjkW@_ST)Mb
zS?0}fl<rC42uj=eH2dp_MXLR)F48(S?(<=exZ+`Z_ITrzb1E|Bus{f)Vu1;4M4%cX
zH*S-a0i0F3eI4I!$+2qP!?|7wSHNkgCuKN{7x@*_){B?n*uQ^G95Y@^*zT@gm&cSC
z#yYpHtd~om51~@=1Eh3w0x;02o#XQN#`sPTdEwKB=KT?Hjvy%62P1+Kw|YXq{2g?G
zDHViV*SX9aPz}zJ$JkdjscjI=8b6_gf|aUd6d2t?er0FP1cN0!1ep|y9~z7dW@mM4
zUC1Xp*|%2eM#%0Jy|)Yq@iYg(b<AvKjQK{q#td*PVTRsPqqiOeB~?xw@R@e*jHf*$
zU}h&*$31Q8LgIbL6>R;f!dGNhy9zGu4RM|M4SW{<4Pej^$C@qEe$=9KwL2~OJvY(L
zEXtmRaq9yA(hVQnG0J2BwniOo333oWzZ>$h=dgNBv_FI7u$uwnK@lWG%=5}pot!wp
zb?8EVJVITs_M}Y+&v+UMJwm7V_fBrJp<inlON(o~JX_r2D?q_Hv*@&Zu^{-;RYkh6
zY^a4Ylop75Ps}60C*UN&t;cn96Cv>H26Ah|SXFa8#jfQG@9$d|k*ePA2e}~|vKaPC
zAmVaKe8eTUvBMX7guMK8ue`ir|3VT^%Y|7jl(Fc3a++oNOx;{e5jJX=A0b|jHxl+Z
zu@fnhXFc<5zdZIPDI>H?&Wc9LQXvp60=$EZ+@uU~^lWGVCz4}@iylL%Fou>t;9M!?
z+t}W<tk+{%a1M1?zOFcOZ%5Dj;`5S)-=uTFf+%|>{|BS9$g2NSB>fYIy>_J^RboQ~
zm7*nsAgYGelVC;^pF`C^&N^alzZt~*Xl4F90*AB(mgVo=)|d(eK|#3|8~n4IAha^X
zv2J4<Ly`ux9L)@4P8X;?KO%4O17LKbY8I-n-#wC6S0M@{Ef*a}Ef{!0-W}Xj59Fm~
zS0^|p;m}<!7Qib%V(XOr<Kv&^^EjFR`p5VI<P@fh`B@PrT<m_R6mws9bj@U9g-Kly
z{KXDWR$~#bgRh59Zd$NDB-3L?qIv?lYM^MiTu>~xFF!UN{ayL2C<1N$Y0gyw0#sq5
z>x;k+X6<^XGyxjI$Z~Dm+kOA^$7xU3TT%7pq;g;BTyH+~LUWYC?>)$hgSE2zQ$%}z
z({*fjZ41f2SeUtjmd6X^V+2QtFNGklnR0n!$?QFP^VZM<oeHOg-aeHl%JpZ;l;~c$
zjqAAkt&Rm7v{4J8>f=?QARp+Ha8u&9VD@ZHJM+8lAH2I<(4v)I=UKq+%J7!|913&T
ze++%D|A=p3V*md!^ffl@HrbJUU)1UTss0_4w<L*5foPKd8ixPsB+$Rll`<Zp+p;W0
z!Rj`;{of6r{P(>V>WP$&7iacl28+nT@KO-HAD?#*m-jslMs$n`U~>4_iTIEtlDxpN
zaTP7Pw}?#dyXoy}oOWbYGRj#z{V;zJw%g<Da`5<Q^eB^(Rmr4Xjx2WDf3?f=<G9ZJ
zH@ZX?$IT8GpE;`z9885)ama)WSLlV_K_)H04CUz?DAP!~XgT$yg?_f~yi22&{E~MV
zaX~;SbW)zRp!amLi&k5M*&TPY*(~N(WlOzn_QHo_p{4X}#kE9Pccs+Q*4KMG{pV3s
z_Z~|%DCqRW$C^P#Pj*q~*tCf7pg`SJOILo;Vb!-;NcLMSk4M$h+GdG&pccEFYP=E5
zt^oEJ2xBy8$fsegU$@Rib3M4C7SF?DQO~-~;P0|p9@UZIhDuuNwkL}NX<Zs!tP7`0
zy+=7~ik-u(UYPyLb*CHK!uNi>fa8hQNoeh0Mi<TXp^nb#m>KYEF-q*nWZSYotS-@Q
z@1bc~RvrbXKhzEVev|`tW%XuWUR~c2Lr2d%cy)geyFhA|t1wk8wLWxf;ZR9R6e+TK
z$UV=27|x(B@&1j?MIqAYuG#j!^x$yY6<aB=Ulv3!mtU(}+d@j3ED3DuQgG0}a~_lc
z+A(zoR5!-O-@`?!oCkwnbA~rOyW%mM3Na>*QFb9Z_egF;`2C`fTE=#z&^Hg8!4-@f
zT~~?BV=k&kf@;F#ka)@LyY@FN>(QSB$)(-H)4`T?pj_{pW=b%GAKOOcQ1U~IHfd*&
zG$yyK(UD3%UUqC?qH!Pq6VWIv^85RVWrKNy8te70E1Zzd3c0oM0537#1$K}F(Bo&f
z`eZ;z^x54~V$^-<rA@&kl8QwCI@1#Lfk&#4emXXX$|!UbSU5ez6B`$nLgSfo&!qiS
z?S^3k>V!BMxsjMC750!29UsDCKrneU9-0yQx!C*pkSb8y`5mIUEocu;HS8H^APVpm
zqBxNhl(BL!0ZiX2B!XDp+*$j!+$bJBhDw-53>+Y_XS~hUlc!9O81DrP28?-I&lvje
zX@Drm8Q*EWac5)SFW;t>4U9I1hB&uZYap@2wsv{==-dn2Z8k58dPGO(%m<+JTIO2n
z>T6N&jzcg&j7y}=oS?N+t?9O-&Yu3W5YTE(|5rktO<$P0ex|>xCDGo>GDOFUtThpg
z&^lXxTW{|p{G2@)2p+WyUUb#S<@3?vHrADI?FEc5r1~3=Ps{7rTy9%;oo9r(>D`8^
zwr*Q{MM4%81x`=@Z|$}YqL+V>#(;L5gOCp70rF=}O*x*^zIZo81W>MJVqn?m^@l!A
z@5AjUMy5v>*k1PzMl9M<n2)|P4zrbb3^_w<a}MDh&6gIbCLS`U6kw4NK9q*V`GUaW
zxNaP6E5y6za1R`|tzjI>@TwV(W@S?e@<iU8ZMM9KIJI2^YIznwV>?Va#DO@jKCi(d
zuoah!3Y65VTI&@);gcghdZY-jQGGH8v^ud(WjtrTD^y3+h;w#YYm+i3_1^kg_#SG8
zNC=<7Z<T_Lj+RG}^r**i0xGTq1@4xv>qm9*R5T-o#luBQfJ?t4@7nJJ53eqf$1~ss
zrB@~{4qGH7!EEch@E1Op#bXl*Su~381`4KKo8iUBT^b!JR!v!Ht84@Q-;B8NV^asv
z(HOf*F_W=dvYpD+!$`S3_He5ScTl)e7``C|&}HIl@m}><=^*}owhH@<Fy>5IdR(+E
zWjg^lxhOMcse}e(_a5Lp^ad`jw2n4?Cm<tO7^T0IGnNlccm1+K_5#gh(IZ~Q1SLN#
z!{eet!uWT;wIF;KME@p)7i|1hB%l8-5e}~>97khCngR7ighn>VTcOf8=I*%R^y7#z
zu-h>ovF}J;OEM-ofsRbc)O*3#L<v$LFSSyOvs6KJY=bK;D<W_>VG1r_bpGYH`*Y*r
z{1sEDet)=Md(sh0(DiD4O*ClB(GU_|PyMtY1b$aEbyA9WL<VD>&Y!=y0_Tv))Jy_Y
zaoWmfV20g`2+iyGx}0YAXTg)c_pJr{JFh;KOxA=%4UHhnk-{Vs-#1y>1D7SY5UY&n
zb#5cYTnINjh1aVriy5_4AxhfeWJrXQDp<jE&3JqAW&hd9`B~sna2fgk3tsGu|DlL7
zaWMVQ+SIUycGAMXJO3U1y*6$ZQOQQ~HQyqaw1GZmQ)NA>L8?|j`a9qqAn5NKZx@#n
zf-xatPDnzSL+X3t*UY&J!Rg`luyB5Shfho<sW?H6sFp>v4;;-kzyz!2c}tHLkv|M@
z#Ovwsb^7`IUNe%yM$q3`+gW00i+-uXjK3k73I>zgOXLqlisD3fh~zpvyZC5pv(Mf7
z{@e>b%H0`<-sk0F>ixW0>OP7tg2qLLi%j?8A7Pz^ck6iF8Fi-){ZO31T2h*fFbp9@
zrmGON1jsFBMnT@Bf*H>#jdzr{sdbte`+wxz|Irj=M4@YmfRc!bk7z9AFL$eQsyoJu
zj6fbiJ$5QTr?oN?lnQ`_B#Xe>r-sI&a43UPLx-vrXVj39y%yDo=?ipDgf-CbQm_`4
za-z4(52Du4aCvq=RjF0yy~J%(J`tnvHtlU5pDMgC72qVco^?zLZ`@?J{d+?b(GU@F
z>uHb8d{%pLL@2(Y$QLooDWwR9gJmP@>y)Tzli6q!R|JdLh^J!i>)YA!_F(3P<Pkz4
zq_)-N9UqK0#?XgFg5Lj^QMIqwe{y&4SUro5W$=vDBPJrc;-f&xHWB;J@0fDQVDdvD
zQKL@#e7aPo$a$uz+JUGV#2JTS%X|1gg{z{a?H*Ea6t~XO-Ff}JXd&YKwv%aMwlR`}
za^r470LG2An^-#O3;Xj=S&@t@7*_|7`EhvMU(Y;4o@(VCZHqdl$E@T$Ph$Wpcgdwy
z{EwFKcvhR%)MZze+4uzN(4C>sQkQRio)>lIHsGeisrp%+0L*{IH>^Hbjc_wNa3v|{
zUM+1^)fLYr=R%ssjqRMko+E<a6;f2OPF3eqx;eob5CDO5jgbN!_XLjajm=3lwZ6Fy
zYq!Sjj+&y&E(E0Z<8AEcF}-b$604G)tLXE;KXLXj<qqt&%Y;JtoUc<jR<HNaWs3TO
zuAJ2*><qmiY;O>Y?5xtBK~Mff?(~s8!`mBb$NZFxWP;jUQ!F?7?eBC@MK>%fc;e)3
zu745RJHn2Bi@(F<jcha59hMd1WLG{Z)c8WYZfgn3NnYW(Kh>TRdfMc5*SvcIe!pBL
z^{%~$k=J1;XP!Pk<BR1I!Aa-L1Y45aPoMNKouP`LZeq*SHyRjkZAkYz$jVt`eH=sJ
zc5Fu0kB7@liw2a8tG<jOUu@EMlq$t7!<RB&Y+dJt1wv^_fz_CV;pk`8lxi`jkIu6Y
zyWch3YO%S#y>+oHPz5SZ!g%PjUC6aSZt(VPWLY~DM2i3A1aiC2=a(7SN3lgA2E;6s
zAZqaeh<47*NYP}nl8YrpYpr=5mBp1+=~4F;eAm+&Z=3Xea12~ns!S~_YQwd;2PGUX
zu)QHow*dy{=X)L;AFCt+hRGMlyPV$b^yt|VBiI+4^&0u<BN*LN&h~P7PzR4i=buh4
zY|&+i_E*9e%9U@A-u#uq;|0jCA@7zMdU>h(W7V3N{P3pVj;HSvr#@^$r&6Q?;A92S
zDJMYeM#8~t)CHGi*?s|Roju@)mRoFwm*w1wg(2aH-4KOsH0d}zkXYYH#zLL&BTu+R
zf<q{Lt?AwcnXb6yWAkp%-6f-VpaVJ3zltl;PU47T3a}vfibkOGGN_;D)y688Y?hqx
zu1ErnJ6R+$jMP5K<>9UAh#n|7q_Hg2>D$Khm9?jKP|1Q`<N-cg-a*7z8Yd)#-5Yh7
zC`@uTSOji&I?D*D8~vua^k&{E=-f0Zv<ag*%!vm4nQm(8UkAWcYZ~;^l2(=VjUh1S
z=5N6~wKlC~URi=dT3($p2Z<w8$ozxLVw#T~A5iT01YiMMY$6#OIfXOQ8{DD2+OHP@
z%-Hn-yi(|^3$`6KQCc{pP!Fu{X&y^Pq1UflYtc}8!|Cai90wRbYxP`V?(P_I#2rSF
zXar)seB=~IL)4?@(1<2Ux3H^jGhhJfk<Awy*?RGbUz(V*?pVO2Hirt&9k*xC1HO(S
zHBDr@W`IDgIYiqn&$;wz*@ScuH_IT6Kj!&*l*V<-Y$dMpvA)KgHv2|W>6q|7=aP+J
z%8IdnNFV8wRqZ)s^hW-*8$AF@+(!V6EL?}hM8NXM@sZ$lH7chU^FwX#tWOF7rTpl+
zJC>dOom|JrhxMLDpvt{01V9wF!p6KB-Z;&YAIHEmvEy_^6YmdFMn*3Ac=F(DK6;`A
zzb*Q}qRKzi`xW2E=lM}>JMX~lt2N|OtcT%fwpty&Fv2lIujrcPU9a~Eg&3i~;S~q=
z8U70a|4Xy`S1tL!zRq-OtN}LIF??_8b~oy)w4g9h$R5ln8%8T^OfM|=|J`j6MG{2K
zL7BLl_4v^{0NeeV)U>)6MFI+pcAn7{%)~><?CSAu@A=qQr2s^!047bk<wIodWl%DY
z#=X)$vybn;)1KP&yNqIF%^YW-lJJ&qlX~7iqVtw_y1Mo6vP;3+^^X9h0^iMk4_{gS
zU4IxFUA~S-2{*ajS#7#jQ;?-;vGSoGFzNyI)zZt98lTey%hQ$3JXR}I3-(LPHdaeU
zZR@_5w(j$`;YBUO-yK`*Uvf7D82VYx>Urlsm9^}Kp_6>H%3Duq%lpYq9py>akKm((
zsteW^Q!j32s63W((jHzsokcJdHq+oM&E^Z&^>r-Qn#C~s@@ANq6&sAe@O1M%Tbavl
zn5u2uJ~p#fRlsUOlP<p1SC7S}>)PEnS&%+!S+MMnb>}X2kHKwOfJ(gS85$v{nZK7_
z@eg^dxzAE!WPeTSRljIgD;*|kAs)_UZ-`W6vAFa0rc|x`YUELErv2Qw`<69;TI1A6
zOi;bV4Yn7uwPGRvIJYVTPR9QgJzure-Qbb>I5h&0<$m+7xu0(W*&(p!YK4n$5tz}0
zNC9OTFsove1>*l|6;EkztC-mIu6m;PGrDbu<`LW5%Q_ikQ$Q1|Y{QqfG-{!f(loVj
zRgl|pNu|>%-G??Vo^ZbQ38SE;k3honttSeiLufNf4kDZZM<l4s3~pcCx-<FPW?svg
zA}wb!42V$=`5ky%6G#OI)hsFfIstzfzHo@mdHoq}*tU_yl!@H7`l%g1Qh<0p?c@O7
zU9+P+(V{HC7(Nl07;4O;bTG5+D(ZsIvERa73p*JwcEw_pPNo_oQo00{x5CQnXl`*S
zyO2&vSvwbx7p4eDs6W$g7?f<9QB5@G7_iFnn)~3>{|aJ)#8YK7XUqOEXbx3PHAI^c
zQAx&(sCNfaz3=&cqqpYu6zq4-tED6UNpz0R{1kzX%#==;=V6iFv)~O*GQ54K{Zr|)
zcn}9{U4OA?AJ5`Jz6xE2T0%wwyS_skt+-NDz79SEp=Ic@>aa%^Kot}+b}tZeNf~gY
zIi}<{56fy2c(BNt_L{mJ_*?P&#KYdPZe!?#j21OfwK&1%Q1|f6pkgeKWyZop2W&*l
z1P><`!mGaGMQ(2fzKyLT!Q(h!j)ik~Q1x6)sIgk`fhdc}M*=g-yt!(N^pMO|<4@j_
zF2}3RR%qRnaz1dfOREDX3hE1)ER-5@YIxVz43Aa+U|6W+j;o&6+>_c3oA{L*u9<J$
zx*nT<Q`^G)tlu@s<G}nz5jRgH14v8Ozs7bPn6m5UExmK+&(wo{L!irMFYHRNl;Wv>
zeOiU?oTe&HD2oND2d2=i;<l{x(I<w{yuy3j5CYCOa0<h#rE;8p%iq4Lg9Tx52Tnv?
zd5^Vfx*AXy!A{kP13^7w>DPbQ%9a$YWvztuBnQM4CZ&cuV4OPQRMo9zSyH76HlDMV
z=wdg=^za0ecXOM%tGyNp!ya^D;go-SY9<1HrKOYqQn$f(jWDexfuCJ=QE!H6Pf+1K
z6cMeuWjCbVRTCY%(d5oY$SGRm+@R}W_H`GZrS0M&^(%A&Fv|CQha@Thi{L6#z7GDv
z3OytXr?IhnFIib>z(s(q5Ck!4nw>g&mV=InkklFmo7Zp$k)o#QV+)fx()(1b*Yv0H
zj?w`9v0|RBwYi&x1S_J*96A=hcLMe!h{|9*9eeJMo4|ht=7Fs{28k-5iN}n=0*0JF
z6<oimL~FW8+2^)%Ag$0HhP%fu@2+(CQl}t0gD11G>GpYg*M&OXb}YEtvbw}m_wPfv
zJ(7r_3rF%ScraP-2>!!9<ijH~QVwRCj!<R)joLc+3JKvvlFvL0x1Pr?LW~^E1%;2v
zb{WbX=<$LyjR4DE1(Fx{#GdU`1i3+nWZyk{a~ss8>LBZ#x>j0pH*li~Tr%L$7UKtt
zQnevuD9;8Gvm%X&DRW?Nx*a*0upZRwPp{wKJSK^32|&$+fo$fPFCY}*vw_CUTnAFQ
zYAS{(@_yW@D8&{HC+T>RbFZ{By4J(+Q8%5l+!*d+XL!*2R_n$WdHr;O_dhCkal?HR
zKzn2x`v|t0k#OasSG}7BH3m1Vhd!9!YUNF!l_~2F&CsuYPIJy{dE#pBo2+O1KD;%h
zLGxGGc<kv5{wteaGGJnC*giUkEgewraa*c!Nngc#oUkV}s*ji!)<0@idwETJsV4L@
zngD+CZNN?_;0_FKkCaiB<D3{@KCtywl|N`TSJ3+sj~vX5R@Zql*nPCixAg-H_3tiy
zMwVc(?4xH3LbE@A6c{T2C<;zTr~xhHL7~Rh=&^4!8JGJYZ0671(box-K?E{u5(t>C
zVO*`j3kr-|Ii4vI(ETuwxzK&kTSfx72ezX5C(JH+;6gP8?Js*tJdae?4`ocw(HAlQ
zC}$2=2lAPmyU3~h!*cO?7On(-h&-7Y$ypdO{0`(Nad3is^JF24K>>cQB;YP??CRQv
z9Ot3Y@NeQz(Hu^KqO$p3VR)9PcDjP(bSvV{AV-Lxal2c*yq)TsZ^&lx$+iCgeWw4i
zWN@(l&!FF;u@U#rlHt2kH&=wA!BQ-F9@JB^MW)Rv^%sk3t{KWCx;6cDU+dnELw|?Z
z+{HQj7ZR)du_-6sOa==JGvYHaqMja~zCNEfaR?N1>aZeSwR>O;XpPZG3c|A3Ex$~<
z*$OTGgTvSPvlqH0K~MXaukWw<+bi7c-Y-|*m-p-Vjh)&bt)I`!P=K>*CJR04pK0rb
z$fWjtm!DhQ>f)FT1<m>>lvFb&=^tz{FJi=zdooEOZ@L^eVl2$H{4V<{#1CUV9Y)YY
z5~PDbD6n|VO3{3=b9<c?2N!S#Dc4uyjm@kg?~QF%?cnJAWxNN+b@pbRdEo^DYmTTv
zH}}?~>W0#@{$Luw-i7kY!AKaGlQ}I`Xz6hFFfTmo3LQ9JmuYxwQ+GGVsg-@!IBhmt
znHT6W1x*NS@$R0ySz2Btf&0zFx#*}(crS=9=pfMM<riF!6j|Rd&~YOs&Rl1HFJuxT
zWIbjx#bktkfp>p28>m!)j*BCI-NLGofM$pe5}7Ju__EAMJfJP}V)unPOf<u&z|<*N
zfoPXt_Nie9`Thb*Pl3Y4g2N?n;|yIL04gYV0)bScLdw3qmhpfIVd-LFR3*<8L$NJF
z0<LBc(Tx+>Aw|I06%UAzLnVZOcKor7NSZ{FPyz{sHoTjLBD^=&(`N+ofpGkNJJ6s~
z95~N3c^EUP^^c}voalsAK&ndYwN7PqpE>+xf4zWb<U74pz7fCKu@IHwOr|HUM2jYk
zefE(6nvKYIDh<sgeHB3RjB7keEpC&ZHtJIsV@d*)x6s2|@rb0wS%6<lrAgQre;|6g
zQa#lDyY%YA$k5B3LEck_KNP-pg=}iMy&gL_qtZnQ-P(7|S6|5j#jnqkxL>%rbCi3}
zfHALrDsP7#f|%>I72knlg8;OlOiXtty3^wiiRcFOn?TEmvymnH-=~f?#N%mHteeSu
zw-T^f<Z^(^F{Xd>YhsZR=PoZ;c9fCRx|Nyn=$h-eyj}3jQQLhg6GbWWdBbAT?WrVi
zF9<+04dws-^FS&u%-aiV-K3ZMTy7x-t(b^uxSO|?`)v9Y0lN!gl8l2GJK)hxn9Hra
z3srGld`}?JgoT|DNV3ZWVKgEVkQexCA!a;gj$Zt_(&UhS5olJo&4jTFP+#>>+0nlh
zK5sW!LUa^SYkMy%Chc~G%yH30{0lPP3zmcpe>3~^dHs5l08!~*Fr4xfoIJ`%l8r;N
zdnO#MQA+>iX<f*X<IYB`cPLkSX+6K7AtY)g-X?-M2jR4vq_-1+H{&1>1a)D`my8RD
zdev>aqt*DDbC&TiKx5X{?WW1whlQI8-wp6rOXDNLbHj7wP?iJjEE{9giBAl;dgkpc
zpt1bdoy;)XsDelzr<s=7>U5W^pxuCh;;&6(m!V{g*Ypl@;=F>eTwrr^UVN{Rt$nEW
zobVJ|i$!d(1xU=Pt%b}&?MIA{)?XN!e=QlOdl=Tz5zHul4}&>A#AO38LvwT%+sv=z
zdn|9Jr#HnlK24uKhbP_16t4dHc_(qoOyyST=08w1CGMfDgd6K?nss*w2nAk0{kU#o
z;z8b{jw}xkLRTc_oTyONTV>-l%kLo&-^9-vDiyYK>=O++8~Cn;mCpa|n2>KDyXo*^
zr-a%kd^zOgG(M&sGsk1-XA6KG!UuQ_h1RmlKgl&7BhAG&Ed^@jAiki2T&{w~oahSf
znFzPwanJFKJXaV@FJirm_&T_JE?R<5S%DXSKy<NhD+`xAIpEJeW5^X#8BgUafZ*4{
zxr^+eY+miOp!pj}^OTEtt~RGaf0i50!&yeqe*QY1MZ!3lX@E3fkvB0TvYEiyP!U5q
zgGId$!SRH*@*xbvB(cKvP<$HHbM_(aS2uxsVF;J5z4ryo3r4zEJ1aZ*Li4zw&CUFE
z4}+DD$N}@3=BcB{Um91DA*NB5E4WN$dTD3l!7t>q58SAfm6ZS&b?tE<3b%ve+)k_2
zDpJwsp5?}V<qOAsAEGL%bg`oO7|&nfTzXntQDo^`gpV+cbZLx-?8Mi)bpImKY74F4
zc6;UmS;Bpvl(`tAgnYpU#U{`Ef^K}B@I*`;;kujoqBJ6SPo=j&oGS~Qjg$XD7W#4q
zi#5V4GydU)sEIO1hD!>&2LTI?-v6?!l=dRg4<A^hjDaR~!_ezQ%!epTN{oEe$JYsW
z=eyRN5RJ9RPqAk#sM~CPI)^RS*<V_W5dl!iFab4B%j0{V(5VYbs$PM{QN{0p&8_EF
z?r9<{ANSN;Q%_WQtocnz=g#vuVf!rQ6mWlT*VcW5Lu73IN6$4o(|_^ee^+JyD=*H`
z_<voC7{9cb$es!js1u5%5<!52E(^mchJ{-#wL{4|ye`c?E_eosO~;?@02Gk?yq}!$
z_GaZr{q(#2_Oq|!ukXie(e9yeqBF&kNxNpaLz)TYK_y}Yn>InS1xD)sY>WIkP+$ir
zX1qv3sZ1$k5_b?o0;z-FuqTaMr`I$n*;Ti}1kYy1n4W*w{63cU<}&J3l@W04fx5`8
zQU_&&6!g;&JL)E$X#muSJ8jxqXKuE$-5zN}Lj6%PX64^9-JGVH#6e_o=}3S8yJs9m
z6}%ox_LX4p5HS%PE!O@AL0}yWZO(9Vqu~&x-}RuY9W~V@@-!q-A=G`^cm&>C3s=D_
zmS;Sh)E7Btm4-WW^EQq$0|GCmokhR+JfvE0NpSLl@09Ah6kLulbDhRZLKHt`YP)vS
zFZg7ua!0OK1I6)wv3KaZFG<^H$3z2F%2WsZ(8&hdVahzn_W@?rraZCC@>oMIrOW5S
zrtN&GHX5qa4&U*(X;T%^44~pQCs>#dmsHUvIWMc+)UC5A4c$uy8cKX!P_H#d_ebU%
zUKgessOlGLt^*?Ye*H!{|J&}|he8!C?y<g3$f4`GBX9LEbddw@NSsb(G`|LLI}R)+
zia%K8T;;hiU)AKBlT>R-iDj?OY9Oc-xthtl1y5Wj(CJx5_+x{=?Ll7M_Q~~!UgA7r
zPBiJ*snoMJeFApA%RUBJU5hGoB6CJh*t)TTAxq?nULn<jux>}6^smJzGp6|d`yJN$
zfqY=+M@h4@E!qPJbDL9d?IWYY1pXQVjmVZ{Q?g3fY4yuGatE93VrDIHBeLgj@3)!K
zxFe4Zb&;dtaZAn$ZxI$<b=f5p0S9>41C(N>1QJXs*b+)Pv?Na_IzD-obXHTTRJwQ{
z_dsM649aj_#b^_H{CF=d4HU76SZhaV(uDfQV8^01awy}Pu_YB+y4)kg`O1g-5?2s&
z1=~0wU<ovO%S0q1YT`Lb3|Z%0UOgIT3HY-@xrJXxVj?Kr5Qey7uh5%HTdG-q>k^&y
zi(@LNG#(5xIAMbNA~OO`x}1pJD~<)-n&|G>!X;VQJMeR(ilQ_VE05OPZ0r;ZzMZhj
zyun$aQW$2RA-ii8?DG~6DG#4kt9|yBN6_}ByihW=w|;zrcK3O}?Q-RW>p9)JE6-;|
zQSxaY(?T_zC{mm-idc@7(q~CvckJ`-jZ*7{(I$iCJstfjkf@dKr$FjgQ(Us)!o;mY
zTGdz8Qo>4W2rXutGbvze``<&PsrGJtaR=V`9x+)EP`@pQ0b>1s6rwK!@)b<GR+j8;
ztVnIvUpq#uDU(Gc&POIENYTyPr%vJtHr6lJ?}ji1<g;5W{}ygR+14}HT4#aI__DOq
zR(0qs5lp&Bu=62HWjIrKb}XK1hhAC4O=*Cjn>cz^+I%nVx9~hSwr8;nqwICmFYL5w
z0Aa=Z-r=S~0m~yL(gT~PiejJa>8*s9S6@h6W_FR}D}nZ7YJC0C$|?<nFmVp)>??3>
z+CT6@YP44BJnuTn-}!sbYTV_Zn{DwlTmBKK<%zSM@+7GWuCep?R61y*#JfdXK_CSC
zYM-Z1Bn33Kwzq7q*nR2IK?-l$tXCef{^UzjkBn(Dh1S1sa@%7@h#h}quUK|=r`TM{
zzV;^{8qm2soAWR?eL69U)0V8T9i@2?XLRbIt-Nx)y^bcbBYHmEdbb8l0G{G&ylWJT
zR<ERC;b4ACMH#P%lp!E4)W^ik&%V38<Nth~*e9hgT+hP~_fcV1-GxdqyJ*HYd%cj^
zNFDBUU2~L{N9gNy3)lOW{|JHX$G^V-+l%R6z-z#e&(Q(_BPWJi7Ac&zXQRFJend8X
z3;BGe6dN8*WKG2TT-c`n<i)EtW~e_M#M5vvr>Z<L71_Z2&rWhY1P?nQE2w9)wKMH|
zNQEX;bP%en00>fE6C1B}$0VGv^BDSxclt`V-v&H?o_fB8$xltYkIY#LL>DnMVilyB
z6N_fSrdEI<KvV2Zt`c+*A@e}LoPbFJ_FAhEKb(N48`$U32%UOSZmtfdd#F=a;7WVg
zs(LY@ge8Q(k%Y?_@*Q4ZY?D7D8ZRcMD@b&{>PavS{Py1v`+=G!j(^Jcs{_GsWL_LL
z)PK&wW!|ZVViYFTAoW=^X@a2$n_u0@hvb*`$2PLee}E%q=3@VY?J@tCLx+*^e?G?F
zQk(wA_OQBcl%Y<C#c9L*Nh*N4=`zu%qbMME*kOZPEwlsaut{2EfdqU6q8``py_)Vz
z)1JxZ$-)Qsd_)dRa&`}2vNqPHvtF{vxhBBT$8Ay=$&t*M$2omGQ>UU1?Eps$A8Ujh
zoV8vLxds_jA!77jnaNBsAu>udi4$TP%I+I)GpTNR&F}U+vfv|0`h(#HF(MZ%Gq?6;
zC7lyONw2yF;mJs%DZKrE?N~x~8S2yxyO$8n>?XZ*GhjJsbv0RVss!Y%tGBOdSo$1o
z&BX~VE2>P@EzOUYuF<Iyr_}s&)T5WyXMUTnuTWLT!Ti9cnN*571BaW<78R38OyQm+
z%W1-S#_&rN<XrpDCMx(yI;iIAt`0Q2E~u(TelkA}v%BWIbiDSJ({J6%6mhk5&}Uvb
zS64F)4k~3B(}xo6Bz~g|mO6LN#Pa*3$E7|ScKs>QTZwf&a^eplbty(N0}pNDYFo7k
zUvq-Ttqb2)I-qT{<-K}CrQYvT)p|^9+{9Qu^Xjp&+I5vJFUG7L!%Wy{7RPie{+M;*
z8IIV)6GSiMg3oV)ug}NMZB{O%80na^R@=Kk82roU+sWv*mI`Y;dwl5?V}@T~1O4QL
zEvtS}`fJN9I>?Pm9Sv;=S}I+V>V)bcxlQo)IZJ=QVBy=M_mx0^s9~Hk;$jwSp@Y#D
z%WDV$b-g0fr3D_Jpfsdh-gi+ztie!DGrS<j&Q&$zy#6j(hMnRNNT~iWguN&N1;@gj
zeN2Sn$Ze&0H|R*wf{O`gK>)drOa@jfad6yV74AA{2-ith`$P1$cZlpNjcpJ2CjS&+
z1OLv=O!Hl2;5_#W+IdVPrK%6lrWHzyfT4B=3ZCHgguqK55=9aD2gB!_?fJWl^rhhL
z8PvqIqH3-xQv*vsvoQ`?D}6ondZX{C2aS0K%RFu3YMy=Xq5o~LzF$pYF4}Seo%Q8o
z|H6-u^77?E4bvA>kKJM3uINBvXQbglIe-aKVJP6>;A$(cjZladVS1w{YUm*Ti{GCO
z0d6k?HnMmktRzQlSA}Ri<qezA4J2C1Mrn#jOm!zsBa_u+2#6XcuxV@{3l*}M-(UWn
z&rnzlO^rzUU`*4$cy1KbyP@dXeeJ6PDm>)=luSM53Hyo4xD3BI)Ym&3>N9up4He;W
zl$RX&C3QlNllqH6Q@Y!L+5<5i=(p)#B{->4dM%oX*dtmxF)%`n^EpE5$COkM7n#Nt
zjdN<q=U}FiT_#kxD$MN3Z~5Fps@d)`k4H3#^zxgLWDi*xL*|%@)qkI~Foim%9;thr
zk1xF1xji6y6)Z>o*Xa0<Etr|<e~+!2*8JClFoyoG2Z3k~vy!ddnX+Aw@N#JTj|Pu8
z$s-6QqeV18Q=ynhNwe%W{)PiniA<LIx)dA+^!Q&XQ?_%DGk%l@ZM;Ws7gtvA+nKNf
z12pPzGKapGj;B(20@T?w=5|fp*yZFWgMhArw09Njuw#(s{qxt^^ylXRT|V^PfdfWF
z`YT6M$Xi*H<m<x=X4=Zl`iuI>!wccI=KFqd7H@a^+w0y<n!4Lpdk!y8PGH4P{DXX2
z*^1lw{rmps!Q0b@^RETm{p0`#vitpna<cgWaJ7yAF~e*s+5_2j-Wd?F2!Glhjs25J
zX8Zf*`l3QD&HX%8J@!-`6Vn`yoy*<XXXGJbN@2-bF`$ot`Gn!WhV;*lYbG0-eURDg
z8|T{rVxK1`!wU~h+P9xw>!q+PWLTxc$4stuaI%}bh|lSl)>@04;eq@=PnUyq1|Wx;
zYCLS0#e2@WY?`)Y>WWyy^O1zwcO^WW3Wj9+m-59&Lj&nibVML#5!2}oG1Ef*i&COo
z)D}?amr}{_qv8e=$%Cq)F^V5%tbXKk-l1%(2rt^-C5S<#{HRh6+^Wf9FrXdbj%n<k
zs>=d5n8N%`#SFm<j(Thi*%Dc--Lug%)~TKMLV0fh2|A#YxmeWWKp_(H1v@ymnDA_t
zNqPiJPK}!_2$hfUk$o28Y8?wqW5BC@m6Ck5oA#jfkSWr2aySzj4Arh2%H$b~u@iG2
zN@)FjmVsDi^Pc|ovU?VH-3`XZW81n}G%@08wjtXc)$6dKD2hDeD^h^YL4e?g^uGRK
z$%t|i=lwPnXouE%8Q)gI<W<awUrTGzQ?13lfibUeJcloDg1j)kM5KR9u(t(^5@i&L
zAxI0D)PN*4?S!uxvdopSGV%9AogIoNvG&z$@sNMoFJ*_ex)$4wlonsBkC+$?&RZl&
z!R&9pp@Ji{RRtQmh7tN&lz2*3w8YI|c-C6&kb+PlEN54}+Nh~{UnJ{bq86yke?hXU
z=NLDJ1LeU|?{>zp3sAnpM)4AnaiX*vU_txfK49(R1*3kLUas$KolPACqR6=1@hLQD
zK*%7YKuzF^NJl?cOtiYLc=y(A6d=nAgWb~@+yqBsNHKu1y;jtEv<axX3<<v$Yj&3N
zo6PmN7oh?%Yuk#m#9tC>2_R}nj#MSlfxYJwIT&w+C(DwNs9w^d2uZ(c+q9Zc)1DOo
zjSQ$(Gzq+RZ8D*Xdh(4yotLr}ae16M)GD7Q{_k_HSYjeXr-)p$+`b*O*J)Mn^b(0&
zNCDzt2SCChWP#*d&>1KnP$9YM|Ll(2XLuCUBsirm$dLt+lGSf)oSm;nNrT{@HfN0G
zu6tDDjMj>%as4Wl2S7xQp1VJ+HTV_8x`Ilg=#?brs^y3LB#P{v%@p_%Ui4cwNH|dG
zlm^*#3lvQc0niD@OUnqlzJCuJog_@o28(i>?o0%?q6?F-DzVl939yit^`8#06(FU{
zS~{X5T6hK7>wEm3=ZfZVX6z(6ZX<JbIZRU%UUcy0W3iGPppMNodUR;4qC5f`Cx5zS
ze|vwD=v&R*&W3ZM&J2-6($AS*)4$1t&}XK<E_H(-gr*Th8AM&mJBqw1GE7E&oaaM*
z!9>}KQJ>PdO7KGM)CNea3+~aYtFHY~@XWG{o&eY$fHXQP*CI|`vTtRE(NjMfd%?Jn
zq4nz&elxI9%X}jmtzmg)EuDEO<NY7T-myCrE$GsX?PSNcZD+@}ZS2_Iv2EM7ZQHhO
zr*p>b+udWl=fn9AYt>p+HRm(QU9=MqPVCk97>sPjJglHPZtUF}y7DY@TQRP<wRGF9
z+8`pGGu6D)EyH(o*Lx?6$7Pckx2zHcaBCdeikGLW=&X`i+O1M)E~NNqrHs+n6`!{!
zK;?C4@2X!7OxyHEb)WQ<JmF)+0$`ehduxNkjCHX^Vi1XOths9qE*cr(@;&_bQV>><
z85ubLY}ciI?L<G>QSUzw?n5<$adOtJvyY8{sh$wsVv~|F=TF$+GOZ^*?Jn&zYygd7
zbt2ex#^~4=pq(`YL*^L~_m*qq3-}EYXu-hS5ni7AF6Hu)h6DQ#ozx<GGjpS(&-!N~
zo8}(e!jl99H@L3+8gX+OJmQ}_tVF9H@$AWsP0D_ptNO71tykh>&~ooH1RN!CQq9@#
zMV6G`x}Fd4(iYz1w6J-7U(s}Jb*H*i{Sa`;R&zJXg7s#l3Uag6;dBp{lp_gG)Gyv}
zSldqW=K_@$KvFaU3C_fOe3C`2P6gM57OH*d0R9W-Bjw8|D9EJeMeXb4H|`fF>dQb~
zGfzShYd-uhjkP@6-${dPlCX%O`-sH911o3-1O1y7liqD<ne>SkrxMkh%x)PfH}!(w
zjh5_brJ_H}EXR@H;PFNsK`Ld?WSjsM`Xg!A+w$fynyEPFLuJsW3p)Au3XIiHCjjV}
zw2PiJG-n&W+}B#7NG&l3{2J7hHC-pyrqWebjo%I1hG#NE3BJF%M`g(aqzVEg-I~1U
z8)b#RHd3U7r}9}UHPC>R*>GNcM5revxx~h8oYQY%r5w8e-Bq!sXTVR#JJgq7eFZz0
z5t@F-5g%+AFFd7F(L%x}MqaK(IniTM@VlsT2n?%o6pcriRd4upmQ4(pc?H>bl#M(%
z?n0;xiH@=R!g<!twvok(pEnt%J|Jq1n?3LM#JcwvjZeOCnc+W9Bko7%TD}>YeGhTp
zVmJatRvk%%Nb`hp$*1?qNJCUPpI2B$i2&0oyrpe#rzg8PnDTAn9ZP~f^i@`Nx4JdZ
zj9WjtHf#)nTq0mudQF@ct$DYX-~Tn<-i-ibiOAo5S}?(Z%~?rqIGj2%KV4Rzr}MC4
z`NP1d_>^R*D7EctYlqyz)C5^=Q)V%L@JH-|R;j87@teWwnneaR&%rMrbxC}aMa0zk
z9PtIZ@S%ySI-d^{{?yL>RsSOpuW9=JHT^HJy;9M?kZD+djL7u(HU<_@+}!^slY@bY
z@ju5;R;pV4L#Fx4(k0N-HNrO#4?`IQe6OjRT*{>iT3SzbrbkI{VmYxwS}b^fY$qZS
zw$G>-Jpw=!-yC-s-oHhJg#(23?r?4Ga<v;ohi0CIl=XY$4T<Lu3y>r5f24guFCr!_
z7b9i<IBagFBI$cfKUq#Z+I_aZo2T513%XA%XS>$v^uAvH`*PEcr`759{?>mM{Wwim
zCB*GQMv+?!#5|?hCwJrr_cC-GPBjTU6p|E|ufkj=tc93)hOyA<e#2mp0Cd+?<F;V*
z#GVs4OD5sg$aD!QqeUy_0!BfU;IE8HpG-6s5=W_)yz_F1CMV<C!t@TKrM*WDc{%IK
z6j6o`$|Zhr0V9I`Lw~V1hh(^N2tJnwq0SX@XjGz&<XPf^FTBQwU_<)y6~O}jraVWJ
zi36azq=bDhbGT79tGk@7UiJW~7xN)zJoFV~K?Im}>*hF;5vZ#%SCtBUtlH*8sPS9(
zA;L9*?UL?)W2IyI^Kq`7zRBLr*i{c)Co}f~Sx|pB=eD{~Iazj@bJ>wfOO@VtfvqGa
z<KC?Lo1(AfiQeTAHbtKk)+$n#MX&!?a#Y0)axkPsw*Lrb)7<Ekc76;vdz&+%%yf6=
z&A$8QdQo$KxV#>@>I;zq*d;;A(6!gA-yZl)@Z;%|9;z~-*ccLCBHeoY9{i>`1fhgM
znSV%OKtF@M=pd+L)6M2ViWlzIy1cC}ny^>ADmdR|MR`)?rB}pul-u%Kl?c)#P;ajl
z_rk~lYqfp4VG$a7lau<}TR>=aT)MTV@+x(;48JcB@fKYDR8HB9TFsViYyW{XD*ce)
zwh3!>xyPwdoEa;M_-2lnD6W~v7yrBv6?YPfLOFzNP>^_tEEg?P-~0mI-}*2lGvR%V
z!CRNFTJC$(2er4FgtP*K3&)gYX%*&S%Uz%}5@YX|P>Hv(XA1N>a4k;5CR*0V$82ax
zOc&QLN$G(-(^Tt8uX|V2({4@0=SrM4%TyH;!Zcp26)%T{?=0im(X|RsxVCuyXgAaI
zr}3ikZ&qlByV07mt_^1GEvq1JC#19c3xP;l=}@xZ6Z%F$N$oy&#v&^f%Xc8_RO#6d
zQAHd_Au$gp@`+UXY@Iq|=XQ#!=V0r2&G|3c*S}XPW$#5&cG}ss!`)poLA)>I;Rdcu
z#eHm>E=Q<uE*0_g*B|qbZrDF|6FMqB-n0n@RBSXiV;|NKHJE2)ojz!37qT6IYY@+^
zeY_U-tw;XC4IC#47XXT_rZBex$ukrt9FxIZYX=YNPZZsz%tz$Ew=U840yy!=c*6Ap
zC6AXT_pX?}Scrvlg{;IRQHwEPJ#q`+d;nj|hfm?zJ6@O$X7W_1oZ{&I^NakY_`<U(
zP^{CSt_jGe0qYAtgMRaXr-h?s9-XX_qX4LSi)XW)--v+ACw;^LU;GmnQQsCRnf$E}
z<lI~U+q7oEe_$}lqJ$;}D~ip*r+=($u^YdZQ~MASZeoFT4Gs1|&~u#@7hg-uJI%TD
zaw}YlZUkLe&@TX6OcjEgCd;gKn+wYi9gtEKO|SJRfh|o2oJ2cNf>CyhXGh$>&+5m3
z4jFaF%^_6XNB@|Bx`lE%Le8#SY>Y;;UzvNJY6=H5DAtXKh89E=#%2HE?@>}D3BYFg
zwG+#os<c}ktmw2vV#YW%-&$s%B~?}{mHPF@+PE`u5x^vy4awe%%51x^dOQZ1oCztN
z&Bn}qfIL1m9;$x)%erg7<oNmQ@of+9wepGiUm8p%_J66qtgQdJ`X-{STCcO;bn5CG
z{Zo9!VhIaD+xX)+t-@F#F%97xgmcs!E2^k5lCYcG`ML$+oH%68ypT+pGu{btcp^?t
zF3amr%j2PYMHxbk@kQrfL&oLFiE|0DrRQf&Wx@8V=by>YB5H9HIqY`y#HY41x1`9j
zhj_zF(PC@q2+uYn_ItQ+sCK2dwY79Q&A<L;NNsC(_udaS3TAC>c(#4r-x1BpBDgto
zg5_@M#;=dhQweG7XiM=*S51+L6ss;^)V_lnmtqPBjHzK5do3)%ql6lSa(**DkrZwb
z%c>;GAf}i|ep2C5<id2J<VsAB5mq9+{1$YNUkG<oDR9>Y9Eo7XUp0*YiM0S7K;e-8
z0xDh~Unir^BcoRP;0`xeCv)pJBuj&-0_#bRN(rrygIc+tb%eZ#MARjRjleMELm1Pn
z9F}rBG)y{!VmK_D8tXImak8N9O$@5vAL~o0!N;zviWBuxH69eKx>mt2dJDA>7=cvg
zdb2`Xwe+>HtmkGc9<g^WvpQQW8oI~+rAw$p*Wg5UjzJ3w5tZqz!mv#ahW5Md9I~fk
zIJTLBl5wOyKAat0z-NyhCIS?p$_IIhBZ!L7K#cG0f(f~MJ$t;pX^4Bjh}DN6mH{gi
zA0!uXjZqujh2MN5*|;jc58xuE1~ks6Id}pNMiZ;iCypFR5LZ3$&C*gG!}tt|f!0w)
z^&IlPTAojvOw)ylV(zLqGM{2dhZ=)e5?^lPpntV7)cwS`QHRXwhrSvPB~MC2n2eem
zPZ9tgm@W`y6Nmlx0@cswh6<*8y7cBt?F7vWl9?FApk(Fyp|CbwAy4gm5Dg?OW`G-A
zt7IALvks&*1Y&h(jzNto(IEn64PR`_1?OWXv{I;t4E6dXGLmA?DhAL=lWg$y@;ZcB
zX3~nvmv4On3`a%=H~3w0*j%S7S^QE+NqWI!YFA#U#e8c?ZJeKC+EJDpuw417CWT+Z
z$&prarvNYBGd|_41lvvr6`tz=@1s7vIm)h6rt4H)yXr4KOd6-m5Sg_5lLo9RJx@LW
z%jE`%(m5?VX;pVucV`!&t7kgTS>Gp3&FU&DIHWdF$dynWkd{lt@V&7`$5mBPsiC&j
zkw{bPL#O#Mt3+Nc3f!~!;-cF2W*qY3gJgoHUQK|bRzb3fGAv}*Xc;Ksl~d=U8HgMW
zzk^Hu`C^?ol&0ofd(BnDK@{uay;1z~wDv%~@=;+N2=Z?*NW47Ak>H84VIE;533N`o
z`boVfe%UJ&z?>E8>vR+CDR4m`K&H^Z)V^Vee4eB+BYw~f#&k2EKA_s<G$6p7(L}HE
z$@s*f(+ei_l(|c0K_VA4q>n!yvp62VQ7(zzRCW>ELMhpWbZil*ydkK(5vc4!eDjPO
zO<7|5bFq3Npd+u_bN`r3O%a>7BV;g)*q->o^z>cQU9v-^)0)XNN|&m4p00_UxC>wV
zox6J+Tw3w~!W1}X{$`O{WjLeuDPV#%lcntoQzujR<yX=3V5JHaeh32Qw`T7dZB%yv
z4YpH{DQds4W}ps=q;|Hlx4?xa(Wz`uw34?@R~XCCShxRlF@}zC)7-4?J4P)RIc&l<
zZ3Eb>@K976UE1;VssHPK#PWFg6piK%6Ana}xbErrR8#V4(c~j-x!s@&h$yhokZPQI
zw(%Ht`MVkL`n-*jDqr)|lHt=ibJK=X8clqjOG}ScIM?BLSh&vbctT)vpfqT6fak-z
z9}SG7)_Ie~S^c2Q>$`<*^s4TD?TYA0H*Enj>qxyy7rq7Y{=Sv6fH=zi4<0fUE~d3Z
zYD>nx70|skZx6;M2i4`Wj1WI(cVAE~BVFbSRk!K5GaAGu(mYv3Eg9bl=B^T&vdXuu
z7PdljTg6y~t(X>-^I5E&0AbmiWCWzk-~JVGJ`b~H#Cv4c_d|Tm_>Wwze(S~!w@Pj_
zOHP>kZk5p#F|`Bm%)1cW)?r)GzF@D{kpwqE;ppeb^?Yf-PLMm)wosYtHNWz-KiSeY
z5c@ai0@-mf_81i4OnC;hSP%+?MTj;8i2hvZ*<Fd(^(j4pxu^*v^agc@6b40@Bj9cP
zVKO-Z&ES88$x8zuy7TOSw<3{om&Dgnfx(0r!_+q@FzVRPCj8CB7MakVXyNEWk}|}n
zlxF<6QMAB<Q%1}R0xanyKrM}RX;W6u0O?RoF4oc+<4?=K0e*R93H-~8_HS|a9RGis
z{eOULEvicXgFgNt{Zo6!A<!`~5u1)|tqsX2fS9k5lL;ayc^E^>lL*H776^ZPh(>EG
z+_=v(OPBu+25r61dhJpduv6r}{Wxgv9i9KlhsX<+(t;rkJnZ9Q-SK2-p{4ccWuK?x
z5xl+}U!K;@OtyK8b>Ds(d<3I=BXK)9WJ9CV<>UHxbNSwH?@f2{Z~olf2`)yPI+Ec=
zugPPu^s3n$$>`ywgoP}C0An8$X={Lscx)mUmkQ}neA~CoFMuM+%f)YBSLFUI(q-ho
ze!e_6V_)eMTM3t1UaU*%6QpV!`TVe8(5+0~KkaD4-Hc4$*QHw2ShPqs1Mqw>Z8wA*
z@;}6XXTl6kI!8`HCs;W;FaRjkP15`(5y^)NZuQZ}i@{2#&08o|I1JIYA3;IHLhIYh
zz`Kg$s?j!et;g0ac4YSLAux~<aE~A;EXL6vodYHQ)623qf*P2X|2W0yRDgblg6PL?
zjiSCleeBILNeXAm%^Rn>`MrK>(W&N*U0KwvoNab@kvuwS(a_f2R;D6Nu{8c(ru|{v
z(UmE6yu5ff&*{XBC!)H@jnynU8G@eHWkorPt7Z0f(#bF8zwaNsfI^yy{jEACLB3N{
zYXM1Z+&^Y!b$3c0<VYxK)-Gb&)Z(U6+1;%@q0>=~Ol1z-u%Z*Fnt!{wX@8qr+IPrm
zlZiL|Dh3+L3pahHz=BBON)8izY4v*g>8n1@SqAwhYm_7$j74M&N&qR7*jI+$j}<Ka
ztel4KB)$yF2Z&kD9xi|v!1&AeaXF{LQc^4si1N)6mz4`)q7P1o9rxvlFx6DZUZc*)
zH^7Q8+POO_!)dpN64`Z3t_mMMLlR_|)`)wT=C@-%Z9tT%OY^=^fhnv=_r>dotJIZg
z{o1M#x2CeT7QJo-)*L&}CahqWK!MZag%!HtnpM6dHHo>V#NG4qT#%N*8r1h$sefz*
z8x-tCT6lkCfoETAfxF-8k^oA0f<tj+7N%aqiK89y2_+-}paz8kL*@!<&mcPazQxAd
z_y<*W5GJlAN<rqpYE!x=m$CQYCJOGl^p+V%K>Z!gx->Xne4SN7N#9*-d$@s;@w!f7
zaQirgJCCw#+H`_p>q$wJcMQ;-2s`&mwGq9$P-O;BSo+c>F)0*BvB8x9(9Jg-*4Mq>
z;k|$0e)aJ()FU1|OjLfNF4B+p61>Yh3tr|6#LpnO<=^i3zu?U|SlRwRZP0(X<DDw&
z*6SkwFb@#i3FRH?pCx^HSE7{K&8Qbq!%#!@>E|R2C}V+=BW6)wpReqbk<7_~6mq=g
zY+X&JQlVWy^PNEQ@#w0K(QHT)grLZy+?aKR14JY20Vu+c#!tK^%<q`LPbWbL76duM
zS72xkj-D)j)X&5e3iL_%;-vw%)bi|I5+ISs&GGH;&71Eo`>&S|mCa=HI~{40xJBKY
zo%b)bE;_m-Nz7kfAFeJ=*V-=6@wfdiUnUpK+C6dszrclw0S_eD?MhXX4FC)Fbn*ct
z^duf#LU^@FDT3*};r(xI{xa}~5g0kd?TBa6mQx5r=uv!eo#lg+qW811{jD3t;ZpFi
zMI;uZ(|;Jp%fAM8^66C#v25Qq0!ILX5TJGpyYYoooLcMw#t1GrCg8^>ev^lR7)^Qd
z5(l6A?E)tULM#Lf1Vlan1uINmrK1lq=!MA27C&PD+pUVI+@$?-tg>7r`)cfI-@+;@
zylx4{nfb1fyoUrM>F&@Gj@TCM!v*>z*g2<TNjbO<Fu54jd;ulAj@~KpFi?m0x}I{~
z9Lv<*a6bmo@nqop-O*w!m$z@jaeC5F+j!iNz`S{fq9L6;3T4LoP)NS24IYr6IN=jw
z$_(PNUQQyl;00ZRelGz#o4_A|X=~rvC{I*|J|P7xm|b;-l+8)~^VRaM8*aQ`W_r;k
z)f^St3XVt3So1!ya>Mk|BoR$g+e>h!x$^E77U$@7uiC#i(-&YM30`DG8&8dqA0_mN
z3p)Giyo0*$*Id?I*n0gcXdr}svRJOZ((FZ+cI*+jV-_4D6i9CW31zIEVhS<I6D}q6
zB@9Y>hnkJk+1wi&(;u&zduOkQtO_-E%g$MXw$f+m%ggj(-OgRVrZrh_wN`ups6*uy
z*~TWkvP9q1JB-ZQin`<-O=ZY9yVi3A^~*|1M3wGl-8*+h*4@F2+pAjt6|v8}`^Tat
zr&X?GIoxTkIN{d!6g2hqw<Y;1&Q`If2YUz*QyrV7BP}*A?S-R_9%68%9DUordY9Iz
z7xmsxH1KpA1Rm}w;NmM+ehJkIbe@&!mD=l;iP@O@Zvw(v-3eQZk)x~LoJVz3=y^K5
zN<d>Vve+U>VCOLLmrv_EmM*;*;3>10=rMmx&e6X;c^;FvI@hhS*dUF*OHgn%))jjM
zQB&XPgE5lb6Is^R_^>VX`eDyY$*}PWe-B@-;*9|JgGYqQ0-+w&d*$ht8}+nL)H}(i
zaq7afXueaAk}0A_fx>)6ICby84)Q(5xe)(LU10gQSt1+j|0Znylm2VzXl##9Ct|7Z
zXe?xGXk%my#mfuj;An5GZw2MLmZ`d0^^dN8TJ?m65rHYMq>&%Wh-MBHG%mcOK$t+X
zp;81@JJvqV+sl-_fclhET=QK(E4V(B$p{78kNjm<py7?T-Q(s(83UZ#ltDIr#~Lr>
zAI@Rn59g4j_r1m_ef!!w!x`kdnOkD-W;4hES4B`WRpn8+d3DGfrR(b1KE{itl<TaC
zPwxIS+@)H6W?wZFLPEvFE~wZJbr7Eos)dJrQcb0&myG?TvebPg9Yx=R)8&rXBfnNR
z$h!kGlC$@f#fc8uN>4j{&C(T-GtySlMNaTfRIzA7K~<X!>w-6DC@{}*8VMwFEB>TT
zgGT}<B%32OW}2_znAu@+#7Et3>oQ1?claDcCXlWS5ntrvi-2>xb50lYLnUX^Dpc$`
zjopC5#$Hf`roJ*<SJE0KZf#u(1+4*UAqmA-_!%tj^!|+-iz8Gz(1y{0{tanQh8^<a
zD~v>P?%RT&E-`FceteB1+eFD`b{*^G;N<mi*O2*x<HRQ~V%Nv(YV|XBMjiFBK%Zn1
z83sl8(wO8rd-5Uo#e!&P*XX`&@le;TH@U$zxLUzf5l`b^NonX)HT34;HI?=1BJv33
z=E}r6d5g8Ii$HO&s+|uk48%j249PZyJY<=j={aU}xZVYCLpx+syOPMV=)|IoK;D;5
za{~VwaN9YSres&>*#TzQ>x#CTA14@(b;VkYTpHM7R#)R0T|zSWF*|z2$nP^VTUk}u
zh1%Sy5wxYDa^E~DjAKv?F`%}wKr{d=C9+@(zqYDq0C@{d?VWH4?~?9Q)C~cFbUhSp
z-?HoWeq*;$(^X)cSksCS9FLghxdTdwc0z7V$pv4SxiuW}8N0T{Bm+IPRQwPNISIwg
z$pYLEU;H-Ur?UP5%aUgQ9#3bFlYFWix@26HOZB;9#5PWS3YP_gLxY}Hcy`KKcJ3}*
zS^#4C4Ysw){Aq#GAXJ?)Ye(p3i=3cag~nJ@jUqr=YnYcAdDay**hO*75z_H>?lFI$
z5H94T7SqtUp@`%tgMw1)8SXUgHk=lXgmLYqy%BX7CGLKn2kxs}b30k-18|T=BDn((
zRlEo^2lZt#QF#QS?4hhePh)e^A1*3rQdrzC;WKLCYanUngG!u%fkPizBsBQxHu`;a
zG!c>*fVwgkp7x`22~)f3u-qLt88R`mte4}lAr47jrQ<;qvjI-Xva!Ta+}WOCLrpyM
z$<jAHH_+sELXbbX3Qyq4I1~)vDrB&M!!$av(JYlWMaDW-H@s$DCRkaF<c3*RVYU~I
z?Lz!sPP-1RSOlP)Spc1H{#aI<Df#YZECEmT$c^za5>)bU)aR)BLbK_J>q5-Jb;M~c
z6A0kYd4K_t+Q$g8iJI)F;xlxvefroYD1I5m?!#=0g+K~hPLd6zS;^f2UF_rN(Q}0&
z_b_dcgXKIigl;O-&A<oIV<b*osx}_`?uclR;|i*vnO_j|)8o)Nz=x{TzU##NRO(P#
zdIaHlH{|X2s7SZa&TiEM8Ht`GlF{uNnKC}M=}DpRT1MrD7`P&T`vY>L0jk6=oo^1a
zi<!pnQy^=}qyI(%ShPtO4Rofhv8LTv9du!G?Tb_^V{m1vEac7BW0J}z=2q`FqTHd|
zu5c~48L(t$zx*6LY*RbbUX~$w^d;`a*D{{8d$SD5HxtNGNZNbl>Fut3Fj5yXJkI0X
z^x4FVB^vnMc-tgs2LI?9ifH47HDawJ$UQkgNLLFFz454_n>iU8VuFvh^dKv`I1w$#
z1>LT}IR08L?QPVoFi9{p1qBCG4r)x0X$T7+Y~J+!hp|<+WvQ<-C7$-hm-JcYO{Jo=
zciY#MJ6-rR@_to}1NWLiG|i+r<Vz1AaF337!D^o%4a^pBJWg26_ojH+qC*(6m2}M#
zzj!EA!G#NPI=)eN=J-~<SL!3hxr|`v-*YnR#FdP}VZ#&f;ZhkOuf|itja?h}i;*9D
zZtgUm3A*a$&oFCx#XaSbx%v?h-msNaRcR~g#Y{4CPz1CruzGT)n&ZDF1ydcMcg&YO
zn^%h{x~r@XoDC8W(j2*w<-l;tR}ut;j;qu>Bl#uqdkVCmQszD&Q762p+K-;FJQaNV
z3-Lhd24^bCpF!e1*k{t+DO1u&A#rBdOS@Qd0B)7M&trQPmAKSPJ&fZPz4re4^d~b-
zypcWI`MxDk`;XNJLZP66UbZy~Q72EZ8<D=BnXL~>o<Z9ubrx4ek9Wq^-+-)RK{dL<
zXX6bJy?DLGn6=_;r~X|@h7d8QJz6Ksh^kvJT_^nV$DKo->E_#N$HEf`2T?1wOIx#m
zzgACrIWavo7$0v3{|311fra|Fm-AnK)C|o3U1(IPy8X`<&U;!nmj@=sH;v>`6ZC1P
z$@(}~I(&f&tRYZ9vSC6A6;@s%>GP56j(qqBH*ucC9g;V)Kiq!Z#idw%SMvO}_j<Q?
zT2B68A4evzN5iI*;%b+GfH}BT_rUuq&iGpQae4Yf5u2qH5&S)mZn+{%=bhQ9J5^0O
zk6+|1Lpsl%`!x)Cacim?zDgYVak#g6cmZ3brqz|EJ4vTIar^iTA3na!p3(WyLVv6+
z2aW(&RGN&Kc2LUUnkEte7Rv^!5O)ZPOP3a=nwKGkEb*y(GuznNGi!=DpKQ?o-sVrh
zbkzHhz<^{#ff<u&CfaWJ_OfLz6Gv?i#>hNV>b@A}UbwVO#FZfTG)<m?XxD$9Sh8$k
z<R|a`#G|x*`EFhKYOM?XIWzHmHDHfKdWs&6?eXvtp3t%H{I#M8w<oe>VB}1*Ci3V+
z6S^u=)o6k1RBvg-wuYS%S$Zg4>KvUrw+HCu$6Rce6>Rr#9pZ!da+GorJHBtPaFSkc
zs3oDpBH>b}7CBM7BPALlP4(~;%u!?v3a)DCY-~ELzfB43@fb<95q$5t8)5EEq%v5w
zfMcT)XaTT1aul?zk%7VQ(pj`!*=9xMS~_j?4GdUF4DO2Zm5Me*47dq%2LgAO#8tz$
zu2u<YaO9$?59SL$!l#I-<TcI(`OvJ}thKL4BjkTBQToi%jeLmwL6{zC;o=NmZ3kJi
zfsxF$s>HqpSt)?us5vL2b_9J6$O<6cfgjEfgd`N&;mI1eH%z=bKujXA2Q~D-_mLTP
z`w{(oUp~10dN`K=kL<sizI<JUI<C9T`{N@_)+~jupcGndjsjgOHJK@~j=|g4)?{cZ
zQ}$}CyxwS^+021jsx+zDcj@Q#e1eU!(GFLfNu;7Y?<%{3h)ULMmS^P;fTask1($k0
z|0B<Cn@I&an8cQdA07AfedC4@!|nw7=rk@#f%)TvDcoW}mM^WA`S~&_Vsae-xx%PB
z{ARiemj&mnD6BoaTK^-ds6^no<af<hX<v^J@QE;yE3vKF`>aWhPs@+<O&PR?RfEWA
zPbpG=*S36>=ThbZ9>VvG0OB1Gj}~Yl*`u-j=}35v;YcDXp3M|Jh!I*k{zg*dek6ER
z*hLSX#@5I%?J4JHNsi~N1*P3!A8GE<U}AHxsyvn89LAtF9+cQReJ;^hw|?d0p^x&)
z)EtxS8J#SMh^{Y7p%)AnbI4$@EjZ`X8&lLJPl0UU3P{DSN};?rGBZH4fqs@Qj2fQG
zpzlcZn%ooJw$o@yI#GS&HLcl&EP_{i*uWT1CT_;V&2I!4BF#BOUbFimBj*<(#Efwb
z)GWktpX{S|jvtDp9#&*~3ZsC#MMVrP7-tOgymW@@#Qn5fCb}qX>6X&%_BFKx_AYPn
zvFC^6pRD0Vyc$|<**A{GPhe{@&o(Fru_rKxY)XlHC*N|)5>PxqM+O8{11w@-n-Vx#
zJMCT0?`8Yq*UT~^g>V>@j5HQX?+6&iF*g?+6X|RPqJp?j$>p@ORs)ocm&2HECuDP{
z?wgVNbvO$7G^h^1M{AEXKmgRWIc5+(uqt4odF4yRnE@o}1bf-Nh`^+eGSA(*6oc7<
z9``M>Vp$5KgZwzL5yeHMAjkvj-;<%t)rRZp@s|!f36}xr%pLb=s{Xxmr*dfAc_|-b
zAsx$?2{>e;xm&416!U=E74%A~4@M)9D6bv_gjhdckas>Ni}Vc;RkGJ()>Tu_dwPY4
zpjB=$3!lHhI3aouy!w)usSsAzk~Mu1d+cawZfhcJ-Cs894Z@6wn$m$bVL=B^BksNb
zw$q056rvG%qmf7LEa(d?n>626DevhLz7ny3K*RPGRI-N4K|7{wcY+{0TCNBXkic$o
z3QRf$@6Um#M%Hm&Ton=qL7%(b%u2kmDZ|SihU3+$*WcHLUrW2-CuGYj^V!|ljW6~7
zx4Dp?V0WHww`>$&7*GH5O#Ew%Dib}!e|2S5s;*k^v!eVoSS9eFMmYv8bi#@~;@sw8
z)WSoPl1MP<n$7`hak<8E6U)gc6zcW|Ms3`o1=p?PyN3j@{>A|Q_5S;(y}M)qaydSG
zDd6+M{H{IZ;4yW2obXq_K@re}Bl>Tgji~!?CtqEO&DeRmyD)Vl2q*N|yV%>iII_Q-
zU)>()KR<7fNwkAufW!hqRp0mf*0A}reIen6@qb1tYUd>k(KP@HcHo?J`&!B3NnhMy
z2to8lwkUS@JdSu96W1TuUoSf}*989ity>8!TN5#nHU^>)fYgKQ084;=*WfwZIQ&<T
zHv*s^-C#X<O8DOt!rA4k)U;uE*dK~2g$`_<5#e)@U1!R_-pTDsH%~^;K6bilr&Pr=
z+QnZB6uUp_LQRdi!KKY20Fk}lqxX&XF2fkpc~0}BOsu;m9=SsmvwdhyOfEcsrYV=n
zF$t_SF+=wDcAi#a&HZj2l#gibH^w;Yo?BH@p9X1(cJdy@{0a-(fXhs*ks-iBG0_14
z4SWzAQQ*$-F0~2b(SbhtO=#!kgxZ!ow`zR}!)Zh;8!Qdk^=7qdH=SyD$xgK?aWts&
zG^pXZEgjiRe`>|XqHFTdRLf|R_TiFX52-!41v>^6+R)DT%Wzd~HR%WsU|BJ(W2_fL
zta<e6nk}+9&Kl8j*HL|6`WFzdu76$n?4m{-;I}O6saZh>m(1&#v0v&0FA^^J4eHn;
zyA==0flqMiokSlW9z1w@9v1ZA*m7~UcC}@0WJt?w4!LTVJPvN#6}NiJ%oQ#p&@sDr
zs=G0aqk7Z4qBcMKC(jFKRbnyR^PNqSPV2-q(th@lQbpVmtHS}Hm)#${-9?9wI@&`y
zUx@?)8y2JYKUp`Fi2)F_I54?f`nQi99nx83k~L<XC$<+nTBII|zc$c<(V7#{uDtjF
z*1-c5NCoTH`~N9_FI*J*EtaAtvLvDh%|@5a#P+flDGYHrP__-nn87<(J>{H5Eg}{K
zT|UsLc->(#4M*)kMErX=XA^aNMaC~K&5De+*G(b8?M`hKBm!EXudF6a8pL@^cz^RN
zVXAIn<($~c^6Uanbt`A(MBTFPNFiXjdddKo&asj5FQza`ovo+4$JP)<+C|P;0U5%M
zMC(JHbf(&5tH(L{l(=-8s4XD+FE6F>o!HKJRPl~9cpFVDG&|4$yA20JC;~aPw{6KE
zk^Bp7!{m;COA2F8$_C)EbA0k*TUQ#xVbHQ`(uQCNtg(qe3r%geV|y`ay&$ook62Uu
zjLb3-PyvUy2v7t&feY(ZEpi{*^+c1S(PFt<f;Zc`Sd%4F&*W)=iM)KW-oe>x{Y{{`
z>PenPr3ks_BNGpmbcIdRd&O*ORTyEd=)@GN`V|-rSLZ&$Bg7s8xgts*62z%O4f7h0
z=*)$iLQDL%XQnJzy7$&>AdUu+?Cf`cISIk?sjf#_XhTEVF)+hq0DjIKTRc(1;+!R4
z6r-T|WSY7wD}g&ItA^4~gm{^4(jrgNR0m!@mD}B(H0(Zb0VQP2P2}?3(bI3%Iipf<
z*=p&FwPZ!Bxf0f&h8p?(A;?v|)$}9knAcExc>P@Mvplm_b~j%)uHnN;PZYiSn$*U<
zRJwI>O+H5DT$F0l*|K`Dtvjt7H_)bYyzFN3@x%!{eSN@^BKD}eXj0hV{>>sDW2{$=
zDhd<&geTjXtNHYgB;G~*nkU)3e>WhPP(sxItg<XDjQ@`)L&?q77@tnoz(Uc{3W`n!
zpMjqKpFx(rgCjmWJLCVA%(1imXKLc2x>Yn*GvZg3t^mGn9Q>Yr&_N6}9AE~>GMJff
zkFzgnkiKO>XT4me<6`*k!}4=#XS>?mT2e!t*D^d%Yx(ZQ1QT5bfuZA2D*7l03_)6j
zY_HrM`e2Pw#HcG8y~)0R)YeOcJpVbN2<zSkei+*x=r<Y>WPXUa0zFu~=9^^*MEFe{
zxep<SRGrKSLVZ<eN=VmjkSDc&$_*)h0YnlCR1z60N~cVZD0-X;wJeA{GH+}GyP?|?
z0<EDNAOnN9Fo!AYMFxz)REkldZ=S0$VM{&yPCPOxr30WH0Cz1&zq~PR0Ho`H0sy>Y
z6ba^(jKEEhBp{M-L@>51a>R33IMKeL5`&{3JQS*{@7N3$@~;Yj?Slt8DMqpmz&p?y
zXgxR~Z~PZ7S9;_fbV4diT;!Axc0IpedvTC^3<MvEdJF{A)3|1}GD`FY$WW&sYn(^E
z2*mi>pgDQqObQ;Z=;{!8AUi||EMPgb71l957)U~i!swpFr*|bmfS4A!edhd?qyWBZ
zEP!Htl*V$HDij1@U4e-&x_EwmJH6UC@#x!%{vp9@mkNoDwO9~JaWFIx+i@@kk9?!_
zvE<J{&`=MuGRAO5NTp*n27Xj}angH9>}dyy4o2~Q<G_%E1toX^lL~!MzN&>lPXW!P
zJ>a0*<57-;Ddc$MF^ky*Eu%0EAUDP>eo)r-__$hHFlE^jR8AzOS?OoG-Ctn${e>OB
zT(2#8GP`g+zdwsl$iql-t=h(4EsbeurJOK)>e@AJL)H5I|GFwvLp;ON!{5wuuq$;_
zALXM>j&(3D{@6j%&W!K9jlYb(?xNSh{w^#}=DZ=|S(ucQMBq8S=D2TBAD<7Fs5hDC
zdOb#)aF_l~>1L}|iUU9<a@5^tdIb8jtI%4wKEn_Mi({D1&&h5%+rz-23H0=SB9;-D
z{}-y-DuQ+P9YEe`6qEIK7jJ07P#b)?OAX4YmfEWpr5SIW22)QdZqfAzc^J3QE3hZe
zt~6~@&!N_7!G@D)d>nuI?SB^hYOq1Ap>yT**nD<;Cd$p<xcv5rI1pvjc_fB9KH@;4
zDVpY9e~8WhXz=+Cg<ak)Yw}%{rLEQQQEr;QXvABtqx9|hRsYrH)vmEg`>mkNmcMJS
zJbmG5ryY5mZ(Jzb|5&F^9lU%#+ESWUUOs9zHzD#DM)iEv$k}SU%DQNKV`6yHkv(mC
z2-!IDx!;`?H)<)|<<X+ktjpg$Jt+~mqB~Q*BYNq$BRPsg9X#v(B&OPG=*!k|AagZc
z;lx#OufKz$aI-|ho3Bcfxu=nxs=bBxx03PlAi?uct|x9V(dC9_ag~*W&H53rW%}pW
z=CS`(y2o%U3!V1lynLF?!3Z_kp%`H;AZ2b>lV4xFLQO=+IIYG)ZLq1hbdZh6xQor%
z`mM)%cKWT1%J{2XH5b#`pvw&H!tylK^W2_dYi_X&c~3j%VHd8_+Gp#fr{luV@zra?
zCR(NsY{h0~Gsz!h;wB$9cgl!UnD^_vR*Rg45{#{ddMBPrm2mf_jx;*!q|2f9U0>0$
zsLKTbuiu~q8XPCqE0lz*d{(`ImCtL>u}&w8-yoYF=L%ol0Gr0?c~AZkWx5^)rttyQ
zRPI>k#@|_W!uP3BCc$txJlKEu_?$<EJoS$EXYD+4dAwqrryW(}vMJD<oSAq!*on~~
z<;l<r-#6}K;$_x1ue2!(0;J&(BPM00jegUjdVTEDvQ*j>KbE-FSJ+q=Phut9sm7mY
zN?)x+vUy<X(a4f-KHh}VAOoTPYD=cuy|`arJTST2*1*U>FTKckbWcbgsSAd~aAS<g
zT2hC5P%ijBdM-aVUKf=PzI>P>KNps)!x?i&Z>CFVHb=-ql1G@;;S@1hWOj1IsQHzU
ze%W`lDzu$bAY)fq95@Ve7&O$--Lh2#NJ+4MaxwsB|DJGkz$#IaURGjLHdntCKSo5D
z{+q$KX0ul-4g4l{2S)+*Zzw`rrK(LeDbvVf?550g%rh1TUnHEwHMBGA=)!VmDqb-Y
z;#C{bv$JSaw%xFYE@LH=*9KRq*Na?7KEH1(Fj7ZSY`{R3DggQ#<()e82(;p-)0e$9
zz<%?1(W6v&<`Y-_rG}@^pIe!y$P!KMHOb&A79s?^S5FboWjvn8rUS3<!~M`kc*@Yc
zX3?!P<rzUL1p5WlUG7Y;X?H0uicTlT6$3-6T?Pu9X?cGgvCGG1nzYW+iyEpRrh<TA
zrW<L|xPEMkmd$~qc|i7p%oZwC<;{F>r4ulvoGx|{Zz%9ABn8#t6-_cXByjoKKgF&Y
znXE{<Eq}ZZzmlqK_K=&}iot`4^Ih&sR|>?zL6baFm(H|1QyQ#9c{<OflgeLDtU9%%
zMf%T$d9wI-^*gck*$ee2ScyC1TiTcQ{63;?lzDjWm#r~=P~(81tar$z7=~;g2G=*?
z$FSY^%wC&ZW*j$gZ$y~hS8#8CdFv@p|K|(N`m>AvpQ3hlCi?%{Ck?7?JM1$fcz1XA
z*|adBWqNC%I)E!mz>o3kU-jME4m&1@h{HzV7!>oooU6MXE;wg2%ro}K)2Po?J*|H_
zFEeY1Fuz?5vdyDaZ%H6x7$9W{%?lI9iH!!xmFAstzoC<zj$Y5>>2w-LIH@pDLM%c}
znW1KS{e2sM_}Cl`pu)tmXOxGG<jd&EdU@C!JPl2{efY81&Ewe(4j!)G|KzWl9<zTu
zehwbGytuzz21C<bgTg{JP5UOrFCZkHnf0-VOZ#bf?rA-;`Y$PoQd~#~<+Bhaq^g`A
z4oy{4wrd>!SdBTU&Z=*Ucmxtsm;-T`4G<;CcL^g=Bq8HT)2en7#xE_7Sam!wlA*s*
z?ATGr&<?>&|5App?QW!Q9*|76l})KMea>X<IE|C);KhW`3AFgGlb4A>;j{E<%d6^H
z^_qRmH<ir>F$F*GXaUOz75HFp5bF=nZ|QhnRMo5yS7;0L_incZZXWGkRP>P%@dyIr
z0@?1&f=Ko6zM^nz@lxRh24C#;DibS8CnN}R4;bDe!0+3A?X&U0&&qK|7O3nXk-s7<
zH)MS7R-Iog<))&p1Bn;|X0I698DPT@Px|xU4g(6GB2X5Qi<3jE)fVe3ud`JjA37(s
z1CwKH=7A7cAYq6nA`tyv=&0Uo8~ZY!7})1Zrm;DzP<tZ?Uv@TBtr9_TgGLX(+^EuA
zziT<ek#Fs8s%rUZa7_vlhyg{-h4>(R`0GDZed|pBUN^qE9obMb<e{_KhHsW_%fLim
znlx&I+dsBfJd0v~pMG@#uh%hWbq}PjeVzL4EjqK-P7WVBAG#fl`+6fB;_#ZJ{bbYb
z@o~6Q%sUyXQeVzZZg*rQ8eds5Sh)UnJ6E9~VFt7`hZ+qdYTCpq8)z`~_F@F2TZoB;
ziL}a^ygjd+Y=*mTG`+$XCiXL%dQ;$CO0V=-WnT;M$E0}SguwTM>=^4iN__Vc`r8Zm
zhQkZ~VIqWsSIZ^nc51l7l&5xlk#EKT8Fh!<zz8wzPrMl@?wb4)mc1q52qEy!p=qUR
zYdE=|M`Ae{W34~s%=2qVM}x_-PMdNaKZqt+tpQZ7m{`7u?i(;OFgL(%Yg;f@t*zYW
zt@Nb&Y+iOud~EkB)bIXx61p*DqbSl21z6pOgIFHzeR<?~H9i^6{u!-W8lp(AnjAn8
zq3gT4`UWR4kZg^P{^c~7P#=A0ti0YEsZ@_UEac`A88uu)<$m5qMc>=t=Nn=d>Tf@~
zV)DPq3V7-Kor{q#(JxqHCe=&&N^wfz$0Tz2*;Kl>*yWB9=DxllBDovCs0qHBBK4<u
z{su|}8U!-#B8kE8ykU{ionvLJgU;la$1IJL%t(~I#o;NalEZ<j&us?)F6e?eVGRh|
ztO7(ZT>EVG*C5CYpJ6k<ILJ{pKe}u3sU<x9hN!A?8c+mHn0#j`EfBN&YNgHfSv4|F
z*-9+&mP>xgu(`nLS~b|JYR9RzYe5y7dH7pucQW%L-36$KS?C&X+dvGX=GrG9j#x)b
z90X{DNA^|na`6eyjXuejw*g}Po2OxieMe@}ZXRoSeX-+jUc#~d7v`L(%A$aI0sfi-
z>cW|ER<cVjMBy!`!?JY^D^(7vHjCXtbr3#3<exGh!>%P}y0Urx6qLFqAi3Y>m{s^W
zHY4{lZ7prv<67_5es_n?_2maU`)%S;>^04-<CoMn(-!ZU{7u0UI5X+|h4YFhOV)KF
z5tHx_24%>1rMQ;CgkMD&Kbj}Yr;euA=pI|Qc%zNBjRor{u;xdAf>cct;aXsGcp;el
zNjuRRz@^M=?d0-bO>?q(2zZ}L&+^8l#GIFYP}mpBE6jF~{%>G5SY4M&fzRiR+f|tK
z`g#oUdjN7yZg4{!U__2-kmu$x)zwu)>VJxxP2q>6ZPv&04RWds^8U_uXzB1!OZWtv
zg(QlGBVkKerd@<;@KQ?@3jID`Mf-JuXBqZXDZ&&<=8zijmu3Z~)wjQLtaTjNu)*ql
z{xbxALkzI;gqPz)Xa8rkz!<@W{fm^+b4%184l!pPVx}FUXO`zz6b2}zA2TFDcvskS
zx1IA=kd9&b-4t7nXr$eVEMfr=B+l9c0LQVqo$OiTWS9j6G;<6%dEU6W82C9$+0Xq6
z#4Md&iw1~F1++&=^9KYj;kOk7gkp)Wa}r)80Cf9XP2}4f;ro?GSFT$dF+41zeiE81
zrvHiT{HZ1?tQ(@l9!NGl%-w%;Br5AL1|;3)o_3on2N1cE^PTGMLMcc&<xpaW58ifj
zzE|Z0ksh4Cz1;MY$BB>D2a=xFRGe$~u4z<U@xHO)L^qReFws3>W!24n(~e1y4@s{b
zPg$!o8$eb@(?$x3!gcUVt(}};*OYx$aZ%Ql;-zWC{@J8C;|2a;v*hDP81WYnM`Gjd
zU!GmI|F?0LGPX8#H2dk7nf^0rqfkY>`e%#P)vZgAgN9Xu5f?|!zz;9Iu9_3v#Gjz^
z2ISjhmhOuxn3s9D?9%cx>{xY`$pLOw=d}phhJkAWg!^)JPv+@#BMkt-wi8j0vGR)n
zwHrW)+QX;v`!>N2H0z4M$ohQy+J222a(#LZzQ~4O{V;qwPCq8+UXEnw(+4W5q}&%d
zg7h%$t>e?311ZelD0R`zT>Nc4kh=igJjZoPIw^1szc#5$UsSlQnhDHri&Ws_69@*F
zi@0Y8-*cMh<dOk^6->TsP8Sv-<bw#VhRzF=ED#ZozlU9DhiUXn%viC6Ra%`jo2wq5
zY$LPe!;-Ec$?6hyPr=%<)Tub9gYMX<=<KrZGNrsZGnoo*5jRTa@0cDLk2R{0x!)eS
zeTs3VG8frt1vqJEqRIUJA3!Spi;cIL(LL-7^@C+;W#7)LOY)k}6>wmttQ5Woe-d#7
zW8Qf0G&&5hZ;gdl#E#OUDXQDijn#|q3$QaFz%r0o=Fa@GGs#8Ug7l50z?cdbB5luM
zt3{MLZ1HORtjSzz<*TWy6N%10VX=4e4H9NP1Y=mP;_jjsMtlSKq+?d|l%3@wyp$AY
zk#U93AAW`(=^ncDBy1PLjs+Z%lbp&iG(>+voNtM`DpIYe!ZwcDE{2Y9SxLr~nXJxO
zT4<}t8eat1^AJ3|uGz-jshaFF(A#)ML~AqzRdYk|qW!JMIuebyDtIIDK>M~4L-WZU
zkfZS6{PTx2<-I;x5Pu_o*aRpD`R-ND=-!PhU#hbm=H3l-#^=RUrfYIq@$wD%b!{ZR
zbL%EdvR)6<MC>3cTw(4b<<lP;=5(V3-Zjwg8Oa{Ot;_%7+5FRby};I8=Vq=xs5*;j
zA!E#@<Aw!)lD51JABB#~#rTIGQeXjTb9e0Alns%^=uZ=!V*mq+{uR__ePw@>vwY4;
zwY6h!fH*;fn?xuiUhs|6$btGrIOm`#3o2@1*=N()#>(ZP#%ebls}8yJM=fqREF5fn
zTWV!PF_Hb?rD<dO0Ak-v-wIV|XY?`Jp<h5(@y%Y~Cjto~*?~yfF7ITC;+XUA=gg(U
z_Lxjo*MrhOsTPCu%yM=sY--I48TRS&Iu_+obzL_{z%KR2l7tCC{-7;K0cd>)9HTA7
z#1Qj#)!-F=W(gV3?$oTOaVb>oiIF6RxLK<wSnMg)=^D_=vdu+XXyWKlka%3}Ic=fP
z89fRkG4wFQU}u7jbq|hUwPt}<lKOvk<m$WaIFz-Ea2+GDFEl**xA|n1MG4|LIHm~Z
zj$^oOVuO>E-Rh78%R8>u-xAT=rcJ~G1it1T%Q>L$$H{7mZ>G$o(8kemMSd%|;*0wU
zW=XAm@OZr8aW$)W{aayS{h$ANX7>NR_V^Ei(?Eq@>vejV&SNUa1Yfm6Bgwib^J7Ph
zu!fNhouFJfLlVTDHGzwUlP>SNQ<z6up4hxCqV0STBUzt~sjBeRMMmimeGy3EFdT7|
z*iwTS35A3Mv_*tZT5l~Hj^k`OY$Py^AY9;+qxZes5jJ^hV!8f2HOd@W`JI_Tlg1E7
ziSWfwce~(_MfX&jyGQA$;o<D^N1CE=8`*O2Ef^7BeR{Jv204*u|NZd;HM1m)z<Yh%
zKVV2ojM2V!00hSblz?H+m%qzX97q{I;j2@UVs3EZH*D!i>4nCAl|T{+f*Ha*)2pMJ
z0(>I**v5$!!H|_M`6kBfvDAe5Lw6|%!cQe#1+-R6b~k4Jc{dbkhd$C;o;+Pt`HH5h
zRbcUuiYg8<ZE&ll>$=+nN8R>vLEkn}i<PIu5Yh(m)$=<q%{zPf_((*IrO*5rhPFxb
zfeiW}$I8!jPB?pVe)ml`95PDQ2`iw4HH2we<yg-IXM)4);97yJf38(vML##`U~le+
zEp9&wNM!T4Yr2=VqN09rKd$-H>C~wNw|~`d?Cs*k*S9AAM`oRNuR*{hw9rAjwFxKU
zMwYZdPeOuOUo>~?>b&Uk19f_`HkmoI;n}9jBp%Nz)Q#aC`Y?JG`Cm>9`@bQ6+5fw-
zE>lhYpVZ|aa?6ZnBmczB%<3?TcnpwmtkHUM0YnSxGO0>Z2d8uWcbDLl9BXG(EQmLP
z&D8BAo8XWC=jLD&+WXGS>vL~$U4t~wNZ~-;-YdO?aI`G#&?!}mZS+{l*7<S!F?&2o
zh@O_1*^Mxc(Nvul9?Q9AH}iBNhiu)dN`;YX9=`sHluMn-pu2nU`rVq*AA{^-6YdjE
zb4?rEbeIi=WM5$i*5;Bq6-?!r`?K6e>fE`*J-xKR#EAg97|hg3ShOVLSECqXA_HLD
zJ5g0L7#pMx>8}gQg!Nz6lT1mz>UlI+3*(-heB_R|I<d=AC~lr(Q&Q_&<FtyKO`rMo
z60~#0-b1~m!!|8&#ZL1*kzU1`8J&Erln=cnTNXR%j3uy934T9Y`|edv58Y5olJsLe
zhh^dRV?CD_qf@jgl2kM>7ria2*rpYa2K*21-QO!(MJ6m-%BLGn2UpaYO<-uz?#(v4
zLkfi#7KN^0xOimw>B;QTnC*(-=ZgW8C+~*azDOM=>WQ(hZ$q4r@?D%>^!*Sfy-}j5
zdVWGazP;AeYcTHFq|(sOB`+p3T-Y->nc&VtOw3RIXl9&@Tn{#D-d@dWUF;B*XK|2F
zA>RCYT&p(OTI~>oet_Mpw+K{TqW#gD&?dSzwQN_Dn2SZ|YfsTDH)g)~OP|(!vJGw=
z9Zb>eUO$-5Y%C3BjPkoZ4vmcW04YA)jfxK(G$1iCy?~CW>5o&0I9mHU<k2JuAc~oL
zO0x>Tp|PJ~U<-QE1J()zC+J}G2|=|{p2Vm<!oc>f&y{Tq4S5ip7MrZzA);y5O@^k)
zisXfb7GoOZJA_CgtG_q4er3;Yev1DVh|E*ovHXsJ&9U2Lwhy96vy46mIx;IVJEU(1
z_lc6kvC>pQg?uqO@Gt#m1a|rc5V+J@cU+2lF$8uE=}1Z-2jQZ8ez&~)HMjcQUo1Y<
zuKB!v`YLrA2jtA9RC7Fec83$<M3@1L$ZsO-8EZJ`{x1(pqM7^ig9CASLR^G}#!da#
z_S1Gff@~sZKsIDWl2jRtam^&q)-iT_@-f<p5N^Y5>3`gO>@K-4{ug8K6eUO(Z0nY7
z+qSxF+qP|^%eHOXc2$>c+g6u-YTvu>)9EqppO5*FIT9mct%w=(1G@5Q87*%z%(7Em
zm$Ny}J}Z&BBWu^^h`s4}wD?=uy3o$gm7x3+z4o{I_;Y;t<0N+SrKy%kGPBKh1zH>X
z8Mk;w2v@_+c6B4RKnc_p`ry@xFj>duzaUVD&ES*yS4`eUN_`KIK+xO#EM)o$a_X9|
zJ0&)3VTM5dTtiULJ_;&f;E?xI!K!hZ^--C*@Ldnw9Z`buTd681t)bKP9RlW(hP(<=
z5ON3r{|VHLbS8e{$vCBu2WR%BaOXbC<|~~`)a9p;l|R+4vP4YgOVyDdUaS3`B9d~6
z?o|S_7RFER*G{}hTOd&;C#*W$u;p8-8+D%AXwF+UnXegC!N|FC_ZZr{15W4q_kyu(
z*)4Pt6KPO^Gd0+_tLXRj<VV7&$?O({a-+~q*+F}tngu{)lv5|1ly`;7**hfpGG1)m
zQ3!^$Xt;Ow%e#b*?7r!*N`iKQX(JO6hi8%Tk3|KJ6NPw4@rKU!B`J$5=YF(!i=fjn
zkP5djZ%s5_B9mXx?9W$(yJwC$RY4#nW4GaVM)L+#Owl^@AKoOGrd2;i7F)*{)avj|
zrmzpBpo`ZWD$`b5HQfdjcMtNzaI7R4_kJ{c+A-RWg?84^79)XaS$3T+Tx`Y+`rqx@
zG^S#ysrrmTQFxaS8FMjnwjE*i0p?x$Dx!~d_kYBd4N4xzlx#?~l^6L;uf(PBtRO>u
z1F=|&rW=n{Qe*Y_eRJ;7#0pZMKN&u$tRH-Kz__Pr`yZQqm<pGKb6^)29&Guku0l)o
zIxYsyZR7yw*0~neZFboDxuQM%E>_6?qJg}>o@&zDyQ?argv<X*+|zhozM*rH*15gf
zlX=+Hh36&O?4$KzG~x#qH|EZ9R4Fcw*D5T<^xg%j))uk)L$ofJj;Z+_MOU1{^cmgK
zn`t9dDS7~$Om|Lql37?pWej*I_2R1bi+XE+fa1i#dsuF}pWD7lDlyY1zch_g_9Jar
zAn-dBLiHq$f76LXLt-dNabkxsAe%_Q-tSMQo%&6VjLgU3Oc)eYHTMU)gNo3LQi9cn
zM^@NFK`5T*C=cxW9vzDXD*J<cf?frZKrN2t)xn1Er2<+x0hFH}X11eyQZSoEI_m={
znMSsNLppQKIls{$3uuD5#I&b9gX=&XRIBOLBi-R&2!FBwB&;8~-O@|nXALgXEm4;&
zF=7iC@<=dMd)b~wae)ms^n)+ec<b$7KndH)*++h%r^!g{iU`Hy2|jGz7ZnMD5YaL+
zrso_7U$p#n%W@?;H2aDOaU9p5s3U!G7)YA*pwr=rhUE^{Q?yb-sN2{0F(DDmhmEpy
zK@RFTy{94Yi0&W~q7A5U*1rERBVwkNTAyTyNw74cO<5pfSEKg<-&SaC_OHzRuaS2)
zmjA9{QQvU<pCN%2HTx2JW%2-rJy7Y4>9*+0js^DU>K1reQ6Z(csSqO6gouo@!=27H
zc@lF?#mz1VbfCz_PBV09gOln0>DSZJwbR2%BiWz=Kd@=*v%0>!ZUBQBH&f`7Vc4^m
z&&!oWW<hbY<V(i5yp)fRiL;x=vn6*yqD!QxVdPk&=ii<j-JH0gIb5A>+$;DyUk;ue
zooPDKwT~}T>9+KA2Hy{!C1Pf3cDfhalq}qiGjyhnXVg58Z&Q_Lc1ZIO*tnih`Y3k*
zy!xT+lVZFjF>>rgx`&gheowUygW8Na^11CBOw}@MDbXi26O4Pd4r;^Hg=}!RaDN1J
z5M-nD1$j^s`ci&Vk9VyaYjEn-A}cRB)QE6H#Ri~UB5I94@ycBPDk&PhV4tFft?pry
zPZdc)-C*fmxOXE$H_?s^=C%n%P===r8|K~LRU&;mAe+8uk_}D>+uEoS=jC4bw9(wT
z&{eEbC90ju=$Ng3G0O%-V=m9%9o7z4w^o+pUa1jK5cl3WYdy-oUoEY?C%?{^7R49i
zs>b}ouRJV7t`U8R1N~cl0#}gYZ^0;Ka3fB)H<h3eM-)UI{v3qQ?-Di*W;L=Vh&!%J
zKt6YFK^(NQJ_{Z;NQUc<kz=>mz&r^QFqwx>D;gh+cxJ*hBB-WpKL}B7S^~O>mqr&N
z>22*EKHcl+fZSsjxrmG}?%Me!XyQrm)i8-n!-C+SYb$Dn1je+if_XU>*ff_1U6W)s
zfV7<Zmz_JKt{f}XFVfaE$H=tK+8BB8j1Q^0Rk$}#2Ne%7f}Os%_gimv(TsP)6@$Ez
zSbn_1y|eSJKYbQT07$tu%y*SS568Uj4Sy|9E&*)N_!rG2o}~JvnqN?oaU#PT#yM=*
zM1ZipdUf9x+phoCICfLCIijGv2~GiVlQ<P?DWc!WkYF`K3G$#^_H3hxTw*&2Gg-mQ
zP#o0QY2o8e2W8>sUgv<NgLd%qcjU2K?qxZ_BZH<H_FkOHiNV_UV<lj{cx22gB=`5>
zKfnj?wsY6CQBBB>7@*zg{f+=$0t>j6_hDh_vSzny=u8&q`N-9BX_L<2uhI{rl~m}1
zfi#<HuDSVJu}qvU7HaFQLQ)>ybYAU8z}UU|Tk^i&GXj`+v?gKGBxt0fbQZiYXv(pO
z8ZXRkU6n_G1Urgby|e<ad?`n9`WX?9$V<wJXHu=8haL@>zatcB7!ejoD~NS(M8?t@
zQuVft5y81)>WU;pF5jy)ercIEXz_%?FRuQOegO+A^pz8gTp1jI39PV3BZE_^AcBtt
z;Gm-Z%$At~3<_!Tp+{LY1Sa&Fd{ljK3xY>6V@$S28@^oaPIP+E)16kQg3PDqOG&PK
zAk9>e2cyk2_k#u)-<nk;>27&aoK^((-V!^j3lVHc(aB+{VAV*X96I<9JLhx=VNz65
zD|#cyhVq1lsiiqHFzDEynzRW<H@44!cBQ{#d-@{V^Bs}cq4K?supfe=Md)o>LnU=Q
zP{(j-1VE4d$ti7t1}j5?f+ct=DIi?L07M%16zui3RPHs#U^b4m_((|B`5P^GiO?B5
zZzU?;*}NSwMR|~WIlMbBrs8&GdhQPd{geIA6y?0I6xfJ|ubqG2+PxQ^uZc!+^2bXU
zgaNzK$w`8=^|z;wRRzd9jjE9Zf8wlCCf$N`m$eQdUtVt|5!=vZiLTvwezKSHSxEhK
z)MBhW|0^AHM%ZS>?1AlkkU0k&YGDin6;Lc4kVyvVTUo0;I9{?~8Ld!)P)=?i^ikum
znmDraJgnP_2&e}ktz;slvMg#OOZkG_KA>5be5$tMZ%6tB_$7z0p->#^rKVi8ds}O(
zT9<|?zxTxgw;&IwU-129(#ybMjy177Db!qqtJ}SyLi181ju88hh7m2T7eNjsK@;zJ
z%L6=!YqwxdgF=>(?~=kOUh;e|Em2Q)$idVzBp&R)vi!@pJ^>S{6via(lmDzRMcXld
zLZv1vFSbKTK&Y7MLv$cWpk2J{966nz^NFwa$MwT7YAtVn1L1vzHVuF*AsaF@1U36!
zGO&pN@oebn3dhZc)8<zPJ>$=W(NKE2hTn@=ftjGkhi~~xN~3|j(_0|9k3P#}Rdjd(
zgy?+Zr<YEVQD^inkm;d9$MgK<2W6UmeD<%ie7exAr`D%HcLojO0+ffq*WJz;vFFR@
z2I_KR;|AhtCd^!k{(akpG~u4ObUwx$<>}@R8>uEaCd>>|li2dqH?gq3F58Qy59v&!
z`c9xKxGI{5Xcmv}p7VWb0AI$M9sLCiH{SsIkL>tQoH#T4e@4?xscYCB{OntwYY}GA
z+v_m#h)7L4K|PIuY1Z>#KjaCpK_ZAY1YO)+NFXA<p5frv5@@(;HVUBo@gg!XeQ@AH
z2l}VS`==id--izOfe8oW#i6LOhUN|a?B(#U9!&3#wL3wIg2<d`>p!2j-uQeSl~$g8
z?jH<+$GQZwi>AHiDNVdc-$~cuf`;_<AQR9O)wZ3`zGuBr#QlNr{*?_KwN+5gCR2SV
z;3^AI;NNRP0BBE>$imun(@8{8ku&v9K6w{=$7DT^l~Ghm&CsEXUqK>dY)a9bq;ak>
zPK%oImI9U{DFvCcbkPQfKkm-Vij`gtelgOt8#=g3(wckKYfH#mvt(|CU-%2dXzOg<
zVq_E>-0vA*c({3gx=6G762Z#Md_oc)kN{T*6FQ8s8X_OjY)~8wN4;JsiH$jzn2^UT
z-ha^LFV4vV$59S6wJDZ@Qi>AM7JKbaXWV9<J#eK?HsZJG8njR)JKm`8pR{F{kD~3g
zPRI#{$cr`>{M>=2aOwqJv=(N#!e$Ge<h==H&ueM9SZ8_Y5OsZiF#vjid`<YUt=YJ;
z)vk<XCv~ztBTGK=_Ixw~S2YK`)O?Y!tr?lL6^U#N^;K`_$kXR^ixKl+x~XH4`O+*d
zA<ddBp<LVAnWPR)X<oUOL^B8>;esfl@i1f)<fR!9@UOm~v1dR7Sxl~pvD%Y@CV$9}
z3<196+mUp{^T$nFA(GG+(qMg7VWt@|X-&4vRkh&U81In+{mNhjs4!t{=I=UHUu?un
zK$sRnAS@KdeN0dTjgZ=I#Qgi~<1goKH!Poo6=2#9EwK(aRm`7+OPOhCnDppryJU@t
zmc6#myBY~?R9VuHuWcjP4Y(wwy(Cexr4#yGoX<J9Zfam|<#C@fLSLc~Yzh8s>JR}K
za>3pA<8N3GS`6YqwDG7MD(};96)J=bNTa1A!JGDzY(Jr^KD~HaiMN-P#6x@>JqqO{
zZFsK2r#Az=ArQ;VfUZE?UcGN#yrU+%%;#A=DDIOpT68IjZ?fk=D1&B9^cHeUJ54%a
zCdE@6(%|-dom%qC{a}o(Qk&evchrrC@S8qvuG;KsLpgy%%gjy+!PE+A!+xYnpiSJ+
z4>_g3H`b!y=Sof>JMKR)81Yut<>%m`I)6*`(oGP(uLSVVxCRp3g1YB^Aa&WHpMHrw
zZ_xEV_EwB3f6OUZqR;evVFOzN2HdVfnt$GwL|Z<6;K5pblTc$8H40y)*yxHw3Lyct
z%fi?}-MXBE{+P#nJUtb0C5}I+STty^oK>DpK_$v)=HM_bMW2<cKzUdoo%in<jFj4w
zy|@DyVnaSL-@!R!iHo0r6@VFl^ImE|(=_cal@f=g#o50T5r+-$&Bl;23IsS319?Cg
z+eJZUYf^Mb(K_F(vs&}|f*|=zCyOty*ulv<`r@DlPel7gy2&-^UK~+oEWMqs+XsX+
zaylrV3D8l$7o3mZKtG2P1wzXt)J|R&64^9#arO|w`8S&e<s9Pqqe3r+i@p^Yd^s=9
z`lY^DTeEw?^D*J=q3q?`rwKwV$Bgt7!z?}X)^<M&jI3L8ZhKaruzH}Rx<GiHE&ZNj
zW5^Rx${4=BRQ!mW?J?ljBcKdC*p>!-p%ZFyYlpA}l<=Ah40<G8oW5V&hMA&8(=wi|
z^ZKR?U({*sxXeurg<?0o<hVmb-RG+=6V3!KM)g%&J5`mo?jAh&lNt(5{#g*7oX?B6
zoew3M>q|}yiRi`j<>L=#XT_0&50c|nRy0Q&e|gTA1lnWlHH7M01=mPha|_`5q?>F~
zL=xP|l(T!q?&SmU{Ll+97E`PBuh9K3(Fr5le-GWNTmLh|<@+<k)jGyMb*qHbS;N1+
zZV~cp4atVy0&$e}3L=hF0AJ#1=Vr9fn)-fTAATSYd>hwe<o#t!fn^nhg`Mv;E+=cc
z7=mREF*d*%H;h7_iAcHPM;X~_KtsMWpDT9~RUfF7f~4P%`}*>Fx#H}~lg=a~BsykH
zg^ZOmLua;PmZ%<bHgv>c`}Op`X5tyIJ82u&n{nXj0QuIKiPHZ1xP)FbtuNus-YdSa
zMvmg${>Jh?)Zr}X4Kv`$xx%d<WnSF&=nx?*>mlu2FsE`GlNj%6&t;M{Z@<i<vS3(M
z7$jfNPZt~uDi|-`$gd!3hzfp&CSCG8Y!)^w0Ay)TUdC8kKoZV@Us-iu!(7?PY1MG`
zTw}b;n|Qa{rzYuPK&DjV-gdPytw^KoE^!PT6ib21lBqAI*XehapGilp@Piq^pJ6tH
z-^JzgOx(R~xz;@Ffco|B>Q<<g0lARB_8_vH*O$S|U|PFI3g%djgf}XT3^~`?hIe03
z2ep_XDY)<?63WdfCA83%T_NCVwa&VMVO;U6W~&>$s?zlR3BN92sUSXv@gRY;KWQjP
z4DSUFA<+V4)m*w|d;*q9ox4Z}G@gd}b|}}Jir%`>#^%EK%;cQ`14H<R#sU@6x`z)g
z$QXvF97G~PO-Ro2m#PQoLjvw!q67dqvA2=9kpUAGR}Eb_={dUzSA8sD5iK=RGKW|K
zJT%ipRB>Ko5xcJCj=uoo0Pt`uL5l4&!qSAu35cj(yKU~$W>{AI=k_L5g~ip0paD*>
zgJCcPElcAl3}qmJdyRMmPvo||Y$=2oJo|rJf*H1Pus|H5>o$K3LU6&45Vs5Y9c9|)
zPa8bOd0O!AufKW)UcZs|?lATRp(z-4gm4Oy-V4-Kaq~ppqQq0<(bxSc+gz<be{C_i
zkARX3ax>clz}^-L?{q*y&)Jx86<%ST!Uh$<Y?!FNbXPqvt%&0t|At48XCO)dv8zD)
zK?5t+ko(1XZRl9#<J{R_BLJICCRRp@SQHKXj-D3_jPU#VeN{~uHLGbR$+j<W>4-IE
zpw&KXr`;QfQV=7gq!v9ihB!hjV_r3Tr;^AM;1;k!_dB1vo(RZvM%>LHqv1S~s5ekj
z0|K{vu78Op<Rtr|KDy=4M(2AKHbr0%Fk%A24Rp^PgeCrn(SUx*gsy9tm2BRIsHMk+
z-DM}MUb-1w^<=|s9%qIIYCO9kFPDm5%{sah6$B_iY*XG+q?^d2c}DIj_T_v}jla=N
zZ=%lBZt<~^mA*h(_O~@Rcr`$-GhrDx?1ShE8>#K};<5@LT|1<4Y{x75#_O|n02HKM
z#)Dlve&o096A!J-){)99nv(oOWT%T!lo{XZA{_HAq19h&#;B&v8GccAWX>l_?FU?>
z7Yfd~Jo)|e01?2bdhqvajceC3p!bjqHL^{<Rwqo`dSv~h99@x6w17JY1ndQ2;k#R|
zuK_}$5aI-q1bFf%8<iEQ`E|a2VpZgqJLUaHRc{Q?NzS6A)LnSM`OTApxA&Ah1VS#&
zNe$1x=z)qfD*tz&V`KRBe|O`U7=O<GKga*S%|o%eOzb8rQun*sDcUcFsko%FSP%{p
zQC^(E$RW=<jzRx^i+Z!PvSLxGjJwO)9Wk`xl*4qq0gmtjb>k(YrAn&1Q`Y^tncA20
z*MY~%T3;#?5=XLN$-)coBx425B<W-iXB`pWLBYDI+?8HgqB(r6ekIKBlGx_$^eI&;
zJhwOdDjTN@Ri96#3}I4g=JfW@AA(c>H<H57?;Sy*f;j<Y?c4XX;uaDFGE)eosIgxh
z5ccyXjueaF$o04@m8(V<ig)`T+-yLjoe=Tmu;WK;eqBn7F=yayqX-@F&~nnCv0R<y
zS+S+m^LBVgw0)YxceL6FNx7DsMW_BsR&oLax|LB%IfSGv_8Lvn*hEIC-K3gMijZxc
zWBH05*`noYO4is3&9r-NidJfvJQ#QV9NUJ$Mg(tWu*keu@L^!-Z0Q%Y!B(16b^~d$
zZj@2M>s_l|ejoT8b!|Mj74U7^D5a4hIcVbx@V(_v0Wn6z-+-f7dos3o<5By#dj1|3
znhiSh1g~uWV-*kni@?37fC=tpXCvYov6D^8BH`{upVeCN4`$shWVFcGUlwS$1$Jd9
zc9GcpC$Zr4@y(69;k)xzdgE0*<U|Jpk2ZMny;;<N*`ZPilDx=xG4#%}TfkG!+pLzX
z<lwaO2`8{PA^p`}8Jio;n**wB;yk2)=T>~&F13ej+2K-1dFdK}AvO3@@(qtj^h_E?
zM<DX9<oXeS+BgXJ$V+R10CCg;0%-F_mdtdZCZw8uHdybrU(p<A>l$%?NOyNg66g`+
z$n}Os!R9RPn>R}PuPoG?ORRd%#dJJ47n<X$`hr$mR&u)POSoFL<@y4>FP<>EVUV{c
zK9T}^2`Spt=w3K@JwUTrDAk9h9&vg(e?0W1WzXO-JCgO=MiGzGk2%aN!h~*g0IC>E
zh~6Xf+6u$EIY6z&^u@3v)WOHVl~z~^()YXEk3iGXiH3HJul?Q4bji4C17P28oY`<0
zlw+m9vNp=0$o}>{za!ep#CmPO=Cm`mxbUz?bsK(BEo<|SQAtRuqF73+4ti1H<#s;V
z<9PN))?AF-)-&SAmYiLB>r1spFc4;_1hM5D7j8eYd<gJmOrQ77IU~vz+{J{0!v9pW
zYgDJKQ#h2`2~c->eBLuSijw52ee5F{ETvcCec$_rE+J|%z*)DcfVV1d$L_In*l4HW
z3<+gZ6i;*p2Gh!LjFOp6f(;dqoP1Q6daUrx?J->4gvkkm_{&RW5aIVs@DAOJN8?Y1
zE@oZ3Pz15{gt{^YWMmFSQ6fiXJWp|6CNx7WiTfps54))RaT^AkV0`>Z<y0iHMrt>X
zc~1Vi63BTKn65dd+%m04L`eV~S=H>C<?soP?H6c&+x&4hyT*kEyLmF*f&*8^-D0bp
z6;J0UZAArjMQ?idqFfrERJl9CncPa(<JCPZZAQbl=o0rA39>SgKq6Q}BJ8$6z#|59
z{88x>*$F`A$zhRYD2d%aci>55v`%G7@fnDWxN_-W&rq3syk|pR?D+floYSEuV`?Me
z?-oNmBd}c|Zd-ppClycoM-ouEhy2c@GdW2WMkn9TXTHz!JK^?@&<HWFydxTWt6T!&
zlM)(l2bx=5mP?{9)S>8b<PBM!<e)PM&{!;@-s^p+>8|f+*S3UCX7HC+4|8?9lIH?x
z;O=Qx-<O)kq;D*K(OubsM(s%|TYNcJ@9cF~R;<&mj_I(!zhBrrfGYPTeA`{Xr+Gj*
z`k<c(EBUlGdGajt@;sC7&3<9f*h&L|Ov)!bofC_e)lWB&^6k@xpd=+T%k0VT(c;`_
z^(UtvNZ7xIZdX1C_QRl!D-MFB*7yVZCf%lwW8W|S-f(M@i!-+edN4Fk+<PH60L?~*
zG#$aB6PC*re!K~%9TCB8#7&Lv^;`9wQs88I&eZYbx-rWm#L7(i=KAA2fhb&BBL`J4
zLPW6yg&LZM+nc5yt!((7Mr()x=bTJF?x0zvVk6|YfP|=zLc*(b3E(@LJ(IFK&D4TZ
z31zz!K-(m8chlZlQ%@UG?uyR+%ZdW*cS&WhJaLR}$7S0qW;Rn^V5yDkBfm(quUAOE
zNJtJja$^qo9{Ns>vxx%+lO`P#)lLEt2UnyPN`rr9D|%v95~NUi3+IK1oPeq+W<_>z
z8uePBMbQu@rWz-B@K%LybN8!hqlyQIMlX|2ez0Wf;)PUUr3ZCjKZkG%$1P@Me#@<>
zyPGGSxA+NP;zBTSDJ!6g+D^`;?c|6u-8r1g$Rg3?R?jGo?f2kSWunh%&U_;Xp<HE4
z(KNj>NulxV)PcIzLt#VHQo%mK=lkBF7Ja{vSRI&K)i0g?4!A268)Arq+{IWwC$_~=
z5&ZLmg|Q-x3vqv0ux3_Pe#We+eXZig7uP8Psl<it$Q;7xizn8Psa1%8{O&onmX^<h
z2YN%wy^76@kgr&r9?b5{H`*Roqa1!E#aY5$N)sIZr*D_vTb4qkUL5j2-l6{W_q6m$
ze2HW=mkFzubZ#zm>GY<*7MZ9e2%h#{_6DHMB=`P`*T-!(KiZp{x_o|w3UfUu0pIFh
zc&11*0DsjT7CjyeU8bf=L)5dkc+hEN`FbVQENMg3|01PSemXsDAmRFs`2u>QM5xtZ
z^88??p$1h#7Sge0X(dvip%rcdjU*4x@)?>iTc33MhQ$Ti`)s!<(GY*VTP!iS<fOAV
zWV_)tMd`M?=<v4h{rtR`%gENKZl2oyq;mu5y_R+wLV~pPFWy9rkuCf!LBprp`^I-3
z$btK>mB>H%JdA9N|La8epH?Dv8|(<(&uS2>kuZxe5(-I8{BRZ-K7=-b(QePPbU?!e
zxTp<>5svduD>rdMYYCU^{nz0_89380J6)LRU=w~|c6wbI-RvI31q8(DhfHa-0peiw
zaX}I6X-EC34DpG+rueek48xwC>KRD=MzedY*}Yty89E<~>3`EgqT<I`QdqjUJ3F|!
zdi$0x8e`Q`Z}_^K*LCH6z1d={(;9^1*d+rf<-{{2x(=yN?slE~AdU#;{6G^D*rfu~
zii5&w(yDj+E@kCP6tY4hWi&J?A){>_%5`_yYb$G${vMSm*eIxdjWj32;wd~=P%PQ3
zomF6?XwJsO=S#!|TroMu-+_ssEQ^9ATk2G_^ReExok*Z4{Xu5rZlAx>$vzkgu7_$t
z_dmf@M_5QJx^=hpVm7cSTX9)9#~G;A#Hn7AZCjcssEDu(SHh0%zj-XtvHKvz&Y!QP
z4*Y=!=}}BxEJ><W6ArYd1Hn*C=u1P+yxix^auUr44GQKF=W<ccgV0g2RgU98b4Dr(
zns6^FEKg)UPEFmFHIk4jM#)xC@p)>cB#^B2UcM`(-iZr784o1V_fOC~!pD{Dl~kOY
z7LV_zdI>=QoQe{zBvY>Ew#{H$JSs6lnIIwK<$EqtH`ME@uayUe0|P~Ee5T?x*%mBY
zg;u|b)jLAoBds9Xn^vFLIMhz$BIiXSm=hGF%Lo<13#f|AGNt5sJE^$z4LTTcUUnok
z3hI`bsARUJc_C=-WXo64tg8Ha4vp5VmI8S81wKl(kUoR+f;$ZB&gU7Qk3dy*sp)Q0
zjSjG0ir&#yD;xwGX0$1csBoH@R4ES)$>{<-=v$_V;XpDJo@W8&jS@%k=-hh_i-I*n
zq4DjGUhw%c)XgMk7Y7S~N-)98yrHFFI?jE47sVz)Q}?7;?<|z~L$kaluO~*A@Pz;P
z!^wVo2wy}@s$<8j@Urk9779l5Zu|OxEUw;}mVH(y*m_Im@GNw#*+g|Zkpg&`^sEnK
zk~Kbe&{U)4&hgUO?xuQf$85A=VjR|yd}gD?PDbxb_Sq_IZQaXeE$I{1&q;W`+3gMr
z4)0Mr8s?ajW3+OSlpygISv~NDUOUJu^s+||-E+MgFx|*+0CSKYYN<g??+1WS6RxkF
zO{9UcvWkO9?GOR{L?^D9Yl9fQm|_qvxR+$F<)r0+0y8I6lTFkcd-UF`{1hPdJeeWL
z;dT*sP3fJNQ4!KoxJ2wjg0RoJ&lWXiT%tM^^1(+H%-xGgU%j6&86+U&9?|O3Y85%Z
z2t(z~(aEoa&d%QY8mvzzX(4G7?VGmwF*EdeeOX0K-IM)8LC4mzjeq5BEQ(<QywOwc
z!Q(=hy}leKa4m~Tj);7ELA(3aWW(wiG^Sv*dSd6A)tebkTTqy~s>1wEgApGy&1Sj*
zXY<^7`uG&nktdeH50^}P++`JuOZD6s2yI`ou@Fmp2IUzW*k&w0Q3&Vl1p0_*E;npW
zO~SeBna^qxS&2Kej?anLbb+c&uC1f*T5W%9nZD&$UDChw2j|9tm?*y=V>JXlvAx~5
zVn#-r6BbU3Y^KP>(M&QL#hU^LXwsmEi>@Nr#=`mnaE2(@En#Amb652@8z7RZ-P#e?
z<yuR}e4;<Dmxca&8awoNjxhl6heu>}Hr^hm!7i55sP^d#$z#<DG<V|HD?hv5(MmOG
zZ`jbDPWRsl4}XC)VA=dCejyXS+TGB#Y?o=<5Lie6XJ!L~H7ssbWp=7fLvFw|Qfc$f
zj`?xB1CqOUuPftaCxE_lPO^)>FvVaTc`9kbtM-epgdaotfL(qoWsDsrNqoU?P@;sI
zy`;TTKC_htH~blgkPx;l%oS7H@#Uzs^o{Z4B3uG)+KZaXARt~#%n=uVxC@|1S)4EJ
zq2GB-QooChA@}BenG%s_o{tY(sFArKJMm)aqnV{*VgN9wdmF^r!I@9Ff(m)e9~C-j
z@!KYvrABIqT<Kp;tHgktoy!__VmOMmVA-;>*^aIZEwoiTZD2uWh*rl%FFukSYQ4^F
zV)F7dI^1oAJ7SI4>Z4v^s9WG56QHiMCBNLV@LX)Qe|VlEYqwU0%PKKk)s9<H{ATHG
zj6LXDr=dDxen?CMfgZzR?Ptv><M$+A_G*0Ay~X2i1_^h!>Fi{;-K>Lu?Y?+EzTs7Q
zaeGz(x%&xIcpFI9Y{ScZPyN@o3U9XGZc(p)r*Z(})<pCB&8GDXH$V_^l~CFxMSICb
z+t}i(G@?&NuKmdFCGNj9{*eV4|6#kx^8XQB{)0L;TUjc0fdQfCSxtR^Ub6Y(hkXMm
zj1K8PNEn(7AX4J6K)?oBANcahZj(ARtMe<%H0f10XRtpH45Le@iw-T5OAvx#RGdnn
z9$XrzpY)HmUja~VN)SPes=RpD=63Jf?cJyAW3hW{`@yALLLUTkGK(BeG1(-Lo^}GL
zDnBtzbjIv0AY8fN`Y7|+5EN6?ouk>O?E`Da{4+JhLSpI(Qs`WE=ZyO9_|?^^k>!|i
zN|b(tWb<1!D`6&8B0=m2o7}C-aIJNdTvWL1z4w96@vhIh(T&dPZx^AB=xw%G+>FK9
z*4gG6QrpZl%K7$m@D0_iyY|KY_TTiHki_)}Y?tJ68`1Un2UW1yl;-F@lN_rOm6q09
zu7s?eOo^lq$Ft<w{a4`VaIdBQcbDAx<w|j!hn-op3-DXcS$}g?bii7#sC*f(+-VLB
z>|3C|J%E3!rFv`Bl)mvSx!e101xFIIgh5NJ#1RcgDO@<Wx6cJcS~bWM39;)91Q7`{
z+9KtcS48(fsac4lq~AtJO}Ym#lX5JrG!O(e=*<Xip+)KiV6l1r<I6KK{@;}tM%Mq{
zEv!}_OZX`{ch9KZj+SHf1Blz-O*vd9dDm55mLKYuYR5rD@5&ebF+q91ScnFrh|joF
z9FtJT?(Xv0LW#X5!uxa`LhtR})aCKe7#Ba3hDsJ~-JlpNXrf4`x<74+;2jifxZtzX
zlT#y_9vA!*Cq^`mDwWKwvpNur{tc784b#%zsVh;u-|-kIv{Li9zU*38QcAj|%+hw@
zY$C(_@PY=B6FK8JTugAc(K0~IT~pO#weC}g?RGXc!4fX1YRV+U%s;^tCXkqr1W70{
zc<=q_GYcCLe@6z>VWlu=iowMuw{Kwf5$@`X0j@MRnMvG_(}f`YCK$P$*SOKol!1Fy
z$K_kYmlnN)wVr-e)`UVs53Q(~xYUhlK5KjSoX6G87I@}~PL7U;YD|&hkFJ=*FvOhT
z)Y&-wYGLh-;Mv*Qd~J@{WFJ}9dtb5Zq3oyC2(pFMgv?~bM1NA?Ej&4}m)3;Tt{*Ia
z-}%`F#&^^2%U0CqiPI%|D1q8qN3;)IO%3u#_nkR;tAML@HyE>SUVo%rD^Za_f$X6M
zen1_b=hex~Jlyo{_k7d#a~MRg(wt-%btw28ux%S@(51ywywlohKga`EbatAl5rq#O
z?g<0$0(WZF;ZLihX@EEfj!k)5^7j66dEr8ZXddHs2ll|fj?=$oKCmPnRHnM-N{wfA
znS2l|p}KT$`;=0O0*t+3fsVaD62@xV?FHi#%Z!iH<Y0aPG~7WUr>@e^EFo&3wlt-1
zm_>?Nkn+RUxf|2!Dl3C?mUCxddZ*A9Y=i}d=2+vyGLKo7klpVrbL6VKS{ufo)W(uN
zhaK#wb*kHfCL|PXRE6Qx?Mht@eO3}$JTj>%d$-|<5&s<^HGI%>Va@eYt_}<Q<c{M6
zNeo(*N?$0-p(z1UD#5tGyq@3;p1PkSHn$bl^Tqa(A|};d!{Mi>pb0r@1`v~oP7#nR
zg5+2I&E&(uAs5r>5IgwkT;m7O6C6>8w2P2OK>l2qWNEMxpaTT3c{k&^P*d$AZpDr+
z8Mlk6WPv$`M5BOrM6Jva`Mv7@RZ<Aq&JAxc2J}li9|s^kqNCFstJg_((5N;L%eJI)
z#)$8dCm+1bPu))vK6r+;L^A|vh9g54Gzt6&aFf)X{}}Wj9K8L&vCo<gcvL@&9bqrS
zhV}83A+oY1UqvjT4PhKMkC(dPMXhnE&&K&wccqZe^xdwWiy-nygK@>`RP_Ezo}|K<
zJN4cQzkOf#M56^{fqBUBZ0zMwyRA9sVZ895m;anbJXc%z?Z%I~@DJ%iR&jV%Uve0e
z&r+LWkkDlP`Ygr;(^t+p(^U5TCk7^}FQp}ey59+57yv%ho|iwK`wV~30(W#SVK<<1
z-i{;HE`}Ef4!cfUR=689^HG~!fPEbtQ67E0xoR8ZKJbndI`_-l=LQ=zmAX0+DYZMN
zi<A6dd;w@|?oyrtu(XSvUwAOWolIGIj)2X<1Yhp&>#)Rz`@tr39mS<HBYflEd>0&O
zqsOoBJM>EFa-82cr#P9p!)7swGzr`$k|gARr6T?0(sORZC(>f+L~TWbYod1Mny{Y1
zrFnYWi>DxIn=a!&7-U5thajLKE%!M?wI^W<ivjkE;^ghhtOuAZ*58?PzU*Ouz-+p?
zubz319e&qxRjxD(3?72Ag8e$0H^M=Gn37WE58?72JId`4Tdz?1iD>B+uUptVyEm7n
zZrg9-=bNM%DW?Vuhd`^nq06}yp%3iAjV>~oQmK?27Y~Gb2`@`eRrgTc-+#DYoE6(^
zBd*>Wzt?vi%5DehVPPq6ZQ*4CYO*PbSl2ax000E9dtN4?7&#_bzbzaaPLkY|j-T@0
z7T+%=XjqltlR)xO633OcNB}y%DW79gHbMSGh0wP&;(5a_P2&ME$^-j6<ci*wdHFdi
zE6#Tp*TCzUPZoUu_bE_1ORmYOBs3U@PY$5kU4s(r#f|%>0E3uyv0FF^_>q5^PDlsy
zqEbOlIL2<KR?A&_Uw9hIq0o6K=@sE{X!|tGhBsrz46`^}!fC1B(mHtnTC8#jyqT^^
z8tJ7P<F$7)LURYhyW<-<l_Y})*ox|WLo7Y}X?@2J9Nc2R4DxN9m5A*=MFt>zs~6Ih
zit6yrmkE_~guceIp2L%jR51h$*x>X2Sd=<$EW3C;d@tU<=uZvPOs|KG<FcEt?VUi)
z>Ihe&s3UM8FH+E(hU#wWenPvg^w0h)&Hba{z{teL@}C<HYBK*Liuq|cKx{<948ll+
zBcT$c+IB)e%)4G};br{Eqg`JXOwm@5byaxibt}t9QF6$tjg)}n7+|v#9bklI1%&05
z^~sW><6{^Oh%)|R)V*?sCGZbI=GQ|Ud)%sxvhNvlB16LNeBYUH9t4E+`s92+dGLOE
zeN7yNvU94UCCyTRZe-hLtT7{>+g^<5qlrKoIG|0sOB$!_o?%Z1yYcD=Ii~&W>mf;r
z68-fXl(KI44SJRqGzQJvc#}vc&C#&73d(c~^;Oi^lKxKR|Ie=8*2(S!4E<acwOBI+
z3w7Bg@Gn}G8y-^z7?QR$(FKu{9}yOEyeV%%!#2&6OgFc{R6Ko3F-nz31=p7C1x_tp
z)Gr2Y&>k~C0eJzEpi-`F>2#RK9$U?9BYEY>!l5mh6i+2(6$Kl63(UKx3!zRvcG3xj
zv%~RS_8>2n8`lm?*t<ezqg~`cf&gspgBj=o7m^f8M2D!DuG)zKxcxuHFs+(L9gOI8
z<4oWvgJ{}DG>R6$p1wT{?1C@O8bF9<`$LIoMSVGL9_@NG(<2cRLH2~de}T)ZaBSbi
znbJ`~sDwbJwyBik0yQnKrNt{F;N?3s+x{vdh)uMw#hMKum4_coUyWkJdcKCrymxJJ
zMiK^06av+-TNg!DNyW)2hDIT1f)$*UvVj>VloBmRm%<!`B*&CO+}%P`Yz|yK8orZZ
zt=t^O013Mm*nN|Q2k9Tjd3tZaXjsdpq#B(A4;Pd88`Wg=x6FNTR|F?92>M}vf+bLE
zJn$L^eOIM^ecxGy4S*(?|N4RwntH@Q>!9U8Bj{&-EpKWkP|}a9y+(S@N9+t1)mdC$
zof&~4t6~qUitC`3FNQrlPtk3cEqWaKg#<TNq;n{JqPJW3aVT$Cp>oL;CFpAa9i`b7
zp*&xmK-_50<xr?Nh6`hqQ)rp*p?`U2(N^yq8okj4Uk|7CyoH_?zH&;^Memx2Rj4PU
zqvMZx0!_2tj=ME$zmZlhZ;#@|$?IY6nEV#8I6g)K%wIk@)z0u_v2ee8rZB(2yt#hQ
zzwcIrPjW;Ch@&20d_;nza(^w{i-r^w0y)u45(ev`2YbJ`-eC$4n8q*>b!b~P&<dB_
z;v%X&pRqRZ)-gt-jCWA9qc0Wo(R@jSymAgfR84bdwsCONrS^%TvpM7(fpQ3*cSke(
zakY%>N^HSq;ANEf=)R*Tdq@Gr=yzbwd73nsr}Gu}ZG2^!@!;^Sme8>vQGFKW+o=2^
zb+Rx1o*RF!D2XQ0G^ciOX-K;7d-hKh+g4z$cNlU?QFPl?bu7(IH1uUScq?<>O4n%q
zL~4cei+;R%Z1?NE0r+bn_*3-u*A{!bxo5%Dc8?M%^O252tTTKBnVplh*T+!~Epv+%
zHGy{qbH>-zbZ2&C1x`;#T*d*BD5a_=zO!b|B%9~ISv_^kuER;VF|WKvN0pjO6M`x^
zwi~Nv%pO2;=X0trN%f_V#icvN1DHNG3GQs%2@J9Ql@ZHIC-k7J{J`=z(*i7-?<BVr
zg$YV%CiEr?&0q9;z$&BTn>~Y#)2g~<GCzeSbz1I5xf0{^4U314InxVrpkn%AqX<}l
zwHHWQJ!>#tv5V<_Ko#2q5A9d~a=1>BPMcldI1iJa0lkdK>n+}9j=<gxqIP}ZoGQav
zpilUCpd09ah0=djOn)){clEuR)c?%I^_-}kQUEgwDI$@AuHt2=DX>t?Vw~nvLamqB
z5H%R5&3m@?_Aui}HZ)BaTF?`5?8h<haE}#w4;H%3yjXuYJYEiMD2fu9DH>Zet&4<5
z%oiAIq>EMBJc!XCwY)ez-uEFnsG>nidyo`x8(ENvg%C4zwRq~WO)tL>DVl10EWE3+
zXz=Zq?W;}Q70oQu3U4k7Te>zKZE4Wov9f!5`j2C#2GFf)Dz(D3cDrYiDU3;yqDI6d
zfbuR9Nu(nrxa;D!;%i2?7Blghiy=n`hOMlx!78+n{NAW_6K&n{ARJ#s$mcmtF-6x>
zVc9sfTu|x2ZNcDl^IaBDm3x+q`8Kh(+R)VT2o`t>zg4WmfzWkdnP0;c*S2#Y+K&cz
zAdF!D@K=gu>0Eu{&#VC*tvTGB&wwG?kIj-oAjN5F=HhlWJALpp!qK8Zjt?-|3JD|E
zq4%n<Ej}idicBinRzE<cKWMg~A9h!|*L`QQrH56p?rzTX(_I6W<^l)Z3w04*ZsLCR
zo$B~&37oI?ejZ`e_zQI|!-<_#|2iB`oy~!LmWa}cEP&aIb=88(EJrf<kDdhRK*z?j
z<Xo8U2pIIqCoQHDDU3&dL|0LzY9!L{Aew^8iuO1TSCRk-O~9i?cC(e12Aev35qS0y
zSL^~fxWA5XJGlV6yMRqf(pxy$BW=q(_&0a;r0#gA2-;NryI26@Bw(A$Pxd*l((=b@
zyHrC=$%q?sO`*eFMo5XwH=V>U)C#OJ(K+om(;dx8lnDr(1PvF<$SL$;FzLa`1zFqC
zRK6nG>A*UsN<9<(wr~P{*Eohn)p+5Hi=S0XxyLoh_1gP!OnvFHohvGC$hPI`aHo!v
zf4m8H5ZGlf<U*o(6+OB#ILmu<VH`?tnyXnr5NJcW%+dcGn6XXvkuN*amx3BX+EdHh
zQ$msTPju_!7f+IH<X}iK(1Gzxj;%q_c~#I`Cj!d%lOEt)LXD<G0Q~F{B*Fn^jVEo|
zJTo@Jt~H&YnBQ9g5gLgicibQVkyZ>cO8|io5=OlV<`vQu?N!`Her4s5GiGAw`5XE7
zgiEJM23FNh$-1fdrE?o3D1ns2FHo3?D0qH44MiaRPN2*OYRb9?EXy)!zr{vPY&gAB
zX1*kkqmwF2DvwH$ca|T$VOo8K2|qXGPVVlOoHkYvhMZAPFbV8tIIE&)hvXhPd34fS
zDtqY=QnKlGBJZbb^T17Gfi2Xr|FsUP)~%xmh3Em>gVB!;@il2nNl&C9m{v?^ZnsWq
z#I3etk}|{PPJzrMr;!t55@3`r&fcMm=k{xgk%{raJDoq^`H?$BF6ugGgtz5x2GSFN
zf?7-rN16Q&{0_)sBQP*h1IUjlew5}~dv#-RPmqSL8>BITD%scqY13y#-z88I_1sV#
zC4s=BmhJGf*9W3JL=zA1+wkJrU_ZfW<q#@!Hk%KLQ&W1){SZE~c%R#pJ|@;TI0axu
zXs#gH_Scka3zJ*jl2XpP86+N{%nV8AXOG`8>bSL&!w#7~J5aDPGT}EpVn2Zr;>c4#
z=xo`;fJ-ln#2fE*1cKX)P^&W&cBdDPhi$U9N*xN$C$)w^iglI7{Wv@sJY*cViK-rN
z=$Rt_PRgH;oejLl9_$PJgsww3oS)?S4eI@h;%x@B34VTuFv)wQmvK*ye%q6_tr;Ba
zm=znXrEh+o4m%adoAZ_dM_FszQPI%!0MPXz3Qvk_E8yXBu9oe*I&lmBe&zZHSuGZ#
zKOZXm_QB&`pH)Xn$0p;4#0s^;CpHel&zo)IZql#RU?r!6-NMF`dDd>Z;%@JUwTjw~
zjbv_rjFUsRX8^_0Mfae$;sLwe9TEC0Yhr~QHP6ai2qS2~cXVwLAkoZZG{V}XB`}DY
z`#H>>M%vpmA{x0l1rxIF2X<?yCHQM({Zb?6L$2hTQbrq>9X30b`yjvLkMi&GdxRd9
zplRaxAFf6T)l1OP)cG1;>zL@jDoxD)n5X{5!T$fQ!2d&yu~=EhZh-;C`=+*TP>Kq1
zX2%up2|$z0|0RyV4u-JRL9>vgCc!z%Hx{>(zav_Oi2iwfnuGUwwFkl)5Qr^u<!FCD
zLqWbz5CL!;OP-XV0LlKC_#m&UFI9T)Xyxw}5b5I3FHt0lFkwPqvQm(KLC82{NEJdN
z^Now~3|koaUVnL*Lkyt;V{Nwge%2bTW1##?Ca_wgFNZ{8%oSTiF}0HAP}7IAqS4Cm
zBr1ic?u+k8hEgcW@sp;O42Ob~$08(Zo^$Xc^7(aEQ_)UV?-Mu~BLlzwo4{w&ga?RK
zSJy26v9%1(EtMK?$tC^CF7*J8{@}i>uYo49uuX$a{2cCg@S%~NJ<~R>_d4nyZx3wI
z&R5pLfkW=!cIVqLo#`-9v-c)HT3~7VLTzt4;{D@=%1X9Mc#DUxgC$EK#xjg_qwRK^
z4~w17X5Ee97(RL@P)vn1+~hCV)eGHLM8it2+@MQYHt@v1%PJ)r$0)B|aBjry3Tqv*
zy@REQqovqBj+#A|Xx2hsSJ2YM>c?5H4zh%z5!o@LEJSgV)9GW3P=n;rB+ekm)B*s&
z<M|W}1Q9}-^71YH?b+xO>@#3nC6lhr7g{{)v#2HEgu=(s4O<v=rfE1j5n=((OC=v@
zg!4rnJcH~y75t43HW%3Qw@mdyz@^H@qIN0l8$++duE)<`{fr-_FyX&369h>j)rWiZ
z43pFAZoHMl8t`2=hond9+(PN9VO=*wS`Fc8gYx?6F5z+Z>K2>{=<ysDGfQJG?3d$W
z9aQLi03CN675=fj`6o~J*Z=yy{fFOFWttKC4+BETUU%>Qu<PirZj#iA3ia|t{j9-!
z`D8y%^Dn*@VUW_yoM$drw%NMshvyo@r6+$s7N*2N_h4BFS`CJ2;&6ZV{_^eE+D!__
z4!n?jju9ONwC)#Mg$U3!4EdP&A@pV+=}>N7XzpWYIS!p|=x?6cr)B=WzCkIn$xM>G
z<2?=|DbdqdYr{uC$4pxBe>^73zl?VN7qdk0A0CrgUR;eq-u{HCPXHLehyrx)U-$Tz
zoSBX3KYQp^t8dtD{>*7UtDROeAvt&?3WM4Bk(dXd(E>GTp_tEuYZP<~H_xl0*KXC`
z+?FICHXi`^16sLfV#3X25u%H+A;$1(?=C%GJsIBX&W9a_B_xS$T!fh*q9YYE#kQzC
zP?F*3e06_%Xf}AriDSj1IE)495FpA?My;KF%Ci(})Gbdi4}wB2=<1L~?YetOK!kS|
zoLv~sPEDBzD&c&?ab~eJUB6!sL6;^riW8;JO_mLcOBN!E4R&OkQXua>;lXrms0qME
zgWCx_mQ3w0jOKTe5ZT&w81B5cSHEK=y8d1D8qv2d%|bw_YqNd^*Qvbr=Uj2vJaaVD
zZk;w|HscVZF`L%7vbiX!8j(~f5y@+A=vYl`F0xd47!MBF<wiotj7nb(YC{M|EH$>c
zfl|(eH@l5947aCBnV`CgN)g)F;-Zh(e&yVGLEdJVj60O7W*+1|hcm)}Buj;k??;H`
zV;z5$51HMxbZKJD4GdIRjdP~<7Ov;%)`aaj;;uJbN;%N|1)SILsJ;l8$aRDa^d+Am
zx+TLh*6ZCC##_z*A!fH5fB(|C?eZJLr5H`Z#Lb9DG!3j&DG8CC{f~SO<jD(mAz>)L
zUgr%C$Sk2z1cf(b!RQB>!8|1EEV~5Ssp%>ZZV+ZtesOoXQE8~a>#ucRz8>nQUxT?u
zXm)N4OT^lfKrxWFKyE%W%1Xz|1FbaY>jAijGq*Ftx&DJXdUG%RI&-xkJns2Ctg`m`
z>`mQIrz{Jhd$Yak*0v9I*tH1m^^B|u;Q7-;U|C{at;ro*?0V7^q>;6_9Wt`e*G^c%
z#|E_tl?p3?*uk4roHprTWZJFS$UbvZ?k7v1nn}8jTtO<nx+ORTTFqj;v;J$RN>&t&
z+(H4KIf1A$TWI`PRvD}hQgPD$nKG;*kS1Y_to57%)DUDKMmXP#Y1Qza!y|ZiIj@Z4
zN*2sX@_9b-Jc6I(8-B(aM+O<Nv!Aa)=&#n9G-ARfwwZRx;YH!i4&k9Nyb2t842<x`
z`UhscMzs~Gkwe^VhNv=QGY6lm@DoGDjj*Ui4ONMpi+puWyp-edeg!9T6?fdsDk#4V
z3%u`Q-=Y|+y*qEU9k!N#`2n?izHa}4>;M$>$q_+-K+~^96XT>_ONn+cjp!H(pq$h=
zhZ{`mk!E7|3L80A;0zG~QlMyaojZgiV0#?>R9%kmP%`5dXE0RvWGmCcjR73l(^~Y#
zX7ADfy_a_mOr-Ir%j>RE3;Dg^x0)Uqcd~@JvU@BL408_{?!}&3ip)fqKu~u)pCD@?
zF;jZP3UH*XcI+40?b3MOX%80$*QlZ3Ph3RyonDGinlX`wRg`XN&ePbIh*A?BM3>A5
zR@$y4?ALJA&|qV$VkY6Uw|d?K#?unjzmm*&;iXhDpsxfr@ltvup_LL}ic&0J0%8I`
zj&j&&Iqqr=w)84|bF;^5e=@>pFFb66u`m^(X&U3XJeQ*s(@}%&WIJWk=P0j_REaCv
zMeMcoJQU5T)Ew4cZsc3(PqJu|XY)hj;*Py&fqBL_U7oU3XA5<(_a4d4P)8H}NTTVX
zbLN>ATW`S0PHgcTKoJvWI4o~~f6|70SpPsn^BOhd=3_pbnJH9qCwu5jtnZ+n+?t9O
z5chB7@*N0``BRH|@wPqn?M7p1gR#s(b9pr&0Vilk1&CZB{JRj#WU(ea+c;$H?40v#
z2gcpgUOIJzV@G}Ou!r!wMe%WM5s6a!t3+)f`8)%yr6}IvQCjtf*=t<UbWva+HCSQI
zPw0VXL(UQyI1y}iGnwvw<z8R4`TXvvw}!o3vsJl5x1$dT=sJlGEt@>dnppI?*VMkX
zx7*<z_cq1S@@fci+ra<3k@N9~<Z{o|ta7Dw<NBxzU^xS6jGW2biE@|hMc}uH;zh3^
zF6sA18ot#PZf@ZGnh+dw<_P2nDpz7#i=AmRi}&ZP`N@KUo0&hY&$c`*+mQ-Y{4i%?
zyyJAMm^`pkPa|sCm%ZPh!_7nMSoxiyWLb#rrtJvW{0&4V`wNt@w9yJ5r1L&o%$<{+
zGR{c>-*qS%%h@9&&Ld=g(n+J8x}yNcnxh_2`cCOcsgG-hd(jLlg5>q)Oo$w#E{7fs
za5tA~9z}N^Q>V&3k;K=#AEq?y#~i2*|Nk)dj!mKjN|$ciwr$(C&E2+bTf1%Bwr%aU
zZQJgiz7uyoOuQ#zPW^<6sLIThxt_J}dORkQfli_#afP_Zmf{JB%4P`LRm!dN9zn?R
z^<X&Y4&|eVjgxZp%wGCdkt&}tj!8uHF|o@qt{AfrW-6E3rtc3Vs85vdO(5z0gy7Bi
zGI}-q6L<Xi^B)S~|AmA9KU6#B|3jbpe;=7;tN$Ng58<Pvm`x~(Xgtm=pvBy8vLSpC
z-t#P65D_&Sq9zl`nw;))>#w!Pc>LF>cN-Lg@7DEk_u9z}n*s=HOQ$<SPgjd+6hOum
zk)Z#La7%*VAY@Jx_(t@K)sQA)%=9ZipY(QT|05C{q<@Hr8mx@v@O(HMJaKEayRm<K
zyIz>uS_&J_eRxS-VZ{G~mZU(pr*EbK*?AJ^sSYG&4uS-m_7FfhV2Jx~rH+*r-8@zi
zlY$}5-YV|xiinUY#R!!v^9-2}1}^@u$oQP#PMLlpW^6Mz5(uj8q^K^trmB1z%mWy?
z0Y2{uqBO~#Qf%~(%Qik4F;U2OzVKAVIvR8qO(j}L6;-d|gzWNuNw<Z1_BJz=oo>$%
zE54-ejU%|#O>;;#RKe~AoZr9P7dBf|ItW`F?RbI+Q^6{dUt#&<q;m7vf2go3P8FKH
zSg3Iu<&BG8OPS80`pVTzbe)9)>t}1F#O47>b>mzDkSO}^Z&XNpiU_Ia<+hJ*Du9l`
zm}MiUAG)>@lDPO@`u*-GBbFqBv>;IoccaafevrK&slndBE{9d+yT?Eo+@55=*KJ%s
z9JX>gnybLzqRyP7)m=tTbRrc|8V;^6M5D#Ry?0jpzWvsngrB8P_#IUrqX=%+eOc!?
zIJ#W($eb^$LH@X!Za$Y>1xG|U7=5+}2$2qf)+;xtRDBPZMNmtsOUHF%%vnG)2bm<v
z>^gTP3VB|VY4!L$C7jJ6=3hFv4_db+qn=Z+Yq6s4cp&;At81CRK%QU&Q=>bIuW@|u
z<u+DO3o(z+)F1#N?vXZ@6UQnEx8e4ATCpH?(gu{IN?6m<HVqV@?-$Vy5`)fv@!{bm
zu>02AuMzG$P%Sf-v4J~;vNB~s^E=4F7v7X}Y&ITy#t6R*@xmS=N(@3GU$l%YKz+`b
zv*2#x1(7U5w%M;m50bmDm8OC623fn_fz)ZTPTZCHxK%x^J!~BK(p$QjWh{}DS(h{S
z<3Dh^&DlIm@p5ACt7KMuyFD@J`b9Fi{-t<CLW4KiN5T7Nb*NEE7VyY=J9^*<cC5mn
z;G}O%Z)P}t9Co4DQ#tA!a>lyRiDgGJgwTnF6<;Uc?oszc8h_fpVE*GQ6@g%NeJ5`V
zBv<GOAQAIiT1R@y=TUGF2(bbbwFiJ&%;6_P;O^?N+>vnrdvI1RTGOn>JQ1mgpJL{F
zv8@Y6W^)||bh#!WB`6)?0&&~}Cl>{X)Hf_T79IitFu1}L37tYa0~i0k8idP@qS`FF
z<`NVF3UbtUZ5OYHHjrq*Da(S=YsaQ2ryiw=Yvuc)aD`TDlW!F79bP6B8?-8WupK`-
zjmI;-r!Um4{%}Q@7m<zxhc%TEQa{of)gx~VR>E&izAs%&5OZ2vDGPE2(zSxiU@>RT
zYTq6eMEYXyP59Yc>DBvEqz37dbN&}r9XCLILtWZ%M&~uJ9;0+V^1!D?k+dP#TWK7#
zsH0Q%1vm3%lbvFpdZlrvz^EcWTIO6(AofRC^a+>z*;Zks#0QtAv*UQ$?Tn(MNN!*t
zDRxS6>N5Hj4B?sE&tN2V#Y?&5iFu;1VaQ^C4!9W5bJ<SH|H5n5#26dHyeZud%+<mM
zM<>@~3+WYL`%>a&^7H&4^J1eilfE?J)}AbqWseJw+H7nL=$-D<7JJ160{?zj>*~IP
zZ50t%Lt#u^SN9P9ti{}iX%HTRt1MNJey1P5sCGJ>R%!hubm|p<ab?be&Z)$lpIn;h
ztGtv+KM5j3|Ge<?dK`zr>&cLf*+k76{q0|a+Qa$HWLSmgos;}Gci8qc{WswKMlr#E
z`fL9up_`NAzs{m9){wS4WJBut1;CN|!?bY{kxCgpv)s9(Y3SFj)d%1t5X6zTWMVWg
z&J8k8*X|HiQYc=zHl*`L)KZ~cyd@``GD;Qr^+-<l@Nf2dd<aH_qs!Q*i@tnNOchqC
z$W#z@*cId&(XB1jzWn@=*!JDk7OTrBZ*m6tYogSb!_6nB%04}6#LA=9;B?J{C?jzs
z@8{Pn?SbUi_2%x>jLa1A!z*9vF>u#eW0@<?Oq5F#jiTH>hKrrDik%oPUT)mJuSn*U
zl1lJ^xVnwc!XP*Kdc3+BATSm{E~)|)$>Ty3o<p8>G>#dBUh974nqzj`VlC~1RpXBN
z5-aQ?!l<u1^u7qc5Dwkv$8=z9nQ`+02<{`E@DXBklQws*fh%8)S0#gu>p#^HRV?}!
zXfbTUImLzU;9C-&qhfs-?mh&%#uXMW#-~sVCx;3_N8S^I-R%DLrTv_Y8S^Ur=FW<X
z<MI%P+_ibqV0_(wXmczbD#ahzot54aA$>tP!gD@aN3pNX_C)O9S(qQ!{?O)NI%)wm
zR#<Fe++;=;oCvi2UwsD4RVVn`(}+r}jvb}qs(?a`M*I;v2<XZs_a~7b1`Sm+vdRyu
z-1IgG^>LQ_POHt&uygFtW;~1Mt1q%UqP(VN%3MsUjDSizz`H0!JMYtj92tUQM}d}W
zzCQ<_B}_yz1z=5J@$Oiyy=<D&_FqmJf+VFeq?Eq%lY^<P>Z7SUE=0e}eq#Hh60CK+
zFgCPIsNAp<Tkxply^6Ry)|?7^=pr1mn@`s*Ow^ttc0~12-|G^4<sltAF|Jt8k%xST
z?oZKasXbu4rACj3n4lZSC-n-*{aL5bA_C6g{jJ5W6?rz9EXR3mnwC2012Iz9);W2w
zuBnbyA>mlpp;|-%1OiwJuw*Vb=Y3XKiT*YvY*-q&fgV-N&rdW81QW_}#qI9VzDC~!
z)ZJ2q09?9mFN_b@916poTFW|cH#0yV7>G*$I*6weHgz3sr2vXF8V)Eov?*Qm=K{M0
zu#tT~?Ro2`Hpnrto67cLCGy@vYeHmxrTflF#*FVbn^rfsUfzFKC%u@v)ORZB&zTWd
zQ+BQJU;R)AV)g*w<hhjMIxpsFy*?YIJ8(;f^bILfP$|cvAg_VwIZ5Oy?<RPQH@Hc8
z+N(2;j|(A8wHV&&8bP4`Rmyv2pmet}<TL@X9hKv5)0LB8%@S$;y$ykhA79S1%VvbS
zj2FC^vZ4mCd+Lf&3IIWiDSJOLhA3uCv@MI>L59!W3a?Yah8R{_65_|^)PI)d?iUYQ
z`A**-sOqpkfu8W=aOP#fhlJ8%3BV?s_n{P|D*oQOMxA06e5@jh)m?@<Ks5*%++aOn
zT~jS*S}O*J3ViMo3@*We1PVe3diM}c4*YGc`a@$?o9}F>M-W6Fo<vG&;@jP7x_RNr
zB>zi5C8+OP{mf4v?$!LXP0VmlVfwA!3gC|K{J5Mb36sGlhSaC~n-mx+^DMq1s%pq$
zR#4|6n{@^ds43;nlW}?+#RNr@SBwj+C1eU<m-1glgo<410cY*odSVKIb+{~pf3<D=
zUxX`Tr{$4>JljFF(rA<1{xM8a-(P_24h9Dy_iM(L5q<PiRgmp6s-isPz+}n5kL;%}
z?xk;5*x+>|yaPF(R0PBQjWdN%0oGEmkV5QiN93D=`!kQxXJl7_O0Y2w)?K246k%Y+
zcaZ|u!4P@_e3gD(ZYgWjt&+4Y-AabAt7HNvppf?|z|iY%oHGz$=rxNr``~0UVNBU_
zV)>#gfh$MxI$a1n&rm^u8Q|wXeL*nzi-USknKp3US8%jiKuDzrmt7AWUpH>w1VeKM
z2wZ&`3^5*t7C0yn!F+iT_%|<|EU+3H<f-X4<i}l&y`{k}Y4W){k6ZOjT{~qI4uJ{;
zT#E3TAfyyYDTieUWjUhBJm@5Gi&xcatt*#@k*RK#HA8H(xKx(z*9CQ-C9N1Z@QS9h
z?A`-EGqilH0a0IEpMPt8ylzCeQ=~$^{&Aj#pUL7(dL9?boL;Wge6no&guk8dZeG49
z#EUA_Me~3Q9W`(XzQ91)?@XwX0gGYel|Ib8ie5qB+&h!A>U$)0G`JS3kg!&Q0o(C{
zccOJjgMfj>Q+`;|=;$)0CWwPwp;p>b!P?xA{mrGgc9Gg3aWrkl1)jxgEf%^m6pxmA
z%xny*vSe1k?hNT{HbcKjGoDVtn{my{#FY~mWn^PMlFIQB=`htz@`Jw?l1B}Fypaq)
zP0WNiLibMwJVVs1OHA3H-f2_T=iA2!lT2p9+&wDgjvSnpHW5FS<ho?fuM8n%{r5^*
zUIX7C&~`5$9%i$boc+D%#n>YdVU|dJ>smV?#MnbHS${+bB7-ySeK#|-9`9ceJpMZ&
z(g-Nb``((k`T=k)5W>D9<#G~2*LfpEj<e6{znez5syLgUR!M>1qx#O#$wu@L)3C{B
zvmw?R1@?dK;TE7=16Dj@`=%SA7JCgu+-Vx68E3IlV7G8D=fCDOv>V23oE0dKz0`E$
z2)nP0{C6fpn_9~v+^w$1Bh@)OT<mep<DIIOS)~*+E(Hpob{6p!h`&CCqRO5f1vRD{
zN?s76yyP?K`zkn--cHZY-&1~oig|gF|Dh{j`wwdjMvnj9y4<X;Y4<B?{(o3lX+WF3
zVo9mRIcN#oQdz}Lfi4?Gk`O{^?OJ_YXvNIJ-d{58>AH1RlW09WHx^vj4vZOJ!e-}7
z+sk0$4ZHkwE;M0RQgM7#*VHmJk$ec^LH<2k$jb!!U&xj}mEj*dKkxe|X7n|F8r8-7
z-^ZhqOSzf3)4AY0_eg2@psqc;fs*#uWINV5x;vkqE>Dj~_~xD3)JQac6Z&8G(HXjY
zLqiQ18j&Uci6p^=N?P0G*h38<M+IrY@$c<M21hVFv9Ye2Zb^~!fQ=bkp4fEjLH^8M
z0HJEJuP;AS-+$n)xoAGr*{;`2SKQO<WbNIyR71Uoe^pT`SUsz*B@T|wxM{9oBsKEb
z_4{3-n>fi+H=(ry<4gnmy0>24{KXMMe5cy2Cwi>sj4pS&M}KzIgzWWh-m9n;3%vqW
z`*r(6Uu)a<6Qluu2FmxxAC`uvNTo_<-sc}pVseYq@0~G>xG)Crjq2z?rsti}gwnud
zBqIphNLGmuzb7|zO!FxvF@belPqoHiY-N($Uui@LzoLU6*p3xeG+*${VklraF?b9_
zmfc)-Tupi_$M^LCc=}}yVN8LB$B`$Tt)hbn|9l|+NoB1lod9&nm)%Zwqa$Sq-3#fN
z<FtieCjLxQ(sjFuzLzeWQMa9T+33`8I7X8iERUq~vPyaSoX}jQ;2*p?S{~T^LtqC{
zHLjhmjumNx=9^Rp(lQldm)8`#{In+I@XeE)gFvGywa#|9{RfN&;xG#{CJuT!fCA1$
zR1{vsz%3#}50JM`qj+_R$CL^(tR_^)Hq^%Lcyjn4PLLaBq|S_$^fZkc1D{>Ho&r^u
zLYa(~bFHj=!0@Ml&Q@Rav#b$(#Re7jIo*TcA5Tsqwg+0gF^`(`iW}ScC}M<)z}TE)
zPIYg?d}s<@ql=?SnJrmhVKTqO)^Ep1B5l%^<#Q^1dj8Vpy{50_jU<6kN)r*&-~}Fi
zN9gZ@05<IkI|<(BZ3`WVgiUj!<qr!UbR*&aLGM|WiQ^Xb;g*4tRtD5t^0wr)2qPlO
zdJ=pz?a0sk3U932Y&WGI`JN4hZ*;7PrMwr3QBXvD*SW+MNHqif`9oF<-3ff5qjhl;
z=c*6HAe%r#ra?m?m@^&Ac7^Yp&9mh~4=?-B%eK{Hb<<h(T~z`-5IoFgAiTh<Bqdd^
zF)~S+t1}PeN|t=)_^dK-0aUj|m}_OEofuGIW6fDqf1`nc;=<j~jm!6FPm~umY&7eA
zG4150=U)p$8%4I@8tK^q0z>{4<sU>~n2!YE==+92ur@hLbq5bSF-;5;BmNSrj$T%e
zqjzSD8pelvlc`Gei6%ZiYQ#e%oU&RpgXqI_W<<jjU~#4>K$}n2)W{R)Xf0HEbIj%B
zEd%N=3&P0Ny%K%h;CuScef|XVGh!TYnZSDUr5Q6l7A#ZSkcEQ#{iDqm=JxX!CS}Xd
zcXMqFK`XCXtKab<w9KE84ZmX~AKu9pT{<l|)G_i)=+#e;5>!P8ASfGlQ(SPz<SnS2
zsc~V*dF&)buJSw$JEb_@esymbgq7h-E*Aj&nT_0dhL76I3KI|$a>BrwDy&u%K?v(=
zhcWagL$(c8&IGKLJP)x0mUmPtuE+PJ;y8*fWkYuXAZ5L;ZXW)wf^+*-Fr+-Crsiy!
zTUT+MUan$8NIka2EMp*&T-$K*=&Poe^{vK+deGXIZ?2eaY_`~|<2JN0%%AgePl%r<
zlq`6*Ky=ycfvnhUW`k6cSBgOa*Aj9!_ZE5+h2Z+_lGwAJfW{<KcxeHc^zKXwJP4h=
z7x&?B5+t;er&;tS2rQ$+XO-Tyi{B8ryBS8ThPLmH>z7}<Xch^~<{7;WZf=Ksmw8La
zf~*|u-Zqk2-|3d>Alj<BlbH%ztI=7soRKrueD%oE0|V=%8%1}>3aS*T^HA&az#VI}
zv5wE`4C_~obr0&J!PCM}ly2kmhOwP<KNy7)6Dl(P=e2f1V~dZ#IG;s8K9Ni44#DU+
z3As0!`16+Z<Y|`;6f+!{^jDo!ZuY;4=qt$9Bl7;>yC!pt=|fBjRBxOp34BnaP5@8C
z_$+E0cFeJi1z}x@)~8|SuaT6VC4QflKVfRsJp{_b;FK<0-0mJuPF`+hc&52NYHH$z
zC!r{>DXn$O(Ap|9OW)6~&2pFTR@f<8FblY=7Zzhmfu$HKBKpKSlW8!elo!gV^dO)l
zh4yiVpb6Q8dFA7*_G9Hb|L(PK^1*jOZ3VfM2~yIC4pNN7n?-rbs*T9SWzG`E+!_9q
z20>7~`ZK|Zb@ltB+xvjnF=cgo+U=a6S%*&-7f?>MeLc6ukX04Zr1K0&S?nqsFK-cu
zHHAO$NLp&!XuqIL>DcCc!fxszu>H^K%JzR%*Z(p&lZoNizVyGJ|F@Dl^=Id|`Xc%6
z^acv(UE`$?h4F{-V1Uk}jXc?dqewz;>L7qj)LfP|ld8`j&woC>M0dfuR{!RTtxMv=
zC{_}3?);0}fw8!dz%<EL-Y&yv*sn5-nkWf3Q~F&7cplLl@KT`P8drKhx^tn<k~Sz3
zhKU~fv%ND?hsqe&CIompQWXck6jc#o5>Q)-5{+-p%a%ldq@x(y#<HuiVK>&z4w%l-
zA!12lRgLN2r!^40;pNlD2S}x(a7N0nQQw2*Hi!;Rl8*%w<VF$cYYp-21FwcVC=qx=
znd`%hXe2TJGk~e%4hmI`$)mznnil3}Q*6h`Zs@xq3I<IuN<~j09HiOGb{r>TLBhi9
zrU_;)siR>7H-={RKo$*6ZbdMbo<Jruj+6%(UK==^o(mkOxB-l_QQ#9W0|S`R_00b$
z<2i-~Lo*^mHP0f-BUB6V3S^$wXVFJ%!Hhm+Y*6n9hC?<M>houLM`ywy3a`fug~%%6
z_LU@2*3Uwj6M!IunBD5n7JC#l7bPa<ZrcRwbAUj*XJ`|_?9aVKpo6n)5NF{4f%I}8
zBQ|u0kuV=UMF>jO>l(@%3!mhrzr=t=XbH!E)y3Ku72F$Ey+xfI0*f(;{k@<*VTgg#
z10fFDmj#1m0093Kb&ms8!U7*UA%O;tpOF*-DkKL!ae|;C^qvc+Y7h6Qb5x4E9aWNy
znr&oIlVV5^)h|WyaByv*=uxI8CSy2nr~3ZN{UP#1Xf8q?C|T3i>Hg#W_2lE2&O&6R
zciCkF=HgoN*C~>tKD@ntlrwJ4=EjM-|EQH-bKh!%Of&AT1$u<q_jG+J(o%t5^={DZ
zq~f*SlfNuh`RjE<HU$32q?w{;xJi}y@pO9LN0F!Y-F2b5!grH;B==jj`r<zGx!aR6
z!+l6(_Z4a(3HM!yF4nQr{{U{6;hI6YaLDIixQaP%$0(VX`%=?<hN?4VwRujvl0mj=
zz_doSuK6=x6%PZrk#U%kj2~N>18qSOFDsJr4YEW8en5rCDg#Zp8;4WQ8SyW6?znYs
z>vw12#HyX5_k!z11@;L;?cZMX9rT@jPzJpSHR?Wt`QYe(n3dIWX0r|b0qHD)$HT`{
z{&#=<mZ#{48yg!P9{1NQ%Qi~f6+a(8z1}{g`IDwDRDZ5rJQ{s#{knC3P$&K^kh|OT
zkq?Q1=UclfKs04Z1cAn=pItF*KKfDry;xFeTCj1`tlCi3b6ye!#jdS5a#^#UsNA1Y
z?R+m@u?xGXtfi|gIiJY?L+`hhG^GIVvNSWgAIK&YaoBW=RSbeHy^h=@dxlsyw&A#-
zuG8DKxp-00X7Q1Ke%_+B?Hpd`kf&u#)S;WAZlg@gAL2-vyK&r5ojI93{m@T5Gmo}m
zA+ajR3>4K=#C2;`^6618-`iE0;^WiZP9<ut$$qfPtzm*JdCj16ISxSGKbQhHp*dhF
zZ0Rvm(llA{qErnlBu{RuV9_Bs+@5K>FTUk`guLvxX?t{H`|t0Sq<drL?vyLtwoI&4
zmVBUF2Hz3!LMabnp0yki))2RvuM%J5v_4Kj6|yxVj8c_3ewejbFHZdNk-PgA9&&9b
zwG00dv{1nr%1u3_TleV5)MHLkW<jKNf2%y#c;=kN>#SEDi{oR|6^tk`MaG-EcniC*
zkB(oNZtePUK@ShDo&B^qw0OS#veE^2aC#y+d346tUa)dc<u;s&`o*NDd#6X?MT<tV
z&k}ql>uJNQc)R7!qKWcnmraxkv!0*I5my=q8BfESg4n&i*0#l=ue@NIj*GJEkqhbO
zYyPK1>c^xc91R?8sZaWc#cfU2;QWhtR9M$Iw;Pbwv~GY$2hf`!wLQ8D*Fc;Fe4Cr3
zbhMuYTnM)#tsBu!j&@o@_HAVr{ZvzGF~6|EgmG!w{ie?9c0(sl^%kGUhfJx1)rgRz
zl7R?rk@w&;&51%%A{?Q2`qV)_wz~))LRMCTCHPKfl(%4vTK7T>o<h|3X6qC&O2Kpi
zlIUfeC?K#rmr6;Lq1GcanNR$_hGhi}YOx?Fr`lEUK2#B)j<9#ax%=rvNw0b5#kcaI
z!0~O4*fED=8}SK`#mrtF4jP|zd=CKkEDJv|yIU*qPw>)^b+Gbx;lXQcb~XV|;12~a
z2|a&>TnPa>E%}n-B^kKKrFji*R;DU>E?zC2tm4ob#cLh=i*a2+A`a+Vn1>Uc20tfp
zs#k8h{#O9=JvS7dQgmrW2zYr*qN25tq3y8%o28@STM^|eqY=PCOfs(M4@&xANh-0}
zo5vHS)Lq!*i%&X#UJ|#XoF(-=_|aI9>027d7Jq}s;S*K~A<n-`j`7lANGb33H#wuN
z8wA`{rD8EyHoq7G%$ZvC$%f9$n<42}<xM?@G$j*`1ge7}&Ald38)7BZA?N!)H4)?B
zQ*+cLi)<gI5=fIJUGLSk{OP~wFLVBXC0~VTyM;I@PFZAt@uptxS`eW&gl*k!Ila+l
zvJ!l}%uxI&@*`Gefu9?p&nY>ZCwhW28=WXgP?J4fJ?$0mN9Q)0?5*t57d4+dl<XFV
z&I!QrJ9jpbgEOJE9c^VTFJi=PCVGBl&U)GIk}chYk7XB!EgS&FpD%FYQ}Rx0obbt2
zbMCKP)_P#u5rUJTlZ2e$YVpQ*g_}QsI4FyW|FfH5XJ_~i=2jM_|C*=Roch1_u6;WX
z>ZdkrYK@-={7tPi<}0w$YKznUM6^IlB{l{r#uAQ=*R?zG;`6STSpELp8sdMuKDKag
zC>E@f=Ppw#bIsIicaDjuCGy)QejQb1GRJ>swM`pEpBb1;KRuo|E?R$yimD-moXB|t
zwe22mU%t=BgYE;QxQvqaVR=-{b;6LX!u>HsS83BVHzhBvtt<F^Im-0V15GiJenP{o
zzb}>I%Yom^k6-xh{rdgIp<f)ov{x>>f*MO3I9ml6KsCgbi##hxOUyqnf)eE%kIM?(
zmAeWbe>5>mnG1;oUp_67LZsh6d2{kle^b8&mYTiLRjlu!?%|=t{d(JPj8E3#Hj|`_
z!lXW-0Pt-v<3O}n#j%6Y<W+-}I%H&o@wUICpx|O~HdDk{Fqkm{NF)wV{TUAuBWbU&
zah$oLPxICUTM+R5NWZbtc9QI0<bzTfmf?upu5cFJQQd@5&unL~Nj8C}DMrd8o{2=8
ze&AB)w?{6L_<kB<BzKLp6Ndl<N-NPgM~)vxt52C_4sxY1rE<3e$$W2Jnf+ZBhavN$
zA_%8MCq70b<!Z@>GT_*I9yuxv9(kdo%X<&3WmgLP)f0r0AQT5=1(}9o<-X~CL37on
z6r!{|^Q0yuKzQHRqrnjO&Aqu(*Igq<mA^wuD%Xp-kb1YM7b`<7c{-)TUekYh!n3%%
z<;y@y=?PKd6%u@ycZoz5N+9N?A?cLXs1k^bhzQkL{$`9~P!<18E#lZ2Rmi~;oIy3D
zphrB8P(XIU5}rabz|#sh4Dp4`j|zS>fYF?2kPF^N%I7<69u}+>rfKUgpwNuUPudsN
zRQsU0SfIH`Vmg9bf0@abk`*Yf>tuopt^rdF=B7u2O79I;C>7d%@yo&2g#uJ4u90cy
zp<L%ls*rS3sZpMzdMGGhWz+~gOqCYWhEmH3G(qZF2_!X8l1$@yHY&!w6djmnHGc0R
ztXzzXT#Q?A;FQSVppflCy{BNzwgv4!J<>Axv%C>~<&!2R-_l4y+i_nm=&+(b1vTu#
z3lCjs5UtdR?ddq(-D(p?ZOsX9)oP;tYrWp0#_3I$!|%Oh^V=)4M*BF+uVUk<uvxny
zLNjKi(4o8X_SeD>>wX~`TjCe|BNqJCVp`V0?8T=ji3`cI#%YGMQ7$EP#O$jS{JUxG
zW#0{*?yEGr;OD8C+#6f_PLH<X=D24HF3L?s>}rg}9m<Do|2yN`Sspm?)=`yVMhP|g
z&P8W|{pp>tfm~2(iCpCiNpEo0T4%TL%*aT~x0f_`8{qS}#TVd}j;6njpli><?{wWB
zYh1U}n#o}9Ht5Qpr_G9R(;@P+KT(1IUgzxbD4EUJhOujf70tkMA0ceT`R9ot9>8;h
z?($f5_FNbRjE`0hqhcN3oBXA-gRT<_$*E{!I1#!3ezK)*gi9s)FM&B;U$<}B@sDh?
z+rf{*WqRVzk?E4F!UHQD`D+~m@>DyYe**XUSBDQ2r4w};rl#JXv=-q3&WRgayW8*q
z!2Lc!<`l@+m;e?gxSc_$80%;=u%&*sK{3_XnU*_b1$d(p@hfDw7Cbmz{Lqmhl))}~
z26Ff8wrEa14BGwUihdmnM!ZlvaImYED=?gy4Y*P=^6Rpwxj=i$*<{}hHt1dS&DZ-o
zc{5>qSWC9fj{fmw>PMby)g_#2*1|J|(DKZ)+%____{>ic${h`$A59Yctz<WjkLJxw
z>$>aA4CMEg7PC5NQ7A<M*1+~?ndpf$X!bfkyL{4(5bSZq<{`}@kbdkTx5(w^DR|&%
z;X77p=wb3;49x~24xhT8*|GTH64>|6UVe1q7kGrXms_x&Lv8qqubnfb_b@B+RfL87
z*ctZ5N<ToCs>S~#$NpM4|3^;qFG=#h4~*2%a@uG^{gq>PiFuI;K?DL1ORi+tx0=Qm
zB}-j5!mLMtrR#=tbX5`>4!+-g`v6IW)DAC?BLE<9vOeD663lT)=k)YG-G76+w5B~&
z8k8VWb#_FRL^Ufk5=`fwsNPW?q9k4G$5fDd-`qXkPT>7CKfkuZa_P{Br1xWg+I(N`
z2A76UE?#Q4OC06Mq&0|~HuuX=inkt!?q2Zx^tx$>FK>lPFr}GLX+@_0Dj`NwZ&_O{
zEKF%_U5rGP;LB-GYZ3{tM{vCAR~#y)NT})0xBv^SU;`6{gjcp;QR^%mS}s=O-n~vO
z^;%f0Ep?XQnm1h0yCb^2ELr-!*IZ!ps_95WFSuo$!h6LCx?nt=t~kViiP5vC_%WQX
zsVe?)Oo>R)(0O{_3lnfb3?mUmq;>BHA#tjB(+sCs1+66WP!S-O;))g^?aHcBw6n9F
zuB;xzpl_s@B1LW>&_&zjOx5kcF?20lz+FDKm9^59Y<+L3sJx5+u94w>jR^OXYcOto
zD&qR%^mkOdR*hk-yR`T6&uJB|L*T~`@s~#dTXG%JeaSPml3{8R{<-bmQ{As{o%2n1
zbapgvJ&J8LebO{WYON|}@rPmfMJh>hKUpKCzK6YA7yC~)3-OuV#?q32m2<Xjs_zUk
z?8fl*wr4u7k1DHDzfO|6?9!#B`@vtsGrNxR^Li_4GjN)y?(bom(HcLVQogs&MfNCD
zZj~Qqa+lNq$=8;h4g%W>SC$Li*lc(o>Hb)OlW^f==7^2n(ch)lKPax9EMrB7l10Wy
zTW;9SPRX!kxsh3GcA1GDpQ4*Ezu#NcF|i?oW2aE3cZDWx237A|^D9}K$LE@x<x-Q$
z_?K7{E4QhuBp33%@V`z&tkpCw-A#NU#vGqw^-w8gH^^2YS#!0rAKn5`0Ze&Bni`UC
zG_?`#h2U7fTSK$(>A7O$WF8aD1|xwGekUhFjTlditnzDbGTx!gOm2@V*A>&Jc{;(F
zF^n%^)+`elDHNT<-ra=5bf5^RbTV->g<F&OY1y<+%N7nfS}IB+NS8yR(nLK*Ci%Lf
z&ihacgZ<G&FfffeP-bVX21ASl><5;;5N{$t6N@{Y>2PNkX0<6h343M2>?5uLlxKVj
z-2HL1Y|FEdHg6}UCL)s0eE{;J!zu}H$e%eHRwcygVvEn+-frm+j5uE<&Lc7!SqQQj
z!;A3p47TF>o~!GxZ-W(<b<SGSll$$Ki&4U&J+jfZE9w}S9*=P)ko1EOZ>by)C&<kN
z^u4`QGb7FbGoTAF@*K&;hdMR?(FmR%gyL?13eD{9_Tc;p7M}O1bePbvA|jrzbwhu{
zfn3opw^X8+>%e}PB(yu>)=KqJ2o%tyyMfZcAWnHmVyP05yQ>=hNTei(ZbR~ozysGj
z|B^m*L0v%*y3XYC+(#f=Thtq{pDjQ>AjJ$q84$J!Pv$7{wkEdm!EM1+XFZT^;gN1#
ztZ2M~C)QavgtbLMefv{_XYnkmleci?n)}RfalDK72xs~;_zF?()#X9=Nzg619DgIy
zG=-iJi0R4AN^zp36}D0&7?D9#^U3MTlon35E1?q*4bC#okDt17f6gtW7C7IB?rUlv
zEo4SiE2MA8K`JE5rd-)?yOib~%k~MBtpr(ZeStqM#Lyv51PEhcfyG6Ccm70~`{lb~
z7+@WgdYS4f8C0ej*isyLsk_rF?0f<t`n09{N=Ol;T#)3}(>j$X30fCIUslr738!p-
zUFO|95yPIqthUb-kcOQ4Q{bA*RTF~2jTwQ+E|v(2rI{z-Z5{3^jOcM<SY6L8#-eAD
z=QwF0EJis6WRZG{mSv4f5iRl3V0}UFT`s0~Iv<NYw}$(hO@GZCHKAazcJ--Tz;+ZL
zm0X+$tc6}XFJ5t?P>LhFi442(E-wz*VLUMPRCgZL4z;#^*=H?-aGeByT-(QjUUzg{
z<2^I}-f!kZG;Dx8RJ$RKeeF@FN1`NB7$ZW2K57(6Xmjq*@s~Z0pm~MpR<CuFeYA~C
z>;qXcRBw2D{-m?vwgeq1=JFW4d;%(^eaDXHewxY-$rrUg<{{-2S5b5!_d3-u@apO7
z_b5PLp>)j`^8LCYfdU`mjz7%YchP(bs6fIh3>-HG`927O=?GWVdA3Y{CmBRrFIgZB
z0NMIMouoI0i9<pCi)?V3ppf-Yvi498AhvX`8_qBTCIeA84+pl-A^uzi=zZySV8B-#
z_TOxmr{hHvq?(i)y|r*phrgasJE<Q+Q!%$Rf2lqpH8m8vkD?xE{y>Zup**<t0nqvZ
zRfSb{pXJrNHkZZQ`DAY?;lD=xD0&=0K9W2Ld+-)2mm$rAksq^pjjZJ95p?f9Cb*OH
zW8lSe^u*bi)%x%an%@wc7IC6lLTcwobU)MW={D81ZNeI@te!;74t8v{ZAHuLb75h)
zLjdF>;hDV_P$@`a&))B*Yuj6>kqKO;JE>rCwkU0Mk`}PhCamo`wnqIlGSXg^uYX*t
zQ(>9|SIp@xx;`$^-PO9aoWtZ;@^A2{4r<zpBGV_{$NiaEoAq?V%+j!y3)=dE6<AA$
zvI7)1qZ@Zh85rp&jf^gDozWgb^a66{Ya91)QfO;DEF;>RX{nXafm<USzs&XB{K(k?
z(t3@!c#p~R_qn+&Q98&B^Y{uxnt(jZq+jBwv!fD})bRa211zIOa=JA*f%9G!G$UU7
z4eD8V@(tNd#pR5<h*5<*Za3IrWto3;unFt5vgFoQ+-NRM|2caf{sA0?rg{4hlP1T1
zoHTzeZ~vt(9sU1rg<onl4vU%e8F`#bEC5C&*HK&yjoE2|$Qe)~IxB_LMXR$PSKP)3
zn^U^Bk0piS%#JgIm{D}B(VXut#wVA?7uW+dib$f}n>4D?iT(r&ikKtNsSWS1OP`&k
zF5SrQFpT^n$&is40IB|^NA675s@3vmz4H{|K|@MGV9_GQg$JjE3lQOwrc^$-O~$6C
zp5XK0D&dy|57%>l!l3HuYU>g_eIJ~kA3&WwxSjZAm27@Ad3P>W$d$RkI?rPO%u+-*
z%jG)RQzHrtkw_9Re!Xj{xg~=jM=76|6Iw0S)HHs7l;ih{ZJ{-EjJIm8Z89WOZiH=Y
zS_fG$Tri8dBNbl1Y$>!vdz!jaMkMszBOiXOVOl<XS+h5JY`nDf*!Exqde%xUsiqGz
zS(74-&?U$zsHL-oW?r=9O0x6-?%@kCTEj+y67Cr=Jmpz;Z0WnCBsbZMZ-^>hN5j$j
z5(W#Ghcc;y*f1d`Fyyr7#MTJk+BV&jRno3nbq7P#fuQ+49TA*ll%g?zYSa8|3X>)V
zb5w53ovW$gYUHgCsqUb4Zej_Bj@PSDZE%zlCM_$zRpX51dw*<|^(6YxWdizZRIRjX
zHYc7(RyS61?!3)U)?C?)F;7~32t7B{3C5*t)G*qC7i6|7^!Fs_ESm)(Tf07~uW2P;
z^i;>c%Lq$Ba!c<$15S<LoGi&Dw}t_~N!2v!m;@xLnYp9QH?r^~xu6rUf=MJ}kG+dc
zLHlajW;p5`YcelynXYtF7&sdX+It{?k;jf4IhJ1T)v|499&F+siamz~9uN&Aj1D@r
zjZDsIId#Rca}hQ>VV-X#4{9CG(9>qsjWY*2+Cyx?;>5~%s&!)TuH!)j$qq_;)%4wJ
zCTXfpS(l4Poqd>?yhnW9>uF};R;1D?S6H*j#Y(F#+qpJ7r`uwmoo}&fmikgysROwc
z4?;3H3&8z;u#{p0o*Ua!TuVDfoKZmCpWwD<?q)awCgSVHN=!nW8YR#rVI>Gj!F_s5
zTm!Vus=4C%r7g!?4-^c|C9M@|;zG(cVOPovTx%0}m-3pIlAUE0&(kVh;lZuW=Xl5f
zk=g$yXObMRg7>Jn4W~GqAJRBa9Rvb{W$!*_HqbRzg-AkSU8F#u8G2@9nY|o%$m$6c
z)965?Ax<dIzcjt#*C8PAVi6H|9N^BbcpAOlbIxyuD1C%d;u`VNM{y4&MFP|0173IY
z8Nc>Z@g2qd6C{R|Np1XC>J*S!NGKzZ$?9A1S~JB+04+Y=>tWN1B9w6wD3*MON%jG$
z^aKI%`;ovz`i|1VWx^JPQC?`6+&M)ZSdv#nQ{AtD9{dfK-Se~}EFGjz<o#TL8J-|m
zatw2-I`_B}EKgm$IM8ILA-lB6AaZy(;j|zq`pd`CThQIzHkft_EN+ROiURq{>I|+I
zMh_4y@jzXW7iXqQhUtgoI(zZ=Js&DrWaSb;`8K0w2yusH3+I9_WI}ak(oB8XhYQo;
zjf6BiX1ZoTwI4dR(q$k5uncOM3HHt@E}830;`-wI$0;8Gt{oT$fI6ry963S`cOBi$
zDe6Ll!(1CI?bv+umpK!+Bkd95(n3!oT=Xu(FuTO7O&=%>bFKqv?&;1AA>p5LQF2=M
z21HfakVv0oZdNpHo7W3KM%WqyKS;tYIPrUb!R7QaHU?4f=?7B`q$(uJl2<Ya8z5o_
z93#Vg)CL`+kiRkL#NKbGMiFyqpL)!@w_?)f44_UoqOM(tp;@=-mpsmm<71qpEOOQf
zT3d~nWLmdt_^HQvL02PPNsnMrvpTA~XC=7y%D$7N(?E+Z|8_9I`#?tFJ)w2TV-m_w
zo9-|Hu0Y0$)XZV9tX~`?qk%8a4|*#;dUUft!LckMasJb<{y(9FI2iwHS7Wh;X5!-S
z0nm=#9y^b!M&)2C!7y+zHL4P7g9i{^lhITZ6h1P(R~&zC%Ae;|(>2mtCd<~XYPFY@
zuu!lMf3Vt_Q%fIT2iI?@dw-MQMiMwW*8qtij#QF(O?yQUe4&(tsnW<DviJw}C{jRa
z2^!IB5RwRFxSP$B5XFy<v)^EVO4H!=8}mLOH0cB<-}MThHS$b~2+>#S3Wg_N<OmkH
zT<oQ@fqBVS(?KEC^aeYkPGEB(AI<%!@CBG>{jwUiifs{|{JqYSYt{sM6)iT2o@%#_
zr6)Bvb88wPN#hA+D{4}uW8HR_h;T?F9Ms;jzrF8@t!MVMGApgej;Gb-DGoO8LB)Og
z#xUKKvVM{9UxH84pyv&XuRic$)^XPCH1N&$RM9)iOT!Xn_s@}ds-ADAr?JAlx3RoH
zW+Xw(?ZhnHNfws$n@w$Mu(oHorRy_OsPh{^OwFy$c49f(zG9B-mba!Fa++)_bWzWD
ztB)Pm(NtfB8>Iol>RXAR13+XFIVXZcjst*#aO(J+@+o>g;Yd<|H0tXX>p_lo0+r{}
zk0s@u#UP-`SrBAB)l70G?)C93S!Wu<7y9o@g48ZX75P|hZr=+bd_tc$oBTaL620sz
zW#=Z3pG%#s5_;v1vs*J~3*1)ga&}(!SPfjNxRdd0dcO8y?4~y{Sm9X!|A62qbEAo(
z3_a^fvX9h<AZ0^zO5}HT;zJYq&ZQ^EK_8@ukP@Uc^phzAKL)@;9l{NZRbc@xKMO^>
ziMe3Q>Ls)r2n)6-Ebr1iyggH}lm~eQ2kgS8Rmomp#@i=2`ava)>=yHHVz&|>oR+rJ
zDq)RiNYX4noLttj8>7~R*SH#~T3bSyh>$7lTId2MK@bDwm(t^1BrrnKc<^O3WyaYJ
zTq<VLdSzAD41Z~wlLy6-9(c%5%|C}mIpH3RC~=UvPvTM@XL`>SAD4yljRfFV2u>T2
zBW!WT3av*<PsPojfQcG83EEX4do1G0=hHVnvoAJN2Wj>l^L3ZHJDb_UD--TLN!iJ6
zfckw%Z+VzR0V4_I)bIo;*U*w9l5z{pPY|;<O=3hddC$r_=pQBBq5wkjCbQr^yc;pJ
zSNFki7c+(;F)Q1tJz=n0E3bcu031s)*@IMzbEi<IIa!(u{ejzYvFAYqzpasr{ZDNI
zF6cKZP5~U}^2;BY-BhUiAi_$1;ZiLGpm4moqtAXt=)K(C+wd-5PHW9lbW{=p4a$yN
zLFpn0lxjrq1S`@LS%(%KNjTL}X;E7YhUz|rtsD~!>3vE5+U&F(1;bA)v>bHPxG`cd
z&awT-EYP*^re1yMmUMez{?y8+K){m6%IJv*Dpa#_VMS@N&tk<6jnMWAN4{nHAh-A~
z?`6N-Aq*bfZO>=L{<ahQnZZ_WW1&ZZ0B2_SsND0%zQK+=Eu7eGdv+55Grry^UPvSd
zD9eP6HT2JH>872TAcx~Uq|(Gengwnc{1_<g$d7f|pcm9RnN_;>)k{2%e^-ZSySsNG
zmqwc_?t`;qaI|V;mCh^W-FUGFV#FUUG`v=vfz^W<B_!kCa~ZdNpiX>_!gRBff)n*p
z%AA${Lx`)sp`E}-hNK)>Ur@m3;zFfXeAE5xE2VLR7dp&~A)=tk8n6Lw>fUdcFbb_6
z6KZA%ERfJ!!qJPMZHo5in-u2Wj)h-hd-JdWCKMyD{c!1UxG8CxyTubDTM_;u44U;u
z3oau%xHx)S@3ilxTOb-RE_Bc60QN;|m3L~!p7*L*b##o}gCqy~&GnjaL^va3`OZE7
z5)^3Di`9T-9sA(_z~Au>X8ZP&zi+4wX{I0x$CVb<K_hht3A)ThI>!g|#kfXHIz}|G
zzcfoI)6DkZx^=VV_QLs_sPJX=Q||s%&+>^})`oJMLw5{K^KU@_y2g?(b1-*M0)DE}
zj|4MU{zAkd*L@h~Zgh6)bF;;&d%5jBlZN+*`ER=L^8OM<*f*ip0Wq{3M>epny0tLb
zZVfh+hZz6z3}k(jfUb{LQ#Vamr^T(W4Zw|2YA#mRd)0Heod6xZ$?Hh^z&35GHYtv&
z3sIfjP(N(Ble#ha?O&Ex#s59;ppF>O{2OsFN0apG3@|Y0>g&IZ^#?T?`7mEX>A*Qb
zCSr#3)vIMA+i?dad}l0AMK}A&k|jq_$imcK+RBC9nCKTQvTq)D^lIOq?b5=%oc{h)
z#2y`w?4wN+x@rYLrfAyjdhvr><M)zVgT|W{$h>u04z9EPLZ2f#L6*$xk6LN->yE@{
zBM3KalW*|$u(2aVULj7;zO&8s?cHDMeui=cB1>15^1gj9n-Y3SWL|9DZ99ZH`tyz~
zm%}da2d1buH9jeqBLp9~p*%#2sI;$1qNM?J`wrh^djFsHcE28*{~_u!ax(n)dpixQ
zUs@m9SB_pjy=XE~f#f+aDB4)jw9H?*l1vPWzpfp5r0e@Jcr_wNjKkM<JJRt<w6RyF
ze=bu`Gh!#5Zg_b}g0h=~vf0gQ(ajn!`rMR~#8@2Bs7GV77{U?Ygv{c(d)p{^f9hU+
zU1cHGFpFvMd0>QVwspVT(sr`FNQUVm;&OHm#w7JIqJkh5qwF(QF?Y7Kd)j^c`1udH
zZV{hH<7=DLnfH#Lqrbo2kMDiIvs-p_dph6tc;GgNIVCq=TG>-uI}Wx!Hioq`)l=R0
zyH<{;uAV+^eZ5$F)hlxi5+S4J(?atbe{5E!>uWlDxvY9PKE9&*8deEo;O~Th0>cTi
zisaO?%U$`Q47~S6xUgy6)XaW9D+$~pW$n9^(Nt}a@+cx)I(`bB0wGY)=TN?bd!G3$
z081;Z_hQH|-Ghm8LDm(Q4!g2QUK@)k3mhyjKEtps+>`R+ij>u7PIQCoKlHxJGg}Og
zYB`3$s6w63_#x*M$Y}vky>}r|%-dcA*9JSjd%*n#dT<psW__Yx&VGX2Zbx$2CXyw`
zFQtR>!U9*p_7S7@%`<!^!+V9hcq0G{l%>BBL@DF=^C-YPG*s&L3lO&2PfCZjnk>cH
zFH%%10y<iPco`HMOvayPN|txng?g@12`~+?+C!2uua~S_&>ButDXc0q$u+NzJ`F~4
zHJT~g*+qD5haj533ke`uVgd%XtTovtHHW#zH#A8*tXs}6GE$F&lH{j=SWYvaeq*F4
zmvXL`Fk+Q1(O`)3NL-5Hs__jKc(a8_&YQ1w&#t9j#se>o^OI{ujW&7Fa59uqu?e%S
z8L-hoIpu?RC1yNFis-lzXBh+pd9n)<_z6JyQnBDGog(XQ<P2^t0{7kF2Nx`0dhiMH
z69)+zO0JD3MW&oGznn|sd@~9_7ERKKvD1*1+{T|;kE>MC92kN25q1>hp3dkU9)5Ah
zi7v2x8uc>gj1BsZwah0j<cAL{BfrV%`M*>|YK`vWk}pIK4<;SKzxE;}1~`;!Y}Trc
z`gZdnfJ|cjFz&AYO}<El3{X!p1fOC@8Ex6GL(L{7K~Jkq_+d%{4BmM`mNf1@R5VDG
z#{JIO{02l3MksOJoEe})v}<+5%V0S+U}{w=AGH_@GPi$vVDMzY-#(C>S!8z=WZ(-M
zu|Z`T70W&dnBs8v`EFZ`%&P#mD1k^KMtWRbz=;Y)31PMbe1L{Y6x}+;DU7$xHk(+w
zHy@Q^Rqj$h4@ZhHfb#a7Rj&^8Kuop<<6mj@v4R}Qzn#BBLV+?P0Zu|=?fB74sR2;r
z-~u6a0Y=~#_F7od`K|B9zRq}HwQB;JPa-yQivlZzAsR3(@Z0q-I?i4Lo`MiUoV!N)
zD3#Ctgt>F1EUII<^tTXb_hfAIzSKx?NEsb%Fi-O$5o3vZM);jJ`ca;*L0FkHM!X+O
z6!5@UR(Mm6AYEDIyFlO6fpwx?mY_BtlA+NCJyR%w*-<MtD*bq(xNP(E&`?DR4;$lQ
zs)Nq(!SSdEz=#v#SP#M@c#xLcAG)*zJY?A!2aHUj3$jQyktd`XS3R=#VOU9TW$rq|
z_bR<)6Zb$9%E}ug{HZ5bd|KL`e>bC~;@+kcuWK)31>^gQPJ5D`W~*$eXAJCc1t->M
zy4>MOl%;Nqc303bsbB6}Z|X8^u=B8}2`s@8?q0mv`^4hHx)5>B?e|5I%1c_huc4Q%
zEQ1|s@W83fHd$awE~hkoux@MFW5cGTGWE7~?^J5t;W|KHc3O5ch3o{kc$A@pIyjw7
zuWz7m1ibG!uCsMRpd&1m6iQSk*>kaN<jFedVEUlOud(U3xpgh#X@Ih(h!^C2p}c+G
zO2(XB0}nmZd@42x>cP56(NmARsrOnav#MS_hP_rXT=UaCiO-x#6p9Hk;M~B#u(wDd
z!`6h{Uz7`o3qozzyiz8zb2lUT(@oZLkEqx-1*T@-yRGWLO(GyEa0y`nsmlsaW^Jn3
zE1U$a6?Q_oonnX4#osouQ2O-=D-VJjhdykN5E|MI?qYM@-Q(+4XmdjvjAzwtLBb%q
z8a-?!@Z}w4&>5Ltwo8klBV*BccyNt3SdLA6)`7~BZ$9DOp`PT4eUDDWe0lTEx@lwL
zmQ;nxY(e{*%VIrsVJ1Dps7y>Wp2u?^a@j-zq~->s{j-&@Z^3w@@in(F<B?`MKi^+C
zwR=#BdfHNVcQ%jZrCkb)c(sUJP+umtgr&(@Z~?cEA>CyuRrZVazGBI=466S^-gVo~
zw~i6j9+XQZ{smTVL`RL=?+)S;2eLmH-#3nbEzlnhn$K6C*-hzNg=#UcXWT8zw3xmN
z+I4OXN2hJRq@;#uU|$3z^YT+v6PQF-Bz@^blvY<QwRhX}mz!7<aAi4&YrZ-yWWu1!
z+PTGg*8JTht3SHH%c0v}Xq$CfXyyL=xO5(W5JVqIbg5h`z?;_K{FK*!Iqt9GuYC9V
z+*|`*`1LM78*TWr*}80(K;d?sP5m|+z@<P9CjVhd=KN1nGBd+}O^KXR*L3=&MEb7O
z?rN3CKmvjE6KC@s-p6B7R~O3R&qsly5k?^6QAxf@lzv_5=nHD9m}ohaEcC<?#BABs
z<D=o1H`<puF?q7$zKupPl9O7Xh)g?Q-;N;~h?rAkSsZ&iYl;9dJ&TBJMe30yHja<4
zLW8)JmfF?d9bX=eAAW&<JiULty-kj1-j$wl|8N{`go2VLjqa|j%C_e)7obH5OJ<x4
zo-+gmkOqg8Gt(MeYZga?+qVAG+A^%^NIt4i5MSljny`#lhsL=`f$E&(QV!A4JXq~$
zHbUyH?6^0rI9Fjdji_Y8(H<8|f#Q0gZ<xG*Iw`p6TLnLbBW76zk^z<6wE<XBbpTH!
z-rJT~fItLv{&x=MJzSIF;;LT`LxG`%iSneHKY{j~C5|H;Mu^ST^oO;w5@sa>;)!f%
zF|@k+@l@*{P+@(d02WcrLLXb4@&E<)eowdc;Z)A1N0t-wh7sT%%@B%|fBlTGRNLy|
z+cjV!;6JGkNu*YM%EgRD(^o<*?-SJ$H7Gn)Gc%W6yp^?b)vcrk^Vr)9)AX4X*Xrhc
zW3RAkk1hQJU=2bEk%W-|gz4)}O%8Zioh%-wrH!-GmpL-TECr5$ETG&pW@_<{OmUzR
zayV(Cf&DRyBxd}tPwlwpD8&JeNAS8U5qoWx)Am&rkoshK0&$E{5;$U+<v2a;i%u4V
z|HarjHfPp_+d8&w+qT`YZQHi(q+{E*ZQJPB>DW2hRp)%#U3K1{u%5Y|8*7d+u0Hzk
zKGPTY&T%<n`z{7TLa9(~t7Mu`(1SvBg~*4ueftAv^CzrWLVE-AwNLXa1qo0-q|hfY
zUUXB(j?TBmHG<A<0^rnS0VMF8Zyt<$TRvBzQ!I^Ufm20%%&V8ak2?RPw5Su3@PKa?
zG&FEOBH{1ZOMTNm1B65h3G4cSe*E5nAPoR<cfHfyI5&<C>^+;C*{-Flb9SmS(0xDJ
zaD8@VhLMp?fiD+HO{%@&cZ^qefj|yidw8b)aQ!m`WL<o)PrVn*&Pybz?j0IjBRK{{
zK!Is<9SPWL!MOfuizXj10TcUrPT-z6I>$DD$K{Pn*sfi#67jXcX*Pbes?|=+<1hZW
z%cqDj*Vk~G#Wpwy%xkylgl>@c6e(+QG3zJ>!GdH?L8N$$8hB#!5;3;F$&VYBfXxbS
zAh!GjwI;d}bJ{QE%Q?Qehuddw=b#k3ERLx<t(JN(Uqm-47t~CZQhFEg^h&EezYFjz
z2fPF!GNOZ7L~xEpn!Z#N^j?dt^Euv4c4j_aJ1Ue7^is<Lr9a6w_)=gJ$^P((0R9bx
z3Yk7Ke_M5!P~dVvNIjyWpAuiY*qaN^dCI?lvv?q8$#D2nLv0<lGNgEn8d)}1Q9IYI
zsL2?BSZqRxd|lMj?sk_pQG|#>$tRL^<fyP*y-^4i)(_+&#HJ06n4;<i$Cg=2B71cs
zq*~eXHgi4%V>^&;GdXeIYm@(JC~&AFv<UGfX&n*;Ic8;icTt?-!{&!`yoUwG@?vm(
zp#9C(<P)Yypxa5ht_>Is64C8g@@Z~YF}c~k^qj-%CRTYKN)zelzLo|9MD)kZe^Y(-
z&T5%PnN{gZH!I1Lqt<I~uVVtv|8d33#oIU3<_4jt^rW`Br<~h9SE>`c0*J@Wi_z|V
zZD+0fmd^NbeUK>4@2puF`e@-tE9XRjB+?7J(qhoRf7ZEbHap<3Pg`uGJ88I$esA~J
zI^<JM26o+!opNEh$p1|I6V#xZ#oe9;f9h2cbKNU?PwvDHE)c=~e^*(KpL+Vg79}PQ
zw*O2oi_!S0vK&ah&$V-#^3BA-01$ZZDdTGqt%hdou9kRAN|b5Aq$_uG;YGO9wY#<~
zeZs0;O1=o;jr881tT;6-mG+Nce^TG0*wqrWO=zM-yE^=|$K8ty(n1n6x6vzW1<|U%
z+rC#bsb^lT(OO6@^-H_nZ=ZIbQ*&3bb3y>)Flplrx+D#-rh-5n1_o#zseQYB-@l#~
zw~yMZiN}RFXFkZU&e61YdrEKgYIb*Yz9g^U?d)p!YmzA1^zK<+oV`AuQowBaws-Ut
z@3%K+@%a_*-=5Cz?vJiIC%Jh(iJVhzd+lnN>i=Y!NJ65R4T?e(m{rC|sHd2cpqW#y
z8}e4<tMV8}OpL`x7CT%%Qb=~&Y|O-uOcE;(8DhSs&%VtiPq%2<1w#REqbzZxbyZs}
zqIA%2mb}+rCXuq28%72l;*fx2C!I}_JFR7ul-j%p*>)th6anLVIpvW_D8hRsx+4-3
zVgPmH*KmwTft(Z}(I^dmxpmvxw?l=_I4`~BT{N6;mK5c4v{_{C%&l?zlxbRi;Z@uY
zPIJHPFU_~IY<usjT^UR__Q{L)9{10*Hsh{m<n%-}T)-a<`g67Wgh*_Ka$oqPn&sc#
zQKtlZqmuq)!b*hsYz81oaySwXl2A<D7*Jp-?U~{5;FN1El2b<Gz5|2lck-TQaD53f
z_nBmHNhH3K>1e>Cwl%FQ@=Kb7Pbq0};@0*0JcK14?8i|yb6Kkyz2Cn>UtlucOHZW>
z%oeJW>&gqdfv{ePlJv8HfjgjTcmAqYsmMzU&yrHkdzRh2F{vI;MTCqiJno$FavLiH
zn-O5`B0kOs{AzN{cM=GrNk2m<rU`+o6n3!I7rZunT%WTunqw!L$Z|J`ztQZnMwx#A
z?zq;e{Ft13uz^m^uOAr=(0P}R5%*EgkuMCEyB<9CKCzY4$aE-xtxDfwGKH#%_>J9t
zmdd>?Z0E;S&&y-Q)a)XFv31kbS!nWn-;{`;G~9SGM(Fwjo&vZNQfLC#=D1T^&XzFf
zoZgbjcUsg|*14e$4@gz-QyO0|=gEqpi#9FUqeKRfSwYdmnrp_#3ePxiJJtqK<&Gfh
z3x#-kn@i=?jsljh<K;HPVwcU}Q7*YI13x5TJyn|J$BVfJFC}@5cO9@IhY(VPCmawQ
z8>vj3VSz-QOiX4;X8?O9vip0q@3W~`Sm~lNAFMFztHu=pvDpQ04KUQt-DiA%0w6DA
zx|h_qgZorS*Etg%YR`ErVw)})K7=^#V~Dul7`;b_q2J^{3}1X5jm@HCNADE|6iU^m
z`NG)Xi&CxzjplF6u&ax<9|+~ysjW6LJ16rB@$O2mk7+E+nKTiQdOaid)dV0f6lvaG
z)~79=W8=7(9kpW-L}S93W?kQZU(>WunhxJl+Q6`s@53*k14*@Sc7D(E;GaMQ$?+y}
zPa%Sgjlm!%hhL}S^iY{=zCv<~qiCE3$nGfrLZl!^FrU@kz2s~EC4zJR<&04pXB(Mo
z=`%aiE(SOoq1{3PTg**|3f=f@Yc0sEFI$sY;u7@4^TMj4k%}P)I_<%s=(m1|lG;IB
z<{G9T5Ii1j8awjCJq}Tas(|jk=|wF$f4s=w;|NHmj4BGE1t8VIX7M0mTre51ag9#^
za`$wko#KD7f)p0CnfgS$OB7fBGf#-B2_~L9e|}#)8`i+y6g&p-uuLm)RicsCkG8|~
zQN2Msg0C%fQf3d<tvD3z1g`E+c7&|Wck!D-6rd2SF5X9_V2%x=V2P7?K*Cz^YA~VW
z3O6TokyB-0mh-iNB0&g`dOL@Ciw|812kF--LB;g_{v6_?M^R;<1jMhkKtiVclZM^-
zU>ys%;EvZ=TcRh;OqW<z*4WPGnDqPd1iN8lpf<z@P>>Pdg}XZ%o{%ij*oPe~ozLJT
z!O+Tl3+WI<B4l`iHM?z426(cGuo+#U&F#5b=u>>9&RSPi&{>S=BfSj=5sVAz*ddJG
z)U26U&6sV$mO-Rd<X}J+L!Jnol4}J0yt=n>!l8G-xwy~l1>(TV!Tj64t{bR0Y*=N7
zGM{z8<mpSV(ZZ=`oZcw>C9K6}<vl=xS839jyWZ+XiZm*+heu(nK`bFVwWpz~`dlG2
ztVaq6=`+g5*nwb>X=RH(@BkFsO$G6=c|Pa+8DrhXeK1v^*&tIw;)3u9^=@fdUdB93
zNlmzTklrgvAlp|*UuVs(an9%X0KQbl6Ql?d6L~AzVIXgcs0RKVx8Wk#rCOwHMm=a6
zW^lew=3nd@dgC5Ps59;n@#<im*4&(6={)63Twom#;tV=p{?O6fPNeEnaxLs}5cVFB
z<&W-o-=&Y~`cz7jkFwksn6Z~It($P*3yc#y6sG+)ov#h;dcvmQM9uA)ZH>Iaa{D;s
z`aU^=uQ!QYDSdP{!rl4s^2m=)k~6>*@#TO=tlCe3e0Vs*L)=#vW^z}64`vUk;O_<x
z(9gpPHO~(+-YIfh23^G;O;me@aUG0nC-O*SePi062WN~VPNVg(ARBw*;3BQ9#7#av
zr20$*o#g8anr6ncA!jZc?MJ>NBIVphXbEoDM9~Y55{VJ&k=}}QQqSfN6V>ApJW4dq
z4uAAm2oS0+6l&ERV!{ZO&n)YI^#IoxzT$%UHkN(sP9i=jKZ78H<;E0a`<o(kEV&eB
z0xBN<nvY{N0_Fx9)3_5~vv*(&uxOeEStiF!D7C*}O6Bpk?kiA_RX~C67@j+#(CKM(
zh6ne=qR0el;1Ct#QB`HIe?R)NZ3kB$V-kbPUP}!vDFFgk52DSsPfbxhEp4HZZN!rk
zB2yB7U_f`l^|v`XwO@XF2E5-sY^1`h^m9GtrJR8X7?|Q*R3JDapwV4AK%I98hNpJq
zwsz||P>G#Hp80bmxL0Ope6IFdjj%_7O1s*D<z7bEtxo~)+&HD{p$}Z2neUrd5cC4_
z6!QyW*S>@KSYBU)jyZ*7yrBXM`P(8{Bd{*m?;S?yp5S9eL!uV!qlv?PSeGclesWnc
zrPqRyf)+EiR4VSv;Yfh!@Jqo(EoaGAS>8UOj>7eY$!O3oN|BOmfwb`LG`EtDy;pl%
zTM64>S<N}~=wSL4TrX8#DDRH9t*7ZXO-Bi1*p)+eg#Rajme^$cEK!Olnf*<^k%Q|F
zuLAvTA@4FCrg|;b`SGKGTzVjvwJ*T0=c~<cL>d3x0S--?M+cVZ5c``wPw%4KVkd8b
zAw!W9aq{;<Gs0133<!j>rFHOC<(SzW3l=iS(?>GNxrIy#2{ncHXXPxQa9wFT{W+zy
z(gicI7m$42%Fqx)9wg}fEHAydHw^m(@hF!SPkTe&!%pTm;6$jiiRlUc%KUd^mNGM2
zeU%Tj2<g8zo{7vMo|!)*-<I0;ld1NO-iOCNxd~*DNKhk^zPpFd-^Vz6l|F>yBM6GE
z#%y(AE)%MeuyIQlx4-yrYO9IBeX3Mh#x37?oJI-B4&Pm`rrrBQ2&Iuf3PVVgAg3$D
zT*WGY&&5N|8Lm4O$WUF)o7GjHK_dZ@esrlSyQ+M^DA;E4G1&!5N#BQXGNPHG2s~ln
zai|JU9X2!QruJ`Pz$(-75%mh!8G>scs$a8T>}0y-!6pN|z|yDtXZ+fu`Q;Peimo}`
zFZ^v<mG~c0slCfzz2Y?t8Gkcg4?uVBf!Z3pW?&loe(QgI8j=aHN6*rd&kp(re_c7$
zg-MWy!nlg@<Fb!wQ0|YPy~OsMp`u~6gTk0QzdQla49XDTh7`&+Q%d&!KKf4Ze&yM7
z{1bHuq;~EHjn{YcuNj=<A2T@Xk09(nO23*lCY=s?V0&)#_O*VetLIDbZZ7sT$dr!H
zSTijOw}Oup{P0BE3X-jT{d@p{noLGFuOA#D2yY1?K={1%hYTALqUdk<c)B_}WD|`Q
zhE?e9hX+0L??h9M>m?e8g}Yy@k`L=6>A#+CODnBN??fSyokEaJ(9DQJY(ARg*l&8b
ze7e7-nd4fe4N9s8rh2uNnLNL*u8*%LKYu_OM4o=n5y5u3{J!1YpWe@BpP~t3@a%dw
z*A~6X(=AD&s8NHNl`tn=$REF=iYG*9l<U<ei0CS_Ok=c!Z_M8}SiXYLe$kLrAx0VJ
zH*nM@ifXDjr2mczLrdX#A@xA*KFx332E&(JlZY#F^_fKODEDCjj(N6PsBYPCjXkWc
zv5<EwVz6zh#*>kW!+)rBNCBHX^_;7mf8s%X;%;WuD>6YevygP%(L5RJVv4ivaN+M;
zX~IL<6(!@--rec56T<omWD0_Jsyg>V%J>Qa*}TTG+WH%rwG9zi^JtsZZQ8m#lhxQX
zW>`KZ?;Mk8!J<oocpY?CjhFLgmaOuSpcDGGu~KyTmY^mhEMDPjxEy7HrFd3ztg*R+
zxlnDo$@(<TQd@LVy;#>AITd%5?oR;%B0_gSMLG7Rvj~s=3v>TVL$pU;1M|w0J)?gh
zZ*_GgCdemijFY3k%3<A&@bD*G9z+;PLG<r1licDdSCy4nXR3mYm<_hLutA}Nm5zJF
zj`OYV58nC2MIu1EZZo@|nflJ0JC^(Ty^@;uqi)t1sHY*EP3>)7hqPUSu>Ot;cv4wK
z<ayKA(o$2(POMz7x@5l~ex5YP8>{;X8MewBAV{C4&6L2SjiPY;0{Vu*?qW(_Wu&2M
zO&(V81YIbn=ADql+8&xu<F*N&+wE5*gAFK^R@icraVhvV7Ek6bg`hsj6s~N7K0Mz3
z^La<26F^iYlLifP9jd^<!FaAuK_8uSixn{C`xM)PTb;FE5RE6aAyEw#>wORbz@6{Q
zid-hGSmjjW*j2PI&;|Wjf&-1m`;PI`&QNbE35fsf^K;gB-Yo=ZZ-(y|<C$&K!NuRa
zt3I2Z2wReAD5RewNRL<g9jVfv-XMBIp(0X!t^2D%$Ho=x%bA$+T(Fk|Y5$U7joUM?
z2?&rcjCA`+N&Q$*<aZr@-Wbe<eCf796f;`NZ5FfnObjKwc&Oud_y_^p!Quh+XQBqd
zmT%XmM4~O*CqVs<P!CGH3_f-tc<vS{nRrH79n|fNt$=hv0G^+rf$oT?7}!8Uz;0!t
z9L~bu`Dk~tc#m5?fRa<+2TX7$GW3~X_8-rNlmeYHU_i=!xp!*B-domDqq?@_ZPXBc
zgnZck=aux>j5TOkc%?EPX4OTnoUL`YRQEP-IxnZw10kvJSQO*-0x(PYrNVjzgZ9iT
z`Q!nL@8`R!PbYBnqlZq&5C9+!dLap<hJuOj%RvaUf{^HgER6Z6l=s|VpsQsb-E*_8
z1i!eWcm+C#T}s}I{5pVMGBfcai)<0hS`>yz2;L$f!6>jA6;J^cMTgjGESY<$!H^q^
zhyp-lABMy<DCS}$dJUG|^9R9=ARgdwPLBQC9=&hImFfW{0rl^pVtuf~X;Pd6_Q0YO
z`>xQTssktS-(?)oA@_qzJhW#@4Z-yuHlcMc-#2+OSs`p=G93LG4wcZZcU6LJawIMo
z8n84K7LnjNxr_p&ND?@x7<u%qLSim<>O>3vR8sg2SiJQBu%boz;1k{mA8v&Li`e&7
z+&crSZG)2dlj2rdK>_}WVZ3-;F`NZd4KxQ#$qPP5o&H$SfaH~@h!#o7X#jm{Jfy$9
z3UPTYGWM;2@=$++mJj?LgqdWH@Y+|miC=f`8v?wy=E~`Si-PT?wl|Ex<4xl3!}wFR
z2lAYgf>fvU&JZ6_DxdwRK>+Fn{>V4K|EvKoP^bn589i1|J1c-TCJm7yj4&`2FBglq
zE>e>ZLZz66(7JqK(K5ypV=~j;V@C6?TX?1hxDa1nB^MeOL1bB{n(2<rESP^Yn00ZU
z2lQvvfHv;=I&%}aT6Or>vkUKA(3m5K>>Xw}oMHf{YX2$F20EE10bgjuBBKQ<t)G!O
zh(5dkgfz&hg*)&QjUTcA=A9^Pk5uAP7||&M3g@>sG5VAm)eLh>?oND)fiR4Cui1nQ
z+Q$NT5hynVivVd(M4R<qA50A1SVN&ko!f=<{bd|JnY?9S=F=KjlJ_{Cc_2NN9U^fM
zczA|-aWux<R^-?ed&oy%#@Gt(jxb>g8e!vtJpQ5~UO6Q(&xb9K3cDu^TmM9CkqE@j
zK+As8QzzqFZ-_Z%UR=s@KD%(NEM+8!Fp0K6l%uME&t@Cz)G45@|F)A8%&ZmB>bnn}
z@T$Z#FjIE`<zdZ@_dInPWgo#I=M9MRHOq1$Ra{hXJ6PcR9m=X-Km<OyVVSwSA8u4R
z!ncL^tvE!gQ!6TIT>z*ebx*`gDvdNE^rYX+Kwo}5ao<W_A5~XG`L~xTJFo!&Lf6}`
zyc^m<P~d`vR?;ELX9S_k#j1_cbc9A0=%kHoINYFBdHFyDioIc`a;h0!dE(9na6^a;
zU|I0C%->NU8(Y%2`L#tYu|qzn&JR-z6$Lri%5yT=k19a`ilMII*L+Ol{X}tCL^SjO
zrtCaT^i~8;)f3?qT{0aY*gJhun(sk1{sSstlBYHoO0@8402@Jr#xtw9zU%da0?hQ*
zx9ZQ^H)6U&5QeuH^l|spNe&snwQ`^c){hrgLSIr>ffH2WCFd^(v|4NCp%+dVza=Ql
z17Z>PDD#-)KT|N55&&AuxC69`LG<_#c=l0v7wE}(WC@3aj`lms3QFL`tQuG5kZ&zJ
zj0QH-LPyE{{<@A>8}|l8A!bU!eYQ)2I2IQA!rGHlqtZ<I`<{H?`^qFcWaO+JwbQ08
zW1NgpxH6pTS3ffv6S$<89ZLQ7#fZz4DPyX$dirM4HO0ix!^8e=Xs)D9H+f!Z%+>lF
zY*CWZl*>ZC2A|17x%-?O$WcWBSGL#oJRQ{3#C2Ccr5pN-mDX^bxv@04uGrIV@RLa=
z#n02u@Tt9)>OP$ld|O{h-m%+<E$By|U$MBer03_%W2&()eL|xmtb8%@*rkBawx97!
z99rV~sg>F{Q0okA-B~p>N)p5RY~^;^1TX%Xhj#E2N>A@Cu+wg)3VP|;-se{hoAwWY
zk#hggumj*wN>U(5JsAkdbC%%#0F92?==ae(_XlS&w6-NLtTvzkWRAAX%++|YiK>fb
z1V=qc1|3I}a<QL&&suNVN!?kFZ?*Y!dnVAg9hxnk5PgISy?**d<{UPGAqUn4Ojcr0
z;*lK$ay`8}Bv_S(hTQsmwS0DKKG(rn*sn#?AeF#t76Cv~oOcr6G#YC&?us@$_qmgw
zKqrtP=p*xg44I7ovN<yRKVHZGKljEazCQMcN+R;^J)Gi-*cdxaR>Is5DYln4gxhiL
zyh>2^>ID<J7*OV_n!}DqJ3vqHFwBJ3C`-*SATNsGCxAG`3c2JIzH6@^`l7Zq#YK6<
zzbaFnY~R{s)fs1dSj;j;5&=U&VxCC*iX_l1u*ehfyVVDS?3q%NI^cY^eXJb5FO2U-
z%8tBs!qdjhchYK_C>0hvu$7SiG8924r5d+YIEByI);Zs0@#+}}Kh>REn37sL1~yrR
z7E8E&`qjDXRyj2S{fSeDrUqY5#`Ism&_6Uo%$)z%2l^j?CN7CGcAE?^p*OFnJX`{U
zk#DMYI(2Bc@-XH;o1XHPt*m<(f;8x3z4SJ;f;d&=zew=!mZ72C!2zG49UQ(W&{y3A
zfjE<mS(5GYXw32FK)_8v{h(xN^(OSl58q<P05F#XUm+pNVh*>Idp6m1umUD~fPzkl
z06|D#JwxT3T426V#grdy^?FB?)jkR`k0c>=cxv!hCx0t9yzmxn$B%HMw(3}`9pGH2
zc5*SF&8_6eCuiclmCPD81^V|(dgY5^HU$V!$A~Z~9f@ZXBEj(}sm3fh#Lt&?XAzt2
ze%U6fofhXB30rQE;l4<PLZwWUQi!3a97bIZsmR)v#O10yMz?UH@Ea|T$Y`CM%sEh@
zSGO)ZMe4-*QK5|GBvRrj&iD;|>AiPco7%=&5kTTj<|dnpcCz)XsF<>cd6!(S7#@OP
z0l&L(Z0DAQq_3ZXKsk>3oN4ItTemv+y~O#X%xU^n@picM#Ny=KkYh(nTBuv(&wFLg
z@0p77PY$D#=63e|<6xQo!KP<q`M>Mce>hl`>3XQ2SKoiXTh76V`!GX0v2(IaUz9v-
zj|k3C%ORnlhaamdaFO5V1C37KKxe&uQoC7)h#+bTg)nX}nL!=^A9R@d5D)J0+S!jS
zw4j>({Wz@^A4OhSTX4innUo<R!A0z6DX{%~<l%vO<=v!nmq<%%llQ$5>5B(*-ql;+
zZ<tty21TGu_}9bzV|ir!e}C)$@NoYD!^6S+kE?hlCWil<ne|(JJ8rWL(Qj6dFb{sd
zphdQ*2i_*Y%AV2@UMo-Ifs|5Hv%p68q5k*bw$s<!jD0B6kqaB|OFxOIV`$dyhiRyT
zJ;a|qJG(!X-@SZYulqivzzQNi*v8vB_(UFwNveqoB6ssZxqOPkPCwp{1{z2{8_}$Z
z+k>jN&g!W@x7(wBiU-{mtgWH-O3XFy`#u>UD#}MpquZh=qokL};6xG)6tyDprXI<u
z1l7P6@Zwv9w1fLB{5hu9==;81$P5ef=*ulEA2n9$kQi4F-Z&W%T&4w>C1%U}=q4Hr
zc!9;A@LnicGH-=hqE<Ie2F|TAf0#ZZj<reXCizC|t()mFlZ)hHGElF>PSDD^*|+R7
z#C|2rrJFSu9*MqqFFjRtZF(R6^)#2$nJrahFP#Iujn2W|o;yt;<Ev2U$KjP#bNcd9
z{IoVLd>yQF!;+HQ7ZGz%O&BQ2)tI&z*-Y?wX{?Ade<LB$;#Dfed(*(IbSE3kU){mc
zfq&}t;GRFdXP6PsX)h)8@^c+}V7Xfx@ZHq-W?SUgobu(CDdCU5%Q-HHjAS{f<YN-J
z$iF~gkF#fH%^7p$AJzECoz{9mX2~Oa(_1O%cG;8%zg~4VZGJ0z*VL;%+rM_09NWG+
zcQisQKJ_UhYJ=OzcTB1NL93(75s%4WV;yS+qV?B8m3pu|ZsnA(=5oRMEZ=<`+*aiZ
z91uZz96F?t4N_DMkT2)3+!()g@Um`H;EJpK)>)GECj+tS#;!U;FdGEX_gtV({%ueT
zf6ip>$s=fY%-k6Td|*bI^lq?IarIfgxnGTj30E@1FFozZLr;Mr9|t1|`(j+L&$H6?
zfI@#)mK*{f!P4xuf=@<7i-;}=S@oMdrP<vkT|^_jdUdz*9h>I_vI@YDNr!%;iwyJ>
z6+*H`EzS441B1{=s<c-)cB+7;e1-b!#YfNrP+9&9PtgAzs;5fTx|hC8qCi<U9r{&_
zWM2*iWC6k6%!huOXRU;fYvI;|<-W@6T%9OjJ*Gyc)IHTV4KN$nZzCI>UWRAsP#>*7
zYx}gDtC{(AtdVdBsAez0Hq8`^a$R(~f9g_E#MWWVzIqxY(I&Z^<BQOM;Ug9FboqD=
z+c^shBlB!@a~1cV)QdblT4eZAhd7g@0KzTfE1^gbRsX9+w6~v;3#ldf(-{Ej>_ja_
zJB5H)V-hIlyPG}iY}PL@?QiPltN)_X@t%%hH=ZqNlEaIFmI~nn{7g;R#`7Te%p12E
zQd>?O6ktVnEb=c!QBsc9-%_{v!pxs?9W4+Z3AkHrwv+^ZhNjLQiO!d_xL!&pWi*tg
zM28Y?OBMO{M4$RpIt;@YX74!KBxc}CBLd9pd09khGb|Mw7I~_rJ*Qe!qbej)3mAr#
z*W{DS5aWwOu)&01n@;+`K@_@kV8bT)5za%?D-Z5`SwqP&)7gA^&+7^UYSISRsOqFh
zeSvj%C?A6t0+NXfTot{CNQZ={{f|W|c)q80Ltq+5Wx#`Z7BGu}2>a1$6bz^V{oHkG
zQa(_+4E-Z4Sg6;I=1^;h*Ej26)_wOZ&KZ2#3#x=!zw<=PZ(3a)ENKMSgX(D(DZ=*l
zOSuU~SY##{h~G5!h`UH8O$)FsJLRE~5tyS@neVY0#mzODi13DvfPI?63F{us`16tZ
zOgn8)H}9LVCxpcG1s3qxENd)bB#g*Yt$8cnqPPgQd`gP~!HD2S<fAG2Rmti~wa+mf
z3JW55G12`BhGH=NFJWQ~M8JJ;gOoQVBr>DcYHjAYPY3FU%X=SlB6P&lLp6g;XwB;y
z1uiPd3XUO(KfS8Ld7;a%9v~8iH^M?wk$||p@;oJgGD&A%@c<%36dkfQvnl)k&`|0n
z&Ib;z+WR*I^6{3>?|HPv`U7k_V%VlQPRVLprWBLq|JC7M(qoJ2`;vSx`A+>HJz$EI
zN~P-m82Bx=hM~f5kgtzQSz}3fN*_46#aTa{y`SVqaoQn_n!kAw(iC+4SZ3<hjxK<L
zRA3DH9ji<Cdo|SG#8XW<4<$80eP~@0@{IZ@zOjT>0YyQxN*E5M+PaW@aSd3#qnRcU
zaM^V>?xC0jy}8PEFq#z#?#4G{jQta$n<}7>qoPp-{mCZ~e(==u5U&pCoUx3J0%{n@
zB85$J=61oGXqG4e+LX2^L0dNmz;RrZpT)USM}8(<)e$PYUAsJ1ull@9HwW3NKX1Rd
z8w{14HK+U{cxBX}?jc0TOLu|&trfet*k8F=dcCt7W#I(sr~%wx!K%l1vm`!uDD~lH
z<YMA(N<Ux#3sJmUB-3KrmMEaB1g1s6yTK4CIFhO}!=+vv4iZIQibOrsn9t2)D~*9&
zh`Ko~eJWVB-yxHt?HnC@CPH;GuX!$h8>HQ|IohVX!&M$m3X;%-L}?$Vhp?+!wh8s+
zRI#651Z8T|4*z434fy(7K&x&#<xfBL?RriM0^^gvAf<gLd(i2X;+~#soQC(?L^r)C
zSZ*~{!PfFJK3(Y3gjfAGB}9yNm(xzj(KTQLn=iF|&Fv{eI`-Rz;-5;X&W0^`13Jtw
z79i_}g`+R*8xL6@0raJ$am(rBb6Q|Q9$9Tt&i92$%$$dm@=2OM`o=FE!@uKbYiX(-
zO;$~;PH20ax;R`iM9T%ZnpHbo<j1fkmSo%mSSVnf^q+6m-WJ)W^M^Vs|J*)|f%@Oh
zlX1LlvRn<7qv4x$a@)1H=fl@r22)NHTQb{9PnYAmhh>{qAmC+O$-LtYSaSPUlA=2f
z87h1oY%VjWd`E(2CSNAp@DR}*s36dOA`6c)7P{0iefjNJOVNu`GxC1A=nDqHPU9GT
zFinHG*+-sA6)}8**9@1IJP(GC1M)Lev(>o_Y^dPQ5U!5p)gj1JLd-M~258HCz3xwc
z15ko}fc|5iV)=(4hLMfsKhHYV>Kgw8r~WbL<%?t1weyf@NXCb`Cdp{An8$PYjDrE%
zjJGIiIFfp7|BnwFQz0yy3^o!pg&1#;vllbNi#yJX-p*Hhdu!(>nmd^ECz^YCMJWLt
z4Dmqlv|90|o8FLY-5ODB%g^Qc+_oj2@Ok&^_WZGyerIoVdviD8<!bx<v2*!y@H}<%
zT>#btjPSq!M2$#_ZmI-ZR?$7s0TSgoxErqqMet8GT;1~tlDl|BZ+YZmz-iEz@-j2x
zJ2cKh)kuhqXz++2U5;Dnq$`<sc50mq<tL{!y|5m7h(_FJ)fk+{I$K-khVBk!czrj+
zU>uR%28}2eeT<}%jQf}&Mm1T-y0J+{YUjETXu@t(>|I$gz-v?i7j}y2=vyOZDn)Vn
zX>k)w3x}&BmZqlIg55Lu#`1OJq>V?qsh@@ir&MAcI)?-bbg@VPyO+*2lTng95xy{s
z*g(lJyiQ4>$&pqyRox?I?_a9U6RzHBFk&{^;54KFKRaHC#dq)4*>G7W`OfjmhIH=y
zpn$o0{lrlU=g!$H*nwS>J!_pqzfNjuZF@CztVaZNMNRscC%7fF?hW{?$+yh3wLOe6
zoM9R;(yFN$5EQ`y4-#VqqAr3%nm<tNwWblbF=PmsGNwh)hJJe44xvln=i1ADEKWs)
zADK9fLqo`Lmiw0#&{>{KE;Xy5jJzjgW=}8sPYct=ssu+{<JDkLRDCgNiR#k{kQ=Zw
zAS?;>xK36iIyx7#5*s|h#X+_g-bt+6vK^~rt*<>if75I>D#!qBB>G80k6(#tXGSY$
zsAlQCNrh8rgeMF;VD@Ho!HgQ`Lu#<G`?vhk`Atdb%z-9V7l#l<*OuS>k@f%ZDuz)_
zLv0ir@LDmxwL48DA)2}p_p$4FG9D;2u*hTAam4Qk(WFO1^jDAFz_N&BY3ZdgzW5tu
zW8Pp}YN^Q8`m-DVc*cMgNwYu$vpNS*AhU7=)ew`X;{sD=+eNkV5QM*Zjs!^fXeR;0
zhsweD8BBibWp5mm7q?&c0Gbb?yAMvp{>b1ZF!!^uQH$mrX*f>bSzYoGMvz^qGYfSu
z06{4DGe%-Hg=h~@xPh13n8BY{F=In?gCx&dU(@4!U^*BKzK%l^W068jIM0rgGl9<|
zVbQWEbNXQ82P1~CB@Ap)#fJRjAi};?m^<DhT>}>B>^DGVJNqoqIj#b0EyyK0HHc^{
zGQ4LbtVNdj5;hm>^NO8Zs20i-Ij04hX1F^!_E=B2kqxs;-){Aoz0GqWidzGDZ{stK
zOFl-b)PiWjV1as9I(*-%L-RekoG(f$7(+6RR1#-sG3nco5Ii{NcU_{x;f<^A`F0Vo
zo@`X{d3UjS@-C4X*}r9g_9!friA!3a5iW)5NMJ~h(v+nsSucadV=mr9_Pp0H*xvwZ
z%E&E|ymi1c8Ji>#>*R-p&5xuEtxm_n7V2shv-mYZsuvbVRE^R!)LIJcY3}Q?<uS?R
zaR_iS1{#FTod&p&$yhBQ{b*piJx@5_Vtz!QvQFL9l_k^c?in01mqu&(Mzu|ui9yuH
zbNNQ=LP0wOiBB<Vb_&a>BerVIz*f#ahH5e@yRky$GWRtS0->o%k4hGbP1Vzr1lfVD
z;r&Xt91VZcspJ65JM-u(w#({jAGzr7*k&{6W+{=J8LO}KBw;tf>!Mmgz{1p_(Aqn=
zrP=~$mRSQs>*9;^virs4VX*|8mA}$kSexL>nE3ISy_JR@O4MCzL-~|pwnHM%J<F>&
zQE7S2Q@+rS_xkcr7+B5H1YARP{^APmm2lzoRqd+0uU&CfN&y@aiMwAkfS!$B9%!tF
zE&j#K*Yn_VW=_2Kg^OdI6=8oRNUp}$HIHmNtH|3MlImhzt}f}Jq;D{j1u;ciD;){m
z2!|}aXUw^w@fy!Z4z|e{rC$TKw*=TERCb^oIeNQ72!DffhWZ7J(xZwXqz~Z@e5=J`
z0W2or7NVv0tu3op=_g0l3k)LdA0-ZYDmY;eMMb!Fd4hx_gYNSpka7V2`08BlQPjz>
znbRTC#a+~`Qnz&}%L|XNbz0B8+z}!^`=dOd&UX*r+ZFwwT&#IYlHVOZX@_l=S|00=
z<%Y2=TGY(NrmA{v;l>5tlhMOlo;P=|_DD(a7q?vdv}TqqwhDMPbh(F@T57NVSMOr`
z-xtY@jO_opcV(+>#{Tp!pWRx7xv&X=sd$ufDStle*3K(18%5)o9%JYrt9dph9heAR
z^3OLoQBPW`2Bl&8W1AUH<}Az&jx`XDm&fyy;pGc%G*HY5LoDzp7ldIzGsb~v@F<=a
z3?ykYKl+`X9^&AK*o0uhj3R)A?2gx~x7p#Vu=OlQoJ5rAW1f-yLBo`~+qZ0;9*>tv
zxa(`bUYEjW{I51PGqlhHC~lR}n}8<H2*$_;eOcsW(|Qx%qGYcTKrs!|;C+7({rG*7
zJ%Wfj8uer`E1i^vg*<-~17>0ib$m^3t<@M*GROKwp~{X1>=4W53O3gT%XTN?$69AF
z728zFa_r`Y10U*pKzr(m?Pvfze~*d?r75Q55{o`Yamp>x(ibay{u$2D7a(;56F}!3
zh}%JU(4pvp@H(gZ29DBj0BL$6hRw1QbfD^q)cT6P;svoXskmrQc2pPLDZ*<MSFWdb
zNefaatKNp>*kZsQ_C^m8g7|4uKcvQr%AaTq)7^)%t|9^mCG?K)LX5=ZzZfP7otGB{
z#L_)Nfx{?c(Tyo}SM<NEjB9j`YxR+m>T0&mW!au+hICys$?!P!omXMWhb%$SWfK|P
z0HuU<v@$c9a<SQ^Gw>QwuG1Mq4JWEk!D%b-JCmaas}p0En>Kvwr5-oD$+MTLN_S}1
zS)vDBzgeEUgvq)T+<9*?VpW@6+h!Z)htZ>!t`8;C!vo~R=*}!+q-7B+mqU>fji}|=
ztxWocM8;6ZC~HbI{gvD7`<q;7!6aEjGGmB04H*K%+u-iQIln|}hI{51NQl$>6o<y7
z#D$Cm$)=2@O5b}bFPF~P^1Xn*uQEQYp`>!)qRXA?Kr%>A_*|RmQ*FTKV^1$F3gnM|
zp?Hkc($y;jD!_0Hw*;9_d|qQ4GYbFkVKr8BrhIe^W%8rTXE!3vUVF`n7w5}v3iaGZ
zqa<Zrrj)Q4q~!y3pn1lWzUq7(vEVu2;WrNe4HwF_g0N1Jbsi9uf)CbCmb6r||6Wo2
zR&=K_I9@Mbsh~O3)FKtyL@my4+A?j>@j)Bb<>C)+v?;F2k}+}9#(RQhlw@{?*O6Y9
z@OtEa%VdtLf`yE2EOLVHa1AuS-1){gy~oH2N7GzDdJ~53cG8HHJLYA+;9sY=q(_v7
z476_7uc*kUythueZ{f~Y)0#!o%cH%}M%y_fUsrzYp;%hD&MplR_trYeak$V7#Eg3=
zKtv?rO^?Y9XUIBbkK!Zje~BJIXIAsoWnt_QS*g{GyG&G21OGa+3pL1VdFOQr4GK7x
zSH-SRcva2-oDl2VJTQX*vie4rQP}MlYB(jd<`OVT8vY`7?Ig4uARjd6k;ojsPM@jw
z-l$z&sGYh&B+hh%i98sW2Y>mcg8y-Nyog-8wiL6lxd6P8QaSD!J^jLF#um`I%f*{w
zS-VLYcz|B$kTb-Rtmv(K>{!~GWamTh@*-XwmPlQ%F+rK=B{A7dCHsZUB@^>1XYwuM
zV+VYzpOTxr5`>ebjB#8vQM&O?FHe_e70uh;8ddoXu!XiU_m3*~PZlaC>wm9u>c8SH
zeyDoSwG0gCn01(Rf6xfRrY4y%rlbOS%p4`aV4G!HS>s9LC|pjCR^=2##}m5%1K_rz
zhBf(oQFOZKPXD}@_S)t2yu>oL)D2P9pk;mvm4!&7=9g<*yGHFInL?0HHY>BdzU&`_
z9c4&mc)z_J$bG*(nOS?kJlIuzyi6XK>ak6JTYtR{yOm>1N|WX*_NP!JMlzg5h|vmw
zR-lK<o4BO)T%XjolZ{YmKeiraU`i(W=(Zhc3SLpu=xXbz+9(_vr?O(c*49oC;EB{?
z5^tRv_@rg0dN6Cu)W2j)zQt^s*~|<4mLW-t2<D=8TmCx|EGv4mW!n+fJcVVY^yuDL
z-l4LYwQ;;-+Isc(vhkqYKF9iVlNYj+$BUMF_i*!sdtj!K4)nadZ{y191Rhy1dLh#+
zHw2n=C0IWn8f}2_C^?Nq$2zud7+&Q-s41d|(^#ggc>KNp_nI2L%tDHZZja$RMCRee
zu{?eHFYOj+TgcDSnnKU&#=_B@XW<^+ytWl?WvIdhD$+pJs?~5PqTuujl}?>bm)Ra(
zKCZYX5eH74?%MM-gj+;P^v4UsvhR=2#XqgGqrzXF>V?~a1vnbvSVW-01|eQfqCBRm
zW}^F%AG1~CKd6MFp!CX}`1t}8FAz7Q?v3Z5?yG|QKCb!j8xO?iaVR~5E7Hh`Nt=1N
z<C#fa_3QVXuJOhKZWwNS5p_|HMfZ3c7oLs=L2FKsn`NnNG7wtp*R=tJFn^`F)#3fH
z3iG%l_x#>JXkrvcL0aJr^V~$9CPA*4fZEDH=QRBgo}&Ro&oQHxp<;jtr-`*AVz)!{
zta)QlsrZ%IS`9^gE7oabu>R3ubsT|^+}H{A6&XSamP8zuTspRkAP7)YWHMi?rY688
zu7cde^VM^?sSKS?RdxW|N;3xz2A5SFG#pVQ{V;0#dow?{&E7U0`UfF3d&*O(av@Q-
zn(;gPV`T&Rq7LX@@qYM4Vg%R6-F}CV%n&E?sZJa8vo_;6QZSBaTln?zGg&mTgDUsR
z{@YS}`#v~^@H{Rokhg98Y|*!=#smK&93b(%vp~5huep4Jj$!BN*Sk9zRbB8n1L2gC
zU?|St^1p{q5Lj@KN<i*Y@U&FU>X_PYb1b!i@c@IRMS}J-E(2t@EDOxSH1|j9a0*HH
zw4$*tM<l1H+3w89H=-{mDG)bCP6IA*(OK4|5R;JMuHG+=5L?U(>OFYD9jrG4M6|;o
zfs+cB-wZBc-VS=e-DGz$Jq<J+vE8Gyj*e23)VqhhC<xL(Nfv`E$9f_c@9(KLZMVxk
zLIEuDj<DVCUUT6pWMD;rL*eA<bD>kSpIjcHVsyTa&(#3r$~uuMatAN5cY?i5EQ5mt
zz}x?bC<c|jY#0?7Px1{{%xKP`G8=_dKc~@u&F%l;yf0agvfw27nhsB<M6^ejc6P7_
z4J2&|tDyR*PEN;@K=?gVl=tk3(l$hM%W1&Ig~%yt-zhrf-yj}bgE+9z2~gOl1;~0E
z+c-!k&ud&S7d#Lx|2Wx6#sK7Amv-)CU5lR%4@ntr&?s8jxv36~*0_l3R$x+x0=|(j
zvtr^t#|~gPg$2D;AtcGl-U#4n2ezTu!xj)=p%0EH6>g(;9S<zf*w}Lo8Fg$QqAwp(
zz88!TO18D&&*s137Dof{D<=EL4-#CL0dA@q5S(5fTqY=x)M={%BJbVMFpk%zA}_m;
zK0tUV?Wb_YTzQVg+oMy@ohni>)({G<BOfGyPL&-oNLp6U3t(X`eHyyg_ss*>1yCqh
z2YwH(X@E(E4(25Q`<RpLqBX_N>NfWq66nEU#*^rNAnC)Tinj$KA$T}(RB?zYrZ~Dd
zvN*O_16H?dLKgG-UZ!hgFI&0$eIE^4&CR=Q?vS@K?h5jvlgq1jSWmo-wtpDr#Y)nM
z>??AXNMHkSKoPjAdy?t2TK2PWPc@GTq6Gj3xt`C<X|`-v?J__!i^_1HkRP#5NEA#J
ztTwRWl7oaLG)Nd1en*CHafYFECbNWJOGzif9_%nplZ`+l&60T6f8v#`f%jQJ>ZEPD
z<%|L~+Vp)4{Yrc4g%cpV4_kN$x+GEV_Qj!*mD~Gv<Vr4~Ffe*+*T-r=0ie8TCAFng
z&AK*^%K1CD*iUK*J0f<C@p|ibwcY;o9VWI3>aJGl3l|2SzQ?*$)Cg8Ig#`=I8E?N>
zt=kC7FPX#dZ+kQTi7|BJKXSI2;E30B)=UxwOQ+=`?u}1i`^kI+z%)Gc`};fz{!8#n
z6k9#*erERW4YR6Ydp>p4sppP5&o@A9F^EK?h(t-PRp#5ptV;T3j%49FvOyQ>jp;{f
zgRbE-z@#^R<k380N&_tT<nZjj&b}Zf9I`>U-E;_#c~ymdH%Me-cyKLs=mT-smEsTH
zXx^n4gUV*`8Mf}pZON8!GqOECK28Emh9sQ63Kv96CU0A3NQ>xbNBtn<ZX(Zk>s3Wc
zLoZOH@O1>^P|7v>s}n8rgjRiHVJG`K0vTIPm8&r9wTX`iMu5bZSBKMHujMC{J6O%A
z^S}rc#ThdwXq8wIogBvO*yX;#oit;q|JCu?|FPGw{dX&Hx4Kr`77L2+O)W-5YBbBi
z244@+-?}C(O%Tv#VqE~0vsgEFilnMU>DJo|&e-gzL6)fCAti6#)En<nKDg~Io73Y?
zvt7I#UO!TOqX-pk8@N_%d2C9Qd?8!7(dk;uLc{m>{r8hv45LO9Y3PAwLBh`p^x9fq
zKJDJk>-Hi{X5-G(NYfj5E!u|4R)d$}cRitl<xKn<ql;2Ee%)7A@Y(~+7}b2H3ihAI
z->lYx-*olU_$%_9YQt9~CvZtuuwY+~^qWn%dltDnl!h3r+VI(w881}wUU-c-^2jk2
z?{tGfFR=Rpel;3#pPJ<HT&dWTro&07^8pV)vjiWg0~GLOYcGR7XZ+;LRd04c#(bpl
zJA0@tsl@-*pM^N0h__9H51K1e*__@FcGSzsEa=$07M#3fw39sv+jgZd8}IN!7u)=J
zknb3AL#cQ@*mAW=OHXmW&ZqrybdYh;p32o;MPbL_F)?vnCT*m0YTFSW14AZF2(=4d
zI`xM!jV<xed(ZA+%sA$LZ1>`DyQ(V+uPzXZmIX}rSI%mK3~wRHA4@G0iqIcH9s#Eb
zKBRsKTe!3o>bY^N;Y3y0X*UvrNwW5g3lCB4n{Mra#E*OcG8rb-9Nwmn`lzY7V|vg&
z^As1V?Sb7bcstBsi{v}9;cj9;$?er;j6t$c$nWkWHf~H0EMjoz#|X5*4Rk(A2*J?U
zwa<MA2$>OIReo%zf!<YFFX!uEnp#qMH8bB{Qb|QuNlprrivqz7<~fv!wH`<#4&uS>
z1M7LbNPUaDX5&xQ(N{|Z9^;m_#Jhxh7^0OlmdLcHdYo^S6wDQHD1F**J-a3bI|$O6
z<r1SEFkXn+@ir-J)_p`^+=WDhrp2$H@A%53*ZC9p93YKSr;n0|3rnESpv$b+CXgvF
z&~U^q`PxCPtyxgsl}D*j%!23%=yW6Tzy$)UgSkzw<FdLP)<XW5Asq)6gA;VfkSca9
zOym~xb9~!*(1WAMPq)RTi+79^58{w<Rd55_PiU`rUt{gi=0t{-hT3-Uwh(&P)T+Tr
z8>?zGl=}Mm^Q?-{%I^B2!}{PQ4QLIM<0b^PyVjSX&Mw3SAQU5g!_!&iBBvlbOiwZM
zI$AeouEZSxK_5YN<~o_T)slzNN6w74fLHdnDE5fTQ~fpXASxgZPR2F(E_8nG?CziX
zTX>^eNi=H%#KOvCu(~0F?L=NVB`gwB>2s+HMR$9w1r^~Lx?dCTQGR<q*y9}B-nl*Y
zU!{|w%p;_F>pq`)wW$x+e<Va1T+uyGiTv@r(l&sgYuA#76y=rTG$8BE!oj9TeW0Gx
z9f}Entg3$n)mf@Xq+&B`o}>CHC9yC`HYy<IG`O<(-moiior>Bz7=d3rrz!(fsAa(~
zP~>~whC2!Uq<RZ9OT+O&6kr~if-ZwPMx^KQbiT5Lu~uVv!aPTe16$TVV0medjk_2(
z1!E#aE7fP#Ly%PEXMi{bXj>xdGNf^vE=-2t49W^put9gG46+fx1pCjso7B1MLW2?K
zl`UJ9k!498WD=#SQ}*dSA=?Scr(wvFo(5=(XbhU>7-j<ym%!j72-WCkkoEJCs$orn
z+DFK^n-`!eQ8Z;_kx)1|gL;5yuCYod42dnMNH1z<m-YqP|8iE^Q>6au$~-bfQC)^s
zc%TXicEO{+0M#tEb10%=Y?+8o35ot!rcZi$ND~HMhgfRPvc#0BSkW4G@i1CyHkL(c
zu}K0mHIyBcu33UGn<MvQfvD%W_oL`kZu@~l3kPp;ey;S02sU(GqRtO-7NvV>rJlST
zAiOzMS<n4^kK;Hc+T?Px<K6`No>X}!WaXf{GPIt9vs*@^O3a~-bJV2%V^DR+I(__l
zW<U0d7O;X&IRGa<1X}CvuEth==`3^ePb0VQX|A|jJF&OHFfwNk+2KLux)y#^r>Kq*
zf@yg3vys9`$GeetY;NRA@ABBhE)1H`1JljjQEl|g)5W(~bliKroAur3(-}D>6+$2d
zre0APA6DRKE^z)8^Z--rEMC^yMd(rP-QDFM{g+{;T+y_O30&b)mu~fNv?rs{7p3Cd
zINSIuJd9agV$!tftzoR&x7$m*uxvZde{>`^rhjxK7S{iq)?2J5^*^^nJvTpJ+M>|v
zFcXPL83o|3$SgET%;#J-gj*OeGM}L4pbMI0-!Hfq<QqsfGcUpT!i&c3bn(qXmo~kY
z-XG6XmzSejqZ#E8QSauouDS>wiXb3S7qo1;XZfNa$ycQ?6~70k!!j>OQ<r)_Uk68@
z?Y@S_r?2!lUJe$|4@#FG`73wzx={Nulq7^qAcznXCtMd%T{PXJ<*D=~c#lC7$I`rz
z4w>@1RKURDz(Cg)tP{B<+t&+F#E9v^7!Bqk4d8@8gF~6ayT(^k%}g>5-_@cQm^EXY
z^%;o`5M&sgdqjB^8!PdZ{2RlEoM6x2Nxnms=`bP=(&w;DR-R`dwHg;?Hfo@b_)nV{
zY0sC!v=cZGln7iqXc6jUWLs&$9F|$5d>W~5y_N?08I}hDjLJZY=Gr&BubPm3FU&9`
z@VjhjYApPtxH!_RU8)_#$%{DDvuq-}R(`?N5BQk#e9VX5W(WdMWf@86i>q48L|PTs
zIGb8-5zUFBHS<~0_>&|$Ewwe*GtZt*cNb<HK+PZtEV?Kj8VhZGZQlEbQdt^Fc9Syr
zhh0O_8X<*5HiD8SKpBzgBQz2y8}{#sqeqG`rCc?@kf;cjT0;*&fb$VZK!l}3=qCZ8
zZ~;PcA)v_*8>BN60=XiJjHx1%3Z`%c-olX(!2~x=@~8&rOQ;Wv2ip)5l2e4nS6Wei
zl5lrWf!mp8H`3MdVu=E7iKs*?yeJPO#4fk2TSqr{k`2F>`0J;kIWqRZT+OgaV^|#F
zQ-voDJ8V~i9Mns?ciCrprWKS=mhkOEpprzATCy9rvwas`bD3ZC#Ilq~Mr%>Ij$XCP
zIsk<llew{XbfVDbij*Pvd@Yi+DD<`&XDNQmR7gQXI${nO|294*g`;fK{WFKvrmq#1
z#7%X!+cA+M13B2(1Zci2d<%`xdOhF`FXjwrq_V`t3m7gYmiu=pI4iI$jWf`Z=b3B%
zy{pcpA^kTs`h$bK9Asy*KZt)I$dZR2tMMPqyEII?pmmMA9aIZVoIV+PGgSG$;iwY!
zi`aXO?2oSGz4G0=BH0-x%VZI5-vG<c=HSjl*)8MEP0h~CyiVfx<8Z*pR@ci+C&=2L
zcO4KpuJUE(CmEDXk{9}fbtMdWk?axh7<|u8siRE!O<yNR!u9iP%`WTzF!oNtnRQXy
zZfx6j$F^-JPi)&xI<{@wcE@%)cG9u!WaqE_RqccSt=eDJI$bBLR?Ruaoa4T(@vvey
zu3oA*fR)ymCsc^$c&}l@TYt6E$S?MJ=b6;6U3Y*xPX$xX_Uzrlv?;<xQkamSFPDrv
zm;mvv;DcNOJ7D&(SfTJaqG#-V*oEN<Nic#izvl|tXX=#^JM!~hha=$Wa}aCc5zHZ~
zPsNN;eW4)s0l~_teq$M3(@rPOf`gmQm)^Tk57Y@J@qMm$5s<hnmiy55EW6{R`V}^G
z^{Yn?qCalDe+K3aTJk{zl2Dj2qgYgNMU9&V++Y?nNeho>H6If<y86)BI*17tG_8H=
z;)BfPGa|u2^olv)+~`&5G(2zt7)|$3+Ao4x%gf)ldJ7R8IdOKTcGg{>r<*D;*THS1
zhe6XG5!W;7aU1882LT1QD=cP<g`3e45w~MhDwAFC1<4qX7nBX1&`9L!Bh{YhC34so
zWGn9xqIqyIE8YMn@PS&yH8HdrJM9-sm#+lR$(XUA_EgB4&3pUuEC^0pvT2XJzHADn
zB1h6=!$MMwFQ++dcRLm?%*eB=tIWv?blLnvOML42qp%35&a$=kIgVK?dx_i^3uE;!
z<JOxMzUX^bE0>O{NIu!t@t}&QJ_xKVywLB~u4}m(%tM?%-XHu;ac`RFIme+mcg^CJ
z(@6cSB4gpfyqRQBr{QeK<j45Dw3;95opf{)qt(B3`x92<!~8PNz;dIwX0itsBpC6Y
zlT-clb~#v9UB;4VxcqP>$4Y_Ea{o~VU%)i-eQWtNb-odfm)QQ-MGO0X4PCOc{jb?X
zhnka48|-LX@4woM&1k?B5>IBPSEaaZlcZSJ$nX(Gg)poWRK+Hw?;blmq{!&TajLp%
zCHFlYt~wBrB%YCDw{<?*JG@*c0b~iX7ASH@$TB4TsKFiK6ip{GvJT8|-goa++QO3N
zSW;YtmJm^^N_9JW{^^@Ocrg`yOyhGFv?zG~nbFDw`rn6{`u=YhdUiXXjt4v2{vY@2
z(HZG{we=Eg0isBybFrjG^Zj?hlNQPYy&N?9(Ie61w_c)+?c14%hWrbfbv9XcE=%lo
zirUFyuGb@IOWk^Dm2_tlu1Akf75D11!hfvQzg{?Fm;;g-VaE>&0D)sjf30+9qe|TF
zV{?N`!;QA{)|xAq$ah_lfD9l9;~{AkAryFSdp-W995m~;8&_d#{zk_BW34MO-+o+a
zmuV!rPc8Vsl1#Ypw}jtQX9@`w;D6l|E5jusT>P2Qb>L+J`*=dUv~=g~aWBVcbGgxL
z^m`9z_ydSeXFxo;6Mj9URmG!wNgnm;7wxR^r`$2FwevmOC&AmNpp2Q-STJ9NwMq)1
z?%>_|NDt%*FFnU(t0tFSh?uU0z}POtoS$ic*&gHtNJO+aFKK?%)LvV86DQ+be<Yv!
z=0=_M=LnIw8*DUG>O5AmIJqMN58wql|4k0y30i*THo-B|&v0gYZ#^qUI)I!|p`Y9-
z_D7ipB9?a!c{OqjiIql;KpCD~@&fX6U-05)T}mGN-r-+RrW^|yC*mX?CN-p+DAuxL
zU#~gXafTJHppCgwAe~onG_;~Tb_ix9)*eG4xGdCB*DtfMKC|$o3oH)r1H`@O66YV?
z8rDDrZyYmyp40L%h=*VVQZDQhGyR!5-8I}|*dQ#zwuoRHH3l-}97E5$rk|w7`C)T8
z9_-NiY*<f98|SU)rOjrXB9c>OY5M6RErfq49Q4el1Xy-#jf}8KC}+xDuX5~-99Oo>
z=(w9wda=10;BazU*4y^TD#JdNzbGTrcK+=*zgb7A*MV9AyR9v=!4<zbQkEdUGoedm
ztAq99G>jw6;sr^Fh~7r!<3+k|!{Ovc3d|5H0TVTn43F96#8(AzgWS&YL|F}3pAtxy
z`-%Yl2R(BpRI*tkhcgntS=J-J7FhGuP3Yblqc1rYIno;d!jDSWO)Z96kF4P~Rtjb~
z;L|1I6qz~R4c?S*(QjzGw)7xf_i{DY;)30J1-UPaeQ8n^NChN=c9`mrf05wtwyE%a
zW(S#1HNF5P2$WQqFe;K!#pbUVN<B+nXz8G~eAZm<l8l+-<Wv%~>W!^klgs8o@g%gL
zAwahhh^*`4qu;A(wW1IZXo_7d<@eD4_`7l|<VtD$q1NFZ5uRW!%yi7z8Thn={t{F5
zPA%kndn$>JV^)@51tfj!x!_AviF1Ooh}2FJd4sVu4P|9GqG)6}RmR{toOL6tpPdRg
za9QXS1x8Y7TvZyp(X4ECf|my!bVGJNXY)R~pGzv<-b{tX7tKhAL|+2S0j@<m>XwTc
z6pJGej+k~-U;PL?ju+Z+pV=AKx3eu!s%hZgMi^Ros&zZo9|_?=HAxbg!@eSzt3AG)
z|J0s!En+ATQ)oZ0tw`l?6RobWLddzt%62}vy2ty{H!!H_#yqnlbUtTnUH_n>vFVY|
z;N5eLeR`(&;J8=a{mg^svPaXT>K(>mxR%k3C@RMgy!CkI%s`CqWgRmLljm5XWc<pB
z!hBozltElG%d%k19#;;^R4&{i^P`%O{42Eb*a9kO6OoF*Br@U^pGp)F9gYFUpqUWx
zNjXCKmlfHk+eEV(f{6p$pVRUhg1}`F1tY3G6+I~7T0!_wIB1n)@#AmqNiyf58f^}^
z;9n(3e}N#@R`K_^1F`Ng)!Lhx|8hKFjtOIpUTYi?&d0^dfTS;#^1b@QC`*~q?1Rma
z#Yuv(Bh`PMcfLh3^<3*W>>2ITWA|XcU@yIr?+o}dycoa0J|(*HpGeb15%wdMys0N=
zE%?$#LH5vMxL)$Og0lWDP*5QZ$5xz0RQdK4g$mwoVW~uIt2pT2?3{S~9>d`!zI~Lf
zS&xOA1s#>AoA_<y%!9?~GnaZy@SH^B{+(-P-2(p|y*WSGkD&0i5lom^qlc-OkzGg8
zs*=Q6UCPn9aK4rMZHFcr)e0pwF<#E4EwN@QjyRh&zNC-|aklawEeywhv@p!<|3k?A
zUjzqy;|1)47}18@`G;{0`(D;o`;C8n6{l!v4QC`GNODq(6u^jozJ+i~{w7}Z<lMQw
zQsM*+8Sq}#OC}|uDkMwFSLh=T*F?dAonyoEkqsvkq9S`-Oe@>(*)$az1AE+I?x9n8
zrhk4ju5<5C^NYDl-x(baR3w(jlqt#@|2MDw8$kRI4|)PjxRp2}amLR?miEH|3cTd&
zSh1zGA?6hS4-c2EmNh*vMeph5zJBsAh;UKp{!%Ssr$jQ;7}S}y&%fvhr(-f%XPa(4
zj>&|^f4bhp2`;V;-qCFw1F=IA8A}KdH!>>;Fm;ORL^}VJ#Kdw%J46#6qoq~a$#&7O
zTbK6ut6RCl{@17dkMe+-iT!_48UO!Ts}#+1`-2hWZ$E$VJq2b1+b1!n3_q(NosIks
z46b7l2TZW$=G0Zi;(;(<tNHI97M?CNsl|o0(E96)HO~*XP~LSh-nVlee(P_i@2jlm
zpAqqi;&4}&Z>$Qk1qx%wRPpDHlSOHhi>H-$!cfxrvX)(~TECZ<S;RJL?E9tnPOg9W
zkA&x;+&lQKgk9YOMLXRqG!vSsV_?`0six5J`w0$H`@!T8>G%NvPH?gzD^zY@>k3p4
zn12VJk$o!Fk!P##9k$ztO-o^&8!Fd|_O@I*X&TyRc=r?nt2691T^?juX{3%VW;AMK
zS8<|}5mTrGdh^N~n~^T2_P?2t)Dwd|nr|EmTQ`2*nK5B^U2>bQ0X$|QfGI3RZq*uR
zL<4fbiyg1HI)G2C&c|2Cz9ikqvEx`0d&(ng$0qj$QsIt@`{lC;V@n>*9<=tV-d52e
zSK*jXm;RED1%XZOYjT51SXZPgJ2?<F><Z)g6gPNksPgPPxRuXlPcF2=_Oi5!O(mjb
z$}-czZ=I%Xz>!_<ePvKQJL1p90f2wW7%%X+gp)q~JWFS)&vMq25?ZI)l)L{nYZ>~m
z#7wnu1XdgU#FyRMtZVuf=U^R)%Z?P2z+kV6;mn0CPza;HO{xae@z;5^`Hm^i0;qL)
zdZA;TtxgCUSO@tWmbv)dbY3{Uz{P={LOBP4xi8Wu>j@cT-GclvMwObPLcbI&Jo*>M
z0$foJt64fV5}ECjk4sN3?CI*?FL!KpeUF&J^3przhNd!%GUN@1x#o=ZZ95BV$GBqU
zAs<oG`?@Gb7Keg9FZ~LYl!ShC?6}@*)+@QEo9>5=*SS3{d}_L1!_V)7#P;$Ud@EEI
z#fBue^C#maBLI=onO4pRpb%IO#OQ}<{yMi*imJ{@C5~Osh2p9jF~st0GlL~d2(5uV
zjY}wZ;n<z;m#sWY0UYZr9x;3KHVNJZoj5QzltOdT`tt->sCG(5hvy$=?aM}+uAq&T
ze{Cc@5T3}tMACGDLFc{MH$&?zlK3jZPaU9{-f_Ne_s8p2t7{9zKtlRDq6xWrX0wXm
z&|etVLL43wliRI3@=L_5=}11}SQIQm{P%zfk4{+hB+mtCI3kPUh>VBMSI)F7?^9kd
z?;E0qpt!WBhAbf$m5V1oyG4$}aRG0+-q$U+e~FW~Foi6&6wIH`{&n4(djyfNsb)zm
z?sbwBBZ)2k53gA8=2lYVlYQ=ZJ+O(tre)1Q1byCIsdqVLlIjoazDCwSIduRUGDE0?
zE;NJl0(5h+8O$sTLFf<!Svd6u5q>yL9<=MGZ!9kWnLb+EeQSP>fvo%u%_dQA?N8F7
z!2Q!*+ek`&c1L!Gy-&-EZv)Gv5=O?*=NV(A?H^QJzCq*C=QctI)Wytbkevdl0IfdE
z%_TV@eA*n!+*+8)%7dGbw{pFv3CpK^g;gpxQ#XU>8vTM)-}ts*+1cbBy5m3V<*8jQ
zqVue1Y*<vO3hZY~(16wjYToGheAi)A8cp1xUyQQX67<v|a$N8mAj_4yry4n5mV@`b
zs}~hA=0wWGwtYShoOuxEzpq8;d66!0#!^0X`0zAgoNxA2-x%xO1_zb3prwC%BTgh>
z&MnENknU2ioK@pPU!y}y_{ug$wOU*L%S9xpV{8j)?s#f~ZxOjV693A%GkJ;+x2Hms
z0KtM<Dy`Oqh(*0}4T{<5^r!kP5P0isy3~~fAC_eqfTAdVwx4vnP?jwY#m<qN-Q-<<
zoOb$Rx<SF)=CCb=#2zNEgq?Z<n%hYAj6zS6KY0vEt%ThS7$=!obxR_odxU;7aBOFk
z?SxfvSH!xU;F~v>acl|zfYh%#=;IwBM2B`p#DxsK77jM!7f{Iv+aY%8-~k!5uof?F
zmUXNOLH4Eew{vnOqKCOm8?^mYB>iWyx<3Oh%0(b>n?4xQ2X)`LA9QQK#MePzczFyS
z<BqbMr+~^rZLp!~3y^k)160L$ysKvXE(vPqc$USMLTEGBdp_?-WK)cwgSRwHgl=V>
zVdMj*bB_2q@t`cXIZq4t$-i1d?#Ao}!+`bqB7uu_mMu7je=AdSnT+vkJZP`OrvkuB
zOK>xx-$M)Y;->IGhoGQet8c}o`8v4O0pu?j5d}>Nn~e4PI@Fh&n@BT#w0m2JEAmzd
z1$+n83y19lU?JA^U`yQmuI>ql?RGW~*@#IioWKuIlGr<a<5VN&m|Fhelu-PsLWQr7
z*=S^kfNa|O#v`02%|Oagdqa$}UA(&zkjpokrm2#dq$C@v;(t@GV96vE1liYk1QQZ0
zUfeNlH?8Q$i2$LeWj^h5NKNr8DrMC%t)&S*-&7-Ua7=i}@<)S$&^y_XItQ6_s6!!-
z-E*W2D9(;iWWna!=0~;<et(!%P;|YXM?v&nnwh~=g~6wm%{^<>Uy9AYLCDY!(f&y_
zTEzuW$#XFR9aSRi)FgTUNof}1B)mOrA#;y*K3ZbOMoy@mh`D#7-_#mB#)G`9>iXHv
zDi-&8kgLojDp~~bFTbuItAZU19U|&AnRCjGI(Ut|i0HcwH#}@Kz!*G8-RfV_++_2w
zDgH?+W?agl(6`0!GIcZ_R4uPeGyPp0kQnrkt?~kPB}^NMW<(Z2`C$L0$G5HXg-6O|
z&OM)G>~3#{z!3`h1#bT_;nhhJu#aQ3K?|+Ofhr_Or|a7uE)n6BM?cb>jb&{PH)cx4
zA}pdKh##_pugg8B`)ZU@s%S%&Aa6i}zc7qN(1I9rK*QyzgH`s0N=wW80&jA>CKJn}
zYe(q|6OnU{BawYfktONc)N5xC4ykHHaeP&-3E1Tyi0s#gLvsNdhfcQRcSqioW%u|0
zzWRRrBNG09S27$wAkhC0531_vU`oWGU}UA@Vhh6{PsGH?_<w2{CubKTPEJ-B23b=(
za~BIDCNAdxwLEuPOE+$#6~*_tZrAwTAd*nT$)kZPzEn~>A3Tx_{HjL?sa%iE(OO#@
zVAuKi$VY0AyAi3;dQs@BX+Z@a#xlq<i0!4~GEt7!)HqX_T<yeeG&KPp)}WD#k}HHf
z&PgL3ZLZwk*odbF#q37&k5!7gDxElvI0|&T$tYp6xN*$1tortj#7HQO2-u(A3km5s
z25D|M2@(+^_mga#5~bm#N$B51HB}%Bbym2bAkhyROjJ;x-C%|`4==hgJtfHrVMis@
zbI5Z5B9A1xHU1(|D*aF>9tkWk1ZsjwRgNnt{EmX!-T)!1y1PZafsM!n1A;*B_rLOP
z=iLT}f@d(2zBKwE2O<*IU`XE+v)^1sl&eCqvMtz7SOn%#E$N4-W`}4L7m^85bC^#_
z;4JQ7&<SW<zoq~Qn9ljY5iHm^@tB7_N@yk+t-ndKrwN++I1R7SOqB+WwQbSt4fS){
zhdhyT@Q6hTnQ1(3o6ONo+(oyLa;9L2>dT^`iLnI|VxXl;QTZc0Lf42zz(<ZkXd}U^
z3L{9u)X*Ruh|u|<5e=5;kh=r(82PY}=H)m=(NzL)0t}N^qJ>O$qeyQIcX2~$#wYy5
z^7}zn;I1jNLVzl>%Yoq&!>q9j;>V66Z1hN=P}3<Ugn>@`jcdXc>Lr-+lM1Qg1ce+p
z`!S(*qmqJuV+!XRhY$&+<ZN!rMw&lXs&Q~sdQiLw$r8cj8UF+4Q1sym91W<ZgSQ>$
zn+r69#U&75weV1a$n#~V?zKK!i>g$sQ@!27oK?MTe5+BbqkJKQ{nP!Y>ZZU)TQ0Ip
zu~4hU+G4YYnL1bF+-q^F(lrgFQ)5TH?f&lmjpI?;0}2b8x_5-xG^k_uJo!`Q1M@6}
zEJ?oM`|J9<j{`+iy=vO%HVuZ*W~b$vNkC^%_zaM~sR^6pzd1mlkeok{QNT5H3~!03
z34aS9WQ%tlYIvXFg(;fcn(*B%p6`?T7T<L^0gQdo!bVY6slhANb>1tuaA`Ka^HcNJ
zWo=ZRnq;gDYYeXptH2p-S_6C`T5J*UISo?9?}mRPwTLJ;v0LMj?Rq!77KBXgXT6>s
z?~hN*wF%#vEYF;)-4?!ly{_Cnk9W29Xn&Vcr8b>*(rI+sZDX`O+USm1bOJP{z24kd
z{?tU4cK*t80<@^HeD|O+z3F!8_`ra=xy4#?5<PhkK!=0!+b5-x8ivn$E)#hBf-#LQ
z9uuTGWQU8?h2?HaC~(8`bzl-s`5XJ-bq?>Hh)_ExJa!QBti`qPFIFrQ&icizcAcju
zJT*4>dfx+{uKq=Khg7N7Y*6Ya^Z3j`u+fDSLzCCI7Nc5GL;{lFkIQ1MN_J-bb9W+i
zb-&`3Zq+)n8P>hn7&PpcpU+#hu1q3X*u)<zYE)0SER)@+GNokZujQ7R76p<nf6hwh
zt%8I>AFJIKlts&WSjk4L&Fc(BR6aF<g`2E8X@8DNfMX9YNV&A$3JUOgW}SH@o?mR{
z#t|o-r?haQiTQ!e;an`;yo*ivQ;*57r5M}|33dm0iR8B0yD7JhZhA}jzL5hzNhOYm
zv!?aTseW=wd<`sXju)EOYkC#*boY|i0RJr(=Vql2dsQ4y)-m6HG1KCHK!BWxX-d;S
ziu~d65MX&IcQ5Hg-sWxEG*jm~f50|DG4AAG{ni0jqTLMr9zFf~>g%>1m-m8R18f+O
zi`^=sH*X(~`K2-cFd`u>p{<R8MhEBm>Tey_{M`VnUJd?<$3?En%c1C<PN~|S*4c-f
z?rL&p6gDS5oNTjJ&V9*AbX2j0?8}z3$}+Zno6f6=q}EQ~d+Q$>`nj>|_+VnC_WgnS
zs@;u-<4Izx#U@`tgm47?4`V3t5w21Uv6mA6Nw0i=^3AlSel4%MY6c9y`CV04F2OeC
z%r{tr1xH)8iW8BSJVxH9nSk=5h>1H<r}Zr0(gk~inj=dbl0XHY;i%JBqsO+=%C(Kk
zJVj5<b}t%!xKV3)`qM>ednL<F>zh;84&&0ti$`C*imf;$%5shM@bitR;BAk<JMTE*
z?J{6?xEwSm%^o4E;uqeB$@^`t#OS;6<-%(_s}c_xp*cK_E|{drB@)p>5t&mQ2+LVJ
ztn95oj$+iJ(50-!Un+z<sWHhqAAtRb;jDzc*@Z2|{Hk-vz@gfvt35wgd#-PPH25#u
zewsfXE(#NAk8k7WCvE7tLQtO>GvlVVRC&2G|Jnhx@lynud!Nkq|9&FLIoX?nL@t`i
z(NGmPZB(vq+2%!u$0XM!x?y+x#<a(U&F5}JQ(K1Vklu&CswM}Si}`H^E$iW58EHkj
zd1>QT+jhbU-7Z#6#*734im=ZkEW%LVepU^_pm((|iI;Sxz|Z@>h);gvk@+j`J$cg7
zpm*Og_<N$f7^%tY9T6XDtJjrK>pj<MZ56^0OIDagXe!yK-Lo0%j-x-jVMFt!&&Vk=
zB7bL~GYRBp&}toJu8Kx~Q=6j&GVJmKX+=Y)>8{MAwh40(*4jipPj}aBk8N-W)SNF3
zGCe^y6ZGo--&;N>=YP;Lm|2-Q|F<;49Dug{W;?pyY#r8%)QGswHp9Cj5nW~>>=3NW
zRw3*tVk6z?*-Z3i0^TUy+s^wfNldwgSvId!p=9o$^XrU6Vl98o$r*$Fm&fbT=yP;K
z3h7Uvsk2)jx<^?Vm1gPe&dybhQG%0;A*FP9_Rr7!k)D%Sl)isgJ1<A0$MfvN=QASV
z_UENe`hMH}`DC8I55usFbOn?e4$49F-&c8i3uO=-#iquS+K_8K7<uR<J2fYf>P5%D
zfY+5v-zM9)v?g%mXvHH&T-^rrK?kctX(rs#j!lCXZ7LeI`K$dV|MJf$`pVv|p;ssJ
zRmKd*V5J-d_DUyV)L3vT{zb46G*t{tTYdGzCrf7)${f2YTb~Z8&dZq$y0g)jlQUcW
z9r6+$hSjKjsl?JDA%8LSh8qJUv}GzXx=z0p8v%S>@|Gb6C+?4n!cI<uGa(w(RLNAt
zdyFUHB9}dR?Nw@$;0g2J7YBu0#-cbGaU@C)hoIHyA^x??3a*%097{4b-BNCAtoq$z
zpXbht0z-LpKP1EZqt+Dpz6@UJz@0N)XYCr}k7tHBSBb`Qy%gF|Vd!J$QDZd!e>#U-
zK4m>bq7%|MV(sBh+n!-r`>k8NwdV};@AyONT`xtav7wQ-1t=mEeowQufVy*dui~ez
znJjXcH8;7HcSYzw>+fVm*ZVTr_|{L$zm>8x0g}%9=@FUUDI2!QMVgK~wJVp-bJDl_
zN93&|74vg8l?v0X>m^T)_0LqO!|(BJFG~Iu-Q?RNN=#8I{`Psv=(4kDD#;c!;wxKl
z1i3D*saQ4lwb|iVAva7;!BDW}*`f5PP{Ij_Z<5q}0kB@8RKbZ^r^862)CvJ9Nxuh)
z#>3}m2>)g$f>E_a9|z@bls#@$<%^?Fv%5*I#-(B@p+e;hCrdNAk{v>}0r9uX)q8*&
z57uNZ6h4%;>uF}Bic+)uP4Mg`woRU=-dRoXIcV|`iEy5Yg${E{N&Pi4ysH*w0xP%i
zBC5!&vbi*G<vyo}b!P0SlA`;md0EN7A4R_`S<QOF&;P9wHh;;f=3%#={U{n$m!RiN
zB?muG<s@_peAqrRcX`=)8Tj5Db0uQ^d|hUhzUIGkQMs{m-h2OQnl3T5D)7oQjgACV
zqn2q31k1XGV9Q}?auLP9kw*T2MsSGA>d_1rA(v!8WHA&L&i7JXD0d(s{YLKxE>r-U
zeCnqa%_7*Mu!KgXA=41-!af17yG<`g8qwsF9ZWY<2#s21zE&j+t)S$m&c5;9^e}&&
zduB?<wI@pm<4G)5e-)AbtZ7Rj#BeE%B~mWc@f_#u&{x=dxuMJK)B^u{O~Tkb+wgeb
zECWTyW*U;Q*ejf@s-XiGKmWT!1(;@<-jgqc1bZSTM+|MQ4+wD(#^`sxnHCY!B}Waz
zc`sify52qxO0{zf)Hw$K0ezZ7%Wh7D@)XW~@YrbmJqkiGxBv};7IKj4OsL*9ETW`?
zK|j$#uy8(6Cb{xQ*nzYoQC99L8Xr%~*!FN8OI$wOzz|0UWHjdVk@pm7yD0O9*gYQF
zD1L*J*@Cafh*+0rzjh$qaD*zYJQ4sF-$H`m4N^MyaEPW0G9mfe#}XK%!)~)c#~e?y
zNKa))@%Ua0>ZgYs^T)|B$(_9reG)$!!AMNIOyVE~F_hk+K6XH!nJj4(-!q9_XGp50
zj*I#A=By4@gv-A-@ZurIZXFdlepgW^Z({>Dh?Nn?H4afjf#O1e>yi#1+n7yZR=Y#Z
zJeE-=%mSEh^jh`VV5*cq9|U(YKY}*db%atGW2v|k63C7Z+H@J2vhi`54@FVDIutvO
z1g5@Bcio9AQE{ySk)FLkY1M}s^zBTKr-xek$?sxvoQIXy&+S`Z>x(g8s1&0a0HsRh
zeT(j5yTrh88IG$#KS#GaL17Q4BjghekT#M^7hl>`A2I{sMATjH8|T@+q{jnu73+EU
zhu0H9S|p2U*q)8&LnkN8a8Bj`&r0ES*S(d>T^RLcHLE<xR;;_l*`@`Yk*c{29rQ>K
z#_|W93zhM;^;x*NZ1_bhPrpEM+|VnaJ&P<|XKn}rO0os3lEYfHoHjLM<|13$AvH)D
z5>+V2`k}do3Oxq`f5cWS2;|YCXeT<BerVe_F~umaNPW-(N|E>?3+q1Vn$f^D2{D#&
zowpDliOi$2Dkvqo9wRc*4WOlt9r?jQIu|ezOBH#?xC2x`9RM9__SNhrR7p!{XQz_`
z68c(ca>Xn_ZgG7*CFhtl`@pd*5t757Jl_~WmcRKKa`pEi5x{Y&aqLR4Nx$Fc988pq
zuH}e%T<kiF)ch098H#WwP;xHZo_xa64V@9QJpUOji=^STiP#J=w#(8Er0*<S4kFrx
z#y3FxQ6OxemY3sdq1to~B<n-rFvfwkdbY&lU)IIl!KlILba6~2F7VY}{gQaPMD@bE
zMqT&6K-J>LK=C40;Wag&FnoTHv;lFO9Ir;7f9JdWeIe#gAx;^|xpSG!NiZev;7Z9W
z0nWE4(X>biSq|``S97vaDZz4i-79X50t11QjP^bmCS@V0x74{!JY^O`m@k*bL<Ev?
zfmoBilyZG~Chto%eL4(xIl{GJ+1+kCQ+W%LABOZn9hWWA>L>uCEm>;I&wQgV1Fl`8
z3Z(}x8fcXe*EEqZWEus7&K&IvWSg9oX^|<@y#i9+m*9192359l5BzpKs;C_Qrz_{+
zOy+s8M^$KL^H9+6V{thj!p3&1q3Qszb#_MjV$%o;jbfiL*1T0l=rI)4FAz!I?Xs&4
zehPOVt&7OV7WZLdVNsWyO5WdXcJ5Q>yh-ttD3*F`v-#@VmvP{=t}btRoHJ#sXBGCm
zVj{7*dis@@Ef%BNXn*MaDgG53ft7VNy^#0r<{mFct_eY7J=&-p>cDuRfY-;rVmlX9
z>Lk7900#Q^>DS}X#@ZfLl$iv;XZhJe5^!dJSi0_!Iu$M*i!SQ5?;Y3%jo)a_EtmA5
zANdu+Y)a0!E$fa|h8<iOp4%Y!`5c||$$Aqc^vS~CY)JgAEy@=Zpd3c}7lrPz0&4p6
zoB^g!p+fCFii<vyt`*Ic)9-nw+oKvpZZyv&Yk9b?@8GuDsLh2GH<N<H_yxvX_CyK3
zsd`Y}VwqtpA09Q*1N6pi9NM7NtQbbFeG~4;f0He%)SjQRlhGz?V0vTmiJoKq=Hvb_
z>l+j`BKbcgTAZB!aTCSF!S=t_yVPiIIsfH^^S{<lI8||@u>L@&3@sZ_x=N>ZsJEZw
z_u#O^9UDzF(S{lP_H+7Y!(SRg;J$denrbT5P@MO#WIIKoM}u-!gHmCp<c4)V@|V5q
zDfD58GBS;beTs*59#iLPwVnW%d34N=H75dr9O|@P@5A%S?(y!^Xk6rsMGj+9QR(|J
ztj#{&<?HElc#S0S>#TibJSs|nzxrZi4Y%qjn0xTe(QGRbTE1L)J5hpc8K%YmA8&T{
z6V<#{U$WPfN^&sm({3JPMUNUV<xI7kBWwKo(vwQ3bwVZ^Ck~#Qno_!H(v_U;I;+qZ
zMNQM+TeUP!KbdpzsLu(h>I5xSe70dV%Sabb<GUd3YdLj}Ww$uw3CJk6nC|#cG6;Vc
zjIu-eIvmy6kLscG4g9kY1gVYn3;r@&P<IB-d;QEUwmVt1M){JY1){Ki(*%r-zqWfE
zN`HFX%5DP3Kue|RB*(Ziz2?U)Yn%8ojh0q#75CN1o^nY9<ZRNRj$RT*an4uaow&S~
z_Q4qLCy5880{^xSIA%RnHZp`g@LP-Edo!B_FJjf`KI(e_4n1A7QZ(9GgErFh{Md&1
zCBhKXg4`E)U7<mqkwl7V=E=iV#>FHxvE*F+^l8ycz7TmiBW@B}@*^GicyK;c0p;Pv
zHl-I2jg|6wg%fx2$A1)mwUY=BeGAthMN1-asC2PCQd_xB0RBcajKjtTjOj7UgC(U8
z_kG(={4rxTFziE9%F5fsB5Q>jB#0aXgU-#iM!*YDBbES42y__khXB?e#~-sKDLw9f
zOjJ^QDPW^tLRrK~Lv?@uYbu`?2D%`WpUGY|p}*O1N48=R-Zv@C7D$a50#Dz0(S(|8
zZw~K3ft98EB^a5#H$z1F>p2XgjF9@!b3N%2D<-(00XE#Ct)*&$!dk10V1JVwNriN<
zpRf~h0pf0W;<zzwvIJ)it-Ngo5GLA~w4GUP<G>auqv`4ZEy2@%v9MiBu?nt(ucH-Q
zgiL04M8;;@6aG0^%h#TrQ_{nhDP1TUE`D8!gPrB8Mxh?Qj32x@&kY#;>gsx0-;N~b
zS^;GncU$f#?H$Y^Ku=Q+L68ctjwb8-n*nWdFvP{^h_Z?QT&gt|CM06y5PF2r&o;D6
zo|;A~sB#mG3UhyKsjD7;icJq0QcqP$Hl4?>Xv6fU0_5-SfRbQpcxqIk{7$z%;+4}Z
zfZ1Z7?dG&{Is|Qgl0d5`+S~6%us%R>_9DHkwz>ZCGtjt!vEchKv3-+iB09AwqX(7y
z`;W%(jhTa}b9$5PGxE?y(im=R!_<)f@tG&NI{cme&cJosKC|Mw&xkq#ZjG~oZifI|
zf2mYZI1i1nhkg2^sMWLO{X$37Wr(hQdgN`w+v5DQYKQmH$@T8gq|ar{tDGt_2sXIw
zmprX5mBoAEYLsw*&ec+dyXW2WHO`s@q2XMmP|3ZGQlALwg!f7rGe&G=T#qdQDtgD*
zMXf!|=b{z3Dcy1>1M1G!<Chi(SpZ03u6k9q%`x~OgzG>qpB|%cw&d4$6xyF3$6a}b
zqJIt}BoOp9rO~|yzQr==a^@R$D}1ytEw?q8*8RaBlkrb`xEH;d0?qboK9GxI0lYLH
z4ZE&xh`+Asjjp5lUZ&%ux66h#nMSq)rRft5Q46BODDr}{=n}Od(l5zsM88~Z9A_y?
zZSMS9J-=J^+#O~>8yw~JCW+uwk?>a>^FgT~qHWo!REEe<SXfr9M?fh8YkPPIOlblj
z5{7{nC2-1sAwzg~jhfaB(8kgwJ!a)(d+JgH%7pfcs~h7{vBee}X+AF|dSvTsk}z_$
zEfA*Y$ycF2)%HVSAx^VwT@Ok>H;C#d9Slxqcz$b^8Yq{xd8!kPi-;x81s%HIdkqxV
z^w&!9vikO%d7<NahE&mVoR$ZaP%v%K&aQVk#x&Or_Yyod0MG#&>+YJzzU_5;ooE39
z2hl?bRc7WLbQQc;=V)<^H@L;?3iDfKeINu?bEaftrg)>pX$SXuGdXya*f-5)SzS54
zW%-@siGQIlw8Z$Dy;bArW35m^uc(>5dd9Kz?dO;tE@}LfjxEmMfVFb<TF2U|?nIPG
zTsQlTs-Q$Qjt?7@z}90bSj@Xh%DOhcymq(2jS;OSJ*ekrii%>>Rm?S35VT^=T<(jo
z`39zhEKexrWL~;6Qauibv**&+&U&{M8GsvOf}nO!T8(J}#$!94fzLjA{$TX0#6Gim
zJZRzg-ZG~q;AJd}g}L7PVo;wf0$AH5g)R6k=9{NB2<th2ko3R}?X(vg49Z`YZeBid
z+pv=c3j))XS!n-p^uhHXj9ezx|2;pWMf3kvPCnQ5y+mY+L)jsns&I3Scsi_WGT^Hv
z75pfpMNxDUW1VEbyfGoEg$qsDMx%fk+#7ZXR&~SkwMFO7n*=D#74(&)6Z*xOLMS-G
zf02a@h)S{o@-2nok33jYBgr(95-`T^C+E|p61zR@zVFEtGCn?!Cx89vfs(ba-45M4
z+~~6kp%38B>h^q-zUj3hW=epG_nc1FDrG_G>&?#o^cFXiL2_lx7|DcT)z6bN2crx4
zf<Ug05!W7VfuQfIom*Pd-(J=v%oZYNCwqjc%d|P@r@wopscI~x`W+7w(9J@&Gc_sJ
z(BZm@z^D|Mu^C1t`^HEn%_<F?v=5vliEak0oMK)zuZf&!Qp9wqRrQaT=Q-KjuV9@-
zNL4x;C3Ga&5JAszKdd{Q%g!nL`h&TPS0LiH&Bz(m`3{nE6#uw=Z~{!drsCeow{1ee
z^NPWMj%|6J;dn2A+IOGBc#attBu!22$&<!Bw`P^q0jdC${yf&m@X!$5FoiOkFofbz
z#42#{Ib}~h$Jx1RfuV_5-)oqbDi+TbKB~F1?%8GZ3g*3{kHTi$W^N?zwVS$XMl#D)
zH1FSlVHHG*>fzI8L7^exS9q@Lvg&Y|CFS~}X{ZYyT0K%GrIKcs4s;ccn?<;~wLPXN
zRRzMJ58+_?er`?ASc#!b!)f`H*TKzOShuAyF99tAU85eW*Gc<_>B-Qy(6b;MW>JHK
zhId0;{UlcB;5<Zg(alwR4zxsmWUP!E^osId`mdm;U`^I%gBbx)6m5<`NuRm`gMN4J
zWKlUVI+`&#o9`6|4hwB|YCF`&Z2V+p_ILKx5iHDf99JuATOQWaIP}m*ts0`$Vl*a_
zhQxkSh86c|^Bt8OQma1<>^<{)4wCulNNXhE{tl7l1?jLc6xN(<tX77Ja%;f|SFtrb
zl{+RYRh#}+(h2}HSJ0Tx-tx3uha8R#UPQ?1YYjOjJv7qH>U%kO$c;C`0?33*<~ng!
zIVVCUxV^J(>6ZRr=Qx%P%?%ZJtwG19oxl5=NyVx{-Ey?@2<=#++(*BOSU?qT=nyuh
zs!R?r^)f`REH?@PS3i)+cWgLT#^oEdBT?Ld{P>c1B%4{9tldE&;~cENKhX=OtRKa5
zFy4xV0fyRzASqvM0%|Pzj~bZTVf)qWVEWo8kwQECu`!(?%%}nlE^ma8oq<D6-9`t-
z;@G4h5l4C4G<`J(`d<ytoKh4~d{>lUgkyO?{gl=0CPkCHkMdSO$cJk$bj2lH9Oc**
z&Y*+o=O!fgB|NqS9z+`(TLT#{FsGE^CI=ZcUorPA--Kz$Bf(jR1PtwNF87)lAh@fC
zt_txH^$%p((lCf!)SsY92W<O^rOM6q*Y(e?tnuq?*oloF0|qglHS?b?y_9>z*$-o2
z+-Wkj@Sc`VF+^Xl=Qo~`{mOBs-!4mVKBcyM%_g<Go&_q!<VT4GG!h)GI(~#Btz$+O
z)Hf51H;~d5sF76Ho$fi&gc!+XjCQ#CzB$itvsItGMlgwe&2bHN{atbAp)(}`7W)Wh
zWw*WTtdF)I&x^D8nIq3LvZYaLr;zep?gK!L)m{pJAn^qv=^wW8A`(A+2wSQAfwivY
zLEugZXvFU?CF~s1h7r}--UlHtQthG^z?0Dtz=zQZ5Y$|AmpUk42@qVD{a@16LU1u+
z(X?qUmy8lZhi_?Q(0NA;y-uUh9+k+oL`G8*hDdwjO7T=sq6AgBG^AR!Je@aghku6i
zwX9Aoe-HcHYG;!BPN_e4d>V!~{M#>ePt4Od+(hl7mSc$o8$|1@4|0f<VH=gLmxQ}B
zXEJfn$0di;cG9n1$s>Ori+cmC5vDGjKz*ib1<1(K9^m9aEn-mFl>y}gAp0C{(-G5I
z(=B~>!JeRSu=MJcAHq>D@{f3`R`>P|xli@Cw(IP_fdQ`>P5-qc{09@9o#lVpr~k!}
zWR9ww!iFGH&#{JPwhc9b+0$g#CKaKWq#$^2iP9lTTLTz$qzZ}Z<DaJ6+N4z`iZjd6
zE|&Z4-W+o{KS(Sed><*u(X>EFh-f1kNq=x1aAhh%;vg54XkikJn8ou08IpA=QsRC|
z!z@*RxklN}Lex8z8Hxxq(y+3j$ESn)c`N<(mA!nCXz=OL2DmxLG2-dl?t>dVi=aaF
za3oN|1}I#)ktB{}I+1VdYTW%A^DnbQ_hgeS82nx{1%IbmoFu^w_DBs|A6Q{%<x?}l
zOB((ma$&|oHCC7KHJg19g1q;fVG{4EQ`*RjV-*Hcap?45jx6&iLewwlUlb_2joQeN
z_OAUlt)o_KbOdGeoBy=S;fq?P)G2H=B+QQVs_hI1J44{L;|$#S{lIE}k9#@Ej((Ck
z=;k5nwjHv%_q&y1AfGb&JIZ8ef0!~olSSPXGs$l>IO!;6n8ctZ?HX7PtNWEE2YuiU
z!!Yq`O)5xS9SqSGnebMtGUqqfn}W$w)>coL*L)-pyu&=Kc~NGyHA$H95@?*>(g)VP
zt`wx?NcH$K^`k@GF}ShmAJ)B+NUN5RU>Y{0hFR88u<RMDAK6*6wPctG`#X$HCGPdf
z!n!QXCKP-M&R0`U_*oMn-I_uBfry|4TQ0$QD`5rJSgEU`-ascZ_JMD7DUw@#eKh(+
z(Y{WD3L>bD4H}&~xlLKQ=1p(e9p*Z%x)mFMKxLR~EDvtGGKu$9xGAw^;$s@ZG&5<}
zxjrC2)yD9D1n^%eD1D-Rk9LQ7YLMo$Z+A7GZb>Zy+OOAj30rlqQd1q@ZMFe($1eJF
z_K$qf^b7qw`hO?9{_nq;i}Ane8xEHLb!su^e^QIb8Yk*>2F_ASv-xSoa)A(Ht*fr`
zQc80uwoM7-8gX0ge|kk`BvZR%H_%`i!qCE4vPU^Yu59_Pdc6B*M^>sZ+w7bAP@!Q(
z(Y&f+D2*9pSuji^+Gx5J{QB0WudAL(qn<6PJ+NM$1%Bobr<udszgz8|Dn-0hI3z!s
z83?%50D6=p#sIt_br5p^jD({Cr=il+K3edl&}vxMQdZg+%ETnC<A{;t$W!Hoo9myg
z-~SlzyK=;h6p_cK^GW7P#o#w)lF8{?KKB$ka_BK$er7y*K5k67xgS3Vx5A+R@Ts)q
z*u|@>87@J5%45QcP@qQOuyc?+y6Rm&chb1@+odWyd<$78o2!YF4iZL@)UA1u`d&xd
zwH@-f3MMfATwdlFnG<As8tE6gWTSVgl-lH|)aW)vj0jD6;N*`K`5%sZ2p(4o8`U^o
zXzLO!EpY}mPpv?rt)E455~{78yYV7JW0E{{Fj?uIEw52LXz(oZV}2KFHM1P_i#iem
zP|C{NE4V2;>EAXhQb+1T4+ic3`1uGfc)^8(Z9&7jB#d9ZgU4R$_jqegwaa_!`TIyf
z`W+$bKjX_&+0jBGgsg3b4~6;jb0Se$r`?C|1%|N0Ub4+uhS<rrgcN~$yIn(}0EenF
zBdetMfEWT!Fi*Ji`JD-paV*`Hu(azH&YnG2ilSh2)E@xz^q(4aRoNcr@UcN@lCiDF
z8l^0*rk@z%1O7#KTO~b5i3cVF035S&-W;>X1j9308wscv->BlH0pBx=0>*d+I!TR^
z)7T#=D%s$_W>`|hhp3R?_5~5{_N_&!Hj$ONhp39kq<<MrB&lF-k^nMNRevYlG&9c%
zoI~1slqO%P`lMek!dR@;2@S_p+PbIWDV$1jMX{H(^T?gxlZ8}*$#98N+7O9S*wH6+
z^t&98K6b1Y8n#kf$Wu&4>##&VN8AaG(JJ*M>XP+aeF#Zt{R6dH^$xHWwo6=;pdJ0H
zTL-P)Iv3hyY1$1yz#WvP7XmmTQNgKg_TUP<b0n>Mf3R~<i>VKkr`T-P8?b!umFx9I
z%{h3=Kcd#H(7I2nHoYD)U952kL_S6Rem%vFWIP)ICL#FCJD|}zsOv-qP{kosWC<>9
zR-Sz~h}rU;Vu_+d-ELq;y3Cq1>sjHy^Uh`*#<Q`1qp((51ztTS(BFbu>Q)>o22*?j
z8Gr{nMwhu9-Bwl=#gmg{-ur0p2}vh5icWB?oM<W1L9iUboB_*M4`{d`H9YFv^hk%x
zHYS^|Gy}nF6{8`<jJs=o8UNu^&v~;tg+dWVjv@|iDB!AN5~8VQr$W#js4xa83_B{O
zjw7XoD;V{4)+SZTbOn_x({{s;%H30Rzz!q8!`ojoMs(m(nw8N<f*=ygmd+DXZW17Z
zfE8ZfvnU6aRE8y`0$e=W-b=z)SGn9MA!A(sCCva`9>enTKZy<EVtfQ+x>J!Gkf*ru
zKV6q`q~|&y!Is|9q}xTxF!Ux4)5CxQvEZ^bwT;tAeN^#obw6<mP_?ygT_VV0OgH|^
zIUIbs;i%@Xwrmvz%V!9r5EQK9E#!z2uhS^AE3o7%b(%Za!?XB^f2nGGZ*DRA<0(>c
zHqxj_!HQMoT$N_D;p({mOQ$EcAI%c}um&HCNn%Z11O+30Mh8BW6pz*Lw|`n>v^Pm_
zX5a<wyBsX;a&X+ti|Qhx4{6kcL{t_I`1=4w;{B9IMbeKnM80PI4_U&at&dYJlnREK
zNoJQc1LY^07_S??rs-eDOpe|7pqUs~TnudB0)^}B+Tt43iAy^e0;bS1KEuL}`jSBD
z0kHbu=z?|=5x(+YUOY7rQ<=jgghrkx1N9}vIt2aIXAc+~eTPF`s!f~_gk=(0G>!^P
z_Die&u%Ix%p?o!`MTpw$AsV1C7TjLi-MZ_^4Sb0AAVWrMySGP(IE76F9}uqt`(vQp
zx+A>d2f*|#AB(6rKT&*cm#?smpZ6KVC|Cd0(|d0FZ?vdf%0kEnsx^`)r#VB4v{7)q
zJ}X$6>~6d2yIul3E8jl7e7;aNL7LaBM0w6{6JYBgR8E^fweH(;o8t0xig%y_hHovi
z#VP5MptTh4>9YT2WhtkM_R<K2Nv~m)25S)3)JDKwI`*+o^CvR(79zV36ez*;W0>mF
ztwidr7s#N=hOlUhW1N*c`|@;u%bD_Y=rGLK*{Bp@U+~G`w$u_no`fZm5Zc)ip$e`f
zM(!vI=HV>5W+uu^6H@0&f?$9Lgb=&?Cdz;j|Fc=iT8i=qA}JeW{jY8DzYfXSIsYdL
z^Zy^$DAAO0{y$;<ca0m#X+@}QGQP|Hm;;I&$1KOcDv6+UN+>FkKWWa^xAU&PKr|_(
zMjk%o05{AX`X}nM8s4){ugA%Yv#lI7Jz1ma77lGMVQ*<G;gUNu*z2m<^lYEJ&!?+c
zre$JdVWdhhs0P(*fo-2$UypRzjza!26K1u{H^esoZU2#`E&+lody0Yw^GgEE>udWz
zQ&Wf>f3m!5%SG5fK!9+(kPEmG-~%j!3?@WShH)Ixk}697ep^QByU`)}D2Pe#LzyVY
zc?u)rvnY%(MX3IpW+yUkl?TR&`=I&s&ckr-g8dZJN_Oq+%OgOXp~6t9ZmAb{RK~_)
zgj3VMsM1VV4~{6GO;PhO><~rs)T%yfLH=ucD4QsG&tJt^5-CMxnnwfFA^g%M_3t{0
z<d?d~juVGrgT9@-iD+W57f@o3;&8O+9*}}A!Z_HW=;90dY9NRWdq58Qy^p@B0(E<4
z{@(5Uk*OYry68R5?mRP01jsbqFJ5D1b3v0&hVlEmjse&EUAr!tdAxeoqwr|13%i_}
z<#v#q3rrJPqbpp-Xl};P8rB7E>WGwmzo4$hGO+EYUIN#Tf<YCB*La5-alRaV1uh(f
zF|0i$R%!Ncar8rASPMqe{^Y^b7!PvN<OcchK5=PjVb<N4hPdbtFo7lZ9CNAIEKlW<
zkrI>5FP~_V(+{lP?_!JjO2c1xDtaj%4=0>+X4)UB2K2Z=HzMl0ZF)Qt!+uR-)RRta
zQdSQj!k(kZ9-^YfN^z4ywEcM9v+@0hRW9Z>FHYM|)2`80Brcnn320GK89~+q9T)es
z6Q9Ppy^qU^D0Gm0(@_?tQ$`q5b%+~i8u^I;McAp_UEA0}0KQ&(sOBUhsyONSVdN{J
z%JB^512QqCBd6fK695#sKNLYupSZJl8e!|kOtF3!19N-0SyCec*oH)oeKr&~S~Eni
zi?@BSg>?AAYEbh%FEvf|AP{97@O_uYLj5&{CS-$v6l5n-(hesD)ln-Up)ig{rda-$
zud&Zkx2O<H%i-q4g_Z*VFZ3@to2`=2d5O_e8CcWk_|F!adP){WMh{t*ZLa!1ff8no
zW_?WoDc6DO+h_Kgkl2tsQ?5hDiW{-)!@Xk<1kj(=nm6pCZ`Lsz+zshiQ%eV5MOYc+
zNR#?Xb{rfzu!zc{$K)HnyG})1=w#Y14ezEQH&HjW4}R%aPeLfSozLddG*1>^UlIFZ
z@T-J?A`*X0TuKO!mPuA~dHw`M50r{rVk?4qQZwI{hNeqN4UX;?*{l_23A@Q3A*=L_
zH@VQ*iNSEelCwZtJ{E_^H(67?VsX29)vqD8MhhZMVdgL$lVA~$xw$xt<>uZ{N7@1<
ziz6+Pc~+nsDEnW}NW^njea6p~Uj^6HYCRjX=btm&Q8kyeyT74K@<D8UC<VElvTPk5
zo$C;hLXTwSnB-x`Y20#Mv;NM|02b+%$hrGLp65I-Ifh-5J_4LAPPCJ*tA5`WT&{7H
zva?x|O*%nHg`JN0g&_ybJ%!c$&THb8_#JnO3qQq03Nz4ZzgrhGz51LxJVM%5a^!2p
zl!>EdHD9KF(D=%U;+NqSAR0@$cOd!js|0(Lyp&35%kfk=L)9$xuW#zIt-f{sZiJ9U
z+;iQ-O|&1R9i_*`_o@b4ZLa%x>wsxsp7FJ%Tj?|F&d4Wifu(nFKKv-rb+%jll<JLH
zvS#r;@7kITT6$^t`|P<*FFQ&`3@RyCRTZoLeg^2T^d5yddSOmle88Wu2H{MCk&kGY
zRpKnRr)Ta@FG)Gd3Q1ZW_Mxi|oq#1KNgLd4_;Gc6%v4<1zthxG{P}JwC+$>O|0trF
z^0K3y*7++jYO`KUJVKnhg$Y?e@oJMU+39Y6@KghC^|-uJHwJnbuC427?4x~-=Ia6a
zbnKk2%T*WkxLM~2FM}h?qWL~9K)LRvTr{B=YRJz<Vqh9jA6(XdM?Dwmv-%rzKon5&
zeu>Y>#~rGyD^!eT)C=h4rq*z<Yl$|VT%rf-o_{Mb%B~O)5^818oNm9$m_xvIfxb9C
zz;5+nX6p6Hd5lbCyt7dww5L;cWU<w_?iS+%X;8CLh;>dLVVa{<hJzLB2He}pxJxn6
zL!(!Yc5Pzbi|5qmNE1R`3FLfXBcoJg6XFLQoxpv@mRP@x96Y@iR=B>{c4LUC6|Ilv
zD`Bxcq73Urv2er6za)5UuJKRReu3lszWNUX92e7n1U{LVS^n3l#25`3`-7it?x*{s
z0>wx}j7Y?|B;Ft!#^|Dp?~DRDCBVvZR6W@{o9FfO4a1p8&|I7EClb5M!uDgy#a6bz
zR=PfpQpk!z<cMZsAg}nhu)*sg?4?8n4Nq#{bbsX|FD0rjL{{<r=!72jb@X(f*H82R
zF!m17p|oH0W^CKZjcx13wr$(Cog3S>ZQHhOJL&Yld-dwUH~7`CR!wS9b>6+tInUlj
ze4_L1>K|*l$YG^Qe(l);+fgBy>G`}Ld8!SCZpoqh>)rP4{p9^H5t<<v#S(&;7Y7pP
z5ga9aJPX>y>fz5pU~~w|O)5@#w|XpJ@YzM{Ty=k`+@!9Y&T^~@-`Y}jUrin;@TI-f
znvlfxm6_zZIfJD&rMv+4g;qTS;0LUKEx&SJ!~o`z2$AjqcpyaYF9PcU%?FIT-2W_}
z<npYvXOx24Zq>UD*0c2bshHJN)%<TV1}ZcmS>Y$IJaSv&d<VwzT^%)4tA;EJj;J+z
zSVy2#%J2XJDkU5;C3$Z}nSkQS7sAeJy?vuuyXq$&(HP&ob=cV<36=c=F|>-Kcf|m=
zKPYV{Y5`cn&0<wj&#3fvHbL&Hy~>6(#RfkP4;RN>kppT7yHEj&C<&5ZARQ(zG3hGG
zLJqRz3em8_BHc>2{KM;3m4!^xA->kGcgdqxk1&G?sm_Tw6ZSi}HZ6e?BaQ(ZpoWR?
z&+!f%xA7H`#HZ>}mJKhh1NEmpAz`1}W6hUo&o;wF|G}&GzF3Gp9Jz>FJPf%H-q4Bs
zfH`k060*|eIj29=%mvaz>FH;eOzvT4Fm0$;L_~`HX>PnR4ri$oW++_5oH21beRGsB
z`f$n|IZmqDZ~-)Q3gTIRzOht0eOr_`I$F+(3)HjBKN!Q)-g#pwPI{S{kLVn~E@YAt
zJul1zS}AJ7`QTtj<p4P>fAVvN1!RWN+^BvTK%~DJjQGWJDON+Lxdbb3(9fu~e*&m7
z^|pO4SoF%wtsQF{L`9n5c-8G#XfeMuc7S``%3lYPDZb0Kd)AD@^oTz4yoM-5KN^rU
zit{bj%SadK4zZxDETuih?{Q3DJj_bRSvOpjR}dEY$ng{T!u=(t`~UIz`bE@=OB(dd
z{W!I6KI<ZF5UUMgg;-)HK4oRooT~svg2v@#TCls2Jka7q9dHwyU+T4ZYl69yc9zXX
zd^DilvXjPhx(1O7-Z)!hg`Xf#O{bC|3Ur)?0gz?SQRd>tCYV93DdRp;NhTE|?kP!;
zeD3)3#qI{B{Yc{<^~Q*$`pP+HSa;xfEULXC82pL+yzH}5=%HlEX7o2hdtYXIX7P#s
z&~l_~IH(xbvGq>2p$H4ZsSt6ZYh@42g?UpFxN$KY>|(i}ZCY-3HJr5@R5%?S{MZu}
zPnU?46hNvsifE#X$Y@CsS)hE4tqd(zJ$M&PI$28aZm7*G%Hh>ZzqTtqZ=RvJb^qBP
zU3DJXyGqK{s4um}*-3YR_yVEv<#=GU`+S%t@{h@ZAmSoW<cE?i2-^sS43$Ih`XtWk
zM7#kK;XhYhT3&1L&MJ2>)imyE;$S;vny#%ppJe`;>b!|J*9rVg8@^+&3MZ-xP}c~3
zi+E$1hBid(Hr<7m4uDKw#88D!_-J--5rl`%h^5JwO@cmTKjY0WjNSakUyup_{49dv
zQXHXK!W%GDLb-gtt;~v><t`7*jxD~J8uG;fIoA0PIl%B=6n&=ueS5n}_5YItXcof)
z5dT2Njt@!gGn_7steR4Y7yL=o3?@QD;77h+wDr(Y#3#y_(lQ|eK7717U^=VN-FmV<
z+d19c@q%@R87T@|+Mfi@hT@F)Q9LW6U&c<fXuB)7d$><BV@KUE>IvZZ!-_n<KGsH-
zHY$b`2L0)Pn$?CGZ)>(c-R}@@rvGYbXVS!v>cZZ*xq06mdl>U(YGd(`0jjh5c5}}R
ziv7e#CW40FCKKwLlg>r;py?IWtY~lLK1%Ce-#KZr6`ju}g%3=mhQk~RQ^SZ&T9n`_
z$}j3ZSVY!rRjc1Ft~@#ytOQ#qrK~7NvT-?=;8NgfD4&yA*`N3mxvEsi)D_J6F@uDI
zXHQTw41?=6GVtmPDD*2Wl!nUsdZ|+zBfLhdfMzdJ5g`^*bC?oJ$GuO3SrYW5A(4nJ
zF6#bdBUbD3=<g6qB2Fm4)hj^M@u{PiTi$=<#AV@2jlhF_?xO%ZzD@R~u)@Wz+H`YO
z;~ZsSQjOlcqvfPBwJR-w!K*xIw20>8)=~p|`;T<dn;;p5iuVfPxL4dGyv_co{!JA;
zbY;<bToPfk2InNl{S#6&<hYIW1~<X$1o?+CAL!4^ApYA$`{JNafpkjhG*Na0`>cGT
ziNT~gg3vS|Brh>F(Z-7NQd#GG@ld|gET|YLB<M_~{;{B^1so=MAG9d}P8H{?Q&@#@
z@5x_pNy4a<o?NY}<&HFrR%&LHKQ9$<J6RyX&OjVu@Zw@1+=OBwVzY=!HV6x=)4)F(
zGm&0hiXSq_sf(ybJD?saP<mk(BiH&hWg7Puqz$&we6#NX<jvmQ_R^rFM>4Tg-G9gm
zyn1IL+{1iO%GaMQDzZqFZLv&%!jH%=hKa3|W(PxJu>MTEL5QEw*dRD>UY0P8;|C%k
zuUkNiS7k>M{>9}#Mv$Ql3+erI06KPjQ%s(Nv9roh9U(A+l3a056C{rW=Suq_yDT@m
z$O+vXG5C%Pc<HJmv&r|-A)POX$VrMj)U1>>G2!IllS3N`5crdeJb*CN-%VP=@W@N!
z4&b1(a@0NhGQp+%m`;Bn!_wAH5i<?ntILp%U(0aMZeUJado%;HE6xN)eVC@GW1)yL
zpGyh=BWu+niG|50zw~PL@zrf320$3pDA<o7R_N_sa3-j8^#Mn+Rz-7n;nU%GE<S?V
zz2TA0j~WXSDIDR_!PsEu$vUa1+6d5usyCFCo$~T=0iOcbF`Kz|FDer^7|3)RFx_~m
z9*!bNm~as8rRwZaL6c5v%G8_q-hgZsqR734;6|VaC>sK+543iuuKc%u=`2-#Vfy&{
zgaBa(0FYKEqw@-Y=Dl+Z7Yx-y1X_0CS&*43_v)e>R|S{@w?e&lKe0Ftmd~=qf6>?U
z#LJ*6pv|ghPx?9;$m?I%E#Q?me}Y%RlxI>x!7Z*TxPep%#sSF{>6H-GASK<ESWt&I
ztPkhkS9V~?7o=Xzv^@NrK=xUy2^oCE5lpuXnJ{;9wbwBE;0ZI@!~|?uUVUr=%ME8Z
zLprwa>bv4{E@!(>^T41t$H1M*aA9hcKVP=P4pG!G6d_pX2K|MxM05Ozdb63L+^pb0
ziNt3_BeHX`7cz08<<=kbZL&?~vg=37iSsyeMljk`@|9RG5ZTIonZo<MC?rx68hTeA
z$oH#@qN7DS0z|1mjH)Um7*PMNPg>koa5d&S(^Z(QiczON<aCaQl2Z<DZmXF$^7mT*
zK<>JBSHVcA-4TQ52b|RpuqH6#lu#woJLrSra|tAUM$g4bkPaC2gG}RN-`lE^{Q&ml
zi)a7GpPKQ%`Wfu(|MT`~m71jOhRFXhZd9;X*gi-hAM<l%cKhkq)uu?s{FBLp0fZ?D
zja7fT8#)@i(vr0c$>KlK*RpK8-;$YE!I)p&?tI+uu7M4y#l*|o;F2@=PWi;8OO^T@
zaJ?WoNEq6p5{vhyw(j2#o=Q#`_aj0gJ9f4vN-l=1AK#FYA<<iWbq%Rlz1r0~d%C)N
zKAS%ew!kSuLJA3$%x?$Q^{H6{Y02BceCF82^jgtO+zv-Xe-E?eTa|*@NlccC@uWz}
zd4SN&Bq%PWY67WJfk$w~JSL%pp5%@P<uH-sJU1Ja0aZ}N1E8$Np@J?BHL>dMgOFT1
zDyk&iJbJ3j%tOuNE60+-*3nf-jt-q9A$U;J5VDJ_L-`iEYpDpc(~oZJy?X`+*;g;0
zc4P7MY!{$Wa@Xj7ZAx(<+0#bA$jtCm2Tdk>oVuu7?7ADRHPNEaO}wq&Cd!A(nqEbT
zHJ_8e1Muw*y3bnrS}Iv~`@=kaL@x$2jSQyMLam?IxN1vF1?p{@rS{b?J`+j^Lv0V9
z$~Vt?Xb0j0CHY_1+nIkB4&gw1BX&abr?6kTr`My@fDuUoiupmn#}%&;nja;t&g(04
zA}k^sVmi7!nLb{M_eRGjvskOMm9HU7QfwarKTd0nl6fZe1EHmnASg{n%)y`|!UkUw
z<gcY1!_JwTIR;6HYU{#=f-H|xVwj8eY~wudSL8~<WYh@&y*QI;0WdfKJ~`P+<WVuW
z_;$u|{mvN=S6?qz)d<oa?RH3iysGtmvd6F-1%T2;vNLt=vL(gYaW>563Ct{WhD>Qn
zBE(qK4`HQ<)g+9#?^d(?J%$SZWezgp$$A3|mj)4!8KoDsvTLaoM^N8R;V_KS=B~z|
z;Or`&9<T%4`=RixP~c0k&oEEYc=oRRv(q1~t8)EUS5fKoS=Qf$t9xT1C3)oNY9t4%
z-pa;K0uGiWkpDwlJq<6#y&rBj8br86tTHa$o<@8@0v*en=b9~X-o?*h!(R+xCyRn6
zQT*hDHwZZgiLcp{6@EUe3HPK<!H~@?OO%BR5P)*$Ch?YzOC#1#no!L>2e`sz)k`|D
zzF>e)ZT9@g8xeZ|A16jg^6<$pQ}_0F%|`dw?7pnR(9p(K*3*j$>#waYT19?WI`a=Z
z+OC4lMQWOwtjajv`N*fW(mB--CSexz!HjBZ7f;}Ooti7^Fpr6!Pj^26cDUNt4r)k7
z>-htj;3;%x63l%ZB%#|rfCH>Uuhp<v=zD9^l6G%LBsKgOlOt}7WX#gK_j*I?{e$C*
zT8+v`AW@mc(FSV)zgMnb?g|w_@VH4URyv=KJ}@KFA~V6${<Nr}k$@5>o{3iyL!wM+
zQncsDI(d^IO(V+E@R2%<jnP%xrD-}M&EV>&Iq!%1U&wM)A!p_?K36ckK)Q~PrZOre
zPX>S~2))}jVERVR?;v!Pnb|KtZF1OMSOzYm=u}`L63FloB%fw0fhG`nZ@6{4&@k_x
z#&1Q%C{u@WK8llT$cHlkah_2ZIqWjM)ayz04*UuewHn2#2d7F8dix^C0&>rrn2yj-
zy&2BBfaWG~gS<T~G9X+JFd*FboM1wCv<MJotj@_noM?1qc!kIS5T|EiFQ{^ylz>6V
zZKdppdYmx(%CZrc{xPC0E!EZbz%g=$4>N`!AT1<tvQ+w33MVx_@v$~23_E)uyHI|+
zouy6qH)mS7MEW@zfv1iB8HL_IiiYhyWpacc`HEFs7((hwYo%~vo$h~RLu>V-YEdG~
zAp7Z9=~aS;5bB$Vpio&yzPfoES2baH{F$Ay`9=njPtOakj62j}A2maOeQ=ogG&EoN
zb<6Ulx3xLW__XU$gv_JNH;3BBRKMqY|Am2*+pTWo0`T>l0PLQhHaOM-41uck@TDnl
z7&brMo*gszLi-Q5r=U|)^F1gR!^_knkCowA5dor_oDN$Vo-~XrAyUI>6Xe29GycWs
zigX68&Qu%V>twc5n=E%g?XJ<Rv?)04W!n$>I}m<aU?{PBTmAQzH6$|jN8FG8*6PLX
z-2E!07#$zc3ZIoVU%)RO*$I7RA`%~cTY&P<FrywN3G-Z^h3nlBHth1O=`>9U<GqpS
zH8pr)nRqi}YtQS%R{&NZHf<S_TCDg#1jzd1lL?e(6BgKm+&|x!2n>YR@Vg8#YKhjr
zlr!2`_nno_OHQ?=Oj+#|^Lu<P@VrW5w`pnQ@BH-@={c@B;in**Z6yF@T&8X0fw6DS
z4iDD8@BCf8J|&a^>VEaodux|rfkUY<c9t4P*8Fh9e>!q9S+L=(=Pre;JfD@p9hcWb
zj?#+xejL6~h$2|!cVGM&kyY2#@UYvc8P5Rkn704Y5?rr=*>d@modX}ngO8zlE*ezH
zDZ_$_^NeTp=*ZP?Xisn`pWV>53Hti#%XHs3@<m?d2TmDD1s*>?k7hnkfc1uT*r>0M
zC7`r~g6ZsV$zobblj$~r2(EvC(3wJg{-abg{a36e3q8yKx>eews_D4&8>=}}b2|&3
z8527@lx!Wsn1CWMfg&+ruA94HACt7(ml8-9Bz5L{BSH1HI$W42lS5pkTCpX&Q@y;{
zabNm!c<SWobVH4zBcU)>)YA1B^q4P)ge9`g`l9m|MKaQ3o3qVj7$v=!iV!3M6opZB
z<IC<|=k4Xwxrg|~K{ApdrttGQS2LrNdwF^CU`jS(e`-6zBGc{re0rqv`)KtHNB%Ms
z+cuYE+X#kuLD)qg+w~!Atay5%8!t}Uzcrtmr4V_EFWr!0RgUZ@UdzbKe5Ka%1_@`3
z4H&@RFd4l-kI7tq%<zKM4HdiTSe9$+b;IP>HVb{sVxOJdR_y1G;w$^R$HOtn&rDPA
zkopz9b|r24mlDAy$`1aw;3s-putzlBZYFzx+5kWg=}0W$tta&^MgJJ!PuwoJgv20(
z<hyoXLs%VF<25bxr=+Hu#=p*U=+V&?r*_m!$|~(dc8!RKf=xNY0=$lvQU8xw6-^QB
z8To5gDY@EQ<Jb~!P6xuqbgZD*!TlbJ`{ka~!{wIVa(c?12`?hgYu3aib$TJS5dI8`
zg*jy*{6%dNKFC*I+h76f-4^1cCL&Zk4h~2lZ=_x&%+Ax*&_rCF52AmbL>HnFSnNk^
z8qE)xhhf_TI=Kk}BIA&VQ;#bWI=xA6r)t+(;u)`r{El97lF=rLCj9ZTV4vVh3^0|f
z22j&dt2V3FNX}w4!#?#1KR5|*q-tPY;I0Qr4>DPMR2FJWTL{8lX@Vo-7Ek7+!XNSo
zco@54gasmtViejfyktJ5$PrG1`r+oQ>%LIiY5D>}a$bu93?$tL!LP5Tsy>IRk*v{d
zXlZkiG=xNVK1h5lqYPaDf%SF*v!<n$38Dx`mK+~SYL%ire^FOI)D8cJ$`9?JA7Eqe
z>@54Z1fIrkp~<!9y58kJB6dcVAS-fVcTuL70RThh41;-lGNZTIF2`C4#znMQRX}yu
z(J&rR$KV>J0414E7cGyW0E^EPoYm_YTJ4LMpolssSP2-+oYH11ibODgTCtd=<`J4w
zlpvP1r`l{C!^{sy72u(V2PAHyoG2p(0eQDH?hl#4uqln?S2kwP+V0*CO{aB1p&jp1
zsc_7UIs~#k2yd|$0Ex6RCife`y#(cbVsq;yzJr`c%^`g)<TYCAR_dEPF}psU1n@AZ
zrdbSrK=vgR?<ddX8bWl?Yaby;gd~30@6S)2X)#0Kb(BAluk$Paxrvd;1=ZKJM?oM&
zjVu^sTxbhQo<t=lC#+VqQ~%a=_)I=ud%CK<2vsqU!QVdBE5<Ic!4UF?*(ZRU_nY&p
z)0-kZ;FJa8PUNPf%=Bh62WH2q(-?o6)~vSzLYE}$-}{EqDzQj`&C1IeyRDNOxR>+O
zmuF)KB8wbv{9C)qCcNd!&h`Wry$hm;Uz?F3F=xV-*j41dzwMhEee-wnx2J!&pD;gd
zK+XUO1~7ecy=J{77$sjiMNG9d!Bua#t{YSHnAefp-dUN*g~v!p;@HJA*c0L=aJc&2
z;m|{#5k;)PiEGw8j>?r8H;N)wAY^lO_sAt?L~UI$GNJ;1Z{M`S<?R#VuhWE?IPT)V
zpO?ad`<~(i^?n1Kt-V2Usw()pw09M{gooad-w+_Y1`5e5T`fIJc^a0zS0hlF5!*cO
zk(kbmZ_RcSiGxqKr@Kilg?oMxCBN$pvkkni6g5PVuPlFDWGq>et7Ceb;7`+vmSxh>
zBG}@d4GVr+&XIq6@*M0O3gPx*(nO}35<DPW^7Hl440o#A;C^Dt2wO7M&F1t|pr~AD
zIbQ6t6d)Ljr?~-+T(Gp2QTe6Pqd@<28sW}tkEp@tM`3T&>G^V<xLlwj5fqh$gr&_e
ztoSNz{PUZ3O&s!3D^Vd+Bt&YMUxfuJj^t2cZ#p$BWw;c=OY&Kt`t7Rgr;8LXy$+pJ
z%BkP$Rv_{W0JSKn<(nCCck{MK$$@Vh`wbR#PDA)#6@%sfhaUQW(Lx;b|8KYAe;B>I
zQWdvd;6v$tQB~{O6?<}eE}|5!$`um>1vv(Rkj^r%OC0QT(eQu0<MzB9l#v#bTvVjX
zzM|RjyvZ11Z-ukF^|+zl=Hey`0s+GmH%8C}l>!Tp3~Gbex4-2m#6XM^iTlUS?di!I
zQQEtA0bhcSUV{I*`>DO){&u$Sy}*S_fj7K#;`GSk<LEZ6z5<R80S$>@fgY49VFu4u
z3V@{IFQSH}94=brC4@RJ;!%aorQUk*YJ?&lHBeA+(v2KTD3kYf6&m~@KjC!Z)5iT_
z^04`RNxMW}hJZW59tOd$M0InQmU0$&Am+F(qLBDTNP*slT|bP8WP$m}6l*+7x+gYU
zbcjxUo6B=6rE&~*i9yVwYw(bUK|hFLlw{qw7<sP;ZuUGk;ne<vZ<gGh{R0czrE4><
z?p`mw_ML=Fj_!^@TpVX0DHRxXnYYb=&kxd!bpT;GhoHMdWcGQ>NnNqElLq`;C?GL$
zHD+H6?7%JmRGk)v1m+6Th@ut2V)liMSsapS7J>0+BammkuS}8uOdC|3EsM`X>Yq|O
zq~!LWBR;1O`jcvHUNHn9bad2D9n?N9g0Va53}Gf$>xsu<nhLvngIlpUu?A7XXu>c*
zxmcF0$Kz?4>qFN}8d|bziY|C5TjAThPST3ti1*f-q{f?K6;pt?*;~fK4bfz)I)B$Y
z()QoMrG`CuCGBAYHs=j50b!bG&ywjW576}$nuLl2D!ffpJ7*?w{1={;38R}jSvhGW
zoPU0h#IVenz1al(+@M3RT$cot;%e<@UIr5$K?7}5SBpusbnVsAiIvT37{owCB&X>^
zQv?vVDrYPxEoVmi`dX|2P^zr0jDPp$?>i>g^76F)^h@pB-#>p~<57Y#Dm^vTPsN~E
z@f3ehF6n%Sk^z$-TW0UB#6Titw++%pkUqy2WxefQ8O3j*eFyOkrz-zJ4GF;_4!i$N
zf(|>S&oD7gS6<%?T1&Fcpql@|UjDHT)-82}wyx6Hs}a@N?}fJ@26DCPCH`l<XA*8H
z;+RsXOJ5wW{U(!v7NIf_%8jtqp8`k(B|pjicxS*_Q|Qj#TvX*j*g1Y*M)P&K`2Cob
zjx)QSFIHn}b1CeN(rFx)lKRHtCokV2AftrwHdKjNenN10I*+!E5xV=Z*dKf%nyV9d
z129s{)O6QdA&|G-<G9fs@$wDeb=yJvUtjls3|2F+(*Mttkt;P#+h2nE?|E=`!;0C&
zEsosW8mdrxU0)5F+P*Hd83H4tIe0vtYD_v%|Kp15@Q?9A8?(6Lr7<0;@Y!aMW1cy4
ztl6VS>~$sGuqSwp172OCUGm_Z8yY0$kh)Afcl7YW>6af9<0t!f6<PRpEJpO(uC}Jq
z!(JGfQ*Mt-dzn(WEL{$F-&A3?g`meRC7Qg9!&nhwFT_!AxCk{8WE@B2@2*dpe<#T(
zHXROPfq5ce5=!!T95H0LI#zs?NICUI<Xn6Dxw`z>*_DDi%(1ch=mBt%0H&uAXKE=F
z+X>c6l7;3j)Q8n9>{_nNM({I;ad=1daMP6%%UY;6B^Y|7L4EH*WWTQ&3#$j%H^9ya
z17aCpQ51hJMor}{Gv?&-QX$oO`OrvyKPsyKlzGXfd|COM#?-ICO$xbc`*12sUvB6k
zZx`;mzU_0Im1Z(p%e4ui!2h==c6hftqYt+?{4y-5@<Ve3O2#}U?=Dv^K!CDTs3pt7
zORklvA6y3}dyZLc24mG(mo3h>Q(%?1q{P<BMlET5hZRkZ$Q&y|O@u>lQpc@>LgQmD
zy^ym4YO`tHF`r=IvIgAIQEK0yl*7z8aza;5vFKj5ku$KV$Dsd@-G3H7ldMlwH5V2_
zkE63bCnqL$yW8vOWA0*Mz|y{Jh!o&;v<{O&1|$-n_+#trZfvb)M6?Gd2ZvX1Bzp?#
z_SB%s!$(_zA58w=(df$@2Rd}LP8V&g@bB28XG90Y80gquuly?jp~SwITv-t+ihTS@
z+dzI0I0BV|ME&aC+&(nKBAm@`DLl<(NANjNn1&rYr<#CL`sAHc48J8xJss61=DGzI
zM9$(w(m530KRQ5f*xPb4OmU%r@|K}ag)OWu3x&8k8tPCi{X53+dJXfXps|IM<<80(
z?><(GW-1o@n)3{^WVsp18aiaI_u+WEX}O*30)8Rt4=4!D3g*g<U|75)E*G7|uA9Qh
zjb*CE0h_<GK^72zU=9}<(n4`dK62%q(?GexOyIF5hSw~uN0|ENv5#}312*8505?j{
zZ-rKa#}`G_IK`Z*a#p?oS834t)<|2Nz}eGuHCKKNBwh&+rWv)`%0!67!JLP@B;Gzo
z`Uj{mDx}^APkxl#`#kZ*x_p{f1l$WeDBj(MPA%jY<(-?fjef4TeTxmZI%+NL_9#qE
z)Z_gJOGQhhI3gcb3uYjwa$6jfI!f$5YFD^AdMsVHWUkL5fBtfiM?Q$KqzDcNSzBbn
zZ7*7f(mk!@Uuiw1bPtK)7+p?3*{&*M6_8{xRM(NBWe5A?cefDn&ypfz@VX5;8pcsc
zqdPyCuh<oQ`#@lwa%OBZFF8I|*?4<VynMwN?#62Ic@sRTHfbe!?4akB)4XrDnRlQ?
zcHwNCb9>c1o}%x*d1xmK698z!jVWdm&t0!i7G&sTb@~u2-MQU7=-lW79zhmMs0m?7
zCwm^Nje_vb5KaSfaK~WbPCb9SmB#x-kru-Q^z4x#ilJm?wjyTK8Fhe<L-+|Ous7<j
zOF6eRS!67v&7AdQ#`aY4G1q60_K)Sxm2YNUkPL~RRA$R)Dr($N8?N+633$iptdyNA
z5nRMj4**e%C%+^;{f3mJW-&THQNa1b%=7cj0&V>G)M_&uIStirBz4P9rUngUnZrSC
z#Y_d7!n1;2JCB*%Gi{Z$g;Ipd&=$pgjA4w!%E(m4JE>lr2x3>Z$ld_&!_fNR4femR
zh-hCk`u;p0Y^@Pm;4Rf5t=EeEV2@n?MZ#I8=lZ6vT7$J~edM^XLH%7h+}nl@R5S&P
z)vU0Fd)OW*zhafMt!~~)X!pPQ5natfH0coB!DMH4dRrxTOe*W4+A_=D4Jy&M3=^3a
z^_NjGU;M$>nywp}yKxx%<|(aOq{4Y;h1Nncd!99e%{xHKR11ZqY^h;lCK1ne1INM1
zwZ14%U7A9|SCQ7QHa4-WZmCwfIF4*<PbX{t+bPnbts2`|*;|vs#p@hcX@8kS5fDEj
z8ez!1l4PS~Y6C42J7bB_99cbk#rEC8|Hx}ppX<j>dcz8rE{>2@3vBlQ#KMcCgNt*s
z*L#!RQ}aU=Rg_7LBT<`6oQyec$X#h{XZ(upnPTtyABBneKY9v`tp7Xy>Po}faibOS
z`&u_IEyKKw3wjt@cT3X5D?Fh!K}S=wYx=P|!C!cFKZp>ud-ZPXvB%XPRNP;sP)K~k
z>l7Uk#D8n(+MsSmz{!CtFCj(r?_{iu4*f7Gq__Z{%rP4>q72Yq;zfAU2dDdn5lnp}
zPH2I1p1(bvujWt#T@B`ko)3{Z(<MxJayRF6Co{CuI=Y_&&p&S?6=#1>e7W8CBBjTD
z_{<Q13<Xo|i@|Z_sV0IDC5X)CA(<3vQ|bSvO{&trCY2XcKcn0}5Y?zt7S8Aiz`zLP
zVbXJPK#7mGO6rm9l(QM9FKdEq7)`nf0bz9{*sdGQ85#R;HeltHX&fRANz|RRTi41b
zO0ylKOStFF8+!HVnnV^^CtKT`B!x(%qt=h#gVW?ChufY8fDwn3#DS?VTUWb-Y)(LV
zNqAKnD+e}9X)2r0&&IZbtDtPY=p?<I!+{C)F9?8kr)=$>hSo2^9#`q(ay?E2%L;JD
zpIi7mA2gu<8Z$c5rhkgIk-a8O7-uFv)HMswNr8o}=Khk1b5}0Of<Us1maESt5+K7r
zuj7OX#c`=`d=k^4Gww_!-XV~FTpL-pX#KfSs7MI?wMep2$v4RmBAxLa!<VC}(K>?n
zcHNu`LDl%Htw~akHDI6^0tn<6TF6x|sM(f7TRx~~u2v<`oN2L%?$mY|h;X>K?JSaq
zJAn79<{WLP+WO%dY}|N1t~(5^@;WYLMNrIs(_EE%Nv}c;4CRYmf|4LLT16*CiJ0FF
zDwH!T+J{vy;^}L~D<7={AL&W$_X|#u4CQUv%vP$_tY+h%ST#Wz5EGpN#k*1}SZaCO
zRFnZTm}X?&OZk5@b(%*J`($pvV*0(8)qGTl7CR`KK1_#bx{qr?Enqx%9k?M#4@{l8
z-eyZCQoyG<bMU*k&Ywsjg2U+IgFy7@)hlD1dLV!?@MQ&L;)SFrS=u<>FY*M|!(_r^
zo`p_hb4{4W&e|Hu{sDyISLVZFLqv6JRO}cH5!xyqWuh4>zBAmZZ({C$)Z0*!sLEEA
zrvZd;&wYZ9?>+kAB%~Dq0Hy_T3vimR>739WjLnc%bDTihg{A4F=-?EM_MNpcBQ@v=
z?qk5Mk)IQP2ZJr{JunFTy*DMyFDz&!k8FQ4ig$6>MDrQN63hYNw@uo)Y9uHihVnSm
z2WGj~+Y;c5O~_up)9LnBP+&VdD|5~cYzsvo5VBpf71aph7>La~d9-4)tQlY#e?Fcv
z-5B0t*!8_k5Z5q$yISb7B2^ReQag&WCrphMP%1aCFeqsf66)_O;GK~KA@f?LBMYVV
zomj@D#%idGHf)|6rUoH-s;`}Cu0B|5_?a7Cka-G~RlSG;RPbBaagOnghVY&qYa{}M
z6-(_EBomJ7oiwn4bKLWEcF`2Gn}dtsTZtZn&C_QhL$t~VJ*7*Ou;eOibX?ywqTq;0
zS-rKM2Z(M?ZW)(hT;qwdAYkaf@t>;LB?9=j9-~-MY5q%R@G~JAK+a=Ds6Kk<7mUG3
z5Q&B`Gg@vTN{}XGS1|~UKn`xsE#qwnAs9I5O^f_@)~sEY&##UkH)L0gR%SJR-nU-~
z)FDr!aUA!aH$hmvTDsP!b6Yjz5-sX74grAsu9Dn5tdyhyGw9T>zBXx1PxM$)!~pTQ
z(TOUjkd%gc`dIJQ+g)!Bz-`_IB=OBpLu|JSS!5tbYsD*CuC2DD9u06Ma=q!PuIax7
z#BBmlnnfAVHy4-+O#Tv%fM)9SM)hz=oA*a~k!M<oKJ(MB^;g^ao7Zt!AHt+N>eDEo
zwJBSek`!wjq%b-Y2b%5Mt@;sjs`XeJ&jBQszCci=$^;RmEZdjhJYaMXjYi#mGx&f*
z{kzeKB0-04vWZIDw`mI(etQ3`aP5}}tF;oc6I;f!9g=}S>z6$3KyHwMAc_MaTzB-|
z@znR;@wq8SBQOcwF|eh$L1|t_;>)*CQg)Rz4NMSLqGIwd88WwoP;XT=lwHrtLABf*
z4R5$`J>!2xXar7a^lw9mQ_X799>cRNdCBBcWi&{^5_;C2MiU_p$(v3@-QDzpr-CTi
z%yQ03iEL3pR_U1q3Jv3as-9v!$XbfvsACPp3Z@4dL3>|uPBxtaUUouga<nwC(&fMW
z@wWchD3<#33K)qEeT=_c&Wb4IhHhgV7U=fmAmk<l&`n+EG`sC0Xf5ym!tmr4b?A<S
z!E6+Xfa@~@L(szCtmV3wsxecf0u^NVEQua}1prE{_kr*F`cd^xxyyFq*YbzP2d2jm
z;!GF9=SFtW8j#&?kIiUkFf<VYS)ig%aN*zvzBhLm5_+}W6G>HUS3T!w8h%=X4RCoO
zi`w*70!Yj_bJ2<SOMu0n;16{~kCQaPx1D(J`l;R0+wSx}5#o0-`H!~O-`-zN(Wh*n
z0e8SBnUt3AUw~Uj5ujuou$Kfnh?j#J@aqp<v@3XvbZ@V&iweFEKV9;R+3wew0`FT3
z8G-443lo<zs0HO+owPUYiTaWhzM2{kYEhtBlv14c?68NtnAO<z`_)>be~QXdo(1F%
z!s<OPDD{k_uz*+@P7sY?r0_-mDTna820e`+HivYyk;UE|OosJnXX$pkJNrC;e%_C5
zdm=CyPLy^B6Lt56>g4!*^H9y-cMqedX$%W>?F4E$34<Y#q5;~?Q7vNbtHW{d_|ehh
z`U#0315fSul9tOuh~iQ*MLchr)XSx88>I9NTOZ5pvAZc8ie4K<pSk=2$3NyvVKN;6
z@Lc5gUcK1NvaS&?=u$lZa&zQN?|R5~A~iEBX3z6{NTvxa6c~eR+EKIBsStK~nAP?2
z*Q$V9iG5zm*E$=Yk=lul>y|-;+x2^ka(ar6#4{pCS5Wg}{FDXQRyjJ(=r)~f5O(Dm
z<GP%hQklGI*OZjDAMeEbiTDobGA&0$Y5pe0j&#s5ZRmb&2K>Sd&8$xNZjWCId;=e1
z$k+DFf~I-YGJE`i7JJWEiDC!$c<f#~nt;S<(7$B7x|yNJT}kRe23ERznL>&xq3mkN
zv1#}VY*lH_`FvBs>?UpXeRS)CL&rufIEC8`YF#|pXpQP4ghU&nB#l;5LM3;5c81os
z*+Ig{`G9n5FA@*n5}^YEzUR&EVWOqSVft!0L5^#JlK7~lyp{z?hrLbxaH$%uQ|B2P
zk^qs<&}8@d*%hwb_WdFpRixQhpu@%TJTa~KG?72_V_lj*$3nkl+*ppShPHrmZTwO-
zSVDQ5y*2CDrbo8?a{F*Vh)T{JLj&TgnfT&?wA3MV3@_b0Yun#ZWt5&~koB$7ES-g!
zy!Z{x_p*OItKG`wv~w}}&O2zqnDvH`e^N}H8@iX^X|ArO;VPl(@2s0u6%yU_mVzM@
zP|cS!F}?`(HEWYNtC{?=`cDmY`x9E0{jyF@HQZh+y?n7qA9_Z<Wh{@XWrB}VeTC-E
zp9K3~mkLu?OEccy$wPrJFc|a!`hOit|I01mVECWik|_=M|JyD3`I{rq>({St5VFWG
zZljQExllcnTqW)TV6I=viVHovcRo62@Nwz6SO8BMV=y6u4iZKYame-Ri7srRg|+C`
z@v!hV-G{xw=(X>dfPF$VupeURC}2Ipj*o?oYm(lPGrfb~sUJlHV+co5e9}j_RI2*7
z=kx3H)v3jjv(b>Te#4pl>eK!g4o&ln`vY4etMeToTl4&^bkMRg;_!xBj!|yEkTrjn
zPJ8DSvv?nl5yX$;0~<jv?Uy=acH%{CHzkiHeCC|0wSPc>TaK<D)6++viegYIV%jlz
zQOLlHS_@GR>^o>5B~@RvUSjJS6=|`mj!4YDw~~utXqgH(u0||8lt7AXCuJ1c2f<Qt
zl&aKl^)cwkeZU<8XnJH6e;hAAbG=zyWW}CkZ&%bFA;Ed7kqGzxS0R+tJU)^b9-JQ}
z7G`9Kz$M7AdSp*8sw=6M5K1^^#ymg}sL3r2rucHD?MZ8YmDK?{kx-6cKcGDX9DBG5
z3f%(C72?I)Tg7VwzgYM$`W!@wg=QCDOd9K*mTC^{hLxwzNsV@n3m-a~GuMhf${5;B
z&>b<7!pCbq#za?g6~PD~ATQkf7>!F|!I@ms>Iq|1giR{5xG_iZp{i*mKPdod4XIO_
zIJF6pSBVB`7}9<RX>P?gRh;V#6@Wp9`p5~Fu8T>(T2O#=#v;+?V>G{3-gZFNgZ2zP
z2L@g!XfVV2+ibF=G2m}Z-@88)MtpEWW47W^1coEpBCY<(>tG{~w!Qn5@oC19+TfFE
z_F0BP{eXWwNm~I7=B$Uz5vN~duyZ>r`RM&)7Z>~Lm4gIVj-vTXeUeT%wM3wPOe`U$
zK|yBePA8+vT4gOMwmU+F$`NFRD-Zz=iU6`TYS+?G3Fkl>M1DG3`6&%X!Q&qZN%(|7
zlA6I>u?FMF06$>@@7af?6;CgphR60E%9&mm5?WPm_0fgDREiZut)!du<N+d=={WhB
zy1#4}1(&`#`iYgX;9Yo2|I^wga)c#ZRzp~Wf1}F?M`CkDjL9lL+{Pgq@HQA#j$c65
zheoh!{IvwE73wz%C{DC8ynFBZj1-aGn}sB5wJpJ*pyh<n@m`50b#ibA{C2NVtMlg8
zm#f@ReQ9*AR^kyo{OW}RcFg@`yN>>2<x4~OQGN9!2?1QRwyDWEL~*`!HLOqlq8$^N
z!(P5Ob*`Qh&%BU~w~xW2cwuV1`qeAL=PduI^tLXpkNIl6biQ#>{M5}7E|bNjRA1cw
zT^7{LB@1N7{rHxsP3=>j6k)rkd^wvmyk8F4wRF?WAE}gT?pPEj3(C)549Wd_^icJ$
zU)gsdOz%^5lyt;y6J2VB|B!@**%6PfG||gAOl(rZ7IAy92!6|TmRwS~;fp3cD6b3{
zcK)aL1%73a;;djrSu}wpK1DvPhWk6UXD+HNRPSB}5<x#@?hQtScERL^6R8K`k2F?n
z%sjZ}t{(R$R>omg%}&LTw{C~ILWzhl%aEsZPqWWej-)SL^q3#uRa#S8Wm)z@T#lH6
z2o}E_fe5yH*I9*11fOB99CeyZDSxz1AO9!B7b1aQ2~N1cp5mvnb0d#=E^wS||A)6f
zFp4Yz9ea^NGJp^JdcM=I6|2YW>eDon?`v9Gxp7Nz)lU$#jQxGZ!nx|i32h#JWg3@Q
z-p(_ex;}l-6vY5OgG6a$?<mJ0H^?`&i)*MHZS)yGSL!!^PbEK}MV*J{=81K>aaMc5
zJo~8#hG`CcuSiTuWsYundz^D2%H7Y*G^F$RGE8l|;Bqdq`QK;V)nEad+VHKW$Xwy5
z#%veS5v6$%==N&emIn)Fuy}*8I_;h;=8EHXbwDYK_Zew3fGM1Q-`zJpv**OzrrA<c
z;Ht{tQ@U|>plz*SZhD(N+&nQ|a*`W1Wwf9O^NznM{MAxs-Y?=$WA*-P$(4dGq$@jO
zEF=6vCZ&?A+TEK;TlRv(I^+h#{o;j<m4FAajMt#@>a3lW4YDSu$#(-INrT2>3w)aO
zEFLe~=xL6%np?AAu=~u^pphiUu@NpEp9+JygY(V%x+5-YB!zHGYwEjeyJ?KF<)8#j
zhUGes;zF(eGrfyvPs`=w&yKeOq*cRPlKs!Ie5Fs4V}r=D-r+GQsLz)i6|ZgFeflD4
zv+|EZhKaTIy$U1Y7xwtp49Dw|fuz~`bn~?K?r}%E8T*Xth=*tJDG*Mzj04y2Yg~%l
z-&U-~XC(1;U0kj3y>D;-V&YoE^Bcp|WD(gok(7Gg!--RLwp)($4~NJsVDetN4(DE?
zE-a+1=r;&`oU5@gO}X$*Udj&ZiZpRM*7f=UC`N$m7JtE?TPXJW7{=#L%O?@Fz|2-Q
z><Bviz)O3VmhvF_N?W=HOG~H=3|z?Encb5C>?$i+80?T<;o$SL3|$pQ)7O*?Ve>6<
zz0&|SAZ1AhwQ_6)QdcG@8QETfzO-QXvszi+f}(1rx<mN)Bo28pYe)U!P8Fv|Zf$0J
zCWx2RC*&uO8cD5?htYW6kvBHWzLl$0TK`ebS^lH7$-vIU^uG?nTxsh5mU9H(ml6be
zy?-0&e}u#17aI9miZny?FEk3^FQ5dIPSq-?i{1~2sou9l!|Sce>uBVi;`7|Y%_O+-
zVqJT<wi)EOR&t*oc0<Q@Pe}yCumc$w!dTPHQ_Yga0hnoiqp5D3fB`{5h$8vEN)}d3
zZ9^{|zlLU6pT^cU7WPJKbH9(5&SFjJe7HU;&VLSG*alt_bYpeW-4g1}0nHFu?Zq)L
zTVKrpsB{Q0HbRWer%WBHtv#PKTo%uaZq5Bx8|4JCh5;B6Q$YYL-6Q$uBEyHeU6>7i
zJ4j7I77cwl_0-GIPOFK2kQhJ^Zv=TK&Ru_!kLl|WwZf+A{ey?ogQa%aNmI5=`5ml4
z8rei@X{bb^{sGdLixj>Ycnpa)=@pCtv-;`Bs&c)&u1%jI0z$0d>q;YYCS<P?^kS|w
z_D;Lg_+>4n=t!BCn|xtw)IJURKNe;vttOW0<9t9`+=ze(e92>Py%w{YR<)hGoS45{
zVPdBKfe}g4H;}oM3`DpH?bX8phISMXkmZMnD9)Otmd(!+=w&5Pp$Ya;BN*2efrE3T
zYo?*ykA%{dtI@gTg5Wa*eG+1l278&EBZX$Ypt~c{o;NZx{2H?h$}7`VIG!IC`@tAW
z5uTWB;5dTKA?=%=3NU^W=hpN_DC}gJAfkib59@geGzzoIf{VL0-g`pe55*h{W5|^a
zF)SVNl2fzW%~lnHHn3Dlm5KmmT-AMu{X<|F`EWDjeu~bEap5lEK0t$4+#R-dY1noC
zxIZL};btS^{q2E92K{~Rfam78o!LQQYm~xZYgJi#S{wyc5a)okwmJ0*{Wldgfx(|3
z>)yRUevD^*7l&gq4B><u8Uq)PhaQkl0gIMHz0}uEahe2KelRD+alBiieEJ-72-CKb
ztvG*|5e=t{aAqUt-jD6`K;q$seVraaK!j<tH>iw;RcQ!YxE9pz4vCOg>{NXS8zWx2
zQjZH1mQT+(Z3i?C{g`(PlD-&t^|_jFk9cbH)C}>1`Ovq-@w{Fe=pi!lB<c0#VcEz^
zNt&>?1jf}qon%qFl6@_<RT`p_gF%W9CNx1CGqiGt_v{*CwQvQU*#W&HC3e*Xs}!zN
zHt?ZkQ4s~2ca4dumhBU60m{9F2!wWIB&*07*|te49*N*I5$uFHowh7xXpH%yGfBEw
z9?(75StWgiQ%Ip*bY>|i9v6%^VfM&DFEMmaI9Y76Lljw53a^(nqa;7j+=$BB5YH;z
zWT7ZSjETWy4ujOSKySLGNN=WNhJz7ixjC-XnI8oV@d{SYK<;MbEmV$JgT~#S=LuT4
zU{5M8-P`#zUSF3RQN-K@`G#ky5vJsFiF&7{Ii`a5WRP|`uF|`}*GmhOG$MI}nU>Y$
ztE7<Z@p3I6dV0fYeS;!mE>)}CaIlloFDR^U_v9MQL$P{ts=>__NthkzMKV9E`hCqh
zL{S8MrzzWI&*XvYHch=!ZgQ($L=}7j%UN^^vEKR_DiW&$_j5rDPo|QA&ywuNiYk-6
z6p{>X9nHSxDa?EQZ#OAG+J=I>_2+|hS*d#FXb6b}PC*zgzf*xB3>*+UluOe~w$1wF
z!t*Nu9fCa|*;R>BAoSJ4vH2~Xa!_Y*zC>|d+c%F-C)XY|xq2>XW;_h4jnE_<Dzld9
z)RNmq=>1xik{RB)=}T)Xg)%SI%>3|m``lIH%%OL*-X^0$Z5nx8l>mhC{ko5s&d$Yx
z3K2t=c!Qj9dlqtVO9r7f9k-b0k#L6%v%6T-wuJ%828dcr6B|Q2c*Knuhvy03VQ(%y
zgfWG7Nvo)yFB8R9QT%FA#)w?fymMl)(@L$|j6*x=V6W4pHaJU~Coo%)TWm~-b3a(p
zW%6IT`zW^LPedX0Nlp81MxOSt1j}s~WVD{kMF=;|r{QhXTL&5_iETo@oW{RzxV48{
z27C9na<r>)oU%0VF4^&|+JO*&&DJ(8$cENr+1MWL?mFbY)BA_>(x2lYo6JX5+_3o%
z(@{&TfjuUGHR|LwN2sv1P6NuEmd5#O+|QQF<uYuNK(Q*7sID&(T)OUy`_%E<@QT}I
zRjx|jWk-pXg~SKM^nR=6y5|~n8vLz_XiQcVOv(WWR}b9JBAw{Z<IJg3PB#hLcY&n>
zp)dp}d{`!6(6FXlg=v<vZm8`#!Q?iu4twA$AFBZISZ?wA`!*68rK+3isNKkC^qaT5
z0pNveK5XTQ=Zy%bjc8eosPz0X)|Mj_CCZ?Egi=FT)py#KpQnLo0hq2F-ld>bFHNCW
z+FK~hhH<7g4GuH|S7uRu=4&0*fjGPniT-XYKcX~)lIV}`ssTfrDS5fwbcGUl&Jr1v
zjBP&$D!k`*?&4QxCEl<Vwq}7#N%cLWTa#-JMXin;Wi{n^-I<FQIdWJt3uC)~n<dcx
zBe<P|<-de9%kSO)bNvsf57mv|;C2M>7uDOibRh&<5V3O_>8r*pVrg9G^|ZnRfAUd-
ziAVv3x~JdWK7Mf{qqc-@u|ps=oXcz;wfSA%+`fzod^FXD3^5Q2aa7el16&ByQ87fU
zF}*q{-ry+FgdVh>NgE79vW3*bATmgH`v#q#$Mb>xcNfN^_>j!FF(pc@@9(>l7u#8v
zFEW@O-zT&kogTM6e4nqo_4b~8JOmQ+$s(BHpDFL##o4Qu>UJG1?>2*48dx;~kUTqq
zF{J){=FxE~OLU3;PvYL4VUUZ8oo@7lAFjOwNW-A}Zy3t3`9xwyNHh|oqhcF^NeXC5
zH$8E)TWe#^Ec7bxygZ%d4!Em~cR*|H=#Ws-n@_Asj40!$@eCJhovj1I76$hBx&_qq
z?K7Le=9ahmxhbE1Go<cJDbOQ*%c!Y$3{RwA{B77)8%ltus6cJ)ZA_-SvT4V>Qr7i-
zd%ykG-fZzj^OYAXRf9RS8wB3Bk4iuZrc9lGog`d`)zdQbD})VBGqPSCu@*t~cAE{}
zLk%){#O<yw-3I2Ou!Py9k`P#D_EsUcUR0riLk0_$R2xAp?lOEL3_dLcuQX;6!d%su
zgSK8<rK_^Z%*HezWx?Ln2iuh-j>eJo%Hc#7_A_h4I9CRdZV#hg;vl1Q<6arzrCM*6
z&9!tW)5xPi2?tXADD#k-SEXde3pp=PLk;ETX^l@VPcOKmxf7_H2F^Iu=S!x)y>``X
zBaPEB?r9Q_koFVrzfoNm4(`FcIux^>G4!BhG9-QtmsLH!@&_>djnD4+EjaBavx$}0
z9?Pk5%};@c)wJSsMSx&>2^3^ve?A$04!>87>ECeLIqI;2M+8(9*66@65n^ec9c#+F
z!K3P{+sfkD5kWW>KlLDINMcdQ6r7uX;&L?8p?EjxyiYsCjo|(JiKSt%D~Or}_N}X_
zJpFE?kwEs;-U`j_L-cDPd&KN@YUj(ct<LT6@K5))3C{EwZzvyv>;2diez6f^9P&E7
z8VO6ZztgDX{6J7w{lI9^&L>8_4B5cF&e4PH#JDFm7oF!)Ois=iRe3Hrm`>)hGwQD+
zAc+%P{GsaF$*h0SSO}ySAzTE)<)G|S-tiYjF#AZ`3R0co>cJF&dbl2ts&eyVF6iT3
z%_yJnVAn)cUxoz?SpS?t9J`0ERr9YZt4jUyoLky@QU0Ka`%wV5s*s#G=1y#JfTlY9
z>p@{>GR}lR`!m2$hmkv@q;MtJzVlvsM!3@5Ot5WAP=qL7b_!~yE{KE^_SQLxFZtyL
z24$u~sMG_}Y+Rs`=`g+rC(f;6pNBQlm=HhM54|xZKOGcwzj3$*hF?{<PPWAKIu;*=
z9A6r@BT2b&ut!a3KN$+G%0bF`5y1n*pJeyXDwsWWrZfmZU5be1(M#2e?f@~4fWb*5
zQ2D*JWC`q}kvVjpT|{MWkRz;B^+1P1Eq&Fa*4M4R>5c{pTAQbq**kzubS`j%+CcDU
z==u73(p^Tbj-8M?X~6iJTh`#-4~Yzi{Q`&<&OoXR!^V%qgzMj%mg#K!B_9x9IjsO{
zH~}S|dwI%&VD1PGTGT4**3`wsf@QU7rDE7jqJwW9#i&?;B=GvFsXOG54I=4>^0^Z6
zmqbljor5b_$yb)X@aZWK0E%_A>@niI|8%9tmFd|jh+X)w`DH~Bo-6WqHDDFF)DH8b
zLxgSUg4X0eKPr*1J)4c<ZI>tY=S-j@xs}P}lQRuE1T+9GwX6>M^L{o#cx+$>t6NY-
zAHI3~3`!)D?tw}Rx{BTL`Qe_vG>Mv$wCAd^dCRMD;f`&aDGTZ_n7+RSp#THocaWrT
z?~|WzytzL`I>*A>S9jI8PMOQZ@5`WIb<Thbj`@Ro0`a$f&h5Bn^A_IS%y8swUyQ&E
zT-vI-(;hg2FWAwplsOxa>3{l(u`3rw4%)UyoAAsy9!2yPOo``)SoLp{j%{~{GWHDj
zgQlW>*jSlrkbq0jSG|Bj9SQ6!LnJ*Jfw5$d2NN)<MH!hsBkDzrs4?K8cn0jN6~vfd
zioSS^u7CDAzI5AN@y!22;j#Xg!lP&SpA}wo^2BeQK?!;BfuhYzLm0#@bZyq^5U=``
zz>HbL+Px^$jKyNGbQ1F=H-LcWj7dDWGyp*0;GwU7zY)Th5yP)(#ms6_(8sKY3CkG@
z6oe*H!YafHF_Y6`k}FQBJR0emo1ASr)^;XjKavzPHlgs0*%kjU#@;!`mnh!yZQG}9
z>$Gj#wr%{{wr$(CZQFLAK5ZLw?t3%yZgMB{Zc<5AQvdGM+V$OQt&dE{3hFP2Bho@+
z;);4Hq6G*P-L2ha59a2tFwJRj;V35h=$c!aR2?F*e5gGiqiB6WIAe?5{@E>Gh7_;U
zGvl2$=FCDf)?=TIW#49d@5eebWi=n=G;$f5%+9u2yYex0GU!<m(%^m_C;j#9X4N1=
z<pbg*ZR(W;9LE6VoPo_d7MTz8jt4`En;Fv8Y<WeMUv5YstD(X{1gT6`YYV+U?rz9F
zF`Q)OXCJn18Kk9Oz!Ov@Q0!M+jB~nxac97LJtW`nuwVXW;xG=O5!0Vh-)zT2RW-R&
zuBDtHXzB51m7AfvGxJZTS07WFnT4y%hUW|3O}dOv7cCXC`q54xx+}lqMMyq&=JqzM
zn)EHsQ{p-qy8tI+=tYG2!mgFjGxg}y{Yo1s7W>Ka1`d-vn8JHDGWw_SGD^<My&U-d
zifikg^f7-1WqmVDaY-M95s1j437ZjTud*5mQpO4x$c`-o@_>%x6AMpA+F+9~Ouq>B
zlM53Fv4b#u?|N8mbyuzR=ZmiPH&l}`%;y(DhDrmf*;V*!)kJxc(ifL+p1}Q$#?g7R
zgMNty8rAhu1i4nMA($a-N>XU*ZcajeE84Z{GSBF+&-H@l&7#lZ&MHfz0z4-S?+4&i
zdFVG$>H`r0C|@crxlGw8$SQt^TFrGWQxpnFImlvhkC2~wM{}}WjH+*Ql$VfxuWhm|
zfz0so1u&7<S>6O5e-{<pt@)oEo1V-Xo2vT#`JNjq6hg70+RZeQ_D$#h#77*27!Za!
zaAo|#;$@zG2K2JjnD19!*sNV80Bvn~NUX&w@t;wVSTO2O?;WRFIS})n^s({VvUeV|
zLj!#@{qAy#r-GBZ{F08L5RF&DW>d;6wcnaa*mKG-svN@lQpKzbB|h_o$`Od=BOlXF
zATE;wVT8msWP+;1>*)nGNur6r0y@YrHbqw+9xOUK0nR>;KvJu-@%r00Ej$w-NaR>K
zXnSj*rL0QwB2}~pxs-Be{?S9gE83yU3^=D^z~m9_yP*=tVHS0qJo&o{DIbtrv^g)A
z%fyF^6(%*nameh3^9NTNndV2=+z|W0W)9{kH<5xEngRwJn=(HZIN|H23$C`&KC3cB
z+J=;4BeTY~%Vf-b&;rhlN02n+2UE1KVz+OVcoMo4D~6E$GiCpvsZb_2_^Yw|6=RHt
zsp!9cML0MZ{*wj%zf0fc`0t0RdNghAw<Hk$d6gImvOu{qF=k;g20(V`8nyQkv{`uf
zpEm~`Y;_NNhG#dV=s#|53KBb0RmB@^aN{MChY#WrIZ4g4l@-KJh9@(sv)Lz+5A0H=
z0B)1qgD>UY!@S(0$Bq<e-lIE}@y9+&6Pb#$)s4`Ykob+%O%)!1*hoB!=Sl1|0Wi}V
zCqm+IAjM08dL&Ivd(8&M$;1-HQ!rx8?txpNoplXU5<yDxCg>&(`AIVj!F}0SAps^F
zB8F^hGns+HNJ$3vP-L8S+akEIyVR3k5aSB1YQdgJ`03+rh-QH;HjIQCZ(zzr$YxM7
z3X!z0iSi_&z;q=lSyzqT6hwhYAhpw`0Sx~j&w3uh2qFr*o><`hwu7iZQ`{L-rbq*E
z>k;xmq(hqcum1g6gr)IFs3Q7EoXfM13lpet#DPExe9%w~i&0LPv-kA|!wiYhA=nBj
z)&5|b>4K+HgiKI8@u2%OdEUIDA~GD~M9{I6V8_8Ufa*dM|E60g6gYYaP$UjQK(vOS
zB(U_^FC=9qqOkCNV#8;`MDIyzq*J4I>R=kdNX3OwWJn^0K;&E^QK(>u0)+N66_#KD
zAr=ve+yEmW(D0Fe#NmFNn-Bves^4;Aq5!+$iNJ8;u`0@2Kwy73p%IwiFar`W7&_?C
z5+jibCb1E^pP;liG6+CK7Q#D$Ic|1NeB^`wMoP6?izaoY{ME7}Go=a<;`f3#wJ+^1
zYTcbb(hm#uZpxCBj{~FCEDCTj@8jLEvq@EV+$IHL@6^W}nhCmJ4*j=oHy3}MIi;R#
z)f+%j2i?!_MeJK&M}1}IJ*<$jtSu?+K~%LC{g+|J_SwmKWqa3;H*?;q{VF|0)W(!c
z$b|{3HpX7LCk+<$tna5|Fn_}z^Cu!u(4yUnLxR4aUki}p4O-iK+uoZ~p!{9c36NZG
z<Dysl>vVj>oor^}yCVR_iU%iE9l|j}i3?y1(u%W)RE;sI1juNr-P+WpDU|xE!>c)r
zN_Txe6)TeK1s8h?^^<aU_=S@~Es`M*EV^$k`Wy8nfG$Pv`^w(rhWGu_z2FKyd9@lC
zwa&9^<Df=LK+>(PX3EZkgm2S0@oDwl+?I+CM>V(o)tbBpJ)K=I5wC=+&wJ=`_K>5(
zF&k<F0jiC*!Xs2=_b-|SYt1lS?+r^SuLN6z4S|k;OIL}4_yQWdfACaT9#`Q3<2`u2
zpfWn&Ov4?=yVjUVIqw`05<BB$2;=0B+7>-z#(8o!+CJ$nK)7t&ib;%S2``GSYw4w~
z?R>ylc&4<8vp|K&PXS3J0eG3P)K}FE!NrLYPe&H-+0XWD=1QQfn+==CvwD;+A0qEU
zu*-c1D(mLB93Kl`CQ{SSOXVPWTal3kbxLg;&Rve@dEm2!s(czr+wTPhm1^sP`|?pF
zm0q=xKMIz&`6M!2Sxf)SS`5Q%aLcd|uC|y$h?qisC+$Px7Lz*DJKm#0dgsSY87?q8
zHfJ#nL0lZpIA}gw##^>a3V9u_nQnIFb2@RT=dMb6v0T4g9*CZt&Qr}ntv51qYIWE5
zvJpnMx7bd3P@U{M8SQU*n&ezxDamC6xZY@9ZCoj$FHfKQ!8t?5#|-hc(!KaGow(jy
z&%n?DTZ1g?gWDvP05=u9<z$wuyTd`84FLl6%!-^n3kTmK{I?j_`-GOwS7Yl3Hk2+(
zJpHx`JkDiiK;V|`(_cktPDAd}lQcDsiHcgXtCRWog?sFX#&VzhA;fgMvZoD2pGs%$
zirT2H4fj^>*M~}-=p9sz>eIZNsuL@oUOM~h(IamEX8f+I#>&&Esyoaebyd!#wUw)x
z)!72W)g-CP)xRHVo9_ozDfl_U4ge)wmw07mgtOynz^KJkV}pn4*hC4nuV$}4wa9vu
z{!fdJT6*pBwG3+;!?N*;-2)h?O;KdA)1&zFa+!W-K~7^3KP64uG(wrg2K=uVy7__U
z)-*$jpsOon7^0hGuV%ULb*5g#qJr{aS5y&x7EWFd3Aw{KuDEasG3`IdyAXKwADpi|
zht`11YW!3E^JSK|ZaSO6(IJleX8MIkg#=uc<UOl;_5^+X?llRTT$)yD$lN*M0e(As
zwK&N8q%J@UT`z||$H$@xX?;#KWdd3S_d7Pv97_qecQRa~Y8Vey=jBB>+vxKVS)xAO
z`(5U;9+j6n?;{ys`)?~7E(8ux`_;(5Psg>?>V6*7X!LyVsCRjM*v7s|Qy3k|08Lx`
zKUr=+yw;N4!f4B_v{x}YeArBYA5)ToO&zNW*bW`rCOtb$ava*qncIyGzNFr=A?uFq
z)uUb#&j#MBteGsAqzf~r3>H=WRg%s3>M73&i!O{5;3b=?_Zf<$tF<L^H`}{on@4_P
zAZP0Ts=k~Y|9KDpe^Fmfc4ip*{~1tYWM}$sN1A&yHJyHO{zs&Fo@5!Fp_5e`Rh8&c
zs5LPJB$YxBVoa|4RckKzV$HryXSLm&H0BBnzwqz$!q?MIqI8+YfRuR#s)J*6R$_|j
zfXKkXWX7_4iVPRD`bcxYn)Hw}^i5hSR@9ud8LY=(RG|SnAA}GC1rKnZrM$)d4fGJl
z9uKIa4s5a1D7eEB_@8KLMPhJ7m4F*Z@MP5fJZlI7hi*uEYuW@5x>|pA7?bm2!N?$x
z!M!8_OL;1Ce<CFa$|yjHsZil9hzLvt$9Q~xH((q~hL|OihSXsJn^PKo`IO3mQBU)4
zdc7VAw2n?hHX0!%F9=yW4G(pRKNErpCHym8e3Q_W^s){oc}iKoo)VS__-NW_5u|E~
z<2d+tid+#qP`opK6a-y~uy3acOf)D{zhY#X`y5nh@NKe*6xOj3i$aO;B!ekOAv*$P
z$eCb!BLtO0T_YJWP!LsFLX@fiXR={{bVYnri-5<E*4U$%{;N8Paz4k>;0MaS6;T4S
zNNy?wXONaw>NhMb@TqA#3MD<LOm&V>m%m|4Q!WS{PDg(J?l?F^sP-|W6;^<unt%az
zL~Z~BFpM$0&P+6jfB=;lOISO26QglIseqsaPPB-BfAr8iC<^5X)htcgc${?pyb!Kl
z1IRxMaC!<z5O(1b^+}$P6=jv2_qcYE6r3+`vN+I!CEoeK0Pm`%Jk}}96%{0^80IlV
zancwurvgk#0CEm0r7(4VGndDEGOx#2ZnsXaej^3_dXOiNRq51-@wWz#$y{y@&nsn3
z0RIOv$`fuHI2|h;JRKTcQ@zT;0BV})>!EQ0Ot+kaW|;ZIjye61Tkg<r9XrRjM+Q)p
z=YZS;K+qIs8ZuoTs8ie@Pjv_3y}eyu@fK3g=yI5<ejoJpYNv8lf+sJl%$U<;Q#^mV
zp+(%<^+yTuQ`ALh_D}TGn#UK^ADZ+Ba;RK{NM;|$AFvRV)!Vx#H+L7W52o-ZMyBg1
zyd9M)v+mfFVS;vVSlms6^rvcHER|lEIkPVMW<}Q)uMhAi@s!atsSuNFayZqiEO2{^
zalJ4<_E8CL1-{dc_j|H>9%9?B?`<Df=^;<C-?J~c`{57zVBb{&wB<n<M4e@}=zoTB
zFNYi%-qj|a<t{Rx^7r+X6Hr(Uo%187c5~!-{Cee=F7>zWJ>gu(M8L!Urq<sC!*t1=
zwbo!Fm-Rst)RFiY%G`Z&6snhNT<eDdWhvUlS(Mg+hEu0g8AjHzbl7|Ho^|L01V#!3
z)LELvSme)17W5$cvj5DBNsV1dINld{Etoe~uF3)~x%WmM*wzh@Q+0ahf~q=aNbxc}
z>k(#2u#R#Z1n=S@=AY9;SA!(?+0K0)xgLw3zv!Dw{baet{&?*~K9|%?SIMp2q(r#X
zKiSrcT^y5#iIakqh$aBB7MfNSEbwuo9tj+*G=%e&A+Me4qgBWZM?Dn0S))~W(;mxp
zFTebam%#GHiMakwwG<rjiFE6+hB1{<wK;@;(!(ux%4&){rP|2R5haxs&XqgzyoP;y
z>21$OIW1R8w%8>lwYMOw1yI@;Js@Y9NsPO_%Zri}?*3b7OQh}aFDTlFRomrVw7297
z+H;hmtwasHYK3>(2&~)c$p^`;k7TsWUo~^4#i+4fIdw4-bDkNl^${)+pxCxX%qr$>
zr5Cul5^3XC;B$kp1fEIn2F4lA5Z}19D#{j0Ej7$GYn-PEQr=QqcafV0KOJksl?Rsd
zZ<(bI=BU|!UNeKs7JYAIv6Z#3(Sc=TQ*iMvSiQ>5a?47uhc%}tS4<jb-JrR5#bCZN
z<*=c*pT6xdrBHP*ldWf7cDR=ll53izeC9IBdCsAtSu(_VYs{N1tHVMYk6oMPiYHtD
ztZyS!z*oNIJh!oR^x-M9lNy;}DOu;Yx|QC|LX_CvW$De9kgee-VZ}sNhbHXaV(bXe
z4)J7K<X?S%>2KC$k&w$~?S0~(ZsJe+oIeouFTZK+)%}2bsW^vwpd!?d5nW|*T%Xim
zANe5k{c}r_IqWevI%9WjSEybjLCZFZ1auS+UBTxVNmgATu9pe^gVN8jS8Ai>A?Jme
z?^)K!9LsdQ<tUhPna^a%T^g<Yb-&0_q-Z11UCJ2@U|k-eo)#Skwr}Gfvgf+-g(mh`
zHZ(x9pK@z<_nXZXiJiX4=Avf2bnaA|x}rEw=J%PjAtpH*4;v3prFl@jL9)0(<+f<@
z!1kqkXK;wt9=(5v<=RpI_1jqUE$O~S<{QOz9RJ~9;zxUr%Im+XB_kvAf38jZzsV&B
z)Bh)zY#je>_{fZww$m^6|0nTDB14hCqKg-b^$%TBmTU>ib21VE?|#lvyQXkfcf)B!
zbJfs%zsXpDNIBNE*zuPe!_pGF0?S;M1TP8x`$+}{GT17e<y8c_v|Ut)vBaTvQKsW|
zl4%Eu6qk7$sDP7skpOssVavjsI74+@;rHpFiEX;_L5k@vvH6we5RQqh!9F%8R{A7f
zFKZ$zwF7&-6!<fQ?|J_okUbt4=DtxN80;h#SwdjmEOoKLEpAI|CN$lkHf%D2J>D;&
ze3r2QCR`K$2qDSkNy&VLm{BSSdc{m;fq*uD5IQ+Rj6VU3<0DSw9dOA&lf)SG11$p%
zbpeeCChkMUkL68a@Ct&z5w#5#yn)Z9{xOO{=<st$+CgA6fWf!?ufZolz+r$-Fgc*A
zQTOFaf;m7c4JXfTWkA47LBPE%6_pn-^L5-Iba)VxA*Bxf`PWYMh{zo39Kk6oMCau#
z^1C6h1{rGH)BpG<8s?g7g|+z8*bZtz>r_8UP?dt>gApQ?hIsf>*e~e$7g-`9gCLn3
zED1tTngoxw)whU&<rrot@=P-eNIpX3FQ!A7QX(dgwuMVyfRd#<o|yyJGV2lZ8(4J*
zd?5kxP6gR+0;72&o;9~q#rV6|TL=Z6YE#V%ivfpVqWH^IsivC;01WXR8sM+t1p2kW
z*U|jtE|56~g>2k|%YmFb?R{GzhQNHl&zzx<@_9g06ukt>k_;gOdX#ZHfVgrHzMZfs
zt>bce4CCn6kNyE?>^6>+pZDL7EE+Ot2v|Rfyxt$r2FUhdzc-|3-(YdcIi^Wkh7E&<
zV8SfDKY+ueYdOBZK*Qu6&$DOcRSY&<)prkXdq0k|0TNopBURU*FOEX{E=xY_=ieMf
zFl5(9*VXC?OB{0PywjiSa<%G9Tb!?Vib-Dg7`VM6@PTqs{ZzNrjlXy=Q<rB2S`a_j
z+w)iL6^n{xUOn(+b-N0#lzW5xBe7&Bep4vEd2F{^4lkDRwfkr*qLF(B>+avu4Q-WC
z8YK9tfsgC-TdFvq+NO&m)3zLN316}d7^eI3e(WeJ;Q@ss^A(QTh`yYAYWt>(E{z|7
zhdip9xc`GUvPKnG3md(GsyfMV7UM3NbtMVee7Sc&xUR8DZ<~9E-=Gp8|NamaM^w`r
zrSV?$!6m$F`8W<f{S-xQjo#KiU-&M3Frrif(9?)bsZ#N0F9lfm{0h+#SuK*8X>-o|
z5H4tB4auHI7lG*i-bVWl^)$g(pM`J%W9?FNL=0E<g%JiHt3+k_qtd0S|Ep2;==R#U
z11Ysz)IY<W9L%>SQweH<-4x!cwwrFaU(8ffjp<q}u)6|C5)o7h=hvt#E*IT|K`IC@
z;@jGaR(tMKbf+)+NinJN)Nm(Q<sU^A!09W)I@W%E?o7`{xwC2!%<Vg~RPNY;-qi+{
zH=dVCjSl)w(2!$+?aWq<Xhp7_bTzrUg+LU;1c<?p&|7w`&D5WawB|QycRfYzvJ<d}
z$XP8@e4YXvzA3;5r>zCstdo=HXSrzQTWkWC^TjoO=IKZ9sSh^cOW3#8CqF7HSx^|u
zE3`P77rFC-MIJ+2M-A+z@oZD1^8{-d22@mFW6>d5u-!L1852uuW!aHu+MWKe;n2-e
z4H16N6~MCPy6g{}lNzUf-oldZpwHFkji<bI$1T_}R!y|X1Lw?SkbJY`16Zx8>Vr^g
z1#sodF2xNQqQ}sol;nv^0Y@3}vps9+O}Q-a^9{_ODZM+Y+0{rHH>x1FWxX4hHfhoy
zpEjJG9=rN&lBP9L5JjF5<;vu?6OK*MiS$D6+QT|VBoHG%f>e8mZBE;c-{tL#FJ{%j
z9W@k(nYtz1)2)^9>obnFG)hiZ&uKPlKYxoV<F{+pZ61+|s;NGtU=6fmzlmmLU-hmW
z_S!<r_j%z@sa#eM$BtBPh4-Cx8mz^$BM)EFU^km*C?{77`Cib%t8|B73IIY)fEp`Y
zu>>J#Z?Qh%n8sh<hNKs{siq?QyfVgY9QgB`KBYS^k;HH5R{TZ2_yoIRO;3ErMNjYd
z!pqGquxrb#-ub*EcXHxwCX+o`>KyjP@*8a_b;H6V>K|W~R=s~7Z0AH*RR_xaY}1t9
z?SbZudAX{*Wqz-05#`N$-aneT)FihFjktGR)!GR0>&l(Y2=CUo@kt~jx*gFH+>749
zzEYv_l_+Q}n7^%D1fJGYt@!j;K=t1e+&wvU{2x@1`TtO|&Pd44{J;MYnFv{!nK)Vg
zmjW^qGW@T{e=pb={@(+l|9>BXcPD5({az#4cB@ZjU#o3ytL@f*TwU`2Zh5V?vp>Bb
zJsCGU>FMR+k<%$1OuyBy!gaX|C7@4IEWo3$;ABOvY;r8I{$jQRNx9O+ijfPsj*{QU
zx;6*8u8#WqIbh29H-scG_s<Vc52A;D>;G+pU?T_eWEG_nRb@Q?ft^|(e%y;dX)SdH
ziuHSl(Y^fX3ayGLN(#xL^*_wJPX-eI1!~N#Z5`*gWO9{F8!ZEb`a=g>?*M`cl#K@;
zEo%dZTLa`;>i^?u;hz|m)D#i(OJHzye0E`XcA<Y`7}LNI)`5*3xO`pfb4nTT4fz>n
zXDws@2de&q;t-0V(W$}3wVA=O@MEr5$n_aN7TxT07AW$^n2C{rY5wzK_VLF;8WO3#
z6@v0RY9CC2#gWO;`3v%M?!6F#4Sf~=r)1l2?wNiD=qLO+uL51J5dE5b^d5ay?DS{)
z+UVY1V{oH?b5Npxb$S!5|JGh&WOaRbVRLX*WOsIA@J&B)k%{TkTxMf?aC~xgcmlfr
z{U$!5=FZO<_01gD7*@N<&(h4=X5ZqoFW?7e3P)p|>$lhX&rR=-XWP#S^EZM1=stnr
z4zL!SGaOv>59znlS6j&g!BuiL-c_~CGEH>O%+C)RIQZ2zuwJta94?R+x*9sr7(WCF
zC{EDaPccph2Uq4g*UwtvyLs}@vn=iR!pQLO&M=gj&&X^z_c{~r@1YI^W_uGUjh+}J
zo!|}IcI#=T?Stt+Dc*4fw~%ZJl+%H1HheXPmpgUnjH*9f$uDZ+p|=4AD&9~;a3^^_
zB$&@Ih-w0?M1X7iH@b9or?8@Xus}8T2m5dB@q<535~>hBo1vKT5ZI3xA#~B2JPmyK
zPCkn1yIo!NUAlg&OhA6#nJS5%(5Mn?EAt39ru0Fc*A@}>4K6^G3{&}-`ITY96(A^o
zQAxe#oAwz3FRg8WfWHH>xi=u!VmwiZy2rJsF;_*gd_j%j%1q*+duJ=?y4>+}>t)Tq
zYRIxkq`G26Tp)7PkCSByd$M<jEczFbtSCxeNeSYEjYD=UBk+0`pJGz`Si<ZdJVpC-
zmgPrWz|uv?BH<x+jB}IGCSu}>!*5DYWb7i;a*I_P&i}%PTm7>}Fd49gbTP242p=!t
z@-6RSjc0pILq^bPf{U%CZ&~Im@*+*c`zwFKusAG69acg0U!&b>H{-5m_WYx{5Mrg^
zm%x~1%Q$L&CmN`nl3qTqN6!yI_mO{94UL?=ERT0Gy+@Cuo_c?fRYtvhshTM>=^)c=
zs3W!G6YsS}&7ZawRNqa5;5k5&j*DGqZLsg&4i7!^@#H42#~$|z?xj^)a_R4}U%>){
zbgMjB-{#8j-c16iKeFp^eO^4`8`+lH@0M`G!W6{0^s$LOt;|ot>?aCOGuBjz*<R%{
zsjzTHFg%TJZg1sjex_UaP((5@{$N)_&8#Hcr8|%}%QCXG%qysjDOs9+<6w*SJyg(R
zOY<02G^k1K=-+DP7zt@$mYt6mF)}h8=lu@ui<Q+Z117YM^%@LwfvfUf770L8yR7Cs
zfDBxJ+u;C2sP*`Hdj<4$7r%~g%7-$I`Ks$@l>KCSGy77orAw1=Li@G~c*_de^i6dq
zf~pgF#HOeSg17M!w*5v0FQBOmSzg|Zy}2|=9HY`ACvi`n;DITNI*E&bBY_32Z*rrO
zK?V<71E7cc{c9qJEjwwZ$AGQ=P5RjjRLK@7OncaE-Km|iQg{}@3(`e_+?_c4GT>|x
zH8JkSM3?B(_oRUI$Jj<(&sk8&l!K%`aaI)+aG!#Eewcn)yQ)0nnm9J4wz6W)G?&O)
zRQ)4qN#SjI+b(F#&pqZu&c)q`uT$}5tI;Qf*tyT$BF~?aBg4`ay5QI-0Y90b9p&ar
zjyJe!nuxtQ@DVsuodT?0?se1_-GeK^ZK1CB>{xwOFOqA#DEN-rflxB5fOCE9!cP!T
zL2fZP{rBAWmrtI=P)#hKp>^0c#qv1qVP$;9^j$reVe)9e^-DMp*m2?6ueEe_qf4<C
zIfIr2wbQ=YulxXQXdxoe5JV`O!cFZvaToFSxH4#v+<dI7f?5?UaIr2}L|3wwMk|VB
ziI3If9D_jM_aqn8Cc-;bSsLE0tf%4M2SCdKJ7;=!crM|7$z>%~vhg7uCdvS`9d2>*
zt>Fr7bYWe7<~HUEcYN7yo%O70o|qwqlk2h+#oqoaAF2p@<kjY-w{VmR@Lqm6!EBT(
zq82ovz<R$=5vYQ3*J9HYTQ0^IbGV0au&VRTma@m*=ZqH*81sPnS$V1(!hS~%sj4Co
z!c*d{i%_(^=pV_hp(@=JUuXGcwi-PrD?|81hGGZEU^*M*=w59R`nny)j3%Y36!lf6
z%Uz0&o7wOuFnV_>NUHW83^`l^VuF`)GbkLVp|Uuja|@F}M~};Fb4Nxqb!Xdh=Q85&
z<2kc%B~)XZpkUb*-;l@Pje;2S)d7PR30}Q85VaHcUwt3Rs8&FTv?Ss%DP5q*&I~a8
z%PVObR3PsuSr22qpt3!-Pa<hS3#<*!(nqowP^=TEe{9Ir5anKh<^Mq;!E1c>BJFmX
z9#w&XY|Go6JPOyU)@BpFDi`=>4-kI?pSPo*6K;N0AfL}W7S?ZxG8BW$j?!OPb<{VO
zc9!$VMYY`K(m-PbR8f>>?LA9%)(_X6MWSQ;Vs3HODWXPNM+<Tu2Q~hH#~Kn?#frFA
z*T(}OoOK_q&NsaPR=6qQ+j2y>L$Ig8ZbznyQqKuXnmfqmPEi@D9Jk&G?&ywLy}U&L
zBr+Aw3enFdl{!W<X($&_Yq05jKDyiNclZ=!c2mD|QzR)lL|0&ud1WxX72Xuua-a{t
znB|8F0$PL?&oIj#$CX$w<OWRWt;#Y)4AQE@DNy*){nQIHjxXQgT_!!VjtJ-5xo&E1
zAYG2r<tZ3#bNqT@HI{+?;?mAL3qjf+SD<Dv-eb-N3K}Edhh4qee~yD=QA*U|V1O?@
z2KSphRYa}BMhIYXC9bY0z2Xwl@M+4<pqHA+8*jPl>g3^+IxaI3I%W8%i{?bwAi591
z4vOGY3tyQl0D#`i#YP`m_2IHmB*|)mPYaDcJP&IKaT8wDo~F{P_UNG0<}I+)OfXiK
zXwP?Dm%g9mzKB?=r*EHSBN-z;-oQwHkB6Ix`(0`Ag%~YL3%Hv?JOlJUnp%(6qHYfe
zd#B04VZ1I|iMNT}y<95o0X!R2XD!&YOX!6es)=hQxW^08KS*>cZxn5rP}o(_RgeHP
zbQ;Ay*N!}3sWrw2p~?3yV`*D({^{s4_${ytz8VbTD>Id&QVLzhc@!ar*#oUqf0wP0
zD|t6l7>dG1STz^i2%y?hE2nv7@H6BDH9$nc>Q0RTTxlGrk(50KyCN}<bmx|m^b4!7
z*uwhBio=3o50(Z)QENOLTkl4W8tGgHu&!uv8;DZcu@OPL!q<EC;^tpExbfVZ>KGV1
z@=8&yD7=9HkP$Gr`dn|q5E5EP5wv8(`!-1E&hQC}7QqJAoYh-s6f$35Z1RwWlT%V=
z<aT(?Aa#;6alNlBWe%7xir>{YW9i=uk=RGH5m!yWg?mPWApea=QtqhAQA6Qu;eH3^
zWR$i<KXN4Drv#u8NM8Cl`3*vtY<^0u2fUW_=yM4ddS&OAJNd)kK`hh!dD<|L>7Z17
ztJOOGt*Q5o_}m=N#EOC)yjFM1isyhO<Rh4I(+#?LJlJ-Y115Rz;=DDZEi^8mfXCxq
z&nxAsMhW8C+MY2quc|U5d?}MK39fYbwThF|Lp%Uik%^lWdS4CIe##x1ubIj07cqt|
zZ4wXURJR@OEE#=HO@{FHXt2cDYo`cEk1`Fz;d+#IX$Vnwm$O+7h&2yH8_EFUEL0@L
zLV$?J3P6;BHWei;M}vyu^wQ6q+mxX)ww4#?lYF@Qa6nw6$euE|(UMl52|?^42uWa?
zbbyN#v6Bd7S$_o#e)cUy%L*`0-)bERf*=QMQF^u}T{0<4I?Ar!)6RmQsexLut)2KX
zhKB~If9dC{h-X#q#zU9^tL&dGb89#UR!J?{b7SZC(_hx~#bQ*gY2ye(VYtm5taGKs
zfRl1lt{pXYn>u>ExEau41OC}O-+O#l9v8iy2vhgKO85r*V*F!qbvj~eC?AxwFD)w&
zHnfi9cN2$@n>!fTG~kz!txLPNnc^?#C?mEm0A=C~DZfk}XD|bdBE4iQ-fDk7mlro<
zsz3=A3Ghl{AFh@-z-I1?s+C7ee=&2V|J+aPT=&R~I%N(vr~Sp;K}Iow(kB#5?t&8=
zGncopGM#(czsn29pa)p1Uw7f2Q_-;j4sk6@f;EH_D|{|LQrB;@6-#dUg>+@q-dABL
z4Sh#K&Q1WMN4zs?yUJVYze|SSGS-y%gv13^^L8noywZl4Hx^fx00*a7u&_~*<>xWf
z-hP*Q*e8Dtu($hpf=J=~tUhXJuH#1@jIwm>eUVOD8-^dl0UR^PE$j7YzdnUFlX#Ck
zQq4t3pEMY?Q<&?nrCl$oL!&=Ofjn7EA*Xpxa8|gd9|cO`9k(UvuS?-%?oJs04Nis5
zKvc|);rj}&t<Kj7-_HaVH^()IY41HtMBrYYPVy00;y6vlC7;FuEresC`<C^K_}fA+
ze4c(2Vr{4pK33PK6#~%+#+Pbrj5u4LjW)G9iuox6>$9{y3J7BNF{f`>szE(QAINY)
z5hJ<=PHfu?n#HHoXsBpWv~PIxlW!_ht&;Tl^@6cdYe#V5-W82Q++eb*eT{dyNi2^b
zviW19HFjujrtQ+`bPW}Kg3(hZ;h`&@Nr3PCjhIh6ws#Js;W@#QZ`}_1Gf}0#EZC<U
z!$IaK4vj<gT84Q=sm}CZWxFOL$D0-Y@j0iU<!o<>WjzKc@9``2x8DqFKE-=ygK$~J
zoGG5rT#O1fkuqTnOedXo+D7^tI2y^2@FN@9Z4z}$r;j*IhqJwi{;O4AxR8_H6;V<@
zzD!{>4KzcR;}CgXyFic6x3b^u&27mFAnplJ5_O>pcVMr}2DK$RJ<fxdV_8I#|2itX
zeg>KS%CeQoC$}q5R|U%kb`YeX8R(S`DwU3EDUG4XVAmh*p+(clPCq^kmvba6)0(c6
zcjxrF8vVLbtmhFI+<>y4xNIstVb^3RhG@&!B8*l}4lb-#j5QjG^S3bqGs+rY@Z<Ak
z62tWt*+C!u)cMlk+|TxAePHgM<8?Kfc~)6LRMkflR$b1FyfjIOT9{k4Y(Nfl%!E0F
z&B?w-uBjs&pSybM2LyMa*0rvu(R)^Gvb~C>q-2S?t^PE_+JmedH5!tLHkbJOB4aX@
zIrJE5tENT_mJpPo{C;XamUnLNQC0WIqAtg&9-9%hBPd<?u<I^XWVI(qPmN&ikj$u;
z*x-lHGiXv5tcRAkb0-m2EJTuT)&=jBv47Gz=nkMRr9llloHSqn2}KOYa)4CBUSF*p
zQnnwq$f%LIz2z>3n~7Os*O)7`YLmOFC_b6$VMADZV+}!_@Q*ykn^i=MwmptR(s5vX
z1hegrI#fiu;^L_`+>@<4B&+N5CCCvyNMor{C!hD~f>moq*ZEE~jGCuz*w`w=;(J{T
zWyO%g=mGlt)~=UKo<wl)Cni`3bHj)6{hMy1@ff6RmT#hgwQ|5qu&M1x(Q8V>Re&xH
zHpi!?uU0ao6~aCx{NBMx=kTbgTsE|yyVraB4ywI~&G$KG4)Rf#bpgKQt@}oqc$F7x
zZm}JPr3Yg<h9u`wG(3}IU$nc~&@(&2AO0-UNP%)q-4}{S?%nWkb|bj$8QO^~R`1y9
z_V*($Xk7t|(Z^v#z*@yQsX{<t_pnqld+E5H62K-BCLRk4t`Mbyb?LedF}u52z3#6M
z-d5A220cdTaLQA<YdlFW@Muz~EGG%+sz|f|9E#TqNxH413wfgiIWr#$+|n0@*sun|
zxHO`9vk+U`EAbtgX$B9Z;i>=CIL12>XVc=-Y5audt5i8*CjMJ7{cggdVPgGW#yHzW
z5=8eGeIBK2K05n5p1OytwjgGS)cIs@KKYrWw=gx6I78G#G8%fu67H*M9shwr@W3$d
zxnNKiXCfc{(;S*!h(rjEz`B!Ok0w2H<R4b-j#myYUH<fk$tJeICh2T}ir-pHFSM@u
zyt!&IRA!~upiCOtax}p8I>3J(lAeM8wb|d!D$XVh!a?OG^R0)o7qqfx?l1oW&q>~*
z5xD+6tsYMWrk~3P@(YP}=Sp;JGz;98h{Y46fhIm3w%}DIwhq$D=1)0i7<1lo)?4My
z5US|Zb=7Dn-?<Hxml*$<!7$}tj_M@Z=qb0(jw8e_kB6k_Et+~u=q@T=LTfdLDeDkH
zmdZSgPo}y%BM+NaJuoB#ZHI>Y!Lxx$Ld037F<_i7ZJlJ66_n;ll;;TnpeBmuW4)J{
zmW%FNqtNHMoFTmGSX`iW>-^G(Pnzf_&XVI4VTpl%*3YL#2^aFqQQunEk)u^Byva}Q
zx5)glK>?r>Q^geoRB4K{IC_|vbb*=M*$`=4tricx7S%wp+LDSL&(homkPzFy>x8Ba
z<S!6-yj77NH@D1b|2l00#oEDB_z{DZcVF#~Hv>FVIZ&!kIB*Xl85wq%tKi_vj3_Zg
zLf{KC!hj|soLdG28OI-YyD3xeRbAXVYoTQ)=y}>)LgVPWEuWu9LUD8l^IIu!v%D!D
z`C`UJyA0e6lA>=D?=l@ebOMABk!0Ue_Ja?Do>&rSFNV|i!A6<PwT}0)R%i@&Qe>V;
zT6lM~7}r%O!rV7YF<h{nq7<hug4sj4<uRLJp|$&O%aYAa&IAt3Gqqp^#+IjM$_jVl
zzK;77CQTKY+cSxIC&Lt{S)ku3!kE>0@P!EPfizU$Nt4nL`SMB7F)Q)wV5uy=FjDS;
zc@Wk;^<@+Y6J+8=I?eKpi_-L`04Hu<Maq4l7i(CDH6jf%x?siiYlEn_?-aCK!YaWS
zR#4HnILmsG$2xyNAQh~CY|_QVWd4jEfqujdKUpfF&v0Q=_^Wys6n*J(-;yrcxN+Zi
z>5Fr&c_*cHv1Pm`Q%UL{E&|zshz&9D$b@`7XwhoZ^V?@Bj{3M%8rbosCot})z*9JT
zv6W`L3*w6xhTqYFQFH?ut01@g6E_rl6}uN+q%Z_6<0)M|v?MawHocwVBOjN!&~Uk%
zCXgEOJry(H1|%-48~XaAE9fm4>F$_E)^W6{QCl|{LRk)DB2+I~E2dW|ND?Ry=`wF1
zW~-Z4i+MkgL-}68VEZGuJ~wu?ykSmbS|Xv2#v>4LndlT$Zev&}r;ig_i~g)`M?z;4
zB@}0g<okxxg{2@KCXz0YJ~u8?;ZvTAb<nnkr+1|zRQ=U0DFq`wqSVfTyW)5CmcNH}
z;_$Hg+ZW{-u<a{>@}sd}cga7-_j}O537hCm_KY%5g5_d3ryP4`<6<?$mR_jG%CQnA
z77g6;6+R+fdN8%gOD-4ha3pq_lA(=Q&R>jcDs}Fyq??AqnyD@$WMPC_yx;TfU_1-B
zruguY`4$cWmjz|S=i-S38Mszy8H9LgfDgK;kb(C2vnc$X!O{OrRceaJI4CK0(Z>@O
zs~z>V%82;mm+XM^2DYFL?yj1(|K*#4p}_Z8^&dE<qNC>d>_6SS{1m4P&?h+(E&emO
z$yL-bI}#Q$d<aj1j6GsD1#S0exRu{*8>TA9lG|C76Bc_mP*ftC@A(#Bqq!E<l-1;%
zG!x_!Y%tlV<rQDDV7U?pn8VhxtvN&%+lTZ&ZWjsDbtOh(Q&RAf-KTd>`-Gs+GDNcX
zkX~V`gQIe{o++e!=L*%t6qWkg@Nz|KyWiN|JpEzn4`ook2Wz#+o7w3b4bD_%xNK;{
z;ORbpW6x}}c|Z+Ho^tbSeL&<ZNUpd$(%-8<S2&vz7okboI<M8PDe;i9JV#JJari>j
zq4TNqR*YEr<et!1$C$c_47%dGmNY9=$o2EqzxEgu2TQKZUscGRqSNMRe&OZ#Hki2J
z86AQg<n_DIbQs*{!L<~#R6!Y$wX6&UUe@uSMK+TxxM-p+^IQIc>X!4o#y$Fe54+e=
zyGi)og0xhx*%7%5@Z0%Kdu4{q!t+z#FQIPVAq1BZh0f@e$SQsDF1IU~E<JPw{(PhQ
zcLnrL;7`Rin6vT~3tuCir@Fd}puuI8a5^`q(U=a<L-W(DumM#@;8<(v4FEbbw%2gm
z$IMX)L_Iy3M0Co*EMvJ62cFF!jE|0zlXf1v&JU`gFK5V)^p?4YBxdHhQ<O1rV38*%
zJYI|LI7mWBvYyJ)ng<rPSW=1%fcUJ*7<X@SCq&||S(uV$MUDM3`NZH;<*jq-h4&-U
zBzcSP7wnh8*izRrbb9VjNFu+~>QWNqF2U%f<@f7w!~xiEf9=Z3AWNT6sK(rfI?hd6
zTM-+iX%P;VSbQZ?se3uD<BBMxgCZP0@d#_NJD02IIQroRyCAsJ?YaGQA-ojoBen``
zb<a1$x$<-MMiv;`ff4A>3?k&!mcJU+X8$xv4<z1qhiwrawwtjbSWbyr35dWdYh(Bn
zNt-Z@f?i}lxl(7`J{3()oHd2i&B7Cf0cB=)j0L)inr04Rj|6pgu5y^I>_AZ2+O@gG
z`|uvjSQ(I*&QYB9<~a2Yx`7@WxMVP0`#!o)^Wp&#9n|@8-LqRxq@YhS&>)6{jqmfr
z+DuwxKA2eoKSMsC*v`oP3$fzBF0Fqoh*m*xV80U?i2eeAcS9Mc(NRIH;l6k6Szj+%
zk)0yRV-5=<xa%chkL22td6I1djj4NXdwqwY%0puoU;dS^Ao-6Nue@G(>GciAaW2h8
zzo2e?4FU3M;M4d(5MY{%_3_c9_ym=ZUbSy><(JBd?|@E1l%9fiAEaa>kk&M#)LmKB
z`5v%s{F#1Ci8a2a;31od8qPK$GPJ1e<reUxn{&T9?4BneUpu~}8^Ok7zxA~j#6f&;
zk=To7&80{YYwMPz&}1mjZOMi|Jf^kV;8iK_etBwKH0*@>EwyTGSC?<Al=)`!Lty*Z
z_yB5TX$M_#M~1_a7P0QcG?<mmpXqU0mO-i~%O)TE4gX#}#@_x0NePFlpFJ|f<c1#s
zyJincp%_WDZ8X9?QSj+xPSe}7p6<LlnA0)zBHOFBZn0bm3Np?O_l0akFMaZPNOfGx
z3s%$te0PgUUPIs1+(4(t#Ze<Z1spu!-{58S6EHKrRkShQT;au=tu(|it>fD;hL5g^
zuXl}*Ow~9_vM-7EXXPC=ef?E>TA?6!UwkOByjjkIMvd5%Kd?VrlamqF3oCl6G)}D)
zg!d~mu7QcRpJJ5^2O4O2K|!^QkzHg)bMDfWdsI3}u`WDfAOCQ=Z*X$+z|^%hB-}U(
z&9hxT{GxskCky(VVa{s=@^oIn$hIQI3(a}y{*#q?&*RF%7ne4Uq*{>i-OGy$S!#im
z{$=xkG$LpdF>JN`2A=e2r@>2bom{j1UWg~7NhaGdhYLaT_nUcY9np$*8*Tgb5zUDe
zdaj#kbWo()4u~cWFoCS!y_NLWx6b*cG8G>JZBw;W__A@OfjSHn40xWQ&J9V@+nkkN
zFF9EwrE$ut5q_0n?(SO8tZm*D7>-tC!ca9(7B`W#$E-eD`TPF)9vT*|s0fuVc6KqE
zu46b;afo`E4)U^P^x1qyJgspvB+-}a#WT_GTlQQN4CRPfO+C4)hgc}<T4KV2r;tjC
za^#Z|+@%$ADsFTwCv_Mm<}aU^qABDj=1)nn__#a5Itnm(nZq;E6EwW#$s5MFGv?F=
zQ2a4|LkYp(q}VC)xZJB6`gUIkgL@v=fW8!_^an9gTP0l=nmMi{qeKmxABP;@cJnR<
z-`MxF5>8#B)L{3nyDt~n154eHpzXLOMW%E>u^wlzo7_lypNYa^QBoMtFqUyOsvt3S
zzbzR~s4?!(9ZhXNq2K(JL)_kIIK<KV2HKrTY3U(BJG|>(4sZx4rqP^12(ute(bOcJ
zzmgMx)VZCs6V&Go2jSpx-#BkWs3@7J>NW~{-Dk1T&vl88HbIN=an+%mRkjbDvLQ}&
z*tH%$xeAC#5C{_Pm#N=xlG#az=A)V2#}sKh^XpSwv$Sl_)^LekJtnb=n;p^;C*z{f
z#2=;zV7&SWuI<Ivx&2V8JcrWAsWf*H&#YdPzQ66v?n?j8_a9d9u)~W<;hnen6h~G0
zlJ&gX9ESAO?~z%h)yX_2L>GE`w0$QSm%4B5M_7_Y%z&|M6VoKd3Cb~ar3UE`ho7=X
zgbM539euh}TcnUiR6S@xA?iLFCuYnbhocWn5A2Z?#-{FlW$X?bQ9yDEF$XmuNX^fp
z=*)5JVsCL$r^lBW0|-D=*EX3d{QI5Af(8|*SIF`?JY!Wvd@&jqTM8qt%#-?jlR5dZ
zdBw12r1M>56=o?DN@@e%JlEa_ihXJ;8%f!Sed-{>l6n0$)%K{(EyXU=gS8>qHSgh&
zzYcH51qA)}DiyVYBZEGcj&dohNpYOLN{##)KbGOcSzqd?fZMhIa!pw3pRRcuNm`(@
z6<HMh#uZi3C<UbwwyP@?+|O^oNF}JC+bYuG?G131%I{kkFK0}pA1z4&I~(vqQsdFr
zd|ZFPK_dQ8>AXj#-^onG^^mnrAUasR&6rnZrp7d&LHB%4vNLx))zG2WZXvXbU1!?R
zQ)YNL(e)imkK^YJfV2%7>5LE{$i2tr8RJ0XlpKkz3#iS=NY!Y=i-XoFBjVqo9{DOK
zJZXNQ$CFt4k~OTi>kArl<(~Yk1#PKV5Cg>bY_wiNS1uhWq9B(b_2b0jiYfXE{+pda
zF1zFHl)x!-CQ<l=#U@V(3pP1*kl%E!-^gbvW$H~wwD%4A82B(V7qxJQ95*APn`<qx
zO{Kf;M4%WtbUP~7#t{wlD!&Nxiu|V5>1RFxIKxcnzo1@7r_p@mB*JNuL7SLr9d_rA
z`Pau{L#rhhoCa~@G#w@BWWr>GKA-q@+vs!)Emd7f8~&X|uTS`ydA&6cxv7SPKG4-_
z0lBV?&GloQ+zXF8iZHcq(TAj?C|EZ*$eHJX)$4pGe7gB4nKTC8A<iKUm^>HZU&*uo
zDDSC8$jCe@-M_(SydtmJZ{#UX-q#fFowanVNESw!P1kLi<t6{3WZyd_`LR<o^k(8c
zD}Ys71zX2L1|Ezeyme!Rkc3Z>_!)`yssrrV*C$Tc{k_6}__^9G&sWln`&WxARQDDp
zrI_EO@tqMD$Eov+pLpAHTmN(mu2@W=YV9)-o<n$|fKEh@=hM%SzeK6(<^AQj)F8d<
z?^ct}2B0iD9t5!&K*o37&y$jwrEFNzz~hcimUoEo2Yf{xzT5UBWw@w}!Z;pP&hqm`
zgL*#`B45v9gt}z8JZ$M~ATFi6E3g)L^)JZqs6`)uALQcC%S+uOEPGXUvfaWT#?O}d
z5Yhl<!{{ZF#xQxJ4qscySix{v15^f@)>wL%QBKi3{IJerwTo;M!MFm;o}xP_QD!?;
zjrA6SaWoPZup`R{Fu*90%$RcQ%ZE`{SY~xuB+F$G#=j~x=~Z1?SY)Y<V$r5QOg}oy
zZ4*jd7*xkV$YYS+GiD=Oxr59dg1`6Gy~wyuiahPsuV&*YAd6yr>s8XT@2+_`_W(|F
z;)$Fbag|rJ`v|0C(c1c&hP%1k#?v0~#bvk$P?6{0O0kf}^6}Z|r2=L68Rvd>j9p>L
z!kQu2=S46EG#!(D_X=9rMndYOx9F1UFl}5&P%Zp1SYYZaL#NJ5u1M{{b=@3Jf{@)#
zY-D9J25+mv-hfmOw5(#X-tJGq`L^hMrCCY0?ihD8R+YePoURNFyGnjmq_!Xe>CO{8
zw=U@JEmA7G!pa-%Nr`N~27{K<qf_uy*)aqH{S1@_Fj4-#IxM-Sh54DIb#L0-BcQOr
zgi{FiO<~M=-Bd<Gj!2UeansUP1)C(pQrD`S%foWE8d4u?QBN$&*m=DV%v$>K1Odk5
zKs#K$YSR@D9K~{z%<?(OqbM#_GOceQ5ac@U2{g2mK`HIMt_FkJu&IaLuzn9(#bGQl
zApzYg`G1spZgOoYR|R?LjA@6aqS9c8SFe)$1|ok9X~fuLRa?R)AyMA>HePwXaBG%d
ze(<aA7UogOc=1Z(yOt>HK7uiwaK>-H&xZx8V8-H(J_5h9@D8@<s-B!7Yn_s#2JQp!
zb$KJiMXzj&$VyX!T;3Y<Lik(S9S*&6XR;YytB9)Qf>aLRPMn}|hw<>j{5A`UJS+?r
zf5oa$;7Caa8Q=XY(BAp`9N71E;vBK<u9AcK8?DZnBAwFWNSphKyU-@p7aTmLK5eW@
z7L)9$g@75nPA>0R=6oT80<fwIZ^~x~&OS(f@@4ux9a46urD&j%B+uo3a@I&)BzLss
zZLNBD{O9WLP*ZVM?n&VR`S5^sl(KO=ugb$s{6~4yIsGwR_g>hd^F>aGw%KXnbbhTy
z#ZZ$K0!=2?23}!KUP>wE66{j8pTu7eEFzQ;QwXUnBM?nmxz{Va8NGEn>uQ5pf@tfK
zueiq^`*-`V-f*#KZvm*b*+(R>1M^5*q0YSSEn<{r%suP<BvA9aZskCuj|TxM*aPLi
z;oFUhZD6>DTOX0gqOE-sc)$7rwd6qFvP8;GrW8nGtJi-qz>pGL+YoEk_f%1*=H&-L
z-Xbq-YOLnRclImz*4x)YGYR<pU3EV6^x%`|-B$89Dq}$46{*w{{J95EuK?bC?f}Oo
z=oPM2#3D@20z(4sMd=1obazj`0^}^-<$MnL&s${YtAVim0)kn}O{@eiCqm?CCpbZ!
zQ`!c3M5;7m@ydM6Ww{iA*3ppGG8+i-y@cggIClcc#TKA-57c<7aXnHhEfr5})p%F!
z@sKy!V`kW#Bv01!FG*EHzcE>AC(^1o3kQVBUA;X6S$TL=Ueq}m?8m0vVw%yP?WTdA
zTouvO^Qb=hBXn%+Td|g~Lto?*>*@Vzds84q(N;dIqiZ_bf?YbDxfpYo&^4b<T~3|3
zs5r7a)q34xvc{8A-nzqNA#}smTl>mSC!EBsE@Rw3#c<hFN_2&t9fupw%xGGHa%ehu
z1ptfZ@s3h{%!+_u>AWq8X^+vNk?0`%<MaY<ixJVfWLI>|EH|S_R2%I4gigX<A(RL~
zIsEopFOFn#D$fK`3|?g(E|A_H9c)JYB7-ik=PZx8kta7%cqOYK=5<T+$E=j)%eiiZ
z$GgrG**{Q=gAm53qz!S-uQ;ajE>#2D?603@;O=oedrdr4GOV1}&=qgZmX4)Ozu2;2
zj~{!?%))+ES9U<@$HmS2B{MV-1cK2S3U?dr&SF{~=lM@eZY1;;Sotg0ws?H)k_*5f
zd!n0AG1!6%4j;PtPdOry_+A$k|J_xNgzCi>l>vFURPhn4Y_mg}28Qh+=UPr=(eW3M
zm478~Oa4ULa^g|=h~Y1)8t0W<8wede<}X|;Pd&|<^H2jkKWdY4e<Gb}j85uB%mP;7
z?CJe*92rIO0ITigLbL{c!g2PzXnk{Ao+%~ROJu*ysql$KqB-1{VrO}E7^a2?R*hMO
z7wt`6e*-Vvc)iJ&3e|pj?K)`rD*X=tL_oX0wL|epX*##LQ3x&shxoN1{4GrvV7i~p
zU`P)mU*4-27|43|V(tKQgvjj$_M}AC119qK=IdQG`A>=RWFIh;^zh#m_rtrKbmW3B
zS)DT1DufR1*JzxfJUG;6;uPY+TkFr^8b-L7Iv?i0vY0qfUGWqjMkYk~X4b*R6l)U}
z5NT?g5G#AgH8F`V2JQHZX+Kwg@~%so1A$T?Cw}q98yx>VJ=MJFmVw3Il4h3(z)Fb%
zL_L%o-1^*dp6N~EE8qZW(p;yKZ_7+@w?_H&OQb#RgLzy9xO#3DGR)*ZHA8tJ-?*Q@
zN-q$0Gd>3prjm{QxL;a@0H+7XY{RRHPo4q>lAJuDOfJ-Oaz(TeJ1KYR8zfsF?TPZl
zOFOs5NTvAnb)rkt!*CWOr{t(H?c8ar4t_#mkf0Ww(r)qN1~79lpQ-T=5GOT8SAC-g
zhNV8@w1%|Gv5s0V{6Df=(qB%T@}~vG>orrB1B`jEIg%Ur*06DKPI&R!$|t?6Tgr)k
z?2Yw<XpLKX;2iK@A+T%tPzj6mCGH-Ra)G^qyurhtK{hknZt}=B7r$vD)93B<TLCy3
zmfog&4S)4|xjuU;BoPyUeha`9Q|sy61_c0Zj3c2sk!l@Zv=IZj-Uz;pX<b1~9DmZ^
zLba4A)D7kT)O!5X7qS6vXqaSl?QFn)32(>r2K%G1FLirsW;_;=6G|cp(B9#u{p!AN
z&9u;sXl~Yu`i%3e$*36(R!qv8%XU-eq*!$?Gz^xjM=9yc`|H%F?EwopU?N3j>Oc_J
zQ|3m1P8bkD_%usD+zE6e%~i6>+?83DC3~iJ=p!Y&{$0CKMB17cOGmgwifUZ6yx<!&
z@Ziyj8*$<fGrDA<O5-+?t4UgFtir9;rtEb-j9hbshIrh|7=e=ZfPLSFdWv;Gl&pCg
zYEUx)-MwXQR4HHC;6EQ{?=Jb3$q9A`eVciFQ1SEaAk|N&pj8+>x!-19wL#IEb8l8`
zXkc63w?K0C1&4;jl(S0?^7YPmeCu6O&^%Uf6{mubs=*w}UMJ%TFUnF*P4KK4$WK73
zCJBzm4iUvSnR_&HX!-PXm&i0^u;Z!s_@PR0sT#@`!Tl}KQVb&2MGC`=xAEBDrtPha
zp4HS)-Um2?YAId23bh{@<p;tGY1S7PwM!n!6{9%)_7vn8Ogh<^n_Yu=99Qq*k~k<u
zAV0++yRZ0DDZ$c)fJ{=;jHp#*tD3-tZ-KBXR<9qf3__i@LSMlAO75fx>MX`|@D|y^
z;-unfev1p2F+o$!I&(<bPaGV?#t4?^Nw18~>cDBwcORkgnb*keV2kk_H=9%<m>xGk
zZ#i?EVEq2(BDt%>KzuZwJ~4&9VW*j`3$vYz^+nlQHse)xaXEChNtBOHgx~YhXKpzV
zF1@mONzhvBCGQPGbQ_=f`^Xwu&q-buXyoSE0A+%mCUN!UTIsX(#_Oz|?0PFNer6(?
zYX*ELBrH^!lSKaFy&1>(OHzuCI0Uwb_BYR|kB8=pKQ3?bD5{~w1%vb6(xPVqx@oC<
zinMaBA3cIn4MG$;8DUe+bx*;e`^6KmP6Zebs%hR|nCO_3&Gi=aPsW`|mWfft@-f%q
z@7k;14IKKb+IxR{9uWk&j)G0%SC{gGC1Vo*MEkYRSXZ0E>>sG~sS#C73ej0Ex@-{$
z&Y`~2Ur1MurqjiO!p;R=z7OH*NsvezhIuk8D4jSF<DR5FBf9SL9Sv=HHMlX&#R)5!
z|MsrA&54@=KV`L6Tpm648#_Kw<8ArS{?QFATRaF!FZlD7Ic37)z=2jq%WP43-PVAr
zW224yXR9R~r{E)lwq?HUTx9F4!GH*EIx@c&;>K6VbcB|CA>myv?gWEOc4g&`O3-y7
zg7Wn<)~`>AL3sXE&;;*z-f3dibo1AJ&2IA2d&GM3+{V>EX9>HQN+<VzNVe<`e<z4e
zT^H0jbju9jQM%(yL95zMA<GR&41{j*H(lLd6g}8qw9aJb?PU4#)VhLFT*Wdg*CymS
z-Wp`{txRSgy;L^Ty@^F_^yz(MYD?^?vaeY7(~23scm+3&=#{$D&1@Kjl>)a8h3iix
z4FQ+MP`(6CFIuazpc7t0#EY)`eJvnj`A4il4uI+hw>JKU9z^INK0V9OzO^GK1pn})
zj@@ARo5wJV;I}MJb)G6>hQURNKs7#ib>qwd%7jIlx>y$M-)tF+vgO<1*+>z8x(t6O
z6hS{n)g!7~>fS#bm5B*`yAz<lY7ceou8bk2EX=(csez83?PBBhYDr&2UqYwSJf(g~
zbOe982`F-f%Tl73^}mE}O`5UZSpKrKW7Suj*VL{EUvWPPNW>6Pr}80(e7cogi6-9;
z;Te7vul*H2sHmF}9Sn+Eo!UQj+HCj8JR1@^Q}fIElWwLpvyiXkN9wE8PgGJ_z+<?1
zaBO9D$0Lv5O=TD&8kDX8OM5RKIhgw5my=5eKiRJ(e-t%f_OD$LS#J+j704=Fx)(Vu
zKo7%%(Grgfg@;Jpj@bOS)-lYK|BST<#(Hk-C&s2Nvd%?Z2qO(a&gp6CDJ~?}(vyZ_
z)ECmswam>zi>C@2S)vRJH+Va8TaMZ6R!=Afj@0~{^g(p+k1%&X(E$>J1f-xh*-EI(
zb^Wf!bhYxt^38ojMV&KB6yZZyZ+{aORgnHD7^bZCeYts2Zi%cBro#T-NT)|U()6n+
z<pEbKJ}AWlsVwIuUm(Z-+~#XmZICp@e30^{7TTMcN+N=qLDqmKa$Y~kdl&?CQh#8W
z#TP3(p@*AnyoF1Syws;v{Wzz>1cXCr&PeBtdJ2gHs@ipJVA;PCL+`fNZezvW4vQty
zggfn^xkUNnra9Vl@VqA)GPyzkV>jdDhRWV%hCkuQ7^E`=2!9r+3P-en;Bk70XA69-
zJ>?#bEhOicLzd1xb_=HmYTI6Yl1U^uXf>;~5LbaKvb8<oOu(>p=vcCw$V7-MdvgXz
zjV`wH5rs5N^M?XW(C3J33ukIc+;AHnJeIG!d{LYh{_uHN!>fp&d8~NitbYutZb{U$
zNfwMBuqgg{;e@e4LtoSP3FBBh{QJBjgT)6{O6>#;r(-c^&+Za@QupBCxwrt)Qfd;h
z6{?DV>9IJw*!m@xoX3t{-D1&<eZ_jGy5E{0Gp9t=>4=MhRD~=O++9o+^|a1Mn8Kth
zm5Rs9#SP4GNw;!_)zk3D-KX+dvU?jOnJl##W`v}>QHAaLJn_!6cO>}GkOyJe69RCy
z3iaB)cVcIBk(X%QeJgN)H#L+~&59BY`q%;%#{bIY3o4+<$Ud+0_Pzl^bbBRQ`a2cg
zMCHa`(&AtLl_pD@IVYLS{QW*KAyv83Yov=M^reh6X_RP5VUUC38o^AF6sQmyo3r(x
zId6Ow?B?Z!4%`WOSgMcf)K0QrXXD6t^Gw`JuMx#B&KMr+<j^dSr;ahH-7^_NgwR;O
zQ_0DaXFDoJ;5aICkFT%G9Zpl3y+%H~n{{YD0Q1mw-hS$D;|GXzb$@57Dfd0>UK+bj
z8xmA?MJ!V~@v|}&yZBt&+&Jvm3V-Dzqp2G6owF`LPvUl}*_ENA*LJ5PyQZiM_E~}I
zqFaDA8Qh$+kiU)bM|abc_3B{>mYKWh>+fz)UM?1#y?Y1Tv|Mz@J$=NPj}5qg!^3Jk
zgm`^4eS3X21iuoekl2|X@PJN{%xRidZPbA(7$7F+q7w4($m}y`4`B~GF2nd@EfJN_
z-WZ;ZAEL~{#Kf>Bj=d>nWfeJ!jMbs29a<hvfz?a!T6HD*i=nTdhJ;Nt^ZkjMe*bnu
zTsRUSs8ZmAg<P3w_(T#vC0TO;3J|5Lur`0*Js|-#YP64bXh$DK=F2r^&T{8qT`dlJ
z2jS}RzQXK+E$r?$?H4yd4H1oLD@h~NeKA#@dJpAjNrR~{4~Knn&cR@dg=<$`8FHfs
zE-w8xG#&aiwP>bAx!r|!K>CwD7jAUA2=6^<?XD#OjfW8iu?^h|9LE|THjAbkN%00D
zkxrXs;_kZpY<BR}q@vrMok*^H2fZ--OJghLq^=4;4QG)j>C6719^I<tmZfAcz9TeY
ztwMVfwW%d1+DJc4Fhx(^p+c#nsjIIBU)zGVjEn~AA|?l%qO;sV?(<$+>H`7vfSa}F
zHy4Q;-@aS$vXi66SEukY+oU#f$mXkkpzs~8KRD%Z_-bP!W2LRBgMi1-8Z4ycyKARJ
z8c|=Ej}9NGD;?~|e42_Uv+}!;0EAyvD1RrXNr1vPn|Piw3BsRhQy8icTcB_+olV@X
zZp^RA$1~z$N{7n@Db-6g)rolTYV$ERTCY7lerXUwll`1)NfM$ORemUsrvsR5fZ5TQ
ze3pQWv)(!bz>^GG2^Y!7s0A!>1<@jM{rl0%CF=t8Xq4{nz_5!sR27rWT1&(n&4I?C
z?j8PzJ(_pzJw{2v&+{w=s!{814WSL$m0jM4x_%leDnN@+s)^cWF&i)=I`g8H)TjSe
zG%pQKiwbgScuts%q$x6x?l?C6eEVrp_k=C<o5jzUz&hh%lA|dF#q*7x&b+o5`lBk-
z9u5-S)zZQLH?DpI@5!IbCmI80NLTo$h!2{_L=SriSr_VDZ39xu@%Dn5*R)1$nj3?*
z(q*DziS&@^gx@E-DlO3icOv=Le*@y|aEpzu=z%dd!0{7VVoSkM(zGzQ5xHA!Jmv%z
z(-U^r`Y!4DIZimp8`0B=`JI^_LzT#uol~2qk4$zQ%FSVv#<(#R4#88QJnpC4VDOyr
zx-%W{V<T{{#;0JBvhP`ud%Lo$GAF|tzEz$dTBjFyAbcMTq-J?z36uF;Ot9H!KvUT?
zfdDOc`9gTB%CBhAe-*7i9piTdzB@!hwCZ?am3@i@5rY)Wd+~N1{2TrT#0#Y;Z~tn7
z2c-w8T0muP8ZU&|J-JGVG2zQ=F=%H3_mO)wqTV!KpsCb#D&<xPp02SUJz<I3Uhmk_
zi9tFzEaeII38@8aK!veddfaKu!h;`Z6b4C!{mya5H=O^|5y+=e4?;@mZ{8$_vD(A@
zanu!mMo+jPlR$m?x1>XX(!8NLM+WSpweNi}n!nB`jHU~CX6IJVUkP_Lro6z%W4f-;
zt}Nt1UeX>lkMm%`{prYO>^XcXoky!93UB$koY2nJ+k7ahEY@_P9}z@HK#$B2jWknS
z4JR1hzzVp1<oo!NH=J&D@~iAEV?#!dmwxSSSl7a9DhJJA;S*@ss(!w8dOxR-UX|8q
z4sx6&4t}EGSpM*44Q4?U{EvCA9EcCzO~CuSGd-wTBPo9KMIpaG&!ZF!)YFosd_URl
zuD85Q3?WVU7<^1pYX^j~U;ee0L^tW2B?vLbH5Alp$3q^HvHon`kh8^v;xP%3^>Iwj
zZ98cpD2|wfh&4BHLu<HuHRH1IOF1xAAcQ7LjQ9x+(&vQ1Zf&4WS=CMJ0%j&h8?__M
z%PYD47P_EVq?x&DUU{65zX-U*`I_}b0kN>;B{AP}KOFq63kJmD>49%UvJS1iy=p~-
zu?~1z9UY`j`!CQvxnJU1R~LmkIQImCWPEd&sX~$K%dJZwzl>~XaayYpqGXvPk+rG1
zL~2BrbLP+TRV*28nTsEe9bp(U^P|sxEQJ!BzQ$M!`)8|v<jhpwp>_#Nt~=;T%wZR9
zo{1S|=Psqk*N$a58|X5qB8@pJtDQ(RGzEV#mZ$3f&vOHlZG|Q8{=j}P+iZ&He5u{b
z)AkeNZ(JNA?w4`??Gp@QU$09}uM#)M1m;Wg&_{%3{%N%)B@x=5Y;qB+4MPuSXAm|t
z%)l#``U`%LgSg?r7k?mBQViJQ*FO@Na$A&VFbb99EGGGq<W={a=AxGBF1WmO>{hB9
zTw~bZswg4E<rO$NP?|o6lJZZZ63*Z;{B>hi3_1t`I`+_f7>LOBrQ|h$>bRinM=P9@
z`An|iW^di6B&<7?StV~Dym@2>NtVi53hj%Pnxm9naj}xYk>+a(x1_aiZxZeNNnymv
zvy{zpm8rMFI5Q8A0wq3a6>>7d_EUoh+9N(=qzi9<2IDGyCo4WnpENB22lg^TaQ$dF
zO~vyT_oS$*U?|0r+;~3JVYmIVXv8}`4~wjgbWq%zIDcSIrmd%e?9UeYkQfjjG-?Ep
z&22rK%sEbe;X}H-3(9&$CXx-jEYd;9ERKdS;7T<C`Lmm#cac7*&8r3fm<Hlu=_#7c
z!;&7%`($blg<UKVLpyN_iHoQd9+Me2JWe>1gq+-8M<yU?h4>MN<t?P?VulDZ^$+U5
zSCs4FS9WIrsN7-JHjcZq<8^lIqBOjN!rSZ+qjkLmD=4~Zb`49oz%)DWLVDhN<ymy%
z?*3z-0i!_L56O=eBKQS()O&I)aV^^-*@9&C4K=mHSA?R>J=qllu!a}1a*NuCp246E
zlT=MGXK#&4q{#hyKCbi(HaJpcUz&n*?A>q%rpOhDsb`&G=V3p#0RfU0jBp*HuMRLk
z#AIy5S)co=z^BgqMaRE<biT<&01BdX^o&=LPr@gG2hZWXXF>zoV7fR+mZqIC;#wx%
z?AWOCAzCYZ9>0USFLEKU$u>a5#y_lDnD3YQ1^$ZkIk{bd`3Qwg5dKVZbCDIZkyPm^
z@-~vDBBp!fhY|t1av^bR)-jwq;FCIX$EF&55O^U|WIzI!RYa@(UAuvWEuci{AQreS
z$JTM)%w9`XPYZmtON&yvvab`uV+?!A2SUMUhm;)?R4BZ7a3E1TAb0_7`i#@!FY->n
zF|fJ3aZ_G}80hlBj^sPY9^Mynibg_yq!S|FK<r?N$5VPfZlIVl;dri>u|xFDtq?|y
zRz^o@Po0#TZ<qFO>|z=;316D{^_r?$Q2H^=yK=F5MUIkvb=?f{?W@8u0rcg*X)AUR
zQZBLn+15fWdJ#KVj?wYE18C6m-PJ$`BB^3(h@4i=M}-<@BQ9=cGoiF(&26ATy9NiW
z6BhSvD7K;I;nC`Vvxpc2%1Z)|ITS-tfk@CTY0n-Mls}C?=Es^Sg@^82b5zp0Tb!w1
z>2EUL<KG$2>ZO}AJl0)TzY=nVc5_t$C62m?-cwuK=cF~AE{8)bu<r;!xE<EXhKVao
zc853+NV|4|t@d+EsmfMwuLY*ap@k$qga8-H2Q0q*4rHg<1Iyy0pV3Ebr9)$u=QGN#
z6gWFuvy`b)0sog55U%0u`94s0-`E4qAK3{o;-flY5KkUI7s68Oql@CAZQ$0yBB5oR
zFm7q6$P#YA%JM`uD~?;(8LT!11W8w9Jr6!MhED4mmyjS(O*NniPa!+@sX(q8nVzRc
zfFDg8w27__L&b|x@GN}bd&!6hbU|BGI_6X$faAN-dSiRonzCskB`xH-4l6{}15H%>
zQo5b8b#CM;kW(Osl6sWPc5Cde0u_je3S^h<>`%h!E!sM)_C_=*7ZKGXpyqN4q-P;C
zgk(t<MLYWaSdx*Lk^32YCf1hp;)c^5piJHZUDyVpHNsz>8^XIWtIQGMjC{Q%V0DT%
zKYW5sW*jqrmo!52oc4B*Rhp<ZBX?L~8U2CN*5FRpW2SP3sK9{XKN<9c;L-E&Dn@7j
z#g@{EDQ+YG2ImWsRG3ZKX69l9WvHLDE;*UcM3xZp43y|yyqX{h-G?PJtp|f>=|ZaO
z=jgRKhW6Jx0T9}A_9{b1E@*NdjV2w3jNH+lCh0FpM+61}{d~3ROv|ZIEL~i?A%+QN
zP#2H=_fu85){={g5lbdkv4x)`(RVQg1t@gERIG++Muu1-<5K}@zJkqo^EQGD{mTa%
zya9Cd?;{sQCd8oZKG2(m8)NsyPf4EPS%u`5k=ul5wz3lXQR`UjPdu3CF>4_D98d&r
zZqzKbhByiUU8TX!a8rJh*ayMpi)~zSz_=m=Q6fVq?Me1)co6zI?MT-NHA)FLEA09M
zyNmjVk!=8sJ{E<N>%anD85F9xaVybRjlgvN_JIZe?ni`)$<$mBE~ISsX@CCJm_;nn
z;~E*|n)%PvxnfDuc4v8}zDR{j4&*DO#6@wmdv=u~=~^3Ai23r2sy+=&uG5!U7)vO#
zLam`Ox5HGN*m0@Li?^~rV15sjb1B;n>Au6V6p&w>m=^fL?m{@+RCi5sj3Phm<Y|tD
z3cGp8pDWs&KWyP3>h5XM+7G5jn6+9A)Wfx8fpQKZPc+`$dfU9s9XeBjH8bC0MJRv9
z&ax~@e@S4kHOr@7s~i`VA@&QZCxAdg8)c#Lc{^R3zfqX0>-&Fk^-f#DFffyD+qP}H
z`)%8{ZQHhO+qP}nwr%d&S~K5&0Xa=lD^*XbJkgJX@VIkaGXO9d;Ky>riYxh(_oG{Z
z>e*Dz_cqW!CVI_ib;G?-*=^!G30Tn386Ka5+&m3^P9{~fPqbz3<o1Rbd_xRE=deq!
zyqXn6)^BJtq|~%F9#>1nZKHZ32)okgLQ6HLea%_@jx_$o(UN)U`9Q&}v(OryUC7{^
zMcRvgb+|5z&(g*;%-J6AqgeHhORKzltk>!qf;YCIsho4yFsn`k*G`4ibB5y9am#27
zB;z4!dquds{`M|MoxV>bR^cEo9_woDf^`wM2fTU(TkS|db}^-}_ePDP|NcbYFbUw3
zwW0A)x37x1WNeJ>vgV__XEMJoUF|!o?uRJUlM<R3jBq-bDZLCQ?B!ETJWQFfE)8z>
zIcO~7t;xjo3%z2Z;)W-ryX<i36<5fra_fwNy?``y-?jKZb7&922M!YKw_1_`zbHX(
zBmYY`pU7c21cb%f+(Tp?sNz@K>JQ!CCDKX*x-FS}b}s{CR+4fIegJ2I8&S=oZJ_lZ
z7q>68W~Swdxd`Qy;c`?D$%c9S_ywmgmT?w2NXKGE`zx(b8d(h;^VljtX&9nOW6D#x
z|Lq_O)z2)*-WUM|v#-011BSqO=nMV~2xb#)cdJ7iCZSbbrMuJjke5`4M4&E1SJnQZ
z^3h<Ewg{1sRbNmN8IRTrfo7A2wiH@Si%Pvjo^6let(>fWhpf#jtLewe1~FEcFi^r0
z0!<Ic(*^i&r2@%ogh~j*h5A=@bA(QxUx?(UaE_YW-))D&nJkdkYQ<T&QhCR&nf^uD
zF@ppj78cN#or*q93U|9aAR$v9g1T&ehZ?)%W?ZU5b5c(gzU(dxj#Cgypp1w4`%dg}
z9v7IHlJyWrzFk91cvv&daeBzhMn#P<EUCY~q50p1HQdnuW|-uA%epfcpv(;P933pd
zr8&e+)^uzAfgI8KgSk(<Mp04YPNKL9PyMCNy?7$MOvcIl;ah$8=mnUq4lLhRMoylM
z0z_qPdfa!i<^ZQGg$a8Q^?VboF)(KwJ#k=|o}cRu=HK8lR@>9(*U%CEbgpkuDIZ?4
z5m*ugw=f60IndZCuTL9Z<WPr}U?YY2Lg4tOb<ynFkL-a^uCBjQ3vg<Z-(`22dxfmF
zhhjER9(kMrbvT?ggx=R@CGLmPjzo)^OcFEM#?0IK^*p%iT}s*|H-g@EOds(|9Y|tQ
zHr7k8brMUlb*Q8eP;cB;Ujm9TVgkEFiqXDx!O)Y<V_iDedV1o`?`DYC9_n3FrC{+_
zLtEMjDi{=Ui=C5_2U|>dw24jj$Rcugdv3(nnj$TpO*ic*5?wJKT|QD1%S6Nb^hb9^
zUeCd}7kP%zT&)L(6V*F_#`;RYQGUl*qXvBP8^nFcJgu10n>tb~4NEjQ%xfgDZ770X
zZKNuXwJrN^+Z(G^7fd|bx!ih=QFdXWjm`nJj#gq>TXwf$Yzv%Iy7EJU{oDxF)+)60
zkV#{1;j@O6tx_4y!;i7ipp!#EA@aLqZY`i2d=}|U3}B%4l*nT>%Cu5ZSL@=uDWoKH
zUMFA6+abL`tkvDWB3ZgbqWa8|STVA<A_R;Bmx<SJ?+TjDooc^CVps{A)JN^GOZ2rh
z>G|13gtorF|GG>B2MzpGsfsEg=_?5e5tq%S>{){;yW$3(X2vb>TB`{Eq^~hoD=<)j
zX*HZTjY2Xf?7{kP8BGzK)b&L{jrYa~)$|Tv<3!M9=JLvx4?4Ml;V|BI9;BI<EnhjV
zH|Ivr&;OkQS-3KM@C-EQzyApVo46Bh#;mc!A`ksTRmrJO{FG_yX%oWJk}b$ys~K;u
z^Hp(6#rxunne12$Th?)4KBH3K|Mlqi_FX6r<qt3313zUHOUKWT<PWD>#HwpG2U1x~
z!yHJQHrWJl{a^COq~(6@KKhpxbqkw7;!C!bI^d#oN2rq}lwCb&2s!}H($nd;Vxd+Y
z9PqldHduWX5c%3P6~!t8q`V<2g1|^St+v7u<U5v_c-)DI%FwQpzTU~%zU$H&wFniw
zp8~Oi7o^V?>Gp0O@i1HGI6XS#*nkkwn^(h3k?8XB?b!x7N7rYWyQa);-rA`*HY#ZH
zt5s*Ew-JVjes}wfZhssVb@9uDv!sJ5@EXH^lZ6Hac1e>GJj1uOBN1==eP78A4$9S#
z<Wfh(iWyyDWr2DjNlZ7!atIK&d?m#=x>s$_Ac7shSjn*w(*)VQ7reBCPRE8G&*({&
zbmVpaP2X7Q;%+tu2zXxe{oINU)POd$&r<qRZsmquND1O1q18{);3M?XF7yi!HFD)H
zF^R&@oCRniyaCgu7f$!LC=sYRV;R0t;-2mr2VVGRdC$50K}GPP<Y6K6vTn7WV+)S@
z_XbB}+{5WPXsCViUW9`gY<Aw2X)AJ^M{i(4`ky>cgDulmdQCzD9#+{EpYc%~jRkvs
z9>p&b6|nu*J>i)CxqrfW67yG#0UlkcYvrLeHVWSd4Vj=jZxknNb|IjH<>D!jV2_w6
ze^L(_>My^Zba<2TP3f$`+9X#V+l5gW+j(WuldxV8C|H@q0%XM0SL}q3gHDJTza+@C
z_b5-GR+@)^b=AE4*<0L3OW4?i`SRud)y~yYMSeCP4)nT{T0ace8EsNt>{G!GaW#zk
z!I<v@<QzUH;nPUb{$IfUgi@%xJb_igEG=N^*T738&4o@j@Kch)KR+@)4`st%Q;0g0
z{FK*?{XiN$E_UoEeTE;~?__SlziIT%K+e|Gh#kikeiwx8H=f@KS%ABcem=B}4gLPF
zOdSYPk3^M0^37N#K*n%iy;M6Z4k~?~t{5vS`{JH$%sW~)2M?70wf^Y))RWoEfzz7>
zh0O-%d`}_4u4^wPcDdj$`;AKQQ-pwB2`=;fkTewBro*jR3zNU+9o7Fbp=-|ZcX*#3
zmktyGdH`IBVX{EW#{5yd1FNa!?pR*NUS}M_Y~AVq2)4x*h)yrxN88qAs1;)}sIIhW
zs9*GNhC=upP;*6EjqQ6h+J~v?UO1HRJN!&Vd_G(9ZR+M^3!N^4CDc3ay6Y}WUeSaO
zZ)~F?uqIzCaF~VfC;k!Pb7u9{Y!F)Y_!WYxR&|n9U9Q53M`*P`v?`M>1CkM(AaS?8
zHZbpXYVw*;k&ZZA)6bMnc_Ul3wIZQd={coU%R}!py+*mHP6E#UkimCIM)DxX>M`Tl
zD9Tf>VY7vwbxpP}E}Q`Y^jwPB0Ud^H+X33<m=xMxQ9;}TfSde|%DFP|3k7AnQRl+r
zxYfSh^^HafY4(lJBk52#3AOt;%VQh~Xrcut#S5`&=rpp}x`cNKOGGg$p!TOv)<Y#4
zliVgxX7h~iT4$8-qO3MfIJBJMtF7BZ<Tz-`+bZ_LKv8jsmM3?bkR_Kp#Z+)A<fQpB
zO>e{H4JlkdN*EUxH`1QsT~%rmzvL@heULs&;bJNtml$4Pi))j{zslKLaT^P&ED9%9
z!za_Zs|{b*2?nva`H1CjU=^^~m=7RVCQi=5o2SMywkSnS!=H55c9PJ;snlpInob=#
zMp+t$h=n8FRZB4_hPhsCn%3u+{e&%UD8h&x_Xe{FGrht&5y1%fU?sQRI;Jh{nVd_a
z!YGw{^zJ^16}7Y+UUW|qTH{hqFMhE&O}-`_Gv6I!(l3BD7Gu|NP~BuGatHJWpK^L@
zE0G=m0ZZvt`~b<Rozu?0vw}{wZCjJ;w{<w9h5&pq?EaC@j>~|-X0Wb!>{L=d%{AIJ
zG^s1sg0v@|N|RkOaL%T?qG6_<<p|x;)&G1Z#uz27D}Tv7FMWo?j*aoKW7xb_v{I8l
z1ATx0g+2hhJXjv05q=m%krIMg!!bUd0ts|!YcuOHLGi=`TMlZv@0aGBN(ZWqJ#)Se
zxpdtfMfdUCtI+!y3Bx*Y8@rn#;%ia~9I@A|-{`okj5Yxs>C~B=cUyYWYM{fpz1#cG
z_VL;S{g6ObS?3K36QpL%^bYT$mH!5_NWWQ~PAhDuoCq4uCQJ$y9_s++E=9pB#o)v>
zAcQ)tn>8iDPxQJsEaF<(IG$wB*8G3c+-Hc)@JXbIO=;DdzBkmpkdE#ZBWTOO`tqys
zUO1XZ>5-~Xgag?be)l(Q`y2!dqKX~JU_XCl`mYxI2V9ZvIe6{Pb-fiKi-{qqxezu(
zdtjM+uzdWTLUu(T#GmH6^J(4`H*fJH8I#n#s!lHoQsC_G@W9f9V+zy76sLjKWF<W%
zER~l;VQ{+Z<JnTkO5+4tU$mfls$0LKd#!n3W5(sn(^7r$93O~*CLXupP!u$?p@RTU
zH5?MW`~02a$CYkM*O%+G`L0Yh!3Jb9FVijk*`(i&2E!2a-{%whP_EPN<+CH>6ySo3
z+_*I4>bjE#$BHMtx=l*+h;?9GWB_A>)dYE1G)(8edhTAr2xtusHQF_5f=9viLN-4S
zL$kWG@cxPk&Du)B>zS_x0?#x^UIlmcrunE0$C%lQ5`qB^QfVeb<@_Xxy&U^3kFarw
zK8VNNBQxA&fE`D+Zcw^vmd8p2;$l)LGC8NruOk~X$CER!5ckaL+=CVVW_}{Ka%M&X
z)K|A;^ARIoJQUnRV>!@hKrFAtCwEP7xDRxHAuJHg4YlOpe8{ioU{^y3`b3+$gQvx@
z`!Tg=$Tv!JH$Hdv=mL=G2>M1Ni!OF;EFsGsxo$0m<Y^?KbTmfXmZL7I`r>hP@g)nI
z$7}H1RhI0jY-$IPbQd}VuM&ZKA(x=s7iB1h-sybCK%hf`OXNq(uPG%;0tFU>>GB63
z4_bW1k{A=8%M4#+o{3~Y(nMl1VpiSXHO~7&IPw?AUA9kiOW!Gz*Gs(I$z#W@Kr|jB
zT&h3E*~Ls))_M0iF&Z+3Ym%dsTH|=y+|@(L9lj9!8136t;JjEJ&SuTg{^Jp5=ijFN
z777XRddHis3yGD;Hz(VLCU`mF%*&furKA<WN&>@2FbK2=OmDtZEOK`mk{e9*>$51&
zyp>RB(Vh={ODdwkGi73YAR}0XeS`h$Le<xEmQ1Mit2S%{7Cv7AGa;5?zm`^co#r&j
zI;-!|ghMApvjO~IlLk7Xc__M`$`ZiG_p>Kaq8{z9PGFe6VZC&Nc=`>V2D{uBUdu`#
zkbuo^6mVOQpG0qUi8l5Qaz7wC@q4dqr%xm7_WxOuLv9`gJErRa9nJOfSvb0R1)f>E
z5+&VqF*Z(j@T<LmN#p!=Jy@P!VX8i^*gIA5ecntSvjbY^?n^Ws4mJRL`_9cz+<{|#
zamJP|8SMJ}TSl9_`o+`l!>w<EHrwhuoB$nU8hf+xyEDzOgTN~HLSR_p)hDj<p@Ilx
z)DzBxZ7C{hg%9mi=6Kcvm<pPgUbApRrki?8#@UL+4&#j8yrX9Ac>x0ZMp?Vfo^hT9
z_R@%E-4prV-o$axBpKmN)xdT2IHH|=yb){G{~UuOG}b%}z;wc)CSF>KsCMgqwOye2
zcMHp3E_hd-8}K~LnDhPCYeupunHdVhl^r6DWMnHXxq3m~-3umC5LRwY`z1#({E%*1
z3F$3ZdV|pY*cQ+R4A{+ik!`PnM#pz(7zB<;*Ql@`%rb-DjvZle&lt!OUr*HzVDYv!
zeG>ur0dImh^JaMuzwzzTrLM45HRsZxlT0-cm&{p%WNl3xb126^Mj@EJU~`gJD|J1(
z&z9Hq50;!<uOBhD!>R6a{Vm`BF}iGEvn<J_LMs@HDeJQ+x8+^ok@O;64{|qlz+aIn
z{;OeC7O-<w3#x9Mq{3+Afpc*0?g?gG1*#~BBeE54xtB-gdK9a@xz39%A@@A1g2b|r
zYsM~Ll;#hXO3htpJ&>R)Dp%v6x$liPo8bR&%pCe`v9s?d7<3z+Pw&vbK@8Lh?+9a(
zOozhC2)!g21EZUnU<_bhlZz;<o2gIuKCov!gjF`8MgR1=Vk|yAB|bKMw_&)&;Ps!B
zB&hD=;VfC|AbDJCI^NADH2w~Xe5I<PQ6{cycUkj|m(IEj+tc*HnF*|Sk34}=26DE)
zo^_5T+{}cIjjQuV3)+{?pQiU;!6xOZ9KtiX1VgBp1^8XfI>tqO^kDIOf-6+g+%=En
zy)M(U1XCjJ?ZvLwY14BsN00u_TBo_Gc>6!PTFdUR^A(B7WZ5p6l1;E!I5#{aU~oOE
z_#TCB49fG8J(6C{Lud+ZqP<AJcP3_QxHh!t>0%5stW%T?SOx%4ot6J<|MG$ry#SQj
z6N-YnIlxbO`a(PYrEG#X+!9YA2&g8i0LJ#+dZR5VWt+5RK^T(6jbvfBa?Sp6cxq!p
z{(>{|fE|-4|L+7^dmCpIaJ}bTYHj{gSM(y-a7Kt|dWJuMP{&DoP`HYtPFJ0eVsPDi
z^dw(P49@P`r{%V3paY10n!vm;Zw>gpSY6R9E}!&YzAHG-4*2e?-3s(jduv}B4s-h_
zW$x{cJyZcNA;R!XNWH=*e+Ypt2^>hF&z*rLkQI{5fKdL6hv!h~Uhw26OxaR6`76v)
zp|&~f2s1#jUAR2}@-o5zz^m8Qi~a?-!k!<UEbqtF&2u7E1+?36v1Xr;>V5{|3LEKo
zowASu?R<(-;r+%Z0R@Ouk48pJ-%-JEcB=}R9-$e`>Q_P>;oL#$kDj{*+1Ny#vwFD7
zLO}BM!w=Wa$~BW6fE7mTC6rv&vf9W~v9`v2M=p|mO}_FGcggBHpI$Zjth3>#XAO*6
z7wAkA`eypLfm_u-3#cUrCkLxg7XT_YiKNvlI+DF;q4{vS9JT5e08KJRMu>^}Asj8d
zp2t*py+JeKCL4|~%?=8uiMhF>shGxjy*v>TpUYB5512VlVQXNqntI(}%-2AUmLEnF
zXFktmJIJ;WV*nh+Z(Ce}N0Err?WN~8h7smlcCQqgm_y&KxHH{uY(hK6iR|yy@+?m9
z*7+wR()zaOrg3I#pc^@1Gj|n4YeEA~p^s_^uStdosx;XMyugpsu2~G43G=bicR!L}
zZpdjNg1B{3Ac6B;`45A)iKLUGOs)May))r)z*(71{v{LG#hmoKK%ANE0{%5Lin24O
z%EYscDOz>d>rS(!hUNxyu9^_>fw?%l4SS4QsOln2w#+otT`TWc_&~ho4D%U1q%9%V
zS3kuva(8dzQYms3FLvikMPdu296g^7Ydlb%Pi8f6yE~1JPwSj`A#%!9os@!MR#0W+
zjR(ap|1Oi}HE(HChJGxQ5GbF5dO6+nfO>~6JZ<h^d3_p@{LtL$woaN@LlAa~q;G$C
zh6yaR<N9gp<iq#u6LebBxm#8tj2BUwhqs}fD#E^L!%nvg<%u4L=U-<Gl29Lreu@|_
zQa2kkm3B#XK+OXQ34jPcba_jwXA6W&^59xzY8>wLa1c~gX(ApPj<t<kzpYyLf%x_N
z1B0zAFa~;%o=zD=M253NAI=uTr<2Zd^Ff_?Ced2<+Bl++!fRfNGCdOAQ-g)|5Brh>
z7BLCBHQ#i?DwYZq;WR$T(H;H;+uU^d&@YMe>9A#mMWW~~FHz-bDtrraD!(EdU2B`#
zimJ^gv=wNvqvlY6rrF>uRlF~i%l0}rQH*DU^_J_uyzDsDinxu*J})R^^C(%M{|YM-
zC)lPsrt{UCpeg?}T>_0?5mv8{9CoW8N7it|vX`06K{R1VO=K=y<rkXTmO}tnd_yR$
z5PXG}lE*~yS(P|HFmi~g?SXs4NH0o7XtzqVcGGFqB@I%Ff)`Ts?ejvALP`1|D@fWb
z(L~;@bwA2aq#+E1F%d`<AnkwbX-#Yqc`x8vGVE=!Qt9J?!klboeU`&vl4Mh7qs1j0
zR=LtXS`0UXx5~`{_m)&|_CXSZs{S`NwM4Gi9T|+uQ~y2ZlR*S?M1$(^nKky&%s-ee
zR3bdz9uxiQ5ibUGeUi7vdt%%(LK3mW;}>q;FNtXPEI42Sn#@p|$i?gKu@BcOzIrn1
zml$Xz^S0cK0263_KkA^0fnSXVe7}>#dpb(cR-ux4M_#h>CtCPP9A-T-GO_Uk*GHJ>
z$8%n)R#m39MYI_l5o#~A%xo~0rL%;Dkm1TLZ#*E(9t3ZRUapAPHs+^focybMy3}ZB
zr3+v;p<9tTwR0EVao+2y_!Sk1q&VzwAA<SF*gEga`S?DfnIDjvEE+P%67kTMp^50b
zx)M@Yo0U$6eRePFv93}@Y41o*w8G0p5f(tpL+zmNT`pzPh;I36PdjF`^j*O14B;7D
z(GhUCIDvEf_>*Ei4II?qfgD!52yh;oWV>E17*fGNh#b+SLfqoQ)mZ0t()yipHNEq*
zf7#<DUM*Pd+Mc#zwsq`Rzdq0~DD0=!kkHiIRKcSo@8?+>TpEzE8U#0HaYWIugqvXE
zhi=QLF%o$l*QJSPcKYy`@eO!UY=uXz!bO6vl@*#Tgzw4vnV*oV>M|oncTLo_*!zF)
zlAdw_BEZ9kHr%9H&UV*RGhA%0MV&hTcT&9@{oeWlJE-m#@1ai3etW!#J5+=<0p(&1
zodvHrdg(de_D6)*#d$*loZRjGX&$Jy7S+L}g9OSJ{YjZ=J#15+#47BNr0KlkjX)*R
zc;4IZ&@H?Qp&kDWmXA-Y7Fnoe31D$n;$^#aBw-)*6cZE3FYMOo^s9TLH+TcLs7G)!
zmN3?d!YaZP?kdes>sxf*z%yI++$qg|o3u3LY(b_d*{oh&sj4)S;QzOUeJVup1_0@@
zp#J%#iBxck>9(vYO`G|<jLIcL)07^nFUEuwRU$gUCLu!ygvVo>9YcnW9-9A;?ktUp
zn|e&mc#T-1D=DOvVc=e2vYOtH+)Pr+h%+tz-T|n=X`Q<2BVKy2&U`Lm8*YN9P}FEY
z%CExFa~RUk{}@~1Ebv|4;_~PtI&8+bU|f%WZ5bv+vj`BULJCy1%lVxWWjM)%|3mAy
zDeMCo^L*wNX>`=*hoJ9B;XE61RO$XLu^Q(mK9E*f`Msf`;-xmuVVFr2$lVKeKyj=q
zcAe`oEFeLw7EhchVqa)!OA+=V>x0ieqjBgAr-Kl&{Kmu<oq=7G7j0n~%Zrw4?`Iq|
z3JM$<mtT97JR2lro5fqb36&n`9^KseK5={Ij);y5>bWnZlD3$OWPzea@8?#AiG;fw
z(3lMu;;jiy8jkN3qT3?6?O|pdSXsNvdjmdqgy`;aQi4W+0WLJ+dE0$QV#KJjykpQj
z*=GzZ@jPo(!ZxLN5X;!`o5@@CJirG<1fYCqIDA<^OW%%BfIe?G`wo4EacXc%VWBk6
z!rJ9c+d_|Q>Z7QTKUM9bu|7N6Z*6BzEnr-L3J1JY;i45U&?N}=$@*G-UM{>Eqz())
zO{QF8H!z_l6eCjxm!E3iX&2vqQK)71Gudwg-w-FAQKFA0Qi-)|r|2<;2GsxNc5FDF
znD>o;2{kDip^iAx;~T+KLr*n*M1tM;cGX*|Hh-J<XC^N&6P#97bh(~Al_PxRXZ@a;
z0F41>mj#dmT(wFS>;Ml$x`|wD?~cW!$6g?$%6{GF^jz5J1;-<5j-V<E%Cm=1u4AF^
z(wq#9<;60AUXH!sv#Ab5_PK%XR!km%@QO@zk{tfbC>Z?6`O*3Q&80eW<hm|VX)A%_
z_ZU2pASm+ITW`s~P_L^Ok!^G|tddc5Q2!eof7&DHdHTLN^MeeEf^{z?@vv?mIzbMz
zS2S_(GbX<fA%ExI#d3>FKFcqI1UOeIeQlzo*+fk%<}g8-HUSgPiG>fUq*D?UkAkhG
z?|N>Z_Nd}=`K%e8!ovc7p~?BliT|PkH-Wx$x@VM+c4C+5E02}Pn8=ix9j|}bhU@Sz
z1S#C=nm<W~v=O;U3Nwz+QRmnI$|0lNSKZ)v%5-=H5KwJ<3CHWJ;810@nQ{!tqnqWH
zgN5B++?XhxItlnpN=W2&(+JnwC?P>LAJCVADm^bd<oU33;`l8IPphEMvb^$MCEGhe
z8L_TJMRhZS$6&LHmCr|K(%{`}je{rN$`G^rlhU%CQV9$!^!l>0*p($!i1WcshKs$=
z{B<@5d1p!-bD40uTdC<>bnNEmGPi{*>B$gj59vpgO25gKsTkkAa<V&4xJ}J+zJNy|
zYYt&4&im|b6gpFv{g*?)S#ev?=B;3_oZ+KULIkzt)Ehnl{mc7!g@rjv&_{h#*|VSZ
zJY@MzBMS#WP>jHS1{Bc1p6J>yv2@Wrh2~m~>vBLof}bq>2-EEKZQVM2?h4!=>-Gv|
z!MW%yW58?d>B(H~7bXvsq~u5k4}ys&`LOuIS)MMT;oJsR<wr0hm5FLjFe4z1;^SU4
zqaTS&_j{#VX(T%4urQj1oTBhU1K!Iw*#^hxy9d^B+BWJSI(BC2f+)-0a<T;PLi;;0
zoZ|aME7W)3y~l*cnAkG820&#sXeF;IN@;2SOCZGqK*gsCSN<cte=LjIHY7`<Fi+vQ
z8i$OpPelkcoi~j`XIP`=#~qziGnQI!P<$of^ut5)A6Zz4a%e%bWH<(V0J~B%CPmJ<
z&`9VCFHLQ>m7|2m6AU5q+yMIKAYiTC3eF!ocA8V9`ADdThRM#9>OW;3rtTb=mD+at
z&b%WQN1r_@AiwmKIsnPCd7sgku>Pghm_3{6;LgLGfw)FuPNZC3hnsmUPhPzIqqa1Y
zT7~IQv)i=})K~?ULf=SD1e?&Kl$sO3{hNKkiUTrBT(u!Y`HjU`VZ0|>vg?^0JDIQP
zNO?i0pADVbHc^z^grB!p1cuaNEuTvJn(dtn+IXlXdaLLu0i`Gof;zRR!j~V?3u-7;
z{v^veO}ev7;Ff22jix+gMMWuP8dnQPC3A!L28`ha#llTKF3Rjq#4aSi@w0l=;5_*E
z>0>RlEvz$`<REJNznk;df)f@XcT28+8SL!6aCXg#CFC=i)k-Bv@vfiv*lCcx@gpM6
z|B`-iXn-$PN<eMg!o#=njbxzRMxV)&UZGU`Y9btZe0jKDfZ&w8Ik-WE{`I%yAoyTK
z^ibQhgG0&j_7bA4z_k56KB@%=GErcejsXib@;p3Mv;-RJvq5_7L)kQf0F$4yDi`;b
zxdN$>!!Io!?#UL{AfL-3S^#AN>W$%b8l~hc{Rnne*2zDy)Z}EN3FiIrTCa#wM)K$l
zJ}@i+C+>U9?X)jtE4V`f?_~v~ER_4-eg3*AV?GQj)eM4Zw0zL&+k+_@ZBxk*Ul8}r
zq@Kb@N!Y4F_z~{v((d-uP{%2c90%lZ3_%?3u1;97s2>u-6<pS_0Go72OffPNYun>&
z8G7FJa}cAFhEHJo4uG@MYQ;YnZvc)@OftPAgS@d+T?$>D7KTArjDIW_2<uM5B_Yv!
z_N&ib9CHuq|1iMuxwz*-Gce}+a8w+M6RzicV6O$Q)7^qxh;u(h<J(Se6U4ao;pJyj
z5q5@frLkuo@>W`v3>oawro(`3q#WM7t^f76SQu*&PL(aWmo-$ryjXkrUsn{EeOV_8
z3mH+_8Rj0e!oYf5u{B163MToD)xHT=DASe_h5*;~*A`T5Q_y$(=#Xn)YPsKVWF<#{
z>Z%QX(P0a91XOrL4fCNL1F`w*OOl?A2+OY+Qkf0EN@_xtCb`Dcxt^%O?&9<~>o1_j
z24O)Mc+#zH*Z&x_u6Rva%YWK!d-x>=?xVIAyc`2gU1@(tb5n!Etk`x{fMpjOh$4|j
zk?T?(ss1q15Hp1|mX8my%S+>a_$*nu4bf6}Jr!<~g`DKlm)3Fd-4>Pn<TO~K%#fpF
zxOtw%hoyZyYCs3I>|W@a0HyTtv<jR`K3z<Brds5p4T60$*qU^@<?P5Hj;#|qvQo28
z38vZ>J!UeD@dUUDzJJC81yTx!uj`eDFbu9L+R8&7MNLzctBwAtQ>$w85ALyt^9}Yb
zjUI9Q10(c;OFcnGUhRvpx2vVOCe;#^`l=v|MqDzv0YyR8oNBI2t&+zw9-6+w{Eb%<
zJLad{7Bbn!8$!Tws0+>Op$5Gl9~-98I&o8FP;Oj}M29PskTwD{(#z61k@@2zJayKj
zXV3lo;RaU-x7o+-YgD<<US$Qnb!)0QBZ(r!(bQt*?lq!7Hba0jW$-ET*bg5RJ^XiV
zF9yUXe9<n|qBexD^h1)Fe<k(FL2JN7JR-;>tlN#Vvu01mt8EAWt^i$E_J@(_8x$-o
z=_f5z?`!ZeLWrV2c*ZistJTqudCyZf5;FO@2p6rl)jhsJxX<E-_)OdxMj|QF7+~*U
z*m(($u0wOKmQ2+}V&9FO#~<dlAl@2U?t`wLLe&k2;^s%B)<tE{z=#9#ei}*>-t?Ax
ze=L6rTa8syY~&`_skJ5C=#O&XsTamO=|&x@VEE0<6pQJc2|F%4RYrflhQ|52F-zw~
z%?L*yp@Aa1Tsc}G>A%crQBoH$MP}#Rui1N*ncd>tDZts@G;!XTu>Bjy#|l4wBUUaY
zZV$2(Q`=98>?y|+oYj@H*oWF*pbgSb3x8D{Fhk}gS}xYI@3scKl$h^7S<zxy2Bv4B
zAQfd<LUGoT?vqZ0+_+WY>;h?M)i`CO)|{d93vr8G@XZ!WTtWjK@#CwiT0&VbCW>UB
ze0Cq~E+B>5<o2g2_tOtdFsgtqx|$aXTMV+3vH3I#Kk)RI-pd>PMb<vqVpUZkLKlp9
z+>Z>~06w#%2B2Z~hoFWc1I3<;_wv_#Dj5S!lLJ@+?sPk=cRRulW9~};ye88W@b=%p
zH-l=mAS*Lj?#x7%zKc%S;LGps=FKKC+vs+BC#@p}88v%B@@?>@hm>Iz&|Lt_Y<=v7
ze7V~@zA8T{@wI+9{4gOzjOwFUr4{rzJLNP6Gi&v|GIRyoo0>G2Lvhh?tvhsYplIN$
z*PH0MHpclVKI>{h_MQzo58;pFd>%g~Fs;2*I$c)jE4lNPh;`9k>DTxhD1(68a%>3d
z51E=~f_JWGDw>~^Ndt_mq4lLizNOJAULP(B%7=4g^1GJw;$zp_Pl~u1l=qTa6ZEXa
zL;DsCtL$-rNgFYVwpQ~!2hnL_F3!t9d+p0X{cFn3ANEIF&LSS}G*WB%u4=s{nzHXT
zx<PJ=3Sh7&(xT|Z6jf}M{xd|VBW|~dX8<rAd8E08U2|sXB9+5^UQ)EII{P4Yig>u=
zGriW=PslC~8OO0?rfbKKmXglE)SRl%QFo$dGo^tC^FKlUEEv;n(&R9=@HIT+?bf4B
z)0LReeO9V;-E=OmMW=l*Qg?)kYd?C^kGvaeW|mfb=l(N|Y%{yDG6^&QBteB2E^c_t
z7Nxsc$B);n9vj`uvKr5s{1{20@goK~FhqgYeAGN23VmdP^;FEMwq?8x4@PnCm(2OQ
z5W4z0kCH1(H7U$IbO)a$4=X6B94%YCH?w3D!!@gaD)HFkHQUJ8<SNS}sY6KLta92)
zH-I{K$@e|Cevz+W1t-n>OZZoi11UjXOb~g1J1bTjhrD>`ejhc140x0$g8r_j08c=$
zzaw0l{|sixM&>@v8nlA2gGtWxE<kEnfO6uuP_F#BLAJ$==4A{{h~m?9rZYcW$|)00
z*n^WemK_A-LITbOxfa_GAP_Q?^e@TWcRcmoDy5&jlvC)qgL6|#Sx}--6Mh}l5v9T5
zsby32%j3ez-0ac}Gci{B=?9Jjvaraq&pza3MI_gMmkX#3QNz>ggj&#SO38GuWQb1I
zSdll?<HY{bi0YB$L30phuBTMQ*g%`NodAMut`(6<2K&UM4p@*ssqke7&X1MlHAF`R
zA?55?l5^he16_mF1K^?b!s5^MfEdf&a-!-zoJ*XI-MObU|2iRx=M>^2nFu{>Y%HLi
z^r+JpofX3RCEK<XPkx=aAqN*v;Zmkehn!UPcwlL9v#nL~;6mmO*~0wLlDTcss#mRZ
z4`tQN5-8Xk2Ea`yQTjm90}qDaa$TX2dKHUrK1m;CJaIZN2!+x9NgOAL>jnyH3{5d1
zz|ZQ^MLe~)D|J%+z<l{Het)U6dJxiz76Ce5ZKWX@qft-fguauWR~%$LdjuFhH{YVF
zaO$>87tV1V(}oyp>o{uT4OTYZ>R}EeCME?h9|Iam5ngTOV!nOTduU#k={$3xey=)S
zae;J#vv!Jno65M*mMV1I5NhXg>G`TEX!#f_zUD}?VPm-(8ZI{B<j0j2AzxYHeAt_8
za7`%57GYL+grEF0WPje}tug#|j-e{<UMlVI#zvm2fB_a$VWBPrrWUtvBr?U<(0Rs<
zT_>#&khy3Z#Z?2&suvg??z?cBZlB8wuM{N883;-x=eN<lobTmUuudDTp=#Geb$fRF
z?c(UmC%(etFakbtcCq@!O?fRFhz3ec8Yg?nl7R<e=JDSvi`O<LdF~Rt??v-Eo8F8-
z2Qy_u`28`*U2C+(Sotl#mU|*(9t-F2M9zPy4Lkm`*5jj?oBngX4Hm%(Q+GlF-xyaZ
z2okX$%kQjZZG1Fv2I&K?Mi$k%Uj-fx0$Mu_>O-Qamc61lI)@z79<g3LC<~o6ZV()@
zr$|O@#j$dTzAU2R)Q8O@vxxpFTB||eSuOXJ?KtH+nehg^%KWB2j{K&>7xkBr11ggk
zFJ@r^0R;Xx<Ekd$cR)Uf8nRLdr}_-Fe)X$Zm?R3Iu&J@Q+_E8~Hf}08eZx*Xy0xR>
zDm-4I`*sCKs_^oI2?T;vfnU#q`M?s50teqBw&^1FHbj<6P7duXAWkyBa4<SobHCQn
zta&DvqQGz&Q%B{R94cjo?Z8<2mYQlKe>J7CqRfV40Xtqdrz2H8W}C$WMK_$9f!-tf
zpoHw|dLKDj!l&v$SfHWhXua6c5G)H^B=j!r9#MoP+nk-$%D#k>PL+{OO}I3hj*eDq
zG@QPsyNt32tl_N}Xx^hoKIufv7+#`5-oT)#ReR=P1ZaulrlR+LZTD}g&>5uTTD8#@
zOcK_V0f_4|lQ>|~rc8(;OFr}|apoxhf+qsLX(q4FCmgO@28U*EDhHS80bF~V8jn)9
z<by2XIl5trikKf#Oy%ws8Z2FHx1<SjTQG<QWI34_^JU$dP%!{n0$2MsiJHH>yTt<3
z+j5;zK^w|Uar_Fp40YBewi3{fZhFwtpNOZ{j#}2>pzN5tk)g6DJOqzl7?*NijLE}w
z6XUTYIsI$Pp<#;8?k~2X&$mmMy>g)qgI~~qNX{`L{-hlc4c7#*QYE?bi}GF+Wv2|3
z{$aWxMs=(^Hd3C2NosdX9OX4kXYEh#lNpaEcU#Bpwg7?__wE{DU`pFP)x5HWC%{;M
zhtLr8PW=LFM=h)Q=}WIzwG203!PU*nPzO|meV%J1=SFDcPLhfT^w13NLvg@F`a5Bm
zr#!d)THDCTl|GQmf+0uzb`Hh@*f!N+i1h}jvax4v5Tu{^(5rIdSmer6n=nj0P1oE;
zCd&2)Viv{OVl<kS;c@s}6Z7d08t_#cd$-cskpR%GD3ms7>;L&E^kvFG6=SIplN|e*
z|5CYjNNx~zJJ<L&p75i$_1BysJ4a~u8JP@z0QL*bS_aQUlELK)JVf^y(SD-N3|oR@
z=U4;$k&Dzbhfr_GvSc2&7$?4~YcO?qZ$GLyvrKF8g~(Mfi>a42d~uOX7u?SYQ(}U0
zssbe9;XAi=+g|2w7bK>PEZ%_w-ujhfh9EmE4&~DnCVUY4zkv$nX;hGI5T?Mv`;bys
zO3C(PM(S|pXPYqf3t(<S_UA7^$5;A<>~iDg6Oy`kd_oO|6;~*bEd^FpO?TmTt;R(M
z+~RsP<L9>TO_D$}wyhD*Q(fHfi5*1J#Pm!)ps*@yu(d3LLSH+->z|ZY{Sc*-z~lfQ
zjR6|Z{)zjj8n3zCiRo=HiAh-(?POA>k19w7P=;E=x2I+k0=C6pVa>&h*q-#C6;+#(
z$JnNt^xjDrWCdPl?muR)%873(FacrJb~VP@r}CQ^|3WCZYO6X*mwBPP8^Y***ua0R
z)=#$-+nNEA2vuV2Ta=w!<m_;hW1^bF#!1gMs(oMsq@ibt%0#=!44IoaD#%hAa#Sz@
zoxS7CU{7T_7N;mfK#-Kq44RSw^CXZO#8Z(LFOD&4g~bVuS3`f%cNFyYrwocqTx#1^
zt;|^7_v3ZCqyt+xmyA04?wX`J6F@1CM0!1iOR`Lf?>$+8Hd(T%NfUXEx4+tb81byC
z*)3Z9pBnaDQ#@TaH)U6;j2(fXhuYVloO3-DdJz_ekzlFB4F%PABUco(#fpG(OR-n2
z*iqUq;Qg5Yn^OZ<mgurKspOOFhXxLMb!U1{N*6;~|HFjMBm~8lfzX>r3@sERa}9EY
zr;)2+Z{4<WoA0T&PB(%2D+x`E;Vg-yffQDPxyg2g0C-cE473~MFW8ilq7v&jFM9r5
zZY#3_D-!>nvZ2te5+nG-7iS*?KR2KAUjPDM9B)WM<d6|l{WGotAV3rRH|(b6I?ZJr
zjrpd_TjrH`M<mgXa;%#;W<Hk6brbpz<Z31}3*PTd<e2`AXao{si{nSG#qJcm{knkA
zZMY-ywCFatdBY{c9q9`WabIYB=q+J+vn<Z+VbUZb6?I*GlH_Z_Co(4;qw(!v06G1a
zaXZO6@hlSI=8G?Y7tEJ;tFSj;3M({aJA%qC-slO}kD|Nl;O{-dj))Iyju5Qlnr;}l
z{jj3b$v^vBvBRa%=I=8zPN)fw8Ty1z$Z6SRE@pO&o`FDb!`%P+LKaOkF54%TD?{Sg
z&aYa!f`Wg;phD?LRu$NlW3C4r`YHT8OXl@4S_5AGFb{{z3DQUo%zpoYi#}{aS5Ql)
z=M~@R(Ft5#Q)J-}p@BQ%TUMib!$5Ji4hIh_!Fg~SO;i0x7`J~$h>1AE-NBqqc&tZ{
z8UHBL<BdjGiq->lbNUa~N$04CLF1+?D>$DOPdt|>QOSizcoIeF?9Yn)WW=!MWs<;N
z0(+R9NTe}k;Wde@yt^+Y0!NSNoVhh7AlrWj6n?m31wM7ftm_E$>w@rm?3tW45%<Kq
zt#&@c=}F2`1yv4Mp9m)4PmX6we5_dKh1_NCVXC*y{i2>Lt(@Vav8u1?o#@Y-bVqXr
zBuFd!BIswBEiv9@578|q=MX22u{Y}#3?G2E93X{jy=Q<A6MK<QjtA=R5fQ}^i~4WA
z3sgi=5R0z5L}`R*zs*&70itz$TBI7gYGA3ebX;B4*V$Rc=a6p|Zid&1O}KA@u%&C=
zc}JLPg-kL1^W_35w?L}=>t=7>fs9rgivUT^Uv7XcYS1mR#cibHEpd<(8go<C_SA~?
z3@kI(7mu7Vj{6HLT%@L+#6#V3na$EQ_N)hRtPliS@|<|eSybjvw8|cX$-v8|)6fyE
z9WQo^<uqWynSD$6Y^@&aGUahnlg2%em?gW^kb+UswEx9X#PsRR@NY*LnzClA3JhYN
zD62(KaO=TkA;xhH8xg@mc_dqfT>QBY`LG>Kyr%rtY=Y2USzoYRwqk)CV}5i7H>hl7
zxf;4`?V_4vLS<=m*CzXg@7fSF*+JWC+vH*jj<w+5_?KqD>DmY!-MhQn+O9XeI&#~N
zaLEoI-!rQMhg7VyT_7mU>CyCEJ6w3;kji}67rNbvR(rLNPo+eEJx5FZ1n<c3572o7
zf_GH4u((Wmf9Zih0?`eUHIwe!MdVYvReY8a^*pgRHdVcVfx~PCzWm*7llVyfK1s#s
zA%pz8oAT&m$)TMy|74Q|W}lpA`1<tEA{QflL5u$s*O31gv*|$AUV2Of1D@TSSebqc
z)(X((&(QKT&_wtJep8A-yNkQ?LM*|qN?W{$MMjKs!!~kl+D(v&{G-`uvZqTg*x!cK
zpBT2Z`dXHL_{}cPKJ_l5RrqUB@GdS^DRa|s*Dq=ZaybX&=H7GSMq)Q?#~;Zh_s+4a
z;8ZG$9;?M~V2fvihP_Ee98Gwdqr^DmIDcKG<Ti~PpvmXQD8Cs6V6iG#(71fh61i4z
zq_znJrwPgN4|dniiyk=A?@xzGiYDjC+fZJla@+=&USrb~b(g5|mWhg^e2zs5^7|{9
z?+IwWH6tsB9}lGwZo=$F#hlk9hlTU~y@DEw?$y*&!Z^9V%OM^fwz4WIBwvA6YCu;5
z5br(%I{tt!xaz9c@-Tu*!m$g|LRy*3BfnDP3ik&Xnhb`{m5eY+0NS|6d{(e?Uh}fq
zN#>z{TL@zO9alGW_vYKP?InPf3m=|-ATMKJ`2?$u>SwOg<hAWliFlVGgdzI2lxU|~
z_uKrVf^#HUUPtE`%O8I##$lbt1?Ky7AzDO@3n7j3!9`NXz%HYjj*GaWixbHaR+IWK
zI>&ERm0vlX+ucNJlUbL?X*L|)?vV+Z@_tBAS$+(4zqMY1kc|DQ7RM&UC(k5)N3?T;
zGKmi}A+u$T^j;~5260uEupSQq{bz#r?V@f9h@-X@CF{Gc@=}xlC^&N9K9UsDFuGuC
z#t5;8c|^G)227i*v|}rc{X@t)l4^#)Z74sPq8axC6nQ2AuT={@z`)9dYg?wm0lDSb
z!BP%dIQGC@JVz)J4?ng6s3&DoC0D&Ukm8{4?SdzN(3SBu*a1IJRm0Ue+AzutCCE)S
z^@7EiUZ*6HMx}1V$b$g(bD}Za74FyhYz?!AwQFky_0uX4E?a36s2GawzKCrb?~ilD
z3fLy;eXH6aaCgKg==Q2cV;>shD9J{JQY^ldDuPv#`|;er=Q-qbjgt56YpJa{xYGv`
zo_xC9?n#N_?wcYSq-F={2O}=aiC{Hv{Z3Sapx$5G1yuuN*Xa7cCl*C>>w19`0Syfx
z9zUPKY=w^aKJgdn<ZMvie7EdeDBCSK5sG#(^uc=3t-O9H<Diu{;vXC6FTmd2Jf~m>
z-PrHgq2{^849fJ^4q{kJVg8H*MR|<zSy@N|=CMXe%jz-F7vJypbqjXbNoVgt^p6+w
zS{5fFCEw64$f29nL3f#WS}U??#{iO8FM57rX{*5)zzS>&_!i;_{1`jb!gFfHlfm~O
zlPs03)Bhrx5We}4Z9w@kOXI<Q{1yV|Qb7+rzB;R5j!`4@^YV&rvhE^=QrY@V2%2jj
z`JtQ<A2#}2Ga3@aS?w^#Ymj}1<n{z!Z$Yw#yzYRl%s<^kgW)z??ef;Nae-0sAb`3X
zC-{?=CC0f?U7NGd9%DaFVNTNk1~36aANB3ekL%3DtIaD(+@ooi?a%>#ft@q@$N$`k
zr(uxk-_KEvSt!9GD3NNlKo*#NKt5Hpk}I&o{7U7n^Ws5vV+WWDyj9*9sDnowf%|2#
z9PpR#ThG545(%eo59Lz#1jg&Nt(VAI(cKg`N3MdiPw1bmG4J3a*B6;EmA4b#(Sv|X
z`2uI|uDe%7K8V2I6}-($>Sn`?Es1=KpBv=LO1ly^zv05d?dBSh5X|X@Qd8|W=#P{6
zt#Bt_!>sKj@ngTePz<fz77|`*m&>aZXz>g7M+zkP*na!%bF8@_n{on;{*gO0{yUGy
zM^!3!&5lA`g_&=sMOa%2h_`TQ3&eP6(e=4O70EI5%rz-s;;KL$fOqEehZkg$;G4jX
z!)CRHYkDE;FF~Q$?PlFC0^RDN3)|V(qfMX@`5)7BD$AVmO>k7MJAKhRS2GypVe`yW
ziQl<Cc-sb*)1|hzh_CM^Y0EBwwCehPDkQ>VSz8<dR?N7%KKVs7WT<m0!yTPF-E%bl
zU(NpNk*tEm0Q=u%qe$HakouX3Ztby?g9unH>I}$%z&(6`xK-Yp3$7EEH=jDU?rox4
zHMmKgph=5F+vU`Oml|9xDtDehGQa)%va4eN<~1`BsI>a_^h1>WRBS$MkJ3VjUNYd$
zRTH$%jv-%E0j^-yv0{;AUs*ztVlM=}(&F21CSAs!)hzVL`-+hgg4Giw(IOCvwxPco
z$pz$}?v-qG`L>`Rjf|ScVS`y0a0*QzH?(V6S?`$`GZ~xTRn&LijMIg_bduD=2r5&H
z#1T7@Mc-4%-KxUYzve8cL6{Mtr}ncu;h`>jdxoVxUfibVIWa7Aq^nPb*JPC7OM#o|
z>W9u^yqM}LK7}kM5|*P0;4CMe)DKNJQ4d(NX*j*#87;YbkvW%P(dllsS};EU<PF!s
zwBo69PI;V|zl=%e#wQG%Nk`LaOYo}j43=d1<?VO*fnO1`5ChCg37Mx(x?sly4x|UC
zjb>+u2&1bMz7fGcrahA2R5U;TbZ%JTZps$OUFl4Ig2P%2TYD!&H54#7O*1pJLH7V_
zryQ@l94XCJ5?uK{x)9JL=>%e#^}K0b;Rz0u_unc&hv6VvoA87fU8^f?l@5Ph8QN}u
zNB_gvMg`WL9VCEswysA8KOelPeGKYa-oaS_CXpG)9N}3&W3913a<}&VUb>_Qnp9)*
ztZ|pVw!#uomwgc38z>7ywrhHI5Y)S`ssBOW<vpA)8w;RoLWnEbiaHX22Z!Fg>*|%R
z_D>T<3to`huX&=^(HVLnx8n<;mQR?6c4^&!L&wGmiSNi8)P3}n$~oa&T_Gh#&n1ck
zF@Yoktar4?#>+NiRx#rjChEYpfHnbc13yKqMG<=k!icxeHLJ<?hF~?pHmOHRVwkXu
zPO$JxEx3o4h)8gHZKw`vSW*0NX+H>*@zMq`q+wg+AKAoT4(gle_mD4k0qX<Qk~TUD
zkKaTjTu6J<++ze4901N;G;}&RNAHYZ)y*`1zc?+T=u><C^<ACnH_6Rq&Bb{N=mdM=
zw1<38aXHM!&qdA$yADpfb+NtkWYwZZ4+qTjetq7;^?$WQJ~tl@P}xZJKuCyorQSSV
zgs4XuKHww%uPNUPRPer>YXc|7!rP6(N4#G8Uehg%@V_7_LBa5Px>-Ifb!<ykx+Q?c
zPftV8BLI&_tZ}~z7G#pNNvZhkpaxc+|Kf#2^R1Ujgi#)B2g9|&IMWTrGs!lwP6E8a
zh=UfHwx2DVXeaRPE|oI1C@~AMD=;YO`HN5|UB_Gzac?gPTyrX8RxO5%Kuj|9aC_#z
zuYPkGCKq$sFK#UcmD#~ip-X_rgd%Jvun8{NX%KC>;6wIoi&O-{kpo;~k$ME9%=(b<
zua*E}?F+p_!&AkY5%;Jv)NhvOH!Gmrdz)GZfM1!PzQZ%n>(Q?qcjT<3JFX05iQkqe
zR%Z-wqh7@5GsbcZdvqxk5iFgEG)x@`RP}bcV_!Nb&(5dGrigXB(5R6MDuZrGuPsZD
zdmfUDhmTlQDS)iwvugi{GDY5g?9W^)qxwR{tBmG+p1s3o=bOP13DY)Mm4eGuHA|6o
z&?MFHO^ni$|5!CMmwSm5j+<El#i`|&O!gZ()+`4j5qX;(uQ}aw`=1S3U7!CLvsXn#
zPxK!Y)a}ou9GUfc&$ICaS+aF~s)Pj$aQuCFfiH3d?JQ-<mecQ|ICC3c*nY`ho{Cd$
z+!{lh%cwB3<GL#r**<p0q{&NUtuPLzZKWURNSr}x0(FBPk+dbQg0)tz_cm`pczh=l
ziE(iE)@$R;VJ-7(x-L+~FuXb+9HfWmKuhj@m`lz)W{}&9ijsD7qmoF#I@KpF#8R3R
zwcoVYU^!xq2rS=mn0R^+Fb)cYx!WUryYL4FT|H3wyzs;#<8QN8MJdOmD%_?%mIIqq
z2pT9y45&eBip<Znr`=DsSn7^?RDFAIY!hHs)MCKK0T!JKVT9|&;Y>+;w!Y9SEya0_
z;W1tOhy>j6QsMGzkEU$_6YkucUAdS3Wb2Bw>hh9Q6_%ps_2hx;y_pDxBw;`meh~SK
z&)RlD4{aFEsuZ)UP5L2FbGgGE?OzhchCY{os!Qn^@FqGUUW=~|^D&x=|F5pzfw>j{
ziPnv6+qR7p+qP{x*|BXKC${b6#I|i4Z|<wQb!+B7boXl2w=6Yj1^s>Dt*g(-^;^D=
zd(<bCvsu+xUxCds*oy37%*YjdGa)Ydp77tQr~+tXkN0`i0r~a?^*)-%xjoeYyjkRg
zB==hApeDA)s$a@B7btwQBv~poQ{pe!8wwg{()2LDyRcpHwG|zNIY(h2@QRqlB&LZx
z-&lQRavD){wzG>;lR>FRKB5tjE=n2ePQbLn#VLL!a~0~Jz>Y*I{|uSedBm2PtUkk#
z?9-C^O!!c6R*HG!jeA3#hxBlF{u^yR7v|O{3A75{TVDRA(ccNud$i^tBSM-`pgN4!
z#8~_*ymbT$X^8XGgsP-0l!1}em9OoAUT*?$5MqKw{wq$ucOh~S9&>=@J41Q)SK@`T
zcsaotyZAQP%S=E7Z_m~B0*>SYLPh_@4(;pnb>q{Fwr!J$RSgAfF)ympH|;Mn7knAy
zQWFEF`6(KZo5&nZdW(FvcyzU;CUuF=krUHAiK7Ufa{x}HHpdCdmAOC24G%#qUHQEQ
z+N9Gwafj?>k#3*BPBf4G<c(Mn@8J-?B)01%xUHy(zuzDYA+S*^Z~&%ewpg@s_}x`S
z?bz`e1`A-jdU6Va?f^`H?O3Z4xB4wXF=U*g>LjoS<Q~T6VI8kd*O=jE4@CdbIeX)R
z{k^UHRcby~%OH}|LBV8U0&xa>@(f(eU+&eirjYI6=a(npRv434nE*Ct!EZzP$S0^7
zS{MJSZ&HglE27D#&bQ``jxF=m?y9Du=H=<0dOfsIki{P2Loec~y}SC4<LThvJ6ZX#
zW~b0o5d|JjJ_yPSCf6Em>pThje|=(mGhP5}oMNri7C0I1>WXVvEy*0-U2R4JBns86
zrNd&$1;fu`SieV`SZyQkneF3LO__jSM^b{_pi>TBBl@lpYJP|cSR>w^-aJzbr#I9P
z6+H@RwygZ}E%NG$YS*6RvWC1Z*>QPFN9?t|Oo!7;off=*D3$uf)260@Px10Rt}n>1
zi=6X4+Q8?q<=QaS;Sl@Li+<F_-CqZDNmn_{bZa--n!NHE9b1es&@ld5IQqAO44-Ln
zKWu5QF4^8gxj>v>`Xbr}8eOP<uiD}<M*=^apL*d5xzV?=Z({j4Uk_nl-ByB`8mkLn
z_2^z!ezH3RWk&I0GbL^?0xXh(95oZ2hAy<+3W-{v_*C7gFP1D?&&=_EFhB6P9;c%L
zG1LbtUT6`v-WJd_wv4C98$OuF6|Q|~$1H0R<q5)$P^-RkX)L2Vn`QLpWwy*+-C$*g
zKDh{S3q*b@R>^nV^e6kLcu%se#v~_&V@D-qqXcd|6(Sf)<Mr;tPBr~95($O1V?J>V
zJ8Hv^-)O|t_EPG2-98vqEL`ZASrf!qDKbOL@%=G)STxBu>4Y@jRU%$sx&6WiQFcGC
zIEUq4!_WSQqHehT8Ehnmqe0%o$jY<fz={+k;}kU%mk%U}<YaFOB=8_EK{buZP)Po~
z=(y!_%SixX$JP-Sch?wpn9bE0;D!FFr<6Ig@RQ>F{&Y;XF8fL0zquqeW^H{eJx<c`
zQUvsHAA^DZQEs@pAgtt21#=M09xaJtOIT#nkeOuSb>ci7>?W|uoHHb4BV;#^S;YSS
z<MZ=*Z=P1Z4CTkiE`IYF3H1+f4Snf!*4?N59ohrrhC}s>7cK%VgeEP>Uu6Z=2;Jn(
z|DLi}lx^~A=RUB-wcylR7PV4(q_|Y(tk4=v#c{ERnWF9ZY*9LIS+`uY1s0HY3!Pby
zThaf8xGAM5T2J63P>rr77FeN8#FACnki2vERjVzAY?Lm9vkdFoBiW;J;tA)vR2}xw
zn}7J%I2mhyP$v;|65Je4T7&2yjHmLf3_giGIQI^uZd|tR+3^e2WxY6d8f^YTv}i1Q
zC8mU!2_KB?PDbQ|P7C{cf8KBxKveD=dDtVQs)d_=FoDz?;Z__x2e7!#Kj@6eKmTKv
zHl7EKZU0V%r5_YnD=d1bs)4bCb3b(P!Z2^Zd(+aJ$N*BsY%{_DKz|#fm38KyJ~3*C
z+oo$Zk)o*xaPV|bKImnd&00bJ5AJ3&G1Di5G^OR63-^EvYO5=s`Oo1OjVKHh+Sv;H
z82Y$Z_^?&hC2N4Sd+tDj&iGZal*AzdJRxw}oVrnA_0Tx_Mj&AYD3y_5wJ+%ixUo+2
zVOPep+K*60@}F>S%Vl_BHe!p)LHNHceI*~BB43EM$i-_OYBAFE0!)pn#VK|K%Yf@J
z_`zr$5&81CFS7%G^+}Hi@wN8<v>~E*j)tU4VlzweC#c|FTFf2y&YUz8ado7*xW{vN
z`z~_0F!`yn9HDT3u_DtA(@OKGbloJ_lgx4QZexla;jcp>GgY?GoE?0>q`z2&Hv(g!
zm~SLt%?`ME1ge*bp_}Kxw1doIdiDb$>Qq&LJFjJQA;oCUB;WNS*Wd=iiME$Dkcbqo
ziwSnS{MHCyCz(3wWyyrCfp`)#!Ym6I0PVOq%guZLz(}0wC{kzbehvuGKMV^qF!YQ-
ztCocFUEZ^py@d8-io9HYNebqQE$nFlRms#SaMgplWMOj3x>HI8sTD>w8SFX#<lDL(
z_6XG?x!b+I!&I|F<BOT66-9D|UxJ|%510Jp3K2!c(US$Kr|&2(J<{(T)uQ)V8wbW#
z_uYprPq;D<k?}J&EqlX2SKU5jj_Jq-`pVgDNyo$$KL?0m=~cBDOB7&hR{r9v9A5~P
zpq6#HkVG~6{-Uz8cf9D#aWKW<dl9n2MAlXxEyX;d?Z^+X!*q;xzHUrzk(C;`pV~1u
z@{Rjcj_L_a6-CDHUPx%h2RVBkyFe@XeWgR;NQ-!?3(wfbA<rr?ave*CCeTskBorm#
z6E8xT4b(F2xqJfAY)jk$(U38lm$+ZfLv4ws3WfKSu?2nNze=c~cSYS?&_Iw@v1urf
z@jwr};y{bus#?)DPfIl{^LYFzE2qwC8f9}ao^8=X0Za(Y++(Nzvw{#(itwn5+oHiV
zHNIalN2rWw{~|Y&#hOr?<nLBQoTeL`8lt9V%Ae!$ekBP_(}aq0RhQKiY0U#Fm+~+D
z^lZ<+zkHcArh3gFQ#V}9BX<}EjWC){qk~5@S(*vV)*j6?neB_G1oyjN52)BZyfIXK
zlL`DXwfPHuvj>SWRt?o!*k*e$)GH4eeI7uamo^KNht(ZVZmJ3AF^|+#NhDv!qvhWM
zbJki3unJOqpQvcJ0TOKvPSZVCDBKq(y%6P@v_&GhVZ~m`OT2<z$37-_(J@c|a)34c
zmD+4;^}N2_j@9gozZkwNH>`hfE?vdr!#<mzVg{CQZ@Ql5caU~nCmHRtXLiml_Z*#M
z4r_gh7296pBE9!&|H5VPipQ`_E~H;Us<$@@V&p$rRa@JQ>jOxfiB1BoWx0WH`@zM8
zeNk!y@05@&YU`ifVitdL^6mY|p7*wlU<*!pYBQ>2(^A+#jZO!WiO!n?iHg=Zmt!>5
ze&}Cc8u-wv_|#SZ<iH)KvczPN@Gk^_8nNX9UD#oSyUJa^JF257|G~3CEy4Jh+>usV
z!Nn+S;9Mp@<ei@mMxl!!1jHBu5{~f(@-+$iDYmQY246hXyogE10bd%UG+|{+KbvQg
zYX;o}FEA@2JN{~1k$wKv+`Sg<_J(nw=JQ@;%kdyMOH;Gb1*G^@b~4KIJTdR(s|Ff+
zz2@O6@tI{UMZ-L2Od#suqWk!X#xsf@i4VhYNV&gX0rlG{?<R`SGy-Xt-uPapSP+HL
z5~W;wE^qvx`|Ml}oPE@jbhzvPssszIMD|Ab8Bq$}hOn-|{;^BY6Dz!hb_ZB!c|Obp
zt{3<;W(8sx)8YIlW+Y_WK>b|vz${Wz3f<T6hxBys_hW1Cz;!)fu<Xj;?6BmO0JcYB
zp&Z%M#VVa-M@#G)eND7RvS-CKme2p<MX$6DaUr^)sE%Id={{^1RAh2UJ`%K7B(XXD
zzC85qTf}#OrJSH<Va}Qa&N28*%N$Lr=OVp*+j%b0RgT|~W<W=GPw73h#xr#tnbou<
z6i1`5jL+EtJ7Ueq+1obUC$zm+U~JMw^j7Mk)sYC`6(<EhjU{y~Pab;Sc*(6)<L<mA
zyOcdHouptVSL_miR(wOVF3PT7UVn!<ia5^`9ZW>2{s_uKMlcENU)oRpt+n`D|04C-
zu#v>(N7e)Go9FLlOf8gis~?Fr4S=eOzV@6@xhGv;?fF)phiz+{5ZuYiUaEDGbEm8n
zt2WfpyKuEu$>`BJDm;gq7YMHMDrz5C@*}q5hx{RS9tQXJ^)pZs5m!vLUUlst55o1}
zTTh!6G0*Id;~tpEiKBxsw-<sQEGKx9VnBQy?7g)-HUEtcu?%`|Wt^U9PNSa3EwiOn
zyT-?{hXn=2n5EPr%X=23!C)HuuIx8|bX$PL9K><p8=KlT_h>^9D){pSNSzCETnO)&
z<y2@ogBa{axXzLM=$I?9N=%xqF1O(D%~<+HUx}jNl@4=XbAjQy!VkMLzr|nu!k6Nw
z%C^X8Gs9=}A9)@}<ra%b)<+V3k9lG*xaiiQlv%)9w5Ln&&<6G1PI%K)k!7i0`mkuD
zhPA;APd44flmSeLOw+<3u3)cMPlq31mf(?;?AA(K51(qgJ+PI^gp2Wg(!=W+1gkCk
zH0}TQ9}Fvm!{dqTz#UpaB1Q9YK#T@g5S|wcaZ9E&cvN(G+lPjh4EZA+U9ODskT=}r
z^jIXq`m#DW?|TjIe)!FDaAs+jKUb2G{C!s76sPH7Aw3OaSDJoM&Rp`^NnJ9tW6nic
z`k?dISiO|pD1<VL(PzMxW=je4j{L^$mX!Y~n27Da0iiQKcU<SwyBcKVe_S#dS2zP#
zI;fz!kE1?p+p`?Bg+tBBDMAUbQ;H{V)Nl=BvhQOTgag%*be|Rick_vPS9fCOW4tnx
z7UOzEk<=Xr0=*FM<nkM2a4?n@_u{RT{u>W&OK(PoObtZDncmQq`_<xWkcs^(ijSJ%
zU<b9tvS|_$#$Xe~U}H*GGu*b#YECMei1SV6%0|Bjb$q1Nf`xlZT=7BM`d}LFH5cX9
z<6eYxAn$gI2T`hcsVu3f{wPg4<xmCxl)qrdQ%0wtjdA^a?MLOuRydnwl%9f%q9$lF
z?%XGTAT)S*n`U!q#34?ej<Bl9aer0yBC~!^G|^55<n7l9LV*GXKFlXD+c8LeDR1n>
zPkkaF%wgCQFcp0R`w2<HYFlRu!rm4l>~uqZKlEA$QckkCK_-id2MYq%7sYnyrL;}K
zgYw7EaYu&PvVp^&d{MSC0K`A_ZaDZ#l8#1Tm<br?q{W27@@IWwq~L?<2i8?|!u`!a
z{hK6xq@&o@Zpo>k$?!QPRFlWlB%|rbE6d7I+scUqW~_k3v81BwljK?vvaRG+SvZ*3
zBC10@3>VC%WO}DaNAQ0ZwJt*X1O)!AJ;bgzQGt@BC!noz^i%&$z1~N1RN=%^#j|jY
z|Ho9Ud@s}i1P=~hVC`*tAOn;Jj9h2}cN^O!xADsKi#<MTy;HhFmpE~xq)RE-_s5ev
zm8km07Ph)T&SDb7?vUZV^(xKn)r;H+HXN2{1MJjG*PdHC(z)Thpf3DbWf#9<v!CC@
zq$|F!=7-r%lg>M3<3rA57mg5X7fd~T1fwYnJJJnbw?T;+U(!&IOKhFw2mxc?jf0YF
zX3~wH^e52or_CtT=^Yd<aNr*7uW6_G|15bOU8>FfF;8us;xc)Q5bcKgH?W;<u++V3
zZ?-K%v%O=Zl4n9c`$2YeKW#cni{pc%Gnsq)hwQ~D`K>VAlRH0*7i%iE+(P`4rXf3k
zvjah1shx4Uev#*$L2pMVzh~>9TDXAoZL1GpQtiZ#SHolrO{in8D@c?jo2Qo_;yiq}
z+A6H9oE#zs{d|t>cwd%aC#O>G%@L#y7o>r!>skH`8^ysqoG&S|BU*SiLdZ26@PJ<h
zVXn%aD~d4}mf|_pfN-elzv1k$+VrD*Gtk$Toq7je5bg$BR!&B?DJXM-I(U%-Iw)pN
zrOJDX?%nV1`)8$t5g+N$3X~Qa|G~tO^Y?#`RAT?gEbp`j+fD<M(CA(&oL@_wVfuPl
z{hP43?JwiJmd3g`WvTo(WB1LE(yZ68QgM>CsyYPbKAS!FaauZ_v!KwN@nUX$2O9CJ
zZ@gGh7+SFkn%9Vt{%5L?3_~+(%|cP>9|#7<oMLrKzS@^a@@WaL1_suuXCLYjh4Z}k
z(Avp0UxV0|zoTwY+uxfl**gULYh`2m5vz^M9$br<D9vHs4(i)*k>2QjJ8qlwAfG6f
z<U&4C6n^k~SDt=Q^e$#!C<khT)!~L^ujHnejj1?)x8yno&@(<0c;F$5O|{`r#QUc?
z!hG9>o%EsrSsbE+&NSfs^i!Durm0I;M0dNy)HP#h=Br@kusrfM_X97IH7D_w6v9n|
zmWk+{mVAMq=}Y*0pp8qtMWReSD3=$=kuOm#*FUK0iyi`v41UoDM13{uIg4^=boM<t
zJb%M~6YD^yDH=g^DH(aZ6LN!0I~nwd5NZ>2+l^z-e01nP%(~Gj#lGK5+BU1vTr3*g
z?z`3{Jb(%?C^&g1T!aM*V}gS7)e6vGYXHJuE>ZNAU@Pnxb0|fl5~wNN0SFZl(`OLJ
zFzQjUdGIDnotnxXCZL5MIQ#k;jl18F45*kK?06rUo7fb-Cz8@HH#4D7rwy)vIjZK3
z)P#m?S?U>r0O!ryj;vJeb|*F07c8Jp@y*6LoK>}nY$$o*w^LIBoqlV|QAAm{eV>n1
zlWnL0&LeGYiv&oIclFU@oW(8JZ*mznT|Sl>T5NOo6YdA+c0|ogzydWd#dp#k|N26{
zpEB*Mq2@u8rfp0^2Y2VORx_topYbU#B2g1>ISKk4Q^*t-*iow&+Aojb)$o%EfzJt8
zOz}JW9{-RJ&E?nJ1iEfs{zrLg85@Dg+hKRIMfjU!2Q5jsr?|2(;a7!I69P&74X-*f
zfq>i?rZO=F2DK^gxn7D1a>*}Ppg=&S91JA#?9hIJzf?iP9)Fz|sIWW@PL{t^tQDKV
z$SDWNk;z-E;ez*@RKov#<<sEB7D1A5CBL1$wLz0b;;ru|BHPqIVzkB)r@dv2E3#m~
z6r&@J0sg9{1-&Es5HE(QmNnlM<kXhRS%Rll`}a#NU>MhIxRc!n$E6cW8aEujfX&!*
za!CL;y_T)+`12bo`MjOwuJXRdY++Jz?jbcn(%az=EW(#8yE!E^hOTnt1NMx<nXsXi
z81fV-q6rncwHcCec?;Cp0cMnJKLwJ6CaL>$-eYR7BBL_89^&_<KeTY@7tx}aHyruZ
zD4yn%BKW?EMttEP35E$9co~hnY!^Q(I|Q|G|DT*zY4WVzA%)v>boJN*^+$}oJ_$JC
zZFmwkwv2VY5#0P;GE!5S>xu^&LIT(QONYWEAPo<d;r;iZt8Q-hj)xE=fIR4wwz$-J
z8;z%BEsJ=it@k(gYLEVjI@CAwy6Iy#I=`^1Q%Hb5|DPOzw<EV8w<W9nHr(ojrO^Q=
z<fmjA7Fyhq$RB7QfEWwvmCV~dNI}-nMQx^Uj0%HbN;w|fk1hUX=h$*a#}SNcg8wyE
zD;fYL*M-u3bOXekgetI%RsQeRP@wCrTr4WxrQnU2sOs&{o^8ob(fM+c!8P~_d?=@`
zO5aE1m*_3&AN<N<)x-``LMk?^+OM3nUMx%A@kocG(<xt{Qp1ZgNM<PY&;F!YA}1TQ
ztmF^m--F*@b$hj~WN70Uj=r)0mzFO(jYY|EtDMcwCakXib3yzbswIx*er2P8d|CRG
zUFaqLBDCB25rkFy(U*gPF$-^@-5_@$z4~uFb5yHhCxm3kG5I9F`QK{iW3uWp<tLfk
zT$canOqIzP_tA^n!g=HiRo}<8VSE|mR#tzb*BVyZFEa}p{p-1|I+9@ldK#>l*F6Gz
z3ln&V^W?KTaAKWSI=6>AiJlH1q$ol8E%)EfvDj6Z33??P@GDI`{4i`Cx8n4H(A}UC
z4kzO-)*Dg2be<HV(92dTq4iqoAvRgq)XO}MOxd!WFNNX+nH^f#GW+u~KpcoK7D6_{
zk7YsW5GwUH?<p6*;*cvO7+~9pbmI=^sLWc}b`>n;j!|sTWd@7MVNzl6&f0x@7`Z;|
zvX`V|$#s>QfZt>x?>H$z@Vnyl=iZv#MOzF1Zo48Mtguq<cX3~*2>MtVj|N$co(Oc|
z{Ki(3j#3VSc<cbjDL6K*4<$@;KDi;LokJIL<HK;@Y`vz9b4e>i!k_jr_CHglg4WiN
ziPE#by%zryGp|91h9wFT!^IF6{{R@3cMx7om3cdN&g;#@??3Y?r)EG@VO>TUk2Vuu
zh%FZnUe-(=o?D8wlq%{0tn94j9T?B@-0gov7&DqL>v=UWaWpXjcnJse72`3)^ygBY
zekw}pKWb?q=PdQCGHxsF>A^f+P`n7u6nb8y;B2O^T~xQVb|+$LBCsX&Zo1w#?A96G
zZzZzmL)J?@IA99qv@abWGQl|HQcn`{4gqIexOR2&KiNJ$16$6ciIXWS-O4MIgn$+B
zsC@B19w%)VJ{6^@A2>ek6^jtX4>m`sopH_l$qL=ADufCC&Ys_pj(dE;wN|TQG-!o2
zBs=CLk`bNMr4<3&H^`B9j%;u;JToGL0IJ(G?kC99bXI+Z);t}n$_CySjtjlwKdF%Y
z;)mdpJ1Fk;51r(2u0K&t6ca00&$w$7D0Ubj120kqCR$-gi<UPg$@69PzSB|I|E7If
z8God(SpuzG)FRsL3<@=&!eRL`BIp(a)1|7qs__oq>YckrI;G$83^>3Xu%}j8E>E#c
z{`$^&!>Q5cyt%tTxWjqU-`qf69iXSEeIMB^xSBdE5H*m(<IQE%-RJpOfKWN(7z-=R
zboR>VYBY--yDhV;`!G6r5igf8aXW~T@|@OcQW3lL2QP|@{WloUsb67_kjm;LwPRk@
z<<`Sy8vo==9N)>;KlUri<sra>&nyW+6WT!U$O{l0xUHos(Xlr#hhfS?x&ws*l*+AK
zOSEi8#$^B+PUBD<G&xuTkMOz-mpPv;O0C>9ku=U3``9D1)J+g!Ix$EVdZUOJ5{#&o
z?!C6`#M<jDq~sJKNz(%vQ@Pb<(uo!dyqo(*yHPhec+D)?>-RS!V?|d6L(n4Pgx4Po
z+j=6wN&oatYoP)WGt31f=A{1qU5_L<{`H((^VqBWBzRG-q&8v(O^r8<u1_0MbWX2<
z?%O438cX4bKNdEY03P|tfNjUR$n@678TZ~z)?mK*95SBkF_)LI*<w|e{cjpkqZ_RI
zkNeNTuw%kH=lH`e;|m)@`8TF^C=`=$2}TwO{?X4+j)bV~>N0&e70V>AcU8%87?je*
zpCH3><XVMK7++K28vKQnDe*}H<Cx}K147VqqX=nS&x_Vi?mjjn@|-o#@etC#vCXtV
zkW!!<NpnmS(R054@AQ2FLBO3#UqATxwnNiDTKVGV%)W-VM3mtfT0>qc_kgbE3m^E{
zT$2#|t~O@$zT4fnk7F!3wjhs!;YsoKSUJy(K%n@!K~Y`fgV-?p@#-ASmJats-b64R
z76uXeD>^GqMs}q?D(@Q7#3QxL+MTtc?#XBs_PAgod=X3JDE5zarOOR(<~1gJ*jGDg
zD{Pvy)VtTvJ*33_MGd9=<y;Qh!}8>D$O%9zb13b02Kk>DT0D)3y7zGMLkzkl!)3GR
z)L55C(KMZc59WAj9)iaXA-m{RaEPx`iOd|pE{yPM{+E4SJBI8B>hs$<`8-ljA!=0*
zgT<r%x3326P}IZtk6yZRZ*5oas$@TMj{3A_-JcJ0L?TNvw<-9bu8E!K`P5g%MAobC
zEGIvlIgfNl7}MycI1@=-a!93jq8u}cYlm-t*=BFXykzX{H3-L~CjKCA_K(B@|8(RO
z55K@-rK6KR!G4<ttDud-%(3wO2aNSt{`|}h_Xc_Q31_t%Oi$iCK)LANQT4HM@=FO_
z*G3n#ry||izfpF+AP?z-&ZD`6Gq|`GP{qCA2+=vt?sxdV*`RRlE!3Q<nO2FK9iMKZ
zhw0mAOGnrS#BpgZ;c2!B^6RLbE$P{ZjgDSB3j~K&zi$`O<k1^6_vvhk_)(ZIsXp<O
zEp~$>Uu@~pl<tV#AQ^52LP$0fa$!TUKhwR{muF@_#tkX1wD&s8rOjb3BfhOg$fG|T
zQ3L#KKW{KQF*~>VT&a$6F>WvwsE}6H1T5@(AsL-i28)vMcrW1(T-lmjMa}a=skY1J
zK+744s<^mhkD`OyE^rr+9e)K0A9Fd~>o#J%UEeCC(oGuXs)V)Eis7)Rf?3w7JZWbU
zz@Br0nCHZ^Cs+Sd2XfOZj7fv=(Pjt?kG(JosbSvM0VykwDMpt|(j#Pov)$Xt=?7L@
z9UMnHIJ-L>pysRbZ!oo2H7_uCr+I1g29Fu8M!`{9nE*)a!D(nX)e6N?$tSL@@O-iF
zq9E!fF-BLKbZQLQjBkNv<Y|4y*RR51;!qq{&|glaj{&uPQhF-+qqr^#f&Nbp_BeB}
zasJp)l-c%+*A;nc?)Izwa3@4he5c#agoiPpA$eScRP7vL>#o*8AG4Fa=x)_I*V#;O
zp8dGUXEEOe^ZZ=oz(;KA+yzF_Dd!*9tyz@C>e@*O3Uo_kNc#vNkp!|u{)^j`Q1@%0
ztLDJHfMm+gw{IImvi7axN-?bVZo1Nq7%b9}7(X~h^OKdUHDjC^g{E^P+dsBl&6;F!
z!lNyiGb~Qo)L6t4^eo}sP`;8e^!vH=U&JSNtmc*@%QZ~dQEfOZH^0<D+I%+;OfI)9
z?BAj3Qo<M}*J0+WiycT%1ahDL5xStzKv&_$`3w|hqxb4P&l+8!9RPh9^I@wx7YNsx
z4|lOAHqX8f`i`ErWWF4?Ig2?$Gu=<VfUcP1x7R|Vs@B|G2UGqkbw?zuwnuD8AO_hT
zoQ!yq8Q=$ZtCQW!b!H&8w`ZsbwZtL%p3d$yK}K$+GDWTYA5m9B3&sks4S_5=ecG85
z`U<24&%1E`4o!;tCbPTu{2=6_kIVcS=CSP`+qJ5Ur}07e&8!05T|yvaEHbZF2P#v~
z>=`(dLgE8@$GvibaX`-^ak_P22GoiFvvD^M(f~_x8v=JFaJStC(CrVNPA6}?+W7j=
z#1U~l2CgYOFlO5g>P(P!)*yzfBk?!Noz7lFccdDImvQ;kgCZO*P?BEANHJY2^wjkD
z5PHPL>R{3hzZ(4SJB5}oxcnTWzHl~AuR06L>NS1y6;juPX`ENFa5SPcx0%RThk@bl
zdqQu;&?t(`fkws&;%h4STM%JOlN|ZE1Un-Q@Mc)7kdoi_QfyxD@R#2@8p)dOOP$ET
zW>{QwW?_cDqb4b~$J)Dx`Jh!YDy|nv7MKZYtkBPt<VQ;cQ{7Ax6o|45I*bm7-U&Jr
z(QSDK^*$-5rM<AV230EBO@TZg?O<ZsX;?IbDf?ld*FfY3_KQH71ZP0Zn#CPA!k-f%
zXp+kv!oR{?Vyo`$uzoP{?F_77I1?vSqSwnehS%eqT1u*Z+sl--RhziC?Q>cZ8rUKk
zNqB7`HS!<3E!<{MTXrV_8M%1jep>$JE^BT(_Ix+%`)Ow;9a=elB%Vo$J-P11{Z5<5
zG*U|d-aY{Q!86Bx7Uy=nvpdB$R&<yC*`7&CdxsM#{*XQ+KNRpCzgbJ7Uml~|?IsoM
zr+5~LQ)RCMa=nHR413E@sMd(s!as}khaPEL-I?FySj`8UuTPbjKPP0(DKedGvsRtH
ztr`|M*pYy1e@3p5Ze^h5suBbeLrRaYvJK#XG?$H3cu~M%?zI`f4kczG%*+$(tzm^9
zfljq%raA+#@G#d}fYxhS*a5X5Q)}evNqX4>SEuS;)NXH3awu#ah*qOPxR7A~%hjf~
z%OvcO80y5{Ky>Bd<C_6KZ}G8O;Fv|aHet#4Ls1myS4$+MUICM~8Qx5@CQ4B>#jvIK
ztu6;-92iAQ9=kj_fsGNKC8uN;E_sR5|7GKMs`daiFsicCuH1<cQLaB(6oNvYKzg@S
ztIBr(@E;|;?QOt7&VH%vLR|Ar<k6Fgr78|#l<4n!(gx=BhGQp2r$2`eJ(iYXL}*Uk
zYmV*d?Ih^c{k(3q0-~I{OU^>2Cz$YK{Ku*Re3W4o5nquAkAAcD>rDalc6N-UG9MYv
z#|hqEsUIN;rJ-8<PoeA<ddb|5SyO>^CP;7VB(1a-@@NeeAZ(b}tI~%N7qcvSlm9!q
z4jvpDlmc$lWdu-KpC4c8@!5yy_?Kv{2eU_*Z4_Y$KBR!&-H#H<M!KJfk<ieWwnV$p
zLDKQIA$jHL7Hc524|_KXmGO~*V9wCtd+z<lPpsU?Z0tEEjNT8zLR9T3E8b|%RZuB3
zEen{kD@8C!$)l37_lHOcfH|69n<26&=S$ENDgK1Hhylm<Kz`Hm;ZklM=(N`y!O#PG
zc-NryB5>FxAB`P|;f&7^j$+c=*J-90Ku5BZ<tyq!9{T$G^Ir3?T2xMv9lpTCa6NY6
zzhcDzX#$|Z*2AKzhkc7fnF5Z5C=i9{DjPj0FM7@|2!v7}BPt#=XI0pY1OExc+$tHx
zg=@yKhP>mIHx~$sl}9rlAssP;Q)qQ2nTc|PErSydDh?2<dJS7?wvvF)lt|D?%!qFa
zg>rz|{Ju{Aggk(Ur1jBau9+l?9yfZ=42||%ryxvtm7QqokRcqiipt)!UrF%w5CR%^
z*O`mG>Kka;=Hl-bZ7x~lYwef9_?$Q9tXTU7CgZ`?l!C(XOWYi9W=M^2(WjO&;?L{c
z$OEVMS|}86Ta}VfFjvw{ATWotg}<~Y`LLv2DVu~StG{s1Po=Cir!xOu2kyfVpPM&Y
z-TL~ha(D)AJgT~N;AW)D*JF;TD?~<^qb?Y>spjdB>&HdLlDCwvik`jA6ud+<S)vI*
zY#GSk?egTY3Jm5SF&QxRrDFZH>nGjLz$QM$WGHEqbAf}eD*GMsOZ3MsaAexb$hmt5
z5ku4AUD&EuJ~lFw+0gI5739|uBZHS=JDW2mp=kQ{*ziwM?UVMs)8}|evsu2znwY}P
zxgr{qBwGaqyVssw>c6~XAVR(9P_K0(4~}O_*9(2wKj4eOPTar51!mMRO=cU;lq(nQ
z-S_|Inf{;-M*i73lCNj5PxdUgUa-Qic0WZch>|(1=|8PzYZdcnGGAHdg?whW{4m0V
zn|wD6-DZMpXAL)&0LyAjH?S_!dFQo9J?*O&stJ}X#VcyC!-j-lu0)hDt1kf>Z=sa_
zL!jK2N)3ZVvpJZ|4Y2N?<$E&48t>0sfu6OD)cLMf+)UixB=x=|vR(vP<sskXzcAiP
zQDCQ+hvNhK6~la9PHPQXh`q#(sMbdDo+zpZv*=7>0AfI$zq1x3MBUy$e$3}+%wbof
zv!9Y}#uu;RtdgrPk!(wEEMsUh?AcXTgxzBKGI7jSxsD{WS}Rf$w%z_Fem&%9Fcjt_
zW0Q)CopC6iUtszeZGWwB20DS-uP278gw_XY+>2nvgZm*oqj5F(KdwCQQ6l1b+0LC;
z`iMS747j?V9EcF^I5@?u3kua(qV^GQe>b+$>AKs2kfvoxuE73Ox)*J8BiHxeGlh+O
zg`Bioa&Ny2*zFHg?G)`Qf)1^2uePyr4zUwD6*z<K{iTQY5mdgvnjj*aOBZY2-i*Y4
zA`~azl_#l5YW8=Jes@2YK;OB9(B=*+cD++YUVY4xK|;SI^6M>iC`W^hXSd+Zru}Xy
zr=rXGB{rRfRqGAg7glO)ZV)mq%X8acdnWh_!@^5Dc7qDDkeLpzqAkb9Z+^kOthQNt
zQ8nuxGlOqBe=|3aK_=g7UPPG!+b&NPC48nEJIc`SDCC+~HPE^Y;R7AoH8!q`Q2<P)
z1b-p*EYuS%w*7yJn|PH9Wo~41baG{3Z3<;>WN%_>3NbS=G$1e_Z(?c+JUj|7Ol59o
zbZ9XkH#0E`FHB`_XLM*XATu#JGc*b>Ol59obZ9dmFbXeBWo~D5Xdp8(G%_(DARr(h
z3NJ=!Y;<LEATLI2VRU6gWn*t-WePq%3UhRFWnpa!c-nlp1yCj3vNZ^eyK5r{8h4k*
z-CY}};Q$ABcXxM(#@*fB-QBfu_y4>1&YL%LCt{)^Dt4~Sm07vAR8*0ZD5}s48Qc5@
zirZK_(lawK{RF7kIN2KkMQn_LiuN}2QjUOsP<U>DG04af@Ed3fvW92;SFpUH6%fFv
zC8nezB2N4N;TQu={u3r;ZDIr9`VZUK$@YITI0Nk+KsMF@s(+GG1MGo@#x~ZLZvUal
z+c<)ZfB>rhBn5yR0EPfZdqZQOm7%=_z{Ujd?*|5`xH&iitsDSS)<!n=wl?;LjzD7o
z$Qs~i1_Y>CgZ^cyIQ|3vL$@|&w6O=+{Nu3)7}+>kJKBSQ4h;WYSH#BF%^qZG<_Mtr
z-ysVVGZWpv?kxX5Wf=fMmX-kJe-Ru2%0LI8z4Jfdf9DrBv;tYW{ad&H>Fr;N0sfKf
zze7O|;viR`u_DOP$n2lqoq+#=r5p_{K}JH>rj|eeJu~xvKs9S)puHu?8mMUF0CN17
zsPxQCO#kImH3J!0SOXm#0PO#90j-VytESjL4H<*1P5-5uwXvbS@&66{+pTB_vUXH;
zvjqZ}{&#l&Z8QI`Eob<Tw5|YcCI%)Z=6?eJx&CLQ`(Nc$9PMo^fEplUN3;KDMqyza
zSAYjS3l|H3o{gOiz{Si4U}oZA^7?<%jhyW5f!2=yURM9K@;|l-=pSW(Kv$p<{OXd8
z5l^6bYDSu44NlI`gs5f<36vPC%V5Wdd8pkfh(<Hux84#ZsWkkZ8GS5*aP~c3A%(O1
ztBmJHTbk(pZ<*L25_szBJtKI|Qa5%$1qoWVfMN&yi<#YFNgQ#ri`l2(HO%1S8H_rU
z0exN6w>mnkC#4STX+nt}UNZx6`0ksVkeh8^_F{lwY>G^RW+Cjij*EdUzQ5f}n%TpJ
z-UXI^5a4Y65`gk9E|pRk)RS+S{qA`OVHEGsF9yu6xZM-N*lPIE-P4?S@cI{O`jJ0*
zJCTfUUdxW46FF;9+I*k>6gt)33d+k=Rb%`mW6e@ZW$~vNKyIY#vB?70BZ6z(Qcg(9
zHX#-bKDSk2J>aXTi-iq~7>XWG0q?7;t1g*%vVuKsVauNQm17pA_Cn=Yh4W*F+IbN`
zgnQ4Viy@Qax6t*+@?^|tQZGtG0cS0}w~BFzg-|M7NFbfDLal}YB-Jd^3m|yHEP#JL
zxUBI-rjfE{{%8Ub11n%Vv#%4J-@;Z$rZSmGpwMW^Kbe}RxwRLH1(fb$ad^$>L2AOx
zU+OoJG(2Q=*9{wgjzP~-G=jc746GMEg+=q-xn*M9VfwGXirp>R!3*Tmd#`T4)R2PN
z^B$841J%Fyf(BZJy&uX-&ZNYb#sX9=u;{QQn>aWa_0guCiP^ePq`+i`OR~WYIGHD$
z8qcax24;(&l#4xkhzf-DvLTixtVOTvYkq7@^<3Gm-cwP8Y9y5EdR1#-$6rGrMeBy7
z6r_{g&^@bZw4lIw2ctFUA{Ir<2p*>fbadFsC>|x=`EcnKm)^aZswHF2ccy48pVJI>
zKsD(|6&=0&Y_5q+rz@#vwZZ!_B@scVZ#Iyno3vDjSQJpo$;kbz^NDg)p65nLlCyj>
z0vnx4bN^ASMGY~s9i{!2!12HlsK4(L3<TJob^eTf@Sq%x1+v~_4ox5s2ydj%IVE1#
zX`=i(^-uLT5V4ZAIH=al(K&PmDE1>P)3dmUwcx+uK`a#x+mmuSvlocaDemKI=}@(g
z8}b1nnci<3x}B4k8tSz<@sCS6C)T7`*xnL$)sUa(Hej{rrxazwFjHvQfRw*H5**p$
z#PIQqBb6a}P04F8{}LyuFe{ny`IvgPUJol~!nOs3*jBmkAa`rqco#4+;bvwuFl!#_
zVainwe2>70K(xF-J_>d8J$zED+-b|r^U`&DmXCQX1k42xcF5O`f%{>V<$ld6c}$v5
z7N>tY&N6nZTBu8@LRju?H9Yv5C250D3AZamGTmzB3~jqCnb6l`VqG^YcfIRV#sNPa
zp{4ij_^*S%l`S}h(zQ|-{)waZEOR#`%XZcwp#cb*QZXMZq<dyrS_hY9@xrbqAC0j}
zk#aGc(^rmG9tZxO{etMwzPEqb)O(p}aA18{Q#&ZA5FvA_hc%!%y)jyX*1bx+HFsS+
zZUniQxk%J6Fp#l_>%2?iRlD^!Mj%`sXpVTPC#eJz|KV0NX;j@Bh|c*WK`Z@G*LLL6
zyV3I%3clDw#|cj!D8JHMxV4wmEUp#n3QE~4oqYbS`3D8<I{~QyU%KJ0DupQl#BYWI
zSdBjf$|Y&340e21WJ>Q=RB9wAGv4<NSjkG7X-t$@Snt0JI28Y;-%8{siJ3-LM%88#
zah@y;lzA`AEI}-<1cJiZvwt82Y5j_vxat9W7gwQubqaWU9i_Y`N<wPxE8qGoUc8#a
zpav`Ic!G{Nd3f;7xs?+~!TF|fJcnjFX(jFk!6148Q`N(*!UVOSCVzS}c<EdA?j*U|
zqh$iQCCpI9>2kqO50s^B4#+Z=NCct7(+hfg9`ZISTL%*x!wR9YUv`FIef6JMI4eX^
zLNGdWlWQTyrrF+h_s+}LOL$Bt$*U(~#IN$$v`9CEa;}U@dAjaF)U^lxw32+Zjjh$<
zSp<vK#!ztecvMLP^J3A$NjX8uOY2`27b|5nCg6{O`qB-RMMj-BV;BrnHf1(jz4Sf<
z+llj2bX?ScnZ6XY{FvEV>M~Wu9IpE|t2hXMKzd*4zFsEpp!Mmh2ma?RJinoIpN&&k
z0VMu|Hh}VuxEW=}T>qWdOOlmGXpuD7#gf)|o|QS5k`sJm=V|6591*bl`exv4PIf(<
ztJK*3X}8o$CB5G4Li!?kd}h8|VbASLop5*vGDE$VijPjc0v|7uw-#ybW!YjqpYG2g
zg$BX^z5*Ctcto*2-pzdEiNd&UX~=hdGJGOCwtPKvjjC%4Am13UGebnd6LFYsKb1dm
z4O9Ax8*!n+*k(jDh<o?#nZVrGy7Sr)z7>kBQ6=a(>GeXXLi`A{0Z5x&_)2WzAc+&U
zd#G~yZjVDCQ1u>c;ewMM@wNv6NB*{O0u#VTRMjCRGG4_S3gs;n{TqUB0^^hxUh|E=
z(JeTupyxE)2|ZS4<;^#+PyfQzd+WIXl)b8K?)W_ilc>)Cw@jlJ*~=k>kEO4Eg}fBM
zO)^>CgYx0vt5y@LPWpFlztTY^loUFvFyt#tv#oJG?WuVJ@?7j8uu>ld`OC}X0^eum
zFj=o8=#k6>U)4(AiD$FbCHIDmG_0sLbnKT~DAzb^?d;vCUF_>6)t!tbF)9-32jerQ
z8v(~e_9Sivnf!*_qXcg|FrfO<r)R$e!}<E7cDPkF<+aHTCy{``6CgYmON)Xj6gesJ
zV9exxiG@rFaL-riNHXHXPYX>Y2ciB%j-woZSLLvu0CyJBb2sjHM`h=>7gGUe<XLk>
zhEnT%k4b^_QT84Y0m@S!F~x{MCJ}DS9QW127Kw`}zi-c`K5<Pn^B8^8Z3%wcO+aG`
z?S8G$s6FLAPTy>1kxg{fU>|L+jhhO)uFH7u3_-xi(y?73Bsc601|T<UnRveGQ&udR
zP#S+bqb87VHN+QQz|c)47M@b!W<T-ai6mO5)zi%Abc~>sUJH$Je&xV+%pmSDZ;k+b
z%?^p+=D;w}X49t5s4nUAjSTmlzxQ*EI_UkFyn3ihvD=#sXyL9%*kC48PP+lEsi)W<
z(JMrfr5Qa9)*Il+@2!nlAbMq|#n0B)ti96EEZAK@YsXxULcr@f<1cC3-4rPuWn%1d
zwoK^^_*MKJte|N3_HQk$!LWllIOAV@x!)7P1G``wLM0R+qzV<LnWZK6644whA%x>1
zslT6o6IpaIo#7>(U>SJN*#e%+A1wjfwZ%?q%rD3(8-$A+oB-))yt9<!_s0azQ88>6
z7J74?0jclxPd*iEHD4vRO3q%dIg?dk(Xq`>W_tY$a!F>ArVEf-?4MV3H^m2PsyF4J
z;HpybeH2q*R|8~`K1GD*k^M912E#2sw+pFETJ=Eu0qTf)>i&_zG-qCzxbs26-k_3h
z&Kumx{#addIVpLAtdpNto7R>8tIyu6x8&<I$Y@%ir2Pl61=sZ3$|?{rhGV{^=guWm
zAAyG@ykbe5jd!22kg#MV^=1R_@K$kVxRd&l6mqjN$_1_tuNXrs4E3GJ;BV7zBuK$H
z^91R8lkamj(&R+sQeTBnvy&FRWSvygpw#D3HW5Vy%u$+AAntw^{2lqKFRRq9VIB0$
zRYiV>q-svJgxygZA-`<>Ed)(yKCZ2u#3_d^%*tX%4<>^}B2=*n*$WTPj?CMGGe^Rl
zbbOXLqlPC|HMX2o1#@Sx_`&RmjK_QlNG}B$KAAiIkbtAKDE>#?`6PdhDPGWY&GtQb
zdTv~a#msi(n1tJr4JCU0O{U(|&zX6aL_Mb%3#!qS(z6R1T=>rhkC2j^?6$!7lV)gZ
zE-G&uPv2>$Qwx{xxj%I+x4vIN^m^*hMs}>5OyS~#)h$BrPnL<&d@1a|6GvB<C(N$-
zKA=AQEwC>;-0+!?v(KvZ*0bR82xzp%828k&@U&wN?+9qwPg?WUZDDH-gv$Cs=P;&M
zhnae}=<TAwp-AOSx?ZJrEz~gcMP`10Uu!*v&n64V_K#jM7s}Q_op}I%Z6t3HiC0GX
zlLt~g7+@TM@E<qfJ-opcFlpwLtyE7Zf!vN3S;KPh25nyoRVPy%<IIeYLpCQsLWOFF
z%o`(=+2t22un(RD#F5C@{7%5y<T0ptXPhXs*z|}_m=D(?f{4xN_8wvkA(SJ93F&6Y
zza84L@2G)NA^_|SmPsZ3Vl)iS5FDSSJEj6Qgu~02Y50Q7s4=aG(37fNZ-(Lh@5P(Q
zwf@HabXROaOBA{n^H~;^uvZY_<U>i^{X@xCW^Tw@V)=!qh)Ks>Sp%zop!tlUmz5f7
zzB}tXj?h+7l>8tWOY&jv1oANm(3o5iqG1%Dm2iSV8R<u(jo-VKHpspT%WZ_)3+-&}
zMx3<kM;j+S?PlZk<tdWyZWK<ecMr^jSw06X9twfAvqklNBm23~p!ldVEI7YKu%^`;
zAM-2S_>QUz;tin~?X&_~4oZ?l5^!B%M6IU44d+;gJ2$+zA@QqPi4etE-7IkZy&X(k
zf(vx>jH`FufwEwgTsm|vg`%Wmi1mFHJdDBYlVYfKmtyR-!UTRUP`MVzjrmcc6YF?H
z0Vd5($8c(3P{{2l;~M=>&j)pGf3(3M5)$&;X0^Ln_0D-k^LYbjK?RO{s118AUf7Dz
z?<nWK&>Ze@%Y~MC{R>c2+}nI10hB*&H6C0<E%64}q|>aU%NkcNor#%n0+=*tLc?a~
zS2AVtde8{evqO{SVZ9{@A*cznEy7!Qm*%xdqa6t^*6TBVGQKQ1()CohTV&^z(h`S<
zOHrCI#hS411chP|B24N_r43zzUJij%oqWA#dZ2qn0wyg?_M<3j8%_ti_l!<X-(jz4
zSXxmBlbLCcR{>u3lkBmpE=vT0%gJ(D)NHrl1PXaYjXa+SM20KA3#VF`tF&MLNIOuW
z2eIO{p4n~J+2<-&U)naey@6y}$KWw+x*Q{pM}gnpY7jxnYnb(xV1?4jz-=dw`-?bL
zk1jv_l*W|gi}A+9XZ#5dk8WdSLQ)H1)jL|3>Nlc!oFk{Kz%!Be7_`(<I0?i&iiU}Q
zm13Vhl&FkJvZ@G3+n3HNj}|tD|5>2>7_lT%>7_4Ce8`igr&A@kXEj0lnjCE&=2wXs
zIU8#DDkp|>UK^#xq1$I7edB0js~10Sq!moG32@yi#Uw`OkkM-#+c!N4-dXt1Lwz<Q
z;WYJ!Dg8H$3zKGXS+4dv>kIiAiy9+@Cg@t{KfF|xdxc$?FQ(ZHqvYb9V?>3pu>MyK
z+SFrzRyAl-QX`Rf7^C*mTfU5oE_A*17YT_;42|E^jtP;Ee2Aic#b-~SL?4lGV!>Y(
zkEp+QML}D}YD9n0RfQ8PW@dYw=a8kpM3oG=mD%vw5QJztVEl0HKwm^$IIUXsB~08B
zp}5-t=af;@g)V8iG!X)AG#FRJ&d_t53`*j4f%kCPo)j)-aeU_9a9+ZTYBjy$wGmT$
zY7fjz86H&-kxIPObdH?4TR!f4P*iLm1?k3U*Q{M$W^iU(M}lyrB=}I&EgXLikGc<A
zMpV&?;bah-O@6_@j<AQeT2qX1L3(cJoxNJIYAt}@+(x5){9Z>UH|>++asNh!9I{v~
z-%i9_fHQ<;=fj{L*hZ=|Z&Ch8IhdtJd^EtDJ^~-sw%>nU4EzdMm@kIUfB_5Xmms(s
zYTUf0as~WkWntjrqK>aO+k_ReUVw&L!(qVanNSjT_qr8-!(g4YjAnKU2_Qf7*;gfN
zPD^MZPDhVc>{4ZrH;$Y4>my8||8fB`#~#wez@E}~Ax8QWf-0%RzsKJd4(ra1D)+~R
zDx|xYK>UD1a2aQjbpXpAHnJAaoBNJPC->Xn2u{v`of?~?{qlqLK=R7Rnsx%yp8(%e
zQi)BSq6HiiAI}7yhj5jQtux2Sc1ppI_qc;hco}`mVUrtgD(Cv{?96jzq+aS3yRv33
z<iZS)+_Y(?ukGE^P}4y#%!%ptMy(c1m(wa;@ne-4UKC$pBR!Mers13Xw7(8?2x1b_
z0yf?Qf0{w!B_l!-tOz+}-eTvQQu<m`;QJ-n&AVEe3-WShs>pGr3;Y7N5UpbOWh{Hx
z+FkH0Zzn&Gh#;MPvPwSp_TiyW_e(4C?V;H}i@t||IXj{UYQK>>uBh&Jh3q2#(Pv9#
z?*gKGpk(lG!@t6@&iry_3Z!m3qNh|t?uhJ@X++O2T9S#F#2mGXL=R-FQ-LHh=F9%f
zt~D1?dCMCdyMuLtRew@|>%n@Nj3qL<A({HJy-b5lfnxJx*`9eY27hrJKnv%k`3W3G
zL)89zE6(wJnZm#Ed5K!49pWRnOF|w+Sz)vhPQ=QZpLRE#_K?8IuBK>1Yheq(Y<Ox$
zCUX`0VGOE89n3`vTz*fzNsW9XQ1EI4CoWK0rw;#_XrqSR(?r%UNQoCZ#&JrbasJow
z$KUx3dMm*)b4HD1J=kk3+F(<y!j1Oxj-R9>AvZkt1r;+qCAGf8mMV!+N~IM&LMw5(
z&jwW<7>Z%Tw$>VU^0gGGc&A57$fwh$l_Ecd%L#ZBwPhil6&^`6ow2ETBGog~Wv(HA
zp&I$YzA43#Cs*-8i7qVOV>9a<zzsiE%V;vregOiYTh@39y!_Z@XK<-^R3h+&G1RK;
z$wMKguFcIun-mM&C5pVq)I7eV)=$D1n{~miK-rn3nx7OhG3vTw`Kw{{a4<nfkF*&I
zAH}TWD<`ham+_|wvgBs<>U+y?bnN8XwvpR>P0SMH`59)gO(Q6YV|Wo*1$_P_mNPWm
z{XXZ$9`%I#nVjaumL|lz^hHXX^r>`4XP$<1jTjESm34N0gCz1{oUFE98f&n$0|?fP
zWIJ_-hL}+y#*E<m_Ni9qR9^Ul;WmbhnMP)#(|QK=Yq6a_cmadvcu6NpZ*UtZGcw}e
zUs<V1*gD0q8^F?6&Sv;NK7-AAzO-*W4l?u6(@*PqzBje{1%S%KBzu*_3xYaX;JUM{
zax1uY*{Hnpu4$tN3rGg+lt@|RXAh_iJBhxL%88^G`vnL5(9%+%>pkN$)~Y1E51kVQ
zf2)MgO${|W#z3imxj=z7M88N%q-EE;oSNr3iL#ax%52N-GzYH$#v$x$t}YgpTp}Em
z-Sqayg)1hH<n<v(*}rAzm9omZD05doHXdb}wA>IOxtdK&{VWBW5jsX-K55J;xV<4O
zC%&rJR=!qEF81EGnxRJ9LM~}4-9sJV$Dz#s>zim<zYL$#8&2c&NI8Q0y>GE|om_H=
zHMqVts9s*9zl^Xt`b+W8)Rrjq*=-8St_7bYtE*uxmCHc-CCk@vk|q8e#q>@H9b-jU
zWRQ{xMwLFhTU^rjDYUwE@(g5vi43c{EE-Q}XOStSDUl-D>>JpcUKXc>G=i`o-whXt
zK~|b3A;tJ}(0E1-mWV?MSv9twz*VOcWzxd!r@uz^v}<D}T!HRDUaTv{YRq_{+a-<9
z1SNy480>f1VV)J60QkpN3#!?Cb0Z@j==S|_RWv9|{rtqE_2vr}MK<7v6;pMw^t$4C
z4h%(q&SXTuLS_1{Gu|PPnsdVrA*?vr%uH%kehUiu#2TVqb432m1A=b@F!z2`*O=EM
z_$>@A1zt&Vn26FvoE)up$4vJ4x6lKL_XCeRaUkWjun`}Pfn&JEdNs^<Xx@Zf={BCi
zkEfXNbd?D+20(y(|1lL&(Qr!H{V|*jS7KRL`!&&&u|;?Sc1SRl4DFkY>rq{abI1f5
zPXhd~9Q6sOd%{?!!xdN_N_o=|_kcYGWu5d%y*AAIjSo~TaxxOS)kymbp&+qeT;gN@
z8irG@F*Hfk%|+fbRUK(N*Y_w7<xA_OR!iy9;P0Pnu7!CccV1#AcWNOnBdp<@;3$p7
zkyeKSmNTuu_*pp2q!cYUs`VophR9CUNEx(AeG40_$X_c%&p~IT=ApGG`AD>*J1dD*
zk>0v1Dh(-GmX1o-2*2)vJl-0<NCUdw_r5Dgii3`wfXuGjEIvRey~WV*ocd{dmQvWa
z4md_}rfnYiGAsC>y<uZaH4e_wXpIWj3L8NZ>#b1lL~q|&nf!S+u3sr>`MTd>V@fF4
z;6+YnuHww`-)E<nG^Ft}xlW16&zWJHb8?|J`O=t4uo%2Oe(!u;%%cezGP1Bkg9wFj
z6-hVXUSJ%gMol299{u|Z^xZxj=nL6z7}r;`yU;nz@3@OTw2YTx`DC-kst|uZQBlOd
zGS3A_VK%9)28`UpBg?@3GvWO#=f=&!X5Ey6NCMrhhmclIP!MhnVv;nlGS8RTVmfCG
zXMG2ZH=lJ$d30wk$kE47Qqc=;=&<k3+#I)rd$rCf>J>?)3E4*~YeP?EvlP4;r1a{>
zaeOeq$3`*F8o{v}Q)(LJsB3IH$0e;7ht;6r(1I-uE>xQw9(oh7YR9&1eZJ4QT!0(D
zN!dtrVw)|g2%=(oB(gn^mu8uerHRk<LxKBffcg2W!;ou@>5es3hw?5fx|8e05c_cB
z{G0j(|2j1b#*^Hh>0dCH;>PPi&yq_AGV3LBr^%@}Vf&K1s+365N5l%53;j{ISpu|n
zNG%K=Le&Sls>VGqXX4*5%v3Piy&pAR^Tg&@GODqgD=HRfRasM&A?5Ic*C<oDFC}48
z^+|;;uaE}Tn0`_96<$@>c_;Fi)?7@uqVBd|!2)lvf=jNWtsNJ1lCJ8UN~--8P&{f1
z=#h)`TYbwL2>OZP%)BTU*WLuvpSyiAr80EfvtFLWBkIpGuRUq5c1@d?EKAi*s?PkO
z8e{G2q_E8qc&pf;LE43k)We;Dy&Wrkmxrm7m(+zLUz64L9SKJ8S49uSP|bQAISLzv
zNgIwWN=8t8POfa8UNpfK#O6fB=i~%r3x)%VLyBvA&sTFTfv#YW^9(M*&BaexbN^0u
z65XhPd{<S0$heGLGo;5xCqN6cqtJNQ+n8jtJ6X{{|C?u@$8!LTaCg`21mDyN{_Fvs
zsgO1CNx$Ubu6<uX!MKBnJutd=X^TN+5dDjAR4RU6zxc{P(diP6%ocwG^Sz?B$fU6S
zuiVph2DpVxCey_CXn#wahbO0Zp0(<JB2<d;l5xGc6Xgyf62P01aB%F%D?70KtLcVk
zOcYG(WjDY2y$Pf3@9XBR;%Hd8(QOF7SM^afF?Md)A&*+8e-p(CiX!H#ajS`%hX2~`
zN3WI$8Sk9tA8?CXS#Wob<Q;E^bXFF5mLx;_wHv+d^*AFd3)r$SI%YTOWasUnCbm{U
z@I~ed-;QVPLeE6;Z0!azrv-9lx+kx*8w>Et$;776<J;&%b9<8K=C+`Ah*zUe(Dh$;
z@D!+XJAs5kYD$D(#rUok&|&%ubz9$R6Zzn9hhn-<Mf!B$#I{=ho*_t47j3P5cl0Y8
zr7L+~_e+sr(M%|F2%V;@nXikw?tp`QWLFk-(+>)}2DhEr3N5#h@-!xHj{5v)v1bg}
zGV3JFEYo{c@wx+_c5<W3-Wvb*^ZcfjVaR)c)M3v`Z^On(HA&yt<SHa$2=d~!XLZsa
zudaJOh~CWJ#Nm}pHml||9IjcSL=fk#YEa`I?upvWMZlFaG7oYh)aX^@;$BV5?4m&N
zDja;ho7;;vm63dr;;D^R`*8S}NB#6}LW2nrk4t-#pq^T3obC1-YR2DEFQ687PoJ6E
zWFyC8d0>>P4Kt4k@>`C}SwnrW=a9^x|4w{jfz;r6NNRq-F%@^Iuet5Xh2y2So&@ut
zzhKF<4Zh)2@oPM#Dq!`AsWwQr+@&-CE&7M~(+_P0ti(D`rc+z;(cA2UjvVJT7W~+|
zYxJX~gTYlzl{VQ@-Byg(?Tl~OSj?(eD{TbzIS>^;yAP$6g^QEP#skH-BNnsqy3Af#
zsKwR-+HemZ?EP|b3pPxSZ-bEwl)9-;u8?*ODA!#u#5pdQH7@V=QYyNb!)=rH%Ow@v
zTdIfhh_VIlUZEEBD7X19cryqdB$+j%7~IgouF3KHscWi{n?vQIPWz>{ku}Sk&wa_*
zk17*#lgT4UUC*;y`iYs5b>Fh3e-ADT*|}re#%}mUZ>FzzxF=Tw=wk<_D`>e}_o0`W
zmWjS9zLnx{Wno6jj>J4FRRO}2CA<-+lsl~;mwjrU7g7_=9pV-G5K+Be<mKZh6lO}2
zmuHMD4NHVR6}qq-U~P3yKQ;jwwy1K8w#IQJSN-6xS!1tKPy%xazB^@$sXciK5z;@@
z(lt&y`1ur#1UDIwa;2m=#y8p$74ti-QUrb7)B9e@F4II!!!yKr-P`k1MJ3AAyS;*S
zP|iLj87du4!=K|&)Eo{D7mzaWS=hUnKq764@iUPPif<XLB=->hKB?prmPTPrNTcd>
z697zcc|rPiA2qJdy`Ub!+vq|^DR{?5IYA@EhxgPdoD7CP2J}#!zuP8KhH3<g2_e_E
zbeV3^sg}8R1g~Kt9bFQ-SuY|Jxn@NKZ0tn>i~P7G0Gbk~O4L;SmI=ULanjquHa7{*
zI>SSp5I~JV2^5_iq=UG>2cM;D@^lP{{Ruybb-23+CmJGovhDn+q#;SOG>d#)rr4)7
zSbB!A+nI1L&PC3dp$@ZGyPz=3rqzi}l|8-1MLQ|7W%K_uA8#ezM0@HPpXG#5hJf0w
z8CX-z`j$d$xjj9E9PWByRc}9V`mcRD^w~3J<CZHACBH`bqmvEah&|A#(qO=gOCwb(
zi#NE5Vy!>TrjjHYA6gHODiW=D2uk)!Ixs3xDl<OVe~xI3l(o|1*krdx#Lql&o%eHu
ziI_V{C@p)q4k8Yno><_%M;X^IMosL(jwaw3pH7&I9ai=IY9ryge98ZPZKr^@M?QND
z=0kDxt%^qO<-z(`mr?;Beg1pBA}Lp;?oXueIFcIqtDKA~+JrsGxf1}38Z+0`jOuVr
zi&ct*ma>o<&wrT#oFvru+^-7KXY%3o$=sLTyw#B%Yu;!*!<<<Sk(`|x;M{eZzhYL+
z3X9W4Q|wHt7!v<8RS~O6{r$W(;^8NQzfS!Xji)n8eN`>y)N9O+$OL)=O${r){`YR}
zfU^*u2cg(f!dTe^k1;(AuIs0by21d=UksmHSMEpO8jK=>963EWomLrB@!hG!Xx8e4
zg5yy)_p#k9@#yZY=7sywE7if%K#A343<`eQDoQBnk1(mhOP--~b^UZwlJ`9Y+0Ldi
zRJv8jcUj6+Al}yVrZ>NS>=*b2$L(BZpFFZDbt&w1L@&med`Fe-I8iU69{?hCr!&qr
zg~5G1<wb2I^$8J~uvjsy5v3^K6X~v5#a=_Y%<HG$nnpPa^-sc#qr@b@Hlt*fZQY=s
z{&iJusb|e{X7)K^7MC>c;o;rqrvgJp!M5s2gvF#AvJ5KSTUt~k2nFQT-TjlRlvjk&
z!8jv{K7uv9AN_xVY0Rj?2cjkL8^(U2NVH)t<}>G}xwS)KRN=^-NhuG+IXJ-BnGwJL
zNd(JLbHGC6`^lpvyl|J2Pq7n{c@WImhXzlr97!f+K(?+PdUwIa87kE43V9z_j_D|!
zph=)D--Dj`eTb?E;I8%4IM;G7NIeZA7I)k<-zsJ;GbHl_ja`LN9QR4TcgXIQ;3d-F
zIHkAUwlpE}-P4Ae8^hl9K|MSw8_5!<bm8#2rPVcdJI;uvmGJ=66VBJNIZ;-Xu9r`r
z^0d8~2m{KlMMvnUxU}6nLTi=kE~h1MKQuVYKG4(vu{b)Fqa6ETN#ggku6ay_57Eed
zT(u5o&0tO>eT(%_#sNV${tL#3Z^`zkt?tUB!A&R%-Y4>9l_T?OMjd_c%S0q@`Ir<f
z<g$qv>`-()gOBYWyC!R0%uiTS8+fdzgeBS6zE6pz`2tmaPB>7_#1FoyzN(eL4cpMB
zeft)3<z+;Ao<#AZjSkGXiuc}UV$JgM-^)ZMI-&*O&?F$Zhu@>}--`J(kPNNl=WVa@
zf3Qdes(mgpRC*(}XG7o*6KEqW2-zi;LR`U&WYly76+vtz>ZAI&R8Qgj&o_V>4;HRc
z*W>X6*rMCD@)Siu#%+Rn{^||qFP&*~T?BxpNn}xj4BTIwh8g2LmQm*y(O~knx$MWM
zT&vme4_7-3K&EV7=}cWTGCC><o7i!h!TVboh9>2@uusgNm`&vAo?w^Xlxd3bC;Lkg
zRIy!P?@K8SjAyWi8YBmBE+sZnAN77Y(fQc4K9jnM0I>b|QLHgh3*2rnXzIT{RbQ#7
zPxGQo?g}RC7hEB(L}GS7dA?|?z`nzT=_63OVpwkvI<PkXg;qdCogh54hZ;4;Id3Sh
zqw3l*emSqvmyASVWVagQ$0rP3%`R#1fr+;{B{xxeCRUAr9rfddu)Lt$RSb{|)@R(2
zYZ+*@5)?U1vw&%s=R8xVlkuGbx#$-4$hZY~$gg<T*5j@lAVCwBPRb;J+<(QplW{hX
zxP#@-Ub4YQ(+{hMm!5DxNNgZhwgq5**q4P#NcXOkx555){BhlU9hqY`V=ETa3L+;q
z?dn)y5uanKsC%oKZOWq@nQe$?S!wj3s>YY%w8EN@vs}zbGqKd?OTAv{5Y6k-yz+xL
z<Xb}&a$yl0g$S%BgL$&*XMwX`K|i0QbFx%^o-ruTbP6A%T_?o;?aDi@_0<MjFs&3D
z&O+jj2haVR#kbSLAOt8P@^<ed^_jLNr_aZ@@-nCg@|eiNG!2cRaHAtdD98-r?e6V0
zrurlB_$7gt-7)puXEpw9mNitAFOqffFD5$v&wHJ9M;N$iR*sS+1f_mkg_cYKReK0<
z*zwALw*A7H+FdES%}5_gf#8p1c|ykQZPD+h{mnm%{2cwVF+W4*R+SMY4!Q!_L|Gn@
zooj6nvp&5O?P#{p@I+c1ryJ@&F<PrSMUv)G)8|9paBPmC9#?U;{yMTA(p$~7Y@n`R
zuot2pNic0~OCBzvT66p2;q8Hi!Bvljm?oL26HASmjAL&n!16pc#U_5>1XtnT@JP|A
zEFuz-<*)Z7M`R`8-wHeiO&HLH(&f>Q842!Dt!AN50LxszA2JfXe{e<oxc^1UIHZ~o
zu9&euP&qMK%rQ-7Or@X`&*AE7=eA@k$bWl^iyMdI3jP{}E#}{c^p1>wA2XrFop^z}
zitsAdOo_E!LRGD|d2G5eAUp14j^3mWfqt3~gZh2WS!<`;F4nMJe4(lpVkhC2!0M&B
z1Dm;xDDeA{_gMlJT~HMB(j0?c!b}wVKsX^@FF4O-|F7``1Svl3vL)4kb+7XpGQvZ*
zZm?0$C#PMd=q69fR~@f*p?5iPVmfbj*+&=qqY1{QIPsv7%I`I^Y#P2EuLudbe2{vF
z@);?UhaV)1xgsqOxr*<;`2bXZf9DwPQ)A-pT&fEmQb#;x`^uBwATgu4K)eKs{w(!;
zeX2(Wql?XyZ_EXT@o%OkV`(mYY?Y!@+O@?<?>Vl!Zaycm)dZaxYQI|-!*q?_L1zSi
zM41kQz^N28GbS|GlbF2=j$N&s5aE=gw!aU<=lH=2jwWXrn$w&NWzlItwoiNA42Q9z
zNCh8BzHIoTdRV85BuoLb^fttV`JmDykD_i~;J%;ZsA9s3BWM0TiY%U%R>a(Y-y@D9
zj<MiQ2_f;A)*J44WbMc@XPpm!(-!@UM^ol#DZj}n@JB)}!+PF?Vw7LAoa)zhO9f30
z((M~OEuI>WF^gJx_bNAXzfR7S;a7p2vQS4H=I2fx9U0<^Y36Mc*w`*>u8%sC|6mW|
zGvG(7tR9>?l9{TGG^jz+c+#LEBu9hN7q~ji=Tlx%$_)(_gu$Fl7)1vksju|+j6!Xl
zPI&xpW^t&kHLx7QEIlqmr+=xfW(}OSDz$xNE=Iw(1})p{BU(^Ajx0=JW_K_z(2bvp
zk!L^R3i6c&yYLCblp7~i`tj%`rF86tq$?k+H^0Z5s6b9(Jmh&$KD$<WZ<o6>=7j+J
zxJf_)d3pP3F|dZWt*=f+F;!Bc4n7i|7fan+y0O3Me(2iM0Bz^pwDL)cyXRd1z1-+H
zB7`L|9>Jui(x7A!Xd3Dtm9p`1<oaxvhwy77PL=uxCp2c5yV8j7!d2c$rr&#a)w;i%
z#)tJuCvKIHU)_cxjWAKijSf9(k5fsX7!>sq>RGr`-6&mGxYAd`nd!7G2Fq!&bt{T9
zLIkEq)$OH5h_r*4t+s8a3!@OL$1*>E#`Ka}ab$q$-~5o-#GX8*Ott_uD<0C*AvlI)
z(XTF8A%kDhWq`n9$!n6Ty;)$E6E=evO|fBC7$qx_am0!<;Jf6U$*Xsz-EXZ|>Ue_f
zPg`EXI`?)$Ujb!qk3kunnd+)lvDN#8I@xS^sy7K-ytaa?TZt3hT2UY*>*O@7FMqt*
z4n&ap1cNK03vZNf>QA6*D_CpZAMX|L3~KolF0<fEp?o8zsObcSpj>d68>O+foeMT$
zbLW!PP2*X~mxLDoCY(XL@_H|`)Tm(}VgEo~d&G4@IH5Qh&MspUM#MN$?b$kmI00Hn
zmCGT*ekg>H;(N*ENS;vo0man6k6f`b^u>dB5w#K&HvDQWe!nZ_C~v~x9ifMdVB5*!
z1r8VMo{88reIqK7u0nfCN42WTNd8W~)dJj8kZfrH_Z!mj@FE1qLBdtQQWUUHhX9Er
zf*+&zv}-eOF5lLewPr+~qvxXO;F0MyTF}le;;WNWAf}T1S8L<hTk1kzr$oW&Kb`5l
z=&r8mSXx>M`q<_TFwBjtfv~9u47l@LD}iUsE{{Bnmal2o#an}vRQ#@%l$pNoWP;&R
z!myFyyP<D*+(lLtI3Bf*f;{Po)X8u?Lec>*^H39O1d7=<*FZhr$R!wr@soy2gJ|jP
z>QNU8ZNDy0QoN<1E|ZEm#AJvhZk;vQz`i#6{0FLdszH9dsJewGS`{>Edad}k%oco!
zj~crN5imfo%fX*%vZz5IuM0i&3W~klpG{e>A;1k7$)-P<R9%hbpwq=5P7rH@b<YM;
zZuLM^D8CVl`}>H*k2(gZBt`$-H(u?by8Y%dlm`@q$yrrD2-I=pJuP_oMvMvjq>x`!
zk38X+ew0^Fi2{7tZ)?|lTza>OpDH&^z7b362zJ%y{bDGAjm7;_Z|*T5_W;_yhV!O#
z0zZ!~aw)bI%JcDEOtyxSg#@qXnx88!67hH&x2w}8PCWi8F_zt9M2hSoDkCWsG!k#p
zgQpETe{_p$^R(?Tm`J;)!hij|QAGEM%aZkO_EQ70>I@)p8Kt7Mhz;kZ2nYHvkT-V1
znp?~%O{a-Rr%ft_Q+7_Km}aCPjxfJpRrL>_MjFaocf16HVkM%Kgm=HZJ>W|ZDv&AC
zPFrUnwF!Cgfqd!bv{J;+5+)f9sN#Qx1{NDxbV4#zyanH(KxHZelL$(6B=|e0%mOk7
z>)FrN<@YX`D2M{fxze}+oH<0Fk_4Wm_9v!UTvbHLOzGdMu<x_DoffT=N*P(*Zib-<
zaibOM8Bl7_P6Pb0$NOdNU(be9aWF)t{y8)a&X*|ot{419RGf(t+ts>r(_Imdb?|!s
z`v<6Q^V$=|`p%`SV!<AW@ciRaMgfXr)Oxj+0oo0zb|u`kbTqub?S$;w5nkY8<S>r{
z8H@@2rgouRf_@1}U}=(1a$?93@EV4qyLFWGtuOaF{*gB8o^=kd$Xc416zhQ*h%ZC*
zDMfv=pMg^=!-1w}_xd<tNH~BH9oqK7STY=GbIi0lKO?&?IHMb1c#;i?TSD%K=90sy
zN~l8VIQq2G{_9x=tcD9~nJPg5aFHIH&Ntk0JIsJOIE8bx*KO;zEE}`%r?Rb0sm4~S
zViQZ_b!$LFe%&6Fvt2u9gc=K*;mF&m%+Y(=VxQ0vdhtZss>6|3m-~R#ZRwl&AT@3D
zTe?p^cF_Vk4$kS!d}i!k)xd{Zg18dvzMY+5uQuQzE&=WU)9}je>nI3JjT0pD300m9
z^gQlv<}hY5bpGtgO!%bgEZo!HbkpLI)RB+KYtp?6_CQ1Ik!jL)^+DXW;H(m^gdQLq
za=*xXNZrV9&k<NAccIaSLoqfnpz~6S#R?l__QjAGce&7N7OC)`ToQ4W9n&qMgN2IP
z&aeETzaB%$EKb&{ox+6JwwO;Kuow9DE9Q_70@lSAspNCyZx5uci=jrHv3%t4-vZy0
zKaCAwLM`6Jhz&yY1-R8+Z*LB|AudHo<=k#x{+}=P(0kwz#)0q&llo234gu*4OXqt|
z6*L*kzEe3LVzN;jMerZVe+&N}n8vKLn<ClFRoODp#C`qvvG!wvXo;7gPdA&KLGOE`
zN(WqNIe`P)kc!kF%@}KlHUDm<UfpWhV@6St0R8<%v1;=fgbp;-C-A#3*yu9@4gB=t
z)6zGN9^7377LRs9aX4OkkI)dkF9BV+QtLf(eu;&rlY$i4yktZhX(nBet=VloTyP7<
zFY0;dWQ0Jed@J<ond)#=fjo4*m;MV4FD!S6yXelaq{iS5`9e_$BOkmeHIyB7EJJ$?
z;^EH1Tprz-Fij!VrsmVIP>}}DgS=D6T+*w}m+ulLZ*hs9!y=O=%-J(pALWiJ_a;U`
ze`5x)+eL_1;gDf=xOK{AiVLyqfn$7He$+_vhV;73)_ybP-h=7%gf1x)kbj5WeMXLo
zjV7^>BxoGx3X5-VVapKQS7PWJ$_mt*K`X}sD`FF9q=vARYtI?VH&~IM0Ruw3b2m2A
zl$Zv_m)^n;zOQaXIJ>WBh}Sj9?3SUpEO_`DkkcT-vY@}R@>K%-)DyU@>=oCl2&EiP
z8nXQ9!sKf8^gUUgcRDM{y_IpB8<r;d4G>+nuFbu*VWuZXOO+{@Td`0fqBJf~-s;FD
zmy~~)mM3cR)pfTd|1f;kxNfe|M9(I|ra($GF}pohBRP1{=#r%Z-MbWCo%s``8YZct
zEYacl@h40%k4f+emY@*(e|S%1X&IuuIr7h2_0uzGSsg`o8pH83L;Ws_GGRms-<AA<
z)i2qUzA#9fWWW=}emZ)KFBg6a`(kS^T$1x*%t=0Ti97xNM8gc>&fN12gGJbz=T_c#
zbquv{a?2dk_M5YDRuF<UIZVsjq+8nX=49{i_g-E0l<DDyC~nvr%(cS)q(jZJ0m2hm
z+fd<BUvcM!+H>&MTuwEEMfEkEk0TQ<Hc8(Jo1mftWz$<j#0W7|b~*ZzAwBe4bzD;N
zb#iC%<$Yf&{CIWkLqtXbKI$ojg%G<~=`}SDR5FBIWDt0MuUC#pL2eq00BZy4=3D(h
z5wfS^UHLHk&|K#?8svCv-2GQmd+tZczl}cx&ZwJW>%<W}!t4A38IiHX4e>Q?f&;P-
z#0Ik%Y=}cw9?W!<3Z*Gq<(TfXFJzTsm}Y42NkYV#e=9O$JHR+4x<GK1B)`-Tfe$Xv
z49H_8E1v5khI@GA>4|jyymq0>uHuZD_!)p(%CRy7KPE+<QeikjfD0wtQst#gRLOKS
zCj<hP!I6fC?HK>+HWpmD`MXZ!i=j1=A+$bCe%<W9rPKSadNGcMlv3Wsz$jvbH+g;!
zDP!M^O1~==>a1w;1sUMH1bJt02jv@ip0RWR-t)&vbKQ99F)>Z<>>48DQ=&Iy@GdL@
zC}h~*e_jvN+rOs0?Fi~SG!WZPs=Qo@&jDh>QH{oJlx+^p-zBq8BKN6ZFJLBf8e0n!
zH+>6V349L+)GcwWD^gYzW$^QfT#-T{HDG<MkC)Z&ee~!V_xNLh^Urp%=N~LpuWl2o
zB~jHqMKv{{zq{zz))B084$&^VtsW^5LiE}SR`=a?t27;5gk-mK(U;+SuRqZj7IBPO
zv!V7&69&H@XjzH6jlltnOL4o$b9X;6Ls=N!6hD~HYEtd-rcGUHWP(%f`wlL}1Ag9K
z>EVZWF0kEiP<<k@m#2vfq4D`laMXFj;+|at)!A=!5>i_2<MaH;cAxKCx%AZ9z~?s9
z#9?Abd0lMu-J7}P24@7L`IPOLzl?o)VK_41d>x?M^_(zfu&l`Criv;*=62o88i&rm
zS%l$RSEbx|DCNGhBjMlcdhl1HIB-JJqfcFe(S^)Ua=*}yJ36qmmHlD4`~b?k(T>KT
z?x{R!S!?^sf6zpCskyUE3AsJ>MWmA@@3hwjry$aKbsa%th}CqIY@!Un2YBO*xw|)o
zKu0?J{<BcDse;iqwk{i!?=|Xk<2Uc^+gVui#pb!X?7Fg@Z$!ANGQ8&X@wTbpNXCzY
z-GaXlTzg3XfJ(G&^=jG_itzV0=HOj$kI+W<{YN=kAF8MMh}00iraEfvy5woFnlJIs
zni-Mtwaq71ca7SZC_sr8GvcE7ZwH=eYmG}yQ5}3V^#W5$;I2U+z_+o%$nRH*;#8HX
zwB*|#>ZAKJ0;Y9WN9>UX$MGNuP;2oyk@sn)BR)SOv0P-oMGe&1JH1C(Tvz<T6LqN$
zwj+Aia(F48K3|<uzraKls2>jJ&@D8KLKMGIhWvNh%iwd|9n$=&YjL>r2IN!S{zTC?
zF~>$|I_IkGIWR8y{=|lhKOL();u%Tr-A*47J7m?+C`Bed;{C8)^il&wz_-I_M5v-R
zR_w9G<%G$~2MedsD(i8(N|*HM^mN*OZ2d8iHAE`}t|-XtVEVbWPH}U=vrzqTv<}dC
z)bhEFD4k^W?5o1!Zs7P59Hk=S{6^W>-2?ZQo;vykVUfv+;YTvC&X~ea0i9)oT!H=(
z$QxGh+4&T~OhtB32}^l1*@~t<qSzrDI0|uDyYLMi^S0x{hGx;{s;?slU^Eo|F~Q@0
z*QVgFxJUU0hnk!%McFP*7Wb{A_9c(jVMhLM7QG;}G<N4y;IC3A*~*-t%(KBFqM&EX
zH6Y|Ls)bEx28J2m0eokXz?+h{`H$<t5On^YoK0{F71qG%?*Pa<BJkIc3zw59x))hq
z^2d8L7BH=@U&N!?=$G{#xV(F9@|tp}%|3D$&4W=W?DH!HbnvhP+sg}6q=upVcxMYf
zTqWAjgN8N5Z)xuC6vd8CT?mWNPpkZL{~RLpGzw0HIFn;~qmw*laAJ{toU-u9x2fAA
zBb+QUS<gc=jt3+6F8J%i=-!emAXo8Pn-hN--(y>c8~+J(0k9VlQMAq$>{I}T`SsY|
z9a3_6N3N+JV#Xi*pq)JhSzHVAnH|e)rOf;KizDSDe$h+4>!0MIJpM4oMnT-*r7z8|
zuXf?Y(J`J^A#J<1F6{AA`&we=2i>6^HJ2SAIbI6%%qF}>u&6?fgf)mPxVP1JEc|u3
zzW(l~E(8hm`fj{l1MWo1n;$6RE@`#nTgEE_K}J0{lc`>cKG@1Wf5{$ylVZ3E8;1@d
z9{6K~57@ZRHRYGg276CV6*~$xFEqD<jrZIsB)pt89-S!1gBMj{Aafg5O(R|@v+tg*
zCB6Ymsl@(37HO>}_SF*%-E3+tvDs4@+lxM*Ge%v_X~)RVfqv5B1NAbO1}`-*$+Y8!
z5PM?y{kJ>54~01-OSce$xpC(2&vBjo>TjIxFmbv0o(qNpzGo8yh=Z?59LwrAut=ec
zn0WRX!Cds`S@cw*qXQ~<oj~FExH}1snrb+0nm*5^S3~ca61^z*0%hOh2KY*{J7}R&
zwu*h$oP$@1p*qyG#__GTc2RZNaODqk_hnF6E{`YN>`0Yfaw?dDy!?}t6@k->`DsMu
zb)mFMCtH&Utn~XdgW~pLK_(Qjxq79{Z!^pTOqGp1n;SAybjH=-j(nVE;0Z}l<ApDK
zCBNi&JAQfo)DGUuM?ld-8u^|kVB)D1cdGSNby8^GMT%%`sngn?u4F)2dbf?h=``@H
z5i?vu{U}K>KG!=P=HL0i_+SBAmQi|-y0t2KljoD!On=x<mW;f4g3B<c;6uHa?0D3l
z)G<B}V8tf{owS5kxnfVf??{?1t8uCM-Ek<u`3A=b(k%N+`oTGg4#R6y?@hr6QXwgQ
z8pg}}bbS+1J>x+lgCnK#F<`g)gMHy@uvZqqJ`i(B;<jzmm`Nd@xm`C>yEk|5kY;z0
zZzd8KKrpcHbv5JL$)QP@WKt3Q+b@@;)7}$C_wJjru6i+}y6y*_d7tC_VMjE^llPt`
zmXyONKD3=L3_P10{?ScBCx?oWMZN45aJd)!VW|9~hhBcqNm<om;5uCNul}EebjIW?
zYAH6|PTF@<Mjg+c*B<)L^3Z+h`T#PlB;7wTk<M6za1As!zvihkehS1`k&X7$;9@b^
zNxQXBj`HJlrn5D2uj-qf$&eczz<^g+1`ed9$DuR(-ygF6gs!8jj}}YrQxBn^R_(_T
z2_LrV%w;xRO=Hp5{aXjKLjfu}OeJC>=5-UfRoyWkKX=*}8Ny|b`Bmr=Ev;3sUbXXN
zPtk{2xxFOcRM)_XpVyE{NxBvlCIU;<)EfPjh_K}+0woChuJZ|`nI%)VUqL}s-kSQ5
zbJY?{C!7`PzCBB2rqM~=NuM)_5U?EhR%VOP=Agk_A{X-jNI<v0LIxZba!J;sR4T@L
z{%YDG3m+t6xE5>k!~2=Ci(phH7smHfBtk?Yy)lZHT*~NdzAOkw$zwTdHxOrlnT=i-
zDJ(AmtCCB)uDEfam6M2tJ%x|DkBkUsb0wrC%rzxoH~}LZG6%{Bivb#($Tjdp6`v%x
zB?0rK3hBj;<u_+ZN)p~ghTjovkDyul9~VpH35R%B8PXq1>;&GwLu)FMD@b>VWeHfn
z3&Pdp>xNB+i7#s`{Rqb7WjW-TLA*Ph^F7BPI)}H8dC%kLL+4wf>GiSxzg|8EGyCiq
zqmz*zhm_qyGfVqy*RpAOY5kNwzY2)aGi$MvbcBCC@pz9Xq@?@66=uXK_1zvWDW!Nl
ze)CQ_WGm^7yk}NL`Xz=!p*yTFxl7dpPD>V}b-tcj*91@bT~qfKAM3r?ziSd8fCpl8
zgJH^j^1!O#3MHHnP?#Lri&Ni`;8r^y&VtS5j!;Qr_+e)JEbZM}m9?ECo!v-o#AM&3
zG?YCppS)pvVnKMQ2$J!CiJPqEtEW{YCvVb*vfk&xeI&OCM_@EM)r61Wr3;gsAE`ak
z3~R~|dhs+%J24sbHm35RJy`2S@+4&_5uU_`Cn?WBsDAAs8pnp`kPE!VhYlCz#vD#^
zK0sRS>j^dDTq(<MO7J*TM{bVeyU0hmiwT(?Jyun<@vUk=)4&9iK<gtYri3j@#9kBq
z4UdVYJ*`MdGvwY;S((WfTL}jCB-^~d?tiwsaWR|8fX%QSq!g&NsCp9eX=d<$y85Oj
zVE|ppv2EM7ZQHhO+qP}nwrzf6+n(8TYwyFJ{RMeQRl3t1e6*n$$XsRcmk6Du9iSuK
zUP*~K*uN#WmNOn17?yBGgi%u;)gx?9+@Q?<3dpm3c|#9Ci2#7|i|R@C`|~BASsbYI
ztkrH9P@Y`k60KZSdW~7|w;%hnF^pyE6LN)Qc<cNY`D;xPq>TIaSC%?AW?*i2nonuC
zwfp9OGdo1so4IrrUK_V74&2%uNz4u-5}En8D};(@nKffTN@Vzkv-)+gT8<2?XdR#L
zf`cRUTF~AJo<a=cSd`Z#Ll!w0JEMZshoW_WbY-725nA={WYqQdS07vsYt8H`91)1!
z&r_yY70V(FETQ3uR=Zgg)|*f#l>aBjNJvdx0?oxeQ*@HyLdqP$(N<htYk&@|_}3ag
zhS)J0*sdZhjW6`YF8UqsLNchhK^Dz3^o$}l+{G`Lh2juR*PJyK5%sN`+eY=MY(z!Q
z-Tv=t_qqMMVCsxGAyP3br`gXP!mxIWCxt)#UEo8ptK<~8pr78;2p(~}!%G9~pC}Z>
z7*vY;tTFev{bu5gDlJz?Oz`%I5`yEmI`(})%<HNoqnU09H20>MiM<Adly_Cr6H1dR
zJ)09heKmV#4Y|HBHzpW$3pC~GPS>8Ds2{DC{{_JCX&eQB`Hwa-Ys$4ECjuVTfw2x~
z+RN!yoQzVyGp9qE@^G!4+}hnK#Xq!yw#fsBRp#kIA&{Q8gXX!($A2!Q@A*`d@hM!j
zNH#Xav4`>u6j6cv^qFkbGwod2=_)k4pM_G<|Lvu*!yp`HJb3<%ty?t<+Bbh&;6_>e
zlS3A8vG?dGNtjgl^8_-Cc)c3>7;Degn9D=t1)B+@((*i{V>LbFH;-#G?<_2n3<F8b
zL6Que{62=KL?$itwIwYkH%x^;OaZ57%14cwh@TuU&6Y9YEHXNtZmgIr4~~!0A2^WT
zbOLYTEqdm0Ii>C_N(xUw6Zkxt`obgaXfLvVf#weRZ&l?JHxEXXGwEVnlNTK}aXp;I
z@2#eo&cj$1GbT>S9kQ1an0}f1`BB(gJbtEKWLCB4VC%%F(jT2yGTEHU348n|C?>)d
z$L4n1b+9RSi({<6dind0fjd=@3bj%{>SnVVGm*ngo>kGZjHRqN_-feuquyKb2l)Eu
z@xjLPGhKT=uN1+9lM{^0%+$G^=4^?XobkE&?M@pJ6-H#gyLBA;p2<8pG~x*GcgZ|r
zO~SNRD)wsO;-}G3s2*4{0AA2LbFAg-40G~^-?dP`a#GW)y2fbRVx<5m6n{?fv2;4a
zaApBvgT@Hr17K?e=6C{5V>-p&gpU59)HA^;2VMegL*!H~?!iIMrVJL{<v6~6Hc8yx
z=y~&STFROEj(^<0*IK8uu2unEoh0@EUDr&{5_PE5QQuP;Pd9Iq%(4ri;+U|ckrJ!L
zxc~&=toz(M)%P2cPw7K>>zaiBP<KLK<XmpSvI|0S8!^jiyuj8noIx@JRmhgOg(*gr
zJWL1U<8u&XHKU=idS#IYxj>EyjBE~@;K*}Hit8q-B5+{+T}0(JJBKH<m2L3-iw3XR
zkTChqX));|8XPW4zCd~OEh2g~&|dZ<&acO{-)O6^399EMe>qY;e9p91;6BKTo>a9~
z`r5_Yrk3(r&*$A)LFVK%%37FzzqgowX2R(V*ws}Qb4l5oru{i6{DE383oaATQPru4
zFIc=^+umvg#pV_U4t$%gHu~23>rl1UGzafvbm$>Svl&4Kc@>gHFxQH1TgYBXLY;Ex
zYQqZ$gs-p;Mm_|%kgV<Y*}=92FUe&;$xt19O9m$Xmy&THF=0N9=WtGSXSCM7db3;@
z$gIa>TmKi2bvtjar7;{MpkI%gGEjUNGd1e^E4v`p$~odlOp0THZMmR~*472v(b2VA
zmiV2X_K?^2yOWnp_GyL6aLa-LnXOGk0fyRYPY)fg;eE6C^pW6d$H}i1z-8#0?QC6l
zc@t&VOfV~teXFm&wpi1|1Px{!#!Y=#bnl-&<cWx~>$1IL%Gmw3r&VOzeK}RWnF&KV
z++?(GmRas7C!#c>FkCJ)nYy<Jc_MFDIldx>=`f(9Te5&taEbS7F@b&FG<<b$&E^zW
z9z%X^R&)Zp!l~D(UK#p(vZ5fQY}ou=XMCU}uwkQ%9rMe0v5mPwfZLD+6$M+0dW=BK
zohpnQB7);b2%n#Twyye<+C^$Ds_iy#4xL>disal|X2#fS*U<{vWA`-bMrY&i5?887
zbFw=Jg1_u!kLyo)c4cTYFwS-3aO08;g<|4#uSc~IWI}0#X17*H80+vsN2WwyKGO}f
zh~iX<L+jfqAL{5q^Oy`oG`h03vrSAzVU!k|;_d4)9XogS<ha~^k3^{KtHd1@jOC0L
z8IttQm_zNwS5Y(g^mQL$;02`V6<mgm3Fc|`AqP+@oefaF<Q@dHA_uQH@pE%Y<!#o!
z=lpKRa*m=0B=BB#V6rQ|LE&Ii&by1$PR86}p}OT8$~{qK`Q(kp;EVqUUQq$0=N}U~
znbg_hTbYPCGA<S_Sw?*q4OD6@Z%5>0{D{=34#o(AckQj$^D}7xGxuVLBy_iL^-FL~
zTC2r3DnyQ<zLfNga5tMYk?vz;(>|-wB)a(fqTuX*_xA(NNOkcZ_ndF`h_5x7LEwgu
zzY&gft!%N2Rnpq&A6=un2#E`R`(;WI_^Uaa0A4yWi_#z+OF4Zw?ORdhN`n^rc%G8R
zG=>R<CU=oSt!0~oGfm*h6uQmxs^}OXUzQFR1F_IMY?Q8EkeX5^DnlXw%)DMtJh5>s
zG0*6h6IMHJ^j1y)*d*R~y@;zNeh{_?K9p2LnIr@ayLkY8%U5e$fp*}`!(JB%N6(<3
zwREmu(4EgAf*a`opLqL<XYgf)q_mTTWcw<hJLjTAKqUP+V=LRbh-zT##J+CPBH{D#
zT3$9UpGp1>SSq696gxcsII`E<u_Bhpw~tZmmev79PLu<gO}u8E2@)eIy>AbNf?b~S
zl;-_YNqmmwH1uWSp&0w0-i)fN@lKar5Y!s>P4?VUOdyGaDk-uF_w+^Wt%~qfv~$CV
zwdfB5ea7KN{RtbjrH@lsR$b@mI=XCu`}Kxu<NuEQ{L`h&LSGvNQXAlE|D5q&nb;F>
zl6Bd&={*f`IfkSu94>(ONNt7-sVdZi-QMEPnxHxdzIaZgsNN^UCxn!rEK8pbif-}k
zXTX=Mi{=IJG;-!z!)Uj!e%NC9A$D^FW%S<60oPy`N;0crW*Gq!Wapi8BU4=~YIi0r
zZ%TcL6Q>61`r`muQih?D=zReKzNt9%*~Z=vXc@9dRCIk`PS;*dnQib%P>=s$0q)9+
zE1vxB!<|V%_Qq<Rn~R5Ey|3{puDx-Zv0yNVx&yeaJ-Jg$)_NOLE3^VLmY?%=F(!mq
zf40^$b7k>A9yq@=HP$Octj9}-3Yf=wZIv<oy<a_PMBSF8Fu4U2_few8o)aRrZdVNz
z5^-5v)df_A#J=6Yv<Mb(3G*kNe6`nP7CQkq4I`jMUJ?TOJ=x3`!PbU_MQWKZjoibB
z3*K&4{OqU}hMVmhg0|Bz&6K`+NeC_Fbw9yydtx1!jSH)!0?god7)*LlZ!VvEe`e{X
z`G<A3ys!`5mgDm@$)CXlpjzgpN*AWa4C>uC--R7vVDgo3vBxdswO!9Gde(>Ug14n}
zsoVSi+U`LpMnEmxS2=}nm&>#rI0K3(#PaMuB9S)*THCkQtfAKlr$I+Pc%(kj9Ln|w
znl+1}o{FS8_;*%zqwf&e>brJW!dmTgG+98ldw^s{lvY&DXS1s?%z+;+fy0uI30OIt
zxkhVHH1=wAr4~kz_Zx+Pd_s*NhE?>OE`fca#dH-+%69MWxF+6+rENH%{~}x8Nxpgl
z4v?4sMUsf?r>yFhs4QFs%SvIdOLXiqkbLiB>hR6>b<H9AXr5%aNcz_E3mz#U;<eY8
zvK<7}D{&9Z9@f|n4Cjx1I{ob=AAi%*-cj%w`pD(wBmYq43pksLxr6Ym7XfbquY04!
z;DT!ks<YEa{SZD>?LBQBC&yKuSKb_^bq4|2jukXLGM1uT!U)|g7R!*O=~UHFpa<xw
z8^_OnFGTM{(xM(tEsHmf6oKrKAUH>q=O;{(%z@enL?%8xX>&I3vqggB28i));v$N;
z6d@eSMSA`4Et>aM_|TKKBCnkSl=%4oEjE^5>avEeBi8!F!4IYqC66Tk9-Fls1nbA|
zxrGI&OPn!^wHP&wB|25ON2s&KbEwOa5C9-ea>U5w>k#(qNX4udiAv+@Xxf)IT%}(M
zxfo{1x+jRGowo~^S^-kN|F?v`WDJiUC-{-at0Rh8afL2Kxb=buIVA;og{R$?NOipe
zAZJs8fSm1$Z^*uP8O#9kY?0y+(=l-AC53~dw7f3N_hch0OCx!<CdJ}LH+;I^cHZ+7
zzhsL)9MZpbLmSnM+bmsr+=saF0-lrz$p7O!C?54G-U7m9lAiMU&Sjt7d+d!C`9{nF
z^#~u->E%fsg}XiCn}^@V9ir?OS&YUKkl6K8iqZoaR?S$s+|P9jr-2|Ob;*(yz;I83
zm6r0ys+G0UN1S-v!U`xsgzSW_#p(@*1tQ4#syzM_MJN>HBD^up+=U;UvCXiI+#B()
ziBKVI8w7Z|wmbycy0%~uuL9wC2!eMbM-GG2Zru?p^Of+UM)j;6!fjlk2Y)0Z-b{^i
zh4&xT3k25W#cOu>9)9EsmZ`SE66#fR<ZJ+MF_^q`9Ho*05y9Pz4^lW5c~4W3!X|6@
zAiG>RAjbt|m2&vZe)h?l-(g<i7Z&pv>2ZYfd}gAY1`FhWY7%9CLnEn<;{x+p1Bs9l
z!MuO9&mlep%E)JC@iM8CEdeIlYH-g#TEfGjzD<*}u5($UEZ}fygJ$=MPm$<DpVD5R
zFN2B|msVfnf9y;aQBvSt$Y(gdQQl}Z#O%r@8CG}S_zF<0<-5&#=_9<K=zY_#X>2_G
zp*u{Guc-3$V|`~)^Wj5WG&YPpP}b^gjM&T60BrRrkP9j&%G*^1AMon5GOVlRBi(^5
z4+tp-W0DJOTZ~lAqjr-i9=vDS`zp5wU6=wxmx<k&WKW^5v(bfndjrblpbz&BPIsSw
zh>cWlQ(L{w=Mftw%bo!Rvtf`>6nrWVOPm+uR}#ZVcZgc)?C*iQzpP9UB<d0}b?L-N
zXRS5xNJq)G8j_5o^WX{S%_q9qqYing91Q2QR=;lo+5vN?YL^vZvE2v8Ou4Th3P54g
z3v#e613KcN3+1UFjCAE&)J7w$y>}FJx1gitf;vE4Ij>N~-^SKu1%SB~Nly>S`+jA!
zdnSV0^OTcP&=VuPnoHx_Aon*!U=~=}Up}tP>Bc}8QqVtMUB6nsH#1<9X(#g<_H+@`
zB_T3Mg-7r6uOD^jV7nj_R{;s;S-^84V0aW}i4-HfrS?eV<Vx&~QiTgX22iiy$G<}T
zP+K#Sp_<EtSPPO_+IicorGl~<iloqbPzZ8UrDDClOzw%{$HXDW`N=Hn56rod0UWI`
zaH`ZcGdmJ1FN8Jv;^}YQTXpw?v{ioE+hwz*-gTfRg_Yt&d>mW#6bz|p5m<*D{bPj`
zNuuL|<rpOi3W3n+2@T=3YfCXGmbY6#bE)$t7Qd)iC<>7$7!nl&Fa`CyBMahd`OA4G
zm8rpmF9_JdZje2V%l+eqk{Xre8gV!A?;*TDGo;uG`uPfp)z)an81&IrgJ+0!hU=Gp
zk>bGe-0^1ulIbLV0;E3pcMP=ZvOd86!IZ{1_Ndw27gbK0l)?i_lk#6Ilhp`CF-yqH
z?fCvw)m*ykFe*zyli>nC>OHFZv=568!hvuavf^A=!Qp7y#jo%bTfT?*qC{w^W{O3H
zYLNO-l56K>67=|@m3*<g4Ldp3n7=Ig^YM7BBBAbYQlY6%_bjU8TiUSkN*0qvK~n%H
zspc$HkBvbTRKre6(o;=CJfQaHCZiu8FZe9I3vO#kU_X(tV_II>RF{aAEWvP;@x(;N
zxM{u%9j$S3Bnw0WfSxVxOt^bE2kMzhi}`^~NNQ?xyUx;~$j>x&R(QL?HOcBm2<?Vw
z0gyjNC}mywXV3J?+~@oUwgS%RV$X$UG08n8&d!wEPdg$$`)KT_)Rhtk*D_uLf>9KQ
zogTw5ADP?d{kiYcK~yOS_|re{QLoz}EIiy*_WhSmpBxW2k)i2LKVObRfCyB|{Jv*>
zI_dcjI-jM+VH*ozYbkGCd(CddO43n|Qe@%DDt%w}8v&M(c+(7op~tQWdT6Y$SDk#n
zQ}a-Edi((uLCK*z@gSb8U9!Wb@G3yz8*15{9m<9RbCV{W;a3#JmMxY8M}gHzqap6u
zH0D(}PR(9a3y#fGm?bGpBRhKLBFl@p8d?Zy5ns;HkPqd1`?N%}J3$A}gj6TZ1j(_U
zdB2BDE0|OrJl12vtVPQyMBo&5@kK<tP_sNM8#2R4r9q1_K%21}N_)_Sk~T%XZvA1n
z^K?@-zvQwDdk?Bgr;Ujo23EG$xw`h-Xz%9`#{m=Oyee*E#_f4%@~?kydnrz%86Jpx
zk1swBqURwDYY#-DvyV0oSHqlrpG9b7NlIr3V-i`TK1Ua^stF4?mIOeJVpB*^k~oYX
zaZkrqQh?r>D1)Y=l41%gA96FF9=JB#a90)>_bWNnLrf4iGH`wQ56ly))?nB&B*kx1
z<YI@hV1rojgQ$6kkWo;~%ZMD#VlA$pDY&jX`7X2bu4yB)04#`(|5BIoyF&#0`zo<?
zFvRC$Y*l0Fz(}FrFb)0O%gY^Lkzz}w{EQthcdx=+e6k`*h*3DM9jkHaYaju0tM2Px
zmuyIUU&BsFZeJZbxuqjH{jye|ubHHnsx|fL=bRuS#501sVGXSUs4#UiurDTm33!xW
z0xsh!+#OT7<f;|RyvqUf^)&@LnzKk`@3+Iu?4|uG98~a!U)@zb35h2HzSYO5!pe4_
znCzL1By^ma12~O+bIaJ;O&v7}jPBGU+~`8-?F~&}yabhgA?@!f@2EcLnfiQbJALkR
zJdPRw+>5ei)hgOh(Vm-e#*Dap4?e5OC?YUk!*<E-s*8>s)vhImlnGSk#0W{T2+nI1
z6+q@unec1@&W)RVpMMK$A$V%dpx*Li*;)$Z6pZ`P*M3I)S9VyG8MF4RVDVh?oM6oE
zBz@7g6PB&`XhDQEin4SPweU?>-*k3(2y<n%eY=IDtovM5GppjG3w2M(9hC~RqCxeJ
zt;Env_?DLTrhDfF&@|<rwS`rT>HkBtVhE65X4A{xtsR3UD$mK`JpR2l7wN5OoFXOq
zJ-<ag{eG1_N&VguDx?)Lnw=!$`-w?Y#LKaIscsZ_qS3O`putO&2GJb3n>nZ8)w2Ca
z;DPuP$o|7Bw&8Bg646ro(%=}t(bPNpg<>*0v0JJ>TNLBJgYss>LdM%t(Tw|Vb`cqu
zC@_ZvPO{)LM6!3M`-%M6Ph5eJpQJmCKvyzbl2LVsvZ|R9pDXRxa-a^ce0$gzns^z*
zi5u>-@wgq52lg#gi_H+S>Ib-2qE-l|_s%ljDE7njLtl?E0zcEmt0ybKodZxAsqpr?
z<k1WCa1n$Azzx}LW9vCwBd$CzS6(*ZjJarFT^@^yU7Q>OM!!c^(qcZ(Jqr4TqH{38
zlCYE-!5wzc@44FHJA{pE5)a4^1x6u+F<*}jj6;D>d*6pZ9cY2HRh1dHt}16?N!y@{
zRuoS7P)R7;J-wKvKwQU%Kx4j|iTV!+1(j7zLZ=ae**_9|*RfS1Dh0=6k<b-Axox}G
z;a}nVpg)%u=*3$mu#FI#VC|7=Dy)Ql1Te~3#pYV?C$Jh_YE_NuqNZrmrE6>`Be8td
zIhR|vy@%~h=NRY{gq~{Zpdw@toJL1sXH^SP3bp1xKvMT-zz#U`B;s-0RlB0f<8bVH
z4KO^Q<&`J<qov+<1PT9SEZrGeaP<c94MeWH<wY-U1UqN{zH)d2=Q*XO#OIMj+l&26
z?HWj4mv848ZyFYqJ+Q9vZocR?6HUi`*1xFtSWGq`H^SU&-aelin)W<DknS{7y~h9%
zCzGT~4*V+nGvNTLC~uLJg8C{hwnzYmFcf;PoEl{e7MYR1^T+Mf8Xj@ZhfCwItxL?p
z@P~-YC3`Z0di<~{7freNViv$OofPZEg^oPI%hkjI-V@E)_7bi1xLX#Elb$w0aIiq1
zycx;IX;krWO50KhpKMC-Esx|Hk8LDF@>L*=!~J_yxOheP1NsNlZ^$-CCL^$YHEPw{
z(-)}5*bcbS2wC-V6u0p}sQ2`NVJm%}fccx8I>`&{C(5`PS-ZGQRTOz=Xq)IFY5>OW
z{#$)1(D^m)q;tJ$1qRnm5BpZ4Gf@xXMmqT4I9yFlN%&1bQw)ebV5U_!k6ZCdHF{c+
zcbI(-aJlwdr2JX1^K=izKQA}xc{6d3Ti@Q){QI*C6HR40zoYwV#+B;RFm>h;FxhA`
zskRByZQ(Sm!XdiuH(aJ@*8z-N@d$k-uDi34qgqrNOIgvzAGRXYg-`p0%rk>+`PFE^
zXYq~7MZ`n70-)MQlic3EMVGJK`&Ue_>t}m%Ns1T=U@czt?K{Vc6{u%ABNtM0gM8i6
zR7->YSEeUmsxEG+l#ByR$@dWhy}qoiU>nZ5r$2WP_1+SkNJm(%m=EqgIkXjYZ^m^S
zM)>|Cs5m2_d1gSd7xJmd#_GO<u#_{14qG*wmol>ti_WveD)2^zF2d0)U~RGm#jc8<
z>p_vwPQK(9%V>2hAm&)73~~j*5V);2`}}nsTwvT1djH(nG79#5pK|wvv9q#w(2GJ=
z>XJ%ECxwg&;yB---HkCwDCu?oL2w{Gd?oU%Bm2lK-&sVSafQahD1Xn&-Nb{d6Jsh}
z@te=lT!X0Je4%IQvgcx4wDDUhjwkq^hcBRJP&yF#CqPx3N`>KC%heqP)c0(JUlOt3
zWy~6CYL%WB<u@P;N*gsn?JVY|+hG^3F916wdrbf}15Oh@ym0zp^Yq;gQ5cetr7sfk
zf*_CdFab%s5z%`OUdPip7L_OaNzRUv?sI`|Z0hfHgc?YI@4k&1byZ5dfSZI@_)fN^
zmlw9d57Yz<i=B{bH#AgcypMEpEpP6zn2JO)IfiJ*_T_C4ob7k!g1yy1;T^v=R+;I`
zQu9CfbupVaMNeJtLEi91DIY{#div;xJb)qF$5efEBGTLBTVbT#<XRa-EI>nh!Yzpm
z?1uM<nPZg0Y0gfEZ65=s_wXabgzJWbh~g7Sj1wP<@Ze3J!k{M2RvdRpfp+PGR3af1
z6iXmx3t}*fs1=v~`cS<UD4HjMQ@?Z38mPW8S2*9Ce&zRZ{tmhW#4Dw*cRvR*FwfZG
zS!!NUuju{XKdRp@d4%JtCcZN-ALHzI6pbA>_d9UP*E02i8?_$SS!N%&YCwvt<Lo=h
z#_z_SU71b7<gGl|%P`H}8ch3xsdhY5<~e#D{MpqH;#${i&9@N0=9CIQ-lWTb8^dtA
z=!3%@yC&GY8BJgmyE{REB%C=2hWzD^DwufH5XrK-iZ{8X?Uhrh@KY|+S~dES_deX>
zsT5U*({8;d_%jgiuX~!}ztXkL1mWWq{jwqD&m$w$%%e)Bz)ND%L6msLm{)uRgAhVE
z_lmMTZj`b9@OAq)PbcvFJ-!IRT*~TTV&~|ns3wy-8du2Xe}^*g5t?Y>(r~m~ydL2b
zA8&G#hBNGXHP-F#3pg@#(ambuR~hHe?K=)F7nmIZ@VfXE=g>m?%y%z&c*Z4~iKv_<
z>R5aXO<u={hIj>6r!O_Ma-xLc`6T4`yhHziIi&|nsD&=unp%0O{#P}Bd#(L5qQv+c
zOl0zE$?hiSwGH9~1m6PtWe@?pW}p67(*>yUJKWMyAYb>4Awe@}F#wFPz?}olKze$@
z2*w`ZHuE>2(S2A3b#Z&<UEOF<%$HCdDQI{@K)*2S93&Q-wUj0#IMDE&4iFU{5wgRQ
zAFWgj@^ZeM%2;6ZgP<lR*#4!d>uiY=Qe>SRko7a*{LIAIxlAJMp~&`fc`R0@DGg9Y
zQiz7`57DR1BHH_+^S=^7!ndZ+!BAq}*!n2zEE+hb$x$TTt6zeiXS(@)xJ;;^gF)y+
zs=Fx9mRsb3nHyWkM$Rb+BT18NrQgNlN6i00{8m6{h{&4BTd7H81Zg97qFzPdXdq@Z
zT?ob~KxML@FP&u=lR~3V2mJul2Av-KF<Gudjvy^Gf6}dV;>o+q%8Snit<pv;2M<A=
z^^AlmySZt-o30a*n{uaydI=CLJIWFwgd@e6KXk(-xzN6e*>}%j0k7=>5APCyG4J@p
zG?5u!eU(cjaxJZl+<<`X&^a^KYUHj05#P2xUj3e;Vd5QFUd+v9Z>UeFPHSksu~pmY
zyOgQPi8BYH;D9ql&i-k#25?@Rq=#&@kLQe^s+mme|K09$4H1i%Xas2i?t(pJ8c66z
zk&(Ek5hnLG+Gn>Mux*DTBWq*A;M`jp3Gy9>f^?0Q-sLKvxgfAyXCGoGlF&WV==*3G
zx&j{~MUJaR(r&Wlg_>ROt<E10x0hum-FeV#l)w1Eo8ME_)!LmHu-#)h2fKK%rF$J4
z&w<F#A8QBzImFI*H*~c)TiRiMOFMuObIIj%8eo7EWq}8cU<Y!S)u8e3u7_U7j{Y$(
zGr3@0HYJ%tm&^QF0?gM$66KYG&m{dF&<#A5J~iSQENT;pXyI6B+}WwkO&AteapGtI
zGnIvW+`AFzby%yKRAE!B)cTit+NaN`aB@_2R7E`<U*&%iG?s8-mJFTqp~gIlb#3h{
zia84txw?x2tgeVKOo8qz{{@7Vo6DD)W5@ADO;TXF6kDc9O}GJC_lzXqDz(bz){gkk
z#M`;x;ZIt=Xkq`tWC%CHv<RLEQTwhs?@u&btfze8<s)r3lJ8<lr2&+9+4Qwb&#Dxo
z5opWTSph`_MP_lh&eMLgNoxDgJMe7{_D(UX+tQ&q)J{A|E``a2MiYMznkzq=Y)pul
zavHa>*b5S?oOXcD3G~Kp=(DmQkv@tC9L><fy3_)acWuMG6=mQ^JB%L(5h=zcaSv5B
zR#1m&6i5E5Bjcu|>@f|-#h7HRhF}^eo@~F=+fqLVL!UZLY-dpS2zZsQZ}i8@U{V3|
zNlCZzt=vSN%^>F+un_VF&Bf_y-)BJ>`*U5o_TQH>1cQ<qp-)?vG`~g!l<<#NzAh(O
z&|J-NP_{z?>y=0_r6^=ci)u_zjt|B*!>4m$1Sbc8p^LtZy%<BphC|d#zx_VGY9zl3
zsC<YbS7cR}2DU2_joG#Lfxqs-I3p9jD%zf&!M)}^N@uSv(l9R-<1E?WEBr~Z2pm=f
z#R=mIj(V5r#89_Y&ZWmrQaJBShTRcawM;6z_w<Aqt6tWbac8Vl#QP`;8LmG~g72Jr
z6j9F9$<s=!N`;+uR00e9_+X4h{(5CC>tPD2qM5^WVCjKPNGq#sJv->Qyb)Lgt?T08
zm%xH`sx}^Cq=%|cmt|E>w~><o5A^}%<wqmlDeaqbYN1tgCrUywiw|CmqFjKWLDFw4
zwfOj|8S%;lSOq}hycP^iNC(1!RtQJMzK<i@%`$%Lu5l_GK+eUo)~MJm7<7QW7Nr#_
zLC;r3Qd}1&-P>Sg3l$~S8gAs*`=}B|7w>u1)te<3jHf$e_ju8d7<-pl8kjD5tvg{U
zqIej$C`qEXt7*HPeIhZLTcP1!Zo3Qqf`4!yh0v(Mk$gAB;kkM%{B@=Vvj9UC>X@zi
z%F}g7Eoz9Oh^E=2Y#qR2Oj4T(CbJScd`DUlzRh@{>u>c|c!k=)J0A%xlMKN;F}brk
z>N_2}^Kr~%cbcU2GwT_-?Dqt$MrRi4tbm@U5MS90D~3E!8W7LgMR1@o6&RPlLFBNB
zMYT|#o=^KiM0FnZ7s7f@uwUG%T=B+XYTcj)l>=fnb7N>diPOo1>1XYy^wWC;Ti<x(
zrH%8N<$Q`+j6KdZ%ZNNmAK{a0o3SSL^+}m9dd?RAT6@_svvb$3eJAW7DmoqzVy)h0
z>=P50cZ;<gv-W)`bEmFQ_%aZD(=Oc+GN|)CdmdKUitygq$O&<ARw6c*dAbJ+0@`qJ
z)){Zn)r&rs#$8Cu+`L{rOr2-WE8LTqr4PHT-tWW{{yw6e0$qrIc4o@QQw=$NQ^YE$
z#tqMw*D_=JGL2JSTF){8$ZrX~r?(-_?!EJ9C_^Dur4Hp*ZtKWU1MD_ff4dN{nqL{~
z|94TqKHeQylaH`ORa8nS>YS)Ip|n4=%(^|Jqdp2)wx%1e#LV)A;S?3vr4lm_fa`Nh
zYM^!K`umZYv))J4#2x<uB^(Nlp}-yBqwsb>ZKdP;6!+-7U##dYkLZ&@r6$UoG>hc0
z+q#Ev0f`DU4JN;_>i^eNHr36kovdmso+$?Cw`f8F+f=$rcMH*_cK4czBmtn7`+(>+
zOl6*|6pDxL?_?0jd;u|SA+WH<&S5eX3#q7tZr$0r?zha>d<F(Lw&Yxxu;R(TiVwE`
z4L)Mn#{~D`276Hks2ZE%nG?lqY^W>m^97NkWa>J*27j~3fFt7xSGf8~z{WxmC9zTN
zF%YOdC{!_|<0>a#l9&fM3W1qB`?KBPXPrI1&!7(t4%o`jVp(UcAg0e@L>$+Vkfvmn
z*a<wrcJhqg11`Qn7XVBjXK8X1j6Tq*xx!9U6u_>h`6ui)F^Io0XGf@)1@<QV`>8_L
z@?RVdSbNRvg%3pVkYVS-ATf}za-YQTel%hwx~g>R5QdiW!i&nHtvHv)^_+KaFRvdJ
zjN@B3KJXdm6%7&2-I&0(PJ15p*Ix6OxaP2=Vf!b~MjTJe$j@sc`*+})bMQI2cSAIH
z0Qm`CL&Wj$zjm1Ge|W>#Qr(NRyMals?8hLqq|VrRo_kFSV^bhlY>6<om09~0`k5MV
zK?qVMZ4zmeaBY4(5&k!5M<2+h<2Yeqv*o<=lljrSiax0R#W3DZ5R%aW{WeQ6l4Y-J
zj6^%6Vj@~xA<FQl{lLX%`rmEN1I04vCqFTq$!SzdO)7k=NR5=6;$<flx`=td-yO?S
zCW@pNa-#JIbXD`E3t3lbDArEvze7o~r`#<%*>YQFv4(kF@I@YeOdX?M6Ay+m9ZN)b
zvz$4OY)pJ?V$_F<Uio7i$NXs9ZeC~LKIV5XvFSbzH+xXUEQg)m82>6sq#kYZSGk1=
zG&ARxNzrVA2VSH};%A`I_42Hfa$?L#5>lXy;T8Jme3pnn0WDBgEz)d$cCzUVNjlyl
z$rBigmXvt;HJaZqef1#Ux|SXz>sX6Gc?K;ES58@{H|_B-9m*RD<2D5KsjoJ?D|y^4
zpZMoZA!APz314)!5(ydlI*(3^wL+Z_;z1q+AN5}g`bczarQ#og6Vy&pikDujb?awB
z(DuH9x&B;fk~n&Rq$5R?1o|0Bv&VImW|ZWYi$(YO1N*=x&7-Zu6`z@<7@p*Tmq`t1
zK04<&#Hrk4F*;(EyFgm$F&fP}-9jfik|y#^d;wj@J%Y?`O8!Z<wnx!E-*_+Un+nvF
zIBh-RdpOSARTd{dYvocp$B;%r4kE?#nVUA+I10hd^V0TK!>@ipnS(eNeTFS~93R42
z5mfWGS1|5ACw1*Y3BsH?n>pu%d}aw;P$3BDL4FpwUqb8V^)q<S?=Lo98{8{<OZL{6
z2+WfQ!BIgL87B*m5BKA5I5YV&-$VsKPJ`Oaw$z|FMZ7XeG`YOFDSoXuJw7+4_<|a!
zv?eD?BaKxrR|P%<AZo1Pf6}m~vLp}<#o#?a3XohruBs!Sz$Bq`-}~&`Kx;5FM8U%1
z91?v4#2_us14NWFQ?zn5@f(b$l@0_pSo7&{5*9b#Cpd-89Z>@DRWPjr##)YSJHN`-
zLW&zZ`?NulRt8hcN0x^vM|<+5L;wcA#0Pt!zn1T1MHH&_$v&k|wI?$4to?Cmh2;ja
zWJO9o=+k<05dA$RfYSphH*Gu==L4HwSO6dJgxN`%oq;yS^$qMC0!Oq&Ua9Q$z4Oe*
zlU>Nfh|LrT<T9Y5BM_c&9qvf>bG8dW`oFZdE9T?POaVPkn)`}-g6hn#wEV{fV9f!C
zvdT}>mz~19FRm-$UWhnI;Am+VPhkC|P2W&*KcmC2oSCK_@@d^Y*yS2PFiAfv>TFGb
z*+TRKNd6CwYkt0Nv^z%I!UBsHZx|tlHUfrja;((}NS5rc@KEOk+qE4OYHPbPIuc$0
zvowQ60QBar>91PK!-JcVcE62SY7>30;x$X(x6&ALf)XS4K*Mhwg}5k+Y9IrEOXu79
zN{SCQ>!$iz)G*G2jAj`Rd_Dt=Qhz&FBzH*Z_pJX~z&K32tzi-AnMt*E+IzxRhi{$2
z_i|FXkcR=>O_3;1Z8XJ0wo6Ldg;4;sYZJ+j=HKmZlYCB20sK{^M66$|&Yg0GoP_Zc
zJ1+BtH5#x>_%i?q?t=@YEZUg`3M*03m}*dL$HdNFDv4Zj)+Nw9T=aE;vGU`?n2;Ve
zwu!`H#&W-8i!~x%SpkXk1nDgNS37fV7zEkqY1VJmJ9r-)y}eqiCy&u-FS62>F~mEK
ziRS5b?T}}x5{`vNP?_0QKU4paHky3U&C#{XU@G$=29pM<)2_lRsDvk-uDN#KYu#Gr
zXW04Il!+)R{^bAF!c>42rH}t88>Tl?dDgix3X2?tZV8A6@-gYRLP%x`r@#E)l;z`}
zOg7_}XdC`5TsM5pftX?-rS)p3f<Vbz08K{}begqw0}-oCAuQp*!?&0tUk;AN^+IX&
z6-%yK;(BL2aZ>#)UF0y*Y}{&TaLZ+M?$F0T!nIAB?7B9TSc*UkpMMwbj8~Ab^pmF`
zWjS7DmNJ1Jcg6aTMgj}e%>KVylzT`jOGbIq6rmbpylmg23MmTwIfi3&@(eFluwoE9
zez)g-HTan)y<UFx=;*W~HjAUt>l}qAlH_3dq?BFuHoC}5Nv0pr_ye22As4tw-iRY0
zNUF*bx3c@v-)V_u-^(9;(yDJbS^XMumHM@4k{A|IUbUkBR+Bu~+cNL43>JV#nT$`S
z3<o>1u!rhozYOsw4z{I&L3@@y=A*I$@$`j-2Ns3|jAxYVHg>?BLG5ru{@9m-*l2Kt
zG#5B3Rfa22aOjHjemB@8ZlX60cXKIaH8y$PpUKqfznUW>($-NdYE{sTR(BhTU7I5E
zlGzcIXqVIGHCZynM<B_oK4x>s=BV-Rnao7EQKt434)`d1LJ9&{Iy8T}+LTWl1%3Cj
z`sN6k%e4S8RXgo}oUz_3=1*;ASA<H%=?Y6S@5Ub<0W!^{J~Os>iv%@jd;?{`<1kT9
zjXHn17zC#=lkmP<9l3rNkw;h;bkk~#HJ5B6%cjo&8-VK+_QvdUO%(Ud5k=&gqF2;g
zA;^GlhfNNZU(J?(;r}iwq9a@c0xvcC`p0kbC}d`>S%kPZ<XsU}*q?4(b(aTDe4(;M
zCKq~)+<Pi`_`?J4bPr12K8>&<mBirGZ^69z-e-9Z6*PnxA&G1zJrAt|U6{a;%fcD>
z+E6|W4%_e;N$_-AD|>noh)X!xMdQTT3nKg}2`*3R5O2oq4hd8Ly{3_%@|We}jwp>s
z>8cxIz5`)??}B^SW<j`H#Y`#*4+HcZ0=-i!8>85w1~jev-W9a(G|qjocBVv;9KUcy
z+4$HbP$-0$S?+X+TQZ-7J4f~2`T$<cP?R}6F^ui0#f-&p#=RZn-&t;i3=XSd#vWFc
zysF4*)N)*L3RoiKNUm4}GU8aPf&Zd>*<4Dx8;Xwa&kwxz1o7Ki=c~8Pd|m^e4OHid
zA~Agko(-1>ikTtWo11%@3(WT)w$6%-N#Ee_b)@X#f^J-7MHoq{&1*2O!HIMMX18E{
z(VN@s7;A9>B1e62r6lR($G#T>phJ>$$%mvZNn<EN#bl^Yr{&K<vgEJ4Gl;xB_wDLD
zhj%j91+wC61jsaq&Rw!V_&Hd@;?6QLOeOi*hO`l8%blFm)r6*rhSxHG7xHrFi1eIX
z!9zQ}C5grc1`Q9U-^0V|TTcvm{Km?`9Mhkosx6af6LAF?--n{-k@+05D=#&OHNtuR
znG8$GpV;I2niH?MtG_jAUG7fniHg;UO^Hs`!F84_RKI%=s~phH$nHvyqNbG?XJfW}
z1q8T}3k;z$UBOYBhGJY*@r{X}^u?{n#;C{pkK^$ujV~JaH@{<F#D+-n{rbK*1zD*`
z7STxD*rd1Z=UDdt+OM9}TZcn}O^N+3?v%%6a-t#^Ii+^6YoeFS3k25J2#3t9ZOBFC
z_BDP)GJvbiQp+TQBD&fNQRVATkUU@YOMRxhG&6aat~e~1g()>YH*q<4NhoaJc|4^+
zFNE`*W=NO+o+xQ*<Kp&n@aCn0@O!(d;7`)n)LV{Un#6PlKUjX03w%Qv(m<}kxC02W
zk^5zi(%wAC-9WFD)|{~?q4u?|?KBa?<U`0Z2=G^Tp#gyM@&1ha#?;=cP!fXHNQ+L{
z9X+03JV@9U6o3}BL$_110lurRpVaKP;b9%CVpC;7`8a8_M;}-n&ww!E3r`hNAx@50
z_Dry{L^y)`lI;8ZCFbwJ*b+Q0l8<A$4AVE{g42VU4e-WKqYce26zvw=<FudcdQL$I
zmK27b7y;bl{nNdv{9S~sQ0)q{ChwU_JN9*tw_Mozg-zmM1N-cnzqE9Y)Oh~QnUJ?h
zG6jxUav)qvYsYaZMXIQe?hi>tSmzpF_Zh#(X()Dgk7~S+gwe!PJ(h`v&MC$%feJsz
zu^{RgT0EGxcO}Rc4Rotn%Ms7odBJOcB8R(o!zf#Ns8lr836EirlpOf3IJPz+2_&|~
z1Uer5yil{GnS(*KS40{VPw}XXpW1OJkC+g?b^Dt)W*yH1+12Dt!d&P!KAXI?#dpYC
zoBrJzJ_V4svQFq?b;#}zX(qE|W_Bm7I=G^wZ=zT*8p{xTfRx+k9cvE5azQHmc*APS
zo*MaM({Z$Wkoo$+@xz=N_5SVz_z|Cl!E?HqoD;o5RX#ZFS8#rHL=b&+B@SplK&MB@
z&ECk^=}e|_aK@8^iBXplN^$lplGM>lMG2_jH+rAlIFSZ67!D7U$p-)dhn1Yt)=31e
z1ts)Z_ydK>yooMQLy=mX<~dSTB9CE6kv*I*cb-9q#n>C<YKOzEOHe@Sud*YJf2`{5
z5?+(w5r^^KWXPJ2BZM;VbT#5K#m}V?efh>(1`s|`foG{!I>dp)2?MQlyf+X`ka&Ki
zKnLr{lgCA8Ln<ZMoXYUJ<7&e{CCZQx^MJ=LaA#J7vDmEaSr6&H@$u4oG?|uc75Kx&
z*x&*~lvSo^p0f`A;}L0|7+83J4bua3+;j~Tap$z0SgwQBZx@2yeh5vL*~>)ELsK~1
zXx5~IYYZ!O%p4kH>Q&%(pDkSGEaQCO#o@z^Z9lcLz1KXnl}Lho?kO@)5x(5r?JU{;
zm#!sDB?g{vY;|r2>&YxPp^tl>_vpF})uffpCi$tlXWy&f<@pZul4Yu1Nlr1#n*z4=
zMn;|2IUgw_>qj1_fRDa&cg3}j|7@CWdYdUo4H3*RrJ)=sdg)^AhfM}Pp2R93lD~)#
zZHVN+aHg|Js*5>bKh>6EI|~NI?E$jym;ANOk+>;Z2)duc#5m$J(W{?c_3(0v>wfXw
z^HcOla>uZS`l2w%N;)K7j@&K%;kQhbGZ`Ct1Z&AG7N-lNeh~%Z$bQa<<85D2uYEvq
z%n(P6X7_J7HBsJ!0-AaD-;%Iu1`k6RHmyDv#d;1m*OPDdUTNtFnrfr?O#5`yN6Nm{
z#ic1SKaQ^eTxXibWtllQM|nB_+fLzci>E@}X8fddaO@p1{;|{RgZ8rOZm_dK|94<7
zri@fqdxcVf{bC)2f0qvSeESC%YW4=^xu~Q1(`>g@Q|?kWuwQ?ejV7OM?rO1XW1**E
z+%*?0xDbeuhYU#>4tL2tkLhfVr>0kpkU>iIGkr2MnO*?^3NW)n3Nvx6wT-vWoyGbL
zBEe8K_(MlyQ`^rK*0JT(_u>F3$gBjCWSmR-zAB~)C(kQwvZHnfe=)yo4?{vNofmvA
zB7#&t^@3^0yBc@sU~$nGBUn<xwNX!Egz&T3xMUU!gqr#tHr+XrL(SCaUlZS^--ZA?
zh8*=%jh;tXGh8G79~Y0QkD2`kgjJ94IN99}${Zo#fEmf2^3F8Vrt(M6d6DU6CU8VU
zl?buE@w1dc@H2D9W>@5PIfeb74{AWtwk7{em>AED?GOA~yrIkCG@0e^I`!Xu$vzf)
z<+1Q;_x`B5gw^3)z6Opo=qu-|kg!?3G$oW?Z_G-aw%OQH|45>xoKgBgGHeL^P7Jm<
z8y=MUhD5Gg=yB1l6%ZsoetJV*aVG%lS!_Alu<1mtgmT><Uu%1}Wh18@N5q>YYu;e`
zlP%Qwnpx1`6Y@U78PFhwbYHI<rzXPx3@ZupYd$|0Etb7?W1CiMn(+FEMXMK7I%}x4
z!=8x2;iptWr#DqT)d?p}&iqc9@<0HQIo5vIdbU4M@~y>HkXx9BH8%C!jEEac$3-WX
zlp{rx^s{!;*(}2RHYvH&&y&!K4b<>cst7n<2vhufhZU}ir7t3sRRB)B(LI{7H^@gC
zMf4xCh=q>e(o~%t0?OQuH|~%`*-Uuy?&T~$U$ApCPJ}2A9y<#t0ktzcL^q@AQaLbt
zhpU^XuLgWuE=#8S3(E9bNCb#$Hsgx+@Hf@F&XYn@Y1BLe4TVz{Xn}s!871&^=d6FD
zRn~(B{H*SO{N(>pbidtxWG~BgLF<~Of*PeMp&hjMr#8AEcHRN0`e0I?_5g|FMqtY~
zeCp(BWMqbKUwdbQzv;TQ_s@<hOnr6V2-~Lt=CC5awP=d3G;Ab)cJGy`R++9T|FeK8
zN-&(CYVnf0M7z^U6Ye>>Sxjweft^ibi%XUO>bvPz$M;fA7l53fl!DP95L1gvUH6EU
zLk1v5A2M1k2b||0RU>hl>7zCpI1+JQeS0OKq+$WB)bc<$Id5jeMF&*~Y7z-LV=NG7
zNgWyo16D+bHM*;@9UM&uam?V><v6j?!|_){(%W#IZxnu^=HGMO9;dNkjfJid2Z*us
z7fg>KZ-E_P3>3m4&t?~8FOH<AbeMUBx$`nh_kILj+ozp!@d?Tk9^jOY!ifzc6FQ~J
z-k99tVa`lRzM8H%ijCr#3z9EyK%A`3$s=SIwQP(ZtBQ#-sqWq*Y~gv%;uRl<p|>FC
zr}tZ}4ZpR+s^5O5DJWNBy3J};H|lJyu9bfb1Y!oOe=!ysbe3G$@|Xp9x@MGJT?c_D
znaHXa<alr<`ST%&p>!cu?+rwt82Dq^z-M&FTy0vF2zd+thR`-YE?^yl19tqF{e6p7
zl!s18X?Y0|KKrVZGFO`>RI!tH>Eryg=ta(g5;m%bp8^oV6|Njwg<N<?4uw&H+E#oh
z%)0nYR8)oQ3v1icgGVcXr;_C?9$6uCU)2*jkI;^euGa;7<gAu;QP_~$H`nY2q~p&V
zq}~VB;>{a^uXH8mB@~`7)20Vc3VDMG2t-`NU8Gd2nWpsQ(v^Rq$!H%GkE3-brl*|h
zDGS}Ryb^!@@{oXpDIUbuVqqTKhEqwP*RC+PFD>EmlgFSMS#i?+y1`<5qMI;DM_~GF
z)s6V2SBh~@Xf+1Om$(~49Dn@~hPBOUwpY=yxJB6r*Y{s+h)Q8`_MH)rP6!0$Zz>Tx
zGrHc{a5D!+yDM-mOFx5FI|o!d^<?o{@pRh`-zoDt*tNs=3NBKc<27OVhJSh)aeYT#
z-ZM*BC!$)qH9)Z2iaw61=)FX@c^aQNfbD{P_iqmEy(kyGk-f{bj}va73+xN{XFL)t
z&GM6aa)f`I95ei#dMmi42g{Ud@HmDd@C_QE!PWhE_Tz1a$|6u0p$+F;vKthgLexd;
zjrrArqZo{2<%r8gu+=Mj%8{%f4h!D*7q}Z>Ye(4Gzjky_B=viIS~ue=1vOPRpwIE!
z=}Y9iI)EO+<!N#K0vekGC1CQPMgvuTm9BlhpS3>I^rRvOl=*ipH=$$vN|GZ*w+I#f
zwdhs5bD(hf=B^%gPSW7oIM*we4L`0BD7ZTeoh!iglq?YRwM(Hfz@A%v-SEq`=-!{l
zM7W!;<n-ctoFzu|T9ES>jGLx}_kpTBNPjmas|2Zh1?9$41q(m7ZzhY)ztMnbEAp-s
zaC&<GE!4CJOIgszkdPEOXCYqCHjUig>@zn-2R*JVzQ3c1#bjIzQ^j%pWmDP6g@5kt
zKQ<|;?pdc;rF3w!U5?~jv_Ee|xa_mSs@9)1aT|{mdJV$rlmu1sMfZr_eDWF~3yZ`D
zeT*7J{(?7GiMKXEXAddD7HkY4o`gW7(s|Ikk4n4Tr1Iwp?bct!tyYj7%S}HG4r&R_
zCcI+2(<Z$u$<OSPq{K3#GikNiDH<`Qohol8?z<6DC@@RAAX@f(Up8hMC!q0Dr>du{
z5-(=98cDx1s%12&N)&TbC(IW+^v#yW&@33MU79Plk@nv6sgi!SRSX{LA#Klzc9_kX
z`VvB5X(DyaL<1B)VxBFvPX-4+;>p)eiLud*7U)2+%H=N`WvRw{`_$-lkGB-5qB>3d
zKYuhH%9FNGUXcAv>L>koXs3tMa>lq$q#2N)?jM1^rhy^ph?8a2cW~%kyd>)3g>`fG
z9f`{w2e?oq1+4s8!2Xe&-G&>V$0I1?lG<@t%n!b37fXcGd^R&S_GhrBY-d)bGiN`a
zcz%wnNE&Yie2CvMte7SfH*2XY#Fx>lq3wTPH#ENN0@y{N=_!-)M-6?o>8|!wudP%M
z`HtHHe%1_5t&oi3<`XznW2IFsEh|D5%b&cBvTW2Dn-7H-M@no?5oe5Nj=mcXg|rH+
z4F<>y{vXhbAD#+j0C7N$ziwo6baG{3Z3<;>WN%_>3NbS=HXtw{Z(?c+JUj|7Ol59o
zbZ9XkH#0d3FHB`_XLM*XAT&8SF*OP=Ol59obZ9dmFbXeBWo~D5XdpE(Gc+<FARr(h
z3NJ=!Y;<LEATLI2VRU6gWn*t-WePq%3UhRFWnpa!c-nloV{oS3)-4<x9Xq*W+qP}1
z!#lRsv2EM7?T)RE?WAL#XYXC_Ip4SI)Tvc<)zp|{&9T;xb&(S*snQFZ+W$0@u(xxe
zXJ%mH0jS!$I+>V>+MAjwIoZ>zm|3{m7&`&}Nx^XgOo1jYfS+a-Ksz|be<2i%ZOs6T
zGRkVAYRa_#cc7`6`G0Dq?ab`~T>oid>gw=6ZQRV9oPqXs0IGi&r~yu9#-{dmHlF{1
zDcHLJP0RpP|LF_>Is=RWE>6a#X12yo)&P5Rz`wT+Q1x_nF|&0BNZXm%J2}`p8M~O7
z0)Tb^7fUmMx*hOeo~p|~!GGX(ri}JZ0Q-M3P5={oS34IcpqVqnfAfmkJ9s()Ei7FC
zRR70hVPa;Y``4G{|3S+D5Vo-asQe4y3{Ww1Hgj_OC-~p}C5&x>HlF_$?|)|dmu!H4
zbo=j8ptA(f!^~6(=wf2|&+M*d{|QUG7~23%gzYSB%m7UP38~wenmO43?aY+yoq;a@
zl9ZmA>A!PomOvA0J2Piz0Q-MZW_G6k#VG#IIHo{5i+^cmXKL(Z`o9DJPAeG$?OfD6
z9sW1(zs~mGKJ)+a<&FOl)&ro+#K6SF{7>OO>whf$|DscMak95I(*&BjSpI)Iiip^I
z0KDm$xmf}9Z0u|R&VLMMCJrXw|3BQs)yc`s&gI{i>Yq{m$2SN5ql=lDhnWf7`ii{?
zZ?IK*R)$MGR^I58m{vP6q&TblNcXr^xZ@d+Ml0y2!3rgb4BTH!`gnMe+y|Tz3OBDe
zS)Z-W46%csvhg9raMX4CCUBe;p6q~XVw7A#rEa)aOUI+~1fn)~%P*l@=#i&$XbmPK
z`o<WrMmmgV<!;Pb0?B?pOCt%mzPr1yyPW{`GJsHgnrxC*2@F{G)$lg|WgnAP?pTRm
zvCS`VP_`jSKvgf7Y6bL<Q?TqIuY$t}iVvt)BW4fmz9|t*b==s#Sxy`{!z*>eXdb>E
zMAN&s%44Wx&IaVpfajtT*M@r`1=-p<v`aG99OZPDK#F0c7P@}>98d#7*p_XTq?BB9
zBC*g52UXTX{_4hfn24y+*xzZOgN==K<x|hLFemM7xl`JCmNDwDR4%nxq`TB^%K$>`
z2PS<q*}Pv%y+t;slP1#!F`|lC8<_*Oj4Lbz(viY~nT*xy^$bAiHqik9{xf<p+{@v0
zeE<@Tv>o$jE0D;n8m1@rHp%ThVtsrjn~4|_g_it_scn{9cd1NJ`ENWHpCvs|U4;2t
z<1U(pm#p!&d8_CI_##as6u`^CdgWhIy4aUrDb5{X_@-UvW!(i@tdKcybN{W57|LGo
zlu8h+@y#DH+#%xkSXq89EwM5gq-Kplhbh&{!NF*VGV4af){86+B0E-|3u?s4JmuPQ
zUWYt9U-qn0=F?ADEMkxgzA|MecH>k}x;4{(<FNifMG>x<RH5%%r-PYz3yv78AC^{}
zNp?r~qORGF4C@z)(yWhA8Y?Swk{;CE?I^2soc!0H%b={{@4JP1D*9qin&#>S%}6(7
ztDbb}@heYTeRL*Wc@wKW4(W_!6rG{vaE^Y;N(n+~Pz5I=_lw>a@^MvxCjoKZ>fJa@
zY&OlqXPFK)`1nqY?t2o)BS)~|fqSSKz~Q`yC;ri!aw6W0^#Of!3Lal%D|5j$`L<CD
zS^F$7J<v$hR?hmcPAgCE$PJ)01iwnp;x68f`;G&?QZnX5!s*6dEJ~+zfUBcN)%Dw$
z9}vy-ao^nMmb%j1q|1qWQo%X3A<e?}p0uZq^s=x8qeDNVBo~36M#E-CsqLNQ!j>S8
zi(?wC0>NiNUXOlBl%mS4Y{~C$;nQ(Drj!lS85HJF>$Qv2r)%$5%*2G9oz={&b!31p
zUo(ssg%*WibA@yq?h<hHtX{L*nP1?m@A;w-_f!H{2qNfKXqW{3g;AORy`bzpZ8cq%
z`Q<Xt*r#T#A*}{(v%lT^7+{&A3;aQ_Q!Se9*&uK1&}+klx)~Sm@uz0buPN;}fX4+&
z=D<<lHWaLK$u*p=gSw<Bf!e3i%a|<JO^=ucAY?(se4?1?lVf8STA9NKvz~f9$tq34
z#cV}iGf{IA{B!;ryj%Cd>GhAn>s+%l>*I#{VR5x6nQIe_5zX11$qJPIP4c~!$MQ)F
z(B0BqvT2Edj6G8CLkg$PbEqW>{`ydB+*cz-HI%4`TgkjdZF@L2?~53v;!{J{h0Ea1
zAV4_uY9AFVGIhA>#$f5*NlL4%LA*C4ZNFmr<)>B=G726(i4lLM@tzum1wQyshGH1a
zB7BwdjC2M^{u?so4_hjA;?p_5hh~gaWvvV*N(_vTpT!(XmznpHg(>0|(KRs**@T>@
zOT(3ZOLHsWt82l)NcLP(_z)fK=&752kPit}+Besr_qPekTf!8?w!y0Hud?Nv1vF}q
z((Y%dsME(szr1^S31loV&65Qbi)mX4PjCjYE9lyOZdE48gA9eUyOHaF$`4nm^?n_5
zh;0#uN>2AH0eUkz%C?{!Q^{l?DjdC#pU)9ri;7(+ktvMu5BBSxFpTe^Q)@TH7)o$P
zH*Rtrg!l}H`@Vrig(gYw*%SqhWVFO}Ui)^L=5Wrn327gXeelMvU>;kkCx`e39o}V-
zcwIC_5AP?{3=m%y9jvrd<bsT$RSEG5MpJx&ILL4Pa5<zOi<S%{!73Kamh0Dk=ODWY
zzm!~-HKAv(#cV$3wpV&B)X>Ko!RD1l;0{S1YP>cp6`XaxJPbgK-XjZ}D-PH=MHGP&
zq_jble~H?V=d27r_<W^Ud4-orLfvg>O&3|2^C>w&w|1ZBt|ActdvEVXZdPQsWBJN0
zU0;qX9aJ)#ZSEwmQYYtDd(}?d0n|xHM?lLTx6+BR={KOiOBL)yI|f*`SubXX@<^b}
zpaI`O46nRm7@r@O{tAQ<T=z8Od;VGe(cRktKKUkf4aE@e44BzrVxY-bO!r?ZU)aWJ
zgJmt)kP&QiqMAel2Tn{No@{*uo$z2Kq8n662CfDJkZRzh!S(<d%c}s%KUjzo1YO>0
zoB=z(!QrU}4!5yENsjrtLd?c5+d0jWK*!ZIASAQiB$`VUtd#<rL+^qUl$YKLO~L4v
z+*DEXn*Rzv)#Vf{HnY!aV;j8pUjZuLRR8S$dI=@gm^0h9h*@T@0v9=vxzP@LEqkA4
zvV8y+!ot<AC)J(~?cRT90?R2W^wy!t*O=xz68gK+3j`Io*u!9?KZ^@jSIGtO<`&Rd
zZzSmvErs6HD?W(kb2X&)M@=+sskZc-R@*5zI2#<DJgMECnxr&bO{LJPlbS{nvu0ZW
zCxlMKp2gV$#@rM5@4L{zri$m6UrEM`O~+lZ>*^{S(_5~hK_jPTa2PD@iWZRMBt#=|
z(+B0&vgKy`0m{cxQJ=qbP*n5an@;69s&MyIkB0EE=OKLd5`K2q^z8UD6>~<Pw?$<s
zcP#ds7t0*y?i1o8KMxX7j2mSWVz)1F-z;qtyNd|~^#3s=s*h!!q;I`1$L+ieYDuF#
zXb_%oqCCJFoX;+`kF6aUq|LYYRAtw9|2;5=A2hyl;#dN~4RePE$j@0NT5SE26OSd3
z!QIKK59VJF3xE?e_Ee3Bqg1>*NPc`Kj@9e%u{1fG#IK;&L1A26J9L;dN_Z+*zz5xM
zM8v;4G!C}^W8Y}fnDX^bhW#Nh@U=l5@^M05H`=Sz=f`Gd?WIJ}Y$;nsy9K3XpfnUU
zAWWR26*~(u5ac4@r;A=JcH^icz&6;XyVl$$)K^UF$Xtbt&*w20DCN-C8Z8rJZt8x%
zO6dmBF2e&UF5SDoY=AKubG8Cyyu_9NITbp*2eKtxP60%$R%M!7Sz#|1%d-`RKPi>I
z{Q5POLl@T*S?&Xph4Ydt=(GCS9<<X?=Bm#8ij=lRu)M_ykcq`PPdoW|O5&Ul$8=|*
zx6&Jy#%p@^uimKtE_YCN^L@*ku8oL|Z+o^h7-Eo5v6Qk{g3w|Ax}m!(J5*P@s{)4B
zR!AHmTbT7WLzEg)MEM*$y#VjfJPY%C5i4cX4kaETkEv%Ko)|3h7Db499wnWOYUmdH
zKwX_qG^AEjQn$!@1o(7mJp{f7oqPw%zt2KUW(CW;NQtbuX5ZJ=%>a{FR@(+%T*6IJ
zI2a;pHbl8N4{1wDD<;zK_HfSc)#t{$>8~kacWV<|pc-&WakL_kc!Wlmt$WcxMbqq4
zM7-93mt4f@spyr#YX3G@9eSxo>DCeHuhCpWifZWN43l8&gB-ZO<m&;f(tE~@P;)oc
zh22tWd3BPG#~B0ya!vQ(G~tEV4vvy%9Qx2}%iaCx4A#kzW#(kByu7=z?~iUANeeQG
zITDPTJ{WbF^3v7JJ)shZ^W(DKi{(IrG$goG?!+T}j*8Mm(#DHvfqDy^klFg32hhy?
zgmUY-o#;tP&trQ^)TX;^gBhN=MV4d(*EnmciL{FID;jLLuV(MC^1Iy5;E&TbC_64H
zKYO2mS=Tdbcf5QaeVc8(8}I=iJ=*B*P4gLST#&|PsDtTBF`941gAbzEx~io4jey4=
zkC(+xl}B6tiwRCSHGT%xyxu`Ac4)tSbgX?GnIpS{+7D7T0`%M2I)Wi{Na>s>4I0tY
z|1SHvD{?4Nxshzvs^1DX&wZ0w9pX0FP2#f20dfOl*Q|td^?uAfg1)toH;X2!AQu$?
z)eeUl$H4=CoAVvrVGEkKamv+bWD-N{#ENcUIQs#2u7&GTDNeBFe*cDOOM-w5*9}`V
zg|BohEK_73ISoo6mbL$xgz<;hsQQC(s>J$_cWlyPqz)l?d{&>|C}S9bJPCAIA4B2&
z=(baLJ)|-|V1J}iIz^k&I5bP>_dMMRm05EnoUEm00ML>e-IfqFrOxAS3@-3NqLp0d
zGJ(fSsU0O*_(9x%RZP-JQIwM(IeD*$lC9Fpn6=#I8%GJ9j=82D#_&+<1<fEQJ=|(<
zKHxWeXLV`nqf|Wcr<E(v-zZ3PdQF&yQDR=w6&iVb2!%Fr|6ax@_a-901$IBYr=t&H
z+HnZwH_2Ha8(#o#sX||iNOF@`a5nVn1#tOD7^I6WW)O^wXQ^4~Np(zUaT|X_XCN_7
zJJa+pRWJBEd>P7FHIzK06tNWGw%UYRU6C8sr4f5!Y<WvkyH1$^*-gVTc=MwRL_(4a
zc=v*BaNCW%WSd$sdLfOhtY?gYw+<S?VEIKc+ObD5`BrTXw-Bt-faS^jB-w*;GOh@n
z;izXkGdv>fd7O2NTGanZoj(+7G=hkT^!}&L%d&3wqPp#(nX|YWOCj8zJs&4xP55Vw
z+hBMe_ivk}_C>=hU~9trVhKKEAZ;BEY*Yi$7Ra>gyo>t=*8rWlrAQKp3~)-*epfq{
zGIcX#9P-7vRqLq9hL`}@iryLJr?N-$R;t;Bh!gMol{lSPnHueJCekOm`$lPl#ml86
zLy%@iFnEegu?!v|{jJ)GDoHO7&#6Ja**`nnw<Zal5h3?k8nXkdhdFRgCvWJyUpgkE
zq=(MTw9lsqC-+76)Z3UN3eM$fGb?7fS9}VIw5Cp8NC+gume_+;E6G<rXnLX@t~LN)
z^WDtubL<)P5O1pJT-ez{v}k1T9<x}Dlfa?C9qKTOBH=U6c~7!M?qT3|RKV6IN;jY@
zj69>UApT~&Gxz`fjDthBwKgTC1HbMUt55Y4!79OpQ%>-O&~Fk-dL@z=d=XjGJg`=2
z&=68g);v{B6sQ|OXInrE6DRN@*mr_Zo~`^km?1IhL(|`*8rr{}q<c$_vIwnRZb{De
zBXXS+&8?u5QuE03E1AA!qNT%EfG^q>I@a85(>BdKPVb1(_cvxhW(u5}NYP_cE+fG#
z^``|r7}}M2n}i%!SEJpP!kl%z34AM5gIf_FRn2}$FZ!!RF2e-5M9(B)2@Fi&O|vfb
zWYM}NZCZLX(k^4nerEf(Y3Y@|pW!kA5wWr9yZQ+M(y>2bjCNw~>}l*VF((GxP1(4{
zM{f+2O}u98H(hNcky3W9_eCCA=4(v(sAr`;zde4KmNOcuM>py+!qQpodH_N4wkXBl
zT~JP0C4H#!_G@!t;8wF~b^IJX$LWX^PA_Obm&0kvat_B={vGEvoS07Q8%`$?wU6%b
z+>G&YH6e-QYkkl7xtGn;fj33<&T)u-oNoQb^>r3!u3a<`TUwGIS;N|eXKcc2%qFUq
zRvasf$a4A{?roesyu*%Sk_*CT%i#RYmQ`m7^zJ?u<@4tz61l~oG_MyJ8B*ACokABO
zb1~K^hNC}&MsO#I-lBEY6Xi&bI??ekU*<SmMCZZKZJF73(9&WVToyD)*pMXt-_e#o
zw^SYg9#$3xE-vcCCd)rC!gfngkQ-PGX#G>lB3{1t67OiN^ER=}u3<ss=l%z3WNjHq
z?L?WVu}Zya3<{<Ri@yd5(&)e4&6wklXyRbb=z9^O0||gNR1$cJe@n*n7baAO;=>g)
zy~@E!Vd33>v&cDv<c^uxNfa#LA<)VHG&+WrH)5y8<mkHoWIdF+@wcO$LJ!2p^^sC$
z)1YVvMaRW4hvOw!Cu8f$GjW(v{Ka?DO(wF6y6yaj8)qi(_V4+*&)RsC^c!YP{X*E4
zB_Or+kENlmUwd<XH@yfay5~E!dMI69hfFo;Iy0OYuHsf^Hobi_n8IwJ9#j}Y3gQwb
z&Z9tvQSvn-d<u*xIc33e&%1KwMr-iLHQC*VdZjzkYE8Q6NsT+)61Om|Qr~qvd&I`y
z(0RTd0bWrdI;T|CLhhZTBjLW+4y5}d%YQC?KR$DAR6pcF3w1(i<IifjWr5?b_L_ku
z1g~JJ(4WSaBJpkl@|KDup1WdaRHI%9?9&;9FRwaM$>>BJ4N8QMWUMp6#Ijbafy^HD
zS1|>vTO4~MjY73WX=a<D2HA||vif1!hH?W;BTT{Ki<7xN1<*#9SkAzui;BV|4wF&p
z!2LD1ME-1%pSXO4?X%8_QQYNWPhzaly6~ssm2J;^Tdw<vpky~QwBZdf#US>)_2bj|
zN<+{_bz;s|qQ&k%XWpg9!SEG*J3)zxl{cv)d6MnbG5cG|hJ+|_!Y4V-h&3-RT}Uq%
zv*>MwDy<kbQw?BlF=#_AbV|0mF1mS0#>4J-ABwBzc*`3C#%xrRW0WhZ`-RsM@?VT<
zz0s5+#vJT49TgfVe&C!PD<hrFTGWX0h*aV8CF{yTxG6pnYq?=k^G0iAXUg6}X#X(z
z1@o?)K%QF52Pw9+{D8@<cL+Q7R41#&IR6a@f@<I3!}tBgE;olwy{j69D}tt8>qH(7
zK67hj72c{;>?K+1H>vLZExmaf!Puq`as$lGCeiw$kd4#OpDbLDpofJHIewzeQv57q
z{k?YT(RQ79mLx}R*`%?*`cB7AuImuJ!{5p*NnV&`3DY``oIHsWg;C5ONNh7l!#(7G
zVd~vPaFESuRc2#Ov`1g6%t@b4XL9aiOxJ?uJW$i<_-lk%A%c_D!B=wwhISa<j*)D)
z@yHlGCd`x(^uQ_I_JYb6cO=r@m@(VLa$?rNsA(g<hm;R6Vuh1(s{9VSg*+!Kf%nEr
zP0ZFKj@b;7xpqFs|M?Ya+5fG3?|qnEh?;rU*pJuR@hb>e6(KdCEKwZN!vfovW1C;i
zwZ}%~S8z)kGg3@E?5Iq_qA-6*ZQMf$Mj|hoS@tV5h*U>Mk#69e-&CiD<RN@P4D`JQ
zE<Zio@&paJ>GcX3${6)3C7G7p;Cg0}_cX>%UO2lmx5o;!+AIP7Kx=)uwEP<Wxbkjb
zC?QfQb-Z8@DaPqN%b<c)&RvDO?y2QC$GrWH5YfYOR+^^*WKQ@5o_UvwuoltBVDTdw
z{nIXJvvyW$F@Ck;zEQFwo%CwHrBNmlzh)60nuE@z6mERU#tml-h44pbP7sc;=`5?o
zUANjQO=qm$@h6|3fA0~^ACwTR;p3RKtimyKKj==q-wWL6V!??ahJ=Uo!6o(KNiF*l
zdPYBV=d0v(o@W10pLu1|2Uemhaw3k$B0)R;Jbu3qdbea5#Z%Hppva(=mLAphcER*u
z-teMo84SHBrwb=Xf!zK19Um7&yFcSv-90Y=0gD66K|5<t#8u+v|JBzo#^LvQb;4=E
z?!z#PLDff_gIvDL#;3sL-M{yB=fg_$6qZ15G#X<55faZ_=O7Uye|XLj+e&`hkMlR5
z?j@^oJ)<0uQ*8yr0%h!j!u6yKLsN!W2LXdZv!9M&(sjtzUpZkoW#6uAYO9T`qp-n+
zftzyHy%fC^UW?r+Q1*b9e9>c%9|Q4Xg=1wQ>+EAKxxGikd&iLqw)=)qB$&P9D3B~P
z3|v1+^)eytpboJY_np@zeAXH7?A=-!9+iHJ7*80m9&C{5^FlGd*!p?OXMzYiM6s!n
zC*H*W*rUj+R3aFBOyT}FSFzlvLiRevF1C>#wWO<UC}Lvy7G8uXEn}>$SrfsYpeuBM
zo?i3U_%9_+78V!h%MpyM9$t&523LvB>CK+T`KGs}IpNORb#$N>^mLqvvPccRwj3$=
z=Zvj&Pg>6rv^}LDhT8m(df;kU26lwA{mEoMyDm=LdWej4L8(}Bm}E0)acRGzbjcB=
z|LDaz$+T-tI!(j3A0U|RmDCM_9!+~Wc4IuMNy|nvB&AH^*UMqtW0R1{5yDgGJ^w`S
zgpTCRhTfKvQ1+i9PzwS)S&rOllyd-)_GUprrsjg%5b{&C>U0>lG6+%GwUqu#(ph`O
zyh*!v?}lmjA$kn+(%f2gnrJ$c@PVs^^PtGIUfPS_?m5<AWEU}Qt!?Bj-R>^RY0al<
zlKcI}__?HxFcuY@p-hlUB(?VAl)sEM%E_iGrlVSc5Y@un$*8G5x(>*zx&bT}hJ!fG
zSObsekBXa-W!*Ta9l57hP?D5t@yFU623#<o3Wo(-L<!=E*<G_C#!S?>TNcb$CGO`k
z0#Z9nghUHhJWFpxeRC(q2fcZW8$y-(j<+?%;gSx|%jrD>dIx=-c|{wCWP>74ei(PS
z+;mjtx;eGIi%UqpynQ->N#7y5wgkyef6w~|YSvcXufcD$)_5nse)5J~>0jzSKG;qE
z5&7|6ivHnV6zc{<u%|s`G0uX)cYGW0W^7-Nw|F`I;V6cO{E-uDTkK?l2bMdLAqNU{
z*RjpY(1GrNVpW7lA8K`7udW3KK91(|sz=XK<+(2q;of6hN)I}+E7S+^?mdC5Z%VN!
z_Kx*EX^xt%@Yi8&Zcxmzs4?kIacyVQjS#-K*&o(V3CT{&!}P+HAO&$zX0x1+Vl|@D
zX)+cB<`?3)21IUD3$@5;y_px3jiFA2LxdZAZFfVohaYIK52eJ|%8EI2G!ZUEVPO@)
z%m}eqqazaFG1uzuMO#5qs|W@BseJ;%FwRB|tY~dLrS06=EG91uDTE=Lv1&TsoQ_PA
z88d_;!Xy;VQ1S}AsaGR`;0jxuKmYt`-b;abUbq3l$b#PVBHYU@x-HH0OxxeHBs#6q
z$3lkNllo@%omN<Cr}nx3GbrhQR`ZUg><>YedqK?IM;tH*ktOfG>?y=O_{z>1wbX$1
zN$@LtD7ZR~ai@I;bIQD1TaL&Qu-G{T0<t0z&ab!9^_Eg7!Qyr;qpJBylzelg%+Jw!
zea8Z6EK7)*1@2d?X|qyZg@sn6kVZe|hswDVzxuiD+rEKF8T?jJ_-6@y&MIeSwTywD
zCOL;}t6)pNq5e^zVs0DpZ4!4|YCYY6ibAX9IlLFc(b(WBS|&_Xz!G9ABFa`%v(laD
z8Oiu?AB*0%j^*2Y_e_NHTYj)85pF2o-jzz@c+Q>ElG8hM16@<94bd4W;Vj#$_r?fu
z^s(1SVB|tElaXkPMdPeY8mfN+j|ua=yzB9knmQ8C&*@~me9lZg(AQt~*GeU-=Yn2j
zy6Zbnnd3olo}=vf<QP4s=GJwa`&BsE4%)P0bb{9w5Z!>KRk4n=FGe<2=kIOqJ>&&7
z>OQ#kBL2uQPTvksA@UEpz&_QB(Jrb)$;+Mk>PLJ5E8FEOfwp?QK+1r1E4^zQ79aoz
zbNRXEhgTNeH?UN)>+HAWRV99n^x%8-W?Fu~kpM9pK9h`~6X!~B=P3_hy0g6DdrwMv
zx8aZzJTK^7JKmoOegaY($`tH1W0m{;_u)S}bK$&AzYqfFE2%{b)bG}CyuEb|pWhhz
zql$DhnQjc&h>jKhR!b6x?mbR*bheC!!OK(^S*8qu@?V<JKy`oL>-&mbchu8~xYA&q
z-OpStt(2{u!{%I&y3~o?&7;B=3(zlb9U2+;BT>ilX~8QT8UQm%kX-2Buf-Dbj!)mh
zydnd`oQZNEe8E09-^A46z}KTcAWk#ok0EM2rKv5-%c`f+OB_g)Jd(f3OAjJucrh{@
zArJJYBxD>qKQ?D$!gC;nBl>#O)rqg9AeTQmU8!e(Ah-3c=h4?J*8~P_WzUb}C0kZq
z@@yb+xV`g8^>FX0>o69Lf)y1#-PW81ZD>-Dcns!97FP--??$(R)pMwyZVB&pFjWl1
zoy_zDvl{4FA7Qyl=U%PbF;L%~vEcold<|5zy5LqlCV7~^tDo!{eSgUD8r>mOF|PC6
z66}ME{6JBXUho*60(ozcL!2x$z3yPoqZM@4sM-h7X!;|<d$VbuSq(7b{r!y)WLYsM
zg7l)P`l=xM%Mo11wtM$aO9f0UuOFVBdv+s}AeOPn<dcB=S#faWYfPR5P)=F)Wohgy
z5?P4MD+-=Y{UR`Ys?S#j?)eCPw&#BE;(<25cRpC^2n<e<jtFG@Og0|zt$~lj4=Rx_
zn=2y!sEbm=s8BMOe7wBaXjO?pVlQ(Z!V`TD(HWfPqeokDjY=y8TW|6$&+dzh@$kLt
z<hq9*!IYohMmf%wy1A#B1T4V#ah*(RM7&A7!6(<fFHGMdY&@>fN^;F}B~B-zU>%R-
zmx9>9+RF3wE*V?Na;oV!v4uaOWMuqx)`|r6m~4gn@7s!2oACjkkwGNa=f9D*$+IF;
z0rl$CZ`{bD8LVGc?P15xF=D!)v3yi%p$$(6JpmH#rfJ?j$>ftij`Lisk>q&pi#q(k
zQSMWJEZH`HblB0-NJy?SP9spbZ@gAfF}GrBIWnU^=g7C=OEPXm;7<Nc-nH6O=1M{)
zF)!Jr*ejI=xh^XC%xSPfpQgmZc6ZWDxQlk*qX`pT7r96`x3;7#@|{aiUG6XAO~fX)
zz?T1GSYVI)#ZoL=Gc%3@4rpX|GbnGT<?ykDL{JHGRcNL?wqDiNY~&+Zt!v?2TfZ$7
z66)8jYgYIj3QJyQ7K{E1pX(Y!tbM0>ni+%fY8GO|Vvy}u4lQ?GiRUYksw_r18%J-A
zmM#fbtRF{dENKAO6uN6bbcq$sj0%CmLPExcGrxA6?`&06Z{hpz&aTYD+-{mGye;Xw
zezVY_4j(~k`}f^>QL{zQ-4nhVJ-Yhu0qTc7Hdv-qb=4-ywcNkawc>pAnsq5DTk*Rd
z3Y#Ot@>?fVo9ZQ*bn`LjWAAJgW@r4{k1n^oyz_C4ZDt_^Va(74-Ak3}4@Iw!q8bO5
zR<<q0Kf{a}QENI|2bG5I^V+Q6mn6~pjGA_f`-<n&2h|M!9=wIj;wAl!Vyu`1Z|$2I
z^l9;fxSq?BnxDo82-A1@t8?}2hx7~G$CkKA!%ExVa9|EyJ#+=^_}n;F@0POx4Wq4R
z11w0Na`USR4?Pofka~Og9v>(`%O+n4`u@zdIe`lZkD&*<`ATXa^KwHD%g#)j440Q4
z(dd**5>WZ$L$q$!Q8#1<a-8K}Mi4ET#)(FGe>L=eyt{p}C<_cD&PTyN34r_prF22W
z;~{mgyAa8K_H2Uu#gD7){#JSZ1xJNeq#X2@OZVG{s3iKxGF2{35oQfw8x8l<P=VSn
zV7&E?noFVqog_Vk?ruRYk}PmluykuYgt1r-ibwb}ty+R<9i1$1oI&<kS1;A>EC;8f
zR>lo15L7a?5?OabibRL8*DFCz+=OGA9GN5p@;bJSEWuqBCT99266El7|IL*8<o!uZ
zfQpClH}bZy%2UEL`WmsdSD-jjm#r{B+WxvFut^3!a-W$mOs!p~O<q<*DwW9PL|4V4
zjsr%`!*p;uS#uug*nCX<-mA->!bVM#;5PoqL%Yu#j*A}Rws*IGJ%62wpvk0IcPNHl
zl=~)on>U4R<ULyM*g@}3R$A=dnSWl~eXU}gR0Sg)$wX)HTA7sZvZf;syQi725mTnf
zGzV`2Dw)|ooVVxlY5`Yj0ja$IBD$@fgjI6R3wGp7VobM80AwdlVARL<=f<o>JSgDf
z!b$=tX5!~NNJ9i_+zRTCrIjT%hBz9vIIQvP6LN@c>z9gRp2A1x@FtSn`j%0;h_oH~
z(l5d0jsi_aP>nw{mI;>jwDMP!5umc2VAk@Mo>}|O7;<I8^Iu!%12AJz)-f-Km$QXI
zCpHC1vt{{aPJA3tAsX+DP;hFDAcdP!E(JU{875xkQprtQvxx}eTk_DBi@L3zkn{Vo
zdDd@Lc~v#M80-DHoZ})=vRjXj?I^n7igLvjGAJn`Swq^uCnrZOpaZih%9HYhP#BsC
z6P~ky-}seP4!@lQ<G1(%2|Ev)mlQJb`Lph)lmS=LXCaD)#I|~kpjhdrLalx30-8wW
z#Da<thMTz514=CEHAAz@c<jR954Dym)UP~_l_(YBg(7z<3n<fR*k+LRmVMA!TFCyC
zq7u?uPIu@5eAm3nlfXtQs%f3!XkF&km2`R<u%|>CLL1Pe+sPpcud}l}vqEk63;u94
zc_%$(M~4y8ea&7Iawq3Tv0rY+N-SQnoSyUw2d{~u3$S_SDU1_sT?M-n8HO6*&{CAS
zC@YmKXIi<PxN_LE3M<Dwa+Pgwf?7^`eRQ9>AhK{<el&L?JEq;SWis|{T|0Eb*hEW)
z?jF5-4;f+IjBdsn%+`SoHhGF0M@I0otNU8oz>Ll%tlr9_m(%^92G^%^%igdzaes-d
z3XJl$Q9{jG+aApq8PU99#jTu)u>9>auA4inp+FS%-J=WQGo>9>wqf7F5+TfMMcfL|
zI9#QX*mg}e5|W{=lVv~V=H71{yyH3;O5#?`Q6Y`hjhNzOH-tf}5YKs{4q5|&LD#to
zTtiZ{6@ucFBEYm2PBsGPmu^Ffns`ifO9n(27U(y`k??>Rl?NET9KpfZJae{FN^QWs
z7v*YVaCYA=J)fz+6(*4I>m=`QrkXPRzNrHqObuyK<PAHw1Rr(pbM6S4dUgDE#dJ;g
z_hO^x_H<%xok67zt`PP}!~F3JEZaT#i_rM*Lp43uYySslba00$Kj%6$*(~T9`oZ5S
zG|8+1dWZ#klw^xulr};(-=j0bD92+0{I9=nFQIc+*s_0T(>g+=yE5Q34{?lvZ<u`I
z-Y)#^=P*$;vWyT4FW!MZW|mAR%8wKZr?-jGNkvVsP@$iJy1mR&ws3e%M|kF>Kb?G`
z#I`;K1dN5*6fO$!E29QOCr6!pms;vV+j;<Set50ealvx77R+f8Sf7+<D(5=9<&C7}
zCsjKZkCtX;jVi!BQ*UN+`N)%&6+epZ7W@$bl;Y1J<F!zUu}xrAl@-<t>w^!Mx0vyz
z8|=J!P)C5nU%|_;xEp#JsaNcH>2_RI;#Ie4c#UD|t?NY6_z66*#ScELIsd{~Wzl|T
zL-$Pyadsa7d-WM#{8^QgP*yd4XDe80K|eYlgR9@C$|3t_bxUb%JFFP@`^S(q=w6b*
z4gP&-3DxdvexDhkU+hn1mlz&B6^0=ljl(ueyoKW9F!@Xb84nkD+6rrkO~#3Q5w<JP
z(NI0j+E{{@k`M<j*ucz|yufZUU#y?>SZU=MbrYa~Prp_EcE7a3Anm`ig~b!<I|QkZ
z3sW*e49ST1<`BkwPKd{gjl2R1<x6ufqQ{tjE~Ucw2ysA;kEh+g6pnZ0NY9}y@wth7
z4hC||SqoneZ%3G|!QPx8Ia~aX1@|2vSpNEbQ>|D$GkhoO<npqTY+#mp(JLay5u}*B
zJM-x!s$YDIWMY+;U~2A0WYXvz@49oy)GoA@G0n;*KpyRgg}0a0B%;)@qaC@S-%NIU
zWi>-s82!Yd_-Xk5Zrh-D$~@z)5Kbh=35leGv++kqC7U>41J>;W1(58o%l>^+5kL@R
zc)T0@!wcav9$H!Xyw_Rv;<KT=Z>>F02;~cZw`r_A2mZqGu-)UT0MK*fTz)Xg(c;@e
zzhpn}<Y2Vrs{tWW;JOky+yr)@UuSmOZRJI;jfLd&z$6RrK@<4C^(0rTi;9dXn8-tt
zWT-1+*Oaf`$UGHEQaoP0oZerM6eUBdo~e1(Eg%5C-=x^Fl3NJ5l@Ie}#t+$VEhSpe
zngP!=3UIuEVpEF^5bZiIl<`A(x!D7qQdh#{yj&ybSAiz1OogRmxdz3nu^hND@Z%Qx
zq9=<G2Q|pK?zkYAH{tCy@fQjc>+u<x-*FtXXGB1{`m`8=WugSPQSY}1r$~Q+HkPHT
zgOiSQrK8&(VqxT+<u}e{q>he<XMD_KTrv+de)jwBour^fG<v66U1P7|f~(R=TH6>v
zFeO_9*LL%!ZL^W-I!r)O*j#mKE6xRz9CSTYCG+J-GoGwj%{tf}<4@*&%m|=^nqOUk
zFx_qs7u1X?Mc5X4W3?BunZSn#L%D%KTC-CwB3(+ec=KmC`fOE(!ue&5rt{=?(u;)0
zEx3vH!OLy;gmRgtW1^Vz7J){Kt#7D=&W-g57Yh8CoN`gG8}DZ}fSZ}1i9Th6)~eka
zR4C<g>tZ9Mt#|ht_7eb`z055>+STY<9X?vcZLTbfK3wGwfnEIOC-|-=5iFd?DdvMe
z7wCsmSMDA>iERljk7Vb!6W8duu(hGB6^mPT;Ne+xuv0U}NpBt)ZW<V4e`?tdOB<Yn
zL~zK~VJWuk;x9i^7H2Y;ZRgpqGs7J+uCs<Aq0h`y`6Cjr1l8(FL}w1px#Vrh@k8ud
zev3-@6iqQ_b5qzJINjL#GKfna#_KlXB2hhT>tFTwxTCr5_n?J!{w)YU$=RF!#x-UW
zph;0<sF1sp6DY=(KfTz%olwBsgeL4<Jy`g`iJZn(dhz%q-dh!6st=h~&j798fdjs)
zNc59);2TEb&~D*VI5!tEYgjE8YDR_+aPlKZ7(4I@Te4^>J-2I)V}JlD637+MGr9{0
zCI3@eekb}qv#yH|jP_mPCJFNoMLIs$)a+(t#dN8%^^hGq9}&9ndU0OwGBr%0A+j!;
zNR3=OSCHDDzF|)7XCzKy_Z}ugPb}sR@IXDalkrYtMdQGl>R&VAAR%#pI5YlE-omXm
zWF5sX8(2noOn$98X%yg+@lW6xbWp9;Du472?Dw(?*5RMKXllHZrcJEV&jzvvyq~@V
zppAJTSRy}ZJ#G+$J7Q4wnV4UWw{alexAb*2)2^zOtFzaVALT{L^IlCk$>*-ICuYyL
zQ>pJ~uf4-nhR1E~7rl$&6W_f6E(p<%n77i?ozspl7k?4`71>uovk~7BV<9thF|~Kx
zf-@_tUH@8Km}A1l71f?E-UO*aTgmf^GF1c)!Brq7KCiH@-Xon`a(ZG2#yktq3(v9z
zFFfkRbUm<*u>W+^cG51BH57cB71~rSQF+YB#I1wN{8JFbeGDx<h`npll(OzrSReby
zdBx%!<Uz>e5jFfb&(%zX7Z_@Hn=#O=i@UBaNq*71E{su*%!=!&UWR+y2a`5Noutzw
z7)o9XuSuCwA84z>Vpopd(`tX6`HFv(6opDi70-Wu*56FPAcE!Hu_yr*Jc#RRt%G~f
zHfQ_<pW5`7gi*tIx9yFH4||ILWC5}#((jCVXe^)~OTp?v6L|ERnrXU#XrdRl{Rj0P
zFZBIeMmeK}Upylur2$Xh2S0&{l@z2yQmBEZM8+mxNGvqSj$x+PBJq}_qAr=-un%lC
z!`pjpcclzo0v&SOrRExNqw0!Ab&${r-&7tBXLkAnNAW;=o`yzv)JW}onCG`TaArqh
z$@7i__hr=AV2Ma;;A>s7<Hk$q9GsCvfOmA@PQ>^x{$~iz?5moVbM`OQn&+`x7+X_Z
zM3M+nB7~Bv=k<$&r;^=HQR^P^z-+-Q(WJUlF2$YfYo@oL6anNUm)}(p66Ut%<_uIv
z_7rYby)l<HJvrWsNf<b}Y>x9YNK6Aw=qoq`$~b`~YZq7U6t!P2nT^b=^p}Ae1vMzw
zdY)xgdX<W2Z%Zwtoxz1OERQ!co^>|EgcFhUb4B`W7trf*KT@VN(VIHCsWbX7v@E=T
zu(=ltKEsoui1o||L(p{rc#}b9F6WDX_8$RLSJFN=YP4nra;RlT88;2`I0?>JbD2cI
z7gC?PWA|>V-)l#qNm}&&Vg+uXBQF|3!{rVWWOTs)?rEBmX^2c;a89oCH>7;Gd(MZq
z2+eSMmaLlbN=vnHj?Jz@vIx7T{YIqG^blPO1aw8^YVY;w^sD!#gN18u%birgA|-Co
zLD!ZF-*u@Utg%1<UgRe#$nUqlrSPc*d<VTS*y2>3EmOM^UF+MlUA88mF5XroF0mH!
z_Y7Mmgc0b+3en7zw`s~yGbq<2Yik+RVy1Aj?2~V$uNnj6X$|90X()zl{e^h`=+Yi}
z>k!$ka{4$@RBwHNwGG(R(e5o4Fdj?I4ML#)E?degT{6oJOtCwE1*%v<41fpWOBHsF
zWG}VpG=)nX>W*iU4o+$7?-RrBc<WvB_}sfugM!v7^rwMo66vi3aJd9{CRO1J9W8tg
z`!8*=sWNBoSRoR+Btx`3IM|DDdoCYhrN01pX^yG~JvL!c36_==vn)w)mopMR;8{t+
zOnLSqHtGgGdZ#rx{(dCP&wiZb(Wj4<tz+})*+^5DRz-)DS>DtLoUIgPrTD0(`;p`+
zjwA-l%<Prm-ci{$D-t#XTV^HH8<0NX7?n26(p$w3@Eqk^Y!4%svmH|j&8w+`ElNFP
z+P6UoRUA0>hX*%17GLY6QiqiE@a4DiG8gFiagn}g&lRp9nliF}$t#(~V!}SzxQa*0
z?Kd-vM|DTHsNn@Un^Jc8fB|%O>*UJI9@xNM)KuU07MKa`E-FXRi)qC^bVW<ZkHI4y
zZ^76S;ZyALlU7JAW5R8ErGi2|$Weuxfj)=Dqo9VZR*(dt+T{t~)4M5VSP$9DH_hy<
zL<A~i$#*1ae?xMp3u$Zy45icC+4DR2#%yX-2Sbw2lvjS)S*^L{G&Rk=$0_KQm3uxb
zDZ`roYI6?05F?=18)1M;tyxi&ez2yO-7<Em9<sxyR#`qq6zF{qz0&!;<!?^(1h3f!
z6x@1wKy-|uC~Z}!@YiCGEp_m<hOPYpwtU*(1a$doAvTlilreky^-lZNEp3%><)uI2
z<?)3v1)c{Jw#Sc{s#8Pm3?a?VIL%%I8C{ryd%HZ_0nt(BGPTmYULGBEsQqlMQM&-Q
zl4~|$LUft3?2Wo`-Mlx5X&`NXhg!XeajyTFLfsSJ&S7w$$+O;{T;LlwA_y{-gmUq6
z8P!fDHV+L^(>QCwln)V<)DzGKVRjkHV+o~{CKNR%rrN9x{~$7JGHApJ?ou?+nP5N<
zfZn#{4ngvUy1-fPl3K1zygAi1yu3vfx%X`sWo*X4j3j#XoL@v3+|0t-FU_N-EV6&>
z;C1P~j^@kAWTvSF{TgTJUeWNktw;?Y5Fg${X7I~-`3ZB4v>F>KJ`gz+YKT3c0`vhk
zS^kKOpm1y;SmYZ3dqyT;ya94VOdarX#pyv3CE8CZ6Az<@$9@IOxb2pkDRR)FO2)Sz
z!RyU>4?P_9S@LJH5`!*zpM-Tc+)GGsp%%(_^^};UHc@T<1$qX^?->8J(Ov5_ZU_Dr
zgd!DD+3!;WB%f8nFNPQUN6#-9V#Z4wT#H5ZjQ~b4n=`iTisBaLQaEJ?R}<DW`0wg8
zdx8Qbe80p_WA_!7sQ&VCBn0UH*oT!-!GS4f{y{=RF1FIvo8&2H!h?eXX-4H<D+}I0
z<y18q5kqvZWW)-k0WxPfusED&O4aM#6Q)vZOFbHF&Rd(#7O)$TzRq5s$5E)WCf={f
zy~Stg(-;UjieXSWt5ZpPxw7UH8L^ZJ3@w8<NM>?kJL=$pHfmS#pxs7c&y8Bw`@<$6
zp|LIhTe9TP4|#zFmFiFXuU0A(F2>hMch5&?D9+RQHL20dDSL(-WWOWL6wkiD7;InS
z0n*@@S1oHiVl(6xea9oZ)G()QqQE8+riH?=zL1^RE$wL{spFUc$OMdg!{h1gmIPPS
zDFw(z4?`9a;&yYfG@rloBw3c9NFU#BRG_>d)5_Ov-J?l@Zms-DJW4o<_h9nFlYI^z
zRzBbvwy9PJEP&xxw5nxOJA;rnF_`ZgVhdI;#rJ46=NafJQC8@q){q~Xpc~@|5tzi3
zWlj7-Sx{@9{JYU{^YSW<08S`G_c=Bn)P&Br+x|YAZp{Trhy8hBYF14iMTHb+k`8bM
znPZex;l^yn8Br*0k*2T-*rX)7T*T>h-QhlR4zymIxu4VRyrvTyBwT;$2Il#jopl2+
zruK{CSeAcc+HMFajNI{+gc9)vDwfk1quQLE3e2IaDKQn}JGBS4{!RNK@TqDW2@ohF
z##xY(%Q^bMPuqFIevs)j;N2A$iPqoe_r06l@k1!a=CrgB^#1*NcraVG@*s?aMX^~$
z`|h*vRNWL?3S;DWY8il=z;p4a5`?bJ5B7UT*rq^=L<#g4MFOQF#QKmicoEp1d>=jE
z!507XN2gpZQ5i3CPu4i#Pn_YFq#t{2s-J8s!yDFtAiS%IEU4%6K4$s$Z34?E!pMV!
zs<2B-oriDblT6Pauw;tV7PtYfcsPW>%J8WK?&f3x!4<{&mX7XnE5N+SOTjp@%kw0y
zt{Jcqh75CdDdq^+n4Xglp-y}(^p_tvM?ooJ5JXXZe0#wWhy_z`JY19OGF8O9Z4|Ic
zLCE%D__l|N`kFMJO7{z(JoG!#2?9DyWKUu;B4qC?hgEIB0A=8HEUC~4c&ftLjyG|3
zx->dfLXV@7FyVUY;pZ_5S3)+0m&h5==0M);jT|TcNJ1|ll18fNsH_hnySUy=tOL`X
zdrwz>i;N*^5PBz&*gGav7xs}D=2Y28U6_*$-J~v+wgBX<CWy<|xtt3{m@E<^@3apc
zXHdFA`?<atc4X0E=4^z*b?I0|z*r3nnp1gO_q21hc^92+Ug1`i$G;FM42bUH9$RJ%
zrm-?2PRe=2-3ciaMb(GQikkGafl-jD#9KvF#Nb9H$vkMiQDmM+W6nug&AQ~_6hjc2
zLh-I%;t@g|7tkFzNx;3KL!}&la@U_J@&CM@9Zwk1Zhz6c%PLlMp~8Q*Wl?l@wAgFZ
zcuCqfy^M-EBoXz%q|1;!r=HblZ^EHFuguW=RLq&|3i+U|Sg<?ZPt<0{EAF_u&Ze*R
zvn7k_yHTpFP20PXg5poSc_jR@TA@3+^!{KlK9)p5asS$(H4+sjEwlNpwbAj!e)!KL
z-n3=7H`ng4nJrV9N6`z%j1Rl%Z6maCL~C4)2Okn;UJd2ILFwx@p9BPEDk{53J?e-4
z!nGTwdCp4TIVcqOFNaTWj00E7M93Ve;;`Fe*P+CUyIQ5N0s3XhrsO$rFdrLDw7qLg
ziB;z*#Ut~rLQ4N_7&gI>^kSVQb=VJX<ClcKW%C;^JVeIl44fD4uLZ{z1<?Tr&9EP}
zW@GnG_P^NMJ|a=Vvy`xkhb34c8A`U$d6h^zyD&D>Mxh=jnB4-3wL&7IJ*|GQ)cC~T
z8VhW9aGkE4hj8&rA7eD?hu=OK;e(9cpV*-g{jTS~Kg2jN4|@jAOFODZsj6qjRuCz$
z)kQd9wY<|C4Vw6ebtPVSUZV1f6Zctr9yY1gmYS`TPTx7Tqi`tBWN0QnAiS|GM(iL_
z4Q+9=ky$eN_&gsu(cGDOYCY0s$pAS)hd;IWqL+i=G4t;2Bx3ZyB@_*B@&KM_phaN;
zM?KYMkt`{$cC^&O`?P)bBOJXSBM{`7Wt}vK2&s%vn-A<(GRw1H4uECul`3*ao~o<&
z_&Oa&p7baf)>U7U(c&{fB=G5Xc!B+xKUpjUWTK(Lip3vTBdR<Km|y$QvVu9*syL}-
zq)eCFUlbO<YQB7!i4<Sls-Nh{nh6PuL{0-y4zn@3nrI!<%Gw2nxbqACRLuzXI%NQ0
zjiqHz<>sL^U&*r|m<Y{wBPGHpn*F{mnvph}m0OfCpsmJg9HqQvlU~n`KX<(`>2XQr
z@m9vjxCC;`<COxaYGz{Y#txWDyKT(PYMRGpA;_M7b+C>B7+#xhLjt4Z@O1xZ=4HJJ
zk*y3aXl=`a$8>}{#*TNHng&+ZZVOyN&mN$7xE>X3dGuX{Ve*Z7ZiNgm7M?*Qhj<{&
zfJZpEl~(2=OX_8v`F6m2q-{^O9!-N4)2P7*ShZ$ytA+&9n+Ea(dy*&OeyeV&cfUTg
zuUc>W8Dll{G=%L*!v}u(Nq`fk1`T3hTP4Qc(L*01v_em+;QekzE!D0{TT0NQSrfWU
z3i>><!FJ(>G8&K*hfrIT#rTXkq~5NPxYp|cMP`xNlb}_vbuT}cUK8>uIZh{d$I(Au
z)DM`J$G1Df8Il0w>0*6XN2!*+*ngYIac{+=&BesPG-~H^;63-xxnM^j1PLq*$7cMv
z1bds^r1XwzZHP)rEbgEl|NQKfaz}(SxY)T$`+I@nCc1d|_>MTiKD~IUd*uE;=6gNU
zb(Mx0#OV<0xBF1+4+Y>=I5<HyJ2|1UU@M{XmFNc}o8B5q=hT=%)T}<p93|r0jNAY4
z-`!j2n0!t%)#ELlt`_x#5cj1Bb_cGka|}XjY-_If<_(e<a#N9y#(DmrtS{(26z5E>
zD%w7Zzd4huF_;x#9%vM%k;h~MH!K~d`7V5P^C(8<7u94zl_9CyPP8^vHiMoPdBe<M
zoTJ#y3`H!uD&TSisfxg@c!DW)0|R?V8zi^e7I*vI1B#<8pI&TQO8Iav>Ma>J)h0Hr
zUbwe$j1}ThZ&@RyfynUPKff6hS%%V}ib#+x{yV<?$@fdD&sG-&%PhJSzWhNxfXs*~
zzkx*&@dB9JK^!FtUR~-ra6leRoU9;sFo9jfR^v5Zg%Y~<1M(^j!t0nl5MvG`f<HHE
zmR4J~xI#sYF&70DdT2bv#f0p@^L;wMQy+m=4zXn_o!KVY;mv8z#<y>#TT-25$G{kz
zAj%}ADdrZK+nrzdW8}nr6@2yV2f6R*PiPzuV-#0m6bJ>Izz-;BG+Og=DDC-EP+U5D
z{7^}16z)}F8;DF9<DKA-wMx;o3Yjev_eqItHC2qGw(-~xut%rT`9>}TskpGWa$+(^
zC0B-GI+)muLi~X|Ey8|^b#-n(C?#ZR(6@6BCry_9W@%Kanh`E#qyRWfKE6cDxY)^O
zkdkJxOyg<PY$ENLXmo>0&v5^a8JlWzxcvk5eoc~!X>v@*Z7F+SO&3d@F;B8;=!`_c
z9>l&+?~?^r8DWLx+dp(4(_$bxuX}}cAFXJemv4V=@A|9)OV0tE)VtLd{MO|W&*6k<
z6D*ZT*4s2bX>o#DXUs@EqM|OYJFiR9n~pz}N&Wy<K&ij^)e`V$l8}0@DEtcab;Nvj
ze4g4MK5|+Ikq;>xM@pFgUAqp}mW||vsorze^j%4(sgI8gQ8~^lQ<HMdTSk;&SO2Vn
zArU7qqjNpE6#nSR{%bPo$|_!<un}1f{Jm>$xV|0%VU8gsoW=;{g_TH0I21wf>85vA
zRG@OY@qBcKh!^6IE*PowD@^$#$KN?3dYWbmX=>x{w)XV@#l<^qi2@)&wq@J4ZQHhO
z+qP}nwr$&0r)=AtzIW!O`xo*dGBPsOzIzTm&gog8Y1Jr0C0=4o!h31y%~1aE%xSP#
zquppBFc-)K5?`zuFb(UE<0eDY`HTkQcQwUgp94-~(*hu9dngdHeP^*b9<H76LHQ1a
zRH^z68!ayBa$g!|(8s@AgCfK&b(?pp0Ago!A~zcKaAl0cQdo=+1<6=GO8VzXrXTV4
ze&HcBsGff;av)KAhWU&h+MXEal-&J60ofG(qndkWgwV-vFgt(yDAOfredCAF=_mov
z4A!eeHRK4Jl+<5@if0#EF_Xqeh19|rK}P;j$LB|$KR3%G8-4B@U2X6TfhlqJIjst{
zE<0W%V9patiZVN|$#(WSMo8j*Eb?b7+`{(fE#uH#)D_|)P+bDyRft{XUS&S?>)2!<
zpg4RM<4Z3e7Q3UrhpDNlC8k}1)zBcl%Lx*V|MbCrVPKujsV=CR1x?pSjY$qy=HVrW
zOO#Ui+%nbFkOFqH73jWV$Cl%+=8L2U8qNDtz5Y}ulV<xM4_oUP4TTpHyyujqtOqt-
zfasv=N-DF~mx88jnd=rqp-s?3hKg$z$;U{M)EG00^Y2Nmd`NuuZm}@5x?2)QkgFb0
z<@#uqx9yFo+>VDJ5iW?^7zB*q=meMJj{D^!lne#oKAF*!Cv@v35UndyzSzttnm?>Q
zqK!3i0|gMArVwDomRDF3$G}k50qYo-50Oz9))Z_$`IcplaXl8)?@^TRtz<v<QDsH0
z{DV6_^V%dcEynOcWN#04x;rE&ZO_ZLqAo{{VyLV4J(ojXNmOk7&pvB8Cx6-i=5aXT
z8ICdb%EcqjhJwf5S`bVy?zk4!;aM~_0ryB$YB{t_YA>e1@E50UM3-OgC*E+`wcefZ
zo)pYA!0e07{8!A+WyS{&N>mhCa$e7<jxG>g_Zb2F&mK%^!}@Lbc+6Uql_d}UvFgbY
zp1y25DFqzN!Ar!)KZ=5J4v3#3Av*P;!*f(m)oLchXUo0l)>#?WMwrN00Q%MTRmPR3
z%dyqDYmzHpvli)I!94ics^Ui?0kXp$0>m|Z!-x)vOkZ3c9T7f#bI~8Qv-?VE8qBuI
z)+MQQdp)inGcK7U=)4Ffb9LM9k7G$B%E($BOrwRv>E9I^9VD5?@dX!K8{U(Q!%PIk
z5^pz#qt^2PSkT2J&SgOs&TRV{2>&%d1AOGtx)m!fYX1CSn9=YcgC)5Lv=8s0SHr4D
zc<vIO7|1LGRN##@3j0_rc!6BwJfS$P^zi582TB(@atzslj9$D<fI8|nw8kh+x?zos
z-#OFonek<Qg7<lUcX&I%!?@=KG9n<p?g#6MhP=q{>m}ec83@2_PQwpW9Hidi+)zSV
zz_C-~VVl@?<qPkCYv+-0hNV7}WA@vUwpKrSQ6HwA9<td@ul8oil7$oNAEayYE*EzS
zTlx%q?o+2!C^F!EC{6*2<igXrw?ney?c|iNTttoCBm$IRJhnpZ!{*s8CcLZPsr=#A
z%3g|!v|XOFa0nt`gY2QV1*SXMb*8cjL2zi%2-T!ytP_{@ke={ri#_g=!n^ROxE()*
zr#Oj|ZL-M*sPlGY#4%rapclk`i6^q`EO2V;=^<-rjLF7wzyNe9rf~fT$e@%BJd5C*
zd}Gobx*<nJHgAd7WiCPUuy6t->igRxv=kJWfi3fZb52J*h<R+Z6))1Qh9O8Qq&!k@
z3}D!wD7c~B?ItdRSBXgvd~qL!K_4iuG0?<$kum`h2fS8nU`SOQ#1thTc#k?_z2ix5
zZdxyUofsaokAO)820CTa%zqR=y5(j2Gl6>W6<c7)BWP;#jb;-p<78adw{HATFR^yQ
zVwsv4^cb0|$&!e-mdXDEt7!71PxtovoQQcMWM4(hg88MY_tBrn;Jl8T!kwuYPAoF%
zRnhY*x30*1X`$!p7mPoLL}J6^Q&2Isza!A+s--q8h=~HrY#dm)iTB~DvNbf~E8r*!
z=xTCzVxVsgsNbU*K*wtC>#{zZ$d6^c)Ld*uTXlmgjh+dEm)m~K@6`bP4S&o<y$2~+
z#!#%lh6y^YVy4MAxueL7c9!7CypX4Hiny3Lp3%4gmwnk7IVGWu;a0XX4LMv57Rwn0
zVD6F(PN8M-*pSVwTs>%9*QAR<cx}FuI`ka4WNz>LzCR>BnKFkexFN<SA_;Cvyix)T
z7xYzlpwC*apiCWEV{o`yf~ccEmRP!JH?0z?h4<SwS}9Z?<i~R?8mOz)Uja{UsLXuG
zVa$*4AH1W=Fy*dldxEHCW~@u&HgaA1rPE)Vo*`F$NQ@4&_N-}GE^@3kzzGE;q<gbA
z3Jv^M9=0kYQyr(`J)=bZ3lcKK2j)wv5!gyRBp}ky;x7rk)T`Vi>vd+mol+tN#QCVQ
zlEKQQ{x3E9u%!l#r1Z)CGQkX63z9Z<=J3tYN)HtV{IgflHW?=)5T3u;Bw4pj|CS<Q
zMBCh0&o3=G=HE0oWx1hJsqzI(Xpt3(jeZmnB^8|$hQmmz5CjBIab%i_R>)^{8@ekx
ze5*HBerCaWvThe~$GjEqif#Y6_d5{IKVU=At5~5jtcyK@cUZREMGt+NeA}IumYXD}
ztPOIuc|P5fyy!p-rmVeEgrbMQ4}A&_ZqPO)aC8@GjipNK;z1jrya|bg?0F3}THk*2
zB=TI0>83p+al$}T$=y8<T&eUXH#ji={4Gri0m-%OtRBS4M_IYZ_$5hc1qbO~awjJ>
z2Cz9@B{KN=8Boq}puq82`HsTeW{Q1|k4l_lXhUJPoc`Di=PTDHAl{GyctnrQ(Gq|Z
z>kiYdX`lb-gAe-{@HrI}Vm}pcxNx8=h<ip02}GLux5G=dWdb}4lcA@)S&iB+8d_Y^
z?tb{f3o#>Eqsg`@SL7AGfBG6tZwGD$I89jS(~I|iPE|&$E5FZE=MXa=WlJ%CFw@?u
z59AsYKV<#oC7%Ga=IdtVCg}9O%6PM#&!c|F(jNw3c0z4PcoQ=N8ehc()W}iB5LGH`
zW@e*1_cn>@%tHmO@wb8v_8@haDsI2c4K=+SuL;EOPQ2273mP%jWU2O>3#Pv<vM?Pv
zZ{8!dzw3ngss!I^lGp*tQNsaVj3m*TC%85d&)1$&DI`38Bc0<&Z6&6tLFc?{z#r~h
zWy$8}2KmJt!XTOH!K;#Vft>9vb2Y${0CgW|IV@sCjX|`)Tch{G1Pwd#Bq_JwU$a9}
zyLCY*ZEn{59+Zl5w^^M6G4gGN61aqZh7{#Y3m`0IUYKb#lZ3U2CTdB4l`XUB;y|D@
zAgX>%S~$hE95<Ta?{|^}FDFUbYE&|xs7r2NK?KCBVgtaTzB|+&xIrBWFc}`ZHB#N}
z!8Du4JwV7n{PGu3=*USL<6tqCz;q<^>21-HdUKs*5ty3?^b^M~aD`5FL$LS?2T;Fy
zwdCq!_ss}@ApnpAtNn!<^vXg#Sj$*-QIE882oH~HOpeDh56=ong|5|0JZr_sN{(HN
zv~_BjwL+(?*)#nz=kFy<uLg~eAsfBsicVH;T_WDuvaBv+t*>fLw}1`8=4zYe2z>w~
z4yxkr{YfT2-JknGPUIGwvwf<$CwFFRLdIk06l9&9@$H6sde*ZhK<`DAb|bXyp7>3x
zt6)eCW?l!{ch<{dvqF3uSCa|2##8+&?45}Go_T;z3}}+>7LqqF4mqe7I|do+h!?om
zL+G06)PDzmN@#MH6FHQUpFrT;L%<&eeZp2Ixr3I3O3rD>)U!uTrgu_;tih^#k1Jc(
zC2SqSG_cm^E69`ZwIST8lgrkJ{?=?75N3PYMTJKnMK9QbR<O_V8?cLHEMId|0J!1m
z7_H08g`s=0JNA*f`I9TRdlx-c+h4lj3AxdCvG1b!Sl^EUV+vM`KMOX@4WF4$2%PgT
zw5^L8?fV^8M~?52Xm4{l>>&gIUGpH<NC(YUs29$-QS&#=1A@zb-)3hVE>A6zac*T@
zP@*643Zdv9&}w+-7_id4btp_6r+yG4U!t(kjEQJ=3wd#D#Z{UgX>oUc>OqH1BJR-3
z!aBk}hp#0gCra@?<xTL^uBdRx#9S&jC3Z?+D{hk0ch}0XEYoKp{l4~<nt<|fS+?7D
zi3D~Vfuic>R2+nC%Ya#g=2Gitd@vmuhA@S}jf*PTb~*j^@}CieCP8teaa$iVF*s5_
zxJ)LN_2v;03zrTj7tIQjpZ0l9LmcJ3aQaIZ%dMkVx_PlkHw@<eJB_bfSEWPyk9Brm
z<%31PRgC@z*Jq^R3f{1f&XhMP{A`>Xm_hK2^y~xZwyU^kupE8qXvk2elC3DvC<&Y5
zs&NIkSL*t|1)Ax@=kta>#HvQ!DDnN-(wOI1Eg>$8nB~sE-$;Vu3X`*ez%zEeXS+j>
zzQGiS6x*Anou5?)m_*ho`REvLfg1_&*BN2obo|_yQ}SjOt?$Ix*ckN%yVmOj+{%3G
zwng&SCRJG4JWl=0kGC@McnEM{%&<`h*;YsI`1I2wccVbudrESogp{jL&O$tjgEsKi
zd)1-h66o|MR`CAx_BXX|f1iGJ5f#B1oUoXjXs+><ObmucB_%v6Pmr1vA}@?Af^{%6
zN|OAr2(;8bHan0#l89|p=@PoA{2zY$&KHG7l^&{>UG7LBUIgjkrzSPb42ozZWcg;8
zxbi!M&p~OjC?)qC0>0o-)bg6Qxy0l9f?PK|Vt2*uKYATROL8qGRch=%Y0e9R*0o$*
zT^M`J?)8AY#PLsaI<QEP9J|h*CRZ3tPUd5x4f17O+T-uwP7}V2ygPT0%RWfPDiOMY
zms$Mb@!mXIx)@`%Qz4MruC}B^J+Qn}Q*-H$N1}t}EK+>~i<~F#Oona@r|J{|N}xs-
zfDG%g3}sND(dkUQ%Xa0PX~@^+I;^Ipux$vd(86;v<?-Rmo)B7+&=m!>@olq5f%<xK
zm$BptM?Q5AQ7k67KZ}q9TX@V={7byk1~r~qB^9fZj)W2Z1KW{b_S6gdFN7li7=B;s
z>|)0J<R4uMElTMEtCt>MsssAcZynR=^JiS?z4qD3n*u{VgE@K}trvb74dMf#9iQSD
zmr57}$P3Zmo(UX_Q#aRbFmlM0kp9VoJon(|LP{V{7KsU@h1OVmZ*g<b1!3)eJ%;-V
zHdjv02s{HEWjr9;;>HWI=E()fqGw$!lP$<Vs?mxtFx2pM`jXvsffEMMAroIt>c+4l
zllt2x{cF_lMm=-^q<0OZXp4@`PQX-9Oq8C`9*jk|^c}1Lg(UxRcl*K8?!blDr)Qc#
zttg;2MQ&o%<)I;JKOnQ57dXQ6evxK--<lT_pd~3M-Mcw<Px|wO5N_xg7Na-XOjad0
zTyfzx-IdA*R-qalBauQSZHm;38{W2L0JBt12r>^PHe=C@xuk;;x)+Xq6xmf?rT%xe
zUnp%gd+XQI*65T@PXs32w9%4q?Lb`EAP^c?{5r$HTrt2XnSAhe@$vi=<h-A$UZivR
zBsmds2iy9N8zV5}Fb?Ex2f6%&I<bL*@PUk&Q$p#*Ki9a#<)E7a9GZh+R^sWZFKJ=?
z$%P{5aFGRlh2=%zJ)|_|md<4awgNvOXgNXQ-E~bPW)}(U72GoiWq+8j;a;;(sUi2y
zskM^;AO)luwcB;byrI8FA6vE~=``ep86x|;CKChp?&{SG?GNuPW)je?5BOXUI&UbJ
z5#6u;?_6qzWBRTX35T%#K5?RoLQa_bhrcyJ$T)EIbIS&m<j`*ek}tlq7W%vo8XPl{
z*zYl@iz)4OL%ogo?#0_?#}i=B1V+k3YgkEEky{nAW)yv60qki-G5RE6w4Yr7BbAGs
ze?)XCd4%T>tSek*hw^RF-S@G2(K=aGU@OVADO)>;VJ&6UbT5p#pz@vJVw*Y5+%@HQ
zIP&rD-6(J{7D^ik?V*i^mBqGWOU2I}r_0MtNlq=jOz_OPDQbs__z%nNli~}~HgIRx
zc=T|2wAlfxfq4++3avPHFwPdEXWVnQqLulmoSo!jb9mngJmLI|+|;eIt){a=dkMRv
zpqd@T#-?A8eOYBM+$fmYTcw?~8_VSqQsx<o^vx`*Lmu~`Jn#h@O;H@l*7dAx9+Ed8
zIeX+6r8*MayJ`-Z-cHlSH8}YG=J@tx`~@oJ&!?Nl8vZWP#>6&0;({bdkg%q<x!L8D
z>2h)<)I84|H>}8kYqp@#2mbqzTFtp})pf{}ClR$VYZj7h5hE%>T7`yHlALZ+8T-a)
zlgaL8H%Qt#0!$bFThS|jMfO*BgV91Z401i1a=k=?#w_xNZv4J?VVvh(W3`O_9tT-b
z_~$`t5C+)1VA1$8znXeZ3W!{xWNqbkJXZ>%pj23{<7`D7Q}&!0jYCZH>Ho~;aIhq-
z83->KS*?j3o9-sCiMl(f(p7OhIUT~>JkJE#hdoR?kwuF%q8R1B=Uat?L^k=`3pRku
z@QS$}ove}#rclI1PN4o-O)*jdBzNwH5^0dH%qZ|(SY`;)gZ9C#dA+mfvlleMMZXp)
zm&S4~!EmT<v&@x&%)SUfO+fxTG57_+hLA5}GMu>HpvqwQ42K;dPb9aeN)Vy9^Sk)(
z7)qe&h{~Q<_u^qcF-ZcvE22VS>3ZMltSHU&v@f*w+qTVx#3M9IBNbfvP`0IK->l8_
zMlZ}sZldR623%2M(k2xGmZ}h8cPeeS#dL-Y3TefLwo9iUOerksJ@W$lrB_nG+6(L4
z#>U}$s$cGhDy~n}i)l{K^9zIE8|HnUR)A2l-=8n}%;GN07Gx!C-x!Wp805>jacEED
zbAra@5<<ip_q}rYh7fq8ARK{8_-~0T!X7|QqB8u}Aw4yHh_pMm2q>2niIC@THjRni
zVSYGuVWAgh(OUb^3C9U%-38?v?Q2@9FbsX+uSw1s6@sli*HOc$H=eXT87jwePL?vC
zsP$m!@i&~%t%+qVQ|>AsUax1MIv+SRje=|#`8f{jH9;yMlU7e^HZEnpr2-LjOo0IP
zR7Alg>S{UlEp!@wODKo|oZuqh->82Utw2qdFM7&Qv!MXWV{5!1gT@YC^u_D>ejGrR
zXaA%&#O<H@NS(Zvn7he~4{E2~c@$Lk6Z&G3|LQ`KJT}V3U1@VIbm4;w$C7u4O5lOA
zM^RjC6g(o(A^P^5bd6Kn2RGN=SL+g7ZrP`t)`j3)Wz+6nMsj|k#gEa&uRo;5klB76
zers+4gjL0t!FXVl#MiLzMbzKyH!xhV%#D3bc@3$Uv`9l7`657D(-s_R?u{Y7qZ4mj
ze3?y%FK}&l9?Ar!n~-V-4+9uszPOkYqa`}sIfoEM2K-qtjtdLE%%e!ayBB`pP}+kB
z;ID47S}NLB_}J=4$EXvGkZsDS6R?6BOo_0LsP<YOWt(1;l8?PEN<-Qd5PJO3(fMb}
zR%K`arcmq@OzCX1UYuAIw$VgV{x|+5P`iOwX=uZCP_x-0l|LC`Kz!bBf)Khwq;~=v
z<b97pKibU^YvyxHVYT^yorF#OzSAbW>jVJ1tS(^=bFQR#j8?Rbep=^$zm7y1|8>s;
zN4(+gXnAx?yfQdZ9&}gjPq3$t-FN%^v!rTu=8*Q0SBBpCjZ{xKJ!a+u5!?9a#sev)
zweaFSR*K0poP6y&v%~)fc=Bt?zE>Kf)RnTKjatt0mJ>dP>SYQo$qrAmMMaF|NxpYz
zJsG+YiWqf?W~k&(t1~yvp3*$iphh|eq{uDe&@pWQ%TCc%KibDr4HeHjX8N{~W1G4K
zQHV{U)f-SLcp?FDRm8(R>+X-PM<93!eE2>=Oi>Vs8wiy6Vez+WU`vx4s*x(wTCjWN
zhZF!~LH@Q{!bYp=%0tV4QRYkmFbHG&2dt!C-PW!<{3Qr%qN`1E0Lqpm^lhK}@7Kfi
z1fy8nNUf+5H!BdDW7;c?AIpU`fEe2b+Sj&c<$>3fQ=tk|Pt%U9&DZT_cg(sywW4Gc
zQR$zCF-M0FziK5{vD^U>ZF8s+BnIQM{~S<H_K(vXY*~u6Bh7036l4d;h?XR-*A+Az
zjSsq8s9tq~KwvFI8*<)2NgsTpc8fW#&r&gw${|!VR}RGY?~4w}40dX4DcL`&eFY3|
z%Qr7}r47~Ujyo1L;yuNl$P49z!|!nR|E|9z-);ATLLjK6-_fu}<slp(;00^o-R004
zSA(d=2n>n^c5QN0lUO0v<7rV2*q1n}+e%~;b=Ifh*1kCFK^*67M-?SDxw(ZzDFMAP
zN{3j-TIi?Kte$EG(*H7lIx$H7v6o%Ce4nN%&!<3;%JMNpzVt+*DR0@tyojJ$xUs-U
zK@zp=`8{I^Moo@l!Adb=&pSmBN*eP%cUMROda<SL`#EPZGs!Wf8v0B>`Ge*57-9*w
zmiD%Cr7Kk`;?rkL{*h?qY8@9yE9H#7g-zx}s(X|d%K!A&p>7|hC4`5B#YDuOnn?|#
zk^91&j|S-vYkD}pd`pfRfK5Lt%}#5}^=^sd?-nW=BxR5j=&wJ|Af8Zevvt|}E@#0h
zUQ^?z;xj1%ze!qk=QMWk=%zRW7}5dPjKiSCQvfVohO_GSo~n{ISYiWQ<P09&q8f2s
zN6v-p2#R!R76?(bUGrC|1G(ov9yB4h<d5>H+(Ve+R_5mx;_Jz?{gMSzAH#*U@;?QE
z&<F6aU?Hb*NDQ%HGi9?BLYFWqN9$5>_P`VcjOb?=CeUy4Hgp>(4$*YYSyK^F-@3VN
zR8KXbmv#o$Ehs;K(&R#Rt?U;Ma40K(jgo^<OuTp%zHeD<`Aai884vJIIb<6|n|sZR
z4PbAx2S<0fJ7HOKukl{MuOq&A<Jv|js{t3$lyd9+7+}^C(3Hw$Ya~j)!=Zp4&B{fM
zb3vhkdA-d)V24$d7xK)MOz!vBSf>8ip$iC$?j6PNq@|656f<x+&(+Qf<4yMX(%)l#
znM~BV{>Ck>hv&oufFR(E^7(HjcCnaSV>uY{a?&*1hSXa$(ii{h-(2oe4(IoiN*Qg-
z>Y43Zw@<CuM6^36RLR=*#YTC~H>+>*Ai_GD2VOdxn}CgY!%zPyu)(+zrqVQjemB(|
z$8DR>41Y?zVE1q-fH|0s?iKLFy)0M>>%R@TV;WGW!3^ZX!PrEVWq>#z=6rC8vWr&z
z3xDLut`X^;o|s#w_%S6b4^~57pgIcEVmVHUF2WzcTNHjP4U<ldKN!8|1TBQo%+%ds
z-s%XvXbg7Kc69WyP!_AqvNR)Ex;G8L=JFDD!%#<-&)0`30qFiIW0**$9oa(IiYI)4
zkRXyL*6I!ooDSJbCI+WP2n^%Td8(4Br9TFl5MN|b2jeCZkr*2Xr~bu(aPIli0aIOB
z`*!dnd6`4#%tL9hf8V^!hsbmC(Ob7t#(LyxrP>fTyn?s@=VW`tJ#2K#1!2Oby3d%G
z(NZM=VkWa|C2i>c3j@2Cm>|f3_b2KB1~rrfZZODNu_rEkW}*XP%r1JgMFmRSMUjOM
zcuRMD+3AFAp6bpMwRx<?_lX8H?bhTDXXK$~+F9RpdC5x3xec;6f#_EXQY3d{1bEPD
zLAlH-6~lf6@JQ<gD8;4CUPiCWsAMTN+P?Kk%(0+k&yXxk0VIQR;aPw$q2{IHU9O(?
zAn8!8`g`qOxusv4)$MRg-V>u(9_>P^wei9LuMOVoQFZiyGL6b9%)clJ?8yoiat|Q!
zivqWB4HKHl{hDH6cH>RHBc2)@k(x5S+d8sEBYs5^<Mfum42Bg}2M<;Z3yRAIn4i)$
z<CRX902@OQ`YNLJbJF+3!AsV&#&E?UxyFe!gj7Ij6mAyV5)(yAB!Uxo!0`8#`9VA4
z5pT*oi%<ry;jtQDB@xr@id-2v{bjVPRQrgK68gQ2f-`#`QLQk_4G8!bI%mdOjX<%^
zu`h5B9^>Mk+|PUylfvyQRyrt%$Nf|zBgDxQBx$gBowy*Nvvczbfi0zrlGX`!baiTS
z1M$XLk*X_HsLUiRW_};p9G^kiH-%?583xljL%Sxu2l9?%-GR|ac*$uIt*!gF=7?qV
zzN+}&3x|*R*--SJC8w%^F-%Ie!FdeqdKxi=>FetFC96g>-{gT=L_b446)P8llfE@!
z6Dx~Z#tTS^oC724<BMi`RUBaijV$XHl9Yh>)$*Voq#3xa=Nlo5)#DCSi<3iD!)62*
zWq&lS=j_r1uQ)!(0jgRg`()-#!{Ch7Y08I+N=VQ4=}|;KTfi~x7W}LHY~f@)G+V%a
zuqZZbZe#SK>o*U#df`iZhev(i@rBw=UtqByx#kClccelKQo`g=4c*j*K#EBg=N}n;
zh(*;MCtp4lw-=v~k}f3#Y)qNNStn<n3@f``GjZdz-Re4LQFDU(DRhg|)6_m!|Ee~2
zJ3k3}<a7V>T$+9Ib?3SeXvC@E2it?Hlcv<N`y6Z%&_Vj|W>lDN=kS2AfO>{kMAaSB
zGw1tstBRRj8-}<G<9T9szLw^vfFY?ppyk>sKHD||=j#50Zv+PBHke%~s_Ra%yen`S
znNIv1mwieoolNU<V}M}}nef#$VkQyjDx{}QO$ZqSf`%lV2KXb^ts~0279ASmD9S?N
zI3s0Oz-MO<jtxX?Zvvd{OExN3%a22TTugH0WlUowC$|j9<Vrbvh6ha<g?0%DCX(W*
zDU=0qe-PV~8#O!0_kbQ_mO`Ql=`h3s&8}@8C|T|+Jz6`;egM_Y=k5Zy`q(7oKQ--*
zH6Zn|Rm8~G(N)6ilF}@>XMI;ljA2hQgc6w+-D!NOkx>^7qGg)YETgG;9l*gr2G0ym
zGG2Q-t0Ie3#|SPwRudB&pa28V_|K=jk_xvwXKuW2oOe|)-Z1UtC(>FHGnqW!YQ1?=
zacfpw0VeWTWOezW(f$lam&uC=Ys_fKhQIIB7ej=UO;~$@1~N~A(k4=3^SjMAx8bJ~
z|6*QzB55z@+_l3G05&~pRXX{(I+Er;euq!|uMYifFu(2^)gRyvX~2=t$-R{TFfak!
zAeNm0Pw&SdwcE$>NK9Q<GnfE`%JR7euO3d7kMe1EkIYBgKaz8p0)lZRmLD5+L39-R
zQ@Yb+%H^IkSiFHi*6znnmxW|h^FkbJ#cfo%wtPP6+|PiJQ6vFBLCXN|C;{WIxGKly
zY_G>{wc{@(sE<lWI!b&)f4r!fc4&*<BcNw>{^gdYKsl8JebPyW8Dj=frd*N}L9Y7f
zSWhR=tBH?cl#mqYY|8@qry0$J%8YV`JJ4+q_WiWv1&iYZk}W<y0K$Q;V)A$$?mDY4
zj&}8}K(#EmaEE@Je+-|_x+^UQ^%hw~dx+Vk8CcgIbY_oqdb52J(y)Ztn{Cp>z1R)?
z_)hd)LMMB?U`7~)@>t2JXHEUWbQNbSjl^k@u@yZD;Yw>s8lRR}nN5hd)x^GiWvbE+
zy=HaOzR=z`+66)GHZTam$pnCZVt(P?T;kr+o3rNUiqfib#{S}&FW6z?`aC04UnG!8
z3A`8wN*AMBO><ta{yS{%oM*puuPs^T7$dvEzkgxEA^i&y?U&OBTWpdO{5EB(oo**Q
zV9CfIzY9y!fDp)cs(6neVaY;}_%Mx8+9@+mu0g3U(YR=<Qw1~p0}iIh`uQ9v*wNC!
zjp1#YMY}9?5x?wpM0MS{kGu`5L%_5!1T{zci30~cx@WQic6u3Kl%2Ck@ZsTs{W)pq
z(;iit<XeO1Ck-ydIZuUV!^2(g6XM725)^t|@O}QKKop~e2|F|u&GF><8L;Xv2hsf!
z(gP21g+_sb?~(DAex9|x;^q``##~AsVUCK8I6EE>>0<;=j$S{S$Em#<@mQNEuL@&M
zzQ+Zz`~J--lnW#2J-wXDsF_Cl4LG`=k~z#B!RZu`;+t!0#b?dG+5NSb-mS$adm&?p
zWR%dg5TUE{w?Fj9Fz@tO&g4lA4-_U%G&1@@2YgGFUS=RP3-{j@OQRvR$DtsZD(*#o
z^JLo|6}fYrYVV+3p!Oh>ih$WR0*}?EG{3zEF##3%74$3_@H+G})vi)*zxr>0g%TWo
zq&W;&(<}?#5Ml*Hzv1I!MpW_AktGkwYt#1EV^G!)6FmGg6A=FI+^W));4BqlH7w9F
z49^#VUrOhFR2#rD$8|G5O`)FcoLK?h)>QFWDV12!(&y9ahjNl1((M>IbG+jBvG7+O
z>*UZchC$OQ3cp|dcjVLH2jf#FlkOyTVm~ke`bMDp(6pVW`QrkrH)uc`8c`3If!uVR
zO%+Qv^yzx`4qq3kK-hoE;2xVboTB?^ENU+Qo~_dzMyVeVeSg))&2InALs%NX71aeT
zZ?>}cq&yAgh7z(4%9??0^tSBK3=mOKX2-4Wn1uz33jfJ4Km0wYpCbO#w^lb2j$uJt
z*i%JK3Ys^8QX)Mb`0J*X=d!Ssc7_y;>lOBgLAGI0|8)U%nrd0x7<2bdHhhk!f*>BB
z`5Llj?I-7HFF%x1Oi7iBS8`)@KeJ!qT+jRX)1t*DfB8a?9G;ZL->li6jmRkt>h*%`
z1mkc2=Qw)af;}Jy%9j5S&C-pADGLiad~mlzGv6<4O6WQBls-7x7KENkY)77$p&p3$
z1A=p{B3bJT_h~p<G<^0TRkklEbwS}GKUtCUyY4}$o6SabZ()Lhj4~&YN+CWvBmycm
zN9$j>B`r-QK44M&s><d2DbxmA5kZ0e8h43tx^Jd}q!!(L;s2!8*uz{t(FBS}*B&uS
z;1ftL0lm?o5#8isfO8tsj-)t^fHY;wiLPrno%ZtMVmknjmnx8R5x!nrgSc+zi=y64
zsMuvQTB>^jd7K09SyhHBN|gfEjLQ<8-*g$0dJT8&1ai|k6IGZ%Ias-UwBvQ+G0*}k
z1?5LQKuqBWnsz=8gi~8#@I!JB=-kk8KWj~8Z$Y86r0HIu&6zAnou<%)rzS}9Hpwi@
zl$1^~k4O{)J0>ii?t#h_C{A7RWw-&HxI1R=iBJKd*6{+O;&q8t_)Q>0p(V_4!o_QK
zN9t0C<nhv)hIz|YIHBE+Ngv)$HNX4s3cZ%Aa1jTBdF{4my2x}i01B0(x8cHbG^zho
z77^-;DI$96K)-GFWw|Q%g&ZkaFs?_0oGF$gF6v9(WFHQ;vGG8l`a>XI@4rjt2&~(_
zs;sfIBZ_bXj-&%75F*ycWkJ{}hA-Mux%>{%d}pDW9_+_sK@^e%vr$G${S+uj`^>Ex
z$toc<Z@1-k^tk?F8(J6Ra8c&g=aa_px+Ijz@>#ETR$S`HT$nW=oipnGI6@?^FE2Ea
z)r!1T+}V41e!lK%dbh0Q58u(j&7-6{-Ae_@F6o5yU9maO1k46ZVCnOmrtI71-$y|5
z0-{z%67YTD*pZgR?!?OK{YAa&gSc)@2Xij3Qh;C9H+yvUkN&cW;=XZY4PD|Af)U?<
z4J>&e>J8&qn0td44M6NR(5x#BMlep^(^$i1`3EZ|wmXQ{x)5iYr;TG#^E!wR7GWdW
zD#w?m`8qqKGkL|>_zy7emk&s2|3G$z?xAXO8yC{wN=WHJe$rhKZ8HCeBRaQGOq&08
zDNxJ0rfzmxK-`qG$Gxi$_a%P=9yG!-4Lo^jn%cO)UpjaEfa}5Q(W^i@fF7CvxpY~*
zHR(pIZzw=uAMX#_!c~#gA}s<K0qWJ0U#PBhor?*(Fzw;j>{Eh=KggB|i<^|}OUn2r
zDaxFl8UTcO-4npBjPdl?pDAFFh=+H`q69bjj$%o8Sx1;ULFL~7l7DK$mHPNVQzJDF
ztUf8UUcv;0MJYbQymzCiaXQRK;Ga$j5nEQ`$%7Je=FA=pwrit=Mj4b@K5I<WTSsYM
z0Ajp?g@Yg2wRRTL(Tp*@*|7-<Zs1Y^YlIX*BL=A)w+<UCAKt!r*Lt1%<XhFq$V>0i
zjFL$GmkKN`9qqkuPiuHu%aT%|in`*EB5pdjeGGhOY`8*}*L~Xa|97fIb@&_*uX-eu
zVo)nf>f5{J8k}YgEXW%iCs^lF__y%Z;TfMO^_~S1%no-~Vd@oLL>t6ARrNfTx6X+)
z1B>6f2@e18SRig1_NA``Uvq^~6-%`;jHYm6x^!Oq;9nh{q)Yy+7Sm9!;I%hICs=gF
z@c7@OO4qDs+^MXLj<@5s<3T)GyJUw=;Z=abH}qFMy3jAUx#yI__NWFJ%|xrL*pZX<
zcprPT?R}iDZHLtfoao;r^LGA!$fo;yM5l+g9H>X1{Z1PS2t-)>t1*->DWdM$*j7(g
zl515~GyDDxv5nOFSjGbFAGT(+$<O(_Fhpz`x;QeR19em-RZW>LrfrDT!1hHZj~n(~
zMB)@|8vcDc+&>HTMm7C?I#Q;->$kAu>eTC3zSkpaW$QdICtm7~h(~^ASD(5#O%3VC
zK}o`gBJzX+f-s$FNh?<qwu{)Ji|e+Z4}4x19ndIIT<_^<hh0>S5^TJiDEKjw7TIg(
zJ9WmPbjdNgv%i=eG;wwQ%|Ox%N`}e}bxNFhxcuA^u9GQpQ(|2{w?Nms6Chd*2|Rsi
zvMB{Fl5?Qa{O&gD#QwsRO3$R<aH0zOvdS`jaRXh^<9_?luOMP0X+_94_XmvJK}K(Z
z3!je1hCdRz6j1PAhNgJi`4~)qHA!!db|YK28&uI;Bd<KuLMHA?)IoSBJz0x`1RXBr
zY1a&(OI!4p6l6fBEv{84<N5S*AUd4tRg4Tjq}CY!_n}pm8*SioL-z4qhk)b|ZUupk
z_yC#$q9`^y#AZXN%`HaZx@v>{-ABl*thwn?!SR{PR)u%dl9C&<{^@CNXfmut255x?
zA@jVB<2d$=AhN~v8@iQw01gn?EWN$%vyMAJbc?qiItdRFi~NI|SI69yedoNCRm<G9
z)VV0Tbj#u;r@GrMw8>DqW|!y>Rlv{q=XPE{zpacNFn=Pz$+X>Rc1MYt;~rXGQ#Wn@
zeQ7JRMkaxFA#6EeX$PQFnW4sI7@#FoN?9TcEckqX?j;A-1KBa1$Gja@*cqNJt~FM;
zZpAWG5b<D{aG-RnnKv<W%Bm;XtN^1bS_FL~fPm9Xrw3?cLB0`gfdjH3S0i{LTDx`+
zFnHF2;W^ez9$anMnn91~Jfj&EmFsY$2swG!pyB?Xx7U$bHka|@z&1hE3^tOM5+fG6
zZQ*uRoa)WVji@QpOwAW6d;=4NpZc^8#H(2OfJE$`$QA$rxs~__%G}7X4-9Juce9nr
zcQ7Hx$YeBS(U+rP7aYt1&fg|0k`UxnDH|N2X6y04EqFK;m3LK{aT28qh3i7oE}GQv
zAkyv>E6kJeKCnu^=IN|@$p+PBvc6~j;f1Sz;VLOS3~mB0irSn#XYsLz?iwwFqJDFA
zkmr?UEhw0$8VDS}=%ElKB@#lV@2DN1d3Rh?XN}i}L5pUB`8=bLc{%%;IS@z(EjEAY
zPj5vfeHg25c(rAe1MJVY8mEQcJpI`!s|OAMQ-UZYXBo_0pFpdH`Z@JjuK|Ge4V9Uq
z3dI`9&!Hr=?K1E|tYL%d2;IGknJ88cOmaMb?l4&ps8<;;54N8_omAMOY*-c}nAklJ
zm_DAg_)4X*roNY%e#pF&xdPrO`-nCy^r2M8_xouRCAxki!lsquPT_~x0Yy`6%UkXJ
zCPxm3#EhzgA0NP(#uVYffq&;G`(UXWq#1o6_>kOwPA|ewc0CVL{P`Kv`OQ3b3<F6n
z*yZoA06woAR17k-T>E{a#IzFTUXo&JEi_U(IwGuJ09y!t@o;cmX^jz@@UVjM-_1b|
z6i3L-<C-xpxop6vmrog7g;w}NYC%;VLPTT2ct}UQ%&pel6xnS<wi<>a2H8ATd$<y`
zzvxk<Ouk8~1Q+ryEw44D7mL2*aLD>*7T!JAB83=E^m%(nP}zq2u}XDRXXlOO<TD>o
z0U+J((DeWdjgT$%P$Pc`Yh7f9MwP4RT`gA|(CeT=^D~iLN{*!`P0zf?$Nc$mV9hs^
zQ_)m}9;iqmjGg7l`|HtseCO)A-=wdQbnC=3ZOz&GYp#^A_^7!2ThQEH^OKd$jFEF|
zYry9z^o<~mD5zv?k*JCOuELgPM{yEZH}pc5E1*6~4z}<%BV_DAli?YXc}B10npe4k
zPCO*#*o!;&Z=^ciEm)p!P@5*}NZgPG_|8mmVdO}}(w@-R{*FY3)G5zld-$rx<lSLe
z_Q>4&n9z-&!yD<t)iiQYy26oh`>u2E9cdtItOL5H^7jDVRu*quREtJ4B^l}!(lR3A
z8L|w3ST<ldzTthJI1<)egioWz`+wo7=oEjjT!I53E)86O?CTLs`QX4$L1QfcwB&~G
z40d(7?W}qLV#1~YLvbak%L>FAN`2QhuF2eI%Oo9ZLAzMMAc+fKk7nd7y~8LOnrb<t
zGCrLhA3y=g?vTIGbsp+bt&Fc38(mR^l>C&PRAT$k<tS(0?52pOxVfP|+GLTE5T7j(
zo0;<b{)vk}=G883s$~!fA)68iBNgkP<lP|d)gMVwex2bkcncMyOAXoNGoTg0)$~5B
z1%J5k<(gM;StMzxOjvUvKE6ilp}_9T5iC0V(x0vW%V^P4nXRF1;JWsX=gbZF)3&*L
z`?0Bqag_P#<;Q?{1J^*dBiW<q?K@zz8mggLd69yt<xfd9rzV5Zmp$b@bX6#Glm5{V
zaIQfJEa<i+mmBjhyDn7O-Y~Kaed?ZJyn#*X0?RP6zhzVkHh%_Ia<lBNMCrGP=vUAD
zo6_+t3j+nS#*g;6BCUKuI0KJ-$eG9sX{|~8KrBHho*k5BI7v!Zg5&@_ild3N?udp8
ze-sd4z&^VDZ4=Jru6bh-xl7i)!0yZSBITo9QiN$I=A`4#6qQGi@JaDt?IeUqV6x4H
zcHR)gqk@Xod78C?jZGwQeie0P9WghEgx$&YfiUREl|ew1y~Iu}6<EQ}VX56cBWVa#
zPSAo4VYkQzr92Yt^Aah|HY2|GVg=e_$#^h^?#OJ&eK-O_?J0qv&e#C|?;daOj$JwZ
zN1I-UZm-*cLY1*5aP+s~Y?tGyfY_9adUCb;Ud`kJ{#6#!<D(XM&wb%CUhu&uLFpp-
z9pNi-m5OykC%IaK@1EuC)s!johnbusnaBY{9uwBEo990HOF0b7Ya3T6r5rQKdx^{#
zm0S|Oxfr-9Aag@pP|cn%H%gwjHu>Ityae<Z8RHnxb31<YSsSs+)q(b|O2ansJe|if
z^PQzKDhslus+fhJSE427%mygs{mPz=1<P#Jf|m7snqiHGb=F-XMg+~XuF|0lbJt4U
z;#%t<CvM?q@AEKNEGq6%Fg%Kh@ufV4vJ7StGkiO6H;&*r-=o5gmNI_x&|$k}h@i->
zkyy$nM-DrOb}@g37Z5YU&aUtzby=gkRS3nA8q~Kt{ds*irAF2}?nib#y&1+8+LkS%
zpVXs>9uRaz^U(lIZ}&s~xbVY+MQahBI!u)ASQ}FR$!bE0CG~TsPC)~HTH{(L4fIC^
z?Y>+wRTnR7Ruz;#C!p*oooJGApo(S}t!1{!v5b=z#^*KVGLm#!cBn*VZ;ResW%qez
zZhmRmq}mrHN9CI(rQWi;sJ8n(_X&k)Nio!2P-cC@To>#dh~#Z9zVEDnLB@SLrYRq`
zCcT8M?a-cGPPk&>_VbJuQm8)P)X(5eCTeM770;$0GEKcuH;9E6k8-{~f}$9((GseO
z>(J1fte_0Mh}a1$GzCtT4d4P>)mE(`ECTpYjYY~mPt2*2=%|_EZeo*zS!}f`xWF5&
z3ZuxhQ$tuu7)O5#O$RtOX^6S2OHCE{9+eFE>aTLpz85ePZX4VlP@~us5|ktk<44@n
zvGI?oZ#}Fr5{l$(yoa~F9y(kPH89|_bX(ZxtF3O~Pc#{CgW05#3)J`9daW6>v3fN3
zj{A4vBjH0Iv^Y|8KjxqTXA1cPF*IbRG7l3Hs7|}2nz`xbxR1aAD>po_GPAx;0#Q<z
zNk>hK(00cahD4r&*P=FXE0XGc_eXGa3gaR<ktL%M1Vv}QJA*b?m3lH8@}geqv+k5{
zszU7y9^BO&QyIBOV%}M%BmEpeOM`xOIIKTH%SuypeQER|5!=Z`N99icQ(?zBx9C3y
zUa^~><?`$2+)Oi_rrPI%*$o#TVz~h9n=9;^65P5a^BydQ1MRHC9N<c02T6!xm{~%+
z6o7>G=&T;M5GNTk*J%;^MxF%cy@}v@AO#Q&?Lwo3=XnZb<MzE;LR+#M`}{9m%OxFT
zb*OLBwp*a5Qr_<)iC_8n`|p$KRH>{Bo6(}+=8Ca`y|`E1lo-Df1T9iJuJ0@5bobbl
z@^wI1NjAn`=x)e~nqknFbu;+?v7YpSX>RpyJNX8TP1+(*CHHj3mEJ?2^9{x8YAGKb
zX>K`ddrXp?^%Uyo-efO2R$L%U&%eT+vTM|;1CO7UOeA0O(jR;OBptxb8&dJAoB}v?
zCxa+6hJIWnLSb?w#H#1N5@a&(S-#RTAsxLIuF9*U3zg0cR4u;}Z#t5uhUH;+W8I>R
z&k4w2bxi>fDXehhGcv%F=g?p|SAi-`KrGexPyn-fW~AUeDWz2BJMIs_NXERqp5v`#
zr>+Vgwi{n9XrV~bSPeBU5TGkNKrl1bmj!E$FvKJq0fZuntybh|7n{ua3F=Q4q;B0q
zPlvwZ7CrtS==l%Kh(0=0P#bs@OAp9bV=!DS0??*r+uTBo8Q2XYZFFAUkwq(a*#xiI
z%#_;pHaqDgo|?JVG8~>TB>Pdwu+)IEJqQ1vy!f6|A|37PpEjvtf85)F_0<jF{FIh+
z5AGFe%$I{dL}4H+tleYJGQYv4?!xwo{rhf?^A#6`@rJbnwUc?s*V5^`3dUIch{++l
zre>f_m4>PpH4F<vjH*P%aav?=AzTkIauf=+ByN+MUPvClM4X>~n!Pm#l>u*etTmNG
z(Ew7)Tv}WSz6si#brn86oTB9fwl&j~fJQer(j5=(=^%pVaRKs!;*)x2&+OkimylN?
zS>?Fl3?m0xAv2!WGrx+1syA`x`Xwg$a_tV2YoIuY94nEp4rfZgcKXqF?3Zl5&qIpy
zLx?)YuO>niqB*s1Xk(jlQtWK$T*Jv(>lDq1;y3(FLdy?gSx#LoWH-LMEDnY;aUXcb
z#Jsi4NK$eLG2=K7fQ<gC*(Q#c<A9;WnASTy;B7J5y@($(Oj7q}hI#YAW6ugWkO_<@
zH7v#~A(%V2_!!cgC`=w1;wntWO^2$8Ue5yM#|8KcI^#%C1<Rcu^ujzbQrfF`ymz}w
z$4MA;hKiq3(CKm^{a0RN7OJ2}%U}p2H|h)mteL;XkV?dz6V&w8c6vp(`c{lt806of
zrBA!32>2knLk_a4#1MQVI7j^o|D#;Hz^F;P2lFhcBdgx85d^}Z&clG;B-Gm>Gu3X^
zQpJ2BxqV_1Xt*{OqB=CFqi@+vGeMj`S5&fSO<~o!q$&Zwv!=Q<_3n~#nC>`l5{YYh
z!TkkGS+AX#OO^I3AJVDlwBa7L3>vhCe6J|9y<ZaqBLr}S#uxpy;0Lj@*AKmcx6?g&
z&}q=2i$dHgKc>_Gva;RLewZf4kRX3_<6K-qpMxtZeq}L|gWovH8MLvi<Vy-Eyr$iI
zMW$uRGp?Tq@J6GYtE#=twX0*~tT)i3SBJl4MvyHsvwZekirQ>Qq>*y*X!da@zxX&|
zFH>w3w0Lh&=m~8g&e&EPk$FXeLgOh7c3xd^(8g7nb=h=#G55slZjZNlCoAMW#;@nv
z9?-JmGI5BZadH!T7*ToYViq0L7nCdUY2!2^&d0Hdq>`3o6_bU)rjtQi`+%^Wfo_Em
z)Z?8E3$Q7qV-@FDU&ME3&r+?_94(!b?f(aE5ANBeB|IGJTOcTfuV2Wp<A$42Y*72`
z95gu_xY2OrKp`p@Uj>NwoT}8rfpFoY|NfVho1&&3*zNvDzyrodwnV(E)(wKZ<?h-s
z59fY&dfgYXn8|<YIGIgPeQHT}MFq)a2+9_hxpiNEk`a^)k?^O3-y_~N()0A5T9}eZ
zdpU;f23D-QXr9XMCCThi=*Pj0Z2VaIajU}$nY;}8MfgSBs-cRX4c;P-QtM*nmOGfB
zJ#k$uDSiY8ix2F4_bB&B@Y9T^jB()yXLC4n%vfhYP~4L28c=BqjfVn|dM@Dka+aq&
zKnKo)9Aauye-M4klZXqewv|jw^?9H%-DX6=SBMEF?|Dr{3YMCeQPa90cv#^HV%xzJ
z^CH5Col`^VKnB4g(3eS2LdSiFlwhdce0l`AUq8I9e(~{~J^b`*Kk$$8`57(tuPHTU
zn%VRg0Yw@uK<GCYNF2UwHuBq}JqOu>^(&yTaG~yOYBBRt;hX7Xo+HEGvu-ol$Xld0
z8F_W}<xx8TD8a+LY2M~Gg*F`LpM_6iM!+{!v7gu`A6q%O^<58di#dmrC+Z1rtXM)(
z6dt}W@r_!gug*V9@Ip>Sx$GAy0x^{LPI_{P$gy`Uvm^+1)*UEo{rn+qDu+qJ!^awG
zq6llPv#6`viyjL?0U2MxVW0%Ytl6gy&+4D$AKKj8##@Kf=YZz2y2kdioh9|Agkhm4
z+ja$W8AIRW0(6Wh->o%OPRbcx!IW`yO1}aQH)-L#rJ${}fvp&(0E+z5fP7ow*^&%i
zf)3C$TF%<DlN>X;p=prbmaoH{5rdvTj#8U+qo60N?>9KhZL?X7TTc(X?a}5Ed;o~P
z6WH{n;k5-ou<k9kEu144#W8t~&j<&1yK+#+KT{7=7_%RITF21T)F|<zjrK_L5a;OK
zGH!mcO~(S_Pr;(ANxN<2LF`zpd>4Eifo1i15na=>+kG$f&yr~2#=-^ARoB6Ll5Ko2
z@{wohajk+i+uj7BU2hKQDn$omN^X-~yI&DM8iV!zHa2M@-N(qLeO9AMbn*9fCj#~*
z%rZMEd?M5rMpG<lyzjN5Ay0PbvpFyPNv>4Z;o{_{wat_vQ33?e2-@vTYrsPcj!9di
zD0ij6AFpUZrqy8`@iorha3w{Huj)|EXV61d8*x6=IC==N`}``jtCn|PuLPkvsEc^d
zLW_B>{OoE#N;4-Q%a_aG{&2V(NX#+HO}ur^ar<KAqNuE9ZRz}#%@Rt2r_DfKDk?~2
zhj#AoI_@<3Klg(iLCZVqEDbmD@YJ;b!R#7FKdy3_<{42(wVt>h+6Sn=W^Dxu_4$$U
zQR~+~S{)iUbv1;0Y#NsK-e7lLv{mh_1a67PY-RYr)NYtQZiKp^<6$PjerStUu+*{7
zOeEZ3C}WDf9K{CCnRe#>PwHJs=kPzD?}SiV{lR4QM~VW6Nvys_$7zao+PDvtx<p6q
zq&uwT&1ae9;z9vp-KhPzO~P8j4ygM*?%P6Q^tui2G2**EM&gba9R0DHNTnQTI#}Ms
zr*HrLoDc+P->e&za{^J8hf9cy)q0Enn?I^k*zGpD9u@>Doth!0j$2u+-($u+etgs?
z#JE<OsaB`M(3w=ZbCsg*-tx#mfS7tD2OK2Q#qh*QRw=0%ur<|jf{F<4*4x0tKD~Ku
z4-uR<uIa==&}QY&A1z;$-n0AY+W=5NufI}QQ@>4~O3DIXNul8MYs*~4TtlYOyrlg|
zcF-!eX*TL#=SI4CSRRCRTn^56h*Ro28>8t2KkU-lwEi&I3*OI&)VqW=BqT|iSY||=
znZwNus1h}zU%~l@l&7#voNGX<#f7Ek{t-l1tKrNDlvWc{&Bo9i&L#e%1_8wXh|gWJ
z`uCY$A|diP68^NJ_u}y!U_4nD3Ax2V7OtLT^?$fJrzK$kCdsyK+veA{ZQHhO+qP}n
zwr$&X_x>}xdol9_b(0m5RT&v4E-Bn1WQ0qY0~_?3kt_;XCKZ^&fX@2sg@3sYt&^`4
zV63$k98T5*IOGdBEt&2?y}y0;IG1Yb&;ZqNfZclplDe`!Nc;mfYwnTPX_~^44A_Pe
zt`2HCne(xK<Tv%{#~AxcFaQ&Nl*YArxXb)U52s6?`O<#nN(=5_sQz3uWZfH`)dJq@
z;GkopQ~Y!eu$Ia^=qs=*MQW$*56t}$u7x5YP02>9+yKW<th<)$uS&h%;A{zDXj>xo
z5R$L4mWS<Zr5BbCQoyK?K+xSSTfsBGoP`N@dn;&>&y%eHHU;P|mkj-lA8y#1?G`#Q
zH~f5RGJ5w9ZeW9*_<xqffk8|`jNsrm+i(`_ZT?Kis2FV=^S=V@^&5OJi8BQs<V^*q
zt|97t|7R<HkkxFI?uR6m*N6ZCf77i=LYo<BujPNDJRsIU3ms3nB%<dPzr&KS11?QK
zKupfW5UUs;H2&R1nAjShpW-d}kOSC(S9GaW>SKGzI)w5|EGSzBAm7{qc(XVs^mW2i
z$;%_$NP0_l*K?C6Je7m!400^L_P8ax`g|`pcbb+Q?hTeUwc)tOEjp?7!*HF^CgsIG
z73>gKCNIhMV<*tW$*JIJ8fRfid+0Grm<INBYV$_{E|@>_t3iIAp%oA(mpjE&a4O`a
z`7%y#!)2gkyKnE#cPUW6)h@MU+F^k&C3~Sa5OOr>z`Uy5R#~v)WKm@|TKQ(n1SLR{
zjWyKiBqE4&J9(Du2o?H8=}zpj>?zxXN)<R%K-wC+?V0L0WPi-$zGye1Urs9MG-$)F
zfKiO|?=)viS|c8_F~z;j2HExEJ0Yvl%oM=!zF5iMg;|@uU22IXfRG*xlqdwd2y$Xb
z2sd1|iM;b+%^VmpE<jf_afgfa*M<}e8}pdU<B>k^cjM>Da)EqKM=}2nU~_yeFgQrW
zq6zmFm2A~~YcV=d9K-kUZwsHQ#y;4~T9?A2J0HiD8}0ak5n5Ktth1qo*8QGi4w}9Q
z;1NH^T4OxQ1zSKO>hCRAIRA|<VmQ4|N#fAj(-64~FX6fc52%`}O>hG>$pqeW+CNuY
z#&m_j?(8v+UC-SAE3XF`V+fiEuRzVIiap+cUYKpk81}iC7Jti{9i6v0?(T4CR<j+<
z9P+edf_43#iVr;0U2cK(X+)><f<Ce?Wvi|^e8>N{NCNl+oKaM){K(@lDO%9OlmT|i
z#4bKz60^*Uk1jT)Kg8d|6gi{Dm)e6W#6Z)ZvV@T=yxFckI$){~J!Jet=3<Z7yZIMr
z18gf<Km^oDey}_l?uvd0F<$4{=}B3Ir8_JipQh}vD<gG*a!iw(Mayif*9i94m;pA2
z^%I~vC2O+L+r`AVR77YW6Z!LQ*bG*FB7)OPcbE}o*g6*85f{+}+R$D2$`SlYr?K_?
z!h-gF|IVV>RY-~_-4(ahmw;l7n7}5HVzh5vIt9bSg`QM$$2u!j+6Eq@pSE-C`a-DJ
zXrYC%T-_ZcOjVG<Xaz6^vvm&83U7vqXLb30b2jzePYh1Ifn42!r^w9RKgno%GtbDg
z%yo9+0467yne(B7t9tAUXtOceeU$||PN5+iqC{qhWp?KV>f*d99foy83`?hBsWQmN
zjI|pZ=s4U_h-r8=f3b2?S?W{|Ecv6ptFoP*`1Mi{tKJPl;Gj?ya8DF6iXi9><a^G}
z{%V1H^n_bQSyc0(LOwBXr-qgwrPZ-#e{p<hK2~}yKMUiHU-+6b@f(;%X|g6WdxWGt
z_(-$8RZg4lAk=}zG+uZ`JXSYx`#KGB`8i;pASfl}L79#1H#7%fR1#s1xe9j{m<3R*
zDwCOQTmJ}tXmsa%yqr=<(k@b!cs~k5jwghEa~)+#Wv}F0TnCN~#vL3K*5Nwt(gI1j
zlxm&M7z*d<&Q5{`F_WvU(Orl`pc39QPq3;~=K_3<{M|yef`*}?*}7PjHnEK!n?2b@
zIvC-g(e884x#Hg1fF(9dcP_8u-dj#~VUtbnqK@GZ^1>B!nmuJMpRg0PjDgg)8lL=B
zH2+fUJ;RSapv)S4(HDj9I~)isWZ+m&4-t@DT*F8R<7jDIj${3w)u#l2C`mFZddyLQ
z9z@3J<=wJ~k8T_YP<n2Jx3z<|_?lb@4Mfb1y(>BEz2TOf+8}HaU(bzO>-r~^^#~WJ
zMXHl9yXQrwPQWje6zfJ3K-huZI*Snt@-vYu`1k3okdhn9_Zc?*>?x;#V6rbKGFRFG
z5r@#z;n^k3k>aG)v*;3{akcH=;%j5sgJXY(>!<TW_qD~6mrG4>*qC2GD2Z&{n2UEW
zsc?y};?uO5hFIP3WN-S9Lp^!CV#8`6c!QTPc0sVAuvr0xjOP4u)Zr)dVoDxzDFy0=
z;Z(;nA_)rzGVHl?T`-Y6dA0SU#oQX<s%wd}e%zr-7~NOSywjP5_!w_Sldp_TP|!%F
zF3Fg^!D7F;Nrm<HGV#H)Z#4MWkQ>H6%lPTO-;%rjKqGjuUbws4{}+eQ^ppC`f6QwK
zzeNE&{t5w~lXY+dDooT}PV+g)o-jfSkP)prUAl<`m|fXa80x<nH{lYdloZ(K4&bHg
zI`(JcExphResQTh4OIQoFR|-#P2`Z%^6ozz|10(@#@_8*23YW4Q?92ZHOIYzD&jb=
z{z`iVfEd3WZ}{q)bwibC>)&!dZe>E)?oCi-S6PbgH)?2p==rTg1Xh7~<Q}BVv(^WP
zq5QD|6CPSTCCY|mnv+C;hELEt5DKg#uyV^gca+kmrv3<aJ>g~b3rjK0gO(Mxtvi}R
zdN}cFqLh4PE>@&pXsqwlAs{&0|NDs-mUUUR!EssN#Jah5<Iz&DQeBpA%2HlB?uQDJ
zu)GiR35vss{`hR~;ewF;e3h!uk~*~QptflThmzy%B}7|+Y5RM8=2@+>QHZ_PiDsu}
zNM=VnH-q_G6ab&kD}CR51Zj)Mpxg<L|7GNv+kSDA>x(U}K=`rdMcAnQqgi?2KCW4I
zP^s<<X_l_R2a}NWgpb?g?HR!wT+5xv+n!P&dYOOD-Hw*dmM?GCRypW1EkXiyXxA1v
zEvxAAjndqpcuH`cgQgpDBWOcOv>c2*353|cpM#@=j3s~bxd1yw6Fw6s=?ZpWE{MqF
zbab13_VOvE+~7tOX4|#4v`Qs>IT}$@>l}R7HBvH{=Y`Bx7qtwvyFm0K)?)sHn^S+H
zlTk@TZ&C|Y^Icle0OCVxGC+b{%3)`hd?s1Vbd(WtM@cPVdN>w-|COfYH3aVg2{;>2
z>~Opw>mxxWKo=RZTlE<q#65pzxeuru$nSk@4V@hcG4yju10u_{c-E@O8B+rhA{u0f
za5pILq01o`DMc=QIo$=EASxS(FE_C$X@A}qT$tavFV5|G=YzYC&XN3~fFcKRUP%3S
z^aHN~Dox@)kSTYWoNEI)&IRGty{@iqBNDId0ZVb6%6w6v99;{s(YvdMq(MS09)><A
zgOsd5_ICS?FAv31w$r4WHq7&h-e!1pvG~YERwc_Oq;--_g^Sqhj`{Py4(2p_+%h{8
zmDht7;?q9wgf*#;7<){Tn=?s8<~47=;|EMkKEu~yL@Fe1)8G@@<RH$LM(+8Tp*xj=
z)~V>f?R_TtZz^QJiGFuavr{1zVKtQX`Wl~kJEy770t)*(1`)G6FMx2{jTI_FNV`KD
zGdbO#<Gfl7vRNd^9po<YV?q{1S3vamMmLPK1ME2y0KTP`miv}?=iIP+ILr5@4{_K1
zJE~RR{Zhqn{FUqt08ee{1(_4<?lxqA<n~b_C3w}23-`_~2Zs;ZS-uLBX333zFb!Kn
zM`bydkc~jcr7_t8GRaE%D^}DC1X%AH(8&vmCf;E^D!iWvi{)>R&7~yaT#~=s91($q
zV-v||5xZY0o@Nnn4%ICI;#)=R^%+CWlfUZXX;*;COJp{`X;T(y$%xaq!}s*lt@cKT
z`p}_~O3czi*HIbUa$2}O-9fbaCy|b+?<BWWTzqpQ?+5~-%-t{GXT_r=Zt;7bx?V1Q
zVE`2-A8D{&gU+3000ThO#?2n7aiBwq(((DmP!hsq@|SIED0G=mVjbILb~)K2s*SVb
z;QNrL2ve_S?f0l!QU}SbKv+AF;0Q(G=lmEl>j~!Az<04eW6+O}{UEMDXKOnJ5XsX@
zS)o@<99zA-gou6M=AC`ZfbgwL+$$~_l%!WXhI?I?I1b*vG2f9+kA804CW&zB;2%mn
zG<K=v$Io*LV|B=4FNMECwiE0^dJ?Gdi9-4m6j9M%napMoPB)j(+|UD{yxYf!v<+eM
zGe(Cb!V53O+vr*p1kX;zLbc+B@iQ~(Ckm!wRdh7N;>{Kj*z5GpeIC!=-*cNn=tyk%
z*x|^kF<1OjIgwW+1MA*psH)bPYo+acBs6PFUB@bkq7$D^0pM(JQ7On+9(iZ6c^Ma8
zSPZ~}%#*ewW+M^khPDuVkC!JeuPj8c*4N^jf^^JaDq2z10#`xUOQ44~|IVruv#U+|
zAy9L<qmlwwe=Ks%pl4Mosec>g<kYIanch*4IEM4{pf`eiGV%fW3V^GimR$yG#sT>i
z1TdG3>dPy7G(Z*sC3iU{W2!M}$iqymD3^0`I7<;2S0X4Y_IV97^$Tl{{?)^oT*ExX
zFa)5S=ejkIO421mMCn)>g*bbjfPQD|j`0h>OtYrY!`oUrkM<y|<R<tYK^n>}8VyW1
z32~xNg@sZg49`m;)>O(Qi<o>r$R+OVJbJ-{AVPrC$B&d^3K@<B0X9F86z#Y^nSk;~
zb~}y<Z$g(34^hkyQ}pwWqiVGo36gI3RI)JAKOx)>SxibYK9HJ*{8#{359*bHDZYVV
z3*F-&+*uxX?Pfb_DD-UE$h6<f;rLG4$6rSp?ceqGsYPG!^I}Ydrm)N{Uy}r=JfoJ|
zctWY9&Re<wgG;%mb;czHg(rov@(pBq3yYmy8t2IZfYL!!bb4p(v7rUWQm-9M!=kK3
zLS=Ye)u29`)LSo=u54K=;xQ}Ho2MEK)qCBP%P<7Ttn*|iC}@t~ywaB|_>0!dN>Y*8
z%8h|<pF)3oGHZLsmIoj|suUo?^(xh7PTbLyF)MFXS}?ARC_wFO34?YgY7m`yt#zEA
zHoa8e5?{<M(jeK^$%oQEgi%;+Sh|bK*6P`axS5C1$?TIR(h2uvh!P2JqHLD=6k`@J
z(~MQ!LwWS5#_Qa6ik`L&yYV$LA)OoZ6eDsrx?eVM;&DDHhLj-@%0Q^TFOII30fS#&
z;U#xGm`BugE6r3AUO5n7v4NSZ)GB9*c4*h)H()AdsbbylB@`2gG8c+l!^q+W+*Yv{
zD7>$TxxTE#p;tl1C0)+TRLL4_F59J$C=TYfG^C3X9h|eCw-Miq^F!BzN9#j}{p}Kw
zYzf~=k2AQ=QpLqF)*%Hmo~>TO^B}is)pg&%Y=P({?e5l**V}pJpz$#tgV^FY5i^9B
zXplFs2>1@B1)~_Zj@1R#mbx+l1@Z?AP<^55&@UbA+oP@C^1%{@*j3B*JUDdx`FIzZ
zIn$axKa_0MF17BKJlZ+SXK3MdEI60(_xALtW|m6ae@wKHl@(@Vcf+@kCDs!UKY`-;
z8T&y`;jrF9CTH}zPmPx>CQej5s17xA+9w==9X>~sy-Gb9!mNS}FH~kuEjvQ-MB=$M
z87ELGiC{$U1yUAP@W0_FQXKz`c8(VnPA%drmC_lY?L8BsSqqbpr&vzAa`1_{&%k?*
z+8|c1v~<3mFeZ$iVQN@tO8j7Iht_i{n;1|8%%vGP8Wn6H?`+9I^_$LKJaH+*B`zV=
zK0Hqb$LzZa*u+I^=z`mO;vY#|mU%A4V2%peh^VDvT9K){pvRPlG_PYq4*8!v;3@x8
z8TFN%-J%(b>F*A|eByMHuGq_id6KS>iOnt-9%1Dk>rS>{KVIl+L*SH?hs#+0_;0HB
z=9{Bi<H8izHYdSNeR#7xbI*7NPJg<=^gO~kjUmJbz~(T_5k}OF1$W(%+3+qVI4P=@
z)9O-t_AkpTUZbb2qV4!Msv6S+NFa$pV;8CqxgnFuNa+}>an=0;8r;E#@}+O8opxW;
zCF}Hyx<>C|di6loo>(p20!zg@`*3Vkz|?n+;jC&eO35A^sgd#&N+%$Du915dG|t<<
z$z8P3c&>*K50X@<^_GR7LGX0dp`s^ohMU$*eRR5awivkG<CEHlu;aFDPHwf1elKs~
z@jx+l{eZ+|zhb5NbE0eJW&bO{G=R8BAY{=yf&ii|MV-%bc89;A;(w9H={eFzJU&;z
zK(`Bi*~Fufm)+}EvLJ238r~9B>2RSj%z71Fj=`EM;nZCFbI(<*HOhX)TWa8=w#`Kj
zKhE%V1s8EIt;*ghRwp$um&rFq^2By4MUK-rQV6j8<Msz#qKbG()pbXjsh^Cf8<5mA
zQuexP*6V8U-CKMIAF@w=bWF-%s}0aXAh*ekI`kioam!bz@bIqQK}3ub1p%xv&s#vv
z6xf8|))SVoEglj<5n!`ohf)u<k<1fY<+<%kq4J=IpVj2_jo~Yx>Nnh*o=XgA%kzY-
z?V4S4;<<A~FM~q3w-dvX3{w<E%M`6T3=-3|7ttvGbENz)`oK&v^GZI?5tZ=bmep>x
z6QUXlMk1F6VWxJWF`0Hbju8|&=LlB~CNQ3BOw)y<dhz+{^ijv=Cph(3NneZHK;61O
zLDNu%pop?G?W505)@Z^l72iR|r_HVLe0}@Rng~W}X(R=g4Bnwf0KaPVrrIt`eskcF
zBYM$X|3YOf+A$#6W+<mwY03xzaD|3?;l5(tfD-6%al?z^E9jm;rCuO|2l3!h+PTUu
z)z64#<pTQ;Z$x;k%cB+PnNAB*-a<JiWen4kj)fJoOygi`6<5Ml>{_OsCw}X{lDm5F
z|4A1)olRwz8wRg5fcUwb$TJv9l0=g!#H$<XWu4=mTQ9OB9<Fu1zHb|Pxjx>n#Ot=9
zKmW?Qz_Np<Gj@`MZ3*a1&n29b87;P%l;635vdrTrey<>t)rt=+T1NL$*$EFQtsAeZ
z;w0{$(-Zo+bFR6=qPZh`)vngx&G<v_w%Mn~yBz+jNAe8Fv?U+I;ecVh&gfMes`%hg
zZyFAhc|umdCxR9S_D|+tG{S>-xC0I?73u(tF7<*2A2gAoGIhX@>n(G`M_YgiWHw~c
zCpQ+>Mt5W(U(y+1*;FbGz|T(Ij8XOp<lo_Frw3bf51c1WZuwz%tviKlA2tezAd0d>
z`E?6x2kF>6ff&#RXb!wLB4{OXC&p^m7i<G16#+Ry^l?v?D&ZCfv-W{LchkjjP~!+~
zdn6t8^g|ob0`uf277Pm~8lxnKUXbJmWdh=03^skH$8#8BU7pBCLDFv|a;yaF6l-yt
zbdn7oe^J~wc{?q~8J^yD<)yI=wU~K6UM=0?$1w!D_f_}>Ze2d+DxBC|0AjxLSzc#Q
zw?;ANLE4ZO^#0@wazIt?(-W+g_V64b6{1g3>#@mI6^pjU0FZv<Y)30Kicc+Gtu7mM
z!&4*?%b03W0J?J$Bx6ik>{M;Yh0+vw$>H9L&1|xtbjBZ&=Kb>)X&Jt6lb6*&{DRB`
z+kTfhii8Mq95B2XqCYRpvNdW4go$7{U*(<~gX}nR^Y<oT^VQ=DZPBa(DSM%dS;o|+
zt16%H?(AtZ(`}NiM%q1(O}|uq!nYY#D7o=nHrpF4be;P31NvS!q*k9St4beoL!a&a
z5Ftihm-t3&eS*^=hGD3<R-&s@nhXEg{NJibIU7V@=JSZ%-{I@0C2Mpwi*(ryyr<uE
zUeLrQ9pU8DNGn28cW37O<-Dn=J*~0ZcclPEss!TSB`ZEmzDdAZzw%3pj5F8Rs<Rbw
z>vg9;X{ZYH=D-bst!A%98d2_?=IM18GO^qHzdWwJnIGGM-jbEiz9(Xm04hEXaPg#K
zV<mA}(o<nn`9b#<vEp^s`kfQ)#o?3}#{9SdAxNil@lVbl^+;Q2Bw_^E#cxWH?M~x6
zZ731^u`*g)3$Q$=&JC-G%*G9ruzJxsdaQ}M&1dt5(o&it9_K}+JcLK%Y0`+?)@k5n
z>w)ABL2f&>4xoaM4C|jPf4Kr-j^ipCv{zrz!E1C1b^aV-S~{G*xDcbuy6e+SSE-Ie
zZSPNxKDQYlm=#n{9Y{i{6Sn>ygZ989L)TMUB>YKb`RV85?oqkPcFszv^vZv8mo>8a
zleI=Mxk$>OtJsw0_Gj^$m(Z92coN9Z2(RJr(zAnA0QkTMm46{<f9|d2^;w7H+viM^
zc-rOI>!`YODUpXeH}{RBD|!_J({zHRKbWE#Fe)ue8p%Bp>`L<%sOBUDR41bpSb*TB
zp+R1fNrkFuBM%f>n0F3t95#Grfgk}`&vG7*wN{^xn!2BzRE@<1l}bwkr=Ci0Z;$Gy
zl|<IaFMzuJ3(pV@BQ;5S(v=_5<<p#o8p60(N)p=VcTamxqycOubSpBacJ9JE&U;-2
zh6mQOccF}&(bOTW@}V5xytRZ?cw{P#;~*Dn#`)+&elNi&t*qpUq`3M+;605rZlcmY
zy(VcpFHmFf_qjUt5lfDHw?27+K~h%{kMH?KG}ooJPp61~jkre^YzFXHe&ciI^<(;Y
z(L<C0XI2y=<U<sFN&4#lWFM+Q7dD{D__FJXdf?9T4U4%#SDDi09Q=K!WopttF;e-S
z5@yj`l>w8*fo(>SanMEyBxQEXb15>9go!EL7rq5*^oWp}dyiD^Zv29`xTdK3MdQkY
zWgne|fRysSS%SFm7(E%ojaPA&3PwEA$Vocr1@8Fn$%eAeI&_o#d*3Th_B9ha-#d7%
zrGhRGziRfK5fV*fIIyNIE>fMmiNBq}K&L0LPFUz$^kfB6Cw{?<kA%&H61%H;L>h>w
zB1uf+JKITb-%@rUE@ivF<4EN6!fsqZ%0%wY6&Sl!ATmMs@5?DNgk;PzaOHq<0u^yA
zyyfwILgaZ4>|h1Gi~VpZi@ab){hTu`0ng8Hr3=V*ORi9sNy8=J3CN0@&E~j9K%Y{o
z>>P58$onPyn^myU{<TkwZDGs1`8MWnM1l)W!z8SN8msjrnPAM1K*O6T{VBZ$jr+4%
zDH2RioZzeZzvYsMM4H2XVT*0t9UBPo%QAiB&>yCPsuzPBmat*~fxZT5mP_~%NT6fL
zImDi}BTcPoDM8c6rs@5B)~tkt2fF&f-C+AE*czNO%G0@*6avo})*T>$k+6tu^+?zc
zfVDt|z>i^|?u3<3*^n$KMP`zu?sqkWroqto^qzvDotTZ%uRel{Cob+X#*~NJDjIYt
z{t<!eg>*rqAfm};<a4qlWGqw2mZADXZB(??sZv@n&+}b@>JpCM9*fdT%<ZEde1%+$
z^kqq+rZC-?xr(?QWnnBCM<3bfFhmRIigdclX~kGMFSf|Aj8XP$$^AgwR^zFQb?N-`
z(7%oWXstn7TCNq0X7GoEylQaVnDGUMrRc$8<4Bm(I2{=JUH{5EgfDwXKR&fX)UY0&
z1(%d}ptn7U0oP9-Q2D*gLPz+dWOUNy)Fw&n1Ve4%h~kP3BWmq!q)n6-<8sADdE#Gj
zR$>~4z5oBQ=Gx`we(uS?PTaix^q`|Oq<JE6De)Nr=kx8NNXvr@BFX7Wvd?M3>8-cY
zGNn5QawRvsGe<lbyK70SBdK**Y$(1e1-@#+OFmejsRs&VSxq@{6#iFta8HBM8)u{g
z`~;bA6=JxikLylxQsykQ?)FF+2N(o8@D^zR4UYQ~A0LP4h0G}xOPL914S*Go{T90t
z0x#8Yvc{M6s|1NAG2#_k+o~&fFB&C3c|4+Hb`JvuN4lu@z&_YP&=a6QCL->{QKSwe
zjk(;nj?d<($#7K`_dMsUW3W;by9#xO3pq@V=eDILh6Oo~1ddrD2Nrv%^GTK$2wEu7
zW5TQNe#&|Ox=QGdF`!r^P=#VKJe}PHcSRtE8Zx4S2SMI}lE-mtnKVgUp(kmrg4AdR
zpgbRf*Nq<$nm@l{?7(!?V9zyC)QyGh^z=wQ(7ZKG_)ELn<lkyxi_p6I|2T_TOW1c0
zU<v*czLi(9o5PG&-f2Ta5nIVwYICIXGceK%&W2YUEa$*b>}fBS7SfVxWVzj3>C})5
z7c(eqO*kVc8@d>T=e@$A1q@DBUjH@TniYJ&@!W|I;&){*m~?qc&D&~8h2wa_I+`*w
zDr-1rn@jDK5Ki9dOfpn^$r7oCYsr4#l2JETtEjeQbJ@y!pZa3v!F>A-d&OugFfjt1
zn0u7V*dPu+yW6`h|CzhwHr3)-M$xj}YzGPYspM88-+g~C{D^|ElU216SmuBT-7YUf
zczJl?y&bX*MwAvvI50RYA{>=QJHrrqR(HBIHo}l%us31csE<;k#ONIC?P)o>E2%?`
z)&6iM9<4T|kV9FojA}rSJ(+1dY;S|z2=Pb!+HX@L(>^PBqYjUQG!-sZ<)66vSCPcw
za6@bJ<s-Rq8!$=Ds?(Zy+F`(JZl~Snox1H#pv!j2MTVi}vfy@N?03jA4nI_7`M`r+
zJfWf&R@)xKE++k;*I68F6{#`8M}7gHx<389Bp-;2`Bfa?Bl_2JF3e9pt&;N943tk*
zh0FV0&^jO|NR~NT88+?T&NhYLJK2uRvqbJ<S_New)COG1GRmOuR2Rx}RqNb#+U#Dh
zYi^b9<naP+Q>A)n)h>mx{B9ZeKL36H3?PCl>6sjLOO%&1O{7f?Lv0}+X$+VrsV2|x
z<6Q78`p6-@_7Uo44~35a(jTpCKr1FwKvv4GpBs~lMZv(1qT;@cy$7Q=-=|D1+fKFQ
zodf(>*X%z?(i}+wS*WXnc9UTC?{_u&`A5nKhd7EqDUDX|GyJU3wK_byEvv917yeos
zLA~UpDC-&yRmcjN&Qmy^^sug-&(e3V2^uj2N5c(%u_cE9&eAZp%z385BvZ5+Sc3+g
zVy?pRd!a8-5Q#cF+!P8f{%Nslz|HD-n)`&TsGFE3dyfb|8~;kwBC_4vYkTB0jfJP0
z7_&H~p&KJDw^E|eK0R*3y$z#{ew+R~+Cl2sIi(TnGihR3R;q-rCtr?xdzGjIrpAG>
zwFB2-^HcUTm;z*}%wWG8G-dE)0)J}CkHRA$)21{5Hp`Ac;jyM-S}o+{WE{7I`5)~#
z^ttI=G%@KIQr@j2(qaH`<iLF-DWqX^!PblsVh^J{YcM_QQaTG|FxmT!-|!tP6;XTK
zWN?^9m9oomhbTv4>odi=NYBSJZ?13<AD^$S^@qt3+%EP2pi}1w$D^X1+gIRSQ;IcG
z?SZ+Ul=re_UA?Ka^P=C1dC4<vP)Iire$y64kqVFx10N2`LezQWb0cSxAe8)A!O5pi
z9Ba+ZY|V-Tqdoy<({9WusTBtU{GHOKFfqYT?`s<PB42YenqaJecCC$S*8Y>r_;@Vj
zv*F8zFwdDb7$H(l_#IiVeV!fc%_%I%FIh`QyG1yW+qH+OF1fgd(@XrM>|R_LK@aE~
zL_pYX<DELps~kt4P8+Wi(oe7Y9r81Nan-cIT8k>db0sa#;jAWA{N@_rXk|u{O<e4w
z>L9tgI4SnX5gntV@d&@Cey!Id>b$^qs`4#Iqd6s4F6)rC_HjFeGT}X)Vflhc``*j(
z&Bm*KUg`8giDk?f+e9%l3*45|^g|kSS|Ah=WOK^<QO{~t=5;(wX^<H_3bf6CD|}qp
zk;2DfpJVElHh>K5I?M<cJt6Q3tiLN8b*I?lQZJ$~?=4-s*C96lM0L+Mv=W=18p>e1
zJ$AN!Qs-@}u+D1LDF|KqgO>RF>=T*SxHUZEUfNYU1d>1@b4E!JB@(s_W~|Z`SaHXM
zU#;^ja8+G6gy*Yf_+!xPcJ}-i+=9O;KBdOS?j#{6F_lG7bTKYf%rajbTkM)=`;!cW
zf^UB$pc<auFPSQdWC3yl->T1nF5z5$I~K3sV%t{BmfP15{($Nqw9NG9D<@%+u1{Ek
z)SmtO<=I8KkYS~giI4W!KlX9%_D?K%e*+$3kRl$qRfJs7F{LRx7<jl_aPV)jMw`HX
z3(XL&=O<jp$}x)KgMY@a$6LfJeOA)p7B+gfn#*?-AUX5{ks)$K3PsVM0$fh)BV&|H
zcCY7@4GG;O$uGm~b95te#MmI>DuJ03(1}(v>93h@Y`)zWv@`J}#j*5hK54ek+Y)b6
zZ*+HVmyY+cX%&$t>DpOxH_vXr87}BkO$l8_=oGr|G^owJLM=HR?}3#RR<zAIY#`=(
zJ?F_+xVh{rG`w5wwC7aj1G((fb5xO=&NgHFB=APkIs-jw*sy#^0#jsh_PYp1=X7J=
z5-_IB3T@&xv%|8_5m}VzMG;ibt$1H>hky{)yLg^Yxm&LJ#8^I@yj{q)#-LqDo~>~z
z`cSM7QVCjxv4QsnkCB~5+<0p%Sa`hlk8La}PGqG!9m6=acmKKRg0j=?&J~)wh7KYb
zO!uq06&RkCa0%0unE>6amS_3I@t48R%C~PV^P`RGw?ZyR-c-HVCCcIljBoLk-`yB5
z>sSx;xMwo2vd>B%Y<w~P8aP|ok{1=vyps3nf-aP9AT$`!WavRk`~wo#tlXNlzoG=4
z^F`#Y>LRdR@^A=(YyDI*5*=nBNY^LXI9s$TUGKohgtd<5_1`_ZpT|*jIm|P$u^5bq
zpG;jGREAix6xT>R+qV~{GJ7N#FAkZ3R`igND2)NjuQDyxq{f1EvC{SvAOlQh@C)ba
zxxQJ%5?_QX??3flilC$J!7l#}>_raaZm5xshL>}!?^q_RuQ;O{w>OKqI)5Z5q+aAG
zMJY(IoI$?vJoZ>b6Kw}O@M!-ocAqqq!d<zCBtR=kd;Y<Ja51iAE;+or&CW9kJ1S{L
z{aKemkPpc3o{rj&NRy&0Mgy)$GTQMW-rQz*xPwkqnDKFwZn~=RnC937lpvX#(Yw>`
z8e+qK?n`R`-nCxu`8ZDlT-x0q+}f?fW}OV+SV1SyC?1W3AB5d2F~59~2vV<@H4hm@
z528R(#g=i9f^BX48=*`Nz9*2adc)(`Br8P>+|DwL$Is3MOZ3}P7niu3f>HlXG|PYS
zr}GEDVgUUyPn3m$6f$vj$d=wEQB;2x%-GjKkjC-jK1?-iNjvE*XmeWls{=EztNK$l
zWi#D3NNYw~lLJ|t<(AF)h9CcU8zcO;+gVIAvxT4qrXp%`d%QGRNlVT?hE<O|qdS5J
zZ2kH)VvQ|3XS<14xq-S>IwXE0P0V1Sd<3udelc8Jnytmd$zr9#%rTvINoLD7k;Y_j
zG=5|-VfC^6h)+~QQc^b7A4x4QloLGT_sc2lh{i!erg>@4NB@BxIb29}gWy1-tifuu
z^jkxnJfgz4f1^=<FvjKA2~Sd%DX@r~Fa9(l+-{N(%Q+I=B<1rEaE>kSm%(xdHGp@`
z8)4T7lH9LAeaw4v!5N4LN_{Zp*9cd+T&QC*3RnH=Y*qwU=>8Bs!2K4MeeEpP>hM1I
z90EGcUrI7N25;t$D66u9=-j91f9HOR;4OKgpeAG&4luZr9O3YxFPjzF&syIC$aPts
zdy_0tros;&GhCa;0rz0n23L!(QSI0yNZyCYkFDF#)}@p=__hqKdjtQFJ$;)o@6ba4
zllM_N(%*^MM0ktL4N0`Nz_?2e(ht=<bN`Zqd`Z*R&;dE^+miL$6S?4vdkog%vk}|K
z=GRF%hXQjfFl4c8`#Qr`gN`wRNKKX08dfxme<f_zr%k0i9VaW@0#3o$^Yq6+g=3!T
zwm+z2%&V_3+0k*j&<c5Hx&Lh0OCN!6QL${g(IbdXkm_uv74h2$PV@G4#9cnU)DF?E
zd4y+DQ!yfBC4OwCHzY32esmQz7WK6<^$u_iRg0cTJku<n*de5I3N8hTx<$y)#z5`x
zs1HQ3pdTyGaR|5i2QWuG0x>?=z%D+d=2Lj)Q(p-9ZEKYRCE+KHV!z=<$Jf~$5(%#|
zErP+&*!v&pPE?gqY=EwIBROWAwihzxV52%dTugYTUaZ4&#eKDhIun5y|7SqdGfjr@
z1E5$=@nRxZZ*02Inkp1pcpG-LT$g)_i#c+)tH}@bZWO{3UM8iV#V?m@du-+T!Qac5
ziwKkAiIb5ws0rvu0z-0K(CM`^&^}91i?htsGFp~CqTDml7(yZBH)fRqG`-s>IXvWZ
z(8L~1E`)A}4jMTc5x(_l5J8IUy(YX=K)T=5@O`2p*PDy?)m(h9Z*@b^uo)3S33ou>
z^}#i)h|gAPY)?T}@WI8iAHPlrcJU2Mve1#%S%PX#Ivc!Y4|eA7u-z|Zkip;U9NopY
z)c%rJ5Yu<*$=>}z0p7#2f$98oa=0cl<Fx_{y0p$nlN4_tM$o}(hz?eCs{?zQ4e`DB
zV%VD<@S87|3%LYnlk6fZYZ;(nA#5m_HG>3&=oQF9C$-bx9XW5}&;qkG6V~z*pBm{D
zVE!m0q9u}Hasea=#6P1tP~$mo5#;W6RDQ<Kw=RhCr83cMABq0@z^q@_xak?c<b24k
zTBmduot;N6#SlhsW6dDBxnF*!p=Gv{CP~w6Iso-Zr>@-jK+g86B`>-lpADJr#!A`7
z+u^LL99S>3KO-)~z-P7Mlg05p;N(NHCX~#!*6Y*mH3;vd7EejcmM>25=GN!TVPvqs
z8CG?>9G)sgd7iu^o}0$|qelOe(0rwW>zuPQ+F8*@Z9V@Y_dI*j`Mtf-Ag>-S(kSuB
z5241vtS`4>;UNWq*eU?^0Qymv)&i0s^ain(GG@Z;p^3icv-?9f-zkNt-8wSL6*q%G
zYke~seH8V!WY?or{){Mte-J|)nb#P;={HSke&i6#{HcHU1a#cs$7P&uRaga$;j;#k
zhD7Q70C@di;CZ%PU*S0{Vw<<fT0SqV*uk<Mpp!oyr)tBTD^o<7g(4K=G!a5<Vz8ca
z*U7mJ8Wy8%+J1tnE9O&N)xh@-y@hMD62)KH`Sjud!#%(oY%cd^hrnTqB8y#t&I?H4
z@$nE?p*;!6+F=H1GbH#;og+d=adOb{QS7!DXe2RMNamZZOb)TI5i7y}#gBlrN(pC1
zd~?vg2KgsOG3GKNx7DL1nSzATG{#P}18mJlyhpwP_gBKnj2r}`7f!Uo^e^b$0uU>v
zXppXiWui!Jw9Czai+IVaRSZL`UNh8We{Vk}vFVK&riX2G&EWRdjsQ4kPR=uwC++9>
z9jt&$3|#pxTI8+d!5*rk#uybd=8{g<YT3>oP?lIoCk>#5-8gys3+MTZ@mTS&NS+}o
z_p3=Mgx?tuXIrm`c6{^Qdu4bn5$alZ>;mI8E?&(CV6e<1(UBNwQ@jhr&;_^7d$SNj
z-niNoI_`lbp5+ub;fV=MHW^i(FJI2J0z{WhCA{aYhlD->I=R#IwWQ&FBIGvtpR)pw
z%u(FPP~XxwU#}dsu0M-^SO9=TUghd<+CM{nP|4bUN3{6C{v4=D<pqV~=EYXztVmo#
za_|Rx7{vJBpYe^;g7bE}VKDtK_yD`w?Z1adbv~nlDUMma%v%w0xUPzT*H0)*XD`ox
zm+GfOR?zcIISq4rMNH9l>EeAjJxu5`k<fi<s<ex)V3dH1YfeCXio&0Ay;H(ct|oX_
z4M$k=#WoBd1lppLFYx}*vBb6u*TX=j^Y8Nj<p`r%iUZ`AaB$%w3m?-VLzT(tw!W9`
zQsFg%szN?CkRI4{xwzrBFu?Al3%}f^yym-L3R<Sv$))XFOx_m+yzBm(o3sCJy$;ur
z303CXK#$<RuzS<N;gPm+kgpf}R%I0+HdyN*HN=v<tc><#DNUUe9Z)3DkIbK`b&t7I
z${6g>;ouxK`I<edvK`b-FDm~FH)Y#tL}JXeb1W<h2IhHP<h&boFEu?!Miu<4iCyFO
zdphb+F^%V^(E`HcxsAUv%pibwLxCW|C3iL5ax1%t>q6%+rFnlNsBymxjPlg_-V%oM
zIk#>OX{_Q6F=C8ic|$oE+hKUGNa(vS_Yb-y7vIcL^}4JM{54hHRc;>GL?hg4UYbHh
zt)#t4KDHn;b<Kvmmv9{y`EBkJ{*Cf1GdnZg0Ym{X1;iW7QxA{wR+*;<?c-=$5z>uB
zl?lgqPSPpDKadeI%f3=F0N+x$NnwBvup0cTQ9KMwZh@eFBpxhl(c_6fL5}2MO1QJ;
zl%cYp=-^B6TeNI4UjY{U3+avgqYLtYk;j-#4DMKPhM<+YadSA)D+ko_!=N?1Un+wI
zQ1`6|Tah+x*zTD|X(B^lxV5*JbE2l}B^5nMVxu*|AG44rhn-Vw<y0Bjdm1T9Mm+yY
z_T?>BLkn&7t)?`D3tV_nls77UaubzpO`kxs>yBk@d$^0LX)1;)j!r^5tcw-qBD%xg
zB!8P)Qo(?pc~)kwI2dcLuW$94rcQyYDBTYLmy%E!^i8k@l}W=Hc!HqoV|qU3%Lt#i
z6QDWiVoNR7_Gwrr_Ug!vpVHK<D3QP;(kY1e0s3}1TA&uiyv<a}RSb08bhgLi=OPo$
z^)EXRe|XuXszO!YeWBgvX_x~n6b9-DtPt~Yn%D$59L%QW2Tc^ZSl*Us{Y~QH&iBGh
z{i__jTTE!ctUuj~yce+%GAF-iEShO?LzwZ6rfd?S6LlfSl#VXgD>$l1wOzTDcj6Fk
zS#}5FBT9daPg`s3C&ytB%hFMk|B@}!8AZ0*lAYfz`V}%EZ3q{IwnZKEwyrf2$G+H?
znOpS^1HMO9=Ec+DP`w7BZ7pM!mUMU|#r2UX+d=^lGXH)#_C4O24GiK#ZF%kTts<<?
zHhDJ-+Na3EQ7JH|%<NK5TdiaG_ZSU(G3uEG#GQ*^dR+AuSOQk%IZ*q$&ACly3Q(0L
z_1e)Xa7}bZW5xlxbh&@gu~v2jA(*_>$+~lshcMr$>H2DIpWY`vh)*7pX;}OINm3=_
zyl=}*p^5xiGt<$Vdm+fO4NrOIed7=F7{>U=r-kxg8bJ-8_aW<1iw-}Fd8i3>K4~LU
zzlv?hVL$Qp4|0nJ(mZBbu5zWy!P9?SoZe_RrK9|W=AoFmn;QjR7N=D)-Zae)8dZSz
zB;S)MeWAAcl@?LZWGC7iiGc_&n9VfLVXx&TL0rXBtHI52-ifRM+EY1Af>4L3j7Fy}
z7oU%YWT4>{THm<9<dIqMTG2hWZMO6FtUDc|k|hw!N0SzHMgI8!agqyKXesE~r30oR
z4t==XgZ*Da5Fx<pZz4wKt$T%861%agDL}MygA4X(=;W7)ZJig(x~;DK`YM}}7B%O*
zmK2}SitoR`f13xy@-&aS<n_(RCoj2FYdK;HK<g&db6i{E!7FSiDXb%-s5v5?gfgtw
zP)0+l`ef6sG}#i?q4rnLV3p4)b@#z5kB)Lq4AvsBeKA7UR=+zu6OmKJ`AByJaW3<Y
zlgZ$9-Ty>(aQELV90#w)86n=*-KTpj==dJ+btEX2>}VM~Ab`aKMy|+>{NJRTeHau*
z&DVIiGFKy3bCw`YNo1<OZ~@Eo(584C_Ut1C+(@aA-fce)<*f@?i!?*w1Bn;zXi#US
z6{M1C;2cc+qQ<C8xN40U=J4|m^U9apa<v6^1%D%Do~?<}iMcyoCxY@Ju@N^$&xfY{
zVeQp*?dPH1)e(9Cpe}E2xe<i>!-#WB(;`E&0R1MD05IYlnt>X93cTyYLOMq(u>g+V
zBsv*+$1&*LsVB5UBzskibrqIqFq=Vt6pQ?KUL0@L;;UszdJd;{4XAaE#7+6v!mzDv
zHh^iXf6ilI8p4*j7O<$=8R{{ZmIDS+%3Zhcwv}8Z{QnNf=qiTBc;SDrfzm_tHPc-@
zcG&<(NEtTG8;9f=1(SyQK4$-p|A`|jhG+~P4mQW|03v8p&E*x?h^k*D#{fpB?0Zfo
zdkAo<7s!I!wave`PtEJpa!yfljmsxlDaxQ9QR~`SpzJ%x7F!f3=DZ4ub6AEKsH6J6
zGQR%DBA(%ZdQth3VQ0IK>QHyj93^6HvYUR5GF=0P?$bzx@YhGkmKZ*&YKod2@%mI&
zrDKIZp<O$>eZy26Bef9r0w{I!TdLG(L5iCxGl)Fl2f=i=7$EfsP)-FW$BFVE%B@-{
z+B9N2x7iY*LVu)`g;4*Hl^#SM2~?&pbpd_E9J@*6sR&(#K;r>?Oz>-jX)nl~Rjdp3
z#H&zkX$o@4<P9qB&;VTOhQH89h2FL;pN>?KZnH$lER3LnvhV|k)I9{caqtt8dcgxd
zNq=ntSF=847!Tj{z3SeE$ZpBv^*?^Eq6R9|;`b-{pHA%6R)<P`{8UJo(E-`~<eeXL
z+WZ`8PDTueJ_W!y;?>B(biDO!gu~bkcx+|Q3{@83!ETBXuS-!US|JnM&z)Qv1Roi}
zLKtDY&*#x#C~hUHh89TG(+T&r#%F&0=J;{c&1>xazeh<Zj22WZ3U)Kh^$rd>Exqt-
ztE_0M`FCT|bETZp@)M@SPPed$&6B`O|J&7NNgS8-uzWVzQ(h)<D`j-#Tzxk|nGh&6
zU#4*Z_Ew>^$(B}o(qLzeYI!D?ob4}Vy|gxASk<l(Z<N$InpFke{AP#LVKv_@6eb5&
z&NMOsei~PBC9DCRm*YKd*j91{E5A%YjrW$GnUyIUZju}?`~})Y{}>;?<%RF{kXy-_
z?2SECnaannB+nPBu+8l?`=^XKc%S1~Y&Hq%BQl1WHH#aY!k`T){WTYP&~8Pi=Vh$<
zEe%q9j#ZgL_?(VH{>GHuNQu`!R<msp1Z?ge|JLuU1W3P9ja8(sL!Pfd827<(g*8)o
zM~JWIum#HgR>x~<zfbc_eb4=jyDjNenm3tT@7vCc);Ss_yvJ2MiXB^l@_w11vRxZ9
zLiAM7k#btU*cW0(I{~rMvM}C#1ATQK;tM`PDoftLs&v&{wEYx-s7Q%h`^A*K?Wqpz
z`beT)I<+rZuPlVc#$5NBV-<Tw_*apE=5AtRpQuy>8ESuyLiv-@hz}7t`wHpG0?#Bz
zv_p1NLk*^2V0SjnRkX^()%vg8G21oFn&Jj))Tclm%=mWeWf3f;GCmS%$V;~y90dD;
zyF!OS9tq0O{RY79C9-aVSj0&7sblSe610V|AJjVN#eZxNPuBruN=&S2&N(6IsQIOa
zM)mA;pA;v%tFjfbJ2PvPFaQFJJ=K=)fJw%$USarhcf#RoZLOb1ZUWVY2`NOMAvoq?
zvjRD*6y`M*)jEAOUt%L>VL|lM|KhgE9`hh5P9+HZtRULK8teRFEVph?HBBxFms(?y
z!QPt=Lr(Il@t`W-!JWmZ*T^xEiiwi$=k8NgYI>PF-6>+pl{r(Rc$iCdq7z28*QA3I
z3dmebw&+A6DtHUu<_IqI0ltxukCdIUrC*`U={h+GwaHir)`3s?-Ns`z+k-ZX`vA(v
z6WZ6N#6`S>hV8U=!+^ohwQk`|3@Ce#eI-jeJN3jBk$bZms!Jf<nUxP6-X3}|Z%w%e
z7Qb*#p8EisKB6xEk~qXsRf2-iAi663q@KZ;s^5mrug)UBA=!0qN~^LfAnd8IvDzpo
zx;cA{7=_Q+zEFw}(L)4)aNM;liH{ID&h*(RPh${E%U?jQM{M*n1ss9<PoURv8)6Yo
z0PVkN)d+nxwzcn%7)Hy2ixFqxR7}O+zghWubQ>yil0ovap3CiW0Y^L8pb!+1lk@g+
z(cq4{QepYj%^RXx{I1~z4{${rbL11CR@ph5qa3RTHqJQ@Ftt7ba@X<)Jm&J1F6dZh
z$hBRIz#z`J4;FkgM`=k+i$%vtF^UgD7{xl&av?R{+_=1(y%bI6wGl#nXV%AJl2~QE
zdxu-XY0)^M?dM516xQr{fu$6i6|Z>4<Ew|}6fAjmCiHN-vo)d_ubkf%Wpd&lV!d^b
zX?Kc8E|!eAY;fd2@nSSSti364rW;Azizo+V#lfkSp+!9eyddJl>3+Tt4#N*VJU#AW
zc32=MiSwMnzX$klM#${iN3gufn_w#Lu56@Zb8ggtMNzbOXd$%$c7{xq4=)F;E<T3W
zy$Vr4LpEkm<E8qCwSUQd!dN+>ulb9dN7LU$Rj(pg1l3d31gI^5KI{eLiZ(v$ozv8a
zh99rb18F3lgrz;*sJwk}bpkkwljim3vJ8edfPYGS6}sJZl)E_a;f2BW?ZS@C@?;v2
zcF>&twE^V*teKNbE$@q}TJ4ujoUTFcwsr(6mpDbSg2m30^|r`6{A8bE0d8-Mfy)EQ
za7*5sKIJ065?T0$28D;wrB5*BW2c_^xxhXqQqPVSQJ3YZgA1$vU;$Alr7RPeQe4Jw
zJrc3<*EFn$G}6aqy*+~|ex2jQbID>Rsa`A5%Y1?Mw2{GdeJ>7ft0j{^+(Mk4mG1@}
z%~u<|(viJEy}9Wci0j6r%3UmBwK$`fQu`dUb21=`f;*911>{a6`Yojyc*I>GV|u`C
zohHWxdB}OxRn05CW|G2{TQ4!Ic8QDi`hNclYe`5FN1$--@D(@G7fhL0B~g`MbzEtq
znIx6|?4dq@9jrY`T<yEGWkk+Ud`%24{P7H-<v@wm1#<4ifvSMEVTE0)Fc2;yv1G_K
z`B2v4F4j9E%srqaS^_XVJVX#^H2hJ__g;nLOoryhK!ijy6``R&*W0vKCMM2(6@k76
z!A+HSDr=yV`oLF9Bncqt<$S+HW%gL~*vjW-f>g1<c~7L;{vLT4y(%<2jLh=KGEp6&
zgYE6ID#%)OKZK);_e#n)kTtx?cRtjTsvQu}r8(Tk<mX|Hhc(z{(wlWKXw{?Jtt0E1
zYuyWEAJ!XTmClsjKd^F)62Uzrx5u6wrq+tVqU(ZHfd_!{Qnk5{nWnuzZ6xiYSS#fY
z-o3=b6Cz+0C>SS{eB48Xdh$#J$lVtMLj2_y+@)v{U`icL@QRJ<W1fOE1Oy3p2M*Ko
z%(`t4hYS(%O#mq>wD8RV5Kc0=%>J%v#{WVq)~3j+?aFjb*HM8iVn9Upn{Y#GjtnC3
z8ds}gmi%=~_=Ip1<Psfn4Kh(OV*AuUe2TUsGB*>mU(WLiyUXu{!w;6Qj_kkge5Z0V
zQI+Z1I)Mv<P8`<NtUF2sp!JBUygM5>H%%iJ^{yrhD>~L^(8t2iBkWeRG`+e<<}Lm1
zm};kLsYssK|M@+x0LpKG5mJ*_^A!Z2jor`@p*1DGxPd*kTBz#(SYs^6Vjo~F&bt$l
zh)fsvLXG@O&Az}S<?rHm!dh1WZS4e)(Or0$%4<29q#3vq=-9>F-XExuoFKl^@0f^3
z-bF5Da-7Ge^+JW8@6FSlNl?<PHaKeLcycR7-$V>)*t(HEvi@Kb!=@3RA_2LKxqN-N
zrLW=X>f$djbTTGHY5*26+SkIh&FM?xR2zE<p!hg5QvqspU(#;w5XF3A1w?7q=c$-j
z{21%{7+u43wxtRHUq(EY;_?6^o5GB|P-$=}Io=>lb}i(QozD2t(++bwewJ2UG%rxd
zoRW_}akGq_Q;aTg_od&qb=tOV+qSLKwr%sYZQFMDY1_7KW6pmv-z1ZK7gI?kdtX&u
zRerUfz1DID6<3>T9Hz!~m#d|lc95U`+Jvx?9hA>5a!RP8_C$AWJKIMjCIjsa<L{)0
z@J)kYBwe&jV*LhgCEZ?wj{TVNJy|Y!Y%V($peekNkrToGD_u8qM7;ajaUyb-yt4@T
zQ8qrRuL+p0A9(5lG>c4XQ8S{LEE|BB(BWq;nW}|A3>WKft9)iXnu4EVjI~L?{h+~L
z=1ssZ24!mA*t7hJ5QG+e@sc0Re*6Jd1gZIC=}fS-_Uzm4_vR`+v?pqtjd|OTUQ)&t
zn!4YLO6K=CH*e6Tq{fYx880M^+<IFds)JrZ@=A9EpY~Itd~-mCQbhtSiP(Ab9gMGO
z_l2jEwoiK7=`J{8C=P{y;5K7MeqG}x(7}PQ)64(LC&K&{JoWaLuu{Nyk!5QLgZoes
zC%qzMuIT$p5`VLI_*_QO1dSEcX}Q0Irqc;@{eV!|#DJ114!C?DMw;=oJkYd;mgItT
zlc#U1PL(^E&)<|@xMGP9?7H>twslkKp~y99J}f*4%Fc6=%aP*u!$~vDs256Tz^2L0
z+M<Z_-N5CV<D|uM+kr*i$TP1~QN`mhQhyKJMVz~*-2}i$Fv|6BKxw^-_?Bp{YYe2w
z!~Gilu8qJT%N;)_8b6VykR~Xx@N^!c_%*d|ctm9!Tr%YMvItB&$7vFq6-a>At*rw+
zsjrIjc<~6;p&}IYr06DN>KY5RP1Sb%^aoe*X%zCN`N&?fZ)-afh^s=yiJOa$SK($W
zuR{s^53Z{&bb0-B@`_R#<{F)Qv2d(B4o+hQe5L&Kf{q47h*H<RteW-%e1mJ_crb_d
zAO~&y4A<sJN}%;c&0W@Re!2+l5p0C)BFqZOU?HQ-BOIj<l4}1zQxTM@m*1&5WV9~t
zwQjs-t%mZtA?o8-vlHc8rY{!C6}Y;BLp)2XeZ%?9<%0aP)y8ynx#f*-2ggC0Knn(=
z>}#*L7Y&zh<EP}y=tgHAJ%#gM(l>LlUzF$kpAb6(=}<l*u}hGG>f3Tx?Dwhtf@-r^
za9w}9f;<AI8=zwF<CW5ox9x>jy%}Hin9E~dvfj?~LE-x@3Rx@FuwS!Ck_O%_TUbA(
z8d=HqDs_eSoReiL7K2FvQOp;ArNi;Hh86|&_8SI8(VKdU3XP%=Xv9C#_1V>)sF@DL
zC5xDv<ALon27N{y$@Bl&e@H2c>^Wf$wC}=tNjW2SJF1CoOPe0@ptBDTSrI=qjkgvX
z%8QB4IU|Q(p6=4#m|0TnD^2k9fJQ2QGNRLcWR75H-p`-I8?@c<dMHqMgH8Q96XDRg
zqL3NMi<ORF0>`&rYh>PAi_tyR`{x<y408%kPU^oQvxyydx#ySk>23m?l(%lSRtjxu
zYvSbWXliKlKb4)4B{bu|$X)^lg8wObc%bQJOl{4a%^3+8S(sV>(=idSF)*_Ir(-5y
z_)kf|#>vS_@V_r;dNB)YXH!Q4dNFH5XHyYVV>=U5Xg)sZ|MN8W^%z$OmMq<s7A(SW
zy#hEQPt5L@Uvpjv#;{`PhS7>*H*>^VM3#_qo#u0rkb;OJF~4|4pyqlcb9in(_-=k`
zpL{O#tgb&UFV;UUeV&_Vu;A5vh<!;R+IU59;f3!(z=Df_sJJYT1ql7)<HP^m3`h`R
zj3{T|ZyRz$hR}lUf_n^SKL%z0-j-yz1X{=_oXZ|#H=q?8KmiPhlm;O!3JC@b2ni|q
zA@4z2_)|fGxd2Wf0eF3ei1bA8sBK~Rz~_h2LzfRP<o+;|Z~!Su$=fei2#R)LL<I&&
z@L3?jY{Kg_+<pRfe<U2(5Q497q(JFq%!tQ$G?d4?J0w867gQgT-2J9E0H5ODCO_m|
zAb|oRL|<(JPzQnAXzn~{I0s;OTA{WVK-@+=4-xDSU;t;QzyN+2XmcOLA%Fz@cJu#J
z!X_{S0{jtG^@SdQaC>485CGck9r(KYMumj_xPk=^+PlUhNW(`t{%0lDAqen1Uk8@Z
zC&34RI=za5wtMt5eiNWTo`W}xAo`ZV1D8=_0tC{ae-=-`!~}Tb^?>yhTJ>3>zoCOT
zf3v0qL2qv10t<U0eT~bbM*90Ro?Mf^mbTY0v?0KLy^JG4JK2A#h6c5U<w3!opTW(k
zd=2)eg@4;P1rh)m2`LTy+TRC+lLPi)T+u$81L|`j4%s0;O@|i|!=L@V062`M0>JsP
zh?7x}S71Vd0Eu9B62!0EK7Mr(0fIgtXffbRLOBJDx!ywKNyL%f<+N|&ydD5GVf3dz
zAYXr`x5t_R+R1-#^YdHwXUj;CR2f;96=a*A^an;oE%XC;_H&>B;A=?fKnPG!kwGCK
zB!q?hG$Zgqe3U2p*{gzb3I-|q9!Gal@H>fqCkC$JM;?HCHKWnx5bHn$tn?<^(vpyY
z4u1lEW;}nc?tc>BYAJpy=YP0~UF_IfcQ{t`gMT0(+J(11zZn?jpTvxY1t4NE!B_p1
zrwII}YT$=}&ab|A%E*2NV?vOoC2sV>|2i**e+uC^75-yz(u?wtx^=ix)h=j==+lP&
zxN-%&3jq!Fg&vB@P)B_Xem5|B&kou@Im6$o5)`lp-wQ6QA%g=DA`tKuNn<lsGXk(6
zGXH@`+}RNVhlK9im57D`DC-gQA;l8BFDV-m`seWigSv(o*b0gl@h*;?q3_YdzS3X1
z;r{?hNdkjr3rgUZfgHA8;cEtTBSB8Of2+Zh9rdpC-T*@mp#zg;qz49>66shHugg^?
zV|c%=h#m4dR;8OdJ`|cEUTyPBLd~~T=7iA_$tBHiyChCkmpR~4PdDfkv}RsF6=>&R
zH040(tn#tpXj_qH;1!?!D700ptc-|^j@nhV;n|;I8epAAZbWb0`4i79XeX%j`rj{>
zst>l-)IL6Zl$EmvJyAJSYoiw*d_s}kE$9D5#(8^QOLVqokN=cm_n>(8v6IKi^)4Fy
zvY)!&Wu#Z!l?#+WT6dJ>xU-)3>)F&;txWu;>n6oTzGnRUQSW7b6!Qqtlsii*3PVq(
zfsD&r!-`TR7cKm!W_yP7ulZy66HwpHpc-)U_)XG+Yto)j`5?)S>>m3ItBH`l_siJU
zZw@;3Ifq#ygQ1uXGaJL`-{Q%MZM`5(TMEdD=5vr*ljO&6$o1#ck=|rT^z*5*#6C0H
z>^ou}$+tkkpK5!fsOMdoQTx)cu|ps<vPv$zg5K)Vwy0UDHTm+N6FK)1DCu5>-Z@VR
zL+CRi*<4$hk<{$%tA+cj)eOp1=h|9XWyag4TDWP8GtbE${f0rKWa>0J17vM-lcX~%
z1QL}x&}+Idj-PJ=EJoUn@31jasw<pS9B_PJ^Ft}9D&MI5rV}Ta6`vD*B9Ex7b%q}m
zZraE&QLA48?H9O0>RlogV^e&L1kILmLX*I{y-V?!n}aq-ge%wVnCA|;b*7q)(!39x
zOas5L;t-?E{KK(xExhj`_-Lv5`oYY)_7{urHe|on1QmBzP_!0Cx(!aM-vV_gZKC_2
zdIkmUVWst@jZWI~4wjc|D^N-zJZ$u>JB7^>ftZgnU*Ccf7`wLqS_Z!RUWsH|6v;(K
zb(_r9>D@Xb-mfZoTw$)9fjhU4-f{YAn+GNH<%<!zsgIW9<j~Hw29x>?*0Wz^1<}w+
z-4`vNvzNq%vV8e%#(Aq=xv>}S(gEkT<rs<y)D5qgYABIJMen{oY&Uqf^?qhQJ#Q%S
zJCEcPzMAzG>d`ppV_^3x92r_JEy7RQJepK}E8VYcza)OfA(Pq7_dOqudu7fOZ#der
zB~Eh7kL=FN%0A6$hgiz*Q|0AGxm715;g$Hc@eoE}){P@}b(j3+pm})ivu740OHQiR
z&#Jsp%|;6fgubZjicVyH$&@sLQ+Qw(qD<?jPI;4CV-Q_``YN-glR{h5yi5zG2?cCZ
zT=z+Sedn7~?RnK&;-oF`ZuO%_T~cQhO*hA2^IDygBThMeV-#}nlw@J<L!zyoue<=A
z5-~_e7vfV1s+8S#d2N6fqgUHA;G0R_?ak%Ubsp=HlD9N3X3cay<Vbu;>BJQcQh%a#
zR=b-2g~wZ|xQ(ad<8A3RR_uSNa4&A@A5nvj3z`A11z`^4mpb)5zCIAgb3*5%QC>Lo
z<VY<rcJV@D*#7q1SgjvApK^s5jQ{y0<FdO(3zRXHYHJwWNBk5muC)jBmeppp1D{Mf
zTqQiivVvAJY&$Wjxn_8ruKvmwS2b7jrB1kv319A7Sm7#r`5t~m4(AyhOSg40xi4)}
zj=WE$;pdU|Ul0QpRtD>mv=Oqy2b-Y#p=h!2jvFt}awM<&tXEHZiRGw_JkqJznp|Al
zf94&eb_T1ozddPhft$9`4dz(r%pmy)=2;Qiu<uw?H>PH1udFq9wBXj?!qnS7zEp|M
zhj><|k7CJ_d~s<yP87D5pSCgcWEFTA#+7P&UJa@r6iOi-KV2H|{uNf<{ER(-AU98~
zel!F0D-_?34{MGpHC%sq;xj49+sQhkWg<=H*yd?i4y-oWyy2b@0nJ!zL{@%|05hVC
zOVuC4wkRDqb}nw&D_3M*Y8VZ^s%gb_W?b$W-?89;{;^$lX9ePj`oR<^gvKT<dIp(H
zHMpFB$uSb26wQFYawr1$)G3ZU{p59VRQLhQH}np^#Zd)!J%K7Qx=F&`VJDI+?y-3J
zmtThG-rAP2eY^Y_8@;Ty<m-KQ_Q`ed4!ikve)V<~yZf(96@BL>4ep_e_O6yw98-v;
zVB41M7tOlO6t@ccYup~c!ti0hXaP3)-wrZj4`~P2yK?8zuzN4<Pf4AMJs~0457Lj=
z8Y&r36}9~K<TIz9$Ge9t#~wx9-~BJWy&Yh1ME69rvBh~kG%pc+(h}~9rRuAAMjfh{
z$WT)m?uf<RA}@F3;!;?=p7VBm$l|zB-+{6jWsX?ptZqvWv>z>~P}bwrzY-d)2WvaX
z;=liWdoVm^*mR3;Y^SAKN@C8PfBR`tnh7cm9sIE)ufUoUWA}u+jtz4+U`~<~b4CoG
z(DUQ1h9_DzQ{K9?&zJ;KqMhp_zXkOmMW7Z=JvyOj*ci0GPUAnR{$%J-)pcOPdpn-X
zq_YrTXL54Pu4%`e|74(x+$T7UA=r7{jv=;(vMb;6xjj%tLEUiLR@7N``6u+!A!BFa
z!h~Om`SbhNbKR?187|vE-3m^wi=N8?WpTK?=vhG!@s49t(HhBYdCtoESk;7;ZR=8X
z@2S3%rHM~Ytr6%}+Vxm4cPu=|+;GR7hxDN6{pWnT%<~Bh-<5Wd<DEr^$*mhlE)B0E
zjk3<PXJp2=5}uc-hvt!^w0iq8Sy#BFe%}r2<El*2NtOaEF7nSa9X47q?BFqZ6ijA=
zNGFdP#Vx=`QQH_(5v~*0K3o;g&~hlO;#cm)2N46nu@}2;K#-M~pY7W;b%R5r_`vEt
zn0g~>gos9k`c_L6PdHKDJ3@J%v9ba7vS6!#{c7Jw_2b2>esA7Vd30vdJR8!%_3Sjd
zY}|~EPaoYewoxTX%4wcM*ZTfwdZt;#_%<%>c`?cJ5ZX$tD9rhhOyKLFtK|5mN=<M5
zVT!pu=(l@@@tIQS-(n^j<4PNF80`(C0g#HzGzfkQg}EU@MTm23Funz1NW<%IMY0s}
zN#(%GPwF<$GAzbTzI!^+-d%MELnS~a+Vh;%rM$7-41B#$P{FLmjs{ei-fM4S^AJ}k
zIW)l)L=vv{^(8h79oP}9E4qqPbSm~Rs0K2%aI?EJ=W|d?y2Qk6%fZW~nIjHTLv$r%
zR$V77f%xDdIY>9Zm05(5-QkI|1W>qbeIrSaEuT<yN4ibeY*TX@)ZBsoc(6+D$s*;Z
z{A{s}5Ix*w%>ftl;j`86$@)&gi(Z@8CEon(c8?b<!FCGmxT2ZJkPPD>&wE#Xg)PMC
z<nAhI`2bjNUo<qPPYOvVR>$bWrFru6*H_j|Y?(&qC0*~f^DzFR%9|7UQkizm__ER+
zcIfTGP^KOqZ}OHzJucaxaigPCrh=nRG2E6Px@F>~bkY>0S<B|uKeJiY!8`-*uJIjv
zrA@*k7g%Z%jdF6UrCSyKR7q?(ch#wnC>FTOhn0k5kG_E#nNDHaB{|CW%zo0hsRZ=`
z%Q2<HEVNTD9dN<yOG;A6<aX(vJhIz(4*Wd(`lLE|<I+)b1{%exQ8l^YweeCpMdgf$
zT9PIukcY@L2GZ4xnb<CkcEog(#r7;ksgrm~MeH%^d2Gw|DU9`3S_{*V*Bk9eMj*Kn
z&PG(HqMNH94AaUO_Um)ULa&pQ$s8(V=;ISdCv76_C)VYWqS4Jai0fI~L`^d-o!e5E
zUql-LH}$FC))B<53pqy47_p%_DL)QailgVAX3Y7D{exeN2X`oZXoHd?huOo_&9)a+
z;}_l(uB_PS$&$nI(U0rw)IlJFaaKwT(KQ<Ib_LVkTnpefJwiSdrKi0T48Ttt^j535
zOfI;;u|v5~$4LkDRHz|LbgYa0GTqUS7fKGVzSQt+HcKsu=N!AgY<CPjAifT(_K!Kf
zA~6%*z1m_SOjYol*^;7|ejsXD@o6K%iepTa#pMN^LJ(G`)UKwdA4N|t{vg8PxP~^$
zBja0jbYN4qQ00uRX-FvC^2lci1!L;~X-vT$^NA6>$4bj*)urM%7o~CY4_x0iov#3&
z?=X<<^XNx5>!U<aY^d1Y{q2M!B<52D#p1eX-^~@9-;CyI{QkpPbqVoIIQ3XMMs|ho
z3=}Vjg~jLF^&?5xZkJe)!!Sj1^&2sH<k+j3VR!?(lD-^lTsH`UC0B(mKs4K=lug@0
zouAsSIZM-4+ijOZnK@N7k~7Sd7^~2Bp|X24D$mtjV(mnE9XE;@#Kzna0%a5t2L^ee
zp7PNKX>E0FdA6$obQVeCoY`30E*9s_gzp626gG7~ew?_JchWJO;TJ+q9=p5(WOU?-
z?%15Knd1HKpfJDa?%FOmC5%Dl1g4FW*NG8|lJt#U%S0*NX<fz`3U~jItNwnhP%F%x
z(?VZeqO)Ov3p-J_C%YG&3(S}y+<k5oYiGd{un!3$@qDI6GaG3GYcr+FbL8pE4Ro?b
z9ayxH#5U*n$RM|Tlm6pLlo{l--<8DGS>a-*K9r&sUGy+E2bLjuqU)MAaYbsI>LTT~
zede4|v(sZfVI-7dZ@Is|e5DtcDOU|A+W~^L``HJBw!X!`(~2i!nfDiqICUo24tBbc
z^VjM$legsHZ0s7xl%&atZ|bSzGx;=QcnV>J?wL-R6`P&%8t~UWFlM`SpJU??j(7P}
zPt7-+e5Cl;AwrgzU3e_SjV)cbKyLkCzrS<c9s&$MyBn8HeaHQTeAjdH6j9)t=R<M7
zR39`1m^aE|*gXUyqt{J`iu5)D*h}^z3f(L>6cp{^I|Y4Xh@Ik6hXba=zbIZ(@<VPN
zYrTte>jsHAyH@syAFIh)7vOL22|G6y(04NHNeWRsOgvDq)K)a(5);5&KOJozvym-a
zO2sCI{gZBZg*|(!iD@XZpDin0eqF#p^gs^(QL=VH!d=Vg&BW8R@o9{?Jyy|rybV>j
zhu^AiB#3{B22S%~Uol5^dqsOVh#}R4mK*_+#iR^Ae8Vz*e446TJP692iOH0Q*}n2^
zjETK;^}rbE<r}sDqYi#kywL24U)Ju<P+*Xhp@pN9^CjM=FN_7yrY7v6LXg3%Cz#yP
z$5b3U4p?boKB>WR9DrLjiOZodN9fDK)MdQJ`XY3|F*^TY{)0`a^rTZe=h>d-NI#-&
z`G8}$%4{W~s^*C_tAHxrQ)RP5M=<YdkBbAO*FA8O_S8)I@hAaa+fY!6p6Z_4;se*T
zk!Ep<o-BR8=BVb$ls|Iiu%DS2mTy)k|7}uKB^($IllYp-kx2g{(*BY4l%eDd>5StU
zHkPrX99UZt-WpIa@)US{5|_~>u;eUc?d2|Ix^V;tGJ-g?a5Rk4Dxzb$wNdX8t}}L7
z|7ICgD+tM7&T-Pl7eZV2AS1_hF$0OlfRqfjz49v2?cVS_!`~l*GU61F)4CJBck;VP
zoBdnn#PTlFq*$`U`|WmOl8r9a)!~3PFKsj$hHCw1l+!iK!BN?=k+PY|l{ZMxD02Q@
z<L~|<__LOfuEc{^cTKU91>bT>k~DIKL~<uj<Fp!9{ua(^0Ti~UqPX3v2h3MaQB-5h
zI5yUOqI6>~^UCpQ-Dp0Ygm3BlXns}TjOE_2sE4U(lV`x4nx3I~^g=u*hjmj)p>9V!
z9vv%U_V1YOsIt{wTc=|INn9r)3&>;HwB+*0^C;N&`RkA$mEPymOp14&esT-6)|Tu!
z_lD~2PeERGmIq1lUcvkf&5g1Ot?Hk&G<En+|2sHUg7kcaA#ZDsn8&xRH-T+?(3#&M
znwIu*%TT_fLd6k2BTiP>&A*b+1qI>)Td2EQ7!+$8?DwI$@^Wq|mf3C6r0@Um+rLGx
z6Ly`H79jxT3p{80d?2=*Ea7Oe;|u*I@+#_30rp*G;C=0g6sFF&32;7(y7H(>$arJp
zVr8*SD{I}K4hDxYIhe~J!1l+aPx|6U-76rX4%Z4J9slmsikTG@RY#lV2#63$W;Iyb
zF1R)RaD7CWx)plqk2VI11mf0zzv)^?GL#SIr7gbGQM1&uC7N<RU9W7GGJPkleAlPq
zC4XLTKk+bxd!ezAWO3<d2iOZxi|Pm~*If_9BE+3>lzSfTFRglbhEK}KMr=H=q&~ut
zw4MpYs?j+dk*5Ua4<!~jKVJuL9fn-bKI}G|1=$UQK3-W)1&5Asgdh|KFP9$Ii7A}$
z*#fi0bF`Kg8uoX}nEl%UsY*t}#QfG&wiZs^JCCXf9hAo=7his*%2{ita~d+W^xO!d
zGgwv+ZJgV7hJS>4B^jI_<qRn~$BjxAi{PJm0GpcJGJj-Cl%Tfhm?fnf6Uk<c#k62-
ziQ@rQT4_0?d1?S}>$ywOTesOEC$U5#hVp#&j^@-F=zlxb&WbH(!d{r^dmoDk&psId
zJ1nfwA^xq#8>k`yN5m{Zr4BL{>!(+<OsTh|G9ZTTQ65feg_HQHSm5fgS+GWqSngRb
zW11!(FgrliR%ja}N!iWiLT7E}E|-aevt9w7pjl3qA}WK;J4Eag`4VYv3q53}o*Mm7
zmde8{RWa*XTHl%E$a^sjDyQd^g+t{Y{KI`xQo)s;*qw#_wc*4&>>`va=$B5SS-q8d
zj1VCfB+<Jm-O%Po#?d#bGwg}#wRBVcZCL7DjQw%zYEy?gVUPbpI~9yDF8RqJ^>T?k
zH?EfrocZbm^LvhW>-1v{#&<x_#3s?}FDib=F=)v&Pr?q1o^h@>qWAnpHsvH}^ASV^
zPMP!7V_D_jd~jfKtariwxPzCBp`CQ4XzOgmdFw@K&<w(f<(f!jlFc}|-@q~|#^^br
zJvaz~;2z1iSF9$y37~6Lt6;2>o!eV8v9?drwg;kO+F^74&VqGN@IvW0$$;{)r;p`z
zs<PS)?1&mguRY)%J6EGUHfJ|U7sGZc<!hj<^(_77cHRSy`epl#UOJqBF{k6}aLlIZ
zUE{fV*0wz9uSL+-HPn&7W2dn_6vY!Q77j#G@Li>CLfwaj)AzPbj3W5cu#H(-8mC*<
zpBy3dwC&_TXHET23+=ih+=(X$#c&`r@#&2fv$smug~DnOpAu+bF>r2(j~JS{Bn{Gw
z@}=`#vj#auZ3MDjm9%+6xy0dNn=HRcI(j1X?nhDIzN<)jyQp3bM9G0;eQFzW&T3;M
z%Urx@y?8{Mg`WAhlzI__!#^ibs!QgGFzxcc`Fv}4E}o3Ja$9*IN-^Ikw}dO1V6`jt
zbWou#-^ZEl0MquoB;uDrMay^tcw9G(Rmk&TKxa`wg^y`7dwV1+F9g(9dXsvpmLXqO
zM0qb3Jo2@Vtqlt0Jfbyh?0@CzRA&CJAft@?tca$}=5l6;RW8CyW95~sWP7a8ec?-2
zWJD=H=?61kyle4tw?ElDqFz#@>HI9v#7AsJ8=$Br6_d2j66W0O{v5irJYNGT8PDj+
z{19z~$rE()Mr@ToiD{aAE|qmW%ADX`_na0tu@^U-A&4RtZfj-O|M_9Yuv($zj#DEy
z##aa)&KgoZ@4|`rD)~uS88SEYGVBO3(iAc#s8OKNIv4Vg|GYLB7OAJPvfj5n7@zmo
zshCva$W}z;@Dh=?aM|{*-`5eYeK6o)o<*r`6OF~0KqckJBG+e=06|VJG(@0q=z-<Z
zmO%cpbsto#vGl&;C4@1$=GywSC)#9NRV-c~a4LOf*2GW>QXFZSF|RIHBnkXUin4RX
zxboK^(eLCHD4hlzfcxU_YP_AIMBavJ!M%*?>q|MvzsNib<Rgrbyl5P0XaAls33n2y
zu(aYc+c2MCz&%)fAxk_U*1^2QxNqZ1u^He&9b3<?4B9WEVmpW6+*xp?){k*)4HO$&
zbjlN3fMj~Gn0rt|>&t$VaW{!{RYpFRKiM<Ps(OpJ{f<?2ILcm#CqDS|sax0@&3|x$
zIbzg$`7{E}WxjcH43WB^T1aNWRN-OSt{;aCVNT(4@A6>=^lRSrDkrNn#g$z8vaD&T
zx@b9mv<p6kQbI_#?g4)LjqC(N7P~oyq7(XOK7qeVQT@vX%*7Y$=&Skea<BZ#y2)E*
zoK6+4f&?<JB9vwbdJ{#gyLqDZCjB}FAGwyeIk&)_JF=Q|oMDuO6k9NRrjuM0r{7e@
z@Y|g&C-MU1if`oyzz7<~{NGp}6T^RFdCZ)g|3P+41gxAa%>U{BAC||#$^QS%^4uU)
z5-%Fq=Aop}&qTo3vV+OGU#M`4LlFs*NJ##`&wJhA7F9mN6XmG@DMpq#7fC28dLp$6
zZk=TLeEU^<sV__Qq<fTfKTLM^*l%}Pk>r`jH3gLuB545C5lBFnRaI6(fddH=APFK$
zVDybyz>ahbd@qA=$Qqi%;zY#1BLp*Hf%oq_Qz6p8mqm#Hn(^9z`-ueV+E89&Q6NE4
zzynBrz=`IR1W>~vHu1`+As2;+3@`wtOuvp|Lb<sN9X!5$K^=uX{0$_>xIcS010ro1
zSa6^P0WO3Z<Z0M+#3=jFEJ8wi3{@}uLiQ6JhV>MbgoN_)@`B;mVu%s0Ne6ono&)uA
z0>T*Jq0jznAoSYc=YpFCep<+&_P{aOyJ<V%#6=*>$bq3i4$$fxD4~73I(P(^hUNrx
z+_MtO5M^<JguX#8z96pwe((^GXrMpToA@^q0`^1nW{@b6&ccF-+xEf4vq0C-Am^25
zkVf5gp8}8r-@hOo{t`HRBO*X<q8f#Pd{W^8t0=PogzRy@&k#UF2NgA<3Ba#C*9vD%
zF*X@1i7u1>*xE9HqJgwk$%Dm)3Obxz(>|KD(LolG54_lo0mat2wOJ2-14)g70=fAC
z7F6DX`pW@7%bWvs0*vuPxJXcp1e3Us4}n}zx?~K`V*S3D{i&8Z2;V&vB+#e+Rv^a+
zZvK0|fF6u+2-t8@e*d?R@6x^8UjSI}^}&3NJlsj7SimPYmmHkaEA7E?T&NcitpLY+
zKtSHaOm9w=J#3>mkxusy+z+b}A&9IhE-h&fKkd&%SyAvK@H=!k<Ug$YS3rP4LIv0z
z;$`<|hQp2Umfz(wtd?v52&xO#Rg?ZY@4II000FA)#~KphHn;yVp5uJLfAi;e2dM-s
z=wF|JpBe9;qx+x2H`An_<g*{$KkFdsEBdb2JN`c-@D4$po<D-SWgmbjZ4g}&TTH|s
zIM$%AZC%|o-p_yyKW)}}#7!;=A{>V#G(Z$|porha9tLY0`s6nq4vfpYZ03$<;zWp3
z2(f|NAB@){)Sw68fbWUW4W=@fYpnj=wzmz?{#`Zr53Bf8CfHV$pDJNgAklSw+uXV|
z4#Yq);Vz)Iy#o@)JbH}Ql4pim{Y!9YD57|cEEvR<M*dl#2yh*EEgcMsLwAqVy;vYK
zz$Y=sBM?^1#CH^G+oaz_?=uegL!clKerdeInZn8YY;2*H=x_OeB}L4Ft4U%f!+^P8
zR!Y{n=WSC;4T^f#P0z^9aoq;2Rws14q%Vb^Z!eua)Nq$E+XWE7mzEd*e!(J?(IM2y
zl)Vu&ZDkc(_Ia74)r}IV#9#KdOk?uT$r4-LE}8d>9zQC3PJ|=L(RD-CajWa#yQi<$
zBXw{+(Igk>b7AXkT!V%yjl5v}P`w?cAvI&0qsM0_dWvZmJYXfBtMavQFtD<UuT6cB
zkc*Cxy56&iaipuMct;C}q?C-gv_<Y3^EQpCGly5?vegVHdT!2u<=+*%w__K^r&E?>
zuNSYm<yy+&>v!q-MyeLeG`WgpoDJ`^n!aRwLAeDd{B_0&hnvV@YVI6+&XFsl)I`!+
zULH+mR`+Kh)-f~C?B0dj-&_(+XQ<VCUfCV^;vVcFN9=wo*|I%FljHbO%@%eDi^_5b
zS?!+vxvl4AR)vTb$_F)StksyBcxSV;L=H0`cpBG(h04wz0Fr0jAo}=o4+UG(a-=mD
zFBupjII>kUMuS!Qu49FXHg&U=V&-bPRSj{a8}kABF)Inl)Be~Ei@oEegR++w`(Ed-
z8bv7aygzrp)t$|8f(*PCt1Hrvwnu$hq&`Kwx}duUO3A$hA8SAU;JOn5;v(}_6KQiB
zX+N#=@BmAv!jFg7E*1-SbmDgb1$>U)?Qx&z{D<qzk08w=TmQn(%|eN$lx@14gL<;P
zMIW!ISrQhd4R0YT?VOd>A5kf?Kf%n++74>L7BFE$tx3p-Tq)?Tn^A{uWfmFbeY@8c
z39&_}DBfP3u9tJdiJ2E{YiPhaMN8p6Lbkx>kKf%$(H0Uy62Es16OM5kwJ-HC@6<CM
z9QEjgT-oAZmKfe8dGgCivo@XwxNUOkJ!dM&SPq|6KVNzirw>V=wxkPBU)MDtTzm9!
z_ca6sH^EvitS6Q{f$tixoFl1Qy+$7f9`r`p76V}pv);S6ac*K?{t)}KEkw+fkyeFA
zo#ykyo$DBQFBpzvL2_m{06j8xM!u)nUe*x9x}kS!*z6lXv%+<~^7)=gTKu-}M;k-R
zc@3z-DYh?s1&bcd3<gaV^1aaLJiNT&8?-4xyx(I`#3tQ8lGP&0SX=m>L%wC(S!MaR
zH$6`OTlQ%{n8KuH+~RtHB2bRLoiOYL3yvJ^Qf;^bf<CJe4x$hw_*HI9UG0=TU@cw6
zspc0o-*Mwu*?V_j7EerONm!%c^Avj~Nb!|uRAh`aC4Z#k9eu9*>{h5H`R%lv<ilH6
zyPOf8w(x<|6qtMAkbzNuqdffnwjFs7bnF6;1@=<OK6GMOw~t#I33mJ0>@v;6mga}q
zCuo>-<wZYF=~NDO7^-T@VS|93#NIg^dieCMaLL+%%mKSwe~(>|dHN}0h~nMN{Zfg~
zRdjBgGY4Xr-@BH^5q2?i&tKHFEjSFOnM55tDs3Ll&*dfpa(of)?s;>aS3|_Nsaa+(
zBS~n){uB_a4RFY-c+uwC#)d9~ycfQ&b*VjkEsOx&Og4IGi(VIU`w?91j#K;n9oa^<
zlyPNjIze+24#S*dpCB<Z^aYB`tP;DedX#y&ziU3M{Uo)#)^qB%Omby)4afBj*p7y%
zEaRGIKai|WgWrv{$lYZ+EV3*S&81RnZeVHsIlQ9u{bZoFes+_diusk2IVdiubVO51
z_(h8=C8N;Xyr$Tz8@*Btu9Z^Q9reFK0GDH5-cJh{oMhSq6G9vBmajvdB)11PD71{?
zDPt5cMKpDM4Dyo#uRY-?!g??dd6T?PjQY28V2dh=^(g{b8{pMm7ZlLgzwt3Jm|jHe
zsC8DMp~Gl!!>{#5Z5w;U%y=JebIKhWj(zg;zv`!F9B&4@DdgA5$hNADkbJjGw#^Y3
z{rbs^6HM7HVQeFWxv<J94PMw9zL24Y@L@iz!u)&I(w67r_@YSDF_EayZ{RUvfv?&q
zNX4IXJAD%$WOCSdlHQ6O=xbVXXD5P=IjjA4BQpTv)73cK9hsZ5d~{Ucp{^Qdp+J#v
zCke~jJm~kRyu%=K6;MWV2UXo6p=&T^qT4vyBX^KN<StJpo2I$NVL_H~HR*L>RE3wz
z`qUEmPZS4|18i-JG;PKJQU1YL<hY4Lam@W$_c?uL?sc2zib0985!^?~spFk}dUX^P
z9uDv+T(ZCIkfiEPOW{}gKCF5Un~&EZnhM!{fz!MTAM%$y;Tb2|(f)YR919cWavh5h
z%R4ML7t4~D)_k5U@g@n1Uh$2vi!t`RjLQa?BGI{CCyxU%?;rHOMgu5NW=@5F42doa
zebp}euqvyPS%2rY<Z*Ue*07YQJh_n>gp}9|5=wfX1u#xgXWA`#D{+o<fynYfS{xv^
z^|&FvIO@}wkUoGfdcCiJS(!t!-a6(?t`@f!Eq->XpTs<RZd~eo(GFI&ZJ&CO1r)m>
zBkYHId-a@o)~PWQZ!j|Ae-JdnkW|eEZkxFd5jN7rlO1#ik=<m4fE3d7(7|<&f+CoW
z#EmSeP+rSdWInY=bkKlZ??juLOzg~`1DW5}<-x)K%G7`xf{(<<VqQIag&B*9Y@hWu
z<Va$@rJ3jN6{de?zGrJxS@-%TLdN#Y?!1?j1rcU{OsZUoc1GClZ{5+&Opz|kyY_R!
zX!mq^3WL1yI@$Qm>p_+I>w*!78`+opja_@O06kQ!K>MHbQi+RFkqsX3CER>73ung0
zg4Z>iqn|;Z8?&>1DKE`(gMkFReOdug{Gq%b)%NXKDBNq2J5e99#JBYYV-ynqA07@t
zzThNpHz|h(yZ()2Zg<TSsSWJQMj_Wgn?#WeJN@Oc4$RyR%IJsrTr#W$NwVp(mc7)B
zGj3>^Qu*%i;;N-pfoett0haJJ+do?GNm<=Rzp!!*U{qb$NbjdW#IFLyg`zL!Wn89*
z|EXr1zPy1)+)|l)K4jUAC8e^GFWCoVOe=l(B4ymL<t@ORt27tg*s^<(>v<n2c*F^3
z3CYx5WT$ctaUgw5y2<@@XT~?nsw>c+$Gl#MDVtE}feDOBV!_zkpQqRO8YlFkhN;a*
z7B&m;jhpwzI&+)3EkFvDiWvs<s96IdeLSkWiV%^bFOdteZs!#}wu;j*w?1#H<ZGj!
zstG9-$d9dRgnL&q>Y&{`Zy^ind_Ae$!1=IW4yRTRG$xyK;&2@ZbLBwi9*UYltvf5w
z;=mxOX0&=Pm8-&3K;L-e^<@Rrkri#}Vx5wxcnMef(Ie$PX_hKPdJ(cR{AN(grnC6@
zpnVD88wOUvff8QkKN~ib+b&K;VbI3g*?Xv$a-kGDoq4P#Zsgr4b=SFlO8&7TH0DwU
zJLQ7cmGzJ7k1OOYmZDFbPzI(}NmQ7?inePGmA>1T^Bcv(q2or%NV8wIk$m_s{3pNG
z-OeR<w8o$f2=Jt9PDN;)aZGgVOjXqA3n69ZytEatcRF{;;TFl>$mEu)+w3_w4Xirn
z&uB<RV*NmFe1(N}|F%nFlHWvdG#-U*{Vr}UPC6q=&!1ZAtjD1MKB?%t{)U^uv&O*h
zGQ!i-EqF3v>ODFkxowxU*tvC^eD8kdIbrxA@;{QU3cV*Ad!A*~PC*nfwLpEf4xdDa
zztHW;5LKXjC<|?OdYDSzkrv#6HMa@AK%(g3Na)P+ea~mIAlqqOp{b_9ziG`W<nY(w
zFC#cgiovOnPgr8v-@J84vz<&0M^rUJhgIL>tid6WqfgB=8`o3{NWjju5d+J4>jXWG
zru6>Sdb?k*>#(kUQZ7Zv<e_8bw6SM`>WnBN28(`*A*O$V=`XTZ8m9&S8oa&yVBE>F
zN?AIMHsjvbJ2>&T4r9QaUd%DTQNoV;=Ly*XCZyXp%GJ{xR)j=s1rr=Vufk4z*<$>f
zT)E}u)iy9QY0=6O<@8R@3^g}HYnGu8-rMo&YzZm6?ny%5TzF-dci7vNUR5*Z;$F(*
znca4Mr;>>|2SnxVxDwJvVdAV%AD<ohQJ8!Uec)v&#YEK+v?g+qS5fcwZsAhnWVh`(
z(Qv7%@ey7=49@s&p9O&++&di<svBLP2sMM+d%&Gs$arWtLu?b#q*t2NIpo|FQ&!0(
z>sURZj{pUNwm_%GyisyQOD!?HIVqw=by_44&0Garf2dF4(_#5-wD03!PJ~OB$2Hu9
zl@jRv+xo?#M%uc#r1;{*X>U=ky;Q%cdsEa?3SEEAU_!Ifk?q&cdrW%wNi-;G+-OZX
z8RcSADYPw&HhlrMR55g$ChH-U;l+ZR_%Rk)u@yW&Y?mL(`OWW?PQu$lNjFJ{MF``l
z`Um;RxG(qRAPd#JM#+9>$UU<|`btsP5T$eS${z2Gy?5H*+ha;w%h%&T!LkJ;ogqP9
zzGk!QBEc=gyqyk?eqJ8x>8HyJFmW|$%RVPgmR%fglYxnAhN8<6DXfhB(9tTTOMu3v
zP<xlv2TzZcnqf~0+^KQS?XCI2?o5XH{d8F%8jkNE#Kc*Fglrjit?|qBiSpTs8{oTP
z4JszHrg%1%Js1A|`&1<;7rAqaMe~u$>Tk3yd?jQH^6<t~9EHL@+TZn=@(B+cf);bR
zC8z_oCF~GYW%06`De*n;w{vm_f9XQwoBKQ))s_|CycCs;XR|)<-j}`bf1Pn)S0=$b
z1FjKr#iV4D5TD6CX0AKh?hm-nWAALzM{3iGDdLh%NI9P+ePOK`?GJBT!pX04VY%`t
zU+hBT>j$ZdSHdWPN2cCEsV_m`mu3Vj(xnSc+({pxcax9jIq2POn$D2_*@qBUv&t)1
zmX#>v;z~c(xO-~T%4i->8wGD|^<+<+!1z)u+46D5KX2`@+akl2K;O{0rz_N`H}Og~
zOKhGsG_P9qo<0dJNUFLL6!A1HQmsM5vB+IN_bz|SGQU1r#opo=KV4ZBBL!=oWcr|)
zG-LX^jm<)sZQGEA3erV^Mzpafl;l$!m(~V#3<!U5P2R|%W$@h`Y~TKKs&;;Oa>v^#
z3_GI5c>4oaJfJ*Qfo&*7*7|3-LuMN0@X#i*!7(M39l2Z;$IgsVk2m8(YsNW5K(W`q
zlMClzJU=$k>luZ$f~n9!IxUJq8GtHV(~YZuZQ9yJ?x_46a9-0dW(iLDk`fv19OPtk
zqcAZxAU?HZYzT{MrT8#VZd7R8<~eS~TNq$&-g|pE^Za6t)qmT5#C?-=P@S9%oocjf
zb!Kuh8)4jX#_il*bFI2Lpys5j>B(@+bM>U60CIu;cJF|~@pTK)?xJ%H54PoExzPzE
zhJM+w?u33{F=Dko$;qh$=iRCJ*IXLjkO@BOrCvx+nc|OJSM|RMC!T6Y7$WQ4{svk>
zKpX<<Sf!E-$veb*d_)(;q-x56RYfsG4c(C|G$csR7Lvtak%ow`V^ZBq<}KFyY~0kh
z^~`cJJD+(bR@kO5ReuvoW^Slb^7$jq_ym5TddPfyDaxbD^Qh&uok+43+^Vl%vgNe#
zX)|wb*B^=53sEvQGL9=1hs-8j(IAJ$U1D=Y{B@Muf*8HZ$8$&bi+`b3N!)}!KY59*
z&pHR~8BLSjQGu<7{2-p6t~o6<@v=$_*O)N*ozvs1RxPMrS8Uwik=XL;I@FxHKF<Pc
z@qoz_l;qBF&w}-oDJO2L`-)aS45DF#1zQ`s_RSpXRcSQ_DxSLqqLcR3g*p-wvXX}r
zdE6I<Kk9$>pm5}NTByc2sozqbFTEU(O5fm~T0s`+mBMee8e%w?SzQoPx{PldNKZvW
zoHf5JwL5IlN7o~npOE*ENeye~Vx`((>-(dYBTO%27aU>MYXB)Q7({QYMW(e!h!rzD
z-lxQZE+ddJ`_Q!~;f`zi9@TX+>nbUjAaiih-s!$U*Ut*PlQ%3*Gd0J^+Gua8RdT^Y
z8n_h36!SMzL%&P;uxA24HdLvaS!gsne1UxHtJHlgyzmt#ADa<O*{pmT&?!~SyfIl1
z-V-YKBEN}Y<7mcGyQ(SQOuMa(^m$kbJyT}8ZzXhcSq3yMJT-;*WcVH*Dxl{YIl4&D
zqF#CJk{9Ft4WWLUvb82&84;kL`R4E}X0${lsq)%G4k66PdC(D*C7VTekTR8(^)(+?
z-+oeml|~p48tpTDUON}f3Ie-EM{Gv%*tq4|1?RlTJ+usCXA*;?^Mcnwi1F9a4!ueZ
zMo)0p!+JSQN_L2O>xOqnVV+M{bhhrbO#(I&(AzI_{)u-<_T-<~oxhu3)_dLr8p&QP
zw;^x)?)=))<!J5I&rM3{GdYKOdK_c%l2I!PFZY2fF+VD`zD^N6niOn@#s-Qv&T-Yf
z|Amz%PhPvXH`Bz%$;>@;xH9hEa>f@^QKnXDsCtJiEy3av3e$M{7IB*lnwa9Gqu#2Q
zh{Ii!cLlKbc`#nu_>L)^527EY`vNyB;8G3Dc8eJcod??S0R=5CzP6)Ja*@mIxsgd%
zF1}s6A)r(IU8WUrvl(wmxOa>J#oY0rLb_!jx^q2#HS$ZfT9<};zL>gv`nQ;!-6W?y
zu7}cq;=nzkyIOreKk;s8oc@-7KuM6n7jHG;YI9%}9c|G`u6G1*=XttC;->cs#4;Y0
z3l$KZkr`xs62>pm7&SP%Tmb)&0;XCDqbwEH;dSoE5C52uj>S7~f*KFPIsx6LmB(JA
z(*Fz7a{$jgO_xufT2D<Xo?rJw`*6^#Va)8aVV`n$QxN#Bhod*9ms{>PgrDBZgr)}1
zEoEN0u;vZIx(S;hbp4i;9^#FU87d8jnCWtRh4|Z?-P!poi;3*<^AqC$&roOW4232(
zXM1x_axa0Y6MbvrPco>EtGCj8kug0wHflhvY=V>Qp2}hN5j!z2WrX<OM?cPAwn?<{
z8KW}~RWexX4zbyC#Kzjd%qvF)e`{%9;6ZOq63)m$a`^BVUQoPfovqx*W=Jaov4=+q
zhZt&~nv}D<To^-a41PrUU?kk|JvN|Ne|Pn-bvAm6_SMB}*Ce?;23vRSPvq5U6e*9q
zyLRT5<gEsww}87vA(RGsSo`rzGcaJ7Z|{CfI#Og14aYAV=;ATqW66|uDg;HFkfJJ|
zG$;Gnl$B^Z*wmBXer|;yIDV2<Y#4e1;S1Wwu23!`lr+`SSV+HDY;}qn=Y!nZx{zC`
zoZB9sD!xWBI!bsSl&mci8L-@mJb60x{@!SLMfxc(xr2!fZ+lR@$%;eTQK~U7D%`cP
zLN8G$^`-__uSpLPx8!wFvY{`3lxE|wMG#BeR{2xW7nMH@b8y&Co*%?Ob8_lgoCOEJ
z*)TW8@yE==LVnPT2$d0R9j&F~;O|$lE{dq0k9t8+s%ZSIa5O7Evo(1m{*FPhF(Ul>
zG4DUm+>Ga(T2VoGG75$mZ>B$WzBQ8?QxcMC-(~!>F5#A)ev?f8bMZG^Vd4l(y2UJ3
z>x|w3kDJ`5?v3`i19%=!jgff3-P4`CjtW=G8duIJ_jjX2;gwZ!)CC_#Hut5S1}>Ac
zNLDMG8}s4V-G#;4AT{MBdPE<?9`ZGK_UEf_3`iT8Z!#?71jWC0FnPs6gSY*?-nUQ}
zecZ5|bkm`fk_Vr#wjX`F_y&SRW>(yyoV5CgTsid#5^M8f#1PB-n6j&p^_j8qevL5&
zyhk1>(n#zOddzOB#B<m9({lxEn$=uZuNGnJQ4x`YrAQyyx=0lVOFMOBABJ>ekri>5
z+cwP1{_=4P)n$YRdpmiWva>km^3I^wIewEHD16SrL27TJT+U`_>KJE=i^X4o&ZU%m
zPuMk$!M}T&-lqwVCfY-Le#@7lA;F_~A5Xq)6MhSwxS|eD$jzyDi>UPEoj5IO?*45s
zdS&;NKxBBumKPo@{iDdR`~j{A?1f(DtS&#v_UFq#>@U;d26ZIV`c}?^QyAwsz3pIS
zvOye2g9`8kl|Y5D#BeN$%hO4)>VY?m9{j`757798DgA#XZA|}-w6U=LUvb<&)f_Ve
z`+p;C>`WXi|KFs|8G>a~d!<K@*i)84B!<f~YP<7v%YcFDcN%6k0VH7osYp^86fq=W
zN10&3i`*mVTtTay(zVZy*UwLLkKL43FXzeH%(mH0W3CU!8f*+$Qr2FqfeQyLDF6i}
z1*q(F3k(`0I7AdIBt%q#gHx7(g5niHX21emjDXNSZPE`(1<?QzI;cEQ0V4+w2YmK9
z6xhE2kQWTApd_Y&K|n-9^M#HS8xK?#=uI#U;1%>|!vPmdu&?Oj{X#-iXGb|s>rDaR
zk?R2zw6qd_Qn>}jK%)IQ0t^7Wh~sdV0i1JCpkNq-*#rhjKm0<sqwV<$IZQ%AJv}{v
z1PDb0ID1qx69~5<d7A*YehG#Fh)dX4O$Gt{L!fWnbOI>yz9G24&lv1MU4<S3Pz-;d
zJVbH^*oeE^P|hGEz~mmlm{!+6fE?pFv+5(W58>5`9gs(Gdt>j{<p&<v=gS$Cm9>kP
zLy*P~dJJtJAkYs0anTh_!jHrU1akh61m;O_4C5by1PTboVhq$y{-cAW!~jU;-Q4?l
zImmEwXv*+l0lyTAM`qaLlqG{{DtiP32G|gixAa^pP$1U<ozC9bo9XSKejmdCKQ=nV
zw8n`UqO5M-5Sf3HLc6+@b9`uUqBn$NXfdFnkdP44Vo@Lidcf+;nCz*UJF=hrEj{rq
zd2~4e$1L(Slzl%65PWExKcgQ>fJ8h3hQE+|P!Pz^#_bF)6e6U3C`UdZ)<7LH`jCH)
zkzw`d1#iI{bO^zKHhK;Zk>K;;<z*a=nBI&7_WTF`z4jYg^UT}=yZrX6=G88<foT%~
zk%E8>77-B%9UR~nDf6ol^4knU81YaZ>*uHv%qak*$j>aQy`-1v`HdKm!#~&$^G_@1
z!y(pe0GRO;u_dOVy9|E&r<GIsAJV4$Iz#c3Ir@_w-@%UZJ~8<)@%@t~h=*AAgAXj$
zv`sMyIPGUN27ItD;e2|TKlM!`+Qs>4H!BcK5>WK(<ooFpgO<bJfpB9BDahIOF5P3Z
ze{q^BK*Cx>iVOMjQ3HTvtiDYLqf-`mCl>;SB<@q8qSLed@XEoR0=a*17?B*H0FVKI
zw?u=C=(PwC;D9tuXYh}&UyK40h~VOeI)JqMKp;AM(6O0VDG(r`SjhMVUWVf*C&4e4
z?_trOVL!K9dJr*jp`0{wkY|Jrj9Y|vdM2_+@<r@7m{;*_xot>T>?di>w1V&TA5Zc=
z1O7VXu|teaNw$G*qJq8G>UlW*faD(Tf1Z}dGag{T<VZ2wr7KXmcPXa-QulDEcrR>=
z(=?S1I2E#tx*`7!v`C^Wq}eL?VYpi5ySUqw<e3-+Uh4V%gwJ##nD1nPkpC{_!PF8o
z)_Juxui`gxHg<Ov*I*wj;eWJPy>?#+$iML5XftZjCe9M_iX%V`rSr8wdGQXUAEa@P
z_sRMHG9c@aZ9Wj^bF}A>d20BHROS8?N8-*b*F-RDDR?+GX5M#+jf`Z%Obk(o3c`F1
zq&h>Oo_FsV$@b-LVAwx70@bV44!oRd@>$Ldj1qW3bY>dgSzmYF;+UH=Mz-tt%q#%<
z5)-#LF?J-n#PSU33bNkd`JrK{(tgUz@Zgykb#FbGuNf8Pbv{LTmhYzSkHjS18Z(=b
z($^?eyLK&)4;G78vW)Ad?x$r5#pYK%n{U6cgzm0AHO2M&w-#A8Cn|o-`Ckh*2&*}T
zTlqJhM6MAC(Qz=oRaHJyV7g-%7;VYDR>POgPr4l3rVo;~B1`dJ3&UVXoO5YYY+qEm
ze7~btLKidg+&Nj5DZ&FndASa)iNdUQcU$5$X!|7z4lQ%D{9rkS=FwbLb_+#sqt?Vi
z0U7B+8`G{P?&0Hr-j4|9oL#rZUkm?>v2%zK1=`kh*|u%lK4sgsZQHhO+pbf#ZQE9#
z`;zXw!5wt=Xb)Gilbx(@t?&N;x#uBnJ)Tw%*m<t3r#rqIb3Ola`aC>zU}(a<QD@(j
zl!LTuetbi&Crv`^h{w7O>$4usBEBdeIqc!m%(5YR<!&`8ydE}P#Fe|Xbu@FXX!ls<
zr03h5N=5EAP^=p^qt;QLFR=-^9LP4(RuJhN2Uqz=gbE8b)}y}GM1G2Ie@Z=Pq{n1;
zO~<CyYrNQdXzhsORka^Tb@NUB7HRWzfHYRXJYw3a;RVaT`t6WK+nfk}^G)$A>ieCW
zC(ejY$I$Xb+9!po%r><w&KqRb&h>eZa+k;&1+_Hcc5?~j2;ISK>KyHEUOR@=tb-oR
zm@2n%WOa*DWJ7ss9%oDhZMVUq_3uf#no2m^w<u4&C*A$kaeSU=EMj%j+ej3~h>1An
z>DnSq>%Ab74kFEm<7G8^GJh(Rko!7bm#`)efum+wH0hvbBKC3Vd1biVdXJP1MP=#`
z8X<v_?6^#aJ!ihNPFA-xf?s2gkVqH26n4nSfhBY@iFiB(c?YT~+Z@|mHzQXmrzf!o
zziRo}qRX>`nx3o6ec63)*T_7MpoyA+J#j~GORuuxf?(L^?J<*v|DcGVu!WlAvmS3N
zb(I_$(@E#H>q#T=iBxYWl8Q=?Pgd>R^jUsAeMY08nB{I#jNNy&h2F%hyetgIM7JOK
zvR`!stB5Qab%LrEa`fCim@g`?mqBG<2)u`&T15m^1qKcr-;<d>W>@#^-$4uJ+ZJ3)
zCABsxQ-YI!K5|>XVwJQ{*H4{w&g)T?GIg?jkCOvl-fQHZwA4X6&jTMI(i6<w*mayX
zKzjuGo=&~`D5f3GL7z+uCgTcXj-sKvnyj{vJl(@2Spo{pVMz_*s)XQ>b-&D0&uF;U
zAvQ$deS@YYMM&B|aByeqtBdTcK;uu)E7LI8#Ctn8YlpIAM4~yl4s?jJt7Wk!fC^$8
zywMh4J9nR+cETH&KquWv3ynphB({C!eczGQMJd(pcak4}^!B@jymDuhxX5)wXTQbt
ztX;MB3MJc7B2#mp+I~5L5Yw&S*vRVG?2WlEH_MlbPhVv)ablmT(bA*d8fv9H8qT@k
zQ|Pw+?Y8v5G9h_y6nl%*e-53aWH>5@nZWVv2OC_N?3XYn)^#5Y>a$i}C%4cMwKzI$
z@Z>H13i6L%GPb;U#nunIxBz>kQSZ3sg(I026w><s&Ig@TZ~`=OX&8V6(MfN)&!$z0
z>l?k<XrzkvI}|*zhjQy|NlfzaqQh#Q+J2O!7zIp%Df?f}2DEqb$rj-?zr|OoO>661
zd1meMY@k+6i{Ym-9HqHWaX3QB@s0NRT-+H3)$@{j@M(-4=hGr@j9HYEB@z!Amj1P>
zuBL9L3M>7jlZ{awSDeV=p9r2v1@pBfJMK_nHwb`NA{RF#j~w=lMI}UacLT~5USF^k
znr*VS>kT?Acg8G423aL;DZ|5-Wi8A8*9cw{Kk2J(^Cd;p%+|48bq5hLX+WaH0$_Ix
zW@=b+)5z;Ze>Uq#X+09IEo0})pEqz}w2HaBKdGe4|84Mu0Axf(@ll-Gsif6|8p&c%
zie*>p#q2g6$?|76+G2$@IGSCD%kt{+NsFuvt<7)H<*h|V-!@ODY~Lx5V#l~}i++t9
z*y)@dq{&R|b@~lcaK+4s*k2BibAN$^d_(q2JUvEH;vo7<``%2po@L8Uc+dch&k`ST
z24U^YlB!>KCPYs-QOs!v@qh`V&vIST;`iB@c2cCm7=O&>*})uJNuwKAw>Z!IRA9YF
z(;3E5{Z~BMc6&<Nm8h(M-SdqLQ>7vxUvLE|xKB{B(mk5gyXo4X_;Qk)%?%A@eHmTW
zWhqX^5ob1+oGl}SH{d})M@oyS3@1$Tz;xYmWxj~AD9@%>JKO(nwEvTx*9~xF33|^I
zl%?(~3P57K>ualH!i?O4!74D4Ryc&z!^(rni0y?bD8mL{!ImJgxDk}&{661DrA&h-
zR^kGVW@o%qt#JwDDz=`!Jfz<hh0TGxOHd^*C+<NC?9h&P5V@ozOrAF$S=bb;pi{?6
zWh|}Q2z3YfXZ=(enCf6hrc=m}GeBB&Gkq|@-}rGtJHt_$QdLx>ikTMx(&@nYw8Ax#
zpFppbb>Eh#CW`z-H%P9>%reEk{1)}x`{3}RTO~enfJ%zrF23}Lf$LIBxnUX$a4#(E
z$?s!B4L%9EEN?x2QL0H1Sfnluhn9t%jAvub$}IY8GfO3w%Glm3Z`fl4g-)=VQoJuX
z=k$aJw5`{Z&t8^s-%fh^t|5G+IUUY137|Uqtl5{LYb4)TZvoDgVQ~{U<=4w!zQ$ZG
zc>A6w@#v}t=12bW%tD+g<F+1D`#i$uma<u0XbrN*p});T5wSvo!HrMEaGL5dC6W<P
zcg?sVuBzQG$`m22_Hz7|B&w&ht}pf0x{qxM#^&iCl<<J(e~rGs)791*q4l7$>35rR
zZ4Ag$<g;gYZNn2j->6R46~SFVSh3L8za%wlC5dyTMWzWCuB>V|g^b*|$!ioG^E?5!
z>Kl9<;+c4Yfpg}@_$a70?2RdS>3S?s!;)hMPh8ch3<x)cr8Dxz&iHB9u7u$5o%w1^
zRY>e1`%(=?t@*IWNr_IRR6t+}bdE6)8Fx!-?{Z?k+ARpK-Ba2IZSO8fpUZR~TQ@J1
zTbqRHaK7f_VkRPry7D$Kc&&O0e6e|f84EUJwAvQIakw^5&pCrn`B_d_Jj(RSh%xm<
z|GKf;STABJ#QW};c+3g*n4E*c*;3@Sd(;rTm=7{d+&zIy<Z-oDXx^Os93SQw<lLPo
zI8Kjlp{hKrsftGW9BC#&Q_OT9S0zWAUJB`syVjOITDS43pR_oR)TM3_FXpejlG$`^
zp@qjMyt5Fd^Il=rw%WqsQX<Ml5=jhezMKZ2laQ)>Oa*XVs5~xmyS^^qo@yHG-YnX1
zo~0hIBgq6k$T-zoaRbd3Rd}Pdxty9^s5%WN`T^uQ1<MojB-i#5=gl|!ZtVtXOdsyZ
zR`n@4KS6rb1Iv{;D8ooU;S-Y@Dk&Jmz!tj^A+f+Wc=mNnH&RK%6YJPNMLUaN@wuHK
zU^H{mTw+U{E~Gqu`HQLifh*}nQ}*KTQMCM4uJ7VK=FOARGxPT=kmI;S|AadBUyxIv
z$Nxc4xWlv{AnJc_-L+@HTQ&`VoiIu=#F>62kjRKNw13VqCMF@|O4U{%Z&iL`Z|&*B
zen%p&IRb%F!~6Uo%SdM>T?~_XGw$IM{XP$}YI-c&jaoVpT$zX0j)ndHE5u6NWD(Sh
zn3<<GA!}71Juk1W*c3(ei;%a)ugCjSXfed6Z7pLDBYiHl>4>z-pDOA#5hZT9O09RL
zMi^q>H`?uX{H|~$X7$$(m()(DlVUlmI^e1tGRdH4lF)`)a0u1sVqb3py23E}{`}`+
z$J1OFqiWMvZ9VqAmI>m+VCXi-=cV1GR63-6=aQ-Weff5gWs@;@N$x^wtjabRB^R>e
zRE*A9`lKFomXqhZw_w7=m$@0bhsV@dVNx)k^Vc(`8@wb%Z$SWOft8BG!ox|;0Z3FU
zTD+kCw*fK>(u$PGiitx{i1Pdp`Ydrk(NGEX?)voX>Wa8b$0D)v#%$gvIxWUY--Eke
zrMw7-wT&3i)^C?-NHbOa57DnP?#I9w^F5)GgY6iD-Q&<}2Du8Tmykna@JRI5>XjwD
zajL?<u*{ptUGuNV-H@t#2rKWz>#Cw0wueCv#ijSBTu8TTS+o%ARPDf3*tNfnZWvIN
zRY20M=HTJ4OeKl#q_J&Yu&BY5G^21J04mY%nRX|bK$|yB{#_2yBAyZ=Q7uYOrF<eh
zK@V4OY7PhuO95Rms=+)w&%-Vx4b$}fjDp=Zi^xv-7}Lp#pj8|DOpLc2i+k|6*H$+C
zP>|-*fI-%IN;G*kzC&CF*aIs>SPj~gIKJcMLmfBD`Kqc-<BqAmVjhiRlP{v>jDU9#
z_EvuFu?yFpaV_(Ibl96q`X_=uN5ZB2;yDCn=M;;~odr@wu;_0}X?5obdPvc9tcwfX
z%@?k|2;h;&9YhOxEKIfBUD_&3y_D&OkrL4aKl|;-LX1`FReUY=tE*|seswL+QJQs9
zsY|X>w>Fld+PwBabkhb*p8cJ}E@mu4wn%BoX+(@<YEe3<ze0;|-SzP?Qc+ygOFx3L
z*rZI4&F5o%<Rg%>`uv6xr*0VVWm1u~a(Cf;CxAzNdnOyi^n$emMNX9#*!lV{v<+7&
zBnxOX4qfN}z#+xfYo=%Eg5=?=s;sKFstvCvgvLfYb-ubbTCmxSVwvkAy1eokwD5c2
zQ+MI1%(tN@wwLQ1COP#GO}U<+L5`9TW_o984a*Qp^T^N*rtb%K_+CD4rs@nIA%#jG
zpy#uv6acYUND|62!uJX7pRA>L@_5~YKvZ9>ww2<Jfa|9Q+$_Qs3d-Z)pD2cL+0vT7
zh>QtwZb}||%nZ~UT$WAumvw+xn#{J{K6wPLT#FBpZLz`J>~k5b`pM}%@bLAh5*=Uj
z4+X4AWo-MM$Q&SD9W9TL?;QB$T^p{rvOGSGI6R)XDE%N^ktws?QVEMX(RYm#H2P$g
zE#X-syrgJsSD!Cf`TNWM#Jqk8b@_)%Kc#KI_}^8JIm=LE&$utr=0%v&pfv{glX8cK
z%uYD`WEXspMX}zB+%hC<TGLRlxX+oB<1RwNM2X@s{z)BSMKwbj)czQvzggZw-4f?-
zzuSkqenKcdxM9_`RH?bm2>!@xViG4?Jl1XE!BFy0TiCyEnYi)E2yJWc5#`!<SSGn!
z#^jVdyN5VG7Sf(QfEy27D{|C_q}lk9R^eMq^0!+Cy{5Vz8SvIz3$UqbuW>tIFkn}%
zDsZBp)|HOG-Rw|^YpV{q+fb!KaZ&hZXO)1w_eI{|>e^~u<!JBW6gG*-f19ZGw5LGE
zXFr}2a&H50LyNBtF6?e4^q1W$+t;ZEE_m#^+(P>-sS(rl(nG|oV-&w6VHA~b%{oH~
zpy5dM(dkOs#SYe@KX@D7xx0XweH>jpkm#_qFGE=MiCM99t%QPeG_p7mhmqGQX1lhN
zxvOc5KD6p~HzHkZ>si^!+f4xcs#gPr%>ZHFJ^Sl=$EW(wtTpJ>G%m@_hTNn3eo(ZX
z{5R%SNaH8S3@$$hr&3MFm~q9#hY<ztj|{J+hDS}WH0@g!K`QpDBakYgS7$ziXQlqD
zoZV_C>g<D|Y9P1J*O*${^zI>YX-$7&yj(8ykgXnkw=S4UJr&eioX<-Hm1?i4jAhia
zl7L%$74<EQo7Six-k!Y1*!o3$;-}BcA9kb~`P0XapQ4=U_o7XSW+Qabyk3+G?oQzc
z8FrRw5#xdNh(j{HEF^Xin42zNr)yAJeu=wU<yk!GB&(uUj_14@!9L7>&Yo@m#bZx*
z($ar)Z8MWVwfw!1o!!U#HbiIAKaCDUH{pKE+3NSFr6{nt7*OI}W+(3b(;~Azm?pb(
z5Q$Xcr7oZn{&oT~BA@Pwi_0&YKutF!cROjN@h#yfQ7xG<yvu#r&tXK2_kBXh=FtU8
zByF#a-g+9?=BllwEQYj>T~4Dh8F5>Cbk{YH%;*FwQ(1GhcSNll>${*eH{3GvVn#61
zcLpjfST2MyDT4I4NNW+Mq_aZJ(5J4ij<VS3*$LkkuO%nAe33iL;w<+5v1XpoV%J>z
zgnkWj8O1}Tb>9Wwu#5MGz|YCiaQK0YK&S<x#9ihILi(BL`?z!G-@D@er*Fv`YSjJH
z{I%?E`?~y1zZXKwta3e{ztqVE{jfK%N)?_?_Yp^?BdK#)Ob3I`G5eU9iv>1$Fj^1K
zeue|Ldm+U#UfWyi>ulLH;_P6|o)|qj6)Z)SCAx+4bWJC)T>1@|QTiJGH$`D|xZmkU
zns}a0OruYWRZd;xRU^jBc<v@PS!*$ymEu}hi11yG6HINFIUUi|$)G0Mx7E%H66?qV
zy-lxGc|BB!S<ZfVUj}8xiWB&v5#{ScJ7jAqz7QkJ(WeNlhIcw)2Oqe48j6fbD-!u)
z^U3N%tvGlDjkam<*wN1YavN=G6k-r-Sle=0rE2@243vB)WAAPQ^AH}YUU<+ui2hBY
zsOHIP$sAm{-Y*$9K{Ih=dl*~|v^f^%CY%jsJ-$0nz(z+`(-+Y#yiBdnF7IX)2|d;-
z6BT9BCP8@_^Pd4$+?Gp}ut3CeC=;!L&56Q~ZqdfnFLaTX|3mVlHPxPB+dc*M-|s;?
z-&%gI+&{W)du#FxgT9qJ{xt?X^gbxpZ3#KAcmzKq1I;As<00jyU17wx=ovFc!_l=%
zRZg2#5aUYxbU{KpW-c%|4q}O#B<^g_N^+RCCKVVBR%ov}Uq%g3)dOHGe-yOW?YccL
zX>^eelp&A)R4z7jsY*J3{nmhMq?qu+9UWn`a+_U6o;cgWY;C$T-;lhuRphjhZ3w~3
zPY6%+7H#PW4(BVaruj#EWG<BrK2?|L#r(7R;j$jg0Qpu$^o5@}OG<o<z{5A!7Dd5=
zmzmsb!;sJk`F3EbIS_f5@KlKNKGpX-y0M}CY3T4S)osn;jOL}a;U(Iw_L58w+cfPO
zCu)j29HP{N%GL=0?-Qw;Q}?ta#v{8M#X_uTHk87m35Ft`+I>tUPCP}$aYeCXp8y3H
zO`2#Au0Iu`{vBqU__iZ_d=Am5a~r#qxSxa1N>{(z5MxuTGI(VV&aTWVkF&&&onfh{
zm6nMy--2RYNN=Rkq{DiLR=!|I$xscyWQ$<GYi~ZmOE|Y^WK-C!v9*d2PoYg#K$Yvs
zeI_H~H@W$AeZi#QL{}&LG3wACsHvbP+$6Z#+M#*I3kdS3epDer>jD*E)7eHB5q6b^
z=d?(g7+NJSs;fcBW}}6r##Tq+!;pLMbU?>8;;TAT4bnHa{?mcYJXyriY6DvGMrFUZ
zveqKAaOI-1BbB(xx@cs(m3#yl8gIp8<pXipgCY=nFP2W3H)C)`i7@RoTJQ<v#eo81
z27LBSKS<+748>QqQY@!DLUwqxT508e)BPzjDh9}s9IYzF7jC7eABrXtJjm|dqjCK)
z+s&I}ULy*BFU}$PR6H;2ZQw(_zVF<pep$AW@G8vkJP7{K@5(|OyVXz1W;wF+TZFDP
znT?p-*o_8Mhqq_eVJ?$i1cw#bhERRlwYmyBw@HBg_945PPiKTc!Mfi2lE#!_K2(YJ
zh$Z7bX#tv4%`URVz6d^<R9kq>64&d{2jv#GIaAW3S3xe-xr_D^Ts?xf@9`+h9eEAX
zaHQ1?Znitp^Be?rrXsdAt<uU98;_Z$$D7t{UGQ$iNAO{XNuet?x`|S?g#6sjs#)b)
zjrh<@B@S=g51M5AWv=gn-Nf^n+F5({FjHzY{Od2LN>e|ISoqaYSe<L>jJm?9Wdt1V
zyCfIxZ}^=5ghaTkW*V~$KQdJ=;>DTW_@NYN+dLaygdzOX(8^P*i=<A}VzOU*Bg`VD
z-uvD07F0WJ*4)%n!w2fC?wQUl{8Dnyd6io$Pz;6#JPf?EBS@f{DB)wukIj*m_xq=i
zI-dwfosyl)_T(mkR4Fi`ZOTkX_n_6r`c9v31YMC&aNA;W(SENN+aKsf)^(npgrUZ<
z4i*8YHq>lx+z)sk9F-!6DiQZ|Li)a1>r-n9K6k%?vS~cURPrQo$*y=y7+#*y&GtBR
z=`-z$C9{PZhGpZwq`*~6?OC&0q<`5zf-Gc5pp{w{@ELVl2MN$u<r(F4mg1e~mvJ-x
z+Sz2?|AOD@avlCx*vax=VJ8R6|HYgCU?(f<e}|o{Y@Gk!u+tq}S?Riw4Oh4lLXwrB
zhwN#`>sF*V7=Zu*Nm}T(M_kMkIhhVfv6xuLGszT?QW6~mL^vM%`)h`C_x-o!Mn{8s
z{g!!liu-Yf*^7K%pOpns-cguKU^xLI8bBO@0#<&Wxd8|eP#`~?0D(e!V!!}usK3DP
zuo&k=9tA5@So|0K4=5<8QGGfUMiA)*AtHc{90vdi5I}@f5J_=>fIkBQ#WOo%L?T#u
zV2^(6|KLs}K#?TJaspm&1qL&87}vZ$dcZ9NJOFV?$-v*Xf4CDjGElG|!XFRg5Xfcp
z1_E#hkW3r`Ci4DU48m3e9cJW@ruf+ArWV3gEShj(_OZts;0{AHbU#L9XkmALtiK*9
z@YBG}9DfO9q9cH`j{aF644F82G&=?;s6HSE43J2E16?kJ7zLF7Hm-hQ2{gl<KhY1-
z<&Vf7_!lS6zoMV&{)0O~|E(Kx2N5C^cY$N<L;v9b&epf5=oipYLpTiD2lq#e?F-lE
zz>0VV_X;quuLn-O<91;X&>4pS7*TyI<6{8<eeBN%oo9Hf!wT^Y8Nh488dLy1zkw6c
zpDXTrPzou?N4V<vg!;Cq#RfW$x%>4x2KmFm^-VQ2zxZbi2GDB{a#``89aO0JcO<(2
z!yiLlQd=7a#Gem9z=wcNsE>`|`Pu)6?I*N>739N<fEPe^jVV8Xft=zcXbB(oB@l>#
zLe4z@*AMHbg9tPLAoibcbq~NLlu-N^7Z>_Z&X1<iXI$VrKy`rndjLScUY(!Lx<2Yj
zxNyJE?|<Z=1zxFS7aOJCev&_F3L2Q_z<1~fsX!3o5x{^12;w;Z+@N6p+@Lsr7_Q{-
ze)p?F?1g}ee<abpWc|vnp2dK-{J?_2?`G9M#H$?(0B-!Czxm{0AO?B^er6s2;|5jw
zA2;aSf83yQ9A9lYpY(&jM&MmTd%b=b>c^gj^-B3+e`kOW`YO=o`b5>>FCZSBf9<N!
zz@l@&NQ0SG=Ya?m1oO}IIAChv9s{_r1{o02`jj8hZC|W*_835x5F$dqJBRdp{QG|!
z4Z>g!LVKd}8y3E70u9a1@;)mG3B*VBhRA5h-~fu`_4O33>l&yT{rMrdQ)1=wb`8NG
z{w)+@R6_%384>hDMi6~3t2-F@Bj7~8FSEg@uBdO?VL%`_h7t12{{q#sxXE#Vodj&Z
z+DYRO909%x|0Y2I8B|czAKs?^OlqMJ1cw^PVMTw$d>4WI1K=S!r@NX%E3dB~uhYO#
z&jCW>=%?`r84#=Jh}UK<lF>Urc_ZPeYFtK1H6KQQ8L}VazIB`<oLEZnDK@!Pck#wq
zw9CUYGokg5z8%?aocg%wohkn`Ap{kz^p$!O$g@0smv3TD9E2O&=f;mk>lE$CF&dUO
zw~^JKNH(Q+g(sgOuDk!dpPP_-9_B8=_2bHW3JqrtgiFX%oox+xnGa9&24c2$%B8pi
zo8%l@RGP3kM7$!Tx>s)D$^CA-O>;4Ny$yw}dH{cC5QPnM^{^f+E3iDVe>ZDDiz<Sj
zTPQs=>lHe1qzua{Q!ON|pGj>MairZ{lcxGkp};qnz(l&Uaa~pg>#{>&**sK>MNW^S
zKGEPZt=F7p`l1u{ZD(U$4(er%*G?%MhC4B3x9AZA(O;3~8v!CLyY+T0WC^CG<Y}HC
z^w>m_!$F}peDipGPVo_vgOXS$(SfJoe7y;4Y$41XFdi7+zX-A63JmoN8kL9C&0cgK
za5|>x-EAiMs;3<1u>8F@eY||x&piVs`9wF}IG;0~jLpvJy-rTkG$gkxpY~Y(%jm-O
zI7Y{vTO3%Np?^3C1T@qt3!C8;sR*59EpbAWEK<7kSlWrrq4BX=T}+%JWLi#PzSXPD
z!0-1d?fulCt=^>FX$-+|wtWOxTRCrwq_Vx?Qj~ie+&T_d4oql(q?I!`<3;yk@}a%)
zFSJ;FQTAcDv~k(_Ea2~!V!?x4lRkImL;h!oEF?ctpA8QC4Z0@P^5|C3DBndUoNkjB
z)5Mm89YF%+<?@lO9k|*@Xq@TcJLCXh%us#k6M>d6WOpT{M77HC5P(f}#xsC=+jn~g
z$TIWAlJi#%jIj?DaVOq|LDzbD-|c<TLqH?d9vJDWP;v5@W{WOFSu#JMMzlvg)jRt*
z_R-$>2*W;SFhOjT67o5hw%{qP&V4NF=g<kq#fKAYTu_p3DLhM!i?q;2kZ-)wS0aPU
zIKFNu(Jeoe23C0Ai;9rzvJd3OdVPa?e>qm^st}41X{WX;%CfYMbqWi&rnN=6YrYp{
zw<-CSV^3!ZP({X#98~8qvy7jUc~(``b`oDctc0(tqium-*gA*U^}NBkuB+4n%+5Xh
zjF^?i9ruNr^J|bx#eKAnz^uj+G)`}-svO9J$k4dTj5R<B>8)qG1W|f-wrRq?w=8no
z<CE*l%@TO-AZj{6w-lqbUZqV?OrB3`g>#%{nI|xH)LaD%My9Osawp!e;mL4Z%;s)%
zh2MsODUESsmuR0UtS6P@Zq47j*wpzcZFaBhQ~dgu>nsHV^`j(DP8|N8NAZQQJxt>J
zMGNW5(hx-tx#vo0-b<~=lJUGFs+uA8q3ML@jN5KaXjdgv&kOwR1dn7ja`p$^Tz*6S
zn(E3&KCR`~(x+@Kr4g@S-4UvDXi^Xl(Mr1FCluyL*Duf@XJU_JTwC@tB%JI$Y)MVC
z&{p@9=k=W^PL`5b%|DsGwp;N^3HEvT<1`L`3cARKOhYxFWq-I1QH{()#T)63R9UY@
zyo9=Rw<VBQL}>ZZ)H7W;VIoo78d-;m7Hr4pFda!P`7u4vuTq#M`TKYJPYhK{%j~uL
zis}<j2Ds4)_eB`}q;7HuA#mGE1v5(Dl=)SrUIa*ZSDuu5ij}NT{0COcS-84~{mBah
zzA)ttPaavoMUGk^U>nO$D@J_f%wuD-Tfd6PZ7NiBcJL=_(Lc@1>Zr=%tY<-U`>@n$
z1faPcDsbNEYF0L2`SLt!&4CRPq^tfgi1X}0mHkFnAv+OXm=d^I-I}EwWMdMveDP-g
z1`#S=?=RHM+Pd2{dQg&x<LznnnUHSq^7?W1)^j@HucMiA&;BGb|18cIz_yU^@LGJ6
zc7`w$6>3p<xGblVhb=$-OAAi_g0MSYPa7WpK<%7^LvAI?D4u?TmeayXQdcBx7_xRS
z;k<Ag)(v-CFSR0BUYBq|zP3(KDnn&q6B0H88VLAdR094J8N!U|4WHqtkJ6roig4Es
zZg{q>IW#q@@hj>09m$aGgAVufI#Ck@1Ry*475SsoS^q3IJGEOI2OsJdK<#K9%Z_UL
zK7RKChsz*x*Gz*>b)QTfgi(jr#YA*iGw)epjO<U_yN_3uFnZo{V?45mmi%Vn)g~?R
z-4i`s8{kq0ibl!Kq<10QDLSpJ2agH1T=32+kV*ZGM`5S<(a$k>vL2=l&JT+U$`)G>
za2AGSw=&%s&3GT2fMhfz^D;}XRb0ola|h!>Qn{_OM&oegx$;>NrI23|an-^Qi(tg>
ze`sZ>|73`6^fmaYWqd=N!1_vu@e%1lNkXoROHU>s8jD39&yIedJA8a$nc8i!?BVCV
zq}-qM9P=g!xckK&o?_?utF9$sKu1Rzle1&~Rnp^c$jkU|{A`|-c-3J?c>72Kbfnn(
zjIyL?FpNt|6A0_ZzUpqQSBc#xO0-JN>jJtlYa-woQ_J@G)-xh4=CyryMs^K~-d-N?
zKADcrZJT(*{$P}{Z4dBfe=K9Q3$^e`G*8o_ia65y(E?~8amRDJqhS_3`Kvj;<TgQ*
znB;ysO5{tqCm_e8-~!pS6L(68<R^rTh1_ypzQkZs>UR<>E7YAp<DBjt{a_F%OH@8#
zq2@^KKJE>O&9Y-vYs##TZqy=~KS(P0oP$ww>q$!pb>n*cdrtrHv=Rk3qHzg0+0kwL
zMy4)mqwE#gQ*XEVjJ;;li=HT-R24PNL_pd9;S);UX10hjc0YIA<1t}jPx6U5G5uqe
z`U!ek@-j}dlbhjgrqf}1lE?~YQ6B%PU6;|R*nq}v$cUnusA8$28F_T`+UCh7m01QM
z4n)EODVo-x(TzVEcdYx4xw)WnRU2@{<xsSGcRut;z)wIeqWdtVs?I`I$lR`|<8$wH
z3iSirq-cg}=!sH{@)NpF)5*T%3u2QmW2Br06#Cg9P*drnsSsi^eyZCmL-8KCJVK);
zOhkMYtBE%U84G{4P<6oSOW4TF8!msN16^<2v&C`0vh!2)^-e@cE(-<}VeIf_2##)D
zqAJ#esJx#}{4AQNSDuI^S54;1w$$ipwQTG!j30K2q{mC%+r{cvwg$P8dme$7gDHg0
z;SR*vFT^%BHD;A4o+R1xr8_pKpZML_A?W&od@$t9@bFRG>(x@idFCh>ZeMRFhZA`p
zBzRoXrv26ysxftooNZOTm+ceM2+CCb&HY6&e}YMRIz~<;`3bc-W^V8-3-3$z>7`kz
z;bI|g_5!k4ysSZk)<5-8_GQ>!r6H0i$lR@J$%xZQ3OBr=9@(`&HBEFY#}|>QtyQ17
zM!olh{1$QG`~&EV3m+m$dI$EVU5@K3rBnNbP7Aw11s)#jMYqcF?nP>2_lp7;bL=4m
z<lpDArNuW{#r<y9=&sa-T`u+`@J**;kI_`;jGVg>By+D4Mb3eWS|;zQpJto>Gb^4X
zyA;J7FDn=Kwm-Bko_9ML><;g=JKR_*hQ5<du=mHt2cG{Xa6XAIcCL=xIu9pNHn=~-
zv|20cAEYi9fGwJ>e?Cq3@mA{C)20e%fgK%~ZjgM?nd8<RGMS<_ef2T0+&vj_3EWWg
zg1-OxV@(nC*QA;wW)eO$M-nmGe<)+pk~3!4|4;_^rS@sj^|wf>M{Al;NE5wo(r8@x
zipm(B03oZ?@1@3jauOfPPeRml02^huEntO_kBuC`k~$gW-xe~!c);l7IlXdx72?KN
z%U9QEO+)lFa<8do!l6-Wu@-J$D4H_z!x{3&D^lB*NV%pE`<NF2ly6MEzLHfp6rmt<
z(klEO@GPp(XVzu+bmVqbDeJlzl^hO4MFb&Ea8K5ymI`A>9I6-Qk;(Z2^0GU3exGbK
z5&v)|{RrJ3wzvwFj+U31tWe^2_FVBNwy>{!i?_RMp<X#v4*05!VJI~65@M(?G5PvD
z%}+jP-I8-Hk2c~9T3$LT4j$R6xp6n_4E79hFDOuolQi$kETLP`tWVUmx4$ur#^Ee%
z6|7awRvhPQ-0eqnlU`cAI-ji?3_Eb(h76-_Hg`mxoM`kOVYg{DQ3f~mnt6QW9_@+!
z-6TE6{SGVv{&dIq4o&REXupH5G-W5H%`(r`@-#~xkRcn+tE*;0LyUnjfD{&9muSj9
z?rx2k3Zm3TM-ai8Sl&d>rVj%$P8<EqGtG5vlxDnI{+DE}TYcmOQcw<Okw^>maO}7%
zGHIG+6C&wRzu|I_g4xpd-8!REnO8Mm-zSt#Di^S-$M-U^T)Gc-ApTcTw`)w!K*5Dx
zSoW6&sci^Qa3*{2cP}*vwT)-9kmUAcXr_R%%~3Cb`N(n$bH+<xOriQj6W$}S^3`%O
zV|o#x5pKkI19x=v$Fib%h_na3=MVb1T4M$1y2Z2_n>Gf62f459m96a+TOl`VGYB|B
z*-7CeRwWxMP!!DYP(l$t(K~qtp4*-%YusWqpV=r9M|JWdln2f}8y>GIlZ@9~1s|z|
z{;QK}1Xl}m!yluk(6`Dp9rF3SMLAK|)+&FBrGMcDSY(V`f~)mOuSeAA^xMb4lL$)M
ze+flX9oTlR<->&wgw}F$%k+&0&)Z8(>3vLE-lksZUNvKQN40(>MJt(BT!~Y82nonN
z*UOIAEdF#v-mcSJ?DuXw^s<^>)HDUaDFx0Wzc{_1JF;W)g-)MRT60rH3b>bB3sYe-
zQM8(6m^yH{KLfF5cw#oTNkSPb<K$HvBf=H%aGhBRvg^=8FIQlecN!8Tn43y}<bti9
zS5ACKtHR?2TIM1b=6SffIhIFxT$pxf%{WP6)m@&qB(klD#q(p_9avXl@~LS2kQBx(
z7JI?#+}c>4o#zRK+1lll^D@)8kCDgg=_nY#Ff_@2XRT)tYGuv1E2HSCd^W3)D+h`K
zst@P*3IRD88&A#K#5??}@}hUZri3&e2foek=Q=9-o924NtV9Ay<~;N?POX|H16RNz
zrphKW=x2IoKWL;?Rf+Qn<*PVueR*=%*u0VsFO!+V9S7-C)}(cdMf*TL%?kPw6GuLl
zj7x)ua$!z`!MOr{!LHtxY!0h+GZs>TkdvY0Q7hB6v(-VymF!UAON64fhvKE88Z2}1
z#IjHOz26q!+&d<mDf-Rk?)w4ZLdC~`z0B|qaY0E-h=f1gvTJ|N%B1U(L;W--#4mSe
zDlvS}TTv)vr(oL60@^bpYf0BFk+`A}f>zB0f;joMST~BcsaOJuMSDI`fx?*82MZXX
z`v*y;Ey~Z(wCRV3N5yu6>E`r#Qn?{i&4I8Sfj7FlPFn`$R<k?^%GWHO)&smWPXSUz
z4dm34>ylokk7KaM&%0Be<8Lu`WE--sary{lz<~Tgj=SaC_!&o3FK?!#euBsO_)x0~
zz0=PQ32)J8mC@1OL6)R^k5HPcdzO49S5|w5KCQfO5X5czZBOjHG)R1|jeDu3q%HXq
zwWpN@{T6tIO%s#eIM;;_q3xX@!<~U3gsB;cdh|#Bnzu`FVeT!a%lM?(avfbd>A}VH
zoB`+TDHWJEJ0I+831Sx7NL9Fwd49IqbofX4U(xlXLgqtGV9O+|t^@JF1%7koT6%Fb
zW7{P3WrTANAI%fjo)AiCJuR?!zZ?ot@9oJJ%(ip(LF|@xT#NenL4-qDLX2OMhL63>
z2A9*Cahl@Z>dWm}68?@swvP?LWfo|ZI%A+)c|75rRp^<pDF-Qffusu*S)!s9^Ba?>
z#<Ra1V#vx*c6`?rYx%_r(X-4xFAxlqpnL|07;}*I><f@z>R+DMb4sjK^t7o&cj3%O
zjdF^TcaF8yHbVgeQkaA2)hokg0&FCzSVNF$T&>oO$$7#Y_<D<qyYZ@L13a4IWK4|A
z7-2B-?NC+iT4~<ZO!4vXE=?S6)ReVn{n!oF`%r%OAsxwdafz*BE7e!S-9OwCI>kko
z+)q3q5Gu_)8ZSDcp>>!_7<7})Vw#kEtP!2S-!4Rn)}!tczrI;R?ysrd1?C@@z<L!>
z8z;yXQaG61XZV`?B8VC$*<am<F;DrNI}@hUs+Y`DY4U0UJGp00ti{Qiyw#?Y*RHZk
zv6FH5w|XpJiS9H-)>g^J@(E-QJyB}!o#H<2<b|+#-^YzB0K+dT1v!9-Pi5cO3)~u4
zb*a}wK%m9uF~J{_eCNx5{y?Fib%KYrmRVstXq3C^tlZliNaDhE*6P@{*r=`BKu^l%
z#F)U`A<ADHrh8IZ)sR127@oC6`Kn@2@Ey-d%|mHz5RE3Y(YYgrh)wU9nIk%?XI=0v
zF{_o~f<@}YJa$<4AhpIPx=lb-yF=z@@?2VRp$!PLtLU_#b*PG_ynxlGuzK_TB=U%9
za&Q>Sa2H{pDu$<-y9Bl7ngLRh?TdDOF<;Y(F*KN7_~87Q@f`w=o?$i5>PDCbRCmHX
zZ+Jcd?XTrZtv{!F9VE38J8ZqDm1;|a(A;c`7xv??_D(-HzBsGbdVlw{{gi!qGIT3)
zrG8HdSg+gj4tA~DGO5C1j8-<0AaQdJntobbbUg;;DfV-bw*cB=!n1Yc4avrP3}~~%
zKJ8RbjLDv9Zm|a4>bYIn!6YSn6q~-Y)2CTQ>?)4Wbtn&@Kh#dU^+{+y`>q|qemWK5
z(<o7Ojt+}aPCN5y?5y&+rZS0DveO?UIVOTPh*VdVjK_(9vW%}G%fBRfeD85|{o4#(
z@`Ps+Gj75M8;&ws9Ah!Emr<`JioA>A?stEmlJw%jDw*@xd2H!=A6!bHt$Bz5-W6>V
zOr8en9yujWYk%F(Y(IuZt)x@IMFTecxG^u&wlx#6!!A>Nyh^~kyxz;tagYCarA5<{
zBt@(<TLn{tb?hm%O3w^-=ou6Lp5f16qtA&0K<D3bZud%zvdXzU<xb$5lm!6g^*Uvv
zPCYlTZnm1;#;-WbqY*o^BHVRBXrlAWRczX?dh7VD9JW5rK6=%I>d9mZLimja*3dHn
zkAUK!E|XKd1S~KVdCK~BMgklq3UtHGz1)0t@Ep10Ycc`NbM)}Vgv&Qc@gsi#UW>P{
zYWt43zIqgE^Du1!{nEoCX<5RW%7IqGOnYUK|LES?BJ4mmT9R|Ts=4CHXx#9%a9mf7
zpJda6oOR*gxKFaEm|S?s>8WFrTjOoSy$qmoTO{_X5HDT>KK6cAC*g(O;l_z#<ow<m
z9l+@vZK6Cf;BjwleydEpZnZBWw@AfeQMrn)vui-palMU?3OF*=ky>x7j>Cpy6s5<A
zg}0R1V~5iDlmi%lg7N*U`^2FvMOY<>6Uf5l`}%B$HY{STU7fQ~VT1fwCpaBaR&o4f
zOK9P43l(hn2W*<3FD2KMK?^0<G8G?ac&bm<sOXBR(cA~e*8bgwN+fx;|5mJup!j_Z
z^`U$s^3m}9dvFj$X5prFEB0@Um}*-Sw(N{9zV#z@1Yu}P>hj@2$O8(U=wYMUm`%U=
z0GQ8C40>tJvp;3UQB?yOD5p}YMQEIL*3Be9s@Y6?oI*;NZY=(j4};I;7Ts5+)O!cJ
z*VTj<TaWmtcOTPywek9+4Xq{x0VIxzP-QMNs{`6zcdg>Y>P3rNQ7t#O=$q%~(Sjf3
ztsVO1g%m{&qB}Ea*+AX%^lyfXtgESGQ2fu%Yt7Oz<VXcninfQH0$07|b?Ryxwx?R1
zzLOp#9cK|5D`;iEDW#Z=;{9kZ!srG^gQAJ$&}oA^*_Dr2B3{m`SEo~P9dhazW5qJ+
zjdbPM^FZQQj$b>Z=9UHv-H-6-FEQc9L&rAuzHK7}@z;c%EB_5LS=RFlO*Al1l3t$Q
zSaq~=G_t#8+a|9zF=5IYLx`Ob47xSC)IHm>r+w!5IX1oO{)>fs#gPL1D$I?j+$oB$
zQIZHMEKed~9dOHV+b-$2l1J-O8+WNy_t;|gtF&cO1IlxI^JOjnO$+J#kjD45#n*W(
z?W@88<S&`k=nBVBbTyZ1khaeYP(a;wpv2TZp4?8kAeMZ91AV)^$k4H+mof=DvwbJe
zPxroxNY0OZftBcECUBGiElHZ1$-7ZUdR~`c@J*?vG%h7{yU6p;1Q&%5;W~p9`M3SZ
z!Aje9FrP9xd(W+64Sqd^jse!u+juUB3fx-&u|TS=lkceQAL0+?e6|eeDB|1s$-Mjl
zwZBuMxCNYrF-r5JpL?k}9%xg6jGKbd$n;NTU)?tg*n~k|m$A?-|5~77{Qxd^8df70
zD+{MUzOF%*62un5yTjdA*o?G8>Lm%`=cj1Q3$-DuI)}~rnmci_vpFbnE=E-m<i>Dj
zvL?4>3=?Gy{5O)hu&f0-mv6C~Kb0~Bw?A|Gj`~7}PHGi5r4FC=7V2WYCmz!L!;@h;
zc?-|JTawV1RE!33u&Kg%=U1};9ysxq(Mt0SJtnMjk@%2Yf`9=01wJ9(beQuVDil)z
zjHS^v!TVpPAL)q*B7%=+k7H32Z4=#=JAkl@nM1W^@R51qHixzIxSH6HV_gnZE%zH9
zb)mXGeQCP5TQVMnK4qJf#k~AJA!jl>#jMQnu;)}S`O_nhB7=f2ol-ygNKV#!0&E?j
zBVGX#Lx=YgV&e&>bs8lp^u?HC!J?1a4K={HzB0;kTzIUw@v${3q?*U%`q-aR7F>sk
zx4?a1;$4T&pR<jTZ8drvRV*1puV`6nPVRpiyY1XYdvhaO{oRU0QN@}Qn|VJK(A%kB
zBmn6v_q;xC6@&ibrwtx@ssM?0)gn47e()nAO?G$kQGM%BoE>WE(ETVfVvoouG+sKX
z&@y-Y0vGXhS^ZZM%=X_%FeBUl63jBOvHZW={}#+Lv#>M%|B_%AC1@VyPc%O9hwuep
z5wkKBk0}^NyyFZ7-Er#M=nP&X46l*gYsPYCqCXGek@6_0If=p$5+X#=W;h<A9KlFj
zgWE-3wy&<Y-_y73%>OoF|L3F*%3^b?5Hk%ylQ0knqn3p(gIb3Z!cY?()OdmrE?AHN
zM2SqR!y<x$B4dycv4A;;hx>N|0G0x(3bXkS0dr*2!PQmO!j>B~Tu{u-Q~8S&7h>R8
z_8S6bliiAegaMDSAOX1tC}WWX?knI0P{07R{dVQx^nY!lLk!X+L&|@VcOm&ZCSfAv
zz{U$kLW~3t7!R>jvI9|2B_{j-VTqprum&m@M%9H~Tefp)$t$YHgoptFiu^_U{tZ&#
zh7PdM1%wa>10rC801jt|NY~&7FeLyP6z0M@999}bl7VWN>)wGFTpHF73=CW<m-oj4
zE;SUVK*cR4z@SR14Fw^VwDd=WCIF%#6E7^F?AJ#cJBp|TVh~)58_~4M8UZS2Ks;b1
zhVchWI0H0BC=t6xQD*grWR~y4zYtVfXHXBvg)C)6^aliH3xfa%5j`!4fDkYgplOCh
z>(9>@X9_~u124Sy`$QmuH3Wpfh`w|07I226A<G#P^_W+dPH)u|0*C@r@T;%NuT+d+
z$bf59eSA<BbV{)3W>fDL_}~X6WXf=Uv7|9VeWv%t8U#8|{>_B{H{IAq_E8}DkczNp
zLwM%{{R=JeR5ZWNU(a(kn|_-hihzpxM@38q`Ey&^2+@3iU_CT6z8erKA$Fz1(rqs$
zT{cKSd3IS9Km`ZwGii1wOY|YcJ?1y(hU3jA!CKG~AP3@C83ZE1uNFtezQuORLCK35
zJMdxlwKb9{L>z`mj?+bk)*oU>xD~=v^h^pFVz|JFJZPYM`5uhH5&>B74}d9vBN8q=
z&0rT1jx26Jg9T5~nBT%?KC;-=x->%nPr2E1%;&x6WPrthKQ$UJZ!TC>9{%t4ub9-x
zNVGnIGHM*C^76gA0C*ML9xIFN4|yUAz|h&7S{$y*$?fY!XP`cVIMVrre*6>BKd*yg
z0=VL;6&73vTL|vlT)d(hzNRNvm6LF7@B>eP673vupYr9A^O_S;^P6*<VbE&lHe@dF
zz1uP%v#Y{#0IP&t?ox-$ZO58Et8bYJPeHloUl2^C+PEWT*{D7{Jav`FwK&FI=jUQE
zJ&%+0rk-l2kiA3QBs8}!JMgZ8b4y|JH`S(I>7nBH5xGM&*ml^IQ+n-9XshwLTpj1k
ziQ9E4`gV4(za(~x2P1m)?!NX^4DhEvVPofG=n+f%7#0>okV3#en!5apehtsk*GK39
zJ(Bm(e0*X(q^MVwdb15*N+$UD);Y0|xCfOTQI6WS_`^;~!~A5i5?QTc!QTn56#4xe
zV~&rH&;YCd%veRdGW7Li!>i?|$=P)J%n)GmElwUA+r0DI*U!%-_8na`I@#lP3YMl1
z`6}6%ljxJ~4n6S~FAxLLbb2oj<GB1D{xW&IzZ7z1M=RPz1R;1Sgk*&~Cd)V2c>_lm
zKa{*HXY0vHXk^(TPM;54irT(EFIrdFrM8&aeX{C{d*^l*lWm^fzv*c&c&QwkOT}>j
zu2%dRzEj;F@)xClQzaJ)`)K(l6Z)ve6%7onvLAJ|4LVk994a>A3UQpeVQMd*e7M15
zP<w+gO^}wViyCg9NqZ#W>SRR!>b~1HztMK(7JIFDIaVm;&z*JC;qPZ?+TGow3<vk^
z)WLW+Ro{!-wTx&$@Vi^NA4E(`{;BxsBsITY69-e~Qki~}q@)?F;?f3{CPfGN+;N?^
z0(Cv6+C9r*ULa{EP5X>(9OYLf(&BTzoPpm!+e|&Juyfq(&Bm=u*sN#pP@wsfOqYqi
zMmL+cmQzN5vewy7*CT&r^-!E9?MhPm^?JPYUd)T`L>48wGmh<O!ghsD83JcvN*tiV
zOuj`_cWv7o1<|C>XU@#2`3;3Gepjx9!(!To;aCXDTY)pEr>0IA>wvqtN-9b_QfL&F
zX~WZVj<Hm~L|VGCy7Dmili%x$GPQbRU5vERDW9g{X8mDJ_7!mFYervHSA5IW%Is~V
zmu)Kh4LiF208IQj9+`9)A8hinDWry-l)e=a=DBO=tMGC=EzySH;$2QHFCiP>PU+zl
z*V_@gur>*d7OEwC^=Rt7U1P$wMd04m*IL<L{NV3G#70Q5heuBHPraf&uBYZKM!EJR
zc-hGpboP?Evu3GuU5oP?J3uGKEq$bX0+Z&?DvSD~IT3voKUNR3gR@GEI!d_`8?ojf
zxnWe`aYtAE)k;=GpdrwG`8G4p$6MnNru6FF)QvLxWYvRzkFUB1W8zJ?FM6+hFL(|e
z+x0u4f!uRDE@XP#5SV;)iE9?NLrMO<zcS!n8C(SMAhDc#otX)@S+b=3*}=5pwzTHG
zXziyXD<d}HR0f@Jl=CVb5!lp3Z7QgS>nao0V80ahOl}u`jZP`0jmT@CkXh(PR&pTk
zy<1bzzhF|8Gl}i_C(eo1djWsr`<xdx0|%SZaqP<#<V3zKTaQ1nZL3S=;*K$He9d{$
zNQhnY0w#v<OZSle!@$h>jdM2G$6i&uMbr@ru&Dkg>#!3gWxkXqqtMCi>2Ne6?EAx0
z_dP3>&+n&EL8!N;yp4<cZ3`HmD2z}^c`tSedX$@O91Y6^QxT*{w{M8_g_R$<+N?W!
zv!1%S-gM4=u;yKgLy>oPDd73R5=A{ZH78>!#U1rIF@qulWik1s-p^=>9k9(I;>Ao0
zj_V9&Ev)V+SbW^Z)7RcB;@=$AsTi4o)-5BnP#s-W1SBQwY&GGIhnhip{cbhuFiYY;
zrI=HbhQJY@sd&ZNf?5gcMviA~(dH>q9xGf)qT}Xz$iBhB_$xWmk!MsiPYoH#2nkb<
zPAYwF&g0gfZWDO2cVETEn|hFfdkmus;F!OirBHsm+uqbONjPJBxSQo>M667!Txsig
z5BJU&L)zBOZ~3n-`SUDkV>2<ll!%3H-19~Zhv)r1HMt$pMq7(*?}1lV&`<E>&Fdl5
zarIepqNhufH-0ogskf0CJ&XKu`BcX;RaDT!cNu=ZhUV*T1a<5rwML7O^SE@L?1>ui
zLW@%&++;IMIvcM;=@~A~*FutS+Yx;?DTJi`I}&Ocf^HJ&0au|Z%ta#i7$pIh<lj0Q
z6?!;%n%z=QbxZjG$>A%xk*jcTU3HaF&0Lzg8MsKFmR|C=Ndb22Gr}*crH+6k8;9*W
z!L<$#d=*|fN&R;+2hlR=x;^iz<m2V5tuBpevWI?lW@4I-$Dup_g3>-ye(%xZ`cKpj
z+6SM0?h&82sUD83h=*;XwAvASE^-dAee`}Lbi*mUZbf;UE)|9&tz=xC_v}JkkTDa6
zAwM)7J!)r{3RrP46iwva)=fL`Qi^wEBbfm#P9ybHI>I=1bQg2-pp+rjA}#eKq3Cn@
zT-$`&iF&@%ipZosy`$kbMPh7XuH#W+*6HrNV)gHw^Sbo*vDS}6HSNkCYwnSPm;>do
z>hKeLu38Ly55b$lsPL)>9~nNWUy*)QfER64WnRLLmXvi`&^J_#YnRqE>r<o9vqbw?
z!ZvQ)8GK?bLIe{B#VoKY0bd_Xo_H7PrLfCt(t~!GZQviT?jdW0X1=lj-(%r3C){|m
z7Cd`LTuk3Fprs(?{4__h@-V0}@YK7yHVCX%9frSzM?GsL*6}=E7EurGGSTaB&m%-t
zl-tR4_7;Y*W;?=TuQg2{NoBMTo!#SIi;NkYM9vzj^y5YB)Tn<;{^j-FV&AYyz3oZ*
zAw!SDLvyZWv!9rlFRrZCFnF6fz`L>t9g)he9mFtYsq!e@2XnIWzVi8HNkHs5AIo6x
zlxW1VaG{G(?91_Y+2s7@lw4@#^aNUYU0v}xxo}fFyP@d{d3_G?bc(gLc0y-+I(5M8
zdn+$s*VOjXB%wDa0}{WAuI^^z#))4xD$l(lLI$Ete%ZCW=QujiHY#x4=&S1gAHv=#
zIFtW@-i$S|ZQHi-#<p!w_$HaynAo;$+nCt4&HdJHZSDW!SG6}?)fe4&U0vr>=Xp-I
zV;i8rYSCQmC8Z0*@FAm>l{YsvKr&l=2aT(c7eR+~`Mg@_zQXp1O`XuBTOqm0^^N(3
zvG8=Crm#b&Ay$%7VT4oO#?I}bI79Yg>}&{+RP*b)((Lcw+G(uK$32)}h)kIqCl)pA
zTEER)UiZfY`OqcdhEj&drSZRnYX22JWaFLfTxcSz8OXPw#dsHVMw|;C(qzu|){hKv
zG;M&&Dlk*C+pL*1jpTH*jxpsr3LBV}N16Vvp{Dr_hl!kq#n0QWU)o9Lai1fb*>9?Z
zd23SbgUPD7*+VUJ@Ak5S?GzX}Z5%ysO@|u!M{}(5TVOMyinu2eX^N%Se($*h7;A)L
zzAmOLGNLp^4*4lQ&jKnVGk8o&SeY%Z2kNQqM~!aDUsyX+w$D{maL*-}HyhTlp0=*k
znS)?j3hEi)Gdoj*BcGyCkgu5>lTCbRTj~sSp&j?vf{Q>EhJY9h^d8dpT{vTLv7BAv
zDV8)w%Cwp@b{)7)OVsx7V66HOP;9bf>r5HmEtHiRSfHxMZdDxY7O;B`itn!;{`W{1
zO2eRLb<8I`f`&N_`uVek*=!jy`48wZdf({(%z@nhBL{M^a{PaVU>0_6PS*eT_W#U*
zEFA1C|4$HvW0JJCb2E1#W|Fisb~6_@H+3{KhZ7Wpb9HkuH@1iSy>8#`>STVi%#amp
zU)X56!O7RseQ>$jV((_BOHjI4Z?8+h_vU|+<+H7OQ(e0(q^da0GKychT;WYEO@1Z9
zXp@5|j0uG)X<-mhQIvtphQ!7eke+~$6C^qXYjtU2j)cnP0eTO{9%uogo(CQfg3&iI
zH3g;!irnaV>(<av=LV{jx$wycGgrsK+1SwN0%o<kvEHejt^@YEwYeF*m8Fl9tp|zo
z!&3wY0j5VE2;xeoODqgfmeG|IffHsZ&O?j`3kpyeX#t_oMAM~H4aiESK|X{p0!`cC
z1GW1(00CTySZMqV98SGDf)I1%CH>b@n##Vs0=Y-EAx(^gvjKr+q2yo*U>Soaruv$7
zj*nuy5&9>rij@kKmqK!SBbr=(1=1(|)PGZ59NrN{q2z=>8d;d1gE52ZXa(-3e-CHE
zk8Kk6z2(-WR}KPIe?VNmp@{<Ci4d(pEQB9`75et}7|~Y&AT}&Me9)R$d1%7h+wi9l
zV4qPwo>Y}U(jNwnmd1vs&mV?wAq-Z=k)L=Hu!h$#!Q-Fcsw|3X{s0SE4aT3_tNlZ`
zPem=X*W9f!@aZbOt8Z8QS0t(*yMI%JW*2Wj-fy;v-VYX;??woBiH;wX{@z7Y;{S3M
zT|A`lzat{SnCz)<s@t2U#QJ|g^hZPbA9eJAqo4Qt5k>_KLJ)n65ui-)o8c^w;p<12
zg4X6{Z@O<{0N%*xU>CHm&e<8PZEel(2TM;?lZuLX>}aRxhiwx01&o4)c?8KAsAXKs
zybsYTZ4Fg_l|i3Y6P=$1;FK>i)cfk@T(&PuG``zgTl{v409J<P-o#3?6IhYfxz;nd
zZiw38BZ>3l=7y6*%oz><8A&ReZY#HZ)fs<VhcmYYE%i0BsRwk$b~%sh=_05UWE;)S
z2vjD!)1wI|nMAPA-du)>Zpmt;_d?6epQ`3tp81l3y`0)8S<;LBss#k+cxv_!KK$`S
zuBEw!XZS@2Mu0fr4u{KFJ?z$;Y{6rj^;3ivB^k9IslRb3w}-d}^L4s@3WBkxvvv->
z7cuM#Ak2-*hRlcHHn`u<`)BPu>*;s&Ed4ke4=bDhSi}QP@sGwclU>K4u{N4DQQbk7
zFIy~!G;Lc6m*XsybfM(x@qN2A8bjs6ZQ5C$kZuMs8D?jd&+SjhJ9t^JhROIgz3l*w
zxdBa%outgKaGl(b5Bf!tLrOyIoLiSL%Pj(_*kI$rXI}?$^wo>j5<{*EMdUlZA@<)s
ztDWN3hTqgt^cW`QHA%U*uOH1=TgvCK4|1mQ3p22h)=TP-Lx(ZOP$%nLH!GcKUTm%Q
zxZ#~L1<ZZ6e_^Q&-_P23#^Kmb)q9t$gjFH#(c;7TK%iU1hHLkeIx7^*-<UV>%s6R*
z;=V)iK`Uc*sUlU3L)s;*+V6+N4^ULw5cRGDa-D(-4tVWvN5^jD@VdQ*^#lzTRx)w#
zEkIXq^JC;%%*VHfrsBJ?e~+A<JbTxl7^M(DA2&vnz%fs6x35LtTP`?AlRp)UT}7lL
z$vI;+5)DTc>5Chv7MUq<w@*vp){C`Y0W~@GLIQEW_QSJ`k{jAOV#Dt+wj92(#bO3_
zXRKJBYrRqglY1s77AHLEJt!PBR=4o;|L(<%cObpRpUE&;5(@4d4u!z;u`?t=FocNZ
z_ph*Q=bK~W)`xUgn22Yd_9MA_?ZvCtLO*AOy@t%pYhD5rd)bwHC(32#njvl6Rt*=;
zm}c++8Pc6_Y;9BP#XICDBrgilfCv{W(dIkO?($bIq<}2TmALGn`d;|u-YT800Glm?
zq<V{g->e*>Bh$?Fl4nu=J1tQhaCrzhYhifgLcdXb*qi=)=bbr!w}`<(4}N*k2fz4v
zBj-Eq@mz^p`-HRmeD9p>S1@}e;-Gi=2)WCezZVr$XzrEhW@*&UmfZt_6r1Oi-bf_z
zlPN!u5oQ5>OU&E{Kym+a7L^m-wsVQqu)m`B)h1#*;aaP64{t}}yn#fZ7|~0wP-`!@
zV~CYr7sEVbF398B*0L$IZ$6?Un;}HU*sLIdD8)cdHiGqtil0{(O}Hfl2JeLk6O0VT
zN)s9R1bV`Sj{O4`s<zP&&Abr-Z%mJKj6+-Sb4H3ea779nhze=o0g|J18MSx67A7t5
zkfe3&6KuoU|Fxa2iMm`}I1cP3u8-rrGvm>;qQHiIgG~khj=lDvakekT9f$SgO%>wf
zTIfRjCkHo;L|M6|sqo;kXYATBc8$fi;??&{T=Fo(#5rp9%IgkdJRa90ky0p30#YL|
z%1Q2iNZ6a5tz)$Gb-hyBv=k-d&7|27b^nrrYn#(Q=J4?y-_fday0{pK<n2xM7AwHH
zC3v-ZTu3o<bbGz31zyv}9RO$Mu8>1dnTD%+yGtu+gqV!FM3-9t>AwQ^6MZj4L~hT`
zLvYm9hSy4E>GMx}gqh()^o&}eo}A=vqAq1<?xMlnoz7WTtV>{Zt`@$>%{YUJMz#8G
zdDj5SB_Zb%6Q)O?KbrYT_mhm)GAwBLAv4b0=qs$6@(qXHkKKNdf3Wofdu<x`AKKN!
zm1!Zpc{Sr)HN`Jy!RpIwsV|#fp*?*wfeS}&J(006EjEkS;Jkz5mpVYGozeGjSX?wv
zd2ainQDyZbxz;ZS;k<+c$=p<fW<J0RO)eFAnd|EpvMyo%wq09|;hU#~NN-*m=lhs=
z@U3JONpzbQEfko&zw8nr2j!0?UR;;MaZq_;l`0gpe8*?m;HuT0mNCD<XJ#Bm&jmmD
zU|B{f!DPR%iQyJWrN_A9XW7yI6y4UDj+~fOiZ@R~MSb|2z&%RMyh$7z;*3gv1iLEl
zb`4@sr*K9P!Y}5q{W2oV8S%lOh~$4>VxthYcsuUiCK+E2nJ5i3u!&IMYCMA;zNGN!
zNLY*F&hyCu-l$h^VP`L1!cy@kE+thzx%qM8=szp?{4D$KvJr{o8b)3Q)cN?2K0-xP
z4jt1oNdpZD`qxUOCkxy^J?~lLyeOqclka00O>BYzDD|{(ymYOQu9mQ6{C4vEFqm1Z
zFu9C-`(Y}kuLH~O-9+v08(vg0D=N}vSO2g%k{n}BV<JAw<vy{eJ$FtI^yq#j&eVAy
z1#nlQRlIG;V<fe!lqBdxw{{FH9+?yrzx=70UGsEf6ofak_Jm*Zpdq0YqYGzorz*N(
zy;o>Y=&Le%ro=r^-Xwkk`QGtjmSDc>U?q3N3o}N{;Gbj!36I>sa|amWB*7Iu>3oDv
zZaHk_8pUMB*z%7UgTXz+$WoK?+Qf}MWQt&0PP$!C-&)&Xj0<bW6xQK4##0675@{*|
zp%9Q-xly~H^qPCRePU3}{lcMLW{XN4>#-5&O31~)Q^G;9^aexhi0VIo{Wv$U7f_~e
zbBMZ_{(aV^YW6N3&zYCKm;U{w|5Edgremj`PnM>USiihHio_%EMMBBAuM*Ss+%Bs&
zA8ftln49>wM%9vo$j2k0hOGtAgyYKJ*3A7l&s0t)=-w^qHQrNK$|e_#`~fgL328-y
zE89$6-y^?zoz+XmQDXl<6)@PRSUWZYc60Auadts3exs&~k}ZGhp&fa<!K9sIxPkl-
zhgCLA;~`t>L4HIi7drPA&FuzQpOVB+(0zJ6{RvLj-K(f6(Ht8aT_lx>WS3XSeys^B
z^y0||TSh~`Ng=&&*=Zo{Y6P)_J?r9jN?g!w(5o;X^}{jAE{9#u%j?er+;RR$A8JgD
zLHr6)F|_-5a3m)<@i4KnLT=T?*Q`q4^XeW*`|u2meL~HO>svC=kEcu!<B!(EL-z}i
z$W#fT(#ZDURzKj1<B*_^G}34)q*u;dH~dd)&b&I--ANm;7X7q76s-BLb8%Q(lJLL3
z%o*dA^CbJ0*TxVf?7xkjPX>@E=Ppx_BcrH1TY+r<LoI8PJ`J(+U!RRVbHD1qpd!;_
zSMD18e#_r`_Gu<zTlmr&03OaXc)EniWK~$Azd}3wSjlTVvw812o1eD*<&|9&N4@YJ
zreSp|C0of}>b%VQj~qjdY!JggVq*DsY#3YA(h>XWTpb%U+9&9qO!btb{feu7U&pGb
zGfT`@?bpU;V;U%hcSwdac5HhC-43R7rAnYQ@gBK{2ZZ1f9&O0DzqIlrx>8je4bb#E
zz)#`HFjJP2&T33He&%Of>qI;_`Dy92{1{TYzCKO};SMbbKt+JDE0O6~i0{XL?~85v
zQT1~qZVjE2n(q)}9|^-hikJ`H1UEUCZ%Y0X<>&Q%w%W+Qc1j#&0vP{LEIySKx>y^7
zAm5L*jVc@lfmb)^xZgG*Byjk!mYQYKSoo2-lB*eAx*nvFhCuo-`mLHEv&7EEfgI*6
zj?o93((^uJos$%aYdd#BtR$t4`it`^STpV&ItH@9PLF71X<3u+N2t8zfFdAbSS9Gn
zw!|_R|JXM@uVvJ`Mv=(<%!N_KE}sUzw$$S3CG5{tc$%sVSWKmT$>ia9f&_b~@&bU9
ziRCv6lxg#MB134(zw2az-e+25xBm`Q{kX*f@;`3_p%EhtMD&;fHn=F<fF)o2N?jmh
z`mg0zVw?>reGeS}t%_m!{8GdVRRhEpz>;h<cVe_mZSZy)OqMyFN!0QMVit3*gm9zE
zq_T2PzAE)aZ8#Kxi^&DC-ZL{ZonYm{YV%Aa9mdd*PYSP6YavkR8)=P@vtURRfQvGq
zL4SZIKqeJKQ`AmUq_Es2WQHCG?wt3%uz-G*eHBE(UdipjQDyt5&C%v|7m_&tha>Id
z;*cG<M6j{9wjG$1Htin#iV|iVt!q8w{c|iP@~lJlNf{U8hnt>q?Iz)=im3aFa2ZsV
z3%%zS7QXUG@{KhR&ooDM(XL|8a3CqIx0LNJCh)0#dk3aUv>J31uDE3AJ$a1y;TfK7
z;~=)N)ciRS0<XcLWOto)k7d>jd1S#CNUm31`nkb-wkB~M_FoNh95tnfNnAHaISPW1
zgYmR@QoN2}PfV<Tm26&<NmGC^Pwk%%zn$t4yf)WkopF-tXLGv6kYhfMqR^4zRl7qb
zodh2xQ8+Qc8(xHaZuu!^7?HRIeEKgt*V>_{c1(y$$)wQ1&1Y_;{7Wu4zPq(Ahu(!j
z`q0vJ{Ki$I)WqF-J_>RB6BOCR;4xe!H8caCb>PC3uLi;l#DK|`S8{Lk)2G%QjEPn<
zJs#56`5L|Ru&>}B?u!Dyr4vcnr8Os>Zs-lk?5b6HD|$E%O9xD@e{+GDst}@={b6Wf
z$7ZH{D>!yjF4LFZ{X_~jFlSHsd%RY2&sihkTw=%)k71dfNYLj+Ggd_bmKLX*-p?X9
zt$6`$qBn+crkISb_GxdIvyNTDc&f{hR?o@JQR_8(sH*{-;cd*bjXu15Wdn=RRIP7)
zrFY0$(D!m~29=W`kBlisU8kliP|Zk~R&Gb@Q3m0SWCt_XESL95Iv@uj@81f_NFPQi
zz9i_8n-IOfX5%5vAiz1^8^-VzAd)!vRQ0us5ql?YbcsG^#Le{(WxiTq3DUOTA-ITf
z7qY-0H;Q^bMI^TIMu#cKcO~zfl{M&=qiiZ3HMdK2P(8eF9kq&&^)3il_Szqr8Xk<Z
z3H2a(`#Hrnp|;;+?&R7P+^L$p&|$_+HxqcK!!um?kxFTu3$c(kOF&#`MgV~gWTK?q
zWoI`evbQ=gVo0yU5w5JI!-1)i65x+3H1f78L%NU?k}^1IAcIFAI|+`IG>e<i$(Zv;
zZ+C&*k&a{wecft!4YT4BPIf{P^e3Oh8i=<ECf!p<^P1`P281AldUJFC?dWM^j$8hh
z*~uyC;!J5^g)YBxvJX#^R_i25xZVsfH;r+T)X!d2y=VH4T`u3%ElzjV<zxXdF;((0
zgBDG*<h|P?V3O7^Av(o{p^MJSj<vr}ZLMMAV%3{8=o$gQ(;7VqL?jjc3E4CJg)j<I
zW8j>op|78Hz!F#D9>9d#NA4>-Gt7*ehS#^3Oe-A%{63t@GjqsO<E2haE;npk6SMxk
ztBrA7g?7Rra&h-QycrR_c<DdA$-k32Dv1L)yw|7>y*e~h-yh=Q@c?4N6VzX4rw6{V
ztfy({;3}QWGl^b0T>GqtodPU?fi7<dy+C&(SWqKq?XqT={;x~h9-mPkwNo?E_Pq-H
zhn<QdwSFS3e{nj3cjVig4Wm%8rb3n>GtI;4<w>fVCtwP6K3mLjB1fD9^%DoIO-o*f
z#=nVZc9rqy&#MGP1RKoYL1ATzfsQL!#S|QGJ3}S2&E{FU9&zQr_4hhdO$oE!{5ITf
zLnTs^!+LW8JYwy}@iV$qZrE%zO^sV54#3O!`Ry;pLMvepih2Xz=gVXi_orqw=k{Rh
z4;(Qjw*?Yqd?O}nJ;VON0H**JcNzb*w%X#O``}CI)jzOrO?*5)^F`dQH6yft<WYj(
zqx}A4`nllycnnsUKiUa>TNY758rFcl2o_dZfgclP#c%#88q;0dChy7Yz0mup^*`xR
zZpsN#^0wm+VKC-w-B2+w-f&JxwVykZk0J349B{00Y!rk@P9EdlbVqUDjK#ftLCch=
zhYN8NjfjrHJGK3TwQ^7cOdDykot(iw*Q4LR(!RD+qJ%7-Q#yQC{LP=m*h{8Jm&un3
zSPCtTrYSkm8TrkEl7+T-8fEAFTWz#guT-Oge&|e53!?2?86c9!w6&NK6(i$>avhPp
zy-xv}u!4A_+O4Z8AA-$Qy^BUS{iq#kD#nyWy&w`G%2-`iNYNLV-M2?`z`9aZK~E*c
zg=Ms5O&q63Er{BBQ)xw~;swDmW4eiRJ^vlk3;djjb|;DzNPy^m$pEvrc-8K3z?{f$
zHNPN)Hw`~%2JB1uOy*Lv-QRI>AV^=e$f#`&PUJuKO8{VUc|+~+|5;?x)6?yXAQE8u
ze8$Gnt}h!SP?}S7mMXeKXhH?pEJj0HAkTs?F{2Jx28+Z8Vs-NA%e`YXp@px1-X#XE
zHl<@1BH@XI7l{%5{EeXjdVnG{8fZ?Pqq(8#X$FHhN9ZOrni>QGf-L-tx^&InuVP!c
z$WBC^%t|RKr~h8QTupIm02$(nAh<%W3#FyVmd>Klb~Y#FHn@9@mC_V|KcRN!7mmr5
z!R;l|hAdJXZS<n|QbPYYo*9<@EfKcr)HgA3-EhNVpY8VTp`gw)iV>iv3O+t1X^bs0
zZ15SxJMg=~*6_iR4R_o#?3x7Gv;^=5CG6m#Jv36K^!Q!YYxk%-Yc@e{VKImH8`qlw
zwM3o^u^Yk|JL@lt_WBgfclbrteF5I&1^M*R9$|Ya?z?A-ZB-IwEC$8VpZZj$;i{xu
z>wQv#8RSMqiOLtJLy4anW3o5Pn6HroeYQA90S#EVI}!K<nMgT=;B<`1Df{{4bl+SC
zuQv7q-K^okJg=85K9pArT^M4Hq{G+V;Mtv`GjIBp)Sw}iocpb<M+jWw6lovw|D}0i
zsac}cuRgQ>s;QSktC`#lZ-3TKn%40N)m*{^9%xS3F)m#l`9nhDkz5wOI#wWcGk`m4
zDoJWDK5Z9{zJ1t1HuapMs{%1sG&DPh6t6kq#^`z0C%@03X;l(5hPwb@LJNlr44KMZ
zk%4wBdMirD%^3EvgyZi1G>ukx2XTSiQB#*J1+R;TbQaEpyL+0I@7PB@;K7Fz&BV^E
zuYH(B9pj1jbS<qqgiCv#{{lFiq4!=k$}rFIG^|ZyiD%avzjJ>rKqRvr`!Q13R!mU(
zYuON0X&s8VoE>*TaL3JC_=r->oI83E1PdwhM8Rn&aOV6+p&$R*T!N%ZdUJbJ6Puz;
zm3s@+SYtr64tvg9y#X9q(L{KnjK*i+U<LUO(vmowm|J!LeBpiPDCLYoxDj@SmOW^G
z7}Pxx{gm{)I;N4mZWNUyQr(|sHI3w16EXVhfwf@vXuc%cwFzQP5p2eEqf;q;bNi=`
zk=ac04%jYC_(^#L%IkD8F<iz0-2-|n)(Oc!I!$)8G<VabstXm$qo^KNQDz#p`?<W)
zQ3%2CF9Q%uO7tU)%5`&|l<Kd$)g7EFtBaQ2(&`M|ZM>uisy)#tE2{lFKZ+Z<@Eqkw
z&S6d^nj1RrHoZ9<G`9}^O^TLk95<*>WLM-KEmtR2V#vy##ezBwHbQknzf8}iFkhlF
zZ0eklDXG~b?|E7?yQvJYQMo?#>{auL*h+z|r(FMn**zSqiYnY7Me$NOG#M|;gzI6S
zoZTG&qwKbb8dAhc(p0dAw%CMkN}h{&W&7aPD4GN+v!+^ANnF>$5%oBWvGS9pq(Dt|
z2oZYzzGL3Jai*AgC#);|pJ-Csq}P|0ao@4|<|2o?PtjCnN)<opC$4=m8)M%WQ8jy|
zm=ISXS8ZydfAb%yDNq}4g{s0A-1_yYBxoFA2hnUB^Pgy%`9OUQ-TnaTg#M$XbDK*-
z`cNUUGY~{zCJi4@QA$(3*V3C=S+cTihnWB3!))>$kxvpbv*;O*(EM-I>IAngm9x_b
z?*qT2=j}be1r*G+O8dgptvW^H-)v3wy2eQUXxNHH)x3x?Vu3ges@Z^xB;6qibFroC
za-+^H^`tusA-vK(XsV3BgNuV;mifFP*PTz~8pjINh@#ztsl&}UlH{xHzmI+)2mehk
z`KDeQa9Ts19+4kiE{`^dcHHE+N(Eq_(z#us)1tFoYai$@@BBuhLyI(^`?WXnfITV^
z`Coryu!fTF4e|*;lPY^r8cR~X&fVX3y`MeU7f2R^t-&NfqyBCwe}C9v#Z_6GZ{Z@1
zequqMd5E59X+Z;O!*z*YH7W!@QV|eEWP8xM8#fP?Cd~&|6tm!AEs6zh&f5Wv`$9C#
zgE{zRugg`yojNsLWSkYQ7OzBMg>C=<?Uom%IPW$b35y5UsJcE`$~8~0<%`~#zYWtC
zYQ{j2EZ}v(%uyW*o4@;^7brD28gK_B9^ur|`P2W}ON>4p3Tud`(a5>&_-c5m>%2_c
z&GRSynAg^>-vw&{;r#gf2cN8<eI8CobHUe&-sabFDavMA6(diBY)5`jE6UbYG{dEb
zvVgj^_S<#G9QRhtLf3F8Dr;zN$Bt#rj=a~w7jkKSs?TJw(_iD3_6A2j8-kU68nkez
zqdDD)!N5aC<Kvx)7*4q~NcH+|>JgZGu%}d4K@rh(_ldMvnw2anwNs+l?fR(o+bRLC
z10R%ZStrbilzlmlzvB1}ffbYy>DovJGPVHBYD?A}Me>OWtWW4uXO43oNp4A5!e0|x
zyF7FO0w+U?&tjVPtWG%|=AonUKAbMe-@E?hz#kAS*<b=y>w_@g)o`LgG@L4E`Q~U;
zVEkxg{SSmi<y*uVMpw*+E7wIKTq$RkT25bV5hkPe*7%|T*dqMrg6VWMKEW^QYdbwa
z&{twx4Rz%4MUN#g)CJ)Q8#tR7>jKKBFSK=JJhcge!c~W1X3;LGXu})%_(>48QtpRv
zlz2K}=BX54l#Ipa4irUh8=l0Pm*vUB^$stK;I30te)UP*v1#n8I}}s6Ed<1nIcVf#
z>JeTR(7$@|zn!bp4%s5kmqGStO3WOLVl>sGVI5nz<$LGsf#7~4BYN>zny&<avD|<8
zl$hr3#q_hxQ^o(~I{izZbf{_z23i>Mn93?I)Pl=c!>bUOZmI**>uTvCe0>*J>L3vb
z)+X|%o|4H~%ZcEdz>VXob5a|{(V}L`WfCF8rCpll>eI09vBI9tXnz`&pDzP03fEb+
zKTCD<5PU8@0{|6v%kyY-3vpRU`r}5htD&JUS}`!04_|dZdGhf`ncXj@vsEb0b;l;A
zi}fPbv0z@uO6iiZlJX3!B0Q%TuOfup90&T}%oa5S-zM4MO#^{XM=)JYoephjduzTq
z!B4(jS{%~%7hjWDN6>(fv+RoLB|WHKxyI&`mYn<bP>|AEQn-r{nS+a@i<nLYn&Die
zaug->S!_i}t7efU)(J+4{tOc};TII3{(WdrMJ1!ZP$GIkH4l6Ks~QZU^`k&l3Cd1f
z0NM|oR{Z)kUxz!_+;?Pki~g_fkfb}_uMQX=oJ*f+-f40yJ`#<)ZlB&_K*c+f<PIG+
zaD`MdsKSZXkcpX?T0Lx}a8OjmV9AD_H(H{OR`rG?9BY@ZYO(-YZs&17&(QH~Pm&-y
zu^TOfRX2d`T>l!{i1D}ccjE{y2?t?FdM<bNnyPCa-r{91@++-A!9Am*fs^>lPQFtm
zzHha@Q{{47(70vdN!G_zYmQ1FVrrOQ1i&8}qa>&B{KI<P{4q#cXkb4HHG^kIRa31(
zKLiz2zK@ZN)5>-~+S++s2HPp_y{t1a^yDo7HoCJUzV1WOR*Y4vNu2)O(R?GiWRN0T
z%Kiw5XynHK>9OnKabZ`rY;VJXGlGC+?7ZG<Xdq>*cJ=c#9Xu!DMa-$~F%3ZnNx+RL
z7q=DMkUBxZ>ALGN&m?l|6cFL)c?ii8&cL@iPNmbHz{g9xi@%_ViV-t=pMSDP$VAYh
z+3A0xF~inYtH)|#rxzHTcTBlpR0qNT1$KPvQ)VjoOJ<r%kyBM}AV#gQbKh@q!tzdC
zeb{1mYS3f5=bM&ncSlymLZq;TYuyNewMo4qW*MQzn(rbg5$}(Uh!pT-ly(<;wkV5N
zJ&d}{gl6ZX7%M+H4s*bFt8$%tz7Kq{kGp-M?vf+Q-JoN~$K$u5%ziPR82hd2h<PYI
zqeve9aUpA8-_#%JfOCR@47b{p?6Y4QUirajG~X9I%~!3qJNJ8m0<gJKDfy(K`Eq}A
z?xC5s?mBQLX1Wqghy-aRP+))D!-EHSii1$VY!#o$*jdtyc(iDDWIctLLM?^JT~*+;
zE)pYv#Yb=bvxWq%#-(Y}{91(0!EX0ne&dtKLok!$;=WwBzx4f5W1Bxy2g;)TXxuE>
zQ6!5Z$wj4#ke;yO;vccAGe{uPTkuKCMDj1U>h9{0gVt{;@6v8k^)M722&6?QP|ZUO
zkE1^|8?^MrMsgNt_O+Kqg-krmFU<L5Yx2?b48hFP(R*Hv&Uw^!pXTfxItI$%bS(kP
zk>jE-5CSB;aTk3h&g=L{+=T?)RarM3aLCPbsa}G<w4V`hA+YQ4L#qtPgM(`D{<$Un
zV9_-T?mo|Vb<CLGlKGpF^9vM3dn@m?MC}M`q>_y=n&`iL?@T8iwEMG|T5(Dqo66N&
z4d1UhgOyp74%SK<&dewNym;ysy+RbycFPlye(mQV4f7A?st9jMj&L@vh*y3}&ap7r
zp8c6(hL8!3+>#|t8+oV<kAsy&x4fI@O4~Y!TlmhW-@}X#j_JWUMtoH0-cYlP>}33f
zk<H@qIa;Ut`wDZ~i3KeOh4U1bzw1qalyc6A%NwnR@h6Zt)_HiLy%{d~;t<++AwY14
zEGq@^uW*&_-!s?%R2d?e=eSkksYK0oY@X|^`q&crOd)9ubH0Hp&z_$VFW}b1lb8Vh
z%Sj_V#lramWR<GbzCBrOCv<F8(~%JAN<;RI&ttZ-?l>xkM9MyT_+{Lf^_zNj(^`+?
zl)C3=)dx03wc^{MUSy>5UJ7jv!}zBi1jP}|2k~=xe@u(DApB&T@G#s;+$!@0S=8x=
zWNr%@R86F>j|O42S4R2}5S>Mp4WI}|t320{fYW9JSk$-!28|4;({x!TaYXBwV)POj
z$0^tHNI*2&bOTO<^=WRM+kf#HJCw2;CG<b|d=nm+wHAVDuJ$OYe9Hx(pf3`j@h3<&
zU6?(Eu|eC2Y8i$&LRP6A0(W?V;k26F?1HODazCjH$ObdFTJpSQkUtXqd7oAC|K1Ij
z(Cm3w1Zc%Nnm%osN>RMF0`zR7h#f~`wFB*KbFsA*7<=|{kzQ1+JCS+5?muh6pUB7O
zHMsy_pz4g_gWL*A+{RF76z!EH%Hik1D+t$2@Ziwqtn<X81CjNg_47UK_1UvUuXH}w
zY>F;99VoQ1LDpUaTYBR{uhg@MNV@;xpYa`&q%u?uOOj=#)YX#iOW_lcSSJh1wT-x$
za+Lx<4{w=~)PaAk(OP?>EW2qpX^F7R9<)jJ<vbSh<Xodll`gd*y_SQ`B?~;ibXaFe
zFAg66+DN?ADA#89;yN#%m;0gX7xKcYYzmZ>I(X#55t9d7_rJZ4_xGuY+Ygn*!FU?V
zQeKH3!i(Fx2PYeSn`3p6x?9fYy%w`-`Lx(;Var;ni{e(vS=8t~{4vkBWPzd@y5@E`
zL7$(~4d@7y5!wNx6oRX0&~c*2Ds;_1_<g3Rxucg)3FOUjDmQJjX?E`<AWhQ2RSxM$
zkq6$0RiTx;Yar1Fw>HSaOs&TGeq&Q6=npXoOnE=KkD&dQ!|`6zZB~Gwwy*PYDNJky
zFfmQAVyzGcOPrRflUV93sxrJ<`M*n_y(NT_Rfl7FL!CSSJ(5-=8~LMl?_tY(Hv}0x
zyHZ0O;h1VV`napkjNb1K7Jh|e6W;vn^Bc&JL>RR`tx8z~Yu6{E4XFV)jTLl;PAMde
z#DPGIlQK(Mj;3Okf92RbZ`EL7p9>I0{iB4~{H#7m9ZaCleTNMLw#|k`({zFRCC&bM
zNOnS+E@s+g^06;Nw5j%`gg7?2Ut%-2D2c}$60VkAHaDD#;F^R{tmMs|!80KdP%d`p
zMst|k!#Ehy3$>;_n8A&N5O+Sc)qYJ`7e8mQjQ4pN$4WrVyef<^ys`i%%~k-*y~O~;
z9-Ulw73Dk69Gwf^II7+#=9$kOQN6d6Hc2+L#`Koi_qwaJ%~Cx5Tae|fhPhH6EUn6K
z2Dw$j?c3@(fp9EyYiZWt=DNZoD}KoK_^vP$!seq9jtP^A(l``c+&JmQN)MefK=9kQ
zOCsy$-8i`;iMlTDa@IgKf%Zbt`{jlX;&}!6!ZLqr7+-;~4{0WbIv#2B%*y?m&I`!1
z@9)o~klJ$G;vf+o7eod2qinl-Tv+0i3~_71L7HD9b>_sE^;uzI>V?yu@R5XexHcCm
z)-9)KGq)kn*VJdGPgDq?#OrAgE{)`erV~URmd-`eGS8#U_O!p9&%BbnO+BtUy+)>J
zhk}DC-*TX6DPOESpwfJ0F=l1fsMqHlmwcb&yv^Q0JS)bpf{Ier#BoWE<-3225Ws7~
zYtP_NF|`@cbi>x>IMq|y`S)@(%AuTd{EDXM!W%Q$PGi4Ot2n!0LwG<8FSB@s@mcgS
zF1<0T-5aD*$PO{4&-!1J{eulM&RE}Sq{Skz%rsVSXuosb?q53yNqJ?W90;!h-p(ax
za9%dYXM0^%UFDv~!Xy<Ti;_e3lIH9mc6{~N@x2|_cEOyq919B%BU9v6{?pMKp<=e&
zW@$l%M%HR47Kyvsw#{a<PfWjYLXt2PfCC|U9qYWol@W-VnHAECfM}=Yrt4i{{fjF@
zH1pA2-M|35+$F^Ero{9~?<!yg#KfWT-2d*+Vz-PzDZGm?E<UYUsUo3&JpFN0^Ul8K
zI59_^>0^(GwNUq1$5{GvS-wv+5Zd59>Vp?qx)sgV^K6i=$q2w*9BQj7bT45XTH7FW
z$727_bfOI-32XX-p)xQbLXG+CJI7;@T5-c;jdkczDVhlvY!meq@s5z6rDM4P(j~D2
zMk9@i{9O%}Vp1Yl^0c1alDgDYwe_V-IJ>I?87)oZYH`JYf56wIW(5srf1SB3?4hHN
znoD@7_gF4Mbod5!pAIC#7Q2J`$#-u5rQ`e<Xv`h%xAE#XuYh}$$rYeXT%5#)I6E_A
zGMU}qs<?tianaL!NWj1X)Gpgu5P2I&637-_wpX7J8BUIlMtwtU0mXV;5#`a@-al<3
z(}>S$kmpOq3Cg0-&wDZIiJAcaw~$sTt8BZktMh&41fngfnRzYnjcFz!*~93?fOh#z
zGrIL@w^42V)V?z?yMnq#=tx!@&%kDEzq-$AGvbXNa@VINoCO+7aLy7m`m3f{89%ta
z0Xfs=;35vvrnvo$(b0J654M}EjRlyD0mBCj?x{Lm4#m?T&wwy4RBt)5Ot^s3zhFH#
z8Mjw9u}7$mUwj^CCR$C(lrTJ9PJV&QhR)HWkS*px`rAG!MIbI19ooJ(=OQ_w^ThXJ
zHTS=1L=mWX5U)4r>2MOO&u0M$&XBhvqgSPR-(P+@F$OX10ln#GTSveU0}C8~7h~?i
z>3SbbYZNy0@T6nCMb?}~iBN(wwaO`BViOS`0r?%|;N(<{TSasv^<&iwOtJlZIwH_L
zhH>vdTAgMH3&SCS|ISo$a#{?L-ox2i-Qb;`t_*cC>f)QhrAK4im3|XuZyTeiwA!NE
zD<?bJysj<MN{GG)waAOUm+$4{KJ$|e?)SWs(%nF>@re!&_R7e-$w8mw8`@xKeghuL
z!;51&P&PhvMeC<YGkgV0`T?GQ;;+aK!6(|636aNlLKP5&et)!#c^0MuFFjf+IOC&V
zzKIWgIv%TYVeq)~*?jiVIw>960%AXC=#a1FWkJM3&aYfk^CPO_`WsK1Jzc0tT&0Q*
z1@U6FgsPK!=_q1tzSjK?Og%G*=c6&MyLlPmUK_Yet5|a62k4jOZvC_|0_Ahiz%i#R
zP7Qe|mSWj6*+o7tOU)9%e4<@v(fAxZwARdG=|~8t`1_MXZffYQbT32eX8hrv9yC4I
zt1{Qd&6Zk*a@w{HM|~fb3nr&=;Qy5S7+x-^wT104j5{N#X5>ZS{l}`9uWL;XlajXE
z+{$u>D*nNcJ-HL-6a9Sx8_`T~lY9oB>6*j4VpM(bLAS?`%>dY@(j_~UGx}16o?j$0
zozDuJLA<W|ZzoaYHT#d0T07RrEU7A;XJhebNd<&jE!;&B97o}JcDULmkj?A)bt3sn
z)Jy;7B=I_?g;qs`?{IoN8d`oOTDY-oz8xX))^sjKd<=`aLh+~FxpJefp~qs;hV_qd
zLPjT94`PVFL7{Yk2$}=)KiEZ{ro~EUumzqOSL?l(i1va+0mOXWz&k}J%89_f@MixG
zS=eY~l)Dce`c$a!R*N#6ZpC1_VXsA8;*FFs7u%U>8O*43A*nW_rz@=)4(QixQMhKk
z@Il?Z%<WzE1Zc%ZQ7bE*3a=zKXc|wHLi5yRxdhUqSEHdT##U0u2F_XCpw3v-9E>?s
zV}YQP>;3|xP-)CIs5)*cmi!~)<N<sB*R7qdnQHa^)omFlS*tvAM^7Zo9UZ^o9RUo?
zmtDpVzgn7NoEh&tpW}=N`1Rcpo#aVT+L(y1e?zgjy|TRT0sGry&8LCdJ1OlBjWh~s
zr5#Zh&QN@tPpB(;yAhgg9mGWpv3Xj5JU`7?AY1esstVCn(Z9)GFnI{Dktm`!<GySF
zkz?P1-J$c#9o?-F1-Mngo8->@wZjvbAuh?~Hc&8#44?9Rv}hMYwu>*@v?*ilOVR-^
z@6|jbv)2pyV4koS@8H^wQV3ub?^U!rR|7i{+vQ!b*Z@cez7Sg_$QqQ)upji()V@7o
zuKJslhHYUxoET6sOIQi|USa)dO0s5czxspAznhHh;-l;CszSFZ(Itx3ZS(!7xUK_Q
zkg7j0%2WC-Fq0D73qlZh!{CECi7(@fOJieg8!kA~LY|)atp7nD-D<`|aWAnaN`7VH
z=(83zjS^omMzvthvR`|0yr%#98K>gkD*6BEEVlhjmAS2&a+VY_rcBARiep)|UeA8U
zZ|t7wXe^&SN8x;RoZ=Rm>s1$M@V=Xt1kBUdNxvtRJY6$8mCy8}!{#@=kWo|=*5K80
z%RlJCZX%70CD7l}g=btP?;^BwE$cchT}Htf<{^6Q1QfpucXWRHw8|o5ps(KM_U#7v
zX=KhCJnQhXpo4SB7|VUK75DcF5u=CJV?MFc>cQJot1o33ZE5IGZ0ly3IHQaAka-)y
z-x&e?ctsaTr$bL#+TpO;VkUl)kr3p_J^UXueiJ8v4ITdurQW<qZV#(;>>$iuLU^nC
zVv&x>57?zzx~ZSFvFx?#nQ{@*>gzB;LD}H8i5!|@jDtCr-lk$BfSbMZ)lGqv5XnX*
zlScs^o-X_wQOlYT2{$KW=P&W7Gk_cu)8uFgBToVLs*e-3>1N+DRK((*dS%*iUc9xF
zs=VHnafU-M*lVdSnWsGW!papVYlEmv5zcXMu)*QVqDBS~F`%@Rr{AA{<;yua+^FW)
zPz%b{BAcS3E%CrxPqzMzaXE_ycN|H>97&Y=a7||r@k^SivT%+jua1%id0n*@^EEs>
z|Dr%WEM8TVMiO=nC=EmAcg)~&Gb;$G<64j`ysqY_o|n0gHCV~B<LF=0wV<~NOtJU1
zgx^86{39?b{Yk-$-<N%n53xt?vZyu>RRbXdW^-N?&O|uI)*xUxHyerVfMb|@^ZOb?
zkuhnnv-{rWDt{jhz@B6;84rBT$bh=N@j@%U6Oj_#TIX^HmzgiOfqh@-&ANRP;DNd|
zD&sp0<}}qi0&IaLbF_V@Y`(s?`t)M{b2n5i_{8>91qi6m`~F^YP*U^m_X>nZf2{o*
z7ABHOxH<izTK0SV=3SQIjU}9~=z*=(3?pg549~i?7{+$+csop310>8kJnrZ!i|OkA
z!_n<W_k3B8zw568jlZSI62ix+@mvzpn0ADdceX>WF89$o+~L$U+nATD{!is=OTY%J
zButGl!f=28&S(m?-(<2DacYF4m(BL`>=HWZrP5?FT_tfFsEks!tJ)0l&5|`e-g@x<
z*?TkSgH&D(tRH#(;+m!ef%>>lcPY776M+WNtQJDgb6yg4F7;{mjAZE2ToWNaPbhj%
z%!K`8@G{^Rwd%^y1P29QTNzan%oS8s0!MkvE^IcOlJKOtpr&P2)A80XJN0vv;gX^I
zNxutQ<P4SXV-y=m5F)%tk~WRRq7W=USBl>m!cw8nnO{X>LnH+Pb~VWCcuQ6DkTK4!
zJq=gF9PHH61$dPd&_5AhU0I78SZl{6o-Tm8yp~qaQ;R+#W?I3QED*g20_LI&G2;dD
zMmhmFa&PTgfytsg?)YI{48=g^*E^#MpN(I!U5m|Wd<q3cs4@m!`@bW2GSbi_^b``A
zkO>woyB7wb=ML8%Y-Yf46}I=nc1R=uDPVc^J=9&aoI14LV^{<5MNH2Cu2YH^^(3dy
zOtu086cFWOI@1I*;~R~W(mMdnHD5eNAn448Wu=o>CR-3qW^VnB%P)$wgnX9G!mmrW
zYZD{ha2wuF+Z^{I_fHKh6Dbps5%;f%C6~18NCMPI`tmzt?@hV^NDrJk%zi;dM~)JR
zS<Cue{K1j}40G%^B}U5iic%m8p^iD@>01E90A>hPZaOSieFy&Y9GoSzsLs?)P7yjo
z=*N!7_DW6jfwDjFcaURjJ?Pt3!xkh1L9v=Lb>n+yTmHI+{=mOitwrstVD&IZhG*>;
zV{A^8R<HvF!$T=_aSjK&-G0V*V05_yrXb)bh!?C3S}8ii0A)S<KY`9xD?E~LjvniK
zNVB|LQNvvbqabd60y$)BYy}x(<!ECoQI($pf!AEyt8s52&3hJNCFYwlPkm3jeOSmk
z+CU<A-F=N}pDuU)I^TZxm>Qto@Ho7NDvWO1u(z`p>r$4Lek7UzBSk`tFW{tmvDE|>
z)ZTqOA&xVms!_@ZI}*o^j&x64B{CH)m!0v}KP?uZ#`{Pee(sa%o}EwlTVXEy%dApc
zP7ZaiCy^Aemuys4`zk3kHl)!Ma<<qDE`;Xr5k;Z{)sB#r3WBO}p(o8T7J1oL)d>HS
z|M9R?`Oo4+G7XQ)akOOT5L?ep>u+x?S1K$`i%zimLqierhi-M_3OX6re<!?8%hUb*
z|Czp&3$-=nz~LTBxH%`c5=(7;=htM=Iu|d1|0(s5%?^2};5|r*k>|NME)VQfNF8zb
zV5BLJId)#PRca*^O5FG?lmM$4lB{GgHgk#~%ps~-$SD=cXkHG;QQmzC;w83N1z&T5
zr{&YM^P(ASVY>CC-tOIG`b0uLm<4e;Qb`ps^W{^aND4{?M_g8ajt#-RS^A747*L0k
z)K7ezQ~f6pQ_JmE=wwbAj~1G6*(+XGFHDy$*PVX>?=_7e)k39%hY-UZrwUkc@vG*P
z`}3>WsT?{@snXoatG4jZu}>xkd5$3xHr`K~>{nPo5HB1<j_;rPK;bZKM}B5IXNxol
zZZuMRN0v41RjQAx*v4#-I`Oe6=ZQG5Ey6cyO^!@oJdgOW)Yi@#|NBno!_<7xTIkke
zUujOz+tfKWHaq=!$t>dYD7&Hh-c~vLoXu<w?p+ub4t48;C6^MKCMA(?9oF{QxZHD4
z2Cg2%F<CYf$zq*E6dWaSe8#6_ylWXyqB3x{enxx%xffB)SKE9L+FUDZ%LCeS2hd!B
z=(3x>BCmRM*P@Q^WAw$I-Ezk8_;@l4HLf-26=v?t+G>Ab$K?viysf@kajMybv(!7N
z;FmchBVXb72H4Lqm|f&qICmg~w7hZZg^90L`s*ViycMK3z>|v5cq;XVQl*mj=LM>3
zzPngSv5_9klrRex2R7zCQi!XN&YXOL7p>d(2WxVWV{T8>*r9vWeE^H3akPb&lnE=%
zoIV&CJ<4;-V>ZELn8x3;Gn?Y8m=jS5uN+q5E)e{c<tl|)S(n&^+K{z)rJlJW3D?$j
z=cvp$02};0a#Tfw@x?1<rsMUJO;XLsfsgv&!3m+0(8hZ*^WqfHI%tk`*>7KcurQ&D
zzE)g?R00=RA}!%%O1#<M-Nf(W?_C;?P~$&-N1f>CX{rx?dFi|1*xtd&L|n;AxqWpd
z%%(4076y&5#<IsHCo%-qj$3}`2c~Hr{Fgokw-|cx^C(e6hSL`YQ+4#coSc>y7Q$zB
z)%~WmBzCY^Lavhuf-)<ke;m>be_XlZx1*<cBQDrUzsiVpdAFxl>3Gss@@h;zJ`ZL+
z3!5F-F<mNrlD1M-4L_Hr1+&nTp;e3jFyUs(>{cPg{K7bJ_-Hp7s!(-WGUhGYjnGZI
zrv_C#i$%&LaN4>Y*0T-fQ17kD7CkK_J#`e>U;67o_EqIKnO6#W;vTdm*uUxjur(5d
z&21d6ooeD;{GOL<mtDIX1sTfY{u&3mbE#|oAJvAA#wJGXzSBW{6x}-t3_HC2JJax6
zcfU4@cULgXYQ3|8X1RXLj;&eO0r5QQ2OB{eXB)Yn&YH&12AL?q{Cy7dJXf6)JKmTv
zMUFZw;VAf*eyGd0n#Ds6v{rB?1i<q36@~OKrJLOPiJ{Zq@9ds6<j80Y0-|m8oQ*@=
z#Qg*RjSb|y4MF!iDm*7Wt2<ug7)HXzQ+w2B${&imjV`G;;@nxhL<@J1CgHk0r4vA#
z`9{q=)u4a>TmoKF6Tk+Ae%rNbAOJaC#_6z#8(m}oGy7VAk6W+I607ohiXSwh?A0ae
zX;!u+X_k&o|0D6Q?eY07rj+2ZE2ixalc?_M>NSK}XP6!R_AyfH_RSoQ>%#R2*7g7t
zRsHUQY%+=Ca;ncZRWALNgi^3O@M8z$KgH%t%3ei@_2E;0*u7BHdef0X5jOKx&~E^%
z`4VciHOhw~A+o2ERDGQgWcSaeZh72Ie8fW4hd5BUPXG=9N>w=CeoEzw`8_=#bIP@c
zND41BC=k6Jds9Ns#PVhY`&EO?_Q8V%%aw#13#e0hQ7$AwQsKZ?m?Cp4S`~X*INZvl
zdoE~y*-tmVBRcDbCsT=P|6iej<9PKTt!>2xvcBrGl{<O5o~nJ4sGZ;B9^xI@fQ*(C
zEAj75!E#TjXnD}EqZQBEM!KjJ^<Jcjx9sKXIE%N^tYH{xZ0cfK@gMpizbVW!rH<qY
z6SLoLSKWZ4rYv6_WO<q&Id1|Bks>KBWsjVw5dR4M5mhL=6(m*Cu`Shx&u<+|6~A$7
zV#kJ)7fg>gOPJ2hr0MzKKuT_;#_56b&ftv|3`9kNq0RBySY`INU82AL?R2k%mpmDZ
ztz0^pn#5DZB6_<c;th{pc&BW_zrEeKZ6q&t+^p5Xo+*x>GpVekK0l@>U*avqt;i-R
z`c`p+)^)mIXy}`gv7v$2XX0`4+tJDxf$3Fzb~vS3GS00+ZHT%rk4EcoLnAL?rW|ht
zmi?nVu}YqXF33uYUR{2CP=sEbD(WwTb{;a(J?TvtlbI_7-~R%b=#Y7Dykbb^{KNex
zK%Wq{2v*2m?iJgD@0EfhH^$1Lb3BaW)AW|9l=7>gy#lNdIO@OWw%JV}`cAhghPwU+
zH!yuz7Cmm?7(NCSZgz;>O;;d!R|DS^v_EU;=%%FO4@ab}&h_QtQs1ff#IuIdcBlAP
zW-Pl%W|apBo3G$TU4|8rZ}v~8K`aE}BfQH$BjU<5{_pz4Y>9M}J_)Lcbkkc)*rd7c
zF9|jZHb*wGhqst<_xAZb2?!BOR{m{R3XB`)fd!2jrClwxH%I$>Jw~^mT{I6!O~wqi
zWh#fV#r=UF=p3>I(f`FNvoZe<R+)v9<^NbOVrAjr{-5(jtgP&;Z2z~l>i-{Bu4*xL
zE7otKz~Astu-j<4f>ZqOs$5UoVy{sZsg1EQqj}@=@qFdn34B@+n!%nmNVBcsSlK>M
zil?q*2~Xom1EC<e)W^}nPWJ-Et<aemX=xd;&_KM0FF<Uu2<Wvy<c5P|vCONlO#ai|
zR!<j?l?TZV9u6)KT2%#tNk~Y@4HuC1r!ciKIi#2_;GGEyfq>u(sQnEKn)dsv^u)0w
zDmg?T;Ql6zZ*IAN9?|ON9c%xGR~Z2YqD==3^jM=sI3O*tAvU@IJwQiR8Jrlh37Qvg
z1MKe>z^)nqdaqF;?1M#&(6`x-WAs4;9Z>$KJN=(fbvyI-Fsd^sHn<*sV2K8BF(9Wz
zcoeAj2qKl&n4KIox#Jj+{Ht$ZZFT$U240T<MB}-n#lhkS`PG)o^EEZTuyJq)rn9%F
z*JuKfv4*i^MkBT+pjSP4nBwP8X=QQ!g;2A8)prLd;xUkH8XvIrZGC3AW0=-!plM3y
zgB3tGxA>H+Ln$1Po*Y$JUB9xk4(!LzuKq6GJ2o+QyfcJ91)UN8WopT$n-?|t-(#yk
zfRo;Lh)<?1JU-ra@2||OTgRoJL!+Z}uqF=6U{m8S-NmIRNBduNKHMch4T+KMP0+r9
zPa+RCuF;>zcJS;U0^}zlKCy??;QI7G-PKrq@K%VUMW?&(t>gjkKMLj}Kc1vN{}%vE
zK(oI;kof<J!2eI={r@NKf7j@Lxy1i}@AH3$mUOeVRW!B({+k5&&o>4j5BTSi`QL(#
z?W}CQ{!iimJyHw!9~}QrC^D|be+MCAZ}C?QW=7`!Fj~1tT6qG^l&xG%Edl1nwtuJe
zuU*654CriYWe@zT=D&jkFt9K)|2LnyrIo3T{Xb*p_zxG*-t50q|0~eH#F@mkR5Y}d
z>HhZ@_1`$<zskF+dpZ84_CGpNa4`FyhksB+MIAf=J`5}ztN;c!W|qGf?k^u6HV(i4
zL&blASpMs*VC?E_<q6RH+xFkgf13XvzyCTJ{5Kd0ds7Fqe|AO9)!5$b?|S`D!#{3Q
zH)rR+s{VW9|2F-f=YJO+2=oM+!mX`1nDT{MXJ%)))???7PKoOc_OpPGhB=lssh1Q5
z);jbBWJd1tR2jY3g)uhI@P2l9EuT7m{FbHtGv1G7OVyVSyeuT}BcmB8LfxY5;-dQ#
z%=E1cG3!h#*5_(Teg{IAlUei6-js^!w;%6Ngfm@i&M;R`G=XI9MS&RPkB&(GrWuV1
znB6GNAZYc)r)QKY$4`$`!xLkpBYiiJ1HbX<@)BSv_>8gr(@sfJUA>H)@0(0NAz$4&
zk}6;o2|Nk8r|7m-l|uhS%b$SfAL|4Lff?s%2?&wY3LaPBmj$$X50K9saL@gLbD_5J
z{!sd25iE-4xQ%95JaCRc5`B~wZU(9sHkh7`o+*VdJ@8_f1U1M24_7ll_()S-cRDsm
za1|2MlV|DXQzw}dd}Exa=X7BBHxPmX;THQO_T{A{?+vriFLykLt+uO!$2jN#bohbl
zfOpXWn$PYi@Za%Y6^t(7fmPG;0~)SyVJ#bA&{cs$4dgWP$3K1}dMZG6+!@U?KZNdl
zdtdC-%g|`nJDG1s*5ABr%mOGg7k#d#Ae>>wM;L2$UN)7S^1T~0kHT-{Pob9sw<*gw
z1I8&o{aql%*~V3ias;A|qE*wyY!D`6IIs3l1eI8m)cDRbyjeMzkzzlHNb4GDf?`dN
zw}Z@>M4Z3+*Z6MbxRK~m%)d3x6Y{|HOT=3`N~*eA#M0Ay&zqwKdNMcUq;UN3hCQ9I
zO_?m3YmH3c)GwEyQ+`kqJH@b0xsK!%pfAc{gJrw}v8)L*7B4K&(PD=B422>5K2eYq
z9_9?Q--#^sKANyIMntT3UaVCkQ26$o*=p1C!_+=*dK;e^qMQp}XbUPE>(%IKCw3)3
zyy6JroTw^S+;5>caFINM1UVJm0tS&5|3kC7g{)Hgo{#!5^Vm~Osz7kv_rf_*=5@w1
zlvhs=S*aKpG_fwo9rF##_3dJ2j~jUzJxrfr&tuMp+IEeRM{;VPG9XBevy@r3$2p2j
zJSus2@T=yV39BH!@-c*j!yR-XSq-$9aEA0Lp+$jkkuiEC8i8Fs3k5@uzp(0xFunND
zrlj_WGC_$cE!x}TL#@9Q9Q2+!;_5&_R~G}EjktnadlCdj%nxncoYP%aI=jqsl{|$H
zs*aN|yoZ?w45{QgYg%OXxL^E)WCWt}$<g7BcsRDric#=EC<d_28-sEv$$A6%-=tuD
z4=m+1xf60qIOPcSLN=$=ioyinlS)q)ufF$7Z=fQ3=v{!id>DsWe;M)y2{(z5E{c<4
z<?Axng#R`VT~9vjI;t&n?{+>2(zIDo>c|!2pj@yG<Oh*+<L`Cz`OT1xROFq^@KZ_>
zB>;!$xvesVP8r1t|8x)YNbJT*c(Ta~w>Q3?#}Xt>RCl#`)6`OMs}-|^tGjGtUSR@`
zr|F=-YXn|C1QngFzuJ#Q3k1~g8%tAor=Kd{g7%^djgxX=PbrpJX=J7FcO_-KI|gTN
zvhy@#OJ18djqXQB%X>E?9LEM40rj-Dtda9sCNKUUmV$FgK_y95XTmsmFs=Q+)N8cg
zUMMSaRb89k9xz55{KB?f(euCM=K6spCkft=$Ov)>y)ZS7-vq#<{T2+`OQr$j2WfhY
zL4F_3jdg9P?!c>VCa4SN8pmmvwTcO^#WnCOTxlNKV$#ke3DBTRH>8g17@JYY`15H>
zXm|7t<(0Org(Rt`jw^e<w<2aWlW@KOjbr&>W#j5FWx)gMg&17OyL%<i-bUQVQ!efm
z(z)vegZc(-4V}BQt)xER(>B6wn%Lsxf!PqnEFKkcfC38WS{q5l<^{%$s`=T)eZyOd
z50=U3WFzwvn@?By4rSuKBAMDe#~uM82ia;xiN6O=ZF|Vmqk;Y=bn+XnzhdPdl!nPa
zI=|}46}zW!D(1!lv^D8ySe3X0fNMSF@F`oR53m9DuR00Ij~i@6d}5ArAFZ}CsANi8
z>w<+P>`&qct(cGZ$<K?GXl?t+@e$$z4!c%vk=)VX9i5w;Xqv@n)$DgAxAvWkx<PkM
zF^gc~yX8&4nS6VP%nuQ8MC>VJn0F}Oxe4PU2ocjY9gt*D=AEc2k-X+TbVZ#uNQtIi
zDNJ)dYN{ITF;MIcVU0NG-B5JVNZp8bx&&jKGc5Q@rR7HlL3bjqZy?7qgIBOHGXkH8
ze*S4o-@1mjGC$Z%Je2OYnz$|k*foZn$dpOl)}Q^vY9|M3!Y^n%%)w2JGn&dZ2#qa!
z;3MT(>eL@<B^MM+HMnLmv@eVHj<SqH<s#m--W5;L56xi*A1O#|$xquzO?x-n6=ktF
zZrR{IODLGG{Av$wEC?f$UJo;b)e!764!+TnQpmX^hWm5F98K1FMs8yq+bd1_4HZhb
zoB)vvGbbIh+02$rM<I64!M#GzFj$lKi5uRlOj)i^??;1y7H6<P{P|q|S{)h3)TT0>
zqVKHA`xf7}h7zsV8o6PP)9Z&dR7F7FxA2<_lx;}^|F{_l5HpisJ%r3xm{JzXr~A}m
z=A)Ijjmf^RH0&J){y=q@K)=ys?i4xEkU_R_SEc3iPz*U5^CzZtvQu7f<+qXH@~>J%
zn=~FWN)O=_AKkk)nihX-RAbjR$uw3P)*)t4E6YluS-aUYJiY5%*0kArBmGw8pPF7H
z$MD7pKU6;ZHV`gKQw%;Z9|nRn2bN)(LrUd&d>(yB(JCoA1J~a$$O1<z6xnbre6vdu
zTo7nuDf>?b_*_#Q#NV2-bF<yQh6uq%p-BSjnnY-rvaqB->Qp!y*{Fq??bq)_C^!ZV
z;LhMhlEG8}qnx%zel?eZ={<P<u`JkDke;m_M{EbtDdq(8X)3MYzl3U$9Tn?1nC9W$
zuV<07$bXtk#GsxxHq@Di{DIY6fXv2_6{{OKP&-J;G=Q=Rr*$+AD;uA%3tmIc!B-s!
zbhKxkbnP3BGn7M(TJ21$rBU~Cu#c@`6Lm#AY_>aAit+wE@!Q-)+j~nC=v1iY_FR$B
z5KqXbhLqqGGg)R|+WFpm)71gbSO43AEvlf1&E}FC+fV5Yq$B!ij5yaO646?!k{(}A
z2|on8zlMwpBnIVGyAqr0681KE+rk*EzGI(OY8+7?JVX|(2{%|RziU;0pXepy58L!z
zZxC-D#q#KY!=sH3<gcXQ#YT>1O(;@x5!m!zN7hc!Z6Up+`1wD`TH3yW>NUR`(zo-B
z2X%-mx*I*9*2(4KKE`X$v%+P>gul)Mp~2|V>$Mf&KNE_%1G3A_B!~*Ka(UgUdMML!
ze<B{k$6SC$^HvO~Q95rDq_(yA;VV8tkCYIXJ=YwniUxWI;DN&uuVg3cn`+@PBf{7m
zy$F8;J8bl8ZEx=Xpe45?IRexAgvd{Ng>bmDQc=%NAPnlX&o^&jT(2c8Un%&N-g8!+
zk|uRDN63+s@n!;6i<Yu|DC}+RMfk{T;J}J0ctTr@B02e@XYS7Bk^&hu2w^}B9`uA(
z-w+c3$Ey|PT{G*w9M@UzJ8*QV_`;HXu$MG+6@n%QwT=31QT(aH%}X_?!ByybG;^^0
z4bIEm{o_sKaEPXt)`q=B!rxvRgHZdJfjk;uA9yhm>yeA>sc|_V^P(Rn2-hj&29Xzf
zmsp!2-W4I9!umawT1z^!L=45K-=xYei6Xniu0~izyP42)TXtId86Hu(n>Aj#9Nj-r
zS<n>2=eL+Q?!3$Aw4D9*$u>x<+q^h$yG@)1l5WewPW{o3nW$6W<lWJS<Tknch|Obf
z5{Ox!JPI&8$n*<_$JFC40w7aR^4T2PgOYkBQqeoFZU$J<KvM-PY*<(Uc30EjlM=cC
z{Q57AbXr>}Q%pZII*+tiIl7#!2*O7!jy1NzcanuX@Q~WACY2wpKGm_h;>)s@k4I15
z2F^a86-b~oG+&EDWTYzA7W5}pBTD#NP~C7p?}QN>#)m#ebx9I-fjv4VpnBa}+nyIG
z2T=s4_}b{$x*jF=AWGwp9vWvwH*gGsG7iPveCCRajTE|UNfWsXOA&KycF(lKm1Bw`
zi5a>WU?Tnc;<<4LhhaR@czFq<<X_?ow&+}BJVmXes6|&o59jvsF3n-UuSz)UKOQl(
znaVM#1TV{bs=w^=yiv|o0>BL?3`LN$F52jXP)y-ViYPi<*tO7mikMqD_XCttIiUWK
z`!7~%rplm&TIOHdH7bK6@yE^HK~nMhPGjG-t}-abp13SXvNN4Wttj!-V)f|Ls5IEQ
z5_`WP<(MVKAwQ^PwqlX0LsFV2WKb@)j8$+D7iw6C9A<b^GHxWYXeO`p(#Xx6nzc3c
zZQ@VQ0PXG@(eDNQHT7NGSQ~5*=m=3y()dN=C2@NpXJeA6w<t|E5Ddy1cI{QfVE9^Y
zJb%VoCTOFrCm$Tp%nt82?VnKAbWYjtV}j?yWgWg#rD@G3l)n@*NVdP1#+CF>oaw9>
zsKw^$E>VZ%*oKTdz}&j~LcXvjKVtiW$%*>T(Zhoa-_un)su_keKYcSu1XsHnD${7Q
zG_HS8k<?;zN&U&=8b+SHB(0K5-fof$Li%$@;|I};Ja6#4wI?MmRfe0dfIHITzCE$J
z@`Sp<pMD89Blw|e5vW8;RA{o-lpD)daaYOUC&#wv+5%o)4iIT;<Ka$Z&AypNmF8aE
zd|uIoT|7`oNakr>Ge`(R!|%1`O{r(i67d0!0%RT|@xImEdS@$$B$N_P23`T@Inhc;
zr^4);My{%pRcQ{zV<xta4c3tCX}9A6V=I@6Xc~PqTp6T3Igh&FdbPu-kF>jWFdL)E
zfY>)eC8KuvV{Br=&cd}516{{ImEzK*P%9AZVb?nYgBD(KH!#AdEaC`741K}Mm?ga6
zEx|d~1*|)2480@{4?8~ZIhp5|QoZ6N72j-}6H^8tNsy<9E}tg2kP#^$)Y`Dm+Z1jI
zp@jBh4>uzQ!Njc|e>J0BO0%afR?nZ&6D1b~2Z2NiSTr5Di}}>kE3;_H`(+aDV?I%x
zL_rB;e}5vN((~~3zIZQ9$;i8bn$Gu_;j*sF1eT-)9Z&hlHdM58n9_;vLoaEyC)Q-Z
zlj8_V40D^i8^WEP3Iaty^H@Itvv-W-+`;2V32F-68aUf6MaFGAqL#)2$n6XCR_U2S
zL&XPZXp3v<NnWmRs#L-w&zj!ZHYW7*X%|(JSzV{udeOFy4|aWnS@d1EMTvhZOMFhi
zixRxC(Ojd5M9cUjPtZGZ$>CW3gwu&;<Gr!1#BlEimD@@uOIQxT6HEE6szhoU!p7=@
zMT1I=LhbG>s<W10vt-#UX{VN70ShW*dOY8?Ah<dT9{CO)lO;fs)c33C9TcLxV~)qg
zE=Z+sVwuRzqgL9z`7O>JWhTiD^5W3*m)N?KMA9PUKY#VBIB`mNxDK1$t%`zwf(%wJ
zru#Zh?`4Ui=!!GO4MQx`!0*a8A*#!-CLPWv;YF%Pbf%X(^I!n_&gZH<XXe@9&&xDK
zhARu$&OE;)>PCP(GSm;U<Y^jdY|+)`{&@2GB%X{d+eYor8HoCO3lrqDC0-WB(jge1
z`OI0t*r@Coezxg?MNd7s2F4Rbhn-3u@a*<o?O8XEv5nbEr9F9|en#)0$w=ux<7`2_
zOenBRGc2fC2E!*ns)XNd*{Gy<e;|d_od@6vRyl9wcV`5h+2M?Q(pE^<y2RY<>5*zd
zzT!dLXO70*io=|TI8r=a#6sdUT;1V?iUx&dcH4EZ7X0!P@u$DdACy(}zIl=`BKw)T
zb~TUfN0claxst4D+|n}q&ai3<j;8=``$f~mVP~?Ioua~aT>u^^IEc`d_)}fht3qyo
z7s1$GN23x6vtk{%Qd26(c1FwQNZla0ujmMxzq7L+nL0w^`gR2p+Gq$zP1*a00P{nP
z-M1DME4|bzKps+$N9PT<a?3{dS>(nV6>a7vUeXGiQ;cpc7__j~rjHA`_CObBDdCLf
z#Ii4@tmtil<F{MdnN_%QjCB?j?wMJTBM!8WngveYHUfXvHy@=G1ic;qo?e>w!v}|y
zsi3l};qQmKVh1`8fYKPHni|Cl9K^h7zMEvE*(?X#Dgxh1BU^kR9r&S{>7(a><ua;P
zNqhGQ>w?Y94b;n0@(S0}X3X;H2{N35+=DpC>v8#hB=qu%h7w&+GId8>*kfd5dAJ5_
zD!Yk<oKxz=Pvt7C?+izg1s&$CJtuXfM~B@GTr9@WR)aMW>rw`be!ebOTzR%$>6%q9
z+v}w|CoRDEO;1a$4oE5Xv}$qupWtc288ZH7+}Dgu^gmbnE3Q>w6eZ|G9dB$lWW@_P
zLJ9xy$aQ@h!6(b)T~HXkS<-=$VFYUyBf*B@JTIf_*d4Pkz?8PiTU!S9EUy}5gtJ;x
ztdo&oe-<Q~7a7sU8~Vk-UjlupxV0nRyk}o?P%n3Omt~YdCo+ou5WB4y*+{$nnZ#>T
zEWY8Lfk$x+Zjyre#1M~QPK;m2*z_6rszFxbaRbg$L}n8@n$%aE<>TW^B2K~^{3x1a
zTrx=p6@Xpaklw9jCBeexcoBk0+l5vKv%t{RS%2edfDw=p_Yhu>zbOcw5ZL@}S@PO4
z!95o+Hj*xiGKF|dXF{n3$Fl;))jR;R>*EBi;?iqEgEgQVDQKkj#MC!iU9xPvkmSLD
zzsK%VeII5Y<Q^{&Gf3qJ{tHxaXspnm(|;$gGZ5(JLPo+HIy?&jC%Gt58x46i&p0tI
zM2jL}n)e;pabE_g*R&Sd^da+$V_~W?3JAT=Uh>i0efbn5kS}x<3s?o|7u&+@4o-4I
zn)soqQHFQj2ewBi9J%C?d$StaWNmRxUOK!4OJdhD#XVF*^ZiOT3@$-~E^qM2ndqqz
z*x|??;o$4eS{OAeUO)jz=yF8&Z;o%<HXpej#&?E-jr>du;AtX5JqN}mQ>TMe)4y+d
zX7iPq7&547SxjgW;fG~`y4&LUs^5}{?pp2}zd~<5@92B7qeo+%Jb#-&b7Q`VyRn`O
zEEO9{Ausotv1r!4L+_hVKY>yaJmDK$nGOD9L@*NZ^nKST1gGOEG7oAFYDC+_t0>52
z(SY+vymWHxgqJ4xsJA8~CG*x{-V$MmW$a?{ck=|ZlXK6sjP{=SdDf><r5gKB(~daV
z$m>q!HDvk;-%LW{K~9J6xp&A1K6&Y|$kb}q6U;JhvZdw9-2zgV67Mf!gBWyz)jWKo
z`kq}g?ioc!G~V(2blRNIFa4mrT1M8p5k}=;{35otHAKFCk}hWO*<m`+H$HPAeyM0}
zUOd0-!z|WNq9OY}UANxx$1Ov9OR;o$iB`X$J`Yt+lt;KyrYNDjY&QP3&qE#?3)gmf
zhnOT|6?o1@<}1Mz9poS3=J<m=P~U|l@Dyg%#R(Z1^thjwzHl=}RTcAz7jt{$_VcWA
z^TgYJoP3|j+Q5*|8@;FEo`h|so-#8gd(WY3@4(3E+zv1%?+pr&gH%Ua*9zcTzw$O|
zb)7o)!%q{7MJ_ts9YoES;Zy;C`i6Bh1cPogRZ5%dC1yLK@TELFlj5y}h+QEITtOtY
zs=Oh>AmJ>e@~22YNEcnuLnkSfaEjQf<~ApL?pqY@u4=#3b=?HXbln?x=4yakYn}A$
z31j+b?5DQ{8x>~LCJ+R#%4@aUS4;e1D?7Xm#UV>Aj79!5N`d8tq%VWYjOaqCpM8!B
zNGGG#4#audocpG^B&fG=eTu38lfFQ_Z=ZW&c2wg^y+vj;yG@M!u$~*=9n>rdf1R^x
zHv6@Qi{D3bD_LC<b%7rG95!<OW0O4odLv|h-2*psnVhqnhKQYRMpXEe;R*)8H5;EG
zq4NmL2iJi*Hx`($3U)fIM)3G$w}rpz9TVD7b`zy=+^}@;sa&T_;)U-V?!v-mkI)Db
zYQaTSifn=!0dYUERd{*7a};=NuCkVs?dVQ1RdV0qSep9i4Li(<TAg=#qC}r}-gba^
z&ENYHjIpuMSy~IRjaAS(ap?UMFR9{u-`ZPw`*(FA^OpM$FTh3%*-tnl*+~K`>LrOs
zcd|<Ajr}yeP9v?2LM{da1qzL#<9$Jz)$x(LOz4&Z?4pK~gQ|NiS$b&=$DrxARdhw~
z1kycB_QNk1G)U1?3QHnT9pCSPmcg;~ZGrSeXEqDBE^C5Fh6HrU$M8h;ul*SXyr*{A
z{YdKTJv0hxi<xKIk7q+oyG<69b9GVK3?RLQY&_EunMmPr?1t3?Q++731I53Z01ccm
z%j;*AdEl>DPp{{Y<nh@E<d{F3_mWN$+T71hFRjofL#xnqx92~w>|q_`$o9cQa@Y}L
z!6st71=1vIv7z8{i;Lf(i9`gp$JK3m*F`Q1e^`Fu`>2Q#e*+eky`#6q26%svO`h1M
zHp1=S%u7|af}KR>6#r`Yb)BrVVvnG2T=9K*@r*EpO{W<qKIX~(B;46XnR9&5v!n)<
zq6pW{uPd|_P{vx^_g&W4)Tpqe;9Kw6XinAi`dL-&;^OM3x9hR4r5f>@R@08NL9r<%
zc;6p_c?F|z+A1bmU53KCNBB)d0p+L<J#AT;+aEtfZfl&)R;n?hj5w~I3zdL@(Zv2r
zl18MtDw*JTodO+{s*`W=J>F=MRnDQbce`;?w-Je-xW@=KE>#NRzxo^uu$wIRph(_T
zA(u&uuAbaK%%%_WlRC9PW<2L+*zvd}f|J@7QTSy@`EB+aw!B|i#kyU7Dh~-QjVG7q
z(>=_#&o%f|OUlVv21ZT|_}DIK=~v-?v*#OlCG*UdNuUWNjRlQ-Nh2`k)_38!X$6~&
zHn3^hN4KJ~#mB=n`$;!|lBVx)QR}0=GS8PNz_h;z7F+^{65e;zZNGs8`%HQf(6pj}
zROk3feNorHxGhDVXd;$TSY9J9I$#>SvT~soEM!Cd?pN7~Y83PtW~M(0F&MhGoC*b=
z<$<5JVL>QY!DyF<@YQ;uFCf`)i&o{>T;I;->^?dcM#Jv@PNsFTw&K+bN+TT6+R`>+
zcQZUCHDF;d`j+D&=>qrKY%~39NcC=R0S0Wf^RfS(8KdoJGggHrFC3{4Zr&N`6wx>3
z7W-MHVjs|-(0F|h#gJ2b@<i><C(6;lK~N5IWvb$Yrbb8YwuyerQd~~Dq)T?q)tk(Y
zrGtxRC=e96TVz8nK^Z`@e!<c?{je=kX!KfAW~M$~sIj|`;6@Ll{Nc+Ce~M4FkK?#K
zkUaj9EtK9e^63>t${@hcoI9J6kIn0Fx1&RcjV!6+2noIXVh;T&hS2twtrp9TS|GR?
zN5MogF>)hmU8|Q^PnVS!60S&9m8S08HWm{Wm8}AY7{rQ;-dx57!4k`lBht{*86gNN
zA_K_gn9|euUG~gZiB~n<stc$LBxd*CD<sff_D$<8x3e{3qFS1Faa_P*nNgs?9yj^2
zx7%riOI&EafP;?w-9}SK?Ylb1wUbyOJ27Pht~iV@CbKhQSA2p%M4tKD*VN>u^I=Xd
zx#)dK-mXZI13Fi60mqwK6G#bY-I{G8KS1ifUGqp^u+l(rptupznBu}-VtJh!Ss}h?
zvmHUjAXF_44x`U14Hpk~5s<PKNZoMI;-!^zxy_$Dl@4rIWnO#6@{uLYvtvYl(bh@3
zJ3$Ur$zoaD5Wn^J7X9^$lQ`km%FY!`%u?aJ{YLvw$qa7ZRtToiYrM#1SBP-R2n)t7
z``AU+1jGrmMiFx_albdeWPi@0-#@_9_3Asn_qn#?uv0~g`D?NqauPE{ueCv&9!E6@
zHFhJo!NYWv64Dgcc@kq5()RUF2Tt0Oro^WCC)%#3o^P2O1$%6zR8?L^fl$o)!d7vp
z2d{i@ec;hV&hvZA3So-FI?l5O3(9=6x=Te#3!D=7M4y80%gi|!{0Vn4LHM%tE&Wxv
zGSsis3;sH96^_L15oz=R^o={Ly)w++a=Z<Y^)wBW4)e_clVy$Lta^MNiE1u@KJ>bI
zE?s3Rzs9j~NmpVS@G$TQ|AX5F!RdAgS$^<IxE7|OCIY|QZ`#HEjlw}b{mco(90Ko)
zp?Tn#zQ=QwNK+XpQJB`*O*_V{KBbBVMh6n{KKoHA4CRMK@#o15V@dQu^0)qUqYM*g
z&j^Tpssbw0D4D{KVcXUlarT(w37K)XEG&TKha7<(Ptv;34-*?4ls3Jvd;7#Cqq*1W
zH}haD2C3g3y%s7+^0um8aJ6WX$npK#8YgKLCTq!VeMvYuI1iLAH&hgN9C%ywv+U8%
z$I$oS%S|HUxDUsw;x$M(g41%a{=a>p-7rUx@DN#5I{gHL--bzI-W#aks5P9;TTWd|
zGLqCp3q#jeb|qgH@vww804Q37tKV6FMkn)LocixN(3)GBU3qBHaaH7Sx%h3YF-4y*
znqsh?#VoEq-?TpB+u(;;$#;Q%F3+VS+L@6AhalEZY6X>bgTHwzCfth~rDx{A9=+(h
z&(~OE<XeQfkZJ*WxFQkgfA*(>6F7Bh97-U!#38lY#I!z9L8!D;Bnzq*v(%xD4potz
z9fD_9y@150+A67gi4ZO2zMuM0Ohw4l`+48^$n7y%fT|5JNMv6=D=v=nH-q%)s%Xc#
z$LS?aqBGhyf27|1((c0eJqe~VyoLY|RvjfRy6wYYnJtp({$h6lrr>}Ka>a&k`$)wX
zD4uR^y<VVqNo+!KD@db+#Qs%IA_rb4XXoX}Uv!!^;X=MbEIxpQ+$7#=PT3ENYEFrn
z#D}ju>arqv0S}BJWE{uqL0#_7%ekf~r?wc!PjqtSCoCKrK~kIy40be;$Ud#Ap`s+H
z{Gw8qJ1E+dinFXA#}0HtHmxu9Z#bBB(vTeGC}S4a*rcSCY^N8ei=agENLW2wpdH3j
zinRh?dZd2XW*+ln|5Fs4^X)EXvIm)u-Vm2=)x%}_E>VXIB|=xS235h>Skq`HRO%*0
zyuL+4+V(E6f#`emWH#vuf!#x`Iy}6F?1VAtW$XAB1W6QtmNA4)D&>O6iS66{>*b4%
z(L6>L)WXd!>j`7pzL~|;O^^=5o@13&nn{ZoJ$84-?6@3Cz3l3ZySG9r=^K_%8-|CI
zF31)#q_^|a`>DeIHi`Jy<|>)Rc1Pl!Y&|yB%nK~{y-)FytaO>8)Sk!{wDuwkd;Jx-
z*t?AFOY>rw<H&mFV?l^)Gz*t+TX>}z=6U5wAXBluaQd7LXA1qc4_e<B{zpn6VCDJQ
z<)~m3`Wa3>b#kq)Cw_W1EaIX*LRMlS##Zsbvx!~}25a*WSOA<!Hz3G{Ju~j_2}K*1
z+iuy4(BwVMVxei2;KTfmabY#r!)?l>3n`wbLqwOPpjx8HIVQI0)KTehqR{J3QKKr;
zj6Zr)Dwe9lYmCnd-mH+$Rp5dWFy>u*Y}TahiVl5XfH%4~f4huU#Eo31klZ3SoAd>S
z$R?nFO<T*QiO!H!*M0#shLsY3SmdwW@;1qa=i{BnKECWt9K`iCvA~7g+G=9qXstK9
zU2g`khsWRqx&&4idnz{};!le-o{QKxz;dYg1t-K+GNcaP{cL}G^Uj`SPau5m!i4<=
zjp>+c)#g4mD&JnrJ6od-y>(Q;JQ&1mww;IKPZL<^l0rWHtL9Q#3A*}czLW$BdYCpu
z4w`Y72Fcq9SIw^}MVZBXKjEIHjC^i$B*u%qx@^~2rmrz+qUIADSoHV;ocmtovmUAt
zEf)^?PXH0)BPlY%`gg0^Tv3gW_x(g29=^zYsqYM2WQLtW0{C+MD^=Wrw|pu4F_9YG
zRl1A!sekft34sz?KR&;cmQp8^!SsiwBxJ^Lhmu_%?6O=|6cU!Tge^t-&Rc_u9o=~2
z*1xzO%K00QDeM1eXlZr?j41Yh3ft%R!eW<kflpxh7a7dj`zC^zxK^tgoYOmEyv~CV
zkyD~XJ)ES(1QjtiK!#VHWd=AQbLm|hb9lT-W929-kVBWffHfQUCNa&Ej+P6S)%h}a
zJwU9}Lu(1tQ-ybk|LEDl4TzYCYUeI_SalGC-Z}D?!<r98Qf!+EUb2AM3ee?rlf@8-
z0{z);UB@|1wa(uG3tvi)XE~GxQdxFybx6v)e;fuR;R$DZB$Pm`bh%uKNESg>-#c(u
zdrY1$hD_+GtOYWl&iK~kU?dFgLbX2}Bu*e+l@VmR_?(v+OS)WV>V83h=vRC1W9!X7
zPq}n>v6V8%*L4ucHJPq}7N(#hWn}P81-%J(<(6$79owP|dWC2%4(S-F&yao~scL9G
z@*kKN?@8hJcsw|S*{G{3uRO(H{w~kOwz0*ab1P`H^517b^X?j_`+V257*OJ1ELXmu
zq*7s+2dAHI<{yii{h@s7?qnLU_?_X#dCA5mRfI8i;G>O#v`)J|0DqcUS9g|QQZ6<L
z;-~&&N(l)k27?FwAJqG?sa<^gHk3qBI$yeQln7a><A(L_&Mp(B$3Gb}jLPf_cD9Ko
z=yET2lL-^o`r<<Jo!b%#mBnO%OHxB=8-{a(N(Q3qGfZ{O=PV~LO8FLKj!LM<75>e;
zc?Yq+`PrzkS(k#raU?xjC6)>;IEKVtg7RLOw_6K!KsA%!&X{4~aaZa~ES(jBaFBis
zmiNdA3K5k_kJp{p@Bx34!{MZU<oZ5}uH?To)o|A#4HOzw>bQA5s9Z}RXJ)-ORFM-Y
zrj4&M`%@W#wkNUw*3yy^Lqa&z$PG!|>dspg>OgCuj<?jV!Gh*W9Zw<1u@w!RqkhmG
zQ>Xm!Crx(fqL1%yH$k;!34D5V4ZMuQ%`I0)r7z>@;4i~q-<q_mV$e@3EYLChZ6MJ1
zQ>fnDvFr$AL1)cc!O^7Z+s(1=_LSR=na?0qR_$Qf;qzgn-aec9&@^#^!uXcf?+L<q
z@W;3oEjNh_K^6*u8W`yq0;fs`n9@%n6Vh6<!~V4M;BtZ1PZ4b61n3{a%+{-BQbxgC
z1Fx&F@MViP4R%k&zHb<X=B$jOk`Sf*zMW#H>Xcw4X#FZ}gAQ>$^6?#|7Rj-QW2PeW
zT_((T^4L20J9oO8^wbzTp~@<eS)e1AN8X~QdkuSAf)pu5In;9%9fFN>bjOB&f?hJq
z!CV*H>hUX<xMYA&ooDp0JVm>s<{%*jPY=<xkY0jzZ3t8=gkZWOWIF6_t;I#dNXqc=
zGv?#j*IM@%<kvi9hsQJ2OjwD`37+_Xz$Hw3$6YU~yuA1{Xv;FXo>4sRsGo!0wqpq3
zbB`i5Rbk1<;%OmbS?TlXeJWc@`!ndc2f<Iuv$`G?!?T1U=%fY3FOd=k-p|W%c3-nt
zMIb4yxPNd(`)L@Y$4te`MKbTz0^^CsGQqHB4gD&`J2UCJ5LQ0lpEGglOtXhe8`$UK
z>H=Sc`D+>9YCm$e1B#)4a5{qW^Z;n)4s%Em4>?cn0(B_QNDHIjJq?)Yxvyfuo{a&2
zBA)Jk^SW^tCfiK<_zFSf3F?s<WGWS7oZQt(S!X{IC>ljHP3;3!5zPr0@!@?PTi*>|
zaASD0X^u$X6}XZ$Q_~cGZ&=)p^Z)uX=bFD~@7M$JL4>36B*no(Fmmtj2F?=i-H*On
znA<eC)}do6CFWLme?h-OF7Sc^!Z=#9K&#?j1?ZTZbQy;neB|y?^joCboUX5r@Q~-f
zz)NIyBsBu*hDv`U%Vu^*RE<Az(FS7M$r4%0p6YkK3KGAVD7-sElr9GxS!BtGoRt1r
z)9o+w{q@!MgAxjuA(%;1{)VOFE;c@Pv$b+EUff4uXUqA(?1p^v!thfgF`n!7{s&GY
zu4*0j?l@L$(Svf^s@IAWCufO;IoxM$W>sj~N!4?COwR*vQ7KZH1uBKwn}r;*?)B71
zXF7>M5{zQMJ%p>7NmF_)iB^_--UY-lh!)G`=(^Fi8tFAOwHPlt6OKr$QIq3NhoEIn
zAk%R?USFMAc*TZ$A9*OF^Q?=%@CUvun-`@2AJ;-{H;KNBuFG=o`m^ljP#=Aurov&j
zi5>2|IzwYX91Cc9zm|d!9wr!iRXc;lkltaom0$ZJHpmgquET|h+F>CR134J)q8a98
z7_<gx_Po2lp|T5dgd}0mVb}nw_<-!fIT#4=JC!n`-SB4IZ+$4rXaas!@(iC;vz0KG
zM$=3bD$gd(sHL-~ijUpn=>Cb2vzFCj@EROzgN^X}C5CEKekPolt~oW4XY73XXnUf$
z;fk+>09Y|~T=bA5|8xPS<?BKmJrjQD1@A|?`xdTwee>3z0O?V~V`%4BpCA^I&Y3sf
zAU<26X8WlgUt-0ST@1hOr@9A?wzd>5iIs5s?U?WR8)5Vaw<C^%z0f5%kKtQXK<jHK
zC54<uIAdqG=3W*dg)k@_{$6=3BH54>RGFC`LJ%tNKAcWN`Eh}ix^13*VJ$|)sXs|<
zsvD9C;>j?hO#``kt!Fu{+a{WJ<6K1v2KN)f#*vTJV2<3vOX^(Hpeu_z!_Hc!(AD6O
zH9Xh|xO31b+$*zM!Btpt5g@Yxmo%$Pw*8;(hO{4DDKq0wB~$tguxdFAwOO4V69h1K
z7bqS#MN}FnAG>tl{R9H%o1S8I0VgFXu_VH7d6$-Pvb(8$#S~7z$sN;nMgW;R5nJI|
zAb13>eLHE2=t*eKo=NN)*o$*()BzQK5NZ3d?faj!^8Fp#DthcXb`?0ak%4IiZq(!K
ztjT+xnSwQ8MG_#8XI);r@J?DjOE3faB&wULYcQxJV{vSq)5r?0W}><n&L~Dq{SLRQ
z8^YnhZ=(l}9ZI;R^1SO$KJTPO(5zYZza*=S<^or+@)e=wPf|=oJQZAs>+B_>@j5Gg
zk0(oXbM64_(<D$HeBvFy*y?1Oe<osT9`dd5;PdnI2M$|ECa>bQ6^m=Y5wcZpJeZnF
z?iT%kLsXJMmh<Nua2qP_!cF2D+U$G^{guOnwDVmssc2uo#)Qi2J!5{*RD_VO?WJKq
z<K@lRO7;|HU1lGCn7);_hT=%ldjdwtHHg=yQpxvrN~>J{t;|XD{QYi<bvULu77xwP
zu+-qQ_C=O57MVHR#UZD!3iDO;_SS4B;BwU(G&Sss<u_DHUbz`iacsxX(UGk5RSjx7
zmVSRGIjTLr85ls!_n9HWgl&)Hfi>oRUQDm+_GA2INHDR-QQh_M<24H*%K=ZlM(Ao;
zS)aoM=sR)VNXiU>O-+VNMpumyd!^{SaC=tk4RYSUS3$b6y*zwfvK`RazEi;z8{QGk
zS(Ro#ldrD{aAGz=AmtkUF*zTY2+%I1(Frf!S%WAqxRvQHzvye4%Cvgu{rVa19)|!_
za2iG{ROduq7Z}U>fR?5-zxTynE$v+S9UVd;8)>wKbJ^p3(s6G+O$O_6>E|!#Ucg<V
zAzuQeEVkSf6Y{|)^O`5_6=mS&4y~T-I*|Nb&K2NPEPiFMlPy3Z#)KVRa4eM1PrHJX
z9c#_}l7`i&;|)fQ#g0C$?yQ?!+oIKa#7Y>Gu{&g~s84XAcVU^`R{jD^KU^H=u-OU=
z6urTgYh+fx0G&0VG#fcYDKoB_FUwy-!<hzz(Gi2J5GbLZXs|RY58bJ9ZuCwKPux}e
z6+HAtJ*<oGtFVdk^j#zntTV{``khJ)X@9(^rHIfsOQ@%aD9V}SXAyCVt~{;;x22ap
z+G_ht;=Pu|D!s0Tzo{?k{u6m9+$u|OJoLDkQkX54k)j7~A<NdG=jKU*=KHU>C+toR
zX%~Izsr(fl#Lw-SZmA(C!h0#5XZD;9`rsBsMsya9jLKiP3~FajEkgG3hofQePRG%7
z5t#ToC_WN8T8whoqNCI4Xn$TFFgwXDOa)LoldR1QRQm~z$G!p)GmmFO9@Yz!Ks;rb
zN2f`ph}WI^u*)cA_s)1oi*9BVhCA(0*?AXRNDs_9`>#p)q-^tvcw?dgl&Ul8+rFF7
z>JsQNUJ*@+RNs*?ZAkgiV~>uR>Z&L!LhFPq86kf>6Q&Olcr4=}Q`z*j8c58SehVZ&
zlzDlNaMiEl;~!CpxUNFp6!i4H-CTq=x+;Q6Z<2i9GSCk-_=u94Ckt)}*tD6j;okuO
zSoH@ND|<~iRdOt<703glME;CD>ZQAzt;Q7xwZ*CtF<0REQeFD`{`sbYP;4Z>|LA^h
zs;?-B8X9MrJKd2Q{20%ZQUjkt1qr9jZ8BRUXBBPb2|-Cc*zjxwF|ky}l^t@A_HErB
zo#)uGl+yTGoxAribmMIzHxwmQ6^nXYL8M$w?S82m_(`!gbNmzA<mP2@|4mdyhr|1i
zP3<&A#kdmPqjp;t{37IO6S_+w2HZkPlFmvp_;2yb5ZW(YXu+coneSAKY*Pec^c2J>
z0lPVBS8jI`$9<N5=#ysXf$Eyn@k9YU6d*)qvx|;j+II(21LJ5XuR{&p?yZI%Z8H3o
zCTxsqB;El1>w`Ey<f*1WX6z><nluXb7UPUz)Pw!}9`E#ZUt@tH++)l5Ths)7(5$R-
zL7!bzD;C?e&*NJX)G2-ZEp#anCN$q^#)n={nITJ|!_47{Y}7RnP<4GJLEnCLku_Do
zt9>8i4V?R%uxp@MT0to^8{aL7Jm5=$Eqps)H2j)T#y-ZVu{CkTpcJ!lwOTpCg$?fP
zn|W?nhv^N2?GpC+Yqc`|EbjLBlg76TzQzJY`Rk(=7C4r@Y_5H0TBT^S`3=o1O%Od#
z+d1V6Uk&v7qjbbHf83CO1*#_1E137*%|DUiV#@_VDr9M0Cx-=>KL?nb4(4rN6~4)G
z3#S&Y%4QbfujJgxTy<^bEtG&j9E~RB&}p-`06M!aW^QaJ#fgps3d6I^tH>r&qCrpv
zStkCZji~1xKXwREx!e!Hv+MYfm*gP0hY6Q=VJxCcuD?P0Lz7+R&88Lf1(B-n8$3K-
z>3v(s=$diruIi@M|Fu_pE3ckJ+Zb-b7^C{Iz{NG<tB@iy8oqSlSxB?$+@O7}Uyvsz
z<&1$;am=PeBFiXB+|(?;BDvNWBzRpq_xY9s6J7O|&QSCC#%fcrw6=P1?+AJzYepHm
zqS$#4gB=F;TzXUS)rMEL8z8d_**))cU2{liN_?Pca!4@=NX4GszhBumjsRlA1Q@2=
zkI;?D9(@r@+jwdmH3SD8J><paCI|5;KnT*i{(Q{e#7Hk$w`7x`lo4SZ1xroT?HRgC
z80X%kyVlA?Gm@EuC*{_YnvXCe-o_5ZD0vc}<j!E=I?zV6ZQ5Kd9v-Vu6*icLB)EHW
z;e{f7sCap_OZ4pQ`ec;72hu{QV{$n0IQ`)@p)}g4a!pK(AYNcG)f}3IsvwXeQ7Lky
zD^sTz_8`V^>%i^bO;tdmQIc7OT5Zt*Dp**qbx+wyF0e*wNMN=P%`e~5OvCT_&RBAh
zB}-<Qdg;-{X_AxgnfOC;Ft1XgYA?*|THDCR2cUd#OW(eh3nFj~YZ(L~`9nge=`8<%
z<-a?$ir@csCuz3|7h69Y)!>RSiMqp$oWW8)MKNu#=lU%Welq`CdH|(J9&w{VGUJf<
zpl_R^)mN-{Fd~!b9FNsi9=A_ZFS_2by*H7E&RZ`AVlA_|%m?PLmLyfGZPF^Pj*V{N
z$1a;_Wuxf$*|Lg)Df_6EFO!_lpt;P>_f+kt1&&~p7i@b>U=;c(=Vsdr2y&?~$D?y+
zTIKea5Jt`s<Y-U^LC%lOh|z3&IMnEboD{C&Ka=Xm3}h*Fb5!>VT$W4DW5s>Y2E-QZ
zslOh_{K+%S?nIgBaQmaCI4__~78Q=eRF!D5Nf&iZzs5i%)CV<Oe}VCE`UHVZj&4|!
zvMBZ`(Srwex1>2BRkkH9B6q3;vaC;^(jJb_^>;F~uXI}p3z4WPhEb|OhFxCSZA~1F
zVzr0CtvfX&g?k+xsFzkLj!%C7F{f1&U)zKtlR00mIAi8!RZahv;S}ZZDciad`ka9Z
zJ!e~@_XYb!?sU9N*z7AzrJ&kB8$>ZU%vV?FfuvIFG)82z6RN0V%{3SM30tNI)<=U0
zz;U`AW?g207!4X`z+45bg@XCyikt$TR8AOxytHvYN>aaEKS_37OE^%?F)hJtX=Sp+
zRi#fK78l77JY0MRu+O%jV7IMVk5F$XBYqY!OW<l*Yb?0p^W%UD-Mpc67GNh}@xxEP
z-W&Ay<16Uhr^b12Mti))wH-K2Ir8(qlPPP|<nBM-EbRsSB#{G|x`nPLqdB*Q`c}kO
z714pD0vjSdKY3Uuytqq(+I`K%X`=Px8Izp#HcEDpEuxVt+TE7oJD%ihIA=*5PNU?S
z`;a|nV2cjP-X)j$gA76$m#27T^Kos4!9(-C@04o7wW5zCRi%G<<{`VpLIt$HVLLf*
zD+*TZ<8LL`Oy0d1ybhN+wbey%&^ws8l5#a^FWxpws@#+9TT-wfAyF|ii-@R-4bjOY
z<3y*8ka&5NRLtJ!$&Zdu)O4!y4gzj~OZ29T>*Zqn&ky!kjEEtmc-pof0N5`y49Qg&
z0-7h+otKpovs5O0Rbt~FJ1T$e0+su=xI~nj#*;rW**&8MA-?J%@8czrQ%d?zd3%E2
z{BTOg4S^lM3<?X}LcMH6$PD9AQ(N!rZF=@=v7R41WqqV<fDdw@q9nr`h+nU!?c7|M
zDTdS4H;on-DGoCZ6$D^%-Cda0MN;EQk^}0E1o4@d<&_OVjO4lp8EN%yaf{PE)C!kP
zKmsc^rX&IU^q=-DF&Fq$uCc>H_xUC?%lvK?ovR(eov~EUo=Tib6L{kZnU@7(l&YI|
zEKJZl+7l7<35G=9@8Ep$<oJruroT7-i6tc1_I2o1&cZwP%N8`*+Jr`S3|n&P1ZB|U
zu9RWI^K?L;B*WJhjy<5eyN6tPC)e9m)a;i9a45e^6|^Qs4oVx)G+~!N-M5A?bpz#2
zqm;m9PL3sImeVmb`*_?g%rvc6Bnj;$2}w)|I}NA`)xJD)HQvAObZ+_9FU)KF2DW8N
zQ-c^L2`I<ppNa7eq^!ZSF)X>qq+e~3!G~ADFd|A?phs+$x0UaWFWpRRl`kFOTuPfs
z8o_8a3G*6yf|J6jvL;hWl6I3JKUMl(aE88G!$mU1ebt^exCp^4KP5Z*3i1Pr&CTFj
zu+MyX^zuehR*BKc9T@4zm*7Ea(I)w#JI1gfr!~c$*J%7=;H(!0&#Q<f2$R6ycJLL-
zW|DK}vjvWuv~xppv=`MuUxs=Oh96}O@yL=-Wt10_hUB6G46<KF>}YnTeaOvlw%!z&
z0vVQ+ImgilhX$Z~L%O<C`@6uz@VP20r4{yVWM;Bb<d^MN<QAN;4olW^FIWcd>^wMW
zw9Y$ipyUE-WlNW&r5ljaLtvGQzmltEco0$#;KZ~*!lhN21K1%M(M~;KyRLEbFGf{x
z9IPBJ;vVw(Lun77Aylg|p*sx@@!K@5g}nrqPDvev9N&=g{rcH`yv`$^=T6L05`!2M
zZ<j9iU{nGe{BQY<D|a^t$-3Lf?xx^g#p7|mmtHDJ;dS9>WR4sxT%D(=Qp!p?V7j_4
zVHw)?GtLbUx{{$jGGL*0qwtV<ERHR>@g^tMIXfYz&E|9#xv7`DdS?NmZogbG!?~O>
zaqrOvhqh>n^dKzAdV@5zGnk*Pd~<9q-)4Qc_#B$pnBDfL1F|uyL-`Z)WFBQ&5t%#1
zKU^f&9$>M*!szN0e3l@%nTMwh4u@PImASzz22@XOkSpYg(!O2!9R~C%$cRh){T~Dm
zKbEib-6`;@v=HXkyvruDNcDgy6J*rf0C{wQW1s8BU>#jzNoGr9TS@5S1lM4S4A!q^
zvnb(HczE0Bwljy4l_%W@p&b6Jsg!eMq<&GJj7s6-56xKEjx-GhGwp~NGCk^LSJEna
z=G`#yBnqO$EK4*}P01hPMa_o6k6D{CuqVPv+KPG$n^s8u<}67`s|^OZVwe(W3#7NX
zUOp1!#>>)qvBux1az&GZB|N{296qhL&0jhEG@AT6i7ChuOe=5H&QBmVh3}JV6W|w(
z6&I+15$}AJhnAS2{=p>quw0<OR_1(36Ub-Ou0PUf!ka)xd1Iv*ZSDur!E$R?>l3kg
zBTrOzf4G^Tt9M$Ug`Tv7i*2=vJLHW6t!T0|X1+f#_S)!{C4<35n!MaDKT9?~ImbEl
zOcGzjv_o%eUZ@N@b;nX`@Qyhhw~7yYEZC;u2?={O*H&9MGXZ{5_0xI>lRC<y=MT&6
zuMg=9((~2%`bQZ<PqpedOo`S@uaaS=E7vmRLnAbw>8Tb?Nd{lH?BI(64u?at<r)@T
ziWO^^8nr`}2VpIr@5eaKs#S`FP%^iaO^sqE!hLgj{pP)a@@LL;K7XWQ<sh67(_pNO
z^G9Zfn$HPeYw9NrB2n0ls#F#64}B8z=Y>wz{1Z1L2en1DUNPncz71W^At1PAYM~D7
zXp~}t&ZL)-2`j^1I~EaqHg(Nqe&vnDWK%Qoxl<bOyy<=pN$Zx3^tgJyq-E-zT*M)R
z3pKNtI}mIx6M510r?Ai1o!1e0g<0Tys0g%*+|zpceedgyvEcr}>3rX>YnD6$-J_0@
zx`D5CaEW#US7_;6UVnI=>DVf7bpN`xY!)ZhEAMNdxDQ;HVDK@sC!cVrG3d(0NMEI(
z;BEY*xsK@ah-cA(SSMxw9_lUI*IS2n&CGKT@a!2?7ue&txe*7D3o<d5kg%>9{wx41
z_3Uyn)2QkOhs^CWV>2&PDo+?u`&hvl&I~C?#a33;B$oh5?gBj=nWz;@pY@V0$yjcv
z2;iK7u}3GSi+uBrz7aRet$;6r*^TLFs8jv@c>>1aU#NtEksm)FbJV8tO9*j;j%ZHz
zn-Tgr{yV@$T@7~&Ql{Qk=VgpJ9zY=qC2U794B*>w)gRE<p?(neDcfLnRl0D&yA=#D
z{qngcGx#VdpLX>01RQ3KAehV1r}q@fkbk-en6wHu8Tc4z9bZ}W!=Xv!&iYY7(jyvW
z;qPO_aO_IVniRY3n<wkODa%fChG-!&N+FH4s}c6$kBx9D{(k|S7Gvq=Ggu_<_mky~
z%FSX1!0pspJg(2SvyCS;sWmL{rOqF!kih|>W*JCru=bn7wPZnM)WqJd=7+=f5?2tR
z%hy0GbX+5#lHb#JWl#9ZkDN*(GxF#Q5#YF|*o9V3Lsa$j)kte?TCeKof9@Eb#Dg_}
zaEjl8u(`$q1+sGktBV2CE=Zod&OWx=dU7xgIXV7qE-r2ir1SifH>@VRu;g4Y8W&5J
zVqq$u%?9hTJjWCPJnWgi`^EJ9@{wSH{!&s#Jq%|598xJ-kbCxy<N)&jc$g>d!()%&
z8WzCLa1zV&68G~c)=g6ejoxrchw5|*6H3U!CEt$U$?zeME)5wB^1RIe*S$*fL0)H%
zc?R*X2kN}7Uc@2a&AUGSL)Eu*>42P+LKaYKYHyC2*sccZD|k4_rN-4S5^e%(&#n$Z
zkSckA<KO8txptKq(j1{J9m<F-+w^#g6^4tDUQ_3c-BYY6L4YRUW81cE+qP}nw(s#h
zwr$(CZQJJFncdm!CX?(wRyv*ZLv^K6`Ro6nNO}nTy*`Wl^e+1ey|ER8-J5z%T9pfm
z2b=m|OB~0nWyH;@u;E~{=*ui@9`lavAA4uTInJh5&^fW_LaCh9uRz>#q%QwZKbC~G
zaJ6STxY3(-Z22P36AHs3o_Up+*QQ^*l#N<IZLAJdSnR>M8o>|Oog1v24ed7y7+Mnc
z9Pm`XH>fuv?BkSXaRM74ONYgcoP=Sr4^##qqWi`!LVU=ukxgJ{-TG&)PR>;>k4jB|
zo`0Z%W;Qlo`W?JkyOtvq&Qy^!xXp~W@i@wXbBK;>xGty}4=gm?FXnKykn3}{M45M3
zOyh!6ia;I?m>nvSO{DVtk%g6X#@ou*HXhRM{*3kpk<W&L;E23>Yk7o|PU4!HW{qSE
z6JxNupwWD>+eGrpa4R)jx5s6AccgHjPT{2U85b!GM=lq@*+JhkPzLzbqiK<$$mN#p
zppX8@0S>%+35xX3hxPX@>FVoN{tQ{QKWN?MZ|~LL<uCNwGO8&3ceUWlpv|DO#EMT!
z@`>f4@*0HNxH&j$|LRxw-4fMLQbg2CEkK-1hNuBe4ZkFwlGXqV#Wb2&L?}@+?)IcU
z_*beR$}`p0y@Xz$h-e~DhfI(U7U%{`C~p&~VB?KkyQK&OvZR+zkN%pC>Qv7s3xO7j
z?P14Bp-SuTKC%mjb75~mqE_`|0b%)IG$1-%_df49o}R*tW^X}&@CZszDWDpL%Gu-}
zQ9Q7L(;d1pi63eMn)$0e_8*}n5W^vAXVd+<X0mb1oQg+4H%?%pU^NbK9F*}cM%CJ5
zHx=bw6wO=z>MlM$>JTy5L|D8v?M24j$*2(Kx2mEPWR&H_?b;t>O9Y6L8yAqUZ;A~<
zLjuAlyrLGec~wp7<aU$%@=#(!wkbOlPGEX~K+13`U}go;FGh!VaU%Nrs?Tt!fGoWr
zmNDFyGjpx@KA-#{SDA~hOY#mQB{p@zt7ZT4jvr!=&AII4ycLh`NM`|Rb&PJ~g{z{|
z0kR02k$e3+Qz}sG*wts4fPk+ZPfwOCJnoMp@^t}D@lw5<Y@}<)e|LyPq8Q{PCS{=(
zsM9n?ZQooa(v)vq%0MlW2}H^5V!%jqE!x9<V0wor;Kz>I`9|s-7bv1ppc1>klaMQ9
z|Ae*VDYDGV7M{Q_@CF$YX>T9im6sd<Myo}kWYHbGuy0;y(EkxY4n$-{$h74#2v=wl
z3B3P#mNXh58q!M-AcAx?yK(qLv};#rg45{$1XL<<KcZgfJHe}+<s86SKq4!KkK=A_
zlxu~YL<!(Ktx_jTb!&p26EHjmpb-x8v)lrUEEP_%(F!10@}aX`&ZtKf9g4+dLn3-<
z|6}gh%V7M7;6dB3iFy}S$WNux@`(qu5(0iKZ#o)ULAYxqD+kdcfvaKiI3j;@vJxcm
zlrl3kyg{FagbycCmS3W3S{4>&>-_`jrWN(X1|v}0MBs|if3b1$;l>zdJP&xq$Vh70
z>ARWN!x9l}=rf&8!z?fhfQCxcd@=&tm<5QFH)&(SWus)JyCbdy!p`Wbo7br#eR7X!
zKWit>)0?1W@Y+Y-{~2)a{gYVl{!7yr&SN%HY2V$(e-*LyH?7|?Vwtk0>aKs-^ze~>
zv)hL(V6<VyK!ixQa$n?5r1nv(LMZJCQEYc}nMAo=Yw_nA9R`wal$-zA8)jPRLrynH
z@SY3;H*+#=Jx53owuIaIC0emOUVUnZO!sP<9>X=&8+OTqnRgRmc!qj<`d-vSwyg1o
z2u5y4cpdYNkQZysbGR-ZN<vyi*)~Yf1agYvj~=!sxMnkpVQf5y^6!CXA(=Br8l)fA
zf@lNc(fdz-!o$R{TFwYW+93ap!BH6Q6Ma^NAfY22C0b&>0;yqqXVJ%#(Z}pHA=}hM
zT>QTh!OlpUv&^mF5_;3wxtr1}0w3HW>WBkmPLs_;&>XC@&(X}QCT$$J!RnFq6JV8Z
z4gIcilZ1kJyk(!;FeP~1!&{FxQ!u|Z|260wa4E_jbxPj#Qlnt)ybe){_ZGDn2GW(M
ztvea4VDw2mnCrjFd;>{pGL4@+&(hY=>?~>O$^3Ri-yzpE`eL_c{>OB8=Sb2OyQEjq
zmCd#Ac&z-W^<huHs&;>ECcI@sdEE-`ym%p%cx?VkIU(R(Q<8(TfFq1jplPT-RspT>
z=rxZDuVX2<SY@MZn{3mH<NxekE6Cyq2{IgS2OW_tG6>&aH*AFwWOgazr>Dm&17<;G
zjN8RZ3|Bb#-Xd`LEYe*6+uhHIV5Q;xWZ$$IDzNtIwnOy^VUdi44lg)gyZa=m;5}*H
zvB1<LM-GCAq4a)k;atCh787w0t10UKZk|+3(HZqQ1g}<-l9r07`QwQdiZTzLqARrA
z#vs92BKI3BO4up{Z>V3u2Gw_tKSHh=iyBidGbeIU-V&9sU}Yzv(=6>$0}b;w<J{Ac
z@mhndeG8d-sAwH1wiSn!<vB~^{w8>x%q1)0mSbJ2wNx!l9_@)v;E!W&8mhAuspW!A
z+s)l&LpNfGvvP5i_DB(JNB@*k%-8@~M#Tm<qKM%!d5dH)6A7-dIXzcegO3*!C7cJ~
zP35mJZb<!79rBc(UZ+ds<_Pb>g!%T$8Q&z>KvfW`-4zbdkbq{mAIW(FPogWKItvL?
zO)@r4b}UIT$76m24(;bWO}JL$N`oJIa--?l_1ft2Q45bk^Xt@-P$=B5%SlLdc=mUV
z_*A;$`N9%FST<eL&e#053>#M{ZQbEXqO)|BxY=NxyCT|@hejS&MgkL{VElmbR0!N9
zcRds2KS&+G-8%;JGPVUZB+VRq4zUy@Ku00T_C4O;aGId@sDP$B2FbAWc`ekt$e2zB
zK0^>{7s}sIM1!iw`_juG1m!|z343lSe1tU+DP)TK%NfNFCT@=ib~=y}ZPOh&R=~#R
zRnC19i`5U&5f%QRf|Py8g(XJTsju4>@wp;#TFF3H(GuJ@%i&bIVJy-7kVlt<Jxfh2
zFXDIoVQ+o?t3Em#s}Mt6bNIT>(7mlsl9Y8B$?5%*0hecKu|v@3q$u>5zKsmvHO*u`
zZIY-q;I$vN-<xQk%<)dAw?Rv@3zp4VXVcsTtdIvP?#xv&Cy)X4cy$*j%Pa;f-cYm1
z_LTyZvU<AuHgRu^(s{F$bw_G_py474^UE823hMW01{MVd2<%0UC2dxq>yxxj8fevV
zE%H>Yw7P(U^b~~@9j++i-?pL3XDaaXI8YnF@<2d#NZ&q>&rpsrsEX?^&U%sycX3~Z
z;f}QH4SHQ&&wZV{a6ktIlX_EKvjdZ&4m*l1xpc-$YPfD$_>-mhKp`OaEk{STpxC~!
z*N6<xK+sxg;JRw_fzTGf3QAa~p`{5jI-c_j6%X0JUn}gjDdt{Gu3|uPgo7%YI$Fvb
zG_|T{9$(@ppMt{&T%#2g2XMZ@?~vz-&Bq~@!4+s>Frlfolq&p}Iex@ji(kD$NI!5g
z^rdhRD9&De{N(2gEt&A?bbmJdUDl=T`nk*-e6q_Wo#wLiVaXz9|JcN`D!=SpN4{1W
zu#aqfe1V)a+Gjr|QVxK)V?+mZi<rM#D>LE%iIc+GP+|hYL9rg4{^gT(1a21GQw{dP
zRb1*&9XL4=NW&7;FMXWYR#_>1g-@g+-YJr2x~8V{p_Z|Pyw(|?#$uIpd4rT}-B1B9
zWAkIUdk$Mj%>*9B0ALNt(T-}^l4#ysjQ@E7uCPWq_-fV9X9(ZnD~y6hom6F?kW7KV
zi&b<=^B1J$=@<|k9`sJFP&%SWjl`FunQ=GKeMmSKi5zECivC36b=Nmy?yNVuCGk0Y
zT5Jz9;W0VBWB&{MV@-w9)!&Rn;j4fez0?6kl9|*m(~yKpjVvlyRcE#&9xX`B5bb^0
z0%xbUQiks;iB74A3!4Xrh+D`8{aYGD;wdvg5V71LiAqR89JHtfEQ}Upo)bo2RVy&h
zMv!N8?GS`+g(Vmw8`}Lkcx9o*9Mj;?U4fuv3Q>K<Vts`hy4oA)X<tol8H$33s55sT
z+9NENIYp7NO<i}Rt@f|Rwh0fxsaPFfVBo7d#1kXk4_RuiWx%pomQ)vZrE$+QOCGlk
z!+LEqgKCnYappBk^>`)au!8+GLJyVzRzV7hRb=>~00h1=XHV?#_6W5$t9XsBb#TpO
zCtD~5+SSmp98}>I{1U69aJAX&<U3`>o-7hL2JCv+pE-tL^o0I}j~5wVbg4(EV&@@m
zbcu)P9wpoOztV6?!BKG+ju1B%nm(JOPoI0sbty0733iy%^&k(NxWWuX8x~evS}^7j
z4QHBd1pru`S{?A?xy`KYL2^^`;3|FSQf*AK;f%Kt)v+{7tfmFe;MTR~y&Th9%;WKP
zoe=WQ6amzW2!KCf@aMxDJVhpaHnUGx7p5e?ZU~xwHkp2RjwyT|%ifFL`X>4-{wV4D
zzT8hC@9q7?n;}O&rAWX*Y~V5^<lA`MYf5?O$jE8M7>VVqUV-TK#!`UVEo*Pqd*X2b
z%mp~Y7e&=#5G<~AL<JSCc`x%?DG{?n)5cGNRr&SKas&rm%aalYJ5yFAm&gb^nYwA1
z61uIg&AxVcO^?i0Y;Ic2E{gM$*19K!ScnJ$Tam%|0I?<M%bF$_7pFE$;-WAc&z-qj
zvt?p5YOBu{IUN2{xxwxSoe5R<&)Z<>+yl#qO*Eq<OHWi*ZV!&qgXztV+~Twme;t-w
zo&=zuqSPuE1=+-jT}ObxL{!0=oY_@BE$^@xiTdM5jm~$9ITV=8dl3mkWw2!#PvIBb
z_I%`8E(7ad=It2)Pzl&XQS57uxs!Mf6+8tTat0&1#PJE`rn18uc78iETksQ8jj*e+
z@2(OQ7Es^LHuF=}`QeFG4TuG0r<KhoO>(if;i$W_4|F2VK_(#gp2^}}(TEZ(*C`Kb
zu8`gXghG<%OAI8F`_ej)j7`x-Fkw;xRn%3qzi9+8$--QTL9q4HBH}BU?%y>!%e}qc
z1><q!BWuhw<8QvSedLjA>)YaIKXX0$a!-i{3VvIF#1aEkTMG<X<*(`ByRQhBB{c9M
z7B9BZG~EgAKPuvt?cH(s3uao$_{S8#{|0e?%H$p=P{qPxSiu4(4hOwJ^rVT)nJw*-
z!?t(X`|3>agwL+qha3=kvUmY`-x#n6f&UQldE63c>08$G{{1|8MC`@xJ?i6|Y_%W)
zU5C?l43vU_zw~$msrx+Hs@_+1IPgTHABAnMjD-K3=f*OHI!Ot~%moTuOIM3dFIXN&
zv#ciUpPI8q(L9w}K?SdO`XCcFG|Z?qb!A8!1X%une!;5u2=y~DL*O2Qc*;-=Hfuaa
z(#Es7THseDDRXNk|HHSxjO_2>n^;39SAUWV>1(BtpW+dKf;##;upENEV)cG8u6ann
zd#fi6+y$h5WF7q2@}RZKOM4+8<*`e&bxw2Aqzh8fSSO}@E8N!{j;a46{5#peFp_D*
z`;<uQNN5AvDLdbSpk*f?AC>UX%z6K8lPHx0%{YMe8WCb;<%21_D~WMxb7$<V{0vvQ
zp5HG8;*7%m>rRVl-04Cxj=PtmP~^RO!x(a=o&eZ^CQF6S`WKr5N7cvJn8;L5^5#>_
zL-6|LnTz$RN-2STQwc;#v*^(V26VLFl=9`<@77oMF7sodd!~jgv~N2%P4;Xm1F4A&
z=RgV|0|G(tkH8k}w1!jgKRlU0K*OJyaz_(%TR<NRqal(sV6WFS;k<HqD<-cQLP>Bh
zX!uO-(I$njwqiUN+u#Q$^WsXY@wnW|r6${~LX5phTaL@5J>^HA4{y^}9nj@6PwAtW
z5d|CyN1k<9N)PNxTCn2?eAtdb`#ikfkv2Ki!zkOBmt~~^s?)xTLZspf_@V~^3nmAi
z-Bw>6+*XZ;lSPkK0Fe<|crx;fzoYaRW+!n%Bh9v}g_l%Rr5EG4F5F+>A(z29#_mfR
zfOulF=vjojDfXBJ%404b+ytp%V~+!OUTP4eSNfj(FB!qmZWOAIP&_bPItnajuzVVu
zdO0gf6Ejz5zC!tddPW&@yb)yZ@#g%*BQ)OBX=8C1*kVvkqG@j=-qnl6;=i@#{`jI)
zLlsx`k6)qOP5$zd>))Vn0qGFk>O&&&P0IA<xyc`M+F=3Bgm5de%P1AzKWE{|?rKJ*
z-?iATleeZkTU)H`MUIGPqbF^G$ssiv>?l-x8M*pkW2=tclurDAdCM#IhK0DdsYMi1
zjG+4YS>B8OUU(IT^<+=tDoHenjb>@u6y~~-g(K6&wET_%C!qsp-e6ue34D|^ggCRs
z;DTt)IS*@uBdDCu7}I1~PZpcg&;K1i+YY)3(lxYXuCJuBxoZjDxO=f1h;Cc22J-Hj
z@*`f)f{rF4ZpQ$2f6-gJA?<9U(Iv(1<agm7^6L7-qc~{N<^gPDH!z(#46Sd#uhho=
zR){mKUE+G2H5yk34rwWub6tko6+&`-R|Yon`<Jm`dFWKF9-=0?SGv2gX+l5ZJ%ebW
zqEg0c*|4XkN+nbSB#Kx;YZchvO&WJ5ndqn6DhS*EfWVw!fpI5K!Kzqoo545%c>}Rr
zf_Xd!IhCE+M8DuwlkE<ajnE8{xov0y5U@sUH~@IGlBpb=tc^S)4(C!C?3$Et)SK$K
zXs(~}Et^xn3x+i-81{T=SMhbKH@-|$F^5jKd68cp@Iu*lz{#>dUZ&1)x$%Oj;H$aC
z{H`(Uf;1>1fqTU-@Tf~}<-CiZu$wf;22-~bep5FJD8?{Ix*5W?DQ*Su)0!q*@x6E+
zCA3H5u0~x>8~1QRLJtQU(h(%K)C=^jF?fZO0|X<if3X4BI(K2YQ*=@s8917SX9u?W
zYB?gwYvlzngo$EUUpgcqvW%<?0Zb>nYq%IRn|UDp1|t_bJ_^y@b3@N<a9IFQV_-g1
z=<u0!4i|)5sg^wIV$O+$%eixR0}3yTuoSb!PA`3px!I5Gu`!(fNVx(0uWK^qAwG$)
z+=%JA&l^ocsLt!IQrrxP=^z7cUnD&X#0%M@lTQ&`v{-UXE15!cn{285zy^-~qEKRZ
z_9w6&h|PWSdk%RP9{ohm7J<ElAPla6&-PI&s(n61@sFt1A+0s@gt0`#Kshb->1iPx
zi{p9+Vg2Bg_=czil?1QIkaKp7-mUP8meJ{ZR+fit>s6UAxF5^4C>Iq9@@5!K5QL}*
zEj`a)8??BrCjC9M|F#gA>Ax)mW@dn*moc?7cd=k1VCQ7!{I3x+!T)|hz`@SO{-1Tg
z1pn_o;A$09jil=qG05|q`25ZUFv|19n+!|?5VHWx4q+jsc)NHoNI=PRIzcJR_(VEO
zjx+a{ozst*AJ>^5M(6I?j@6FU?gP)XvnH1|69ES6Iv|*+y@SIORNym9Yzs|J4vjzp
z239)FG++tLe}_sQ0Xc?DjN|PA)d&a}DmzesXuuc&f&iWj2z|5gd&sD$!zMsu{aHM^
z1TAh50GGh<*x=x&y)?bL04q7a0(eCG2-*|@@B4$WR{+jVAi{|7D8HuyxC|Hpb-=(j
z1aJlcStTh64Pit9s*1w!c@SYix_D7@PR(us0Dx8q1Q<}@jes@=bO7PKZ~z1_P(fdG
zuruF%q`=@BfO~QV1V@B0SAZbg-(ehpat7$)5I8?~umOF5Ep%!ML&ER~3;-ZdE)JmI
zXm<v!f!~NnM?l_p4nBQ&V@(DQAUJ10jzR!<4OZjSU&Y9V=Am!d*l?Z!fDD=7@(3_z
zADu<~YrxaMQFTEA`v4&S;^^F|tQY`6ec3owJorxh@to}fKVuG|!g(#fN_~(HfSX$F
z0XTgQZ~xwTy}kLWl<WH~+nE?2?|+nMxr$rmXB9wjs0)T=C&T|0{<kS4Y^4Q>8h`ZG
zP0D9u0KmW1cZ;uF<Y!>7o6H5P|JlK80Y_9Lz*y=>1_0VLc(CY0fTskk`fa6j{yh!*
z!pFGxH|hILxc~VL`+UXxj+6fS?Vj~7RdaB4n5%OF`Go}fI?M(v1N7r(3D~8F1$pZ=
zs=^_J%lW0dxXl6E>n-{r`{aH>Z-rTUF(lI|*edYXBy<3c0LLf-6$H2jU{iyG-~KRO
z4G!v~gF~QrGpz8(g8Da0j=!->tib^{g#BB8*Yy!FpguZZ!Uew-X9{bIsSE9<?r-&1
z`FphBwFBm!U4VdpxDx`oV4h|N?a4GX)4=;O|H!AI_mSQagvY6C01!~!O}*}#d<ze~
z);4ya4uBp2GVC9p93KPoeElB2*8W}Jwa^m?WSG0{o)^>*6ojvQksq|jIJ-D_VA$tx
z{msAg&-$mYV_X4(8Tn;~0S*i66&IBfRg5)whabywZNA5Q>vN;}F8Su7-vO?i<RbOZ
zXqkB`zh`v1sa(~}J71J8Uc;llJpZ^Ufv2>P-1*Tp(*tXZ<MjhMH!_TRA;fti%3Z_?
zCcX9-Z6HO}Bh>zUOx#J^J`P~BWXzlWWBP)fTL8dXy8w(v^yS+8`i?R*64Ac2NCqmt
za=E7<{}rWcbbx<&d2hrjaQi7ZnhYB;L~19zll~NWhu<~1aDGE{33odohxwO9dM~(&
zbz8%WpEJ4kD}`BWkBA*;wnorjOYy+TW+WW409oP?GUjQw2}H)Syf?&oQk@uj?|2v4
zY~J<wxJj@0c;`=*PDEw76!e&YE5AHa)K;c*R|MSoj4z0Ex&6^pT%0kFL-$xTQ(>vH
zh>vfs8jW+bf_Y8U-FP85dF9$XNkfF$9TU_}0Q*Fr=-#1sFR-RjmCa}`m86&BhT&L;
zl#tCyT}{LDwyKoIlpgOw`fy0r4Jekq@LXrq(m%m+K2qDY8#Nk$&vo6ue^<a=?eF1E
zAZVSb=em|D60z#uUBw)2xkqyint{@<qAl<qvyF{oSFSq_Z(@|a@_R9yQ<tl<&N_4m
zor5mOTNSQB0iQ!-91+g(EXkoSk+xvn#HnhwJK0fnmJJn$K;D8!`W~gC)heL#bYyJ8
zy^j*5`TD6_dZnexBdt1j=Omhy?mA!zwGJYg&@JhPPKQv91+&BxFIl?}RWfjdZH8RE
zhWppU8z75xvvK$}nig<?K}(8CLm`z(VK}XmwtMfM!m#6W=7|*|6iO&Hoc1h}kFbJ*
zd;spG&pU<LM5>4b1ufTKu1lnA`-}{{82eb4rIF3Kx(_yS#GDuPdO(o;9GZ@r{*8n=
z8yHY?l%xm`n0a*7qTOMO`PXV<B)5n5jkYS|CC@yNvBK=?6$~SPq)ne8ItuG*RxK)4
zb^Cm;7mDV2aXW|p_}#5FUBxbzkN)Sp_GTcfJ|?#g?ZQ64@P)%2)N}SssftlIl>lRM
zT>z$moV6fn(sPrns&`U_xWl?tc~GA4_{yM_GMqm1@KpWqQLpJBoT|M6Vyk<>K24Ya
zy_0B{G*nh$!kl&>R^u6LnMNl8Ydwan(Mi-JY2)+n9>*aZ6O;0byS8P^tA~2qCJ|Ms
ztX>;VekI0oH-%wzDXMSjW@ez-3M*STVG#Mh-k&%#qq8Zu8>d`s(fe)PTeaEIjNQuP
zS?pJ67j?tE@Vv(^3H<XxsIhm6A{=)Cb0MKh*)P56McL{_k2DH;=R@V-iFQeW+-%76
z@yIkqbIdN8im8Q%p;8Y?0X(b|-V7ULg$0a*DTCGRb@nCE+&0{4{yjHJd{z>lM?{pS
zC$CP;1jq<i4fd+P>tLTgB8!eP^?pLiG{UUk7LSr=-zQkgsE36;SA&ZdPCvAWqAwZ~
z?xaZxWW9C>tSd)x)=W;gnLS6%#R&>{u?hzE#-&}|1@>Qdq*V6_zwc81pX^SRPkQ?1
zv5)g|xySO}^FE&GKRWw}&jHucNZ2XgiyOkZpYBF-#SS_*lDakc{nlA^A)YuhCt-Gx
z>rj3eJ>e`PnNisv2k;lE(WF3U2C@^rRa4g<d|i~%-%LPef$#Cw?(H4U<x}?TozLyM
zO!LODD@nn%pnpomK<p;bl0u`ovEE8ug4E82O`SV!YN;wc-FA>r=r>^fA<8;>@4U+w
zoL2Vo|K`g44n+vvDx<tC9gGbvD|VH?N!89C=$*w~nzwxd-k1)(L#y+Ea5kp*MiH4p
zKb(K1^v;?|?Kq4Z^%7;JXq`7OF=-K{u>Cmo?U!mcMpebh`*B><Xsue*Ecbsr{e+-=
zs-Q&<PtTxI&LxsXP1E^bdU7Xj)&DAcbR$~=R!}ucbO&#{OSqcvRV*B}{t1{0J>t8s
z_G|3YS$w?<UV@gxwj`z2CYMfcA2k(Wrk+qYnTt_q8jRx#(SnPMgF4NRa7VABP*mi~
zXl)6(jm*5S9l3Id;G`=k0P5g8yl3EjU?L_45LkRmnYru4&eBc(n6&NDKOR_TiL35`
z&of7;XCv}eRF(ymZZ^i7^^0g8tjE5wX?mn-joBqi@+D<BP^6KMFa`SYP?^ZBYC^`h
zr+Acd_|i}gTYN_}_Y5JelP6PhtqLwKif#o3pl|Epfp)y}{xs&w@dY{b&}a9o?3R#R
zsV8L<a6V3m4f6Ls?wkBTw{9CW?e?O+b|z{f3`%r1VG-;19Gf>TTEk&oyWX5JJc(qJ
zaPBz=*0x|(<*JutMO3b$%4g5%8Eh*5d~4aM!9KOgw4T=bSQ4_R`KY;tsk6*cD!b%&
z^Y)_L0dmy(z!*J6l@8tCs$Hk?a+}Vx{gpA^G*9av)SdSyvi;-26gIWwxokOEX5Tzx
z;!4h>nWakP4QeWSf<SL_e~;Y3#_6uu3lY5c8eE$zppX8bem+Pp$&VI`hC$`BEyYbB
zCb`yXH$MIq3?h(H|3?=h0q8HiIiz%S@I3i}W1I~*-HQJC7GoXGP`;l0j3iT8srfvO
zsEMs>?Ahi}2iFTrtYSE;xo|y0cpMmG`ovwT<8n-4{kyV#=yqeAfZ}5l3m&W4_hNU_
zTTZ*X<iD!li|SGCy*ZUJMvZu(q7wU#dF6O&#Of$gcVDUcJ`>N&l!B!n>^@A<1P_?Y
z>UN4ojq!XA-O+MOjZ+8Ka85R@zWIj$scf|okFL`)bYXnZJCK&owi96+5akLVR%r12
z`RHSrn7Xp>zZHxI1|i}LB86{SW=;gpxJ~&Y^N&-kgPt>|9tmR`XwLKT#xk);;TXHa
zdYS`4fTVZehmDSdW}Y7xUG)y8*pV+I$gg(LM@oaGwH0WCTB1Shg@}9dpV=w375+Zo
zWIt%!h1*_=?FKY9=}4ZnM2qw$(Y@6BHB@9pSMPQAfptwfDBKOvdjkxP02^ehkB)`M
z`%|7nfT!%hj%VUWWvlHP$6M+IIIAqr-L5C(r331j59gczaMT;=gYduaycAtH&G?(e
z?c34akP);w;xwu?F$HenCmG{07Jac<feA+!0w6v63)4tO3=mUkMyU-_f4zgxZ}fbr
z{gj0S+WKe)Pu{(<9GCeJ-&s4Dn-iSss}iS%RpYe|2p(xKV>qv|?edgp&X2<BL!KES
zv2alq31-yuY_~36<2k3Y_`r}vk>UVk5hrue0}2|lQnkaeV@RMD?V<;u4kIVS9^`sD
zmyWGr?fAtfA8U5yo<Wq0-pa%&d|%c|10MNzxXAf@W!2o0M$Gc?1geN`CtTh4?bJ%i
z6@=W%|INg;xjDJR_Q9O2OfW@y1mpB;I+BmHwdaxZr4hSe$BDabFkJfu1XA{YDd;s^
zk;*wAi*g!>-bg>#-(YpEj0oi?DW0f&p`mn(OkshM^b9zue2T+ma)5^UKVBp6!WX=A
z#wauy2!ztP{cF@*gz@n$)+H`RsCL8I(WYW?b45rsZG4GH*WZ5a+nvIZ>=?dPG&Ahc
zbk=2q%GWw^b3Sd8|LQ3u3hj|bDUstI-{7QlgcUs_yG<Orz!u7mlREC0=d+L{6_iQs
z;iT5hj#u5b7xg~Mw57DoPc!2*+nNxrF0|6r(ej#L8>tQe3KM)7glh`#Rlcm|YZUCx
z92!YVbBX(RxA5^sCAU90s8cjD(nH}2ZFy&wSjOUV14hHvG89=Gkk7+A!&oYaeC^j^
zDIwx$1nnf8Lk^muA<Wfl4yBYXWw3MNB^tz*JR!@*;=8<p<mz!@;nVp?j$q~5Q8M(5
z@_PBC9bKOhk#77fm%^lJH;lP8TWmZTJ+3u#6Xw1a3$D~FJs~D&Sd%AbYxa|n-exZ@
zr}9thKGVX)=c;6vcKS>%qJ{npBlWgqiZ`ojBhvj5vE^bzDCQM!;u!@pF9h2a2HXNr
z_=DRm(hs^sU0q)ucgm-j-@{jp=1H9M3F2PcpXhM&BQembQzKVhjy-=Z_&i`QH?qH{
z{;jZaWZLs6)e3Q)i6eF}<83W0sT9BCiBRlQ`nSG<mpQ29m(K~|Ducd8=N%?@elAvH
z(3Zwsk4?8w7^`~0FXZRnUl(ClD&M6F(Fs?Gu41P=Nia)`0w|N?!7AC}ty&|>DptSd
zUt``wuJ^Yv-$RI#v9+MRn;!G4e5gAe$IkbAJ&YLpdVEvcax4`9o$REzEM<x8Xa*3p
zF=VHhW@FtZr|cW7)GnI0U|OQN&57r+Z;Adh2I8hdKi4v9OjApNujzpjTp_y`M*K2c
zLWbpCBx#Knq4bLE`<p5As4Lh7Zj5$HQ;NS(!xqy{Fcr=PI>F}?H<B-hb9dOHszRs>
zStE{oL>cEv4j4;UmIVtR@cmg+isvB>owpZ1+noHf_@GLe8*d}-e#7Hu-D`!KZ$5W+
z0%+~%qKX(?#+@~@+FPVZ$oyV~Lk+{kJmHL%>4DY_U<Lp7Q(P~Xx`8JFD)PwAzWj;l
zAZD4#T1$TQDI+wkr)XIbaW`{Ducjy0t0NJm*WCE5%I!rtyqm)XeHkAitGI_FCS1LE
z_8u$xndI<6dcm;fW0j}16%MBcBo(k(gvfS}R4W6heJTY2E_n*z{@nc9tVO5T5h-yt
zl!0HL_?@ba0MxCN2TJ7|#Tk}5r}`*5!^p6f#aD4aA|duddjjLD`xs?qDASsdfFtZx
zc5LTun<>UfYxSV9m3S<3YMI~CiLB?u1$*|G0>Xa9K5}opYqZxK;#3xv%zfv0Zl3b6
zYmuqaZ?fS08_UQL@J?jlhL*+#iq}2LIBi&!Ajg}Xp>PwFbhrx#mwBf-yKv;CiDaC`
z({wf+YHC5agk5;PvqUf#VTtI+N-wA9(>U0;Xu1J?nAbzJkv~V3i)mqL>Qr{rEkIYS
z-#oFVvZ(*vAiRF^g*Sq-({E8J)yZwUEzOgvomxz5Hj^t`U8ZJ8lEcH^IuW+O0iybr
z5UVB6^cjnK<?=paC+(v_9s_YLEo}3v846D8+?BS6O|nm1$@LWJ-=S8q*=Z%EieWCv
zGox|dWbSHFq38#@R8I{xL|;l=(;?9^Ys4ly?hC055w4y)8Uryfx2Q`iY;Rht_NIL9
z+Rnl}*XSz3p+J38A=jDMXt<Xuc|oiGhXGez`ltiC!z4&9Mk`(h;}OAPP9k-wbjKPQ
z^dD!R=Mjr%y=Hl~?iPZ^6!pc`&jZM7nj8;8NIJYZ;~?2Ghe<y#ATqFpp$8A4zy3qB
z9;p|XU>qh%hiMb>QjkS0TQWE7_#I7|Dwe%yJC(b#*K&$C9WqGKV1BIin2%E}PGwk)
zaA>sW)5*c+z+l6%$>pPL<m;!dexCj&s==WrO*_pAH;`lI#9T$+*6NO(6AZq|J92=a
zmHE}wv83#S?g)U#Ri25)h72-A973rIRx&c0OK}Nm?_BR9Iy|?ES+4p<#da0RU+an#
zgw?GPH@NLyJy8sb5AqKuNM35;N%IF!D240KZ9F$aET6|3z?sz3=3x3^;4gAzbb5EP
z+Jfe@NpGg$=EiH~?%A2ux@>x{I!DD8jM!y%1Mx$HJ-V}}`qbHkgvpxvK5X28&<Wtb
zSn|HP1%_Y`)&4nD+9Aa*`icH8+h+-<bDntdsFa7c{wV<lClseS8+4z43Tl+9XE^8R
z3x~W|0~1|ThTbz@Yo5VS{&mbFQqAj_CE^SBWHF<U-9*2O5K_ec6*JSw4L?;j>U^38
z+;Ljug5$}gzYv4)<s;|S+~<RZhQd10O9K}yD~`tP^`NQ{gLezBdaDfQY<)g3=j1Bx
z389G8p_<4w@ztBNy+Vwb>rDEVktl0tSy%7lkT8U|IP<R{Fk}198aXfdONMeqdSV5d
z2DQcaj=Jq#QA65Kkhr#Hucb;_SMln*1(Y4>JwDbbEKAtxQ#K6N)_m5xOZjVsA~+q&
zmXc_+X8W}wqW5~xmG*HurWjprBT}}z%^6Hzb4dG}&(kuaqU<fa;&*s)tYJU7(E3V*
zi?PH10X7wa%b9%Re*l7c8aIL$_eEnUu>uKEBAsy;aK(=mzJ-x985944kM`9zJWELz
z>rR#XDRS}KWaC1ZfWPkeuQzB1DEmv(=ojpv7Sy`?tYsSGs`?5p)y9+iRcLl4?tl|O
zsn$ES*mw6TuaEUX<F|b|F{;m*#4Y8pyBS?`A{*kG+cG%T9+4qd8^=d0HN`y^9Z|(^
z5lOAiont=K%CrL2ebakPK7OZA`imr>kp`rYj%X769Edr<%~m&5e$$(38<OH9Y5ay$
z8g%EKPskkeyeGl%nuI409j&>$OjOFd31`)EyG;%?Y6d?hKcy^PhWavcN?K4i-3-a6
z_Wn_<n$>vobyUCsxJ*w%g!H8<5X#pnHhyFDcoKf>T%dV0yP4`Z$8o1na`RrbAS25*
z`!&hSK*;{00nAlwes*8%rRc;gG+$v#HuP>|5;NQ}R6%zhq+oD+5pTV1q&j0<7S81L
z>1<XwxfzHK%=|Djs29veBq|A&j&feTd#-_{+lzr|y@lV&uAZB0p{{D<@-3<Gh`rY{
zs#l^!A3-dD2)-G(U=_4sG*2SOzF>?G-NuK>PYp+)Q&p#8=eukhM?pNXKn(Wlt%!!n
zb?@L*ftb>Pxk|PAN4mFHo+s}ejX!h*xzk>u!lflqu1xQTEusN%(af99-u6{lw;}R0
z4s$vE{u|6E3nJ9Z*vF>e^szzt{j)7mRQ(`PsEy0WHOCN5NUEp9m_tjSZmn~!kyBr7
zMUl;x0nfq~@6R8~tPb{KG80GezNc5(H4QVA!bJBo@YJ~JQIAJ(g#kv4n}yVYj5TM6
zcht(Ef6kKF%$|<IZUt^^dAc6oxe$L#&nro$II;hAs-sZ{&3tw0-djy#dXw~CFc9@g
zSHb?OohocZcA*q_4=*6!Be&vX1)z4jMKv}xtafcn!N1=6?~|<~`jF<ohnO2hiOlB}
zn3|qaX=TiZS*5u^-fFy3XFS#`65MtC-pkdC#XCN2g%5rk0@`6qoK@5-Gp-@c{Ui>a
zMyctz$*XQo@2n0IN3i4H>@><d0X5L`fNV8oT@tkH=?jW-p|(<VS2^UDv7?4WvCHzl
znv4UX#mgD#3RdBX#((&_P^2cvX3F4nUClMGr4IA#>MU~?WCSYc%IYU%(^e2RV=!RT
zk!ITQ-bIT%UT%Vi1A`g9jx^^Pz&=~ITOSyay_E@gvem=(i36F*Qu0SmOXNMD(Js>K
zsGw+r{s}M>=U`2HV5+)zZ6}!O;o7?HIxlB^YM>AHMel9CXIA{+e8u@?#DsErH%8iq
z(As4iiZokP;(`T#yk_<vt31LYV>tDZt2~L97H*+xxqXp2GW!?iKc{UN<)K$sUZ_F!
zAl`ZXYDsqJ;hE!5&#73*x`_<ImR7Vo!o2Ts?X%eLcu#Ow_ox6^Q9npvOiOF08a1Rs
zEP+k4aH1mYu1TR(+@tZXp(+*$-&#NOUc}Dm1!~_*{=S&@L`C_a9B3@ztO-~pYYr~t
z#vE7BPG9pd*$&e}F{F4alU43**rF^9$fHjE>a^5ItR)0;)UabPDNDs;EMsu4t<uIX
zxSY~a4e0)Dn%56TSv%dAUHP8+IzD49C1}KhDHr<amU?x=ek6Bn@XG5~8D1TKR9lRA
zv>og~XgS+4;OSWgXjmXINF&g0x2gD474P+ahr5g+a=owc(rIB*)c46ly|rbMrKu!M
z;$Z0LLB}88BFk-Pa&jOJS9eHGwqnP+?n*g}UGd>VzlM@6i@J{3gfk|=;C&^PFPx|!
zi#hb`ON_UJay@rf2Ud+8hRmk0*%;+qGZeIf#3({c`!VdUf6Q61mPg^(t?~v_Ks4yw
z9)QKF=sQ;-b})Q*9Eh>&8hoMtBI#+*Q#@#jutS0-tn;cQK3X>0zUY=1=Pq^RnOE12
zjDA?gDX-d>G$@hK`iG6M%%1!aa<yt7Xl`I#ZBv`I)zUJ|e+!dYbqkpNVj=(N`oTyp
z?{aJVvxGP9ElY~><I$;{{)tD4te2x|<8ahKhdMTSH2jd@&{QfmM&u(RC5SUnW3_a2
z9LB6H(;ZKa85R7|-?3`$gEw0cSxLTCI;nSa(L&`D*o4XRIZ{bP7&&0ecr#j9mwQv<
zOF47gc*`dl-T2<u!CVWHmA2sazEZ7O^1`W<!rfzpm%Gv#N1p<kYlZ_gi-7@Wn8q@X
z;fZxw7mHe|!?^5c%}ABK4wtv_WC%F#VCI@(9}LfG$}h^?<wLi(KWSMKnDkfH$jSR$
z=k-ULlaHKx#<kNMJGGe@cOfa<YiVK=#eF8d2mUQutBU<wa(fzkQuJ_bY+(AePd#zA
zExEV&PC6?%Rkd|WH`X6=9>hL`_Glo{46cX`(TreH8ua5Kbxy`zAN=ldUleL-pb*-Y
z5<o8ysHf8)W^*c0Fr(}wI>S@YM|&m|&Mbq$!8dz_V2@|<RL3bLL~k+fJXuUGv<39G
zh8Xp@jzM==DGV`wBYtnP?$k&AG@3$Eb8Y1VlRq(DDu8X`HNL3BJx@PHo?<EX1j-Yh
zwm7-h%$oC)l4IPtxhu~r>v=hjQafe-m3VxM%D*TLx!m1n$;5R?^zlY9S1;HWXF{A=
z9rZyd@e8sZ5k6isx{`2#44nJyjj6z)WPp4M1|2=w$wF=*B2SXdY9+M%Hs{$*q>+$R
zoV&CmkEW%z@+s}=y@r1NzPuM}{V?h7hvQKFDh~0kOx_vGa>yo0yu^BX5g*&pbqtFg
z(H&4(uHlRbM#Oi5hdU~c%^mWgy$9YI>2|1A@8flJhz^C&?SxI*Wg>5MMMGD3aEE5M
z%1GhiC4b<7`5g;8^mg;D735eO<IB<IE%#EoJN*vcr<pa?;Z#4yjL^IxIajrk8{2xp
zR!OF3!ioG=;nD7s)YMLX@!R8hF+|_PSkQa88c~@;YWI~syk~+{<}Ok7phNWbA;dMH
z?n5x&{0x7`Pj(vW_8hM#dh<?|<&W0Dn8allCR}U6U^(2yoNv1`&hE#eez})nQ>9QS
zrDAY!6fZ=pzW5OF8`FX1!fEdNF$+8P+6)LpYT5g_FXm1zpLN+aa+9TkHFv#_FQGzu
zgfD*-1Vv3-U$#2ojTop3@2*ry)ZuA{DfzJB)q`+!kNyQkvO7{-l@XH4XE_9Yz5?7r
zeaKN<E|aM%PlE;n>3qK1lmll=5NfxlgF2#ggU~J}x5euZB^GISkKBc^{gbrgCD|q7
ztO#7Yw4%=m=Hm30+s}ETY<`1~$DYEGAo(tcK7^v=O2cB#fscI3^+F^3BS;`L_`BQ$
zOS%^6L*-yQCjx)5N!mldU<aL<6|dJVCcS(BK@;l_pA@^sc%f?`xd6*=6R!r!;5-R%
zO`Lq_H5Lmdi<JfDhNwW9$aq_Fer&dH6_h$cnWmP9mx)vtJwa-X;vd#Tl?NfL<!A#`
zx%@Z?7W2=I6+IH{nTwBP7_6Zcq2_>}HX^G)%WIK8fa`B?AX9Y)Osb`SUJFs+6^f?C
zP;ULD1Uy+3=0`hLz)!qN1`~FTSKitnadMh{y+suqfaF$d4DBiYC=wNG`HL@BVbTw}
z!O6_f@gW`ws@bf`Mokzvad4acL~E+g2#j289g;0rQpbpFQ%E;9>}>-J142CppkOEZ
z$jkE1y(gwS5F9hFmA0(G0*oHv^QJGh_DN#Wh7r4J^A)cBJMZYw**Y^CJl7+oA;Dvf
zv~S&w<88Li8O|Gm2?q6YwTY@a=S%;j<A%IhT2=e1ct@Nw6mP3#es$cI%xx5rv>~0m
z9Ba~ZG8N|}VcC6L7;F}%HiJXA7$h2hq);M?9*mM}l;^v!mC6|*2fA@NoXB?v@BXx4
zy*4jSqk<0JVbf~n*D*v8X65^^gm3O~9LzA@wldMj?n~}c{HYv=>8_<LG07Kl#S@re
zr(a>F;9)yc-xRsA(AbTCu6pSajmUT?%mM0!C!4Q*daY%-#KKVUI|laJJeP8?mV6DR
z84JD4W8w@&C^j9UtSLG;h`Y<B5ec(i+3^*LwxYiIbG}1~1I9fm3*Cg0rgGH`SZtJV
zTcsC^0UP31LN1O;?Vgk9gwAbEyfip4<)<9j-^U2gKUcgs`eMdt*{mM$kQY&f<YE$U
zyxKJG>oAld@(T4ySbM>BDT&yNXl`s!)w5AAq?2S?I!8MyfCw@y{QIGsFP;p8XnnNZ
z6FZ(5JH%CgG2K6QZ#RT)ZTw`^Ih5+HLN)!$aD>9y`<&o{xFIc2m?BIl(Jw0l2FZL3
zo{#szftVS)gWP$$5X8pqzHBVt%Y0;<w>W9BC>)2T*DQm8G}Sx1WZrL$hv@MTsJy`)
zu>4du9q%sUk0q`!{FxF?L97EHZr=kenU(Q*a^-#Xw(erm9yD?Z@Rr%)Mas7m^|-v0
zcd_4Er*83GfAgGcuJJTn$J|6M0rZZ<TrG_YN)5VQePfI%f=R{(24S8s%5aMm{BMO5
zky#-X>VhyXF-XeXM87fa6{Euv>gJ)%FviLsP)UxnMg?(Z9mT+U@7=XpKg78$q%}Dr
zk>F+v$*FF{m@!g#1l@s)r<GU;krGSKHErHek?YWr|3+UB%4PX)b1N+WZEl5;?Z1*L
zOa%Y9Y6hnN8vdWb6(&{&W|se7$hyPW!%Vl1>>p&Ev*yOtVtW~ByD27)yD2x?5)->!
zKkwM%x65okqho%u>hox<zpAI&eA7{RqjfpEgF+%z0W)9*M<xIX?yU}nCT6A=&=FOV
zSYKUTpPn3y6C4$s&A3;a4jdaKw1jDKWp!*oZ*~QB1Zn{^4PYdo3_wjC5E>p19t%Wp
zv3GWCV`_K~mG|>UP*`02S3kvH2!KrQ_x!zcad=_?PvHG^7}w%*&jP&3{Tuq;zgrRY
z|G=!v3KCnIf=B|ilmu}JApaL;z4f1%wbTENSwjc>Pt3abKQU{!|BYF<0`kfR#{RaQ
z0mcfLb5o<I_TB<eqX&dG)^}g-;<a%BR9(s1ZcqIX-mTi*+|d(?n}_!zy7~M1_b1`#
z=;+F1SL5gc`ZrQVN`8UMt4kZN`a5-Me7geUO#O;waHe&9D=++$uNnNiYev*w=m6C7
zN-z1_WdZ=0snNx?jca?`e}g!=w7;bXr-vtY4tg=>fU^S8MrO=f1pa*rgMa<+`HcTE
z<ePp4yPHeX^M`xuV6Wou-{9m*=(&|F#MsbhdpYIR-germtNpYeOjJv6XK(=MNB_0y
zEu7!QQ#hBOR>7s;DVS+}q!BZ?mliz$<~&ZEzd^y5yGZ~Szud`~zq^rNzXAup<Uqgt
z39rAoZNIvOzsCr_zr3@*xLO0#D{B2~aBsI{ef~R)0GxgNsnOrRXtp-yrkBtDQNOnd
z8GXM5@BSePZmd6>P=?-rdYu@UnSR6PCqxG(kS$6~j;*X98Q5FCH%tAmFJ8<9%(bb#
z_w;#xZAk+HL(|`T-<N2rWo)m0EqO-$+X7z9ezQOIBY#Y1ib_Zd$Oz?Mdw0u!M<~CS
zcT}e*@9CR=6{EU>*MD`sf{BO@_8{*J4fc=zx!6vP-@Hy$Gyr&RZ2vgxk^kVVt!eH5
zCujYw&Bi?0JNz4W+N<{Q3oEv@v($f;YpLkU+QR<HZScEz7q_`Iz5ZTX^PBVDEBbr-
zv8xCU#2Z-RU`-8IvMW^0+`y!JS^p?jGUNjOq0uKkfHRq~6a@`zjq3%>MC9&S5xsLc
zfRSnL!N1v^GK<3xXKE2Q@=<3~?zI@84hrID!txzjrk#|n<09TK17;eLewtCww<_(#
zjlh*Pg(dE-JV6Ck{2!Wi_CL*f)!zbWd<z<waV!@n;Y)TA8=0X~AwzKP;?K+srLx=)
zk0NYLTrAA%j?fGK!QwGlpOO#r$csEL@%zn%>CxBRKXT^%VkY6uCUIz~SaUw-+!u9K
z!mR$#8m<Q$kUg@qx<ZQ{Yj?PWD8ATWS6-AQRlfVK)m&LdmbaX!3>x&VE%foFfsPe=
z5A;m1c=g$wyAE$l$!7MzYKI+KOH%1fG(*ylx$~bLjy;5{pSlvROQYmPAj6AXq!M~6
zie>ghiHK~OOb6=G;>RF1U3Q-5G317h#nQ)-V2|*dmabr%6$*OUZy&0HSBJ4$0qpph
zFUX%Nqa|sW))qH*B&d|}5o+08_(FU%{Z?FJ4J~=1QRZO6uQD##nmU`IHuN>qT_3)(
z2{)d4&|o2`FGc|toC7n_o@TSXp?8fW+UNUN?<%*$Uw_W^U>+0ul+KxlN``Y7W$}F!
zatxmYgb2cY1W`KMc86>UvP2=WC+u^tOSGG3Izx@*J}_8YrOy%?qg#gQd|n3Jd&yI8
zyBSSyg*l`foP)3=f!xbsx9X2o=kX{n)65Hmc?ln*lpr<oDj!p_u^1bF#)$RuX%k$`
zjlhSGTZ`+DGqapXiO+3aFfxvSRaUvRUZGFzoYQJV3FPzPcjH7wby^U~>MxDj7;eqJ
zpw5wyl0o@E>04^OtD8-@InD_?RgCuTEi#*y7_4m}Y=DmIclGMY1=?)8{f3~&UH4^F
z4wdv#aKq0dI1^$sXcfaFktMq+N;4HKwam-o!XGpneRe}uC)GWoS#)996<kk-%1$Jo
zZ3R1spE0?YLR;NRgu|@a+ranA34>t*HgOGGw%IXGF+Yf>AzPCkcco;jqIDAEd~Dv&
zzMSZMbL%Ghm?3WI_GnAsNW`Qt$8m8zgO>}yUkyoC(;B_#{v`3Zi`3&G(Lh@uh8Rjh
ze0-$`u9MbD2x~fY?faKvT=jGCK}-KMWaO#uS}u>Om+#F_G~l^&`NHSU{Stj`A5_bl
zu;gp!wb6=pHCO3k%i@nn@B;j1F%y(eLNJA<-{WP@GFq9=hZ*GkDcjFH%u}8n%sLPr
zK5JeVz$Yz7U;j#92@ep{E|mwr-d54~>5&dD;&OvS<L3>-M{x`CTy5n(I$(saPshSb
zA_E%FXqO*qQc92GGwR^f?dCObJpP9bvxl;jCVAlk+JbxnpTo&ZY4hARD+KbF%v=yi
zU8GzPB+R~!(QL|R3q(30YL3HSJKEuD7WzB7l5!HC=Pe}i(Hq6Xmx+Ye!!jgH7hxDq
z;H$V{<Vm}3?wesyF_U{HLNO!oy21I@SA9<S{kjPiKraaiap||+mFi&z!!Wk|CY^%9
ztK8S}Y;$5*IM(t-?*${5afr^9SmiU1NGDAS+?MfDov{-BNIfaiau>kcrBNl%P~6-U
zZQ3`O2F#4eUc3>Z@SR={+42m)t;Jeoxox}(1%C{cn+^Ger3PU1`Ylp01vt|zCjCbi
zatQ;alo_0Eb+%&xN201BpkW|gwXTKAX!`aE<#xx*#m)M&*NQQ}TYi1+5+l-vbLC`j
zg$$Yoo@$hdust8~g?iXkv`x4Q!6sNG2t(0*BS!p+9AWB>Mp4F7bLL~Z#lm7@#h<fo
z_u?2elLzC__kk?P3-5j$$?Vo_l!TL#<IWlD3ndJpqg4|zmW1-KOM?tXv}Rt6Fk!7O
z{uPE!YZkiKLuZRv&?mj61kH}o)VW9asMfIN;-K^$hltrqn|5XaO^55}V(xE(YhN9b
zu@+H3m$8Fh*~0@1ZE^np0aQS%zYq#nWWd1>*{r?8czgp<sId;avQ$vY^oH$v#_i63
z)R2HHdjgc{3K|pJJ~YAL$x7n1DY{Pj7={`hG+9v2)kS47fb<lw@k~c#Ace=V8&*Na
z+OZInXPp}H(7%{goW7T0%i6_hY`6ay8i8-Ns-~ykBofw?fYe4e@Wt-NDj@vulY01c
zDh<yTd4CXTknfQ=Rb-jqqgzrNxD4hcnd0KeHRo8LWI$A|IpLW(Nfi&BN)a3sIT}e_
zTBCdm@0FTHm;%6o;CW&0n~rL!#Ic}xJ3KzSSP=^eG;k1<gIt@cPy*HUgJz8T;5F_N
zz}|>aSBFA5t_Js>4Z}c5G*7}VPO}YZB)L>bORaYq@@$lrOu3Bn!s|>uf*pn#=L;TH
z6h*sALD!vHD_EWhl$#EkEda6}aBDg|IEb?6Y!533Wgk%%FEpC_0|3nT3&W0V@CHUa
z77(=NN?@U}H@Q@ZidKd?k*jzmy}+3x=E90CI^7bOSub)BNIl1DI2}svmm}eD$-BHQ
z#u+wS)r2%1#L}=){ay`IG$;`?gL<FK2#q3xFs|oZCLeb@lYqxhz5OKmY3Vj+9(=HB
z@vJi|c`n5A8!x!)WH7Kwkx{~2D&`992rml~2)FnlnIBeZjRGppM}Fr&$PCBINx8)F
zy@u&f@TdzEq0*I*`%(f0AL7mwLj(4**Brj9`s3ac>3k;*pG?9rgcLmrqfXE2QL*_=
z6YA@dWq-CxWW!0R1^#^=Zy;o!QtdLhkBrrD<dO$iZPZ-gBm?KLQQye(t(cRY@hy0M
zRAp2;n~rH^5^ToMvyMXvV=tjl=JE%R#ksa=QC}>lJ!f_8nkTsZ<u-vTk4jCR4?a_5
zf7GvUQk>ccTFib&l6WFVdsR@by-gC(Yu?*It9uJU!3i29@wl^J_nuJjH0hh9<S^02
zG*-0|%hF=VZ#DKB`jxK%7vxF3<jV5(<_ML09(CJ(B)hK0O@)dPd56pa7E9}t3XAku
zF>gj0I~SF8KK!%#RosrqREN$;lNS|lP#B=u=@UaC!!@Y*M84yi>?DXeCsuk_b&dy(
z2*>>dpCvl;5dKHr2`}IVMfH)@6FhDiQ3pLM6rH|@JxMMWA#a6cOUTQR1?+;A*TNfU
zf#_Oj>>>}OhOZ)!a;<8n!v#Z5Elo%8F)|FxbXro4fcgo$O4zX8rYj7&yU4V9cnt2#
z^H&_AbWwsW*p|wNIOvgM8s_8BYq%RjFSwvAmC71eDStCfpqq^VKL}(^pkaI|Vtt2T
z`W6@bu>dz><XQi?^;Mgey!BY*hr%(;n}UzTKIczGr9B6CHX2yclTsP&)v7MnAiwdi
zGaBLaHocvwmJYOOWceElxEKefiO$5}3sR=g$c2b`5VeNi5-*Ke<FJ}2&)?F#&c4_T
z_@mb$G}Xgu<OkNp_11Pw(VkF=nGn3S2wSHBs;4cUCc)==YPp-dM{@3o;OeT5dj!aP
z3AftU0BwKb8RlreYi;#RKCttFv$;PkzdC$nA5KCy?pdGIBWFM|3>aGQ*T}c#*fGm<
zCS$xz`mBixrNh~G=Ah$?jKO2sP>ne?P7gl<EcuqJJ{Fc#gh~KV8#L)%bo>Q8ipr0W
zsyCncVJBn9JKr0s!_kcjlL+7e9pp*E?Vx?sAmqM}!!(tk1bvp^O++u{x^<{8aW|^%
zlr;m*GqzjJ9Tn#T=Cu>%xHQZ&=$dE&pGHLSb2S;KCzYpky-?y0Gae;a?<tayYV@I)
zB3;BAlp6<8GxMoJL?=$z3Z|9>>9~i7lC_~m8l-uoARbLe!)mm&6j*{gq%xLXm|J@G
z++<dFZAPM*ZO`!N=KI@3aD91$JviQbukn_wxyt&U&D9;Nn@)8J5&10)h)kBRtWAxi
zm;IF>5>J@mzZsq`bd1YRx6M1#EE_9m3U&w~e6Hk#_HFAqe@h0CNUO94%df6w$bJbQ
zBz9)d>4+M8RSkE&I;U`~boK=Md*7BfYQqp6OC%2RlIp~bdh6LcQ2Lwb&>jO0_{t@Q
z7g=++Q!<Yls8%Fi>&uN9CJ4sXz4<{yDFReR&Cdw%!}%5{Y{{6STt-F@;bv1DpA`Up
z<2|E|!*25SUAKKr<QS|<sVX&uzQ6l6_@w!1EgLzF!KxUfP+9F>Dqv~-be&x{46dd9
z*Vjku($g4wW(;Q37L&>(t!IlrzPli7$%T~&*=JoaWIi#6t(`1iXrnsDMo)plWx{B7
z2e22fSYbIaP=X12^`|w2WA(*^R^cQN_C&{P@||}3Y0y+uk1mZgV%vY64>*to`sU4>
zX3Ur<b@U~XD!D=Z+zT_sY)>`^F9grn7Mt3Rcyh;90nRIOBA8nrIiSuP5W(mln0Mif
zX^rKUJdS*#0wZ?}DOE>ct_FXJH!$(I4WGRwGkNK^uwEk26E_j^-sMXfv5x3EPkmid
zFu};%w8~XW@Sah>?x7OIBb>6Ax1X%JF8@AmT-le|5{#@1D<9D=7sy{@_><xPJ@{8T
z!<Of%%GG5wfXY~pmu|s?p4aY*_y^Tf!1?|9T4=~Kj0)X_idEakJ~z>6<)$6w?ILsi
zO*U+bW=sTpv~=r=pIsp3en-`@_VM=T)WILHA)ch@O*(z|toDAZ$g)&UCRdAL+4A-~
zlAY*--+fgsV#ftmG^I|V=F7jvJuQt6^_TMssV(^t_}v)7nN{J|*adYOdXuam99V2-
z^e#gW$Td@DFnvQTNp|^q#{{bGuFeBIe$3!!O3Q=OHN{)L5>-jDi&-7Kl7wCk!)DU8
zO8u5q(7X#S*ak>9m=;q<TV2|3Q*imwq%o@M38VbU1s@CBe&VQw#@`&IG?$<0NM1?)
zXqV8Xpep==KHrYs!vwj)k%=KH^u06ehq_oAStEDHMNQJ(O8u!=IV>Hrj(&Nui9>uR
zx9ZOF>8qlI=d86uSI%l4YwCIu@7klE8@yGQ57%81zIBTG(XP@RvYh*y_u~yuDJ^Zc
zMW#{=45p=$V8^7%#siFq{{mk4*Mqa0x=eBF5ik#K`Jp11;<S&mut7Ha?(om$i`YV1
z=`a1}0A-_%aDRq38ryo@g#MljGqmU&J<9yYV%?Ps`NJrNsa)4o>QP+@zCPs0;T}FW
zAM;!Eob)(Zu2D{af2rV~@~k8R-5tvIVGHJSO@n((u9o6SP4N+ro`+I5PpJDkj}e|a
zFe^UEx}{zL2=zy8ehl(LADBaE&{Kg6aC}f8nmr4O?Ik!#B`WyMohz-($B#9!If<T6
z2z=Y)XS_J~>8@DZ2gIxrec%18+{~}gfXZoA0*ERE78~jFFXA*=Y?sfOB(rNxWJa-`
zX9IRqav^Pv;E@Yirbrb8yGsrPK9M^J+CgU(*`XgK%Dx@}M-6grY`(WwW)p70Zg6Vl
z_{iVho6LfD!cIGfB07k=l{uqsXQ73{D}2L#UN7T<2#w+v4ec1CrzxTO13d_b_53W?
z1B$lRst8hQH-f!1p988CK{F%X@x_I>M!C6P7i3^+IHwY3y9A6EMf~7E^(!EF5)6r*
zcLoA|_;wNcHih0Sdrt3nMG5lfU}QRuJAt#&THBE?sou_o4!*i?81}ktOMa3<*pKDw
zI`%T4`+C_vPV?p4h!5EPE~2T(z%3=%79p=Ajo{|-1_0uB-Kj-;;=`2Ng|CQuNU><_
zFFbJG3vb52Y$Ygj_w9#CSW|U>bwA?eQG%y2xg2{<wYUN5zMpI2G*VzYol%z*r|~pB
zjK0YiA1r6m;3h>!KL_Eo8bOU7qpK6g?+2}NZMVjFMzP1S!aSk~$s~R|Hij$Ln?vTc
zGjZn8J`i7h%gkzxaQd2nyGn?1)oH9vXr{{UJ58?KG7qs{aK*!sY9fbjaLrCp@Xjwb
z;xX7P;n-=^jY&P^(73;pljp1_KL>b{%qL3Lnz}a97EGyz0l%?Z!D+__M(nGagjrdN
z{h>(V4+z3=OfCnH*Dnwbec(=%Z5N$9$JU4=K@}$}0<r!4R{wkz#ImafKeSfKIIbS9
zu8Qr@p9Wi|Ci3z5dO9FE(d)w<ht`(omVKf-Wuiu*)%&wF9&GZLl3N>#FVqxHr9`s&
zV*<$g!xB%w<lbu0_b)x-LcLLSjP<+YvMM_BYG{OPU6B28Io@~QrX5K^CmlxZbCJ>D
zqCKM3#%qryvaOA7)IGvu6P-+4^=O|zsXWrW?I!KsGy5D@F6HY_-OkqjR(sc@gc3r8
zQwv#O6Y;Q|&zH=Je7$R^C{#_z_VBT17a|^-5%`?GJKt4E7oFK@In<QxPBKEB4nrCi
zPZr9?%uw@Wp5kgIuzV1#&J;q`B3sh}m;FJ<FACf!!4iMfdp6M~4?ppGy)EmKn3LHX
zjtqqMNa-N{qS*XSO8yL!#iXA_$sna5Mqy0_lpVN)RTw#f>P9M(8}e<p(xYH0<*4g>
z;AH@nY=CT$HfUTgTH|1$HmDENb+xj-NUAiRz?ionu>sxGwPPp1GOP1;_X7RxCF;c#
zG-T1EKvR5KgHRnx>;_dbX2g>3-a`poc!>EVlbKSoLW!M;=Y`=%Sfx~+smi5t$P3t$
zP4(7lu24(aj@*p9o1;8nxKqgr!M-ZorEBMfBZ_(F892V^iozNi&aR)q%K$%gpL8`j
z<OK=L_q|-W+=R;W)IQDHhmze4)b%A#UHgWBE9^z$OwQYkHwF4VHCF)^u>=pg=#cw7
zsmr{QZ2SQX1@-9}IpQ<sA@{^}^d!v9rPn)a9FmR(YRt^Z!`tavQ^ng6>QX>+#=}6S
z>K1TG&E&GY{5kWEe#k^n=*}j_fy_(h(}Ikf(Nxa7kCGyeh)e0u-HF}o*ZqxPWYMc$
z3-(*+8MWp-!Ck%FY?-|03$#SzS}TQbG-yOpMe5HPcX3ZVTQeG?(d>0ynuO?hysaVc
zqhLv~&c5NdHa<CMl!m)94kWA>7}Yg&=EbZW`=3A8DqHVq$`1k)(ch!}3z3`;uwa-J
z*u@I6guadSVG_sa;}&XYIxuNxDS2#>H}05=8eQ*skyS&masARl3C<Jd@3p~u1V|v_
ze+P6;t$rm_Lk{3I6-ugtop*Zi!aHgDs2NOsx*M(BR#_&C&8X&p%1;oGZki~uusbPt
zD6v(~#SYhVNU7~6tJWx}B^pxnchNeMWwao#C)uu7%*ZP!c+YuVA@4*{nBa`{n|3l4
zdrym!0&2Vn(;h%?X?)Rr;795Z-P4~bXB~$b3JjB0`t8!(Di_o$@8kLO<4VkK*S>#J
zG?8#m1>GsY)(23{PX4|b!O&K=!v??-CBfllXu%VF7Yqe?w#P-h9&8(81`mq8evj@>
z4koSdOiO5?GD{OkeWQI}6-ZZK&nP8@mgWC;5SP~sPkTtQK;!3M!r2w22Pife+8WjQ
z6Kt!Au*EB(NY1QGEM;kpU;%QoQSk{OQEUDANVehx_>#Iq+!|!JeF&S~mr(P~lBR#c
ziv0L#<KUM|fG3js?v=Bli*+W|x8O(eoZMcEh9%SkLlol?16NmK7(A!z<k20UL;mNZ
zQz=A17j$r);5~{c*V3qjK~u>$0gEqj4uEx0%waJ`O$AZ>+%7@=Tk+)n5l)HI6oueo
z;JC(}X^mjz-RePpk@t(0y#?)e?a{Mb<+pG1T}N2x#O^ZRak<WhQS7!OMK0cGe<Gn0
z$bRp$9QF}m=+=>h-U)n%GKw(D3dP2?Gvju)d2B0H;JJ0;|GaL=hO-VE2iauK-<RO(
z@}{Q9%GckNF^>GPeG{Ol$A6B;8I%xgy=E<Xy((&Mpbb=we1L1`6J6ywBD>wMJR26N
ztfR77J~cy+`>f&UV$2pYbfFv<o9m(-=|si^QOVoDLVLklT}!|W=Mc3a;Lg1P1}Zgt
zMlXWzEAk(Pzc8)ida~uuM%GsOvaA-_aBrtF0Z#Grh`(G>H0pOv0z$4oxn;kXm)&Db
z_Z@l+gOqL8qzO`t!ylo^d;wn__g|_6A)w~Rj~<v?P&@HZ*V=ceF0JDg^td2JE^V^b
z9Gr4~ZsxJ3gdc)IH(FH<uGx=!BJ(Z@Z2h6~-%mQ%DwBE&XfB<EK<|wbp65wx4~Tc3
z?e>`5t<X%0J8<7B*4S%P)0(_7@C}~r6^)QkvV(m7y#$Xh=iMnWp|{b4_8G_)O2p^Z
zMdyll=uij<wqipdFFqIyTu_=u-geSA4kX@#b2XWq{~eJ&0)yXhPYx9Ia0B4Ph^5_O
zIWRB(YWkVJ`T?eRz(ooJ=iz|p&4|%%>>iJKaScg=N5VzRQxCO^IvV<7WZ|6L5&RNF
z|Iw~+e9=6T?CU1+5|=TYDpL=c>&%b+ZR@Q#b^oz&MqF>^t8HRJf}1%q&h2=bNk4?H
zEUiomWDZVLz<Eptuj5ZPo=Q+&{`!ffM8JHRXGT_h9pC87bfuS`m(GD{*r4+SY;Bnd
z6pJLEUS<g()*?eYc-fYj^;=WW+G;&Xs{YE{pzT1~)Tvr(!EqOi(0um!2U%x8O2j2J
zi@6k@5&bLuGHw8;?Jz<3XD6nJ?042KuyKoqZ=NdrN=^(gLhL}H<$)PMRNhqbM~w0(
z$+lA*>B~`MI)o=exSD5FteHw`6TORUM71n32!c*UDkG=n9~?QSqWgx16o==Zu+%cI
zO*4Z~9&?~HHy0MU1g0nC+%6q|1P)l=6o*{j=Kf#@1u#Q@*raI8L;J%FSzrA)HFbhg
z9yY=vHw>vQ9>xR_eVOr8YJn0?;Tj{oRO*&a*(1yDzc3Hny_^&-o9q3$0#0j>p9aeK
zE_!AZG;eff1oT84C#5)z+Z(bMOJ<Pi)sJCj(wrY>nF`M1e_JOkjamv?RZ;ECdJ_45
zav1BVy4DB+@ykhi)=jq9$~@SU;0h4HK3CU((VdGYla9gxgJZo+e~+)_+h~oes=3&8
zc9E>LOz7m46*00kT5BIjcu9?0xq@56fGHvcpBM0Gh=#a(CPXF88)h6K?))7>45JYm
zcS%yW#Yh!-r52{dGEAeFc>)Jco*pixn&BNG2Cco6a$zcb$gsl;`IedhF5rZ<ft>4&
zj}TnP8;^LgAl#nKPYQ!X#3DbleB?8sn*{26M@m3n^UBX}s_iGm$1#?~tutXda59As
zxX<kVNNrN+Xxa|(_6rR<Q#&=7wsXZntzCG%F{8$L&Pt?o#w&Su{7M(PS4{0XVt0-E
zTi?go)?mRTL|oc&JnEZ?Z-L5aP=Y>-3(ssm&f}ASL=)LYvXrfM#M@tL8p>xL*2C*m
zfBN^=e17==Ryk%+=#h;omSmrQf}YPs3Es?zYwqtsuJ@48LLg5tm-HYZ<oIr#pM$6<
z(nc2&gb5`S|I*cdn?h(87CMTRpemqRGD2*YTBN%>l~&u&q4P)27IoTqutSD|<bRo`
zyB`jmZx@=5vq)w*7b@B^J}co01Kr7arkl3}O0$Yy8Tk&MjN;^a&f#h>=plV_J82El
z`Hv7$<jbd}tOl8Olh_H&G|o0H_=}y{EY8FUdL(=w{QWWFH2nbqv-^!&tLrluUbOUM
z-c=e>?OUB562%+u)WpYkDB&8u9EYntWyvl`uVr|1t>P((Pca9F8<eSov}+QJIcF{B
z@TA@EKipTT_=eLEf>e7E=e#Kw1Yk-73M{zUm}sz`SsJ}Bzv`kTP`cWQF+JB-x({@J
z#EG%UZg@)1CF`r%!ueq0h{y~1L-8C0TWn*b|2&ur*24t2C>?3)3tE#E`j^iVesr$(
zWyrL9Mj1%x;dOM&ZNj6CoPRmgb1ku`l)0q8ksyXoy*jK@op!7v>;%`ob>c^OQ?iEk
zTPNc!C)Pe!#??_QXvLt8lgJ&+MwoqEkK2oQ#oSzE5gpIfv^uY(>uNDmWT8Bz?}n-E
zwGagLM5bLmTtvg?t`@%bt*-mvu$yP}?k9h5$y9SVF75nP_(K}~LFk~VhX$7R!Sg+$
z(D}8Qh1wq?*usY#QpLN-W+*$!Jzi5C_B1J3oY&IwYqTdYPIwh{s`ni6I)O==m@Dvo
z;iM+9mpkFPO)6!pg(dP-#|}hvE(YWibS*a5I;%>7s-;)X|IJ$6_Y)FLcBMF@Wj98@
zzItq}LQ#m+w@a)(XHcFles&V7c}=XGlP5;{;RiY-rR09J#M8RxU}#VL)}QLs?>Yqi
z{@mx*F&%|`es`27D)Tn0W<b=wrmm!dY&c`MLdQNf$C5(kF4Bx#nx6WCbnm4<v_7Zc
ztm;D=?%+K`*jmkKjFCs|vxjPzdn$G31sWF^uJIP0&&&Q|_+%pEb!(i}lqbUEMHO|?
zFNtr3PT4Wdc#4?_XYf>^aBSg;KhOdV6gm;J@AhLBzS)QXptU7`4hnD0@mGb|ICi15
zNuSLLsyVT@GRrB*L#M0yg8RDGIKjo@5s8iC$S4-LlC3R2bznZln8e$<4pWth#-!da
zhB^oRLg>#VLE)U5f{{Ylg-b2miR<z0h}tP@548_A`*8NM_Ra2paHdj8WUs^_^-+wL
ziv(tuVo#v7$SSr_W7ZS;cfLfAp?;<krX9ZKKs8o(02I08RYRJrRP~%_tZ{Z_%5?Lp
zbq15BeM4SS@VqkboE#R7Sr-8X%$9a&ROI-G7!43fBP;6uJ~$PxfwL}Tlk+xDj5;5m
zaMo-twk7$RG6S+|><UNI3n{I*#9*YL=OXpFqW$!om5-i28I}cs0@IVzx8sZ#zv75m
zt9=hT7rat5L{t)1P)(<NXYGU}1Q;MhwJI42{uszmbqp&F1RwiQe*T;qk<2Z7d0bc*
z<Bnm2k4|bnA?xKV0wv>XXwU*N%<+a+$-?=m%7&u^TZHOan$`zToN%y$y)jF$flcOh
zENJWx7PYK|+f2+u3Qy6f_hKmY+7~3oug3X2Lq&iezEXb7Fr54KH1?xd8>_sn*+ZC?
z=b|Q6-eT}Ao5xB!S(dmYFaUt*hA%*v@5C#TYb4ipiuL5N_KdDKqaI=cMcY4h?T33f
zqr6<YiM@G~nI|U2*Mi=eomDu}s{wSdIyJAFdZvP)mGpMMQNt@;D0d9USpzuZxbZ>=
z&VbP7j8fZdhf2gCq9I|?=wi>tZZzsV$}XwQf6^gf<D~@d<0It^qFA?pj8pLWBij<6
zrpZU*H{LYrOpqEGaw72(#3kRa<M*R|=<RPgdxg!<-^_xW%u%ra=@zg1JS^no-m%_8
z{VI+W?<@OaL>Q-;ei8Ev$#3ptVhE!I?ejgYv?95scFv!Vd!cOk6K4qz#YKvGt1Ga|
z1sR6lO}n-lYbq~SUM3fV%Tv0}{r;R=*!8Z~iEB(N6GOm@qB@?>>h4=+J{1|$#4nTu
zwPhaeLGl5?ZdRtB3vZiTf-Hn745`JbIT>9T=1Jd-vrrhyP$4qS)5%$6c<wiI=$aoO
zqyql5rld{h7~f1J5L7%+$ly4S6;2H9G$7})l`uf>7GaQT4zl5wB?H4?eCQmktGN>S
z%M};8;{=#t^3Arbs(h^#3nuj2kgy>g<T20nBD^7N8u#%t>G+`+U!olHfR<ux?AM5<
zWvPIJ;DD`Ft2v_1Q0VsTR2nol>fTI!WlHt%{;Ypw+W{`!)y;CXe&0nwGOTPG)d^d%
z(}_tU6H0*)<Rf{pb-JcbbuGNxPF75N11!TMng2e(FvwYp7GVwKMcLi^UNDlkE3-fz
zeTSE^uQpnJmG#5IqWE2srPPyZ!wnnkE8b7<Z4Gkh`VV(yj1kDA<y*su9gEBlnKyR~
zF4WgxB&+2q{=i0%?~hj~Gm-(i!Vrd#^hDBe8acaYP}BpPKcoep14|nT4`RPW2#snC
zV_^*C!(i30XG8}AsjoA|i8>%0qmHJQIAg9?aM*>$&iUSNf@#yu5KYA-wGenMD&X3`
zQ}*fz4Je024<CYbWgaDSCliXxdkl)%RM3isSuHDMJib{m9b|Ej*C}SpI>)9izGq0>
z#{SsAlE+4nl*NVajr`$#vDea}NulI;rnlF6wW(9y77Ye_6Lq9=7?n_j|Ja*nzkp2v
zkWRK_s(roU%N~ct4^VI$m3E|Z^NwyIe)<)n9hRk<{ZmpaUJw4QNw3GzsM)G98>TA)
z`5e0Mx1gT@LnDPJO&B_t(P8y)JVH9d<-xIJ$)P{;DnAF?YzgL@IWGhVq3D{avDGHl
zG9OehQOk#IDW=Xfh@<GL0N0Oqoh=tGK5o(KG*qHK6AYg`CWC1V#p`WPN?Pw)t4{C!
zRIL<i(Dy$<2d@DAR{_^H2!T>EA&jxjv53V7{rMBRRpw(|1t|LBAld7pFSdqz6fYUG
z1Vp}HkJ!lvr0c(?`I<>zSB8(gw<$j`G#<{hi4D-59NiXp{h6`0OigZai{Hi=t|wPd
z%Akif4=GW<`~DV{X_IcArjD~pz#WEfEv$elPr`9ePS?Jw7b^v7w>{|O72LOI2*W)V
z8!4D_Uxl`yC3+yaHr-e#*H)HtKE`FK?t4zkrn_uePSXJOP8FanA5PV0iE|EuCFA>i
z@V=(9WsbOhln;|u4`$MBfH@nG8^THxuJ3VxLQIn+!yl|QjU-?@Ki~;8vzRcvoS8~I
zRZ);^pK8FLBuxDY8CglDo<U*Qna3>?!cr0l9ht-lCcIOZ>(WR^8`SN%)64x}8(?HD
zIa<Q@87TH8Pvpo9IFLnfe%B5eU{ZYjW6_|Y+5HVi-6&#pv-X&Kz<0(>a&H8D!dX|x
zH$&E@De=!TmmdSXW?XmxRZ~Gx=>1{0(GI$BrdiI<ddo_@`+U+Bl%=^}HW7z~kM`Gj
zZ9HwA3pG$-6^D{jOpW_xpE4ehfKNwY<4<IJFYFB595o8CS*f?Q=5=*^VJ5eE{JVzH
z)tnJfF12?^21vRvKq9aOFV}a)r6uQZtkIiKwq{d!|EI?pDnfEZ2wlr13h;r0Bl3p&
ziEO;Y_K~j~iPmd5sU~PTBYs4E^>k7I-@Mw6YtYT+8K}<g!Me#J<x(GrKU%e#NI!re
z!=L5wgrCWpeB@kKo+4lej$PlmT%WpfQK-T#lLC<|KhGYkb|CJkdn3e&SD#?x|4v5c
zt}B65C^)&eNqKchRL##(RQbXlkA7=GSf4r6foTUa4fu9>YQ8sZ=8xmz|1g&aJT6Aa
zFHTMmzQ0VNkR7heYIa;?si!pF&f2&zAdKh0_^srmh#H@-lt-)!7QZJ^;w<}d=hx3@
z8W7PPZErU|HENa83%<Q8(eynUY_LT2l*ilym@8NmkZKItWqpoC7fbAZkY-qfE0RSr
z!bIl-4T2KF1ESIY14pC1F-uUs$bWI{!>Fv@LHd0CV8~SFOo`JbLYnOAQ}qwfUMO}e
zx)=RDcZiG}n+1ddhMZrb<j3Uws$(`9Gs4d*ZGB%Yv@WfZd4Pdj=kR97N?Sm&>f8zG
z{z&Wg;SzCx=8fJpx1`<90otmP(9wG%L{Dlj7x|%M?mB+h^0yhWVy)rFYpfV<lX=pv
z@~C2hBh)7atUa(xV0JyBjScB=4Lb-_4!Oce?@JCcugQK8D<L5_^5y((iV~&l4dV0j
zw6+KSgRr8fSDp#yo))rZ27zi$JY1S%(AEu9I%J%Q23S*K<I(SE8ZJKTK}&oH6^BOR
zSwM*%XVACPr~u*l9@&G<&}r#MsPHfv0yL4!0gQJVEXNKp9lwxAj`hl%+sOvp<O$ow
z27;uRrZ_A;So)aVViRqdH*{LqHpw-h^A#m3;z#G%7+IZ6juj>O)1O1VUtyg-P-88P
z8CEYN-1rP>rCjfW9b-WoQj;fA6@9%|?ncWN;XE3WKqx;3M^yq8$zvg6RFHZOv%F+e
zgDlEm%r<HK;>zV%oEs`s&1TKCZn~KbDs-G`hj15-sh@|&k#4d?x_;k9dQwYPABz!f
zuo=(_U2bWr)s<5VAqQiK$tw-&!npYiyHDaY>G-6EJGSf9u==jV%Y4d%M%A6`-1>tO
ztO@y=vO6@!4>1>ZlR`&9Rby%@f)O|X?9(Jr9^WLVxbQ=HViwP_GbXd&*8!l$A6deX
z{4G8o{LZ`*oOeeXvCr_oD~htL8C^kyJJD$I=5ua{Px?yqE^CyZVXw%wTVoQ3OlKEs
zO8IWaG`s3cr=x1zq89~VEpsdg%^9$MFm%+r6JFV(lyJZ};sd_PF&TL}$r6=++?+Rp
zZt%388G~j+``w<Y_FIyPU@FH+@e4H_V`hq6lIbKhR;ofg&CKLJ;hn5lX+}vRTC>E}
zm96We5-9K{dLC&ui4Yh7<P>D<g-p~ls5C^%bv0139bY2*Vn9(XCfiuy8CRCi--3$Q
zy5S*e_-oZ_6XF`ibsgp;jRcq(_-P!@9_7Woo0Wj#uUHoknn`{24CG`z+Od+X-qMwA
zMl6RJz8eu1Lnb3USY;VXOH-Xc%(pxx)5K2+s7yj<;i1XL2ID`ZxhQo{)LIE^s~U!i
zL)<R-)B?U7OImh%KMIA_-aD6%=8FK7KUXjVx$eL~ySrlJnJC#4Bh1?-Aan`>r`!2#
zDvm(bI#_jb?SmZrb_y=|GrC**3U6rU=Oylq(525QAS!nH6iUx3t@{Uu<cLRwp2mnD
zoZywHox{7Vva+wu3}{lv0-VxZ-Z9o+Hl@%gCWa|dQ-VFJ;xmw~rjjSgJws}`<+-D3
zI+*lbYJ$gr=0$}ujG2lucQ*Gj;70=an)g&2oIGQXY&E(WT`*mmw@Dh0#jHiwV+wpT
zaHgsJ3l`Hnk1ENjH<JlKB*!Ro!WLij$mopX46-PIUAU7OlwtyW!ZOjup^%JMyqJkT
zq?l339$g*QK~f&g^Ljb?_kgr53*8$-ic28;E|WPLw_K<IyWKdyzz_8E7`i~4Jd|i0
zlRQMT4!4Lu4%04aIE!V)ypdh`5@=^kI=0b1P5n0Il>Ir4inE*P6%1*Nwi?CFsdcj5
z`kTxTrKq?rf(~}Oy>ez;<<g}lwm)~ur%jl$5WEgebx>xTa`%;LkdJF-aCBOT&VWjY
z5b>y!(eX+Kxh&NPBxWSThk~C$A+0ej<Lqql&%Trxi<)Sywo80d^;d|we44?%atXNC
z8P2DUYJHPpWFPbhcrRBGF;2McVVw*hxZYR2S%Y{<rIUB{)6JrF-B*u+U$+i?X}#8s
zhVdfSx^KgN^M1!@a2z{Q>=?q23gHmJh(HAQHn%SkvB8y!5iA(Nf~L-ntl0ccZqe!P
z!JRR~j!2h^RJFCqR6$6mxGm_BrGhDL>eX)<54!Ga2?3}TYO=e}h|RtV7AlD$PiYwF
zG!9(iT1*qtx%-OK;)M54!}Ij*+inGFxiDKX=%|LP52FB=JD1{{Xn*?XZh%Qhl(?OO
zLOTPUr|d;GVOlWv#MV&<IImtwQn1I<m631KD<gQ#*w~)V8vBqk7WWu-__G5~f7adb
zTs{ljpWY9s_LpoW)CImj_o(Qrz4FS}`-ZLP2zRB0NI}1F1EZMhSqLG48%7S_oX>6y
z35@#62m%kjuPta|F%G>z(dH$A*345B6^OdX-g7LnYj-)AGC+3vzNe(3(Clh&V#T3l
zPOx;i9UWZ8h2t%QctXO0s?mtYvx2w4#kFyGW3z=$&g&JWz!@H37<s9xZ4==91T`Xf
z5@{DN*A^?1dv4^2VU>FYEdt?oavvi#kA1lbuWdz(eyf21Ne=Q#RK2T^gigSzH4-_B
zt4q7s0BK-(w;3wwCrw&~&1A!XNDU#@EDM^WywTwcdk}*2y5bGdO#C@}U7~TikC(F!
zx3osKYq~ah!39(M#fxOG8Fd4uzgoFKnttN{`+L;2(fFe__HtNQjJ2fEI@aQX?BH?r
zr$NhzgT!#y+!{@0F$Rucx>LlK{nSZ&jN;Bq@mF<={g&BNO*%Khd2>E)0^ZR#kV1r`
z@>J^>=z(GD`~q}Ful!{5AVN?Yd}FFj`m3zN1vW1x+|JFkJ!Yo!mf^DfIVt(kN3nzV
zC;yV&7RJC5*6#tq8K7q?zjEe2x}@5`56%gHpUT!J^p<B0e)Lvr!liIjGg^@_RvMp?
z+w>x{AAdIo@%5=)P_E;qyBc7buuT1MkKgPZLlptt<uL|TPwU;BOXw78z~o=E%9EmI
zYe}bgWHWu`PV<W5f%G*2yw&O#Q=bIpK(4xh?oHahkT?l=%|fFVvOz)29cf;Ho6{}g
zy+B43>l}-mt7O?ULrhd9a+g?1^{9ME<B)Fke?W!iTkd0Wyy6SLk4j*m<y|al4)#QY
ztORyo(~^WY;wlHQ5$3NLl7!^8X$FV;*(j?Ugn7)7rDeRP@yE-d4~$ue`V=v*-U{hL
zSb+KW@B8!&#VL=tj^XqR_lP#Z+90~sKv$xWVWEb9Ak6Ugf4p%`G^V>V{Tx#9;9=z$
zQW7UnNBZ5ISuiL&P#o;6`iK6!Oxo~+Cdw&+4%z~Yedr=`cteBwfpAjiB-q|IKrnwg
z5|fIHz0Q*n%8LbB#%{J`uG-Y)WdTP+<&}}AB)*j0eWI9(B)J1B!^Bn$)diJgKn*vv
zJ&v#YcoH6{-~#V?$Mz=ei-sB|W(onM2GEMrvjc9ccqjY6ztenZ=v($ib#oOS=C|}X
z75i8kAtNNh?H|BfAEI%u{+3D058tE7VFt3L<KB`z`QK(Z&h%H3isx40@T%N%R+xjl
zN8*zB;2Mk^zRut027Uw77)g^6HeTvd*C5=K4O)ha<mEW7BdAxDtZOLOp3SJejs7f;
z7#x)mq+OJvr_r0Pimn!ly~wd#;vNfA&cLMhY5wte7eSl&sW&|~P;YiMi(mzAw`<B|
zX@~Hs?2aI3fCYhyKY-ck0v{cShDLRchQta-KgHuySk@D|TGB?ttAhUXlcF1s0bQQ}
zwv9&JfZK1_IsDPiLWm0!JAs0!*NH*Vx<V#3A9T@W?mU`nji+puXnt=3#OtcTH52me
zn^sXY?|_k=jY!b%gt4)4h!5J>TNBS^Thc2qx=b)VD;s!}uEp^A+<2Q1$w7g`48vnH
z&79pSEsAzX)q6)asTf@8E2YX}2L{c@6sxbbwwhP$=UDuqJOu2iIt<}+5aI#>F-v~&
zdXtP-F}=X`AND^_9$tNwr24|+C>zg82bbl|e{TJ5ky)jagaK0S>|bj;`Cb>0-9Xlz
zxF&@@1c~bAgy~GiND$s-k=5ZUATva0u8jyW?QGVoG+I~5Ra1vQ@5mKW>ZT-mI)TEJ
zE%OSI$T>KY$Bco49+gw9Q?`_>oaTbbh7zv2I+}!Q)}f1kf=@IPBiTyr-M~52Dg~6)
zw#3Np0PiD6WT`kZYn1HPN8LliBEoQX-UQAvK{3xvjbh=9FWl8&-LX-0lqx9EF+Z$R
z2Oo?t6E76)+3RJ&QJ7;%QM_Uz48xUDE<+5{!MnCPU9G$xwg?GCMNr|{?u=CvjAB!2
zTxR5ry)HY>Jg<psabv^-4dm#e*A=VuSD%QCjN&G~vG(_YOrY--@CDZ=<w)6;p<(Bq
zbJh-i{hT@ZW$~vX#`-0N>d%}RyU|4|<@nl&C`P7Qo&2Dppd?u@P>|Bb#4z&eW>PT)
z43$qd^sFa1cj!k|UY$T<pVdC*!S?_7AzKfw{;X)N6}4n}DQa1id+U!c$RSm!-*+(K
zyfF_FA25`dJ<(DCE(DYi;nH*jhAzxFXZ=AR(g(&53<vY};ocJA?H&dQJjAP#X7!MA
z{~T+>CHfd{s8=YQSj#_{v}Q<E_?w*9Z28&wMu^5ptlc9#I7ninFEJXqLjeFSRY_mm
z=k;4Z@wH&Z>nc8GFyk$b4|PTlbgTU%IjCls&;6vhkMOu5Da7tD@Z33K^Ho;?K7y7x
z^lcdc9wr!iRU5^HJycVn6k@(xWpcJ0XD=$HXLw!)t`0;Ahi?1Gu~F@><R1!`+T5I*
z-(`KdbEIgPx-ZmgnFU*Gp6l|>tc6%E#*07})f=`6Suyl^$oU}?E075k21bQ21rA+l
zuzG3{N$z9><T32d=QErGZg|>A{`0&4BBhp&(`^_Xz0O@t(o5Le25S6yc7500W@|nP
zer4o$6IbiOm>P<W8Mq#>Mv5=x!HH$6@}8pu*<LyXgozDpe=`)dsH~hji((U_PD<mn
zg~JZ~TE<G%FHzn~J2VOhfGoh}5Y(9~SpVmrrr-!PS;8C9`u7&D*)ErUP)b@SY>`1G
zK^+2$KP%;H#=lS9p1$!4#=+1I_xFjsMhsba?L%H*?>rXw$5t9_)a6y>#WKI#qq%g)
zlHQUZa@O7WP*`$>jb+2S=-JJ5i6h}xjl7|k5>sR{lvYiR{~RTRNi5|&j$O#XYBw8e
zV)u}eUL3gH*_{NB=$TWgitd6~-T>?4pN+>=3KI=Ykw~yO6^?ztRbBk*#*6>-=18n(
zrBTgmecl76`9dms-OqrB0hPDo-*ov9+<c3?-@EmHS2K8bI;T2uOOb1Q&>PN2{N8L$
ziO}r0VNE5rI>zR7_i(n5(xY%sqQ+3$s`}_i`#fuj9-)S$)!LB#S-yY*A&-<X>3=fm
z!MOiL#IlAdQA7i3A&=r~<bz@h<##<0u-P;S6?{7PG?5Jz!U#Vt5Q8A(c-F%_`E3h1
zi=^nwsal%jIEPg{tb*Zsrtng%0<Gm`3rQrnj&JOQDf2PdK|c%}Cv7=i^}F;~wj5F|
za`M7AopH@NjDgx2E!-+)+bbjj@B1y>*!wnSP#Kew&W1OHwd^=KM66|m)~_23&{b~Z
z;?6_c>E!l6?-o)%Q<-A-kXMu@NA5u<L<m>JfdxIZBXLWjE8PTjVfn3S4`DSHuACYp
zejh$(7ad)yYArz)owfmK3&H?PEUXz}`eu}Fusy~Xn3v&$a8SL<u$e@APNQfI8}Gyc
zs=Y@*fy0H(u;>v#8{-b=V&hRIePq+0+&du~w7L<MLUd#UuO%_c^)@p@!C&Z(CO28f
zLbTdmt*rQ%g20eCu;54J`Ev@(`V-M-TmDrUR7-fCa{pyr5bQPV2;HUIkY(VsU70*8
zolu2{*UAsWMreyqBwpCP<73EK_$4*<e3MxJvvP}toj;6xcPgYT0Y%sCaFPi8^@YG>
z#_*REm3$tfl=7rQ$W(ueA=bs2QL=K?ZY=AK67DpfMEF$ALicc2)AU~iS=DSeFfroq
zU~+8yau(i<Q+;bnKN4~3e*(%4a_*FWnX1&f1s6C500q7YhpUJ7UNtWbJ|{18oY5k6
zoKU}Rc?{a>`}`TYToxw$psxG`AhMFv1Q;U}pyCLfA-cPa#<RVtwz&cb8zsjdNZY}n
z&bEX}vBo_NXLQbCFVW<2#Cn1trfY|{faZOhTjby$bZcZ?EJY&MBaU;&n3pN6Z|(F7
zjJME8<0=X8qn8ouvJfz_1|Ol)wa{1V@+cxbi1jRZHIRof^epr$$;anH_}1MC<)wT=
zA5U{qzm-d)f*B<)nIOT&0*pSU^(c!y=HZm_?|V~7*AtDRZUzlTVU`jWn_5=QaI&H#
zDbg6l!GD(@izTLfv}YzM`8v4?{D#P<IY4B`7uzX-2&vC5LJ8v<U`pg-a@K*!j;f0I
z;gZs3GM`gI38BCb(~9Iv0eMoPNna`Xa6Il*xqZZ^A;l3azj-U$>SPkU%NrZ%ZDig^
zgU>*?IPoGn^-yYfMqFw3UIe`wndCn|+rqDn66<jjnx#~h4OVu^Z5UJK+xanBC5YAA
zMr<_V=biN)#%Al<lGq@39_&;53VN0T^MduJrjj&U#RvzXF%Mo7#P6CO8eUetZZ_V7
zJ?w7zTg<Bz8(-`-My!zIr;<HPn_5?#iV(`7su2uFbCXq-bt&1gYe{#`nbX|SP*Vm*
zY_c~*Qo~iM%F@XeCT#%>+bzSYN`$65wO=Gw=;25)v^F{i9yOznm73rvxK`Cl+zj#)
zUY~@Rcx!!aFw@ci2BFc9S>}6m`oqJO4MngtWW|1j8p)_^isMv{ji&BU)05PKT{uta
zCYayPR+l!_-?6-^y@6;9tx|hcUT=$lMgy^zJRB{grrsN4rHTqa?RS=G3Wx!>HQoU>
zxC6FrVtmQ*<o^C-!V`OxmNF|ShxbwaATS_v#l^3R9cXCr;#Z34fpDH#9xd-kLrtxG
z=ys@u0hts922Ze(Ub$qBeF`TX3(cm<QXA@@GKP1L?B;0sY@B!5rTnG~9uqgNL0~^N
z(<pPG?FX+Ef%3vwrFZglPy>^VlS$2gcwfrHSr$V|)<8Fwc5#eapLHm05}bV+J%5Mt
zN=_?J4aCjB1{JBJQE!5ms}U9^vpM|g4-%)baOGYH*;MGI90bSh^&h_A_>vJW&N=eb
zFcwkY|7wB8EP#`ozfvvK<75!&Bj$t_jUjq$e^rjVcRbf7x%q%huGJC)^?KzAP=gl_
z%+}AcHL@mM*d%)&G<n!uoZ(iAV3g^!x!pHS-_}@&{`Nrn0=1RK{Lr7AA<e4b36$J|
zaZ0Durzv)YJ~)F@v+l1{-p6mPEw0Cs<(sL0ih~f^(xw?IQIop2^$jQs<L&=N;(Do<
zA`lyvKxGBLk6Lu6y>)ICm(=v_gtmZO0wk^fh3Eyn@?wtFF=M3bypP;cMu<?wAbHX*
zZP@cvN%Nb;c3HMb3WBXridAi<nd{!eT&O6`NOV?e552r>3gl^Z0n#{5ap6jeZ%{+l
z#nZ1Pet_}3E8Q+%UT`v;2+|k_513AhBzE!mDMzFPmPzK>XF>}Gg6{i4x&^X*7+9{s
z?`6RS?}xjXKZyG;4Ue2*M8jvAlp?n_-$R_wo6pVsiCwJ62@A%Q?(Kq`2x4UnOp<k?
zDfK<B+w>Gj(@8gep9C#xgbg@-R`qgtrm1W|Y#8#;2<OL3nn{?dYOB~Ae|xj5s@E)$
zSDmw0ojH3hRH?1(azrf6Z|6jR;y6QW>ldB140d2?tLQmDu1$?2I!a5iADt^?6&d{1
zBrfLO5%r)n5=#VL4{8fLY0cqtvPW_O#}Jg{s*<`cX<&sC1DdlWzPtqLZKA56ZKl?+
zh{@Nm7E?ax(0Wq+@Ck_{NF&&-`IQ*Fb_(w`Xdv_d1;`*b-*$`e&h*rBV)daJC>gu2
zJG<GXB<=4nE!hp<*-LjuS*)*I<)vuStSkb!^?%h4m_qCRSzod&Md1Tp4NApNuIwMJ
z$<c`ZEpy7l0j=G5{1#WghP!)AR?){j#!NSSAcm@=iPVLzLS>Wi0}gV*%<*9>2uXuQ
zU~B23UIMCuIdDBnT_~w#9jitKGxWUn4d6g`*qf`Sy25e+&lCD<9pB64@I<Xd)dIZO
zV{=$+Xf=MnP2RrKMB7lTtZvd845=;fB}75X`#|~1nzb`s^Rs}&L0pv2C$JT?lw){1
z(6`FpQ$(9(fQ7zm{m)yFodOD?WE?g+pnz3N@W4TbsGa{Qs9VJ3Yvmth>k|RWQ@rQE
zD@$zM3WCC#>#2nSx=})AR;GTy`4~~OQ~*Bhbd&E&*C2WFI!Hm8bicb2G#N^;rhQpe
zt37yBYoS?#ha#CygoP7b^-b2oQnu6wM)4USWssNAu3@&3tYJ(ZmV*d9$1Z8Fg^8^9
z0Zq~?IJPdLFFOkiXq<7;K(p!3!(~f`tk77)s^HMAwzD7G6q$^(`~ZWIn8$*h*ftRM
zE|Xkv`P7TE{sN`cJ4n<GJ>wHU%B!0lApa{3D?H*=@ALi$<-QJSt?|ZDoERVhh9q}c
zp#}Pr3tX*B=k(3mI(o6*52BoF6L^0R2uA&S)`HnX>xoor&i>FyDf1iBwL{R_*Dv4B
z<841Cf{O-P!rlcGki#><dbYXM3)D&}i~p^E)@EnjeoB^yw@b+Umt)Y+>`P{(mAAc8
zzi==YgTo7SI$OvI%r(cXs~h=bbQQH(S-7tEBHj7m-b&yZdT#GKWxsT>KGsoH?nmBA
zr7FEL697*!W~c5{jW-L5V@n0wdTvRzChiU)cEk4nrzQIZtmk<B|6}DTpG{6ahNs3c
zX*5J-h@bhI+WJE!R|)xSeoq~Ulx?~a?RY^xtH`sFGZJz#mY-}nU2Ymr)JeO6@vBoN
z>$v3{1rigtk=D~EHbB0c>6pgZ{s6~evBv1~3;aG@wHvRZ=4ZR$DI#Pq?JrtpK47aT
zFjXT33N%XZ(*d`=1$ngigsX#l)ZdWtN(N_On3ZLUv?=cOHO0Wmo;Y^(wjgpAU(B|E
zC`)rMQD{O7tHuv<&Mdb)7a_-F_6-!T8Y!smtKC&u`T);=%|7nP+b<8=ZN~-igZ}A%
zarvPz{tqq#amiTMEd6U7%(@agELOZYP3n#Dn|dATa>BNx+2H>p3!Wc~v%n4eFml0_
zS;f9Di{NP?j|E5P#8=Mm6{#rigtfK>xJ^;nf=OJrC0>5g$9bItGBwWNRfs0X(^NQK
zuRxr1N{%7!Y%NR5n9TGhtLS6A;x>^B{%$?GvAI;~Aec2ZVdrpJ?=nF-H01=a!A0~h
zmt!6uF%DWKLA3D^&)xE-s(^7Uv5DNxp8t=|lv0*LQfHvsofq}tz^m(BplhSHieA0A
z5qqv7qDl(IBR7R0kpO4s9^!z^;e{Hef=sKthr;z*baYuCs$IV+zq(<POxJ}}#%2FV
zF?bNagC1sWp5D;~cm+735{B`E?@CryhGR+isphl)<2aAqNYIq`_U5W;)z)1Hz(A0a
zRI46RXhKwv-g!QiCwESGRi0X8Vp8OliA)07z=|p;Eb~Z~M+GOnk-b(=U)m<qTOT1m
zWTA|c8m4Oz6;01CsH*H-r;z5z)phOIjtNN$k+k@hA~=+w`Wprz0iE*IzAqGcSS@xT
ze-gfQpXT0E1lgY*)1h<9G}=G!|5vu*xe!YB<pV#yUwL{5u=3w!8(&t_;sY90DqGn0
z;osRvdFZwuCy3puV81}}w1D(tAN-d*&H|F<>C7ZnggQkLBM!lsSH>mvh;lA_zbH95
z2H#KPIN|7;$q_p56ba4Bl&se~&WF}#j(($dCntDvc36)0N`Nr$jev7QH5fn)Z)gFm
z_FO!tiZNno4aJ1E40$1E)Aq5Ctfj3IG~4;O$>84j#`t6b-A2s}S)#>z$)-B^0GNlN
zxuQ#`u~yHC-(CU>C-pFX%d7$B8(@|2sPhLz&^U#D+*k(33D3(7;!dWh1AAGeC{dKo
z?pMyV5pHorowZK<>xXt?FZ0Q4Xw?74p^*3U#quQeS(g^+hsX~$d!=$sEvt3Nx57&T
zzAQL)nhAJJ->d!D_z=lg>(#XJS3Y|6m7Dn>eWChK3{vYMI3i(n5FdCrt?Y2pQcN1i
zKB57CG(<_*cZTOD^iI$N<9a3)J`dH_U>v-ewc!!J5pvLs+M3E){ChKHDWwpgXc4;8
zVG<k>E{seWON2ai<>~6p7@A+&LT>mif}qe9szGeyrLkjv<pqM=CzjSP*b_$wOqA5f
zL|ef2(wKk2mno{JImP5GEkF$^>XwmIeNGA4G%$MXOiFCRDh-Tzan@fOlJ^^Hjwkez
zvq=DOEH_pjuAA3Snj<TCU%P2O0>4D<{2+P<V3C0@6CJQlA}^@#rrTz%l5uc}YR=3x
zNDUX*vD6{U%a@@1xgU9gm|R&u@$N<^RU`p7$Az<97=|vnxO8etmgYFlEWz-CXJ_zC
z*5V*oZ*GA6y2s2T5%u)PYJexbtr@4pOOn%wbht;EBYBRa{cz?Pd(+059bR*p*k(I1
zk<517aVo0D!8kXxWh`jX*)3m9$AtyC&Gy=-;y1WM4$@#H#mhWJk&co<f6Ug8gK8Yd
ziurn}y8b$u2^xx!VV7UoF^ohfumNP!AA$XsJr)}`$=7J+&sJ0RDpKC75a#ACgA7;y
za3;;Ap^V)_j4({JAkgmq+O}=mwr$(CZQHhO+qP}noPQQ?G09uZvXV+_cT+bPXOHCO
zE+C<SD@0~2KB#+HJ-1_Zf8m@nnX2HdoZD!D?AqAp%RX#VQs|J<qXd(4t8)s4<xpd#
z?Y|Gm8V2Xz#t8-W3v8@b5vbYj6pp9Xv&ZXx%omqtIgRBQHo9vn!scYNlu3-CDET~f
zK!f7`t1=kkZia{jGNzE(OiPTm>?RIO9a*2t>*WYJGY&H>S7ohTIhoc?>Qs&71Q8io
zloa<OiNfDp56|C<e1YHfTTs)1pX2n1G-%Kl`}^qPFZ1!KFxqagdw^yFyLC(qC<V;S
zkL8|jmmeY3Ns-I()|aBbX+)$$8T{^xoQ(<=!Cj^|k3f^M8@$~#W>6?MZ%qm0zI{4D
z74EtedF!f5?czat<un!d$)Qsct_*-n!x(I^R99lhcXMLXA)I*1)HiM43buwf<|{nG
zACYf|vh#l!>B0HE3t!?cvPOuWr!i#f(cwerV?P{!>Aw`(73Obvma}<0kkzzb{MHxH
zva-6mv{jC<%~2i)5Mqj=+)}Gd;HzJY_yMfEvYeZNk_b!!mB!H_sapv9fT?~V^NFiz
zTV_P6DkQqLM!Z7D|J3&1gZC%>SZajf^<2eZE}cwzr#l0g0_U=#6xUu5QgM745;?a+
zY$U-J@F28j!#EfFC3^;@lOOR7IJJ8%N>Ec<DO)&%KMMyaXghpmlyN>rqA-3px}68)
zgpnBigqdCH@~?W-g1O+Vwr8M%kaheie4>`6jtt!^09~F^L3by81OzGUu*WKR!vwH;
zw%@x*YtDw@(x~WOq8WZDCVUh(H|lyj^4j!bi|s4T(K7@@nQwvz5?&={KVh<#Zt#6Z
zbp?5hhu7e<J1HSN%@gH+&5VjUi#6}POH!y>XXRl*h5HToA#;(<hr`|6eThibQ2+UX
zgOJ4tl)2CGnkZ_BZEzmJeM^OLh);l)4jtclYlA=nTzk&%4h;&$co*jzI&Hi{J=F7R
z2VfV{?#D==O92xDd-oAn*2w>S6y;S9$#%nv!V^abqz?9gvL-5pw;G>dIXdrw8{|#V
z+c})UQcPT@Xu=>Bt0r+N5bz%ym#wUl1!zw<m3WwRIGly><&Ou`E1I_oR=H5EU)>|C
z3%6sx9thK(`*&gm*|~t2`R%$<?r4`PW8JjQIlDO04?aJ%*oRI6_3J;K;6LPeXunX}
z@kl9OFovi%CR;;an_@uY3$YeONS3UC#)s?u(*XgE-^7J8K2h_g%~LYw{(%`RYb%6_
z`d9`pu5ytJI&DLCR63=M1>W)Rkq%b7L`O3kj3oQ?fCJyPfHi3YMEa4Ct+FN6054%R
z=73n%xH(&=da$0z`uYY?Io#16y+OwSLJ3nSHFvvt3^=Cr9MSIsH~%TQN(r)d4GR|>
z9Ee4BH6sO%NAG@A*sD{~G|b9FBp>ojGs^VQ_OwynDog3By`<@IVy`&vf_7(IdO;oM
z-yJFbH|85DqthQ+54JQf|5hV@=pgXqm=<f!eQG0*A3Mmx^*JyQ2-&Nf@=o9gqm0+h
zK4x$%mQaT=v!a#;BpwaxDC@K<y7rT#83^}S3%5m(#~DO4$Q8m>4BjOXlJ2G+N_s$Q
zBjyY9kDxc!CgzOJRKCQ?yE<O%#!oqeW?tLYfUW#)71;4d;mvXCWp^-YjKRnM-UwxJ
zUHT;Oh8b*{>d;FgS`G&ecF$Y!h4ScN7j@E&s7k}d<3-_K1g^6z8xfPuB)AAVoVHrQ
zCr&i|NMuHdyD5$%3|!!6qnRmZ&u`S}bS{$r-zi|+RZSRt0`;~cstPDl0VTNOg)IC5
z3u*yy23}qVG)1I6cB?`F%YvxoGVJl+Yf4h3b9dR43Z$bV)2Cq!o72EnbK{PK{uA2i
z+^t|LKmk}zL8z_D8z_4{Zc*E`i$l7#xiGQ}<1H|Nc}mZKK|*Tb0##BT%8Ji6$*X(W
zm4|Ewg&5&uBOx`<u<Ua+QcW$(O56rE^(skg+@3K(T9{ET{SSq*vJO{kmvpS<fiscm
zqeeXo&aLlT@w(oCB$A_8CHk}k5%mzo=1UtEPypNz=_h^)Om@8SsYI4P+|*<({wx$n
z#(Yu?l`A=O+nyfHd0A5kjAGpWBki?9&)u>`4wfpLL0gYfG(h}%EOQ(bp=q36S^=)F
zISng}Krs1bp<fIPs%>U{@Fm^Hg!}*se}j+Z%K~f}2K7%|ulXhQsb;xg+NbN1^9?@9
ztOv?pzX~!wK9<|(al2Tn*gKwww6{a5P+j@Gi(tvX3!o-O-G$s<l#3{WOP(0eznQUg
zIeN}}iz01H01C$_m8;5kZRYgnYI%>K9DHelmfXXHSCh{Bc19~>QT;&-lN6i4==8#O
zeO3&|`4D?+^o&J*aDx$(u~HF^I*ud8Q3$u%54p{Oc`8|tv<R!1$g4ckVxqLQhm>M|
zR9()UgDYjs9!RVqY7cY~2wd-6T$W|$L%9r;2-<Iy??l{AmutzL>JV{UVtu4+47G3f
zWvNEOX9o2kv~|=QHAc8KtQY3*;=zsu+Gw+P7R93wq=y+}FQ9b0-ppZ7t21wh=sX&{
zkz1Q=(0L?b_y-Dv{BBKk4lUO={(Z}+I8h2`d?Q%icSGK+ZkTsWr*$5Ls#^js2<9#1
z9(a1>nKMNqcca0N6zQX`4fEL7qe)6&*!XZoPRhOcQy|@K=&jK}0}z81EWpF`E{+bF
zS7{u`6Y`%fi~E8=t$+E`xuAo`l$GY8{3KIQf>-HQnSYZiCF`n<7m0lG_wsx6r`6AT
zBHL&j!nP(vuipey1<+wZA_3OHSYM7e1Jx6Lb?&CQCck^U(RcTG=OiC?iO%wl%tssg
z(l?bwecwiVSx7+B<b0A1%<GWL%i(+<5Rmo*AyL~6gS-)mr;oJvmqU~hDsKV;!Uh7>
zp>2j0<l8k7u&0F18N|4(wQavW+ad@%hwIB;nwR6hV1}3ijg;VUAygbmSVw879cGv!
z4<(~DB6Us}pZob|Rd%}GFP8RatK4<hUb#8hQlFSH_BWg#@@muC=Cs&=3G*I*yP1Qt
z-OD)YT!!!6^a5SN9ZZ`^z1LVKwr!2jy91VX<nf|#$Yc#Z(Hmbiks{?sTMw?1DV-RL
zdrxE~*c7EgV^k?`Acz#tF}N@9?MLS)=`rd!-F-pIiSKNY@C_MjliKp27XE;YYH((y
zC(!&0aPp|247F-taawf%Or;k~#okFT8alj}^xTd{KCm&A$nI4UGEkN})n^}|_1VJ7
zfhFsq#wJCNa;;lH9YMET|JKRD8t2N0S`RmM9M@_SCEWW_bS5aTRC;@GP+yh!_??}#
zwMm0J@nsVJjlcQPK4KdyNGJ!`PjF?c9RAP^(Hm%%_fzv*T6Yp{<squst-zy?wMOUA
zv?uvs3a5QvHYa={j#O!UxJCoem9XCABBi~>iRzs+$P1?+P!fb|a{ESu6AZ%z+xQ!5
z%fUdkjuLEzs#11*M1VJ42~{9azS*IqQ?SeDD+w;x1LGVHmL13qwRUF3xyxzoM(q$c
z??o74?TG@9Rw^E5OL3vNE_vd+iteATt~i-R%bjSaX}`%+jQEB)i>f8U_$4+%5aAM~
z|I76vyn<r^#w3MRFnaI%kqssq&A-iL@q<>`))W~UBtiW|i2+!)@eClURCoZYBaMn^
zew6^G7gfk(SUM9Yc!r{m;OH%pAmiG?HWCN$@&cPMvAf8be+pFQJ7~dABIx!nmv{oY
z1}7F($V74hS7|H+5qTTga}V#3NCJf+WjAsJElb9}*G743iU*+QqI3R!higaTR)v2v
z*+AzCg3nvR{92?<vVxoXQ?!koHQfOfD-ZW#S8_1saf}ulXW0QK%?`CJmSjAbu;H)H
zKf-e(K}5HXCZWjUbad54W`@eSC?H5B9*FBT)u9p13n((|SHD6j1esX{b5ApCDWv%C
zI`^Uvh*i}K-0k#7Gb7U~h<CP-bvY`(57d>x!=xUZ%QlUX!?wU*jj~g!6S)SRQu@Bv
zfg2&JgX6)oy1Xb1(<~Rs<t)}N6a((!1ZIs~EKG`bT%@Zjx`$TDw1mISa6BU7$7Yl@
z^AkrcxIcxcXZftu9c`+VRcCBZYIN|h!H=SPfYeSwj)~E3%nJUfLA3!Yd+k`zCgh7Q
zsev}+WNG4Y<@g~7BdXx3ou(piW0<5sHAw2TMGVP(;-0kRlgxdbVSZIcaR>=e)H7Pj
zk#m%*_B))e2n)t~MKsm$No}d=gXotzfX_DMpp52P7j}1`&fUx88+%nPJtN&@Z6E45
zKX)k81^h{_0Cazoj{~@k%!{F_3SYqm9pQ?;oVSqu%@)x>{uM$AE-69YBCrlbbLlJ}
zZMohofJZwkKwWMhy0m*dYHlHR;96`zRWHHiv!ypF<98O#9vpPo&8q2#MIO8OCgdh~
zRQ_Yk(cHX*$Jd}56<5~-qecw<?QKT&r#jG(=q}gV5oEp}jHF_EoTvj!08<dBydFtr
z0C+>0T#e8q`RiaYHxeKV*mEq&SNlCTaRkA;rTOBJ9}Uh0QANO}i|Ub4xcI&z+;0!^
ziAd!>pFY!GtH$dR&fnHwhO`yOvSx8!fny0ez=Qq-pFD4TKC&;#a=4kj8$i|E9cu+f
zyD*opzD=xPY42|Rp^R@n@7Qk!`G?xqI^b_jG1bd%Z?>M3TW_TVnbsk=tIbDmY%5j^
zU^5Z#3l9CYENZ#({!=&eGyfD@lfj79N3&gzR*ti%Ci8=>)=(N}^SKeVtC@6&IX)r1
z-an9T-|-R;%;*rm_G{{hB=UB6G$Zy#;)-Hr*zVTz{)_bLI6u(_SRT1Iczq<{WW@UY
zFN{8HszyCGK9@~Z7}f%)ikW<~R8OHMCjbjf2vesCdUrSp4g)qSuVit-V-~m-2=4X!
zV^jPZ^MWe<LiZl_?V<@GVP~ofj|1zc$A;N`P{x-he9<2>%ufd33ohNS&GL0EmyCSY
z$dZGIqa+=Kc-CW98jrxDADzdaGAEE!Y@da$T4SU^omNu?1-Z+~Ebdwc_@i?G_T9zb
zw@vTp{iYxsGUSJ6^-kw`;U11I8|~I%MoqlGY?N(lHv3rG11~>kCHh#v@%B-BTfw|l
zFI|dvu=Fdn!v!QZI_XaYKPuRitlv3OqX|9?S})Dnn~iH8Z8RA|Q+q%Ak>@a2)PqJo
zq8kAfx7u^YiFaWbt=Y}u%qDo6I+|*$8JpzDCG^YJ&vC$Mk=bW<U=QN^tGRUntj}V6
z;ZjE8+$M626`u%=u}3K5Q<DDSjoAGXAc<BIhOik>6YSQt5Ek@zMXH>?u8F_iqpxgn
zD<B-qS=l*;J<C^o+YiOFR%hm*$dZjGcDW_pK!@cVV&lbLHyqy-Td%(<UjIvO;zQ3L
zqu}!pGCa5+#Pl(m7C7|qpX`5AyI5<c0@95FN$lIOoK5?}<S?WihD;JaZfZGP<vsht
zn3hY+vmjLrMv+w@9PDMpttQtV=tXcu{{?IT--k{&!v+d{2AfMr`8&2=fOS2WK0U^!
z)H)KVLGw$oJ^0>x8qh<Dl?!?ZgDsNc(eyd(>qQKws*7BR34Jj15*1N<Dx-ubg^ACW
zLt2`g3~kp*A<)Y*s>8kcbYMtW{?TgoOVcBea$wu$EMo+s&T_Yk>pm-mrk?ioTx^p2
zYLa3X5sb%Sj+JILE(1506$W6!``*oFEmoC&>@TMjPz59x9snq$3T!%`WF7>65@${-
zjX(t?t|)7Lp$-ZMVo~NhJUI246G_OcNV+%uN?)!E9H+qp0T@2JBU9Og1W8n{>w>-L
z`)bEJF)(Hpu$pX^%AA^6)Gei4d=~<o0JP3#vc)EnSoby3Lpezavd}Y!@J_*5HMPZe
zN<rFTxo<v?=^mYt5^<IBHbh~xr`x6WllQO$>d3h>tLnEcIk%~c;VQ-@B$@MPXa}oe
z(`qn$os_0pQ3-b01Nz}}0fi#X6=a`yVD%Tf9y{_tN~o$i7irsf6Qa1bSoW93<Z=>3
z^6fALdI$Oc?1M16^b7$)k0l95+^m`h4Wq}^HEl^;!vIHkD?Rsck{cuJu?QdLUq^5H
zxWbjp6ku)!g3#anyQDzgHon8BMaA#?oIi4RRf91_^MY=)hJhP1-hz%XbnsU)dvgU`
zK2@Sh#s{-FYvjEP%uQ$B@WOGhJ=E3I1=VZdZZ(28S>WE|pUGcYe18*4R!Kc~u4D~G
z=A7Fls*oomRPYQY4Cyrs>2uIyJK(j(Rlc^+Zf05<hoM%h*j#9<>DGP`FLx=QsGRlx
z=G*WbIh>=mi;CWKT09gvRwqxL3pETKuh)gcjc5NW<xDx(=HSZ05D#3ZpT8*09>LD@
zkJ|=5X@np&g`xh39j|5NRxD;H4U9u;xamgI{Otz70(%!jY>HNjJZ^3|KWuGinb=<5
z1T<V~o6M!W@DY2wTh^vK(4y%CkaqPFhPsp}8OKM^-(Hrn^3l%TwM`3&dsKEmnps*_
zgH3UZc0}08l7+e#03B3l%~g_&2jBqZ=O@3{de@?}DlYE?2o;~tJd*MdP6S*d4Sq;X
zonzcN_TcpssC@fUbahfI-u?qo6DZ<#QBuZuTsLwb?n>0RE4B(&e#Cz5Rn3yz9DOjn
zIO<@Yd#6+$x(!q-vm;X$Y}#^#8c&84h<DCC=~$%sC1*7lz%uc-(+7yBP5&oHEyD3(
z2;;(ouF77Nt8rWH6=3^$0JohSp2IiSswMBIv5ZJdL}TrfS>Q+GFI;B#AySnQRzOmo
z)5nGfzVp2Kb>3}Pg9ZLIls^SMaP=$wJ_~YgJSi@M-hX6g^gbYBmI<w=elt}Hz%igR
z$@`(qKX*J@!NBfs(1|O;4cR}iXI1Oj3-pUObp>}^A;zmpe<t)ZwYS_e0*cpN7l+<G
zTSvK=t>mVKFQ7fG*!Y1AxPJ8rL{L+ps2%2hqjf3p(X42OhKdN^leLF3ufebdaqNX-
z;FaGvq8h%FK|w@1qw@P2!p7|q7vSL4){W`PQe*<7GYKG)=k|~U0W{uFe@dY{$|q9m
zK1SwxtRD-({rggsXyt!Z<LX&R-dwGg!E;Q~EcE3+3#ef(a0qq1e)*f5kaY56WDV<b
zQiTTDq0I!GTrv#4{)i23FB29Qa|G#><2zc^gN3SFa9dgKZhBM6k_-kpeO_z^PJCXw
zztsK(r2Ia^To%NSnCu2e&}+@x;^5R0UyJHL=hu4&4=FlSKQHuoR@|0*%Cl0|$}G`I
z6YD`3T4qBka_D5&;fp;Yw&Ox|rCp#l5wy|+YXE(HD_s1Bqr>%jRf-TN5Hn4(5A>WS
zhma&e>?p;u{kD=A++Z_0VVbma2;KQlh6F(`lr{N@DX7>Uzo13ALIFG%<s(A_Q==GO
z%9Hv`D8{t^*wRd57rIw^n%+585f*iCIVO#$sZR%VC^H{`6e!l{wSnW1`zb(I#l&S~
zg;-TWXOCzHaKi&C+=9duB0ya8<f|L#A+4HVX>Lr$b*+HK2(R+yvVurh#LWq|iSNI7
zpcmLt6XgFBr{V$BOTTp`{qVp#J}k7sA&k5GLqVvGhq2RnNQqG73PEj=k{S~31*HAf
zseaUJd=W=EY1VbD=^Jhi|LC?eazTrh#$fr^C+}-$NN9Z)IIhgYA}B2OBw@+y+#mz|
zA^KOJ7GSYsa0G$zb3n;UN-$l^C6mkr<&+pmMwq_H3Yz%O<*5*SU&IsveACoLS@szn
zFl5+#b@T{R;z;S^pds~{{r2|91z#ntgP7}T_<p*~_1>UWF}bi8HYfU0;fqQ7MJfq5
zpS^jJ-g`;OGQK*;%*JFni6o9X8(1xPf-M`x#(@NM%QDl#Soot%;GG9429);IL0CCf
zf#~zNAzW^FYM`Iy+n<@MEepbiw1DTMpmF3nJdCPs4xCUaAP4!EcahL4=IlGS<%9?u
zCh2N@(r)OMbeEGk%7vlW-i}RIB_C_^*=&ull%V*_5u>8rl1|*KUSp@lFDy%bLd@y2
zCg2gBw&{v&=jH=|#|$1kztGp&snEcie57+}h;$E}gU2>%r+_@i_vd<i$4Q0?JoBFl
znVj0}1O!4l)+Ku?3!Q7rm6H28xGJB}H+gL^k7TC*tik3mv0eA;id)5sQVl0J62%AJ
z>%MMT=FCR*rj9g?*6*LZ5#oF1_D5Uxhk?nibkuIviUs*m72GF%&qdmndG>ixEuh>r
z5sef-2h^N~DL*ns3BJMo2q%v@=I%#6$EnoWwD%|Nu^SEYkV94CK1NEdOBjL__l9Bn
zGSrm)8YH}sVqPQkDgupW?(kj^UQa0aoE)<Bi&paukVS5UoRJqvD52@GI|7m@6Qt9H
zG`8gbYDlpDuZ9E@{r{j5{}~bt%p45=bNqh{2}UL+`u|@;qV4}05=}JNT}Dr_HeE(r
zr#M@!He2WG7weboIBwQk6Kiu{zFv1uH%)#yPQEoPR?0Dz8bznwtAu2#$^~eQEo`8X
zTU#CT|2_Etu$;_-9Aj-AElqP{a*=Xl69eEkb8UZoOeo7wgeK+1r9gy+56Az{91<QR
z6F^-ZV0wFd+x{2;!3#`|Ov(ug4PZqN)>2cm^RMyy2#Egm8}8VyxT&NA$N%sy3Tu9J
zcL3G=>@n@|k6j+*KS4qr0MIm*mES)+urM{b8`xh`P!|Rtu=yV+#s&bgPK*t;3_xQl
zDL<DPK1=`fE`<Km9Z3K1?(FhoUJVD+_X(mSC^}#^jz4Lt-;h5pDSwzBKBxb``H9uV
zjgLF7{zVM^TVvz1$2Kt9R{$w4BrLThKgf40wx&Nd0a)2rKOXBpL%#{+n;e~NnQUqt
zZGis-iO9&mE?G1vqBK88F0J1Xz*kd09+?{%?Z1u1zWP39KLs=*DK0aAsQICjJnd0_
z|B8~PpqPZSy=}h<>}={kmC&lDlAI4e6@D;#piC@|j7@-?oWI5IlixC~zZSXvzeX=2
zFfgn+Kas7!sVhFKdS?gF&2)x-hNkZsi^E5@wvG}U>?6J+LK8DnIDOO4ww>jb13r4T
z{G~oj;g`8-`9GS$@#WpwQ6zmx<{+cwHn)!~ME)1QQpTY_u7p24@$Wz3{XeqD-@S;h
zJ(f>B*oVIxkKeoH!&+9>N72@PKM=s*qyMr9px+MT>t8lIA}uMbxxb0OJ}*mvKb_-0
zaiSB$=Qb2U*Hgcf{4-PiBg4aAHvJ1~+Y3Num1bv#27q*ojqjV4zH3X@QhqWlDXx7z
znZFfT03If!CBGqM8fuy9n_mmwuRenOoD09YzdGr^%3K8vVQmqipjW>xwF?#N-`O-k
zCa|+UBIG~Wz^KZyfAIHzWlQl*Y(H}&eBOhDuebnqo~E#{06b7=S9|})2ch>(jb486
z9RH%iKl)Z3MQLSIaRG2hK}ktZ`~KU{lJPt7lLzX*>sN%(#MHp{W9F)?9epkR7rW($
z`5Pdkos<1r`CW&#fA{)l^iwzRA1dKbYG2iNBjYYq&DhAKbJ_4HRx0cQ`k6i;CWtec
zr7RT*WO?KD!$jEHQ3bPiC5Vw(=E2LwohFOJ`&wcFJNij)L+-WYpBD1oTfF=uLY{@p
z$6)~%i2gFwXOe3%?5;;MX)9pvL|=h-ElojjmDVGKApK|T64>Ec2#fDR1<6n5<;?%g
zBw-glb}ncODo{hm%TpxI>+4yJjfsbYaoZDcy*E}g6BR~s#xnRQ#X}<N@WAZcX~*~x
z?CoMC=gul?e7#tHJ^tJoZ9~km_sH;X4>l}o_@k&$kq&Wfq>?(G#%@hcf?bQB<?laK
zBKkj6VvzhP<epu`$(3=I<!L{}JitWlrA({-Ry(nJ&ahH<?I|-N(R>6mqIYR4?8yzr
z5VkJLYH0R!f(M=y_aTu|&}k56$rE`G(Hxl$xSg#Dov=)$C1J;?GwK;jAA4doqOp`r
z8PiOE!1I0+KLvDpjOB`7yJ`8tG)cU290t*~kD(JOt;E+r-KN46qz|KK+1Z9r63>SY
za|1@fdHLd`sSts+%f;%t;Gt3EnV^q6cP;4`H%3+r@<ds%)6}HQGk|)(^<kcs>MU*E
z8D&ASh{nYd)I116sE=5ii`UdQFp(@+KM4Ea9v97%7$OWa1iU?tszYWUt9jf<<T$cV
zZ$h_#*790#B3(udgT4BxztMt?v#tcq3Bp@go<!o^uusEk!R+-gFWu4)fN}W_tCTP`
z_AVD))}<H+f3bk=`gH?X!iCVAmtC8~_-1Z3krIc)B7ayC>AZ?!Z?jy7!XdlGfD)kl
zOS6>&Ir&juB+Eld<Yu1}`sFyTgt#oKXKL4it2M1mqRmkb=&?Nb4|i2*v^ZZiQ-2-g
zt9pkQZmj=x#>y^`%Ig&6^b%l#uJZTCs91X3b-FdYCpe?|#!lozzJl}X%NFcD<6vj!
zJ+(q@9V(e8rj1^e*f7+%LdmAUU8o5|o6+PY71(IhDoxdF_v<%Entvk)kOfQ4LWFaj
zXKF&FDZys)g)-{fVK&hEy;PGicIT#+9Jfh4#<tf*IM(j@cp0{Hi%D#(UykZKqDrRd
znMNluu<f1B_7MANe8FyygwWVV4=g*QlK`f8<k}mNVqE!C=xIyWtS{~&a0^#R#mDEq
zCOGJ&Yt_>4KT`r_^AJ?s!dHrS`=!OYYCcQxyJHtO0I~$FQ^o@Qo)Ai*^Z9g(r-($p
zW!wtz!5QgIAF@-8Ez}q`F6>ZGA^ZjP*v_uVP6q*2Lm#6Vi_CO;fssQ?OuU15`MQ-S
z+O^;W{Q1UGeE_T>N=og!i98r$j<-w=rli8``6$u%^Zs-WS3DK>iupF#@b!WDpKB#E
za|1M25<>`pkzW3Ug<gi(Y24d1*8+QN9_=_QF5)dXk_r9*u*pC)T~f^@p9X19nrCMS
zkI@RM3rRv>1Z)6r+eX>l${?QTf;dYJ`)ooW$gQqvD}h^wEzuTMvGbo{p}ljUh+7b2
z<7Q)j-rw@CyDEdO95Cd;W94ha@{)=-ucW+^>aTRvO_b|RQJI71Z5d5e0se(CpL#rt
z2k!;&Yh<{$6@i7f+ER%<*&m`<fMz)Kvj@3h$=str8b50@u8#q>KnSoYOD7XiMHv)%
z3)qX3pvLm>GEES6W%YBa+Y^G}TXd>BVay?hBm-u3y|?@jf#<mvyy!rIvl`l)tP>}c
zE^y7+GlJw<^1JKex8|-5&H?d+HcD}HmE)rzGG_Y4;zrw^0kv!+LXVV_rz$2cr18}y
zsT}wws8Tn%BxJC7U<DH5E{n^1d4)~6x}ZR&Z}~~)`+QujhTX-U?KvTwTY9^amzwp2
z=8h)A{KmvOwNc59T++yV@HZ+5W2gLd{;mge&3r<3Yi(UwWwQ2rLQ16G2c(;Djka8z
z<<6~NTS>79->pm;2wJtx6c{mNz~i4chaog86YDto*2e`xuNRMdzhXAl%HyZ5LQVoo
z97_Ovtk#U1QFz1FegPM6j7n_U_`=hw=MkEhVOlWYYp4?sKQP%H67Y$`-mQ#`8&*R<
zi!gjbGO{Fkytw0WG`~VDz?bMt3Gj2Q<YY$4mTxlbbgm3UL70)Q9<fyJUgW23mFIAx
z`jlmLPtP;0;{$QO4zCb<8xtTdv#9BA_LW1O0i{Ks3wka5vr2q2hYiCMscSnO;P34V
zS8EggmoJ?wC!^>!v&JoJA^?F3D5n?i4oC42>oE4Hpa~D-k%^;L-~#gB0`#m+_Gdrm
z5G>PQWt}?-kM1OvblperbTDR`x8RPb@q`XwS3PJ4;#QbR3jZKG=nZo%H^>Y0a$;Q)
zV1keb6aA_`x8raHn*&>^NGMWyT1OF5Oa_-@Z|&Y}ZSKb>I|sbYxhV4HQEA#ipuM+3
zik}j;RVFQHg0yJh6eC#=vb6U083o8pE=Gkys2(P=`bb=|-vFnKjtq|R9Zri97D(<!
zWCHO)tO)5;&`}O>mjzsK11vj8?ZQF~QCe7<Dn-l5vJG{>Ynto;)=oUOLCsb{q7-vW
zTGkG^92kd0)gM66JHpfCd{2@-fG#Q@>;_h{l=XvEKi6<=%nb5iOC(0ecqHGIxXQ(M
z&o?!x3T+Ks>d9vL&erILQe_G{5~*@1q+v3M=hyQ&nMnLKwC$Qg_qJ6&F9KP7*C}*a
z5Ryg^D^v&_2}pNDD+Tb?5bE0Fc&D{6-<(6CES8nU4NX&bM~O9$Msem5N=MZOT;r*v
zBNSqHbIW_PTE{11f&KXzOa_A}n>5Me!H6iZIn$alBW(aQS$fWdmpNv*6g|&Z74hl_
zH`u2c;<Cg<rmmsVt$wT!^ys4Z!^@>K@{u%&`U(~=#pZqueUG3ZLbOnRH$@Q;AtE}|
zUav@SM;QiIA7!^ARth(PM1UMU<^}MsBIrjk>~XZVQ!oaq8EY#+&%6ONumM#29x^5)
zRq>_y`-LF*>4paCph3-e!Ogr2&<7TX7vx7w%{0$MuNMElu7%9ZI44y?Va*nMtDyA_
zGYFLL8I4Cuf&b<*j7dn6Ku=QnTEd@WoXyVc&iAsHd1?Z}6tzB`$;cjvw*iG~_%zkI
z$CJ%7uq+I)lPk3V)7T%+-fTs=H^w<xNx9$2JaNeJip;`kv3MQ$jgMl8$F<C@W?xlp
zl@s6r6(TQ+?ecr)8+&l5aGd;JPRjH+9la(j+$yEg5xbX7tR#decXl#%)s755Z|`io
z!ZuHvz_FCYMyTtTZtBp*7nvpV$|9DmDxuov*x(aUEmx_9lK@KDVYdx`S?LvxVw*x|
zG<=g(k^kl<0ca;%b<_@TzRwD*fH{J_x<8&Vy%_NFOJv5<G2U}t>>7;CMVeqet2IRQ
zK}_&R`y~*1SlOC^1Gn_9vq4fjy)=nD37!(yC1LpVyOBy+4uFPCEx9&#tZ@)m+tzY)
z6)GCm&uypb1BVkdUnas_CF^{$DlHFAd<_!XI^zu>X=?SG+w*v-Z*wX;afcBut|Qq7
zzH%xud<IGPDGcJF=}p_dz4}yN4s<s`HE1iBy*GzFbL>Leu}{}Q{AsnJO?eqn^Dc7n
z=vL;_0Gna2!eUAawiA1Tio8ZkyNZN3Q47f~0Ea6u5Y70TH2*i&!c2`Kar|f{-?}+X
zfn*2M_#Bb(UX{E}!RJp|JWa~a5YTa?wwh?Jklci*9vSYS{3##3L`4n>NczR<rwybu
z|Dw{Uvud4_3E>D&S#T||rjiDSh_wP{ilRmMJY|+(_PSpp7dS@r0sEC2pYw>-B?oMv
zeNmPp1ox-^x(0dmpg5d6uXq@*^*K83;f(dJ@2@RkbLO-9&2An)Vz>-Tz=g@B1)?zl
zy+~J7v6S!in89ZT+CeGX@PXGLJ@tZ-vx!z%+|~YWR2{$o)P|>ZnK_6HcB`sn`3LH9
zucu{?K({oxMJT32UZdK0Z<9VrJw20J%El4+Ye!mE;J57n0GLm%b43&Fjdy={IPh%l
zb<@RH31M%c4<n>q>C_SloT_VQ8FYMk&W^`@um1pTp=Xk@|D_Z!$Ck<QjGkq*r5x#?
zT=zLlyA<k^Z1N&A0#b3h0j!{ES~f<0UD_;>hyK&fzec%P6F4V|MWehl@uVAu!T;uD
zr%zUWUC1Lq%qpgGr6RhHt&Q=)4j~2>oGi%(UL3#!PCX;gtd8u?1s3Bhs0R6Q;HH9j
zg3xV4N4vfgl1FL{m`tM%`tvtAW)!u=4n^D!l1{R+yOS0LPn!|vw+}oWY|TjT>o?KV
zkm)Wrw_2Byjii2S9qGrOn7oycUpxg#%976HLnn}__J6ew$Z3X1F+Wcv=xW?|p9>bV
z60sLo=CZ(G%UvY~^*~Juh8{JL?=}A0YQCN|%Bfba%XQq&%kBr>QYOd^H^Lgb1B+kQ
z|D>2~T69ihACz(j_-GWih!Q5@xD-O)H3;IFzo`L3xcJt<&GZ2>rE!Af)->Ueqwp<M
zRPVc}U(pzIDk<bDhlSP<z1|vvNW%Y~+^I8^78HiqZqSwI?XeM=0|O<n^iS44^&By2
z{~vnszr@kQTN)89etPiAT4-D#v&p<w%#KW0RT#5rfg*}L_+UZDPr=5T4y)0U5Cj)C
z7ck)%C#7}NEIdRW>HM0hypVj@j;!`{$hxZ`Fq-$}ul)bq4li0wP4ce^xZe&uneA-@
zPgE?CxyJ|Dnu5-t1eelIYmc(nB05hgMQ=65?F!%(6TZ~heo$hi#gMg?fab0n%_Y<K
zP2%NPrq)wgFtJ?rDFG=wX098wKmuRkeTdH})UdwH49LpgXXV)xXhMn{=8oB^fnyyC
zYcR#k-`qKll+@MrkNj+jx7eDidJ0ow>Vd9aZX}&d=As?jop<p{1iT;OHk2`Qn5Eq^
zNt9{)YK&Y{?j3d|JCORnT8+`j{ZLhV-w*o0G&#o}2nDl_pP2cgzq9@K*DwcGx5%8=
zRI5l`wne-N7d|&`@r(_KgWn&i#VwY~ejmFh&i$KY!$B7gF$R>Ra%JXBL_YJ7!_7R*
zTSJcYH-Me$+Gt3D$yR1mNV<s`xv7z{xs@UGu!py<rJTd@)@fQ_EDLgE9{Lq2&0pTO
z$J=7D&(KNnHK=i|=dbt(EeIMMuSeqUIiCH%ltue}EX0Unc(@_;B>cGp^9baZ<#-T6
z6<dLw+NFRMmV0JO5ME~q1a0g)rn7VUH;K{?r`H46bfw@?73GcQwGmaGC&g!fGUJ&f
z{~%_W5=Yk~nXk_?D>Uyzc-Ez5RMM9nLmBlKb{hNh5$Z2{*Co9aeEm4u%{o6$7O`tx
zzp6hO5*os?;pj~XzTIo7TOBipa1TfV8*kzv_0s|jMS>_4-W<&&qf-`VSSoQWB9(2|
z2Kkg#lMZiOw8>4hXtlhAr9_8e9Qu3WEM_!LG^%oSC@RlQ%3Q^zWke~`mMV`$@SQ^(
ze?ubg3ie2q`7JsE$|E2@%W271ZNjZHW*<bZ(c%=y)@wqugNZ@CQ$st*KW=-W)`V9c
z12+P`EI_3!R+`re`^7D<8otD5oIdF1HM(6qN*We$2wvbu4YWt|w3fbstL)XCSym65
z4s#%fz&6fQqnV9vkvl|l;okom@(ct%A>i>>#coHX?P@%rjYNeB0t^?Uy@<QvlnT@x
z$&`>^FW)>Ok4VERJ%Re_d)i$KiZg)+ygWUmhx9qZG)D~J*h7b$aH=37i!9e%%HaTR
zGKoAFDCc^66k$|KbJ#r8hp<bS_1gpf>F31aXLs2A{ZLI(+)N}i@`_X=#jYH|<D;n5
z_t@)z>|>9r<1WPkhvg9JI6D6wLs+q!MN4W`m^&CpRFZk`i52~UD3us`b$@*fWW+ZI
zn=t*>xNZ}h1U{<J!$Zd)8v7y*p$^CPBHgK>BlH;O6m{UjO#6RlM>h*}&MJpfWVbz@
zun3cAox#lA9x(p%ko*{x&~=U@%zRJDFwQl7JHPn_1t;Se^Ab0RrqOz{RnG}yZ;F>a
zdAz^7cSgftSArg}Y8rE;+wAGVP{l_H#2a1AD4v1z<I-HMJIBQKEcFP~6CILi-#+DJ
zI5wCn+)4N^+Z1DEznK72G=OONfNQs{-F^8fXhd}hy)9Z;`Hgyv__da!4e*n0?g@#v
zX4Y+d`_J5Alb+PC(f=;98YwZE4H>0gL=j-^GTJGR6?>N>IcV|1fE~L?CO7xLEYXyb
zLY0?UPtR#IJYtC#0bNp2lnHi+07qP&sd^_KI^G+0!}aBN2NBrDi|KDt(jg^7`7Sm*
zVZ9&oR5#?g8sIGU=2^s-MDY##>}WgFZn*j~#vtadwB7q+$UB!!&@t3{(vfK5fTN||
zOtyMT%L03`Ocq!lS<~EFQG$zRgn$FB_GV`w5o||rn0+S??r0N^qwiur5wIcl`W^;n
z0wW~v69fSKHV|ovz(~6g(<Q+2>UC}waa(NPLJ&aogp+PZNW|07=z1tr8g#kLB#tph
znarzLn$i!Y)WflY_>}zm2QpE3V038bgKm3Z0*uOCZR8_=E%^R`p#Q}B&e+gc(-qH>
z!fPu9;9f7R-LWbD#jsG0)`4XVDTV#yK!fI)b(DnH8(H{RaWUpux!cKL&&d{^@S;&H
zh}G;lx?HT?H*~(soSq5GJ8W2_=E=Z<r)vGw=W?ueQ!|l_V4N+zK4Rc}*ZMiRyoL9|
zNHHs<d%SxBy?ik=9_`K8AXJ27HLUx?Iz9KpZEn7oi5(Q(IfRC48z^hL|Dn^3CGi=O
ztX4*1p7a`}pLTJ&J>q4!4Og~l=Br|&=5^c*H}joI`Usj@%FL*6{9due7ZhvAXrJ~`
zu1mTxmDODxCc&`(&O>!#_-EX)FBi$|hwdVMB{lfXjwo>pWc?R(3ZkZ@z>X79O#YyO
z7%Ij9#VWgv^~1k!-`aQ_6%E{x$rVP!BMrmBbHSVR<^CB53e$EsQ!v&y@-x1Y$B^vX
z1hMvX@zvl~azCtA<ChsMGW;mxScqO33w^mTTVK^3p`KAqJ4*xrap?Da>+tSi_Po~`
z(Ap!-eS)|z*ve{ElT-cJ@&x9~ZR%L2G7Qj2)=iUwHYL@?w<FoXn#!Tp#-quga*v>?
zsbnttsU^#0s}E*_2xWokA|3^PWkXFsH9u-U0-?rY2O3GNCu-?pJ!@|9ZoYoBebn}(
ziwlVk-F<!>a$l*(0IW1MZh}EAYU?n9T#e~7%!ia5gJ25kbPf#y!iGqcILSfeu&->R
z;jQ2*7e;|=t4w2@h5_gZoP<Q|&hG&-fIU%SeUQO+av`{hzqMV%<ZplLZLQ%6mn@-c
zgbYn$VH57P<b7y0ckh&FRA6Ym*X5+?sv;7ha+J)cZfv)g9Fo$u%CecxQ%7k!CtR0A
zef^xFIakD7G_F#Nq>;WRIJ-^59F$TG`F+x3<iyhx(irl7ch7O2Gv`#oaFBq}6^P<a
zG-K4tt0rik;d+WzSSO^fr!)I$JySCA5ACX6E=<{|M;^JbTl62Ds31aC$3k2b(c4xE
zkftfhD!jjW=@d_%S~)NKB*q|&lnI1k1_);Lv8W<`8mY!HzO3A`&`a^A_m&;P&0Xak
z<v?auaGehQ#p(e9rNEM-Lr6$I3D*Pqax929jIFE-G1By48|9;(dJk;QmsVDznSbR6
z%4My&WnKjEBJnR_U4w5V)YQocQ(F|6X<y4AF>pF<>`UtR-r))BOB=PnhnC^CZV<9z
z2qzp86~^v{6wFRMQUCI*m6OEFb79QZdVK(}7P%9l-8>zX0m<IL+f^3l*VB(;P+TQ>
z?~FP$9rzg;Nzus!Gv^1R8UmN(KtQ_W(JXAc(@~ZH35`(<E|wf;Y_=5@S1S$rN1lw8
z9L<w3D<0-QZMTE`9CpXbR^2zqgv`R1e>dGVgI1*}IM;dBsOzqMh^}_ft67Twb+-+-
zhzOONr=AM*YoZiW+iKH=yX;M9Cu}Zm2Q!ZNj$N&xlUJN_XRs23+eeU{R<el-DHPKT
z6=5qlvy>x1qcn}2DRg02l-^?-<(+nK5frX5;#}9l7F#W(=96I0=J_o!Md((O-OLID
zE6=h(d&K72>Yg&Poml7LHp&C{jytd>r*H4rhTBc{VidEs`iQ~9H~56=p=2Fkg(|oz
zmWeL3e_VJm0SS9)xt#9K9SB0diTPq)agFgr!_z+B>p0GqBR1mRa1jNRxtGtHbeV(i
zx-D<j?SPX3lAN4f!@T!!EX<I%_^MdW0irSDKMYNcAS;Ss1Ko9Zd%y_%1_Z)Bzas{c
z-hsERWzP`?CQ=tD1y{T5jZg8&`DV~}vTa?Tc3m#3mg$edb@xV#S^SyBNC{(e*JJV1
ze{kP9{aCw><E!r{wR&{J3LcUYQ49;uXJk&|w-HpZHlaMa1CN8?*`7ie)l2*dj3Qcf
z(QZAmcBAhk9;iNIy2!O9nkKb?#1pGFkh0GylYe#V^a^#p`Cp!%q9rI<!n;QJBQVum
zNiDq;@Rtnh?^^+g0$sdmmeY+qo*;4W?$I7#DToT2O%k_FXVsxVF4@y|#frDfQUy&H
zyd(&Cbo=ZLrvMMIgmEs>e5`SXg7k=-<`YZY&9jV$$Nz|5uV92=GE^i*Q|~BgHzSD%
zH>Bz^EyepB;`JC%QQI(CG?KmMl|kJOo(3y=)rwxA`gdnk2b^yvokAz42QqB1h?irc
zQ!qpk^^1&0P;M1Lur}J?1$x9~O7u8+$b!0YSv``|nZ;;xNHD*5|LRAdW+{Curk&Wz
z<%9z)13w;no30@HW0l$r+$550H_#y;KACwRf$YIV@wlF)AMReW>Ie#4ZbXj<ZRd?y
z5C5lQ(p53adXv6S60m(a8cptd%nKh??YK%$sH;3ZaleaZeX5lN7OB{B<1hLj<-<{e
z^g`p?0^gbyoJ~r840BXJaS9+bA8{<Mw3>=V+AGe`fJQPTOeRt%NpiOjRK-f+6CYlt
zsGaVTG0GX8FK0EI@3PPz=xQ^E=9iweUL_{_w>p=!$au<^LwI6uz+&AyA`!Cr(N<My
zc;$y)o(+{~#^xksm4mSiw9>+0<S;mCz?4eum20<yKde;St6v~<bO;kj>Fv01v=EMu
z;EP5dgpU3>$+(Fp*~Ukst_Ln@>IHNBL?xr)@S$J-lL(UzeSwjp)O{dnav?&;QCX-!
zH5%7WU_QTeT$s_l?nZe%YU4rj{8uo+C`jO^*;c!<Ua2-%YRy?=lBa3jCeHqJomm9T
z1h~1J!PdmFYGPgs=M#bt4#yd5d8qVf)04oYS54iOy7{=?f5%%t);_$cXw{rfuGTL8
z*+L~k$<et~^YGmIJ{=Jqs)5a^r&ziNTF!tNljuIX6QPi^joN2p@gT!JJL!Z++IL}{
znjnNspqcVUakGNG>ZVh49*>Qz4nkgbU4*(Aq)b6A3w{=X6-BDqG#PqGMst0_g_waZ
zvwU5gmZPAJd4DinjLi(S*gZgjI+$h2WlHZvF_p}>qJ=5`60z5hJX;MGx1IHOe=K6Q
z0eBWetiAL`4fuB6c$jS!%&64-IijD1iD!kHw`6+rz4YyTBF*?AwzWD6h?Tzoqnk|c
zg&tKk8IZzXH%OmcL4APw^-T_;EG4HCPcWsBhNdcBVrRIPj@HJ!-77B{TnF2!QkqfY
zn`f3GSI20nEpwM*-^N8K3ddr8VW|Fs{ITOiQAJDE(1TB#21jshz72N!5hSH^nmUtn
z@j+K+eT(iEA;o0G(XJOn)&59}vs|6tm+a&n!|#?x>InWdJ|5yWL;_-*_aj};i-hMS
zYGn9gBTzmqBkkS=b44ZUB-+MQnKA!>T`O+XFIfB}a1oS)>oLx&8+{%n0UnOStviP;
zT$+fQvOq`{XrAEZ8Ho$MJ9Or<s!{V{8;cnOf|XY;`<}edYRR{qVTY?bGpuq12}Wc)
z{NG&gqTeHDh66AJ`AD6sDz13q2zjJ%_CmCahwly+$f6cs7n^zkz3Ig##_Gi4d9WjC
zr+UGKHC?y4m8&JLt*_GaIH*-rT9u8{DMlHc0{pF06X9ja{6a!7k55^_E*!5H$8u+i
zuMyrcm4|FRR*Y4ACGP~U;YST}2?3pXii-?_A#tDPv=E@2cS*6iD3_ccL6$Jv7a@YQ
zV0E4xZ@1X>3S|0CcHA$^@#e*Sm4XcF-=~Qjn^F~^1BR+ymhQ~C`S%uSpFU`ZmHOMf
zR(h37K($yyc8Fy<1$Z=KVVK}!TMUc#XD!XzV1lktN)+QBa5`2RZMjxb4IEdII{29)
zba%`H;8=KhWQ*{w5NToI7`;CnrNUq6eGD}XS9`}>ApDg}ZWyZeq8XAcu+T`k7#k{S
zuM#zDm9Q=5t&zoAdijOhIs)RPbN2Xdt6+A<b4gvuY65pP5So2=d%<pn?-t<M2#~g|
zrM;A;Z_6Y~q3#jI?TOMk;lm7nu--|n@tE=aG*~B`r$~473<|1^7oYHX>`lTDAqgEy
zk-(MYlBza3ui!`Ke02EcQY~-4hJw%DfYFv<3}K-kGImjrbj=pH;p^8l-%}|OANZ*J
z*HXnVJKsLug+|oC!A|MA=;!FD3XaSVu#dPum1UM?pn0asA|Xm!^s}u!?1&pTStigX
z*~~1F3x!E=(oP+fXm}L&`NH0*{GH2948>t0hMY`0Uh6D0q7)-w+-7b%$LX)?d<Qx_
z`_GSAY{n#QH~O{LqgPW}q%6WkRwA5>N|cLrO&XkHcX=O>rEq>$?s({P9^-$-=`dg&
z=-aK>*G#^s`G!xh6BctKv>{(6U0`Ee5Om~d0AKFAWi*u)>$$vxKCRNJS(RV;PU?ow
zn>TT^e%AKrzFMU<V^@$PBiuA|_~uTn8}(9X4{k9Lv<3Meh1k-%PJZ<Tq)t7jt0TTm
zak>H+E#IKpR3xK1HOzL>3%>cuhTeBv%w2>l6eEsKDLFjuWdsE6Sc_^NK!O`&;%XN5
zi9p(9^ZHS|=h@T0m}rgL7^Y1;U@ha{VxM^a<~dJEr;I0y+%M}p$FgIRdI^j}#Asob
z-JLOAZx^1wTf_v5dtbQBX-m7lJ57qkUJSQ1W2(t-HUHmw#F5xm&FYgZliXjW?cx}K
zbsDzbL8}<>rto!_DIw3~{rXs;)Ayqh*y%Q48{Q`71@Og0lE%Ok<rM7k(>hOt&#Ymv
zp0cuvd8x-vcs9t}Y)Zty-r?X3W?frGXOj?UnF|Ox8Klmqnw-nPGJvNK8}APSKEKIP
zv7?E@A{tcN16I9d{io2+d$b3ar$`JfkvJ^QC@SxqT@2T~gb9kl8(KZR+TRe;I=%xR
z?q{b*3>>GHG|leMkzkL7@P~b#i1E)bkNeoYBgEw#aArjXi7b1}S}<d!SQNy%?LPQl
zaaKV5e2W)P&R=(GSpx#8_pAJp4G8naB{%*8kvMtI>8pUlj#p;MNVyZdghYo+^qJ&h
zrCX6-im34%5|R~al+xCP8T{`I8q(+!DKpZ16eBzdzd#nUX`3BoEJEJ56~x@-GsEHl
zN_W@xXg_icUdJk%%H8m+;(R1t^?PTAJSpY8XVv<Jnm~qSP{H)zU+`Sm*@Py@fW5je
z6v+#D!e)L;kUl{z$;31~Db0W?txPtB?W{+dNlj*Us|m~@hOEXm9FtGLCs19wWtOEi
z&dc|RQ$*&}CFIJd=BuT}>>F03o3oz}CtkqfC&1k(z!(l{MVwTGxs-e8l-PExX=%qb
zXiM&_Tto=){oOKY*ZqovT+C3O#nOew1AO8i)=~)<!#^FX+X$xK3*6sBqg#Y<n=5YV
z3Z~pVYRDXysAt|2PArS%_#gE77JFRN&-*A{9aD2|q_RDq9C~Gi^01yO%|XL~;e1u3
zOZIFb%Ai{Se-6kD{Lqj5`BMovxp@E}=-770yzF#4QZcmUj(>V4!;Ty|+D$u}>Gb=8
z6JDP=a|Yy0>l!itBnj=-g>m7uus2C+*Q5=6t?{!MEp|xE)y{qq{WuCYTEW%M$55vw
zaoe3vWY*D{T`>*qz`n{<CS!^VHh6Z&YJj2-UhWgmG=|*ay}Y*!J>C(HXh@1YSr_Ot
zIOn}F<>G0={|U8n_>X-;449l$#^k(4i5(k1(nNV2`gA|g?nyoW!$4%3YnPD!eXltA
zcF)e=Az1l$sgeXn{g35Dx268@ex?ZRwxF>%ms^+U`2u#Ov|z<JwJbtN#8r)TU!jFP
zO;*l9u&1XtQA?UyC(L)F9!h><jFN3)o_aeNb$UD6@DRZO^;D3cDm@^Ao-p#dY?_-_
zN^;7x4zAT7I+YSkcQ0KW`pP2cu8ot}i@fohpONcFprh)&IB=0kJ^THPMwd<N6!A@s
z3tsrn16D(8!`AA4)Ad%P9yucRI_PA9NN619KLkD%k}bh8`u{L?4^5f?TN;4Nwz_QF
zwyiFkUAAr8Ho9!vwr$(4`ljz@VitEXyZix}kr5}(`@A8Dm^vVJjOq{+qB(ClK(B$|
zFDo&3F%ytQi@+)J?9U0<YEMuXJW7v10YrK_hJ#+8Q9pSS+<w$|^6;D|8SW{6`l{^4
zDou_Y+DD)8lu?-%+{VH(E~+e-GYBS21^R2Mvjys3C_L!?M(~`bjr+Mpv(4}h$h^yQ
zV#>FaiJxLAlvm*E<vNm|-3`5SZ&As^eqZzzT3g!b#8Ob#7YmEu!dbch`%b(I?imw*
z?r*|FGnm1!1ez3{WS%9ZEUju733TlwD^CW4%Gyulmk1MTPz8dAU<I&aZY+8`QVvW8
z4}j=ZNA_HY8|t5Zk0Gl|WxoVFjnz=VZ?>&tmlK+$W+`9m4Ctf4J&D||7<$bt(|$b>
zwdz`gjI?QJ6@^BQ+=ig#U)3=fQR|cvU!2}?XJU|1Rf{I^YvSu-J?K=I$%`O!tcG^d
zkhYYv1R-8sR=^ha|NV$lHKDEx>EznGIMC&*KuiKP*q`F~Mh>cwF0vUzM~BwWS45>n
z!(QE7Ntga>G%f8AACdg#FM68+zReT{mmQ1u4OQ_<Y)|nmAMK2jga5#F7XDE2EaueD
zR8hjw0f*tn9o?93?9@_=8yUUZp@s^_(H_adVI0O`i30gc%O>JZ72+3PDQ)bWE<>JA
z*Mnfr=7WlUzMrve8XdQ@oV6ga{ExS*MGz)dut0qC<$ggAR~LD@eL02MlrA+)D~b;H
zvl5QXXx++G#$zWgrSppr%wo6;pbX39WC(Fb2li+HpLw|+Y9)jydc5Evgd?{fEcS5Z
zrrdd~{n5>8-1C+4#)J92a3nW#-m2zb{cC)HJAcIalQi-4q|Z@VD`-IjgztGpIz|iy
zSz;k0NbuUZhhJ~#J?Imn5AjNi#nAaJcoZ(_Q)c7A0{zqFE)h@hxWm3<Fsy23<ri22
zOacivXcHGI-iq@s=ztAMN{WIm;%y9~BS{uWno__eWH$Fbu8zmopy6LO121Q_Zm&G|
za2B7PO6IVQJH6s-XjuYfmtRRrdPO4g$2lbzMtf1*_D0ZOl;MK(;RIdA1PsjC3=fJ>
zJoI<JMB|rI&L>10ADE_Ol|$q>Vi`8zvvxZlNC{3*W8koBv-D2>1fKqFkVb`R1ECRt
zu<=J{_<q_sknaOdXZHu+EAtW(oY@{K*Bka``}>z9yOJtT-)z9YVuA7u$#)1^sE+#k
zSeYfiX(80?uUMQ%`PH$3EOP$pIY(8hyIh?lm6k}S3z`zfleY(>TkYG)0ZLyKmuu6$
zAbpR7j^((pw#4G0=Dyh}kk#Q?pQHSir3O&Zp~T~7d#ZMt2q0&A+7z@Ki|t6dx_=Z`
zh0)TlHG!%V@K?L|$K*vVXB+9WG$x1oq)f}$jmHP7vQA+mdP1r`XkJHqp(X+Ad<Yfh
z$GCnv-L=%&KN8%6b>5otO@fJ(^kuHQ<h!=U#DnW^%U%3!6WnI2n2(#y-HWzWzA_3s
zBZ~uO;;r$mifJ@i;`Xj)C;tA>5Pgm<UDY#4q=CBVcm;Kmd8-ev;dSWx^LT50&_;B)
zJRzfwm1AEt9HVSHLuF5YT=A8f4RQtB;)b;T8rG)0>Q3(mrSwb4FdjWNJ{JUv5p%(`
zu9{J-p#|WFgstMy6Q6k)o3#94=Jpc-tgBkxs7Bn#Ym!`~78=g*hmbuU(VmK%*?0{$
zg^`>Wp$CnE1`E9jBqYI8Jg;Ry%Envu48*<q-MhB>``fFSOH$~Mb*!^r_d8Wyh_;=I
zSM`kEDtsOtUDhdG-`5!QqWUbXVS$9<6w*WZZEiCv+a&xy`Rat2fPM`I=U&fBe!b$V
z8lW=Wo(p>$am(uDuz3Tio6Q5K+GnKx_sqY8*)v3V5-F+G15-BmU8<L0Ug5fs0`}4Q
zd26T*%n8nLe514w+&<Hm!5BRFc2fM(D|CMp(IlaiF5;4`SA-dhtAA+=qkP$O+Fq|U
zw_I%$vzRxC&wg=pcHtSTxK7||Y4^f*QJkj!-<?Hu&dIw*Lt4iq-8F{gsb@F}A;vb-
z2n~y3&11Ka3&$8~-aJHIX=%kc5s`kTRuj|$i&R+f>@-z#Bbt%j20^@3i}40NMz@el
zCDSw90Le*}k>p{4AmMH+TWVyok~mMP73Q7GAZ({7DY~!dXO0nnNU(3pFpn6SC-4I-
zt{N6YJL*!qG2EoJe~_aTrX+}JHtrMYFg`csxK*yO?_1ie6Eh5dXM{(@5p)g{mWy5x
z6a@~Pr2L+JGuR|cbMYf3+Z0NDEz^1QC%7&nY=EWYq3zyCr(sv!1S{gm1e#Oc)-KF(
zi4YB2l_DT4f8i`>rOV%s*m)>1VU0EEb;=5>(PfM)dM%p|DV69zW!ZAabAMtJ@u(tR
zl}GhBWniW1s5+}M#X3K42>0)1^Cltout~;j;Quu}w-)asy-vZHt;A0wex&ALwCa3M
zZ^XBFU<vI{<@wx3rvw8%mYvnMjJ!n4-eGZux)amzjR`T2`kl}2DhVy*n7K~i1EmS%
zHnzFYC8Gn^3NxZ6D2>lB2ApadevZ6?DfvlRhJ<Z~6IDBI0DZ-&38k5nkW{n4`bDh#
zvzYRkFw$4fs}h0O$c?tx9-C?Y)JKRy+i&zMSkuychSW+DaA>Sx8S+OVVZHO`5przZ
zP%IF;@W?8B3&9y-`(t(-Rg!zS5?u4$N)C9o%F+jM-mCaM;}R?LrCVcP6b@Y*^<tqE
z>JKWiBn8174zpEfL*g~a(9r>8Ot8<ZlFo>`86nJ+ENt10KlnR}2XB$4h>^#0EsnPD
z39X{i3NFxTH29`w@6xYCZ9g#^gI@o)q0CpjLumy|(jMSSdGEDXM|d;TH9{AcVf~q-
z`Ww*FISliAM&nA2{v=z6;F<esO@&qVz2ouM;&vnHTg`nEev{*;4G(_5y`jNYYD3mD
zgE^f2kJJp@s>$16D{7SrdHt{ld!PYYx9}o*75z`XbR3zgta^zF1Jd4njk3YfmO*q)
z*Cu&qQ%vS^Ul<xsnn_F+@-?KP%2Ta3_X(DSb)aHT)i(Mh;zm~E5smMtK)*=uRWnrf
zdc-USII?@6?`Mm~f_QwNVuvSuh>7$-ktUZaABXu9nYAY`k1aM?Djs`qr>RWx*Revr
zF(XpA&VB8WB-h5+G81NmRyQEa2vfblE^~DX!m!``>7}ixqew-B`K>5I5_m%G*W{bI
zP@d`+u~Vu)H;2Cz1od4USp~9z-do*V(&0J|u@V^{;d^s}`opEC(9n27LeC9#oAAm>
zErmK?2t*3-m{H$Hrvq2uA<|u5H}A#$gDILf*JisapPTDVV3z}vFrYThna6|WHY1e#
zfOHWS`Ug9U(6&lf6y_S4j~upMot?m`k#E%~xXA+E>CHBk_#RF5lM%SfaB@^|Kh|Q=
zmNzc3k=~~3Gn+34Z!;o6QYwl!uvZa1E<xng61rt<g;vfwEk`c}8nV6X6<8imu+gKq
zUYxUC=NC=QWhk3Q=;U)Nw6nX?Fr_Hb3yWq#`^7OF?qxn<5q=Zb!i^F%#qT>pNi33o
zC{IU-?fZq(4axp?vt#LyRUOgX<^X;Z@RtEL4wd;tbDA5Z{Hz?mp+H9LO3Wo(Jeiia
zplfcuF@Vvl)AXd3<tT6VeR1@Uh&_epxys<?wWk#frx%sKu>0^I`_{VExaZLmHqW)!
zi&=1HmR0%syj*N}My=6DCiAXvtLxy~Ep*9pMFm3WlhwlHa^a|#;riF(P6Vy;#}diZ
zipZ!+{g&w~5hHlUv><Ikj58~CevOb<2nEU~Pe2p?N%pLI{6Zv$3{D?oCP>@XW~X3R
zWsw?8sIb15Y>P+z9ut6oSaQG?%f)1^-)!CF&yg@4=HAYB2aA*nS00t<4RH?@p=oo!
z+;!Y_N|yYK+b>B`d8RE^Yw=LRx^e&DE>+vX$&p3lwrCBbS{fuFZkfo4!YQ7OxG1F`
znm)(I04Io4CZ}qm<n@_zZ04MPXH8h%9v5jAV;x2RmUU^eqxr_ga2>JL4jY<HHoFHR
z;F5v`_xo|m9XyE)dHhCp<tYKae(s?IElB!p-MRiPCoU>?1O57wMBpPHUwFw`@y9<>
zcPqW_yIq48Z+cY4#T`=8zzAQI;K==8CShRyfl~(RU8Z93&-CL1iZ(VZ=D`CxE5v&(
zLXAK1`$i_tXe$4!E-FbgEO)5N4dz=C{qB7HH-Rqi0^?=NJmuiZBC`q+0qjz~?Vo$S
zJyNcmZJ1k*&gkE~*aR9~Bgg~udJVvxE!2}%3~#+?C@)>T-sMr9J8t3X6EQLv3kfcn
zsD6@IH401?KyxAqi=DYok$!V?{OXMVc5@CnF$C1<MZCLUkG_D0y_nlP!w~Q$q`~r7
zr7UI}pLd+sM)FI$$H@q44Fvc$W;fZ7<4rUryBpG^hF5@qLrP@*b|WFI=VuRn-l7T8
zQvqmaM6s{4z;u0o0vC>$5+XflLkgOO1umn5j%H8u<H#gLW?D;ZqK)oQkxw>qM5Ru7
zLCNii99bR0G>g9L^3KiKBtD)o7vk~?2o+VP^M<kH0U(<J*G-7g*@0Z}t=Hrlc>FO@
zg81aN3;LJxBK!77(a7NXl3qbZ6><p&6#(YI=h*SG^#S(hNbKN;*Qwgy0*JfP&Q|g2
z8=}$tm(WTSTcTeobOAVHC84(Cn}LnqjftCIjxYG6+JS+?x=?~6MS4t{N|)D@nUhRR
z&eCxv{?~V(9f~SLwbs8;=sjABz+!nB=SB8}pGC?Gv>^v%PxPN`rc4h*3bL9!_05bS
zO>e1T2kyZiR1QW=b%HeZ$2$Cf<*nT^DeLOT3u+Ro3tA!}*&UeH{>(>J^+VXo+!(}H
z=rNqx$;c&3nb&%+OW-VH$dyi6qV(<sLKvB^#`aQl!O?+-Y1Mku4SB7+9wFnXlb7Dr
z2yFCC2kGqwsuJVuWkeiq1%6UL{PN;!;tODP^_5R~h{55PhNp%Z(78hUv$F1sE+o@~
zY5id9jnj~bi>x?fkjRgZfSE|Zz|b~wkLkCo`$Q08cLLlt9Mo+?7t)Y}`I48bnkkn`
zWN3Ayj$c<qMxZ|^?s;;OsDv!d?Mv}kj8(Nxz6j`(H%7Af-FnVF%a4Ein=|Qd=+Z5A
zt|bGpW`7V7=iWyoza|#rp@>8D*tymqRNV8otWQ97-iJMGmS$3qG<7a=$qSY*!I-uH
z@Yw?kH1%hOK-ZW5cRfE#L8^cTmb6+Fo~^D!)lg8S<o?oL`BTnZzMp<B%W_X{^Uw`b
z?(L*ozG$80T&61`uEctw1PqZ?v=sq?VHM^jY01JopRXabw;NTPdKl}t*sRi}PdD(6
zN|f>*s18;HZQFyrkxbB2mDHdIUMkzT8xYssoPwRWUfatRJfk(`WnDrV@`Y;E`a|(X
z2>jtGnvRqoo31mFR>a0fLj)zR$|#5HR9$P9IM?T4m5n3~;)&T9XkLKSow7~4=?$}y
z%-d^B!3>-x`B7wD@vwHNTZ<M6g}9S*MO9^7-@?95F=+$alQb#q_!Qv$rBPy7In2EH
zVkqSOvFSjgjn%tEb*puMQrm;K4ws$kc7urzlHL>)VbW$vzUs+iUjbEkhu^SAK@o)_
zei_p~?-gq>o1YJTcANTOoHl%B%~u|qS>tTqLgWVwyceRquZKsW4%!Nn63{mIa+4d|
zt18E<#!*D^TN}+mR;*-nSl}&8Zu%pfw#Q%i%De%(ha-?>QM9+2{J1gd+3P}OVt#%R
zJ@i8peJKg6b1tR`NlkTur2`Lq_F0`VFI&wIMsLj?pxc2q!3d)CjWkcs1AF-(JQgbh
z62vN0%f-&y?8za9In`;8U7G=R-r*H1hO_v^{`#IJ;JOrXuVs-iinWuLwZ$F`%;iDg
z7Vra^Y$RMfg}Jw>>exMUR;1deC1K0IDPo6GBLHU8weZ6hs+^hfx))iG`Hrh_$c<bm
z_Obu_t0)3^g_P4;xbaDl;OT9`%U=DE%H(*?+BH=b;_)nsx|UdKO4s*d$r>Twc0e}e
zcKV#um5sc?Ovb4?MO6r8NEy<&J$qJLbwMTXV6TfRo=`RY#Q+_8e!T<!%t;$;LdKA9
zU<dP;LWa;$dtl$we!i=+?t^K{Wm|yjq>heI1E(Ks$iU?2QD@BxVMVFLdNa`oBsTuh
zf(PjMlExrjN{Lo#mc;&XA@~r%Cac7{u4ag!6O1tHnQ-&J0&sZ7`NDvEw5sq;ZMV2P
zN@L67V8oOVEZ<s;S9^A27i`%H9;~Zfr7f(x8lMxpP_qzu-DK-4;Re18C)CyxxuNZM
zO_OVNzP`U?Ae|4$pEK*Q88-3sM;#Q-ZKrDFNbvh?(8C-7P&bD%rMz@A51pcWP_aZi
zi<(Oz#dc@2sl+2R@)^u@DZV%ppNzLPv&R8snIal0bcD|+-d+zE-}T24<Vm_Uv8ayj
z7@8nP^EI>K`8HO_Y<yO!P}KSJ3^A-R^|@5D%z(%NGgq2%ah2;uzXkUOw(5$YG~2@7
zM%QopxH)uh==m@rF=Rw~l3cVj)HrdF*pY%NB`?nQMuxx4F~xz=l|RAiUC4)<1-<iJ
zO<#6@;(@?4(0}<fkY>Swb)ebRhsC`{;?plzO+9pwV3dYrNUk)m$1X!Vn=T?M7Qrjl
z6D`KTt}EZTqgTRSLBL9$4OTD=tCqXvu*MG%Fic01f|ddCb1PhTI*H8xI!>f<`{V?!
zQ)GtlCYa}r!i0&!<bkN?zjsu@_*<7(9Go{0-XZl$*8j86fsS#55`0}YOM<?>r*RUf
zj-(MDH%4m1M9-`kBgAiFsknYN+H|~+O3BbGQ+p57d*B_~;zxOyKk4D8zOB+mNJ?f5
zx;7MP<473eH%N%b7>D-zPa%h{mWe%|Np#-JUG~HWey#AQj=hC{|ATZ$I@%i38~5?<
z+2a@^6C{2z>Z8=Pn*1BI{s|{0b>lh9352Fa>wZb?48)%y2a=u&MiE~kCPG23uZoe%
zr3u8YmF$U|qvsRL2K4((9pOh#&(MH@jH>%N@|p(82LbYMa_nu)-Q0)g#h?g-kzz5q
zy$m|4zdSxhAiJl9X`+%W%E2$g1+OUED;=YiV(z9y9|=f#{txmay;ja-Py5EKU|QVu
zR>aO7K+BX<kBV8-({A%T66C&7>FP?3Zf<x~hg{%9!1&6bgJ*bZ;I<d|QW-AYT?{=+
zBx1;gT5;VkmGxPlJM!h()jLXUtr`)=1xz0Z9wxMgs?CYD<un%O^K}DF>6qV3<Tio#
zW1ZNx9E@YV4Ejm86Pd2Bf4HSFSL?#XuBLT0Pd4*lw&)7R&EVF1Zc!C=lq$5LD8AH9
z`F(_ah`9+yYM{Q*lR?m91O&`gX|zo}PZ9KYF*Trg_fbUSj~vvq*ozjH!PyMmeHh97
zSHItgFFq-T!9w>A^T5Pl4MWa*U&efy`#p4*A=?Q56>ya9P->LNk=}^_+0r|L&5PN-
zj|;&E(tk9?IrKViX<AbCHv6<TFkt$|Vg(^lfqpGa2nQ6;ryk!rawOvE{!*J^>2!`Y
z`r{EHEbvN7!qg^I`Le69D3#bRd@VN7vXX>hPAchtc3J+2R=FtQRTk%wuLXqvg~eQA
z9+!JW72G8ZS}$A=omlVAxEGEmgF;O#l9H#muVZlQ?M%UHR@fw-cgdXl)&RYV!=uUW
z%q(2^%=3)FYQk1udgzrSJVEkwlE{RrluC&`il*M!J5q7sQGhwYEOtJcRu#rw?qNXf
z+#Liz{8wGJw<Hv4ev1A{|46af(*`_x!*3MusPj?MS}CpMYGd}w^SP`a4R@3~Rtpsr
zCR#Z?ZWZ*QE-K7r8$;x7Dl4l%sayJ(X)4WA9IzmSSwI%6K1dB?thuKFuR;8Xze#lk
zV+Cbg$A#XDXoRPE`~H(cNfH*j-F3i3=TfNovb6nMf(Zg~UYZewq%=`hUJAW6Md%2Z
zyi3&}GRP55F5R1mOY0_r5J#i<23X@z2m%*(L)|~WF$!kz(+!yJOT6hvh4#_;*rPb?
z2Xl)jF65S@u{{DS+^^7u?dfln>{%2csOuk~d7Z$RCU<90@NTXellCaav|IKyOWS+G
z{QI+zX@VTPeKMxWU9E$Es*>+SS3{1Yl96kCu26R>*FAP`Ui=I<7qhW&D#y!N36jiW
zuu@x<4GXDasZS!#Szz)?6cQ3ndN&72>RfZgfl~pO#0}lqkD|@riK4vlySPmT>IV~8
ze!HkR7cnL;-dvDNq>gmv%)gtpLz!>UFgsz4dpO$Uuk`W#>V1}yOy45?8a$*tkC`~A
zH?trrF46vUpnORDJVvkIg^)5Kg<$#^X~Jm74yQt=k~<eLC%upPBeYq+*P&byEbx4z
zGZ*16AB!C}-d*!JG`3r!)O!$E>zf>}P-avbj!VA(RaFfUZj!B_sa$DmbQ%x2-;_5U
zFGiO7P7Ztz)X1B|OX&5aI$ov~|JB#Or?^w%n#ac;s^C9E{UTX3R-D-5gLnlwUZMv^
z=P3bMkKTU;VzNVWe<tB=D#EcL?G6YbVo>jTr0lBbjikyuRzjiCTl5-AhShR|c$HQK
zT;OJ~gTp^E*5zKAEusV{*{`O{)PWHtf2>}`Dg$5s*#`XL#27sXB1?kxZqQWCg~J?h
zVqIyE7f<Btgw>PD^Uvi0H!W!JyU@X*B3F$SJ;S9(0X0#MmuMWuj2}qNq;MBhs$}gF
zbgrnFeEF-YQtSpXjH-?lB<sA4zbQ}uM{(1-OpJtUZ6AFS@ab<aj?LL?)@nnXa({za
z3*BGNw+YFtER@q&VT@f&g-b$vtnf=N73;9h(>-_LyEWb7x;0G8HU>4)te9_LrC^%X
z`_jsM^=z^lt@hW)$U6IAJ0hwK?tJfZUIN?>jLUb}SUx){U4HIWp=tLO0Fb1>zJ-Hq
zd!))k-X`$F4?2klUEBO%1go)F{})P2t;(Jktm{S7oSoFMz4sbM61h|m({q^|)o6DD
zNYyWCB$yB!ywBI9Vd>*<ZqGF({bnTYAH&?GyBJ6DNDQWT6fZN>!U!IV{30!q<7ctI
zLAVgV0B&^8H#R6$0*B2$7J$-_yH-O-%{J3_tn`7k!dAP0Mv<j6biP}V`l0Nr@yw3-
zvp?)sZ1j1R;=m;&2aS3550L%M%%<1`g)mzc#EsIX-1p3MVF_Gp=&Dc(?2bOy++mU}
zGHr6ffz-o~1%>PSQa<jz+dE=LT*7@NmFl7HL*nsMy$G5!@I-!Y1{ELjhB_o>xuY3o
z<oa?>jQMv+q_I9&^Tg>*ut5oCX$W|1#)iLu3dhpseu-}ljszh?zW6-=<Q(k~cUNw~
zE`qg;R$+AyGhPm$Ne~iYa=M&XYfs3dFrru?Vc%$v459TyZ5H|i4MB6n)7x^*7)Pm7
zIoL+htc)?($n6Fu^oXuX>2(NMR`%)?rd&bX<q0Q=g*`;t9kRV(<P{mc1j+RCPK>4|
z{QPt3lTq5f-Vh3X#(6^X$u?H(eo`MQr-twtTp=XU&Fh_O)}2XtQEO3$%8Dz1%en0#
z77Qax{K8t1;zH{>xl+_^62+We?u-F1Q?(+yG$<}wJojA*io&%;7%7Kq5N{%U_8%Oq
zYX=rsd2rUShOBt~<5N4ai~ubNHsBlBUR5KSGD53m$$MIu7ED{K?!5Yj1ZYzU%CU6>
zj%B3Cz8nP;A(j4=2TtKx?O{h6zt+xZb)4U!n3x~xVb`q~e?+(wJvR0?5JrX3%<U4x
zBABuAXVa}e6cU-|++tmiC7)BoL*^G>)IiO3WRs?Q#$T@Wa)&QZ<G67qag+eje5&Nm
zKs*MPucNEl3dYX+;po?f_=Ns<4DFMY!{hesaXxQ5s0tMc6ViH3_|MaLjIB^i`;T~E
zDrojsEU63L&dM~iOFbv8&0~)<B3m1-g6ZrC74`@VPP+AtNElDm8YE1}-8{Go3`9&0
zoxMA9Q_!q!Os;AoX!V;)#fyi}c@1)#x__+>hdS4G^iYIXZqgcccTjHLhcQ>H>)@J1
z-*0ia|GkgZ((Lo3N|+1f@$Ge-O2K@rt<o@fx7BnUSrzb7R?A;J5&RNZ#a1~LQEZu<
zem)ic2t;+<`-*Z#?b5&&r#o!p1%Yd97gi%N#S<DFt?91`DrgjRJsFte1?p(593O6U
zk$qVwqNxzs7%Uq{rKH+Bw`#q=Ewuj@9M6b<yDiCbw-YFN|Ipfp?$s@?_F&16|AxUi
z#-ve5A-k+-7~UU#nIh6Jm6>}|KdO@W6|zfS*5Mw8PLppF;kc6w#tX(x3nW|aWyPFI
zF>>xRSYH~5FX_*qcrhZqh)sayVnw_-d|iY&KvOfKPUn4=3w$M;uq0wJZ+Phs=CKQj
z+(}cHRW2nkWP<^0(nRRlno%5e1X>5kmh<xXbCkZkt_dZ(@aywglMmHhd{eQl(#6WV
zjWXD>ott^cztctwP^V1S1_TZnII+o=WQ#6t2NDTo%JJu4<%P8LrA=YZV}J7Qdkk^>
z3#|@SCy>qEP)Gni@v;&fUV&p9`>WE1`D%Rcfl8WIlXQkCZzIqs>{v{5-IheY)_fQP
zF?2|{es4Gs?V=r^gMkkh?TT?yO=P4W%X*s@y3TmLXLy_S<hIM3<24Gkj2bT3b88rL
zV?6!Mzm<qCb%pqc-O(l}5ef}-_RW(A*Pu6kxtx{-&ib_Zsm98bx3JDM(-00VlO&X(
zWm0Zoh0zA+6`x0TvK5Jd_=SvJfu`$j46>!eV!k^S^;)+%b+arRg$r~KhCTJL(j^ZC
z9&;h#j~=uk^wm2^JT<dA#3j~4++ye!ur9G65&bm5QYIC3k?d)lzx=u<;5oySaIi|y
zuR%e&rz!9)2et>wF1iqehdRrWUZ=C~z8Px|i%QUsT<y=QdX`ERdrvi&7I_nU?=3v?
z!xa=e?h$jhuFV9;<zyZLPKl=RxSuagwlO(1o{z?}Ln!vO)IyZGP}(ew5`;DZG50G5
zs5@1;+Pq|&oC*6GRL=OtU4rU?in8=T-PN(RoIoX^E)S_6m?CN=2cDVSVtMf?l$Hjk
zcLJX-4)niOikD-O5M)2jol4?0)gN|^SSo6AMIei*SuLzJ;fow#S}NGRTds;f%XWip
zE|gG}GOXXzb8nhXQLfVGH!Vv`?f(o7>Ab3s=|z6=04WQh;p#~4+issZl!T6cKPHe>
z5%|RVrjHDGGIZUERy5bs`)9+tEeizk4b<D{=4A~k3TADJ!d+RHQ3>q_s_3t1u(yvr
zDWt-fHwHTDI51)l;IYaaC8kS;yi_n$fWO9+xl`gE(hpM5xMj#}rthNYsRZu@6i2b+
zhZ^}0x#;~aiThrDPr&%tYD*G@=2Y=mnS7+eO0GBw$(;(;S^((yZ>Da$76*TV+nPEV
zmR%!nH?KFi3WeXsU&Fe?iE(j?Xx~(Q0L2a;8w7~glO!f;Rt1|#^2l%G(czFVl5VxV
zdjH@|S1`jd$<f<}{><X+Mje4uVN_=mkL5ys#aVNI{Zb$y+qFdRDor|%pNc=@YzC(h
zZM_F5oL!$x@*jC7vJp|mrz$+~G<3IydsDXLUR6&#$@)fSCCrJ_D3;1$n^6Z1%rU9i
zuWICnJOJh7fWrMN$P%qb=gi`AK1l0+oU>eN09e0yT4gulU6{+u2YJb|NrmjLD<}4L
z-g<gNaJOmVo9~%kt3k1nNTHJT=hUl8m!+2X+GL)y6TW#%#e+}=aQ(SEz2$;Ia<Twj
z`?|Q*ezblLJMk%NeMOx!LPhL)Q4jFhU^?6=pTUketFyFo+ma=&`Yl2Bse>Ho&5Jvk
zmD1#rQe<fYhKu(4B^ZqX-NjB-_b!+okp|Q7p>%R*ram978)GentsTh7+&lTpDe4V!
zmOEAwn&po~h2S(9!G#_hBuq|J*!Qokq|VEW!s^Mf{+PPha>9Q^tJgevi4)H<soap(
zUO;>8NE^9XP0vIA!Y*5N%#0|<<9w(Z1JD#Wqhr<Q>t%VXvr3aHmsbhF)@dvF9IJ3h
zq`}3K#vNowCtRg7h2>Z#usT(6RB&=Nh|<-9x*<@TSW}vE5#K5biBU75#%y=!kbO<_
zZ_Sfms^7Q__H|iW$oWJ|@M4^q+*xCTj(mh{mIS$6$;}nFm-DhH&|{?J_j)}GZ)bLN
zHzKHK@jf106;^-SSDo7m+18LhmaK@>yxIBZPp{mFPUp)q^$@2{MMhMo<rwvy@TV_<
z_Nb5yh7K+o8!sJyd=Yp*p{9)sZ?X*|v$X&GJ*}O2R<{Rw$k<I>f|xv%`+6hGaD^iy
zQJ@F5(!bcj_E4M2;;lY&!C@NNq1^UiXK%-?lfi{)+!!jZ?NqX<Y$MCC=kFv2g6G5(
zf!0LPW&~187AwZ2Mb5CKvK5R?DhGs=f<q7e9*3b|VgN=^SfuR+icI-Y0g!!)<xG4O
z%dT=sdg{s~ACj83WfbG}?;<t{sotXhlcEz)#mEsl{{Yun(tK%iHmwdh{NAGy8@#gk
z#m45ah7i<SnuZfWm5d#N>z%}62*063oihHOp7a?zHN}y1W1^urjoJ81ddPQkXY}5&
zx*pV{fCU;~ZHN&8Rt#!;wNiAwp8%{=)CKW#Wgu==_X!JE92qtVUyTs5C@~g_jg@+-
zVr^!b7O%vPaO4jYo}Yp%bF7Zo?>%^!t5ytw^g;88gp_LqHb|*6VW|+=cr623_~4MD
zbUMv8VPV$CC}dSxL#bzpnSq&xjE&{ArzPu<nsJwbpJuer6MYdiqR{lD+SK>HIv82!
zQF;j_6YH0YyLY}YiA&sauA5bKtQ3<&%D%G7BUjc?xyPUN5N48zT|9T}i?=Ki!e7)l
zp$};DM2oIhzVDf0JX$Cc1Wv85jw1VmP33JVVzTR?G)#n+%~QXMmq(B>`8UW;GT<{)
z`{Mj?E0aYiuM{GBNak0dGJk?CJ$e1Enz8ZGkxwtnaD)4<>1K>MyMj}##>jYCD_~6I
zLb{93sDpCJ25g1uEwk%j1A3&c7ts6&gq<!y-Sk6BT0)PKZKD6~Z1GCN&e_aofs@Ar
z1WwDRv}nvswCQJ7bR*}CysyJ$z$S!B2bhm?J5_?_W=Z5qisGzA1JWZ0NR|Dy!(8}&
z>3oOx)xSIk{?xi^8hE}#aVEQm!n7C&0c>@a-gRt$)L2oHLdI8=t$9M$JvlBiXqd`n
z9y27Duo;20d8gkaDDrN6hq@mI=k}=&h~=4(`O^0Z1O>nY8ZJ8YcbW}!Wx26*nuW1m
zR(Zc!Cy({<Ze(G2nW5EL62cC4J(K&gQI5+K?HC3&^`GG6*Q8kI6{E_z`-Li#5d5rY
zEz}ci)~6xTFm0Xb4InA26ez@8G~#M=5&0~tOO8DuiL*BuMEv{gfDQ59U%FAlMCI;A
z#6`q&Ibncad&KCc%^%3G_3StR(?)+5r4o=_3dpEr)gc$R1c1N3z0?fvi9W9^E)6ri
zxU`#RBypCvi!QX<(H~d2!8}xw-;0shSg>E@u3eDxFaIUcb;<e8UkfLmz_#ZA>|^e&
zXrnwDG`?&66(!?5uq#w6S?_BT=3IWSG4X#gCbLg9guBqnCjW5}pi%l96e9Zzoga7(
zw2uT_cPB$E@Y%~Lc}vwBwvSz@T-QIiz1_3ZIy*kq8PlRamAu<qTJ4fJcv5FaGxjG>
z1?v{A6tFu~$V<E7bgc}|C10fI^Oqz!TT$+8_8I!?YBR5a@a#K+^jaDb43Vpgz3Km?
zSTXJ!VR``45MChlJF*~$Wqy+^<@Dw3(xV%_q)o;ro3KPED};i0YD@<9<%H+u0WxzP
z=?7l$oWvD<m0+_|t@>y4;WX=PbDMIoBJ)racy>Dw!>SH+(d+y-MP=0qO@ouyv<E!^
zK^8f(&G-_)`3cn}*rr=4|19Gmn*Wk2ukcLuLqZuHgc5Jpay?B5N5F@$+ygr5BzSVY
z3QHBLI1~DEmX#aB{<8_&FZyJ`8}occdF!Y(uQpzu9DTS>J%ZCm1-8>Gj8ypgl%eih
z$885E40Mt6W6WhxDEb?`I&?t?a<4~mu&*r^B{c8IhP)lFyNbhAaG;D8e4t@I(f8-M
z{Kds@l(b2OErUWh1<$4qvGl3%$(_x>M5hW>cT`JI)w0mlAaR5FWqU7FR`buka3b87
z2PRo+A+4X4rqa_&Z9_NdJNT&syqytV>U)~vSO_!BflvpC5XisekJL_ZE#~tYpGHq3
z<{m656o6puKxC|<RO^KR%GbQa=_f)HEx%lZLr{vWr!)UeTkl>-?piw<f=qJeS~xAB
zAwf8T&D9ZM(hJX}koP>6o;Q=vhpgzeoyxV8wzbTM+-+ea!9sPQhb-yVmXhE_rdMxu
zGXqTPt3z-nYDuu;jBROS7wsHBOiClbhP0n$E#$<bSOY(^UaflSbqhd~Q;AlL0-@NU
zRpOpn=zSAFt)^pS%$%yOvpc@K72y4nj2hwNamtgVj!m1HsvYxa?L4rkEEBAAO*gw{
zD)jd*pG;XZU?DNRb6GV_A>vdsAMA+rvbi@LJxeHhCI8ru^_`fHZz{H76~HILG&DP|
z@{{?|3L!70OtrSHR02BDW|IV|32u(1Q*P#5h7SA<v(W;y+qdd(jIP9KD03{_dV9Qf
zw2Lmv4tNbPzuhG^ck7OhVWESaav)Nv_?w4X0DK3jdLrHvGrE+6@TtHN7$*{?1qW-s
zy{O$U3yf77m9w`}YQ!wH$0L5B--ys|R?V3nn@+Ux5gR|?KPs2!Xr6B&n<UN92$?E9
zG`p68Fy~%XoY5?00j*;64x-{}iWIn5P6jHqeuSajdnqi!VyCMYv^mZJW$4T6ql;Tz
zd*Rom;j}GYIAoG!Sph^uqc3%?*_Cz)S=lGpSBc=#=0cqkO))lzuoi|DZ`NWi*##j;
z&cyAYB2OBSH-qsoVgGQJYFXDA2m%}vmuX07iiUDgZWw*@<-&2`gtmjtL{A`LQSDNp
zmgb$NdiNH2mIU4m9eh0nZlJbf`f8`d{M?=`*ic38J9=ohBoN_eely3nAIGl2cUgPj
zX&upnT*)tjV|`1${7$7+3^QVa#NQLN$Se3BL{=TWxI#T@KvT4FaU(t;1Tr>X1>p%b
zbCv$kcsJHRmPctL>eZ8d9})M<!C@%Zh)u|c@IJ0-NQIZwsIZ-}?8;qhrDOLHnZgG4
zq^Z!<7Uc~4kZ3u1rL0~!Htn}3n&+$%eR@4z37*NSY4gfBr#i%frm-C=jdwW6R%?*~
z?Gb;rF79uNj^z!*SII)A%r!cO7)p6Xa@=NMDWWZD%`tBu)c#;a4IvDF|28A;DS?Sa
zKZb)|16L*Vc|sR;PZbTbjinz`r>WmaQ`<k}XU@N;H~w79_l7`19pd75cXNRulZrW)
ziDoP(hD4EG$C-`3tF@zWLhLhO@mFyp#Ww_o4+^2ngnqX~ydi_T)p91X?ZvzOOY`5A
zhYJ{r;M7s0J=G{W0c?+|tsY-aX<9nv+X9=VG7No%TKy70<H$rvMnS%f*--L(dH-O`
zS@NVS#YYY;#dn}utxsn&(hTlE%SH3cfmFdmm2aIO0^W~Z3~`#Ehi}*~jUiY|0-n!b
zaDl-F#lN+_t0!Bvvbm3(%j56|6n^5)-A{z*M4t8(`Lzm(Lk4j76OIT%UN|tU9x)xQ
zzz~NTR}bTZ?~M`3)>?lgk`x=@2QKkU85`H+QKB#AX@6W#L<}-r;Ckm7#oURx;#7Ze
zMB|;nB#}L%!s$<9_gt+7LmT@%ulxF8-HYLZHFuI26hvndU9#LbLsyEivkOFau}9)k
zNh*^|LCk(g!5!e*pP_<{iGDr`e>v8a5lSNhIcE7vPd1Po`A<XGAYoZ2CVN2#b+u%X
zka;(==exB01v=Mai?)ihjA7O<y7|v2N}QNlI7LsicUL3koyUT#M3(ykvqmk;*U#PY
ztc@+vPTKQPg$WqdC1~veBg}SNV3{D9(aji{$n%Q2Ro?9R44-iF>@7v}^Qz{Lva7~(
zq#(FdM@jT1YU1u`O-!Bi#@{RXq5>FJ&)*Cq!1s(%&G^D(JhWP=h?w80YJ5y-@+jvL
z&9HOzx{`Dy_K>#=xHPP76IiE58Iw5hJ6EEVl85IHue5DlWWE^Jr)1E7YiOR^1J1g}
zT!Z@((W{iXq4KCqAb<X=*~|G~&0gmJWA-w!voLb}H?x<CgNc>%|7-TP8^@dKrqN}$
zR8R0){D<ppxvpx`{m+Sq#zqZWpYQh5<u#_~b+!B3_;$@ImTPoK!)iE!vp5)LqA8I4
zMZDDL#g)D(+8Is<NR`Y5Tm%P41XachOH1mP2u`0q3k@7qd0kR@K~zb1PIy5S)i1&A
zB``uGOE?1~5JpEw$ANf|;A=eI9NK9*9UwG6F3R$9i=WAdIM}|iM}hd>gqhS6SD=Po
z0^3sd7lf8lS9(9HpIXoqIXPf@j8Y&L8Jyq#bzOm_#o3(x!aDNkgtnGu7f-Fw;OZne
zGRFSVCH3VwjZ|d+Yxb_~%nUu`tz@77I6`xS!U4}9l#cu}du92e<?wz2%z+R}<y%r`
zX8d}AQ)hsjC`5SDdRgr2AvHFzHrq14e)s8jW_<U;INCP=E)73_eae&?++H16Uzwcm
zfxKm@3W;9rx^!q@c6=pmu0Jrq)p2}YS{T_~eJIEP27VU4#B|~4uCsyZOY!LeN6gIN
z=^ETTYnYxt)8A=R{o@b%rfuBX8yNs~a8@9S-%cVFs5rmPPXAfI?SPCKzh~$-{SFRi
z`?~ut^``IMuAfE-nO|nVaag!9#Y=m+Vsw8_*jt-|3IJx6k{cQ8pqN-cHXIHvZ-`UV
zOXvEag>PoFia&G%bN|@le`v3fC4>lr+3f=dng0Pm-n0Y2l>mSw{UJbp^Ap_xa5w-u
z#Q~E<KYo0(fE=x<<q`3jxs@k<A%8%dF%YxAA3c`LAz*rYZe?NQ5rB5L%eO2C&@D+U
zZ?B9gSo8}6$T66k{iq{r$#bvxY18El)AGs6q0#C_(*gYnW%o0CX6UTQi6~EsDakqi
z$M&+x1JRLy4*p|%X=!7suYa)kFMEl8^CSWEe)KbbRC&9TGegtUnV$i^wLg;z`?Wvx
z)VT<y{s4i9j;ii)Kl&d-WR!+iAos<8iY}0O29^c@zq1!IE-7UG6%c=~FB%PKPGxg;
zXm|oLXPIMSWCZ!?0UW;9rri9KdYf5VINr@hDLXMYGQPG=02XhM?5-}3AC=d=GvB+V
z|B-sHZ7nYhE#Vq_jO?cHj8bw__Lajjc$mow`Q}EUxKS<?H_DV%kQ9w~FZ1y0M>Q?(
zlyIb1SqGn&Z`!?Y?mQ^PTv%7a3`KUbEIo7LPGw}*_5Hi#R>tLh&_rw1Ve%Xb@UMLu
zNgH8Hhez^Vn*>VAiq#*0?~O~7)}b#>GMN3>stJK&j>dueXF02h5V~Q@p`l7@?!E$P
z&!65Syv@CxP1;@&x&TON+!P3z$(u-v49D4w>%)tE=k@b%uwT<L+YS&f)sQ^f#0Ml5
z(jPjTurz1?O50lHl0PgIb?()wNSk>Yu8Ea&<z5aOwlz@0EJd&bb{ePw1`CNigJjjx
z&Z6B>MzJQJ_w%fKf4kvQCB@l~<4D$U=lJ>qixq#_z)y+1<k9IHfAm+3st^OMdB+HB
zBSlKx<(tQ4%E#FXd-KifSPa$7aK4g-12fK7r;MW+qJa5fSiDxj_VPux>Fx`@h}-zV
z8h<&ctMg&^7wqyHLNxSLa+)^0%0J(Pw{;If&8hNAvkSs3qNbbPt#VtPiWqfKR|UO2
zPXcXh+nZpyr6$6OzFIk*`HZS=8n>p>2J}iItTV(#>&uZsyw!0D7Nmfb)h8h=Ct)n!
zX|+#7=+Yf==fE=?d1vJ@f;6(b7%oCla7!9uEAhJGbj~zul3k7I*Lbp>Qc!1;s2*RH
zo=d1dqavnk$SNgoGfa{M*&ZE7LwWUkg;J}!=joB!OM`D4`Vc2w19O35KLP{3!b{aL
zRe^slzyU&l##X+>7c+`j4goA`4$|gmiU~^92aS9PFAmpVLxSZNcx634x9YNMv?%&*
z|87CEi1)b<VHrhaOGEt8Ng<u2Q_obAmW?OxQaH<G&%7o1@X9sm85IiJ9*u0lJk@fl
z>dz7^e7V{^0oso{XY@Xjaao%~BR}dRM<)C)AiZhkhd?af;2DAaiv3;zztzILYb!dF
ztQ*&yHol$fn5LEt(9>S{WZZ!*P+aJe_S7LQwEGQPv#~cyndFrOo((WHXnn1`xM+H$
zm-?1w(b!LpWhwYHM=UZTBc&xz_py0JjxsU5R0}ScbbK8#>WT^W3G23OT8AkfPbgl*
zh=oLv9?XxWU3Cg5*#>uX`$x=45Gd{aO~=rUkkbJLFBz&<#G0t=P7BFdLQ!iA9zr=}
z{j&tO@(3BI7MwIUoN-6T*{#`o-V)h^J&%>2;_`T!iP3C7J!yZX>sV%!##txHvW37G
zB6zO|NQ)F?TwcP1m;+7qO);-!xY5}>kqSf?r_)Pw?{JnTLFh2gtjwcO#YR(gJ--Mr
zv<=#B<>4xuU{j#|z4oUYqjXj_Z;j^t40CLR3Rk*O#dXFg(D-+|atr+QbT04((r<cB
z+w-?M=4E}eA|P}hO31(+$wU}C=UGan=af@25^0T=RPurvgF``5eDvrrQU<@+y=?Q0
zks!kyxLbRRUPS}!sLVXD9{VPQE^4zNZ>{{?DTS@9#=s$Z>Vuyx7y2;*E~CGH<jt`g
z(;^V@9?f3ATC7y2i%^bQclNqE>*4#sumx(E5}pC>vOZfBy4HAP7lER#S-gEKwjoAP
zu)*4_7#=-A;C{cfVzrb<nd3BKyp0Rg#_Z{(FRbr=scel~;_x4o<l6VtT}f6#q<%e*
z7|rP8{9EB}`rz`IYRL`^<$W-J6a?xhJ4=~}2W4*MYuJ3*Rtj3A)V?2A**Axq@_$65
z*n4Du!qdIP<`Xux)r#PFwQuIEN{x%BHE|KCgs&<}=RW*Nox<cI?8W2|_uQ~weBtre
z1(zl?NhN}rGK-j8pD0IScknaqEmcReNG?tscdaJd*lgX&2FCR}1hpvOZEH7x!1d1j
z7dl3FP}<kjWd{G355{#j>po3m8?qKYr5-*o>khp19WOG<<mMi9Uz}ykxw@G#|8mIg
z;b8Hi5gIg>`^oO)oPWWtSa)+Ja-g7+-%~)>X0oA@N^mD>wZzorHj>%h-%eYaG`Eg6
z<$f7`4x>S$eTKx(&QQp7szj*u^-zPPUv8*K9D+UD?r?2R7|tUjz(WQOMtRSco0~KM
zqOW&l2N6Q}7e#Hc9%XUXX0V~#!ke>&Ys)pSByZ*r0VpQ;v)og-s^?O}W1`q1m4-2e
zr%16d6SYRohEH@>j_RFOb?{x>z0D6+_$>0`^6tojKbnai$}&WQ(ziiy)ucRY2My`W
zdY32_G+eaoCNd7bcF#H1!<*ZdEQONxnwIxq^M`als~qISChxYH${)&ilXq;vz;N@U
z&(d#&`*g$MyfXIB@p!x)sIvHtz-jH8yb7lnxIRB+71WFUgzJ*VrX_fUw=+bRr&nlJ
zkHgxCMz-Ct+N`eRdai}vRI@RL_ke$v#sQmp@NN4)1E##Ynnd(lSvD{pMK0B=6xA3q
zb?E<CA~)p^=irv`mz94s$#DHKL>_&{wk|zp{V4GTN>S}QbNP`G2DcSRE{CTkku)yn
zho5M7DLJ?(9J;tZmR&*@q9(DJG7}`^Ey)Xwq}jHoO-jafi4F<7Yn9>tj%VKG3N0!)
zSRC)`0y1ST)d7eL#d4m2v73-x#?|xIy`+`LLvAjnW#i5l>s&Hu;P(@(+V*bf04H-5
zoY8_fWE1Ng8<mXQY1rO^%^$%$?6&gN59v*=qA~)v*SD_sQdp*UQ_!=fV%M-Ka2+?L
zUY>8Wg!W?mO|Sk`<hLMQZR{4I5z2(`C5u~~4L_@*AmDIO_ol$u@K8RtX*8Q<9CrAI
z$KSPQ+1P(ecM?|9!FbG?)>H$Dya}f_?;b7aVvnDa!K)<;khC`RT549JdP@#$N#}{E
z0A0x#c<KACiUr@b+!Cy#lgIbLL&dS{rAr1DFA<v6_nod*P<Y)M+|~;Xwe3nE33f#y
zz9izj@%|lR7VGKy?Ji_iUTM`}lY*|i>e5$REhGzFg(GmIiJ_chUf`?ykOZBYG9hZ{
zlxl77-F1ZK>{I4!griK0$B(6|nC0#h#4|63%hC@u&lz9i9C3z^=kww^VcZ+cr0>aC
zQO#^Ls&d^-0pIc>dJ(@`JQv3zcgdC7Xq8E#8;!+8JjZ#?vO;RiS<ZeZb#Q9Xa*wYz
zw;|}N#7#Y1XKkJQshj1Z)O5S&;PH0ber2vv;(@@h$6dk)U9Y33v{m*@bzsZd*mze;
z%Jg9xj?-WXI*a?gLpjmLJ7fQEBp&BM<j(Dx;Lv=enVLM!-=YpETiG+bD4`#e{#_po
z?knf_sxyGOc3A+6EGp+51+pWfn{3yDUfp#o6tTa`k&RYUkYPC_Fw?l0$iJ1ym>QX%
zyrTz_u)3}2O^bFMVK!8_0g0klJ(1d~$vz=R4&V08)Pjdo2JvFszc8~SKajtn3%!Ex
zN$s4(e~oM(&`93HSG;M2lPf)BAVn)0%X~<N1Q%&iy5XUv4^w{;TJWB%T<2ai+w%g(
zzE=2ynTb}_NC2~L(9$NEDZisWrl6okLhP;;LkR~hx&~SL6GTQHhHmsdVgMIr3^MVs
z=rAoGHZdmbfvd!0`FHoUUVGD3dY3^X?9@a`rGgole*EuSz1bfYGtUpz5upLn?i>U+
zn)=*f3sU`7{p0{}$XN+yf)Rcb=LeO+45LU6gso4xrM_6rMAPlcznv$lSRWEj+zISM
zWzOxNM`3QR-@H4e#UNZ^L~|(1awGX_`y+Qu^|(;{6M3%?X}KR6277G^_!X2@9Eb^<
zA#Sn{_$Ebf{wNeZAuFVQiBR)duf>(sAf#ysV68;G?A~O|QF~yMTR1p><^oQ~f1GF|
zLGxRiZWWvKIZxx12zwo+jh{->GYswUWsv%jbht(*cfn=8%{((PNn{?;hA7g2-^$>s
z%^mJSL8#g~Dnu@lQ+3fCrE{;%)*gkA`_GHBasGnIcA9e&Fx#XOf^r8?^7MdO*O(Mt
z6gpJTyC#uCfCrd{!xRtc5+*O%^*6NTG{~3fM_I+<q-G1`U`eWoSh8Dz`(MgDui0(V
z^VP7~`HvNmd5ofd;Obzl;9y5#?l(e=Ogp<t;-nNLj0q8JuZwna@!pS9D6P6yB1f7E
zqb3uwTo$<e*+Uh2Ns^|fX{3JiuvU;MhHN+|2yQ&;U7i%KYn;X&u>%t$5D3Vv|6}Wy
z7j=9@M=*mvFHqLZ;SHVuzhH0C+9wFgb0Umld4(`BkmmbagfM)cIh@^OSujIdzgw#Z
zEe!9_PiauRVgN=JdKjAgtR+BcOJ*f0ls}pW1-hanH?ei0$OVXb;sX^%yv`1%=x&oI
z?(Z!HWPqkh=}dJ)GDVYQ*KXg^oM~L|EsA_Qy6pAel)$f)zVW5Zsbu>UkBxtG<rZUU
zc}X_Vb$VhpH*92-6K~rrsdSCGtbxf)y)2X#zb$|2*?(18N85InrAa)Wp_mKR!bz*9
z@daBA8}7X>HMyLu4C=EbFT4^ce%8Pd91=(Ae7^+oavnwhB0nYuHN^!^cr7uQoIspw
ztJ6JVmq82%V6=?>+Z5lyct*y0yh>ebKPPDINv)*;Y<(|XwnOVl*^FI_2qP(cTQj(j
z3MmS@@^UB*zf}nUc8t#`wdd+iUC;`yrDO>iUi*(yvhaGgNsA3gySo2}vUds=B-*yL
zYumPM+qUi6wr$(CZQHhO+cv9r_x&07Jaj*-h|Kw#^UJYfWzI1wGFP*q-cvO13sd-u
zjgBWxWT`B%wp|hJ9<bldKbsGNPN69G@HFS<x_CgCh`vUKrOGf0)4srY<BrtM#b2W4
zurknPT<9r_Pa;h-nbf{(LK$?RX4@>v23jA|u|>a~;iycSSJR9*sZxXn+R(r>D&8z7
zN>z)cU7uZF&)feEF542xxmLp3(Xv>mW8*#c*A`$JMt<hv#wEKs%_>F=okC7*JBp!^
z6FQ%uT5jptjY37?L+odRL(J`##lm#A0!c~c&(c6;rwf18HA53JCO7eL-M+1|Uq`2{
zsF2qT(%5cTP$-E)O()0F=DfbHfs^}K_2aVKg>R~Znu%mHU@t$E(P+4{A@iKQ#MH2R
zP0@P+YNw{FBf2u)o*;RHBSB9i%;=<?%Ou2+Wbf^~w$$4OwgQE{2gQ=vf9EZD&>O!x
z2XK36k1)hw63!k75ECb!W*Ud%!}C8CMQmv?MFnF@aA!>)ahZ}l|Mi-9oZS|R38)>^
zQY+ViKjES+sa^#FRec82LmhmZDs#NxDlJWB(DH_VXw{vA{s^w2Rc>oCvN|aj?a@)c
z0$vFDO5h{6R!c2{X}0=spP-LkcZ@a;F&ZH7Fm<l@K8Uu?{YvdZ2nw2{a3+inK=KC7
z60xTzpB0arH<1ay!UH?L5qe;N@wN#%`~I`Oh7OEpdzpCL%HJ}K`N2W=__R9)P1A+j
zI^2?he>{J3)?^@`wIEQ6ePi14F*_URVEGxs=LRMahL3+IfPBH&Z(47tV|H9_<jR(C
z8C+OIo7J)oSi%SEPw4$gy+o1w#Z^8!L+Ljsx_o$PwbsnZp1Gh<+%R@du2Hqf90K(k
z!9GMHwR+%&uSmAbsz9lDS`8Garzv+?*j*@c9T_tKcFxyoCw}<M8}?lZhR>YYTgp~#
zpTIU6y=(vIUwCo$&Oc*$iJ9fsWWl<kX3ELl_~we2sAg#EAPOl~cupbT5jCOq8d_B+
zL13#@tZ`o9txk2AUd;#l94}VqaR1@lBQ%avL4+{1B`6i;!J%xA;&QA|W1XcA@m-k&
zRoIeFwrmipd^K3KxViH?_SKshMdJ=YXfl<E$oCP?J4d-OGj_C$prbJ}tNyG#T@8_%
zj!!Am3gH#}W_cpJvW=@tqa;$DeAMlxq#GHH7HPYr2*4CgmqG#LFV&FlTLBHOybqM?
zwmKJn^etX}^iYxfgh6X<Z(rg@Za#9*24^DByGK|N@R3SPHe7QY@L?jZYW2_RXD5+}
z<zWx~H1J$eppZe133iaw#!YV<ATXSiV?O}zt43AG9`((eg;~Tdho!iev&wEVooDf6
zqbvw|q&#J)zNn)`9ns>@Z%=_XSGNIb)&Oq+Z^_Yde|H-B<Ec>^%V#Up>?(YAl#=NU
z@_JEH2&pE{tE*RqF=nKg3CYCaq<-Wdz7k0=PKL}|bhIlVBVNy*+*);JodQ5!0ddMm
z5efK(z*0AfMt-_#eG+YS`gPjWz9_I$X{1K>+WB_%yXi|sA*#LFZLJq+!d)z_Dnkph
z6we2Sldt^mz_m35=4y-Zb=Y*P_7zhLm2=Oii4WsEkn742)jI17b#(v0)gRZac_=A_
zq<iu$jdjaiF4J~Em{<$@5}23KrC|J!ZWbF0w}c;%kcN5jegv}>?6jT2Ze;@IZR|FM
z*D73*8>-wKluJLKh%LF2;1}134;Sxe$vq!mUt=Nf+2mXr=OSq|vKg!MWUVjzaSwz%
z6Jd-%-Se)V`~gi-qXMtnv{2Epg2zp)8=e5Nf>&@-+$ufsUir*UxjXR79Ig1qi<jA0
zslj7g76S%Kb?|Pn?t6mfqD>^>SryaMQ*%r~p8Yi~Gk1I4ByV2W{fEE+l4HQVM(T73
z`2rqe-*ukbQBa^je>vop3Q+(HEaJ_M2(ptkw8;jh%p^}h`=(a7bcDM>nApX>Z~yh^
z=muE^+hkp_h?{}#-raQSzAj^asBY>#2*4;sF5Tdu0=?xIUlmW~$f%fqn*a9MMTSG~
zsQmzYa}gw*JAISO3dW%RP~>VFvt~)=L!O>cAQ7WbxkdvF^~J0maVPe{$#23Mc&(~=
zddq(G+hj}^?}yb{4?TdrJa8`LAtNjmN|5Fa@d!D;jaN(_UCf?DPFUNG=YRoL4J+Yx
zLij^!o393~t|IPvtHc(68nP*AouW-fk$)GXKvpv%G>%Fu)4ED2Qe(PU@E3+~hNKWn
zg3vWNn=)4StG>Nd$WVH~y8*?BP-dLa-zd%ri&>(zt}eT!-#{;Dta{5?Z!Ywg8kV+?
z-GWUq&zf$f-eU}MbK=+<b5gb~+gLQE4|`?KpXN3wroo0aY#cxGbsEs&9B7Tx{;gvr
zC{ux7t>*Svn6k?{{p#a|h84^7nDF`ymCz%Hz=4)~U~0!z;PFFeQyF!Uh6CXc6yaG^
zF%WCQV^HQY7*c<k0o3KzN(9;11T0Qu172Bk(Y0_|@Qzaz%+OpnPBXkbH*2v4bfQ}7
zL>mV>(T>vrkU?O5BY{(C?19MqIv#=K(shuwyLB$AIaS%<=%n2ri;!qg+Ltec<mUSx
zm-rp@3-qaB8dh-@_xEZmK*I0rgU^?n1!t&&^~d<O7#XnHLrIQeikyOU(+tqs_hx3C
zB=rl;R?S0F7D<+}8-aHUXvc_W@meKB+|PVSDkVUj6%Dt=X1_!LBrYgI@&`u$N#~+$
z3sAKkJ57&}{CL6rxw|*01t9DBU`nLquJN2Q0Yf~4Uoz}fBO}&822|RJU!!R!XWnkI
zEE5~_JfQ&sM}SM)xHjzwm7=7s@r=?rlWCv;;PL;=r_PY{9LAAn)1zSPDjNNClp*^Q
zdM@n+OG1eJ-nw)F*AMaNyy4U;<|X4npa-K?E!%}Ouv`=zdns~?66o+epe+L99#4o_
zo)CY&F@r0<U?Uo%+KhDF+?@aO#2&{=L|x;PT@_3kUzO1+ymMy&L`y>Qz)_*~9P}k>
z8Hb>Br;>RdS&8^>ua#Z7Ai<A09xph8?;r-Fj|67~l|X+y6>$Oav!q6A-yWVx+L7qk
zCr2M+g~@#$bM+iXR94igL6>#pMyp+sn7ssuT{{dt;w3--8ylEpMF+z-YCe)qN{%|j
zzJCoIT%<u(a?m(7##X5>YmULzf&x;mI<*C-m)qgB5V}p9RC>hwEf&d54&A+*uxVSn
zo7mWY9~hx`Ql5<dc}t^q6$M<MZBfCbeS+&GPBZOJlYB3c_YTz%@b_ztwXs;7M5^V$
zK93KQYW5HGvL=!yX!5s35n1b;a2&>L^u}MX={jTgr_z4qF?{gI0Z_Qc@5QGd>%GFf
z3cwXKA-q+LDnqo^3T^6-`~yMzc_r{?;?T3a<vOj@muWdyr{%bX&<~gZ8mkd|)v(Vl
z|4?~2f#6vpOW}OK@*;g|hKftc-)-=7p+k~>79XMUaUm;C@E=}a?I$H8abU914!vtP
zf1Ujg|FMH4G?sei882o*8z@y^F65%1^W(Ml+FD%XHZFQ!n4IOL4bbbcvC_P7ht2m&
z(iZs<ch+TIKj}^IYAq(ZC8J={FTb*`Ge0eAj{u_r;ST!GgRL5(hWXz=m}nS7-#Lv+
zZX?P{ldC@g<2|Pig`DG%27VOhT`~xuMUL4^`dU>_MPz4T-```OE2RH+|Fx-4E{?uk
zUu@o!#L!oIN~7kK4liy61y<||aSJ!<k{JSPFLbruNqS&qt01c5F(yG2Y*9w8O>82a
z@+4lM6314nQbk)mz3U)eKj4_zP7?KOGhUiY^coP_c@vo=J}s!^zr~IcT?UkYS5my>
zws@p3RNUmWVd<K-dP~d@deleDS^m>{m3MvwKaoD=bb1R&OjIO}jCXp;m4l9?&{$36
z2@2*mc}^psMG}OJzAZjkxCFo$@^?8xU_E*D<=357zGID`<Z;7AU4Z;`t_pT`Fu9|w
z)mUlMw8j`cyHV+z=TNLjfW0sMQ|U?iR^i6!tK@M`XZWp7vNvfxN(vkTL>Qad4k-sJ
zV<;zpo`V^;%)Si`NG6Zt-;eoHk1&qv?E3}^w6u(@;Q^)7dVFsTnb}1vZ^3I*&7oC~
zpxsTpkCE8G;5r~+yo+=A)QL)6FnO~~cPV@SSR&J7<XYwAb!-DFmKIks;mPB32b#39
zJvCuCk+XPjgnv}Jk+!7=BPyafTd%58K-Km!nfQVmK=hBEcxq#X!xNOx-}VU6XCYDz
zkA#Cd%*7}ssY<zOPr3P9*`=eYevSF^CskYGc(+yd2+Rz0c+il^J2y9f0jwI-x34pw
zTV`h94vDgUQB@RwG`TY|p?J6Q*nHb_UNo097uaQU1ppkHCrS-9fHruTg}TGVjDi&E
zFLSe)Xu69Jj%j@#z7(}qf+fO!E%F`RUGLjv8T-j{(>8tA8x<3vb!Zz=H2?x+uLCmG
zlkU|_PhgdxX(nhI^EEu|)uDTf(wJn<1`K{a|Gpr=K1I&Xpu)*pHWHXfrERpP92LU;
z%2zN@E(r98$mqnB=4yKT7lO+7GoAXB^)-mQ*olf9IN;A!k;4x}V<P^!g&M<f&|I#Q
zY7?Q{1AtT7axzC95B4pSSm0atXKglLP~JCV7^OXWH|DpvSy^b>dU2iJ#i8#)h<kFJ
zPM!jYM0_#q2U_=ngST+=XJ=LbNCbH`{|`Y_P{jr$3yH#cM)D5xS+jOVnF)~dzX2PB
z#n2tDQ6BS2ZM`cD(1G*_BIoaK#_mz5q^oONz#}f4U5%djBv(q+u%DmzsT?-CRyg^A
zuv)Y|#VfZBMseH9D>1S5ITNF}MXZ7=P_P|;<v2Z~tN3irTknOs`y;#c$cO(5Q@tDE
z-A$%;zhVt+Ki%z#^e=9qJZ&IqC<qsWr$fjMzQNk4b+l`;q$KlQPLkIPL~=4Y41?(j
z41K0Sx=5hR6@QvNW~3UBEBn~jh^udNRJ$iPqjl~3$MVOZ+;~44bxYN1aUSi1xz}qm
zpSKoB<R+kWSrl)kO(7A=RhqWd(oeN8qahz;^uOcZLxL*%fvKSLr4Bx6shu7K9{?6^
zfsAb>ol(hQ{WB@nRC&cmNd`i7>Q{uNK?9CnK0-K7sXroz7B1sTX$Jq(dQkSpm@O!0
z@nm+HeCMz+{4)I!q$eK34vlnkwPb365mKJzO}$M4ii_{7;Q=*rUkC<!NV-bJM1ZMs
z#PrN(Qy!4;Qg+8nX?lH{3}bPYKbMwINjy^bifK!uZmtS9k2AUUX3qARL*Q^PJ?YY0
z4{sx_r)Di`$s-3`>4}WC>2Na7)*6jNF(;7EGjvzBp@;flYHrXa#{z*=TPMz*iG`Vd
z#$HdwnI!rU>H>?CCROCyA;BQCMDeK2hFvoLtx^|GW|Bg#tvflw?fJvR^(45+Bv!u7
zA7^@7(oTo!0IL-M$Dl*)MMe)Oo<Qcb&+W}OF`0z}A^^07?SHIApXha=4S2xnGStaI
zf^4b|Pl3UcuUFgC)BxKM1T4CyAWt-YV0J(GG%$|mXNd;%g*{@h{n2e+oKQTl*D|lL
zHhere09B&3z*+)PsK3I%kmz>#P!(H9wacK6+Z^JY?@`gu@^&+5-v9Rox%g%rY|Qd`
zqgk3)bij}mye80ZL^e}BScN&ZwbPYBu`LrhpUr2&6yD5@OFIv=ZXwKWUb8)3^eVBc
z52`I~akFA`5AoAG`8ehTLu=aKXT9nTcj77=H?||uh~JR{)Y9#Q^Mzs2AsX?6_^UZ6
z=120p0!UXcl9yojGGm%l3#P1?b7Ol0xPwEOVwU$4238)<o>bB0?}d7?-d)q5aS8oT
z527>%#XpttnYQ=JBuB9Trlw&!@pD#~?GT@Xb8wVer_<yXw|s1OWX(P2Gz&a1H+&9+
z_1!v$9FFY?+a2LEg&UsXQI@R@r=xa>9^okYUMI4>c=6oj;?LP!%Ni6$MBK&9vHgG)
zqoayan=+1a_-15?TLtF)Or)Uk(1AVH^8LSKBkk1ag@qpiKo=eqELs&96e&|=4aYSd
z&}qE{L=9cTvj@pHJo{Dfz=TFO_@MGQ#kue#p2@|lks3;^C%Lx8CaAH=Hma#C{s`bl
z9yo*%;Uq=#p)}nN9w$|3h#28-Aao|Zu)sz(?OCp9D*GqflRYVJlrO|89^49><d{1V
zdFIx&5+;$}3N3)kviqg$h|FblxYJEB!|6U)$W-Cw$*E?{Sj|#o<;Q{ZXoZh7k!Hl>
z`zS}Hv{Jpc@H5GQTZiXVURHks&$qi>aL~PgZAg_P@;&kTg3g}lX<}}40)KO)Y8`Ck
zq5d{R3?Nsv@GmbL=#spN$1ut)o-99={8(JzG3|o8h%H;4<misCzQHrV5+pc<vr!dL
z9C7k*q}R=O#^aFN!vR`a&4Qj-x=sm0oTyvu)YcAHN%EHHJc*kQXm*A{$3v&;)!$kr
zWtNS*U~MbCIT>=X#H+Rbpg!<;Gk@X#;e;j0hse9(3n*=$wtIoKKKRFKx%;@<t;q4r
zKUp!PMqyu3;Mt^maK%D^8}63pqon9*3`2A9eIk&kN7m5&;7Dh=VL4JjFb0Ba-+SUJ
z;yjT;0f@M)j9@|^IpSWro8r*X6GO}Yc|;S}0mGpXAMGTRid>PAe&_(F9l2u0;YU|C
zY5j+EwCW(8-|6E>o8yP(f^SDEialW%{Qjvw4pG^q^TBmR`%F|}N~$OHdl&z^xR`&M
zEM1MmPTD_FVW9FBbv{sqNX%AmG{qmL`S`e{N?mMOe1ex4>1~Q)#7vd$lJLvXtfMl5
z@v#h?WmVHF7h9+<@ax^_wc=%K;&UL0Pb`*MAp<kBTDfk#^=zPcz@A}_rt?+RipMxe
z)wIvz)_&F?5h?eL5Y1U<sf9D0<$=pSg0}WcH6Y|ht*mmRlE<%!ESTW2p0G6Uh+~-H
z&kW*QGGtZpt_NmK5{r_{Yp_2`g1sPkCARGLerOA>wS+Rv^2bAVt7K&hhmLq^g){fy
z9W0DcBM*BEO(6u7{%v>vg_PDFF-#A}+q8{L%wyf;w<V%PhL5YXy`uz6EdxkKc_|@2
zi&OXjkh@h2YP$UB3^&e`05T{y2Jn3F#r@4-8na{$0vFc@QJQujCZ88P$eH#wzl5S?
z?re5Vg*gE*CRQ|M+jgNC3MlK151d$=knCj9+YkL^JW$F+$IafK{PU(rhI=h8-iDN`
ziuV<mui%&AR(>Ox37mV*EVlE7$Vz7(&#Uc2x!(uy;~B*eFU8JYG_=HuPZe01Gp-tR
zG|#b#O!^`skzz{m9T|>z3(>AD-}7_Teo*JN!?Sq2NRH37&Ph$%nDbcV1y}Cm3@~sz
zw$W{#b^Xl{!i_%c5H6C83OI3+m(lrqpoR<;q2Z!~FlN<SzwFS7Kl-ggFGjmP_4oYY
zd(KRkTll84Al%z?*oF%}xke-wI)fHu;T#RE6;6iy$Rm1p_%>$w8ih8hWFaH@of*w2
zGs>3Iy84)sBS&}lZpqp%YLORFV$Kr{vp=u4G);l=u@3wIz;vO##<s^*`HYAlv<PX9
zvcSw^O(TV9O9MK|Tb)(Ed2U0C$gZE%O^F?Ts{Rkq1LvPGY0tdlw@sd^&55TillKVi
zDF6I9{lKw$m9jd#ylon{BQ=Pp{$<&O!AR^r*Dlkf`VIa)CMtej-5?xbbW7y9fh6N8
zis$f|11`--%Ae9b8w)7g1YBrL8Dueo4HtR<dfVmrKs1s#F(fyi^|NJ*k`RGXzvhUc
zEfn5c<JzHW^yfgjSD|HwsmrLzaF|L*8~+Y}sH#A}S!zh~F^L5`k7`7(k_-hA8sX-2
zz?`zZutTB_soEk#MC>b+c3}E&W^{+qy_ZR65UeB51mi(#Qw^GbLtVywXHG_V<^<$o
zh4;7`D&Nk9BrCo82W&*E7z<Sb=YAM1AxUM#PsCLp1>ZPJ1q<37PQx8bm8wzmT4dv~
zV(0=y?$vt5LR0+c;Sm-+n2j@*v)@&4P><#t>t@!fA0)`_etga_9=z3#2*xw-DMUc^
zQ{%=qj2!A7`g6xgg42!4(NsdxlSNl8Xz-~Wm(r0U0Pc1XYI1j|h!GphW{$^iu0b#_
z0Gk?h!u-frwkwVd6Tp%#z0<ZeTb7^+pDcTBsc5twlrev`yv9zd)E`k3ZSB&`RaT|#
z@>#xz!?mO$d$Z^}Ua%ke8Wc^5{dSs?Q5PBDelsCLIjY8{Fmi!D*V9oy$B@WGAobKF
zMRIYE=c&vQytzq9Ge=qE1><wVSi?d}Fr@K;5wk`b8Dv2$ssjb#GO1URV0O5|gaGbK
zXxKS;H`f)>k36O@2c&3b8qdnilL(}zi?=K^2SjK(<1}$8;MciT_(!8urCz12<y`wy
zy*&ArA*1Ua%d>a@N&nh=MLVX2(7Z<_{978v1$#u>W&?NAu-$;DxOqT$JPo31q&=H%
z>c;_~ddd7QqBNV6utBii#feo@1E9GqWfWl*AF2y?Z5$df7I5fF?x}<`V}_R>dCHo^
zpOMHju1$4D({b$<2SpHPmb?5sVP-WU>=8|Dj}@;DWLU+0tXBQ5%UNz((D)E`=u*d>
zl`lmN0%TrYdJo`KVKYQ}q;)=E!l4%yLZ8yem*=>uGAG>|;P*El!M_{ATjC!*fKH0;
zKbn&Ace9i=A%rZbwQ-_0+vahL`9kNtc?qE5WLejPF5xg6pbx(kIx?aNkO4<gs-ywH
z;)Wv{yq`ocFHyXi@e&PTMZ?<uq&8U8@oJ$YBOb@PvPFAxby>g1IV?S4@x<a2)a0E_
z6yX-D>-+#=XJReS#o94))KBv8=eS~f3CCC8wwyPl?4%U{e9vB-LppjSmfY8LEn@XQ
zi=_<|S6+xx$}1wBsmmo|4FKXX7vjO9`?}L)^J>o=sgQ+gs*wdAfnR#cPtcmb^=MCA
z0WDYzzJm9;l+PrAo!&<I`NDBDtT}!%F*Tvi#~hjXM7kWE)j*>QDXdi6aeF=}wt*VA
zX&H8k&qKWTJn346e#nqheKqE65xHc!2!o@FYX4HPMlI8QiqInO%@b%}tR@yhBs&J?
zwP(U4oC8FStag`aDcAB->2n5ySN}mDHP%BV0FmvO8KbEIK*6M;E)^J?wr9*c+yj)l
zzxVN`{jxjZ$yFvqOZB$o_Cpbfizo-)*9IxZt7!1t!cLx+E-ZE}PvrpRFnez^EIJ^4
zmyigVs=irG*1AHy1l1@-KHQT`3&Iwpv-{i;^$s5^SI>j{<R0H{m;Q2+UO9;PXT~&i
z+2<W*`fb7^vV*NOLS~Y7vB{JE=?H1hrG}FOR=<X14_z5FJ@-=ViiBaoGfO)1>tbAI
z-oOiy)6!{z7DQ)rhHaZ#FOqTGS*-2-^6f8gr3)Um6TkvBQZ5<2<L3mMEvqrK2NLyc
z3h#ux)>3Px+}kY5?eF!5u3c>KhRSYQQF?u#`GRH{qtY)i?-BVm-(i(gZtAABA8pnx
zTc^XH)ICoiU!M+0mnhV<si*Z$glCV!ovskkQE$T_zGBKZclAen$Mu^ujz4*1BtFg<
zV{47k2Aqddy)%ScQ~D!OT$tS7)B8h2-3-*l1#P1A{WZ41d;Q(k0%%2$VV`9s`4iGp
zsN?06)~rR5MwL5Jrj$RxFxD~|br)c=M;(PeHk-9g=taSNSIzHwu4~hoB!uCPYz16k
zu0jwk9(a6ucB;g)eY%vH!Q#*?ZLXW<!Ob7*v(F><6HxAHZ0qwvp-1lYM0HmHe|&5V
zXy5_(;Q8T0{@=mXnmjqmHu$5DZr!7hGX6>m_c^Ptq|6JF$6KqPb1DBDVjk&%`&f;T
zMr6ED=W99UEM0AqRo+o+kYTwo;v}lbj;CWB2Gd(t-%{7iv58iR3vk@Hw?QR2-8PYY
z$@Hcf=<hgh6_jXfY(8EiUnf4#MXot2cFLg}d<>#AyW5iZD;x%r$~0#)@o;f#g>{vu
z=C{L?q&53^0>Z0vC1b#AVEdiuW2T&hJkdV#2#WTwYN08-;C4BK(BhjRim9{wXDk7{
z+2`B!_dYzD#EmSwd}G%F9ksNGdudHoO7CbV;f>A)iC?|QN4DyD3-QEjB)lKl`@bPA
z(gpvyc|a+~ErS&Y&&2^AV1!$PNXCf;JDQDhr#vWMFD-qA5@zK;*y>}Y5%;F;2K(?g
znLsEbA1F@zt?peIz4w+x3Vvx;$XucohlAD>Dlgq(J(g=ay)i|VE|pJ}Qh%ugzAp|E
z<%CJV^v%haq&)@(q+)*l%!glb)%}%v$GNx#QPS{ioorfrcJSGG-d9}<(_Hc0Y6X9#
zDjp`}Uj^3#kVdS2Oaxc#p~Uj7R}Q;7jyh+_F9_iCF$Rp)i~~rPSR2;$*=-2V$0gWt
zJ%QlPMtz7Xq%%SY48LzEYt_?d8PI=Qa-=7Nf8g1O_Ijw~FOy#84c4dgNU`k>YzuEK
z<DL8GH~0#C;CSo;bEq_^3OaFmWu#?~N+wBWaM>ARPF-4K^LM=so8NQrX}4*=hW8Z@
zkv85ZL)j7KIcqDZH2Er*i7yQPu#w-kIlWIAa76^;QMY>rYKebGaV|=AU6(SH!@&oF
zdgJlbX39ElcBHHuAwG+%W~biVoPSFBT48*9P_wQvI*l4e5G=Do@ua@Ql{Jzz0j0nh
zei!W+&ovv!;oTmS7PjAoqW6loGcYHGEK6)knx)&T6Lv*+z1hc$$WntKCPgFXo4M8t
z30q-*V29fkX@*HM8{7+KqjM+S<d#wr)z*(SH^a252tfDW23!ealsg@`=-+=0zwr*(
zb_AYdL1a65jWZjoTDhPhcWJJKato;SH;Rds#La_oNV#%z^Qa(csQlZ7c)A<l4==Qn
z({IMO8m`lXa2KSuOv`&SkSUtpz<~sAgEI}uiOSfyOI7YKo38_+5=JFi&mOy%LL<>~
zFKw@SqC(O*^Xp)l_#+LPxe((=KPS^^HRfj4I_1VJ#U7p$-vZ-{0Z$&J2m<JdQGJD=
z1IeSKGk-ZEjvgLd`=#XHI_Y?G<v(LCr^m^v!3q%ci-^<$;^KD?+`G2QkV7d9ni~wu
z39d@xwn(oiILAi{EJyMd3c)Q~`p##r&p^7^50l!Ga;6x>FlB+oScnTVKcKnr!=jJN
zm)veqRb_TG6<tLQIaARJIxGr8)hzdnJK|5Kyt^&mFMkIRo@2AOqF`rTxLI#g!gCkm
z|MKjBud$t4Vq=?~S#>c|Kh+wlUL>qwVWnNo)ib&$zH-mk47h{_>FB^XZl4UpL=>nk
zaHQCgM4QL_jitd%!H)C(VbW)v-ikSdCf)iTmMWnn?I5RO+1h>C&fPYnW>}sGNYyxX
z63Xc`bML+Dq04Mh`75jLt;(-^7!>=4q+RDLkT#>z9nTK;MJlx2RyG5)=%&Kx48M6l
zymc5UW~sU7Z9m9)e_xIhcby`~?YICBSa@1MM`NlyM6_jelk>&UjfmptE+Gm*tz3)|
z0_HHf_*2rH^}?<9oE!wQ2O`;I?ov7IYYcqyh{(#;^mt%#U{qRM&Pt)~-PkFB!qUg3
z4ICnH&4)?CNyZ*01ANV<>xw=)i~n}fS|Y&r>;<@Jg@Hw#7X`gd>l(>b$V%?5W`I1=
zG>TPiXPziw`_E364v`d3+X4l5^t;>^(@pl1{W|Z*$JB-Woiw6%aJle9@(OwrDePx*
zG^1K>KIWogfkdpIgwFz2lD8!oRYoE+RY&eMugl38J}m34N_jr{Q)(@9U$BU=+VBF*
z^f2&7GD{~fsD~+-^ZQ=>l89Hz1F*s!_hy~W^Wq95kh7GIx{)tm9ZUo7Hrxt)P5Ky>
z%Kay+R`|LXNXv$(&PPhP5sR8=+f@!T5hHBE-w6Z65p9Z_zb=aKz4xJX-hhIhMi{VC
zn|Z_QOH^_69z+gJa!^a!W!+mY;zXiLt9rRgH(QH;4^Xz2zLt}paTH@EaD{&Bs>*P$
zOaiO?g2xfV{UNI!e1rKGshWWBOyw0RSXqB0nUK!&jrwUu>R?qc3WBHG>afW}khOG$
z$eUT}3kjBWCYaKq`wq)4kt-LU-kK>*Ba3rBHVestvCbFwU<#I1h|56vFHyqnuQj`Z
zxgipw|0WX_pm6Sbr^lDE%NsQp#?&l^n1n@tZqZvNkEtKLSXx(b7!knLTY71mkbB|m
z-#06Qbi<ofE>O&9zq;kji1`34)f?j|^+4!nN&Mv^mWA}3DIhdrQSx;}5aa{AsBU}G
z>C<e9gtFyt0|c68$sbG?x%%4JmlGbF7f(qP?~-%%lxz{1{5Xr@EqfM%d-E^0Oz-&_
zo*F)Dq?ie&7h6onxj?KA<N?S`$(EE+ujiYQ`twqig7Aa8M-Eydf)f>>ml6!{Pa6$>
zpBYXwl=c=eO35l0R(`c6%^ttMXmA!r!4>GfvmdS(@w9x0L_qq))|{FIxW<8-ulQv$
zfY4OL)3>Kr^J(?4<7O%@O1yaV3~d=pxg>spV&^6ROZthbaOv1fT5>23-!Tct$ROdY
zdepg+Y~iB+irUo5B8R#hv!uYYixGp2IU5lj6LvAb&2_JpK}sPZ&{3S59Ud;EKl%d`
z3)FE{6Y^PWaQj)J)TTfus6hL<^=SA_P{OcFoJlh;`#-M*qe=9%)%hzbY?-6U=OC(3
zKp?ytAGa&ZqWY42YmDgrT9ytUAfic(DjktY^infNmog@W31bvj$*2chGNtBYGpYXi
zLG7&>2|?xT_t;AI&Zf1+iM-@9P>`t-MQfTg!r{9Dvh7u|sEOzX@tF6jTp_!ZJ&xM&
z88efQ$Ks90KkCyaXd~2UERVz}Y2TnCF?BJb7W-SrwSqoSN)8=BF{yeGB#Y2SZ#{B@
z;G!py6G)rzL;qYARC%F`7OSXZkR_EnfExzVJ?JU_bwYcNt>_8k-vhOarpQxD;ZV=$
zOKK6t!^HWsKgk2anx4_W3MrFGPp21cs5ft&`=en%6lg-i%<{m^{#6gFv}s*U?h(*Z
zp%lf9UX4arYqh5bXN9`5?6*D!o)F9OkWMbAiVrv8?r>=73sr<^(dZrSNgn$1=L;)4
zEP8HK+$$1N*d{P*9+Vb$@Kap;^Wd;We!N)k7XIAFEgt|1UM+%fVS)Nz_MCk)1eXck
zt&V3@Nk=J?4eF9BlTIV)0wLf_AYP%%?}?r>0c>QBLh>c|cjh>rS9S?K{BGGQh$R3`
zy?>eVVI2D9G+h094O9kC)}8c~z!&cmGUsI*$a)6zHY;ckMe{e1??y6H_nD-@;|Tt8
zS;(xhDtIC1uH0H@47nP`DMQTT$IwSCL#ykej&L}Z=1uH&-a@^{=;WpltG$@e1UX5B
zv1{rlt8UXB;7Mwb2clRl6=EJiM1;6KlSF{aW&e)1CNt0cFmP9&+dR$U-OINVyZQ%-
z0mIlLAVWN_nJ=j0*r>883E{<wjSL2g8Gd~C119wenf@YMrhMn|eWaJxBdy{;p_2@7
zM5UyobF;+v;efdeLGk(y%PgyeON=uo71rof9-(&GTZ>v4*B|p>Ifkh}^c+@-7SOB@
zWMCWIWiV(=s0hNje5<*pgS6s;;|_3~jI9uFz0zU#E~;8jUJKI2iL70hJ8fN-Ej9b>
z0Wdie63m7!YfEieI!~QRwA`go<i5?E3MVbOQlTOn<qEZNf)ZGFwck^$<qODA8;fnm
zA}r)uf_eZc8XRva+KLhWSH9`qWIH=iX&-5rf?ctFy@~HTu6L31$1`_ca1GL^eiTo=
zk!1~`E$9oM_~2?4d-SZkvmLmr(q-e~lF-S%ci>IkLP>5)(ArmbmsjxIM<Ep(TfjU^
z27>G#LA=k9`nLogGONLMYylH}Zg%MP>{WLbSTpxP4C&6TE`CJb;ta3ZguVFSpe#(6
zEBjjtR`c4Rk--DMzVG<+x!lNf;v|Z>Psq2<>;dia`xll0HeU8q!>LP{9c+Ae5pLy8
zs*&DPZsffh(hu+DK}D%|GE|^cs&uaC^U7vfY<|XOyRN-5N@^hq@7=d=WICV&-_96N
z6~Z2tpLLPhkoo#iJb;HXmbm(2nEra-Oui}QU(|Y*C^v#o;uRF?mH@s|$9>{{EBP{I
zYTEnQ@xt{aP&=ikU+m_Y$owyVH2wQDd{Ta}rB*NFm0$xQ!zF&^`Ws|se)>kRKCKzn
z{rCP2nlO2sD6VbVN~DGrmt|ayz}&KrKvA}}f!J2#Dq||jyENyU<33%8M<tTeF_a0D
zq&%|DHZi9rIeUBupy*rjxs)N~5xMVBICudX6;_PMch(foj;*&v<N|W98uFx414y=a
z;c#i5FH|SDhtqUfBs_HX+EsS&Hw|$~OUfG&xtdb@LO#S|4@Ly@#GC<Jt5c3M&M!l9
z*9Ar$T^Rt7YK6qD-p`n<eK2oaSWhz5P{p1iEsSdx8u|v$pBox<;9E&)8dtZQTZR+c
zlM9|JMXdrosK;jXn-PO7@}n97dD6oafq5e{m>e;k-=FYiEQbUp6e4SLqE9p<()U9f
zw70oo9M`g+ixDEsPeN_NueChzUFM_KR15z_#l523P?ob(aU?WH;sS2gau*q1c$p8q
zN-x99SH){}Okx0q>7raFX6H@g;DnM{Mc4T3tQM!n=bAU0>nx=_zU`QpL1&%6!%0fI
z59q-Mymh;(EBvhJXf&!dG(z7il_i0!PgGO`FHx4P4KXTk7Y~A}3;68PuhPRsq8SrI
z`hus*Xd2@x(!Zk_AF+Tz-VD)il(0AA;r`Ag;zO`R3D-XCD{mIB7xQ`<{#&#5#I}bH
z8KMM@xbv-71G4=b%rKIVE;^KJn?i6zlmBI#r%9viphVf6-`Id)_~u3lI^B>2JZkQ#
zthsV*O_?%TR0au|VEr)$&4nTZ(Qhz3OHfRuxLvc!1-#MU=7dCDlE&4y*iS8rY}MZU
za-@5{Xs`&$+c54mjnyw-ntR7b(2NzAmyt|rSWAd+w1B)_r`TqLJW3xhb(46J(|xph
zT^6eijUd3MtuZ&Z1eQn6pG2Z~QpL1RD}n#~fxy)mpB6QN83T_33Lp9W_PM=1`(ALl
z?J%M&H(tlSIAGR1H`4u>EIQRAEGJdP%1K<(w>Lkl5G<(jGGpX)qwoiDJKtPz@=tCR
zUZ2o1Oklb(ylnyRP@!}#^9YYC_Dpmcc#!Jt%XHMKN6yyyMZAQEYRaH~ywM)jlXI!i
zfhbc6l9O?XA><lgF#Y0Hji;|;E+6-~x%&#CL8}A1_<42$@WMqke{#1NU+Y>!mL3=7
zPS-4qs`K)Bk4vD=S*o!jvBpM&Af5ays#D^pfaw+8!4S7UI^VGj*kM;VCJoYk6)93f
zw~sms<Ah{{U!kiJV!BW)+95cU4KDDFQV<yvKOglJc~jwFz4Q__MKHGhV1<i$<>*m8
z9BqJmF{9?x6ZR9Dh$?h8Olu__bm`u=6-s9L<5_Q0e@My^DTY8-JqMz!hL?df{CJws
zJ-7y;RVIJ;1%08B(4I4QA-Kt7J<RMMMfKJF;>Gc2i@2PE<=#X(9A{`Wql>&d(8}B?
z&f}}LsCM&x*iqS_QC;z^nj*O82>0w)vPC`EBifG+jnox)uhCgdkXrR$dypJRE`g8(
z5fZ~~2q+E_$E!HO+*1f8W~$frY|}N&cLvHdF^X_zLEKtZC@|}{#F-X?h&_r}k)^hK
zi>Pk!lgS1YVSCO*?;8T)XCbQ$bW7Hrv%7p#DYrif_TX)Q!}r%@dY{04aED|@nMi5D
z#(V_Jun9uGFeFb=Kpy&6FHV7PXow!7s-P0lYjX=(P@_jrPa$^6GaG^yaVc()ggiH2
z!7p4C0(}xYOitXz^swnCpgJEs!wTF=Iq0-192B6R3c3~g=nu@WX4KEtTBhFknbdS|
zu44L(DbbZNPv9=0Tv1?#1f3BVb(JRfypUY!l@AV{!Od82bknH%z|YD-nE6x-ak69b
z;2oAQ&Py2xW(G)`+UH(rG+BM%)R$ZuYmFD7723*`HCdPP|5}EIJdwT9E%KdiQ`xUd
zR7OK(Tl}6KnX;ep7#u&pfnjcVp*cCO4oTwSuIu0f_9;8^EO!~lvy9tMWGy-U`>|LK
zrQIm)Nei~jjEYWmG~SNHa~Olg`cwN2;v1sHCQ_Elo1x-H<jV-B!9T|y0=u54U=C1M
zIYd|d$E`gdFX(5SIo1^D%1O1D*>6HkO0-daX!>z1)(4pzPt>e6=C2bMdUw@V$Sqj_
zz56x|jD+R>ir0qtxUU2-884l?7`3fZzq^H8MJzr9Qm{=jF)A0PY&jV%EP1*bZ8qnQ
z&T=L90<R~dDq?7FyYSl#s;V+K?F;A=Na2St@bU*KOhFQyNWn4o_oy!>){TljNOWR*
zHv(#4f9lP?;ZtLo=E?3Z#Yl>dp0Ca)!B6j>9{_-#JUaLI*@}06DIuxRpLEC~Qs0N?
z?Wd(+xZ(5*poHn5>uHKJ9Xh5XcZD=9Dg2(!Z1%`6D763^2zp@avhc;7Fq$HJ%+h8a
zJVl`D9a6N<pWB>2fSvaTW+tFQMdUS!JR!F#W`SC^oi(OT*$?N(6%tMU5|@?Ao`!kS
z+umSm7T5Fvtv2mlQL@VxO;`NE?xNoj+}iZ3@>Lg}94`Xv{@wMVJJab!iZZ1DntVj?
zY}7^1oII}{GbDYGev<CDr`FbljR{<7ZYjO5W5x&&9U!jthOyN}Frnck{vH?@4u2f;
z5M>bdeP0VCM{Tq1aukY5kI!(mTFL&w+QogY{!uy$U{ym@RaHF?@NvN4a(j0q&fO&T
zM>y>mn?a$YIJvrEV9NiRT4_6|nV+6P?AUV(Fa91(2yd<DWQ|$%SRjc7Lf)X@(>Eca
zt5DG42Y~Xcql3B|;$SJ_U(M9Qu)pbT3={n>mAWeYmk0K?P>)#ikIR@`M5T@IQs-{q
zG_-j7P6a%OqN-KNhQ4^;Ig(-kX4!0X@YU6Mu3VoL(>9tTA(!oejY*5XDl>~^`rmcE
z@h<HA85}00hOexv;5T6uVY>B%DfP`Kb-{6E>}EaRLfGOLV?btxKdW6?OZT`Q*4;em
zl{qY^DLvC8K(yvuOmI1}HU-3sA9@er$>H~Za@}C2|8K4v`0R{q|I14QA0LuV)ZEI+
z*a4qT)JosUSlHOm*2ox=mlx8}$-!9P8q#eeS*71*lO4YMlBzBA&$$4Lxd<W(I;i+K
zpu<{AS)hn*FwuNDfEJ_Cdt5Kw?i4!|^-CPC6?g2;RM#u9lB=)xW=Rd6u3(1u%7tfJ
z)M`1D`S<Hh*48$Qp({_Y{tu!c-pA(hWZ?&E^fv3%MURg~PRQUFMM7@KjobBL^z%Pr
zJRP{fUgRr<*V%osj$%ok7lM{zAMTfPudUKXiN(}ylA_Ms(M#@W+wzgL(?PGOYu~kU
zLT&G)8nbym{SNGreTDQ{%}ez)Xm6ez_z4n?FoxzfjcvRTA4Jy+5FMmO<?sB=V>EBq
z(lpwPR@+pLtelYVv?^7Ldmf#xsYR^!k0k0IS?@I1!OD(*WoK(m-8r+PfBYrME`LL(
zcyA&&eAlxBG{5u>ZntM`PxDfqHtSMpUvy_WX{9^&S%|5sS1p!XE{UPPeQPkx-Pd>9
zGRKy!zceqZG|8v}O&W^=$`!^G$lKzE2NwPm<~a}5>k-Xp(XOeJ_mP`7(Ds$}l~E!~
zLx@Jxz|4lRYq01sZ!P%M6Ch2k`OOE+GZGo3b+*?HPpiCIvPt~0XW3{XxWe<(m!i7_
zWGC9{EAYXO8Pv&k>;|Ak=h{o~l+v4YK2Y1*1mvnA61#S%v$v@6+zUw@l1{Xzy9i-%
z8^}@)IfI^^s&mzFpgTd!U^q!l;9=H!uqjv$gE4e#l^VITVftNo=(c|2odnh`dQtB!
zM$*`m8+rW^vKvyI(?HBxv*>b>)fhs(*652~9f_o)JPkC4XI1LHG2JMav;DQ|eA|l=
z&!=-r^*;d~8omIxD3^5Mvma0gjWdWDR4W>Uj}gu$OOu>T7slBkxP~+9{gH`T`|}br
z9Cb(1*y9+TSw~T)LlQG<m&B$==ftK5;xl(Q#HMGi2~19$;}|>HM^Rf@N79cZ|5frB
z$MEbvg6>Jp>VHjOdU6|2_aHHK`@Os;F+I@wDMd{2<?V7C$J6OLiani^`eAO|&oU?d
zcgQmx|J85qrWS*(*<aQ=Wv<CVa3<U*5LWo}{d8Tkhzw`qfj3tk@%Tg^^r|-Un=}{6
z`&853O?iK5nfV2}33IdlA7NzpPcSn5AHkT(58wTWvME$_#NTBJNS$nK4nXuiM_f@@
z(K_MpY%vZBbtv@H^Jy|tuKYZQYsCX+YqI0HcYLz=Nb0hEcX4>6Q(61N{dZ(#bUF2!
zqNBY$^u^)n`{~U4wDWTL(e#-c_IgC_qD6D{(uZ#w<e|eBE_Fs%!<Q`i$9nf)iC%1n
z_x<i6j7mlRV?K?J+`ZG<aANr3>h=2QCSK?yv+QmCNL=c-6ZrJM<x)V6Z!C@Z1l!+y
z@z9?}##hZt`QIrW7Hou3j*M^;md=c=>rWeG-3t_2A||!%p6nwu@7B>cstmT<({a4j
zZ<{bwD3(u5+MLq+Z0>E5#?t!7!yaFy)_<fNO;)s}i~T}53-b=;oV`$)f<3XDg=&re
zCN2tJkH3$WvwW~#=0k2C%{M>nH+$dB@}CuHGE6Yudp-{{;!{?)lRF{*ZU@!2%SISC
z(VMY4Ql>!$BL_!8Nsv#$&_im1VMDU+HSGt$1Vcw<kzPVa0Y(+@_aEsguPHu8`4x)5
zh)&b6pi?pSN1>p@pu;c$hO}2gWom0T^R*TN!>bm`U=17t(OaFnCj*cLS*kC=4cV_%
zD%Gs^VHkpCIoiTnwab15cw*^|Iv<ru?MP)sT$0#LOBj?$w4pf<VoLsU81T!Xbe(IC
zBi)JVFNdf9#bLm*-~Z&$@s~lQ|HWbL|KgC<FzTJv^#9M{W&%SFsun|tj}}wOyAzpA
zgr^bK$ec#q53(oJN^XE|oli$8$faC<soponbHfkx&Tkbd_t|vFG@FANt%^zP1mse>
zEY+Fxmqi$!k*vl*G%}9C!o(~kz43JR1SS`bag5pU#H{*NiJ7rQiJ77Jto?0?nYr6v
z7M&89y12$M+BwG4Po(~<L``CPMMGkGCpC`orRWfnqv%kg^GmDLc>3Efi!}d~ASU_n
zbR>U4{9Uv96z_{u8IC3CU$J!NF-LH+)18V(bKs)g`m)<&5CUX?u!7i+c&4_V7^8O7
zi=BS5Dp{qUJHB1CvV#!XPx-G?&miHtaxza)7*>}c{}DvS{{-UynJWH&K<xg%K|G!-
z`oBPoFbWL5C-B|%W;#@^@EGr3G5j()={8lj;q~4esln3q%JN#d=;Vt{Eq}Iuc6N=h
zvAL-4#{Q<~^VogA`M7+K{FTJ(mD!6HkFsB}>VY~6&FA^c>1+I*tNgL*^Cs1g>->4;
zJEB#gY<Mc9^|jrr1KUZF1NZ61YX8;1_38F{X_sF5PP3Bsv^5(_zkfJQ=~M>iM6UTd
zD0k|E-9!GUVTl8onO`|Vw7?+-WnYG>Hv8#wMp|F4qsPf8wfuZdt@gByD>P-c)01uL
z>xKSe*rZu5GPjs}dLF&CEfQH;{;0&)Rj~<o2J5{!dG2&KSLW2DQ#o@x*b(=BcyqRV
za5&!$=jOxZ-dtok7_&it;I+~0vjJ1ft7+b?T3MnY=5zM*J}o{u`7^N%Tw^n^sz^4>
zsfodyNrECBEEpj;5=@Mo0*Ve&4Fn5<Ww&A1A0h}UDx2sGDheQ~kf-NBPia-|F4{L=
z0a|#Hf*A#qt~UZ21sWBC0Wi3=96Vi9v$5ML0SI0>UjnoLfQ#Pp)G*<PG{8cA;d#J*
zwM?;UxffLrG}F!o*7BFkr@tGf?vT@AiR6|<cK8{w?Sz<Nk$5YzLm!4jUzSq9Da4d?
zt#hg!?U893{n2q8H-pBVt>2<AgrRGb<QGVY-}M^B)_1fW$C}A7;+4rn8e?odrzb*o
zO_F_rn_)eU?iyK*F4S9%vFP>j3*)~r)~NhOy>E&a)G9uZUcK+9+~14foPcN3rO6K_
zH#&7+S+A8`gk0y7xxaNY*y+DS=IgAa%hO#lJ7GD_!0KcjLC4BE$ol`vnE1@vF|p|u
znXJQz4jBiL92p0am_vvw8HZ4BVpjeyoBvZFk4}h9GwA-S)R>f2-<p)wPxoIydPx44
zD*oOcNX-njeib36ko`X}zW%>34y7`SJ%E*-Xj3?t0T=Gjm)aZ!<skup5x{=E0=0EZ
zE3s#uSM!%r&Mf-e-O|M@x(T9rpG7-*J04t=hyVO(m3+4Ok1#U*cNkfj82-0aF<V*2
zwvZmV>yfG}RPx$?#D(0SG9u_b?stc3QL((E3Y<l)2narw|9QFksFNUB+1(Txw{L91
zeY-w0-I^MmyR^OX@#smRbG?=wGVyW#+VE;)(@?{SW6i~*p|$z(aQQ~A6H>FWCb3m=
zX|>I|m`zl>d0|&^9sKN@Q2$x-ou`Lm|7qg;fmKnYZJj;k+x)qhkNtIC_gJd%{?me8
zt^2d_gE83iN>a6xalS|0lZPrx$V-IKv<Jr(C9}#4=}cGb6x>z4coPN?y>qoyl%~D0
z!(z>o#f75vbWK5XQIauZ`&1QUvjA_qGdhmDdR_CFBE{h7pJ9)=Q{KyE`@+5CVz$xU
z_APBkww3kqqlcg<ZrhC5yxso1wUs+jh|)+WeS2~DvZIxMu5B-u{D}V9&>*yaQ{9^6
zK$0R-|6rx{?{PB#Wpg_n6C?(cr!6!^v!=_q07n+}hM;(`2EH)Pj=%*3$hELc51%0}
zjRWN*4*d<|#cmBn5!r#ZDCsf*Up`fM`PRX19j1|xEHO?Z8)Gkuu>v?PJAlJZz-|9;
zswCpd9{I4BU6M%_kXRq3s2vKEs0mIT>z|Xeo5}qohA)=d%Kbv92k5?BCD!V_$2xpD
z9Kh47T0f|fF(gGE=y)c=o-86c<a^#evI1ugt)XtRyKsk*weYl0$v6VoH8P-*?<o`$
z^w6n(=tD<Vw~>%K!5Q{L&X0)ke*|JY`#VTaId-=Y?4&2%qPT-(gL!C%Zm>b_yhYQ3
z2S#b8{cY-5asbzyGZZ#Q&jHG%`gX3NH8eOL*smqctC`!bJjze`y+M<`SUbPj*mQ%+
z-lYGrk(vK{BQtX_|8F<4b>jbQWcZNVFUVU0MUT9+ST7au@F9T-{mqJS)o#)kVi8DC
zj5yYRK0LVM%POz}v8Fp1y+DcMwXfZ|l^?)9UsLN~P0qH&Z<)KJBbm~|G@p=8uaJ5C
z(d#e4-I<y<`G$+rvR6*)JR0}1la{r5$~>sm`-&H~5Ich<U~f;xAihF3xgQT3$s4ES
zyh+>35xBoT+YPw_ClNn>=98Meb$98GHb7d{tj@uPO*-`I`u%1%e=Uqb`L2R)y66om
zlD6#}<TTb_{f9rbHU*j&PdB8LP`%cSHq$d7-;O|RMg1wBgJpa*8_T_Snbr8QqC7Q5
zkh84Cxc*(%1EO+g-PdqzVd}+mg&s+*KyI9n_hUbUbR|7T6^q2X=PmoT=N~xX78&Rl
z8<^v9&zRO2ObxqMe{o{cMQg7fyJTy?;-EPmg&k-er5#Wm1*H+qhG}1CJ%Ky~9B@1&
zunX%T*v2^?pu5Xl-E6>%jRmOV+lC_@1>6Ycj_={IjPLn|bH9ytCiMRsc!0DlBp0^j
zSaHDJryj8H667Ov!i@PA5w^dgFfhncBOW@_m^zc`<aIPyH5IwCu-_q%IvHHGv&HKr
zzS;JcFtUx>#%s;8p|K;7gn9k8z5uP<nx*-9lN(&r+W!Uumc|bKkNv~)-}{G^gY|#E
ze|}$t2lVjWr)qA$u5LiXc*OpY1nUs->{eRI7R_qn(rkzl^5CIApU+rGRjNGjGC8|b
zUH4Fqub0kT^D3?XgT1#3uA^DfMa4`OGfTE4i<z03nVFdxEM{g#i<y~aG0S4Mn3>M9
zH_W~>|DD+raUbr(QD|l4s;;i8?*6h?R%UiqGs@k_N(Q%UN>@X|>BBkd`ue`MySWu7
zm!{^51s6AO?NS$2vy0xt3i)Qk-R0IuMiZ9~C1;%vIo6AOvwr~3>HGS7R-&{bEYFJW
z@vdAm3c|G~H!j}Mu;T&WJ(^MC?eP6!n`KU_+g>K8-nxqVl3t`EQ>TRfocvmzP5Jxx
zkh44*<alLgcCF1{>MZ9xu^n$*h)pXEAv78^E4LGP*&*)<Ogut!DP48m>pnFv)w63e
z_YNF*)JAmpl`0i=kg78iA03uoGv3*Dqi){a>!fDeFWbhd>JI0s8s4hjYG&eZ&@|Bp
zDeJ87&ZHc2>Ubm9=Ix(w5sP+{K;d?nfkoK(a??Y(V}ba>FDb|&Yk+d0N7#Voky;^9
z{rUzK+z6fhgxa)_EEs^qQi6fL=S{2l_qKx233^CrKf?7H)EGL&iG$!lW0skc>X|vs
zM9^gpX^8}FM}p8H_G1;|paeS}<Hm+;!&w+{!KH`;9kXP%-(7i_-dwf)@NIuThEs(t
zv>%C$FQ7`n&mp-_LWj5-AGRFW0D*_62{ZWam=)u2g#N8C9ef-70Da&v*x<rCTJ5<j
zNOF^PxKV7ARTS^&1wH7c%f<ksgohl#Vr)6KanCu^zq$hp`8nx|eoZCqA!%oA7ksC2
zI84@=bRoyySM?W;A}txcr`98;n*c76U&*}W7c;T46s76pgY3X|i86#K|EX#hkfpVg
zUT(`zud@t~Maq|wQvtdh2tG%ok41A^DT}?dS&-e4a6W0mmsBXM-M@b1(?Wk(o#w61
zexdcgxQC{u(m+u|a<;U6Ii1gn73S3l+C3OtmHhMuX@3?t@RyRn@?Rx^m6845mjvbC
zl7QOr4@qzl15pk#IuzWgu~z=RphHBzE*6`Mgz)xoaz)~{au0;f8<S&t+Pn7VrY)<q
zZaqQjwRbv^+2*L5CVube2zP#dUBy8l#mR-EnBkGh%?*%*@<kJXBv9%$PF(7ZmN)&(
zv1F|4`fPqP(~uqDz5Lex#9fBaxs`C%x`HUpxxLlJ+P~q@m0{8}WMeT-{bc;CIM*mE
z#pN(*oo9AZ!P(PKk*ecVb5v@z)Zpy<slO=}9cuKr7@LOVK^yN>U+?PqhAFlG`?@oK
zdfib*78m?miQ*%Yl<xD-Tj`hHbKAs~w7Vy_HJnmCE;Sl;YnP_HyY~)uOqRFFt)%OB
zcP6Q}T0&)I)r>iN#WbHepH#wquhH2$Er{aEj|@q(K^Im96v=%@uWytZ5M^-G$@DO5
z2AVxarDb%@-)b7z@&g9>v0)=n_F>Q=%>zbB1xaY81@l5Qv|MZ3M8HSAl$739ZZXcd
zwDhn<4Jq8=zLp_?bCFi~n+Cydg}Q9@(CaT9*41mmW+{j*B}fg>^I)3#bo8)2?85Et
z5WJ62*jjEuAMyvW!2%=RlcMN}YyCo*1YKg(JLETs-sKv>)d1HH>?4W@I)-tRvBVp8
z-)-HU#zVa%&Sd@6+fArjw3|>eZV$5~rgy!4pA=w&^ITMDIqr|rcZsJU3oawZEodje
zK?2V`Lk7Z@5X$z6O`hK;kBD*<!(%A(aXNNS36ht{Cz=QP$jP;g?P$QAa@(<>a4HJM
z+h%_JP(+xAmGZt-K5QyVfrmA`E<aaPyMxo`{H8E`m8<NUMK*o&Ks=(_efHWi+v(R$
zp(&3n$F=)x?Now{SN84Np*XEqb_1Z`=b`xTLl5q=-=-31yt5h3ZxW$%N~E5-jO$Ob
zT6Mc#igvC0I6b`YzB_KNLe6kEo#*wVEHxZF;}f)#sI<m-jtqChN9v4i(O%t@%KfF#
zu>Mz}VW*?}UkS}B2U5q0Y7M~D+sn*Lsz3t=T4)5yehVsnd07=bum>G#EZBqy2tcm4
z%&?9Vj9jp5<>bT-cFpnOne^)Rjb^+N4%fYE;g!|SWZrBF**ULTcRkd`mEANvHT2cy
z_Cez<@qWOR-RW{SBN*DgWhdi}TwiUCS*nOh5b(kq$_(Jyd8c^iJaHLzB{Q8m+&*K9
zWDMIY*}hGV=h4D|M)eH{V0l9+*{vbyhPe}#Nb|c}W_`gMVne@yMWgSI_rx0VX}kP1
zrKkFZ`Z<qNyX#l;Dj9dp)76_>czei|vF_ItgLf)weAzip`&X@TUn%d*;49^|4{>U0
z?rg-6W+-d{^zx)%yOQ5i4?eoqh3P-+hBPPVUWj0%H^beXUwH5lS7~Hcub9y#DO3$w
zS1Q8U?tLrGhc&Zxt6H7MlBAz!Jmf3Gi)JZQg8hDid;wMhTSoQ+-vV;$2HpY&+sFsb
z&9-olfUJS2A?g91o(D>5)W8Iq?_t;ilH(>W)-b}Hr_<^W6fING%4<%fph-q!pq!j8
z)h7y4?f@V78eBiWDt1|ut49T40W{rD1JuF{1SvCF%lB)i=GU}}BsIzwuv-q<ZR#2$
zVmetae~tQ_pGV$gI>bB_k0rJoNkx4z@Evv<vo)P+fo*zme^<6eWmU=ATXPmo@U%u=
zv=XTr5N#QS4aEV8Dr!oQ?U#cD+a3jt4OlEv$$UGe_-KtLYJ(lEO8Vlp8i_sV136;R
zt2niJ`HBE@Bw9AeHp@{RzdjB7NDd9h$j)^3M5EmFN`WWXXo^iosVC|SaMsQPH~-Wr
z3IB||u35cm{p`_B3X`Hy(%{B;VEVKc7^9hR!|#F`Sb|Sn1z@cGj1ps}Hy-0QkMClb
zBQ<Ae+vDo~zN6`wdrFzr`rpDYs)xR!>9%8MQs<<A%AT-6u>SDIJ6@H<n{QZ4k+N#F
zmAp#7rW!M?lo;Qk=>mZi)o=&BX^T*~j4`ZaVrz-N0PEOWR&{#Pp5T!4(TmxR#J#Fb
zcF>5nr?Hu!*B*u+-ctNHz@jZf*py3pyKuhmW?lK{8h?LKzVB{c`PeZ2c;R^A9Xb8T
zt@{X^d`BdI;Z0omATs(0A%E$ncTX<vm=FV~fXwf$yA*AmIP31|vk3whXq~8TB*F9>
z*$w^ge%iO9w%d1CgG({eSG-kf$M}CK)&DhZIw3PN!@n=pioG^~MFqfSydhY4%V$Wq
zRUA(2FJ6e-RGDtpxCBP7UZ@xAFZk}m%{Z;B${xfOv)SR!*ksDfgW#jVVtQF%O2Lz7
z38Cp%?VGJ(%km?`ptSea7Kkiy&gOxV$4U`57sIi0>Y?~J2iM&;d{wS*z(Z<mOC!w}
zwSwe8^zlshr=~U42bWTF7eklOb?vcj^sSw4OYS2+YT_-dl`8XrGT4-Y1jcf@0xNUV
zjC;gPxSG@fO%lb~FVW1IER3N1if|vTndZ`UF1jbX3y;!=gITweMm4XtiK=GT>&HFn
z4QuEo237hGn+a>Y;HFOCpGdQ-RzlDiCBL41ITr}g5j|gba4)DsV;6y6DASSK?v?zw
z)2Q4mDOe3USFO3|_I*7+nP$?GFhk=qL$8a9sMpy0!-1xQmcf>e=%Xp}i59XOi&`j9
zy=PW5Bh*$Uo=<Iuk{Jw0_HoYZ5+ewv5z2hYMpg+@{j`&4C*##rH-;Ncb|cS=!1f-H
zYFU(uQ$L4#GE`A5I8aIeL9`gI%jO#4Ht01kt|m$~aLTSpY%+moL0*xDVlsubD6Pg=
zptU%z655Ip<(!atAjl}*G(KuSu$D5^h5C?`M=*GpZ;xP65y26DVll63sC_DIuPxoG
z#VY6%{+uV(*stDqGn{56&Ko)J_T4Gm&Ga(ZS9{o?W^rA~1HuXT>R1_~OGje0Ayt^y
z!(c9BXyIk@1-3<%?&dF%$Npb`Pgq#~eQ$;Je_D>X2-Hvop(4lr$8tnLJpZ>6#%X&x
zZPSP>WIM!ovk}Q_1MnA+t1sZMk#?o*3}&rL>b8l$J8lZIvI=YJ$W&qBXO(L)S7j3n
zIBq6$vWBjDX<B7?v=nSvD|YEpvWmR%n(c2KA25P(?c6WCWjn3E^xmR8iw|@z_H$PH
z>LMFOlvA_arkqSUuW~^FY%iq4ZDrT@VC#ow*e!!jN{$s7-w{CufE|d7gq2eSYaTVT
z%>eJ`cxOVo{$v?u(fB%W`we|c@S}I^JpV{hx3$Y=W8h2~&D|BF4i^;)i#jFWL6lOY
zljmGkMyi|DS=;CBSLU=8RdulymFxL^(aFO^Mg+P7&7eWOx=2-7S7lkO16y=0#o!n_
z52d`MLxn%<X)4B^TyRa@?F@(XNS?foV>YQ2g3vh2hW<z#WuGGrsUj+KD2v_<&)`sy
zbss`fzdzIoiIP+_9CKZ!rBP{IrMK#-nxms3Mi^c@<;SV#F>tE_2Kt*BElCmY>QxE6
z!I3cOFR4i{{GZ7~yJ$6cY@j;skvzteit^Zo@uyRVJ>%5|BP!$7hWEbPGWwr&Pjh7_
z-tYJeB;Ic6*_j_Qe_KwC-jf~}OO}M&#aj3}*$&rg$RhiF_)`!ZA$=*iMSO>L;jklj
z(y?#oeh%y52YV`@V@oZ|nK3eUul|~QR6mG(A|1uVmKtiU7_R6bz=pGJPq&x;Rr1O4
z{O7|+2hI?*)6br7ZaDW*h)H*?r-)jqG4NWYhM*XenbfaqJ;rI4$&OMj2wP(?yv@X!
zd_49oW^Qm*OU)g(-0N08^pA@>U5ioeJe1n2L=hja*ijdF6(!QQ7)!|de18RW=D+Zf
zv#|Ux>^4fhHh<RLe=|TDa1t$}3BW*$;kx`|-3<Ub7{c327ORVdYdf;g6zy-IpRIX$
z{*<o!&fOyJMLSrJ^YahTz5fYx_7A%%hlNKMHzZka&GR+&wS!fI)%inCm<^Ww)glMO
z`Hw)A^G2I0qUle~wC78OCmRTjN1tyKpuEbaY*8=wvj_#WpnH3HwRz{QZwTi*6XQdj
z75C4e7emaj_XeGqoPHx6e*if857G~7c1+##u?XECcrw?`S8-R(YLVtlJ=il}T<={l
zqzPWe?woHZZK|6>H?+-;^!P{2T2YfY6_}47hTr!FZY6aIE_63IKQmP)Jt#IM;k0=l
zJ$v6A=YLC-tX;NBj<fvkyYU}T#|J>2c(DhE`~dvZ7?MgdzD5c3C+vOZL-9T#LW8EO
zAR=PrphGRO;x86R*17rK;Ze}5LC{Hod_i}gD9MQhqrM(uV_6tkZla-B+Sr^D*BYws
zUwF%M6RIBf<rAD)7v3H^dua=*d7waZBrxWLG!<RBfm?A;`l;d<$fn&WB@U&M-aO%^
z-YLVahTl1lzkWVLJ=vS%cvP(KW_MF6%i$p}InZ&0&!kVT=)UH%^w4dH2!frDaGZqK
z>|;B{@H(P%{Z%q+cx%cvxz+ons*E17z#HyBzvZ*>c5e{;H&(`*_9&jH9OKDU+4#uq
zUZcz(KI3CImwIli3z3$6yiuu+OI@Oi>vS?(a}d-6t2MS>r4>`gJZAblEM|Gp@EUB6
zO_z>=$;OrHll0phd{k7gYvFbvPXpWe+m4s^zCR>4=N1lZAvQJEmfGwIa+0dfVh4TS
z@uLX1EAUxo2nrVrrT-G?f8l`s|7c1_YxqeeEyeoZPU&53zmXon*__hll<<0(7hz))
z-qew)#mL2>+G?T2%^P~yMeS;X)cn-6%yMTZ+Phlp)vaU``RqGA*10fY3gtF%w()yP
z_q2V6ek0wvzq^z<*-Qaj*|C5{c-wz=TxUTPSeK+2t+n~y!Oj1`o(^F{kGFt(QhXc;
zbOpm6H=F<c2a(z<wJZ}5=>^YZAg6VPAr3XV!LvZbVcv(HvFprjS+$lftL@JH_wbH(
zSak%LpvYuN`BoCN0v!Ux&!3AoGumh_csV{Zt&Wvbtd3>13m!cS-W>c3&TTCz3p?<K
zmlF;4aS4)2$=CrncNvFu_#&vL{%n>-zBm1)nq3x|9Fo`s=Ynoe3~ieoBc380TL`Dd
zEXTmGk3~OZZ2NCO7Y)W%pJZm3A6e|8a;a!-t&JFn)=u(8;J)+IDua#iX-J8i%{hCV
zgQ&eFMs^}M<w3NT+`WQX_Q(XT<`T(aI4t`TMK6AT=%{y~N^eAUutfjTZC6s0K<hGH
zV)F5vD_`XEoSK>LImv7{q{i?u_L4StlCMpwm8+BKXtRtwq~Dz@8PXcan}lD$yK4^t
zFJdnp<&Jq~2(V{yQ^3NLTYYIrOWU-*HGlo;IOG*?1P4#PtF>sTvVjO6+^jv(LVi0J
z%I<pRX|w^q3&vr+CC~x?G88iQtaT4bGd>JKGhgcuYdVSUW3yE^&isR&Tnp09_*;(V
zCmF&VHZ3MDh@}%vo#$*DmTpWhOMBguA#H-c0o^BR=mBAIj{F&xk551LFM<9S4&MLm
zl>QH`x#&K2h$U!Q88wWLc)!MaL8-j6c*`jcn=cmk8KApZrMt8%1Y{HZp3<$We7Md5
zx=R;eN~ao3!1?L;cilx*Rai4Sriy)?699BC&b6>%0MO4#t8T1ot&iv6!EH0H&$L?U
zk3aaS9yC}}kbZ@pm_B&UI@@|%|3*4U&dD4OxR>YWP$UciyGz<7MCa8tXlJS-QiF9R
zch8_ZqI3v%dOUc%V*t`2uRuiqAYD=Is-9~WF1EvyK<2!~8o|0jE%XSb6IbSq)3ZIS
z7}0zGxy>DI2}4upVzR-hhR~=%D|`&E6#LGzo0o(BZE<P%<?vw=dct!5Ezx>E*haU_
zXS1WtjA&osD@%s)Uixv4D*)=aLw}%-2!Q%0%Pn}6n*b<dC@M+B%DEt@sO$6_Qhj11
z`c=0+q<>Lf<o>R_1dZ_%_B}(Pp~U5hBs)d{G?>PV@GxdpR+qTt+VbnEt`huMDrfzf
zn0pq5=Npc$3Zk!_VPILKX;Xr$vu+(AP57ogR53Fp<L;CahEnlQUI^pQl)=`6&m1Q2
z2`=Hz*XDSyWh<%(_{zs6F{5RzOIo0&kz^;;TvO>A7-z)WAa^^Nj{V7W2ruBe9OHP;
zWw)4bSTT%lbm5j3<Nam?f$o^+;_7a6`hwdN<F7Uavb&~fj0Xv%hHi9Nhjj$?*IZnf
z8!pVd>6fx32WyQNan7ufD31;NuvaXnI0Td@j%kP*0Nq8S)G!eJkM5Fdto}FMW%^%q
zmq%Q$wax3b2dBoTjNWGShen&VVbdZ@PNS=wZqZ#)#PSe;@`7+<nDv)P|10n5|5ADR
z{|C}7|1Z-2FVX=>Pfjvy|4XF*g?p8i;orA!djE+N09Zo)9w)#Olu8*sP?Ff+q^wwZ
zpB^wLYykw!e!E${2`8>n{vC1}5VG@nurfXobC?AP63Dm<1Oy43{s|ITnex6^U)$4m
z=J@I2(a_X%=kDx<G$p8dc1mhfZ{1~`aWom%Y<*|{{VeT{nWFJR+OtTH!2ZR;_eGIe
znjv|%I{Mx1_-sqI;F`=qC*!E!zPkXrnY>F2MQJAmZ>ZRkAgC=xp^a~5QQcL@wxBun
zE^JW!Di5%!E;_{q;fkWhyfpKK?rl}{Bq>0R0mK9}pXJ>wTGqPRN9;0r#qV(1UNQZS
z32?Q4DH0f7&U$|D7l?}8Fd?z<wELFD8uL<s7IA^yEvL)tNyX$%rK0ZKi1*@c9OC_A
z)7hAaQ%mzB%5?R)yAjQ_1JA?)$O5T7B@w~X5{1Jm`Mjt|gb<iS9u+DD$_xxy!hRVX
zCS(o`5(Yy|p4K?HpC*<_&W?~o+Z0^hNd@2+K&*G6AhDuGGY#q?T4L>q{p_3#I@iUa
z3G7H94xQ;hbH#X2(!=hYj~tR@0yjY!U@9-;CTyQ_Py?(&e_NVLpRTf>RFJ10f^9#+
z!OUcNFC>@|jHf}>hc!%*qCK;$>NtVa3e3haT|+y<j&T<luscnlpe6qn>yU!wp;g6(
z58NN(_o96e8K!7Q3)5nR%ZB^5>noJ`MW8C~l<1Y|__%dush$q_C8j)xk?zwAEM748
zKBvi}M;<|cw#2D*e7P%mrw*N<PwhqU6s^CFB3-a(aHnLA>iA)Lt_<#V*EB|bHrU&6
zy=W;`eKruYVV5Aq0ui^cpUzI=<Gdu-OEhDIcz<0k+sjBde2#qpqfDx-Y0ji4zhL@x
zYu~2RBBBqzybXfSXp7c_FZ>(XV>NvOO+o<&d>{@bQ;V6H9v4$(=Xg`sNt`eu{NRn*
z(30ZO8+i6!6y#q>2m{-{dH|H%?TiU&WeqG8ovdMKWe5SA#QhHUc5rk8u=LFT*k+NC
ziHZGhTPyw>(couXmX50|O`CUAkF}kxEMcs6C39({QT5^(Oto3VO%rQzWJ8N8$y#AD
z32%3vlfZobwCeGz>#eRF_#lK(nGc{q6pS{`Dn^l-DbT=;fB;H77B_Mx9(&@1OnpI<
zMJ%h*aWohl4#!CRl}P#)!+bvqi4#Dej6VbJi_j3e43eDPBv_NvP^11bL)sTM?#8rO
z6$98(dk8*g6(AQOTA(<lAp=RHF{puFrD9;{`!p%4fMG2Yw*FSS39)b9T1Fz41F)fy
zhM`dz;|%G*uNwNbWNLsnjX^X(+)@1mftqoj(EE%g(U=f|Q3GO8`E2vxN5MU?P_Q!~
zGSJJY`M{*8_{z2oAdCo^fc#4+6X-H9G$O$QYcnO2^jUY2M@Yh38R(?)$l+>;#5T~4
z3fRP94#Z_Z=+cA`!1D#h5^D$zag8vc%6mf46UCf<^bu=MZxIqj7m5?{pO$_Odjyk^
zq5J}_K-#$D2b7v1;xBGyrB22e)*(?c5A}&b`3wB9KhL-HnNUG{V@B#3rXYtN3@8cl
z8k#6#MVJA7Fp~c5+%F8cTX?A8)UaRG8W8szu+>wioPd|9c^0Gs6{dNZmRKD5NajGc
z6EGW;MtJ3du=y$rYR#m{L7953gbhEHCb0oo>u?^3<O+eEd~Jn{Wd0_!knfK@#KwZf
zGDFUe9-MH@r#l{d%a-092Qo}7Es}TUh>Eu*yigDfj|9AoR7)>LDS91=W~L~{I|3|E
zD6}hQpw-t&&4dMmn*24Y08cHwcZQETorVcGKgp~KmwSzpMr6|~Vc%5!-O0Uc{Q9D<
zdU3xnof=~}6Yr+F-nJXh`!0eGo90GR9*_`oO;6r{zSkgl5Ho@uy&JWz=a~e(S?%j_
zo-((OI<K1H74Olw;#VDms|Nhb%_Bc^coS_w?FDZP<{f?~cr#;^Jrm@pZmAFz=JN7q
zEa~d5b0`M~=B~1_1z)^~DoojvFoKiC?>5$Z!p$euqP$c#(bnh^RMEn+<~@pY5oikd
zKh*s9aWmwK>l<o@t4bQaaV&C^sWBjbqWrN;A3>(ZsiSdX=qPGVy}g_Vw+niT4XJ-i
zhRJ#fOpk6`xIN4T3Ms1KjUhzvy=fhdsn@?y31~YHAqfe?-YixCWj%S!*~A<3&FT`t
z!ny`^pph1>P|hR$2`j)ZWaZLoDdIDo*HNB_jQ5e}hzm@_;y{@y?7Tm<A8+V0Tc1P&
z$v6zQdH@>)$*UE7?Fv00T7oAXCl@w~QA`8a$GW^j1TjVRG*}*A80@UO%6E023}3oI
zp95!Zt-*ytJ#Kmbe9i?(LfF@Kp!rA~!uEVFca}5pnmq~^${76B{pP-ifWN}S@i461
z{ejSGNIeu*!yM)s`{xiH)K8u*9FtGs385jxq4^4)Y+J#<^3O3bpoRSd-TVo0Ju+NS
z+efJYE|;hswt(qhwL)MXP4)O##^7g$ejSYQB?1YF=55A_zSyhj^5avR1C-%)1aud`
zb4vYcTAMxCh!}e{a^rS+->f@eDD~-6^BmxIituq2;a&$@wVqtC=_?0QO>o%s2Bk1}
z+$!uh>m5;5-UTqPi05J<-~$(j#W;5cS*}Ch$n$`640%bBk7Eo0K@B_HlSTGWnmBFw
zi3Vq6HsU(SD!FxZ{EE@ftXKn!yKX%ZEtmK?yIPros?)M`FW3-~Z&X1angT4(Eq7f8
zqs<I!JzFNNQ>tQ{W>bq+cW4N7qro%y@T-sEUtO+~%xpuRVSZviNi;lxcq`d>3GW_9
znzpcuW(Ire+SRDm*OV}!Z;R<#!Zox+qc&Ctjo6$!p%WvscFmHckkflCOBU3$#`1{q
zuWInW)l7l2@qInL2nzg0ZnX-I4E<sc&dC<xm>NBK^O+s?z3A-mA?4<^1lMI`nHtL;
zYD0w?MG&fw{?`x~GD;x)C~=<{3h4cdvUI#NL=;|>{9`0GiBA@FQWZB(leJTjTOKQj
zKNI5dGUD!Z57y10KpcqK61|aR#r0Iv0Hw{b8sH7-krS2*SYlj<U0$|oR)eJbr*(*T
zkO!xQW7zLxjRPUe@~eDxY`Fzf-T`RfWk?kUE-?`D)n5g0$WY57f0o@X?GXM1>NU~<
zsonUMU;A8@65s}ehQ1D#V>8?BVXkJCzpWc~_919+vi0y?Y$-!qQy<gUkDe>lb<&0r
zjW=WR%paH-LiVP$83dINMctEQR4x6d8%bHM><|yGW<4lIbEEcvj|SN15C;bJmS)=U
z^>5|tRW7`<F9!mvp&qunVi<#w=vE5eM3!i`+9KbgfF!6}r+sH(7w{G~y2OIV%6VFO
zBD<bU5_5c8De7O1T1&gAmJ$|Di#4i5oIV`9;AJ<me0-npj{@uW&ny&pbYy<0mhaCL
z?0dlk9_;>bCV%@S8k3w;pA$1}8*{0)X(0E7X(04<+e3kda{2D3{JWdiO|yfGJ9u+=
zl*2&SRCKCh!K*rR_7piHM#>EXmB^F2vQGEl5@x}xcPeQX1g7flg1Wk~?*6Wqm$zco
z;_e_5L3OzUv*t&-ZpBc7gN{#ym*qE<vL{RRh4txsMDiyic3qm>jxPMexsLH4n>NEY
zuZBKpx>aNJKoiy1<+QA-`@}i>H(okbOHa;SBPmbPK8!j^upkoDh{#{ksY{+FUR|<2
zlB=(syJ)IS@h3cZbaeNXeKIy_ius<5S}AX2pDvzKsylIIO?tRcA_E08qr2W*K9jxO
z?%l2ICM3saD}z#kFkVp_)D;nXp(K4D84(ihn3hJvqeo9NQqCvQEadP=<(=&Qk+Uaz
zWBL^FYW!~aPRTnlb$#<H{m$~id2Gj>Ht~JMOXC1T^Y-df3uqDP*dyA!f4bV5j;`NV
z)0EQ0{rojy`oiSg$YTrTWcRz5y7J_yr<B*6eT$XZ=rvK!Dcu}i9^duZNpp91PJ-%Z
z<$=sgN(Pw*C8a_bN|s{^h&9u46MXRXOh1=*h-dd~-<%IQ%OmIf64xis=>=Kp7|>2>
zlk(9s=kNVdBnKE<W;!Dny9@8p;<>z&Kc=WNQfUqeIhf$FkKhYmxzb{)XJkxr#>~EQ
z<G&&@Wc-Z4V5~{uhJ*UHeto%V^Mko7Cvd2avIN6%Q*BBP#rlJm7Z~ki<uEI%dg}F-
z=iuS;EF(o$hu7_OXVg;#E(i?z{Ol6#Ggj6%j%R`(gew?M5pYHVc`bXdLWbnLH+_4D
zRcR(i)zHF4z4?m4@_GN|iK(oQ+dCAd&y`iEzFI4u{RdC<hFzh$1dBi}TGUsvRU$c&
zVt4{04mJIR&pM}Q+ugqRuXl!9M^2$+YM7A8c4Z!Hl8xw%tHKe|C1(%Vb_wVgFUHLe
zrk|&#{2%JO5wUB2M34Aj^?SESD5O&&%zegw2YR`&SDqj@ZtNJXHiZoZUTVIvaAEuq
zsCcegaO23}Q`4BLGDWJ;c)dvC-QZ^TeLmjnEx+PTp&eRSi2K~4iv&^GlRDlDvAX5b
zb#M411UD_i#lW=@o^+)kj?>13I{F?){IjfXm;23bV12^PO&^*d$2aYhPu?#?kwYn8
zLq78_OdHSKPF$5-w`kO(^Lh$rUTk#w+?+22vX)%vL=iDdoBkkZs)$Pq_$5f7x1Bu_
zHm?zJfQ=P6CL*hWSQ*zUhRJkV^_4eqclZsKF2@){QA#aUS1>&5;p^yot$L!q!MZ#9
zea2K8Q;>#FsBE?`-x#W=@G3_2Bth3>S-C9gyPwR=)vuejzR+YDcf}G+RaOq=z^-9B
z^aY7Qb)2I7>yC5@q-W$kC88JV?1%8W!)pAjoDQ>VueJ_+#LIGKBwKsuv8#E(!VIW|
z$d)LzK>bqRJdW`*eb|MFZ6;@X&YTp@l+?xr-Gl|m-RfpyJd9=gyeH8L8de?<R-Xg(
z%@zA?_N*i-wc0Az`;tzE<)ht%de@7#CTAxW0^08%#y?+2Rhq88^Y)G#46Lb18xZwp
zECzEoxE5<RHY(G5lI<MYZsufWGNU<~#`Yu95VCTP&Jr%E>6m@S4vT;<>n{x9Y!Ess
zEy!x>K$Zi3I|>ilq`Ky=`$U=dQ+SR0B<OR*r~^y}Z*0%$uIOd?V4;>NuOmaaJGs>D
ztyIgr{>w<fe)#9n8w@p=29H&~{80MB#SCb5ycL2pB&>vHsk%1IPX<=GnEQsBk%sig
zJusJq-C_c?M*SkU6PXeP<K!;x&K_v_@3RYF_)Wr<q}aW*0SScBy1cd#2}I|^W-_(h
zyWzGwn;}_BSa%1_j|47Cz5*pkR~`90jAE07eNQ6ekauCQr{d4n%{&fIyf&0a{_^S7
za|y*!M@Eyl28>7SN?b`2)6+zYHk8gD_1=%u@U2L@Hnfy&gEFE1F+(TFIaa(XwnCO4
zfiRKuPoq|iRdhTiAo&M~yLSC-Mt!dJ=FW}x-D|nMaYi4OkjEUlUFv?=C78{7Jsxg2
z6a6LQc;U(+1`D}rfasaYa}oO0oE=UX9ZsVRmKCp1i&y4x8Ks*KpoKm%quq(%<9x}+
zlRS-6S?8D3TEXHIsAjQ|ffPZU1SD>z(KmC89Iz(yoeJp$8RlqKoOB~Z>0=;Ti|jpT
zv17J&t5ESm(j)x%TVT>ggk=sU=QVb+F#2o_%#*y)Q`xz#Jk^kv=qD8Zx4rF{-L}1z
zx6xIHojG#3no4=_sbY(x+GyV$+T}LX_BV>IU8w>==V1h8lA{_CHG9suN(3<u&1f<X
z(sJv7U}GIhvnHIu5Ynt|Zos8Ab8j=@N}h>a0YjUXoVka%)I*+`u>M38<6JM;lkxin
zg~dy}EHp2of}t(oj+n>Y`fK5_POJaj**&6KOT@0bj~f4NHV(*(H)CfJ*p{K&dH_^U
z*HR$f{v#jODk<2T4GQ$Mi5uy;6v6`Kyizonm-xPs2Ui?P5?v|d8<{Tpwd&(NtVK=}
zqR<YBBRU*O|7MUvrEzIibyO=P^2s1cX0`Ej*zFISXXnQ1^=&aDq!7696o$2z`P9*#
zyj$G!NFmnPUqC{Tq~vQEex!Dkm0gg?-7Ez$30`OZ<$6Grp^|45RS?KzAavbc#nq{$
z_tByRjZe?`nWcpI#vN!BIOmlZIS#<B?U0{;UZDJ1<<$-G{D4X9k}dd2ut*R(FuWrL
zX@5jWQ`y%E^PpqJNkR`#$^ituEre4|f;UiXP1twkNq9w1!N|kC^+Kona*@ZdyP3XQ
zor@r95HK!nCxJ3x`ihQn%ojBE4oM7MI|f$5M5L0Y=S;4aRr|0_n)>}JBVin(q@kC>
zx`yM&>9&{^o$^QR0fjfaL?hhDE6taB&+s2Cv0x~TMD8oG9ac3gI0mVWulQmCVLy@j
zD|kIdGG$-nUJj$z15q$Q@?g`ZpHySye#k)<UQgoEhQtMp{}SMH>iwD7HJ2_F=#(O0
zY`=}<;Qq-1jiFGPDD23SFpel%3<Fs-HUtKH(I197!c%XicfA4DhCIL)@`ZtDRkOQ!
zJs4cI4pgn8yU5pN+Zo@&xg?3<6^<yB`z!i!i%z><<pX*lfo=Jdm4A<!d<HwpVr9$s
ziC?;g)7A|OZ0;)ClCbA}@%MQjF%0un?DMsT%02;ym5B17U?(g{6*dA}$;^amE3o`z
zNCQAYDf@?GAD%HhP|z4ngTkbovq(*k{94S?@;bo$u}hz93mBl=hZ^LBmw}Vyy$7(f
z_+nl8w&2gDX&>Zw!D6i;OSMt-c)txo<Tkki%}e`rvgAvH$gKm*F#%aD!OD>XJ>daG
zkn^p>###f34C8}fIvRj|c?KFHlm@!P12RU@n}MafLMA&0T6V%V_5dcADF_Gl`y^1k
zJ2rq(xN`22Z~i%%xsy%-w_}>AGZqbRJ`82F2}M|IX4lhNfn0}V#RV$=h`Dab(rAad
z4*@6=$wM4DECZ@hJNv3r0&F?lB5sFpiqAf=<d>ezt9Zj3MSgf+>BNohWq^?XUl@TH
zS^m|>``;LW*jZWrX#-+lWcz!#!~dKS;G6ntH106G_h$7kQE3o$quQ!qeL@T{cc{o7
zgllBXUNwAojAYku3tynF?}$8~1?tK@vz7Xd@#0HHzB7`*fq}wd3Sd!=b=&vg8u<%`
z4!~A`hB515mxj>>6E*e0nh6e(M{w<>8}SQ@KlZ?4Qqm*T8Col&M*;&I4f^Mr=qNu4
z($vMT3)SstCq9y(2S;Z@Rxh)wTv46IYj|y;38Iao@WolDvWrBKe6h70A)sIIpF`$_
zfX@u50IB{i6k!d;-=sed`kC;_eub#Ghn^UYC~r%CTN#a4h5;3c%LSnb3cnu10-4K%
zl4(2092+X%h!&1m>|S8I((E8szy>!^oRqjMg`a}_$`G3Zyw?z)B2aEvXjY7YE<wp5
z8EqKT90H;!c_2o6XM~ZC?32GNXzC~%GVHh!T$+(EAe9DZ>L_L)A!7o)xOGqB0BIyB
zBUb%9#JJH{;n&nrwIUX=Sy=y85EBOAPy)s@z7_zdpNuzd1S21tHY)U`RSs7HwaU*<
zj6z~E4}l^E%T545g6In4Lk(*jzl*zE_Li_5L^2s+x)NSmto0h1jw`F!kNA^f^bkqv
zGM^8PC*2TV<s0h+5tSh0YwQP4uJJ~-Gh*=k7%JQ!3z<&s({g9PW=zGIn@jBhF>J1D
z)sw0-z!C39@<cTeQ8C||8$;|E6D(tTzEa}jS6|43II<YCE{(@{z($m{*o6rJM6rgx
zR7Ar|pRW03xeuz3<X(rL?8G2$dfi-}6UZ*bV5E`njlREcuM-%huUfx7+??z`dEgvm
zY~NkytUksNbhp1gc;E;<J)Si~#7>!jkGldDfCtg&Dfl@-BEfGZ)<D9+AVmY$YV<6F
z4m5!ntN>?zs)>k26S#%ytz&w6-#@y$Q7(FHzA<&Cym-I8{^r^}ks((okOx1Q5?ZBm
zE|1rm7m+Eo7f$Z5CKSxfpoHTN%`9xQo)DOi6fbBtn2GDuj*>3`<%-m!gxCt(GKtvK
z1j<HotI(%{Ccsl%^wT831kq^HjU$#~_leI8g}rRVrY>3T#$S*1As1;2GcR2}KBWs(
zZArdQ!v)AR$VP&hD#%UBbQHY?lOrY4)@c&!Vpd&TScHImb)+)mg@dR<*#)WC$nqCa
zVm{KW%pIdZ-M$R_#J&vhfkQvJl2TsZwlHx_%lYb)h$gY=YV+A8u|9vZ$fV6dvK-1x
z>nf@Ggw{P{$_%j+4s-d+XaiuQp)5%Z{t21jUCMm!7m(y`iovJ7Vz>6sDf6*pyhbxD
z780Fn3D|Y07lf&Yl8G*eauYIrhCRz)^>_uDC2G%sCZX2l!z~gGICuys?|aS!TaCD~
zri6NJd_|enXLuEqzG6*JXiGyi;ssHxEPE%3NkgrdZH%>iTQBmsp`5sKX-`nP+bc(m
zez3Wdn9E;k!0)xtIJMAFN=NT3zFT-uR3Jw0483y-IJ0%Ap3S@xDw*@lP!%uJzO`@#
zJc^-nD@ykWoedkLQ<Qb(_R@J8kH&W3hUQ_J;N$zt^u@Wb^%DVe#oM3rS%^%Y?Nf)m
zIa8TePCVOcXa%zkznIUEEzTotTq&g^bmnZ3B8vyBiiL~Jdo48}p;UJ68NtZxdnMQ4
zbMZd={RUM7sGFeQxhk~W7~0RP6WDWh*YBRAZX9+cZ3!_Iq%>|}uqvK>XUtSz!eDiF
zY>>uZ`6{*f6#U%a)Y8FEO<ST-+=$wN$I)vZFau{Bx17%)rbo1XNm$Drqjh+!0|kJO
zYx6NP;m98YIoA5E=gDKEb%?A3iNG%7VUwx%N~$dFPA665l2Wb3Rp=>0v}_{uNG!I}
z9_E?z)Wuh9kx{9MgK6_aNz?_TRN#PdGHH*({NeGZB^Ep37N!S0Ag2+eReJ*zYHE~g
z5CHrsLKJfG*0&>wzwxtWeAbY5BEMw90GxXX*<rB@x7jxeHgz4Ay!lroJnwDm^Y&*i
zk_rD&iIyjy6-yrlLs6eo6K~cT{I*W)`Ntp!$X+ov!c&m%<}mh<b^^=d;!(1IJ-$w3
z$beMzsj*}ZMchjLh)P=pcvbamsZX0IzVr7vLG~gbyFfxO5bC#pnn9X>7GxHijt1_X
zL~}WWnb+dq0LX0H6nv>YFb9v`%Olv!&2;>w)5dZ%I69z=4Qeej*?*FjyOO$KPQzaB
zxm?1steLaAeRtc2o3U~MAH7n|El4?|_xPZ3T&VmGi3+D;l5@^Ny%xceNyBM!R=<*Y
zF$=q7y|vl1^85UI@06%@b^uv@ZNJIA>pN^wrAC80#c>X>Cq``RaQa)P^U?D=#$Xf7
zV$+M|NfdUuY&YkJ%M;h@$wZFsM)%vchfimAF#Y-EK{bJnPSq6c(hpTP>|e!a4a_x%
z(#vqR8@$=P+O*Z(C`-SrPu&tQ#zabp)`Z4mw6^DLWmQvxT;66rbG}(es=!^VLY3%0
ze^bO-vG)*3q7S8KCkd<NFf`h&ba!f^ggtA^u|~#pSXPhZ*>u1z{^*t)8q55)6&V$^
zN4eW)gWQy1&0#*ZZUqYE;572(@PN&5ES`~Gf2C+j-&R*qB_T1oy+HzG%$Zewo01LU
z`?403iN*6#jJ51qPjw2S?T1@ycFtzEmwTrV9-o<QD@Tg24t^_Hj?Yv!!3xFJ(k5@V
zu~c$!I=AoJ_4|+Nlo(#_P8vPy>NW?C<(5k4``69gr>7KOUk+UYUM}truTLL@fojEl
zJXyY2hs@)j9T@47f(F%=l_+pfF3RaK6_FHIQ0+fwIX@J~AJMKcGRHxq4Jt*2c@~?T
zBn{3-hLu$edMMLdx9A#H`*=b-G|yu*m#zi$)=)MqTEu^<{#lWdiReckmpUusb5{83
za-Bd`<_|w%#mvMtvOjy;8e@Glrq_g>=9pqxwBBZzkvpLC=7PMJA0~yFc8Cs-=#bT5
z);VN4=mD)c+F%iT&3JpGi9<KAej8L_MaK>!Y`9Hmq$qAc|77Bh6D9lU2ZA%Qi~TA9
z-GQfUthK-D4`2e*V7HUM*NU~X7{<9ilKV7=EzBJ+Oi9WR<yUDV5J8{>{D>%YL{$H3
zf^mZ)jT1zf{LU0)tk%C#5lp_y%uP%)j>k`kEM&@0j5ppIH=Pg_qonmLjh;+YWG?c7
z7^xJj^%oi(6T`oT{L4Am8aW#RQZ#+BGcr*!R->k;VWOj9V4{Gb<#%>6vvv5Q#mLG6
z_^@ix(a|aXCQiSJ{%>OQn`qH70X}qeLcfXeZ{qS#&gwU3`<vMPCf5Hv@aN1w4gKSE
zrGFm#)3h}Ot&+KumGS?hd%=G;V)UE)??$8iyMp6C+jRKNeg0?JKROltUE=(knExh!
zS~aGiRdhD6FgA4hLP0AC$W!I?pKgNG@2Y<W`5$9e`(5^rR{vc8p9cTb`bY7fTks$E
z?N4Wa?u{V@t)#KLtF42P;}^hv|I??rt&I>MWftKVAr1yQ1{OMIdUkpyfRB|;15iWS
z*69EJppt{Wot?1}Am^68m7_5Xt(=mu8jXmvm6d_Mjm_@|tz>5INC^1<+XMefsABBk
z2<VQGkp|EMGdlwdGb1$v4a+}}064)|jZn(m#h4H<5(Zi!V*_)28wx^8JAEfJM`L|L
zm%pFSLc>S{zzrb1m$8kJt$_s$JpeWT6661$7Yoev|8N*H(*am1V;fT^Ge95Ae}5(U
ze_ksjssFGcYyPwj$R0KhuNU@}2Nn|wilLpY=@TcFAv*SpfrGz0|18+$W;?=4$4<HC
z=vRn&{W#}I{rZt}ODi}DDnmnsAx$PV^|Qjb28WFTeWht=cm;YwHuXV>UU3lT5&mm}
zG+Z>gh^{!5M0Sc{anVrr=p^wVw#D#<Atx(TtQdW%@QR5}mAGaI>XI+Wn6;_{<oRfc
z%n<Bsl$wtY3XAsVXwdvlSfA%$>w}WzWjG_yDMS<L%QG3V9P^VH50WHl0M(bo9oX=a
z<VNkq-E~d$_YHfX3X)7AOa{I%M531wqU0MZkDO-NQ`+HMc7*>l{#t65nS@I$*#}os
znBVCBwfq;Zctk_^1~3VUrJpboJmnZ~Lq#}+X@XE3s6(u(csnkJQoL|HE^3kR(ZNJU
zMmqsM4<O*A+V|=Ca@)&tU7TGVzw*6^WdL*=9Xep7k5vpTUV{|6Z=K5v83THxJ)`}C
zHiu?Pd(m>fJ>%siEmcU)%8?`LNkeB6iujxv3R+<Oi)SB9Qz$5MN=H=#saZ&J%dRnS
z#uG^FBJAgLHv$czwJU!Ou=607L@6nAYmE5CTxM!S$2Q|zy3dV$+~eZfaN}_k%Y<|x
zn(jIdkk-i^v3J;wf>LYH#gd==otOr)__X0b&CN%c1}gpe9U#k>I{J>tEGDQh;Vu}E
zV(*_|3}k;Ta<CDeeFtCOZO8m_Z!Qj}UM<f#peOrK3bs7QO$mJOltI#$%|hCjy?h=8
zHE8@~s&jRV+<1VdFKclYR#W1sLVS8;9?C7>tVtBkUF%6`0Jf7n@V9ZChRtK6V~J%b
z3D_wx`7NVZ{~*R9JTpv#f~s~env<MQS4H;Bx5S`eC#>X#zJH{FxDSfPJcY)i{}O#d
z9agbaif2$$QjP}a`LoD)c@7ORn7iyQQD0V-3p1R(!u{&^r9=#77p{IYLM$X9zb~wE
zt65HH+v$ZCYyUaUw@5X*ls@GtIwLH?TCcrlTf)hQG)Le4sv0}(faIrR#>3;~=JoiT
zqv?2Jdso?s=31%E>B?RS`L%_I$B@`no9qdveUa7o2Xb0Uzcx^2eM9B#+*9-6kCZ82
z@DU=HT%#dCA{v!C7kiZE6q!D5{Y<RGSbTn5VG}U>*IB3?uqMz_{;{Ai+u=gQP1DLx
zR(&I!ng;o4dP1V<TUK3JdXUTTbV;rKG`{Pwxq`q-U!o+B9pa71F=xzCw<H@fws6lr
zV>--r0Ud)A`sng~X`PEDO<d#Z9hulANuq;}%hCfm{k#mafr0dV$b$zXW?V8z0e$YB
zNREktkL_%at~UWD=vM?C3}opFZY1OVS_L#q_OOMl$NqJs6?!VO56o1?K43dq8U@bv
zfU)KC!Y~1aUzv=oxSgkiNdSsox*h}<mtdkKa4TJ}UGE?$@HW%H1<c5B5P^c4(WO0@
zy!;#>`RN<f!HYNIK^>A=Ov9O!tLV4Ss_}ZmM>m6QkdF3PQ<(}Hx2WOkm2aW>(Z}m&
z$88RQzqA+ei!r(5p!w!PMREfxi?py9^J5LogouVuZRW;{BR0tJ>mY}spb3W`KYhbR
z5kYm3!kqyFB$x+XU(d>;q74O9sMqc17Lt1uX3D>`tk|KK{LGpk%aeSsRCjhhw#>bC
z^b1(_n*lb8Q8-tAu35-LvSk}qge%R_(9chU9gAn5WY-zHstz|dXl)YqK{Qmo@dTj6
zmtqgPQY;Bpg)x$7nokAHjdYy$pA9Fp{9Ib6c2)aUsbED0F%M3==pyJRz4PX~3eVbN
zrVkgr(Y<}vLs$t4FdRgOBH}^>1Su|$G{Vxml!~P$t6m7nTbv07g1%vH!0oK4;ZFWC
zOo$G$iNqj@2|7UP+v|||8r0h5bLZ_|&-l%m;GpnHyawR+spqv>1uPegRJv|@NRn(;
zznx;J-x#zFCLE1vaGWJ;qidH#-$-zvBx|SI*)wOiOzuX5A9(MeM~5h&t?i7$>{~s+
zp5;{}uzIX~laHwUG$y#M`zf^*Pz@|t;k<KKSYcAtoS?rBQL9tOWxMWhq7GJ-hag4U
zE$Lp=)1HJg##5STV>nQoe6ZxI20g}h&z>5I;D;ee4fQ=6__clS%iA(jNJMMx23E&B
zXR!L-G!;Tsw!-AVo6qT&g!`A9F1wevN1~9|@4UQ6#yAzx^GWepSUU9iLHp9i4`Ey8
z+OQKZpHV88@(x2D2&Ps&ZNFdKfPcO4(W=$~AAe!_rD~+(&Qee{`)>cFErQwZ9heea
z39pLN#UfxW<vQA|>+^Dcco;H#`0PGora>z146=c}iVz!I79~OYb01QmpnB8GJdRjp
zv9>Nc$+=?1a3Xt=w=`D!ZCa7BaEwX0xJ)D2;#w)8<Z?`w<;?ZOGFVMib*PNt;21*{
zUFt<QcPZzG1$I?K>+FF<M!Bh0ncXu%w^gh0_b6VUk`3%)mQofg>lRfV3s>51%{Gso
z!<?pOk%}lKoTS+gZxlS4Z|U<&!o`#t9eb0^ZoaZ@XA>+R(Ot~n@fGh<znl8dTR-g0
z9J76o5Kso0xkAQ`EpCh4i-w{$ar5+vr>jr83xB^JUaD_<dFj9DAEqU+IroVV(G0M9
zs&jOF*lj+HNtwc;+;p~RWLf=I3HNlEvN`4Y;lr_X3Fq^1`Tlx!Gy(Xf>hkt@R+ij7
zY#)B{%)zm}xY&gL)=to!&HF)ZsakyfjFMbFfBBa>?!UfV_&3EvO~Kd%hF09h$k>fg
zla7#{iIGklhE~zs!}t#mL#s-tNl(Z~NDnAgu(bt*u`vM>nbHv|{65LV_V?G{iu3QU
zLWG)JtV}Gw|L7QmS=s67=>!EtL<AW{=mhBb+1Tk=nVEP9|K}=zp8l@&Ff!5o+fj04
zY35dVs%XQ9<I6siJIi6aY#Y1FGTF^+F9%;`PT9(Lxw;4JcS!F^DI|fNQ6x1<3w^#A
ze*%{{f|B4-5-9SbY!HhCL6g)d<BP~8bQbW3RfaYwqpVexroDMN{2_a|;mCcl;rV_z
z-nmN%1wsJ9kEv$&YXRuT*BZP!0XcJixpEUQ%D&vX8hop!T-!=qa5;E%zi)b<)Zi?7
zkmX>HdlYKm<^%V0fh&Nb^vKLWCUU_W0$6IWD?pUdtvCzJ8v+rEn-Jjq8{KpIx1W1d
zYT!>HhkKyuvTXU>KdAw0@KOB&E9WOrgKP@0)+0y-I&>_wGPIB?fOQJQuYsUTse$qU
zMuB<h1;PY7=|NIMWbupALs^Dv0$Ti37j4J64D1r1Ujusz;n|}@2Qmaig~&;bsgtk0
zI?Q5iw3DO<Tng^2M_~pu;pZ*h4%8lp;IH3<UpU;0Wk)ow`V;24mk`jFAD$i}CM0Ap
z*cX4V&;6u85R_?qz;p-*0fJj}Y#`Af23RtFyL!k-KWo5>`LHs<N_v>s;7xke>0l@Q
zZwMhUdT}uu^$JDWUHr8b&8Fx^J!){B{FwEqzXSmH;FGQb|B!VF@ZLgV16K2405|Cs
zwxbBCh5!sp4dRiAwoFT)XMBu^TXd1bVL=(_eo7g5c0R5+n#sAaj6@pG{8&s;G@auw
zizoySsWCmJ7}Z8ASb%p<9yIL8NRxfwGQle*azJnNk=C&Spp^WC|NdB<{Ai{jpL@}3
z(5yf`dwl4iq3Q~3*p7ow!0>Wc0<M*%Q#5u;$)batNh?Uc%Qzb*f!BLuHwW9<0u|<v
zSZdD%w6eI1n~EN{`FC6tKZ}`)B6G$s#|=c2x{%hASgyx{-_$~!<n2|H_~JdCjOdEK
zb6hXQwb_J#U%dIh2ROOB^n4*CS_6hXcT?427>uKcezi0D#%a~t8S}$4(YX{k9__pZ
z98qb6KFkm$mzW)SIF1HT#Tb?+&3ug_SZ)M&Wp@Ns^u7glD-s-W^O-1qG)~F}KUON*
z$RVXP74=8mJiBRc%7x+_9~7~izRFY3Z>)Zi+aDyXewB3|ZAX%87-L5FdU3y_){m#0
z{lS+7eni3k#e9EXkF2zL>t$+erboA;*8<EscQn8f{#xN<4A1Xb)uwr?Bxa-*<`(Gf
zOhXRrEq66Q_1;{IRt?c1NsB;klW;o1f)m}ir32auqP<6^rZ<3h{)n~-2z~?DC6HT>
zmJKMFy5*`4BF67c4>i(15P$=?XupD7I0?PAfM$+cKSRGdooukrJ>qugQtFo%n%nH)
zw+P(uCP1_S*Wh%Py_~jYYwSpYTnDVWa+_Q~OzF!&d}jh&U-nvz?#C2570UUHR6Nxu
zVED%80@UX`?v!s(UATmCDClJ|(wyJaH^_1!E<I#VCZ2t1AUVWl_{UaxteG@FnsBAL
zVmf8R+<%%ho$Wd_fXaZIv#J@hbSNVrKvT@fdX9_Os1k&WF_-7B^s+95ASi(35C=Q;
zSY6w&vuPH3ppf{ZLKi}@t?RkmOJWP$&Q(fTFQ0<Olp!V+u1?TQm6m0#ncx4P0JsQ8
z_r;&6i8fLKpYAq$^f~n*l@x8HoL^D|k|K~4S)>QduH2(>;{d&+s3ax&>0z@c|0(Y=
zrlEBppA><lQvflfC9`kc)HrmNBAOI!q$nqyHv3+;yfZvSOF$bb+DPZXtl7134dFmg
zL>uY6+0U-4?;_ycSmf=JZ*F-0Rapg)OXWBBsaXLkK^0g8s?A>Smv;eJ2x<WD)#75a
zzckCc1k?gvy+vNV#T6iz&s(=1z$?F~y0IIp3FI33&(reqGAy=%X0x}~sILX^?k;wM
z_PpZN$*~G_gVh#RRjRKGaF{4^jwr4N>p+i%i+`jceSiZ-k@H5e7i<6<!B#M2Vf{4?
zR&Ce8i{nRen}sc@9GpvvyTD#BY~kwH<j}JAgAogRpHv^GmLl)-;t|04r8sKg;2-5Z
z0FHyh7H)lBeaFC2!0DzqY2mI>IZiMk{p2Z-*Z<K5IZlEZFm2)Z|I~LHaD*swgedBa
zdB(z{OXcN6QH;Tyg;Ph>H_r(w#JXtV%-!m{#838F_zV46f-I8A!rQ*0rmi<5SZE_D
zSoon^<2ZsPi@`z*@48)ms3}1u2`WiYNrFm}22f++J&$V~x=Ij3vckgqx2mrZpp67Y
zBuy4Ra8izDfEbci3m>XgUpuqXVR1D;770>G2zSzH;b;G=aXkQmB>4g#eL#I2z!Maa
z<T`l#A@w1ogi~6=p*-PGo*<QE7;Log%B{cM2GC2g%fhd}UwtSdK@rJb3%~iS`bL<a
z`iOp#i{a_N$$?6e!(iOPXC6@>$NdCFBu6ZKR%x1`jResoWiXLDekaEyKpP3sPfl8R
z^@beNKxc1skjw!jkjz;4!>?(`S!OlLK~v517QXn5EL4)rTQohVqiF>43X<Jwq8Cr}
zeT9X;->*I@h({`(`r^@n&vkI)MS0PIM+csJfX@T}ckPQu7M_yfms|L+M)j#$=rl~=
zJtf08TKLX&d3Ao~;^fhWC(^#zqHsjsc7QfK#ls^8Pi64k7S$-lJreMg3{O4qMBn#X
zw6a%Ts)gq!;*o$Ku&7Cy;t`EU41R+}E#2zd$V>^SM+bh3MXe?WQt{kkJY~dFH#|Ax
zx93A{(-2hRQIkg&o_n6(XHic{UIj%@MjQU1MFTZ*P)j^w@Z1MI(f3CzTK^Au5r{`Q
zo-*Q3ShW2OIZDh_zky18*`mE4m4h<kk&j0&o;Qh~vFN}@<V8LnG58A>9owlsgyb)A
zeF%U8f6=0eN7XcoUiz)9YEWg-g()=)%%B&m1}w1XeI0Tv1dH(=#99iLSaipPytSYX
zEC<Uh`tY0_^<V{P1q~K`Y*3CyunM$U^ocR`wS#WZY0-oI>RSy4K#xVADXVV{SPP1v
zmsuUc+6V?M`ut6eTL*@~dawa(2b(NPzM&zT!7$ii(Ut4!+X?o9T^2p@UG?n-2f!#8
zvFNFi9Q(jHI1I)tdgf1ZXqyf(H93xfTmo0WBkxgg0!&!+<017O2UDO7W<klK7ax=N
zB$x)L!3;P9&VsyTRF+O*od;a&OZtmBz|pj%hsNL{P5lr8$pVXhJsZ$hqG=$c$^0^|
zv>C6H5|`1e5>(M-x>*5mW-U!%EdUE?vfhkKaE53~tT#ha8TXme5iCTL)zM^!G9;BP
zqe(0?l#?x|$xdbsU<FM!J3~I%N}2>E<Bpv*(;UO%qMfzUBwAS;;N4T=Qk`)F&sNbS
zcv&Y{O>+c`t9jN<6A5HJfNOJ!TX}}+vLa0sk&!7`AI)7@T;{U@nn)@e1bo9QA*l={
zW_)ri?Z85g8MUc|CbLaoGfln$WPB3HI3bt#E|95jJ54$CfSoiEaHe0|MN?1F)b`LM
zAu_FlS6fLxsYUOnNmgXqp)s18pSE+HCfSo|?@@6{i&3DC&?F@@MV0e&Nh_BGj??7R
zMkc**GAe1CB(5o%WNn5Vvy(Knr)W2urb%{ZXguRgT~YuLdd7ROM0v<ajZBhPw2*uz
z$x|&dB%ew03N4b)BzekBmP@|k&Rp`ZeM*w=plLvoZ^w#&B;STr2_*SetOAhaTd)=Y
zNxm6t5s>7YuxfxLzY?n!Nb-$XOMxW60;?WK@(oymB)=R>e<8_hoE|U9FT>JYBzZkl
z6PM)cu(V1^UNh(flKc{^E+ENkx;;RWUyQXDNb*`+FOcLHV)X+_UK=q8B>8HrAt1?X
z*ERr2z6xs-kmR++TYw~Afwc`t@`}d}AjyYV!$6W(@OA@9KET=wbedDR_W>kdxQI0h
zkbEJ=IslM-VIJ!sP~o1#It-9};R4oCpey5ftO<bR3+J#Vfi45HSS5hu3um#)0Ld55
zV4VU;zA%Gz8t57+OG9+@o5s@fkbFUZrXSbG+ml$DsXjpHH?%AyUzozuRUxqIEu1GS
z3p!;OK=K7mHv+meR07mq&{?1h=%WOBDJ%e}y?|Z{i$EvP(VzzCgH<g+=!K;qKM=hB
zuztMWDn6%|O<}oJ*4!(r!76{5$Xa2QZ(f$wXjKbdmbKEV)?Ag<WL0C&%WAf&sh4E6
zSk=R~p3-VnH$E+^%?b<m%WAj62OpHxVTEfuWv#O6!qfR~EZUu8$qPA_m2)(-=4d>Z
zqxn5K+P<2j^R66Sf6cMxS2=p0%Q5g|j-fB)*m%F5CvE6kq+~Gxq=-qYKKO)+jOyBH
zppte5sG7|J6|(a{HEa&3e8oVOiwMT~PnP-GQyCMynBc|4EhcU;af^vtOx$8Z789+Q
zXvG97CPpzaiiuH7jACLG6Qh_q6xRbvP|W#0ruxK<fWi~=F)60z#4Uhw6Sn~>P0YWn
zVv0<RB4YFsBcB+l#65tj60Zdml$euSOf8A~0c9i}1XPfi?};(RBj$^BywR#lm-9V1
zcPRKjCcOY!3T19&b98cLVQmU!Ze(v_Y6^37VRCeMa%E-;H8eLlF)|7zB}Gq03O#wb
AYXATM

literal 0
HcmV?d00001

diff --git a/docs/2_39/cmdstan-guide/bib.html b/docs/2_39/cmdstan-guide/bib.html
new file mode 100644
index 000000000..5bed838fb
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/bib.html
@@ -0,0 +1,821 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>bib</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Docs">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#references" id="toc-references" class="nav-link active" data-scroll-target="#references"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/bib.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
+
+
+
+
+<section id="references" class="level1 unnumbered">
+<h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></h1>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/bib.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/command_line_options.html b/docs/2_39/cmdstan-guide/command_line_options.html
new file mode 100644
index 000000000..e821bbec5
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/command_line_options.html
@@ -0,0 +1,1151 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Command-Line Interface Overview</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/mcmc_config.html" rel="next">
+<link href="../cmdstan-guide/parallelization.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Command-Line Interface Overview">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Command-Line Interface Overview</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#command-line-interface-overview" id="toc-command-line-interface-overview" class="nav-link active" data-scroll-target="#command-line-interface-overview">Command-Line Interface Overview</a>
+  <ul class="collapse">
+  <li><a href="#input-data-argument" id="toc-input-data-argument" class="nav-link" data-scroll-target="#input-data-argument">Input data argument</a></li>
+  <li><a href="#output-control.section" id="toc-output-control.section" class="nav-link" data-scroll-target="#output-control.section">Output control arguments</a></li>
+  <li><a href="#initialize-model-parameters-argument" id="toc-initialize-model-parameters-argument" class="nav-link" data-scroll-target="#initialize-model-parameters-argument">Initialize model parameters argument</a></li>
+  <li><a href="#random-number-generator-arguments" id="toc-random-number-generator-arguments" class="nav-link" data-scroll-target="#random-number-generator-arguments">Random number generator arguments</a></li>
+  <li><a href="#chain-identifier-argument-id" id="toc-chain-identifier-argument-id" class="nav-link" data-scroll-target="#chain-identifier-argument-id">Chain identifier argument: <code>id</code></a></li>
+  <li><a href="#command-line-help" id="toc-command-line-help" class="nav-link" data-scroll-target="#command-line-help">Command line help</a></li>
+  <li><a href="#error-messages-and-return-codes" id="toc-error-messages-and-return-codes" class="nav-link" data-scroll-target="#error-messages-and-return-codes">Error messages and return codes</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/command_line_options.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Command-Line Interface Overview</a></li></ol></nav></header>
+
+
+
+
+<section id="command-line-interface-overview" class="level1">
+<h1>Command-Line Interface Overview</h1>
+<p>A CmdStan executable is built from the Stan model concept and the CmdStan command line parser. The command line argument syntax consists of sets of keywords and keyword-value pairs. Arguments are grouped by the following keywords:</p>
+<ul>
+<li><p><code>method</code> - specifies the kind of inference done on the model.<br> Each kind of inference requires further configuration via sub-arguments. The <code>method</code> argument is required. It can be specified overtly as the a keyword-value pair <code>method=&lt;inference&gt;</code> or implicitly as one of the following:</p>
+<ul>
+<li><code>sample</code> - obtain a sample (set of draws) from the posterior using HMC</li>
+<li><code>optimize</code> - penalized maximum likelihood estimation</li>
+<li><code>pathfinder</code> - quasi-Newton variational inference</li>
+<li><code>variational</code> - automatic differentation variational inference (ADVI)</li>
+<li><code>laplace</code> - sample from a normal approximation centered at the mode</li>
+<li><code>generate_quantities</code> - run model’s <code>generated quantities</code> block on existing sample to obtain new quantities of interest.</li>
+<li><code>log_prob</code> - compute the log probability and gradient of the model for one set of parameters.</li>
+<li><code>diagnose</code> - compute and compare sampler gradient calculations to finite differences.</li>
+</ul></li>
+<li><p><code>data</code> - specifies the input data file, if any.</p></li>
+<li><p><code>output</code> - specifies program outputs, both disk files and terminal window outputs.</p></li>
+<li><p><code>init</code> - specifies initial values for the model parameters, if any.</p></li>
+<li><p><code>random</code> - specifies the seed for the pseudo-random number.</p></li>
+</ul>
+<p>The remainder of this chapter covers the general configuration options used for all processing. The following chapters cover the per-inference configuration options.</p>
+<section id="input-data-argument" class="level2">
+<h2 class="anchored" data-anchor-id="input-data-argument">Input data argument</h2>
+<p>The values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:</p>
+<pre><code>data file=&lt;filepath&gt;</code></pre>
+<p>The keyword <code>data</code> must be followed directly by the keyword-value pair <code>file=&lt;filepath&gt;</code>. If the model doesn’t declare any data variables, this argument is ignored.</p>
+<p>The input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model <code>bernoulli.stan</code> defines two data variables <code>N</code> and <code>y</code>. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.</p>
+<p>For example if the input data file doesn’t include the definition for variable <code>y</code>, the executable exits with the following message:</p>
+<pre><code>Exception: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)</code></pre>
+</section>
+<section id="output-control.section" class="level2">
+<h2 class="anchored" data-anchor-id="output-control.section">Output control arguments</h2>
+<p>The <code>output</code> keyword is used to specify non-default options for output files and messages written to the terminal window. The <code>output</code> keyword takes several keyword-value pair sub-arguments.</p>
+<p>The keyword value pair <code>file=&lt;filepath&gt;</code> specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named <code>output.csv</code> in the current working directory.</p>
+<p>The keyword value pair <code>diagnostic_file=&lt;filepath&gt;</code> specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms <code>sample</code> and <code>variational</code>.</p>
+<p>The keyword value pair <code>refresh=&lt;int&gt;</code> specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.</p>
+<p>The keyword value pair <code>sig_figs=&lt;int&gt;</code> specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8. &nbsp; <strong><em>Note:</em></strong> increasing <code>sig_figs</code> above the default will increase the size of the output CSV files accordingly.</p>
+<p>The keyword value pair <code>profile_file=&lt;filepath&gt;</code> specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and <code>profile_file</code> is unspecified, the profiling data is written to a file named <code>profile.csv</code> in the current working directory.</p>
+<p>The keyword value pair <code>save_cmdstan_config=&lt;boolean&gt;</code> specifies whether to save the configuration options used to run the program to a file named <code>&lt;output file&gt;_config.json</code> alongside the other output files. The default value is <code>false</code>, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.</p>
+</section>
+<section id="initialize-model-parameters-argument" class="level2">
+<h2 class="anchored" data-anchor-id="initialize-model-parameters-argument">Initialize model parameters argument</h2>
+<p>Initialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range <span class="math inline">\([-2, 2]\)</span>. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in <span class="math inline">\((-2, 2)\)</span>, variables constrained to be positive are initialized roughly in <span class="math inline">\((0.14, 7.4)\)</span>, variables constrained to fall between 0 and 1 are initialized with values roughly in <span class="math inline">\((0.12, 0.88)\)</span>.</p>
+<p>The initialization argument is specified as keyword-value pair with keyword <code>init</code>. The value can be one of the following:</p>
+<ul>
+<li><p>positive real number <span class="math inline">\(x\)</span>. All parameters will be initialized to random draws from a uniform distribution over the range <span class="math inline">\([-x, x]\)</span>.</p></li>
+<li><p><span class="math inline">\(0\)</span> - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: <span class="math inline">\(0\)</span> for unconstrained parameters; <span class="math inline">\(1\)</span> for parameters constrained to be positive; <span class="math inline">\(0.5\)</span> for variables to constrained to lie between <span class="math inline">\(0\)</span> and <span class="math inline">\(1\)</span>; a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.</p></li>
+<li><p>filepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range <span class="math inline">\([-2, 2]\)</span>.</p></li>
+</ul>
+</section>
+<section id="random-number-generator-arguments" class="level2">
+<h2 class="anchored" data-anchor-id="random-number-generator-arguments">Random number generator arguments</h2>
+<p>The random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.</p>
+<p>The syntax for the random seed argument is:</p>
+<pre><code>random seed=&lt;int&gt;</code></pre>
+<p>The keyword <code>random</code> must be followed directly by the keyword-value pair <code>seed=&lt;int&gt;</code>.</p>
+</section>
+<section id="chain-identifier-argument-id" class="level2">
+<h2 class="anchored" data-anchor-id="chain-identifier-argument-id">Chain identifier argument: <code>id</code></h2>
+<p>The chain identifier argument is used in conjunction with the <code>random seed</code> argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.</p>
+<p>The syntax for the random seed argument is:</p>
+<pre><code>id=&lt;int&gt;</code></pre>
+<p>The default value is 1.</p>
+<p>When running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.</p>
+<p>For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the <a href="https://mc-stan.org/docs/reference-manual/reproducibility.html">Stan Reference Manual Reproducibility chapter</a> for further details.</p>
+</section>
+<section id="command-line-help" class="level2">
+<h2 class="anchored" data-anchor-id="command-line-help">Command line help</h2>
+<p>CmdStan provides a <code>help</code> and <code>help-all</code> mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword <code>help</code>:</p>
+<pre><code>./bernoulli help</code></pre>
+</section>
+<section id="error-messages-and-return-codes" class="level2">
+<h2 class="anchored" data-anchor-id="error-messages-and-return-codes">Error messages and return codes</h2>
+<p>CmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.</p>
+<p>All program executables provide a return code between 0 and 255:</p>
+<ul>
+<li><p>0 - Program ran to termination as expected.</p></li>
+<li><p>value in range [1 : 125] - Method invoked could not run due to problems with model or data.</p></li>
+<li><p>value &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g.&nbsp;return code 139 results from termination signal 11 (segmentation violation).</p></li>
+</ul>
+<p>A non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/parallelization.html" class="pagination-link" aria-label="Parallelization">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Parallelization</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/mcmc_config.html" class="pagination-link" aria-label="MCMC Sampling using Hamiltonian Monte Carlo">
+        <span class="nav-page-text">MCMC Sampling using Hamiltonian Monte Carlo</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/command_line_options.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/compiling_stan_programs.html b/docs/2_39/cmdstan-guide/compiling_stan_programs.html
new file mode 100644
index 000000000..2b151827e
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/compiling_stan_programs.html
@@ -0,0 +1,1191 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Compiling a Stan Program</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/parallelization.html" rel="next">
+<link href="../cmdstan-guide/example_model_data.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Compiling a Stan Program">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/compiling_stan_programs.html">Compiling a Stan Program</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#compiling-a-stan-program" id="toc-compiling-a-stan-program" class="nav-link active" data-scroll-target="#compiling-a-stan-program">Compiling a Stan Program</a>
+  <ul class="collapse">
+  <li><a href="#invoking-the-make-utility" id="toc-invoking-the-make-utility" class="nav-link" data-scroll-target="#invoking-the-make-utility">Invoking the Make utility</a></li>
+  <li><a href="#dependencies" id="toc-dependencies" class="nav-link" data-scroll-target="#dependencies">Dependencies</a></li>
+  <li><a href="#compiler-errors" id="toc-compiler-errors" class="nav-link" data-scroll-target="#compiler-errors">Compiler errors</a></li>
+  <li><a href="#troubleshooting-c-compiler-or-linker-errors" id="toc-troubleshooting-c-compiler-or-linker-errors" class="nav-link" data-scroll-target="#troubleshooting-c-compiler-or-linker-errors">Troubleshooting C++ compiler or linker errors</a></li>
+  <li><a href="#c-compilation-and-linking-flags" id="toc-c-compilation-and-linking-flags" class="nav-link" data-scroll-target="#c-compilation-and-linking-flags">C++ compilation and linking flags</a>
+  <ul class="collapse">
+  <li><a href="#optimizing-by-ignoring-range-checks" id="toc-optimizing-by-ignoring-range-checks" class="nav-link" data-scroll-target="#optimizing-by-ignoring-range-checks">Optimizing by ignoring range checks</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/compiling_stan_programs.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/compiling_stan_programs.html">Compiling a Stan Program</a></li></ol></nav></header>
+
+
+
+
+<section id="compiling-a-stan-program" class="level1">
+<h1>Compiling a Stan Program</h1>
+<p>A Stan program must be in a file with extension <code>.stan</code>. The CmdStan makefile rules specify all necessary steps to translate files with suffix <code>.stan</code> to a CmdStan executable program. This is a two-stage process:</p>
+<ul>
+<li>first the Stan program is translated to C++ by the <code>stanc</code> compiler</li>
+<li>then the C++ compiler compiles all C++ sources and links them together with the CmdStan interface program and the Stan and math libraries.</li>
+</ul>
+<section id="invoking-the-make-utility" class="level2">
+<h2 class="anchored" data-anchor-id="invoking-the-make-utility">Invoking the Make utility</h2>
+<p>To compile Stan programs, you must invoke the Make program from the <code>&lt;cmdstan-home&gt;</code> directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.</p>
+<pre><code>&gt; cd &lt;cmdstan_home&gt;</code></pre>
+<p>In the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the <code>.stan</code> omitted. On Windows, replace <code>.stan</code> with <code>.exe</code>, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:</p>
+<pre><code>&gt; make examples/bernoulli/bernoulli</code></pre>
+<p>On Windows, the command is the same with the addition of <code>.exe</code> at the end of the target (<em>note the use of forward slashes</em>):</p>
+<pre><code>&gt; make examples/bernoulli/bernoulli.exe</code></pre>
+<p>The generated C++ code (<code>bernoulli.hpp</code>), object file (<code>bernoulli.o</code>) and the compiled executable will be placed in the same directory as the Stan program.</p>
+<p>The compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.</p>
+</section>
+<section id="dependencies" class="level2">
+<h2 class="anchored" data-anchor-id="dependencies">Dependencies</h2>
+<p>When executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target <code>bernoulli</code> twice in a row, without any editing <code>bernoulli.stan</code> or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:</p>
+<pre><code>&gt; make examples/bernoulli/bernoulli
+make: `examples/bernoulli/bernoulli' is up to date.</code></pre>
+<p>If the file containing the Stan program is updated, the next call to <code>make</code> will rebuild the CmdStan executable.</p>
+</section>
+<section id="compiler-errors" class="level2">
+<h2 class="anchored" data-anchor-id="compiler-errors">Compiler errors</h2>
+<p>The Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.</p>
+<p>The simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the <code>bernoulli.stan</code> program, we introduce a typo on line <span class="math inline">\(9\)</span> by writing <code>thata</code> instead of <code>theta</code>, the Make command fails with the following</p>
+<pre><code>--- Translating Stan model to C++ code ---
+bin/stanc  --o=bernoulli.hpp bernoulli.stan
+
+Semantic error in 'bernoulli.stan', line 9, column 2 to column 7:
+   -------------------------------------------------
+     7:  }
+     8:  model {
+     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1
+           ^
+    10:    y ~ bernoulli(theta);
+    11:  }
+   -------------------------------------------------
+
+Identifier 'thata' not in scope.
+
+make: *** [bernoulli.hpp] Error 1</code></pre>
+<p>Stan is a <a href="https://mc-stan.org/docs/reference-manual/data-types.html">strongly-typed language</a>; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program <code>foo.stan</code> contains an illegal assignment statement:</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> y = x;</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Make command fails with the following:</p>
+<pre><code>Semantic error in 'foo.stan', line 5, column 2 to column 12:
+   -------------------------------------------------
+     3:  }
+     4:  transformed data {
+     5:    int y = x;
+           ^
+     6:  }
+   -------------------------------------------------
+
+Ill-typed arguments supplied to assignment operator =:
+ lhs has type int and rhs has type real</code></pre>
+<p>The <a href="https://mc-stan.org/docs/reference-manual/index.html">Stan Reference Manual</a> provides a complete specification of the Stan programming language. The <a href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html">Stan User’s Guide</a> also contains a full description of the errors and warnings stanc can emit.</p>
+</section>
+<section id="troubleshooting-c-compiler-or-linker-errors" class="level2">
+<h2 class="anchored" data-anchor-id="troubleshooting-c-compiler-or-linker-errors">Troubleshooting C++ compiler or linker errors</h2>
+<p>If the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.</p>
+<p>The Make command prints the following message to the terminal at the point when it compiles and links the C++ file:</p>
+<pre><code>--- Compiling, linking C++ code ---</code></pre>
+<p>If the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.</p>
+<p>If this happens, please report the error, together with the Stan program on either the <a href="https://discourse.mc-stan.org/">Stan Forums</a> or on the Stan compiler GitHub <a href="https://github.com/stan-dev/stanc3/issues">issues tracker</a>.</p>
+</section>
+<section id="c-compilation-and-linking-flags" class="level2">
+<h2 class="anchored" data-anchor-id="c-compilation-and-linking-flags">C++ compilation and linking flags</h2>
+<p>Users can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.</p>
+<p>The <code>CXXFLAGS</code> and <code>LDFLAGS</code> makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the <code>make/local</code> file.</p>
+<p>For example:</p>
+<pre><code>CXXFLAGS = -O2</code></pre>
+<p>A recommend a set of <code>CXXFLAGS</code> and <code>LDFLAGS</code> flags can be turned on by setting <code>STAN_CPP_OPTIMS=true</code> in the <code>make/local</code> file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.</p>
+<section id="optimizing-by-ignoring-range-checks" class="level3">
+<h3 class="anchored" data-anchor-id="optimizing-by-ignoring-range-checks">Optimizing by ignoring range checks</h3>
+<p>When assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.</p>
+<p>For some models these checks can represent a significant part of the models execution time. By setting the <code>STAN_NO_RANGE_CHECKS=true</code> makefile flag in the <code>make/local</code> file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.</p>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/example_model_data.html" class="pagination-link" aria-label="Example Model and Data">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Example Model and Data</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/parallelization.html" class="pagination-link" aria-label="Parallelization">
+        <span class="nav-page-text">Parallelization</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/compiling_stan_programs.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/diagnose_config.html b/docs/2_39/cmdstan-guide/diagnose_config.html
new file mode 100644
index 000000000..e590ceb56
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/diagnose_config.html
@@ -0,0 +1,1102 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Diagnosing HMC by Comparison of Gradients</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/stanc.html" rel="next">
+<link href="../cmdstan-guide/log_prob_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Diagnosing HMC by Comparison of Gradients">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/diagnose_config.html">Diagnosing HMC by Comparison of Gradients</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#diagnosing-hmc-by-comparison-of-gradients" id="toc-diagnosing-hmc-by-comparison-of-gradients" class="nav-link active" data-scroll-target="#diagnosing-hmc-by-comparison-of-gradients">Diagnosing HMC by Comparison of Gradients</a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/diagnose_config.html">Diagnosing HMC by Comparison of Gradients</a></li></ol></nav></header>
+
+
+
+
+<section id="diagnosing-hmc-by-comparison-of-gradients" class="level1">
+<h1>Diagnosing HMC by Comparison of Gradients</h1>
+<p>CmdStan has a basic diagnostic feature that will calculate the gradients of the initial state and compare them with gradients calculated by finite differences. Discrepancies between the two indicate that there is a problem with the model or initial states or else there is a bug in Stan.</p>
+<p>To allow for the possibility of adding other kinds of diagnostic tests, the <code>diagnose</code> method argument configuration has subargument <code>test</code> which currently only takes value <code>gradient</code>. There are two available gradient test configuration arguments:</p>
+<ul>
+<li><p><code>epsilon</code> - The finite difference step size. Must be a positive real number. Default value is <span class="math inline">\(1^{-6}\)</span></p></li>
+<li><p><code>error</code> - The error threshold. Must be a positive real number. Default value is <span class="math inline">\(1^{-6}\)</span></p></li>
+</ul>
+<p>To run on the different platforms with the default configuration, use one of the following.</p>
+<p>Mac OS and Linux</p>
+<pre><code>&gt; ./my_model diagnose data file=my_data</code></pre>
+<p>Windows</p>
+<pre><code>&gt; my_model diagnose data file=my_data</code></pre>
+<p>To relax the test threshold, specify the <code>error</code> argument as follows:</p>
+<pre><code>&gt; ./my_model diagnose test=gradient error=0.0001 data file=my_data</code></pre>
+<p>To see how this works, we run diagnostics on the example bernoulli model:</p>
+<pre><code>&gt; ./bernoulli diagnose data file=bernoulli.data.R</code></pre>
+<p>Executing this command prints output to the console and as a series of comment lines to the output csv file. The console output is:</p>
+<pre><code>method = diagnose
+  diagnose
+    test = gradient (Default)
+      gradient
+        epsilon = 9.9999999999999995e-07 (Default)
+        error = 9.9999999999999995e-07 (Default)
+id = 0 (Default)
+data
+  file = bernoulli.data.json
+init = 2 (Default)
+random
+  seed = 2152196153 (Default)
+output
+  file = output.csv (Default)
+  diagnostic_file =  (Default)
+  refresh = 100 (Default)
+
+TEST GRADIENT MODE
+
+ Log probability=-8.42814
+
+ param idx           value           model     finite diff           error
+         0       0.0361376         -3.1084         -3.1084    -2.37554e-10</code></pre>
+<p>The same information is printed to the output file as csv comments, i.e., each line is prefixed with a pound sign <code>#</code>.</p>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/log_prob_config.html" class="pagination-link" aria-label="Extracting log probabilities and gradients for diagnostics">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Extracting log probabilities and gradients for diagnostics</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/stanc.html" class="pagination-link" aria-label="`stanc`: Translating Stan to C++">
+        <span class="nav-page-text"><code>stanc</code>: Translating Stan to C++</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/diagnose_utility.html b/docs/2_39/cmdstan-guide/diagnose_utility.html
new file mode 100644
index 000000000..804b2d293
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/diagnose_utility.html
@@ -0,0 +1,1226 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Diagnosing Biased Hamiltonian Monte Carlo Inferences</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/print.html" rel="next">
+<link href="../cmdstan-guide/stansummary.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Diagnosing Biased Hamiltonian Monte Carlo Inferences">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/diagnose_utility.html"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#diagnose" id="toc-diagnose" class="nav-link active" data-scroll-target="#diagnose"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</a>
+  <ul class="collapse">
+  <li><a href="#building-the-diagnose-command" id="toc-building-the-diagnose-command" class="nav-link" data-scroll-target="#building-the-diagnose-command">Building the diagnose command</a></li>
+  <li><a href="#running-the-diagnose-command" id="toc-running-the-diagnose-command" class="nav-link" data-scroll-target="#running-the-diagnose-command">Running the diagnose command</a></li>
+  <li><a href="#diagnose-warnings-and-recommendations" id="toc-diagnose-warnings-and-recommendations" class="nav-link" data-scroll-target="#diagnose-warnings-and-recommendations"><code>diagnose</code> warnings and recommendations</a>
+  <ul class="collapse">
+  <li><a href="#divergent-transitions-after-warmup" id="toc-divergent-transitions-after-warmup" class="nav-link" data-scroll-target="#divergent-transitions-after-warmup">Divergent transitions after warmup</a></li>
+  <li><a href="#maximum-treedepth-exceeded" id="toc-maximum-treedepth-exceeded" class="nav-link" data-scroll-target="#maximum-treedepth-exceeded">Maximum treedepth exceeded</a></li>
+  <li><a href="#low-e-bfmi-values---sampler-transitions-hmc-potential-energy." id="toc-low-e-bfmi-values---sampler-transitions-hmc-potential-energy." class="nav-link" data-scroll-target="#low-e-bfmi-values---sampler-transitions-hmc-potential-energy.">Low E-BFMI values - sampler transitions HMC potential energy.</a></li>
+  <li><a href="#low-effective-sample-sizes" id="toc-low-effective-sample-sizes" class="nav-link" data-scroll-target="#low-effective-sample-sizes">Low effective sample sizes</a></li>
+  <li><a href="#high-hatr" id="toc-high-hatr" class="nav-link" data-scroll-target="#high-hatr">High <span class="math inline">\(\hat{R}\)</span></a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_utility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/diagnose_utility.html"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</a></li></ol></nav></header>
+
+
+
+
+<section id="diagnose" class="level1">
+<h1><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</h1>
+<p>CmdStan is distributed with a utility that is able to read in and analyze the output of one or more Markov chains to check for the following potential problems:</p>
+<ul>
+<li>Divergent transitions</li>
+<li>Transitions that hit the maximum treedepth</li>
+<li>Low E-BFMI values</li>
+<li>Low effective sample sizes</li>
+<li>High <span class="math inline">\(\hat{R}\)</span> values</li>
+</ul>
+<p>The meanings of several of these problems are discussed in https://arxiv.org/abs/1701.02434.</p>
+<section id="building-the-diagnose-command" class="level2">
+<h2 class="anchored" data-anchor-id="building-the-diagnose-command">Building the diagnose command</h2>
+<p>The CmdStan makefile task <code>build</code> compiles the <code>diagnose</code> utility into the <code>bin</code> directory. It can be compiled directly using the makefile as follows:</p>
+<pre><code>&gt; cd &lt;cmdstan-home&gt;
+&gt; make bin/diagnose</code></pre>
+</section>
+<section id="running-the-diagnose-command" class="level2">
+<h2 class="anchored" data-anchor-id="running-the-diagnose-command">Running the diagnose command</h2>
+<p>The <code>diagnose</code> command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to <code>diagnose</code>, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.</p>
+<p>To fully exercise the <code>diagnose</code> command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the <a href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html">Stan User’s Guide reparameterization section</a>. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">9</span>] x;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(<span class="dv">0</span>, <span class="dv">3</span>);</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  x ~ normal(<span class="dv">0</span>, exp(y / <span class="dv">2</span>));</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan</p>
+<p>Stan has trouble sampling from the region where <code>y</code> is small and thus <code>x</code> is constrained to be near 0. This is due to the fact that the density’s scale changes with <code>y</code>, so that a step size that works well when <code>y</code> is large is inefficient when <code>y</code> is small and vice-versa.</p>
+<p>Running 4 chains produces output files <code>output_1.csv</code>, …, <code>output_4.csv</code>. We run <code>diagnose</code> command on this fileset:</p>
+<pre><code>&gt; bin/diagnose output_*.csv</code></pre>
+<p>The output is printed to the terminal window:</p>
+<pre><code>Checking sampler transitions treedepth.
+18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.
+Trajectories that are prematurely terminated due to this limit will result in slow exploration.
+For optimal performance, increase this limit.
+
+Checking sampler transitions for divergences.
+11 of 4000 (0.28%) transitions ended with a divergence.
+These divergent transitions indicate that HMC is not fully able to explore the posterior distribution.
+Try increasing adapt delta closer to 1.
+If this doesn't remove all divergences, try to reparameterize the model.
+
+Checking E-BFMI - sampler transitions HMC potential energy.
+The E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.
+If possible, try to reparameterize the model.
+
+Rank-normalized split effective sample size satisfactory for all parameters.
+
+The following parameters had rank-normalized split R-hat greater than 1.01:
+  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]
+Such high values indicate incomplete mixing and biased estimation.
+You should consider regularizing your model with additional prior information or a more effective parameterization.
+
+Processing complete.</code></pre>
+<p>In this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters <code>x_raw</code> and <code>y_raw</code> are sampled as independent standard normals, which is easy for Stan.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y_raw;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">9</span>] x_raw;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">9</span>] x;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  y = <span class="fl">3.0</span> * y_raw;</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  x = exp(y / <span class="dv">2</span>) * x_raw;</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>  y_raw ~ std_normal(); <span class="co">// implies y ~ normal(0, 3)</span></span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>  x_raw ~ std_normal(); <span class="co">// implies x ~ normal(0, exp(y / 2))</span></span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan</p>
+<p>We compile the program and run 4 chains, as before. Now the <code>diagnose</code> command doesn’t detect any problems:</p>
+<pre><code>Checking sampler transitions treedepth.
+Treedepth satisfactory for all transitions.
+
+Checking sampler transitions for divergences.
+No divergent transitions found.
+
+Checking E-BFMI - sampler transitions HMC potential energy.
+E-BFMI satisfactory.
+
+Rank-normalized split effective sample size satisfactory for all parameters.
+
+Rank-normalized split R-hat values satisfactory for all parameters.
+
+Processing complete, no problems detected.</code></pre>
+</section>
+<section id="diagnose-warnings-and-recommendations" class="level2">
+<h2 class="anchored" data-anchor-id="diagnose-warnings-and-recommendations"><code>diagnose</code> warnings and recommendations</h2>
+<section id="divergent-transitions-after-warmup" class="level3">
+<h3 class="anchored" data-anchor-id="divergent-transitions-after-warmup">Divergent transitions after warmup</h3>
+<p>Stan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a <a href="https://en.wikipedia.org/wiki/Hamiltonian_system">Hamiltonian system</a>. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the <em>step size controls the resolution of the sampler</em>.</p>
+<p>Unfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as <em>divergences</em> which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.</p>
+<p>If the divergent transitions cannot be eliminated by increasing the <code>adapt_delta</code> parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the <a href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html">reparameterization section</a> of the Stan User’s Guide.</p>
+</section>
+<section id="maximum-treedepth-exceeded" class="level3">
+<h3 class="anchored" data-anchor-id="maximum-treedepth-exceeded">Maximum treedepth exceeded</h3>
+<p>Warnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a <em>validity</em> concern, hitting the maximum treedepth is an <em>efficiency</em> concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the <em>Hamiltonian Monte Carlo Sampling</em> chapter in the <a href="https://mc-stan.org/docs/reference-manual/mcmc.html">Stan Reference Manual</a>). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.</p>
+<p>This is controlled through the <code>max_depth</code> argument. If the number of transitions which exceed maximum treedepth is low, increasing <code>max_depth</code> may correct this problem.</p>
+</section>
+<section id="low-e-bfmi-values---sampler-transitions-hmc-potential-energy." class="level3">
+<h3 class="anchored" data-anchor-id="low-e-bfmi-values---sampler-transitions-hmc-potential-energy.">Low E-BFMI values - sampler transitions HMC potential energy.</h3>
+<p>The sampler csv output column <code>energy__</code> is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of <code>energy</code> is much larger than <span class="math inline">\(\sqrt{D / 2}\)</span>, where <span class="math inline">\(D\)</span> is the number of <em>unconstrained</em> parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.</p>
+<p>The warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.</p>
+</section>
+<section id="low-effective-sample-sizes" class="level3">
+<h3 class="anchored" data-anchor-id="low-effective-sample-sizes">Low effective sample sizes</h3>
+<p>Roughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses <span class="math inline">\(\hat{R}\)</span> adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.</p>
+<p>Bulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g.&nbsp;mean and median).</p>
+<p>Often quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.</p>
+</section>
+<section id="high-hatr" class="level3">
+<h3 class="anchored" data-anchor-id="high-hatr">High <span class="math inline">\(\hat{R}\)</span></h3>
+<p><span class="math inline">\(\hat{R}\)</span> (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), <span class="math inline">\(\hat{R}\)</span> is larger than 1. We recommend running at least four chains by default and only using the sample if <span class="math inline">\(\hat{R}\)</span> is less than 1.01. Stan reports <span class="math inline">\(\hat{R}\)</span> which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.</p>
+<p>There is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.</p>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/stansummary.html" class="pagination-link" aria-label="`stansummary`: MCMC Output Analysis">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text"><code>stansummary</code>: MCMC Output Analysis</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/print.html" class="pagination-link" aria-label="`print` (deprecated): MCMC Output Analysis">
+        <span class="nav-page-text"><code>print</code> (deprecated): MCMC Output Analysis</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_utility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/example_model_data.html b/docs/2_39/cmdstan-guide/example_model_data.html
new file mode 100644
index 000000000..7bd1e760e
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/example_model_data.html
@@ -0,0 +1,1091 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Example Model and Data</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/compiling_stan_programs.html" rel="next">
+<link href="../cmdstan-guide/installation.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Example Model and Data">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/example_model_data.html">Example Model and Data</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#example-model-and-data" id="toc-example-model-and-data" class="nav-link active" data-scroll-target="#example-model-and-data">Example Model and Data</a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/example_model_data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/example_model_data.html">Example Model and Data</a></li></ol></nav></header>
+
+
+
+
+<section id="example-model-and-data" class="level1">
+<h1>Example Model and Data</h1>
+<p>The following is a simple, complete Stan program for a Bernoulli model of binary data.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> The model assumes the binary observed data <code>y[1],...,y[N]</code> are i.i.d. with Bernoulli chance-of-success <code>theta</code>.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> { </span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N; </span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>} </span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>} </span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  theta ~ beta(<span class="dv">1</span>, <span class="dv">1</span>);  <span class="co">// uniform prior on interval 0,1</span></span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli(theta);</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The input data file contains definitions for the two variables <code>N</code> and <code>y</code> which are specified in the data block of program <code>bernoulli.stan</code> (above).</p>
+<p>A data set of <code>N=10</code> observations is included in the example Bernoulli model directory in both JSON notation and Rdump data format where 8 out of 10 trials had outcome <code>0</code> (failure) and 2 trials had outcome <code>1</code> (success). In JSON, this data is:</p>
+<pre><code>{
+    "N" : 10,
+    "y" : [0,1,0,0,0,0,0,0,0,1]
+}</code></pre>
+
+
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The model is available with the CmdStan distribution at the path <code>&lt;cmdstan-home&gt;/examples/bernoulli/bernoulli.stan</code>.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/installation.html" class="pagination-link" aria-label="CmdStan Installation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">CmdStan Installation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/compiling_stan_programs.html" class="pagination-link" aria-label="Compiling a Stan Program">
+        <span class="nav-page-text">Compiling a Stan Program</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/example_model_data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/external_code.html b/docs/2_39/cmdstan-guide/external_code.html
new file mode 100644
index 000000000..591f51b58
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/external_code.html
@@ -0,0 +1,1197 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Using external C++ code</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/rdump_apdx.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Using external C++ code">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/external_code.html">Using external C++ code</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#using-external-cpp-code" id="toc-using-external-cpp-code" class="nav-link active" data-scroll-target="#using-external-cpp-code">Using external C++ code</a>
+  <ul class="collapse">
+  <li><a href="#derivative-specializations" id="toc-derivative-specializations" class="nav-link" data-scroll-target="#derivative-specializations">Derivative specializations</a></li>
+  <li><a href="#special-functions-rngs-distributions-editing-target" id="toc-special-functions-rngs-distributions-editing-target" class="nav-link" data-scroll-target="#special-functions-rngs-distributions-editing-target">Special functions: RNGs, distributions, editing <code>target</code></a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/external_code.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/external_code.html">Using external C++ code</a></li></ol></nav></header>
+
+
+
+
+<section id="using-external-cpp-code" class="level1">
+<h1>Using external C++ code</h1>
+<p>The <code>--allow-undefined</code> flag can be passed to the call to stanc, which will allow undefined functions in the Stan language to be parsed without an error. We can then include a definition of the function in a C++ header file.</p>
+<p>This requires specifying two makefile variables:</p>
+<ul>
+<li><code>STANCFLAGS=--allow-undefined</code></li>
+<li><code>USER_HEADER=&lt;header_file.hpp&gt;</code>, where <code>&lt;header_file.hpp&gt;</code> is the name of a header file that defines a function with the same name and a compatible signature. This function can appear in the global namespace or in the model namespace, which is defined as the name of the model (either the file name, or the <code>--name</code> argument to stanc) followed by <code>_namespace</code>.</li>
+</ul>
+<p>This is an advanced feature which is only recommended to users familiar with the internals of Stan’s Math library. Most existing C++ code will need to be modified to work with Stan, to varying degrees.</p>
+<p>As an example, consider the following variant of the Bernoulli example</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> make_odds(<span class="kw">data</span> <span class="dt">real</span> theta);</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  theta ~ beta(<span class="dv">1</span>, <span class="dv">1</span>); <span class="co">// uniform prior on interval 0, 1</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli(theta);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> odds;</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>  odds = make_odds(theta);</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here the <code>make_odds</code> function is declared but not defined, which would ordinarily result in a parser error. However, if you put <code>STANCFLAGS = --allow-undefined</code> into the <code>make/local</code> file or into the <code>stanc</code> call, then the stanc compiler will translate this program to C++, but the generated C++ code will not compile unless you write a file such as <code>examples/bernoulli/make_odds.hpp</code> with the following lines</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode c++ code-with-copy"><code class="sourceCode cpp"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="pp">#include </span><span class="im">&lt;ostream&gt;</span></span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="dt">double</span> make_odds<span class="op">(</span><span class="at">const</span> <span class="dt">double</span><span class="op">&amp;</span> theta<span class="op">,</span> <span class="bu">std::</span>ostream <span class="op">*</span><span class="va">pstream__</span><span class="op">)</span> <span class="op">{</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> theta <span class="op">/</span> <span class="op">(</span><span class="dv">1</span> <span class="op">-</span> theta<span class="op">);</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="op">}</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The signature for this function needs to fulfill all the usages in the C++ class emitted by stanc. The <code>pstream__</code> argument is mandatory in the signature but need not be used if your function does not print any output. Because <code>make_odds</code> was declared with a <code>data</code> argument and only used in <code>generated quantites</code>, a signature which accepts and returns <code>double</code> is acceptable. Functions which will have parameters passed as input in the <code>transformed parameters</code> or <code>model</code> blocks will require the ability to accept Stan’s autodiff types. If you wish to autodiff through this function, the simplest option is to make it a template, like</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode c++ code-with-copy"><code class="sourceCode cpp"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">template</span> <span class="op">&lt;</span><span class="kw">typename</span> T<span class="op">&gt;</span></span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>T make_odds<span class="op">(</span><span class="at">const</span> T <span class="op">&amp;</span>theta<span class="op">,</span> <span class="bu">std::</span>ostream <span class="op">*</span><span class="va">pstream__</span><span class="op">)</span></span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="op">{</span></span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> theta <span class="op">/</span> <span class="op">(</span><span class="dv">1</span> <span class="op">-</span> theta<span class="op">);</span></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="op">}</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Given the above, the following <code>make</code> invocation should work</p>
+<pre><code>&gt; make STANCFLAGS=--allow-undefined USER_HEADER=examples/bernoulli/make_odds.hpp examples/bernoulli/bernoulli # on Windows add .exe</code></pre>
+<p>Alternatively, you could put <code>STANCFLAGS</code> and <code>USER_HEADER</code> into the <code>make/local</code> file instead of specifying them on the command-line.</p>
+<p>If the function were more complicated and involved functions in the Stan Math Library, then you would need to add <code>#include &lt;stan/model/model_header.hpp&gt;</code> and prefix the function calls with <code>stan::math::</code>.</p>
+<section id="derivative-specializations" class="level2">
+<h2 class="anchored" data-anchor-id="derivative-specializations">Derivative specializations</h2>
+<p>External C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the <a href="https://mc-stan.org/math/md_doxygen_2contributor__help__pages_2getting__started.html">Stan Math documentation</a>.</p>
+<p>Suppose you have the following (nonsensical) model which relies on a function called <code>my_dot_self</code>. We will implement this as a copy of the built-in <code>dot_self</code> function.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// both overloads end up using the same C++ template</span></span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> my_dot_self(<span class="dt">vector</span> theta);</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> my_dot_self(<span class="dt">row_vector</span> theta);</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] input_data;</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// no autodiff for data - will call using doubles</span></span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ds = my_dot_self(input_data);</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">row_vector</span>[N] thetas;</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>  thetas ~ normal(<span class="dv">0</span>,<span class="dv">1</span>);</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>  <span class="co">// autodiff - will call using stan::math::var types</span></span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>  input_data ~ normal(thetas, my_dot_self(thetas));</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If you wanted to autodiff through this function, the following header would suffice<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a>:</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode c++ code-with-copy"><code class="sourceCode cpp"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="pp">#include </span><span class="im">&lt;stan/model/model_header.hpp&gt;</span></span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="pp">#include </span><span class="im">&lt;ostream&gt;</span></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="kw">template</span> <span class="op">&lt;</span><span class="kw">typename</span> EigVec<span class="op">,</span> stan<span class="op">::</span><span class="dt">require_eigen_vector_t</span><span class="op">&lt;</span>EigVec<span class="op">&gt;</span> <span class="op">*</span> <span class="op">=</span> <span class="kw">nullptr</span><span class="op">&gt;</span></span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="kw">inline</span> stan<span class="op">::</span><span class="dt">value_type_t</span><span class="op">&lt;</span>EigVec<span class="op">&gt;</span> my_dot_self<span class="op">(</span><span class="at">const</span> EigVec <span class="op">&amp;</span>x<span class="op">,</span> <span class="bu">std::</span>ostream <span class="op">*</span><span class="va">pstream__</span><span class="op">)</span></span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a><span class="op">{</span></span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>    <span class="at">const</span> <span class="kw">auto</span> <span class="op">&amp;</span>x_ref <span class="op">=</span> stan<span class="op">::</span>math<span class="op">::</span>to_ref<span class="op">(</span>x<span class="op">);</span></span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>    stan<span class="op">::</span><span class="dt">value_type_t</span><span class="op">&lt;</span>EigVec<span class="op">&gt;</span> sum_x <span class="op">=</span> <span class="fl">0.0</span><span class="op">;</span></span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> <span class="op">(</span><span class="dt">int</span> i <span class="op">=</span> <span class="dv">0</span><span class="op">;</span> i <span class="op">&lt;</span> x<span class="op">.</span>size<span class="op">();</span> <span class="op">++</span>i<span class="op">)</span></span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>    <span class="op">{</span></span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>        sum_x <span class="op">+=</span> x_ref<span class="op">.</span>coeff<span class="op">(</span>i<span class="op">)</span> <span class="op">*</span> x_ref<span class="op">.</span>coeff<span class="op">(</span>i<span class="op">);</span></span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>    <span class="op">}</span></span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> sum_x<span class="op">;</span></span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a><span class="op">}</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>However, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode c++ code-with-copy"><code class="sourceCode cpp"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="pp">#include </span><span class="im">&lt;stan/model/model_header.hpp&gt;</span></span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="pp">#include </span><span class="im">&lt;ostream&gt;</span></span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a><span class="kw">template</span> <span class="op">&lt;</span><span class="kw">typename</span> EigVec<span class="op">,</span> stan<span class="op">::</span><span class="dt">require_eigen_vector_t</span><span class="op">&lt;</span>EigVec<span class="op">&gt;</span> <span class="op">*</span> <span class="op">=</span> <span class="kw">nullptr</span><span class="op">,</span></span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>          stan<span class="op">::</span>require_not_st_var<span class="op">&lt;</span>EigVec<span class="op">&gt;</span> <span class="op">*</span> <span class="op">=</span> <span class="kw">nullptr</span><span class="op">&gt;</span></span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a><span class="kw">inline</span> <span class="dt">double</span> my_dot_self<span class="op">(</span><span class="at">const</span> EigVec <span class="op">&amp;</span>x<span class="op">,</span> <span class="bu">std::</span>ostream <span class="op">*</span><span class="va">pstream__</span><span class="op">)</span></span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a><span class="op">{</span></span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>    <span class="kw">auto</span> x_ref <span class="op">=</span> stan<span class="op">::</span>math<span class="op">::</span>to_ref<span class="op">(</span>x<span class="op">);</span></span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">double</span> sum <span class="op">=</span> <span class="fl">0.0</span><span class="op">;</span></span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> <span class="op">(</span><span class="dt">int</span> i <span class="op">=</span> <span class="dv">0</span><span class="op">;</span> i <span class="op">&lt;</span> x<span class="op">.</span>size<span class="op">();</span> <span class="op">++</span>i<span class="op">)</span></span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>    <span class="op">{</span></span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>        sum <span class="op">+=</span> x_ref<span class="op">.</span>coeff<span class="op">(</span>i<span class="op">)</span> <span class="op">*</span> x_ref<span class="op">.</span>coeff<span class="op">(</span>i<span class="op">);</span></span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a>    <span class="op">}</span></span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> sum<span class="op">;</span></span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a><span class="op">}</span></span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a><span class="kw">template</span> <span class="op">&lt;</span><span class="kw">typename</span> EigVec<span class="op">,</span> stan<span class="op">::</span>require_eigen_vt<span class="op">&lt;</span>stan<span class="op">::</span>is_var<span class="op">,</span> EigVec<span class="op">&gt;</span> <span class="op">*</span> <span class="op">=</span> <span class="kw">nullptr</span><span class="op">&gt;</span></span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a><span class="kw">inline</span> stan<span class="op">::</span>math<span class="op">::</span>var my_dot_self<span class="op">(</span><span class="at">const</span> EigVec <span class="op">&amp;</span>v<span class="op">,</span> <span class="bu">std::</span>ostream <span class="op">*</span><span class="va">pstream__</span><span class="op">)</span></span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a><span class="op">{</span></span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>    <span class="co">// (1) put v into our memory arena</span></span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a>    stan<span class="op">::</span><span class="dt">arena_t</span><span class="op">&lt;</span>EigVec<span class="op">&gt;</span> arena_v<span class="op">(</span>v<span class="op">);</span></span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a>    <span class="co">// (2) calculate forward pass using</span></span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a>    <span class="co">// (3) the .val() method for matrices of var types</span></span>
+<span id="cb7-24"><a href="#cb7-24" aria-hidden="true" tabindex="-1"></a>    stan<span class="op">::</span>math<span class="op">::</span>var res <span class="op">=</span> my_dot_self<span class="op">(</span>arena_v<span class="op">.</span>val<span class="op">(),</span> <span class="va">pstream__</span><span class="op">);</span></span>
+<span id="cb7-25"><a href="#cb7-25" aria-hidden="true" tabindex="-1"></a>    <span class="co">// (4) Place a callback for the reverse pass on the callback stack.</span></span>
+<span id="cb7-26"><a href="#cb7-26" aria-hidden="true" tabindex="-1"></a>    stan<span class="op">::</span>math<span class="op">::</span>reverse_pass_callback<span class="op">(</span></span>
+<span id="cb7-27"><a href="#cb7-27" aria-hidden="true" tabindex="-1"></a>        <span class="op">[</span>res<span class="op">,</span> arena_v<span class="op">]()</span> <span class="at">mutable</span></span>
+<span id="cb7-28"><a href="#cb7-28" aria-hidden="true" tabindex="-1"></a>        <span class="op">{</span> arena_v<span class="op">.</span>adj<span class="op">()</span> <span class="op">+=</span> <span class="fl">2.0</span> <span class="op">*</span> res<span class="op">.</span>adj<span class="op">()</span> <span class="op">*</span> arena_v<span class="op">.</span>val<span class="op">();</span> <span class="op">});</span></span>
+<span id="cb7-29"><a href="#cb7-29" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> res<span class="op">;</span></span>
+<span id="cb7-30"><a href="#cb7-30" aria-hidden="true" tabindex="-1"></a><span class="op">}</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.</p>
+</section>
+<section id="special-functions-rngs-distributions-editing-target" class="level2">
+<h2 class="anchored" data-anchor-id="special-functions-rngs-distributions-editing-target">Special functions: RNGs, distributions, editing <code>target</code></h2>
+<p>Some functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.</p>
+<ul>
+<li>RNGs must end with <code>_rng</code>. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a <code>stan::rng_t</code> object (a type alias to <code>boost::rng::mixmax</code>).</li>
+<li>Functions which edit the <code>target</code> directly must end with <code>_lp</code> and will be passed a reference to <code>lp__</code> and a reference to a <code>stan::math::accumulator</code> object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter <code>propto__</code> which controls whether or not constant terms can be dropped.</li>
+<li>Probability distributions must end with <code>_lpdf</code> or <code>_lpmf</code> and will be passed a boolean template parameter <code>propto__</code> which controls whether or not constant terms can be dropped.</li>
+</ul>
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Details of programming in the Stan Math style are omitted from this section, it is presented only as an example<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/rdump_apdx.html" class="pagination-link" aria-label="RDump Format for CmdStan">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">RDump Format for CmdStan</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/external_code.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/generate_quantities_config.html b/docs/2_39/cmdstan-guide/generate_quantities_config.html
new file mode 100644
index 000000000..d2c731c4a
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/generate_quantities_config.html
@@ -0,0 +1,1159 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Generating Quantities of Interest from a Fitted Model</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/laplace_sample_config.html" rel="next">
+<link href="../cmdstan-guide/variational_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Generating Quantities of Interest from a Fitted Model">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/generate_quantities_config.html">Generating Quantities of Interest from a Fitted Model</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#gc-intro" id="toc-gc-intro" class="nav-link active" data-scroll-target="#gc-intro">Generating Quantities of Interest from a Fitted Model</a>
+  <ul class="collapse">
+  <li><a href="#example" id="toc-example" class="nav-link" data-scroll-target="#example">Example</a></li>
+  <li><a href="#errors" id="toc-errors" class="nav-link" data-scroll-target="#errors">Errors</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/generate_quantities_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/generate_quantities_config.html">Generating Quantities of Interest from a Fitted Model</a></li></ol></nav></header>
+
+
+
+
+<section id="gc-intro" class="level1">
+<h1>Generating Quantities of Interest from a Fitted Model</h1>
+<p>The <code>generate_quantities</code> method allows you to generate additional quantities of interest from a fitted model without re-running the sampler. Instead, you write a modified version of the original Stan program and add a generated quantities block or modify the existing one which specifies how to compute the new quantities of interest. Running the <code>generate_quantities</code> method on the new program together with sampler outputs (i.e., a set of draws) from the fitted model runs the generated quantities block of the new program using the the existing sample by plugging in the per-draw parameter estimates for the computations in the generated quantities block.</p>
+<p>This method requires sub-argument <code>fitted_params</code> which takes as its value an existing <a href="../cmdstan-guide/stan_csv_apdx.html">Stan CSV</a> file that contains a parameter values from an equivalent model, i.e., a model with the same parameters block, conditioned on the same data.</p>
+<p>The <a href="https://mc-stan.org/docs/reference-manual/blocks.html#generated-quantities">generated quantities block</a> computes <em>quantities of interest</em> (QOIs) based on the data, transformed data, parameters, and transformed parameters. It can be used to:</p>
+<ul>
+<li>generate simulated data for model testing by forward sampling</li>
+<li>generate predictions for new data</li>
+<li>calculate posterior event probabilities, including multiple comparisons, sign tests, etc.</li>
+<li>calculate posterior expectations</li>
+<li>transform parameters for reporting</li>
+<li>apply full Bayesian decision theory</li>
+<li>calculate log likelihoods, deviances, etc. for model comparison</li>
+</ul>
+<p>For an overview of the uses of this feature, see the Stan User’s Guide section on <a href="https://mc-stan.org/docs/stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction">Stand-alone generated quantities and ongoing prediction</a>.</p>
+<section id="example" class="level2">
+<h2 class="anchored" data-anchor-id="example">Example</h2>
+<p>To illustrate how this works we use the <code>generate_quantities</code> method to do posterior predictive checks using the estimate of <code>theta</code> given the example bernoulli model and data, following the <a href="https://mc-stan.org/docs/stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan">posterior predictive simulation</a> procedure in the Stan User’s Guide.</p>
+<p>We write a program <code>bernoulli_ppc.stan</code> which contains the following generated quantities block, with comments to explain the procedure:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> y_sim;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// use current estimate of theta to generate new sample</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>    y_sim[n] = bernoulli_rng(theta);</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// estimate theta_rep from new sample</span></span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta_rep = sum(y_sim) * <span class="fl">1.0</span> / N;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The rest of the program is the same as in <code>bernoulli.stan</code>.</p>
+<p>The <code>generate_method</code> requires the sub-argument <code>fitted_params</code> which takes as its value the name of a Stan CSV file. The per-draw parameter values from the <code>fitted_params</code> file will be used to run the generated quantities block.</p>
+<p>If we run the <code>bernoulli.stan</code> program for a single chain to generate a sample in file <code>bernoulli_fit.csv</code>:</p>
+<pre><code>&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv</code></pre>
+<p>Then we can run the <code>bernoulli_ppc.stan</code> to carry out the posterior predictive checks:</p>
+<pre><code>&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \
+                  data file=bernoulli.data.json \
+                  output file=bernoulli_ppc.csv</code></pre>
+<p>The output file <code>bernoulli_ppc.csv</code> contains only the values for the variables declared in the <code>generated quantities</code> block, i.e., <code>theta_rep</code> and the elements of <code>y_sim</code>:</p>
+<pre><code># model = bernoulli_ppc_model
+# method = generate_quantities
+#   generate_quantities
+#     fitted_params = bernoulli_fit.csv
+# id = 1 (Default)
+# data
+#   file = bernoulli.data.json
+# init = 2 (Default)
+# random
+#   seed = 2983956445 (Default)
+# output
+#   file = output.csv (Default)
+y_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep
+1,1,1,0,0,0,1,1,0,1,0.6
+1,1,0,1,0,0,1,0,1,0,0.5
+1,0,1,1,1,1,1,1,0,1,0.8
+0,1,0,1,0,1,0,1,0,0,0.4
+1,0,0,0,0,0,0,0,0,0,0.1
+0,0,0,0,0,1,1,1,0,0,0.3
+0,0,1,0,1,0,0,0,0,0,0.2
+1,0,1,0,1,1,0,1,1,0,0.6
+...</code></pre>
+<p>Given the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> y_sim;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// use current estimate of theta to generate new sample</span></span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>    y_sim[n] = bernoulli_rng(theta);</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta_cp = theta;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// estimate theta_rep from new sample</span></span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta_rep = sum(y_sim) * <span class="fl">1.0</span> / N;</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now the output is slightly more interpretable: <code>theta_cp</code> is the same as the <code>theta</code> used to generate the values <code>y_sim[1]</code> through <code>y_sim[1]</code>. Comparing columns <code>theta_cp</code> and <code>theta_rep</code> allows us to see how the uncertainty in our estimate of <code>theta</code> is carried forward into our predictions:</p>
+<pre><code>y_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep
+0,1,1,0,1,0,0,1,1,0,0.545679,0.5
+1,1,1,1,1,1,0,1,1,0,0.527164,0.8
+1,1,1,1,0,1,1,1,1,0,0.529116,0.8
+1,0,1,1,1,1,0,0,1,0,0.478844,0.6
+0,1,0,0,0,0,1,0,1,0,0.238793,0.3
+0,0,0,0,0,1,1,0,0,0,0.258294,0.2
+1,1,1,0,0,0,0,0,0,0,0.258465,0.3</code></pre>
+</section>
+<section id="errors" class="level2">
+<h2 class="anchored" data-anchor-id="errors">Errors</h2>
+<p>The <code>fitted_params</code> file must be a <a href="../cmdstan-guide/stan_csv_apdx.html">Stan CSV</a> file; attempts to use a regular CSV file will result an error message of the form:</p>
+<pre><code>Error reading fitted param names from sample csv file &lt;filename.csv&gt;</code></pre>
+<p>The <code>fitted_params</code> file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:</p>
+<pre><code>Error reading fitted param names from sample csv file &lt;filename.csv&gt;</code></pre>
+<p>The parameter values of the <code>fitted_params</code> are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in <code>bernoulli_fit.csv</code> so that the value of <code>theta</code> is outside the declared bounds <code>real&lt;lower=0, upper=1&gt;</code>, the program will return the following error message:</p>
+<pre><code>Exception: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \
+(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)</code></pre>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/variational_config.html" class="pagination-link" aria-label="Variational Inference using ADVI">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Variational Inference using ADVI</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/laplace_sample_config.html" class="pagination-link" aria-label="Laplace sampling">
+        <span class="nav-page-text">Laplace sampling</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/generate_quantities_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/img/logo_tm.png b/docs/2_39/cmdstan-guide/img/logo_tm.png
new file mode 100644
index 0000000000000000000000000000000000000000..48c9769c7de2c7c8585b52a5fdee686aaa47c8b5
GIT binary patch
literal 21065
zcmZtt1yr0r@F)zki%W59ad$6Xq`2$iwz#{yJH@TI7I$|RDJ|~qQrz7?e*b&ld%yRb
zclYdhCdo)9n<tsfOd^#OBtId2L4tyU`XntSrUC^8(ERU*0QXUYYqKl-@c=lhNQyvJ
zj}sn#6qHOgrOo8!q3Azg1SlW?6AJo2l#ee|xFr<K|G-dC)E`eMDCj)E|830!{y(h%
z%{=J;3;%~Wr3x+c!9>dvr0Jq5FUMzWZ_8|GVsB*1>|yKh9|F`j555o3*3`w2)Wg=s
z&Y90cfc$?j_&(tOs#(ZM{|CjzT7X<rUWruH-pQ1di<y;~m0S>sl$7+FlZhFhikQUz
zCjTf2kXyL8IPkHsxVyVEyK^wxJDIbv@$&Mru(GqTvon2QFgbhLxfpsd**R1EFDCyF
zA2CyBV<$@o7fX9P(*O7x8ri$L2#}NiC+Pot{#QR;EY1F}Bs=H-t=5NvEdRBzuraf;
z{2$&Qq~HFl<x_OBH2skLAAdo%Z~ue&|El}na=x+rC;b1fGyhA||Ec{@RS@YL%l{cR
zL8Q<6Tkud&!cfv;-$5RL(`<MTLh<^C3iBkVRK~x&`-D<(jIzH0ay&dqGyJU7;XB0c
z=l+m$k8iE;lK-IkG0p;ZP=4bF7n&?B?t%%-Wm8SH5>uQPSKp5~flJFNT%#xBC+hwi
zejWTXGkUMbOUoH&M^;cWVgU%~0CHiN2q+9v;QvP<4;%0c1OYK)T!i&n6SRDCks9A3
zLfu!YAY$*`l5297u1ONMcf2q`MRD1T;o>h!KilKYUbSSem2e#W-KMr5GOmf`!kP^#
z*{PJMX-Rc;JBaIQ>v((U-jl6dt0|rkBh`eFg1|+)gv?z_e1}}%M!WnP{FlCJ7tVHy
zioa?Y+NeM<;I>GLwDHN7E8a$750ZY<n^c(2sW36-NIh>geF>K|Fnr6$V>I2MmneEa
zDB8pG9&nYY^eI6i0NbYMwl(HdPfs)5r4x(Urh(TfV~n+U4(T<MF@6>dhefX{zot`a
z;i>&qGf{35+kP9T4(<+GdmvQ8Y|&w&;M&S`Ktfy`9U%XTKwfG9T9{3{x}LCVs1+Z}
zei@}kCed4fu$M1Gb`gSLzsTTk!zbcGH_vL8%73jdFv0RkAU@W0KJN512vcD$*CF`O
z)2c8xh<@)lG*nrQX?*aM%g?j$XSVYCHetkhdmd8$)o@}VWJ(CtO4SIg5raA}sdZ{E
zP;2r%czBTdGDz-2pZLIvy7n|=<QeUitdTFHK7R82nM(7cNu#KoGUfU|Zysd7BqNM1
zT~6oB;_ve%A*!#Yq-oUXM@rX&(Me$mP=@I%cP?EC_PN{-Qq`O4+KR`<T`Gd@V0k<K
zk|KtV=&&0Xvv3$MW!^InKjN|e-m45ARuKX`sg?D_nIv31zPx~(X<o|LQDkK@!7Abm
z_$GAeNS9H<L5PH+n4Z$_Jcm1)g?oR!ahd4vk3@!xEP+K}&NbgIWR_$rJ%$%^H3nU^
z)j#kFDa-sd;Ox0VSAO~wHMEEwYZU6Wel(_GGE)M^hws(l|0&9#x3)rJuPAs{E2y;e
zyM}ibN_1)?`AO6A0FcaLc#a|y5O%_2*`+Y?zCn?;PmG+$(+)C)IT$LbN4=EHauq+8
z^G3_?Ggt>MkXUU7OO-u|vfV5ojJU}m5t1YVYtTL~U;IhDYy>jhbYZ5$6oH|4@TM!?
z?=DjhX+~eRe4Bjq^=J-6=~2{}834K$n}i9&Zq)dvBWdqOZ$tIC3Zbw+py(F!@>5F+
z{B2h8bxfyNhS=r^P`QP}lKvI%0V=jt91!w7UB2k>%GvuIEJu&K4jL%cwNh$xe+)_o
zMCp164~Z*BZ#~c7$)L*FEa~S~KV`8L1-d=V{(#>}reEF3!aYmIvsO`)LG&pVY{iwn
z_f-<pLh@I>cXNLxl%}Mn@~d^ESPvs38@Vr_`?#IBm#uow3D^?#@<Eah5wH2bE;tz;
zW%TC4W;cjRckT;}<SDw!Ssgy%Z3q!BOVacc(6P-M6xvsb_sb*7PDw+`OLfFu%Db4x
z_wVK|s8XuHpacubsA0q+W8sX*&(p&wg<EI*7Fp_?UE)aUuzHt$!zqqU?Xzj_a~BK*
zd~DrcsmM2J0>y3w({m7yv5Zc+ztsi=O*}*@H;+g`hQp@sgXXJ5-_>+x4SK4#Wv!Qa
zne=Xn$n>@Xg@FnY^26ZVB7!1<vmVBhzoJ}xm8<}jS|n&UjhB&)Zktj*!R;DvtB3jw
zw<yHTSPY`1aA7BG9%}RYl*Ig<)V*jt_dlxiDcejRMC*x(WF-4puX}|-G;ZqsdNj)x
zF=}u{ppZ0}0sPZhMvt?5ynR6_<*v>UYEdqgT6W=^UlApObgf>Boc?vKh?re}6+I&Y
zFku{~M4lh~+Ts0Mc=oBNhD+ELsf3a4K*aU_-bXvO#<H2V=cm<ymyJg<Zk?b!5}1KF
zKOa8+R3Zw!+vgEyi(>*Id@pZAI!$}&$Mo~Prj6eq&CK<uBbbiHnc>KWxEI=vE0_Uh
z6fBM<4d|VsXky(g)e@f{2^!F^7dQvkL-W|*Ui5JMOEaizvGZibhvi8H!j{a;8YKO_
zFIR>B>bVuANg}EL=HDFrK|n)TI#acz_Nm{5F*S-D_??WEIa>|whok<}<X)q3wQjF8
zWLWHK3Jp1DRT*+_-J}#?{WD&Ljf@mL3!R%oR~4&ZuE9L<w=P^bq_#W}+a_S`@uXdT
z|GX+iov1Q=7Am@6Br?DX0UQ(DeEIs|+SB8){$&oK`1^M$$kUP~!+q7)R6=P_^AY9N
ztmf7`EONbHRzM?68x?9B!}hS~9PRod)w#ZM5RXc&pUWtgHTT52>X0DsG6JHXu0q<!
z`ijnK2tsk$4+^E&-2D90L3uiFl_CROS6n19=(dhyK=Zg{@1@&~;YZU`L~OlvfUp^1
zZ?YeX0TljMxB~y!_^CepXL+ByRh%ytVOC<4u*aqv-#YBp0=6Y~N)E_aXGdgd2{4od
z0EWe}^895WbJHEv7T+oklZ%`gvrZ;UR$ILt`TY-3LDL2cIIB<MpF5RizGKvM_KItP
z6!J!7aFEnA%OyylO$w%I`}bRpcN(09)I3#1nThEZ_1lHzOnh5diZ3dVcrsJl?toDW
zWf{WQXWZYkzFnEKr3c2ney*#JINpqTw--_G_vdIHAd`(W)I^}uM9Xd`_kS*}VPMQ9
zp}uE83puFGKy+zQDJf9lq-~O?HrDxR(!nn1?Dr$XGeqWNuG;;H{sHHn%)3!}4D)hc
zA_g<7*c;-0q<J%tlQOnh34p@SVYP^a=0{cV6W{J#tqE2ylb|Pu_12broj7_)Ye~H<
zR)^2${q)IK?eoWI7@+`opodh3N*FAPy%y~1!Y->rPF6L#Mu<PRNkzXbPS{uJq+N(s
z95%2pJcRr?@UkkNM?he3bqcFhQH(8tn~}NAzzO&NeSNU9*xO)f2o#}!jYiMWYw<j4
zp;^3}Hj6iZ*f{P;V8;YZe7qjcn)OP*0+LiwBzDjP6w2yA6Gjh?Vg7WVUD5_kcR`Pi
zj+%Oymq>S!w>o}xg|lsmZqA=7p0yi(NV8_UWM>4*xPGCZE216aDi+zE7I%bpa<goP
zWk(>M`XFg~d~bWi|28jOQE$?kQi|Y@AoYNZtzLd1nH_2eGM+p%AVf}@9j0n!XJc^p
zO>Z39Hw#hJcv`#(7g~MM-cojJ$bR-9Go|@3nGc-uF=f<Rl?LBB2|0Tc)jZp>9sM+3
z0y@*w=(77n#Iu6*#@kFhS^P)L_fC95msBprdd2geoj~~@$s#dge7KjT)|1s|)|~HR
zKfhuWgWtQ&<X@`i7X?}YTbqA62FnH^<de6HLZ7N;%1V`njKt_Mo>;b%E?7%tUT`wm
z`f=$P$$zcn(&}cql`=l{=6m%jGs}nx>jNXH;b!Vvt?7=iL6zJ5RNOH5mC^2E1{id9
zx)_dloBCRxH*zM47_FeP6!OGifHae)#bJlLpv?d;hQNNV=)a9T0={n6_Dg+Rx{C~p
z`$Ba1FRq*?bhg`<Ta$*fVUrjU+&L&2+bZR+0F{EixBWsw8`KR4JDqqQhws5oDIwjv
zwcVmQdTByuBS;U;oj3)eaT%18Ujon6%MnBbFwzRG`lk#CiX5A*ODI488#2$<sQ47=
zo5pT~bFwkNuehu~-d%37x9mE1*Cliq1@+%dALM%nJz1m1TjZvZ%<_}5n1SuYb9p(_
z$ry1?I7sY&c6Vj@e+SfadPu@)nOs7n$G&lhSAukRgpl+~`~FM`{rUP@T`gtKkwwD4
z+ll3umN$*uuK*EjsdE^;Gmh0$vkizwrNao!fe8T;1s^||qoMd&^yExWCXgT!Cf>Zf
zwEB6@ZHZA<VxMqWbE7@|${?b(4kVT6M=4n$^I0L!dKCw~FF^yw4<`D2Z<E#eomyt>
zMiEP1yZa;{%<o04*HS6h7j<S=xRXaUvUY9adDwN>x4MlWoF7LlEt?p6#P8EKlVc-)
z{#YWzqZsCc4c0C^hs|mD;s`BwrMD*OX5rouBlMKTJ3$ycri|lH;e(&fgix#W+o;y*
z_5OtYOK(~ZU~c2|fA6YwGF?UcmrMbF1SN{08ub;Zw^L?Rk<8&(kqp>9CB$ukqW>lD
zNR%4A^hX?*iCbz0=`M{y<A}YIj>_8#9G}V`XmbA}VA0HrPw**MU~_l0$DyhKI}Aqb
zB<RiAe%2`fU*7(RJGtcKAu-t+|E=DOOZp^={<rcV<=jsmX&kI|vNtba!~g|KwSC#D
zN|3xk_1r(}`JmTxw5ttT+E9Wfq@3Ne%m9y<+z@CX=kR!y(@gcnf1ACJV~EgK>$674
zoMm8=$J)*O7nQg4{xb1hN@shR0Toc){Cnqjo?lVNul?WfUF~TPbnyUiT#?t}j<45O
z9jKOjdd^69WlaPuYYuYv0Nd+=6rE9i-ocaiRS1exI27f~73e%qT)i6i?_m}Zx5GLy
z6Q_<U(p{(`Qi%<-vyUtRw$&l-HDkr1+ml>YX$CiTT18@Twz-nauV~>illbxJX}hKj
z<yyrmX1S;4b!EpabxYSZn+zYkxR?Fy9e>m_I=0vs5a&`j8SYc`!>=b^n-@&$h`oz)
zuQ3k{!c!L?@!`1&K{ZqAI<&bU!+@xiKxq#5Y<gvhb|v#B$`ZjJ=x38a1bNL`|ChYY
z?5g4yD<&wa&;Hb;zf}g)k6sK>>bLy_6>=klH(?fQFrOxTnJ%1?nZDm5-PNISZ`770
zcUjWJ0Zh+fe#)FXRcW)n><-f}pD0q-f<_NvMir>DGn2EgPY-EkZ}jH=TL(jp1qJLj
zR!q!V-NF+y5D~0h$xs1|_}_KPO51Y#q`5%)rdiT6SoO|Fv&+nL^o!;TRUEzswtULd
ze=G|{n$`tqH5L|+R>@EXFx6$mNYh~)R(=MBoIE|oFge0B=-VHwoVxU1;_j%cRIj@k
zJ6uPaR+t+j^hQepVt)T+_@Ds6Op1)!mdU+B&-Z)-vEQiI&@(Mc3a&c6|H9l)dN>!8
z7Kg!fh>JjWL;#b0f-Wu9kiw=6=&!S}^O;|@wJOZ#gN=Tx7HwAJjsC*dW>sa~<XSL&
z7128v9R{t-ml0G+JC-qg9|r~RmbLOJX|9<d=qOp!vBQ|4dR8S{_szAVxn?**yqGTv
z0t8im?Jhs<MsYv<d=V&T?u)zP*_E+ePLF)2SaD^*@(qr=3{i=4m>quys4ED!>7(mN
z6;z`{4%8;={oKLD=6%L*m7u|ibT{>JWU`~1-`6?O<zz1_P70zL3V=cB<qhoD!Y2`a
zXiy;sW>ECzEkTOt{%Y|STgB_)2o)|MBNrmY@mR^~1k68&Y*wQuX=wZ~u0~|6*A$%u
zfQm<97Goby+Ig)U)>H;H_qVi%B#U%SyN5by6$W55zhQ#^YG!m^0r5<yPym7*#pI{N
z*!N{h(S%OOcL`0IPA$n&k6dc)h~e$t$z=wbW!GWst#IMcDg`2b$h0$3^qd8Vn|f<o
zlh={buU7-99AvU&P2j;_q2boW_tI@}HSa^kfK}e$!1O-_@>Rh@*;Yu*fSdfdgp(7T
zaZi0(Zh4flAR?yXx1AYkh5XcDhx15zU`=N4COIKy9%3aaR0;j8FxYM+l4&8I^c^LI
zmCRev=O{^Dm5VT?yQIn>`TR+1393cWoYW2dV>2R;5f+$P8X7bouR#>=5H*7kTf_^J
z?Umc4MPJSp&nVcrR@umd_%M{10znS=Ibw$)R0KEJnA}%FzDgGl5N+X(s6-VQGc*_v
zM+`cR;s=$sWwL;0KZ6yY0<a-1CF&D8<Z7-W$t^pg#+G%)Nt23iwFtV^ar&>6hD@@e
zDJDQ36YN0Qy~dvNL)QfH*UR@7@QN*Bg;X2;%;2lT&)-<rAE0QO%-mN1YQIk54|mv(
zPv4usB>$N*spN<J(ZQ+x!)sGJj^Xng*<-?D-34MW5dk{Ds@Bu{Qn1}K5eT`tQsF5e
zxnd<2?Et*46!4xcI?PEL;o?Y%9^tM{^iDJy1}6^@tZ(1^_3G*)#Y%vO=*sllA?SX*
zpj0Daa$#2DGXkrdtm<CsA$#(!8U{qUb|g*q`RKv9Q8ZA)ciZ*XF*R_ND`no;^Ha=+
zL8WU>;rND9eStK5VWEV5auvBRV6ZNeP`}&Rkp{kDzyL&Jez<f5v3N>O#kNcZ4fcj$
zquxc*YV-&iI==lX>rUD+l9zEp*NostiSPg>e>i|HGJCfkPM#cP!oPx6gPPI{HR7Fa
zpQ#TVw|H)FHTixdTcfeM1Ry*}fI%aX$rU~w1NW;@Zz^d$x2ApfW452<afR>QWdbp!
zqTxK;1felVGym%W0E;$uBK0?WsxewCB^oU4m*`$An`eGH4YXa~Z;Wq<=Rc)~=Ok^C
zvk|gHdSf|GahRY?nFxn779)mh9yZwef0URcUPc??o?N96sB~A&Ei}g8(#yK*v7QtK
zr4+N$haiOT0%+)`kK6*L%|O2%_~v7fa5fmU0XE~g1aYz>j}O@1?<F+`4%_>MSH3Ev
zL4&+i+}ey;VQ?k~3>ZCXHpe272<RUc!Vy0p7P$gMtGd7+P9YD4%R>CZgbGg7?*Yo{
ztv2lT#?f9?_?pQ3^-RQynr7l)=ZYZ{20qE6C)D{xkAOdn3@PeK6A4)m?Id;V`KW0M
zv1swKA##jN79guZ#q~ASAW0QZQ(B-#eIUq|ts!l>GAr3yaovxo&8E>HZAx5BI0+`C
zuW~@rbAjhDF4fWzIv9>a@cuoy-T&<{&wymvhSc?lh-G`Q=cXb+pE+@j@o8_^S3=g-
z)H^AZuAtM~QdTqPOSHFzKn{a9cPJHuRdO@J<_<66P~FtDSy<dX%|_hLuBxr%#Y%#O
z(H*0&<aX2w?ud5BiHrkEg@;e3cw*P;CWVIN50kF)LK3N{UW5q8X^I~-L(*~im^is3
zQ+zl{%20aq77N~B*GFg9hR|cl;=wl(eFkeO(Ixr93WLA){Xg3^{l7>rWXT8*xci!7
z6v`th4vK7m<npjc&IjeXJUUsdOqh|IU}&-?Go^0nc)yH}&9y&NjqF`Vr_4>i^gm~w
zFRM!ot5x$Ksw@N4nUg`WKp3wU8S&8k2B}v|J7XF#d7{0Cnkp#&yL~VBg+JwBhLxiC
z63)`Iywa`oMKgZ)(i1d%D~m6G^Von!(`DFIN`@&>OV1Ra);qpiR+c#wq^yOrkNg6g
zKEn!iwr>W#6#5i4zctLRyr#tgT0QG+)(Y3d5>3j<HihAR*dUZX($UJ1?LIP>*N!fs
zQAHcKyrNt6k}M<7O0lsCV?kAiAwv{pSi1q|WheqibI}rsPPYT$vG5AQ56Mw#3X)Uo
zOyWJ$g%4V^(`OY~z#8#Ps>a4PMolg3O2!_eI>x(eE8pxTxccOhF}noCBi>*D@wX(U
zVacyfvPat9{!YbM;YMspaq5<8I=k$L6?+o8g8Y5>*^kqw`?EJGEj8Emnzg8tLNdcm
zEVJvmJub=jo^&)A2jryG*#nwowKAdN^nx0jtz}B}$(=Od!M%n{y6m`ZoSv&U`g>ke
z<yuK?;mg*jvq7|;LM#sa7%9XaH|6ME3Q`|QYu0$_h5uMYUxYSnCYz8uG$t?s#_a*W
zfluB@nvl24!k|!{XwM+I)2IidKVCl`bT+Y}N}f~N%`TLGMyV+!CsQcRKz7bv?^79a
z03Iyy+CcSM`Kk*CQs88h5Z0gHp|Bhc{%;Do<k7<SrAp#IN^0CN8G-D-$dP}cRN1!l
zlGLSt_&kL<=-g{`(v(~V79;$euouO{P`ik@vAP|^g!RaSu|XlzFd_UP-q;_hU)cy*
zaLzZqEDf8Irfp_H-TFfwRzbYaU70MER~pp;jEf<tJN-qvuj2kku9p%4on%G2ip2%5
z_Ie6WB9Ta@^xMf%t;IYEL6M9qocFbB5H0^COWxtuQ3GCRmD<*S>C`PRRQ}7Z(GpPU
zz@r~CNJTZ@X(sO}{3gW<kf14*79jqC;ty5eSFWFDdTnl#XgE_tgf}#VU_4+%C~OV0
z87UWMQIQ|8TG9u_VtTFR#sNCOVZ8)fA0{M-Fpc(y=cR@-+T#p+kL@q~h&Kv44gdK2
z&VQYdB3*A4<sppA2$mNa*#WB(k|CM6E&eJG70Gv6rqEgv<AUs-{+7&_vJrb&ou%K*
zzt|}MhjMzVfZ)|H{XmP9=rUS(!Q)W6N5#G2!4F3k&oWEcl>gqlhq=dZu18OcXL|%m
za_TR*?Mn%s(PhaE^g}MMpQ~sD)q>mYsl`tu*~Sj`AB}m`%V&^H86jqQ?#p5Vr@JTZ
z3Ikk5^E)9PUOHF<!;TUELyPT$*yyaGRe8b4pFT2Ty*j`tz6DlD)B3;j;r+Cr!SqNx
zHyq*fH!S@4Ua(i{`@&3$0OIC6iBbr`E>ZP9DRoYW4mK0n>EFoM2Z_9TXxvA)!N045
znTlwgnL<a;SPl_k@GlVGBTjb(^IC_2vY~^_dzb;IDT-1tw4L&j0<Qqkho1#vo)d9S
zPZD6U_OE+Q$NOh@d-S<Xnn|c<WP3-*M)#hBeL-!aYq^Dm0*Ddb3NYNfF;O7RUwtA1
zVH=?%*i_``D8RbJYr)EWGlQcmEeMxOedw;h=6{rU+liVl4F$V_0qIkCQ7-<cfAN1k
zA44ry60^@3lB2@_<;_y}tvLlhVT@|&m&VC7>G0Xa7@<ALwCng0O?%&pz#X!!p-MbR
zURw0@ob4(U!C!R2P)JNAsf27`(_42($ZyWdBI=xvUh?U-DdL_iC7mps&{FMvvFOR=
zfFjp!i$eDazfd-yoqgR-ATj7O68LFh3C2V8(!c{?eE8%ub~37<7&=o%X<)7EY6-fS
z(yu(J9i&6r%sA68*+|Uq^pKixgGF}=g{nfXeEcUr-(ordJK%~{B_OFSbw9tpebL;H
zlQLxG8+B(q@LJt4a91hIx#6W6`C7{#Uabxtn4vD%kH_0d9NhK1=GyOH`W1I(VgRpG
z;QDP}3uT;O_PSL)F2Yfz?#ofQRX`UHnJW}c;Q&3|$GJckF-F=gM$hs3C3wAY;+QOm
zhwca!br);r6G+ihg-v^Y30fd~?zdyUl58gLOTe>j4yb%>NU>=OE-?`SMub#Qo2CyB
zDXmH{X#uHuZrJ&m@dx{j%w|CXjVRV#hIA;+AjNyNRV+)u+JhC9I~~p)W+XNv@P}Dz
z?@D{8dhOgP;T$xTKO#cz0&IU5al&OwMnH}8wx|pQO>MGVHgKCOctoX`rvzYJ4D5>#
z_C$^e*&CtqxqXy6PYLSUx$PVE{egP5jIv4nlOIO7a$Uu^yz)HY-YpsOt0u0mT3`@t
zXARv&hg45J-n+m!`B{y=^^0MmoS|Z`4g70R%n}yYCaspik6RfC^&oHc9#6veze2!1
zyDp0&uSyMg37L(6iDwc&b);N>cl`1Z1_$H^)fKar^sR>${yeMm`*==ITod_pK;nAE
z%HC|y0RaYrD6k)^XrxZ-igppT!#}~KFTTP+J|hv4m1n9(*0iIvZVu%mKnb*NCJBOu
zj|&>xUK1`KTLqXJ*oKyqbcddlnTzx|stQjh6r@xcmy*h3Iyi)<Cq1be^m*UPLTLW{
z^~Dj}7r<&Y0a;Wo47a0tG(e>sH3vKFIw0~4LRIAcU0#AK&fOOs9G@Vw#?)@)@ez)i
zS4?;L`491{tISTPrC4sy%jJm1mxe^u(bh4CT+gmD*6AygAwXwiL?AY$7kn&dUvw^E
zND#~AR8K+uP+^bmi@akbD|mS+UeF83ifDd8ixG%G9&jn<A?sqk>fXY1c2KqCt6hty
zAZ2i4`jbNHD4$!o$@9!e7q_C(smg}6|JoX@cezhM$+Hs^ZzY>6uN;a`R)>S|(M2HH
z|DE?H*A3Ss<|j-{ABa{sBML+TKuP*LE+3<PLDUQ1^>?8p2sYeT(Du&)fyl(Z(I+TR
z(QmnEyK>IqIB$2ChMG;LLdm#9;@L;<JN~3j2;w?dhj68<ube1ZYy{I`J4sltMDRh$
z)pH1dGjO!hY%jJ=iC%`pCH%{=<JSF`xqE|j#n76^dP058E-bgxZz1o#M_Nf%c6I>;
zL2-FNrseM(6T!P=2V7%PGf%T^$y%x~%QgXZT3(92Z!ejnKAPv^lUnXD1GT;46sy^n
zubOHNIK&1`X!ABt#9yqfd{{^$G+e&ULm=`b!g4ZvDCvHvMW_T=1IuOf<mlgku0)lJ
zYl|I`5MG@N-R`b_g<$J$w-sLK&o5^-1~q(~@9p*sTYpnF`q4U2^O~%LZi^(k2hH!o
z134zb2Ts0!Nk|`{1g>K_Xv@*@`?*&SsR6sRYg>BMd8|n2G|nyEN9eXuACgX56E-%G
zWef+)?Hz2I7IVo#uJYDuuqOu3s0iZx7W0Hsf2KVe3Dl;j3`OKgsUYmIrnEPdk~K(N
zY*Tdz8Q$DZ0yu<!hv2sgf)Ee8F)U6S`>*^@1Mh_%;g-L!4fv;1<y+kQ-f&T(eF?-p
zwTY1LpTkRz2>6X?GT6|z2=k6zxkjBWi>2GW0f=@G6}5j)U3M_MK7X&nQun)o)#O|D
zt&ua<_?n6%x^qbp|IGpaCf5sBySddLm5Mx14Y<l%!9C^O1a?dvt(ZgVQbV`gFPP%A
zKHKcpSu=C&&D~f-JFi?v&1M?&H9%M3om6b8JBc`^3PXKT<V8+|ANfjwuaH&BR8BK#
zXI|)yRj&cNV&k~&oYm5-X+I}6-j)SHF;e>LC1ZrxTX;?s($m!d%a=(ZSJxg$9)^l)
z=RvZ~Uavfr6By-&U5!%cJ^qQFuSf(qW_IApTD!pSN#=#Vc}LQSIvpPC>jwxX<AFP7
zwq4ReyBF&EJS-dA&M!X{A3!IF0X-}i>StWiDngBkO`c~4Ohm61$Ln(Sj9tpkB4%3T
z`0(Ni56=;?PYewVr1I}+FWCg@<Se2Z_`m7>QP=fM)z&j-tQxC6&`as)cA;Auy`1UA
zPs?K_lm0?}K>=8J`YZOV^y1Y^gdK-;H|qC`j-O!wnu*FDQTO0C=kd8tEF?GQmGbJh
zR#)ghvfKzsoj{r9t8jaVx;~48@F=c8>Va>Q1^xMy@t={g>ijmaZLp171(-fUW&F&a
zxxapEc%PtPuF6K$y!wOt@Pv#4j}fxTus&SY<g<Dh(7ydzou)}djcxn%$1#s1y>T8p
zky{RN;ESHDnMS*HjwJH@;jRdS>Q!QIP(BJuHQ;l=xvch9xlm=;6BdrXzWh=r+2oWG
zvSf>WNIgr7$Z|v-X&9Q33P)#K@)bH@5aB@h>Ogu^8KRx#(`tknss+I(!2B1-vsCIX
z$P{cKbO+EJ%Ql>sx&1n1*dTlxs8^ka&-JGlkqzTIpt(_5kwiR^(rsLTS@EV6%wHu;
zys5uVWp2VD)lU}G<L7}-{JbK$2~tjGV9Cr|vSCu&Bh{A_T7jVxmH58=2QhxupH=*7
z+9u9Q(IJ}X$un`eFvzx{&g4t*ZOG~DPPBxB$9BKy#o{#y(0#h~w#+5ydd#p%<aDd6
zTEj&J>8z0z(EtmAPeecnC{2x^nzS<}Ueo<Xuh{@yK%Sb8_K76Ece8^;MybtDY<Ldd
zPR8ui#>%11i|E~qpYqcexL1dk(4z1CmP4>q`H~Jxw(7(`u$Ec{b8c&^ztJzcRm@0L
zv_5rPhY=8Ly$jUAr7(m+3X14~1&Uu}qjgz~U9zm-Yf&1l3KLb-sAv!iZGmnyQVk?4
z360Qsl8YiONN!sV$8xQ|?}m6CZMpTO59OuKtK7MFa2M_H9BH*UQ%)%Wbqv)P6*w#y
ze`Rfia=1iVz4M8_M<T#T3ES64`=1gc%~xSIEy`O0=J7Dn7DyNZuP7wR3#zte|J=3H
zF9;fKblrYah_!`tR(%K!M=5!m7SVAT-QuDS)IoN0;Uj6TQAZv_s)}hUUPg-`(hkb6
z(vyo(JAFkhzw<?tUJ@mB?)yquLG&GBCbGhmQ7J$-yjOlZc-YsWR?FtKj}F!H`}A-@
z^2S|<=CM6&@KXT8X$F04yFU4%+`o`Re0W3lDqQ5~@lOgA+j5C30tu`1nnW{(|A?`S
zAPFGRCj0`Uuar!YN1p%~Hvbm!bn>=4CUlj%Lh3nx|MqsK+lvGPC4M)oa&A=pJOV4k
z#WtKFMa~){E)nB7h|wq3TDGiNIm$=)zK=&M={~SUn6`f`d^EsB{Yrv^pR3xnt7W3+
zHX5YWIU17*xd@bn_QekG7#y?7eMwO6e2UCq+U5OaS3^=eafRo|P;&$LY5H-`td8R~
z?OG^QZIjK|dCMWnE&$vp&wgX!4JRXFgst6V$z8abeJ|wp{JXWhsU}y_mpu|dcyNav
z`@wq9UW^gM^RJfJx;?D`iu_QVmnW?03}0aN>p<cYS}1^HBJjZvyIJ(u_e<%^$J$0?
zvZ_-{C<XD7&0@p<>5V8xlXP6$@~4j~_u|{)^OlaqR5u#YE$7xWqg6NwOqRmGyRwa{
zAXliuJtjfn`OE_F>UM>t%{aW-R<fFnc(?c5v4h8TIN0v2E}bSTL&o+}EFf@Y3aSNZ
z{%@O(?MaypA8K`}{0>X^mqfUTQ%_oPLI7c56$>vyhkAEZ#Q@j#fVuYoS;Pfg$C~VN
zRnT7Vs#&Z<fG?MUjWN5~7Eblp)6=Nj=f1=M1XfjFFwM!+0#fPXBNY$PU9Alb=Me~)
z#dgl5>B5^w7Nq*<`T9ee58rJah~Hp=h;O{A!04riYdXses5=%}TF^1v?_lXz?X)o9
zI=w5_NOaB(LvZ*PrXmcJ2#*>pt?`1C7~Z;M#g&(MXvBn3+uYQSOMmiuz=ZzYj5d@8
z@Kt8*NUlW`1`}Gx!wA8y0%eMg)$DDLmHu3Xb&v|K#i?9g4Mw|y9z1cxtvA&n7XMjh
zOFV$>U$g33ECG0C%?y==5@IcXbio$S?vl)4l@w9WLQFy@P7J8aj*nQMbJ#^k;d4eU
zipxOvMDJ@H#N4s_q0)j|8JiH4@+=5Wgwet>?@eZ~hmg#RAw|+~gFmIvq^d_hz`+YX
za;NM3sgf1d+gAn=f?TaR+tg}X(*jpUiDgT$^s&$reoE}>BB<7A|GVM>H#&waob)jn
zt3E0(zOGwW=dojVDy(!dc;b^gRDd0OQ^OFj&+IYWK^_EhT~AypjYtjZ&X-2f_w_Z+
zee0a|Q{=Fm?jVwSw(+$u#dbK13Qkkhs=iPu$1Z6F*-mp+)e%p8D5RovR|t*jX~v{K
z7xJy%6;KA{>e8Fs;Pb{^tdrZ-Y9AJl)CsJsrnD2e>B&Ba<v0W6I^GoUzgnZhdd{ot
z|D~+xHw!IX;qD7`Ww4}5my-+DjL|8a4Gwah=IbH7le;g!`AvHubSszcHpY(6>&`Y%
zZ0;r`pGy%CfS<W`{boI>>&XbzCLosIO?)o#Yq8Ugkv+hW2egPT-5QB&F8{f#PIFMI
zS?@tUin=e1PFtZ#P>PaqWf1ONo<e;V^22*l6oiHw|CZaSfDL9ePe7${CQulw3L2c$
z8QQDN8b*o<h{$819)QdK3GEtw-!|VqgG(LAHa9TGgR4__ENf9A0HAPUaf8L44q3&Z
zrR0N4?yzoNAq&E72+omTokF@~>zdllK*)R<F@S`~)&!%}U6&>FNrbfOVApiFZR4N3
zZjv;zq;D6!*5u7mx6S?Ja~G%HUn}jeia2w0kBu!04xnQOsKCXAR7!J@adARtx$$di
zPg!PD=gm!hXP4dTmB&AJFsMyW{u-emlhWow&#VN*<@c_4>CeO|N!xLY5I%C451g_+
z37V-XIVYngJ-mtkI3^#PT3Ux?h<ZEX8?scPBRTfaDtF=g3pE}><NKIZTdU<^x+?vA
z!vL%6@1M!sYhzRNk9z*<4OFjuzCM#FFfJ#bRCfj*#h!#eQNTl%*URabq|$p4P2{r>
zMaMIcg#*ul;dG*i5)6vN@28#rW}>f3h4BP6z_u-pS6)Kd>GRz?@!yg8>bUa?aF7<}
zm-3r%<2#8FB0h14XLmrQREDFt@X*4rx*@Z4vS@_+y9vM@tfFEdlo`dWuG)S0hrS<@
z!!kff3`G7xEm&V@wJ~oDw-<C(XFJf@<OHn-E4h4_3Ou{xh7WDX(tqIsTT18)Q1Ckj
ze>mYh3eYd|K-*Ws;+~f&@l^`|Q5LSLKH+bBMRH#J7>X5U(ry|vm2|Mi3cvdzhABE0
zxK>@<)IXskCc`CmFWN}F{7vmt8S)!nlS=>VpP5h>-4z(h7*Epe*{HNNBR8$a+d=rr
zJ|^h88f>al^itjj=#PEKaPsrMwd4V7A~udqQ^`LMP^|<Jf0X|ZbFcRJ*l;vJ-Kod#
z1L^9tHLI}S(?*SUm!T`l$oDKVM>I-x$H=Tyl!hkFe~<9EMk&7@0$huqca<UAy^ntP
zX_He9)TBma5mL|FRLR!P;$_zSM`4BGmOrb9`QHv=<<V2pVIr4;)tD_kB8Re1@=9<$
zLAAEH)Z4#2+hb!Tp#ah~;kytDU6%beHw%&IW>g$otk|{bX6~a7z`&%!)IUzv!>i2R
z2xMhi*l^*7vB>sm->XtE8Ylmq)b9IMjy4Q2&M+4&-q(n<bad<A5sF6m`S~}jZGL43
zeziaO^E{4eSkL+@ET<WyUR759-g%V|7Y2<DTUC};$!#?*_F}dF5nw)a5z=ZFHJ|7=
z3}Lt6@j_3r#h}x{S;ZxH`+f__<{{=+gPjiPL7?Uk`d60H)lO@0rQ)Y!3_sx2o*s)8
zcm^|2T}ICCPeH%@R2a_6B@!3PwAL_V4vUCiP+)kKB0JSVstwUDJ-`>x<X%lOzY^;!
zgk@lsiVUI$)%FYM$qhvcy94_sxX7&Yjydtu9Nh|NT%ltZxQH1<2BY=;8*yaZb))>6
zfAVh)oYhRaTE`aLig|50O3Y6Z1ncA2WDyUX;L|OxBTVzg!4xFT<o9^U^UmP-NO+J(
zu&vBnq40@amk_u8buF;kQ1~Np*Y_zdksZcJzO%F?(Oa!HZ^vL;>M^tnM|eLXfE#8Y
z$9M@p+n{qmOiooRk^x=bc~Dtf{NE>Gp(_N%xb|(|cwWDD5{((Eb#$=8!><;fF8ak~
z+%>aRzJFWmNO{g;y&Ax<()3&(+f)4gM#QwE`9fEQ)Du1jL>8(?ga{(mkL19i>zl+Q
ze$BVTp?3dsHHdmPxZ`vC%+!poyUC4~0?;TetPx{Jm!$+6k1IJX@i{FfVC6_5-~kIO
zi|(&tUO%MgylPgz3t=EMtqU}%vH_txkhRA?%sPPynR$`EuMuUgfw88sbva?#b&Rh0
zHK+I1cf7}ULt2{$|Gfr!_2U>@Z~{ZYQga?q*EK6z5)p9n@ej63Lx;EP2e2uM%dJrB
zE|RYHjE=Q(qEgi$YOZK^j>~@TqKU^|yZ=|~`|<V)44Hysuq*FZ=}J~%+6NnP>?HeM
zE4p{l;n0O5a@Wpg$y#7Cm?sPi9c`)jSY;*zsyC0yYZPHeC||vxhB)=`CiyWCI00L4
zoZxnkF}Kr3Z38EC0Zde3bJZH=$b4LWliTPN_vBp~a9w|Fx6G+3#BXNuQTZzeka;>^
z3<e}gN=-%Nt}Xv@SRX$%TPmoxt<=b_%*UjqXQKkCg}ZjqWDw=v%=HS=w21*UI+Hha
zX8;F^!jT9T!NQWhKSm~+QLA*Eop-gH$!ni(j?c+ifIsb5EQng{u(5QBo1+fPs}+&e
zP2^(D;H~%`KI`F4@nb0Q2d2-Wpw|R8j6W~cGp&<2_9MJX{Giw??;Xp)L=JyesV>m2
z!LHQbUj&jS1gYf}mP+j*x{x&3)>H(w7rgR(cOyM%;V_pAs^gdlP}swe)cqt8nXhod
z?^z5U$#gm2aSGnJ)g#;^1s+kOXQm=<b3Tl=tu=FOgd5`aD}IvC;DzNYw0KdA<gfoV
z_87>$=T!_YWIuAxtVkrn<u5#xvM*I3zQsED_e4eRAGz1w#1ldq6cJ72-jpLZ#pVDB
zayASfI(=wd@0<Kvlh(Vm1b!GMK<v!e#e`kExDP6rHWTd26rn@In4||_kQQ%QBrr%t
zEyl?Ww9XlMJp_(`J8Zu;YB$k#wVRH|><wwGb4^qbwRF{k#hsH1J-=IW#0ROlZzQ<@
zFeg9D3BeZBxi+@pa?BDw0YWr_)TepP<wNSQe)qQ5G(PU<xs9{Ig(<^z$tJolL6!5n
z)_7&6U10<|kb)y3gDWFBz*ji$04(ON^(NTpW#bAnv*P}hcUM3q%SRM_R$y`Q@7tm6
zl0F&+_^X%O)CLlt4x_N4XJ@hOV_$4Q!~+#De1Ui!TE6wNFhgW)aB!$E7Ka-kM$>zG
zEDfP|y#7oTQ@T_Xz0&?|=tTZeU#GTjpMe7R&O?=Uj|{jaBRqYmmya{V+>U9c>z-yU
zi)*ZV5IlDH2~P8|N29mb53x3|Qe)7r=SI_QQfH4o<iZPOK1C~Wy&r0o13M`Pi5mEY
zoHPu5z>Q}q$*iq%`RJ8D|55C-b`zwD_=f~V=j)$6hu;r2wLDPGd4}@RpMjX>6+U!J
zG_c;SQLn*=TP%<+Umr`<Jt`oCrk4^?B|OftORV>EzQrJV(AxH#s);aAQc@BE|L#sT
zDzZmE-E1tJ0}a49ecg`YIv+9+mtK#9ylI}}u;9(kH(occ59T$9k=7&f*r>bQE~BM%
z^r$o9il~SX`rPcMu=;8KR;)c=8psdn+j|=j9Y>R<p1+yOZu{GveE5og5?B0dsE^u!
zKCe@Z6ch0PMdF4p9jIE}2`jOu2Xs9AS#N|d$ODaJ(7A=Z*WQCG)ix~iINOIQU;8>s
z;p=~G86k|lVAm`VEZTxi(rn+H9E`BJT8l4;f9%zi#e@3cZwZmm<;0b0|LRWu?R^s%
z!Ik}Vt*3)6Rpw%A0FzPq($syG<+r|122}1fWx;KncfV?#K`uQ1xV&*ApV#yYlfL_z
z5XaVW<7dZSL@XDif5d6?-=Jy-k-%deu_)a~`Vl<8eEbcVVhHfU(r^g*(({%hj%zT?
zf6FpC!tgxE+cN83AKf*ou=J4<a3Db{NI~$Mu@jiVFp~F@shD3}-npW6Pco4sh^hYV
zD@pzIYDY*F1NGifPS!WcDRI}Km7Xj4wy2C(mD;)N7wIe#3Rnik4P`nE8B#WHbV_}v
zY=&OB!+lt=-61c3N>&ES#A+HmUJgee1yv>%*f?@%#zh1Uc#7*WpKJ8;h1{X)$g3`|
zY7m9(4J5OG4j^}+2h@qkc@Tn5UI+_4m&o|2aZ{Oj35eUh=+jZU;OiJ@VLhU@QL~Uh
zzwM{I?3nB=Vm4ZQBIg#2UhBbn747;Ai}ObRkbSG-udo2*lSu)ZWLzKqW>)Zt2fm#E
z@ZEWhG9r%cX7u16g1lW@Ao)&y41TlijzBkq23e3l5-y94<Avnq5A4c3ssHgx^qK!{
z6NReIOoy?Yn(1ObB|SUUTovZf(TBdKFwR`+*#ao1O!(F6U3K|euixF^uTS4*uSb3)
zTg3HN+63$~L$%uDitZihd^YKN`aQC+ik~sLs#&tKdF-n9_}UB>2WUCkxZ1+6E~?>o
zeROzJAaJTRb7V{vUry@&E#r(g)Oz0B)TGPJocw-&7GMA3ZjIN?f8>)CM7`x&NT%0r
zKw9p<OFkabc|u;=dO`^hUg%fNiuJ14I7dUkW0xCrLGMro>e`iX)a^ucp3aYjt_=U8
zO4u-0ele1dZaRpP68YNs@y0@4@}c;8;P(}984JwTZZJxEneBK+TxmJs6e{&o!=ZWJ
zN&*-)m8<mkzd-pd`(qr)kfX1|use?J3{re%Bm{sFl<M3P=y^T@)zP6U=gKi%=oU)=
zhDFgIYz*5sM2eaxZ{CzJ%Lwf`2&D<VQiR5O?5U8?uZ~>2o92VLI<yjcLiCuh(+|fJ
zpxswWH|A1X7eNw>GQfMT-(~MoHZ&I4IWK7P_9M)M5YrFN@&A?vQN6~`;uO~3qsRC4
z4ifc5rqiH0PaE}UW_oYA&Va6>03)j-Os<-}e}aA$Il2<=2(DhB2N>q(EagI2hJXko
zYJVI5)g~6JE(HnucI-?z#mApQ_MUq7(+I7@@;|))rs#kw^oYShb?0#QRu@X)9@2p{
z{`rBe>K|Wu@^)Vnum_KX+I{WXHS4}u2z-R;MG%q_gkltE1czi%8UZc>7znU|7<4Gi
zebN=zZ~%O7G#t_j59>)zhE0R6O3Y9LLuCmMZPY#bKYz?1QBbV^nKm)w6K6YRp$0-q
z(d01!7f|R}>d0Xn>C-F^KkS<wT0NNpaP%5!=|oAj;nnS2i)^08Ufg)C;>Ut;z6k9#
z1!=IM-@iH`NPP6X3BA`g5j^D(fE)&(%-C&^_Oab)Hwqh0Ingj^drTCf3VYo#xF0Um
z74N#D(_}nRYP#Gi#UBxPKqD9kr#84y)+gU20$eWo<7!uU@^tfku2SP&tL#kLoYf`&
zcic|#Uu2#9oI7|RoAdO-n^!jF_59)A0#^<1v>2py?}U<`!2<@4YP)C-{I6aeOAb=e
zQj>8N2mJQRNw?U|1MB1`46O9uPwf;BOm=)~s{gRiF3P!Hv%UV-EuJMi`rNye9r9zu
zCj@w$<Z^^`RL<p>g6C2*MxoA$NcYN3dJtqAex<x?pYfTrfI|aXI<;y~9R!3F?F`Gl
zk&}LeN|C{QXGPiE*kCiRq>nr>-7D7=3S4Z)4Q7<IYKog`n7~n<ZtJ~iPXYHNlv?&B
zCQY)q!6RL2w&o3K|8{^InjSL^z)(J>e=?x2$%;#&9%(Nmh5PiW5oKq<4|&q^nwonL
zF+{jTG;z%f=Ze+{5*1|cCGcL6D3MMjs7emZr(|XPyy*if-~VJpO0Hcd5J?u;bN)|B
zv)oO(U$on=uJ*la<?1)X)E99^O)?8(f*#naMUp>JY6UB$qumGy3-tM*>o6!3DW%ca
z16h>Z@#u}=d1FHTB}DA7NgAg>(`N0H5g-HPdM;|+bD=x&eeF^`{a>3GpD4^nZb$Nw
z4N5gRc7>*8aWQBW>Y~q`FzO;(<pROvcl{cJ{YZ8J=tq=Rt;i;I$BqPCyTDiM3edfB
zn^x<a!3hc}cEx;h`GEji<ZaC>O7qS<%S|KR>a*4hEVmB0EU}6HjXTb-PG^ec&WDQN
zyQw5D0HarX^d&c1MomuJk`1vY;2sTV^SgHmwFFlu&-dr$#ABvE%b2Q5j+8R8y36;>
zPj;*ad9(kN>_p_B_2iijqI+v8;mpx?9mb*H@clI#q%|wg&*0ICfhKif)dW+rXc?n-
zIw`5bj>zg?!G}Uu8YGp$vMF{bDI|h01JkpY_^nSUF?C+l%vw0=73xX&<SSZ(TNE)n
zPrf*G>YO*mZiF`e?w0GGG9v}8unb=uTJ_zXV<4V};XjOTKGx1&0RUF@JKph|`4*Qm
zxreJt^0SXNFMWajDbKhe4k|3T(nN-yjMNAf)O2i*5XSSp4lFKe=VLpPB$=6oZOMS&
zRi4kMQaB~^s+_MNwnxA4D<EG?DRQUcIGWdn9$---4$MD6YH>9ml4V%$ay_%PQylhT
zrNym`f7JZ^TodBbT0AInX=HjPDvv|+HWj+_UddSZnKXRsKmn)G-xF5)=V_m|0Iy|J
zPil}6Lj+%5fg^m>T^RKWGyrG2fAD)(ON}4X`|l`fkAC<^T5EjtgSP`AX!G}(Nvlt3
zlrI|GdByFY)+#NeE1tYL1ya}Qj+$5Tf~hQ(a4IckIcu;9Y3=ungXV<GX}-e1ZFb{w
zTuk{(;3}vWS8%6C3AvPiWbqPs0kD&uN-b_nVA`p%VJ_lc7vV>o!iSrPVz_M)meIeb
zv7%O7<9S7%7CMCKiS_O1X_oD9ObX%P9Lf2-EL>IgbPz+T;=%CrbBa?c0%8?at7Iy-
zY=+Wq%f;5gWxBBOc{;TP(>m*;{`U?7x^#l7)6@XR9Tdo7nJNe1mnnN%zdtMYJ2G68
z`tgK}s{v7)25fMfit8qc-`jAp=XEmuhHBeK#%Cxg|5)~Wf@xGY0!^pGEe}-{I=Y>z
zQA3YqDX3(Vfr_5;eIjZbGoiA(eDgvJwioT7!K%b{J?aWoV(Y_};5BU4!-{EFsG%_F
ziLX$%X)mjn_j+{+Jm~O3Ku9ZdJU9`;VFDSJMJ>I@udyN-VT`Ev2!}qw(yQyDi!xMV
zn-E16G;9~~zjwKuZHOqlOZl@bbY{HEdm@h_i1(K%d?d0g9wz(Iv=rl}TE#$o1*=GK
zV&6AFKp(xoTTt|B?4YLg!m~Za=s;hR)l`1AMdKR1=@65Myo#$~xV7yA3v(@+s*58L
z%$s|Uoj4?%lrp7KZ3L-PG|IYiEBi6&(5G9(Re$MJz4O<EkZMF+Ko*w6SC=DZq!cHd
zQ<O4T2AI~j+o0!I&a09EbD3c*^tjhTRmIr+sJ~?x4~ry6Ol2unu0-Q6{?WKc3l?Q(
z)xWD^l>Vwj5dyU?UlRN`_|EV1Z5;o6%$t{{nt$i~>%bimL~`9@++k-a@_x>@)!PFh
zK>zbkk*C!HLQ9OhbJljHkDYmEM<_Uv+Ur6iwz*waAu^0z)T+PsK9v#iaHE{w&&n2e
zD99qX8b+h9RD{@dgU?rXfispq`H1pHUjA#OHR#e{x#dFeANtn2n;M7DR;B-^@Yvk>
zw<M=q4AVuE1=-D$#wC_F)`4L)D7CSgvm=~GG*BbNp#N7|@FIF0|Ixt5wT$f;AqAz|
z{>X9%`7vBnZz~1**>hGHFw>9XFl&wHwQ4`d>D&z+zI#PKJkp)ea^>0rd5tcW3tLl$
zTxR3vzWP$y#_Pet?hEnk7$h~Gmx^)<(;_<vDZMwF$FRbU^qSh3p5gl+t`rQ1bet68
zPePU2Vcs@uzoJ&e)Z%uVa0P|d|BPr?$w6l_z@LxLZVk-hXQg8fyC4Z(#eF1C`leff
zF!p*%1N#+xb~k>+y}=*zEIQbvU}79Z3&#Zhv>t7SwNkDY(Vu&jtMN3QDTq=(#8P{C
zy_qXS%E5!EgoM%)KMwL(xuic8B8BYXvx>*$pZ%76TRV`sz2+$o3Mf0Mr3k1Ct<m@O
zUhlB7N<YG`3~=7DhOHs8-R7=9OrEYs@CRXxz3-}k;6596Mrc-MVa7ch`Sx7MSDh+l
zFsG25!8{E%psjas-a9olsj7*>=sEChAK8z&6z!@t_4QU8Z{0D!C=VL<Xom4ICGD_-
za8ut(jhK2Oezv0xDc{c(s0_MCU6I(iJ6o&-Cv`SajmI8~GK4-at64&jl<|5xJ>%#4
zqWDkIxzL96&Q$dPs(p^1%N_m+1tGZ#X_q`ZzJTPErYL%zNl1u3Wd$e?5s`i_`M2Tw
zmeVVn)B`g|uQ@eap}+1l7U{OXqD>TsbEevdr&gJc2~in%Q{75kW-ZE9`lW|kUaL+A
zRBOGh_gYE9*z3G~<!AQz|J8Ax(Qr6X8&`uBMDHac#Ol$@#_B@!)uToiB^HZBThVI>
z5+$tOgJ311too1MiC&_Y)oXUWoA>kkex7^g%$#%YoSA3t^E0ldsjxr+Tuj@u?hMt)
zi5w&9!Sc_$RoFBApX$fym#L(6dNEfpiO{1A-uvOUqnF&Me^w4gv;J68B=~&1U!9KJ
ztnhKkuT-lrt?iI)*`0!a1zOrUWsbT9p|P`y?CEYYLl-d6A)|D`9%TL3u@T_JP0<Sv
z?}5}m7C>r2NsEY&<MYH#HAlmfE3{|!@x&ZRa$LV1uI3h)?4g^WZCgy?1kM(1AcGbA
zf==}MstrmUk81&PLKm5xxx6d+dp!s`m7Cae4DP!yi?o{sG2|9aO?<h5fad*3<E5(t
z2W-T%@=473wG2vikQ}d+C^Z|pF?g=<traEp(L*{3!F$hNi(A1n%4>Cd*4xZNfEfs!
z<-v_N|5eAU@Da=vHqdfP&^hxp&PSuX!fHL#UROoG^q?~M;|qKLP?nn_U5#R(pmDWV
zB6k40SFa0(;y&fndP%Sg!v5@O+``vCI;xHW%3Pof#oFDbKs|oLW`amJEmYiv4W@sy
z6ln1~*ykHlR0P$;8)?XD8j!c5`vfAyi+F)6xlcok>;D?};39P;c{J3cPjMHwi~j!N
z@7j%%_lEe*@EQ49ry)&8@r3w7_5}a|KXs`HW|v~Ou}G~a6Iw6n{fK4`84-RK34n7=
zT2H;IUAsGQP-0CYgn(JBb^|l-b2>@q+>za;zoAvrMg-%TyyKqud_ph$J!i>u>2Pb}
z&GnA&*>jNW7eIcaN6DVNo<(6#nSP`raG|(+dYmt<bDh^}>YujvH2kBBf7zm&d6DD>
zP}FrCP7famM2HHjIV--LL5*U4^He25>)8BwlZuB`wT<7OPbkw=oDf-*tcl&l1Pu1U
z5c;mM<qSc|4j<FN_H>nd-X~(ju#MHh@QYL@`YLcRDxOumK*@Armh9MSyQlEZA>rBM
zs!4Oql%mo`R#+z)pV$s0iq-%mG*Od*6YpXK379Tt?Kg*G^OoK!eK4`CaU~YWx49VX
zB^#@|IS{W8NUb%DWKFktRUFq`;<#-&Cwa`no2T_=`N>>@UZo;uswF=GFZ=9ys)RFU
z5vpiZk!1U^GLTq+>o`~-vHH2I`_Tm47JOPlS;xVddG&+iIG*LRJGa&ole%D>Q5;O9
zKr5=bz+4)k=6?7EIR+K>$RQ&q7SJ~}sY#4o7h^qI|Bg}@aXLs|+i%s+@08S1IOOa?
z)5R*zf|Sl(bHmIuQ<w%quUE$dD;t&=9)`|7F&q07^(dS6+(3{sTT<nnDK}ZD;jznd
z@r#}<A4r+KAnv^6BEZtxHYb}{(6u;D9YAeP+v)tQm`wmC`5xthF9;78;yC-_(cwI>
z<ZCsVLcI)EF366>KaoIR^%rk5x=a4_%TAz4i$<p|Dz3&w1S2tESL@N0{HXPSg?dtT
zpW^XCfy>~erQ643QX*aO6GVvEK2jia1N-=Be9~Ci3}_bKTrJn0H-VBOjE*gRgtpKH
z`ip=BO~9)yOe)2{|4uv~gK_5z;;&XFwPoBj`j{^j_D|Dgjdz)#0698+O?=w-_M&bm
zKW(r``?U}NC05E!xF75gy?j+8_P5O#xdCFW5$v=;#p$-3qz`3IBD+XT_$FMgGQ3;Q
z%d9@Uz>DBN?q>2|kJg6XF%PqKTBG8ITVD5+oXQW5Z+g#x#d5r~+$;lVN3~ZHlHu;t
zf&;g(0Hnm=LPOAA)qu@{!qmP)9@2uV0{PB23C9iDAZ7AhiJ6cUv4+kN?|rOP13lJw
zPdp!aTfcRxA?thOVT7C|w%Xdx#l<?3#mo}OZ4O$6UJ_?nV|U4mV+?Bv1`8Mg;yS5N
z=U~1*Apk!${=e4SWA*7-&!)WAbJR6Sgh8IgZcKA*o*!%zBAHcO=Q`#M7Rm+tJP7yZ
zO@YgIG-VeXd?<OTmZzQrr%zl83e`1IY4jWq3+TWT`)6v=j{{rWe`m4CEFZWRA<M+^
z?8LB}f;wlZ=kC}EmFEqcQk2bASToNVj%uR~2a^`{c34(c`Y(_UWNPehqKhC$ls}FW
zJKgmfW|t-A%}0?$RDbT3xW|UEkZEqO$0=E2Pce?)^~wy|Z2MIK7aDVnpccEQGdoBG
ztEs)HeAq5G!Kx~k1(nnNIKQ1J(U=XOMtchUK>c{XPul4$e!`<?h`7&R{>4A7<p_?a
z>qI}(babltb{noGQm>(!5tH-hWb#|yol{8SvqrnM4w(RRA_UG)6%z&{dTKwrTd>=E
zjC*QtsJZ=ta3}`1`1JHvW(nSe`pz}4mPVX~<D4x<q}y-aW)h|OqN9;p&qDjjZq~TL
zfTm2=)syxOs;&oOdDFglhkDxoTrh`3YYC$(c(*?jVM14Jw{LF#isQ3zUwu|f)~h`H
zPoTVVoPs%}qf2g0FF9%gC!dkv`=WsRwL4L1J_qb{=C!S-*E{=;BAYhzb`hpx+>FHW
z*Rg=$7zqMi(Civd+v{_y(@v*l!N<2LbX8Qok(sD4RZO3;$GZ}$OaojhGZ3W~<d>lc
z$o&`cOlcS>0LqVWfC2pL8&m0Jn<@NcOBXvh=h-$*HT45>5<{Kgt)j^ik|oRfZiM8u
z3cm=ync|Mk*?7qZUS5wXk7z^)gHk3n{uHgI1d5Ct_OfZG=Gvy*+Ke8cnf5ZozrSxl
z_(^IitNKN}b8;+KX9$A|F4~x8RNQ+_bvo^FBZZ3Q_nT!k2h`2TVX1(#V!C3LjYL(&
zK^VUxdl8i?xRA)Oa8=Vv{^d^$>L-T91Mb$%ODyopR4}l7WmhWuG+l}{KE2Ki7)MGJ
z8TP#IwV3)9T5@WhGw~0pnNKG!lut*{eb^}SIU!B1SElr-nyxT(6~acXAA?KT+p2QH
znuQavXQZ;RszY6oE&c9=njT|Lsq|5(q&b%*0$66o{XjHa^$~>7S^8_MJS?~)(f}Vs
zI?7{8#dE@D2D3RV<(xWX`ouAWj*$enC=?s=tAaTQm`p}fS%yNEElOFz{l*+6hguKG
zr`hR0CtDp|_0BdpEar_B?{UNvo;_vM>!HT~iffpjrlYkqcK^0ehxww`Xl<j(jKL+)
zw8(3$mBiNXPJGNOw!12qa9Qe>=%~=`Z6GsgKCI*N+$m(S^m3$-LeX($M~Aw|wiUqx
z{Sr9hkwsmaY2DSaMC&M=759h$MC#DpflwD%u<+Jfy5Qj32`j*na)5Ej%g)m;zZK+V
zErNdhNYq1v_;Mc;w9$g&8;d4}(c6x<3{EWaUPQ1?I=%=cww&k$|Gt{xwd#$w8Fka<
zN4WK|r*m<(9R2<aT`ySqP3zI1|M0)<uKFK5w2--jztVe$xtBPC!EWE5A!jyg)G46>
z83dt$@E&1F`pRQloAxFo?OY#4^`)kCeTkZjc#pj_|BuCAq|lbePL2o!*>#l<(M*3&
zkKbSxFIT;fpt#(uIV`pt6qQ30E41IY5t-*%%v<f~rM0@&DTJy+M?r}I=}air@^%%@
z*D^fTg2rv~X86UifX|mDA9c5c)4Cs&GRr%vZ{al5M_}**8*|<MX^{9|r-9b*d2j!_
zn6T?A3xpA@hi=>#{ve~#(;pR=M#*RUTE@9>(Q}TxXV3C}Ary^*>OqG$e^Yr0UzOwC
ztkB}$9N)YbX#PAB?4~lEZ-&X+9rUKXcOylN^smqvB2pF2=zcOgJ9`*lyBBdv74%3T
z37O<8cP-qYn>kOlq3Fq!ek&*HPirl(*5b4xKxspfBg;5X<$28}w1_;q4U}Tl?alZj
zz3!~_j3z(w-ZKJnr;CG25t-wQ@t=+ixxzy79N14T_KB#fyj?@ia3(@`?byinr-Tkw
zZ6L1?$bRF7?$wo<70X!6=zhDu*XIL$Ea|()`Nr81eMU@~hxA(>>Jy7=En^+R50?3_
zb;IbY(1Y-ho@sgL+)}F~dbP1}e!@5PG_}FyiVaC$s!5`Rjva;k$Fz=Fn}w;j`hNdz
zt%W^m`bk25*k5^R&T?<!C`)fF%x^1d=)k{=F<>bnxsj}<{_-6+b`-Xn0&h-A<s@=3
z{=hk^?EmXzV^}#XxfYwe`PatIshy|Ku#!DPGV7s$!&8HvIO-V5dXvJXYkz3zp$J-N
zh@YZ?8<MjP8w%=X6HOYvCErRR9rS;O^^k`4Vn?fMKuirTFOlf8F%le`CpfIQq{{M6
zycDo;{ViYvWF*sKQ*}x;#jZd?U!$bslPcIoJ*vI%fu`%3*Z%uED;s2R+D|(|b0MIi
z*mRIqrv^fa>savq%~QNS2iP8a6L3CRRyE|FgoGbJl(Q`bevY4E@ojTC`x~KSwun7P
zcQ2E0@|0|*<fm)Fmb9Hop!lzIaeOhmbdEk4GHdnwX43ZMa=VxR-iWg!rh%5nIXVAG
z1{)+f-5HugB)>hn6G=XURvSt{LD;whwBD8ck;h?n+Qf|x=(9t_ZB!3#q62cz|7?x9
zj%}*Fk=x2`%oUyA<r*;2Vdv{LHhM9<MoIE^%6_c%0&NN1rOQn2a&}3qdC*pqp%NLb
z`*p;}7D^pN`LSS1?nO=)B0_ug2RmT!`u9%YP793aKb--2-K=3r%toV_c}DX<K>-_5
zB75~kJbQt}B3B}SGx8yUk?4w6W#rvA)b7<KBbA^Fsi#qfcGQ@B6T-O><u7{8hS`p?
zDJ1yhB@aoNe;jQGSsHL@Y%=?f5xF;t?pb$#9ewJxC#Q#Q`&a0!B12o)jsEl>w2rPu
z^|U$KqjKgiXNEQ8Wzr;;Mbo9Y_~T~s-KTCUqyj1>d0ppW$f0ePrTeknCDhPpdB4=7
z`&ZUm4$|3FTjZRxMyVl#3*W|!_b3#e{64$juk~`D=V#uIpTSjsP$E!4d^p7St&eh!
zhn-===?`A<oGGWC#_f-Pc^^2xFEJy$t7%NbkW23L;T~(ow^pB&8RH%j@XfS_XH*>l
z1IVNC{Gakc3NJezu=#=rWo+QaU+dr8vU>+qo~SSa8jUD=QsKrbB4wKQR<Xs>zehS;
zE*bN`IZ#GH*c@SaIvL-09*HGox%f=0?-Yw<&*i#T<uBKtV>=eh7a+wQ2%Fveg{+zO
zDCm(Sy-kGompv|TS4Lo@<aAfI>AIK}J`Ij2#?|57{%-iB^NB`*N054sC!#wSIw-|7
z<>u(R&B|<WNtvX567Z-pUP^{){@_ThRn0<#x%g`jcZzpANm($KHl~a-_U_Ym{~-~>
zvzFmWPw@nC=AvKaWA7;M0Y1;!MLevY<}}6g8>vE#j0G)GJAX{t4MV#%)q5P6T(q_&
z9{l^Tu0n05{P&Ot43K(%)K%?<2(V=D#ctn93;ycnRb{?*NDz>l{AJDnTp45_*VB(!
ztyFQwJ^{JMjoaOa^u~Kgxl&<e_q%?z(PWRkJevZ=TXa@?m4OfsaT7``JgPTY&2}HR
zG(3>{0(`DO-z+XPMugD`Vk&zjNWPw2A{N_&C!A6%_t50)x*R0vUILz@vNXxBH3#0G
zQJ;MW6V|(!-g8z3v6~%GZi?=DZ&~-U>+<b&*YC0E`WH<68nBczi1l$6A_>l2>ev<X
zH^MMd<t#FPU<W7i{$D)4|DV+Nzaym7;dfNZPC%~4;k{8Dml5x|nyzY%vMv0709UFk
Au>b%7

literal 0
HcmV?d00001

diff --git a/docs/2_39/cmdstan-guide/img/warmup-epochs.png b/docs/2_39/cmdstan-guide/img/warmup-epochs.png
new file mode 100644
index 0000000000000000000000000000000000000000..7ccf22e0546ba69d66eb041549e34ad0e15e159d
GIT binary patch
literal 16881
zcmeHvc{r5q+rKsYQWRww9-$KIQNq|tl7!?bWSvwJ8rxv3P1&MovlK(iQz|i)ond6k
zB%0~5jctZ8YV3otufJ>R`+a^#@f^QD-sATk$NPT$!7=Xpp6kA@^E%JZd7hv1yj(wF
zWhNrHO^}0wL&W^(;gcL3yaWyo&a3<zz!l}!VNCFcs;7y`2`>{f6Sz0r-_{rAe*WM&
z_}TL(%?|C=)z#tP*m2w4&F$pT-8(+QHo3Wd{H(T9&_DQOeEf}*Zc*>6YZZtJH3}aT
zGSi)$dZfjB@EkmQwrvitsIBIItT}YGq2?YzsPyE?#Ji|Thyu|{W5+I-n_C0N+mw0>
zLE~Y01;_0>Wg1*K&Ps90@t=cpzG+e%mT=oIJ}vk9!7T+3g$3?gpVN3QXrYFCE#hAD
z8hbylR8Z(ozmt9MCC5Z%?4zeoV^pq3D)1jWsdmh`Fd*{ad7BHNN3LI0+Vkhh&HAi6
zQ-P8<goV|mrHyH#HcPtVvROCIdWfIpzapNY^sc%GcMpa7Dt4&h&omSaCCbgc@<(*~
z14Zs@R}8uFRl0RgP0ycjm%e{vt2F;OCxmp&p=XQx3Xq(D?@@;U4h{_^_77*sQ==db
z4r31U!w3He;T)j|+>k%N#5b117eiIIdvt=|Ovv=`uAIYHB{_9@-^X+E-Zs&Acf{0E
z_!y6b>C>wpH`?rbtngT2gUvxM&az!4Cz3vS?(}%`R5ElpDtoSiJ{;6ZyxgrjPxQFc
zi1f&s(KW=iJI{{pVO~ihs<-irD;Qtl;N;@@`GfH_o(<X;^HPtkagY7<CQe$Z5Nh@J
z*jKixD#*f}>eSQMz8y`^U-h~lPhQNt;*0fDE^+&HHa~8QCsl0td6xfv!9zc;n(^9{
zuB!rTuEtnrblaM%<UrL5@DpV0{DoB35&{QUq}M+P;D65|QSZ)g)SvCS2E#2)t9j|r
zB^Ac9R~A*VOVgd^B$HTS0e3=@y4$@fL|NcK<Jc!P%%AXtfP{d8DUCfI#LUU}{eL~d
zpYh5|sVX*kh~vUk&0n&Ad(M1fdZW^jXc|6lrbLaYWPHW=5f|&*XCa|2yN}k|yWiYj
z)8U12?hnyZ{g>eP9W-wJg4-1}Q_(GACXYO~LLGKim)M~$>xvEWvxiqs)qHtmf`8zZ
z-~N`j|G7TOq2Pk@HAG-w^lc@x*y|FisRw5Lk0g*=7Ux(^1u{N`Hqu%bj?Z^k1*2w#
zlI(Gu+#90D>4Z2$t3ys0H->-G#En)HL!$(DTT-{;NCb?D3{c9!9b*kzr0qt0bW+W1
zB35s);<y~T;rhSS?28nqgNbY5q5e;AdquK~2YHVP5!~NFLi-<$RWc(qk7sD-&8K7>
z%;vSyp+)<)pHGXid2%FXr|q+7894S0DWY&*B!EG*9ChoCn4#cDO1kBrS!c@|PCgPV
z3QaE_fcI3RU}?@l0>Ko7ozB8=0Uj17*Q)JRaBGl9#qxn<Z_)mMZw<bS<AgXZl72yX
zo%DS$F*R`=`VFl+QFigTkDV7Xd}{JasO48%tv3E!;rlk;+TUk0@0#IwxlDPui0q<B
zy?7zlrCN)Si$QH&(_XF^*8zLF^G?KUZRKwf;$Q#rPm)^~*eYONSCU`7*t(LeD+xgC
zb;DRU3=j{kyV##R=(=I75o@h0$-2i_H;i?|_-+O3V6YAb|C7N$I6qP_cSSf=!@0c6
zR9JeC!h0jju8Gn<Pg}bRYyhb<KYz%Zf&BWyGD~~J`<xF?g5*cmR7#HJ?SSS3-wDGV
z#}t^PQtU{=gPce}^qr7G>354cwmj4D(oXK>?b5AEsKqT$VesWc`7#?swiPyH!|<^&
zFKl!2d--!2oh<gN$Sn+X7&Qpt>;WaIS^G>BdU_~%DV_M!I17W(_escnQuUoeG`8VF
zM%8}xv1`_!vbe&5D(CN;i)?$Tjc_PH-cwSLbv0!uCF%TzU7jDJU02OIS~H$MS~V7<
z8I5Fpi@`08z@6i^O$E!R>H7>Oqy~>7J6jOw5q*q)QC-tB^vzc{jc;=ytEwgnclsyI
z|DYaC1^Sk@rj<C`u$nd5Ow@0;7z?o6j^^{9zu6=KTGW&%p=u?Z0k1-KIAYzyZ|-kB
z>Usd7nrxA!jhL#3&R|Cd%ZWi*c3B;Lo=wCftw$anMd`b`B0D5?%_1b&WDx2x&&2YL
z`p*34$iXzANQjLeSxc`^`3%0jz}j)zSciJ3d#`)#pOrV|pxPApyw&&F52pW@XU`^e
zV~VfwtuR@Z1q<r3XN!h+dEKUvG*rU$E3$oT$xFlG1q%j@41+WZvz<6J!TWaV<EV$n
z9fK?isrZfc72_42x30GBOWyEyI`}QK@SUr6Lr8vHcB&tpak7|p3c`6nF<|GujB>Rz
zQSD2^<pqR-g$SR!K5(1?Z!~?x!7lf=H7Q=yW`2Hx-4(jzcNP0LY4&$$T9^Z7Z!mP}
z)Gz!vkRSWXYc5XOte|h^Ka4#sDV|i~;hdN7ug8Cx!D$7cNt{WxYmfaU3y^TruX0cQ
zP38SPEB1$KL7r&3TD9LljNM@ib~Fo7wCmSo>&H~=QCED!d3F9TP59H;#$kf|gwBHe
z)23_qUNC@~xOftVgRpb$u`gP4(n5rW%0tF`3rmJ0Ltj6FIwTH9rYimS?YH`?p{}xk
zCMq?Aelrj^07W`Y6u!uHr<O5%?8t~>7Ivap#X2KM`>i&YsbJsS{HLa8A9V*azRN~Z
z`PPhw6E(>6vn>wn%=1L!EM{EZCEK+98D{J{!<KeprmmL$(qxk-{6KejK*$UwxQT2@
zsiWV?v2<RVAR53c%JU1*jFcR0WJh7a?V^o4PAZVdc3Ik{M4RsKoBePH3Ts54C|0|a
zHLnbkL1&(ca3k+i8Rf6q^%FJ&;XY<!BByfqgiiLqo@#ZBduoyPtP&lyJO_IpzY&&f
zFnu(6zvmC*-~V4Xels^;U3?*PNA_Z$s>!1Jkj5?Yc|_4r{rqY2{vgc7g~~w-kxuy3
z>BA30IbE-{kgJ8+p7^02s%lYovRk61rS2mSf3w*`&F%ws+Fspub}G%USfjG=)0KIt
zYiqt9ut^Q>e^8(GDiAQQ^-|Y*J#W2e`JY{x{f{U=uP<`FTVlN>WWDcYJ>psKU|Mg}
z`Va3#`yUhxoXD*1{g8<7gyWpMIMfFnbf~`_rToOA{GIw5^J6Nf&$I<LS>oG0veCoj
zmcu$O<WK2RYjRGfZBceENaak>ss_7jsV$I$4mvYS#x0+!NZ}`V*=>?olWZ~$<KJ!n
zTEQuV76a>jf1MVul*F%fq2c-GnrM2_&Y0_K&g`K9Z|ZJ`P5I@6FJIG>ps|(H9cqh<
z3?s4e(TQCpoN{Q#trDd()G#~|%9It=?Yp%nY<l?j*nLM5c`tl^kmB_E+6ECpr6V`^
z9pROTM6>>|;BngWc!{NwS&}fj+wZQbwdj*0HwVMaM34l~Ylm*Wyu3J92Ony62^)!h
z)%AO#WJnK+ujUFTcTG)<5I$N!qGooSo+xX@rb+rsszIZh#J7g~5*x(iWn)F<wiUNu
z$cUWzdh>6_Y(-PStG_d4>{|vrmH>&;_XbbEmZr$e&J4RJt%+OI1y^PXM*g2eX2)o-
z;kGxhhV;F#w%0K8TJdfj&&EHu8<Sehjbu``By|UON3fX{{*5Xd9u&7&KV&d%&Ptz-
zQ-@;88(xTNRDM?U+Du{%TGqX(5-iQ1-}TIZlk1egY~-mFtlxLim;)8GWsGURlzSNq
z+e;m8&v(RHEY2~qnXHLk%d+t!sk0uR?(L*K%eXuk0Es~2d#d)iq!d3fOBM|qEn~b4
zq*ZPfdS8)OzHAp~Pnzw_cv&)=n7Y8k8KG-WJNkSjj{o7_j-HM34fFx6eCH<|%ks&~
zDqmbtaga!zGB5qADEiF<qB56DVNn$mhJag<+%?t}KvE7)A{^|Eif^W^TvZEv_jkkP
zJp~{+i2vf1PoNvIBns{#LfyQQ{8>c1?IB4^?!aZIwy2dQJSU&5_%t<)X^6vN;5$Z$
z<5gD2!`9|)?3Z{ZV(Hv`qQcOTLfcKT5U{ndg^?1?DRaK36y2)%lrJ}a$mv(Lf7rrh
zn`0$n28jr4gX5NJLWlCmgQEd6s<I@@r;5Km{ezO6G*d9NUL3=ViMh@<rqzf-hreWN
zoLQEqh18oPj;>8x+PHz8r?t5!i!?p66w2@^cE~#ctYcSP-I~_nHYF9GY@s{VY>VHS
zlewwY!2^9p<0GGc>)YKbYZ9ZV7F)OR-Vd<99-G4YzdR~(0o-xx*S)%H%WKCWH@A)G
z!MlIM#Lpj^9gKBc7=!8TzSLEPFR;14DadL@pUPS59;l6Y016hJ>wftkZpVX}iJwMq
zTE%y>1!0T^>>MILscdcH*;o;fpoi=c|Iiwa*dm@pF4OyVWbHkO0d%_%tYzogt7>*|
z^F^zAtTF8o%{}h5H}*u#Pj1!-WN%|{qNWQ>Mm0>Tlol&f(_-g}&g_a@7{M2MHig3b
zcc5a~o!zH>>jAy`i(oC?06>UUj1)Vj8-xXw_T90l$<z&K+-Zf@+aKPi1QB4{P-p|d
zAbDip?Xg7H#ov%kpoLdamMN3n>a!c<YxLeT+kRbv8$vvN5A4uv7sD{^I0EPa4OpLL
zjvRhMy`<~+2udfBjb^pU5J;@>*2d0(VuRLZM2XKq^;dylvj+M$Z2xtp;`Uj_*hKw`
zsc;J7GEgd7-(73bk}iLX^!;9^{mFA1WW2v1%6(QwN#&qtm^}7XGfH>weSM^VXNSEs
zs!C9}zrSmKUS(trtCLh5Ysy~UCmS~W#*=K3rXj+;LHMX^@CX*L&1{a@SyM(L=>US_
z`DjG^;-l*7Ez?ic*3iS;?R$KCR(fAbC(LZQfobHR9tU-%6~T~K2^Oml2e=h&EM-<L
z)D(~mWwvNIwfPXE7U>Xt`d;@&cdxvn(>gs@xEsj^!C8^>1HOP%j{~r%id~$jr!6s~
zqT2ija`@2-raUO^HRmD2z3Z1|y1Pg0!#5gj-xqo7Vpnf?<XpY0N<^)g<WjE`E`2g3
zb9g+NARo573wYi-J(uwD53g&td2J`1>IR+FH3#|Rnuj!b=F3O*14j4eUDl*;yU>un
zQ9gX^-59`?ICf+CNO7meoJ3x%7=)S^K~BWdYa^o;I=f?4v%`NEmeJmQ{QmFCvWR9I
z%B8lJ9Grlpa?hJxrgh)TYd>4wLlT#rlb@;1w{-VAGao8y>F%fAIxhAQ)iBuK->dSH
zvW<)kHsd>nof+eEbZh<b!=Y7P40JG6fQ?jA&Ve><9@!B!v^a=7aM*VykpNd!32eW8
zS<y%MEpBCr87JiuqPWP)UKt+TB@3XJCKfeoQVJGOqiQ&QeXP+!^C?54YPQ$6_qysc
z47&y0m#E9wPS*$cX3l}D*QkgV3-94ptV&vXUo=|SWJLpC0kl6ZVxC&sL`Gn|i4;8&
zHa=jM7+fVm)Ej!$cX5#y|BeqLd)8haKODxaJm8HTuT?Z${)UGv`+h^O;40qwd62T?
z3|aR{+i|k}bKfivQ1&?meBNKDxid72^}u)|fRf+-m@@yxL_r}>6O|G}cScV$+B-=`
zE8#kmUrg>oAG>oz&3#SF3HH=<!RiQ<g0QM%5AfpR*Cwcj!jrd5L<Pvs!Op+mlupV9
zN>nE|xOdS7m<B!<Q}rNoXloL|w`_>p#C>cj$7pHi4VeS^Ub?jTA&eMgso(4_SwiLb
z(a|<}XplFMevZ-5s4heL{Y>$Po_RWCC8QKYNO#7T2Xir^adJWS9s8CK^@3{_6f5BG
ze3$1KIkBGjmC;Gr8MfVXO<kONVI8q+j<*_%WYv%&rwhiaxy80KJAik?D<$i{ZJ;a|
zEKENzR^wTo5zje2`AlpL1KYn@=j%P&J8t2V{eRV=>FtdBV)FVr$FmI$1VZU@jPcO{
zYO#88+iwQ4+)Va66sKh%$rIp^#`Z_>4mVw>(d;|1a$foy2P_n7*r<GY@B0;?80o_$
zy^wZY%yXO;QAbIj);)togkl``$S0d(LhKfRzejhKK^W2y<pQia^m5^K7*hBH-QBM@
zjLwItb=BxXGSINxyy9c!NN^kh507MJd_@fCpi)8E(DBm?tyK#_=?QlE*Q?h(vFPpU
zRXPDex$=g6bo!y*0t|f`2v-NW1xY&6&UipDyYZfo{N6182E@qFbG~Y(62ox3Lmt=t
z#y-<A;nC?+GE!t2-f)bZ8WC<mezUxcMzbd7<C9~r2PY7G(1=z6RdnD<j3r&KYQ6>Y
zj8Qd_f{6|}5C(S)>YDAE=$AB^0jDw$Uh-ZWvx}%j2#P6ns-v4HVUu{0lya`?>X~=&
zpW~+FLkq()oEC3=E>#CXc9$<+q<-IZNLG8%hsa(Esa@UB8X8+M?(hwK@>e!2&|kj3
zQ0e3BP6((M$_xwZ{*-i{fyEYkWFaqv4g%Z_v<EiO1fXsjsLG@jGlZ}hBjwUoU_PM-
z9c_ML+eE0HY}N7Fl>P295LR{1?;G}R_|@G2pnb{HtCjE3kbYYf!t8OOYM=XfYqI%C
zvf+c#nPLeDsWHGYY^2C!u_;?!Muor#ODp=o$n%Kkjq}a-?qmmd4fnO~UI1b5%%=yK
z$?!$vmi)XFoOJ(SQ#Vhjgm}`-vtH>{Z#C}Z*PrXV3#+@MJP_K8rm!=cdA$l2Gin7I
zp&LBf6^M&R#|TNOkff4fzFpbCQ*Fe*DkSNQJ~mBk49f6+YhI4SqP;WnZ-pa$Sh6I;
z6$9%qAO6Bapk6I#8nHSOYd)<I`O!V70+`>SY0>fFvd;&)_1_!$4TUa;OfSA1#f{g>
z&=S=0`N7#4hmhenggc|PH)W($j;D+FmhCFeq6LLd&wncR)>)oxvSwFOV^Q1&?~vy&
z6Ui1J#KMH8i>zK{BJVd+c6T1ZLZfW^+?Tq`d=?>oUsLR2!6Y=;C!-W&>zi_{#m2XO
z1?kVxBe<TcLNyGdmAOq98YAM<TqAW2Ubmj{nwp0_^}Mz+dIPjn7)*jBDjF1wwaz6!
zRUHw3{szQgwIBBGU#)Vdq}sqqo^r${{J`515P0mB4QSX~N4Ge;Cz50^6CFYbk^64+
z=p*4+3+%Os^9O^c$F)Fs<lwmw$N}ZKw7q8q@!V$b@tB_@A0XR@Q(Ww%tli@bfpFzs
z<p*eOMb^;L6UfqY?RO!8ZI^QM%Kh|L7FiURnd#m#MwW%xOn6jijVS5?IP>G>qR<46
za{V%>{b>c6AkDDih1;`jch~@LFgt@=F`#I=GeBz$zjN>z0y1js8f19pb%aAO^`JH|
zpFRG49}AI-98D=c?b<ctDW1Bnb|46HR&#gx*>3_Zj};0G4BTz!_IW?qX|#-Ll^Htt
zVt`h!iPN9ssGBr?2c7ATiqbo@_~p~A6EnppQA%^P$lR4o@lvgn)gOM1Z+tS|-mliN
zWw*U}Bsg96)Kb)7@=G9YGgR$mnA$v<52XR#oED8+9VP9Oiht)7<Va<`;{!3rh%VWn
zJ8eX1l<it*227;2zYcXumV08^yWQ{Yg4&=L$h!vylr2)Df!8#MsKlZ{H_%k*OUzIb
z`W)jvg(NyQ0*`_&xqTVJ;0&>ojTY@qlxc8aZg3B<o4z(+_;|%Q*~XD@a^kRk`%*ru
z9aq38SZLRt7AQOe%;S>H(&~_W(orE&`drY^oL<=F?1*V|S1JgU1|s3mYkXp&;||?>
zosv51RI_FCfv==KP$5u(=O6p1l)O8@5SmSPjf%L>s+4OB>h!^b%)7~!%{BlIL8_7L
zr@CR99a!K&n0X7>Y4M+qc%l^8!f<$&F2O7}e8kJuA~qp7kDv=$y7nQ_ON2ztD(R~6
zoAB*%kH3P%H`{>XV?-)-hKGz4k;{WjT*#9mLHv{K&J=4M_-akMh^_`i!saZn!DDH-
z#ix!rcFu@X=~^CE^RU$RyNt{LD`{<Sd;g0ag`TwhiqT_Sl#mi698z8eqP*?FKJc=y
zspe6Tg*HM>)xxNZM&)Fq?@(Sz>`;D%hLi6*UfO#JYvH}Sms?A!QbeVgeUShxYJ5s4
zsq%@WO*r5J?~av~D3s;WZYi47SVvCW%};{?_CayOH+E{e=_mw*hOiP!kPu)R!e{5n
zxU9mksTRAtW$&S--M}SIR{E(j%NO(*Akf#!45lI#4a>iRBa7c%Qb`4o9_L%@whSmB
z%Pd+X>+^xQ8Q|Y2-?q3;p6{AtF;wNFmWEuLB6|_y%=~29e1cQcfi^{u;1fM&*E*;B
zhsqZuAuzmjR&eFUR;_2`ImCt+a_HpgqutO;&CipFd&=n<-HR76BlLGf<KibQVt<mQ
zcKh}SSMn@6I6XKpvicms#!Rl|udU&#5l(@BnbsE$UCLgnVEHN3%V}lJf>^Ta&D^()
z;GtaB);>8txxf=<40K;z81&jw5>ae;xb+!}`*7Mo@cAL476XJd{`Vv4Y&N3Q9l+GE
zv|)G%0UX+dit3M5C!vEnUOZvpjiCUW$$m8gs)1NAbY)eECfy~d{XC`he$j9TbMeDL
z3|;5w!_8dWXI0Ts88?5y8g}2)B{en+HP-(mS%vgzRQMqtXu`JMIM;md#N@GfNcz^5
zERX`F@AEtv9~>%G^ua9kJ(NIqqEiNJpB*>Po<hrMQB!^ZY1#t7WNcT_WH6=uO=<~*
z*^w07V+2CYN31ka7|0NX(6rf!2T&`wBH#}(LCRXNRcQU)ci5jorL4?*pE4VZT>Alq
z6IuXeJ`0}J`xW1Q_?{D3;0;3Dl~%jI?-}`L(Cf8dEq>r1ZjXa7y7%VIRnYt+z26V!
zPJX5E%;Z<({&D*QK=#A~+FYxg;CI<K9|u0N;A*<)+P5V&0S!TI<Nt?MtuK0&%>N%d
reM_$55AY8H&awX&!2jP}UxZxPGG=MFsOt#c$6;=2bvXZ!+tvRA$LlzT

literal 0
HcmV?d00001

diff --git a/docs/2_39/cmdstan-guide/index.html b/docs/2_39/cmdstan-guide/index.html
new file mode 100644
index 000000000..db95069e3
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/index.html
@@ -0,0 +1,1077 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>CmdStan User’s Guide</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/installation.html" rel="next">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="CmdStan User’s Guide – Stan Docs">
+<meta property="og:description" content="Version 2.39">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed slimcontent">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/index.html">CmdStan User’s Guide</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar zindex-bottom">
+    </div>
+<!-- main -->
+<main class="content page-columns page-full" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">CmdStan User’s Guide</h1>
+<p class="subtitle lead">Version 2.39</p>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    
+  
+    
+  </div>
+  
+
+
+</header>
+
+
+<div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
+<p>This document is a user’s guide for CmdStan, the command-line interface to the <a href="https://mc-stan.org/">Stan</a> statistical modeling language. CmdStan provides the programs and tools to compile Stan programs into C++ executables that can be run directly from the command line, together with a few utilities to check and summarize the resulting outputs.</p>
+<p>In CmdStan, statistical models written in the Stan probabilistic programming language are translated into a C++ program which is then compiled together with the CmdStan routines that provide the logic needed to manage all user inputs and program outputs and the Stan inference algorithms and math library. The resulting command line executable program can be used to</p>
+<ul>
+<li><p>do inference on data, producing an exact or approximate estimate of the posterior;</p></li>
+<li><p>generate new quantities of interest from an existing estimate;</p></li>
+<li><p>generate data from the model according to a given set of parameters.</p></li>
+</ul>
+<p>The packages CmdStanR and CmdStanPy provide interfaces to CmdStan from R and Python, respectively, similarly, JuliaStan also interfaces with CmdStan.</p>
+<p><a href="https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf">Download the pdf version of this manual</a>.</p>
+<section id="benefits-of-cmdstan" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="benefits-of-cmdstan">Benefits of CmdStan</h4>
+<ul>
+<li><p>With every new Stan release, there is a corresponding CmdStan release, therefore CmdStan provides access to the latest version of Stan, and can be used to run the development version of Stan as well.</p></li>
+<li><p>Of the Stan interfaces, CmdStan has the lightest memory footprint, therefore it can fit larger and more complex models. It has has the fewest dependencies, which makes it easier to run in limited environments such as clusters.</p></li>
+<li><p>The output generated is in CSV format and can be post-processed using other Stan interfaces or general tools.</p></li>
+</ul>
+</section>
+<section id="stan-documentation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="stan-documentation">Stan documentation</h4>
+<ul>
+<li><p><a href="https://mc-stan.org/docs/stan-users-guide/index.html"><strong>Stan User’s Guide</strong></a> The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:</p></li>
+<li><p><a href="https://mc-stan.org/docs/reference-manual/index.html"><strong>Stan Reference Manual</strong></a> Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.</p></li>
+<li><p><a href="https://mc-stan.org/docs/functions-reference/index.html"><strong>Stan Functions Reference</strong></a> The Stan Functions Reference provides definitions and examples for all the functions defined in the Stan math library and available in the Stan programming language, including all probability distributions.</p></li>
+</ul>
+</section>
+<section id="copyright-and-trademark" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
+<ul>
+<li><p>Copyright 2011–2025, Stan Development Team and their assignees.</p></li>
+<li><p>The Stan name and logo are registered trademarks of NumFOCUS.</p></li>
+</ul>
+</section>
+<section id="licensing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
+<ul>
+<li><p><em>Text content:</em> <a href="https://creativecommons.org/licenses/by-nd/4.0/legalcode">CC-BY ND 4.0 license</a></p></li>
+<li><p><em>Computer code:</em> <a href="https://opensource.org/licenses/BSD-3-Clause">BSD 3-clause license</a></p></li>
+<li><p><em>Logo:</em> <a href="https://mc-stan.org/about/#copyright-and-trademark">Stan logo usage guidelines</a></p></li>
+</ul>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/installation.html" class="pagination-link" aria-label="CmdStan Installation">
+        <span class="nav-page-text">CmdStan Installation</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/installation.html b/docs/2_39/cmdstan-guide/installation.html
new file mode 100644
index 000000000..abdc6b9c4
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/installation.html
@@ -0,0 +1,1357 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>CmdStan Installation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/example_model_data.html" rel="next">
+<link href="../cmdstan-guide/index.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="CmdStan Installation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">CmdStan Installation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#cmdstan-installation" id="toc-cmdstan-installation" class="nav-link active" data-scroll-target="#cmdstan-installation">CmdStan Installation</a>
+  <ul class="collapse">
+  <li><a href="#conda-install" id="toc-conda-install" class="nav-link" data-scroll-target="#conda-install">Installation via <code>conda</code></a>
+  <ul class="collapse">
+  <li><a href="#cmdstan-install-location-under-conda" id="toc-cmdstan-install-location-under-conda" class="nav-link" data-scroll-target="#cmdstan-install-location-under-conda">CmdStan install location under conda</a></li>
+  </ul></li>
+  <li><a href="#installation-from-github" id="toc-installation-from-github" class="nav-link" data-scroll-target="#installation-from-github">Installation from GitHub</a>
+  <ul class="collapse">
+  <li><a href="#downloading-the-source-code" id="toc-downloading-the-source-code" class="nav-link" data-scroll-target="#downloading-the-source-code">Downloading the source code</a></li>
+  <li><a href="#building-cmdstan" id="toc-building-cmdstan" class="nav-link" data-scroll-target="#building-cmdstan">Building CmdStan</a></li>
+  </ul></li>
+  <li><a href="#checking-the-stan-compiler" id="toc-checking-the-stan-compiler" class="nav-link" data-scroll-target="#checking-the-stan-compiler">Checking the Stan compiler</a></li>
+  <li><a href="#troubleshooting-the-installation" id="toc-troubleshooting-the-installation" class="nav-link" data-scroll-target="#troubleshooting-the-installation">Troubleshooting the installation</a>
+  <ul class="collapse">
+  <li><a href="#common-problems" id="toc-common-problems" class="nav-link" data-scroll-target="#common-problems">Common problems</a></li>
+  </ul></li>
+  <li><a href="#cpp-toolchain" id="toc-cpp-toolchain" class="nav-link" data-scroll-target="#cpp-toolchain">C++ Toolchain</a></li>
+  <li><a href="#gnu-make" id="toc-gnu-make" class="nav-link" data-scroll-target="#gnu-make">Using GNU Make</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/installation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">CmdStan Installation</a></li></ol></nav></header>
+
+
+
+
+<section id="cmdstan-installation" class="level1">
+<h1>CmdStan Installation</h1>
+<p>There are a few ways that you can install CmdStan. Depending on your operating system and your level of expertise, you can either:</p>
+<ul>
+<li><p>Use the <a href="https://docs.conda.io/en/latest/">conda</a> package management system to install a pre-built version of CmdStan along with the required dependencies. <em>Recommended for Windows users</em>.</p></li>
+<li><p>Install the source code from <a href="https://github.com/stan-dev/cmdstan">GitHub CmdStan repository</a>. This requires a modern C++ compiler and toolchain. See the <a href="#cpp-toolchain">C++ Toolchain</a> section for further details.</p></li>
+</ul>
+<section id="conda-install" class="level2">
+<h2 class="anchored" data-anchor-id="conda-install">Installation via <code>conda</code></h2>
+<p>With conda, you can install CmdStan from the <a href="https://conda-forge.org/">conda-forge channel</a>. This will install a pre-built version of CmdStan along with the required dependencies (i.e.&nbsp;a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command <code>cmdstan_model</code> to activate the CmdStan makefile from anywhere.</p>
+<p><em>Note</em>: This requires that conda has been installed already on your machine. We recommend using the <a href="https://github.com/conda-forge/miniforge">miniforge</a> distribution.</p>
+<p>We recommend installing CmdStan in a new conda environment:</p>
+<pre><code> conda create -n stan -c conda-forge cmdstan</code></pre>
+<p>This command creates a new conda environment named <code>stan</code> and downloads and installs the <code>cmdstan</code> package as well as CmdStan and the required C++ toolchain.</p>
+<p>To install into an existing conda environment, use the conda <code>install</code> command instead of <code>create</code>:</p>
+<pre><code> conda install -c conda-forge cmdstan</code></pre>
+<p>Whichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment <code>stan</code>, run the command</p>
+<pre><code> conda activate stan</code></pre>
+<p>By default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and <em>newer</em> can be installed by specifying <code>cmdstan==VERSION</code> in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment</p>
+<pre><code>conda install -c conda-forge cmdstan=2.27.0</code></pre>
+<section id="cmdstan-install-location-under-conda" class="level3">
+<h3 class="anchored" data-anchor-id="cmdstan-install-location-under-conda">CmdStan install location under conda</h3>
+<p>A Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command</p>
+<pre><code> conda info -e</code></pre>
+<p>The shell environment variable <code>CONDA_PREFIX</code> points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the <code>bin</code> subdirectory of the conda environment directory, i.e., <code>$CONDA_PREFIX/bin/cmdstan</code> (Linux, MacOS), <code>%CONDA_PREFIX%\bin\cmdstan</code> (Windows).</p>
+<p>Please report conda-specific install problems directly to the conda-forge issue tracker, <a href="https://github.com/conda-forge/cmdstan-feedstock/issues">here</a>.</p>
+</section>
+</section>
+<section id="installation-from-github" class="level2">
+<h2 class="anchored" data-anchor-id="installation-from-github">Installation from GitHub</h2>
+<p>Installation from GitHub consists of the following steps:</p>
+<ul>
+<li><p>Verify that you have a modern C++ toolchain. See the <a href="#cpp-toolchain">C++ Toolchain</a> section for details.</p></li>
+<li><p>Download the CmdStan source code from GitHub</p></li>
+<li><p>Build the CmdStan libraries and executables</p></li>
+<li><p>Check the installation by compiling and running the CmdStan example model <code>bernoulli.stan</code>.</p></li>
+</ul>
+<section id="downloading-the-source-code" class="level3">
+<h3 class="anchored" data-anchor-id="downloading-the-source-code">Downloading the source code</h3>
+<p>The GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the <code>cmdstan</code> module; the Stan inference engine algorithms and Stan math library functions are specified as <a href="https://git-scm.com/book/en/v2/Git-Tools-Submodules">submodules</a> and stored in the GitHub repositories <a href="https://github.com/stan-dev/stan">stan</a> and <a href="https://github.com/stan-dev/math">math</a>, respectively.</p>
+<p>A <a href="https://github.com/stan-dev/cmdstan/releases">CmdStan release</a> is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do <strong>not</strong> contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-<version number="">”, e.g.&nbsp;“cmdstan-2.29.2”.</version></p>
+<p>By cloning the CmdStan repository with argument <code>--recursive</code>, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named <code>cmdstan</code>:</p>
+<pre><code>&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive</code></pre>
+<p>Throughout this manual, we refer to this top-level CmdStan source directory as <strong><code>&lt;cmdstan-home&gt;</code></strong>. This directory contains the following subdirectories:</p>
+<ul>
+<li>directory <code>cmdstan/stan</code> contains the sub-module <code>stan</code> (https://github.com/stan-dev/stan)</li>
+<li>directory <code>cmdstan/stan/lib/stan_math</code> contains the sub-module <code>math</code> (https://github.com/stan-dev/math)</li>
+</ul>
+</section>
+<section id="building-cmdstan" class="level3">
+<h3 class="anchored" data-anchor-id="building-cmdstan">Building CmdStan</h3>
+<p>Building CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:</p>
+<ul>
+<li><p><code>stanc</code>: the Stan compiler (translates Stan language to C++).</p></li>
+<li><p><code>stansummary</code>: a basic posterior analysis tool. The <code>stansummary</code> utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, <code>stansummary</code> reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and <span class="math inline">\(\hat{R}\)</span> values.</p></li>
+<li><p><code>diagnose</code>: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.</p></li>
+</ul>
+<p>CmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): <code>bin/linux-stanc</code>, <code>bin/mac-stanc</code> and <code>bin/windows-stanc</code>. The CmdStan makefile <code>build</code> task copies the appropriate binary to <code>bin/stanc</code>. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.</p>
+<p>Steps to build CmdStan:</p>
+<ul>
+<li><p>Open a command-line terminal window and change directories to the CmdStan home directory.</p></li>
+<li><p>Run the makefile target <code>build</code> which instantiates the CmdStan utilities and compiles all necessary C++ libraries.</p></li>
+</ul>
+<pre><code>&gt; cd &lt;cmdstan-home&gt;
+&gt; make build</code></pre>
+<p>If your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the <code>-j</code> option. For example, to build on 4 cores, type:</p>
+<pre><code>&gt; make -j4 build</code></pre>
+<p>When <code>make build</code> is successful, the directory <code>&lt;cmdstan-home&gt;/bin/</code> will contain the executables <code>stanc</code>, <code>stansummary</code>, and <code>diagnose</code> (on Windows, corresponding <code>.exe</code> files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:</p>
+<pre><code>--- CmdStan v2.29.2 built ---</code></pre>
+<p><strong>Warning:</strong> <em>The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.</em></p>
+<p><strong>Windows only:</strong> CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory <code>&lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb</code> is part of the <code>PATH</code> environment variable. See <a href="https://helpdeskgeek.com/windows-10/add-windows-path-environment-variable/">these instructions</a> for details on changing the <code>PATH</code>. To permanently make this setting for the current user, you may execute:</p>
+<pre><code>&gt; make install-tbb</code></pre>
+<p>After changing the <code>PATH</code> environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)</p>
+</section>
+</section>
+<section id="checking-the-stan-compiler" class="level2">
+<h2 class="anchored" data-anchor-id="checking-the-stan-compiler">Checking the Stan compiler</h2>
+<p>To check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.</p>
+<p>On Linux and macOS:</p>
+<pre><code># compile the example
+&gt; make examples/bernoulli/bernoulli
+
+# fit to provided data (results of 10 trials, 2 out of 10 successes)
+&gt; ./examples/bernoulli/bernoulli sample\
+  data file=examples/bernoulli/bernoulli.data.json
+
+# default output written to file `output.csv`,
+# default num_samples is 1000, output file should have approx. 1050 lines
+&gt; wc -l output.csv
+
+# run the `bin/stansummary utility to summarize parameter estimates
+&gt; bin/stansummary output.csv</code></pre>
+<p>On Windows:</p>
+<pre><code># compile the example
+&gt; make examples/bernoulli/bernoulli.exe
+
+# fit to provided data (results of 10 trials, 2 out of 10 successes)
+&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json
+
+# run the `bin/stansummary.exe utility to summarize parameter estimates
+&gt; bin/stansummary.exe output.csv</code></pre>
+<p>The sample data in file <code>bernoulli.json.data</code> specifies 2 out of 10 successes, therefore the range <code>mean(theta)</code><span class="math inline">\(\pm\)</span><code>sd(theta)</code> should include 0.2.</p>
+</section>
+<section id="troubleshooting-the-installation" class="level2">
+<h2 class="anchored" data-anchor-id="troubleshooting-the-installation">Troubleshooting the installation</h2>
+<p>Updates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the <code>&lt;cmdstan-home&gt;</code> directory:</p>
+<pre><code>&gt; cd &lt;cmdstan-home&gt;
+&gt; make clean-all
+&gt; make build</code></pre>
+<section id="common-problems" class="level3">
+<h3 class="anchored" data-anchor-id="common-problems">Common problems</h3>
+<p>This section contains solutions to problems reported on https://discourse.mc-stan.org</p>
+<p><strong>Compiler error message about PCH file</strong></p>
+<p>To speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:</p>
+<pre><code>error: PCH file uses an older PCH format that is no longer supported</code></pre>
+<p>In this case, clean and rebuild CmdStan, as shown in the previous section.</p>
+<p><strong>Windows: ‘g++’, ‘make’, or ‘cut’ is not recognized</strong></p>
+<p>The CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:</p>
+<pre><code>'cut' is not recognized as an internal or external command,
+operable program or batch file.</code></pre>
+<p>To fix this, ensure you have followed the steps for adding the toolchain to your <code>PATH</code> and installing the additional utilities covered in the <a href="#windows">configuration instructions</a></p>
+<p><strong>Spaces in paths to CmdStan or model</strong></p>
+<p><code>make</code> can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the <code>One Drive</code> folder.</p>
+<p>Unfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:</p>
+<pre><code>make: *** INTERNAL: readdir: Invalid argument</code></pre>
+<pre><code>make: *** [make/program:50: x.hpp] Error 2</code></pre>
+<p>If the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.</p>
+</section>
+</section>
+<section id="cpp-toolchain" class="level2">
+<h2 class="anchored" data-anchor-id="cpp-toolchain">C++ Toolchain</h2>
+<p>Compiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.</p>
+<section id="linux" class="level4">
+<h4 class="anchored" data-anchor-id="linux">Linux</h4>
+<p>The required C++ compiler is <code>g++ 4.9 3</code>. On most systems the GNU Make utility is pre-installed and is the default <code>make</code> utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:</p>
+<pre><code>g++ --version
+make --version</code></pre>
+<p>If these are at least at <code>g++</code> version 4.9.3 or later and <code>make</code> version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler <code>g++</code>, because later versions are faster.</p>
+<p>To install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:</p>
+<pre><code>sudo apt install g++
+sudo apt install make</code></pre>
+<p>If you can’t run <code>sudo</code>, you will need to ask your sysadmin or cluster administrator to install these tools for you.</p>
+</section>
+<section id="macos" class="level4">
+<h4 class="anchored" data-anchor-id="macos">MacOS</h4>
+<p>To check if you already already have an appropriate toolchain installed, open the Terminal application and enter:</p>
+<pre><code>clang++ --version
+make --version</code></pre>
+<p>If either of these commands prints the message <code>command not found</code>, you will need to install Xcode’s command line tools.</p>
+<p>Open the Terminal application and enter:</p>
+<pre><code>xcode-select --install</code></pre>
+<p>Select “Install” in the window that opens.</p>
+<p>After the installation completes, you can double check that installation was successful by reopening the Terminal and running:</p>
+<pre><code>clang++ --version
+make --version</code></pre>
+<p>You can read more about Xcode on its site: <a href="https://developer.apple.com/xcode/">https://developer.apple.com/xcode/</a></p>
+<p>We don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.</p>
+</section>
+<section id="windows" class="level4">
+<h4 class="anchored" data-anchor-id="windows">Windows</h4>
+<p>The Windows toolchain consists of programs <code>g++</code>, the C++ compiler, and <code>make</code>, the GNU Make utility. To check if these are present, open a command shell [^1] and type:</p>
+<pre><code>g++ --version
+make --version</code></pre>
+<p>CmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your <code>PATH</code> variable, See <a href="https://helpdeskgeek.com/windows-10/add-windows-path-environment-variable/">these instructions</a> for details on changing the <code>PATH</code> if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.</p>
+<section id="rtools45" class="level5">
+<h5 class="anchored" data-anchor-id="rtools45">RTools45</h5>
+<p>All required utilities (e.g., <code>make</code>, <code>g++</code>) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the <a href="https://cran.r-project.org/bin/windows/Rtools/rtools45/rtools.html">R Project documentation</a>.</p>
+<p>The R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.</p>
+<p>Note that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a <a href="https://support.microsoft.com/en-us/topic/update-for-universal-c-runtime-in-windows-c0514201-7fe6-95a3-b0a5-287930f3560c">Microsoft update</a>.</p>
+<section id="installation---intelamd-64-bit-x86_64" class="level6">
+<h6 class="anchored" data-anchor-id="installation---intelamd-64-bit-x86_64">Installation - Intel/AMD 64-bit (x86_64)</h6>
+<p>Download the installer and complete the prompts for installation:</p>
+<ul>
+<li><a href="https://github.com/r-hub/rtools45/releases/download/latest/rtools45.exe">RTools45</a></li>
+</ul>
+<p>Next, you need to add the toolchain directory to your <code>PATH</code> variable:</p>
+<pre><code>C:\rtools45\usr\bin
+C:\rtools45\x86_64-w64-mingw32.static.posix\bin</code></pre>
+</section>
+<section id="installation---arm-64-bit-arm64aarch64" class="level6">
+<h6 class="anchored" data-anchor-id="installation---arm-64-bit-arm64aarch64">Installation - ARM 64-bit (arm64/aarch64)</h6>
+<p>Download the installer and complete the prompts for installation:</p>
+<ul>
+<li><a href="https://github.com/r-hub/rtools45/releases/download/latest/rtools45-aarch64.exe">RTools45 - ARM64</a></li>
+</ul>
+<p>Next, you need to add the toolchain directory to your <code>PATH</code> variable:</p>
+<pre><code>C:\rtools45-aarch64\usr\bin
+C:\rtools45-aarch64\aarch64-w64-mingw32.static.posix\bin</code></pre>
+</section>
+</section>
+</section>
+</section>
+<section id="gnu-make" class="level2">
+<h2 class="anchored" data-anchor-id="gnu-make">Using GNU Make</h2>
+<p>CmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.</p>
+<p>GNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed <em>only</em> when there are changes to those other files. Thus Make provides an efficient way to manage complex software.</p>
+<p>The CmdStan Makefile is in the <code>&lt;cmdstan-home&gt;</code> directory and is named <code>makefile</code>. This is one of the default <a href="https://www.gnu.org/software/make/manual/html_node/Makefile-Names.html">GNU Makefile names</a>, which allows you to omit the <code>-f makefile</code> argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, <strong>Make only works properly when invoked from the <code>&lt;cmdstan-home&gt;</code> directory</strong>; attempts to use this Makefile from another directory by specifying the full path to the file <code>makefile</code> won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:</p>
+<pre><code>make -f ~/github/stan-dev/cmdstan/makefile
+/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory
+/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory
+/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory
+/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory
+make: *** No rule to make target `make/command'.  Stop.</code></pre>
+<p>The <a href="#conda-install">conda-forge <code>cmdstan</code> package</a> provides a solution to this problem via <code>cmdstan_model</code> command which lets you run the CmdStan makefile from anywhere to compile a Stan model.</p>
+<p>Makefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix <code>.stan</code>; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program <code>my_program.stan</code> in directory <code>../my_dir/</code>, the make target is <code>../my_dir/my_program</code> or <code>../my_dir/my_program.exe</code> (on Windows).</p>
+<p>To call Make, you invoke the utility name, <code>make</code>, followed by, in order:</p>
+<ul>
+<li><p>zero or more <a href="https://www.gnu.org/software/make/manual/html_node/Options-Summary.html">Make program options</a>, then specify any Make variables as a series of</p></li>
+<li><p>zero of more Make variables, described below</p></li>
+<li><p>zero or more <em>target</em> names; the set of names is determined by the Makefile rules.</p></li>
+</ul>
+<pre><code>make &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;</code></pre>
+<p><strong>Makefile Variables</strong></p>
+<p>Make targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile <code>../my_dir/my_program.stan</code> for an OpenCL (GPU) machine, set the makefile variable <code>STAN_OPENCL</code> to <code>TRUE</code>:</p>
+<pre><code>&gt; make STAN_OPENCL=TRUE ../my_dir/my_program</code></pre>
+<p>Makefile variables can also be set by creating a file named <code>local</code> in the CmdStan <code>make</code> subdirectory which contains a list of <code>&lt;VARIABLE&gt;=&lt;VALUE&gt;</code> pairs, one per line. For example, to get the same effect as the above command every time, you would put the line <code>STAN_OPENCL=TRUE</code> into the file <code>&lt;cmdstan_home&gt;/make/local</code>.</p>
+<p>The complete set of Makefile variables can be found in file <code>&lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags</code>.</p>
+<p><strong>Make Targets</strong></p>
+<p>When invoked without any arguments at all, Make prints a help message:</p>
+<pre><code>&gt; make
+--------------------------------------------------------------------------------
+CmdStan v2.33.1 help
+
+  Build CmdStan utilities:
+    &gt; make build
+
+    This target will:
+    1. Install the Stan compiler bin/stanc from stanc3 binaries.
+    2. Build the print utility bin/print (deprecated; will be removed in v3.0)
+    3. Build the stansummary utility bin/stansummary
+    4. Build the diagnose utility bin/diagnose
+    5. Build all libraries and object files compile and link an executable Stan program
+
+    Note: to build using multiple cores, use the -j option to make, e.g.,
+    for 4 cores:
+    &gt; make build -j4
+
+
+  Build a Stan program:
+
+    Given a Stan program at foo/bar.stan, build an executable by typing:
+    &gt; make foo/bar
+
+    This target will:
+    1. Install the Stan compiler (bin/stanc), as needed.
+    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.
+    3. Compile the C++ code using cc . to generate foo/bar
+
+  Additional make options:
+    STANCFLAGS: defaults to "". These are extra options passed to bin/stanc
+      when generating C++ code. If you want to allow undefined functions in the
+      Stan program, either add this to make/local or the command line:
+          STANCFLAGS = --allow_undefined
+    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the
+      header file that is included. This defaults to "user_header.hpp" in the
+      directory of the Stan program.
+    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads
+      the nightly version.
+    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.
+    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.
+
+
+  Example - bernoulli model: examples/bernoulli/bernoulli.stan
+
+    1. Build the model:
+       &gt; make examples/bernoulli/bernoulli
+    2. Run the sampling algorithm given the model and data:
+       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R
+    3. Look at the posterior sample:
+       &gt; bin/stansummary output.csv
+
+
+  Clean CmdStan:
+
+    Remove the built CmdStan tools:
+    &gt; make clean-all
+
+--------------------------------------------------------------------------------</code></pre>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/index.html" class="pagination-link" aria-label="CmdStan User's Guide">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">CmdStan User’s Guide</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/example_model_data.html" class="pagination-link" aria-label="Example Model and Data">
+        <span class="nav-page-text">Example Model and Data</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/installation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/json_apdx.html b/docs/2_39/cmdstan-guide/json_apdx.html
new file mode 100644
index 000000000..f958afc7a
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/json_apdx.html
@@ -0,0 +1,1273 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>JSON Format for CmdStan</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/rdump_apdx.html" rel="next">
+<link href="../cmdstan-guide/stan_csv_apdx.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="JSON Format for CmdStan">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/json_apdx.html">JSON Format for CmdStan</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#json" id="toc-json" class="nav-link active" data-scroll-target="#json">JSON Format for CmdStan</a>
+  <ul class="collapse">
+  <li><a href="#creating-json-files" id="toc-creating-json-files" class="nav-link" data-scroll-target="#creating-json-files">Creating JSON files</a></li>
+  <li><a href="#json-syntax-summary" id="toc-json-syntax-summary" class="nav-link" data-scroll-target="#json-syntax-summary">JSON syntax summary</a></li>
+  <li><a href="#stan-data-types-in-json-notation" id="toc-stan-data-types-in-json-notation" class="nav-link" data-scroll-target="#stan-data-types-in-json-notation">Stan data types in JSON notation</a>
+  <ul class="collapse">
+  <li><a href="#empty-arrays-in-json" id="toc-empty-arrays-in-json" class="nav-link" data-scroll-target="#empty-arrays-in-json">Empty arrays in JSON</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/json_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/json_apdx.html">JSON Format for CmdStan</a></li></ol></nav></header>
+
+
+
+
+<section id="json" class="level1">
+<h1>JSON Format for CmdStan</h1>
+<p>CmdStan can use JSON format for input data for both model data and parameters. Model data is read in by the model constructor. Model parameters are used to initialize the sampler and optimizer.</p>
+<section id="creating-json-files" class="level2">
+<h2 class="anchored" data-anchor-id="creating-json-files">Creating JSON files</h2>
+<p>You can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the <a href="https://mc-stan.org/cmdstanr/reference/write_stan_json"><code>write_stan_json()</code></a> function provided by the CmdStanR interface.</p>
+</section>
+<section id="json-syntax-summary" class="level2">
+<h2 class="anchored" data-anchor-id="json-syntax-summary">JSON syntax summary</h2>
+<p>JSON is a data interchange notation, defined by an <a href="https://ecma-international.org/publications-and-standards/standards/ecma-404/">EMCA standard</a>. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:</p>
+<ul>
+<li><p>Structural tokens are the left and right curly bracket <code>{}</code>, left and right square bracket <code>[]</code>, the semicolon <code>;</code>, and the comma <code>,</code>.</p></li>
+<li><p>Literal tokens must always be in lowercase. There are three literal tokens: <code>true</code>, <code>false</code>, <code>null</code>.</p></li>
+<li><p>A primitive value is a single token which is either a literal, a string, or a number.</p></li>
+<li><p>A string consists of zero or more Unicode characters enclosed in double quotes, e.g.&nbsp;<code>"foo"</code>. A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g.&nbsp;<code>"\\uHHHH"</code> where <code>HHHH</code> is the Unicode code point in hex.</p></li>
+<li><p>Numbers are represented using either decimal notation or scientific notation. The following are examples of numbers: <code>17</code>, <code>17.2</code>, <code>-17.2</code>, <code>-17.2e8</code>, <code>17.2e-8</code>. <br> There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.</p></li>
+<li><p>The special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string <code>"Inf"</code>, the string <code>"Infinity"</code>, or the atom <code>Infinity</code>. Negative infinity can be represented as the string <code>"-Inf"</code>, the string <code>"-Infinity"</code>, or the atom <code>-Infinity</code>. Not-a-number can be represented as the string <code>"NaN"</code> or the atom <code>NaN</code>. These values may be mixed with other numerical types.</p></li>
+<li><p>A complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number <span class="math inline">\(2.3 - 1.83i\)</span> would be represented in JSON as the two-element array <code>[2.3, -1.83]</code>.</p></li>
+<li><p>A JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: <code>[]</code>, <code>[1]</code>, <code>[0.2, "-inf", true]</code>.</p></li>
+<li><p>Vectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector <span class="math inline">\([1 \quad 2]^{\top}\)</span> and the row vector <span class="math inline">\([1 \quad 2]\)</span> are represented by the JSON array <code>[1, 2]</code>.</p></li>
+<li><p>Complex vectors are represented as arrays of two-element arrays. For example, the complex vector <span class="math inline">\([2.3 - 1.83i \quad -4.8 +
+2i]^{\top}\)</span> is represented as <code>[[2.3, -1.83], [-4.8, 2]]</code> in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).</p></li>
+<li><p>Matrices are represented as arrays of their row vectors. For example, the <span class="math inline">\(2 \times 3\)</span> matrix <span class="math display">\[\begin{equation*}
+\begin{bmatrix}
+1 &amp; 2.7 &amp; -9.8 \\
+4.2 &amp; 1.8 &amp; -7.3
+\end{bmatrix}
+\end{equation*}\]</span> is represented in JSON as <code>[[1, 2.7, -9.8],   [4.2, 1.8, -7.3]]</code>.</p></li>
+<li><p>Complex matrices are also represented as arrays of their row vectors. For example, the <span class="math inline">\(2 \times 3\)</span> complex matrix <span class="math display">\[\begin{equation*}
+\begin{bmatrix}
+1 + 2i &amp; 3 - 4.2i &amp; 13.1 + 2.7i \\
+3.1 &amp; -5i &amp; 0
+\end{bmatrix}
+\end{equation*}\]</span> would be represented in JSON as <code>[[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]]</code>.</p></li>
+<li><p>Tuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple <code>(1.5, 3.4)</code> is represented in JSON as <code>{"1": 1.5, "2": 3.4}</code>.</p></li>
+<li><p>A name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.</p></li>
+<li><p>A JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: <code>{ }</code>, <code>{"foo": null}</code>, <code>{"bar" : 17, "baz" : [14,15,16.6] }</code>.</p></li>
+</ul>
+</section>
+<section id="stan-data-types-in-json-notation" class="level2">
+<h2 class="anchored" data-anchor-id="stan-data-types-in-json-notation">Stan data types in JSON notation</h2>
+<p>Stan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:</p>
+<pre><code>{ "N" : 10, "y" : [0,1,0,0,0,0,0,0,0,1] }</code></pre>
+<p>Matrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> d1;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> d2;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> d3;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[d1, d2, d3] <span class="dt">int</span> ar;</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>the following JSON input would be valid:</p>
+<pre><code>{ "d1" : 2,
+  "d2" : 3,
+  "d3" : 4,
+  "ar" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],
+          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]
+}</code></pre>
+<p>JSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.</p>
+<p>All data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.</p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 41%">
+<col style="width: 58%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: left;">Stan declaration</th>
+<th style="text-align: left;">JSON encoding</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><code>int i</code></td>
+<td style="text-align: left;"><code>"i": 17</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>real a</code></td>
+<td style="text-align: left;"><code>"a" : 17</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>"a" : 17.2</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>"a" : "NaN"</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>"a" : "+inf"</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>"a" : "-inf"</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex z</code></td>
+<td style="text-align: left;"><code>"z": [1, -2.3]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>array[5] int</code></td>
+<td style="text-align: left;"><code>"a" : [1, 2, 3, 4, 5]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>array[5] real a</code></td>
+<td style="text-align: left;"><code>"a" : [ 1, 2, 3.3, "NaN", 5 ]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>array[2] complex b</code></td>
+<td style="text-align: left;"><code>"b" : [[1, -2.3], [4.9, 0]]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>vector[5] a</code></td>
+<td style="text-align: left;"><code>"a" : [1, 2, 3.3, "NaN", 5]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>row_vector[5] a</code></td>
+<td style="text-align: left;"><code>"a" : [1, 2, 3.3, "NaN", 5]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>matrix[2, 3] a</code></td>
+<td style="text-align: left;"><code>"a" : [[ 1, 2, 3 ], [ 4, 5, 6]]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_vector[2] c</code></td>
+<td style="text-align: left;"><code>"c" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex_row_vector[2] c</code></td>
+<td style="text-align: left;"><code>"c" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_matrix[2, 3] d</code></td>
+<td style="text-align: left;"><code>"d" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>tuple(real, array[2] int) t</code></td>
+<td style="text-align: left;"><code>"t" : { "1": 1.4, "2": [1, 2]}</code></td>
+</tr>
+</tbody>
+</table>
+<section id="empty-arrays-in-json" class="level3">
+<h3 class="anchored" data-anchor-id="empty-arrays-in-json">Empty arrays in JSON</h3>
+<p>JSON notation is not able to distinguish between multi-dimensional arrays where any dimension is <span class="math inline">\(0\)</span>, e.g., a 2-D array with dimensions <span class="math inline">\((1,0)\)</span>, i.e., an array which contains a single array which is empty, has JSON representation <code>[ ]</code>. To see how this works, consider the following Stan program data block:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> d;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[d] <span class="dt">int</span> ar_1d;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[d, d] <span class="dt">int</span> ar_2d;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[d, d, d] <span class="dt">int</span> ar_3d;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In the case where variable <code>d</code> is <code>1</code>, all arrays will contain a single value. If array variable <code>ar_d1</code> contains value <code>7</code>, 2-D array variable <code>ar_d2</code> contains (an array which contains) value <code>8</code>, and 3-D array variable <code>ar_d3</code> contains (an array which contains an array which contains) value <code>9</code>, the JSON representation is:</p>
+<pre><code>{ "ar_d1" : [7],
+  "ar_d2" : [[8]],
+  "ar_d3" : [[[9]]]
+}</code></pre>
+<p>However, in the case where variable <code>d</code> is <code>0</code>, <code>ar_d1</code> is empty, i.e., it contains no values, as is <code>ar_d2</code>, <code>ar_d3</code>, and the JSON representation is</p>
+<pre><code>{ "d" : 0,
+  "ar_d1" : [ ],
+  "ar_d2" : [ ],
+  "ar_d3" : [ ]
+}</code></pre>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/stan_csv_apdx.html" class="pagination-link" aria-label="Stan CSV File Format">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Stan CSV File Format</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/rdump_apdx.html" class="pagination-link" aria-label="RDump Format for CmdStan">
+        <span class="nav-page-text">RDump Format for CmdStan</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/json_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/laplace_sample_config.html b/docs/2_39/cmdstan-guide/laplace_sample_config.html
new file mode 100644
index 000000000..65f2b4133
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/laplace_sample_config.html
@@ -0,0 +1,1134 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Laplace sampling</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/log_prob_config.html" rel="next">
+<link href="../cmdstan-guide/generate_quantities_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Laplace sampling">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/laplace_sample_config.html">Laplace sampling</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#laplace-sampling" id="toc-laplace-sampling" class="nav-link active" data-scroll-target="#laplace-sampling">Laplace sampling</a>
+  <ul class="collapse">
+  <li><a href="#configuration" id="toc-configuration" class="nav-link" data-scroll-target="#configuration">Configuration</a></li>
+  <li><a href="#csv-output" id="toc-csv-output" class="nav-link" data-scroll-target="#csv-output">CSV output</a></li>
+  <li><a href="#diagnostic-file-outputs" id="toc-diagnostic-file-outputs" class="nav-link" data-scroll-target="#diagnostic-file-outputs">Diagnostic file outputs</a></li>
+  <li><a href="#example" id="toc-example" class="nav-link" data-scroll-target="#example">Example</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/laplace_sample_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/laplace_sample_config.html">Laplace sampling</a></li></ol></nav></header>
+
+
+
+
+<section id="laplace-sampling" class="level1">
+<h1>Laplace sampling</h1>
+<p>The <code>laplace</code> method produces a sample from a normal approximation centered at the <a href="../cmdstan-guide/optimize_config.html">mode</a> of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood. In general, the posterior mode in the unconstrained space doesn’t correspond to the mean (nor mode) in the constrained space, and thus the sample is needed to infer the mean as well as the standard deviation. (See <a href="https://avehtari.github.io/casestudies/Jacobian/jacobian.html">this case study</a> for a visual illustration.)</p>
+<p>This is computationally inexpensive compared to exact Bayesian inference with MCMC. The goodness of this estimate depends on both the estimate of the mode and how much the true posterior in the unconstrained space resembles a Gaussian.</p>
+<section id="configuration" class="level2">
+<h2 class="anchored" data-anchor-id="configuration">Configuration</h2>
+<p>This method takes several arguments:</p>
+<ul>
+<li><p><code>mode</code> - Input file of parameters values on the constrained scale. When Stan’s <code>optimize</code> method is used to estimate the modal values, the value of boolean argument <code>jacobian</code> should be <code>false</code> if <code>optimize</code> was run with default settings, i.e., the input is the MLE estimate; if <code>optimize</code> was run with argument <code>jacobian=true</code>, then the <code>laplace</code> method default setting, <code>jacobian=true</code>, should be used.</p></li>
+<li><p><code>jacobian</code> - Whether or not the <a href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html">Jacobian adjustment</a> should be included in the gradient. The default value is <code>true</code> (include adjustment). (Note: in optimization, the default value is <code>false</code>, for historical reasons.)</p></li>
+<li><p><code>draws</code> - How many total draws to return. The default is <span class="math inline">\(1000\)</span>.</p></li>
+<li><p><code>calculate_lp</code> - Whether to calculate the log probability of the model at each draw. If this is <code>false</code>, the <code>log_p__</code> column of the output will be entirely <code>nan</code>. The default value is <code>true</code>.</p></li>
+</ul>
+</section>
+<section id="csv-output" class="level2">
+<h2 class="anchored" data-anchor-id="csv-output">CSV output</h2>
+<p>The output file consists of the following pieces of information:</p>
+<ul>
+<li><p>The full set of configuration options available for the <code>laplace</code> method is reported at the beginning of the output file as CSV comments.</p></li>
+<li><p>Output columns <code>log_p__</code> and <code>log_q__</code>, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.</p></li>
+<li><p>Output columns for all model parameters on the constrained scale.</p></li>
+</ul>
+</section>
+<section id="diagnostic-file-outputs" class="level2">
+<h2 class="anchored" data-anchor-id="diagnostic-file-outputs">Diagnostic file outputs</h2>
+<p>If requested with <code>output diagnostic_file=</code>, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.</p>
+</section>
+<section id="example" class="level2">
+<h2 class="anchored" data-anchor-id="example">Example</h2>
+<p>To get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:</p>
+<ul>
+<li><p>find the MAP estimate by running optimization with argument <code>jacobian=true</code></p></li>
+<li><p>run the Laplace estimator using the MAP estimate as the <code>mode</code> argument.</p></li>
+</ul>
+<p>Because the default output file name from all methods is <code>output.csv</code>, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.</p>
+<pre><code>./examples/bernoulli/bernoulli optimize jacobian=1 \
+  data file=examples/bernoulli/bernoulli.data.json \
+  output file=bernoulli_optimize_lbfgs.csv random seed=1234
+
+
+./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \
+ data file=examples/bernoulli/bernoulli.data.json random seed=1234</code></pre>
+<p>The header and first few data rows of the output sample are shown below.</p>
+<pre><code># method = laplace
+#   laplace
+#     mode = bernoulli_lbfgs.csv
+#     jacobian = true (Default)
+#     draws = 1000 (Default)
+#     calculate_lp = true (default)
+# id = 1 (Default)
+# data
+#   file = examples/bernoulli/bernoulli.data.json
+# init = 2 (Default)
+# random
+#   seed = 875960551 (Default)
+# output
+#   file = output.csv (Default)
+#   diagnostic_file =  (Default)
+#   refresh = 100 (Default)
+#   sig_figs = 8 (Default)
+#   profile_file = profile.csv (Default)
+# num_threads = 1 (Default)
+log_p__,log_q__,theta
+-9.4562,-2.33997,0.0498545
+-6.9144,-0.0117349,0.182898
+-7.18171,-0.746034,0.376428
+...</code></pre>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/generate_quantities_config.html" class="pagination-link" aria-label="Generating Quantities of Interest from a Fitted Model">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Generating Quantities of Interest from a Fitted Model</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/log_prob_config.html" class="pagination-link" aria-label="Extracting log probabilities and gradients for diagnostics">
+        <span class="nav-page-text">Extracting log probabilities and gradients for diagnostics</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/laplace_sample_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/log_prob_config.html b/docs/2_39/cmdstan-guide/log_prob_config.html
new file mode 100644
index 000000000..b60d63e64
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/log_prob_config.html
@@ -0,0 +1,1121 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Extracting log probabilities and gradients for diagnostics</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/diagnose_config.html" rel="next">
+<link href="../cmdstan-guide/laplace_sample_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Extracting log probabilities and gradients for diagnostics">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/log_prob_config.html">Extracting log probabilities and gradients for diagnostics</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#extracting-log-probabilities-and-gradients-for-diagnostics" id="toc-extracting-log-probabilities-and-gradients-for-diagnostics" class="nav-link active" data-scroll-target="#extracting-log-probabilities-and-gradients-for-diagnostics">Extracting log probabilities and gradients for diagnostics</a>
+  <ul class="collapse">
+  <li><a href="#configuration" id="toc-configuration" class="nav-link" data-scroll-target="#configuration">Configuration</a></li>
+  <li><a href="#csv-output" id="toc-csv-output" class="nav-link" data-scroll-target="#csv-output">CSV output</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/log_prob_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/log_prob_config.html">Extracting log probabilities and gradients for diagnostics</a></li></ol></nav></header>
+
+
+
+
+<section id="extracting-log-probabilities-and-gradients-for-diagnostics" class="level1">
+<h1>Extracting log probabilities and gradients for diagnostics</h1>
+<p>CmdStan can return the computed log probability and the gradient with respect to a set of parameters.</p>
+<p>This is similar to the <code>diagnose</code> subcommand, but the output format differs and the results here are not compared with those from finite differences.</p>
+<p><strong>Note</strong>: Startup and data initialization costs mean that this method is not an efficient way to calculate these quantities. It is provided only for convenience and should not be used for serious computation.</p>
+<section id="configuration" class="level2">
+<h2 class="anchored" data-anchor-id="configuration">Configuration</h2>
+<p>This method takes 3 arguments:</p>
+<ul>
+<li><p><code>jacobian</code> - Whether or not the <a href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html">Jacobian adjustment for constrained</a> parameters should be included in the gradient. Default value is <code>true</code> (include adjustment).</p></li>
+<li><p><code>constrained_params</code> - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using <a href="../cmdstan-guide/json_apdx.html">JSON</a> format. Alternatively, the input file can be set of draws in <a href="../cmdstan-guide/stan_csv_apdx.html">StanCSV</a> format.</p></li>
+<li><p><code>unconstrained_params</code> - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called <code>params_r</code>, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.</p></li>
+</ul>
+<p>Only one of <code>constrained_params</code> and <code>unconstrained_params</code> can be specified.</p>
+<p>For more on the differences between constrained and unconstrained parameters, see the Stan reference manual <a href="https://mc-stan.org/docs/reference-manual/transforms.html">section on variable transforms</a>.</p>
+</section>
+<section id="csv-output" class="level2">
+<h2 class="anchored" data-anchor-id="csv-output">CSV output</h2>
+<p>The output file consists of the following pieces of information:</p>
+<ul>
+<li><p>The full set of configuration options available for the <code>log_prob</code> method is reported at the beginning of the output file as CSV comments.</p></li>
+<li><p>Column headers, the first column is labelled <code>lp__</code>, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.</p></li>
+<li><p>Values which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).</p></li>
+</ul>
+<p>For example, if we have a file called params.json:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode json code-with-copy"><code class="sourceCode json"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="fu">{</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">"theta"</span> <span class="fu">:</span> <span class="dv">0</span><span class="er">.</span><span class="dv">1</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a><span class="fu">}</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>We can run the example model:</p>
+<pre><code>/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json</code></pre>
+<p>This yields</p>
+<pre><code># method = log_prob
+#   log_prob
+#     unconstrained_params =  (Default)
+#     constrained_params = params.json
+#     jacobian = true (Default)
+# id = 1 (Default)
+# data
+#   file = bernoulli.data.json
+# init = 2 (Default)
+# random
+#   seed = 2390820139 (Default)
+# output
+#   file = output.csv (Default)
+#   diagnostic_file =  (Default)
+#   refresh = 100 (Default)
+#   sig_figs = 8 (Default)
+#   profile_file = profile.csv (Default)
+# num_threads = 1 (Default)
+lp_,theta
+-7.856,1.8</code></pre>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/laplace_sample_config.html" class="pagination-link" aria-label="Laplace sampling">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Laplace sampling</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/diagnose_config.html" class="pagination-link" aria-label="Diagnosing HMC by Comparison of Gradients">
+        <span class="nav-page-text">Diagnosing HMC by Comparison of Gradients</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/log_prob_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/mcmc_config.html b/docs/2_39/cmdstan-guide/mcmc_config.html
new file mode 100644
index 000000000..d418b1d8a
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/mcmc_config.html
@@ -0,0 +1,1597 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>MCMC Sampling</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/optimize_config.html" rel="next">
+<link href="../cmdstan-guide/command_line_options.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="MCMC Sampling">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/warmup-epochs.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/mcmc_config.html">MCMC Sampling using Hamiltonian Monte Carlo</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#mcmc-config" id="toc-mcmc-config" class="nav-link active" data-scroll-target="#mcmc-config">MCMC Sampling using Hamiltonian Monte Carlo</a>
+  <ul class="collapse">
+  <li><a href="#running-the-sampler" id="toc-running-the-sampler" class="nav-link" data-scroll-target="#running-the-sampler">Running the sampler</a></li>
+  <li><a href="#mcmc_output_csv" id="toc-mcmc_output_csv" class="nav-link" data-scroll-target="#mcmc_output_csv">Stan CSV output file</a></li>
+  <li><a href="#iterations" id="toc-iterations" class="nav-link" data-scroll-target="#iterations">Iterations</a></li>
+  <li><a href="#adaptation" id="toc-adaptation" class="nav-link" data-scroll-target="#adaptation">Adaptation</a>
+  <ul class="collapse">
+  <li><a href="#step-size-optimization-configuration" id="toc-step-size-optimization-configuration" class="nav-link" data-scroll-target="#step-size-optimization-configuration">Step size optimization configuration</a></li>
+  <li><a href="#warmup-schedule-configuration" id="toc-warmup-schedule-configuration" class="nav-link" data-scroll-target="#warmup-schedule-configuration">Warmup schedule configuration</a></li>
+  </ul></li>
+  <li><a href="#algorithm" id="toc-algorithm" class="nav-link" data-scroll-target="#algorithm">Algorithm</a>
+  <ul class="collapse">
+  <li><a href="#samples-from-a-set-of-fixed-parameters" id="toc-samples-from-a-set-of-fixed-parameters" class="nav-link" data-scroll-target="#samples-from-a-set-of-fixed-parameters">Samples from a set of fixed parameters</a></li>
+  <li><a href="#hmc-samplers" id="toc-hmc-samplers" class="nav-link" data-scroll-target="#hmc-samplers">HMC samplers</a></li>
+  </ul></li>
+  <li><a href="#sampler-diag-file" id="toc-sampler-diag-file" class="nav-link" data-scroll-target="#sampler-diag-file">Sampler diagnostic file</a></li>
+  <li><a href="#multi-chain-sampling" id="toc-multi-chain-sampling" class="nav-link" data-scroll-target="#multi-chain-sampling">Running multiple chains</a>
+  <ul class="collapse">
+  <li><a href="#sampler-num-chains" id="toc-sampler-num-chains" class="nav-link" data-scroll-target="#sampler-num-chains">Using the num_chains argument to run multiple chains</a></li>
+  </ul></li>
+  <li><a href="#summarizing-sampler-outputs-with-stansummary" id="toc-summarizing-sampler-outputs-with-stansummary" class="nav-link" data-scroll-target="#summarizing-sampler-outputs-with-stansummary">Summarizing sampler output(s) with <code>stansummary</code></a></li>
+  <li><a href="#old-multichain" id="toc-old-multichain" class="nav-link" data-scroll-target="#old-multichain">Examples - older parallelism</a>
+  <ul class="collapse">
+  <li><a href="#running-multiple-chains-with-a-specified-rng-seed" id="toc-running-multiple-chains-with-a-specified-rng-seed" class="nav-link" data-scroll-target="#running-multiple-chains-with-a-specified-rng-seed">Running multiple chains with a specified RNG seed</a></li>
+  <li><a href="#changing-the-default-warmup-and-sampling-iterations" id="toc-changing-the-default-warmup-and-sampling-iterations" class="nav-link" data-scroll-target="#changing-the-default-warmup-and-sampling-iterations">Changing the default warmup and sampling iterations</a></li>
+  <li><a href="#saving-warmup-draws" id="toc-saving-warmup-draws" class="nav-link" data-scroll-target="#saving-warmup-draws">Saving warmup draws</a></li>
+  <li><a href="#initializing-parameters" id="toc-initializing-parameters" class="nav-link" data-scroll-target="#initializing-parameters">Initializing parameters</a></li>
+  <li><a href="#specifying-the-metric-and-stepsize" id="toc-specifying-the-metric-and-stepsize" class="nav-link" data-scroll-target="#specifying-the-metric-and-stepsize">Specifying the metric and stepsize</a></li>
+  <li><a href="#changing-the-nuts-hmc-adaptation-parameters" id="toc-changing-the-nuts-hmc-adaptation-parameters" class="nav-link" data-scroll-target="#changing-the-nuts-hmc-adaptation-parameters">Changing the NUTS-HMC adaptation parameters</a></li>
+  <li><a href="#increasing-the-tree-depth" id="toc-increasing-the-tree-depth" class="nav-link" data-scroll-target="#increasing-the-tree-depth">Increasing the tree-depth</a></li>
+  <li><a href="#capturing-hamiltonian-diagnostics-and-gradients" id="toc-capturing-hamiltonian-diagnostics-and-gradients" class="nav-link" data-scroll-target="#capturing-hamiltonian-diagnostics-and-gradients">Capturing Hamiltonian diagnostics and gradients</a></li>
+  <li><a href="#suppressing-progress-updates-to-the-console" id="toc-suppressing-progress-updates-to-the-console" class="nav-link" data-scroll-target="#suppressing-progress-updates-to-the-console">Suppressing progress updates to the console</a></li>
+  <li><a href="#everything-example" id="toc-everything-example" class="nav-link" data-scroll-target="#everything-example">Everything example</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/mcmc_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/mcmc_config.html">MCMC Sampling using Hamiltonian Monte Carlo</a></li></ol></nav></header>
+
+
+
+
+<section id="mcmc-config" class="level1">
+<h1>MCMC Sampling using Hamiltonian Monte Carlo</h1>
+<p>The <code>sample</code> method provides Bayesian inference over the model conditioned on data using Hamiltonian Monte Carlo (HMC) sampling. By default, the inference engine used is the No-U-Turn sampler (NUTS), an adaptive form of Hamiltonian Monte Carlo sampling. For details on HMC and NUTS, see the Stan Reference Manual chapter on <a href="https://mc-stan.org/docs/reference-manual/mcmc.html">MCMC Sampling</a>.</p>
+<section id="running-the-sampler" class="level2">
+<h2 class="anchored" data-anchor-id="running-the-sampler">Running the sampler</h2>
+<p>To generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument <code>sample</code> or <code>method=sample</code> together with the input data. The executable can be run from any directory.</p>
+<p>The full set of configuration options available for the <code>sample</code> method is available by using the <code>sample help-all</code> subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.</p>
+<p>Here, we run it in the directory which contains the Stan program and input data, <code>&lt;cmdstan-home&gt;/examples/bernoulli</code>:</p>
+<pre><code>&gt; cd examples/bernoulli
+&gt; ls
+  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan</code></pre>
+<p>To execute sampling of the model under Linux or Mac, use:</p>
+<pre><code>&gt; ./bernoulli sample data file=bernoulli.data.json</code></pre>
+<p>In Windows, the <code>./</code> prefix is not needed:</p>
+<pre><code>&gt; bernoulli.exe sample data file=bernoulli.data.json</code></pre>
+<p>The output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:</p>
+<pre><code>method = sample (Default)
+  sample
+    num_samples = 1000 (Default)
+    num_warmup = 1000 (Default)
+    save_warmup = false (Default)
+    thin = 1 (Default)
+    adapt
+      engaged = true (Default)
+      gamma = 0.050000000000000003 (Default)
+      delta = 0.80000000000000004 (Default)
+      kappa = 0.75 (Default)
+      t0 = 10 (Default)
+      init_buffer = 75 (Default)
+      term_buffer = 50 (Default)
+      window = 25 (Default)
+      save_metric = false (Default)
+    algorithm = hmc (Default)
+      hmc
+        engine = nuts (Default)
+          nuts
+            max_depth = 10 (Default)
+        metric = diag_e (Default)
+        metric_file =  (Default)
+        stepsize = 1 (Default)
+        stepsize_jitter = 0 (Default)
+    num_chains = 1 (Default)
+id = 0 (Default)
+data
+  file = bernoulli.data.json
+init = 2 (Default)
+random
+  seed = 3252652196 (Default)
+output
+  file = output.csv (Default)
+  diagnostic_file =  (Default)
+  refresh = 100 (Default)</code></pre>
+<p>After the configuration has been displayed, a short timing message is given.</p>
+<pre><code>Gradient evaluation took 1.2e-05 seconds
+1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.
+Adjust your expectations accordingly!</code></pre>
+<p>Next, the sampler reports the iteration number, reporting the percentage complete.</p>
+<pre><code>Iteration:    1 / 2000 [  0%]  (Warmup)
+...
+Iteration: 2000 / 2000 [100%]  (Sampling)</code></pre>
+<p>Finally, the sampler reports timing information:</p>
+<pre><code> Elapsed Time: 0.007 seconds (Warm-up)
+               0.017 seconds (Sampling)
+               0.024 seconds (Total)</code></pre>
+</section>
+<section id="mcmc_output_csv" class="level2">
+<h2 class="anchored" data-anchor-id="mcmc_output_csv">Stan CSV output file</h2>
+<p>Each execution of the model results in draws from a single Markov chain being written to a file in <a href="../cmdstan-guide/stan_csv_apdx.html">comma-separated value (CSV) format</a>. The default name of the output file is <code>output.csv</code>.</p>
+<p>The first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (<code>#</code>)).</p>
+<p>When the example model <code>bernoulli.stan</code> is run via the command line with all default arguments, the following configuration is displayed:</p>
+<pre><code># stan_version_major = 2
+# stan_version_minor = 23
+# stan_version_patch = 0
+# model = bernoulli_model
+# method = sample (Default)
+#   sample
+#     num_samples = 1000 (Default)
+#     num_warmup = 1000 (Default)
+#     save_warmup = false (Default)
+#     thin = 1 (Default)
+#     adapt
+#       engaged = 1 (Default)
+#       gamma = 0.050000 (Default)
+#       delta = 0.800000 (Default)
+#       kappa = 0.750000 (Default)
+#       t0 = 10.000000 (Default)
+#       init_buffer = 75 (Default)
+#       term_buffer = 50 (Default)
+#       window = 25 (Default)
+#       save_metric = false (Default)
+#     algorithm = hmc (Default)
+#       hmc
+#         engine = nuts (Default)
+#           nuts
+#             max_depth = 10 (Default)
+#         metric = diag_e (Default)
+#         metric_file =  (Default)
+#         stepsize = 1.000000 (Default)
+#         stepsize_jitter = 0.000000 (Default)
+#     num_chains = 1 (Default)
+# output
+#   file = output.csv (Default)
+#   diagnostic_file =  (Default)
+#   refresh = 100 (Default)</code></pre>
+<p>This is followed by a CSV header indicating the names of the values sampled.</p>
+<pre><code>lp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta</code></pre>
+<p>The first output columns report the HMC sampler information:</p>
+<ul>
+<li><code>lp__</code> - the total log probability density (up to an additive constant) at each sample</li>
+<li><code>accept_stat__</code> - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory</li>
+<li><code>stepsize__</code> - integrator step size</li>
+<li><code>treedepth__</code> - depth of tree used by NUTS (NUTS sampler)</li>
+<li><code>n_leapfrog__</code> - number of leapfrog calculations (NUTS sampler)</li>
+<li><code>divergent__</code> - has value <code>1</code> if trajectory diverged, otherwise <code>0</code>. (NUTS sampler)</li>
+<li><code>energy__</code> - value of the Hamiltonian</li>
+<li><code>int_time__</code> - total integration time (static HMC sampler)</li>
+</ul>
+<p>Because the above header is from the NUTS sampler, it has columns <code>treedepth__</code>, <code>n_leapfrog__</code>, and <code>divergent__</code> and doesn’t have column <code>int_time__</code>. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, <code>theta</code>.</p>
+<p>The header line is written to the output file before warmup begins. If option <code>save_warmup</code> is set to <code>true</code>, the warmup draws are output directly after the header. The total number of warmup draws saved is <code>num_warmup</code> divided by <code>thin</code>, rounded up (i.e., <code>ceiling</code>).</p>
+<p>Following the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:</p>
+<pre><code># Adaptation terminated
+# Step size = 0.884484
+# Diagonal elements of inverse mass matrix:
+# 0.535006</code></pre>
+<p>The default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter <code>theta</code>.</p>
+<p>Draws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.</p>
+<pre><code>-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853
+-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295
+-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299
+-6.88712,1,0.884484,1,1,0,7.02101,0.188229
+-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596
+...</code></pre>
+<p>The output ends with timing details:</p>
+<pre><code>#  Elapsed Time: 0.007 seconds (Warm-up)
+#                0.017 seconds (Sampling)
+#                0.024 seconds (Total)</code></pre>
+</section>
+<section id="iterations" class="level2">
+<h2 class="anchored" data-anchor-id="iterations">Iterations</h2>
+<p>At every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters <code>metric</code> and <code>stepsize</code> in order to efficiently sample from <em>typical set</em>, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.</p>
+<p>The following keyword-value arguments control the total number of iterations:</p>
+<ul>
+<li><code>num_samples</code></li>
+<li><code>num_warmup</code></li>
+<li><code>save_warmup</code></li>
+<li><code>thin</code></li>
+</ul>
+<p>The values for arguments <code>num_samples</code> and <code>num_warmup</code> must be a non-negative integer. The default value for both is <span class="math inline">\(1000\)</span>.</p>
+<p>For well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.</p>
+<p>The number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is <span class="math inline">\(\sqrt{N}\)</span>; therefore every additional decimal place of accuracy increases this by a factor of 10.</p>
+<p>Argument <code>save_warmup</code> takes values <code>false</code> or <code>true</code>. The default value is <code>false</code>, i.e., warmup draws are not saved to the output file. When the value is <code>true</code>, the warmup draws are written to the CSV output file directly after the CSV header line.</p>
+<p>Argument <code>thin</code> controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce <em>anticorrelated</em> draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.</p>
+<p>The value of argument <code>thin</code> must be a positive integer. When thin is set to value <span class="math inline">\(N\)</span>, every <span class="math inline">\(N^{th}\)</span> iteration is written to the output file. Should the value of <code>thin</code> exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of <code>thin</code> equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When <code>num_sampling=M</code> and <code>thin=N</code>, the number of iterations written to the output CSV file will be <code>ceiling(M/N)</code>. If <code>save_warmup=true</code>, thinning is applied to the warmup iterations as well.</p>
+</section>
+<section id="adaptation" class="level2">
+<h2 class="anchored" data-anchor-id="adaptation">Adaptation</h2>
+<p>The <code>adapt</code> keyword is used to specify non-default options for the sampler adaptation schedule and settings.</p>
+<p>Adaptation can be turned off by setting sub-argument <code>engaged</code> to value <code>false</code>. If <code>engaged=false</code>, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is <code>engaged=1</code>, this keyword-value pair can be omitted from the command.</p>
+<p>There are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section <a href="https://mc-stan.org/docs/reference-manual/mcmc.html#hmc-algorithm-parameters">Automatic Parameter Tuning</a>.</p>
+<p>The boolean sub-argument <code>save_metric</code> was added in Stan version 2.34. When <code>save_metric=true</code>, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix <code>_metric.json</code>, e.g., if using the default output filename <code>output.csv</code>, the saved metric file will be <code>output_metric.json</code>. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument <code>metric_file</code>.</p>
+<section id="step-size-optimization-configuration" class="level3">
+<h3 class="anchored" data-anchor-id="step-size-optimization-configuration">Step size optimization configuration</h3>
+<p>The Stan User’s Guide section on <a href="https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature">model conditioning and curvature</a> provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on <a href="https://mc-stan.org/docs/reference-manual/mcmc.html#hmc-algorithm-parameters">HMC algorithm parameters</a> explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.</p>
+<p>The following keyword-value arguments control the settings used to optimize the step size:</p>
+<ul>
+<li><p><code>delta</code> - The target Metropolis acceptance rate. The default value is <span class="math inline">\(0.8\)</span>. Its value must be strictly between <span class="math inline">\(0\)</span> and <span class="math inline">\(1\)</span>. Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of <code>delta</code> will also allow some models that would otherwise get stuck to overcome their blockages. <br> Models with difficult posterior geometries may required increasing the <code>delta</code> argument closer to <span class="math inline">\(1\)</span>; we recommend first trying to raise it to <span class="math inline">\(0.9\)</span> or at most <span class="math inline">\(0.95\)</span>. Values about <span class="math inline">\(0.95\)</span> are strong indication of bad geometry; the better solution is to change the model geometry through <a href="https://mc-stan.org/docs/stan-users-guide/reparameterization.html">reparameterization</a> which could yield both more efficient and faster sampling.</p></li>
+<li><p><code>gamma</code> - Adaptation regularization scale. Must be a positive real number, default value is <span class="math inline">\(0.05\)</span>. This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.</p></li>
+<li><p><code>kappa</code> - Adaptation relaxation exponent. Must be a positive real number, default value is <span class="math inline">\(0.75\)</span>. This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.</p></li>
+<li><p><code>t_0</code> - Adaptation iteration offset. Must be a positive real number, default value is <span class="math inline">\(10\)</span>. This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.</p></li>
+</ul>
+</section>
+<section id="warmup-schedule-configuration" class="level3">
+<h3 class="anchored" data-anchor-id="warmup-schedule-configuration">Warmup schedule configuration</h3>
+<p>When adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:</p>
+<ul>
+<li><code>init_buffer</code> - The number of iterations spent tuning the step size at the outset of adaptation.</li>
+<li><code>window</code> - The initial number of iterations devoted to tune the metric, will be doubled successively.</li>
+<li><code>term_buffer</code> - The number of iterations used to re-tune the step size once the metric has been tuned.</li>
+</ul>
+<p>The specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.</p>
+<p>The following figure is taken from the Stan Reference Manual, where label “I” correspond to <code>init_buffer</code>, the initial “II” corresponds to <code>window</code>, and the final “III” corresponds to <code>term_buffer</code>:</p>
+<p><strong>Warmup Epochs Figure.</strong> <a id="adaptation.figure"></a> <em>Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.</em></p>
+<p><img src="img/warmup-epochs.png" class="img-fluid"></p>
+</section>
+</section>
+<section id="algorithm" class="level2">
+<h2 class="anchored" data-anchor-id="algorithm">Algorithm</h2>
+<p>The <code>algorithm</code> keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: <code>hmc</code>, which generates from an HMC-driven Markov chain; and <code>fixed_param</code> which generates a new sample without changing the state of the Markov chain. The default argument is <code>algorithm=hmc</code>.</p>
+<section id="samples-from-a-set-of-fixed-parameters" class="level3">
+<h3 class="anchored" data-anchor-id="samples-from-a-set-of-fixed-parameters">Samples from a set of fixed parameters</h3>
+<p>If a model doesn’t specify any parameters, then argument <code>algorithm=fixed_param</code> is mandatory.</p>
+<p>The fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.</p>
+</section>
+<section id="hmc-samplers" class="level3">
+<h3 class="anchored" data-anchor-id="hmc-samplers">HMC samplers</h3>
+<p>All HMC algorithms have three parameters:</p>
+<ul>
+<li>step size</li>
+<li>metric</li>
+<li>integration time - the number of steps taken along the Hamiltonian trajectory</li>
+</ul>
+<p>See the Stan Reference Manual section on <a href="https://mc-stan.org/docs/reference-manual/mcmc.html#hmc-algorithm-parameters">HMC algorithm parameters</a> for further details.</p>
+<section id="step-size" class="level4">
+<h4 class="anchored" data-anchor-id="step-size">Step size</h4>
+<p>The HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.</p>
+<p>The following keyword-value arguments control the step size:</p>
+<ul>
+<li><p><code>stepsize</code> - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is <span class="math inline">\(1\)</span>.</p></li>
+<li><p><code>stepsize_jitter</code> - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between <span class="math inline">\(0\)</span> and <span class="math inline">\(1\)</span>. The default value is <span class="math inline">\(0\)</span>. Setting <code>stepsize_jitter</code> to <span class="math inline">\(1\)</span> causes step sizes to be selected in the range of <span class="math inline">\(0\)</span> to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.</p></li>
+</ul>
+</section>
+<section id="metric" class="level4">
+<h4 class="anchored" data-anchor-id="metric">Metric</h4>
+<p>All HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a <em>mass matrix</em> or, more formally, a <em>metric</em> <span class="citation" data-cites="Betancourt:2017">Betancourt (<a href="#ref-Betancourt:2017" role="doc-biblioref">2017</a>)</span>.</p>
+<p>The <code>metric</code> argument specifies the choice of Euclidean HMC implementations:</p>
+<ul>
+<li><code>metric=unit</code> specifies unit metric (diagonal matrix of ones).</li>
+<li><code>metric=diag_e</code> specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.</li>
+<li><code>metric=dense_e</code> specifies a dense metric (a dense, symmetric positive definite matrix).</li>
+</ul>
+<p>By default, the metric is estimated during warmup. However, when <code>metric=diag_e</code> or <code>metric=dense_e</code>, an initial guess for the metric can be specified with the <code>metric_file</code> argument whose value is the filepath to a JSON or Rdump file which contains a single variable <code>inv_metric</code>. For a <code>diag_e</code> metric the <code>inv_metric</code> value must be a vector of positive values, one for each parameter in the system. For a <code>dense_e</code> metric, <code>inv_metric</code> value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.</p>
+<p>The <code>metric_file</code> option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.</p>
+<p>If adaptation is disabled, both the <code>metric_file</code> and <code>stepsize</code> arguments should be specified.</p>
+</section>
+<section id="integration-time" class="level4">
+<h4 class="anchored" data-anchor-id="integration-time">Integration time</h4>
+<p>The total integration time is determined by the argument <code>engine</code> which take possible values:</p>
+<ul>
+<li><code>nuts</code> - the No-U-Turn Sampler which dynamically determines the optimal integration time.</li>
+<li><code>static</code> - an HMC sampler which uses a user-specified integration time.</li>
+</ul>
+<p>The default argument is <code>engine=nuts</code>.</p>
+<p>The NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.</p>
+<p>When <code>engine=nuts</code>, the subargument <code>max_depth</code> can be used to control the depth of the tree. The default argument is <code>max_depth=10</code>. In the case where a model has a difficult posterior from which to sample, <code>max_depth</code> should be increased to ensure that that the NUTS tree can grow as large as necessary.</p>
+<p>When the argument <code>engine=static</code> is specified, the user must specify the integration time via keyword <code>int_time</code> which takes as a value a positive number. The default value is <span class="math inline">\(2\pi\)</span>.</p>
+</section>
+</section>
+</section>
+<section id="sampler-diag-file" class="level2">
+<h2 class="anchored" data-anchor-id="sampler-diag-file">Sampler diagnostic file</h2>
+<p>The <code>output</code> keyword sub-argument <code>diagnostic_file=&lt;filepath&gt;</code> specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.</p>
+</section>
+<section id="multi-chain-sampling" class="level2">
+<h2 class="anchored" data-anchor-id="multi-chain-sampling">Running multiple chains</h2>
+<p>A Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.</p>
+<p>The preferred way of using multiple chains is to run them all from the same executable using the <code>num_chains</code> argument. There is also the option to use the <a href="#old-multichain">Unix or DOS shell to run multiple executables</a>.</p>
+<section id="sampler-num-chains" class="level3">
+<h3 class="anchored" data-anchor-id="sampler-num-chains">Using the num_chains argument to run multiple chains</h3>
+<p>The <code>num_chains</code> argument can be used for all of Stan’s samplers with the exception of the <code>static</code> HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with <code>STAN_THREADS=true</code>, these will either run in parallel or one after the other.</p>
+<p>When <code>num_chains</code> is greather than <code>1</code> (the default), arguments related to filenames (e.g.&nbsp;<code>output file=</code>, <code>init=</code>) can accept a comma separated list of values, one per each chain.</p>
+<p>For example, <code>sample</code> will specify the names of the three chain’s output files.</p>
+<pre><code>./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv</code></pre>
+<p>This will write the output in <code>output_1.csv</code>, <code>output_2.csv</code>, <code>output_3.csv</code>.</p>
+<p>If the model was not compiled with <code>STAN_THREADS=true</code>, the above command will run 3 chains sequentially.</p>
+<p>If the model was compiled with <code>STAN_THREADS=true</code>, the chains can run in parallel, with the <code>num_threads</code> argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (<code>map_rect</code> or <code>reduce_sum</code> calls), the below command will run 3 chains in parallel, provided there are cores available:</p>
+<pre><code>./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4</code></pre>
+<p>If the model uses within-chain parallelization (<code>map_rect</code> or <code>reduce_sum</code> calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the <a href="https://www.intel.com/content/www/us/en/docs/onetbb/developer-guide-api-reference/2021-11/how-task-scheduler-works.html">Threading Building Blocks scheduler</a>.</p>
+<pre><code>./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16</code></pre>
+<section id="legacy-filename-behavior" class="level4">
+<h4 class="anchored" data-anchor-id="legacy-filename-behavior">Legacy filename behavior</h4>
+<p>If a comma separated list is not used, the <code>num_chains</code> argument changes the normal meanings of filename arguments when it is greater than <code>1</code> (the default). They are now interpreted as a “template” which is used for each chain.</p>
+<p>For example, when <code>num_chains=2</code>, the argument <code>output file=foo.csv</code> <em>no longer produces a file <code>foo.csv</code></em>, but instead produces two files, <code>foo_1.csv</code> and <code>foo_2.csv</code>. If you also supply <code>id=5</code>, the files produced will be <code>foo_5.csv</code> and <code>foo_6.csv</code> – <code>id=5</code> gives the id of the first chain, and the remaining chains are sequential from there.</p>
+<p>This also applies to input files, like those used for initialization. For example, if <code>num_chains=3</code> and <code>init=bar.json</code> will first look for <code>bar_1.json</code>. If it exists, it will use <code>bar_1.json</code> for the first chain, <code>bar_2.json</code> for the second, and so on. If <code>bar_1.json</code> does not exist, it falls back to looking for <code>bar.json</code>, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the <code>id</code> argument, which defaults to <code>1</code>.</p>
+<p>For example, this shorthand is equivalent to the example given above:</p>
+<pre><code>./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv</code></pre>
+<p>A suffix with the chain id is appended to the provided output filename (<code>output.csv</code> in the above command), so this will also produce files <code>output_1.csv</code>, <code>output_2.csv</code>, <code>output_3.csv</code>.</p>
+</section>
+</section>
+</section>
+<section id="summarizing-sampler-outputs-with-stansummary" class="level2">
+<h2 class="anchored" data-anchor-id="summarizing-sampler-outputs-with-stansummary">Summarizing sampler output(s) with <code>stansummary</code></h2>
+<p>The <a href="../cmdstan-guide/stansummary.html"><code>stansummary</code> utility</a> processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file <code>stansummary</code> reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and <span class="math inline">\(\hat{R}\)</span> values.</p>
+<p>To run <code>stansummary</code> on the output files generated by the for loop above, by the above run of the <code>bernoulli</code> model on Mac or Linux:</p>
+<pre><code>&lt;cmdstan-home&gt;/bin/stansummary output_*.csv</code></pre>
+<p>On Windows, use backslashes to call the <code>stansummary.exe</code>.</p>
+<pre><code>&lt;cmdstan-home&gt;\bin\stansummary.exe output_*.csv</code></pre>
+<p>The stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable <code>theta</code>:</p>
+<pre><code>Inference for Stan model: bernoulli_model
+4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.
+
+Warmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total
+Sampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total
+
+                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat
+
+lp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0
+accept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0
+stepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan
+treedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0
+n_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0
+divergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan
+energy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0
+
+theta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0
+
+Samples were drawn using hmc with nuts.
+For each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)
+and for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.
+At convergence R_hat will be very close to 1.00.</code></pre>
+<p>In this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for <code>theta</code> reflect the uncertainty in our estimate, due to the small amount of data, given the prior of <code>beta(1, 1)</code></p>
+</section>
+<section id="old-multichain" class="level2">
+<h2 class="anchored" data-anchor-id="old-multichain">Examples - older parallelism</h2>
+<p><strong>Note</strong>: Many of these examples can be simplified by using the <a href="#sampler-num-chains"><code>num_chains</code> argument</a>.</p>
+<p>When the <code>num_chains</code> argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.</p>
+<p>To run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell <code>for</code> loop to set up index variables needed to identify each chain and its outputs.</p>
+<p>On MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:</p>
+<pre><code>for NAME [in LIST]; do COMMANDS; done</code></pre>
+<p>The list can be a simple sequence of numbers, or you can use the shell expansion syntax <code>{1..N}</code> which expands to the sequence from <span class="math inline">\(1\)</span> to <span class="math inline">\(N\)</span>, e.g.&nbsp;<code>{1..4}</code> expands to <code>1 2 3 4</code>. Note that the expression <code>{1..N}</code> cannot contain spaces.</p>
+<p>To run 4 chains for the example bernoulli model on MacOS or Linux:</p>
+<pre><code>&gt; for i in {1..4}
+    do
+      ./bernoulli sample data file=bernoulli.data.json \
+      output file=output_${i}.csv
+    done</code></pre>
+<p>The backslash (<code>\</code>) indicates a line continuation in Unix. The expression <code>${i}</code> substitutes in the value of loop index variable <code>i</code>. To run chains in parallel, put an ampersand (<code>&amp;</code>) at the end of the nested sampler command:</p>
+<pre><code>&gt; for i in {1..4}
+    do
+      ./bernoulli sample data file=bernoulli.data.json \
+      output file=output_${i}.csv &amp;
+    done</code></pre>
+<p>This pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.</p>
+<p>On Windows, the DOS <a href="https://www.windows-commandline.com/windows-for-loop-examples/">for-loop syntax</a> is one of:</p>
+<pre><code>for %i in (SET) do COMMAND COMMAND-ARGUMENTS
+for /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS</code></pre>
+<p>To run 4 chains in parallel on Windows:</p>
+<pre><code>&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^
+                                    data file=bernoulli.data.json my_data ^
+                                    output file=output_%i.csv</code></pre>
+<p>The caret (<code>^</code>) indicates a line continuation in DOS. The expression <code>%i</code> is the loop index.</p>
+<p>In the following extended examples, we focus on just the nested sampler command for Unix.</p>
+<section id="running-multiple-chains-with-a-specified-rng-seed" class="level3">
+<h3 class="anchored" data-anchor-id="running-multiple-chains-with-a-specified-rng-seed">Running multiple chains with a specified RNG seed</h3>
+<p>For reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.</p>
+<p>The RNG seed is specified by <code>random seed=&lt;int&gt;</code> and the offset is specified by <code>id=&lt;loop index&gt;</code>, so the call to the sampler is:</p>
+<pre><code>./my_model sample data file=my_model.data.json \
+            output file=output_${i}.csv \
+            random seed=12345 id=${i}</code></pre>
+</section>
+<section id="changing-the-default-warmup-and-sampling-iterations" class="level3">
+<h3 class="anchored" data-anchor-id="changing-the-default-warmup-and-sampling-iterations">Changing the default warmup and sampling iterations</h3>
+<p>The warmup and sampling iteration keyword-value arguments must follow the <code>sample</code> keyword. The call to the sampler which overrides the default warmup and sampling iterations is:</p>
+<pre><code>./my_model sample num_warmup=500 num_sampling=500 \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+</section>
+<section id="saving-warmup-draws" class="level3">
+<h3 class="anchored" data-anchor-id="saving-warmup-draws">Saving warmup draws</h3>
+<p>To save warmup draws as part of the Stan CSV output file, use the keyword-value argument <code>save_warmup=true</code>. This must be grouped with the other <code>sample</code> keyword sub-arguments.</p>
+<pre><code>./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+</section>
+<section id="initializing-parameters" class="level3">
+<h3 class="anchored" data-anchor-id="initializing-parameters">Initializing parameters</h3>
+<p>By default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range <span class="math inline">\([{-2}, 2]\)</span>. To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument <code>init=&lt;filepath&gt;</code>:</p>
+<pre><code>./my_model sample init=my_param_inits.json data file=my_model.data.json \
+           output file=output_${i}.csv</code></pre>
+<p>To verify that the specified values will be used by the sampler, you can run the sampler with option <code>algorithm=fixed_param</code>, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting <code>num_warmp=0</code> and <code>num_samples=1</code> saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.</p>
+<p>For example, if we run the example Bernoulli model with specified initial value for parameter “theta”:</p>
+<pre><code>{ "theta" : 0.5 }</code></pre>
+<p>via command:</p>
+<pre><code>./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \
+            init=bernoulli.init.json data file=bernoulli.data.json</code></pre>
+<p>The resulting output CSV file contains a single draw:</p>
+<pre><code>lp__,accept_stat__,theta
+0,0,0.5
+#
+#  Elapsed Time: 0 seconds (Warm-up)
+#                0 seconds (Sampling)
+#                0 seconds (Total)
+#</code></pre>
+</section>
+<section id="specifying-the-metric-and-stepsize" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-the-metric-and-stepsize">Specifying the metric and stepsize</h3>
+<p>An initial estimate for the metric can be specified with the <code>metric_file</code> argument whose value is the filepath to a JSON or Rdump file which contains a variable <code>inv_metric</code>. The <code>metric_file</code> option can be used with and without adaptation enabled.</p>
+<p>By default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file <code>bernoulli.diag_e.json</code>, contents</p>
+<pre><code>{ "inv_metric" : [0.296291] }</code></pre>
+<p>can be used as the initial metric as follows:</p>
+<pre><code>../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+<p>If adaptation is disabled, both the <code>metric_file</code> and <code>stepsize</code> arguments should be specified.</p>
+<pre><code>../my_model sample adapt engaged=false \
+            algorithm=hmc stepsize=0.9 \
+            metric_file=bernoulli.diag_e.json \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+<p>The resulting output CSV file will contain the following set of comment lines:</p>
+<pre><code># Adaptation terminated
+# Step size = 0.9
+# Diagonal elements of inverse mass matrix:
+# 0.296291</code></pre>
+<p>As of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument <code>save_metric</code>, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the <code>stepsize</code> argument as well as the <code>metric_file</code> arguments; the former is the value of the <code>stepsize</code> element in the saved metric file, and the later is the metric file path.</p>
+</section>
+<section id="changing-the-nuts-hmc-adaptation-parameters" class="level3">
+<h3 class="anchored" data-anchor-id="changing-the-nuts-hmc-adaptation-parameters">Changing the NUTS-HMC adaptation parameters</h3>
+<p>The keyword-value arguments for these settings are grouped together under the <code>adapt</code> keyword which itself is a sub-argument of the <code>sample</code> keyword.</p>
+<p>Models with difficult posterior geometries may required increasing the <code>delta</code> argument closer to <span class="math inline">\(1\)</span>.</p>
+<pre><code>./my_model sample adapt delta=0.95 \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+<p>To skip adaptation altogether, use the keyword-value argument <code>engaged=false</code>. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.</p>
+<pre><code>../my_model sample adapt engaged=false \
+            algorithm=hmc stepsize=0.9 \
+            metric_file=bernoulli.diag_e.json \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+<p>Even with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the <a href="https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature">typical set</a>.</p>
+<p>To skip warmup altogether requires specifying both <code>num_warmup=0</code> and <code>adapt engaged=false</code>.</p>
+<pre><code>../my_model sample num_warmup=0 adapt engaged=false \
+            algorithm=hmc stepsize=0.9 \
+            metric_file=bernoulli.diag_e.json \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+</section>
+<section id="increasing-the-tree-depth" class="level3">
+<h3 class="anchored" data-anchor-id="increasing-the-tree-depth">Increasing the tree-depth</h3>
+<p>Models with difficult posterior geometries may required increasing the <code>max_depth</code> argument from its default value <span class="math inline">\(10\)</span>. This requires specifying a series of keyword-argument pairs:</p>
+<pre><code>./my_model sample adapt delta=0.95 \
+            algorithm=hmc engine=nuts max_depth=15 \
+            data file=my_model.data.json \
+            output file=output_${i}.csv</code></pre>
+</section>
+<section id="capturing-hamiltonian-diagnostics-and-gradients" class="level3">
+<h3 class="anchored" data-anchor-id="capturing-hamiltonian-diagnostics-and-gradients">Capturing Hamiltonian diagnostics and gradients</h3>
+<p>The <code>output</code> keyword sub-argument <code>diagnostic_file=&lt;filepath&gt;</code> write the sampler parameters and gradients of all model parameters for each draw to a CSV file:</p>
+<pre><code>./my_model sample data file=my_model.data.json \
+            output file=output_${i}.csv \
+            diagnostic_file=diagnostics_${i}.csv</code></pre>
+</section>
+<section id="suppressing-progress-updates-to-the-console" class="level3">
+<h3 class="anchored" data-anchor-id="suppressing-progress-updates-to-the-console">Suppressing progress updates to the console</h3>
+<p>The <code>output</code> keyword sub-argument <code>refresh=&lt;int&gt;</code> specifies the number of iterations between progress messages written to the terminal window. The default value is <span class="math inline">\(100\)</span> iterations. The progress updates look like:</p>
+<pre><code>Iteration:    1 / 2000 [  0%]  (Warmup)
+Iteration:  100 / 2000 [  5%]  (Warmup)
+Iteration:  200 / 2000 [ 10%]  (Warmup)
+Iteration:  300 / 2000 [ 15%]  (Warmup)</code></pre>
+<p>For simple models which fit quickly, such updates can be annoying; to suppress them altogether, set <code>refresh=0</code>. This only turns off the <code>Iteration:</code> messages; the configuration and timing information are still written to the terminal.</p>
+<pre><code>./my_model sample data file=my_model.data.json \
+            output file=output_${i}.csv \
+            refresh=0</code></pre>
+<p>For complicated models which take a long time to fit, setting the refresh rate to a low number, e.g.&nbsp;<span class="math inline">\(10\)</span> or even <span class="math inline">\(1\)</span>, provides a way to more closely monitor the sampler.</p>
+</section>
+<section id="everything-example" class="level3">
+<h3 class="anchored" data-anchor-id="everything-example">Everything example</h3>
+<p>The CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:</p>
+<pre><code>./bernoulli sample data file=bernoulli.data.json adapt delta=0.95
+adapt is either mistyped or misplaced.
+Perhaps you meant one of the following valid configurations?
+  method=sample sample adapt
+  method=variational variational adapt
+Failed to parse arguments, terminating Stan</code></pre>
+<p>The following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.</p>
+<pre><code>./my_model sample num_warmup=2000 \
+           init=my_param_inits.json \
+           adapt delta=0.95 init_buffer=100 \
+           window=50 term_buffer=100 \
+           algorithm=hmc engine=nuts max_depth=15 \
+           metric=dense_e metric_file=my_metric.json \
+           stepsize=0.6555 \
+           data file=my_model.data.json \
+           output file=output_${i}.csv refresh=10 \
+           random seed=12345 id=${i}</code></pre>
+<p>The keywords <code>sample</code>, <code>data</code>, <code>output</code>, and <code>random</code> are the top-level argument groups. Within the <code>sample</code> config arguments, the keyword <code>adapt</code> groups the adaptation algorithm parameters and the keyword-value <code>algorithm=hmc</code> groups the NUTS-HMC parameters.</p>
+<p>The top-level groups can be freely ordered with respect to one another. The following is also a valid command:</p>
+<pre><code>./my_model random seed=12345 id=${i} \
+           data file=my_model.data.json \
+           output file=output_${i}.csv refresh=10 \
+           sample num_warmup=2000 \
+           init=my_param_inits.json \
+           algorithm=hmc engine=nuts max_depth=15 \
+           metric=dense_e metric_file=my_metric.json \
+           stepsize=0.6555 \
+           adapt delta=0.95 init_buffer=100 \
+           window=50 term_buffer=100</code></pre>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Betancourt:2017" class="csl-entry" role="listitem">
+Betancourt, Michael. 2017. <span>“A Conceptual Introduction to <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>arXiv</em> 1701.02434. <a href="https://arxiv.org/abs/1701.02434">https://arxiv.org/abs/1701.02434</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/command_line_options.html" class="pagination-link" aria-label="Command-Line Interface Overview">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Command-Line Interface Overview</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/optimize_config.html" class="pagination-link" aria-label="Optimization">
+        <span class="nav-page-text">Optimization</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/mcmc_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/optimize_config.html b/docs/2_39/cmdstan-guide/optimize_config.html
new file mode 100644
index 000000000..56a348a61
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/optimize_config.html
@@ -0,0 +1,1234 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Optimization</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/pathfinder_config.html" rel="next">
+<link href="../cmdstan-guide/mcmc_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Optimization">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/optimize_config.html">Optimization</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#optimization" id="toc-optimization" class="nav-link active" data-scroll-target="#optimization">Optimization</a>
+  <ul class="collapse">
+  <li><a href="#meaning-of-the-converged__-column" id="toc-meaning-of-the-converged__-column" class="nav-link" data-scroll-target="#meaning-of-the-converged__-column">Meaning of the <code>converged__</code> column</a></li>
+  <li><a href="#jacobian-adjustments" id="toc-jacobian-adjustments" class="nav-link" data-scroll-target="#jacobian-adjustments">Jacobian adjustments</a></li>
+  <li><a href="#optimization-algorithms" id="toc-optimization-algorithms" class="nav-link" data-scroll-target="#optimization-algorithms">Optimization algorithms</a></li>
+  <li><a href="#the-quasi-newton-optimizers" id="toc-the-quasi-newton-optimizers" class="nav-link" data-scroll-target="#the-quasi-newton-optimizers">The quasi-Newton optimizers</a></li>
+  <li><a href="#the-newton-optimizer" id="toc-the-newton-optimizer" class="nav-link" data-scroll-target="#the-newton-optimizer">The Newton optimizer</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/optimize_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/optimize_config.html">Optimization</a></li></ol></nav></header>
+
+
+
+
+<section id="optimization" class="level1">
+<h1>Optimization</h1>
+<p>The CmdStan executable can run Stan’s optimization algorithms, which provide a deterministic method to find the posterior mode. If the posterior is not convex, there is no guarantee Stan will be able to find the global optimum as opposed to a local optimum of log probability.</p>
+<p>The full set of configuration options available for the <code>optimize</code> method is available by using the <code>optimize help-all</code> subcommand. The arguments with their requested values or defaults are also reported at the beginning of the optimizer console output and in the output CSV file’s comments.</p>
+<p>The executable does not need to be recompiled in order to switch from sampling to optimization, and the data input format is the same. The following is a minimal call to Stan’s optimizer using defaults for everything but the location of the data file.</p>
+<pre><code>&gt; ./bernoulli optimize data file=bernoulli.data.json</code></pre>
+<p>Executing this command prints both output to the console and to a CSV file.</p>
+<p>The first part of the console output reports on the configuration used. The above command uses all default configurations, therefore the optimizer used is the L-BFGS optimizer and its default initial stepsize and tolerances for monitoring convergence:</p>
+<pre><code> ./bernoulli optimize data file=bernoulli.data.json
+method = optimize
+  optimize
+    algorithm = lbfgs (Default)
+      lbfgs
+        init_alpha = 0.001 (Default)
+        tol_obj = 1e-12 (Default)
+        tol_rel_obj = 10000 (Default)
+        tol_grad = 1e-08 (Default)
+        tol_rel_grad = 1e+07 (Default)
+        tol_param = 1e-08 (Default)
+        history_size = 5 (Default)
+    jacobian = false (Default)
+    iter = 2000 (Default)
+    save_iterations = false (Default)
+id = 1 (Default)
+data
+  file = bernoulli.data.json
+init = 2 (Default)
+random
+  seed = 87122538 (Default)
+output
+  file = output.csv (Default)
+  diagnostic_file =  (Default)
+  refresh = 100 (Default)
+  sig_figs = 8 (Default)
+  profile_file = profile.csv (Default)
+  save_cmdstan_config = false (Default)
+num_threads = 1 (Default)</code></pre>
+<p>The second part of the output indicates how well the algorithm fared, here converging and terminating normally. The numbers reported indicate that it took 5 iterations and 8 gradient evaluations. This is, not surprisingly, far fewer iterations than required for sampling; even fewer iterations would be used with less stringent user-specified convergence tolerances. The <code>alpha</code> value is for step size used. In the final state the change in parameters was roughly <span class="math inline">\(0.002\)</span> and the length of the gradient roughly 3e-05 (<span class="math inline">\(0.00003\)</span>).</p>
+<pre><code>Initial log joint probability = -6.85653
+    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes
+       5      -5.00402    0.00184936   3.35074e-05           1           1        8
+Optimization terminated normally:
+  Convergence detected: relative gradient magnitude is below tolerance</code></pre>
+<p>The output from optimization is written into the file <code>output.csv</code> by default. The output follows the <a href="../cmdstan-guide/stan_csv_apdx.html">same pattern</a> as the output for sampling, first dumping the entire set of parameters used as comment lines:</p>
+<pre><code># stan_version_major = 2
+# stan_version_minor = 23
+# stan_version_patch = 0
+# model = bernoulli_model
+# method = optimize
+#   optimize
+#     algorithm = lbfgs (Default)
+#       lbfgs
+#         init_alpha = 0.001 (Default)
+#         tol_obj = 1e-12 (Default)
+#         tol_rel_obj = 10000 (Default)
+#         tol_grad = 1e-08 (Default)
+#         tol_rel_grad = 1e+07 (Default)
+#         tol_param = 1e-08 (Default)
+#         history_size = 5 (Default)
+#     jacobian = false (Default)
+#     iter = 2000 (Default)
+#     save_iterations = false (Default)</code></pre>
+<p>Following the config information are two lines of output, the CSV headers and the recorded values:</p>
+<pre><code>lp__,converged__,theta
+-5.00402,31,0.200003</code></pre>
+<p>Note that everything is a comment other than a line for the header, and a line for the values. Here, the header indicates the unnormalized log probability with <code>lp__</code>, algorithm status in <code>converged__</code>, and the model parameter <code>theta</code>. The maximum log probability is -5.0 and the posterior mode for <code>theta</code> is 0.20. The mode exactly matches what we would expect from the data. Because the prior was uniform, the result 0.20 represents the maximum likelihood estimate (MLE) for the very simple Bernoulli model. Note that no uncertainty is reported.</p>
+<p>All of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument <code>save_iterations</code> specifies whether or not to save the intermediate iterations to the output file. Allowed values are <code>true</code> or <code>false</code>. The default value is <code>false</code>, i.e., intermediate iterations are not saved to the output file. Running the optimizer with <code>save_iterations=true</code> writes both the initial log joint probability and values for all iterations to the output CSV file.</p>
+<p>Running the example model with option <code>save_iterations=true</code>, i.e., the command</p>
+<pre><code>&gt; ./bernoulli optimize save_iterations=1 data file=bernoulli.data.json</code></pre>
+<p>produces CSV file output rows:</p>
+<pre><code>lp__,converged__,theta
+-6.85653,0,0.493689
+-6.10128,0,0.420936
+-5.02953,0,0.22956
+-5.00517,0,0.206107
+-5.00403,0,0.200299
+-5.00402,31,0.200003</code></pre>
+<section id="meaning-of-the-converged__-column" class="level2">
+<h2 class="anchored" data-anchor-id="meaning-of-the-converged__-column">Meaning of the <code>converged__</code> column</h2>
+<p>The <code>converged__</code> column is used to indicate the status of the algorithm. It can take on the following values.</p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 15%">
+<col style="width: 84%">
+</colgroup>
+<thead>
+<tr class="header">
+<th>Numeric value</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td>-1</td>
+<td>Line search failed to achieve a sufficient decrease, no more progress can be made</td>
+</tr>
+<tr class="even">
+<td>0</td>
+<td>Successful step completed</td>
+</tr>
+<tr class="odd">
+<td>10</td>
+<td>Convergence detected: absolute parameter change was below tolerance</td>
+</tr>
+<tr class="even">
+<td>20</td>
+<td>Convergence detected: absolute change in objective function was below tolerance</td>
+</tr>
+<tr class="odd">
+<td>21</td>
+<td>Convergence detected: relative change in objective function was below tolerance</td>
+</tr>
+<tr class="even">
+<td>30</td>
+<td>Convergence detected: gradient norm is below tolerance</td>
+</tr>
+<tr class="odd">
+<td>31</td>
+<td>Convergence detected: relative gradient magnitude is below tolerance</td>
+</tr>
+<tr class="even">
+<td>40</td>
+<td>Maximum number of iterations hit, may not be at an optima</td>
+</tr>
+</tbody>
+</table>
+<p>Note that not all algorithms can produce all codes, and some, such as <code>0</code>, will only be observed if <code>save_iterations</code> is used.</p>
+<p>A comment explaining the final value (with the same contents as the above table) is added at the end of the CSV file.</p>
+</section>
+<section id="jacobian-adjustments" class="level2">
+<h2 class="anchored" data-anchor-id="jacobian-adjustments">Jacobian adjustments</h2>
+<p>If there are constrained parameters, Stan makes <a href="https://mc-stan.org/docs/reference-manual/transforms.html">a transformation to an unconstrained space</a> and runs the optimization algorithm in the unconstrained space.</p>
+<p>The <code>jacobian</code> argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of <a href="https://mc-stan.org/docs/reference-manual/transforms.html">inverse parameter transforms</a>.</p>
+<p>If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.</p>
+<p>If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, <a href="../cmdstan-guide/laplace_sample_config.html">Laplace sampling</a>, as then Jacobian adjustment needs to be included for correct results.</p>
+</section>
+<section id="optimization-algorithms" class="level2">
+<h2 class="anchored" data-anchor-id="optimization-algorithms">Optimization algorithms</h2>
+<p>The <code>algorithm</code> argument specifies the optimization algorithm. This argument takes one of the following three values:</p>
+<ul>
+<li><p><code>lbfgs</code> A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.</p></li>
+<li><p><code>bfgs</code> A quasi-Newton optimizer.</p></li>
+<li><p><code>newton</code> A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.</p></li>
+</ul>
+<p>See the Stan Reference Manual’s <a href="https://mc-stan.org/docs/reference-manual/optimization.html">Optimization chapter</a> for a description of these algorithms.</p>
+<p>All of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument <code>save_iterations</code> specifies whether or not to save the intermediate iterations to the output file. Allowed values are <code>true</code> or <code>false. The default value is</code>false`, i.e., intermediate iterations are not saved to the output file.</p>
+</section>
+<section id="the-quasi-newton-optimizers" class="level2">
+<h2 class="anchored" data-anchor-id="the-quasi-newton-optimizers">The quasi-Newton optimizers</h2>
+<p>For both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the <a href="https://mc-stan.org/docs/reference-manual/optimization.html#bfgs-and-l-bfgs-configuration">BFGS and L-BFGS configuration section</a> for details on the convergence tests.</p>
+<p>Both BFGS and L-BFGS have the following configuration arguments:</p>
+<ul>
+<li><p><code>init_alpha</code> - The initial step size parameter. Must be a positive real number. Default value is <span class="math inline">\(0.001\)</span></p></li>
+<li><p><code>tol_obj</code> - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is <span class="math inline">\(1^{-12}\)</span>.</p></li>
+<li><p><code>tol_rel_obj</code> - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is <span class="math inline">\(1^{4}\)</span>.</p></li>
+<li><p><code>tol_grad</code> - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is <span class="math inline">\(1^{-8}\)</span>.</p></li>
+<li><p><code>tol_rel_grad</code> - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is <span class="math inline">\(1^{7}\)</span>.</p></li>
+<li><p><code>tol_param</code> - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is <span class="math inline">\(1^{-8}\)</span>.</p></li>
+</ul>
+<p>The <code>init_alpha</code> argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set <code>init_alpha</code> to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.</p>
+<p>In addition to the above, the L-BFGS algorithm has argument <code>history_size</code> which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (<span class="math inline">\(5\)</span>-<span class="math inline">\(10\)</span>) are sufficient; the default value is <span class="math inline">\(5\)</span>.</p>
+<p>If L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.</p>
+</section>
+<section id="the-newton-optimizer" class="level2">
+<h2 class="anchored" data-anchor-id="the-newton-optimizer">The Newton optimizer</h2>
+<p>There are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/mcmc_config.html" class="pagination-link" aria-label="MCMC Sampling using Hamiltonian Monte Carlo">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">MCMC Sampling using Hamiltonian Monte Carlo</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/pathfinder_config.html" class="pagination-link" aria-label="Pathfinder Method for Approximate Bayesian Inference">
+        <span class="nav-page-text">Pathfinder Method for Approximate Bayesian Inference</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/optimize_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/parallelization.html b/docs/2_39/cmdstan-guide/parallelization.html
new file mode 100644
index 000000000..a6c4315fc
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/parallelization.html
@@ -0,0 +1,1162 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Parallelization</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/command_line_options.html" rel="next">
+<link href="../cmdstan-guide/compiling_stan_programs.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Parallelization">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/parallelization.html">Parallelization</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#parallelization" id="toc-parallelization" class="nav-link active" data-scroll-target="#parallelization">Parallelization</a>
+  <ul class="collapse">
+  <li><a href="#multi-threading-with-tbb" id="toc-multi-threading-with-tbb" class="nav-link" data-scroll-target="#multi-threading-with-tbb">Multi-threading with TBB</a>
+  <ul class="collapse">
+  <li><a href="#compiling" id="toc-compiling" class="nav-link" data-scroll-target="#compiling">Compiling</a></li>
+  <li><a href="#running" id="toc-running" class="nav-link" data-scroll-target="#running">Running</a></li>
+  </ul></li>
+  <li><a href="#multi-processing-with-mpi" id="toc-multi-processing-with-mpi" class="nav-link" data-scroll-target="#multi-processing-with-mpi">Multi-processing with MPI</a>
+  <ul class="collapse">
+  <li><a href="#dependencies" id="toc-dependencies" class="nav-link" data-scroll-target="#dependencies">Dependencies</a></li>
+  <li><a href="#compiling-1" id="toc-compiling-1" class="nav-link" data-scroll-target="#compiling-1">Compiling</a></li>
+  <li><a href="#running-1" id="toc-running-1" class="nav-link" data-scroll-target="#running-1">Running</a></li>
+  </ul></li>
+  <li><a href="#opencl" id="toc-opencl" class="nav-link" data-scroll-target="#opencl">OpenCL</a>
+  <ul class="collapse">
+  <li><a href="#dependencies-1" id="toc-dependencies-1" class="nav-link" data-scroll-target="#dependencies-1">Dependencies</a></li>
+  <li><a href="#compiling-2" id="toc-compiling-2" class="nav-link" data-scroll-target="#compiling-2">Compiling</a></li>
+  <li><a href="#running-2" id="toc-running-2" class="nav-link" data-scroll-target="#running-2">Running</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/installation.html">Getting Started</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/parallelization.html">Parallelization</a></li></ol></nav></header>
+
+
+
+
+<section id="parallelization" class="level1">
+<h1>Parallelization</h1>
+<p>Stan provides three ways of parallelizing execution of a Stan model:</p>
+<ul>
+<li>multi-threading with Intel Threading Building Blocks (TBB),</li>
+<li>multi-processing with Message Passing Interface (MPI) and</li>
+<li>manycore processing with OpenCL.</li>
+</ul>
+<section id="multi-threading-with-tbb" class="level2">
+<h2 class="anchored" data-anchor-id="multi-threading-with-tbb">Multi-threading with TBB</h2>
+<p>In order to exploit multi-threading in a Stan model, the models must be rewritten to use the <code>reduce_sum</code> and <code>map_rect</code> functions. For instructions on how to rewrite Stan models to use these functions see <a href="https://mc-stan.org/docs/stan-users-guide/parallelization.html">Stan’s User guide chapter on parallelization</a>, <a href="https://mc-stan.org/users/documentation/case-studies/reduce_sum_tutorial.html">the reduce_sum case study</a> or the <a href="https://github.com/rmcelreath/cmdstan_map_rect_tutorial">Multithreading and Map-Reduce tutorial</a>.</p>
+<section id="compiling" class="level3">
+<h3 class="anchored" data-anchor-id="compiling">Compiling</h3>
+<p>Once a model is rewritten to use the above-mentioned functions, the model must be compiled with the <code>STAN_THREADS</code> makefile flag. The flag can be supplied in the <code>make</code> call but we recommend writing the flag to the <code>make/local</code> file. If the <code>STAN_THREADS</code> flag is defined/non-empty, threads will be enabled.</p>
+<p>An example of the contents of <code>make/local</code> to enable threading with TBB:</p>
+<pre><code>STAN_THREADS=true</code></pre>
+<p>The model is then compiled as normal:</p>
+<pre><code>make path/to/model</code></pre>
+</section>
+<section id="running" class="level3">
+<h3 class="anchored" data-anchor-id="running">Running</h3>
+<p>Before running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the <code>num_threads</code> argument. Valid values for <code>num_threads</code> are positive integers and -1. If <code>num_threads</code> is set to -1, all available cores will be used.</p>
+<p>Generally, this number should not exceed the number of available cores for best performance.</p>
+<p>Example:</p>
+<pre><code>./model sample data file=data.json num_threads=4 ...</code></pre>
+<p>When the model is compiled with <code>STAN_THREADS</code> we can sample with multiple chains with a single executable (see section <a href="../cmdstan-guide/mcmc_config.html#multi-chain-sampling">running multiple chains</a> for cases when this is available). When running multiple chains <code>num_threads</code> is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:</p>
+<pre><code>./model sample num_chains=2 data file=data.json num_threads=8 ...</code></pre>
+</section>
+</section>
+<section id="multi-processing-with-mpi" class="level2">
+<h2 class="anchored" data-anchor-id="multi-processing-with-mpi">Multi-processing with MPI</h2>
+<p>In order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the <a href="https://mc-stan.org/docs/functions-reference/higher-order_functions.html#functions-map"><code>map_rect</code> function</a>. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.</p>
+<section id="dependencies" class="level3">
+<h3 class="anchored" data-anchor-id="dependencies">Dependencies</h3>
+<p>Compiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are <a href="https://www.mpich.org/">MPICH</a> and <a href="https://www.open-mpi.org/">OpenMPI</a>.</p>
+</section>
+<section id="compiling-1" class="level3">
+<h3 class="anchored" data-anchor-id="compiling-1">Compiling</h3>
+<p>Once a model is rewritten to use <code>map_rect</code>, additional makefile flags must be written to the <code>make/local</code>. These are:</p>
+<ul>
+<li><code>STAN_MPI</code>: Enables the use of MPI with Stan if defined.</li>
+<li><code>CXX</code>: The name of the MPI C++ compiler wrapper. Typically <code>mpicxx</code>.</li>
+<li><code>TBB_CXX_TYPE</code>: The C++ compiler the MPI wrapper wraps. Typically <code>gcc</code> on Linux and <code>clang</code> on macOS.</li>
+</ul>
+<p>An example of <code>make/local</code> on Linux:</p>
+<pre><code>STAN_MPI=true
+CXX=mpicxx
+TBB_CXX_TYPE=gcc</code></pre>
+<p>The model is then compiled as normal:</p>
+<pre><code>make path/to/model</code></pre>
+</section>
+<section id="running-1" class="level3">
+<h3 class="anchored" data-anchor-id="running-1">Running</h3>
+<p>The Stan model compiled with <code>STAN_MPI</code> is run using an MPI launcher. The MPI standard suggests using <code>mpiexec</code>, but a vendor wrapper for the launcher like <code>mpirun</code> can also be used. The launcher is supplied the path to the built executable and the number of processes to start: <code>-n X</code> for <code>mpiexec</code> or <code>-np X</code> for <code>mpirun</code> where <code>X</code> is replaced by the integer representing the number of processes.</p>
+<p>Example for running a model with six processes:</p>
+<pre><code>mpiexec -n 6 path/to/model sample data file=data.json ...</code></pre>
+</section>
+</section>
+<section id="opencl" class="level2">
+<h2 class="anchored" data-anchor-id="opencl">OpenCL</h2>
+<section id="dependencies-1" class="level3">
+<h3 class="anchored" data-anchor-id="dependencies-1">Dependencies</h3>
+<p>OpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.</p>
+<p>In order to check if any OpenCL-enabled device and its runtime is already present use the <code>clinfo</code> tool. On Linux, <code>clinfo</code> can typically be installed with the default package manager (for example <code>sudo apt-get install clinfo</code> on Ubuntu). For Windows, pre-built <code>clinfo</code> binary can be found <a href="https://github.com/Oblomov/clinfo#windows-support">here</a>.</p>
+<p>Also use <code>clinfo</code> to verify successful installation of OpenCL runtimes.</p>
+<section id="nvidia-gpu" class="level4">
+<h4 class="anchored" data-anchor-id="nvidia-gpu">NVIDIA GPU</h4>
+<ul>
+<li><p>Linux:</p>
+<p>Install the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:</p>
+<pre><code>sudo apt update
+sudo apt install nvidia-driver-460 nvidia-cuda-toolkit</code></pre>
+<p>Replace the driver version (<code>460</code> in the above case) with the lastest number at the time of installation.</p></li>
+<li><p>Windows:</p>
+<p>Install the <a href="https://www.nvidia.com/Download/index.aspx">NVIDIA GPU Driver</a> and <a href="https://developer.nvidia.com/cuda-toolkit">CUDA Toolkit</a>.</p></li>
+</ul>
+</section>
+<section id="amd-gpu" class="level4">
+<h4 class="anchored" data-anchor-id="amd-gpu">AMD GPU</h4>
+<ul>
+<li><p>Linux:</p>
+<p>Install <code>Radeon Software for Linux</code> available <a href="https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-20-40">here</a>.</p></li>
+<li><p>Windows:</p>
+<p>We recommend installing the open source <a href="https://github.com/GPUOpen-LibrariesAndSDKs/OCL-SDK/releases">OCL-SDK</a>.</p></li>
+</ul>
+</section>
+<section id="amd-cpu" class="level4">
+<h4 class="anchored" data-anchor-id="amd-cpu">AMD CPU</h4>
+<p>Install the open source <a href="http://portablecl.org/download.html">PoCL</a>.</p>
+</section>
+<section id="intel-cpugpu" class="level4">
+<h4 class="anchored" data-anchor-id="intel-cpugpu">Intel CPU/GPU</h4>
+<p>Follow Intel’s install instructions given <a href="https://software.intel.com/content/www/us/en/develop/articles/opencl-drivers.html">here</a> (requires registration).</p>
+</section>
+</section>
+<section id="compiling-2" class="level3">
+<h3 class="anchored" data-anchor-id="compiling-2">Compiling</h3>
+<p>In order to enable the OpenCL backend the model must be compiled with the <code>STAN_OPENCL</code> makefile flag defined/non-empty. The flag can be supplied in the <code>make</code> call but we recommend writing the flag to the <code>make/local</code> file.</p>
+<p>An example of the contents of <code>make/local</code> to enable parallelization with OpenCL:</p>
+<pre><code>STAN_OPENCL=true</code></pre>
+<p>If you are using OpenCL with an integrated GPU you also need to define the <code>INTEGRATED_OPENCL</code> flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:</p>
+<pre><code>INTEGRATED_OPENCL=true</code></pre>
+<p>The model is then compiled as normal:</p>
+<pre><code>make path/to/model</code></pre>
+</section>
+<section id="running-2" class="level3">
+<h3 class="anchored" data-anchor-id="running-2">Running</h3>
+<p>The Stan model compiled with <code>STAN_OPENCL</code> can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the <code>clinfo</code> program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically <code>0</code>. In that case you can also omit the OpenCL IDs as the default <code>0</code> IDs are used in that case.</p>
+<p>We supply these IDs when starting the executable as shown below:</p>
+<pre><code>path/to/model sample data file=data.json opencl platform=0 device=1</code></pre>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/compiling_stan_programs.html" class="pagination-link" aria-label="Compiling a Stan Program">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Compiling a Stan Program</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/command_line_options.html" class="pagination-link" aria-label="Command-Line Interface Overview">
+        <span class="nav-page-text">Command-Line Interface Overview</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/pathfinder_config.html b/docs/2_39/cmdstan-guide/pathfinder_config.html
new file mode 100644
index 000000000..3ad17a27a
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/pathfinder_config.html
@@ -0,0 +1,1269 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Pathfinder Method for Approximate Bayesian Inference</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/variational_config.html" rel="next">
+<link href="../cmdstan-guide/optimize_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Pathfinder Method for Approximate Bayesian Inference">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/pathfinder_config.html">Pathfinder Method for Approximate Bayesian Inference</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#pathfinder-config" id="toc-pathfinder-config" class="nav-link active" data-scroll-target="#pathfinder-config">Pathfinder Method for Approximate Bayesian Inference</a>
+  <ul class="collapse">
+  <li><a href="#pathfinder-configuration" id="toc-pathfinder-configuration" class="nav-link" data-scroll-target="#pathfinder-configuration">Pathfinder Configuration</a></li>
+  <li><a href="#l-bfgs-configuration" id="toc-l-bfgs-configuration" class="nav-link" data-scroll-target="#l-bfgs-configuration">L-BFGS Configuration</a></li>
+  <li><a href="#pathfinder_csv" id="toc-pathfinder_csv" class="nav-link" data-scroll-target="#pathfinder_csv">Multi-path Pathfinder CSV files</a></li>
+  <li><a href="#single-path-pathfinder-outputs" id="toc-single-path-pathfinder-outputs" class="nav-link" data-scroll-target="#single-path-pathfinder-outputs">Single-path Pathfinder Outputs</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/pathfinder_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/pathfinder_config.html">Pathfinder Method for Approximate Bayesian Inference</a></li></ol></nav></header>
+
+
+
+
+<section id="pathfinder-config" class="level1">
+<h1>Pathfinder Method for Approximate Bayesian Inference</h1>
+<p>The CmdStan method <code>pathfinder</code> uses the Pathfinder algorithm of <span class="citation" data-cites="zhang_pathfinder:2022">Zhang et al. (<a href="#ref-zhang_pathfinder:2022" role="doc-biblioref">2022</a>)</span>, which is further described in the <a href="https://mc-stan.org/docs/reference-manual/pathfinder.html">Stan Reference Manual</a>.</p>
+<p>A single run of the Pathfinder algorithm generates a set of approximate draws. Inference is improved by running multiple Pathfinder instances and using Pareto-smoothed importance resampling (PSIS) of the resulting sets of draws. This better matches non-normal target densities and also eliminates minor modes.</p>
+<p>The <code>pathfinder</code> method runs multi-path Pathfinder by default, which returns a PSIS sample over the draws from several individual (“single-path”) Pathfinder runs. Argument <code>num_paths</code> specifies the number of single-path Pathfinders, the default is <span class="math inline">\(4\)</span>. If <code>num_paths</code> is set to 1, then only one individual Pathfinder is run without the PSIS reweighting of the sample.</p>
+<p>The full set of configuration options available for the <code>pathfinder</code> method is available by using the <code>pathfinder help-all</code> subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.</p>
+<p>The following is a minimal call to the Pathfinder algorithm using defaults for everything but the location of the data file.</p>
+<pre><code>&gt; ./bernoulli pathfinder data file=bernoulli.data.R</code></pre>
+<p>Executing this command prints output both to the console and to csv files.</p>
+<p>The first part of the console output reports on the configuration used.</p>
+<pre><code>method = pathfinder
+  pathfinder
+    init_alpha = 0.001 (Default)
+    tol_obj = 1e-12 (Default)
+    tol_rel_obj = 10000 (Default)
+    tol_grad = 1e-08 (Default)
+    tol_rel_grad = 1e+07 (Default)
+    tol_param = 1e-08 (Default)
+    history_size = 5 (Default)
+    num_psis_draws = 1000 (Default)
+    num_paths = 4 (Default)
+    save_single_paths = false (Default)
+    psis_resample = true (Default)
+    calculate_lp = true (Default)
+    max_lbfgs_iters = 1000 (Default)
+    num_draws = 1000 (Default)
+    num_elbo_draws = 25 (Default)
+id = 1 (Default)
+data
+  file = bernoulli.data.json
+init = 2 (Default)
+random
+  seed = 2790476610 (Default)
+output
+  file = output.csv (Default)
+  diagnostic_file =  (Default)
+  refresh = 100 (Default)
+  sig_figs = 8 (Default)
+  profile_file = profile.csv (Default)
+  save_cmdstan_config = false (Default)
+num_threads = 1 (Default)</code></pre>
+<p>The rest of the output describes the progression of the algorithm.</p>
+<p>By default, the Pathfinder algorithm runs 4 single-path Pathfinders in parallel, then uses importance resampling on the set of returned draws to produce the specified number of draws.</p>
+<pre><code>Path [1] :Initial log joint density = -11.543343
+Path [1] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes
+              5      -6.748e+00      1.070e-03   1.707e-05    1.000e+00  1.000e+00       126 -6.220e+00 -6.220e+00
+Path [1] :Best Iter: [5] ELBO (-6.219833) evaluations: (126)
+Path [2] :Initial log joint density = -7.443345
+Path [2] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes
+              5      -6.748e+00      9.936e-05   3.738e-07    1.000e+00  1.000e+00       126 -6.164e+00 -6.164e+00
+Path [2] :Best Iter: [5] ELBO (-6.164015) evaluations: (126)
+Path [3] :Initial log joint density = -18.986308
+Path [3] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes
+              5      -6.748e+00      2.996e-04   4.018e-06    1.000e+00  1.000e+00       126 -6.201e+00 -6.201e+00
+Path [3] :Best Iter: [5] ELBO (-6.200559) evaluations: (126)
+Path [4] :Initial log joint density = -8.304453
+Path [4] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes
+              5      -6.748e+00      2.814e-04   2.034e-06    1.000e+00  1.000e+00       126 -6.221e+00 -6.221e+00
+Path [4] :Best Iter: [3] ELBO (-6.161276) evaluations: (126)
+Total log probability function evaluations:8404</code></pre>
+<section id="pathfinder-configuration" class="level2">
+<h2 class="anchored" data-anchor-id="pathfinder-configuration">Pathfinder Configuration</h2>
+<ul>
+<li><p><code>num_psis_draws</code> - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is <span class="math inline">\(1000\)</span>.</p></li>
+<li><p><code>num_paths</code> - Number of single pathfinders. Must be a positive integer. Default value is <span class="math inline">\(4\)</span>.</p></li>
+<li><p><code>save_single_paths</code> - When <code>true</code>, save outputs from single pathfinders. Valid values: <code>[true, false]</code>. Default is <code>false</code>.</p></li>
+<li><p><code>max_lbfgs_iters</code> - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is <span class="math inline">\(1000\)</span>.</p></li>
+<li><p><code>num_draws</code> - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is <span class="math inline">\(1000\)</span>. Can differ from <code>num_psis_draws</code>.</p></li>
+<li><p><code>num_elbo_draws</code> - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is <span class="math inline">\(25\)</span>.</p></li>
+<li><p><code>psis_resample</code> - If <code>true</code>, perform PSIS resampling on draws returned from individual pathfinders. If <code>false</code>, returns all <code>num_paths * num_draws</code> draws from the individual pathfinders. Valid values: <code>[true, false]</code>. Default is <code>true</code>.</p></li>
+<li><p><code>calculate_lp</code> - If <code>true</code>, log probabilities of the approximate draws are calculated and returned with the output. If <code>false</code>, each pathfinder will only calculate the lp values needed for the ELBO calculation. If <code>false</code>, PSIS resampling cannot be performed and the algorithm returns <code>num_paths * num_draws</code> samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: <code>[true, false]</code>. Default is <code>true</code>.</p></li>
+</ul>
+</section>
+<section id="l-bfgs-configuration" class="level2">
+<h2 class="anchored" data-anchor-id="l-bfgs-configuration">L-BFGS Configuration</h2>
+<p>Arguments <code>init_alpha</code> through <code>history_size</code> are the full set of arguments to the L-BFGS optimizer and have the same defaults for <a href="../cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers">optimization</a>.</p>
+</section>
+<section id="pathfinder_csv" class="level2">
+<h2 class="anchored" data-anchor-id="pathfinder_csv">Multi-path Pathfinder CSV files</h2>
+<p>By default, the <code>pathfinder</code> method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a <a href="../cmdstan-guide/stan_csv_apdx.html">StanCSV file</a>.</p>
+<p>The CSV files have the following structure:</p>
+<p>The initial CSV comment rows contain the complete set of CmdStan configuration options.</p>
+<pre><code>...
+# method = pathfinder
+#   pathfinder
+#     init_alpha = 0.001 (Default)
+#     tol_obj = 9.9999999999999998e-13 (Default)
+#     tol_rel_obj = 10000 (Default)
+#     tol_grad = 1e-08 (Default)
+#     tol_rel_grad = 10000000 (Default)
+#     tol_param = 1e-08 (Default)
+#     history_size = 5 (Default)
+#     num_psis_draws = 1000 (Default)
+#     num_paths = 4 (Default)
+#     psis_resample = 1 (Default)
+#     calculate_lp = 1 (Default)
+#     save_single_paths = 0 (Default)
+#     max_lbfgs_iters = 1000 (Default)
+#     num_draws = 1000 (Default)
+#     num_elbo_draws = 25 (Default)
+...</code></pre>
+<p>Next is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs <code>lp_approx__</code>, the log density in the approximating distribution, and <code>lp__</code>, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.</p>
+<pre><code>lp_approx__,lp__,theta
+-2.4973, -8.2951, 0.0811852
+-0.87445, -7.06526, 0.160207
+-0.812285, -7.07124, 0.35819
+...</code></pre>
+<p>The final lines are comment lines which give timing information.</p>
+<pre><code># Elapsed Time: 0.016000 seconds (Pathfinders)
+#               0.003000 seconds (PSIS)
+#               0.019000 seconds (Total)</code></pre>
+<p>Pathfinder provides option <code>save_single_paths</code> which will save output from the single-path Pathfinder runs.</p>
+</section>
+<section id="single-path-pathfinder-outputs" class="level2">
+<h2 class="anchored" data-anchor-id="single-path-pathfinder-outputs">Single-path Pathfinder Outputs</h2>
+<p>The boolean option <code>save_single_paths</code> is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When <code>save_single_paths</code> is <code>true</code>, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name <code>output.csv</code> and default number of runs (4), the resulting CSV files are</p>
+<pre><code>output.csv
+output_path_1.csv
+output_path_1.json
+output_path_2.csv
+output_path_2.json
+output_path_3.csv
+output_path_3.json
+output_path_4.csv
+output_path_4.json</code></pre>
+<p>The individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.</p>
+<p>To see how this works, we run Pathfinder on the centered-parameterization of the <code>eight-schools</code> model, where the posterior distribution has a funnel shape:</p>
+<pre><code>&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json</code></pre>
+<p>Each JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.</p>
+<pre><code>{
+  "0": {
+    "iter": 0,
+    "unconstrained_parameters": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],
+    "grads": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]
+  },
+  "1": {
+    "iter": 1,
+    "unconstrained_parameters": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],
+    "grads": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],
+    "history_size": 1,
+    "lbfgs_success": true,
+    "pathfinder_success": true,
+    "x_center": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],
+    "logDetCholHk": -2.6839,
+    "L_approx": [[-0.0630456, -0.0187959], [0, 1.08328]],
+    "Qk": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],
+    "alpha": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
+    "full": false,
+    "lbfgs_note": ""
+  },
+  ...,
+  "171": {
+    "iter": 171,
+    "unconstrained_parameters": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],
+    "grads": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],
+    "history_size": 5,
+    "lbfgs_success": true,
+    "pathfinder_success": true,
+    "x_center": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],
+    "logDetCholHk": 299.023,
+    "L_approx": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],
+    "Qk": [],
+    "alpha": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],
+    "full": true,
+    "lbfgs_note": ""
+  },
+  "172": {
+    "iter": 172,
+    "unconstrained_parameters": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],
+    "grads": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],
+    "history_size": 5,
+    "lbfgs_success": false,
+    "pathfinder_success": false,
+    "lbfgs_note": ""
+  }
+}</code></pre>
+<p>Option <code>num_paths=1</code> runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments <code>num_paths=1 save_single_paths=true</code> creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-zhang_pathfinder:2022" class="csl-entry" role="listitem">
+Zhang, Lu, Bob Carpenter, Andrew Gelman, and Aki Vehtari. 2022. <span>“Pathfinder: Parallel Quasi-Newton Variational Inference.”</span> <em>Journal of Machine Learning Research</em> 23 (306): 1–49. <a href="http://jmlr.org/papers/v23/21-0889.html">http://jmlr.org/papers/v23/21-0889.html</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/optimize_config.html" class="pagination-link" aria-label="Optimization">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Optimization</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/variational_config.html" class="pagination-link" aria-label="Variational Inference using ADVI">
+        <span class="nav-page-text">Variational Inference using ADVI</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/pathfinder_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/print.html b/docs/2_39/cmdstan-guide/print.html
new file mode 100644
index 000000000..38a6da455
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/print.html
@@ -0,0 +1,1034 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Utility print (deprecated)</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/stan_csv_apdx.html" rel="next">
+<link href="../cmdstan-guide/diagnose_utility.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Utility print (deprecated)">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/print.html"><code>print</code> (deprecated): MCMC Output Analysis</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#print-deprecated-mcmc-output-analysis" id="toc-print-deprecated-mcmc-output-analysis" class="nav-link active" data-scroll-target="#print-deprecated-mcmc-output-analysis"><code>print</code> (deprecated): MCMC Output Analysis</a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/print.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/print.html"><code>print</code> (deprecated): MCMC Output Analysis</a></li></ol></nav></header>
+
+
+
+
+<section id="print-deprecated-mcmc-output-analysis" class="level1">
+<h1><code>print</code> (deprecated): MCMC Output Analysis</h1>
+<p>The <code>print</code> utility is deprecated, but is still available until CmdStan v3.0. It has been replaced by the <a href="../cmdstan-guide/stansummary.html"><code>stansummary</code> utility</a>.</p>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/diagnose_utility.html" class="pagination-link" aria-label="`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/stan_csv_apdx.html" class="pagination-link" aria-label="Stan CSV File Format">
+        <span class="nav-page-text">Stan CSV File Format</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/print.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/rdump_apdx.html b/docs/2_39/cmdstan-guide/rdump_apdx.html
new file mode 100644
index 000000000..aea8c2afe
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/rdump_apdx.html
@@ -0,0 +1,1290 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>RDump Format for CmdStan</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/external_code.html" rel="next">
+<link href="../cmdstan-guide/json_apdx.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="RDump Format for CmdStan">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/rdump_apdx.html">RDump Format for CmdStan</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#rdump" id="toc-rdump" class="nav-link active" data-scroll-target="#rdump">RDump Format for CmdStan</a>
+  <ul class="collapse">
+  <li><a href="#creating-dump-files" id="toc-creating-dump-files" class="nav-link" data-scroll-target="#creating-dump-files">Creating dump files</a></li>
+  <li><a href="#scalar-variables" id="toc-scalar-variables" class="nav-link" data-scroll-target="#scalar-variables">Scalar variables</a></li>
+  <li><a href="#sequence-variables" id="toc-sequence-variables" class="nav-link" data-scroll-target="#sequence-variables">Sequence variables</a></li>
+  <li><a href="#array-variables" id="toc-array-variables" class="nav-link" data-scroll-target="#array-variables">Array variables</a></li>
+  <li><a href="#matrix--and-vector-valued-variables" id="toc-matrix--and-vector-valued-variables" class="nav-link" data-scroll-target="#matrix--and-vector-valued-variables">Matrix- and vector-valued variables</a>
+  <ul class="collapse">
+  <li><a href="#vector-dump-format" id="toc-vector-dump-format" class="nav-link" data-scroll-target="#vector-dump-format">Vector dump format</a></li>
+  <li><a href="#matrix-dump-format" id="toc-matrix-dump-format" class="nav-link" data-scroll-target="#matrix-dump-format">Matrix dump format</a></li>
+  <li><a href="#arrays-of-vectors-and-matrices" id="toc-arrays-of-vectors-and-matrices" class="nav-link" data-scroll-target="#arrays-of-vectors-and-matrices">Arrays of vectors and matrices</a></li>
+  </ul></li>
+  <li><a href="#complex-valued-variables" id="toc-complex-valued-variables" class="nav-link" data-scroll-target="#complex-valued-variables">Complex-valued variables</a></li>
+  <li><a href="#integer--and-real-valued-variables" id="toc-integer--and-real-valued-variables" class="nav-link" data-scroll-target="#integer--and-real-valued-variables">Integer- and real-valued variables</a>
+  <ul class="collapse">
+  <li><a href="#scientific-notation" id="toc-scientific-notation" class="nav-link" data-scroll-target="#scientific-notation">Scientific notation</a></li>
+  <li><a href="#infinite-and-not-a-number-values" id="toc-infinite-and-not-a-number-values" class="nav-link" data-scroll-target="#infinite-and-not-a-number-values">Infinite and not-a-number values</a></li>
+  </ul></li>
+  <li><a href="#quoted-variable-names" id="toc-quoted-variable-names" class="nav-link" data-scroll-target="#quoted-variable-names">Quoted variable names</a></li>
+  <li><a href="#line-breaks" id="toc-line-breaks" class="nav-link" data-scroll-target="#line-breaks">Line breaks</a></li>
+  <li><a href="#bnf-grammar-for-dump-data" id="toc-bnf-grammar-for-dump-data" class="nav-link" data-scroll-target="#bnf-grammar-for-dump-data">BNF grammar for dump data</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/rdump_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/rdump_apdx.html">RDump Format for CmdStan</a></li></ol></nav></header>
+
+
+
+
+<section id="rdump" class="level1">
+<h1>RDump Format for CmdStan</h1>
+<p><strong>NOTE:</strong> Although the RDump format is still supported, I/O with JSON is faster and recommended. See the <a href="../cmdstan-guide/json_apdx.html">chapter on JSON</a> for more details.</p>
+<p>RDump format can be used to represent values for Stan variables. This format was introduced in SPLUS and is used in R, JAGS, and in BUGS (but with a different ordering).</p>
+<p>A dump file is structured as a sequence of variable definitions. Each variable is defined in terms of its dimensionality and its values. There are three kinds of variable declarations: - scalars - sequences - general arrays</p>
+<section id="creating-dump-files" class="level2">
+<h2 class="anchored" data-anchor-id="creating-dump-files">Creating dump files</h2>
+<p>Dump files can be created from R using RStan, via the <code>rstan</code> package function <code>stan_rdump</code>. Stan RDump files must be created via <code>stan_rdump</code> and not by R’s native <code>dump</code> function because R’s <code>dump</code> function uses a richer syntax than is supported by the underlying Stan i/o libraries.</p>
+</section>
+<section id="scalar-variables" class="level2">
+<h2 class="anchored" data-anchor-id="scalar-variables">Scalar variables</h2>
+<p>A simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable <code>y</code> with value <span class="math inline">\(17.2\)</span>:</p>
+<pre><code>y &lt;- 17.2</code></pre>
+</section>
+<section id="sequence-variables" class="level2">
+<h2 class="anchored" data-anchor-id="sequence-variables">Sequence variables</h2>
+<p>One-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.</p>
+<pre><code>n &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)</code></pre>
+<p>Arrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.</p>
+<p>Sequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as</p>
+<pre><code>n &lt;- 1:3</code></pre>
+<p>The sequence denoted by <code>1:3</code> is of length <span class="math inline">\(3\)</span>, running from <span class="math inline">\(1\)</span> to <span class="math inline">\(3\)</span> inclusive. The colon notation allows sequences going from high to low. The following are equivalent:</p>
+<pre><code>n &lt;- 2:-2
+n &lt;- c(2,1,0,-1,-2)</code></pre>
+<p>As a special case, a sequence of zeros can also be represented in the dump format by <code>integer(x)</code> and <code>double(x)</code>, for type int and double, respectively. Here <code>x</code> is a non-negative integer to specify the length. If <code>x</code> is <span class="math inline">\(0\)</span>, it can be omitted. The following are some examples.</p>
+<pre><code>x1 &lt;- integer()
+x2 &lt;- integer(0)
+x3 &lt;- integer(2)
+y1 &lt;- double()
+y2 &lt;- double(0)
+y3 &lt;- double(2)</code></pre>
+</section>
+<section id="array-variables" class="level2">
+<h2 class="anchored" data-anchor-id="array-variables">Array variables</h2>
+<p>For more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a <span class="math inline">\(2 \times 3\)</span> array:</p>
+<pre><code>y &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))</code></pre>
+<p>Data is stored column-major, thus the values for <code>y</code> will be:</p>
+<pre><code>y[1, 1] = 1
+y[1, 2] = 3
+y[1, 3] = 5
+y[2, 1] = 2
+y[2, 2] = 4
+y[2, 3] = 6</code></pre>
+<p>The <code>structure</code> keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.</p>
+<p>If the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.</p>
+<pre><code>y &lt;- structure(1:6, .Dim = c(2,3))</code></pre>
+<p>Sequence notation can be used within any call to the generic <code>c()</code> function in R. In the above example, c(2,3) could be written as <code>c(2:3)</code>.</p>
+<p>The generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,</p>
+<pre><code>z &lt;- structure(1:24, .Dim = c(2,3,4))</code></pre>
+<p>produces a three-dimensional <code>int</code> (assignable to <code>real</code>) array <code>z</code> with values:</p>
+<pre><code>z[1, 1, 1] = 1
+z[2, 1, 1] = 2
+z[1, 2, 1] = 3
+z[2, 2, 1] = 4
+z[1, 3, 1] = 5
+z[2, 3, 1] = 6
+z[1, 1, 2] = 7
+z[2, 1, 2] = 8
+z[1, 2, 2] = 9
+z[2, 2, 2] = 10
+z[1, 3, 2] = 11
+z[2, 3, 2] = 12
+z[1, 1, 3] = 13
+z[2, 1, 3] = 14
+z[1, 2, 3] = 15
+z[2, 2, 3] = 16
+z[1, 3, 3] = 17
+z[2, 3, 3] = 18
+z[1, 1, 4] = 19
+z[2, 1, 4] = 20
+z[1, 2, 4] = 21
+z[2, 2, 4] = 22
+z[1, 3, 4] = 23
+z[2, 3, 4] = 24</code></pre>
+<p>If the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.</p>
+<p>The sequence of values inside <code>structure</code> can also be <code>integer(x)</code> or <code>double(x)</code>. In particular, if one or more dimensions is zero, <code>integer()</code> can be put inside <code>structure</code>. For instance, the following example is supported by the dump format.</p>
+<pre><code>y &lt;- structure(integer(), .Dim = c(2,0))</code></pre>
+</section>
+<section id="matrix--and-vector-valued-variables" class="level2">
+<h2 class="anchored" data-anchor-id="matrix--and-vector-valued-variables">Matrix- and vector-valued variables</h2>
+<p>The dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.</p>
+<section id="vector-dump-format" class="level3">
+<h3 class="anchored" data-anchor-id="vector-dump-format">Vector dump format</h3>
+<p>The following three declarations have the same dump format for their data.</p>
+<pre><code>array[K] real a;
+vector[K] b;
+row_vector[K] c;</code></pre>
+</section>
+<section id="matrix-dump-format" class="level3">
+<h3 class="anchored" data-anchor-id="matrix-dump-format">Matrix dump format</h3>
+<p>The following declarations have the same dump format.</p>
+<pre><code>array[M, N] real a;
+matrix[M, N] b;</code></pre>
+</section>
+<section id="arrays-of-vectors-and-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="arrays-of-vectors-and-matrices">Arrays of vectors and matrices</h3>
+<p>The key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.</p>
+<p>For the dump data format, the following declarations have the same arrangement.</p>
+<pre><code>array[M, N] real a;
+matrix[M, N] b;
+array[M] vector[N] c;
+array[M] row_vector[N] d;</code></pre>
+<p>Similarly, the following also have the same dump format.</p>
+<pre><code>array[P, M, N] real a;
+array[P] matrix[M, N] b;
+array[P, M] vector[N] c;
+array[P, M] row_vector[N] d;</code></pre>
+</section>
+</section>
+<section id="complex-valued-variables" class="level2">
+<h2 class="anchored" data-anchor-id="complex-valued-variables">Complex-valued variables</h2>
+<p>At this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.</p>
+</section>
+<section id="integer--and-real-valued-variables" class="level2">
+<h2 class="anchored" data-anchor-id="integer--and-real-valued-variables">Integer- and real-valued variables</h2>
+<p>There is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., <span class="math inline">\(132.3\)</span>) or uses scientific notation (e.g., <span class="math inline">\(1.323e2\)</span>), Stan assumes that the values are real.</p>
+<p>For a single value, if there is no decimal point, it may be assigned to an <code>int</code> or <code>real</code> variable in Stan. An array value may only be assigned to an <code>int</code> array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.</p>
+<p>The following dump file declares an integer value for <code>y</code>.</p>
+<pre><code>y &lt;- 2</code></pre>
+<p>This definition can be used for a Stan variable <code>y</code> declared as <code>real</code> or as <code>int</code>. Assigning an integer value to a real variable automatically promotes the integer value to a real value.</p>
+<p>Integer values may optionally be followed by <code>L</code> or <code>l</code>, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.</p>
+<pre><code>y &lt;- 2L</code></pre>
+<p>The following dump file provides a real value for <code>y</code>.</p>
+<pre><code>y &lt;- 2.0</code></pre>
+<p>Even though this is a round value, the occurrence of the decimal point in the value, <span class="math inline">\(2.0\)</span>, causes Stan to infer that <code>y</code> is real valued. This dump file may only be used for variables <code>y</code> declared as real in Stan.</p>
+<section id="scientific-notation" class="level3">
+<h3 class="anchored" data-anchor-id="scientific-notation">Scientific notation</h3>
+<p>Numbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as <span class="math inline">\(1e+06\)</span>.</p>
+</section>
+<section id="infinite-and-not-a-number-values" class="level3">
+<h3 class="anchored" data-anchor-id="infinite-and-not-a-number-values">Infinite and not-a-number values</h3>
+<p>Stan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.</p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: right;">Value</th>
+<th style="text-align: left;">Preferred Form</th>
+<th style="text-align: left;">Alternative Forms</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: right;">positive infinity</td>
+<td style="text-align: left;">Inf</td>
+<td style="text-align: left;">Infinity, infinity</td>
+</tr>
+<tr class="even">
+<td style="text-align: right;">negative infinity</td>
+<td style="text-align: left;">-Inf</td>
+<td style="text-align: left;">-Infinity, -infinity</td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;">not a number</td>
+<td style="text-align: left;">NaN</td>
+<td style="text-align: left;"></td>
+</tr>
+</tbody>
+</table>
+<p>These strings are not case sensitive, so <code>inf</code> may also be used for positive infinity, or <code>NAN</code> for not-a-number.</p>
+</section>
+</section>
+<section id="quoted-variable-names" class="level2">
+<h2 class="anchored" data-anchor-id="quoted-variable-names">Quoted variable names</h2>
+<p>In order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.</p>
+<pre><code>"y" &lt;- c(1,2,3) \end{Verbatim}</code></pre>
+</section>
+<section id="line-breaks" class="level2">
+<h2 class="anchored" data-anchor-id="line-breaks">Line breaks</h2>
+<p>The line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.</p>
+<pre><code>y &lt;- 2
+y &lt;-
+3</code></pre>
+<p>Also following R, breaking before the assignment arrow are not allowed, so the following is invalid.</p>
+<pre><code>y
+&lt;- 2 # Syntax Error</code></pre>
+<p>Lines may also be broken in the middle of sequences declared using the <code>c(...)</code> notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a <span class="math inline">\(2 \times 2 \times 3\)</span> array is valid.</p>
+<pre><code>y &lt;-
+structure(c(1,2,3,
+4,5,6,7,8,9,10,11,
+12), .Dim = c(2,2,
+3))</code></pre>
+<p>Because there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as <code>int</code> or <code>real</code>.</p>
+</section>
+<section id="bnf-grammar-for-dump-data" class="level2">
+<h2 class="anchored" data-anchor-id="bnf-grammar-for-dump-data">BNF grammar for dump data</h2>
+<p>A more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.</p>
+<pre><code>definition ::= name &lt;- value optional_semicolon
+
+name ::= char*     | ''' char* '''     | '"' char* '"'
+
+value ::= value&lt;int&gt; | value&lt;double&gt;
+
+value&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |
+'structure' '(' seq&lt;T&gt; ',' ".Dim" '=' seq&lt;int&gt; ')'       | 'structure'
+'(' zero_array&lt;T&gt; ',' ".Dim" '=' seq&lt;int&gt; ')'
+
+seq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;
+
+zero_array&lt;int&gt; ::= "integer" '(' &lt;non-negative int&gt;? ')'
+
+zero_array&lt;real&gt; ::= "double" '(' &lt;non-negative int&gt;? ')'
+
+seq&lt;real&gt; ::= cseq&lt;real&gt;
+
+cseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'
+
+vseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;</code></pre>
+<p>The template parameters <code>T</code> will be set to either <code>int</code> or <code>real</code>. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/json_apdx.html" class="pagination-link" aria-label="JSON Format for CmdStan">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">JSON Format for CmdStan</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/external_code.html" class="pagination-link" aria-label="Using external C++ code">
+        <span class="nav-page-text">Using external C++ code</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/rdump_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/stan_csv_apdx.html b/docs/2_39/cmdstan-guide/stan_csv_apdx.html
new file mode 100644
index 000000000..0c202f475
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/stan_csv_apdx.html
@@ -0,0 +1,1279 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Stan CSV File Format</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/json_apdx.html" rel="next">
+<link href="../cmdstan-guide/print.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan CSV File Format">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Stan CSV File Format</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#stan_csv" id="toc-stan_csv" class="nav-link active" data-scroll-target="#stan_csv">Stan CSV File Format</a>
+  <ul class="collapse">
+  <li><a href="#csv-column-names-and-order" id="toc-csv-column-names-and-order" class="nav-link" data-scroll-target="#csv-column-names-and-order">CSV column names and order</a></li>
+  <li><a href="#mcmc-sampler-csv-output" id="toc-mcmc-sampler-csv-output" class="nav-link" data-scroll-target="#mcmc-sampler-csv-output">MCMC sampler CSV output</a>
+  <ul class="collapse">
+  <li><a href="#sampler-stan-csv-output-file" id="toc-sampler-stan-csv-output-file" class="nav-link" data-scroll-target="#sampler-stan-csv-output-file">Sampler Stan CSV output file</a></li>
+  <li><a href="#diagnostic-csv-output-file" id="toc-diagnostic-csv-output-file" class="nav-link" data-scroll-target="#diagnostic-csv-output-file">Diagnostic CSV output file</a></li>
+  <li><a href="#profiling-csv-output-file" id="toc-profiling-csv-output-file" class="nav-link" data-scroll-target="#profiling-csv-output-file">Profiling CSV output file</a></li>
+  </ul></li>
+  <li><a href="#optimization-output" id="toc-optimization-output" class="nav-link" data-scroll-target="#optimization-output">Optimization output</a></li>
+  <li><a href="#variational-inference-output" id="toc-variational-inference-output" class="nav-link" data-scroll-target="#variational-inference-output">Variational inference output</a></li>
+  <li><a href="#generate-quantities-outputs" id="toc-generate-quantities-outputs" class="nav-link" data-scroll-target="#generate-quantities-outputs">Generate quantities outputs</a></li>
+  <li><a href="#diagnose-method-outputs" id="toc-diagnose-method-outputs" class="nav-link" data-scroll-target="#diagnose-method-outputs">Diagnose method outputs</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stan_csv_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Appendices</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stan_csv_apdx.html">Stan CSV File Format</a></li></ol></nav></header>
+
+
+
+
+<section id="stan_csv" class="level1">
+<h1>Stan CSV File Format</h1>
+<p>The output from all CmdStan methods is in <a href="https://en.wikipedia.org/wiki/Comma-separated_values">CSV format</a>. A Stan CSV file is a data table where the columns are the method and model parameters and quantities of interest. Each row contains one record’s worth of data in plain-text format using the comma character (‘,’) as the field delimiter (hence the name).</p>
+<p>For the Stan CSV files, data is strictly numerical, however, possible values include both positive and negative infinity and “Not-a-Number” which are represented as the strings <code>NaN</code>, <code>inf</code>, <code>+inf</code>, <code>-inf</code>. All other values are written in decimal notation by default with at most 8 digits of precision. The number of significant digits written can be controlled with argument <code>sig_figs=&lt;int&gt;</code>. See more in <a href="../cmdstan-guide/command_line_options.html#output-control.section">Output control arguments section</a>.</p>
+<p>Stan CSV files have a header row containing the column names. They also make extensive use of CSV comments, i.e., lines which begin with the <code>#</code> character. In addition to initial and final comment rows, some methods also put comment rows in the middle of the data table, which makes it difficult to use many of the commonly used CSV parser packages.</p>
+<section id="csv-column-names-and-order" class="level2">
+<h2 class="anchored" data-anchor-id="csv-column-names-and-order">CSV column names and order</h2>
+<p>The data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.</p>
+<p>Stan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is</p>
+<ul>
+<li>The column name consists of the variable name followed by the element indices.</li>
+<li>Indices are delimited by periods (‘.’).</li>
+<li>Indexing is 1-based, i.e., given a dimension of size <span class="math inline">\(N\)</span>, the first element index is <span class="math inline">\(1\)</span> and the last element index is <span class="math inline">\(N\)</span>.</li>
+<li>Tuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).</li>
+</ul>
+<p>Container variables are serialized in <a href="https://eli.thegreenplace.net/2015/memory-layout-of-multi-dimensional-arrays">column major order</a>, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.</p>
+<p>To see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:</p>
+<pre><code> array[2, 3, 4] real foo;</code></pre>
+<p>The Stan CSV file will require 24 columns to output the elements of <code>foo</code>. The first 6 columns will be labeled:</p>
+<pre><code>foo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1</code></pre>
+<p>The final 6 columns will be labeled:</p>
+<pre><code>foo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4</code></pre>
+<p>To see how a tuple would be laid out, consider the following variable:</p>
+<pre><code>tuple(real, array[3] real) bar;</code></pre>
+<p>This will correspond to 4 columns in the CSV file, which are labeled</p>
+<pre><code>bar:1,bar:2.1,bar:2.2,bar:2.3</code></pre>
+</section>
+<section id="mcmc-sampler-csv-output" class="level2">
+<h2 class="anchored" data-anchor-id="mcmc-sampler-csv-output">MCMC sampler CSV output</h2>
+<p>The sample method produces both a Stan CSV output file and a <a href="../cmdstan-guide/mcmc_config.html#sampler-diag-file">diagnostic file</a> which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.</p>
+<p>To see how this works, we show snippets of the output file resulting from the following command:</p>
+<pre><code>./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \
+            data file=bernoulli.data.json \
+            output file=bernoulli_samples.csv</code></pre>
+<section id="sampler-stan-csv-output-file" class="level3">
+<h3 class="anchored" data-anchor-id="sampler-stan-csv-output-file">Sampler Stan CSV output file</h3>
+<p>The sampler output file contains the following:</p>
+<ul>
+<li>Initial comment rows listing full CmdStan argument configuration.</li>
+<li>Header row</li>
+<li>Data rows containing warmup draws, if run with option <code>save_warmup=1</code></li>
+<li>Comment rows for adaptation listing step size and metric used for sampling</li>
+<li>Sampling draws</li>
+<li>Comment rows giving timing information</li>
+</ul>
+<p><strong>Initial comments rows: argument configuration</strong></p>
+<p>All configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as <code>(Default)</code>.</p>
+<p>In the above example the <code>num_samples</code>, <code>num_warmup</code>, and <code>save_warmup</code> arguments were specified, whereas subargument <code>thin</code> is left at its default value, as seen in the initial comment rows:</p>
+<pre><code># stan_version_major = 2
+# stan_version_minor = 24
+# stan_version_patch = 0
+# model = bernoulli_model
+# method = sample (Default)
+#   sample
+#     num_samples = 100
+#     num_warmup = 200
+#     save_warmup = 1
+#     thin = 1 (Default)
+#     adapt
+#       engaged = 1 (Default)
+#       gamma = 0.050000000000000003 (Default)
+#       delta = 0.80000000000000004 (Default)
+#       kappa = 0.75 (Default)
+#       t0 = 10 (Default)
+#       init_buffer = 75 (Default)
+#       term_buffer = 50 (Default)
+#       window = 25 (Default)
+#     algorithm = hmc (Default)
+#       hmc
+#         engine = nuts (Default)
+#           nuts
+#             max_depth = 10 (Default)
+#         metric = diag_e (Default)
+#         metric_file =  (Default)
+#         stepsize = 1 (Default)
+#         stepsize_jitter = 0 (Default)
+# id = 0 (Default)
+# data
+#   file = bernoulli.data.json
+# init = 2 (Default)
+# random
+#   seed = 2991989946 (Default)
+# output
+#   file = bernoulli_samples.csv
+#   diagnostic_file = bernoulli_diagnostics.csv
+#   refresh = 100 (Default)</code></pre>
+<p>Note that when running multi-threaded programs which use <code>reduce_sum</code> for <a href="../cmdstan-guide/parallelization.html">high-level parallelization</a>, the number of threads used will also be included in this initial comment header.</p>
+<p><strong>Column headers</strong></p>
+<p>The CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the <a href="../cmdstan-guide/mcmc_config.html#mcmc_output_csv">output file</a> section of the chapter on MCMC Sampling. The example model <code>bernoulli.stan</code> only contains one parameter <code>theta</code>, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:</p>
+<pre><code>lp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta</code></pre>
+<p>As a second example, we show the output of the <code>eight_schools.stan</code> model on run on example dataset. This model has 3 parameters: <code>mu</code>, <code>theta</code> a vector whose length is dependent on the input data, here <code>N = 8</code>, and <code>tau</code>. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:</p>
+<pre><code>mu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau</code></pre>
+<p><strong>Data rows containing warmup draws</strong></p>
+<p>When run with option <code>save_warmup=1</code>, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is <code>save_warmup=0</code>, this section is usually not present in the output file.</p>
+<p>Here we specified <code>num_warmup=200</code> and left <code>thin</code> at the default value <span class="math inline">\(1\)</span>, therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.</p>
+<pre><code>-6.74827,1,1,1,1,0,6.75348,0.247195
+-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195
+-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195
+-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619
+-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517</code></pre>
+<p><strong>Comment rows for adaptation</strong></p>
+<p>During warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.</p>
+<pre><code># Adaptation terminated
+# Step size = 0.813694
+# Diagonal elements of inverse mass matrix:
+# 0.592879</code></pre>
+<p>As the example bernoulli model only contains a single parameter, and as the default metric is <code>diag_e</code>, the inverse mass matrix is a <span class="math inline">\(1 \times 1\)</span> matrix, and the length of the diagonal vector is also <span class="math inline">\(1\)</span>.</p>
+<p>In contrast, if we run the eight schools example model with metric <code>dense_e</code>, the adaptation comments section lists both the stepsize and the full <span class="math inline">\(10 \times 10\)</span> inverse mass matrix:</p>
+<pre><code># Adaptation terminated
+# Step size = 0.211252
+# Elements of inverse mass matrix:
+# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757
+# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028
+# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353
+# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097
+# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245
+# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507
+# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946
+# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661
+# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104
+# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433</code></pre>
+<p><em>Note that when the sampler is run with arguments <code>algorithm=fixed_param</code>, this section will be missing.</em></p>
+<p><strong>Data rows containing sampling draws</strong></p>
+<p>The output file contains the values for the thinned set draws during sampling. Here we specified <code>num_sampling=100</code> and left <code>thin</code> at the default value <span class="math inline">\(1\)</span>, therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.</p>
+<pre><code>-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093
+-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431
+-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341</code></pre>
+<p><strong>Timing information</strong></p>
+<p>Upon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:</p>
+<pre><code>#
+#  Elapsed Time: 0.005 seconds (Warm-up)
+#                0.002 seconds (Sampling)
+#                0.007 seconds (Total)
+#</code></pre>
+</section>
+<section id="diagnostic-csv-output-file" class="level3">
+<h3 class="anchored" data-anchor-id="diagnostic-csv-output-file">Diagnostic CSV output file</h3>
+<p>The diagnostic file contains the following:</p>
+<ul>
+<li>Initial comment rows listing full CmdStan argument configuration.</li>
+<li>Header row</li>
+<li>Data rows containing warmup draws, if run with option <code>save_warmup=1</code></li>
+<li>Sampling draws</li>
+<li>Comment rows giving timing information</li>
+</ul>
+<p>The columns in this file contain, in order:</p>
+<ul>
+<li>all sampler parameters</li>
+<li>all model parameter estimates (on the unconstrained scale)</li>
+<li>the latent Hamiltonian for each parameter</li>
+<li>the gradient for each parameters</li>
+</ul>
+<p>The labels for the latent Hamiltonian columns are the parameter column label with prefix <code>p_</code> and the labels for the gradient columns are the parameter column label with prefix <code>g_</code>.</p>
+<p>These are the column labels from the file <code>bernoulli_diagnostic.csv</code>:</p>
+<pre><code>lp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta</code></pre>
+</section>
+<section id="profiling-csv-output-file" class="level3">
+<h3 class="anchored" data-anchor-id="profiling-csv-output-file">Profiling CSV output file</h3>
+<p>The profiling information is stored in a plain CSV format with no meta information in the comments.</p>
+<p>Each row represents timing information collected in a <code>profile</code> statement for a given thread. It is possible that some <code>profile</code> statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).</p>
+<p>The columns are as follows:</p>
+<ul>
+<li><code>name</code>, The name of the <code>profile</code> statement that is being timed</li>
+<li><code>thread_id</code>, The thread that executed the <code>profile</code> statement</li>
+<li><code>total_time</code>, The combined time spent executing statements inside the <code>profile</code> which includes calculation with and without automatic differentiation</li>
+<li><code>forward_time</code>, The time spent in the <code>profile</code> statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation</li>
+<li><code>reverse_time</code>, The time spent in the <code>profile</code> statement during the reverse (backward) pass of reverse mode automatic differentiation</li>
+<li><code>chain_stack</code>, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass</li>
+<li><code>no_chain_stack</code>, The number of objects allocated on the non-chaining automatic differentiation stack</li>
+<li><code>autodiff_calls</code>, The total number of times the <code>profile</code> statement was executed with automatic differentiation</li>
+<li><code>no_autodiff_calls</code> - The total number of times the <code>profile</code> statement was executed without automatic differentiation</li>
+</ul>
+</section>
+</section>
+<section id="optimization-output" class="level2">
+<h2 class="anchored" data-anchor-id="optimization-output">Optimization output</h2>
+<ul>
+<li>Config as comments</li>
+<li>Header row</li>
+<li>Penalized maximum likelihood estimate</li>
+</ul>
+</section>
+<section id="variational-inference-output" class="level2">
+<h2 class="anchored" data-anchor-id="variational-inference-output">Variational inference output</h2>
+<ul>
+<li>Config as comments</li>
+<li>Header row</li>
+<li>Adaptation as comments</li>
+<li>Variational estimate</li>
+<li>Sample draws from estimate of the posterior</li>
+</ul>
+</section>
+<section id="generate-quantities-outputs" class="level2">
+<h2 class="anchored" data-anchor-id="generate-quantities-outputs">Generate quantities outputs</h2>
+<ul>
+<li>Header row</li>
+<li>Quantities of interest</li>
+</ul>
+</section>
+<section id="diagnose-method-outputs" class="level2">
+<h2 class="anchored" data-anchor-id="diagnose-method-outputs">Diagnose method outputs</h2>
+<ul>
+<li>Header row</li>
+<li>Gradients</li>
+</ul>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/print.html" class="pagination-link" aria-label="`print` (deprecated): MCMC Output Analysis">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text"><code>print</code> (deprecated): MCMC Output Analysis</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/json_apdx.html" class="pagination-link" aria-label="JSON Format for CmdStan">
+        <span class="nav-page-text">JSON Format for CmdStan</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stan_csv_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/stanc.html b/docs/2_39/cmdstan-guide/stanc.html
new file mode 100644
index 000000000..0a8d4f5ff
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/stanc.html
@@ -0,0 +1,1064 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Translating Stan to C++</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/stansummary.html" rel="next">
+<link href="../cmdstan-guide/diagnose_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Translating Stan to C++">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html"><code>stanc</code>: Translating Stan to C++</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#stanc" id="toc-stanc" class="nav-link active" data-scroll-target="#stanc"><code>stanc</code>: Translating Stan to C++</a>
+  <ul class="collapse">
+  <li><a href="#instantiating-the-stanc-binary" id="toc-instantiating-the-stanc-binary" class="nav-link" data-scroll-target="#instantiating-the-stanc-binary">Instantiating the <code>stanc</code> binary</a></li>
+  <li><a href="#the-stan-compiler-program" id="toc-the-stan-compiler-program" class="nav-link" data-scroll-target="#the-stan-compiler-program">The Stan compiler program</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html"><code>stanc</code>: Translating Stan to C++</a></li></ol></nav></header>
+
+
+
+
+<section id="stanc" class="level1">
+<h1><code>stanc</code>: Translating Stan to C++</h1>
+<p>CmdStan translates Stan programs to C++ using the Stan compiler program which is included in the CmdStan release <code>bin</code> directory as program <code>stanc</code>. One can view the complete stanc documentation in the <a href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html">Stan User’s Guide</a>.</p>
+<p>As of release 2.22, the CmdStan Stan to C++ compiler is written in OCaml. This compiler is called “stanc3” and has has its own repository https://github.com/stan-dev/stanc3, from which pre-built binaries for Linux, Mac, and Windows can be downloaded.</p>
+<section id="instantiating-the-stanc-binary" class="level2">
+<h2 class="anchored" data-anchor-id="instantiating-the-stanc-binary">Instantiating the <code>stanc</code> binary</h2>
+<p>Before the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:</p>
+<pre><code>make bin/stanc</code></pre>
+<p>For Windows:</p>
+<pre><code>make bin/stanc.exe</code></pre>
+<p>This is also done as part of the <code>make build</code> command.</p>
+</section>
+<section id="the-stan-compiler-program" class="level2">
+<h2 class="anchored" data-anchor-id="the-stan-compiler-program">The Stan compiler program</h2>
+<p>The Stan compiler program <code>stanc</code> converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model <code>bernoulli.stan</code>. For Linux and Mac:</p>
+<pre><code>&gt; cd &lt;cmdstan-home&gt;
+&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan</code></pre>
+<p>For Windows:</p>
+<pre><code>&gt; cd &lt;cmdstan-home&gt;
+&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan</code></pre>
+<p>The base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (<code>a--z</code> or <code>A--Z</code>) and contain only alphanumeric characters (<code>a--z</code>, <code>A--Z</code>, and <code>0--9</code>) and underscores (<code>_</code>) and should not conflict with any C++ reserved keyword.</p>
+<p>The C++ code implementing the class is written to the file <code>bernoulli.hpp</code> in the current directory. The final argument, <code>bernoulli.stan</code>, is the file from which to read the Stan program.</p>
+<p>In practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via <code>make</code>, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the <code>.stan</code> omitted. On Windows, replace <code>.stan</code> with <code>.exe</code>, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:</p>
+<pre><code>&gt; make examples/bernoulli/bernoulli</code></pre>
+<p>For Windows:</p>
+<pre><code>&gt; make examples/bernoulli/bernoulli.exe</code></pre>
+<p>The makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable <code>STANCFLAGS</code> can be used to to override the default <a href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html#stanc-args">arguments</a> to stanc, e.g.,</p>
+<pre><code>&gt; make STANCFLAGS="--include-paths=~/foo" examples/bernoulli/bernoulli</code></pre>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/diagnose_config.html" class="pagination-link" aria-label="Diagnosing HMC by Comparison of Gradients">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Diagnosing HMC by Comparison of Gradients</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/stansummary.html" class="pagination-link" aria-label="`stansummary`: MCMC Output Analysis">
+        <span class="nav-page-text"><code>stansummary</code>: MCMC Output Analysis</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/stansummary.html b/docs/2_39/cmdstan-guide/stansummary.html
new file mode 100644
index 000000000..e9070745a
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/stansummary.html
@@ -0,0 +1,1212 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Summarizing MCMC Output</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/diagnose_utility.html" rel="next">
+<link href="../cmdstan-guide/stanc.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Summarizing MCMC Output">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stansummary.html"><code>stansummary</code>: MCMC Output Analysis</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#stansummary" id="toc-stansummary" class="nav-link active" data-scroll-target="#stansummary"><code>stansummary</code>: MCMC Output Analysis</a>
+  <ul class="collapse">
+  <li><a href="#building-the-stansummary-command" id="toc-building-the-stansummary-command" class="nav-link" data-scroll-target="#building-the-stansummary-command">Building the <code>stansummary</code> command</a></li>
+  <li><a href="#running-the-stansummary-program" id="toc-running-the-stansummary-program" class="nav-link" data-scroll-target="#running-the-stansummary-program">Running the <code>stansummary</code> program</a>
+  <ul class="collapse">
+  <li><a href="#sampler-parameters" id="toc-sampler-parameters" class="nav-link" data-scroll-target="#sampler-parameters">Sampler parameters</a></li>
+  <li><a href="#model-parameters-and-quantities-of-interest" id="toc-model-parameters-and-quantities-of-interest" class="nav-link" data-scroll-target="#model-parameters-and-quantities-of-interest">Model parameters and quantities of interest</a></li>
+  </ul></li>
+  <li><a href="#command-line-options" id="toc-command-line-options" class="nav-link" data-scroll-target="#command-line-options">Command-line options</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stansummary.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/stanc.html">Tools and Utilities</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/stansummary.html"><code>stansummary</code>: MCMC Output Analysis</a></li></ol></nav></header>
+
+
+
+
+<section id="stansummary" class="level1">
+<h1><code>stansummary</code>: MCMC Output Analysis</h1>
+<p>The CmdStan <code>stansummary</code> program reports statistics for one or more sampler chains over all sampler and model parameters and quantities of interest. The statistics reported include both summary statistics of the estimates and diagnostic statistics on the sampler chains, reported in the following order:</p>
+<ul>
+<li>Mean - sample mean</li>
+<li>MCSE - Monte Carlo Standard Error, a measure of the amount of noise in the sample</li>
+<li>StdDev - sample standard deviation - the standard deviation around the sample mean.</li>
+<li>MAD - Median Absolute Deviation - the median absolute deviation around the sample median.</li>
+<li>Quantiles - default 5%, 50%, 95%</li>
+<li>ESS_bulk</li>
+<li>ESS_tail</li>
+<li>ESS_bulk/s - Bulk ESS per second</li>
+<li>R_hat - <span class="math inline">\(\hat{R}\)</span> statistic, a MCMC convergence diagnostic</li>
+</ul>
+<p>When reviewing the <code>stansummary</code> output, it is important to check the final three output columns first - these are the diagnostic statistics on MCMC convergence and effective sample size. A <span class="math inline">\(\hat{R}\)</span> statistic of greater than <span class="math inline">\(1\)</span> indicates potential convergence problems and that the sample is not presentative of the target posterior, thus the estimates of the mean and all other summary statistics are likely to be invalid. A value <span class="math inline">\(1.01\)</span> can be used as generic threshold to decide whether more iterations or further convergence analysis is needed, but other thresholds can be used depending on the specific use case.</p>
+<p>Estimation by sampling produces an approximate value for the model parameters; the MCSE statistic indicates the amount of uncertainty in the estimate. Therefore MCSE column is placed next to the sample mean column, in order to make it easy to compare this sample with others.</p>
+<p>For more information, see the <a href="https://mc-stan.org/docs/reference-manual/analysis.html">Posterior Analysis</a> chapter of the Stan Reference Manual which describes both the theory and practice of MCMC estimation techniques.</p>
+<p>The statistics - Mean, StdDev, MAD, and Quantiles - are computed directly from all draws across all chains. The diagnostic statistics - ESS_bulk, ESS_tail, and R_hat are computed from the rank-normalized, folded, and splitted chains according to the definitions by <span class="citation" data-cites="Vehtari+etal:2021:Rhat">Vehtari et al. (<a href="#ref-Vehtari+etal:2021:Rhat" role="doc-biblioref">2021</a>)</span>. the MCSE statistic is computed using split chain R_hat and autocorrelations. The summary statistics and the algorithms used to compute them are described in sections <a href="https://mc-stan.org/docs/reference-manual/analysis.html#notation-for-samples-chains-and-draws">Notation for draws</a> and <a href="https://mc-stan.org/docs/reference-manual/analysis.html#effective-sample-size.section">Effective sample size</a>.</p>
+<section id="building-the-stansummary-command" class="level2">
+<h2 class="anchored" data-anchor-id="building-the-stansummary-command">Building the <code>stansummary</code> command</h2>
+<p>The CmdStan makefile task <code>build</code> compiles the <code>stansummary</code> utility into the <code>bin</code> directory. It can be compiled directly using the makefile as follows:</p>
+<pre><code>&gt; cd &lt;cmdstan-home&gt;
+&gt; make bin/stansummary</code></pre>
+</section>
+<section id="running-the-stansummary-program" class="level2">
+<h2 class="anchored" data-anchor-id="running-the-stansummary-program">Running the <code>stansummary</code> program</h2>
+<p>The <code>stansummary</code> utility processes one or more output files from a set of chains from one run of the HMC sampler. To run <code>stansummary</code> on the output file or files generated by a run of the sampler, on Mac or Linux:</p>
+<pre><code>&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;</code></pre>
+<p>On Windows, use backslashes to call the <code>stansummary.exe</code>.</p>
+<pre><code>&lt;cmdstan-home&gt;\bin\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;</code></pre>
+<p>For example, after running 4 chains to fit the example model <code>eight_schools.stan</code> to the supplied example data file, we run <code>stansummary</code> on the resulting Stan CSV output files to get the following report:</p>
+<pre><code>&gt; bin/stansummary eight_*.csv
+Inference for Stan model: eight_schools_model
+4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.
+
+Warmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total
+Sampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total
+
+                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat
+
+lp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0
+accept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0
+stepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan
+treedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0
+n_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0
+divergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan
+energy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0
+
+mu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0
+theta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0
+theta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0
+theta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0
+theta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0
+theta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0
+theta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0
+theta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0
+theta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0
+tau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0
+
+Samples were drawn using hmc with nuts.
+For each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)
+and for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.
+At convergence R_hat will be very close to 1.00.</code></pre>
+<p>The console output information consists of</p>
+<ul>
+<li>Model, chains, and timing summaries</li>
+<li>Sampler parameter statistics</li>
+<li>Model parameter statistics</li>
+<li>Sampling algorithm - either nuts (shown here) or static HMC.</li>
+</ul>
+<p>There is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g.&nbsp;if there are no divergent transitions.</p>
+<section id="sampler-parameters" class="level3">
+<h3 class="anchored" data-anchor-id="sampler-parameters">Sampler parameters</h3>
+<p>The initial Stan CSV columns provide information on the sampler state for each draw:</p>
+<ul>
+<li><code>lp__</code> - the total log probability density (up to an additive constant) at each sample</li>
+<li><code>accept_stat__</code> - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory</li>
+<li><code>stepsize__</code> - integrator step size</li>
+<li><code>treedepth__</code> - depth of tree used by NUTS (NUTS sampler)</li>
+<li><code>n_leapfrog__</code> - number of leapfrog calculations (NUTS sampler)</li>
+<li><code>divergent__</code> - has value <code>1</code> if trajectory diverged, otherwise <code>0</code>. (NUTS sampler)</li>
+<li><code>energy__</code> - value of the Hamiltonian</li>
+<li><code>int_time__</code> - total integration time (static HMC sampler)</li>
+</ul>
+<p>Because we ran the NUTS sampler, the above summary reports sampler parameters <code>treedepth__</code>, <code>n_leapfrog__</code>, and <code>divergent__</code>; the static HMC sampler would report <code>int_time__</code> instead.</p>
+</section>
+<section id="model-parameters-and-quantities-of-interest" class="level3">
+<h3 class="anchored" data-anchor-id="model-parameters-and-quantities-of-interest">Model parameters and quantities of interest</h3>
+<p>The remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The <code>eight_schools.stan</code> program parameters block contains the following parameter variable declarations:</p>
+<pre><code>  real mu;
+  array[J] real theta;
+  real&lt;lower=0&gt; tau;</code></pre>
+<p>In the example data, <code>J</code> is <span class="math inline">\(8\)</span>; therefore the stansummary listing reports on <code>theta[1]</code> through <code>theta[8]</code>.</p>
+</section>
+</section>
+<section id="command-line-options" class="level2">
+<h2 class="anchored" data-anchor-id="command-line-options">Command-line options</h2>
+<p>The <code>stansummary</code> command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the <code>-h</code> or <code>--help</code> option, the program prints the usage message to the console and exits.</p>
+<pre><code>Report statistics for one or more Stan CSV files from a HMC sampler run.
+Example:  stansummary model_chain_1.csv model_chain_2.csv
+Options:
+  -a, --autocorr [n]          Display the chain autocorrelation for the n-th
+                              input file, in addition to statistics.
+  -c, --csv_filename [file]   Write statistics to a CSV file.
+  -h, --help                  Produce help message, then exit.
+  -p, --percentiles [values]  Percentiles to report as ordered set of
+                              comma-separated numbers from (0.1,99.9), inclusive.
+                              Default is 5,50,95.
+  -s, --sig_figs [n]          Significant figures reported. Default is 2.
+                              Must be an integer from (1, 18), inclusive.
+  -i, --include_param [name]  Include the named parameter in the summary output.
+                              By default, all parameters in the file are summarized,
+                              passing this argument one or more times will filter
+                              the output down to just the requested arguments.</code></pre>
+<p>Both short an long option names are allowed. Short names are specified as <code>-&lt;o&gt; &lt;value&gt;</code>; long option names can be specified either as <code>--&lt;option&gt;=&lt;value&gt;</code> or <code>--&lt;option&gt; &lt;value&gt;</code>.</p>
+<p>The <code>--percentiles</code> argument can also be passed an empty string <code>""</code>, which results in no percentiles being displayed in the output of the command.</p>
+<p>The amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument <code>sig_figs</code>. The default sampler output precision is 8. The <code>--sig_figs</code> argument to the stansummary program should not exceed the <code>sig_figs</code> argument to the sampler.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Vehtari+etal:2021:Rhat" class="csl-entry" role="listitem">
+Vehtari, Aki, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian Bürkner. 2021. <span>“Rank-Normalization, Folding, and Localization: An Improved <span class="math inline">\(\widehat{R}\)</span> for Assessing Convergence of <span>MCMC</span>.”</span> <em>Bayesian Analysis</em> 16: 667–718.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/stanc.html" class="pagination-link" aria-label="`stanc`: Translating Stan to C++">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text"><code>stanc</code>: Translating Stan to C++</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/diagnose_utility.html" class="pagination-link" aria-label="`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences">
+        <span class="nav-page-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stansummary.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/cmdstan-guide/variational_config.html b/docs/2_39/cmdstan-guide/variational_config.html
new file mode 100644
index 000000000..96e9ed10f
--- /dev/null
+++ b/docs/2_39/cmdstan-guide/variational_config.html
@@ -0,0 +1,1240 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>ADVI for Variational Inference</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../cmdstan-guide/generate_quantities_config.html" rel="next">
+<link href="../cmdstan-guide/pathfinder_config.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="ADVI for Variational Inference">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/variational_config.html">Variational Inference using ADVI</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Getting Started</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/installation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">CmdStan Installation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/example_model_data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Example Model and Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/compiling_stan_programs.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compiling a Stan Program</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Running CmdStan</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/command_line_options.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Command-Line Interface Overview</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/mcmc_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling using Hamiltonian Monte Carlo</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/optimize_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/pathfinder_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder Method for Approximate Bayesian Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/variational_config.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Variational Inference using ADVI</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/generate_quantities_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Generating Quantities of Interest from a Fitted Model</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/laplace_sample_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/log_prob_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Extracting log probabilities and gradients for diagnostics</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_config.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnosing HMC by Comparison of Gradients</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Tools and Utilities</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stanc</code>: Translating Stan to C++</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stansummary.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>stansummary</code>: MCMC Output Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/diagnose_utility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>diagnose</code>: Diagnosing Biased Hamiltonian Monte Carlo Inferences</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/print.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text"><code>print</code> (deprecated): MCMC Output Analysis</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/stan_csv_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan CSV File Format</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/json_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">JSON Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/rdump_apdx.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">RDump Format for CmdStan</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../cmdstan-guide/external_code.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using external C++ code</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#variational-inference-using-advi" id="toc-variational-inference-using-advi" class="nav-link active" data-scroll-target="#variational-inference-using-advi">Variational Inference using ADVI</a>
+  <ul class="collapse">
+  <li><a href="#variational-algorithms" id="toc-variational-algorithms" class="nav-link" data-scroll-target="#variational-algorithms">Variational algorithms</a></li>
+  <li><a href="#configuration" id="toc-configuration" class="nav-link" data-scroll-target="#configuration">Configuration</a></li>
+  <li><a href="#csv-output" id="toc-csv-output" class="nav-link" data-scroll-target="#csv-output">CSV output</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/variational_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../cmdstan-guide/command_line_options.html">Running CmdStan</a></li><li class="breadcrumb-item"><a href="../cmdstan-guide/variational_config.html">Variational Inference using ADVI</a></li></ol></nav></header>
+
+
+
+
+<section id="variational-inference-using-advi" class="level1">
+<h1>Variational Inference using ADVI</h1>
+<p>Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) <span class="citation" data-cites="kucukelbir:2017">Kucukelbir et al. (<a href="#ref-kucukelbir:2017" role="doc-biblioref">2017</a>)</span>. ADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO (evidence lower bound). ADVI optimizes the ELBO in the real-coordinate space using <a href="https://mc-stan.org/docs/reference-manual/variational.html#stochastic-gradient-ascent">stochastic gradient ascent</a>. The measures of convergence are similar to the relative tolerance scheme of Stan’s <a href="https://mc-stan.org/docs/reference-manual/optimization.html">optimization algorithms</a>.</p>
+<p>The algorithm progression consists of an adaptation phase followed by a sampling phase. The adaptation phase finds a good value for the step size scaling parameter <code>eta</code>. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. The algorithm runs until the mean change in ELBO drops below the specified tolerance.</p>
+<p>The full set of configuration options available for the <code>variational</code> method is available by using the <code>variational help-all</code> subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.</p>
+<p>The following is a minimal call to Stan’s variational inference algorithm using defaults for everything but the location of the data file.</p>
+<pre><code>&gt; ./bernoulli variational data file=bernoulli.data.R</code></pre>
+<p>Executing this command prints both output to the console and to a csv file.</p>
+<p>The first part of the console output reports on the configuration used: the default option <code>algorithm=meanfield</code> and the default tolerances for monitoring the algorithm’s convergence.</p>
+<pre><code>method = variational
+  variational
+    algorithm = meanfield (Default)
+      meanfield
+    iter = 10000 (Default)
+    grad_samples = 1 (Default)
+    elbo_samples = 100 (Default)
+    eta = 1 (Default)
+    adapt
+      engaged = true (Default)
+      iter = 50 (Default)
+    tol_rel_obj = 0.01 (Default)
+    eval_elbo = 100 (Default)
+    output_samples = 1000 (Default)
+id = 1 (Default)
+data
+  file = bernoulli.data.json
+init = 2 (Default)
+random
+  seed = 2790599354 (Default)
+output
+  file = output.csv (Default)
+  diagnostic_file =  (Default)
+  refresh = 100 (Default)
+  sig_figs = 8 (Default)
+  profile_file = profile.csv (Default)
+  save_cmdstan_config = false (Default)
+num_threads = 1 (Default)</code></pre>
+<p>After the configuration has been displayed, informational and timing messages are output:</p>
+<pre><code>------------------------------------------------------------
+EXPERIMENTAL ALGORITHM:
+  This procedure has not been thoroughly tested and may be unstable
+  or buggy. The interface is subject to change.
+------------------------------------------------------------
+
+Gradient evaluation took 2.1e-05 seconds
+1000 transitions using 10 leapfrog steps per transition would take 0.21 seconds.
+Adjust your expectations accordingly!</code></pre>
+<p>The rest of the output describes the progression of the algorithm. An adaptation phase finds a good value for the step size scaling parameter <code>eta</code>. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. That is, while the algorithm appears to have converged in <span class="math inline">\(\sim\)</span> 250 iterations, the algorithm runs for another few thousand iterations until mean change in ELBO drops below the default tolerance of 0.01.</p>
+<pre><code>Begin eta adaptation.
+Iteration:   1 / 250 [  0%]  (Adaptation)
+Iteration:  50 / 250 [ 20%]  (Adaptation)
+Iteration: 100 / 250 [ 40%]  (Adaptation)
+Iteration: 150 / 250 [ 60%]  (Adaptation)
+Iteration: 200 / 250 [ 80%]  (Adaptation)
+Success! Found best value [eta = 1] earlier than expected.
+
+Begin stochastic gradient ascent.
+  iter             ELBO   delta_ELBO_mean   delta_ELBO_med   notes
+   100           -6.131             1.000            1.000
+   200           -6.458             0.525            1.000
+   300           -6.300             0.359            0.051
+   400           -6.137             0.276            0.051
+   500           -6.243             0.224            0.027
+   600           -6.305             0.188            0.027
+   700           -6.289             0.162            0.025
+   800           -6.402             0.144            0.025
+   900           -6.103             0.133            0.025
+  1000           -6.314             0.123            0.027
+  1100           -6.348             0.024            0.025
+  1200           -6.244             0.020            0.018
+  1300           -6.293             0.019            0.017
+  1400           -6.250             0.017            0.017
+  1500           -6.241             0.015            0.010   MEDIAN ELBO CONVERGED
+
+Drawing a sample of size 1000 from the approximate posterior...
+COMPLETED.</code></pre>
+<section id="variational-algorithms" class="level2">
+<h2 class="anchored" data-anchor-id="variational-algorithms">Variational algorithms</h2>
+<p>Stan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option <code>algorithm=meanfield</code>. The <code>algorithm</code> argument specifies the variational algorithm.</p>
+<ul>
+<li><p><code>algorithm=meanfield</code> - Use a fully factorized Gaussian for the approximation. This is the default algorithm.</p></li>
+<li><p><code>algorithm=fullrank</code> Use a Gaussian with a full-rank covariance matrix for the approximation.</p></li>
+</ul>
+</section>
+<section id="configuration" class="level2">
+<h2 class="anchored" data-anchor-id="configuration">Configuration</h2>
+<ul>
+<li><p><code>iter=&lt;int&gt;</code> Maximum number of iterations. Must be <span class="math inline">\(&gt; 0\)</span>. Default is <span class="math inline">\(10000\)</span>.</p></li>
+<li><p><code>grad_samples=&lt;int&gt;</code> Number of samples for Monte Carlo estimate of gradients. Must be <span class="math inline">\(&gt; 0\)</span>. Default is <span class="math inline">\(1\)</span>.</p></li>
+<li><p><code>elbo_samples=&lt;int&gt;</code> Number of samples for Monte Carlo estimate of ELBO (objective function). Must be <span class="math inline">\(&gt; 0\)</span>. Default is <span class="math inline">\(100\)</span>.</p></li>
+<li><p><code>eta=&lt;double&gt;</code> Stepsize weighting parameter for adaptive stepsize sequence. Must be <span class="math inline">\(&gt; 0\)</span>. Default is <span class="math inline">\(1.0\)</span>.</p></li>
+<li><p><code>adapt</code> Warmup Adaptation keyword, takes sub-arguments:</p>
+<ul>
+<li><p><code>engaged=&lt;boolean&gt;</code> Adaptation engaged? Valid values: <code>[true, false]</code>. Default is <code>true</code>.</p></li>
+<li><p><code>iter=&lt;int&gt;</code> Maximum number of adaptation iterations. Must be <span class="math inline">\(&gt; 0\)</span>. Default is <span class="math inline">\(50\)</span>.</p></li>
+</ul></li>
+<li><p><code>tol_rel_obj=&lt;double&gt;</code> Convergence tolerance on the relative norm of the objective. Must be <span class="math inline">\(&gt; 0\)</span>. Default is <span class="math inline">\(0.01\)</span>.</p></li>
+<li><p><code>eval_elbo=&lt;int&gt;</code> Evaluate ELBO every Nth iteration. Must be <span class="math inline">\(&gt; 0\)</span>. Default is 100.</p></li>
+<li><p><code>output_samples=&lt;int&gt;</code> Number of posterior samples to draw and save. Must be <span class="math inline">\(&gt; 0\)</span>. Default is 1000.</p></li>
+</ul>
+</section>
+<section id="csv-output" class="level2">
+<h2 class="anchored" data-anchor-id="csv-output">CSV output</h2>
+<p>The output file consists of the following pieces of information:</p>
+<ul>
+<li><p>The full set of configuration options available for the <code>variational</code> method is reported at the beginning of the sampler output file as CSV comments.</p></li>
+<li><p>The first three output columns are labelled <code>lp__</code>, <code>log_p__</code>, <code>log_g__</code>, the rest are the model parameters.</p></li>
+<li><p>The stepsize adaptation information is output as CSV comments following column header row.</p></li>
+<li><p>The following line contains the mean of the variational approximation.</p></li>
+<li><p>The rest of the output contains <code>output_samples</code> number of draws sampled from the variational approximation.</p></li>
+</ul>
+<p>To illustrate, we call Stan’s variational inference on the example model and data:</p>
+<pre><code>&gt; ./bernoulli variational data file=bernoulli.data.R</code></pre>
+<p>By default, the output file is <code>output.csv</code>.</p>
+<p>The output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:</p>
+<pre><code># stan_version_major = 2
+# stan_version_minor = 23
+# stan_version_patch = 0
+# model = bernoulli_model
+# method = variational
+#   variational
+#     algorithm = meanfield (Default)
+#       meanfield
+#     iter = 10000 (Default)
+#     grad_samples = 1 (Default)
+#     elbo_samples = 100 (Default)
+#     eta = 1 (Default)
+#     adapt
+#       engaged = true (Default)
+#       iter = 50 (Default)
+#     tol_rel_obj = 0.01 (Default)
+#     eval_elbo = 100 (Default)
+#     output_samples = 1000 (Default)
+...</code></pre>
+<p>Next, the column header row:</p>
+<pre><code>lp__,log_p__,log_g__,theta</code></pre>
+<p>Additional comments provide stepsize adaptation information:</p>
+<pre><code># Stepsize adaptation complete.
+# eta = 1</code></pre>
+<p>Followed by the data rows. The first line is special — it is the mean of the variational approximation.</p>
+<pre><code>0,0,0,0.214911</code></pre>
+<p>That is, the estimate for <code>theta</code> given the data is <code>0.2</code>.</p>
+<p>The rest of the output contains <code>output_samples</code> number of draws samples from the variational approximation.</p>
+<p>The following is a sample based on this approximation:</p>
+<pre><code>0,-14.0252,-5.21718,0.770397
+0,-7.05063,-0.10025,0.162061
+0,-6.75031,-0.0191099,0.241606
+...</code></pre>
+<p>The header indicates the unnormalized log probability with <code>lp__</code>. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-kucukelbir:2017" class="csl-entry" role="listitem">
+Kucukelbir, Alp, Dustin Tran, Rajesh Ranganath, Andrew Gelman, and David M Blei. 2017. <span>“Automatic Differentiation Variational Inference.”</span> <em>Journal of Machine Learning Research</em>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../cmdstan-guide/pathfinder_config.html" class="pagination-link" aria-label="Pathfinder Method for Approximate Bayesian Inference">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Pathfinder Method for Approximate Bayesian Inference</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../cmdstan-guide/generate_quantities_config.html" class="pagination-link" aria-label="Generating Quantities of Interest from a Fitted Model">
+        <span class="nav-page-text">Generating Quantities of Interest from a Fitted Model</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/variational_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference-2_39.pdf b/docs/2_39/functions-reference-2_39.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..4a0ef0f1a01e74c5bd6cd05e86f5a0f82b9e16dd
GIT binary patch
literal 991980
zcmcG#V~}V;kS5x;ZQHhO8@FxSwr$(CZQr)7+qS#6XBIILyAf~xyz{3jD)LmFlbQ8Z
zR)SPPM2wb^junb@eraS4ik*Ogz|PPTiiZb^UdF`M%-NiPiII)qe?Cz3ViwlUCXNL3
zV%7%ECL$(AcE%=9e0)$&&W<JqHc;*xA@RKOAcP2^UcMvA3N&NTFmBB!JViND*3ly-
z$I%f}|F-|#pp%e^K?~o_o-#9IcoGSt4}KPzNmje=Ag*i#@|FM(U|kU*2f6IAmWX$z
zwN3Sw4>Trpy)e<uIdQtYkm27e8W1~SHhD^@<AUR6*bIS|tp|mQJ@qbLqX1;Yl}(;w
zXp6VnVr4G_HdL-g^Dnl&5%*2kDbq>Alc3g2e^jtWJ&-8OF@Xq`Qx8kP{TTwmBEKVR
z7VfH6{bcA<vS~Z<wt}stBcFFgTYdWaZj*0$l^=sLu{Hi*zxdnx5B9M97dEMQ*qad0
zYsec~nix4l(JQ+cI{)X0q>X`@2^77$h4C*FnORt%=p{@n%*>ttXJ5#|*-62~QP|GL
z-p<y<)|r3<ieA{x+Rjng-oWU0J`odF3nLRHaiQNCWeuFH2pE}u*HrpHQek9Z;^6pC
zvHX8a<;usxSw*$ewZ;WC`(xu}@^n?*#wJA&Yh)DA7%D>;LJ~rfKmY^S01UDSrbLDi
zvZ7cirE*dR42g<aGE&9zh6+#+MI;$a5G-M$0(bM{YsO8EiXgtARN1BURCCTz(yi{=
zb=7zG0dDTYHUtRZzdZ3wk<;K0w5^wFnL9Fz*tFc(<{Q49ExFz5@@=cyY|HbT-ecq6
zQ84Vwh$gU-c0`zDRh1*DMr>uvivv=`aL524QAkqDWZIs@iYtlBVvX(D)0AY&_Y%da
zAjS!d8QF{yvIqBP*+h-ifwi?n*fA=a(r#>8Zp*mCbpQZ-!aXyi>7TJ;jjs5bDq=`b
z66K-N(l7m@^A&H-29&x@aoyVD%vu<$&Jca*0QNw<gwe7#m4wpNG|CSX?KFZKJ<K>+
zl7^4QzL=Ulao6VgZ)niIvH<o-BW#<qVx0}4kT({_wW2g?H!#F5)-X*Csh8#i8KY{S
zD**Q7Fl^dl$lD6CCn@uL)<mS6Hxb`dOjY65<*nDYgjhoM-~j7H+s3B7AJJFfrrJmP
zAH0kNQPXs-)$D&S(W(l2aM6L|dD89l#E>=hWYdlHW?OfW$;~Gc9a}pp*fQ+30nRh}
z8gog~kTU$<btP~bB^)ug)#P0>tlRnh%M};AIa(6vst5=<8En|k%ZMUEjJNqNxLZ)z
zp%4&d&}A%Squ-R%`P&c&3zZUAWTMUTtwBIywoETdvA86`o1TcsEhjcrRW!{;eSkqi
zl$tng_;dQgU=0n*_zoX|8e0OGCeUpO`stgOx_dL_rbVCP*h8(xJ1r}jG##)3cZr)u
zI*YTHhx(B~mE>gTh_x>Y{(yM{8g(i9_|wrCJw4R(0XkapakId_0q&75ZCYz%pQD`l
z4fkX4Dz0ccDRKdD&;Vh?Olo0;(D|=n6qvX%8MhjiLf%4<B~jKaaIFg%C=;zvS&f4c
zt1b(tLA1jRK@N32IX42jlB86e>NIx5Qe|b@w*i{X@9gAmNE|%qbN{WzwzeWkw*>JF
zs|?xh1NTbk2Cy(g{tJ@N4L|Q~?*R%D;JCjg(^wvDTNT0%V4DJzDQ8_vN{EF5e9Yce
zOf*Gp0cHriPTZ0(T8KgMQV1x!4X@cKQ^1^+rrHnHFGrTPpxY8pItBf!<Lzi^_}x(f
zmW?assvGfS68M(No1eVx5Bhe<P@B)h4DmG4X3-TN5i)2&L3k_h>MDsTVl;e-SWKBT
z%ux<Qf~UC_Q)JZ7N)Ka?X;m7L6cv^N>0qna&i(B?jYSEWu?sh~^Sidr_q*&1Qt<+;
z{lTRGYaeBK9fYF6-w&7gt$y@&fAxAtA(}KAHbuxU3gCdeJ|B|~ANAC|*nYTQ?bsLu
zVhMT^_q2#b2KRJf)RLAzB$f$1@1)DD`>6eTBsVq{c~HRb$M-$XU;=gdzWDz*Eq>n2
zB3A!=qfsx0#7?78=fVR7Em^PV*%qbI*&BSkh=8(J_x%x$CD5_18Uqe={CO;6Ulb<i
ze1GJIfZ2lr33E)cnBhwXF?m0iuOJ!~9>&h|1@D&Itiug?f-L1Um!t_69k=r8zcehg
zARUui$g6wX`n(}}`|_G5zKx3pR<lM*of=i7UDpuCijA*Idu{phdi!u`2l|sEfD9R+
zpz!xmIf0W@#K-^Z@yAhD<HS$>?iM>4`gLFh3_2GzseAY(i+b?)r53SO$m+jo3axE*
z?1l6Mnh`e>i-iqebdAD)V@<~S1PzQ1|2k^K6c75!Qghx@zMXe!%>3+-4Mrf>E8n1j
zSj$SzLq+BxYIT;xvfz@g(jil>`J!6@zDMi?&06Pb0}%_2$m1vu!1H?pWF6yVqGf7;
zl*GSc)DmJ>YUa$PNQ)e_?v9{-dcw+AJ653(LAu-P1m}GVlY6Yb-EPQq3;uk~5U$aa
zkMjzYo)9A;ESCLLMcuc2)qeV6NZn(s&!}A@A@&Zlx25=gcEB|Hy1n(OBcQg&;WW|0
z-M~S!-pkWHt?1X0ynVPSV0NHFvS^l~_aA_eKS@>?qJ9%4zCng*IG!~c){n3iA)Naw
za4@D>R}g2{JSNajykN*VV}aI)o1qUu-p-we<XnC084PYUd{MUpjm=}lKuZ|}@sNLu
z2IumEXFdIe0kY}S4=i~J)F^UbGzEXv*Jso0ietemVDAmAcsB+EJe1%yn&C??mdA`=
z)_nOubK%dEj@G}~KG^2MD7^{%F+pYU+bIHaM!Fe!2IHcGM*FABMBd$)xP)JD_6})G
zbN7O659%kdY&`18Xs7-i1@VRDs0a8js#bFka1Bf_3N}itXBdOGYwl+WGd-(bH{enf
zw9<2vB^<27R0FPS>qhwTOO%9)gZRP>IqfQB-Eo4G^s;>yUT>&zc!e+i?m{B#f&l9%
zG=WR))$N-HMrx^favrjnP2+QcV6X7Yj7!XKu)mH4EWviNN+Ev1I4eDB<PZ-Oz_aO-
z&1XWu@iWU>Ct$-kGlY?7jtbzX6RIkqdbYkt7mamjS(U-e4WvKI^89YG9z`OyqyW1)
zHbl4UwY?0E(tisn*sfTIS!H>%(0Lp}XYIRy-3UY=moPwea4L#H`I3Yjdu`c|0%+Wu
zO*2tWUz!$Xi_r|F(a4|Q=nH*3GO9Qg_mU+eXlWe6(?|%A*FIYRiQWBF;9zc7qAW(k
zx{1^<JbzV+APJNJ44?}YQ>rwkeFwtm#sIcGa~Ze@3F0{wS0_!#wZ5l<mH?Y<cSb56
zHwI-}U>-AJpPtAMm~_QFTN9+JFKwda=uvaaem-TunDK^nTO(-W7cFKCqyZ_w0`D<p
z)LNPiHrH}wDu~=+5uEu2Usq_v_!;_OT6ZVPbu76;vtT?mC=m1p?4ezww+=!ryP=>Q
zlX1O?OD_d9_3%@m;rn~xbZm#NKs!{>BF<!+m&HJUN!a4NLrVe1zqSH6Jj1zVQ}eAm
zJ_15aj{cGAXeso1&7!AHZbNpf#|aWDtPl=7ds02H5}(hwvjRL0h$#cUz$H%xduwtB
z@IX5c=5L{fPHI(U=lGYRrXfETWDCR-1~gVsYK?M_xv~)<0r=n}LD-Ub__RcyhE9&h
zQNRT9z~Dku>6g>+vnf(5lQT$)*XoVtfV@FLf-DJd5v<A*8!=COklSMm=G~WFJU(h}
zI7XjootJVIs6Zx<u0T4%g0-^d(*SsOh`Q(G^2^WQFM307YX~DjuZ!DUFV)sjSfK{u
zYjnw%9t-lLz9GrO1nE3fMLo_-YS|WXV_7&-c6Hajc)hxk{-_3C#Ta+%q`_F1i!lIr
zPFAIz?aNz)#B3|K<tJ~e9RB0Q$rDcYvVj15J-D0O2Nzg<y+ZZ8XR+ZJye^ZiFhp0j
zcSK#qdV_&9j4>m#G<o;tJnVF+8rXRoQXTy#xed-z6;iz>ZS=U)8H1?BwL3qYhQFq=
zO}43@|3v%>Sf90&i&G{VY+FIR2)z!KI7fF>>C?>$S32pENx%jV4<GQ@r7h^|$*ij;
z_<EHo`Y${$F^p=U6f4R)C;$pO5YYSD=e@%%bn4|^fizv4Xc5Z59PST6qO|PI8l@%A
zu<v!w`R?5;M^IfGx^2o`Zh$p%ARy%Zsnz4u!{zB8J5%n#>C->a?A|^-^1^{7UM5&D
zuT{5T0nS)2VRC<Y)!uD!GgQ6PB;>nwbWXVW^I2NKPRCg=)^wuu1%@-d7a5<O@{~jm
z?u$M6ftw^7ufyZ0f|-2#UXLtZ0T-4Yt;yFicS^OmisIj^jvPhRZyppS;uQ##$?My3
zo=LY1vNzXxA3EZeCSH_THRVxv^fX&N`n*GiQ{xOAgf6$6Mn0{(vz)|%8zwJef=p1E
zo`99}agjz*av8i{0uQ=*cb<TyWlI%0)F7zg3z_y7>2}VUi|-(+rcZS%$$M+n*kWjH
z?{e|o%fw2-v1<wWy<s07JI&2*KAgRPGn5BcdJJZB?L*jGLYg8?JS#WUK!C#^MuyKp
z(dan?jz>!!8b2F8RgdmPHEPEkzHzTUBHZCNKeCw7(L7Cn9C~z2fG4>y8o3?5!^+G@
zJ}#le$!60+wd8elj&qjuO%r6eLCa<dDDBw>#Ixyf<YrF!ST)R=Cmzyt)d0!odHXsx
zZ^s57L0$*tD95lNSJbAQH~mrPkRKJ+gHj;#sHr*7kVw!b?l6#{jK9L)TBfS8m$KGp
z)*RLTMW)@o>AVNDRT*bqUvX1(^<>JF>sb<KS(0!kkh9NH9#&fs%(2U8Ud^kX$DMln
zrmo<a=1r%oii{Ri(sR<)>{VOU=c~l#0LfMvLx0F*+T9#ZJ;P$X*^3dfR|Mq!Mav}a
zA|iY+N&dT2KtJaedon{QC@$~X1m>#+*9nIG?`Vy3O@sK_b>D4T{0I5DsLz@{;mRDt
z{1k~!DM&TNk*D*pY{cwCoet44yAnSBNa^gQpT9jk>e|4&_c9XLNlFCm_Qt%}{49vq
zj^dsLPeaQju|Tf+O;}L<RX*Ev^}0t_auDG*Xx72EVZCXcsA@}&MoZkcla6DP-z7A8
zV-1Qw(Pus<qlhX_D3J8HQ5p&g5>am2j%T>H0x|X)@64j|A^|^~b^M||uyG=$?)BqZ
zuc(C<m9z$&ZAD)Q|0r?5vQU9)+X*Nt(#DY*?$Y{*5QF2zV{t|2>P(X|4lb(GPhc;I
zUjB8H<#=2L*_M8sP}pf329j-DMsAv;XVJ{3CGV7~VoMLP&-G(5fcXOk#(4S%kMD7E
z_!PM<M#^dg6pVBKz>yssR&88{v);RIH7KBNuBAJz2oJ68wIFTG?bq1wb~j&&YEl+2
z8N&9*yCI@*a-yAp$!IC=X|vqhGLyYvZu%1`bInb$hWXIP&W5pph~587XfKt^ZCN-V
zRaEUN?sb@v7#F)T?P<2T2pF*~_Fa3QV895R<Qg~2a3~dY9WxEX%+m1>d}V3P*hR{#
zLz~t9+IcDw7t;?m)8nU~U~Z<biX9_9nbFe?n~v>_N?oNR8>@Pqyy&j59_PMiY%W=+
z%U-{Ss>^RUkPkSBll{l*^$gpAo>nPtJ*Eq`Qwh+?XV1JMdxvqF&Ep7g(((l+ulFEe
zz+i{%m$GbMIh1Aww`nsF8;)zJqqZvINSWl<fRxtjEP3JfBCxO$>sx^2`z7re2(JiA
zGq;}*42zz2X;4|)!K+N2QaAN!nZnDjY`BcEpvK3*nXChEPx8a#en?3s(s8MGi{0>8
z^I)&9R}ecC6_?|*`sUXyIUfn3my<w?eRF#P6IMq*Yz^XdC|n6y($|gA2@E(78P1qF
zVyDZ@K-s(7zYBZ<aBI|4xUYvj{=GJ+(^0q(vPx7WODD`5blskiK4jBY#C2m)Lc!mH
zgM&RmYu*0KMB4JcrQ@$33t2H}<UvCTJBqifd%OgsYa_iG_q=?<NbvBd%NUG5_T2Kj
zbp^QW#!t_rkLo6H@q5AZpFmkLvB38d-oo3H<KyGHyOC}wVxBih<|fvufFC#BUq0QI
zMQ2NonF}G690lGg5%s1$c^kH|yRr`D98dLyNs+5z^i!sdns^a4d67fU<+j3NHS2+2
z^8dUaNF(RL?sItca$$hVP&!tL(0}98L;=@|-983aF%dPS2W=nb3ljRadu=SBAeoGF
z@Q`~8Xkl8D#i}6Y|I@XNY>mz7mBVftC^EnwlfL2i=Y3q4mH{Ed-ZetmSKlS&@epvk
zrVV<Rf9tk*SH{EBpZl$UcPKAX%Ht}hhXHxQ=}&|*QlN=f34gfH8_@;GW6B789}*zB
zJ^7E-Wwqdh=2{q;!7P<)rAfRB(VdfqnQ=}(T1{=;965+6G9C=(Pg*!CeV?wsf!Mi&
zC7**F55Zmc-n?X^X4Mrl1;q1pd|VIai}yk6{wB5o-(O5dd)t2uTJ$FT@d^T}q;Ec8
zA^oGemUNHMnjFR*a9;5w0zWi`=5}sp;DMyP4&4>aCk1Kb{mt8lL&8P5F}ts$CqDIL
z7V`LEl3F4t#bk3adH_z(**~0)@BYQn30=2qkrJvQa3mq$8A@_c-wjmoWYadBcjxIX
z5bz>n04YIA`-4oBKZpFdO<<!JF_@?1pGl!Kiw3chG|>y3@klk5$&Q3)6L-=Et$%Wi
zJPlUhB#=YqGDP9Dz^q7U$-F)PYH3uho~DNeH#x~@Ul`a1nj{iaE>KDGYcse<fVuvw
zQZu^MBU0;bZ9ozu7oekXLV)kqr937qQnUca!@~k4emeJTjt|jDrC^z*Cyo8|AC~3?
zJ}DXEdDF=0+*@`jVGszgw`~aG+Lx+;$BbFG@gQP=44r_^yOsBJp=HB^&|fMaPrxsa
z|0e8iX?kneSJqGAsZ|~GXYP}sv5!ZmNg8Vk9cp9$J{>=2ki@$HlMki!Fp_4KNyO*h
zGq<`+bnL(x`PC445CXwH5AXAe@?{KDlze;*9l9d3SaGxTU(E>Mz3+uOf5Y4qD=~VA
zxZxN|h&5NtxSTZ}pX?43ebGsHB*P37g_oKg+(5>E3lNG-i|%wBe*XFrLL0;gxqU|(
z2IJdv7IzsGHBo5|V|H&i`GqzHvEC>!LHV(2+>FQwQ@%q$&X0iDNWslt7|n=AeIEZ9
zL}ruqpYgPs9T()OZ_YkQ4z;WKksg+!VK~xrLy3P3Md}fyKMTj1!KB@U=AWNn->vbX
zq7uEI%ZAMM0T1KNYqy!U)sdj%99uvJlcu(px&<4o!jH`WT|g{*0EIgUIQDXo+K!Kq
z%Zt@~0NCN)Q#NJj{pogQLF-uOS2*a0uP$#sNcO~$k+AXg9c>N!L4>V3YA^)&DuDm&
zpQOr8G(tU)hHk>g54i)$)q<=O>rGS&_d0_-<=o--ckkDP+f?{gQGr6P8)Wn0FMYKB
z6ss0>$!fw~hE!KsvzMYHgQopDE&j3?R!J|5d)xhXN+5Qxu;VC0C_5t4)78G`SFFOr
z2)ykN!%hqv(FZNWj-F`Zavu|c{qi@XMp)Dar(bx4mI{TLF!Ko}oN{S*r*qIr1?ERc
zkIarD-XHL4nUnQ;Sy(kPsN}<tba!y2s?C0Qly`bOa`5WXLxYQy{x;HGXKuJyuz6(5
z%jm}Y2Sh&MeU01ehLak!)DOKS8(K^0L1wxwEKELV+A~~MsFW|@nvGJP?fJqp>fdx<
z`PqMezX|FWp%D;S$o~5KxqDEaX6d06+5OW*AnJS-6s5Qr-5QJZ&WgKfLi&9=oq^+R
z`%s?yBt5@CWPmw4*#;GyGXson%!e4iWQKqP#O}@Ji+rsrpRKl+jIR;9VDx>04n}Im
zi`iOyv!}4u7={zyjLlvdR$z4cDM0!Oa6>74@r%!`p#Tew;L;0>q4PcEYYUBr+$X8}
z!vX?u`!~`o@}cxYl`QP9^VG;;ed=M_&UQZi`VOP;(UW313jR!@%*cd1s40Q1MODlY
z<(NC04U>*k^x>>ngrrFKOAk}j4|$PVpkaH?1q8UOqpFArxCx#(X%Nm3zbDcUtLqSP
zox!T-&AQ-lIIe6vgKKDI#Gu7A-9SV{M4erwH=o01-q>unA3<nVrwy^~FlmERV=mVX
zcxvk5u*d{^@-qGO*3LeH{<j-%!RyMbsSbQ7bChm4c&b{V3WaU#T7#d0qgr3SaX9ML
zCBGZGEofIj@jNI$!-)S%f%6J!PqW0A-(c#v_xgRFIQan+L-e=<A^olL#^}L3NgILa
zY|a>8P#{KHxD`VDu@Fx`ffl@KK{Iz8B|Ln}s?|+Adgl^NJF5#sQo@K-u-PFy-+L%c
zDjBSl+QcV%UZTA26L6HYB2}{b>7CtTavW?P(YcaA=2lej2;CU4^khT7POP_$(XkVx
zHl7!=oTsn;o_B{lf!vJSI(&Bp+gpV`dN{%X=^=P%EnVuH-M=&gwf5Kv1@s`9;M>`q
z(^6~SM_31Oh~KR`(XEG&Xyaf?YoJbYB{&6ei9{IAY~oJlnX;7>2oyC)OG9HfbPV3*
zxE6ZfH)@NpK=%A=!U(%!9M@9N5naCpK)@aV&I@if7$JKF_kZ<UF|X{sKq<F%oPrz1
zj1sVRhUE(zC~d}@bdrzJUr|e(-xCXy+m^?FY&H0y-6?ONToW}Q5CWdjh+kV`CTBm1
z=qKjyxXQ4a53xG_SRBBhg@9}HqW}66eKs7lhkgO-=TQylMy}s(Wpb4GH8nx=8V#Ha
z93HxW3~2kb%Ibv&wdMWy7{5IL$$if9Je&faaE~KS60xix_Z3*!;S-E90^!AV@wQ0F
zoIv;Pz&*ZOC7R2w=6?VU1VlR1bm))eTvG&dOrz(?Yronz_+!Vx_xi#ht-Wr?es+z9
z!<~^SkBVpHOX-^EkMn{KM!)j;(%$iBqxAJQ{tos78k78XNHr3QaLB?#ZSKoo59^n1
z2a&>-jgQ%bw{1CSEL4sFHvwQMb4UV;YZQvf28xbh-TwikXVr!UR$_9`*mpl-wTO>L
z$HEN=#tik)eL{p*)jl}n_!c-nEh%JMrZN8Ewfy`#4i%jI^Fd_WhTQ%Vg^qR#6EyoR
zU^q}MoBRwLPFHt}-aMd6j=*(c@`f(UjvFxaMr8=TK6;!K>KdVDs3MRL1+SX=3O)H!
zN05L=JI|pET>o`h^k3W+W(J9b?BEg^+?RkOCK#-M-mj5Z_=Ck=g-*`EX0ZP@hjr}$
z1k+N((Gx2tNSHPNfDM0S0ET%KCY*YK@AI`ITluvUS;K<W=?M_Z|ECDq$-$PU$dU6I
zxVj#}4EnxI;&b!y<sA;r1UW#RSlRXZeP2quP0rtd7YvFEqp%f54GUGrNDeU;{U~MA
z%FIr(%b}o!9A5u48%amStIVIG+kfJXLpz?FG&=G(EJLSP+~Wlhith(UK0jd|qy7mH
zguz=Gl3$mhBr2Cj2*>N6lV!ONt`T~oFlQF!>AM6#4)Y!>NqTff`L7sUpliM=C$Nhx
z1(U{?!+0K@*m2l7`rC5|l)Rte6d*7_dolngYcMPQkH0Wj&o~Be1(%3l?@1<jzw^Fs
z3LvkdJ7)SeK+kYHkCLrX*H5h{0CyGK0gRofGT1*Geg>ik^LQ+ASjUMD(&10Uuj9@v
zJb%I@K@t=KN`1_9Q6Obns2gDHNP*gP9jS4Lmzf{7b4VOh@-lR#5P5_mixE?#d12mu
zF5<5v7I3`&Srd(FpD5ct^#d?dl!Jc3Hd_lgNxM??S*H^~j9fn9TZ2X?LpqlmM&%a<
z_`|ULsBC4S3%>lH&dCz+dM9n**cY{$7Jn-5i05Z>+*n-aX@8o=f=nE)iJrGjk1Hm~
z=}}HM7&MclJ?Eyuw)Ch5;(9ky{{R%E!L^Ds9{-$BwymMf1M13KPO&Q<7=9Sn#kNVZ
zEX(l$d7gj3Tod?p3lhU6ISy$dkUc2g`mTtTAbUBu)Jh^}Pg8%`e>ST>00soek)d9H
zb!Tvi?T6+Q#L+A`)rak@si~LQkGt+qp)w=%?m&YKMrk9(>rgkZ{W#NcJB=$*`xEg&
zE7BTr?pNS4;h}mv>BPw*WW7+JcN9JuugUYdriK_5yk`%<8gv$aF(KzG?KQ-Nv&sNF
zg9iyR6TdSfX&RkE@APgiv~|UhoddcNtOy~1$O!>Sy!(f|k3WM0N8HKWZQ4ciKE!~r
zgSuFyHtJwMOR`n5`FACn<2}pViQVbMqBC9TW@ilyM!66J#A*iNZ7SH|!&+UR!1V4P
z_$?J_MEhrzDo91>ad-h~g6}|P?r*0{6x8rH=FfYc?FEUk0><sCMKuVR1~6%WqmV-C
zFVqDdXwHWa|LI*k;NPfCI3UinZPS22m5_V7UZkElPSa1+lte|!*B;d5r(ciY?oV*H
zE+qUON*M$_M*Wby3G_@rwM0J5q^Sx2sPM0^(IA6=$WPbA3sA#y|1S%mku6k#p=knY
zUqU`c4RJ-x(HC+u`f{FdF1NfY?WH5qr&28_@S9qQ|9)As?EHb*L0>daSNgAKaANUj
z6<dhq=cdn8JQyw2ATFNUc&qg01*Sc+LiDx7J@zxl13X@wv_Dp?JdiipRUrf;K0b~r
z03KZKLDCeO(>W1>`M&L8&_hdqz?}d*yO_9lOW+LhpapH3*z04luP{-iTRvFr;pTl$
zKO==q5Ncd^6g|~QH5~lmc~AZD8*ZA0M02|A2f5Dr<0ziJx9|h*?}*76fFu2jQaIEQ
zj4TC)<6uNbb$RJNM2BMp#QO8qkpFVnngD1VoL3N`<&=4gBq5v3gw>GUsjXW&4Xv4*
zx6ix$dsV&7E*-#*QOU&XyO*x0wQ4vrq&v>~NV%OjF)MD-?jOX?tERfJ?2_m*DLT=c
zIr3w{9`nhd?QzJz)`SWe&y)7B9%Y$o_@$rMFZO?4U%nwTTk~Yt_SznBftw=YmxJVJ
z>8L~T&&V$IRK)ISj)pDtXkMPP)995B3DvR@f)$zmF#T$*&}Vw+Wxi3TnHpkMK}$|P
zzc($==Fc0}*BNuAjmeMQ;pt(5A{UYsi&Q_nh~-S<a@@M|cdC)M?US?1iQBKYZH5mE
zY9bHdERS@4-;jM|)KQBOx1i)VRO53K@}Pp!I`>Mk$f$OW9aOESAB}1gmp{f<SUfU1
zVCib#yEMvodRK*Eoi`zy4nx|ML55biN58}gloRd+!b>P1Jjj~3q02@WwS81)XSC|^
zMH1?`OTT)4H@E(Nbl`{uu-{gXnJFw@rG=^_BIKI|!p95wMp6@@!B@JgizvS4_l&dc
z6<@n89)BNY<q#~?nNB%-;M5+vru-eH7diCY$1~^p-nx{4!D}rQY!cvI8PD=-9dIPT
z<S`{Y955;A8?<>q^i9o?km38uMN!ID40_@(tY(B^W}+yG2M-_>hPJ0gY3-`EvM7pZ
zY>H3Qqn_uJGMwfM;iVha$D*c#?`R1B<I_Jq2DSZQ?<+jwm|<IZ^I}*b7nGZ|Q67g>
zhIV)BHr51fnR|rfx$JWe@Y!dtiP+}%&)}o1nl1i=b_RBJrF{Pe^AVzR)dv1~%=T=f
zC-XR7DyWZX%25!H|BKg(9lad;^*FAjagZM&-IIq=PR(}^+bwysbY$++(zY;*^)416
zI*4`5;9fQTtUd8QjSusZ0kzjZ-xsrS-=UVR0)X1_v||<HASt4Rb|h%_84sZ+@vve_
z9{~t$?q7~Mzc^kAzi0z{xyd{JI3V{`FL*7N-(;G8Q?cW&hB(0~5a%uB+e=0>^lPnP
zUzxEa9WfylkLG}Pp;c4x^yB9Y1Em2QjaIK*8}Zi-T0Ap`{z}D$ONuG$ypSnyl}i9z
zSEvVsJc4H#^Mn&qqyf~=QHP?d3zL5MGjAbl-Gz4_Ys+T-4V0KgCio+)2bRCaCUNKQ
zB2(Uxl9I&uso9%f*2k7dYgV3(p*ocC0CXp%?o<gN7Esj<+5neR@pbL^?zcXEoG5>5
zyIblq_FfdcV|$IEj4SqK54wn17BZCiFGhS)@(cflXpK-qm%}Xn`DQkU88VZP(`I0k
z*0xQ!D`Vv{e!C9+o08-`iQ2UfE@p4qd<8am%|{Q@piSUBL*k0cV$bR}d)?K26K>%y
z3)W|bKV$AuP{GUgw|YHKJZ{4duF~~xgZ1MA;UU1ssVZdTDh9sZYs+Gj6Y}xC6U@yK
zAdC20$i2t+{W3{Nj=wDs3lFsdIebot^@!5vC+^#|-ea7`_}NV6N(^54p3g8rQsCez
z0<^W4RtL{q3oyy6;LQpMoDGl`KGz0fs%hUMu;2{dD1l-c-c86jX$3tnjz<CZZk8ZX
z4E#qpdKRg6gcUf17HyI&0c$+82M5)BTm+X6n__D{pdN<<P*1;m1+8X?h+e@SW>ujN
zhM8u`>{fZ5jF1W-q(#2}Ak84i=6wT|!UpoK-8Y|rLsDMM7%xQiK7`Y&To26=Bwk+Z
zB0{JlnDIU;gS1tjwHTBD-*JG32{Y4RKtHe(^-E-a3`Q|Yfd&}?V@HZzyvPyay`GB0
z5&b<#NThm!gBJb*RG=<t0u+oork;b4m~RdzF=&$TjC7!$R4Odcp93Ljsuv+Z8rpH{
z4hyN!BE%;`Mpwa~3~1cMX<i;QBc1}Wh8RN;gx5GeQ^aZ>9l-~mkPiMBXkJ!snzTqI
zogfv$6h$L2pA|@V31Z}I!!YIyT0+{Y*e;53&ofAlQAj$FH616px}Sh84@|#;a!uMy
zVI&oV5dJrv#F-vUSG0b!)IT<QR)AGNEPQeCj<X!YzbzAJokdBUk&Qw;CSa1SCtN5D
zM((fjk4!m;Nz9$5Gm$B(i8lOFd^nCfKt8aP$*&MnfxF^Ym+YHk-zf<ZL9T`DJ2?vg
z<clYSFn)t(;#IQk4+3$bz%4dffOkPX5^cl`VxS2Odnkd98}-W~j?%yq{#d+%{U#u$
z+J=l-1Q8hdWuS&W;Gj>K2+q6x`9ByKw|mgOzmv#As3*PhXe4oz1veXeyT0J<^#%mI
z!0WG@37qcM<BWeY0=VIwK_YgnXhf5TQ-h6PHxRmS-lG0`6vJV`qMJT~Ki-E>0eTE{
zQe7PJwCzI+Ad`%kC94D8&0feL+xz?wtS)i$zWH$P`u)1cKNs}wgfzmUAkOST1>3I#
zZZK*rb5g@#$p^FG|00IXK|kXz?SrIC_`5&=zU*e8z?|KJwH1Pn>;gF~1CHT})!0b!
zMa&n*k24E9nI{K;m}mb3-CFG%8bzJl^xeyK1pEji+{bkR4hHdGRG=C&X_5VFpw~I4
z41gLlQxsrXC7s#-VT=8-?dvZIb;&xk2^!1?j0g`<8wPICy2gW)0S*1cHN?WAWanUK
zWj-d(e<sJ@zXHY`7#!ONw*4Kq%?{uTV2>ZL-f96l?FI!l^|c9@^dKS7DjpZfl<M0B
z_A2h9pZ{eSXe$k|_6S%a)Nce*mNnG8GU56BJQ&PN12B53J$kAkcKzP}up9CO3=G_P
z0{Z;EpW`v`H4Y7M{U0WH@GKW=9Ef4y#K<E**D6w?@F!C{4rDVs><1i%+pjmm_Ty`v
z-<2u~?CtOwDBCE#z#?#CRHRi<0@Ks9K2v6bd>?BnT!y{}r?~s6Caxzg1qx(qpNQ)q
z;M>=_AfG=Fu)piqUm<Ryad<CbjfYt#w4(1v)<2lumlp}coPGQJ_Tn?J`v=)QKdNDu
zeZH^Bi(?HQ9rYIT*No=j8_QJa@691+-!%J9FoI2;r$LH8zF-s}qd&Vugn=-`5yJNx
zu-8{8Lq3Bi2e<|6W6{=cCG91ayPZolot*Q2IlMQopCs~c12VcnLk8R0<u8Z#1o6-J
z4*biewTzY|Z}dQes1+q4DrKZ<9GLng9zNR=t*FRPo9Ykf{{f+#HG0_!QbCy{GKMA5
zu8<Ut&_XIEfA~1PJFkeTX8i*UVCNO8NYz@8mLgq-5?}FVEreJ=YQ*fL=#G<S0=Zc4
z#Jk`*6^z5>%5L>!XLyJY00B(Q6!%_XQ{<|h!Qm}KmZdDiR!>J1`)0)#!9H(}i(H;0
z35Hcdoa9WRBR#~5c4C*#&*VDusXgsSz!x{#6afCQIst}NOI*Y}akgzU?MM5dH~T5S
zmksz|vh@CEmn8HD{?8;->;IkY_@AkV|2r%6|Ic(saYqA>|C8a!%*4R(-_1L{+V-$d
ze!?BQyD=q%Boy+8C=fH|h&fg;$PdI2DMKj%HK@mk#Ca-IMXBe2b63E?Ef^RiK$t?5
z<Phl(#0f&`Ak-HzBoZPJAVP>00FvySGJSbN1=-%de8TB7d*S4}>3elLWi@zt`^w!F
z3`P(dkKW#Po5p4~o12-`YI+mHZfZoE%i{EK?rUAhPf*-MWM)6gt7-1i=m+sSmVe7S
ziIojV4M!5AnWS!llo1N7p|P77=Qcpk4=eRkwKQvV%HF>FzR5pRy{uO)%Z?VFZKOTS
z8hkK~P<B&is8zl8ZOWKX{HW<1S2_Q;!V)DZn<YEz#DMCAHY3M|{|}s$1}0~Y0of+$
zd;|zuJ+0}65!7<}adBU`NA;X#C~mgXrq%>;JcWI}zZ}*~S0AzJ{*Y``<tpXE5OF4D
zbJM-jf;<C^>nE~6p+=dWF-C=#ew4vUMRPSH)X^CGAd1bmBgukhRcw21W_a3(0rq4g
zKHqHl-HEm0{o!&TlB1_m>u()1?5V`+Y4j>{ip-&EUoro3vQTM|+{NXL>=swRdC`^#
zZiAc~!}d{nDf+V9h>Uv9>R8%<b-+1|fer6WRcR_St{SVvX61ray4;jSzQIaZ9n<!+
zk+yeJWz*3fs7)nVo-vnQ3mQ2>-tR2x_u(hL_0LSiN1tV?%GJ1&8G;)-I@0u@Rv89V
zR|S|WFs8+qa~8raih#@-Q_T!w?)jORwO^8mX1E8&V3*Thh^9F-48pE90wS7e3`J@m
zhDRzzt_;BJ<Vtl`HykrC+l}vt2D075Fgv9ZJa(*ogD|_X+=H{8bF8L;aG_oSTRuNV
zrDqE6FU)QYj7g7&x(3EqC;Fuo<tA$yH!aM_F3qCp8rsVY$}Su{bv=o-j7cs0d@$9r
z)amDw5vn<?-$2X;+RHoH#<T{d))|vY3?CViUh1kVGEI$42Fa1tb_@?E?mDb11A!5`
z>&vrL4Ldi+rPs+^b{5E%=@Xk}lMZBK>E~H>Ko?N;Ol+y#7fk{^2{qg|6@ugnc4~ue
zUWU`C@jI4Ts6*;A@aE+H534#ATN>J{fs4A7ZL8p<{f{(J$QMcqo@t{YEqMTvH&ohA
z`Y|;?#@nt=?rZNz(Y>+?)9)AvUn+}Y=U#(Q1n<=1`(DGx1dqECCUp?EGzj^Qq95jS
zKnx1&P#R_d^!E2|?xgyE2qegj>u4td|J0LBY%50N*W~Laz3qIUoCT&4M!B%;2_5|-
zGkc&u(H3eIA1$}yTN7dzVY12wwk+T#^B82nHaEl~OlUz|F$n+HKAdF451i4?m(ZF4
z;a5w9Ok(q}1;EhYLm1E!0{y2D92`k)&<Bq0&W9|M73(1gsS?#EU=BxL(jr%5)lx%p
zg6tmi*+fC|!p5Q2BtCHkRk6AcJb4K$%3&rfX{i^TIe0{y_B==vF^Yw$pQLA`K(|gv
z(o>aCZyW|7F*`KGaV97s8%n5$yk91@Td$Tx2Dbd~;9e~{0ZCodKpY-O{1z9OlP(B;
zhY6Td4+MUPF5KV-9MAm)%Gwk1*|?pD{;K2IcpVXe75E-Sg%!c7N4?_TzXG)*zl}Vg
z#&2j%U>!%0Cl*h+1ifBYI>q(Xtp^O#MD@eO1{$Me`(ln2mrm-Vjem8>UU37_ua?zb
zQOhPd&q}Yqu$!*8HYvZb+rr&bwS+gFScP;KHETiRdekVg->hyxKmT^zEa(L&{nlJj
zcNLgnU(r{0Rr<G4#Z`koX?{~Vy}gP>*k3H-kkm^l_3Ek(g2-$2?7|Cy$dmKxQWFZ6
zXWgmga0^!dqfX2o2wtbeql+H^UYGmqoC5*>XJ!zUp|W{k(FL8y{VAvHLel;AUv?R_
z?ebpgsv!|JkNffc)<%q`kgvxy%BVn)-qlvEn_!S0e24CbfN%=ns)YINg-hE7iuo<|
z_7xX8)~D8it1b+zZ_TKUS7r0WRVNtM=f<_`7tn8W<GKw6>+5sqx(x~Ii|g8T9R$_~
z<=~!6h3nGg3;jFf)O8{hqzCG5Q!W*@RlAvAF!^vnW`1i$e!Z?(Lcd*0T`*7=W4kKz
zI~23|MEtwv=G8kI%PT9kx<l1pk+zGX`Mp&IoV#Kv-tF3$U*cB0E+dC>;65y`YVzUZ
zyDcOdj!cCZ>)eY->Ais}>D=0n{edbW`Bl{0y_8knGv)mO>N+0>`N1G{i}o>zl<rvH
zW<&k_(Ru*&9H)!Umku~`g26H-yVaOqe$&~h^O!qOZEv=ai@dV*Z?d8hd0Pfdu`P}N
zdEuEMX4qf2>#SX~94?u====sH<H<a7+UJyUKk|B*Rn51@7}4_9X+X;r2JL+j>IO$W
zp1E`JL7<wHesXC!C|4TkF#Ou;bHSmLqJDC#TNi4luxa66z5dl5{I3qJOMk^w=9L@K
zvRwf&&5-Y#;~W@F6WNEG$Sh*i6}tYA`}Rc%MN~iyM)0a?*Z>``elf`);0`5J`=irv
z$g@s7uKN!z0_qYW@t8Wm#=UE&b=^3)OR21V;x_uA`U3&;H&~dH$hI)Yre)}Vkx9(P
zt&9}ZY*^zpUiRu#U<ui?A^Iii1|(&a1v=PEK$4!bqXjy55Y%Pa@zsyo6q59;&D7Vi
zNq;&-gD)^RByB1o)Ef1x0+uM#LJf5+*oupv@D;0<K&bafjx7miQR`rfTQo_iNsU&!
zla+=$558e3eaJIYl=4Y?eGhcc0osS3gFw{&!O%B6B^*R%$qu_}AW^#nrnh@aIF;ZM
zATJ6=R4+lz)f@h!T1IuV54T7&R+H-2>?$w=(~@=@X;wR9GOkymIWZ7c*;a_9jtSJc
z9Zl%=1<|dS(>Crb-P5Gj=#jLo@<(s4Gj3ACiKT~4|B`#FZ9k*1_ai1Xpm+3#qK(u^
zV)%ET)x=79!HV&21x9%VPVlZ~sCwzijyC*0+vZGqEE{aJU5XfY4{)yA8xm#3e0lvi
zxsv$$2tvb)E!V9DB;#h)(U91JlHKRb7@J4kGFX3+7NBN!pot}7_k?o?TPF0agJ!1I
zhyy}pB>LphrE~~Yuhv5F?ZWis3Rgg1u|Uz(B}d@$KEG9spQM%H39J+=%Bo%!QPvte
z8t*xh$4q@uVp?oEBio5FyOX<AlSGMcWaX8#A!^5nUsc6lp|?f17y}-v#9?XOD9SLX
z0VN+1QtIfFGN?L5gxt_8xL-B$QflCnG^mN3boq)ls_c$^!~}XF0p(Uf=ALr)&K%MQ
zZY^f!!Y7ze9yD+Yb=gD7M4Ez`@if_kQh@~H4eg^yBwFkF!%B5A3x=|=!+4TATul*a
zPYm<=uI-*fq_-+n)UffMgl1`5uUB{2M`&Zrx>VTivQV56ifyWJ;hE8_lEpl#-}6ae
zQRO)m-T&?aWKoOQnHS~|O-j3>b1EGdwGl6tm4C<>$S@|AS<_=nf5EYu)SUg~qx(J4
z`L{^&u;oq8ubTXt822%HR*un8CW1&exn5Gf+UuHV`*B>er#u&5z>~@{o30n;Qu$xP
ztYv!bTaZEU+%*YqYParY^=@gr+(qsFv4i+We5{v75qDCgU7uw`Gd`Pzej}dAhq-;W
zJ=@iD*oEHY<UQ3|0i9>E?DE?EbzzyUAH$Tl!N5#Y2$!vc%+hh2;M~=Qrb&I<Y90Q?
z&GWY|7_4WbIx!YKPwCvvdMfl)X1d<`Wy4ve&kVmgC3US~nwR2lg;36S;7Fef>EBA#
zpZksIH#xy2SoNq9`j6yH75K%gCc&k=XR+H0fpKglGwYUb;?8p^##UMpubtHgBlW+V
zh#T$RHiA}p@J{z7S+&f}vAH*L&HB9OU14pZJ~aVOtN5QYAb9Q{r5BYs%U*Uo{k1bB
zJ*&6tJ_g<udt59ha6frQ`<>dmz*oA=wx4&sLSW?s^L0b_>eMK*aNRua#Wja>wJP0#
zU=83C7$Y%jwfBL*AH5E;LEo|0vn_J79d~ijnh~eW;~bh@KK3FtHu~sH`0yT~8k?0l
zZK*JiR9uMiqm_~yrq|@6D%KxW?ZCra*L8nh5B65oI@QP%0j04X7ZBz8j0Z!WjI^a+
z@tLm|i9YtH%f{<Ro^u;c3kwZ`>>cK+eeZ(3qdfGTO}|yoi&;a~=iH(~^dPp`;`&;h
z#~2ZNXYW5-Zs0CfAWReS9Fw}9y*vjmQ}nK8XgM6aAE!(1nM=rHQI4*+TmKD74&h&9
zZUKlv5C$Xswf_N+5R&Zszi&(||52><zi&+ctB#I?mGi${N0*a2W4p<L67u2!MSC2g
z@vW@GCdj-f2(HW`#vVjx9Ol5OIxJAC+$_IK9g!>xe^IW>2^jmDGv(zWp>qkP^XU*W
zbo2E7Z@GpHR52WBAnHkYz<9taCRlCcz4x_Erw)S9q{(8YvJfck+Tgv%A%{I@w~nvo
zCy>>>AR$!+PI|A<5D9U#6!qJK?W>7ZkDSyjL`chki2_tuY58`5g2aH<u~7|`1ocoN
z1a!BSDv*(`4!-p%<%&$WEzI_!=HmI5UZr1(i-tYD07Oy|94s)dL*aO=fzIq^uJ5}~
z+P^?6-aKh85jjhW&>yXyt)3{de<S$cy^kF?zr%Cy-3cNa-5B95q#8P>Vp5<vfDgk@
z#uZVbKz2O3L=qWkgFvT^_o5uZ;tRD>TCHBc2~dJ(t-=wOoJb7FYJrWX(%-e#HU<}N
zW}vEy5LqvVf{4|b9b5EAWyaxk9#<yqo_+}G3>Ep)(A-XUKF0aS6+d4VzD|4jQoHU8
zCprq56pm+t3nEJ;bm6v!(c_DmofdF>xRarh$co@aW_2AiYPif~)DNF4xa}_>gaipo
zOhGPwjU+PAJAlrn-87V7gZMxFGKtAVK*y!)UYl2`mpFpyGms%o$Wxq#nWe~62gAM$
zvL7EsRFKjC6670G8nH}SbCB3(sX{Tb=MkW5+jg@6lfm9NMuNr`1$&!GQC204eAbmw
zN^%Y5zP4ZIP7tT9jl!)1j<eqQB-sOob`Kz?@z)5X0vx<=a;I0sRy5A$N-Im9co3hO
zI<Q12DPnK9(L?`PK#j;YHPk%q{<WED03B{mq^}R6Fo`1mXfH^Ur_{OCnXdN615unz
ze2{fwW9&onDGm)+XU_Xe!ym5H8wvY#Qek<OCb~}^f8sR%_tOSlgt8_Uz5T;-@M&cj
z!+P7c(jy=Ui_T`$4Ezlcu5c*;Pu>A`jsfCx=UV4fKk1`=3cW&y*masN^TrV&#{@j|
zHc{a<Z#O>%dsQs=M-*DC=p93k1%FA<*)#wI`o{XjWFjO#-aCL<YL<dUVW(`YB~z&(
zq-;7-TV4p%eUv$S1biU_5DjLleJtiG6xj$J*FXY@C~llF_|)wS6Z@Liom*Xn7%A^*
zg$%!2Q|}9)N8r8i|K6&M4F8Xs*Wc<&_W!Q$Aw_-4aZ?PT`-0k@1YYL`_=8KUHCde0
zRdslh^bQcn0LuhPA26Th^n#hd6<x<%LXcdd60OtCwX{qy$wnwi&(?O8{pK38C4yAV
zA*H&HOo)_8HMo>X(=O2u?V>~u{l@O6UAwluIBM=?=wawo{-R6%BJ4Qr>E`BI+2I>(
zB0L`mLXMtdsu6r!VJxE>WOcZdMsr|MgS8*T;M{Qpr7a5n>06;~#8m&?v--Zax{6I<
zz|U&ZlZuR_K+KEE0D+SlUaKLQ0&}g{$cqETBjd_6o>5rIgjaJ?feAb_o>KhzVorF0
zCPULY%2Ur-HUqx<Wz`*(Sn8az&IGY8K-h8b7$9|25OTg$DUYg?I#=C~9@8kw+U;?q
zXJ$ha35GZ|1_{K|*&;ldiOW+8qdTWV8Wa*(eWmrdbpu`X)VD4>cc_BnsrNdCK!_v7
z-s#+h5Dx`<KRq62c9T`QiQ`0fyBV2PGZ)=v^9NUXFoN}yl@ypljc7iR#pDH6Yfw)d
zHYU-9gMBfPf>(`Dv~DQC@EWCV@EhQhHu=-Vb||8m%$_=VGr-U`!r@1n^y=-xs3jl|
zbBACR#lZkVRWM$R-mrhz6Q#5)#|AtSiE6~<4}4?<tC$WC{y!)<obj=OazB5IZmMu-
zFjp?9Mt4Fm1)6dF!2X<87u5BoWls&Mc8F&RN-cQs$a7k;pPIJGG0)SET?%lGa8v_6
z(tnvckb^e=bQZbF-DX~7lfIm-ZkGJe(Sd1rW`IyFLBX)%N5RtuaDJ%R#-hj@>iMI)
zzg@#3c6Ye(S{^@fY~J)3&ue8NjaE{9pM7K(u+>DO6d`x+;83Iy=}UOv)becZtFI<4
zqYxd1o3Wsy>fFF1As5qv_ntgfP*q*e-GL&<gQ}>y18^x!A7-Bw*vpEh$B|MYO8)FS
z=BG=C<a$tLy5eEO__VW)EPN6SfTF~a3%*IdG6xlS<+4I0o;7Hq?o?+N^RyD6itV>m
zQnw0=Y4PTerhH`tH$>HO0iXuQ8dcC)p8=vo-D&QBZrN$gLzUV^_e1H;!=Wg{Cm9mB
z3BRyuE(@X1bvbrsGX7OTV2v#sEE@?hRLX<8b`|Y&@=Lyoet)iGGD7Q`HY3xN3GsN-
za8ta#(6z@J+!6|<-X+jD+6agU`D?{e!YxB1)w3Pll5H_cb@{pJw3)@+nJj@7SxxV~
z50jt1K>rkH0v5Gc7Wu&0DMdqDV05&K7WWBEL58>Lv20>;+%)B};@e<*TMmhjV${go
zF3B}NEL;B58T5(3!}PdxSUcM{*CYoUf<tQ0W0A*z3_492&NB?{PW65hli@W{t}Jnf
z_4mwaga2lQ{RcR^{1NT{fhp|&wLtuTo&Ssstp9CH$w~Xg6gHI57hfpak`P24NW*g$
zSF#+E*`;h-F5P~VYDb|2tVBm9pLZN60D1@tO3jGI*gHQCw}WNd5cM0o9aDBYJPvKj
zVx-LK#xzuEwLo$Rr-sSaD!-MHi<8~UId*uEst8n6poQcCO!6XY;<xwftFez6T2%M1
zSAG5;_s{HYzuDY(WZU{Am`7VmVhKecc%%}l25_?Q`59WtaxH#bnLlTLbK6&SmA%{M
z=*OsTB0DW<39_jY64M{ckanmM=F-T5W<=Jp2Y{u938u4ey?m@Uh1Auqq?-uOJ7;A#
zJExZGIs@{YjQaSq2MUV)(12s3_uY#C!nm{JL`|HHTAVwW=VIW#^!=(@p9^ZH)E7k@
z7=$Thd1DV0am-NmAn;Q=&uhbh(e@PRMHpg62e3eeB9Pyw9@`ky%@rdVif;%`YD>m>
z&VU$Gp0B}LmNulC&;8;Nmc9yD=mz*?6R+;{v3kmBPzs}CCy6{|$rPx}R+Gc~??7xh
zk@?niHdlM}W#SlvE85G=YEA-|AbrcZnwZH8Hq~|UVx}^^p_pjfT3>%62^4ms5)o;&
zJa#N&M-6+^i3ECFey(KIkDk4Jprm@@5BIi=6JhT%oz$EV@PH@Spz5%7Tf#bN<pMHD
z#I}bAlS}8<U;PCnSY&1`V+jl)N?E*=U3Md@CJ93-3oTXRR&$#&tH|ZPbhl@i9KEKy
z@;0Kc=HhWTNjK9vGsw~X`ym7+^}>RQ#M)K)K(V!~kGOD@(&BQ0MZ8jrCGHq%jHgq`
z#TJ?AI^q<|R~kkoyn{KaIt-z(G#Z|EvBTdsj^go(yqtID5N;Bh5Q8u5IZGKLO6%J)
ziObhY@@$MB*;Z}ba(W4nznmre+MwRCT(3TK3*20ePhSk(IHwBog^)NM%PGQ|pg6+w
z)J+o5m+2&SP8(<Y;QZjo&gCt*mByTsr2>ry>HIyb0G4k42V?IT9c#C3jmEZ(if!Ar
zS8UrWTuD}JuGn_6V%xTD+sV!T&UfxTt?jqnv#ZS?PkVmMs##S}pJVhsMxQ(p8w}Ou
zI_^cYGAcrk2w25PQBAn!=q2@ke!gj!rvN0cYtNK0Vv-wR7(L+f+uL9I85#48Z2S_3
zm@3r#i6^fJx>E5;g~H`RXP142xy0d6o3bgJ7+*j3PjVo$&X!QK08)dgvU?Ap@ToT~
zG_xzdMqXxXukvPXBOEC-)BE?6qD>^he^gWUGZ%ES5N_27!wgTeqId(p(p~I=_cGUh
zOgy~9FhAV&LZGx!qCJw{VTaM@ppetc9|EiMcV~xRZFU?PuNo_4j9RFUX-y>S1_;Tv
zm^Sghs(WD4YKX+N2R7yCE-(<x6a*J~c_WkJF<fvmFFyeV%9i<n&8%IRbm1`GYr$m*
zE|kDxZBlpVC`CR#ud{J~nWDr49A}6(sfi4g;y-G?+mXf@KE@cY?c|c641zTrzdW-y
z-hqzdA;<n5asFq%@m~tfw;zRyi!BU;JP{Kk<G(y2oSeT+CRo2O|J&Dtjq^XoqYiDk
z7`!Irj;ZQxrkh_AWq=4$1FRZSN#t0fV0B0eSp5A5BE7X0<VTX4FCV=)KAP%w0=pnH
zAg+$N33`5h^*FU^Y^R<uRHu<RQ4PGL-7;gd;%K&JNqK->gbHu*aow(lQ{f_hX+F_8
zB@dv@_9B3@%z|>jQE*H)8d^>`SdBGocR9`E9_0qA5$lU{CVHk2?Okv#C<Tk0bq6x-
z5atiC-!9~~^z{C8gF`nWFUkl|xF0shjiks*828BW6hV^CwE#UtBv=!`P?2f($2Dl*
zUt|?UURZ;IR*Sv~k7OqQ$Y68`XD2Gry>7D6l^uN)te^`tV^=H{X>eAE6lQV#ZfmJi
zm^M`!p8LWerQEQ(LHxlo<PUV$w8+nM)dueIU&#ifJe7Zb;ov>1XaN5X!<RKpM|X%P
ziPQ)bHPt{RDm*R&LD8$iR_t~XO9>jOV4ScJT{H(v0DA;Nx<mGtpCLdEzqC~b)hjFG
z0g1&4kgMO0F^Vx@4$uuh=GkE{!p+=)4`0+4#eA+-x5k^0DR~J>Ju;bya8(_R&if_V
zElITuR`LS(PG-`VT2W7ZC{PnHRAX?3nx1Z0AQ~`|7u3jTZ?DNi#yr)hpJOfo-O*Eh
zLAB~sQ`+d;%%}ACU@&}_Ujny{saMKSW(Lw6mk$rYnc@-R(z(G9mjHNiBSpXUV4SWR
z!5-1hb7vCV@nMjTfx&*OrMk0sZ{zlH3DK*``|eZ!ZR7Uc!L_ZcwZm~`^&*+$Exm8!
zVc-)&Y02*0_Uq%|%{)YZ-Lo6MFR`w3ZRf$dJ!I$Jthx%(oe-(*G2eCWr<Of~`Pkao
zlLiOjd1i$E_jER%wO>;Q1D_M+=60SJcGXvpG|e%wT^~aNbp9wMt*vh_NA7Jy+C(7D
z9<u#?>I_Vb<|gfthpHFgygxcW-rNN)8D8_3zQ{4Jvb~*snmO0DTAdy4b_=yPFa25?
z*tbtw+M3=dJ+C_zmLqoj@~+Ox%R$dG)^5!_O8WBV8CDru^rgGz-FEm{i+p!{Pt=>W
zckF(*p<n)=3$a%86ZHb^IJ^IGk)*I}nZNQ~Dz~-KxaHT;{%YC8%cJAh)n&&U<UXdK
zyfk*_Y$XuMDLc|aN{BSWRNEw02$s!MsWd`DKXYpR6z2YQvK#V5X!o?F@9|kbu~S4z
z=`~Om)cgpYOQ6NK_%31CU2D&0iy&lm+aTuk*l5)s%X_3EY6A^GXV5e#h#h)08B0*m
zJTHJ?sEEmH?ukeVp;2IQDDhasV0{Uo({Z{@lVJ4l!V2?d<CeaKZVUIvhy#XO*<{;B
zS@z}v+d_da)k~6Z9R#D7<YiZ-W$e)?pzQ@3a*2n_+QBi%U8PU2Z`w!3%}wF$t>q)s
znU`TCoZw$RC-S|EMO-0}Ah$oEu0y8Pl{7}~H5`^$Cy)r|{{aEUJL6=frX#^Qd)U+4
zJh&_kLl{Tp7|Bn;g-wTQ_B&96v^#^k9jqPzY&mvPC_*j@3;EEqT=JY*aSO~d3W~QT
z5iucdxl;dAe9*{RD;GKX;t0I7rAdbX9M27{9ByGQHs&H9oK0xWK&rgy$5<XY{EOt;
zo}dfbo(K%Pj@6w?u=nvJClX<0>!_w$oPFM)2!Odau4zc9l+2&@52EFWCeEx<V~An8
z&KzsaLcXu~K%l5eP`+I>=!{`f{lyEVx{|RML7kAHGZx0*1u&w0A@q{%Un0yFZv$wx
zN~lc@_y_GOLF0d-E|!#Gwbp*r1od#>Kk8!F-xyGdso`5{dI(rBSG3cNp8Y*VL`P4m
z@)M{SCai2zSXx8#rO^1pd9Q`ubPvP%WW@13+?QKOCOeiVzQ@P!SqVjMlERenLT#!I
z=@WJ7Cgs+o*~&z_LN}TQ?a~eD3w5g|<+h~N%K87g#e`j{Z`pDCGl}l(E`JJjon}xS
zL@bgsmsuRZ<kWQGKE9kR3?#ISK0gvRe8dP=_mPa<6Esr0oaL<u%qSMX)tkF!jZqS}
z-~d)3G9orVtUNoy>)u<@zgz$3PqYt9nm@jGbn!|k>g_t^3?%9eYMpqHL*%g68Prfv
zLuy&fAY?DtnYt(ra)^sf_Tiv^-q119K71n%s<(^e(Bn7bgp7@^9!ZAUfMwkWHT1}2
z4Wi7-PwDYy0w8<i+!vuZ{z2)5Lnvuq%Oz>1AME`-2TX5ZWK1aY0>91<Km5xw|5Q@4
z8K}e@c8`Q;9Ls*bU3HDplZr4lTA<G!i_&j}@8V)Fs-^P?pGH;;LGj_NgTJS#W)=`+
z9tBDc2rA3NoxUKGP(xt9<r1oG*t{Mdb)#5jzlaG@|7M70UC0ofBFKE9TU+e_M?!<7
zmuTdRlur%6VQW_-oc!4?42@bnCYmP>hD|z}=4+htlNFZ!&S!sHO%0t|=Gq51{IXG$
zGZWo4bRr=RW;;r-IS$HU1Ex~61JP$50Gh<{2r>?s4gqN4AW9Awn4S@j0R_v>G@M5U
z^%%T?G4La<{2{t?e01~nN7srPs8XF~>WwQJ{^15;x}g;scMg9tO;wk?Oa*z)=mr7`
zyAm6Dk&tb)i*yLKA>p808-mcX%>Pj9v7qMBV9fYvYae9f^$XEaDH8UXN}AJrOFKkI
z7U>ya^&$?@UX}OviB(Vd1XXNoQYka>FTa)AHSb;L7L{t3S}ngu<!Rn5Xan_~c@`q{
z*+^_NHSdjw!M=wdA9G!}t=oL5{*nrbqgv!_8CIbhlXQ89+HD;o;NvH6Slz1J&{6ru
z-&MJCUEPaQD-Dgd*44R%EP|NEibt1*;&MLSivvcbo;q(K#`t1NO>U7Uxt?%-QtD+?
zG8L7YU`CTH-=rG0x=f~i0=7fr0JlWCxT=paHujUyJkDU^Sm^L9-3viNAT3ou%$?`>
zFqo7=v=NOm3z(crEkv4LR*|6=Ojyzp6^?oq@d-X#1q@b!!fefLWFUhCJk3gz;LrHr
zB#MH+PdFG(l01R$V9+#EvG>s^$(r+n6jGX=!39myn#v76Xx`yLNfe2>0U2bCs`@pd
zSXCybCtrv|A1@W*2N^%+or=gSYo~#Rz$}>zm@Jrq-+SfDiEUJ{cPLR+D^APtkB%Ns
z@S2C0hzVVLUn=Gg14()pO>q0^5MTbp8nb|C2!;A1jqPLoDi;wZ_ZBx74b5aTgdl<F
z=Rb?VP9<Uli1zD%9Y}}?qtOL~aFRnsLLuAI2VmEcc<Z!BL#Glk8Ck=ML!l7WvSrEk
zV}q=GB4tB^t$rf%j_^}(mtAX7QK05R7dt2-tmV2O?6RD4^Ejow{JqsSv+ydJWqTlr
zI;1G6_8um|L;cIQc8Wuo46KzJ_U-^?{I)kuI6&Kbzf4T`(dY7UjJM{;EUdnYpbEfc
zVG9#}?L(Y=I(oRmeE0(Kp^KjWhe(u#^WV8|CT3Q)|CCRw)ReQ|V@K*Z(kMy!4PSv3
z51_jFGa=C?)<(rH&alxNYdV6wn5-k^#@_w;j6kMbhczgT$ww78<Q>zO`V>64IoP*3
zg-M8knc@y6qX3{PxyDGvgo<JL#kjX=bn(HaJubrD_IkZ`v0jWfb$Qi#^zczUa8W$a
z_kJ}{o#IZ*sr!osEeH%~6+D`HAH-U83K1lLQSg;jG!9J{sV8*82T+oso6_dvNx6tW
z$t`f`s)w=RSL=zt<l0H)W;t^YCQK?Kiv}Ey4fCsuToB8#=3XvQ0Sl$QxF?f!*0!uW
z>z4aM?;K%`e9Qih)b5j}O;Xq`pURDU)QDnp@G_*1oYLr)j6rI!oiiMa7kWC}g;K~_
z|JLDArqffloGZ-i)2AW0>&JYXNL&f&0)nETP(X&ivIa9$7}9}J23!CXX`l>m+p68H
z8v0S#<e)%Ef=KDu1yg0lGDyAyN-Jz#Y$<>6n_K-^duX|N!i#LjXe+B=(VfRx-UGO^
zx1TdCBJ)6kaP;*_kAr#7F=d2hPU<G7$BQ{?G1jI|hru(9kT7HbUYEF{m=>7Y+MVL$
zuv|zyrd(_3ZjlQyvMe&OiJ((<YII$!tht~3QV8k8SzndNxQkEgbpVt)3TPI~1Nmq#
zOUJ8M`%_y5#RMy$u*QLr=)0$mG-n$4pcTF_y08`gE2Hny+%ajWyb>cw?5!+3f}5s%
zV}vT(Rz3_F9s>U$u1y3iyw{-Jt1O)b#N=2JROjp?<p<6w>GwByXhJl=n$f$cgu(6u
z>jsi;m-8H<TE2%oS?rP!2SOz&BF;&+CaPr;KnMUeS3QtAod>17nlnZor<zYBp%*C^
z1<?g?oODX}w85>fFaz}!?Kq$bx27X>3eobwuQDsjT`M&-I0n`7@f}PUHvx!$VfMr;
ze%aE9N)C6fjQOmN_7(;LVib|$sDhP39VBgo#mAzq3&xx>>C3Aly9jBNpM=_@-5c%_
z7PwVv2uSt*3d5NzXoi3)Z}}V;GuT(DN}xG1rCOK<$92Gy+&dPZ$W$&ewp-TiAtsBx
z&fkOfE<;Xg#sw++8!^L)oKklPGUfN<p!}ilm^!KA?JT_%gu%wG$ylW^l&$+nCEhJ7
zCK43Br!RzTK00Vz=fuEJq^Z~{Nz2her&j8I8`L}yYXQ~PB{xko^^6Z3>#S>KSO}#Z
zWQ)%n0f)Te#ZFjnff5R}#Wf0*KQc-HndxADU%b$#pAeL>$c|!p1n)!}DP<u+yTpuN
z77et_(`qmKZc++neA!y=>F{4;?s+{MxoGmy`g^Cskj2PenOi9LMtY#+X&+Z$+~hvx
zh9}27D}<42n8Ac<wl0sI9*A?@(DO~}1~CVfhexwW|7f$uStrYDJni&4G8hC!LkK+a
zT5^ILIsP5w_x0I0FpsYUs9AH&&o~<#9(+AvYr~4C&mh4-E8S&zRHza=Z>5bcgPJdm
z&U}w4(-|oDPUo1KLy_oVh0X;{nzq%6#TnD`ysAJ9ajW|ZVEQmK3pp+4`a0^U>Sa#P
z;u8q~kQXn1KLZGCDYXfJ`QZ(+kw1QZxV8i;EFNkTikaoho<rT+^BJ-ZEJ_$0dr^ot
z2vz3;MWEE9<i4&jA@4?Zxeo$He~vD8clG@(RJr~EExEY<9WA+-ng3&@uu|i{(ejuQ
z6g{Qj+0&*@s%-$E_1BKU7O`#s3(SOSdMK~0tJ63k-ZUUO<1fWbuf=A1DhCUzyc@K<
zZ)=P0mbZt2XI>nJD(Tq~DJHZzh~Z#@YwCNdZ$o+4QkS;}k>*qd<xyxEo;P09BF)Gw
zp}oLXHS?(B>-~0?yF$#_+t7nOBqGPM2!sdr1oUiTw-BK4x3eUtQF$a(*&L+y+9qd6
zqc8UBBr*sCie=bHPsCy<6Z0&wvK%`LseK}HFj%%j^H7Uza4YSqJ%Z)heBp(y%FwhV
zJ<`Do$hrmkHgwEWA&bb$Uol;FuMjQ7w<yaEPjjm|44*min+8$gX)<qn^lh0KKaTT|
zvf~Sl?6p}pA8U?2VvDCk&yprgl^pE606Jv{bzpFCwC2e<0rjBRc^hKcw+>DaiC9LV
z$jR8hEVKJ?Va%jSl681lsyu57*+5Xn8NPov8rky0`6mGv+vw<`MWP3gT|fy+ASNBK
zfNAPEoua13tbWH!BH>iw1Ga2R8Ui_lBFUp){9u550y%9C5w#UbPowyGDCbtNlS&ZD
zR=AfyFEuuvwYEK(84R!C6Pm*ONI<xPPo8MxyajsfsEG0k+<JqAX;=CzYJCHh_tkm&
ztm@@6=HWAo_u_+_7cIkYi22hT%tX|~ZeO-GH!)SKR)k%D1xZp2A*9GbDimyV;vS*}
zK|v?9X~WJ)7kyNdf@Wm^xq|0B=MSApr29Eff&(dqM!UzxlB&c65Pfgx8Ex=T7v!5v
zD*&VHei&+4`G^*H#;yO{7emqehS4l5MrizW+31i^k>K(?hEJUBrm70U=8CM?4f0N9
zz<ZST?U6mhA=cI;t70_<vm<(A(4->E`7tS0CCIs>@g{o5N>ACc9xkIw=!sUADcZa?
zc+M@#G0*QgwyBTK=c7Ek42!C11jyE1mxUQ_Y62{9#$N${5@Sv{eWC^Tzyhp?Cv{R(
zr>lmG=?gelf9I<S3VMMkNZij-SxRp%|5%#wAJeIlXSK)(u7{BTKoIP66WEtJ?ydL8
z<+e84Y-*Wn#x5<T#xCu8aeZ2`=Z9mmz;jTKD~NiyJW&0<HayR|;6HHRx_USiun~xg
z<K1Yaif*d)C$0{>C@#8qvT&<!yIeag>J4RV3A(BT@qSf*my#6&U#oE=^kh2tksU#`
zo`zWf)-Y4-BT~!#Cc@5twZ!oS0#BQ){_l@E3;X{h>Hc3);QQ|XdHwG;u7&>+8rjh{
z-e{T=&7{HXk;YbQSuUI9%_``*<uDWbNB<}njD@W2pLv-Af&A`fTBx!Z3E@0irwFRo
z4X#JTOev3W=KwpCC7~-FSNAiKH4BglY8h5vJ8v4WDt+D^8f!Z<l?T`JZm*pxk9jIT
zdt2zceOgesvg1t8>V_pn0}V(4&5Gjkr?1%?cM=EB3o_wSE190{mHf5LOc8%N;L|_@
zd8eaQ)L^;VL1q1BY5khZa^}crYuxn`30}39`JuQdvcUFeOElphX3muxifrbYHCnZ<
z;m@o;)<=BJ&6TDoNh6Cu9u8tj-VQ14&}$(NB>|_%#aqbvx*Zp@u03q&tJ+$gRz_oe
z*RZ<EU^S;|YW}s^v{r`jh(m~EE_K?Qx|@8@D!vhq7GY&h==0So6~F`thfE<oLBWC*
zGfwGkd7n?~d>Z*~`U!~rWswU)EQ7+WDoVKpc<%w94dIk`6-Fw=VUh|Y%0k&!iX(c6
zM~wUy4}m5ni?DkmO)?w{s3z3CfVf-u?^q{!GflF7kI$uyE40tD+@9_;5+kXfsv?pB
zPLyB^hB+E^nsfBkMJ)4I$1f&>=qr&A5~`F(?s-%TL<3QB4|bOd2P10T-Vx0NnjQ^k
z4Yk0WY=vUmsagO~dSg_;b{E5JKMsGC3``lrPmWC{2Il+OzshOI@#CZ4OMRhPBEQeM
zbV)@8nH7Js0q9W~0F%(I29ib(7^Na&1uh7d$2=_KSG>e=0Z@ddGbU=>s-$D~;ZH>@
z=|inJw;wG^GkECW(D$@z@Mx#e1f#$>k}cp)US6>0SbYdKxZ8MZwF-p-#hNJ46tFzx
zI&$dINNs7ag<4I<^$*(HZ|3B^p=8uhu>7-hYK;BhT+k$x*JQ3+Zf`A)fh)8v-Vsx|
z0REr7Dd`t4<qogaD|fLNcwr;#<|DS3N^utY)=OW!4RYWsRA;YaO6f0=Ob@N;FOCIp
zFzF!+WAW+J`2eZUfSYUcho_ZsupiRfu|njtcinjav#xu>U`N>6;gD1@&;-%nX2#6A
ztQhD@uMK``(dk1B^%05FjHfH*Vm}47_?<{qeVhO8tyZT}lM2l5>b|I^q@NZZn;m3e
zgw{yyyS~;LDk)*T13Ck#))38DSdbiuQ+2LAUUAM8k@MXm2?sw~%gpaA@lYmCti2&s
zd7~JS;uGpNeHp;*#c&B1EkfB96g=ECN5oh&!f+?ES2u&&CR^%LyJMhwC}t$x)6g8C
z1#}&W{DZu+=hk(IM>h*xN<wF2j>;txp8TQ;@hez74He}UkrdpUMa0^DdpFsq)RuU9
zC;X6dSp)Ro0JFANigBTPvKxuteqM)uFx-hm8at;05hi^+s-%+MZSH21?x7`cW6@^`
z*;}8U9j(up5d~O|{|-CMjQ>-M{QuDgxR|~zsQ+6VU}s_aPhn}||7rtQ-`YTngQnU$
zB}f49T0lGyG|@6QDYA>=z8Ii?8C{*G`{l_qW<qN$oNt|ULmH^E)x1^9TbFim=nSWn
zKsk0bP!oDqS@B(phM2;4F}5D80V5oSU&cXaZeJ*#Ql3%K36UDBP_J6mL3^+xd`vlf
zub&fGTnb?~Sm`?~by7W$)Uy&|yNh>b_eRm7$I69~Z5U2cws77Ied87=1((SH!II<r
zeQ03Tt{Py<n_&g7o&YZ0luRtE&3J@%>LEQ6yDpO4dn8L-u9CCULo7l}lM)X7i)RU;
zJy>RviZ*1Ort2iP0HIx7B$uG!TLt)#zR5+if=i>7dj`V@H<z7pqTKK-4Wi!Lu`2li
z1h-v=EPkg5o29Ic*r4wJgl2ViF&ZDV3M`~6ETf6#3NMFx-Td3k1LOhk#~_qUvudgF
z_|lJ^5}Y*mXsaGn)}A4^Qt)^>PKlDIZEVD1K5z!X;><IK`JV`v^O{TB8=f)0=@ly<
z@miIQHDl~L9K(beiefWB&<4GJ9K+vfqGteA3)k0BI(viDNU6dsDJch=YW?mhh=0Ma
zpjP9hPR2GzU}4$|)jLg^33i0((S$*+D(-O;gR93Wh)C?OWnByZf`=fUy|#hCMLF^2
z)C{*6<?X|(k;@tV13C!w0QB49yE?sB%TaFtPt#~MBgUt@{ErCFcrVN$qhL4oA>(R*
zVY`8o9@{0KWVat@IlYELmQJSL``(5tr;n$zvzMRi>tmn(<<QYqSI5iS!b$mB+f~(-
z9=XlN{ptPM+?Bk3SK`gi>HWk*`rCD#$KO%1Xl##$`Q>I9szYUmxs;ue1cz7l3}5HB
zUG|KuVEvAvnx-ZWCFMl@5Qfg?-mVP;)rhFJ`k&`D`|5ecWN;h!;+IM#@F|*xBWigS
z4Tgk$9yoe1><%znL34y(_eYG4?uZ<OonM~|FInbI98V(KEI$-h?R?4a6Q9@kU4JGy
zMhg`mxi{H&c0f!z{QSJv;_PVt+~IG3J;@jF`&c?FpIeIc@p1kls0#Ld!{y(8TSIiW
zui8SqV(EPCdvLec_j7#{!J-}!pHjK`Go`;}NE_+hW=na;M_#Bwf%fy^qPb@tTF+E3
zL}DwNJ;sw@jCaw~=8p4VB(Yc}OWu?F;3P3+YD=Mo``|P&ed<Z!h5O(v(O$(j>(*wp
zo@;V)%l=mX!b&GpEly8Z=2sH*yI<7u49(prq^4(xCkuj`jjt#!vdt%1Y@GlS1CtO5
zIZg+QX5*hloYFU7CLqCr(QAPu1iVkkRMBAv!9#(ryRG+&hqkw~Hp(}zo(w*AGe}J&
zq5cuJfn-bsNd$+X6(`wo-q!x{-y>`Q-ztB+?O1SfE8N%s-E29mYIQ*yKO(9jh$zV!
zO=Zv;n2BMS;m`D?)?V>ga<BeQaC+0>A3Mu_S2&H5MZ#%45nQNdlqS$7Umezxy3ESL
zU`ZvE*s(C;OFuoYF6~Q)5sXDP$t82S$D#wGP9Ey`@BKR>&XEJ+bj>+cz?4Z_YnM@m
zfpLD#!$nax%KjEs&8tQ?YQu#(#n=}7l+&4(5I&^zUEzOpC!#zvD$RHM-4ZQKle|b@
z-iJ_wI8Rbba#N>gY~HeJQ6odt`qfHQqKel-pM-v-nLQP{SAo=@82!=`XiYVaU|j79
zz7Hhw`^6ojl2XKpr1qEX3Z~ZZC$hK!7B<^W;I9Jl4xL4b%{k>_elTsY<dI%3gt`(E
znz38#Um0x5Ekv4sASO70JdJ#sLF&3nVT^Ru2v}t2f#Ez?evrI*u?U?Q$h$$Z!KI>!
z@@QM{0~xFBo8F(!X@bvGRkA5Ra;Ue0&5tP4#`luev^$kc00|XkKuDxPrVvXaX`iHI
z4u75yyTOb^ep|qm3X=ML)G?1Dx1%Uu3upa#z*odJx<VNQHkFN%{|U+E4E;R0)cTyL
z2I%{gR6i+}ogKoPi>W}lL{Z*6>DvAC^9wV&wCO9-7=X;*oWkDnXCyBUd@K6sEaA*D
zVeh}j%|)D~pMCb!5Y<w#&eUD!-IGeE!VCXF+tl0S-XX7#la%h^Ls-3nr_bir)K-e8
zK3A-^(doALNDdU7sj2ahl1zzr&Ca{yyfRnpkJ<A$27dxEx+K1_IhNSqg6L`ir)W^j
z@Vo}$)lW{3eQkJAt4>rQc$B+YVQ;@Oxt(wy8PmZxHkSVKY>nZzCHE&tGy-vwgK}|E
za9sPx*!D!Rtd1_zSf<<*6bksZgEct}8RZxQI7MJ`41QhXoioa1i!vD^NJa>7e(2Ae
zjB#vRm&&*g8YU6wKiv8QjtR>^<WirIg0W)J8ckG)t1~s)HOBPmr=OF3D6>9bt1M?0
zyS1)u6)|pLp1`Q+^*xYF+<Gh|W4a(*{;U=?a0MucV$2*jKS&^BA<&DsT3Ogt$QUg@
zSU^0tli;}cwIW$TqSQT5bwNfTuen0C&+%u_Iv#|AUcdhbM?_^PFIFnt%%778$Rfkq
zAS(j17G<@uI0CpdjrLmB1sD3HPk88Y;))p<%DzBF<&tv<!@)HoM)<a3P_6~*#|kFU
z{fIfqPUn#wz5PD<f4(RE0J4H!d(2;^S?+9H3@!MG0?Vq1*>1yzjw@8V-oP9IZ80Im
zTj}b)o}$Mi4YwwRRLl)^X8(CugX?`yLmaU<=GpzZ7#_)1%QvM=y8?21ocK`3{q9c-
zM3z+BW1Ute!r=pL{KcKuOONe&CN2b`X7y>`LsUKY@s*=JT;@BvuPGN}<kU=*t;pq*
zuee;o5nVqKEcbKGJX0cl=+5Wp%=x{beViQJc<pZ0q|&Albj9Yac4binV8SEs={Y*4
zqL+1ho2XIL&U}zWRW0jjI!dFWmwG#zsPQf+ePnYGV}*GO$~+#AMM5~9(L8$;bkluI
zWc_i|K$c2=MuX+)_nEiD$qZFQZ`&<h=_^izn86%|ItxkqY>OnPl_H3cv+5LPn<$De
zh?fP$WK|glTEdRnuh(d#inoKJG%)u4GdkGXAmoWN*U560R-6lJg4~F-CNtZDp+k9@
zFwa8hTNhkS8Fx!|97{Tn1lL7KNw;NmooDa=5aaGu_d6!5ZNnRIiW`t=^#wZ!&RB%y
z*@~mDdjeW%Tq27FrQ`_Qh$0k79M+%?jVk%q$1!Lf-cQ_7|J{#3%l0MA&StSg@*2b%
zy_Jw7`?L-*bT7kjI>}!4hKUbF?NGcI!Z=<<D0XbYU7!S|_4#-fw*2CpEP=$GhG03J
z8iXKBw{(Q;9bH6%MV}Z*k83?MxA%<4m-kB>)q9o^xgYZ}o-bf(h6JZsD&2)yXPsMm
zLi5}@Y991Cqw+-sm$l~-)<S#<hs$5Hm%LPj<X*o!$yk@BA2HW?lrawPMP5((h{0N2
z5Mev1M#|=06%mMA9pcyEA`Vx8hLMc=Q_tZwrn`yz=>fmt##Z?)A1i7dM4JWej&ZB3
z?y&^?=NA|w4zd5g$E~c)?Ek=wvHs_0(KMut{i}vP_+K^jra5Mw$o6AWIFKxoYX*6f
zwBhus>8_!p@x&vkBfKAYGJi=~Tq?mXdSeao?%t*{duO<NXC?+rf8Myo)ku=ieTVM$
z$f(E)V7<^WvugKTzJC`dI}7~Ub&)D0+!CEd(F0X})Vq);JU4xpF05oY;rZ#fd?(R_
zvuB7WAD=<6=Q#w9^4LOYD9?bnZ>x-06jC#1KuNTH-30mBr1nZ{>G_hb@mQV;xP(*z
z8Tng2lrd#o2j#PbJ?k!h#UZIsz!T?*Ll<UDQJEa14&u4H`LZyvthf!12@g$7rT;`>
z^jDT84zB<Hi)hGBR~SJK?YV>bfl!09htfo7_{41wD{3wlH1{nlv<MFkqBNr-wBpxz
zmXq-kB5E88QzZl`A_)%-p%mV2<2Q~w!Be(n1Rr22F~L+HN3wGDN(Y0jEiuf~YD^~Q
zt>`#oZzVO@p68bruiX;PeV_EBbs3(d#aCU)O%!DGyCgXf3H@H5Op?X-TtKY<;D!DS
zWg~GeqZq`oY>Fu}X{dnW&&1rKSb+RE&^Ulyx*Z|hK|Kp<(H;G$mJM9A)0EyutxdUW
z?37!ZN6ON=i7DYV69plO?Az8l#43;(yk5Lzb$T+nWwn_)Vi2RvNh4%5=L35A;*Q04
z&WBiE*iacosuz-pnb#Yozj-OXqB=%qP8P3TG!LxanbrJY=9p!*kuJ_jTb<569t61D
zh&3PxhP^2GP-7j|*9oe#GP#YnyvUWv5G`31F>z-ifNgiZysrSBaf$ZRS{&+x<#lth
zDC^L+7YR2h4K5i686>Q89>F-2$3A9oBIb;qNRXv}1<4Luj|GJUFjD)}&QPuJ`$Jtl
zJ?u`8XgO!fKWj*bL<ohUtJf3Hs%WSW>RHtrAF^O9Qj6Wx)#_cE>2C^LHQ_)uX8ZAn
zU)Pu+qDj&O4u>HOlmxruj7VS~B~N)y{Vm!wCXENnrELA*1{Lrn{^c4kAX$x2Z<%E?
zwMM3>(sRER=ochxp<pU{1voemcaoEG^J+D}2gSpp&(C|7-)ERm3i2>rBW|v?^}80I
zbaH&Qc;{DR%5WPcl9Ka1qo@OZ!6o{b24eb129C2SZ)Q#3wvyQ%xeE@-IIG&fhtlhB
zMTeI{IdKIaPl0alfJr+%7P^e0oV$Kf*G=q%XRP(&CGQvMq_*=jWer{tOG7Y-<bd#R
z2FKg!ZLLkPb9~9`4pnq$JP7*ShkJVH6=QyZe##7I`~w7V{sRPjm(l*mscGf^$CwqE
z0l=<?X5G$NEz?}A<k~FSa>z!7Wy7e+yw1IS7^M6W;VcdsPzIg7r+R!cV0ZWFdq&pk
zBL-H^tOMkXV&hcFBV2n;0L`L*B1~$fKjkr52IlS}5H;$#&^bR`Ha_)yUz?l1N9sqL
z$Jc9Gy&;&;Hr!46SCazUJ{-@W$z+Lm@3DzI`mDdTq62w7SKNGoy^U!~IzXUNl*Wrw
z{qb?fVoX<}C|2puyh)Ih2`}bC(Qho~X;4F^ym$~2A!UmgvHBKd>JHj+YKA&;-@a#U
zh#-ZFmkIw5r2Hi;OU|Fb!Qv>)6vD)laCn?h$Il;80`IS#rrH6C*fS@Rlcivi-vVec
z8mDTGKR0+vdfhPM(-hqv;z_P1iRo)R76uKpDH^hrj#T838=2=gb^1l}i@T~QHfhxH
zi&#4Vg=ui`g;7o{i1|dHg;?2Z*<@kKjaV$QM5O=$J$AO5*ymu-PPJsKo_g(YsJ%|V
zuGFbp44p7hWH8ZFH7By=(WA+0ygZO*kmE?4+9Fgau*@T{u|yE~%M5%glTrk^ur8(4
z*k>$MP&m+&DH_c>_v#ZY4r}k;yP?kYz9O`N`Y-n&z!Yog*DrGu3Anb+uYoFc&m++V
zpO@pw18Rmm_$jL1R@6y{bs(rnGW;V<&JOSmJeHvr-(@-qb66SDpVcQNAZ@=}7Y6y$
zY{l0<sv4WP)*TKi$k#Aa3GePcI;peI-O5D!4GLJ#PF_?g9Hj?2$Sy3>lM5bMCH}Tq
zx5VI5<WX+uH);j8)$8)XdSzp}sU(*ZtU?jh(0YNR!3et?33|mS#R~MZT3!N>OgC1a
zAz|?%`jORi8?*U@5%&b$WRG~V`Aq&G&7k!WFDfyi#BH_<!i6<r`&(qcmGmjW+xWvc
zT`i4I@awNW+`&pAs+_kmwOVv@zqPU8xwD7Bpw+5`C-=bo6oO*fh918c-i0_){Doxi
zkaBRj{8(t+*wIvROtg3iXI~PnEJ1}&orCsyx#Fh(eAckOMg+TJq1ifr&YfMJ8QGgr
z-a3G^v;KhOp<9uA6zfSF{{qC#M^JlWpfY!})$dZ$$(U^%etPmPcZ2VNRviIX7rDH@
z1K&Zzt!go$wQ@FBlphZS(nUMBMR=T9a(InAQ+XpZ&OG4~&q!pl))<So)oMq*`5@@3
zbnURBqXGWRVBBj5w9)}Hbjs<y-OY}5_*ybJ9t&=01Cy!H#HujK6I`foi}{WNYRSaH
z|DfjLQ-;d64UN@RNK7D3+h&B#<4<=9lH1SZfzaDq*w<gMW9am}E)`ep2ntzj`uy3D
z16zV2m<zlg$_8gdMB#@MpUM)rb=5oAuB)wsv7n%Ot+BZ$JMpW=F#iHINzQlscdTLi
zZut4PmOUmWj{n$#q)<b~d5s;Z^G)Mg-Kt0!2nDPM?_xg7ZQX289Ra@>GAMvj_C*m|
zI_%2_vtQZ!Iq5GZa?AjK=Z72cOb(QVgR{_$S4)eBd*+RQA*m5GB6}MV25<_tR2}QG
zO<>6No>UV>epl7!>igqm+K`YqGcq)Vv=q4zJiI0ng{TSbb{LUx?+A>%-Q!@@q7p#3
z-iR2Xt>ZD!ZJ<u^iq+e+<g8s^zE^i{63p;G^C;8OO0{Mon<~nx<DspHV4$ZD_6UJY
z3gxeu1HsmY4j5v`vrqB0U}0;AhEf_sr4`{#Gh`7aW^UO`BsY~x8H1@k*zQ0HFwLY?
z#vs4Z72F5Ed5QaBm>LejPua9QH8s=I-ZNS?T|Bj1sjD8epAG<&pwVF-M?NA=HUjN&
zA^2(GY5#q=ZI^%%17kM$P62L(1b-z%E8C8M`t8n_@K)R}zSYtwG8bC%(%4PwaUOVv
zh5SZw;&_%hfj74D2RycEl0vve*?1_)<+(>^G%{E!;S&R0AYWGaU0PyZvDVHcR!TH>
z;xAR*(^2zW1f>J5=MiD3v!D=x%IzV;K~oQB)-_M8h~2bPb1CuJ&<Xh11@0nj)zn8b
zh``xg1PB#Iix2e~+_R@xZ3@k9xeqzD5R|5JvvQer2g?f;h^%2`PT&O1!60ap50%zi
z)^nH6>Ine}d$NJ|ULyNObK*5%LR)huwb$P>-lq}JI!=xcwM9GCnOn+*wu+#v6pT!!
zi?M5oC?hOadEW-gpp}e=4~=XTK1Dx{KvxLva-}JmJ*CDrr<^STUzj#N0>xWtz|pjC
z2+86Rg7HvTlwUMb`EmJxh;JhnW&re+4#nVY@X1vfzp%Vs@lj5y6fj-vgj!o<SaZ{K
zJyN2)VBku@QulaH<OU7m_TrS<rL&Qve$qkr2$^g~e5kX1#|l^P^|Z?(yitnI#-(Z%
z9p@m~gXx|dFNFN_<9VER*PXjcxZD&SF?Gjvd-QtvZFYO9v+;wBo;cM=jomdlizIvM
zy`4n7v2`*_ticNZV&?0iyEk7?o=AcIOktJF1&AAq`$66!=3iI2T*Go*48#z19G2#(
z)N5Z~w$CEGv105^FyBU8`;KB$e~Pg@OCbeQOfqK+T9ZMzjjz;v-MPf%=Of7cm7u_B
z;Gg#9WUdQ%1lr&5_&c{sO<TA0@k&GMn^R^A)2E#YaTPw9l^4YZRYIPj?D4W{*06jQ
z%63`7)A{+{YnCLItG+Nh)WHrFRJVxMu*iQq!#QA)pVF~s)QoN>KVLQr>t(89;H9#s
z@+;U^2P-AjnMt1b?_K|wV?~vKo=>YEztq&mWE1&PgKNjOA312njmghJ4z1Gx4w^wM
zRxeR}>H0vlcE9|<$qdPO0vlBi*M`n_)WFZ}t_(%6`WQ>!;+PG6!5EYCsQnIt2{je`
zXnHcNarG}nQUGy$CP_=X#Mhpy9<_II2f8SAXygj6#`2laI>>?Vux-grWW>(1<WXFk
zM1+F+DIvG%7qbsL2*ia3X|EU%Vk*?vZcXYnLwbZCFhh!T@;^Y_H>dya5Xbsor$`bJ
z{qF*#+5{79FC$X$PDj`7pxf|+lN9~L55~L%5q3wU0p(bj)lWYal7Kl5zAYI>pq|FO
ztH)9+<LT}oz&X)|Afzx2PUPKi<}a7u#y*?^N8+{lKbJCQJFy}<?|!n4da)V_aS5Gx
zGkq{`nzP9d44Af#?-~EB{9LlnXG3Cy8Y`A-sl--AI3m%GevYwGet_T&q`m&9MY8?F
zBLBPd`af7Ci=w0kqoTtRbGHyEkTE6r&Og1!KNM{+GqU~1dqt9_cJ*E(vhS9@;l8PB
z%@CZF634!!wo3%Y0AQKZM=By9+0%xtZvWJY{)?X_BiA+TEU<l<bbvU-`Q;h@;IKI^
zy~{VY{7c}1-}|dj5jP3C5~Y5H&vw_v_*R%ytZ@aOy-QS@_ucF1sl5L%r7~Vz;>%l6
zTFlnfWBF86Tz{|#udMOJ@%Bkyww+2e?i5W!Rk#c<=s^5sAwji(tHL*U!4Jm<l;#iX
z>Cjl=MR3;7<j<^iSoZjmuI9QrHU<3i?a3<WgzVL5oRli7;4&^{hDbeur-j;DHa~^D
z8s~+l{QN^1wv21AoRflTCcvc8tUD(c_jRF{qUSsTkJlr_s-)7sMyY6lv~8sP?{PLM
z->MK59s&guqQYG{1Pp0od*b9Vn79KPF25pN^~|ec(li&UCl7m)>rBZ9tK!LzrKBr6
zQzNAccaRAGv9!Jox;fc>&pI@py1>e>y;cMr7eD<<V)BHA$wwzgjLdY>ndx{Oinnu*
zAa8JAKWm6L>A(xrD8AqRKBRuIrRMY+wZq*xJqPwg2}0c-**uqteHLeF0&+Ldnfm32
z(<*cJAzMepus(zq)67cx8bfnPi(IM-h}03=6i)b!f<c{vXxUOA@E0MEc3v{y8c>gp
z_(2a2daz%`3<1_nZSc51%gy!z!meSupt5&qt1nQ@Ao)&0e)q%v@{OC=3QB+QC2`Ck
z%G7`t9If)w1uXmz9M48w)2D{ETxYn@e9Z$h-m4C`S&zG&u7^g>=*@tf(h<}9guoBP
zk}~RmvZc~!?t;_y8Fn@W9qX9-AMX-d$(PL2SmQ$d&AHCW;W{kQ3k%!a+FpXpGhHTt
zOwRB&E=>!i{uQUxD|v+R#~Qi9NWIl6es6bNg$K6Mzr!r`I$&-Z1tF$B!#)Ww*E($~
z2NE*@oFRR9KQe@-t$w_^D{$UH&-6TL$7C>>W5LZ|IKrnpT8ZXiu_IQUl<!34ums^f
zP^RV1=yeBx_E~Q&Rvo_dh1Fp6B=D<4X0cN(Mr^e}31U?>E*A)c&U^B?QMl9SZ{T9B
z;qK7~4E&%>rPc5rq`B^l+3sYa)(%&T1U0@wSb8tIhsamQk5ZyUG+w$cZi;{{)&CJ&
zI2okqH8dIMru%Vfw@eC(zKbQO!`%7umsgdHe0D!0tCazr2w522?On?!V%s0BkDvNc
zQtL8H7pvecdMz<~fL6e8xeeG&P{ANO4mCaY^(=H2#sWIOA4HKD<<VFTalV4H?pQo_
zE>m~ykaIO(n;8cnD2*nS_fc8Da;oVc2J+Q1;7=srcaO;eZlZYTOWKX?QgscRxib`2
zMG7-pRUACkD#}ogFi2&r${7<rdC{{#s0E`k%|XA}eX)62p5`pCyAeK8?RwwztB`R@
z)8?GMGKO-w^;Bek*rUk%N9a5n4LMq!k{pcj_%ss0kSxOkZlp791j6QrouUD9ixZ#b
zJpQ4)S{4@h7*PLbeWk=ucZy`^!2-%Nr{X%?q$#%y95gM-NcY`Hr02_roKO$g7lT8O
zP^devG?1nN7%Su$)~>{=V{@d)v-ImK%Y)BjpMZ`K{IWsEkc$s3$1j>g9pFG+3W!)O
zs5*aIo47i@Y@uShYh5=qXIoZ<rqM`yp(1EtrotJHM3K30?QN))Jf~dZ-`<U~4kbxh
za&r*jK!46DCMYkQn6}+Km4YlSf;02RuGRQpenIq1UeX}?J`i;2I-;?>QG!BcOvXnq
z^W(A?9`6`|RFx(3#f!f+EsUEAZnV(GR+rK0Uw*u1)va0>8r3JJZTUU9fmtXiHqg1h
zzivi@kIU~Cs;*T;^MO0#Wa@5#hU!IRrDl@p#c~6=gnIVlSzN?AGyBO8$BP}a%5J_z
z&~=dHbh!6LIK#X8>t%Xo>^g~G`EWWS*FF6pR~78l5A~%bgO*k%)cAV71$FhcD?>53
zhv-t_QaARN`h15b`np3*ME3i<4`V<x@*&=ypqmA9B$BJsx!9u<AXhY%udO!vR{u4G
zO;OBzykk5i;|k1Lp1|VO2vIuRH_$CEYI-3gz~q=pMm^zm<ba}*F+>O54MAoD+O$xP
z!lr4!-M=WL_GxETlc5t*1riCd6K@#nkT!gL%&U0=rG3d%XZwT@kfpyGJ@bQ`klsm*
zUcIm>h?!Q@X+Z{}&dTETn4Q>>M%?cGEQWk0V886)6ce5CXRB-V3&RwtjQy?^J?rr2
zi>r(gjr*^*AeN>K@d9RXN4j2Z3PWATHLIvv`33s11&!6#e0)02(&N)(PoFH&=#EPh
z5mcqOij=pG4|Wh8sf?ba&4h<uA~!@NnRs&8_SVN2@3Z;ldLh()bM|Cuv9_acA72&C
z42a$J#sl#}7-dO|D;!?e+u9Mq2-@ayh1xV#2T!`mu8Yj@__1gvq|j#av(a*>fn`~;
zneq{YD!odCT&^O3Q;T)8(~qmdO)O*4RL)AB=|qTmV<UOE9$*oy!tpKARbFnLlb0K(
zNFoZ%KL;d)p29NB1~qgLYhMC-UEl@&dQgp>G2!Cq`1Q@F264)p%d#a~sz&3V9A{E~
zRDzHdZ^5?RQwGw>_^Fm{hgCUGk1v7yw1{2NFdq!TXdbF271tQ)s+?`p6G(oD>cK8)
z6|!7U0gm!k7Ka)^lO{FOnfP$m&&o!3Hr|o^tYkP(63<jxi6hDCmIeD^QkhIyBDnlk
ze4gRq=gO-8A*A>ww*NggV`FFiPpdE%H2*vOJjSpxBKEjt;w!{&o0C6v!+S_{V@3nl
zK9~Hi4zvAuqpg9BQ$-uge4LHA+2;4(u*b|EH-BQ`f4kXzI~ZuQ?y1)rtkWE<KiH5R
zV97TgY)GYBs#l0J2%o?tKb(2sotg1^GCwgVJcPCMaB=l|JEM~qpofer*RQfmz7t3h
z=xSTEvf08K6oa_z@Z1xn=gfpeAepBH-_hjS5sTFu4l2}&<c@J|5g$_?j{7@LgbT;<
z$e`^)$<9_rnlQ2h<mb?rV$>PR6_YhUW_*r<IflF14vg%$OCI0|t^O?goa5N1jFrN}
z_%mV>lUhFC&Uk|~EWFe55D;dwup_>*LVh*cD%{ilWA9T$MSW;A=B2TT&aTQ<&D||3
zVu1i$5@p;!fwlF@eY=>?)ejjV6NmiaetW8zepD+N0dpbMNL)JcihWaiW6NI8DT$Jt
zSQB;OC_P2DkSQTG?X+1@Drv`bqu{bFq|^D1b#<k2;_5_X?Im!<8}e99SadAvcP4G#
z6&6`Unwp$6qUTP)`6103pO(xgkqXZ$1I@gng#TR|Dkwb_cQBd1U?xnnUAV;KNW04e
zabA)DwX73fqp4}F(`Ch2@tU*8yTb=Qy>Inso}TuP_JfSBVg`YBA6y$wG+PnAW7L@x
zsouY^2hTe8@&{v#Ndzz=l)9O$V_GPXr5wcx9#CQ9%#@=fTb()YCu7NjWesweMzQW#
z=&~#}%U1XSnbEme#nb4lhI^AErzPnTP!FsSgq75Uwur&Yzkr`wKLUMfB}FT=V06uP
ztzIgrq0LdHMm;iQ8}zhjl!3x?4&?YYGj1r!G^OuV%ZF(q`@*ZzHN4il4N8xVQ#9;<
zpqvsr9W-hu7N`j!T8@JBu>FM(Z3|1y)iDC|X+4D)a^0wI1Uv6~OGD|_l}ldmLF(+Q
zUEY_|0}8g>+VN|+vN;p%rBuAedJ(kzqI4)KNTU;lffpu%+nP!`i#uQ!@UR;z%PBPh
z@clFwRDw#zCSi)-%u|)6;`;eC$X%}h3k;5>dxS6Azt$j!db#e#bq}?)s_#n)7tzx&
zs*Di=1;Ap#--tVq<r=#zy<Rlr(tHX?Zy$ih2t^%sH!0ZA(EW_OtL{HGz1do~-XmXz
zy6L!u2&jHUxtru9JkS%;S5E^<#{BJbOjMdvmRqXM%3fY)eLMI|t{6drW^J1y+bt8n
zLwQbtpnwtB0vJF?k4I0Gzvk=X5aQ9x)zAaXR-BgwB&4{PV~xMUgL%wzG%O*A%MIoI
zvr&xDuGnWTxcUznvJb(L<e5~o!3hs;s5@Jg$b-NRd+_MXHFAaJ$*P*l!P>=VIgjA6
zF>od^!bZRQ2%Mhp+W9C<#+yws@tRbk{nc5~IXjOI#>1tdH4kj#uUoNc7+2c7Z0`A9
zJzyi6Szp?Vgle7VP(kTDqB)IWlkivMyh~|%7jw3sFl2Dt#%XhB@O5;X_tU=yuiN_J
z!HS6QdzT*vg3Z!fx2lHd<RpBoEsLuf=Qpz0liQb%;!!Mwu{p=`yRoE`m_jIgBHfZ0
zv&??^<bYb}@tdBj;vUWa*%#O@BT&w+P0#q5#6ljwFF;(8<q^_$^XDY-bcf><j=PAl
zhtxryL%>14N%JCxRDXi2(8XP<xHEeSU)^ft@2pDY5`VEEuV^x!XFJys)0`|W+d?k@
zU-zw>-;iM0oHED+>Msj9%3r2y<q%TrY5L3=7;}XBetD#tcwg><%VJPrMd6JC#bESb
z+6aFh(Iga}tGr%@V5i)Lj=J3q37+jZzmA`gdgc|^xEKxPWH}R)ltC9<@3FZ_QJifu
z7?IX~W7q2bfw#fL09^w&23`BjFznV5Rd1sqWI0fqLGZR!tDBIS+ZJNc!(bj4UTSaD
z7LF4YgIBw84Qe#+Xp+H{U;15CkqnAt;}6@Yz@@m4ypROCp!Vx@^iEZySB=HdCBR;3
z!=}UnhyyFlqN9<0M`(;_RV9Ywh?g9KH@p->NmuGB08Z6CbY9SP1CeRBMdCs)EeJS4
zJOT7$fRN-1Hm|(sOA&f+;dU36w(TeWKWsMNnd-l7Hq1<1|Cb5)--&=EEt#0TX5`MP
z>T5bH*>n>Ad+tlHR09mnea6fIkaRqgR9jdYnP{?x;-{tJaJ59yhPLnajnFBhS^|Bw
znVZ9!FpjTxr0ULWLVoX8<>9FDp!l#AUS2zwRwhs4c)%L(IY*a-^!mrs&XMKN;ZtQ~
zRCwq2ahc0wndA8v)$YopB{t%1U2A2e(flbsx=MFr%jzN<4SdQk_XEdr#Nn!i9;&qA
zyr5+_;&v_oWrKdwQc1(>9WV(2yfcE*;}Y)VBVhT!=<vtJYWj!C+x4bgo3Fu(4NZ=|
zMmx(Ll59EWzFNFywNJy+2UWdOyK~EzYwt(aHi_3{haLcTB*hxSB}6yX9jp<v_^~W%
zwdt6<JD|4D!?IMrK1b&W`uBVY3K_Vh5^0!|@`_Co#t#|>KTi=*M?8ydOJeWlSqI*g
z9_t*QH5o3K6m6VXK(du%{&MIzW$O&7MnW)a?$y0forT8Zr6!h>QLsJY+GC-ywL-+>
zMzi(#qilTXY3HN$*XIq%TzXJYNJIOLfMZ3QQ^{IivI&C`Te=KJU$J6uNd^|byC8&H
zkyB4~^P1K4#garCR`g=Epk2F}Wz}A(n_SSVUhJ}ptbgf6>y9K{r<7mJ1}U}IR10f;
zhAROqG-$MKje0Q_)dM-?M8T?MutvJ<v1SO*{02KX=(8l@L%80DXDTwxK89j*7ujPv
zw|pM5YNT?cN_^g2!-JO@RdCzrqw7Rb1w;WDoxlkx_#edrSgFEPg9rMG9_v*vG(Czq
zN*lM9CQ|AI&F|v@<JAGa8U-?gCcpSvNPlT~?J1HO=Q5+>;SfO6w);?e1W^K^V~}2+
zbQ1L?lL|WN<|E{mL{1`1w5o6*sn?8<m{ruuQG_uPgqV-aoi;qD21|6Wh5zVNo)`t(
z-OA@-(jLnW1Bfv?J55V@ylAs&MU=PZL-z1>Dm~4fO9^Kr38ATb6fHrbrg8rt#?CQF
zkhR;=W!tuG+cvvw+qP|^%eHOXwryAS^mivFVkYjHnUg;=GXJjJx!<*(WnnLP=G9{0
z7Ea;-BslIVJdHj@Mf{*G_n?%92T1@?6I)9Y)A>30x~5Y;&$}LFbc98RVQ$1(hmGYZ
zoXyEp?aRX$1MLH8)%mT#h@P|Sc%9C5xa8deO|rP0%14nRu~3Yzn{$I?XG_MFvU{0i
z1SE!QF$a*LH-MNBJ^PaHk2KKlBhXJfuwIXuE+L<s6PeH7g79()ZCf01S<Hj{<xKp(
z1Dzbxe~Ki=V@BMZ1aFzrJjnqPP<4_p!nZy$&Y87m`>PoN>#2sKk)|ODwtm6j;>wgv
zelUsaqk#$gqWREbv_G)TqX+RUQQJlZZe~`!-PPz&8}`LbPMi8LAR;l;uHCRjL1F}|
zp9YfNhGgRvW!--O=VZ%)#k?0jEitsBu=?lBBHi8%l9(!}3GVk4ISP!%b~FlU(1dqT
zu0i#w>T{h?w3CDh5U+otG>6N9419RZqo`@(=M0l#D#@LqY8?x6g8Of%HhG|Z4Q4W*
zM%C3DKv`pL4vHB9#5z?-YyG=dSKyc(xFvhr^b%=dDNfUYG<ScoW#?pdL2TGRLeoCq
zM@i%|v>V^fQ^w(d&06RTkWZMUAI!!^GblPk0?@XX{XGIN;-_qq>047eXvb4HO58bF
z9TkvPl+-iXg)h`O32Jk<^FJ!la&AF3)dahT_9Ek{a=IiAkAK$*?U$FoeTC}gF}=8B
z8;`l`=B+gsIUFMfR%iHX&d|#f=h0Ds=o)sV+i>Q*iMKAB5Re`g{p}<Zeikn|N7SsY
zUu*y!S;?$405w!V`~2`XTzSH_v9pgNmA%A?AY)h9W7R_DJoZ<69gDKPiF83M$LInq
zYOo-fq2&<g^W68uEP)CyS+?-Ol&OO*ESj3b5>>f25u@41Xge9wP(d*{GF^@lZopbf
zFpZrQZcN`%b`voI7=)>mgw(k%@5LT_>rIo2Rh2;;$Ad=9=XIo8UFE;|$GN$i3OP;-
z(8w`H6QEA$?KY4@_h?`>eRgwG-B>4%69Gfq64i1n6UtNkTligRveQiljBz@<3=I59
zRS(&H7c(M-j`oBBb8KFSwI$|QzJ8#nwPG$(Pqu?~1H>NU58&^MkO=Z0dIyrg78xxN
zeOkhYVC)#pKHK^-E@G-wO6+f)BVHPr^MNh$04<tj;`7~Hno(G0_B8A-B0RmkTKr?L
z`#L8}&TEp{2CdJaISzzay?HeU7Fi4TLXj?+KwU76j^$&Zr4gVy7u#(qRD{wNh9W0q
z&o*1I;l(YHr3^R)^zx|57N>HN=*@L51Kg%(#Rd9v?|QV)NEH2^%GIC*5pKDb%cOzX
z>m^uvff=VLS0cPbI|w~A>af8rm%eCRsPa5Q^-62DL)ugj1F;)HvEN)S+OaM46N)JF
z#?7TXBEusbJWg0H^oBZIddO7il4;JUl1Hne#t~AxaSkEV^?01C_X)t<?&R72*k)ld
zv*cvvLjWJ$i(<&VGtaX^$4@2~^e8)&+S}XbK85gontL~1=z3Z}0cw*GrnGlBDgt1T
z$On%kE6u}Fe76yX5CLc9;rX9TNn0`VTAUf*djn~n3O%)1zsl_LOJd|~CbWwq;P`$b
zieSP}c=U-wT-`#I0IW>_0Zlz7Xyj*3>r59bk39r4GKW$3A4Nd$jAP)>$lf)bNwLUk
zV$R5FW?ag8Q8T;*?cUZ0dp<!}6zEC4M&}p$`4k51o-y+uy|CmrE^YNH<(W(@PpRa2
zJd$wQ=fuK}My!M!f>0S7f`Nw;Q{GQ#O{Dg`c$v%sU-FQXyT;(^4`%YH=Y&an6aYTR
zwfX|l;WFh0rWt2<3hpcfe*QKuj`TfHC%!#!{2ReV{~*%eqo*&unBu3El~;1Cm^q=;
z%Nft-g1&${kM^}|?L4h!e?q^#YoGfI!0QCv{)&zbV2^;7=qP3g{c|l5#QMkm?aq(E
z$8O}mk2%bo|9jx-|CxMX{NIxg|9TAS{<Ud~&57i*s*bRhHV$Oks6nHPHUiL2&<T76
zB$x<l$~G26^tv&!_Sa?>baZ=)BI<R;p*HI3=Fd??;`+RAP;W~Tg<+;qKxo=D{)RDc
zI#`SmYaPt{%Z$VnxhoyxE(=9?W<MYiM$tbw5%xXV^aBGFij2cdB6yBam=WE%>-QBh
zmvxy)oE*c>{`}sM`)r^bOK#E+2*rW3Jy;2XX+P?ra;Sjg2zumMya9PM9D`CQ4iFib
zlxQCy18~H^FgsowGPidS<O#&ME+eNzNfCURagNb|AD0|HY%n8H-{HONk~+hPNgSl7
zb7aqe7jexQ`*`>&<bm0Jf*Bz+1dPC0zcEO4q-5fOqFhuDd13@y6*vPDZyRJjFO!iy
z%pUi-J_0`Dq`w1W2orQVz}}hiI!P&W_%MmmKoaw~nRn6{i72{&IZ{aGT9L+`3KW?M
zO}mslsC+PrELoZ<A!*RN0T%&LV-GKC5@=P}l?)hx^Iv0R5R-wUA{kTGLB2R3JzbIA
z=<M)imk$f56$A2|mdH9#B`FoVezP$H6#`%k5iTsPgUXY5S%guSU>`XrVlles3DV_J
zQ_>TlAU*Odh;<@0LXDJkd;vf+@g7!XjIKim!5Ly5l<@6%5Hld^8DL8tU%vuvfNrIu
zk=%zlgfH1MwkSxOO6focFo-lcAV>b0hUc&tWL3n#*c>tB#|+JevO%#x57a#!HJtO<
z=lcGBmDiu+0`~34W5?zkms>i1o6A$bJUYKT7kahJLf`LxgV~O_?cNT(jx_u--*gwd
z&*2sCcxz0q#O~$x6k|pc<-p@$!rAHl-{(6SH+hw>)QjCr{5E%U=av?{nt!Igz8+!z
zhyf!eg@AeLSpyK4uN{R6+?^2wS8t$DvI%PzXt0loVBmt7K|dbgt_N?^&4db)9g&5r
zp8$)ee62Y5{=D#OuoP~3czC$J+*?jvE<0+0_N<Vu_Huo>%;_wB%yken$AxvUUoIc$
zz$ce@Jzg}Q@4V)<o*Z6B^k(3{Hr|jLzfN9*xzb!5a&=E%ICh`0FRv}v{w(6>%;;6|
zWl2f&p(Ud*PP0F<=CV!*<EpoB?Nwihp+tEm=wxYie(LIs)fBIMtT<W;TNO9gH<vfJ
zH_PfC3Vx5&*gTYt7m~$g^6ci8lCA2ETRfD1&!{_Y-uC#Myx{pQ-|W0PXoHrTLmLe<
zZ{jSUh)g1dK2gWk9Q{VY-Us6q+!Sm_SlWgKk~{<+A0L?rZ(UC4hv_g7dw3H_Y$c^e
zXpkd@1b`ZhPQDKaUn`=6m?DW$Sz>`_7P9pw0aPhV<S-*H$OQP58B9iq2@Bo_&K9%6
z>+wgljWmJ1z(~zI&XPbQe~hx}ZSnOchcq%_J(5@(X0?yeAY7jfYh~$)qMAYaigZI{
zG5?Bf4HhIgL=FsJwA{MduM`FLI<v+w4Hj2N7%6c8*rU{cB=7>rzdG~<#pL%lxJnNO
z$aS8zR3cl0<GeWyDxE{XuNrx<+~uvpbZQFE5A%P!M;{Sn-n)bp)Iszf0Xj~9rJs5e
z%o&Xp+z6n^LOc~xG+2oLzCI;hyQj^F8zx#3FS@QGpcetNLd3f{6&^Vi&c~O_k!kux
zDgb1i5Pf|rT(TF`JuR`k+;ACt59n&yR~Ls5ch2@(X%e%Aq}q;v>i4(x5M)cW=P%<$
z?LB=II6|@rVTO_2rex+oChH9d_9&ggI(G`#NaQzkDTPU#T1Y{)nrnpk`|l_wL_~jW
z#)3YM6nu!I$OQklH6uxc+{DxueRW)Ty6HExlsX5_dp&7nAj3jI0;ZpZzC<pQ14|1H
zX1DhMff@`453>d5h$e;*jc_nkV{aoc^|Zdx=K%r^83zP$EFXit^}v<&jSllEswZFr
zZPgZyAm{WKZDbjds&!aP@mzwCdhKLbbz+k@y@P-^79Ul$g~T7X)D&B^MBB9VTwS#Z
zw@In@dK1#>=2b^&YI=)epVby78EccgwUJ@9mx$^sWUY4c7AARXlianVVlsMGkMeP8
zSzXHqrKGgXuBHEZt9^V@R@e4{N&Xj|)XFN=r(1H=0*E4(BiH1GOFV!XNj3zo%@nY~
z1_~<{2Y0C^-hvz$mB!ESCU$v*5<%?!%ObUk=O^He!YnU{vA2L<3Z>X?A^t~B)pJF^
zo7@5uQ|QBEFCqvVsA*t~azYn&n1%-k^j8|#h2<&F1P%QQV1P0Tl!C)K-*pTXXO*L5
z2cu;a62`10F$4ZY<RYMsb;q89{SCW^gjNNqmIQ;)TZMqZq}ZA5p+7#mLP1Q}FEoL5
zoWc?lNp$t1^dG9LHUlB;(@89~txCmEd)x-rWf47wJ}(J7#&NJJ59H2wH})d+BNuhm
z<#>Kopi<Sog2#J3^z9Q9eL?yY$dT$*D8^}E#G%Bs!2n|Jy1w~)(^^o1jQ@4*-TRCM
z35>Mj;wd%a4PQFc<pP6*zT$&02|mM7NmUT%+Ng-tB`ns($|$f6;>}M&;<c+LDojo@
zO-W?u;{|9dKly?R<$E`a9Q#^v3GU!JZnr^^UZx<}iv+xE#z9Jn;IORBdHrq4gG5VC
zp7@!dlXsvHhFEc$`&9nrxq3;oXF{-lObKTK(eI%o#xgXalrcUS9c{-S-yDuK0O^}5
zE0~W8W)OY*@;mQ9HnzDZd++L|T#1$R`_qPu<#vYY#04&TVUFeWQr;7=QrmALb4}ZC
zfh8<Au{KIx@AR^PnuLc)FU>VZXyX~uVwJo|CoA67M(CF=eJ3>e>(TJ;FCnvh##QOE
znRIFwPZ<GLDT!ww9e=sYwh~lT%8t{puxHrq#|ruF&YNC4zxm!gxt7hSY5tCzH)k)$
z_i;M4(`{33Jv^LeKj+iO>+wdfWLpfwU}*uO7Va6rrs}jie>Ejfj*iFSfP#j~XNOmJ
z5y%7;*ym<b+2$h^PN%!vjNH%*^X9dy^v<oP%jy36<8#H`9J;uV*%xCk-xrc4&&`D8
z>N6c44S2A^O!A^Lae?leKfA)!=->VC-X5I%*r2T1a06<3v-MDjL_jTa^6^bO>dg1x
z1G-h8?=Mercvzl2zig?#8W%--x1O?Y7Sn=`6~V(+bDl>Z8;38-{D7R}#<ZfiBr<c<
zm{}$UdR{UUmL^f*gL^svCM*qY3w`fzmI3j*c(Q~|@Zi~vMx{4$_8L+PfP^Z8)oLg2
z7N3lMEL_^Je_AwjyfyRi@~q)cCJTeE|5j1w&93(HcOJtUyH(%#+S!}QELGO!uk4<(
zUeM+d<~d9@ZaopiJwtb5+qHfRdBp3rG48uo+tO3Ej+k2z8BGRzNwzb-_eyv?86VFW
zXW;X$*ZDQ*0pbLz-3UqVg31Ez2Bh{H;Xc$k`WoRR;mwcsJ75J^eZA+qgh*WqI<<Wq
zU1CD&_$_2@_=+6EyTLI&ZM+go!ZXQyR5KzPF53--jV8nuS_C`cBf6zEe3?&7xGAW;
zT%PAOOcYlA_wkjL^`D$9EG+-p{k*ODzmA0N5;3XR47%`>UiOMyGvzj&ahez97J*5X
z*h-nsBa`v}^gst1OJF8nk;G=@hL0NFUmtAh2n_UV`UkCg(U*FKk%MIQ5fGQdltlTD
z3)n|6p#LSP{J>BWnIl0)46^gf$LZnciS^1t5bn>%_2T)lN0jyL&!1<Vzw+bC?{+zU
zxAx-X;mM-Fmm(JowwB(K`N{fm@a$Lh#vni-15;e+B%`hm!d6Y+?ne-IHzOir7?cEs
zQx(NHQ|X#|bal8uG<@-nJdL<SPCBEi4_7?(5kM%f0f85>;V>5&_<90<?6^)6QF&n0
zRxy@el2~x+nKZIEY`bI}o7HIMtP&A<H}~F(O>A~W+VzFOc)KfAZJDR0sb<eE0&lWj
zGvUqrqR!t2YHeeWcNC7#A0ebgQ_~G`NHkVeJ5nn(+<&I3tYC3dLmmKvHI-J&+BPRr
zArgI3*uG7YFvPj(oW<Yl76TDZEc+b_9Cq5FL65x36IW{q>|#|8wHxqVqiSmAp<UhI
z4~5eOO2-O@I3AW|;&B|`a<kzV2O2sbv|(2VT!}9FAAm10v`i4nx^32wgax9#2?A-s
zHtqO{gK8doA;}cgXR!h5pPW?NvJN`!K}-~Soms>)YoCM@)pN9Aj|EAf);>@V?nQhz
zEOU2M`qgq+F|olyF$Q`n{e3b)5T$ewZ7E~@IxydJXC0H_4(%N|g0pKzMzXC42LE4|
zZnDyppP>B3X6F+wiC0wA0eaIa=RIkYZNE?|qy853Eb>TRJFMMbJ~ky(xzVlcao%Ny
zO$HMGb~)6;w%u_4_1nyTP1%Orj8dj~AHpY`RY-Sj+I)~6!<o_30=U3BjW)`vbk3u&
zZ6Z-@gElDRSDNs8&=_SwBzL%^Ws<aR7}mo)D)1^xL44z{8cs?n^SmW2!NPGrTsNPO
zHZh#QCPbUHYYsoBy`Js9Yly!E+z7~)7@&(A<`N0V9|hwn(67}Sm}8WC?+Y%O!M_$)
z*lct1C~7A~MPL$b*mhXLEB*C{SO}X%VKRDntK-iIPr@H(_}TH2KngTkfWUoGRU*_5
zR-!<X85V+saBzEEj4=kqD1r3g>TM|0Ae6MI8*Y_>22{qj@4(Kucv*SfeVURLhcL5g
zQ6(x)tmDY0#GG5A%_d)xnJGAGBoSYj&>7cm3)iuEVZD;94GKLHI<n_5Ii)Kvm&kqt
z7ScC(=d{4=7(2J;s#d=7Y#X-;Y(+C{-C(|J%H0e)^h6yRl8@p`6?iFUZ|%3?9dnR~
z@<=^`i~r6$#lWn_&TX6>a{92(<^RLPV}{LR7pc~et4_S`i8z~ueVJ-ai=h4arR~c%
zU{sX;^XK@oTDv4Jy8#lX$@?SO=Z%H3vr~OzhX1l)V(WND+|uh>MvU3jhmuy!Cz3iE
zqDYlM?3?9&<8ygiJiG29Nl}<!&=9nh-Vy!Nf4dLPVvnYe*}ks%YAOr??>O3;J<&yu
z<H;gE{Gt=FK-?zVi>)5M7Ty4CdAl)A5W|m4sTsFiXyhpo2(<^CxMf;x?A8|o$X!5y
zn+S)J)C-llpo=&1_pi`9aD0CX>MeReHws}u{{+4dIEK3q(8fXy;XSKwh*Q5_Oa9o^
zQ6&3fmJ<t&0ddWTY1>GaxRHIkUn<CLMjO&RLuIyfq<1$u#g7=dZ}i@40n!~#ibqO&
z9Zx^OLYRhcaxZ?aZ&T>LAd>6#eLqSmBaja(j021VQt+3vGhD87MK?|FiUloov@q2u
z$8q*fmoRvis9=cC<)HF{d}Q)UcYA%-O8018@M!Qs2#!_87mX<0;$HdDOFZ9qnWaC_
ztafJO%6dc@<J+Y2em{w&V5D&|X*~BXsXWgPae?6KUE~E2_{C#Bm)kk~F0rLve}G;r
z6c+zM@&Az!!@|V;Z%qKr>N<9R2{AsiwR4jIm@yi@5lD<zgx3pngLAOeooTihXBrV9
zTz|I_vE6#T!U>Kf>h-w#K17rWuTiJDUXnqd{$RKCJ2LdTy-7rKBGfg3=Q;!`2-2cu
z(Lv`tZ%8Fcy`{Z9zN2L~Hij|?!zdz6UOPXI&!bNtha?A#u$&np^#%LWVdu}EmvQrX
z$Rir=&zq;0dE24v>Up+19p`_ga!1b2_s`upReSypg?Z2m%0_b222rIPG_D8_VY!;z
z`LYPf{>YIuvzDoxb`E;0rWt73jw4w%je&_GQJeozvi)>+ewX={+>E;&d5iu5FO9mU
zNoK=`flP@^gt0~IGd-3vHIq_u=%zIL4gb|$7AIj)9kUQ8$(%nZ5n06FO67PBA^>=V
zju3SVEWYH&=LYA;Z4fskA-+{EZ|$R2=vUs_?KK)EcFh!*og?ht3NjkVM(`x`$12CX
zDV;^x(|5AG)8TF{kLpGpJ>nX5D8!wu_dJtEj4QU`tq->|#4vG58HBGD!f+IsV6c-S
zgA+7SO%YL7yQ=y-X?xt2;!l&D2BXX5oHnOt&-JfxcqT{#F*V*37dOY=#+h4acIsO$
zLS!_)amj3)&J!=u%3_O~iHrXBh4<e;<s9=g4suhCs-X;MuokVPlyesS!x8$9I2=rn
z7!2u5D@5Xi1XXd_wpv?*p(IK-32nH5@B>nZwu8rQ5V=>2F6@@lUhMLz491KDZ;R<H
zbOdy~riwzzJD0$g;>CAt<#AXc<ub3CVlja1BpfU(q2o>TWQ5wJ@-y@E3`YEMhjv$-
z2tz9Li{$_({kDyY9@!F^c)jcd&6{_gGGKQJWTUbdyPMZZ(Re-w7Y?`pWZ*E#d~tjk
z81@gy>_Keccnas|$=v;U(8|}@$*L)YV&0c(I#f$WI9QSe%3zX6NXwHEoSu}7{DXrD
zQsLoA21xSgd^E8+69e4Nrz0hwI<;o9CTM#gtvg&V2$>4*2!MBN(BQ6GSWzSG`!u@o
zy8wR(%7+D&9-GpZfC&g|qVuK_oATl}9l97p+D6j>(CT^M?VePcU+ywv1%Ye?9r=iy
z`Rf)B-33q!56Yt{w<7bKrYBlJ+HS}?Lk7Nfr*=A*_Q@(O!;mt5B}Zc{q@a*t*fQcm
z@!Ice25A&XRxI5yTr>(~^l;XwNf)^D@T`X%1gpV~90Tm^S#P@ZX&r)LYd+z+@M+y>
z%S2Y$HnflCZ)>QUw#F%N6LB_r1>ZcR9}|J@<5I+p$cU-uU(X1VwVJ{yt?)QEhG@Q9
zm|>U1ol$v58Kz}eY*wwgI@{DXu}#`;W8I>=#L+i!V#+2V+UH3j_4f$Ne_nhlnW0D7
zvhcB(sMKih4v|=W4$hC9FLQ7n2;TjfBVHAMb5NMEaZgg73NsBJp|5Ou+Pe?pHjX~y
zIO(1gog@twhn#@G%!cd{fpNyw{ky|yx9vLWbaAkXS~l-44rj}RuAVhkWyVGY#+2#t
zm$T6afpG8Y$_)6%Zqzm#l<FkiUJNAVRcb**4wu?O%mB_&`NU{jHq!KY=5H{0@gLdd
z!?+6LpkcRpI9RpOWO2s}2;b_^ZRL#1xYuK&-Cd<!9NG8cQSSme3^fHO;~X)_4D^Aw
z=6k1zR>m=SqyEOO@Omm=BU?rEIr~E)8yfZsbDu31gQtLp(@;ANvZ}~gU2<1=DlH?g
z!aXcOqiYE%eU~k8$im^$CB?v#`yP<X8=SW|NI4hn>S{kx@7Gupx#4jp1y0uEOyReC
z-Rj%kw~z+S?c~nwEyo14Dj-;05ps{x&d?JumGvQF%5q|*mR&=CB}3Km1oZ(ctF`rl
zdbY)PoedSt`zN1TBL<B?u#7{|zsDm-Tb}+0xSA!x>Yw!eAAIACj12!;->20#ZU3TC
zeO}ei6oH87!qn@{p;>Gi&GEbXMxywx&_dV}i~wVCG#*CISKTP28j6&9tg{O4fGxb-
zPYO7lD6n`ySlqcgb*FZ=09XSE#TZd(QUTQiP$CrwgNA+X7>$FIC{iOsS8w^I{2~p>
zEZiRrehv@P+#Wa2UWaS#Y(Z{;B_t^@hH7WAt(Xd9grC58B<3_6tZ-Ua7#B$xCvwK4
zCIFfBd5t6>Y|YIc#WhXjXY9kNIDYe3(DExvwrY078jAic%B<ak<W>{-+X{ifU?fr*
z*E3tO-D1zFUO8oBS^I3;PJL71y@+9*YP;P@N}JXEjtxA6*I-j5>TdE49*Sr(M+gqo
z_R)aE1t8@Pv_~baG4Ns0Y}$|(P9Ws0y5SW-vfl*`jG=EhaEqn24Lq0#JlKeZ4K(5l
z{wypELo7xQAj=dEjUt^(@8iHE2lK-Kr}6|W3&GmK`o43BYU8l^Ysmq$-sek2pC-_}
zdZV|8DAkxLRX#a_dBDM^{<$3_Xc~q%4VWS3`geybeK$e!ToJS--Z0j{(hKK}l@syD
zO7O;-IY8>M5X4t(%ul$uUla@cqD=xi{O%hD=ybaXVk|*Wqs{EbNIvq_MLSr8Xdt54
zuPuBh=uG4YUY>Ke+!=W<Q=kMyKmUdp0!ZvkX}W9SB<<i?>LRK$2I)zo0EeTZl0rjW
z1(+J-f|!mnjF<KOsX2}o@S74VeQLSRh^^VR7xMT_HuIz>9sqJ6alz|-w#LG^2~#j7
znz&2US1aAN=BX_fXJCAOkqP}z1n19JJG1`L4{v7Sdj)Rc$pE<}QzjXHJJ*!QwYJoK
zpI>_zwA*Q(Cvtn7!KI}xxy#QuDDek#!<dSt>NXpa-%8cmYN;(LHUXwjt-@b?)k==~
zhTrxgwD=qpckja-3>gMBbNnLOVR{SRIlj8B8wu!c)2WDDiDE%9a6=N~G={+8g%M6G
ziwBAs7v-+q>jtwhEaRIkD8MQh`J8%Y_>Tjd(UAb`KC}(`at5_q8{c<Kr0qE1O-wOI
zk;I8nf=6+k;=MC18NDI^2py(kutaLH9~+)y#$}-twop^6V;Awp!SlA-dhdkkHzFq{
zLK*9VC_e!#@VG{_?hEFPxibJHfXg~rUs4tdzxJEV-6(cFO`ys33&2v!RJE8^%3APH
zEO1Cqb553Nd}b!6q@@_JYGovdq9N%Bj(6P`iP^RLa-9G0@5y!aCUa++^L3=-DPbmn
zV+-9mjS<Twi06{TamnVY@5jE&w=?(L{cLVY&YZ?FemOp9mzWRKnzz2^tzSa?vPj@n
zfC&$VRBIs_k}|@FxB{1pD7awl%Du1Pf)K*<QoXKd)BNbN5=Qd<!ach4yziR0RhwpB
zXJIdVpwlUR4q^I9tZ~cub0BqJEtu;F+^vr5M*rMGS{Thh7&`c-qQ}H)-r)M}!T1Rs
zU<wBO5Bl_fsV(}~=;mTI>HnacdrsB=!bqt@<t2GcmA$eGrqN<s(yj{PZ8p#dq-#h{
z=k<0G!xF1S>d`5@5qNQC9<$<#c18+!(kf@im!D;z<dJb7=A>70A$UoJg^C{%Q(n)Q
zx!x1LTzQRD^2`~@B*uj53{q_Uy5E^IX2=26mrjk5%HluVUdNv%o$<a{4Wi%Kd;PM0
z0-^Hk@NR4!I6qzevbKD>{mgT;7132+yK^nTYQohbTXlnGZ2E``jACXu6^-q;ykIbj
zmwII{$*9@REPN<ZW!2DFOPI$p5GP1*it!rXmBolqJ;vdxc3JIpLw(f_-Sk#mTE<|1
ztfao^`u}1Jd|3?=>=tm2hYecO8Fp2k75J&K8qbT%KTnEdUPKnGbBGzR{}v?_zxO{m
z82F{%8IJu8DHhUUlS<%>4koKvB<WTN6@?eg-{Cx|&dvFmBQw<;+!0H8ERS5SI-LP<
zt0|9UO4vI)+k7gby;Ilxa29#I<S6Z|9LnNvepd<bR)NysWiCaWnA+4a4KLXIB%tZX
zMWcoQ`5O)ZDx<N_0WTdy=cUq-)!FyiPV{;;|7PrmjB_r1&7q2H4*}?!V5`vS?gPUv
zX9a!BC6JRU%XT)&%{oz72N6wcbjIahv>2g?D0DZ!I1aP|8y8E2hi6?Tc`^94FjA$B
zLJF49QHJA5p3so4KOx`Z%BO$JIr&KEx|#Xl;l8FqQIlQ$Y+j<KRx@j;rHu422?;JV
zuEwaJ4I#k6*6W&2C9O@<AWPm)e_$Dxw}Ap1sq8k_uW_fJ&|b~8iK4;eNt5tIi)OR#
z!k<C{f+7VaC>5H3PaUoYcjBs<tL6hu$D(J>+q9-KHv|tbCpr*JricRz-=bebFZ)&q
zYo&`9nR2MXceIwo_;QW11{@_E)KRlLu-qQRW8HN#U58)-p5S_lLqH^(?ybD)>&;&k
zFc*2}?hfVUI$$Uy?*ALZ%?v{KA}h_gb#}V6z6?c0t&A|CDFs!;PXLOL*kDb0f}`H5
z?T@D`60sNxqC@MSalvMdn2du&h~K;57}*gFfnkE)oI>m6AV!2h8pO-oH)U9OI4j{z
z%#$WEeaJ{`;?r3X;r@8~6n8((#H+}D+*^fJ@*wEKC%xE*-w<byf98E1Gakc02tYzA
z>$CjQMw9*$BT~LM8OM>eqZpKLR(K}EWq>4%SnIqpC&c&b%21>F+u4VFy#+XBXwyu0
zv+aK`wP4+N9M!xGEZIjmFTzKvru5{RGEqRuU^+)M?>8el%jd2{2%u~ZCoo*W4Q7er
zR6|E518R*i5tJ?F2r72VzRJCIxzvJxSuGr+xyabMj=d4=l+Xo+ForODd5ZR!KvhV<
zj5#jfirIhtau54NZ2G~u+U3nH9bfk2p4;#K2fg9=2OlCME7QN$n`;eeyT8?oJ(uct
zXjpYCp9e;WPjdh+b3mest2XeGa8?#3jlU&Qig2iVdm6cgBZ^45CFj8-@sMt;@II&6
znW-Po*be3|G4s6jbY|)3V!}m6iX@z!yCYSw&bR7@jOU-3JdvqgUv54+yG5Q%>Dp57
zQyP!I(wGTFtS76ra;HrO4%J&|(Kx!c)S_z2Dp#WPv@KCFO_ee{vH!Z&H{3~xQtD7}
z<|HQ;aAa!wkDhhrbEj1&4u^BLooo8X7n-9=+8mnP8fhnNbyXOoWx6)9ptH7fuFZ{i
zxp90zM2X_8vX_foV1W~coyoHOD`RF+R8MUMv220pk`DcoR%A?|a#{qRTsvJ=SBXGc
zh%^+eT5T%!x3umXz2Nn^Loo}9(j}435fA6JsY;Tsf{0>+Lmn}R572bBQSht5L2Q<&
z8?FXj`NP*P&KVq7!;}0u9HyCQ?<e%qwspF@yMbjlhD_#N1uJKa@KJkLLL@#?_!2W>
zXC3TpDsf{2o$OCaXVmC_X>*IL-og}IvLjlBdA&MoHQ5}rz&RXDVYq~+FeO@K6dsPj
zWk6uG<HoL}g0aK(x;9cm2Sd&*|4<t}t>3~8$cnpSkNVRK0R>2oj<KiLUYN!Z!${cY
z7Z3g<Ivrsg8ul@#IFSXD?RsGVn0%<YgaHe&M5(vHS{$P9Hv}Jqm7GOhwD3piy#lOi
zqZIwFx}4Un5;g%XYseusdj1w<6zxla)8rqAPce-~c-6m|a~+U$?CWV&G8KIdd^l;f
z+1wu!<b^!xQs9-!Jti6bI^Hs_cBU<NhCSgs-`J~iJfy)g*(HQ$VQvpUAu>Z+5eE9v
zUN>d$0VaK<c*Q$KnPXxH4kxH3DMfnVt(QjPULqvcx7o~6DRlPF5e+A;B6c~B@3558
zc@OWn5YR!@BN9P!H(1SD$axJu8tjuAcWeVAMec`_0TQXK=^F?QzL^n$$;~B35h|kq
zr%^(^fVM#4GLvLmftrw0De6{%Dh|)8U*qf$3V#}(3JV7z<K2f<$R)Krh^@_<eIsV>
z_in};$32IFsH-K2*>3)P_zD)MT9g3>=AS3#gh<}=w{{#thiDC5`UWgG{buBPNf=N9
zOj5mDDD%05PD%jt|8*Zc0H3ucwrR7zzm*i}+|;yyZ*!d~)<Sw%Cn|#7Mh{!0O@-G9
zU(yVR)`eaU$c}#OiU}3HtBGb`0lKr^aOT>ERuPEbq!73NLD@8E_qiDYe5mm({l?}$
zt~Fx<oKh|y-j7TMj5fw&Ahw7c3N_Vn16IVn{u$BCzG<t3f6riW5v8q|x^&EiGjNd8
z?S{+=gdwI(^6|kzxy+<Yqq=0sBjDVSLxE@LT;U2OF+oTelc2(L{_NP?`*Rf1r1w&O
zzqa=%KMq2sBt3%4tLWS9lsoUiRtCTT*?$-HlQa<GQ$9!<?S2R1D&+0$F0*@kH%3Us
z0RMH<Y%<u{M`M1sPBx9aBQr<{J1MfuE&Q7U|21v&mi@|o*vqXc*?bsj4(~2(<0F+E
z=O=<VJkY}AqcAKj^)&I3QsKFz{MwJ=mo;9O9iI0RkLlb4QwGWW)8>{~X-kc+-|r>%
z)y7>jF9KC}aX~jmO*GmS=H4nLiA9>QJRSuV&NHWZ%a_A6uJxomQpzgoUdUnsh#0k+
zv({9Z2SJr4eX6)3g?vTAJ#=)9j8t32jgkGL>Q$DGR#^i(sXrKxiwY&*eQZG}m~EgW
zd-3i&xFNQG-xlhLO$X{_2{yh%5%Bw9WBw}MhvmCVL)ki&y>hw8RpIL%l{#cEihT_K
znS>mhpHSt<mMPx#>0mzlGwv1P_n_-Z<@1BN(EsPER#@z1jA!<la-Lk^nZ(R`L6}eQ
znNlY&gF&`z*H5eC>ZkC$;gerSj_~nTE-i!>{V$3=nQWgB4a`PsORd!D2=nKbBhohn
z-_wF8xOGJOt8;2CFNcGQKn{AW5Xh1K&P?n5r{K`d)N(Jtn&_<=uy^mAka`FL+ui#M
zu<4t@@SnaS|3vO!W%$==>}$;Nzg+F7n%cdD%mjogkU1InSgX<-<HQxkji9V^1~wDs
z1c{IW<H?*KFMKp964~*H<08H%p?H9uz_3<i@H9X;-51?x&)HZ`XqE-Un80}seMJKI
z;J>G2XG9#(kk`HV-w#*m`Z}AU7=p<FP@SiDr$Kt+;-%k&s2s*+{3wWU@^Nswuzt75
zLIXkSx3RvCmGp2sxi}rGhi<j-;^cO^y{sUfYfejs;KbGXc;F=ZK#(%Pf+$4V;DwY5
zKZjJ~JbuBTP<=$E?~e?K!ig1RGLFnn1e*|)k!1_A`Oa`MPN*Y|>m;&tTjCj?<5O{>
zh_!n|G&t#yxFsgI6NM6F<1FJ7m(IEVWf=SW4!Uo}S?*j}B*5hP4e|q(Lk5{?_3usb
zp`Xu?(>?|PDAB3&Z^AtwAZkSt(o`aIcu|x3n+xZY;6k`Do_StYxflb%Ali@!1kf|3
zyv(yifS;-Y9EL(*qDawRy}Zfe9Y1WfxyVB*eXL>{=B0Vm8~9=vVf-C{D0}iWcd#Mk
z9zAho9WJgE3M4KJMKFq0IM9s+@+#?iG&9v$>6^{!(DaY)7rR%v#y#oDH(oHF^Tk^#
zL?f#!tp&IjGX5=Pu0uA71Obwsxt>f(JbAp3AqKyQs3f%v`obws;|dut0w@>@u<!yo
z^GBx2)v2V+WsmfAwDpUKY88J>EwSdhh@|2PIud*5Tb88Xd>?E!NPws-6pA&2=qNZP
zCU{XC-msWU-8`7z)-u@_`oNL--NkOjVX{21H`0?E8+!4*_XxpV0I>0pBKvsxR1WZ8
z6W_#y;z%5SIFQm_CAOsH*^65X%+C9+Ki+yaF<02CzKIC>B_h!p{Me)u^bDF)d}y0k
zcO-e^S75TKMG+ri+G&V6>1}X8<Iv2|Z%nY#L_H75tG&9l(-F9CX8@Ykj<68PSAazq
zK-GkUxrolIrvX4<N29K~WFDuuG>}G;JZ^Rg4r;foXm_~t7j89w8KagNWpJbtO08}A
ze{A*TmpcZP?v+!fX91B~g}IU;f)o1csc)nH!CaNFs7;ldvl-<4#A-ygF}D+CpFGvO
zG6iTd5Eu3RUL~$0Ypqw~31+8fvg7!?6Sc?ozC+!6fk3h2-h8@@72}L7`#9OC^X%BD
zqk7r9j<6b>uo(O(D2Df(oRMhV3jVt^5){ip7>0|tARl_#OhIF^OhE%Lxp>0#<#RQn
zZE7mq?*M3@iIUfnPO#D%OPmuU)D|WDYAQeoP&D6Gq!)>w{eBe;r-&d~Vy=N`oDS%g
z3Ex@%L*HU<CKP8d=4dCpVjV?Dk9V{D2mSW@p!?c6sZ|Hh`kohetxYC!0=amHTSrwk
z3SKf%wTGNjFL``8zmu{6aT`UmHOp}%E^o)}=!>YKnh)MNM~)0tQU2kP=tVic*)2XL
zA6fy_10>~+D}0mbnhqGMP@;BUv2kHb2@n?ZjklUpga}kh0{2Q%jVw*F6F>^PaYd}B
z2}8+p(NG0O@`^i}Rj4VXl+6Tm%z9Tk?45;PXMD)}S=gs>QZg-c(!IexAR;4hbGkXQ
zS`-bPx^;L}#|O#OX;%bShxVj~Ig(-9o~|D`i>r^#VYJh8Set+<3qVud%qB=T{Z2TG
zWC}CcH2`D8kyY9iaT$sS!m7X?cjwv|67Vubwj)SK1On6yRMi`|uhOzbl>YK4SU~%s
zJ64iWEIS-5c0l__ZlYvjsY(*lHjBAl($l|9-WUd(Eg=9sZ!BFjX-waS>j4I4e&y7(
z^t^hs_;alqy=1TkG4)4YK&&%kq-ufKUL=rRpQ;qC)6_TD<JB=CaC>Tzr_mIEniaaa
zfb|bTcKbB=rMAr}K0H);7?NyiWoOAIz1K9aDoT$taL+M1Y+mjt4d%-1+y`SnvHi=y
zpEUCv>+PFZa#g;}RCTS#d27aRh4y-=9d<RU^_k8gA&cN(m6V|C9-hsyU!8k8kam;I
z!b%_XZ`S1T|M0QkWd85bCr0MK@TdQ}{{PXZ|N7<N_z&jktc3}I**n}be@%Ep7EKcc
zHnUAfe-Y0(OENfxw3cMl!S@@EGnuHoW$~nb9H?93NMX>!2WL8-D-F)Y7rzS^{v{40
zCOce&?=|N>gxFi$lkaz~^9b|l7HV+9zQo7UxA&VZkMVCWN6)X-eulfP`}^^*{c!1v
zT%7&Sr@OuHv&VSG6TGfJ&TrUCaeO}RrM<k^zPNwt{f;79I3yua0nLqy7=QO-DpcZ=
zD)tpSf7~7{thYFSDAWwz@8tmk*Do2wA_Oy?xc?@BdCHjq&!qE)+zpF`kVNs6`<)%t
zKN?20o<0UL=K%sw8DB2USv?4%5`XzA^f_%L5*%xccRvW=gN`oRb<+kPekwtL@-!+w
zrXUgKXvud{F#EIIuv=T`#P>kyl2mnZZ7_!E_4v)LXg0)wIOQekib^HNj_Dc|`s~tU
zBHZ{z-Lo`(0G+~ObZ5<TkFoS5hX@=Lgw@e=_|6AO>NMh{aNHKCd?tEBrB}j?BbV$O
zW}8@Yfpo<*MAdq0xJ^1_lA&$t?%GLEB@Yb^Jm9^OvRUsjIBs`2<pI!hNx3sNd$qaO
zN0F;X@VBF$PF1m|9!rlB2Zc+%+)`6x#j;z0Os49X<XkXZI5{Oby~5C&u=hTi46C23
zJi*r0G7zyo)c}tK9Lyd9Pz-1Pj&0oVP;KqnFhjPIWY*rbx<pl%q_v*V<_XE}d<rvy
zN9En;p|yn#qH106jIJ6be_3gQH$AL}JC3>2&>I{-Pu1Z?I-*Ik8h;L3PP-$r_WB07
zn5k%4s&XAKFi%8rvyp{J%%|Ek+hR`gMot!|Sl|heP?u&wWL-6-qJ5%5RcS*N*=IU+
zw%N5=tj`X{KSfekXTL0a#6mlnV;uU+I@3%w+qk0c6A)y==Kk^r2wNGbE{dBdP@9@0
z)5%msQ#;>cX~HgO7>+fXE^VXkA$LuR$bcvi!CY-APyNhp{Xu1~(8yn-`YJyGE^5`G
zLj}W0_i&)qo-WNI1PAO|Anp@eVhh0cjVO^Dr7Xx7-Rav3M#INZoy~m&Q!Iy!At!tT
z6nm}>R~84`<%-HP%Yd(pJ<Moahk6e#mBWr0v!tAlfx98SMF@(h0H9jqh&rdQrHCG{
zKh0%5FhLFDCF4VgayA)M9D04D7UtsjmZ9kO(4%O@3*Mu!d&93VNL<17S{DxBC$Ar6
zRPF`vj=89%pIYwZj<tQJuaY%HdpsWt=8GS`A5Ak$D1Ai^%ExwGol#LFc_r0@Cm}Cw
z_=5r$9KSNK%fgb^!Cf`{ErpM}3A4`<c*@ct<rL+I^@?(AF}(&k?_>k+XwUhB)k~+_
zZ-7=t0{0$wr;kO}P3^_}lb`H$`E*|S<?+OLUBD+?t=Gj-U}QUkiMT&B*EN57J&ehz
zR``|2*=Fz5eYEpSa_AP@8b*gu8cqGDpI4Y2;eS?}oM!c<)hM`LLS2gIv^l*8C)pkv
zLesAIj<G#F1}F3GD+*Ux7|{^oqSO|f3ddgmw$7bB_`h^^HX@`C<WClPQ6Pr8RVU=0
z?M!v4yUwu);Jt1YSvG3?*m@yA1Y8GbR%`+fR(H1Vpn?@9o~<lXsHpkVb*I?yk-tAv
zpVJ7%xzn7nqfG@lA@j)U)jtQa1k=fer;S`mAn}K1BccZ{A~b8Jr!wZzmQL*O8g$mw
zA|JF{?h3Du=p5WSBey+2XN+FhqQGnen8nj(`8LxlZ3B>$MxZ90yc-_MLW;cV(I1AC
z{FDCv;}ICgzb?VK*3e4YU_<f!yJDW7qKT|Y&B@`5zqZYiBeQ@!pJZEf=)2-#P2V_K
zk^AEnM}rxyLDt#VB9nzqB7J(M?M#<Pa0Z9?q;sj-`+Oi~k8&_)#&;2i@S1?aM<!J8
zgRt$X(!~Gb=k6dZxt3p$Z78-M9(i){c@y`2soOb7a?fGwRwUoG<MJwdddgP3blLN6
z|6RTAv+VQdV?xG`udDl_*Z1|M8*PqAjW|$vxJcwCefPJIR=03_&1dzl)&^7{#Q_eo
zRKgu398fISVXE$F56B0=I}S6U0$!owWvOtAc9sNZRU?8g3A>CIHLn+AG!u&dtdH(4
z45|jmlDkjxu~u#ncc3!cQ?#Mv;76=4VIoQB58{E&`qHS%k~k^y=?H}T#FxMAvWFIR
zx6byHhj%shHWxga?k~|^Zw3VDG;EpT>sa~l*VBG5H;`apc!e;gss1Nm-k*1hI6EF^
zqO_<*cWs~k!xgXnmRGeawsr=UJpRmkY-X9p5Z9H&1hBt=156tlOTjcT5K;|E@gcn|
z+TEMGk(mUKOBJbnh>H!v`dX_neT~rgT%-6k*-4H(5fi4kMbu=$n*h}45M8DP;zuH0
zd%O4;e$q_3tn`qooz`8&T=7az3-i?}1(*VsiEG`Rk^uCq7)tM8T{ij9`?*Nk+e4ia
zkK~zy2@4ctHLXM9enO7lSE{Kbpa|3_K7qn1X`+LID#@nhQ&(!%Bj#MPN$zV~N4hw$
zH`9@b>ax_-tct~{71P0Nhl0TS0Z+CKTk%7a<~*`SHepcgFoiU-jiOBL%pdH@rBHN|
z7{^geREG{clzMo`faF&6JSuw{7cJ;zm)aGJO=%S_)~7wLPzLE>f;o|154Cb!LMs`r
z8%5`51>l;K`FTi{(@l|E)*RSFApu|~u?9*aM%>KvI#UY^-kZO=l#kN~w5t*zF@7Q1
zoC^rHR8bPC8+QdG;E|}wWST58N8lhC!qsp~>ks&Z4l4M7lIaDE{{c&9u`CJ_d(&kh
zk_!{tayX3ugB<@o$8_aHz0x&uAgJl0R%Qz(-)D@I;?d9)rCQ#dbzY*$qrk^{qX@eP
zyqroIqH4Y$m^u)hd=v)-EwOJAo@9kWT+E&+3x;)rVoEI)HDzu7q;GN1WVsPl6b`Fu
zVR5qD&ssUJhxPs2M|&;k!e!5~4=|{M5(pX&Yi+V604g=zRbef(PGZ7qGsf}=<PmBD
z1*<~A@VC@54~+_ESS+AIkU*i+Z?I&`i8cN$#unwj=ar<<5eSu7AV&_#T<cn;>&Jyw
zs0_Bp8IWkL6kB}7MG|S7YT?IU-l&CP)Tw4sdh+?%8>QNO{MzhVMX~cyNNL{(kXI>?
z(N~Q&Uo4C1eSFHZc<`0)>tcf!{5Y`G$gnq}wI}dEFcf6ySx?)0P@|^yYe84<UxW$=
zOmX-}+Y)&Et6B0u0FFD}I^P>^p0<U}opZ+DOxnipc3!*T(BI%d1lf%@6G8@hH_j&E
zh@f(xuUQSgW0GEyC-fPGOwrdRqlL&j#%q)h+PFEVgu6+njgw}54+!jjz?hw-X#ZXA
z*#3|5o&VMO#lrlrcN5#1n*Z5w`>xjRDGLHFc-kexe*E?DCJ==ys!IlR2ntM6N5G15
zk&xp5-Ee0Y%PhDmX)#efiFRgppKN4&i!Eh&Eq$8Yc`|Wun6f+&Az9RPk53_QC@UdT
zrLec(&3Fr9V8*7D{EQZ|ClNw_CoQ)6;lTl)%kSiN;m5WY{ERcUl!}Mlj^KwyGyb{T
z4lB${@ci2U#^!!~-7g)l1s_#vic?NjKm=fawvbup1ipAHB!OyE((R=^*)b9>oz-z+
z!Su#u;7IkHLJm|I;lZRD@ZUXm)IS%ZPRI8oB2lF19&m5$x0^L_Bql-@I7R%T2#DfJ
zUL27T2ytpgQ?pA%MD&c5EEoiE`8#WRFDl|A%*G2)+{N(|4hsK$&Xeiz`#5jw;r8(?
zCx*57hp8s=$dj8c@aJJ3IdxTA<e-P(x$l~q8(@eLeEJ@N)h#NB-@qwD?^_gAC+6-k
zQ|ZmP76*QFt>g5Qf!O*>7SZNg)(VO0L^ic=dB2J&q+4*J7ERK^Ye8>_G;t<oyw0p<
z+F3Ot3Q@PVCXWSfnuT0Rd>h$z7K!|5x90odJb-)SAgM7PME`DvO<gjRO@=V-<dVv@
zi0TiT77vVsK+uhE3?SZKja8D}BNdS-$o0Kc$GwSBx-G(1aDcxE;%gC1#6nyD=9QzW
zf%Ye0)&Q8Oa&;t3Z)KLWvTqnGLo0}r@FePuRbk29bkjss5evF2^>rS(ruJ4oky)kn
zFH#S+1BBcUmlOf=2Uw&^QSJ5=kLEsILrgmHxB3Ys<qq73scNZg>EY{~!zEfyx5g3=
zwP8uN8u~dg!HuNsBngh|^a2smBj(cRkbXH#`sp0sI&?F1TZVSlcoUEy;{<dG=GFTL
zCf33ozD6F|$4Zi{&J<PcjN)-BH{0l~LvhO!hmS)xbT_jMk-K>RM9&Bk7)bPaJQ5^H
z-~kq~L6_@btq}WXiMMXu-j7#E&_I?$5rznB)c566+&VO#kze(ocAaCM&X-lA=SB^*
zd1SwGu!GQFL$jWRJzhPxLEc@E4`1i?Y$L<k>DjK(H}88-`IeEr)+M^FA-e}#7c|xx
zSWg1&FLZ5cvv|I$w&IVMeT~7;0w$ijm-_n9g>bI~ZTAv_Jq&A;2BpbDS^vx?rGEG>
z1in&|4*bgN@Zm~U_9|?H(ptF&{RjfydX&xX-|qWC0#ev+moQ~k?o%C@oko6@#9MKz
z-QK)w&nQ8z$g3T{JI`#Q-$m+t8P}#>4Q=$S@?~QG_`xstn${zetmDeJB}gsUG6E#0
ziVv6j(I6>@{lnzUDGVkDs~-Z5vnD`?VI@q5k(o7x;*i-f&jtkT-I}w>ZhZ$Aj}(S+
zUH-BYePg=P4B?;dW*%u#_uEyUb@s0X6`X;EH7ynOlai*aW?WWB;B7qqy@p#DWLWD=
z(RrSD7eI6;VSrS8M?koKf17SR!TIK72(v{}Eu)<h0S!svJ%<KquMtKe73(rW?6CzH
zVtIw#@>lG2J9cEfMLdPCCO(s~T#wEv1gAm5MXnEA`f5?0nWUmxUN1@CxR3LWVGekp
z`EZAqMjU8(1z^!Cf@(gb^ys}<uf7iA*sq`%7ZYhxw0G!z6{SgbJ=#n2KGet3GL@6$
zDTKsEFfSv{a87&J!gW|zEQ#qa>eGYNHhbJsbE~RP@V48LNQo-yh&!Nr@mK}6fneIB
z**#V=jo^VIPQK2fU6rz}cO)S^2@STHQk+jm<wAvKGHt}d<{91J9aQ;Q75*10pvKIq
zE_%KV-w&6n)EL0R%FBPdqEVqqz(p&6q7;}5UJEX8C<hF7_<F(J0CJaB)+DbXVp~Bs
zd&X*8YVL)r&aMLtNoNtHyV5l}8CGjhlCBPRgM6VA4)L9GBz68t=Kpc3W&YPw?Y4&O
zf8f#Gm+E_J79)Z8g8s(byd1IHB}pX_Q%zwkNxw}b4pc~Nkd%)6c1h8bMKf!RG;@l&
zA^MOsuK2pPL5Ie&86Fn*L|HiaUP&M=%0kHIqa|dMODP{wB5KCM2v-Z^=IKXIvqQKW
z+m2N16%=PVzu%8+oY^pALJWl^#f^&B?HyeoJik7}o%nb>zJIPreSI7p?=Mg~Pj7d#
z#cF-MoD^*s^k;~~g!L#yJ8T%q;e9f%IxlFjwqU5}G4@5ICOF7-&y?lgjWgk>+LxeC
zz(vd}e3USlu)8)_N2<fL1GR(i3Q6cJCCP5IIV6DnwI`t*<CHNGhIuQ1@g4NC%BQ2y
zya~SjfhPKVzaTko77}GH#w$8M4$Z4Dv3F|fzY+gxV5;f6MJh`1_mg^}&i6RVCW-i)
zOJgYEpZ@F5Hn+OfbZDyfFyEqj+QeytZw}9ai;W2I1c_dd2wIeLI+WSplH_!wkzjrQ
zgRyg95=3d%blbMOr)^u)wykN~wr$(CZQHhOn>#ml@7;~~PQ;!cQI%Dh`99A;U93_q
zfZd0IUN|CrFexoV;bng!SO<f5AAj=3O^(5QU%q7ddV#g0ovz@m^k7N}3G-eLo=(Qo
z53?5Q*?07iFyapH1l>=!vW_Lpdy-*u0q~XtR!d?qNxeRvgL5Ns0;PQrlyfSmHq_{P
zztPo!fzY{8e0rtSedVAX?|^f<39AgFBKw<iGz|!Y^X_L!gv^}hW)JNS*Q!rOlc7&-
z0PE^Wg{3*51rF-5aT3Zfpq8fpR%>d_u9BreSwg!}FDz3?q9*+1RTIbKug@1~J@;Eq
zS5<gHxFG*v2}o50BV~ACO74`JJmz`pTYIqOPb@(r5ZeT^&cSd9ewB6h29j-S&7x#Q
z()9YGFqB@lG4Vtdwyi5hUG-Jk`qiG{Vi2S>Dz=HKY!0|~{V$F>=+D@rj`LW*DY!Jn
z_-e{%m9Fq4W$13|3z~9soe%jw@TQAtM>M{~S5x=4blP2W+BTeQ!Q25xz`*Lxv~(gI
zAhA_(F4r&`?Q4eF;fX8q!eT%W1B;CdJYSb7l-t!?N0T@QwC65IG%s2t3+d(5D67b+
zpePGKe8Sha&d>!cc<+!P2-K%Ep|3iwXJ>k#HpkKo(J3bd#CCnV7;t~Zp#fvcs|D6z
zs+e6UARG*+#VQjtFn}sDXj*4d!Eyah>l5t@jHe~%iDs*-2MkFu-|Q6`8n2UGML4&4
zGwyZmU$|L&uMU?gNoB@q5m|-7ZtcuFNmgT&S#8)9vEyeBoJbeP+9Y=7l@wSLsPCss
z$sbB@>-?m%)d_KrAs6iG!zq#_4N}eLphjN0{G(89;E%<!>RG0Ey??IuX$2@7%@3^I
zZC~*EKN+p7Z(t8MWYsNIg%i4x-?)Q))e5@^+!r7Gn;&Mt<hcO{?~7~mWq)z!2iyWe
zx!`dv?LhkpD(ruj`sH`3Z^?GLS#OMnX<L?C1iYX^f|0GjOH<{2A~WN0gBZ<{sy{+V
zW<yM_5Wr$gtO1HQz%yzB!vRlqQNeVE{(!Y(`Yrv#<;C#-3L^jejmpOKuUn!+Y7^1_
z<)xibHKi!16Hr(Th@hP(q`(3Zo|Wnl4+o^Nvm|uKaGx}6KW%x5Pe~-v93wsj`HNpA
zzwf@;ixaPf7rUKUl3h`q(UB(vLJUhR>U*+6DB<s42)!k4r44w|!x@@0&L|}Ie0e^a
z>FIWDSgQ{V^sy!{+KXWG>GW)QyLf+kQ02~s78euRhjCQHWM=ZH4qZ4OIjI4GC65w{
zr*|;Fh_O;f3)7l}NyUfXwxRf?J0p*(p<+6dRVlJmK#@4i57%5+7boFQBTYn_*(%eA
z*bybqx-sl3RLa(B7J}`z5cwqIWWlJ2M>2M}4IYp1I+w&}`M@oqa2r}rUX1hdvka%%
z<}}ZQ$18|V4EMEJRczIa51|)@hf?i`ogQMo_xw3T?;J&GXJncs)MfJtxVBr$+0pZK
z&EH<-v8f@QpE{)L2xnJd`WPXd>BsqJ>`zhWX0fdzTHMm=%werf=uh3%Q|`@Tl(jX>
z>vt`w)@&%**$T1WJn_ESc|F2`UyBSlXD&ASF}UWcY(l)mx~hj56Y1gZ03B6}2L+@>
zz7I3mWthD{PZ8Oows%YB>$ELr_g|m^mj2Hn<lFD`!sH=-w)5jveq#46)*p&r%!Dw_
zalSzAq7<$u4roS@Vr6@WPkroh(C9thzGQm`4p1{v*~lUWwd#hFW;xxK06|@m4M%C-
z)NrOihE}(f4ZZ@O_7dov56t7KYT0bvHJIZ>({1pxCCY8;g9ocNZ+jEXOhA2Po@E){
zb6O&9hO~PQ4ke0j^Qqh!m(r@B$f4P{n%<s++6nk-U5zev(A!Af$--N@ytzjEXz$Hy
zHZfJ!T4tj%@Yq;*<}9Z`DeJ>za(O^Hwz<5dvEYfCk**vRR8UwazU&&?2dC?2YGZrN
zSLbBH6EpG1WU+v_VQD6$o>^&~W-rr&)mvj()i8l;K^t`x!DY($W_-zY<aukAfCox}
zTx%Jw>1TGAyTy(dS2WQ`Aua)atR2w)$w61$2`sdyc%lRFlWL7mb~uasu0~q7C4OX?
zoAeP5Y-qtHfRcIdp%n;g7?kn%a`0Vg#CnjX{DNL4dfhRR?8m*J&FWRg`<B_D@MvUJ
z<nvKTpang~dDfE3dyTl@x0NFyfK9|I6XAO}oD;D!2Cil4cu&Td$yUi+fVBY5jncmz
z6^K^+UA%r#fnM7JRbsg0G10d<f}t77M?OK2Y&y5j!q+zx`!jRwuJ1s_XgG)y4{mCg
z#uG~(`^l6}Rblq2cr-ACDgp`!m71TD6$ZlufNxyqN^Z<$wWYvYH+%G2A>seA)H^Qk
zTZ|vWeQI;amoni%Y)~<JR=9fCvnR@F1H>It2)5=E%h-NwEU9fNtr;$PXqClD32s2o
zCS&%V=v!deNv<)PyKa0bQwlgY-x@Vmqk`^L284*zqm`^sJ1kXu&G&_r2pKbK*R#vM
zriPO)Dy)Olb|4*ArIT_|*7O`KldkY3_Es1&)kU2`*ee4UcV|ow=aw)4FCkBlxFvKI
zg>+-yexWsfaW{|{@gfxS*z51pULJwk-SRso0N0dywRf(eO&tNb1_AhGBAM~64&Ivp
z4?;lC!(~__RN*jA1D2)JpIFhTo`<P`z~tkAs+PSS#Z+@zpe;|ji%j1ZYc9O$tIh`Y
zq1u^=IfRjk>iBEmLrl6#Z&c4ZVk+`jK>7NsGe3T|nMgOL!RK}ZCD5D%1G8nWnR9AN
zPoJG7yoO)xfo0U|3h<jZg8)ota`j081-&GjO9O$k11^PN^v+x(l~M2M%zV-pXd$s5
z^~^dgg6e#Z?qF*Gli4+pNHVY^5O{oV863D$-$g>|$fZQ`h*Npr_Fakrmgn4j)!fB+
z40L%SlEES4zDT2${Mr1KlXhmFSwD>(*|ruPW|e+4W_x!SFun1%319lxi}l#jtU9^A
z>ohJ8$Ixml^vK<1Ax54l7uBM(YUT+q$8l2;P01IW=+rpLgYS?e&XhhJdwJG6vM9Z5
zdykN)Dqk;5L*xh0Jj<r`5-{4%YQ@If(oa>*q{7wk*S6E4P#3jr|G^i(^vl*~BG_P3
zFT)n=6@5dp%#w^Wr6id+@OL#xM&WJSrS&gas#psmm(*TpJQ^)SGqrO(v3epLWDoV*
z_)mH5{kG9C%G~}gdxz0o%rx{{o`a|?xr58!6i1vkcpMK?H{?GufK<Y?fDzvZol4IS
zT1j&!-)A4oJ}k8<%Y{RetkbHWNpXA;fOcA+33nY=K}?BlueBVa)5^{DP&N^~yLyjG
z`22~L-^w5N@8>$+e)R$WR2kX-VOYrccT4(T4hyGLcmC$_5WHt}@<cJC)W6Y!Sb*oL
z&on@`V%a^+g8A)CMN6sO@;8TnzM?rLnsLZR+rhSwj~33ZI}ToO&9J73u)4Thxj0<C
zNQD#Ql^k%TjpJtH;?w2I^2e+mDU6LDPYwqzm_!z6=twd4U<4X@$$Id9csaT-qT2H7
z2ny=vnY-C}c=3N2^dpBxPE<>Jb~jo^Qf_2#?=H^|#ty%1@aeq#fw2(yNl6KT(;9En
zhJP)kk_{GOA+xIUw7>#k=b8lCLkJR~ktC*qgzE*<r->whdZtKYpCnwgj{esCDcpyF
zZBXX0x-S<hps{(dkG8pIbky~Y2x88J>KA65d{C|P#KJM5K<qe$^WsIOvs8H=%jl@&
z7jp|p7H$?1R7Sh-6MrE^-BE+j6+LrFqSV*5Qmo4C8u~LTsg=PM0UvR`jP~lPsI)L+
zXyT!O#vr0l)LIkPKMX+_*c*i?1p8GWYal`qHB;fnJ0-ALy>fj<3aD1bvZq9N>af|s
z944#X<!=lL=GpG<;ubnw`AAYk+6@srK}vN8Iqvi-WM@XHWyKL>f7hU{!j9b^tR_B*
z_W-WR!+19iv{huBT4}|RCuOai3v!DFZCi0vU#<x?PVA^<&evkj1>BrbPH7o|<t<e)
z#8}~=8nm-Czl`}@^##X&?H#18d<Br2%bC5Y9%7hF%(Bsy6j@3PIZh5mJpJdH1cHcO
zV2D=JQTC#K>^i`D{gSWUh)3Vdvg8@9eIaZdbUyhM82cGn!&tcs*}M7N9=#3Pu*O*L
zDtzq_XI3LFuwLvw(Uwk9CoIIko&6et;-qqGkE4Bxx*CvU_BWLtTQD(RtdF3waI4>&
zA%v_OM&Je)SRd<c4rx1KdSASGep5eVbVo6aTJE8>B!YKC`M}}2Eud$Q;SgmbKx>BJ
zRo)wOV~fb|<2FJe&vfaVV0BJ~)@*)kAV{JxIhZ?Eovyzl;6J={OuY;7`BOA`h($v!
ztzZ321$7T-aB6HUoXiAmf$q0?(KQsPsH3@=F)7Nt`gUvF5CZNb=QOR5WJ(x;*v{p+
z4fDv*E3lMn3^;+oJR(t)mgcrDtuhVm(^j|DDzF+h%dJ5A+8w$=?kW-GkYi445Ugu1
z6w_bv{a9dn5x$pPN#uLtoQK39!m`t^&u>N$+=0NI8&_-}`VxY~Oj#E!%8Ma57OCei
zA+68BES<(Q-*qy(NPzDMiKgY+Fp^u>6RfFGMWPKvU3l0BLcO)t0e>C|^RE_POb%dH
zK5iHGG}KPo_2Pr*mRke!#N>5*J?PqM+8X@gY{95tu|Zt0z!cUsYo-25-+__)RH?<<
z|Ja-qHlN1{jd%OJct8b<gK8q+z`ca*hFx<nd&Kb})CCQ`j27O>-%2-&Vr!Y}GBfTN
zHA8y=OzwCSUjmcB5bg#!1hRpde&c1@u89?`mo1H~0+~ZdCrlqrb(7or<r%kZ{u0i;
zsl>gBCl>b=0dDkGWvkN+m*}i5o~0n5XaomC@OI-yf*fU~tIr_BVWz#A!AX_;7^2QS
zAUA%71>#&sN2w+ff4^CRX;8fX$v-yne<S5^+3n8ByxT3P`|?=&qmff#E0gFhv@807
z3YQdm$bpse!xlw0v()+yhs73S{+Vw725e}({@eSv#$QkLVp^3B%tV9G_wKV7C&w%J
z^*O=FKU}4ML(KnyoMrv1`24q5=~arf>;@k~@XZ_Q1Wnv6eZ!6iJ(4h}{I6J3cS@et
zAwJ^7#^J`>8?Nc~VIi3c=CD8^*}qN1o9>6ZLCA1yzkSzRYwlGpLKs56+WGLiCOH)h
zE5Ihi1FAc>(%xQj2DFfplM%8I@R9-{EF3&9PA;6B%p5NqFHE0RuK98*Jp?%pUcG9*
zyl_liuVS(Ek1)sGfGBx%w>LSd0`?&zd)$?g_X962n{dQ8z3e16M7HsbO7B|=+O<bY
zZCx$nBP;FC^vx4F4#YsPNLb52U;mhsneG!|mFe)WR0OjTyIgMhq<9sCbWhqjU1UQ}
zGmBZRV9wLJ5?V0gbW+_G9Rbb_A?_zmJ;WtDg{1KiJ&lp}Xbd-kGf)oeVy(vSehhGP
z8YJGJQd;l}r2S5C9$u0Z!Hgg+s;Fprc#L(-)f>fMK#vh;-So+}TNOA+^SfF+6l$6l
z9q*M_Z{igvvI@HrcC1=>eBrWqk%`#flcX+dD;?a?!(WE)Ro#iRjY&#M3Q0&NgQ&8t
z>5Pj;nNkxN6ynZ=9D~hWW+xO^$ShDevf3v3lv9KD8ZjMqR&0MQnblxyc1VuOa6!6r
zmNQauCN6Ya+&7KRXdX+cM3CW+U}f1$TyHTU<Aa~amMps@7!83hjrguImv2;;vZEI(
z!JQGK3p#OZEce#1hhGYZbn@xaayKdk*$i1d8mwYGRd5>b>7}wdbWfW6oL@1IoMIhC
zVkI`oCs|c@mMtd6UfE0K+_EM|>$(;MH6LT!>H^K676Tgu8#YHqRvr%gv7D9w-b1rL
z3$m`)ROL&1Ha`}u7aw=KY{MiW|9}%@{YOI_6WhOT@?EL@&ugF)g%N?_lO$Xq1yDTM
z5;NRdY^%%CL(Y<P`B*HCgzawjx3?%>ky!I)A^$w;03J;E;oalb-E~KTnJwz<?nhR^
z-Sl@E3crNHkTQKYOdyng8AOsDeKuH6aAveO{!gpVxaoSFzCYqLUr++a_ulQv-PjxX
z--tdartsg0{@d7I#%T(o0X4Vh!}lTACJ(2li*sa5$;}R4%#64D$8@)Dh`Z3m!ME8s
zCtWm#YTyHNsH=cuIYO_z`n0r4?)LpKb$z>*5=5s7E>WW0tR(O;b&<a*0*1~@;%H&P
zpfoyBv7jOyx-~Z5gS#<WtEgQo_17n3hU&1zog|EG6XBD7;Xo0fEM$3q!@jY)D!F}D
zmE2)}#b~%Px&W&UOMnv~Nm)0WXmRA=F1Cq`_Turl-rv2;Um0}LO(tY!uF$)7ZR8QB
zHnDnUYR;f{Z~ccusJzQE!-h?1`vNLW*vm*gYOh;YoEA1Bs&Q<@Nw|{nDEm|*l9_^4
zHK{NoOb40R58J4Hj>%rO*o1)<eMLr1j^F+x*q%3oQqm0kvc2UmAej`C$|@&DeiuD+
zsc1&9zym=MovHauy82I=la&Fg6!8ktt~C?o!*Z1m73w@XZN#esMdIq8b=0J{)rinu
zgYLa3a+_}nrxLEI$4MzJc`4h_JL#9Qic2+Les7y0KS$|WboiEFj!1mN@q-#f))vIB
z95TH?j_A`^RJM!_6#lZ-I)36SY-@yn@&Eps3J7`!YxM$oBCDHk*rk*vVSg%;6Vg)d
z+s0ENn29TpXw|VZ>qrIac~`e!w1v%H&rze%2iCTml<oVG&l(2yB^xK=gF=q=6^PG(
zZ7~vr#tUm^L?*3+0Ut1HQ}=}`T5F&Qg#~(75nfRu6Ol+<e@b1(*`_A8ZtsYqk#V|t
zdhOOl&23koI#S?Y6M*ws@VRXxR1)u{TdMTBUw-X2BU*6GXMtRk^#Y2$JsUn~vROW0
zRW7#ssg%FLVooUb`}%uE5eYU6N68H`@5z;Bk`JOlIh3;Ciki08i2O-N><~0BB8@*2
zS=4W##69nzF(pdkn_Bk5&tgM`R`cD9)!BDFDM5{nu5BW^q<xc{A!uIRI!+G64WcLD
z67z-7Cph+8C0I2{gqTWgfG}t>kS9WOiL7;7azC|}la~rSEv-{IW9GiC7k~L?*29_B
z=&5~rdG-YNpV$5sR3RmY1Y?rO=>ofZ2y_|4=6g;nCry4Bf-WD>EC<BB6^a8BM4%+R
z<gAX;zz+o^46!Q_PmD+F0kvsJwJECK_tILML<`pD+Z+1FB6*#4Py#Pn2nd&tcZT-3
zn?RW(QP=Y@5RL?3cfX$u)u*mdqt*KSPMt)Xy<%5?kMTP~y{M<X(!5JVoUP_>#`nqS
z!5nw$T@)UDmXwcFC=<EJ%Tj7;s$a^I?8}{x?olEvIcgh1&{{n$ZxKKcWNS1(;8!a{
zqW+e#ytY!)Sg~O&Y-+4fQM(Q{ej)G<spdhg!AlMlYgM?2L@l~5>?tZ(;;=vQS+YiF
zh(Sl~x(E>@S-i!@7=rM@Oi`SX-uFE|<877PvsfWX(vrb6!Gb9S>1%q(P1%Qk%U&Rh
zzb}ORJ%!Xl4BqL15>YK%r9YXKL{)PvD*7eKs>JhJI1jYunaMTax+Z?V;Y7{mtu{Pz
z;<3rd;j3baoX!ES4{P8~a=TvsEX+IT_st7zjFGR*!7jzF_Ypswt42UKNlwl%P&PP{
z=5xZ2n=Upyu?_=#tSl2?qW0Jlx`}^riUjH$=b29djD6qmP8ZohmJh&fspdbp?pT@r
z;l07Y!tk%zu2*XRd+`49cks^lYuWwg2yH6BYEcAe0MH!N%QW0S*))TkrkwbwU^M3D
zk_0dB5ZEjn$V<bVhiinbVP_gcbOkY5<|W$L1Dp)x;;%xyEYesNF(GajNsl&vjqsht
zSS)u;Nem~)i`&b`)0H(%!S5hJ_GF~t1%HN)v}cMJC)0LW`4@e^lpkn?9}-bRNY_*V
zvNUJ^0s&7E6p2PVQ-r0m(L>`Hn#^xN7bJF~Hl6^)0_ya55P3OcSsa<v@SFKgF}jG!
zH?@<Awx2?Xn-01HH-Y$7ix!VJMlpu{E5zeV1_PTxt!~RqVv<H8NW$AaU^-RxBvzxR
zqlr|7zSw#3*sn-XR?ZTngIB$ss*EVnS0+q;_j2-mX26xmi_Nlk4vXcan9x^%n$7oJ
zpJnSmbPc~(%=WyQi!|#r$H%K;W;pyKZRCMo#cYESg$U~HU|Pc=cGD2_fp!rjH7URX
zS$C{baqOl8W(OePfaIzDd%s(u$_-;(ah^aEYop1)+J#1LIxEk~1E`UqOi;Cu86CS$
zFd+P~6>A0f0+K+wKnyYx5Pwq$rbqBs29o^|MGkxdW8%Pq7D>DyxVIi;hz*({m^=0y
zB*{m#5ylt&qafzvkRl~6jqKmp<$*&{#Ja5kGI2~=as9xFeFPB?kO<UU;=AG?zOU2y
zrEFZS{PNzywe$2{pvu6v3jkv`_QSFJdlb-=3|bCSeSaqlRBjhADiv*J-_P5FKT!w-
zDqf)w#LLcwzsJj^?ksRUx^%}MlYfpZv!I^3evu05W$eRFQ6=gl^JMweGmGu*-W^1%
zS0w{yKQvV&1`3u2J+&1S+2jqM73UbY`0j3_<--~rz*#e;3g7ZTlt;vNFyj{%QFKCg
z6`M`AZM1%HAx@@5O2nUJ@<KJHx!=g_CLJVTPkBQt$(ErM0Obe16IbP<)(u57)K8Zq
z>nO9<P?md0fo9;g)v@$P7LnBUI+W6YC=;QL?*vbF=V&h`qQ}QCe>5tY$kMb~)Q_B*
z8nV^{H+EQ6j}MrJ8_Th4kP^#Y-K{oVxQ?Z3Enfe0G->N#$~(9V5a9Xry0al60*^Dh
z(Iaw%T-3KtfH@gdnLM?d>Za-{4c~adI(6<S3mvNiZn`fkys!!cU%7LLkXF?s(gzOp
z>S=Cw7VXhEf#T$x=G|>A^P$1--PIX8^|&jvyk4gFs0lzr6o0dRC#fS$e8qysJI?Ex
z<A*M-+m$+yM}g0XIAx`Xz@G3T7nhp1)WyNU_2Suay|_6jR%5rUR53k_SgAM@jIo2a
z2=8prxLFh&M8A84NTKTC=^VE3jiGMh>(09RRPyfe#ML?@-wM@{Ky8<P^Xr)NY0glv
z0Kv1Xc)?y79kwe%$Rd#aEhwYb6B-0%U?#bvX-LrvsWxCy=+Wf5119jaCFrGDzE!P*
zN;u@PxSZu#IL?39p{QSJ?q|RIxFm7{xD8XAVP}p^9Yz+5t@t#x=e`HvB}nkO4@{1n
zc;UM)mfSgPa`i9n&<WSL6MZR4Nfih{gY*v4IP{V2=FCxWqn=IRKHDsDf+L<bb5c_#
zgsltv%O);KyC$|Aw$H^P_UGynd?Il|$m&!Z%I`P+?uU%<UhfjYaDfa%U+*xwa7=V~
zB4Wi?%yLdnY@iY!G1A=+yS$O6U0<9ooCST{!|lEf?45yP;m1i8J#(UQbw*B#Ez0qY
z8g3a@FnA>r@INM>nQl-If+8vPIHPGKu^1lFVh7^5@6YPDQ;cleihLNzkfnOhK-bd$
z(q`KAWXY=aG34C)Is@r3=G+@NGi2KQZmpUi3sBDiE=_&$@T@NyIXM7<=p>SMIWap)
z31j5hL~=a|$x5U{1jgMepx^(oW;E9w(3Gg2=meUM?+zz>SI);kiruwRnOS%!xl(`&
z=l4<-Z?Qp<O7(;qKKQ_1;Ws>+b_r33J3Dv<B1o0E^a0QY03n9WkEJ}r-oWeW!?HQe
z!>;Z+Ik*{+ZRqL&RFd+g`G@Z+^FJn&{|Z$9W#?~6_5YF(h!9g`<@@tPh^3n~spkpr
zp6fMH!lg(0N&kYl!X9q84&{;$$K5P78iU{qUooZ+GHiyKTOmzv_jV24Z?Cn&2}sN)
z4|LlhgmB0)ktvnwv%qx(rpbC_eY{<!P1m`GK!`H^AY6_gOAd5OOi~g;sj#3J!<3{3
zCK?W8=WPB9!}WRguolzmVek2Lhf;EqLnl*$@6*;=504{PGZ)PnDF=7Ktm`L7ETf*Q
zCXgx>aE9B-{RaynIlR_-a(vpgRSms?lnz>J#sX(T*x+-%oz;Om{Ua%12EW>NIkB~Y
zaVfv@+SX15?O`)~(z=U&b^1~OwppTV;Gw==1SGLgyn^O6jxxUYlzA$meJnhU<Nn%^
zP^@=cjh*U?1U7ob({)bcIf*h|c!i$6llNBsMtRr(h!7lT(#yGH8q0vEPk|XxBv`Q9
zRpM7NC7}{0N-9qka7iYy`-#l8PEXliL>FwSqMK8)`duOzH&_m_JNeZ+CTt@C!&X}Y
zs)MIp%x`j-%;3LB$R$B-?E6F&0;OVq^63tOsMAG38!5=GwnI;#u-H9aBcYZzwtyU(
z=!=)<jF6@~?5TNKy!!Id2<%T=wPppFe0&EeZ=tf>cF5+333%|nl&b(P8(4)o{0Wtc
zjm@G+uTDq4n~z_Bl(}Ie>&8KV)fQyaTWJ|d6vxG;w+)Z!)%BDeOaOuHPAs<34o!{S
z5#)$kAc%!rA&o9V4%~>Goiq0vm+#A5dDFkjY{wySBh+=^L^(K4_hsydKjx$@?_Cup
zE`HzZGjo8D;oM%$tZi0C76qry+cuDi5`n&gKt*AY`HOeDoBb6;boF^+ptD<lnrw-@
zP!)BKV%wB^IVvyrO1yX%RoYTdN*k+Y;bHXjk3_#GbU^qtY%Dogo7C8QFib=I_;1kF
zN3eaw_Kb&kMCjl^kWYl&ZrY{mWlR5c6yI9^VRJ4Xum~_n@M?I1OEJb=<o*FnKd98&
z-7@a)FDmOQ#F5ydwxCYG%!|g@G`W%AI-&AdXfLhuD=%q^1(9Gi-=bcbZeYwbq6MIQ
zY+q|~(-thVy4QzoW&I|&PY>*mn{wEV#j4?<j@mh$OAulVxr)}@J_t6EePi^#oXgk+
z;G*P8$1gZ#Sl4RA<Y0ph#1*ezIp$NK|F72YL^8p?Kg8Y~&{M=Elo5)rH)L##C7=Z)
zN5#D{WbrXmjMPd-qVg9^c5T`|B!O&%HxueZp498L;k#7~(8t_8PvCGFh~x!U1(K93
zJiMJkV$?-yt{xIXj>Ac2s?A1zNkkPv`VIMoV>r$}VP6v${iyxy%WVAM@u?j=hcA~}
z#3HdXcOMIh*4r@>jV6W(o1+_?7UMH02I-1hoN`QS+EktBVAit1>ye^?O+*2S_~P_I
zRp98_Ef91d2s1w}p4ghadWc0y9n~JN{hma^;`8S-+A@Mk#C87fIl{(0Nu}30n?EeH
zQF_37astWQOQOzW&d$-wmu+wiw|4-i{l;1@YJL>wr_0OL3t)9>FRa?(CK$_lYKpcw
zAk;0Lr+cg0^`LV^HCD~%b#lc1Vz>S^G}zRXA?)1I?AawkmXdSbxBZw!8-jDj)ZU4w
zR~4eAxick6M`a!kAE)2^uw%M>gkAb0fuzrO{CgZ`{OwjmtOZgGO1$xW2do9eh_fx6
z;?<kci#D`?!BwbLivtX0jKAVZJS1KbqqoMpJENG%;>kWD-kSxhS6sH2^}KeHAF^Gn
z5LZ|U3H4wC3PM5J!z@8(?MfU}@42#w;}jfUF%wE4V_Iz4agh5hkFD=+*(rq)Y2D!(
zza-;>e@FUJ{Ynh?vcr@}@`vH9NJ(dRMNldz7kns%sNT`viSD2tfBJp+$p3r9NFp9%
zT?|f^^6ljt(@j}J_$R0_U9ba&7&K|lZfW`0Vu2e&(NngZ|1GFz$VRNmknNnuN3Nai
z_S}uRzNIk3Nsc+&{*nZe><1h&<x;v;wcVTSgu1+k?BrX4Y^E682WZ5YaO$6G$v>$b
z>DmAF6lqEIf0o0<nn(jd`8;vl<ur?&NV6KD79`WBXOWQYMDXEA)>M4FQUEC8m}%Bj
zvZQ1|#*VHB?z^@?iE<!`dLF$aJpxpFsp}RgtHnOUAcP6a$P*C5N)81HO7p+y6{s;{
zf^GVLe*hrB>~!LPZNj?^`tQ9mem}Lyd%h*~_<Vera$j0t4tmk|ZQONykGwwZCZRlG
zAQniO2JQ;j%CA1EK|&$=5r9=Ckyj3kb04N-ReXxNrB8NQz;zl=kr>IjY5ys4t5-kT
zahoXCF)Ia8O#2=pJoXF@uiL%u*MXOHHU+@-fMh9&$3_Aw0s7)HIdE*4O5y#Lp?;43
z_wd9mIdZMNf>+3XyPwdF*`K3xr5I@FC#M4x^aG&m0oeMWDfUsUyaRMGMw`zPCqXoA
z?))**JHKW=lE7E!6-vRE4%)`P;?xsPjL;wjWYf326%_66m4kVhg*SIk;%hSf%p0+8
zKt$@RD`jjwb?$58GzgKlX%)790z^c5-DSRqNsQ*w5?Zv6hW;du{|?a8Io%~4J04r=
zES%=llo=-}sf?I}K)DN|W#$(A6;M-z;3o7-r@2BpnRX@~oKy)7GO858<U!<e2fm$~
ze0&NT+F+IjhX3kE7A;wVir%dKoCZ$W2IZyu5lVm&DA0nxh(U-kR|GJP+8v*lk#Q)c
z20C6-&^qf3NKjJKP)}z2>6=_rok(|2l1NLBBP!6`s(?gj(tA!11@sHnD3m1CN0200
zVIsjYE7^Rba^1?(DdSi=C|iNpmYgfuU3;Xz`v$9Lfmnn8??;@LP7R1T**AS{!n6T8
zr?==dcr|=$*flwNNgP*{Zh8BeG+hL%{d*obJpsmorCy!@JIO8cJb$><6$4Y=Dyo0V
zuuS0kt6PWktvY>RUci`Sr6Q;8Iv)2M-y*hy-4XIRvyaKMK6S=Af8x=heoSSeSm>o@
zWurnmdvd-!W1C|Q31O7UI)iZw8TVxQ)(`TE!AfkKNHW4BKf&WyOXXe*Lp!0|nE9T%
ziR(f8*h0Tcd{V%L=Fe53N<7o`J~A58mn$dBu-_|tiZ7#fvu5UR)NjF1YS4gfv{S$s
z*GaH5kOQNiR#pNl+k|5aS;|g+1*hFo4R#2>`4;JtDqQYBa2VZ+<tdxHPYOQJl#v1r
zVN@I8DkM8@@{oLO7T(Ea?~<?{6!9+Ys+F}R&yyVLg(uRizpP!D2kx~c;*gk3OS#pH
z2!qz3Xe~dRKzLaKwBg+`vF(A&M6ES)AlFz?q_&n2pjhhiCF(ZX6qeVPhDD;67Bj}Q
zH|6F`W41L%%|uF7)8>c@b0pih@^D$sqAaX)-ERs|NY^lmi;o(eB%Am*R;w(F@$rZf
zt*}&V-gU7Yw(p$kj?L`9QEL_?+ihdvs6ju$HL}G~4_Mq8kRKrFDgBRsxQtjB{&Nn)
zz{>Qm`yxwfn*W8E^}eZ@Er1!Mri&-HS`|zuu?nfi#&Yp64e_fxOAV!Qi_aYVdg40#
zWhJSc335SyIo64jeK8UvT0G|@RIoy)Ev?<zDJG;8OL(BpoF+;bF;7gyEM721{f_cr
z_&9P{l7b$I2Cg7Q-;Y^p%V=UF<CB%?gAwH>VZbkPz+7+_^044?up<&#_xiXs@_=vm
zvUGN|!Gxpj>FLAj_A&7xX|W6Xgcr1vf({yiK_r}RZ((;S9o<GOtevaJPkfdc%^QVT
z#N2K9<lf25h9y14cGSqCyrxSnY>xqzSLr*ztWWDY9x7$TtnT-{BdRc1PAl#9l_d*&
zX!L3+6E@m0LI%C11|R>}3!G|vr<ByXeY&ocVca%!)}P4qr$DMe<)R3)uC;R*RC!vi
zT%*O<sGGb3LG|&IsmRmZkz*_WwzIryWLda7hIR}(p06v*W2Gs*FU;Aq2+K+tomu<L
zS$?WA1vKLKLZZ37w2X)WzuV{c!Qnl~(DtXNvp-`Et+tLo&VUosh;kFyiac4w1{B*$
zxk|3i3i!D+#~*9s*qvQ!Bk*kraSXcIjb;@iVEL-d^}^F~$@|!BLTwerrop|c@~C!=
zbFh+Pp*iO~yFK|L<*17a^Wv^)s#zZcLFd^BatZ+1?UTe%e^K$|*i0iX1T}<k*FP~B
z<RCUG&VD1seNoPN>-6)Qn;rK38e6*}<Vn97HcU|*%9@~y(^QNs#-q%lo)s3%8?@Vd
z`R=D1p|!sA=i<7L0UHZ53u|}xXD38<Xl^Tmxw82wNvCxHTf9{EY}A|BA;i!y8^pjc
zfC;JNuYgV|<Sh?kO+<%lTG!c=%9YXF+-jt{x_lpiTaYExvOQui6jsWIm}g&WqWqbI
zaOA0t1@Ve=zl>ykU3`7iWB@uxf0w9AH*Gd%8(A;eGfA;WhLM8{r`al@F_Q?%g{6~{
zk7E8tNF+}2Ggw;6rA<am2Ep{mBbCD>==E?R+wz0!0`?CBn8sp)>j`7)zO}9h*)KbO
z^^gUUO5HTz4e*3W<5x24?Yt90NAhdvDDS?*#;OMFy`BaMuCxZeQzxcUeCL7?YE7f%
z27XV{<gu#28VBo49IlzFdIm%JZro_Xn^d`N%OyVG$pNnib>Z3xNjtU$sdgsC#B0;U
zI8jeu@9-SVmfAg|Y!9)x5|fa?pR#+9l4}jmuh>6Kl?KBj%RjOEv-R~x>~Je{Yobnt
z)IZ<pp<lGVzKP#5=musz-c6CBB-(9;)G>l|TwcAjY%_J&0;;dB1SjYW=42E7PLbQP
zSD%u!wdG#>nmik!?Km2ERZN?~3k*phgV^rNx=EYM|J(l6u>iMVUx^9?1@;A)40<Wh
z@2`86HAZVqW$QHlvI8w8V$cr2ky37xHX#ZGKt8R=i>9mNmi6No+2bI#?Jk23P)<Ag
zk~jos;tQZ6+$*pp_@s~v&5@_Eir_mtQJF=Xb708JsNFW?_qA#>d@TV{h*fZ|KXF>k
zIk@VaOa{S$(^7ASU4@++*dMq=%WJ9-b^j--t!=ZbnQfPJButw;<YKthTC7mu;Nr8<
zm;h$$X|4=%s=cY<<6Pn4#%M|#=Zz4;LR@VKc;QE|abcb+!N8`X-5aklCtniQE<pX>
zsq9w@$DLfKNr!kry^kg!!O7>o$TxkvY&<lJ(l8k<(et3-RYH=?>&8r&id{x%qz!%@
zg$WZ6S<4=%sR4Lx6+C~UkYc8o)W8@5=glXaQYvsq!Ja-tGJxd6kY$p+w9Sl?fOWp?
zk}vx9R%ytodjh~yj9G<Oy!TKxjy&P+m*A3s?Ou|x7<Z;=KB0DA5efWmP!*XqJoNyO
z!Cijp%NCqRI6^N%V4gT_UhZXa+yb<1vm&(=RnkxPie6FJIPsd6e!5*=i~f|Xf|pxJ
zmsx%Fc}(}-c%W3eAI{Kd=w!jOXgrsx6<b)}C#dI66M(#XwGtQ|e&)My4oYujt8w**
z&0}_t>4-YUqn;Q2NaWKW*P7;zmj09}*k0I%^-6cI29x>To((c&TIN@z_*IfkuTq}=
zqUfw1&^_ZAB<pwfy?PV+>d%nK4Kj)6sL*@gW|yBJFX_bk1Q_24fYw`DHs0P-flIjG
z{pp}mLU%fI$8jG|KY3Gs5M&FdYD%v?@Aqx+ad{d3scijIz=G}HxAXtIY`y&NvQ>lB
zrU!2|XL+@3ExCD>M+K%{jGR}0coskS@x+Bfk!F9W<qNl(P5xkS{W<Ld7Zp$e0KPM$
zO=Chg4XNTMK$$OJy9q)K1rmdBOsG~JAp2LxmQE)WI52UPp@a6j>+5dHlOGgr=C`+N
zm*>Yr$i&;pQ3v`-=9>VBNa|2-K>~l!8q<IbRVm0BlC$XMrFQC7t_!~bv9jSzXJ!L?
zr+7WL`q8?1Wdos84vKmjK{!2CIHo*7NWCsUaelZOf0B%SM-tMxRS}Nfyk_efsf*%0
zIlq6Y_X@t+uFNK3KYJ%v!-1YLJocy%JDL^&KkBM5BGD#P$E>s_>T=fc`WvXR;fAw?
zPa-zg_S!Fs<!6JRs)!Kh66=KW(Qv93E9Grd!%7GUzlcbFx+sK^%j^nnI0zvrr1>|f
zVd24-rX5ml9rpdqmyB!DOw>^QIUDv2H2#Cqow}Sx&_MbbB%o?>RPU4L;cWU=bAdfe
zm{=0{3|d5!V<dV)p=@Vg2qD5Xiy2nMe3X9hTbRAXZfGPuRD4g2Th4evh<<w<Xg+A*
zMmhS#0`_PTJx<ody1rQ?KBX<=Cf?IKg-G%sLO=lu!9;F`ky~WNOMS{7f|P@jLi8uZ
z7UF|Vjd@YV2uhTPg2{I)83;(H^+e_87H9DUE0?zI^L+92O6p-q<9VBY&VE$Gp3a-W
zQIG*5*W6HP&B84IG09NOeFSD~?suGg?9m6le1g##&fvFK6>SxDhpr{*WDWT)v!}!w
z=;%aZ++&RB?4it~_nzL&rOeUV1QYKqVmB_0RZ{i3<)ng9=>?zah73y{+Y*EIrSCQ&
z3kbAHrnRB4R`fKY{GfC!VcB0A&lqzeF-TRenWkfA#kdJcfU1x57ek>Uco%xy!fP0O
zxcu(Fs84%i<`Ty->;?iEd>Nk?r?;t1e5N9|*a>_^>80j<F$E6!+q1mhBLi3}yO=DC
z3~e{AHvL<)Fr|+BjxJSKz9Jtn<$=y24gqwvB8g5u5PktR#p@FrF}EVfra+-h(AI(3
zyd+48kc1bV(-XG_yw``gv@IwD1+ro~0&)w7Xlh9d@{0g3G%qOdeepS;1%}zSJj$V8
z6o3Ry7#pd-9Rsc5n9NF?99?P6AxJYC!l28nK0w@4sgx}!J^wCO=t9;hk%;rDkZ9ZJ
zBH!5O>lI~@u*+d3AZn9zbEQEjYlmvs^iSM5VSuLL@L^h_a1$kI^w~W-Z8ZTw&qlmW
z2r*ONJT0%Y&`$Hy`IBb1kv>0dBwSGAs60>+ria))Z#KWYUcF}MH)XVOxjwuM7g%<J
zYUq2G!_^=a%|M&?z*cZBjI9IcMvYcm-=bMvZkPpVeOXhvnz*#AH=5aKykfQOsGVLW
z&8BL71vz8Vkb3FtX3S@{9BE=a4>+q9`4tiBt2OE5Ov;7aC&^+fB+u%spR(u#?}SbF
zqz1~SpWgkx><k3u=DaX*_C<e3iV;h&3zu%u@%sJrvGOdH#b=Irk_8^<h6fQ266<)v
zugOI)dJFD>QK;(td28q9X3Tr=O?i|HR81lMR4s0)(tDHDM7?ye{BB%g)4<)LW2U|q
z)L4g{p)K!Hb1b-loBVnsoGnPgqHb}i{Q@hCu7=?PjE&!`=UpV8-RuS)z0OV~)@q(H
z<~$BCjF`|d&i#@9@TE3q?r6MK{I%o?cvg8s6Cwc&OU%6yYcJ-Kr$JIdJ(DiAA}YJB
zwPGz{GbuyP%FNTcs6*SHA`c|o%LkIN5+muXsh9_%o+WrZ5SjB9&Q}Np{YULQGq7zL
z2*a3@xgR3RDB0YlFd|I7vfS|ISOplg7f8tUrmUsPGG=UGiPF@q6*_u`W8O|?gqD%}
z29OEmFT@#aJiQW41~d$nFgSyU=b}(tCe~!fOwW}E2<S_(3shZiMA69HbeX(12qUa@
zKB#URWI1uJXT8vr8Vav5n17v;1FdO%;&EbaG(K%DDh`3)2^D4sed|XqfmGnNh!5j^
zU&c`J8?w^){5$OadYc@%$*DA;k1WM~h}{S#efH6#(&v@CB4BL7{(gPWi*V_brP+~Y
z5;BKg$T?FG2#?tKDptV%T^ipm3hHW|kZD`n<gDpW{Ug7J_b<I{$I|Y3Cj(Wupn=ME
zM>cTF(a<5xNmE}@>k?`}T$FYRt26NF>Al1+sY?J(<x}k@B2NEnOk45L3jIIPJfE)@
zX(i(2#+xNe>peearySgiQQI^yMKX@RU);;Q16yW(M>A+u{X-dK{Rh+?JH!7;lN0>^
z=|xM)JF*+}2qQZms;>-5p?5K>;*BN!PzD0nlzMcQ@YWXVre(lpKuh26fAm9&XQhP$
z!ph;DFS&fYT<|wYOoJgfuW(7a1`>P=Fh=6V6E_KH;~<Dh{rJXE{&+||`w-{A%aVt2
zsU#{+l2gyE;`m@`5f8}knE1Qmym<WzW5a@*{<86sC}U&q!SZfw5Dl1UFxcQ}d$-Tb
z^!l<_hkgv*KqU8b1d(7YP5RW3u#NUa+b2&}v70@~bE%n^iej=#(JVIJZ-+?$-so6K
zV5FPZaX0T#T+ooa-p}7Gztf1)o#qSwTVfX_qAGC<Dh<FjAvaFRIH#JchIDr-S>9zp
z#NWG#2Q~R~G(w^#6{W<iWD$h5bRn@J@)05c?rVKX;@xny5!n38($>W8vAJ3~Yp({l
zvnH+B{%9oBJ897>C7&6c6-$?ywAbSQ#nKVUR@7sfW+v7twlU4QB#m3POjgNLagI%O
z(MpUm1_qg82MY0%(-SwY`CJrQ9)K{Zk|!y@2UHX~J9)|lxxpwUSvM;QuZ4Blr6!tL
z<A~l42;}Mq<9!q?;I%t%zc0_s=g908Ge%w9-@sV+?eie29VVq&3(NJ#h3O3MDuU?2
zPO~AP%rj}My|d8w^<t5})+QyR!6Q1|uBNIqpZOE45Qwe&6Lb)Wf%~fk5a2@Cvmvy*
z&ga6lhFR5&>T<=)qah4q$MqHUHu7vHN2WUS`Nl{~>E=@KG#-q@Vz9+71?Rn-t`P&T
z2c@SqRu)@3xlr-f7)KyC_o;~Ttr@4QIw7)JllTd$u)&BfT7P%x@PcIfMz!~#>Ep_i
zlNYh0-5OS@8{kZLFC4P>WAh@Qknl|kf6ww_jkD)b=adoBoJ7#X@g895Cm?8c5-m({
zk@rz`wg5ob;skIGWP&ZvY}v-jx?WLtWSvmX<e>N*vEEneGrRVr^c#NuMq)1;{z%|#
zXY$|++HS@&Ne!ZFll$`B<rCuCZ&~R*Te@m4+Fcx|*6G&-NP9=N5lK>>`_2Lx+B8)x
zzUtLh1fQ#I`CV(!w8;B$^4Yb7ywp5U_=3F&nGNutM~{W&U&@5p+5WFgn1SuzyL-Ac
z`mC|p5V~$u5l%@sQYSFu1?%a7*daH;O#KLC{F;N|!>foza4C`u3En<B)w!BgEZmBc
z;sf|-UBsq4-EiRA!u!~f1z7=wVS?2p8Ke-67;zwr*tF!J(g-k%dcHuBCHp*xcva;y
z;|gmU#wv}NN%*Ka1ZfSl7=(yT`NBfTH7M41Vc#M#K?&#bYzPIz{-7!{Ck9DRl1Fsw
z2@Y8lVK}EU87nx=VZ(y(%ne`W*%&LutLXZYDdpmQh$&CfGvo?Rn#vCth}@{r<9JhR
zk3^LcTlKL+mJj`1ugXP$);@}t>ti3X0F_8gEQi4D5>&ODcJ*J|A9YpfnCoMLK|hIl
ziIWc-4vCdRRLU)8XLT(=jU-ZvL>LxFp`5%>T)d-~@=q4PC<dt#dYv(l5rGWY%LFLB
z1A;|qikG^Tp#nuUh206-Z9`(m!Qw+V#USK_XlTb|Vlq^KEA73*^zV<QFwcBf^`eA<
zSt{$d^>tU~&O9np+V?Xja11eKC_Tm_09o2(tDKjR*D<IDyY*eerr*<qET@qIi3;>*
zEsa+PpDr(KCV#+MEBq}5N-7DWk93RRB=bd5g6b%1i-SXf2?b$Fqdw>VQG}qJ-+&SX
zZ7B(2zz`^2IaN#T=7g}CM5u@LO2jfL4w;nyke9ClT`abihmI4!n|FR&TdtF?h!J!2
zJiO@h^FY@LEo)c-lcS~cCrf!Io5g-)VXPkqFZ~@O(?3W$#DJPO)k#rRiG7nhPw~s0
zmFG7m-DFvrP7gQh)5cv&P#gBm)yj-PCx#b>>JxM%Pd4>2Uk_$$`A+1?!o}K~>o)a6
z@!8DX(9!AJV+6+%h5tL#7-li%)GRt)XI#f$0Zh2w7&R<Tk(Gid827Fvq@|_-Cm_$4
zQ<M&v5oYLnDgzG0%NH&?qJ5i3KT@B@&v&4&>h&(j!Twjp>}d~l^X)rj^M0^C)nbg5
zBv_+<omz*NzEZews@jP^+JVNk8lVF-8TEko&1fHc3hlHRRpv#(m|=LHOvaXklX_ac
zo6Zv1$_?7o1Aa=vPx<?sqccM`i(p5?`^&rX*M4F;xbiWMO0QBY-+)~mR!~}|G7V6I
z@TLEze87EJnNLP{7S`*@7!D6z3G#)NdwD9g2dxdC|K>gIe!lc3o#=j*;NiOaFF0+_
zi)76Zf>m!r8$<Ct!Q{b=I1=#?YPP=<)cR~@_&SiX*gUS}SJSM3Bo2+O#n3gz<^Z|)
zA!ldjk&a;Jr@1{nCVsEQ+d|&5ANL0dA6Z^nw49Tk<)@N|fH!^Sx`nlncYSx>8k2(;
zgH>Bz-dE;yt(fZbi#J#2+i&;zNct9h^qgw%HVhgO@P)CZm4KgL_HW<z+!?>SPIjW9
ze-Ta_guuv9yG^ynJB`Kx<?)OTK<~Y8&1=-^Ik6D$^k8W)c5O_(tF^p2JIvhWp-fdv
z&UK;+rvQ>;`zpQjyYpT5H68(S(Dfw?bp5TcQx(|D&Staa8yJXSZ->^`Vz@4c?2EYS
z?^|>6fSwS*?w)4}0Ta~q_kbU-8ocm9m9KDyLBRqlt@!oHQVza-#VH?02vLkzoo6&m
z4)~E&TY<%eeYaMzOo>UEfNSA&MjoV9gZ*1GWc!?dyIN5p{COu=yyWYQx)Ay)(47t%
zja*cNNwg>#bwDR4a42rT<SJ^2Lui*J9m6Rk<SNg~Fo;lJrC!RV6D0e20P|p4ggtZJ
z?UW@GGvc3dCi~A6u3LMS^s{ZyjWA%GRsfFXz*iVyjW<aP!1keBOzo)OFMkom=fKyN
zeePatv*4Qzpf)H7;QDvuF)zM~=ldw&2a7pdO-A11VTJ~qE}{!A>L=h0(=TM+R+w`e
zt{3Dl<MLsw*Pd7GkKY!!=M|ehJ#QH2F}v;GtC!>ySrrUj8LEFC&Anq(4KurzpzU`J
zkXd$qTgLknRC9D%ItVvs5Vo|?XWJ)y#wQPZm;_L1YA>F>M0%fBXU0Qj#3g8Rz(p!D
z-n5%FjcpWLWE_4>!}&Y1WA>|LY5|>eWXAw*#m`|N(!AxkEwt@?Scffw*1~|&w|fI<
zGR7R%QQ7yAVS8RD6)0K4)o-A+dP;V_^m4B%ZV_op2@c#$m9!rtAL0$%SS?OOO0~Yh
z#VAPKC3|@?pVviwM06=3`P25(=vOb0oL73gM^6({TVs13fSGKx+BZ)b{wdbxFm8_i
zioJBXY3m=j#FtP$JMKdYQFuC@Ad~RX4DK?9*`>blK}1^fHT>gpS^8tiM~djK)*+N)
zlh=*e{-3py_3uvne}+BS{}0yxUp!N{RCW{>_z=8rbmB$~gK=6se#d*<0-_>__La%w
zFWLctj7o_Uz=D2WY#qnX4+3u>Ivt)}uV1z7ylh=twYi+z*|<D%V_IDUux0Cs6g<Zx
z%bLoHPqiv(+h2M;gGQ($P)>chfaf%i@P~$fXnxRL9Z&ULFT2vYJuTz8@Lu+aJ;Ock
zfa@fzWN;}^15-`6hLYO^M9gs;4{DMM0SBThHljV8)<VNG=-16XiyA~k_)6wIegQct
zWUt5aVwr-~N4)i;%KFPWT<^@{;b;jaO8MRi@M#2n{jpD0PZHM9Q9c{QBR9}EdRlAa
zNhWjdGBb;ff`wh{jVInm6quaN72bjpI-HsJjK`DF_%zeTY0@<G9Jn@KlmDag_HwzJ
z{g594uopj>#y3~&zqBQYjs~RIn(P&9wy;J9{)}6^Ks~tQ2>+e!(c13OoAqo~uG<jp
zBfx++1pg+Q_Kgj^p`<Lrb8s8$9wm4(E$rr{100bZ1e@+;H&E@8v$=gzarg)ab8eq0
z&xW35HKkDW)`+Z;a<#rD3)4`3^K1nGz?)(6=!+*ig!%-)7@40(iRs3-JqQp*J)qxJ
zd49cqttLq-E{yGX5L<+C4>}%<4MemgyvbRH5b8@p$v3Zm@Y|t}rKKD^@%R^_B4T4E
zh!Ln<KxUV_kacRI1uW=R2rHG!p+T23e4>Vr3SZA(g}QLkR7eP((%ML|P_VS4H^bO;
zL%z&KC-plEOAQ-3mszU+D;X>gm|=jn(&U)3-C}QtHGrd3O5jBoWGUXo*kr_74*p6a
zaYgYOa_UUs;l^Wq`5#w>*`1=M;7Fig+nrdu^k!G!k`z@lRKClW6Jwp^C*;)qWrO$=
z%iknwhWk!h%e^pBe*cTHcWlnIjkb1U>yB+Z>9}Ltwr$(CZQHi(bdrv3JLza=?Ru;B
zhxe)4>pxsI=Q*x9#xch1-}2r%U(>d0HYQFX!Qt-+{_E@i&wD@T|2ijflrHfDgF*Gb
z`G>}#sDv^)6f5)tByXEA-W)Y`ba2!sTR)B#?m+H&T73ZknIx7XcOB3#<qAh2+P!Uq
z@JDMSi~44FPmsTlDI$sG9A_G12%8rcmr7G7&P;d=UOh@Z+kRuFUx)_)O#-X9XAI>b
z_)f&imQktE0-Q6cXA+l+f?m$bsDMZzZ}{*(k+yp|IlR0u@#FY%aZT;~csZ@$q}3nG
zKR5YY`GLxnBH7WG%$biuxh?_?1n{nE$WN#kM-!(vbkfqQEL&*LT{vOc7;GywbZRSQ
z;9$TSFq6?^phHDjj_9fNFy4s69d_}U=&3eBeV^=mD#wx?Waum-4f8S#b@eJ2y54(k
z-(IIcVff{jrXY%DC6ouDSTYh;6e2*@kX>@opJqBO_XCT2sRxG&%jqwwjx#8#q6VaH
ztQEGb_g@J6leoP_wAEa+8|fd;9RWwzld4Ch*%(>Z`?VtC*G0XS_`Q7k*zWsE!i1r~
zflAe66|_e(33QOCB}Dz@^*{g;sHKs=krfE!XBEN)?||4y$5PFJ9tX6|Q`LQgKN!ly
zT)9A_<~sL2$*fz?^{DM+`t@=7H(aSus|YmkI>&p<Eev}_*7qC5A;SXlRcHwiQ8FCN
zTC%f!jLbh7T=9bh5;t9EiF!I?B{Nt@wl^f~P68}Q2>~L98cK8aVB>P;`GDaHR)z35
z!+B-#7@t?lmn}?r?ST|*B)q?%rDB&)%we-m%{#1Wy3uP+$vmw`RxAEj5OG6eYh%Q_
zmeVGenmsOL#ylglHia@w_(sz1E{S>4N(1fEs(*_vx{z_4E>a1|n4@`POl5{P*^0*M
zc+z2#Hjj^pzge59hK{RE?*LlX&Vvd8WZ8xuju=nbu`wJ$_20w6Q~7*5KhBV|RG6c8
zQgB)D@q9{atdVNnnyH;AkMJ^WU4}>#g=LRlJwtW|KumnwqMM*+<siS2Mdu)|z1c-7
zT>UWRf*Qub|9$kPgtZE~QUvn-0yHh3)`n=IkL#ivBD)HVeM<RX2+h?N4w)BMN0#}2
z4!KI{8!Y~^tBd|Iv^#CGu*I1|zGAM_dSh|7V&vL}9m%&K6Y5sYF5K$Hz4Sr6bhK!l
z_<?G(I7X1{f2()D#qRY7?BBhS`Qt6yV(eb9j;iUhhpEar^MGefcQV>VJ(Y%=6%DYg
ziFBpayxR?)d<(w*79Hb*`(9q?TbVfM5F*)OV#TiA)dfPIGg57CqYQ}-;E)2xKm52N
zAx`xfo+NPvZon3z?LrZArJN!%L8FrLW3lAjph*!-jLNbvOPmOiJGbQYe6+rS5X8Lk
zVh^5kKeNGbA!+l~Zk**Y@*mv=Hu0)hdl6garQ>de*=`!ileD<{wU|<=X7puk#29Kx
z;zi@yG1j-A3p!ctbG}_4Tafe<9?&{QJkb4WA)iR{W4sl7fpmzWIR1~J^g~$vzg996
zBlG`yC~c`pxn;AVZa>g8XLFj#KMLlHV93Pga8R^ph0!`@)20K*xlHz>*8f`4+leB@
zw9hd0&g_&?6F)uR?S0>k%x5J&ZFymS*3DKbj31O_j@Z{4CW8qc76zk~U6_eBlDyqS
zB+ck~dwp1YnyPZcs5|C`9jEQ>b@y@ccv`}u-K=^*Osm_Es+xLvn7iA+I0-}~g9Jmt
z4>Tl6uRuQln8ExN;EV*K^N4!>-99IXOJ%{=+3JX80!o?vqL@TZl@1<r#00IV4EZh$
zf;{IVF>c*gFW0s;0LZD|J^=|+B+6r+7HDLpu)3}q?A+coac!t^ZdlwPZQCf$TrzKM
z%u=nR|M7y(M&rMmC{{NWVi8|<N-E}{J^^cNub*u5Dt_iK(Y9ojcy(wsuaHdk>!Jbs
zfIudGXW#*d!)fCvSs$UKD=-(A1_FVhL7M6zuPD@E2(^qM)up45>fOS(is^ulz4aw7
zyoycvq|1u{<i%rLr2}FVY`FYb#e(C2l5}TlZ({4kDJ01HTkv5Gc<rZI!yERvK(;_i
zLjkZ8abeuWz@p{Rt~Y6U7&ALSJIJ&j(>Q!PiS+)_Y0pa(O-DknYBDKE1k(84LV9Ka
z;v%N#hG)1Z#_0ff($%0bjS)<ge-H*(Tb<@@oD=>mK5`Dl)aH>Te8qVxTvCV*k8xFL
z_2Z3>smmEu1OxtZbs=35HI#Z4+J|X=q5~8mL2KNe*y(5)7|hj-Z<fjw^J*<irexhz
z;ihrDeVy%1luFT5VJ@{fNC_pA()sXgJ@rM`#3+Bjazw9vGXSVbk;f9uLG8mWt{bGL
zb9{8U>-|D94~C*yuw0hlK8NbLrcmOIw2m!Y6pbfAOa24Vo~C#3irp-gm!RoV*t8(y
z6LvcGOMHi*qpZ}%v6%uF2R&8}5}4e;Iav2)?AtT!m7pc~&|L0yw*3>4E#%N#?bU1Z
zzRdppvBKqL)5Ci)fAF6`h%6%Yjq}~bAvd#OqJBwm7JPs*G>pnmwp+W14N`XPvUU35
zK8k+jDiu*UQ|b4h>9e(pEt^Tg&DDaB%@&xv&76}NCXUozEF@ddtFcO=8(0Pdv%JAI
zWP1Iqb_B8bk5?q7B8z$E9{bumk=cqJL!quPssJ~3PZsHj(HR6*$l^rF=v{z4L@PJ)
z?<tCo)(_b`?TQ+gaiO<O8dEr|nNezdT(illscANw#AW4uuZ0Z>7pz>f$kzwHpe#&0
zG`KK+4~g8tXlo_7T$@}f=fd=6Bm78W$z+dG3S!Opn%TV*lJUXHSJH20s>L?9pf3<#
zZNx&etCObcOPW2ZVfA{kYwNg^q?iiwX$NJIXHWGBGDitf`M<`=e~~eMd_@0SfBerl
z`Tv>%(b5nTh!>RD{$GDQ%Ux@jjlwwd6;w)za-8*|-cErySK^B7{-6#jd|>!tcOMo4
z1&2B$hr8RotMC0Wv1x#yq&#3+4HyMD4t$Ua@px)iY5exG?G4e69!eFIeh{h<O!<-9
zY$s25X0FZ*aKvG3j_wdJbn&o+xG?uiP~83B-ocw&>c8j5<IfBoUGHx0tiJEN`%z4e
zfmxDzuu?L0<c0=RET!xQ!b0doN%g3*U#Y<_7P`ezf?6&zx~8@gfUnWoYnK&Om#;M4
zbst?%gh6fXU=6Ej^`USDMv?zH*($B3#vya<f~p}C_~>qb>ndM#Z<R2nQj$+LeMNHS
z&JhPVyaHJnl8L^8x`xw0^U4p4TOS>_(K8#BSv*XKb#E(oz}hUQXk@9;_dj0Oj{E$v
zG3i5`9b)RoG^qa-`DWEGh3L;7)V}AICRvLlaYzId1;&)i;lBo^UtfALYEzjqG^f;L
z>7;fVL%-4NyJ<?wT#Ls5`Pj~hxk)R|KqT%U3ptSw$8>DX43-}q94H4E@IO}3p<5Ku
zJ-jkfz96=FD!$B2_&a`#^30@|$e#|aIGnO>qGGo=IIv;EC7-j?MVKak;nI7WONghD
zMVd+|JwS;d-p_%opB{lJkq!1R!!|;U;G#1pj;z|UW$V;h)E>ShMkxwyY829`%X0&6
z0OODuU2;E~dFAsL1RByOAieIT$zsVd10)a53mrcdpC&)HKJXKY#@-4n3`m$V60M0>
zS}=BG-j5=QG#nJ2q$BF&AjKZ71UHIyH<OWQkX$ONTl**OEsVTveii$)l7LbKMlu`Z
ziD$jK>I03fO3<w#7lfBuRZu7GI8z9@M7z@Yrv<~U(lZ$XX#`U<qS3dn7xikE1z$X^
zE70cG6-FsmK@Qzr|Ej5$OOhlS6!ti<PW<Tn=bv1&(mvuv-2{R1+Jj<|gUmGolczQI
zF4IMtof4g;Zqb~ZM!lY2_y>1k&c<EN3YeOw0&nv3jn79$V_i*<OF~4SalEhZ`Y~98
zm-P;laFzG!b)O0HekMdpPsEqXcvXpYwhw-N4bs~z98}*6@85h>4?O8jV9>teDMsm4
z`(<fx%>w~A9JMYR%RS%IX>6cyk+XDk;h^igiuDZ0qY%(W2CdpW>_YOut19;md>Fcd
z*c|`9PgRkA9|H>olJc5Hw>b=wCdTK=5K6WI{S^D8Fhw^_{PkRyEws6IP7UI$)baLn
zwOBir$Df-Hf)m99Pfj=wOG7T=NFEc0deeycp~*-x3J^zxHCbf`OKk~0Wty?3vN_rR
z^|aKiF0kC#h}kl=u)xBI?XB$e6%H`D7B6l0&2Yd!Ub9gvdyc21y6Q(}q8-ZPCSDQe
z>!|&`gn@Izqbj!y`L|#1_vcyRKkH#(acAM-BnT`^ukDl?TYadnl|}A<F(su@=i5vm
z$zNlQnshR1&8Y*S#=RM2CH8HrBSlB_)yplpu|ABa`|7U=+bu5;R~A%ykIO~hS;u`R
z4@rH%1=d!%|1sM*|AU9e%*6b^W;XX|*x2E80KTsE8g@Db>G1W^q;_lhky%6{*=$i-
z+O3emXsm1)2rX1B@HSp9-t8xR3WOzO9eWTXs4O{fXSshp#Q)%Z1e5x-yR&sZZZj2<
zFb$b=$(pi+jUy7V3KYqjMDB^uf}xo+Dx*v5YQMhjXErcU=$6)kkIt66*craqG8{#g
zR=+*pEuW9iuihuh*7W?`=|<Fx3WH%{;By2`P3X(nf!}{69LCQEr{ucsE7eNaGPj-P
zEKOk%MC`xoUhrL4&W)Hh)ez(umn%aT%4|>g7DvvKFGVjt0*ufbei~7UWKkd<Buv!{
z%ZWL&e`yH}UDjqxhs2-BG+14V7rQ<{FIjx5*F=_P);H_^&iFU_X|qY3uw7R*Dx)E#
zI+B!So0gVk;-E7++uzNjjWyT-13!2tmlt@EZQjy$Ub@ZoyZX#SbFRwNdeY{%*0rmO
zTXI-sN7MV`D1Q>mPkJY3R3+eGrw*4g@~u{Po4w>+*>;Ou>t(@n<V&QU%H)30XFWxC
z-SgPwv%Y$z&DY}~g+4%Zz<iR5)#h>iEKByL&^pqx180BI|8OO$F$9AwFv@t+aS-0k
z-JJ=*^%<x#;(gY(qjF)zcNpw>(3`v}(^KP1hjH!Ha!r~J#mJ+N(~mV+5T3tiH?z5-
z1{A`M`!@&U!BT@&Zpo&DHq$mM>;1_k2<g(WJYCYz5uyZn(x=gS$`ES`gC5#M2YI^6
zrNQT4ceUM}9maBw5}{sPQ>i99y<IN-){ZL;`@>tUd2WIXP&7YWx`V%m8rzFjRBiS<
z*n|d5(WbPO#_#j&T`*kxD`$3lD0&REVxSBr(GJDuO;;@oMANp<aV)%DCUq_oVGARL
z!D8nV-}A%!;DOuhhG+5Q^6D4(8)Ij}p$Kf;uRZNRZ1Jf1-{>B+C{$4wH>t#+G0<wh
zTqqEd3fvy;rEbUFwP_}MJ~}(>t522D9#-Wb!3D$;gjtG5$s=2(2B+gzO1Azt=@pUo
z87_gd@SgTUm@pFUr%~@V!Z^rr6|(-#yStH$w0VxA>b}$@f@x-<;^bpgfCpH(#h!Vk
zKZK(1OFFF(1++!narSO7g&w#C0Ts9T`Xo{|1=BMBk}y_`>`uZ7(5;$%T-}@dHz;+n
zkimXcJ8g<7{}lGc{7y(}pxPoY5!Ki5l6CpY)}gLEs6~ZXZsRj^K~YL6I%?=~zx0>_
zX&RfCV-i;UeM!16o{cSA%xJ^9!u4)>gKNq`&2|)K56dFKF`msf4WSBWSqQ-d!N7wP
zFcx_T3<e4&>KgJ?jZtay+q>M=q}j!(Vj;es>BK(Q-Si>WrIiU<f6h34=MSUV^n>&c
zgW0Cr6v=h^@qFkWYL5`(&8DB6eC49geZxBKluCy!%qsd=_9WXPW>B&@6D^-det#zJ
zp`ShfPGC_3Alul+;1$MjWcgRn1FfhMU-HG)33;g^5C`?{H>g9$?|kNwgx!3QL(rH#
z4^igi1^-d|%-`5U6|!~u^8~K8j>n}C*hUn179-7ii7&#NC;~4wI4ceN_ptlEl|)XU
z8>p~0A01jmFX7RLK5wr|$~2zEryI*#SabPJFn@~lFARIktG?A`1DpM6!5mp^4gFAz
zNS8d?568+!^w~k}{UR(yfKz!~eYqWN>HlHtNS7wM%EBe7iX#!23|9NpPj5RahCSzf
zKABuSykLBL&??nn9F-3HS^2H4vBtX*`>u6v*V4*fxkoWr`K|s_w5G{2`7LJ0Pfmu7
z#KjBHVGDf&s9G{|c&+kIS%IaM+1rm>A*hJ+I$BJlM-V;~4U_$yV*@Fl3ctoq@X~&L
zxZLl<z_V7j6^HmkD2}|m@9A>4ha!XiIt!u*lz>=sIEm~^e+is6@gC*ZxtdU~<)K1|
z*L-oq^TKx?E+f|w8ffu`A`&;M!~4Q=&KK0jf=&Ct63~W+s3QO-JLDRKY8x>9H?yJ!
zgU2HF#XXXy++;QNQLEwJ0i_Hd%NAClG%IWXTVcyzBYF-<HrV>)+&mh{pjmc%dzoc@
zJc1Cq@ZpkBbjY@c`iz+yE)gcEnE5n`cqi;Je^x{>f7iI|+OwZ^kP~mh2Tf~9Y;Bz&
z?hpHfmp6~-%rRx{j%lk5#{x=g55EIsftj(AZB`RUf()CWB%vaRM?vFGKiPS=lsr>2
zCooI^&;U6nDF=1kVa}+}dJ!<2mYITG$ZL$ooQ9-l>|OdzVIi`b;-Au;phyTv=%tX@
zha*yOo2VxJ9YKL1jNi%<MW_TKA+W??1bpFf7-R(Yc%c+}%QRCJHFlloKKEC#gMX@P
zZuIV~)!0O&F%}h>P|Vs#9G%o~>$R)SG;Unru81#kqu*b+3c4U|uA(i)ip;pYsoAoi
zrP5NsCA;#UA65wYv>71qp4}j9I#-|4x2Jkm&HgZGPlCegjS|?-->Fw;MHt<%jbM%K
zHkvU3E`6|;*0rJn&C3zgrsK4;e`Vx#amHCQC7S<f71DYZk^nxuuKDWJwYkn(<gIr+
z4vGjcN?)$Kn(E4FhwvT{!g-nCL5{_Er8H<RcZDCf)SS7|sqc^GIcm{U^Q{trF6>`D
zGMsmxw7GUz_}hCxKpjO{AhvR0eM9WD4;_XO%-#5V^%UJ_EQ4T8zVQL)xxDBrEb3*;
z4!5h3!4%`UI+-%X+AhiWqqc^wCI!cfp&b9^Vs=}OphmB1h#GeeI`q9~{EwR%IK6iV
zm;Pq~4C~|ZK|H$|A%Q)GV=<8yJ;|5c4bllgglw|em(CU@oRHMRJ2%`@*xw>68wgR&
zy8%SK{OE7crg56y|NCLW&iG&daM)P>*P?6Y|717j>gKntSxi09Cc+TJyPB;OoQ2M>
zOYmSyRJDN{t^YReJ+0ozlaeizo)<fW4In0-4mtIClgb`L%j)WI=jn3$OlTA)$XaAd
z-yo-w2bMuOB}ki1i~))c`pWEbdr8$_%ArWYk@XkmyI()g=gyvH+y%k$nOJheQR8fD
z>$YXTuu%j>)$I7RzfHe=e4oCaVBi1U4|QQIt@?F)*@0>6mjT2<7GOwAHpMeB6#^TK
z1<4e-T9E5TrDI(+6inMpY}afe&!fCb!ePo-4N@_Fbu${GVgf?S$x|1&<|L&gp|27Y
zW9_X7?7zITRyCq2%%FZ9QSliB;G`y@9w?~R_L`<W_h_+Z$D+T~C#io4a4;^e+mAsr
z73@h84-7EDBDp55Zn&68lNbH7vejELDinS5)-t`=*WztL00vEsQps42EoUnA=-N=#
zJ2yOQw^obYSmb?qPN*3*Zl;Evs(tTjnCnq(ie+UEB`q#%MTO$tG5_R!0iLQc#_@S2
zB1(MXw%D=L)gdLoS+s7MtkP&eWFGl+Ytz>9&G4DNm&07DWb96eQcqV%VBtuS1IuI2
zbIt2ZQ=)==I4W|oWR3r^{;WsA%K(hRs76}ksFHZPyn80D&hKaG*ZmBDk!3@nA)^ci
z4f5=7<K#F_k>S{6dyul)cng@oiZ+j#P9B#c?TsXh-gs@x%R-q|62X5*#w>_?NEM@;
ztu(Ip+N2)Q?|91rK&}*(0JpkpP~1(31Xd^$z8{_n8;E0*Qw7+TdZ+xccf_i-rTxM0
zE2x<0brrpCOajPds`D?E)3uK%B>Qx9^3(_Ltc)C&Wp^Ot(x3njb?BxfGHC+&!6W%A
zK~}070?n?MN@a3N#l!j+SiN#XWs`dIznO84oMDmGee{=Q@N#i~=YRmyKnC}NZk~Hw
zJ*Qkfw>i4)vU~yC3(5@^Z>E}exr4YQwK@<U9?dCkSkcfm@$uZ751V0n<Y2@lP!{VX
z{D_M$^<^;kPzn<~3T%uTJ+)wxkq@AIX-us)3qqsj4)q6YmGq>%Hxyu;C2%U@FrMYo
z2V6@kZ7#hHdi1Hz%5<wy)wI26%bT)$3{#d$W-3>>woz?u1+b&lfB#(UyL#awq43Lx
zP_9~O?>5+E>mc3|4v9y&O9xQo3l54S*}Kk($on+fAoPvG0OKaNGL6-u1qcEH{86ZE
z!M72h4oP{a;#)!J3Nyr)=)w3E_|utruE7j;D}ZZ179yzq3vR*E9MA(JNrK`M<-sL5
z3M7a8Bhzb<u0J4chJQ)F3->eHMCihgCrf^YzlnK+tw44f@pvLH;nl-^Bf4ND=RA8A
zze0l<wHF!S&WUrte8vqD7nroNK9a7}0z2y82_e9|@>d#iBrvPFFCu?82Msf}M$C~A
z8_>JCejN&)#Jy066&GRos{^skBzy}XUGzd-kVDk%Iy3kI>BvNY&LgPFRQP=TN|ur)
zOs~&JvN={98rN0g$EJG^Gj;OUI8@2m!Z{4oN+6^*6y+#j39f1>4AEQQ5C``qD)^pt
zeRp6+dg9iiElkiaMII8*=ALL)M^8jmCX^%`!R>E=6sQqZHGEhHE!BMTr}Ad$N6q!(
z&*v_A`MaHueii2x%3OEfA|Rt726ksdDDU4DVJz>40KRsGye_g*ECt#&V@3hji~Y0<
z%~4HXdF9yEu=CshklY+6=)@SCpYqk~pp|uOPTlK85)j&MSYV2H`D|Fussg&D3-Fo0
zf;^Z4e&Y{>JWp`Hv2H6Cw*=TWG;QND=8iOz+rvmBF7uw2H&!+2#R}3C<T+wbvQS`G
zWbGkO1SbQvD)*2up-kgOWbDQRn^<5B0voDY>%xa50WFqs9rEJJ_;6)*5AnYP^G4@j
zBJ)g1z#L#C$qU#kOSaH)8yTtr9ZxA|IY=@L;1J(5Ejxl<uN|dUuG8B-JdJiN4gUM$
z1lvV^^l=kYp{tTmSbp2YbGa*@JjpOw@zyX+JL!BMO1l<4x93&9x+46o)<-wDYDaPW
zO8Jq0;Hu}fwbDi1zo!Ds+rA%!v7;n1e@vTw?*?9b5@Cuw#uFzH<>aNC4eK}^X|O7l
z961)lB>*Iopuc#gYW%>&anYRRvOdG*fC%x=ejV7Rm)^Bkve^}{Vx<4egQw>eqa#TP
z_P=)5e;q~HIRDq8dRar_rw9qP_g(Flvf$YG`6wPlVjdHE9ZcqrZeqT^^|}F}LL}`*
zg`ZE~9vr&`>I7XpF$SUpPp{7#zD!e9w5b_|oEb%2?yqxZvFu~6EFDupGf7Lql1WpH
zwXO(4>3EA;mK<)c&y6LFx?Jg)?CH{JS=~O5yTetBS+pGP?}O#0wW!sf<;6Fx%j-!$
z68}G98Z;oF;Fg7gr81sLu%e%AMx7*EW3YJ26n5U0HzmMN$J%yfgnM&3O>Lxe-CHww
z<H&^L_kNRM1Lfe+k&I7oMJg*TQnZXIoy0r$rpoGGH|37f<i$@*6W!{bVM}){+p^6W
zT3BjNtui30X`!W8!<%+BK}kUj|EY1d%2P?#WqtKXWq}da?XG6j22Qr^HY;`8j`K&#
z(uN7O9}IZNl@3bVMp^25XNw`Ms`%hDa2Ay9owsuD(5^1~SV^k1p-KyDNP}q%PGLtH
z3jE39AHyP;$1?2Vw_RCQIXDC3qe60ARqQQ`oGtArb-qVirA;7B=Gfp;1;CRPFAhhf
zLlN0)KMQ7c31D_)qE;<;ThcxUPucrE6_owsK#(DJ&rE&o<knkBI1uTmwT)877pICi
zrY`SHiv4DFLp+9f_w4{I+3O7E3VYg1P+Ar7>Rrkp8icF5rsJ0<#1lSQaNq#K6i1(!
zkYoWvMxEVMdq3d8g-%tt1Hv%}BqjvV{gD~m1PF<S2)XrQ1xE^I&wQ0M^lX%hL-o2Q
z-qcXap-lGMF2`Gj1p4RAfBr4yNZ@|4>N^AJ=4Kr9l_DA^8x_>mgoz`{J)!V6L+6%b
zW(UmK<#^v*GmZlhqv_(qDw)wlLbOIQTT=V_%Y^kkBRzD9_9R+`$qlDKog^GGP4Vq<
zsCr4Al*GgQC=glIS=_PBrkKB?abw-!NAaa9wjrkgRLZGRh1GaxHz7s8H8GY^W=po_
zACQBkx{e@1DZJL?*WW5n3ZYN%Fj5socP$U0`274ba=bY$P=Xw!D8oZag|$+07-YA7
z^4Gw9z0WQPdfsY@bd0qDYtt7V&(DW1{)D3c7z4P<#J$N9{*s5ML=X;bDH2ibE&NoL
zTR#y1{yFc~fL9sU;~C@>`AV*kk!f4(aKyy!Z^Kq7Is}}|FGZNxB4M`&SPA2E77Dw6
zaUNO6(yx_VACMfhq<qK@`uK}jkI{K<(CLKck=<h>KqMOStNU68-#!?NSQ4{j|00#a
zND080PAwYSGsDDy5M`Wdb6mWh`H9*f#t>f^`)to>)-pA|Drij2_hI+En9!}pLtsQV
zrc!vHhu`B1fadE|?N%@{`<w4jhw|3YE6uvel(%_h;B{`LnVrDI$dVUm&*I4oO0qB)
zF+c(l9=mg>J7KP}FDGCo7t7PG*?V4_r`B5a5YLRWjywbLdh-!Df#uciWA1`3nF=mF
z8{CJ}-_#;cJKDxm^IBuW6Nc7k*M0ULtO7femyt{D20!Vz=#V#L7*RT$X#i+u5_T`e
z4_%N_2n2PE<Y$S8BlE3Bhu&WB$5W(exhXespGY-tJ~&O(Ygz@43|Sd~`PoGJOyLJ?
z1$nG}{8`y7#>HVJvf!B(eNf1MU4)}w8t`XSk*F0H>DwEP=v5u)+sAoE;o*AsQBa;T
zbVIsp^K;PRAL!eVG%q92oOuH0S|b;Qh&>$XbRZbp4=;|@2K2Ps88DPXbRqki7e+LN
zi{W`}g}v|5$7uCQ#pP4x(|8Yal0ba6EVODhff8?R6^~Ut`0$rucctz6mq)A5D)R0G
zl#fRaA*tNkwvs0yIrZcFmv1`r-164IE`;&{6|M7Nx^M?D4KM~cA^w@x2eT94RSpF$
zK1pVqztI{(X+6KYd=xh(Tv#|JL2v?@o5qa)z|p4lY*#A=)L_R}w9@_^@4$YqeoV^K
z`em8BVIF57H_qvh9cReuNbPHOYZ|-n0{9|(AaC{#=v=v@kysau#{4iaof7Zb8He=$
zt`ROZ{|jnh;<=P>TfFEiA3D?e4wjr4-XhwbFkBj_weN7#NV$ywR=fw$wq29u8fT(T
z!2@~aB!k#!x^|ymVu8Wok9jBc#*73$R|+Fz#iKS8L<#D=%@-8!0;Og#`2Z@_pY&5$
z`bH_|U>WkolxJc-41+||>%}N%1Zf(y_0#%wb;69oB!WZrt3#s(n?GO*kl*j~f;ZEu
z3xQfO6oe#=V%le6)@Q*89=q`3v1Ax|tp|M-!TEx?bb9UhAFJ*skMVz<+L@V{{txT;
z|39SBqp|9UBZlsCq87c=-#CB0$)j~qQ`nqHn>Mwbvg^~JO^t-Bs3e>Z%qXsRvt%HM
z=qYZFr^^xGW{Xh9VD6wYzyCDxur|S7S4=6XXjvPzW<um#P)x2|Y3Qo(!3?;6TwQzF
z={9V>RA3r_GSlC)(DinE&Rd&mD^sdH6^>u3SaiOBK3YRu`p#h<B{$>y__*KM>zkX~
z?(qQF>V1A3M6G4z`Ffv#WzUx{@aI${6%%?*!Z(y4fY`BhdFt(`1T1jtu3+q$_Gr{q
zRe6ZrW^e#Pj)$=|q(exnjRTgJy<0N<Ug_2;Wd<k#sZ11FFoxDvhhCNDLZzPAlEfY6
z;l}~QKT?9sM(CDyE)Zu+-h1{!uz8P11(UYv=x{^a`d>psu!ENUIobn(x*Z=~DZ27P
z*thjIlaM3;`g`HrE}WV6R!NxZ>=hb)A{}vUYHh-fVwSI2Ki)E8(S%4!CfIbu(B*!r
zuDQLRk0>O35&b3CS%uMkIHQYpJLiMFm@ui60npu?8q!SGkO-Mk_q&e}#haR=Exd7B
ztlWqqABjMi(u72xNUD3Iy#)hACOsbfAXp29a)+(^-y}{vbTJK;$Da?LH0m-8bIJo=
z$`VYlzC8T-0pcLA1?o}TJa(las1FlU{YpIcltt-b*1b&@7HiTo9hv7~6;6Z1dBtay
zbFk7*60=U?^%GK<!4MhJf0lOPNs&U}<^|ZUPbRSh+1tym7gTGA?L_XesNa(j1=p<w
z`D=^x&}zr|%~$-<r$>xxa#8msQ7h`9p`nW`<?~2kSI^076Q*8V%eQC7P2cJ^Hy1j~
z7AIN$a9b)&g;JpC=pC3^UuzYNNnIX2*?9~`kXy?aBHbd3qN=A5xheM5<#D^HwI?!m
zF-HRl2Y7giMNeYu-YBkvgjq?GxdlfSk`#z@gD#^(PfBWaXNU1H9^uIXBQWSaD8~vd
zf9+bIPN-+2JYAZqRhK>E5*jqZ+_I+ZLtY%unKY0`tXJZj-;_YsL$B{$dv2K<8~EFn
z0!0Y!4-6NKr;-NBSKqZcyenAF>3D6~x~h9X5~Yz5#uy^R-VJF9qvi3#!1+VxF=<Y>
zno_2cE&aMePQaJ9584%y9z1(AmLE*qd;Th?(KjpzycX7GGrp{kOaS<T#o~ZBXUdd-
zQlaz1n?3I*Rf38@nA2V7`xgW*&MVj15ZT+_@h$<`tO;(GN*haB1t=P5kJrkFjnfsU
zx@`h4a!>TyH*HT}sEU=ZzP(ekpQhc8v(SpGQD)}ax*vc0hm=_c_YnGl2<pS;87YbV
zHN~LM)6w!o`yq-(V`d=Gqs5wC;KfE9lQs79fF|^}WXvTm;Jp<_+-^8A5Eq<??ml>k
z$4-NH`;Y_rud+s+-gW$GPd9tq^PxK<j*s3p_Y?(hdv$GV2ZQQNw8tzv1Q??SK*^8V
zFIR;Pz2)FnMaUSI7~T!ik{&jE+V5g-&w8zt;aAyEc$PO?u0*dwhk<n}aHAX*L#mUK
zp03Vui<wk1%(a8=8*xLT>Y-4SJL_Huw#2YQ=CxOs=2W0!y^P2)mJ4;kw>=%3vKcDz
zpqKY5%_&2aj3qK83)ZMs$r~S-@0s7gl;lhk5RKQ+vHY;H@=Pqe4xFL^YIMZz<0p=0
zXbAl1XvwFqwC$v&>9UqP#Vv82wv#)gJbh1rQri8yi@?rf-b`^$C)}>jQIigu>?G&4
zxo!{RJKD3KNDud0M#SLpx<CN}yi7Bs5}96_H-R8OjA8^XMebuk5sqa!eCIy`94I>d
z!qIt7inh_1k5n?!PU<sOWSXIRc)3~}X2mS+soCZndkIcaAf;)h?)DPRZZ?^zmAmLR
zop&GYLY(tBKp-j+Vxh8Vv<u&i-r_nJonBLv9I+Zq)Ehpw)_Si);w}z02yw8G)9s65
z!%zHUn-(6ssCG?!5`>GY!h~XtNRgnn5`R{d`ze(1Aq&z!nr<)CcFP#TnvyC1z4Wgw
zWKhh@JvNx5Cn(cC?GlJJH59knewY*ngo<rga@M(aS-$NxZeKqW)LQ8$z&#)rbxm!7
zt!<CYyBiRx8x0sp11emiWrbznD`oyYJJtQkw;&iE-)V;iS{@2o-PswTy6(LVdLIM@
zXD8jqzdSb8V4tYp3BCZ0ND3$jMWr4gE0*{`9eLtONv#(UF7fr0eg#1gc3X=b&BaAg
zL<oBl_zf6~g+7egqwBhPs6uI5${<FQ%%JoIibg{diMwpjpz50s)MmKM!OH`OyQzm3
z!3u^N^_l-1^WUrVc2ji-sthlE=3p;B*v$X(X6No*WwG@VSNW~MIrj>0rYXOhO@i&z
z1S^$X{Zmxw4Q2g0zMJNt_9yE-$VX`?vh$~5dEDu1UKqsETrV&}ZW^f`t{Sz&es4Bo
zA_94+`ig;6%6(`Ev}kfDq?kV4zJ7E#hV)?Cjfn>P%+S~?;?-zawr1lQeDNY-tI{Uv
zbtA#1NJdp_s2_Cy*O6&?eYK0kS9$?(mo^OOtkvCVV$&le2ERRPCb4!wMnL4-)=SZE
zn-`@3UhxE3%_*1ww4Y8nFJ%RM5?In?tibiyoS^kd2340#+v6XzR%`mzb4t_4b4thB
zvQ+T}uDmD@@DQ#71NNyD9?(&5LTttjQBP#YtSkQ#o^<cIW2;=jZ~Z6F;*p;*%V%W8
z@x5D$e&O@Ug48?fxbQi0y!xiYfbyaL$;(*5F670h4!2t$=$63;Vd+GQ&5Itu105PC
zrgx0Ew~zcegAl)}7aS!fE%`_#!cB_LfA~x>dmMTH4-fzRH0i${C9MDVqlAf#jq`tN
z*qO=DvO^I=9ozY&YPD<CM0D!omWP%Xh|rWDs8I%>X1Ie^acuu#lg*Ud`M$i&VQLL3
z@VU2qxy*3Na`J5|od^piXMm8S2ojsaa)}_3b;^;ZM=3xdWBoa2Cu!WfQbM5^v#4ew
z<n{c!I-8iHAR!4$0yGV&)yd_4aA7-%gw1`szgauppC4r{bWNNUohUZ5QcIRIsxwk5
z>zeXQq15xHHpZ_kO2zeXL}NM$RWI%3P>>jeW?8W1_2!isjHtA>Q!;SwQCCl;LsbT=
zf%9mS8ks`7(Nb!bE_omwMYjoekS433_sXeNIs5iXGTe_`eo$*o8?NV&qr0ZwWGjd?
zjd-%GBqOl%h3pu%?4KjdX^^rdbBq-Pkxz^1Y6>L_<JK13yL5gvYrA8N*nc+MrXc2_
zw}xb;<)LF-&@zbJPvx4|^KH^*Q`XuxJ&oOR>-{60T{Lm8hvr+#%JN8^4IJk}WWO1l
zqva_|W>{2QCS&O@306U%G(-}yKo!`8C9u6$X`N*=DZCz>WBD7%5rb(Yq@1QtS&&cn
zM+k*zEP~PHZd1fCeAzrFEYFaObWc<TIZINI91hJ8>wy>|qCRDOE<uGbYDzK?{esl!
z5|s)*av4zcC-y!--%2FUa=9YUrgSRzjOS)e_NGL+2k6;aK{z~>4DPJK_O(5k^@>q_
zkYk(m6zOuRR7((mRXvzr5dTpH2zm%jLEyML3U_WyD~OFrq@+$v{rem*L529po@fW%
zb>VH6wy;jxe1`3|&?$(N&6!^-s9}NHa|jhlI+{Enl{Vr#LXY)}ev5HTRBzj~iluw7
zf@pCUZ#G<?zS{{F@U4j5gzuBbaZnHQT=VC*qH$af@f|DKnhRK3HW{{VveqrjaR}jN
zYDlT82T@Y$U#6aL8-1<Sd6C}8N&!6kQT#<3^*)tI%^4YIsvv_&KK2ZWjnqXQFUlAU
z%!W#~(`JR>3Ek*tb=l6z%*!*w^rg?qn+;Oi+bQGQm;6lEX=-b~3~}mtSQ|$N9{2ZH
z@?YHCPp1J6j#}ZmLJXPt&phtA-B=&^^;Joj*YDKi<tW3iYiDWTd&5x!b!Bl%a9vgc
z!<Gf0)`V2%iy|j?y4NDs?;sl&m^B5y)1dk@%|>hnWEm;dY<qhzd2}17L2z#?_9)jd
zs%-6iuz?Z`D{stQa#Gb3wRXp6E}iOfG+aewISYZa#LrA;;fAhhN_bJ(LhY7$mYC6{
z;+lpP7opD=1(^gu%d2m+U0~|DX}vn8Q_G%Kz3KMj4Vy`|H*GWaf?D=74ISmoI_>C#
z5QgZ3Xl$N?PZ5uZ)PJ8>;jO(S@@8kNzF0X>uOZf-)dE_h*s<525!v=^Wi~<@xi`CS
z6j^E8P@8Im&k=%`Kn6io=1%tZd&SnEwEq$5xF&4W?aotYP(VCd5M{B1ASp*s^K2kA
z!ab;k-v~9*#rEwmgEp^a;m|I0uedeFV8u%F)=*wv2o~cw&@pMtuzUQrpVJ?HWvlsF
zBYvkWe3wvx*M-7QASVfwM1Ap?IK<R`MNXST(SuMB$CV7T4iee}mK&w+HIbHuA{IDt
zB48{+JPj)qU2z=PFa;-N;U?1rWLV5n$)`m%wh^4epb;X!6im2h+m&NT#|I_+wr{~j
zy{-CxOug0*W`zkh4A-#WxGb%?=YgjxNe65fRNM?oAMLVgICt*9=FCY6f#7cZcYCeR
z!{@`P<D>W*^7lucrw>p?SPnT?M})rxj{{6p{Zijddqh(h&hBZ+j4}{)WRa+(DIV}#
z{%R0P0;iN}32)uU|7mLe22vOa99hVCLusu;Seuk%z{K4EtJpTeG0$d<60)Je)}iQO
zj=27_2BnsO7x_R}*Y=4hL)O}hA4qgw>Cq6f9Mgh$u!n#Bk0XVc&z5uH_Bd-sMH!ZR
zg@^1Dl+Oymz&c8-Mxh<ATfhYY&wW}%z|Bhm?>D+TAK2-iRRG6_N$-+3B*@N@&W#oT
zZqbqt71BlHd*ZfF=+5J7@1y?`O{+KP0N#gS_&}LeeahYu{qNUS-Bx3<<#!BhYUpo~
z`%H?~pmW{EFM9foa(s`Lkn^wFm%o1grQnJ0V@=#TA!bqCsQ);&lBYlxysWRgmFQo8
zeml%}No|Wl?0;2b5@T&@!2DCkd8IV5e<fc@PY$Ws@kO~gFlggv<*G<&h|&HwQ28|U
z{DP(T@cxg2Hg=}}!~${tpQPFUztld;&~jQAMfKnLq<N$)HE3eLIx5I<Q2Z^EC7~7e
zNH*)v2&JVIPpOadapCn$Of&#9CA;rjDu=vYwB@wn#7oD~8EOA`e7d}QdPHTGAgU=3
zfF}o_C=7xRaz{WMoGMKV-#qNDrj>zRWTF~?DS#-cXRP+&elu}!UOZU@8ijg(3h%yN
z9=_^)*WnY9^7_8q9h$NKJAWP8SiOFJJa}IXYU}I!)<@?U)Fn}ePF1Z>+%+xf)u1+@
zJy(iIMz<X~z)!lnV$4EdD67b_%{U_QlLFASkWbHyOlxB@!Jz!lP1Lqfge#j%LDF0%
z*)f>r-E@Q_%lPbdpIL`~>1>1<5AE%B|Av+zEEO0ONyh!{nN!w2wa(Kgz`^@{Q0J=i
z`5J#TBC8Yt1p;H94k|Vb7D5w&j6k0;kZg8#ZkiC(z#XBsI%-K@oqB?~<a`OCt6{SB
z_mG^Rrg5-OQF>G9+Y<SA&}L7|B*KnKmDTrPp_nnZEXMUo4GzKV2@kSd)1Rq27<O;F
z>IiWeH)sR{*(V1}=jp58NR<)|_UsRa88wYs1Or}fSjW0>A8EQJx{oX5XpL2*$PUCs
zyrj?zmN@h$;m`gKSO9TIk#H%qNHC12@DxW0I#OH(jgTicq$V(c_RM{uO{C;LZ#kTp
z;9@2<ShXtG*8=R6XDp>KWjC!s-{MfG(5w`EM;w4?TdvRMh_H>c!4ZTi(I0m|lO1kr
ziFtswGQ!1CQ!NT~BOqu$MV<|Mqa1TfD{?h0(77V{jj9<gbi?vME2yxSP~1_8gClWK
zE#Dc~OE4p6Aq;6AJ%&Rl`lhOQU72ev*=;$}G92y5UMPVjWh{D2r%bhur5L_IUO5;@
z2$@m*<D@A+;YB9QTx$|MeeeCxrfWjdzyhC`$q~^mVL2*bEdDW2v}kRiY>oqaXt%Bu
zgTJS}sqj~QHaXmN=|Wf$AE-`(c|1@YQw-5qu~s6)5Oiih$u7TY2Ut?P6#tv>s*HPA
z4x(nxmCJk_K&27@osJl0G)lQuA-lt&ZSA*o?JfjF9$$lsPz!PIxWTX!8ol8*<VGzu
zg9>U6v}}@wiUMKu8h~4*%payUg9}<|ai)&GkUL*LE<09O9XZF{p1g*`f_e&3st{=-
zm~I)Y!7fRHM|4FiY+f&NhykoQ<ORBKZhiG;&A;Wcv>2No%sh;ytTzrM>fOk>DI->8
zn6Q-=o-}j3<eYk3RY1Y$ZVC-o8`>WTW0}mKb#1vVTbA5f`kF@b2rD}Chm|lqw2L74
z;;^87L9GADm?OvZZF$7ld`q==AzOPWRu<d475wxGU-6_yAZT>V2DBQyCeVTaajZnP
z>Rzlulf>Q1?JnIK>$%ognNN>>UFT-L_G0(Qv4cdX1TZq0RLgzYbqs+XFCECEO~X+a
zyduzEg16f`*=4^4aw~rhu&=`1O!V2pSrS}~!0Ysh&he6DMyGe*2nB0(+%^~YFlUe?
z&7W8;9?UKWY+r}F#A~AsYVu<1pu{AbQ>dAb+Z@8kKrQj4C|60w392RVC+Q>RU&syx
z)nqZw)lN8#ad0O8K{AH2lc_dTRk*nUYYj6}&4ab$vPme@gi6s-UU(zI#cur6C8nE1
z%I5CpPbYz=%M(mg*i8k6OZG1&mw{M!Kn|Vu*Ri7@d#rkmAeh|`%|jrxjKN%A<X61Q
zADS3Kz{oL&g?*x2AJmI#?R2w#6A3S2MV91-HE2oyGcWe5xZ^TZ<#ubX;wx936;B6a
zgxUP}FBu7!kP4aiH_|QqJHWIw40{{}R!VF+riue^aG0s$gR`bKJVw9FFLu*F6_nIF
zmCQ22OF?T=`ljx5lcS3w;+%982~ag2<haH^*>x*snq7J571K8|ar#$tZ~{(UBi-lg
zxRK&zZx!Gy2<zw`d~$W6_>1kecmjj=sdIL!Qy8|x1Vt+;EO9xBlhkLdEP_u(>gncZ
z!8iRlRe$(!>4)g*|7Zl%72H`aX;vt8YkZ&6MGMRY)EQP<vNE0vy)ph}fXDlLl9ei_
z4>Aos#pHSGjKI=Bx~oH7;x%M`T#-8L6o%&kj3EBm<RrPazV*+$<psxW5Vi<+P|TdW
zy)`zlAoG29l}+xWo+n7w9JD|pduN({DFmZK!?T{k#GyP-1&5gl%0jshd)8D)?%G$w
z_;G+&ZkH3G6}1b;cvso3tb+fzqVlhuW@pk$<&bd8^&j}{89o|$^_Etd;8bVz<J%G4
zf4cpp@F!%h#b46|m|3Y`8o%fNzW1A}GVZR@?XOf<fkN+={`KYdaY6O~Y2ZO4?986R
zQ@`4HFLPCGvA9D7oPS^BA>G|1RD0r1@b(qqJn;MNbtVhM@C8oMcw2qImIw&^`VXb}
zUrl{X%<TW`I`A!Zsh>+NQt!q8Tx$1>&33%Zfho#CA!}mDEg2l-%h_yD;wzL@4cq#9
zo3wjX3rR&4yiX%&99Vw+j<Y*Pk+MZe?Z)cKjMdF0c8UNk>yRyFkC=~0#7<Y9c;tFx
zME3_FuUNGH__x2RW_Ch>LNc}ly?!^+)s&=1PNin;{qOzl;%sf??R{eGY$L5q(V3MR
zLK>496BV1`mthL5ah|kB;*U-&)6LEn?*P7Vew~>RPQOdj;LZDFoeLZZ>u#@%#f=B=
z3MVc{)<rwpBbHbIal=IIfpx_75Q~Ll-JYqtMvn@smdjzxPpJZD$S_pOF3xPYTZ+2P
zzQ6ql(>0VMRjlaOl%jV)q6{kJBM;}8tyPUO`Lc7xt4-eV%QSn@XvnO&7XkEYZ3ds5
z*RJv;BcQHtM+xhSLy&4sZF$qzO5zBZWWzQiH1~6az9$ZB^%~#in`g`%`@HQUW3s)d
zZ+?0F%9?u^<eQ>g?$hkz17F5ko{_U4oRoi;X&@8Bv~B|Pp%FI_kjj$2@wqdunv=?d
zvzQ`HhP-FMoqQBu?XBpp<@;0X7JT(e#f|^4ti$1>$!$87U*71Pqr*HJ)yUjkIMSHv
zJ+AhaMytp|Q~Q~dc#YN$L1f%MxcFj&1)5g{B56PpprvyxI`yfLMlLW}D)M;Oh)#O4
zRo<5Jr-CzHaYay$$3U%LBdH7HliEFNt8f(qeAE>uc|A#s{i^|)ZHq;VUzU$8*tz}M
z>ktb%@q#Kt;JlWLtLHfZCM|-f0m-W2Z+ke>0uMGTMDhR^eu*AoPIMr0S`%22%3{K#
z-M?(dLrz-`vLb0vC;0A<KW0%24GN0eJQDDrQmKkJ;T5~wTB$DTs?5v{BV-}TvORlA
zTjxt;2af4lIQ(NVpgSP=p6%G7PC{n0gy}yaN-x0dqB@%=d;ighC3|m)_?^B<ms_Yj
zsLirB72-gd{q??@89GmhScsl;<=s?~l)LlY6pbdW)dWPJrU-uDAP}wUnoSYaZnBYn
zxtUUc|Lk?N)~2jx(EfRRFE>!p4}ZN;tiSv(V?ez<lekNK2zs_-3UvS(Ab&zSt`l7g
z%pmHjL##?X(B2lMaDTKZ-`qg*H$)zmxAN&Z-QS>TcIi5?W|B4hwAcV1FlNp9QqjLN
zoB>WPa6H>41{^OPLV5nMafQmfiZ>GV#l-PDR09_|;FE|2u_K-F6H~F@&1jmW79=zc
z>_HH(1rJL=t22_{LUzJf_D{<-=MhtLOA9+io_)l^kP;6mIDN_whv&YJdn8h|XGdCm
zu;u*tQVU*k1hk%I#}n@hc4$wLg~ObIjnJ#r$(tw?UgNVZ9cK2cedf+#Bt=gGVV)+W
ze9_?b0krxIWH{W|Gwcq(#L(s*o9vsua7MH!49v6!e@kFf#10m}OD2}acJ#P+GLp)S
zLj`8@f`}c^4G0K&%;~LgWL9RR*K%f7nyWpCw$LC*h@jS__`uj%MV4wtmX4+rgcqJ=
z7FL88vPODRVh;c*sP#q_o;eEFDm4pi#Ffyu-G8@yS~Ie!(D9iD6^NAA^}}ojMP%V5
z`liB9Xek;9EsFh(pC)>grdFsCH%-cA>Iuq)E_<#MKc;-Fp1gb6HPeEe54bJuwSX(A
zQ>e{7;Tx=ZiTtI^G%7F@-mm^6HD3*e%HO2geHTlg2#lBPWd3WK{s*y*iJ9|%&D$PP
zm->MLApMW;MO;P(EQ+1W#te7WHfN$8PwR~AEFkqX{TFJ*xq;Z%lMZse_@q2pTgI_4
zl=kJphMk+h!l{#Rk)G^!><`@!24BMvO=;nkcb_^@5I}9rk}FDJWGm_yyVKJFX6dEa
zAHc|u7Bl@;XVw?TU@Hop#Byw$EG}}871i`IK|lf&KDX`H{<_xV<K^n*a-Et+P&+rK
z)~B<x0wG)eQts2+(ToI)L_=0ZN4{-5$kH(Cc>)E+%z^JC?XQTPr`F`rGF>|R=d&}r
z`o2=TQyl_WNxAHfHimFNk&=am3W?5zCAT2vPDqtzkB4k#uG@KTS!nyR-aUi$X3Mn`
zGa%w~;j+%<D;-^<<2=vAovjKTr>@L#PJ*>-YO2|+qVq)#RnNJX)P=F-UA6VPN*q%P
zY;^U9%iZs6zk8=*guMSoEW$vscD51}qxFOv%8@q{WgYs(1(xv=JL0CRuj;Z-J9~XW
zkVk2fv<VpjGi4#(sy){z!YdJFxC92obJ9ZIU+v`EA$b!yePs6(pK8lGV@YG8>ubAX
z{Og<CO_jXL@W6Ftsz!YW9DAo^%6mVxtxENQt!Eeo-`zN#ERvm800et59vA-3HX9Cf
zC266UNRcGoCu5GVIGoC`+Gd1TL1L^(n8>Ja7aQb5IzCw$r+E$va&Oi{aS=13QMWkc
zJwUVy`EG6lehl}|mBRSc(;}VLPW?2_j<8WAawIxLLT*fZEWuLuX=6aW>_;NN^ISC!
z%DZJeZzFMPN)^N%Nq!LPr8NCl%lP~aP_1;n7M<Zs6HD>W?!oF`H+LG(?KEp^@@RPI
zM)VPA7*mhqeOa4Bawgbm@|uo9UsK7#9&119R1ZlCV7T{4X!GClSh3AamC&})KaF0M
zi35~A(Srzz`Ey|VLx9MruO?YRb>g}KN0joo7cc~blNl(tf|MDcNmr>Fq9_9^<0&_=
z-s8fFA(ePAVDr02$M_A0{k*+|@{6~j5rfZYMAMVcP?gXZgYKm#a|qqIlVxl}@fC6W
zF#eQ<%l{J@4muaI5r(Pe3l$+6NqBNAZ{I~5QIT#4!Ch_9P5p`Xf0cGVFDDKgh+Net
zJy}_3aXpl}R-Snx*W&SM-VQ4rMbXW)E&V@~y+e?uZ4<3q?6Pg!wr$(CZQHhO+qSFA
zwyiGpsV`2%K7-wZ{|w%UXY>v;?_9aomA)$1<&P;Wb-DJwmJ6t90EG2s(2-Sf{q&pA
zoAlmQ50Lb9%9sQPzp>}Vyq2kc#iW4f1{R710*vF-g?^4qr3<S-jD1piVf$rpC-#Gh
z-B68lg%Y@oT>&^b`~u$07`U<R;(FPLQrgf3<$750hK`{%hyXzXH`)VkO8>S&zP2Gr
z2cQ4t5U>jKFcjfFlTQ2Tz8RbhDFl*U6!zR(Y%V&?O7U6MoV3?cS<YJ)<^4>Lg?j^M
zDHSmoju@N3fDo#0*1f+v9iRNM5%D&9EezKML65@_3-RPkf5Sv9nfck-lefXM;#2iC
zf3wds;@GQq_rafDuJyxmw#8x@nZuO5SGwRuch-5h8sX80Zy{*5ir(LW2+jmvGoW9&
zGRJ6-ZB(_nBuqaZ1~0tyYCAiCBEbq$+%zVl3gSH^t3LSa%q0X-**Xn%Xa;d@`N!&E
z_R6qr#vgJpYL(sqZgA??ouOI0Fy@`Vr>t(-Y7s~>z$?yI@M}oVjFS4o=YszHZc!%L
zd>!iuEbHuF)`?<5$xpc?YxK!@zGszrY?c{x5$x+e3{$#q<xcdxPZ(LN<d%Pw*q^Ih
z|Cir_f#rV}fo3!|V}DAl&+ZQ!I(is73@w2=6Ub%*nPy&Qq5WD&2}D<`i>g1Bt(SPm
z*C(u)Hd(sn+NSX=iP(~viQ9=CEsK;5LdLiI&DQb$fJOeluwE)j(q*t&1Tsc}AZe1v
zgW4nnI3woNPgrl~^Wie9jnA78lSyJ}wNI}%^YiKC;ri@$VvbH13YRh=S-y5Ybk(4G
z8nmXmP+~7a^gLSY+<}JIG}1ijPn_?gCq_L#@-CYuS`}9Nqo9rtwa}0I!Ap!JCwdg4
z<Q(^w;DDBszkG259euxi(ub{M=1X+y+_f#2_7fO0r#9;PdXcZT_6z0n*tTKwt;^>a
zeeet$X%^NN2_$nN7~^Edo9lq`i=75_sY=AeUMachA{F6%Yu9JmZ1Z>L8(H4RX<;>q
z{}_LMFq3XZAQQZfFQD+qA%YI-@p=5<lECCF!TE<b9!cfAus<)DY6gPEvY)?;%85^<
zZP9Mg6uFjd)Oo9ViVn%&D)DTQ8?wk@hC5(YKZvJ1-j-^oCvHRl()k*Epxe`t=Vy6%
zMuB2w1NdxBys*>njoK|Z;a}h;<w2zSRv2KC>GTuE795F0!>&$}%I#FRd$Cmpdv^;0
zv5YD8;tO90XHT4Zn^HFe)*xZcxWY)F7j)XpNaj4Iq6kAJukM<&cv49p2mr;y6=>1m
zA;!mn{5Xx2n6_L3!oia8jJ1H&>57Q5Tm%*qtAe!Q{qV|yAL7!)2FfcybBC1ala>)5
z?R3HS{YYB)%@TU5jg-ayr2Wo=<OtQ3CmZ|xNex9=dLNy-z)|ibZz{t+pP~Xl7@dV!
zcZIx$b;3Abp6)Ko2Jo2lq5fh<$q`EaXgqsx@`pOhlms}CCc%Ksip(c0k-%X|pQ&SC
zq8E8MHo34k9x-_=R|2wPJVnALGH{R8sS}-A#*GV)u@D+k5wWEBh?uW^5jOB5Kr$cX
zXPNRy*fgYI$O_wh=-ZMY8HD*Hu5;}|8YzWS=f~#N84Kr-N@MRz4HG_v5RFB|h<NBt
z-pahYpB`D8%!bhK+$ZTKQ89kgd{@?s4N{vadV?TBGESLX8QrpG{T*ByI9LW4^&k~W
z9}o~Z1-vjkc|Me_89qoO9j>7!q4pJBcx}eFihvlPC-D$Lu@=&>Iy3E5*(8hWm0QSJ
zG^7@B8)VhAZ~~)DvdBTe<J5}I00ffJu#n)6qK6;2>>?hI_}r852^BZ1a0Nwi`AF<0
zjD+Pkm87V+0W?WQ>GW$Z)2Lk@H$R4*TYC@I?V}cdzvuJT^zHYOp1wEAh`O69vp%an
z*ATFNIyLOpw(<S2`aMAegW+@i^reBlwUcn)^h()jyP8#ae+FSy$EM#Kw>Ci`e!6Y(
zRpx9Z^jPEhua3|UVm4QFLf(Trs8Uv6=PJBeH{S1On|rK}AS45s<d0ZaqP!=!Q4gKu
zQ8hd(v-9)!C;W}AT;Os|n>Zv3zG)b-r$IUE_r9YB2kumxIb`{p<Fac*8g|v_3Y~2F
zpZlJjzFocDT1QgS^a-T%xJBxX2;vp2j$vKr?SayTg$TVnDNJVMm?+hM6Z6<WSVL~t
z>U|#K8>nA^t!Fv~EXp82EV+2n8BpEeHH^qg4D$Jrb-R+l>a;vyYa^Vg&ggn^10<^0
zXuWI5YeWcgPSs^wL!^?nIL_?6A)TqFr8)(bs@f)Yw|6@pTROFBs2cA~Z+n`HVs3M5
zJ|3}?@Tjatbvk%KqHtZ6U(~pw31ra_-nwwmS+O%mG=+GVF@3{b=QPWMGy>sf5Tisx
z2o(d!m`!2%Z4qdyd~pjy(?@&jP&~UCvP#1de^Zz$I0siH1f<`3#+*DyFMUOzECnI}
zUgBknjjnYmIUwHE=tFyRZ(p;!e?!xwpEVUC1I`9JRE~mL%jS6~n$<Q%1kLA72SYG~
z$4Rl`Jq$TOygw?^19_v{Z^NeYj1{6_Z0^k0k@<h$XS?*Vvm-U+6lg|7q+6tI38+OJ
znt0a&uUB+rZ9BfTLCQ4m@04;I0N9Pp=A6WtRhws9B<F&4-&^yLaGsXdH2|=*Yh>dx
z5ybv=CdeGUlYkvF1G@&1+_X+~aJmv!L`n>I{ywa%<CCWS1`U}B`}40}^shBL3)BC+
zkof(7p8NG|mX?I>e6Sqs72Bmr5?ry1QcL&qh&B$Qh1v;@N5&67#D#t^^2u04Jx?HU
z8dtub7k`PpGuHmuYLePg5*k{z$S({D6Xu?X^k|SYa3=(r(}i7zu_V9JzP@(+wWs3$
zGFtKqQ{CU!S99loj)@C^#-L>o2^_pVd|z1qSP6nh%<cHH{~h^yzqtC!TDA4M-`~DG
z&*IzZY5OHxF9wp@-~3A|QA;X<g#<kcu`-@@QG+{+9%Z-e4`Yw6sjN;08dG11xaWK@
zhst#2V7S7(5OZNDJ5QQyEaFBIFjr;AhNF*H%cPspMY^D3y67Dds=Fj<#4vj;szyp`
z*^PnZc4s%0Kq$LRz!U>xW*{y$!lf@QOd*hTb+y&fv2~YpRduu4p=!$}0DT*v^KVxs
z0?wDd>88TzDzI1ktX;Dabk75r!QcBh5H+1p+<+;L0pv{k6HBPdy6*mNEbY(Ro=V4;
zgS&@Ign2Yut=iRXQyna(zlDOi<iI1K)yrA*R#%DYRr46zk+|p4yWA`o+~$n$v~AUH
zWLjQ;%5sK{bO)dzh6F+22OQFx(<vCl8wMJb%+|r8-`<WiAVLPgqsrhH#q#N=5mrCd
z$>T4=;gN8=<k&ov7>IRkrX~%#Y7v}C2cENuBtp1SkK^I7)U+zJ#2MYP);r4=yUl%d
zciVrTQY4#nK6II=ZmB}mm0<TY68LwNqy%Y<lNVOjbSv3)0II*aG)dqI${)M7@;}QU
z4Voka%t`R5o@$GLtO)|6TFK^fo^Nv66~Cvm^B=^__YviDENZ)?!<EC++(!#0;1_zz
z<L?&WAEq738paRc#Ce2N${k`GND~$M-%1oUn|SXTgai}HA}=c-k@)j_Ex5&G2mAU%
zgBhpRQyd^;ODC8vUm9yipG$);&Z{EE6h9S;xY^|IXWrHrrCSVbZ?`Cc<d@rmjHlay
zj%$ihjt0Sp_~V~;T*HTB@%9%D(VL&i@-)PuHzlBx<BzgBHEwU4OG7woQN~Wz9izM3
z2g+P9+y(H&LaPwGmC}Gt>{Guu9hNEhKm>qo3L@@7<3Aasd!CmfAP882`u1`jp%oPD
z3Nh0Yro-hh-N69<burXwgyX(m{NkeY*bRvcHW6=#@WWLC+do$Hk<x&(<z%G<KuY+Q
zQCMPSv1`}DhggC+ln<5U`p%_w+;!ci#Eu$rjt6uooBDLJ)N1lN9anTt67}rqecfd2
zqgk}yIQu!I6wx$|Ri0<tR#jczEwdVEz^eW_rxvx0Rm)qWgC^L{2E*`cHKi8;v_>iY
zt&r4bqM&v0_ckM!Pj!Nzx^*rHB+vDxvU+LNrHAJRv{mu80-FBE-9w4rI?e}THc@Yv
zzKM(mqVaGT<-M5IeVy*Nl>XUeKsgdQHmPPLzs~dPsWE&ayaU_rqFq+fc@ylU0Sg#q
zh=A~UgMkNIdJ-P2+{D*<yrjJC#AxU}V^A^mXK~xzp-fK2`63Shce+pINF^V_Q}CvM
z!)+Qr%$BVVzvQXSK)h?`8fa755`B6lIkeHD(are?k!M%Malv<}V@j<-q<?IpJQv^u
zk$ZAnLoT>_+@q`JT8tvxh@+h^LfA=Km;?3Q2tBc^4J=C(t_Gjvx%FSiq3`~g*S7t#
zt!6i|QP`JI11o@BUt0a1T9>sy!1~p@pF+F^i>|(BlFDv6UlhCpzUX~(roTOt{ANGZ
z1a+Aaqj^TypTI{aqCTk9A1$RRywfmI9b?O~fsvEQ(dG4jId;!%byz%-AR%9o=RNUy
z0jRf9G^w)JZM~RkO}|Kavta^8(=*2vy?t8zMS97O-hKaTnY+6cZNW-U#q>-xyqgEM
zWiLcNPi`BA<>gi4b+14f4)>|bg{i9+L*L&bG%j;uif}rpD$f>Ji6-v834+)PKMoO*
zrT!v&v&FIctGiSH3%_4;`Hk0JyObMyJh|tNXZ_Cz(+YtGk8T8}(AF|TN_l|=#UJ1l
z@vX+mD4RcjD`#Hn=1qC>azDWi5CjnZ(NCEFG21Y*F#OMv-)k+cANdH?_q9%c_sAk-
z&v!kC*O|+OEk|~<=J$lq;`ux)j#ZHs@rdlh<tw(Xus?{1WI~R#nI5$iP=M3dT@G+K
zX&qALr~6CY#Cj8ysLnz7k~Ue=rtq1tgiN{e#1W|n2VyXVBd=;^EUn+i{>SO=mpAW8
zWGwCX+wsT5qn=*}_g|-nr6wqfPYz@fWps7IJV8cmC($X9s#P<zSdwUjGUTzc(ceZ!
zw&6{)db@tS9vr60f$UXRo|&_d4z{~AwQWnshDzZ>>EhJmMWD0NNeoNDU5=BxQ4HA*
zcF_c`WOvj#dM20GdT`9FS%|~zS}{H%C^^}2EZ|SL-)0C-^nx7FWe`G_l9%DZw`5N4
z?|-~mrcTxkvkLFZBXMc$?te$Ox@mN9zYS*9WJf~4PcWZ&HE8X?o2Y;L0~ayY128C}
zGA?}yGLl5N#AMX1oFBZ`(inUX%!Ld@ve8A-QN?)&*=?IkGAe#|Y0U2YzOYR@wko)h
z<zr4+2u7JwYZ{>@#71yD?GBRe-{TOCsE^)fmmqG5V1w?#(n9<-81Fdpy3m1)FbI{X
zu1{`A@H)w`{-M&NNW&@I7g7SUFwKmj3=cqxR}vAj2Lf8sKm$t9ypUr~QAjP<tx(|m
zTp0*%MEL$&Ndkzo<=F~d014DFmgl@LW|WyoyJE$&a-1`x2>3AB9`dZ8x|oqF)Ypx6
z`iwVccCkuB6@S3Avz_+uF})QsT=-sLb~s)ngcB+#s8~7&JaYMyAP#-plB{yR$?Sr!
zGJUWG((G%Y=b$GnNWi`VI7XE;2}?=B5#Z4g*UY=Jfj+o^O1p0iP66&8K3vy%ATIN8
zoaRwjOnoq#h9J>~Ac*k@qJcvsY7@-VNc?E2z=}=;(QF`tHF$pn1b+ddqh2}}r@=ym
z<W>ZtWqINVBL<>ia*%|Bxu-XBV!2>Jprey3bmR+SUJ`5vd9^#7%LRv($6L56>n~2N
z*`{kxPaVgG_)ido`90o0ho{rAm`cl^6n<H76SfBC@n#nHjhej5yVi}*%$v@|FV~Jd
zaesdENAugqfRc2=TcOAS7CmfFrgYtlSTmxRc=SH24DGlY<APCn4ZRrS?jEaPN;pC6
z8$)r|Z<m8#YGk+v0So|Shy??rjugU4wAoZ+RaW9}=;U8ecGlC+sy*d7&~&ViysKzS
zeSm8MZLqJZx1Gz|6v5h1OXuJem(;3$tRbLn+yMxVZgpZ3G?QVT_9s4;$zZhCwJSf1
zuQ_Fdj0+a;)m!0P5wuEPs1@Q~Qbn_d${1J^)J`(%3=haImwvR`{Vl+eH}}T@a9>a=
z4d8H|GIAJm3bvRli}ECcbgtJ7&5&NOie~*A6#<Mep8b-sA%6`<Y9!?&g0+ZS{JjpB
z)YN8%(iC9-W4t8`?DAf1&&})kOd@5wJC<ut;Fsk-5;~0tolkn(qVC}SkkhyfYjh}o
z!Whoh&+lmRA>9{!Aqb5i5H;gp8$eYU0utmg<Zo`VJ4^*HzTOFUC%0*Pr>yOz%ODu<
z=7+L020p_)kjP!)-p0@YN`N`g=Xd5vZ^zIAtos#6gnt(_uPIDX+0bQkTa`Tu3;;X@
zlsGcbghI@_g)&SMg^5xQa{Sp|ZEFKYI7$vippuWF%ulP2y(ANa|JB;v+a-1Sh?DhB
z-=~?1Q<DWvm`c}}mFQ`6Fi>T{Q_I(N5HkV2Knu`w<~kXV=_IVE^Xo0H=VK>VQ{XW?
zRA60U6;f~PWFG#hivv|i*Z?88HJ(e7bc<rV-G^crCB6)mAIy75f_l$ha&#RjtGuY#
zd6?c<uY1};v4lFL2E++f*9L+iiE>`AZYj&6b<I@;(v8PNA_JHN09i@)w&~%q3hr-t
zGsMPxAUcS=Z9q^%WSzzKOU9_;5UDo1Va)a^6X_$V@{rkgC4E|mk3TTs@pDd}B@+z+
zz%m7SD(K|gUR~{m#=Ucr$UY2u-4^`&q~!Oc=btK&Vz6&bmg`;Tl&_wr->qj>`da#R
zocr}n^^EVl&g(;ghXjZfn9Vh-z`M@#rZq(K`@dO`t@C3mF}pnHC7bKlKpGB%2U&rt
zHp#PBW}j+Cu^$-KJ2MsU%E@pN<eUBRs-MYiY{j=?apa>HrT@lwpzm!>41h{XQ<Mj^
z2$e9q&Cn_>Imj|acjSm$cn>);w{<QR?0r|lAC8aTJx!3s-<t-}RC0>QpwpAW#}0Vq
z8#ijAdJi2O!6;$*2)W>SA@O>7Er~Mx!Xz)h3GGKedDi_N)kh45h*gen<X;sfu(h<j
zSE2rn*I<Vxq~>`Y%>%70|9YFkU>=_hz|SWA{PE(i2ke_QbiWSP6_MIMSParthX>CR
zAc8S&t_TN@AG`2$-7CWlFJ+sr0OLVHY7Nuy?IPLA^C`ewl2saf4Fi%U=y~!fT;Gyu
z>MeZBZX&%RUC;X#ZtTiA2Q@2yeGQxA9M`dmn%X@SDFkb53(4Y$W>Uopy(B4J{QT@W
zBCYQCiqo<gCu1MX$+%EpTqy^a&@5f}U~wicc|`)|LWRa!lfL7G^vGc&1lXPPQ-$VO
z4r-KUyCd@!&G*4WnmJ4DkAk4a`kao-we`-&Q+Pq=d1cSb!O~(0%Wg`moRj5LI5!4L
zintBV$-|u7c*>oXhor>rBT9(W{A=5901^Q)kJgqnI320NgA)Ta>ScfaL{6PEbBqQ@
zZlwcVn=`YF0!PS{@(#NE>~O^#kEk?SQ62Zz8}Ji2@S2B)k%QufxRgB&I3NQgjV-Ii
zE8~0&7JF460jB;pum7DYXrEU1eOT6z*==--CQ`s+yV7R2kRQ7_%V+4A<sj^Rmj8^f
za4fku5~g6oZ+s4X=)Vu|Trfn2L1Rn@fI3)Px&#E)+be&)wlpmfPy?cCabH}nCdr0-
z)=Ka%3C8%l&Mat^%n5^xMgSW{{xktkn#c-t%xhZ4x4h|F0ak7t+KRUX8@(T`aF$;-
zoVvmDJMk0fZq4^>l}0@_#G5ux&b_6+E@5@6f)}uNYTDjL-aZ@(exA9sj<1|szaz+v
z-Dj=+A@J4B!#glKFMWF=$eJ0HV_oNwUF#Nx-ojOO3yH@1`%1cc{<AD_@1mBwRXg|O
z>hWchntFXbztVZMJ*woM&`utEx06_fQ|+EQzpRF@@!-o5YG`-b*0E=JqH{&~x{nk(
zgR6x)N;*~hy<``;`koYrlRHg!s8!*_OQ4RSn<fM7vuJQTguI0diUf?%B^9qS_62+@
z?NtRnxbr(~zxqYeS_EeE$o9}<(`7%PYX{VAwvgDf3;ni-od&K;hg$mSRHgp-FP!-U
z3+TUIi2qU?80r7#wBfayjNN~CXMXG&D=7F3M|WNPKeX`<*udtN6y3#q^(iYaKpzdw
z&2T>-)3jwz3WbO(WF8bw`!vj1tl4gYT}DSCbdKvDoY$S`V*FT6n9S(9i20w#VTQ7?
zw+OEZ;(OlM?M^hi6Al$EJ7_E|X4c_u>^FK|_G?(aI;pu|IrG%4T<i|q-wUVNrlMCo
zIp62My*!=lTrY7&ad@#~W_^C`<$#{l!olL&@KNHL%i==-EFeNPOStwbM@o38_Gjop
z#nmwvFH?=`+sIKG3N@1xeDntfWC?y{6(c-hEPhNQn5RNB>Zy+berM7Hv&+Y|zmeV{
z%oQ9tLQmnN5O#<&D}H%%3!O!oFi=>5;j6GttcLh$YJW6z!XVAGzHgacuY7(!e|amW
zM0NOiQCDr&>r@+)1^UB6zFu2n4>~pt!q1l1XRk^@yO^URK25o~g8urVK>DyDN6yj^
zAQ@Ku^Le3#07V)PMyu=8MrH_eGan9WZ|QNBdJ94g@MzK|87hq?ZEChjVkw2t-S0F}
z^0Od&dsd4wjG<ySeiTvR3%6E9PRYxpK7n&7fI;=-r5HS_td3O4_-MNvnXK!iz>Dn2
zqdJ}Xu;eM%lMXL$^(F<&<V#(u$eXLGOs4^MZMg$<D{NWwnht%ZPFtcYP`ejPA*{k$
z<ju=_880hskh>{Ef14k_Dr>QOwoI8@MZc!rEiSo!t^d`K)sRht<w%DM$=4te^lb6g
zQBztlc~_c=c|Qa(5dEYQhMYqDd(u!K8d|j|n2mYyR~$}p!?P;41fv8AZUj!#`;~Z+
zBs>=DiG(nT4J5-sb+Nt)nq57dG!DoLDPS+)3g$K7kn66$PU*l^UmFwsHOE(T8Sw`H
zDfy{5!6heIzmxHn;;1ksTvswfnEWWNIEELx2|&bW=CC8REtrB5flKswVtBhK<1?R>
zpR*bE50=wBTEeA#gA3#J#&dgWm0h5?QQ2IKv}vboE?Rp(@Ai%!#eJKXwUkLwPga~z
zwDhR(wQSXj!SH(8L8{ofeB(&f&$>b2@$_X4o}d>4Lu8<ZL2?OQB80MPMW%7d(t>|a
zXSkV9BqJ7Co$*7vGpF9#`;v*mwXJQ)NqS8xE&&ym#hgeGi!<K{I;slU@YJ<Xk+{9K
zoN1KGmQuxOU);z|wotTuQWQK$=zL&75!^WKu2js&pfRjF`y_42-l$fJ4($%1oStGK
z$}8mkt_d02ucsK5h+SGS&O?4qfFh1S{4|aHG&T|u7Adansnwpf#=cd6-HsVvwm76I
zBDjf%pB;-E+!|X@ll>5dXrGG_S#vvDCxkm<E7Ar~W6Ct_8bbpoKMQ7ubRn~+b3Ggw
z7!_b#D|~w}_CQ3YG&sEG7r|0nFLnWUN1bR-*2$NnR<vQV`&(rFxhuuuXF<Ahl*)Xj
z%(g=|>LV6)$zRgvpjVyOk^Z=y2}}s(f<RFhEx(#Y<5v>RlQAzi-CeJ#g>ApzUeJ61
zE@iMl<w^yqWh33{$4d_t;1x2i7_)Yr)f3!tk$P&Rl<;E*AFn(rCli!(ZP{BeuH+!m
zmKfwulLtdmk-NSp)}qURdVqxRsA&Hxga34bGyl)7wHfp&+e6m7+`4%MQKL|}ghctL
zMv<oe9ZV3fb<~h6`dCd3M-q?6w|7_$BCX{O*-TqkFN6^S4!hYPZcPzxPrp&J!>-Gw
zs3~n>LKDR^;?CZf<cLISGfugp%d9h{?E3ZFRXkU+LOG`aB#ggX)odv~>3-P|1P6s1
z8(HhxmbiUXosJQGX`7zwX)@a!&}eB9k&dZY7T6dA>~jD_al*ySO#Q-zFiaTnI>76f
zU5&uh61a4?kur8#*;{M2>#0zi1R*1>*%9p*rzSD}n>Z4EJ$+Q3pH{4`G@epPnes*O
z?#(;3X#AqX+Z;gF>$`ve?rvL?0UGmiiS#Ly#?1hVlv0O<r3muI_m*9)E^+bS4o3HN
z78Y=@zGc^KHa)xJrjN=Fh`WGOo<Dq3>KXr{0BHNR)}EynvFl^24cI?83{}L3oMNVG
zt)Vt5f0)&)?CiPoBWRhq-y`%pRu(bh#hfbA6&>x79Qqa5@vuP|&~(&lEtO;PR)@z*
zZB}BKwgA=kspqD5R2V^pXf+6#kS~T>AJ3ZGLY5)Us@tgI0SYb;5C)*Qtev#;ArTb=
zNUe}j1o?sjqeM&=Fk)+alLv>%DG#N?LH_U+dpQnx(KBLP%pSloF4|Bj?hmFuM>a<c
zB|UqMH~!cdQd|~zZ(z2U^#K}Dr<C;BGh?qQIa{5dzf;P5J9s}ntjj<~+WiFr37X71
zc%Af3n62jsY0{^g%3Bt$*X!zz9l%|r$^ECWc{TS7031aO<N=t0%a|eu^BL`DR6@UD
zXjxqBfOlZnR$=%O7&*iP@}%)A032Jm%E-nK@F0SI$(c<`3_=J2bU30ZTv!iV(5fH2
z;OC*CHN`goAL$k*R`8w(o6-_1%{t*UtmuRTYkGi(qv>h2!uObd+7x5m90LO83HJ+r
zwUBtf3kv>y1cvpm`g|dr@-GX!rG4;&giKl>Yx5b+FbgHAc)^f$qzn!eW?jZWcGsGI
z6@$!CIL9w&o@9&&`0E%qY_G>&eD6OVl&OX71G=ve#)=(SK@X{%8L~!Uh+`&ZoG76V
zEz5E(A?N`VEbML^4m3s~k-$K-c!y3ZGsIwl3CQ9vrzH>eHf&MO8IW>%q$ZC#Ug0X%
z>pe(ZVXToGOjhtqY2&wkZh0M_D5|v5-JNx2ClZ2RXm`wt^tdwsG^FDAC`+IhrQM(Q
zZj>FchqZ&^iYl?sO;=r_o6qZdQz1@uJYw#0J}zylE+I$$T9-CM>M*42skg=76^i9+
zyvVROkq8tx*Gjd3A`SjqID0;vinB0g_zPGmVP@Ti{_OX{!vekU$R6jB3Jd}*7B_x}
znS41$@Coz?WgLkWfx<iq2ffvl2Nmew^1EKF4vg*nfG$=O55n^#PNGVGcFg<B-2nY#
z2%)Cy`El(^Hc$%?!3^QQQGo?-#k;za=XD&>9oCv)kSu*ZdG#C-|8`(P#M_$^yO{MZ
zDl13LYX1%M*KvtZqAN?yl}T4@sq?No+7qzh<e)ik4qSt?rD*Pp^DUrSK}}PIN!_FD
zeddLVf)W<j$(kpEF#bg=XtUX?2UM}tPWksClgYusNh$F=DOvrOeVfjc%|4A&tkkPp
z>Ji^@YJ`=#v-k&vxXpuo@P-4elD){83lOrggT9e~x0<VpGEylUZwWKo<}v=(EYlc=
z1pn`xQ$1#)-t&ou)jLx9qS&1}x6R}2+DA?lEJkrGQ$dKtlZ38zDJ<;FVD21j;C<Oq
zzzGoNa-@O_zPzFo^j{Xt++uTJvb{R`p%9>^VBgxwDOMm23q)kghg*kwoTD4R3ZS2w
zasgbtZsPs-7?Y8vO17WCxAff7nVDm=79@uG#<c3^O1f^~3CA(xjQBj6GXnD@7!00_
z8F3=OXfGW5-_8a(TcFWQwm_lEK?neobl9>k9Hy<+&3mpG1YzP39zf#|_E2<oZPSoJ
zN9Pb`fpo;`DCSh6AleiBIXO)@oSXGlA^~%f*2j2)1!CcEQiKw9d1oJparxMwU$wA?
zwe@qs*ndesdUt?}#0BId&vV%BJ9q}05`<{jH{xI3XjqzQKtE_2KBi_Pm<>G+b1>*^
z?$?pL++u<3<qRD#-eJ|ZoM_mGeB;ewXm;F~J!ArobRax<^_ha_Haff4czIZDviZ%M
zs!4PJT|CF+S+P5gT^Bz>Q9Q7nyMk@ddG?vx^e=qsb>9s0pR8@pK3aG2zjXR?H!E4~
zjw%w&bR<33$dt%tRciM=fWj)ywKXJhT98@FOg?X-OXsknlkd~ys*pk$=f2~5<$}ix
zDMiAzzzsa%$s~yn{G&&*{Bu-b|DQ*NZFP<RoZUZS@*WYWa8}QRkFz$HOaiGuA+pF4
zVg7lFjDHxvqU)93?t0MjV(Hbgt7bl2sHs{L)9x10;TiYf!ex+JRZtpPwg@DJgh{ie
zn6zl1RB%VQs@XzX`r{8CM^85wz8Z)&P>C1=3?udY>~sjO>M$;f1cm*?ya5?;OzhV0
z>+a?4_1F$dP~7|N<?&uFcK`Q&8Jk!v_n*%j^XHSVE;-Q&z?X7<ECU6|4PvH^)Sylh
z8Tra$rA)TprOUh6+>+*Z{zk&>nDfsws}Q%!?fT>j=lvgKsOaC}huQ9?73hlGfXv7Y
z@CJYQE_^pvX43^Cc$POWl-pu^Dc_;QTA&g~IjOcSl3IF-=m%rRFPOxTRa|wKHN1F+
zBD-<miLAid?&uAZ0XvZuL~uHXK&1|*s>z#{6uVA!6i#!C!%LlTAwU!Gt}a(Ng$(Uu
zzG?d&Pai&@xfA7wK{4)HJb%F$ghvJ(LM#wcAY&XWvGMWfKIuiTMg^ePMqc=#4q|{(
zIOB70(^hiaqmvjq-+4b-OeP#%brju#eleiA_2+Q2UQjtwolw2Y50ivE^`bbiK{)TQ
z**Fx5^#R99uKKK?#mI%3Uh{rH@n9L(#^ay!T9NVxgY?BzU)tngv6#zVB^m~_%Pi$5
zt2&(^(8C%ZgbVwJNT4v3S1J1!q{vlrY<)*4hZh^agK&G+WhN66l*&C!cTz>LAV*N4
zW8_)7fgLA89}^bz_V<a@Y|i>zVrQ#^X@$~OeGb720xOi#4`9(Tb*UA9EUqHJkH*4t
zSR?;LSM3k<;X<sFiJ57m5fn$&&_sOq$GI<|(K+^ls`B6WoikFj<`kT7EH~GKfh=td
zfcNpp09ssbyJVE|P@X%~fF+ie<2`Q9{QUt<ezuW!NaKX#rcrub&eu1km0fEuSvG&I
zi8A*mESt3w^4Ef>!Fsitj>xSmG;%L%t?)lzNiotTNyZH;n4Vo6Dq_?S4V>`z1&Tts
zcFPIENmD6?7d=7O0+ku;Y;5~o){ORdp4Mg)Xww2lPUy7H?0mA0dtfNBzpw!&UxVqn
zkvUT{N>Cp=DW*;)K{2TDD0{PLJh-1Ta7>tbk{pA_G8}6^>bOQLxMBI%id|x1nXVs9
z?b@XE4`pBH#u&O@XyV-#?QIKZTZq%A2@8M^vN~WR)gSuVi6jZYci=R6`?x4cd%ly(
zuXO(6VxdYFY%{$K*UOuCB}a@|V=RdZfh!8m<MSp9q}}xeup62`kiG-;t@Sr7p)LzT
zsCCjg-e$(wHu?6F7#P38Y3%?>MDgPnB<H29WJSxG40@0);nFoq&8@-FeP#n&9!gTI
znr&#nx!F1cG(aUS7;d7$3*2iQpSbt74E1yNAC$KF8?CV#7(x3-)yE71Hmk-uo*QD8
z@_Bh}omY}eiRE}Q;7k+l674HjZ=0=fQIyPf?m$(#lm~ps+N1+heX4DbbFMX8wLr$s
z(^j*5J&W$PorHr#D9hwt?JfbaNa919p!EtxktpkbnySeHXpMM^*X;MJZ0WJ=*Tl)a
zU1FwZc6Q^E|E&cF>r8m`cbP^`iT}IBO$wqYEREZXeqBkHWygM;h%B)W=EOwp?LG~8
z)dci0ZChE2=X$l%Wn3pi9T4OE={D=jRLX}d*Fu4=grwVGD?=OC<kp6?&W-zW!!<R1
zf3atWl~_skkFv`858e|4J?sDM`Pf$1a>^1#{;#YmNX~fPriJmcV{o~YVVG$ro=b0n
zQOD0j2Q!O1TK|2=RgVXt8BDApO(KK8@;z<bb(<Af>w2vJ$&LEhJGx5L-$OGdO46Zw
zNqhcRSnmSIKfL-8goO0Ox!>%+&L8p6{9t@2@Af`VDBrG^MC_jjzI(sD9h`WTz2TuF
zaJ6?oIX>CmzOE9RIGg*ynLtx%8pl_&_ZHJ$p05_;H%#*!0_x*)<xmzyHKXY)(s?xC
zOdr{y(weLdXpjcEVY*1DT%C-W@4bQc_Z)A{S##tL#JaK(HMM3L2)C4DRmc0<<2yvN
z&n(aR)b}7))6n5^_aPjP23Ij5D*A~c(f3_nWxBOZDqze!@HVqpw7-d_AD(t<N*y|#
zJ%rH31n(wu%)(3p(lx_eH3RLB5u(#ciS-0A2sPP!ssh1ut?QY)S80rBs6*a1a`bf2
zgYm16Y2yfpds_op(M8s}$4MEmEmD($h$v%pJb8^R@8C(3?pzf~of3+R9#RP_btphq
zDS(Az$l>qkOW$@wC@j3uXp=PlxU>4gB#@heQ>_aT_v|YKykGNTY;SJfQ#<9j>H^hY
zlt452WEU7-ZyA9>rdSEdz6>5<LM9`Oiywd^uSgJ9x=5(#`zNtD@gg9)qUYn&OHEuP
zGu$A?a`3amB*`2*XB}wTl}NgkRzj7Hr=HEFiUvn+Tb8Wh=V!<X7zOsFs8@hOYBPqU
z7U<F{Dxq2EAOBe+kYfjioN5ru?<1JlAH*p1NXh^jTqS5;72pQ)nsCB~@ev*0BbY@E
zk;R~JzMwoJG1Z-i30?uP5t%=RTYzr)n(*Qfc4}E7A3=}KJemE`S8Nm^nHjn}P>7X=
zaW=vqt7pxkL>#$18tE3@8={K{hejglrv@KXbVnb=S}?m`kRW-D&`SX=uZyeBX^LN#
zJ?gQuwV@{6LIG)^6qh`jd}#+QT-@+x4=ZC)b2kpq&_r6ik-}6cq!-CO#0yo}8~X4D
zCn~j-F(@S6use<kkr74NG<L|)M6-uftjYrsV?CNyd*pnd5-#qSb+O>`bhmlXsw|ou
zRC5?(u4b0v0r0kwkOt`8<W*4pTJE=0Bsi2^zgK3aLI~lJ$7C=Cw3V|eGngyt3TT>Y
zmn;ec8(S=r$J2hTW8#scYB`xB1y-&KuSu%QI=Un`w;d3wzdZ2`#Ctw|zxu&ei7o0U
zM?XSx(n6hcer~$|p^04*a)#j1sCbcIk<oZjO<GJS$v7Qzn?mYjAf-znyG$y(N+~;w
zOS~$Q^Z2=vMYo{sD2A!<ntIlSCsz0v=-Fh>qS;%j1Nr8>E|lucMu$()JPUO)$neaU
z2fKgSN^P+!U?ZW(@YUozofBD1_s5xAJ)!fhcSkEx%jizT{sB*p?d|4YmFHhjRd%NT
zUG&iSKjD((0#j;qJNTUY0f%em9A*iml56QiSM!WSjui2*Lc^)P-Y77Nz|!Lli4ASx
zcz~CXZ96vt>=`0#o=q>#buRW}>P2xW%AyHsLq<ex!f6p^j<}sM?O0w}UtV?+bvIRn
zZ2SrK`TD3>KTb~`4i6?wr2wN?QKKKb+2=*tDxK^%`d?PU;9)Vk9{}4KDmRbk$LAW1
z+-?tdF1O$FL(Pvq)}!1XQ%??Pn}}h17`mNKnQVTZE<1I6OxWdmpdiSF#V#6rDcbqK
zShd($4cko$Xp@UCn`n9tMr|+TM$Hgg_qxsW)*fIoi~wcLCW}a14CFy!*gul!o%H*%
z(*9xvEd$G2GtDIzrNxWN?8f67*TUiE7nW6r4;`zTzUwfe;1vXX0CSl>VgUN4<SsQT
zP6zt|ewo_6C%?Va;2a*bLVP%BrA7k=vN--Erzvs1VBz5qpe?ZgdQ+DyS@9&s;`We7
zu_tnr>A*KVMCB3M8JZ#yro&9XdE{fPvKLuWkDM1s_+GfMGQ)5Z2zERn3@?w)P0!!(
zBB%6E4xm=fh}x?~lKPVkbpwQ+?T_Ch*5mW@iukl)Wy-~6q4qbkHBpMcR>06o5D_Z2
zCyNl2?oXNZip`*I^Q*ek-w8sR@FKC}5vUnMmx?pF8HxGyrT=c0AGH-NhUui7RQ(Wa
zbY6g78oMVe={4O%$m*HRK}(e`_GhKs5dr5likk)?s_=KVGx@^G0Z_^N17{oUkcx#K
z7|<IH5WKoS7=6b8dxlWgpO5JYg;t6WeCrD%UqPVKfAFHJ4<a0RCnH^mJkxESqWV5W
z^>4BKjEOB#3-q1hF~RhAK;`&vLVG1iUx2dU0-E3(0~nM1$`?_H!wU6eo^UKc`uz^T
z;GvBIVa=0jC2^0V5o*Al2E|Bl)AE1}{9+^hQfZ~_78Z%bv@26h(Bx}r&6oprC%TPt
zm(B*EoU>-WJEV^2$s>k`qPCJ_W+p!Sy#T{UDOjA2v65zPz0#*13949azsprn<+;^)
z2p18swgQz7hIb&rDz*0WX^lX}lA@MZasF8LfjKQTvd{gToaG0%(nD$z6#g<}iMz@4
ze7s8N3cMOeAKWWYmI6=|k2#@A`Xz{Xh{XtKslgoN{3edj1!|w})gP~H6%L<=V;4R}
z7Z!HoA=Ed_T$2=b1NJS0505?O{X19ln?tABh0wU<_CP*3Fh030cBDV%vviD*Ac$?`
zhP&!-CB8cJ+ctcd&bATc$+_$IMACJ4M77SxrpE^V!77g{5cw?DAtZ5(LxyO3KrYcC
zICn&yV-zYq&|19<IP~VhSoieG$ntPajQI`0cr_g(-yNQI`Qxwdlgknj#19HRm(`1Q
z+yr;03>Xx1t+agda|xSj-eeJA6B5byjn?B4da?|BPz}DdL?QF7A&PzS07mY$*4yVA
zoWMY}scVYxV9{k_|6dYy4<P}@w1-SqTqt^)n<-O`t)N8HYk>>gw+(2?vH-f&#_-8A
zJ~EmM;K(Tu=J3`$Q>G?RPxu;SrwMBSys7sV8XXf?oGj5VIpuqVUDGotj4guw*C)6L
z8SwJJAzs-^61U11x=<PHZ1O-dlRTs6N7hTP;5fEVmVrM<E#yFi%nUy(yIx6$fx5G}
z46rH6{eci;R_$+DsB5bf5bKy;9on8FPRix(Oxxg*m6kUh5^AxT?%=PXpxmvXz+X3z
zY3ZGHrwya)KlUu)`^NGF?G^ixt$KMXLT#0tg0GS1MS@_MK&a1!@s~Y_4afoY8k7C-
z6rlZ8jz&JixDI;OYmu3!2)BHX9xK8W1Ugety~D7(W-~R(>$Vd~A?nofLuq!We@AC&
zom6ptaY3ax6|AD_6a&xGZz9Ly+Mlh23L|(S=!@j6tD_8R(ABZsTvKP2UaFkvW=0}}
z_Rj~hK?^jC(!CO}ZNYSQm8$xa>hrM$_hzM7z}y~AF}jicBBrFDXRLaxxUQ&nJBLS)
zm=7n1D^f%Ivy7QMTJ=0#pWCs4&O6O5+{p%(zRs+7T#H%epxS|w!o>aDh<2)RJW}k?
znQmq_pC*@ul;4vXo@qNxC3BEc;9EsWitQNlWUidt*29$ZTgEOALrq+UP$WnV_q%hb
zjlkvwqK-4Xqx#BiHY`jC1XM`Uuf_`&6aawaD5ouMn9JRLbB%)aW54r!Tg-4*ECch1
zQqN?_4R6uEFE)RbUzqIGt=wC!%r3U}x&kiDZWDhIZ|eezbL&D0l6UTwyL;!D_{!*F
z*LnZbuYKmMTwGaABy`$@R{`8<5v^zS4k;d+(@Hp;PdwiW>zdUk<TW1Jg4T7Hoj9Ie
zGYbZfeDAOZ=~^RNzoTU>9m%9;S@;{BlG^f{%3r!X`EMP-kLzo-=kPty5F{tMqo@B!
zAq{N?>AlnsfJ_(k{vX|!?H}Ge1}0YK{~4*KZe#b8JNnMmEAUBDirTLdHy97AmWqP4
zQY6M@&BysQVqk61c%<<oDSqIHD5fHloHbUjlPNx&yVK3_xG}?`0>bX#ebRHWdy3g3
z5|c+*U{Biup@IdD{kXlSYlG$p#DHW75f1OJ|BT7;<$be8l=ttWQP+Dod{3>%)swwk
zPaGeIy_(_4&6&RZ`TU{5Ob9>Nvl38p_qAt#@_wA3;h1d$gNmk0)Q{zo#Y_Xbq6T^d
zgF+S?nHCe#;VMKZBW}hu7Pe+m6yO*X!dqt_IM%RP=-BphwHEsH1QKMB(^++ArB%7q
z-jXRqJGk?;Slu4iI(!z0B@~A%p$G>P{wiapq;u}sMrNg^)m)F!x!pE0SuL!-7X8~I
z<J46n)pc`~h`Q<;bgj4|)zdLW$>Y-NfB_$!PB@H?OeQX|5-hFQ;kE!36?>2Eurqq)
zJKTj{!TthdL+{18XPy90*M!v~fj=ouwgZnOkaXvMBrK^=_*Ud`bqRVk>aZwtWTg90
z)(!Gle+RVWfTA4fcSw|=gNH$=4k|2oyijn>Glk6PDr^r;Few?R8|VOBjvyr9GC0z&
zF{rYt7-GMes?U7^Et=fsfFs8xpgs~54_zo}-yLT;9ARA#ohUGjasVkW`NM)iVRg-I
z0TBBL$~rif8BhqvQj&K-K8qxI*-U~++PBF(K3A!=$x+r|aMOw`$E1hhN7WV&&R@|J
zm6@R*M!&W<c+J{tV!5p;1C@&jC7ZNy=qVd*u51)(SPY@Eg*&t)mWCIzo{s<qdI~qB
zk2cyf25spA;xd&w1X1%Txw}f~EF;`q*C}CZ>wNg}lv6hvG&h>F7>DRwj>VQ=G_TAQ
zNnol=_7phuBkC*QXO4=GDJ<V(f~IJ-IYEmJYR;vJu*yA-%D<7cOwF{WuWD$pgNCT(
zUz?@5T<9KKG^~|E`HKvb!};ZaqnbW1TIiLwpk$S`an_Wb1PtBQ=)^$be4q$u6R&(<
zI`*>oT8gy*@GlKxd)sE>L+v71>YP;yqK=oEBK9{3A~Tvbuuu0E_~u@HNC<sy{!ZCx
zr?pqJ{dU&MVFPTti$fpS2YdCnz>MtG(>mNsZ#=_OxLxtAK4gG}URe~}6!797CYRvc
z4_K)JY1;?<E0+AO?|8=CaN*amFh7&aErjSFcjkAzVB}DaQ)<oU>~644uZT7pfxN#=
zvO)4mJLN>>S8|Po)#9kKqI;jhn(^#W8q~-#2gg5zY#HN%vyIP5g=&OwfNq+YYI*qp
zXVzs9BX$yj?k+YCYs0F1LJ9=ucr^vQ(NC#kEYi7tjc`HL&1Bk-9x@6IA_|-->yFpo
zX5z}jY6!S*$Y?!Ngp^v~_OJzEMX%(rYlvB4RJQ-oaE_Qc;fr0r?hGj2<=nILr*d7%
z6>ENzdH15Va#Vkg_zC7+`7+3&CFxm#1O%ZX7gtjK<HMbZI6_!^5%7<<3K4eZ;k#Z@
zI4>}Wov!cfFJyl7djL?`i<qDRTj*c%F2;O4Z;1YYKTxLD5{6tfCY~|JyfoaCt#vu<
z?bo-2W`PJKrwWhtW}Y(FBD?I?Gn6IN9`lVetJ3Q`tg`DnlqJ`<ztyd$@ka}OxvVQt
zM+;P?*Jsa_*LgY*2X*a7DpovmpM6Z<or*DWZ8C?qTgM<_3Zyr>$I~fgKj(92p<{Zm
z;`4mFm@J%L(4x<`ww<=<xfwyb=CUrE`h@><DS4`-{AqQS%)&6+z_SHKjQo{N2qbPh
z-J@mtSkeVo^2JeUhXSj{Nt4Y5S1a`eCC9hr%aeoi34;TVlJ>76{V&a%jqU%(-HHEq
zMEbGEh|AD`*!hRy{>1N-ERvWNICHp5g`}E?g$Z#IBk}rrp@85eA{vi|ZOO<4_Z;0C
z;{KbErh%|OIo&yNxj7E0!{v!9phwXFlLGb-4QK=JnceXmc|O^FzmKB#%0)!#MH3zN
z6p(U0T|X_o-#yt2fx@t1!~_rSj*ju_eEGh-o>t@G5pn;1zdWHy-95d{G$D0ydwm=j
ze_kC&v5AfYaMkc*8U*~ZNW`!pqO%o7Sxjgaj3U-m-zG1hv7AI8)ops~c7AtbM~ncE
z;$8rwz9ZJtrA@rt7dOP-8H8=vb%_pADBdT+j^B_~$A_7_M|AI&tyD=^S5kQ#rsenS
zRad!LU3G78ocbK(wTqih3cfF#zMeg$?sK$0a17dJ%&SGEYRf{iWy&?NXvrjv{r#Ie
zXZpGdD!(BN%&s-NjTeYe8HC$VT;4Ly;gU~&&uEyL-~=kf-KZwd%L7q3<cWBdt6jc=
zdqS>9+dgY*#$MPt{pAhtj2I!8wnhhA7(HMWU!6En(yp_w3UXikvGwDlc|W|8MWN4t
zJ-m>H7X-Sls3k&K%M4~EG>(FhwxxGjz>gEx8^b+Ko>KbG5r?4Ag<b=8$8UH5d4~rs
zdYQbvk|GU|C>$$|I*=`h&j#16g{>Oh6i^Luc9JU=(1?;4xdH;TQ2iIO1&RjYhL>^#
z`W;@&8LXp09`_<u-JwB!&k2r2xhtbU`uiP<xTrqW0Kb^&kQ=w$<x}`zlnQIbVz(>_
zj_&b@*(;$2@f2}Y?IL3X)}Phl`YMTB6^Bd`5er9&#d>r>fR%_GQzogyL<Yzy_YC!u
zWA0w(EC<$S%LCFGpjiVU4HUjI{v{y5#Ur;&om$GSHr0_{j@OSzZw8l6{W4sAFnn4f
zX70!Ic<r%YGu-V@g!!wmPQ>`!8<d>XpQl!g6%PkKFAiYHiz^TnqLS>?+Ic}?YUhRq
zs)OWMAy)CoZdM}rTJ?lFtznfy#PrIQ9}c~h{1rrXDlBc9#0{WQH;04Y<&o@EfV^oF
z%ZoX??Px1%Ka`w<(f3Xh(0<%moMcix)rJWE<zfjRdEGQ-aq3BcYUV=tVV7V$xQsx;
z1lohgkf?CR8B$vb0-{)Lf(n>~9a1Gcw%p)--l|rUg%9L8!x#S%r(p&{e+?pkeo<`v
z=hT@g5AAoz%$aVQXnem#!ran5g0OZiP!v3QPzsZ9BL}u#q5Z>;%iK4;OI&1ba?H$S
z*p+mzT@3s5>q_75`v<pFzSgEs7nQK!QLByi$fj>bU_kNlFUd!lICd>#>;VNnsM%L`
zSjaOUS+UKsOE&K0XV@t_9&8h<SDtetK>Fr7te~NVa;MH-$}M3nzGwvjf?v{l{RxaJ
zZ}j+^VTwn%<SBsCulGeD5DzvyzEu-t^EgZctXZF`B<wmesOQWP+rBlh)#*dTH}nx~
z>BB;sCH+m5J{%ULpgYo#+|&h;Yl|E_jv-4gaMd%n^|E=E<@f$ue4b@=0#o4~f`hGv
z`cU%r#JD}HeNb1Q@QW5ffWau))6AU*wY{7AEujbN^E{aPcJ1TD2c<<`q+n2MA7o*u
z+io(jn-A}#VW_`hj}A<2{VOF52$voT;-{X~a4kLir&<KokWBM%;d?e75&Y2*$PK-L
zyPNG=^czeq{fzTeo7x?3Ek74)%L2*o-GQ|<49B5eEf(pe((b8e4h~B9bIj(Ix^Abu
zoAgT42_zzZ*JX~ST$<5&VzI^u7k?M|%G`?4HQ@C)!*pHK63gR<#oHonMFzz90;0n|
zzr7-FdepT_7rOy{f<@oI>UTFF?k<@EXhw#~sjixKUrH(cxMGPVHcdxvUrf!mQ{FG3
zV23D!>Pc<IS=37Cd)i95Y7I81B)t*<`5vOU#!N-y1HEtQZIX8@pb((7M>=;iq(G~`
zQXt1&o4BhFrlv+>bTW_ngoq{!d<FPi^kVRVFEMxj)Vh=z5ud)Ci@G;y#@($bs;<h+
z6DSLA7v@yKvRR{wcq{Fyy8XH?IPfj%aIT4b^89No2N$I*3LCu~G&O0Gz-O$1ArVOI
ze*oCz#p8C8m-u{*wS&^EPB6^|*hAEs6Xc<KjMb=HHo-WX9sl5%y9U(|R0j|a5^Q<W
zfcFOJ50p9o?x{*3QW6Jv8Tq|f7=NXEz%47a?gnc6DSLwo%|JJ!f=~;!Hw%#9|6tYi
z(guThZmNC*$GC+~Nb_ZaRH~u*NB!ZolgI=am@Df>hhyXpHqB_3RlE0GVsPbmk+j)*
zEiS+LJ&oHhEnq&A&Bt$4tZvgKEMH`im=P!Ok-@EEJeTjj$GN5ZJw4Ey{`&vBb^Zs3
z{{OfN${zM6_;hlHmP*bxkaV*64D|H>;WZtdoblP&e?Fy6Y|Wg_@fq0Y|7Wz#jK=PN
z^N;VV-oV&P$U%Y_E(ZwcqN@g!HUGSEe=9zm#zpG@Vply*^T+LLP9kfHidJMbDKO$0
zG2X<X+wNC>vLgL#@o*^6VP*^(qlC15;HEjyVWe0#WbJ@(Iq3j=*#E=WI|X+dby422
zopfw>Y}>YN+qP}}V>=z&9ox2TYkIz#nVPD3tG@T@xqI%;T6>>;_HX$|igj&AkExJj
z<{1B_1?V_TnT$q`1RF&#k&gpkc<}TGBQ^Gmph*oQB5{-xh9w0<b7G@NWuSF%LV%J`
z9+3Y!MMor}fU;7D9!9MY81~+RN_!EUq8amWgLc~=helKxwItT~TXj?sakpk@i@6Uw
zc&qfsf0c?3Gvk7g0}6S8Bb2lpD8`5o3Jy3Ta*$Hej#z}85-CVhI80W^X0;#JW|RR~
zt-xTF`|CFjLa1Rr^W<t?0%n|E!Bg0i`%ngRXru=RF|&&g)12qmZ-4~;pQK=<r!pvD
z?t}meB2TDok+?rd`JxaSAQ6QF6swXcGN{lFw!`BCn5qF;44a1rJcitnJfA`yLLCc0
z<=x7r0yW>LuDXA7MTpY0prk^Wkk#r2YEYK&){OB*(Lad$Km-_hsNzKP7H7*F)Cpan
zo3-Ji6%hW2F2XyUi5S@i%9YDAG*ZeIssn$>|B*w1DDN&5h(JM)8Hpo09cTvZ#$RT`
zj2oYji%^xP_q*kT08dW{o&YB@f$}3^2ccm8oj|^oRub$k0wyemB~NYy2^}57WLw(+
zsEqvhWoKFWvjOVJm*$>!s@}g>qSyG7&)<CR+&VZMp`_fkx-0e6>E+6MGF9wLPpF+y
z3SIMYlN)obn5&h>v~RGaKFsWARK!c)P5pS-m!uy|A`LMfGU3&7v&?76UpKa3sEj}}
z(ApM{2?MIdtRzhEwk8tNk4Z~H)^;+^<N1Gt7&4(=rfTWw<=M|kQ9*zL2(Y~%AJ!`M
zUVt7++?jU+kTT!>y~CkSvU@iJV<eDT-E&z~PM;AD!)wPgdbo5ZXW~m&J1Y?M@iWdn
z-aSp_01o)M-Fb4A7=Mhov}uun^unnvCf=U*eLAY0&A(1t&o-hPPabAU!!lXL$bef1
z8^MRO{uV5MKmZk75pLP`wC_QnX}5Gjz=6mw%*k!YUtKx(UP;^8^1C?~7X&XJH!fAX
z+Q)5r_P{o9Cujk|*mMb_TK4$p!cjXH<oubkwK8?5x3LfMk5p|IsG==9APw(RyD%AV
z0iu=KHZ!p_iP*0oVNTGOhtKTT+MX2*HLZ`4nkxxc`)kF|;Dq*6ED~(2#m0Oi7O_b#
zf-Mz+J3<<w7G3R&VqS%pJ2O;R1`RD*G^!@$VO}8$eI^KPsT7p7q}C*jn1k5{z%Q+2
z`P!C25!O^_8w%PFZ`ft~r#m2|AZ4^^eN}xUqE$IYztyDHb1{4XU}P@I`30aT5`Vxu
zbiGnmo!$zTnk9qGY%LDs&5-xJ@``>TJJt{vQsflav_q5eif!qL(^7G+M=aw(5>IgK
zCy2uYqH;h^5V_5b#IL;XFPcNgyv+wflCsAei`3!!s7Tup-6`2DlkC!x_IRZ9dW)yv
zM|eM{8yELKae0a#v(1UiC1veuHxaC-Ta`6>Yx^OtW*OK>q64_5$$NwuNmqs$UIrB8
zhc7VG7|T~WF?Rlzp+B&388KMZhhm2olq2)gJY)&BhkDA^vfgdA&;6pZg)?RN7igTk
z-Ll_9>*YQ&8;%~98)kI+HF)np7)Fv$Zo9Z|u-LKV$?A%K=sR4hE<nq1<3JE)md}=D
zN}1KqyF5`J*RfH}YRrCHpervl<AqWjO7vro-RJk@MQOEZrbj8PXuN_q0-po#_=(q@
zpW`F<Vt$Vp@xn@BbGvL1-{d*@a&WgLsKPI1%je;Dqj3a>_3l+Y|B@;YC`!Yv-wQv=
zDhED7<^TSaua>t^`PJR+6=>DVr>QEwbYrh~@so^LwKQ@huj=aB@_uXC)8=E%KSjSo
zJ6(JE_F4J<_U_c7(K9eQ^l|oS)uE-6O+Y_wHsSsNE@n_!^7W^q!)KkR_zshi3xouM
zyJM@X9uA><ljPB9X4d+=UVfS4d=XYpxJfzov6^nPCcSXU&nj5HH}--y6}BJK-Nx03
zoA6+F0Hu9H9zMsUtB1+#e6i@1wuakalW_HoIluT>lk>RWAPwIj&97L00rt?=g}-Ur
zhO;O>L09QxNbdZ_NL=aS-}ifWbrCq=w?^LJs?8<jdKQiw{rJ3HvJzHQ>7zyR-x}YM
z2R7dm->(im#}+95dGjt=d`vhm?mH<=xOZbWSb)NM+@yaP&i?5CYmV&f|F=2*Z}Vee
z`tQx}e}NwUzrK<f%pNHme}3_rG<n^!P>wnISYV=Q5-UnYVm$uyoet{uH)R9w7%4^2
z;lH-><Ej;Q-vDoScV_KoVl7%Z$TdGPO$B12Dqu{&A(8>3|DNd;&8c|ShlmtGKj-J^
z`f6<CEeHlT*T=`*&&{dD=j*=4_xAo0vD3r-`)16$ljrN?a_Ro*Z7uxm<OqOKmJNRe
z6~mys8^&4`I|;`<5)|q=6AFc{QXnBs?dqMA;*1g+xujomu~|v30}e)lj%<|j0CgA9
z((yA0(rRS13nH#^r*^ts^=}Y1Du8%*DMGwvDM@{?7PM;Zb;K;gTK)@lf+Pn9di6oX
zGPnAGtJ@I}{~+DzP#T3D<P1=QK9?2CN6H2bg$;r>O6n&TQbDbHJIb!OC6{Y_CB87q
zs@qH=#^%%w(bSzh4Z59;EL5#uGaJiEEe$~^=KMDVbfe!Oj+V|-q9U5|%4lGUY<v?k
z!cCjmYW4>4t~mgGBZLAh9t2=AI79cN8yJCF?NB_`PEK8?C>mna+&GX7`n0wK#jw)b
zExfPSvfXjXUu>0p40vGZtBAH#Z6yOZHpsg94{8#ZaRwCs7S>B2CE_VW5p%r_m*|CV
zT(mji|9!(&dZoQUf+b2cTc@%VI%VXO3n7qENKTnAGpMM##VETaB9ed=gv>95V;qK9
zWHt(w2QnW0uHrN1khyLV%{_Fr`ol;SW9w+G964B}+f%tfTo)(u4^CjmSc6PHKWw?S
zOS&&C93WolneofIwSkRfZVl{h+swH$^5$xvxV~`7u5)VsaGuZAAXqhCye-g(h)U-M
zefSQ#Ns}}&2r?Fca|=hqNSsY2W};!6{HqP%$Er42C5v)m+2a=U<f^0H@0vtmO0wD`
zIX_Q_1DQ3z>|SM*z9QKkJJG@93yd!qQ2DoGC0tRljS@lReAg{GB^5SoY%XCqny0h1
zN;TUwasuDM*G|1wX`^nLyHS0lPA)GR>we*v>bhYnUs5gM$#zx}A<JPUoq1Sx4?ZFJ
zX`ur6@UN%<*vl1OGI+KQY!IG!!9wsC{Pk$+x?tpL)z;WLe0wmX?!E`M<FoU1RT;&1
zt@N{X4cSMztWu#a*lBw{GKJ^9=9<whU@{dqw_ML=0i&V(XrZvdMfDv_NhX_8h)P%~
zN96<9b%$A&En{hzF$o^81!ejs(vLQ5YhVLIsyIYCTo-?4p9Ebu6T3a({JQz^8ze!c
zL;@yB;9g7ra^gPBFVBJ?a+X<kC-`Hjbz2uz+z70Y;Lq|5L*IOCQlk#)j75O`+6O|-
z+6{N`oW38$j;|uwNTHX(et;gum8AglcVnr##cQt`H&FOK%KF#w#uO0AuP>%<gMR?z
z8yALU<KuxI0^DJ;Ge^yL&bpma<v#MhYmTGn9~<b{Cn9knrYjB+p}~3Ekwj&Y*eWat
z#j&`sw6pM$=Rag~+V_wIyIZZb<~?n!-5S^6cG);v$sO@|l+~-q=nTa(<wd~H-4+3S
zxe!o(t3=O+GK9bW*ga~H2LNOf`J<o~|3Kmz1X-vM^#=`(DX(X=@`rNK^tZx5ixywt
zA>U305n=Mz5Z<L#h$Zg!3tUES0*_b*py38Re{yNbHYOikPE_;$QOL)DQ<V(~LhXQ3
z1FZdHl(&VX+rufjplz}h0s@4(4IXsSeKNjmfAV7uH_!KxQHoz+&-L3qFta5LKm?>W
zCJ^>=NZ{~-6pJ(7H66m4uMRr5b3V+Oy*z#v)lQN0hb@<K5ig9CaDor``g+^9XP}7?
zMnWu3gM4Iqzn003kpxuuOk1+iP=}oAb|QqojxlhuU&On*3uHaBCS1_)Jsv^B9?RmA
z;e^1U{Dnx)zQvl=4U&F<g``s&|AWYLu>4;g(Ekzn|9+`tNn^rkgAHx#g1W7k962bO
z9bK2Bc8ffj+_KPxOSU1wbbUWYgbKy9^ZQ*7rC3BMpN7*<Mm>abcYkxGAK20XYj(GF
z`@-;MsbV*fOpWmu96d-a$Zxa)A&9|;U296qpO4$CD&_nZHrR5aDA*=N8LscI(}UYX
zW&b`>c<?jC;=;YVaDKiX?l<pSbpoJ)B6oc`KUY*e0$)Gp%dj&2Z34s{ACDi-onmom
zl{TE)zFoVhlTP@`D+n1><J65u0>~KPD%qhQ*HEimz4TR_WdURjk{>ZLJR$#lN_cy{
zj+2D;lipVf4%k>E_zXfF&5$7|@r6*L6H}Jyc~Kvn9PXF>H6QG$LQ))4$E-H@vruod
z<k4t34dQ;Aw^;!oPmapp7*!@Z8Scvrj-jl3#AR^6F|$DF3=^ACMSLrJOO&~B)1`^R
zjWNlNoFZlAh*)IVck@Ub!Urtwx5MdBf{f!W18tx6qdsg{`Hd~PG2_J(65<d63SrR}
zK%U(&6%j{lnA5YvWt7^qQ3{R3e>rxovxhd0;+HJuAo0Ym+2dq0N;3iH^MeC8FAx99
zz3ZVhqm&qmlJ0-g*0@qjr?c%;Ks`ew9pjcd&PCgTh1PU$WICxkRHFSYtRPx4!jEN#
zgXOE(Z|VJN8L8|^^|kw>QkN(}U|y}-(##KEJoz#^mlnMhJrsSp@wMc+M<a&3xTFXV
z>yhN!PO9)sHIf4%AfpnKW)!1r2fgQ5!Z4O)F#Y;1#+MO5Hjz(A4*40)K>uoY<+$J3
zg~DdUF?=Obhbs<z6%<XQq7xsLkb6(~(|#F14@+a}<czo~>$7ltwJnZAb85P9D9Ob|
zDe#&vhea;kJ5j4B-mWtPjypUvYAi&%`z&8;jm1Y8&Jl(6?6ReL>O%6e1I)Tk02@;_
z;CQnp!ldCUbk-vS$qoqjBXRRBivTdb**7fdb#?iVo932MyCE1QE<ULLX(IKzZpeMV
zd6crifsje29?rOqQ@SmL7g!5KV<_tMTp`Zx`1^LRf9mgGkd1ia`fVs;F|khd$u$_I
zFOV#=H*1JQh#os!T6KFF4(DDroHxJMFL>tJ5TgW6^AIEHSo7|>9ZQEe|DLIjrL5hK
z3@%t1zry2-(*Bd}JbdUOQv4Z<U}iBEb~|mg$Dr8^k+2Guu1hxlCnniQGYH5q2|uWp
zYp&+Cf-<*k9`u!4-}d1N7VsLNIczB{Mcclbh675ACD1vTo6B{zU&fPxXID&iA>bz(
zim+{YKC>yv@$0Mj&PkZfY$vV3%7<LFde=r^#oK_+Lq59uJm>mEl4{8SZl#cgz(HkY
zK=%n_gRHHDa8P5wUb9@j-S=-RWdXF~WXEFZ8tcVnH$h}m*D4R}Qm`lilIH>;xM~pb
zP$@n%4E)MY1s1-}YazIFkIcvU0`w50Zg!Prh=;UskhfQ=3Gwh_BrcIpJeLUALu0FY
z1hy3S#sZr^hSUC&e}aBQaO@iK&WE1)cQecl=fjiNyO?Ld#Wp1VA(W_$6itJLV#Su|
z_`qR27M41pe%%b9uo6p*-fgRI^V`0=>w~=Nz4`huE{LT0hcFCTlDOxxg`viF06?Fw
zoX@TmN63ES>P&OWWc@6toy+gtc6f%DRwyM84ym+giW<N=dO;7n%xNo~d3G~M$5&}v
zTTXFl{P(r}d&(+cmE&DNsv&^zEPQTr(uD-8*T}M|q@z%UWu@8g!tD0Z&MPgp6(0aX
z9*z_Xatjl5L;}CP+vLC~O=M9qytev=Zif>FgR}f;!=~tGE!Gr_?&c7<uN6Y^mg1b9
zPjqo@cYU>>)ubQMtqb#iebOz{?6F59rCSoSrt~LSq$3yEQcu)K{bK3ewm-%z<j$ho
zw}9^oUi!DzbnopWr{SMglbccg1Efwz9PyuYpW{F1J`>}AJ=y<9aZ^6{hrs_(pH!@k
zH9Ps@I7}!p7+j{a(I}((-2lk|qI0Akvu5-9^j4f`GL{4mfxDV4E-ddZ|Kg+XJ~;Zz
z@X+^WEC1y=fLWgBNeq*zjdUKV2F_F}xgVS}7VW8tz<RK-`#?UA|N3VxKGbZk&fjmF
z-_!l`&Vk+sQ4V2;fB?^beS~<2>-EdR-oeA?vE;|~cJC@MNtHTFk(@SrG{z!#{88f?
z-Df|ELjosLY9!QyhQ@U}M0jw`gio#q@+v5L9C(OQy=T_XjX4XVqC=~{;<#i{U#3yW
z)CM#9`p=-YPdt3tFb9gWAQ_rjnS9dGl@T^wPnX<4p-Q@6HtqYS=-_Q4HjZpBSu}vO
zG0Apd>T~y!>2v!lWt}3M(+Y3o=@;udi(LtgNV!XP*|(#3=AR|&Uq&5Hu%e-=p|iSG
z#F&i@g%qv}Z>J|ht2LTOs#P(~A=Aobu0+hay3HSTl*UHVKgQQCELJNczc=Ybb`F_Z
z3SkF8{Prk#6Q!maU)uOjzqSrI#-J#KJgs(v?z5@N#vJCkzs5(wzh483mwDbOP;-zB
zDbvr|8Q<nVX!EPl!<EtxQaTNiTEmt=A`7vWyl9wyrnHz0XO(O{@oC|^gE}@dRude+
z=_1sPsKFNDt<GhzJ{6@4!PQEDbSp;t%9zkf7z4a$aJ$q6AJco9=2{!)UN3#JQ_3-q
zr2|D;fU($-P&`72XIja^kJhe71$hZL`=+W)I&6+0&C3k}@&08VdDry!A2^Pi7ed}m
zJ`(aWE}}Pwvn!6$QUN84cuAL770z~|gqG{n!iJPkrK6Z((7~l5F)T_p<2PrCsrqVc
z%x`p*CQ-1X25+Wea%rH{uXLqNu~m6kWbY&orHGOFD_|ae8ESy1w4PTC?M*d85}`Yh
zz?xw^$(IPi4g|Q+>_Iqh%o>#FgIpC%de(@BB8C<#jZ5LE^9suu$9;-=t73P2yXVqO
zn;$4ni%bliTCwHGqR`yeOl+JvaAwKcuQSD`oa2@&DTCILYN6-d?}Ev}F<mS3bpUhf
zf=zcqakEam>b1tyq#j)|<I<yTPoFdnF^x6rPNN$#UOtp}5%!d|CUZ2YgaX1O6q=zC
za)X^MtwTLcT(PF@YV~_Ms>^=q+1juMN>7-mrIIPfYA@aq&!#=(q$j=O2*2GoQziQ3
zwKBt%Vive9LNe<o$Zm0`M5*>JpJm!TOfFri>1>GHNi3Wb@v?MGHJsyVIlDw)*HoeE
znl2fk&oxxFgFn)`felNP*_Ql4A^QyqM3w5IY)?5y_d1ebq;5~yB0|5|EA;4QpiTls
zK2mG5S-?onRpQ1+&engybXz$hbU3zn&*M84A8KpJxQ(6MOd>zT#y(p?%X;d&T#wgb
zvK4C^XxL=o@hLs!dn%wu)WTiHBb(X$G_KPwYCQhX87N;Z%H1}1widHFG5-1EFylM3
zQ#zPJn(<77;NImISvV+j@)uJDOFR{cei<j!A6ki8r#Y$Ro)$rPLl{t^KH7?SQL<>g
zuMHePRfd1NdtJ7nqx#g}EdbHEq5DQsY%%4S^Yw80&9?|+B2lDP&}R8lzSs3Sg!6`|
z<#fUQALNje>Hn2Oj{kknPx$|zvbCfxtBfs()crtx+@>P5!K=^hJPHg=t|b<0M5Q_m
zN=&5+6*Wxy{r-3f`R9*M#%}+@wUGSF1^d_bWnmjSC7)`Gc4`e?O$elY$*&sff@+3!
z5GxX`^nDtp2Cwb)jdh|&WyKJL>iKLEXe*yug66D|r7RXq%kt%zp>oI9Ba7a*?&lgQ
z5k=phrZ-0T)!MvEw)u^mg<vu3m)qTcBVL2BsH#MixGK$Gl}q<dtXRPhY_#iG+uKi_
zy~-X*hl?Pw0_!KI>p6Ravdf}Gbxql}xX+;Q$DG#aj9$n3l)CEq<z~jmBaAfAQK;lq
z@tyJf;=xT-BLF)uW;qOeyYEd`ag0H~x@5jqaaAY9XQZeX2>Go!VfKnFTK`k{29?38
z$M;V@LKHqTAxw}~gXXLo<&iG&d|bu(a)_HFmm<%M%ya4B7*+g7oh*sqoDdU%oOr&J
zKOzW7BlTZ<u9iacxXtKH-MCl}0-k?6X^81$Xe5lm?6KZC;5+5RYX5<E)d!T&NQ;FW
zPD7r-*0JW=YqE2S+79_&L@1Ak>PR2eG(uKG_s*h|$5&J&8=IWj>WXLUA@9#wy=auC
zjq8SyRk8+w3Uby2n~_pL2ZQ-w$EN}Ye!~;dfsm_%GMn~|lb#ApFlsUg=2~b|5T|wt
z9tN9pY<uQ}Emu*+<*s-8eH7+FN0Tj&*^x=8AEUx4QKfNH+&?Nk?0wdm^SwV!nio7F
zVKIbdXt;qIA8%y;0_Lz}%fH>_y}$B`msLrKozabaI?b?L;^BknQl6*|7pGWyOL;<j
zpK<<epTW{A<7Sw(Nb?E;jP*_o+7%}KH$@RmM<{4QdJG(=oDo+uxec7THgFQg@giTf
z-O_O)J?f+Wi*0DCt*$e8#L1V0u?)HyX33}Mp$<Kn`Go(d`?^Y$Lk@3(ZuxzOw8vqS
zjVdV>Ld`E%s?c>>S!-fv@%)(LK_{+!_6uw-42rUY@GW1Q8VxzP)tPQ|Be#=~s5kI;
z>m<alXzVoz&%&Cb-}d|VxB`5g8l~ejtk}l=U`2cCoT1-jV$9#!V|AOS`p2{caR2Gx
z{fF=f4#xj=$M|0yydm{5yA1}U?h|T{l(~q#N^ss10p)xF#r&tKDZ!RAk#aT*$!tGg
zaZ)nTxGWFJ`DC%hTgT&WtR&)72;|43m`fi1sTsOjp{VkBtW=OG_6TnYFru`A_+#rP
z2hv>KncipS*?|%fhLZ-GqVUWKQpX;Q%9wMGnaad6;8Kp1q~<=bb{!*q2t2wR0rJ0Z
z1<O;F;hk+LxS<@687n~wGMFByBFF)re}Z=T>`3H+o&Bl^N$DNfn9mO-lXgbdVzI`;
z3HP?#&re&QV%KFZEmSh=IjFPN5lpR_)l!&YIFPb)G=Mz{W7M}7-Nbe`MJ0=<a&&O$
z3%G%qBQ*h2NQ?<>lK>mM)HEef!%JB`?4qb24Ra%W-@|pgBJ1Ah*$VDO@@Q-0t&yji
z`WR!?^3*StljpT5t)nxhX%=nrNIGCj_;Ck`Oz57IMVSv#QptnM&90KUWi8cYt*Qx=
z{Dv?|ZdZoNJVDRE!2sfPh>ZR=x&nJg^fFm1=vl7J(5G@bd?`*Cpuq#$u#G^d$;g@;
zCDF(`(}b1X39#jHO9}~Vn0T*@`#;WB4=YJDwdS>usnrF|&9)2Kl;-M4H2EYt`Kh2y
z!)MmSzS((n(8j}Qw^ssH&~mR4D3hR4t}Y4|eCE0y1Y8t1ia@|L+@yUW+2@>58%;IJ
zUG>o$8kH17k)5H%1yc6ZeVi^`1RCI(z@6ES1WeJ%7MbM;l-iofhD<|TCZn%yDErm9
z^GG^8epz9(_YWvS+NizB3b^zIewFi0VZOX+HXzN^!)Q4&tlY)=g;(5<6Q~WgCpSCv
z{cZX_G9~}A)JJ=@$XClu|DMqv7{=eKFW)7kaY_v<Yn%E1+Nm@}Sj*AlY9+;G<*s>c
z<VT8DA_zL+7S~AHc@!^`T{bIe8H$DkwD$HbJ#xOafbdIJ_Ii|ylu%Vwoq$sq7ksSE
zIgnkcs9P%XPg0glS6wAPvc8NRyz;x|@n{6ZyX7|6S%U*ef;&z9U!v3Aa=dyQNB8?g
zn<W<AqRa>bqHbAoxnEBz(sI8w-w{<1d*v3xZNmb$gNy|)d6n)H7f#;R^jSWgCH~(d
z$ieWRu7KEC|Lff3?*9XHzLl5~fyxD&wd6WzkzB{kUR*9oHeqTK-N=?oAyI5u(BDA>
zlL%t5Jz1Lr25Z}V@mu{RONWxx#q;Xwc>5TN7D>i-pxMA7FOo)?iA<&1ph0f0OlE$A
zQJ33euqdW7Ni`4(9v^nv%8#%^lS?pjtTU&QxTa&u!pp<Qm(4gl-~07;!9L@dZ`$$t
z@H7{B;3W`@ZGh+ramjqPI)mq*dS^1t%E<PhgkUaIw&K}7F0USX(oS8~csm-IFQ`h+
zL%-*tNZbY_F5G6CJJ&L*o}0pqpXek>X>HrFZkML~$w3n3n$}wsb-ZEweRlP+##6NT
zV3%um7w3rpia1k3EQVPQmYVtKgPz_WQ*{-S>C(8FxAbk`e7(|9La8>?QGTAYlbWh;
zp9gkdqp#_K*V6;Pp?lHl56UhTlsE{a;30(2{haYZ@Wl6Xw@qL5E}c!^llMSGWvGG_
z&6No&sbOt5$?TyOoCGzfq3jT)Kx-8J9`NWTO>8Re!aXK|jNIr5ZD8|GGkf0V1#hTB
zk&QAI3@c*{BzqiN^|dy0g|+ZEO5jzQnmZa(OGvR;CMgDT)q|q#{AyOj5}nd>*dW$$
z7!{vYD8Zgbd5nidE!3sTizE;j9*V`zLe^(lwDlwx8KDyd7;sNQiR3GHIhtG_7r;ik
zpN(vJB#<sh-me9pN<y``Ui`OuVx=feU^vhi4lAMgpL>lJdR$rsd!<{pX#+}IT!ys(
zCrONc@l;XCG7a40ryr@F&z|F(frx0}{j5Qd#SjNO&;mVRJ?emH#LlmGVMlM)AvMi`
z-L5)uM-Jd$a(BkHA<UE%3eU@AgB>GiX=<L`J6v{lpf&fln)<b0%Y(4id!c2s9{2Fl
zh9jjQ_bG2y?gWw{vNQ&$uo$537{U=LSbsa41wrA(*t&{R<Wf6<U7j%~OS~rehe2G1
z658eG^!(DG;yzivZ6VuaHNw)>>w7BsS#ySb-B)r_-l`Wb))RYZ`(G5_@&^+JJ1<G~
z^_$^VMD;?YU$F(J`}u4htguX$JtkWXTWM-1y6D)GPlZk?JKM}exV<|rQL;-q^8<Es
zK!jE`fQf=Z;z%hOjJe<`x>4YJyV`3&9vXdITk?pkA38MNi<oAY(gjcbT3?nplr*t&
z9aJ(G*Q<Pj844psCXeUuA49EGz*sFMr5fPECv5#1okzHM?Nc_Zzvy0?CY;pmCl-8X
zF(_)GHltLIR5dYjyPm+(1r-%I<T<qmmJ`9i-TbDimf_N#B*#j~Ck?|I4fS)<-lpGz
zCTM8gT-&E}WKsg?#k1chhXT`g<F31+t_bBiq~yn08=O1u<-GR!IqW+)eP+4F6d@>B
zZ+z|Rdk*|hW3`yatFF36{;qm+SayZ8-s8JD-aFv%L_E-~5M%dR>qM|3`grrV>)w<R
zItyv1^bBAx^{D0`p;V&s7pgRLOLUwZq(c_r=YykBL(zHVBA2$EOo%a$_=1m6IzTb7
zDp<`?EqV3xiVq#MzzeXB#0mp=`q>aia&$Z6KWe8px5XWmKXj96L7nIwL(*^w5#YPs
za)^N)kq(K>)Bp~%VPP_QE&7ggY?j|YGm@Ri_kNna*HV#+h(h=U2ODOEuKSyRyN|9y
zdaxxISmkPnr}A0*<q3CU-spUZZnPNrL-TbZ5kmn|3NB&pg)EmZNYZb?!-zvyCZN!X
z5RuZ6jqHy|Qu2|ah_wRD@e*aB2Z~SE*_T?(#Ij*_I;}r89vmJ$9Iqa;X}>UlyqmaW
z7>Q>oa%vE_d&<U8==Iy>7PAgZiF8Tkn2*3}lahmbKYu8TVu0)5!23woVW-6d<4)iq
zRtn4*Oe@?{IL<fd!ZI7$MRc`e(WltJ4Z*A-X*oG9`dx=(3ul>85#>|{*6jfmXg=o(
z$rc6p_*I_@CD#O%30Nwm$jEoUr@-G`=>VPsRzK%6$bXwzb!>QZ!{%$*NhAc+8?o`8
zfSh{jra|(J-GkM9VF)2K9<Y8&e2n?IMim8mMQ0f-_!Dwuh^;Rv`S&crZ!B-l;I0B?
zV}{+0kT%KR)s)d|MX7>KCS5?FElHt$hi;Bs+i9v`S7KHlAC+;R;sm|%NIuua2ncwI
zr@k_cG(Lye#QR$GG<>lRG97c;V}}Hx2Ne6iR)j4sOkw(-bqU#!<%Khrc$=D;=f1Mc
z<yzuQ8X;AuY(UvSRSBZOofrIZqrdG-jZSn^K1_3V(k{%Z|C*_OD_4(COqrd%IKo;B
z=W%7=Z$2M4@N_=Lmp3cofbhC(7`RS3s<?evX};=}Vf>PnFmbw#Liyo8K5_K^4}Qt`
zpO)d7IR0zDT+-OE`^OmZT~WVP1Q{5c-fk_2eMG_d4T532NU=eQDP8+(REfBvWufkd
zCVe8hsp0oZ+#H4EqQmoL=9+Vu$(laagdhGV1AZ5u*fA5NoO4brJ#zsw87o6Y!yPF%
z9+d#*kyPW}jlJLd!N}c)zWjFl^hj?PUs|Crz^E^zpn8w)$QYtpD;!-gQ8Q|JK>UP}
z=LDj9VXt(lDI<Il*ONL++5nSz2dzr8zAYoH&ra(%WkqbFxlFxLGgVJ^ZoJc!#;Cqu
zXVDKseY6e^Qj`G^q+rDa&q380<w+F$%Ft7p=Tmo>;<uJw*o-V(&UD$z!jKuB-xT`Z
zLtF;hXxPYvDtMlTo!Uu^S}M~%e<UN5FvI}yJE4ey*ht0=3LxL$9h+p8nm~v<38AT8
zq#KCZHA%*Dof;anQt)6+CNPcA@`>t{E>j{9sfI=3>hIP+$~CJk1lX(HAslJ?9)$LQ
zx#snR&A>lWFDuI~7$Qb*MFrs|1n4hp7TSQgO3f|pG~Zh@o>!h*-XW&q-bhAkCQEQA
z^M>e){xi-A@<)mqp>!KJ?J&J`=$e~9p3i=_+OB>P@ykgF6%i?um<j8)sq;$3v^d{P
z3t&>LT?}cf0T9y{!l8`Oc8aC_NVK@b?B+t1rE;?5u%o|JCya2E)0lU)6c2)xF~2of
zL~Z({3j0L~#k~{O@rrZ!^3@E}^0iLz7nHBID0fMrIfGoBV4H2}EBdXt_+&N0PnI(U
z-=#i4V>;Z2QzZ?HSidSpwF`JV{nDAC7l8_G=9ps25+rw3Zo7W5b;<xc2cYMPaQ{&j
zqYr~&c@v6D%c?m?VH>YuHbcZT0a00CppXpfaEN;!*c(pdNk~f68XX<mkR@wQJw#7E
zlvA%lUMGUKLpn8m?V>z>5NkkH3yR4aUT?Z^VUPkRHD|oNzZn`bYpGW2&9SZR@Q@Nd
z&l>i6mOsLV!={s_WOB(SRZk-O+ElfA`I5jCVg+Cj6q41LHxoD-RLCMf>8N<#0?p{k
z&!Yt@UiY@uC@ccCIB9#vZ`86687ZZtW`|HGpG{R;M8xS_Kk4pp_LK%?j)N>MZ?||Q
z(@30}GxFOl9Ea-_$};dLGVtdx@y{{wj}J-IOw6lxexx{YL&MOnbpi>S{kd{mUa~6_
zi7Y>G+j=--AyiecfR4Mvz@ti<kJLaE`cQ2Qx5ye4P4>*%inFk)y2PhROFlF}oMQQ6
zO;*jS7<C)_2{v1=E6ltzah-$SJ=YvUkyK%EQ>{(cT9&7tQik2y5h#l@ZUDTPw0G#0
zm5-kB{QGTM)PlYhMhdbrRy*BE>2iO2rk;&Y>~*R}Xy4qmXTE3n-U9u4$=kJRiY9i(
ztC3d4bvV7Cge4&oi{erCFnM6-XeaN4?#4;pm0%$=y1G0HOKIlMhlx#R;^yN08{-fO
z9)mdncCL5K^kRSUSIwVzOF@d~O;913%HZP0dPAYhtEHxt%;FS<iCLF+ve2At6(Hrj
zta5{=2X_)wSMjo=pic=L#|*6q3cOkbj=QR(tGw_$ueF~=ir+?H&wXbPLzrAHAiivc
zxz(w!fg*QxX851=$AKkbNBAjDoudTS`1gVjt0<M>dct!p2eV(E?s3QXSY}dbMQ^;y
zP5yxKqi61-I88+=v_rw~rNr|Kwdazpao$q$?wHX{Ymx9S*)^>5@~|HedtN1+|D>${
zX_JMK<G;p?TxrPKZ*U-W|4Wjj1ZAAqV@KpIj&^AsBEBl^p-CciuxV!8Mf~fS<>CJ|
zosdNPyCDZC5FjkyQoIMq9NC*p2K(p-`)GV9*ztT(q(=AnGvDy-UP{Bf?o<GO5x?qb
zX7keU^R^|9RA2<&?@$3^BvQox&Go&`&w24&jGVc*m0M5D&Bx7s4I=Pt-|=&F0ln?O
zb>aW+^d76gT?kc#EF8aE<5*3B_tH;Jfyn+=2+KDS9U#CMFI`<o#*|+`eZVP8&Z;2d
zq<|fBqa-r{;Q4CD`BDD__{;akhvJYDcg!&|9ksc_kUZ)@NEnRh9yR0z-<de@Y7S0?
z4e5`iT10C&kl_eK`!GnBqL{F9P7QHAuK6&g)}Uz=8+4Ol3`7il?wy<S=}cp6<^IY*
z*$zRs@`7jsi|;r1laDIUTYPMv={I)gOo@V7O^1?fpIy*z#Hf*@PwqwhN>6+)TGGVF
z4tNL#0x-a@v%2jb`f*^z-}loH#SXhq^kcg-;QZ>byMc`MS^Q&Z*ihzq8$id>(Llcn
z(9PF!Ebwvlwi@zI@vC}K+jt)1sx9|#x?Yf@t9>{U&Yc<et$JvT`noLFNY{rVGAGkr
zp@s<w1hVUns*#3S$ZB@87*2(hz)QzDsI;`NG)+!A##W|@!Pj;g4vXPd#KdhDmN1u!
zR<)*CTq;0;F`!N-^47=fgvnCD4KW}6LkqLZwtOtv>%2h(*B>Cs<1{Z_+H|NV!Ited
zO%IN<H#p<c<7<F8=Q3V>BBkts`dsrMpCyxTyF?OXIjLya2&tc)0jtv_x$JmAi0zFq
zZ!nH8WJ*3;ki2XVcxwO?vK9uNjqJE7(jkq#T{pI!Z>;9jZ8|GOM!Y+kQh(6pVSdrR
zTWj#!8PA2pNZ&PL3ZA(t!_c#Z>o4m(4{b=8IPeu8jwHlk!EvJ#A-hLbFoKL@G#T2J
zi8MIq2#lm_5gfY;ThK6V%b&BJH+}T%=1nVIeUFZfbY=~zo}Q4$Kos&b&`XC<zI^S)
zg|lhnj+pMi+Iseggo5X}ZIisTGs}T`tS9V_X^f1ZiE@9@S5Ge<Tp9AthJmCZKb&*M
zWx%)^`CUMel@h1kHrn2Zhu^0B!`i)Fj)9E{S%9cxDC*E_)3aaNK}v>=3F;2hW-i=~
z@1X28D)4=NaA`#@0>6!(B+KXunmYZ#QzVEls_>^m*Ld*5d#Bk(8?DLJ5?vf`&jZ&7
z)jq0)07PTxCZ<fbaS`aKA`PBBNTp)&Vml1p@1Uxj@x=nE`DfLXinS`4D?if?)r{p$
z!^-{(+LKCAua8!eZfoe4O3I`rpi=TeYyn{K_7e7{X5HUA)>hbP`&?Yv(;X*Pmc6pl
zKWX#!Qc}j<d&sA%)kc~|knlZp_3*>ubrLh`tFPno1kltM>tJp}$%E`hG(+bw^Fv53
z^D9U)1$AcxXVzYi8Fa>tMb=hsQIE}VZ0FWqTKftQ#TVsu!iy^|lJ8mgH%lwn;%MmV
ztbd;i>S@Kvc&FpNk|a$>NR3oV#_h_jeOM753j)$o2x8VnpObI+UTeSjE#z4Q|AXQ&
z{ij-z<-e~br!+MGXLhdNO5Lt@1r}_-AkyLCnC<0aW(8pMT=>F-DM8|QwnDLtQp=M5
z&X8zgwa>8SyAkeR{8e3_o$*y|&sA^kcTYAiS1E_fAxZ?x)2TFkSYG@PD9B9!*>=ec
zvgiRcV!c044t_i|(Rbbf^wDZB$H&Fr$+(%Fj4?ptj@&f;v>!kF!~*}ce`du}*;=Q1
z8Sq8c9;P9AAFQ!xIWvjP7f4y3A6VI-Ku&J?!R}fJ4BAjZA}2ocp8f<)`U4Lu#6`8n
z>Tl7346+a`Ss6<Z))Y@v6e4WYqB1E{yEr8wZgI;121kTtb#2X(u@llNoQfFSSv<j2
zkyBDWqqw>SQDwCQ$%uM?Z<FE5uD8vtoqj<lLLpS4zO{_O5MnWqBG>03T~$sN^29uH
zvjdaP(MHnLP|cRO^CcczyZJ@G4%T3r>ceThQUa$@M9eGE^MZ(l+k%9J8~l4i58c%~
zWT$5mVL5%O!^7~<&rqr_f8|wLo-0Ebw9A`*<iX%a-_`NLvHQoq8t4HI5Y|-^7#FN)
z1+<`5?7!O=Ow`F&#5Milg!k6%+ZHQ9tGx|US7oA_nU&S-rims)<Jnfjf9bDO!cpRC
z8?5oDGwcM!UXOI98+oi7njlilIT{Sr(N~xtablgeNbtlhBRr6>#3wv+)K(hCqyRxZ
z%*|^yQUFY5o5`!{vz#@WbhE#~xNNm>P7lv&<0TZ<WTQHuhU@LnVKR<}c21@HopR9-
zK4vPfO!}#+%RZWb5(VKy`dH-7w~Jbd<%R$!bEH4NnL2MS%j@z(<gSo`C!tL)NT?kQ
zohuRCohn5dj4<7>ozaAWEzRk%p**)tU_BPz{ceu-t8_}(w$Byi{2QW4b;oV(jE^TX
zmYlR{RL&PWMIapyu`r#MsNZ^e0)WrNbuT4MNZ`3fE0a1W<k7iOO7Jr-5wCk~uR&52
z2q7YQV<-TU4@U|tIyIX;m~8wlpl>frh+tUPA;G8o(f4?e>@S8Mgb<j1Hj0AO&OnMR
zOw@2Pt~>4+OI~ZU!NPRlZ=HcA;+@Io!Ggyrb_gZtOgLfkhrx!OFhL>t8P4mHjHVHc
z%a%o*Oxs5x22iJW+izSS+<5u=Nx-B_v>hFg-WhUQ76+6n2zOA9MuA%D{6+Pwg=SQ0
zRw{oh#SbFCRWcoE0aTr=524)b>i2L)yy}rMsEwU9u2M@XWXA%UTnt6^0=VUMCj>f8
zYN=l0p04&In8Y>?ycn2I%Aw)!EiWEG4pO`92t%KW!;Db}t5Gm9A*a%<l5@YuT`G_R
zw&F1t2AZSds<Gc#Ss_h6*jQ?8f4wxx7ptAx<w99TYoWI&)S9MVQ9qC~dx(8r4cFtI
zxuAoApjV8mO<VB?&Bjme<|4>!C@t86ue2GSBk={9@p3s@4JDH*tFN$tS*z1a`PIpX
zFJ-&G%f!#xP#eqo&eaN+7a<6vN+DsS6>ON}uSS3gB+GV4RjpU+5Q0F$S098B{{!8Z
zRW>apKB_g|B&wSHVq-7K)DeC$DY9waSl*#Ya!GXehh@-T79NJE3?H6jdT#i)4PDB|
zgc)EY$oUdx+<bri&f@zaQ<1!OYPIMEJ}HnNA-wfMvrN)FC3H$7aww>e0aGW{eRMkL
ztm^{L)TOXmc58)XE>_;k?E2$<ii#oFr_CrVNXj+I(I@*e4}okv^BoycKev&2K4?Vg
z`v_(!>p9169TASrrwNZJK4#@0E11s5V&Ci6MZS4*e1o)3p+B)d)XH(E+hFWh+w+|^
zvfcvWey?%r2uwf|4P^2QVyK{E=GXI%umPiBW%$`RH$taxhF~CX0dJhwK2Y?_D#@%8
z1Tlw&!wXa}vM>?PS*=GNFh{eG1a7Hzl1qj>?QwF8G%+7D!8;zA<CQo@6zFM0n`vLu
zS7zC$GHxa#7q*LDeaXWBD`39Cxq?@|PZ;_2boU1w$oU8Oyhm{VKRLvIs);%N`|a-L
z|K$+>lW59oOvC~0aDV8Q-IOz?C{B+@kz7DFZYG^fCHZz*cM}AA%S6U$`JN99X5@J3
z_i;sM;ea~3do?CMk*scC3}u8O&I6MM5D*DWLM~7OpKx{fMdkc%`nFNU9y1)QNuw4L
zjk#<=@%1_JHgYs{bQVO7P36>~fDAuFPd_7%WxPkg@x6Bd4rhTy_m6S<oviy5(qSw!
z-3N8VAP0&LC0dvP9)&evf=hE_(0XKhKK)n3K!6IRCIwY73W{um*!5ZdJ(CKCKdj+{
zE#~*0SZoP<r_qv$Dey~#Mg||7YJ-*4=SrX9x?Fu{ryTWVIu1Ll{LTD;{M05$q<s~Y
zCl~loSUzI`hYrnyvhK%8t=+rt?(P<CO-?y6#QJlLY`gB^nQvB6!`HO=Yb#y-6x-^4
zMM1E6pFL~Kg-a^M(r;$N4!2~hCiF;Fb*hY(CD}hW3r8C_DIFTQpT6^3sz~D+gs~ui
z2m@DlqY=?!gC6J%Fz<_eso<%xn;mZnND(O@r2rrSHbsz7F_QFY^WoX#7${VUJ_J?9
z4lpV=C@0J=FUaQzO86arM{M36U56zp70V>~w0}=2SAK@r9==-@p1EM@pvoZX7J!Gn
zP@aJfnf0JHk)8J<i83A8WMj<ygw$qeQIaDh`H!|H|HTe$xd-{(#V{AMh=}r<(zGFW
zQxgy-a1h5#B=T^)tW0Mj-a|%HMoxOTUxd^{X@dkZvwSm<JyKn6lMi*|3=ej)RJfn|
zFO$T=CG)t00LEa1*&!7j9-zhlO&aKU#W%^Ps()rENsLI*-!t$!Z&r@FweQ&O$U~9l
zXCQPiQRQvMN!#ebwS(NXS(#AA5U}%pA30M83W#vKbmcKaPR@?c+;RzJ1HBtF0>dnj
zLWE-yf{(UsmsZh?0bg6=-p*Da8bTUD7@9Shsq!sC7S7yCOhP3b7<={&7w~fa3K?>u
zX{35cza9f06}6(#-%0D--1=OvFj4o`Wg%;!L+=o*>u`mD(dP*!4xzTz!$_zh=8bQ<
z^6ju1ABSJ3Zq4Z|u4qt=S_W7JK)^;U;M%M@UoT-4uU1<qgfR0ByP~;n0#oF$QkaX>
zw}}@PO6xq`{fm5d|D}a|F8pA+epv~kn%ssim%iOJZ4c>-bZpR4DA_<vZFlL<HfLGJ
zdfhZUII@Cz#4f+#!c+1QqWJZ1*%w$mO_?Mz0(Gi=5Ggl}3rM4?O=nN}fdB2v;x&P_
zb#=pbi#^>n!@OP0A7{SPqsng$f^ZiMikRdeHxCAHn?f^J=R}u=O09MRItP5z0PmW3
znupFnYB99t#D$+oTry=|*dr$29gKQV0UX;r2(y-4A+l@DrpzV@x6?Q2Y!dd0VeF7@
z3|GRtM)+k;Uh}AvRcNxbz|+ryNZ@c^GbGel`GOdM61{<z^)JjzBe0~xfp5oGS&s6^
zTweu-l4KP<b)-YE94vdM3M_jsxf|_v_9~!Zp#G7uedD-dgLwm&SQ=PT>+i^jKaUlu
zUBI$&@IoVz;~1vyZeJP~xz_m|b_B-Ryx?BIL2{SyStt>)(ZNvPG9)TCgnQiCA((KC
zHdrrEjgGWL-&z+NCcp5w5OuI#9A=O(NI!X{FP9N?#Hv=r8|Y<~_A@@v^D-NCqHr>$
zc@%nKy!wEqM91D5s8C_GXu5s2>+j`i#ohHq7vNEhQhDztdQ0uFca>wlF?Y%2Wn;h4
z6z4Qi&B=h5#n!v>_duR?bXa<rx}BLkTr_N9a;`o31FTs%i3N^%7LwY9xC;nkJWFi2
zKf`5{3~QoY{YK74(l!YT`Oa5mX!6W9<5-7c>%en%2JJFuw_cEe*b4F5P(-?4rr|W#
zw$b)r&j3JihhziqzJT7^zpe0neAr8kAEdn&+VQg%I*2vKI=PJz>{-3AtbSWUqNlML
zfw}qDM{?Z_48&E$-?)5_Xnxt8m}nMG4O93qH*!N2q(QMFOzx%W^Bn0}lvHBYKGd-<
zT0tE#LH*&;ctwr{0Ut-ND7pO4Z{tD^+c314O$sU>m7Py+Z^n&v{oWv>f~=5BU<9*}
zx5Yru!|tNQqmR77iVEAFd1QL(rVV;zuEx%(gFFGLxTp%y{c?j+a^V1`nF+{rrHenC
zrK<C<GF&j~NX#x7AT|5`G6oLUPpB=JYAp;E+GapDui+}y8->kNWhrj$ev2pF>m`TS
z7na&Am~)-*2Pv56`ZTt0mg<8BVSKjT5C|abRm!+{`eLDorcW5x9hSPVVGO;&a%wj2
z@NTrrIe+g}sADv(Qm_W}yIhl2woY0s0(*-pfsL3Tx0HoVE)CXBe3#>{xQ+(PC%JLb
z-LL6$*HU4;t@s99c@RXCcNtMrQ)B}=0+dTO>bwc~NOgiU+vjhp>1j6P>Ss>&F>h-}
z=vFWSgICh`H4PUr6a{8dalkU(?37m4WcN^U^MH81<^3)<qL0RUsiV&y{)0U;|EC6y
zjg9fYC1h=B>i=ut;QU|8W{Z5xnLkOS7O&nD?5yj&HSBZs-R$hjG#YBEXgt#{Ildl2
zp{0<o{#s_aT)i795kdz60e?a%+n*`kT<t!&INXKJtwu}Y-p>N)c#yrAprGN4*=4&k
z=_3gT>)|T=IWu(-u#CU;2N}~B{dqn8-zH|#{pEJPzTG@nzpnOW3EyK#$y7xCAl^ql
zWdk>J54Om7z?g*#&EU5#jPMXsAvh{FY>~GPI!aXV<LDZiM|nFAdrt_94Jb5w+cr+^
zv@OcGy2GGBYp_3orn*GMAd=WTH1touXDAHVxXmbz#01__g&<IXxG`5#Xf|%7l~<Xi
z8k)yWJ<+zft>!Ab+uxgocH89H(lSy5S^$f#Jv5Dr_cecbk-SwH0Gw6^hlL^zz)HUd
zjN_Da5kwDflDi~oVX2jK(w#SsLbl5q`3RPor!GN{KC^XAB-*0%SLq1y3@Nb_BE*&z
z`muz4UErUtE{AhX*iYJG6cAua@E7dx91N%p4d5@gRW|lCp!UqeO}c+K%=wZVx;a)y
zz!0YknoHaMUbP@4nDvy0Dn+x{I%-@)$h@y|VV|+0RotbZ%qTFI4AfU!@Y1ng-Lz|P
zMu8=lL~G;Ti*0Uot>VmE$tcYKDpCe8V+U4->eXWWX)80XBy@4g);Y;*@&q}P-U_+G
z0~S(3lUS!1k|WSTy{Hx}Y{wG-I!5vahyQ#j!PC@?`}No&p-pD$wd})#780xw=TQ4K
zt$B0p1MDN29&61<E2q>FRd6ZCNDL8ti-{4KXC_V3b0M5(tO;=(LF$ghRl20OyB+Lu
z{|D-Hz%(#zNh!dz#@n}H+jJ7+MF)2ZGH2SytA|m4rahAlKX|ll-1;!#*cDW6<50D%
zI5!xEj1Ww!isTBO=l0>Lbe%0S&L&9uhYLW_$<=5aYio^5LoL0kf@CX`J*%&Uzhwk9
zP$6lV-crhdTAkeQ(F%d{$H^q89st?a|8K%6dV`Bh0+-^8Vd0T~4HEm-fWdi>p{l6D
z?%D2Qwl~rh^qRRS8><{`lZnPL@0E1_P$+J9gCdi=@oMN28*Cai|4f=1B|_DZK3Z;`
zv-gChfFiCVb<?&**Bu^4Zr|&$RJYL2A1p@)%m=AA0n%1H4yc37q7#CCNF5OoA@vw$
zdq5B7?@^J89-su4<C^S`M{64`kkN`cEwvI8U_j2#=Ph0q_Yla|6jkB?`lbv6aHK?L
z(!{IluPwz#pH5|GY#ST>dwt@|M{zC`e6y$F=}okyj5k@))W<_e6tA!gI^3sjn|1;I
zRwgkOSAx%L*u3_BxXIPm5*FZ6nOv!7(f;p;egqB`?|b@Sq!FgDi1c)~0ZStfby04a
z#)}&m^T-AKS!t89O7z~z-|Sje%ww&c3jD=|_;GTrGYK&xA@%erleN6Y1B3p=YY$E@
z$UUK0l`lw!juHcX!~l1QIT#>a;bZev_EA-LP6MeB{RUqbTOc;mGfgx);B0|Pr<S!H
z(9RE|`h;XkzN9Gu<5f;U&F8f)#t3AIb$_R%#D{ns;7C;b3510+k|YNpMModv(RN#q
z0Ns%d4MT!8SbQ89gPumEkE3!{DYabRaiv}Fu0~Yj=a$d@qVREi9+a?f=OgD2JIS~_
zZXPbrOkk9{sSVS!p2YQv2dm`Pn|!xw>XOH$!`wvBWmBo9X{R0VBy6j7$9r8MXNW$X
zkU2(%aB;U+-aaRED&i9OHup+@Ng!*xjc@Hjt2-N&uV1XmZKIp}6^DiifBc;aG3c{T
zGlTY$i<?JXa|UstaeUk^^e%28^yn^QJxFiXWwVJr?@41H%t?xOfdSoKR7J07D?M%B
zFO9q7!_OweHg)1Cc+eZVocd%(D9SG@fmn7!FiI$~1Nfs*#yHQyM*yi@BH2WRH7rOQ
znQvgH?L|$)a2L(-CblOHHZD~$6lF3L^)yzHG%t0QS6}1W>T6wXMmY_)P##*P$W(jf
z0hk+vQm{xN?R^_ND($8LB{i_lgO(PV)Z&W&LD@HO2ih)M#<p$Swr#6pyJOq7ZQHil
zvF(mKP9|sO4Awfm=6?763vWHOt7=#6U49|VKD~lzEH#!woRtBTMzvVe72I4l)~k8+
zQ@f8zs~)w!kpbWiT0f=7Sp2Ir>lb$?FrZdqz2h>Si$-Wwtd$`mv0&coE=(6M;B&^h
zBbnMm5uMC0S*4k1XRWBX7LzTyw<l<?n>DqIseq;r^1><h%qqg)-OaRzyS{)<M@j8}
zL$klFf^jhZW3}ZgP3eCTFg+*gr{4t3AbYv)!>%sN8%NRBuT)Vil!$2(OPlfL!=ba<
zoju`}qA_<C$P7@AdJ?{nLAwy<ojHz<F#CtmZM%c+K;fAJ5YmJRb1ytbOh`J!V<Ig1
z!bFYz{<G)nK;eMI5sCx|5ltYb&Agp2oNoMHUObrczIuVc{5TFgT>MzDpt$EBJD<Nl
zw9OTG@V+m^PXm)XLIbfdNF$&SopUE6&DB?KpJqI`x8FL%^)+$g!x)1g!`p_!$f+`>
zo~*eOfkrMQm`F66Sms~0`FAQ0OhbbNxk1O`^nH{nk2o*_1uDU^3=I0xk%yc}AjFlZ
zW1Ty6O4w;1@@|LN31H6O@MYhLs#A>EHJr?@`3?Ul7~)LU1T7NmeaeDCN+-^63F&yg
zUJR1%nM+U@2JxdaT#onMkW0Hwk8u7Rsc%!Zm%zfgv)<*}ff?8wCN~R@)!2QQpXl|2
z6hKt+Opas7T0@QqOj3iE;N&P0VwWSxha?jD75{rzA3zR#a%z!&<a*UM>1Cza*S@pJ
zHyb5#D(SPE<7u|3*x+dl{sL8d#`_P2IGMcIt*R79HSbk?K$(!F`Z6+4fP(~DD=!f*
z;hUZy5{m9o;v`Ei!{UjA3%098u@y#q9alOei4D^MiXu$+&pBHsz4YSgUX~iVXm3IX
zt&}xUK6ZChu86pOZ6A$Ni2+8XpqAmB7ffzd!W1ErY{@rV-kOEAiRYGkL-$lTl#Q06
z!ol$qzo*ojtWYx_k5~^CVZC6lorKh6J@k8L(7jm;H+H5d%Uy6>%CqsnW_=~%ViV8L
zN7B{E)73VGPE9B=L5(?p@|EVhtci=LgYuU!#_4IN9pCFdWyGW~SgW_q!dxM(cn^V;
zC)-?t0%zWk_W_R~1ju%i#yOu{j2NVD&9lk-vVmN|<d^$0BD}l^X4MS&w?n}Yla2Zp
zxMIB0#4dI6Uy`zr^+HI5Wzw=q)F9=sdKl*7A2|LT=MA6%djL^5gg-O0P+mx?WpdOs
z=PN6Y9FfW-_Dy0wFrXYt_qwBDOCy8FOLN(>YN_QNNrp@>*hFK6Bj8Gzy|&v_wmjvU
zSMm5t2!id8;CJfF)h0J>qgMK(oLx>|t$SRJYM}W?EVN%os;KkqM}5u<oq5xjK5)o3
zJ|mN(H%OB-1Yj@8tXI5L)jqFiAf$id^C6T(%L-@`W8P<SV97Bfs0Oq+H%VK0e*ZX@
zqx?PL<YPjzZ5pBXmNGrFB@>Ku{ClVDY4x994!=@LXjavnr4Ht4syY-4FU5Gs`X(ii
z3O=ZUx&XqCbkpc2=d^_N!!=pkMf%C}BAog9v)_cYN?~wh0}BzaF$)2e<H8y=J=mZf
zCuwqn)S*tyLVu2h<m|RbfSz#B1OQHRWz(CEWINWg9=J1RqONHA?D3Ni|9VE;E}Tr6
z)OfmCBbfTVD$}G>@rPYNKV4R4?F0y;4H}kC0TsQZ;Y^H?<5`%_#7jlRrI^G&cyXe)
zct+`Fp)To-?IK*Wk08XVQs7{j>ITtzPbBsZFx>9_)LpHE4w*#PtWk62l?Cv@q3)AJ
zH@PGx_0u8+4ilcerF@-tR_2PV=Zy=#{|x*GqXN`Me9?6TV8CW<uTh?qkc8Q+s??&o
z0)H7f`oM3J>Rb>I-*As@Gp4(*9_km1-8UE+-y;8EdEnkQf6X@`S<j7X%?q(=NoU$4
zt(6rRtM#sR52fQ((dS@USmCmtl=7A>^sj55p@y2ETELGdRJkgwr56VB9+@ZMB@%OS
zC)h%M*v0ONOLrI?eJpbY0<lsU*E^dgLb5?AxC_Zg%Hm<`P45j9(^uoHjO^(LVmY3v
z$PE6fmu$sEQ13B8=UdfT5P7$mm^ZZ8hkA;m%>rd+5p(S(5il1V+`VsPA_NIxe=5|y
zoww8^r7)1Bnors?flJSfO{Jkxxvld_VR_3}X71%j+4m}I5c78SAZD4~OufjVwe3X(
zBWECsDBM!~sGvfrqx{;jM<5pFwj=b_S%blc{oAZ|73yeVdz93j?Jg@mKbzqJ-r$or
zQF@*CW}TIF0?HINs1TnZ7XUbR#{Apsq={5zs);N0Xx)K+H50bB4jr0%R$n#+28snI
zuhA*?!QRWA{Y<87G(`RFbI$D7T^@m&MS%Q@)??9;5%fl4d10vCr!1n{*4GNe=O6iG
z3o1j-VuQBup}m*M9pp3U4-T;`WSnOHEv|#F-Hr>)NKuXSnvF*EoASCLsV@HtPZ%~F
z;8&Jr-w!8bR?nF?UO$4KZVONMRj#i+B1-mDe4i(6iC6vOxcWLbnyGFqWKg`6P3ZDt
zeGNQ<C0GgJs_nD5h`!^)3xSpcvk$-3P9|r1JRQQC>j9(EN=ksNdv(CE>cA&^0b2(|
zih?g-@?V;4e?!{8++VUY{2OTh!QI2wzwaL2sGXAh!uISFovb6dB;euVD9BeRj58&J
z*BuEtf8Rag#wT_C5{g>&i_U5@=dgR;wwF-6UMt*|y*66!{0%JI6+{w2N$-8vc2}ej
z1a(%dY<DV6B#9s;`Vrds<|pps!QBQaL{`4zi{m9_cMS&LPx0Znh3>T&T@-W&p%{e@
z+AJH(r2_)`d)U0%%?6k4m@rZzi=vZ)dh&AC9wjJ3srT2~)INhgwzO(c{RpgFGezo7
zgQX#Dt00JY*zz59@`EqHqV^v9YGWw^Kw%WB3&!5*28V}$#N_B*XA3Ux;!j-~YHlr_
zH5Zf2t~y$=@<<Ck9_~sezt!CHh2WU?%(-B}k|5@oHbw);#x{IQMm0^o$r0vI?W5r#
zXUzOJj@1WXPmP(#BfSKF3TKXSN%HaLFLjQR01*~PoJeK|81nrz+|+U9t$s%t5S;Xh
z1c+ie;h(t2?(TeasXG$)fa6`N3^OBlCMr=lM!BaIm5xJhXCm{6oU)Qiz)JAv(ow{Z
zG9-Fj;&x9SFG|ZSOgD`s)!iF4ggDyh`M6P!cd5%W7#YpV8}$>6At0Yi8**xZemAw}
zg4S#`nXHj$qF=Vw(TQD<<)|30p&RDxg9%|II_uL_&apxAa?V4~q>*3ek{io-&g`vo
zZ*w`+NcH!YvvLLA>glq65bl@bCsrc_BjO~Hu7{}u6Z9t3{4O@tsMvQ0hgv@GK(^8G
z_@g!ElVUzu4RS^Cx!5lb4axZY7%#LD@jI&>nA}doyql`{L|%S@%D(bv{1mXOqRD>j
zCH;b|`pnR2L{Qt3Xnq7)5eE&sL&JH(ZY+vji1kwu_-~txeMNF`oVV)S`|7&#w1SYk
z&b^OTY;NhX@!8elOU8^}?G#-2+RVDD(+4y^;cJ)bu(BVbmp;MSuTGkt;OC9ct^e{S
zSpIs1$MnBODgWS2RI5tcWikA#`m{GL9=^qXn;UGAdgUNl3XyJ}_P_K%RzV}uXSeb7
z2IoNt3oB4V1<$;FzGt_?+5~5LcKgG}$=#boR5xT<6KISNflE;inM@_&(ogow%OTT?
zv)A}nOjDX_5HvJK_Clz9W$pN|2cm<huqaym!Z%g9QVD8A^L~7^@`GLgVUpp+%iDRe
zAV;wX@J0p%Y`58buMm%apOI1RO}(MuR)^z&*u)gVI!-bfjX&-NB%ZloweN{rm8H7k
zVddBb(L##*)b^)V55lj!NuK>7`S_wBN&PVn#2b}RPiLOZ_K8gG-ya>c@UbR)5hXbj
z@;>9b(csg<TD$WlJ)*<}FxiL=o{32Ppga~3t(*MqT45YK#hN+WM>9Wa8x141<xZi+
zwb2;Fs+77!MFi-=5FpCJ$P&5x$qZ3=>RrgyzeQ9+xA7xw5|c%F4{)Stx`g+PY=k5+
zkvCpX0iM~T*_DH0mjcr3cucs}J;#|%pEi<wwgL)Zb=Nt*a$k$><)3##7VNI=ukuEN
z0IvXRQ@8eJz$1(VL{1%DPX|c|0JS;;>QRLdO{!Z98U=yMb2YbM_-;kkvqG^DqC!$l
zmM)4zM>lD^^;&W#6~UZKNiGr}SPnGGz^XlhtNyHVw@xeD-N3o;N}l_EXl4}FJ)s>F
zg`HorlSOHGj4Lnm=rfhP%sUH)@Wcx!3`{`AJ_eB`I`k1_!x+&j3ZvlDxPwcixTvxl
zAzMj%GeppO@-UACR+}8R`<&q<5Kc`1LeK>=3WdcDnjp{KsU!_~1aWI$G*mkM(k!6P
zW47qK>k0e7EW#*(Z5E*p+!8dj!*X1hc0^Q26^r)#rLA-9T`N&_I8+pd{U(OOP~<l0
zY0ZaHV1sstFWhg6`BNB#rKmRyD4m*MOr8Q^xI^S<-D63zQ2v+<g2cZV*ohT#{tyt+
zRnglff`mg@4;vG#LTGd`zTRcJexN7zD8f5BhT2&hO*R8nxU}C?j9mA^y$xoxhkg@Z
z8~2iTxQ)_H`C}`eb|8xE%j(mf5tHo0BR=Z`Q27w{Z!pC27keEx&VMi5{evgBr7kVM
zA%M_xPyOdq!fkc!2~WFRrCQ}N0b6AZBvh*bH;IGgs`h%-DiZ?+gR-fqR3^*aXNSjd
zrUugtKlZtM_UXPbQ8^Zd7UEaLbgB$iNf!7OZRU^q#<9N3jrX5+{WFG#QX)SREVHjn
z>*si0JrZjYskD<b{Vi{zcAL?f&1MY0_heT%M-W%C#B(R8y5&}wofN$?4>umDf++t8
z**5&Jh)daUoFC=m6iu4)XBTTBocFuo3!0jsGKWO*)H?M8!E^f7n~y)Xh)-E6lP(mn
z+OWv|Ic>Kn+N?EeDBOO13JAtbnodq_U|G2;sj#iU1w>Mz&-1mttWaG+0#6j&TeObt
z-VBrL%{ARxWgX3RLFYr;dh^S|CZRx8f=#$=WCCf~kFBTDs{6GblTw%fA=S0HwYXrR
zUv*Ki{st~inra#G4@T3@bWyZzB8@!VeG_rLPE!E@?*zWY{0n03IN_4#E&pAc#*tmL
zxR)$mC+;muC`4CHHnVY4EDR}0FevG`$$X=Q*DP5$Zzc)5^*UhYLz6)x+j-TSjp!Ym
zX?+tJmu@rPg09l$;hDqLxX7Ag700Ec2;{H#KpfZEBvKH2twZ3<s%e&T8}Rjoz(vQh
zt@$z>VZ&KMI7cP4wodfjte>|yn`k7i!ptuO_=SMMSeBq1v4OU=61#D9uX)0&v9m#W
zY%NBzs;yJsnIbW!rgdU{O_-O9YOFkQ?z21DwYBbgNiH$wbNEEcLBXkPbmiO7*=pu{
zW|OqV7D50n=dTxp&%axd9<y_(>|hSFkt+_sYwNvGJm%Zt{Yo>oPaDW%5{r+s`@x^)
zTO;{{;m~#E{YIFMd{_LoRg}|W<SRKP3-0V<`eU(<kM9q-&3yMt%ic>1RuR&4r}oe2
zOjmYJh8G6@S6B&^Fx|gE59{B?_zdj-eDhR&<J(&s!FyKyR1wUOE)-oCZ;{ACS)YCk
zO4R?#P;g^Y!;c0tA&cYVHAZ40r7Oy)wDU?jS7iQBX9jZ`b8(k1ub&>B|Jyj;76e`?
zL5v2fp%yR}h>}Ps7^Xq=hB9jC@aLb6Ozga~U6i7v8xle6<GI;hJ@2uaYcgdr3yk=2
z6)_esPu6hJw5_dAgQp$eww^}q*6SVb%PKtubg&F?JQ8!h{p=SDJKqVzgrfMzgw*Qj
zU_fk~kf7ZGQKs665m26nF+T~3H)UTXN%e~A`zvECMJ!tf;LB!23}wOaI1Dg`f%%;F
zEZSz-vdt>fj+Tq_O7DWN=?d#hX}xu`jh$0(my2p>WPdgTsDn5pD4-A0pyH|5&Wsxl
zF9r8})8NL%!=p`F%={iNM81)1Tn^kGvM=deu${o1R#k5}uYCZ%gG%t(1LYzTIhS}F
z-an+v5?^n{fm!B`?tUGy<hrNHbQ>em0ZAbHZc*FPG<Bz>SgWyl`89aG@AAX#D(`Xh
z`|lx+Nnmy-Hy{ZkA~`d#e@x+6y2^1NJu1~Mgr1Vz*J45QVPf3-$Z)UW4r|^h-{N>{
zB>{Am!5MBKfvf@WP4F7s@SbPr@QLc_s}|55hr*-u#|%d2epV4s5-bcp3wlr|FpDAO
z<g66hNWwk9NhA0bDK!Y%Fv)Saq7e^_oyy%<K%q!qTnPxPWVcQx+upZRXExWM{s|aK
zBq1$Fw8j!3zsmwLIAzh!Z;Y*GQW_pGZPZhDwj%KBdDq}jj(+S1!y3uz?V{RxfhF_)
z{#BDPDwD|2D~})v3YoN>g%iQujWroAb-nBs3-EplRIhcU;RqZl1h^bAX>*Omd(tEZ
zfi?)&&h!=<G63N$?z@@Oz_U?6zkjbfJr4BD<XpOkg8wcnV8YzHL#%J1W=UQ7Q^XX$
zW!C|BY^!fbcEh_&6B8y=47*A+Zym9me6y&<Zd4&|!bG8Lj4*<GncbsKXTj23^913m
z3<Hj=W+#Inx8-%@G_;7=PT<h=Cg3rsT&qGNBH;L^D9)b9e&+sZ*fHsdy}RC3zl7>;
zUjuZ<qRJ8fT)>}qO!7S@afQgWp$#}WZI^nF8B~+^vs8nNTR8bG2GuQr=_?TmXZrDQ
z2k(*@Uq73q7pdlwLn%A1PEZ^Tfn-%y$>43{8o3ESgGp$ORaQ@5b16Ddo`>1!&j->d
z5RUob*s&ncYzVMJfMdV7>+}%u;KqTRvY0WA7`i8h-p#neC}Fg?`V|D>B?Jjf8O2Xo
zKI)uHxpHh2TwsQtpGomryjOFKv);0hm58aMNCL^>nBhnlzUR<b90qsxV{6Iwg=!y`
z<;xwp^g0?WUOJYQI<~ed_a<NDuJQbbk<nB-;EKZ|xaZkvwc_;e_+TNVMm~D{d%H`M
z7nl2XwbGO#f6GCC!D3d{e=G9*11w%j)wNq^Knb~iLG6)gS;O%iX*Wi;mZ{*X*O9TT
zqy%jwR4fzH+TEEQl1t84xQ!QdH5tX({`vlV*bH=v4)oCe$Y{~*iDHP7sKFSVz_<@3
znGXRGo`Bc^iDE!9q$Fs+1WAlY68IU5zziv9HYeYs<DB?O{^a#h&23LexYYl+>*<#2
zW%C45TOm!*N1VACJ}D~$?Zvra)QP@0k;{ml;HLu77<Q_`8U1;Q=2xu)3vZDl`|kB=
z-oTP;2CWWWjAKz*?3Pa|QAIhRf&iy?wgzL<7HOKKTF_1qA2C@Wz4zoiAcQRFjOhW5
z)J3BP(0?9?U1?>7En=l%FdbgkIh2Gb7mO_9gWcLWycoE1yc+HP61k|?20lxa9ldFc
zt0V=yPu&`-qe2(q^~c;AONP}T#WTJJ`HsR(j(qXKM~f_HD#T|3;uj&n0Ok|d)S2=9
z(d?F6u?<u8(<rw3ty<BSP*-;kGBfCgYSZD=({Yx~x>t`i0+$s;2S3jj6U)}@5w}{f
zuC7%_F7#FY+W4QTYOgZsUFnV?bEa*lvsIL7OT89PySrk>#@0)m2e*?|WOv~iLTm%y
zu*@z8{QlI3_`$NTlkGd(Suysr2Nkom9&~)&!I+$o7q}fgm_>_>r9|5MA4R#b^+|=2
z?8aoJdp^@9bid9+GbZN2?K_HDhI(FY8u*@AUt_b1|Gau}2K?OB>)ofRT(6z9U3dj9
z$n2T^OXz*`)&Exq{og8ptl#>t|CGg^PSLf?{8#ARQ+q*f*nb2wrk*9)W;ZoYw59!$
zO9XEcHdFwVP4xYNEDS^`D=Ac{v(_M|mwTTzh)rw{kMPXI^(1D79tDKnk!8QT1vue9
z%^@-crUzgPGc&!j-P#BpwJC<;f><O(5p6PeQhQ)!$HYo&NE|4<*ucru6Du2vx&3gL
z%GcYGBl9?av=p0Ld=8;%LE-M4t8kZ<e5Y1^tq()9FR<t6J-J#G!<eW@^EHycJXJVU
z+X7NRr7Y8oLR+ko6GSBl7#`U?<Ev>ruLF)OgVmoT7~1KqdM2DUPBhKubjAdTsR!65
z%9at_1I0!g6UX>fv_j+ky2$gROrYvW>ldvcp|Z19nm|N4+T*8cF|RQhniv+(YcQY0
z^@G+g#7EpCpBnUZ5RXFN_Xj3>#ib>&SOO);CO=%59P@0jXt-y#P?1Y$U*v}h!2GBk
zf!O$JshYLL?6s4gs3Rx~`9JsUYl#M-K9B{0Tn6Ve{`pTCZlMC9T%PjpXbo3rS+7Hp
zpEp~4H`J~hCdWJRKH1X_XT7!a?2y;H%-~w@Y?&5Hd>xHld^>^4d?P<BGLInL#}P~0
zQ$w3L7C$)N47O!G7*A1Z@iH#JOV4LI(6E)7=9}+nlKtuLsXi#$MZ;7ymxJ!<2G9a)
z6zJ8blFAx)$nS3b&`)B-n%3G(b_U)Vw@))+0w~(;*5u>QS+jyQol(d{yQ34T?S3J%
zm*?PWEaJ@+ygKRo*n@{O*P`RE4q@{Go6V%@|1Bo}mJV5%|8Yz%rS5#ELxj<t7wY#k
z^T>-GNxU@?b~4rMGJLY?#Uvs9cyp%UB(kp$7#D=5a)MqA?+)gUtl7`g3vL7r0FWLS
zzVZGgRx!SteEKi|N4PS8EXXhna7Qp7Eg&RwZXJRPQ}fQLgv49;4MH5v!Ep_rZp`sp
zv+kY`zKl}@Y|rJZ!<kRgUR_=0j@)eGu(C?FmGLdvDS&iG71}f)c~R(&aF+@D(6+hz
z;7<zK9N*eOi<BY{;DGUMb{*8qmm{$89R^sHLvUP=v>f+1gIbtwah$7S7Z4$}PsI%u
zn3Q#D6^t5=v_QmA2<RgOi6yp;tULL#YqBjJ$u<uyQ;+PaW?0ik3O)1QIWp-c*cAgy
zAE<*Wr@~bcTw}SM$AT*FgOlSisKY;F*4iZ<@>44DcqDxDqHs3nS{DRbVDQq3PN>7m
z2S{NRzu^R>%EklVP8vGDaO2U7<PHb0?8iHglyHs$xc3d8`M@->j#OnR0&C1ChKjy*
zrecMTTl?o_$V{3`DS=%pqjDg+)Q0&IKOqjQAyH0zGPiwFD<C;a5zSYru@E(Qd5tX?
zruCo<YJOylYB8L%M#SJ3<>+^%3`)Z81Ay^n;_TZa8Y?2<A6J8@=;)I_We8bv8{B?T
zW}*$Mjbfc6D{dG>=61K@ATEMV8N!kSp4^$%%Ymxe(GXt)jdWs6#!!@O*Qtq<bn?Fy
z<Bs?9lmDh2{!6_5&FhYn<KHey{{Viw)xWvV|EY*~qgGM3{C14Q+aA1FB7kS_4iN`3
zET$z5LZU=KxiL|D9oI;m7?Cv8pmNIahT)TuZs%f}D%Kb#=BxLfqpQ=MKqMtu(H_&#
zDRC+xDO;wZ$QI2vvL3}}=KG`DkmbsvJ_*SHM&C)w_uct}ud^LDy1k&DtgwE8#pfrl
zlyTNp&!^wT_4dx@{=$W;-Is+TeHlCOZh9C@pVj<8EbwO<`A`wo*U`3YMuX5#jssqA
zLdk+fec5)$WYB49yB7d0ETXmL8+jE)c<$w-k=RrUYBO|>bw4QKJi)9HF$BnzXiRdT
z&-}p2V>s_>v8kaNJ$kIHR9A}97jM4lNX(h6?5U43^9oqFdmt0!x&ZI1XT)Lj+Rj@%
zAkFO1hQ=tQ2@(ub2zgE{7hWVuFT4BTkA|B%XFZwsIOGm`6#+nWyih_pRlax6G_Ybw
z_R)p8Kp#5PC@OnSd&1c<(NI;sEI$kvB14{j=?#^agYnsrQ1AIQ4s!hdo<`+4YSkND
zBNEk!r?%Yyh`shXk+^>Ld2tEFFm*P=sNrPVb5+d~Sv?O|AWBam$$m8sRf~u@JC9Gc
z9iSs;{vm373y8eT{gIAsu+;(pzhmU$cZX4wvPEON-`zCV3d((A^j9GHhLtaqJ-(_U
zBxn6YJXEm4&zlU;Po>MG6QNXgSSNut7v3oiB&1?t`WI|0j;GOU&aP=CQxpU4j{91-
z_oACwF|C8$f7%>BiMPR2UpTu2gK``%&9_13!#U*dy?=k4sF+=4&UmHVyR<d)1r9Zl
zD)z*Q$oY%d%q)?LoFX~;rnYIektSV+@&ujR>dLHR{kRV2Bu*M!VKKa#&bAOtXi=xx
zg9qafio}DHako~l90qfxsR04iv6f0&4d3t%4Qty(TJ7IuVOX|NZ3`xSXkhhRQ5X(m
zfr1@oTdJxA%2Y1sI0iAaSM_sGWg%d+&{x<n!I-|Pz@A`Z57BBFOt02H*mEFKnlB-x
zA-feAluv0{-8DyeSq7!@;0sHt3(AL|-fUr$S_}+;6eImwfBccY4_cy{At$^MqKSPb
zH>0~M#z{~9yfw63Gtw-oXT%-q;AeH*zEiDhc4POfdA0ey*N6QcrliOk*>a#g#=m?Y
zZDptu(8F6iyf{DAyN<WZ=HKZA3-0RKWLD!B(bOk-GT~Z_CZ585cd>1`oD+uM^Ym&J
zYS$c+cBtU~v$GV<?#Gc6?^^rG>svW($=>!We=ll``;a$z+qV(dzRNOv?-7PZR>i%D
zduD~;wlVq_U}%VZ#@~{}UqFe4^`C>%R_dhOfB;I!6VEV>woz!it7&-&M$~=)7P_TO
zaDl9!Atn0GHlgw%9>RpEIx{Ed$L!P!e`zU17cfW-bgECj#()GMXfgrNGpqpUu8hDC
z$IZp=I8vP8#vYL7cD3t?>iekUQX}p_ysAf;&dbr8)RNDwT?DCEsR>6wSVu>MqGDkd
z(8T$9f!2?6>wdAdnQv(~moVsJq~mnx%nmjFu#A3fW){>=#m*EPNRbknkzW+^@d+t}
zC>$C>1$FZlpcY85vP}KPROw8oP^^4Ln~;eF(a3HsGHJ@ERGc&3VmKxpF#=;)IhoE}
ze~J@tkXT}p6gplT{#6qpKJYhq22a<awk!(%ng+@q8d-?A=+?3Xau}@oVG6nK+&Est
zA+FX7rfUQ?HlrUte-_FbeSxkYndv|6WY#A)$_6aG;{S0uY*9qqsYLp+{url{$Y6eK
zx!jdzd#d7MTq^f$bG^tv%CacA-+y6Qqa#3cw5&tE-sqDZSxXXC&sV=6Dwg$=ngeS8
zk#<kQ{^xH<&*08E^cJVz<=#W9J*GALlG8fiEte(0D}?2rZ}a-Ia=O-lqmO5^y(?~y
z?*r_IZlN20iHdKO_>c0Lh3TJnZda+=@&gPgemgIy-kV73a`YF=;w*V-6d+IfL?bd7
zQb18K14ldB*8Z}j&|(5ZrEKKnKjeF}vuynI5b`Dm+!ZL)IE0a;1wo;>!XOVF{rx*B
z%nQLMf!`S0=+q#_RM6u6(yysUJQs#l%kh%1TZDD%Kw5Cs{hXB>cx^$W&%0w*&W}Ri
zRIOTFp)CPrFrjK^?ost_K({q-c!lDGc`v2&<tf*u8p#Y=(fOW;rg5CXX&Z~vGNm9F
zDOv=|xhRzd@RcQ5%#fN#910RC^UIA#>d6o9)NUE1gp$KE0UoGZChN%6kur7z1i69w
zGP2KDaWgAP3KTIfVy=yLkSpXOF!!)9x-u$7yUXxlM;3pNfrQ+$NIkIqK6S21wVzwg
zOsC|yKxLrt7!k`+&8SSq6R|=H?zSoV)`}_SCv|@6?m%<1%+45EES)yg^*{4y@)p0u
z%Dq~Jzm1N7R1}qd;=oND<eJHa3ayilo%&rmZ;#b&g{RDqL%y{p7g8f=wpn82`=Z-o
zR=wdv@cv}^8)dae**E{!(vML`TJT_Ys}f<Lzy0{X9Otky{&U9aDs@NxTfJ^{=S}^Y
zG2V2=Y&7OW!<?H@7C8FSgh5pksYN1R(88~w7mr;|j6yW(bBwEr{A=#z_r<KP9GoaW
zg6Op;xL_?mSVcUU7E^ct^bjdh4#_WMim>@~;&1{wv;?lfgT7%Ig_Fbk+nY%PdrS2|
zd0}o&UQAp(SUx;&T-emh9#H&OquAC$2q}uE=rF1<c<4;LY7<r?-zq>r1X>X7tbu99
zDX)TLTS{wuaWf1G)`0s3sCKHmXD%u;7gRYT?sQ#JMJ&^^S`xj6N)TZCqA0_0^m@tx
z0?0+@IU-1As}aBEsWfp4mX(NmN2NENBPDa7Pdl&^mAOn3s{4WAg|gh<MkCMb8H|3Y
zEDeW!3kwpaf&NCgz})a|wfDY$&?61rB{3g^(HErjPOTcs0I3>vP{zw>JD3T50BoaO
z$DMY_MVj}VrFqjcmWtA-xnv4Rv#vjb9V42Dt%)ndN4Nwg+CosBA{|lJzRAB~&ODM|
z=?!eC9`_2UccB)YlP&#3-Cevnvt{-2V3H8Q9I3legNUG>*S#i3nwER)ybF+DvC&_8
zObj-vixX*dLQLXk@XhkNpF$sAwBNvT$Y*)?rMDPZG(8YInPN|9yU+aq0SVb^-L339
z?L>REFEd-SS|fTBD}MB@yn|lK`|yejUgP0i%tih!p#I{y@_+os{y`DC_5TJ`GTOn8
zoE}*{BpaDcUP7qkMvzcOjD(~zDJAXOh3gq-u?enKzD!>t?weQ7+h%Lc^qBm>Sm<_V
znX7#j38aO25aE27gm4p65~XYb;mfh+{1>7(7soM;PW?V%ArnO3=)md6=wRci`P@^m
zX->ZP5A0)azDw;R(ZIx74j&H}mcIPNMf#KK6}+DYex9rJ08e7gzdx3HkFuTz76@3{
zCJlaQU31OcJu}JAn`K8Mub?KiOSFrqH#mbjKtaL*QwhcRRItp<tJv2YQlXf1Nn2>J
zp*v^9vc0GzFty8kaxY$TeHXFkZy}6!aKJdJ)$xzf#Y);?q6*`vbTKtL&vp1NWU9H~
zdYoBgws|-P8%`x#rZ&DY(==8!b*Ze>inYQfaXDJzx8_Cr<8f+7mxiuYcY7N1`Yrq;
zhO-Ubpbh>_Y$_lI;0%=a7K}CYzlHJoYBw{oRi~2kMk+=pSBKm9-oULG=m`QZnm(^H
zd>ps6rQaW!ck~YWXQr;G?ul$GtUnFiuU}+dxUI^|Sf#y%b$^4~_-zA!B5^LpWx4a1
zySnl3EFMBi+WAIR&8N(U|CyYBbO9f`n0ven@bOd(eQb5Nq*}{{%8_3niM+p5=el&%
zIaIeRxS+-Y>*BTRkMBI<?)jq8O;j<_ZQW*fUrt!JKk0Co(C7pt&XSOYCBX)O0=COH
z)D%G7EYEDvz(`9wI4P>iG|XUMBbi0R(LcWfJK4Pp13t2jB9=vswbDrF=?~|k&Z9}r
zq;Wt%orfc-<Gfrwj!|q{qh%)Qiri`1Z_Y5fEIi||XmGr=t*gORur6fnE@mD24U>Im
zNFJYAAmMKmkc+EaDrp<Vehq?w;MPIAZF>U)5NwW$BnjK9k<f~LgKe3AP^&Q??T99%
z47;=ym={~HZ3injXY!wqf}P|4olgCKUHVx6uS*{b$G@wxe<0qQE^D{Rh7fZ7M(t)z
zj}$6A0|Q7bz_(i>01YAuXA_OH;G_)#+ZB8TaMpXx?YLj1_DAa2HsGOdP67*V+)U5M
zLX~zX4^5cXUkPRiMTIqP-x7!8r8!6@;0eW1vQ3^)z1I?<yltP{U6c|u5>GC`M28wQ
zBHn-@%923=*db(Z4?+Dl>H9c1VAh!yQw*`If}x^{hYa-(xw)~Th`7SSAy`G748u)@
zWe6dmXaoa_m;-UY05Z*TTta^+1<7&;(?F#RXf2GGzaS^ccvw^k$>UfMN6b)|@Etet
zP&ujtKoGHq-6^NCV-RQviU?bsa3Rc=fgDBfv114^tQaE?;Ev8#nLnro%!6hkxQGKP
zirI6TdViSCG^N$pyfSDm4k)hiA#k3-E~*xMVaU|5G$f4vPqB74Po_{BIg};iy$cl+
zMf;htM)t)9j{yIa+X^&X<9I7kdPg|Y25eT8J1VH)1gL}%)1)96Y!)R9QCe)6Xlmd!
zr8rFCDk6!D{4Y@+I4LxTu2ad|tjxSNPQ5*aicvw<Vk#Jfc!;xf0T1I2rUx}70W3Y}
zz?5*%@v=1*P|9Oiqf;jY_N?J>!R7)v8N@jq5_5kQ00wi3iunnYikP@M%n*R~5=IS{
z_Ax}2cohr;6S*01B`{eY|4l#$uq;qV1_VXXD>ZL(RL$(Hs@K%7&qlb9#ueXc^^b$;
zm!;SDb0j-zv4eZnFE8(Qci&r6Z_cRUz4_S-{_P&0rWb@6B0OWW>O0F|1aykX3hNus
z8-0i`lB_d^rrgla)E~azhNe~=zF)t$D!ILNsEVoedG5GH?+3;uPvsdyym5Y@b58S>
z@vk37)O-9q5o(Zk<+HsbpA8~ZCm6ZtJg?}Y1T|T{U@z<Nb*Xp1H}poG={p;@IQ>z>
zuM?c9wgra}z6KgLUV%uU#0hPG2<#W$83gD6-iO$LIOQET!ZA8FvdT#K%MDs?(_j;o
zJE~fCO^R7tw@*#nTC?n)(8br`w@IgXn^8_Kq%C=VvZ*X`#)|6nVwHR9aVg7==)FDY
z@BF4|6KqXp%<KfpX=7tOGh+OHEK}V4x1T<mMkmnYRrtd(M6<8OX&o8Av%FeIy(>FF
zBXx6@JdE~f<j8*Pcx;h7d<@>-!wO}?VsqO~$|VoJsebK$T&oX$$lJ|+={?4>q5D?R
zhs@ooel>m7`+A8DMzbHodMGNNOiLSQJC-%yf8dqIRJQK#2|SCQ`mT0uf*c-qmipSV
zsh3%9%|Fc?{9N|B;4Cexhc~TB>r|Qa=|$oe3!u`8k9>pQ8MgWzLoqVwS{29RbltLc
zvkN&#A-7~0J-Eqdu$q~~B?;At>&b};U%}YNyE#hz<boC>&><s%+?r6rBkyq-2+5$N
z{<teUpLW&beQ4;7&v#N>9#O72+$1pERA^WnBWn6lmO{5b_p7e4_e-w_^H`<KB3*&@
zjBct)pv;0sC1E|h<N(T<rzJKzs`OQQIjV|oOAOD*MdlN7_dwtJ_ovfje|hs;m%ho#
z*1g`yFS16Ta7*z}YAftDw}p*fm@Q{1>k(V!%Y`W2-kftS?kw_B*2((j!SrKiwmbhB
znuE(XSTA*RFbZYAc1(RufI1qk@8hv8kFw4WgckU#9MlI0hDoP|S!nW)0pv7jHe7eJ
z#s2m363yPC%@_6p`m%MST&W3_QqtA}T4*ONv{R>Xm^PL4(KOiLk~m9hlJNUDF}~GO
zi?es_7jzyH{2G#UD#?swlB10u#Lg8^?YUGfd|x`gjC{){_#=DdreCr3k;7lh6Bx!m
zkHUysub`($pgXoR+&Mmg3!o88^@eLj$n~1NC?C{uKGn`S@dU~}l*x)zxMTt4=gpdr
z9xzMvICs^!EoWL)&@20~UotkXZFqa!pQ*oZ9^z~_*A2AoC0gIVv<a)_2rZaNv;i#M
z#py`Tcxc&uz5u&ihb{gSJ2<{~S^lFnV`Ti_W4-?x8@?r_{w1*TFG(r24r0P*Fo&}!
zT=HxuE=e*qdK)ghj3OF2$$3)Bwa+JAd%^UDxrU^58+cyYIvqOKCtd*Pg<Z0|{#f#p
z#zgjO%;s;4qo24_$NqEv2^k9I0h(CvXqE+IXDSw0dR`gXa_t3sC-k{=KJMI|Lu+fJ
z&^S6iK2LuJ*Jkm3yj|;!p587?*{_Q+d_)z6g8d0PVlnot<;vip4GoFH`m~_Uc(0~Q
zH{-@Ey~?9DxO&A;vTpsm<A%&PxaH3)ENm~RL>FyVn1~WYyU)}s%3QgZl0u&H^ojSF
zg*dx}fx*%dY&7%Ah!B6-ss&`O5d}h}QFK+yr$bgcB{!9o9aCIq*KOTXOiifHAGeC;
z6hA+_0-9{InX7NnDlRVHHndk~SOeXU1<C3im(tMeW5I4rY&;^8VE9w}y{RO3ps-sA
zzVkujN>GRG<1nYyrbuS0OM@^y-9Zb6zw`qP+7Wo!t>#53vcQ|93$h|-{YiE<ip5S;
z1c-A<n*pA|V9Y}lVTQkhCtivT3uB4fpFhue_>E#n&Ao0f7p@|JjfQB;xt@~^g<Nyc
z6;=zKE0#!)%zqe!7u2efp<w{-22DHYu8Ojx+tcNsTwiV&D1T5O&|Nx=6*sB;W!YGE
z?t)0axUH<qwk@q%NLN0wAh@5j#>+LHG2*S?5M4q<weO#lU|z#8(kYrl!@~uUnTsNO
zw5Rz9X2RMuc5Hi>tdw)O)<9}B8Rk+6aWNI6o#R`x|1eSP`F_WfnTK;eldI9ND1iXU
zlF1|ol6Yz_Y>X~~+sdn~%YiI!kq|$wVIGBLSbk~5ea)N<j14Mt1>1w|3RHYb+w-KE
zf75thzqN=^%@#A-KFd-?TGGnM8TZz?3PsnlqL4j2jM1QjG@*yE>YUOP^Wo{wnuX!Y
z_6~o!1Mo*1{Yo`sl-+0sHYqH<6o5_~Ko8$*6{-aC#I){(isKwi;d>auee4XcCfiEP
zZL)z)u3Ar(T6P&||H>NC)-TQ|r<<^Z#IZxkh?wes>}mes)&xZr796VEHM4@+VXxX>
z6*@%GYFc<qEE<6mv5$G~7tfc%wcp3Pms=vIyQK?zri?#>UxRX~JZLJ5&3hP|N#wgZ
zc0UTM&BC_&qT*V%6$E(T5xu}zmT3MpMo_xRFeY+G0OkOhiGypV;7t;oLzmPXMD+<1
z+9!eGtRqjVXY=Nd_H(K=q-rSg0X9kFEetr2jI$tzGj+~O_(ZqOR+sUz8`T%aHsCeW
zE1UK0#)OL~N_=Cc_hVJ@&*@LIIFgenjh@8v*+Y*d!w83r<C=O<zQLmdz~pfv=}+@H
zrrbiFLGgC=gJAs*&qsa4!0o9YylgAMHR!*w=1dyz*kM%!ayS1Jp4bv){&D$<iFxt}
zf&IhMjHgC>UE1Ov33m~$7?gaX2dy}C=F3?MKG5ApYH%r0{0mpA_eRVcp?HH!uA#EY
z^2dv5>l4r%sQ=*algYY)L42f-uXxo3K2Je~6!ZGz;z-}nM0t@)QpJf&ek0DLcwyYw
zjp)l|{wBSi%<Om2y1m~%e@k|M=^hzb{yX}4(y`ykuIIY;rrM>>^ce}CLP39##GAk-
zpc)=@eL%tbO3aV6J>hceVa4@1fl~C3M<DEiuzb<jp1tYwZD#(>p-TWczfPBi4v+8j
zng}s%SsINknT!k)9m4UraedA^W3tzumHm^GTFB*ilz9xsVu4W9*U!;P*GTo1Mo!rq
zmAo|`M7P(;C7HXk{nyf+t>+Eb+M2C7c{|^aY4h&BZ5RdlRS|!W{-{^?IsQ^4O4?02
z=M>v4#-urk496fzhT`f))s}b9)S6UwRb96O&vjHWp~<>k`K(<cn~QfJHgpX6S~^|q
z*lqyAeU--mR&oyiZcrx*!@BULw&ys`*lOAu?XBm01~ZzfMYbR9@<tcM7snLBsYWh|
zB7<b!^3jj<LJDpqQzS@|t^zwz=YtXv1h-S3*BNG|AIC5T%ar1JoB|_MCu}UcPzezN
z<|qXjT=SKfiHqJDb_-9~_+e!9!f_M>MlL`o6!#uuD}Dof+RdsT6cjgK`SvhShv9hw
z_^`nnRXniZ0h9$IID><eh?Z_Rx(fbCDOs`czgYF%)JikobFDwt;_}moWB7xoCO;Wt
z0U`Rt4if=RA^1qdX$<6};A=<mryztdz%!j0I#$286gpqpHdS0TyMsh0$u>3x$Z0ce
zmW&H+r6Iv5VgQdnArK%2IQ&K3gve+Bp_pJn(g&>(g$=c5!!pz^wko3i`XBvys+s}$
z-L<xVvz32st2t+?oA~Z^zp<nUv9NbK>utk5n`f7{s;1jk-P*QxjbX*swlKm87BxfR
zK|mA>L@1hhJDdF(8gGK$YQ2$#yo_tHC9{LkfpoGJ8$4%4hC9ic>v?4-28^yE)FDp0
zpdH6<t!|e~b9G&US!=`4u6|-zD-t~pS>APC695@JACY`cc6EQ-OS*+|QwQ#O9nlWc
zNfGx$6jud8zZEoRZhKC-f(bH5@$GXIN<y}*R5o=O17kU>ZR595ZAvYWfLQtgdwC#p
z$D~C_2F_&C_B^2gsvOrTz$m~s(~}fyBZq*F`+_2T0vU)B<L=wqq>~$jdRQpZ@R`AG
zHtI*w)?kd_<jN#q@@r_PWtA0ff9p62nWTsh3D1XxXNQ-MO$BKeWnpJRhX)e^WU3J+
zk+sifdvnAS5hNp4G`>Gr6ZA1nq_5CV38GikJ8ykP`}>f8=3!`i>R4pVJ{FcFZ>&|7
z&uu6!B+%bVK)wxkEf0mrP#>H}0AsnYU>SY4cCMtxqGgSW{fy=D9l|IEf2m2AbHXO0
zuNw^g37n3_ycKt}i-74Gk$!5GqpAVyQ3x>%&&q8f4sibD$mN8x$B0~bSao-+;cn3c
zEdn5t*?cGT0Rq9PlrzVa5k=&E^4Gx$nP#F^$_AHqh&6Y~58#SKJw-h9-M9FFN)Ex&
zI{-B*eF4un9kZ*Nmht$|U!x%}DQ`kF2@vD+u=q0(8$?sMu$kK&yrOoD*LX-`#@8V*
zs*kQ1?V=+tHE}oU&;44lf*_Drpj#Hr{Z1M&QjLZrPV2_F4jnb*DuA&*`5m9o!WA{Y
zQy8nxNB7U|SwyVdwqXp02jhiG;i-QcILApCYTQDo@dPBF;p?MTUh9K?cwD1!=?p6!
zv=K=W(d7cEKf3o}b=AiZL<&MljQRRkRh00;(*9J8`4plM^C7&P6-HHVk-*HOqWXZX
zcmP!OxG^InDe&BRqO8+_4>CMG)BD3IAg_IDF>5|=c)da}D+#5F)a7hZkp1z7Nof4K
zi>yjCP~b=wvOp2;O=pw-1Ni)ck3APjC)aUpP?QD)a=S(b1`sSbom(d#tO=AEO`l<p
z)OdD-xVeQg1@+8EjYr{!RR{?W!9FQmOT7c4HGIfwI$R3_9;2ys>lsX#0tv)Vc=guP
znd}e1ePR5);&M&#(7rAsYxECvcZ9E-HllIOXMUd{vw7-o<lSes6?mHDYpO*<{{kxL
z{RH_>bl_xQ{?AgF^?yqZ{)5@#RjPFCzZNR4y`b;~ZK8I<W4&oqEBT7pcx|W;h4hJ!
z>Y@Bgc6x+yu1zfzgwb7!>BzYacs^&}M-Bw$kPshCZ<uhR$dk;A>eXcO?fu(?lqh6^
zMD}oW;ZW7w4kblHa_RMay-|BU9&RU3J)doP5WBi%kYw86^)6I<HrHV!8e_3Xk!2B!
zHKfN|DtW$Fjro@SZEe={V(56KASl9)FElYv)+~FUv?YOflnxbdWXd1k`F_Okkd50n
z%UzNDSY@oJxu=<!NREzHiUGPQa-&^rM@mCB(M64ytfDQfGMNM+`&CRho~pcR3ANq%
zE)`wwo7iY4k|4`MeQ$V38P*bj-bB2l*H=FWb;vMPUKcr7IR=)H?x^e6^a}WgT~tT!
z3PhXhfPcaMK5|mrI$?<~kwoeuYDwO~bhH8z<-8P5<yV3~6UC#sU%IVE7lwmx8>54v
z4~d*y)^rBtU=g(u$j=dM{>vnSi1(8Rm_mLS4=Idt%E-3YUcowGf%ocrNpodxKmH-j
ztd2Grw${X7DvujRf4Wc%f8U6smgwMbUV!uqfq9mdT;;KmX^L}e`fdwQ>AZT2<XZ+_
zc>^S*4e^=DEi#)GX#oMP{!wE^ZV)EXG2hU@E_B<A)25Xr6*+T$sceMWMTFve4&~xl
z_P>m-a#i<oT*W(3x`?h;|LO{i%9UTYkjp5?tKwwq?&fdUrDMWZUh4>fTob1~6_(u$
z#Cj(rcxQyy`e7-49&6?Z@3ReO7cU>Q-rXZjKhSTNt^}0%m+1Vq9{i6+?Z2~vc9kM+
zw=RI-cl{p0yROtd#aF+CTumTb%!c@^Kv_<OPG3#RA7?}H@}%pzf<&Sc41gL<1lPIR
zc>iq5!|mroAfVpuYEa963<Js$r&3@alO_@%4WpCCAm%zF(D7q1_Jh%1UpNBfn@F?5
zt)6>>AKb8Kap0xui(h>V0y$Zn!?h>R`tF5i?QF%7?@wQk@0;e=fiFf5n)%_zLRe#;
zfqr5T#KP6#OBgwIOi>kbP8c<6CW&@sEhRgrGaOoxCNxWNz<CuoJ52ReP^czhX-b`j
z&>Ue%pzAbGxttLrA)`47JrU>^n>s6Y<bc>OzXCFjw%mB&FrfFi1R9F9apHWcjC_r#
zbW$FhR(G;jQs`1)+a_c>e~dt-nu5Nq=jd+q{z;lZdklmXA#a5ZYNhjsOJR>=Tw2wR
z4WYv}&y>XSmgn9{@G@;h8KfiXAfagVIOPOL9j>(t6*ff&PtSp1;sigWshBkW&>9X^
zb6E6rIV%K#zKh}`tegE5aZ~IVa$VPR1+j1xEE?C8xZYv-gkOIsu()2~)YH@bvhhOn
z*mX<lw@9Mp3f7D-@5<D~>-XSlh)T7voEbjH^E;NzMsa6{Xnt=os7cURyEWfNtZ+%&
z5z;z~6{~EXKf^}@N|tqU;cq|B=eBB$!keNr>K@m-SRli?;@s9uo-G@b5q;@BZ_+Np
z>I6?Z+K4IU!v1yDG*Yo#thE=}h5t0WZqf0_>R8Sh_%gTsz|Jo<oz-K?h6&2E7qH@S
zUFa_{$@CYtpYvO?^?!Z;V0iGo614bz9dn}o96}s_#doQzo#lo)0VstHSEgc)%zXwm
zDrZC9^p$hVht%0@u7nr_aKv!&$kS`*%G1BSIk=tKA9=fbnzp+dsW?UXG{PL~Ko7$)
zAb1*$d_8ayR=7uTxXSPOalSZtJ0ALUGB|nLS$N?0t)>{ofV``!4LN@bQx^>^2yss|
zcU(1($N7keN)ge{NTr;Pxso)O!Ul^#Hgm6)B6Ym@i^63K3?@rV?j9QMmMeQ%loQ7x
z>H~X~fVVuhO@Jmy2SOC)-QwU@l?!Jc;VoH*hPiWlXjy)EYV6hWo=dwpS+jC;EYoaR
z_WWH?tYVCo=SJAzG#OUL#ru!xuCBSNX3E{NMzS!{i$%jiiz>KwC6CpFHJYTBbh7Fg
z5D1T@ACN9po*Ymaix5jm$-;>+8G&>1cz9)_QYI1QfVYIv5JjWUSpO<@<n@myGlpj~
z43ge3K@SEn^ai@6Y2QC^<J){swNkHLh*2UK*Bk7R&n8&xKv(#;ZHSCSfL*%Q6fn;z
z0|L*gwD%&G9LV=<k}Fb{QcVG-s4cLS{;Lg8nHj2a8A~JD=R%ibRI40n{pWycg6#&V
zExh~Zhj@XtMnh%gw}1`T@ekUy=QYOH>!zL%rS$@g&|j7V79h<F-jJofH&@%m)KBr#
z97qrJ<k$eP5S+VL+a&e4G5tQ!USSwL%mv;$<6UZ+>+;AkEoaF63a2yY<zjgNATR@p
zk(H~xEpk_t7K&JX6(;fIRJqvI9un^Zbg*Xe{OT7h?31Lc0F}s`is==qP#Vz~F`H+8
zVbJM<QLl6}LC7M6KeNpl%NHE4Vju;UWJi1!E(Rnlh89%Jgm+xbfUH&T4`-s{(ib`H
zl-4(>1xsLpEGiTH3eq5y$0Y-=iKeZ_(iJ&k*19NEy_*?g3lQV^nFTM8H#T@Q$KEM<
z0QJqrF26-zQs*#mwNdwc@dkdd;cEu|oU|z1+;3u3eBoWi6vJUoymW_+Q*X`?vYLO)
zo;YLYbR||<q@2l7v8_lSS2bn)X)v{DoqR#P%!2f5V+L=2qJ~@iGoO(;zM18!>K@nA
z^6*bJT($b7IBemhqW2QGKQ{q;ujOf>;i#a%Q)lF)nmg~9kSs?S9Q4(Ph))c&|Ev<h
z0Nvc|)rXzz_99UzSypDqub%O%&!v0;=#@f!&h~DjwY%R{9V>dArIXtsA09l<fa|q-
zjDLCP%zwE&Wn^ObH<ABeKlCMa8@unu`7`H7ek>D1qZcxzLGA+K7FV8Cyt{yZ6NRy+
zL(oJ5`Mk@*j~(LJ1kx4TC{|3iRe;!Ex4+$HZy&<ICjG#A^}4cjyWB`c<Kn(Km?>M}
z`~V3mArq7-w0_yLgYnU|YkED#445Vjq3|%Hy1gWym#e$2%X1H=BScu#bQxmNLH1N+
z+QGHQi>J2FKU+(zbl<Loqq)oHppOQA?l<Ux?xfW(<;}I2&?P#F9Ezt$jzmsvvXM;L
zshsQ9k>6}3xz3wHx`6h;dVoMXKqTWHq_FV+VC)^3GYz+G(b%?aJ6~+OW81cE+fK(e
zJLuT9ZFG{3b+dNuTlZF-t~%@e5AU4gnL~1UY31YCgyr8QSQpcOpD1c%f#c#<Q-T!d
z@;^aiPf6I%Oe(41v%Q`yv&~EG9Q?je<3Qmrwxz6B$7g>-g;dOoisBR<i!w#pnYn)!
zUsQC6RK(9soonOW_>LZ=`~YO*o$QUEj?XEsvz-l0jr_iiGr^n|HGO|S`{2m9Imu#{
zE{^kQ3{{W|m2wzXoID>d7E@J`$;n2Dzoxn(hO^Va*x1C)7Qnm2)3|Y~$%c*|@XDKx
zZe$WOckzW&4p6CJMmZ7UZud}xjY%YndIr+|cDrsDkz~a@f^jN~6BjaObL*N|TF*XS
zm5$3Qsb3}HeP2T_OMqcTMbVzom~|Ww_{^5nTYv1?>ZUy)uvWVOMs)t_Rg2#Y$thRG
zk87aOr+}`|h2(M%N>q9uwyD6Psa04;O_F1^X5*-z)RIUb(>{-g3iCM4nK4xIvU6df
zr0{5fn<^4rKw@^~76SzD(lMZDPfUBpX@j$)#=Ts!Oe2Hnfq8^I(crC2JvCcr6-tix
zvibF@nxzRluWYXl061P?Q@nE$!Y4|Hv_HaXq?%@W8cX;h7<tbwSYVK&@ENXA&naT5
zSpY+zR_ocGc~jL4oK-Nv+v(xQOoFhp4SnU1+ZG<`oI{x2s5Jt^8j^8ORJF;DP4{vK
zlEx@W7O5^*rDH4kVV%YU$b+ddS_?e`E;Z%#SlWiJl}Cfjk&8m!z0*^r6$m(@+CI1Q
zFpw9>3&%EHEw%Ws6iXO2XW1$%YL91xY*bGBVz*PU_YJkeGd>D%Sx?A#1wTA_`}Y+6
zF9jtM_<Siamc||#v6NI|H@-*>o+*QQZmX&OPzwM4d4{v|;mK)O{zl5VX9Jk-!Ku3C
zSrT}>{Y1R(cP?Zili2wBv0k;Xc2gtpT@7;J8~pfuv-h;wb{eqDu~Ft<^X`nKZXV##
zNY6pJSC!WZ9fu+kDbrvs%8W$zoZ-5-Usp-TztJuH{48Vv+ayIlHh5O+kgrYT>ViLr
z%sxc-yK<;*iifqsxH6AtAYN2ymM&PY$4%$8D~ys3>>F>}b~DL1i{V{gI~BX&-~aUT
zS0t*TxkcOf8>P(Ps?iK61<>IwW(qXL1WIS=B1}6Ags)GedmeCdLBI!@Czt5EEgTIy
z=|~J1<^$t0)DR(#@s(%m#)AA2O;(6kRNn<9c`V+*2G61udkSq{)tJ2s0+p8}XgwQe
zG&H#?WdroKkXxvBIyFa!V|~I+K<}j2Ev;QR^S0@3>Ywo0s|Ke0;leO(mlz+|ZhL3g
zdKBvl!m$e@YaMCFtuw>a2S&$>F8S98{0G)yX8yM1|Ihb-AA#NKTmKn>{!jXa64()%
ze$j*Vkaf0E2-;EPI7|ifaB`)#hOeV!T#k<`H&oJbhoe$C9KOh7r14L;T|9a9ZIAU&
zhIcXmPiOi1N4EJSX+|ASS+5|Okd8n7B~NSVl7(lol9#=2fUiGiMxPrl;^2?|`3j)d
z;pv25P8atAjN4&^0=hbKy<xpQTqo83WqnXhK%yM#5bJrFyK2xeE`p(d5JsG?@HaYr
zs8`+`?ruT?G%k_jzhj?n+fgC{dK@_#TqwYE)CrU&X|Gd<wdFoNDw4cWp+UKq{vb~>
z!&9Z>V+ewG{9w7@lt)s|Bf^tH2kdNRv@?tceg(X)AtOP6Zb~3inhJK<AT)_w*DYiY
zuI_x8a0UQK(U|*l`*Jtl*WgP!^Z@8R27wBR(g-DaxqVxseg&N6f_m!aA^J)e_AwQ~
zc%ZtLm@6Ps@Q(Dv?%=omaK&M`%JPwSQ!Ws8cR%02@9+DOlYeno1D26C1g(#^GQ?an
zFE{sIuh&(~nao$~VA=$J5lOd|;vi$%UVzw1?^DU$;5?6Z?*l^&frsZP%HLjuwTwos
z)6Lb}UE%`|Z}SKfRTAN$9b&ldIk={qLa==yhG2A5DGZ2aJC3@7=-@Ha;j3HQIA@`(
z^GpF?imxj*m$vLMsv1WzEFG=K!I6f+kav%?Tz+v7@szC_WNz$YQ5Ue1uooqHvxX%O
z+FJT)AFArgr}oBLtA~(sy3h>Stx472t6s>rRdNo?;go1=Ws1IV+j2&w;+~YVLvDu-
z<87aYBLB2Gy?RLj2l7i2_jgZMks@eQhB1Rvw3#IZvm`F>ZtPBE8N|Bc6B?k)LWaBc
zN*BE#iJwPiZS%DGT{|pTza$*g|H@YI^T??IEYeur$wNAWAHv#gS*>W_ZSE*lmofe~
zK9#5o6P?xl(frym(&i3Zkz8M5=ZYB@39+En7v9Q{5EC7w-4#i}O=)le#4}demhV~w
zZ=goKx&?1a#B#(Ml6_DlFKfmqRUhKdyOBU2?!HT)-Ge?D0K!&3YG2Y#44o&4<^xzg
zFd0*UzgDUTipJeQ(~8DOH?KXd$Dblcjo%DH@?lGx8FzMY8p&CchYw9k7&zdf{>Gm_
zsMb_|seL{=?=xpS#R0I7w~PytKoti@SgrCMJ~%Di55b2HMa%YXa&VTzhHn1^oDv0|
zch%_c$92j%LqWdKwLNsZWG`knuqt@R=N)zr9A|=cUt@p5pi2h!HkCr=mOYfN*4F6h
zaO}pJDoxZ8tB2zJ5pejUw8w!Cjw^lRg-yv<af$Ujw=D`HKn_n`@g`B+JGF2~n7rMw
zRcZyaqz2Df#W7T~v7y(N^HNWkPOxYv6{1XLr83*2xrT5EJIp4plL@@9dey4B(oi=+
zCDM8@^69D6n$tzqB|nv^ovPIB*h%GJ<?gL>SYViYXx$dKIU5b3!DXz{*0ElCwi>@%
zKu)&~;jZ1%fTS8+%f(<vvhg+?yvc|AapFNJ{|PC=#=7v2F~;`)*9!fwlZ_eh-;c2&
zO{s6b7p>=;@0E0w#p3BZqaFj;axYN0x}M4=&CQ~e)F)DfM(aM!?revLvJfk$G|C|o
zAZ={?+4DwZiBm(4^Xv5B?{4=R(@ZH+`GGB$j+~shi(<%7bVlNZYt8E^@Nslb{@c<X
zU5eRAD;TcpbZBN~Tei-g0va8VW|0IHd7z&!hZbMPS@-F3P`9JA(_W}v=l?b!`5*El
z=7(5l1j2?DI(~NN%e?N&LbL)V*y4VKXbiEeB$-20F~yvZv*B<iTD_h`2<fz`%F*22
z{u3j6BcY|RMlo?jQYMiT2y+aWMWR-RAipe1q~m*?HOEyawb@bYy2*vR#;%<A83l$|
zYnE+eDa|Lx!9y8GmD8JQH3zm&!u5>==1CpP2M{rsp={UlSVY-yEL)ac-)B)^(=j)L
z)k5j+@AqKbHJd8vImhea`;xO?phEW~j31?D7ZhJslw`HIE0!0VQ)WH-819;?Cb5KW
za~&?pPlQqOJJXdL4C|iPwwC<Q^scRAavCJ__rfJmv}eCG4o!905K)CCeO}ms-_3xI
zgh{g+zrYK0SY~q(fL(yHyuthk4<7<Y8pMkD8nfZz+Oo)E3;?IqF8WWZb7)87tQx!8
zm<GazF;R-^!!LIOyS81d+gynw)j!3lpiQ@N-!9>?8p5Pkh5i<@AB-1^2UR)5cLY#?
zWLoDu_?7F#WcCTdsq&(^bJ~F%;8>o<W;&$L;O+tqZKiVb%2oz|yT*$ZJJEsEm?Pht
zYKUrL1a2#11PV=%%w{za#<I_WK>rqAdaQQe=*-l-+o~a?>*Q;n0eFd{>EPf<v$M#q
z))=}rP%mq?_&IEmw;vt|c~FWHg#rtR5<y59ngWC;Q0>*3XR?g6ae|3<SgQ9=#EI1L
zXQ)S<?JVAy85y_Qep;MSYYnw|E@WgMEO#s?Tw_VMh~wSQ@{bBdV*{EpGqLhe*~?+d
zQ<IURumJYhrFWcbDR7Szh*%)}&P3MwzL)GzsR+93#$}XDNipvGvkAi?Mo|F~s74VP
z?Wvz?Yey-w+fKuYNtToi0U`5(!kG`8A$RN<36*xi4ed+;9C<#x82li6CqKgWy!Q<D
z?|w41ZMEdW*RI?KzHBq-YGZ!6lpzf1GxFwqtJlmoM(p5=36;Tw6%7(ae-y_b8*@`Z
zXNZkb28Bd}y2}HDT21x~3S={(@0dE++O2q++j2eSqbXshB&#zTk`hy)2jLVNP=H6K
z5D#Z0Qz=CS@y%_77(_fq+pd|2tr0u)&|j2VogCM8B${fk>+!k7tOEXI&^J|%6Kwam
zpJ*Jdc|p-ek=96Jrn)LEciH4Inz`MZ8<kwn=Cf-mRDEdK-`rc7^lXu*MWB!bV3i;^
zwoFu!0z!V(5Ju)$W&G}U6owMFvIhJ%!U{#1J!R)RDa&<NG`ZzwRG$i0x(f@Eyk%2S
zkJFYY79}~_%ab?#tf0pb+cGh23cI@}b}ZYaTzP@0Rk-qMW6I}z^3;nvRx&^jLIg3I
z*%gH=L^+BId6Pe~QT&Mpkl4I)Yr9y)(_!>S=f7~e6)JuLn*(mfl|%^8MIkHkgK^XF
z!}|l)o{qv5_52}7c>E@Wn{Y7Jr5~BGE?4RU)wc?aHDc$o<YuYU=RNf5csACW^A_7e
zE%2d{T*s}oW}J6<_QZCK9Rf^?p{0>`2bI@aL9|!8!MZ!4TrPx=f*)*Ofb_?^;2Q2P
zh&9Ux*sqZ&I;lj6xT_nMM1_vX_s*K=F1+bb1NbrCy_Dg)w4WyZf&|Nfyr3-Rk>HB+
zpy@-2S9%@8)%3xxhcTQ%Abyf!1&hljT0j_9Z$ls2!!foKn{TJmyZ8n3HDgU3L$QX5
z0a45n8Z<M`woTUA4OD%sLD<`8H{8c-g-^zbI3``)xo9Hro(9&+Hh1ePErWSE-2|q3
ztDO$jbl@;4g6;~u&qdD76$$;Qs}XQN_OQcYHx+)1C7~4WM()WQEP8mDum-r)AH0e7
z#tXrPvi&Svu6`?opeit~dA(b#VqaoGa~@_7Lkj_zBpT%VayVTn+4=Qqwg+OIue7{{
zbnObHg$o=aCG%%~als%ddmp!|Pw-M#LPw0xaZbeDO<tbv^imSS#&b3F9u$D2BtZ2@
zfCI9~rPg6t2P7T16Ug^oGGfgz<dA^-VaZY`*@6aw@Ghk=1W`@lOWJG5gz6A8K3Xl&
z*Zsb`&(0?&jg?OFep4zJg-nJ?SqwF#Wb`Qy)VeM1<>QaA7)E8R=AYc2$7bYLh-mow
zJ{$+?+1{9sFvHZ|yX+=9wO`abu!>c@rwO_;q8WHqt4dB#&Ti1&hBOK^^8IE&eMNJ_
zM#4Y@OF1}ielexnL<Ye`*mm4{<nNf+v)p5{v0=vISQx-+W%Y*3-qys51I)tnE6pka
zeaH6lBR0o>E#6)|{M{G$a4~B5*9!iJ<cjk<Xz0IvNlaz@Po~hw%?lbo#gFbw=U?8Z
zfh6Up#F7T`XkvCoL?kWbJRmLkJqNlck@y?Xg*i5qP4}xi)9<;8H*-I?pSIiT-JO1y
z-@-BX{e~sv1<p$nELYJ3++M#5!eKwdSbSYtVIhF533TJ-=_C5oUP^H8JzQyec(S|X
z>ue%iElln5zdJe+4|Q6LL8AkUx8||a+ITfjf}{S<z^M_!SoSlFTYKguS`W~6=0+0w
zBF~yG(r;Q7n=Bh^|8*21gPG2}rxV$a{<_zHpGmRFbV4=WcsjH7(FZC0wqSCd<~Y2P
z222X%m1vhW__NG~U-%$VR~LPZiDap<KVN08zYRXgM-#k>bLN@7YLz0NS_n(sV<6IC
z{)Kx2LiXvvLT?$T1J>-wSnZlve)x<muN@M1L~Gq7_aV4zdjHVzQ$-bsRgg^|EkUI?
zoGlH!N4~1tK^`PTYUSy*WAn1^`iEN2P!8#(;F+9S>}(-4cU?f<&|GzrS3vj|{Danh
z01qhw1w;MR-=Ox@WDiS$4N#u47zg^S_9nI6BmP}*r$d|Ml`hve8Q#1yO>|pgDIl}#
z$Ig$^dqLuYU`JNG5F8vF-ijzja%3w?N?8X_uaB|6?m34GY??VyuM5h{Ha8c4SZtxx
zWuEM$wEKpa*K3+gC#I+8SCsfnz&+Y`1h@5?*vA|Zie7yN$-WF`IecEv+kmI9q;y3!
zuQX(i1)*ul-hDF(N%mgsKDL*(ufp8<4n<TlK@+Pw)&Y5#1V7i#*$z`9un+{1IuABO
zeUVx)Hh<+IJ}MiyNbu*N&=)RUECQoM`O0Hiza1}zCJ+Thk#?iFDh~UMG^>Qw7m#9-
zMkERi@XRQg9lRb|BNjbB6zxjp&MWoZK|IW-1Oy&THjn~UBo*`h#(Urtp-p|kHMb0&
z)Zo=M6(_-X-ed_b(N3~$4*T%0aV$O`GK()%vmrjo*R|mQ(0>JL>dXvm3nY;R{QcLj
z^G^W}+kc(EJftD>AGPBDpx3DkC;~7jJTFNWsqE?EF#xyA@RW%KbZh31Vu@m#lHMLz
zI!RO*(-A2`$^h6)(6?td@+=*4wwIT~p}WJi3mvjBtRdCLU$P<6Cbdvgrj0t3Cl$xh
z@e}&_-DZojcFZD?vp6QtGK~J7pMEdrHi?c>%3>IqD&GxHK1zpYcb8K)lKxMhOV~-<
zzuvBblf8t4v55%&S|ZgsQSv+A!44<IV2a{~YUMGCCT^#x>gp|Gn7XMp!D!2S5XhQ>
zm_k#+!X1H-S@pw<2(`H_wQ2vF#Lm~j;G#~!&U;ATYTYV3yV7VLk4*WA`dN^qCf7TP
zR$>XN59dkpWHubY2fq6K%Xm8Mxczpqn&9Uxu{mx&;zIrbx1NQ)hhoH%V}4;}oLy^`
z-6&qMhwrze2O;MQ2r1*nU1zX}8z81G{4pf|erw@?)@*yjj~^eQ5^@n7f#)<x!)9V%
zn1_L$CIU!-GerI})tJ3;UlTwl_!K~&sI6ewxH?u=x3#nC%8!L3A4gM&f+I}<1(nMX
zp(uo>Cc|pwf{NdLm-NusUCu#M89~VkN4JtI=QoHvI9Wuvc_4IG^^fJz)^1;QM_d~u
z?iebB`1|DffQ)~S$^e&c{#!fVLz<;u19_s`W$fq!Op4uXF2vtpIfly9K!x4mG=%FA
z(gGqa2#VW`Kfj8$94<))bpW47*QA<>`Y*H@fMJp`tF+9y)lm#>36kAS5zK)SbKjfE
zYCbF=k=dyDFxl$cGmXq7gzVuzn37{2BP`3mxKwFX3sJ6da9m`k?jLtJq1$v$?B4nE
z%*-L>Y8K%_;C%fxE8=9%yK)A<-)Gy>ivNS#+q#WFcK1LJ*w0IFOwP*`w|(&^aU66E
zKG-gPY$?I)vZ^O~0Zb23ep~9pFr~7xr4S&uG}HT{G-RTH>{0FT{B7i9&;kQ=b)t3K
z_Fbg(6YMmfzQQH}$U=5$A;EfSxk(kKVY<xVNWXamW;1{P)x$ba>ri$Cz5W*h^{NR1
z<EGtFKJLI*!R*yjlN4efB1?YC{a<~H35TZ)$j}!vOuM*_dZ0^;Zzr|{rkyxunEWd#
z|DnR&p5uEcKcwUjsv;3hfHUDK7z&^{0wA7Gu<+whbb)stdz|o+jig5BHiP%d?y~ZZ
z{W4;&zL&w9x7MrJW}LCTqu!(b;u442G(iR8__ej#-HqnYXZ2<0d8K1FqGOOE%}qnl
z@5!95!0pV9huL2ra)>|th=|g1AY!aITYC`_Oy_&V?zUW8I>moPru@b#*u^zC*a0%P
zh}G>{3_>6+ht32we|iCdLtXi;6iOr!d3!%@@ar(dGHy8C^@J|TXJ3D87GDtV^MQv4
zujY#XFGwB-{jY6m9}B?%7!=vzsHc<cc-Bnjmedo)S(5k{AMSwtnXX!i;Bf<M%_p~w
z0QNQvX!q?{2hb`gsEu*Xct{)9PvkI*l96{;;H`WDT5Bc^#0ndzP;9P!_c^2UhJHgd
z2)I)|<O`_Nk-y9c>~sk26xc*1Gh~dU=gP|1OiC;#2NiHo=cHM&Bw~g}waO}I&KS#z
z?)CSsC(8zC%ew8`24$(l_4DL7af8eY8E%la*}v_rsmZe(lpsn1Tvo0y8r@IZjYaBS
zZ)}ekH%*WYUWR9g4Z*S!(Z6OA923m;TR+``PPbf6qfGrDP!Z3A2S5h!yhp2N*?U4>
z=nX#YT;fNFHLiMg(yI6;DSzi|iMcT+dST}p(ULTMaJp$&XvNDjlv|4vzzwYXc%RfD
zK<D@3*l)&PUt80pXg$dZYytUM5+s~XK6cgEV~VXaYMcY3Jnr^1=nj;*=POx0<8h}s
z`5JGyOX6E3ynqLB`Katoo-p~u51=CnkW%M-yU<!S-6ccKi*jE6`n2%A;5MT7PyS=>
zvvdBR;s<6X&i}eaX6ZYHb%P1b_eA}W`P=muKPc5qX#m<lO!2LbX@xGa4JhEKL20A?
z@>}U}>!DICWwBLgt;@aGaE^=DJ>LW@a`6D<5fBuxV{<bGlVm76+(KMH^ij?s<yl;N
zx{{cr#r23P-j4R)58v->4sQcGCDL!|oqcZ44vkUyKfFAXa{N2rSza@J0K5-koUVe<
z=){*g=5Zl<&Vq9Eapos{1W8nB?Ui^9qq^^vOukyO4az!l7_6uv+g5riU~MLfo?zb6
z`Zs4vB<&8o7{~;yvYu=xg<bOZn%sG&8AapKk!McAZHv=TM~W1#SnqHo#HobLD5JAd
z#|bPQHN6;iEjY~XmqJU#DmBKN7&Vk51Q^k%(WYvpH2P@h(=!X+Iat{?RW#W~qV!Bj
zwknzODN;7`(@t~*EPBSH1!2Kr75+#ncIWAFjK(0W8s6H-u64Hq9YVO-o|WT1#<Qj%
zbLcqeRsly)m)zM@o0or5ShqE)d{D+$Sr7phudSwN#EC*pe<0z@*S3;VI<q-fk3(5<
z4R`IZsA{DZ5y^~Z>S4+wA|wTbRrt-VRIOCr`&8MxX3vnJ4sh%&I6D*T8C1;*rDaqN
zcZRj(4H6D8so=5Po3=A=F7X`l(3O+n{Sd9HyaZ{|MWF~M{0~>_uOmG_`~G0!X9RK(
z5bzr8pF75aCH*NcL)Q2z7cc6z63rfy$3Zrf^g%{oBQgG&n6QjvSrH~kkD}4=yAT-~
z3@1@=aaijhK*;Xc>d{~`0xZZKtDg(2jU<q2K;t`V#;E{{7_`5=AyD0Yarn%Z=a&}%
zneLSG^E*SMgDSHpX`xZHt+?guXb2vGJFhcU2II*Vua!)vzBYD%x@+o(^K+KZl;+~6
z_Ct(N5EL~d?b80z+QQ|?KGrqt_D;P=sieFJ3J6tqe4{4E42*X0LjhWL3wV_)(i%<%
zNL*O2H5JUb)MQNmtw?}tni@<VaFCMs3h?z>Y^`ws4+0NhsakV>m}^i*X4A@M9Nstm
z@;0dAY$GXdAa$y1*l=#1q2A8`HXaaUo-=N&)d14Ou7@;>(c&u6ftp~VSD{wE5<H|P
z^4!DB)T-jnuenmTpWk{@hGo$G)}7oU-o2R%`DZ^uJPh&+nt*$zP@S}7GZnp+QVp)F
z1=;T_QjuIcvyiORzi*Prc}s<-X`z=>^Hk*Z_lEQzs6-6j*05S*UA4a&BGu^`ElLwn
z1ZaS@#`%w8t=`bM@5adq&Ad#_kF4b@D@znc8|IcDjgz>YB*{4rI0hdJcy5(_7xS7Y
zELPXkAUdrUWT_a#SBoDk7c!;Pp1AlR?O~Lid!%bY4uYchFzXCF47>R{MEWBLZtMrI
z(=Re+mp2nJ558>@l-u{2{*kBx;9p;dgFGyEG#u}{XRrKR&+yi%dKCYdKOFz$zd8Rm
z!QlT+{!FRw{D=JUo6#?98#^7m3zw8yRaPsuhIFCV5r7+w5_Vh^&7l9zza0KH9v>!R
zkkdvUm=5H{aVrtVzw>YB{!h<jdQ(28dEL%1$*5#$I}FASNFfL|t*|jRU2X6LQ>?i@
zMsJVbgvIfc5jHuA9Pogw*ZuYT!oCrIbbE0jNs}ZEjvpYO+_62Q_Y42yn+uwpd-y!C
z&u2f!ZZQPnMHPbiBW>|Kx+lR)Y+34%`pfce>nL5bm<k>f3&VW#P5wucP!2~W1EV^}
zGcdQgXyZq<KzT8#$#+URp`)T2J_1d~IFf^d$#->>1)IU~-*PhQ)ghd|6R|eiN=-5&
z0=G7Bgh?xa&a$d@D=UVZscUfU>xRWEjXe*ZW`U7vb#$fm#|zy42Q&mW?ERFPDh9n*
z7mm3tCq|3g=~PUKp^~VV2+|AxCjeE{DehK3z5a*xvTBboirgiA9I*#re_YgVTa^dT
zrAsc56hEi+EuribsjmesE`DebS_UcDTY(M&YSCa&oJl9ZtOVq!IK&*UEN?~>Wjhe&
zt|}G2oW80A-$}TvKhU7P9suWf9c8M{sT_pLEVZa7ZR_~cohy_ErRY%58Do*^LeT2S
zay|MWgNi|79d$7((%g=AIC8&f9944Jpz&jMA)E68;iOr2`!AAbc<Z8;tcONYP@h_e
z32Dlmpj!~?Z>%u8B~5#KyPBfmIFu<`HuNkg`0TV%um(o*gwoH?2g5{3F%Pe1$shV;
z1oeqR&YeGq*h{7Sw@J{+%t$L7s1tk{JR>MS@w*wIKKfT;{le38x){YQ61|2Wv~$P}
zU@g4&PTRBq>s6t~bTAUVcZP$74x;@~KfhS_>M2tP6sjy|e#|(UUJdGy0w{Baf9il@
z8|L%t<Q|+cE%nhRi@tX}Y7eldh)kkx4?Z$&ouYGxjJOQEy0A)WF(Iq|tP5xZ&Jwa`
zrpJqnn{-aG&?~H%6}1oqBpT50x|ggFEn2vqJ><X5^HkQN$iqEe!&Kv;YgocuN`COl
zM989aMVc>di3$(Nerm4CQ7`tQM%Avo0E5OP%1U4BB7f@-Fcgu)FrB}Y9I%;Tgg+A8
z<vee);;r%iW{#`0n{K14ZvQ40sIXvQX%|wP%kTR2Ds-P<wzpmbYOF(fzYKihfPYZQ
zDP~MSIkU|?n#zVtKnD-6qonJvyIe&n9EK}dSb^mScgxpE^w-b9`sFZ4(S`ye22Axh
zIpPckzw4JNTg=EWnlCz?PFD~gpSKj0;V1f?yGj=&#I$2<<hRME*<l0uy8UCU;(4==
z7e&?WnbBQ+N+W2z0O>IFUEQq||FgFaFGZ^Zf1D&!?ziA(`mm*?WZxqVBoPB@$t3u@
zo`ov;<0p=;x1KFh`k>1j&;|OS3yPgZYk;=)F8pd^J^2ekY<D2ZW?@GoM*ns#Z8GuR
z&W#(2QMauvs6+M+dayVccc_o8c9K<B4^5Wu3iEiVtUE|;X!DlLn-RG#!FLXU%#V*{
zhbM<64uP)g1eY~4$l~G38K(xelpt8@&=cQA&=UmRFg=gUvy%CF9*`!gaTe~Nrp4eD
zMS)|qD3@h-_c*L7sB`KfMz0a&8U2V?{}=u@Hh~lV_;U3>mpHA!_k8%_lTV@qW50kG
zY$t(6Np~cCa^3;t4i=Qk{t0W080P+<)uY27{~D$LfO1?c|NDafFI3h`X&ZJMl5pQq
zifElF0an~O_c~@xXGO3!AZX%B1@tgON?O1TDyz~)Hw%p9ahDA>V6L6a>s4G4`-<;=
zi37=X&<da2FUC%Huka4}9>=lYeV-JG6m#QDwJt0j^8BS?!1Z)1VP#)AW@;tL?)CyE
z9kaZA-OtGsH#0L7*zJGs@M~rm7^g=c--CwR(lOUg#&+s<sIc=(xDMuxs6LeB-8m?E
z!NL;_kq~NP&#UjJ@dhSBO{a0aPS}z;x*01s_Rm|37cf=%;FW)~;lmP*C13Iu;u0t0
z;h_?Y&Qx2REG5N&M~rIFn8gwr(e?N<UR*FP9J@_6q)mEpuKMj~?W8ZXW}0{G@#^*z
zF>h-ZBF7X%V1-{R<_U(2a!oY%vrW$4XtH)yo?oM;LQi8fYs@PyuCQs&IiQQjyxOUe
z%8ki!zi=EIy0uwYKGw7&LR=YiP?;I3?M&5<b4_JuGx)_gqt-0Jp>9I*+=R^W(rJvY
zaLgm4d|}Rc-v1#t@d~NrkZl(!U(c{Q^p3u~{#17P=>k1aycvGcd8jE%Ku;ncC;1lG
zOgdj{jQlt-T=d#jTvKU>yH0x!kvM%o$|6ZPY6I2lE#7%c%%}R{3b0P+q#6bv+93=o
zkYoc;zacr!kAxlEdHg_<ND$WC<W~UW0y2ND+7)Gjrtxw1_ZwA_JhGFvKP{`=xEicw
z0g}|8E8?Wy=<<}?bg&-q_1Kn?siQ?vJ6Gb+D2e5QgOAXE;nZdI=FeUtwRt@H>=zs)
z7F?cUw`$bB>V4DdmOJ%jta92`TVSxw?Xpx>h;FTpZnZjR#^7R#Pdp-;+iBvaYkX@!
zz<sv^j(f^0|81n)AF`cNF9+S4r(=jitUEiL1X1}bjfS!IP0Ny#d;_#-0)3@$ZvW>l
z%`<c^#*;|dP`4Wsd@e@YaxTWnTdo?FK^ksU_WIi`&lU3+!K@)^=PtN2lan`29Mbr;
zZeG!`bx~`(b5-L7;mUn;uh!b0l_}{{;*Og~=gUDGLOoe5{`_t|(TgBA6-6zN5HmoV
zmV!oDB0bDQ(~?(h4e{$%2xMV9-r_*%Q;5;Jfcc1UW%%b`_t3_gc_rI`5ssJrb);U2
zwdb`7F7xd%IVbwC>&L%^s+;wyE?gsG{`L#NWVPySe~bEjedeBu61u4jERYpm;jTdR
z34yAB>$O1DKzH#2Ek*!(X0P^G13m8afc%_y$SpC<^nd*Cod0lDFay2;)qkm*nbOey
zjt#`{`|juN853m7g*|z!M_ePhYK^w-Y$b##780CrkyAlaO^mAfeEkSxB{Mf!kIOBk
z59&j4J)Z3F+u<r+W5k~7;d@yb8dKjnBxxF?scnqDkS&)zj+)gm!&G1E3RNn*VeI)j
z#}&6F8=@oc$2vU??fHDVAY|Om$z|v$h82@f|1me@ub+SJ<LLOfrSSQ(LAaCiBC*r`
zvA1c-SNI#Qe8d;xT=ja9G1qTOqxCQgtLmM;P9ap!hpy7qv4f#|x^=y}UqarbJ_Wi?
z8lYOA3`@lpqJ8|}L#M9Wbl;zIE2(iPO3UM+YSVUfd_GXyk*iDxKIazGH={CN;9fzc
zyV-r`e_##?#au@N{W6b5k{Xtn1C<t@z>%#wmx}xT^yw*A^W}8%g_ceGwYU4lwlZ)(
zfRw*8aBH=|J`gYGu6%7&!$3<0$fPi$@X61G_%vOpr;UD?NHQpakwW^t^m$HpD1wIE
zzxSI1{&O#dxz*#S_fkGnGiA%Kt%9SMSKu>6v1GoU58ZG0E6kSZ=x|{_e<%KHFYsBP
zLw9A-66X5z`LyXoQ~-nn&jc<{il8JAmPE9iIh2)*E2UeTB}y^ZenRzV|5#;3tgT8I
zUhjsps@96b)`_me8OBDw8aRPPjt*8nhjT?&cXP^As9zOI1_b3#Kc+~*h64?=A(I`?
zYLLI^xXp<TbpA}w1KH2W#mNQ8vJlo}=pv2wuqmW|lXMW+zFNsoCbCJBNyo(dD;)LJ
z3M>&)s3}spG^|R9$R_HGWXCY3oQT@j8QB6w6cz=zC|Ku3ZhW9b>gllP6BQbxmsOXE
z*2=aE7SXrHCk|~ZbKKgaFen_a;@IRdC;V^`qTD~^R?}PyPEbd|vxaKOe<LK}gEfa4
zKDiD;kwTcE&Hpk9rMPcs=E5mMr0e$s)TQhh<o$RIeUTeue!hk_U_*{#RWzyk-d*!e
z)&4nns-;w7ZR@tkl|ln?K3QIS8T)z<7@UPwqR2!=9xbQfV&BtCoaiv>e=dT{VVV@p
zbpItyjisLqeD0U4Gmz<bEW{MaW-KK;m$yjU-$4L&Y@mt)>kU>od+)+=mu^bxhBa1T
zXEzXgmuEib%QY@)p12D7izxDxITIf0y%Lc~<lSVHp;8AY&oR;huv<&)dcm|E*qe@!
zBwYj@Pxh5`$LqxMIBbPfDQ{c7xthKa`BzP@+m_>~l*crBqY_?WPx`B#288VZipp{L
zvVmD~0IfQE)6Jt_*sjm{oGJ$E75pb@%U^eOL#?g0H>2)v(@d37_c~*8v%$%32aps?
zopo#>XZ>d%)6bI&3M5Zhu@XjzZ+vr12c8xdFx-u${Ho-Ul7cnd1I4RF8WtwkQm8@Q
z^0bKGfJK-8SBX0vNjd||*8O^#jnvtOGPgR7!;k>d5{`JL<=Qi!)TxxPCuTQZWIYM^
zQBZjJ_u@%Vx+mT*Cvyr{#2&nKyuy_e1NE?9Pa21hYJ5b?Hu&v@^YvvHeY|Z5{LOE#
z10s$|l3D2fibGA5itvWxBOs>|Iu14L+@ZFmOw510>thLK!fR)ebxjGzGH6jXa?7(t
z8dt%>=p<bPoyu!D-X2f=AUTE@fa%QwgTzbB1*kt?@<nJGe(Id@8%8!<eKu~OS6u^}
zzs;=pa=-hLW0*E{a*5k5|14ht%+1%wO?g0~7XOgA#(&_;HgnMj$e2d-Q)O58pXef=
z8eUAjYv`t4Asgy&Ubs5*T>CFvqxo%3*>E0sr&d}^tP?kRKTNotr=+69f<y;3BQh)^
z*395QF~n^XPu1BSB)#{+5=I@&tb1BEc9I8ko`a%>0W}D)II)g=e5_+kZT>-Ms#@+L
zuKyyY$O<~SV%9e}Sx5;b4$X96txPg972T4-EcAktVEE~<E7Lr-pPk>F6Y*BH+1z+-
zj?CLdr#{pJ&t2(IfSXcV)p`CtQWV=^+H7|!mNDG$F;XgKg61ig;W7a*+hJO=p~yPr
z6vVYai{*p<FgAKVb$a4=-ACb9d<(ukfKW3Mi>&!o)0r*72l{6DhB9(aS&bZb%mg-A
zS!XQVZXdIY_nyS9I5ULU*Dv0zin=b3(!N2Uz;ad+@KDjgslXNNv`i7ZAlHGQ+6uw}
zywa&oYAD|wa-`A~IF*Y9JA-Q0rSINj(&glgUfFJK7yWrsRjh#y695bJa94Wvhxw_u
zl#JHQ=v;5&M6hZ#ZpZ9bY_)ilk^}`8%F5g(drqM+lhL$JGDec160E-DUAf3#o$%4O
zfzy~V9UvK799y>>cm|!VRngJ8VPiX~Ko-wP^@kkwasr@F*3J2Ly)ujiZ#zpFwxh^d
z)L4i-Aa+X1&rbg6H}JPuX3K`or~VFc!oR2lIEGkw^%Cmpev*9){hT2eC?n{;(PYN)
zm#W7Q+Tjmb0%=`tK7`zgmw=D}+e$C}GE%5aUtiqY(=T9+)3B(2tmS`2GBR^8{nr=O
zqwgkDT+aV6X_t=R9@q%^6Cwz;$helklo6?VNkkeG7aBDL70pRk{yW4SDeWi^5ynK|
zUh69k#|I;OlM<0;J1kkcdGE^tp02=)Cf`(=%5v{S`}luK)-oULn-n@o`lRX-qK?ze
z3G3>1@^l|yC@eoe2fFS#S5`jv-na7kfJSHdcc1P#KaYk+s_s8G+|^gQ89#RPb)g8t
z6H^pxkwc?@SXny@YwJM4c&HV%>@Rmh9fL0vvyCHaEeD%%-%n%LkLJM^eS>6)1$+II
zg)TE4KCgpXXG^G47A*hLq}DiCG0V>gMk40GeyIN9aTtS;p8}li{`zkf3YBKeRpOt(
z2kK3R&H|6|3q1a|L5_XPFE|DJ+<@|aa4`%)gUKYO0iVexf|Zv>oVdztB?mwb+`6{7
zMfe)~07TK6WWVJf*zKDaqhDC6Qx4cU48!+mh$!K^l7R$&Az*`TF}I}DuHIhr#cYY|
zN`E~f{y;MI(8wD>2a_uf!{AlyLMl$74_XY`b4n$Nj*<5Uw@5a23O4)%x&HCvJuRR=
zz+$qS0mT|c1bJMfu!OTqT97$p?_|^hup=AE7gL^;as)HHP)9Dx+#P6=jpUhtB*vDJ
zk)W(FFyPB~A(Tn!NhPS%Pwr*j9xT&=3EqSig1l{k65|4K`V*{P4F`1!rIE!u_YOP?
z3)(G54R>25tYsdm#&?QDR#Kf%di#25MrSl-CJ33_0W(#!V%g<aSOodZ67K}6E_&rk
zWT<*b%?jv`VyU9gf$5=bHZwHtnAqeh3Tc9Su^&e*1}Z4zGTMXCh%H@NQ<Y^3`aZGv
z@>weA>KS@g>^rW^FIE?lBuV051qZ@C=%)TskhKjNk*%fCgf(X2G8}m7ja+hah^VQz
zc2_r(_G-ohQgh`E*}ZtN43oeF;Kxm$54;4$$bPbtyy0CMGIZ`JW>pD^GvqceMOu4M
z{bmc{bzM@)8e!x4((R+YGaXx#jzkc7xmHa_(Z^cqdceZTNn&{4l_n<*^>L7<`7i~5
zjR`B}e8eiLswVRVAwGNX{uh_XAB|c=-5RnGQx4ox)T~4)h5A{*m9{pvP?F9hfY#O9
zD6-2ZC4lbnPQ%_gZ^$5KFn}y;r$)Fne`E(!?yUxzzbknBz0{82+-qMz-W~?F4HQY6
zGR@myJ~(@&$#hA453k<s&6*v025WJ9{=L<T=F3Ts^K-Z~wwoL9?I}|kbm3<-GI*ST
z$!PM=uxp3B*)EJNfJre>WA!MX$_aT<#MX3_CL6~wcQs^Y4ouEp{5*C8)KWU<6!B2(
zMo&E5EBl{V4rXJ7v$`)dyK@xuR%tmd@*5=D3AF!hEe*C(nL^V0U16!N<DxdQUpklp
z)jH*s5|8lMZe=Fo&azh{Y5KysDUbSfeoRfhrfm+CRgiS)^-fyCsrW@*HXFo?wcoSy
zqA!(cmaDmjY@<)fylYf#J-Z_%p|OD-Jt04yNF@9RG>h=Q5Z2F|RUD6J&FaBlcH&Hr
zBsH{?+34$q)Idr(A_>jBr*&OP*LZflAYwrmDy^v-g-<p>6PYePJ0R4HH#0^vvpaCY
zoW}jlf{@0re%d}xu5dqu#7iEnx!lIF;N=6@My+ohRqV2tV5$8E7jd?Y7-QOZ(|+GK
zqr+r0L5igV7BYQ3H<@DvVLc;xgFR|WI()sm`TRG6l)BPJSpSQwD$Z%`d>>g-=ehz<
zH1NA5vw;LU=)(>*M2rQtA-`O>kruvtvtQ!(t!@-4E~HhlmbGwj!EtlEeNsuAeV}M#
zq#epPCY(R4PMB~&$d+)Rt2KV_#Ffw)(UwpIwN2p9d5b|?{N6<@|MC)QIgMro=LVL|
z2@Rf&ahN18*dT&_QMBV`R5&Uo951aSkHF2E35q~|8}jc0rbwmv;o=+$k~-}SfzaW{
zc!A8NcL%kR2pP1q99JYV91WE3jT+s~5AlI~MGG>NLDKIc^a2(dyd`6#UNno}S>>5|
zkfk0YEA>RGbjgo5w#kv@;@22rtcBSbZUS@3-?pmX{Cu1xggeK{kY%Nlkq7ZM`PFLA
zXZ&_yypCb>R8ynqWWUj$J2)uQN^i*RYkg4{$N7=$58P~T_48jF=^wHvX6FAlEcU;-
zv41}lIk0@cABx*1L`ELHaH}F5WV*1`S3f{;6Tkz_Xd|`se~fip?%YVG*hs1En`{Mj
zld-+rcI6I;L?zPpk7m!sC(rFgCJidvhY%;rkigH4E0(flPL66GZa(|mUH=JT-D%v<
zQ#!-*Q4tsT@cHZL<ii~oYL-}dpn;3=;l$QTla<r+!8hsu?BRn`y6xxjRq>bG$Q9?~
zE1WW~t!QY|vND6bp?=X5HHbi9vpVg}DVO}h1w+`UTh7e0ak_NxB%Mva^ghAAY!vId
zI$U3jylU%%W0`w=dByO8q(jT>Vnb6)eP_)MeMygf2on?-58cf;f$rk(@!4G^w4!YC
z%W#JsPx|7moOyU-dW!cN79N=U#RVJO{E1r$j``4&2-jeik~JW*q%ds*=}gAh5}I-D
z<QJRjBFr4wJ>&wNhU*{YD?45rP_0V(IyyG`t>GMwT{BQuSSxl)?d)I);Y_~5l({R#
ze8>xw#lqSebm@cB%=<Y+k(AvyCqsAYxj5@wVZWRz12PEd#YI}N{WSDH)dy>C0GZf@
zBY{$n<N_l(vsztfmgMp_<8c~gz+!}PMS9Odw7iTE=iXz{3B>FLd6D3yoy;1NcD=}{
ze^jA2%)kt2z2taCzy5$6Y}(wd$zRvOL@?4EsIxf{k(Ar|CX-IA++0?cfqkAJhwa3-
zaTg(jF_p0gt{1jfS3&(y3>6$${TQbS9+gu9VVBfQU3`l&y>p<b=AMgTx5%?%lAz!p
z4?>@{9H8`gmoe3GzZ8>-%3vP#zy`z`r<-tqfKOYOa~!QWU3xn^faDmZb2~C>ufsNZ
z#ADlRb0zP1UoI|v@NkEVc;$`_|2oX(Q#V$L&F9f)M$-xffR&gP<WCtwgLo@hTDuy!
zD`5THUaALCI%~i+$)>A+lXKEW(`+1MPG{&G8`%`AveZH|1W6_5N-Hawm&V-+p8%ot
zof=ZncD{A~*@0#53S<Kvtl27=F4n3wPY+)?T~WCt*P4aXth0Y-cqE;9JBQ`DNxsJW
z;;eV!pBM_pjHXggOxLHOL<7P=J@Z5iKTbRIg<9+he{OtXgz2&=v)RFL%JW_>&R7QG
z!(HQ6-X|*t-dhMlQ5uLVhU_jGn7VYNsDK045#Hxc%6B;?%OnwdX8TPVD}*3>P#Een
zCaj`D%^%s?D^EG1EuR2N*YV6dAJ*5MFl<d@IJRkygEInOuz085Jp#&irk}glDQ=5}
z26{tp-OpUQ$D_?@*PXE^s8KhTVW^4qjKUkn)nZz3hmDE`lv|J)V_y4=D&ecGA~Dk%
zx%>9rdZz7;TiP@+o*XgT9Wz{rJ8+Kn(G9n0ZJvvnl6o;FyKpeUrmzk#^?agTBm2<H
zwAQ~Uk=y^M0Hq-ygTy^$&(s_?AQoGBbc|obXPbC^7#$(iiBn&dr|#<)t2m3enGZSV
zODejMzxz2|`S>3^ym69i!2F{Vl8ftK778ZT|C$V3`u~bBBn$Vv%=I_vXM-=+#l7mI
zC8ZE7iI~uAn-V3{i*+A%+|y*D;+c|iNFbc{od7&__IkzJxf0z~$@-X-W$A*k<`GJ=
zp3&Ozv<fg{<q6wHN4|g1f#uBtRTg`>y?OF``MdJKWq*9db8?<w-gkI&^?GuG0)1V=
z`w8?q3{JGIoJ6m1uH3JU4BeEa2R$uGN>d@uMQKHIq@9FtxOt{v#ZsON*&<nh^m|0Z
zH|mMSn`Vk*pBIe#Q<)jO^U9CUBO8%f2ry0%P3ISZU>^+nj@jqp18pu?MC>mkSu6%5
z4#Kh98utImGmWt-T0H%cQuA10NJr*jUmSt2IGBhR$MBJ?_B7pK@|IvOrB<SpT8(#h
zGh(4Cs*1Cre&#P9MW?Av&Ld)@Q9*{2-_p+JTwqJEqWW4dzqvuPMW&f~tSN<4$HZt=
z7t+F95W;#5Zh^@<Ddfp<Q22#ba&f9mTOx}O<zaIAq2luz)CavTs+wqr&kDM1j1eQL
zvT``RHGo1|-^{>>d{2Frg<|S9_DGrfmpZcQS6_}&3dwBUrlF)Me2ZjJ3s36V7KK&0
zl9K3`-?IiJJJ2xn!Y~08<lbMGr)4QO^)1E;fEH!>hoBxYSrB!9heu!xp#?X`0c3jI
z<a1#WDXX8u*_Zh*AR#IqhP^HCpQs9n-T<oXU#%XU2ex|cA}B-OOEz2vGEbSii(!pl
zfl)~iQ(OvwK9Kc&Ma+o_C(-8{o|3`*iHsCTM$$Y+r<meIvOQD$*Zuc%in!^8>Q`ys
z^Zr!n?0D{F!YC3RWKgvt8?=}*g-bt(SihndU+H-dc)4?l{2SwD#f;^PGv;4;x<_D&
z3`~bh$Y$(sZ|CqV=%6SlfK_Dp6@I`G9)unWR&1i_>aXjxw9U#ow{;d(DQXYPkFNE`
zU$0RVu7b&mjdf>rQf*jsr)ndroIinh(!VCqa@@9=x)J52gs0~CEwTxm;S-aPaP0~r
z!hpv~y@id?SuUqzg#_e0;@@S*L9bbjU(M*WSOtSHLqdY^h?pv|(8DZ33;)m)6Et@R
zFcabA_JgX$bFqD&39?waHN&8LnoYgl>8vT|EHrG(Mpo316@U!zRR;D?WAkbcv*v<B
ze)nO!Ops1ABgag=ukqa<EyJ5eF#XXb3YueK{^ZvX+AIzHj&kDLWP1dLb=$w9Ai|=v
z<l3fHRD@Sq4YG!AAdPv1c{Bae3tY&ssDP|&g`_i%F)G^)<_Uj<0S<|+m90hhWk=y*
zQPcj4o6x*a%Wc9tWHP5OoYI^~HH$NIc5BuPxTas+v3|diF?|HtGHWOO0`MJg_}lmy
zc;r18FBClv1$j0AUJw(Z!Oj1Hfw=x@GG+PSqOt!L1EKxrSpE}El+M$Gu|C{JvI}E%
zy##`T2p(ig4X>uhBrUo9%P+i$EV>b;OjrGK?C;!Ohwq6G8FnVBxL&+pPK@5Kq#(dK
zn*~b;g)k~PyXlX_u^1h)xL8I?d;MSgldjF}#wr@o95f~Qe*C@yTs(wj1sX+X0M#t=
z-ki8tS&9?;PpCWho0;k0)2S|f0gozDweY#<K&2Bo*8VDo_7WX#7R{ovl{H;Wj5QcJ
z>FiCr;Z2GQOT9E9CitL6*|Li_W`#!hb9v{lIr57OZghqSF>@`Iip9KkRu$@qtt*e!
zI}MLMY<A(H?8c=_Y6+k?D{HT~-3U_XMJr5uFRE5f;+r)uTvZb{&KmIiyD1v|5Lcyi
zThF#tjJD{_E}Mb7RI@0p_nMfQDAOLA=!eQ;SB_N_RgU?QG$YUeBI;pK1Yz8qxKOF<
zCZoiUj#1E4Q{?e@nSwr&#ivuA?J+H--zK&hkOSsj8xJ)4w$V1oHcOiLX>_V8IN@F7
zT62a^H^wDNiZq*@lgCv?-kcNJ)twu$#p7e8H50&)SgDriyz{AZVm(=$2;-Kmqfg`)
zSeT1<Zf0Rcm?=DSR5@VI8`+(HFQj7S%VIB1qoy?UreVn{RO7Smrq9YhEuo%_Cp*+r
z7lGJ%1XxsOe8FNbDS9_}2FI9xLCj>!aoIs@4GCrm#`XtVyUG(dmc+LwX}RituxDUS
z!ymV-^e+OICp!<r3laX@gCEy}cKXGpViBIOiy_3;7Rei-x;G>BtmmI~vW|jB8FaCL
z%LM?*nY3m@hZz)FUBxj}IS5Bb$t*hbiw#9sMO}eEQu(+6_2`kI;BwVc<62>E0-sw8
zH7f*aSjJ9M{OpyWuaS<UH;x1=KpA!6>TEYL3ESBwNrW+(50`e90&DGzw-Vxy@q?8x
z;AJm>Q-fhX4q_=Q@{FAYl~t7dGPc!*K@z#le|-+=-<L)dQPeZrkW*R#S8P{ri?aHi
zJwIh%$GzZxG}8)<DX)IHg}rCfG{&#56Q?|-(KyEE_`Gyo%t#&m!g3+)t3V8HOx^Nb
zO#re%yI5G%Srv6usfhz<OR&dUs;<>JnG`RLnz2DM;CM5}`4VVW`CvO(dQ>E(`;IBX
z%~$_mafiP(2G9f+4ETqV))9MO^56+qVjV3HClD0q*W>}SmBBr(K+dL-oIPAyf~drF
zV2VC9WB7g@oyp)Z1&h>@sCSVj0`KldddxJ?;!IbHQN(WcA#=hoDllxLYOR%_w=@E+
zwKjffESlccY<49P`;%z@tuq<n$?>m35ExfM&+e={by&Fi^XAj3Yd>y&zu?Tnb`(U4
zdUi7=dG)|uT5|UllI5nt{fV!{Q(WFBp=G@KI1!(58W|C<Op_fh^AuVBsvwFUOxps9
zNf^bsrx3$vb(619>BoP@&_lI#fcwcA&ys8MVd(a$1x!3XfGN`QOV9pXexj69->3H0
z{{dm(Vo^Yq4EP`SL}sS%)8YS-PsPjv_^+k;-5N5=2mezKM#b6Dh-2opT`?Z0D0rm?
z&S7ouM1)p?1#I*|lxCRgpHC2LVr>%$=LjG+^Loub`+POU*%D{tzWQLf#Lr=}B9*BI
zL5H^|sU@LCD-;$VxHu(I3vrxRC@#7^+gtg)d$6Y|04#Mdqdy%V2MPK$Gxu`iJFMaJ
z4E_`!_pv{=()4q0<i1Jh%jx-SSM(1>OmT=)Y#Wsloihuc))p+|kcWw)mLp~E;y$#6
zQT1Io#hFnaW4DvRYyzfWaj4eI)qrZNPljC68HvJ(`p>mw0aPOmvB_8A@7mz}#yKMn
z#@EOqHoZo~Pg0#wE7!ad<1f|k9**XYC=I3lc5)CpKcKw`?Vy;Zv#b5<O01FtIAG1_
zT&f0$8&T3@Ueac=fN)BWXEtBDDBfIdZIO4T8&=er|A(=6?6P(3mNnD1ZQHgv)3!0w
zwr$&*X>+D++qUh@ll#=JUDevV+Bxgn_yZ&E{`4LZmzajaOK`j%MFhxp=9Ge6(rfmr
z?E1Y%a9<@^#wdX0sy(y*3p{&ox|7Ez5)6Rl9!-!x97Zo>A`!<uuinP18LRdMn%f6$
zrd&ha_5?MvEZ%VBhe1~IF9X<qDuNV?ntQHX<9VW7S!hLGz`%U30E$rwNb#2-OdyH+
zxx`TK>is_~_<ww(nWw*qYG{5L5eTeD%_<LYaaM&d+U_>4(f^ruS5(l1Ff1*jpye6j
z`8?ipxPPf|pMoU^l6=7vfv<Hj@vt)&FAo*qcHre`B(HazgebI1yvV6w1>;)#zMZPz
zxa2wEw>5SAjxR2>bS*CDwp#h!_hS=VIs`<moFs$%LliZvHxG(6P_YZX!%u~6z0Mmi
zVxF1_YFQpJj_*ympG-@3ddZ^4qigm!o-y3iV@Z5QL9GdJW9mV=b8;7DqGtM=;&^~T
zDc8E47isWLb0<)(rM1H9C4q44y5pq`maQ&?qNlHdL85W_vkGkiF$9o3D%m~Ud}k76
zzZeKU{OCRPH;-(QX)=8FT2TDo)?-?F&;nAhE2*P?X8C$xDT>9Qzqa1!2*5J{)W~W;
zLqq&6@c^sC0=9{H;T5oG^nGXA8pQxC*P5^{x?>_~3bZF!YBQ=*3RJhKSXxa~8S3ZU
zSv_O?{7Xb7lzkY8NZaLMrNAfoFdh2_*VkKI1_fy`9K{kGXwhVyk<wT?`&;A_+CbF3
z&x#HKwv9vCSmJ?Owq2tz)`zVA@BP$9H%pm>v?t|0n=P%_qwq(v6Keu}njx$0r#5Rt
z#yo&`RUb_Bx944hdtmSClf(>2v{dY}Q`Y?l7}Xm=Z1BWb6i1R4L^BZ-s6~l>{woH|
zq2cc0YKHucH`4_PuwMoWmb0NqP}f^1+Dab%WEjFc3mWZ!s8Ugxa1j9t>cCHZvzm+V
z@e3DXiYxI&<%vZO^dz8F;VbZ>(k=N3ElN&CR!&onS#k*L3mxuVgF_`@=9}hq<R*^?
zH`>0|6bfu<nDg$x!8<mbWd_2~r9ych=`NX7EjysALY_>=W!SY$u|lIA8jjCx(92~n
z`1Cd(I=3r}>{gmDeC<iJq6IJCYcG{f)jJu<t=oK;1;Y`c&Q)`EtpL}%A?z6Pmp6sp
z^9BT6VWb3DhZv#jQgB^?dm1hJm~|Wz0EBd4;q3B&#i=H`5b*B2Tx_dfFLX8TR%w^K
zTq4~~XoruZ8N&_g>K1IGBt3Jibqn0hy7pE)so!(W@qT@6YFMgekVUC@&%7TMFHT^S
zk$+21EfItVU+Dd|Mw9%{k?==Y|7j*<X5jd@28y3dnC%aG{pV6s%8nV0>F`1~BdlrM
z(%!?D!8!wEflwNgKyfZ^$+`Nl?Mfk)Vm=>hwRurYKEGfW!hApZ2Xpp-EPwQTORj(G
zmbw@DM_ZF&YLKgu{!3fQ`YR52nr6B&x_EsCOD`oDr6JkDoRx+6za8%E-8`1?{ff$n
zKK*|(Pp)&zM7TMfubbCDyD^WWmMt&0?@^1(**C0Gyh}{wnjj_qpGs|`Q2-W3Rd;G~
z@9O#CN(?jK7{{6hnvy~fNDy;xm7z|$E2mvi|3We09;O=e1lps)M07APTm$0py-NG@
zqc0@U``4uYL-Qw$c^e7^Rd(>7;1l0b;h>f76L`ej!|BNP)V<Y(nu$I0=-H+f-}Dz>
z_eoeDXh*rCn^_<<K^A0<N0yu|T1)C71@^Id$m=*rS>eFnsF|QP8mO|cLpID`HxK6*
z`81JMXF|nu@9w#)PHbc2bGCVot2p@Y>g5m8vu3AzOW*El!96vvjIF^k{QIhWPnn{h
zc6SWAD9?PXOr2C@uZ=XO^Yyc<6cbeJ8pq{A2%vL%i`<%jmfg=&mmAvj4pvKj{x%An
zO$1>?kbG?uL%0x1U=?uUAP821Nw-34`7ib&wTSpt)eYnZlbSORB$L1d)`F2~_>VT~
zC!CJhCLIz4CCivC*8#8y=r19JX!%$D(Mbg!n(4ifXX{)uh2S9Vt3?&A9go{d)4<NL
z6rgog39qaWX;!IM1kncSl~%D_Ut5K>c4Krv5h(gxnz7-SFT~~mG!ibcn&!nn&g0x&
z=zy34)_ULL`H|`=Vd4M!;Vh69DD+#k5@>V3u`ba4iGTacy4Ff@%@h>1u`UsV6miKA
zn>a%XpLlJ+oc9G-GROr3t+dR>+2_1nf-Od_KGUMkt#u50`tOf4m!rtWCnEy8F?XWa
zrL7{Ynyhe*aXQS^PeyL$Ae*`q+Z`;@+MHw9K#Mfdjo}-b*5$C)`57ESy$f76`Pzs$
z_i4X}f94yJ76AUm*)I;j3AX{s(TvQHm^`Cl+U25~W62a9P$SSJs~VXFleP>c%{N}2
z(^;VP!X(huDvRg_lW;kWWkezPU@$$WI9Xc^+5c4UPn0#P;kK;3RbvxmYs$lJ4-jJ%
zExQ-mFOpA1j(pn%NqKqcLS)oI9JWGoS3sOqp7B<3|I@#v*GXzkC9syXcp?}rA(puH
z3MyNDv#UkI;II5xjFg?1xYTO2f1q@|JQCHQY=TiQbXgj;1#2&u`q~jmf^!RhRFT(Y
zD5+m87-xqS$dn^+_9+#VZVk~MIiS7wz6dBJZP=(0&DoNjhMkoqgYi&Xnl9!f*%-DT
zoy3!uq;jJbL@CUIMF;$of^u3_I&ot?`pE59rK0ePDUQ%#RgiIp;+J7j9&Ay|0nO(q
zx2B`pc{wt!lxo;<&Tx-94D5(1WOuzrdr(-POag@io%^gk61~*{p-G`A7U~<zy&`^-
z7s5<;0@Z@khU_NR!KSiSjKLyhx`b2QmqGd|8<4OrI}<1}bwvyEISr71M+LDWXSJ#i
zE4gDR8j3Fm`9+|h2(4cZ*^dVTj#j7wFkkDAZ`V@>xeUx=<Kk<ppM+So9K!S(m2dUQ
zynwe|&6SG*qi^$Y&##tW<A$!u?B%*vP%)3D^D45KjpD+d{@B0D;`>y3<1pVOM4<7M
zOOjyyv4xS|>ivt9bfVHLKwJ8Ih|ts8v=8V5JT^l2^gq>>gY*Ay^#4DhlsFmxKa>(1
z)4xV3O=`&6;;_SXUf0ajb()JtB#HDu#K9L)&HxD}3TqLEvaF&B0^1sJ3%p&wbLb9p
zGe1sUxfvbTVnnADln#qW%r&YCLrs#%)48Seq~A>VOAj`HN@CFz=Zzj(%8zL5$!qp2
z30c2d(yaCotS`)5|NW5cWy%4;l)}ViMSw1xWhnm}2C5kFOUggO5WpgDt>X`f<peZw
zlqG0k81U=C5Ee&WUMNuFIichHY(HgfE8dXanJN=_b=2zZe#w>+aOP=WYj9RQG&I)Y
zWO3Z6_KMIH!@L8W#(<i}>`j{kl*TAFmIESwaCbRzVt%cvZX7$-MH9^A<RxNMJ=COn
zchMezIs#D05-SMoN063KA;>7|jFBPY-R?~=6RnI=+}=tUJBpG(JIXy6%kl+mRCyOp
znP3LI8pQ*9{~K!?^bUnADzq}J)B>6+1qCn?HKq@kW>AxbODvjTzmBE=sC{k<k}_jo
zEfTG~EhrQ>GZc!8sWRLe2wo+INyKVJPyzt`A{G>&Q9n@~mTg`ruHA~x3L@~*V0=pm
z9%#Xm08cO+0!fk(W=3Z|9q{BNC)gQ^S+)QA;4Zs-<I998rLx0hZ9NkMTbd%KW|fBD
zsCill)lHuNg$g1lm`+oxX^<qXUkp+}e!%TkKuJtJB+??UzF*m2LnlH!h~MJSvj>F1
zs?fCL8d+I1Dbvelsk;XxG`ZZ}NJ6*hjiW|D-IJLV+AG!D2vRi52`~YaUQUmudHz0_
zpnq4AHwwig56ABx+tn(@?XXhlb9v#%mn)`IcPD{-+i+{XKh1*qx4xP*X>$8Y7FQ2J
ziCUGK)xYZhs{3;Jw){Oo5?+41Kl==Akz2~`FOws@X<E7ioA2NqUYYSL9k*WKhyIge
z$TjZsdFcuQUuv{_F+UuuC)4U*(EWkOz>U`3xkG3k)sffd3fw=Wzp$#`=-s>08;PrL
zMejC{2Pi)$>kDp9b4bXN^KvY>zv@lSLX25NXcJ%iCCNFbe&5(t$!bax8u+>j-IJ|-
z$vxosAlvHsmardO7az4D+>z=TwU)fG-_XFSob7_$y1w`Rym&umv#!s;P>kK8-u3n{
zZ1VgyAKK83K2Cmq|6FlzGxlKmYWTW;cS`9@Z*#z1Nk<+5DVZH*BqlZZ3j;M1i9Eaw
z;Q-II%MCS!zf!LTxZ`CLRLn;`S2X&e)5sY`Se?XSP*;mW)s0LkQa4k65;3zM7()mr
zb8Dp+X&pFC!QlaGe+{M;oVB0O)NMr}2ktRvB1Rs6BtIg&6drE6+)>EQM&AJwlx%#X
zhhz<8kBorRIw#ZTXp$M9S-wp=nFEz1W(CZFtQNV3gmIL2&cP_eJSNHZLckqI@(vgy
zKxEK|qLu9@Jzn-~Ch?0jf;b<m9XV{>zfjiS+vD4`781wN2<)#UUQg7-Eb+k_oSYfQ
z32sKK`X6L9IWR~_G!nYuYWJ4&u*?Nkc0v<8(xR**GLq&(vJA_3K+&pr-dPl|+>eqz
zPx6n{(64cp%kCVQsi~lFdpG_LnOu9X%**cEkOcU3`DVbCz=Jk92L)xx4YETA#sz5C
z=-|t$x0lR@84T&n$it?|_S%cl7eWwHyTbh(XxD6z>tFnso9Hi>i`pYG_IkBu7n8cR
z(dZr9*?whFXd6#E64-VinNp+k4l7<CY*tb+D;230b;b%RU8ZzAY^+T&JjGngN}46S
zi;4c=MFSb@E#kXOTX)hK&@!t=dbFS3sG*}a!nsNPf@=3_vZ1bK8kh~06Q&xgfhNV%
zOwCxCGcCjQTwR2fP6!NI!6v+CD#(;E06r}FC|dDXC81a`5V*TvE%jcv#~$9{tE%y#
z$tIioPxWv%*wT@0OBS579F^c9h8ce&rRftYHsq?fCKlV?TyyJ}H%*6H+Yu{@8SB<I
z!YZ0W?<@6go3Q`~k}!;hajbjv&L3K!A6GeIrQ4g$ESANJbrLPIJegWH-f_KhTSQyK
zrC*>bY+m&p9uCa+6U7?ZR@U&yVX?DVWQ;~V)EwaW+ctCP-gH}gz1jTsjcP`%0G>IX
zGJcWy_e&@0?0R7va;JNM*eFUFIjm_rxa3a$>A{6~B@8txcJOU;1Y5b^#Z^G}X2RFo
z1mEtREV84qkiOw>vC_W(&v%TG;pcAtAL|&-|F^L9U&JgXRi))N7!bPdsU2iU3@>hp
zU2IVR3lstCq&@Q>+S(1m|H3EHBz?Wd>>IN~`h$im{z{L!bobtNlL?d*2<)$#?yt%Q
z7L(NvQ^*V0b?#9i2qsb<(dU4p6B5NBSD-<Kg?F6>4=mi=rvQ`epKnO){Os-dzR329
z#4y^u!mPf&^gQf+z1_MwIdDVJf$P4&f!J~)rp)K*rI4ZzY3J2yjG)#J0+M3Urz0?c
znN<Yu$F$*XbQO(Xbw_cW*|o-F#HACnP;eU@p^hNJqrI|@!4@ymlI$C7iVbAMGlvne
zM~A4`0FZ@Xue?0>e+=3RgY`qqUpKq(`l!SQ4BC@ghqt;RMg;r#jr*2OYwI<ocgZw)
z%qQIjtosVBa=Jj>P_7y2euh&OVBfR|8876PxNns*6=}stt^zd@zy!Jr6hXP$6PRWJ
zq+W*9x*0~k+c{bCbj<Qti>{&JKn?`BrWRE~coSmkl5Pg?!^~iAbre7{wjh(fp*uO|
zM5;~L=e@*AI#yZw7%1hD8<P@=gIPcUi{JdoE_b;gEy|X@D!yjHl}lq=9TA^hVqO#r
z#b}fOQ5=v~D5u#pk&#~wMjOH<nwWaEy?;`KB~G5o!IeKWJF3yd(Vj`yu>N<4dLbef
zBr`?C{z<oXm8Zp*jt*N3XWFO)WvEgCpl?>o;tjM%(eUA~cx&E@OM&ONzFNvj2J7GQ
zz-l2UVl)pw6<~c8U>a<I3Ml<WAayo?YU?F1>y~Ur4J)%F;m2^*o4;0_Zk;2Yaci@(
zp+cFMAhv@9&EbNIi9!9?OkpZNow5e^-aBmru!#Cpbm40DWsDO77+GFNjOoVD<#BQ_
z%q@2x3`SzeA>}DGp|ZVz{T5)C!kHzQTb@D<F~lne5_Td>jdF`cDVcvoRCS1!b>=L9
zb|Od!xS217H)CF);*NYB+q+nJJN=Poa6P&$v0TEYjD-NG@9;!yUp**iwL*0z9-0eN
zkvg5vV8f?^2tqRGX4|i!&iTKMaXQLPRN+6~+8zDCmYNcSlxEqyGLBXav!XR)^K()$
zy84mCb^XbZd>=dyq}0uL8WP=`JFx#aOHJ;F_qd=qyQ1>Kkw`vR_CG0s>ZnxMBm4<b
z?+ko1JPyL#V!S*#$&EGn>I^|usQoKs<7s57><mM1J6E#%%ixo`;bj4O>C5h%o{}Sh
z^dJA%{}7*IWMTgA&h>wBAyU_<ndd<9xvse}qC(vFcE_*)0PHtc2YX|f1I?d#djhL>
z%Gj{>`NSFimH2R^NFuOD+3e}ne9b-V0=Y&MkVB7Gg9e`q8AH<?LD|S9u1!dZLMBwO
z2$oH%O2c3I<@FMU#GO!(qHhCpauSsDaeFd(cDimdMk0x#TcQ|$a9CK3HhOj$>D%{q
z&MvndQjA`s*LkmrPVyJ<tamGquL<mTzXBDesVWGuUIDQX@mZD9nJDN|)T2)ac>)Nf
zwD=i_spbO{dnZ;nOG<_d5|U>VWiYKE6})2N0z)OKXy8Pkr%V}gNl=23Qm)y);fjEE
zAWD!JVeAlD_v4+>c4==C?u|LzR_x_l6qJmGNXldoMI9<oF)^iUL*&!LO4iVr!~h2w
zOW!;x9*)E=`{@4F@Ml4v_Mn8%^(Xtj8|=wan7)w{J_8{$p5}@{d$i)CeUb8M(P%|_
zY1yEn{eU2SW)LqV_)9g-MM39`XWxi+25QODeNbNr#<LBCcpz$p5K917+>uKFqr8b-
zv_bVGj0ym7l`yTLblaZ@Bp0FaGRB&b`#S>GA{#@54NNJMC4wEiR(JWwSG>T90;96Q
z`Nh3$P*7kHg#i?BBn6W2zB#N$I&?F9h_#+r0SJjG!R2i8Qw`&yAOrs`+Ala>3r%Jj
z$DYhUrhw#}BX1#Tf7;_OVn|*BNwW?k7%hvqS}5k;uA}S{^^HgC1ym9Y&@G6<+N+AD
zw0xq2eh0nfBOf9M1ag9+A2U{qQ+0LD#KK}>DLln~B&7N^H)SH4qNzMoId3ZWEYt62
zcVmD;D-t=#LCt59)hoWM4Za%o*SGz@T^Y>!0EiXCupH~x4eJf?<I0E$7domzvy*}Q
z$z<Cpt8R<KNkXtfp%0>9Ea#chyXDrNw{;q3!R>|Qtt+e}?Vmsn1HGb1$j?0ORvA1M
zeesR!^ktG~O}{g00R&AwU)(SIKDcb@3(O~k)4c89$50Kj#7&%0*{$#9RAhD4+)C@{
z3<`unP6oYHusPPu;c=HoRZ`Gb{}566yDV`7nh~T+f2t#LgI3n7tG_Av6(h#K`<S^h
zrhs@<r5k#M!fl>##|<ylt6$qy?BaMpXeNu3><p9ajB?Si$Z{0}%bne~$vnyiYeJ_Y
zayL6Ujxhp2n(otTG`(gZ4*A0YystoTn~3xj_lMo)<g9oJS@99E#SR?t*0~e7tO-hp
zhnwh+=dPK6@3qY2d{V5<D>bHm?dWfdzEii@oO|C->R5#|`#)&+G_C!T{2q6%yQ??9
z*ZFY~X6QjFyW!s8Lt1Y4mQ-A5$<e@dCa>Z<8}@EGEHsw}u=};Gw<dA+>8!qEp@;c&
z+A{*FVNCvMw)yok1VByr=rliiWU}+-gR)@tj)k3og6-RjYa5aDq=7*J+W=f_kTuMM
zOX0oHTIyGUxbe;%3_KtEn+!S^z?`H}oE2vJ+#93S%KO%F`g_(9ldTVXK+4=&__l`^
zi_2>U0v^?$Jaa}vYF8)jBUz0ulFw{#$$M=WV4A+dnVY0X7?Y<glDfr~Y>X@?O0Roq
z=BmWf0_M`BbT2KfV&j-}D;1!PCTV?OYiGD@Z3YQ%lx*i2sL`P}v8qDA8jy5&S0Sy%
zfYtUR6%dt2!l$^h-O%3PXew3t-4$7D&N#cB>gqh}3`|r>9Mz`~4jXgB-H*KZx1+v@
z{g3sEt><Le8n-J|wrHCKz>QdPn_(<}vp8!`q2QykbL33{w<~_O#o(#z{S79>UDv%M
z>l-_y*4TL$F1|1Jhdp|+!8IN6!QTQ2-E&mW2@?wq1{Du#7WuX&3fI&%ZABlMqoji~
zlNh6Ld0x~;OQ=V1M}Q~koY?Z3EX_W3+&!knt`cOH4nE!X4VYM83h_@m|Fcr~kNNFK
z=l{DU?*Fdyi|YT=I3VsGhWUjciP(>KK}*Z^vuki$7uEuCpdkk`u1Lf&+tlkOj;$vR
zt-h?pttC$&a_?xr?rLwaaKfKGd0cgKaJh|8K*lsgQJXc!gG3z_LChF2XNu$-h^P&M
z2~nJVIh%YuJo(fgA&nMStDzsh7H@TXTz{PY;dxfuer!F`bF{g;vNG^|ecpXe?#v%7
zN2-ONbdgZz3k4Eg5fWdfGoPuf<GMcLM36XZ!nKi=DNr0(H<w3t-#lv6(Pb29?v=k~
zEw7fBV=n|TA{G<KK$1rx8gak`K!hZmot?U5Z}-A>^k80V$3@&MG?us2(+kxLP6icU
z&c-Uw#hm09u;k?Ix897cze2#(`^|?$NeQI~K!WRlN@epLn49Qq@K**LP1|R^?M8;4
ziR0Y!<3~EH9*wKKk{W|ZBPW<IYBGwF`u60jSKt2~t+K#v!qndfAt`N1*1wS8Xk81(
zgbj*W)=)5=pTnL(8RzCd3<^=s7qQztbgS&x2jU11GJjOe`s)@Y*Gqw_Zwam7Kqa!D
zDg@$Bn7HJ3D@_0dNp<=Dgdlc@AclATsSRcH5FdWvlqQB(g4d}FqLe<}9WhrLJNKZT
z)CiUvruuEx>Li#Mq5uRLa5gV*c<Nd(*Bw|kT>A}Fatp~&oPi5E??x|UK_w^c2JyqP
zHNz!*Z`DzebLZ8!-~ov>9e*EYs!%vHw3W=_r@L0IwFJ`rGob!pK9Khd8Fbo?INYB(
z5%g`l^N1zYeo1GdVi?P8^Vs;9Q@#A<G{4oeCwq8Ek2V|CW{q8V@SMok(hj;VSj}Dl
zIA0u5ro9K?H_;MR4G|n*QLEq}JN=B?vcgla&yDdkFGK^H4K5>r{8$<^h6ThXngllG
zdhN<zRE{EdPji%{BFTiCm@EWx6opYX4s2!E4vv0}T|nI?nsmVv#V#7QEAaJ{hAmxB
z*RvM{SXdE1zZ40EJ_3{tY0v0CLG#x}?{SeY!V*|G#hEOBAznt1)SJPl`k`)V4af6@
z5>A{F2l@QjLWcNX6aNxNQ0_p7HUiR8zW>F}8(x4!SxqXE!nusfg9kC-;XMF$o&mqB
zg`ZdgpVspqfU&Sej=7)V*00*(E@!e6=8tHc(|Txw`5TLvaaFLS1U{E6)hq^&ECJ27
zrlbD5RC?(h0D~QDT3!M-qoNh4Ye7yl5!n?DVh5^e+CSkg%u!6yyi14bb)V3c2RV)W
zg{=i*f)*Ifb}n+I2>oh4vYU+SG}x04dic=s<3N&TuVVsTz832vOSYnE`-Yt3VEv*?
z)AX<bMj~2iyKHlGF^=*o6}8%oWQFjdVxmrVqtEn9!{>L!J*~PcbTwp(MtL~PnU<3y
z0$PV&ErxzJRwt#pU%iA^KY>UK*XlJI%DZ!f^g6?e$&syr`{L)Fr7MD}CkiP^wSlYJ
z_`8>;*`SVUik{u?br%8=+-%eX1X1HN=!^3DQ89Nb1H~iRzy>cHzy`_p>e_^z!iQ5F
zm(1jAT7ABnGtrsLF9lPAVRboYy{iW<{qvH;)#dCfV89homc}yPRlh%ZU{b#8yk4T5
zgeZtNk#*N8F=F5W0?Y*>^+s9^k7sb=X+g@R9y&s$nQz*Q72dK(N_8G9lpQg}{jl_P
z&)NYq8>zcw&zBXzI78jHyAdGmV*<lIwSNw_TOYrN4)~7P1^hOYk@7`TH?*4$j(8o`
zC-{bIjBs$;b{#_khcV_@>oY1+7PBj27N4fzDhw+6EnAx9x!SCg9?fk)K4fh7VI;F7
z0LpfW6R^l)7V`9B#sfQ4<zG;*d7WvhhQZCNC?DD^<Q_OJopCBCYcrlZ?SRUV#gMGu
zLfqkrj`izFujz)BP8cq}GrnfOIqUkrQ+0n$eE}+x*~a{%+W%9pArt$*Hcw3c_l@sZ
z?SulIF_m5c#ls=OsvUGq$Jm=<2F>4UoDEt-igeKCY1y?X-*~Pr(TdH&aSu3S*Y4>3
z$9B|9^#frxL!WPOy)K48N}#9?IAubh3M(Q~7VC9F{J<l#d^mnMcD=@w&8qpqtIS|I
zIpKfa%pc4@+*z_5{oqwFqkr%!dSr#_A@ddR-Me4Ia;a*qkGuOdxms0b3mJfC=@2ZX
zY6U9%SC6soGEN0IN#z?lwhit|s^&;$8}Z>NyIiNVYcvx}H1+3Wl}nS(O6lhv8>^Xa
z;*3L*NaCFv_$&oeoJOJ+9Nfe9i2*sjc6|Qum}mDJ!w`pgkzh0khzgL<L|BNhy4f}Y
zpHA8S%!arwF1w9Qjcg|V6%MO0k{tbq7jVNU;0y$`sx9uCyB@#mv#I~|k3Li69#vSW
z7MtcAwJpYfdjD>=$U%MDDQP6e0A;_mQuCJq3Fi=Rzpt#`HKOga;xxqC5P99mBU_P)
z1pReJmVHQ!X4<NqvUpTFf5|8#=!W_{osl2f-QSC{URmD2MrZfxa-#j+`tB|R8QoO#
zj&u3b=ULt<_7rCyKt=T8PpplN2ia{bG89g4Q;DA0^oWp$B#UVTf%E-ajAOCC?!6T_
z)n=?*aHOXInl}m=$E8GLA4bk!l8d@!)y-f|MWcxGO7Mm%aj~_uVJISr!$+_}%8(c2
zH19s{;so;G07$(Y5<JLtAU#S1$3`g;7~rC;iSo#0403<i>ym0OrER317zRCI;H(Ws
zjjmsZOwS9sz4)hU;iyjQYL%APMS5HjQ+LPjB4xzJy~l>L){{Q;BIS9N7Jh}6w*8RO
zg}GIw;2Ybe-At#0p<QQl+OVZ!2lK*ZusdXvg-mLkecI-W#pQ?jx}EB*KeFYGh^YLg
zv46j5n&@AA|Fzd-A>)}|IjC=d<tZdL<#NRUr8>TDxW!e$hewa~?u9x-HXzAzvj}7|
zL7F7vT1R6`oyB?F^xp5j>CnP<|A=_|MwXe$!md{@S~FdKPs!HZ&=>q!eVAq;X`0av
z$AWTm2p}#Yna8pNDsKTY;q*!I>s5in2Ziux+E78bn@k<WM}4nraPc#?xAe(#qi4O^
zhTS@Y1E9sP2;#t10??mI3c7DCmUC+ZvMc&d|MeB%nd_W0PAjc&{5!z(QeMLZs=Y<J
zq7S7a1BX_C53jrV<%e?JR&9$mto-|=VE;vVHGQRipEW~{jyr$}iSjBXn{CsN^)jkG
zE%=izxaSHcRcF!vFs>vdtg-sn3*?L6fFA_HnH1v1Yw-yH<V-v=owdgURZ?Pfh=}RA
zd}i0xA|ZRL4sJcup9q9MNNeO<?c0<rX`7o=EZlqd81<;w1+F{_Lb<Yy2Q>>2**<3-
zXlR|yC!^{INKvsLXcSj<ZZ}=wp{`?~balWHv|#$%{RJ_<^bz?$j0gz{aY9|?J|J+~
z$do&rIdnT9a9*W*<Gs(%-$BTbPx6Q!1K@3ky!4rFz!UKpOn82F1MbP2M93~?tru~S
z<0a@1aggWxBOu{~(tYB%GzQm#>P*F9N4MYb<6o!V@Hx+)Rr*wTpd=4e5#42>d&2rr
zdMD8miNFN)7+HtF_TSJFXI<@Qb>`nk5)%n?U!VzZC}@wAG-t%>3$gm~0h1T7`nLzd
zw@LeYoT6Kvo7%sozd)x-J$wI2PJg`k|2aG}GydyvYp2@(1GPU_t7R`kuD<=8je>VD
zGE1n-E~|P%GUI6!;m(Su5HH9+*ftaNLy}Oa+jR0k6xDBQcYCjOBN){YN@xiz9NpOo
zLdsOHjVLpA$h1qA$oGRtmoXh3&Ih8>Pm&T1`ExxLJB(hKB*}!6+r`WG7vnuOv3vSE
ztK;3y2lHrZ`-MLh|Ce|Q>^EmuD~@+Y2R{cQPYekpLN#MNSKQ*1seq8>=5cR$@$4`q
zI#QUj<QNM^MMrEyr6%gO*UiWJXZ^%@`N$D(FQj^)kE9SIy!?q45*(N{8MG72YZf<s
z;~^;cik<s8IhAd12sC~JoFHP8H4GMzd@zCKdJMytyP~pmZGECj-BX=yYK16UuJ1GC
z#2vMediwOofh<OK2B}Qyp-~r>0&_=y@JrqMp&Cy4<<e{U$g=N;6X4JekU3E!z+|)>
z*|tx<prxiSp@4ZzcM_CdyuDE|vXmdTRCaoAQyoe41spavs=M%!M&U-NrlzgyV9-W}
zBAlwZ4srmX35z5paK8q_k=8!Pcv67i+#j$ahONepqj0Zs$V1isD(e^bR?3y7c%UH}
zBY`)9g`T7yOCYuSK#8n)FhPwzK?U4HW{w<~E|?4|h7cP}G{I$sKast|awO@y8=l5+
z;%8n}wCIx2{_E=Z2M#xCD~DB$7-j;>L`h)MZqNZ}#H8j88auXxj$=B2$$NON0p>(e
zKVeQQKjtkI7`5i6V+q10z{Pc)eY*2|T2AwJNMBJBNFOE9+Q}V&%TZ3zW~?eps+$l!
z1K~unhHNwi_?Jo9J*>smur>N-YNu31V3>DN3bN(?;mR~K_Im_<ouv62bjs29rQk)w
zj?0rnKOK6!BNh8tojh_NQWy-y-}p{7AIx~%Yrw}ge-5zL{gdtg95`XjQI=?^&`7hR
z-8ebIo^(KS{qi6N_xQ|6ZrF^C(G4Aa)$wT<V+gbLo!U+f-DAvbUkVwu^+3xpYCW{0
z{#lbXJ+PKx5l!4ba?R{4v5GH?{1H8gE4r%t9QipaM`n9)MDddT=J16teIl@=glH>+
zeQ=1@gbwITvOc10u1l@6j^;RCo<79!p{|VLY#Gyu$AwdY-}i|RR^Cku(~$5`mjLRO
z5TTgkT_!HHyvWT7QE;)TEg@Vj&Ub9h4H5C1`zD92Km*Mcmx|b&J0n*g#aI#yV@uB&
z9{U554Kh$B`>a0c-a9vcku733%$KuBQrZ*Q;h~;<4mV@+$mz0p_KaQl=*>rahBWX;
zw3}m(0zb_;4`6<Gh3}kugEnP5lBkWCjYuH~bZ|3K%q6tbo>GD?gDd;76e*JKcQkRo
zg$y=7S<D|lZpbFqe&>6<Fa?`D_1|uQGv^bAieBxH*y1{j#%UMETftK1rw-_<jdf>X
zaLURuc|pcd?~C1BLd0TC+0=lF8GmA?3MzfJ>J%?_vQC>}!YOg-<W|I%blVd*uMq4X
zA)dk#aA|KBLX7u{v6(I&<_jho;(nR?=S%Y@DJ|PBd*Zazq){wZ)m`#(V`_j8vRu=$
zgGO=|Q9Qe=8+q41BxOLNSjb+Md&9-EnpRxo($bKMDwsS$NaI)166*y&`(fV3?lp`0
z6`X!YzMs9HZ<RS<#st1$9mimO{}Im2MDSHLb{($=JXuzuaaYyArKp!pV|P!kK@&|+
zUVQ7Oxv)BKdEU~${adc*W4p8BlHmz_ox~2;=Bc&<2ySb`4-l-%lmAbG%kod#3Qnef
zU-E;o?y%qd=-nL<tNh!*9(*O8ha-K{iMw6av?ASwEOT@Q$%MIBLcFNMxvLAD13+}7
z(U_gyn+O0e>m8M)ebW5tGt*YhcB?Cuh*~1QZAx`r=0rwPwp4l5GVwDB-R1WF%+_|B
zSyLuYT`;^B&CV_F`pe^U|7`PkFo8PbcfC55?7iGFH1Yl7G5mMmJ-(0EGiEH`$IY`Z
zN2RT~KJW}MFZ4;1(@82^w*ifogRoV8{Me&V+DCq6=3Pk{f|*{+dF??E*spW}<-U9%
zzBYiMl$!v;9B&mih%wAuzaKhaBoXW8C4Hu4=~Me(+I0iCreYCvGTj#;)Xn7nuCHms
zGRn!CUBBHXkzQ#r)S21ZpAjs)Ma@+ZCNDD5CUTqbi_-S3niIz0S)MKQ40D8X8*;!I
zfmnqN)Wx{GfWOl6tGnVY^~;AjG)LhV$H`^#TS-yv71E?S8H#$8iK2Qz@uX^d;gRS<
z2h19hLN3#62ZyCDhy*6#VA)ab4wCzK>$*zM;~Ny<ReKlSS*T@OS(p6SI@X9E0%j!5
zxr~EsP~3<O%4%jJkI^)_ORKS*V4_>rfC<vml!3|_CH`8d<+j>q)DdR>2SqSLhQz3R
z1biMIp{(xnRmFY%s<Su8Z|a_~aU}j0ofg4X52fIS{Qm3wnj1CcC~@rcyLQ=5J%&uR
z^L+P##7s+_kjSC*BSt$wJ2Ln;x%A}(2AjCoOq*862w;~r%X4||@v3vE=<2)r&7E%t
zJ$P6BD-RvEm}YM7#!rN*ygTQXc6eiUw>=O?yI!L#o1Q;wbDWa!WhxNKzI40SJ~48G
zVZo)hUU04^&^M}ud?9~32dk9wSrv>N^U)ZXEPkfG%)HHPc2*^@zJ(&!4lx`HyJv$o
zZWdwhZJrpIO&%g~QXBjp<xRZhThDZje9YO4pexdrg(8YMYJ}8;`;Lpkv#I%sAd0ZQ
zirz~v)4SCX@)G{tV<74%MhH4)(PO3{RQJn{F$#R@4Zr$w13SRIsg4GWW3f!AZ#Y+o
zKLku2KAR3QSOB$U_-9>Lnz6WtPQ&O2Nf8tZGHvVm>UwE$^hoj;zX)!BDR9`*B~ADI
zsjdw32O`Y-U>@1Rd&4W_@PHp!05AlAQaU)F6KAgU67#p!yCzt_L>F_5WQz9<R_7rn
zUO-cnk4Ng1w0arnF(W+!eNZ?8(F^_sk8{O^0^v#?4pJYi6i%%>e`A`U-jA5v@JfbK
z%14s(mF0)#L0PCsBG}dPvdXK(ANz|coRLX%w{oPhfoMeOfUqukwB(fg0HyxwX=ca5
zSFh^4z88NgKro^z(p^vW*?DyfFAGy)g6#)zhbU`la)b4=|A9L6uI_p(I5skjX3@D;
z4$&Q528Cx>8@@C};$4`jwueI;vm=ySaACn|NwC8`8|FtJO8)7<*rlJQ;MrhF!62Pq
z&(||}eVVEE2Fa4%5kiKel4S5h&%NJ)^fuLZ9P65<j~E6Tyeltr$xEnVwTyXZ(5Uso
ze~zE0-tg(HSaxj{A;*M!f2ur&!T@KHTfM0?lg=-CPiC2)&Gap=v3oq9W_F;8gZ-O1
z^9$?`uyE=KPP9Wj=TyqB%N(@qZ{@XXs4#5aIVcNbxEcnLioHeV;PI-B63rJl%}9G~
z46{IqcTvaQiuaM!82y@s1xI3kDhTGMrZp)oGF87T<rC>xt_GTuiC@lOe{{`mI){Aq
zyw{lBle6A1{H^CM(HBgJAz*Uj5O&g%!zvVE`nb28S%lrVQ&yAAFnH6`hKQYN*6WA6
zdYtR|l%#?sVcHVUEcl`FjM$-mBa{rHj$LtGp`iq{5a%nD2BqLBU7mjQv7q4L9i^7_
zeE<2?(shmn$CfKh4Vu5V`ttF@6boDu!6744urNeHTcThooD}1GX$(fwW2_V8=%x1A
zjc?yVtAnibjrW%%wrN-oAJRuH|D=t7N|5#5!j5HJ#4p}cq8Uic{$(h{T?P+qD!evo
zw-GMbOnHI5K_Y*@i7LQLn<l^Bmj?)|04CBeIH(%fJmW^2DRnwLNU9}EKqo&`AJbLT
zGzR*atY@etOaTM>p~}}|txxJN;Yk>Vc3KP$&TEk7`0#TcC%@mFiAmp{7Xk_p1zL?~
z!tUE5u~%5lbw%@9s_Z9sb0W*s!+p$z+gn-zOL#UxGntLP9D|gn;;pF;^e;%+TW>Cp
zrCAKnSuvUD+hb@iCfdBhc%=cpf*bE%`_ze)h6($U4$>T^XBrJVP$DNGG~^RX4PTTu
zIrT?`+RpL#TiL^OiJ+`vf(jBGT7#ub6)b83;B(BX;t>Of&@J1u+}xu;Y^f5))Bvqf
z#!5O$=plhHT*l;(2_9PsNJc!evgeiBUXXRX_Nh;KVJGe2Kh4{4hw7gIJN4;0I!!{P
zPJB9P>-nt7GY+yf+S$c`N`!?DF+)n?sq}VEe76eaN&|IjH9p)6@FvJ$=vP~?Q||4-
zk{EwyG|(>F)OivyF53n;Seb+EHA)Oy5fjHb-fp~<gpIB;3JLHA?zOKxTRiUdqNj0j
z{_#!!hn*DDj~)Hr`p%2hr4!d%5&w&|QCfy;a7$LF!a0&zx1V$=yE=A+@TZvoGQ%L9
zh@6CVe^swL8eMXx(KK6|1Oeo2Q_mIItOavQx7)4F`*xo?K&2!uLD_WOXvmnXRU|Fa
z%5&taGa<~*sR<3KW~<BHx8c>UsRXEo%~o%_<D>Q#zUJ%o>-aKLZ_5|vdh2_Up6~1K
zNosbnh8>?z_tWhC8DGzy?U00_j|N;2hq(A69i1Z#%)uXG|CDtf%6_zgby6X}sQY1R
zo|I==f7u#;H(|nkD2}9!|FIv)HA>mQeLu=rh#+TPCJrb`W%UWf)yY#dBDpT&Ci(Tc
z&1)vP%=3GLefOIVha-tx$<;*3^{b6jtha<Kt|t~Lz^zv?<a!936NwYD{D5FZB>9rL
zT%IZ(g@_dP%ZNnJy0EdOgsrN2Y<sfTHt8|YqP_GYwf;O1Z@A<UT8Cd_DiDfIrd9y-
zgBy)}v*j@L_w{r2?e`K7v}qiSfucxvx7JLUd{Af)$0@m-jTz8ueHX9V-$yywml~O6
z(dNHT75i=FALWaWW(O>_ONNDL9RR-_ubNuj8LmnIuqI!h=&GajCN9WIT#%95!_C=G
zFOFQC+Ns}FTXyXYYajyyEeNOH?s}cJ)*y_y;#twdnCPiQRx%twUB{<tVvaqwg(1Ki
z78Vn?T^(MFJt4ni-nT$JMHN*VmQ^~MLe;OFap^SgU;A@wdulSzyhK&Gy9>_)`5)l-
z9goTuXXhfsPeGCLpi)sNvBy<G3mm`>!Ww`yvZJbd8+IvIf__;6(X`htzX&_ezl$y)
zrX0?GyOLjK%c(w>Tuxv9dS(>#PyfO>0T&}_HO4!{k(b)jf8Z5as!A6_Awk>GixRvS
z&Ap;Vp6kh|<pC+g8A`&amYd-BW`$y!#-V~%w!>pKYKY^4LUkLV#pS6q;vzw0(qDM!
zavGQDp|RVcNELSYY9u+zP>cveLTf}+Wr<%hWbF+@4Qf>$4oi@@q~Dd`8w8RF5Hghy
zu7ybAGmuJE2y7tI5+;oUq#Grg#u>wo<ndLn4@j08kyZt8cPQV3^k^c(Fq(#cF!Q@+
z(h3xby1A7gyhDW-KMknv5CGV~91KG-o=K*y6DMYPSuhM12|(kM|8@o!Xf9G~CEkUT
zt4R}8LZuisJj$_!1X?K7r&PhQr@>PMVPrG%_AXN|$B|@huWh4&=mqUV02;-GPJ~H}
zbDQ8Cx4xLr&7}=n2Tu)pGBG3YhTk1SRh~YY?GAH#6Bn3SEtUgVt^y@${~fO=j?_%D
zhHST>!0gx#uuz*NOn0N8T^<p(Z)0*Rcvw+h8Ev!gqJclr>5m_}Si5^u;@&s{^#T&s
z?%~*h0-dxLrAqluh!w$tRfLhCLI(yz0;lH7q)&k^p_t+DMP&$3K>(rQakfTeTZvRS
z*K2Nc4+ihN|6KuA+YWnf-UmuA5r9ILGpVl$eHsz0saD&Afd*J<Fxfe+Pp{AYV~_$+
znDH2N-FF8wnMl@eJ%z@!+B^*vaH|};@hKgu3rrt@_O7E4l7!95$2k8o{6d!~W}jq_
zQ;gV4U?;C%5sN4hxz*MMMTj;Ftv0g`I>*y7Gx{)GLw$|5>(ipOws(5@-aNu|;;9yi
zTpUk3c-q|F`oy8sdUg?ZWU2U!It)9?e}X^cL>*S~-ofKW$KhsLPdNLm9Kz4z_8;L1
zDapH%?xe1$hsV9zvOVWZ;1j;lkS{a(u4P6bBz?f<)XU#t>LXxGPMY8ZDgWhPN>t~L
z=LsowazJpef$zm_VeK2Ka|z@eKzXnuk3l;y=$GPWgAL6Etx!+4l;F1~DY_vRIrV1A
zb6Fw2ZaqY`3$ho(G!YjBo*gO{cvu0hqw1md^MT7pBM6u-laUzCoE}tFMAUR7>;8LJ
ze>>|cv)ZX@As~JUfPes+0v6>%)U%LD&&QMD9WSe8Cx$1N&1Q%xApHIU+@BLXA7S+V
z`cY_&S;7P`tzQPWcB(B<0Vi08`wP4q9VC?19<I()iEvZQYgSxJk~|Yk5yW*ce|fiz
zWID;U9Dw)W0`3k6oYb^0KA%iPMA0Zm3G9+87U3*rbDhL#79y{Eo89IaK=P{P0~|mN
zY>b0Yt;KZt%;9=FYf43LOYLxrjGb_y9|yXD5jtAnrLSjh8n(;2rKwLY)(Q_^DXsuS
z(+6$L(CUT!aHe>EF9@+Dl@Jy=ky8RMxX*s6e%^aMHXN&$n_#8W=G`Gz=6K1D46xE2
zj2XkSn>r9vH>e9TUt`M~8!Yv1YVgf)PckOxMGq<VhUH!Yj&bBp@<<^kb`p?YHP*!U
z*j1R2rb|lZeQvBa)cr%5`{+u$Aej45`~$M=m^rbxG^av&X3}x2*Wjc*k(*8g=v?Cj
z2@uxgBX9OJF?~2Hw(4i!fN?Nv_`(u-ZqTDMM{On%mG{cpSE67MTfvg{^<4_HyP4Y?
zy=6at)A6K`v=qTuG}l_3)&nt&<ZNi7NweAJXZf(1|2y7c&PpgasE*z3DbR5UzznfL
z2%>*Y&F0BD3NDOopfs$LmBz3tJt56t{>5<W#ZYA3AOYEhG0k2%wZBM{ggwp{MVJ|0
zgXJ|3=>1k$`>qM&mtb8h0qQ6;#SxEzF)SNvocC6coXL_F@^6@$u4vvsF_^pPJ`R49
zChpLX)!j)NL1_Ql2>bzIxpkhQfyQ2!t4%dloxv(P<?w#a1JUh?gvR#yQS!Rsl24q>
z^{?49_qsFuOiBPJdG+>V>e1Xujph~TEi1hbz%EI}$bXtbSpR92%g*|*8xekpU$z_n
zJN<T|idl=<MT(nEWTDh3yLXW<pagP9*GVlx%}G7_>+PKT0H$fq!<yXMFGu2*aoQFS
zMQ#aF&WFpri^Kg%v7$&nIGIgDn|zE+Olzn~60PE3TG`I$ZFw)%j2@JpY*?f>eaw*n
z>DTAW{b1%`<oLX!g`4H-X<Hr7=j9&D{OR7^B7W=F+hnA)w#Vb!vun;7Bsr&FhrA@*
zLDk*CH1VfndR<5D$l((0yG!Vv1R?oxvS~VP0ZX-J`%!Zx__Fs&mzCrfW?&Yxmho+$
zr_D?Ah3n$+yyLjL2Ou<1&194#U66r>sXkFUP{8h!;l8|J0{z7LZyfUa33e$Qfvzgx
zDd`Uq$`}cnU{fw7g@Z`qJ^wN$M0jP`+tZY44VB4^mFr)kIn_An3Vyl3@VF0&NDoZ<
zF7eD7BO_8Kr`8&ahbpoeKLoO(GUJc5)*ZgGuowX@rOg}{z1qY6FW@lK?F;^?hz;y8
zav}vV1Og;$8fxWunr2VAH44kLZIw?jGhUm_9^2KW{$|Liz9VHzZ9%#9w#Z9jIy_Ne
zr$e&LahKa9zJvyu!+U9n00oD*;*Gfp;QTOwm4@c<r3?*=I04C0*i!q?+TR;cYSmpk
zE8-7&@RRj5Ye%J~inZFK!@y2(qJZ)!LlcA#x%m=tSD~3}mkN#{wSw1|b<E&pvEJ5~
zGNH*@&Fh)Xj_|vUwRtU{>>QiiBB*grB<>GP9Domppy@fso~?a`0=QAn3{bzpF@iHq
z8v3djU1cRE9MQ94um@TR%yWr@FjA589S0i;Pzayy_o!=grB%&&yM18ZkZs)I;P_hm
z;<gQy+-D;px%xfSgJ-?5sN`o@d2`O>MA!Jpa*{w3nKGI0OZa3oeP~-#Yf-Jkn*-iT
zfN*I@uyzG^8FW7*XjF|SG!si@7<xe?99iA0qwtthdkqWc|8_cZYu~owzRhJ8Roc6*
zy71amgM~_aOB+piZI%ou*Ak&!UJ*}KYx^E+JEqCzWz^D4%hxliJ-A-7)Hn%mx?Y%L
zaVU!CNC8RV4o9Z-A=gJ9U=4a)r5w(@zH>>p5J)>K`SwPXd57%`4fMAHV8)S5A@8oE
z5}L{*0$Exxnmy(W`<w}$7Ez=zdicl<AgtpDPK6zVACMo3>{26rr6HwDP2ZqY13gi)
z;y{||3gDBKXb&0cqoM&*15fQRmkdS_O<nJhZrEmK&?!@gbwX7V5tr?PxdJ`xv=2;I
zYa$mPmRK#p0ci`OdM+Xi)>T2~dP5lEQ;OE92?;f(Mb(~=Za|wDgZyb#8#VDjo@%Nv
z2yaDK{xp88rq9IxJK8C{2x|fKPHVT~Y}R`6cv=wGmIINFP3Aqf8j?bJqhPTg*MKE_
zH#YmLeLcnb(RE`up=Pdnb1s~lfG7xN>Iu8wA&{Bo>VTKvbn_;iOBD4ozu?p$1UGuf
zsUQGy9Sq0K1>=$F-Xcp%)hb-=awWLnwxZg)w437hi0nSw;}A;3Vtv%m0X%4S(hP!T
zCw-%MKaE)d!)=uQvt0}=q4@|6OYeJ%{2`i12p!hpP3~lE=bgDy8>H#&Z&J8KpK0Hr
zy$iXwzx&IoQ|nwovEL7TkNc>%XYpl{uDXuekuo1!pD#&RcR3|A^2F84Q^$^^C9w!E
zPmk?`&IkMF90H{|_@Pz^HUfF4MAE*GEX;Y~=p{>j*yho2&z5S{VPAYOmm;cznB;yn
z<-h66TZWK2WO+#q)!pjhF2X>^0*yDX@=B~BSrA9F;%G|CqdCe6qFl54)AkoGdg^?b
z5yE3-J=yt^hU{=0Lrg;gdx(tEYm<75l~r_xAqbdTQ3GKw7n^S}^9$blM&tYmT{fx;
z%_9)CSpi{G<V4xU7(fY7j_Muf0UiifE5WaiYVi@8olqvbyi3!Jo&ytvfw5t?ZP#L`
zOs$ad+$$ZnBWhO@z3-fIzcB&K#@$;!Q+!}19QWht&UtiIXHdM?3vcJIguBZ!;B8}P
z7dcsezqYvSHlzS51=a?fu1w}-Hu*~Vvo0@4oQ|F&9;y4maw@&Q43b8g&OYfjEj%~{
z-R{8*x67jDT|wFq5`%q%l$+p>{uTCx)?tV*>z~q(iuoi?{ge4JgRRy0q6pHWqP6U_
z$fq{%cH4=?7Vmq^4}l1ix(yG*N1F!BU^ivV1`O;p(myGCQRi>j7OTX}*qTqk{yvpq
z2gnSA?g)=QB+r9202+mep0g7ME!@iCC!FKYbe4zc3xuXJ(U=<|S;HcWDSm?_z0cym
zBmM9VJX3h>^YJp~%5QLvAq*PJsKgUq=}Gx8xOae0Sdd8fPY)&=!#~_UIsZK(bMb#Z
z+fPQy?(`x32zrI#?o2xlllH4-WgJFJni-H-4b$P%V`s;@`8Lh{fJiy_WnU8_gV^70
zou9#nB9BEoUJfVQhuQf%kL|lgG;$)GZd-LmobnTr2E#RZY|y?V6o-&q94qEH-d=X+
zHyxZEsg-eP`M(^VZJLUv-hI8^UT^zra`84c{&Xwu?=Hu(w}pMVw|KO^KJRb)c;t@U
zl!(-+7ZZeah7IGOcZO$MMW!9L`I&a?s8Ur6CVAbe_wT7NM(ND&vod*DhT=AhG}^ZR
zgh<`mi@y!jW*$Bxz}N==Ry|wm?d{oBg{4Hj#WsqIbr05^y!o3(mX}7F-Y`>r$wIip
z^Fumuz@~nls?H3T=L4}C4zhe`u?lSHtT*adQ*8_suEFTMf;Wwa&^hlpn$IA(M%+k?
ziWp}^>s>ZK){ek~LET9DtCac)fXdrU3EU@cNYbSv1svc#zZV29hQ|o#&hx}3pW4)u
z0G*F=fS+qnlejnHW;P+ip}o)UjjMNU+>9ubz`(%L3I;j0&G^MD6N$AV3jnAZdys*S
zeHXvuPW~^(&M~+Wwp-J&JGO1xNyoNrb)0l;+qP}nwr$(!c=FYGr{+wZu9^4Ou3fc%
z?ON+uzV5p>9;c*|f%Wut<|rb*+BDLNpk#Uw9zhlsQ^F<3-g$^V@kF6uL<G{rFqj`o
zmw-Qfq#t_N(UlN8L&BG=uD{cJKg^u0%sQQ3UoUXfeNIYNW;?U?Y36$+nBY9K4PFg6
zEI680k{Fgnsbk1diQq>=mWK&-M1d6z#-A)UhNOX78v@9o8Hyt`Hb$tODRqpp%0CI#
zlPA|$K;!SyWc&O<a1Qi-p+VF74KnJhr@f(r7jNitDcIhW=?1{w+xcW^c?ItyOa3ek
z=(oJe5Jkv<gsL;WgMBJS_iy@9GEoeHzLMsP$vUv}#XWG-L-RqmJ$KY!rFFgG`uGiQ
z%Hr&w!aH*KdD6jdSlVT)v&tR+D4UA!L;;#{Z|s<p%iD{nNxPySTB==+aZZL+FjI7a
zI?r?2i(|U+b*J1Bay>Bq%XZfrleQcW3~i*JGjm^<a*{DK(8V|Ae)NZfi|Nzga?^&K
zmkdxfF)bh2*3+CmxCl89S@J1Iszux&tO;r`4!2ET6b~{imw!Y+TZU2GKqmJmloPD6
zvi!MulRnR%h7ZmJRubgMEwm**VC+vrU#5t+GIoxQLbtPw4c=hn<1i9)x4wl%P9)%~
zPjI0?5a9z1MfE3dQMn^fIkQ0_N?@-&#o{U19>1nQzW4Z>8T>zj{T6^<#~MBGwehyg
z12T8)F}S2vwMZRJd~ncWHPp`owg#lG=pAB`F7H1y;-<|<%QJ)ka_ZDyd+VG6y*ng$
z?lQ!yKE#u(b{c6>bB!yi@VV%XO9@c%m0SeIaX@<kIP>wDg<?WN{Z@Lcr-8Y1@L9pt
zpae?$<YQLd3h+%Sj)xs0!!s(6IIwyEs=+wMSMBqL?$!y2Nq5rE^rEUd$+Vk<j}QH9
zX_ibbw2K{M5ARDr=2<_~+E3jI9%%B*PRRCPTsV5;9g@Vsr`t-_uR^9khnnkAWfV;H
z(J9k;eDQe#F=3yXRbsodpt~oTugDQ;<v&Xngoz^5{CmE2<&uiaPjE<!)Y%CdUi^R=
zk#fb3lJlVLyNeN#C~09XZ+E5E+dYsiZ!NMCDd)3$kY_rOXNVD`ImVTaW)D<MQrZfS
zHjM2MWG{B<RAu)g&MD@%`v<jz%70xj8keR_8O+C%yNua^j&op~UaNUOFUDHXbN`0<
zGzLhrsMtIoPJ<naxDH-L8Iq-4o}HhStTx4beU6bxm557)kb~GeXLd!@DCnuij}2J=
z6KIhlM{ql(^62lM0-K+GK%2>1po)6-TV1!uSnBZVgDcUllw=Xi;|zO!wlhD%6Z9uN
z`}x$C+@Gm4{uG#iRly&%MZ@~T_U2TWo=v055E;;?nQQ`>AyBjU4KTtVw8uS23Vxd=
z>tEpx?~$CQP<5M@SluzuVj?ipL93M1=rkm-IF$rc+_LnJFZj-(Rhtp#y<puz2OLi}
z=8OG14CbttUbcrbmx~CVx#ZnzP)wc+5HKV;K*g<OMJ6e;m?MLQxAfo^=VkA|VTrrr
zy@0i3=m;QhF2){x0BXfIP4dy$f?L}B1Z}m-l=zEs>)LEDOAT}T56xL&)GplZYIs8Z
zaV-3&Np39P`ht7yCBLL=gB0|0U~brl=}4hUM(fTrBY@5MHXg1r-WE|t9VU(T`tGP-
z#?V1Zv({k&<nvJ((7Ow7i*QO91jv(hq0=^#>1cScFLbc{7<l<anmbuVG2JLPION`X
zR*K4^j-{lE@k@w$4CmpB6|SlJU5|9pI;(#SMP5&YuNzU7W&Z}Z6#dFLmAX%z>_UJ2
zx|dDcxO>3BdU0LuJI#(>mRR|f9ijx(I7?E)MIL2_N0axdwhUSZVl9PF;R?TH4G)n=
z5$uO5Q!*<w>n=%P3%0HNlub&}2^d4UzfpesD=6E8;m@fRkEiiM?=WaDJDUF9-2)%!
zn@0YctojO^v|S)c&j(%;?`C1Hyhr1XT3Fp`CfxRpy;r!L6tE~%Gdzc+I3-IxDXeZI
zsmvcl`yUX_(KesJXs~}0{u=E6#)e^G|JT)Vqng!!bO}GRbaSH_;p;sR#Bck-Ts&7f
zE39JL27R0$kkg|A*Az@g#oKke^Tp048#ay5?+6k+I^3^X_c!I)+rn)h&Ijr4!%eFZ
zQ7H;QP-Z<)7!YoS@H1}Ax*~W7LFPww@3*r%3Dq74Li$1qDag$Wvc6c`)4Dsdq$q}K
z`Qs_j*u9w2M~Jg@yFcf<yqzEK*1Ft2t}fm4AAz{AiQpBhy(CB^5^F^6+Umlxs>dU%
zx;&a?B|FKRb;l{I@9S#5^|IBf8?`ORz&q1pn6sAJxv8$6(k)HBTbIMzHaO6FPhmk0
z`Tg;?2Kg!`08GOmj&*X@=$5uDRoo*Dc2L0*Ct3ny^P+a2b(BN~T%<!{mBk#JyrFi9
zKmuIBZ6qij$;IPDdb9*_#25!^LlA|;3<|wr*@<ET1{=r%GZ_1N6iGlFYURmy@x?hL
zDC+apVM0Dk4k>bZT$AuA5`tT8bK7s_=LxVRauUS5kC~2EPP3sR_Axr}0l|v`5DFlI
zL4_x{82GY*`#m&_rA3&}?O7xW&fc8L^}qUf*BaWF=L<h07-;U+MqN8hc+T^dTT+<y
z82SM0Bou^DYUNCkNcZ2Uw~d;yZ0P7vt;G>=Gg7ZEV4{m|Gr0yA3e|YUUM%aEbL$<%
zC<-V}dTvAshhhN=AXP%hZ}it=CEY(2Z1Z25N7Cj;_B6@ngFMW$@AGIcDII`Zom5d*
zkK_s?fms|IqU6rSew$Zzv^chf71|LK;x>h(j215tS~t)-vP9TuYOuN~uP4sVQ#-wE
zwpr)f$#GVAVtb=oy|EK04JP--kMHsqk^X@y?3$VK`+i1|%jbXhEW|z40Kkmyqk#eQ
zn+=le#jQMu$pP&hrErbk^Y3s-(AN(7Jcs=hg2z1s-bcL4O@@sq#~nW!t+D)cen=$~
z&GuAF&!;eK=J_>M>Rz(!3Wy_24D=wVZ>4i&y<O(GR!2AzIV=GgCkOMUw+*&&)jo`l
zsPv<0!;8<HSX`bmI6^Cvp-+(*1IKbSyxRTv@Zzq32k@t{9xGsdpS2g#6k!;mYg77;
zl1uC*$eCycFcc=5exhx`O^#4-11Uejqkv66PQ$r=vOY?vRL6iad^o@542{4fAR~u-
zX6|R~<GeuPSH3?)4>X0WE1+6{<>vQE=Mi<IRtKcHFVuG0>0&G%5M<~8BJZBLP~*Tw
zw}DcOJx%29iw~(>j8FB?U+2;4#Ye!pgw~c;a2R5~mX86uKDNY%wYxc-7ctu55I*?m
z{k~SG8KBf-Md<9l9XCIkHBJm5molD;=4c*i*@@sY1SAS%QxJz-c58Y{Ska-p9F5Aa
zlG*SCI4KeU3UP5Derma;u-DwifPFY)TGo^0>mQj<h|Z$Nn+N2Kz>2YplA;{15J#Kd
z?|%&*v6rXM9#S|8to=@IH*^%Z3=(-N+2EEf4wvlwYMaG@Oils~-{e`*<<tqPKq3$6
zkc7<PIVt52;7N02Lb<o@AO<QbPgnu-s679uGCLu0sv=p!XxB;*&O8)$|J)q(!zSfD
z6%E4<PZ0b`Vmm$xKXw7*NJ)Zd*>F1d^QlE7X06W!79rkNGC<{)T!=bxW>6&JXcAy<
zBoy~FLy)yK*Go;=yt=H9dZQhUIee+HfBq-lE`DIFbe|C;fc?ZzxD2Qu2s|r+r3a5p
zuKse*+8aD3vy1Js?q`mkdEv`<a@jjLWD;bGD{BkV=5z^bRB3~DL*}y<aK3c6#V&4p
zd^aQ-twFuKL&5!KcV}O>kzdb|12{k2DQvb<#&^4>BKB=Tv9)(0{Hv}TXvW4$HDBE;
zI>nx@Pk%WF+bS*yw!RstHn(}rw4smE-6taGuF`=-8HSfq@6-y2(%ync?r1?8Sw0*>
zv-aIovKM!1h+#mq9YL^!<@XkyAVq+|XFMNH0X|YF1|q@(Kmn9&3<Feh9apOS5zG*6
zZ!gi-WmCa?&?~4^&BS?2*ZD&S?r9rY@D-HTZLC@1osae0=7AolSFk2T&91jcCX=)?
zB&|XtK#3G|Q}ctpvgsWAyi<-(F$31!1}$!hp>tL<UO*7&YE$c)k%5Z=4cFt&m81I)
zTI`(`_20tKw-ERL*tcb4{AW`9FJJ?ks?s*;{P3M`s<s0PD;F1N?ZZuF<xeqUkQ}Y@
zLKo6u7J{Gy=5Oa*HG1g-Vo{NEti>NU7wM_bPD9X-S3GB)n4Qrg#Yj0rOatUa9Vb)!
zWAOqAlm%#$ou$<fVvMi_d7p<bgDJsY%&;j1p9g0bf)6k67xo@Fo*bDFT|Wqka}{!<
zF@edeBSaeP!g&#N@>>xTUFBBqXIucGLs`c%-BdX=2S*9JwAGI5#pcwRi~~4_0XBZ#
zLWLntwV?TsPWbeX)(@vAW+tj%!P4t#f)@}<rL)cvj*q9eq+I%;X2?}(j#n$q(H!v0
zpH<r;In5xT`|AgyqY_~U0JjLK(k9+%#>Z=~6iScxRn~E_Rc*u+SPz_7a>~xV4N>Cx
zKOHxTUqRM#ov^2=9NE579Qp;6NuF*z*_MU+z!tmTmT~}Y-i=1@*F#;>3#{sP6~D!|
zJqv?|t<CycpY2Y-dMNL6M|(?GYhQ_LT<xg>%1b!^k*0r2rFnhz9iqhuqU&r=NEw5E
z%&}<HlFK7OtPu)<*5o}vafW$s!6~2FY%swXCZ%p}X*|qtjAUpqJ+|IEFo02ivuejL
zXYjd^EeouwZ@f&kyd3<rwE7iO@WZ^4(j?g2GEP5D43IW8awYd>RIYL0xuS4c!DGAc
ztZYg^Fu{eWj0pvC3Jwtm4AC=%@%mRWSV#=`BFDA?BrwW*>45_KXrY#5l;X0?VA4TL
zYb>_6?~xP@M^$<kwdSbZxbKLDy^ds5`mfR3U=91@!laHSE2oXR)9qfSK?+`|R7qn|
zhcCEuPw!J*g&WN~e;7!^cj*y00eJsy7|~O>{<Cc&jx_SEvE<>?XKh=U5RiGzZL)Mp
zW!bQ<W~9$VfBMbh`PKeAREpD%IFr>o(WIxYEYv3QwPlr!jslhWz1PO<{C<Vk3hPf2
zjb*%|%Rda5AArA<roaB`A%79KVE*6w_5Z>{BIBi_=jq}3uRl<DNrPQ@KA&oHXsHn|
zDAR%G{NzsBNEDW1=NdlV*l~&r3UjjdB8I1#?h!?n0%GCnofj{zbfET1=C$H2gab$=
zEXE257pYuFeMaBiA5X7vQ_>53zj@AB1j7~2c*nAfc@xl8YDMNc<db#N_ZuS>lz3eR
z@3-di1}D3>-;p`9R!MYV7}{#!7spPxuML1$?~g{AeVTi%6iZ#U^w;AdB8T+=V-5Q;
z+7ws_fZr0gz8KI)gg~GuenfJLA%ik-vB^N_hU%x!UF)QG7HCPH^FdZ=AQ05b*tNB4
z(yb058*j$jPAMb4vWS5NBJFkt+HE^f^5|uY6t`H4-vcKcs`X)jomwWlz+EMKtnytY
zpMn0uYiI2qS_=;l@18*joTYKrhu=tPHAByzsur6z@9A(KsM~9cgHD@=di9UiBfVnr
z)?C(7!7I2#)m>O!0BNth$yqSIk^2Gb2@ZCPqaX=N&X>Ex-J2z5Gg}u!GMiF8UyuDI
z4wSe?AVJwu-*hK+h0aWKpaqb8(kOpRIP{Nw7}|_p^1y5k|AUK$AKu>|5xSiIrA)s!
zF(Pm%$|4E+IL7!5RhwJ6KHT;l<R`s=os>y6|7s|I-Mabr?YmBl(Wrgar!C#wd}s<L
zEfUIMkW)P{bSWQprr;}lxmGn=f;y#Hg{#*$?z`B!lDfjJI3$WN53FsOG;GAQ_lKaV
zp=)QGF3<G(0OJ5AlK4$Q!a1R{j1q21r{~djX%TC$^$&OM*p78Z%9o?yqd!WiaBP}Y
zCr5n3+eu!Z+hp^oo~sMmj<B=zaa4mem}!=QgsF3Y(Ct3KbLi8G>)2S(>((DnmB`{U
zcL~L)_(Tauwh5%vfFl$+@q|Q4r9}8EJW1bjjH@PJYTS=Sjd+;&-$;4$ta0~!@8@VR
zvgabYv9f;8Nnl!Dm`B{y`}FWF8(Yh_{7I%ha3i9Z8@mRBwe+i{I2*_#Q#U5q2Yca`
z!y&|X%O}tIzAc*RI%)asb;xyqPJpifJTJJ`Rj5fH;dzXAJ8I1h7em<=k^^Yc$rryH
ztk54d((=2>WFOB&I#om*GtiuU?4uZ>8gI#DWV|he#0xSRMnAkZ(w`IV*8X@KM-t6}
z#OGpRYoug_p=EG0)5tt2qLc*hXVy~7CXy)B)k}!Q{YtV-CB<IK%=+NAi19zF`W=;L
z2G>8k;OQ(pKA;XIph%daTI<aNg0N%8o8y#Pg{7Rt5Nqa>1O+}RXrP2t^nE9z{&QMQ
zqD1aC*OTAtm_!rcuAlr**KIjFH=33DhB7H?Q0F9xg=-ODZ3M!#O39%H{HRH=wJ6Jt
zQgUZ^GzXR-ni{!?Fto}*-DqF|KbA2Bt};B`yhe_|r+&k`B!@=**J-4>)r|6p^s~`Z
z;i6WshkX~_Cw)tytq2nG5KV5iT)90mw$PxG)IADEmMe&E@BJ!)AH+6iI3Uzm+j+Q~
zjWyb=e*-eRL?%JZ)pV1$EQ(bxaL))Wyg=s9-P;Uc942@fYs$<~Y1h_XI6Afzl<=n<
zjM<FA9vW>bX1lTwR96q{GSFtRsm&R8N(=1qHa75Mu-)qk>)3q#uMqEWYv~dN_*h7x
zDVrt=p;{Z;9o3WPU`1h&Wd;DHA`j)%yNCm=0aM>{<A9|=G26&liUuD3Q%krZvKs<3
zfHj?iz0O0S0?HuMM~hULDe_s3q2Ugj6U^yumMM~AVy2MdI~I%bxq3~tnJe4Koft(1
zBpA0q4+lo&z&Z5(w8$PQ{Wt_r3E_kzy#~ELN>TdaqL2Q!n)c+F<reAzk)l_6=t!i+
z_mUVOr_q2%RNAe|b3Ar}-1{27?_C%NsTLoQ6$2TSs|l3;1H_Fb2(*Re!3(2t0-yq|
zK|5Ep6)T$ulM(?|h9dwZ$iNkxxDee}c<Q`_d@3ukFIpMORFFQV=rp3U=ZtdoGESt@
z$L{<9S}c_MSHvc4K`5jGmm)935V{c-Q{wT3G%XByQ+sPeux2R%9k93Nm~2RZPY!BY
zE%M23#(08zw@BEyXC>`prbMBlje82da9OEry4)mU$ziMNxY+1JC1CK=h}rFG!pWYu
z`U}i%Y;J|5GP=EHE1Bil@&wz;FWf9Ngky?<i*pYuG0>8C!meqQ2|kx;)mX1c_ozCs
zlzHl1zjg${#eRhfDF;H|f|CJj4tx`g-O{M&iG#t6j=Ru~6^*tSs6Izb!H?f0#-kUD
zL!y?Gt#OG5F*+WIm?Oo^A8Rn{gyj46?XTMs_8<#PoJeyGwi;t1sdZ^D6lgT;3+4{`
zG)%nGE`E&m5=vL)9aN0^0mk{=t;8TLzOz!6sLRd5n~9z;mX<)y?pJIBsKn&C=1nI2
z*G*7}c~nku{WOtaotgxf;~PUM1lkH>H^{5RK(t#N!SBx*!<SnhO-Y3wIL^=KoP_Sq
zZ$pom7U@y-(H2|$SrDO7dv<kClCCA;US@)Erj+&FTkshV%Q+AL+xp`I+x4$}tT>KG
zX@H!WtiXD+ByaRr;bn}^Do?H_KhCI$KaE$iU$<7X1y>EA;3W62z#SO1qa7}EcfnNG
zd)LVST$T^%on;UszkN$gTai_qz0Nv`<h;jV48CqxrNRaN5NK|qj7^W<U#QYFp>~y|
zg}UGi6krrMH+*hLXu+H+UQmQ2O&PA?{yqapDw=WJ;6+>1Lz$*^I^4~pPN>Z9(6n=d
zNAw~em)I_Q1gaH#$LTlN*|?-WORg<%M`ptZv+Y6E&n&ddrbe{(w8HJzvYx*|eoqaN
z)#`x_u=4axRCFe!1Wa<6h74@A_ezBv5hUgcPRD-tY$vBBg-E94ZN;xz3~&)hsKlUX
zmr7(~;xXUr6FtTemV;NjYkXl|@s;p(A$hi=ErCs3nFy>OHUa6?>h2zTEMjF(C8{$G
z^ZGiGQ2WE4_(3!Ug&m{QURg1o|I599_yjG*+fvUB(YSJTJ8Iz=>CQ@c+{3@`875%f
zDMMd0$|h`I+Z8>nx09@=&1N2Y8xRR=J`c&__FhSa&2ry&!?ngr{*|@jrV^T=7jlk{
zN{>Jh%hb!@{=UNuEh_5royJ$59MWMMmjlN5nreIu`S~~~<XS30xiOQ6OwGx&KgZoN
zXpC<8h~q&TH@vub#_$OZ@i**MP~BN`ZazKloyQebKYC$XVR6L0=3L#lm{s4DVD)ow
zX_Bnrk`xM;mVbDK?b5;`^i&yQek||$RKHwccPxhfwh0yJ^QN&HR{r?8D)st>pMb)z
z>kIiM0qLBz=$#)I9WC9?`U^v#A)#dc<6_O31dY~IusA0|#(ms1Fh7k3qZJ@N2?=e8
z$6kq^X;HOGlB^vPgn&l3^=O%XYc(f#tO^9%dBE>-h%;r0i7Sx~^LF{Puk+;X!hml(
zM78Ob&kFc)%H;0(<H2_e#QG4*`e4XRuhGhG{KV3{Oe2TI<>281X2c|0<-6K?2UQC4
z&iq@_!v425Bm>*ORz%wTZ|Cdz49PYM!;Z0*ijL{{kGVb82;{7W(1|31aY)78T|{wl
zG{R9h%s<z%CYE8kQ(sJi7-5e9VY@Qk**RVud)47{#S}24u7F8_drA7VK;qSJXo%IK
zK2zS`PyKexj?wioDP;IVc^+Si>(57DCe5TkVNoz60|$1G%d3~WGrq1+yR#yZ;mFvM
zy}W*VdhI9CR(pL^cI|<<%kc<-=tG^e<D+8FhTM>+Q17mB0gn+w7Tcwzsq+-=8Y<9i
zSY%-|GfyU$FIaAz`KPM5?y~M#E^%o@guqH?w5>i-L+;O=vzfbTzzMrFS8ZrC-te2e
zICe6xwlo#PVM7<}E(2BXQI0aj%u+I&GcpxoaS3Y#LE_6)h}US4-cwgC!D^oN|5VHA
z6^&EJdvO;lK-h0{SMOpJhH-ko!GXT;v^$pq79<|pl35{@dC~cz3UOPgFLc_2j@dzv
zMn_D5%{k3E5m*lvp&&;86%x!jvC2qJ$i--wM~>*%$4`VQPxd-!R3ugu74>{>sS#Y@
zI6vI+XWnMz4|InhA=hXbJK&4`ioVdL8XZFzx$}|@-C>yn#qwm~V^AkDn4SE9tk?Q>
zs*ywUF<>TZ`w#9FjUnaO2j|xWWHEU}+p1V%*o@+K1A-97kdVM!=fwVjIePC3vRPI^
zy?i%~R7Cr1rjavYg|<cF^rBy~wwv%;*DIfh&gel#syH#X7RGg}m5&pEO(NJ7)SPq0
z!6jqF^aD)pNz}o6l8}hR7U=9HvH4=ngLe9RtGFx=E;OrduD=w37Yi}JC)D&>I7-Vp
zYAK64cz)gNv%`W^6Tn7(7&8ONaeF9Ll#?B-H#E#7-l;?eruD9}p!wJgOId*%l5mtl
zog!eyA@InHjm1@h5f;Qg+J<BNjn#rsvsG$0R#Nmv+r}!Xt+l7DULo_1R^T5J2MU{!
zm6M(Jta@lcoND#l_RnOI-4wa);(+4)Pw~~|b9L3m6ZgFPQ12$P(s;juTnYeCg(GZ>
zjp}Zilw6J97+?U0GH)N*yPdmfsJSE==@!JAL7|4Wq**!`bzXt(R)4SJBP9q4G|AK6
z4d2NL*hrJyB-E9)G$sm!lcv1r2yL9!BPi|y(^GdzN+V5Jo(1oqaHES2n90t*fg!<s
z>4G_?72BZ5y=Ft1De~w*nLJP81EO=wRSK*VYpWSA1fc|3sp(txmvQO`Dq}5al*E3c
zwj<bXLi8kp&{~)ES?wa5X_sp<Ho8{^ApMV3(Dsjf)Ds9{CcBB^T?%gNgOS2xP%GRu
zm=|UMeaS5ZI8#~#)ShYoL^1FjSc3}9YBM~&uM6M<U5`(d9)X%oTW7N=*$K6Jhx=Uz
z$NK|iFT%H4oL%HY7I4p2D(%I`g!E<*+yD}UzV;BdRPOTZbA7xoep6lt2*COSZA#W&
zmLKdEtw$2kyIB8#qh>F2-JQ_hElq8BwLb8i<N0AQ$3BcQJ?AuUM>IX~bb3mKUc04z
zILKaiNql8CY&N||PkQ&%>X?WYe!ay%tm07|$H#XAw%ozU{Oz6lzv$3X#@42eX87zJ
z|G=OBMHT-a<+txXSJ&}BeJ<b?Lre~<vZ7FI*u*)?K3KaLdNrT_JafS9>^nPSwecl9
zwHH%U`)@(1z8Z?*8rg8Xpx~m#{E3Aj_Nl)>ol%`>D6|9u8~ilS*PbBAfJNIDcw5=K
z-&x-oY*PxJUT%Akoai~3u7~sMcgA1n*|}z~(USXPQ89dMbR(+3y8Ws}i~X9VA;n7^
zXXDop<*E@mbA~t`U|J6ITgIvEiesHqL_H4AH0?Aj7GcyR(LMa2^IhXe0#VN_jQEL5
zM9#<(9pfgC#2Y&P)BOrIQ!shyv7=~Ac7fm_VE02L@EiveVLz)v0ey}IBha20lCN4`
z&GgEn3s=`TNoHT1u4|1|H8YNWEwszgpo{zxlf3LMsZ<pBnPZntKiF`)5BPi$OfS=C
zXOMtu$H^f15cfpz6Vh~p-0vC02{p;J!{bMZfsGti`uZ+}O(4-a?l9pePABZ}GTAX_
zRO2MrAnv*V!nCey%f&1m<a7GHpg&LO$S&r)XMnIcbNF1{iLsyrP}cL3V~_h>t!C#R
zSql9xW|BA66xbh~g$2V1hX}1aY%2=iv&9CA>siJM#M0b~WKUW6u^k)K8X7k$-<5=Z
zIX<KQavMPV9pHr<NT3Q(JYyaUdE7q|bT&k{8EWR)<FsURfvmwOg6V2M75GbjfBlTx
zN;rr>$S&6&6Z$pK6AmCGKB_S~%D$V?e3bo!bG`-`*7`e-?<1LpbYlO`3pMd}fOwXB
zOt1n*Sau|VoJ5YMF(<7fMF=g`6qKpKLpi7@9wq#Rdwd)dzNVjec74BQhP5ZFA#?SD
zpQfE(s7SiJf_$0UkEgmgx@3znZBDZ@#Z<WTnE?!%9zjzkyv)=7vY`pZ)ZCYXRtv7X
zEKDrQ2Sp9<7666Pn9NQ-&GW-<ck5o;;b$<;=Gp#j{1$7=j!cS?GV$UOQBVb&ymrPf
zHcj5WkrT>7PndJ|;_$MK%E3|ua@0_cZ?KB|>|u92SLFf8Sg<vtS9DQsXp0}c4N-d%
z@~?R7d?~#?x5<)4^LL)dx&G<wDoQECo>MDgUWA=%ZdCw>&;$E<pC!mQVaQ9V#)JJK
zhLgROvlSjSN@efhH~rQ}IUivRZ_5E|7wb=Be7w);8rH*Z)i~?Ly=RHJDY||vveR%0
zMi|&n0LtexroX!LUs#%K?Ehpf|H_@$t-syb^I7!<!fNFras1nxE6a<b6qJ-L$}9|*
z7~+1Q{#mQOPSM8RHy~3n81!88-OL<M<pB4s1c%nd)5ObLdO<N&2@!~PQ(0(EF@?M$
zm<U7}T#`-S>3Ko%<n1P8-(VL}^YM`B5qCYin;(>3==!ZiS#wD=jDp)l*S^fAyT<o0
zehxi+?Bux4Fuj@nROz*N>h;kCHQPr@5h9#|!cQGtY0#=npZrP=!Gps%XD%<m7S-3K
zdzcWZOiQMImG=IN*VTs|vOM2eZD+)eC*l%xMQFQZ*H0c@jy;;jfWenrOjOSosD}#E
zH~bnFq(&JIh}2uDgEDTNDG^OP=`6TnI{|`$7xG6egwW=x3M3a2p_dQ6$>aD8s7S!(
zDjsWr*q;usdcVEUk_31mfo?B03dcqTIx3@Kb$R0F{z9+#GK0K5L}M3|rU{2uEh44w
zcFSmdU=!JRE#=fxB0#L`NKB5B=??3mG<sg1Me`KGdc5zP2!UuKON^mm6;l6`M~;0k
zNYko(9ovjb+|Vd>Xt8}swy8ZkOK=%}sauR3JtGM~^%!DHswIDRFD5>wLgr>*zlUuC
z%;87P7^S%TP(es6rY5dBQe31Ts6h(Wun^Z(fiwDi;M4rL$bPQIyVn3_^>C$`-}>Q<
z+YbKj&Z=#VJtNg|S8Pm=7`CP2-G__>SbD;SeRc4iRJFAGIL^jz$j2Bx17iS>4TzDp
zxB1x8?rs*ewN~>2K=Gnl)}$U61GZ-rBB`bs@4CtsNaP6g&9vyUJMQ#rz$_PBEl-L&
zEQY38Iu=M`jHBLT9h|kHcCL0{mcHi0nXQC?iKF5bas``&+?CoDd4`ryoOJi<si-+u
z*-N31$`HE}%vQ2O5xY5EX{T)uyCwdO@=>kQhXRza4nKj`uT%k?rD{Q?^U6OCB|6_f
zR1JRIH)4+Hkm}#US@6#fT~?cR=Dw76@?Z|j!XE3uxT5qu{wcdH)Yu1a{W|RPwQ3>M
z!+`ouPP0O$B)U7m_r-8o@`%>t&kOPw4`cWSPXCX=5DW7^Sb%@wVVlX?HtF=pLD%<G
z9*~Q+(R>Wa=84v+sD)!KDGkynL=E8yNg*aQ?{63|IP`He;^t*Ks|3}aQQP2e==^LD
zGjF}AeC(b0I|Oj<aS{i*M5M7hh`?wg+4<E;;TTOg1ihTuGPHb=T`RCCdeT8n<22r^
zwBB&0hF|?g2hr1$*QbYOCd91rOO8*gpI6@8ojVn{VbCxFXHsEeoY7j$K8zfh82V!7
ztwyaIr4O**Sk3|rl%eKw;{=^euydo>H1dR*HFnkVpUU!k$_XIn>>LbpCJ<HhWU_NM
z_&FI{PQ*TmT})O~Cb!h{7IU>?!E&LXNUq>aIF%`DUl<tKD&kkN++QeFlm36!8qOu`
z@W96L89o-6dpA2s${0uA)LR*)Yt%64<^W&<&WVwWa9PA*td2C`dJQn%)l_;>o%L52
zR|S$qeK3r_8(*V41W=9jo(v^L*8;Ujz+UkF6@xirV>d6!j|Z(4#dI&PHuayBp?Fwd
z&my^}j`4mN?|+)<-8qOWH|6dNfy0kaW3PPG)|ReYs2&FJUFg~kh@v0ViWSj7U{h?X
zbs@f**%JE365aDx44#kXDY@3n8$Ex)TlwknD84x8ogtIxMVFg|ZiPrxOKzvQkv(!s
zxJ~~q^zwJ(A?dXKxc&atTuQTatvKb}y~3)^dywMhPWQ!|jq1S%r0=*rxsy8HKkux)
z81>$31L55Dw=oOD-?WlgSXut-Wc6Rz+hU4#bQ=G+y?vlKK$>Grd(ih3w$D1HTeBpE
zlWrF2qtJx*fvKZ?end^o<gm1h1fDH=u-|7nxOf~2@jHN=C>{@8JA!9?o0}ih!8AP5
z$Po~5b1a`Q-w%PF_e1YfSO}L!!+_S6U7@gqVr#x<5if5`jRmm&-XddYdh6il{OS)p
zi~BB9&W)R4XEz?~$6<?e>pz!oq(b113hjo@JFyyO;q*Z`zSL)W%Se>I>CBKt@p_5h
zJ1SlPIjtJRvig`t;%#$*@IhLoq-v!@wa17QV+_DFtl~+s;q+gfuw-{>ZNiY1oAyR#
zw&iE#0C-QJJG4g!*R$X?`l4Hi03x6Ak*`dR<Ok+?%+2WP0Z^7q?o7QAwj`HdkjkB<
z;#x#nJaiBJVkQ<aMf2oY<*vCw)bfTcu!@~1L+UJ%2A*XKRg7B`4%!v(c?pAej7&ZF
z_q8G6r2#5Oi%hA875-)j+_Vi-lwo9Ul+Zq&(Ld<(rJ!e7YcCFK2haGaFCeI2M%BWc
zYiAtv3D(u-4i0}XXOY=(l5rNH?x8P>>DRcPS5YL!Crj4J-B&!Vo1y9XSKQs#QLY&_
z?w1?K6v}U{Xo`*N&T5v7{h^1REgz`c=KZbm!(dc)NI^^5LGL3w>vHBu@9{mQu$Oqf
zaYE`tjt0weJB?AMK45C8tI;%rJp)`lVPL&Y)X6z{fk7>Kdoy!UH+kkHwy9RzJpVcB
zbn6QE>m@wz+F1v%tjF_jIFxTt^8bAH*cktz`TYwsoJ^LM-KU5Dj~TYciVfg+lXBES
zs8l$agK|`Wlq;M45TX=M44C_T`)yHNFqo&CaA;IqYf5_G@g&vjYTNU&^@8f$c8pF)
zCJ>9?7kZ6Di9H1ZG?eLbve5!)Fmm4t@Z9iR{&j!Re2fHZ?eW{*b@N(VcL1$Z@<XxO
z+v)cn=tLwBc3UUawd{Mo`;^%_`#waAT-R|E!_Z^ix34Qvk6D!@L95U`pFc8(U}uAn
zM+9vB-ka^oBt9i?#S*`4ldAVkMHKq!K9pji+0X<Uf_V0aU5L1EZ3f>PGy34+T2TSz
zRLO=grGXW6j3z+3<wXruJQ1yHk{24GKfJKHM}M$`4q)T77I%{CgI=+`mPC`66v%Gu
z95OG(eEv57WiC&qo;q7J(QE2jnlbCTj1WV~{3*vf<#gApi@<wM)}9dhUcf||Nt%gA
zB{H7$kDIkZLmvcu$xISrJ0HPaYCwfqQjL{Zh_x6w1}vJbly$iW5_PJFSQ+zl(xkfn
z<We6YIHwE}ha3cJ@Z#nbJH~ysQ%2A!vF9L3eDD6N;{3o0kO$j(!gwyMd~A0|KWW&)
zl%6ar4O;t+e2{Ow>`uN-Uvf3buj#afk2Mx^G_mnQQ%PD)ZR<)fA-f)Q;f#1xd;x^Q
zf<?H-XLdGTu*BMVQqTLR5yj3(O=5M9squ%$u`3L%qzS(;t&$O<OViN}$|l#WEe+Xm
z*1%fdTzl%^*FlF?y_l1q7$w$#Aq;CHrPli~%5``|s-1CRC6y{+KC~*hs~zRmww)Qy
ziq`QmFJ4xSAc4PJcZ3zW8vj@idsbX_BAY9V1JhDe>V_-dV^yq>P};i-L{$U>1!FTH
zc=WE_^<t`(akVF={Kho%-Yo;8hXQ%g#gy}oT?+BZ6C>cToLp<VebWQ%Fz1DJKv;1D
zPbW-K0D*%ofT%sL<mZP3^6LyUX384O^wzAVi2hrC!0?x`INSe54Ezg6xm44XU8nyq
znms4N@-n;Eh1_b&5I;3s{0g%x34a0eX<EW~_c5`P*}Mq~7IMl085i&Kedctd#0&}H
ziQu*80v0wz6!Q`ww49@_%QyCq9syDK_+%nK;n7<t8w@<*goNe0_5uMVYh3yq-WC5=
z8GVyNl-=#31KzvD!^`x*^{fvT?e9X=adbQBik1~Aemc4NGrmkI6x~vg_S-gxA`kf&
zg*wW-jo+)3K8^*JhJL8#(aA({0RJBkM(KS`ELxd;@r*<^afEgMa938^cy21o-<7!V
zH?UO@c_EV6a$&@|YT|);kV4p~I|A@@qk4yHRMiIV^yi9~gQShRhKd93i%FG2)kbiw
zz8{2%8YV^a4^t8mJm$z+EgM-wCLYGJ8E<8XLg@p*!~&6^$SP!+x+`sw+Ex2@snZu-
z>SRS3B+c$mz!s35Eg92w#O;=;8sZGHlBQSO>#RLR^zkOn#zoe@!cb@w-&!+NQw`Y+
z67u2YAY3AW0B{Y!nB?{^G@j;-b+G8il3)`02q4pACyH+;d`P{P54G$o9{jJI2}i@c
z`Fs${(&Q_(>$vFiaQcB2p@<|&EYXjYRv_dRpVshY&Up`x?{D+6aUd1J;#*3tnur(^
zsxezfr?@)JCJz1QG8F-<mMQy==rCXdt)0DBj>p<r#Vx!Ff3%3RI%$$8I0}St>`x*5
z>SrIfVdWPG-pO%cOp^Sma{KVB=}!byi-I4kT4s~?09F|B8pqBT$J#~fj1EC%r$&>g
z#&LIi!+0HzdA876cb7Z3&D+%n&-8e~>)}R7ZlA9Q0t6^5V4G=BOc!|o*F}>{co*S^
zQ-^!v^ORouHO}Y4^HfyfIrU~lLlAfB+8E8783+YCs@BEIYqX^&DhylfBLkZ?wBTnF
zgVvj&De~hLVEJSh{@=Xedt>T<X3O;dz8<|)owZ)0NA^6XdPgr9(ps-Y;%g$8%GBU2
zP1P{3EQ1k1gM^S&hxG2o!iK-ZE|uf#9g2_i^qhL+SdId`6Z7$~$PkWH3HJS#5E?`Y
z1rs7C<Pq+TCq9d}iIOF;YE$r??`nIuN7!Q3rtO8K>l?4Z+RV@mHp#hfT#uTnZd;yw
zJ8>xA5B)7As}KW4fff@H%2)!ZjG6*e2Z~<R!MH)QA6Oka?ult;tdhwZc+EY0O8?0#
z1#_&(d8Mo^1Bz}U#HN-v07Th>PYX@HJOS`nH*1uF443bcRvf54fT62-KtuyF?pQA#
zH#|>WOoQc%ys%up!WDA^wXF;wV`reTzJZO#5N<(>6k$PH>)&per61y@#3d6nN{%6G
zdaB5BbFa!OZ8z{f2G(SWL0NXdxZc+nrum$Ff7@aMI;%9IJktT?YEBT3t1+xY`Q-)u
zqT6U|x)loi6!qmkeC=}8#O*vS4wqTuO3Hmk^$~394B)}EP6>5Bs+nE>n+`%w{88ne
zbd`Ekb&BS@*pom)&FrIOmg9fG8J~FidGs-#jF-{%W&WWea=3btt7BM74(ErnaDy4C
ze6@W4h{oG5eZuBl@myE*EO;Loz>E4^gf^0d;+dk``wG6-Zgcty{-F~65#3s+`W2mQ
zh0fvyjM@px{#Q%-%hfb1!@swrLsj|jx(nWGx_SqpVE5J1@WybdP_y`VC?hUuqvC*%
zr5Ldgf4XD07bz<RejTf3veRaKB!{>2wzKn!-hR4oWV8SB=gvWdydZurB-?xI&(>Y~
zD0&ANK1Nz;029yK?d%Ld)E`Vaxb_%gF_nv_?Yjeb(lqJpD~_?oT-fQw`{AR(v!;zM
z3=LW)uk#Kc+*uJt7N;KXw~B`bKELJWk{_<bbu*N^EOUVLIhsSmxqB1GbA`3<NyX-N
zeqmBg5UK~cNx%<sTIWE8tEi#N3y|amhX8;!K#Xy(k{bdQ=u2VPx0Z02l(WN)QH!dg
z$MaeS6RNs2zMweO7&}Bv(rZ@CxiHN~HT_Cgt#m%yRk;ET+!zbJbRBVMh0&Y?D!Tc&
zPP51u3U^VjiWi`(sQaB!ZqBN@Uy12wTtJ#;EovH3BV-eLe@q$QU1cxXA5}v$i&sR)
z6V`PA?}=4OfCTIlupbjN6F?}zk{8Zyamhey)f4xTUa@d{tnP@DBusW>fD$YoL&~>T
zGHsn@KE>5q9$eX>ul1nfTQEkVoC{R`{FngH53y1Ol2?&ozTF9m58vVmO5YyZUl&$a
z4V!lTwB!pPGVke$x_N5qYg+b4L%HUZ+|B)g)?Fj=-tuGacaY)f#^T*}iRs;*<0qU&
z+sCCD<5euqfnDUaBV7H1TP?N8^9fuEKG3@lT$pyR5?MO;&z5=KG`uZ<xT21xPivX6
zHKlrwqD?#9A!7GnUY$O#S%%+p{5q#<#b~hSMaPNkE@=ZiPUV*`fTvcCCQEMDGwpB2
zJm2-3EFnRC`d{ppp8mgYKm9-C7dhBi|Bw74GxNXJ!d?9D@{1E<$Q1HE2>5umnmM3;
z#K8?gaOQR75n%0smjI{T*ZECO-0k)4V-uLta5}ak$ilq5^4J_M(^Ks`g0=*vNO}1a
zq&4ABOW@aeF_5FRWsqk=NTkTg<ZI9nr80=goz#A0_GPT0D=3X9VwM|8_)?DsK_eH*
zl)J$#{9$rYmTn07@{`oNvE{xYhXG6~Audu{+r2tv;ld|}XYv7^TH0asWm^yENr_gs
z5XOxV-0?(oM|k!5Jx9<H3Uc8<V?vACurUgW3kboOe$K?bOYjk4YV@`l1u1j}{9%Ix
zHN^(F5ErK=$mdDuwL!>eo{N%zA*!81F|2N9gsJVkq2XA}3PJ)M!l`xP6icl$l1XfO
zI1VKuF+J3fl13d=Mn6sLEl>N+e3j7psRtBEh*R`<%xtkeGvNtpg_9N;)^mmtscl&S
z9Fp?Fb7myK8k$o}kJiB)q-!FNl&p;COV*SbXq6$aC0LVtXc)i<#Nh|<s}yR{C2Z>e
z^MuG|)qt@4axPI8R?;K)2$XRX4}_FhLE;OJTq$JiLG$#iA0ab9Qc+R?Xt482uRA1B
zNQrw2`IX1;+wA~#elilrSSJ&D0SP6yM1k3hk1)pKTPD?9?m|lX&`_+KK<UXT5aI60
zhT(OD(I&Q`J}FjJzyQ#zs(`h`(-?Jx(<+uCHfd5P(ZdiLhS%2%dw1z<w$(E&XV(u#
zTz){+ap-%0@!Ac3U5&ExY?~tz!WVelgKXrwvq7g)?|zQfx!N#bZ9EI&ORl1O+!@Vw
zcj$C`cXs?ntMY!?+jz7|%C^}&1QI;DN$>vBFxc6|8yEh<o}yE2W>cXWIih=29Z!EO
zK7Srhm!>JZ_EPt1aeKD>Sk+#E2BV9al%#&R!mO9KD<gA;?yyB`>{CA8`h3M|g3F+>
zemvbD9{-RSAuXJ;=a75%yM~_!Tmu#gHf!GH!fE0LN@wm35;%9ZlcZ@Wq}ga-3c^)=
zm@}61g6o+C!IirOFn*4Gn88iSfV`N`l#)xyg3Jq&f=Q4AiiM;S7eygDj54O358mKS
zD41V{rX;dG#<2Yby}>CY4;{i6I6X=4rT=VHKzL1vV$P|+z>3cz$0;;HBRsc|>&aOa
zGF6RaUK0XVC%K_lTQIdYT)l<V59Fu^SHl9hA>@l>36EPU%AJpEC&{y?pQ@kT-sh$z
zZ|9e>qsP9mnsY7ferFHWAUcLHO)k2j5T(}q1S_YMCbtMz_ted$U|e7?{_8NV1Gua~
ziFTDz4EHyf^fS@YpLlrs7laB&FqJ~ka>k9p+Wz7P!>8#t=c}47nP7&?zLY=bdKcMM
zD!<g)S3Ek?`83B>0)@X`=dYU|UyjQcv%^1EUtjz0+AyD!KcYWmKkmcj@FudKLx|1E
zR=e=qfz$++B%~j;a{&sAQBIbtzVts#E@Sept?4f<I?oE?uYV*2k4nzP?D`@?alX;I
zfBy?Y&QAV}?TdIBslj#I56WDrXdDqYc~aXtt9W4E0;{~9ACq-hj!{g0M_@xg5@tBS
zz5!dmNMpgI*NE_0?B9>MflHJ2`F<HjcNUJUI;UY$+uvkv#^Wdxd#pI{R;kLhJRyGQ
zhdpJ&#v>IF{npkze=RzI0h&+RE7csV2U+-KEf)c}o#>z$73{+6_9ib<bsd%+Wn<pj
zO{HJ_wMhLVY#`IZb>lobICvfsHiyxER@DfshX&1)#0mxWR6uI`$rGHn*eRxt>a$dA
zwHxLLZ!r=;3XU@pWRl6xy2M7XkOy22nr9967>NL-4?@A1Q=ymb+Ykf*`yi>uqL=S}
zFhsInunfENynSIz3fVQYFx#Zb!4O>MjnI(aeZ|-uh{nueh$Y&5jT6jVO-kdZ;WD#q
zY``Q|xKH97f^VMl#nZWO{v|RnF#h+*@XzR#h3y|XoPSX%9;&WeuYJ#Ij;Wqzlg=i$
zI+8`9it;3&fJDRG5c2deiF+Wh@x8O{mgVNxD1HPIkIhG|rmm(sn3$mXdExu<Ds2TP
zhJ%rSVOSE!lVcOm#?ld!_KEJod)gr<tvm*h%5<W%;j#>vJ8L|iYpgWtF3SF<a$Nse
zsorVn1$!p!7O64})wrbm#@ONuXO&T2Rvv)ll-8JY?gh=K!wfPlnvb3&Bv8ZWpSvKN
zNJ$)y=UVEQ%X^p7w<cDOx+PLI?6C*4KLB+fDA|YvcMY}*cRNxws`49N_Nu7<@bhYj
z_G5Jh>X}stYR4=@K8#r;T2BDf&yk*KK}7C%tG&v{ZNZmk98L}PG+62gJy|N*SS4AO
z8(|7l*R>hE>~Xx=yr!@}v5p#%i29V+&pja<1dV}nW4O7iNi^1VKSSo3a;Fq865f|f
zWLd<Kk4)|Gr>ou&)R{wh3VIfE*cc+8J0s4DzE?BemDQIVDMN}$KBT-b2*7hjg^mTJ
z1s)VN$ku3=Tq_Fm7}5wIG{Mz#>s)g_^$4~>nuKlldm1I70Q{6N?ehX`>ip4?k?Nfk
z+mB@o6|lM^HFQ_*q4j?Ik*eX*%ETdY+E=CpE-vUU?}!3_`XkZh+b`gEC1(NHRG^DU
z!12wP=`5_yVRrVLELt4mwP7d$_O=l`cgznUyoKVJQA!P7)xqpoAOG~7A2BzU+BMFL
zUA#amlCH-I1ayERRav~=0Yd|5WL_f~T~E$mtj)jRRd4|udG9&@W|2&PV;<A9{(H@i
zWJ#NSR(PN57ZlDh;fcdHJ1Kw5cHw47L_8=<=>En-ZNNC=k*2uI>T9_IVu@HJt`#eH
zM(H0M-+GbQdnXP7`PsBy8QLB0)9OYsu}UKKzXyzn|AR*~YYO8X3>X>BuGih_BwGP7
z2rw^T_m#FNJ<qPbGGTj;r$Av>F=IqQ9A0k5yY}jQ*_8I`NF&FczT*4y)$GjnY<<>$
zX#=M+2>K-AOWQ~Iqm8(}tthgNLRk#!REaj+Bb`L{o>KiPd~Q_?^2ZWI%upLcEj6z5
zy>e~#b)4m;_E~A{4vu2vA3U`k5m0KY1u2rDTO<!+>u{ZmqJUokpuvE`Fs#+iX`bR4
z<2f5Ynhpa_TGVioyz<oAw>^o@KyaNure&{=88*Gm4;^faC7iYh*c;afbkhA1NSDwf
zOLC`D${`i)8pY}dbyJrcCD|nTK+12`%;*ggULr@jaX&q3>E?^;1*X4bbIFbax+eW-
zvGn|><N?3YE}ko-<BqX~-el!NqiemJYAMCo7}Ay<dig&oU-+Xl3JG>((=W>%Z(X}+
z;$6K)o=#ccIEx^XYzc;^n4}8N=014YdSAs5lFPgcUJwB2TZ~V85o%c?%;jVW_&F+C
zWuJoW4GsI#9ZUofTlmiS``XkkQB%{c$s6{6%OBCMQ0ugN-fbD+vdL(~j!Dz7miq?v
zXd<q_mh$5}SghYmIu}t$%p9s^V6a1(qToiZF9mg!<tvN+0e_B<9+(m<P<K}{LpBp&
zK@95vJ!L8zwF81Q4zY0H^&?Y$0O&9Ej~4JFr+vHOp%SH=MaL#@dQQUa8Nq_rn+c9}
zN9Te>=o<k>cLm7Dai$MiZ>H~Ssn`gL8j`cf5U~J3O!J1u9G|x(P9c=g*^gx~8t@*@
zM-{94DNOD3mh5_QxLQj^i<dBP6~v=}SRRjG07KceCkP?!e6J264W5g!5lKI*4{rFp
zAFhFq9EV8_7#pRe7;oXR3WEwh)dT}oKpsk7IHIGr^ll*l5vps5&?bC>oo;)~b0u<q
zph-m(JTCA33-NR`F&3lcY{FTjmLgn=kaAE=-9tqoIslmya!p>GSVRLRCt(bj;4z1>
zFBwzf&kl(awn)*nksy5SD3r8p3AOhsvjMt-#B&n+g5W;T0+-+!9nBd%9`fOgH5dAV
z#`*x%5y5k02@$Df{hN&aI$JXeTG^Z0PvJ72bJhAFLh%7{bwymB{dh|DjIb)3mHKV6
zkE0a$7D$jEh!Ad|L+?>g6bqV2o@t0MQ(9a*p3B*4eRBr&T!oM}Zf1cnw~$}9BOmd(
zl_55k{loSdhC(jsd+ma}VeKHf%a~XBaO4EB^sggtqZZiSw94TZF)q{dk>1p<Lt0<q
zh8KL#+X|(+i3jT+%h&HAZK>Mtz_fyLf03zYU|{-7IAma8{-?tK|I5Q|s;WDDAAWl+
zSMLz9AQ7@d_>E|utz{Yw-4dJJ%dPMGQmmPb3zC$IomF4&<p@zWSjeWQ7RV0NzUe=&
z1LHs8a$kHfzEPxh@)D`JA&!QK=-6M^j_m6P*p74>?P|XO4VvNJK)hwk%EoboeuT?j
zdvSfZ;>tyRG?S4PCnuw2UtC-)!;y`)hr=b`ryqQr*T?*@6O5vd-cbXV44DyNx}XMR
zg19$;k4`~_b9?de17)n!FxRFHw=~?P$OR6BCqf*<?b1$EF^iBOgg-N`Vi<<KX)0ZA
z)~#zOPj`jazp&AZmvR9V!lyN)!$)TDVdWn+H2GLAPW$O;)2-coFN{)N(YW7Hw>?Qk
zwy{Fzv@YYkUnE0CmAtw10#eJs8#qK9LC=m7gDx*#DQt5r)syO)dlk0T)FH76FEy-q
zp*KM8z<+>+q*GQk%0@@P^?o%%`h~Lk5NJe~?#2)(4!I{JGy74o5d{$^!2R>NZnjaE
zPR0D0&Wvmu>GB|b{_~>g&R<T}8~ZGKQSUPvgRe#5XRJ?f#{B=H?46=Bjkay=*tYG8
zlZtJlV%tf@wr$(CZQHhO`(&?n{;OTB{p)<~z5H(Hm}B<wjNSpi69|V^cI@!aTvPyi
zdsdFw2?tnH)4vmGE+DC->zgid+jW%8l9!b@`KkF+fbgC|ykzCzn&fv$NpW1eB=C;`
z_vh&Dco{j>k2BT6ogwv-))i69%zWk0pu07eom{<ZQmK&1f*VGv7>W{`RAo7f?_wPP
z?lRPa`^MccDX~S28KS4OvvMpT>MoshD8joIrd9qlnrbR)i;epOHcQ}SbU0>@o7a2>
zoJFoEdE8cnaSWKr>Ws-N+q7B{)s#lz6%~!|^l;z4Zc#)%#i}_mHs_Fr&d|CU^yB*(
zFG&)e`l%Pn(niNYG%_XIWpk8`(dMUUj*UhqfczdV5uO@0ZbSvwX<0d48a0g#O$48`
z_nEX(FrWDigmJ`vp^X~9BiGUx;daD#R2L0)%a^Kr8(j^O@621z064>lzyB_-dJ1O1
zg(uCV=Athiln-L^g9086ti*|!Ov-PSCP4J(yA>?)BkU4Ns|tPAudSw9B}}EL3`fjJ
z4WRry))5ZwDw}pojh(-Q>aj7z0)*ZY5U*IK!vd=)A;xW$LxdD(YMbhJ*yu8d(V`yb
z+RhZU4;THpSJ6sN>}OFZCY(^6Rf~jlQNm{Y9vxFCHw~O{-IjC6GyEV#ZN0rSWLMzH
zuAih4-G9Ux;kz7}ch21|+Aue(4+iJKE)`Cd_bF>Z?-~B3=(2OV)UT+LGnrFAXNob{
zF)1Z$DHV2UWA7<983@pD(r1Wyaf*(H46RLFYrypDT_}V<E*8Y<0t66ws@qwC&g}1|
z-(bVN=e;}R5qtef%#Qfic<-Yi$+i&3^PL0rHh<sSJlB*yp?v(|G+}p_cT38_PdD@R
zLiIDV`oSd#NDSKNz*A=7cXoeaGde9F1nuprdc1IuNbO&vN0mGd{*_r>-pp+|O;KTO
zqP)0PVl(Y{OjIELw2(R_1Uux6=yi%_!p)EA6O8O2!G^yc>jtC)r^otN8^!#O5fB6O
zkBilRe*gcK>?iuneq98n>s;kV+1$MHx%0rm&4$S-O=hzt3@+1_2`?<qs0rmXuEF4S
z8T~y+XfA<>({?TViC@<j^afB>&-zsF`tqcB_2eajiB2fDU@l$DUCPZ*GNARw82WmS
zoCL!=<I~PV#Abzvn3^vPnQcI>^Y!&W--!cPwxp*^U_Ufw#ryv2V{dt$_u}YhdTr~?
z;(5GuO(#?AbiVn%&0~s4e35V#J#0c~7Mg`H6ce`<K!p-8M@284H-U$92+b+3GRrub
zZ)cTkTiZr+M^cTehC{BTl;Gs(nLO;MevsJyOHUgBh?EJ49SRu8j$MCcG+m1(PD!2|
zdYli^iUA@*peCt)lCiW}qH?QeHiK}&FX&E|`o@wg7ULbBo;)dkACFBf-XPjk_BMuq
zU(mV<=&JeDGp^sU1pxa*{%!E1nxyJd+t%v20k`g;_wUmAx}g}|o*@Eh?*cb}{ioZ{
zT0oyWm#DG}J6Ue_f^(rlO_z8Dky3-ig|Z{^X4_=t+g$dx<JGO}Vk(K40E+X%%9~!X
z60mOoiB{9AI#XlRQwtO#X`dGu{IXeBL#|;p2y%<Xj&)xhy4MjJJ(;}6zOh=Jxi?-5
z)*y`SRJx5TMoTE)x{bgU480$_f4XiF8^|*S#4&(8KoTK0Cez`zs2SQMQ?Xa#QEmF4
zkat4QpT5r#-i?fwSwsgI9Y&FO8wlNsKX@;1@*aOLuX2dVX4V+H2NVQ&$lbP<m#!NI
zBZ3I7)RmWB@deY6Pnwv`1(VUuX&uOa&3-EFPdt|p$4u0P`Ke#MRSvx4K@*yIhOkk=
zulULs3PnYTYThJC>*eCpb#%MTDUt68mMl;%q)Xih6p}%g-*fL$P&x2iS{Q+MV*~AY
zpI?7S)AOQx>M<_Bc>UChrluAgX*vUPZ%t9+HWE$2Ps-hUpfI{O^?H^iP2^zEYkB4N
zA{&1Gg)q<aESesPcX^YKo8?Gyj{o{;-{QJv!s`asKCA2z3&iMYj+A`s%6ZB+1)U4V
zw&BSNvgI`Twkz4^F;DAMBZg2UjUwo9LU9yB!&fvaPreUn5c8Q&y`KD=um%XC3(jJG
z<@D~MC7-sj02zz#5p9!+r;uRTv_>Z0Qm^<Y?Q4|9!TD5}Qk3Lm^E3{3H3Y?9#@A3>
zY36>Ewvsa8$FpAx(1r^+*Q&8GYgSLh5Cjh`j;Aq?77|=NwiH0T7rbJnR0!xUK}I8n
zi=tU*swY#ieZ2>YD7<z^EO|ps@v6zRWjarGT=v;j5CSSN=gHF_n4(q@on#=T93LvM
z1e~1!By!904XtImgE~(bN9l2=-WZ>j4;)K>K=3FH(}0Nb^PFfrOtvXP5){cnqdhT?
zlI70?<J?2iN?P+=qYUEa0mZhLxj?%fgEdP^MP8fKuAcGgdEeC(^7%&6(40Slh)}Bt
ziPL1$UoLVOh0W9+u^fnvs>mxh#8tqv=ReDK+}q6PX_S+VLV{*;kjR@+Cx}F3G&D*X
zhH9X)npGv<U9&34m3fZ++Nd(o{ELn(vHE<DVn!m4R4qD-@*m9>*)3gS<`+Qr>%@@n
zee_^~la>6Fc@+w<!;OUvlf#52L&8qYW5GQ_pn_yVjY~9UE8Uc1549r}#L)<Kr`2xN
zppmgM3J(oJ&)G0gTQ4s0<zoj==B$SNyr@oiu<2P_C>G$mitQS*%A9qR0raM^xtSRV
zC)WTcSNLGdS<UyNi9BtqV{po4EV=9*nTDFp=_+Q8mtpBn&gWs5Tl+TY95-ig=F`nX
zghsb!(F%PCY)b2to^RT;ffu3+yHTUg93r+S*Dv)r@y*)%Xuqn@5Z@&57*cO{=|dXK
zx)F$i+mCc7*a{|=alhFzzn2Mew--oZb0Rjcbxt6jsG>%k-wo{Fjv047ldBWOyBSBa
zKC)rx*1ZAmUbD;pbqoCd2QkCW@_&7d{tJ0$@<-mu5c>~`J0d1oT+PQj#bJA(q+h+K
zW>!dorDrNTV^<3hed+Ou#vG})5kVf`$Rg_?=%_VSN>x@{Z(DQ!b@046n$F}`NF4i6
zdR_4s=ZMqq9m0OM0_nIOo4{|j51GelJBw~xN1K|Lnv1H&-R<>u8vgogDLr~zxMy>*
z_}<I=X?gJU?Bj0F{MPYtIGt%~2(5)r^<{_8#0Oy&DsyLx+$Myc9K<%eaW5<O+0;ki
z^jyI<b=&w%FH=-Hk;n8`!pv+dK!L4kh2ohL5oaSI2Gjw;J1Df}gUaEjTlrp51zJ{g
zt%jBNe0YQD0};5bKIA-c1OdMPb^sCB5yxh>qP+?#R>4vFT(~B075-!W_eT4A<$hhs
ze4BW9<W@3ke7vn|dYRjs*^D-g*h$i}n}Pl*k7Y6W*@L@x0Ux(S3@ENdM>qtr2<8AX
z5mjlFNvp1Jd*mE-+f;(Q!oVjI&@i%Hwpfy3Kq3J9)-N*k+eL${UjRGq3k61c#FP7C
zK_D`=Y`8edXpOabORP$%)%ADUN)|49Okx<Knsyp#7*%Tls#ByOR>*+!W&#;N_0o+H
zo9!*LIl|l#offzEmMyahN1Q^=Q_t6B%v3Bvh=MnFji#3;Bp~91rsZ|ZVwn07QST!!
zG54#)7&mU|BI$J@jPYZ*c}Y3fVJ*^uYa*~<&z?3qc*$V$aTUlm{Y7a^0ctNur>@&*
zM{gpG>!cL?386fcF7V!QQs&DgbuH*ND2!$slGJ1bb3PcpMH}L2hJIf&D#jY5fN7Zc
z4~J;}a+AF~vD}9t2_dhXq~5A?-1%RwPhZ~;`25`Kxn5azfO~$Q<_W=c^ms2j30y^!
z7X}$0L=oSLiOaSL!cJv`%KTm_=8Z~=`H_~1rJAUdy%W%(Kt$ekE0iDLB`%lx_{-xw
zXW)EMMPGXO<3blT9cwyHvVuD&I?wd?*+U1sqPHVJ=Kia!a*{%RJzxhdOjE$4KsJ&1
zAO{%D?+vfU@vfND%Ngz=o>#47b1oPs`FYo%%W9b4OdVcWZVyWlSJTA>s*7kOkO3=B
z;1@AZ#-apE^Z{z@*5nst=GItT4%Uw;wbJodNPyO7P5uriy~Y<p?UxxlbDJeV0dJDx
z?iA-8Crj#N3%eH`*tWo&rlLw^na)F1`uL$fW-7rg(j_M`8iB3&5py366IXWrM?UN@
z7LU<e^l(<Jwq}>i^p$bNwN_2@7uqc0tzoe^>K9X5_7o|7Oh$;#w;Jo;<!E89T^gAm
z&<&()ssAhu|Dg2Q|6j~yl68I<A%vdoR}@z%68#T4rw@n{ODTgK+BC9A$V3)C)$}|P
zhtHSxmkbt4ng%3GQ`yJN%=Y`uX_I8PX1qt93tUonaPmJcK^TLH$r~;LuCa;%V<VCx
zTkzSoSRYb6xEb{3>f<RPp-C_iD@w!o@N{w7yb!KJp3;8we-?Ooc-vn#-lN9q`}rh!
zPqR5*Kk}AE7f)Z4s~f|)&-{RNq+A$Y{^UKOvP=n7iCdzy;mDA~?>l%YNqOr`&guQ8
zp_iF*IoB`<*=d7BUTT|kz@ZmTp`Xl?#F`y?F_c4H<$^#h{PMu}uJ7pU1yN>fSQghC
zQ59;4((MTakX^F0OWvCz*K-WcC(^BjCsfmO+b_x-cjI!N6=1Mzm`f`TV1yhp=z|sA
z5B8n;vW$rYUp6={nz+lw7C>o;-!&?NwE)HIZ5yH&>=)fXy-9|?A*2CQVwoI=ZE6b$
z$Vgf7`3<T3Tb<M5vJ%InpSoMi<ed^=phw0RSj?pPjaK8Fs(PfZXxRzdLdqUg=u-R5
zQpvWvY{Y_o{BzA@(wAU^EFV<J;?#zNmbGU9;t&DvQPq=~bUr{JEgll%q<GOFRuCxi
zsodukjUFR8dKo%|uR#~{kh8W8Td{<qoL@e%#LB^^K!k*UWJ8lj0JF~{gDv>E#Lj+A
z_E;$`T4JH^Xf}$bcvu}fCyoMn7Yyf{l+mXsj|`tpB$;&81_1<JOS_p9q8z0G_Lf#R
zT_ZfmSZE&%SIiCfiQ`-V8FkQlXhnG^1VTM7RHY{taVKUEfCYgT3Z`xl%*(>Ez`Uju
zon@t`WtN5$69UuA_YtJk>T%0+d4i@EP>+d40E%e;xK__fUmGl_8;#yF`90(|g~DiU
z!LlWR`giq4Y7HjkNd%YEq*&f3t{Xx$`>uLbPC<mAxI*|`9K_`niz<WYY%NF_sAP_n
z!#Em<z|oOqEWdUQ$e6q0F3yN#65qJoGGxAXD-aqr>0`Kl-AJ#-QzkmyWE+wE@?EnJ
z!x6FEv{r`UaJv2akW(_8h=sL`2vJj#tjX#hR`<Dj7VfkS$x6XPzp4<9HeGs)LwyMB
z=`r)vsPx3a<b=Dtij6@(-cz>B0#_v`-qcmD?&Ypo&*3ZW_#{V-ApQ5g9$o+>$wpx#
zj_XxcDur6B%&$lP9H4CP&yA2}4%Yo%Z)Z~{x)DxZW{ss2E18M-@`~h(PeA!UG%^1=
z=`8=CR@mA9?{NRWXaW|KG@|F(P=>dU6%H5{k@h7m$@d^d`uog(V%m5rKp+ZnnIad+
z@=0iDva|Maif!j60Q{XWv!beplfS2jgTYt<VZXTC8F<~_;%1q}#LHTsj<SJrfiQmw
z7+_9R`#a~7MZ~L-pu$d9Z++Z94|dmwNMd!l^32<^dB3kMPFK@T*PiYnU7nYZuLi&B
z$m$IBJ>yFX6#5-%JV!8bo<y|a_By3NAYoJm;Nm&%6IY4HQDrnzvo>4Qf44A=7ggAx
zPPMSg_M4_TlrJ0^8-#&@!=WRjQMlr~WJz$?;ZaKTZz4Cav-S41<oc9a?sY`S!EKrK
ztd$qWk?0;{t6o%^H7PNK2juA0Gs^9ltyG$0CmYEhJI154s?zaBg0+ttTV33q?Uvw1
zm8MsmX&J(<ePF6i_Ge%Ki6X{Db{srW*PQMz?JFXTvj6Q=Q({UAID9e=yvt@jF#Y0F
z$O)aH-%9`(Q77F~HkSZF=N~}oQ;2fo%7(<fs`JRCSpp+pcCB>~mI29}TW}r-8c;ku
zb}tcENGbIW!?N~G3P0ODk&7c+Lnep{-Lo5kO9ryti3_IaHL<*8(c|=pvCI}8XlhR-
za>#N6v80PavH@w}7H?XPGt{8@k>KJildnajVo3PdhQ7$-R-$7L1_@I}FvJTF@cw8b
z5VqiBev0?DhsYy5;HdE}r&pBPses`osTqV-jE?KmJJ$S-=7WKtXA2>tv<G)pQhrUY
zGp&2X<&hOIJthefJlnOmNq;v6nEY!kZTo4fR#mJV<9Fh4_uC=3IprGnS-~^dJ<I1}
z6!C_Mp)?k!<xp6g%}2vTazAE0e>BX38fS(&BhbK-74*a0ua~zm&7pR63GPOziOD}0
zjE-=<g<Qpw6<PsTXEDTq5JY^3@6v1Yyuw@S<~`B;o#EEzS!Z<a>&XS^V930}sp5?l
zbzIrPRG>1~E5d!orKd>Qx#X!$;nV@2-CjE(-tAzD0kDz=nYgy~NGNn4e3kM@E*Fn#
zAyD}ZSgK|_0aE8VSjdL~8Ns(PML!SyEFn~T@7gfWk;Opox6OgxuKNcV+D-Vt$E-mw
zK0F#c1gey+n9O0O>VoQO^nnA6R85WuwF4eUrP4FagItM>A_cB+;N59+4+-hhAN<i~
z8*iBj6py}JS!kZAaPHasCodv6$gm(Ae1>$qp%Z<(EACalVWBuAiB<6`da?M{;=jqP
zH?-C}DXce7I9iMus8zL7-6>A^R<Loeb?KTw2(+Xrs!-$5$Dq~sDJl3vC8olPdq{Yi
zzEAuEJ3Q|#wcpiqyHnZ|(_KJ4EcEFAtQ-Fj!`c4d_V{Gh|1%r@FqL#!P4C<i4HJ-5
zg341?Dy@t`hEbKcps+QatG;%j3$?O@0G+#QcurkqH!(J;wcpo0{@iMJx!y@(5+%Z%
zn+dJPMdJX3M?CUH)?we)5-5FU`f{<6s;RIdrWWu+M#yu*^<?9U21n151-^hfc;ye-
z%&(unm-V-HyL*^H-iV}=nRvPIu*of!nd0#-RlLwMKsh0r2<1~5q0;J7I#Tc@!Zc6J
z*Js2n9v@e`J-?&HO&pY37l*9OOhjU!sz@DmNIOzV&eylRT9`qmCowd9{^|i_;o#yO
zSkbyVZxXCc)9HoGJ7nQHLq@<sfWr;LFjd@xPo!K!m)uMUG_zLNY|U~du?z25=-*Rs
z{uPa|cXjT&7MB#RmUYPLMl}kJz))U^-u=*%0<A4N1PD-X2T)Bt2oX$}f=NhK$}n%x
z*C|<N%ZP(rex%03sMP|0bw>hL%tRb=fm7u^(h%d0r1`<4S&rATgiS3st}+zD1)A^Z
zV$}x&VtIwz!?OkQT~4IV@vB~*acNFOm`7raS$3>qq+RZoJq|H|YL}oIwJg3M!)Tm#
z0~CZMO1%X*LbiW%<T_&MO&$N?$DqNkR<MkDgoMJ6B});qDg~y#!>%-@Dh~9E^h#5f
z+A*F2C`60_WT-z852!1zPX(!G;h4?FU=UN4F)t!xRA2k-b^Tz9;@9PpD<gx#t3IRR
zhA)o|EbU%Q+9o1@#vd!zfhqmCY0lCE=YZ4j;rH$Vk9C;MzD#+P8TXd^nZ`B>bSS%&
zxfsD={R%K;vqha#5nI5iRn_E{J)r(R8^9!%JqtU@_BL@OSWtoZqDjCo`9mv5@cu%#
zGDc0mtu)ETjz;&nna#MPSnxf%I8am&yzm@<=FS0nA$4M{U~l)ihq=7T95T;#F+t!^
zdNS)EW(|`x3O+*OYUwm%BL$^Oq`Ml+B}7llz8mv8SQH^BdjEz~lpD9o9@s5=ZaF1G
zU7O%r3Dz({dU)`{Jp=_Dq9;T*xgyt?ot8u3W{({U9-HxCv}}l%Oz2#?+wGZx?iY{L
zjXMP4rgvI;KLH#TSNn#l_BppyHfs-pIORZd&#Z%A=dIgEE=lWmjq;&9e>~oB+21Px
zAEXQmVBWmJJn>uWQh4^Eo9pp=3AWyWiH7J={&k9f>_Yw*{mjVlV=eMu3c;6Z>rv|?
zh@RQix>?G=s+v!kuod`2$g(@S%kH&6AT{V<qIuMWNQEH@7q9Q!o`kCh4aSBSv~d8g
zM61bXXJuz)_GxN&l%sLoTAgWHp6;TInJ0>pRPn2fdW?vgMVRQLma!e#B?&bTZd8UW
zt)I8S*<IeB)8~|^s|TC1)z;e^9@jixEv?EPjPHk$*B$S}>+7rO_}g9WUye`D*4J#%
z=sQG{0RwN!Jh>AvZDSo0Gr=_^!#1N6H)fn(5=v<nr^b`c>z_p*cejdD>6^Wk0+6sr
zlo3;RPKo;6bmfD?{a5waoSK3)il=`sf~9`!&J<FO;_2u%=AT=m!6V3GwPYZPOP7G?
zn2l5?!@6qC*|51=MC<RTl}@#<oTcm6+nzV)xF_e1TwBOq3oN68&9yx3q=g&%7yRFE
z&ic?xx5+n5pGPDAsHKTx#2rvBE?78BE*vOUunv9>{gSILZ}kp5TOHv{BLU$=_TvbR
zU~CC0jF_d@?Gqn;rySKSUth9uO7vvcLC1csrAg~wbEb}7>`=Qdb0aXw;j=c*I?K=%
zrCJPVdAts5yOnR@#9OK~OB0k|9N*N_5bY42qh>-(B^|&4xbkFO7uaeqD(NaE3^<N2
zjGuHQz@E&sTwcEJc|`lB`b>Xh5Lt+6rKn&<ud<N$8evqgoaAC1dgIAxP+hji2Pj<i
zpG^akK+crF=%?B&9S(t7G!R;QsT!YquKbWJ;kyLeO2*DWUvJ-VRa>$$KImwQwZ?`<
zo;McZreiVaJwH>#J1ZGVl}J^T*_RWKmA%5BKvpi1Sgzq*y@W6b{>eg;t6!-|2Vd5>
z*3_w|!dl*25}mhe?q#`fWw<;SR{OnTq}G15__yAMTQXTJN28c<p6r>vc$W)C$9e8d
zl|`P5G@xPi00WY>fNar-Y=xExWGX=BGDkruL`-e@dv8L`mXWD6aZ;T9BGTR>3ObRe
zn=UyL62j#)-<un2nBzDN=J~GB9k05WYnTb+6u<Z-K!KUY{M;_tU(6cFV!`qESfB$I
z+xqXx@HOYPpH10Z152vZWcg`1C<FR6^~<9OixJYUQ=FHx=v&xBlEN_8<B=(L=9?(b
z?of?)Fn((HXSr+qcMsI;yE7@Wi>MMYQXouBp2dw!-*oBvHC*0Guu{tGjm=39Kb)d2
ztwvWH>p>6gE9~d3cWC@`b8pKt^sD|t)^l@@Mw1E15|1Jn+=#b)tKFxwunaPo!_mJ(
zC-1JEx3>1y`9V@os=#JO?G|H!7LZgoZ?-~+9ysMiTo>Vj3e9y=7nn+(uv^!x@LIx8
zvyvT3eQ`-5{iV~ais=GWV6R8|5x~}>gOVtQpExb-nY|2TctSb4fB&@5F?OqxFhT|L
zC))MeHYpx;jg)%1spyOelc)+PWH1n!l;${;oSe2s+Cq$p!~?jgCCP44yOW?!TuJrq
z0AXrH-Ea{Ch_&*Qk=38z8T?@lpu~b$R8GLRU8gJmlN&NEgT#apRVa&KZNQ!cI6b#<
z;2LWl&9%(q5H2=mx!=>_t<73~5>$&kCoFHN1`{LEKs+WLI<SeC!V?P@AW^_YD3Taw
zgPKMoD=dqqJ+!=7C%Gz#?6%2|5$5N43UD~=us71@q3A4XCj>RI9v+H|%?B`x?!!W(
zSPR;VVz+=y52{E`_&9=*`z~DO16f3JNQR%Tr=-Xq06Akpe__NaMK(ZUZ)5?tYZm0e
zSZG$n<?;>CgsOpQ(6Fm3@{^dg?u5fCJkQAF!hCr@>&(E96p;I?N8X(;U^ic)E*3s0
zDF>L-GG7$#=>V9Mpmw0M&=sBkFu4=GsE?w5%3co-lf05XOE?*nD<|m1$cL>?gAL2-
z#>2a;jLB6{el}R6-8UCUFIDR)fq_1ybl7eR+b`>92ti#0hg~{<+QFL-H*3J)g_n~%
z*Sk@^Twm_3(@2S)3hS6fK7!*D$Czqez4e_&3hjW!)r07heK}JZ>X?5<=IZ=d-+hij
zrDsSPfPIKWI5E_PbwL^2r=?GT*y+QGK`q4&`ZZ(FnR4|w#ty`UZ-MrP>;Z40MUtQg
zjfro#$**abXcq;W14D^#;T=i=Z*H2}eVfI5HjC{?sRZ&40yEgB`e8u}IYhd;4I9u5
zOLKSQ@r@)Vn9rA0;3vc&?H;d7Kl#K}1gT0PKO#=dN30>_hyJGW{Lmxfs_UPGw=x1a
z%94$?5}+T{0|x5}IV5TtV}F8h5E-!cH+wZr0OyD!LysVGjI3TMA`+PoGt+&;l9fhX
z7e4LcM$lPyw}wg1^|Foo$O_TJbA!36e^d7mIAkMF*unj^6?_-$V_KzZT!Qf#{-|GJ
zU_)tEH^_-eq$3kTKQ?Z7HhsS-EZ7|*|3DV%N%FSc<S)Z1%%UilJHIOz@-=`MJ9I=m
z;VC5plq;cn^~lxOP!k9ISd&TwB9FRXI12+CM9NVScwaC$AU7ZyYBGUGs%e4iX{Wo#
z8;XyTe^@ZG4Frka6M_`x@~5$yq~T<Et!+J{{4|jId3DD`g#}~jW_jm>dW-K9-q>7j
zo~t=BN-z^R0u<_dSb9n^ycb+<kr>z3gS{acsSg%ZBHSWn!<B*NV*e9w{MboC9*3;I
z{5LSykpoyf_``MiPiC~PrIoj`Pw@fK4}n`E?cOGsCt(YnTm-C!K9jePsa|0B$wjyK
zrmvqMY2(U5B7#2aTb(Z~$V%`2NT;y)1sJZ&T>%#%ilkREA1^k2d|A&Io#JgZ1D;c>
zadWBG6ofU{8}a2nx)#{*$Sb`(KrqauZ}Cm8pNz;bK=ei`TBp(-AFS7(!`>%^^X<tD
zvq6B4jw5Y!m+uu9z>j-x&*j7djuWBwV9K82lI9ON-7D79i}#7$<K+cxE|>Yw7Yge?
zUMLJK|NW4AP*u%(UF1KwPQId0dM7-JmyKq-^bC&UgIo_Jka`|95X#|A%%%7Csz1_T
zMM6I1n{NDufzwg?s&zN2%oS3`2ag*AHh1?PHN%M5|JTzHO$ak;kgysEydLadEZ+a}
zleYekKUz-+w&U&aYy0ZqqoD*S-Y@1<sL)|_Yn|@*`#bCHmp6~R7Z;~bMvrb!XV=Sg
z?Jm#zEn0@Y-cn!DpZ@!jjSv!93@Z*I4{?_)wNmeNm8!Z@oM|gzFy3>1U$`p+fm_>H
z=|o-m$-Vw*bCT+QQbmPn+wT1U+lkxJOXaPqqvBrJX{K^`b^o(yVvI}{dOJQT-1pKg
z3(eIv*wVBXqlOQw2adrI&k#bBFgFCAYXSkP#-~BoKY_#x%r-=8N+esfT@MM;OgBXY
zCMLaFloj8$&ZkQSQe@)fpf^&JoL;V9GPVQ%^H9=<uk9JypH#4t6ly+V`X)Woe|ADF
zkUln89ux!sZgfaZt(g{5?GigpX)R!kh<cVqXel}kBC-Srpiz?rzoy5v^%F%~<3;ph
z+mMO@FFW(q#C=VR>Nk=iK2|e3*Dk^$mhA-m@dFoRwADtg)c|1$@|lzjB!P5`s9S64
z2U@E2(zEc>32)PC$0{GYD9!?ye1Cf-Mz|(nW!sm5Jl24wQVTzWHK=~O7m5_$DjV`f
zpdf{C0+fsI0_It8mTU@^^?Zkzb@;^apW{QW4uK7}BRCXlqh5sDARwIVB{PM?XfGtv
z23B$Y43}dAG1$n|W0+k}W)>^pb6OLN<e~UL5XT;NQKUSlN1M#wRz|6UPKpFHftJ3Z
zeZ$7=CxzXw$3)e~HG&Il76Q|YMU{6IQ2XT#6}c=+q7w_xt6bkc43lVGxxetoc`-0)
z+-aFKtbHZkT!wya|NZ+r1GTrXJIrqk&7a+dmM@XB$^U=}zcq*3BZt9;v0<r8FLt#=
z0v*kxsPXzQ$1h){HGP4hNNc;DmD=a$!T92wuZS(T1`T)!vwHusV3Z4JVA!c|>db0g
z>IJ3>$m2=>@0*S30<D|}Y2ES03zG>};dWYFAT|Pk%Bgs=XHMXuYf5AeEtD-wYX{sQ
zx9;4==?d~VQ~zEl@kwVz33tu2t?B&@N4#}l6ynLl=9)09tC60ht5>3~gYcEO9x$nq
zc^nWE`nj|(2=Ggg5$zCwXB>o_{WCmrt%ma^v2>4hHLGREh!apL-&7oGF<vH1Al+9w
z=@)dJ0hf%RrwjXZwQOhk=NLpj{vgK@E|Gx%K9Hm=qypc=&bn~6F~q>P6IeHvR3xK|
zDbk=fa9jAuT)I6dQZY>S8$$v*gybz$4*|`dTrkZo?1NVT=k*t6qB-T>smS7QbOiCm
zo_IPyL>*8eB4H`~7!$mpV5OsGZ_7u@Sepc>8X11!w0M~nY5x|;zU~XK(f}VwEAb5~
zNJwC}{we(?2naE~VoE)*B@#{Cm|Y3>Ym7Wiw%2^Mm2!LxbcS?=FcQ<RQjtGFx#ykX
zHg^`tx@(d?^<rD3VhLItVlYgAtk}m~$mHcI`gT4m9BIr|YWQ?0`J<Ge(v-OaMKDls
z!8dbd64`%w+ypK%j|5lqF^Swgn~&pVlO%}BDG3<!-uB4b;@(}jh<uBbvn=}H3eTJ4
z*9;Om7&M8%VBW6}iMbP^wZ*)(c)lQ_{C{`c@J0~`W;pI;d(!!>R};PHK;D?&+W-jz
zi}EP(0uOF+z>S`<C4>R*e%LT5uV-sS1_Sva+(^&hmHc3I0)41JN;dSAt*J;byV?)c
z-Z@b9%p1`2d$jG9e$=W>haID=MxQiLU~+Rw^*a;<nSQ6$b6Q>(??l^2zcJwCHcgby
zXmhqezwL*0Uq>8u>3AUN%TRXH9_G>l%cr!I{Wklu3=^X7hnm~s@v_dBN=@FIY{NM3
zbyD9Qhqs`PT$O{ww@BarETpegG>oM$a#+~StXu^XnU<%=h`Ctrd?AoEIhw9-N4{~a
z`M@3S%Gg6F)JwECOZzm?q<w6C(+BonOQG?8!-V1Cif`?(va<D@@_Bn%Kj{a{nxL26
zwL5Y6Zjw3W)obj#K6F(glD}!Nyk6G(E0j2|KdTSfc(zPVN5=wLvRFpR2_$Y;rcwOy
zMfwYW@Ay^j?YB44Yz8Q-!X<ponMPdY)ONo#d>8H=V0cZdK3kyk!?pmh<*RM~>!r^2
zkBJ-u6YGCXzBm5FJ^YaGoyRICrsxd)*5WJi^<rom1*=4kzCc!p6ismuiG>Ss`;Xn*
zxj_k(!coY@+@X0Rsk;;D!{k_8Aj~gzc08Qy?S{;7`bZX4{;Qq@4mky6ie<V@5MMZf
z0pg5V6w;#E-Jf^oi-Y+4kSY!ikB@KqZvOkHi^<{l>6BfXUtXWCPmk-5*8ANiyxCe3
zyV%KLa)rIl2I;m&Xqj@@a05VpiB+o#)p!`rg)zX9@!}|!S}Zuwc2C$x?+$2XXH61O
z?KuyG<Q^tc_%!-f@`P*<rBsDnCT3<jG<gnBpVy84RM%a%gfd-OLG{u^YPK9Iv`L)V
zIkFh)hzQj=o$4sgX1#_MIXHNQP3KxGtPFs~vFK?clO|vy^pd!xaIXPZ*98w_1&3d-
zm)F(xG~eN6AHTG$ujzaCnY>+dxXau%l-ys=_^KM3lWvAiF^pv2#G+20mPt%9bk?{p
zT2P;qq(3w(RR<2hVd?57Bm00RF42qjtEk12{bG(kfIr=&>P>!kWT+<mwTRcR^&!g^
zO3a6oqb(QSvdR|3MUHU5mqivafY2WRNg)HSDi{)n@<!<-Uipj}7$$1#0U_;<m@i1A
z*aA*}(f}NLTI$*Sn*TDgVe+o7c0iqdc3IpjD{PkAf>mGQivfqW<Y-o#!qGzpZxnmr
zf0G-)_=@jatYHw3x+pv?{1k&0zF^o|!3doEy!iISo(cRLhb(r1&96(NcrgYX6&WpP
z7%dBNOpgv$l>lWnh6S(##dhbe(BEf==2I$-4UNqN9>8E}r*Go5mj(Gs@RIs<V|5=M
z8zP!iFZM!&(UaQ+Z&>Jb3(pU4UGTxrARnaGSPFPEowliT#utGvHzJo%^C^Ms_uij~
zxYrI#V|2A!9|Z^{J(;xMT7<aehV~zIn#Yb$2`2XyTTz(%`3f1;PMlVJCkD?zMN}5V
zluc}ByLr;tkXeFY`w0HBRMov>k)Gq!$IDftwX|eL5m)Nn@pVBI&it;E8j0m=jTQBl
zz=iq6hDs*)j#n*9WN=eh5+$=X<lQaq@J@`p4x#t^w6;?G{*S$23vm>Wv0#RE>%Lq8
z<74N%w1UgO>7zu_E4&8OsPUv8*XKP3#O@yJOf^~0IZVuye+OKpb8KV_*aIPrL)*KZ
z=n|50JUBBG)c$qfIb72$LF-VelX@@VcX%z4Ie#p(d9kl66mDvt0-r{K%5!>njk9nE
z0!<p$V)OigICdU?0DB}J?$6-Cb>Gr(O&P*2G3TiLs}lZIU8+89;A%>^kW-Am60?cm
z?+8{Q%y(nrn2>^zhaF~Ai*=d#%;8oT`CSNfz3WKY(AxbmfrSRnsIO#h3ytqH&L30@
z{EI_E=3i7ZrJU7<LPDD~0|4AXGJjFvnN^{OjqywjWHtAxTBNl@!vhgF-s;Xr4<@GN
z%NuC|IszK9pX&d(ljQ;wC^~v>R$4@~6Xt7!`{S80|I+4NfpBfLt+YUBszQzjMzC`#
zv-2i7Zz%s*W2wT6k7C|PDEH1_t6()->z+B2K2;C`O}2J?X>zqt33vWt#LIp9TYhOb
zA_-!Mh3~m5g|Wb7?r;~gwCy-2bJ2|ec;#P7Gtes4d%vNA<j+fa6HOwxD}aASCMGhn
z;ioL2Xurcy-tbYbTauR1-?|{r8~a`FwxS^?ZCVmdQ?jeEq=XPul}NYSI)^uDPEM$6
zxGoG!U**-?vZk!<ELpYPSK!kGVHk(M_Aq0{MoM^7+6_)PJ5;F=n1ii40Efnd-#bl{
z=)ni;&BtmZLe#5sy_V$!pvhRcHOkcgU5s8IkH4S;n)I)d&HfK@m4W?dh3UT(S36ZT
zqSuA~$7p2)JE{Tm=&TmsMKy;H(MrXNPYgwA%aEnk=7O}Tou)g>{%C^DX=yW1cyH96
z-JW(kV?Bk=3A*d$YjyeDM8~Umk;ufAYJTv|ICm~^mU4fMdTD*oIyC0=?)foV$BW#B
zhX|R<Kp9$hue9rR_s;zDW73WkR+d_$N@^_LTD!aB?ew*};pNWG%(2evo@hcxhud3F
zl1{tZEu1LpHV0({$voD#LNT{O38g76$;efLsQs(S?z&|+{RkjG`Wjx?nVMIdT7~jh
zT=QdxMav$9kzZMiP&5Pem^z+OeRnLgycuq)D#oog?EM#EDo`fZqSbuc^JTH1k5OKk
zazr`xU3)byrL*gtiC?EKj^i%0?{bd)APj^KzRx1|^5d{u>zoAn=GH91-GHsVpiZ3H
zWAGaEt;H6tKwe>D-qxItaxryCM~U0A467Xdo?Vd%S;_(CDP#){j%gzx`VNaZn|}fX
zOn{;kHs_(;k7uiZF|~q;35K}l^>~s9t7qrwSkaJL(ty&liH@ID)s^YiZ+J6urWU!<
z;y8pohJq8AmZR!NKQ2-8bNIO9lUQ%Wns}z@cK;c_Ll%t_6zb{Ur?#8^uw!=zRToce
zNaFm^O7_Q5%tGJi?LfX1R77jcR%|VoxHN!JUcK#TaBH5nnovF#lB{^u>-6!Xfh?3B
z;^FN_FW`X>jptWCrU!fMuf1A=z&pvLRhQ*eMu7HWjCl5uM|MN@z&=J&M}7e9npEi?
z6k}yARY2`c6|8y?BW12D11sF1mWTYS)tf~p8eG-A?EHy$O>c@zoSGv9RHg<FtqnGx
zlvEisASP7E`g^1T^hS53CC(Z5q7CtU))84*$P5EAhA~erPD5S+WRyB!hsO~$RILP5
z>RMqWhW@EGM@faLc#^#fz)>}Er2-T?zh>uc&gw}G2jYk(uV)C?tk0-$D4G%MNppyZ
z%fFJGW?pTyXBudnedmpJ7C3+GIi=D%N`OVDy~4jD6~c6Evbd&YMM^g4#Spf@6mJUR
zV`qO4s%Z~YYRE2IzDKGcnRay6+&BF?qg$J8wI7$J2YJGdv9Y8xMVvItkMcPe7oDIX
zBPE3oOwNy6i#0@TF@q?tna6<_R{BDH{tWDIwu1BBgG+%R{M6bP>0X^LF+G(W(v97@
zBW~q}S_{4hbj8pc67BOV+M2_<y4FvF?SUB2Wez4f><9fJOhLA|d~}+mH|GT8`R%Zp
zpx*+-s9A+U;=O5)3IdKu7K@QMv5hd1V<dqhb)PpjYM6(k1J;PeUK?dS)u5=B*KJDK
zD&PAXWs6d8Yj)K6v*m01&8UaPLJz@~o8W4nX8jeiO?KUJ4duH)--y*^EaXQ#8kN$R
z5VhKgy0_S~XF&0lFjDR_4nk9w^>RVc%Zd(SU_Ky2m@?N7$R$47ZFD2RG6<N}*UB|(
zVLxQ$Q=%LOy`-t;95uFbsvw5Js@{<|8QE+3y5M{Ukc74CETZHV;d(9de%<ek5iK<K
z#!7_j8=O;|usyPhEnU9%>_xY=mV@Il;00&ortap{`T(Kh>}4jDRg_@>0)jOg)#T1z
zBzS-|_dmKyWc&O?kb|S=QHt28xtmaMilQ+12?Ch2t+&|%=!-42<5jO-yTh-yAWW|C
zCQQ`6^N*swc#lwTc9{wG7VrRy#-qWbi{Z(#JrP;XMMbaOogxF!Y6Q8_P|j^dH5`iH
z2xcRajox{6*iBm?Wz0bt>z5<@Bnd|BzQpkW-aypj1jy$Fj43Ek73Kwk7}e%Gon(;q
z5|GzOG8Vw9J+8+r&kj!6B$JsihC_ZjD3QC2(^4y56{WXOJ%}Fu-G-H(!VoQ!>uvPr
z;d~nZGs4x%=BSvD&3NPI6IUM>C!Ri(MJ2Z;{;NN?d(vR&!#oLFGp->PWTA%a70nR<
z70hlt0yRt1yvSFRiLg(>j@$k=$P5m?9dYqEt`Q||HH4PbB0gGoC{;G|hI==-KDTcl
zgsT49!i?nsl<oA%2v(ea5o1MIG~9^Ic^>4iFgLV}qB4{*;RD%EK<7baF^Z%S54Sy)
zcfCGy2k-?Zg-Ub?!%dj7om)|b^}Fur;h;ZK?tZg5A$x)pV&2_h$1lJ@RQPtJlElK;
zgPH)6xP+WQsWaCk%JCLl-tyxXb3R*_@$R&1sJh%=Eh#3smoplglV@>#<pvdF8{@DW
zYRBaK`z<sE!u}$Em4R#e)`BYlwEl3W++y<f`e4<#un!B@{djU1-bE*X$o^T&*mVGF
zJUr_@gQzCv>=`-QI=w0KtpIV69+Pu<ocO!}T$Tk-Vvz2DPqY!?1JgZzD&=)hj#*4;
zG=2j%>vXO+^HPsG{*tgc`Md*MYV9UjDh}k6TdI9)!gi^FnVdU?{Cy9j1DWM(VW=R6
z?-_c}9!O}O^O??dI=z$LN({|%vx~3(uJ`_R%)WI4H*)uJ%GwXhRXcwQpSEjU!&Plu
ze&hd4+>6=D-kxpm5-WD_eJ47ce=}Cx|33Z=S8eT><wxgYoWvbOSKvyU?RVXo^!)m^
zy6wol4))LM{h#_G<A3dovZ*Tmqc5U#y{nv104^Cm9nSdgaqx#RLn5_O%%Fi0%d;q?
zI*>5Re0y$FYX;?!lsm@ekVghtZ*?4|H}B3)LPTq4ac5_9yYtszVj7Yp=D(T`9EjP3
z(xd0U+C8kR686m+Qjy^4bbP*Uo{#I0D^T6MJeIR2*FSr4yMNx@y}qY@-|wAG4t^}I
zv*MF^Pk6gM`PI|f(;}cDT{(3?DkYu3fZOt+^@(7Okk%i}r+G=B&WXG!7&&&!cn<^7
zg>kBSc6LdQNk2V$FeYL|cNc`$tpxBgfMyVCQ%IN|V~w#Rr7iri>ZayaqK1jp4NWtC
zcJtn?=p_DwLu>$oGeo>c9RWIb!`Th2eLY{dcA#Ow%_#E`JwHo_F04)VKj_0&*P+!*
zjf~2n1N8eM$Lb4BwTG@!1=u3QA30$}xxu`Ns1TV%fw)JwOz>EXRyQKjb1KgSAt}s(
zM0RPSB2@hjlUeHKVw_CN$SGJszdy2dQqAGaI$-5<wza7Uu<hxgNZQ04GIxbK%w<+}
z>FX;;4<MkWa$=Nt&~>uC>NZrEM0w5-ujR+hYAr5GhvAIB?7DTXi?YjM1|;tpPvmhL
zcvHlXg=(LYRU`T}TDt0>!7pyg%s;C;&p5MpC+H@RvcUNfiXzo8{zwfZdsj%nKL=NI
zJeJ=uFAWN|-EuyGDjVse$}UYEYH1Y&rSx*Q3;$H>BqgjZdBjfgn0FRi0<q-XBmMhg
z9Z#w8VCG5phjB}zQI%5O#N<jjVL0z*3RP@@(s~wW23TT|RZ|J(yr36T#&mj6wy~wr
zVMiLNB)YyhqM<zcRmf!d6EFcHAd@q8ofnKlOi$Gf<z749BAE%RBR-}OKCSdW1dx7q
z`Upb<>GoHaeA2i-ww-Cb)X9k6&7ju(pJZPq@2z!5Ec-o_LbF5qxFyVV{C<Lt)T0+Q
zH;Uo|!m@Af$2IzDG#l+!MrWq^!%A6+qes=as(PUXa1@(#%)~a;W%GlJGtfO^67mWG
zs>pItoOwOjXJ1?XmQc&2RyU!VZB#7IUX40b5~Pe3qYn3gwE(3Z(t(S6O(LFu;A}IH
zL=cl$gavuz%fK86l$525AR)2lgC61itH74!SL?BM+ol6YSr82v2gC{Vg=DAkh$BG;
zdWk*Q7+u5&N<=mJ^;hN0e4-F$Yx*MfP?5seI6^6`0BXre`t1Mu41u8F01$#Vz|-R3
zxZuPQ6pDMCoQ;sm0ovf<+ot-LgM9?(Cd2Ce`nAalyZR{EwpygeiX*nPPL!w6Wb>-H
zNn&&sotRFtEg&W~4S(889|&b27fo2tL=wtn?zG9|4xcQyM+IOLyll3^MUBFFWL|}M
z_5NNHIl%iH0ec}^kRUJii^Uzj8!Rjxi;c4=<p&A157U>^rrKf}Eap%3{mqbRo&hk@
zP>PE_3ZexfDzP?U%0%*PLd_AJpAb}B(H8hsLn^ios|m|kXSRZKOl*c6zym8LtA&UR
zG4;$Y{FG5=K~w3zq`%+9eBP)v<H#TeBp*3<?gX|*VUsm9N%Q-2$++aa4Nt6kyca3K
zc0hVTi@UCTF0(2O5v-K3;0Rg;yA3I9`8W1@y147l7cZ`sDZW`%uA{EaWy(glU(Jb=
zc4x9ihmX-DPoh}pI2iTJ3@0ZPgN7?u2**}d`(o2NnqvyIHh(tBuj&A;@@B<nt3SR8
z3jV-5W9PBGlQTge+=a?^qTst9N-XjifM_6L1?RKGGzN;x=J3|u26ncz(8#MlHkd_Q
zz72Z$Od2#$y4dYN=0G;)aZK0w2CwZlU3%etzEJFy8%$mrjgz$&Ff@qR4eyru);fKK
z+@xnaMzJ8m;Z;(<KTiUYk0pq5IjIcFwAJV|>(ld9%CEaz1syF8ZtwSrM4*?<L;jHJ
zs`y|SkHvsIlyHSaYYP1v&W!Y?@twP5wUQWFggz;VrN({yV&q{GNMJ}brh7aZy;ADZ
zB{ujr@AEPW{grqq91Iah2c?D;7SCitUTE5dE$qQ0${ji!`6wnr@D&0-_q8X@aVSRM
zoIVI;Z7R&E6N(&_S>sBKJA0vmH9LSt8=h!4A77|130tTyD4MGV5te9I5JRXh*$H8_
zkPaURoUL%^v`Kw^bw2@lUyF^Ivm+z>6SgrW)a4(y72}Vk;r}9i*ctzOce+YtN$YjC
z|B2B!p#VnhCluz6+9JAinq~1V?+3ahlC$JMQ%<TwGHU+r5`;Yz3yTL^!Gp0AIq2xJ
z4m#TE9PWHZ_claNz7{wwq>dCiFw&3}$$}yt5<E1*zAlF+Qu~PibbN(#;otyc)U$;!
zE#Q58nm&|vYE%aiu;n!l=M&!WcK<v*+F#r*_-N1^$jz?gVS9Z$G8Zm`U(W7UMMvKN
z{T&X0fnIWq_t4AVM<CT{8U6R%|L+*bHZ>6B2A#U5>)H8t+g?m^8)E@$<2XP;a?YyQ
zlX0fBoF6|`QPA?^SMI{L_4;F##b4I7F~{<5+d*m1^DcF>*Pwl%CsCO!SneaKP=~Xl
zDyHI@+b2r0baEBy-bUW$>U2<&uA;fqO4Ke}+9go3QB9{1)od+OZtx6o!cKrrNU17t
zv(}he?~!oE?q8XuO(jczf2AZGpE0w57>+okbWM6u%=55H-+oRKg|Iyrj}4DOAQ0F4
z*(d3bMDb0dd{Em6)Tw6u<%!IvGa?Wk$pZl%9w2Z)bpe%$MGtaOprHDcNUti5QIja@
zS~Xj9P|~|Q0*9Qk=$<a0{5a|NznX=<QPkdaHrkp=rejfA`+IKuy`b7jsw|Zs&qA1U
z;zpN?B=}iD0N$_6t5`=btQ8wp!Ltk(f+UHQ(x%p>$)q=mIC({Xmvm{Vu~yQg)#w%B
z*PuUFq3qSFW|Y9V!z3F8=-aO~_5}e*2l-;Z(RTR;)|8OK;OsS%0oUcnn`;Qf8|0Hc
zZ@6$wCH`Tc=j0SDl*;NcVsKy~hCcQdQgHe{2R56!p-f8$u4G}_e5(1ZziPS(o7p*T
zW^5_OvozV=&^~`u;`p1>I%*1N)4YAIq1&}C0*p%r2_T7zRD>ApU0uS2!8m8-Any4$
zKomhGGu-D=DjwS+M=c&|r8W4k2b#vE%hl7MP@xnt4qQR-P8QSiiH$IWq#HvFo=E7U
z2%}J&vV)BTyCL^6+$=psfD~^G;rUnL{*bycv(krH7Jns)Zbnym)WuI2Bds&;rnFz^
zGJdBfc@v(^*Xy(9VdfJ*Na3mQ8iC|3?Duc^okXGCZa2RHl#(KzE&%1Uy~!7E-`{pO
zM7l3&g=e+t!g2J4x*tI+IKI06fGqrhmC*&B0|>?#eE@MOx)5;v^Cm8SglsT_ID!}k
zB~u12X|iF$wil2e$i#{qTu{}$cMIM9Zi!kSAR8{Nfd4%0Z2$YTGq5xK*9+c*s*=(B
zEJ$6`)ib0aSf|D>MInQ<^pH+K)G=DR#n$ocC01=TJxMT<S6Nv)*;Hthl1**SF`#kC
zI)nPPXXW6d;lqCD-5vYrm$J{-ULQNcggEhiCwpEB6dEUEO#524-LyQ3sO_&F-_^FF
zx2G~TG~?01Z!5v+oMtYs7Th@LbPM0pggEE9r)zl4pD+C7eN5Jc!drf}keV)pdkjO2
z8odneY)695LSRa((Glf#(Ge=S-GRBNhh+0y*6E8Z@eOnf(CszwmoxJvJX;VdWIQ^+
z5oL9f9%;Ts=|b(-MFV>N4c)f1DzG>Cqq+G{AbGu=wAOjQ6=Y_m-vm{EKUTeBE!7my
zkuvZKAqv1#$9w6{A{Y>yofq>ly#w5J%vf$#nbq-bgfIp$bKsh~W~0K}Kh|P4z0))Q
zQh$%c-fD1`SD)&5Mz&2f+;Bm1Zse>?e~PVWh8sSD90;$&kgq!1JZm6!_082~b(Xlf
zLnkl|-P@OI-hOUqU&S8*v+burgwgop)h7{e`aUlRuK_-zh*p~>M!{vjW*oTmYopSz
z1LYLFFqeG>XJmVRXE$(7LDsqN5r*Q#gm0I-h0YtQ{4HU*4YrYh-ci7_bpbcXc-upG
z37TLh%GX%@$kBBFNJ1)t#E1V0ZXxOz0*$du=n_#PTLfo6%IR5vo8HD)4SvDx3->ZI
z3J}@%0;p5;j~lnoQ+|aBe<G_?h`Hibt#wdxcdQy-UzK~&RZZkDpEbCpb2BEPmD78L
z6xO1`FY7Zrc(SD8fg|R>7ex(_*8(ANG$l&x<5a161E%zaRt->1{b>Awoh8x^a5zHS
zxvL@BE2I@YS_%1VKK1W!fgA-q7u*G~HY-OBlE)4UalFf+?(t3wl64xB5pwooF3v##
z)p6nd2W1x7eYQFLF!DTuxERX|5vos%0inRQtR|UzSR*i1%Q#MS2jKL}BulfxjN@e<
zQ00J69E~}i{T0I^jJpagEzx61O;jrDZ}hr~97PNO&lt-g&B9v@F=`QKfMP1r|A(@7
zfDt9?+BMs@ZQHhO+d6IAw%w;~+qT_(+P3ZPIp58HGdG!&nQxLxrBX>%Qpw(Hui9(9
z&kIw@0*Z7~zS_S)wup!m0b5pN_a0E=ADo5xp@_|91>BY=L^of<Q%NPf5TvI-Df`Gd
zbSw>YgWA@p<;*$jmrMuxFd`vi+2udhhw{#ADqSB>&q?oh-PbL=bCMLdHtoq#;tMmm
z=I9AOlQD!6^McV>TT^0N_B8}+ct-Uwv}gz@7)2{)x1I)EA|`zOfQE#gO3B2FEQrQg
zV<7SUcHpP%+Gi?rgf14JV$3&=*RzyP{@0H%LoA8D2rPeE1s;iwTT@`nOwUT`@k8be
zsz0M5IJ_v2n1?O#lQ*G_l3u|I^j>46orp>Dt+XwOXpBNQiA^4hV_kPjv!tq+v)h1u
z5+|D=pGcws%1@Xe9*)ah6Rvvi50UhEQ`muFm<0FCMVd|OlOMtXVW_RE+jI{SGH2Zr
znPk~g-~x?%ysE+xL-k_!a^k{f;D3%Oraz??X-Wb%>4}y}b-XwiEn!w9K@FD1?jlRd
zN!e`A##Mc?8z4(9ziu1}IRIV!fr8mqJCRc|Pk%Qh!$lezU4!;)&RO_5Oks22)s9xF
zgdF0jBo3T^fTnOTq)!_3=<c6_k~r1{7^wn1JQQdP7#5fw2g7v^$uOUY6};>Wfm7E_
z&~syn2#u|y9?Y0{(xKayXMm7y0_M9)UkBm}Fu4%f*FUfsfR?QQ{vnsydz2}sZj(@b
zbM6AN2#Pa7_%Rj~8)jjL_(eWFgagV4nO^Sf3_Oh7Wh@bcI$tmMMjD7%sSnW~*Uq5j
z(z)0(4SnH6nX=89w3}Yv9!v%Jgvix)I8|}DD<5^tEJJmd)-T{k4U>q~Sl)Q$V)f94
zby6GWf>58wzUSKVLp^HtW;=Qk&KE`u$ai5QWuQ`znFCWj8@e(+=V@iZ%W8|YR9+Hr
z_weWwJZTpKX9>_gFm?3S=<}@=rEJr`fgG6FrGTk>&De|W0$034UX9RK4u3OnVE>Vd
z6?|l%iCJH(<l*B9dS6q~BMN;^AJUa2tfwoOt<{>G%{s)?j&t4vd9KiFRmDQ-U-bIC
z3A3ST>0P#Zq0=d5T)ezTq!hx2I3EW|k1x+!bVV3EUg<dABqczoI9>e8Z}Sd5kWR%<
z>j138m)KHk?<9~V;j2+$&w+e$ORd+Qxl5chd*aPXytk5Fd>qv}%JnBY`rCW;^}s4y
z^64!_!wGH*%u30U+_ZoZGpfW|{sTO+pA6K@NfUHn>TzdoZW2N(?^ffy8#UV08*IkJ
zVkcE&02wSYX)TpzjpF&WE{t#8R0bjK$3$TPyR5Nv>W~$YVrxpitD9(Gbs{@df02W%
zzj)iMh$QK@E)yx@7ZG?IgHJ!D9=h_yvK;{r?2^vpc_zef93!}}yWWM=pX<=O4Yt&!
zw@JAX71Z%XH}%ht@!r7oqmJ!V9}O4AmLq%uPw+pE+Am9%-0O~Sb9Kt${qKV@6rC7)
z`Umo;VmdZNxlf<+h9U%E_sT-<{wi5FghjcCKM&tqH}mgvzyGz4|0RC@KeBd4PR{?)
z=>J!smbkyjssEL=7cvfpQ7E2TX4$ScjR9lhLy-vP{n^w7Y&q0w=+4sLAq-0v5#4SZ
z27`UBV-lT-+VN&4#@Z5LdA}N`k0PJdBO#iQ{C&@&r6W!hW0HuJIc8BGxjiZ~w>CPA
zKdZ-(3E#pv9xlZJQQ!Id$M??F!Gkr+Nzz!<%qe5xb!$U&<{+xlEv@bQ+$fdwZvXs9
z|Mg>fW0G3O^X>SFyYL9o#Wg0nfMhm?4vX;@fpRdZN7EmFtL_zS-R78t@q_nuP3UL%
zDi`w<B26^BosTcgAqvDuMkj$pnWOD?c$Edpa!7~^B6~(ka$-q*lK4UURml?!WD-Lv
z76;86SLPC<E$a)b5Z{*OH~Rf&wa$=@1zWAXj;IMv!M$=2FX5EYr0qJV0wO$iDxITX
zO3chx6kOiz)JbjUM(gNo1R5cPnfu$KL;V|R5X0oK4}g(CK(4=?i+ocWI#n+n8P1U|
zuO!hy)hbA!8qJAvqmeK5T)$B(Z!jN*-mjl|a+F9U;rQBd!;s4dhvO`7&Q#QQNS9A^
zvKw$_j%41qKug#gj2e5{difM4(u$=G&+l_a+MG1iWjnLV-FvOcm9w0PG?|z<>IjGf
zO?`pSkcjRmQc0#YlgpyVNCZ*nc~)^4pf4Z#ToJJg9gUap)NwIcn@`FLp5#0D&6&|d
z_V~S3t8~{fZLYgZdmF5tV8iS9Q=dG0%Ri%1dIp_VpJqCSAxwQ7#U#xTBby8BHWP{*
zRMj8oEtF%Z5#G0W`BG88#T#UqE}Xllugw)uXc7A-uRM2lm31xsX0##sqd6(slq1h$
zDUc;R%SdAW_JvX3gf8T;2T7$D;rJ)#vzM)U*y6VgTdAtChmN_MUBk1O8P0(!XdOh3
z;{~>A?+ilv<vi82VM4R&T~YaQ0}QwPRRgiEJiE*e4US1mDV+=5)F34%>pL5lvt5<9
zRBP876tXvWLl_$zOi6NGFou3)Z9ILcb8*_&BQdE=y#@^<c;YXvf{1@-%eWsc44ci#
zCkauGTa41y%Bi%=xs*sijlrn+_1HB(ECEh!Vua?cp<H~UWww-Q;RCUTcPM~Bl?9Cf
zGWu3VrHH0pLRsu1B&7T1FQeU4w1KFg<gUKF-fQ6_E_5-|G5Sa-?iKXk2m{V=GQdMm
zdmhP`<rOdywO>-E+#?HKNA+EtVYuz>LMqMyf=wjYd~aucQ;N?p!7ZQ0jsu!d&62JW
zwBSI=kl_|E#;JE$1;ViXDWU4&u@EQD>W~%|M#{43W_D}Eo0CL?AE&s$T&1x8rMl(H
z66B@+QT!}+IUzis3cghjhPw`HW0uLZ3Yb+jYMk}eLEC!a2q}IsahK?2UN%(4^?_wR
zTvB$uB{hm>)+%e~J~#3h&{nd%QDUx018COGA#hq<?E{$z#gxEA1c31;ZF`*{lc}B7
zHHJEaZAN+u8B0z1CP6P@-ajGBHtSHy5hTL0Gv5f=0uxahAmLW#(RYTT`0k-2ki}|_
zbk3o>Hg%-m49a^xyEyoyJu68LubH3!;<>Cl3{PT&lw`88dz%OkF20?(7x*XaOST3h
z9g1(TgY{`uuiK6<bUm$`9QjICp$H`GIj0?a2%<?hhVoYb#CpHsm=MlT2{9qWksUs(
zUr7`N2B;sCU&Mp&YSYP7)NX4NwA(+7;d0}7Z3RRQ=doe9<RJM7Xlkq+W}3s7I8Xt-
z!5c*^D4alZmlr@*QC#e5erqjHnTa5vq&VX;8I{F50;72%K#~D-yk?v8OaCn+N#t(c
zA3-94!Ms>-2a+kBVVOlXWQ`G=JReK~T>RPu)@TwF{}O#JyHpocgRN8ky=Dt_@_Erg
z${VUBJa5m<KdUr;V>ludmkECfAZ#Y1{BHrHXJPKF!x`5lzHXs)97S6qx5@3A&DYhk
z*yo~_7PBolgrCT?sWYPEgj!lC-4`2vMywXBF~EjAhGukxFu(;dOuLa#vdT%pDgO>)
z;m3$fW{q-H;jP=lwAnc>yR7_=+*p0Q*AHT$w%XMaYza(7LS{$*;K5d&W%E6scAG!s
zn49JnFPR6wY#*}>GVfpq7#}{0aij6XKx~;~nW1+PjX+ubH%?T$Q41vu3GD3y?U=@>
zSbRWpKS-im^^ErvO?KA%?t1YLs>&y$K(il@LPn3&V}OySR2d@^NQzJp;{?Y^x#1~l
zaL+9(qd#t!W1X@q0qyJ{t_o#a9Zl1T^(LqsX`n>gs=$Ivx8ADwP<$YzK{R?QRH7*y
z{WUaca}XY%%MLN{f#LnchrGDCb|p%}7w-pQM5g6y*q%CAcU52kqW^0>M$4An>beMo
z^r3RtAH~RQjq4yRGkA8>*!zf$^ad#*^Te;TLrmw0=zi-Gv3B5@5!bPhaoG4^_5z13
z9YmdmuSYw*Rc18(j~v_C2RHRsf8^EGuWS$XIC)UqQ<FeQ$V(#OtSGxS3T7ZnR2)G6
z$Yf%H7*K#Yj~7&9g({3Aw8L}-bHVh#&R>5tP#Os%*DO(H4aMTnTTPK?DBYpdGjKBn
z8J2`!auPFA-wE|504d^;1Ngz{7T0XCRWT;~<)+H>o}fB{`yr#+{I5FGN)5_nm?9kH
zxh4$$BsV<nq%aebnb%GZC;0&&!Or&oD?woT7s-w7zZdW||7~*L{L7?#shulCdt&Mm
zC9(i)l~_Q5Q26tqKmZGKrPUI0Ev1e7i@Uc+7|unjUO8`=?-qgc!QomHYfF^n-TwLE
z_;T7y%*G^QSw}o;L*i6IQnpNG{+R3?2b*Zjs+uLH)>mH>PhUr`Tf4fw;I1@EmFQsl
zXq4aH$KlQ4;iA+nE3UVy$J*EHK`)X~vm8PdtIe}QhQaML8jRrKZrW>7taVjL=`R+d
z+CwOMS>o_KUCkwQDhp{etLn7H)zo34f{Q*1lqXpOINM@~C_fy`HWJEccj_3Rh@G0+
zSUhoEptN5!Qh2Mrgp999HiduRx7mnnCYGZQ5Di^+ode?ANm;JU*}yPTWX2GXJ?jI8
z;tz-S<OStTSKCIIe5*~p(;*~Z0wW@(dm-9&OqEw&Y>}b}Wo#U@{}=4c>uIP=xU73=
zk*=;T`%$yF2X7CHu`uUemHJh#%6u1qgQe8LAZP5C7$Te8S`8(2Hh*w;Q^i8*tiBn$
z+)1oulUXE^8H@}Eq6|^v5jCxW0ko8P@O^>|3kF~sMbc!ICG}h)15ep7;}yvm1&*qf
z=Mh3MA*?s~0K>5d0}&D<M9@7cL9#`La~RYitG{s|@>jGj19R9dq20BETJrGwuj8xg
zp`5$+Dt`~PFllOOV}xcCENjmlyaA_&eX<>wFga>4G9Nzi!y!`Gd6@d5Vc5Anl&fy<
zKGCfrRp60`%HLJmO?HH#b3>YJw$ATv5BA;U=P6jRd-fI=VCEb4z2qvTVYaD$(_EX$
zG1=UP2`HKN=<IO)LT;w0*^e4P3gDg)#4zIrVoBEyT1{cJ66>>*+C1R<_(eB-k4SD&
zUFnL#TO1cWqcOKCZl@@*ZKZS)xzOPngj`#t<l=J}<cF7trKQe0o+Au9skGj0kWu6~
zIrNUdynfd$htlhH*|I5Yn-AwOGqKNp+jb)uoP!?L?A`*y!GM<vg5y@L#O4J_Ti76S
zfh>|;9WLandkesC0DHE&4*&~>S{H%!6om>5o^g0>nvR04wwkgQHp+h$qI!-mLIrBo
zd4cQ0Jr}~n1viWMi@&q;Y@T#_bzzZ!H&&BVdunp39R-u(3!(#Z`GMt(j{0DmnI>>*
z2Yf;U()UaNWpF((@{=gM3R=Wfbe84n`*K9_dC_&=+YhNIt*%XzlmwUsOA^e3=U(fs
zX&S=LWNe!z;ao{B$R!$-lX@zoxeL~QbV!RR@V&wi`n-qUj)98U0?Umtoa$DI0BOy>
zoVnsl2pDaIP?AT>rGEqhz>L4s9N)K=Lg*0&W2jOw_aU^M70I0@HM{Xa6oyreyG1pR
zRe>e!`oU}iP$u%$@{c&fU-gW~eR!v_?Pz;LkV%ZqVB66S1U|paJcmSFJiF6Wj&ufp
z*iTi|lp*BvAdsbFT2H-x#BFbatRNVv^g=LGf@g1wSyli;QV<@*3g~s+Q&CC+wR<`b
ziof6mV3MyX<|&ixF$i?9l8IczQ|kGbvZB~Jdj9%we3vI_Yfjg@yS7rj#375A%v!-F
zw&oa8ds)W1?23)hguvO9;nb;Gfsr~@uM^Z25B{wif<cfH$)9hP%V0=rB>V`!1Fd2K
z$C~@9zJLr#3YoPf2<j!#$N*d(P)2fC^WMHllLxE-*{U1RWMY$wVf8M>N*ll5Qu`$g
z6tGZ;*Wi_c7JqZW^BNKb_=M!;o|}v|7{xSZ42@25Q^!EDGoq?5^y>d2P7ZPIG*@tn
z$~ZBFcQPSyYUAw$3=j_TCN-d8G3cvB1Y#5t&cfBKTR<9t=$)`~2)Qd_C1g!eaPr9E
zzTxDF{A4kg`#T=ml*b&&N}u9@(-~Is^svPS`7t%tNh(eE?=JM2z9!?<s~N5OJWUi1
z$;5CaQ<x0AXwCO?XO*4HN})1B<V|NoUAue~>2oBqRqsyrq5^*jAw8c{#;_}Vm8zgG
z%OVeVEI^}K9I`3-G#*l~?Y*#?zP!+A(;jfE#L&<DtzjYfozAUhp_^yDxHWpZb?-b#
z>>|n#8H6EAQ5jtumv|aqPAqW7kq8uP`)v#yLSUh9H(*hqJa!)<!w6==l3a!hDwqMb
z^>)GZhXFaN+N6CReW~I1?)Rj9b1%+2U+WIL#Wm5R4^6g|5XH@FGWD?*i`#^o`@`)_
zUZjl07!a@`x@XV>pUXC4benS0&np{-M{2j;+(JzrYknRsQQu_Bug{(3VzYV;?^eza
zaG>hxv45>|%>OX8<zW2pO>K`fHvR`U@jRb!+tBF7zfhr45lq$WL5<4PPg!Gx%}EyE
z>odB|S|8-Jr-pi$JneZrbu{JRIygo@Fm?jm^89A#ZSRZ(2q?@+AvvN<#FD^PK*eO@
zE$It_i2H@(*ZxAlFJzDiUV3KZ$O+YM)W3RB6kXXgcG7$_AAv$rkZ{})3Otc6Dlg^_
zA<7DtB&BSF$S3@Ir^+OcAf9R9nItcUz)@rd9#ATpv8?aRKzT2!G@~w0pJ<gq((d)%
z9qH5><Eol{Y;peE#6h*BHSBtiC__b*lZogS`v}5Xf#zb0aqxo&c9f^MXaFLu)Xf?y
z)=A$(C-rVCyQ2XRq+X*H&1Eav5jg>Y=19YFQ8F!@`M8gJL`bQSIC|@k>afRZQyevI
zih=ebb4hwDEUU3#C^kc%kOe5HwFg8tRQ$}Fw$fH|du36F)TWV>$z+*fM@%-X2;&l#
z1?+;=@2L_SyCO%41=FAlU%Oawpz(-BO>f4BjHyNM{j~H)`#4Z^yfE%;$pqcm#EpmP
zZnkvNM)}Jptk37)z^lmz?^kR&SLg2+a;}x_9g|U`GS8#Os*2Sv9!gWTOgo<NyzyZ6
zV#SO$cu}yle{tRVpg|g`R?Av5i~N$x%b`*vg#<U+E)FO5`P_tqLq=~Z=VkQ4=Fm{z
z<$2Eqo(c*#85ih_jDD)PX~XxQcUov0uQq-yZirJP%Xw~yFz}eyNP(X)22tTwu=2*<
z)We9SnZ1^*H?pt9C|`OQe#DvAzy0Olx6Cuuza^S_66{V9x8dSl{LWXVKQuHh<TfiA
z`-HtWc_-oTuN47q=Vny?lBC5gcC!Qbd`iuj@PchuCYw6eGewFlhE{JIa3$@@%IXz&
z#^t+vCzjCTXbtc9gjYWwbOlkjU7F6g+wtY`AEFugvRK`CKU0sdOsm)+8(&+K_Rg<^
zX_%xl6yYi@sM=(1XGnwg?(i|zbr<2!e?0fPBe7J1h2MybPKmB6LuzKBCGw$LYmqwm
z<Pn5cee`jieST8{mzP(QduxDJ@~hOXsW?ar#g<dj10ZynKUsR;qU-b-Oa5c_F*5z1
z+fe^+7Ex@R|F=aHE9-y1O{FPqhs}=gAD-|ZJqpLr>^%T6f1daj0ca3OIGY%p1rdrs
zFtt4nf%lsa*duYZ#O{j};zNU9V_##Kz7GWLS<H$OgyvTL!`Ymm1>1!I6`t^*sdUK-
zMI8#lvH<MR&2?@PWz}u*K?=G87J^amRFqADf}uOdP*QMWd}Rs@XF>kb<K%Rvg&2u#
zyxs7!NYPC=2DrwKtK+!ZhL)DLwx9)8PLT4Xol?${F8y9IpO|2jZlGY6LFlbDvjHWV
zL7`RwNS&U5_>!H;z<a$1R7Eo}41biIAbqM%8<EOP*yTW!u7n{IUCKcds}%CnHMXXz
z{+_;JG8way3e*3v$^Y6-o;Rm^4QzxWDJ7=%K7b2c0&;rdn*(%&q0$On^7g<`;I6d?
zRKGK257kb}f~(Zbj)S48bb(xCy&$HktFHy8Gr3R^aFf*35e-=?_g^D*qcX4{yz)VU
za--Xg8pwhE1-N3!3o;}9WU4UWy-%2)-w>YuV%U?~heF5_>EZn<Ec_=AdQtNtF*H-z
z0doEnUPqXe?S&L14o(U}flM7#?$?>qA&M2Ndc0HFxpYx9qAYi8V9}u>#C35Pj;W|~
zkVUJ`pKjrH!3tb80#0L*cq@2q3lvLGsyYtKEdY~@bY?*kLFI?$0oF(Yc)4Itl)TA6
zrE|^k_;=?qx~(#UM-PEFPgH8{>};CU*;RA&y+ab3+|Fi5k1r%gSNwyUh7cVZ`)gC^
zUxkh#kF4rNCnZOGHM+i!7WLi^J-#niE{Jl2_rr}RhHsuvM?_tky9vL1zjk`5dE7r<
ztm?g?sMXBTb9s_FSBCG8&KKVw-a3~{WnEu-cGPb8yL=kHk>qrJ1Yy)u*cBe1aS8#a
zuy>0?&G(wn%WXGq{5~<ezzC&h`TRcJZAM@L2Swpe(<S;YMF&~AqmFs}=4IUIJ<$&(
zSB*rgu|WrL=PMUE7*aG4?OShs#QU}ojfC}i1p4lN`yT)U#crTnZ}?MM(Xv$|`&Z3Z
zZc-pjMuHZfe%JfTT(cj9ZKD4k^ReXKH1@+k@dZGteMU4j!WmN`O4{VlwhQlo@9yDT
z5(-1qdB6*_;p=CWC2$82GPJ~!&hanAfgCMYB$!dEuaf~Fi{Po;#b<`BxPC8N0rn<Q
z_9j90X$nD(;(&?)E<e_I@~Zt(axF*Cl<&R;IR__D$W&&r<|q^xA*+x1xDM2OzAPyK
zXW%krPhs;-e7iQ1e;N9Xhv3CU`s_^otQ7(3w&aj^_@@eo{Pk1v;bOxyN-(Nihkr1p
zW@}RxB(2DCi^^v&jaKL3{HYvQ4VC*gyT+?PyHxCDWP)%~xj=hE4YN*SWh9^se|~2+
z8dzGW=kNQL!@UloyD}XDx+dovzgC~Eo2zGD^s^e4TefIf<bl-q;paKxV+v5-Twm6~
z%iUT^w=t(rwo8_{H`axeBR3DVuVDrko?KvN&28nE>wzm`Wy`r*noHG|!XIVh7B=2|
z2M=_PDXvDcd%rR#vv<10Gq3}@^bU{`;bGgaI`;<2N_$y0Nu}Tjq)HKr7wPsl1dfrE
zJD*5CpYR;To3X_<Zp&@~^3zPy=5<OK&+(PHZtvFCuo`o0w|P2wzlI-*4b*{jA+yFd
zn5yL)G<x1YS4Hi9-uEtZhwsPVKflhqu;aIB{PeyXygf6$25o%9mX-1?#?Y)#FqgBa
zhm#!%2J;zFck$c?^h1v1FIF!A^uKN)iTSH$iFn=$on8n+?FaUdHxZp^D7>warr*Pm
zmPmLD2vI@$;3nqzL;T@|KoH5(#YdTghF(6<>?6ZNgSdXhp4)tC9UNdPNn0xgHJ1&1
zb*+6e!_YNVyzCd6d(<FXjQ|{q;|B|509C>ESApm1+HDn<^c-jDDp<VemirxN=qj96
z#OnAGp6O@c)p;{ah#Pmi&a1AZ7PZ;@nQu(R78fGfeFr&dQ!T?EZ5(JEOaf6$M^`K^
z`vSIU^1e~YhrDSI%j;}er6rJ#A>snmu<!5siLigaxZRU{Z{f+#QIzS#96s~Kd}8Iz
zCU-XlSU*Vl0i1s>QvSz6#KinB6(j@8f31SN%8>n!1xM)3H!6=5^E?=|!JUIMx2(D6
zbM=|KN|t0Y;<%d9;p2`QqymzlG*IE#bkS7=N{G)3XdQfNs_0!0!tg9R|K?VgZ=vd5
zKelCLr(1|4M4*h|y8oQYCNEmAoI?4Tom=l_INjwF_F*j!IOZ1EbTXO33Oci8zXx)m
zZRvzu%fqna_eVqa3`xYery{>|D^jwBs3x@Pl2Ual0I4dYF!+KoO2u*3^1c%ejp8#H
zHSCJ}Gs=if=EwJ(iaPh)sjKh~6w=1M;bK(@Gx3ng2bCWB6WO-PAUJE70g4=2bQ-ia
zctM{KrkrsAORdzfrmyw6O}i0bxh0bwz1e~(?>+)IsxQ=lB31Krkr-v)BEe6Iq>@_{
zG9E828~#k@?a)W=geoSylEiJe<(yTd?E+7AVTi|<J6+`xS`p4v!Avn}%ef8K;~WjY
z>JyjTb2WlvI}Ltem?&=U2w|huH{<o_wrFQes{bT|1gjxb+>w>l6A0Ib<rhBXy3P}E
zOx1T-$-A+pZx6q0faz&)XMAAJ=j0p#EQJd$*1J-n<teysKxogV1=hP@+Wee4hO99E
z=`TwGSfL&g*f%`_{U-FP%+d=)0(k0D7{F_|L|_255UN9Ib%YmU0R=D~I}4(A-Wz{*
zvc9At5I8uvXB>@;BLOD$NHEW34|pXm0!EEUJn-ry7;v?K)O4z)jH*vOVYOmLa2O6t
zc;Me2eg3(w7%-3uB>jf$JN^<BoAkkgzPZ(Vp7yupm!45nD7o>oy9jsTXrAh%cY_c@
zZ41Fp<=4+$uqq(yFkK)*aBR=P;ohaYj5>Bq%at=;yR^VpK3I#y!L)}Egs@O96$<7J
zSRe=Ba`1l(@rrdJ0^uAx6m3ia?qomva`n`qEmL+61Tgrw=UZXL*rMS}NLB7)5N8UR
zMP#Kp_}jS%U41I&Y`~zNb%zV-0K<(_dy3yrkZ-yqDU%EV+rV5|Bai8M-m7yw4$SH9
zv5;49>G&qR4E@SijQwap$W8oZW*;kgp3D3uB7eOPViSftjg{@QH?JdY^%=#hkiF19
zr09M=S@dg&DTwseBW>95CzcnDbbBIaUk~)}u6^3qV7aHHoEwoS&n$a?Rlaa63|%Mx
zn8Cu&Kc|tvxAL#t(S>zwU1DPSzNE(5nQ){o`rwH2Vb%Cu)Cl?#u_k_!9AAvfABrXB
z4(W*a7BKwh#>62prhVu)`A_E;>}{fLjTb1vt34BWeu*M~7z7<+j+6~^d4hUkm3$CV
z#Tj%tTHaNkNDA{FE(I^c{E1fFp4J_Egiwy;MqtdRxFC`<k~J4B8z|T(V;ye*;aBeN
zyH&64gkMXcKHPgkY>k}r;}^e#>uT<`?KhY;DrM)t0u9SQ*yxOm|C@mP|6O>!%Fwmj
zU_%MLd7yTt1ZJHaq)$j&OSQ$VYd-a6%%cUORnvw|ZCa~Y{Qkg=XqYafOfaHe^S_d6
z7<#xr8g-JG;CaWmJN<~VLqng{r@-Vqj%JwO7bcM{N-@+qv#5{W9wAu1{MsKqq4)31
z!{8V!nx|Muet!;|T#c$^Z)~6AakXY!4ISn0{(L&;{yAN~_)p`%$Z%9zt?zd+TGwuM
zh}>5;XXPzuT;3f}YJlT>LD{eJ3Eh_Ml|(AJi>O$)wdoeERy08|w3KLTnieX6Y*HP?
z@|?J~MytL7E(aD1?SzUb=gyhlSMk=j=f!R^9CNC(bc#?Jj>)3)bdA}*k>#tBW+l3t
zN;L`tq#cA1$2_fk9N=G;bF$6vI>|QGHyr4dsx6OS@oC?d7uI@3y$9hewYN}odS=rA
zxH`)R6ReHyHcb!kJ~nk6R^z^bhJQ;#FE`t)nWN)tsZ-yd*)?pw>kfV%qpY<pxeZS?
zJ>$RM%wHcWdq2``UWQ|<_iZZOH*Sp!-4`je*`ZtlnpYCiMmj2Rr&&9a=rWDrp~$=H
zCo^RBbu()hcGNP|*AI_my{OX<eY*QHP#REy=$B?XOX>g+QBosGkLQp#&ber$TvAM-
zE_*DtgfGsYJBlrBIpW-AKUwKpze`T}C_3Tc3RPZYW+ea*bnwXo7D~=a;5H{%gZDtW
zx%x4W{GthTp+a*+P27Ol1RTJ+4vGQ4EDj*J@IiF_hY+R9^cW3<APfSbu~7W=v@rp;
zC`j5CpB>BAMHnh+VYvcho53V@Yk84S9^uxcD9wVOjlfm3N<0PAIwaYRC0l=}7ODkf
zaX_8Jo#%v$qI~pG{Hhx0SKW~H=?6=Jde!;iKRsVh{DW*AOpXaMpF(2~+pGzyMZ!GK
ztp!X*&47_u1lb5YjT-?WxhSy_Lh9H2L2|NSO@RL7-}<p}m~WxSbzGNTxV%4IdC5O=
zNvnU;;m!cL*~bj-n=pWaIEH`RT%+l|`ZR@CGG);}5K*&xsIDwS!#W(C>vcI~2D+;7
z;no4enRkI;9kxD!bBp(Y-tz{z12_b=X;f&y{6UZiS;EEM8_|P7qytpf+$wLu!GfWw
z5-pt!jOiwz2*(Pco7rT%Ac(#sE#f<WLSA65oz<t?I|JaH`W}BpDDl0u@J4v6_m0n#
zd?ko9ej^YhzY|0mxfcw?8w%@NGrkmY=lQZ_lFZcI1W9jlk&TAp7)AQ9DEs?N+23ar
z2qTRZ2?fdXg-}KcgaYw`q5cGyA6lv>`n@Aho@rNy`C6U@5?{<_+|Tzx9*GAbgUJHR
zm2qHYGn}&63x&+@1`7`EE+LU&$Q!25Jqj!G^Sxv@#o~|Qwgg#iyIn(d3k+mj56L+p
z1$`%Jut)6zZcr|A>Tml9e9+tQ&rV^AF>gQ2PKr<xtKerb4x~#WiBWK{=?62C7e^@=
zShqkJgCz~`)i!@USvlOH?~V*I{WD_ys}jWU*W=;8R2y&o$J6I;tH_7?t;Dnh1R6XL
z2U||b;%^seX1w(RTr-LWRLjYk+M7UVDG*|5DKS@k@<8az>(SxXC{Pc_+}p#|%9qJX
zH~eH79H)%2#3NlIT85fNk;-U?;-n+0xAefJM)}E|T6YC$4kx-nLV1FOHVhizqr@XD
zbUQPHMYB$g6aCC-YeE+JEWH-0fnOdc)!iQe1jtdpna_BI>%ez4N{q8wWnE;SJ%=Ig
zFrYDJ9z|MTF9>V3?x)V~z*oD@_|UU#*3g_1_b!-YuA^)<MnqLnseiyMHIq>4d8#gk
z;b!34;6YZW_8Z;0KCGO<X}p<2n|@4=^65E}gvg+Ssk8Gmepr&oe5NL;dds2i|D-<I
zU_gPrE8LER`onCoUGP`#LmsTyOpq%f5sV;$_cw`Le!bU3q>eoQCot44jneN}-=D<g
znnguF^rLqlI(eqLhzC8XD8zyA1wxP6qo)i97$7a(0vay*eCqviZ;$I%iZm5czVx#_
zA2nLMDU@*XWO1q2NVzDmoh+H5USIxZj$N3Sif)gZJ|W-rAEZe{pmvc3AqVNix0}Qy
z0FK*3$S}dkAKQNifid)#<PAW)Z-g~CUuLXr!%UAxZhb1BMi`@Jgv2bCz2%yr*<+q;
zG*D7We$?@gLQ5Gtc0)^%c%hAo6uO*QTcx2&%+*xUqa#ydOlp@#Ns?{s_NB4q?w$Ys
zs`tqUFo=yAAW69FGtB&%c!Vsy4ETI;^7(!lJdRxftx1UYZ(><AzcrW(Hkh)sjpDvR
zh0YwzA@|Vsr5$;;q`KW&NA$JHs3${72beS$muJSLDaiBH*g?1;T@zX3D1TYVBl>ER
zj(W~NpLqW*a&UhTC2(0tb?#Lq9_AfykB=V6u~#+pEQpB-wR0lS#OsJx1vh0UuZ%fU
z*ieVgjUV&Za1Hu*+$z6>%^u9+$ftQZZIu~O(hs~9G;y7}!w*2Syt?Hbn5rjV^4S1#
zwcV>8Sv|(EEJc^*7-d8ZoFI$lx(#;Xkj_ZPqD`heqKy&#n}Y&Srwg;_3Md38=e7f+
z=tjdy47gOd%lh$!zYlmj?E8|2=oFxUUjV0RH`Q+-)6wZ@#j~T>C%?gRzFX7PuUv8T
zF_1>7E+$<fdh-R+g?0|zeF6(>-jJ_Q#+evQ;&?{rl(WZH?##)HlR63r2mm;u)~4M*
zIl|}P-w6F2sHCJ`6GL2`55PXYPo<E*G2J^P5WK9LM${cWcT!bF3$yBoe#k_`nDUME
zg%gG*kqB0+?bgK6v7)4_V^zK7eFNj+;#lm#d18=pNXSrO(i(q5+5MihTUeQb-~Kpr
zP;=D|bEEv280U3$G_#FU)w?6<qutff-6eRfnK69%v337$=z7N%^8l9qXR!H)_L1%H
z&)I(+Y|^#uvKanS;~yy9gTZxn7@mz=inYmYs+I7_TvUgUT8;Oa5HnqOe8GglQ-~$7
z7tG}OfKL4A_;Z0l{-z~Od~(~MNlD65G=Lz3Q&6@v12l0=0OjVO)E5(C!JMo8CrQd6
zqQ9__U=WTe6RcK0<QJn1*xYn31+@W2pD>zwvwgCB=F(>9%^u1Rl({*$e>^pz<P2~F
zmBfbjr7{wf)@+J&FM;XOUF#Aii6&i0F!ok_g#u<pkR=jm*R?wrl7S=L8=trz4Uo`0
z9(UVdgt!&a<G(daiRj1y9P~j$0Tn&D_Gq@oy5`nU`HR`CID>x80NiUHa51(ij=%pj
z%z5VPrD}So6|L7aR$fxr?6@V}sym5Shi_14q_Kj#2tZk)B2+gZsi<m}RdMtI{pwW?
zdwnL0-oF*sDQ${fFtsrPkS!M!U?_KqU(qxhztP~MX;FSq?TF`3(jw*$#~Xk#F#YBI
z*gF&!x<}mod0V!%LZW}a4s+guwW@>W=J6h!nqL)4fp8BH9B0ACds<*bdw}gUjz|V@
zqjA@|%n$^Cfqi_1*!RKpV6nuUYyJ{tp6m6HUJ)wnx})tT0VrHmdu{Q$w9XYB50zQS
z+X}{K@eAg7$p+qW=x=>IXn+U)M@RP;O*@RSO$fbR5lzZ9m6S<GY(vvhb%j<3v(bHe
zL)O>;dM~hb*bt9D50e}>E|5|(+5wXi++p?9hd*T5Kv5V08Hd3f|Abo*T>H8@d^fIM
zdCV(b_v5be7nmr1k59BHjOwT$(}IpQuNKVhl1hKqmaBZL+f&BG?^kZW!JcpR_%p*-
zsa(OFAN))gG?{Opt#)PKe+_5We~6>l8UBA5UH=`<iwr3KJ0H}a=~6=vU-uB<gm>ZZ
zY;!2<*3ec+$!1rtFWKw1$uPCkEj8Br<FS`cUo#F@NX{`3?>uh|S=l+zu{14_lq{V>
zn+h^TLP``e!6J>I9TF6)i#Lt*$VfvoGn7YT{_*?T`N_~{MM@?0^%K(57x$}7y`9$Q
z(??vntuBqKj;=El7vow{oQruY<4sTc8(}3VZ4;eWJK-eU%%&@B!&@w>!6c+K3iN#8
zbj1X-$^MXoI4In-fBCUF>{l$Ni6mFD*W`lwL*RuB%T_1CnDxM#ry++=roI*>OC(Pt
ze!(1DU1gsfwP~%UDXD9-TT)jhXdcYVME(851_wMybad1V>SzN@>6*u|y}6D#RgFd&
zUmBt5YND+0%Hg>~fyDkn;G#>^>+lLxQ_((PeY!|3yDRX4#s*-=c-eozXUOg2cf;|M
z5ow%H#@^ph@Pmaj`laBVJP~FH*)DTg);QFH$UfjTx)a6Dul$6VeWo`yN;*c#Q#-@y
z74%i2kCmG$y*-`;z9U9-c<u?XGro~uCfsrAeLX@rVprfUNv`5qWuHIxJLP?o%%%A!
zM_h)tI|X%*&I%|VtW=kl)#|xzTb>I>(SwaTKZKqWOVbk5rM{W2oTPL_m-!-7-9MVP
zkB1EeX#ent*nU0{NYykp;1_~M58>@g!~rHE0!Y3>Z((^y?mWRaKuAv48I*`GMCdk%
zPDKXeL)g6IjCTFrH60TOC5UVjN`|^zN&)uP0`u0o@(`NO3u96!xS^ze*(k|2B)%Ev
z@HXInNZpWq#Z!Z?dJS~=l~vFUxzRn!q-kn;+l{OPI-D|Ta`MIo3_<~QflE;#>Z`7@
zT9|8%xG;MDWjK8KPO8K*Vo?5=6m`rfe`9zj5gTGMTAtHWlTf^f&Z&Cj{PX*AM80tx
zpQ*|1_Mda&9}*%q4%YuV#9wLX{wKuy%+~HA;@SJySYM7>OSj6cYc=!8oL7XCHW@=H
z%(KY;d|6v7CC;acl-X49ItY8uesP>&0lU`VTzcbqhl7EK0lJVRX4{i@I8(T3^dZ8f
zBg^JEp_kp#71V-<`I{PJO2kr_LltQ~-A&$-VC8u78U9Hbd%zsMZY{(wls{g%8$WeE
z4GF-;#r5M~^ar+9{u6$)I>rZ?5Wl%w`o^MJ@|Ge<Ed?LUEh`lhZtY2Bn6KcjkZ&LY
zscHa(2nl9<IzOj7bVl>dDh)Fyn<-x<i7<=?SteM3RSrr)H~ISgIq<+aVw*9<d?4=6
z^66gYNpb!?73FZm8fvMHK!zAxIn2*E7-hnwZmLFzb`FNo8l#378{Uu-Kmn2(Nu*D5
zw5h`SV>6kB(zrpx-ZE@E4h$hCh_MnuqZLS{G!Rw+V_<DLj_j9dCh41m)_xxOCZBc%
z*hee^h!`H$(qBZZ+=;+S*>IuRV0|E^05V<CayUU0*Xe+UKz#EuplP=#s%P6Vv2WWo
z7(8G4r{cM2|5~%d-k;<4<TOz4O=%6<9i4e4r~TTgg?{Slpvd&DOD2s>WM$B4YQK31
zO=noIKIy3iR?RGvz14(2>c;}ml_zRUfKt&`avz&3YN_!hcM_Iv91o917A9DMS;llY
zP5}x6Im91#jvgF=Y1tF>H>*dn2$o*6q32dhkQRoIi6PEP{eG!1a6&jl!*wAv6`jq}
z=QalQ^^^SGw%*1gloP1Va=R-%oAi(3j4*3Td2*qnPD*}fKc$gOruC(hiL~g|eMv*q
z?*;E|DsqAB7jkl3Q*71%&yxG!OG`uN6(uFJ7nP9Iwn6NrWcLN+-Lj&~2`vh!g%k!w
z=!&fQTviol_f&~mK3e6qg%-EywORp7{~G>m|7bX1Wcwd>(|_TMqZ-={H^2ZB{K6M9
zY7%}iB$*japh$9T9iBfI7zvdU2~sQngOTptaNWZlh>dXBXftuYz$6mA?Jl9do0pIc
zZqA$N5^u!;HLdu2fv}JjEV4SwW&#DHK;JrON$?T>sqFfTAjN^?{Y{-ch^DF!GBM3_
zD71cVqg(;8!BBxKFTxYhh6H_Iw^`TaIurEKDmP6P%j7-lp|dmD2QXiYG2ovc{eRYZ
zWlU|&oh=9$ng4g2#sBxN*R;lX?0+DGFMmbC##+|b(1Cn1sc`X)CoISyR@z%=MB1@5
zFZVB$2}?qvsaJyJV-bm`6`oZ`s`DJ4vG$J_uz1z{bavNxFcING$q;z7xkO~dHLyrx
zqLVI<tYjFpzWU!Q5hD7?ft)dZRIMNPZQArmRQ8|k6BpBLl@BRi@|`h*gnrvLjk;SF
z?OwYyOn<q%*xDGF(*o47;my<}iCCAm{nrZoaf<@{c2A8QS5F%;!Z|CN51#fnm5v|j
zHw|hN&@oLZA5~HpU7M^JfYNc{PD3b8;T_CRY<62=R)WsiS5eL>H^0B^HW&KuYDbq#
zmJ!{Xb#L$>3=_mO*uMQDITo_TB$4qcbp*E2B8E76+@k!bUu=DT3<sto04x;>5JNPy
zNo#ijz>Bs`t_`(2jN906H~4gyzohXmN*mVoK(@1N>Lvla+Iw>YpfwGh@quss3P(eC
z0U3b6kj(AQT=h*_<u)@|l^Vo)Vsa5Bn6Sp=->;Lj$2e(35KPJ-P;3|7z)u;Se&s@C
zxTU!~!N^R&76{4geLDLQvQ&^V<@uBW(;|skbZyc|WaHy>tTwza^7A#<$V3hBF5P@w
z?ot65azXfmC~y9%W~U?pEd+A(=%uIrSx`qw200*_ajV>KYdM;`+@NVxs%(;={;6km
zUtm=tj`av9$Sq0)G^yZ3@+M);qFgc{;6aj7(bSs426`b9s;L{y1Bp<`7g5#`eRds5
zMvy>eY%$f{dP8%RAI5v)2-Vpzx|P&c=PbfVdiREj7P6mF=$CG(AoMXqM7GM74-0JE
zCX^v65Z9=@e~OEm5}akqjGHE%!jRr{ng<U>QjvYP;G55LWKAWv;R4?{l~^2J^ZgQs
zIDP=L_ID!i2~6g^!XVupg)*VNkUKx#rCazeQ02I=Ax<y@F4ir~os-MAW8ziwTPP>o
zw%@e}hEk|SSaM~qAW;HS17ovbFOHAk1Z`u^R$h4GYK`H`R$6GM1SN_EPslU#l-et?
zd?x{e?rUQfGic2L=i&<;l|A+xl~&`l(KXjgN!j4KoXj~;`Blw3X<=dcYG-R<DT@!?
z`*eEMWmbdQhL&4BuVUiBA<>oGE)M`jgGd5!y)ZegZVLKkEmUZLFhi%_V;GRu!@tw<
z6orpw5cOw}OiFOiQb35%&YeNKiTB}asO#*jh(Re?tcj7%NArjeWlewdla!WAeffyT
z22&&&Qgs>@%RC;jIzHG4;&IK4Za=rt4r5VwMc#<7&1QF|Rb@IpSdnS2CtDn(Urgsh
z;TWczsZ8Q8kYOy%gIU%M#64d|9j$SMgQ2bDKp0lI4n~L}1O!K>DwUveJ)1)nTZe<^
z`8xTpMLRU~;X`2gE9rF9WBsj*tN-2@{{TR~?FZq;Gx$r=iLMsD&zjrjG8o@Z^FGXM
zVllW$d7_JoBz^wG;9Y&qE|fj5b|N4q;u!n<^9Z5#IUzc`-PQ@AqQkNSiFFL;g4^n`
zVIr}MSGdVkf(T;U-MdT%|9)5<=@(p!%r|O_37|HadOhhLMp=Xph&EUS*&0)OBrUKH
z{!D7XGxQo$e0wx4aKit(h@=${%QTog>@}8t_cNaKtuslhJ9^ewdg~ceKXXGk-Yo!2
z8A--jEbeA<hi2ew5zCz25;d04lyU7^vkvLvY)he-&9dstA}&Fh{^7hTJ{=I&6V2#0
zjwrB41TNSIsTtq?;elDBpUmn76of*f_*cYbXZVM}i-G08W&}>BPC90DpoG5oLZxn%
z6F7@&O$|gOC6`jURz#a8O$B1?E9B>w+PQA?fd{XLn_1Yq&`}4^z0J~}oeKZye5rf6
zzEpKa>mt?g!m^YsM)$c(e*{VU>%b}+@!}xHuy1h<9bsNuyW`i^)7_E&p|ZO+e8iu|
zp0=ufe6{JCyR)g&)9LwhOqKbA{e|0gR7E=jIv^xuTR1JrdJ4nt+6Qg;J8&;ib<l@<
zyOdB#E3-^DgH$XFX&p<Jb6`j+ZPFRlZelsAcS=EBg8{|LbkuhMNZd)^h{wy@@^|a;
z9WLK88ZSQxxP}K`et3Wy?yH5EARKdt`3T0cN1b>VLSej;yg>j;m17e?h_#Ix?U!BF
z)jkg`+p&-G^BP$Uk|7piwQl){%)zi0_EB^hf+SLhXWdzJ_?qZDq_I{}1mknDPKH?a
zNzC%iq%-efDphC^N{uJWWw=#fo06@J4vDabhHAwEY&)*pV7QRMY>dZiX3eq4^g_Q0
zkBHFxb<w&u0;l%e2?$#H5>-h>TRD|BGJLtNk;bEFX5I*V_S8CjE$LsU>;M265oKa~
zKJz_mbaRhSPWhMtF6!${GEow5%S^W;qm0rx*D)VNyB4$am=F%h%MtSsSo1a~3sw~b
zK`jWY)K)f;zb>pTm<w?CeD5HG6<uK9M=chPBx`iq=B7!MEn9aT_bp{`qkC9LPNx^V
z33%o}kQ|s#u1!PjV0WAuS5NuGE&9)Y1<uADcoWO5E{jb^b<&j!Csr+QnVIr#M-`ei
zrkW(04t6~^rHTQCHgGG!RCg*<TWc9M^sOEbXHHx?kaDK`B@koG6quhVQjTGX{UlOM
zvI#1s{wCVxX$k!#Vl78<v12W-5<)G)BY+LKjEQkq@sjS&<V(V%yQ}jFpO(%hUR;rk
z&N2^81w+Cuwq0~BZ)DoXYr0(-5VC)arhoShQHW+g09!$=%Sw4ADrF_RXo$UFRA<tO
z<&PSmV0lHq>l&${YU{R8>C=w<ZF-iBm&y%LdE{r8_gBJ@Goq>k++CbhVR3#6<|l4e
zNI{_D#j+wPvrtr3l2(0v$#@E^AvOqUtlnF_E_+Hu?Z!BYH3}3uK3BB}3A8FijmB(}
zTF7f|@Pw&`Q6_0~TrM;eR$YT!)~q$QHw93~>tJjZzh?l2LARaa=#eTTrs$Zd(&<}S
z22L**=Q;`kqbrtUl`o_XR#W*npf9}FZ1b79Y67}Ko}pN|<yH;LKH9RL8XvJ?+ZM$3
zBBJ3ZplJ*~%85D$G>b@;2!s833Fqb8*>Tku?XUI!#Q<3&GN62p+`p`N5%!A#v?g|Y
zeDAz0F1Qkz8L38YN}`R6F==-_W7(n&j32+I_lWLSu3bSRzUlJL=0AhXKhioTMuz|C
zp8apqI_ad%HrW5rAQWpdOC%!;iVhExuePQM<yM@uOO-CsI<;A?%GKj6UwYvs;*pBa
zFu6OMQ)qEtK_CMAzy(;i{;Ej#Vr$lWc{jViH0r`lgBw*u;6Er562_nt(KH_w#CnIK
z6WD)zUq3{Uz8VZwZeK^w?U%N&Q%AGpbs@W9<t;xQtX+kd*}C&RJ`_n51);`^_hcj$
zC3Xdg&EzFX6F9RS1C{yUD;ROyy}^EqZt|%XUpt6pgbsW8^xrw~wKvCb<F?3@Zo~{K
z!nWnP0?ReS$~6rlM!d?1Y+ECx4)M3sR@68Q$~|@TvZFJpj|E{v2qx0>V*?2PnOTU)
z#T?gIyTBPqvJJ&!hoPmDMX7{Fy%wq??Lx2?F~eGw`U+9nK3%AOnmq8@bSV!tUe?&t
zB=8u3l#Nnwbm#ejV5w1c!dUK~t}Gdv(R}K5{klTk5X;$Q^P*q$^rh!Tfpx)KM_VT%
z7R_}0#W){IP9phg64@G^u<)wuGq)k9%575XhnwOBgz;qaRn)1CHV$#rsFWCp-2V_F
zh*r)z=FKLNj!Ac15)$5OtO}nsVWSd&@b~A#zgjNbtBx^qxQ{8y#cw4qgj`o2mpfz~
zbbsZJrNi6ST~ZFx7>M*MtOv$j6*KqI-dg&yQ~kjeEXS_}Jw}Z7fIget)T>Xh8lC}-
zgF~-XS2grA+Sne;k0YnD#&fWb(2CkIbGj=Bg_>L1L@Zm#eBMImX$EF}J}Z^|cXeY<
z_c^#7ZPJ%&5HXHd_o2y-#LYVLtLB-k^9~nSX$yWf=z+2AFCU=|Z3{NiL_uPJ?=q)Y
zH?hV*Iy`mvtbv{sL53)0qn39g%cs9*FJ_tjNQ-?Am@z|zxRsG9kC?DB|2f)rTAM;$
z^EglP<Vb-RNimIteO3i0qE>`NFCHd%>4(=Oe#+6L-3VO&zUXFKiyYx_njW0WEkyDU
z;`>Dokg_v}S-na_m^8Hy7F+7d?u>0>DSwHSiZ5NE*~7p?&cVtl-;}wv_N!fJS<713
zn}Tq!7%t&1|F|N{CESAYpT=Q$;TOoAaw`=#QMsEK1AAD75HWMnHJTH6XvnGj+-*Pq
z(7mA(Ogo`GCh3tx{S<Q-yAesM9N_NW;!4>>nz>mR;5#Ign=fv9J*JFnC9X!;I+vu9
zH!bW4c$gA&5p-r6ZR>uG+;+a=;8M6k0EYERsP`PZ4_C{yeCMcBhM-n}QPcU*qhNG8
z0dFvY>gUx8PWN5=t6zJDi3XHAS0H5asT%!&yCzh?RT60d0HI%~d6&UHL(~x9VoJ!p
zh?P38Y}E(zba?$!AOrv~RIRh4mrLTF8*X*kbheF5bQ)w4yA<y&wr+7?a8PLc8vE7`
zMr8Yo7J?|ArwF?{wxanp7}5ne$)w2G+FO!U2Dbu69-GLoW*gmj37lyrv}9sH91Fyx
z5}KMk)0h<qD+m_`!>6qeLSbNLkgq5@>Yj*@(K(AT)5eQ>-dvL+SQF<K_t7Q>r_dc2
z7jD!9PqO!<<DgByO>nmB`Hg0WhDeN0>p4t?uvqK?1(6w=dWJo6GctPSg9yEzG~LM{
z^&!7_Yk}?pi>yhxFk+`bG31uH5rIq(WR^cHv_S~^k)@c9jh1>?DFWd7b$lVKCK{N;
zEc#cAjtCXnXnE<Xv#pNzHUl!c@;|6&bem?ce~q@|+h};Xj#QWGYrz=A=h5--me(ke
zzV~zupOH>8Q4B=#=sRX5wkZLRoKxm=cV9aun)t^+%LT{Vy<&r411yjx1h}0c0Loog
z3I_%7Ms1;x;7`&#h(0E5LkI!qB`WBH07V_sLwpsD`EzKslIir+fy_U&D*Cv$eLUGB
zK~`ZI<QxHTvBbENV^<(&41Dt#1ws%O%87q;`hBSp`lt}!x*`xkBYz^3Xia@yPCOi8
zLyG}$Xb4IIIj)znciN=eN^lK#|Gcut)9E^JeYXJAI`(c8i~nDYy;E=|ebnw9+qP}n
z*2K=lHYc|2iSBq}+n(5%@Q!VpU*4+y)joNu_IuKO)F<85wXXkK*Sdan3;_N2_y_>m
z`x=i9)}Wpmfmm1Z<nMLlnvSNiA~&zLP%{X<lo5j&hdGU(7B!V6X1X1CqMmPIiral)
z@|O&O`0VC>Xzn&)UUffraxzzGJ^I|4w)cJ05d>n3!7letW80e^+XExzIm>AA>)0(9
zMceJ+$YC5G10;#QU^1ebVeB|Sj?`jJ`Iinjx-9~!tNYbOG*k@pB!^Oqc%ko(jC`3^
z+XEZH!XW(4axQ%))yxk6f!|Rt3ht(MMyAdng-WKhZj%XtDhQsQ4N;2J?qtg*pwoC^
zxvVxCARBM*^BN}MnvW9%e1;a<HIMz&DjOoPo0~kKc&+5ivGwbHeNmw|bPY!hs<}<n
ztCs{rI+D^p+c0$eNkvJ}H)h|~s{#8IQt$Fn)t0Mj`lmz~c4<ENy6c(UgfUdb(=IU-
z!1wwU+q^Y1qS`eHui&P<HFI?#J&FKUEiy0hZT9Ty{7~vJpvPnTAP7wKmme|wYU4?n
z-qg>a_(j+82I^}OnoXb&jj~aYndnLZW6T*Tv3RoXuAGlpsoiL+bmXk8rW<nuyRP|g
zDa`Tvrt_xmMuZ?b{Qx4_{!6VIH~Z)6yOy~Xv>9M%%5$wy9oNEUR&k&sfK#LKuJp`*
z$R;9STFeF?1=z?}_HwN1v6zxI#AxA6sqQS)Tbb*2^@s5@5+8X)<2lU4GW&0%BvaD#
zM9i0y3*Et$7BbtC_%BuPPhG^5oKtRFnFwtKKSJSj{D1T$#Y}U>_5P|4$5%vTp9SEl
z0iSPVCb-nJsdD&t;)*a&%zD>DOd`WyU@ZUKPyb7Y@cd^_jqCr+S^nQT<lz5W6f|vy
zgKx-S4-bnJsd5zNhi!bU_=t3iD7Y}nql<oiy)gws62qBuE3%&`C=Hs{YcuqGxQm`v
z3!ZlK`SkE~dX8D04^bd_9W!d?!gPy-hmQB@6>T~SnT95Pqut8`UJe%*@1D|Y4JQu;
z@;?`qgbk<3PQLTbeQc#`wz@Ni*1rx;KOWAXhLEwDi*anyKnp}+@b!jNtqRzqTL=P}
z4TkONyqPE{IpDf@6C$I<EB7T;G2E|82Ck?Y<BGH`Ti+-Q*BtaQBkUZfOcXrfI3fnD
z(kL`10qn96lHCz@tLijUxWqE>vV|<}5iGin<L5Qy<Fp$E-;o{J_t1E=8hT6sf}$hR
zi`cI6t1~KKE>LuRz3Iwv$zF4rM;r!}&~Bjpj)y8(+6hh(BE`Me)?WW3(r|scqj4Hp
zC(7h`YNBfl-L0WfvK7e3yxj7#l>NoOZVliB&`};E^(8q+aA>Nss0JoQJGJSX?jI;f
ze3FjE?6-)Zi{7CDUOCj_3F&qdHgrfLVV?ixd=Hgaiw#5}B<6aZ9rFn$F<6rXhJmP>
zwpkoen46Ys{T&+2DPZf~;JO}VzvS$SFNVK~4|b-2ck?AdBrK`2N2rGkBoErYOom4+
zAgWngNGy(cI5H`@&nQeN0<l;)G70*9;tu(!Q)uj>OFB3H9w!$I>SgA9Usgtub&cad
zkc-amDgWVA!<{~+RqA|JAUTQFtEv=hpc%;igJJLqM)P?^XBEw7@4wqWyuE+Yramr&
zlzLI8>Ply0yL%nQ*l#(e&KphoNDL#4p?2uY6jx6pjf=$gMNA_Jjvq&Z;>F{!{yHWQ
z2wo+E+eZ*sAti!$Bk|5kwct7{h=2_V5}nmHh2xPm1b&>xf*1cF5P+;hH(zMWT?ih@
zin};-2mJo=@v^v(K_DW$ID~GLSMVuWx?hgJ`1e7v?#^CC{Ts12cOpo$sZ9_!^AdDt
zL;(p3AG;bsyufk+<3V7DT>67bY@q#~2jdS+x@BWKOy(9eOG2pOJ_sB*ew&w1mPX5#
zUgAKaGxR8B3;$1s4{vT$|CULur|vFiKyGL45mtY<3MnOPoWOMcv}QrGTpk40pnr42
zRzYt8;A54_Tl!*MhIjI>L{9BQk#+wejDt~d=g;HoKvWutVW16NqX1_~w59l?fKCcT
zbeT~VN-jw<6bz#ug6WDZwz6VjB(PZUQ)~x#cYdz%FVlicu<F7*d?KIzqvc&&)1%n<
zw{oV=rNvMNl1N94e>T$sJo*=ldHG`OH;EB<(nQV-B>`axARVMx$~oxyWE3w5zW&3S
z8VMw&$ci1fu%EhrDE1~v@=qRz1_2X^6?6|N?1(%XLcUm)C<bG0-JI4Xg*l0QS;HM;
z_n-eI7t__2F~wE1`vqN~l6Aun$TPKBkhRGmgsYS+4^z8P3JgoRqM$K3+9PLryeF5&
zus|Y&?wK~fe=i`uii8AB2H>yj6kELmA=V#7*ck*a85QY=N-6eFG0a9Lf1c-dfOPdp
z_SC^*@LdYJs7?D}l?x&U2ipn6qM(UK1I`Urbfy^CRsUnMzkaHpW%kIZkTLnQ7SJV<
zvm94RV7IRoA@GY+CaXHRMWmWat}Yu7k097IpIBnFVpuzJpoL2Xx!PN;S<(jB$`ixU
zC?8vZJB_E#?w29F*3r*e>g3B%ZR*BAH>bj4ieB&o@k35O_20-<sdf@d&8$$Kl<XoV
z&d7;63S|E$<qEEldFe~;FcZK9L;yo9ItvVM!CH$dN1GzQbj2zOvAw4*_9llnG?}n=
zeD@QzvMrE<Lp8u**<GVj$Rurg3U*3iL%ma9D@Y6yag125<qo}yc%V|#NXIuB!6_;j
zjNa@IWr=P;n?eEFrE?Puyt#tHEL(11)*QF--;ZVbSk8~AVWhzjF;D~Gsy6^ba*20C
z64#r?g@nhzLEuhV2m&gW2TOA)#I+{-2ot=nv_$el4}h4Gz&TFwYs29pmn4$jD&U31
zsw$$sq2Z&DZrNS2*ICFt<`-gWKz=GSIt@mAva!r>YDkpkX0#~txZh!vq&D5<<-)e{
z4S(dp09}^`n7_aWwF-;yp~>kgIVFp?JW?@dnd*@>Sx$qcy{>QB0_1ln^Fa`<voI;l
zl6;(1*GOr|L<eJwYV121sHsS(U}^3S!GD<Dvbr?$yx|jV&&mB~P5X~ADI3TC4?_Qc
z*R&iRJy&20n*aCNTJHom22=!vVj7@Sqo_xX-l<e-$(9NsYv-xrRloQ7%2Wp*U35T`
zLJhhNwgTzi`+;Q=vk}a=+5LRA5tHs{-ZWdFDaLr&5+lQCQZ7cdH?(1K(;F!7b@SDU
z%)I4NB5fjYs2E=GukuMN?{d?(M>{$LGCKl$Lc4zZ<Dv7(QHWXC-~U3m`(y8LapPeG
zpx1c_FQxmul}RU0L`j;EGNejjT!HEEP>OP6_Bhc>rie|epq8YD$g#I(yOgmjuZ3;(
z*USschEX(@@mZ|YxN28&av@Q^yN%xXFksZ7?)IKy+H5o|dwOjNF>CE2&i?w_$zc=?
z1`X_J<3@~M*0}MMV(i~#QT6WlOQ~4;pfmgPnuA4h>ENk}e8o1aN*5rndT?VjibE8q
z6x|Xffmb2%qt#l${qkR7M%d30>P4WLet0K8KQFeZRl42^u;3*+tX>hDRXa0ytNp}d
znjv#?Dmba1;rx&haSs7Aqh(^EiXCu-#S(|_$_=sM&96hkyNT4)7@cPQ!_`q&{iCea
zVodCmCN@l@yEAJVuc=S7XULE*91Fv`b+w}sFw&3v%2Z2#e$j}9Pj6~vrT4T5WkS2l
z;A!8YV>mniq%*g0al{eBepPb<RyY(k;ZDjC0r-KXQr4D0JiOvmD~m(DE?c1-`41Y=
zlG;HCvH$SEs@MrF{5p#Mh(O*6x?o5t?5w4B=}{p!N+%Ih=v9J1juM+KN*whiQC}_%
zDW-4ixEv^g4X*xX8zH56K3Oq#%jrG_pFG;pAo$w2L;h>DS~!-&i^^N-RT>ul;@A_*
z{@7$>M2^Fik5X(3J+*Y!B<L4fNpPio#AjvjFMS>Qz73CXF{z}Hk&(pA6&vVqW@v=p
z;u|IcL#7^KS2%toTKAnf*kklC;Hp0-@1d%cW6uZmQ?m&K-GC!4?38{kpfGV=je6xY
zWvAgi?Nd7*`son}7-6$i#@OK`{O2sUkWZPu{X`kN{SWRuM_x|S+k@pqXY^{d1A#)c
z61FPBMy`o?7rf!s6PYW|A`8a-aje(^G&eE}Bzm<|1rwTKuu3K<5<!Y4>UT#hWCzA_
zC`s92k%Q0`d^;OxiV9>B6s4rWjWnrPQUns$dBKrvZ;!>N6U^Im5swH`5)2BIMiBC1
zkQH8uk;yAeDe2waQ>l=3o8yv-<E%192+JPg)881Oi5@(HF?zh!g}#-P1=2ZcFt7=q
zj4T8FBGPs>lDv=%4mimwDm9WNMSS7&KfiEJPAH;X<fgGu34&STV9<bp8M}R~)O6!z
ztKN>f%m>Cs-$5hh7)0BYVe%2C^tyZWQ;jPA1J&d#8AB4VPIAlKm@m#!FV5LB1$tR6
zQ|^b*=p{yWBoxxh@#2sUxU{%bbh#{OkqV>7bsJd5i+?Jl3<Ix_-+a4ANJX}ge$ij!
z_o-{*i5y`;vi%G(c^8_1$KPF%qyiC}zBnJ|d?dbg5bj4I1Q8Yal{SqvG2PS(StEp1
z5X-x!B5=5fG^S%f1gDl&F~RWJlxxp{ZZURp86mzkDWnt!krb6gvzlcrEyCG)l5fqY
zst%Sy5WG`{SrkfiNQm@nq1{Krjqv(lSinstw-B|qWFN$|5<v|c!K>b|*3m)uGu%{9
zy?c;3$ME4X0kFKR)jQYVhva+dvRklC=1dpJ4j%m4g~)A~0efLNGwY9aUu{i2SXaf}
zohPVi)3gNeJ>oS^0w3q#I=jQ|AewX4tq6h_aRHEAT4Mj$n|&72=r2QAvjAa@hoh95
zi}EA3nUzQgmO2WfelZ}Jfn3X2fP~_HU#je#?D+GTw!}L!h@)N;m0{!U$Z#ZNN^15-
zlbdN(AWuvDYqQB+(P@+vLL|-NWE<tIJ46W9-os-S&zVxT3{;@>BUe&jDN#M;<TSyu
zAvd@Iw_<558%+({PG*Ib($Hlf)ePg@Cb`5gTsWX1!yK9{V(Vx70h$b6b}U+w6U`0H
z&05H0Wu|Q_bmMf86HKfIm*(GbuUiQGyS&FI;5#vU7%N0&Aqz%<&VpIO#fdlfZ|QtA
zRNr}w$6AF2^Y#dmUTFxB_}`uq8l*<351?#7f?Ja?D=NV($!#RBbW2Eg_iYO%S1}(n
z!&zcgols+Eh+0snppDsmFYjh75T|m98MU(%0Ubw6Zcg`f>L>Nbp7SYz<Lbl66)~55
ztn9@EeX58Btc8g#I`?1THR|3!xS$*gc*G(5U1}Kllm*t^v8+r4BqvW9A3G!P@Ds-$
zABGd=^Z8Be;*kd3r?v#CYOdc<euD`vshO`A8qz6PSyl{lDBMqCbsSi>LO2HoON^Cv
zerV_9$O-Bfwt5N5MIx+(&kUm~8wj+?>&3zY4$d<WZ%QuGnq7J2b=Ntp_Mh%0Oy5$Q
zE@CSTg@SbJAW$UH*y)XL*M=s`QZZ%B(Pw}d`kYup_}-^F4_@`wg3!6no3w&tVDe(%
zVS$e{-|+?WuzsL@?E&j#F3$3t!RFX2O?9tAwOr;)O+Ex{gr63WE9C9uRjh?1*_B9}
z9s6)#=9+>gF-EwNW+$`ckmEddeWBt)T8(Cei=3qzX>etTl>Mv|13u06taVlPOS^!O
zzoce|yOA=|So@_wASx^t9!WcNkWVj&!=pU^>h#^3ij8bW^@SM@`#^n&b)a49AV2c2
zjak4R<)z6jyrq4}2IKaSNI2>ruk90Nlie-9O$0J4Eab^nP7rn;+nY#m$v842XM?O0
z?>+2{30^o?()ERtV=vHl()vMn!>1t&xFXN)Ly_jc!<I5xpeKvJiT~86nEf;S3p;4E
z(8>=H4L<!LBqO3<TVUbQ`}z9_fx#IR8%8{ZcaL$#G-aF3%%!EWwyKbZXd5C6DVwZN
z*u5_=%KnO2>qn1hP!e#8#|ey+^BVVy&Md@TTVfD&8MYDB#_$G&u#BL|u%WQpULN(W
z)wC1@6(T-xWi8;l9}n>c>pVFvNziCYG;&UJh@>?WFTe;2mRXkhM^LnlbSL3c`#OTH
z)+pDht;fbks}PqCm6|e+d8V=>IG<hTZhs81Nqu~ix!Z<_B;z(fYMURb#d7n~FACy^
zAz6ETcacYTIbGxUD`K~l@95G)^cs@oMM!{XLRql06i<vbg`-lS8G|_G$Vq3v4d?{U
zIQHyQm<rp!KhbRQx46>n)K>c(JUyT?45#4HZ`5ahy1;HbKC>>YZm*WP^O!i`YYaX*
z_Fb_Yax|k+2;PpHh;4!?yE2eTKR*qS6VM*(l}pdP=06TEhkA3m1rh#B1@ivK@&y|k
z*Z+R`LgyRx)`s~%D)0|NEUtk>lo&9=&e9_3lB<#586IO*vag<}1yAqxluhUBjcAi%
zd`ZCqtw|fhpM^PMzui{=8%`PkBfZu8>hN&-z}fGW`bVJHx~U^U8}Eifl4obGB2s8D
zlq4EUK6m%|@bUJdle-l@ZRksvfp+BP0C;v*39R&}Y<2W+BG(fB+M6Pi-(3U(JlyxO
zCvT<>FyB%6uRjk*Bo+4p3^sZU-#cEv9|;XRq9!oT1@n1)4z(@RT5%>o4Gl6x5H|6o
z1WTew(DWhcsln$V8{6&d0Iq53T@k4-7#!zD!Y`&GH_#KS+m@O)l8|Y(xU%HDl5jVu
zh3U`PUl{c|<Gbr|p09hD>$mT$Ep%G91Is$4(rl-C4ztWamovygzrxR)$cdipb?n{P
zZ+vdr)`{H@I?nB`dTN()H&KHAQN%26B8?6u4=x4|hEb+FnCiTDGSRn=82kcu)WM(n
zNY8pa5T&xaOiIV6WRohEAGFb2HLG6!UNl(zt;&8+T^a1-2x<Kl-Jt(u&tGPvyxL;d
zeT#JPI#8}b+Zi`J^`yFSkqV4Akt2w{Kh!YB&$_%?iU;7ugL)x;fnI^J#D*Pg)PUpo
zm-6XUpY@#B5x10j`*8Gszp!ixCuo=H@^I?-v~NzXl_x(H$VVLo&R=uu=!WE}IR_z|
zNh@1R=|#tGx3ug|8l+D4bck4S@E5dT&orHvO^>zhSZCy;h=y6KW%BCakNcgDMn-PE
zYTb<*MC+YSB4dY4)-oMr1YEaMMV;T8Xv4|RqL<#BMDS8jofklM`B{RwRNi&se;x0T
zrWM(eX0~6B4BSDlOf#TI>YfMcD~fI_NZ}kjNW3K9d>H4B+gqzHS?CE-Cc)$V`5!Aw
zXy?B<M*3Hv1Y{4=e4gD~936jWbbil$Dxj9)g8CQ2upSv)<LA#qul^X479zkV0Wb;X
zA`39qV@0VRIdmZ8GeIapP$j{xyAY0>h7WOMP^QqX!aBVW;-jZv<RAbpUOTPkt}f;T
z=DQL#TNYl52%tnOi>>4%H-?981`QHh(+s+3(KyELb06eyY=LB-?wrq+8T&M2caaBR
z#XqxYjHlbONkw1W`fpCy+eCPM_9J-(OA%BeOHv?bH11Mj13%ENc~j!z9L;&|5odLl
zHB|gjk0l(RBiR&rAd3R3H`ts~2M=*+WNG>!%!L;EmEaXz;9_sEk8}5F$A?^hDdpAr
zM5{@#@S2;C(IEM9wB)|Q<}6_RORd8OO|-zEq}EN$5ma33#S#EiER8=lk^lSBZ`V@l
z>_C`T^e?Q%&Wxv!Ez=0XQT!8lK6A$eE+$DA>54nXPD0M%LG?b93m^q?#YmjWg)axk
zI$<Y0Z7XDN0Y#6dfQw`rK4uBRnVLs6!RewwwX>pZ;iN{I<MwNFnn{;T2_0~P=_OG}
z=C@UE9DtcD41}2?gkGkb`LO=_apb@lsC4h$F+}pbzPRJ0w$R9^h>+o|bo1~@DtSU8
ze=<)Ba22^U#%G#gYleGwTvYymSWU0ybHlShBg%;&0#2)A&pwv#ROzQd1m^oufWXaV
ze<~GT={)~{Y(stRx&}nie%!!YiacY2+`q6BMfFk$>RkU|HA3i%lFg^$4`A|OV(wv=
z3MErMLFnZBQD2<XeN~`HP$7%+{4;A{Rc+g|I!iIlp@otA_xUTVjV~{zI(}GjS|{G_
zY!TRYB)pBwTI@ltdB8467c2NOLhS0_`@W7Vr0&WMJG;!l&YE=xD%8A@=JMK-Hiw7`
zP)qV<T8gPejbLskjU}6;l$GD_J}+abFkW`__X3&?jP{Fk>wiUv0vyJtlTN@62DtR2
zm>HPe8OnMPqJTu|pFYbn{Kuyo{>YGFrND8fZ1oV5F&deYMkqL;sCKe*5}-Rs1Z#%%
z<6;jJ)lla0S^jWO^QGzPW5nN@d$+r3lQEdEXdg!7#X6<PKh4ZQ>fel2!3e7^6{u5I
zU?4oFPzja(0hV-{Hd`$B!slyr{C<UGolSMFBJNG%wrY&xd((_}{5gyE4hJ{+O*a{S
z-p-veZ(ybrf!<KbJt`2fs3Vu<rOFx`FK^DJ?hlq88onkFnD7+O<xEK-0%&ATbPXq(
zFa@~Ks6>i5MI`s3;HGd`wJp=aI9_%O1{k}H3Ns>5u9~=yH6bOHDj9ofXe?R%gyV*E
zpz<X1;&|*jhdC-olCw`Q1bpf+5-xlK^_jpSoo8>T?&c<>l%cnti=XoDXVvx{>?>td
zhyO4pbIPM2ZN|ko>zT`GAvtsEgldbGH9=pfrb;Wv8>LxBSrbQnH9s`cd3+DEx)Q=b
z`MEQjkiG`P8$5nB+wpq<sJ@52=<>awMbInV%A+MoBgnl>X|%&PA-WztWd)&7@&zj<
z;;P!%r>1vfJ%o9=Ni3n#ayj0y8MD%TYcXTSB^SZ-BP}U8d5#zowADwH#)-axs8lmZ
z{!#IY4nq=Hjq6Pc?=BQZ*V2W;QU#_366vybzGG}4#h+nWB}}vZqehBWz-66`UuLT)
z0_A?7Gb1EO<e$8J>rV|1kK1zT!koqBxA#Ts_BN4Dz>gu^>wIl<p|#E`vXhqD_Jp_o
zy+4<mkBd@_0mq%wsKC}!k{yl`1%#g@euZ|?t=z9~gCPtaw+=_jn|wkK^voDtuK6(B
zh0Ftl7`rqRia}5Hl^(7~B12Q{&r;Vqqs94JbS36zFy*!2hNR6)rFU^_*4AQh=tm_+
z#3u*(__F{dq>lRZ-hj|VM7FZO^7V>1`=Fs;qV+;i{Gq4Xl2UD@41xS^Vr}9ca2_}~
zLQ^6@%8}XRu&Ll5x(I$UG5l|{mrzIzNp<Q0%q)KNJ~n&*#(q=ah=q*?udEz9x(BS#
zyC1GC=Vd8<Y1Mq}haeRR{(T}yaG&;Dd9;d1VpT4k-X&dC#zHhS55Kj0pB?>p2(0fN
zckg@`<$~rX0wH{T#UJ?z=hW=|Kx{P?%ljWPiJj#?+-|(=|9c40M*5iQI?H!_hd@|D
zZiTi?1BIUS01+K2WD0r#4eKKdQ<uUj9*kpNn)zN>fi~ExL&N&1Lvzn3qv_Fj|MchZ
z$jR^_(H&t_QT{;YG@&6+Hf$)aM@f?FAsQ9c(Sj1d2Vx=@cc-YOf63;}=<Lkx$!u38
zUSDi>a1X2#I^qqwlX~-JO!3X;3YdOzvU8KJw&x1YtrM$yt$Rd-s7w=v*ne-5au)WW
zmWqznq4INW<avbchecG25_RP}-DA9_E<JzJkhh9N*(x*-5^0*zFNRp(Q}Iz06<NEY
z>L}tiZaUdTb38p{jpUguQb`Ug0n{z;+qhEBcy*nM&c?eaY8H3L-OTW&v)`$+n{vUh
zaKSC9k{E`QH4|s>`bB5EFyTDNdS_P|doOf(Cq<+71lnK5DD_0tUWMqt4b*k5J(2xM
zHAqR0P@phgLUkg{tLAsma?Gkn`osC;O;tFllGrqedzknBotHQoqxB_7YNl(<2#qxp
znw^eF_KX`L-!vdbpai3#=#_y=2BL(atdNFV596R=WK5IHTPJOPWA_v<PbK|o1_U=(
zo7oo$Ax6*lyJ(~>9t<wl(MS;1cd4RwN2jX-Q!0n2nYJ%mDzG+_=+fN{inSy&k0l@b
z1tJ{O;3$f>CYHE=k#Yfvp3Pu{2m_Udw^N{eZM^|aLh2Zl;n->pcLMY-`kjobwjJul
z!NtxxIw9=`ST1L56w21PYY2dDF+j#ZT1sHM5;y(_`xI3Wyx8kh3E6Lrv*0=u*4xUQ
zrRLxZUPCgegT(jcUu%2E33!!%2|c72x8r@<_OPCER|PNIh6T@e;`!Ho;9C?pyJv>z
z7E2VbNPEv|$&M!?K`>L113lIK3noALM{B>n#+WT~3}zF_ZRmJr4Y9bk`})9OLwrb#
z?Sjq$%_0p(RlO_=V5^2KP-rE1Is`I!(gH<Dh&0IxYaDt*gXBYlrvG#Oxh$h3LoTI#
zyL{jOL0QqM5EODrz;nAp0OjXgk<u>fFEnb0ct#}7Mp2g$gnkp_ku_n}!xT<T(tjXK
z(*Wf(2dZcmxZNn&V<e#blPa&g&FEQRs9U}9OwarZ(w)j?n!AR2kKj=s+wC5adNw7L
zqfVS`mCl}N@(@Sp@_dm$ASraAHTgQ-@1J<=b}n1t1uNtcTGIB_N0qQze*|o7)B6-g
z7c$TLlJnRnWnZxb2gS@0Fg~F4ytl`T==k%}ZI4>^8{YMw|NVa%&{$ai*Qm^59H0xZ
z<#xlcU)qLB3M6<rm%6o{|HLgh>d!(ZyX=T~OyW>D8D?_wS??7S771*$it8FSsy4%8
zcz3jq648@?&#SZF!>i);<b*aBbbaflH``+&i-kaJ*RLJF+b|<gJ$-4`r>(7CueZxX
ze@gVPfXXwH-A_QPhezvE>(!E@qd(>7Ma0);Rr0K1uisZZNuJl>R&RS)0!iP<Q*TW}
z-`B<C=icel19R<_81DhGRID^>r1be;h@_gfB+pMjzx1fF3tL>5;gr!%B(yXOR@8Rd
zPpk0`>PLOG9Ci8ZutWp6bklO?2nP8u4@NPn_V%n+`_J2xr^R2B?E$KQ9#$=IRD~M2
zq&kFit0D)uheY&+h@NsHF2CkM<?BF5lUYp0X2B&ukDZMPjGF;()k8<sge?WM98`7`
zNE0S#3K<2PU#1qae%{kJZ2i`ztz7BN#SxhhQd(i~7(p)S>1CgQ*EQX|lCJuw#X9e<
zg{PDIw@|=2({AofB2&VM@7+OX=44_s`nE~l*NZaolM&S~5Ke4$JZ0#n2<b&(dJSPa
zcbmI|zy#pb$$(gMM9KIKBC*kpDz{JFvpLp&GJDLVl4S+Uu$xSdGzv5`l;%1*&Lk;>
z6M^NXH!^`RSvk_K2$X-XT2tV5k>nGY_oa2Wq&%P18o3kghj=Gi=2rH&7^a)&`S`1~
zJ1KFYgBe*f7yV9UT8q!(RhF2ZGz(Fz%LIcdTeQq2Ba6ZZlzEKUSLs&Znshi|)X>%)
zgmAS#e5HRtqSL^Z?DVWYRC3qW{F=ybSbC8UG&C3d3S+RM>I8&Q^y)-gw*<E6hbUSK
zNYE_N%|Dv}&#**w_E7M<T&uCd#<pikwUDsKXf7F$mhf;<kUpj>>&BU08u9IwRgK1C
zM}q{DMAHgaTWJQ^?Uc~#A2R+dx}~eKW1OtGX7^1nbtSU19G+knLU@?wp+Dip2pqH_
z#FimIvlfodSeZq-$pumRk?!{sFuf_jSoq1>A+@5wu}qW;woz&lp0}^@2C0609O<{O
zqLwK`Yy9bc@cLnxjD|A9O#Mmf=z6VM7qQ_%*a5Tq6emtX>hZVDOjD5+k9YbS92Y=A
zIfH1LuUgJk%l&#W5Ly)hO%iW2fhbC=%E!lRgNmXM6s;cllkE<u3=jF~G7y^p?7u02
z@vNZCeF;_2P9CPHdsE1GtIJJpnqzoEwMqd<8fY|msvj1W3%e2t;F>_PKcKE5=@X0V
zCl;hdkWzei){Xcp4p#iLZyUz@ITa+C``w#Dy*t*KCtpKh=(;EgP>fU9o-?3~1ylDX
ztkX8RWK55FlgG}Y86H)pTF#u&U(6n<J(t%+X2Vdb%P~P`uTW;M5Jcd^QO3;)MJM6|
zy&kfAakL%1vgqgX;lZkzGI)$TH{)a(h(c%s8BmH{vWCcb>!S$l3q9UUmltf(^X%vh
z?A#HMK{e&rPS~>CH7zw;=0%B2RZ=^+#{&9j;{-k?MEUs`;w<IR@#=6k#l-XQ$!in`
z?rAoR&{^wYDKh$&lq;Rge0e*n9L6)<U{|0lMHl*|2y~Tq48-l3GC30!!pnV*3Uwj?
zD&wIU8%4SiOuzB%@#!=YiBd+PiIc_Nlr3Y8s!lpQxr5oZNE`qaHo3pc%M+N>5_HFf
zMPb;0Oxvlg<_5>Uk+kO15y%5;;wcz*!&Dd*^dqG?QP;Az{ne${vRr-n07~!_V8to)
zI46U=2(;s;eO{L%;l|fHvg1ZgZI1z>)VHyWg^A6R=fwU|3ntwe;<-G+R3X)~`nE-j
z__I8bD~`>H4`~d|pG-ELV6>Z%itq>lGG$wv&5Ue}W-6YE7v)K4)fn2gi}hsd4iGw@
zjWNZ&rDVmmF*WuhqFi?CgV~-BUiKK%Q-GcZthW6^tVOvD<oMO1)iz#3l-eP<iHYb`
znMp~xO39*DZ<sb%6JRuB-U~=RLo#9j69IJv<9?J#Yv5t~rOZ8nH#dAJ65wLy12chU
zR08@>JY136mqWDpl$w4vbLG##ct)RMc%;QVx5$$hvi{CM5idnh(2qe&JF3Fd=_9A}
z2&%G7()eFmGv~_4o`)?&wlN*43s5fRt*PM(C2#!?dXavhIgtyzen`Qi{oa1qBAhv&
zQJ^Pg&&YojdD*##@C3-ZvrcSs-N;(`4i=s3{)zchQo9#L;DU;=|BAa0h*p|uBoThJ
z+DZ+$Ewe=KSt03AywC<QEgFmup{gnOhMkajtu=t2pXDC!u$M#6mLkOo3*85(vh=Wt
zvm*45-%rld;N|k?Wvfrm3NO8t)V<CIDg=D{#=BZWo>w^5=5hJ_nsTGj$juSR?jZ1p
zrsr2;owRjWciVF=|H`!fvUprAFKGK46nC^T$Zm}M=XbY7i3dBi{_&q+3Jc@J4m3tR
zp9<+j(ed#2%bx%}j5b?SB}xg~A9tm!GFfcE$DijSD^Cjiv!9I!X_q96{^T<Z?jrWB
zc7c7>+Ne!LN!b_-IE{xEB8f!V?I)9{`R$};I4&sBj^TYHPz!ZER?KdG?UD_81Ni|2
zAO_qI59gDlyl&xXSr<>GUzxI4@(VB?EbnKxDGHx9S<-S=liX3ky>I+BOPJWt>BzA5
z>z$Wh;?g;)he^CG4F3#PIGdSeUJo0TYXA&t!H2iZaCMX~|D_pN|D&bK%E|M;&Xm3Q
zzh*I+8Eu$N5?5C=ep%!nBN|Gw!{YLZ#ZqMYNeyQKTaZ{qFuF}yZ4pE+WRW91Pe4d3
zF?6|9xt3Ok(7nnhY7XbGE2T{ZB?EoAq<xCzpVSMIWNw8;RO%IGd`d#43Kpe?rM+Dp
zW*k@crqM5g7#`gpPM^U0-BY&#%pbVqrRc@huRD8h$k<vbN?KQI#S_B*e*7QLC(B!e
zQ|NqT8zM!r6LtfLV=i$FF)w+G6?Tc2r5@NkWt9)z8&IcZ<mZ$(5a<*WZ217x+;8S@
zm@DRm_pdRHG&7PgECbWTpo8jx#q~lRGTBTj1vejqJVl7JiAQa2_mI7znrzO>3V|3(
zS3I@{wmHr3JrB&YvkuxUaWNYlICc~^0=VFg6iXflqAM`eVTlCGVVDUs`Gd{rEE}J|
z%<WKsVpBm2L%Kxpei>6hQpc2ytFj)f=tEV@<u<E1ZBiF`L$lL8E%kJy$zH`bQ1^)W
zwkoAbXi!I77vV1woAD8uU9Fj)guIhf?oFCP{RUP-UT>}<N3kj0$gtn3zBmO#?_g(W
zqPUk7%~NDQ8Vmhzz9b{?3s1XXNC+j{x8><H_Oqtx1)HEP6Q6Mf?@r)KL%DN>4i;iO
zk>(j^yu(0!XJO3+!Em7E6xy_fb#tfw?rzPx`rRV7T>I9pj0U5uqX{+bRB%~Va(0#k
zn=HhBC^lxl=%J5Oq7F0?gdnfJ?)n~q)}E7fHw_lDo%^qlJ3wJ3<KIi_5jL3jkF_8e
zBT>}?in-X4!uXCtHDw$SO+g+_Pf<q?^rn$MtW@+0%*%5Dv5-cqSy4hz6=rPp1c3=X
zSvN8^(O6WOSF}FZ_{g+;kioSjr>0@yed?zPq^%DU3=Cn1moV{`?AmcXf)L}pdTQ=e
z9FYl`lCu)NR5d?N65GgpG9iy8YgM3oWIu}kRkVG5Jt7TG=KAU&3)ooj(r^YS`z8eb
zma*Gb?IF$-2P0P829bI!92q1(oB}VZPWpr-uOlu~1(R<`Ix@>#G!9cdiKrsa%wtAa
z{UQ0uLtT@q*5i;051tiX|6G?9Tq+q9BNHj&tzVFK0*@9>;UY@;9L8Cp&PFR#`I5UH
z#Bngs$gyAs5>BpEiF99%C=KnOp2l#mWCLr336~GgaU3782Z05Y?Q^SNZ=ygrn6;%L
z?uSTwAITtRMHeViW{R=cRV_$yLQpN-9MT_HI(Ytb^Cgf<8KUxk^P8KlEygC>ih;f%
z$q>nsY_S>L)E$XXs)7j#Uhy5jAR;5;^LhD9A}hfDY4P=iO>~IrOEvZup|4^D7^Zh3
z-e)}42qBoX+vJeaj8Oy<|DJ_^l;apgoz<~-Cz{rQgK#`=F$B?$Mi5ioR#)?FWQ^xP
zUP|N9zsTy;UpRAj!VB`UBAo^7=jWv#K<fO4FDzeeEu0!H79mn950Q)&mt5o{dBo?Z
zcaMDOmRpo`ii4LFS+XQs*eG?OGd+E~$7htDq)7%RlRhGw6g&&{kvLhSBNzXTda~Z(
zUp|ONUQZrWtIsKqk8Qm=wqJa#Bgsyt9_ctJwGAs(MjF5^qw|XkYU#}8vHEc%)qUe1
zxkCn~xsK#~xxs}1J}I!W#vk0k=T!p>k3me*Dxfgecxz=)-D-{0ce19JwP*114nx)R
zifMVEv28HN(Pzp5Al0qwa!&Svph<h&ULYR?lb&4A)1!m=6Y0&4`#>AMCb4;Ip}}1^
zw^`3YuO=<}3VnURx7jHmX7+7Y^fyzT<0=HzL<Q2DVAkHxjB?*Z#1F{6!8YFGT4wb*
zgO}=4qo%br%RH9k8GoRcFt60PFpHO<Zx;ya$_#XD9x{$FG0SJ|_wC=J7##8?{4xj!
z@{qfbY8a3VVVMke2*9f7R@+?6*S%&)78<FmM=EX`69Nrg>BTn!o4`NaM8?1QkIw3_
zEV?aWrNyujaoyC5!PRKx_p@lkFijX9@nifZf1&1MZ|^)SXPdjgRblpnZ&8rslF!Dq
z3gg1v$AkEg&}{-5QJB#rP(JTF3ChrIT`v?`!Z+GVOV2Tw5;+sKZoW)8i(gncMQpaC
z4Qkg%<z{OG9e6c#zfWvU4iocVWr*#+Jj=Yi|Lddcl#ZSwaQJ`xUEAc4+Q9A<hFR#N
zK_Qtrs<J<7`Brct$xB$MN$TGp{q1b#e45EIsk-S$ppbQY`s(V-Arsz)G{@^zOnN2u
z)=%~<N!&4x<+)QPG(J^TbUKZuHL2d=SX4zU`C{w$$HT=l?sRVWq`_W>w4Za&i)Y_c
zJwPRQ<#M|4_T-|{SLaU(HiK^eHz><ZHOZ}>j$q<&^sCL!x>3*fyTz%_A;-$A-u5&T
z-(K}lI>hsx2ew9lf%1&=b6lDj%YYjmqnFsc!rtRkcIBAAClR^|+<tT2{tj=ddC)7D
zqw9W9rCU>syqFr6JK@x4OZ8LhneW|>ssDQJHeQ*_iXu5yJ%mtGG=@B^6LGa1$<2ic
ztv23PhTkR@@-STRh{yuLoz0%`)#)F?$d#?eKybaEHm4sU?RBQOtqIT}1d|4bSJ+V`
z42uhI2~=CcGcBVOvd6I(@IED_Ro_@QG};OoIGbH}9#}r@Gn=nAqu8B-Z->FYJjo>c
zXR&hwy1!PB!zY>I&s&)iSQS+y9#F7!+eFW0T_ZQHe%gYYkkuw6I&HB@qxpRRp(3aK
z=4;MJ#eo%Xda=TS3{|vKkVgV(Df#G0O5$JRA3ka~zBw$kik|Zm^9%0`+p0#YUi&NQ
zl8Pv>e=@^9HV=TNo9@!CySVv&)_qKn=i3LW*2z)*!=8LCtZ6^$8uO4)`7_osa!#&V
zTorF)b2+&j(NyIM+r5{Tc*vta5HUamDd%g!gUJ+*+g}9W2UY@A6;)^H$2d_<SV)E*
zMgv3VO;OHE!TXU)NLDg=VW`&=%;s;m9em)3reQJ1^7RcdP$;A1dXhEz7OZ+JC%@XN
zzqdoJrwVxOQ8HnSK(xxywudogpLtHdx9r{%3eV*bC`Fv?<9TYX&Y5F?g&gu-L(o)T
zfB=@ZKV?oOWxB4&Aef7SChTP)5z<9QtCIk+&c?(Tj_~i?>4ueZNfj-gZmtVy(|75O
z=7^dh`!w=w!GBJcA&q4`EU+O-n|*Qy_JTwCs<K^z#e!g6LVple-w255MK{IiK)SNo
z;1EggYUx#zS4@xYttv>>1W%jymgH~U8`%NK1{S6IR=bb`L1(ez+ge7$c10{EMdSl#
zFhwLG-uGpz2|EJ@)D~#zD5q?BObUT_#?vQ%aWO#H1{Tf*5O%Kyd6AqIsVv!3j}q&`
zody1J-Tzx;L?Ei{V91s0C_}ISsML`lk>Z(f5K@wI#BjG`)`x1FU#IUNTVS%RWSZ)<
z{ruZAt#rFuf=Yi0WVHc(rXZf<h}$-hDe2}4J=%sWw)SY2M*JN-{Plv1@V3EwjuD!v
z5P*r^z&ff4{{_K!1R392mUWaYRRDX$1{Oj|hreE>;5U3okQ&)J%o9NoGK1!tM;W4o
zannDwbk<EPNc*^Pz{QNeZV6vjQIl303xCU32Y+P`XlZLqy|k-7Fdyypc{%l_8C10Q
zT087x!1$8mq9*Y${~CSqTF;$%gx;qBN5n>LO`a%8!%Ey^dl+@jgg}=&ddRrnnM5j?
z)zKAnbAQn}Hj~Tu{a7ZK?h$YaO+1=HAS3^c7Y!jN!6#<{F-UsnW9d7U?IjIyvnBcd
zR>Q%hQ>S*p<mh}~zjEc=^f-d#6~bbRvWvuee(BM+qG8_6xEXTV#+ar~D3g`=FzXgh
z6&XUciPG&IXd)=K!k~hE*<oYaO<T{JHe8wWbc!`t;&IG$H3V+H+{9yeSY63P*}x*0
zX5vES^O$RgAQD}Mo7eD4f$<$6E=;=EXljM{v|&9}k{Dt`Ea5E?UKZpS2s%Dy0V=A<
zLbPut3yh_N&$cS#MFUG(Q}MUv;?=IaiYEE-jU_>3+BYbgfeELD@|}RaAr=3k2pWHd
z3J!*lQ;3IbZ__fouM#A4BvQ%H^Kob$KGg|}rt-^TG*xTrObmQUAImLvCbg3Uvtv29
zqqigN&mjqhBG8UX)o@&Kdg$5owaUh_sSHbK18s<_$aO01Q;6pc^>aPXW25Fj??YS(
zz)?q9Sx7XHGpvQQvxH+U-gU*Tq**u{#vdE2+3w1OZ4*;van|65s*ul{7ubKB7Whpa
zvuPJOZMWZgL00?Es_WWcy+%ZZRFeN+K~5=HAYEDK<4sBb>cCfr9WvJ49_!P0%5&<~
zIY*mANdGDk36BTTr~5z?`TLn;OmWHmiN)Oh3D-nDPW+hBdnRUpJGu<--X_xNxX}e}
zmbrWEz(ArkUI=E0UO!*b&8$G9_?su_jivUT;y`@0ZEG*o{`2vZ@z`ivBbmpx5mz@c
zF>|(S>NQS9zt-z;cCoLVyr)z+2w%(ljG7{_4NHejV`d^yXeeu`+}esqS4rpHD><E#
zI#^c&Tf$Sda@*ntf$C-xTqTwkyAcpw(c0Lu;^yP2PA2X!+j-IH-CHZncYS`;osb6|
zAiW7)uO`$xHi4?QrI-RwRiMhF6)8!Yj`LK+DeE~}T=zxCsl8V}O3H0}-I(yo5}sW)
zoRm=JjZyXMs*LH{)i!=8<cy9h10Ttq7^a38+GSJrX)3cvUWVc!-aFFM*Cq3;9)mH>
zHu|A9=q1<h{BJ1H)O748IY^iK`F94qR~Pr9UnhR6ujjGWU&IjVX9?3<P3BKz8^xIE
z|I`Qn<^N`9`CmJKhIGDTS9#FB3BA`eS~@sgqUslvxcn)rf6eIUa+Aw3DEhH%Ddx#l
z-QTZDMM1s^8Vwwo)q<!HZ(&m0ueK<bPq@?mUUv>Rhow*1=HD3(<}Ga#8I%8$;gH0*
zw3?2K`h5B{)EnTfR7aGEDyoZjUhe;S{U{s*RMK*#yN*S1<HSar5qf`}hyW_xCFD8m
zmOg!bTpSJ65(m7UoAvbfM<>C$icyj=%p9s0{FJXPr0I^dF#X8q<3Nr$<V%9+GQys_
zo<v<7Z1gWyFETSc1l!Z`T4B5plIgsg=~yGmDvTbF&Z+*GO|=!y*!6CzELnpdCr>M!
z%j)N2(S#TMD8wCiZM6u;`P2Ub!OGE+ZN&0X($-1#az1@&zlveOAJEy|I?+w5X;A!e
zXY^|BI|R1X)g1<XqWJ=2YsJlVBPhiOEy+X9gw{3~srsAI-DvV?n8%qXPw+hF^7I>e
zG#!+ASbiJw!@1Ww2Yj!BQg39rQ@q`U3No#eZjxUw`gXJ1>RNveL(9yf<@00C4!LEy
z!gA5Y)}#4LTX;p|2Gi7L*@4!`)sa_A2K|tc+t}2*w0)#$&mm;*bkOzJRYz&z>Y$$i
z@XB=_-M{6n7?!E^2MmCb0-W4aij`6k?zV(X0eVR_j)Jm(W{x;|Art*5x7sjU4TE})
z)`7_G&XwObd&=Rm$FnJD>RI}t${8HHHXe^4-=c+S37R<t$$i9<fB%sL$gkFKm*iZC
z^0%GagBVv`ip7MNG(JRp{o0d{XG<Xpl57!GP&!&5q{?yOlL<Z20znpS8bqKxW`I<*
z;U(~rA#6(G8My^gEU(_Jd|8DGa2>ZV6P@JvDX@!QJ=({;-$WWsTD*@cU*ax%5NVXm
zCZMDXc4@v}sP8>wy`~ZN4pwU&lwtA(UW+=3y8I5-ikkh0Ku&sKPGM~zG4T{kFH)*q
z4zl$ibgpa0?x~?-{8v(uaUpk>9Iccx7g|s{<!TJeGe9LjQF#y?dXP7aEga)WwuR^p
z?f}E!8M0ptG=rsPCJxK{6G?>>MGG4O9utq2F`i=e7$(F_;0)z0iTzxfOH9=B@HRPa
zv>#6F%U}&`fY(yGNer!*#RlnZiySJQ4M{qiPDhL^=wV-qaA9hjf%vN0x{Gg*xH+e{
zY!8gKY=;1e2oXBrX{<kuo@NoDvE`9X0lY-u7fR3@50Pw#RVi#2@*_xo-Do4|YA@BI
znnaNheS-_i5vUFE4f-ozLv7-bw8mjNB)i{1)o-wa<kIQRT#m-O)`8KSH*qA9-zeo2
zmK%`~s0;7M-d0GYeNt!66n`OMOV)$ikO(zj<j9O5FU`um0%_Zm8;7U{)1aGBXWBC(
zj}#A~K$B3mqmIA8bNRC(5MGIShD)DZTqF}iRV!2Xn(dBo4l?u$!{8;#gwpb*oA#NY
z;;oERsIl!kx9&w=sOwNe2>QMEmrG6X#VoD&=A=Bg{;>Q=LWtN|>!_)0sv$odf6}@<
zA+|pjMLq&5Wxz11Db+W;4P)%s>aVAl@IIp;nVBaF9CQ4sv0r?Q2CrvxTVuiO|5t`@
z#``QM=lBj{S{&EyxM0Zf`2$lZoy-2U^Rvf(0LuOs|0P~Ud3b_spO9Q=c<l<N-rh>n
z_1gB!&G6Viw$Up6lfQJwq1LtNF#I_zm&uzkYynNSfdwC%=W)bYEl-`o=lNfVhJd(A
zhmR45FIM7zToEQb`9!#{|De^D-3dE?_R_Nmq;}d6cHRyK<@6#oU-cb5?CcopNp!rn
zJDf8;7ytzpR{1;G`WDAdf*}pGO0sWR)Rg%K7Ki2uQ&RtG3U>w_UplRu>80<$n%JHR
z12p}Q$Y(kpw*t>sm3W6PjBr;9iyoeHNu7j*SKNttiDhuxvm^WQXtLR3hqCD_hb@C3
z;((`P$Ug?rLPd8LRU0tmN3npEg(zYR&=REtGW`JZmb3i2pVT_k9|P}oQ;1*HNQUT^
zQn<j5Tk_u52<5)OBE8+a@?WeGI;3bbdRk~WAbc*TxbLV%dvb206(Fp@s@>Vt%!|i0
zr7NUg=u#y*_EPIP#BSGrw}CW8+3R&TMS=&ZhJsyQ(8sQXEZWoeucVvW@_f*$G%wOM
zYRq!uJHO3b{_sM8^HJ{*as9MpiYLrKKg*W(N5T>4gKXRa95MGAPEp2d)#NTdkO@|X
zGu8jCw>72Sph1<rRtRw8+j$Etu0hwo;7QGJ{AG#<40yWe%MMHct)i^SwfJahdE1@i
z7&1+5;EIbnC5**yX!E`gW_bN;JzPF*Af`74Mh`LS>=p<OnZkyUryf9xyXRea*bWj@
zw=u**XF_>w5@4+vAn*bKqtfi}g^quUl_)OV%5V20p!W6h(C?T)e|ObFE$-3>^d9uH
z0L50d?vro-{Y^XJe(ieu6OTZAma_ltF}i&9Pmo;qIuvoAs+O@#kQ3cBh=+38XB-cE
zG;DZ@G~&Rso|=8Bx=2Ba*cW>JIeCWlV7cPo6Pkqt$s37ns?E&!U0Ot+&R?IrqQ+?t
zfrT^x3I4k^kHa_aKBzjbHz$MnG9hc2kUtrb&huqNx#73TPo)(h6Bh}Iq&g6J0(l^P
zto`2MzAbJ63g65$9TI^wQ*t|nb$v3Vj7y2rehHcGIY*SnN2iVt^Sx_r4X({omK&o-
zC~f<9;X`bJD!~E-8X>^s@QVaiPI`PTCN1btSgBVok=>v1Cpp7|ETv>wL_!Ceu^#dC
z>CB8vLQf}Vah7chnV6)FbgKDQi#AaSGKZi4M#`kHLD)oR(c-6u5)UYUV!;W_e$3&H
z7s?^N!BLXK1i!?2HTjz<1}hJtL<sqpnR#9l_Pz$@;s#F0@8#<=belx96YxxcOE?~J
z^;b-vOrKlqdHorJB>_BBAA!gm$9+WO!YdQ!QZEDU+}p)rFL>|Ak>Nq7cx2mZl*E0a
zBV}->Z|Fdwp3_b<*>{pY#_TB~_g_uIF#bM*XJpJCE0?*JUWsg!-AZThBW83SqD8l@
zRNGUWbJcHztNa;G_CJ+;++x~%hqU#E3;QoQ^=$@XA#pUZffEq;za%+sj{jp3`hTy}
zRcp&T0(nrr<<zkiIO_!W(E5C}nwyH-T$Km2y$VfBHcaA$da~O7)>WkVsriIMN*Hp9
z3=%K<`yqhT+KqVb_+tn-Gz7r^KRn}@UleXd|I_NXs*mnbz#$J-u#{UE>qO?}z5~9L
z4n>?LG@d0~9XUGI3cs`FI7?FB&3ye-ipypYdVhfR=~0#3(x2H$Fdo?o@XOW}dhzh*
z|7dx>n~KBOurDT4x1NRbfF#Xu6YVZUCCB(zbZx3AiG@yO-?Y-V|G+fkA-sTTM}Dz?
ziThO<fq}0in?zY@7iB2RYJRit-z(dyfonmf;W0ZQNM(+TRZX-$>Jac1NND@am9k+`
zhm8{(=aMd5ENKr;vtIE7Girx;anmtfqXk&#ktUmKSTHq+SEJ7H&V2c&z5w{doSMbV
z<o}Zc4o(a=#;mHZl}y)x>D0dfb25RPDoUA(Zr;l8=A~Hfb+0HC_V~!DPy!a?^kE&!
zPRtR3WUdejc6<I<Mi)-G9wxQaP3Iz7NV8{>qd0|(v2jso%$C5rH+Ks|lVP4hxW(YN
z{rtq1_9$#zrIx1bLvQVq-*2o{HC1z^x~4O#&^8!f$hvo-kif9OeNgUbEBLm#tifo~
zHf$y`w@4T(BdeWEBPD8(?MmO3i#rVuhi2w?ZA{HGCRLO4n2pQq*ZlX{rP`xxHSWDJ
za;QrLadZ$zU3vs)`c#PY5vw7!7CEgx|6n@DB=r=5jZH1&mCfp#Ubv9zLm5Ikf=pY-
zNAm+c*=ApyQ6i=)V80O<N8`m_51wjFYPJ)T%9WkoMWUuM5nli5;Qz<iJ2r>b0PUKw
zZQHhO+dH=HWXHCX9ox2T+qP{^&Z(N3nmSL-yz2+7kE^>^-+f(>(-erkCCs^swwxjJ
zL^8Na#$7TE+G@`$7ZHW(!1F0Vng?zMtrP0gs&>m6*wy=TaSiVZ))FAs4Dbt}W+F|A
zbl}aNAkL3+ISJmeaF=o-_Cf{TbQR7ED?Vhc!u?i@Bg)+Q{6P`8OZ=IvEVyuTWl`5P
zV!R4`#S@p*ML4${3^NE5`+Q)!hPt87F&a=ZNM1=yyLQ;n0u`V@FNCgZ&_H-Bp=ur0
z;kX#comm@$K^!;_|I%7~0@#i{GMA$9VYQTiQ2-#t-Z9=$(LL>c{dNMg#JbS7%wXk=
zuxp%TNC|=mte}N)->W4XMGMNFl@#bv&DcHVRVMoDJ{@aS0lc=(S0LJq1^tS>BJixi
zLom;*2K<FHyFvnN$lVMTR_T0RKzY3U`yS_G!so54@F2E}wCrq;E65YgS7{I0JrWRR
z<I!Xa<BzOd*H_iJo68=#^ezg_=Y;%LCzVEVh5>`xdtA}@XYu>7j{*w;;Py+FTt>4Y
zrtl6cLlElt3S!Dha5xOgE>(;b(mdt9<k(S{vr~}dO8pZO<S4G<mvIVx5gam|K&N#v
zj)72xPEe_Vtsz)`<twDs))}gr3BLFf0KNWdY5wK9rHP<ImH@B#z+RIOktP~NRdAs#
z)TOX8<XbQrlTt)_l198vzP!YXMLZ{nq2|cr$n>uP8yvUvIGSdxFIR@}--tfF+Ep95
zEe_b~1-4cfWD<9|s&+q$^sXrAyY*?~-MhUt2FMnYFvimsGNs2HPr=)`Wg#p{mFFZW
zxA%@%Qb`iqH1$&D#^*PRBf_xth>;@%LzDtg<!37?K1XZoV}GCr8|>Z|!?Wwq<z|OJ
zXSWL}5>DuM<gn~83k`WKd1`9L-&IN2(@WiZsPw^97@Na!2;FO_YP+d=TVNNC?vD3>
zwfGx&MwYYCt*F3c!E5jtIA{nf<sqA^Dz#+)k{OR96@P7n9vJ6M`6YsC&#p&~%V0-%
zlxCk%`?as3!9Ag)!Bqzb2WCVYI;dCD;q<RkDls6`raz(b`QQQY#)ZD-F#duj++D?)
zj@>c&vy(<=<LxJmc*w>poRKBYTQTDXK#X|9RTF$_T;DA80h_Cuc{#&K`nv9a`yzPT
zwP!qjULGa>_cW9-va0#~K|5c**>-4V1i5S1Xx_1QDS0hoKRcLNJM}8YfYbKLMZ>>6
zOEo}&6U;{G_7?V^bUHWbZUVlh;w*s^QM2zdom#Dl?3wEWLrb-X`hyBCDa>HoXAU6;
z3?(8tntn02cZ;d=Zx9f>LZwS_8gs}V5<S)L6AMIlpFs9osAlDY#N=kJX_|6{OH~o#
znlGnC_~9Evt8Bj6Ov7+*93fV?0{hxbo_Th`mh}>j>e=uN!)+bzz*FX1%RK$HTHVt|
zP59AQoLGEwb7Y;)|8AR2_lM^;JxJM3pz(Eqb)TFu?mXGk3kk(uiatZWF$5@iz%1`D
zLS6`MdiA<Ecv|l#(b0bZ$(=t$|Eog&r<sbA?SH(Gh}Mu!++c(G-xYGE5>pB3X_VF`
zfh<dIW39EMFfosGeWHw}&GT~k>j@N6A*J!T_#luJ3W(qD7v#pn0mAcowgkVSUc*1A
z7}vl9nW*0ZpHC9!(C<zYXoKmk35i+#v-$n`9=Q7^A|tV9%pvvldUtiR@$zyZOYtTJ
z(Y!qxRKw@<es=KjF=h91^H~7z`}uzGykOVk)6AvkAhh>`uE0hl^)Jq%7E(Fo2-5b6
z73S#jY6V~DA1G#LiU=GQ0dkwTtICgeMd(3G^-2jG?LV7-&FnwS?gT)H4k<$p1fs(t
zncy456z&b%B0?Ix#~e6SzxgC6w|3FQbl&PcFxxUa`!aQ33+=VDZ^;X+)QmiSSl;mT
zf@}M=MKru#0R~8d0}FySV@#ZgfqDM8Za2G~1JM<gSU-|zb+^J>?|El^qI7rC*fbdC
z8&u^Nb#Fo>viv!x!djIsBI%3sPPAR<5CKehNb23lh9t?`tbw;cy2V+G3(lBa;_Sdb
zR9ofkwO||I<vNkpE~r&*K$(0KgF+p`&Y)kQv7<t0W`^Hbwr&_P?bqB4bWq5VS5f*g
z!Yjm+xO>+j3uR30gQ}vr@>QJa+dSu1d=jM4#G9}}A}z(WW*GzVEw6g#HJPDJs+M-L
z(Ld_|2uOX*&Go0RV;j8gH5=?#b3ML@YpSw$>fj#eBYTz?mGu+B<rl-aPc*Z%%Ou8a
z5{Q>E{pW>RE{k2Z-1xE$xMoNw$Y@%bHP6C2kHuvj7D}54dZ$@ne@QgQ_clq)Aj#!*
zoolfx8B2<Pg9rk6n4h5{@&J<p0n0-kF?KQ($^&`jP<9c;bg)`Ztdlq$2%8|!5u(!^
zy{`5xawa%61b|wdzZaQCoo?ijrt0x&#E}}+&Reia2(8yovww<laQHOtriVMS=&(o-
zTi?*1QO?TyxR>aiJ(z0->EtyMOa_^IrQGO`Oo%jqF8O`Kdoq(K4@vYojRt5^D@wpe
zqW_7fOQu$p%`Br+5=Cp!p@YR`^}d&{&<{Z6trux>`RXklNqdDz_SegF*PgT>)&D4Z
z<ZGxk&#GZ5)FKi_ibN@v4<EPzu61k_<3QzjpO+KQ!#I*XSN@&YXD~yz>@mR`6==({
z^jwR9!3t*4L4m<PYOl4El^%9ec0(4rOF@Wkxa?Z_nq4maR8$S@Rvc+?&rU&9&`u9o
zXiTEZx?d_~6pMCm6ZeeZp&jg=9#%2q_P)HwL_;l8u&J1Sg$M_240w{2cGPP-Y&pLl
z^;Yl}c4q><Lh*fb>XPIjPu^kXZGp?Q^1m9|(f}so#0Q)jPUvqbV|^xUjO`e1;F*gT
zqDp@k=z?vhDD)c`Hs$vM<DJ8lNcAKe5hbYUi#1CCPA3!$Lo<>l4&hQ2i)${|z1K=$
z?i<=fx-6?U)<F@K!)l;QH^oJo`iWH9>*Ro*YvBWGouGwTnv0#4PV=o11|on{0P#~;
zT23YL9?>y$+OJm8<JqKjWqQ)E@lu9KPG=3$-RUkXKMseBtAUfC546)VyL_nRb3Nzq
ze=<ET5g3#?r1TKkG+!vml<u-g>w`|nPuhoAMNq0|g@Q5;ZJCzy`pwJEB0*-UtXL!f
zUt8ZodCAw&<hu@`D{erF8RfPaN1!p}RZS2>Rg1EwB&gBk>EDhoT}S;Kb_H>q+$%9{
zZAr<k!+VUC4&vY006q>JuOI%h`K9|;PX-I%$!;jinWpXwQ;~--Op~kOzXo^&8maKI
zLp$I{6)9tQ+#dW^XA7URTHos9dwpia>A+ob`OLYbWC`@Vtc0f6<LifhTVeofI-A<|
zK9q7d(+Lh>xjaG5+BDjDXt>#-j=8n>+42h#zMEm8nCM#Wj+ri67DRb9iNRm(pU&z)
zM7=M9Ltxo^G+|u<9iEC!J29X0nEuXkl}+2-a6mmgFU5)2M{oNWpMtyB8SB+nf#S^j
z6M{GJg2~}WZV$L>K)=B_`D_nA3n|-eHPjLMAJ9Vd8H@fe5uTOIWd<~VEZ;)hC>;Uo
zCCrm(i`D#@y^%tgbzuhGoet*O{c^5ToeCMo$h#o}PC<OmA7zJwN*dWJt^J?*%d2_a
zf-lx0+EL5fNe)#RDU8W2g(~Kmz?HfwL)G|Hm@n#y%dJ%sB{&%m+zad~1>W35+~1j2
zhG2*>(g~_pFX|%(kn9pm9AMc#L)HOIKRpIdh)%6nX&pA!euq_@7j^mzCt=3SK_v*b
zXE<XVZY`$pQ*(e{wnI(t>OkP<AR}=qIZsT45SbFTEz5<s!J6JC{whq69*c{=XS{Bg
zjGCLTgA@bGD_A^EhC1o*^6V?!6*{PMCl2p{Mtf;#cS>vh>#9+m&`TESuRC$(@Iu<l
zh@>a%H5Onc1A}7<dTUVOcreq0e*@3j7Y`c*Utkcyr7f(+14XBV1y2WNQVGlndHsc3
zY0iZH@;ya{(~<lA`lfGIaq4~24B_jA{YTSiPR9}bAN<q;S4$wIGw^d__}TEgZ(n9W
zj(4Ep=q(brNd_$6(Om_cmQjJ%9##uS4NI+GBOmIeL=~@?_r0-qx6XM3>#%Rx>0};Q
zeh_RQw!L8C$FU_u=T7%Kd}DYe`+sx*O#iPL`Txtt&d&1x%+Rdt|KsG`?@A&Tn;quY
z1b4csYphr(0;vHq3Ty%0C=hoT=Nd1JCx*yU8~GRDtmj7DgZ-k8oZD!|+EIhvOhodN
zPYI<cZ?q3oJiehoL56}Z#9GKv5hDO(L|ZmYPNS-fOR<i9U&%tqwM_(hm2wF*XkV&P
zm$DVBDz&s?EeTl+w#rv3MxsiuK4<nZZ-Ve&kT|5NP=FwjsTr9#C8$i?T-=b%wE5_e
zr}1YgEIzXsWw;-xd=UmkqpHT@tST%5im#1aTJSR)9|{F(RL!Y1GWIA2wJ-qjq&v|1
z=sMY8uaFoz9^m98V0N05SEw8*hQg2vDnL|_hAR3nJh<gM8RjyCTw1_C#Zd?AZwi>_
z3}FElQ+E5_@lAL7+h~!7!qS69YpNv&HL#!vHB#&vLmp7vIHE#J>(tBET1ix5^#hLO
zOA?Ym2=vf(u2SUnaz+wXU@9%nIiMIKav&3lFgQw^xP6i#Y<8*b=KPufS_zDrkY_kg
zRBfSt_a+T^jSL}c&#<bvJxDT%IdfyBRE^QX{RBU3iMb+c6o${ZW+wzfI$^I7v^+NP
z!H#t6q)|F>XpR`5G6HQ(VB<-P8z7P<P5{Cd0mC5KV{1II0+5SwGRZ^;C1z4X)IxJ*
zhT_>OJTN*yzzJ)s!DaZsDeXlVQHev#LPjhU>WRYr#v#4|+h8M^2sj5Du{9uRSdx7u
zCt@W3GR(;ROY1uCZcV+m2IdxTxSKq}*3>v!>;7mN<2vsTlxThh&xH`YfF(ouSUlPk
zdO9|I-Ky%%wm;qO&zrP*HQm~6Ezm4$+BH8PZowgLzBOnQC24L=-R+H!xy!zF-fiuh
z+Irp^I<^gxUN0l(2PJOuY<in*m<0MjbY79++d8@IC`Qbet|4aEokzu{TN(U6LQx$c
zw#zFy{JyLQaE{=|a7iBmIWSHezVN2JxW`p4;{A;f5KZW&sR14WCBm}|C(g5uNT=}(
zRZ|HvX%Owte#`N>$hZM?PG0YTY7Pm)vl8j#fpR=s3-EvqkG+$~`sXm+e@_>a4+u|j
z<iN8mEJOHkrLO&nkaG(^@^Gl~&clX~KbcCywEF%W@swgncEMFBBIJm4i$23w0I)oA
z6v^02;T|Nd?}<4+rMqavD?ekR9^?|SanX;$?&x=PgvjrCtr(R!?KBWLUk1!(WL@8R
zcJ-N`Y#c<bEyyvGT~6(=piii;+|;Z;+~h3((4@$-q2sHg+oasOzFpk694$Cm!keyw
z3*=Y&cuf-G>oZ$DQ#FgMh8l|T+BSAOPU!uH{o(N5&X9QdZI(5x_`a=SrVJs;@$K7t
z9YF2^8Zt!WmtypES>e|DI$Rrb=NKVHM#$oLMZ?$rG~=oa)=f<;PTm4M8J(?pNO90{
zxf!5i^|^|l@io7oI9Q8QEFPeL(sdo6_bd^)W-%EZX?d1sB#_Tm8J#ZM`b2B!lMF<x
z5_P7q2r)*`dwT0M$=|kARBDda@R}8kmn7<$fV|-rS*Ox%+uDJl#&7=}RW|C9K7$K?
zE;R3;cEGf8yAZ(245f1Cd_*B>lCd933n;u!hKC%|N(tqLn^fdmRXyNSc13lweA7qw
zWmbcgDMFpq{%bJYb7>_iMy?n1@p69g(qR%f^MT#0qXhR}U*SB1R$nu!(glL0Ml-Yl
zy9d8&DF=tjHEHL%!$qaiJ0Vy)%Qn$6v4u@Ijm5Je1P@LI61xd1Z>UF#i_iO@A-sP@
zm8t-~vGi9ccto<hg*6CuXsA0CWCl@{;Tg71H;O<=d!E8+zHdGsky}lDOiq)+!6>V<
z!)>`ACY%vZtr};*VXhmRJEl~-LyF_(WEc2;yVg+OPA|I{PxV9sx$Y4ml^mhODC|Nl
zN(ppnsx})p!x1-60<c76d_?^0?~ziAz&V=yPLuZjK!@Ikxl8>-ns{qvfiX8ZzV3t!
z^q%Iuz>FC_-p5*r-O9m+POP09?^uIwW8rD!UlsMkQ>m+4k!PF6+LOEs*mS$l@KK#P
zzT}-%hmNn9dZX_PJgFgd)zo;3E943=y+}M0@yPYO8rM*%QL%hoYIthLq5)#z8X|?Y
zlH;0)AK-6>m6rb~PRy+Tm6u~-{T~Yr{_oiKh5D|<R2sxD4lV7^I@v#GroZma(oA?7
z0VHyibA%QnUoSeH5&^|aqplm$^hALBb+L9mTd0EEBC{Vu@D)Q9Zb>Mmy`m|`=1rnB
z!aP$9lAKyLvn63lnfoS_%t&)NDGvRTk9(<08!xGLDIQo=DJmj8x^z92K0Z%$CI?PR
zF_1I6SY9w@o^t`M8IW}MK~5}x*|;B9OD0OLJv+3#brf1t2qG{Bf_|-aoN^v9%b~&9
zW|*N*K%|({j*AqHI-lPvm|n1yfp3L=_|uog`p5Gd_`l>CTBGf6mZ5*6ReM;Z6}^t9
z%9)5r$s=-CSRR|+WSOg=I}Ur{z&G^M97C7xkA+Wb;$y|}zH-wMe1APC!`r*g4>5%x
z0~(0L8D$v?%3KEdY4*QVAM`X~b6MyWAy@OS8&87zgRl$*83{2XkVBA(Dm~P_Ox8@i
zytLg+{x%l4>g}jue{7rAD{k2wbi4RWWL4U3S#@Q3UmOg=CF@diABEIs(Dalwu})s_
zE|&)aF{0|Ta>XlSIk>ifAF;!@rCw#yxrTb;VqphlVO5?<BYBOTknHu9tKT5eiO~R7
z%9!tsJ^6(Aa+-Sd?C$$3vYm_d#iD=7u@!YzZ=P7G;<@?1c?K@Il|vrIT7+m43Q=cO
z^m>W`10>iNW+`WyU`iz}>N?zV%}HprJs;V^+gTuPX;9ZpG_&)J^1wu;>g(7=$*#HH
zw?5wnUWuyU<A7!EoreR)N~OD$yMTch#pEh`#$toKEhhSlk%&{C>K?W&85Zy%C*Vxb
z1zDMhr*E{l7=eebTQ1W(3pGi-p+mD-x#bURik14d5Z0uAKhH|wUo>YqhR68wxU@WN
zI&IpVx(^@3{S{>dy^6Lz6m%)a9BPNI*;pQ!-{=p0;enJ1V>jibD|OdB)nR3h5Rgz{
zQ#B^$e!J%);Vl0AH|M76Qv)XkaT6h!M^5QqS~jSynIdRHR+}3h9T|Is?<q}1q<OXI
zLOjugcdYnF-AxSOYpj+_X8pmyRN0obwq<<(V>Ov9X*cp6YM;ge#NF;Hiks3*tnH4n
zcs96%oFCc$-PVQpfS|Bx)o&doZ9J^%Up=q?2DH^_?O{PshUW+6TpeqKXM4Jk^)pwz
zVM5Uf07s!8po(9MqMAGlihog{{kx*t5;4yCMtFR(Sz)xY1TYGmEMs6gVQDzvvR|q}
zRoV7ON`btNx-+wzquOrfvnzwjM#)8Pf42r)az<4L-lTR1rgFR>C|Z&y1)XyQ%!H(R
z0$;rq1tdNlgQlrcuB@_|`wF;S__MPEN`PdDPsi|dn4gN#;9_g2eoAee&o+%F8~{y4
zUcUJRUmqEVfzx#w=czBh7e7Bjp<?poIJe8vQnID<+r6DoA76L!mXG5&oCsG;g$s4j
z&b7#21Cl8Q<=32$<($yYM^lxq7BeK3)ydvHkq6{JX}LxSGi@eYw9VcZWEa(?opoj0
z!38P5z7OU$fW*NfZU$<OaN<!|jYGPVF5O?Iuwu;sWi(Z(xh!S{c&-#)SyjN}21{|%
zY{NRrJLMJe=FYZ&L;IJLM&(tJUSnY$J$^!M{qMhzR<T61i9k&hyiy=XpEho(Q-WxL
zhPG9!acG!71H`OurwBNr2S)K<Ee~}@F#i7}`Y~WR1xB7O{2_r?AVSJUyO8Dgrt`<1
z#H!MrKNnQszSR23lmPSV#q(){pq<%4@fT=uzr`@V4J6vay04ZBi!hbr?D9WSzV-_z
zuXU-Jy{ae2g#TUTpz$G4$mJDKK}`EKT2r$>#?R`;K%j`pg?6vqcjaFTTYWr&=&$sO
zsNzYr6yE>_y3W}^Q7h4F!u0|nJ?1Gil~kodOhP{^;h8ny&4`OSc!e<M(8_U<!;$k*
zdAHb<pR_@5SEKm?s^L`?|JUdL*O2yq?C+RZnE%I!=OWFGm|qL^-x>ASDyCXB_oX5`
z$qZSgM#=bgirHcc;|we4cvAI*m-(8TIAN)LBCjkRnpz}EUDFU|v*}sm=B)U&4fL8+
zdR<*;KIs7#!#QgY=7Twz2nrIR>6DB$J-}qc@0a_M)Rm;dKm!#e=8)Y|{OzuHm$%~u
z?>-Vp3fL29%=q1{kCvNlZ!Pc7$AS;%ua8jtlpB7ZSJ|4k01e8`ha(p$a4#APR6^NT
z>t9@@B=P1sv(a4mS~d}ZXu2zq4_Si5t9Vx#vjWf}(XF^y(yW36XCez^nS|K{@8raU
znvE*#d?d@Nv5jkyULpZC`|8RyO825uI!HP1GHJ72!+xo~{kFRivUbz;vVy1eGS$YU
z7EV!UU?ai0gaFY1$n+RQ>y({p_%N?Z?R9d^F3_~37~^U}aAw4G&^-3@efpSjir*EH
zt&O<(xpitDu={c{VQdHB&J22HLaqWkt(4N?$7Z)0`)eari-)SN!3>{QZ)$SCSy}V#
zms%TS4y1v}Hs2&=k$=`AA5PMf_&7KcbDuC($3eBKnvBF$FGiQHk#Zf#V$52DbyNGS
znoIW<?}SEZ`B;u^B}df4E09s$LUV$F%u$0i==YY~8(Jj6`Un`eu}iahRXMg*d3~#d
zx|I8<()ydDjv2h~7q6a3P^J<OQeM)X<P_oiK80t>@ypL5^2a<y;U~2?V=a&)7A-5#
z-;8fc*hD_M>>9pZ2MXNmTXM@Bz4E{CJ@QJ~2^1&`X7JSV!IoI_tqZ&&udt8}5R)bt
zid@JtAr>@<etNRD*5m?YvV>BdkWExDlV1yJ!2u+-YRb!}>_rGwakgKgOb6&22fF@v
z(9<Th(^6kV0q406#MHea$AMxnTfPj+<s04w$kE@9%1f^n@boNjm%T2k&BZ#Xy5Igu
z#1!xaVfipiRos{|*0%|$8|Bj*Ti+jw*+&KmB<o%>417aYbWGTg3xoe;fkykpvn@YG
z!<*C*Ese_U$3nt$^%hv74Fq^9Fdk=N`EZ#A{4}s$4$L-doUoC|9QMN<iC)2=ZZU*l
z6g)X5tR)zcLd22#cz}cnIl>z!bD<M%L3HjhaZACwsXm$C4<TSPX_3-vM0NerM9gI2
z!u%Fertx`{7OM<&=Z1w(^7!A^W?)BY56u$U<nrxg-oJ~1!?mBUx-|tOq+ucc`753r
z!MuDZ8?4eUzgnkvZoW^jpB&)#$1I>IWFEBXVt0_RP!PUfct=MUPpNqrKmbTar<~b?
zGZ$0cg+uj6nht@}&PxS?N^6Ld95YB^c3XE4K5TP601*K;hVTH=;Iy|DpWTins@d(C
zj`;@=p+w~vr3pjZ>x1ZP*zHx6sd6iVfWpy#Vpd3=iKdA-E0E2K8^<FUGA#UQ+rpfN
zL%Wh&3p!y)-$Rkak+Ox3{HfG4SJy|IMrBJE_79p}-HI{4nS{QXd4$?fwJ6!@#j;Si
zn#o2CWWlx?fHC286Wia!gna~V1QFs;LJ}!tgdFvV=fDAc?bjfA9}Te|G_+QwDeJ{p
zrut;60Pq2{&1M>x^aFDh^e^>X81?8W^gF^T4OPy*P8Bvt19+AFOctKXH@+qNbK`wd
zx4jhgz#XVPkS0qIJzv<NIusZ?Q7pH_p&5OPWHm<_7XmU``dO^<dr38lt5gR~pdrCA
ziU{%((bqJc(OK_NI*Vu-IbDgjXfQP<H0h#2S8m-`gg4^&SE|!24yiUsw6}nuIV$UC
z)c}<K<BCBPpiJ2%<LJ4frv!)4pOBh2N<*krPp!FNti_}{)sLRm6?eEeKFi|q9XAYC
zAxhHnzB6T0S6jm&_yI&Xks#374QKY?k_V%3Ji4vkJSs!17mMLnl%2|YT61bhw7PnN
zYcFkwidShJ{Ai-Qm2CS0xsQho%etFbcBEvqB^ohOP48KLP+t4Og*?z3(&=b<L8ukk
zR~6fJ-8SH~)FPT8%ida_Dmmq_w7k}?N9rWBU?3pGh(pRY#p0#^<Y`~&^aE8>gaqsm
z9>VcJIxu|?0X`eMh7SwU`_i8RcMW}5jntD?SdgQ3{uk_f9*-TWgOz~Dg4xK=USL-@
z>!{1YQeaXU2!$nru66ggLS2HsmaBt5Cy11}iKxKOV~)Tm=%mm;j#%+hzFjY8ha%)1
z_Vv-7IR7g#k%Yro)~KA)eWlwnd!6}7+HE3~XWaRp$8nsqQ>vX?B=EBCbe}Unidr=y
zcFE)N)}i?a9oJB=o(x%o=)P2w+c^jkHz*)|!C;Q7FVMgy8zc(UW{%-N=8M5uVDc>!
z{EaYv#+GxWX1^gCd`$E@OrGC`Mt_(!j+n-(-{mrkII0{2eR^T9Cyn`+k_*Kc%K%W$
zHg}L2PVCC)K%@Sv%FEif!YZX3aHh%Dca%Y1f7C4dNPoTAaEFhZx>d7F;uTd7I~PG-
z@#i$0k5`b;*9A6&0$I#&%5uPvAST2cNvpsR*b{7QKZ;v{f<%<$ix&?X2@Bdh3Z!EK
zR;6$Hm_o1)CS(2N&L*mIc8Rmw7K0^52N4c8>mUP-V=%~p2vdm)hn8Jfd1lKH`oUq{
zSKJYn9?SriNvMDs^N?JS$jnOM{Jsz0pEe_N5<mww&Q@<@BjyYZl%V06%i}HDyRiFo
z+3nnTPsC&KuJo+oknm#kUqv(ak}Hdyp^mJvhMV#;;m7T1^7haQ0)cG8`wK$NW>*5G
z8|)8|u5-B~*<J6T!uP@=)>{{y(bZDgNAOS~v(wY;4@00j{A`GXt-Z+02CQRXa67+1
zv|_V|Df7K<Q-|e3tBV+RXLb4TcO|e_n+QGlW6&ACdjf+bP2soKBpbn4T_6_VCXL>b
zW;AIxnXJe&v~&lh0`UiSQV`5H=uxT`vbSIiG5HGqY3|u^6;(O#H`P&HU!RS+469-)
z<CW8aV|E0=M*?8@1mYw&62z=khdCBBY+OqFxySZM;RnU!$cp{CR0k)#^^N>F;U5!G
zHC>O4(Ot6+DB{=1oh6Ad-C?9coKRyT0yG+7XT)4PzeyJsnt!f5!Wl%knff5pP^Z9v
z0S#BA7|^|WY;ETWxXWmhMN({1(c7xM?d$k4&AauI!tk0kjG;C(SpDoHQJOgczG`pf
zlRwxt_pGK;rJ^JaD+x~a??M)X(jvFbsoEb{Oo(S`VL0D0Wf@tg?Q@aWV(p^~HYn;n
zUtn?HSz83$&B@b47m}s!4c>SB&ep|B|JBv8|JP8@$^1Va>YFrvySh=tzc1>yv^d7A
z*VJQdaRAsOiPnBw0Ck_n^k_B_31LMN{)MJI*oQmHTdJL7^`;F1x`W+rsxC{Ms{3<%
z`*S<GUnURdPl3B^^^?U+(QF+OMe%c#2kRmkHjj*G0hq3-X-OkLo-Zr#b~~gu{Gaxz
z5(5XJ@ci9d-aj+=H$Lm12Ty@&6UvPSn}zm}c!p7Wj}Z%9icEA8#>0*o`}H<W9mpP?
z)Z9Om8F`Mssxj-Pm#08fL<y7V_-`sWGm7FQ?eUKGOCur<@VkktHj2b=Z;?B<lj7SE
zJ;Yzwf&qPOP)q2vqp1j<O^p5bM;~h~|IEH}T>an5W|LTAIQ~#eUjftvHt*(HXy)08
zeS(mik_HF^*OhrDguCBAk6P=m-06B%K3pn8eG_;NrVqi57ULu<6tq~59B$F*ifaV<
z#}0a}>n^2j2vp;OPb#2GXC!^74O-Q)8$E5(N(@KfpCrD+cIxf#5vb^n;ZY6Rkxv$?
zvHdt(XDYUe-FZWG_jQ-qWtyqtx1g7e9m#kuOZ4Qs+&nqse>^D@nO!KuFSuSl?@IKi
zL?kiQNPxP93jX|U&CfBQ!sq?`UeCgUAs9ye5&QEO@ntb?Q9sJyfM5N36czc+Int{6
zw4=JZjuA)7dRO}8#PiewGDe|<VtT~sD#X!Cb$*b_v=KDRiHxwj-{wPp^OhdS2Ndd6
z{ByVU({hvg-jF6dic+rbaTizD0ymZLn%{vgvfMht3`Z1Hd&l$S30QE=1MFb*<z!Sr
zYiOhSGJ_w6<7v^gN1*}>4*ev~id0?|KXVv*Vo68~PTqV>a$H<Jrwg>EZ-hNS0tgGU
zj?c_@OJbo*NUJzm%dR}6yVC9D+Y-^251vAJ=~4Jhf}qBN-xJhu!Na!~m&9x7(xGq6
z1)%y+7=ZojYbtFFHqn!I_8L|+vf0NLmBxtU3tqzjbb(oKGf@nr;3xw%cSgo<|9;W*
zbkJP_h>uKpzb4CVnev%p)^z1-VZLZ?3AnJIFhDY_lR8ue@biz{huSWU-I(q6J*v!S
z^@lE;ir)okEJYC@5`~hlJpyxfq940u?1Wg%EU{U8|20bHgMxtPYkRjHwETIjK?N3|
ztmid%f?I$6wG*pzbwQ*Ygvl66o+}}#Uz5sV;^F){Sn6{Xju43W=BrL7?1FSa$XZEx
zF><<VFX~Vq>jx6JqKWyWIdp73NbJuxpT~IngtNV_29SC2I^$ukuO&4Bqmm#E%kC|R
zQZCrkn}(|@qOtseKHn?a)tximI30``15P#o85|u3k`^4li%TU!UanKWb3y65B~KYY
zI=D3~IQ#7KnPR#B^b-VEL_2_;1#`-Mln5-?xV5+W1#R;<eKm#v+bD%mPP;v_9o*iP
zK|NRYd3hn+R`!vx5{6}9<0}#a-YZW7^+~{WGYGd|$s~AkWC!^J-KU1<ZM~(5o-^{_
zyC`j9Vo94_?Mb;58s}KxLy+S?9(dRP9y(F^-?rWBy_{9A_*NQ!`K_P9r~i!E8V)z*
zY4i6BE~r*uzOYb;^EBYxn+$K|mC_>*-+=bI>MkKVbP(RmatQT`O2e0fOflKxS<;I*
z1qZS1`gp#tc^aj2ArKdvc|oUTco-PPEij(Z^6keCW~#;crzv45PJpTT$*F~l0JL|L
zlz@Rvf<U6V)j;e1H~=)Ahrerg*>OvN#-OVT2{mgluHCTh&V21g+jHY1fmH`M748cd
zHeeD)|BWPsCn|~Hp<8meiD3y+g<;Y5jchZ4BOmCXwR1m9F7Z>3Fj=mCen*gSk`mZ`
zKHDl58r1ECvUe!9U}p>gQs*~-DnQJcDlPcu9F#}?5(P^WrEd#AD0tCgYu}iMV?WxU
z<HC%eSY{%W-lW2QRi`2;Igm@#t^nC4Tcz^~#4T=Q&(Vus?sI}TJ8)7eC5Ljbv2m}$
zwy|5)jqRw_6GQ>v<#N%Sy(DbF7f_jV3V*d$P-i!@d;CGA!BvR~bgT_Y2&n3Z@N8Ae
z;yq44h-tPnIt<`qIzZaLpe5xlGW7O=TswE|5i!)}`F7RBS&|PG$(i!*3oPKntaNJX
zT!@%`g_S*}&9@oenmYs?6tcbR0|^2w>sT+^&jYsOxKee9iKmm+6xEUXh*Yal^|G1B
zLt;2uH&U$b=nJQMPb)buVeTFs-`BGl;|m*<aQ=jbZ>S;eF>Hx#*sqw@R@m<Vqs@QL
za06fn`kwu2J5{2ucC#qb00(|o)%&%#NXyVq`{j(4aF#syF)f&^@}<=c;EG0R)AaH!
z1xM0*a?hC6`?wTAP?-!OIt#xB?fns0pzXIDjK`;XqfA0Of-)9^=#}3aOlq_b>?BZf
zr5s!0$1K!RS)o?~dXSu!qIZZWB;>+>jL1yWYMKTm;5t?7oh3V2l!4+{Cu1oCTYZdh
zfgD|uban%CBMYyQa9KL8nH+;n<!?THGNZ8)WhaG<O=b69XTGjru}mLPQg$v?A9aMo
zM7rCQl(ErHNkJOy>A$nWRV}^tQxYSHP3V+TjP5bPI@BX{>6W4X1KMj7&GDb*{l9M5
z7&-sP<$X(2Gx4Aq(I@AZ&nCMBG5ptB6%WS@{-jS%HmO7>HL@VjR3n8#H2-+CMxGr2
zZ&#}D@{W&$RaF=P1Zyk$x(0sklCWTHq@stey8DAb^f+um+bmL*%(?h~C9ka=a{gvB
z7S;G_etvbpN40M=6{qP550vBc{r!m@`q;SORCgZ6Xv?OSHOuk&d_DWR`aId_Va?(D
zyePp(<Ny9#;Ez8>jn;<I02s$PwixDNpF|?lX<qH+Y!UstsUCXKrthi7k7l6odvm~d
zbdg^^@ZpN8X-d=}paw0^Hg7D2G@vQv{uE60?&!q3f&vmYBm9=rKQqwdT8(?lK@hb2
zVt=S9h(Rh=`lvua5R|!vOT062W$IQLne8rsT*k1q0IoXNzo5LT$IML)H^Qp0t28xE
z#pCwIr&kl{;@;skRFb_FeDr2#@`TXsj=E2<bMvsmSmkgYb`*G^CkMS%Kk<)nJ>6MZ
zb7>=F<HJzBx2N28t$KpT@h^uhoN*Ls9t~nb(PH9<;L{Z7iO*VQU*ed4t-VEwyIYXc
z28IjXA}Y<xEr#CBz=w_^Ke{NYuE2_c0{X`kA<V=pZ7BvN4tq4mE{85oc<A(=N>LoY
zbnC(1<xNfSKi;DwuAV>a9}||Cp_%rK=Yk3L%L>N+`{>h4`ta(_HENxlD-b<gd3aKp
zE{bnjzzbA~Qpn`_ygkAL$qh?&TFXbv6ZEDRNJ;xQD7pf{wrZ6TUwml<G9E^LW_WmK
z#{<<GDHSWW4hqGmno!kTw*C4Yfd%&Lkl72jP*KW>l9b`c<$_4NMBB5?9c8%$ufS*Z
zU&6~VcKo^~IbZeLbB&zouHG-IQ6VXElLaZrMxR;gPU9CSh{Za)H~wNd&zo6dB4gpU
z?6h*qHeGvUO6B>m)y;~3MZ3K(D|>k9S-(%Y#cK*O5_PL^TF)@pLo^1`K_@ybaTwqA
zQBA7!+=iwEF0>i)WeZ&ILKnYlXC|P8`#I)?Gf99dWdi<ioBf$lSf1wdn;s@g1Td?o
z*=iTBLhZ@Vi_W<q==Td7JI<Ipz{zFymUPvoAWh}qUHxNtSWgCKiV^fDw6<9=sRuqW
zyA?k#ndQQ$@E~i#(vn${(Qj0s+l5C`$DT*>3I_PGG$|7hqdZB4yeJmcH{8KjYG4;$
zr8tRAH0BMWzL7e$2Q>TK29|J-Rpw;Gvjb2?YahiFDh?Nuyd(<n^Pk4l{KE4P4;ZVT
zYl(3Ladh6NXX#xhqy<SU2<kt9vPsfoBG|lX5^TCBNbvaeQsaxi{`QJ$Y1j4>L@FA5
z8;A5q!vj>4A>92I@RVOJ(VUqAo(7<K<zo;8v8`vhdr4stNZW$YBorlsb3u?Rc?M7n
z!w6$LU>Bi@$KY~RA-&(hO5Oa6f@F<9S8xWbdLKNG#_dSkcrm+SPeDv+hya7M*YwkM
zMd0y!Z*^gi3y}FjK<*t5=>hgGn4R!*xJ;f<bl+&s{I}9gz5oIh*FDMiRb*N&<0m@;
zZ=W{fKUB?@2f*8Ztbb&<vRk!Pd=Z=yjqYmY8=@Jz*b!z;sHdt=M(Y1zq4UTdarF@l
zT)Srr!<1+ht$pT81{OPw)F|7Ytb2mbv(k`)D$?Ye7f!w>c?ka_U1Q0xi2-C;y=3&I
z>Yr~XtX;2O**HnFAim{O;Y8W6fgJQ=J63Pxo>_PN)2L;58Fxo%5KSTj&$bl=>}1~~
z*bQ)$lZ$XlnERGY2&%Zw>;@s6V@rWr&iYfOxyX*=J;J_#A-BqsY$_13S1}nBIV2+1
z$Eo4oZo&fM)E6F*#KRjTUu3zvfzh&HMDUQHkTpqI36BRj+Bl04VdWg1TXB^KQ4ZmG
zS!UNkk3UWowErL$=U^4ybRE!$79nN(XN14R6GB-Z!$DE}#{2mI!{4(>JcZS^M>ogm
z+MZ-7S3s}AXh+%!1Eas`9OH3l`9x2Ms0|_(UH{}-VeuDO;~{D;ySQ69AUPU$uowWV
zC2K!G{zvZSS+wA>MvH+B6t*-b93dYbOX=l<mMZ`Z#xobtZW}e?J$*{(BwBi*G1}9M
zCvCjh!e!ZyaiN~aWTe7K2nKGgmW^x+BKTufldds8fmYNnhA~+1t9VJkNLCM=WN?D6
z;^^I-A}7LJxc-cj?3hkC?BaG*Jd6A?r1H<eAH0=rBL@P)&ZayfNL@X4DmqAVUap}I
z7E9oAEvubHLd#08avZclNl=B6@I4z%F7<Ix&ceV1284h+<A-H{5RB{uuYdY|TnI`f
zUibFvz}kkWX{Y1T=&;z}8Vma?s}@^*vdP_~p^4q#0TsD}SviyjoP|~itfZH9ArM}H
z8JEVJ#-lZokTCkD+O`7_4Zv`>))pfVLv+Q0#hkx5Rw#>F@UH5fkr70g0e9zC7tEv3
z_<yZVl`d6Mlycf_yJ?Pvf5t5z5yl9)?wu#h<N7U=%=t?Dem3<@67*S5Nt8qjQ|#{d
zoMxdt!d+|?j2deL%U&k_iRme2#IS&b&ax_^JJ3NIO9vu-Y5!nbBFUvCS)hG$yEu`l
zz^ntLTA61305&<Bm?Ak9H=0n?W$IMYnuR@V&9Q6Di<Coa6gEv`kQHdVX&76ZF)7ui
zMPTPd9C-1g#$J@zCjy%V6p724?c*eNCVqN6-UlNxp(sk<N-hQ<K%cn^IY>!|{gF#|
zUNGrRwto8or}O#O<{yNGllRf4+VX`BL2a=IJ|kEHmMZ@zBRD<k_f^Zrn;vuWcPsy}
z9Z@D>tM%tM0n&EqkRz0}kOGiGniqDQ^V*(TKH>ji|1f_j^LjA{No8qXW5?5an)6<&
zOX6v*y<Yyo?}q<SZT)Ji#2;8$9qToKkG~bBm?AY(c0>17I(70mMhA`m&bJqGFALk>
zDHY@Q&6;}n+CT4k4^!IvuNH^nKlW0L9KVdY|NZ>`-<3^i{7KAUL-Bd3*{yOPHYfEL
zD3rG#%djv@Z^nMHNE|a~Y_?$J0K0MhxbSiY4ywzt)_pgO^LBQ&eV-<;ZMmy`v3*`}
zbaj%pKQE$)wz#BS(Z%qhi9n1=Yti<2;UlCFd+-}t{ds!2>D$_1pYRaCe75`X{xN-f
zu)8_QN$J_#6x&k}A4D)^w9Zql6@i+nU0dR_vsK>r|4w40N(FV*5kOA5cQ)^}?{dRO
zi0Z?ugNO52w)9#aq59<^rq_lNL=~~F5|8uCDD)%QwE;ijYbp8%P&~baa_;$jX;j{e
zRiM1b+SIKbpU>o~#iB+f^~zl~g>B&wp(0(<w*;#yraUPZ1iQ}bw(;rVHf-upjOYJN
zDGm(J%0rY_1j(il-q5o09(`i2P=Ax^K}+_*(mQN6^|B}BMu9i#bG1ss!9J5hdgD(6
z%^+@#oi3?>w*;Xi(iPyyXQ)wnGRQF-dT%RbE3oN(=xW*%%4R5S+-61zTn`%xn@~zZ
z)Sz-v+^kS22-UWOsM+GX|5_a#4@>JojWx}ijV)eP$>%-SKBc^Ew~UlY@+=LY&G}=H
zv}nc6%qH|EC~@jwPXedfr%A6AWq9scmu=4jp-BP?ETjmLuEHXp&{S>0=rX5sq)7=U
zl$s@lA%#4i#_og1w%yiqz1?NMv~hGfQR&1ik|t2UTgo)vY0aMaU@$3=?ZqPIu#f9#
z$NO<m#I~*^f`9`93eHcn7@7HxJJLdqizq<oS~O{(j=lC7q7^XI1B99}cl@2$ban9I
z-4<Q-s%2;&o288Pq)8CBasT3JYg0)mRB)GqXp^=s**w^5xksuFTW6GBUK#p%-yUIx
z_~GkAeg`W=7*5=Q2;#?^uagJl7op^8bM|xLt=P2CjjoV!IC8^ziJKCOP2%FuE+Z^5
zU=GkpDCg1x&4^D>o4G6i<)_8!nZE?xx^t)(<B|OD2ub*eW|Z7#8)u@K@jQ1w6gm$x
z61ev+^Hsc;gLla~_`2J()WCO<Q5bGo$;RcIV!XioL3JewHpY@Q;{c9er;&F<o8GOm
z@BA}4gAj>my3Ha~MSVbA>K`%a)97ohtiYzEI*%Wk@+N5r)tNCGG4sdkc>KBIZP0GG
z&h|0*K~{98i-d9dIZMHhdh|uop+S&C%b0J0I#d(crz89{^jCuS0r4;kwpks97n>su
zNPuMb45L7PjqFMzI~Qp_jB(cbVy53b1>|FJGun<BHGxuhUFFI;M+j^#fC|&{D#P-M
z31uvm|I&w1h2E58$#Y@ibzzw=c&!jjU}8VKm@9pKPZdp?$_2Z^5(c2v--4C!aR^}8
z7Br*n0Z>_jX2tTUs@N)~M6vLmWpLvXaom%OH<SIi*+Qd?y5kcDBb(%#_2n3UvR;7-
zSql)^5pbj!7A#kumJ=Z1r~(HNo5@<qjNEDWaP_V`hH=s6t`LjoR%4fBKtnYo){ep0
zJVm@#Da51dFw&%+$9a2jxd_Vwhi%<*vKz3njeK*7Jkh(oEk8t~xPpW<FRX%?jlr51
z&Res;8-a>0f`KH#s#}({;B{+<)N7GXIp+eUEycXQwCKspqkK(g6LHTA%o04(U2Im^
z=Q8t_V<ad9^%RdXaWTfuEQ*wcVu$gcpK^R4aJXnX*wn}eA++Fk*yvgw?t}3G$V^9@
z`|xVYBmw1A;>+V6`g=wlquv!^%R5NNSIw<wauqRqK*+Y_q_*}fE2f?_cukj?@uiG>
zD}fm~5tt>l8!m7y9tA|muus1@3@Y3+*`nfL#iCIwd3E_k7!%SJ?D~$g8T3YZApZmU
z2|bfNxw>kP-Y`MRxA`1H)iXz_{y~8d_-UI1DFCsgcd(n&OZXTpf90qVA6O8ZD%9qg
zkI#Ko3>q_pnn7>3c*%J}Hzk&2Ow8#r4~VR&)d~a9ELa=_7AfJCkj_a`aBr$~oW~M-
zALyilAy#&xsM;%s^X66cB&y7h^M!orAA5z~U6~ltPR0gsQjDS5<jv?sN$^`mcb`5X
zKhRT)NgW-Xjf5ct>U}yYI-iYP#>Z*n#-L75UmJ3~{yYbD+D0?^W9p3~E}+ZagI(<n
zmO_|bR}|!}$5I<>l=S<a;_%bz@*gfzE=YY&K(!Q&V+MdLl10|_Q@0|BY&xy&q^9qA
zkRKS1tK>pQw2qDJf@NSUi~dD+c(W`R_7UMjaBiA1ZfrL!qPN(BGZ=6@5o#Gk5JyK}
z(donGW}B#XSV9=lTNSJ6f8v1QoITTH#MDgPFwbrU8M2!E8oExTw1^7ucsgxu4j}Ve
zsem7Kij;O((zP^Q=N4Luyh0w2hn2cH%)fLaNM$`#%yX~&t3d!5DRE=IR&of>F0U4M
zUU>hch<p_IB3PxlEeHxC?}W21o#}7j!}ag~@}V+w{@1F2gYADDKF`vSj{OaQ{IByg
zir)k761))*^9ZynQq9K%6hPVqE4KPWDaVAX{}uljPjX%P1HzYP2gg6{{h-LwF=PHb
zLEZlD>-BP8;>J3mA<3c4)WL0>KWeOsteU#E#rKiMEXU@%^~EMX#%GI}%=6<kfHCo;
z|5VMIL95Qv^B@~|JuJ+h`DpwB$Gxr7KD;E>GB1BqIBKbc-f91#sH9%YFr`6;LP;`F
zNF|}z_e{R}JtIq-sKYT)@&q>Pn#S%&H8V31vgGlzbUlFM#np0b+%7p&ps|f_|L6=;
z;EKK>_Xm5~TA$$3q@$_<YZ6)vh))Sg5sGEj^~GIc>#7c*d9cjs#QHJpsq0OI@<t{N
z$ct$BmRQPU;4fS1j@RT?DSRTFUpG5Auvw+!0xRGrzq8xi)p@yNK=bca>Vt=%`Ow^_
z+ZVut$v^BZoVr5?f8Rk+C`dYm56>=Gb_!k$IiZ2t@;s&Cap%Em1zCF7@+>|Nj?Kye
zkP40o<r{#Be*`9qc}M2i`|8NG?^AN(G*?%BZ-6+VN2U;NeGhm08C#&O**B&n8CCjJ
zzJb?cX`@uPGeR6cQ7_XLpruRF*nH$}e>Yk@${-3Vh($p9Ib%Hg=OuH*X>ex~<sE{;
zOBbp9gPDkQ9@|;Xyz=d$&2ax3mt@fQ;d_NTU#-Bi16AEw1}6ICM}Q^e^|z(PheQPF
zK4X2@0+CaYyJ*r^eSQL2uN;)6+ov_9@g6*{wt?ECC(a4;<OWCh2|>bN0>U;S#lkl|
zz6Og#<$8pldS*)LJEHoHx#~?ZgoKbYc=eyx6QIH;orH^oajmdR?s0ZP+c!$lq<((3
z*uG~i$>(Kh^b^1<G-ak6W-&o)cTtFZUzy|DaX4jsJ9o>#p%PMtlvv>rR@XhMwx9=g
zc@7ru&dt2X90mn?Obl$^<8r`CV+f-pP>|sTXQ)g&k}!_aU#%JyTwz5)q|(GO=CU&6
zP5MSMyqj*Zq%IUKAv=~vEjnf8-8c8>$?A|I=W!UgM5A59BJ7O0wr7k^`bri;6H3Td
zB(<Y*oX^E*a9wykU`0l1r1<OD2~0_eNsXuc;o%Xsw9E#P$~&Bp^ppdrwG)any^%#`
zzUS8`ZN-^AfeO2!*Pz@r4=%OUurQ;Nv<qR56J(lVtJkgbuZ+|C3uf?!8&i&F1u)WN
z(OuD}xsC=-4Ebe~4teqHN31W;tv~0Aptag@Gxa(ePEl*jp|K+G|0;MH6&fn=8b9(<
zY%#DV9FRtvl_CY7Oi0|Mv?-i!(?$TWPXn`tJReu>qAx)P{;0-KF{j`A6Bkg+O>Y?&
zM92Z4Enl8c!?KHk^^`i>WzS`OzljV+1T3-#4jLrfqH#f2l7(5j*$<^5<N*$9)HNJF
zr>Sk{UxBo%Siys1!~#|<Pkmdnl>@|Dk3AJ6tLw3+PeT$SNQ+Tmnjc;vp#C!j!nXR9
z3P&t^l=|Rt&<MX=&0D*~<H<FR86I)Spd@Ih@1Dljx`)|dhOn&rXc)@Z3b)v}3Pj27
znLX1#d&s+zIxD_l9sxwMtpr=n91~vtVy?v@tHI_-OOjT=>t3x33W|<{K`UE6_f=j0
zX)c%7mh(qwkOV~VK{;nx3}N5va%aPL>bq_n?s|eGzA|kKT_Qh$62A4kGq#39?pK)^
zH8uO}dGjkE;i$O-o<4atSOdY3n_dX5&Fw3hX6NS8qT-zL<=xGvH`z!}2%xX>X|2M*
znasVxIHIiB*2@GWEQlf0+~2XjEJl$Nrb#flA)=1r*BP$ur+W<PeOO~jfRMvhvl@98
zPJ7j1SN^Ddh6hust4yhZ#5#<i?^~jSwC#&h=OLK|m}ssf!dd6#{Kh+oH(ZRS?+yFz
z_Y+u6xC+whv<0PcX9Mnb)y5yE8Jbs*tH=*~l%F6F%fV=5H)3AR`WLFZ()orXD;U&b
zuNp!iE%&vY#bufX|J@8!ek{o;ne8ZpC2$0B4j_Z!cNj%};4UiyTku68ADqK`ZdKIk
z=yi-oMm6okgy*)37B`5ed*D02c%A9<<I4`oI5^Rp?UsRcdMNEpfBG6>Z}}?5v5Z<)
zs7?1I#L*9(xQNgm$tm$SI6YAWQK?UsQzKcVzl%=I(ygWC1ciyHCT9)L!?qnp2f)yW
zh4sr1%be$B3vaD0?yhTBY?qn1xJ+c1`5Z(F!tfAlHz$sryfAY7DEIo<!XUX^tLk~v
z;^Yc5umtV!V~FNN?AXg_TCh6r`i>36#{GI^?%LX_B=j**de;BiFu<7R;fr*#@5S?|
zF&xtx8Li)l%{6nUUtOQ7B-41%B&hM-^Lsf65=GBe7x0|XK0C|M^=W@`PW_beN835W
zn@57R=!Fa&n=ztDDDDe=h?X=U(uAUC4dR7#ta+s-v>>a;#DQp>{Ea{}?2Ce&Ca0g;
zH<wi68$UmYgp&pb3_?&4m?s6&NorZzOqI6y)BEL|-+wS=?Z4(=PR{=`pUKF~^gqVc
zEu~4?9k8MJ-Mpc4jrXSQQH77l%%Nb&{2_3VUoa$?kf|&qD$=U|{NSUj7Zu%Rl4KNx
zGj8o1X}Mh}aInYT-A!$Xjcj<34lp5z|Gyaf$L7r1HVhDr?cA|#+_7!jPRF*Lj-7Pu
zj%}MA+qT_NN0XVoYig?IshYQHeOQ0sI&~<<5<qVch6x!g3FCrFuhE~Dpd>bH%SejP
zSi|Vw+4KH!zH+ga4TmqX;$9Qb@#5>`#{Xt>clMstSir`Jf{9AqI1suRJzcjzgc3^g
z*HH+JQi)ueSSlDMt$!bBnaLb2{A{NJls^$hb%U=+ytIl%L@Ob%CBn*XC^89FhU;0J
zUoqQ0KW7(48tv6*-r0;D^E+PI&O_UcxkHj$+GOZpHBF8Fxv~_SO!^4v@iB(Hz`C=;
zHO_Kn+`XY9Y>Z27%1^!K{$2yw<%Fbh-zmq!rdqFZQ2NHU4gjAbWTSSZ!>l-<TS;pd
z+s-^&v*29YoSsjPBfVjECBaVuhD!OaA7B!+q*`*VTlueKJzIEf+HZ;Nm4)g4NgH{{
z*v*tG2HDl93fjngsh!`>Y@?x5)h(A>czKfE;l842dW|PcHVmS;+cZ;woya!#q}Z-n
zQJzy4?}$>DGXv}A6~k7{zf#$)1nyY-75nKTbS)&0$ze-ZB2Ri0vumCz`RpPDhYlRr
ziK?YhPH?L^%Q==HQF9K8Yqd2F_sNP7v%L8#ontGj@CEvd>kd-8*iz$vK!=sz1RR4&
z!eOyn7|Lzj_58+AXU-RLto=bxB-IbD+Uc?gUnP|e`U~fSBxzcKxXD_2rDOPHP*2l)
zq)O5Z9E|4lP(ZJOVJ#ZO^s^2?y(g`BcyI)U8r^NOzAP+3RWTZHO?7<psg{z(u#=^7
zyrg=;_aClOn2FSnR@KtFC15-=Ql7x-H~*a8t?qd+Kb~*T+ZI7Y0v%CA;>Z{UtxqQV
zr7#yMNQrdM&T*~g@i<of(pUr;e*glq9QZs$C`Q`GVI0{yilTh+5K*RRey={$HL1yJ
zyU!Jn*eO(t&HK4ZU8$beyt^@oBcm`jJLOU9FvBF=x9lB+ee8e!<rLtcwE4S`o7OG8
zkNWK9(4?B}cRFz(;snAKTLMfWXoH9Qw6~JSd6R9+!SYF%KXQeA30AdQ^3tn{U8=pA
zPUjvnEb3xwe@Q@b@7IPFb1HEV!5Nf!z@quHa{Kf=sT+E1!}|kV6cb72kA54j*)vft
zq{~tefHcdtc+g22;M83iqyl72qJ9V{^P^!8<N}^lSY8;yUO*2;-0_XnaPa~35yAx!
z#O9F9AT4CJ0v$o9iVDO=?<unX_PB^iM4E2$?<EgbZDXR%0P9<pG(S6jkY}4^Lu=Z$
zqg0_ukZK!Dv5tZAPjGUA=rjp<R-oI;6%#4uSwC13&R<y;4t}J(S0;d9eNvrh);!Ed
z3P#n1K3mXqKWjmn51tI`L1Gd0>t`jU_=&9oo#Me1vK`mvpn;B{eOt{`Sds+Vp#BFP
zE?8Ab2nV|JMK`|>-HjfuOMd4knBx2#b^7vG4Rt=?@v;Am9NSg^6C#8`kP=#s1nWbR
zIGx%qf;>X>@%BI<LW+J5QE~f7_^<ix@7kX`x`Gi;_YU|zVW>x{@205%mt3DgV2X<i
zppRhzN*O*`9MUtFwN$fDp|hHRdDJ)v?slSZ5VjHDeqLo^XfA-M2N!+=5Q(!be}U`A
z9&w@d5mh6I;hCY033-O>Y4;83<g1Orb!k`hpJ<Q7MtleVV&1E2z_6q3zg}-zSR^MW
zjJuQSIXp?<n~O7XK7WJk5BNI0U-|!0dfAx&BLoY`%F6t|o*=KjUCcMWPmuptM(`rN
zL*bBQt5gT3E)V&-!7@pL8IMM(R4$|Z{f&o|OpDb-8vS-1qG+I0q1UuW4H0s^$BpsL
z;Be>5{uPEru9%9_-mptNEYmKSF`?((<l8V6SiFC%V0SmEmc4B>3@IrXf%(GQ4W-~n
z8s~2yVx1y2ZY+4vN1kLTN`Ui&)BW|E?3=DHYkN=kKi`MJ9{^{;&zINr8p99Q4}*X$
z>niw7ZRq?^FVsM|65&yT^Yk$4qyZC6+A$j&QVLg7&R-E@+XG#`YD@Z^dIMipYnY9l
zP4wp_G7W_;Gs2{NLTtO;f3fh5qFFJq-)0;xT})~6u4|fA6Ua2W(M@BDkO-b4M#0F*
z@4)c$VMK@oL3H?%HWmF<^TUEO#RMb17{+wjEp19=oQL|2C6^wIU1^lXTEzO0;jCCe
z8~;!mgeOE2N8HUWB`YGECMS>v4+`Hx+#)ndry~w3(^>YRl<ICbNmk{H5KFbtE%uAD
zNRr>HFKCFTMVl8?8Cz-h$Ioj%Hqe`EJJlL%ln7~v%J4Dbz)?5L*nP@13o<si8XK}!
zr^{aCWUdXDTND0TS4}7Bs)ci5{u4?vuGyV&8HwvEiSjpV{lME5xvW%;0ng^JtLid3
z7WSo%{(>jV+-sR{!*IOaFO%6C`i(0#G7cfs2jNFQOe;*IfPlPJp7>jS1${G-Y*q|g
zWn;r>h-Pbr|AHTYw_iTFV<Fv5LeTu1khm(abYqv)S+`t}PshDASQbqTh>9fUv<*f5
zji48S@)MJTQBHzBRL;VSWr~-=O`~Xw?gba=$V3&@kD$^|e%?8gVz7BIe`~NeevxZo
zjN*itU2@5)=s2P)coCn!G-(~>rBZ52*^{=MTf~Bl-3SR*=_`3XAZU><nAvS3RXo$A
zLzT`T{fEri%m(URbW?rU0GLwnG6Pv13#6ckA%9t6`Bmm{h+Uqy1j~eUW(%@iE*s_S
z?SttueG9oQj>Oxzk|#X^DpGMXvp)H7x3kLwLpjS-U-O5Yx#32~+z0`cvo|9tRV~?l
z3m0EL6!Y(7KfTYtGK@sjaJVFBa~Np{aT6IvnG+fLqwU#7!KAny9l1z@=)uN|CpB^_
zh`78Lwpz-uEr%#ee*v%-Bv6<xf058<*0JSBKvUw#Qt(M`mGX=Rep$tXilau6w!r`J
zb%XJB6mFe&pFzQP6qnrSBrqTM{(bDd%puMH@Y2|^Q+?x=Z$uCQ0-S7JcdD8C2nI5E
z^He7lLUs6;DOMQLawYMb4Z#<4r$9nzO{siGuOFMLTje})ib@?9k2d6SpoZo2mRAsN
zv#Gl2n5j0ecPv|pf&{=!SDbI_cJKJMt@7DH&QVAvq69Vju)=wH;Aa?I7Q9t$l}P8h
zQfXJyTR1=wx)$)iV~nA{(nTo!u~ZZ~@-?VDMk?zbOZ9?cbKibzUb?BuVofuFa+z7q
zsI2D7Sffs*J~Rnmo3t_`8ZA}B<{nP1I3XK?!crL5M|F%xJzt5o-Nq+Z8Bv0G+=tnK
zApl(pA;FVqejxGG6}`j%n~DbNam2vDb1W$pJ19h8r`;{m3@#@4tH#t}C02fGy19Ty
z!J31u2n`49!hXB!*$MmjeN(dyE9bp4)@7M)Q)nlB1*;imFA>LfW|lOrVnJsJ)ypcH
zYk<irkz_HgHRAWtUreW=&k9UqFy)Q~QqeOiM}E^oB43i(U!+nmH$5n)vW}j+Lj2FM
zyNX2Rybl{@zIV_wGJyXIB4h*pm%Sh>*Z(X}I@0=o)Zj^k8#JO$MjQ>JsRZmTQBFXE
z*yuTc2E~l(^_dr7CEjFK%8-dTm;k+iHuJ%|!wrnHC&T~v#$@*|-;<0mXo|>x&6!{T
zabOuyk4bd^bED}>Vso~P<Z)Ol4%<6=2Kry#tXZl-rd%C^8<o$TxA$q!Sc2`)M2y>v
z=H#)pH4f$hjN3Ve&Kw;5kOLyY(9^->PxClW3gX!1DgCID<R)O!;-tbVVr@~F9>jI&
zG=vrbR;IC$T#ZD0+vZCWyf&eA)Q~Po+Ac$t-$$tIunIz|O}s3sbmGKo*pFR=Ot<!P
z1srlfC>kLgXXUTCEm?C43B5Y$Ms5AJHu+q2D)166;qvtb=hkwA1_c~pxSKJr#?4~g
z7uRP+!ez4__ffaw=!V4(_jr!SuIHvY4d@;E$vPn#-?*iN&kp%2#)Nppgw@fJ>O7Yi
zY_mChz28whAl(`D5brdVw%S&u!zpgp@ql{rzLi%BkSL)D5hX0|+C-KiY=RI0kfTxV
zer#_e37E)@DvrvMZq{iC{4Ej(>xE`*Z5yH~jT8Na%(Ku+HVEz$SQRB{==)|4KL~OI
zyYntO;pR6UY$rpvtzXr4905KS4POoR6%}myu#jcB-YW%ZI0bj|E+;o0!p!8;v<sf1
z7gaq=Z)OL()k=gITk96>^N)6|?kDC)0^w=X#o7Y|VNThvLre0l6DjZ(<Z5iM$4J2w
zwx@r*f|&gjG6I>h$s>eSpQqn@mIY9$GDRq<jrjn42E~cuU|U6K7&56NA)BX#@eDyA
zGCSaA`a=7>S2wN~RW3}fYucKAygBDa(ur~<+5A^@>ab{Xp~3QaY_~*OyKf`u&ai(@
z4YjPPqn)=;*3kyaPSe2++DlRN@yet%g5bak`C=PergAq?D*}?xNl1kh^KUJs$fzPf
zg2)jeV_Cgg8@m^;A9_ZG8yVnb$v`~q-_H7$!W$}(Ak{Yy*V75pZ>2QX2ON75(}RV4
z&C?W-8a}Q6;DcWAg+nP94Lx8xRk<u-cKBs*4=~mZ^&5Tj9-)T%MfQ#I?+sd8X0|$e
zW>(w2wpMZ@w09D8w=&OUe{y)~GCn#!YL~(jd?uhM``{s-*bvV-eFk2lomm}L)N1Z+
ztXS-d&S)ax#x1%LO%XDUTEaPS;iC$3&=ma6!haT>W+IEN>@T%Q%2BF;kvsW8WrAl#
ztPjh&Uw#ukI4~7SP$kv}8iux~fZu|{W6QegMim1{i%Jlp(v!xK5&G?MWHms02?D#O
zk3CYQx?{J_)6_|D^j5G5?0{~8E$3lvhl;9n;J-o^hbYV=+)m)aG5MS@e-?I=lm&Sm
zfeV<vH`g{#hEy;HRp=@1Q+w)q3!u4$z*0uar7#&4|DaccE3=d^8I`yWGGtYQTXt5;
z9twfJowNgWyxOXf`8c2t&oNRpV+8ZOBEyzFl)o_He2#VTk8h`uO-RXQ_=rQlWZ0-;
zV8Hi?q!VzE<3xFT-jFf`U+Q})O^-HOwT<M9x*Yxe=hz*6uQXBr=HSQB@hGZ2799z6
z9$~!DyICl5<pBvqT<fNC4XGH&L;$>QH`_nR%O89YlMb;ooz9)$g$&0lwV<JEd}CjW
zxI)zQ_nqbljkn&i{VN<R-Om>bSa#5%G}Iw4=V!p}Hhp0X`p7y9{Es<Nhy=`1Z`)S!
zK|zS|&u!NE_rC-Ch#0}$KJ>tMi29Y_5ZxJ3354&@>hBjstVG-`nPe!LWQ=57_O&XD
z3{K3Y7+CAu*~3iZEvS+L15T*Z(9&~X<VUwc50solR4)PrSx^8vIW&TAJ=YuF_PyfF
zf6dSTnZ(1w_CG`TS6cEe-^fDuo93<5lp0KrG~ut99LfbAf{e2QtI_X7xM4L?_V>D@
z&wq^V#NskdRa|cKq9f2j#BHm7oj0I1=bJo>UOrDed_Ha@KhYC}T&_alB`qbT<;&C*
zxI6^f<sRk?_ue$IVxV%&1jyvxjD8#ZJw6$8*n4$lJC5pzNuJZPI{y?fiD^$h>m0i<
zoW9fN%K5JHzUiMlI9~IJJAQ3LuhH>o4i%-9*0rqG-5V~ksN=@DJd>zVr-9s2uc2do
zuq(S_Z80u&m=M>6_(0AK45!Rpv8M3_>%&ebmI;YTm#KAf@i>>^`Q3)+Ir{F%-w>t0
z*Q)f=D$Ar;ef@~ACc2r-t)Py5R}Qn-GDgSjjs#yUx*_PxNd?}Q38zz)oo$%+U4wc(
zlrbea>5>mj*}n9S#gFjozoUYE1BE8wf`Jsl@D>P84}oMa$%(gpa$OaZ`M>Z7nnV>s
zEf|17sbXg3HW38{ysRNwE_zcqH&v-kBWoo-$;VEpI~f;jSy_{lwG%&5DZLWKomiT%
zR8uTTvSz-DqBwkF*^93+&9VJgWw|JXcb{1qe^CZmvIxMn;eBkGq-j}Rqy$>HnIB`G
zcn=ESTq4QAQTxp8n7oDRs#}6G|9(zn0IW6(!{rfIYIi*>dxzO$nx(FCYUwj`Rn>TV
z2PKyn+C+Kc-zFwm*32T~u`*xPn-r=%+3Yr%SmSh<(qgZY6h-7W7#3^Zc{j4?CL@EN
zSNi9Q6GyDvQtD-PJqdv|C$6xo4Q!>`CYsy$S~!J(f<U|+cWPqV!oK&i;^3m!)wEjL
z-)m#;q`Q~vH?wh=zmUmruT?L^l{(C1D?Z-GOY2=W8V1VAOatMf10k34`P}+L*?%Nd
z%oJJ0;TaGenS&70-Y8oO@QF-`GGO*W<asf$mYbEJtjFG31kKyro<T@BQy=qBAdlu4
zi=aq%e{#A#RWdjQ$Tr8BdL?I`7YyX*KI39J$#d)OmuX2r_HZh*T4+2*1|vK9E5Qs<
zMZvkAi~ycT5X?m(#C)p8pa>$C$o#XaX=_Mcz)0O`2*wF-8~Hl}){}p0_}x2i30ACg
za%j<MItiSHhT?(w48St2J``@QR?o+Al$c4rBgl{fEY&mE4YPsa3FW14eD%*kxdYcZ
zP>(MH2Gx#DF}TGF$&}x$>QXE>vZ)LCMP4J^G~JZ_AF7xk(&m|-74m(9=&-oU_Wj((
zQG$})@%oT2tn9}wE^fGWH-m2TZ%pK2J~1#Q20|v@sfsD-+nhUwCT(CyN>8FD{!)_<
z<mM5M#4Jf4d2e*nLUCo@iJ#}4(dVjm7)PP3AH#Ovmr5jDCH>icA4)rh8+KLsB5Gam
zjga?OJ4idl*~HpQ_l_!S9bAh`NTpMzzVV;!oNEb~ei#gUO$EGT>jDpl$h+^#I~5Le
zD+i<mU#1OB(YvO0o4J7y_-w;F7yGW7LeAM|Y|a&+D$jUqOMt{{cc+S8spn2EvraBQ
z9jww#!P{P_vz}=+GYz0wYyvUh1-V8++%LJNM$ZuI%<WggK@I`9J9O^nzRyZJP1|ln
z@}Y4;wy37i>2U^+JHEO}f38OKX~Kx7jR=^;@SKmB&kprG2^06M>c8{Vt<%6<=oB90
znNP@zzqTa*RW4b9|8ePM1OCsW?$o!F_QC%R#13*64-&R)6rF;}2M%pbw7>u|T7Io3
zk&Jo$UGbn!iuy6(PRI`-5K;JNkFV#|+!B8h9RJ<@e&X=(GI*COLYclCfwD1hOvo{k
z8LR&X?mGgkI3RXJWZ%*G<MIA@|NLKk)?7Lq{=h6!kG}qP-{=n0Oh7Zb;B>FN1UcOI
zs4Eem*psL4xXd36ANmsl@e(#EL!4@r1|wz)j8dGr1WvrZX@WAks$pK1OlKvFWkr?u
zA%CHR`sxL!%Xnp($Y;q*>&!W;=$6s8r&8-`u#wiJRgF%!5?KZRtk+hp-Lt*8j4Psr
z62zUvevA^C64Avy1CANOedTfZBQOLHXOkYARR$H64LccTE`7DwP*UzEvUH2wb*e%%
z?Sv|4QA9b3AG1lyBc%41nmap1Q)8y7nCIoMN{-5nj<yfYwBakiBuo$OB;*|7IuTNP
z(HAVt)jVJ-cK8glC=@=Bk}r7G1v~IxXy<I42`=TAC6*ypenxEg^-oWsv<MRMA(ChC
zD8L<jl5Yp%@6UF9X7w!BJ*R|4-3UjRiuZZ-=xk61nxhEG>dY5*YG5641pI!Xv>)ZE
zVkV3csfL40XdtniF%&ep5ty_Py#+uTLr#N9nrL*-mX*{vcUe1eBl3qf&eT2(Kt|XU
zqh&6TyOM$a?OJ(Lj>@V-+b%WI4b9ho@_cwXu8F!DyuvL}vC-H5@vqEypsEybw^m}%
zHSy5pW-B9FM8%xajK8MjIje$W9P#8+XuzNE;S8E3Ah}fB092@9@31t%N$yhqvy^|V
z7tgu`WP3VM2y8VukW`+9)|>dsPY5Oq4pLQLTQOXb3Ibfo${`vYEPGShMyc<Je{utY
zQ+1fck9v@r)Nk*lP9P-+dv3#68xHBB5m09A1ZU{(v@X}$v?f|D6qPB70Q8XLTA{rj
zgCd)3$v6%uYiIo^n;;o^M_B>MF_};Lx4;{1kHCd;Bi-GidpFYvk21@P|Io85_)vA-
zQG+{+^SiAaH3z>S@s*yB=)7Vs6r4nd!!~<tpqbG^yuD#3`^yOA8j@a}DNpb9V0Rr_
zU&k0>`4Lmz?N{nrj(u8(I{cMwI1w5t^@%WVPuTR8MPnDx72Yo?ghGY{mS4eiEFP64
z!6rwglA--sr$6nF9ePg0j<hS@?DcYW5C5bKab9FACwC{GWP5O&ciO4Ryeq*mu#rM;
zqg8X3Fd~NgD%xLQz$b~cpyi1zkl8{95`V#!?IF^BB$dK6VzWfJF+=y<*MLbu($o@8
zJ)I<00htB~GKB>RGIZ_1MTmgChX5(&b)Kuk$^GD&$?)<a;)Vv=hNRP{bSj487Rp?Z
z=*Jz59)#=ahVM}KEMLCm2A{6rWMp&^3%LEe_4bPRucft+ZL1I=uuUkg$j)LR6%gW*
zT{62J=d5bx@pBm_3;)=@wM%A4kYwer2-7xHrCXY-n+#0(a(lLN#VIMC?I0ROg{LXf
z3qldy4C(>5Bv@1f%x%B{85N0(Za(v(He3!afEqK`zSRgo%hf;7xvE)i+iYOob7dVZ
zC7IEZ#6I38Ut^OcN`K$Y3A(dE(K9bx`hy%}iOT-mo5t-{0D7fft2_Gq59SK0G|I5@
zgxe6On;yIWPstPW(IcBSBwTw_U&X%<sBSQ^yAlM?oNIbu82{{kY0CG6;+N>j!jQ?j
zA9qslh8R`(`amoYL^KcZ4;c-Z|G?@yCd41!w0<NxL`=a0V<;QCL^mvuv-&x5>|$8(
zN3!ASlmL#$>Hdj#G5rs=HJ8s65%n9otaD#D&m!ORC(E7iS`$H)Jh;dDwxF606k$ij
z<y-jZj-<OuRsPFR(@f}%?eYv2)fcF1u-f{6R`&nG@VMFk=RtVs|BRgauI%4~@NCGY
z3=AOz48F4?i&GYtJ5SuY?E2U&njJ84D5dEA<=bS7I=F;-O&Xbi04BKa+NA#N3K~X|
z6E)7C-<zG^+X=T(iiDypqGSaX9W@?u02p8LaJmo?@$zzW)`?XLQMt@4je?5yBVD*Q
zfcMq!-|@Yt2sa^WR2-Gw?p)qzXu{ms(MGc7&CaUB!bEa(d?^v*jAC4}C!62f-oR7d
zr_-HIuis)vTwUX|pkX#G8fq}$M2#+w>k2)SnBo^}3TasI*<2}XKXwW(@3^bJKGso1
z18#v8UZaWrl38YLWBZj>v0KJW!#E?aBa(1>uq8g~0o6AJHrcWR?2+g}y+>~12plHS
zRt@XA{f{kKH?4hRy$gkLB1AzjPJsexBrr+sr-{S&#UsVXmSh(hY9KhH5&`ql6}h|^
zK*_pR>aIn11U6g;>=!#wuWq1U2p@qw0w(^hY9kbl501ZhO%RV!t*&j%-s)sA7E6ak
zCjcjJx!YYq=3Vdfr$n{bl&nJy*dfavrj~xECSvt>o3*V*x<cActrsDcFs)N3g)_AM
zq>nPw@N*-rRqZ@EE^d1=)=6DsseUl|C}KX63QWa*k~vRVuE}O>fGbRLV(Hkfve5Za
zY<PF7t8osf-;q1b3{xa{tMHhKnO<qt)_z3Jv{<T|tZgtq)hKa8dq+7@_CjaBl9keK
zkgw&6GY&6n=Kectl^1Rh5<aQW)a`oR{bRdgvZ^HnjcgH%Xw?qY4pQpCUdGbKauPLs
z$W55VrQ|#H>P7oz^90Jw8{F{Qw0;%-1>WWI!JibQk8tEEqt<4FD}<!T;sDr3a&gIL
zEOl0EeiE<FC)2Cr;rIvMb4a-1*`i1#VvjuKu})I3AC?prnJQy-j>nj0O`fAH#Ss#W
zhmLEbRDFmiC79c^@_yRlJYNOUuZYYjquSy8=XNUefoCH$li@m*t%tJNzk`eoO7B0{
z%wZ<;HWmkDHg?_LK~lqmTs5pvAgX<%B#JZ+!jlPd12me*qHvyIZoz!=k-HK5)7kE{
zs?hEo@_Ll2cxj7522=%+;MPXiF-k%WT@9uL{#h+oWCoKY-{K%#cBhF9Iodz|W_Gz4
ztyUSO7uw5UJ3brBx_Q45V*ToY@n%kG?bgZ{8Ju`V(D76|M~|fGnpWxMX-KC%80rQm
ztqK=>iVjh39gBDDEgY?v^7$|_(07_y9dtV^sMVGNGtGhvX5k=9{zm6I+B`1FQvv{w
zCyE=88}pn_SPo^rY>{rDG9arXyx_9F?lK3z@rP>`m_vsS&p}1k9dPmfkU7S*FWW>e
z0A>WT0cp6+Xx@dH5yg>s1kxop#3?B73+dqBrLVvUkQO!7xwRAx(WWNp8iWn1tB_`;
z^GCw5<AUh)0EAR4;0hJ=&hrSRWN-pogIYQA#EhpAW_CL)t$oE+8>0sH6;vikxn+F0
znaJLGqiMUiaUX7XL@|4cgHDLhwBZaGILj^}rba=Xid0a|CGn9Yr2AN{HS_Qodt(Sv
zxF(D#Az#}sC3Q63UHZV3Rk0jaDEi`WG>EcuPW9g1brJn6ndeVDoX!{T?$#$CZ9)$^
z!HkwJh`7zu^#D|W3hzUD9EunkTtpYKO7NuwpiZwFSo%6>`1O)45lONZ_nEg@J7VkN
z6RvuLl0#5Jpx=bM!BS6Du|jf08SCC+0A*LNRb(TQiH&P{z^30lVbkN%$-mvpZ83Fw
zjn?S^eRe)Ed>bxG;hyR!PC@qYt!m8?piMWJ?KC~(OS(r5dLB(E(#<H+++r%2iE2co
zG-#G{H4T@NjT~zO;NR>XB5eM&$R&G5@Es9Ga01?T1m6d}5qaiEkd8NMXji6rU2p-!
zt<wae!b;ABg}RhPC*6*;_G%6Wl|@f$cvq+^yHdPEDY3v0jSRc5!<V`7JkEfB{c4sw
zM_|x!580JJ37vgTr@)F3qD=qwc4!mJ5sMt5+hdAA`+XIIy?)C6+~D_sQ)I9|*@}w}
zamq8Yqr2Gu426px`=$!Q@;o;>iNNlY!io!C3Tsy6zl`kRXT1e`atLU}ML;;icA`je
z?fh0h+i-r>x|q$O+mkZJm~LNn*{ZOC+Lih`r<Z;BLz(HjY83=Vb8bavmQEtftAV_O
z!<;1ac}K;NqHL!_RiUE|7wtlE1B_KiP`*5`7&9<d47`O=LfUS{f(?F+4lUVe@ykoz
zogiilVQI@@7vcfABe~Hw+I#bh0|~!(RPHFGn5y@;(N(n$8Tg|d9CTwu{;SjUyjRK5
ztzmr<vzj-sXK5DOTGOS+yO#j`EgMBMlsw#p4eg(sJxYNjTFiXwZnd%Bc7l*Rp7YTS
z0gufeHg<4@1om+H$re<SvVI6WPw}f&NgKazCwN9Okld>=-oDZ~ipK!uf*y=)f85U`
zNMB=^P_bZ;2$2Z}@~?URI75ihyC>|A$4w3@J3h{-JpGwmiZhVIB#6|Qt6-#8_)^vf
z-#iXbCphRte`EtQ{pnF4c?yBMX{D?Vr>1@MVooo=;l$^fXM;9YcZp3~;uLq;45v_S
zW-s5{*QFOL|39p8S-#2tYWi%P|FO<-|L^PEH*(>!(Sr7W3Rr%-;uf@pc%F)W%X)NH
zuH_0_TMi1$bWG@ytEIS|0{jF~$i<ZaM=xxwIot=({kz+Bs~_%IY&zdb)YtpH6VbIh
zKx;DDs$(9VZAlSH8EcW^s8o-+8P<_V;)kE}t}VV_|Naej=4~7Lr8Bh^veJL-Pbsbt
z-QK-jwC?<itz2;-6aM<hY9%`EWej-vmjm!4P(iI=le?vDau&mCs)wT#_CauFAb1^^
zgE##R$w}U_?$VtL&*~%BVCUQ^-@VQ<p*e-7fmfRW+%tovibm@fZ!!YzrFsb?OE*tw
zSe=b=AMiU<YizcJ)o=`RFa!j2v@2?;*LP}T#B=siRPJ-g-UH3#lnZ~_t_2iC36#V&
zW;3xBH<oICa!%0ow&B|Ag@ng?CZLX!q)O1EMOosDkAzXf2Jj0lINP7mzY<vJI8K>I
zzU?)>ZDkKPvOQ*PCn-1Ljm5AQHhsS*dELP_ip=o@s<E3cg={+g(oSa;8+`OASr~G?
zoH=X#p^;G?s=tmU5#p!Nd@BpyTv$V-9HWZR&zagfR~}vm&(_HbI!37aPY(*TWyG1t
zRXu*0DrDGFzy4K)z+KDXi7jPOzJKibB5?shf5#OmLvU*}0-ocvUSZT&$>kbE8y2;r
zzAhHuAUsiZCR4MjE*;+7I#;AtF^Lv(3^f$yy?f=|LDQZg7|T8CuGPS-n84rz_)cKB
z)P9DGBsq&(5H(t(uWIn%m$xWxQ-%L!1H>(6YngWi3P!P1HjF}5TsQpCKOB2f2RC_J
z8mX<2SBWJIm(5SNGCmpwM`WC+%alme%7CdKz-`lulLo^`>stX|OXoEiKoN*-`6pns
zxr=bdlyc8ksWLACfzh3ifTmoTIZH=7N@4dE_&X6#bs!y~U2#{j6cN<sAZR4qPunXt
ztEkyC>)2*<2Y~Voh)RGT@G}M+1t_6JvADJQZxM6Wf_5q%^ccJz6(XNM-V07mXNa~_
zs{!HVy^i8tp^igEb`iB}r_UAucVr4)cN~bK_}7@!Z82U0&9$u3>J}Bqhnuxg=2Rap
zdA3>c!)31k5I{7?`3Qb!F0p-YG8w*g1!=PV43>C2)AIE7vkS+Brn<(VKiOL9Pu<}%
zB|78-WGjPT{xkia@=v0=7Z+DQn%4zE{44$`y=;Pr<e92pcCn5NM!hpmtVqWFMtI5Z
z{z>;@wKZfBoBdJ^icX-A>^`>T&<16J&rR-HtB!l-@cAqe=_71+!?H|8qI&cruW3e8
z94S<;;Z8%^#zbzG*3C~lhcr*YRRl1{-{y*S54_+ZS#uYW0WYT-4=E6Xn$%=)i}4g|
zY6pSlCewMMFZNZV))B;29Cp}o4r^cxg=>O-w8bxgfCKIM+b)o^K+bK<aroLc<~wCI
zZg*AoZ^vM4)Lixcf=NJtcObvPw3BY{<(RRN#781_S&Z}+f>cK5qY4)>f#rqD$f~l8
zTVH%57&RGNh^6Fq;@6`8jf|<uWgkcYB4t^FQ|<V-aVfR$bENpH-!in~$3O^*0N5iB
zeg)(cCD<%Gy6^|&3a=2YD}HDcrAhEPPPXB$r7iZeTqbb{^<D$sSI_;odQAWx1nTrV
zf@W1*s!(qyZj7GLd?g6R90r|%nVDmyPf}hORXDZd@E~U%*PK+IOUnB=Ki5S9TH~=s
z4b7$_L{SL(LKvDKmcV)aC-Aa`zv+i*u-;uReZlFa3C`l|@8AK)EBp>aC6$h#Pv%uo
zA*DX2$^4-%;j$9y@*WmZ4X*6GjG*>8YEk`mMOkcefW)tErl;aJtZJK%P9}HcPu(1f
zN0YR}Bjes_I?0@)_;jt&$m@CM<_Y_SxWB{;c|-x}-Gy5`=gT*}ou{Q3D_{|*oah=G
zjDPhVmzMh1XUcDM|9xKR8)yw0zR!a4+Vg4b$fenZUC*DM^g8M(q&NZF<b$lLbJ~M>
zuBjL)^6tWn8G<K68*-;lEda5-$|R~GjSYT-=~WZsB7M6MX(vtAX9HxrUPtQwn?(Zj
zl)rQ?>w`ez7p`@3kwiy1cMwX`CXBqfU3e3i*2;2n;=7vl8ItsO*aZrS;kl}bOclK0
zxV90ysB(EBlnbMGRavwIhvHSPW}9b4SDd|<`OU2*Kvv4$^7u~&fIID;_|_MtjZ8~j
zz4WhChMb9%<Ng_&7GW##fnQ?R&<(j+Z=egZ&Vnenlbgt%21##0flO&{rtUp_&pD$_
zA7A4iiN`;sCCJ&o_FLheTNJBbJbKmZX8bBo3%1%h;TR)O&&?bFduE+!@-}gM@*@k1
zfOWKxvGN&zZXT<?^FjE#Aw0<1!P_6}f9auz90@D@N8o?L<Jf`MC@y%X&$`t-X|F@W
zwNYJcs(s;?vT$@(4Ry$y(fgFsA29!dmUDAQDBR&!ifkd4qwqw3CqzC&LAK&eTSyG8
zY<&~uzCt_-=c8-vou4aD7zNHv3b;faM?IIGP9)OcCR^IhD9%lRPu>Ce+sw~cZ;Up%
zm|@2EY+mChvmMf~0>G~o{Qt9?fBRwozid*0?B592|0qYj(w0@-ltAfupmSYJonpS3
z!}slDJXnRu4wh$>9;XOR{nbdPOczAbE5Og{9jkF!154>dF)XZO^$)o9;kUrx=!mhq
zyS06Bw%w0v5hS6bf+J@EnjEx?X3Q9TW_D{MRh0cM@ZsnvR(H+t%?4uuBXFO~B0>tt
z?BUP4R1FdpnfqZVh5yMy5*iwAaf595^|IWZl}nJDOXSDL%biV4sIc>GpF9)qFW_hD
zpM$6C)AnTV$<H-kbTZhJ6d^{tS_&twvs?pcy`znOAa~15b7~$>YI4oGZIMAK$;bM?
zE|s$k%%Z4O8H~aUuDPlCp|EJv_V-2OQC)0;PM$e6=EK%!+v*SA%FR&^4L*b?Y4&cZ
zJ0+FQzH??el$DpHc4_WB`oz26e9Nku;HS^aEkaMdHwN$;STr@>yzovzD|d&0U*%gr
zu}o1!+=5}@NnzrCP;zmtQlS>20Pr*fV^U_fRT!(l-5AJ+uc$izU8{njh9FwJrh^7!
zpd~cxTyoOiGQ4P3^=U^sd4A37H+DI5j@E@<rbsXX=Tla!vMupP0@AUW<p)>wq)z2x
z|EXV)Y7tMR*IxQJ`iVe<#=NoAp!Bg_l6T8+;;94jIh7o{hRakU7J?>aE$%1qFB7Td
zIB|-1-5|K`0rd0l6`5I8n;UH#fG%(K(gA&gHdz*?C>D0v_>1g00+nKz`SdEUsqz+7
zaO2u>uTXNQ%FVcnEOmnwH@+5Lg@l1Bn%a77FdOQ5`nEPAL{!n>GdmgY;@)X~kL&GF
zD)-hE@eK7fyY5po(r~L>_1pwlqY&AAlM7UjB1x3NdC8N#(DGZ3Y0SIfLJd{(YCHgt
z9nNWp+6OE>ATD_ytc5?$Ig0v~lG;LtOgtCnD7S$?`Fe~G>L@q(|Nc1<d7Zt0R+=VV
zJq$oOG+km~C*`u%Nt_fDwmB7}BzK*P@Ihf3AJ`OE@`Qzu|9QqTRXNY!I2Eb=RM21M
zhfg&&b#IP}=eW9J*kCP?AQkm%LfA|HFiqRPIhCFyaelxnee&SFVp|XBqKQXETtpd^
zmgg`&&c&%<>&H{5IBA1;5;KP@5gI>-`_~um4Qm#e<dju+A{Ud;19@;T)mCtA)wNDp
z)~u<d!GBA%&YQ>D3zDF~xUlEli3|m_lLQmh1kb&Q{VIFdRgFPHtwLVvlovU@MCRWJ
zXA1ucq6SAj;8+Uwn67(cABG$tUExb2`)33;Pr46YbN|-7d+6#d!$*thd^*AZ4*ZL3
z`dqqz=7r^rl>nY@Q0k7PAEhSj)ynJqCoA+?AQ-q2%48?)1-%e>bwqQYF~wgMM<L0<
zO}A*~q{J{NF!6>kNCG!V77*JZ6xO%O>NpG*kG``mywT=Rw?p8^c+0Dz5ZRF5fTt?G
zed?>fHinRt*)NKCY%rSehrov$YnY-KVjazcSEC22Zg6W~u%=q`C;GZFX5b}KlQ>VM
zo{JGXAIf?u0VywCUN*0;xpYEFu;AqIGq1-l0JvP4)&*@jpro=$<>?jF-|mD>1XNb+
zt)lfYJwd7D7n7+M)VpVdg-%Ek2~v&cK<V$%;P&FqY<ODhwX+}57wdl^H(Re)v3)%l
zWClm{mE?adbzW#{VHv&ESk-m4c?bpKMp+7xa%D*K)4O-qGpw*F9AB8_He1+DiBDXJ
zd#_-8a0FAykjk&0^Ai-Si=OCt0|XIB9cgR^<s{HcjGT=+58<QbU*{cZn?B<nIEt_N
zUQ4N_+!$B699a;jIy~WMnJc(nDb;q%w-e#`GHQ##6kpcL`x-y1)PyjasJ>MDSCsp)
zF&#rmtG<R0s8et`zaU<aO6Gb8{WeNeu*ZRW@tHE=hI3QTUft9F<3Xb5P-WJ$7_?pR
z=rXM7o)XtBexR)W%8EmjntRo3QpPN7JGB2AkVTT+h(Zi(-6K^V5QTiKN9)&lOd`hK
z!}F=?VLNoLq?*;lq)-KU;;fb&A;O@MG~nU~t?h2TdAHu+i#1NG|6?O$W&3}+HvNw=
zKrBGE|GCF3>1^2JaiaLGXrla)eIRCJ8q_rgNa?mQ%z}w0f!W3KRUI`~Yb@igt7rbb
z@Ci#YxiGJ$P`3)AK4uI*L^zq;OVB%LQWDXH3ugOVCok+b>{_PLBsW&y=hR4m*CQ5=
zY}*ZOjDt65+igb)T`33`0T2oTl5T0RQeaZyw<ECI+Nl!3;*aLS$4xz53-OUJfJrh{
ziKD+;w2`LV>Q1+R4_HQf2+z2R!Nx<d#)Qy|69$Dlg|Q=?4msU$<eTP=BA3C?cmzMu
z=%SH87c)a#0%?(hUi-A#Q$vBvCNS>^&sM+DN31nS3xkg8?G~wz;*3ESjju=n586T4
zKO^I+Bhq!Sr=2ylDz!aCK)-M%8GQRgOez?(?O_A@6feOhsY|$4L>PatCu4EoSc;q`
z-EN9X{~pREF@wO#*%(*|Baj}49dAhvc8J^(gKHLkK&Xg7JJR9eW5%AP)627IMpH-}
zz-HdY8NUgH-Y+-io2>$snQl<Q-e^x4j&aOY4-@G_oS{FX){CKs_V5}`mc*T{A2)>o
z*J|ghsvm%!g2E9QJ2K#tY5%D_APK*dBu>*#k0q`I>?K1*5RD3?)EZF*u_#2}2cpR!
zz=Fwy<Sp@_7JhG~5w!6VaG+r6w*ZMJ90RKq?5j|#2#&jK?ZpQA?E;!YaM1*exkkmv
z!U*XT%!t;*d){yrlV6~PMUk?hC_n@a5hZdH*QA3w<tn($;)Kz_Vj8h6Ufv6gohW#*
zP!Y9W%;DPJ4@P=$R%xWAJ6AUEmACWC>R=tJvx0@BT`zw7;k69vY2%<=POcUF`j@HP
z7*pi=LZoZ<=$aYe33J@UY2AyL3i7sEbsNg?%ua6fzTIO1M&5|`+g*={xR>f<lmoLp
zyFH#c^5B)v@I^&=X|nLrBsN9wTX_d=s^JRJ5M-~>iAl$falUi#GRmSA<%>06&D_Rt
z3{WfYmvN&|ttV@@?ZZR-;9Xw#{)M-*y1etrx&EEso#@@v&uQeU8IV+LXO=)9Dys_k
zL4@@fGZ=Ak9IkWi{d+g+u*c(uj?sCqWl7pQ@>TewmNkPkt~8TZ6yY1y!oz@I1XCIZ
zmy<LMn^D!@{EVDno`DVGl&xUg4$$LHNDO)o0&fHbm2;!$)^y=s#P|tpJ&_YMu(Y|?
zK*>dByhMYh0iS_Mf>3pbmHKlx$*O-i8bmV@oCL&m<Ph`hLZosVG!YnzxNrmzG{HEV
z_QJ`57*`(f)-o(t!tAowKEdv6Cu~51g=S-LM9Ah|8_~2=qzTkLwlOjP*8DIswB>7y
z(dh$NgUH{Ka}8hQBZ$eTd7P2)?R;&f&fEy8*^s7psoAeIREVXL)4M>s;(KVq@ORwk
z5L4b4?kWg;fb#Y0T|F)&a@&qVs1GlzJ#nWe$l5UD53J6lttet~rtadc;1Rs?n6Fk5
zi`dq<et9c!PKxFLtzjJehcGh(Pg<*WfVPzUPyatjwh61Ijz-+JH}qCkiOeoP+EKRs
zqL%P0hASHz`GoFhB{l9!UXC7--{1CI>~6A{qWA4*9oC)cFUI7Z)R8kyRP;6LER-H^
z#_yW&F|La6IaeMJYZ6<pwoZ>k)dW=C6#t0iJJ$t5(&FFqPeqp|kDmvKx~a7;5p;4?
zewuI*%H6vJsO|i%t#D$s6@2=_ZE5{z5Nu-@<y7zOx$b468oYN_WB<56kGEQIj{cZx
zVOBh<PM-8x0hvL)DF-4d=k=K~Njw@Z6$NN3j%t1=x;am`FllOWxM|BLU;iPX3)aw<
zt`8xmT4J=-cPC;ULWjM$TbNJbxD{Q^BTC9z5V?n*8^HguoHR)#9`vl>d%S0+RGR-2
zYvJt;@=ryjKP$E5(cx~slMH3Zd%q*@9t38}k0~pWWAjx32-&ZAXC*_Vx#5<wNc}>}
znoHb-EbMMwW)jRl1RZ|5W5%f_2e<fPDTdJq0uSc^oT6pY5XA)~I2edvcV@xyKiXZt
z`tQz20uL8xP}B?zkw=$K55OZhenNI;#iMuxTWCjcMu5s$Bg@;isvar}6<84f)-REx
z+T~xn{$jhrdb_@c^8X2ptr3?=3g(C8YlspW@;Ox{Qb<`(m^?Y|AM*t@0`<I6G#>Md
zrMG?|%vj(^X7O*n0=cIk*|ts~ul0Z@txN^Fq8$EARJ@2EFW!h7ctXN!?1A4ief=0T
z?|I$z3G#X`8qRXmA_jFJRGUKDeM>u}AnHw?XPKj!T|v0HmRt&_dRXTlBY~hN@eMgH
zo@b5mI{2!UfCMAqLjBDaxH6#ZyJ)MBtiry~>970zRYUn$uf-&G#_~#olOQGTj_v2J
z%W{*Bv{7b8A{Ki$KWooT>C=Pk`?=mLrmJWMCaLvSy)9&76@bPHkJji47-yN2MY`}J
zPY>n4v{D&7!}f{Bv5gzipnUOd4{c&z?@K%QV>@}$N_ntSfz)rbH2#s*zn=Gb&`H?o
zPuSPG>=Uxdh-vAQ@&F=0i2$`bBC$@4b7DGRq}ULKXn&!+^!{2OpwGgFE1-q9m@Lgs
z>xIB(LnSJ=x>sJyzADi;lNem~jvCms>@R@#aJN^rlo`odG|IS#(n(NX4QJ1vYlyL^
zTx}S{|9oEeX0M_OVB?k95!|G%3g3`R`xuD51Qy0H@$dyi?7z0bso56s$SX8sDiOI8
z-rkq8b<VYt51q8&U03raI-SHE00lH^bK->-?LVSo^+xG9K201Dx!I`B*Pz*;(cK1m
z09h<y2`kK;6w;-8byTm`Gl518q@NWNA({H0&Tv*6(ku@xgqx}Iy!ft2Fe`g4uG*(G
zJO-$TU2=23XCgj9<M&Py{$rkD`>*H_AUo&(u8;aAENyW8|C}R!qYQ>qN3N>2ZqV!W
z(ImVqGlGdHn!4H1%#umJU3iog5UK%qF9h6?e<Bx34rY*8^u&uQpDY(l=TY_6Dk#ig
ztH+Vj?glmL@RB+f5uu#RTDTqfy5dj;)AQjddAQyEd%wRN9KLtnbCTA_-rt=He~|Tj
zy?(zUqSop$2o*T(^?7)5zFRr}ILlxf5%85RLm-B5MGe(C3(YLUuT{PElp@8*U+>VX
z`y;RFS|5GE;xpt_KVtFAPkxRd<Q6&rl}s4mc0JAe>yL3tL<F!JT=e>x#IG=AMjNa;
z;a|?Ov$<W=A0;=qA9bif61z5t@%_0nM7gMYQ3b~rKVi1sxHd*sHkx$hbJIx<`?d?O
zy}Y5IpiYutc!0jje4gt1<XLSMjX#<^q;;l3XJC$uGQLTsNf1WE$RdS1SADD2$r>n0
z36uJqUMd#i24}sC-LeyI%GbC0-9Q;PoRcNu_y@ej>HD{)s^eew^RzUL7fqT3w);qZ
z@lt0gtqOuYfrZBUt3d~{dDMbK=}c77hH-GEG8^oxOV9Owke~$$x^BbZpRZyR42>J|
zh~Vb7e2wk;*BO1A@ZH^u<l&~ymtu}1k`*AJV{bCJU3$NU2i3vwkl#{|TokZ1LxJ3j
z4Mp8#O!<o+^AdttzPNG?Gk&DL=JUNu2}8ZjIM<T)G4ZI3dwKmwdRika2Qe>fP4WTU
zC~*ep$i@8S!~8G4NYCA)AQhqVn2q63L(!IjSHD5&+$o5EZMyH62t&trYI2$hY-=2~
zn3*qgGS(Z%Z`4oQ^KJbDMNkq$^@JKJ*aX=(PA*ie`d&{sSbycJzr^setZ&&Zj22uX
zJ2Ao=!595AE}8(@F_jg%^Uq7vbipYIjWh_EQe`a#i;LRDw)Y2fBW<fgBtU$gaZ=H<
zxMQ`%A_we1Uy-6*QGS3l>6REqx!cV5ig&WVp8Vc&{!*Po8l{X}WolBK6Ij@U0j*x0
z-D=v%2e5I%ADaY(1bgaYcmsq5llrEA+IPr4bckqg+*me06Da69Ws^_Utc|#B$n;%=
zS8dCglZ!{5m0w4_Q%*Qsg+j&c*4gv{>_U~WUhS+)4+R)m<>mCD90WMkt3$FZ+HiFK
zx%<B@yF+=8btA?qk;6SJZ5~X}#D&MeSjOcRxAAMKG41iz)%KMio!p^Vghqw#{Xo0V
z;ra6&Eg_`+L-CV{*s)1~aWCfn3Pu}xOc-7&^lejvb+SB{h05el0BBg7+U|(bU9HU_
zaOKa0`N`7HXLg3@kZ%d28XG9VH!)}6P4;lWR47>8p^-mIX_Z1O{w>RRsk?Il{K%m6
zQ8e?InjkT}hs}D<uggw_wVWY;yCG6*w9NNy^@mK7Qu*~NeTDmqr-zcp3-X0BTv>-*
z<3ubVH8`ZB&!V+#N6Bl;w)JJ7a!EYV;`q8sT+3qszy{ON*?g3UTsl$IrA;0|gwsWS
zbp{UG!69vkg^1hN+2l{yC3}GJBJE!icvdrm=^b-tHRiHvCeT!lb+-PO!q1KIpewRf
z!%qQ@?l2+WC<-n*j32aEV((D!$;jGW)6J>);=PXoZ?<$`x2z?L29HqA8El9pKjdxx
zs@p~cT$-5?-V@oV%n^VWi5SXnkwU3M5_<H!nO*dvMF^8kXT)r;i~Ub6k~uITuf4%_
z;#pK|`%l!i6#4|6FSD&EkVpStZs3CC%){SN>0@LZt!cT<w^%v^DQXfBA{jxLOE(bc
zeNAFvid*3lJPU@wT3uqi2;u0v1I%CGz$TSg9|k*|!W@JA{16OnPyZqLhqZr~^L#(K
z!=qC5skBa+EW!5?deDbb?7#NCEL%eCoT5XN2iH|3&^XoqiBxXo7;F7OsBbxvf81uR
z9Fs6m_slO1+l8_>o~84MX^*^7O~Gg<`STaG0^jVnjhy1I$wQEz()FFuPLRiTlJn+~
z2&tYci#~Mh!BFy!J~W2MGd;$yj`^pmGns5Ysqu`C!q_};xUqfsVqe3f;i%BpANN9_
zJ?C#q`8|T?-7jkZt|uRS$T$Xk#UN9<aGG0b@Mm|od*lOl(~bY^IsZYW;^6-Ow;baC
zZ(G1vweJ(Z5Mb+s&fE%~C8yOQ)oMwI&QjTW6{JGXVqQz;lGbT`V8!K%SK;apv4)G^
zu@lh=KI^>{_A4{DB9(;;Mda^k7}0r*#4t-UYSqjl*?a!|krBa#i$-pX8jne5INn<c
zpVR$=OIh#sh*FWp2zXq@X5+cmbQi)YOwGmz#k96GMsrS&hmGMA?d&q@?v8F@-_OV6
z!q*NPXL+qMq$Fo)dUEy#X&yxFQqqEDoXL+6t0(SU+>t_$Hpf6NOn%E))sd_uoIP@2
zN@b{}y|Ekl)`09r)iy@0Qo{UP#X5UXq`|Uyk3^M$&yYK?Rr#?&l8Y`4=49!(j)B*;
zvEL|Qb>Cf|<aXo0FFol856`og`|4XVAnUyT41{*rySPNdxlp?ix9XW^qB5M6&M8zp
zqxI%rHPb~EV+GRA!FsV@eBNWn23L)2o7UkPC4L#$4LDpVikj6LD)Q9U$`XH<{?>_l
zd##7Ia$GE-*dE%J{%<>n7@pGYf2?unX6O~a20rJN3N4|oNezj<6SYX44`m)si!Ch>
zPkDTvBGii||MZH;M?kKlNG^kjFQZIWUa5wuB6KoX)rAa!D5WTdd)u-|mQ?+Y1Pp5U
z$4>s3?O6brvvrAy;?|jDNg&Sm^(BQ;AC!Rztiv~IQO>z&*fxa6N@6|}N29>eCf5((
z{1Qo`;S%NwdT&~o`XG#lr@p*un^%M9qzn}GA`MCgg@p~_w$8u10xDi?tOUHI=ADY^
zSf>!2imh2fQa~IHu5kFkd6Iz~c|)}lXL`pf?@VDvOk2kUJLxmm@bgf2MYgp1ZG2t*
z>l+KBHZDsEI@+#0?Qp-++sdMmNqQ`r>0l&=3$~aW?|mT%<L_@&xa~H&gyUeX=sJr~
z`c!lpCGu`HC`lIl;4H@ZuQ$N?U%GU5PR{=oi<r{f`qrhR1k4x~q=CVqu)Zfymk$zh
z6cKc!8CyLqVt|^?aVTp$Q*hUOy%Bm9*^Vi6xUNFK6aV`|7Vq)=w)KBd_D)TjwO!C)
z*|xfPciFBk+qP}9%eHOXwr$(CZA^a=Gcj}U9K7=%_Fj3dT!}c5iBORqt=X}cla7ds
z7Vro)5q8OhQyXMx?B8zu^TZL?<EMg?U~QkSFBiEeUa1V}x340$8r-T~v~;FiMRyz4
z+&9vSe_fuQHiU%0DF7)hvJ&C`0jyDK00rW`K|qqwvE4AvbgO3C^mV^s)7oh}v+k{~
zar?IHWQ(=O*uC0{SCgnbGzR;e9(WAqHjHxGHl3ofBAtCJ(-(|rg-J+2cOTuQ(xAIy
zS(}L8*;W^c$f@3+R*^#?G?Xdtor-2Q7soiXy0SZ11>XHYOjY{Hc~s4-M}ynckBH$l
z`Lkr!jJs~a$<83lTmO~!gyP1H>8|4Q>9xt<c^sk{A4y%hac@@^O)i+V@g{zGhEhTt
zgEBC<oX2bik;vH*cQ%q+JE|E>$h`HwzvBxPhl;(PD;o)=ObnBrn$M?-*LPH>?%z;3
zFaZZ=YQOM?ZnHTRm0c1?5FvH>%$BDLBuP?{$bA;ALxAglD510?9@44DmN-;FH5k(l
z!bI<musNkyBHLHjIVQ2G{4(>%A58j3)1Vy<ne_K7+xgPUzp!Zy5nw`%U*njg){U7X
zZGSaa4OeBib*Qh77fi^Dph%+7;%tXg+{d^Qlg1`S0=Ki!KAH$*PiH(0D@|t*=2X~b
zfKMzo+b);0IVux@&$@O{^rb0!*E7YQWbroNg*^8~Lv|@F5o<B<Zps1OG5GNF4X9ae
zaCp$z;o7oK`+?n>(f@h<VP7WA>|t^aYZUZyqBihAQ_^te4o_dtIbEQ`(@>s5=sS;b
zdHZYRMJhhcU4+n<U(Vn|vcsfMNbz?QZwiQ3Ak^=A4ZFXD*<1C+9a}&pUG%^@z$dyL
z55RGNTmX=S1v5^M1#_|2WW(Jm?>4XDeoT1e^0W@pyG6#<7Y<)(!wxu6e?5tWO2s|v
z>CP+)i~KpQ1kEvS^FJI?dwIwG9ClRQ7WtDsSrqx99RP~@NoIsW^*7^)mA)<O2H|2I
zex7L=2C4ko2o*X}L^wd|&p2pyESIU77}O%V2#h2+DN<qfLL#1CmM9Z{SYerLzyvm`
z-2?Uk1LEDdm6<R?T^)X{oIX)AZ}tOn(?G3%`+!O1F4#CvA@g2+Z<*Y;IbL^3(Jv!S
zL|E#z+SZqf6X~v1nfkdXJ4EFvAn*<SoH|h#s;jt*%s58EEQ5VGftW4+Q}vKguyE;j
zkdW9%JfvR1S?oU7!~6}wJ>CWetSK{mR&Q(b6oKjs!{mAd51c;W4AZ(e5HceJm;=?7
zn(0(x>^KMdp+)T)<u@+SsaCK-#~DG-T_T>nJd)~ZtTAI44k+DormZam>V+5SvYQQA
zaSM{kHzR>0i>)?nhBP;FOcL@S!TLx*ziQBq&35j<w%by%SN`5qg8jfmFO$^h?Ctw`
zV&JcYJ}he5u~IMJ9;&BP7W~HQe=%~LXs(<sBep#j{lG=n$5n1nRNn%ydQ}dZDa9&R
z-s5!CV|mlB&F90^m*YVQTW#{_?x#~H*5ux|GtXsq$~!@8Cb?x+UYu9h(clG%9Ib4?
zFN}N=2=L7_DKEI(Y!P2_1j=HU%ewt!UbWAw3f%QdZYuJx(E6_X>Leq=5e(w&FJt$p
z-GFv(U%e>973E$`t-WY{d)GIoXOP}gE-qswSCl24Cw730Gd(;3_4h>w_0J}>f-J~$
zkki<HL9jKNI;Gg&-lH7LgDhB?(dlmRBkyKd^knXA@!#$uXlzSta-<JeYt!VW^On><
zH4OKKzRTQE#kv{AvQqrZ-Z{$|9T78_15e<2NWE~!<|tDu2t**MWux~{_$aq{=FW|+
zuU{V0a1`*=Kn^-;qtHq?;x=Cmo>KitSzIAO(8V1VnoT(dCj+84{kAzk^eM9oO>%f8
zQ`t$NzchNg^*YE}&ZjhVyB4c`Ufs13Gk<)TV35`;FtL<ZSFjqPBT@Ulvl?F=cD_}0
zm{wmiHH@sp-}Y|oM#QVI<G4CW%^;@4VblWgVKY$GBO`%{-(W0%`kDU=dj2z&`Ts}p
z#FeU~W5K^O_=l=pV)>ve{v=QvM{}lX(#p~AlBeWwbds=eAWXRF_nndxKOk(*ZD*}J
zec8!|kBjS?4#-Pu?#nNSdq)QccWin(k-FS|NmhJRd_=H*;FtKx4`vdgY#$kK-Yx(;
zJ|2mGA9k!hLe`JFOBS{5G0%~!2v+)s*6!NS{kw^qH!EU*dVObYMv9M47Fmy1>w9TX
zwjSgG6evFy>}oIm9hN9DDLxr8G5_Un@KQ+^ltS>kfAqn8xHUe`Z`h7j^NFVMiutR{
za+jR7EO(jwG$re+XC*SS1ZVG@^3Uz5im5q2ks^uprXI&8+;Ge5Lya|XW%CMV&-FY=
zTZ3rgSDZsiy2jMl83iv$E<@#BfgWd5Fx3|~QYO3K5orxr&(HOnIA8Rv<RQenI%Zc1
zX<PMFXWP^B?J@+HbNSYZxJ4FMU>5wtzW829OoJ3nr-<Wy@P?S?qA#@L6CEnff_CGt
zdYA-8PH$KYddY})MGQ?&672<LcZ1C|MrAl;$Aq-D!}r0ehq5HzZU8<2LKcQW)270B
zjGCC+p)j9d9~?YUEZ~~mX5Wxee^6Nh=?0M|+RB;#*KYF5SvR-qx!rB*7|A0~L1@FN
z@~yQi=`AeUknBDI*pqxO0a=bqkNZ66ANhODNQ&5p(N3fx0FL9XW<f&&rB|8TU-K*s
zg|T%RXQ4t8#f{WOH|O493@_c@?&@=Bk5InAUmFQUdLNo;0)j%X7yb{<Nj-~brI-#I
z;Kw~pu-uf>%}w-u@<?<1minpd6L1Zn9Zv?j7LS#2>0%NG^N^60^BoxX{W#u-zaV`@
z@aohek-3HriR=W3Fg!gG@kVo3ZLJ>ce5)W?+A(|m9A?-Bg@jc0+34*Q+hWI}RB9SL
zvnCwAW`&o$SCoOXi0E2G<JOnWlV4c8%NKKzcigx6fjq2uZmY>ECORt0A90g#^<?Z6
zuWpK%qLEl{g9viji3tgPn*o(T3#_~}JTqQ?ca_Umi|%o=eN$?R*-qXJT+4@Q%4u3l
zKH#1bn$(jEy<2TVcrS1Z?Yr0g(bU02U`*o^n=RIfLPX&pTl?e&fKzvscHK0b?$lOu
z7iTYToe-WmE;k{n5m616fSeTQ#q>WwIv3A*igt|ryl{7CZxSTIV9qoBF@MU_1(=-{
z^ZL<&)i|6eur%BIt2r?=NUr2~dkkJa>#3{_DL4~7<d=jsOun77QY^^;TecNA*&##w
zuh_zZAm4i!B^Bm=>sj{$j^l;a%gWsu&6^TGF)GgBNH7!C<FO>jT+{GjAcE?y{PL{F
zpKF;xvG;F0U-dWZH3QIs;ZWBaA&oWCZ4mKB2UbDEs8V$<0y{BiZ@dbJ$om=~>lQt(
zI&SYp*qMKzsOM`c>S=nh@2lP9pOH^>`LVheE^)tt;^f#^m&tPV0>A>t=z;bB$ndGv
zLYs|TV<L%|Cm*Fp93#RWUz95y=TtbrhKTYTo-DgyNqt%(`^g_#1YXG9E<E3G$nH2!
zqk_ol<H?eKB@zQ08P3Ic{#++Ri7<`tc!|PsJc7JG!{5=wAIvd1rDcAC+BgGu{paZW
z|6XsUjBQMv%m^768UN?boqwb6|6x1L{^jJRkTbAhiid9rxA5(2YGOtHChH$pP>D}C
z-uV}lt6oohBL7&YeJM`jFpBNaZ5u2YaN!Ga6TM9vy`9Na4l3QaPpMW3n&&S^DPW96
z9lv`J2;D#CL~8iw?P_Ldts~-%IX+=%?S9~F?S3B-8(-Em^FD6u?d<bk>>(p}^W)VS
zhfLp$VHQOM838SANe*UD4@XM=Nlqj&u`<+kfd_Jfgy>=pRP=xlQb0S3ir4b_jH{4A
zm0eO9<-_gW>w{A<d6rVy*c^&!XA3Jr7)oVRBD%D@b{eU6vUodl*xF*N*$!brrd{#6
za*mnU%7(#`k+I*h-V=kwsqhXonXxcy11BUCM!*ash#xAj%4W2*$-VKGFUbfA)s|I)
zb#|RT%Xt9hGW(NCQHy&5>oA>|lwtPf*F?R$2KsgV(&TRNs#qrEGu^}|Z56NuQB!yB
zZmwa_d!Efa{P;yZPHEdfzN@CXQPMvLCn<Y<Rl~k}t)5K)r-!c1!?4H{g7y4yu~cXb
zd+dO))}K)Mf5hG#u@;llUmo<i!)c?ow(+lGJ<}s$zEE}%2JyKfCve~=TH_%Z>1h5+
z+nAqz-(+6!*?kYt_qbys8D)yh%E^(ZxnX7Q@;as#2HoT#7tocI?PbW+L8u}Ws|DY+
zOIjx}Av9`MFO{*?0s=52bSO?0zsSn-il)l=I78*dIv5cRl_yL%b#TE^%D-%tPxJu&
zNl;{10l^@fZPO`ug?@mzNt`L&=y%O{KEK3=pQa-ADofS5;9738X|+`nwlULhrI4Mt
z$qtDnW$M}mVOW3Axpk5k<AkcK=8G=SNI7THuxM=R<l3<ci7XYN)4M^tq{<O1a}gt~
zFVp(|8A`|G%yJSkNC$^%h)YkcheB`tDbS&mrB14b_ls?I<(eed#iGzNkn7E$zcE$2
z4wGRJ^c*RDaZ{{pt$yI(To0RLy2?Ut6RsntY?S$zKINYorT3n|=)9f2&+Ow)LS_J;
z-_iAI@nT(>`Q$&N^&c!Tr>b{9ROi>7?b|lHG8?zakQ<19o82oT{QNh$t<cnr1Y+3@
zup+>_qxzvCnwTI3r!&9uqb=mouP56G{>r@SB+;sUfj4=unZ4Nf>M8<C<r0S~ue&Jp
z%#1Ap^Bt;B!*=>zsV*32CO*<Hw+<2X6%yE^L5Fv*FwD8u$)*SRO{WK1cT|C@R%zn|
z5Io=CbE8TWt~S_#Pa0F5))E!D$C!NcKq)^7seSR$>!E`;yZTplO?~XWGOFvJ1WPWj
z<DXOE9NLC&Vm%d)74Gog(D3|3W`&^plT7vAUzY0{@#ib2t6>+c>^l-bONZ74lA`f!
zTb&?Gs0Bc+!|FJ0<^K4Xe1=(lTcNb19U3)yb4|dZ$AzBrFbF(_V$lb_K1S9kW{h3c
zax+_^rlr4zuT0<C6ibdL6RWXkvdon^Q&h3hb+|RgV9U4R-nl26B7n%T_HQHmJrWsV
zpVGz>)@q-&1gMO50TgzOnmYVW)jKW<G^9YKJW-uZ@2hO<taMn_87@!b8^^DK96EGn
z1Q7r}5kMcqbd(*>Hfkb(N25Af?IZWp-8;$=K96ShcgO>JgTwzCj2tZgA&q3^_@C*B
zF10a-e`lbs4=UHha%wOjVE@sngw+N-MhZ<w2?bg)e`6JL2|)!1W?kN>sEo4FG=L%x
z)ZqN7cUNbyePguU<HZo&Lxd_Dbu=IzS~~wt5RDAk6uLiB^g;PiUjQu2JN@VGJXmjS
zO(Y)c*%z+kY|2*k=47R10sF{Rz<@dH4or4ir>CdO+mQv`im}K00>k#({l(<_Bj@Hi
zau)V-391KF+ayq2h&{?gDOZW~x;`fHe)~+7EwMx%W+Z;~Tm*@f@iZA-J*(`qyIi>{
zi=s;s1#6i2(wjSL2rNWU#?<3}KI-CN8kPufy_tq0G!KEOoP7Sex8Rhl$3?*ch0~To
zgkbT3n!+8?tCyZDR1Cx)UJBSTcFVGUqyX=>XytI7wu!_-zlwSq71jf>de&?{O2v}n
zl8&!E0W(M>+}3!3W61Qs+X+c!XvOK^@Vou@$(37o*ECqq87OA~OELl!->3MAWzwVy
z`QVbX#vXgpK1(g16cYZHE8%Ki+-DY12s3}%uK|?>fyqq2KdZ|*!)cf4nGzt3BMnbg
z)4d$2fsuvQ0%IH{7NY)A7!u<R2<sISd*<G@wQ(r0jw2`ujIGTNUt-!s$zVJIITYPZ
z?m}WKMRu>6z#K?+Kx5*7*>G*!>&4y&FC}th1<{li*$>4!H<S~?3I^Kfox?3vZon!@
zfQJiBHG6qNHq<y>YgbawQ#Fy1N#4kB2CkdhD!vNP^nG+d_37?%VUlA-p!G8tpL9R^
z+c<V)T;5Mp+PU6d7Uc2=GKj+n{8McALW#ZTuS}F+^L)k1mmMJ-yTm7=xRM27)|Hyj
z2*_`(XcL9%^UGW>YAZ+Wa)UEof@U5q>U0ymxyDVaIcA#SG<xEAV%k?Wg~_u>+*m|F
zy4cuEKdT8mTe+|^$c796RU!^R$_Ln=EnXX?H-ohT!`f%$3vUcz_WV;A6r;&>5F5<T
z_ounGaG!ae=#QTOUqdzmncfX45vmx}8n211ls^d{_X*T!=!A{{085;@5{}plBSuLW
z@<Wy?9B{c_O7}A44O&&Qt$bjzac`zQj4jbO#UM>{$4Wb}IXcGr0iQY@4-KWYdJk==
zXgo=$MgJSMb-;T`^QVHxl5Jcnp}+NoQi{DMJ>gePG!Zm<X33_x<M<wu6+6!gC<yA{
zeUPx*q4ung*CdLs%yYw)%F4$)n;~S<jk7c9EIacmAivCf+%t2^y2+%DPFzNBvJ~|+
z*(yAhA3iqZtGayv=r-^yfNK9X>rM9vT>aZYF_~EJTnk?wIuPTbyEqz>u4kL~u~nLw
z+DtceFIPw^yZ-$SyGE{6O)k+@<1SRg$?vkfvb@3ID(hu-(O?7V!@3+F?aHDgB~4#;
zAYFRqlQBk+mjdGGiS~oc%*bIA0WRdY?O8ev+?*QD1Z`+FG(!HeXbP7`vo6Ey$a=NZ
z{mqr5m6|GD^1`ZZMmZ(~ANPLz;Yty7dHwFe8&m0T-m++Y5ZoVsl`3SOl#})!^2I6{
z6~$3CqE6GTj#Z3+07|B&(WV*_PG#j~<z|0o<=>03=TOEYm_;F8b<EX0PB~QCCXf>+
zDv76uib$>O&HE5uYP0dZ8;o~(>slsa<&Ns(oU$qd3fnkw&F8Sli)?|Iq1L^oVJ}pb
z?1<dT*AwrB+-BCu-K=nwCIX75ObSA-Wact%%?qt?r_kHbmzk=(lkE6j{s;z*`Vr9x
zzdnYUFP6KDQ@tzDU!|ywKBI&o|G>@iE1%@;KIWuTUv;;n!Yu+nOe>{Y5Ma=6RJX`e
zb2CBc*3U+EjFnK&Y?<w;ZWI*D;~a=SAX1JS&cwh|5~)eQf9U;U#sq?MLW_4iKSyrZ
zH$j6@xXiK%N5yf^(zELrUvZZ&_o(4MnS!ltvC};*`2!ZBJD#jP615~kPkp5bP%?nB
zUY9F$tDQ}K66!Y%_#GLrl=R`L1nFav=e#GfE6Xw0p+!S8XmWMxeKBI)YEV>RoJR9+
zujz&Ixn+(XcV3LNrQtDIQ7%x8S&SEhA^#|AIC#J5vQRv!4k`$E2FoHQ<0}%Lq(6_%
zfLr3m%P6lI?#lI$D2?r$Au$~kwI30mJtu>M-nr9JckA@Qz7hJ}Vhw#{B{sHs(c9I(
z`PqRmZ<B;x&efhKLPJIP@Ns=^*YAX?zibM2vo`+(1xd{Mv%+FQcYk%y<RIH092KI^
znXC?4{oH3Vp`-%-tc3|9kzPeqBbw#T-Sdix#ns30ACSjF|6i$u%&h-2<ozES<o_7@
zFH~<;EJOru0i)aTkqP2t2ZhbW+yHEN8h*4A<}!su^q&t}s9WH4iNjTE+OANfyLXqD
z4cnAeNv_MTtWFInolYY5dtx+j=35Nh81U>};C?OFh^Ch@GB^h2tOukIx}BabSLY`O
zx39e~=KOC0Ox1?7Od&(miHQhaOhuG;0@|$KH#!+TTjLjJDMPk&+jhr#P{<Kre9TCI
zg$UBs{9*EVrQXL~b<*v!vI>GN=epESZX!FHe_^gz{eiAr@u9G1RW@Yr+^}Gkh}<~2
zeC|c}tO+|lXoi+V`nTC4w<M`+hn;N}29UObx0=0;JM8?_k6?S`P(^oJSsKd1XRR|I
z%!u^Q2h{Rx#+{oxH&o6a*sxYAS@utCGhI2{`<d<|XAR|5<@*ma9(Tbg>kp8~d`A6U
z9C@LA4bP*0s`r7jra9K0m<Rtv<_G86s}H|`etwtG#XJd??ORR4>ZTkOc)d%-&UfO7
zCW7dPa$iwM5g76Cr?QTb$|3RBk3q!&4W44asaY*KP1D<sW>i1w?%^V2luZ%oDx1e`
zq*>dKE}IlD`WKc9nxB*(H+Hadl0Q_hrmDKgw6Z<PKrV34^P>|zxdK%p4f1i@EG)VF
zDFOKq&wDmeLO%{OJ$BnH-;7RlyRH{B5Gl(li#zS>9z__g{g%ISLNx1YcY!Xq7jWf2
zH~|Dgp$e`EcT_z~IJ7Dwoxex9_;GMF)pDk|a4y!+GVnQx0k$N06e%;6V{?fw$;byY
z8qJ+}6=1*&>b2={Q$0t2#v=^7|7yERy=bn8?zVAqwqe6eWrE3B3czN5Bi(8gi}ZF-
z>d)QmGj=N{QBLiodo248=u~MRO302R{UKWh?MM|uhR#7lLdr(2=+zH|uPRr5+F=Qm
z5E@(n0dIG%=Fjz{H1}Ure{qUvgbA)5W&keMUtVCR1bGR}U1I0HCIgsd&S#cKE`X-!
z`9+OPu5#N&2l124dJk*km56PH$m($yTz97hsZ|K+1G6D;!6G!~MmhopMl)ZbaLylD
z<=0Lxcnzgg%f?k-s*A7TTD3C=mD9-rW~=_)clrjksWvnlzGJOsFt)0@%+jFau6Jkr
z(tc%Gdv`}STuQmE#G7pGH<LOd02ABMio!t-(AupM`$@=Ye^`?&JO+}<rVx$_v0gZ@
ztlHn{t8!#K<8m$F)1rp{bZ4Zkhy!2ewm;cvG`S>PfnL<*DQq0@$u6+SYIi)^`Pmo;
zPff73$MhED2@&rG2du1<u>5jGMco??{HU5z65|DaKRoub$}!JYG0#lnjlz6fAJ~qs
z{rfz!U&`3w;^CuMZ0VtvjON(#0aS{FVc}88Sk)ZtkHq;JUST+5X`(ks?uh;3Fb+60
zN*`iVUob6v8%LlZu19t?ju)xoGhep(a~2yFSL8$Rg^vnqMhNwVEyet4h`i720KKaF
zy@}^(r?l`LAPxb>L@1F&Q$dWYs~f}G_4XXT@p^$my>m4~!bvK29>%lAaq28BcAZRO
zGclL=+GGkIjnlnCb=7dx@x5@Xy<5B2;pjQ=m%G`MAT~1)iAYZjUcQ8X>aqBwqLO*!
z!ns@=Ns~w$-OYHB6ZcW~gMaLlZ9UyF{)3k$YPVl8j@Umxc75JeU8xwE_m@_d>h08#
zl}AaBuK<V_m?yk+*kjV0?QmN7OS{kvhdEEYz`5reoMW9e0E!tUzAYuH=$@ihX@ngM
z%T0=ffw>zu!uVA9i*}lTF?isF+T5vbyLMt<`QRm;nPeFZ?av7+i+vjt*RMJd2iAE0
z2KcCD6OeNB-~ijYg>-TbnBN?J0M(ImA$BeDm<5r7V7U(gT5T3jx%Qun_egCnZRu|v
zQImw4s<G97-G?<}t*Qz3stI*I>YVSG9;jv`&1$?EjB?*n;LkpcS=5b&SFb|`_lR+O
zEB|E;W1;^KNdg1g|1N_o`hWKy-12dO+Bh(;3HE2|5I~ZdjQF^)Wb7ccu)B~%^uyBv
z5YDF5kc=lWE33QL7Cep>2+pVd^T+<>QWF}##C+;VpTVzC_E00K(n$Xy%GZDKw$t6z
z-`>tsX6u~82EeAi!Q4)7Cl@af6^IQSF=nG+X;THM!M4@gKCSJZ+>%DEt?mb`H9u}1
zZLb$w8!^GB*ut#7e^C8E&g<wGPsF^ZQS;|lwd=>ztyp6y;9veweU`SGm+x;*70Xn5
ze7LQ(_R)U0D6+j*e@l@Z58{&hA&>hp;Lukf;vj?X3?a0>s-cT&<LbMvS~X#Z(o?a%
zyW|{6w&?Hz-EFm{joKv;IERf$wB)3?CE!9#89B~6ThEB><xNjSX?j=OWR!ox+peFP
zXjb7ReRQtjWc#%x3#Y-L*k7wns!-+XCS60|&{ov1&7{ImX|+x~W6LdmtWT{hZNtkE
z;w+1dch%wXUj*=~eTZ2SNPuDLkqWzxXSwOj2%+FNbdK7<%KoNHVLuF<aAeau<D7`n
zF^SS?YwCO)`4>7MG9w7_ZHgc=3M2?Yy)>m?G64Z@SsteaD%fPe<KoLPTLO0q+d2}b
z+sN$wl->T*!@HWEOJ4r*Zb0bS3(z-bM0;R5nIm^~Ep+c2==D%P7ewV~v*+O;qWoPZ
zAZkENnmNID9}1P1$)-Jd+&?&%r-f4j;=+L#f!8_mjw!{L&Ae#YwZnN6^t%G-)#RG#
zv#on!!t-nPSF^?ek4>iSUt}!M5N2q=sy~Qha+w9Z)>vtcDUosM52w*__WISjSRJl&
zH$}rSP$p@v+sPd0;9cEd5P%^nqy<=Fwm^q<3E$6%e(d?<a&I98!F9R{%rO}XrGx@5
zvO$7b^oM>o4mcN9)MlO4{5%RYc9ifGPz2DAW&qOYBX2L3B4{I}<}e)yt!U!sYp@bB
z!1%g+l=G&k_^2O7V(IAIaCT*AF?iYphbKuSCKh&O&YdSAj)juSN^zB9l#4p4cUbx&
z?9k-rbT4?Zna?(L#a~`Gb1dxZ3?&|^WS>k$WB~P1*I6#enIzf@M~g9%AOxq}=Ewln
z`BX+I&}`g}^7ZMNO3`d-DAOvB4Mid@{EnP~en}O$4j&l4IyyT2!xsZ85~E%nyRU*+
z?|Xu`2?$f11S7%f?_};x<=7trn)E?2!nd))$hg&ui)B~%-VAzJ(fKZ*{rMt9{H3wf
z|30wg^eBQH7mIkCo^S^sg#S&mVkm6q%NtjV*+*lCz%qJ};RY8Z^=E#79Z}Wzj@~db
zNH$C|$oSTHAICFZqHFYa$ENt&m_grwx>7LL`|~GrmV$)RLi!WRiEK6WcUXQQYL9S6
z96s8j44q{Cj?Mi^d7faC?mS2VgX?=ChLrP1Y8d%%YV#yMPluv>DttEs6g2;>rYQch
ze(j}Y3i=Nz3c4pxP(6k9eE8*H=iho2;1a8Ld$0jK-@T0$lrx(2)pw_a7;yHQ)S5WX
zM<OkPy{O3IxgrOI`soIB`}bPTldA44@{(nT!rReX!;bpXXEnazhoxRwYJ6*p7(+34
zHtjsZL0`Qz>;*z=fZ#|O2ryHC!5KLrsh1sCWcZrZYdq4O_<bOV#}j_QxuQT$Ut4YN
zb%mU`a!;&Og5Mwg^F{OJmezSyGO4aVL2+AZr$&ixm<dSSu}YAYnQ}Hgrc&e#6#+Z9
z5*62Bom2eUkp|j-=TQQ*zAS_3yKLQ7+rD3*C82I9DL|7Q^3_%l0C9QoHwc`w75PN5
z9Zj*tsRKZ2j)zzARcd-=W8D@l29kJ3U-%gB!3tSBjwQZz)v5VH&FCj(vAcoc7_Y^B
z_Mx*Sp#EGHWLYD*9dY(Nk>d(qfhR(<jK2Ft{tmz$h^kv7cz0lDeVcO)f+z2FH${50
zE96r01umH@u3%?MEk6R_eg`b`Zj_9Kh;xC69Ng-mkH67oX@p3SqmB=-KAKb--Qix4
z^<vNDNp2#W7eyN0;?vg^+?l-9N6*FW;Q*A|%XO%oWJ13yVY6%AM`E`G$rU$M1|UJ+
zrDP4bh?MU2o8MD?R-Q^t8O<BWo%V7ta`8)LncwOsULz&qNml_{=UrzpT7Ts~IE7lL
zqT6>Ipb~tqFj0K61btJYtoe$3aFM^pbslw^2WS84UwdH&7d%Te&JhHid~aE?c;A8u
zg&QJ{g043fNXI=pRyA=hM)3yYbp6Wc2&NqBvGH0#eM*ej6%wFgfZA=2d8TpL$c*tY
zSQ!1lf00LxoX*BnbKGifywEt1wNwSa68U8t0UWFpZEGieN55<hH+t6R{8<|$%M5xX
z{qeZ8f9Z4BQqcZiixk6ug`}`D{m+)SDK&B1O;)t-nHo9SB;bfW%6MQ0^K`JOJTp{G
zZU8@_B{P!*b9}+?<{vKtVGEWk`*|Q5I5ph2<Qp&By<I(m-Q6XKn(7`ctL)B203nJz
zv8!DjtQwwOQQRzJ>ng;JK>yrYNW{uP@~*ZJeaT+NrJC1Nvnet<yXRVxd4{U7VKR@~
zn0jtpw7C_SL7_67rahng9<fLr$et(?&lM;Gbtli^S>Lvbar#=vC8@1OMQD(ldCgUy
zm)8dAwQ^kI2klzNi}>2j?jF(Gxem)4`2*v`X7zjIDSZcJypx>#(l9i=3uBt&AfSPC
z00Mp5OZF1?LoTv;|2B4wPV$p_8_b-@3{e(IJ{YTjWYeh24+RG)vT&9=SBelFSLg$N
zq}uGEV*lAt{)*;Z@(qTB_ON!ncb-mT3=B08&sgW!*m*Kx-%!K2Lp%FeeQld82P%yD
zSIc#>6rt8+DUy6u?4G)iflyPkCIsnL5kf?qo)yy9Beelo$V|ujzDDy<m2i7k^`*3W
zlDtMzM>)^&(RTO76i2ZBhRdhXTHrwO26osDLowU5O+N^@%V4D_*)UW<$8;=ocu-R!
zA^&47)Ef{X)PkQX$PfxGW$>$-MOPgcm%l0gT|<`yaZri}JbNS}v8N1NASv8sWJ<>g
zz&!?B1gbBH%L=>*DO-&eO#9?&kLW1TvM^aQod2x=orw}jU7UZHcO~X7X9y<_-97kL
zKIhq&_<~n7kQf5QD&8%AFR)FDI8WqyI|Q~!%pciCIS#+Pd<;Js&k;}@DDuq(8c5FI
zR!lf^$~fab{-fe9<hYxpQM4haXAk9C3n{4%_7Oz*7Y666*uQxOXqzga53E-qW|!Q=
zQ`^3i&nyyc>F1xR8Bfc)=kh*3DmWcDurA4G=Lx!_ltt$odN+$0v8@d*n9o1@2=MpF
zGDCaqJcV@D>^h8xEY!bJ`H_H<)*t9>CGV%gKDz}JZ0DLe<hek!WTOzGhv3EudW*Sj
zMsk#1An76ga|F+HXYfwjVeK_l8eJifD}1ev4{L-scuhNPv85!SjKb-?gO)mjP&CyO
z(L78qbc8pu(TMnpq>DsNu#dp{r2}$!2sN67m!ApRr)Hf&H{K<QWBI4K;@8~z29@c_
zG(zUzNk1UeVwqGX%W!+XD9H7uMtqN=0c(VYD7Zb{RCq^J&$(9BO}eoO)(z8ZU??R_
z<$CMsJ}Rqn9ZTNE5Pl<TuxVn$pw<}a`iacr>8-}3u5{*)wz_eTA56z)K9p3!Mj@<n
zqniL@Y3IuKR1Q4lRdn?OGUQ0}OP~2i&k<7JhoLEDWe_tr#%$DXr?F<4b*`ou^Cy<F
zhrul6Tlf5Kwr_e_oCz!^>^y4k*j-XwtR>j8V>edx3{ejW1EsctI5#%fhvn|J&L+!s
zL6*@i6KRc;l!5S0C!GO;UX;ZTy-uHn0zH*3iS}Y(!(~Vs5Y(dmO`M+)c<*$U>ToVy
z;0AUl89veCgUa>(9aDio6R-<5`~<JGmMs@7RLocjteQu5Edc(kAKildSQgXXVM@{m
zKYs$L@G(iLk=)M6TGcJOh0vp#?zE~ffKxnsMqS$^h;xfswo9CtVv!y=!6p)5U-CIK
zAYq=4!6ZFe2odi@{aS4KaR|Kh7P>Ok2r@5#qFE9bhs!@bNQ>WA<H%P7_AL`JeX~F+
zV|GxdX7+LD9NljucJlrnS4gKc<2eY#gG_mYVpyF<0gQo?+W8^8xY8I4;aIRn8Ct$T
zc5)5~=f}~aCxAaR1Y6uhH%!BJz_es|Mu*!T0U@5rdiI9d=xkl}T#F20PNhIQCPl}x
zl8njtOJG+UWLra&fDKr2?zuM`w6}K1gW9(FZsaAanjTa^!_!D4hE*Dc!SEj`EbrPy
zp_ZC0)Co(2B!FQ`+FUL2-T!LW4RMy!#qy;q`bFR;*2sHQmS@pps`VNOk}^4?j|^uz
zPO&4C$apm{<tjE2modY|(_1Psq9WK?35Gq$1xj185S=B(X^b}{QMq8FA&i{my~h4x
z5dOGG*rEf{FP;v`B#io@(&NKtU0kxjrla;XX2~-5&RMCG#0GrO-+#H5XM|$xAmVZF
z9#?~eSeX~l7kX)NE?d#2-}fTM#9qKp6k80!@g)>Kgg-j&10+=}ocCXd^<Sa(9RK*h
z|5bc@rK)bbz=!00Q$tTT8h37d0~!|c&*v}nPu{N}*96#D?yn0tC<eYe!1h=P+iJ%{
zIYN>3n7a04+F?t;5-mZF#=qLkx!hbkp=KNvqbwK`A2b5A3Z+JvF-Ali<R6}19iC-*
z`Ad&Pl1Bv+-HI7L?cQxijJ?_=%J~`g&ygdejD#GR&gjMvyl;J}LL!pNW)pN)=4|u<
z19eKO74B(|o&QaE-V3BGjC!1zu~sg&j0r$luQSD?FxS9ZFmS$HjyOO(alw##(MiFB
z&eObLQ`@a?W!GslS5BOeCuWFV>F}n`n);kHiD2?DZsDTZ<+;;w$CirZN10DW3dd@-
z^^~5@lm+E$G`4s!R|}a<e-qZodP|%7g6czSo2Xb+u{mxzUzN}-Qta`(8Q(Pi*U<H9
zA0=?hXMoCOYo{Lw4XK7BFUVclg{*itUYzEY0?s$#6c9yB{P)Z07~?4PbXB@V%sm+m
zQqq={PNhmc{Y7*X2A3F;L`2<J)F%A}_bBrxEi-{P!_j>qbECeIAZA;~<casgM|cP}
z3ax;5u>0M;gp6iLRp!(=9!QpD0#4rg-N`3KhEUJSyF^OtL!z~zm{*qloxmD*ApcFK
z?^BxlkN|Cjtpi4Nl>Gg{M^Ph9>+g3I0w^)krBA?{G0wSX3=p0`EV94LAlR>*x?5Fz
z9$T<;?Qp36reHfEV&T|iVqy_Of1-MhO?Io&181G_g$E8J=d%>$y?+~!bV(FEn|+;K
z>`{A*)uRNxLjNmb&2ETAFc_Pc=+)9j%h#4S^{>Q8%hsazmcKpcvCu)KiJ%CmbQ<O0
z<^Lk4$8NVE<am7W{ri0ZO%#P&Rc66Ra-hm18V9Vynwr|qJd&J5FaUJ+p!B0zRZwId
zSXHW$2}0rEyYHcnf<Ryv4<My-cDj4n>Nyp8+tb{J^j`O*oSB_Idqa&Es1}Dd-CwH#
z#SXX)upN&b<~nVRV(YLtG8+TscxSJhK~u$*dMocw5W>>E$>acLfE?eL-1(h{ZqzN^
zM>#41f~-}ulNou9T=f(|woDs-yRb<~waNn@4>)j~6;$<ts2!IMUA<GfieOa|Dxd{W
z<5kk}FQdk_ci<@J=?$cPquvabD-cb9+rB8r>b<w#!G#rz%1QR9!<KGc2GcDbao0~5
z#fxJ%C25NlXZr&Ciaq<%6IepmmE=GF?EjM0vof&%uYY#Y|6PQAQ!_gUybPy{^80ZS
zLfjWGym?I%1*{(6v2?z7UY7wL{{1|COQcxeY2S<~Ss0ud!}Kr(CmkTj7x2R7boY65
zef-oB2rh*@Er7jw1&R#LC?l3yx=V}5W348p<JI;-PXo<kjh*EAP;8AKZbH6M?9+5a
zv~{^L-TArP)ZwJ6a&&7%x*a8L0;XTEa#IL}5--Fh;2bH3@G<XeKBX^|kHx%_`_^bh
zWzZ!4(LFWpMDBn$k}=}Y%BFi?PtfkbCewdMdZkn6s+h{J3nya7l^&MIa^R(@d{vOu
zK{siXv+Sasux#weH|u*HC@(>rNRQPJp|aWpCEF8a%wjb$9EqW4PicKWv|mT}2;A&;
zm!`0d_BOuXrW;9sE#4nnpl@D#<PX(H##WCms?h$%el3rF{$|p+X#SEC#LD&SA7Ee8
zh+HJ0>)KEtGeb916LX12Zg+Tebn_}IYW#9zK>MY({>#C1z{E~62U!0y#kh)9HTs%S
za(xJYc&X&(teP^CdwYB9)9c_(RqwF^FnXRLt7{fJ@EPKZ0v_nPsK6q&*8sz_{b@ta
z%mjL!hbE6jVg`ap4cTF}Za=CK$v`W_-ZDXnKLl;KhY};TPtc3n=RezCXi<e6y>13u
zzuenU6qlB(0I6$Y#)3Bd5gA4Z!5ug2&Q30mOgLxJtuo@?gYjF&W`Z4AN(YpZF5Mf)
zwUu2Bo`XGcYBl=Hx%3`0cYN5raz4+Kd^)UEC<)*XT9Q*OkkADstw~UkA3T`gBFl>q
zOI2JA>549Zd;Y7izM1tsy>1(wHRo4&@##G90&aqq(2B*4)I0YH7<`ymeu|@$j!SF2
zP|#%ZudF~+s0Uk$^fwpy(t-HFKp2LdT=R}URkYygC27D^Zwg)0;bB#xT=L5o|2o)q
z&LpkSu_+3p(&r#ZWgCIbdF^MW3F5`-d_^<G^d)ahaTI(J5h03VnY8n>83ZTA@b2F5
zcR!1NRg`xoLSZDw5@*}`Ci86-GdfyO`!Sh8-@tO7G5VVh_lhU1%DJ0BOQ4vs9~VPW
z=xezamo`yAQJc;uG4H*STdAW|uR{@<K_bOV&k<}4EB~$|EMq4rOU->U2_0uTfz0$B
z-YvA4OSQ`rgL_gPVAY<izpWm2fSSXU9O1z1x@e5S3d>&|(+Lj`47EeK-`De!1=lhe
z598fgvf_ZlQvIi#(zUhmzD6K97T@4;YMT&KoMCZ>LKJ38OglzpC7R{W797&!z|I8z
zNDGhZ+bPC=_E4x&qECISr;EH^*zPPmo+F5;6hbsM@w7nJyNy*T@M0;Iy(q?uuzXhW
zTKBujD`JGtAHEET7Thg&YMZ_Y2dK<<E|UA*C8cSWjez(YiOF9-V`JBtNSfL(G-1xe
zI%mX|#wb9e1Va>txnXArj&v931xC-)wI3y{UKM~tbPTs*>To3p&@%g81Wk@PxlDNz
z?V>&#X(O`Uf-<$Mim5qi6CooVICT4|iqRwcQjkEM`Ainz+fGU9;-4#-hMFEE2|7p$
zF|O!swAC#@L*}L@^hdz{aa~gJHO=Z*eXVN_!S1Z#OWu%4hT>KfV=t7<PklBygSq#D
zbMq!GT%d?pJXTd)iEILh^>q=Oz##(mtitTtK&ti!1#VX3`NW+Cs^?<sdgO)n13BV&
zvqDp?^Y;hE7cnC*5GKLM!f9l@I+|%RRD@EOj;Y?41OpCA@24lbAw?(!uybsyB2-)5
zn5g*gE5XLxpyPS6^o$ycGEW;z`CZFL|CyAHz4No`;BNe;1a_4=@D(?<^GjeI2ruyb
zaWw@Bl3<F>AC!0QdL*+ZSVpdFEuvE`qFe3qM4DyUq*Ew;9=k7VI)HB|5&G&qT|#!}
zOL(_??3OxLFp$Kl<&gEmy8ZVXLkWrVt{KD&BmJwPo@b4EkP3ce7dNZofkA7V4ih;j
zEnB`U@rP$l_Z>sNAAKfShx&%71jW$wV9*1DstTLIcoN<Gw2SwfIh6aq^1u|^Q<(in
zngwqxJ|{hgJTAk+-6LG;l(&wBjgO_6;OG>yhWO7ZbiZxU<7ZJCULk<flL9{IgarSf
z4U8LiSY2ijbd+wmbutZ!j~ea@!+AMcx#cZN!StW;?pt!4{uf65S4RjVBLm0(j!_r?
zIzmu>vU>z}WQK>Rb=~t$fjc6k>4fPzQraQ$+Cv2D%tcit9iB!D*dJ|g;6mbshs2vb
zbg3Q05Ki7Q^4w}+U2m_}HkQ^@F*U%kCk(J5E39+(3FjFO{=leY?3Ty@&h~jLZo1o%
zxH7CF&=2I{xGszT`uIH9KYhE`B_bmV4`NJ^AMgDm*?!!8n%Y0@o<OU#Jnb}jbbMSt
zj<1ehQYw5pk_c+H=R{gBB0104Cs443`#v`={ZEpsdma}*mey)aL8_1ZpwP_7gplsw
z8l?i9-GZIaMQeHC`S|fo`3?QA)_8lAM&*U`d}*0p!_a=+ldnvRD~u9?Ka9Xk)qpFS
zBSB9YrN<+-cdn(`)J1Y2{=7$-irV4da%D%ON!B~r=_s=qcmKHgtXlEjh4P|A{LGbH
z&X~_)bTEcE-?7wv_cUe?;a!S}E{yn1vr8JNQSi4<z>ol7Y|pPTMD1IrH5al`n>tv8
zm^Xq(`>zZr1RAc*>defmD$KKOS+5h#$Jf?Jp(aB!D=F-|YyLJG8kRSo;$!NKBY|jz
zF>B4`kyloH6>KQ`5}LQNz<WzIv6@x$(bJN_kWwtf0l?)YelOH&hTgK?Y=Uu6r3D@n
z(II*WE`ESm`x>=m2h-4EE=7-?@}^W}4?rGBAP)?s(e>Uq^*@XX=}>@660LSJVL<`-
zl>fS^E)seQ2SzP*@~}7Ee`OniUSMk5pB(C}C~q7ouVgdI;t3)cO>>R1+9UM)IYS|2
z3U=#IHysbOXWKP*%x39{r2E#|=q(ppKODzRTU`uaRpXn>*`9rX*1ks=vaG&m8f8Pl
zK_`uef=wer*7pj^=Np$h`m=ZTdn0PgWAv>3&OiLn-tx7aNS>N!MfF3Xj+niSCGAt7
zEKYP7#bdZ4fb!OWL1=TAR}wZ2b(8<)UVr~UtAZQ>e9hb&Hi}ku7yMZ~igN2rR6;no
zQB7gvt^h=_(XDYX*{-FTP?EO+fBibxw)rHq$!@}8ePuJI@*>2FXz7T>L4iAyx(033
zENvA^Fv^U{C?KoVOC+I)l4bKwXEQPw`v;*9Px>0`VDk@XPdAejN^@$+@vK51$<g<+
zQX{U^tyqD)aAxBd@IHw00QPT+=<;@r{Oe_G*#WI0><EZdlHiVRW~4SCI(r=C&QhV^
zEiY+MSTE*6Ams!}zUyhExdV}p7#aRx(35-|@J7g+doSM2Sk(--+zNocLobk6v2l0h
z`~7h1Y;y^(cpA{p^3cwl0)5}$CXz8zAek3Gobt(O?NJ_Wxea@yxqiFBms=Xq$YiLD
z>F`FoC@_8)_aLbOipbEyr*D3uG_xvnL=oZCkG?(#+}@FY6}d$@eI>4?E-r5Eu>mLs
zy~%lT0W^OA@u|gq|DP#ZrAE8a47^D_98t(lN`~e+ZZ$44hw)H~?ABcW17c8mPAoQ%
z<ziqq{=hM61iD)jV9*DkIYTP=eDuHIP_lA=`3~~H*2iKWnI~3K+p02VOSDfsMvRqu
zmB|RgMDEgn(3A!I&wjx*4FLawLf}U1MavXZ&lD2;xo=V2g-{6_stwq1pBk9B8oOXh
z7CG3y>pxtB?Q#1O2zjfn*v+GYW_y#aYrzoOpYvz$phL1hTZ9T{h29KC##=+}0reIX
z7?o-W?6}n0T7MR$vIt-Xcr}O(Y$MHWB9~H>d$mUEbyPDzAh8mPLqG-gUQT%DC_W9h
z>4Mzes)0=Zg4xEn(A208pYjzdMzOlMbuYeYgi;$o`tCPV-Gmp^T!wHwvQcV9eX0#4
zJbUXa+RCYFxHD&dTTS;?Wb-w57XK_UYh^;m$ZB`1diUpOG!j3Q@ROFP^fumdQ6e>E
zQR9swtAPtZOl<CmNmA<&QPg+YM@vt5s=B}f*5^`=1u3VNRiY_lUsyLTSyHBwZYK_O
zYAVoU_Ui0@Pxcq90RFU_D)%dFZW;Q;fP%o=x!Ep4E#2tzGXw9p3%f<N+YCZ~7Y&S8
zXfF~h;--}VZ?Zr^P`EJi>B&!hik~JEG($k2Al0Ho*)$>9%MAV%4o~iB?YC=|H+<;v
z(%YZMvK;Xa<@#oSx?3a$OsyTm<zFLoSEt?!{DxnLYpNsZE1qr=NG*=xpDlP)mju&%
zbK%T0_r<I;z!^0XMkndko+u-cc(^Oh;531*12q#MxYd3FJYDM4Y1iVbgRXOoUqe@k
zj(Qg857{K<+3bnXa?7gEm7yI?8Wf8EM0D}j%TOO~hZo)JWLRX&3(xN)?yAtte&O(C
ziBN#m+$~Sp6!|aBSTa!Fqc7{Y8eYlwFje_UNjM&PO~IT}LQA13ZLM5>i1duB2b!WC
zJm0yb-FLpO+c<IH!CBY>3x@$V^B+<3A<KHc)M0lMgbi#ltEL$1jMQg4`n498sOw#E
zV7jKA(k#?wcuRA!B$2=o147avzRB=DhM-(UX=6f#KB4pW<~%16PZS|$x(jnQ_@J4&
zi}ASPrc;FW6HiwNgx`#RMI+{4S*EChML`AFpsjCFq{#eiq`j<bUu!rIG+f_d!Mp9;
zB1o|@Slv<(X(op&!Um#sJzPr#*E39UZk-fd1BXWmQNPr#B6K%>=$>zI8UnrxiTqc>
z@qX6B1gJ$DXqagnH{#C5e%RTrrFL*$O629IR-@kc1MbEY#rfE%P*vw+l1npI;Pg+}
ze0zEtbB>6tf4CmF9=SWsMVAF<Wp+229vt*?&-QBXE<f71tvEj?_Nbw%RJ7Ag+iUxf
zH~$V^*11kL^M1Ta%CuPC6l+rtm+TC=?{}&U8cemykiusiuCrLN*Hb<asrk;T4lxkC
zg+U;MWite2v)Z*<ZD`u$sIqIXzz=zQ^!}~<ssL3X>w1@KA=Sr*P9|R|o_vxO0iOQX
ztN0Q5^-X|rOm^@lo^neP0eIBGAUk;fI=@6jVU@G`&&SVy2`xA{{^w)SzYJd6Lspc3
z=EKuwMm_siv>`!^RigvFXL@MTD4+)@kTy*h%RJGN-yHU}yxtMi64GP$tpQ_F4x?dc
zIm$!t<va?^?Ga{=Q$r~k5tu8H>7z&q_sr{?BE(UNQqW11>zDa|8598|DPa=kp93qp
z*0kyvJ|FK#8$Q|=*tR;}ALsWYQ$9LclE}0~=y{YuzwE%RFLPnf*`ng@`TAMLq{0W9
z1-X^nbhj~6pjDZ-leB%PZG*e5aIM;YrB{6syR?O(BpBBcMx~-_;0Q7%31{Fex7)<H
zz{EIehPs-Ke|)MPd-+X`bgnhAJhw$>l=BH=oqEj&zCEcAhk-FovAQ+nyOX0aMH<a=
zW;{x+-N*+I%&pgVjoFTuUlPcrZtTSml^%s0A}gW%Zg`syS%8yHG_ire(nu%QAC!x7
z@n%9;KrRq6irixtG)9B-kVV%<<wvPLa&1IB2&LAd3KIM!`uP`+$i(trqfn0OQQV*z
zD*~|kD$_7Vy#KU?O&k@oMbFDfG;K2bi@x85kscj|{WJ$csP3|W?$^gVU~|6`Kqhh~
zA9B(3@0DY08N#H<1R4=Srs$&6z0Y0LN#XM9_E1YI9M1@nEbS_x@D&NTT=*C>prJJp
z%_O(kr&E#Y?Mqy1T>m?=bd)RcWOVb_QlO8~)1i~CVjA|E1a#!U;mHa+Pu;!Sh1cK0
zKf0U&-qd1V=}%m5a#<)A7}jsQ$BiaX#z13M1AmBp<X6JTc6OK(nqCDM1O-ig5t>sc
zA@f=z39yrda;<U}W%x>xD~RzL7xoisUQ(yp8Dvf%gaq;q7JbEjY+Wc$O@3q%GR)pY
zFmhDD!FZpvS%6nY7b5vaGmnz6HnW$v=dH9USr_4PB>zYg7c%@kLIIwUnsgQ4u1VTA
z^J|>!{z;Axm+P@TFk|H?X@1?g0uH0zZo1PFi+(2v(_D{xpr;jxcxt)_QEE!em~<T4
zEdZ``1e=uoCYsvivtX3-kp|<1CRy+#YwnLmh_&&;lW_My%rnMB?nXUErGJnxEdJK!
z?~rCIcPsGdrJIX<x)yK$mv!%)4`gHBCoUBMMRI0<fTiMy^LjNEjk%=ju)G7gJPANW
zShaO3uMPV=V}EsT#A~k>vdMbMB?6or<9RTaIdb4n3t9xhdSA;iyvj99kK`J-K;ytx
zx4YV$P+y3cpCYbbI(<9Fu&4`&kUD4va+zNmglVQH79LULai**sQeSNEglzsQ-*3!(
zHAK*HfGdU|m_H2YMw6j2bH3gr?i{NJ$$B|U{*}=y2ev)V@)LV6_FR(&$@;i9Fv1}l
zB_}nn(MTl9*kFHb#c7TizFjBW{*%C-GShDXdk{3-UWDHXJnfYPXkgNWZXo^vI>D(W
zF(GBj*`kVusIIk%1~Clq622sNlL?zC%g8JdJvR@-C^}U%d%|#fF%MGYj?^X`klmOZ
zkQ)A(1w^226KKqsZ!R%SBY#u$Fx2lY_0KYr8!E7!iGJHQ4<)ftWtJEU_k(vwmQvia
zT-8li<TQ6^UMSXT{u>ay)aQFJRN8K;C$fWLqzWV9r*2?>3QeI2XYlR%3AJKQ?#WOa
zgZ?kZ-od#NsN439ZQHhO+v(U&$F^;|W81cE8y(wDUe2p?-@8?(>)vnw1G{$B+H1`@
z<{ZDVV!lfW$nSYr*_0s{1d>q3$1hu<cAAKoXrp#6ZO1zobTy`(>5W$WY!!33OA#_b
zIU)z30#>j0qUcP!T>douiz%|UeSmGU%CxfqE7RCsVf(@>(v}G?eO9f7PEc^+qiA5V
zcZ*kuOl`d`gOe#z&rw=s#or<f&C-AJj5rsv#W-xtOpsZMNk3KK^0Fg-k7EGKKFGS3
z21jb#z!`u(=gPqgbl0yrN=vgO${lPAYHn&ql-(C^Yahm`VJGnF0bb~NBv6Y7+UAkm
z-}8tv#*qF)q%t0eu8bHs+?XK$85MdQHHo2>H7DDLEXfN-W*3iqCK%EVhjiZ2V1Ef+
zI&BhUc2_G6v?RXgx*gRJF2r$E_-ab{D_Qw$2lG)mP|gd%X&1n1&)N8ies73XJ-W|C
zOXEc|iE0*LS;FtGP9^A@!StJ8n-qqJ<ZCaXFSu~^>9X#FmY=tS-`N^Sh4DQp{}Z-9
z;mUUvI#Quwky<3;bjwA(xl*{<3R}E+a}684@TPw4*8p7g6bGiQ9K&=?0?=}%jrG&P
zIbcCZ;rI@keYea3*L7vC(oD<p2g{)DiFc+EcJq|0r5CJSmtA5hmDNSaKpyWC$HVqU
z-RjhHy3b0M2n&Gfd+3-{8a|wAoAYqXGrHtR4oPlf&p5p5j)5@7ZDqP9jV;u7*K@K>
zY27}A>Z<ZC@+Q5>F?PV`0Ob`{6Ol(T1Zp<ZpK2aE6V=;iYn>o@5~o6Yny;(J3fo#X
z4Sarb==HtlJaXMV`N--UX{RCMDQId~QUv!H*Etp6%a3(9SUJPK7nac7BYZ`%baR3Q
zAd+gwzroI)`*m#35AGVy$oR#5bWr{IL(@+Ds@rCpnP%IlTc~WyoV5#jbzQtDBRsV-
zcvozE$f}~-j|<4;SPzTy7sKS0fX(-|mCZdIZ|v3HS9(en&Ad_AtR0GwDPXa3S)#pi
zm_bY>5+~a@o7kt^DtLRDp#%dniA+q~k|d85@fD~=wlqF?1Su*n0Ve2yFhR;cNV4oj
zb)T-b7ygJ=&WnFl20xk7|AWrY#K`y`dvmAMH)1x~ki2thW;ZP3EJ9F7rTY2SfVt0y
z`sK3Z0z~T>Sdcn0Yw2)xOy~Q2KF~A@MIIY@1FZu777fvDmyar!luHD93k23w>)O<G
zd*R*_khg`2)NX-lgGkaqL?`NNf%1pY6s&x&ecey2L>+B2^g+skR2qP7eH{#~Y^Yt=
zuqz4)r^KDh5}CWbUtK=eY}@Q~dcM;>b^4fvdVPLvO`y~7e4I}`?KB^Ew6t&_JhkWl
z>R?{N$U_{p=mpA-BZ&+MG6@=TueBBk2WQ38Z;A?bUcm)dATs6I#9Q=JVr}%rc4mgU
zHr9{{d=N8gIH}RV(tK5Lh3C+2qIh+oS;sa%V(HAJqg`^U6$pVA$=f{T=U2o^mTE3w
zvo-<FVv&^q(<|_0jMh<(DqrL_HYjq@#C(AyEnkX(YVSaMpXg`1dT?>av!qp5x9AXs
zagn}AF~zI`eXtZAM~<QI5QF0|HCw#Gy1>|4)KQFC#eyFS`xDF@SUjOqhzPx(l}HjX
zo8qYvqf_IPj5-<N$Z^nRW4t$)Hot}HTK?+*GPt9Rw#tB^`S&?;vkG^<;gSlLs%*`m
zWR;hI<c6vP-i`sZte@f?`lccrb^!;edkQT&f?1RZi*OOR3Sszww;m4sTX7;tN(NX`
zzmPkHAo|jkPR?Mpi{;gBicH7G-m~uGfy&~`2s7eWvZxMyWIzhPJQOynR5Y%K{t8Zs
z;(ofPWn`4Gdz58hee&>%U$t>B)XoNdcWd7-15Z3-of^{ZuHP20VJ*#yvvoX^zO!VO
zO}yiVBC=&mv%5&z(}sbs?PN(1<BDpRf^UW(R%~e<Kn$A`-B%Hoy>On+7OOIkrB{)7
z0`zqimv6&36c4{FF+4$SAh37zTP!C$$`#Bo4A7DGF7ql5Z*CjYm+rM0jxfWSqzSQZ
zbZiA7Gq>~?xG_F!t|J%xtR3lYh8(?m`5}DR^Lq$3IZvaMXXIA%s%jNoNj*C6+XnAB
zq4>~AN#v1LzSJK?iN*a4utIx*vKU8vu+W~gaszRSyF~gFHUljJ?+K`Rq+I(1>;!O9
z4t9HnczVmu`S&ZOF^bU!YJaKiObpS^<V|uQ2IN6DS(?mkSd<tfp{2X&ejS;Z$yL%`
z`Vvg*e%7-sAwf1$8N(7&lcuK==soPZ3%RumQNHAeTJ)<?1jhi5B6EY;9p&qJzS<dy
z%N_!X`;Qo)nDC4e<U`sJWo+TMf*4cdvKVDxRv?^E;}$VW-dV9q6CXRi1BjQLdO$J!
zMaH&=a%j$WKDb9obLH9BpkF)$xXztGJBdKy!!$)OZOSnhnybnkqEU!OCYj9to#hcZ
zHI%qJ<54W=U1*deTmda2q2O}JDKT)CyqH(#oSWQa<tL55NI*aMJ6kDXc+O#&j6=@4
zicboi!ulXAX&1DeM{7^Gqa-@yxxBuC#yD@O^cUgU8b-G<QW36UmJ=7;G2KvN)L?uO
zOZ0}keM>j1YI~L;rz0)_Sd;CeWM!a~KmH%4c-8XeLMzUDeH7p*^;&-x+HvAZ)Sezw
z{k9x@bMWVW;#R&3o!;PyJ!H>H$2nQ!-gX1VJryt0)FD+|h$m$uqGKa)KT*!5a>stm
z>Ne-U%p-*6{N2amM;v?s?^Ma%A|<g?>fXAJ#l%9<kG(g&@O`f>JK#BI9^Xs0S&x>U
zkM~PaBsM_qNzDD~NJ&%4DUNx<a%ci6@}hVTOshKH_gcvh6>+RaLs2&{PiR3MQXul5
z!rqlG>*t;tCpc0HSYPP;Vh#sQtfU{G-<*TTi-bfgDmd{p^{;8c;SjNbS>-l7xQE9a
zn08PXqbt92?PC>Bt)~osAZRoo8MDa(Cdi9$q~%qO1zJN^Z1A5L$Ajb8m3LdYI6o7t
z8gi7KHBvt{c610Pq}(Lfqvb~@-h}#_Axm((D)&&p?_c$mk(~as3FL;Rg|}_ncbFq+
zr19OmODslzwiUHXde$Y+u^#PgELr;DEj1QHhLFjFag53sJBh>d#HOrsD^ONz&S8Y{
z<9m&fS|c*U#)$`N6)g{zov9S28DkfbYj+ljwg!fyrDcZ27jCBDMgR*Y4dA}8%zJs1
z^NuK%DC+?M9*hD&h*Tg<KBZEjTVQWyn9$+6pxM^^C7)Gn3Bh&sXVcgFmj;yPvc=Ra
zA+$8GlcpQM0g#$AENZvnZj-^SS!ViIbgu3AqPQp<p-S6MO#79w;}XLr?=dNmEZaur
zg>n}LhokwxdwGC&__Z#x3fe>ApB}vOh2lq4$j;j!HbiZqRDdN{Ej#)tcMR-_Ovki8
zUQ&;70pnuGIO*f<F$2mBk0v2ELifLT-ti+$;WT=r?kWYQB;Yi97uNiG9v8-WQOBCN
z$$od6TF8m?t9eb$K+DC^`15I^X0e^pO1tCG+O?G=KK|kz^hiXzw8QmXC0_dEHY6g=
zI_7HvMt`)#G2u?V^$mvcABsb|d66GOz##dQ8H6Bn^v9p}vZF)C<0SYC<>pmO2P)AI
zuisD6nBa%k7i*eb#pGRS^mqqeg!zS?*=L=Ew08!tc9=#Ei7ze|gvi}I6Zj$$pUtnh
z`W#(fQBvHsc(6^G?RGyp)%ZKizss}OQ-m{480o4!!F{AC``w#Tq^V$Y!i*{CXwA)g
zUg1x%;u*3?79V$BfdZLX1?l1fHZ$BIbP{8Q1+a}3GFPtSyF*fTS1r`cE{6`b6%sJw
zmtEZbc{3jWdY*VG#Jl+4L?$F=={Tiy<No>E;Cr34)%vYFee)66tK_4<c*!>vupM5v
z0xaLaav2J0|0+rT85qRD$@rg6Y7RBF95-4KeV_IE#62wgUdZO;@aN?cB=H?qL>8tO
zQ`_=LM@?giP!qUvHnu-6b@T<(bB)L35@k%^3=t4&_0%tHU1bw9gUD)TzF7D>yoO38
zVHh*cjk~3S<dPPHq>?97$5h8r#>QS{EHb_>4|9%>Y2<r6pO5hAr<v2Icf6jTJ_bJM
zQ)>z_65}Eiiu{3<M$<Q(#DEu0-CoCQ=T6h(l4fN{nX}_J%}MU=j`P>iyft>RW?J)4
zXI$pktzYRo2Hlr}@Y$>#b+l7e*LDv2Ew$+1>dQKS6PScJM+J~mAVgV4^GXRgFVbuL
zPd}D{VAG4Lls8;9%17KA`s;cLk-&~9V7#n8`sxH~IUs@>>Ide0HdAs7s(7LQJi%5!
z2f}Q*;4-`R_({bham9xq4#86C2QFnjF=sa{I!?J;Mh?+duZ~yyQ)<^46L~A{95`$(
zGOIuzgPJE4v1_$F1Es*oo*P<Mcs5cQ5smd#YPvPR?L$a`pw`J$(UvQ#5r$8@d%=Kk
z4ojK^$#N#^PaH%!2)1T1K&>Zl>mf<W(1|Zk!R*O^%66(*S<P*buF5kcci%lJA-TDF
zE7D(xrd7vC&g}M;S_yrA`CS-Uq#=p)Q%A$O+XDH3!gXK=*<Q0Txa^jU1nt>bsK}gj
z^Af^@9uu$qogGqSyBQD8IO*KZ-UpWY{q~ydl081joQ&45vynYdA>G9@E{{1II3nJ0
zXR4F6pAO6zrmPJ7BEWE`Ton%Z_H_&dY|uhM<u;7aLT?8MBoQ5-o)$%&EHm;%=6+G)
zTEl<4IPtnR(g5CoJT8EF;|m$@Q^)6cE=0YW6BvqIE<3+&?EJJMkIE#}enLk=Oe9~J
zPjUcgQhw%DOey;vCcM>np8r<2PNT-JAbM^8$j?#Pg8s5s1M6ev`=S@s20U<bW;Q;?
zaR9WBER6Y66uiw?uN<hg?l=bmLYlc8lJP5yqmu?e(QKmvqVTr21~*?^vhM*C1)@_N
zE`ontj?!NOM<I^gm%-<1P=p2gI)p^!qvrmkwga<_JNV7-!uQ4_`JqcC2|&$_bGx2r
z-Nvi4q%+EaSlts<A$*9*A^q+H>`ueHqdhKvByTXqYQ8W}cFt<&?ZfgUgohZ{q2p+p
zyWVL}4Mh8Hi8W+~gabs%1^wF>$|>XP`tnD237<?J^%q?CuU#OxVzN!2Pa=5EnN8^<
zk;vYo>tm@R6W@-=VL~}!T~xeg-`O-LXY@qt1sIGN>vA0*m^JZh<PO^dd+5>j`mPr`
zLjO6v7i(z1A(@oCAz_ej>kGDNqizS9W3s`FjeO6CJ4U~n-4y#P<>%Yw?o-0V>|gu~
z0RwG{&$1kVp*Xz9o)q*yFxDsc&L0(E##0LCepb14js*K`>iZxLTi}>50?&<3k3T#6
z3`>C3Kn6!o{t>3z_xb7Z-UDPEoollQimuLQr*(u%1Wlr?23*NUlLpOY;lLG~wL3V&
z?u9bSb-8cN4V3wDtQi7ETb!)q^qn^8_MI01?iYiwJp)o0Hrl9qR?=g=cIzE#$;$Hq
z?<kZS+n+7E%r76Kjsnp?Xtikfu8&`k9S+33iK<>#B7O3#h(d$nxx4(%fz>62ZeEj*
zd?eC4i>2rJCsqt8^!NrB7gh|+$?&iRFb)oIa5a>|Qf@ufyhcxb*vmr9BZX7VJ64__
z-PBv3@5HJryDN_ZAfbJZ+Nu6Wc@qHVPw3!pE>hl0DyojhA3jQGOo_<ZW(W0{yM5Mp
z+ahhDOxB972(T*yF(z}N0Ru~|UkAg>&9F8g3$YvdR}<Zh{(a6%G(JG7pJ)PMfM*P0
zILHY52yc&S!@0t1P!pWHQl$?<#70QnHQ9=(g@zt)196BWqA};oNP7=?{_2eZxkY#J
zot&**V_<P@ogyAVagZLopm0XglR&2sLJ`fhSNtuM*-)!_z1Uo-#6&)yslu{>&E1$x
z6K2cto&i49UQYNz2g~0we$b}&A#X$>K<Ef~0{ZJkOvEsTH@hl$9_4(ERE#31S4K2+
z2BC5ny=nXa%b;IKph}zdbn7`6b70tNm>_O=k+%?hQKZM5#9%kPTrI=R;k~smAO3M&
zb=K%A=JqfZ?-GE^eqQlrfVQ3V@NssozY*g@mGsN$E;y^`m=h?d4Ee<4T*<LRC#It#
z;PbnIVcb<#PtSX)bLHbYu%s3QzXLO$a>KZ2UlxfvSgh0X>?!9huJUOye(y8XAWM)j
zNq`de>v`vnA*JYG^9P1G9J5y$Bw|<m<puG?k9c0MB902ZIri!g`iA5+7K#^Z17sdI
zFBw_dZ?9jlV(#7(rZT2$Q&dJJ8GS5;s?!62r465+R8ZznL&~TV|DsH2%kII*f9)IY
z0%^E02<@q@03h=>W|~0p3yuJ0=OQem)k}H~D*O>@*A^uC?)b}EM?I5ZxGC0%S@txO
z#&hKE2_e(IQj-S)y+iD*9iI;<3ewuM775gz7S))86TX3@qrlPr@o2NK{_l*h|6@WE
z0~G!LVt{dQ{KvlkA&vhpq3J~WPM<Dx;I{#Y9a16eERbLpkTr0D-QQ><mJ}rmRg%xQ
z?7W(ajp|=Q@4Ln>&vqQ_cyN>s*-46j162Qt-e|0){f$=Kj)RaBrLyge*p5U{nG^(3
z3NL_=JR}hcf=m2>sJW3?9_Nm#(<V7Nz{WxGm?qj2mJl@*rJRsK6vJ>K5;cJwsFWYq
z4D=loIKLb17x>+<nnl1ESv4n!6iufAnp7$w2bS!ryQXha4iZh6QkO=6X%EIG@F&Vo
zDHFFZ97bC%pA&SbxD~iaMt3qusabOPsB(3U35^3^#4#qWo&!3EgTYF?lcSe2Xq{r!
z!Wb%q1h~^bDF9^HZ3t5_1SN_21z=3bK~!p7aZX5#GZI0N=rAoyqPMPa3|V;GAP7;B
za|9IJdmD9(@^9&Gy^w$i^ZW(VIDldRK^|ljRjpxrAcDvZ)IHI!+8JTi#l%jb(orp?
z>_})d;35SC(?jpu-u!&8+g?Kf2&R^yl`u&fuvcT_GnhV9lRK)pQ6fbN0;C9`8%`8h
zIZ)!UJ1Z3e@>}6>&`1X605lqHqZiyX5=Y1Y^p<qOLn6W80~SLLWbJTrOJWIx8$@6s
z`=Go^m5xwF!c+>wc5?(*F(9Jk+5_)GB$v>~{%V+BF8gUhp)j|k{J^Bzu)>(miMR;n
zFb*n#MglHIj=BA+2t_mqp<+V~6TCB{2++`)bSSyyeE5Qc1I0&6E{nz6x^i8~B-o#R
zd^CaPwZmN~Blr6WI>BZOcRJr;aAd<)pe^Ytr;zE2wI5f5?(X4Yt<-}#Ge-VgtV?Yv
zGqpzLvHEMhv;mPXT=lx!yX<!FqaII{Uf+P%;z$US=0JRD!-G#}a`^EhM1H&@=RC6F
z{&=D0tV`xJP90?olZ-3Jps`k^6EItbOxfJiS6JIno{$bP<%oV%@zIN`qtr~ainGF}
zQdf=s#&0kyoYd&0S6AA=AA_~6-#0_CrdF+B9*|qd%RIOK?bPzl%-O+9o2*oBcxuT>
z>Cdm8PH$RYc|O*u)YoTKXI{Px_G4t;KRUnA7j8Q<mhgYKGFHb{&yYvn&!{qU=S=d;
znXNc}mbHC-Y;~r`@c)^~tYAc#x;Q(Z;}x3uL3}B{uubwR)RXOXq%|#~bH7BYhCNmY
z4t$6EMxGa#dOXPUW>#~&jMk+x{CX-<a_sqf;$B7hh+i_keZ5~ZMS!3iV7*Z#M`)@6
zKFI_1QWv=lUeHw1LKFW6bqNzqTA|5r-5|0ArLcx~#SgLI-hx{?K!9=K3Q;ZTiwI07
zo~Lyj{4sKJ`(Foj;+K(B(yi&pE@P^OS!<YCQB5aI&pqimH@tbgxq9D@j0_ceWt{g^
z(}y(l(XJT?F9NDYEn0XQncSTEHjFFH-fmNnRTOz|u9Zi_kjjfky7Ok@>(Euxomn#W
zWN4$!zQhC&;(4bp$=<_$>&RjOuyH#5j-h(0e2yI-5MJzRE&>n3`W)3^R2a>`RUkJB
z|5JVzdfHLeKknrq{hlGQQF3PIl5~-11(+*1{=&bLW@5EE7T1qA_e^k(20>dWt6<vU
zkD~#8-tLd)+@IEsOrh1Gk^SbshllJQdFgVJ0Vi;4A)AwA&M{QN$8Bk8n{*L<MCj>{
zvoSk$J?9A*tEGwDQY*e`nA1AsvI$gL0Uk4}eRKiFLe%y9*8T$D%*4*CK`+|&$mcdW
zGWbx3W9PEXed8Cm-@$APcX-Z<WK4jov9w}|O+rCngS9EGnR~z;!LjISF!Q^dOA^v}
zwK#9HfF0huXH*_`m~yKeN3=80Se?tZD{#X?!Wu7Tq0TitD{o^&mp!#A&DvU71im{h
zI_@70i@-S4P%;Kwms?xHZQ!^Hc6Zs;eLMqc@BWnakxp~xhdgWv#Ap(X7VOb_o|9hj
zY~Z-6l2$C+yTGwubl*^G<6AraW6jzJ0q3jyIF7Rr78k)<l62<#11-(V=LG0ufI4K6
z26E!ZE=+|gP{DYh>p<2P)~Wt*AOdyRp>g{ni^6%)g4~(y{wQ_tg|)w-JDW~*G@&d1
zqIowF4kvEH*tQ$MP+fGGSL*1nEW?mfO7mLgf(nt@xlP&YiJh*j!osVlBU7}9=X&J`
zJo13k0khMkkm8g?jXnct*76nM2^a>VB|LG{)6x1@wOqhUo7q{WFBh^s%h0|jFV>&N
zucJvYrHrFGlE)F}r8{j2?6vSQD>Ib>UIB{0j@GFD>5Wnr#+n^!M!yN6b4jMt3BTF1
zLH<j=lI*eDs>RD=L&EVRj<-zXmdD#+zO$5xtZVx2i0@t)uhr~yq2pi0N^_i^4m=}K
z$er59tH^SENLgsCZ>40GpDkS9DY>4_sbH>uuSBo`TIXjaf4&-s>y}Ll=ZtTjZ3TR=
zaHTbA$2K@%lT4v$ik|^HnF=>Ffi!ZAV2f`F2rZ4)jC;Dkf=dMHb$LBxhmiL_w|ZFx
zHamk!3t*2Y%dy7+?PubX?(3B_N=^LZL1$%S{GWpXMkdz(xC>wUfp#AJfO_74Ks}gY
znDHo-NTTqWERxDJ303E46!W5atM$YJ@M^J_?LFPXaEB5Bi8>d}p}Hat9L!nhoH;`E
z#Ds0>I+do!EAa#&SVO29*5L{wVi6(nBo(mk5NSvE&lUJHH5%q%R#LMHRI__&biQu)
zHqK31^sO+!WT#$)su<eaJD(oTE$v-B@0hOD*Sr%Qho3{4ksB>76+0i(^scF+5N1<C
zwp&gl&_m{>AR0GmmIXIA*9}CpnUz$Es_~8IC6$|bOT;?@X@U8149BJzinK5qsJVa9
z3WH7i=x7(X#%7Kp!$bv}MJ?ZxmK}=&QBue36rJ2NFCAN;0_&;Jr?|%aO?unUdk35C
zHH(&9PE0b<4ik99=augWdGRd0Yk$E{#lV!zzZp&0BmlL#8wbeZgL*>6I^C=`wJH~%
zf=QYjO1g#pSrgveZYX}Ll_v}&V(D{MJ+OIWr0gAOFe9vlBnu0<5t6cU^CTvg<4ohJ
zZ>&5c8v8EQlY%h<Mt7FrJfLX39Cwaf0~L<=ySsJoB3yaD-P2S51Zk#v+QMIW-t20a
z1}#s?j_fmOmGF$@?7I!Q2e!amtF>0YXzr@(Ur|luKNp?m%6)$;jwR>yVerdk$Vl^`
zbA3PUsH}n@;yExQO<J<p=qIMgn%1^k;mnft-O=Siu~kGgYeC|7J)LO^T@r|oY*3y7
zPCOUs#+tmIrLpfJWfGc-zke(W1xr57=K(8SUzlQ^BnhWU*1`QQwf;*rz+)dc6A1WC
z0cJK+J`HX|np95?y$Xc9Z0LcjZn`%@S?yPl?pheo_;gL%BCu!PY;yR!*~16UqoUNv
z^{yF+5mtDxT9<wn6xtr3`pmZjg~HncS_1g7wzchwdUyf*TD3$sTXvMiYRb2yfalhC
zW|+@K!D+}b_{+0Gh^Q-@FC4eRYKz`|W9MvbNvQ$hu7{AU1E7t0)Gm{C$NIC)hKF1u
zIEUmjeOyp^NS^>mlPxXvg+ZNfC${-$t4j@e6eQdWlCiH*M;Ku6=GLy&s?olF;?;t~
zf*Wr3>Qg97u72fLtBGgOX=WF>U}&9=;Yk1y(Xz@g(3I=`q`CQ?td`kqOJ9O5mNc{A
z!sWB-mJt6D8JtavgtK{HMAYt3++S596#F>8HdE#NfP?hWD9qa5semI1k8hoX5SE6h
zv&wLkqHPcA;{{+L_V9tus?8c}sycyx&G?UD_Rj@C4QVHMdo|@kBQ1I;;73`4G<Z{^
zfhjKHQ<hGdkz+O+KfLo~x8ee@BE71x*#NS{r=c-s;X+|(=7@pQ*3TcXaG0?4k3?FK
zeQ&G_dwS^Ac!67KWmuLC=xH_Wf;w%2<s9eKsN%GXG%F?s(stbo>;(i}0<^qKqc8@c
zJgSf;-wz1j2Kt={OiShKE%G#x0(J;-429+OO>k=^lI#Iqr`E~m1`7AAZ?V8Nd|<N!
zE;oCT@!ol~B6UAC6n5LgMwH;Vpz_3)XhnFMChEPib#QzvQ3}1BqN<|h)kjQ3iqg}-
z(FG*5YR<`7CQLWi6E2>ovn=(@!lR2w-vlPqJ4T#<ehX4QRE~bz<3x;5@gIiz?NA!M
zEqCx3u;d;PFRvBeuqSvJZ#pf@geKc#Y^m*FoQ(Mj_m+GE<Ort8{`1oOmnr{8&-_oN
zOVyY*xPAr%k$1156v@5lna6ff1QaSL*RTNMTowY-qi=f#p?oaa+wG_zZi&@jcbOIH
zO*#s>q(~Ovw_p*FRjm41;!|MzsBn*8EJb4G<mcVCG4^T%k=b3E1c(q_a|Aeqb@a0y
zS@a8WAwWkd#kU_)2ltHhF+1VS6OMqu7H>ETQI$Mlo!KVMqbQmD=ExQ7>v!}8|N5!y
z|9T((dz+r%KZp}zR*+C*P_R2@>J<bAFron6`{$SZ%UH(5{2v?3y40j&|BDv)V=Pky
z%ZJ>F6^7Gk)G90idZdMgs-oSX>a71gh7CgTw!w{3R&x4Cw)#X8fi->pZpuM4#zr_c
z<7)iWskhzbTg#G%g^^68VS`kSWI!WPk$A}Unxq68IWwX6Mm$y$h156yUOz6chmM&@
z@>y4zsr>3@GcK&kefkU^K#o>vIDT3vfPqH_X}u&3T1uqBm-i+IYsSQzjiyo(I`UDs
znJT-gE$`H#)$!4GQ}O1-gC`g5%c-)>f=ZSL(wf)EibumNDoDk}+T*^aRg8kc>zLlD
zWvp_W)NM7buq!1KYh#<n43*Tpz-%53o$j-SYV9HPcah(2qA*xqTM=CqS!wYC%B=i3
zM0ys+2pvbkx8j*|CkHxK&rvh2(Q~Y#NiXik9-S`=&g#xvN5*~6U8GNHOj{iULvtJh
zd|E@E6pIj<@YLJB9PaqD`{8UPP5v+Wa0m)iZ3r1={VEqiR+XMSF&^h1d3PP~<C&%M
z1)c$-a~8;55qDX9cHC(rN$AOZ>Z(CKMI!UrPIQAmVR(_bT-%af5v>FM1keF~C4cbD
z*%gEFm~FV$W?GWE&g-g#l;BF_YH-X_09)$0A0MT1moo@wqco6es3B!aEB4aP_Xfws
zfJz)7M&fyqp&C^rxIoeNEcjs;BZ_1Z7$RIOeaI6;8vL};S`v28sKY)mjiNwA=_#2X
z!75FO&1y%}>rO)nDo4gx1kzpHi4vT_IU+HOVOW~eWQ<BZ41LDoD|HscHEayS8T2_i
zB!arfL*N;)5VW7l@+E-%rzV{Lpb1;_saxTku$B-(=ty!RdX1tH&h6I0OR`G8s|fiK
zK7Gez_Ws7gZk7kcf;E;2RLX0KuF=x#a!zC)LT>)v#X3BxSB=nPVci;vqrB+&c5@e7
zlfWx0dh6ty9yiO^z(81xssP89N^F^oW>GstU_J@!JHj-4Rj>NMUG})nRNB6>ioEIO
zxBA>)y_$J@YcD<fI+R|!FpOam60mefT*QCBzqL5PfrotE;Mau;bTB^2g$C<xd|D@j
zvb6%zyhr#=F^i9;eL$;ubwA?WDsf(~8B7X%hb5)IGr-0-wr4Ag=XQfo-TX!d-+$;r
zb>|%g4D${ZT?pNPs$8YRzHCI4bV;%Az&TaPtiB`XU`m^yAiL39pDE%YHhC06As7<*
zU3)uoW1z+7+OlJr`pn^^^yuD1b!JXxUk_O2vESk^VlfMa#TI;y21^mSHYhh>>h<>P
zW?F5Q86?SD`eFE<PuJ6ifdvKz8reFMxMcF+1dSMn^Dvcz3p2Jri!&%o{RP3!X##FU
zFyfJ;nI5nF>0uq46K}$j`%g0J7ra4F?)C<13No6yx!9|R4%?w4I|q7630K4iMzz}h
z^k3S|Q`fz2O>spvn>R1%#ao{S7clMuj>jm29JQ+2)3Xn!?PFo3m4$Oy(zo!>{HR-_
z$|(DHR1+Qmn_Ji4N0*QJM_GaT`y0Ka--@z&fX`FdcP5g*aj548OY@81P@^N8-wJPi
zV{MGI_g}ccwz}j6UAciz0{OO<dCD&>w=K>rx2313G!PMjJn8jv1F##Sy0&H+U8H&{
za$>LUEesiB#o8g^ePjDufB*eEC_mBR3;3a<gY>U^lH*@gP4=I(tN-`&f6sPX`u`2M
zieN+`@QQ^B4{_ERPLV{gY*%>_kpk2cx=r;LQg>)tD4&<O4idapm(|FPg3$ORa|7wI
z_jvRA$g{hTLx*?!uNCSbl#<BVAQR362?E2Ad37);6}#S}PW3PDub1~~QN_ee#USws
zfXLF1OYHR9bUod=BHHtDq{fJ2K_RJXwz}H%K~TPf{JOebTDp2VU$vcYUsK!zj{#gb
zz1iUc!0#d$Pw~h2_M%Q1xk<&NR3HPG_=@*wvv}jObBi6qs>%4}JbG2QFOaUXk_l8j
zd*oCW8m^Z_QWUV)x3}bcQ6kUcu3Ao*vrK<E6tPIlz+71L{o{HwK@+6%1yd(zCDgak
za{V~qDnOxKvAuSxZrA!UzQ5;<*mH;`T{pZ{ts>heGCT`M;Wvbnoey>VNh<0uGwzON
zlRK~|mqWRp_3~34Vso%p)iZ%XCJ$r-I6!qAW7O|ADyN|vQ@3Ddp5l&0H>S&!Eb`qh
zR4ttUj`sS~XjomADcMnix-12i;FRfMth7#yQ+1SO4K`l{JNCb1iTqI4Vi1MJmOYN&
zMU`!jFv#`g8E+$)t7agIjGTuVC0j+DyXnsI$GCo1u}n@1%THwmKxNFwpWHmxcMKhF
zBi+Q9F2-u~6YoN*BeO93GcrvE&i2uP<Qvx8u5j$Kj?ZL}b>s+P!5M8uDDa5{?E4AZ
z5n%}XT$SdkWRp!}Usj;Tjq3u?TrS7rCQ3pM^BUU5u1)7G3>Z>^&PWzxUm6IQnh_)F
zu3Kpl<f}pgt_le>>Cq7Nxwne?F>Nj-2YMX1;I116+nQW10iJ-b=mS71fd*LDx9ePI
zRBZugXq)_TjH`5!#y;k|x^EHw<!Yj)3EKZG+ce>8O}GIiHlSusbJ=C3+UTx)<xMx+
zHB@if=l9r0l9H$`fGX(}1%U`Qn7^gq6k0rdedVLqIHNs^#^(Y($(^>c2M)<&&L;Tn
z^Z7>PW-EE^=d_aKixTs_an4?IJ7remqJ#$koi0m#U~FgeYfd~F+%T9iL269=G#L8F
z{Q&V8l}OBkSP+Lg><=?JGPY__vTi*Nr}wM@EVqPecQr|ESsZiHhe0G4E|nG8eQp&c
zfhVN2zNh86dQJTR)({h`8h$j+{lORiTTS{o=&@=e_O$!hC{J9OKG;M?dQP$tSo_DK
zE4eMI@;6bHq!CYnCJDeuOv<h_^6<8YSF9o8yW9>sdrC!pR}D61Bgy?&`U$hKb!c(3
z*76IB{{^r@G=YO;8p44RdUX692PR>#yoq3%Nfc2I(e6+tnz^QcGS9%_<lMVRwHWt!
z`zRJUXlx@@-bbdsU4K^G1qU7Ifk+3EHAst2daRLSVC*tUd0CAdj<;PChIX_p@v(J7
z$C%B}iu+akBU=L(X7FLuj99j^42W~|db0c$@;Hjf>)pC5B$qyd!#atf`6Psz_s=2L
zqxcEN5)3P7&Gs*Vl=YybgXjaGv-c;4cQ-DCp#Zg*dGbNatWWP40n@-+ol3@%<X~_O
zLLsOX3ARchKZ%OWJ~a)=48kG}J;105!wCIS9TX%f%^UcBQ+fE~EKM%^Q^3!{mAP&1
zPM!f-A@CPL4F{9InoKV$WO8!XJ%U7&Boq@fz^Ifj+)QvL2Pj(5+s)f$)kag!LB4~-
zL7JjmoWw0iTELWJ{niN#Y5b#^3tkjdw+OK~Z<OC?RuZN%i6j)i*F|0X+7OSSf+p(*
zdyFi8j4XeagIlJTwGNnY+?(tO0I;MaNIw$;C__uuD~e>NS{ow6*AVfrMis%oENa&2
z=U>WamVC}>yTAZdk8G<9Z<@M>(X0L(8rp~?{K<fuF$HNMg=C6C2f^MoCn7A~CvC)t
z6EW{@U?uR;!-C`4ZxiQ%d62VV^H8}CV-(U{wr=c(%I(;l74ueHbE(gAqw_I_IxeH*
zl8Ws+1v?REjLhg5&cyVEoJyh5)Z=X|W?C?vLRqkZ@}6G7HF7nC2dYKJUVw7d(XWNP
z!xsIHqOP3|2+NQ1q!T+e0;F(C3?Y8}set>m(Zsp%z(r9}dmYVzngY3KNHIGt0H2f1
zY&Xi6I2p?%P+-K@wd!V>l6UNe+=#%EefQV^0uz1Uey)hFl;Z*(dIn-Wy$(_=%gLf^
zwmryuWLCm%4<keRaf!A)D0ze<g%@o%`KHEoORtXerjHY^e)4sLj&U<%XFELI#(u8l
z`7;7c&&Z`Y=T_O2QNi_$4sgOC$3JE>)2sB(sDp>s$V`VOd#%yb62jdZ^W9uEii=(I
z;4%{wH*=MDm>xeG+w##{zKY@mVGe5Pa2owpF~5bDDL4xE?#Ov?emJyK;Q0NoL~hm}
zFU&voU>W}7MDCxZlb=ol$#<tmD0Z98@Rb%Rl#c}6tS<Dj*v^Z71<ik%+B12)ekP%1
zZQsuK9hU#W!sC)fQ0rkATzS!)BP)Gp(k`q(HK?E>Qtdv{b{nCRIJ&WrDTAJvjJS`q
zUt8q<{Uj7JJ}hq{jhe@u_0LxK=ke5(v)BF8nTHQ1ax}Tx!p7;v;nZVKan07?)Z^vD
z;be4==evi;K?uL0|CHJF(TcAtbHnM9)!Wt9<A#=-IGu0C_dzCm)}!bgh9qbJm!OKF
zIbCfkE(ka#s)DplsZlXxf0`(xY~1|3_v2jxY2pYA{d@;eM^ssFA{9Nlj|3vBje$_w
z!4M366b7i62w{1+5sRZ!(kvb>O%lu9mPv<01AWAYcb&%J<GW59R2>-x8fpFOc&SNh
zk*G7KbQHNoowr16DKN2hl0UIw)SzkrNa&v${|VJ!rX7vV9r=lipIw$94zaPPbheT{
zW9Q)PnmmQNxK8yN?cbI?>c*-j#(qPa2`c76$h_JsSBuvP)CF)AFD4?4L?B^^s`+aa
z!<WT=qCzxzB1tY}HtM9gWP>P#Awk43swnVokehK&tRN(iwrNFzra+K(&vLINDAAZw
z#!Q=!J5edC#(((So05a^x7<aqDlWF8@8&HK4Q>La-NFTS^;5KdV!|!k)~YVawPmjs
zxSR_bAC#$W^}b)Oy)~EJ|Cl9vc2l!O=1>bz61&}*Hr(rZx@a`59&T{$B-!+9u66F{
zAnAAMnVIl1G$AS|6-w@tI2i(HUw~_L?Id5d`K+5RLC2O?`l~h9{^qRghlfPG{H4g9
zt)nBtuRbqMuyt{#$2{}>zIK_6P5WGQs02#@U!OZ%4t(-_D>Vbmr&wSPBtZUOggbg5
zg&)A(rvyhF_@q?VJW@as3StR}@lCNcA_jNvAdIx_ME~Ww42pZYKdn9u+4+?BuT~3e
zhP{UJX>=RYlqweuRCuc;nn=?5#f=J}AzXXwgV22DhUi*9{`a*Gt|mMpLMT;iW+?(B
ztCrwr=2IRiEZ*u$IsaV7B+mw??qkZtJ>4m<1~uj<PgW;BoIYdpMjk73`E?OOF&>C(
zLB@^`n``uy&5kltKB_fN^=mVBOAzlpls^S9uQMUnxX>-}SJrb_{hgpVB*+XQ!NMks
z)=VU=OCN+aa3JUuePTYRAUo?bY_YFhsueEd^TSUuHJ;-sAu#!5I)A}(U6}N2BPVxR
ztwoaIrf;h0^C7}cj?P0HajG*?-W&Y@*unr1$WVj%!5~*BJ}2mE&MPm(0}wsaW->?P
zE|-(B+Y1A9e)SJj2921qILS6@l-wK@;9o~_fy-;JM#C#|_}Lww0WnsEa(=vT9np17
zM0WaErj(W&^g`-`F~$A@;jjmxaLTGq)xkNVVBOD`rNi&&T%F*DGT1ueK9f)v0+&It
znF!Qqkx6AnB|>=aWf_!K4pCxgdADW8B_sTW0$7Tb@2r21WDjaz+p-#Il&*$nygfJV
zNs?3Nh`9PHfd;zEEo>|(s?~{$n2!8f6zwcC;YISZ?Ij$@f&60?8R-jX-x^l7!-kyq
z&H-Kp?oe>>%MWVlWC#;5o<}tS&ISM**L>CtcA?9)VJ6KD!1~fKVg#;B8OMU2H7$3i
zc#(lB&c%7JhQ%XeTvlV>0sYNWx+I~N`Wz{0U#ZSh+!~Y+@>e(va?qtnEDZkwUS*wg
zRPiJX(MgOZ4T)SmNZVcIS*jY$*;Xbw+iL@1B8;KXJz=%97olYdUge|v<u#s_yh!jB
zFEAePH&lc%WBJ~DU07e3Q-w0#9Xbcp=yJi`ZW$0`c9!V?g>~3?E*~odbh{!AW}E8E
z2Vhdd2b(L_ejh_j-f$AQCyx4f5)|%I=q{QY_#3?=gtQ!f|AQG$znDLugm44~ZQ1VR
zqM|ph68G%=^fSLt$o7;&V~aMwNzAUqG(aYve{@rcY}h@AU@<NNbAkNjNxuS1D4ep_
zhe<9(=(x`|a0%08hDh`u%d~V1BK<=Qr>JC}fy=T-k5|wx-LDv3d0}1pP+j?8mSf$8
ztQ*}vVO>B<nRQjtBl6DuMpn^ORsNtq0WJooG$rbx46r8X%FQgKIbi1)<H_*iOl<~F
zT7A4I_?ZDO1bnfR67l&1=14a7Navcc^C<?-6k_tGXwK_RGU{g&57rPCt1wzB0<fcd
z1QyfA)l|}r(Lcy-1D6v8;Ao?1j}dqr-6$um_nrwO@GvyExif8%hMCy=V}Xos3qo(3
z?xP;!LO`sbS$^8WTa78?g~*0Sdl<wBNZZC0@TrV)x${a3g4h}^q3A?%*O}X$S!*ao
ziF%_I$_hj8Bj@oE^sWvR{_cshhx8QlYu#XTs8IVIq%J!2>Y*kVT8?ZP%N@3v4`zvD
zu2iiJhjaEc{X0Ht>hIl-tUO*8H5!;gC1OM%4>!Gwmh+;u@|7cBZM)*MwQD0Got#CW
z%Z0`~hJ<!l$UU2m`mHHJv&+uk&CHrE3Pms&@5L+(?h8FgZp%wvqQ=!rLN;J{RjZCZ
zPu&2IV6dXB`dSZvXxV)XoYph?zgWAZnQdtF4>@atJAiN3RDo1zFFe7#hSdeEwY<7b
zsHafDjwGP2x$nV$1ML4n0SmmMw(ywyET*u?%%AcmNPj`E_?UApkhorsYgFmd8Zg3x
z8$w+Tg)N$#5HIm8kgveHo>uscvG32;{lw+$tM})@3=r*{E6cza49kD4L<e8d<*4hH
zk-iYg{<ahSDz=Rf2fOVHc<TmWWZ`w_8vv-Zx9neiJm)`rx@;W(-Nye1qUvGVj_o=F
z!pP1m^#?}t(BrGwhkCP%5&V`&Lugq=oPUxS6eN^Cs)g?dP8NfbVtbQQz~xwsBj1kq
zbwT9MolSJ&U4h&zH_a)G^;?n@$`&C}!Y&*VodPG18a^P`8`p=26Fi@NQh<>LM4g<v
zq6J@$Kh&(NSswIOV85B3<d$nIazQtf^18zFw%QZ(sr?VV!Slk9$!U{lImZ>KVD0jH
zHvt3!vbv=J5D<)Zft_J2v~9sWv702DOQKv-$r_KO;WNk727=pKFU&p(U5bspp=pDI
zSl7jHl?wgvEs@lgliTlkgTv0|jH%FpQ5$C_0AYVvmHK2D=Qsie;z~@DEV?(g-qsf}
z5XP~G<s`Z|`D9(pcLBzRt6xyvHtz`t*VC{yn+#!4M(lyy?S%0lG$sn?{0^^jCH7n{
zn(z%yztS4*<V{&3pfig+sq(#;22}Z)XDhtiPM2D2np8Jh+!bX1+%MDzK_io!#;M?$
z+7}_+QWP6zviyWkQi<>49L7s#QG&J`j01p2Ew#a%DRJGt9T?Bc7bj*`y(vcnigY!h
zaM3x^4QaLSg<_Kxj~+;Eb+FrLiwqD=Y<a2X=$&^{qe?^)!Ivx9yN-Lz#N1gCra8ut
zqJqoW5s&W>r}0PY8%UQdgV>JDtwR9GM&J$o=%3<LjOZQqS^R^<2WjUAyYZC0`a`@{
zc-V^F6tTVq1^wMzR9n#uCF<&#pH3=?$ENPimmppM$2{-bJA*A^?QpXp*3e~PmOhVH
zieDlLWxv!ZoOems^FPG7&xzHUf+=}Q8tzh$60Vb_#~JJN5`d2#n8ypQYsKBoyy3={
z)(2JsKWkwE^aTTf&DszTBfq}h4)CO7@18tBjH!Yl_rT$fOyj;+Xk9+><$1G%t8%XB
z+id=E{@ECq|7R7>`M(%F|H1iRO51VF<oG%N_i8r~UHNC=39}EB7G)#sHcSV0^egAa
z#+e+G1~-6mnd$Ebwg!2ORYuR6C9*k<Vabi}hYiqd&M9!tV&rDGW0CApSR)7ptq%LY
z#VN$WB4!1+Zu{MjnM6TuKL`r1G+2H@Zj6{P-xDL}IG(&rOSvPa3r<gsKg=Gm$jJ`w
zPn;s2HX>K%K&E<13#aNQ+&-Ypis#I+baZ3rr}yygw?}$<+>paA6<|oIf&>`jW?A^i
z8zXUn%=#hww8Dost~b4_CCyRjsE{-Frn7Eo&8j5`M0*}-DDMpuO-@yR(JZ`KrdRz0
z>YZV!!NnSEA`Bb%G)hjS6m@w%Z{MR9*<`GkZ|>dD!%=Ico{Bf}^NrNORySESch}8i
z9^m9iN*X7EABgnMT{5<eb<#;WWs^2}3ouGWF(P0&rxBAVhH&~Lia1;iQm>MAOY-DM
z9|>ifXFZl39JW8GpP(^wOE)xj4~!*Ol@IUR*he#OMVokJ{xzSrwc@*GyY4EV^2+X=
z*P(h6WRe6y9iUkFK*JZIwjqY48fdckqnHkm{93hu5^4rcnsD>-Kwc_=jx)OrhTcfx
zA6p1B4G9SjsWqc4Z~nYooc_neI(xBdp3daVp}$|pU5Tjmgl+#l9wo!B8Gmd_SSG>`
zgp7kN(FeTHR79CKsq3)m@g0r@bCYdPS=^C(TwGL{G5_j)3+P^_X2B~wZYr{~-1?p-
zle$B_kkvR`%~`3se2ZnEdAOD^DAfGu?q@j)U@9tT-;90Ln&sO5m!Ba59=;i1&6GeS
zdCWb*c+G&uN3PsN?2^TrYtP5C>$)C-Ts5baRW?UfAcgO%?nKPy?;Zkc@QrbR94q_{
z(!$n;OQ`(41K-GtfkhudG9egz$1LBczS?6A1|iC8Nd-9OIN(%>K8mr#`Q>PJe|zig
z=x8K!B9lhKRii83L4*WaWp$0bozTW^L1@gX=9GaKegfRV=hC<UhAuj>=BcD$EZrEG
z8K2WW+iZk0q(TJ1n@qE+A_I9GcGdb`Pm364XN&u@b_z8&;SOjmEvGtn_NEJ#Yr5|>
z0bGDD$%^SNoSQ-?fyOzEDH`e{Uw#ENZhby=)T1y}L?L999bMEwA3Zeu>gEc4c@qLi
z6bS`OiGDS<yO|$&b4{TF|JL|w%(b{0m`Tv{4l92blMsSrhB!@>yl>!x5lYSnIQmnZ
z&OiU?{~VOE{WlEu|1MY`s!RXS%l~V;<b;A4O&D$`6c4KrQWHfi@?r`W%3ZBl^<qGG
zj1ua<)ar=EK+S5`OfbPYU78veoN2H7sldWDZ?Ohfbf+h;hE})F<@zWKAejg&`&9d6
zNTcvZF0IP?OdNh4A9XK&nlT$FG0?xZE;AddP3<%S^B<Y*8dnv~SUyefZ%=Jit{cO&
zKA*pCCXp%x+Gr{2#B!N3>TI~nAj&7vus3b;fBmH{c}hs<DMfoF{%U-^YVNFBaenXc
zq`dDqz9gMU_38NWWWO@fm}leTyS^`fuUdDp%A`dp_QM)Y<Dppr4nVAsv7KB60qdhE
zri?_MBx|u!H7BH{L$uMPfqv_|huXuyunx{+a;bHFIvC3KK$~Sjz?e4fbDh=^W@&0K
zioSfQg!Lkc9>Sz-@d0#hfRrMhTH?Q622gX%E!VU;*U7r~VluW0)*f>i#1c-z05KBs
z`eW891t8OM&^2fPX4%03hie23Cmq?IUD_z8{+`)f3TlgYYg{u=5lB=2mFsct=dAVF
z8oMfaFT2L_#sBxTeud}0=7be8D(Ws{oBNDS6^cPH+>im-YT8n)^j8;*8)mgZ{!hx|
z8R;q}0>GSflF<VZp>z;tl)DsGLHLo2DKxSNRREMby?tVBFtQZ5GD9C!vD0)yIfCeJ
zaq6d?)LJsUd^#C91ks}Y+LHvIB&UBnfOKMTpd^rxn4CFwGSGK#>X-_6Fv|-4$fvQ5
zbiYQH^eFeFeXkR3FGBC(Rtq7!5=IA2py1Su$~=i8q`5s5_iSKSPjQmOhkT9>ZnYAQ
zv&$%OgczIwm}T0_07KO9kshc4FD$VQ#QI#6)CoF3q18&vQt)tETKFgCQL$iSouRa~
zk&{y+mY<nGIHSjEXbH}uNXQ#w+5^japdRZvbqc`8L|PiC2^<)gy3F5Uh#5Z<!!SJV
zgm;x2M0Ke|(I6xY<&#YaQrx7`o$$Rxnz^uo%R-bRKZ!LK)nFk8`H%Gb5TPI>&XV6K
zx!MEBtK}qlG)eFHF3;U?ygo|c_Dwj%!>CC9p3%Qc=2XYfku}CGyx(I<m?Dwau(rt^
zHo^StxLXfzprBN>8<Vlzp6OtT=7zI}75WSoVz|+Z{W0cuRm+1F`Z;inHz%>!o*pgo
zMQ)Wi?i&VhI5wmI03-0j#{EPipRN%wy8Q7Ch~<ICe@Wd-B#Psd%WbX4t!N&J=bI9F
zf|MBqlAuJy0K##Dz(Br?uT|+*394&}iAodk!<XH2TI-#&%*xa?=rVaQC#apmvhG$P
zIoi}8%bPAV=fIDo3>Hq9!fyWv_{{oqL!CiiJ$+Fv0a^7)LoICN(?YN2++^*IzqR#x
z{AT;Sxhfs`DZK9m?gtKli33=~#oeQ$&7<=mZZJ%FlsMW83Mel<F*P&=o>6sE!2Rs_
zyMBjjE?PLu#c`U;)Au&9ry$)wavTN>w3I;eY^>Yd9{%Zr;1~=}6bAkKeQ!6Il7Cuu
z1J*CXoRGYG$X_%BQQ<DD_sQLhrCKhvxuKO0m)qXZF=O!F+`muJ@vPsB?iVuyOEX3F
z3?Oi9eR&5`S}tN>-W)fvAp@VSD3^{D$;bhN6C@qML)qzLF+HNJR`b%ZDtZt70S5lL
z$5(s>u67B<)c3=k%_PR*tsST7rgc_UykQ^Y;oPL1_u*qRys8!+WhrPjZxWL^d~qSH
z_|SFpUbZuH(rEL=_U4Z}y3Fa??9f5gI&*Kt*7-e-Y1P@Alg#{jgZ}(@Wa>`?fA3wW
zZ`J{9BFY@noOOtqbxYL^dO!kVICGvW=jMX9_n7lnUFFK*CD*1F`=bly7pTw-Vc1|?
zg!=V+^X71L0q0nRz?JtBIF~7iK2x3WB0^)5=^@?xLlW9!vo%i=WADK`o~4<fBbM{}
z8s3$geWTD0Y&N$}K&$8WkZ@@J8E1hQmTM0ZjFRwxY942W*E#;wy@Lj;ENb`i-eZ%m
zKA8(y$M^Mltm;d<7lkm*V=g^K#*Tu~7!Zy-*93P-B2K;zby&g?N+8DMRhiL|rPWx1
z^U6|V505EZF)rFl4*>hb-Sod+n;#PK{}6W>*;xK#{Lzy7mhI0PiSLYFo;YThhL5lQ
zs&GJ)V8}A9(zVGJ!)WuqWrK-)0(0RieoujfR1z_}PAVGiM=$*U!`M0a*21;hzP4=}
zyV$jD+qS!F+qP}nwr#uJwXNGP_uibGoHse|AFz_Ol9f5;{5|6t12<-*1K-DmNoE;R
zrh59<iRp|Ef)!FgWFceP4w)30zhq#W*m1?7wggJP86zrENy<vYN=-GLj!d!A?nB6V
z%YDZAM9%R_jp>Qa+0l8$dE0q+8b-R#rdFprkB_&*=lzh-?V^N?ch>jI1*qw5v=#vk
z3C1A{(;!rqzAZ6^m|8x+79%we?44qHEX#aV`JL5)qoM$@`8XmO|6@Ld%LdWLM>G3H
zON)$WsTSm5DAZk#o?k?L8M$JZy+-@7?aIUb(?fF;*-P26*S1k@dL#}yaY2<Z8zCgJ
z7*;$4O)S{ROiS(QhnI~u#ln=$B}bnX>$)m({=ycKCQ3$rn@hq*Wwo}g%lY~|^P_&-
z<oVBh6q=XYL0S>DYO1;C*VUhIr&k{8lq%Hng%i{@VA8Bt#=T%d{<7a)+|3c+<LYfY
z=tt{QG%m(lM#cvsYZl5(wFTJBZcSBQTmY`tT&~zwkr>8fiDp8uPV#N3^hw*sLu8s(
zA8+k9Py2~PRKrBr)I>P}X-5_)4DrN--82Re{f-bA)9mp|LcaO}LTq&fNBSPCM$;qC
zom3VO&-=3|9rIJ0c;m*r^?uD&_#(Y^1O2iyTFAbJ?{I7*t$HXnNcLz?uKjqjtpf=W
zKG2yX`?5T=UoR1XJ`Pl02;@&g2Qfnk08+mP!!YQ<;3DTZ+aPt(h5`$@BX$8f*55h!
zLb6Wh5G`rI=Pb>Guw|)N=b+&JviXB3f*@WH&eg$wjsGQhU~rwC2^fL&^bCr;i^azP
z#Tw)~2A{&=vMT91n`POP9K@m5xN3fv?w*<;tGN19k|Lw$Awo_K8zoua1Eiz3A~5$r
zfQd)u@yl=iez41v1x1hP$Y7}C^UKHMEAo$YrHcD&pZfvMHyS8G!5#@dpsaJOyu5mE
zKH6U$1_gt$WxV^MB9Qh_lkaw2Y+8MqZ;+Bgy>e=h?<M7kgM3-_3HK(#U-jpjbYiM&
z030DR9Of8H{PIFutxBsTghN+Y3;4HmQqtpP^?ExzhqkfxEg-ik`e^m1dDmbLcqwo0
zg<WIO!jMVJM?vsevVK<gs<L;JX2<xsvGN5sX#dC66UJns_+VkpMt$?XI{ro1EwhXz
znBzK~n?osAV8*+b&U<>>H1NeA7ay%lXkK4Ig{p*V9|x!8PD|lyK1Wql{696{Q&Sff
zV(X8&fneve3AZIY<<t6vgrgAr1sVL929d*bp?Tbj{(u81!kO^~GVIq#8hcr7Gv!tp
zf_ZA5qy9N~D$Bu-bA%Ga6p31n+XAPj1@&bv&7uONUaFNqz{Xbt+B!?3X)%LsjV<Xg
z6ZT^D)kNgl!aP!iMx&X)T#zJ8(xHhEu6fMsy4`GXyyM<+IY3<f;vA%Uv9!Y#l@tMy
zkB(a?k~Q{e)_3}oV1U<NaBEYkghkzDU@^#C$KfG;8Fx?GOr$sr7x+%L@b911(;k1H
z_Hl^C&GaVJ5#}HPK`b;~w>8uZmb$39<9-VhGsi~Oa?LZ+PY?{bBCx@4)cS#o)evDb
zgJj;@UbJG?2@+Sj|K@PVyoFgL*QzJb%%&F$9VN&$4xLgXLJa+b2*O|`ML-UqkL0@5
z0&9*<aDOkaNrDjY=9<VfxB(LP7>d~JduCN@Bwbr&zIh24KyL(pMJdWz#z#Y^`E}D#
z?y5Srn+!W%FQ3T2`(juf$b>26jju?QTf!U609*KGA9;(xxAxc1m)o7S*{N^QEZv*;
zDFm8pN-88aI&h?sQB8P=^|Y8WZk#;SMzFMH-6wEelP`%`DngHQI7=ZYkf<v2XNc1g
zFOT2r)xUnHSXl84%y|@&pieYFA)g$~2sbMQyFdQfB5aLS`-dm^COp)Jj1A>4(kc=(
z)}lPeu^K4+qEY#7`srv}^WXVeMEzap-!NHNmBq2MlwdREHu%dH^n7qAQlR@RTjX3w
z?2w0`{>d>#ybGtj`a+?vEW|F3eRv!bWnnXrc!XYZ2Iqj?%Pv6k6XGY56p)TckY4Qp
zV}F^IARLzQfnq3u?fW@mzvzI4v*~3r=lC;C45xiv0QTTN-7BHDk3^Z=P((7?N@BCh
zAWv;7!8~|)-@$2*cC!9SyBPlgi7>GIxG(>`wZ~yY1m-_Jd@<>%sc!rby!}=2tW&nC
zs?2$*;AFE5TH!LHBc?1FACEe^u*5Pkn|m`EEq?^Gmm6ANR2duO%+4Rf6+XAyw7NlT
zoD$hVM&Ou`T^K$3m}`2ccvxOR%R!24HnJ6m#Z52@Ahyiu8oaCzR*rTwBbA*53YM|J
z{=?zYdBj44A{{q(kGr!Q3y)?`6y1!hPRxw1wwC(4D4T2Jr{~>f>~uF~9}cBOeB^&B
zG~U12dLEt8CtjVrwN;GY`=CwS7G9R7%Cj4fI-lkq45rx_(@8KB#LUj5RhCKOsJJQ5
zUn?i6R|~9wQM3R7Ld^v(W|Bk{#zKea)OJhBI5O;nP$MLq7H72uv6Q5gUY$lmeUqZG
z(1G*JgqUKWM&>{^?hG%jcrar5fHGvbj1n=Y=C8q@3{e;wwrBm(V8(x`gT%*+6rpoI
zDXm|O$SJg=ekJ3oDRf6)cL1jF2h=v9)(>F&3&Z>6d5Q}`Zkg$KTC(090py4C5~$}?
z*?A7?c#e8|9`e7$3f|=DrtrD5`myLd!>ZE~@`BDk=n}v#!5a?l35rnWM?O8V^)6&%
zLouv#_E%piZ@HK7i+B8i7R8GvY%EGg11?h*qaYG|D|>Tw#PO$Z{&KdmW*l*5c;BrT
zt^-tJ%-xfbkn_I@X3NPy-k>1`enk+o-CxawX!0UZ6X<mrD`4Po?9NT1hO%xbVE0r*
zeeir*9pIQmfU+3$M22yLsA6phqQT^139sX+9|<U!e1aGt5uW}1`)?LT|1-07n(Rcr
z&1WTJ&+W|FoWm3vNi6cTb{qmxzi5%8O?Fg_H*ftBCOOF*y*j`^8!h!PNy6G(Mhjb7
ze@hDFuBqeQw)r$o=KNQ=M0K?H6R?qwqBUg-y&!^oX9Kw__QE_ls$itzuD{90?`a*+
z?~9`EoCfqIvS}SneZZJhVVNQGF7H(JaSVmC+(oEhdX>-3r;nKCR2DJE`p2dT->VAe
z_^qm?Pc4T0XA}PmT>HfHeym&OB^M@e0KoyK*|8;Z8|Kf--%<z(;Ko@4a)png^?24^
z!IkHq>MJ?5FTA5CUmCD4m==;0Ex{i(vJ=JH$t%>V&f17J>z%YF&ED@>kiaq+8~L0o
zl0am&k(3i8!>2nsJ37QHZy*-y?Z~<zNH5jUt}~rTy~azle&{3H?mDzrcesLJj<i^9
zKxM|wg1sS_scbdm6NqzV@Dv9^Mb=j&EyAuYsOiX%QDftrQ)Av)XimsFMsM9#*TI`Q
zE98EAQdVuVF>l&U8k06wMu5#d9|!wvFay4o%yEdVQSp4UF`)38PBuUFkNue*o2J=B
zVV8#>S$%<nytoWW+b70Jx~jsXp6aswNb9eJnf{LYco)HapXF2M-mJE8r1SNo6@*BD
zx(t|Pa~)8#I8u;cT4FUB$xa)twePdGhjCWRamV(Z0fftHK)rv*B4s~b;ChYx-GbMI
zAGUh9cQa7m^hc{Nc8zkHe3&Z63eD1qJZZC11&*9*?1Hl<oV_%+(I4GU>Zj7hbx;*Y
z=<3<Y6CGU29J*z$?wqa;2GN`!=-@*5|L`m4U)XOFIZ_yF3qJ7D8#E;#$<<D!6oW9>
zf#>glk%pSgs^N)3wq&~ZphwihDSm-Wb1V7VL5`Lz)Pv%GXll}!PU-xb8~9l~sgY{~
z0)56H2Dyw4@7Kfo)*yglSbUdfwGJns6&7IQqRpX2v3H*FJn(n7co)^Cxi9}f)^qeP
zuY@+Ewc*Z?jTon+ow}IZX*?Q~{1vfbJZHpxb5M;!fi+_U+;pECElQvxA=m-=nBp)h
z`@4l2&|H(IeEx25ECAup5d!Gk+x}7a=41-@luoG%o?sKcV5!Q*dI55^`lZ1IyCuPv
zU2;~<!uCxc$%UFGQ;`_){QTup2D%9Ao)qK20ut&d?WOT#na#ePElac+%`UvbZn*>@
zxXmxK^xR*|TV50Tt-w$i3|RxO)-P+clVWHvg7dHCS(B#10eT|4AknUCFYtTxJ1Gl9
zVl>%czd9YR{n61_Rbas+Is7#5Lkp`?QW}JhhLgFBhOe9sa?XAII|6M#l${3;w|91}
z=#((V7M^~=SH$kC>(tzpzwHKHrx<Ub`N0a>sk>oXw=G;GPjmn3lwT{eR#}yNQrU~M
z(^F=%iU#`f0i69Nu2fal1PURCgJELJ=q=jA&zuvT`6prfr$BxNmj4g?|91qq^^-}v
zA@UytSj$=t$@=b%<8FtwCYLuFRHc~FpKGLEOeh~L%t<Nt(BhIjQPP-bEGmAU-I^bA
z|8ktMw#E#13<xI^vmFz&orPcp^q)5whAy}i*ni$+5W}i{*M#z+c{+p4a69lKrug6^
z17@zy92n0;26nFN_>VI&AptvY_?r=XdW*a~F6S|(_-}$OAAz9FIogrnydA73p9}Y2
zLqoe(AU?!hyu&@6oO_k0VaH;OP|+NsFbx8ttv6oxuUCVy!z+l2VB-A#p+BPMDB<0u
zE4?pXFILa%@l%zkHGJNrsm)usK%-uSCrcC^2__v#91Jau4)5Yp@i-}t%vXVSf+4Q(
z%1tHg$VYFGjzBF4(JztAxHv{HjBNokuV!Or-ew$Y(23%DelJ8&E<|}dbRH_QTKf6P
zRx?FGf)XbK9>=AJ(DX;a+#0HPW)56U?&yL%GE5dxSW0=ChMQ1wkU3y;xwyPF!vJT$
zS}U4!zxYgZsB14`*rSccENG^1z*y!GPzy~u<yn}$CXTJ^ZZ1;5f|UBQ7<U?a&W<O4
z7O0xL8PYCUHFHQWl>(V)<G0y8AYE5t2@+<WGkQ)iBTLcf=0PA|q!N!WvxpsAciK@L
zzhE4gJt%-Y7zv}ynOA&uvUn#(^|81!$A=o03he-ryqUN%D#XPler^(PP%*Ps=;oW`
zj#yKL2LSTJ)fJ%nQO1@bFqy_eRRhA12%!y8)etfj2&!h#oWC-`m^835(8E00$CxPH
zUzgUiDdYK04NOmE&9CFJ`57@4p+YG1Wb<wj$tu}p2HJAx2<jpyIsnG`ajIJb0F>VK
zD<4>GX1Ge|kku7;ko{d1A-Oh8a$GWItL&ylY<xX}nbW7~WfoJ^_sLUfhf4kGHfHN)
zkS|ZH`r1sJ+#G{c;!0WiI6H|fEq(>F#;Ct~Y(L$p_Q22@<pqo2fdXJyh?8cq#p&FB
ziU^2ccBw3^%IB66>U<1xqRP&94A6lTCAEau*5f^P7sh16Ks1yXm|+PQWMm?NsoZI0
z*(Sy81iz2GCyjVa1j<vVD#sJR#{ilTg8*x!@DBQsNeiQV@8d)y!lTqOY8lj)8uK^N
z{e|+;+8_O6O7~<d^b>7h^fUtx&`_op@5AZ*lbz0K`2nQjxFzj8$jv%pE_W3w+8@c{
ziKNXZ^xOc<l|zVS64rPfBb8V-fJ|<u1sYD}`323sEr@Va^{2588f7s<kyWIzL&_w)
zM$3(M9D82U%MU{;Bp$;n;=5f9J@u9AfmKKY8q^zpiSm3HH<T=P?)dqf@zRSY@4!P;
zjw=RS*BhM2`#X^62S@9z9yZ!N2-Z(uC5#DrW5?xmEURZuB{wuy>aGSeQ2=J(wjkeR
z4YGY~j44vYYT$|NzSXi&lI>1(a%)YaOXffl`>EHnL&a=!K<kE^cD8neMLZfM58=9S
z(5I*o*z;WoBOBrQSssv5;1({t<IUwk|29|*RL|^5o4MOrccyP?4S1%q+B{rMQEhP4
z{(Q+aV(sP=)5nQU?Tu*$-;mS#g?E*_!m<f8(-RI%!1B6+n8$~&GJ9?u+L$NI{Q0NE
zEn5$mR-8{WXi{F$vE~3zanCT5W`E2;rItjzWC-bF{~ZBAuD4T8OKCwE4ZV_+(q?e^
zz5dEu1+Qp5(vaRX9yhT=3t)2Fu@6$SZW)uN?xHqTg>npk>6{UE0>k33BqwQ`&(z`J
z8VQo%^P(_I1eRx}bi>kT!wbv-?@uQ^Fb!_OFl#IvoT+bSo3s4zm3NctCUKrIoul8r
z!n{Of_115v7`Hn%>SW?)UxEW+LMvO-yP*fN>9pC!TJKaD)APsbd;ly3ST%~%P1$+T
z6ygYv=+uD&X}@xhj~6B}IV+``Eu0nsfif2KXW~~AuG%8G<?3V^6Qt7atVz$mg~fh#
zryU9jUbM_%erY1z8a4)OEYR*|HdKXYn$;DaBkEgQq^Fy4R#2bt90n>{zB<~Zm>_4~
z#trXB?Ge-T<C33Qr&9U|h~Iu98-->OGeFZ=Z*o;=BA!B*2rnG3QW!Y0yL^wf&l>I&
zf8uMhfVvfHtQ&q)X;x+3v6iaK+R>5yZ11dn=Xc6P_kRsAyzjJZRxz~k^1OjN<(iND
zqe%S&Ib>mD{=bi}e=(3gN)@+V=R@(m{s?DZw@#hx#%YLU;f|%UmoX0nvkN7q8)aik
zdOLW(V1OtrETt7cKYlO?K56mpWI$jN1o8TdE7~>iyDx4?g&`=Pa}YINK%a=@r^bpb
zS0%7zu4Lf1#6>XHMQyQOfjq~n>jOKMR<IYDI3F|IyF%RV%f7oS2is#w$%!g?N`P7I
zws6YH$jaR7&`o*s4zab+Zs;GQ-njjy0`K#(873ALHy`5*1$@*3D?xR2nhTC{`*?Kq
z{dswf$#@GU$p(3jnX7fj=1#fcy5PkP@*V_nE%kO=))swTL?jZoBlI@(oi0}mQ&+hT
z-%;9;NOLv%$-XT3t``oC;e>LEuDb)j6lsZVGmajY%hTdxwrU&@5w8zT>4H@H&@6)R
z)<Gx<g5DM*7-7tp$g6`|It=(+1n)YksGX^a*t1-W4*PgALfCTK$F`q95;fPT{ppF#
z2=)5NjicF!auCa^`iT;6hBlX1lI`YN*HkG3`gbK{6-o*cn4w4veD8QVFX)Ws#id(N
zw;h$)2$zlu5)u=I%Jj$$4BrPbuswTV(z5en%*ou?s5ALdr&>qxvtjcKNF5Ssi$|n(
z>g@3ly=hm=&`SVSJ&CoMUN6^VE$Eo>6cW|WUu`HnO~nm|NKr|M?_{cmw9C@j0?T9+
z$M+;TZXo2s-KAPv!|o@=3x)ndpVi?LR<}48V93VJn4b=-<}M`<aLcSSPXe4F9V%wq
z%IoYQeMMhysou<gIJt_Sqkk>HXzF)_|55~@{?)-&dP==FP05n^{gk_ne;GB2gK)DN
zPq%lzvuGIbf$mP2hjv9x)-le`j70|T^}A*s=9kjgcMxV*fW}w-B4ZeBURF_Z|Fvqt
z&V!@L_mSh*(sS%1uOKyhs5>vf`u*&$fBO6XAoBkS&RKtoJpVO(5O>)4DVhF(4;0L3
z{$A`42-3;+B~I%?+-;cfr1C7BNJkQh36qX*8}9(IL<E8nSyJnnwFsWB6VGT|nOjC|
z9u6!xS=>I-wPP@h@#9uZ6F;feDG&~EQN?S@F%*OHJ_ICJu)0xi;m<lS!RRTrK5eeI
z4nBOKS6!X1yzVXxA6|~PPxasW4-5AznL!|Jd}?B?GYh}_@i+!xp)vVW`(aE;41vT2
z8TpxrHjN5EGx4|3J^@QAS#`Hx##$?rBN%hum$u4fm2BCkM~*r#5_CS-uI{&qbZb7t
zjwWN0@NLH5^V4+R@^k>3j_q`$KaKums3s5<*KSLZfyO!sAoVlCDoB_VkyJtC^3cQv
z&(-OwJg@Y+sek){PM^56xcLw;TyftIkn?`*hS<xW*hP~xit$A#-Ka#A?g3UFy6!$r
z_TnK)^a&ALl=8mi&nR8<(4D-x>bOQ!Ns^~$DGa(w;QcUgSi*hCDyZ2uP6<fZXoav=
zJzCQ$r_RL{*88DHPluLN(j*aBBB90}4%8}@*OLl(9eiWuck!c;p3hhw2e+#cVYsIc
zqZ;5RPzoW7RNu_NcS{~-@aqWiE-hw~Dr2Pg5SlxHmP`kdvKJlui8N)T*Y#FqvD>P!
z5`v~H+Uu+FvteMn`pRV|n<z8YS4WLyf>Pb3A&a_gaHpu+1}op$h1_mf?it;Im_Pzz
z5LkNI9|#*fM;ETD3hd8+OgPH*SIUAeHQADW%vCXEg6fUDy><asR0KJek5CW`(p#H3
zwAQYci5us_s=17e_6V00UPA(hp^~d8LcOHrs*pcEZ5bow5;6-hR$G(IM1@ezdVu-S
zG&LU)HR=d1DU>h(WkV$J{we(?kgt$9jvNzkK!AobEyjfwF^RIUDJ~lfu$P`hydc>W
z(uP!l9uuTw8k@;qk<_&Whf<&g&=&qabF`+nlP$<C&Y-)yX|JCUJYO9M8V9G}ioNtH
zlrV4=(Qstm4+W!|Y@E*-3qf!IkYz@Zi|>kF0M{Rhf<FR6V%ZQP9_tPVG;&p-&$|JR
z-Rjf6Vu}w<AAyWUibcE%;OvE=NUmsu*rzxYBK6cz;Whxy;C0n078l-tGu%x~K<*2m
zxK1YakW3B|aL{bwz3#^rE(XV?I13zN5#7WqmyU!Fgf%D++9w81Rg0+@kb?*&93F;R
zncqVhs2|=3zxafgtdt+vi@W_>3W^f(a`VyNIp#0Li3Qu;klKM=YB;$Xt%;Dc$_8@&
z9@a0;Fgu0()HFH-NT~IZ{85u0)q;ke>Gd~%0c#_IM?Mp$K8Jd?jyK;1O3#lmGM$2k
zn5(};3sCy#ipOj34Lcv$q|rT4j7*{9R>d|9cVY_-Ud2A`6<N6?T@{5}El7e;1hjjf
zy#=A7QT}39T7v<#yF6Z<v|7x{RtH@<s}@!zMs^jA!4QeV+7ujPoGjv~YvUL6)n)7D
zf4yVEDTqatNaS5)*Ahh*fINLwh92~~6EOZnZ6fXmdOkli(mt111NT~4L?0{3SqbS_
zA*?+3Z985J4VQigpWmKb6&XjF0*=#vx9`1R1O1ZlK`<m1q_^-=Y^sW5^)REd&L-0F
zmm$&_lmcY^Lw%5u@E7OAw5K|H(3J=wWD?%OrS6I^GGsl0O$nTVCpmK66j6wus>mbR
z3plDIVec!^y*=CcCwHwwK2fxU3msgAbhmofCsr~8=*;-#_<?Slqb5QaknU{If%j3t
zDxLbf(4AB6VXV3dIPAzBCuJ4U338GVRU_YeJn%CKcnc_Wb~;{K)ZbvE+Z@XeeQB_S
zlN>vA+{5Y0*(TiNQiIm{&&dYNo?^OSa_REM&`5nZ_ab~=u3?yqS}4lyhK9;$E9X)e
zmzhApT(r_+Acr_E&^TZL{PPD$NPQ+(HN3y9UxuAMmj!8e%{^5$&9W#zSqu~W9y@<L
zKu7lVDHhi0dIjM4c)3mqR?wEWCCbF`8zD)+CGp=#PVYc{SJ=7S15tJxz0t@5)Thwh
z4B_Ii*s&IAyq5vziwl{f6zPL%wfC+1weE+B=5N!M^sC&S{Km_PU7zo_vyX5-0(mKq
zX7UN?4X=jw&J5f1atn1-X^F-N^#!Ecs@`VKeX`3SG<1M=X|*Jt=yqytMGD!*f*8Nd
zS-MM0eBa~kXqMt?p*LP%MGK0H9%L)9F2^)T$eJ>YEU9|{UgxHv7|qFxQ4Pm++7KB|
z^XkS0h%Dy=`^C?J@hd5>`|XxGmrQ89_C*o5na{XmXGR1;`A53n+quO5q=n4%|Dc6z
z%>TMTa7axu;b%jrd!}YLPO%Q5lYdbK&YsKJ1RmM6GP9Esy(rJfgw%pae%jsA(#L}b
zXcriTxgNPhfg{wlwqx$|F~nxxC~Nj`Y3crTX@#yxjHd3tWzyouw@@~`y`8-B_IQt#
z;#v)aVt`@VEV%6Wyt{il_;4SD7f;5Xr_D2YeGSOE(Ans`tf9Bqc6aoo?Kl>^eqP?^
z9Jlapc9|dQ&gmhd@H>GCYp{fyM9K>UiJ5H1PFmF5Y*^~g!@ISOp=b0*a3wA=a{cL7
zrcZ~YlVqx=K*aG^6KOTpne85B^`<uUtWA;>=s2Ja=kRHvsMu<rWW7#?ub|vZMl5rD
zAMYv6H&({4Hcg*zyz}|K_xb$T`u)A5hFE+zb@2Y-*v<gRn6`o=9EoI^mrO>37;cs+
zWfDvQ`E0E}_R)18?nq=3VVN8p@UnJf@3DF4^xAicbSmgp`h2vnmeMhPt%h4+%!zbr
zb7VSjW2f~iVjz$0WsOm4=arD84Vg*Nva}TI9(h^Qa@WWlg<mbRF6RAD3;gXv{HI<U
zr%IEL71w(!dvK7pGF!__-}pBte(=ayA1@3^jc>-mSGuwJt!N>uHw-{%98~5X%z<+x
znNwZxg(>z-i}<h};1ZbPaYVT`G65mJB(klG<*z<BoNuR&eZtKHbVWpO2E0A=-@qTv
zj15dMAumI<Fbghtr4VM8wK60LQ>$z1fSd<_tFA`4qx2gC=mBVYt~dZNbPoUn@xNie
zX+7(%_9EK+4l%k-3q{?)HYf$F&22wU^4~zBQskyyNY(Cn0ndN8nxHh!eEkAiW~dF?
zCox2<w(fO0caU1V9LHFRPH2wFFwj(T>i+)gG9=eRt|U!dx*-!~x-~2@M=T&FQ3x9v
zs+I|kc9K2x3KG$A8Z959r^=0^4B{LK)(Wi`(i>gjHn9|UNVqDRipE7u0}c5)MDbAr
z{QikDFTap#o_k&KxW+gjRq!D>_u%``S{G<K&iSe=kNVze6K!La{@%mj%<xDEZm@3M
zll18E?RfIdm6^>PnB<f<><Lc<v+lZYk02ZQ=86?^)4SjZ08`CeM+Q*f#Fqd;J_K-V
zz>?HuVm?D;9;-QZ`n|W(;x@XV-SoOz{Z>R=UM+~2m_o>X?xlmP((#D~UD5(OPM)uP
z41XKD()O&{ieS}lW7s!Oe}B$Ub5{RL8_n?Cph5HN-B&{kDz5vXB_sw54L^L~k1h{b
zeQ#ioXD&cD{ps5(y=)<)<d%dy{__$WYTKbuBdF=;ZJphoHDi&PV)4!3p`_PGsKO7j
zcbs{ysg7^4IS04ex}5lK{&M&7d8R6E5nN>8BYia(%Zy4#>>S0Ylepzqaque_4*Y^B
zV!>SdZHU7D#QjmLi{4=kI7rtO99kbjCye!OlsLGLmBt@tya!S9+#%de!D{?jH>9me
z>hj=2S&=p!nxA*vt^K%--HG5l5h-S}&&qK%X?Knk$6UMz8;vB`G>M`nPXG@tfSMQF
zqIs3w{9AGGF7DV=Hq9cA!Qi2PKeH_D)4!Mz@FN0Q({!9Fg+~@+s*wQgr^)GTt8s}U
zne@6U8#ChU$$WyXXwXQVCYvV$DOC`>ff%_Ld-Bish^N2spE%$gRGjFiEj?o_F%wrA
z!N#lT*QG)^5s;>>F7{Je@p_%{H~$orbRs2D#oHgK7p|mLw0l+}0}%mnQgB#aOiU>@
z%u5=NvpL!>;hOPnI<G&8L^V`P);b7~jzPZ;LNSduZDz%s1EL7l-PFaLC<B=dITY%A
z7`(}Eh!HX$DL4i;l0l_OqOUDL{0C;K52L>ovQ%?{V=H;7kV^ljPeM_3cvzCi?c~~X
z9t*U@@mMN+RxA3iv5_PTh?~y^a;5`uPJf%fhq966y?Zf}fiKs5-{V!g#aZsxc)FB~
zqq>Tg^K5VFDNb*dRr|sxBPqy7&6Fa{=HmC6e+sFp4~yJpDRMc3X<6~*iS=95bCZ)W
zFqD^dNoMd{Uv2If08(wJDR4jr2N>COq@@399x*jZ%UCWGWO8T&l>G>#uUD@99y8I3
zaOI!avs9-Yebq4MFwV5UT}YSFvb;NY#a!)ax}WY#+Jxx^FhJIOROSR-n>t;eX=Lzf
zR_mJj4Tr2PP8p$*7%K_3@vN%6Z!??3MVkO02w>z!Yel^fjamo2u3=rQAmeH*_SRYU
z!Y39csV<L-T6)@j^R1!0;sLGdOrBv>t)ITKJVn!)_EA(VW!9iHd4zxIWO(RtEPfW{
zm{ZN9O|6@teD=yA09`(hzf0V8QI6XkM&4UBvJq^th$1+4ns=<RH-}!~nbIyRbeDE%
zJzra^lDl7_KV|OcIN+HuhW8qxgjQB3MVDHHh?Q~M{mz|Ax}}>S&~m-7G!gJWwvJy~
zwV6XAc-3K}&-4bYj!A}EPT1vg!7g|rr4InZo;bNbHDmr!l>dRxFfg(FKS1-p4-x*m
zvrFwxEQ}-=R&0S~RIX5PpWA7GV57cZDp^ZX23v1+r@KIGERlG`X2l8PgD;H(@A|++
zH@}Z4w{K}};&$z4W8jPrIg1Y|@&r)UUyUR%oDE~f|Cum>gs^}@#>~}?A)W}Qr`z4d
z!^`c~8HHbJKy}6CVB<W6H>PH1E7BYH`(;0BmxmJ(8H?OC<EXfZkfT5bpc9u)K`a|W
z5D(M+mlJ{?i3Iu6;!x<s4dGPEsZrXRWkqLdyC^~9h{K5HZ&{>6Nnr#AVSs1sZWkvF
zHN1OkhN}De(n_dEdukNhKqR4`az)U3J0k0PM{{#OKT&9Zxn&EF@78ISh)ONvN3xu+
zX04iI!4FJOpvs0hI;LVb`I!8``P$lv9Ou(!+v#VP`OhfGdVOFx{Xn2?Az-wdR1z~N
zLUzDv|63?J{G9@2H<|(CHhTClWrIPZRZ4^Bl_?W;aq?Hj8vM&+swygv?#zv*TFIs?
znTGOlD#?_NOY9DU6~|K@+Pjd+uBwT34R*b%QE>t!2_^{YYD3E`Hrf^uk9#v4R?D+w
z0~352(epop3usPq`p!2${j?P^@)B$}Jn_+liG8sLULrT}c1_$6G~x!IJAe`TX8s>N
z01=kbB=0J{7Rl;@D%C4)KTX0BPz~|KvZf84B6VB1YbMIL!X%l&VxcjF^ZXlZ-T<*;
z{_q37o4G8&lTHFC_#y`>18@m`D%r?}5-1L-9pm9v3CmH_dfLMptCaZ8$B%;HYITh?
zAE_(P&hNuEY6MvuUDIoN_KXI3K~NW-u+uA*@<nYs2B2`cG1N$kHlVza0y}F?Nj&Z7
zL-X9yn`PEX?5^px0D1CJjitv&CQIF^MUOm*=+hQg*CDk1i1Mk-vx4@-FqSs^gRb`V
zG$lSty0p^eT~`3`j3SOW0U<KGfetqt$lr1|JO(I$4q;R@5{|amDGi+hbTZAw(>e3>
zjmo-6Q<)#U)}4(;Q9H+>U2q1x%#7U*%>F>o{*Mojo~E4FzQn^OVg_aR@SBYfPz#)U
z<Do39BKh^uw5nFEATcFm;m#<mqJ@c&d!HUNISD^I6$F^cia5#lA;b#w<6-@qd-p{w
zcJ4|L92ffU^F*6P1C71)vuSDYp)()|{K_!{A^e<@L~T4<lP#PD`lQK~d2&L&aBSjg
zf6{-aG8uQ0XF=|jFWs`w0LTeEM><^a00^gEWpJ0@ExD(8)w{5?Wf&c9YCQ8e)uR~8
z8$~Q8(QJ-F!LX;WcN&FUni!I)TBuf3e=QkjOfn|i%xqnGEwh{*naF8JDj>Q%{=N0R
zw<@!nj^HEiA=DfMiS^*D&xMgXLkJ-#?$fn%pr<0}izIy(u$Dy&Di8uM9P&kAtFDy^
zK29eH_ZtMF2Q+w1p6|X}$m|e%L)Yg%Y*cj&5kec4!)eG4tOow{qR6a0H_Vd32P^8!
zq)H=)5u~HQ_2$LM3KfWzeR<6z5hIccBNT|pzTg&NYZg~B78R;whn*%Zx4t~}ziq$K
ztf{wy*o+Fd+-TrFY^XeY^j}>SIkGxtp>`D06eyWVB0AT^u>eT|*M4u-I(*C(ws(z4
zdlUIm7I#@N!m$2};5FOxepS>Mv6`TV^DSv{WlrhnBCkAsuKZ>$N0Rx6uRm7I>t?8z
zcgKv2YxX+*Z0^?5uaZ<B<&Cd<f^p?&H)hbjRvl#P-g<$&ZSjmg3^GMv&bNEV++F0R
zqml(TXS|r${PppT?2er_ZAGE`u7hKuTZE>n-r36SSTownZ(a(+2N&0&Se9@70N1kL
zhz>>mh=W^SM+PVgbS^jV6f&ah@)Mhs2yV5d#Ky?{i<(vFU|>NXlhssR-svau*yQg4
zpXTKhdP^alUrFg_yI!K(g%^XXOcO+@jq8(05FLi|*a{X+a1TN^Ud|`1z4|Zif0BZq
zL;ruWN+za%oq;r^CjG-I5xV}vDyNhnYF`Lev*TCAw*nT#6BX0P<dCZDsN#N!x5<|*
z>}-Rg7eb^p$1k293h6=8pmo^p{2D888Z1aLo%y+9=}=&j(lF@dzdDgf7bfXzotkxr
zZ%Ydqlw6<maJdn>GO7jtMuT3Q`{n+1zo2`9UQ_aW)|1G9CChw(d`73Mi=&5){PguG
z<D`b}82!`g>+NoB<MpVbf>q`l^BE17H>Xo<aCb9pH+(2o=yzzEG)$IH_KI{8owKwK
znj_y<p>*?9_%Jq|k&%2@Touv~617~4LA(d*`{|?|&uhqXfN!6>J|^6Uc1o$Mw#qBx
zmgj2Xx#vTm5Pgw!%Mok!E0LvwCzJ+ms}?<6Yochso38p)W3XBGP*%!NeL56pUd3c$
zD7o~y#By3r0j!N48EaLiR$+u^0;`An&jGnEJ9#+M`O0QebyHGspjH_z)#|VVpRqOd
zw7xa=eqmdJ4D5g9*kKBkiV|M^)!Wpk2aqr=I*A?JY6+`ool8Fo;7dgdJjma1cd9g;
zu)(xb7$JRJkcBSE3sNj{>ljEJ6{o607rV_Web(pDMV$;M@ffCOZW8!|JG_(yw%yy;
za--R@6IULrcubfuUgAJ10L?7P>ye?VcI+}}v9-QcoAceYov{{<#a!AflaH2Vp9(G4
zw!u!I>0~$P!vN<&OtmZ+akN{nHq+3}EL7$OLEG1Mj0}707!>PG{W>R_XZ3?2(_j?%
zW4GteuQRh%l-qx<{=l4P3&9X!_I36y5*6oL9{wk->Kyt+&4LZ96v})=X~N{72$hU5
z`w;TJ23nCwpeB>0z;WgWzT~s(=@(ZG6<!EfW|8(jp;GkEZ~(#y#+g6k^2vKN#&HDe
zR{l^)=6+(CR%uIgu%h=v0yjy@%TdGGt-L3Bk+Pj^bdc^mdxd(oTIoOlWq(Z2zHoea
z5(X(_kKFZ&gkN=17|3SYCoSp>#ciyh10=4kr-%Gu0-Ye5CYPUV%XkhDaxlNp4aoxY
zl(lpDsdFGKVt3de4-!n7!!+gej*`S4P>fqxs^+bijh6IaVOu)d^X5h#mTP{E8>kqa
zH}Z$zBm^ynt6y0Hp%skdPjd47Vd~s)mp?^r8}-gboX?yw|E8%=UfC>t$PX&XNyKY0
zPr|40PNHSa#{2AUgiBs|eKJY#!m1(86pz(a5dn!Ly^urnq{Y0r(hum+9w4p-#EAo3
zq1iL(1ydjy1XEazh26~HwAu|E1he_{gU!rb?H#+W_w(o}Ndag18>Q&0UMX-M{lW(f
ze8C?;S`g^VJ&?Dnm31RAiMbrbcWTZ<FEzyYr-+~Yh5+C3gkmVNc`sQVT{xxw*V2Y*
z%GsTa<cdiq@;j6B0BTihTi=~@JgGDC>KGG~q-sT{wCQ@P^2WN;$tq_1s*7hL%nJkq
zv<kfy;pn)pz%C!*tRln%3J_sXV?3ZL-wf)kJ*aE7Z#H%A0o08aAh#;t2<mR@erRrQ
ztB``x!!ti&ZVzUV9I-ci48$UB)0YB~x+i1|fZn~?h08dxpVI*s5{(-T9%SMZ7gBHH
z2+cBH@!x9oHjO3yjq__Q7v;I6N5#;<5h{!Q9>Tlsc`|?w?pJF<%G#C&jxQNPOf^=x
zF}*QEenvSCDDga)1?_~lw%CV6tVDcr_QafIwfDP)<{%~SFIHz%(Ve24SJjgKj!3Tx
zdm(i$p_>&)eKHAykLuf7*ECf1AkmH*l8t0S<Cvb&N@_zzhOu|2fN!>f&oD72@70Li
zLS#L_UziZMR%x&9Vj<Q2K2{W%&R|qz@E<~RYx^Z1h~t$eFJq$V69x_N9$(s9ONX1^
z@Mh(Wr8N^48(nHexM%bI5zE_N-ftstek(pmd3VxnA6?!wowsw}66p60K7dkVChq@m
z?lb@E<_Q+||7xCK{@Jwumou!XAD<C6D?<0V>dmT7>*1uG7?6lB*KRW(G>ABybrjAW
zLU^v9@(W<$clVB*bEP)7b5pV(9?gux&L1885J9TqL~T2UqQ8S*)I5n&1fuIAtMXH1
zp@zSLtD+s65S`)^Acu-3qYWLg27rLaD2pR0!BYylvc{5DYnPIYux?DoECigH4evtd
zd@0!&1DV0z1dJ1M=@|{LfGUV?Fb2Sz#GwhSc9n9K)4M=*DKhteqOn;(oGcxnc>Y)*
zZI7R$=hoQ;hhyP_S}Z7w%3o-z@he%kABL8}LXt464W)uah%1YY0fEzqHxVrQtL;aW
z0vfi3myJe4iDC;DRxm;GLug40pa|I5QZOZxp^hlg$Y2=Mvl&Aq(*{{2`w|d*#Ddr>
zL=>Xj<N6&5p}`3y(^|U{voFV5(`a%;h+|aLMJdnr-Qhq7ns9)Uht_$8HI78GQN|^p
z0g&1gSsVQI`!<0BOc{jXYc$G&2T{I&W`Qkd!R3m>u|f(CSD5k{9&Ad7G6`9;Lh>xs
zb5xZ1HME+jES4-1WNeNL4MHz|z(7qRSw+-oep#jW6Nfs0j}V#*MCNa1Qm3~sAl+Xf
z>^4PESTDY`Y8~v4aI;kmJeG&W5Q6U8Oz2M{Z6H>Z8t(tdcZ?aS%q(Krk3~7`CY;+8
zE)CPrgw_zP1!wfjB5)B1^!iT2M?uVBL0{e^xs{Dm)3Ig~6pdV;yT^X(+UFw-q=Slw
z2q>jRx2Ic|%VXK@q+$Zw#re4X-YjuQ*W*CTYHe$5m)5^7x$Y`*%S9F>0eRo;IkEPE
zpV6cLVe7nNsifuS1AFZ&HPhP9Hm{e%UMnA0J3h%T`vbZY!p}%5G<3GrRV0mTXj>Ua
zk7rM33wQ3?M(eA#6EnIzooZeV=qA3t7?kH86mNDxEr5q6$6di_z_8d1d=d<*bO0L!
zMj7ETcGf5qLU`g^=<`zqPXn9~2jV~97`WsV(Qkp$Z$XEv6-8Ix;Caw6K@7ew?b3dm
zA2mw?)k<D4<{m;BOe8wJWaHS_Ezr$eK)^j$Pznz~*uN#p4hMswAo4&BBYlA_w*GDK
zh;3RAG*3jr%hka2{rpq0`G@EtI}_5(caUGemPM*GtQFGKXFUpx7<YMiyS4ROp`YUo
zUtpG6`wBllJsn7Ud_7Ja?L=G+ot@p^l}qDoyYRa7IMUG>c+h+B^dVow>*Rrc`qBN3
zD<!f0B`MXN2()>R_S91SZvJR<6kCU1ZOf~3a70%jX-hG~tKrD<Tf-ITw+8_=>e<{~
z<blDBJ`kK{?Z`A&II!)<7Zh`w!81!Y&w%^Jk#9;yrg38*^SRA5Ppu;hGPU(@`zlW|
z<nsi22Ia-Dm)=1fZmicdsZRJuezGS%l{X@$pI@xln?H8IcLUNbEf@8*cA^ktI|jyl
zNx=H<RIiK4y4E(fLCA`FS(X`};RMcl|It;KJmALb9rR4>8D7N?Y|CyP($OZfp>LnR
z03OOzS2^Xvac$>rpjj^8oRvMB{KOC5o=a$Fi{1#jxc~}lZ<RzKMIVRMJTQ5{m$~a%
zM9CMOU1x{CrfOa7udmED>+slK>HW&)Hd9n*<v??0kF&9QdU16S$cY=YAF!3l;KsSl
z2yTlmMuaDt>V}=Ay%<RsfG2i&H@+fq&|7+umR+3+<54K6Jx3BfWde6eF^2Nv9La|?
zCrnq%0^>uxW0nH=1IZ(q4FEHjQh303zMMaJz`Ep}505hehj;SSh%c)k#deh|lqCz9
zZ3nmnj++kxPnTPMLdkco9kfqO_YGHg@!igP6W!(ah<G|#a>bOBdm4XPkRy?L-6+MT
z^2onTw8oM$;^%D61}YlnwQR37xi9H`an;oG&HH+_qzl@Rkli0$m=Y!~yP+L>n2`NU
z><%@U)!KN@WAKO=VlJRB!FYc><{g~IH_6dT3B+*E_G@|HcqM5bjXt&o?d*kv{uPBc
z%N`ylL$(l?^!KSJQBgwkLKv61^|R}xDqVl~vNwQ;ElG#D;|L)_bTdX`G1Gk64nBG2
zmw?0p9~_0SJy0xEXFMZZq(T>p*G1IaV$K7cgY@Q>tOK~^mKEOiw|mo=;L(9A1nf6p
zJ(=^(KL{2J>pz6GGX7@y*HuVc>gu)!qln%)x_K>fYx=f3qW&OuCS1Y4d4Mzl{Vzbk
zXyqAQNarOLemmE$?qr_@5K$_|H=UR8!CN`0x~$wPJ1Mi`(Tcb_K3&-+{LF`lRwMu8
zt(Y_7G5AB$DwG;AXT<VGNCZX66x}vF7m`r7KX>?W@pk+%FgC<S1}2KWVtRtS!^PRf
z(Z$}z_WgW1LhtPD<o@SiBV`3&b|CR*-(urjHen*c=QTQGp&?^oW8?mu8iVNITk7lU
zlYE*(%AO|H7Z1mkaTuv<fFCB-uQm?s&lru{=In?Pp><(S(ovIcI?_5J8P?d;_(Gu!
zx^S9w!qHZ}n$e?n8r?_b7h0g4QKHqcdR%w$LOaDs-5^`L?72}04TwI}!c8cfHgE}6
zcLHkVpX6h6NSsJRD#ygtnA7^h_Z~81v{0!B+6sJ6m%8ebA*#u#KfGFGuS88!jx9wJ
zL<DEo(p<`d`Rx1Y58_`aD1ftOLvf*cOQm~fzMZKACe19_DTY9sHCzL^53Q+F<pkxb
z<|)5ZhhYW%6?nHS-e!`qL@N&0Si?OrHJ;19t1BFjHFbWomGd(%byBEK_0p6f??~Lu
zHbFc=`1t6MyJan@YXl9Y3_X;TD~Ah$y8sM^%KPH$MPQvdIWU?)S*6e>yc_S9kKdV_
zDF_{+6~h815)y+nNd}`*3P#RVTRM}g8zW;f6+z4^M7H|UQEZu?3&KlfiozW>Q~Ebf
zSV`r7bHwG@rX86pxX&vyNDk0IKbNPjR2+^iB%S6<RkWIxnfMgP>h)<Ts~mEE01+Fy
zj*KP}jHl)}Ym!e}F-asV?`bZ*-P&VX@pm=IRm=sg{E9d}_tY5h_BOK7ZV+4(I~NX$
zjw+qC5ehOn&}%#i)x<M?9n$kQ;Oc3pET6dW#%)gF(341y>d%9B{upy%sq7BvK`{88
zNW>fKIv!2&l!IJ^YfTI2HJ#CmWtgAeSpODWlp3`Gum0$7>xU8?6DMY{=jQY9^}?LN
zz9q2<lCQjn{`fuL_IVF#2<4h)Vy@${Z}r$JrFz3^?r58Da5bVKGK5gj-Q#W7e07m=
zcy0uVw*?K1u*;15Q}sAKKy$J)URk}cyK{Nh-rza%z3>P1W3`UF-`kaQci-C5nRt-O
zFRo-(hwdDPfZZvcbEGkMnN}nfhT*#W<~Bpk{Ja>oUB@C;sG%P?OUMoJFaAni+~G_&
zq7i5`j#_`9N!ZleLjSr9=LU0dz!iZ+s7V(@cN_p<EZ-xHI4()s4*qRh40jhLO9sGs
z_J&tfc>TD~O3at0<~Iw+jU{+_;{{wB-j3ME2sDo@ce1Bc(3Rfk34;+EVk;VPYnpsj
z1IM8L%d?uvbdlZ)k(Kgm0g47F{oOu<0Re_0AxG6b$&Nn~8$1rG%D|Xc^Q6VD9|CpT
z(H;vd|Mxr*nGKNudM3Pi==qJHK;-z4*+RD;+jju5Wj|5$*Qq0j&R0n9=o*&V-yB$#
z`>p70@rAJ_fcioo$Ik+qD{OY4i;7xV<h1gDG`ImfxEo);`&YSNa}=r6Mg<Bk{eF`I
z5JI~Tf2G}eY@zi#9DVwazPJkk6`Tg~!U?Snm0bEyO#JZ>+2)xEZNPB816kBj9+{-(
zuJTdz(Is^f%=jgOnNUS<Bp^gJ#tIT;2r&ENaT+3~Mlg9LE<x_V;o6=NJA56eY@8{w
zuT?&D;q8b}<!tR^T4t2;^`!3wo*5c$$Fs4e1ZFoPe4udltsQw_iAPu-5l7q1S%ZgA
z?MblsC3yTXqy?%R*KEMh97q>&G^E0s_xl{pDA|Zh%|m&vRVi@pVn@$$i*Y&pOF>e@
zY3WHin1~OlnF`>{z+V1R9P0%M#wFfCgwc?0^)G(@@<Z_k_5;Y(4&#AO(;y%dA6Yaq
zqeMtrSV{K&v3i(*bpgqAY<3K7qI-0F03jg&$wy%U7V(R{WrI5v<)+G2mG2US6j+1`
zs)gB>y8^)=8egzW!#v`<SRhA*2*M@F<yxooXm*+nYBbOWqd=;+5=@hopLlbt<gF^Y
zqqvmfh!&^`P({Kf24k3O`cA$%Z3juyP5)xSt7`9D8_bEjq8*M+j5Va*Mz~aH&ChW1
z#?b%?T+Hab`3YG`1+wj*$emH+VL=(&MUukT5SWD>`plI9>0NhJ5&+;?7xFO=CB+dR
zwJX8_00BA}-gTiOk<`UN#a`CH9*JG(O;jT$&8TQ1oM~Ve>|j3{U_hhl3Z}r;2qGF@
z+A}utxMv1oFEwkECfl7<8EC~)A~@vs|GNWL-mSSBWa!Vr>KZWHPzPmIpZ8`zdZnji
z(45;a#j;;D-Hjb7h;fU?E!z`d$~HuBX;#)w=nOr3Xx7?Z%7;S2&k|><7iE&64<lF}
zlXgpBQUk{D?E`RIE0PyBw6UxD4fheQ{o>T53Nl0=2Nccx!Z-H!O+`+#gD7PIEr)w=
z7v|_v+Fg~cS+aHOifE@_`(dLUnwO`&!E7Xp#K_07A-iWT46|2oAPt@!l;$kNO87h6
zom;DK*DQJcfqiBjyXbg7w~gpr^{RXEi8ZxG!$fet?T=-|W5G3GDEj?DL_}>lm~Op`
zvY;)AeFF%%k3v*I>vNp0&w%8lL7V%L<m7@!k7T{4r`UGSv2d`X&1q<rglj83?@Y;;
zv-uI-I9p*w*tvS)9<QA#_WV|6d8vmoP*wv|O+Wq-T|$VF**G?s)+sF-7_<FPbYb<e
zikc4{)Ed5nqM=-f6HrN{Zs&#Q*#N>|&_&<Z5HR>Gh4I@B+e1J9nx!2$Po&%TS6SD|
zdb`bfKnnupcq8PUS0(@c^pJP$7A^lkz%mfE`4T6=#YVoOWv`2zr@CjyIS({T@ZTrU
zf)_!}^*i<FsMSpd{M%{>emKW+3Wa8sT7vXdUx4sOv!H`2pDAn6{vjV|-HlV*HyW>Z
zF#Y$>hEQwf>u3+XsapQ{rx!OK0>3G7m~-i|JM<ZJ7-AT)ONGZB2F%oPJVfnq+?mwl
zrJ7U$La~%~$zSy~=G`rSmXOW@c}hZ>8J-ZXrb1hC_^hTt^ZXC&khgZrJy0BVl{U8Z
znk+DOVt1NYMc{Jrymvq`-e>SNrQou%=8wWX3LuI$xzoFZG1#*@-|<&Sx1Im#LdNz_
zRVEC~|N269r2b#zxM>MY&6BZ(%~|7l*AeG9(TJ;&WFj+8C`O3$R?6Em1&9)mkyO5(
z<Di*bZr2-d_4$LZ!3?GD?{mV!;o%fGWj=;M6o-TjaUL-#GksaU5rWq#f#H_zSdJ}^
z9h!&912jwt26Vx*Pv<)|d(YXETpb+i>?P$CM+rW!Zgy5q+gd|}>~+m_!u?Ur;o1I!
z!=KyRJ4MXzKxI^UqF+RkVT}~XkL6O_OnQcwcT>TqtI9z*+=lu9V;$w?IcNj&JQHuH
zFYcK!_51Ej*;+C9SL7{A@%~a;RD7hC*ELdgRvL~T>2=cy*xv2>$v!>aI8PCt$KKyB
zQ&C@999K!=rQQ()O?luT{E+B<*VpsortxSE7a<<qm}rzmN7e1SYHaj{<7$J%d3+;W
z@pS+}7^KMIDVwkz+=vx$RSll8U;A4b@9HTVGTsT>MhBIye^e42d0c92@ryr+CQ59+
zJXAK@UO)2wo+RiW1TTTw!`3P23(In%`C$bn=E<^XTcz2O8SrWSKA5&>!-c_kG|{_j
zE>CH9P1^QfZsysZz<tC7TMpzGT|2N_M{!jT2QwEUp#?WuX15+~kRl;9GZxIz*%G5T
zknz@xZsuCk;lY?3So=Sey<>Z3UDUN1+qR7gE4Gu$728I|wry8%#kQ@AZQHh;ROjip
zzx90_{q+6;dw<$%&Nasz=QwjeTAvEj=E>obO^7Xn)Ls76`dNqXtJ$;{J2p$B;K+E@
z<Ffkc!$H_4^Zx3rS4bH@<~iR>LT^FbPri__@MninH{OT~IAPc{A4NF|iO-BeBclL{
z28BiaT1_u|%rZH2I_Z9LVww$jUnEB>267A|2d`^}XsMR#R@O(Tlm=8OGD;0n2bzi8
zN*7RJ1DHkd6UXy{`w%SXeMDy4$uO#xU_!R|s4+4sVM4O`L`bhA6)d^|!*D!*(uNJC
zd2t89_URrh4KPN9=4*N`xC>Sif!;}o(S4_ef)cku&88S+Mkr=%c?yKCtH$(Cxlj5x
z5HUnv?hV@0B4qa6(VirRD38p>XC#wp5^oLP9`aHwCb6H4@`EpARrX>U>Pi9Zv6V%E
z&UMJ&xtvlvZ$EoiYf33Id{QS3o~knHF0={Y$9yN2q$K1XPr4d~ckZVxf=C8Yn}kuG
z6>LaOpmHI)Wo$?+eD=iGNLuX+8#Lb@4uR{(pLWDebj$<@+WQHJB6~uPdH4B%zU6@*
z!qSYVlsBsrOaYo~y%1ExMrkA#+L6L|tIg-be$tFPbqipEeS`*aOz{8ka-GO~2Zq3T
z@b>*<heCyzw)wz4xljG|)@uCqrg(Ex<uCo^Dll0YEU{;}0rT6jJte2S;LtHv6}sR&
z4!v5L5^2f>7S`}i;;Eor&H0;LOom1-Cr$Lq44{CMIj?4TIN3H>JmDqv<aW@I0Z3Gs
zQGU-G^Yu|gB$m?vlEGb1V@gi05sgqo(Kbc3_C$QsF5m$awZyR*Q&YgO68;B_+n*<W
zvbFe`KkvW#j{&E%|9+(pD_zq00^s^<N{f1KbUDu9x63v4Ua*Fhuze?*_xdbm)~lKR
zY1`C&KuhpvQ&anaM+dXN|L3vxUxFCt_qg<bIrwbS`1L=6*#BAHrwZf@pLmF8k8?n>
zJTFac$3<qGW#vsSjZ{H%;rYDO<&#2XeJEw)Mx_WCG`hWPA@)!52uyOruA#!7-bSlL
z{~wrn`Uds?!@QwtI4~2#7@Oh3M2pkY_xU(|n>?80DgukE*7tQk2HqGu`aGVsj4n<S
z59Lr@=lngiwn&5z4hhDs`rc_ibq?r4-vKq0hT5cL+nlfVlQFbq+$?oTP@m$Ube;jy
ziW#=H@K3HA>D7!|>TRk$c3RsVFYYz3RboBc+H+wBq?hh&o(R_Vyn1HY*$KobN8Fa`
zG*{QT#QAL)C$ya3Rap6TIZsayHdZa0B~no0itEr>C=3{gSn<`UA-^6PgS&|X`v)k{
zxB{6Wx9vsHsNxqWb~8wYEm<}F2S-I&<kFMLHOef*ckecH7CmL_6@5~-Q&d-NH;2Wy
z*sERsYRawG>$T77k3IGJ)No!l>R|xb&9WpRx={(+8~N+tGDu~`xQcbwN)bC8qA5Bd
zH^a{thrtV??`y~5ubB+oIDJB4F6*9mgZ<zt%V$nHKXe$3L%q}~_97|y$iw$05P6v8
z1yPTsA{hvhJYkBG&lP>_veQ@B8nY)eZqB*iMOh;nkxFWDoNK*~cF&0fFKXTUoXR24
zyr-`wdaax2Qk{-Z8}&GBZ0CkwfYWBC@D2zq3RF5~@w*$imQj!S!W~lUln&!CI4onb
zdZcjD25xtUX=Ss=ot1YT<@W3g$>U{qN>b<ii2(t@B#aP0sk+3+P>iidoco1G0W!?N
zhc2tK%i`xYHTQ=boaUzB5#B9z-0TFnvK{agl04+KNFnW}-<i7<B`l|j^NsBX)?~bi
zu79|UG`N-3pIMTB!)S(m(Me2&pJWR#8Dnwfm&@k@mdSs?3^0=q9Hyb<AG<Aw(Bicw
z>~4)*vqT#fdXh~sW_!n1y#1<ko%H5*ikbYq=+jZbIi8N3PI_Ze&T8bn8P&f<mV06v
zq$K3AfOa6#OdMlCDzxCJp*T`>U*!XW=YX&MJbnvAaw$&3K0o;aRZBj*uqntE#jbMQ
zl`jjG&fkI`SokLdPNEu@`-Tn(Lw>;#hoh8Iav`7tD`PlspD4IB7K88)X2(=0380Q9
zV-8vTYS6TFn8RA;p#JoxTTu-?K<W+yF=C#Qht!-U680O4?wLwTYLm!`w>mQbsj`={
zES3};IzpEcjY*Ez%*tqBcr<goT^5(@S3xN>m<L?;`!}BM?H!@ZncduX<w<8z>RT_1
z_pxf9+AC2KGx<KAv50}-%!`4xi1OeU^CLi@Pxp53xbGQ}INs9OJ%h2H+%d>83W4Y%
zmylDC)v86X4sDt{Bl!VM#A~CIS;StE!aB%y;1uC`tQCbGqZW?++fBo`w0-}@!0OO=
z*2|Bk*NC;8gYJOej~&F}?RDMdJ2?rN8r?`<|7Pv?hnCX;bZfm~G*ko(vh*7l_exe7
zpL)!;yH$t);w1CuKe8~#hyF5`u*X?oqniyrd&4eYiFIk}mrhp1vXXQUiRJbphxqrn
zKB26kv(@H+vlwf1ptHp;Ay;74(y35upQP7|X9tO{7*B}uKcEuuRpHmxU~DFUv4)GN
z!5@WejF+=YgezwRx^@6`9{AVW^}CSB0+4=hlnBF?eA%mf=}v4+7rRbBk=i{U-OfFo
zjDnYgLM>vsck?%CNynVP4Y_j*UtB=_q{4Jf7b+9Ous3}7Om`Sy+>uYHvscOSg1(zV
zUhH}Af$g0s+ms&@O71YXCEDmQ)r*NhWbAaE>vd&0ov~GyERYn2fWe$nbyaa05rzYk
zx#T{vLYCEjKjn?P#9eDMcYw(01WvyCvL=uYJs(}j9diug7Bz?ca+uz~EkfDxDh^Xc
z7jIjAhF5x;nFK$y{}Y3LZ{u>(2PztwBoyo454cD)eC`6Dy&5Ma3X+oYZ-LRYyPKi6
zl(#?)jm_fQ0Cr$8Gv)w?&WEnMYi)TE@l%5B=S;N-l*J|7o(c5^LQ;CptS1Snjp>9l
zWo|Jzz5&4_mJiZ6-0t)tm}qYu0SC8LXodt1m0>M18ZX+E1RY|5f4|URp;M(=|B3}J
z88@bF`MP^N$Uq`7_uTL}GlUX4y_UooNs^6@vRokagKCl&A(S8<`fG|XK|k5D!Q7xO
z5h`a@pk+I@IH3L67Sf&9#7JJVIJ^!WHQaScQ#q<-52mnorWR?$CY?DEgJCZL4i=C3
zOz0lx{Rm?M*JHD5w(i|(&hl_uGzYG;Fl=Y~=Y2eIek(oFP!_HD;rtLiWz^$fz4o5d
ze*rjsidTZ&&%+1*RWG^!gS5lS%Kg8N0J+q%iCb?){p$WJLfmCBvH$*S9ivr8uYHmf
z_Q^Y^`de0N()+3BML!Drq_AOktw2GkSbWjy1&MTGD0pHjd0>B(o*pK;Z=JP!F*z7D
zILKMUJ^p{2-W-4|>99XgkI25Uq&a37^W%ODUr&=(fs;ZA+3{q|67&z7`hP!b-M?{v
zC1BIxBB@TKw-<x&#M@ook8@o<9*_NQ%QaXv?InYci&4bb>et(!2E%`Q^vKh~Yi0Yz
z5?3RRe2ckNrSMOWDk8o28}oyaf@vjNm4B-)jh9<Q#;JDeViR+YFIMSGvxD|xD3bN7
z`wfGot`~P_{)#q_ySlS;JKeU+j@iE072O+(yNF1bRin&Ih=!#=9EHG=3d+KMX5QYb
zuBAe4n6$OAd$D#q*JHmD++aLNJi%oc>atWnDE#yO{z+vwHr-{foWl8DykV0yaOvBD
zAn)zkQ6bS^v#ZVY<}MAB!w(PVLrd3@UpcRDw!ut6<I;Y-p5k=;wVPMEaXD%GS9ftl
z8&6?DE2E|QdzK1Cv>1+eHt~5$soXIA?Od?q2*J<$(eld=K0ARn2N9Y~K45@?3IpxP
zFBCSAS&fB?>sU&OWdMSc2;F|(2&{U_QU_X=CJxpeof%YVeDkxvxRt3HpvV8FE}GVJ
z4?}fh27><H$m`5s5RSU1@^I42?`a=Jv^NvP@42>K(i8kK#%!78L?v5($@8;4!3Kkx
z#gPch#LlD`cW+>S2T;dY&d0?S$5-Pu?Kf3XpVUs9!6bf;>VC3=)(O$l%i=n)#A1(5
zVEosP@SghFK1Z+}xaoR-a>E?_qqG8s>^o3)NWQryDn(Qg_W?w738d`3Prh{w_`SnC
zvE*B)x2ON^y`BuvT+pkTw0(sIVKdj5*1Ikb0Dw$Ft`vI1+$?hFNt~F9kQb%M>r{a*
zd*PFjM|;j+(IcN7-R70@7oU?E)Y#|k!C`-plGgvua(Q01`%#~xH9JkxUNM#fFZdO9
z?q;{43)9L|?w7k+Lx34H2S<z+7>(kLLX=E8l!lM`yG5AXCl?IdcU!$8A1=zj6XLZ%
zr@e$G72$6T)fMUVdLmbg(`$%#sVnCzpp6zN$$8+Tw_KMpmnZ54tsG83+vIe!Xk*vk
z3$3?yhaX;d64kN`j<`z9VuPN=2qDDd9PTh3{O$5X7ty*Wxw~r3#v_MCo}JJiJGx41
zhHx4X=BoOUqxQ@+>_Xa8%0-kO5K41hNOJDP&wAZZ@WvXG#(s-EUg+_&o=HwC6*A%6
zSC3{UQc4iU8_Wsyun?*mVrnIEXkFG{u26mn$8G3TGBJEkj?l5sGlLwIO`@Xmd7KxR
zM`N<p$!$g<gBX0AoDe#`A5Mp)7+L$2{GmMei~^|b30TQVx>pu630oS4)$|nk$xL%2
zWYZ_)=brzw3=O8#+!|?Gm<nJ4lW%koAdk+4uJp(SWCIs_5tATATZNP?NHTZO%mYj0
zk21{_LlxZ&Y1JW$fuAKU<0j9>DY=RD1GIz58m2?MiY*m};S`^Wt@z}}=NC}c$jcVs
z^vTAh@|g?i+iX7G*Db8|!uCYc&LB{=zTer74aSeh7NG?>+hv{ykKyeiQFQNt2i)-)
zfJN`%FWF(todmJ3OTy*G;bltX2qn82gD&NYWq|X6g7;YVD8O>pMzyw}w;RkB%_nb@
z$fM0__vX9F;UqKu!sN0oX#CuM|3eE(YXLid_AUh5uELUuDFuBjTAtZx3OtR?n5(XB
zm=RPSLG|zWF}+QRO#>T;l{P0zP&i^COu)jYgA7{po6YMgWO1fKA*Zx{YIkLxa5Dqb
z-09c#LaxTT8_I<i^;WvDTFW(q4_qT7Wg6a19PI)NNifPxJa2G#;v=au7jKFV2uUT|
z-=YXWIgWA^xiKQ)_X-f^YY!%YmH&}_C6T^0oW2{nRWe<IkYogIP}u<tNx~WdO#~=5
zDdPj}CV&G4xI&>--D{avC!45^*rDi&!v8xtyq)iX?~;E$i2^NuyXOzogb6A%u<D=S
zvhPp+Ipv3wGAbO|CGrE3pKGy0*Sn|QG;(k8@`@W$9!95%V{di3RXfZAy5xLal1cv2
zQ_z-)NC(oyiEskabX1b3WuxoCYiPK?p`yb^V&elGgcOmS3lY6;y%y%1&1Xx_s4rO;
zEzxBcLc|u58uE$srzkCWRmCr2g|c-evV%MO`4i70z_-QGE@mDF;;a*cg$SZCeOQZp
z<gbbw&G$;5`a_ZIIB8}($2)Dc?akDduHK61fJ2uQ(dW;NZru^K-t!Jk+trTC@$3zr
ze)B3GZua|eAUkgj#v8nX!GHe%C%bLE8&;4B-<pa!E5P9379bS&wI`50QtUVeiz}~f
zj#XZhLaSXW*A|0)?VUDY#f)~s>9TkK7LtK?Ma+>@?Hi8v-i0^Y@Z&`xKBbNlfkBn7
z4!o)2otO0ami{oU#Si_!1L#&Rbw_LkhEJT`G$LO59Ao-Rc3rQXQcBrs(>%!GHAzNb
zjG)9REL&}_0N}$I2@Vk38Tl7>CMOFzmc-|iH+P4F$8Ay!Zf_)M>-P94@*C4_4xW{!
z=PzMKV%SrbOb5boPib)u_r#FLj-&y5MVWob7txnrf=wU&U?S^afs>{FrL7_TYZJtU
z+4@s?g~3qTokH3!tK9S|@t`A8>odMY4~y1Ti9<?%1IB8MXN6A%(Lo9{fBFLky*#LH
zRn*RaS~w)7#`L>EJwXdaLoYBGR`xvN5DU6#XDvon+=KN~b$hwwXxfURco#gd5I&RX
zvpQ1Dg|X9r^<sU`{mos#F>Co3e-W%q_4g)SZk6`fug$OOE1CB(r(7&-wB6Ul0b8dr
z0S55#czB@LfGhdH)cNDT{R}YylED(ki%Y~0?Zz?zgO#_`sUd#0ek2BVL<F=a^0OD?
zs&!SL;KW31gMVFM44HIlekvJup2Yz_`#-C6S4(c_VOlh6XM*F72fieFO0MRk%!Q`K
zC%}=X_=SfW)9<5$<THa8Dh3C?C+<gu{{PWKaefD;{lAuQS%0$quY0INEp4auR!raN
znwbK=7?cgBO@R5_&~2;31>vy#rK>AVgqp_rUZjd^*Zis7Hh3VmKbZKcgKNuW63yMt
zPG->c2FT8=QLv;r)0g-C-Xd?{<Om6Q8Nud8Y#PD;v`|LQY{Z{FvKVCF4$;=3{fhIJ
zlI0RRd8&b(+rh=v*J0rO-&Q~KlGKv#QaAN)$y?wrpO4dJwxCWIf0upSwsAnf<o90g
z+2G^p>f`Bf+_8G<O+@C${!vZMmctS30E=)YZhT8}n2}k8-i}Pj)pdnisCxnRgw#L?
z5uj9LO=|eSIH6LWN@eHv(w3jTev-$H4ZTkS59*g1($R<oYX(XzPIaj5yveBrZ5|Qh
zAxx6i$fRqw?ffh9_$1kQl=V~pov#Z%FaRYo8tswGD$82;fEqpG_GRkxWBt=x&V119
zJbRwZ<MeM8)P$$36iP8OBvAq?x>R65E2)$MTf1g&C1=iD670S=!d^RcTG$E8iM_}I
zR~D!KFVl^`?};XCfZ|nFdZLYS)x=}WFzsirV_&S*V`tsE(Iw0J$MnMc30hP(1@aa#
z*$f4H=6?U?NU{Llf*!{n%goASS9{(d0tZtR%D#k2*TK(Xo#<~jAt@!$6ewPZ5`Ox|
zTdv_RpTf`zo+_*Z#vCDm@u#rR+o^Bi#VL_7>6Fz4X0qY2(z7u9B#r%hwjd%QAADMk
zU4qAQi@%c`V0-WvwWNDzcz=P|1*D`F4@VGLj+Cw(3dPC1^8MYq%RU|`Y!Pgr+3Gjy
zosDTg_FN;?Qo$E%BOH%<<gij~1hc-t0%mGF1BK>(67+<{9$AI2CMBeG<FDH;htVB2
z8%3%);+kM^N#gXoMauk?XOl|2j2^$^0KI29`KF0U`$cPRdX%GL==L^F^LDiByDV%&
z?eGNy%1LYJ31zT`&2r(7q;{v6qsEjHgKn%?Do8Gzty%FiurmX3QIXz;PCI3Y+T{%S
z<<rM#InED*`WGAfmVdZG4T0pqb+4ObZQC|iy!G>eYI9}QRH#S&-Yoz7Z!Lh{AG*+E
z(7P>t^H32ve8+VJ)LvmFIRv53m!k9-=DW+XPyoIbSgVjWsbX+(rlq0=z;Jn3rqdte
zFjI31EbZhdyQY6$2%KY?-Z(U+m;8FR$b$Hw*QQ>KvT0u_ex=Yz@U)Fis1{X43Q@b{
z>kk2nX(`p_iiE0=9&}V;k5tX-kUYb;1qjhgV$_b%8E0-cTOH4;4=6y<zw98l@=fRo
z^}?8j3ho`|;Rb_F8-ad?^|SdiDFEVfgjpye;7+!eAMo;tRcM!5Ir_<^EtO4q@$7|j
zsN;8vOL3VYLATPv+EZ@N{cD3AV0MpKu&6W10IRo9q#ID)m<`;!e6=u=DnGGa=lu!B
zB8h$=xFYSM_j)D`!&s3w@7^Bne~xpKFoXmsoI_b#s%y{$C&Ukr%g_$c<vu`9<SWc8
zIMx~#mSr5YXJIYLp`NCp;26#7{xl=1Bz;J(($ft2HAm$Zjo;sx`9DR%Dwm;X$1n6m
zXw;kY-DF&h^@jlEASwE)&;(=*a72BIEKamiQbXP5Mb~*TZ!f_4G{FZyi%3VTLJrma
zA$W|9T2zEH>xH;`=8r6^n8ry&3jh+GlY}c?g>>jDA18>}5-eO;1l0gq3Ix3tnta2j
zBc{c+uq+)`#eb(|rXLWVcNLZ_h&N0aC<9`OPPP5q>d&I6;mKoME5}LVwj!iA98(xY
zs;bvK2#~eL@4;*ND3i3dUJ#!C8Gm``zrR+W`?x-ywms$;q6I#SLtX2ned8DY!~r+9
z8NsR9V9<+%j@HOB*C^>AcYsa*!=R4OFcx!FwIXNEbDZvqd_!$K1_D{}XKpfUyQ;AW
zVyq{~0PWb{dUzp85_jm1yk3YGo;V>&6wKf37HQ8*I7Ya^zoCkdU-(xV9iJ<j_v;zw
z#Z~`h>v~)CXAz4-z?6L+G5>yLbw|R{o25ejWE=XHp4<0*2-S;MoDRtyHVjE=wvA_1
z#ji{wd#GHQ7MuX;lcEZy2<)|paM1dSSNG=`LBVTCxYa5VKOh&O85?5CdD5>nH$S0I
zO~?{R3L<`RvV3HsNx+9Twn!xp2SI}5Q;iwv^!T;nnUSxlc;ar$&kn5pVsk-pQrxF2
z&ExXJ+Ywo=YX|rPH|DuUh-Lfc*i|ISo%}d{^5IM3MQ;Cr+vADY)p+Usr#{T`V1m~w
z&n2qz#nj)rJ{mCJ4CRa;ElWDqSU~5NZ8%c6ad7j7@E%JB9%6XiVR!D<oE}ED-`cx4
z;1^9vqXDHYC7MqC1i&-Hogz}TT;)$7`>jC|qyj!uZgn8Zj^VYlzp`>q8K<68<22=K
zb-m73KI6Dunw3G=zc%t&W&Zcnzq}8wI)sShkK~e@w?B(YFk)rH$q^EJ&ZU;%H!DM*
zvMUyRsL_V93?4})f#x0l2G2Gd9pH%rc(Y)?Jr+UdM+59x<yrm>GX>2g+EZMjc@1U^
z*cV=%MNUI?l!YOi#t+QKWbY@H`gfL#It_tmOL;6o5DBZGZH&*4U*C-bAsnN?1d+ms
zA9209%CN!U>3tqbQuz$z*+NwR9sg?`uRG3Fv3QfI?lN0tvbTUA&ObY=qlYRstkhc~
zeC*$OtD2rjfI>lcpPIRyj`B3o^3A9iJeenJ!>rUDb>R_HAJ3VsmpV2wj7pkJ#M54H
zo-%>r7BgJv#$)*Q*5heXc0|yeSN`wtpF45AoedOB)9902<D1uU>y<PcTM3qO@iCk7
zj^A+z?b_27#=dM*Zu3UK*8Rr2!K0<muZxToz+nE{<eza}y@lcRSQpbp$Kia|ioRy1
zP!zhdq(^~Os&UiyiMteDNrsZ$ChHr6rqREa`b0c*0o{{CJrHaI!X{q$uC?9w5Y%u6
z5xm}`*}hMOq0Xb8u3A8JbcV6b3+l+uFoL<Q3lG6uOjCctg8s7k*S;37&sJCD`7?Zf
z^6jcK^aE?%e1GVQtE}eVx>U7^elFbz2N`k4Wt&U7=9nWplMlV0N0y2v=Yze2f9flx
zy|18o5mfcf2Yd`6)0$Ok-eaYH#y~pbK43qA$pggNa0DXBQMP;lp$5iGu(-I&zJd-j
ztPTeA(geFdsB7^(RBA2<$(3@hgU?nUrico1z=o;wY9CqbD-r?(h!jrHKaqa#$QkZj
zw(&dOK4)sa^0LS4^=`E?p*6V;F!#tqORVu4nxRywK;tLCn?x4835;C#Ox4?2-Lqg7
zt_?8a1j_}+wtFe*QO@mNX0_lRv|6`UG>{P;Ll$<bYHds4ck_Pi{D}WNzHM2?`bGq_
zEc=$*Rsz<?ts1s?IZHU1d_k!mvfKW%XZ)`$AZF(OHR1X)4Pdv<g%W!Gipet;m2r5Z
z%)=4W0pxFt661G?68eLo?$3yum`T<1^+FiYC`u(~&9%t^^+XXt5UV#e(O)MUq$w1%
zEiadp)Itn_Y=lN%9C*O9Ll-ZCN@o&((3={s$nuu)@%jWlv%=C(oe%1RAaVWA{aKy0
z4JRhOkko<|@0WQ0ynK#eHehPvWcBs%JxZE&m5&npxa9ldFHg+m^^bai&XWWhCIp7s
z0wjn%944Gg3hQO*p3X`FC1fzal;?rod8Md-q_TGU-g9ZY)IMULr?TX<F*9{D3wI)_
zya}1-L(g^lLM{Lrl}^E1!C-ret2~lOsub915%~ePt=z1et~;qdKKQt;qV!NsgZPgn
zb)B>8++c!+I5Ro`HMRyfpOdluA*OEJo)!mF4+nxeS@PT-y#n2d{RE?H6vw6&%$X1s
zi_Q4ecEH6w>Ce4050mW4di~f1BYCt}?_YTlgsFAPT@{UC0|I?lOVdo9n(|8e^YTrs
z&bpq<rwoYp88b`eLMP*Kf9_@-^w6kFB4Tk@4t<?TmR4=2Z(+1fqvmp7|A}%oY(ZX3
z+3#;VZGE*ZM+9#5RoduqJkyL2JKXc#;hI1SbES6k4{d3TP=I~p59>{e?Kcdu1V)9<
zvM{hgeq_unJF~D!V96o#@e(uAkGbZ{>%H3UZ#X-7Fx!An9T$@MAwW^n+*Fl>D7J!(
z={>7)I)y(~D0!~?L%ZR1`)a?59HOp=S!O!|mSv-#v9^as3$ej61^70mwpl-9-TZ`#
zjE%G)F$3Jz!f{U?MdVt=O3x7U<o~PDly%F~v~^CZ0Bp3TMh+&C4U`}8J>kcQzBSx1
zf05Sn7xRW@s-O;ta9IDx&Adn1V_y~?dsqjyKPrfUGyVO!LG;IL+}I-On_!o8&>UPW
zGc$J1U5y{(zs)XhR)&H)*~~1nQ#WBmqWy7CBF$!zI!Axn)dsaRu_}m(_ocSP2)J>|
zqF7gdW&BJqh%gb|+9wnXRF=k5?!^cgA>!q2OYUe5PIY><CQkA)Xp`zF?%WJH@N~D{
zRe@_keYGbNqKe`&eADQ4UL+nXI^S0$HpZ*aIy1gDzL0#ZBkUx}6zMLF>+&^ttD($`
z{ez?a)jQHE*WAP!{u4E`Yf0^BbiE<P!^&5R9-MH0nfjwptr<Dvy`{d<a;}3(AFxaC
z<TS4ok%xs``tZu{Kl>GK8cj%Fko_Tw`<TghX4S(LVfpayd6nr^z>AcB-~4OFJs*|`
zB*Dl%A~uCo{WA}yO`zF@I|4I!^B{$TTYWQ|e@eGrKg#RDZJ&bs72DLlq~*Vk4wnDm
z)BXP$*)RUjb&~_@`?|SP&>5pZuu<2sl|D(a>cFOKu5+op_U-z#aU%aklpuGWmG$*R
zgwhFuL9=Q}#35g3aI?WgY%kjYj5hn{dHefz@4ZFH#w<cr)->uox<VFCyfl8*D9a};
zfC}Wm`Uhr~u3nd~)8o<OYNqse`jE@(nbGdHzWzsJYrFg7_2A%PocUh{-<Pkd9Z~=(
zYzmwv2L*15?L7RNOaK`PXV7|z{ra)3;4%l04G^vRRpFQI!2h0^3iNKH<y$Y`s9sZ-
zNPEZxZ7mrjWR&6BUnM{=#Nx~INwE7+gJM{rW3S6_5xIIgH`1>Zv~}e)RZXpk;{<lZ
z0pB~$F|c<6Vdy*T{E8Eg-hIAac2gVL@}1Ret^VDe5dXW6>*(A{{rmd7t@n0(Y@~Yq
z6dRG0ftttO&D<yKzBisfwcj=Q4wo*{_asz0A7K{nAdP`1Y#}{E6~#-@-0JegW)}MU
z4Z9H9GqgP?ARY7kt0o5Zqvrj{uEVsO;hSl4f_;5g?r7)Q-0?&oXO5%EhLNJl+d(Uc
zz5Yj|jq%P$s$3&gE`iQ+P0yVK&IVOxJL8J!YZt+(4@3<GU@Eq6YJ8R*H5P0%5cuRb
z*<wYkUlI1^8)jR%mNOvO^_l;dx?eP(xZ^~FGI)S%dB&#ItMs?L%c|1$PqVWu5iKf*
zCI1~U=~*aVFQcU!LEz75DK9cBBC7eJ{H<-7HHDrX#ck`Y;5|Qwy<jo#BL$<$c)ksd
zKHUq*QE-R5+SRd|>Mri9>+wrKhh0n6nYZ@4&8w&Gk~y;|{7Xpt%C>5nJAp~6yFJ-0
z&!`ZpsThME9kQrR8VQ>02Ou?LrkKj=!&bhK3L>G<W#}hL?hdU$RTX3xnkXuMeM%x^
z`WV-4rZ^$*2i-d}bYyn-!|AEz_O;u1JtMmn!n{wLcj@>h#~bS&?}~J876>|mHdQ0!
z-lsscwRFeGP|JDPWqE-&#R#*v5K8g-n70G}3hPVm<|Q-S$V-&lCNvc|d$dEoZio83
z`(hCC-!S76<nTQ#ggDpnxZy+gWSS3Dhp(neyEtih4A4MU^u0(+hkp(_SVF-RCnE?G
zEsX})!+{nOM?M}ZdbTj?9A8E$>@}peyM-<uEr-8%Dk2tgwa8vzpO}{2m+R>3I5|je
zctll74NkC<pN4R0X!|snrXt3WhKgEan<|OaS~QGlmVoP>0%jCL42DS(aEnfU!$T7=
zKN%~(k{|es(dlrk-@OtA$#Vx_<Oo;~EJJyiX{hH`VrSg^<}wfzF%3FmCLAt-cZINw
z1|j$!JM&_h($>a4nUi(YlhyJn7!2=nyWu?1D8xx(GDO$CRN#MJp(jSTa=6UGjGk~`
z2T-XrDYKQTnTtzhpkBomO~P0v=0a-!iJ?zQ&~0ZL6maKlM5_+zRRRUJ@AhdoXJHY3
zVHiQLi-fJyA}f0|CX(n7SUaV}ub`|AkNn#9K3j(pYU4<b6?aB#;a_yVF-##u3*okq
znC}gPjsiV3E;Wp5gx(E?Nh#hBA(aad*xdq*h?wdFlx_Q2VI6A&hX?!zF_GV#u-h*q
z-dF`xJMDHYHIa+R3@RY*FiF+q@fymUsjWwIQ~wE;jR>bfbT!!};nATagwk>^2uDO?
zpam-voq!XC=ClA^PSh^sHK0*VVJ#H@+@5d(dDBXQ%r2LhS7Tu5zU#9_#QlK8`SMo=
z9CfTvoT)~>vl~)SSsKbmtp&)C(L&g&J^Vd~C59|X(j80A-R)&CA06NFtV!j-T_xJ3
zBnd-L8l+JnQJJ7)QO^oeGa&N};}6L?c{nR9Ga^eSp;#@EYLEcmw6bjhfS9;ks1Mg&
zXPD0vNz6E@=_pva@l!@mQU4%?x*wM<<VS4JPkt~Q>0_t~sN>iNQ6mt2TKKPP1OeD%
zGG)-Es=~585Ep#ScxZrM@Pmg~9==yaF(@q<q|=J<pvsgi8g8}1c2Rw*hFQ=R5?vpW
zgX=CXFu<l^+}C)@c4|puazdR8fuFVm%xrqH+Tjp<%P!yCDUU0Mrg5%o%v^m<Y@w7|
za*{OpIn@h6n;|%LPqMZ%SuD3MU%t4IIAbUIL5C<spwkTnF*f$4`mVf<w!WeMZd(7@
zRM%_3>a$7gUTwSiU90xFR<{M)21~wk+%xALKf^4qEn?u<kDq;`iDJ#XZIw^uMOYW1
z5V@+HRtr<U9IBuK^-*B^7zn7Nt>AU*eA40ismI>3lU(){dL8I0Mtx0<5`$+R9p7IP
z@u1efIQ3cv4~hWIS1fL_yTdc+IMJ8jDNHIu<Xm^)csmG2tOJ5T>O=NBEeO}FP^qpP
z*5>N7cC5wMM)L(SI_hk9N*B9*(J=~?`Ke(Hr)ct|BJW%m*4b6r8s53(U)_&dty-H?
z@QEr0;?D-Uv$f4HqI>qxt|2wFTkc5A&8y~*!pU(|8_4yk`7l#HOeHTkS(3Wl!aKIL
zw_vuk!6eS86mCpzG;JtTX>i$}NIbR#Jl%%O?X<6(%}$l`C)d|Xc&suDGosnK`Kn?S
z39Y-2r82sxTtC!?iTN9}fvpvgmv8A)yauHqv2hWBXUZwTRx*WnA&mSZ#4!t18PU)`
zeX_uyrVF9EM^t{pt?&)Q^lEZ*fYlRDEfxq53~+(|5-=1QMju|@mF+{8_-)POHrOmT
z>3Os%+n05&3Qd;*x_n*MOIfev&DUaWP6m6owpkubmD?|LJA#55xQk<d3EeR1Y}|2I
z)jDe)yEc8+so6%mw(Z+Yqr!GMKf8m1y*$|b+wX+AHxZn9NRf^WjZ|$d&N9m=C*RVl
zz|1LZfW|Id%xO`2lNNX8W}s({5MRRhNO_;RRz8CI0TYvDH|NY7U%sF`NwF|4t~*`m
zUopAGo1i$&ldXaw!E0|Rn~U{}y(~YGp89l_9=)|Y51t;m<2@*qa=+Eb`KLWG&L>i+
zcQf%FV(x7c^(taDA<))-6a5r=c``#imUamcT?IUw+PCtIqY0rDY1-v|^NUu|=+x|d
z0naM?WAmRq)PH1-bF#DjpDn@v1>$<~&$b=#e=^7K)n5{%qRxp8guKq;8qbN=%8KOI
zRk(KwX&`XoWmBGpuXi#r)>P7)<>s?5c-NmFUj<f}>q(+@qUG{uXWQtdr9>$)B96r9
z_Szy@m&p}H1b73>ZkJE?XSvM{?eH`fj!_DL&TqRf%IsaZ^1;IdSM#?6$`jme-5or@
z?7ZiTyQc>;Jb;^;vy+D-wyvK3X7jD&#1%}i5Dy6HVY3MO6t1aKV#JLl3LOR?(XZ{*
z6ZZLr34;-yN#k|QWN6oU9V3R-AtMzym;rD)k=Ovb0=7)L))N5c57Ul^NII4j>8tVb
zw4zC$<{#?!KDXTWWzgD<<K4`0MIOp=)R8Ra5tGQ)98|WEaPP&l*Kc|qB%zN4rcjuB
z$*}PT!RV)okm;?UC++q+^a?RV?aDA&l^Ts%AkBhD-Qcv*9y@6elH*gq?{U{W|Cm|c
zhfk$cTU}r}S*PMSCfLf_n$B&UloJn7lmQAJ4P8ZGC6Cp3^*EC_So(7AZgyLBw+j?<
zf5g=H6j+_(8Yk^wjk8W-yXKCZOB^F=@kq}>afGTFm#YG7(nF;{75WGr?O|wkKLNRV
zhp@q~Ot%Z8Ur2!H&|bET;Cv+{3!}cd0TW4OfA$b5bguEKdiKW)>Wm&s%RmD{z>a2q
zTTxlc>Rcndk)=p#NW?FA^h%I4272&(42QOLUs_$fTPu>z((PU#B_BlKOAF^Xej-VV
z<m*$=N1Swgb}ncOMdz#TUH+Y4Y`iI|<EkWzY`~OnAf1%C{Efn_5YmkYVNy^HvYVIW
zI0a&x_(9)6@f-kHdM>QHruN8EH;w!qWD7K50re`VaBjKtTH`aYg9aVI<@>>fpuPQ@
zlIt`gV2b#!<g<*RKp2=Du0R=zo8G}H`5$w$@y6=gA`=Bjc;TXZPd$~2OQC$;*FS$Y
z+OeJ>Dqn37g+L{T1WXz#!&4uDy5hY%)0XRU1+(%b(GeB@_2aPqhmVtmll6ZO7WmFE
z{+|GoZ?FIj`I6aooG3Elosc8U+Q8)?ESOBbDHd8$WO5LexS!Ww%2W*XVv2PsU>|hC
z!v$aWtC^XbWxbTeotYT2uf9%9_Fq>9!%+KT6GxXMm@KLDa7U`P4C-ZLv(drh3b)O~
z1JAzFgMIzTQ&07YhS$MhdLG~O)CL<h(@pKB$&wB_ee5Frua72-2y!{6%NT{d+?>5Q
zN)c2<?7}%QZ5-X+lSLba#;B@^t`vB);jAJLld>;o;8uv=pN>hhuBwdGN8M(d>~gIR
z=Y6s&*LCMrZ${S&%VtC#0Y!1A#Drhhe3=N<VED8V%=r!tzc(3{1`9{9EK^#F@AJp7
z6rjZ!plU%gQA;pt>ZGW2P)OawIPrxt=WgFedf1tkDmq=dk^4cmq7}BH0%VLR?2O#I
zo=S001A9f8P&ywkb{OlG{a!VSb}iM-i1^8$%Q;RtbV=3MKc=1knq-qky<$jL^~QLk
z$D)tg>}$tD21eWht=nYI%_Al_x92F^(rC`kR2;^{f}B6gVn^Xiey#ivVfS6k6FsEt
zhUfUvK%!vQfYqPi7Kv7rcu97eEw})CJI0%dL_@|v89hoB%ms)#JR@n^=UPUUetJzB
z(EKCYgBW}LQ{H#2y5y8y|7ejo=1Nry>Tc^Iaut<KRtsRjf{6L`RE%npP#8qps`RU{
z${&D3*~fZBxf9J&1!E@`yqkdZ%Z2_nPt^!bw9=)0wzp9hVwQpH-nz%Kwa=OLRJv%;
zd&K-G9JTT%7yn+9$s4|V)s#n7#4g?4TfLQ_xD<vv-a>_MGdF7WeJ&STs8Srv{wGPL
zM7Ub6r87DdOclh+2ALKrmU$F5V+=35XEzO3ze>gE^1gveZWI?b1}EXEFa8*_86<Dv
zVSs-v9o(5h2t{Ot7eNF#nO3_VAHr$9?T07|htF65JV6v?o=hV=7cp25%Hgl++`9`<
zxC9sQJ?987ju9uB7lHTwMa%GHPh#*QQDe}U-}+3}BMZS+`Hhfl5Q<V$xKT&aSmZ(h
z6eH8LFXlW#knIiisEs~N8*B=aD@*BwUF~?E0>8c3meUw5CIZ<g08|Z&CvR+kqnBe|
zcDG&YyCy7M%CwP%m%2t-fmmPLCuwM^ya^2;Otj<v1})F)05R1=La7DB&{fc!FerK`
zr1ZljEVaxCm!^-)9|eT+>=~k!6_%i`^cZ{7X9rA-ssUeeoe*fzPeFUt5_>|t8Z>_O
zmmhFEP|MQu8vGJtATqnuHMU>NHcslAcoKm*){p(V?UVCHSP)B`yW5k~SVrfn^7ic$
zp;|~zBhkrmIFvc1TuLo+Zc+k1B;i2;B>$51Yzy=R*vp$P0)GL)C01$;{jgctK8-EL
z<hVHI;oeC`;$4ydox$incGu`57yZihzUVR`3^umr%n;~jlk8~2FrV1V?ECm3_Jl92
z(<W;S8X03HK_O>CO{b1}eo@wE8oE@1Q_W=<D#;xINog3{&qlxC%gf2q^Rf&->;o1`
zh8=712)JoL?nUMoHdM<TirXLQtyOgg@$JT5okxcp!@&OOud)elza5D&K+)b}bIc+*
z<V~kYws7JUaxyvRm$yj7Gf*m^99ch<^yGQm)_@8pzwqvGfpTV}*!5uQt?-x%s;XsL
zQCS*jC!2fET{T$egt}x&(DBdqfP0_cLnU>~&a7cD)p7%w{16X>e%>zZ)X@p3$MII8
znL-2#E@^D`4E^h<hXeD9XbWqXsu0PtuLxHiaIR&t>(i$90?og*(}DXu9reOB9m(36
zcDvl!PUzT0&Qm_U3Xc8e;HP=TQGrja?SENwL@lN@EhHyiWzuKgW0+Si7aUhr;=+QP
z5n1z@1h_-o%^a58CuR!n)tE?q)UZy>aVx+Wd_>X80~UDVb>(68t4dx6>t-J+<?eG>
zm7(5;7x1m3#C0=(9>=><2t5rq<$EZmAg-`HY?<Mj?%v#6p549##<F|_hIGO6lz<Iq
zUTov-I$f`U+W|9DeZF>$TWfme9w(*GqzM=lwh}(sQgw6;bw=H#dB+UfI{}=lv_P;D
z!XMU^xlp$VDoP7B%zEl}rv@BMSM@d?*WKZH3=qi64u=f7NbR9n4G!-eh$gnJL?%AZ
z52c<4-XkP$9N2aF)V1zz(L=r;e#m(A(iH!E8LohSI1%sk*?qfTZ;}dJTCSnJXaq*`
zl=d5Uy%cLrA09z3W=Yslp_Ob8I?lWHMLAQ)uU`_KW6c$;-FkzT$vD=qpGSP`4srmB
zpwA6K8>kQG|0;=W|5XyXzscGEOUp@@=Kryre3wKid20xHuuM5*z+T~LuZ_)NvMXz?
z*tj+A94W2&+-cnPo-kTG?Kl9KDYjR)^7*t5O_OPjmigW4YW&Fld&DrRm_5s&xkbV7
z3t}4Oh)lCO>6v!$?BMxOep(+r6MEb+logWfA|qAs>t_7!<L)3DE)y+8Zb~e4-VK|{
z*Vpa#>>~x=@A2vE@bLA{+gn?&+vnEy_31c@-l+%7D`kg*OS@MZmy<f{kic*vk&P~4
zjLmHinOe5^xxMniMU^W;SF`4WTwKMy2-Yb-4K)O{fCPrUklP_}O+?+6mXcjVW><lA
zMI9T6yllE@{Y%$Jqss2cguX}CtUq?Uw#RhSjoKV{$nw(n%T@45N-(dYL)+)g;HCZK
zrxeLm<_JH|A(U%od@yWs0IrTS1(#4$h*N3{6qsD}SUc18B|gl<SV!BUZAZnBW@7bt
zwcYsnCam;(ec9&DWVIf^ZnAQ6z)B=Vgio!|SH~^G97arBn<F<~u)1T>ZP1vSH2}4U
z9m`;rFED)qHPtKk)YHZl(AN<BE5M)5wY~B>su^FSd2X@NOoUWIVh9KMO_e3xP;<zA
zqSx4DNIzzL5tQ{9!r>M$l9|NMF^|B4Ri|0o_?zP>0FP)bz&8!n9mx)K+R$OZDUHzm
z$`X+rbTsQ`n8uu4g<+{yzzow|3I@{-FB}PufSSiX!T;aMK1K<*(d!OVzxl7M5*Brt
z<Zn0>iPJYmm-g<u^w$Z?)g06D73rOsRGppZ79SPUIE2Oo*<E5l=Ri^_!a+<TduPZ&
z2*sxLb-9uDLYJ-JdtUhT3usPIVOQ!@S}u%^`6Ew{V;~K&G_#~tX}4n<F>W+6n+8lS
zJR$chjKkuxSv8qdS(@FytX2WAAIBQ_MwkiBRh-a3Xk4WyKP$AJ*R;G$ceQTXv;4)b
zx}*q|jvW868?2jnhn`Z#;m4e@uO)s$fDT566rG3n8#*0($k=!Te-$Cq&PoKfM3`u2
z+HbgkAEJZxR2c-eXf$KN(JemaJm?v=kLK7qRFrWj;P~<W);nO8)!sdN*wwWc?M9ap
zGWx&Y)Rs2^^R{6TO;=68I~Jk27`Mf>7{86vcBc3Glg^uR&Ck`m;a^zJnvuh2{Adv(
zu?a8Ue~P6N)9s@MK$vUIzw^d^ln3*AcAR^H8z05*vQ|Y_JCT1@RYRLdf(8qrfkI0t
zow(CI*e9Cg4F-#Z<3(YLzbdd>`ijYm<oRLp2|(GhAef(?7rVgyF1c+#QM2(_7Hzob
zFg``=(45OgNwIaMvg*J+mQAZ?6aQ=V`RD6icQ&iNt(qn@jKzy7+6>1ttZ<WJ|Namt
zE@VqVJMimIo{s0QYhj`JKYo}3F`=I^()K{DAUL&@AMMVXUdN8qp+J&<sb$KSo5zm(
zQpxbbK?E|m?n}Se=M*dDQS{O3-KfxH0^k+$EAz08zb|u8xZ|CJrnAv$!Q~a*7Uu@@
zS+<YO{=IihK3JF&yFx~efs#uwyzw@|sti;^=&dyw0<*RqXH9Ng2bV+qo0`z<&xdQF
z^`@h>R5P<b5S=fcPpEu|dod5}4ECM3m1lkIKn7NEQlY>M`&e^uh9tn|HrkyTG#@!q
z9g_OTJoky3<U>N9Ds18)>2kp|W#CUd)I5VHs&N6gtU3nKeNV-t_zzc>iwj-gm1j`!
zj<{(g0r`Ey*5Hf1m5@=_xpL7ys0iOGG32D$wpbUA0?u7UgGaMw|B_A;I*fEpHd@Jb
z7za{j9zdmi$`)=68r7ZWKrj>f2{wX)TJ*9!L!O#jT0Sf1xvY>NbIc@hh_v!av&c-d
zD0AHOcPYC&zuO}mmErAD&dy2+)l>SHYt42rMIhH>-I+hUO^i}Mh7>`o^q$WBuXUaC
z^Lo1X=`Hv*tZ3#I{zGzD(isrRn6@Iu;;ITMYb=b6KUwb~O8kYxG!J25WTZzio4qXS
zhCwpnv@*f1Xyv6iGNm;pGj_rHOYwdCk>kSQ?EFQaAEqWKQt}9<1Ad5Ju7yQ`_mD*i
zu3x}m(xB2)DEH38qj=#nAbv87*EC$yeZ=v+O<TLq4jYRx8Qaf@L8>KaS&aGN_#moS
z`X8~(iK0`J#)r@uRq*|zYk%x^3a|2elya4*&`sNwd6s&y1(xSwSB2b!B9CWH-I363
zwI5&ZTA%~PGUGOPiJdlfrQ|r<E)ih*_AaqAr=oJ_ITwH)K`5lVfYD^^l#=cDLDmQM
zo5(&hB*;J{e3~q95FY8$2XizQhG&I%7C1+^ZSw}yU{p+Gn-y{lEbn^I{kCeb{zcP|
zJK>{bj`T$3`7D<8c1h1LA)y*d4X~;@rPXloAT27DU@2!p{~iuvI&i5<dF<oy!C^>P
zvILPmWw39U_T}E^HJ4_8ChY=()W6Z{jSOV&&`z$h#EQY1tGKW(bYZOU!1tU$ZHbY5
zXt|MlYA)dmUSSC&a}%gU)Wg%?pvx-+YpO!=#1a~-HPpTNqP1w2*Oa7{oXCcz@%R^u
z=YIZ+Bl-4Wo_N(C@FQJu^!(xUUI@#;)nr4FX9T4t9OoQ6N|0_pkV~yNwEwF2?EmGQ
z=j8Y%O#ILD|Bc@s)Y!KBwwU@p>*q=0N8x^9z1D)lW?=nTYW&qHP{{~3sD_64=R(Ud
z60HC0iBMV^%XHPTzQ`+)hA7T$__~7^4*D4sX$xVhBxrsa;#`QFu?(Tu4KyiW2hH$D
z;3?k=I|ZB~JDyWWSSq&NlmscU95peMgmM1ujkn|L*XF^6I;ZE)wroAXlwXeAmVP@^
z&bK&hwWOn?+l=LhW68tU!Q=J)x+KA|)#udO@aS&Y?d<NOO3O4QsXCO4W>>|^m>H1N
zY6#<Eb}}6(V?45=B6Og@KmiM;d)koV(N@0ol^#)n3z}(zl7WZ1T*olpz#mJ|8&_1`
z3YcIsE38?(kW6(m)0Aj97ztTaqhJg)5jC<LmVUyofYv{SXzkJ0(@ECpRI`C<bp6FL
zqbSR}Y9&3oOAccFdg~UYbP0m=3B{&h()siejs4C6?8YfQ(y4I9l7#*pwqYow0Kax-
zl7z~xh$Zs;(8KiFc#eS~426oHLD(^A=zmW(*u@f;fWBvuB%;CPSumb55OAuDcaS9D
zJ6~<|aAIUtCwWq6pfi|FQBY+N#(W<ZtzblMHq*4M`d{GG53_Rdf6#di7uy9&b9kCi
z1|UPge{NWYkc*K~{H&$XU;@RBef)EeJ>R<=zyb<Qk!u*PQ3_rSYa|QZ;Cyomk95TR
z$^nwp2-mGSyYRi^+AeO-|57#Ko2|*dP@9g>nh%41wY!MU-kKO6H^vYVE%hy76{*Hc
zw9QDN`0K-gsoyeaG01B48{VQ|%5`I?<wi7Od23L4bYBS251}SxZJ58e=3MO(VEJTP
zafzpi#Pjj-66bGtsG84L&sSqugvhqtvKNMN7Nv>)%>T+rnoUFHn@A0FElAFfzuK&V
zH>(G2_#;i=Q0u)B&vUt^Lsss}MrbFYOHTQl<7;leiy|Cm8*&2*KLv00sh7edq<N)F
z?A}9(Vw|C#IeGF9x?#NI2h2T+aDWqB6<O#gbJTSKnMWfzPbcGeGBC<EHLfTvq|Jw|
zXfb>VkIsI*x24eC{=Qj)m`fp)^jYYHufcPtA0{MO=Hq8^9_;Fkr>3UXT8c^BEe{i6
z1(_0UAVP#Na=bk4?n7n1OY)F-=hZ<D7`;0@Tz0UbG9$CLSqP)$xcJptZ2p-pNujG*
z_5pn@^3=@$;2>UQuO8U;xuuojx2f`T&dNU0?AL;3l_kt1WI{*CSP8=V=mc+a-p+O_
zX-H>wm1lN$DERMWJ2NAMu9+3~9A?whZkaoytPNX);9t;DrQ!2j^fJUCuL@5^CTg%P
zroUymz*baJ>=^RWzAg`F^AJ5&WS005NIr`D9uY5<h~Wl0T_H<==q0|7@ed_RB-qLb
zV};IA876FQYD+mR1)dtM^WQn68i?5AUeEP`*Fr1nCmfm-gKp0bMCKm+waI6-?NiYN
z6RC{o-X(TeKvhq30{VidR4*J(M8=^Re{Diz5}XQ|L*H%MfCUv!9E<fLjdS<aW>U!X
zE#9XFRJJBc%@wGPEH5R_%kS-|a4<WT=SMy4MV}qpq)CL*J3X8f620O0OCA!jJcR^X
zm`w(fv@GgC+I%sNfMgT#E2B%vEo-11a^B}9*NQMx4S=jEWP_-qgS7W?YA>qX%$GB#
z<#jXKwbopqSTVb!<#5w*MS5U&1h{U_|0r)ykr!k6qYe6dvF1qA1iddn7x!lj>7z;^
zFq=ylYHkBLxr1|wCEQ$ktEB-UcSfW}jTIrb)Ji+vfPVr6_XoY%%zm1%L#oY8sKSx*
zuBQ9e1tdEL>}LTeAg^5%`{6I?+WJ2!DKD$+2uu0ly4)u$e&j`oBot`H&B_v|<&{b`
zY1iu=*iEElJLR*BZ-qTW;Ll8mFWuzx4agRxmsUvb9;62|&}~l?*s(jF2qJoj<b#D`
z`oozBJ{Z~;=Vy;ytYMNZR?MnL#VDDJi*o@HhsC{e<eRIBPqVQ<D?v?5X+F5M2L@r`
zxg<Yfmh+8k=(vDlRTd&K&zdh7qr(XOK!=)*pemQR<NN=`*f~XK-gR3(wr$(CZQEAG
z729@F(G%OO*tRRSZB(4j*XNwkJx0G{^!r@>FZS(T>)CUyIe(ny{%gnKjonI)sy-xZ
z=7cM!JyNgUv!A4ySX^9-BE0Dp?+g%W`VF4-%lIA`i*mt`Og1+_5(J2KHupf?Y#uu)
zgp<?6T9rPqjRd8=v=Dz)9puG9&zG6mprBm#gCa%21BE3XKACWK9z=9hLR$Q>bGOH=
zf<$)Kl6kO23>%U&F2x@w+*v5_tqGS<Y9{2GoLfbkWP6{qA1vKL#5wfoFJf7A{22v3
zp$Q3Po((Wk0t49O+Zul$zDIt*9)4LoZNMj15UZ6w58%fiB)GTqkUWuO@}bV(m1}z)
z*LX7Hzk{}3=9T?p*ZhZx3p3ZhPoy3C?^p7h=82RdIhaGj!YOgwC0lmff>Vm>{H+oS
zUE`rBId$puH&(HONC>8gT2>)*Od1LL!mnfHqYid&%qxhpR&n}>l97QMAV!s>rf}4P
zN<sM_^mY2AW-=l$`SZiy7e+<~=(tX6WKk9{<f$x8!Jc<JKPO+FLJ&wYu~JrIr<>2S
zi_5pl`{<u<@9$F`n#-8?(O28=uT!gUFMt-!>zE+a3!*@ZS^OZ1vDgyrLF8dumO_}&
z#NRlEIV|WKBhomGDt6==(%mkiQBbaV(2f7|Oh%fI&=OvDBOw0+fQ(Z6p<A=e8BNC6
zK&IeuBA-z^LMDk4TBUoB46Ap>Rjpu&vQM<GaM|nK+$L*;QQzQ&8=#D)oDf1eXbXXe
zj)KRR7`U^{vuv)%X)#a4Avj1BbAGVJZ(G0iT~ab!?ghyov+;{d$*5B6Dn>`Dh!yzJ
zkv(J{n9bHf^U4dD01ui&FXI(kDWzm^4=6}vi@+Z-PAk%DX<#|L2ke2!zk<yLbsn|f
zjrdg$PJ<H{s?>6NehDFwO=X5gm)aG>dl7(3+vol<&(drW&uC)QaKc~LoL0fB+O%lH
z_P5*bR5%mUe0n6IPxaS0k$$TeNq=I#boI9cWnYT!E2wbOX%bI98{aLv-s^M~8|9Of
ziXthf0$QU6Pop96O~Dg2ph;&;G*Tdr9aeUO@i4VBgua0wC26IT365ERKI)UvoM2gH
z>s#ZavY%pv-J?iZT9-R+(lz#-0rfxyZ{JF_m;R`mTl^D8TfN@bFcp9r3`0sxap*N4
zTWgpr_pMM~8QUmR<TfSycT9CObzFjY6Frc*DU>C-5`czL>{Jqru`jJ<UngfPDnCfo
zz(T%5Z%9Bzh2#;ODu-eHL=ol@_2mQO(%|7m6V8Vwf@@VXvuj&RgwqrlbHuJns$fVS
zTDnB)<OEB{`0>W0@I(J->r7s6I3p~4tm;^>G!1xHcF*Uzjkug&4LUD!g*%t}+jHpC
zvt$kMCMkoxhc$LAy$r?&tLzn!-?*+11V-V%3k|I{EP1-_z0;^}#Od`u9VWuX*&!$s
z9+azo%R**prJw)(oN{oUuj|esJ)r-~mv9}@-cT*!l#8&+R<z;x0E1KwTxJe`w!^JC
z6A7vz0g07gtN^-*V%865oh=VodJP+5DjmxN(f3*ie_$9{y)!!{7A8nV;u-S3o9nC@
zq(D)5X7ZG2B!j6b-n(O-@H=wdEelbyA4@?PY5%wQ95=(pIoC)F-23IC9@WNFD?vS}
zn3lw4$SD<)m>Y@x-yJB?TU)&0b0rRT@XSQvfSF$Tu(hcwLWx-su0ErqsZ1U>+V<pf
zzXOe3T~XvAee}%hpOqeAy0sDk5>m`dKWUNJj)metAdtB>s~68%C4nB6`JM1V{H1s>
zmQ4)x;@69jjyDaym~-%G;2@bfp&+E6lXp!V`4G~C{gCHpYO&!9ZzmByvYxuw88V0o
z#=iR0``rf#NqcuX6SLty{#s-u21u8AV35YLlJy$z0!sw7YvOHS-)>}k>LA+07uO&e
zArULd2`NJepF=tLrLi4LbMbA8XzC8+(X*_#bK9b{VyihdOYe?m@?clfb>z`kuC=^4
za;I!ok~Hr70O&6g7m->Yuy9zx)B)fZ<2M==aCu5jq0;950523jLYFRH6T8JDidX-;
z{%KEFztbg3qa+UTvL=mY$6a|l8yz~H+CEZuoO7GPd`KOU=MITMUK1;@aI&Apm>#!o
zfUAf)?A|tRP{Fg6kG}w+jUOORMsv0L_%@j}6^0*=GLnK^1qly7=I;H<El1nKSSrhV
zC2=yeX@f|Y8qDl*cl<|i9F)jcbEdsDsrjxtz3KALJ-o`s!UAr+Y+>l(FqYOG?uuPg
zE@`l0gVxw_+FuCQV8-F(7%PKJkQtNM$_Xz9eoa?iDK0n*E*k#~lq=pP^|1Xb-@a+q
zGR$$@>3=~i=Jcm32iXR8JFF*;iLjkfq2FLn@J!4oZtuXe8j6>qs@G)lt%d3OY$a56
z0%);**X&AdSR_&eSo$e^ULJ-3e-qm(1?~LSCu#X=spRL^4UgsVgXy)0$RedSRl3{d
z{rhxNS+iC+#Ge_5tqF&#8Z|krOj&f?$RkgTb%LJL^jsK`lARDWU#J)06$<BKyd4A3
z{!UQ=+4xpd6EzWRmITl%z5{k8M5Nb=_9wxyYgg@4;%XFCCKSoap{_%Z69Z1BHqzh5
ztbZio)dRS-^;}}M#ryF33rZWKnWE5;KiIfZ4A2PM7mExt(AK-Ok%ifB@f$!CoFO(>
z{!YRW_RneWGQMV3-5m0f(!QIAH;}N9Z?`DId|4T;cCFf89D-A6fgBDH@4R+-G|9Lv
z!j(-`_*#QOn7(wXMnyS1IQ16I<{*L%ZnQzS9mDovlF1zC^Qer<zls||ojhZLIJ@MD
zV1m4@jnoM$=+3eQeTqM5x*9^QF*PkebOo>#u3E)9F$=yR3gQ<2<GRFmEZqO=aAact
z*GS|cZNu+vbrk>W(%Hg>YVA*SX}l+5@dcuAHvXbM{*5c*1iRI_8oD<!?sYnZzpD;q
z`E+9?s4eS51w!&F0~y<JmSL=Lv8l+h42E4D-j7e9g&Ylo6nv3a7>QCO<ftOWQCCNE
zip;EHNsW<gY~5|Ge1Z!s-5syTLkF9SQ>F2m$IGY!ZJ#cl9LA}`X_%atEyv56U+o_8
zu*;g9n8z(#pU<ZsL+l9tB~b_&yEY@pmMxfRmT;e6?=NQy@U}*lU)^tS>xe3FbuyLG
zNmfe5q2t=J4^dv1Bob7zMpC3pXW;Ej)`cDDNXswnd;rJtz14Qx3Ra6f*FY95)I?z7
zNJo)$OCxOKnD`7^f;*t9d@Tv~-Kf}>4|wp}c}l#hiT?PKjJIF6&C^G%N!}^)zfvDf
z^sCzu?|(}ta?yXxv(Pru*zrVt*)BAU0r)EMr68jbV}#L}Z`MIFG0s<O**z6};m;^-
zI_W{dgdcgiQ&A+QJQi$pIRg|a%jn)QwDtGPJf$Xlt0twJy(U9+$<D1}fWDJM<m(Lx
zZW_X<njj;tRh?+0NS<1S;h+G6giGim^0$^bi;XnfY$x({bq3GZJ1RT|_L@6G+NbQp
z$w9(z%s-mdx#vMcq-i7e@IpBh0abpWx@fWIbmtI_8hGtI@1Va5Y_KnfTXtRrn~LIv
z+H3B%b>jnGOmwID)JCWbQ>l#{yW<7s!{b!3x`0uH&*A5@FGEoJ#Myuyj#5omvU8#F
z1xq{=6`~PL45r@`P%R{?w*BfHkz%xpTTM^7orObo2C`WDMI)O!-Niv`sL<baf>Yba
zQmvB(Et0JbYBW7{gbbV!W=(oqsl0g6JTr>X<227bHFjnXi|1K!vSlfitqMW}0SZX=
z)eYw0_HfPoNtMHMl-#6xPzMif((9<qPC9>naLzk*ZQ8<Mwzb9y8ZD7$uEfNtj6^Td
zdyn(SJ_?yG{IK2IzB>YzH0^>e_6kT1f@Jc}Yhk<Ps|C>H+Vi7n-L8%fTlr}LulDSf
zfqp=ga*sjHm+W8rEzyuGQxmNT+hV$NA^gMzqmeEX={(tzjcI)f*>Ekf_W7F-><Mmu
z3$gHJDlqdd6D3&5el@J!?3kQOec0FuZ{3n!%OTzzWF6HDUTlZpxb4+z|Ac2m%AC_5
zVBFJ?j<=pDnig<ea>zlyvAB!jUwKzD#s_Y1@H%_td4OMArL#C%41S(bk)OCZ!h8g5
z6aiJ~dvW;!BnJ8#09%9B?Q^?tH+Qh}r*TPT|CxY1yboNl%6Q-QKHi33zSJ|3Q!zW@
z=?TlCW#9zhl0}BhCaj<tEimwyk}7515w#~PX}m*1+0S$&1(%=T)?NTA9dblU5k`SO
zWXA|GA8rACXOy74n`2BMzZyM1h;{ym=>`db7aywSMDl<zO_S#%QWOZsi}X~o&stFe
zwIPR}`w(#ink1Ol24mG7J|yIm1*Gk?C=DP8q|?6ZLOf6Zb!e@zC2W8fXdCkiG3vF$
z4VhpFN&EvVhPWm151*Mpl0E{0SR`0!BX5kNFgu&xsv$B3C-DR*IA<xBY;TXgJaN8@
z%MiE&KIWm>gu6iH?fK)~-K(iX53tu!RPSezo;4C17*i_aYM^l7zO)}SvvNSv<zZH0
zSd?RXu(R{cRXx%XIY!h(m(iLjb3}8k<5i+Q>Pf*vZw1`@<jV6%99LvIk%rA?35T$m
zE*?rn=r&2~n(;Zlc|ib6CRTxd{u)MNeh7`xAJ}7-gP+h;b8`p86gzJ!J(J>&TgbnJ
zf9Q+`8)1_^<XeW>K_1nOwX_`|;XdOs<@BQG;FWPBQTC;CoR#}Qmmn3pHIH`N_SdHf
zHIa(84l!?7W3?O@Z+OMv!mI&bE$@17S@}86T<eZynH1Om-eu?4P^7%rslfr(BujKd
z!_7Av`3%8=8$YejQmYbvDXSYtaXf_y)7>US3ujLmw;RIqvWG&qQ92=4QvX6c<=!7S
zh6b&EWteO@^wq52y7#scx=I_Od9DC{1G<_e@97j~wyyV-`tZnP-*miBAgo9GT{8Ss
z%<Nukxz>ZD_lvcvPxBcf&upXUa-8M<=XUXU4J@`s3rmjWcXHtG%{5jBwySR06M~RB
z<*VpRizY`e!Cn2+9EpuB=-`K8AA}Fav(KrLRo0)+(tXI}Yd(jCLHtAHyYq$ftWTdS
z?CZWAO1Wp@>ZusSga=Bm<(OGF-9SGfXB7GmYb3^)-D~#X&hfW%MV4jZ6^lY$9691w
zZTA$>U)kgzpVe@eHLtUcXHGH}@E)oXDmGZ-Kqh3it+x4F7XnL=`EekE!}yH{!a<+i
zPA<o4(gtcCxFsPAm76KQPCZBUphZG)uC+nhox{CUs_MCY6`GNt(TaU0OPd7NuA2mj
z-(P@dgq7Cpak2;4-u=eOPntU*I}~^r2^%Y(K6}FISqWXXaJ4pzm*jPdE8#NV@<1|C
zntv{kDlR*+L%sQQI=4!rEN8<HoE3W<l7La)WO~qzlKn>g2>sAgfjz4MYJ9d=#u!cS
z<{QFXzD6^Euz{HVY&gM*5@c=-)BO~(g<tiVnRTum-88g7k)dizJQ5QL5>v*qwZ*c%
zkgBW4460`R4#8`jNKAJ;fv<P6UMq2G{ShbGzIt$o0zA2M6|`Q}r_GalyoGPHBya=1
zp9b#(*G>l>N^oD?A1z!Jly)_=S@{Kud~qTAPm;j5Wd8sBtKj}#F#E6H|6fS|_Q0vG
zGktsD-eCxnQ=zOMjn79Vtxkb?j1*GFsK~%(xZH-NsV5&s-0irHlqrkIa+vHClgux>
zD}NEx+MUJno%ybDL%Bc2f&XQjgGs^M^3wBEp%Mvw=nr&X1cA6bznR!#E^KF!|5*Uy
zqCCm{-ucCTk<-FUWE~A&w3!&y%%F%k&7c_HF6s7ua<F&PiR+eI(|?~FJaKby@j!yP
zl5Z|mHrR@pm$cWWG9u&JOBtx4!GxQ+A`!Z)egs3LGQd}5#&z~sNQN&-aV5;CAlaqN
zcQurJYu835838RVG~V{eNn@QdxrQ%peLlw_XvfOYNat1e5w5?!h3H^mZSh8tvyri>
zH*j7Dw2vX+7U1}Y)<=`_Z#b*!D^;rc1BE;C87po{qhQRePv9R@f;2JfBWSq<u$z8n
zFNEu`bfZaunk6?TSPP6z%z-NQneZ;ou{AoYV&8@MJ*y`x$Po*krmo)tip6|J-BjNB
zU#)y*30X=E`T%5!G(Pdo$OT+DF8NgJpEXhePjR^<mmL0t6I1n5kR9k*^Zldtp>?{6
zr9BiDNUhB;%1mRuIcrpHgcb|=T~hVAr4C5)lw6sTdNGP}7Q+MBTUELNiqHrtaZZ5q
zyJE+voEhR`T(Aa5QW&ZL<pQhj)BqrYAd@pcz+1I9>QD(`ny^CQLCFOOumL6tIiANU
zRO<bZFUZUs<KFojNiW&(q<CRKBn6PEhbH7YXjMfIjgdOQ6{zxcmCOijvqD8qS){%l
z>TEyRAVnvYTf;n#-YBlCe}j)RFLJhjv0Sna2R|Z!iENM=cO^DkuE5Mpv{gF!bn@Vs
zyN4xMSd9Rtv%hzEK_@}`u<WIK1XRA2A=k2YkIrx)gMI`X4LPY+odG{nJ~d;c@`pDC
zMqYNYpqDxiVTF}Wp09+@E67RG^L7r{kAbOMbIE6)ACA1QwEdSWRlqI#$1^)e{>qOb
z^6E31j;I220=9Kt-HDvC83;_^<HZ-)`NJNdgb^5CKRUTzY446!s`>!?rMo6kx4TY>
zv6!lomX9W9>FjE9z=Du%NMA<GL-o^1Ca-t=UiT@iv8A8~OH}R;nnfG++QH5vGhYt#
zB|XyHPU_V!>kD{JM3KY$_3V2Uuazj8P1@nD6}e&OG!i0aJDdiZ_}Po+6{br{`iokX
zcY^f*>dKqXahuKS(zD9jhP79Y&fHJGnRWZtd)Rv1JEwoV?f*orW##zqo$LSqZD0KF
zWY%-dI?gs`<4$NiFDnYy5pFj7svNxd@Fc4Y{3I3e;d;yeB(owBiF3^kn0DmkK&eD;
zB6hr8GK3V0hZGu0UQ-q(s|L9iC8nxLBm*MGq#PreFz3t&^y4B$K=6k|_HJLkA8pZ!
zJ@`Z4fhFOM(PPJrfBogAcP17O_Q$9#McMU;=*YCJo0R%vW}$cokcHFxVI;{+K&0hK
zfK#EC=Ze5y#_S+0dM)bpR>XhN&r|gL5KO*ht4%e}`G)o^zDAVYI%g`}lQ<AujIdPy
z=+N2H3J9r_0u~vA6WN<o4Nl=}_(I~E2Vy~oGp2awhZWL%EZaBX2JM}&NwXU!Q%QOC
znm2I;5DcO0-}1h-w!{<0R!~`BwE#>+kU^WOgG;rOA(&JVEYE`yG=AcCbhb|{xmp;e
z)^S_`jvryVaRx8{s<$glR0R@>LP{|OAu8OSaUj0P21Q(%C8&aWE>`SHkd=UuMowv?
z6=dqn<;H8Vbj#k8WMd$hWoxpw6h)LFk;Pz{D<IcLEn%BU(aJ#MM>t1?l;E?BR-H5C
z)||2rUU?~}EDtTKV=LeRrxQQhzhTljwi}cprfG7>c`mLD1^d3+5t^@6{lpx5Gkv=N
z0<oP!DvpGuaR+A2*b%#K&~DUQZ~6`l6zvEi0zzxxb;4i%Pe|RZh2H%I;V8Sq3~TGr
zsVC0zvqFt=QIpFsnFQ82jim*yQ#0MiXo8;^?8hj<t7~^Yj?<JcuQ3P0oHD{8#OR(K
zV7<3S!?URl$zr0yUL|?zh(3GfXByfKMDIAys5OgJ6ImqJE=6i}r0-zKUdEsWbOE5@
zckIf#2mF{cxIeT3Z-b5nZB*-fZTb{S5EO;Ou<gue;N;U2U+`*9PMX=%S7mfk+F5}M
z;wqF%41aBOH_Q48ZDZ^i7&Mi$!w|>!VcYxf!*)42npcHt!qyy225s?PZV1<fTg{qh
zg=#WI$&~E#pUtk$Fl{nW^CGjrW8ch^`d#*!jspE*Fj26XG*5l^L%2@ZcYxnzQc>EI
zoL%Qoq6Sz&4w1Z`K93<{vcZEiVkMw<(KgvHXMJA*RT56N^0aC{g0ei{*;T#x1WXN#
zJc&BEPs(TulfXtCtJSScbocY>wM{ii3|d8|1qI0i3RN_Ti|HC9k6gwG8L=qtQW&s4
z$^Br_wgu+D^IagXYO&OBu#oEez`d`TMOs{9ljOk#-PFV*o!B1ug!Ld#RPz1~0D<rC
zKy%aW_uUg^-fWH_8)>?d_=k~91lVqeFNu=&O(VepGt%;ZQvdm5tQ8=SRQOpN$5H5?
z5%Y=j7}|FOLPu4G$OJMuK6i*7cXYEZIY3ff;81M_J*UJKsE({3<LB;Q<-x#IK{l8|
zRc0OeptlQrk9R9NiVRXN2IM^SOllc1ie|po5bq3~Ln&s<X#s=9d2v?U@TNZ7#1vsP
zE*kjMe<J=0i<U48jK&s7DS7986QJXfJlE|ErY)~rATKB)ttlY@_an2qSbojy+tI)~
z%PR7#tYPXZcQ)OuzdPfWkHXVy_lhKx;I&0+;*@XG`AI7vJmpEUsnKT1jSzM#HS_mJ
z!{|l_zJiz)5c4B{<=N+<?$$<Ff{N7<33B|jnIwO3#~np;b}5mPKr2Sah%N+ug7wqy
zTMWV^#5~NOhS%UOHa)nD-i<`7{8Zk{;aj@8M<&p&nq&iP0n)9_H4pedCKn-}xipjj
z0kT^c=HZLR0v^5s>39GM_yy}Xm-$UJ=m~bF5o&v8a8MZVgd55&4J)P`rK&IR3Z(SB
ze<;!%9RK%}M$OaFoQP4;#75P{4u(;Ih?$A$KV{narwb7`C(Hlg3UYG)YawV+TiF4R
z6T$Cq?X2m&BTga-bvYAE5&b-dRUnCUU<ZVsW41VB>yqxukGzk+A5^udGgn^jd|L-s
z1Q&-^0@7$Ni3MzC7R2Bzw8WTS)%j{)Q-Ni|RNQq5u+(Pf^k{0GigQNbA6(%gd+-=h
zXV9KMtdHo~{|MOcD!uih^Q*eLn9J>cRT*hrHNHeo>WjD=6XMI8bb&(5lVSCkkfA?v
z7%kRo<k#2FEHYVG*ZdqXi`84E2ux-PQ-#`^z#$v|C>Eya0tL4)k(<L=%AV6eS^MKk
z6G_}>lrh(*a!ZjuQY4$6Pis*P!4QxK-+JVgswS$Cq;Hd8!xLCUj4B!^a&Yd_L>-=G
zym=uzGXxoC-KgXWjW4E;8ibPRrD?Q?uLd+W)Cq5TRDrIFBvC$v9*|5{%%F@kUIXJ2
z64}LAMUu3!h8|=jL8n#Z<^x^e%l)Y`<$BkW>H}dRwZ~&DRgMo?D5L5GttVaGVy!fd
zg~j;jHg79z43vhI)GnjdUg4pcE-Jqh1=47JEY$}~ziHAY%t9z`GrRP5ghR(ZLJi*0
zL>c3-TCTj~qxwXU8I)mCuN4)DoeuIbr5-}2V6%*6&?Li94t|qQ3m%60xQj$Z1PbAl
z|5$=y1Qy;0OO8p1K!ZViPmPw;ev71gxMUXAJilrRSN_x;k-bJK`ui6Ii}%XnAM6OC
zs`6G~qdu8~iQ8F{!{mj9<lXZdh<aM6(=_Af;~qZ*d$$u;&---;OG-nZTOg>o9USOx
zE-18LTCSkaCZ26DrKQFmyJ(6n_y$O2SN<)EQ1o1;A7Olz8@ou3ebk+PkSRoQ@6ZE?
zf(G=A2J{-c@>qWFui{t+jGua;e_4Az&*Q-jvcB56u)BjWq9+!cEK_P8#%`RQx;oq1
zwe+@2^0uA^_AWc_8Poj3IBNp9{zhdR7u?f6KqxI|=NcL?Fs}jD4~}0S4vzbW-|;<J
zeA+s_I$F;LN*Z_9cRO}z`FbTSYsSErdRGhj&!H|)5>v`0CgbYtpXqzM+F#5=nZvKQ
z^*=2={JhpIv2qNnw&)qdH3t+Kd+vK?KD%^2oIYNd-u-Ul&lg|J^$b3EHuM;4YQkG4
zq5<3W3HMw3*bf`UXI`1J$E(%ckKw#izet0_<*1!!8t}l&v8{PCpFwinK{{auI471T
zfUbPpx=wT?qX%dDA;o#B>uT{RPNBfA(Bncp1feFeEyUCD!B~(}S%IZCOnyBfU5=;`
zg7=A4IDkQgB|HI#53XMgj}kf4K7D4p>~HcDYcC@zO6jTRV95M(3sLdbquS)Om4Cw*
zBv12scvp7yo?{m@o>NtE-W>fMgx81aNH@pX8!V*ALazv9;vFabypZTi*d>VM)y%28
z^<+^Pxn*nqgH!9=%Png%%9)LVj5fuoKA%7$6pXVE5kuFTsB!DvIZaR6G}e>0iz7#<
z6?ycvBxKwXBtZ$3s09xU1xr&2+4nb)KV0I|;h1tWJ<t+q$!yX*!sp`1t%KPi$55M{
zNni``q;{cPuT!P6BN;E07)wUI!!$K!{BVG`0^zN(0O&Xd==(Eg8n9m--)I0c)SH;N
z$GNKM?e0m(b=qne$rcA%@k<y<rcoJk8R4l(0S_=06nYjrB1pN-A0F^@M*%C(;b~rk
zV1D5HuLP#@OCw=V5ZE(ho9Rf5oGz;^fyUR%@eP`xYtPq=#ahoets1i?D?fSlA8aRS
zfMx2V7TGzLYkuf?iSj!&c@vqvGy~=Vd3n=&0PRVr>0{x;3E3_*<vg?1Oiep9nqVRZ
zKip)o5j$!?mhL1;c%T`y_E{gvg`WVHWy-1Ktdl*~&=?gbo}krTT;3RJaLK`C-q?9G
zK~Ec>#jMl=uSPT8*z1&|*}%^^DgDv`b^&}w$5sxxbN3w#Z5$D;LPIN+^2DY~czik|
zUYEO+*)iPrp`-F@JcuBrL~8%e82+y_^T9XLlt>ZYl?{TCqqo!)wmwGE1WTH@@)cvF
zB|OiCKvtVyiJ^ZNvER;;tw%y%;#QRtq&!73%g@;g-2o3eZ=czF>5<_oRb)&EB3pv2
zo0m(WrqU1PtTSE{81-60+2hJta8Xmthl)86+>+K&^qNe*z77CA<q*OeqUYydF-}Dk
z`M*yLGxcg!CJG-~;s5ZEUz`uqn?z5d2E@EZQ2?XDpnW(LUpnD<Qbo=vK(+4+)$#}u
zk1GFg!KzLhWD<UG7@}+P@0t}W6Seq*yV=8YI4^9sMhTEN;vk~}03A@2Wsidqr6zK6
zHjHeW%yijrNRGTM%Sx{1+Dc;E^7S~@z4v@2mS}C(TY2{vOn~{t3oX&U>XZV7npg<H
z6$N40xq>j|GaPxR$@qhsPjNN>L&xRh{D;vBGsnMo4@_yv#{U<vqKp$e0=rv0|ActV
zex5!`zWqFFJ}-fORhd>cj?AG!cy4<uKg{cUmrx2d??yO<>%b?=!GP_tnc&f2GKax3
zK38@(l<be5_e3n^@6sX)H5vzOZwH_0gZIn3w2>KXcu13cqP1R(*vZD++)f>Xtw~Hh
z<y>qQLn`@?<z+>~?(Xi^&faTG&6XK^hP1TJ^}D09)APf{iF4gbBME%6*})J?p=1n=
z1rT+54<WfSW#bGn?1H@*M6))sl%}#LaUFH>Pf53xYmO_p!fbXtNwDhYYqdGH+R7IZ
z)wAR0XIw&!T)=3-CQY1B`fe)ezA}nVp;El32r@p4qOv&Lic1?P`h+`lmO(GY&=NxW
z7Yy^>vM;ygFr&sC&D8uWf#6j^MZa==AG?AVV2j*dOTsniX9PmeaO<XD?vd_$1H)ZK
z+#Sw#0hhv$mBE|=Q+nG=X{OdaJmO{Pw#uyAUUHoMu`mo7Fw8BGlOtkDRDq1>-k#`r
zIvsQ9JH6tUHF)k_o2al|i3{3i{gLF0iiElyKtl&pkd$K|_`VoZBra*GV`q+xhc=;G
zIGs(^dgq9>dfMrkQfc+zQ9|uoOPKB=P-A?m?xItrj;&>DUVx=pKckayj~Mt@rV{T!
zXsTGb!(=Eig;g7pq@Ixv{Dscfww9OCAk!y5359O%{;9Xwpi-=ALBO2}7-5OnknVUo
zD~kj5zUVksQQfA1tedof<RSlbytWx7-5qlD26ZnVOzr5NG6EV1Un=(CvyihQ*UR+*
z4`W+~3C6gwqns^t#-|<)o2X7tNuG5)u2Eez9AgrnQ)=RH{LS1?a#1d?jOmPPGm)?F
z6QcEg_%>h2*sJ1ypPNQmId!<(M1Ku($^7(LGEdP26m0B@os1}NDlG&cK$L+BgSqNS
zwEfcP-{yIqs(jcc%wrJYmZQlsFdurYh2g=Ldhk^bjoi$ya2kV|PTgUWrD>`0!#t!U
z3Xm!&#wOj(ZUZd<SxN`E!tUU5VIZ<Ww+(^^<q((g(mvqmjMn40ha8JX0+EX0Q7*R2
z{k2nr!bO*4?rnlMDCM)sjJ)iVAb729Br5{pP5u);f7t)(m#Y?^)k9gmPl*+{a@vOo
z6QpV1W$$xua@1ulb}jUqYC%z&?^kf83qUfjN?9kCXXNPjHVwy=vMLNLHfg?!4^j~x
z9KGIc;3hkFf1^<f!D5!No0p4P^^-F8C5lzsyrS}tYUu9608$z#cr0fZ*e?7A*gk!t
z>YCC{S<?sGQ(i%VxkWh>`xLu7bkPfxASB(f`~Xl!=928k@Ce@cq*;Dtfv}Ck`Cd~a
z)q!+<5WdIkbwyMSpj9Pz^C^*Z6}UN|ms<4KBxd-KOujHpV<jPY^sYH=DQTd~7+zqq
z+&s=>J|y``W%wrgYR6_@5sY;_7CeW{h!EgjLo@3fu`AyMQBX`h-Qz2Cq{C(#s4z;h
ztaf~I1X!7o^4LNKZ1@|<FsV}zUIv_zg!!51HFy-c@?Pga#mWA0MFElNbxdxuz(0oy
z&rumbgHMj%fKZYB9SmN*`4;zhEYwI;96WW7x`DpOV=8>+>PWi>x)Toan0-66D*g!+
zgqdORM|JK_^YW`y_DK#@e#?&JkcTt-0!;w*GOV!rxkq+1ppoH9CJ^e`3K3N1qzn{`
znmOc5VWO~R#*SCBh~VdLFV-wW_-6A9wyn%*__ylQi;*N0K>#Rr?i0E2Z@XHk`x#*P
z21P&)G6Z$~w?1~_Afq?+?p{)bj7cB#;TX*vWR3Y}|0)*z?+{QrI5@lhvG+l2<2s4t
zybs)B>l%x+iPbl4&4b-D5p=4$;k_*{PVf6vFXqYB0Cc2!LYy&tB$G1N)X9$QkFc-R
z7<!`?6C+S6&IHroL~@DVDtw;Ym;7CFqInYbetoOt__wiz#sI5ATV``L{)hwvrrd_-
zh~!f$U|VNL{Ct;o45qnTXN)e2luBwvlw-crD%4rW1MsDY)|*id+a#b-BG7tBb0#S4
zGV?q)Vy?tt`Yx`vYLTF7K*!u4a-)<k9!%>Rpf9;iMq5Er1^c)mP)01}W`FmmD}vrl
z@Ye-pQMX^U^|p`r7MGdRMFLmJ3ZGPsvFb^ho3cL_(}Bi`h1BvW;zIpVbx!u9uZ|J=
z(d=;SU%45GkG{aOGW0qBsfTg>6Lg7-mE+$E`Au574uukE|4F2p(^y_;V}dI?8_)p6
zsuYjWjHz}kMhy<;53PV_$yZeLbZt*!!>zR5K9LAMJ91@l|DCuoLcZC#IOteUuHA;s
z0fUw$!3J+Z42%N7mxu}HY0ZWdloy+X15Ed)D+=gkZ*S=7Pczg^m#`+@jIEv?o<6Qe
z-xDfsBk%3K-#uKi_kG76^L7#4kS=fd%SD_oG@dU20QZ%$ohS>r4n1G5zGV<9RH;y-
zCuRvEx|yg~Lt&Fhj1aD9I%gSCA{4)1EavODB2Ow?RLv<AtqqFY#W}L2rzl+ww!D>C
z1a?+$hugA=GN}n5Sha*fuAjHqtflz~I1YW2YsWL$kVF!d#-Y(0$S|JSz+f3weiehy
z$q;-0W-;@ry!86D{8;g>R)vup^>}9-;|YhDQTEZl#>oiq;)&+UO?mhW`LRm+5&c_B
zumloL05yp9mlaSFH8RTnx}*dBs7Nm)N*uw`qS|2cqP|xersuuLsJYHEST(`PwkqRY
zY1M|YW^t@wAM~+0K2O>J2QP!iRFg;?Hav{XA@-R6S;7=>JXuiaYtKW3kzNo!qF5eV
zbe<Pfxf&EIo$fLf=~(J`#56lb$1zSc*|OGX`|xMlL%F+reD*_4y~7wg0ykyzxvPO!
zgLjA^N<lpiSOGd0jL0BzT3Cqu@qK3>_CYR1W$GiaraJ-8{Gv+Fk<HAB*W?Tprp4qG
z`X5|bUIfYX?&PmPR#MH)kh?g(iSfx&;WEndk7zGzhz(Vp!yKYapJ{-kh$;$jSu<lc
z)HYJ5oZ-o11s$Ku@fe3I2c?_=N=f;AgP)eVY^P%!-gqT<bm-qhK07_8o!`z1z5SIt
zg#PP91UuJNCC42mu3tZ5DL~d;`@o<%UH7S3UKtzK{1^BNMz3T@P%&h=(Ijt;4?=^)
zp(=o5oBl8`+NDP<CMaB;#{r4jB9wKKTx<T~41S9$T>DE`Vp7s}hcR*XMq{|H0Q>`1
zJ{0+jx`8@_j$;DuMHK6Y3T!LiM5g?qptZr0_6J0rS~AZBl(uEf$Fo2f;Q~(?MjuAE
z057zkbCDvEzWU2`#_@JD!wi*S3N=FGp4np_BTuNLv-*QgGV=|*B}Y?Me{JfqM`<>x
zWT;W(QbSs&3Ug@A7C<D-c?#@!J&bL=84YH*ztRREZ(vc!#$LL}M+@a5D%}H6Uo8-M
z@1v3=uNPyUtyeS=L>*+_>o7o8b67g>Cl8E?m9>o)+jZdp-nb9+ZMZ$)nuQeQ57<5f
zZe0Y;V}uIOv3o+*lAsDv2K0Jvs=aK6#c4QllQ%s0t^#|$(02U@6}uoVqvW2zN-P@q
z8{#6vmHM#pNXDlm$9OV>jZ^hIW^UrnOHepN7&cXhl);DOrQ|7PCR^R^M?ueFRQkLs
z2YPaNIg=xETEO+samLQ^v~!Zi(^Rhk(Bp==2T5F_zvkqfm9m?2ZX48`)`2j*Wg<BF
z#LVJU808@$UwVh~3kX7m7lAVqe$jYx%BQ1194^I$G{=`%VVpArTWIoeTAl7Xjikq0
zWY`?Dh>LyF4!B^X!RG#O9D}Ft`IS`o=0U(T*52noWsxD>m|{a6Khm5u&8=yPP~eC`
zDVVnvjF&f=!>mb3_K0o8Efa6%Q#O^cAU9rbkU|u`><W)fV$dnf8__Mf3w#De9@4e1
z>p1Ci1+Y`o?9&wyt+q&CFf-4Qr`*jV6N4oqs*3FuJ9vqlsS3W<w8N|Lt|H{LA6AY4
z)v~0-m@+N*9?%97i%fJcN$DMUssq#Xx6G-8&=^GsHhFznQAgJW0S$>~y+bxmZnvV|
zCD>5*Euno9EH)r@YW@zuv725Ua33tgV)-)Z*?;g)ATY8c5sm`3#scv%XqIa0u%e||
z)yQGItz$Xjr!(Zb?IH_>s8U0)tZcn#F-$hHj1!st4ByMvd_gJQKeMd|dB(|0Qp&nG
zu`6j{l-%3lMoh>>O1GlY<DG<O!;V}zWX4m=@R}?^w2c8;8NyPrI0~VoVn5}WLcRO`
zV9K&ic!U`<t}WELXAK%a>^7_Q#2iKAi+>oe-g~LpBO4dZnmSjCI+yM{hZEr^dpqJe
zW5JqAJ}##J5+ZbC;s1hVy(Cz6P~&YQ?Pr{UW9MdqF8n=VH^*g@)2jr<JV$T25+fqG
z*t3pnXai?qFW~(sG4)&c$n9EiB-I;^F%f6|3UlbL_44yY-l6I5js6l|%OR|By0YJi
z0OPgwN})bt<wcP+zC~i&aVvheDd$mU-zI#*eN|NpqWux!Twb<d{TeI!Wkj7RaQk?J
zPB&alX^!u5BA*i<1ZI#V4v7f$k_8fxw>y8HF!OuEkFEpRTHC`fH}lH+Y#UFac7&~U
zm^eH|w_51298u2S_Z{wIPe+b|MAl7z3q=O!y7%%IUgOqSw@xp|4Cj!pL;k7z{u9E9
z^MB@Q|H6;<Jx2ez_FrRk{QPQkY1*+RX2K$QL}>~q1?w>6KH?nabY^=Ui`RWxT@F)J
zRLYQfJTB>{swe$wf*va(s~D))HSRlJ$Wpw7ei@d?cStZ4Xa?q8kQY*yb#t|+tM_Zj
z(v&ipbPziJWRiI9PdAU#zqMWns712;8$UzHHUamA(SZBp&q!Z?XE$G;#lLS8lwW5T
zC&PjrTs_lZJVjL2#RMZ%)QZI6Z7ZttFt71}@~A_mylpW9FMK&`;o9?s2Al+E##o`J
z{oPq7FB{W);U<!336>#bW+G}-a_IDbf7dZL>Bxa!u^8bAfQK6BP>)Ev;IJM3g`zN7
zWx<~!aGgmEaS>eKZNV2<HO4Ax&j*>wiaAe}cFZ@Ordi(g8~_4hl}R*J2+Dvz%n#>x
z>`>&1d0?W6$_`(6GZ|+JF3E(3fOr&>&LhPgr>YEb#-vUA^=s<F2~JKFV&94-!f%>l
zO22fnD*3RTP%ObQbB%!R;${Q^HJieCu4vovA{{e%=FShpAvfraO@W17wVh1MX+{nU
zV!T>rNv}{07e|hHj6)bnoT!ei<4D8=9Ofk%g^%GEI3s|rZ{Hr&Lp$XAvl_T|men$h
z`=)(o?+^#OP|Jf>9K*~a%DQ-~$t)Edh#BXo^T!wm*rHEO0%aw;pV=ELAar4`qbt7j
zUd@s+hyQerN>2W9$P%3_*^_l2-l!RcVGYWXK7^I-k5~UPu!3Hrt!gi9ilKxal3W*x
zla)B%kDS}PwZ04Z$HlO5H?y)G9Oac+rJ-q$^+A=uj8CL)UfJ(g&%DZyA6q}lo6Oeh
z)t<vs$D7T|>0?+U+XCpdDilz<TDuxQbO*dqzHroD47?-jQb0M=cF+M;4BNaH;XMrv
z`rH@IH$M%Y#r-XY?a&1N>YufxcXTvvR&3Ggx3nL0mFP}2^$atVm#zeVUA8@XOGNTh
zO|C{cbYs6|zVvVT2%bDPAC`W>e_{HdA*6j7{W9IYeg|J)T3`Lg%k)oj3>)Xa9&lW0
z$*S&u=Nr9gQgQM|hj|Jja0}5ZZON=m)g;5_Q(@y4rqH8m*U75W^zifJbPcD{!=XtR
zJI}*SKKN{YJWg<>gL846<Hvi2mx2L$zcC(g5}g0z?VKZPJwc6S!wC`D`@9{q%JcZL
zLQ)BQJEpXhbmPVO^xFG$=HvT%Ubs1tZ1EO+XL;r0`!r{J_Yi2s^X1|@ta?Gfgo1@;
zy0fHGeo{ij90mEtf7J~FL#GSPCkO#y^vg5p^oGt|qHXQA3mQ&u>@BbuRHsI6rhn@5
z5ViN<N$z_Pe1p#hNpz|=Y&`dgko!U4)5#jsrEUqBK~5hb33XGff{N0K$(I?X_QI|A
zawc&C2H3_+I*^p6kb^n-jOX63IB-Hx_CI<5z8nc`cg%QXisKVy6y#ywA1Y*3Ux|3P
z1p!HDV*Tkef+Z67yJJwWnC@7fF#9DgedIQ0hGLmlv2MQLQZEwf)mx~?Z=WB6Wjr2?
zWKbH$GeZcUekLe(%ZD=q3tY+7#*m7pJZC`#cVrw&v<FtaV09IzB7!T~CL%GHWaS=5
zuDKIhx$5J@-lO~bConCU+|bN-{Dg9*L_B8N^h*8WCIi9dvEpBCF>M@7`xm#JZtgmt
zU*0!FoIj;PghXt#lm?;PXZYl58_jh(P;<!X8Nhj2B6O}RV4;{YV4<}B00RT20+3qa
zptS6up^TB3i`YbTp{R<~UGs`M$b~_tHQImd1kWB`THfi3Hmg*i%}?_VV-<uvhn$)a
z%;`@S;!hR|<zG&AY)kMvRw_G4@Hb;V5&@J1=~<_cC<_u(+t}hl-IVJSDlstDP`^K$
zc<F;dku5j;Z=5(KWS~*J;LU_p-%OTz*E4<*A0+@Ed-vyzttHT;!rlk9$_>>fSaII%
zPoUaCZe|2wna;@m-eX_cDykcCpH3XEO|V)0DtDVvlQpDWbVKu$$dfgmk!$+WxwE6t
zwIV_`yW*T^n);iQZUfLX$g~DDe40i!<G3Oy?vkXhLkZ(nmBKMg07TO{3H3w3g6Bsx
zPW9|rvc=i=oU-UJ>*+t;I3qQMR&ISUK)Ewb_~4qRxLDRCm|Uidl+IM!+4DCyD3^;5
zT=&zGjIsiW$<o+5%i*<if3McU4gBWI%jY2~p$`KHk*IR2{=pCQ98S)6TIBF$2&DpW
zH90}UA>ZJNCZQQ^H3g(f^^w?S@lctw$3u&qEf;8v3PrWYdjp%@qK%rsH|#sHSPv+0
ztGJdEe5Z0mzL#F;V>A~GZqhAq%nJoioh=cs7K9;rm(4xth6-2I(5vfcCS>F>=k<ar
z?=;VO2xQ}@^=WDs1xohuPiCT*s<Rp)Y*%AV9=Fq+^t3=5CuA)LkE_Oz+QyWw@F;sE
z_t~AQ0};#-izsyqvmaKHv{|Qz1S~h5ekl>8PDPpCy+ajI`xo%6lQbIVd`w=v(OObI
zot(o44iuGWt<iuJRKzN2ylst@{gtexY5PNjAf3cSRK(QxSm5py84Ue?A50G;0crT%
zNCTt?+onSAcVw@R_bmAkQO(h$XoavOf}9ll*uSXtGd?>wg4a=hxiL1dm)muP|KbQv
z{5cB?Z5~zj;}?12C6X>v_0QzXBMLdn_JyzCPyQW=cK(eHS$06XdR@@S<bUWt(FkPX
zTDAA{4K3LYAom*`J32;xZZug~35%EJ57~<bBiZt3Nl{xd05vfziYQuFiKIs)1Rje-
zDAGlF-uX82|NJn4fWI?=*|d^PzF=2xFw@@x{aWO}75=N$IO8^^IZJC0`$2NpWt8`V
zxrL~78VIm<gQLY#%4qOU3k(7?J8!ew3?Fmda^U?XB)T<I@Xpw1UH?6n-s0fD|1`_2
zs+d~l>e%G&U_u~YlB~;b20&NaL~^bboBML|VD8Do59uwlHw$Y)%`sC9U|0VI;HKLa
zV+~$y3)HgK{6k0JV)}<IMP@eUf1P-@nJ(+F&w&*1zmxifCT~836aiC(>jU<Q9*XNG
zeJfF<YN`6t3W)`R-J*#qWYy*!b`4NoNo)rRqRs--F^<1u99?{!1$enT_UKqfrDzIQ
zeFx3RTE)_%t(qgc2UAYQY8!K>YoIu??Zb%6jU(D;z9E<&d>x$`-x3}3oD~Z?%E)(2
zzyIs|?cQDg1f#%4cTZcEezRdC0P#52l2E|6(-k#q$7ue>)6HDUa<6<z3I|ob{F~e4
z&}(P@Q}F78BUkg3hcS3AL>$M3ywa7MB~whbA9+ByNy_%<O+|#)ms`Al)*{qh+Pvpz
z?CxAYvsSZws@npv_))_RRh0|9O7$58u5e$GB}YE&bVca@5)R;DLo3~mux<9N<7b|P
zXqJtETfPxNCVgM!OONyj_BxhEa;xk`j!X~8CPmhSPzD+40GV191;H)V&Z<tk3SBn!
zM6O!aI5kB%Fh1<IQ<aN|TXuh9n3_KubJt#Nc$hlh^hSltS0P0>dZZisi>vqxEFSlS
zuCJXB%M7+drlN@I@6=?71X|-K)li^aG8{DjEcYgFG+V?@pThESn`sHPY1P%GV#*6@
z(~HS597L9^YnSs#L<*6x+yZo{6VQnl-jy^_1XonAE_JKt=KW?!`SV;!M%M&NylTeL
z{SD0Um_CK^AQdcV;)fhF_<G+HQxkr06iZRM$Apscmk+3tD?K+xMa^_icm?lxy^^bI
zTyRCD6PemOc-|gVa>-0T;UA0SBU!}}JP`!2{Bi%)Arkm-IKw*sN@p-psKOG&nM&1B
zbQT*$GXSFm2rhV63+hT<%aSh?c436nVyIF?7=K^K1@{Z9_YM1YO|$m;_`bM~hPYRY
z>sZM3ToK0S55X8X4B@`%9PyWYb)}dMNB-h4iS#=u9Ak|eEBY0#2#|UX&voc*#T1cv
z=20rgg|Lie1T7plnNBw92f?F&!ZM#3yaasI8lW-)8ni3hF>OqA29eIac~5uxGM{K$
zA`oRr5=#HD2t=YF4aB*SL3&A|3&%l3XE5U+s7+2z!QWzs-4HT4q%FQP#<0|_^L=DD
zQ2XAOl5#+4DV!s(g22(ec7=cbN}L<fa1olGPn8CV!L|;|L(-SDN=BFSL|20Uaxey_
zr}UDjOl`4hYku)+8cX_?F`GSL3WRWP9<eM=#jA4~CnBh#*bo+_IMa8Lp}-|nnztIG
zd8qo-&!_B)`XQN|)f4j|nN*I5Bp<Ni9+ZhrF@Yk3wrLd%%S+l<#S_=RoWfwF6yjwM
zKB=i6&l8nKakPV*+&j<76k&9^{n6>%;utgUMxlA2BJ4d6)(GqpA>;|Z5;YJFe9Uyj
z^qj?Chw*4sB7`>JT%Itgcc^gA&T@v+y`O3j&T8Jf*CKkFl@UY=2`;W8-rOu%@cMI3
zDnE9vVM3h#b`$M9guM~azM<7b@l|dUl|WK%rH=LY6a#-%)OmXLFlclK4#J+RI_fc2
z_-R83Lgn^2<a-2VQd1A?ta^Rw>axYAWdHsQpKENin*D=}$F<<NmsgORjh~#YiJ#TT
ztp}XM@tn)fyVUi*%fRN?a;lWY)9}R5OGGE8SiDaYOs)j}^LCo&=JTfpp7M}aaB}|C
zxW2EC73;0_nvHI|O)9PbEYVoft0I_G1tb6wk`$$%;G8kMnS+K5Epc-ZHIx_}h;slb
zRj$mfY_*4`U`OW9hcuENsMd()c|WeD=v79_J;^3pMhOgeb&Q>UCu0jkDn#rWbw0F?
zT9(>6Nm0e(0tK~Vc%Ygr6Frt5#Z-09Bp}aw&&vc%QkSmq>Eq~?>{Z3xH>BpA<fHNS
z!;g+(!Q8$N%+gHd$uBlOEC)Yw*(vZC&~usAnA`*W>3V_pPsN@2AJBcw>|Fm^xt`M2
zjbHm#u8%dJk^gwOU16&GTBSsD*er0!5)`mSbsA~47?s(fRI5F<{EZpmFvS|M5T6%d
zPG>#7n{sC`{ZDa!aDH&%Y4;MYi~JAw*s47uJq9cd>L;@N@!T8p+-ROO9T`E7_uIm~
zv>Z6Ww%*tK#E%vqH(yrYpC1pG4>&nK>D}#KeHpcJupyyfKbkmXV$gARhXbGP&q9fj
ziASM?FPmF8F;zojIJy}>KTmw+m>`xZqT<~ve>*YiIykpveUn4XS~_TgDT7o0emd7l
z6r=~Ji)1YtMo-x}1F9R&wodgZ%}+I=^#y*g6M_-6WPU^GdE}R2DW$xUAe0yOl7EBn
zkYF*}`D%&bC>QfUWkad{E*0HJIN+>CL8mTiN>qSM<TH^<8qT-|QIIV%LiScs)XQ7c
zUju4iL96!K-IqXSEqEj%gc9|T<HG&60b>yfyAr|C9%ir*<#`3ZBe_6Rg}Xvtk^n^t
z#oTBF0O5&MFjos6ViBXM$%4-(HFu<f+VHmKXaVQJP$1zuhI#q*0I+N`F1hfde$f_5
zg|A&}?d$*phd-a*BTHFtM!h2!(1?bdk=9gnqX_dELe)tC_+Ps@q%1`i1KV>#D$bQi
zZ`iC9No1_3JUGyNRvbdPu7%*fH%L(=_$k1I!qhO%pv>5z?o5G5c8(0hNGU?6bf>fb
zObAyVIiYdy`+o2F!<|3)<O>=e{^1cx0=9%o4Hu5hQIzzLy->hQjF1VWLW5!8Cc}-k
zIG=k%<jNd~uO*o^b9rdP(3cbuMQ$w87w0V8&cs+B2$un0ogrb!g8|C2bezZJ;#9|U
z^HE|!3@CUzd#|1o825$&kj+<Rh|G$5agrtdXl5Re5;84s5xA!eXb=6Ru2Ou?R*Uns
zLuFK#kOco|w?b(=`aCkW?$wwUju$G9I?a!E7GJDc?p*;cI{u)*s%>|j1Po^)jvQ`-
z8w_rfX;gEs%}nfdvEg3LU7ky~aP(%==n`%B@owanQR=zs`;w*ZPAo~?{2aMoCQ#$H
z{@CxeC3M&F#)S^)h@QT9i+YKzhkG#9N{JjeHv_&7XCSDMj^i)Wpzge`;AWmWlUa<<
zMbHm3$~^f-wWTJ#MYuB_u97d6tR4Z8LMkT`0Bk_Ki9>3gG80NyD-&)(sQaUy50O<4
zk%@T#4+&K#o~~){RNt#N$}NX0GbPw%HNG8SgeBIud&Me%cy#C2t7DP7#_bG18{?^-
zCn>`CZQ?LnXAH@vN`*9>MS-!H7bO^_6|Nu*U09`9#Aq>KAA#VwuW}dU(py%;*6hv<
ztQL#`!2x%$>#}(pG!}*l1bGw0(HSs8&uHX13G1<P1M&oaQ*a5;B0)%Yobm8mGlY4O
zXg$5Q&M?L28OoZj3$yrNjJ;!wtZlcp+xBd8wr$(yY}@v1+qP}nwr$(CZG8Q%Z?Ei~
ztS5Utl~hvysw!1?#(kY*jN=G|OMpQb1rHZX=@j<f-r%p<G=80XrSp#i6fU^J%am&2
z5F`|-Yh=kbL221cIBif(){A-@c&o<XNT&rujEfCVq6Ai9Vo|IF3tzBk-l~k_F1yb^
z+R_@8QKXiT;?~V5RX>n3`lFR+L2nF?`Zia97fq!BT+!l!<iE~#sY0CxK+Q%`ZSJ@k
zf$83L`40KKvnh0LaMT0WKam=9n;cU-?!<}u@HsvmAVjXIY~(#n-`tZu-}q}TsE1pK
z?uUEGUBcT}BK6IBY0Dzi@TYkgU^b)#I!=^nu}QQLYj($%FioCkcA-m$V*U5hx)OIb
z>0oF4r7fHgY#FMU%Ygs(H2gH*+H6;UFKdjX4dn`D?Y0*XA-XDUmprjMQeHBzt=k2~
z-C@7_Iw6~(PFlZZC3|GDSr)3zaKkQq%b#UjFC&8YED?ZUg7biLO0`b6m6ZO=ciOFb
z9tYK|Zbf=Sl(+S+W@C0qr{}3J5*_C9^`U01)Tzs`sQRJDGO;^`s$e|tptvVTI$y?8
z6k3+SHA};NS9xM-;W*alq+hC&%iT_N%0`Y#(tn;$oElR<S0|<Qu+T;MRJJehOn&VZ
zmGh_sPbnCknz)s9pwa4`_u!KURp54ED%-$%OtVHzOBJA-twu*;@xu=Cg$fpj_FH#S
zQi8W}inC}J8u?o~CI@_@x$<CtLbLWtA^l?w`iIDg@xOPg|3!XlDK~D_M+e_=OzGiJ
zp#LH5aezR^JwuU#bcPR#B<u%53|RX4#!(vzRpzns=;q|a70wS!rwXQk9f6hs7?@Ci
zctWU)2=U{I1v&v<;g_&|Dcv4|_FKeF(H9#^${U{0CdK6t??*sEk3$b6#HZh=D!yKS
z<VtOS^Jq*h43!Q_msXjt)b6Zi-Sh!)7ngCPlc%qTLyyr;FC^6j^t@_dH$uh5M~KCd
z;Yu<<E{A}SkR>V{F3;Wl2ITq)=J20bBOqEz&ZOCcQ=!4<FV6SoP?b)h7Hh*{j2*@F
zQ*@?_EIdVBZ9+ZeW*U~)nj5y!<~v$C=#R24+~to$xy2x4k99}pjR2fJOVR_C9d8Y|
zzHilGJN*$d%~W$zT#?cUXc^0M70lAW&)>J(jMg8YwQr{K5JVyv=<1qeg*T^2@xynh
zDWzAJM_V#kgj=7<g(bOVw+-_Ab-b>U3;*dCV_^KBi)-{OKRNt=J^uGqmY<m|>kXFw
zgoGumm?I1xVF&oZsFq<wrn3ZH?ELuGV@+wEY?PH+Xg?Hm-mtu+g}G`LYPY$+Y#(fl
z+gw87<9?{yxU-{=>GS!?)zp+JBpKvt8QREsR0Xv{YFZgIVf8-@^wHt@=KZ)mi=mQ(
zVcv@*Eqeva^s;gpW6jxp>vX5C%KT!h^^EH_<9e&ipycX8GOAi)+D)Da*|D5h)C%~p
z1U`HvaS4A<m6hiQbJ+$7*0?8^T^0_#BbN<}Sc|3ZvW@_q+SbzM*rxNtnkqHtN23=9
z1!`x<_rB_uPV*y`Qj&~ZbcB8hdBog6QH^h>%fhuLZII2cx32P_cJqLy%B!`7imhM&
z+2v-0qc;8lpue1eCfx!TZo!>TmX#(2H;E||X<TCChcZ{B;Db;Zcr|A?rQiR!DOff)
zbsNjDYjiG_?sHuCX4Pg2#2~U*(X^w)Q6SX`01D7t(7Mw7<Om`JcR*lF_ygP#a$WQK
zW*EdiQUDD?QTJ(R?Z}X0C8QK3ZF+DOlXZ|o;CvnlF3`g5#0}ZUyru-^WKBpzik2Gf
z^tTL?<5YRfQyM3sEM6=Dmb4ZHE3_?|*bjpcTY0B8x#6ilzBTh(J(=XvKp|x%W9GQG
zoDl`txY)uA*0k^}os}H<Z81bSgN@ZD$Yu}W3vdGz*<sRB=6rC1QX%{`BudGR)96mO
zy`jn*H4H4e)!~gq7`uPZUN0sG*du|cPOT8<0^mUe<}GZ|OA50go<;~tV()}*$OENO
z5!9-*awFWymjnZ?!5bU)QCZOZnpNyUi>Qea-kV4OF6iN73F-weGlh*I(i{&4hDC)p
zXnSEErs7}OTFS0Bjp<+t@1Yxrt<8^huppc-H;9uR><7vsCM44;<52}+j+J`%_|rIq
z;+v=NO`?|v0jy&Dv-T6GDh>-l*-IV^4>8%mP12%zMO)F0o+cBV5gT_7EGR=b`5;>&
zKHn^TM?ofQ6BSn_vLBMcgQW$)3^7revfw}G*@%d0mk7#2p=D%y@Cy}<Yg~P%{>NQi
z6nqQ!S_=0<9v3gLGzhOPDL^R_FcLQbc@MnaCQ%B`wsgoK(y)NEFnYRU`d}79im~81
zW|ts;K4%hA(KC<89<PR;fQTCQaH+)oB+&&XR4KoqHo^KBi1finp3VB$hWR@)msxe;
zs-NS5_t=^J%qCmL1P@N(4dD)g=#<M<f++VmR+Kl8>rIDPhDH1w1frj@#DJN%Ksv5!
zqq3nBdzdIe>I6#-^)XjScQiPz?Wfn%b+$fqEgj8+8|alxsH-VMLIwo>tH;C9;VBjs
z`Ujs35XSto?cPQakWQ>wr-gbQ?e#<`qVAu`5YD=R{_uXM$D<KYcT!%%UDEe9uBi8E
ziqhoW4JAWR>3pMI!pf}8d00Z>M3#9#9{CRQ30&t}-ga^dB)Hc)j#PU|$6kr0#cf4W
zlSn;lHYCLvgi+QHXhKX9&fCdl!WAcax!zn-?hloHqKvNI5DLv@_ebS%keQLkbQ2A=
z2If(hZXo@U?zD2qVgA+iq7El3IsK3KK55P9L0!r6<%`X_g9>{NmRx1S7}k06;gy@?
zZVsHtYyXch6heWFU;jdFgh%rYyRh0c3?WcGDdXuJDVIKoW&Dgwvc$4J3$7U=y*MDy
zkMJ<ZSq$bKfhofJGQXD2VPT!)g(VXBz;>(-%)iraObTOKek$)1-Q#H%v%&Vtk+n3_
z`ph+0{6H?9FgNAF1c=i1w+sGTD&7h_Rf~8yrgqCot@r~QIq&nHRfhX}F;C1}`(#zR
zwQ}c^Ud=SN*N3H##RP&WWJ|l~6rn6!Oh!C&I&`<^`iHj3`pYdZ=o=k!yy0!TD9b_>
za?)wq_)(FzjJGcx=;a1<OA|hL0g-%|$#j0I4o24gVw3u!){D`6@PJp*m9?fRJMm=4
zX%NqrxK*1cX8riFKoz~ynkn67|5ajlL%UKo*b^X|5tDFj?S*TEF&8)pXkvIm-}LeB
z6%K*Aq{~@15y3^?>vv)ZeM>PSz4pBY!BBX%#DI&GkdO-xLH<3ZdrG)Joocyc-Hz55
z_+<%Bq9#2`bw~)sIJsP*bO@9$pQ6XD&|o`^T;>6vqS)4-h1-R={L4N1s$sG8CM+`h
zP9Rpbuo*EL4_sT%%?et<A=Ug)s9$1@s!Z4&bVNyR?dJ86_{O(L^S;);`D0mbLIz(-
zBXsh3GQW@*kEzuBn?7mjJ=&%IsoedT?ElA}HUk|y^S>>3i)!n3>mrDr%PP9tB^34y
zXcM4~hv1~Pb?T(jOU6&aPOW%Qc{TWusUU}BiDS*zsnC2dY<UHW8Ec+>02!yJr#zdU
zMdJ0sBK6iXT^X4kw^B59e#l|v7EXzfiKChbY2y}7shKit$Hmts-yeIJBH_JE`49%6
zhArYcpZkN+^~u)K#hP=Flq08#6dbQ##n;k~JgJ$V?+NrQuD3@Qf$rX?$DW=~Pvxw%
z;MVo5COThCobS7j8twa#zc}>&k}&vJCe~=k^lvFtg_0*M=WJMvwckG%D+r$!Xg6z{
zOugvl>I;$2pLVQKue)tloY}qV)oTX`ig02`DCV|<UY8U!FLqe6cqKoznU9;iV{h&-
z8`INLGEfo+?#2i)9!W|P)4NP5!+%&dR<Rgqs0kj#&+&qty}~sg7Qx0J94KsRzfZ5b
zIo4Nr-O?sMHf=7TwKrKUeT_@ClB+0bSbmhxvtQJAJ=CooB}l1T09WQ}RU2qS5U@XU
zOr>j?D1w<@z!_)GtUoyJ9R$(EJ0pEvWp+|rf8?mq_Mhb4O>Z`}J><JxW?vR*T34q{
zmc8%RJ(%QdVw@NvNTO@=UX3|wG&?r_>Pw#dIKI^OJl@;z6(=Kzm@1lp*S4uX3|Gk2
zyLMC4rt()`cjJZ=l%9ad1|3Hy3{2Q)SGNIlfFclr4~f|!QcuW~!2s3nir$nYMB};7
zPMXZK`mAv_`TTpDks+jx52lSAH<EXw?XjLTRiZCX(;SJrX(64O%GCT=*qn3&fgHbX
zUVmx3p!pgN5Q7_WvpU{NMtf$oxfQ5($O6&1R(Xs~CG0&`3cRUmloEw`s<c&oN;jQ}
z2ubF&LT#4^8a~(|p(+vU_e^fF20EvTpJd+`14>}4GD%Vgo~_ut*!H;XQ9%@ecr!xE
zs-(&TZj3fbqWV`(FeG~xn*)nE><`!@(|uS^+2rpgs&hwhFX&`TjcH(;<HNeJS~Ngs
z>nJSpus}ZDEizs!0`Ix4Y=`S<!FC1#@KR}0sfz53HWQU$tl&0-ln9lMGt?BAW7Ht}
zW<R7{_w7I^VkTUuL1f}+qaxp|pw^RUCod;SC4(4D;f@_gbtV3U7`XB@UU{V<d%p(m
zCC4<z)4N@GYDqi4I&7&;7M07gkHUE+I7i{CoisRTZB{8EzTLLvCj)ueEBLnL{<tFY
zkl(9Dh8EIydCzzA0xr7rRXac>@pb5X(6L~Fnqs)NEqp<ud_w|Pb2!C95|0ki8ntO+
zlL@R*#1JzXrSH?cFwCK?uMlYKn;&bq)txl&Z<X@Lz@>Ytm39J_W)0=rp_)aCf}`Nk
zV+a+-IqEEh0rK%umtSA+)9yB^X|gUs;svLkPaV7q+8qQL&lTZ?89-iT0DlGWe;~Mj
zN+IbKh?oL=V!_{FJ_0L%gBVhtxFA79?C^0X9kN{GP4lpo4#~v%m-$D(r<9%&q}SjZ
zMW5&p4?Sa3!16`N&tF%)<#%C7)E|!3Hw^Ic6_KYDtipWn!HgLIe18n$r&1C%2}CAm
zjM(-L5}ngx-gdAa9kW8&Fmk4#NAlyy_Q_cUC#UGH3uN#J0>sw?w}wLvd4_s=q$L=5
zDD7qfjfbreyx5{r$R2TYWpt6~rTGw8xVW;ok^&q?Z7DlP!%$ba6WlOTX!23xD?DS@
zkvjW|xg-+4v5o*bntbrOWig$$Kla5anw&>uF4zki#E^s#op3BvyJR4O%3nPX!Z+ib
zz44{+>@vG??92s!guS_rQ~fj0fa6!(!5YDz=ckaO_@xV8kAqqxkWG4e`o&p>GqvF2
zEpQR8=ZS|<CELk(`9eTiXLcJ}9Ml3%G)2>+E%Gj@H0&G`GC&{le<Os5H6U+)KApDN
z6>x7n0reK|L&JeiR##mNI49h>Vz2O4Rr=!TC_e>fCF*Px5~RNqa0pw}(<EF)T4o3a
zt}NMF^g|w*)mq{ptsm77IUzD}$Pe%by{}KP)jfvTg?89DUDzAeh1nxx#Q=d0j2I+p
z&hTh{AJNFrnvmkkaMTR|qW4qR8w&Lm+%Jy4x~FOFiMqQYlL=>UETEjA?@ow3F}S<m
z(Qey>W2B%c??B4}9X{TAD72iV6=}bH`?y8Mu#7=YExcF+`b#-3&xmU2^wvfgoHP|G
zIH2sG%Zb9E?V&nfDJ?qJ?``QjLr(buEMm_%K6%MBBKJBah&duB0~@1V%KC_mh1^n&
zEtYi@TcYa}rji3R&y2HcOnYY8ne-Zo$gItIswwVbORV^ETuiR$(3U^4bdYUpikGYt
z;_NKEiUs-ud#heqDXTom_L_FPI@qnuS2YnUIY4kbS@p`fxD~`X9K#ahE`|O^g>$@e
z`9dcXXXA|qb<V8!+&6f~^AaGlE{pmYTH(JczR<q#Q0<EokLIJsKZNH8pR{H@(b~c3
zpL(PpUl(!o&Ic*3tCUJuC3x^PsltLfsx<<))fxbYZXz%TSySt>4VcYA2x*>K=r37q
zyf;29P}rX04H7L7#S<DD+vK*v!7BoA%4e>s4ZUoBtQJiV69BZzA)4J&fwL&FL{Yl^
zy?ucB_<$}vx`1EdHl}BpAqzQNV{|=g8SDRbS5llo;NwE*;T@?)6kb@w9ecrfhnnoP
z?WW!Ad#1*+*x>%w@aI@}$RY#wOTi|^^9px;bF45~PlkL73i)GD&X<(w!BZWb67lzI
zKhuq<fvd9)orXN=SD$WDhUq=;?{UFR9E`DkVwuQ;-8E6LI~#G_Vemx|#oO+L9oT;|
ztgtcuLqfv9!0@jPD?cgye;JQ|wvO|eBH=dxr=)S>q~aapTQ-F}B^Yqj^9-$sXA#KT
zw^moTyKn*Ifr!uI3g)j~dXYoyZD_VzcVeE5x!2r1Z}lB+A5ikeY4Jmk+E~+`0<OqQ
zArs43(;j%9)C52<U=oWylasG$zrUiJd85E29)qvFJUQv=c-uPOUN;Z#oBi=stj2?9
z-^bCqqiDRletY12K0G8M)7zIqDE&5Z-^ze)KJ{DLdqB^H588`%<xe4&B<Y+5N-tlB
zF%g^3Myu~h;~%27R-Qi%kkwmnu;wrfoo0X<FOp)Ux*y!+dgsRC7#%lH+^#71=y@At
zpxP=dcZgM5EsGU(80lvx?jx~0sz(B{g{A(vERy4GE=bEt;;C_uQL<7s7C*~M=3H0k
zXqLvyl0rkX1`*T;7$8R>i#sO2&FK~rp0WeI58ZaBTpz$MpgEBh-BAUHLVNJHa^n|$
zR=sSUlW9!=;rI2`l*tQjyXQuUs_)*ZKKyo350?_PB~MfP^)-DOiMe)tAMbtXKHr(L
zi&^i}x+aqM-w@79XaWsHsOThA${%j`k8_N+RzewZV$DPvntn;5ox3K|h6`s+RReZ)
zv}seDWF7@lBi8#i*Kd2WRD~(AStQY*p2lDHe(j$5`1*o%MS_$vi3?qpEUz+X6rw05
z4?QV7Pp29uS}zs#!iAd#M=nATX-IPNwNt~nI$~yB0R^W&kYx88sktIchp>yNX^hpS
zC<}^$RzU1-gIA2Km7QKs@okZTTkwhNpA!NAlM~Ns4)*pef<0+w2XQN?GFD8Up1t%+
z{vlX^o!stQ4S%4|(4;z!);lGEraWgT2c1aPD^xEWw^Oe`oeUROFWg-eNmaXC4Q*Z@
z>Ig%Jg$$_2fkyODBP{VV@BrAD2*0%^W`~zj9CL!th#*;&n2^!&jiHYt8_ciS^MmN!
z=a?Nk!VQi9+NHf$;_1pXS*!`<bC!knE!O8lKSe5C{sxYNazS5`IW|JbGN0HwMzTh!
z2uSk{7G&5_qUuL#HhhC?UcuS|i=Mdm9;WKFU#l~dE-9V8B`BZUozxSq1>zBQsVD1L
z_=#H27wtxGb0Pj1?C2?1?3g9!^_qRVUf$VpcR2!4ZcU$mq$T?n>B(=HWN}8}Uh#*u
zam2vq{}tObDhE)9_p7Q922D1jp0kXM&iAtL&k4+_G(p*IiX72Vma22DfD<2ApFMSB
z=kb9@E1|FhRw!`lJA6=$_u&a3uKcu4yKEZP>~->G0WCbVn6C<;?z;&gbdp`XXyL#o
zfUi(IUm+bXU{Yncy(*W+M@VtdaD<g8olcDNZT^{mPq_)W2|W`l?++j<5>9r4f@#LI
z@9Ma2W?>M*C<2B9gW_*^XAi|Ttr=evVr_-<T!8O{sy?5BD{fc@J5_cth5L99iSaDB
zHSpSJT4mlbD8`2oM|@wrF~<br#JYQiv6bbAy%}#x*Q79WkW5FUa7;EwT|z({uxW(%
z{XlwqqnX_ScaV!yH71<(Q;@Va6`P0tSUThc#I%aIx5!^e>a;&I20pigxT5C(Bs?1%
zB61R>B8fvV$6zVtHY3n6gD7&FZ8FJ(*A?pHG7(9OF^#CW7*h9HqrKJ+Ow?bS+PPtO
zT;S-!DZL-JRHVBhc^4wrVL9b}ywrw^#q*iQ7p*x}|6Ah?-jzv$coQPI_tZQtggDz>
zE9<P;LzuUETU`|aJlI_IJx3lPbafC*b(LlqU%9M)B68oyu4$3>_xfBqt0b*Sf!9{7
zy8*Zv>O+$^Xx{u|9{9JICPcpFI!i0H*ZDwy-1>ahtVX+z<6SQyP5NY3m1Y>WWKCzy
zMr@GW=3V4Dh8LEWN9)Q0Vtyz{iY$3Z-=RfE%gamZ6N$@f{5h}hIC!_?tYGkz=kLsC
z9?g$>9P@_HzrK<VWtzaFOCLr-`t3=3>v|*j1tpBJ`f&Kf(vDcGqHOz!i6fb9rE=Zk
ze&Kp8+ejftG!{*#H@uKrsfV#s%vekfH1^pe$D^elR!=_jcZ77loah6rirxx2S<*t=
z@)#v?wnl8*>+Qqh?nAC|bQjT0j-v3(Awq>y5qQqO*bw=nFc_|$L1s#^T1}wiGt6M-
zjUuci<Gk6;zfug_`oTj%2=W3E92a;YGtp&c&g@szcBZV)V%%CL&p6Fg4u$WBURjAN
zQE?a4Q|Q+m8c<df1uzI4??elvnQ~`1*OoT%cESRsp-ZPs=L`B3q^#?fAJ%3b>aHi_
z-nFXPmL9#dd@h;P#rq5rvK%{c83y5#%RC3YTbI3(e(ZAL74GJ&Lf|HwGqQ-YhVR34
zOrO@vGpM4cQgT^LS~rdxpxLbTQpe8Cr@OUO_3Nk^#1X1W%{$5Cwjr3+vDpJEr8=ob
zi%;u;*-PzeiKoMaa8%NDMrz-5dZ_M(DWjkTb({naWuDwR+tfu~X*#0ZG_X+(%<bPY
zbeHspwi>qC%BBGUoqB{%5R+Y2PP7NEjD-BoLn)8QbZH2Qq_K*@*A#zmW<(H@Z*Js}
zwWjqZrKfucois4H5}bHhUV?uQHV*E~fRN@4H}7*^lGdfQ1YB|P-aYeH{tYp!XDpo8
zO6k|<HtwG{o;WJ)F9vAVt|+&y9qFr6Ie952K2Sz(Qf(SWS`!Ob?yi&$_?kipr-SLF
zu%7K;w~a6o^J)@~o2;ipcWk!97-R;8aluxk$r!hR%szb918Zo_4gb=T3yM}39WX)!
z1w<(te4fb>vEot5q1uIAC-}``V=H>CeKNR3I(oIBhf>yM$@nzGi65K(M&yt@<K5Y_
zioH#onfU>pmN^S*m&Cu~`sSZ)$ri@Ny#9Ad#Uo46FXcd}$j)XZQy@FJG|JQHU^`oT
zDflj2$fWdZ;Df;S-Qet-v3sB{GgU`F&i?+7q`k3MsCe2(oyxk6!aMF+8F2VN#V<_%
zFo4F&!1{mFmw(Y8id0&+`<d_eJf{3Godyx6939G%#uby0cgw}zdmx@P1Yyf5rCmUE
zxO>*pU_e8g<1!<5n$Ftf{##P-P-4H{9_F6R>c@!?3WUmbTi*{nj0Tw7=e+kbW5__#
zG5j)r_-p2*hZ+WgF91Dq{&9Hs*@fxxZK9Y)XyE)E(6M{V`xdM(a+sG3z$SS9wx%^Y
zVmtVD^U=Y<CcJpR{N3fl(oV(XLT<cPFeZ*{!1;`?A4LF&z-|x5%uu|l9C`q$Ppix_
zZkeRTEpY+Nep$GXscYS@W`LVYRkCLfw8mg(nq&_P?uLCAD9Ye%-LNWrPFA%<BH6rY
zwrqqNO>K;7R@`h9z*%4b_5~j!$kG^FGqp<P&XEVpgTPPP-mr^K(@Q?P%p$X_b<Wg(
zHy@gM{<l+d#}&<^c!&uw9W6^M&tTH8kQ{$yXryXf*|RQI(Z<m;4_c`_Z}E3k9rUK$
z@Hzh(>ZQMlw2>4?!rEq`X6fmgR1+U*m`^n2BVI|r@a6KK1o_DVq;bzp5Eyp^qv19x
z6-zS;QFG;AH_~9hUCLM_*!EE&U}C8Lyx!fIo+()VK{q$XMFoKL^VBL+yjMy)DYl3=
zPee<NJ5>l3&PwA(6B6>m)Nrm=lFzKc_#WhOHISTYK!Yd*-$!1#3en)+AaRwKuslyb
zGJSE~`5ks-N%^Po{yqO;OtsyJK^lGG47(b;bNR}8mJ^NuXyhg2guQop@Ya7X01xm!
zdH_dVjrB<F^GX?aF)F*=pkUOS0azAsMtX)bo<5#-i0rPRmcThFC0Qy_HHe~F@=vPD
z^bh8rm#QF%#8hQ+ZfZi?+uDZN210yMOS4yQ1|VpXiu$PAM&S*s1hZPvh-5Bn^g42F
zjLEE75{)!{03<YT6Rn%@K)ujCaXr}BrjdbGTZO!T8f7O<{bZhU5|o%HWM9c+UHbY<
ze9a#D`3va5cfjl)FO+{G%>AIN|7}jG6x9jUM~5u(%rl668MA&s(kia(P2p%aLIN8`
zrlxK>|J=+>z~*_=%Z>#}8)H*(;M`nmzLw)d#1J3&0&)oX3u9&p-}TomVDjWE%aM9a
zVtUp2*<UV{v7CG9$^_Z4m}D7Mk*6x%_T1^x>|j@l8y+cF$!GTB+-&7Z+5)r4ARpgQ
zUWB!LXw(7uniI)cgjL=X_;S`EGkWxY`no?Y;{OrL{`XkwzwuBp%8IGb$=Vz<^zZ`%
z7?6SP{^Ohc!_kYL<-bko|9h)*P*v(bc#6*JYS|yygz<|MlrY?<L@ZWV!)n$XUJS33
zN+5M>QbRhxPpdOqNP+;xb$PwvL_Cfe2X?z7)fHLF3Nh`&g`afUy86d(DW=v3vnx4z
zaK?ea5z9hM!92Q1;@ZBGpxM>##{5C-+HR<LH&{JbW)U{&WZ!xaM$<(Z&Ow`2SGsZk
z?5c!&ucUBnB-aR=)g411E?(AZm7Gh|cq59G1R>0%A0vTeS$l?4uDxlsD_SOuL`6u9
zbKicEV)BCsflVHK@4t;(6>c3<ex<4A<;a&p{$t{tzUd05WY3l`yvCgn=$?7$WH|vb
z6$~L;K-33;rrkWovfkQ>I_cHOtxPWst0E#_=+eAvI0}m32Uk;vp-Zmyp@{R`c}#aV
z!lrH{qy(gtL=`)gK;JxbZV)1JZm*%(-|whU;eH$sL+x&|!n!!MI{w#WgkU`326@&2
z^?j>{Bmz-yovsS$50y$Mq%%q1%}DSd!PZ`Ks?CnnFEF**ROo<0A@waxB~pReqtjeE
zvmdXrRO9`Bd6g&j?+zt<($F`kfXhu_@86s>Tpq>^e(o>T4ulE>L2yl)ePQwic%kL~
zob2qXi9Q!H3(@)KVCc!Wsbob1!|=lkAhMGjvX-4(I6<)9?QW9w30vR3YZ`;<ZAi1B
zVvxl_AzT`t<^2-$?E8fj94n7$5sWXzqUFI|j{_g)gDF0w&FQ{`assD<=B@c)>0gLU
zO)dP+*NB-^@r-7@J!|NnTs;E_s=niC(k8>EZlxmJelR&WT}3a$XWpqsGD92dk=UjL
zXJ6iztW;yqm03Y?+Cs4ppgmMBO)>T!N8XklaWX^#sY|#CoBYRDnPvGUbufr+gq11r
zGT7hQRDcusMA1Dr?hzyQK}ZTX$>;|hu(1~*`zSWo;@4SoOf=~@GJo=`3Yme_F+#gV
zZGMY51#&6P3pisMC7yUFhfl6pY~TOW*Njm<b#B#591{23nb$vXwjfjI7+y8!xsKeI
zkg0+|J{&#LoaQks$CSi01V1x~te?w59uP)4A*Q%D5&ID3Y<zl)_9z`c1;I8XgVA;t
z4<PU0B()M6SER&Cv%K2eH0iUc#M1|_A`q#^R)0aC?{C74&Qum7&PsmJ!^}lk4X>yb
zXxVy!p8FQ-bjCVG_o|6ja5~1vh>V0nfC2pqw=r#aBNF!}&IZ<an~qh}m#fNFOS+8k
zilWoT@#CfDK!*NUwAPyC<<F$HMb-KZH;H%KxLP{@qtAZ-3%-GrW4TkDGkF0C43jhG
z)9Lq-e9E9!(skXU)p>jh+F$Q+NbX=9D9A+mIP0^DkU<qk8q%nBo>_X_mY-fJKd`aA
z*KAx0JwCa5b)p~x7@G^dF80S0UNF@fAkzdQk_ep+iuA${vZ@+f{GiZ*%u8_SA?fRu
zr2AZWp-R)lAoCK_*%Bphe3b)pKG??ulFv8WL;VdSTM&&qHk@r+Mnc`5;IDcf-1o6s
zzr(?8ZUn{*JU0<rLgB_S9IKYHP31eQ;Fk^W5pB&wwC5LxiQw1G!XcbWGVjL2>tE~k
z5BBY<K`dBTTSo3u`pM9xvB8KR&A0V-UEL%}^*v|yfSdb<$^i&I&*;8Ez|I-HKus}{
zzGq$tD1GTC?nzHgO%pygvQ9<1XZ^UFrm*Y)7f-KKcZ&pSRh%eENeJW^BcM~)&r7oD
zfa*2(%_^tS3H?)3_y8CZ-3+?()S3QJ;I3YPwMIaRc2iy&Ec_TD*K6&?!ghJ*&?(dW
z2lN4!A>RJnEV%}^%WiFuI14T&DCD>-6~lauUu`$wAZqPcD3x>jo9s`3{6W33a~S-n
z7l3|}J!)%E`k1G+3S7G61)|QTP%&DT&fG440H0x1lUYBMw<dunGBybTn$6|x-AQ`R
z>;bKk71+eTC;^~BiJ?to08~vUOhyfS>QU%2c3Gp&QOi2&`TgZAg;TuJ%voOBkeXwB
z<zUvpF@AY=c1>=Z1ozi|+(2N$@cDH?ey2C<tT2r=JGjkqyZ-#e8pEk*H?DaUH*#u~
zFToKjQBP2}f}4+XQr>lzfmyCtzY)kRK1Cc2>YxjFjD-W67QAyUT#pKShu@kqF{M@U
z0t~7J+>DZb<#n#GCALF8j_xz=<C=|-&+si%i19&Am+|h{Abo-#d}5-G^XdOn@nZQ0
zE(HU_zo!-c$IQIXg7`D1b0fAK1<wTmRI-4>kUA_4o28~v#XV6V-ypQUu}Fz%G+g?5
z`_2r|MYO6MnaK@F)FYCQyS@DD_JCw}bU40wKOUFIkivH2kU?f>Idlvu^Dx4M14Aah
zx-{YGVfmJp*8_gzF$rUoiRgux_j|vXcK`Xfn6T89f}IolG^04##F|$ptJD3PrsMnd
ze)VO3_3(H+{O03oe(>dEetrD5c-jQ5p&v4uLys~SF}%`FSR?X7o6s!9(RIaE3$;Ec
z5xK5e(KsSpq;k-Xo7T8{l4Sne#m`suGKH>4?IZH9v^;BZe#ZR+{>KOF#{<dHI;;CN
z%i?5*4*)_FMv4L*O)a>LP#)8GrWn)m5AtQ7y$O3eshDuE-l3@j(*Mx$r7}Hr)Tt{I
z#nBNhyT7*2Mcwku(Xp|z{80bpRq&;?*CgCHSbVlJ-MZhpHYKRnSbjo+qlfR1Yc#`9
zemtj&_cYojCl>fV%CKtg-JB&~F3AdQ-tIHfzOe&N;gP`&uRe4~QVqYuc8?_vlVs9{
z@(1-0eV4^gt<cw~9iQs8b|QZ?v6g9|NxK|rnu%T%IRbuW7=|JGhMrVP<#9ig`O^p@
zL>TSMnAL)C{9T^4l0i<ME^Lo`c0Ekmoek5p54CPIu`@pUad2MEm8yT23^HWL()#@8
zICLcC9+2k`jr12ropVd-1lB&$gu1ct#O@>`%NX%^#P79}vJB=eO!5&pL0p=Hy~E9{
z;7?j!WPu@F4d;Z*Mnh>6SsnV^!M`cm+<I;oBDH$#;|=bqyNyPEQEBv*2Xe-&eupHO
z>cuVU@>$~>1*lywx9~R1W6(KTL(UV>Somb`b{e0`70Ty_xDNy9gf4D4c|>$$=<LX9
zmSXtnQ@iYX^CdV#BwE^*td1^KsCtRR)NXQUix&4mH3EfvC1n0f?+~YfqOiIPZIJDJ
za5zy#dRTt2{%%@SEIob3e){B9E+y9GgHkT)0>Nx@+o<6!L}Qisl{er&G~}pS8nVtp
z#%ZpN;c2lRG(M~Ag>30W@oSr=J@cg#b!rk@vecMhk0mss?%;kL{93HtqHkQHq05*-
zhFHdY%jiKvW8m;vAJY&oK_)|zOseB@HFjDw$67B4dr$rR<DeB&E}|ysJn{YE*p<Gb
z*cDhjIPi!RboHz|Nf=Tip#rY&Dtwln&zU1H7CKDNxnS-z{FVG?E-oHmt1zT3(-Kfr
z|Mf3@{oKn4-)dRggOw-Ym!cfTV7fyr#xa~g8`<bXzJ6a!g4&qYpXU!s_x(#@K%a(T
zuBOap)m|wu-*7$lR#8bWx|#xnxNbpkb(&2F&z>%~=WgTUP#E;Gh{=*v7yXrCJ3PWY
zKshNY2K&%Gw^w>~M&MU_uk&;r=3)Q~qaa`p8of7QH~C;{tvT7kJ*MK>ZTfsEF1Qyb
zd@@O}Ey0`X5imoe>M{dv4>=vK_7t#!z5_~fBXY294jYi9jBT-4NROQhis}@txf-m@
zukqK+@Df1;VOFXi-#g7Fg-y_e0YDea>nY*r^#}ZY@2+w1%M#mV?p7p}!#l*KCtUk!
zmL~`UY4pxVt!BdC7QcxzKgUg<lB+Ps#i?zTM@w|aNmOox`<H7>TG69iXUjjr=V8Hg
zsx-#7KOe)ARf<iKdKPkV{x|%%<3kx7LBM3ktFiimFnS}!#z6ew@OJZHUe#II#z4y;
z<X1gEWi^NlZvaGt88QPkYrJFQ8nMa5dqwLi;`H@Uro%{QiLS<MU_^QIxoY7eT=`rv
ztg!NGfGWG$%V)rV?E=^68;RNQF|JHpwbv<j63y~rwp8HD;m3tR1xhnCtr26(Umu!j
z@NO=KmrFhe=S<&=)4X53gY|mKS^*Z=3;B9nWd2cQ;{|pTERhNt$(M?qob(+!7E1I`
zEaM0Sj!Ph;O6=tDA+}Q3Xq91(^t9)8ouP7DeWoe^>NTvMeDuAk_MIb%h(h_2!?|p&
zIoA}t3G^Z?lwm+JxGCCIWG<tb+c%yoAb;G=f`tR1@9R`wI-c*pG*e#dVQz4?hD?#o
zD1+~i(6O?`rh5F|e@@H45drt%=&7$Mszo+dOo<%CTMrB$Ahmzd{kCpP$4+}p+iPZ~
zx#VhtZ6YHVw3v@fcTp1>A06YwwR%ECFv{JR>Su?x<nglv`m0P<`?edI*C_x=z*&D6
zbJ5itHa^CyJS&>+B)6W{SYAXtunbo$5E5e_uvgN6IP){lFd2BNMkhESF;9DWVka2J
zbLBj`UI_Z~iw_Bn8w(l4x-<*jMf^|l*C$_8Zh&J`;!}VX)MM0Cw@fqtXV}tuBiG9{
zFol`U;xH&yE#8b(xW7bh6EFdmv%GLKF9yY*`Hjyo;AAECO?4t5P<Uq<?NqVefVH2|
zW>R9kAu}<B5-+J02S;56;FRa~JSQjjL8`UVs+2)|MV1Z6lW7rUX)&x^Ehl29?Fql;
zFHLMeofIHq8aHs9>%v$_xwzA6MylpCkKRUeF8o9OV~v+3B0}I)r*7bxXsrOAhO<Yj
zf{w4S!@dXU)_DS`iT+d37o_4E32r?#Vs`&I0#T4u3)2-}_L)QpVDvK&So;!SFqRQA
z^nB|UA1nuh%W&22YQBi2aV4RQ!<`{g-Fdr4&1XD|T43qjy}VZEFwl`t%T;*{oX{f7
zf{f}WW)zBoh<szO5PfYdqr-XCBh4oX7d;&$+KYUb+Vt*JT4T46Td3YN-rp67mL<Za
z(woixB(#vq4&p+9ew;~v8vTLVNpi>Iv49L@zc1lYZ$g~BK;_IPpUyxgnQ%%7oo~Oh
zy))0KerX9x7wWtSzvNVwSRY&W6}sJf)prV*U5~G@?*8W48g|Z>E^nu(qsyu)37^Te
z?KM!5L{Al2z3ocEJ;9LEdKv=*zG^JnW`UI#Z2-C6W=UM~;B5V6?RLq$mAw40^Cvtk
zEl~R?jHEkAr6q5C$;y4!$g2J`XU%u8QW7FIZT(^;f6xuwq$=Iu;V)K~{rv`$1v->d
z&4g&d7o7VDvD`o1MX~<F_!m7B%fCh<O#aZl)>-~T_nOK7H30*6Es(A5fWcsK8RI5%
z4o}t;2T?=3oHXAC_kPLQaU!4*wT}-9Mg%Vv8ytMhx$P|W<H5-trA50gLQ7WRn<pYJ
z)u%|^BGn}RkE%nm=>DJ@As=MKg#ZQH^Tp-<eSLcfiRAR;^Rf7J^0A$_co~;B;ydfh
z&6U_?J1izZA~4*AkgCz2##Gh$+toEu!1^fwTB0OrG{8(>{xGtGmgTXnWqnCl$&Hl7
zQTcehqoNZWfQ#N#$ks$CW`U0Zke}@riPL^)p2I^l?zGS2kU#zk1g!PhtEF0H@~Zww
z!+}t75ypF}MA-KaQYF;*=g=&T-e(hT_2I~_5#1Znaco8Cc;tq2cjGR{{W3K6sN%e;
zRZDi3d7W{5SoC?@Sb`huVVHmziqEw&bE)5+wyja@EZPA!2H}s&g*ag#!gk*$D8vp3
zulIKX&Xe=V?I8gr{K6_0JR-~vH7%mY=J!P}Edm|F@P$Gq=)`sN;e>I|QDI8;J=g%c
zH3>vu?v1WPkOiiMV%d9+#yDoS+^_&UV3<&ks<aP6)UI88akOiC&*6pv?=v_4spfiB
zZHcj3^3Aq=HXLu@X?+Exz0%QCp#6Mt+w=4$IzSX(q$U@pDI>DLPlQQdW0GIb4F+eP
zq-^pd<;tg*!rMjM9SlAa65BYe8w@SRmM-mwbXU))=?q3<MUeJuWZPYVCB-_j?b{N}
zxUme5&vd@d-aX!Um9JD5e${<W<g5P`=*{PY*&T?%0GbKc7B_h;ti=VIa^}&wf#v@J
zP^u8a`09_B{R_BCj)X58?gemfL=vPlx+P%NqJ4v=jzrm}NAZLnMdZG74#Zc5ZmrYo
zNc7#-Qzu+FrJMTr^^LKx<LzvgC?O42E@=-|m9L@7LOvDYG5Xzn6Mtp?hEBuaRoTo!
zkG0^n6Qz+v{j$<6nRbS?c`x8EfhUr|9_FILdH4G!8zik=9KCP^$F%qmD5BDEd|mq(
zV9+1^zERfR)Zv-sg{WKyb4VgOY_jmnuUpj(<MPF}H<p51Rr8GmvHkE!(xJM$!L4s7
zu5BwI{BXPrM+F3Pq7_OzL}}Z&Pf?7lPW@ORRkaE61m3*f2ewhim_JpW=xb4JuJb14
z*zom>QKl>A+v-hGdJ70m*7JzU5bO=B+@ghgH1?o4${G1t(?Foz(<aNl9uP6VxwH|i
z0N%vGBM>I>Sc31(82>!>8PScl^^+pqy~g;#gAheqX5$g*LeAptv$rFioAM1m2_PEw
z2jPtbS#qq_M0QyO(6y32Mk8t4f#5G5yPDCG$X^Q+js$7VjStzs`EUdd|LI_CElr7M
z8Rzt^>4v~(S&S~^B=KR$<G*<f1d_2MOL+4pM5$Nds>L=3B@2Cq+M0*z4OoWiE!c0o
zEcwI?CCxiD-@VdabvxdEJnV-JJ?xJ%J?QU>4Da+jUyf1cd@>Rs#>1fJAA{}lGZJSx
z2Cx`3QVd|A@T^QG9VgrAL>Ao8xdp*S7o5{Cdnj%w)OIS=W+~JjT0_n7kX~;QN}6MX
zofg-4voqU&f*CO~d`mDcNJvr4h?oudvkN97+{vzkXAlJy2-4@G0Ynd{=h`HIM<MVb
z7TBW!TWGR%DorUAla`L}Ix}zr==j1w_6y3OHy@5lffjmO+pcOj0w_;W_9VSyt7?2w
z+a{}SkR`gh=dwLd=(m-&gYb0$_1Jz=tjmWn#nW+_0rb#-=Vo1~j8R``AGFkpogn-M
ziJh!JX3(+ZlE|$h7ChX9CJuGhqPzQM#hsdOc`et))i|nkI4DWaJ+o{vNa$rGEOuCI
zO3}Ly(=J~o3!G;OqEPs;JYgLsPst@c;ddQ1{Opyx`l6+M%N^gd8O&~%uoEm5$;};+
zI=8tMTz`T-)vsh9#xR=T-fKNx8QL$<)<bpIf2;#6|CBc~|Ld)n#{Wm)a{Oa(W<~I0
z4LN<7Sr9*yJ~u>sO>p395E`{0mp~w2zWR^B+2sZhGVY>?8v(%A^GDzeibE~p^6<1;
zKKe7v%Z6!IK%AIA>M?NeU-p*#QHuxiT$t=nj@Q#eOut9n^gJ;D(DuV%+-`4P*Uw!G
zNl<Ar5fcM^TRX?sv(3fB*K_;kum1M&(ZR#T_VQ>eFRxqMyWPFT#e==gL(%?Sj3)*_
z%03bDnz)IY-x}HmWwK?3`!BtrLXg-O{@-ONrB=qvi{|wcW9N2_Enf`J93TPA{F)A;
zJO;j`CI+Flc5nuJ(jZN}V#Od&`v|p8eL#gbnN*QtuJxOWBCN4Y&8?UlMu$tVu61eG
zACzV79t!W`n}&;2>lv|3sjZ*VZqLpfe-DrC@T4iB5_R!<1zooqNaYRC|F*K9M7843
z6NOqN4z(yK;^&5Jk|GqE{Y_Cw${~jz;5-_MJX4FG&Hz(GGMz@c3Ddw>1ARdsbJI1b
z!az*m$b00-M#W=rG&H+2N>`GSJEFgC5`sVwD`P5fz+Eogmn7p?y}$y}W$cR^wUpz0
zy0>Ju3{yHZzBK<b((rk5U2Ji!=ODpBL8Q0QE-{7MrnLv2`0%piK}`wnUB#3}NI=9!
z>|ABZ%o&;!q$~!J%V_f)PaM>{+G|*&xtF4!9J}kK22ivf!)m-ilqR(zL?IMeHI<YX
z;zte7x}Hw)V<VHWA5$PJE5(%`5S!o-6j(r5JAj9cttrORLkxYFY7CK622us%_wLC8
z0N|*BXUs!1$!&H`v<BD^0Z%oA?eXl9m~mEc`2HqK(_p1gQGfu6a~T;qn8uQ<j7x-z
zGe1iDOV`ZXQFdpoM<iZ1f#WftIEBK{dYj(lcqN0nnDrO3q@eR}x|H&wJ^aGu@F>+^
z2G+X?+q#>v*HY-(qLII>-GIbi)iUkbJX4<u7YAOJ1gK{CCjj6H_(NR6ExKe-u#r5G
zkO#E~b8fntAYloE*_9MVa)&VnmoBYohlQBGVPL#1?*>V#|Juyu$MoX8>l0`A@(SPt
zQ~3g-7<Jm^yR~a4FL^m_dL+>I^`Q??J(>+n;eEViGAQ;_HPofms`SHPPrz743qgZ~
z{9=zu{98~UdRtMFE(NI8+Z+2Q8W0`1euM3^chYC~+G;`Kt<PF827F`mraK!$3d}ol
z$LDLJTK&Yo(z34hi3JiBI$5{#Au|_L9{wnTe4y11s^W5X5nJqU1Z4wrs+bm-0*QaB
z8V@Uyy(PfCncjWfswUO4QCEQ3Q6!xEq^04PBTEo7=}up4gI(L>(bAh+4@-Fyiwqb(
zS&Z0eh@3Bkf%FY8i}3tGVz7oN&*ez^W1n}rNBu+X@s&izOUxxB?3UvZT~0E_N9*&x
z<tnnzQ0XE&Og<i%dxK!t0>{MZHEVhLMG+acK;8x?(M9ld%KDz3Bpc8MM;O{YPwoRJ
z^MXm6)Z()zzaSoU0SU%>n22*Cl%qyU9$+l;xYKZOmXywAK?olZrlPP?h11>=>}IF0
zcm%R0Vi}SNhMYD*mVf{zo%`$@hSh#Zne@qp&yhd1PHt<6X&2!iD7_bmSkOy#zjR;%
zdxG!3CfIch-WiFUvFE@9<D#l+7oin?W$wdc)A<CH=m7-nFWr`mfuqH-WV-+yo8mX{
zN`kAucfKLxxod=OEboH(6n|KaG@gYUyv_zX6I)*&f2>=4SLAy?2ZILH7XbKTaF=mi
zoDIFgXLVyX=Q0(M$iR@@Sa~^&oyBS%kV8Mw*YNSlc*zHOij+)YBIkYOAbThmN1Z*R
z#nI<B*dQ5x3vMLo${xHxT`+1I#}?UA=^mSA2n6=}Pti~0!-#O(j~U4Nf7MgYhNC0+
zaH$lR(n+jKDJ`V^2124TSYN`4kt?T%ZH|3$R>hJ4bp$A32>PvZMb4%xDhK6t;rxke
z&mvfGJzvkp3R@Im3!UI!6IfKX{m4~>d2!6mbqz;eEKO;$QVZC=>)3`wAK>lDHBvkz
zNL7unwfPlYmM%#i&W5~h0QQVyPdfb!MQmD(sS!Aowy0YAu(MV274_IBTQEv9{mRpl
zqD4Dq6wwBvr#1t%-EA{aZ|9l*dMmZn(BGy?RZnPb(00ufvJ_+IS!^pOrRCcG5KX{k
zwqdYLVD0yew}txH{}x9&6Sc58z``-Ct%xT->8^+;pc`>+*PgZNjAA{}{n4&d{I#N^
z&e2#fG~n>P6)eLN>1mPx=<=j)vXY<RrG72>KzQyv(f_vTSlZ9f$rH-sxQ7BJL4y#y
z<=C1vxUEwt)dBpWcD9t2Yp{lDd{W?b4PWW}+CXn<a*cB=)-tpc1&zI`zy^BJl^p-%
z34|5KT|@FD9B&$}i|$~F5%D|=6V9EcSC@wC*|Bh>lpaWh8}$d$5ehkgiL0LM$^yd-
z)Nylr7CjCCtt-E*v7T|o{u=uEwW#@-99ZK004J_7k^<r=t20UrRM$6UsPy`~a+v@I
zFuZhWct4UtD3{G;PJwt)=T(s$UEdD}BJR?c41c8~?vqz<PhYTynnD2=v&#Xqpb1#7
zF4<XIChHTJ0n0M>A4)ze>;HGr_5W7~#Pa{@fLQ4N^=8mdz#fgs3iA`NpR8yzAqvYQ
z)<DF7|HlS|H=28?59f+5Jfo|10W9|I$vc)_Ve*l3yK<_BmT(j=7Mo2b5v8^Wie|6;
zK}k}RpC&~X#om(OzrVpz7H=v5DP5p9D9)T)<DTNHSz71gqbzob83cl-5JfYzB6|`&
zS5y|0)&@ms%8kcw-lS+3Tc4`3Dm=Rstj3fQX09My%i}>t&kqWQYz9ncOK$;bx+Js&
zZZSFf7MK6Cn&Kctha1PHZyqNw*Wx1-MYK|o%Q3g`perN|Wt2kHd`h58e76Qcp^ua=
zJ}60*+mlbOfh!`+FfQN}N+c*;K~uY`s&6aaB7T#UG7TO3hk^wtsSqZ+&_ckHI$452
z%P{V>8=h%lMFdE_kPlc`Y!X*?au!QV-a!&Vu#YT=cu6>Bfkb?iz&f5NXuwk6K+sHw
zULf6M7ghuW4YH`NR0NF)xiME2Vp$i~;H1prO^<51C>yCoaiSImR?f`xfQWgWom0;o
zk&To{Ru0b;Ge?>!Wt~_+K3$bg%gP+VgS=M&zbADyhOQ~iB%)WT7wKfaMnp+$!X{o_
zhhE0i;rF^7ym+Lgwo!4YB}jy<@;sbLDOf*X-dH6qzqyB>o_gIR_#UMTIQEVS>?|tu
z)&bs;a195zL7OilnW?-J`|1%Tf`9=KRt)=^T}{|T<PSDDX$U)p0D5?Ea4^rkwXMq@
z4Xw%@qgVYM`ifcISjQzMA0e^35gr5#bBP)#lYUtcI(BHenvE`h$huM=-^+A$t6Z(y
zlTo6Anb)=MM0>M>r$cS4_0Mvr+wGjAb+XRY#v)Mm#9s6FSN%Y+vo|%`roC!+M^}5}
zL!9MDo_lA9*5<Z{>Q<G4^!dBx>+O8INjdG+91K8UB!G(S!xdOQb@%}D&FSMKRJ~S}
zLi#GB>^eaB)#T@U`_k9r{Nr}`)#};N{#}t2PE}iWyH4A2axL>j7Mv<%;2v;aq~%4v
zC9(9Di@3UH1!zk#HI%jH^ZeQ6m?P~5T=Vm_HT}!{*_!C>;4tnaC)`noO)@ncE8N%D
zaP|n`j)&v&b2hKeB{;rX;aTsP92P@X^$34{&u@=C-#t;~N0$Wy)odkCoJV%25T)n|
z_JP{f3W~Z(3M}?|wLmH-X=U<vQ&N8YYPMy0C7H!3h)4E=)CvkNUnTAsb+0EtNE_sk
zjLS<!Oi;hyI9WJ@%vl8X(LjHmlnpL+Pkk?%v%|qLJjk}O_vofV!N*PNn~E%#mRzv@
zvU01P(h){&U%kV{zJ_nv0w}Wo2!G=KViPTobSeJ^ACkxtCjlULDUXOMzMK$rU!(7o
z+M74~o47*fr>r<F)uddNO}~oz<QLaQgu8eJ*b6RO6*k<duU*ubK1eqBhS&Ip#-1Iu
z?K97~OrpdBNzYmHHyq5Dy1v=?wXLY%t9;*U9a5;qE{%@R{waXM8#x3sKs3|)1Ba(G
zV+;0ppNCt>|221Os?SL<6ZuR)&mFG$RkQsq%a*Y?<P50UiodTZJxxiv@%Nv)YI5xE
zY#Aoej!imZcynYG8<rI*RuXtQlcS^(6SjKR7RGdh2Caw{%6_?s6;1TnfXQuO?6nVI
zyy89LaS%MHElD#hF*VOMrEDue!AFeb4{o`&3DEcPd4sVjZIYAYh>fj_E{!-b&M9CD
zQd#{mS3duX%+d7bA>WoZ(L!6i!i#%0Dhd9x*YB(GVl)SCfeE;`sCfL*kDOx2M^{|K
zcGN|Ov3<UFewAF?9PVPVSX;a>5$ov{(k)(aIgXytRw_0@{Ao@o*P_?|!`M3oR~EJ3
z-mz`lwr$(CZQHi(j-7OD+qR94)5*zuzW=H3Ty)iWZ}!!$U29eCwZ@#!9OE|%Ph57I
zt~yCr>A2Bq$Z5Fb_uS!4QqAsJ9~{lcu27+E>of+2P@(N>G!FJr;bU-E?ZpQWIs3aP
zA;n#c_89?~h<T47+q+#y0WuVnljIl@n>}jIUpo$IUJ`G0EYF@lYW`k0p@6g15qqt?
zXs4j+U8na4Et0L<2OQXrt2D%BlB<4MRz0jj-nRKBsQx-izWZI3B&!4iJ@dC*>-%U!
zIb)_wWJl4*Dl;F=Me6>SGCnbwz)H-4vb+yh#?yv#nHIriYbW!pyL%es%}wByx?M?o
zuB0;ri4E|9BZ1>HkLgIn6;<R+cf-5(FZ^a-^}t{JLO5_w{+K#FkpB=c4vzm@z+_DA
z%v~%9IDd$K|E(qLIAhoT$2&Q?`$qkM(K`Gz%i(8YrNua}R>X51UMZl2MF|3;sao>)
zzH=yW1e7VwROUMGYVGt^*Xd>d!pPdn@UrpIaf!2&KtmIXGQDHvY{#gUv^|;ueTAaA
zXAOlossrw&-q)A8>jF0C%X5_wYBT+-h5rinJ^!zh;au;tw=i#S2j>g8*E=e-&NkM2
zy72dNtG_cWcgctCpYyArCMyUaVj>EW3Sxs$MeuY%ZIFE*WNi?c#Ht9=h;Rt?6$`NE
z;_!+~2D)n#zU5?>o*Ym~KKJY4M(Gk>eDFh#$fg!Om|_=IzP_FAxJU#uPlS}ohXRT7
zM5L&SQJM5hUVFvJB&DDT@@eM9S4JuJK)E2Ayz&-5-p~l7GO+|2yr9}cBb(mU!DPIh
zb3+_cL0=0QFA1ubP#dh1^46ZmtUe%52DFbV$pn#)pi-EjF*3(053bGAJgOm{mIf$7
z!aQDmngf&tQP5WMN0(Gk%d-3+h8X5owRg(sv?f$4U*TA^9?#kfh%gYdG+=un)a8S2
zy5wi$pSE)jN>nvRF3R!*$(CDI1|t_@k>)w3ARsE4LIA2V9U{`U(n{!SI=I`i@hdvb
z5N~P8-^1lJvqv5%7osC8Uay7WnhCUrOr`Spe_k<U8{d>>qd&7RJ$G;hI3Mc4k)ts7
zHx%_?O0S!GiNwJ<wvJ<hYo&)NIhl3v#xotu2+hO9ymK-eJ98Y<%ww7dF_$tQ*wnxa
z0x=2;Vzs_M<`hVl2rYW>3&ITUWb{bcWQ25$oTO}E(#4I8b1Wlb3+M6J%C6|xWCC!z
z2COaRYZApSC4=(;=+-DV7n`o}CR#PW6@b7o11Bg0jW>*!TjQ8xqVU_G%vB^wN~#^i
zA{TOwlVQ!@e17=gQ<^>P$29&Gr!fK4@=S}z9eh$y{=$r78IEN~GT>ey=5!{xy&lV9
zIhgni^eAsaXdKi)+z=|E0vC`XkIU<ITuGI;5-LPe{%yn!{6`>rCI>^fztT`*!x1Ff
z!$0O5O0L{Uq&NqHavO;;>Ef>>Ml!>2$LB2Sh<|P$Hxem`MA`mkfaGfuTG@FaqEABM
zHl^g}dnD5wiwP{k#k>nL>$(dN!HvWRyE>T`+OQX~dYM_l#H0fm`AON8qr?rr@GK={
zRf~8nV3u{wbl`Shr?=BW?)?v@&*HiG0c$FY&j0yj|HJIb$@o8q690uOdOKtC#}$nd
z_TmXe|3f|3wYAK&mPVhGoT|KL0K+-JG-lo_Gc1GOHJ((6jvgE?v{X+Db@?iuEZ+sw
zCWh9=pQY<=deokw2~82ko%F8HFowJ#m<m0Ht*L{Zuz2izdHYN=P`x-&A|VjkIlfpN
zM!!9&wNwtoMky|%#o0P*rRUy$ewuEv>%D01R_ok6-?==!T<LaS?YQ!)k^A6N<`@)}
zBM{9X6`U~4Ov$xgUlLW~)N;gnO`M5B(z=ks9Rk%pLM2~pz<(Mrv`2~co;P__m>Pnp
z!yz$T7|yWQ*ef?kAX_ZEHrUFzJV>W28=uvCTRnvr^-h!5dn_I=)^A4R6<lnDG>wo-
zRYd^)t5c0%@}(7`$$|Be05(64l^vXzlt9@UF%05_($H=sp<v}{9C6AER^KkdT4Yi}
z?FH%L%VdkRF7fWiA2Z-miybb=qs>QxjUIs#CDISqK9!QAZZ6p%AfxC!Md1T=AX)^>
z51ZUBmCl8v*u%8!llv6ZP5&%}e(RD*do2x}fs5u&YQwV>ISuBAqGAdv9wr?`k>+L6
zV~dTlFCkTmHVTbO7+A|gG7HavV+)1320{kqo%PaX!PItt6rQ*n4-`8^69X=qDl{s)
z{zKlR2Aa20!w7WljZMYZHXy`E8MmG{La%0kdzlSFtu0Mcd=3jEkWnunysJtgwZL#^
zWzjSRggumli`5Pzv(^zxeyR-vTfu2JFFWeul9qxbT3{VTnL8744-lggNntR?fspkC
z)Q!|>51p<DkZpA|WvP$G8HZ+RZ2)mEsv0|BuqBlsBufW@eWGQ{47rzRVvsn{ycqrE
z_YgkF5ZbaJZ^?i=%O*??;N3MaaX**%Q`ULhN;`)M=E~x7DF?bo0Qc-DRL(@=?m}fe
z6)0yqR5AN&s{V(o7L1K*ckLB>p)%!wPMQ7%Tf89<YbI*YUQjwN_1hU1PjtUVJrFW)
zt{7$<mv;igxj-C~@#2w!`Km))5u7+Y(to6QOna{eEFSCQoW*D7131c=<Q~eI)j=#p
zC?`tU$xPyAER*p>U?v9R$<ZuK_Y`J9JS4%3V^T&+EjuLuXS_#W`D(0=QZ7-U&PvJE
zVDMrv461~IlakGJT(^NTo)ab8DF&;7alo!GtnV0J`E~c|Tw?*FD;UX5%e(T`=UsTC
z%%4_|%=WGAw5aQT#t(R^U&H=iPx!xj4Hy~O|7$km_K$Y|SdT|{AJxwk^~}TgMWTGp
zgrhh@!&xx7U?n!-0*xA40W{fYNM!@Q?zy9qh)0~R+Vk3;NJ&f|dPX}x5)$j#pX*<~
z&U9T|e8lP>P?Un8=-)3G_wNM=a2)8Rx@P%?BMSBMp|9X;{d&7FbVLS|A0P9IzFh0%
zIiLA;^5AB}f|a>D+CN|SkG7uW!~I3@!*9Q0!qU;Zdy97CW`S1Hpg0C}0?5@O8JW)^
zR%bjx454_h5aS(5Q0_SrBS0pi%NY2x&h+v@3-hN(T9aqd7CD%G_h-?0k=~m&U)iP?
z9_IOET=bTJU$3j4A`wm`9i$izQGTIP7Dh;&Nb$K{6&;h7f*@jcSdW}{&tQv-J7&&2
zpu3g|i8!u`<xG=1*3mOO@2;Knv-Pf>sHNXe);1^<DMXr^Jq}fCS!Hh-$B{HI-wspE
zxTH6w1Trp=)Jlvl(s@!Ps5EAVS_vy?q!#wcoIBtB{2Acw6h*J6lJTJbesD}5XeZQB
z{gQZ8+zxH7r^_#pqpX5Ow6-WL#H#$2FukH8W)NTPIHfi%o3$0wbS~@0F5dwS9L%Jt
zc8NZ9-;I~r6e`<aY}>PGhLs-OJTpI7*J?Xki*&)q9(jh02H*w7g~EuWL7BYmW!F+%
zsSb*pLxBBY*AE#40675<8Atq^lw&3fx8Kp`3La<mii!h}8I$+}P@IrVm6<v)M8cmf
zzx-{Dh#~7(6!WwccUUqh4S|}Q>d8o}+Ax+dE}!rrz6mWvBGd)e+FC)mh%4BQ>^d^3
zB}b5rG|igR`m#HY1x<+3*hyjmXPj2jdF4tX%h=ez<hPl+q%qD(KiXF+wUx9D#!d+A
zb^xrX&b+X`VjXDsH6l)>(9qyr!X8JGcF<w<IwH>K$XB1^bZw9@srZt|A7@=f{wXZz
zKrFX4YAgJ2m@fcp6dGqYg_Vt@&BhhICTeS2Y;Tik#?5{#LJE@${=axFm^t~+yjJcm
z5Tg;wo)RuUmox9UaupHxb2(@2X6o~VFjDb%-_xv5R2ymfZE>?QjEE&1<+N7@+A;Ib
z<?N@)Yo~nu2v_}sK53U-GKVzY%|MHqNq+qq7ejH`LM(qtXaR~hJIjZ-1qUR4-BEra
zZOUfe)j`ag3oZHY67!*e-QN}{qgdu_o<`D%5mZ8R(~x)g?YNbBHI@JbZB^!s5CeTR
z%sE~vHvvH529ZpAv;WMZwdP9RZObKtC{CwSs#wK9^uBigmMubwY0yyLdKn7<0*>Rp
z^vb$?9i)|t0H%0)Pph))lz_%D)NRnIU!rh;1D;p6?+_3Gl{lP5sqkcW4U8w9MCMt)
zD$k>g$<iSN;u>eONeD%7yZXvOSJ_5)S1Mr8Qdd(c-VJRr?k+eP`R84DVEV40`a;c{
z*Uq^p|H0gv52`IGKw=@6F*0s>??$ylb*{(&jqH)%3~wq1yVaGUne05f3m1NCT4|BG
z5U$~dO(Cw<t2PP^DtPOLV!>Ff0*}YG^_Yudi$Qbz#a&E`g5tggVS)<>A;Mv6KgJ7_
z?S_E+7q$=^S@fT8x_^Z%SlIsSH{EiE^iRJ12QmJR$}^FS0Ek9^Fei+}UBj%*9eb>t
zbU&AJRWB@6i6_akcy50SL_nDkhSM@`l9Mc8(XyuN^HE^ols$K}Iv+iLwocuTEF1%r
zDr1LKi$sZ1C`6WQ`Ctd5gX5dg<@X$=h#X-_fW*p?nBL3l&&=uRof1@HTyR8@8Wl%B
ztCypnhntV{@i+a{JbAHc@^KRS+sDU?(bN0=z{$zW&&^OCV}7J;|FlUx43xZF3VB>6
zDTpjnaVLbmYS`6JSMB&@rYe8t<nm_UjGLbA*yslL{C75#5mQ=;41yfCX*qlBD6gb&
zUo+lnjX)ZMl*dR$>X-zWu0mh511v8tCf<TXS8PTyVr7SN-pEu!$hAuK%Ua<d|F&Sr
zTmunZi5j>vlYvO14Qbv;jj9N-@I`@(U+^A$`6D*qf*}Xjzg3q!g3lsv4Z1jnJW!ea
z9v#{2+cm4To|s-s0S08v`*oLKWn<=l{7g560V!B6NyN1QE5;<-?&SJuz4T?~(W<%X
zofaXj0Ez%Ll`kp_73v_(9&4161)bge#^_U!o#`aNUa;ZdZ>?r9Z+I1RnEX1TJS0>o
z{SgdZcT+L`k-v<?9R6Sv@2rTFJG(Gp{$(|)2qB{}%rY{X5U~(0F_Tn`-|;ONN9py+
z-TO@@&Bs08!QClEC+06gRj*n~T0kWwPys-A8Q8z11`&Ao(L_HvVz#X;s6vKJV}*js
zDhD*wH5swOCCQq`-@VFR(8}P8ep^j7ZL|aH(??}C7{$tW$x|@TQ<kAlX>3t=R<)_m
zap_THu;4~{M=}9~MGM?oF9VWbg^fcGZUGxKmVuLz^2B&Frh+xrUCwes@)dTC1a9<3
zjq4x$xy%dLw}3bTNj!n09>qXo5m4iRZjHw{SWuX0orbI4v@DBA%rLn4`b}~-x)oRR
zc@eM)dF0l%81MVN*ZBcCbX^bRAOR#vAS1vZ3>M}XI20f&4wJ?5wPbpvvXp9w-ROk0
zIlQML=wANLXyL!%R*9mQt`dH-&J_ga0OWyYfWCZD)qc)rN5Vw~CoVz<(~nVxGq(m-
zdFa%wnYyaY4$<-4vvZWK$(wDDMT{wq7Tf5{%7<~l@rwfY`|udBYE2F=a5t7sBem{g
z_>a|3L0_I1Eu=Zu)%er%FZEtJ`^zp(bx&X3Y2Dx`6i}hUZc7@rjCHr7%q6#wT=Xj9
zwAgNM3~=V1BfC?#le{L!miDJ;(wj?12L8=yJHJ-={K@Ij3}jtzgtJ(zjJf!J)fQbj
z`|5mIgzsnk6>Ihl09HBisPS=4<pMZ3QT`m0^x_$feL56Vht!P;6RfN}d`I4|lyp@;
z-Dxa}3D0?&1}PjNY&==*?y1Y}wWG(cs?m3a5w#|?3_g9J$^0-519j99HZMUGh<b;)
zLbdM=@n(PIestWZemxIVzfJiPByQ!jr@HI@ssY3im$idb3~gLaH1zY|nLaN2r+5kp
z40>Cyv9ceqUcDK_Yo>at?s{Osibk$s>r2rrnWroLoqKZYx!>)y{BM@{cU?sG(ZDZ^
z_0g`0#MpN48>HCc3#59vP3^F<A18-s)YY_p$Hj=ENgd;j;WMiVCYo18#8J;dA@(<B
z^b(uhB!<w8@p&XMh;FFQ$cZ30;?|qRaOC6P{ri<~M!#uJ+x*J=sDfS17e9TtXxS~4
z9_6^TmoINa3C8LOA>fHAIOAwP5NHknus#+Z+h3tS>PF)&o0ZNG_>;GTVcNVB_MO2o
zYol$uYaUK+V({cYf}y2czRiD=j}rQ^Rq}=dPh=p89Pr%0K;M_p<~8>>{l3B1m_K7*
z_~T$s<5xM8F%GDJq71r=J>W&Q;YF5(MDp`_v&zg^VNO(GUKEeT2d^-g3(&IH)k&xi
z<{XG+Z>A1$4p02EGQXvX65ry!m2G$}d!=6x+F1C97H~5BBeTHB&i-E?sVr;!?^Ung
zi+*AEy2kWFn*@xv#eA7WTdD4WA-p9bMeBh@s;PQXvIoX<+bx9>)A3vbv<U_n|Cx+Z
z`JwwR?rlV|XGpPo@M?1ZByx~zNn)CcSaNp6`2V(A%$uO;!cmIh#Eu9Z><`P^`{T#{
z>Kkl=Kke;$U!Sm-qbEBpuk+8&@#Dbu&Fk{_9LTJT|KXB{%xb#IWQYP^(L_KDppg79
zupvoUBS91+9B^gP>iwLu<3^J1410344LyzXUAdF_s?$WZmNb)2v6+X$qK|1lLKtPu
z-6R`m&?W7T&z%}37D2lbQn1;Z^{O9BW^R;>u{92^04eGPpz>mAo2n}(ue@~4CkJl+
zn(4~4eaD!d{BDTnJacA#bIP(=rCp(~R*+5RaW!JoC$Zz8G*hEU-jF>VKk!P^@PKR=
z1MEq{_8G-soG4-p5Mym2=*4RA@JKiDbn0H#m&uhtRFp$DR6}OY@(~@P;%d93DBS75
z_|@)8Zr~4UpsyKFH(`XgP)!=J*8K<xty3nc2l%>HXb56oEoyAORs?XH3@{%cA!5rr
z8Y+^-iPkM0y%acp%+L^zyOTkf&wk=RiA}sqYP#~PefC0|+Ow7w3TriJu45&_jTyG7
zPi$6XCb487%OZE?aOGU*_w|=SIQsYq)FhGnLW_1=t<sB~+M2m8nW!u9$R;Ff4ymhA
zdB^P(F4@iPsw<GflBQNmn$oP27$WhImKUt{mr`a<!P=dPX52`OB`}roU14)p+WG{M
z5txiZhPhH4gpZ3Y&}=M*bPNEqRmRF?V}0etG*u0NIN5l;lyYZ<lx@lS5g<Q_h{{Z{
z9ACJJ(|VUbxOBSdGr5f|c@8IFkA>;W{r%^aE`LmwpH&A=(P?bE+|wj-1cOilT98+v
zIHpm|yJwsIY=%KZP&EB{_wwdE@{yj?gZ`zjGd3NLrKkZi<HC^W$EcHMu8z5<S|<#f
zbads%&yZ`hklilTaj3p|Le5_6>N6FK!1}RIxKltB1VCW~CRvw7dE3%3y*}gFg<>n0
z{1ng`Wk9#l6R}Z<pmEs&p9NS%^3c9D_-^e;yfl|kBr1YX5{Kf4_mWBxhQsiz<~b_R
z=P`j8-h9!M4N?PcHgfMAw1xpfZs&a?xa6L=?+Dcxf_AN)?fbPnd#!Jf2c|;sHy$%^
zBXE-wnZle+gzMBJ4Pv=|nL9Pe8;IhK-R3-=VSfoNJ-jciK<wZ_at4drjdN2Qp*RCq
zjSvs9JRTBhNH`>*GZpU(qw*4>C<F7kHPy(O0)9shOds5Kf;6Ck00(VU!T?VeJ$Q1K
zBdh<&JNXYl>xbDnQBi_Hizj_N9awM2M6-WiFcai&z??86gFfVWXhI6eI}r-=O)3Lz
z8UT0>v<Ly6(au)Ku9jE@5zO1o7519t<7_ebvOLv_7(xZT;lvFyr3&Sovi28A5<~a9
zzMN*bLl1c_f&{nGzbzQyy>Y{U<w})^fg0Qczn*p0;<d}6f+HJl4$)r5V3k13i<BHY
z0i}fRD-~ewVvYOcd2yRXyhtfW2<`j5I<9r`fHVZ$kp~lfVfrbNviL<Es4$9BUgq23
z-}H#&@ca9@{DWv29oroNE&9qacjM`f<jw#IpMO#F^)!Vd14d;wf^=25DejK*PMhz2
zjs|%Sd8X*Y0CzS~yI>7y4RoNSq_G!XGjkBS9QjY13-2-%TI|-7NYKA_%J2G*(vj`y
z1mK-$u9>Dl+B&B!jhe_up6^^iinktL>|c3OK)O`&3<}Y(h9<FLn>PHBfCn9qyUS2j
zU{+i%2jnV~2txz03C8lBDmd$-3)$i60I#sJD9M^3oBJD9c^GSJ@eTL1I)vj)OHEZ_
zsN;^iKuh31GErxU$F#~$kAB{DU6}0S`XRD=<?5A^Ei~ckRX)7*Y}RUtoCRyWI$ApI
zZ&hbHHB^cDf(606#)@EOxyTIuJa?=l8E{-=AJq0;<<p(Dox#iX=c^IE>97o<M1=+F
z$G%WCqd7W2Ah@-o7=qo40i^b9F9FBu_ll7PNdR_x*<F!LG?&9SPV0`60iBFcM*zV9
zd3Qktt|2^;8X}~9^Skx@qc8gm;w<(}?k)^|u8}gOD_W39RaxAg9*lA|El|v&c6U#+
zjXBjeMQ~&1@XSwnv!|&c{&}KSv9o)1-c31n=2=hNJ)5bqQ#J5y)8y$4EfE6pG{M-V
z>D;_iJwbfu=$aTb8j+M{Eb0kQ%L1tUZ?AN8!{tQ*WrETR@Ow5zF=6w;H@+c?_01G-
z&)P=@88*FoRdvQ@BIOb27)nbgW5#sj3EF1srRDDmIHc<_9x+jdL_yg*_~7lW_kRf}
z<3Dg2jO_nCt#n#L@@F%J(sQX^&u*nev<4|x9!?^s^I}lm>LbOBsa0gkhEy7nqU_G~
z_u6-jA}I{zw=6j?B1m7GpFMHh31Xa_I6ZmssXa$DQ;H;2@wn~(X=`GN-Wg6AzyBDY
zo!v#qajh*yfh;hn?Rq~S-W<Q2bQA)OqZB(7k-*W<@9k>m<M;G6dePU-k)N5Vqpz*I
z`WRndK6<!1emc53_>^+SFeFE=C!$g6lVX~tV(dc7B1)B<bW}cqrbZ=gc20E)iA1mZ
z((w6Gs9~G*xi{qMbk+JJK^<d~_73ki3}ij=#F2<G3jcgx?&|{QN8jt;+@^*0_<8Tv
zqp^D|kgjpVD~%jNSwHeq!34l5LGmN4I1HLWEXkclISfQGcwWQCRWaMSsjf|w_N}tN
zE3am*TeIHT@#^fZZ-RZNCO-ypY19^dUQ!jq7_n9tefBxLO#is)O*da5-WN?on8QFy
z)nq*B_Rv*J$2QJP0pJ^(+3`dm7Sl*Y2(c8etkI9q#NIwSCGOH}y|`-1ZYL2D3Bz3?
z5q%>-NBbH8ER{cC2Lcw;9Qv|2{K#QZFc=kthrJ#8M!(8j`#m`tx!18{(dy#0QzwA`
zZl|{I)>$sD9Sf&)nnmc-;QolybNjr#)-53&hG+~zI5af*wTwMnFk(Qm>$9?>aNJPr
zdwRPeWzn520*7}(GUCv%e{{|e1~;J!AtiG+fU@&X<9JkH^|mmoG$c`mi{fyqFe**V
z$Nu^&ayXSnOGVI=ON$PDl&K=?+eyEzYarMya6!KuxCO2pyf6l1p+cn+wP7^>)?j{9
z-u>4%ng!jWZkPhtFb%Asjq-H9PDLExf+D7bYfb<&MH?Jhxu~kmgIJZEXwGH$pS}Q9
zXc&&JS@zOS%EMVxXY)4ho?zcgZZWOh{v9h$vIggTYVH&1TfTOql5fHMXpc<otv{t@
zno%Mb1x=M`SXnZ|>{$Sl2e!IXS8|%`=1t>O?`TIa;7wiQ*QV5fHq*J!H&2Ufby<y+
z(Hn^1+k7h$>T*cEnzmQ9#7xu|=VdzB%is*<QUsOV)+XJq5Zf`34qF#vSbJ+ZDAR6-
zKn0!}=q>Sq1H{c4UV6mSki>*V(>zwI>E*>|xxc;O)zz~f;s@say*>S`)cNEt^|Zm1
zit4brbK~IoNI*=5=Hw6*!!V3%VL{L51@b}+`u?ak<p#DOnb+xUobTNJ48HP!10W>w
zi()v=ckR^J-(-JNWP^)0Ra`umP8-DwQ``@67_GHbn)ev}u`@HE#-C2@Eg%BQqNWX#
zL&y|tUEa|%q`aP2Xs6{$&y%QBj<!r8v=%@gCJrNU9N>Y5pjCv5E^)s6NEiaD6q~fg
zMA0M^eM=U-aa5cOdV7>3f*%9E#!c^J>J}%+jIuWs#x8O1@e=&OLdb*rDU3F$1%-Dj
zNc5KJ+EN!n6md;qh=+7)9S2!9Hi?M;?BgO9>=2=>LM$fWsX-#^p0?}>PnkesX9)P7
ziy0%-X-5$1fRQkZN0Tdp$pDEoTxo>*)~TZX|D>C?VNS|u0aw4Szq3d%i{eayTBaK8
zu(z~bA0{W9eJTcF))O|O@`6~kOKWfoCi1?!f7X{boQ*&Dz|>TM-v?l*08lL!o`fBp
zJ(>KmJgOo`L`lF%WD&NMDmr)ibHf{F=1_(W1m^kJ^!s;AL28Kzu|J}2tY1SfH<(@`
zk|V_^$*_UD=NKc<h7xMfcRCG!-!|Om2G&&q>k_&Rv@CC-0RqO@V)bD~Pd5eNyrf%9
ze??DW7}%v~U0Im`RFY5=DbA{E0CkClWXSJyAvj!zV~<hPGR)ew1Z(<jV8(jt#Cp+`
zHaZivy&MJ$U>~2d?RXLs4pXUNI0c;WM`E-x=srHwBwoc}PHqfVFEo;E3V2QsnSobF
zh*pPndklhMLMyM=8lVvXR8*H-3@YslTR}^T;K5f=UKb!DBiDAiL^=US2E08!)S~nV
z@YXQ`lX@Pg92f&3IL%-~3S1L}X+jGN7;|Ck`jvdI7=i^;W5AID35K1jK3=D-K1~8a
z-QM@@8KZ8!0VSk(+m&lh5)>#bITr4IjG%MQgadaIj0v~v*`@$HHXc0kS2{>kIYD2}
zMguQ=5sflu?&c97?&xCU_eT&hfD(df5*hGckNx9&^k_r=z{FXY*+7TR6^!oRre9F9
zrwo%WUecS$z7R?$P>u`KP|GW^+tR%ilWz|h`F#M|G=-+BZe%=FWo{=;g9PN3TkUW4
z4!0+5u;fg7*?(u&isUyNM0XjQJOl#SU_irCSP=?U{|*{;=51R&z4tbGO=PtuA=#b%
z`BK1b1SnQu)CLF@muV;Nmy^$8bVSfi@cII?9tAf1m#8u^{=@sv%E9s9`Y5LVuL-46
zGQ`LSV=i7&v)i_C*2Y9(vKdpe2&Ijv6d|Sg(s}$X5&&gT8U8p&tT?z3v_XT0j|NRZ
zv1>?i)A;1u(NT&4GNuVh>bwssBPQfeYY9;neX%>^utd?hj08@v*ZuP{y}yjP%HC`F
z<K)}$+hp70<F20k?yNsA=W%si=zn&jf;mWKSz^dnqqu(Za<J!5d5m$0YD8Izq0NbM
z<EFVNc6ZA?8cXY~6<cE=^%BD7<2L4TF=X>G4ny^~T4m!SEu|sP8%l8uma)|rf0<}6
z;Ge>$?S5k^*MJy_v|N7=0kKC_oy!>E;BTGkznrV0e4MhIyhT(`tat)RjYse|<RXXR
zM`j2hZ4JW;24z{~1=BvpTOMK|+P7M(uhnj~*;&(V|1tA3s_zkhs<9g2c-0Q?xb(gv
ztk=kDs7HoTHGLv%9{$!fU$xwinsm{Xi;P=rhRUuLuo39mI*~KzJ~L~V$QpM4b|O!6
zRHDJ=?)(KXs)H#K=>Vo0ZIaC8Q(YWySARK4%k{I#1;CA^@K~#Th^AD&cTKWQsxtmI
z+K)y-<obc-7SWqJ_(O+av6Xl-Auwf7zKJ0`yYgcU_?43b5;6qb+ahcnEX4I}UpE?E
zlR!t<B3xU&&}-5)VNJPyUo=lujTWUxeVmo*i{2XL=t4%1_CR-nOdSz+!c-G#ysRHr
zlDbG8a1*k^IQ9$SKJgx61!tfc<`47j+9UOL=!{JMXCH5MyAKq+F*viV%_8XK;IADe
zh@=Dqw5p;Nr7F${f#nZJVJbvq%2Vpw?g-7^>cAb;!;-QP-XGDcNTSQrvWQbq0F2hq
z$ap(ORJ=4iUp7Z$`dG7>1wO0@ciK==c=K-O3<1thjrx4fya9v9v>_1}^IXH*9&f(%
z+#EwST4B#ckcbr+a$<LsO1=SZumxuD`5NPQd+68T5jE;7K?wp+;)=kS`CQJ6V$&CB
zzrDiqr<pu~6|Cd3bDD4pAeZN=Y3-y_HfzvB($nD(ydU)oK8RYvtG^{X0J`>WHIVxV
z`N}k^4N@$D(;%O7?;Ba?eQZxJMiF{mU^}EW#OJtGS#LM!%6L>B6Ok4nJi}D|h=z87
z=~uR;G#>fQBLUnm0jLuki%xniCLqajsE?nfwiV1f&MN*6r3fep8Y(3ig+QqAC>CE8
zfh&c?jbb2PTq@Z7dUlvwPZ36utd@FC^TRQGEEtO%a<&iB`_?s_oHHudB;QoJToUS>
z61wCfE!)Yi8Vxen1F1YUq`K}--)MPn2`xf(*Ui8ob%xMhYLJG!xmOqM){%>z1|%l;
zd{2!h7#~&I37b_r<cl`cXyojKx|eCXm!tvs>n@Gk$pO`+a9u54cURRcF2mH0?VBB<
zz3`BT3D>1m`Wdo$?#(|~Swa0IGhdEP#`oFLtLC*RJOVCt7z}_ct-2rXWCbA(P{MB&
zuD7kr{Rq#6AqXZR1>Jw8e<lgf_%=eFW*~!UUa42;`0!sd^YEd@^31{Hna_%%gu*@r
z9?EugNmVw7_8RT}+gv^KF;@i*=vj}?chFh`f47?*N$j!DBKn$Ljlpo@C6woRnSYbX
z%nO8%Yzr?9uZ4xB-eSFCZ_ObXA*?n>2ef4nEvIJ-wObWc`5s%jAQ*%ZN_Oez#bUtx
z;zBX&R{KH-3~uu!boREzxC-Dg<AR1OUJO9xJHggJRcya^!`f0cCSyfP*#zg6<N1|0
zIs?l~jDd|uSiBghle)``9cbYHism0%Ln9(p2_6Fq3!*U)v<{=nVm5|tQ!xh<{2_@G
zXu3f5<!HjQHi_bdrC~Co>+KC9=a(nXC%d5_<cO^2L9)LBLjFjU-{l=&OuEE+pDI}P
zXT5c5KL)5I%P#xO>7mpA9u|BEEagl>p`=7VYujZ-q=?a7;O?MVkviq~HlKz$Ov09Z
zKG5yD>`oIs%WM@A!s_0LFV1~fav|=m+C`@H7DG|}Z$d*NjSz;GKU>vrP*W%zzuS5+
z;=vHjrx@};of9CU5mK_H5!RfU*pfg(@B~WmxK22-f{sc3SctOj>NF6Lz`R3%VH}zi
z`3MT<nL!bLY7acVp8lyq|M2`UGI0La9Oz+n&Hq&K{(nj2YRo*Ydim_8o8TsgqjKAZ
z1k-gytMS@66w0WtXSm@)qV)+fxDE&#fkA7IoZD=|a4zg{t~PV@{q?t46gdH)^iH{U
zw~$MUb0n%%<k}p2V!F@!+x@yz_872SOFVKY78YeX|F;L$uDp11oq5D^6COo6Tv{`9
z_vrQIr_}S_uI$jo^#eOc`wjKQ4+p+(jgZC8_3oYhm?GKm-N)<zSGT-&JVhD@vMpl2
zJuKwfg+{6cQ{B?etuy+n3aLhtx;Za09DJAdqIp8eJ!bO(5=0bLo>+3a(qjK<pBWEx
zzDGz1k1dYM!|-y#mLTLCP`4d>mHFx~lcltCvgm^FOjq7?c@$DW4|kwV2M+5dudLDG
zj5lKEuoDT@r9U=EHumn=#8=s0+e}(zo6TNKxm6_en<M&`PYfoE<tY^MPw~&b+PCV~
zwtVkL_>Z28h8Zd$)RwGBTc~GCS5(=+&;}UFIPz1|OZ3M3v3SKU*tBhsK^Z5EQ8{yh
zX*l)5fCY`W;~!K)a@ChSD=?J16Olxbt-?NgNeXUfEO+Y(dZ-8|p1MjFHKq3^yZyO=
zE33IZ_)cT!B0?-O800~cO`5L>I1Hw3VpQvi(!wXkBWsy!TP5+5cB#Yj&N-$nZ(?sy
zjM2sBuYG%hKni#HJg|Nv;a$>^dqwdtk_F531t(HPOY-8eh!ckJ7YyPQ2Lq5H;2c6G
zA)3+or^J2)EERoijl{Hm1pJ4mp4T!sO|PsITAu~Wd*J%P_z-CPv}~`K+hoI6FP1&?
z;EowBo1y%~GOr^i7z5BiaLOvpTBWEqvDMO1)jomn;tV|_;I)20_y_1l8VK}v`~k2C
z(>iYJ1bvif&a|VhYWBp3M%#K%bVA$4$$6HRT4(fh;#wB>nP?rQmW92rw9N(<jsZOI
zeO_^6-xM42V(C3#bpGP?qxhMH`*AP9d|f(#M?tQ^x{}$|pk)S@$t_nT*7Lx6@ilOI
z0!6@Gcx@97p4qde^!d_Lr$vBY2)6sk@tV!MX(#aj9~mt8KfiMV_}+IZJo_2&+6CSX
z19xM2#?ez=$h>3vVf#oI1J#x1Z4*FQRs7-*4$H99y<KRz<^aGx1fsaaO!mc}{xxp7
z=gLM<b0Vw~0MyTIAY@oumyGo3y$8qTccj3}5c1A8o(x2V{Ac%`_==HA3CH|o{VPv=
z)x3!>`pTMax2So@(lv8$10^xM9uc=H6M9GVPZiO`<Wucf7AsO7up%nvcm{#1rjcCO
zG{3b6NZ)pTN?CUZ@~5JZEd>Vd_7#*)>_rk2-9q;<d5`m-_Fo=W+G9_zlAt9jhBP+3
zhEZjebaSf9i+q&)h1N)y#Y00uloWg5nthaPa@4c-vpM!)@EIM{3%xaMJu~KQz{?`q
z@m(5^UEzPf491;t^ZtULZ9@Kw&dSOBuT(5M+y7XM|KEVb7L852O)&)D7xk;+n2_vl
zeY^()cDellhXoNps1iW|$>wPlBHhHO?1Hm4_h+G~5{ct50eGVH$@NTTbMu*`0&S7G
zI`Y%UWb#?chB2e6453HtA=D%xLn4x_y17kEVvWN^5h^C{=Y#3D<zdSKl1$yt`-kDD
z;iu1sD=L^zNs`}O-kz?{lb|{a46j!YPoF>e*zC9C=*va-$K5<0^sdgdgbWGL+K|$A
z*JL?~KeccE4_itil&wlkFIo#-=?t+o4Z7l;x5aIcq##`s?$NGjJQH8P>7%PF9)D^x
zxx4ktCv9$DsOm0kd)AgcHDx-R;Y9+lQAkm?3R&78owVr|M0;i~<>$@RRIaC|Yc<$L
zzzaxX83Q>L``3<TGbFs3Y0X;u?G|MiB9NwHoe#k<Z8B{3VU^p>c(KEpAw5`slKY$H
z6trr_aSG9)hSMch1E{6F+o)753u_ovL9mq|2O9Gclf{VPs4EVME@r%N8LRG^i1YMq
zv7YPGM1xzccs^LLz)Utz@q|_kmb5%>vx-KTYAg5a24qC}z}EC!KDLPG15^zqU5uw=
zJXn}!S~&1BMk`rw3~45Dl+3^#Ph=7{oVTECEc;su;4JgL>{ocX?SwxdFkH=eMj2A~
zH#DNDa=$vgDfutHLW5v}C=B%s>evvX42<TjoGxGSIjFq9VFs$!bHVw)9pwOK_ncz&
z@;~vUjQzVuQbMGNP2+)F0zgOyIlxgAwt`N>#F-V+iU_y2)v>6F5mkT=sPqFu8X%1(
zG+4QuBh7=Wa#7gy!GFOU)0~@nM184b`g%cN(t?99Q^u>4oiQx@d1V{RR@4Y!*Pp>|
z(s16M)!xN7gGC!0QZ@F_mZ=b&ViauicuWM*k}))bx83#bW4AVfn3Du`u5)17?F+eV
zoXL_(S#cn(u99rvX5obbgZEOg5ifT^<4VGxs=D*fw_B72D?7Kz_McI^x6DyB-;_n2
zN@KUhg#lh>(lsvxQk2mcwH4>V?0OEP$)%kv+9^}Ty}4le^43m+^$8>IM!OK#aja~l
z*-}|=T-4bQCuH2zUefD*_`4~~YuE9^=oyb7WW5&%4LKo+7{W1n6y)7Hp6qnQh&O;p
zhYj<*XWUd{bjK6-2`CQ9g7X&(&OkEU{v9rTo^exxpT#7+WN^YnJ5Vh8NOguX0Ch{U
zRi^5ZGje`>(zAtH#jw_p&+@0)Dc_<N)CLbLhkD}LtgjtzTjx3Nc-XdGIyz#W6C!6;
zHJ4R`aZP{@oB)%vw258yLrVGy&W*)@eM*3-F}2+Y!vj<2Z;sfghQ-Mj5lvrM$WOal
z4u%02*qm=wr6J9Bo)UL-yxRI`mu`d`W)j=dW)log$b*pN)>VI%-J0^9m@5!ocyxk8
zyT{r|ebwj%^%w|oMS5?@XjGnD1zj2UbxLwsTBcujYZP0PfMa_eH}+VZU2cc~>fM$)
zKi`$KuR7-+9EudI=ITtlg`mNmqX_2np_2wO*c#PGW7!TyQPh;PTZ@G`an@Mm=KTo6
zV%uFAcqjoO0xO>%Ic5wH!dc>o;1Lvr+B^)(LO#=g!Ge)c!eZ{q7^^S1sBlPLTWiJs
zrjXT2_1Db<lH@5Jb;0>@j^9Vn9V>UZNGUf0%dS82Mo5uY8c+7%6J~@Q0)PN-1Md|H
zhe0-)V=yGF6<Yz>SoQ^CmZx6<tORDHHkX^i1EXX>adW^5_yS!Y4&fXPk35#}cgHA#
zdbI@7-q)zSbr{Xz35`ez@oT`&Fuf-;ec_fvJkpLGMsxvSFFhyb@3LYUCXXC`5UpJ3
zE?a=oN9a7!so!gEr2Hk)1hT<&L9-aE0&YpX@xCI;s$1;U*~FoRQOFnI02cpDyEgMy
z=psR;8z}>`TMSoJ639il*_+FJSBCC$frD@nf<g+5+Xyxrd`<t!97*Ja7R&XCUq{<~
zf*h7EqO{(8(w`5>EhKZFpTD+-7)Nt)0H5!-g}Bl4V9myHOZhf*FanYS%tV59Aky3m
zFUJPr!d`2!?g6@}#5=(_G%_%VH5Ni7I5gfU!T3+Bvudp2g67Nx9#x%mjs>!S9;=yS
zuNfl>O#m`fZT+_XbjGq_79G3lWyn`t$&}w-?~;^CIKOQQj%Vm5t>zx9qFZL~95sxj
z>fJ6uHo4CKO_9SD+9JrShc>A|=c58??w};|cg*(WcG^aFCoI|<W3_2!RE&Mi&&Guv
z-qV-;<VDX<GXFLYI=rw~|Mo_movtflx3|Fp=~DmpI=fD&&sE9Vi<QGC`Hl}NpAWJc
z^Le02xz-$i(I>Zc*L6RqAtoIrS<$UO?+ssZHWcY!EBSvh%^6u)82{UO((wPVBrbo*
zjIuN!_5q9Mg!G*=E>}xw@|I-6lFb8ZB{)e^m2&<*=pcRRl%utWCn7v(eK&siXq`j-
zTz&oHN5?)7pBYUALWw0z8T#NdU_sI$zadVl_IxEc^ZYV;ef?w{mJpRuAlC}fJf9|`
z-5=S}7co&eN-$*2g%0j6u8!Z9*Du-Yu3xJjo8HdXm*194uX}&uXZkrkU0L~g`MO!i
zWNYUvpM3mr5~oL*DiR448H%AUBsEa!syUm!b?$GCQf;2+&15ul^=ge?wBm_m{r?#;
zX={<(k4z{DaW|$A%XCFnUE4KC@{##(+dPugEqFXSIyZITaR)1ep;+fUa-*%`D9V{5
z9Xi<af3L}#6K&#dV^J@<q78x#<t+|E+RQ;f!@>T9S)QBU%7;1(n{+ovo;Uoc`Hf{8
zuH$&^n8qiYH(t^^Zn*v?9r?YO+T_@Hug~}$`j}!LGd+_pk#VpCmRhmjYx%?6ux?n3
z0PbpipevUR&^vZSqF=xJoM-(^icTDUib3s&-z8++!3ARx2<73Z0G}vuf`E}20!cW+
zoTEs9NgKiRYH_-Bv|<bj9++Xf=57X|7ihZ}F4H?|!_kMh?gm6)>3wLs=I-u-On~hL
zx;B{hU<d-SwW_+P%0$99v$I*;V{`_NLoT&A7pm1PLB+Y#$VAof3Z$bSc+Z7~krz!p
z18C~mk-xz;!E_9>2;?l&sg+t|5{&vFGIVYlx5?Z90{W<EtB7>3a6Sq4fh(_<O8|d&
zUwbX|Nrba70G43Z2T8UIX=oqgVNv15@P48FIZ8}4dJ``GhIqca!$Q`&$~<C|V-OhK
zGXtaICdN|B=*Od;GD$}jE0~2Mo^KqO+T6BQ%cwOSM)0@Uu@3)!@f5(DiYoUPNSnle
z4=@<B*T~0E5rGI^$SojfpM`JtKKeZi??KMLH~0w<(J*btri?^}88b4x6YCZnfw3?g
z!{UOO0JMWZ(n36}e2U$s;K!qjb4xeU4&<Wlu235Aa<_L#2#;{wdq-%sElU)ta@6jS
zbR>zW4%hUWedg?-n%DtK^O>cV&y%j@(k>kC?cxy)*H-yuBAGnw-8s*;dqqnhP+nPF
zjTS*|nyR}584l-es$@>6m~rA(LUqv9wS$(KJGW0&YqaTvDUDK_`%CWN5pDr~j4Ns3
z8Qq{ipLB^9=JL)%yM5X1nu;=A)a&q((AI*>fKd?!@!YAoUCv{L+gTWlr^Y<VN}Yp9
z{Uf`C$7x8K2RQkpOlAraLy$_bI=IZ&kqL5VwMbvUMZ3hG-0s^Q-JG0QRY6%Y(Cw@<
zRB3y&gKhyrFcG2IS#ponWNMquqkskLjF}h%<MYxD7Li!D1FHm^cLZusQ1Hx}E0pLU
z#b4b9{hnx>rBtS__8hF(P<5FWA$9v|3H(o<2}tcjI?E#v7lA=Lo0U$mKdUWJPc(_l
z{0<%#JNsXwt4#1dY(U{3cbY^<dzoGZPeGIlyn3jEX^@7lNj5J;eJG*ikI6Nbf+jI&
zq?*`uDJ8#0ApB$5p+1;Mr9rW{-){*<fI}(%kz#3T8czI)HxJPut_XlDg%*dIdsU(F
zUPK&zq@TJAyGWc?Naem07Ab;>{(cmr9YNi0zM7msVaQ^(U>{8k)-xKfvQ7*n($g9a
zu*3!Bi)UmELQ)cD<_lc$@0HBX6nQ>59|=-qVvxG5Ag54}DVygvAUJh~m-PYAQGmX6
ze>JBEvW}cz6Mr$-QCKdEa$rb|UUVL;FK5sm4`kzwI7FB3jLPyR+=Cfn-}Ok4r!yt5
z0NXy1pg3AQk*(a%9XU=Aej)wR;&01_z2UK?_jTLsL=m15x!u===+Q~;+>(wI0McPQ
z*y-5?DY*%R+|y#UGhW0zZ9eQHRD`HLe~XW4lqa0Z*1oghfl5j{awnL)ji>PiX#DgW
zyAUUyWDw1HybYuQo}T89_#ohhqD5_;*Vya<2@y1dhAUxp2O|qc+tU}_auzseFn|wE
zX5$Bx{43ptCR}XK_j;~hvt0u4FIi#z*D(bH$A4{}8&=<p`!Q1gY{u^9!Hno`ljh3H
z&avpCO4i!tB^BnFt|3xQ)YEBoQ2sv0Nk~zTJ2hz!LAU`<D;(}m<uD5;7621juzGiA
zZf`3i6+(zdQ8up=*CwDqArsD<h4W2e_7m#SloCDW@7Ao`vR~q(Nt2C#!j8dTuJ+ok
z-0slv+4^;KqIcbjOURT2+2&viVi#lHbI0eqzaU=!^%@85CrPjv2f`7X;PgyRQBAbD
zq~biGi)dU(9hcGj4MA~C0YV<7O%5bZC@M|GK)y5V*`N|K_3ii^)0oE#%Lu`@_s1h(
zJIQ0BVK3vZzU5PbWc)&X#xWCz36|QWh(ubnL5UCq#R%+tL%;5^U1QGSrm<d9b)yz1
zW|a_yJm>1&v8qinOU1mco!N7p(w}Vk^j8EXp@tBr`1s$nmGbQ#$-dQN`xg&2E)^iz
z3JYJOUV4Pl8)}?50~Enj-I=LOzuvw6Pet-yJ7pulnMV!im}JQ+n2I7m6-FA9FR9O6
zYpb4H^(8)Z^iai%lyVG!oO)@nm2AhFf*w3)$i?^C9a?bGTi)}koo&zgJW(KKb6V09
z4B!A)X4$i5LurRQi@#n*6yKBBYiiWt$BV}|@rU1k?KW=ky7oM_55GeSL3k#C3_To)
zq7>Lv_JBB#glF_3w7)|+0nSD4;BZDk2J=D`MMzf@fa5vPkf4xWGQ@<8B7j6g^Hl9*
zqY2ODF<h;CwG7XFRSY7qL=*Dhh7mMluoVr@y}1pZxJW$+Cqot49zlIkmN5X@E-^M^
zvgF~bd9_JXt!+5ZqG)=CBBLw#MIy`WgdpFi>Y-^&WHSf#QD89wb!0^X5^Kzj;x@O>
zT&R371@~a%#ndmr36yJE)z2vbFQs9v*o(u?(Xj5FR{#o0$BK2W>@=hMBM_iD;z*)@
zHG%zIwSRckmNOS%Q4*HOJjj4v4_8@mQ&8sj(d4c;H&9q4Tq}t03rodY<8?PLcWcUl
zGvj<}%F(J9{|h^L7b>jVH;Z#mZ+EI4`^xb(LY#Mp_z9A9Ec{A8W9fCB6BoZu*`gob
zaX0#yfOr&yXp)3jyyy$U2}_K_1pl_m`NzHw;``~2K@_~50tDTLVr!q3b8C=V=aoO9
zN{@8^0*P^->NnGG%Dp%s$|#2&oEoC6a_GL2K@2oJ*F+bwXY3s*UJQXIhJc1(Kr}53
ztOm$i?iapA^EQtuWqR9gP2H?qW+@)Pf?yAN7cg~#3rPi3F8Y^IB4)qd+}h^ZV%Yt?
z#m5VLRv+-bdwZ!~dXqbE$W?K<@y{=fKa~JCGMMP&=ahodiznTkUrX2Bb^^+k0bqgA
z^~5Eef~v0jnA*AWWj@D0`}#`ERiqm`Kz{*16&wX}{%W(cU{+s{MdQRz)2?rQyq#Jy
zI<)QOql+i=piX_8Acu$N$3wGc`czh!e-$Tlvi@sc&GLUjs{ilV(Dwho%B7H}#n3>g
zJO7g~6wd|D-Eyp)b*_L^Q(qiTttiR+pIAo_fJ8#!r$o-Uj66vHveC=OGw8fY-rSMR
zofi)Wm#n8zL?S`*z3qT0k&>hZRnk3;Cw~YE9AC^XUM`ar5hJKbL73FhbY0y}tb8q;
z7{M6=1t$iiWH>u|9lbqWoP6B&SN3G|{oC;L@pOwfKVL7Ner|6MX1>0j&L%S0(k4ot
zczGCU3?ebe8nO6tCh?Dy9mW<9C9K3JzYvR2;^tkaCHw5hWub05Da*K-Ht1dR|GK9g
z2G(FQ@<1@YDR9qzBFRRHF6w&WdR%Vx`@LOLCg=%1qo|OEBv-2J=68u`D~Sa}M$iER
z#ATS`U_0F|%QCMJe51J!rNza9cR*U}H6>_Ugf^zu#3!6V7}Z3;vTiIjanvMh_)7|9
zenX->c${TlgeiVfo)=8I$bm7c1P=G)c+}^3Uf({*3$Y0zBzq(1LytYdu($_SQP!en
zf6h<5mJ$JsO=Vtm|8Wf5Bv#l26kVhN>u&c~LH8Ge!ujA01Q3tJ=ejtpc&7{bm=;hF
z#*Frji)x7oiy+7OlQ=?V));OWk0tYGDLo!+B1XcPQw4kV#9Z?IUR_{l2$=?O3K7h?
zJSnM6q)7)(*SA12rk_vtUV|8UKJ`2)UQMj4L!u76pYw7J@>T`)@IxY$RI`QWyevu$
z23DEpMae_&ZC6`mX}rWz@VP7~JRa~>BwJx26=)*OHR&co{{lvu@BL(nU4os9s}3Nc
zR(0Dw=w1jPtbU3Nt2czf9>yod0<goVM~EB+p755(oNE9p=(+z?aQ~6izNyDv3UX-a
zulR5@hm0B>k~PT|HCBhMG_|&~ePJX^8Ek=vlo2i<Vu+Sw9RxH4-ndR@)wwoPwUbF7
zRWE1EQ7Y?_qYo^qR>Mf%Z+&uj7$VQI`Cfo`9=ZYA7L1mHgCf!NPT_L5q;yTyVXqGE
zAnS@yA>$XY-5vGCX~0F{(;#a#%@*d_zaK`t?$MoA--YW3&We6l%or4{PNW-oe;d}e
zr|oK<@RIzu4k>#W;<{@UC9+&9;AE-&J*6TCN?hSS_Rp39)d%G@8^*|@x0E%i%lsfi
z_(<767BX9(^wQTWc@1pwV<GSdvU#-shyojoHPV?uXy)q~GC-^1AgNuh0IrX)zbgpi
zUh|>|VliWmrXUVFhQ66#`A=~>Hult%<F#dsKHB2Y3oi7|M9Uk>X;qhNCYwy=dT$WU
zDY8uid9Uo~y;ByYw-*0pe8Bu;7`+U`rTB+t!p)l_SwnF6<s=3Xs>j5bF!>B%Po=QI
zDLy@rH?BhEYCZILh)?=ObmeM&I;&e_q1Wp<rxBP924n+Py1-WrWFwS(j*c2A*G=8v
z<|b;nD@#tesoxt<B5*LrO--J$lVR5<aKfyqXlT%P={Wn?O<usK|Iio^@Sq_e;1qnw
zMIOo+p5>i89Jm#rhWp+?n<V}a5JZF%dhp!tx3&qO(N0gX8q!l~qnEQKEL~0E|Dx<1
zgDZWvJ{_xL+wR!5oen#;ZQHi(WXHDAv2ELCPtK{gYTl{$Uo&T(PrGV=+f{q5=U(^v
zU6-WgmXzg<IULcyKCWag(J<lvz-P`_3AcJ(nERR=<^A-3#2jU@rb}0LBS5F6b1$3N
zKp7A=K0+G*P)u=;JQQ2)Qd^v6dZxLz3ye*~rq3b;Kbr8?ko9$e;n$hpwKU89>dlb*
z<OgQFRaSM~bV2sq%Z#Gwk+CK9>FU(f>#H{DwQ}6XfWL8;+}_2xhVA9qq@UaEA-o;p
z`cToO)Z16imRM<3eXKP2f}_Qy@8<Ag(!HI!Q|az|4}Fw_XMRUSYC?`*LH+8mn|E)*
zGo<HvV$!3>zA!dknr=-pb9T9YS^^AtznuBht{*YCC_Vf&4_6!c3v?KwK=Ds6ng3*!
zU}R<I{7+{Wi~lw8Usk(O6f2zEY6DvXfi{4Y0;Om|YxJ`OYZT$MA{|33FQSX^@tk%@
zMU_nKUXYyaM=1HTd$G!FYGNu8IPM2QFFtKrSSlVRCMZHY>CB)BAp|L9Ayf=w61ihF
zL8YIBE0a{RKTiPf%avv)IzqqI>G}0GqfPMj@ick6du#P}(d5g)o1HC_+mk7~jZ2U<
zt;5TkP|y~j97GKVWCKi9IUD9715A<xtu^PDY?ekOBi3gHw{L)B`s&c2P4T0WY|;%0
zqpl-@+4r1jG{IU+8NKInuTmnCWpwj8a?#|AaU=$;aKCVJspoQ&?A0q%2kn&Hk)2KY
zua{@<jz^1JeSbfAGp|rZSl?gd{O`U$hLK><GP!ZR_eMRM+j`C;SrO^pmO=vLD^jlK
ziM6myL3mKf@(DIS$@WQd`?Z~e+vB7*uC3}Zy3Ugm9d4LQlmhW!X?~ikK->yNo|i1X
zF|@Pl4>gjz=Y*zPNo{1#_MDo<@GX^ipfmtBB>feknYu-SAp*6PAd52-qzHrNEyKq>
z5;<9|L;w0|k?6m{qVpPRu46SHsU1jgTpDaZ|5K89ptdSbB#IN!wk0$2g;=|IsOdS~
z$o;dhph~<TaV(uQz~b1dy>i7$LyWNQ*Ihdk1Mm}FRB#>GeF#9~VoD^a^Kf8>3s25t
z2UEzoYb=ou6UJ(p?hY82%Zme^e1MR2SOgzxo*GevS$>S2*Gv!^UmjbPjlx5qAxvTb
z03HdvFn0oP5m~Me5?WBtaSmV)`t=)2!H95vH|?=qsI!eB`SMkJ%8Lzy06BNROy)hw
zQX&xrX>mjM*#@;fdZN%6xmnpD5m#T_Ji|@^agMhE`lUdoYd~!7@lKG0*r{t<x2&M+
zorEW;#a7;bWo&Pmj}fZB_0;H>_+C6>3%P&rtZYC`<P1-*=<GQ3_m8O;MD#$8?Ypcd
z7MCOd_m)n4Glzbw_XS^rvl%4<+09GU1&1?(m27YwGPgBR6a;|0*Nc4EwR%`D<H8|<
z=(iWa&MeSp(2bp%GOz7zP<xv2+u~16vCJfzpf*Fxit%}{mc&6)Ey)_d5Y^rtU_8nQ
z@~gHdeusykh};%fal%BlPMUP9aBG65gB<)>SjGc~-JJBa#8(K0j!<kR(lH4$FF`$Y
zE&PK;ezK~uH#NxM-RyL5JPt7()~tX!*~}^=2$dM2mO;6^J1T9#dyv<hMxClc(~>3r
zgb_PrUMDm_<bD21-&q<y<<Qo%CDg~+SER~wHZNCIztkiQz0w6wZEj;rSzWy%){Rx-
z;}Pgqs}`OKTjkh8%}V6#MvQQMw$7CQbD8A+b2%~!J7QvnbE?r@T?&?@Ls6HPaZ-fy
zn#$&iLA1+&y3#VUT^@s1Fx9UhP%8<kU2cENXZB)f(@jCIzaoUFGA6tVmvK<xJ*q^r
z@b&Nbs^daM7jJ{Cl($_WxBE4D;PYtoS5XuEz#vjYO4?Efg-t$Bm)6KvdzkCH!yG&q
zeDWtD#Y{Xl3)Uea`{a>L&`(#+c$y#rc5w2qXA4Jj`iGZ_&TUM#i+PWgTJ+rcJ>hHg
zkFy=rP%+4UAT*cUoGE_o%#prEZ_oajBSzhw>yo|OKqB^Mq~rkDBG(J?hI!&CkCJ80
z<b&c)S!&is6;k!ljY}qvo^}lyPf&j#vsnx0OQE*Hs*;Y6WjBSyjn?DWCpZrJNRyvT
zaO?W!zb>@Y)|5p8Iq5it5$I!9Lz!Uu%j15YcB|}&De{hf7SZA)HlN6?6>y?mXKQu%
zrk+x;Z%9-!#{xhIH-^VWkLvDeHitWO#W_%`WEWn!4DyX7bKaq;w>HZqo)AS{3fD<4
ziLmnhlg@0gE+L$Q`X{^VV-EJ622qF47~XtO!z8tRyj!KkuY_ZN{!vpicR9V9P}Em&
zgn9m{Lo){w_hC8elI^>i;FbxrisuYF$=;#lcMtq3>Ev|APs*QF`%bbnI-OSrY9Q>*
zRKa4>)2E6(u-QC(W7n;Dty+hG2aA3=1?GA-?Gb(5qDUqDZoM*&=xTxG;)L!{K<0ST
z$#JTam4jYW;Hxu<@7isU^MNxK`G@<Tn$Q11Wc)9n6$jIQ?P2~0*qg2tX~%hn@2L1I
z8W$I{$cgbPzFZO?WoVLQHVOmYkYWJj2%Qh=*v<ZL_rnF~LsrnNMvbb*8ryw_da@`j
z8M(R9c}FryArx8`Np=wnWNO)T7|C{WTaf&P2kp1BQ%=*-KQzcJK^Pv181%V4a^Jv;
zbg18inAuZ8lL3bx2Ns<CyxgLnkpOoOPF)?^8hc+4_iQ>k`IOVc#g8@jGzJ2c*1d$m
z9VzK!i>3FB@UY=<vDhY7a51u0%ycO^L#oi$41}lo$@VkV;I6t{Tfkh3wM|4Ov8Jt;
z9Au2!pRyE-)~}S{2~LX==1FzC*l{&2G02XqQP+0z?W+lt<DdyQnzA`0r~oOu*6H5$
z<`PgfToSGm@IuLpdi@|2tY6u71;~=ZL=x?R;&abSXiDIsO``d0t*o$XyMv2BWHex7
zg%)|<OPB5ajj!44>c_I0$AIcd05WT4^#s1liFjK<>$zsD`qSzik(R9TM7bVgh=y`H
zR`q)VXZ|Sc@PTwl0oAe&<ZNoQxgBKioi!-OdvkdXf>?1}^^~$e@Y@|<?il6wd_`Xs
z;W{-P103zv5GaZOnV&+|tfGIL@G`WlFYMMgD&q}wfx~z&j0L__l80PYS{@sU7r3^0
zAe;iOh}T>LHy$GNcT)%L1-x=e_j3iK*VPwhw_%yOh;t=F1!Lj_^JtvpQToNuU+~-)
z%^UX!jM#Wc{=nIqR>1gW;X-*_f9scDdv33D<ri-ZOCIVxRR3j%v%a)uCi)nmJ1(YH
z9QJ4;FRPrVK(x|EI@h-DXHv_xqK8?fXTH3E6faA&F9FVZjJBa?2E1@kY<2``$jgia
zRr2^9mV_ei#RP9fMg7g(*xnke%Oxq+L2-|2X8o%2BRs;+$U-ATaSknol{f9jlc6c$
zDqG8>A(E76+c1z?+->Nvx4t1(mReIs&O=j@eQmR>>TydqVV)hzuBwu}JE|2oyR%N9
z{~&OdK-+pX-EH~P*AiQGR0d>pli9+GRoYPak4QzovaN)!&epUz_RmjlhN`oVNNoc7
zYSbHhRleBe$XD=~0fvFeK3^b)d|%gZW(&gkZ?0De{2{%~fBZLc{=-@{%m3Fk*1znS
z-PbjGwi+xZ9N(%cVO=U@RME8yS~!rZOJ}W>ddpEsbjr7NZq!FYV~^!^;uFafmW+{S
zZ(R7vKM@Rf^Jtk`JsoWxU;V(q!km<nHdkRYA{ye!%m*fdTj;naj~$t60~80BuxSTF
z=h|I1O9ZLQ%+>0*pVbyii}tRS1D^=z?VOdiv6CKZi(l`@^hJa*T$4?)AW8O4BT$I>
zWy7dNbIO{q?L&F&1f9Ca5ucdg*GKlKf1m8&npZYcTFxii+Yk`0V-|UWtZn&|K9?Wu
zd$#d1>>2?5O{3x#c=Y4Z@_3FHOrmB*Mg3uM#?BjCbO_E7^1Uc!N)hi;Sn2kH!8@Q?
zoB~)seYUD9HuWPJHNh~D0>ZtK%|w)cnfgz}NkSqTbny+&xaE#l;M;g)F<ne}n5xsv
z_pOe1=s|@S7jEdYRSh+>*Mec9EtT4f(==&4lJlfN4WJO92Rfnm91ixEVZ)AD#7l$W
zCZ{}$J*F!+x_eL)!Z9vrlm;TqBtTW<4;-G$^k-myFh*A9fKMn|tAu`C&2*?SMrOgV
z8`<U3f?^F1y>5>yq$cH)8^1g<fGT5A{*BPzhN@BM?g4v+-z~^OQy9vZKXolpZ}u3A
z_G)z#Z9Ad;`U5Sfz!7g`kXMEhsE~80^}_NHo;2rCX2?bj8K-g{Vas$2rVxHXn5J%N
ziE!!hb$fI+4d`z2OJ30PMZ$guRWj=C@_3Jl2k-V0WvFa8P_k?%!_9Qm+zsh$dt+_9
zftJ|H6g7px$VDMK;X_#<Y7)xJE`%kK36Nrg0FP-MXE`zYeC%d``(Qw)khysVujl)@
zfcUdsGpH~G8YF_a)6{=ouiP)v@4P6n7acI&kuYZ~ArLM5%`FlC<t{`^Jf}g+f88M^
zYgKBWsN`d7jO<Y?;ZnltZ<gt5G^S3+ag}w3#Ar{e$w%Ky%JO#K2wzPs<5r?7&Uzfg
zY{gJAoqI3ywGxQbk&Y!=zgp$c1?0Z0!LtM{m}!s3l(LCcD;~W*!Umc;R1*2~0i+d-
z4wd7#C$t{CY;8YgEf*-}6qrUEm4Si4go3d^C~+|phls-BIIXy!RVtu=X$B5k*SJ?a
zrBTjUj6R*#E*fTxRSpb<|8g!`ytB4+QFfK4WxV3_PFPa&!Jj54oyrb*w)O2V{1dL5
zQEa4rkuURv7}hj$D(&E8nsw}Sz_CZyT8gzNA9$cXpGVe3t7z=6(9tVnRs<A1O*r(C
z$P%mE%9PEnRO|PBC#a#&*`lFH`ST-Tx?4d%A-&uba~R>#Yxa@Py{Xj2zDKoIo-T<{
zyqPr?SiEJO%(N(!`@%NHAe;~bmEAzwnqVrvlyh14j?<989@>)~xmoqnV8sfc$W40o
zUKbB<B~$<>hAR5xS^d=?B*};)Ti_JP9zB8_j@57`;N(m3u3C_BN*9F~pm4Jdp_u9V
zfiwLzLKj>!RVQ{^eQVq*bq22NGEoDW2X2V5cU|_b9sPzfi@vP%BjD}4A%NU`P{(@*
zQY6QzaQ$J#p`6NUOCEBN9ZtWxREPNbFXgis)7FMEh3BC*!_5OcJk7oxT*5>{KtPVz
zNt%C-n1p#y+VHHK6bUJdkT}T$OTPq(6AQ$4KI7%{xdkil$Q}{IbK(_l<cNjUX8vqG
z2YTd*-Zl-w`dbu@7#n>OxSFmQ`h6!Q3ll|mD`lQGemUjG4Y3mqjCsVXz2`;}UgI43
zuioWIC;lOd0q8DP+eLr*>w#i5@pCf;$ZoSm|K&V*FT4dtG@z@K{%IJ#q0ptTL`joa
zd)U7=r_+C@t1l&UKJ<(24)&B{2Kq+`cr|4o$W6G>1`IC>dn_rZpn9`l^xkUFT#Qay
zeQKX$eWm>S{1Y{^xb4kOhULnUac~R%0@pWQC;n5(=lCa959@z!!I)H+{&&g$MqZm0
zBBcsT6$KT9^0Zv(yLPyENCpU4j_NcuU!24biq39t?Fus)kFQ(Rg2NDy&JLvSdTqmo
z7Hy3cZXKQsc{$yZ@94-F^aZi&dT6=n6GFiW+2(j$57QtfBHZiJZ2?8&)o~Cap_K{$
z==yklo&0#b95f}uV<tr5M|JXY<jtHa?df70ps%Spng_Hj+nOG|FFLXHWOuu9=j!NA
zCWRw4$>q~F;^)&MQEK|&BJ0&1GfNhzV<t{hXvOeTymGZ^nu-}FWBhm@vkP#>Y5fjD
zQh#fJauAU888JnTQ!1k=L65@f{S?eFTDPhrtlK1)2$my7+i`#4DH0p3%DKr4SFos|
z=qhk+Wsc|`?S9mm=VV8zyjmpYfiReO+yLdnS}P#%;D38#0wfX`dIhX!E<W%<dw=KQ
zveB;>yDuH(-qfNpcdyvEoLeQLIsdx4RA?|?lFasPZlCL{82_j)zqi&GOOdVP@M<D|
z8q-D4cn-q&6Q|eOWcpKC4byIEUDUt?P$O%YBN?xOi7^VS7!?HPl8X!~z=uU*6MEve
zkXUL(8Mzqu%ir(z1QrAT2PawGkqVA7In*)w^?2o<v7%;6nY9-eD8TfzWV$P{?%kVC
zSD&Hc%j_7LC$qO^<+I7jR0OL^iUma8K>R*ESVY6xrk3W95$AN&MP%=ZMpj0-L@q|(
z-;jfD#;GY8xAL;T7TmK<lV7`q|NbqjowJMEXJ4O_sh66o)JS@**3OwhU0fT77TH%R
zeDUd5g%L53ozHQxpFyc_r?SpF`=kTpto9mN2QekWmqhHw2IZ?&cV2xyIUIB`o8_G}
zPyC&z{K{MhNZ&7!t1SKuIr-sU0jqaWHnk@Lgd9rb_{-n^0HbA*0E$HX>qQ{Gj{e%Q
zM@T^m(4|JY0nKX(sR}89EbvTb-ql0Z;yS9`J!L6f$?~(}s_-XgDfC`TJ4k;y%UFqZ
z^p)f=CV=f{q$=SZkQzR48cXaxoxOG6PQs%T2hO%NK%@dBpPga@8j4UX3#xYNgJGwU
z5Lwqtd@KRxM^iC2`5$0+3Wum04lCZ3zkgHm@5T`P{D_rA%7`nkkgx934Wg#X0!2rx
zt#~l~G}$3R1=3*gb$Yg3D+jNIWp>&NIZ6U}N`+HRv}Aa`%n6~Q;(>Sv)xiYm55@q(
z0>VPYcA#(gD5pEPHv7m{3Y>=GQ;%RA;l?9=k>Lm%>bPAwv3${M84xS%j8MFC>D5S)
zlP8;yXT>k2)N{C4ccPIt3PB=TLr#~IVm%qnh#<Rc_+_wU>J!S)Oeyr!gR)|?h#?=i
zA;OZEV%xuW!&xhJWkh(hASg-Du)wPsc?_a)R9M1&fWcxBJ{T8}phfNMDuQ33jECD(
z49xaZMP?LKTBeFG&ZjgI=~z~(8)lT=E2$5LNexY5gw3RjNzPl3&4yGPM$9>kF_mC{
z9JVk_ES%u?y>S|;JVFhJ)D5-}F&r{_#}M~Js$v=175eb}X|d9ELr21tQhDQmHekM2
zsZm^7w~fY=S+~M<4e#=nZnT{J+}ReUrL=i07!M!v{+hk&0j&_0Gun{uFW{Wlc|!(Z
zL|Rek8#2(Mm&jOdPj(x5mR4T1%iPz`?;~7pwx8nOBq&<Nz`r`F&y6@YNT-fne)XMR
z5Z9)W#ATQ4#)KOt7vB8I3HSMdQ_C`!SL3QmF(jb-ObB^EB3EQn>mu(?%qx$Y*BYpQ
zc?ZlL*GzhfVj7E3vRraBST6r%4KQQ-Qu1kJz9@B6mT5>%H(vV+Jeh+pU6011%wX1L
z3f|%8qTs_WJ;GHGa)powUpU1B)+3<La}_5P@)Y-@e*=o0ZvPw+l!>`i4^879@X;8I
zN()T;+0HTsYhVF{#LO)7xpL{?nC3EVBw%PjkBFWon5NYHMrxjsi320sRJoxn?;-{B
z7tZ&|`z5!HBSD>a=X6e48zYSGBOt-g0n*#qkAG6?Wb>o!`oT;X^!P|z7ldYVehb<u
z{KrVRCnB$u7Dr<1PNjg9^AT`uD%7@GU7hVKB<wwD{XeuAj_-X120~i{3m6_A82bOK
zb%E)>Sr=ITV<F$6hO{k?IP5p!V4|*T0zUMB2Z%%f-%c|R9E2jYQ4Btu1XU2U{N~@*
zg&5qSpQBrkQWhEyh@sr?lYs}~@TRKDQcit}RC^Y6e@Br3(5s^6Le4Z+LJnH6nuUpL
zVD=g?8imrcwaHpiT!C>~`AfgF$j-F^fCxEOFg5=`KYsthE>I?;W;T1&#IzucsB}VO
zaSO`|B2FpJ=~kf3^1j-W%1h(mAoucws_2HM)rmHBoDR%itl?TgOyuQ}6Ts&fP@`N5
z0ZaK4gy~;%Tp|)Bw$Xh<Mv+y@S~S+cNVWit?Jz28nSN2c(B_@6(zx{uJ<$S)@t9dq
zrCuW%fZ&blI$<+z)za_*N0pfWt`ZJ-Y}TZF@~XY{zMvv>Ezkfza>h_A2a$;k5%4dY
zf$&5#x^lY`0haKrvNV&M#v%i6@IoU^S|~!!rqnnoGrXNwVu-+>td%8{a*~>wPS~05
zN!d&-Yzje{mip4#=08MLRv#1_$=DM0<C3glCd$$)Er}&8lt^K-G8I|;?I$S-C1L8|
z3h3e+o5AqeGE`GJ)+%o-_QcARwF2^G`ZbZxGcbRRrV$7Bmt)aGLQ1D?VS(a+CA7*6
zPe!l?xHT)xP>Zud3&vxe*|YW?LH0C=+{-ait-{vYs1{m+dp59H9Kx0>{^)YBK?TEV
z7{V^vlL`x~8AD8<#fw0zXDB-X2ho6(Q585@9UZOo{9Mj;T<-FEe)n49)upRGjy(VQ
zLt88^ce1_Vb@A@#&cLTB;$i7&{9|Cmbvh+?^P6X1yUNv-tsf`-0D(xyvXe+*9fEO{
zr7F2U;cV;k3Fj2d4$$22dV6^Kae}JUk%+RKoe)g+#rjfy7;pD8F}SEQH_BhZ?GaTF
zD6W`S|IgLo;oaW%I4E1V1*5dullQOJ<x<0Dsncq{6Lz^1+`80poz(Nf1y)|ogEs3s
zd0M5@=I*J~d~!8t&K87!RuYf87mJuv=_2wnw&|(4wcfamv;5IQ=RjC{DAGJ8Nl6Y8
zE4sXb2O_0L*Ckk>0NnwtAzwzLx211W4R5k4Ol1zHWaE)jtnQUZq7@`%AS2_VKkG>f
zXcf9!5@<(h<PdUUc1;A5raVAYW+knM$aE8s`MiAjdK??cZ?f5Nco_O>oJrPs>2u+8
zmXoXGSsSb^K$db*6^8tx5qOC+s^%t3>PDuyFn^S(R{gI3>T$ZWbX{$;rJH|0a@e2v
zw)g7EJg=Wp$fuWxLd{3A>JNn9VX3JNdwUCDMAi^x|MSWB=R(1&|A`lNViZz#pjW+6
zyo~`w;^N+9C0a_yJAgm#7y(&M7u3ulF3t)PdgxS8$m!2Ym@==Fp7N@W0z<usJY}1K
zajb_0F;=*ETpc6`39=_Q7#wIY&j<%BDVpva5+GE#*eYpW6#h@E1UIJIvz`_w<$k?K
zGiZPbfrtiMOf5ul1Sts$6~+M6*rNc##b>5KR$SLk8WXIDq>1UFcxe-Uyx6=g_F>7r
zMaNu=@I9rDN{nV_golKVV*i$*_zq%RKhv)a4`y7Ri*gxRW5(EB75s4kA0pP0(prl$
z&)x4*hcy@Oai<n$0+wvOV*@u)1;wNkIbt@JTHgNr=i)@oT=Bd1L;EEp{A+*wyBBgq
zmT?{f5vj6C^`{BaP<~OU7Rx%1PT}B-vQ$#J(oEcD9hf1=j>>+Up;{Qno5gDWzJ6tL
zax5$*dg$1qEUOg3Ajjb_Bq~(`!w=K?*_NtNXYpWCXM~vK!V9C7wh5v0Y#NxU8kTq?
zt`ecSI<z?mJ3R_+t6k>JOKuoUZzJG&=Ki!jff01b%wEry2spxD^%O_?gr*m+Z#}K)
zQrtEr$npqwyi(;AMBh297ggk}F!B&rY)9avw-$z9V5MHzz%mOqGZ6kOpx$9zu!G?U
zMakaue(+wShBZa2S7s9v>$>n4mdhL_M~y=hxHi`^4fy+03uid0zqixZz>z3k#av5p
z18oT`ZLejlodqexWMck)A;_{W<2jAS7D~&XG7-x@2q0Mh#t}fe>)?yG+PygQJRZ7-
zh{G}lkHus>+Hq!1^mg9aWSNyG{~J76Y2`}+5G^f@S|1+<wKdUdXd%e-TAd_afouq`
z`z_`*b{@D^W77GFZGc&}dR#Oh(4koyt2H1{)^|JAwY?oOAinYsy^WLepW;%i-<pPh
zKmYHQm~WYE;@Y=N_E_y@Rwl9_l5BPURJdC6u2xokE|s)lAAF!4TsVrT_+J|_rqUqp
zj8vmmrFi(2<Fm7~uicJLU~oskaMd2u!(9|Ln6d#uq!9o;o*zLlMnzE3MhrO+*TPd_
zgrEM(EVw#<?;M=&>-l#QLZ*GO;cd3MJ-O;XF$OYk&0a1x-CpjU|L(s$CaT5XEQ%#4
zl-B}gu}?M1#H)yE1HuVIT9KPxdPfmj$?Xd!Dww1(#Mn(Y5|5DbC(3g+j$1P2*Bd8n
zFcMEE{<JiWB$bJDfD@DzH66+N=25~uAKjO}x}glACsgiRMJd1=v*`K9%}jtul7;V%
z-Xgz^JY#7UxwWcK${MmVZ;wsX-n@GAnOR*i<7u3Zy0mE*sdP7dns=9vbx!wpSSQ!O
zFmtS<CmK$lZe~v7T3$}DWD2|DnyelXMg)RQtX2h3B8fQISmpdaJ+(03p>1A@-x1ya
zQ|2Fz+$YMQ$G8j4e&w}>!L?ERO5CMsWO7SRPu@HVy{{ej60y=!;0<px^BdrcbVV5v
zwl^2bJ&S%t%2unF=fIq}=+N~i<+oz-;np7>$XWzR1?e|Zf$g}RULcwWOh5OQ#V;`J
zzHq&Xw39Gkp0&YVfkNPITYVr;BZ#Q>9dW)Hypz<SPH8|2$5Wd^50n8~n0j2;Z?~Fy
z;-B*sd{ImK8omqW9|l&bDX)T8jyju}s_VOJLJ+U%zpe0M{0-iP;Yy%VPq!k_BfHW&
z8a!Gutt>R$^$EqTpr)NBp$gHTDr6!jtyJhTSru{R=BZh$8f%3eJ$crL(j%yA?uOML
zuF(Obwm2J*0kb!Y6^($Zb}MG8`zB*Uqj246c8VNZ%VB=fk~qd$fyTO-G~fhG43+h{
zxyH4chRNp?>2wkmlQdg8<d3i<K~4nt%UKnAx9zYi!glI=ekQ=*$&<ISyN%)L5@&)w
z^gA=BSOP6)^*@-BMF;rTZG=Klf2XY1ne|Yk@X$LdH`EtVM!42($sm25Eze2DNfHKO
zO=k!34rSZgrLXTqbuP6%DA_`%h!(RTCVEdJ!t&NAaoUYG5@dEbFk*sH!ayO|2c3N?
zKzSthg!A=8f{tkM7<muw<CO}|7iJb<BwmAn83a)qhP{^c8;P-_{V9yGLPV_HL(Mqn
zdhc-)%!#wX-~Vnio)O|m2(nmkA0QH=L`v2PSGO~k!yvhAEHr2-rqO+7t)7DRJ3km>
zP$t=gX1*=cv5ACzDseSd#s~eFMu*VbBmJlwK*~ylBJU?qTyXHTUYaehFNff9^%gL~
zi=G#;Fv0uD;j5=5Ef6O{@bOT@uh)OE;>mgCebz=gLKQ6fBDI28u2+B2^;aU0id=Vj
z79qvUMkLigj}phSlrA81OH(t)Fx*v)GBQ~R3AIq8LEqDP*q$q<Ljm<z;UdCn`*RCP
z-bGk!pV|g^Q4m|!j@utyDCcGvGHifuxWG3=+zXATGZXQq=#KKp#j_UcICS1Q88(9w
zd(r(y&-Lkj?W*K+4rGW@N|aQJx@|q-e(EKjbeFu`qdvzyM|I<!_NYM#ABIQj?4{K&
zlsvP=Y{h4iQ-$wGCbK}8BqFFSbaoayxOjw<S2$IQZ<XH-mI~lk*^)-d1vhSuL=c4d
z%V22halt?;2XN%N4n?X~sk4qycZ$SN!h+&DjrVNz%K2WNkoTcfeqv~iZ#=bYMv2BJ
z0Dvt<<Zxi4b72!pc58fivx@EJ_pBFyX}icZU|>KOGwkE|d+m&_!TbI~&~P)#w>;0R
zVh{f3IVnYK{yN{%w_rMF%3^+9zU6+XbI?&2H$lr)n}U?&*G-M**qpQ(aO`q;-}9bq
zGCJPkF_~8DiQ>nM{`r+FV_}xvlN*8|sX92tSUYgWs*KUr03YK9N`nJ)o9?2AZ8^sT
zm->Fn`qe(43036Rcn@=X*%lmLb;-ePp70Px%ino{lDL(*cWF?y9)HB>;kp-9n<FS9
zF8m-#vym7?k8$f|!IiMd$qEP|`9?S*2%!Amvwv!q&Nn_lclha&327uJh{(g32D2xB
zojPT#_o!Hqn;bgat6%fdanFO$_4WV{KnPU-&j!TA@GbxO--oOJx<LJw3jD{(qwo3M
zzgXMfho^rhkK{qp<PhB3&-z?wi&IJW^U0)4ek#ICa;i|wZ*PGEClJb#Ns*^NktV3q
zqQ85;gTOBV!%vA%9~YL^aQ<MKLy8V~rPfm<bPHm@=<~wrgrid1f4<kv?Iu_EG{_(c
zCYe!$?s_>~UHn*lbQ1!HrxM%`53_S~=KR<2uBSd%S4(H26vLKod*b4K{M5>a*Za$a
zud6dHomOlG3)~Y4opR-;fk{pHx|I;wFRpscBgxLX5Je$g{V^o+CT;U4yM|V2+X#er
z4tX2&@nmjk@t&ACgUR1p`u@>(20Y;m@CZ-GiJ!OV^}G_ufSP^500wYl4nklQ$~jU*
zb!F~}mONjdR6o#0reQL7#jZa$L)u*BmAEKq<kCOuR&1X$TV)8cW|(Nlc)B)IPw5(v
z0p87J+eWRLt2L1C4I%vP$;S1~8HxSXm(=oLMQkiEttQ{1k!UT>(?c7}*|Dc7$KJ>>
z$4(&#60zS5L@<i7Rey7$(_snroKUuHkl&AqC(*CsiFv^MYeK!adoxL(P-KwE_>m2!
zI3vn#2u+kjGB?+Fl&`A!k{+c%yMKFwb;sZIu?uP6M2r0Bc5l#B-pN$^;X`2fZyxIE
zA()=hETm5lxR(R=OVF3XLu@~N{LD6jPgLVs|1_#_*2|5O6wOD{uk+t83QIq&F9s?g
zdz$MIkRvyrd=jk^E2HiMlh2Dl5l@RjU*;GJLb(&ckS^?M^!ZkDxukpd|0W{Q)o3Bi
z0>cFG^oQ_h639N?L7~aPxY<zzI|Z~kAG;@zxofFXX$09P+qLCx3@JEZ>?HC;k26AN
z&x|y7Hpq7w(Q4@&^J3}|t^Wm^g|A#4hNQ@_hDY}fs(X+b#<nOm?5$c&0NCj&58J^V
zt<f9fn+{o|+FpVfWw%CMDrIr5HgVS``Z*dKYm_9{-S~8@(tgq8f$OOU*SNO5w|rhR
zQK7qgTw{)ho*^dvBrR%8(MYX&CWognti-y}^H#ia6yLhD%SAPf(U+8MDhdycUj-&1
z3h=^XnO;6F+_5Hm11jj?`Q)~LtEr}+!h6?NM$4uHuN2+}kRvZz`T<e>8~ZvdfMqaY
z9z-5ln8S+_;6KnH+F3q5Q;(>F&iZ51lfwHYgUh6mv6tu^i6d1cqjjfy%(dm#x?OTy
zI9ss3!2PyCkv@mKi75ut1#h!)Y7<WqPN%Yv6#$M}<TRMXIB)=@U2J%c@w1=w^5n*2
z@mwREWwr2(LJf!j_yIE_jQMEPZ?^r6h&(5WXy0#AgrwT6OP|rKNhlt5yU*g53_wDf
zPk&qGA_y|{Cm0^gYupII&J2C6!u`cI+cpO2F-D&WQbEF>+@=)tEX+JmFN0U$ESojG
zWpe{kXrS=vM}HnYV)Dm*HDXwO3G1t27zS)b_E~%z31OZTdp8%(FZ~O|*fg>Z7qiOF
zdqkk$r#GgVX3ojdy_t`G$GK^E+_$xLd4H<-@#LWZx7Vwn;Zb7Rff<)F6%>yyox%Yw
z<Emjqn>kzSjt>@mZwwWYP_rgJ#Ww&%NME|<!j^0M+Mpp_Svv53pGuLInHHkT2`aWe
ztA|XyQ0Q2$y2^N#Q%tvj<ih1wq!H=jkg$hE8?rB~)8h~7(9g5g*mzVP25p4vy5VOU
z*s||wa8Q^|HCi?yOFY-@V5fo^D0a*azys3Qobrk!j~MSM6@l{sF81>gOU~)z&=d=a
z)ytbHk5Z^HT8^nGfQYqsUngTO!I>xN!};1tk^nuA@(_4|pGLhZevTwuR7vwE@9J8y
zc5VJSijQ65$ZVfk5SU@9NGR=G_Rh&sgGMp|j#F5$0y*(sA|f@yY~zsuB&1M2i$kt}
z(@q)kA9@f(;}Lmk8P|L{&hl~NY?>XE?XtbkM0^OlXv9#+2!E=2|718e6tJ%qV6^PH
z;B+hki|X=|8F;DV{M1dwnTJ%e#uJrw9tJSDCS4oYFH^IL0G;OO?1G9a_Dy?By9)@?
zri-Y})vpN~rJ{3dFRvBi3Op#NU^2uqhye5E+NKRzFs+bc9AxBYn^GJ6;O86d5T8&O
zf^~Q_5kR`-ABPrh<JzXPbOyJt%*ijGHYV}L{pBD7)lalLB!j`aFC|`+oSopX0k65D
z=UqimBSI4I&&Tyt335O5PVSNEF%BHeP&xf2NmowsP7mJ$F-Si}YdqdhvEz%`7J3Gx
z$SrNU8sd$P6Mw2DMY(m!&57AZDMfk1#W-zw!%p90HTe-Zc9p`joe~>YSwoY%sZRlt
z3eRWITD=-`E_Oc#D`eNv^{BbzcviJuoIq6ng*SRuzy-rUwao9p!vAcUEDZlS{7rq+
z_Pb^JPV0UHp8@JVVFh9O2$vZUSRpKbc=QCg!Rl$a*Q`bp4=V9PJ}<e?Br@xhowPHZ
z<qKj;-#<2QDY0olv9+?gvT}HRC(GbjW})Q+PKgmE3HO6$)FI~GZ`cl^B$&dDga+<&
ztGBhXb!F*wz4*ghi*^Cy1SRaL^uAd--EA2kJfByu=jty729jkm0w_`uP5NAgs+?vJ
zWK^{jb^5WeJqlc5t;FSWHvkNj{Qmb!E#n3ie~b0{vC9OMVMS&`5MyYP$nR4TVqKEm
zq=JHy(NCkpD9=hB@+3~2bskN$`}O+vp2_R6^`{o=CDazcrAarY5-2<(o#FE%H+8w<
zyj(%hoLy)v|Lv}|^13NSW3yH@Gh(k8$S71Xa?*?+xKIL^R)mtM5v;vt;MQq2w0Kqs
zz{4D!s}e)vWCNZnBy&f5w$WvOy@Tm&Gpm+`CQe^R*?RRdt}lFTcdNdWp1NovlFUb?
zkq-cmk-}56+^T3Dtp@DKM=DrJ=m0V~UC}06$vAh2+3!E@PG7ZUHD(c>)LFeo2jKD=
zKf1361>98q%y71JWDQz|BIg66qsztDdIg<9W(;rgvM^IDaP|PB%|RrBm~m*UbLRqr
z(isLyvRz(RagiA7wCX6yvhJv_j;mi$eci2sjJeokLI`ENFAl9UgkprE6KBr$Bf9%;
z8Q`OVnC~(Q`2<2Re;?Q$2$fkRRXzv=4xy-s<7a$_@}QxYgIcX2`&o&uSc!JoiTDDD
z>PBXQf}c~Iz)14k817K7ZqW3dg0P7=C9yJitn~2VbAiM@e4VlmUsL&}(zt|>(L@@E
z5|BeljZrc=NU0Dc!()pE#S>s?X=|1<RqE&}=qz1Q4%jK<*aeo53<wWIETWq~GrJ!b
zW(;zFXrLFe8@2|-%kaVjAYzVV`Tp<~Lkn6M$e}jXwoz6TjB<V7<MqQL5OV^7OF-uJ
zMna<EamBn>QavO4(06ca=`%ISmejfh9rq_NrVE@Tnm4Mc5|hspj%790nYR1=DMeT5
z8qv)ZxwXrFG&9iGu5yvcK$BQ)b^~#=1E<nXRJkm$S+X(AamK2q$x&}@w6DY6aa`={
z4eB-8wNg3mX=d2_F+8yBydNd&Z3v97EXiG+ew~MfVZD`vqXygj-VYJt)W!;KN|n4A
zy>i_w8;rAQqtrNUS3?U!+AMlCb~B+a5S}F10Q+>F8QHL7J?()|smJ(ZoPUNFNoQM2
zduJ@!g1gn(57el#AZ<A=2_|hxq}T=OP0zSk90P|^ayTp{!INwlJW8d%M+NRO0u%fu
zw`Y3+O{6;L$%l*kRTp*vkD4@OfsxV-9B$Di<h&%uxWOTq*?T*C!|yuQpmY(PQ=buO
zLFW%028nRqa>E-4WGfy_8k~{E{ArsdB=blD!~|*(qdTy9`0DOf7>$s<E{x=9&i*j8
zVu|ddnm9?Ipb5BS8tVrD2@<7156N9?`hs<D%J~jF1#EBV2Q$Uq%%V%74JEghFA}>9
zmbk`~xZ2<@UExJaY;Bj&v;`Fuj$C5lS77gp{qrQ8Eq}xZQwVq&RIywhLt_YO-BXl)
zT(WZLc2syE9htHNscmn30?QUfm0J&%$|!QH+MhZg*bN+!9lV9BG)by>NsqmJ0txWq
zVtJ7A7*I~AzH&oGeb2~IesdHP*Z@wMevlX+^S@7>Ty152q8;a^rF9M7mHaD=w>5P{
zAjnBj^dI6JxPMZx_0rycr=e{M8wwWX8JKb5hS33V5pZ3R1%L0O8U9{z&0^56>&u0W
zZ5;K%#jHzI;z&xuy@mOUgEbyncu(SkE4x!jp+ZXN2pf|U+pa#NsE{Q;{RPn^jg|F}
z75yLg)|mfeklKGc1X0S8VE!AwAZe|_xl&F0yl29NDmrdO_KS#$;_P9>RnQMrQpsm=
z%E<``iVmY=yQM=N%2OJ_GqEppb>}3Ad0M=?XfAWpO~&J20EQcrAJ-ld;P~D--dtQn
zE4E>;xaKs1;#3c}sn5p^`;oJlMWIP~v+5E&Jp3J5Qzbok>xZ$Xr46U2i>r&p-e2E)
zr5$NLtr@b@a`cnvpcheRG2#?_=2pB*d~?cfh81y7#Z5c)P{@F1lguE|Eb=zyOa}RZ
zmW1SOhXc#KXlQZR4*3R3_20jx%!E@Mf+P?1AFLC}6@Lx(oy9Ji*w2@0fm&4zAGp%g
zDrUzHBk%7{!T(*!zcs)|1O9^ys2dteLySs8s7C%qE~wsKu+`^O&lP7EvAfVzu)|qi
z_7h9pwXQS`ZSFI5N8-BJJ75^MDJ6vVjmbtwPZ@?q&Kc@phXE7AADN5z7tVD)laz$G
z3U<OAEgrjP1=&a;<$JIA=O%HAJ`{IaAt3If3=UZ$3nLE0KIq4~$q+(!JIciy)o#8m
zQto{@Qq@P)Bh3ytHH4$IxI?W1y((hL^ZZ(+<(&d@+0)Dn93rOEBQP~hs=5bFx=!~X
zK2+{RJKXh`8SD-BSd<wAk%@z3y2Vf5b?MRu_vFB^*D<>FvV>9UkohF|h0HI8;BGD!
zvR(emfbzUEJ1$C*bYfjhsGgcAT{uexdbAz49zOH}KeTdaS7(0;o;9>`l*kd+g(+S!
z3bv%U4ng7lV!?C30w-j<W^InGMH%Uj(CmS#!w9X66uDmG#u#184Y|yBv2F}0CR}eT
zCf~f-$j4E|c9K@Ltv`2=3`Q6MtsDMqk?{d-&n}4MtIp5;<*jUL7st~w#uqQ|_rn9X
ztQx&qGD+mOf*-$3L%#Xx#7aS|ISiDMJ?$eaoy}#X*a{Cv(B#6(-M-lbI97FFwzhQQ
zVzr0mh&4VfHE)Vim{&7<(hCgL_N|7IV8V9M>YA5u8$MmY-AC~6xYpug$6;Qd)qW&c
zajT^D<5z09zhn7oCSVBia}(Z|3!?66LpT^VwR+qjZn`;F<5l6CHV5ATPqne^<6<mC
zv_nY6x=Afu^bXmi9>+cu6-Ct5nPzW}k2pk~=7+P_A)_7b)B;z+43pdK_<3I+d#+RS
zuJb26E5RD{?e^mKPAtR#FCsIN87kq4MuWr8V8U>&r(Sb-BrfQs^SooXtBwe#YQjZY
zq5QF%dzx&>dp8wVyDFFJYTQ^dP+(J1$Izw5N`@k8L!-292pslJu-~;YdD4J5>r*Ef
z=uoBLv*xAtrCzR`VrVKyENHR{Wf##Psx@Vcxg-^0jAV4<3Nd#$W~ne#eBlJjpN|6r
z@F-+R$J*~>t=8?<KLXOzv6aYSCY}$K+Z{AgGD`~wNS)vmdNeww5%LP5d~xx<D7ikc
zP8eDZe^N%#zHd1@ZaotnY3~Dtscaf&Tq;%Dx2c6)<E=iv!7yH)LwY}Es{KvvppFk1
zZ6z|!E}YT7zjLWcdq?RnvFK79&@h84Gnr&Ao3@Q9^BA<AQ3AX+&|6{RR<qIx#`1$W
zEmA#H%Y%o8G^V6Izkz<jJ>y)Z`7M3jl=3;xWGMcT=&gHeFzLoniklf+CEJZL)k}ff
zha73>OBF<uKsA8X00y7TnbT>-*YX|S0@3kiVny5AwWMkd5k@q(`J?Ne8mN2uR~!r-
zTh*T$20}7SXE);Jz^_K?obPOf=qv!#1I)&72?YcO2{hTH69wG5jFQC$H~uVSnKI@5
zB;>h$_DN3*azH0Bs9$qDZp-YpuiM^`{=~s!lf>?KK_^0~Do~5xn&zSy%2?;ZGjVM+
z%6ecMw;#Ll^X^rx*qMNM7X@|Db`Ov)TF5`ud8U6_R%2%Pk8fMP1IV`DaJugI?_n44
ztjG3-=xku%V89PjTHM0}2yjtFCdF`N$+6&of}XAh`iP{{H~&^(7=p&E?C<9~GhKV%
z3~$%W<;lgz#YeI>hGi~c%&a>iEeb3Z$`NTYbyijeI>C$y4SA#?5`ln@T#4H5<LYew
z<L&J7_;qobfI|0(KU>#fiM=1ZU#K5{Duht%6!zVkB&zX<VGyYf2RRsOdDiR4Snjy7
zPNufR#@3OE!2S{2=Am3liZlgaekoE_rEAl6uDMK#U8PEHX?pIG&AHxqw$Zt61?Q4|
z$~k2w=bH3Il&nA<F5B~36){HyjHCpLQ5u=l92wpdMOlQOFiZiB1-?)WtboSM3s{#m
z#uql?a|JoPu|SX+FT7g-gzypO*pQ7Bxl-7a(ZB#mp`eeI5jMylI0dheOhHKUU+M%I
z&Fqdwd|s+ui7$fwWWxA3y#R+BjSTtmmw0cEf(Y^X9CU0W(HGvHg`yVNUkH*D1yn(~
z7H6v6=XEQ_M)N{^R`>Ofkc2e9^n{r#1BDTx(ssi1vHH1^F-Ls4y`{|L37QkMotD|$
z#VUmYC^to~)M)J5S2s&+vvx(dN(f0v;?o>Z7%pQ)Hr@bC01A<zE5s)mYYVdT4PV=e
zCY+1PPK}~S1Eoet3aumKdGC+kjV*<IhOlMfeXBm}7p~*Ux7#BJ^w0N$^N9XL8a!zt
z;zH}zZM=*qU=kvslsYo%AisZE%?*)1GoTwEg`)13-UtH~d0~#pTc7=|TT4yfmAI$;
ziQLjSH9v_-LMfh>B?1?~W}?cFWVTou%(&Ld0Z+1#c>>u%@x_i#rVuLu@})G`3)->y
z!N$KEVF|&(0($Z4mCGbJ>suY!!E>N$#=j-Y@01NSBI=&HEDOXS4Vp6itjMGnM&^r;
z28H*7HA#3ZWQ8BH_fr_kLL}kf5h;kSFy>RZ9|^NmLNLGOn~5&j=%k+E?;mzDsBaF2
zFU@>Wd-3y4KDHRkgJyEZ(ENcWMQlmm+!-&$#H_BdLb{BXtJ<ZbKnp=LFA6j8tUt|>
zKd>=S<e}-OHeDq8#*NX$x%E8|`<37uCVQD%6wbry#-HKzKD509{y)|r@TWwqWdWI#
z=SqPN_@<t&j}88Z1=2QY8t|T^G9eZxN7mhe4?BFq-2)5<+%h3vkmcj9ao{moe)7f5
zL@n@qRj?bvCz5!(IBNbE?<w|<i#&oE27~1nB(&<3C%Z1wLV~8k@P0ySJF8TX0tTE6
zMD`R=NAN!Ol<ybcL6o|S`FD0*0sKrkk?aBb6Nu@l7qwG}-%pzzeN-q=7qX*%&|0~T
zE{IN5ynyD8=q`bh$G%CpA3^v|e-lCS-@yw^w=^9lIQ(;ZHGdl-e75vZbb>H@6}RCR
ziUue%8HrwIY?u+uAlU7R9_HX1R-pB(x5iD~F(J@<o6R4I<1w=2m;K<c+9=+R;3iEV
zL8c~*e~4EQCq+KrLtsea*(!mg0E|ucQPFvg5^w-um3Y72$wT2V{m*{2Z#T=04}bU)
z%g`t}#_A0&BK_2OCwZIZf3h?60|;%kUJsRw`PfeM=rKgo1J)#!Z0T(4+gCWJumOu_
zN%_#iG>Tt~S6fQ&`zQdCQcusO_qD_x-iy1VXj66kvXqVoyJ_SXVIh%6uCw9ghK7^j
ziY<Q&6%1WR+rO#ju?_L;qqn|72whn^Gi2e|-W*-Ixu2l^FJu2;@ML0O{-;}o|BoQZ
zq}r&%ccJaN|7rKlUNZAwo4|Fb#goiONFi2iqohM?g`u(`B3iAKI{end1L4Nyp{J4@
z4+<8}&Tn2n3T4P3XJoZ(J!?>y22g}x4k^#!P-v0;q7(|2@fbL0o+Rad+jw_#fLIW*
z5JQwjTeFiQ@bPwC<?iA-V~|{j{i7&Oz1H2?diB)alk0IC8Y$m(Rhk)@Y5Dg<`nlb8
z(c5?aT*s@uS*_8`uHu82lQ|9uk%X+l)uq1}F9o^q+r=XxWsytKf|r-n+;KlLYr%Wh
zB3r*o^8@4f+#vaxAi|50-|L!9**{fq4!9GwSJ$IW<DtQaZ!$8vg%fvjd2!%HpN}&7
z+Nebd8zsn)bV}V77g=kbB`BS#>vwq`N?wrV`7p<9P_cPAvcMh{U^R@PojSTmQmT<{
zY#oRUSq>G<)W^aOTf1*LTjFOXLZ%ml@Gtno;7?GU9}3%jmakdM5k)dfoqSI>sLzG&
zKo)h>ljE3nVy)#oD=Z=YMhtJX3AO_Y-1>Rw=%D19rRuO|Vj%{iW4YV~YAvF4(o$s=
zJa2(gYnVLsQapGaq*jr4xw0ZVU};<Wwz7f+jK8UI+0zk4nCirfJp@FcQHDUnD%dV_
zGbb3MA{{a(>+2p{EU4nCz<N;AFp-bF&KsBC0LH>4(4<k+{fzrUiXi1O5DiP0rWvre
zha+HJ!a82|70jB58Y@ZiiRVkU-RmGQ#62@<60*dIcz8VA{D{^S&L|!qh#rt;0;CHG
zO9CK9E&W>NbWd)z#q@JP*PwC5v{ISsF{1{ZtJl_BoccoSFI}vG?57p$)<29bg^-4+
z!OfWZZWj>W<l3+*_vBEQPtKVzrT0}sNs!lnJlRvbo5$CIJ{qK5zsA#956YvIKfs^6
zqLoFN!@h$ZqL#w?Jc0zQ{n{!vl>w6@V}sK!JT4)^lQ5Wo5GDgFpMbE<A_HpA`Ei#(
z6ve!aI8=uu1!8>BBzeX4*8V4A){2EbAhws$P;568GK4<HD!8McUndth@<DJN2Nd)5
z<JJcGaVf)Czf_}&qv7<9kNau(y^xq&fL#DRxvV_M(u*XHnR$5}5<oIzO&|xUZ;O$h
zC?mG(-k$a{FgIG5JqQEH7Wp;%m4NXqo1-QxOMS6ge<g%=zxklgiM!q^)cy6uPb9@2
zi&<bx>jX4H0cMp8m96i~sPay}&UTQ60_;R%K_e~_fOW*Wjd$DlnRYl|$qzTa$(4)H
zACL5g%(eTGPrL%hz3Ci3dH%lsR!pi9U2l%!Z+D`wLNc?7>bPpeC0aum{o4qx#=GJ?
zJXU8=CUQ{TXsa3?=JY5YhPK<4fayfS7$MTWiT+jCM=!scw&zf7*FJ$iWqs+5PctMo
zdwMqvP~^+|M~bKFI@+rj4r1{91gMf~aRIIv`>O{0b{8lj1KD^BNK|SA<V*vynidm6
ziDTXV+M((E2CR@L=SeRfIMIWrDzAT@4~*xsu7BR*y++v+W#RiP%$r6jVELcUZtctX
z^50ZAt|1~e>UkZS)K{;)H3+K?g2YyEbuNu(NOd3eI8nk;yEF_pN5TU$R!xGP5&X*O
z+rZJkAq0UYE|g$#J;2;{wSKGlkk#P01|AGV1qJB!)D}qy_$CPRG^BXCaOVx^tQ9?N
zS6u+lB%2TU-_I{4m7Yl{&arbh7)<0qm={|th&#rYJBxPn$1^6|OK=eU&BY_0LE<>i
zQXnkRv#P`_u0y<9lsA)j;F|b(jR(MEmOQ;`{*dA`bme&#%3)Tv0d*fpP2z%!4f+`F
za*`GxciVipPYNIU1NsavJ#dh4tu~7%gcDPh;(`N};^vF!uuq4p4ri)!R%l{gT}o+P
za{2}yBCgghzx}!tEv~dbwmb&_IKidETlnF=jbGr11x&>M^ltr~^824tadwvf7y~h+
zZvC&R_}7(gek_}6gBKbV?0ap2Y+S)f`+_iIk!apZeW7fMwxryV{)+9kA}QINqE39%
z4W#1Ut6Qh%716a_V}zX(j|VGG7Y{K<IC9b;ON_~Q*iifqW51S2gQyc*2^bYidIV-H
ze;23w<9df{A{Cn3$J@dF1?rY5MNen0S9+JX`(S!a$d4dVSO6w%fO$Sr^N)dwAW@h>
z!;+y=IUgyqUy@c|&K9UjLcEG;sdCmjI@vXIk6gSpW64PHkHdOMN^&2}=&fUO1vxT?
zQ6uJcT{609b<eNFmcQaiK|6?{47)gDB+e;m;sTDX;KpJbm;iI+pwD)!a?<~cvUBVX
zt=raYY}>YNJK3>q@7T7zlkC{GZQHhO8<lg~{cx+@x2jHkS#AA+wZ@q98Kd`4>4snp
zmG#)lQd76Ts_x`one5489bMARvVl!glv*pN?wTtW)`zjCEwaz19!hAQ?B$iqPjg$^
zSqvl;tVdU8Fo{e;JtFs3WepX{>6VvdXch@)h^$~Ia@6gT?{Ch@Nitgp8rEC%5tAR&
zn*E^jx3fh|dP?PV<;N;r)^{=%4K_(b64OGW2!mtlDXPdoeCw6<b&M&@K7NAS39Rj9
zEZ37h>5KFeEmKVFDb2qzH+tbA5mm}du(R=E^XgR}phVbJzFbWoeL1{+t;EK4RxL|`
zUjm#*`e5%0g0XE4qJDg4X-L85LNF&jSqzZ7Q;`pOc1eea8*r3FX9UbUu|YW^5M?Tw
z(TRjZDcSM4yh;IpZP{e;vGpH@YA9l#MAhj~TfE{az98d|0LK&xAQ#CLgiBxWi$_GA
zBat1eTA3OeuJlv<z@tpgV?W+2(}&Ty*2!e)9xXa?CGsEdm8aTT6lqJ!7~Od(qCq?u
z7s}Pi(E{b4G&)VbSuO~X@Z1!2BOeR~ENTDKQu0DUc70|^1hoZP^=(6Enn7lD+N*jl
zvG|A{%XFAY6H_G+q;~NNbh`-kIgJ|p4KP1K1@QG5F%P0N)69X@8GC?%(M$lJU8jNE
zV^Q?I>BPw1V3c@m4==ufIUK~T-BKNBfv|8{cRNrV>kVn|G?E3A@fl4Rlt=LtJ{~Y$
zh?xUdZh3+=_v2VX21bz*t`R>00Y5rKt5f-4HsUToam=(o`lv59QYuVer)X~H`9OY$
z9gAtK9?aO@xST^|k^y!?#6=v6c2`r&`StW1nmq?vGTMG==c-Pdl{zB$;@xYW!@U7q
zkH!N9eglfTd<FDjh{ZfaZ&^7Tx+nVPB=J$^c!ngmd&o9Gcb$!woNR#D0vP6RTQU+N
z4BQ*Qr@(}N;s%P?!bx|UHXjIupI89pRdjXrH~dG`L_HH>Nn=D9LvR4s443p)_11F{
zdOqbt0jkg$+c5(G-f<woo<i%UIs|_Fdk*Du!^0~6=PUO?sLBfZv9?I6m?VuPM0Y_R
z&fxwDQNgleA<q;k84@W*7>E{)9osCa-1v%~&11>JS!gJO4gHJP{T-DYr|#TnfyHGf
zjXaqytJV$gPGcysg%iE|Lt4fBF5nxGgRo>5A>5xLNtdq$at`jcOyhyZaKOO&iBD9-
z_;{Jbj?GzmrjvF~0MSw2%_lEa0eip7z}uuY?f$}OL<na9>uus&J;Y%*(oOiK_owYN
z`qHwK3IeOBXr00+*Y?V2at-~CT=Wf<GS0)?&~V2s(9B5&+>IQFE!0C=ezo)#C%iyS
zf}vAeNa`g&+#nc9w5|>4Q&e+6vJ`*UwNf>$u_o=jBT{Xkr6-gB+xFbVfydCb|H=U-
z64nFJetd7(tV8qIRnhzf3FRvsClK{^tpvnwU<i3vRVfT4p9+G}q5u{k4Fl()j2ng@
zKqi6RB97g;aPqlQa41meK^|iDCnp_G_$3#|{0xBq7_`NqW)O(HBd)nYU8s9|+p~52
zR;iF+cb`GA{?~aAnzBcYh6|1r*q6Iwy`S^mlk;D_)`-zZLy#*EMj!aTN5i1Tw<I3`
zvX4KX09*$q-L(SQNS&M3yYq$X%@Emgm_UPV<Aj8c{-NuMtp14;U`Sxr#w3mYLGTC?
z8{kNM5B`Y@5%p|wH+KF(xzHpqIs}E67LD_4aULsRb0=&Y4Vy<r^(W1xVM85PL+1U*
zadvS5RicgR2e35Oj43MTZ>?CO${+YWDe(P5J-MXa$TSc<I%<u<phhC#*-pL7_($!J
zRT9zFnB3;hjYvASWtqz_GkVWQi+jSpwLfpPd=F~#_r`Ry@PvMt99vI{*JIz}p_tp~
zCU=Ps*}i6XLw0W?3{yaKIt%d3s2j5!%^efC(T=bHN3$I806E*};K@{1fIZ7C?6EPY
z?$Zw)W*%2qzJk73C#fwt@HgMl``A=--vD}`q|5(UCjTjFU}I<dPX~Vg8|j%sh8E;{
z*vVsLyvn0SN6Vr(!T3iUwrceVB}wXB{O@_IpVchRm~u&RaQ_3t+Mlcen#Rr{ednO!
z*>J<tWlX~`?vFCrzDA!hQHw}Qq%n`o)>G+D;PZZVRwj~_reO$iH0eo6u<qB(hr`F4
zB|`zwGo@f)Zd{A6yFD*ftdIMpZg&R<e(cn?c1NqrbhX&>e)()9`~K|UYoktHXs1FA
zT1Axs?dYYZFfl?d8K&ZoP|RlZ7`W&yd(JCsYl6ABb|M*>zks1PT_#0ou!8D`n>-q4
zE&^*G5kZs5tsrIoEAc0Du)A?g+O|5nhrb^+<~moTA!LyTy>(UBFK@<yAszbke12|z
z|LrYteBS<GeC?dNO{X{`xFLnehk<FdAA=nK16^+EZ4?R+E<vk#c&UN|Je^i+wy8;L
zO;O$*uLc%=pCWuVtkbMGUM12}wXnYqB2OPQ%~E|O=Qm4BRsmEh^J;XmHBCP>em_7X
z^ydFWBcrQ)goJDw3!?TXcxy~O)M<NCUlhFQKU@&zlx!{$V(=O<(sc53XHnI5RscFF
zcEMDjv29m|3JjYAf4LD^k)cjhCK}>7Fk5A!o}&ty2fTHMW(WyGX+`&=cXvJ(MiwBz
zC5Xcc0zCB$Uu(PGbo=9xp*G%%zv)&hlj;uRC<h=xyi}JVpsQf!rcyBO&_spyh*I@F
z-ypR)f(aO?b)Dh+(IT0wqI;-unq_-bK8jC@f+J@4?WFRus;i?a8{u`S0Yz__w01q{
z=XVxe_<{F}Q4mT>%i2JA#6w8cI-*dn=-pX0>xIG8-!gR^LT%zlNF@=mUVy2W`54~}
zhlrqq`l?5$o~(9kW`lB97G+7a5)hLF;!(WOCn6Mw7|>Mb&!mrWpAO(9%KZcb2{x65
zwt7HUo^e(p(NLIQ*i>*dVh+?qjUwGmqso3N!#7PWS8mZnlO9@haB)1$aIzO>aF`e^
zR2V}73qo~ZqNL^|)MG6Z_Sa9ojBMJwS3g3GOEZ02MLy5Uzl5ElJP<$tMshFuwW{Bo
zA$Y`iCBDw<ye1W6fE5-x3A`dn3g_QrOr(m+BC6Re*iqAPBd4?)T5d`M(nh`B#n?Td
z`Q4&7mDmo$uv$<dV#HCsf>4$quGz9WU(3q2XjU4N6njyP<KD&J0yJMW?=GQMvbf!@
z_Wr$OMXnGykPz}q2)hNpf*XY|Y9$9<FgCWjAOOl`zC(d8P!iYt9JZLspFK9-SD^p?
z#VGNUSHma@>+7|wJL&X}O`!Dp7)`<f_;jRgGhfOgU+&D_M09o)j3OC`#)9Imi~5we
z7L4LJgseRl!}ZF|T%jl+xRDpbV5IdkB&S${aM?6^(O0q8KSlSJ58L5IA4%fTTKkp5
z%aFFj=P4Q09M(&zfJkrg>%#|e<7}_yS3^v_;A9M%6n1KraHP;28Im&Dkx|-b>s0{c
z8!QFold*m?!8I{&96?AM*LJ*hhq6K@!`MnR>@OFX3)X`omSI4AC&qp}a6V6#m$TEI
z@)8~jq8>=uCVQjIb##AsM3j-g@@2jBdAV&u69I?E<#r;KSBk!dBW5~o5D^DS8r`6X
zn6^M2r3u$1D49>;l}W{Fb}l6W_*7jVk%MYJ*0-B_5J-vXmqP2XEyP(f>_&ahd#gq8
z5`J$>rY{0h=b0I?e*nkI-1NL*crLRVknwKRM)G^^6FX~hsQ@X<vvJBssAD*m>?`2%
zHZg+n92-K|z~_{PqkGFU$?)C+pft;8Krf$aW>Ymo`qZ1tzMUF}2hiU14=c=%H4Bn^
zy9lJQ$6bU!<k6+A&R3i)6B%;iYy-r{!^dyd(sLII^j6vA1i??U&#@7?)tBQdWrZy@
zU9n~^*`}S-Hw?V%@g*P=5o%Gl&f)JTJ-F(#2G#ht<kW~UtB&CzGa`fwy+_<j<dsvy
zKs`4JMkFBq2xszD;iAOuB9p*n2obUAsGZj3n|Q~Go>5<Iq^VcIYq#2)mv#^SU6()i
zNRElSd$!`(sNdc6WW%xMe4n&Jh_#v6`<?pj#BNIcrFOg1!d|36_EF7VbeXqHC*8=~
z8u#Q~iiRhSvCw0FY9EBo7;|G1X+%u-_QkjZ(eFn*0m0Fy6`KEfVc*b?C0@L~SoFx{
ziXM4cP6<kzq%d^_S%6KFW2VZLd9+)HhLM6Ad8ghi+R)C|v-+Y3Bff(9Pt}d}p8^&Z
z#{YOUs`I}_qn`gY8WoA*BSjLz+hCs1iqST8ix<*BVXW2EI9f+NdddUyzV3=7snKA<
z4jDiMpI176d*##pu3yvqP~*z&#*3-FjZnc8tSljHTX&;=0*Oa75zL<I`OHTerq2TN
zGl_p&PDnWLy`S8izn&cmh`Qp-+?F0CUhSJ5<>Sr&!Q$@f-LIjBjs9b6gPR`<DhUQ#
z$$$?<K9WR$*Pz8yfAtfSD3kY9A+axVRubZ#`!N$#oQeLKP?daiXVWv+P^*^@6ZwPs
z6$V<>nAA@&u(EwV-2FD|j%c>7;^kbYQ#9@j!zB?rj6n%GtpAf$=g1?<>HCueIoE|G
zke7o1i4%PnVIZBdvz<q;+H{WZQF$`}-8bMMbyLlqR!On9ZfRfKt?n+H=4N|-s`oV?
z$+Si^i^=T=Y8*M<d!U&eIqye?W9V;*L@md}#h@c_(h-wzEbDe))s6Uu&C_JDd5GX9
zc$p@lg!4f#7$&*{{s|LO&Wf>e0b7eBChS6Ph-h1JKqMMeM%U{7u&Q5(1S4=$eTt6c
zvf?P|w-xAw=DctsxHhjKMDGa3aiKY=X^30?<%h$=nS<$(;fSJ`YuzM`@WO)}fOpU}
z2q4rg3iw6?m9Y<yGtdPN4$+s=f}<~$6V`^)lLWj%5JC%n!w6o(zh2R-Q73yxMfYSn
zo(syC3+iHS&oSUtWnmx0od9ERXy0tr3*Mx!L|{piEd~9i*ue0C^=uF1A><?-PNbJr
z`FsaMm=(@_IS?;6OQ<+}_hZ;1s<PsEmp_8aQ!J>QS#Howy~zLq>@)L!QK(+774%3j
z^ZXoL2)i5`@Jh#m5+JILIZY&rkRhy&;GjoOH{A5<FXWFcSFdh>tWn2$o|vX@=ct~e
zl9BrKOtUui{mVee0HkXRb>FCvoLEn^iIvwUCcN<GjhuB@W%>!JP#bV0*AQ$?;>1&~
zV?d#I*(~X9B9yR#LCQt>5C`ZxgVkYp+IW6@+0$+-f-m5tRPeoG?v2si?W|PpWQtk?
z+Nq8;lFtayK?*`iXikn#AsnS(2IB||1p&4ev+tc$Fs>q3coM-KiW@d*pi8Ll9fZQz
z`kcEWy>F3L_?K{%XLS5YT+2;1fu!bNr(qadVy$XiNzmH5Q4LMNQ~yZmB~u!L9s`oK
zR)xr#FfJ|<8}jj56{JO#JWSl*xFkPeHX-BC$BfYr*+r}8z&)<b61u6fvF8l!RrZ9!
zu&hyk`y4DtMO55oLCi(uY-!P@?Y)Z-%M!jB=w*xwa?{O4j9^UkS`taLW1Zw33RgU+
zd4|Od5D|1QkzEcem7uZWoIA{)cg143^A_0p!{KE|z?Agsf*2_<$ziNuWHGnKsROP=
z`Ag25OfXGy!cunSUCwBE=hFDbME@5G>C6k^E2TjgP_OBE?<e`i<nCcRm+LOjO(UYt
zvT2;D=|o3+ukBISA%{G5WWfW4sf6M%d|Y5EH&N~rQM?_AlfYrzr~(?G8-gkh=6*tV
zb0;C8U2lE|JU;mwoB)3A(nGK`k=O!agm@D2h|&%!;xX-U7_fw(_?ZU=f{O%ZF(z_Y
zJCD#O$xqujk<YPEOm)1#tRaY)T$ymf=-hZUTEa)Bj}p96;_JvX{1>%Bbsbdhs-G8h
zQM&&75msU*u#Zg-id`6Y5XTbAaVY;+zZ?P*L5Gv)21WL1(gsE~b<ydWOO3VJj&;3n
zX*mn9LToOI@H!t0(E0qaRdXi2W&7st*B%5oGqzfv;B7%FGsan}$DRo|e+WVjdRa^U
zb5+EB)}MVM=sa@mbhz2Iu7k!?A{7*r=r?>0kThbHe6d?%I#F?9>13=<*Yx5IMTXt!
z){ZOyD488!Yt!DefBLuoA2_f7ad<wdKK^5NL)kp1ZYe<c&2$YOL3ihb3erS9zWHyj
zTVGNhO1wi(-bwQ_oSDQQq(ct-05sh=BKqqI-s6p74DfLP@Z-Sa`QG{E3{WskNfU-n
z*c6z5N9WS$z3p*OoL>Ci_YbVSMJA>kbLs)HV<&_3(^W;)Wc`Rgni>#T@qcM*?mT#K
zzn{i<eeB&CarHK|x3t^0>4%r+M_*nioxI%bt&=hIyD8NpYXtf=<5wHSd}<L@B(P7D
zCiW?uvGaqPqA0~v7T(A5b+E|CWAW!MP3_P3PhM1k81pROw`+Tr>=|(+qxv)3KQ8Gi
z(M9XDMfIgK#^L-x3?_OB;Us!M;zK}ypbG$3PN9}ke4+pWs`{Ne8`#OTx7)x?K27(U
zFLmjEsIg<2o+|5a+2aRvQyChJv5Mc`jl8byBcUK&YgX|S)oglZqL?Oyp7^wAl8~}i
z1c0T@w7@nPNQrH<$@X1TJRUibP`tW~fj>8H7>Lm&xp4hv!bN@U6q5NGjraz;V@F>}
zL{x8X=G;WveK#H1dZT|3#6Pm~p2sg@8pJgH(I7#0?T(iepk^Dx0`=}MEz`wN+ZL}y
z6%*Z=){{oL9ObCkL>rMkOII7>+}VP+ryw5~t#X0al4pkXfU<p*JTisw{fxGey-Psk
z<9OaK0KAjsY1PNS0qNl<poDxSR15B3mlry1&A`g=F41_wbVQif=se1fB;N9hEn}2J
zi7m<(EE9&xIjHl!j$ybO{6_(;q1gFcr*ev978`IEIEHdK5H4NZ@YBUfc(P>ZBPH?w
z&KxcvCU7oXD?Q3&pXkApN`ol_TR6A!$JTSiN`-|`Vu@YV75@-GBku}oUosNEYNun)
zma0DI?FLdI)Mlr6warSq)7|T>5$QK{;#<$>s=Mr*oSh0OAWICx45rtLX`PtN^%+Gb
zU?9y-J-*1J@FVVTe)=VNVpolUK^{3aS-OcCDY`|^lRs42HkE5130Aw(WmlNV>l?m`
zS8Vq7U<wK+b@mq(k#4%2&de|$fTV(5Flq0julm*8_`w4s7*}~Mz~JXRuj4`npiZo<
z9Imd->lJr8PGBT2AemG=AJo`Dv*7jUUJ?*YLNiIGl=@R%ox!hgFmP-vL>M5?E}b}n
zA;N}a_gQOeUj_9$x=A>gG6X=uFt>=>hvynBzZBVlHL@oK7KBQwwHBs0AckV2_uH^s
zXkd3#j^!8jAtMUW8hbUe!6k&()F&*VaS;II*ClNH$z{{RSJ5F5yv(pGQ#};96`BZ?
z9jhM&{VYRBZf|)`S1Un+1!Ju=GFd-onrpR2qQwL%DRL^)$f9&?F?ynAzf~}rNTNyv
zLC|mnsZ@x|yiNW-H>v8@8~$DwBB0`itwskBLG9-Ex#>?6dl%0}!&vbBbmP1w=5z)Y
ziceSN0ZCaT`&h*;PhuZYhXsbz=BT`++^Q~*lNaG~2=<He-;XG&90ac}3X}_M3sN|_
zkOHEZ@oJgt^<s5<#si0TdmTc9i5os1a{{%uvp#pl`r*er7f)UVBvPkn<s3j*Q=4Gd
z$wCM;!VvrcC6fVB(W);C0Q40hT$cHSY986F1})aK{*8%4q!GCrxKj``8pcDzp~!kR
z-|J5wGYA~8KeLcq0yTFND2dN6KZ{aJh|n8rImy>8FoZzF7nHv*MbxH7c?uZN`%whw
z1ZXdmwh0_0ABN%RCnD}AO$@_L1jLodnYbgkWR$)R0G8Cz*h>5vXyb;wpMa;_yOf2h
zZ$jD?$$%1&iY8+2t{96FB9sw0Vg*@l@}8fV%jGQEe?u7e*_0qAzE`52Bf+p$`T(vC
zDcNfV%-|sr8470nXpzJ3NRm2zoGcz%lNUr;lQ!ycv?Xh+#%i&+@GiG44NLsTY5|B?
zYV8sgcf0Q?c}g0$Fg<oQ)sa7PT#306_jJOiL{(m}CcAWabz?H#W~8FcUYS>8=&<+J
zdH>7Iae8K-%*L~Mvd+0HV(#<C4gFZO4lmx$H|a9G7E?U;3yNa)vUBMsO1s@e)hEs6
z`w(!(TY^1OD^m8k!wsu^IJ>;$6cyv(w3{RXKI@;#&OZ@e*%|-KC;A@{UK>@mZGQ-_
zKG%A?!^)aG-w_F6<X{a{LfB`0^8zC>n@dD8QhP9{x5JLnnHA?y<`4_YeA}7VeA})~
zkeb8+Mfv4cPMo<UqLi|v2^3m{bP6OSWJE%%NSmNhhElg2Q6!LCo6^EB(>C?1<@O^)
zNF8AAeC=;>lQvbu*>%my$i1<F;SdQDLUicV5%99ad-K^b$q9*gm2j;DDFcX58c_+#
zZ3NA#f9YP!XdQo-)!C;~9#Z%Z#=Ay>$|6Sy{fKLPEGkuYy)aG0yQKAnJmW+gL#A1l
z4zeRR9h~XhfIuTct-VPQIfHD5a!E*5+IK^-V_?cwtQQ86ABD6)2~Serf)kM@lu7_h
zQj~V@Ot3GZ05~$`D>uwzX$-odi#5*b7+t<sw&j+Obl=n1TB*FE;HuEa3)yWKrqbbS
zysdN8+SbD0nC-wMgwyoFoLs{6V%%{ysFg}Y@jDq!u0(&f0?J^t%r?LceMj}bpn)Kh
zEIz&=3sMjOEq?Sd;&BN?gcwOU@>bwfq7G6iVSyc+O~G{n(f8^Bm28nuGq+Ru*S==g
z9H5t5GM=)6^Nvt!SxPfpvVzAPpci3J5~N5tWpYZPckIgi3=h?%S=sxTSN<<Jlg_O^
zY?ISw2|a)B<cYH6X)qi~)6Nu^$6QR3{OND~7Z_pj?men4nK^uzUG@w7N!&w7IB|Fo
z-~sen9Bc@=+!V#zV9M%No@H&UusfY{rB69R0{ddw2$iHI_t*@6tFV=qZCP$({WAG$
zo6)7U<dg67)~Hg*x^yx$i7xur*J7u0ClMr}?5W(FWAmVKHS>vP<(x5i=2A8JsR7xY
z#2!XXKh&-ztPOOnsj<8H!lyVLdfT?mmgjOR)YEY%G#_H|H_OQVDa598)$@+#;!*q4
z?e=tav6*h_ZYdDWORRVR@391rJM6Y`es<r}bM`-e8~-F!VP^P`{Z5PjHP3jicA+3P
zG1-M5iig!eiS)3;*4FATD494ZM5Gd4XOMGZdu`_`c{!J80xlDfMR7GX<->eG8Ax=3
zhWIcZ1e0}bn$UbrE2?ZHRwz)9x(5YmRNH_xnRMbYhSBxH^S**FxQs~}AX9zjy_)zu
zC^|%v`lZQ_(hhA|wA%hAPa9dN_f@*V-tORIKGwF@Y`N8ij-z3Ps&r`rqE#+HD~Mtr
zjw*ntXkfB)2dmB3?)Y}<guOXb^xHQ*)AWAd+N#YN;>?|oewje$;&ZkAgU@^wziHiL
z`6apSu6QGtNbh~|-OKv~?&!XaU&A&6jeqB2))L4(xJTPD)+N2gSV{HpsOFYl*7{81
z+AX3oapFBdBa&%|@wnod7x$KraK7A2jiqt9Gb$qkf}9;Uldgsp#RS3eS84Z&<+A%F
z;|Wcf_jetaa-{4lfuBY(%!WXs+LOT{XldM^0?>Pa=mfV5Rtp+d=XWF}`8WlufZLsJ
z*;v6ueH+aNY7s3T3MUj}-KgJ5+Gb%FSMQxl1-n8<6$uO}RN)+|EIU5ThWI%lmo)KX
z<tD|nXa#7{04Q`5;(xFw8m>fS^Hj|gA{|RmmuN)2Pl;DIP*uV#I<X_;^;5Js!{&Qn
zcA{Cg`=XX!xfu6zt*(ajh4U~>ru(ltzrhV0q!AB@8PCSu((e3?tVIXreyQ;>jlm>E
zwepxHg+7dXBL+lWd7~1B5@)0nJ4uQOgI^#`w+k2!S&-iyry#`u?x->Qrm)Z&zmk|a
zL21m5I6z&ny1ey=qa5sL5b&oYmPN@dq@0o4NXZymn+zvLh{zBnB1#h|FoB9Dl4qrS
z@T7mRp1y$~BS8g*G+8nLIz>(%yfF*@TKRw=%@!u(<(P#vgi=};vy0}F@wxAG7jQJl
z5f5O3{7gGNwzfkXA@AZnBDhH3LbesNQnr(n1LHz6$zd|KFCX?G{D}VoUI5o^m<lYO
z>-D@X+4;55HJWayq$~pf@57cOa!Vdv{47$G!~V5@h-vbN2Xy*&Dyi~1i`bGASZa-9
z5zm&ODp}XDYisK;COAJo>Qo#vm?O7k)BHF9ezEVknRy5nH+uSIJ@1q(ZM_!rXqH}V
zGi8vK!GDW6#_^^~0>LfNb}97I;V|m%>EjY#&o~b>0yAMFAolb~KO94qdrR=yD5N*(
z5!|O^2TLs9A#-}!z;Z2lk?6xXr6Wid34*{6Vn)ZBf-O6(LrOgPQv)s-uKqW9FX9)2
zKcvj7V?CH{%;{e=si+vq1<OrvGxDlf_B;xaaWfz$oNUZcF5>yyRHp}`d5=YSpfB+S
zkB~t8%ffyfH|TonoAXu$_Ke>$cl_(Mqu?)e+i+qQpr6=77KlUUX15rBbB-K_giHDl
zJYs|rqeaa{X3K$mIS}1RYZA99@06*n2EJ+U8k{lZh^}W}xLP)oR2cLDHDKH~K`O^h
zK=LVF(m#DgQ+=3tBr%^n#%N>{BV1&ew*43wmg@lZD0o^UUh`Vbg5fIso(Bya)sK7U
zH4E_5W%ItdoZe$NYTi!txc7RbL~vF(kieUG>_f7bVHl9EOu0gDS&Bc_m`ws9Zn@y%
zsuft#w!3vw5`h)<dy$|ZaHYUL>}pY{5@FOOq`*4J4*+@rIR%J_#H;1G9eVPK!84*p
z0I_j7#V`QzOA^6^I!d+*HaNci2(fxG_Z>jIx4aVzrip-oz54zLD!MHG=~v6~PoN%l
zw*QaZ?teS|+Qw7VNyH7<$&*y&(?zf_n1?s5#wSWC%BuXh^s;tgDC3I^xvdq<d)-f`
zyffah&Nkf^xHCI7r?R?DSni0BEUG#W8_64r3&@nIY+IlBUNewokDv=nefeTrzVG%X
z9j1i7`@MQmACvhryK9MU_^YS+=;)_(KU;X-Z+yPoU$0)ZDBlhWiIgoiY)Z>VaMBcp
z5n6f&31fa+#L!Q;BZ_PVfRjWud@hf-LVUDyOzxG_)Gn)9)H(>uVl=ZelhM!`m@5#q
z+@R5r@E9h^rw*j@j$vNo9>&Wr-L*uI7i?*pmM^|}L{ue-l;-e$GOH}aRUSdhly)~~
z^IsNs+ou)}j&>e$tZxnF^pZIZ_aqMRE*NFWiwHwuf{93$Pz^7RYvlu$epopgfvs(8
zuLrkna>o8L;u$f5uQ;06K3Jz2O~{KG@6_5nY^SN3=YFns)?764Xe^(XF;J^=9Xw;G
zUpDu(EoOf)GI)7j@m~jk`s+``AlQH5qEvVVwgS^8aj#=T?rPuUcpjZ}!9ce(wa|RK
zWX-<jz=pXOf;4tGv+|8)NmoPw>;^_WkT^JrB1$TO5IlTyCsVUinbk<%N+!f1l0hb_
zIMj%!0PbVOw`NFShtcyvM>l^pk=)SLIL-@f@TpIL>jc(Ux+D<dWkdj@32e~5`_3I1
zOqed6=H73D2h1S^C@y*=rWXErRc350Hpkwa^0zH2^2|f&$OU*O>U<>z3v6O%DEzYM
zM<<gPLL+$?W`^D4^Y}Zsr6~Ae+eS)UA41bk64S4oHvPOn^Y7_WFJ7F5zODSq%fjT;
z%M3*M)VfRI`>)Pzvc+&<y}3Mw+Ad>YLZ2{lUHAQ;YvTFT=g~H>7@M(&*nwI8ZZ}^`
z@k(*kKsTtSQN-^S@N@7_Vw_;hk3Wn^l!KlzeV{w!Z>>fn(*X)QuN}8>fwEwf#t1fV
z-+0>Ib^wxjae_&H4S5;K>E$m}lXD)|5@e!_GHpA~+sSCOw2~mOG`00?XYjIO^4jP9
z?w$k>7z+qYrm{gK2W7vUp28qrey-KqmE9P7WhF|kY&cU5g%Mo{{>a=R5~hxm-}f5p
z8)_yv!V<&uAc`;-Cz>H|SR(1yoMduww)BHskJKR%&TeAkD-gj4eE)SnkUj1Q;{w+^
z>y1T5bPm~)ivTo#66RA=dh=tRm5&u*t#ct60D$)^(<fPg%0frX$2<qn$?DdgA};uA
z?Jx@8c4{ATMjEEhk1aC>!8}qOX|Cugew<AC3ZiKB!g4!A0)uY}^85!gbq_v%=2-w8
zMpS|2tTd?i%iy3p1%wgE2nFM%0o761@yV3)-Fm4A32t3}M_oSRg{V`7@BJaw<F{28
zvp@36%N)+6;7DDvJH8oz3y`-Hp%A<x`5xOW6_$t3Oek`&iG55TqEjdjUv7h+1WFCr
ziNYbk02wMWL823w2OB7gzB<*C+Aw3Ne%moZqAV8(sWvRGVS!pDg7K=+LqrFO@RQ0Q
z7{x*FVSj|ECrVhY;h^A@@QI)^Tr?erLJ<k}_!_)o$r?wtgRnsu$b>sW@-B!AyfMj~
z6Gz>QAIQyMyWd!m1Vq#dz$dab4vX;|32{X56G`nZL`=XGh(C%#cstk5XOZlhKGw31
zdSl=<yZgtx$eyk0+?WZ1+SS84`n-4Bic5!=xoTRT3l)I^2tVT1ul?RxmM_DyFR&-4
zVg@NZr8)s##GWNp{i##k?fs52g&{WQLlI&|3RS0bYnkqwW#!0#a=6o_M+#?tW|(zN
zJ8Cd%>n_J=Nwj0N^Ou9tN=P$v;9ff=nQT<*-@mk2e>ssfQWqez+pejP&P1@Bcg9a$
zxR7RY&eq=ugwe@UfCzIM8sr3wTQ%$GVKau(wj-_qVOz41puxOmaoYMckj=b8Em=6W
zkqu+ghK9M7kg3H2VpkRA>%JJ_q9c3v8zHzn{E#y4<oeH<)@AuKMbd(V_#aOT`VDNY
z^}_`BJM2u5kN^zlXS{|lP3}f-N`viD?90CA`fx=!UFm7nOV%^Vltt;34Z2uj&DfL;
zhAxk1k{%>3E{D7FSh!bBQQK4eqQ$)ETJ$RSDh~5ev!Qp}<UXd^_VTlmdO;iPcrTan
z<hyDGKftGLN(xj4Vg<}yL4>37#*M#G!)ROzL=q6;YVTKC%x3dv&h0jZWA)eahGC*k
zC!3qNEpNn(*peI<vSB?YqSoa?v-rY0Nq1(9Y%OMuOrtl1M$mYV?8uunTqoCdFhQD3
zu4<3A&eJDb*@SGHE$ofdR3RuKm@WfHJT^Tq%M+NY$%xU*+PeYG&B77?scv%q(<T7x
zf1U^~rfB_)g`)Uv-%~$O7_p9XQ^H`(&?TEvr1hvQhc-YMYt;si){&21@>~pG%NH0+
z)@Ha};=PF|61m&pPv-`828VW**Dvp`j{lnK0051bqW8K<dHC0@@ouo=^;9uZEMI7M
zX9o#~xr`EuMWsQk<y^I`?q&#j!fXoEg;Io=5n-*b=lvH~$Jb{cpO5GLts#?-_x<kX
z>tL(5lg}q#hIaQCg6?*Qr)@!oK{0w5Fepn{!bTW2bqEFFRti-Cuti}P7UQ9Bcsxs5
zQSD}}L|y8+qDb(3217e4vzEaC#r=>7F&09*0-ktjrWB;Lj0uRvkR>s@0$%d@yF+vM
z_35Y8m!0{u9|4_l=K(U>ij{8E5R#FvS-P*)yhrU?m)iqPwCcS3!;5=jOtgNDUx1vG
zK35pxRid^SL8AI1mp>_l(hRicfDw>nO(wN}59%(|Qt<eg^+tE}9pWsXy5;b$#BY4*
zCSuF<RhLGYe5Thuyy=9ziG<T8>7`WfyP8SU_u04JhCYrS81$N%PRZqZbN=c(VAMH!
zQ^cT!6iLg>!eLTTKn2y^Azrd21vb}PJuz(bZwgSo76fMvOJsrU(7EP#5mJ3d0Y$E$
zc#t<AltczWsBySoAOY<)xY)jD$&cxDCJXXt3yBK9Fi#c4@D#+j5-$rcE^2b4E-23t
z2c7&<40=i~clzPF%HA_6h^O<TPHYFFfJ*^!`>gz!0|-^(zoA*i(T^pJHV?}kHW>`O
z2QU_Q2-d`+eurW>mF)=sM%))fXLLi10lGj4L~NY2>Rc<hrJ5R}*lrcKKGVk+-VA%8
zz1cB<?nt`WB<>0m(6lUf>?gi5=KT=%JH9(OKya3iUo<$nvtm>ySp={vCap0DVt@kb
zga`eL1Uk0sh-zk!_m{3(WC?aNLu9R|Cu~hpFOSCp%|~X;gub@{Sn3Io6#U=MrKD_H
z!O9P)H0YQDJA_K2o<u=P9kbGN8KZTZNH-gg4LPRwf{FYJH}zxcccGX2k|?X{xgIC9
zDO8)6W45ue0_nXz>#QokuT^}|jZ>2SuR|_Z><5kFI4_K2)k$~li96$Z*YL?nefPut
z!6lqlyp5^%^$l7^7}Uqnu-{hu8kr;0QRY!WH+R9y59DfQTbZz^(mKbZ+qLW8E_YXB
zPiOO5PFEQ@Y;RrQWzLzw!9h6!5Nli-^3IqW^q0KY9ibj?QIKFG{8$>{kO`ZgOrw!?
zVoelz!3kI~1T>>#(NpV#cW#i$c(XY6QdSpb2;dol##Y$hDnqLqEsQ%TV(BbnIG0C_
zkd+q6=haO{aB&uZ672&A2g1R{s($)Gsb`kAjdd@4n+u}9<-H64KzXoXl*C8|o!)>7
z*PKDMr+bGGxpz-8aj=9A2=N`)QT~M$_m<io(8!udazR+mDYgsgS)$rikZzUK-i&i~
zTaCm;zCl>d8{e_+$`YF=kr~TGWr(fM!(>4{8(m?BJyum0*Is51x&mTPhb9q-oQfy{
zvuKeZDB2~4RW&C`EZG#fL;+iEZX^eLzV{d_*b+^(RP^3Wm4V0*2q|d2PuJkbWfubt
zgV<8J9%S_N!vK79K}!UFqL|EX68BQ`lt8^izF=H-BQCEu<T6G$#WrrRmhq<py{)AB
z3~Ixns<=9h+%-oS^>R=b=~V8E7I;MQecKO|Xu(pnj|fH4hTcJ1)nMd5A#^OCEhgqE
z`JF>B3Oh>)cLGHT#3$?FhUR&WlEJ80xdQ=C#+marJrXB6q&8Wg#IQ8X@D+5<Pm5bg
zvNO+d1L-nxDA?me?iQL3aBJos+P7-$ZpMXTON-xre3IMZ*++H1ELHmLKvbeCA?b4#
z?ziU$bQe~~_%}}l8J;T<F&IK+dZ5yQ0)C@dn{!9OZ9C0b3ikw<JPR_zgUn94&ze!c
zbECV0Z-(S{cg|};4M`i+EL@6X{wm#2jedt0A0%3~8%J6u=dVo_${}E5?ipBgs4h+i
zgLaVFH;|m71oZ!`AB>!w{~fmTe}#Xu{5SYF^MA~*Thv^)!C{B--mac8de1|pM<VTk
zh=ZO(GxyIK71rv9Z>5RI@o&BXmH6%oy;~c-Sj)OKW|sisI5LkBiQvhjh%Q~kQOsF#
z*h$ohBdDNLpD`3n3(4>hhSlIM2-b$ap$zUjqQ#Ob-6G0OqalztkM0EJl|b2V(LazO
zMNsk$pUN04go5lB*b^Y4Lj)b0>!<{!T1AZmDGNuNW03f}1+rhp8MVPBOZS=)K}{@B
z0x!gn`c42hdti(CO(~|}&$$L}Ad!lEBmh<l-*;lUDccfK-g>DWSilEu6Swd#6|k@Y
zFeNC_QQack?g?rQ^j*eY`m;S3*QNwgHRa8bLflFns1V<-gaWK}P6bi*AW3?dQ%V9;
zjPOIQHn2OBRcQFbS|Tu#*sin#ddP=A7}e+o(Fiix2#!yTnODl7RN1m5E%^J4KDWHg
z)z3ilw;Gt$J``$eP*lpRRvhS-Bc#=B3UyIRyQdPE?E(&1xg*&RfrshqbseU(73yS4
z)f+8X*k&{bXbA1Owoi`K`N=;cxaAkBqJ4UV-=EQTA~DgnVHldpO`{{0OSl@C(nmI^
z!2w(ojB4FZ>)eKmF$TLp8K~vdPzWP(i6dTWP+|wzzl5l+wSV<DQbDuRP_otk>f7P5
z^jNFv*Aiu@MtB0vv;4XWhe9pd0t8_?XDr34c%lqoYeZ&Zh5B32d?=6ERF)@MLK72%
zHhz3QA8F@=k~A#{MiYqQ6Dr*ma8vK@AS-Zx1o<0k+~EEIn%StQoSb6bYaTk+*6AJu
zuJK$K=*Kkk^@dxbz;7XZ!H_VSKscdRhR>@>qnA;+-KDJ)v|r<Od*QC$rRm)PK@Qz<
zH`?)by^X5d@#Wd5&Nn!qT8w_2Wz&{7c=x>c`gV75dG<jC`q;FsdcD={Q}=}tv&lxT
z+vETwT0K33EwJw7<K@vdWIn9MIeodRxDFV3I6lsf{`tE6T>MRwB)4RCt!n0d<5c(I
z{yKhl%J>N0d`xR96iiGJTv%S~kEQMC6e~qp%L;U54?i<r^I`F6iJLUt2Y;MwJ2m_I
z`NpCc!#Wmwl^M-B%}_*sa@E8C^#M)%t_WG2jL(;`0V`yQ7&!ut>{l?G{7exCy5wJ#
zFF-}hR5uFn5}q#12bbajj~&|aiPjokDb?J%+T4)3ae;k`*BYLP#4G8LU@qZJMEMqz
z?a;s#o8~=_pMQil(man;ODqFm8ZJ7<5mzBb&m!gzJE7bfnfh!zd7Y6u&_2<Tb|!7w
z)X6mOfKtsG7txZFXGs9j(HphBWU8284#Wzcvlnc{Lep-abBY_a4pxK)iT&1}u&cIF
za(rL5*+-o|K)o<p*H29qx)NQA<m0;*4T)JZA?vPOd{YLe9)}4DC5R2G2v}`18NDu0
zn9}UHpQy~HKYq~A6sydZZerC=v(J-lQW3hMrtoO&W>{Yt;Lv&}9bum;Pzwlw@(bxx
zn+a;5l;&;rsTcF~i(`8-$s#I!60WuI49N$Bo6j{_n$kjqtm@jDfGCFem;>Q>`sDab
zA;-Ad;sqMH_u`Nj<{j60TyW24(U<zl+DA1mP<h>wBeXER0bqGoosFGdoBKxB5>CKp
zaNe-9QE?S)Z(w9_$UkqyiA#`GD=pcg>)AoemNN1IdW?uyLW5+|@E*EcMx%g2WqqSe
zqF7m!tHPk5h@ubKZ#C~P58Q!JZTwF!<xr7fmI2az^|?wAdX=9v*1Vwj3T;_h0iJ70
zHCwG<do>NQ+Hjb<F`+b8%2N^FvXoLP4UmMBE{icMhpTD~6bCClbjMpbx#~pYo*W}e
zHdot#%Z0W@jw_l^D9O+y-!BbmLGhkUoA|bHCe4PH*)s$<a;>XH!viRe_Q^a+5tqXF
z*Y=yt3O>Ddlx{HUNNHlupbNv!%IDbmeo(;Zx8rqJ6uF%hK-)eHGwJ0-t3nF?q{lCo
z@%}tcy8xNnL>Zn-Yrs#0WLh}@Lto}3`<&&iGz**&-JO+8(5Yj__T!~n!1P{y|LgAt
zwQ-w`@+rMRl_$?YnOS~Iz2t>C7OBs^__ia+g$j2=$xhgg`hqPWhd=7$95Hb%E?3!I
zjgeQM7ZjIC*u_Kltt%%8+QxExU*M$JbPhun+Q+K2)?=0G>u)~8lugrgC|}^CrV4}q
zv!F3CGBEtlJvn9;rvC~w|KG&#OYMKzDv<tVtN7EGmf*t=bP%VXDF7J18Qlm`V-J@o
zvKmDz#aPPm_57|f4P(|;d`uYrqute(mMSZ~s(L$HWi=fv4?x|{^m6pM8*I*a?MvQ*
zP|DClpF`11ji=Pf&_nH}38FA@Z(9hV_2KjU+P}}1O%|#jY&PTX<f!iYx?hMKT&CCc
z`rN!fLq9g-U)cWgdD{CZuDAX2_IZlMm-FTQ{))8ydr?*BG}_RsT2=@{o`<Y@U8N3n
zE;VFTtK6ils$$HaZP~d}(fqSV`>V>TSA!Dw*q@S~mjA|Ul;!t`lbMAI)J=If9}7y=
z_(;-Ld&og^NI|==4p%jHgvya0)&v~Lv;&Pfe(2;?JxzleGED=Ay!L@c*r^T<MHLf`
zvPH&u?Tj#u%Oz3MLP3l`FF_^gFtHIdqf!G&;IGv1!XTy1Lb^M%J8_M2zN9-UMWcXU
zDqI9r!eHhODdG2#)lBVL4p)Hm{`3HO<#?qR`>(H~*W@_qhM>bxD@08d6_kYygNKu&
z^xbuPP`)2sJ!ftY-UOYLI4Pc9o?0CouWHusvj*YGAmL0+#@Z9y4KeW6{Y+viXvQXl
z>J~ba=lq<yT?Y*Dg%q6#ynt(SoO<+>i-O3$znvGE0sw+0PNb5z*2Z=dGr+a&EaD5>
zxQYH~?cZ1gSb;|?3CPZ&<B+Y45Ec`?C|SJXpbDZ%$Cg2KNE|sTCpP<w2oJ3Z+2PE`
zlu4|qi=w-uLee9nir$5yhlcbFjT%^J=vmgN$o4;Z^cmg}(Cj7baB5Z_f<TOUi2f0V
z(`W_f&tIVM2!cdn9V1r^oB8QD!sLxsm33_-bm=jTV$m?QI}ia;;C$XBnSQH}Dq7wm
znk)qqZjd-;*qu0^QAzJnNeA^O*Gy-aCnJF`_Qvn}DM1cvkJ^R>VsXK=hx}9uP6Pk7
zscX14OH2#(>fHV{F){}N6p9e#d{R%vq`v5LBZ`>o9CWS(Q>Vt#j>x8|kfcEJlQ>v1
zv7^)QS%1TH6T!Z$Os&poOQ5%lq6i+(BU0B9RG1kOOukRSB_edBnzBDVi7Ld1fc~PA
zY7`_rLs4QNG^Y(xLIIa&XdMBARFJq!U0=@=X?-BXBwQJUI!L;^JH&QUq#m+)^~U{B
ztxe55_DF$JZR3I)K&cipF@GTO^D0hg=M-zGFa-rrTsZ0*a*G5@nu@k=-=^CG>CbTB
z2I224rv8MFi7gV@!WJEDDV8l2TFWp5U>PDhz$dVKa2h|X8)1Gf$}5hEZUZ1K_vHTe
z3t5jrLbBs*0l72-#A3y#5Suv!oukA6r-;7kL_o>^g%8VZuM%%4Bm`U%+&T*`eZ!NL
zDNA7gw>gSUZSx2N^F@AZUUw|u`9lk*p&?~@&)>ec$iCNUU#P1{P&Ic~GszA|JNz^+
zgb-2ug-=W&FjT5=Qcra9Ph^?}>Li`zQOSvcXmq;T!ib#eg1`&f8kd+#|88`}=SMl#
zX)a}=t$2xLf%uDU&b_@cT07p0o@Eu4UczS_>~RLhZOq630Q<0Vy!5AvuUFEec*dby
zX-pi1G&yptqP&N3aoZVWKrektNhOwKrr3dq2p#I^PDs;-$LCdI`ydze`<ozeGn|1a
z{FX^DRyeT$_7wG$4fn-*CFfX(N=M(KhiALqwQRCt_~6%L+grf)727-f1F)X<wk`d0
z`uZjFhxvqIV<#4Oh&)hgU5whRftQgG@XW-}v~;DB0o9u6;iGx+uBWU~mN6($dl3!|
zmq6vcX5PV5(|NgO1Ui1%_VzW~x|-2(Vo5K^K<S^x95m<*M~!Gm#saXx<W*!LMsV(T
zs&=xNg|%~~&)hGHBebhc)%K02S_-HGQW|+m%)$JEHu1@U@U<{|d&aSBE~nNEJp#A=
z)^HyPV7eU?ki|5Q1`iPK{^2#ks5xSFMpooX9Gg%(J**Z;%m^;_HTc_HAiPp_Y~8pR
z#m)A#sD3jkQ>3q4bx!TYv0*2Xxf{Ax5a!B7J5E1Kz&r2<&>s~Avd*K7a<?r~z12Ed
z8RQ%Qor~o6kua$g-AF5Wn>suNz8m^)HH~?usi!wg@71_=EtioeZv@C9VLUG2mKrLf
zp)|_?#EBd&6kR%N4UzaAWA`j95XES~()f_z-WGcXOdGsKB1x}52Y{(OGVN`(A#+t9
z0ktPOq#nqJEQs`F{W3{vb9wnt?c+35l0@zM)@ux#lM_Tsu*5F9G#Ee?<5|kFQncXe
zlu}ZBMYm@P?I*QtxEac&JEV}+8LNyvVh+E5M~j?(5Cj=4Q^xFZ4ccoLkF`uWVSrl5
z(e*al2f?KN4LlNA(~dG0kxZ2SFy({RhO%o+A~Tb=_!!9i>*MssuEKm3YE@Th&Npg2
zIU^mJnNcx2YAgf4CiZFt%-iK&a!1L>oW0loiqd#kbe&uQx0M#*2gyutZN4PZ*=o*P
z6JP=*oS<$w))+b0I?{q9*kh{9aa7#u7alF%H1)F6@rOq5z>^7Xm@Q$-(`Cq>2E0lh
zOxSlmEmM|*g}%{Auu)k^l)GNiIOzBgP0XU$hF=*&<LqSLqbE}$zbNEk=UN-&F+-pN
z_~?Z(hpqyA2tW1_M6l{Cw%P8R_I$X4mjoB)5S!hlw|*-ODCj}XRlKaql<PCK@CJuR
zUlM4}3TbP$DQW=tV}~#FFAY9Rc<VH4BBDqk?H-IQTmgjC%Scey7_0j!lDgu<lOcvW
z-rf=(n7ARjQBU4G*dn;<aZ&GIH1_~`i0q=!@}^2@bom9;<NY3@p@@->Tf4)o80s`g
za|t@N^)h7p+A2O**Yk1MMz<xkMS3>Y-mg#V$HULF{;P2-x_puW@#GL|1E)--avIR7
zS`i%U5T`9T5))~K*l&R+W%DvS;N`%gae;c_x9|=B$c3d$_UPgq5c+6@2Ns=!XVYBi
zUIY3Fx(td=kpaK-P-v^F0Nt^x4dv%GL<1SrUw3n?>+G$+*(By(3*ON%|3Ge*HwcAc
zALjgk!ixwW$+8z$T{h0?(oNT0sPTElCD$HqqejxL*aYVhTfc&FF7-A{BF*VGDs;on
zpM^p`8=|8LD+Nof0sw*JS-hF{G!UaNPDqddG^y16^;h!ii8)~$aE8lqhN96l*d`*x
z&ep7(Ud3SC<3l?w8NmPiDhI#7gm#$3-w6#J7gplpX;fx-79%i{ZDell!nP{|+lv@p
z&)!j%u`BafD;*mvKKnMHz_~58(_}EeD>aSzb;#^Zr3e-l;dLggV@Y%BcIRx&qxv*#
zWW-%yJR0Y|T7;TN{VS6qIA03r#{-k#jAaUQ_^zLg-+|IeZKxsOaK0DM%6k&0Vh<fS
z1+(q?_7Pw8_&8v!xnmN?^8y~rOIB)Wc6D0{3Bh3WDPmJ0#Mu%hA%679<>W8|hp>$3
z*qD`Owm|Dc2Nsl@c{;r!&1Uu(H?a*9)&s@3%gbsw7qwE&f8=*!Mlb^*`<dh$kLZSF
z&{Y2cL~Y^O&nD$+C(&RrH}le3>{(+W>gKmc(Kxg6?q&(Nvr@pVsQTD!jCTkzXH3}_
zr~8<<e?ED5UdGYZIaGYd;$KrvT{c<u*0CKN59P~~s9gO!etbQ>{rxi{1la|3bn(rm
zv$HI_qw;jS`Mg5zb#J=8>}>mz4Oa9qa{HS7QLnE^!&{VAh`J-*sQ&KC^zrHPP3N)y
z>FX%(6%S6)_R^(&>#Cl8G2;p?MW!d2O=X=O>Dd|f^~~qdrg5|ST&JQe_Kjy;7gqCo
z*8y2J4X%u!i6Jew#1v{OfWp{OBpk_A%{QI3OELO%J?wY~r7X=urmkL&j*9w%$%m~C
zdAD?*!Hk@FEAQe;q;2kZ@&yRdoKioPaphJkI0U)(s)L;z8Yoa?p(6$l;UQpg=}~`W
z)y^z)g}8|8r}P2bPdmC=B(iJg_}mnr{h1C`iqxc5t#?_-$cZ7sunhLQ+`G{xQnES@
zf60+9xp|$R>ZDV1na#U%?eg5V`I@U~rEy;cLrwFFoxL&hqt?T7n;|FXUJ+u|y^K^u
zK-+p$T_AU4+gHOK$dVu_{#Nciz{S;=)-#es*!exRE1UO~Ok6uj#K9i$XOFC(DuII-
z_etciPi0PBT9r_d40fi#t`NOlG#A&&G?vN95}(y+?Uy0#bN!ZY_}wmPawPpt^@y(H
zY!FMr0xP29orHu~^K(}qt)6(Qun4!;^r2?O4Da$8$Fp``WJ;*al5tghvO@mJH!u`k
zu#m@V$c1zpCz;@NXjEZo-^?!;WB8=o-6qsdnJ}tfIYZ8ui|G@8PaleYJHHC<KjRm!
z-jue%t|r@CYpY?N=c!)bs7tO_x?80hwUMljGUMxN`{>jAoNacW!w}S<e()SHTX}tB
zfC&IVADPB(@#u;;X|Qo+3`<SovCA+L<xB6x4-s^uI#whm%SN);iC&3nD&$>I!|2ep
z^wK31d>aA~Vi`Z^s@IV@Ps7t%W75rwZ-o%)PQ8z?<GgQ19>5y5sg1v<fE;GDb3pM%
zehp_GA=kaqYpS5*{Nsu3A0jL!PS*c;^yBh>QT9$jwuIZdZP~VM+qP}noMqd#ZQHYK
zyJp$8jazF+oQM-~SH#^nA4fjrc=&tI>|bw9d;AafIN~qN>K!69Sm|n#!?|9pkajef
zL>gn(&<u_MP^w<QpG<)E+q3t)rBbmw{4l(zUd{$!x#jTw-qb9?%$_;(>*{p;@^n4|
zC4*SbDQ9|yxqz9JwLr<VN%Z+5*xL8=;}Snk4#TCNVKAZv&FL`c_ZNNr>D+F<+&`k^
z*+~Z#3(oiFM_DILCqJHQ-d~H8`nlb?d>(HPQDw8aX!`#VWjw#XCFPmV>i4)GsB;z`
z^=M@o1uKQ(TZ}Ljgg5diPa%=%vUT;n<L}Vc#b)HtOxBFhr_o!9vdE!!yX$ouFo{Lf
zN58SN7*45}f5cE}#VGYRCrZ#L!@VbGiz5;ST52$sKbW9l2gww{(xqly`F1?q8zp^p
zncmfJ%(6#$_#TShO`CbSFzTbvx|_GH`_b`b>)X#fS6v+!{EU&i*S-37QJzqNxM_Z_
z(XxMZ;9s|CNzZJ6s1EJh_f&j!z4grPuDL!GAbf2YB{~1$+ezWiwXcf|d-{u`<V};B
zd9smtt&V>53kbNlutK7L+~Gj4VbAKHluvWY{?pY@ac24<++&=%r!Evj8T>ujp!92O
z3ZgB2B*tKrfv3*I)`Wf6iK%ZW^r$)YE3;8R7_qj`Aio4CHBx2WHMD~2jq$AeI7FEk
z51>S;6VEO!+a2sG7pF;7N*Q9p%lJyxX7|X>j8(SL>f4hEVNc_P<H^GPIw@!3Cj)UD
zFj;^KyvotV`FiJ`Mtg;YA|F`0H0HYpLdHHpZJ61{`Dy}aLqmg*EI-a*3Dl@ue`gT!
z2r7R9jAyPL>JM&*`9v8lDb0ou)5Kc{wg!?VbPWxVQc>FDv6LViZMEb{T}VEqL7ZaC
z^xrxlP;MWw{9^`BMFr6ESR9R5p-~<uD#Vd!VQsLOnIXXe9!h}7T-x8R-Hu^W6V@hg
zNMK9O31B*?gPmd=bg+Z~u8+mWAsNxY)!M$=2PnRhvAK(*-q0>Yy%!INx=mPJQ!*^-
z0)iW!=V-JGhd|?&lHgF0SpZ`#85eNkFH8U*B-)=0N)H}oRu6TYBlM<XrW{3^toGA`
zSa7$KYEWuJ+dm5VF&E5kF{%UwK6fbPm?;V^+BnEknOi#Awv_X#*3K)A!oz(V_##QK
zP#>I(KIBhX5tF^k#F#dIFvxzX*t$~`%q}V6vs#2APJ_yeM+$O@?2hHJq<?=pORHE<
z4Xd2SOQjgB=z|fxu}j|%<b~6!K*I;U;rh>lHO$h+#<gpij8Y=sz7I&O;@M~Y7^aT+
z_ULY@oQz@!IxvlbO+c{>Xb45%AR~pDH1slcGPELx=HG?s_*44giu_~!+uQKx;3w6I
z<Hr7EV(`NF_){&3cec`rKCvR!TrqI0=!w&caV980>g(OX)9@44)RS;W&9GHluU3xU
ziB79Q5h<cvlEb!wZUjhZ%Hzk7vPOZdVE5OcW_|ozXG=W3B7L&oJrIX>HabsvcEyzH
zecZ#7oi-0Ge+`e*bJ8{c%8rLoY%lk$i<_+)91DQ(ylSOG$0m(hR@<;gl~bCeuuK;=
zmn_wGwsh0dwfv+H5{$Ej276IM!oybdh!J8*s6{nSZ<MG}WZ-^tnS!qJWv1u=fMCwO
z;_~memJkMM=ED(bN9w^6xkYmzWXy(CEH|1nlQt~ayX-d110TKy948U2Z77wRh~z3(
zN3HeyzritYXV1c>0{|hVBKd$Y1IaMsUZXT_z=l3*=e+IT3Op!!L^4TLY8jNi!9Qs5
z8}~Ox9<EblRl0kU-!9QxsBznSX?Zg%|4OLYk~YS@8#GcdLgY2Zmf<qIT!V#$4kSuJ
z3jZ*#lx?D?No=!!=-&PA?m9B|YJIt#F)kaaXY~+40H-RSlH~ryM+Q6-fkf?9mY7j{
zHM%&&(dN+IKUF;jC%w`puA`@aaM%m8tMef}aJm~j2iOa47O_^X?%0@pnH04FJlS~i
z{BV2`o@kF}_jKhGT`-q6;XX595M;cFs|9W+1i+Qsw~z2uq}(%;^gbMv5cRk!tmBfh
z5+b3dEtDQ$uStxiaQp;ZkkkaeM<0QMbj{Ld?gvZ62{Ks9Ic)B0%sEgg2_X3yBsLGT
zvK$Yp2r{WnkyM}*GeL@MRs9W&YjtK#uqux49#CU;rZ6;1b*rwI0KeFy!m?yzN5~nT
z@nXUy6T90_3t$^XkAZ|3U|z@xdIQ;ZT7x-?QR7Cz9+0R43!LZkE|V{%8flkFmnn2i
z1{(UkpPzl@n=RyEVT-4d>;`l^dao;X0~;5?hylnlq+4jla9TKt^uuL%9!$Q(4d{!n
z3y)WU6)qX8b*V|IIP<oaJuEJFUZ(H<W0Ll1TW<?h|2vGXNxeOQk(JzWZ3M}%pXFab
zvcv+qJ1AB<;Gb0gV{R%MJTT-v@WK6H7(y<j9Hcgo2C+v1VBv_6KM-g-c1?jxrO9T<
zQR&6RQQk0afoTId--w1Le<Z(Wm+11bk_p%*h5j$Sz(YC6`ZQpC8w_%_SpmF(yqs3u
zi84NxO7T<3qd8=DFnycShna$9B~{Bv-aYY8aHL4!{1Dh;Kf{1*)ktB|M3Tol=Eh%(
zyG*lYv2ZqhqH?~LcK)CYJP|%XX#n88)>vPdlPRZh6(<CiQv_8wtM_GyoFQDS%$)-k
z;1}KSKO#sbf>|(oOUTv~Ihpec5eJl(T{ENZ&Jjd0S$mM;HvC~HG6By}nyI<g%zb}&
z2rvNTM}4p(`t_XcQ^c%|5`xK*&)B>UU;6OY*oE-zmpS8>;kLNYOW6l&VpXhhE5MH6
z774=;x;3M2Cdan7J6lBrw5QN)qqm<UWgb_wb${cYo)Q|>3Oi;=$P@GHix|RBV|3<o
zn2}U8OnJuTC&Wau<oUtW->6AGtl5|F&9E6K)S)p9@(TDRI&~ggJixZZ50|x&#ZyhP
zvxXsy2Aja`)I~WOV5Dr##xgs#(qU)Zrh9-Iv9;Z9>Jyn_4Cjhg()Dt`2uQAa41Yl1
zjK7qNR1&CerisZ+KobNskF0mg>14j003@E!0$;M91xAUuG-RA1m-;ab-nI8yit@eL
zjXGKhTRWxne5aTKc4DIC*VjrCF^GrmR3VM1iiog^d$C7{Nbn+6jX5k;f9^|jkheLi
zm7lkUy}RM)FwtsoRM|*F6F)wcRasvbu7eDFTHL~>!S8Y)NW<evlOyjyo%wqFXKrf!
zw~Gn%3Cp)lLA%?bDL#r!-;Wc|yFpWccr79Salrla2>VfEbsmmDCTQ)6Bmt8;y^oSU
zB~B#R`B1B2^Q4Ac&@$<-zlj;ePMs@BaM_I)+^tuLoRB`PmRk3cK262|pO(B7i*m$D
z6~;Q=BMrAFobLXC-Zw-$-N4@+a_^#2d^8Vui3iCHT34Kez`+~ns_kx8gJFrLz5gDB
zsX+w}q~<Y%3Ae@g19QXSYC`hOZ&=!2{{RyH?plJmoY(p`RzcP&6>)+quCTOqNn}DT
zxyZWB^BP#|tdS-vVM-Tuy{VLdL`llPpOkc(OG}6XgaVR~a^zCnBL#evE^kwFbww^6
zN{sXQPuTlS|6nM8CNm)F{@N-&z?T4hIw2(EIg>B&Uz>B3J8})iQD-`&Mp6Fg#5sQ%
zUH6BXhTCJzjKN(hS9?!R<XJ4n1oyP!aOkgY_wHW=<X4N4$G_+4>yf`H*sb<1>d#lw
zu?I9dI2y0X%KsX~W6-Arr@~DUgQH1Y4&Nw4yG5amg0G&?{C<Q23}D=$%Vp%g?Qdv(
z$tc6NL<y3_EiV3~NkI0z&v&|h6KJLksAX+5>#o%G6n)Un^JWCyDgO8t*d1XB83q5;
zg?Bl?1_&4#cmdZEMo&*Rw7J3ZCy5y`O|Y&OSidfHAK)_saIYU|TC9nF*i?$>0fU(!
z+KDa+D!a7X{c?LrZu!q&_Xhr3chC5L+r|Hn0FRB0{eKH&y3&xe-(*AiWrClsFe1!v
zrU}DC8iCWrH`kYq70?33h#Oa7T1R89g$DdW!Q+ZWXoWqJxdfd~hsikO?7y~!_71uF
z$GeKyABvKg17-h*$n>E4e~3&yGQL@gkr-TZAwx}ZCuR6@^s&lRx0jcn`+u&}dU3G1
zynH;~-nT0F)AM@k)a_^KPtZ^7eRJ}DKKEYZ-sH*6vt3Lb1M1`RBhg7SNVx`DTqUg_
z+xI1<P#45J8%MLey<{F~UT`GyNX1kVLJtZF=OGy!aMU6M64QhG^^YoiN7Gq@fgnPN
zViAaqWD==F9BAL2giDbE5f_6sxI!p^eRZ(qpMH%O4E3lSMl6T|cs67;p}ese*Sq<6
zu^CX0Ko7jiUX=kgo6XT65fmf=rG%Qctu^GHT+G$jg|moc`3-O&Z^C*#krgteMk?R`
z<|Ez`2}Xtl8t_8I(!=X6N~iz4O`J%o(HVbofHHu)>59B!ikJ#`H9Tx?qhCD1KY3>T
z##`62;zhDKsg(;6S%>s9cMfy!tvc?Zt+TaNrsXfDjrH<q*<OyAHu&CfC89e4<1K+M
zXT;(wwB?&Cm>d9B@$27EWKpwZVTbePBmE&@0~^EP2`#up(13)Bu_UDhAQ`=t^!TDk
z13->yhSDjvV?NVKVlW^}X^ZTRC6Kvj015`!e&lBW;AnS}kwRI*&k2ug8>(7|_wfK!
z-*0a!t0(Lq8Y@MS!Kg4PGmHlk=0P+ZB%qdq8B#E2zn>pbSgg+klVK9PY3`YJMpl`!
zkhc;;o1V60pC_DZ8$Uc-#;T4TGU}>qieC|fYrGtUY)%d~p#p^*q5sZKQty~&hfo%Q
zt#&0{Y+kXFjI`))_IG63qOBEbpf4x!>l`1^+2BQ;lbOL+lqg%c$iS)G+PyaS^lH~u
zSG4?m-~T+rylBS87$+m0ZU+CO$UcVsT?Vgoq_a|9IGfbf2mnKB3yc0YDl?7OXp}>{
z;xfL9dZgN2EHMlN9=!l&_{hLRAue)i9~S5N6>rNgI|R*|kHiW&ieV^Pl&{zc7IUB9
z;029^T|xH<qJ;5n8IoLX?053&OduM5Z7^{K7Z6F2o+mJj_bU-v4sXL*fnbjHgSQUT
zp}<fQ<A-N5)Rlfe<@nXy!}UkYR0}Cj#`|(kaecs5Yyrg~f*YC0ARAif8eqV1$Pm*4
zLoD(JODbp*0qTw<JSt4EUZt?rM$p?nhw<X9>jT5k4lgaNDDP&awASqzN`W#3^28}Z
z9k1RUMoATbpA~@FoQwe;Oj1>KU7z>-ZRB-lZ@I90u)!VVB9?$}__l<-3ui<FOW@n3
z-N<<{<9E-&?{vJxX?vUJyc99d@&YIWBao5c-`yKn>VcDox+MyMQ7jhA%kxaRXK0Ix
z$?;984%6d!b*FiqI_cdQL=bE1ny|la+@kBMrL5X-)eC8^&t|JKPZH8<QPHm{1I88a
z*D!lmuZ4EE@i@y_!*T8Uq;p#Id&|aBK&Of2?3-?I0*o>Z1{p79Ce<cM&0^9$Rv)I5
z37$bkbi`un%{e6eWE3VP@+JTzo3d3+{^#q~M!DZq8Z3?n=M)91GzAg}f|+TcuV>e=
ziB?7RH=|HYU`XYI#_g2*w8H(&;JF)<$=k?}*we@ld(y2oUhz@%jW6~3&L)o3i{xK-
zs{wH6=-*RvueRW}f4v4$RZV#Uw-ArxH#8q!R<e`d+bL=0{D@0453^*wv;$ADlDM5+
zZ8l_%tTbRcv1maIIZ!ZhZu7=$N?2N;;mP5Yh>Q8kGy?Ay4G@e5&yw@TEJ~bQ`xX9u
z73q(q{{E*8KuDXhl>5N*#z)1(*n0VNj6?%In85P}O+Q?j1KQnk)ppICVnAdR^nVoY
zH=&fOEX2As2w$jn`^)haakd+>A4k1E4E)0&ioOQiGSum88W^(((}zpLfoU*tChS@w
z-t(NTv0{Dm77JotA>XVX;_cmn;Ei}%)?BjGr_CF59d*r?eZF+p-BLj}qtpBXGd@Hf
z>B&ZaCjCrWZ+@6e_zbsqZqSipNq)B1Mb)b;w+g)uj;4YBU`gE+=m~yw7dk->46N0w
zG8mIJLt(&WoZn+zp|E&=-sa7|9JR#^Oz>?|?0&7EFmICweq2dZ8jbu@{U^n@N*?a0
zAIU9~Yl#flk8|un0rgRMah%UJZFocK$;JN;u3EZf^WTaZ)Bl~C`#(jE{ePdjKGOW3
z)V1$R-CPmO5Y0{T^GfYms-Z>akS_juVLUM@s)2MaY1>#me(zuus$#0C+ju>Y7m@7U
z=GfqkPIOThggC#f-W2()o+BD5MS_ah!$$bHghK=a`aDn6_NdDJ+2|YntR7;vb1P+J
z^Ge^K-TU3iiT&}3`XVMOM=6S|xzPUW^wHDs=l+XR+vmkm*F)7>)MQ(oer~U`G(COo
zi8?y{_6$9pzCJc3^5xSxr6@vaeJruj6EPNIuE+yw_zc^yEprZI^?~gAp;_vX{(@`*
zU8L#!rq>Z<6ia>5RO7>6h-C?CTqclpdCeXhC2<c4_(GIIN%;(B1bGjgWZ9k!B-~l7
zIpdDHB-OS~7vqODYhCuUQpySLw!gLhF<HKSK=hH#p$XAz#xr2h#fCPrnukHD4xtRz
zH#bvT;#y{QjpqKEFfa_zBLK`=z*EK56Pl?zC@zXA-^vXWVJxL81moj!J&MheZNQ#P
zA;l7|AMtC5hRLVkQ5|X$4%ak2GuU2%#9<_?Mh+){YpXeK#fU@+Ql3P!BfQala-H;7
zeJ#!*lCi;LK8DN_#ogt<Q8w?yaG#1n^G!slJyC;|#kPu(U`4Pvh=+I(d_-V~!98OE
zV_dr4ik`!)^>oR<#{I#wD`#Z7M<djR0V6Js>ZQ|Y5{au^8ZWR8qzI<5mYo2;V;Au0
zW=|(#IhbNFuG70;3&9*eo_wZIaFrURQ6(>DofP(;$wXwKaz_61WMS&YZAwC2zVNQl
ztxhL(qI$<!@{?CwUARUzJOa&eOZonL&|$=ku{T?{Y1mx&3rFuxo~>;#ZoR!0oAtUm
zYeCth{(pk~c@60I<Xoi<^v8kWa78TLw;j;39h%JSjqK=^E8I1=IwFSQNxxVJ2f8W%
zZduItef|2<?aoKOG|F6Zw%>Bl)Gui!C)Wv)94hU^c~!A$RPvX<+HqAFi<A8KYhM%7
z<em%HQ#Cs~+<MP9IV_Fq?Qx1Y$s`Qp=AT<bQwV5H1nXrJ<7Wr})qTc}?g<WCcL}0<
z6XQ1+7@S=QkaX1H&TkRgqP@}E!~lbc!OAe}x%|U)VaxsJR@67Ec8Sm>Ue5eAo6h=H
zWd#a_-pOJ_zSR&%khAb%DHAnOs#dJ=Lpon(oC7uXw@yo`_BLgL+Y;C(Bf)Sc#izZS
zD)XS4PK_C(lu3}@f1S}FnuFo2AR55#M@yi@^+E_&Kz)rb?c&=bZ%K&Jf&wJ(T|xB%
zHH!>11Rgx$j6ijY=LZC6Mj8Uo6fi|H!JBZ#8sG!zNI_hmz%azTa>NKLf(*YemTm2%
zaR+Fn;V&U^dumZ`5kC0B``VwE=HEULgD|=5@KW@^%(ai_>X5)<7G=WZH~ht%{|rof
zg&=pAYjK7Z4+Vj!AVD*10X&G0fQ%s`la*0G^<^Bi3vnb;A&Dwk%_z9l6e6yBUYPdm
zB2di+&@S9E7~)@rIkF#+ocuI+Tvnui{18oUrSm*K09iOW4;7%sOAU6k#~_!uXb(Y1
z;@~+^^b(B5=i`M3<B{Y0uN+q3IcX&2AUKf#Lw_GCG&_tAk;Fg>?tISj#sBkwFbsO>
z#F^u;nVk@og25Z2YXwt+AInK15aft33%S!|3gsODS$N;4Tv4zb63d4~(qD^5w}{^0
zH72IV|GC%x5fL^XlytlyR%sL|Wi?k{z;eIRi0@l^t2>!8!sh{gW$-x9SuA6|KYnv^
zf2xdFJ`xdTjObaG^Sv5_1RQ1|o=};o>e~TS3`HEO{<0a$rhLP~K;Y7<XR#{=aE;v+
z8+bjbGsbH{?uP9-*hjj}QT}{<YsCM)+0Iztk5L<AzDJ`S@ty+IW#zkH(bnoWWh3_N
zNfJL9yc-;qz<VK~X{V;_w%?E<7=HB5AC7=J*9|DT@#x7@&-<*8MQ~@1$DhdQc^{wK
zPOE$T*kH{72KWk@6{bHCnOP<NMjlPlVDS44Y+J~93b+q4<?bLWyn>^okWKtcH?G>-
z=_tdmk*RmKraHpZ7c52Bl?b6Fbku1-c+&YY2caO&nu@~7nl{+r_4%S!1$1uw05dRw
zswMCxw*Z}epo%cY-wrmrMkL#*rY)K!ll!wg+)mO}JhVQJqWzuCj*~qAgbkd$+~bvh
zU7H&3vliZcTkNRRp%&c5rFpCswqyx>{al}*@3Ky>3GnsMOt=*~foLT2*s67}+91Az
zfST(MP72Rdr8b_{bMA*2h7^=VvbxH1^uJglwl`vI+17omyZEz0@pg7_-^rBzYaRTr
zJVMU@HIHyhb1QCx4aTpxcOVYVK~eO4aOD8C3ys}iYi@Eb%$9>A;x~WFA-M0{{^Q$x
zNYClIUpU<Qe1;N2<VRA1KaK;>2gc*>_2B9Fa1QH$jA6!dZg7VKflE<=nMS2i9mFpr
zLbhsC%L32Ov)S9_$DdWSzn2yE5(nM(X=}#t*US5_m)8v&8ho!xRURH5`VGF`PmlZS
z)7E9(@NC_-#C!`W>VXt8in&yk(rKvrDUbmU3atkK8u}1SF(jHclJ8XO8X2dJo>?&e
z-`{Xm*l=01j>w)>i?6Hp+Wb}ZB0uLSk8W4oc!QgfU%DX9b?P|NhRJZ$292xSxw4uH
zwwz8_*%u#gd-u_Ga$S<Ji%Ak5cjY+lfSyObMlb?Jt%l6D20b5nmh!n#i<hfk4y0B*
zbeo-Ne=5<(*R=&N#gQxT+to~`f)AD6^BUD!JN#}SJQJKhGa_*Eo_Kh=Q2C(z-5#)H
zAuvHb8j(V^5*Sgqq*RoWRV)T0;<$mFq)?FDdpr^lb&Nl*WrR$I*vXD&W?5Jn?3pju
z(c5)BZleK92(48=m0im$>&f0^&Bw+yN!C}{p-r2FoVnLzGeJL-q7y=>BujI?G<@Sh
zP>N1j77^hf0|M!cO!c~mHE=+3qNNJmJaif<rkyMPWJviv?Zsz=gUfm@n`DAdEy=&N
z1`d-+<%Fw0<agc^>QCgB!UBX%{+FSP6jm`zpA5>60290KSmp?=R;Nn>%~{gc9GrU}
z_e__)h*!!fh0z2YZ*hl-tf5<NLV2-ge4Hd`6QB_J6c3K_Wz#pPI0(>aifVs@+Gd=4
zezWGzDY#V}1fnKcdVvA!^%y>;&Uv*4eHx5KQv{*L2{@57b=QNwX-C7uDikJ&wjPxe
z8(y@y2_~!JdV&)7EwhR!W(2Hcu0|Ka&pI4%1tPE`;y_6{D>n;034j6;oGhiOm6XAe
z(W)f0A*bmn3wS}B(;^}?k}AEMDpRgg_o^)<Vf%zKLyi<wq7E66B0~s6n5eqPM7t7Y
zk3SjF-gHQb%$Cigq6$ukDpP>0P6F-gwl7)ES;wMJ8U^C=4r21V^T=GDT8PWdC?O}I
zTqvrZx<vo3bWtQw^J;6mqIiSOp<DpL!E|1w#M-MEp?F$lu5qm(WIRAh$&fG#4L2X1
z^$6<0U^4Q{D8Bylm*n*=sRAKCDxrewZ%_oaSSlbc4q{*dU9@qQpysoQtFtgo*lIwr
z*4YP7#XATO;=0HutNYA>L@t0xM%oK13fzP%qaP_w{&QN2o1=>m4CUhf;pF6r@jGaS
z_ZCzrb<?9@LDu>~$Voey#PAwVWkMvKRw)$M86c)(?08Tzojj=cJEYxc`L|rT0Q~~@
zuS@F3Us7@wJ;sA2L-+L&4{5fXv|>v|n7~t4y+2YXqH(N1`pfPo%caCXg~@^-u`M+A
z3-Q=)_uzd*im)BV!D_(DM-r5u%*Kx^Ey8c};oqFxLRIlX3G1AH<j_}U;+i#`XlQ9k
zw8rl>&7PXQ8{Q$wWM(x6fz<ttFa^`$nxn=8o9UB!EJ7WX@LJ}<l$A7C|2f-%<~Xzs
z$lV`#>6?u=nC2(E2$6~)$jDpES`+8482FDsBZ_{nO{l70OyD!m_^kNkrJOXSoOW|^
z7O3aX4Iu`3>XcKspR(he!!4y$N-~AIAMVg*?noN{=%>o|S^<@*{CVh}h7g<AozrJ0
za(0o)LFhE^nfISleItDkeA!fw1J^Z8DhmPx`F7~zcz*y)AyAG(R)?+^1DoJxnkjF|
z(O|&!FqbmuWM&Xizo3RN2<KacHigdWi7+F@Y4nlX4K9Qqt1VSir$TRlzm>AJy(fqe
zVayHuLlA_L9R9xO+4o=_Uttwr;bl|QIlSCFoRX>>Mg3>GT<f`-Pq!C>0#+E$i+o(M
zzo@c4v%qYid{XYom$fJwnl$JBd*h(Up1zdaJQM3lRJ8StUh?b3*YN>fCen`Lztsw6
z=Kpsvn~|CQe_bnb)TNz%1t$L^2b>OuzypEAJ6M<7Ivc{&c7&6JM}Qo+p-@7V<esU!
z0fLq)rkTvi1p$ik@%^9`2~rgZiZefXXnIj^fM7}_<$*k&6d(zRMiw&0oQm4j9!oNL
z{`VR8q8@EV+>nf1f^zC0)cf_l_wae|ismROJTp#{2nox-v#C2<cSG;^F`HGF);9k(
zohFZ-hTiVt=)>OI&G*C9+w<mZUX!hOgPslyj*>2IWKAapx)%+KC?7F_)ckHpws5PM
zmZGNu#@cF{^Fc~3B<p<bVs3*HK@vD3IE1oSe}j_8?xTBZYpDg}lSm*&0nW=}z^2}i
zC?bV7_0Qp^cJYY@i4neyxm0NJX`^#E=DL4(pS#3+!}Yac-nL5pIygkP<C$stem8OE
z!FNl&lN#IJfZ2FO0*kG~vODH@-ln^9&!~5;4-W)z``V{<90kfp&!BoXy}7hD*F3rb
zL5czc_M{2saer;Xx!qy?3aH5DV*;Nh>ncEGnFq7CjoD<mcd)yM^%N&f1h9vw4^CdW
zdf;D1vZ19FOdk;uEJjEr5FCvHrlckl!7X=+GeT04)o?yNUT`irB&!+;A!g9fPS8*>
zDa64VK|$QF%&^#1Y#UeEaBKB#9aojGDIqeP)N(Aga}vYQOVJ;h*9?J9f(pE1*fHv}
zWD!{zxlX^SINX4nh$1$XiY)R@v1Ap|jfjq-9_|6-me!Y*z-o(N^|p!ux7p`HaS6)_
zfE%{HF6zDv{0YhLhBn<rHywY3a?+1SVDc5)B_OI!&cd+OHiLgRbr`LTrZqR|M~ISf
z93@L>qC2R2yvt1cdw`Emiq^|6fh1s(2$)Wd^7(vo7~TALXSVKMZ%_=&5`CdkKdW<4
z_;kc2YLpJR-V`|))Cg!zm7{B$Y&VkREMUG&5xVD`_5?FxxPQ9PrpvbDG@&J1+9zj&
z{w7t{22^N7ya0Wg_%;(>iTfFrg2ZW7R|qn(6bP1p<bHuC^=U%f3i47UTvIIh$*FMU
z^jyN5crcoo=<qRhkkJVIa}3NP^+MLv1~D{$V!fyt|F-yUSJG?2$CJSrR)NGX{sK!B
z{o7(q9b-G=30g}BY(uqPd&idtx~Bs=4$hK|cb+cdfTlEh2y#U6n`sziNkTv;5<w~~
z>hF$G?M1EU2yL20n`d)(AZrSvd)*f)hL#y~b=A3ws{2$x%tC}dc_otwkUcVyC|k!d
z_M2@|1`vsi?0a<%8&xm?q^fdJ(Xn;6$E-TvM{Gmb<z#gIUgcYdrB)c~_+W@Mr4u|P
zdw{5W_fXMw|0QPp)lwI;E=q|lwM6=d-K1*G-d(@}COwSH-k$QJ`u6rnggcUnD-Q6$
z_FL1$sUc6CcUW5ilJ}<H&tvIUFU*SXdM~W<<dneTYG-Vq4|t2MJsjejpNHGEdt1en
z+KpbQ?oD*=<!gX1=9B<p+*z7avHlFtiUIH$`^F;uCF~>gpIBc`za(E(F_6D3WMcXy
zI|)tM+IpGW+ywf!MlwCSFo{tyg;v;`>n_z%u+s1(^x{Wl(I(j*OE-pYL{zAQL7chU
zG|u28=?d>!X>rp`fG$VirTQ{me=E2}oK)W|<KMvnv*9FY9Tj))%rFS9?1c7;akg}9
zT@<yb>CF+tJV}?}yL8mwu0!Bd1!ReyGYuFIci7mb>l;E~vRu~s>+5yNFN5i{QRt_X
z1}E43T&5+Nq?L?*@??38$Rc<8WWn(B9sa=L-x-fWE4P`xMB}49kn<o~${1M;2+J=u
zkmn&x{Mx3!s~a%5pO5mw|7yS;%o4wHVC&4xeuF}SmiGR)GyA_vr#P6I{<p)MVNL1Z
z{{BB__UWmpap;YB0E$Isv4othQGVDJQM~O{EtK}NmQo;ysGnCD5oto>r-pNu0etFE
z#-78g53?|x+%;16Cs%*aPTIrx1A8cm2%!{+{Bav(668NH$fCqIs*YTx&|16W8?KZ?
zPceqFMg3OM6!fQ7x;itJ-LtB{)z5E*wfDAr{I<|9UNLzc;D3x|dHn|e$IP(~$;gx$
zWSJcw3`x?DOGgqYnfL~VlfGlFrY(~;AN{8jYlfmU@8NtK*6zZpW6R{*ORv37)tZ{V
z(&mR!m)j3Q9kM^%Qm9&d$V(VjIB{|{=Ym-}UyO*-2)E0ZXgWDNJGP&_LblqI%?Kno
zCN^c^3w2((cd2tq&G$(gf9%6KkIsDM4?ibO5_55M6Ttj<FBdP5uFZwfau3fErHyck
zdCAVwUya%eRlUuQf4a;st_gJaF2rbqqx~@i#2~gSiBkZE(erS&CLs10^7>|~=VlNu
zyht&}r-Q?UIwTo@BQdb;Imc~27rW^8%+wRtN3C?JHFtui<~5A57^yR>8yl|L!jeJd
zfc97s&8TyGJ=P*AR-)y|hLK`yEHe9~+b_F{64t~yutXj}45Zr`_+r^$>IUv;5+%b{
zQ|3nTM#VV}jziXRhsf@<Gaj@Fg+%x3z;F_m$dgyD{H-u@h|D?Q@I{jf^-9bdBmQXy
z@hwPxMgij9hvDIBLkT*##GK;EXMH%B?4#UexvZbrma!-h)8#Q?uEMbBV&)Cny!(#?
zd;mPYTs#0y%Yb`zjKl&emEZtzARrjX^G0xA2r4-Fv|0#DL3~8AOJ*urXHf}n7=W9O
z(AYCM-pKpo7jK7#3#FV!1_y+?Oyh_nG@5o~|JXS+`**p=s$fF-*bB3Q4JE4CQZEh%
z58c$Mz-{hYLBy3>w3$0$AUA~YTWCj#Ci3_SiCuqZz>>w7`2NBD@e0M$RoNO-K;WU4
zwI^KB*2}YDURY151U$T{V+jyOmNxHKX54b)zCMQ#FO244bg+L2#T1y~H40-JNn%Yg
z$lwfNaP*_YBCuQq)0Ye^g+TxX1_<6rgh)kk5+9?fEaW~*eS2o#`2gFx5tBb^7hFT?
zJ+Q;r2vMHOZP-Am-Wy=RBSW1RI_yyv8*2vY|7bPWh+UJ@HJK`0?7>4OT;3#JzxgbZ
zX(*n)p?M^zZTP@V>o$k~nNyQVc$mGN;}4dcLt-R~k6(Ba@zbFMt5F54nTa!&hE`z}
z(|}nT-3XQ&B#_grW`yHfCgH&EM>oeI#IU06qi>ATJwcD`uTKVUxZ7W^07y>$yJYCt
zFsQ>O5p$U+90n=~bfGcz46f7m1*nd<a7$??)!QBi`z`?0rM`hYtaWjf^I^2CT3uWH
zuJPw=uE=?CWNeUOZHxd2i*bS6bYeJzf{~a2J$#h|i|rAHdCYJs#&?YUjcco}R|ioj
zP|d2i%&w<>=~m(7x}SMTingl7KG{v%BJ)9hdjw7nHS0IRe`H}>h=4G&<Y$lhDj0Ae
zj=hu@dq}XRY8*|+DLjQyS_@BHUo(=YriUw1Q@ud7VIq2@8^l!EA8s-V2VW1&wjsr_
z_j%xKQ}zpW!C7e%-FX#FrK}GKR>dv5*z$Rw!fe$Vy=NR~28ThN<CKO^ZyR2)^=Zn^
zV6f3DfknJ-JkH??TnzvYT#ehxn22N}e?Xk=YHDo`Mr(jxxr@8-Z0l^1h7x*e`?!m8
zCXJ08>4P~(45I9G$JOL~TwHf16sAoMeyHhyYFJKG!n%2zRhN<FFX<~8UI*?CCa(7@
zhR_Eh2jFV1oT=qc^eta7DmKU&itjC4GrtlQi^hfGrGqSZaD*+uX_}z599V>GWs^of
z8f*s-BAf}FMHiaUeL(>A0%Rp_zn;g_c{cRA@U3zkVyC!W!r#c*g!D%QZ4L`Wl|4>%
zT@a6CMBrDOj9PpV0}zA+Bl<nq7FVI7!LK`rNx7ZuMfQpV3p}D#>}O=a3)&<Z7CM4d
z^x8GRGPZiIoGidHM;rC%ni_4+0sm^1U#_@t((r=Js0Vsr*+)&`Zv}%#`_ZUf<s>xO
z_rp(B79}pq?%n)u#|v^rM?eb}(j5Qe+(zOG&7DK=Vc9q*c1Et|<nX2M)GTLwi#bJ;
zdxq!?u}WK@UFt}-E0`fZL#x!2YEw9&JOj(Y3zB+*Jw09*{U?9py47-nbdIW0KOx_u
z1sX%x9GrDu;Vw1InVUFz@9a@b!4$maW?ZrBKma2NMO(B37&`2%X@Bl=Ze%0A54J7!
z80N<9n^}XhE(59vMcy4;JyAI>afPGFDvG5Qqi}?y-Xf4Wg4hvV36lspNVdQ;s1mNa
z2o*w=Th12Et7sdJn-KaV7N8}>gDU)<d%W_ai1slFS4e@A&EnkbVE5>@tt84+car1>
zc!J-O_}?lo>wj5eWn%cHHT=)@|GmaK{htZb80znYX_q)etjQ_QFp9Q*ZL4@2oZJQF
zv7Z%4s?d}TiBwtTcgphpoZAtfSs>QBI7|n{A1<8xk$scRY@A)gl-=F!&fE2R4Qs9$
z03~V4)-{<v^&hrR!gT7ItXxU%=8W&#HEh4AfEtQ~F)S0!e4O9M-q+>H#dtDQwsI($
zGNI%5L08q<(ekJpK7Ttrzpih(u71xqN0+zTk$!jApI-LY6fc4{u?^7Y=Kw7Rksz({
zn03m23`DU;N>vH;g^C};0hRBP>VHSxx}2OJotm%Wk@YYB4CIlRiWrO&0{LRi1aflP
zDc0grkgyWJ?D)r?nX1~)W<8o4C7}+nY#75ga<tX4P5Dx!gTGziYA>($FUa0B|GGZz
zm&dJ(V22ROXx1sCp^6~QNqzx74uRGoAYlTw0&kbwud^;a+qF|c6K!tSq9;8R)SKN}
zIBa1qYs>ndX{Oyb>z=-+=e7qF*KpH?`kIQ6HxcY%RFVEvt)m`4&DiM#zd*#h<rsBP
zAV`}`GLofX2+>3|#@T6RPl%zY`Sei}0>$_M`Klt&a_o@zXH1uRUJO1l9_sKG_}|{}
z{;H>S9$Zjd7l?9<yA~u+rbVZ#b-R^s5I#z@Wk=nsjisYi8~@X4Lxs)ya-cNT&nESn
z5WyL|kFy!M@Bw`+6F!Rrh)gW~9z$-nKXGmcFwNYb5<_ViA4CSI00S^dYDfb_W{d?~
z#=X8zWffR?4=+9PFzRZgc(ij}MOM8v?%NhZJs-axbsNON1uEB{8X3xw)`)txpVb-e
zJ_F-4#Cx`16$rYVXmgAcdPPAR5Ql5OULoZYvwR>`nuvI0qf9jdNKqpghgSx4){R$e
zs}dX`E@6!L@fLx1KU`#+m=TFZY@bt8F#!O+z=nz@r(Yi65k$QgH-8DTQ?39Rt2wf7
zjjhUL)&!zyABM#b`<>*FO*FH8?r%+IB`x=XUjO~VmwP!X3efz@;J-Q-x0~ERTXyPR
zZ2^2cZg0p|D78^r_R7yz;G~Ub!T7E5RfAYW`cJbF$bJ&YW(Le;vnGbzPgAClH{zY-
zINtvEdBz#)m*e`n)N9Cit6pc@wac5<tJ0r(8d-^{q?VPrYVXrjax~$%!ctl*k>=5<
z!2%nu@QfBslsE5gUGMEd1&V^TZpJ-kd+C>^>gTA|?G7zd*`9};e*2U<xYFA<_89g#
z(PHlA+!C)%RWAvoQkR}OD)H_*9YE&;l21ayM%NVB*?;!*g=*4U2KgnfcHV%18?E0h
z_p}dRPs`dnj6RNA;!RZb`A#(ne%97YMT{62kb)ryEdB&C>(Pw-Wy~V;V|dK82cIpI
z4jYecUA7k+ffy|@@vIA8!DwL_U}E-#3OA!4uH5t|Po39z-D8z&@KFVjIfc&almF3y
z1?xWRQWXYg@}Pvt%I6OT)+L0d?n_!UlKXR@sig;t;vn)-{J|B=xnRHnYpgdaIgb~q
z|H>UQu^V7j@|Rdg_Tf9gQL{S4QD%XP58aO|lCpsbeYcjE5CcmfV!oOaf0vi^LKShu
zQg|cAlmX(?zt$PX`W3<(4IA_eFTjQ@F&h1_d!dYbnOb9;m;knRB9oE#^kc&Kc-mjg
z02Cyval9fV+!~=Uh@kp$=>fpAMz{V*K^nR~M+P7wLWGmYWOKI!QFtW!CI|yDB7r(e
zxg_}IDTPf_f;N`-a@GU(zIwRcIn1YNDhZ?j(Ls?03oOvFp|iJfXb+6)%P0mxepQif
znJBF>BhRd?g+=v}(@}wGV*HW*@O<Pt3l>R^w_i_&ElxcZMyetWcW^~gL$$}_9eE<H
zkeM+6dmu=2Sg^@26Us#yOZd?!SEk4&3wj+fkqCDb2a0IsL=btK-px`dz9GrRN)aIe
zvfD>cqbEc*3_t9LQ}#icc0@2Se4ojGfZxgpW6nV8<AJf%=poE8@Z~UnWrR>b1d+$^
zO&;2%v~|UhV%E++R<?*6qR9bIV2mWy(xw1ozT>~Vz-Y-7?>e942s*ALL@v<lk5_58
z*T11==Z4FnZC?%-N9&SGLl)YV=_mFcq;Rbr1uzx_J73qYO#blI1Dds5oK0_0|18{l
z8cC#U&GOhvHdCSfN1r7RjhB}u;gB923zd}NKzz$j1|>&ttIS@-UlykYTop>G9=*B<
zyw#^j0GEO0SgZ&HnhTHk`Hd+I5$XsYDNlm_a*|4%(E*%H-1z9tT;C^>e<%%$M$eWm
zJ!DuIb(p9e^q;OV{c_o%=z`v?>wCrRBs9f<D=dxe!yl@vggZJ^XHfTUCbx&k)pS*P
z*<lQ?gcge!dXR}Y2m*j=+>xh9hX<Kd4Wb;B2+i(cO{WEoXv_eRyXrIJN4PscSPi<&
zK%gee!ZOZ_sfRs?y_Sau=lpx=ONGw@HYI)(%5Rof>Xbr;B_2++`ZFV;-Mpj^h%j1X
z_9tntJh4AtUflZ|FY1siG0N`iRBg``bK1>R+pZEWvJUA?@iAQ(MDdg^A-2Hhwu!u`
zm(Q4+Iv;y*3^DA2#fM-LLZ$!_({p^S-Ob<FtDaOQpdj|AOfgl^<7Q%J`?U0WP)Yn?
zH}svqqKj<~3x@W&m6qd@JXxS4Rg?pHf4X14$%oFZi=6IXI9z@x=hXK1`bncE$6Px$
zwzj&|hnebI3q3uPs-B618$j_h&~?QRBV)Z(oD2r%x24Z2?M<}PJZx_cJi$244@xE{
zv9FqsBFw%YX>px52n<QG_d$+)!W1)yz!&;fHtN`a?LPh+_?&_B|A&i4+I~X};TNKD
zD*-s~xpPQVh7kk%f&dQZB^BjQM41sTQt92UFps~p22QF}N-}A=D+EsGcGAMUGrETQ
zoQn5sNB(dA@I2Juh_aX@MX&k)Fa<>`UY5&X6j^d0Bfc);=2>>X!_K0iF8zGlf4qMf
zOat(@w)`5MdN+L{@UCn9n3?@VdiAy(R@$aP2ZRExt+l9VS^DWR2Tzt`h<L9s4>0xw
z+?lcR6(p?>rIFEB-U?|G$>^ijJ;{PS2q~n^k~x(K@63XULvyTBhN-EgA+N@jVj0c?
zv(05cZIx598W%=-YuL_=rU3f^k*wO<eg!q_zPve-(O<`iQ>%OY&-2<kZ8y2oK5^_>
z7}4<wEJ(qWul;y;=>kZJL#Tp*RP<Uw7pII_GaBJvsvW_C`JaCb$usSSZLOq)dN<N_
zC7KFnvaaN8ZHi`CO+hLV;CwcGa$z#0B8w4(yZ<`Wx&vg0(2$tyQX@#<*eIL7MwbZ$
zkh7``##<9*$jfo7;2_fFdGM&i$si(nm$y#kF{N$0l=fFTb=h6X=oGIX&M=~|0Ngr4
z<RBTst_qn->Sb|RR%5$6b0f5J{jMER2yyQ6h@kiz*1rlSJyOo7-+8BKesR$wu?L4Q
zYvd48Cd0yZ$1+?ojvz#$%)QxS3>x?y#$YVI_&^%Jd(xZ8=pO-%JTB9jOa+XlF_qRm
zQ;h&9F!GIJFw5v-vU@Z%GebJVo)X9SL@lz|6M%fIm@IdG*OiM6b5_pd7iiu4oyt07
zV<^N(hCsl1sTA?lt1JS`vVPfgrYKgPg%s<yRjsOmZ7=?{tMuN!Cha8do4JQ%V-ib%
z7-sm3Y_n-Ejg$@_7HNJu1{9jkF*fYTckKhJ&{2?DQp6!F4u(sC<H(`bgo!)=F}oWz
zMW6{r8XJt&mKtof#1#nuIlViu{Rnen?xp;y?4_Le*Ta6mfq~Zq22Q5vlu-;~6r-Qk
z*r4!C%{5xlXQWR^sg}LUCI?kk$9794PsHHTUxZ+uIKjm=QUr<bbIwx8H<Mq5u(Z<#
z?*Iz$cHVbF9OjIyiP@wA0jCH^-1+8F;+Yi4W<M{~9g>~0pN=QC8|5C}*d6>3WOP?g
zuov4FelC5i;rTWbuFvz<f3g?`H`Tt0sA_-2yH}W>@2wl2P+YYK`J37vujX$R>{Q0O
zg>Wfby|V_k26I7=+x^`N+=Nj}OBe(M+bQVet|tZu&1}v|>rAvqNr1IkCKcVesAkQT
z&}o??GQ5!D_k!)R<mcMJ9grNVrtQv271NvE;Be;NTrXvno)}6t<wfxX`-hAM5}bxO
z<-&^wu7!<jw+-toe4o|Bd~=GhRnwNq6E}EgwXnX~o88U+t+2ytb?Ua=rvvw057s<6
z0&Ugj4b9az$g4B=pO0<bJwh($R#cs#DDxjA?`X0gpqyj++y8bfvHh135)(7S{~AWO
zg}GsO*!r}kUogtJzjyB|*Z{gtrfuU7By0RwkYzwPJEphXDI=cP1Q+P*1uv?E#F~BA
z3Hd&kRuIa?!}F5m@EpH)Ijip8@_qMwKFux6RER(tXx{dMnS}YD&ykHAKF@~2oSnaS
zH_jxU4W${%viMU*w-k;>FKAKSA6(X+M(^!TCtJ^GKC|4iO@Ad@CBu<Tv_%khSdIg0
z)`jd*4<DzW?J02#A%6Ow{*1Pf-5*{%?@kAHYhLALf3bd1IBT?~2Mw%?V){6Mb6?gs
zTUw~o!aw7@%HsULLc(k!FsA<4AmKpq5Qt?!0<=tICP6U>E5Wwtpzw;Z)?e}aDv4Dh
z0^3MduRgL6Tg}N4y(xOT&IBzN<`&Q5Si(HXMqp16-M6RB&3#uR)yE0Jd1==7Cj2zr
zu3keIZ2Pb04tw(BBl!}hkZi034uI=-cT+YzWKJT8fhRnTUzuf~paBNKPdXlWbqFM$
z{R4$?s4s>EfH#*Al={g^>GzewuO`NsBOC=&IMKbsJB2`l$Q0H9(B;5n+m9-9ZHm{k
zL8Xv&I3n<b1UuiL$S+4k5b4jtMhxg!Ya$`2(YBHSqFh{Epd(jP_D}->7zA4=s0zP9
zK#~n%<*R8eNFj2jAK8}kS&&O0WdMvZqOvv0Zm?ADYIx+hfk{6spsi(-M^>X$5X32<
zGi-J2?eK{(5_tl|EMI-{d?rT^k)5e(ru@>)dNW_2DP=)#03X4jB<n%HWF}zlOsyt4
zCD3t*Rzz|KIQDW+NB8@`6jaU`2{B5*4{}UpAR~%#*Qez$e8<+Xv>h{lq{FX82rY9d
z!WkNP)#tBeH@Yp+X2qGl2}~1S$jHQmuK1Ka1RAwH{?s&}f9X^EU&YuV!9@KW{m`u-
z(E;2P*{<kA4dZAAbq%3uGEYk;6n*}_eSgw5NazucFa)c|3MdS4ieUi{TU$5R0EP*D
zgkss^8Dp>Oq<?215Y*gltmepFTb<K`>FXQ9j@6%8rMxR1s^2=B#U$L8xtJvC#XMz|
zF&Td*mdo2uWq6&tiw<iYUb?gfj~8_L(J-N{x>}zVl&|8QD<#zsf_8;Q`Yv0#ZWE(q
z=PlB9L0GEg3zxBiGz>ZJXeY?AU#W7q#%)N=YxHw9dLxYiQ7=K2lO*Jx--k=k!6#Xc
zqp0cgL~kN>`J6+<%L%eDDLd%JgKM-w7ACk?`OW5MuAvIFH&nu4Ov&$al6cECbEMqF
zfcML|2scf2=C}|w%>cCP7{@m&$32_^pcWX6BarJRrayl;*n5YjSCyuChWVybC%m$r
zaf!>Ib>HCF1|f-BJHb=iXQ_75Tp+*TM0l!@>`P1=4bV_1Y3U=Rh~>!PCgvDBj^gnj
zI{hVFYF?e?Z&vaqCfxYG9tt*(RG)A*hik|K<^o%l4bSm9|HAEhJnZ5>(2-hI5@uzz
zoMp$+%KaMWh>kafNyQd|H<M5_!+Q{pDRt0u(qTgCpnP~mXLia|x(I}Xl7LGQ3x>&y
zrGPWCh85HRbz(XLj?UKG9B*;fIozYyIaU+$%NfZAx*Bif=o(4m5vV;+NjnsvSShq~
zIR5srhUm1S{S^9v?T-+^9l8$g&Ea{qbx_kxSx9)zc1ynoSx;W#Iv42HA}aGKYuU9e
zh088q-*-I7Fqg0*MUOz7v&ir90izNY2b_kxLA5_Bf2*{YZPMzl@u(c6ZZ2K%{k^1*
z!h+I!<7%w-4A-20>bjsUDyR&!j+*uGDbE*KXR0Qa2c@E4mW?*v%^X_IqgNgRP!n*(
z&k2xl=y(hu)2ZgrFGayk>xiQ8uJr&A5z(Lcl1^lW(5tEK?I@-Ko0k-*!N_^NnZ&+S
zwq&qzz^4frv=T=`ARjNC#YaDv%4`vUN_utgw2eF_v)9jum@pV>iM-oaIG;5bQsbdM
zuOixU%l%%dF>Wnb5hf_*BoF9av84u|!+USUvx49HSfmR>IFU_L3F;@-P=Yuyg)82$
zqvHzh(}ifUBqrOH28wQ>xRb&npb)H%o=J8B(RWr!Zfr2d5zKUj$sK^D;e2r(2#qT}
zD1~k_@;y7ao<5)2bkU5?@#<<)Ru7-aukA-Af<r&Nt)0+q4f~l<z-_$~*!zxEWJ2tQ
z48YUoq#L0ES`fT$5^Au<&I1ZG>~qcWdk!@e^lI~5KRTW%jre@MjNO74cVG0#_ev8#
zKMlvS9k0Z!n*Q)%L}LZXiua^l?aXt*6N6cLZEeTj;!Bo2n)MhP<dP7kXU>Mh1kS}7
zz3sfM3k}+_*jo3S__#eZya_#@=jHUtd|NiODNjSzT&;eX!JeaWwUCV;F>9=IU%f&@
z;4=v=!0lW*^-4%OIIpF<Z&t>DNOcPMHUj0BMI1S%ouehJLBO}%nSPi~et`_-SR$HF
z7-e=up#D<Afl;A|Y}Zf>l()bY7iS6Wj35ymxsH5q;Nj#>ed*+Q5B32_3l};QdhAMq
zCl*!~`hF2)$=}M=2=b>xcFHR*llDREz&lU6ppE|`qm=|M<hcU-r;EY<bKw|JQeyKw
zT;QKq3hyJGP9#I5;|pwclcb8MB5d-V{E&@95stYVoq%LACEZh>{fp4K%6=1Q3IPJc
zH1zU;&56Ew>g8)Fs|^d+4I<#Nji(xg1m}LS1Un>U-2U9?-?+Phfma`wM3=XHXwZv>
zCYn+pzw5wj;;IP44trzd$GLw9nO~g+a~f`yMmJ1tcgkh_!;FU59VH7diXRyX{;~@m
zHVLohY`L+N;T?GE5j+;p#zSUk^k&S8^Zb%ZOJA$MdGARE`&ghg`Y(SC>%<ZzLpf1}
zY7wfj%I%O}hxruap>g7-a?T@ddWK)9;(-(l^hMafVMXMNq;NLt^+E5Y5cQ34qyb`a
zb~TV9Z2fpBkWY<Z8pX=A1SteFs@;hxR<zfh(bBNNO_ESs;s$97wh&k6MBJy1P{F-O
zDkOf_<cX+|9@v6o?}sm0elKcEMWypW`1SSD$p)M?zSo4r`>7?(|Hit!<o}`Uou2~>
zl<3dcwllG9+qUgY>||n_6Wg|J+qP|M^LF32YJZrjefuxmy0^P;pVQ}au`zNUUZy_2
zy`lRFPSp7~@n7Q<8^b>wIqXdTGu~=Y-TEgx9^vasFV{<{Chaht91<6-LLJ*!y`JXw
zjf4tnFbS8Y?n2T5sCwP|J(^oF$#_*`-bu(ev?$tjxczubxHCwsbAEPw^yKxos1Y4$
z?cP}W1W#@UKc%3_j*RJ6;xviDfiVvGS@-AZhw%jPSvN<!r*qi&^)xLRx*45MW=8M0
zYXW{&cT(P@3|Jr-R3JfptVJGT<DCHeNSLp-gbeS}17Tz*1H1$Z75C$27{q5g@p!^H
zw@9sus`O<%GFq<j5Xc=;8ajzTwCx08pHXn+81rb{U#2a>ATGrtrw%RD>tu~Y)$1SP
zI4iy(u0r144DL%5n6hJ)>`=I(Pecsg?W{-b2KV}UyUfx#2u2Oyx&qQLB1A6Uy)Xb$
zDp`Ik2$FOoNNWPN5#Z5rd62H156f`WTBl$8PM<bJa(zQdoyR&)?cgwByA<@WQ`!P%
z)^QHk73m-h`XcjaC4wV3ae5*d^_|+evnusN6!+1CS{4j25FA69oSKLVnMf3#v}t_>
z*yu~S?6`?Y#cKNrhTW;q%1&=`?~oG)Ce#L57ds<H3CeS9FDhbQ01T%3cYoh2e$xD;
zLju{-mf~xDMcMd;hfyXh%DoK3Pu4hs-!U3K9AS->fN;dhFTOwst6#!z<ofz@(Uxgc
zyqU%OA>5Pr$)Ip8SUc1!K6__1=a#uZecHq<RYSxxcWPc3Aa&sWI}`#ljx7-EVHlLZ
z>%lio;NNToda|KFVyHpqPro&wQle97AxeHR{0`uU(1Hxi8PfB}cd><mc)*4R>SWWQ
zS$1!ej^Lzdb}0{nmf0K(Dq7hU{1)VRE8=Mm$dy1CF40p`H7~A^*HS$>2G}pH?;k{<
z%)x|w7EmyZ!>$0F1r+Blf;87qaX?Z$f1Ga+A1`24L+L8*ne+JrH1=YyciL@lqUUQp
zRzc7;eTJ>w_(@R77kD4C@+x#O*1_yLCJtqttM6xP>8)uymWP;V%U>ZN4Be8`1rQ*H
zhLW0usgz!92!<{HcCD?>i0L7nz+_6;<6wY7jR2~ns>b<>QMs|&1Sv$=YrKRzo_n%K
z3}GTvoA#(;HiICwHlGbYYPkS|p#YFO752C@DR`pb22C@eAYx5fa-|Rv4!4DJk_&IA
ze7+wRC=#2}4#mr5<6+K=d7b(dJ4>=llYqzQVK;DsBMLB>Cr8hTZbw9mZ}z}&JofDo
z+ewHR<)O+mbiHZc?+EJh_m8PDofA8xPpTCIy{bsp;7E(Cy_r%WleWPPzc(FnDCbd3
zNFMAQ7AP_{;Jr$#wDmQw)LFsu6rB}qEgzp%4o67V#w5&gEU66@?sdj%wOS$1C0Mwn
zrs?lw5p^bI;#f>UpW0beQ?hD3a&vw`6YM}8%G&y$9;iBP#VMzwD_0A!`wP;C7^1^r
zm*qTSKF-xd{My$^#nh*uspG@b-aDL1ZK!OD8;X)!W3ach!Jre{-LDr8eMVSuwUYl>
zzlJej;CJfMsB_YJ*rl6Z&aYGP5~ZeA@4|Zef+xY|Ajvi6a_OKP233g0a)g?TN8cLY
z91bj794v*{<+wlRvLh~@!ouOCyj@<ucnboTNU!QlQf}MqL3XPK8%`Pjr1oLX4j@=r
zJq(0p`Ri%a)`y!&1tgK=PJsy+hb+p3wJ?(>)n>C;C$J_)L(!kOD{d2u#2<+}ir%~$
zv38a+X{)6V5?0cW7$0CgU`%8jm;ZS{`RLF~Y=(rJOaF8;h-Gbg2ibWe=ekk)Mk=z@
z7N{bp(x#=ZZ;UE9FK_^?9oEm1{@t9|^4ftV2xt~Nv`0#}1(!sV>4z0xjT2_crHe3Y
znPU)&!7nC^Vk_BFs))eo_X{bWme__|3TgZF2^>k@uLf^yVP};+>GH}7{RVt&+2zLa
z=y!18i`NK0&_Z~0Vh_OEz*Xf_IXU!VR#1NxmOsx5Phg<rlA{pY2Z(md>>*D0pdYWa
zIDH!y<2IR!kC$_o>tpm(S_sf|NSTV8ST$)rd_E~)P#mT|27jFBTsp0o+I0zAKr$ac
zMXY?|HSDP(y1hkg%Ae$Xy_<eB^r>j{s2*E#q&A?T&*K1&*wnyJh~)>|jR$@Z`$<%3
z7yV9#kOGvI>KbJ9?9K!ICr0e!N2u(S$|ZM3nUEu7(iDpXET?pW16xn#`6w3}7%H)1
z#VUyH$8^&YdJ_1D;~e+z(%%n$YVvjXu4Mt>1p02v+kFtg=s;KxO<&qb0q3B^h-e-r
zXOn;cF$=S(jzDNZ+&JgKLOZ!rQ$e3iTew67ULiXRBR?V4&wRnIp9Q{V8*nUK{uX>*
zMl>?c(?(w+Fpnn}`cwt*D?n>D&NbJIbbN#l7tOBblR;k%H}sqy_vg^*sikn5#zyyd
zda9ctlS|E*jP;`8_mn)AO#H?uY^?7u9CjCv)>n|e@8EMDge(7SzW)H8F>?HOmGM7t
zR}16#<bJI7gRVZ|*vMfmu3yRO($?k#5DMa(rS!u5T>Lq(^z=Pu!&OWTiRxnr*$*Fg
zSEpuysJ}Am<BS5r;`Liu`h2zgnR60%OPE{4)QJhvPdnRjq#-V&qZ$r5lX;ELDd~dQ
zY_yA0XHp;r=cTcw$I57r)%taL%T24QL><*+s?7)z47Pz<Cxe#k*xOM*oK^7Omse_G
ziMrAJ_$8x-bF6w5`h>>)siU5rfC7v49Z$0KM{Qy*TgFqx1$QB_Fd;lfTFqp_(+Oix
zy5}Z-k@DGhbU9n!JA`1ZPbZLf+P1~~686CH+H^xfiMXKf?uIYqS*Te7gEic8tk-4Z
zjqP*Aii|}J*2RRRl^E3B{_OfjV9)r%^#)Xc4x9VeLudbos?Na3%<;df>OVlw|N7Z(
zel&+*9A05O5A(I6A4A)ImJ7O{)p$Fl)oNp*4z>J4vl(%G{7K@zj3)ERAX4Ehch@UB
z6El%(U|hD@#lf2?@1M^lhdw4KZ3#v_<lHdkB+B%-*KlpJoRh`N53g^?R#}dO*p5|}
z#WHoB!W{Lx{;IO^vQ-wb=C4}rr4WbNg~KY&rtEHSD$v>m%b7|<S<D<(QFg>|%Ljf+
znMQ$htp<*6x++Bi>pBS0wLu5t0-bAkwGHBSkuGvJ6CJd#Td%L3ZeG@(OA`6^!Kyy0
zwxsq$SR?tw#$@X9Hg@g<Y`Qj=Wmz9TIR=1-h4+XTaS0L9jB1s79h49wn0VksrE-*p
zTgG_*96|mVifsPG+JomZ?Q1x#kUh&lLt@-iO!>qGELlbAqQec1w?l**_dCvrvlY4}
znMRl~ImW+WCK#0Sva9Wl$Omx+h5EDUvV`zJ!FHlS7MtqGLImUf5bTI?0tNj!h;+B>
zC<YiRo&36$ML4@jrvBi$gP3Z>%Vbq{sZk>GyE%u!6s#svhT?a`XX5<1UAj9e^0{)#
z4^<n!$_ZN-X`@yLu-P>v6lwvbwDpf(3a>cVa00bhKETQ=U**LTHL+q#+F(@T+H!y-
z<lv5w<c}~V?A)Md8k$6GHNhiUfz8OOASTqZ;1VpwzTrwEu_%3=Q17;5q4a+(QVfGR
zha)YtQG@Ur_@?|!zd6kBp!v^B68&pOU6G%9<OgA9#YGPfxbUXD)26>Ng2qBoiY5S5
z#RX`Vo)2AIyv9eM^t#B*(OMr?!fYVX`jg(U?+l*Ge+$+=S5jU;tl-c`Q@?LSPn9tC
zv-V+eYG7dmDSW1c%I9N!FbHyDzQy~e0|6P;C>fgNa`Q9Uc-z3^op;bSiKzV9j<p6$
zG`FvsGxS<of&Fs1>oFjnhF#?GN(-%y(~VQ89C|f*HTB8vb|^y^$;_r8gZZ2lY6gxN
zom+BlcU_!w9;C&Bfs5z7&e}lr?p)n$O>g1jYYoZ&0v<Ke*p_SEVv>XiR2!mwiRH6N
z`?TB)KJ0xP*Xbe^kKme1S%JZ2bv;hO;aY;D5tkaVBQo_ifBlj)9u?BHxf?9Hai=RU
z>;@gv?IT|Zai$!^33#OW5Xw}uZ|Bu5h}&iW=j+H&gdjHg2o%I;gE;<c7^Nj2rG@Z&
z5f8VI;VB<|pE(jwZ$z9M*<WATk|zyP0Bf4A!(zDMNwb};X&Yv9-p`y!p}T_$zwK$o
z;tw_4)%_r<+?xA*Wg9?-^i@_W$gu>O)rRqzEkjXqY!L9_3YM-g8k|yTPgzJNk>PN1
zVGBp{ep8H30*+(9eO>)`GEPhkAdUiXVa7R?+Qrj*U7MrCz&_NHml_FT`6ARbIXrK#
z+u90VF?)h+!3&Zazs??%C#YJ|ytD$E#0Nh8;VWbuaBzWP(Fp^-Q=L`QB<VA~6e@lW
zXV}-kd?k)aLw+P4c|Q19i65xPP6V7;Y$!YoxbW#Kf_We&6ch@V7Uf)+RcA@|$2qFu
zt95$MeM56#f0*rGXWX-ZDQz(45_mVJ8^CQ*&z@v5&mJ-&-pB@?R;eA}E=bRrT*{0u
zk{z+(rSYD4m*tQFpDeikZ2)dyoDEpnq)^DfZfHhCiQR=yNRvCW^naGje@KC>EdP6-
zs#N{~bJ8Ppyi>XT^-b9ol6zR_(0<kjkM@*|@J})iDUohPIqVVP>-sBB>avFibue(~
zz{0eF3*`sPug};<DVpqx2+A>m9PeugYYwzW?nfRjpx`&`TcGgL`QqUoqTsEZWI|sA
zW?0htb@$rUq7!I@l<zx0k24GjRdqDJI(Zp2sXJ{p=+u-!hkx;9`X*0Zq&9BIna%S6
zLUcfkN!p)@Swwu6IuuOFq-_-6M=eLae;ixXn-!K;X=#I~MIeQa6dHycRT`AXgW~@e
z<5z}2yb*K(d$~15??K65VFUx8ZrO0>{3gD$x{9S^U*vg=w6{B3a#0=|>}cdb{{Wq@
zmBSO?#db26TiVXOG)xGqFxlV${VZ%W5c(Vj=?RiN>L7b#tVWiB!TA{goQ26Du*tQm
zwd|Ji6`55lSC(yu(p1hC<>pJvqQR|2wY){i3Qig<3#N0~q-rj?9p3E}coRAt{c0va
z1uUWYX92@aZby^i21`zjRlAcwXs@dE%PHljNqb~==h4(v@01^?ACBQhN*7CLe7a+-
zJ)C{oXFuNeH4qFU#~H1}ufi!nWZG84HQ1Qe)DU)dy!R}MTg7|ohXl6C_t~B0S$#e%
zPZwr4`HxcgeEo5_0LVzadZI%1vqX47Y>nrn+|<_!z5ATs?j}rvYbsQv>$**RB`Ky&
zD?iADsP>xmud|c|LFg^cn>GrcrlSyH2DmFMlX|#<J?qi|8~S^#aY7iSzwDL2&16sw
z`&S;}kD@d}kgK>zE_Lx;Iy&vL_P0Dc9<=kbAbZ)tEP-|R28}`RLgqRttpC1D;(mP^
zvew4=cwdgn&6fn?KT+2qn|E)a<#$Jxq<;VF<^Cr<kK_Nm-T&V%yG3;^+XH44@9XLj
zBg#uRuVx|;BnY(HP}G^eaeviQ=`}E!c!PM0W~u?(<Cmm?QZvmYaQ;NjFh|d-?CC-F
z_EEd1M@-!hZ+rKrI~9TdM@~v!{*RmliWD?P%eK0nA?fsWet2##_2eQ#U;LWv{A2bg
z@0WMA_UXz4I^NW2bd}?4RNKtmRK|XOyXPPvu3i#qzJ{$&u4W&HQWmF-5|4y}?XVYR
zB#i~}YC(Vio-_f*(%w=Hh8=}>tRn8NhuOTh@sV}f(n(Z%Y{J9a*fT999JL+iny%ox
zaM|>>qtiL}9=}}rL89=a<}ry#%)}!r`?eFZ_&9Mvpr;B7BDM<=0pVc8A1$`ruNPk|
zhuUK~cEPDW=_sXsCJ-V}%f{se<$r)0z2SW@Su#TA;z$S>EGw`OQ~Jh`g$7*l8%Bbm
z@wXX*K+4pinuId25(W|1_}hPN8gMPR9(K`&j1lBrKFPE?Pv6svtP`CkWZ`Up<^k5%
zSdsX2#}L9(mp9?-Rl#3!-sq-CBNQc2rK0KQ-XE|nQ;>mmJBIzjn+>)F?wo=Ul}0vX
zL%4~7ku8*v2OPynW-Crq#*)DM;cy?+1!koY$%Kb?Nt{E*RJe6z$@%Kwimgsy2{P;j
z(oJYE(B9YK3Q!}vAXXvt60W)1Qv(R|5CI%WjEna%5=j$OLQ9b#BoN^{q9|1mv%f7+
zWV{#8h|{BLXb2G+DIa|b%_{>NcXsHuqwX`l*PH{qH;y_+La{@MCN$wuUa##oS#FXg
zsLA?f^tu=A7=gM3ps9Wc&5hb2<!Wc#+50gkO@mB_NmIne<J3(Ho9t#+`tnsM=@SON
z>pM<iC!2P#vn*pGSu=?2p;Oz9q2nkF0xPV!`H^Gie`%dDGv4IGp~%ctq*Br;_ON+`
z^4rHTixcYHr1F5&Wz*itA>Uz$!AxZehySrfOn2*U<9`3N;x<wQ5L3GN)e&xTespnJ
z&A#3<yewabu~6|xYAvNqCQRUxwVnOF*BKT1GW0J5GJk>q>PhR3M*&_4xj%?w&v7Ur
zbaou1LVJN&PN|<mW14-Pjf8KEmZW%+B3&;VV(LCFNEf#rA{t7>gMWBvs|?p&i!DWP
zEU;nru9lL);asyNEkb}81F<{YZJ+T`Y>>1eHyw=*;Hfc%5_-NgP$;S~<}}1(F!o>I
zp<jsihJfC!06w=6t#9NGK2k3z9)t#8L9+QC_&~g_Am)JtdWg|A`UG2l@i-blS$Kqn
zBb4;1E6il^Q7wC5#Xc~1ohGFYRxx<}P?-vK>EL?ydYH&jcesbXe<&5?k}Om@?5isJ
z?4%BHBX|@+LE&}sdGh{(8R<r{9ddk91mNNLH`<dUw2BiwfQ*J{pboljr+<S+84gVl
zB|hldm3ijO0KJlY>jiOAM0%b{{tBSc(FcITJZ=$Pe>K%5s3bBYlJENRlzH8$U%rp*
z(;?6e?Ww&RIJP|I3%f76t_xvMvo6YCrnk|o_re!0kBY`KaG4K<(+m+5&dtxnwHk{s
zyTPS5<RQY=k9x|wU_Amy3dn#$aSlQNm!K;$juT6>$>-HVy=Y3L<b63k6wS##{F02W
z6f*M2a=q+04NH-V76u{Nz|mztLVuJ|lHK8CAW;~dz&RN%aJjrOxYU58zRPW(VoUVd
zc452PosG0X8<vpK&)Qz@x0bwSn1))87p461>NoRTur3eLU)S?(#4ybJXY`F2uvb7d
z(!@r5L9Sy4C7{X;(xR(hwqOTS1ZEBye1Bw=epFR7A1*El)mN`K{<z!uz!hff`RR6F
z6synWn{L(D`4Aag^9311$6&7aqoDdHUTttDCM7W@ChzloQwQzSnrvNg{PY|j>o3P-
zK!HbxXkatnX`0usw&*}x1qwRHL!DIQlfsrl(_xIA9-Fzzj!+;uV52P?&UeE&u)+Z+
z24~UO_#k0@b&|IP-69T)gRKY&c+-Y9(#S-)KtZLcr5o{p<J1<l8*XdlTU&t{BM^|?
zT3sWUpsc4IbD%2H-a7SaszPY`z05XkXfq$2$_&JV1A03N;oHmDz=3-42>W4ivbW18
z;`Di2J0=J=ypFrL4$t5hkAmJ0cmYlk!~ro4*1h$3A8d#UtlE!uT(5OSm1qXN8w>z(
zM_`&Nq-P+3L<Ly6aHads_!xc@UQ^lX=~bFF%in;%GOlkM@88EfDgK@tR{wEwbOPd>
zr#Ioi{VE?R_jac+J>4wK8X4bN^X{y6O%T%AO)IkhvDgHmTiWP<_V%F`GLki|QjPt3
zr$h;9iI+Er6fcYKSQnI)VW8trEOlqN!RE1vx5@C&mor+mko+OWeCt9wKu&7oof>A%
zlspU$ajU~foUVhYhbm=iU|#<^tbmdE6Qs7fviR<fh2fEw{x+qZ&z?%4SHMbB^&2v{
zNi~Sr{)R^SoNjD6bJ4B>xtHf?wT(=}CnqtF8tzW%Fl2^VlT`E|f7XE=44tX$L(2{k
zG(PQPdj{y9IN>b6caPIBS=6<?trHSk>qI~(K--4(t8J!fafXNP4pjjK(tK}Hzo-8Z
zONB7c-~ELya7s_^9}S<8{vW|j9PIz~wfzrUm%|h(+kI99-|H{bHWeUM?RX>#NS?z+
z3lDK9vDI-K#F{!Q<VB*y11qJ6A6%}`>fI1_zTYMxxdQ@zzq=hT9Hx@luKclsLz?WC
z4+Jn7LD<2Z#IfuzTtrDhq1^(e<Aiad)6cZ8k1yc-QE_P~Bv2ry2SGEw%#V0oX`XE8
zGY1K7J@K<Fz0IQe^7;0Mv!ge(V;f#LMc3tGHrM5!bi$QSH~S<HY8Tt0=An&1P4He?
zH>3a^Osw`UkZs|M7j8m(!r6~7bm6P~xuCbo5e2>S@(EUs-M-)+AZ1hxmA@d|cAjX+
zS@g;a2W9*UQ!nCwLjvmvRvc8E-nz=t6;YoD`>nd9-224I4{8pWP#+kk+uAI6^c2r{
zcaLp9m>-khD^DFBr&yJjPxxxR%FzpfWMD3>O`_t7!HgCRWPgPQy7jRW5(e^6^?J84
z|LINDT7aJ4vEdo^vq?G4_fU>f&BX16Q1DDrtGvpxGgObDO<!5ts??nuqkb^bg=J-0
zJY9RjN;rHQw}C}S=?`WvKo<4q4}U`#Lp4=WVX`uIzP{v~>{m^7=~iK-S%0|@f8+JQ
z;KSo7zuqO1!1*YC>7ju0XN!gph&rwwDz=@P+363cLhJ7<yHsA(@TyM@?x~EYwH<qs
z-?*&S{CH70$JAOs!NqC^xI$&cdEM84CKUMGL9Y6u4fs<74MN7irMb9gTWa~a{xaiB
zd||y097aGW@zrf2HE`2#FkGqir&>AzB^y#w9VSL*NaAOuy{1PPpZabZA&$p?V9!XZ
zC@DqE>`_)wBiz@+CfqlYCT?yUic=Y+EGAA2!%K!leBtjmvgOATH-nwBMTm&^g7Xm)
zIV+y?J)MXR1h2A3-k}QrLBB4h$)BT+-#XL@)_B8G!T?w$acinzWFQ`>A0l$uvuM{(
z{%y8+Q@Ftpnkh<;ZFE5<(p;eyQ8<g3>iM{yWu(@8gxwC+&a^oU)*jUFrTFPgC}8N}
z6{d8rl9(#o!IYX)m9_7w5wqwjrIPEWvr51`<8PQT^LdoA&bGRt2{}M;(GA_Q6O-s{
zZ3+W*baoYjrgccEwWb=ySTwxpc~uKCg>slRy&-*^@~frh_eU1ZIianOPv4d{rf9{$
z+?hJjZmqy&FcfPJ@S2Lk1?P5tI(zUP0^IXW39Io^#5<VX+AB3CYp}@Qz%Y#8Z-`!&
z_c)*C_q-HOV(>2kOgqfbW}8{MvKewN>(VxCWM~UHVlDyom7DJJQIw9~>&GI-I}tPJ
zzjVF}MUQUT&{GJsfcu{bGd)4G0moneSt$OQY{UFNrxh%!Yy8In`=9%pe2|p%PIweA
zJ1jDJWG=r-W`1~#24M;-k}(nrGTNW(n<<Cn1<5V-DTRKoJv$CAu3@%eB(i%DGCH`d
zX!tB{vK8VVr<e(27Kt>ezf@qW_|W`8GGJn8p1~LkUPgx(i?Sy@-X}A0AWZa>sMzKE
zEj<YsysTb!CQhfrL;dXQi`B!!rQ5^X!wgL)2iKQ@6WsNevpIPvJ0}Moes+|%OZ>(1
z_7ESh`;<mWJeoKpFcw$T-#TgV6>LB>gd}tQ<cYdsT%Z3$nzrdIzZPhy$a0b}@0j6G
ztV-!sq087SRF#=DwUmstuGf0`gME`PUcR70UlbN28O7S2gjJM=$3*ke9gV}VOAPS(
z(}NnY6I4Fe@?&M~lZ<%BCA%lGywVqH!Z1FYt8`42<$dogI6K`4!11C?{%P(|m#HgK
zPt1g%tIgmvth0PhM5dcPha2-(?X=mTCnk4pYb5mozvxTz*bh}{y;(51$~x!F<}4%M
ztS_yr%kmrB46#@;j&^thV0aFGU*=oZSVfVETh?Vyxfi=M0+zQn>R&%G8X%Wa{E0SQ
z5IHd)L3t@uH%^AGR?%Ts@+*KC_Qx3`fcML_eB9l`-V1e%CPs}x7K<bl9r*?mJMd;F
ze5?0W9*!MgApA&I&VPe82#lLemuHwnAZs=ls7Uy%FI%foU{asln1gH3FAPDO`4YN(
za<l672&rsCp37u%76zz7lh<-H3;*TMQ^cu-NyO8YVhE?_GP^sjh^F^+dSgNnK{w=b
zW5UPosZpjfiUino@!Pi=cZ4Z7Zf>mfd8Q^G9;jUg@LhDKAMB0I{bMFylpuG(C<t%C
zK-hZbi|Pb4AH#|m1e3trS6~dmh&qF;U&dg8Afyom@(@PCG-HTt9Wbco5fIcvb7ALt
zyb*RQHwqPBTjjN&L?;F)y1Zg6$vr`BP_N1{D8$wRuR*aIzPT+fP%(wP)a9+CR5f|k
zq?R0OJGaPP<C(7Bnd4NpwVqE2h{DWMS+rh8m{3#P?%^po8Mj+jKh<Jg+HC0&rQTcK
zg<E=7o>W<)k}GMxeNh>Ufx1bK_d?bbQHDhf2BVpkdcVw2Ud1c8%nTcT)ryEFLKLJ4
zKt0q!o9>!56am=ppdC#BsFbz#1Ch}XtAdcyrKu}oT{KSlxn|Nc&iaYIWu<{FU4sxo
zzm%>oWr$cpygS+~I{*d+q!YOpsEm-vRDPS9Ag;{NP_MCCxW>B?=mUMZRV(XAHKxwo
z;d<rBOP6y<gN<j{ehT22z7~nRSo8IwJIq6!!05__z5$N98;>jYSsr)Zw|tQA1)M=P
zr@^45;+4)im(%+9e7Yq>*2!2bpLP%$MnEXB!F!~8Fj1M?Fp4Yag3HqQD^F4eOvhUB
zETd(?D!>7K5{N4NtbIoWR>7`m3GkoLPJoo&XNC@oKsFOfJ1u2;7+F6Lz0=apXvTLO
zk!McHEeGY48Z3KrR~*E?_lm(!4TCwwHK7b)o_@q(Qi*2c)>cbBAF0a`!rb|ynHZMa
zZ;vd{&IPu2u(T^iy7Uzg2=Miz6eRLDb>}SGkPaV1CelEix|2`~E|&`sAp~?r7%x7>
zF5RR1cbz~;K&4VTvV4a<#d;RXLy$40rK$<X?=JO88Ulmpnsb5rX1#Y)TaL9?hCElb
znVul=TU=Bd4ewfYJsGnl&g;(Zb)*DEyNb<Bq(53mncG*p0KKt{UW|Ds4P*eUe?7Na
zhOYu>PQ+7e^|+_fD>m?A&N_YXzAwu<sG!&N^KcCAB<OXhd-7*Td^M#*<+hUEAsS*n
zn^WKW<bOfR?$I+Kf4^IOY_+LCg|JCAuwhVGcRN9IsbOM%<N$gwxjg*|T3$(0dB=hQ
z;+e+LbXv_^{T>Xn3I~04^Sq&J3CZ&QN9CvI_<vXa|CNZ%&hkHZhDpu;a~pBNz~R)&
zR6pGZu|sT!odX)q4!VZL@xw#KkSXkmUw_zsPEY(P^TcxvNDumd#HXevro8V7C#!__
zVcPqKmb(XQY788Vh0ijrh)Fw%Naa-tQ^Xa*%ncDp;=>B7#z-a7e91{oyo9HWX(eMv
zDahf|1tnwI)Ju^FqR30dC4*DW`d}#C3PF@-QKf@D$z$aP<x75J$|0A6x#aju8jOsW
z7so1=83zNXkt~0P*+{_d7jS}5XFX)ua0bHQ`h!dOAW|iYXA4ReHTr71(z(S(-1P?S
zc_xBaD=ZXeEyfvplGttk0gQ7(mvm2o9SaCggc9{HZ-XMQZsRUFa7KS2Y1O<ymaV`O
z^0o6P!TULuMMbj9pawzp$W}rHT2#bEs9`QyMDf36SU??oQJzLneOZj_P39%@!deA6
zR@Qjn_X{9g!5o<%=qKLEs~Vs&T|}^t$xnkk0kQXRi{n>0<~N2=yf#5L<S|acP@Dx8
z@6o8Ti2+rHS^@P%))6s~gyPkKrW1B4V?&kKP%~y!(2odlCG-R&3@C{nw6<zg1R6X4
zswEGu@CMQ}G)_#aWv+h7X;d~h=~=BT5SCw@A|$OXM*)1S*aDQamn}ns2EsYo-9YIG
z1|g&gs>m6*Ln#kI)TU5{?|<aSumTd7p>)=R7`i}-feA5lrX*+3(C!QwK@uSlk{gO9
zBO^;VYHj78fbxGUwD5laga@I^>(hY9*z|t6@whvXIOIy7PC7amU^W9F@Nc-ww%9nE
z-b9^R>ECOo8jRL+BqWb~vvUT8_jT=rilusZg{q-CSMfJmveV`RMz}|}_ElBa=^Z&r
zPJdl|b+Vw9z8}A5n(s*S)=Zmz>VA9PM}eo^mLZ#Dx-j9?qGfJXmPt4|bYkf4Y)zw8
z#jMzRcYk|&yF2|zXq6|W3vYSB3a7QL!HTS<)cdXOUT5XTsQmuaQhV8!q}1HB>R>s}
zqg5qz5=a**-sF4U-?7iNHS5GZ!X?~siAw_s-Z^hF7pkA`v0_+@XD%eIah4c{WyqrD
z5Xt)HZPTQCpW&z)p@#3;;Qu$^O`m1*;fmHh;(?xYVNHMugW;XN4l^PDRxmdLJ^t2>
zfDeF<hIFTp`1}Jvv;l(kUE#Ks&%pZK`(9g9!q(OS=hE}2>V9=|`gQWuiZ!GGT1ktv
z(M-zPB!&67**59%t@wO3JK^Z`<LL6C2+FlO9X}m?vL3n&-Etb$#s8=WP5I@-YHCfh
zcF~5Wc9O*RyusshgPCtaZxR;nVs`Io5=g50m}%T<FnIVZeG=FNE6oZXMDhIB+HiBV
zX%t^x!+vPr%gc^iq~(0ywO?N6X~rxq!f6}7Ik%*HWAR$tHsV!E3u-0O3Nx174!RGt
zUF4E9)M~r~z5^u3&={!$zPW4-oC+p#(jqTA2=)v}nLWcBI99zBmngB2p8tvoN`K(2
z(9Ey1nFDL(&zk-)yv5DnGw-zdgoLtOj>@<6`=jM#x`*T0_rtsWz4sz_*G^BL_aWWM
zZ%efC@l{vbVjs0V$+YvqJ(o6f(|I2}+n~-H-?ACQGUO2F6kQKYo_>iF_CJ^mX+7X~
zGqkOnkj>*SeyX-a(qbWb#kjfZLm_zwMqqgPiYLsW1}luF0zn;Fzrgextp7+3;m&}<
z>K1EO)UOGwG^99@PpUSrjh~PZ(7+ggAk2n7+!>hGbG|+^GA0at@#5I*ozKKFIDt`N
z9PUqj**oIBg9Q;gk+C)Y!SD#U&RMQL50MkA#5ccD*b9oKm~~SUX9XbZ8^H_fc79a{
zM>PRa_`*>NR9z9lL{u^nmrktgTF(!C0gQwsNmAk?G7z4)E$^yfpBcsHa2f&^z5){t
z-3KNN%rA0mfUD}-nP|LSZqT*UsDBTgD%|^30C@A8;ZovM2<y;be?8a0^I>)fSH`j6
z9rv<GYA+U!@~%g*W$+{*+6|c4gTN`OG&{~;ogz1zL2PiLlWKS6Ii~Vp*1MLm0wo$u
zcx|?Akdb$3UqOq_Z6Vy-(oTHm%@33bW%2paACoc?=83uBr(qw3I@Hx|PQs^$532+V
z>07Erp$SIc*0B!KlE&j_6DzV`ttZZ6k8^Q${>87UC6JM#1g_sk%YQbefZY+om8SlT
z1oGo}y0?c)x@Vj4z63K>Z;2f4i9RBQx3&pCC+_gvI0}YB%8K(Rkmn0D5>^THP_xR7
zJEbiq3fx}iO{&`mR@l8Ab^sqRTFg~GJE!bh*w277uzI{_B58^aUL-ifk;c43{CysA
z_VrLZcv861V-G-N;3NgCd|ov8>Me>UV1{>{&fk<Uuv49=1zC`&tvR_cYcO8*UK*JB
z2f*rDSMdHU;B<BlDgu@sfPTrxN4?vF)>vp7nm$@@x6qNAHNQN7<&h7m4(|B5PEeuv
zyXvNU`uCeHmX2@WJ0J+w7x4Se-^qXUY{q{<CfM2kYwP$Q);gCdQjTl%2*cN}D7<N@
zdFGE#4DfSBAq5HSt<0H`&~T41E0GKZsiz^;x?tE45~Q&Gxn$#-7Oo_pA7apO!GLhQ
z_&CGhA!d<ah~XfJ0v`zSh<2n1rre?6z0a}Yi?4`2JU@~vOE6?K@!?tsuG8J&iPlE@
znG@tN0b<GgAy7#1@$&8B>#dBZx4ZqNRdVxVNqWToczVSAR_UM4%O5a~{O;jmBc!XZ
zmp|32v)2Y~oVZAz2>x_N3^9JoP(g~R3ai}a`mfIZlmm^!ZL&hhzfbiRss~lSGn7r9
z){EI?L&Egxwm=GvpiPa8gdt(uauC9Lf_Ts5F{cu&{(j)NKhINNzmp2c+v3R|hhfQK
z<0M)zItW3=4$1c+1bd_p*DEA$9*c0uiasYWUT(*jCiM0fA@G~u`(Dux-#6%?9yZLQ
zVuZ+D-463;7TsPX*uLm6ShrKHLAU5v3IPU&!36?Ckc$b@g@r*NRv}}auiVnr<NAl>
z8=hbd2DpdUl+m~t+!sv5k3F}Dv_1Wre01k!Id;gd(^y_8GQoXg5b}C*$F0v+zRF;e
zp8Gu;f=E#^j}EvGB*N@a%2j7BgZ4X4L~X?I+;|pDV<Evr4DMpYka6@bm+2|OQ<KE~
z@M2VdCRC~L(5($p!rZF;FJL|tWRdn2U9QB7lzIH6xkETPnbwLYoXlm$-Ey7=TFkT7
zEwkAQD5SVyDK?0L{?z??N?}|Q8WKy2$zKjrOy`Z%=rIk|RF<MNjd@-A^^EqR3I<+|
zbhD+wic}UCv9Wv|v;7W|S$Gjd;44Ml?_ZHMm(o}aM%Grm+E}T=am4slQSHF7CoxF%
zpI|kKjV1_UZ^;;02C<YTm<5uYU^2t;0`r^3Gn(r#n($cy^G}49W=+Bi80ds+Z|cJQ
zgkddc;_1dpTDSLWYn#7<u9uZpQj4nW--EM1e1e#BffNWd#f74n-KcX#v6f<k!RBU!
zJvf?+=DVzRum$-HdfBL7w{l7cHd@Zg$9-hHLW-_{eSbkHMqaS_{!W9!A(CD2{!Mu@
zPFhNb|6Wmw5}W6Qp7u6Y?$4_uf5xLO6=ru{q#us2ryC{`ro5Q6cqhp`Mm0-ao#e;@
zHm7LK$+n=${eX<tm}I`FX}I30Lf;wAqqM??`tWmGe`ZF<`vOh;8qfdFPnqLirgZlI
zdAwfKIRB3+{X45$5Z~mN>FeJSK&sfEaD5i9U(VoG=LqCY@#c*tLaOm5<WJ4lFR{Xu
z4e7Zj!CrO3xG}E7+dNST<sf9+_&@YOZg+tiNfbrQ$@o>1zB=I>7z5h)Rl--6LQF$*
z#zUmy>aA>9hpZ`C?=%F#x%~PBqI+-dRHUH!W!ouT+|0G(W3p>gGqUNeY1yp~-Ug}d
z&zqx%m(B6)%k}lJyq?{Y#oLRQiv#rWpZtRENpH_Pe<nRbNWvQ&kS3r*Nn~98VqkK9
zbwWRGqcqR8Svia7g0YLLBI=|hHj;&Wu*VjC-0Q{3y2`0@=X%Pd<Mg78Y&u)k2&r}k
z?qK;~2hz4o{F<K0x2un)jOKGwu2fTu;^(L1girSycW1{Fyzr*dubE<EB|>^a#XevW
z7+L0x?TWE-py(exO}~Ix+z&aP-j7iPsR{$>FKIohRi;jI0!1OEPF3}e->#FF1-j;a
zkU{yQ^{leUxp^h%#eH5yv&tX*D{E5WqI0fy!^AnX6)eojymdR3%_L{Co(}VJXKOwH
za1)LmXuA}w$417yc~Yqs;!r~i7R*cnc8Dq;4b27U_8N_VlEurwf<hhjzxeb;&Wml&
zu%3){$@iAOO=dkSK4IyMESp!yT2_@G6LnIn&#ko<D!E*0XNnU=9rh$LB>?bT6_Y0|
z&jCPqO#lX9Wf=_r_VP#x@6Ju|;LE6IZJ{mJwWwo-*|$}mwf-px4M%)1AR-Wl6re$Y
z5#uK{nsC+gNq|qu_>~$K7*IU3G4@Ug0IYzA0Z<xY30>6IGW!g%QwZ|$=Fv&rjDJnj
zTVx)t0x7BHvo`(|19yOj@KY6jk3Pf*>WdIfuA6!(Xe&!W^VjtHin_ISlpb#q5LKr*
z^h0kz9&$7{vA5z}8RRJIdoQ-$jz<fUf>cmI*8Z3VdB4<=lIDHG1~kMFp$(z@Met4G
z+Lb(^{6=~VQ4@jnMPBz0*|8TU<3=P*-4RxL<wfiG596pge4Wk|a5W=-%SEijI;_lS
zoj8Tu2!>V@ckn{_cZNM_da(ZRIrxU4KK!F>c<2G$3L#|Vtt(-Z&fDv1V|JiLyqxQ)
z(|Wm=e%l@`-wZharCCcEg&<y)9_ckzBJdoCJ@XBa%6ir0ivo&7LQGr+6~Qhlg5ma;
zJiLpzk0g+8`xCq&&xNnW#&GNCx3I(dJc~W;C;X*m$1(=xM{kdasPsMx(y~li^q4!2
zoAEw}94T{ki{v?hfI~>OZ3|BC2H&`CR5gu|YkBZS1d5supC?8hJTeFjJ|V>5a9W<9
zP@;U=9C!bV%wS{uLW6n=BYtg1`kPMZ@|jaSD%45#j=9V;xZoG(=7SU08FS7~=S4}i
z%U!OGev!8k>-}JYN;V}Zl6oMQtSd^$CM3`Yu;sFTOcbJLsD%~E{7Y5rwuB&v@=Rcq
z2;{{XR|EQ_!*6IMnZol?lr{fJGpAqj2$B?gk|=);m3<XMPkN`4xfl}_nX~w~yAW{e
zJoUhC4~ygtO!ld+fE^2zO{&Ow8t`SfxAJiVv~uh;VGwdZ3846wQWXOW&u@62>fn`O
z>`4f{4>t4XZ3YOdc*k2I2mIX1)2-KA{q8#W!Oyid(x8oC($2~pw`gTH#en)}bui1i
z-N4iX(CBvF?Zx`Axc?5RDDqtFf|dYR5uboGnCV{@@Q$+0-a*0=oF1A@P4@W5he+r0
zg{Ufoz!+uaZCyjaP>2TuhRlB!E5f+>wV+?r+KA0ZStr<(i?r5I6&*3iB?a@t&$?u9
zRJvOOdv;ShL0fcJ&<(?)2ITlLEhC{As4&5Pg1vhOwC~p-B5^L&q@KKitC4T2j2C)Y
zhFhMqCPi(oTFn}Hq*h(d$YfHnXZ_(xD2LNBuTnMH90G&(I_}Fg9c^>6I0Eq*B*m~(
zM#I;i=)WVy(c^zDzMGI+`()p95(5j6`sL7qnr1*gwNaU|;JB`a-I#MXIp0K2rB?W-
z85klfwS6R@-FGmz7I?0`tTS42b2rXU*@0M4*_Eqj_H1)KKGx8*3G6;mF5hp07W#)7
zk|2QOGqVI7Wy1#q;aOEZ9OH-9z>*Sj@9%1@5R>~M{uAvLdMe(DE_*z#L&VU6%CX?K
zcJgP);E$L=c0U0unXa(I!XP-;%~)VRHg`&{N>Zyn0Cc_|dc^}1oEu*3$~9^nCOW+n
z^Je6D#_DtnVY;XREGPDWTg;XMw1pI-UCwFC_=6DHlVCCS^s9kw7pvuyJT|G$3LSix
z6{UzV3@k6v+hNK&M@W#Oy6c%9wiEntg$K|?>z%@|8=I-~h(QN-{T;)ZexVp{eMx%k
zf{bR!CA=SaMD)>EH0SU4qiHqQ3*cC1FR`OtggL5HQkzd|XY?;OU%so|7_JXz&UC-h
z7(So5G3SIC84VN9G0r>m!ec#~ip^<Vnk{@5<HB#_kv8}#W4;W`2lB{IY=(z8ABqEz
zLw0yEUY;rZg3SHZ-fg}nm$83!Y{q{$*%|16;-3G<sIvcPsvE37!}k{or=?%`Zu%U9
zH?gz9iXYAK99VE0RdvvnWU%V*Wqw}IU>pb6#%gxw`GJ!`oGwwJDci3inS6WU{Acix
z*WPqjMPbZQ=4AzMY)D*+QpiNgF|pZe2D#H6rLXmL*GN;U@diRs@|#^W+|RcsA#Vo`
z@s>P>W#d?Knf?9w@{hZR8&1AX_U^vch!}&#$JgD#$Jg6Y1{c08oI~KGA@pRmCr9uF
z@D?;&qQCv@QR>7fvzQSmQjiH6f8_|GKdNwu6-V6WA}e$Ej?ln4GZb!}cu&hh@+!lp
zsw>l7G^J<o5+5thvflpSj~OgC_{fWx-306&AWub=ht$Ds<Dls*F0gN}%`?Y5(IfnT
z3Hu3*%fz9cF9f&jY;27<hnq^o%%#(Y=iH?^Lp$)M$wVTU#9Su}GCm&zJJ<U>gw}&?
zwOpNk!(h8!d7Qk@17&<hJ*CPq7VreEFU@Va2>JnIlGqJ$@28wK0~_px<fWnb$aZ9F
zG}?_qN_CQv&Lq@ikt1rmiNWADl4!SSjbZ9(K)r~Q0BO+bu%+as=klFT8hi7#Z9e_L
zhwVuEZT?ic_0|jIQ8a4QPnub+9o-hO@nnAQ!FFJ!cLe_MY*_T|2vdZo&1f}f{i^{e
z3yIUpQTbbQA;{Ta>Wu}HV03^NqCxmC??7{02ro8`M_TgNVDaQXOaT_2fO2e5-NZL6
zC`RWHx}ytf#7r^X*T{&4q;o~2dsqA@CMxuyIzfx(^Z**%0OOw-8ucJk{JnxFAp*L)
ztIzDa`+k{%t@;YJ{E*do&coa851Fx4?3%L{tE`IaaaZUpyhxw4I}!f$@*{TzfBuAD
zh$_zKdjntczHVoF<SH5}&L^0l8K=5im<S}T22<CgYM7SSb!@Z)thh9pGe>*||0L>}
zKw8C_0CqnWAXZB!tx_c(;=_{55fIaK6;oGJUBE>ZNkzH%!tu)oNM9?P+OUuI)gLF4
zGcEr$&;6n?G5wBX<cs*|9INO8{P41kA4sTX5zc-Z9hP2ko~{N=G7)~tCe?h$0_Z=~
zqssQdP%j6>&{zOi8OKszg=WX70L0j+0KCMTbD>@Eh1S27_1nq{VU~h5*mb}AR<i2K
z)*1~Pdx=1KAKW3VjxACKD@5bX+QUa2M;#^qBQo$5y^~|&Uvk5^c0<Q_n$uVru!Oaq
z+sGtkZcWGk+xidAdRuwT5M{#^6))ftuU4|9DCX`i&BDBs9spmI+dE-T#Kv1q(!hz8
z4`##5xTR7jNnGmM$8U%C`h!*Kb8!I@jOm{DI$EvaVtG#5n$v*aVn8i|r7di?>dvm_
zGz`M%TlxJXGPb?ukJ&W2JfrYk0LfxAJ^v~#O#ix2Wn=xHQxF$_;F0U1C|$>DC*pud
z#!mza2Rl~5YXc0_g~<i<<YDocQUy|qF0d${mtOU;IAaZa)HLzPW<Kj|wq0V}nlU#g
z2c!3^3B}cr(PBSwTZ#FrmIDX>H)kt<)!~642g#lpE@+^4{n>UX9K3$?+DK5`&Xbpe
z^;2y>778Z=r#oAgY`D0|)#dAIu>5qvMOU_{{wK)zxqh^L{uaMpFT3k#B6;*b*+i16
z#=;MIWGfaNgb+qhGTelsd`JX+`?FTi5qjEa=?!gd_5P}(d9;$|_w3}VE)`wUmpyqc
z5@Xb(=%tw}q|T|IChNA|wZEIfxJpOZs#sjtT~*o|h%{`(&KO6_RTON}!%P&(G@WPa
z=h2{1dY8@94{Q!OC+8VU)1!Q@;yb3!s4FujWNKT=D#{@O34!z^s=QDQ2gtF5c7U;z
z&TIA=lzL_2t>a&ZR#i<ZQ&AXP5e%Mw6~W(46Fg<4FSt!p7X^AmV2Vi&$@n(4%Vg>t
z=)2Xms<!}W|9$lkhx6;?Mw{G<QQxU+>9ng83?o2jCt9{coV{eLiHGq8?b$K#gAOx1
zs<GS4;Dk$!6)@xz)j)nR?W3$(Cp?kaFnuhv_Ipp@q`EtTaLtR(%IuOEX9)AJS@PAq
zW|gqm-x8xd#*v`(Mk;UslP5Uf6QFd*QIvQPk*^G{8LJ(I+lDNU%CQBdjKJ6vz5q@X
z1=6nU<`y5D8$bz$A=EE{*X5b_Rhw=L8=F#pIJ*Vwv=QuI&af4P0><^ko_PzW^oMXE
zzuAFhN8%06T{c*;m~am;m>s5n(1H<KH&kbbIp0L<pGj*!gzLA<`%z^u;GPA>g#(of
zD%zeAY&7{O|DJxs9MWETgR}2rh%mg~?n)a41uF7EXY}F;JA?R0vo~j4oc>-BhX@bK
z!K<PWie$*<cnaMJiPiUAS?7gec9nKfNyI$hziy*OvpAekH$`>x=wh0#_0g0|ge%4w
zL<C$%2@k8kte`#`<z6>#{6n5>q1%|bNUO&jvOUZlr)vkXF}wn5R(Ga-erZ?kH|tRG
z#^wDU2?98M-SszLNN^y_ciz;V+1~)o;$4kXQeA|mwdr?C-j<(TvVcbO-UE}RD3k;e
zPfihzJW~-rp1V)Kll&sa>BXbD%rw{WiInU_ID|b`48}=`$CJ4Nl!$_rs)RnLuET)S
z33-_CgLyLooM{;k6BDc<ZNQn$k6t|#LA2gY3*fUVWBF%X`FE^Y<GEg;NtTUO`#96s
z#s{#Se!SpjQ%KmlfwaExC>vK=58U*u&<j}G1R|(%h%kVHD>|d2>$*up5gNVKlD)yS
zQ1I|pfyv@Qp$L%1d&%Jh?eB`TIg@Fw+9`)iA`i9`cyYAaw+g-kwby)jZod@bZ5Zw_
z6z$#_WK)|Q**_q?=Risrw3(TVr%wkJhy|YgnCu})*(5Np2!aqJye9%~O>61z7UbCU
z!sggPr!_J<VeSvHSk^E$;Fk)aj%TSt%p#!yW<t6}AfqD}Hr$U|#q0*;M`>*4Fhse4
z(z(&D8J9#q%Pw&@4F+N2Na>+^_3MWVd4SZcW<c*H&VDg~!b^VwPSRL_GKjWxA86|o
zIht%N2Uvu|l-g?OM%#RT&If{yPlxF~yk9+9l96pH*}>i){H+^M{Q(Ryh^Ws-SKH(e
z9Q1n85QvwReKAg_x(_=l07$^sTFA#%H#)#Q^q&KvM|U#dxU_x*_jqX6e|+CTwzmvm
zK#E3B5Dp>oCQyi<Mo1o{q+I5O<orPHs&g-GJ?T>{f2Ht{@I;}6gw*>~$sQ4_22<k8
zL(bJ0&TcLB_vOHdKwYm^mjlsEvFEllX72sMoR?!7G@ikQA3K6#_@PxI7$e8Bwp&@*
zbR})|D<Llg;Yz3(A+3Yp$Lq%xx>x=N;&Fz&ZaG?M+eFmc<f<37xJz$Kkim57t|0gu
zz}2cU{#V67tb_zQT*s@koNB!`7BF)oqf$1<q&(fF)taupvS_$zE1tWA78#tKDRV_`
zTtPo*IwuOgwV1p>W@JQkx)qT?y3o~3>we&~d`}>5osK*)vB^T$!;aIgai^hX&|mjw
zoE~R{lmngDLMz3djxe=wIBYU&jgGS+i4dtBU$3A0G=vF96|Q-AZkK7l8)z{~#SA-<
z3P-Bm@i<*(Nm^b&;JxwUQJxN&M6?5}wVbOjUHQg*a|X!cq_qCVsi+QtJBkg|IN*bx
zx9XJH$o5+Yzf&dOz5T1OT_Dj_Q^g>v?qOP6FRSCDRN+O2S2m7ya7`smKq(mJwNA_0
zetsx<e!X$m{K-9pMM&>fKDpRW;KlOij&5;N?&_aC^Z%`1|Cde4$ngK;#!*AVc7FuX
zJ6o?;46DZD{mRr=NB?*w40+fnlI1-f7Og>8V<S#nAOsg&!23^^>AhuIJa>8s@l8OQ
zm$I6dvy)1yC_G{mZ<jll*ZroPqA;Wc^YW|{;(tf-w5j~=p^dwUZkg|gO}R+YPHRm>
zA{2Aw^sUd^;q}vyX>{Z&^9h#Ll%JveS<Q8RTFtgwx|(-$<F!$akFC9HqvP}8@aQw-
z`Ro1|{<;W#Vknz;yY+l)CQV=iT9*21ac+}qc1LE_ybrcE>}mPijX?S%<*59@JJH>v
zy(48&NLHA?jwrJxOE?^YG9R)&(5Nf#PqQ4foRK}b>3avY3k#aGTbA%(0&$oLJEcIL
zsYQArRd+Je*7`U}ZiSz<m-^+s9zJV#)z3}=!vyvz(4y&E5LKu6-TLb}BCLqn`-$)S
zf$uZ1{lg$r_wCk*bv;>&&l?Exhz{!S6hx2<#n*rS8vyvR1X^lloubl<r9ujsZ_7OG
zfU%RI`D&l3md5W=Z0?=(u0P$*)s8f3$DgOY=d$Z*nr*ID<#-NaG<MCpOxFpuIyW>%
z4~ypwuRy}o`<tQ*iimgT>NC(|R$Y=WEge$|%S}I&iE!6YN&F+^KaVHEZy7QlA#b-$
zU_ywN^!4*Dx{YBYk;f*#6Ieot34%NOu*$MumqfB1>PI8M`IrKl`@f&Rs4}M=#g)&!
zv#pog>rDsr%tW<e3Gr@BBZ(z%vv}+U0!t_e`zVoc960wIFM<)vQJKvk7u3VPllgH4
zXI_jvRxTaTw>FkdwWjJer+g;vap}AGTHL_P5yJ<slSlyG-?Xn~$pBX-gA@_wTeews
z%=jt3#m+`LNdGU&-l<EFFbmgA+qP|1+O}<*m9}l$wr$%sD{XtS&&=KFi=JN)u~x*2
z*x!Efe9MB6u5>_Hiw{_$cWtM<oO$=$qa8jjWysM}#cnd(tOFGWMPNl$ij=nR_<OV}
z3-?n7BKz`s`yYTv^@mR?sZj`F<Yesi$^7jTO+#A&tg|=m!}hHWjArN(Tj++L4#`~9
z*6geJ9Y4|`2pIdWJ(6Uk_+Oi?eXg%rgfUBaK*tls5u#sL&_{HFLVG;U?r?~{%jv!q
zctxRl45q0W4H5>8!8;}^$NCVra~i6#8!*1((BMN=I-aEHwv+?64O*&NX%@dISrFmE
zKwf~^$t9lzu8f1vX(65n-u9!?nNOuA95ntM*TCw*$R71kJAwpG!~*N>hXayL<k|_*
z%Qoku=bYf>dUb2?blSTMT81bq60v{@*N!fQcr6q=M~w-4xO25$yR^epHv3I7azbQf
zzek#eUS?9OOC*%JE@B#|GYVs1LkWb=F;?E%!rf}^Rr8pjBq2?M-=d6qNs|JTbB!>V
zK^ok)`O!cd{t>(OM`zl)4`tK^&0pzR$ZeBC-+Kd8Nj1@S@%ak{7ab%H6T=rRCJJ-c
zmJ7*wDLrSvKKD?^?4Zl9B?`mvfH!im#3h=j*IrK5*{^-!6$leNXc(ssiYdt_&|Z}o
zZL*?FC{`6t)YPQT9eWVGmz&0Huc3Z(ak~JhYp^GnP_gcdQ^e}X78JN+lH5wnFK>*g
zs0ONUh}5fk%!Y5b$x_@i4+24A@3uXgGg_Avl*ue_iIP+8*BgzfP}j9$QX^Irm}<_5
zG$P2WX_`9%8nEClhX+PnFY)<>v6HnQUL%X043kt!+=Ua)GGb;R45(f7(K7zW7m-&8
zcR{#DYM%pHDyjz9Vux`96Xe`({rvqSKCcGygjU-cg$^{nzYDdb^fuG6(-x`{@Xwex
zGNDgy!=TzYC4Q~qqUD&F43Ie{s&l6gHX@KiQcLQ*LNR$6=oEDEOZKEPpb*lC@rC{)
zM3E`L8Na7^iM%RC0XEM`VLj-b9D!93n6=UDiDL?$Cts8$I9`$VY7zOcIbj6uzEP|p
zMZJzm1<ejr6714S^4TI&<$2rX5_O=M=mEd|ZQ1>F*gY_wCk`feIZe@pE_ll8z^ifv
zEI!c0{`_Jp<3w0K@Y{=&Z#KPY?x$Egdv<<4P_Z>iOrfikl5<A@Gsr%8t4wP2)bx!(
z>h$*5$j8F0aU^ugYpcP)fkqDM$8w{dCXtA}N&~hlYsE*$lVCe1Yb7P$F@hUOpa`f%
z)UCBjc9_I!#2gVF{KvFG?*smhai3DRD!c5~*bCFfl@IsqY7KyMG{4uRTdTBHU}cKZ
z&!Xw{+I+1-sFz~bPsPmA%{FR4hiw1K^1AZqi$!o&CJi(nXvuwbZiOR=A6>OGsZHZ0
z`s(o(<6g)aRPYpNr0e(JL^#O@%RhFfYSFO1$A&>B#>qq77^ssvYJOUH^b4F1{cRIo
z)`5g%n!*gv&OuVPNHMr7K*47|oBgva)a8u{$;kPS#;YdKoGIC?n07MFjI61p!wkoU
zDes{++a|Dugp%sQ@@U)y$aFEFRIj4Rn##_9$Tya;zcTa4`qjOZt~0?BD$y=F7<XYi
zFndI)GvQPeR(Q(1-<#?7?g!cxms2yQCxqq~D#mLYOB}lI#HSx3SlnfZORj?EDypYh
z#B|P%-SE%HOe#FvmNyK(Cq~@1OFmcC`I7R-5b59Z8yPC0ZN=q>%L&Rmd$_l^oCHaE
z0{7GtoU7WXOz~=(*@#H}-B};vmUnxO!~Mbkoc_W_TqRrPEyn(8T|r2&;MUy=TJFNR
zBm~3mT#(_9Z0?t}CCRYdoXyYB-S?u=GHvuZ!S{TPz?Hn~{rpIR|BGMY1lAHPx4Ws<
z^A*zDPS1h+IEQM7Sp910K+2AO18RUQ#DLj2Pb=Rz&p!@-)?TfTcj#VfTCTLp@oHg#
z+|(kvURCJlQ)gFf;Vf$2$J?-i+WzzHWJ8SA6kT(Xcf4u#xW#9oo3<2smgx&D0Qmmj
za>h(7tpB&}$Hc(E^1p5ohc&bkH-Dp;yL$U#W_=ac4S4z`a|;^g?M_O%P>qLVvX<gm
zq<7a1%rL###8MHYV(USI`;;->UNyVzrx<3FFlO|0J!!h!ZxQ<07Dc4V>rU~E>5z1Y
z|A;bY8e2J%#S0_|NoMrA-8mGvb9KMj34@O0^z?YIe)w`xl=nU@-~X*Y*p(s^-Wyul
z@pN`&^z?ilOh4Xt+L80g_4?eSn&s2bA)p~C*B3~dhSAk6!YKpON)n`Q#i57fP-del
z=POEmtSmw*z3D7hRk&7eL~OHOP3b5KmnBO>G<K!!2O}2JP>Y9`k)_PH)TpK;Az{V+
zb`jrk*4U}({O#^)&*|Q+I^V8^U#;BCd?O+akPksY{+`aQ>$|Af{L?E374Gn@@B3ZH
zx1X4ryRqBpv0~Hm{Enwf4vB79Bz4H26(99WLBi0bB4Ky4hE*)R@f54`YW@`=a_rEa
zMNe`c$+lHCOU}2l(RZ~L+_$eYuKRZ$Z`@^I{XLtkug_N^+iP2q75{<Ct6bf?sMuU3
z(q|7wA$D)*TBSRXA_y0%xhZHAQmUwu&}LWIAFg_Au26e+uL48pZfL1+wW7;-s@b_d
z&3tc8S0s1SJ$QOr?Oe3Aw_9u!cQbD`b=FvNjau;9^-(q_>e;Pwp&vlRliYSGVxl8i
zlvERz+Di;WTq1A^#-Rm(0f53ViViBn=uPaPfdF&rMhilWuWbuHb`&)Fo=n4aJzck0
z{jnJ$MB?j<5b}a33Kh&~dwlijbXV)PQiVfc7Y3g}Z@rpg#usS1odQ9cQc0d1-kb4P
z6k>ABrQdanU?Sd&S~xWjm+YuTPYIyJx}&)3DJY7F2`Sowep`Zuqy!-n8fcYZ?W5Ps
zt{t_S74FY8(d$41AO@l^6h~4U=an$0jt__=%HyoQ=PY-G%sIr*YXC?NCek%=BbNV{
z88iVzSx2qfJ{l$=H>wQwD6iMP2L5y|zE(_vFX2vX83yH)Ca@rfv4}zqDZw#daRhWq
z)yLu_U~~?@`gU#5T|<a!7AW=p!Iqg#&wgvCi#=|$xIE)559mpd5knfUP(pyhI6(UX
zsyrpkD=@;?6?!(PK$7}{PE#d<qyQsFfdOKBh5rOPy<B$|$L+s^M1bG2lJxZ(D#d5h
z8z87Zd61%NW>+-fr~m_RCznG(nLeW6f`?cGgm#Y_R@ordhl1A@%<}XXX5_j+;9K@?
zQ8Nr9;#O-2Vz4|f^MFI2Ek*Cud6g|?fl&q`W8)8=GQEgw<vX;7i<Oo$^DFwuvCkH`
z=wnbnuuj&|`P?I$7RReA5C*1>K>2(|vI}7d7?$MN<X{MqfIDCaSmg()H~k>GH&#UW
zH;;o2=?30R;4kI8Bwkj`aGz~LdRlU7q>81BkUKW<$At717ixC5IpLp^pA-S=YuXP|
zr?zMOIX?_yNdpjZl~b0)EyTHeiW^f)>{GGr7L5_5=2hvQ+zd@ABeYV|=olX83T4)#
z^p-Pd)CqB>5=I_K&^a?h-+nDSzXuMixbc|Lh(X5TQGA&FUX_x`Rnnvwi2rDzHv1ze
z-@e-4cgFgK0Nh!0$V5usQ8jrZ>+WbTR2AdFDQWYl)UXc^=R^dkP%wWJIdZ~B;&L2#
z#r_XX94K5I12;@r5UdcqrvQeyxqK8iJSbkCG*!VSMRpJrz;v^cfQXtnq^%`lqPAFo
z;c-L+YNeihG6}q)y*xnv;bZG0a|c?YFeS%=0VSoyJpxb8G(1x=t6T$80ywajja<Wx
zqHBO*-NrA26Ja^<-H6dqETT#MsKN3cm`CNYz<T#3m3sL<X@*%84r>C;?9PAfGXm#M
z%0=ORN4bCUMP4k41Oc{qxo}{_MEHZZ!Y?6sgTVzG(z+1|zSHgJg<NX56kxBUj(W|M
z5+Bp2s&o5Qm{es%lA>oE&8j~S&!SJ`-=Z_-_Q&;r<lG+SM!n<YCzMVz{=NEhH^!7x
z39u72l~1z+&Hg$DPAY1(22^r3UC-p+=0-lZ_Et4~N|~&uU<t#S&Dq&+wTYinQAMf-
zV;fqn)LI{rm8R`AF*K7?%IH<u=HVY>j~5w}E-N=dRSx!0O73b;t|`qQc53F@|JdyH
zj`ul1Y~d~t2!0!fUdFTY=o_S8B0I%{<H0XY=hZQ5p`xe1jO7?F(uKwP8*vEp&|pe=
zlQeBm0|txawjz>pTCIjp_@{9s-Mc)x&ny1y5MRv<uhv)7D%p3igVkIt*Jsm)1TD_b
ztk@d3mKWR5d*UbRn{{IQFI-Vwxkm{_`judaT-ECbcPe)7lV&`N73dF2NrBHUZ8$aU
z2aEaeq-|ln0#+`4StChY;Hkx#Gr&30c$kev@pD{I9Pa7Zx&GYt^FDkSw`HwLiT}x7
z_3^jbReKvSwc^=*tTs^}mXycAk~~A(^EQTQP|eg}U8)zl8yL1ypTuy=T(0+Q+x6_+
zk~Eb$=o}b_M>#HsM|lKUf}~n^D8v8a1q6W6Tpb<d{2-_b1aCU{L0u8DI5Fh$I^pa0
z_<Rb|XZIe5xFQkclE}|?YeJ57Nf707FmL7hdF_PUg6zq}P#ttdes#TeO>DS~-S9nZ
z93P!b5yKAryft38y{?TVmELav4FR$xd#YoG$ptO!;xKQp5BY!|zsFksZ_5qV|B~Ni
zWcc6B_uIclzh8IIf3!w!MP}51mqUzE*>k+h7?G0&7w#x=%B*;bSmDiSX+E!1bwQ#>
z@gr^)3wkmT{rbt_?^kf?L<_-0SF@A#^~zR7k{I@B=L8v~d}LDAd__`?_|JGpYH(kZ
z-foY{vJ08T2#kDvSiZCA+FIXPIy-Qo8uJNOOqk@7v*Bfm&Th`$$r{@lF?N>+-l>jG
z-rn}kp0=*O+#C8`V_?(8%uR)h>nImMTM@DF#yL6U6bO+fqj19%hO-1~I0s{VDfO65
zr<Sx9*hD(AwfVg;iIdpI5|4>gl^v(QV$((nE7NtJGH!R#d@CGAD~Kh=m(>E1LV!%<
z;U<4t0Ocwx-cZ0_lw&f7@8v2^p}rY#K7))F&&#XoHIf#O{o0#o)Ykl;$+BrSmy%8D
z(^?6u3vFK#s0{bdvR7k!^;^s$9d?u>v3%VxhMB+DqCq)N`Wb4Wt*b`fsxHAEOf|lu
zuM}9bPcnry$0Ac~bVHIL=n@Ez0=QyS9y|!7-;>lDKZVq-RF)5G#vvQGC8hTyftb5A
z!uACxnR54)XWZ<llna!JP|A1TK2Wcpz5nugdQp&la5IdAGjpyrxA#@?Z?G0$%{0aO
z-tK3Aq{!9YXman0p*=7#&3dN`ybd6a<Fc?LN*9n%8F9^?%%A);lCCfk{6~O9W<+r3
zO7I{1#{z>O>4_sg|1bTCBd%uFzljG<-<%CEjS)U%43Gr~{7Qchpd%i5S7)yQ9$tlK
zxc{y|m(_iH&rKHv^-{E5rp7o<L7b$dU3*4tH<G%hC!V^0FViutQb=7vC@HdT1MoP~
z{ENLRPv0#nSw7`AvZ!vcXi(fa5g0=jWo-b;BW7?6<9{C2&6*1zrvz9YCG$u5EQV^E
z-r{tN)*?FW4AyWiFl-%;(QCeX1I1o2PBib3*S%)|ZfuG&DfsIo>DWwAUrnP;CCCbq
z5*l_wuzqUKYAYz-#w1^i3VNx&zP^<>NyHp^^D(IyI_BdrS_tg5hv<mq3TA~#NDB;%
z=E`7B%WeS@;UBi3qn8jETixKrARZDS5F$5`;3aVYIFT@cpmo?D57CgOQJ%&g7g)4^
zPR>Y^6z&{G4duLGx3JOLkDk#p-Y#4t>hsYt(4cl_7H52)@J<YJ+SpMN&|noI4bEx|
zYFgs1QBUT^ZLXS4Ot63#GmJK(*m}ACsV5gwNYA*0f&LyJ&VwPw-k-bVRLs`D>GheD
zQ6CYLQRsuOlWi%u@B`sb#YRwi-n9ebhD}_;%SgKRwS9heXw1&1VJ&|JJvp#A72@O;
zuf94P=u&&OK)2i;o()>r&B0^194QUUB2#{Vz@{A7A4kdJcTXzb!WczFXgjnz);fwm
z0CSlLuSz^Ve!ie*CoeDcbeyCP|LZsBzb#Z5{(nS4e;tX~BL8tD9`NRf*mmB8B8bz<
zNQrCaImG*^&JXvrfj@IJ&(aEn^=7Y+@_5(dYL>l^|29xJKD<v))i|(3+CM&?&cB}a
zBL~PB$Hgm)9$tiv{~r}{-VN&uODP-$i&XpYb$#+7{^0KQu@sSz#OL+kG5>=XlXPhB
z)$sHFZv5d+Jd7RlqkOa!{Sdu0RyRhj*VpH=wCC;l;~_D!G52frx69doe$b5kFcH19
z*Ug8Y+k@x(x*vIFBLFkJZHb^YRSH_lDJYN%QFH_Y;~<QkG)y!o)4U7dU>HocbZf|h
zG1$NF&Ld%lnXgrRQsp}oU+<_<<h__&qo~!#QDe&PYe{9TLBmXmZvy&9>p#6@-P*Jb
z_I)h0i_hr{X(l2ql2Um9{Q?sNmHZJ2;`;t9V^>Gsof45u^bgj!Guiayy|vv65N@J@
zDkQC{5)5q-Vvy!iuumoK?1of6ssA1xC(g{5rz0gW6oY_?2e^q%F%5<&puNIQ9_TWY
zCX>laf?C3U^~sD0)X3?sDf-o8(+t#O6Zdv1Qz3E$X++Y7jVCDRu8D;=ig=9C;HNm;
zs+1f0!TxJc1zKg>2BQ-)eSWZmAOh)3hCM-|K|u^|SU3#Qs0#LGz744#I-TREf#6jV
zi_7f;JauGPTi`vX*VN8o!Vsbw4^K|=-q}eu-3p68TG3%K%-BFKhVEAqHG_{%?YGb~
zj!_pFh9tBPCi?}Ip<pu{3Cdo7%=iA2F$K@}#|Z%|xB(4)Lcn8pY5&*TC**uS3IMza
z<3RMTpNG@8$4!3d)O;VKN_TLnj85QkjNm6vfdgiImemeUfYMddzm?6ifAo8c#6uCs
zhy?&p#7MQ9L@)e?<W(FxX@oGm<}d{iMC?Ht!MV=G&khXXvFhu2!Y%Sd*hy_~T6_=`
z{_`qpiuuA|Nn?1DRg!?D-&1fj{Hp)So=$G~jf&iLBQ&6{;m07nQz9-XN5uvqS!<(w
z*HT#Ix>HDQc=G;{)O^nSPaqXdJdJ5*DN7z6i3^$k04?zCstpM8WFT-z)+281_ZkNR
zhGJ+W&#k~nT>JbBgJ3LyVH;`nL;)1J0C>*C?|YE{Z-y0`{1;-O-dVbHu1g(EKxsiq
zG=Z6*l6hN&tsAhkukt=?(F10sXH)ez=4*+~F6F`c>!bRoLqhG0h`Leb+bqKgMm<{$
zwtxzw-jc);lIxt+A)O(rsfARHP$*#7N1e8Iq0%cSHD3upVD)MAsU%JL(iJEk!#vyq
zqU)h*3zVdv0KnXJQlnWwh)h6Q>&{wHps8KN4XMODlU$yk9VFqT^EgP6)a_;K@Lc;3
zLD?S&qxo%l7^n(FmWU5fRMhdvh{y<(*CFxQAOKJhq#^~er{NsDWJNL0Ho+jvAm`b~
zNP%dA9+nR_tnjfC0!0fDElsTKt9fxGlA{z+NEPSw?#^*Fzpxz3laCB9|Dtma_H7Md
z1`1$8cm7MkOBOJ%U`E^{=tn^nWwVdS@N*-Z9<SW?jW~WuiO6-+oSLfsj0nl%@(7rz
zjVWeI!qgWocWWC607AmbEgC}Dqnie!T*xXg5#12!Uu(d6L*4PVPqH18Z({x)aN7h4
zLYlNyZPH{2Az;h}i(05J_!Sc4@Fwk0+%)@ZJ6T8f;IrbA`(Z1gupg)l>{|XKS>AdT
zWK5_AhN=?HFHTLgRlP|iKCA&t;3QBKaMpa0Df}XwAQ;-eE}`Zfu}!r7xYm*o3u2(7
z`<vO=+>O)`Cq{Ne%r5r40*F*2@Tc8RiawkyvQSU&&>YI@*<~UsW`}D+1f&?)h<cX$
zx0jxgThbX0L>C5xSXu}XR^(ra0$x;sG6OvLNDJA8&nuho-sMOLyeOoB3e&kIBr5Ds
zz-ztE3$CW1o$=y4p{6EOhm_QwNb$L)28;z{Nb1p!=0ck+dP6nYZT&7%sMotwSqy`4
zn&XSrwg^&{{N|b+?PGJJQ%7jbH>T)v{Kv5THMm}#uHi+4AA9^)-*tO4vDoOx*92uA
z<f$b$0szQ>FaQi(7=SJEgKCzmH=#~QE1^Ef<~}@J*0j9prrAB7z~n(U#@S}TRXw%;
zxos6w^PbR3g1oIY<BlDz(W1((dIxFJaWNn`uF=;Yls|I|`eWB|-WK$<=d_7rv<(sI
zmc?qpNmnlfN~$k`YH0`l(y%$o&t!nZsx7b$$<c%#4lB?fH71Y;NtY6>DAyCO1cR!w
zz0B|$s`Ai%bzmM^J;7_Tu~b>cQS*UNG;6AtR1I-5ndCm^@VELWkTM0QYU4Tj`ri3n
z%p!c&AGW1=#G&#~>_6`@B~ZW6=T5dwL?L;Nd<k>bQXAZk^r;S(G6p=(`DddzU(K#-
zPs7pjIW(5loK7GG1tYm8`nsQ5wqYN;^It-G3kYx@f`*w5=cx-M$=ta-3C1p(yfeiW
zD4y^2a<!4u#{uBJ^}ZE1E$3y69H{lY3xI3b92$=;4^PlbJQHmrD9edZtp@-Ra5@iR
zoPA}^;R@fv7tf;;#I~uD*E@Y^qV2=3H*XObBGE=oB|hv0+rUHM4hU+KG0`W7tETkl
zV7d~6f&Ahjle1D|RYn@UnQhS__RIH*2ZBT6Q)F17i&{A$EEZ(Zmt2af?;Z^HzGF{&
zx~>*aAkWV^PxAG0hT}ODwk_!J3U)f_!BLKHBLg>vzWm@fptFKv{C}G${;>StyG}+n
z=Kne4u&wcbPBFh&^;?QraflrdvQq9#7F@Zl`ef~dsL%w8V5M+J%87E$uR9%XaYPnc
zro@t~M-aE*VLG(Po6TZ}rU?7Tf0xIPm+Pcv@uHfFm@=t|iSc_#26P2yMDOeck~zLP
z-=CKW`rGPcpb_bQJeLJ(eBG}m9u9wb^MPW^ginwO?@pgKH_xUwQ*TP`YInCgc5C>Q
zD|dQ5-M_xKFCtg3XW`LnOHPENYwT>J+nFzMtz&2DSS{KZ<1vN<#E`Lwi;ExrwMaX*
zUO48?PFiQ)d?G?bqMS;aI&&;U2qO<54(c|GCpK$ILE2OWGduKG9NIo#Fm-Mwe~4tH
zx*l(TPP$}Y`|xa4-=I&@DGBonL)Jd>8TjKbxVO`~oY(_#bGGfvv8j_^((J}=lc;wC
zQeptkqKasbFyx2zKhLxm(KPICPf*thX9@L_BHxZ^<Hd*%fg~U(;aLd_uy6O_Ez(nh
z#Elj{xg3*i7GqDIqhkk#7L6q=NZIOelkoz?@xuo0V9e`BkIiYW9}*6OKoSa)GBi(4
z2%Tm_C#JqJQbrX0)i6_3j$s-{u;B7Ig*J(bU!hUfG=aCUy@B>oetJesMY52tT&#jh
z{I-B{mBw_fVPez_P!Dvg-_CMOm4|tg8<ctQ4(){{haeV@NLg50TeiP@80x|~k-=5F
z9gFUYsYG-)`(k3^#MG?7u*BScbdzKi(ma)wOZ3{u@i^AYhu2nYw+^&8fL9C7L?>WJ
zl6=E)a<3NabZ>Z8`bm%t#-$d<fNt$LpGnwYu?_~wlgbOq8kb6Kwx?pF`Ol7#`#}v5
zDRuDu)ZP%;ta-O@bCYA<tFt|HGgAfpEhAYkMzgd*mtu$=QU_XB%Dw`!aH$3!pp1K~
z)F`yKlf@a??qw>i|2D1A5Ow(>lg(^itn+dw$AnADbbW)zR>45;Su8o$MGyJ{P-%@|
zWzJ}Jq&o4J>I<Vz1jTNq<Mms<0GeQ(-G-4)+vs&&YLB*sTKrG+kwwow0;KBfIPyy=
zh!E27-Hrfe5*cOiL7`z7d3nJ<NI<AZAIg|)fK`v7A4e}OTtbOC<r>}R2-*)^3R9Fl
zhzaI>R5J){a*<cjEfuQBUv?Tp=7-9-v7vyrQvRAm0RdXVzg$QT!k=6jrvk^KpdBHG
z09Bb?hh`oiuo(TN0YnTcQGZr8!Mu%suO}GB`SZdB2k=G1U@_jOxl|8z%p*|3D7dZt
zBBWj>Hc%OG9cunKuMo_>Kme9Z6ZLm5oP!0k%2MVp&^=;cEf0JoA_B{RD+r6E7#F0a
z!z1ul%rY<R;P+vtYqMK-_0HZWCr;(p1uM2X`I1J0&J6{+)=5WB!-<3DbdsYs4&9(6
zMEJtG_*eg-qN^WYgMc~-lb7oO9Of_M8OF))!9|bWH+2Rzm-`Df8g}ck?Herx5qJ?X
z4sVG3qssXj<bWczJxLB)Bx?5IA8!np$t{67igu_q1{i6SBVTq$`857PHoZ8|3NfL=
zdaiJj7xDxLkQ&tTOz-(gNF|D4xy4V(po1z_0b6^nwVy2DXpa}#k~6SsG{o-Dw#&J)
z#v!k^Us^>abWO9g`>44br;ycB&HE1@il>C^dzp(Ll{F9O89ys;M0tOZtU?zNgw=io
z*e_Agy(&)~4o2_E9$-GKEM?}enU@h1+|BVHy^r;&kH+2NZ*h(5Zn^}o)-ANm^H32*
zG8>wrzCFQ80xGwr`|<0aPaF6>7z(RJ$nTtf=kJb>lSpN$PJBM_+Ad%t6fiteJ~X2o
zdr(PbH!jKW8w!QN>kPwa_2qsR{;(0WpwmKl3?~AJ;S%8J!LP|2UhD=%any4bno(4+
zewrbI^InZUc9cL07Py9Ez1>iNkusr`k_4MtjRzrs=+LA=#Cd>rM*NlazmP*{6;WTV
z*lJzjd2UvZPg}mOe)BH_2WwrCg#5tyNb=TTBefp9!t!lC>e#2;wKN{+JQID^|FZ8W
zfSDtiMvehY_~(L|gfTaM^!#B68M|_Cs*H#DlN`o?y#0T{j46g6PuHI3*<b=l6Vm(Z
zU|2RxYf%l~iX~rGU<%M)k9D=BNtQ#X@Y2uD19LrR^27`m@!;ByW5}LIfc_88?DuJ$
zAX~N>=T>yTt4yKn8loVCviJ(0GJkQwM`EP@J{kh3=Tj@HX+sGuNh3A**HntkJ`$`x
z60cKXP-%iP>O6ry5QTmDMzE0&DuU{GL)z?<nKocD1z_Yb)1CB+&v>vj&C|mE#)@q#
zRFSoQmBR7wRKK(NpHR9IGhaZ6*2?Yw_HShSFBR=fj12#KMf>k*`=|x+$HzCAw+4_t
z)-D2n`(fSwZ#xjrXebdS+<lF8<@}(A#Z?Ezd#l-Nf^f1`ZXk&`Z}`y+Oz6YwbX;WP
zCXCsM-M8c0zbnjs&P8!4s>aF8sBvlga0c}C<3ArhBBT@Z^P1vaoP2NBi=~l?v0_$K
zx95-e_;NknZYN~;bi6%$Z=Kk_AFYco_`JAZ5BIfpv+S9X|9}h#4a~?n)Iv^SpC^KD
zHkoCZ{-ar8wK(r-DFr*#oOn#XHQua(7|A!Ty?CoM$!1<DX|XJi$edTuMrLEPZMRf?
z1Z7x7VO%zlAt_wBayf9<pDwna!b{EDsz2#Q--Ab4XxVNz-A=OU=(xsy)n{fqbCtpa
zUdTZyKpIH~Y=`hCnhljnh7l%&G4f{oaiEEIqyYl-#|>YgzNMLW`t~0_Y(f&R=f2e=
z`x6%ysW4sy6PH6=h2fS!R8d6iogt2j6rrx@PeIn_*M>HWk8PAbJ?7H~Q~Dg@caZ4M
z3x>pso<a=P?iWZ9k1`aWM}v?pBT1QbvT>0UiE|3EA7h!s%!V#PHfE~KSA5wQ4^k6^
zdob0YXevBvF&&Q8ctd3Y$R0$(3>}yB&uYAH8LA0M|5wD{cu4J{Xuw==)J5ryKu^ai
z+~;<dWCz8Ux%)he>P#)o9PlS%Wu_b{{2`O@V{w|IO|WMWV_}2CbLHIA=jM3*yTHys
z64Y~QKw6cpgcpMnYV6k-0EwbE;(<4csxZ<|yneH%&&kmyQ3eO6NL1{xa7#Q7IH>do
z>sx1`KOmtgyP(iv$N_ItqUgY%z&b08If2n%Se6-X7!jMo!tdg)W(zmoONC745|b35
z<%scc@xhZ)9J|~orNY-X5r=5VZ$GP$2Z4wq(;aoONIh!kFROV8br1*BN_2~Dgn`-@
zc11PPjyP15aj6ST^V&3&9bekmvQT|x#6<^8Z?jfw&)K>6VhMZ4Mb8!wiqLX%_mv_O
zO!eduBP1;xISwU-qiX$RX9HW8cD3#E)d*9^SxFh&$9q^+L_ac&tB<^E>xIqw^YZNx
zT1RFW(aPZ05<8DwWlLG*kMdPDxfg{ivYos-Xtc=)Bd8i>sSRvLQ=o<BYX!c(;U=a7
z1Evs*c?gT%Sce=L(30<S1&XgE*s}IiuObu1lE0V2bYWOVrauCN;@F}1&*2H2blyB6
z+4w)46M7pV)}qJC76TYcD#Mt{)Ruj}=6jV*eqy>?!^~NF8{)ha^Y;v!pb*`^_0xn9
zsq}(wMNt7*6~k6Z3UgWf=b|{F1mDz%y1t+TVwzTlTCL*XgW#}Sz)Kxne^+68_(8_T
zZiV&}pUUwkM!s0KP<exa;{&n8PpU=y+~eah7uTMzT^_t|;)IN1Ss7kS%CJ3I1MW0b
z-~lKmb|~2wD<d;gxnmxN@zm(|xV9otoKz8Gc_1%)t0Oky5a4P!V3Z0A)vD}sOR(p$
zOpl}ceYF0LSt@auGY>u4v{r_288OJ@=__Y;CYdA&N}87n=zTYV1F9&jj)2-%cC>0a
zXExl>$_^|5r5Fq^qcFRjR@kUJ<<^^9v{Y<>&(&`|`6v6YYKpn)blmj(9sp5sj3^<~
z&99$$y>8EZTy@?itCRr1cAd7pk!#vej1K5az1!>eL)7Nj+V;5IJacitz6#Gg9or<4
z(Qiss;78kcG6y}!O9EI%&gq090@388#;vI*2oa+o%z19zjI^cNPJd?Sl}^pbJ=P>+
z;(RY|F7#TQA3&pWHV8sLRV{U`W@55G|NeB}@6LX2>hrmU!vFFsazP;&b>f^tdd=DB
zp0Uu)N{-$$wBd|&Tvgt7SV>ohomD=F=+zX4eS>bLv#Hv0nC7VT%Ia7qQ$2ipyBWn^
zQod?GvNVi?HaSgI$FMZ1&}u4sw>ayE=Rsgpy=}YUSh!;{6NyXmkSi@0!{7SKH=^+s
zgL;{8?gXz(2|+%V+=}<phQF|bQS<|%zcL5qUkVQA5K9J)PFBryaHU&O^<#d~xTX#3
zgZ3Bh0ag2FJ27cims@hlr9GTPQB^VMG+;w6-J?(WuMWyZ5SB~L*cvcQoK-@drOQtI
zB087){~V_ClbM%F8gU&u8Rd_1yMRRdUnhb}a(<~Qg&X|h=xETbkMg!5w&-{5m-c0l
zGK~7=T!Yf`ey*)s+flI8nF%o&=+x)LJDcQ|VLGTdvj>n{0yGO}dn)zkk^g$0J;pHg
z*DW4f$h)wA@#U{T^1RC1J+Tg$O+j5p0Qx;H)?~zdE}i&a)cG=wTvP_i9)E*R`*=BT
z@9jql=A>dU(Evf7HTZSVE+y+sfn>|wm$|}9^(_0MNj?D-PtRxv_@#cU&|n#-rhFmS
zxb(Q>s&w4Vi4Y32Ea7Foiw$Kl^eI6!K@|QDob+Xlq}z3f*Az~z0ya1I2A|nO2lfZ8
zmFt<sPB3KRB^xmDQVp_mQ(2)N^h0F2dm(C+loCnOU;mmycaP*LK5HuLg;vYfsO!de
z>qP=`|EcZOYXKCTL0PMw;5^`+uyY3h>pw2rU>k(km0c#v&Sgs_N$r+r;aY_L<<^7k
z97tOLWuF@T+yA<?IHUmptq?(<CLMAl@V<tNZ>rg(0wfs7prk3$1}gyU(q&%Gjig`a
zZo)aJvDx^VO7JMU?;Fsv$?EST=wAXuTG=02ZOG27kX==~)-&2&(Swu`gg8VI35pwl
zRDwV{%Rl~2ka8q%Hf$F_3MaG|LS>BC%BBYSZx=`kMJOtf?@y9=VavM`5#($<(Pb<o
z$<Jz4Pp{Vlvptkp{(l`u{!6%#gXRBoCH)W0BFn13ZC~sN|4ATG05VRNlHg^YnAVG<
zg&*Lxh~%k95cefONx1ys_x*(9D57=WUM1xBT(B3P@%m(c2xn0TV`<~&9FN)25l1vH
zFv0_0S0XszA5W7n52%9phIG#m&28Y@)nTX*n=}+b$|nfXae5#9=xu!1G#^2Qb(bzk
zkFaoaaZcUV&E7IoV_S7)cV&L5ysnk|)m~?3X6@Pe_@;?cLfBee$+i)l0bb@53Y2aa
z{Xhi~Mcnh$CDWD!RDomQ<5TwQ!^Z}?C>b7awF?wjm*tPCv#vwEZolz#M3&)op4#vv
zTN^DpH)?&$L(CpmOm^5KuZuOCGP^48O%)D?C_$+}lnkC>B=)rZHuJ$V%84kwXzxp%
z8!;+YK<PLKMLybJ46yw+Vb~`kfoAF*_CgyfJr8Oi=x9M}3@zF+6o}d&O!*O3ap=#z
zNfPW6)ha(7j)5MN1EJZ(+8pE4`ow`Wc0mm>%BcM@I+K`rHK9`JlW(2tr&hh;y|_^-
z-}AjM@l|spONB5o4u#3<cVFVEG*9d?e?%M87H(1b^lq7R+S$K*Zx-JE92=T*HJyy6
zd}AVKh}ZP*2{VMBjb!rStZ6R-YsgaO8M~!PL^A=Iv{%n#U?eIgq)~UBO{Emd4yber
z7kUre?3qP0oK@~WQ8%W>DC1;z|8du$wZBzcPI3=**+x5tl2Pj<TQkyg$=p)i(B+Uc
zz%YPVz+X}h?k8k*f``+~sSLgVsai<ILqBpkl^lNtTuQ6dzeG@PGFQ9kxZAmlL2UY7
zYZky1C{&f&wS|~CAWJ$!bOCz|WKzFGk1tULDS`>5V^cOX>UZ;`E4NO^*iA#eNOQW|
zArm8Yi=@(PiB1N`m?j=E%5C<&3T=4qH^sXvqe6mH5kN*NMG}Z*A~Q^7+!tNEW#1FS
z>hCl3ThA`>U$w$YQ(fHCQVto2ukyE}w$kr>NXw$ILRg&0m~2ra5u_$U3gSlPvK+ql
zWE%334#D`zfw{Arj1z}2+l|YAQgfyt6)b0)CtSh@?X70dKA-*U``0>0NJwIES4%4t
zz??Ag+8R&D8z9U2OC41aFS0HWG`);9xs>V*ky9Cyj@q>HMP{;Sxl$f`jfNu4k33a2
zkJZb=*Cghd8{nUmJ!_iBjR`1m?b^TEv|8&aK2>#%+T9*_qy>$sKJ~j2QTCi?MB6TA
ziaTo(feVN2s-ozrs=BDomA^F0k>3Qb`pGd9`vH_%1aACqUx5F{jACMD`k$9|!y22m
z*yD)at7=ok;cgxuKI0?M)^TfNP>@Jk{CLJjjA&vD#*~Ugl%CQ0KRl(KCo6+wloo8`
zfkW-eu4>s8HJr;rk(mK#cC)8PFR|!myIl@d`XTlu#tsfT5IW))VU8tj>L{tqwqu`%
z!|-2m2Mxx`_v;`3u6;Cqxar#q8u?f^w@cn0+pJb9tI>b{MWX~0%l^G0Hwki#DgmVq
zBEq7UEDmo`KZLW=-dcH3wCCJ(Nqo=zn9fRHe^+`c*O*zaT4#M)={0EFJpaqKclZ6v
z0~*eJqWZq>IKxw$rn4L@!m3<6AAltuSXWMeIDG$W6uw7>t3(!|SIc%+V42)iTblu&
zbnoFHjcq`?FUl*0y^pf5Tx1k5Sdzx9Pe+O<Dx7~4TTbN7U35>5qL9F_zd(3g&|ayg
zH6+{&f)tiG`?L(V*F+sTH6g7@bZlG@dT7OrVV5j9%Cgy3^x|(wF67_7QaxLjuH;wy
zICM&gQs?p4drG|qW`4jl`pFWj00x?vWIh3%lzge8jo$Zs$hw~b5$!uTi&S;ekmT9l
z(Rl!hI?$~EBs&DbH4B5c{>kIk<WF8Xv0;RquANws3kHhPOjlCMbd6k4&s>n~l4)(h
zPIj2Ga7DP%c$aa^D1OQ&bqWRH0%L7VMmdLk>-t=Jnnhmy=~Neo0X|Z+1M*{qO0J-j
zD(V0`17cL4fx{ShUK}(faxz&;dzqYkI|b#hLRL&D-5^nh$v4))pp9VS8QXm=gZXbY
z$cV2Qh^7^GkB?o4;9V0IE~IBHv<K5D5=TFA&Nq87^?g#evZ-lUaV$(J!gF8Ia;SE2
zBLJfR!()sIHU6Eb&7#IO^@(m}Xt<1w?0(!a&@zF2G6Jc1tsd>Q*Us?C%0+LJ!<+NU
zwgz8KZTkqBdD50CSlxF2b<&wGtX_zVWd{CIdUCyvIWUZYJAFWLTjG-zoKuWqOx6%(
zARIx_?}|qWcg%L{G@LGxqA#dG?FCqcrIgV!`O@e`hpvOxK$DQAIaYxFsZ)rtG{TW2
z1XY7KjKk29B#oLj0)yn{&Z)MbRKQ%ddiG&i$?+CrS>w^3;00s<xlhWde_Vn=`nmG(
z3q~h960?;Bf<#!8GQMBC@~e)}aSgJF?fhAXwVp=4pfH4RMFyY!XF`T34)sdcoODCK
zkYs{k&qgOzh-@9L#Su}oa%HKtI!j{PqI#AbMfWO#aaty<3bsQ>a4Ev&B6nuvS$^fa
zSz?oy-a?IoRlzlSlFTOYT4;+IvBfA<Uj6BE7TpZqj7P43Z0ba+M-;5BgRApXJQ7s*
zTYipzzZYb_Q;@Mhuto#zX7(EqI}62HmX@iD7DO#yu-X2BW(%%v;)iwX9b?N5KBzXf
z`Q)<Ltb<mEj+5?}m(vxxRO`?95Z5S?KELxIi-k;uR=2PQl&7du%yWe7SBp8u&{p$t
zGs2k=a7H1N;U}E1iQC2zP`=qoY>w)fcxI^c!6O^R;tAtZ#oCc<ND|x=&#UnULI$QA
zOWrqLEDWdNQ-`>+p2HK+5cusT!~OU3Wb0I;S%bAQbJuL5-<+hZ!juRXwnxE{PX2~R
zZ{MrM7C-2DOHaG{=6bc#qj~cgXp~b3JW0A^WYD3RJ(*WDfpF7u)mn(Thi4-rSTIG;
zR&&EUtG`}n-3snfUd{--e3e{TOsSu|Vzh(?QIIqNxfb`Pw!PRO^sDz02}R@A_}svo
ziA6@y29}+7GmhYji-`pk9-2tRh23`FRJ_QriD={Q%fb*J*TX0k{LHBz+#O||3*k<u
z6`<83CM!CXa*_FJT%qK)Y264@0`FkdN2A^`5(ls_p&zafY_xwboH>iL(2I7qWggJH
z{7qA<p_+NT%27CTNK<|(ePBV1JOxp*3?D1gwKh_!dv4kKnbERWCcTuK(a(gCT6g}1
zB_n+pd&^ziUHfNQr@NAEGv&(1!liW`L6l7hmQSUxZ3e8G8?9*<*An$ZH306DOM7W~
z*(OFvN@Q)WkR{!nqvky)lDn8$Snf1{j`W=3CgDx*V#4fSne14#Wi2>v-85@>1O&2&
zu#qO(gEvmgmq(~V8YSOZ$(cOH0mnF`^ccUA7BPM9#5nn|+V2ldz#Jf`yAc~eFmx~-
z!~`vDbdlp{)Qh!bCb(r9=a=pRW7o;Q=-!MHHADAr&KWnpLzDizCjCJy5wN?Z+_Se*
zwbqc}RaLay@V&F060Y0h<vF+zDTG#^@}abrf?}<PVC}noE~j^G0IQ&x?vJwrtn8N{
zX(t9b0o2m!AIYS!zy$J}jaeid9QY*>qsCY7;k_0hFWko-5qE{Fit#ttLFqw^nCGo0
zxNq0U+q2_(@+R`<oal&{J+{j2qb0NFQ1V&!=8zA%VkPVLZ$&#<Y(r!^&wpHT1z%<w
z+JD^d=}yN=Bz;4Z<jyWXH_iI?XOHFqNZqW@E#I(#657tg!`jdZ1>p%4X>aT*?TkE2
z1$>e$V9^}n^>nRqWa2x^)t*W=mLuD)2aTFHhGVrxX-v}wFrNy{!A=gOOEqC~4M)_q
zb9|#T^WTkJgC|$-%;r`6@!2Y|cMWaNKkyG}zUlg2Dq!h$;of`EWRJ8DXB9>-e(tT1
zbYtg}OgR&MOpSq1!kYCky*X=q?pYtkEP7ff_Lwl;eD;dw;XwZZTwE>J@5IxkCqFs$
zq&mg{8DL9+$_|;fIu~i))nM%kqeVgQE?Kv9?*IGyVOU?@V|{r16n+h0v4YPV!#cJq
zi-1U-W686+=DSh!??z)*!KMl!T)C{s>-nFuQX~A9U0zFG7$~W18SUpA>R*o9ry2>$
zCf|i80-?wO{%fbz3+@`Mw(}ZN*W8N~yJoYRr)LDaD2_l&Gec;zFEybbXT4T%R+=1}
zcby1jgAVG7cj`oU>UT54pLP4;(Ff41fqb&9#T?F9^wO5zgRe&~Jy=$5qYu0CsjlVm
z16mz~@dNHc@1LknZ<sorPQ))QN#@t^tYan(ZBsNe%rNGb=%u)i2<A*JZBkydk3ovr
z0CwpO-SEN+g+G+~XS&M-@n|CAZEKJresy%1z_Rc7E`7ZR|8>&(uSToiKIZ?iNM=}L
z*N(Uq(dR`kKed3#^6oWL3dLJIqjSh65zLiJlYlayS$Gq=$_4dmf)_vH^9t8-Y`eJW
zxt%|A0G_CQaDVFdkWDCq?i$VM?rAvnB9iT<NL}5Kbv5$oO7Ju|Aw#|#?c(Ynl(hBn
zKKFvJ+kLbwJB-l=vJfl@q@2&^Vff{u^@d(ed!Cf-$g!A3$EWT4`J=b%3p~ch`ODzz
z`+c&vy7uRN8@%>DK$KXPMnA|AqZGOYA;NqTEdn)2qEr1wzx>I2Jy4W3b#uje&2odi
z>90v(b;Xa(r<--f=5yVk?8hElPjSjucjVuRK~m}zR~4ld_6B(1D;amCWfi?*l?|<|
zvx#Y*n6`~8?M399neuIz%aRBbggN~e^gI!YzzeA2RGp{s{WvwRdyCfEZ84l~AuGI2
z>Qi(oiFD+_ud!!&UYD|i5t|JiB<n!^%jD^2<25b0aN2|?55j7Kh`&u<i+`|h@8zH2
zOdZDs+_!n($+W)3QHO5_+t|kFHjNsC4O}WxF{Ppy+8vV&C5qyP0I-<i*%18z#t}&~
z>?@H!zoU>aqe`HE0SqEFs0>sXdf#6=tfvy2?2kY|t)a5W*-)EsyNC=743{zZ)toq0
zho)_n%nbbO(4y?L_WKar@-Wpxp;z1W7xgMDfWbfkQNV2y&g*LMH9iCNw&=sJ4R#7!
zT2~oxf5tsamhN>t`5|BGAHnqAa&%Q?w82Y@7ZIaPtqq&$#0^L_yQ@3C<+pJ@Dq!yx
z5NaDGU;yNxb7qYnwPPp;lBSMT8Ea;Mx9JgpYvg7QmEN)305nOzjW<gUN6PHIV6xVE
zalFoE<<_I1NunaccO{Wg*{G4Af{^v?sjL8G1x@ccvSb93CQy4g77qw#rE7w7Bmjdb
zr;7tZcN)n0fHVib$@(^EdToGnl#OYVU*WJkUytBa_`#jzPUXha?ZNBm=v*^R!SJ#=
zE{w=Y_curmFkC^H05+r${co{p?}7E_fNAh&+W`p>6#%roIQQbBBb*j^wx@7hAnhfI
zQ1%9!G!zF+Q60hVa!!pN-U)Vv)BSgZdLln-_PnHO!dRiQWrP(;)Zc;|yHcL9Ss;_;
z=*+Cs=+se$1iz>M@%P2RsR>u1i$kjwqUIwe0;QtO2)@Zb*`pMQdlGe6z{4scImNw@
z=^eg3LKF_)W3!FpDS<EuG?~uVCb>0yea9>MoW9pTAn=hW<)111K^HCwrwQpO%-Y@$
zuxAeYG@xs;Z+P?rgz`5fp_Uj~2x0_$FKZc*(`FQq01R~;hS}p@5vR#BFQ(LiYCTZa
z&+OP$h>(FBZ0T!prWR#wNk`8&J8)u}+KQ0;OGYG6STByO4||ka5Qe&GKFxfNIwO;!
zKo(lureJokQz{reGW%=G%h5EN!(wb60Y0&=(Mt~hJ6d`*P|Y-W+U|>hfSQcWkZ{+&
zHa3q1E2s$%IB#j6a&BmTG9#OcYeam-Ji>>aX&#MLeYdH<9?8!R8eH->&oAR2>s&^0
zro=E22)XL;)F;}n4xU<%GCZ>)u#X;bI1oZ&eFvg9Kpz)`V#>tq1(EP#5-Q{%smAKs
zJZKlbNF(!x3nm2gB4VOVHHRjy-~SI)K*aO64o0b<@GXBvP!;mMQ&`NTC@-pbFW@$)
zT4viNyhHQzE2s$F(`G5cNMFA!mM6A3cKTCzij;a*1+5Q28uTR-0IUsv>MomPq5-*M
zk_3iwD(<7gG_WTLH$_2dBfHM316ThU_up%F*vx1LLlQE}QUdeu{X(O&eDCTX3P1w8
z>2}9N`42&4?VF8bPyxL2S$9u$KP;Ac7kfPC($js@=VUUQemxD=_)HI**+Y^$dE6Nv
z99jD&>*egZIrl(4*n!`Tf#_>hivY9)4dnVmNq|J(1uJc(nvc=5hcfe~6Q4C1Z>xq{
z{?+T$cxoU>FhO<JxVf1$lQi&!oF}et7(k*TZqy||UmGUEuPHpk!})Ob42@3L0)3$@
z!sd)s@#QnKD{!V0Zlg6NgUR_;#(&w=x|t?X{u#(s`!G`hP-y|e%*79oakNbIJp*(x
z3PCCq_Bv}1ntG8}b#_-0ivcy=9wH;J6^sw4PDK>KaXlB(6Mfvm%ZwgXd`ahynKhF}
zvFA}ddz<0|#svKZ3qo%5m}+e0NVPm>5=G_Epbmv7pK*ONR@+X-J$vnvcM>Rvp5!wG
zO-9RU(rZTr-HDAJJ%4{<qVGr8RLz{gJn^GJ2@#^$`%@WrWnJDO(Ub*5&eorCS$jfU
zT^L;o#4`e!?bu?FPUTS0>$PN3g7wY6Wg+mXFZO=E?t-_TZ3;1wq*aU}T}lsN=io?I
zxGalhg<$XZa*xP;E$%q)w632>-M~N0nwSP)*IXTqof}5L{rqa;UZvNXNrKdsa<&&z
zNc^BabvtEc6oO#MI1{CCp?BlUvgfj=FtQ(TXk_aTq}Hr2b_#I>>f8y3D5R+-?;<6f
z9{1-yv36M5e_!kW_Sd^9@5cgkz_~`|e|=v>4#NvLIl-5f<OxSF>nJ^s-ZGtcT%L};
z4F*bvKeX0t;#7Dm5-u(xPvSn>0V&;7$Q9=b{-a0|gf;#9hokLrFM%A14${QT{iqac
z%vrV(u)xqXLpns=^B>R2!R3Ql*m}+<eqSbishhH)G{fs0sbV*E35;ukm2kw{ZQ2Fw
z=;Ey*Sc1Kz^e>qnug9Dh+%R#&bn&_(^i~(7<#N#o2j6WBK9>8)lLLNQ6FrAc=-3bD
z$cQV`a;Jv=8}?0^2kF0!ARNs9w=?Sx<NsbH9jz=AyU2jh{i61oCcbo0%U;i3e+e!E
zQAC;rkZ*n1f@Xn)<Z7Vy=|+}av{~nWO^N4ex|5lnYG5pb(VmlC+So`v1H-svo-ffs
zq?jmfF`+EHkmUs@EX@IX5^vf+d|-O_KHWLVk`-CT%vZt|`5<?8F!*t6_5OS*IesWi
zRY0@g7w8UilVw^J3a0HR-ltPJ<Gu4XxfaJkKqhLmR4G5wuk$bBvZW4Uv%_&6AN2SD
z6z}G8K`q*|j$@++M;k=UF-s~a5f`UWQJm=n64Wqkd98|Ht$#SaQa<S}KWdBBww<!-
zNA>dDi#3_kKsnP5;&p&1<Y>db1{3;o<Q*oq_m0_PCc-0803p|$vT%f7SCl;-(a*jh
zAx#tts|~cwsqrQfCc<(XS6IIKb^x9p9oXAz)@Gg4H`cDQqJMM6=3?L%61({vKhY++
zgNzuVghp;erhe!Pjlftandz{#vAvG8z7&o9N>izGrqy*VzC2vhTCtFG0PdmcAZ)Pa
z8RYFro(3sY%mLF5>5ioAd^N#3VXl!}Ju4ZjnsPD&tkTqF(_$lMKw}|}_0X&AA~{dw
zQKGHDsuYkMF|xs}Vhy}3eY7$Mn$Ep^VZk$Sl+J$MIHx?}leF5>B=hg1ga1}Az5bK`
zRpStjjxQVDy-whe!TXurJB%-f*K_7kq`*Kdjj`F+LSGp!H7VQ}=M~AJb)9FSeBuDF
z*7k$Qg_ee3j5`586qJZzur9?bF`4$$e$znjSfdr?xti*u(s{Rk=d&L8%m30^@5@Gk
zwpZ@LR*<&TcvN)K{Z+T5kx!R8y8FMry#H-}X5skrzs}Fa8XA8$f16)cYx{_m8O=_3
z8^lx^*0+aTDiWG^tW%+h1Q4kc_7d{UNt1tauG9M&353UY(ElIG-YK}VH(>sZZ9CuC
zPA0Z(ClhmG+qP}nwllG9+fF9Q=B=&&R&DL?;@!J*ZqDuLr=RXm8$dm>@p^ml7Od%t
zHR{!zp1!SQR@(Zw&>H=ZN$Ju!88ltwEW(+%MceSg*?RbXFm(|_=E-lQa{q7CqnBYt
zRC#IR0$ZK++idQ#a>n4Z^&E)bCieBo?{eoKt>|&sf=(fd7$u|yt5LdwNFcb9MFHLE
zrr&g?syi>L*s-I6!<5xF)j(=ynFLlm1B&R^LxtN9IqSei{f9-r=D=QkiKsG1sQ%Q!
zv&j#UV8Ud3?v19*;P01z72dpyPX(<2DLn(<#EQkifOzO(S~m3^0HVt`3(hL2ZU>)e
zd74=k;V4Ood*<o6maYR;5G3oKk~exmOJAWJBT=HX%uW|Ng+(^N96~DqH76+S<&~Et
zKVCfKmnQLajNOS(Hh-J?{_I$(Aa5mtC0isT+U-dWQB9;0yQ#+Txu?8;xqAyvq0%l;
zU-s+h*{?+RwlX<Ri`*pZencP^wGKG5p$aBUn%)V4-rn6iqPI`to5!Y-ylVOAxkfYi
zDSxCiAdO*Po$#|wLuFXEn{GKfO*(&%!#<!CqPshXYH5T7B8+Icqi-|}Trj{KBl0g|
zDm!f&Af-Y@&>1WJ^2M!RCZD$FgIFE5I&Ao!<uIK$iH|dSdQ1gaKL!hmSN>x``0OGV
zi2LFIpX;|^cm?LsJJL2LQP^PU!JbCdO6=Z5*Y@;;E~stAcT(hS%{s(D$);@qknczi
z-b4RkU<FNw)`R7^f+ds{Z3pTFz^+jH&RzRg9!XUuPVTQ-*Q4!;XN+OnN(WTV);n27
z_*8Q((u!xnk-W+NPz|Mg6l%E#Uy=LKcPtFiRfSA*B);ss0jhzt=n>mFHK&7o=$!pI
zW#SRYuGOVmdp(2Z!Q8=RMkmr6PBC<dtkbhD_b*RmPSB+;AIg{PFj=#Zmw$Z@6GRy%
zbv)+<Z@TTGClD7FUmx|Wzf_fzLvKG^bZ2e`@AIjJM#-k_iW}}6eN%a#&Q|n#5-v(J
zD|c!;c%jUBZg<Do6W@+uc8n@Gj99>}h&NK#?$3Ah{usduLj@uTwgvJ_ft4Ii)gl+z
zjj+hr&@uXw1*(BPn;UW>f}8H@%A(#N)WfPEAnm{!!og%JS|`5SI(<BdnfUp0EIPx(
z%7G;c0*x$`_6Wl>Dt~=Vjq0y=p4%vU;S%#5J?nwYz3kO<)?`|?`uuCEkw5xE{rl-e
zxgxEBwgDRC8IE;Kx1R*C3z5aF&>(*H^(9Rcb7S%o70Xh>!{<<TLt(KAgJ5rQi<bq6
ztxj#Q*9!a1jLtGiE)o4HXl2+%f#C?$#o0%S&s+Dsy4cun(Ndo>N(!%T|E%Y+Igwo{
z%MAkNI>6~!)o}K?2F^%3WtV!_rJ$`$5kvq+rL`b5<=1H$0Ky2}A~QT|>ktk{o;^n)
zF8eLi(WUiUNPHTlM(o220-woJFkQwSpaAarcAas=_S)`2X%>l64`4k*p#!cZBw)`(
z*+89NCujxTDHn<gumyGKcjYDRf5BQur;^-MWvXe=F_;~r$Rj53m<?UKqwez4!i!{E
z=-B|R0*DPkd3b)~PP0}&Ox^+olPZ-H+*5NM83}mkE0B7yLucn3x5pf|mY<>o*>5OS
zkyk6~H{oIpmP7%I&}fPeHD_=bV?DS*9%7oXlt&W@?XSUA0FoIT1ms_Blnm_#BmftZ
zo}|zq$m6K6yZ>YwfrLyn9&u7nem90w%B1<2=V$YKP-Wb16DC3c%QataK8362lM8ib
zNE$kAZgWP!2{VK$RXo|920M;V%y9SRS_1}QV}qI_G(hwO?EeVa2kIf6#048XhvfQf
z4vN-CgLS6m@2v>e^NkckiU{|Is@6wN=+-m*YJ^r9%OVjPCMp-vaVZzGG^~<wS4*9U
z4w;8|<uT|d*4OmWQgx)o^^4_Y9UQjC&LJSa2FZEk0Me_e{CMSz`m77n9xC23#8#+F
z0G=FZ3=0Z8KYC&{V+-5cr0lS$vo-XU!v(~!D{z>>MbsKuOF6LFB*JtU>-C_1a~Lii
zRp^HobdbIIjE^qHid(xOMjCczm}jekl_``GHT>EWhYdOl9_O~rxAxn*eGH0pxL*d<
zB4Coy?O>Q4x~sYL&U{DE3c2g_92p(t0CiQk1?*~;8zS)S{Yx2qLnhlxvw6kP$MQAk
zZAI8qp?<pDh7S2jnpWHavHgjig0BoX@1(04#9mKX-1j;c_tlV-Nv3mDLGobhp${OI
z$p56rGrGZ*VhyfB2l_UiT-VhciZAKFGW>Ede06R^<ywV~9)`!bU{u?YO(l#el+J~(
z9HXvIzbvG}YS4`uh>-4@2vZ#XX#kkvW=BEyA%P;jZW-tGhiiMM9N&+zeqWcfnxR_o
z?Y-Gm264x#Y?oj0j=;*B20feM*fF*0#fK*$7waX2i3x0|Ko7FW4h!*M#oJ|#KO2fU
z@_4v0Lz-OsXy@+7_6TCdoSABL@d(7M9HG3W3XUHTH3Y#zk(PC+Z<o8FU)8FOwBxrT
z#{FJ*(_C1z|1h<Di4f2-ML5q7FYcod<Qf``0S5w@L9yVOy}l4ywD`%pEf7~t$mQ69
z0)}243)Z+iAxXLk+fez%6s!Gx#xnzaXD^z?bj+N(UOoUt=at@wO;O@b^LnTV7v3;y
z0@bE^;A<frrg5s{<yNM7l;WM&rS;hj*IoX6g%hM{8^oKChZGxHf}GhA#<FQF+pH1l
z$P(`y<%*PePWf*erAq^hg}NcG*je87>lTd25bteE{Ic!n#uDbEAW1&M*d-16nO%QU
zUz80K8rzXzesSiolBAjXL^Pr>a7-(KFq);JxK$fWGaJ^q4}M}<<H=}J%Z8JhsFqnJ
zlwIqu9_m9`D~CNACs>m|Wrt5|d!47D%c$wbzBaVs$XoI?0PKg{zNFBk%-^RGea?$b
zdGTF6h_4f_2Hltk!p)|guEX=B!KrwWoyWeKGY~rG-0Q;2nbKG7Pa6FLAy8{MGxLaG
zH2Civo+q2pk3<I1n5_V_RY{JFjLovwLhR@$Fc|Wy-x8A&#K=(3#(buxIy`I)6bHqi
zkaa^CR)rCew_HM5Y1&*)gR{S0mUQda&`wf)%1GZ<pnlD`eW;3A9WW?Qp!)T_JGIp!
zaY&~2a)vy6R~)|#R}xUX4_|QIJ4j{k<eZ$v>ARAJeCq;U`4Ka|yZMqi_^{K(?Z5O~
z-5fFdSC2XLqWhFSayejb5hL(GA2c5kcjtq!EVsXbck0s1|3eY7{NQsK0k(#gFg!dk
z4F87($jJP^=I2gn{G?YTP=1OgNUMWNu$6;8cm7%A>7*A!Y5LYz2JzT%lnvJQLh%t2
zYj^r}z223b6;w(ISLv4sb(r;<m*T2V>UyeFFgAgdZm!Rk_Rsrtu?+wtP1=2Zd(^)i
zv(%>X^Zf;Ie8I#f0aM3ZF#{{U-)?Wm?}NRjj`@YR_k*5zU0-g_PsjJ2ngs*H!=r=>
zub;zD`{(;h{Mm=4!ylNo$#}U1tdUM|QoBUyO^MUQ^EA7(D2x#x*J%4x2h^|_f6_Wx
zy=_|>+hs&Fnl7elW<Dyc?>l@140iS(dbWH<6g92Qc%WtHJ~&@gdkfZR0s~hnTYhbs
zW!NQRdP|o!v6};{|9BbKPYVH5)umFubL-r#5~tIUApcC8Iii=>{pmJTHFI4f1d9n{
zmEJo$PE}p8tvs66ij})jyQ-nL-%R7aAmi4-@ye*p?4YJT0sH8=fRn8;51hi7tbrLM
zZ{5pf*f#Ytjy2RA59CY(F#Er}e%4FbahqCa<mmi0i5ozuZ{nW(#E2N99HA3v<%k&Y
zJ@D!uHs`|w2OW_4d`r!Zbm`=6u1}y%D9?Q}Dk51px!t9xF(Gio7^T7<upbju1~Q&U
zQguyvY){<8@u>ZAN?2)ZQ{NgYDs#NFGfHrO<E_h;Q5A%pi4bZ%I(Ba)v8(-Y7_ohI
z+k~Jel|HOPesEiIOLK6M%J1`b@G#n$GDg`xQq5-g6CkP11`ZFN*b51bQzI{n@3wUk
z<d><dtij_nLk#FI)=pe0wIN)LeXXzUQ!RCg5#&veBC4AP&eWSrRvzs@&2_xv@X&<&
zY4SCSP{EA^7bC}$gTZ0}{b@6c)l(vtiMy-^W?)Y-5qn6O^<Ouvou-P>uUFre%S&!u
z%C@juZIU&zF?h$nnybE#w!cL74}1Y5PK@%`lryPI&c4q&!L7!UFOi0*#S}>)d6Dm8
z<4UH0sY{oA1~$!vMFyR4Ntr;k9+Q_KsL^J{|3OPOuc?76b)o}h0?BOY7@S460IQle
z{i+)rgo6vfNBpZi0)3G6m&JD6X!=nk9NF3iP8ZC(C489IKV3dlW2h)@LBNf^uZ_bQ
z@*sMMJ$U#!zD)w4@s^8=srD!5Gj|mHbZXQeIGrtWpNLt%BVurV3I58I<sgov8hHGC
z4hTV4%a6)kQ>}@=3pitCD^ZcGAV_mX7J&y@y>r2zO;KAa<=m<x8J4aYk?u?gc9B5n
z0zjMhY`7jUS(77%cWfKML5syD<%4C<h(jC`%6c(_SiO_p8RSFXL@5R>LJ=Sh;yS#h
z=wuBHTj~&Fs3kuhGd?f`F&p~8kERqS_kvjjur<SBIL)@0#E_!)EY&>KKPQ3OZr%NG
zi5@tx^M@90>jT3K0|@IR)o2B+Aikp07}NBn*$d6S74Mw-tFa?c-`cyw(LsZQ6}gta
z**C8RcHj@at@~0sV!9c@Z6$O+zMDY$L*5<Rw^eQIJHGVp@0%-Kwh@OqrdZI4_4zBo
zg4k~2$#SL2I}-F?U3Z2{)*3UI_!2E2+7JI@?Hr*`roPgKtEw}+f}QQGaIygo|5mE5
z-!FB4`=y7gAAGJ}3oL;Tw;Rp_3##%fW)MVU#-9?~ABr3f5Kw<&Rj^p!z(mR}TGy;j
zzX0_%_spKoy%JH816BwDH?-(S<V#gE@S&<&Wp^yxIVhQ9P=$0GEaW!<gF6HlHU2q9
z^M2Wxm(L~jd`9Ch1M`XXr~$T)O1}^RL_1~zsAG={KCZEJ#h2nbi1>Y)MYgi1ojNIC
z)pKk|_P3O|$|h$LjS6j8K};5{!>efZUJQtnIUY|#F<yz2fP16_hfDWk*q~pQC(M{0
zwItulZUdhsatagy?IdSJHqfBIK}5jO>veM*BG$XZ<0ZOsn|8V<3;yRu{;&1x2yO79
zM%7G}&PW^;>V*<{{>^&E?(rqzPQ~FNf0D(=nE!JgRpTeH;!LsapREJRoedp9gh&C!
zf5691)H0;%DMm4EFbnX$lAf!+YH3%H))uGFX%aQG?a+T!NM_HVGK7yDSM(I*e-bV#
zm$2!#?B5{*G7WR|R8|S_!$*2lg>r}y8U>+I9=nJa5RBHw5w<+vl{8#2sWLb9*~LKC
z`{n8-fAQm?=t2n8f}ma)I$<F3rB<vhEOCs0Nd{DX#49_uOc!iCy(}z9D6200gfS{A
zQA}!bHjh<}BpBo5T))ux{mv~WMIOOeerTWI!-ia%_IjMwR><158#SX`Otm);p)p;y
zg5oyM2Q>E-NJ0Y?-$C~A0(Ip8@N<8oE$n}QV!jH3P$g;5=xB5&(Lz<|X1sqw!|hb(
zc#X>8D0+>rCH)wOh`8%u6WLho(j%m&6(-^VkX5M~smNyr{D#tr-sHYPV0!a}gg(Op
z??&n}Yt${xMi4Hbir(*3Fj7bxm&R}3+p=1q7Lq_UPwMzOpw^uGz`j~i(zk@|t52X@
zz=&p8`XBEoLBG23cL@i6S8&W`h(bFfVFs;5<aJwg))WL-AU+<|R6y5u!u#VALR|s3
z%{oye$lHM=3&JG#zs^pWJs#RaEQxSw8r61MKG%C(vyp%T-WE=`*@;1v7s~U*_CC7I
zQ<Gyqy8fB0JU#RGd8ZNTiwN@5L0zWM6NbQ@Ms97h7T4<ar?#Zk1xLfke@}3M{PhL8
znY!!!FWtcOADA6xW=^L6t&_B*p>2CGg8s994w%KnZXU^^8KXzhZ@eS#bV*!v@dwH5
z%Olg;rx)cSZq_ao_4Vv+zhy=<;k5Xe#k4nNbv)MTam-}Y05`bZ8e?yzII#ssL`k45
z3D^krs}z6{6(S1UFu64tL9u_IxcYc#WtD;0NC>MV1%Yoon|i7B?dWdf%Vxoz){v1q
zQ3`Hk=l`vzr_;?BJ#_iJ70bWs`)zfya`R<+veVPqf0)^Oj?eyHk~n*Zkw9vw1aWZV
zf{yoo@&1z2(Z>7s6tut2m<N19TErS>AVfSCID9Np>qu75k5C5S0qLf7<9M66F=;@J
zcc^Lv<8K?oyH~5p4#zSPggc_CNoJtjvj_BtN09vSA9c#KX&~QN&OtEvHG3gH9yD9V
zvcZ43jme1YmA1eEXBO$KYW!4M+8OFJr?Sy0+Pvj~8E_stp(l3&lUe(l^IgdwX$1$W
zy!fVC=MP_c6r@*x<mRNd*tb5k&4OvR^;Mf@X)(S9!SyeT#;NaZUGK!&UuTud6OV80
zx}-t>*nyUF2)t)@$enfpjw2)m-VmSzj~N(;rPj{5+b5#iJesczjSbi2h@B2|hpHC%
z(~i=NabXy_)JU_4;QXz)NZ#7k*^wUNS{%}HL0I^jt&k!D24Prsqv2^18BJW@?@*Y?
zB+&9+4f%WrHxUv%>rN%XCXN5{j|AlyvlVD1zjv3AW6Ys;8{&)Ht!t%oJEN6?SlSX?
z?;CiLKe!q&aqvI!LaY+#v&4t5Y9@@lVDUA@1~8T|1+qsPIsLLzDs|&+s!46L5ZBY4
zMYwAS9#e(YSgLb>K|ybL8I;DN$EEmAACE{l0g;nQ(3W+(rd#BXca71KR~xO}rFLm<
z%pu_7`8_zPCR-So@rN=`N{UTivNjW<k^!XDiXgT@2RH<_OkZ0BeE@qO_KYV);aM9x
z-tViDSBv=Y(LXdr$wh6^5M?!F5iwnf_Kc<)QpL@+tAiwUdx5Hl_$O9&()~L>?DcCU
zP%dO8+N)noWE5|bVF|Qp7k;$jGLo_7VdeUD7oVhTawycLW$q}%m6S}REt46NobkR`
zW(;#&7tMUB4rQJ}p&a#m!oWg)@e0)4P^=xQ@mS1oE%0o3@K6EVLF~RptC?tk$g%|j
z)Vmx>qgXU`P{F}EWo>y;qH$NGEaNaEky1OR1o?n#g}z#k=I0f~@Q)~GZHo3w12F>&
z5!A%cHn|qZaS1r>wMx!JWl~_cU@jwqi#Tf+p9~D)k!(mWX93h~K8k}v(t0V@8x5@X
zgE%6N1R+i-(n1pC&=LO|nhsmM+$7>GXAB<tWkNmL6vU!ej+j$eJJ#ZE|LnGDOh!u~
z4H~2Z6PlEpNiI}>CUnk36x>hAzYx%X#VpuNARA`<w?c_gBVL6@aj+Zn6z}OuU2FAo
zt343TXWq2OK-HmeRtC$?^{>oSN-09?SKE6BwgTPS1#wqB^#MNhNdaFGrt8M89Hh-e
zq_u4Eub;gl9!XPvl_{!h6I7`^O2It)PTZxEF<0otIE%5_?MLuQv_gB3SPZ`8GbiLk
zs4?G}TyjxC!EncL`IMOU(y+<4!HuHCs{O{zaD#DTC`ER6MUJs^E}Bdp=Y^66?uHqQ
zdTDFUIF3Q8T07bIfu+goM$wMSEk5r)^k61Jb9NB7_y*mnxjB@MF!vIrNbI4g&jZNw
zf$I*ib~x0VC_32Ts;Vd<Qp=<UoB0OL#tV$P%0;i#GM@7o?uTZNKF;cFtI0PsKw7|A
z95nB*WpNF&_&%#5`q2VmVmI^UC7WgfCevgXJN~t|2alr&s5W$UnDoF_58jk2Wc4zx
zO?p4Gf4qd34VH-9)*IM7&{{c*1e@9RZ(iNOEo3S)Tyb)&x|_uxL<#}IULscGbR_@&
z3_*_QV8;+;!=K-tXhD-p1U;{jxHu10a;kt?Zev$dPYGJu-c<T@ud02%=R$Iyx~yAr
zpI2IUucq925H4CT&pTLPRy4Sse|Z2)+#Z$Lg%(_q*_3nZp^VH+w--85y{nyDHT(sB
zJ?O5Y6FUM7T<(_C#)1Hxwp`dOBt~XIFSpK?=8%!S%lpXv317mBN*%||d^0*=5y+Mt
ztqp}lcH(h8Ev6ft_TCv$b1E2=o6v<dC!W`{>(lo^O>vL{n!L*PoV8**`y#gr{-$c4
zdvp@hs84x3t~=~z2T7S!6mUjEOX~<eSpNpCed+Ac6=^eXfJaF}9r9exxS%YT%<F?u
zU9mhB&-GpV`~t<PQPQQ*@On4juD{0ci9Fl#Q=#@yK9d6(meU&~q~`^}kR}!}>^{<M
zuS*}6V8*mt7&%T;Jw6s~o(I)->2Le=jQ?ttPWA{r8>L9QQi>6A1&GmJhI^`C5~Lz(
zk(M_2X1sEk&6e<0dqT<uC;zKkj6%?hm55~@O#QLixU3m>7Hj3*gJ?4_&%F1x@y@$+
z+K5Q@$tZ;okzmAEo?zHzBS%`e%JpEftwV^-L4HA95%EH0c7j-Du|lryD<SkRgcv=@
ztXppN*q?V!oz_)22Qy0O#--)c%d}14wU&Sh2<zt+L6%tV<<lTAwsm;fLVBMK3#*U0
zerPzHqQhuI)GE9if4~e!^9*t;f7EMIoi)s#!bgxJJJd~34YHT2UXN$hU7#1uO_BB?
z6~^cB;NcBonkfImMKS!wrgjO{@8<}A{YJz1t4(V7_>G0paR}v$N_$yZ6rCL#+j6j3
z$+2)W-xb`EI%?Hf9=m@Qd{^PA%^5}kDMK+0R=@#LD}^9z37b!T0LO9z0xEE%r5~L!
z(U^MhcSeut9Q+*csH<kGy#Y&YAQLu&;Vju)f53jq?g6HsFYM>gt?X;_t%?<y9S!+G
zvk_zxGg<uEmV#G^m%8lIHhIx~awUNxg1Iu9SwPI<jOJ`d$YZrYAfrNanv8EfPr+Rm
z+(59!?O8bEJaj+rgL_&SGGVpOpJEXyrQWAV>RCBjJk>;Gptygsrsvv64ajI$y67px
z$VD(<McZ#3?mElzu=l*TB|hf6_zw}q{QnYBD(-eB00wzOOJyf(7zQ~26C)!GgP4Vb
zqZ5Gh7svl&&M>jDv;A*H){*8|3@$rT_XUlc482u?Ds4hvpC6_S9sn3_1Cj!iu#3_k
zSs2C?IIDMt#*#VbchPZ^Q(b)E=+e}v7Gb0@4OnV43to4=v5?(xQPg)g%Qhx@*WJDo
zDURMb?k|-t7(<Q~c4I>%^akVvRiJfJ!!mesvSA3@hFT#;Y}-5jbR7u_GG)JET#ehF
z7_x&Ua0$5ZC6)kjG}I(z#vG#{mjDGP5W$F|^}MlBl_hkyYgJ+<zAL-6R~;25gQh+N
zaT&@2CYEh(b$$Q6@<<q?Ne4Eq2nhd*p;<}6Xr=(upptE{7*%MmqYR1>qr(cYSSEGo
zMpCylfdSzfNnx~nuqS0Y3`JB$xb_SVN-_p;@>}XQu6%(CPC)}oV>VdGuz(D7X0-4n
zFc`7|17Ql2QgX1ihD=y*vK80<I^lbw{+Qarx~VWI(|BTzidvwJ;B2UbGU&tTwX?+{
z4Mj*4(=Zh>eSr#7RJ0SVNVs8e(Pb-pLDMEbi8ND4(}F^3e?@vu4ItdY+Q~6P;sVuh
z*~4>k!`ONUB;SHXFCo><wVAo0LnjEUA)O0l#@3;Thd7viF=Uc$3uACF1UX$&C>7nH
zxJ8f%>M}nmMGJ&ns6<lgZ0JP&rk+9bWg+A*2~!OLI--8-<#JI+Sq%%NqJcVztcGkP
zvwKPL`l5tmY!oA^D=`w%BX}raN$6_O@NhLVv&!rAcBpYAv)>ulQ?|W<kbmWvGBam#
zcWMZD*Fghdg?^o206e8N`Wpop=ExcCG17)pfG>ZD9>2`YO1NMC#(!0!PS?92UCBNF
zjjdVPQU=N;UcS6aoBwsPQCd*>U?T$izC#O=FP+fIq-40w`!lw|TfBbu2rcKrK;KIE
zySjEN`0O_iH~ZGbQ|ceblkW@Puir+yTe%miFKG2u>?`W)#}$~B*GKP`o{nz@-pzlq
zp0;W)Vsp27*L;{T&+*i--DqqLPPfQR3vVv7uF2*f_Jn+8jTr2^lb)5E`O7B)=}`qs
zNdl(fFR%UU#O)iHSa(4-P%H=}=(VqA8wH$TCd@;30NwHer+|ZX2%AbekB71rA|J>g
za(-kdO#;?-2%#5&(Df&zwpLw4Ke9E2l-nkzEyTUEF<gdX`!Pxa#fB98l=lst$2G33
zIaB|n1~SW8i84CwbVv_#rv=heQ0aOPK3<2ZCm8R#XnbPem8KZUS&F}gQny3DC!4ZR
zCgGD#{Y7jUe#u`G3j^G79RkJhVoKRNfDvAhY4W{rC%kuBSS`UN<=;n>@yD=}c{`3w
zV!yao>-R-IBE_3}SLeSBFTjnU;Mx#n=p%q<L9sYLzd2}m=u3k}9LDJ|32c99n|Crk
zk@&wW_ac`u*6GDpUN233T)h9AD4o))ydHbLKX`A@vbXI#*_zo}fYl>htVe_@2fMH{
zFda{MI)Dj>Z$9JvX|e0a{>lHHeqDNXQ7stl+t%Uk8{D4Tf_qUowftSC$ra8HXpp6Z
zt8WaVj)Qu|>fiALP2kzgmXgd5NuCAW1NaeC@3awYQmnr$)&Bb}({M;GkwPfe@Eb)+
zx@6p|q{NT3A7+e)(Rj-q^D)YY;`Ax@w;oo_=Wjg@Z!>vR%(|h8%A}r9SdE-KQ?7uF
z^5Bf}<P1<isCmDidlu>hJV(+B0p)~u4#DaJ#D{8b|3{5pq3+GaZ-_xFxA$8<bGet-
zx-4K+$PB#H8^6}Q9P@*Ra*YvP9Cb^V_bWaPF`GIeg}pAj(lI)-8XgWQgNx{#V}CwQ
z8p@;kKi%zbTcu}`9zyv{BoTVss8Z(EL8G}WsgS(&a8d`%0QRpYEiDeUqIzTmzaUVJ
zLx|{nw5-YddVCvS`Hh{pk?gxZ^{@l^`@<0m_fM)*LGZ4X(MQZO<!2_Hx&~?TgP<rJ
zcjAk<A>!rm+@E!a?^T;x11oEb)xJm922F=2!2*78ElgvAzim1qcMpBL-Olt=K=z_U
z#c&&V`w79mxi-=<S6HB}``HM~xrAf*lI;9nuY%ij*O>3xM_!Vi06L3$i*t$;qB!cU
zc^;$}_03BRJQc(zf1me{(r;n>f=|=gclZZAe*C27|M)Sn{N$PbzqbNR?Em}Iz1IIZ
z-R~AgX*4?a@*L*NW{9G&nj3-otir?8h-^%r?@KItU{bzsd`!h9nU>{~7MjQLgbUA1
zw&y-%3oq=~=M8VraJ_zBY%OKp;L)TJWl?GV&<%tz=eY{9sIHHPJE2_<-%~pYpT#3w
zmBq+%y&At%Y|b6nDr-u0=Eim=On-j;xxLk`8eXexPae%rr8`~}qI*AS5l@^b1xtHe
zIt?PUxm^Kj*(UHQGV^WTH|gT^Nkd`{22WzQ914+JDfygn`K^(w>o@mqi}hiTrS5+X
zI%*s&ofa!I7uH*gxfN$!9`aOrOR`JM6Gaj2SDZ^h(3%od;fw`GkE36HbpI3kIF{!S
zisna0JFT7m9r>yB(&;l<8SdKDOgwp><f1TAb>HxZSJU=iZ91#GYPW@+_e0Zfm5o@v
z^(A4>Kg*fcofo{SP9GaJn>sAq<z0MmxwOj+s;*6UTE3ff<2Kpz9_5)7GdF^45m+!J
zFmh!Mbe`QV+No(PM+Y9N^y{k4WhqM)YXR^n9qJb`Q5?l>E5Kx%?S|vv$N@LJ2_{|%
z^JAaQjENo}Di9Np6_A9Ia-@d{IH^OLYBI<O)7taOAETz9`TrCc&X1*C&6?M*F%dhL
zs(QUPJ;JCA(DvH6%Z@)nm3_#Okr#_8?tMbUUF`CwCxZ6MQ<?fd==igbQPJ4sp?4mG
zevZb{;f~b2b>j|};k$-?l!>MQIQ4fVwus{-Qb9?CfMbsLeNReWI@UnjJ$+)ZjmT;i
zL*aapdVkp+$MOaTO~c!L00#FU!X!gFmKKqSb$dpx1D+}6+<1w&t@xz@ooDdDKd-z$
zVazhEsf>8YA|FS61Q1|4tHoinZ#r&mk|Z4rlR=x1r39y?2;t#`%w2jBd9}BE<Dwe)
z^Pf|EW;OF(5C@hBT)KiDP;2OS1gL!tWPYL2(%}Kq9)dRyOjF@(*F_e1^(uPi*p4I2
zm%eQ9%)KeYtOe+XkW`t>1LXsO!?hPCY85lH?fS?l*b+JNAd6)yWp^t&7|KHz`5X1)
z5hDm5J=kMnkYYBCD1!XUWP6CgZDT9g9tuf8J@67iTSUq_pI6uCt(C&($U#zQkibJB
zJ3rx(ErzDFgINANH5jwV)H-L)*8~gRkI4P!^IW<`>A;HKNL=D%+(T33H@|=$=Q&H(
zWwmoEkl{qaPAM<hLtERLQklm?B{sLJ116va3bcG|<l8oBG=s1dX4N4;{Q)zT=4_Ci
z?w)UE;+$tEgn$582Wx}Oxn>>bz+!{}Iu7+Rwz3S_xLehlf7ys(W@{p>=0{K&qu}l6
z;vuBRq?>ufA@qqI!ObYSJek0&2W}Ywq#)QY7$F6xk|d2W$UTfvm<n3RQ*l+rA8Og?
z9dq>aXU2rCo$g<{@TKdfi1=O({+RnAYec3jw<84wV0qMSEzbf?K}GhqNbqjLCENpd
z$tep-RW+rV9DE5b6)jTBsyY_vZ>P09>|)D-H-g8yE)8T5N&UGnkpt*JcObKFgQ!!l
zj`x3Aca|moBxnE(B;p&Yhpy&+QFHKwZcW-VqyV$|mR*#1C&>m&?zp?6t(_p5BR9Oh
zvNs$%QPho#yftGLKa~PuhhmT1Ot+T51sl$PW$%_Zy1229J`<Uom&q>UrQs^58qQfe
z&iTxSe%U|}2f=dZAp5xx(&!k=9K+^FMsE~gu(HnS3+3lPR|xsQa@-!3TXG+R(OISq
zv0L3uci$_s`n?Oug4oWk$2fx{v!oQR*C&Ypbw_zVuJc`%l7uCiDo=N;c*mai)0f(#
z#{aS5807w=taM_Tk6#oC406B3xc&^&-hL<%`<%7g3EtP5i^ir(^R?m7P8+(PYjsAc
z6bPyY1XPSEbu}$%9cjM7k(TTb>&73uN%k6yrKC|X+x_Z9I@+&nj)q>umfw~&E>3)j
z=T}PlvCj<ZHozA4oP5by%0<fG@5P%q`o+A=K%EhqF;})5t5)<Xa_nYW2ZuLchR&(4
z^MP_qng@8!PgdIQ2WtHYU-9f$y_V94ydnyMN5*LJnT*Uw;QhOTd19IkB!sSma)CyW
z$y!~|5uL1_e93C5pvxaiD#LopYo>P;!T%0q6NAU6JX4P5ewnzK^0%IkKpbI(GLLZr
zJbn=QV>*lS=nKKXdshHgEGp36d@PUuKwx#btEDDh`Z!0q#>%{_bl@PzCJQz8uJcYj
z5blDa6+cx5OknTgHjrVnCQzp5uQRw;1y%OH(ydr0yen~m({9n<3HbsJ3>x#)a=uQ@
zofbFU)r6@>66opFZq2r20--OmTQ8NG+I@jYf+Z#R0sLxP9`!z#rd)fU<!841kc1AC
zRHAKV{#M^wmpLbhIp1<L-!{JV2_aNN9jhzQ4!7F^YAxx)`38Zm<iy{@h{OEJPX+(#
zPDAcCJigaLP(8u}0oh7xTN|9q*2*>WM6b_P>P|7u<>R8$B!1RZ-19g|DV?i@^vU|=
z1mCB(Q~))5KpHjowwKb~7*?)ih{~EQzHd#&7WrwPlZMFJw)dp*=WNgEm*KxlX)!$t
zt4rw;LZ)@6A@g_S!l@Xn!5F!OsE7i4au3^mbN1;je!5aHBzZ)hfEvt2w-Uk)C>X*(
z848N8+u81%+SF3FyGD$i0@9aS7qgK45yWB0P_3yn6{8Wn?Gv^T{0&WPAz)8>Wda=W
zqmCJyeoY4>nx@~=*$4_w)>*yGeAqMg;)9UMhXpI5OE$DsfcV$$K^Sxe?uU~|>fkQ(
z;Uhuxn{WOh6|c`?BJ|w1PhuEr@TF@Rf#80rldy*tEPno#${L;-1v`a+jtbrBlev}S
zl6BhFK--~HEg>3C(LCnn?;D8ZOOVwM2J6bdga0LFnExyE%Etb`E{$DjZ8>hVA^Xng
z<%@ev4;b7<pXr!6SX^i5WK0a%Y^6(62yY}+i<LMOWU;b;9&`c(Q1SqCE}OBTeBs3Y
z!-o96oj}>TQ>T}s6g#<mZ_kkZW+@uOi<eGJ2bSc)#43N+&W<iq12ml<TNhGyzJM0R
z2K|@^+69TP%FKMd-j5&OHVQct(MsbA+f|y}o*(pLfb!hlpNAJ#W^z7ieLXyiMD2fG
z^*+D(yEyv!cfQ{qhi7W#H<jeg^$}J?=;3P0gR~v5F;rnp9ZSON#~D5!!!Guf+jWvp
zD#sJ8Ql3QJjDyDd`0eB)xwVq)M&b*ICIcv(lx^52d|&(FG+5JM3zDy%-`&@AQZuhU
zWWAFMY3YEsV2|;Wos^R+3)>5t%4}mlg*7U@`80oTC2=6|G>hOSX^2Nm_fd)cgjxkO
zHvvb4l^yFz+O6^i5%Zs?`<D_G0}m%VzSQg!j1nX#8-Kpv8#>nU{HZD`?sAu5zYUOo
z>($sgWCr%ZDJNkeQHeiY?cp;{g*Ve?t1PGb934+g$p$O6-@-lJ@YiG>xxAHUaPPAK
z&;SWxDPT{lN*z_(isR?SJJr+~kq+ntMoY)+U~qK6vNYtY)Z~rDX^?RV0^_4M9c;a?
zq0Zef@+bb;8zmi&@)PUdHtZ%_w9A>9*MlbQbf3ydLDV$?0}R~&E+UBn?E7l0`aE99
z-tCkpjNoNz%<NsS=APHtxVrDfQp%1FsF=R%V@0RK;!^>jA*nIq8AH4S&<jH-1JZQX
z5AbYk>+earr41UTUY%cM6%QL%4!VRN>N7PfKXVJW!{46DHd0NAmW}_aodWOQ^DU8L
zg}%Vh%{}@L<!S@@MUt}fqp2uj>aWQ2Bgck@4O2dm7Ht#TOuFLeF^+~^x-FX1-q~BR
zcB0HsqJhN&>8Dq!7xoj)fK?6`XF11jH4$++=PUXYCd(;*(?VQ{C!F)3VgXO1LdcP4
zVc)V9DdW+Es_lh9kt9%=p;k&r6)~Mr^NF}CI?eUp@IV8<_C9??s$nAy=s^+FCUgiU
zs*a<^vKCv+syLb}1A&10$&`2DnCh1;%uWl4CqwW?YgBaWWgSMRJ~2t6=8M6}{lx++
z5u?8ziV@;7beSe!RSaN+r?N=hNhu;8g-mCLqD+%C%W-014S*hi4^hc7WNLXHd+4N9
z1c*HfUsIeVp@dj@g^4{4!J)9w4crc0MxlZmEZ#Uu&#&nqPh4iBO>j^Mymg~ZkV3fe
zW;muK6X~C#UtR>ajq)i;g-8rZxgr;&qY;yn^pT=SI}DI>CR3t!8zt^KK#K84g*Uq%
zT_K}c`z<PpVN)obg9H#QnlOt~C898b(+O+A&y~oo(Tmr7-aL0~9-GiZjWEeXLrFDz
zq@DgikAEqN4Q?ySIf4{k*Dtsv^%)9YrXdUA7+J#^cN#ERW1@<%Q>I2E6E?54;m5(F
zg0RyQp3g#0q#dTqfhPnnrG+r<VM{pI>$rixM@1k4w{s?+EKsbZCF2=sz!Bl_6;X>t
z#RR$Y^O~?ns)ifhD?XA?JRXRMAVL$M8}DaOP&iE{esWRa$mAJUk=X^OI-CpH5fBe#
zPQ7>W=+&fbFl6H{%YK{LYwKNx1}iR)%7kn}^7H?El5~FDX+-`54<&Df`N62ov^>WD
z8OGf^P2Ne6ib(g3qx2|76)rF(!$~OHWNFWvZQm(NoRVB3k)<x)In9UPy<T4__A+Un
zYBexj?%WA=)>iun2tpikD1bcuJ_b5{E0qXG5AWA<8sQ?pvJu&2dP1VUjB<hE$JnCy
zKnR2+vEfXn7(vjOZoQB0J1`vFFN9Xty4Tirba>#b{ayd{`a&!bJM?tq2}LE!<anlQ
zQ3H@SGm>6{2w3o8%g^@pSMe&6@%M8V)sXL=!dW$pSKBgLCB}Zz3?KDYkzy(B(-ifK
zb+(YrUQ5+2m8TZYYk-LAi+npZ2M9U-KZ%QZdmWt?(MdQ|3^jBt1>Waw27<bUegVrb
z1r2wsjeoqDFUA{?|12Mg2;jZb@E@7XMqB_qY@%QpkH`(5g=63Obp)u~f2lMO36?W0
zvPqOczIyW<Khnq!o6^jQ^Si>S42@~?h&&Ap*y@V4`+mT3f49*DUt<=muBA&Z|0MQo
zC*$g;6pO4f-kR8uATV?=hiVY4>cA-}A}GHmLWls;Vvs^J;aJVn;SOmZ&nK^zp_hTJ
z#MV}<T}C4`x>gRdw~-52CUK+omo3V8`SsL^0`Vk<VJ09(7Q(;jB(Eo8#0u}ilEl?t
z9T51a?Bya;kA~mzPz!ypw2J4<Zm06!m76W731POfdX}=YMLLg<1!)J76Nv65(zFkG
z5NT{y;Bem&!i^1Ak%)9xQMbT}FWiMLc94R^C6e<?Z;a81jC-nCPBd05U2B4uH-^K$
zWpL`4yfQDZ(>}NGCnH9X5BfT^?5F)hYIudxm)@xt&rYU1oI9_8k^_a=CpxNHp$`Sa
zq5Si+xcl(-Yf0*H#TCr2RZhPR9=0>_k0?(Z=WJe_0cUX|zT-Fnth~E;tiLK0Md1n{
zVmCM3^*!G)ky6X(tO{FR$xA83=3J{Xt4Eh29J<wMvJQ1VKp<a_<Idy;w(s)EWxYVH
z5&jgl+w&cT`e0$<^F)s_|IDEI!*PP`SO(PAT*B95xJZA*S5O%=s6U1BiWM67*4y59
zGws^BnL*!fD6f)u1W6X&`|l!XWb$gIVZOB-$%Wa4Hp2-Q$rPop0*SAnV*kU9)gOpK
zOV|bx>?Rh#UM9l6JwQ<Nhi09A)lX(o^|E48SS)p9=v>YMMV2v~>e6~c%yd`_3SSzT
z*<}6to^?H10~L)tx(o~VkcX8LI^3njEuB+Fmw9ej1#F&~VmRV+u8@Xn2Gujxi;(J+
z25!FDQU}nYSZdD{md7p_4h(OQCdTEQy4mmHz@5uoZa%!m)4&~Ii7UHwxI`zabZ?s?
zECG{|KD|0&R$a@xZtzb}Wa`<1_8Sp?dxJt*F~xrH+q~K}Tz<VpvRzx4-}|vCCXRYv
zWbkCW&J;w_3h2hcS=C?`9?L6lKVmZZR9-Lgj6U8>v+sN2Q*1QIbsoxIOHFoVM(n1C
zz^(i?7~hN%(pA9Hi|T+<^67ndYybPZ1K2esot0%DR<U+F_bt@-JvaPA`R8%->^UhZ
zs8icoZoszk>S_c!FnBzW3B!-<fi2t<xu@?`3%v)+;}6HA-jcp<FVYnS6Xpxq`F1<>
zCwQFqN_^XpOze^RWFzXt9<L=ptOU1PNV|&VHxWmEn%v-U`73i;Qva8+C-$VA9>gG0
z6m}>DYw2}>f2T6VJm=n~>~$iEPk3iuL!!}u<1O}@-uJ-h*@d2t5xF9*0pyjX&^88m
z^f96Q_{Z_X0>w8-$ZQkPe@RT1|0o6iAZh>C5_3vZ`$uB_Pa0;okQJAB2tP}X(4}*2
zaYJrh%jPrg(%3xQrHMxIcX$LlOZL~Z50C_M9?@9Nl+!yYG7kqw>y1tyapz3L<<+q3
z3TCZuo@jb<k&2ATiYvByI$8vpkm*X~<!4ZGP@)CBDpR}HOD(&4Bl|+un4yCs=TAIF
zUAcy0#b@thV(n?^$o0Fc=if%>nfwZW&zInBTy0!Y9L%VTRWn%!C6^=sLXH)nK9F#d
zl5@d4kxpKs;<M8F+%>c0pt@CK>3eCjHrd%5bvuwNL-o?tcBG0l?Ute}tRcx;Xsuh#
zLsdu%FbP0wO1-mh|4^~sh<%XHRPFvaniyUBGO$nO@(*B7ZP<Cuj~ni=Y@OICX{U}D
z)8pTM<^O)>zjk`5UF*H~HN3~>m<>XvDq`U%{#)jp#F!xg>{)Gps=9UE3QtQkRP<>q
zOMGbD-2Ot7$=y5KjhvWSj)q#B=i6A8xz+Hwzq4)K!W%g1!Ch>cAly^YzEfLKP4T8|
zhA!`_KD9CRY^OFZ*1g$5DmJI#_9wrngLYO;8ib$g{p08ar^6wi=%pLAuEy1+pF{O3
zveU-BUNbi#R7y0+5c$K@R|}^Cw!oyR*b?JwwQ0CYKX5q9EqnO#`?v84{)ZWREz)kB
z0V6J49u`&k^NBNXqC%81@whM&Nt#jzR>7;k>E7WM4-ajoCt~qi$y&4O!h5M@y+%u2
zo3G`?m4|poqS5Sl092J*S$%EPn!*8Avkp@Acwya^P^AhG8dRTBaNZ|xAFaqP6}43z
zhj@2&riXSjf=kJ0ZuqW|7o*IX@dY8T)>P3we_sV$ou$59klM_>{hH7HH!)?9R}M6I
z4Jnh1_&#CaKpy?APML-oECgm3lOL><Q$EiaavCMUfJi+sx!Zw$fF2@h*CV6-pV7P7
zKclPxlJ^=>B)3?Hro{lfdc3Q5VP8{<!xByJJL%SJ;}6~LTbc>eC?gxG(fNAIR@Fw~
z4qa`J$S~0*HT=sAo3kF+{Oh|F-Z*FhE{20Qm*({E25Asx;KGwVCd$cG2`Z{$V_-&X
zimb*uQ7Cqk8%Hf5#?+}3kIoB?oz*}@MJqPl>)pl;@6Juux#!c%ewu)!5FzYn+EuED
zm#gEJ4&55By#^jcRX9Y~57TWHeoZY7!_=4bfA=<wCUyCRV%i0>jH29M^R|o<)Iqp3
z*;5Qb5&3&}tU8TxR1zE)lH1uSsKXE@)LgHszKrq`mM}^w%-gf1WNa}2j3gLdrI0a7
z^^z!3D3$^WDeYpo&OSt_1eRnB|B&c-l*halsWSLL6iIQuPd@^(@*hbA8>sad#bMVg
z;DYT=WG3|@d84trnj<KQYMYIJ7s~Vq?(fK!6`4>{elJ7WVpmY+v5(laSX0_FF>4&U
ztxqKfqd>^{WZu%Wt3vWYYn{`u_CDdE3wr1H@wf9a<SBHj#+Ko>sjM1rs-c>@lX!z)
zHXmFVnwr;lr)r%C0}ttV=F87(H~TRuXv#y>c0SE?==8qm&+Hb)IgNYB*Y6bxRxyyl
zMZ1(ignL(iC0lK_ka>N0iyb1}&}uOlh60g)x$>aF7$@)+9tpY17I62;q2U#h=VFk;
zy<7<~kN$=!i;xgFhjy=_qT^$Ps3-kO<{<hG)}<KIHwnYTnHQ0I3<#8%f7Ocx8uJw#
z8sFacKjlSR`Z@`%&uC?zPbpW@tMf*rTE|T>zi$iO;3{*Hl|*73SP%_?9NC(jp5d&~
zAz{R?WJIL?nn-}vmU13_8q~w<I~i>NSIDgjD6=c_4nky55*gw~HE_8}rj6kUKEjV~
zB!&E1MDGCYn;EK`w^zaT+%nLQR)ogUZeApx0Jn6*UUe>_;^G{yikx*M+R9cyN0_}2
zZw-B*r4f^^Gpw-^|B7MLUj|NS2n2aF|MvF?B$$Qz1)>#(B7)sz3>$$*G|VF^MKJrB
z)qqO@-h_d5H-*)q$ZDA$)l!d;#dtXh>LZD-->4CqSxD}`l^_+={t}L2kaX&Y^p;w$
z!WGqQQh;aDkS1#!Ya2uAp~iPs^;qw5Cp&NoM8aLfm4RWD1go4g$m7l@udva~`D(rm
zT>W)Lggm8^?;e+)jekUxr^$r{2)_6xSvLl?IH_a_&v5j2k>Mkee5*fj%i*&Vil9~v
z(^2sVhEf`OoLja@RFyC~pPw<W39aDDJZ6Tw?+-^a3K+=(feZ_ZpB@#~xKh0HLcw!;
zGC*#hoA`B4Ebk_UO5}Vge4n+(aDu`@34+JM7IdDGelrL?-|9zqeCWK@RG#VAl?w=N
zbWNq)T$=V(dXc(LscM~Vc>8z9bRI>#jgl-7dP;?$F41`O*}=uHo%huM;z!jf9Wr5h
zU2C1i)lU0XR9#m<@zU&R`sK0G<7Ri!PiuvV*jPvA+$vIPBaUV%MP!5wlKpwqx<u16
z(W3PpgVOD95?9!K;N6y`gzt!h>v8rNOl*z@_gDZMwa<AMWThGQTHvs(ChKKe$Bip;
z4xWojY((%Dr=RraIL1rdZB9LiMNvHPLTb_XBx24LY->lY;)^@r+6GPaTTHvUgNA@7
zik{l}RZ;MT5M70}6nbmQLW7^VG+IE&BGpV@FYv`bdO7M3>J8UJJusOdalW{A2Lpo4
z#C3+M>q?b|oml^kV=qS%6fEmA^0#9yo`(N#&3U)PRy;Wbx`<I5UTei(q9=cUjFyD@
zRBwGd`xG2>h39dW+mXI%D&Msiknx^l_8tms`r)pifqUN=sqh%>@o#knKFPY%nlIRf
zU?H6U5+kht#mE2uQSzSB`oG1<OzjV<?2mzcuZ$V@fo!8Th4nAgPFb_%Z2a{Ct6$8?
z=q9CWdOLxzl71j$I|*lrP$-Zb94j|?Q1Uh8Zubw7gfqXrzERCk1EY!@99ps%IgxdU
zCx;KMnBJVFqH{h!E}vfbk*x|Wz?6s5LPR&KQhj-ST%XTBt>nNVP*^iz!%G=_-%g^C
zn0bAC9sWHZ(R@GMjIs=un%&5M9gX^Oat8OuesgfY9bJ`gor5Dx!gG*7DZ9l|3^7Tv
z5^u^(+vtB}8g`a{;l7eu75Gaqi=$B%iOHNLod;52ZHP#c21!WPRCG2#^dzUzk5^7b
zm04*1`Wp5)^G&_g8oU^OHzgDwvI~NjU#Ycyc@U`=Qi@9|E5r0T`5*5II-%`x@b<SH
ze|4rMY}O3s5$8x#O^XSjjaIFAtNvhI3In!i@5Rp;TW~XO8Az_OGLv|adliJ`swg;d
z@_nJSqUV0RS7&9arL5IQg%%@5o_{?hGZP}Xx*u<Ad6S+i6-;rG!Z8knZIQu;$H>w{
zU8Qp05M9l@;+>gCNksG-?PmmKKxl{ojW8VrhcKOltaCx-Y0EaZ^0;Hw%6FsBpSki+
zRFNWGKPQHpGDhW^jv-tH-#wxM|I6hN)WcsN7N9FTqIiRT5Fh#r2W5_GOK6Xm-ah?y
zi%n_p5&G9f$BW)=^vnv9Hy!KR&qVe=_>(wL*_wl)o*8T;c(h3+|L~dri1qVAvTl1A
z9Bfxq=NlS3k%A+1Dk3X0@;1F-r$Z!$+HN0i9qg3}vI;7dc1CG>3}Fr9K$|Fa@b{Nw
z^_S@VyT0&1&!NZTxAfRZ+5=L(+Oc96-J)H|%zPd!SjG5C<h|^(=L}4fkqA@iC&k$Q
z^6dp@0L|y99(Xlu?)k^DuZx|_aj4^l)aD<9WDJHSbYq+(IG$&%C5T5RlBjnA76_Z=
zqy|glTle9j2Frk<3M$pvwxxFQ{<qh|z593a)F^|`W6dg>KK)V2dtqIo7%Q1}J@l=@
zt0Eo~KJCrr`W1abGEoGAx{G6$XWz{-s(USHB&>Mc@}030SUg_8Jn%64mp0Stss+cL
zC%ND{_3C0n+hz@WBPXhHhZQ0H@^b1`;=vf-Ca`dI=SKZQlChX;TY7=A#7;uO*;oU6
z>|bkbl28DxKsK~+Ddc?W0FodiYkz|H3sanVkdy^NkqrbDTw?z(KA+%&5U4$IMXgDw
zEZp66k+a)|CAF{&Bn=~JWI?*A9%ba>xB^n$Pg~yL03ac}U;!ztTQTE0n0A-%z3_gC
z%KrJ#Ah_*_<}OTbl-IB|&OBHtbb(qYlT2tDm4d=<9{Vs!;qiqOAz(1@)aE^eds?L`
z=|+BIOmL&*hHPeP^K4Y7jsyWTf&$GgLmZ5Sb&*IW(yMis<y=Q?LqF!f3f>IeN}$@f
zZ-8!QLwx^IR>B5mQaFP*C+rvv@eY(CaUiuB5@88-FeH20zru5Z)M1ofqom=|l;aS?
z9%<Uxpg2frnYf0Zj8BIsU?H=F%&CaX)<QPRz(>C^H>&FDMPLk9vp}3w@9mja!UOPy
z!b`G^LL3tYp6gXvij4m16C2lYFkNXN4dIu&N4acR+}-2FEXa-*oB2awTm@ML{a=jz
zLv$zo8}$nZzu2~I+qP}nwr!)sj&0j^$9B@OolZ`k_dRFy+=G8js%EvmgSysUSMB}T
zdyejRng;TkCx)!lX;SDupsp<SoG|B*DOG0WI_JNa0?(>f9y(HtP6u6#ldw0d=%9tR
z9tVBvEALOi*1~3*$|dm?k*?HIGx+}gpmfu6wzyv1dC0<IpjKkhz@UlMUr;pzjc1a2
zH>(+<W31Ke`kH4oIkpYGbPVDJM^K>^;cvylCOrkX5B7R^$vjy~EjA{y-l=34e&mS|
z4^{W}9GPlPYp^ss|KwS_nW_$Mp>nV{t-oU-cr|<q<XX%4uho9`cgCa)=*)^}VyICF
zLIl+lZFqU1PM0ZMpu?R>lw_C55zPL*^T3+XPsp%HY-^+02Hw>AsBlsSaA;r6NSv`S
z5G!tDY%QR0p~IZrm<*p%!f;PqyBnjM8<!8OM0&Sz!7ELrSc1%0+sG63WW{87Stry0
zp@EPYK{h|Nf>+m;FJ1djl2j-$KesOGt;{fek@aRiSdn)SRm~d3xOO}KHL=ak;CF7Y
z=(c&(B1K;GpUvBI;$^vR8i!er9j+r0etSCta6uxVU>>u;6IZy19BCTAH~|w#iTESl
z1eu>EfHJZRGZsvrEAZ2I&nWY4-2~~cKkH+@Zt9Pz@4n)f;$y+(grV3?U>etQ>BvRJ
znRV!$$N+Fh0{Gvy8sOM4U{TykV9Q$YaVtDstTS}KxJo@<y{iMA)$E9))EW*5^&i4u
zROq`a#6s&Sj5^!aN@yz=3OVxi;if5lq7q4++q<`hqiFCRB8Snz#g>%qJ$RUkAWzVY
z83Q^L@!}Wz)PS8}sJ2CYs0lNzg=E>>mQ{_Zp84RTRm3(E33mRIWf}1CB50DK0(WYD
ze%(}}+cT*{xf7Wg)xHW@1wPe;QvLbGZK#9}J<EzaIaXdTJ^d<1TF+vAon05(yR0wG
z#3>^IJe}Ms#~kwh^R)@XyAiJMWbrheOYIo2@GWL6X>wEn__(vB$q?imQ)sAZO<f`+
z#C&@2&ms<OR;(9>)gp&j2MVQrhh5yaO4PB=9FcDgHh8-H$~bnuwKaz&cym?&oD_qe
zGA72nr*k{qZrzfBtg5rC0d@_7bPyP40NW6T2xybc{SXpzW{1fyw|*j@#kTUhs=gzo
zxOz^k49`aM0Kd0q0y}tKkEeKb?RWVxQ{p+FbF+$&#5<mn0gWJsZRzguqbM+(iAlxg
z%5HXVW*GIPBMsN)bs*HnQ;Mrki~H<5G5Z}{J(rJFZH>ZP1%Z#akAg{qtiPu&At-#*
zjeHWt3y5oZI(RszZHDmrR>k{a<bvb^-es3BPaI?8w<**;&KUs6mlPsy5+b-Rs?}qV
zcg};`LpYS?kWe&QG%E^DkjTLk9B7P%k^?lX_?s1^r%|dIVKR`whpQ^chN`ejOZ@mT
z<(oljccqD8Q$BA5acO6+M*VB3v^+~t)1LRZ^zN`%CEF%J$sDnO3gJm|N;+}8$BJD#
zBwyOO{zYsC)Z1XG^%JOk7&p0qv_QN$!|@K+LoX*sARx$;XPN5wp}EM=A;aJx4Po2d
z32ZX)#pom`BNbK=KL{v`ak0f+Q2NO``|Tr&^?Fy+-h0=INZd`M-`=pN>+K&qt>@8U
zuLC*~OYT??F;7pQX)DPtY@@=z5;i98bHKWT9Yy)m`4NKN<Uh4J8^3jZo|b?^O8ZU$
z*H5JrpUWS6(I>)BmxShjD#HJ&C}98JnN>6CIu1qRsAIbiRGt7UzF!|=e)smddTpEA
zlcjOqDy&RvLNP>OIj$xCulSinzz`Nit(EV?j+rbE0v59l@SJO4oSr_<-26O!glhnB
zEGaWY2T*d*0kT0|fD`g3vT_J#w7Kq~<J<n&h;>h-`S9MYbB8~NftYMq1B-8u?pW3F
z1;N4b7(!jn*-Tf$4aS8-fV0D{M0oha@@JP=w<cXco<wxS)zRO!*X#$F*>K`e!BCg^
zzigTAf3Th#6Y@_n!eP5DWDP}0Va!#D?22_vx7UcXl!At?3f5S)w3h5Ggf)K;VBu<8
z#K)2k#5xYwd(<drcXg!B59m%Jk@t*Ow|8~Uo#W`RX=G<JD_kh#*WGWL^0dGh;X09c
z?;14p`)sPN6D-IY<M#G-x`qgB7op6@MGw(%U@904L7NgM0gR>%F>n)~O*@Dr6@PLD
zYp!YR5tVXK97ZlFokIH8TfetDP5FF-UjI}+Gh`$o7Qn1vB{!EhoizEZr>D8hPJJ{-
zN5WeWuAp~`*wb(v$IT^r)MtbHmLjtu*1;Bva^Bg6=Y_{41s%HM2a`-ZuGwU4m*^oW
zB6eS-mRMcE(i<Y%_i$0Y%s~+=lhm$x{yVq@QxZif#>DbjCw4-XFNW!DK8u$8%K{Fb
z-w@<{o&XL0G<srHWtz(WgVSEJJFBuF<`@6XNDz%pZBtoEI=`+U<d4?Q*s0cDBuPZV
zX;_TJwltH|);5wm<Lr}LRx5a3z4l53wo?lD=~#Y%NeY-g+$T*0w;K6&Ata)W9r?x_
zS7*Y590i$p#?>A;e&!PX4QN(3NwBho&}kN{JX@<MTS+9b9+)>9>CVxHixaEAdrw({
zXGJPXN|#9Dv@|+Jp$cONNo|GpF(y#$YT_MC#R|_<wnu?*_HZtX3e%Tj`v=m1O+Vx>
z{bz#C!J*0c6n$5aCG>f%og<r63zX~SDP8@;8{2sV$;^2Aq!0X!agXU_o+piU&RJi`
zRK*M^fF(3`<9e4ItJaN=lN}yT?_<L=8{Nt*o8ElgV9Ht(!O*{ZAk9K2{2D<FZXTTE
zHyy!3_+4(8u+3-Ti<Zmbv;>~qyV|nMx@D0es(;QBOg2^oKKUqrZ(44#aBMjuC^f+A
zr_9CZsNr*h<#>0hOq^)+RdbFNZio=mXnrhVtM{DaiF%6orO2l4RIzLV#TDuiv0b;5
zjQP81UA47*|7*J}iiNbbA+KpL+r3x+^Wp|V5DC)tG;-*od&`$QXb-^2Kyo{w@JP)6
z0TYyl<o;iyfsN_^c6u_g{l5u=|DnA^ExrqSP!Lh<!*?WAsv%{4-J`-%lqP0GoR&DI
zkq}zyGvEPn;CwL5<(jM=bfEUPL#>GthY$#g_7wO5T(QMcwGdXsCE76H^nT4<gltW*
zLqILVQ<E@mmWNA7I9cRI5HrSo8cVw*HR98|H34k<$p7#C6QA2Z)3@z%oEk#IdfQVN
zX(LDM;dX)Z$3hj+fO&Rw0`(F`v;RCS+kYJP{~Lq<!(mwzB{dio9Z#A2g+PH!D8css
z^Ckb~on`y!i1`0q|0mL6>uTmq$RKWO<Z32rX5wIK2F=e8?c(ZeW@HELxiO<HlW@q6
z+&foys|44K#V_U@#Ag~ZZOS1D-vxI`)C9TI+F27zhn2jQ@b5`SOc`O)y2IiR%bJKt
zvncw6@v?>iT6XQVhU?QlUM}yq`@?nDOL|Qvc@9-aXNOLdD_I=b<oWW9`;!haWyky6
zjKDil)$U5Hbl#_`ZJU=@%SkkO-%rf+oT$0VuGhhF+~cN1Q_EN-uh8%Zu?FQjqAr}2
z+K{G7B?UL_66n<np(Q*?x_cVYZQJy|{{1CMRhR9qx2z#cvDy7x*LBvgEthZwDu0tL
zc9{Pw09Pnbs&ezPbqzyK$Ceya6@*z61f{)&rD8PZMm-O?VnkY63c$NP8KAdQy)?Y$
zDeAU)^dz!^uC}hrAUB3dGA)1NgD-Ab$m&9Zx}RFJ5-UmbZy;@I${1Iytl_VXVyI+F
z^%i}tG`6tHo0hiWB(l{3QP(x>zHSz~&4ZblV;=Hrj2Umw@7OsdFbvAI*~$Ya4O@5r
zsBGHmC-#@Qu0{U@^-0<_vJI<W4k8H_-?koH@_6g~wt|krX-^NfmCic9$xJQ7q{(y`
zXMeK4To;<W;3H9rwCEm};(yVKL@^y%>JWiVa+vv*#ZHH%2U{+={qo|1gkmk4M9bXS
z&Y#D&e8cPgJKH$L%*ndh8@RngkX@jg(;Xjc?;^Xjz)C}z`2G-f87j#lmQqN#3{<rz
zlB%6t#iLBW$vpb_XTGypP+gWxbYEeX%$0(Mva(leYx3B}{n4Kbp>$PWeClvT1RtgP
zDVrXkt5(XcGeiw`X7s|6A=kK2Ek7#T5-Zyh=`Zprogn%^-EkEZj$$JrG;-u-eoAhA
zV`2bS;F!aVG*eFd`eg^<yH7|B;Qe%+KEF(C^V+7Z0vxoUa_3#!n>Q6LJ3+CBWOC!X
zT$Rn_$A1{BGy{Z~<-U5RrbaTS1P#PT3-i*;FxE64JYkVRatH|9=q6R%tZRFu#Wz|^
za!RqEXa4P}ge-;7Tpt2y6z-0Ftp@;YhuHh!h_ZLwpo4aPXak_Wt$S#%`v=(Mx9fvc
z2U5GLaqeMg`2lH0C4<!UVuu><BI-9Z3Zomhw1(4uxo<%#d^Wc71;jthE~<8mzlyoq
z&gL=El=UtJD#o6KuGgBYITxgJjG@L;Vs9gv;z@KBnt#HY%w>@{Z#-pTY8PLnSiU{S
zpD9%7Adp}8W}RQbC7SQx*G~3Bcn~bx+x1=Rvb)8yk%+AtPqn&ntJQ<s6kMG>)~(6*
zOjyGQLS+|2BPfVWlgObAEmrC^$LE@CmIFon2s`@fTT_vUD74`_Jz+P5=Kj)(#8$JU
zi83+a#<m^Vek1}f^dSgT@!aJ8joon|76grJdrFg%&wd6-n5FQ#)_tyuoWS4q0{y1>
z>asLNe24-+ieGt3PI~(I4P<F~cqlYr1131R@2Smhr)o&8NxN5bwsGWHa|L=#-~w3}
ziG}+y6{1^A5l9@5&$~l@rpc&+G7JO{!%9mynr3#bX%;Q1)V8J?FMG)1cv;x}($`ZU
zNK!b@iTOYyc`O{g+=F*`+d{D3oaXUS^L-_V-f2DsVFel5=6Ur1Z$9n_)c&f~fSPQ#
z{FEhA6bKAN0es&SE(zIw;DgV+HnX@@0nNy|LU3`2OH<JQV>j@l-Pq2-Es8|r&3kf+
z-x8V<6o6fv^!NZ2<ieveG*XURab-@GzZh6N1z-QsdHojA4>3~IKN_Su8a<ziY|EQ1
z9c}6<CbNET`eJoi*$WRrn!V>;N_>By1iGuuIKaaX4h9=^Flx)HXGQ!rc%Z-#4a#Wm
z-+zm$o3&NQzNL0-ba6mRe0;|7U$~<63EREf|Lvc*jtDT>@pHJSw#6NPJ@&e}SpGHY
z8(rGd$UhCKPKpPqQ5Kzq5?@k7B$bBho3G;eBaxP>S=B5xGo_8{K%8OtFe~EKhJpPf
zdJWdW`ZzI`W8i-zT#2HE)7)^?O33Ob7YBP{Q=gx(PpeDm;<9BU#(*Kvm+|QRYvy@?
z4x?bQA%ya{%0=LW*c^dhIMA>72w=#agf9;bO_(T^69S1bAQ8GYxU*HxUI{5Q8Thi&
zs|RL+yPHb-!%_G)FVa3L_%Xk)qrL()3{narH=FEV$@m)B!C9Bm-Vr8hCuSGub|33>
zCkMUzT2*qGl?V@pgm?lu)_ujuEYjjNA>pp$$VOu&b8L2ha^lA-ug&|rp?}Ha^}<`a
z^=cjmvt=Z&^&b#;_&o*ToVXfN{{7v(d)kXvkbgEm*QTcqry!}@rVmuP(lhz|arO$%
zPGe+qlT_9<lZgm!Xy~zKG=EJycO0Ce>UPFkY!+5$J)glMqsU<43w5Gm>~&Lk5#bwB
z>=>^ubhO|%=8e9JC5Pe2=Zx5|zbAlHy$W+lO04mqZ`BXfhgZcN@d&x3h(V8Sh1H)$
zPpq?h^~8D_n<wSNwE3nzZ!Qrf7tLWUm?!6v;n)Pjxah_FI$p8$2u1|nn%B~chTG+%
z=Zcq?AQV2AvaUD=nt&&w<uc7NK&MH`8h6~bt=$=n-maH6XN0{NTwW`$rGBOO2;zx<
zu20NmK65Vy7UH2h0+P?5r(op)ffa|PHg9q9hAbC2W^zrxEhF?hpO(%>IA~A<Ro;^z
zs-4U;>$AspgLg9?Izn%`7C&{NqG_?@Cv>Rvx3@EY8^&3rA+sy4L6?&p3o7rp_@6SW
zqY3-^O$1DnA$i@5E4qQMK0ngFeX<>Ib1!3VTKssPgLN|u#*ilT8az5iu;xA&08ShM
z_((M(oq_Dloz1{DoD`#Qic}#-RcL`wBqD=7zAeW$m<a{(p6vvf|GmmB01d$j*nu|j
zodxpGC>RtZpJpUH#=Y%{blx9)y}yuppV+_pb0Fbhs@MoIDJj(Hb>}C5X>~bbQ$#rE
zgJj)dw(;Qm2pTukR!>X|qso~a|HKm)*E7~X?N*8S=j(m!9y$;-4i)$i7}wdXcAbNL
z>fwl>l4>gH9prQ6xw9O_*9OU-U@si^QN&uJNlFhUoZGL^D#HQv>3)rr{1!kTgZulK
zhG7=-lDDC!8_)fO_BA`HP~%{_j}?9U7vD~wr&BWgZCk4I`&;vHy9BXZe;jFmF!%MQ
z4xm0C<LBOm&<W<hc6`3aeJ}j9CLuPXx9^6EdiTvT%6MV$22v-`+#U4t#kP@VGW}(-
zDRnyr8@DgI9|4p4N<ij+`g*d2uTc98<cbq^hh>DcIud}0#r%&6O%BunCoQNVg(`Tu
z0dH1OZ50;0?ic{2?g&1b`3lSL{PTxnfCvvaXTBt$$L&d;L&69)sak>SJNU)%u_SW%
zSC#vpI)m@*bjkv*jp=S5kd8iC^Y!c{NCPb*)!k-c%08_yMn9AjUd6`?C{cw4&tUD`
z8*%$abaY5WO?+xvQ3-svx4V<=Zy;zP0h0exne6}JIALOCV)<WH=Ex7X?U4Qd-#Q4&
zG{jK>!dEQTWQmatqW99Q1#qU_oSamqp0c{X;UOhEN1B)|rSeV|MX>1I_kH;7R1f8N
zyMI|WvOa<~poCZ;sf<Ot0-O@GhipV2JVO17%p41uH4`?xe>x(2cYE*6UIr4O5upFw
z{rdEBG1pWVvrAvgc|pDZ({?=YGlsAVSah@P@&<C4f+Lp(a;S?*&W0T*(fHnCXsBI(
z3-y?a^ASO&lrMY{?OYuoA*u#TGdzW)5bC;%0+cho_~Jq#QIVrPp|GZ*touuvI#l=l
zNUA`BkrqQSjp8WDg=juZsv26L_Q($779y0yK$Lz!L@>Oi5hSxcS!GU_ON6IL9>Q<s
zbYZ5X!SQRT);opI*B7<qD{+(mGSR~=i{*{NNdQaQ<-pc(#v94QoO2uVJBG1-pE_#D
zN|iWR%=<h_qaF^nlOD3Xm#(qaw<n<PJAVy(ER~KQjQF6%iT`K^m8T$y;3DC-gxeCU
zZx;Jg9s1_0TP1bf=;@}NEqXDe?OBqwn{dIo#*Jqy*KVPc>_K|=>yzW+kmgt@aY<=M
z*VkUmwM8}k$%4RZ2uQO&V(S<M@4j-9BRrT4J#a(cFA9AyP6=JiquUivkt6zIUVWFa
zb3X1f)6D|iKA85qtZJ`ZC78roFlva+uu4~-V-c2&w1y0Ivt*-Cd<{zSV;?oc5(s#M
zPD97RN!yfyy1G=wwK@iE+rF}AVa(TPGwa1_Dh3$ySv1HXSdh}Nw~f%R51(FlePmxT
zJiVJZYoMjK1!@Y~EN#33Rd&q9MuiMhGA~0y3_ohplSUJ2q4TUgN+{z^`XX42IbwE6
zq#tEl^GwET!#vuNhF8`m=#UT7UVP#EY|UCXxGQiznx>?(6(2BBNl2{;766z5f6B|;
z{cxedD5DYfP&h3L9U~FB1x|R5F_miGvjyZyq0xK6ZRHJ-QKgvrO(o|<+Y>wxv>Kf$
z8r=cA9Sj)=x6THbtsAQm#s*|&+9@Mzin)JX;z`luzp1xD*V7|P9aS*E2C{gI_SyAn
zlDFtKST_my@CuofgCA$QUy2JpQV{HMKB}e`Vi+ms;9_5azMc4ifJ+Sj;my@izaF8Y
zPa>IvE;AyDOh)9}?g$2^N5RM^Sv*Y|Idp*7SmCK{=B<i#*!XLXq~+@LV1#2$qoOEN
znD;tAxuzURK^fHvd~*+bfu(V*7d<Q5RvfmL2R9)u>r3_IFro2}TA?#EbxPhWe%AVz
zMBX<Paut@>#*(1@q7sF=P&bv>)R|h_jA+!1Unf-LC$WtFHZZdn&*lcm4&(w-jI2(&
zXC>%)A)72ymTT9u>ZM#Z5sgadfYs+&#g4Oy0d;)>Q8%sjBNjZ>>~d)71c#T2ga-1S
zA)5U_4*1wqgf(J4Yo+2{d;7aq=^Y#rO8BE=_msf}TG73Y2`E^*IQ?hk>anEHNx9Z?
zaU=kG+hDlZu9Vg@^8T75%XJl`X`jkELW0~C*-}PQDZXcpICpFv0Lsp-;dS(7{nT~^
z`zc=67(d(0!GvGC^DKl)5(yA(9i32U(JInQru%BFhx1&0?+}sLJ}^c=ZRX-9W<~`S
z`pIPjtN<FipRon-h^@vE@@VS0r6gvk=vnW{6pE$co+$3g?hd((L2VXps*zJUKD=5S
zd}dO97RC}F{_CFz5qaTF$L-=v=WX8^I&(L#6QLEvPhSR`;~lHD?9tvIo^s&%#Z3XP
zos}maT2jY~q-KKRBD;L0cIhBfhp`76a6D(|m^Tw?*m}>#bz2DdT7<V;yiIVu9Hw^m
zpMqt+F7nN3@OJu<wK>T4JfvGk@qT;pdWRb(*!deQaQ#ocm#KqG0;Q+n!~Y!X|CLo?
zXZfG2B(JcioDbXY_zen3^vsOzNCQ=m6K$_-kD0hgd#{cJrBX-qq?RI6`DRzkg22fF
z3n@1nC{!wPAcKKA7j7WVj{Jow1Ok5doD$sdf8H)MLM?`}kWqL{MU7MZ(JZXa+oJnf
z4g@D<2kiWNYd5=}N>!R@%4ctKZlJ{AP^q<iUNvr`KOyV=G-}&iHK>wXP=8vn``Rx;
zMUk<cb%6**w%McQHbN<ogugABl(Gjp04Ln3+2R^BJ%$RMM}mA+1&YKmZ$iNo;A3cQ
z_e5X8%d+>;8RL;#(>fUQ>Trmm#%-kfTzFY-)#SOPh*jL#hgnKWhNMGWHHI03A#3#N
z(Nl#-T05jbS6jTUS|6BU<6N$dMwb}rwU)-^F2C*oH*MX2m>VguswDPZUy5<wPRzEN
zor$`HXCf?4LoyFWsiZYVcM3K88ZF7UbS(Ys)X{jT9BC_E>%Vx$-#sVe7f=z?SpH|-
zS#Gb{O4SzEM<C0h>o~aB<Hq552+?lDqyyT_4~(W1u0Yx(I)$s`SG|E|%i^WJXQDTU
zhqcWJvfoQF4Qd-moqTbru>VmwD$`$I;b3I=FujXJ&a1=L*z%rH20FbGaoh2vd>VMj
z2C}0n@mP`t6o+&M$z-MQ0u~Br37l^fEp*^RFr4Vtfl2%v|D8HGUAx6_@ozMSx_f1x
zr4WGP`3Ia7wbla%oo<sEx4o923^bU!nm86q)I$mH&ErBbb-i{CBfF9A1>S;6ZJvk5
zxTlF1nbcDV{IBSZVJGxlK(iRf;+|Ghqg8|>OAaPAQrn4Qi>0I{xT_!uvWyC~ML9zI
zIg%ArQy<91=F?&a?_%{2L!Hy0dUXNo743{sQcx}m_WbB?s7@6SHtV?3BT;!~8^AtL
zH$=!7nYXwz$fVLczfx;WFuz0|4;p_HQq2r<aUig;9TiS6k4Vgt-C$J?=#b6`5bvsC
zakm{tHcH8RTO@%++L*_7`~`fhY7tXNH{Z=tLHx0^KO#6mrTKJJT?QP1IApGiY%bP$
zhCV<b1=4TwkENo==`|AAw;k2c*IOc>BN!nEURod$z7f|^i!TT~|IrE-qJVS1St8{a
z4VTxW22UJBDY8a=Sgbbc&*ii+@CykXq%NGZAL^J{5BQYxQlXmKH4KCHgBsh7DR#~l
zT>HA0k6^{e_TDiT33t9M*A`wLf32oF%uJ{fF|3KdnS6O`V6VZ$FKBK`!*C#+p+dbj
z;V}rR!I3jev55KPz_?3vI*5dcf#CDMxa|E@i5gy_Z_l)vlXCRD(1=nE*O7<O{PHSb
z%+yDwlc-3ldj$sJR%B6V_c`iRXF|?Pci{Q(7p?C&?bxxlvlZrh<G`IyH&vZvkM5BI
z>9?Pd4-6rRtfJI6Lxhq0^iI&MZ=`4zIgP5vctx*-MF%|cShOmRa!06RBKhH5CEZd!
z0u|~Rk?gSv;q42qGX8jl-Nd8#zdp?rA!C=Is3I~m`WoDG%ZQb!c9`V0WVPH!+w)XT
z4Fkzw_PY{rS^(LnJ)P?GUu(E4{LLMBPj-e{U}k4dH<PR)OHqg+vQtsKcSpenJe^t@
zPQZbH?@gw;5obvb!W${~?z}oCdhJ0O1Ps$i*lS46!y-6=<qrDH;nYbv_AY0AztYqU
zqr*!7VYa_J(6Ym`MZEM>1?-_-fA@rOeGTjfj$8Byb4Fbl7s%Q&{(_CFHa>pZ=q_+A
z7}l!fw8kiLuB<DkU2DY;?W-c5w-ekDncU9?^&nal^U!Czzs31F_)-9~M?L-Rp=@)z
zAOm>~D+Gk-K@QeTZT>8;+!jd4tr@|hW_sVbNvVE6Y=jo0p`h7xsTBQ1B`?yP;%DJD
z6EEl<R;-<fK^qQqlX^DPgqAdfH<$l~U);05cxlBd877*+!z4@38y{^50vIm!(KY`?
zz#`*9+}%^iJE&-!>WRb@$|Ncxn0i?Ou7;1>xI~nc54tE^+uI)<#6fji%&)>eo1B^{
zter-r;F8fZkZz{Z$|-Qek9sk7P%+S;aJyYfpF7(Xi$>lvqRnp<Xx9Y|fbHz2jz7vB
zWkyT(Vhse{I2V>R`yvq0jU(~~8GzkOj(B_<o1KN2#iSkAbUBft2ZzvZN49W2Uqq8*
z$E7D;<V!Gv;83Ng;rfMRa0_YeMoSw@NJeO&Sn9DlMLei)GQdz-3X}~yU<cyEx`Ajo
zLw2=rI~CS13FAf{MGX}jdu(yQTZ&e#lS*+0iYAftJ?+9KX|FhtvexG!8u|0PgPIc!
z@dT^h^#_^@*>7pNmZaawIS0i=A<aj@oa#srm}L8dO?!5*=L|T!^w)GQ?)!@gJn+}q
zXB+6)!Ik3qYOu7ZrqB?*{jzCuHN9UW%+qsvEvOr^{?7#PZqUU<ovE->PTm7XWS5Q=
z*9XGUKVoV`kduvK3Jf^)*Yt=hUezr<GIuWhGeFU}H*be91bpcr*frJd-4^H{n!*ZP
z(h)IE!-Zc9XZQq_o#_yO!?mYC&(8N6U5%<v{bB$iC?DCXs4EmT2GMO_R|D%qn0?ST
zkipW?I`Sq4jxS7$ypwAwBQKW<T{kghSp1sXTfO&WQ=_%Unkr)LSY!xspiPi<rX;;1
zc<_x{&#i?}p?7B8;CmA@e<VVrI_0fD^>d$)d+_lG8aNQ+r>euIV!^+LwK~7-DKsaM
z-J_qNjYpV1?cqT?B-zfGH+05USSz4)>ExKWc3=DnWqdM7LB$*rg4sul-vOFNnnYMd
z!4d^BPV5-ij3e>7v8dG{(Q?i6%0m`VZ?{${c$C`5zJK|qVsd0jn6#Fb0YJupDes{^
zS~7bGf5v4GK(J0|DWgZ(WEPC+3Azcxm0QC=>9RP+Qri{#GNkL?C@-bFs_@a7K98tf
z$S%*Cot&TE#=ny4X1dml_@@N1U{|$BS30(2Kc?ui2qRL@8gwR}cW!W9;<wQuo?t`t
z?tmcH;%)}(`y@b*8J0T&HrsJ-rP&RqB8LaBleV8TQ()!93M29KHKD27c)M{7pnaDS
zrCuyb&W<W4RkQ5wdee$RN%*LInsMRKiZyq}C4@oD#JtaraImTi3|UK{0gsiZetmZZ
zXGW{MuX>Ah?F08q@ddAy|Ah&6nOya)Fmgz-99k3A9TcY+E5Lw>!39Y4O?KJ!42Wy*
zO|__9<IW51bj<B*kD;KO@+Uy2u;Jnqhwshm63E?p??bqVbhL74zVs7&aWie#gmVU0
zs0d9s<cg}VJwUDdX7)75xWE6l2Cro>-<yNXKp*mB@Mnfzr9Ke{ib57v`+Di-b*Y`@
zVJ=xH=$u&UYX7C8ec(TV0?wDTx%7{U?RM@rS?7ltVhXS=+avC-r;?JOZ>5^^icdxc
zW!E3ar4@dSVzzR~TW;}h=`6aadhYdGVxspvGw!nyA$U76=eb*n;*lI=e4S5=p@c&L
z84(nkW=2;_qsaL|L_KmuG8c*Ra7z0v1djAD`N-eQTAz#C5doli@Tu_srB8p@WB)f8
zhKuokE)ZJ&Ki*Py^#2od+5E$T9!w+TYEJOSHrYFyb$t|pC&it)86airl_WM9|8eoM
zHyR@=V|{Sh%zzAu8wde$dY=n=V2QT>_i6n_!(vV!shT)URXF9ZHC6*DGOti(%wKCH
zzgR3GK<=Dq@Zssr?)T?)Uxq_|z6`#B-q+*1uQ*tIoxkT9w!_X_UfW!Nzu(*914r-2
z{aA(reSqi7wZUnh{=biPnpyC=BCFuy!u~iv9i=*4q@b4<RZ^kUM8Re#Qf6Yz22}At
zal%QE5eCh@3OCNeWQf8K!6f*MB*_Za#7}g|a9WtuN>W8y3NI-{qKQB(<^~txBtvRM
zCR*4TtY-?88B7ew7}6{yD+9}kN$&!N(oAAfG3iQ6@G0g&+$Xs{5zQfbW7FT9VFtpU
zdk9SvdXmFb)trR?PM7!@S-YAhT2D#iG|ZkmPL+v?qKb|wM|iWSQU_Cs5==^f7S`h9
z&lB}=HDH)aFfN&T+Py>>NPWsN4t^f<*U|+69srDlJEHiiblyU3M`Pp1JjmH`#sm^$
z*qkzPslaAd3UxYiw8<2h`9W9SL#3AB5hakVtBPgIZD0y15IcT-9*2IbTh`jTP1?H0
zLEZ|_)uZJ}IMdp)jWqA<xu(GshmG#(@A*v^Dq}~D*uV&F8xBh~Wsa`GD|N5jt1R+4
z>NZXN7Vj1~o0$Th@mPdVa;CvdZuYvFKfxpK>@@CLHgBP0)sxX5EA?tyU3t$r!!0(Q
zbYQ~Yee0Un=W|}aTD-2hba=u>uQvL|{`T3W?)2YJHrj;W@46+`U2Nc+c-V)V5Gf^>
z#+x_@ApzU4_Ch7??FST?{?VF+1gS%v0WS^!p8GI6x!Y6AM92XlNqB4N(=<$0K+2TA
z#qW-hSxI;NG%7mf(qwI^{Fl2-I*)*p2QWtLS5PPWdQmIEl2eNe^6>_0kPem*xu1IW
zkY?nbSmj3q(DLCZ@^|RPDUg>u!VKNDwS?C}YvBfJb0;69fZsG_XjpJq3^<^WAwql{
z#ufnrH@bO0UTNad87+m8L3?_eip>D?>y%zdew-f(A@?SzV6#%CKu`U`Uc!A1P2b3H
zrG~8Mq;U9(dDR0x0O<yZH5p!a-xmt1o79(LM_cIusWMpr#`!|1oQtsSXre~R;K&Jv
zn^&#|d=4at(rzJP<oCxy9&AxhN&D2MTV{!#Nz=?l4UU`RP+SL`kTp=0aB@A-Xi<fr
z@H~?W`7&iGux=j`(4%z#$Eq0L_rJlWQ?D!wJ_>j}q)><hLn~{%a=qUZlX6Dn<h=Ua
zo~oS`ZDYejgzHe?-t2CR)?E)&FI!j{!my7XkDU~M9*&FOsWrn1WO0CmZc!euHQCo=
zQW^mbqOlxUh^&wewcUspB^R&r=F0-utFzH+Jz%-iI4Eynjg;O8-KHSehXu!)GFVC>
zLxE#U!QP8{OCB2@boa^{a6tMTWGz;aB+S_UH{5XW?0y%Q>ORabt4HEut@f@Z7s_p$
zA5{D(oFpj8bbyuhK`A0Pne6Yz)~w>~6#cAR&EF{(m2SCSElczf<tNy~gIF=UYVdHJ
zaezW>ZJkx!&7QU6#<z_=^XmHfrijf355*uSq5ekP-G|Pt%MBs#y)96e?y_```W7xo
zJb(mb6oYX3{zEc-LZ*&FQO`prXc>(FGIL4;;DkR;>u+2w`d5oyI<l%sKf4MdnTUDE
zI%*;eR_{Vc6Ve{}euD5;!H$z|E{$z6*{E0s?XT8V@)RjMy*cCTZMNEuOFDmQKeA*J
zaFQa7ZCrLjVl-lFUNF*0($=r(Znk#HDWTKC=@hgUtDmrQ^CaArt&9C<<3gZNsl~*F
z$QLv%R}!nSCXsA>XH^-cOW>5@ePkcD=%0_FWLZ*XB#Ytik%Gfyz+exy!fQ0L&YaXF
zFw7NQjz&aqA+w=>!b@81<4VZ~)acy|ebq+yjGN0aiHYLiKRYZ5;@p5-rs-hsX}868
z_`^J`6Q`XlkCV1pn!!Ld{)_SH#XmDN3iBk;KGHz!0u}gl1dAkO38_=4@_pRbSZ%)m
z$H1wSOP(3#Fmmj-(ZwOMto|PdS1|c&11T@*eBxfXV#Samid;%L3M#&+N9C-`L{+%Z
z(aM65rVgMGSINw=p<5|e85x-3(~jnRkD6JtW%TZb@Yk;3iGw*w2|(tiKuCTQwhWX?
zjQ;8GhVMGLo9*<sM&p_N3r}W=6z-iq04f1}bouy=pJ;`v*ZbV&W_|Ogrec{?aQ%In
z^69Q7a^D2I7<(InbLCGkmR2z|U^~-0rK%A7Uau&lYjBYkVV&Gib6DtGoY!|#IrQf%
z3Mqkv%8bOHl#gf-mfDkv<8Yj3$6^oTWXf}5=`b5#A?4z;%zBJBWY_0<uxQ>=d-ys!
zKNdn)iEYn0x?S)Pc_t=u!=9PQhKfZ{M3q|4=&*R`%n}(i-13VyiBlhEj~Z%6R&-pk
zTy3O6{OqfHD6ODT@UW{lSIhVv)LZt41}E4eFz>_v(gd9Lwr5BEsh0Ade=`HLDsglU
zxK+xNApK|CwJ~+pU^2Po^ke6H*JZ!9r@?;7N0Yk(-@XEJg>#4BwvlYJw^IfY@gOf>
zzji<9K~XN5%=Gr48sdC{#iGm^S6a#VUJ%kgarP7Oui5?<cK|NQ0B8Fp@sBIJZKbh8
zK}q(eaAjLZo8u%oJ6rD>JBE?OQ6`)}j}*NleAwhtFEQpGRxnV`ec!VO(aoy{BgS!V
z>5bB7k=2P~F{-ai&ARQa8}7kQ8OkI#vZj>pF`2jo9Tob#=+Cxxglc?=Kfg>#-7)SU
zaVT*lT+v7`6>1PUMTE@-m0%b}`v|VN>0m13glD!ojhuB)$RTY%cat6b+Ljx)8rY9<
zCd%Ndm7E9RjG2<Lj26;M07W3UAgVADWa(W1kdDeEgdoi=q?&`N(H}SI_nc5-jOn{(
zIhZ1vmP*W~-@3(CLnEN~(U?!;rP&l*rgS&CQ=PRkS}NA!wA}0I);4R`1C}*o-&E1&
zzwFlA_)K>AOC>05=Q>0*<uMz((F}4_C^vS<A!n}HDs<(#VBtYACI})#HT|X66Q{e%
zY0^iZZ&-$e@NrLhhlnB>G1ZV?P(Bs52bovai>%N$D9u#~SbNo997v%w*QZj5*ajaW
zs@l)~AM(mr)!yOLa2X(!M)V)k3J-|`5GYu+tq{8)hfUgQnBF>Jmkv%iZhR9!-f(`Q
ztmVWimu~B|1)c@H))}hiH{#G){n%g&$Z*60{=wAn{_-_UwnbfRKa_ImJzwKPq@rY*
zu-l>^5ZgQZ>0zv)<ahp%(yNK5=X=WyFp~SR&u9N~U>vOGZ)Hp)c|9%!EVW1&_OxG$
zn2FOo;J;+Meh)~4AXS$7D)uD`8{g;I)klkZZu~%%B$qG-5Q;-X>gXWVDUtLW=oPv~
zB{TnZTCzUP(}{cL8b=m8^VanZ6K<1<>(A%`QN_6v7={)HN)qkk)~p2+|9ys}Nfo?3
z^4V+yixFAn2&9z~9!r5fW#xB!_Ygf=D#0sa`}|^gSq1g?K-c2gM@Z$eW&#_^fk=Fg
z!L(@lqXU)lfuUz^`QC&c@_<RucY1Tbu!mw;7*sE3EJ}qXH5gFigm$!fR7Gv$u}uf;
zx@^)$6NH5vk+K~VRUw!eW^DDaoeF<o2~I7O_AT#F6;i4+?&-rUxd5(duFWw9Q<nHV
zQNZa~7f|}VaWPiL=(sPwv0XZE4kAn$iPUa7={IXBkoZVo1&a=>?mk+J(-#d#RNAoe
zEYqN{LgH&p2<siqB=E2y+WCtm>Q1K-`q?{nVt_ck_pt0kmM}E`_WS<)GC;UI%jZ9}
z`hOYG7@7WOh~t%}Ov>hu5$!|cw!8v+U^qq8-r4+qt<glIa7cZmf+{`~%Sj{(JE`&V
z^ZjZx1VAOd)Bao^(IZJXZ^_!(=T9V;m%{pXx14=kwJHFaXIY4*GI39}Pn{&3q&jZe
zA@84mr@H;w@HuutU|C7RToXCpKlsP@d+}iC{%bXxE9ER!#hE38I(NIz$0fkSBjB`j
zm!7{{&o|(PADw{z^ZhGdpigJZ)VTHY?H_~ZaMcnFTe2m*#Tl!@2`x#7`t|r?sjeMv
z(-7Q@MJXS?dt5f9`%i|b&F1OUS*O|Ts%&%P+aVT5^{ENR7Dnb<_d0r#AuS-Qa%<^n
zg9do%kIZNzE@`q5+=pNa87moY8EO=%Q4V-=LWOJ@C)6M_Qe-b6n)f6HsisNrzD^LC
zMq2yHGrCO52(^GUCbS^lx?db|BRic?C>c3-NH%LeRHl$sE_8)7!>e%%WC58?swL;w
zRgqTR%!XeN-LqQ5y^O-xHniRRURf4eRoZ%+eFfh#lANFA)~b~XhyOzZ69k8Lc7oNU
z`!%E?DEMO2>EX)d#%Q#3NDeiUmv}Lvytsw7!=2v(>o~}mnP7~Bg3p#l>DljtW=XH)
zkYH2^Fr~@J#9V9|j~NKT1d-VeX&1+GTAaVlB0kJz?KvvAh~S+qqN%aI#-yOp^F*Z)
zXCm_tp{S|Fs$NfuVD80U#Hr-6&SgbhTiT}OS2)r3X)`A?JJr&v9cmehJ6eI%&Tw6O
zGhs540hZbv2AbltqO-!JX8O=z{nAaMlDCevxrk)v{D(VyMz<4$lw0lo&fjVa#5_D-
z4V>(9R=M~UF(O*Lg<urCq=BGE#JK9<5Ph;$wBwVcWU2^lg2jhk*S84~9Ajgn-MBmp
zzPxw0v8B^MRYGvWh$F=&SA+?KmxKY~IIQLPgC`^=_uP(K<On$-1CnKbT%m<AtAp>8
z0JMM5+boWNd^`pqye|sDB8MT%4q$Wms2+SFi7*Bc)2NhmbzQMX2CVM)uhu!Z<tSD|
zaSd_sa>zlCOPKeylVi_ztco+~ZWFrx?wix9?+3tw{D$rYsXDG@gUELnwv9o0+^Yar
z<Z{8zb0IF5m?CjrR5F}unR4Mi;=;Sft%lXbKo1@Oo(;ADvbYH0zOn}$a<dB|zWh4=
z_eU<ROU+F;`zsYn)<qFZt|F!JL0i!6c9Du>WNm3|HDdxdEhgrU)k1X27gG4kBe8*r
z{WmYK`qTMbMCs?0E&`mYk`ANOuN2xjZe2(6+-E1lUmXR<mJ*Rjv!V)5DR&Lk263L6
zFdIS^1DiHZ@W;UiAtLQ4zYy*TekpIe5ek_v*{E7j?<b-%-^ieNq>B!*og$%0vntIg
ziediCua4=BE|m%^#J%rR(iU6w;)C)Vyw8Klg6<|m?&QF{1P1OkX5?hV01e3<Sx7~h
z3~6(eApMGOtG2$F;-L(xMJ2!#*V6SB<Qd>v%A$rYnhsOi5F_PSw~t<#NwD@qkM_it
z+DHF=Ads5BofR{DX&h^e3bc2;hF{g%T&-<Ddwo8{{s{UZqV>^?YtMDhTN<To5T|!;
ztc{W^IqR!!vYWazw9M2ej|t&Nmj&N}(B#o<#xHq{k1+)IEd(Di)9I1ZG{r0hSR9&~
zbQs=g!HqwkcqfIA!x<;@1u<wZWIN)f)S}aOv^IIiW|!ZuUrFnt{<$G4zd~AqSfiEJ
z7vE8RsH)?uCZOm|Xr*kxg$pL6cwh+Bw1!^gV*QSb*Nz#1BYjg4(^~(u`Ne8BhA>@2
z@h^jg|L9><^=ZP6`)y`U|MFd>Twf09{ZTV&XvIcnY-Sz1a=H<_2nZf5)rRU<8o2!^
z)G2;OPO{Qj#wUCdT>Tr)GlPP`<n}{JX@DZE*n4;0y&sOO3ovA<63%qm`3J$LrFPSp
zlDS<p%bLNu5DxpJAB(@6AB%mK=dd7l^*n9NMp7=k*~PD#ixdj#Q9LHFt;Hn_N~hQ>
zyM2?RwJTItKGqv%!_{epB}h!wF3qDV{Fo#f1$G0DsS;sOZ;P*QVUq(m3rx2Ha6~L4
z6AbY2(OM2qQmSr&OchApu2MeKk{e@N@W)Y_?1CF#jz_V`^<TysyBz%nVSjp4;(HuJ
zF#--UddfEOvs+^dRUFt5oxSSlH*fN)V4#7do{nFz$`Sf~z^$xt9SSL!Ih6+?G@%m5
ziq-<zB){1~2o4j+J3kon(K?|c#I{jlc6s+Mv>BGY1eUz0j`D(385YB|UsfW@Lq-Pj
zjN9;II;r@<j20MG%rVt<S;O2z-ajw$F=xs4VRMN3OwouWM<sq^yLlDWV#R^%3pc#3
zBRk($?_`VY6Hwf3jtVb}k>M37xn+hZxoq8BgnX~1&h2EMwD>(}Z{kAIj5O3RdDh{!
zxxIx>-LkM;r_nXPlCYT#vei@4X7<7+tO>7kwziAhP}E*}YD7S^{VhSmknTHdYrQSf
zx7Ng$%Jr!*GINr^{l}atbn-BHdAcvASyj-yx3B)k@tHAiK1eqce$xFA{SwDSNl4|=
z<X>fZ<){PnaqA(#Ow8nD)BDPc{@Q*h`+H+1a3)KM6=rg|=#;&DNu<etz}>`-y9Z;r
z$tzT_{r<d~2K8daPwVgmk!^e5lFa}!E5pRJf1e&zK-~Vd&!G36LlwxW{J+G)Pjl1%
zz1U;r{GY|b%nu}Ss}(6=)nK4ak%b$Q001RVflxcvxIF59QEtcWGzOZ|&7<yd^ftaI
z@%2vM4kV?xv@V~b$w2H71cn~7z3Z>YstLyG>2-fvmGS^1zDP)BHGQ~ejTOn6s3dDi
z8dAmk%7q#W>D-yf1Q+nR8FQL}@!jkDaPK-m23M!w)1Ptb@a;8;36tN`<@<cnL8kwC
zIP*|7)#f1o?ediI5F;7;5=mOX436Xola(n&u_TSAm|ThaV=xTH^q(h9ltn@>8CXWg
zoUEi%_t{?gIN?jXoVKMLa@WiJAsU?jqf*(gtbQ#06l*in+wZ?uG`r4#&3M9}xL}@x
zk@2m0SuLZjcDOxlosE$TSu~`1KYy}mDni0AeIP4NrkvNODA+uMbt|aCHK>kdrEMjE
zSzAsHnD2_4y-eJ>;(oj6a^8ed7;3bOo=JAq0S$6Qhk%)fR^Y0<o2>b##PM81rIBWj
zu<e|#%0Q7En+DRw>WM}wxA!^l*kJ&c0N3O5M$@6Mo9oN%G>IClI2ueDAC*iy-bxWp
zZw&RI)=x}qihLPk=ym^MKvUepES(9&4Ix6JOqe3!b@~Zf4sFYUz9LW<nnbEz{^VD{
z(fjn(9Z3bpaqQMWwd$4|;E2r*ZrbwkR%3LQO$>^XuY*fpw`{0$l8bmCd;h&OJ}iS(
zu(cgc>IB_}OhsH1hGTQVyQz`TB|#O1$PWR@9RgOQmkeo|jE|sQdlx}Un^kDdW1w2V
zp1^7wjczbsx!=JO9gcHmzWvv);(>XHDF!;CMhlz}PH{X3WK}%8aiBsiZtmG%V-E-V
zPxA+}&hM9yTyc3mLlk0i-bQ5{tZ8W_TOo4izkt*vMZuNr=PFP`d5Tr0TxKxMY2|=A
ziIxqSF{(9H)HmBmNkaL8rYFMaMTB^>&cTcX!aTHqvh33O?oOdu-8O70%i5k{Lk=h~
zn!VuS;KNy-p4H7lupSB}$sGE!bZ|p#30p?C_E1t=gs!+AsvcGvTov@VNLl27609;_
z`{bNb(!vzA#!hRu_3e5j>nl<0%?zAHI{4ao<rJX+^LBoBkb3GQXkyQMtbMHEb0KdO
zUX!xW$Y5@jfZPN6P2BrNZ7%zp%bpr1#`N*iPu3z@2c9%U=(bZPECmxK&!7c$!CQN6
zJ+qNt|JJ?eB@)p03#cx;yFC@C>J&AMUA=P<P;hV;a^Zq-D=@9(u`m!#P=!<t=`mTh
zkU!-^G<I|M5qz?38woCkjyz^a3ev*bo{UT^a28(VVgy#+OK;tq`*&mxJ7CB0X^hh#
z4%Ygs@0j=}DeL-KG!Mq%CsHC$KoE0-G597d=}9T5Fbq&;0v-gvj&0m@6QtgrXtcV1
zX<d{FP*))Gl0l`IlxvVb+Q!CJ!ISbxUxc&!HNqAh>JmicVhj=j&Gfb`HR^S_by6Y>
z<;4s^dZ+v{K7^gtTFv&oBa{qUt_OgCdOVFo3(}uZ7Db9_wvjjdwLH|Do@a%NGQ<74
zU@b8UdU(J-Rdb68t)!&2(IFWbPAp0aDG6mdhFSl$ac4;(yFZ4Uj}WL-jC)%<SsiK?
zb6{ArM}!e7Rf)>{Zk9>ob}sT!PcE#=((y;J{K(Cypmi~{CKkG<4Qy%KUd7z3tE0^%
zE+R!FvxEg}j6*`S7U-iqM`#$5x!EB$Qnw`0qVX`&)8uoe+cyE>B8^~gr3;4x^Q@?}
zMq@g#O3%OXxXa4`Lr&%Vd}vHAS1-V_-}n>cRH2U#Q;D(|>@Ueck$$}NA=XWB7LPRX
zYV+%$DstmIfH>bwn}?=`#-hzljTd&)A_cog!I$W!J>uxWUbk965E$%C<h&KPYE$T4
zR1&_$eS-xtL%zZkurz936iT{PILMOWYiIxD=dyBY>LcT@LXUMmE{Thaq|$3OIPKFD
zdAnlH5B|-}2-nl+@m7?(;dDx`H>}$!11iBdBlH}((W=x$qs#8KW~4w7@ra_Xr46Xu
zKt^^)A2f5fvO(5YAFrpbG_K>Y7<acp2-Vf<2_rf3dpPTifzX4G2}A1c2oXzBQK%%V
z@a!KTx3UH;jMFhOX{Wz-xY|FhHTYGPa?Q>IWnP<uH1ST;c4CfW46gHp4tsrRB}jIx
z$LO6rK~g5AksR=#KA%zR=0AW%5Q2@`9(`BQ;hScS4^$FCkr3nm;^R>di&+i~iwV#h
zwM7azJFnzrh<LWhu!T1x%uNk)-B`OqLQp;nsd@pjXms`xaN4A=?bx`(sKepexKUex
z{a4a(S$dkeQ8_<6(PXmR6$+FmnSaKYaEK3uOpLP9keP==Y9DyZw1bf!1moBt2_Cah
zE!UG}!(X52cw+B=YEFs>0i}8byLw=Zn{$#L1xY}B)stN?Iz{?y-$nJEs}c^Nx$I-|
z^)8k=QJn@-K%VeT!&_zwA<ZU7i9fzfGGXj8bv@}Iyu^F+wKq|$IiFkHCfA()YdDk(
zf<Sq};&GnK57K#`U<^VSXDfz2GRhpcZ8TQbd^O{}j-GyWiBs-Xq0tIs^B!HhY)QjF
z$aeLm8WvG~k)L^8CBt<zelQS8V#dR{78T0*z97D%Zh(<q{D1?;wJ;g~TLubaW>2ih
zBpMOq>3ZIMSGDt&KB88Y;9gh7dba)J9-3I?2*y)@5%z^JifIEZ1nm}oQUcrZUjxyj
zM53{m3<8&u{w3j@rLMTsG&k59E(9PjCvK!-wfDg+mhe<kfPzF~_wOm}n-!Z-$AC1W
zz=*PaJ%rBdOi4|0P&uC`FD*o`MH3V|u9@RQSyCfoWS}saLQ|*7NsK-9SNl1<LufGh
zZGtgntpt{~vMsT~Yga~)zeR)+Tk)<#92pH4>rY$8*!+s&_$ZQcl-%WA^8j2bn2e_}
zfe4&p1$cQgj9VDk^f)fViU#MZ?-lpTlT?2%>H|W(``fTE<T4xsW<=`p&iB&DiS?SQ
zAm`c8YR$#UG*NLN%&{LcS6e2r?DI&}Py{|-_pl{>@fayS;lZLUrdmk#FiXip%#x^G
zvh>8v5?DyeNU(<-5Yh1KHy%0t043&LLgvc)fjz04y2tKgrR}@eKhGw(&3s4FzBewa
z*o1(?U}kx-LpE@&-d7k5L$%9^pcj~ijz%@F3&7^p)CT&^SHW9(%BSU8EK$h*j(ots
z*l1*<(s)eSwUa-7zU(em%wLc^pAe&uAF`?K4h>TIk>z&+0zh~b*D3hjtll=e&3J#t
zE%=R&H_-PvTbA9?+RQTqD{dSAU87oq9O0<icrX)W_IOY66IFjA{6CcaQ;;ZK(}e-H
zZTD##r)}FfZQHhO+qP}nwr$()p8jU$Vj|{^n13$nde=>5<<7P8St^rpG5^^D{+G7P
z^#2FM@PFEFGt&RGT_SEGg4Te`G@ir#$3v<0<JZ%?h2n!el2LuZ|0RLjd4H*Cx&lg%
zIju{*>&xTUs#0}teWB4sTN7e`**@#rJzp}L$c4*`8#StmN{RSO2DXKwRh+h@?j3%<
z4Ebbtk+R`9N|NkTgw}d#e}BAn3=JK<WkO*y`cIz-?7!}hprhAlYjxUB+2r(ie0&}S
zecXL+bxh>=xIG*^I1C*OW#srUC}ir@EQ%GhWmx<}m5@I(-o!m04r^cjjYk?`v3PzX
zd(p1iYMAD==mAqzqD|EeA}_h39B-BZ(m{}fpwB{)S8UYUHjpYIA)q`IldTN}LG0Ur
zbHVE3B@^c=Lmhxncu%eTSJdeoVaeiNyYBp@{cYPAD>nY`dj#?gn?$n-CwAjcqCckY
z{G5XiGR}wN+W>lWyK1Hv7ZDwAwAdg<e=Q<c(9{Tny7i0{X9Cds0FPZ@UHCJHdd}C2
zWYS~lKq#GQiLW_vDi%tkrn?KlY`Sk9vBejt!XV1Ss3QROKKr{B4SjUVl2!9egMC@e
za)|xfq0$>JbR>d5_}gFO>}RagxJG3Zv0`1}Btrh)x|ZBi{e^y7RT&Fcz+!1PHW0<E
zUzG@V2*J+^5{$j@W;brQec8wWKr+-x)pfDmj&@mN%E5#x0rfhCd5u!%onCo-O3z}D
zFI6u!-HC0jTJm}*CCP*3h4W{KLVn;`bz{YtLfLIjU0g&{K&*Grth@SoI8OCKgR7gH
zm7@`gUodx;!U3r-Hl8qZx0PEXDKwv)nUfVeIq7dn%5-u@h<mHeM5SYnrmx{0O@+l3
zxN(z5wJvh6(1|r`u0zMZR`X2r!?XY}Fd$955WBD=aIp#l3nw6$wkjjN5$n`wCygUF
zsy{~of@%_e-t3cEC?aJn0+H}#@xb7lv&!K_C&l&FO%e-YN8^O?JT+E&hVq<|63Tjh
z-&Zq6poubywUAoUgVM5L_nLrd?Z%cwZy$;!{0RyqGGwHDS04d~io)(LOi^dVq3}09
z`de|ivX*Bm9cUkRvAmZElB$n7T8NVNw$xgBNi0_aFV3|UxjiJREK43^YR)LllJsDs
zcos#z5EumsVeGNUUYNZQ@_Go_EHKMO?cYlaBohnZRt0UaVMK|b`0v8#E)JA{sas$Q
zql6=)tEBZtO3>Kbo%&&~&bm6-ba(PnB3XQ3bRv-S1X{)H+QD^>SRlA@1*9I9YuG@V
zU?(P#V;FCFv=fV!%(Rej_@?JCux|tJ=keC)ERpBGFj(wCk*ev0kb@)UbDK(V>Mgqt
zOWoh1bkN0<q_dF5eS@qfXjFRSfL6FBXka*`Xs#JNJb7c2f9fv-TQAm-3s(_Gls=v+
z>@pJJbxG;tdC(n&T+i}b@rRFsvC{@^8noi5oirHO@TPIjd}myi@^1Y&X{9Zn2HfX<
zdk{Q#QB$H4t7qo02wj7jT}XT>jx1Isy#trg!vVEh*JQoJH<MaNFp_jx&;As)l%36#
zU+UzI#7L}<1%j@2w@Xa_-XRpe{+=EemDB>`%ADuu=Lap={r8r~aqdc!&yUmWL<+#q
zJCrw&5hL&hdCN8H1+M-=Th`z2#ZBDXIO*eyx$;Kh$?mg|(5$H1Z6h~w-5Y|1cO;%1
z(j{?!MUFzgm9C(%z<9-cVrfwlE0elKI=MtB{wq7+10eruW^9A*asv=MCJ~rEx&z5w
z(2sK4bM<vgRXeZrK_KihrnEW6ch;zqJ7Fuqq)-<nt&^3Y8qPmE>Nk%JIy)UA17elJ
zTbG`6ACpc!8;4eYyAyzjHjZ@;(tOEVw;ESGPDarbea&wtm9zk>e0F)gG*-1wd!bI1
zAtNF7MK8kD7jkz@5bd&lm!F3@_F}mBM#t;$PBR8%kVr{L`moST7$0X!3aFyaxf)?W
z?6nVmnH`;Lk-*HLY`Wv3V@9w^f=(+#mmED54ad6@lQ+eW;4;L$lk$H=a!qO|@&wTn
zL!p<GKtxxV8Q_9@Gsg2oAIJK^>O-1>UNA0P0~{bSKKNC@o_g4wh*0SJztj>aXr^+b
zz#NT_mYn3Mc{S6J$!MuFoe2FiGf_XzIv1R>uyq~<qmDi;tPe)RSXHK0Au9xKqDY_O
z5a>rj^;BtrN&UVk4)V9GCj1-rT`?{SXWCntxstGjSfT6ao}+3Dp|?;|cesi&?EOnD
zGv@(|?=u<<2y@Si1#2S-9X}OCE`3>sV00VOFDYpsj>isPG1dTPxzv*HdLiXDE*F%x
zPPM*VAS)~Om2YzJVMkGwkoWcz>s5rc&(%#nj*f5UIT^DZn}SE2nm=!y%wzWYKz(EP
zzYVYD1>tN>KnR^NRd}SFcs7*zmy_(|wpx?z_zA)HJno<Kq@ySVffQjC*PDnIZ<Ph{
z$4BY3q{fbFi*T?yb2@Rv5yUyPpGFqnkPX)gIla|oD2WgW_5c^hkm0}x(us*CKGG%G
zKo+Mhvyb5#+Qt&a_hW=N`9pYf9{k!#Dgr2}a@7FGpsUl`5#d=p2eBa=C_y}Txr_9L
zBt<E3srWb)UiM7Wj|<B=^_d?vg~ZJv{k5K5u3MWIRzk-%E}b2GH?J$!&G8KDavrq~
zW+kP3*~k!RU%K_TPwitV@lYCwx0NfvSL7A&lR+d#ndHlP-8td*C_owvAiiBMy<@nA
zZe<zuYH$QuA#4^A`nD*pFH>>vYd8T#1L4atR)w*9Zk3iLY8nLRkm5Vj4heF$x(t|K
z!v5XK(W6X(qFhM0*~ZfXpAf&{y3$$F95Wz=b#0u|g|&r1pKup&xcT%_09i967T~et
zpprU+5LJHMTbs`*^B$WAsQnGl{$UJC@oQ#RZ099oi!u=nS$l3Tb8hlznx+rx0z9T?
zOyS}y_awOw!8HG=cjav&J>3x1PT(l^lpvI7r~7HzD(x9#+ZUlIF4=_G;5M^yBfnHY
zN;Pywf-1J%&x6f5G%a`nk-TOlVg`)9r;x_SirY?nuLJGR{wo3ORVC_hGskp$S{-S8
z9C@Cr74vy`mCmjw%hZ_1Q|~4Z_dVhlCz|+c9Y}?VlFpT4gR}=rQ`xf}`ub78&Wb=3
zTa65H*z+b$m|oD`I!0me3k%$f>db&}K~58t*dK!|!a<84Ve*q7T`VRby~=h3r(P7*
zT(*ztp1hMGum6S@W$~l<&&KV)gg#bQhX3m>;Qy08-Hzz}3w+c#!-VdO0$5AqI?NPb
zU&xjuv&%`322aM17(pouO{D#L+8Xj_Rg2$Z$~r?HBWYE=eA*z%*_wX6l(F${b$@s+
z*nOiiX^S&<Zc3+Mlx)(|ENQ4Xy2J}Zl4^)>Gk3i|u<6nEdcQh}Xp?fgZ;8+(z2WWp
zIT~0Bk;$R`c6&-`!E5@%I8Sj<=I0d(KVvtCh1-%nm2@a|>dBIlY}A%#G_}&Tp%qGX
z_h4Igk^Eet;<(j1)LzQPZE1FvZyMgN5B?dXs|nx{N1q0*|06u^SPv6TyGR9G79~)2
zAS!nn<<VLFoN-b7v~fI&$v(lTI$@$uiei$s%$RjTCE2jH1lIj?Fr;Z0*JzQbVt%M%
z`HbZbFwE*L`m*O}TNKrAb#d(}+`=jow|H09VpT`+z2xO}pu%iQD8n@#D*vR9>ALG&
z%s3$*OCzQPk`#XUEI>TeWpirQQ*?ozeV<SG#ZwtEls*c*B2<)s!uLMZV)zTjMw?nC
z<TKO6bh)$g6diuwqE+t+)B<TuLLzKCf2IB!zv3^hhaEux7p<j8pv=a%mtrr;$Zh%Y
z17>n^gID*qc2gxO@NCm3lr-UBrc!aUc~x9DtgDUy&HPuyl8RB&^B&S%eachcLJ6pa
zk(5Bwf8N*s+FFzBt-^!4M9qU$57_;>W+mBjdgX2ET$L6_F?QHI$F>_#O?|$rnft`y
z>YFHz(+?$f7=`3akRZ(9+4L9pY-yhM99FfBPvq{Sx!CE|c*mWPqdia|!0Io<Rc_Zm
zc@j)EVDa!}kRa0pk(1GfUDe#=jCq^T+Q+ILD|}R@$IUnMZ0Vqa7GX_L{u7s>{-h&h
z#PnrA8D|NqxDZ_^{V#x=QmObH|Ab(081(xXVBYh4e=!1DLxq2aFE0C?24PGFYR1Jm
z`n#I`)<P#B_=1iFwm0j`Hq99Jl*pxC#NVyek~^DE_cDtC#I>)GptciIXvm_QbkdYQ
z(729x*;4l=J4`_pr%JXDlQ0$&@5$swb|(}1S}PM1rF60pyD*0+22tWfY0diWPw3N}
z%rc8fek+f2@4|l;Do~KJQT3vX+n5=LB145maOXV6?aAF9<|t9Pu2Z974O3CJqBpFA
zaB@J9m3j;h@sWAb#Vb05Wk#6$F^Pk$r*1}Oaws>dUfs8IP?qh<TakyfaStfpY2}Ui
z$H&ks3QSm}AUHW5=y$TQ->&MS_*>H4g4nMR9l`^;m}W?yJ;gbPdoh<DD`SHy2dNa9
z$bf{7D1e~@m!6hyG|?3k9TbyENQ5LY#s9VspN^8oCB_h$y=D?VZO^3O=JP2FU(f^k
zs?s0r?T6%3$)Erxl>rQvLL|apLpl(g0ASrh-5R>Yk$A`qhB)!zaGL|pBeXtV{!y=8
zSBXf~V~?RYOe7Qk3xIZ{px2!pqBw6niNXC<9NHh!=r2!oZoYR-5D>EXvv97-pbWrK
zNv{x1k2&$sU#i&<i3j1}HcmlhqC6qDoJe}J(AWrJa3in>R8KViJGygK?e|3~uRj9r
z5%#)q-lYkp3ng1LDXohUT~Ky`gC+nB^&#BXi3x5D3OF|~yEK&Y@i4~#ClpFNG^ZrE
z$oiP=$P+Wi)LS|6PmDXyU1G{QCtm^%_Hzj&d_q9x<Ah)lqC&A5gz=LG(A)c9VK|jW
z@8ZX94-i2W!f-ZKvl2tdg7XKMLZmo`D|}o#VyHIlAgMMZDi}r3EZpYSg6N|Ups<e5
zp;YhOVh>#lis8cjq<~k3+T`A^ixND{CD71g4Q5T{xD8*0ytxPXj-3zS^~_}gXC(s2
z&)H)n23@7|jDrum3-m-XS3`TaCby!DwOcVr=R4J#Uq!zKeG>psV7UXWjL_MB4ccI}
zQ$qB^-7{RzE5iCwWxBKKx&S!>_kw2XmRenHeatMH{zz^fDHNhfH@?Z3#kQ;0Fl`^*
zbTb{KG4MF%l-CTZoGIjmWRDb7$_zzW$l<_E^z;Q{=ZkWDx7xwmUq<-L^Zd*I048%7
zs83~MWp+Yq+~l?PS_-5LjbZ&*eF&Hk3Z6ns_f!!OY>eb|oQk{vf!G4q47nzPK%&u_
zS!xP5;S_$GD)65T(oR`TaBTtZi!N7utnveV1dEPxqG3LW^TU*eYQ!N0O$AjJx*!Gh
zTHGkyz<8=x9qBCrqwLOl!5qh31?z=3tUpEMj!895?zzceIWEmLZr%TWYzM56+VP`*
z$<w+xy`WKHe=lqq36&EwoNz6SPPkw&Y{Cz*n<|UMR{%xeF{V-sV?Z3(mG%m700&YJ
zj>(1Yh9QDKUOQ^*hs+HSFDw4QrgcrHc*sGWt|5CpRF;A|(G;9`zfMg>(hZBJeIeN)
zT`r{d8xrK^olCkk$=y|K13&BsNRfVW-$qhqG=Qi2eoRO1@#Qis25!!vi%y_SfC5SB
z)4|qCUsjTYN+OZV*U9k(l!rDh#4*M`+^_=xcFC*NtwhU6-4c*ktjmDg<9(&4x%_;+
z*Inz)stt<b$DSvX-|!{5_Ih)q%ru0na+-YjD5tIs@QW=DMM4&j<q0Qu{WD@ASSXwP
zQ3i_ejuHZ|6%L)LA{0Rp{rYso%pHsTe#bR4B8x@2&x9(!PbPh62LtIeKx+^;)Cboc
z7LYa3z^qt|FT}I;qI&N`_8vFXke~ki1%aYuN=Z7IFPr0qm3xe-#(`=Z8bZZQG%m+Z
z3|IMKoY6SnduK3qW!$S&SrM{yl<QA<dL#9&1FNB>whX`Ww{^31XbZZ59u!paJZ%&v
zVA(RYy%;m7EGaa2wxH5Kd>N@MC$aBq9Mve`H0HpN+i?T!#7(nJqa7~>BM$a)7?F6L
z?urYS9cUgoW~q~RQqwrOTpi5-!iMr^b$-hb3Fb@c9`OKl=T)?2EgA5h-zNe%5E7cB
za1ut3*izlop&49RRv929YRGeoOWw$qkNXWFZ#~SC4CZP5pck7J=-!RU*Dd^3n#Ge|
zO6;~Q;N1i~l-;pfM2h5<2UzH%95DY27kKc7jyI5jYH>{fRsj00|5X+SnZa~)&SBV{
zo0pd<mg)gm^7CgB{_5g5<hjv3TV^A0I_;kk`gfv!X}|y&g|m0TkIU|&o~x*Kj|ayG
zBOnJT<dQ-C*u}95fL0nk_&q9WB9y3+><B|@S2>J6R%$@Pzf~OK+qS*T-g6ik*lfdG
z9x;E^(m4Z`W-6=>z@GQV7kH4qcMOPmQ%x?F@G!HD=Wl_vL)^!I1!1#yP6`Csy~l6R
z;xtBJ+-S4j>rro~HJQs{;IZ|ZgrWT<*E?8TrK_>)dpGNNf7FATgAc)M<FZ$-dF&nH
zb|<XJDQ#nTi~V;`6m3u%?)e5ablz~9TpUQORQcMM>sp(?+ll>Z#0!JJADa!yMq8>n
zhPAJSZu4l*w$oq?fe_$Ocq7>b6H?Bnv-W7jj@H>cB6y?sjC-gLt|$}R?j~esEjiL(
zrKt@FOvWl1Vf;7{Me)_=Ww9W7jhkr0L^1&GF1l4WgLr-Tgk=G62%{c$**^hRwl*~m
zI!nz5Y@Rso*V?l-@JT~6(ClPB_brE*sS@l7-8vkUfbFe%(?(5or8Sx8*XNtALn(++
zc6lsz5}kU{3lJgn+Pq!{zViTF6{Nkl9}E>}z1W5Aww?x-GJJr^v;zN2;bx-yFJcA@
z)BkyO<w{lhcXb7!>q6z0d`gv$2Li#g1f0xn-7$XH$>~%o{*Dfbq9IQ#s4%m=r&|&T
zA2K{1mGdfZAwpcBcVF-2Y9pww<+=9d;BZaf{(IgeTR=i~k6J$lGUfwFFUKPy>BV{+
zBHK&#)AKuCcPXP7<Q_HE^qA}CY~S3@jXOgjS`RRNs;zfnf^~*Qr?b^FFGNR2v#Z<H
zyNs8ny1IJj`}=%qi0ZZ^LdC|XtR+<!7>Wp0cDie?LDeI~yWRV;d}8SEA+Wc3l?c}W
z0(t)pi0EHy%cj)19Vy#nO17>T!Z~Ujx{S9M)#!6*U>_SOI^pyqp~>O;U~q3KNDZ`c
zfS@O_VJ^<Ech>Ci6%K62Ao^()GNbpi!&L2PXm-5goTa=!gW>J(yX`aT@iEYW8Dh}c
z?5+xyc>TE#)`%5f=~(3(J#33vYoztPRT;SVY>ZQ)u&}By>sBD;SVKLjIQriD^a~g~
zdM$m0T143JK_o!-LrUTV?%z6AG8FoK3;FDvi>jT&3gYgS8>8%P;UMpt>lPZ55GuUY
z3p{DT!Bs||Lch~UQ*|Pq+hn_mGtzWW$J2+BkFYjx^~-Rno+G<Toqam#iVeb@Dw`iz
z8PJ_Uh1b}UY{JARL<&KaiR}HQ!=-X37K^9hPJv`>9}D~9GE<br&XOu7K0uP@V&*W$
z8Zvi>kA#EExI%`Yc}T=Q6gk4mwLk7zqp;~p^<ni$%Fd~o>Rc@mCon{Q(!$=_lcAX+
zwLLV&dES-zY@0<%-*ml?A;K%3r}1;0!@2Ln{TQ}o2+%;#ni0bBBufyoGzdvnoUtS8
zmQ_>+$rl5BZp^>b{m9iPlKZem#yEPgDsh8J8_z*<(NuZcWRST+_w%7h(1i{<@qw7<
zMx%JO5I&S847yS6@koOEoxKc-TXoM3I&X*RE}kHV39br(RUVBPzY-w=i8r%M19Dk-
z+W;^adYx7R?kFtosi2@h{2}(MI1r&QdSeg6!8ut$NHCWlH0yT9zxPphBUMfZ&Kie4
zq#4UKyV*+e9{*lsrs6z26!Ls8oFWiXt9BNec|!_bil+e+AUfS_u9yG8aN#G<1Pq_V
zwreGCY}aKUT}~qAS^4`kD;WL;iT!Kt{4hXK^tGK8dGWps1bQ}`%f#{TUvCT(ok8JS
z^bq}Xz`pp$Y{MHcnnX+>Y^~Q{WzAGOZHy<A-_FGEs!o|>i07ui(jQ;s=*|i<<exTs
zG9R<=_d?+`Ig(TYhY$Wuvzr=edL^Cu(Yu<ubzdaLy@E8BcVEgey40CG5|4TPl3FFC
z2}lbdv3v9_>aB<=o7<*v9jCwyGScX2<*5P;l5oFOc^HP_B>8~5JUr9m2*h$4pxomN
z<gqD`d@5{d-HEryIv+kL7TfDgG>$3~i13`2t<SjM7_b=1Ca0fK^q!U*9})E*Aa#mE
z-6U!Ridp&!YbcBT-R0az+Z7yaD4N^I&@+LFMELvG>3;JS+ew(A_hDBEY$Yf}guNPY
z))L#9>J*MH`^*xv@1ev5qx2Lxci!1~xWdfmKbiEgJi<`s^v)f}(yEmS-M7xUg#dfz
zMrR%L1+lHSfRDECvDenzR6o9c6~C<Rl~xaOKVpYQj%}j}(xfL6x;K=fF{|tv>r41z
zc(V<Rz*rJud?TngMOYfdEJWK;xb%kd%XZ8!dn`ZvptY$HCLM2c>saI92$Qs(AmRLs
z^u8mqK>rW?(d;;5?}!uGe`&?Sr`(Qr)K*VcjW()&S;a;=1W$*awWT^m8Gs0<l$7zw
z*3q7HFnSgiCwf7|J0gT>*kyfQVTlr#DZAew{zGD*&UUwTK%0BM4FCMWQD#NANOt)w
zX6_8sTe#acF3xwFE&7=q3rya*>7y*=o8_8}T1Ncsv8XeCELNSB6^0oIsQRTr49P{6
zfO06|H3f*_#{`o~LWuo+-G<)|nEEw&#Q?aYVKmij6&}R>=|;a0oiHf-Wbw0|F&CQe
zQPNrX<Z&rxJ1ruCO!+Kau~bxBiKx1lAb4hx7}*fuyfxX@^(k{jNBn^I@KO-JV<XA?
zW5&dGycFL?LpCq)bQRUwp8-$dgUWh#?*tBnkZ6#)?FTT8{~|XukL_yVyBIABm4K@J
zAmg_GDSA9_=D4gk?fHi-5AGltElLyk57fRU>DjF<^%K@dT-M<~raVk^Z2#fVqGS4h
z<4ga4b48cbwj2(A9a=p-0ycTYr2aOsX=ecm+gh9LE@?u)N*xETy1(TixjbUoFAaZe
zCmR$EMLey3OOjMN2LuRuw`HdTA!}>$^ztc!7M+&IXULRiLcF>yn|hfs&qTsppeUVM
zxd9KUFO>8_(JbBb?Q7_0bJUPtg;Jb$d-9hJV(<1nb@Zk}dv}m?veVt|>fnu;Li2rh
zf4_gxV)GNcQ`7D9u(ZE_m2q?Yqy72ypdt@*lN2sn1gPAZ?3RztL<pMdOAc&L&Fyx6
z7Bpr;!Ad;Ae9e&qo#08Xsiv8Nc4Mj86e;*m$lK9$+Bh;!lj0h_NXQnD42lp0uby|4
z>4U&1(VywL+EPJ9SUJK<%@N%yd@HIV9H~d*Wjp!0xmxA5Yg=Vvc}a<p094{0#yXGj
z7?nU;1ZBcF0v7+Rn)*{qfMjDU^GN)wugWrjaiaZQ_peFCvd_e!wTfknGrG#k6CC(e
z)sRz1TXRK{)}CVwPO>f$b`HstXbT0GRAF03Q_{!<;;}t|oJD_`P-tz7<DTMBLi2ME
zAw>k5UEk(~{tgq(Euc<@yU_$#*|jF8$r!aa)m1gMX5snJ%C1nIL+&CLu@=!PEDiJ1
z@V=><&Ae|4;`l2-4U#U+?+`IHfkO?|YEP6>0LTMfk?be^Z(8d!TFSb{3cg$7^=9ZH
zjEQV5AJaTDm7|7iF^g5R^Vx4mZa|{0pq%iyk`<AaQcG4LpjnJ2Lw+RxX%bFE8re$R
z-5sp=9xh1`hMxY8>m}u#CPN%m``d7`GbS@!&J)MKo0bjtm3=!xTZr^Orb75B+NU2l
z_iAoN$uHn6V4Am?w$y~^pz|v(+oP%|U@(`WyFfL8HNAwN+m3^6FuP`GeFN*W`(NHG
zjck#_GaLsdqdw-2_iBYat}_hsnl}f>Kx=9H0cP<NrefL#6eoWs$*duCbOScLurf=@
zCp<D6Ep9rs`5#cKqj@H}U_pC)5ld~aQ<h&$kLi?_w{#C->GsO4JBD2QQ+wGcJo)r+
z49!BvUol<`;vg&of^aATTE~ck2QKQhyYWT_<mj$mBoWX9rZD&D@K}5lI*Q8-dOHP2
z?(gVt(SVGeyn!HH>6p6?gZ2W$F;H|kXER+gU<$k7^8@o<PIlAWTwtZv^)W@Cm{ohh
zAA7Cw!r|m6ceh>1(0?6<j(1)?z#tZ?t9cL!n71&ZLdjdmpSBc`2Iij5fb#0?`Bw<f
zTqvW8j)7m#^s54X+lPpk8P*|b^rOz02BBE<+ESf2rTGdY5)cpNMD__I5CRRBws0O2
zX<$?^;W6|NOXUa5fj55pN^tI>WyWeFvI~|M=BNvRIK)^`PQBAX9QgV;eF&MhW*oJ=
z(q;<rk?-&Rqg?7geD1Q2f#m}ZSy;@}UEfsvs?}tLD|iP<M%f72`09RBkG|8f3W;fW
zyECJ60um8{^x#~T;+ns1h(+IJKlWKsJfP;Dp}@$?P&l|m*cf_QEo`aN5sE1av7K`<
z8zLZC4cgI$G9s?%0EXp+OViu&Iu`ltqcI&$eONRQyTuX%y5dMW>&_xs<cVPy|B5pR
z{=WEyNDOKj(NOHtz>ow;Zn&%jYj6Ug3mN3^uu~_)k~P%@^?Nm$rvnh+aEyWW3H&sf
z{a`7fx5AT9h1MSgUZtG<68OzN-yRtJFCzv;h5nCZ6Gl)7sR^vfoN#ur5ZzZ`u{r+1
zveqQTHPMg;K{fADU`muf{sN+8`es01B03&f5_Yp5vI$YF-vsiXS9zYJKFJ9%9BYVJ
zL-kQte2CuC@`J{p=k~D^?RC`NA3!dY149F8^!Q&y1YbFa9eJ_6j$wuFlcJW3Xm!kt
zWAzF0axBcz0tO(G)#g_;&w6!@e;`^c7Bo%KbzAF%oizP!H3MT&udfU4T5IvskbJ&f
zpK_~r1w-Jf2(Hux|9s?Te1U~mfR)I`@?PP<{U)qzkw;a<T$RJENTuM^^}OqU_Qo@i
z0|Vf*z-x&drOJiS3LN|!(U9~`+>5S{_n`|1(ZP=pULsgSePJ6MPSH1*jWw`L59Nx`
zihE_URH;^9i;3a=ImDDxpK1kbvTbYyNE_kPg^T@)7_{2r`a^-4xC(ZEdfVE4e|i-*
z@bB!(Z|D*(Yovfq4Xh7K6${r=vZIfZt95IfI*&4DjtT=pYG;eXazyIX_<nXOaHcyh
z%i1D0ZYuHZiybi?N&4+n_Yx(-5e34RCIlBuDv%CB%&*piB@e-RyuICw!^KsX#}u-l
z_`QcIXUajEvzgy{EjX8j?usWwC>Mtx9l&b`0FbUaTY{bgSo6M2|1%bRbP-s1oC1U%
z{|`5+ua<>Gf{rb?iD#`KM*kYYVDhW0+frAXwpE>C_x%i%0z6?Z3-uvgt%wiT04M9+
z%W`nZc*X6Z-BRn1YNaMX2_*HfSjYrNV+JbznYPkCdjP6$D8Zv+Hx`!?LxG?c3e07`
zx_+h54ALwNnH9`La|Q%01>JbRM*-jib+y*n`UhJ_z0!i$P)XcKb|ip*@V&;!^e4Tz
zPSVGfF-~BZ=1(LqnQ#P+@Q8pX{*3`3k6@$18SHf~7bnX=!4EPQg43h^Q|AwIG$_W8
z2!?UZ$R+y==20Pg%hLiWY+qK{NgC5QH)s(O$<Yn%cPmKg$e-x6`D?DXhfBPDstfI)
zZ`R`*1!<LQ?xIwg2Vd}-(%*lzH>j52YR`YQ6u-F0|1Z-wJ^lY3RlcPvZL`6K&~>76
zN)AjvVG}3XV&~E!)*`Yy0JPPwYsoPulH3`e$u#u+iW8P-#4e#<D2)KZBTE><IKawi
zLw>MJ*xP3#dQ&2L!()$TT1b-A=ZHy%2|<H+OqfdJUqe9Tcx&}O(h&o4fkRJ?xRRTn
z{C4|c%7Byen-H$M>I)`FV8{0T>FsIHz+<;iQZROQI<VRmNt)0=>*Mv#^+$*+)BFA2
z#AqMFqofy2&ksl=Ua2{FdNo!=%<N1GtxnwJM4Y839rX_0g9}3<a=^{m%3d{XW6eHc
zdV)>TMMX-pD)m$U7lWntTErk3oa9x@O^iZctCdmfR#f_m1-*}Ox;u#3FU1UjAA^{y
z{4r(G!yhaTq>h@O&M`LUE4<efhU}278X_Xsc7)s4c}*{Z09HfyDb@=TpAMpFE(F@z
z4pg)-Va}V9drruFM2d%ix}KkoyvH!HrGHi^WU6xFTvrRt>SewhmgWI!+*QqLvU8AB
zcKg7=mHzLPl!kVz(+)X-iU#U8hKD#RhhbNMaH{cvKCH$PNym7tA7_b;jN3NL*8m>V
zF7{}d_+KPaI}un~F@}XeY^%YfGQtiU^O)sTW;)5*5S3a@vRASg3c=jDuc|@9$?@|A
zYYLW~dV7cl>*q-@C&N--2#Q5$%cI(0*7$~6Ce5Wwj>U?zw(?$Swu-m4o<xJm)jZl4
z<o0l$^o#YqdxICVto3qORdgsqd4qe8adVQL%(OwF5|&mm#=cPNX|~#Ed*d`4>~MwL
z6InvAzrzF%o)ZM3J$wZvg`?`o3~iHILse~7`|Y-@aNKd@#eDl3A}9z^F+^E2gb?T6
zwZa5qlq-?rhFGZ<Gvd6v#@7!MVTW#5dS&Ne{R(;*M3{4?8;nD~pM&d5q8)HKutLz#
z`t$)<cN>Le3neHLly5i$VJaIcYs&J9zE<i#WG=4AuH#u^v4gL_02)k4KhN*|anYwl
zx{|~#)q_DN6s$}rDm_&D`I83^Q8Q+#cZsXx9huc~QZ?@zz<IZXLMfMUSMVN@VHo&Z
zW=t}M(x<^8^d-TDD34{6XAlW`g^8yODjj!fX$>t$J1!n<nd~pDY%MV#<n2(icgtTP
zIqC%nIU%&9TJgp5%Wgm6V(ka>86;=Z6!0Wwfo3TsWxHZfPAMfDqou%mzsSza^XhSv
zARRz{F8}W1^Aj}>a2BA_h)_p(HAc6jGi%c{Z0+aNOwo;rE9Gsn?(($HR*LMP1Qzl5
zEtC&*o$<VB_TLTtFGq#d&6d^4m<-3Bv+0IL$EVcRFlNt|N=q|`otXNg?PyNW%^fRe
zjX}w$KTvqq`@LB2$K4_db6{?gDvr<kDvm(mAWAi1Uz{{uYR@;B_hXkb7>lY7b2~;p
zTZujM;0?UO@(sT)YV0#K8gMI#GN&W`^M@i-<s^~$3s`m-aAlr%@@kHE()E4iny@B)
zvc|9aEJqKCMsuoi#m>EM<||M@!-A*EcPe6DF;h;Vc}ivI;8My5l`|nn|AbQq=!KqX
z2;1&inFsV0+7rt1=iGAm?(x(IOd*!}Kdaq;uodaq+5gWn`IVaF{|12n+lLjCrbe+r
z$U5sYd$7IWFw1|AH|7czny~!6J)N;l*WCC6;Ki1zhGj5?WDEI2>(<5GYy{ahM%q5U
zp7vdyeu#oxh!IhjMH;6dCdB;@nw!5y=#I(gPn<3lG<X-l`)`Ice54@^;qLL#lc5M`
z#Dr{S`k%M^skS$#+xP3^=i#Ie$Jf)|+F3%0jF0n8%s1PQne+SO=g{HOWy%en9m{44
zbYKuj(`WM}6pmgEMX9)IqHn9Avgi7^OqoDfvh^}4^V^*vC4F>5{`c#Gjw-^7e2QB;
z{Sa4GdY#r5-8`hJACdml0CEt>$($u9Cy&g<g|uAXTWhqmo=?JR<w^#m`2M|DGlP>d
z+O>mL`^=NlIoQ4Z8Ll$ep;PB1RKkbMS_ach^|<}sBf)NrHo<;V9;9&kf^3l!r=Tzh
zQ8I#l351ZE#V-UKS0kCI-1V>Hh6M3|5z2fN^`E#Njkvrzp#wNtJNEJjp}<l<2A8!=
z(?CpknY1Pmr(vmm^Q`HHWlsl%BImhKe5nm15G2S|3Fcx`w6-=0EYTuAgLvR78GKf@
zrM_^T#aBq7Pn_nJ&WGs@IVHCF@NK;hzq!yvF><6#`U=m~jvNysWu+qcFJa}SjLRn{
ztS8FuvGWSPZPBhK`@7@Tsm$N+#!Y)$k+V9WG6CTnfjJ6nT25ei1(gjUJ5iDbqpT)k
zF{V-Ap}w?liju~GwUz~|LE^$uq3z|~-Xd61cZsZ4N#WFWZ{EFYI=m<Y$U>!w!o1{;
zPs=ZEa9g)_p<JMA0Sn^*Hqy_aw^AV<zr4(0d@Qi+HEYcNfkOtzRv9xl!J{zLtVbRo
zbRZ~H<5Y{IW+Do?I1nxnR8Y!ME;hr-`uI%P`n<7}c6mAneYYe6Hq*h7kr1-9FNDB~
z1<T>%UJ1PBdT;sQYFfpjnt}wI0uJNfu43TFY|m)8@cXvRqlxKn4Y$-YZ)loPQddf&
zpwT`#`kWF@@)GO1JaVFUsWaU<KH_W`E1tS$+g#rN5&nBcwz6`;)f6HK@#MINW_CDb
z?A8U8q2xk_Gzg*qrB~Dv8o+4H7{Ca)XJozETLj&c<YBQt50%L-^CsRwWng4U2$Fy=
zgr4?g-8bj0mVzo|xEwsp)=D2n3;>$2OGt4I#L5tCZHd?3wIxU$ekC8VDoWjQP?(p-
z3qWptN66NDmuh|$RwN=6X4~KOTA@HfAU0rHYd@gl-<T>0_6bT0>(ht)gi>VivmS=0
ziVtDIXkUue<Ly_QQ#eY7k!lVTaZe4cGA%#+^~eND3n&dbdle4CCprjvr{r%y2cUD{
zjUBKCu3IaoIRasUn1A=Bm$ePOoHx!5v2G<UzFH#KA?WA}ODPOd^_55qbwyCgXj6$+
z&Wp9)LdMn<Gs1QV-GtXmsQSzVRD=uyx^>3u=~u)ZwdyA+icnG2T~H9$H6^j1k;LzX
zGaHJ-7AcZzt>_;l@w39kU?h+af<Q{e_3RsYFS<8KSQup+nE?C>@<$O)7DK8PFQ+jb
zf)BX6sr~?Y6A1+O9`gDX1e|(C_LGlC4~1m*9z??j*pu*j=Me^0sz_igxdtc85t}X-
zxpV7Dl~H5b6kor#Qd)aFcCkC;X%b~N?4+_7uO7Y{?@uyR>xFyhjl0;&6HK7zO2TV(
zh@}(vpbK{^Ers)MWbQhc6h%@vc`%&~R@MMpb;sfvt&1;l2%J%BNTy3R4S#`)cOJN!
zTy^{E^%*<U;mCyDEG`QP)IR2izuq#2Kz!?|OF&DXb-cQYOz$|b&A=`LH9)_^Zt?h3
z*+$WCz@Z_hl<whE)yLF2oKyRj7*ZIZh$l&XYxc*qLu^~^`Jc@7AnlVAIoo|C`sJg|
zrO>2ahC*b%1PY?njN-8JsDK!AtG8U$>s--?ZBl4*wMfF+U@iQH;zS^j*NqOFTAHo>
z>_8Wwk&o50gq!YSrn$SADmvFpf+>QAbbT1a@?MEq)_4D7S`lFzAR}6n&gWD=L!c)c
z4d*EAKJ7xeihn@;R>n<$|3*-a5q$0+xDY5!xfz0##lYUJ-XQn=1A?;_cM1^8-=Fcl
zcyU=lYBR!TD#(J}16JyqX?EAKoY!u^ndUTkbVPQUNQSExYlaZc$|$ik!06&*5`j3`
z^YDeiDZN-F)-mS3MZB1c5eCr{be?R-vO)}Ywb;2ffA`uEWZ!QcJJ(IblQ-osdqwg`
z28w6&QUNezKLE<k#LvZL)wUiDyFgb-!EI^4JeLWapZgw3gqr=M2b%HCpv%7pBGy}>
z#ZNQj6<{#>7XNvHQNqrH)@KXWNiri;+=$>ss|ylv9_ffXF{hvRS@U)6W^0>+!wMct
z<f0Y)n0)=H@lL)|#8G}b<eOQt902Aaro2}SH@>P$g_Ny<IH_!V-W5vmd;q9+m<EJn
zii_t(%3bjec1i551=qg8=0pa?u}mVtC6P5+49b|E7Zijv6tswH@mA(#2H#3RC@xW@
z*F<Y~_s;=!uT97B54kOfRbP6w4?knOJ?@{n{LFZ~!n4&GWk=?cn59p9J+U`=p#Z}Z
zM}ij$F)R8|ntdaw^#9#bx-KRl^C>$DrkR-f)&L-z+p2q#EaZ)3vC~cQ0r;lgk@~OB
zfPvvZ?m>*K|7Ys*lIoVtZ)e~$qmw6!S*qq8ez-l)vc>U8ME*&aOAWJLY*k%<C}BSj
zH}L(OB{bfQy=K>dvJnd|iE{JW(XoSb<6j^gSdc@T9YvGfeMmJgM#dB-@(@ZAw+q>?
zDO8~LNMIZY&lBbC^&YR=)gnz6Y9Tv!>GnNw^mKaur7H#s2FQdF4l<%*ldaX|YWLJn
z+tc0o%eMRcjTqL}>G65ls9;S29y^$;aIv`{qOk>7`o_h_q6|lpDj-M_I~&CNij$v1
z1j`K+hAKo-mPGWYTO<5BwmLgS4IR4*&H)n?vDKum5>1zp(AZo}^T{~QsHEgP5f+Tc
zFGl^8AKX0!@B7nXY25SdI{=|*@JbY?EJjR-q>2wJ_oqs|m5Bv&_ke*#Sg(62M%=Sw
zQWA?0yl&Hsm>(w&MSQH|eG|=JNFZLs*PlEFLqhD<t%(aZk=zdiV`a`#p`%{y$j-mX
z+MH6uy;A8q#U{nN{V^s&pU*MVX8F4*dIVAf16%^)4cJl-(pqd~?owPYxakxP*m6--
zM3a%2{0ZxCib*y*a>nsm{v)kYG#vZk*)IDa-XMWS8m@wGW+0BqOB5(?h3oNn6ZJUt
zdK1OuxbiLH{R6_^LOjHCJUy+fwBX@^wz>sMTA;(XXLR2gNe$4FW~}8P3p|x&!@Uda
zWgx%yYdPPDmJ#o>zM;5%@eV*x5}b`-6h<N4g;;VSW;4ktx<0n=2}xOj7h(`Wsaa(+
z^X0PN4aIR3=#(_a%(Wrc{d&)%-Js|N&p8(=xhJ#ArF%qZ?DJ7%A0|3NuB!f<`Q+@z
zSVa($SjjB%<Y~hL@pKv97|E9qSb>GlVU$}?glX2b#^c8=S^zJF6BkH2F+6z0;*G>`
zwq$a-O|r2H%k@~gMeyOk=)Vh8)RrBxAGzN02s?mrvHr&U@+w}?6^t24R~gYZS{W<g
zG#mW7?WCEawjuz-XU1LWtUS-nKW9GwTsxl`uQ@P2e}KIvF%ofU;M<_&thsuzs+Tho
zFXf-?dksAeG|frPN@0oukiUp_SG=YNL*>v==04crz5Mhq$xkXiHMbQi;|L$=U1xA(
zUiOev^&<3iE-$W5jKBv$0jGT!IZFeL+^M+V4W4i)<`1&FWV!#bxlfz=J5a(25D_k&
zf%tO+C4`X7>dm^yp_2m|C5kONQZ{XfPEk)eZKiAVP{?+UN;TaTk`F{54+n&lC)HJq
z`3^d>Sg5P0x?TmG>P)(Cjh^UijW?D9)29nY>iFUhoM3I3Ihwoku@2Hdu;qZFxf&?E
ztQA|Qo#YukD>ik37HV)oL36l6h4hC)-Q%Pjv4P?~E5i{Y%w>(wIZop7xDm3C(&>=w
z+V5sMtt}Wm;q>8Fv!dZ77B;7>(aJCacWJ21HS6y=NGwGYlrR>pS1M>^L?M{V>)QOx
z<vUGp{TqzN4m>rMH5<L#22#o(jf|ixCjNRv)`LNy4nx*iK2K62cXjF=D8ZTfHwo%Q
zawP9aP^AHhiP$=TLeTwHn5A4ULoKR!A0QWR+f*Y$Sk(f0_|lz>d<;Roj`Q%bDNZjx
z<_9R_&bd|<Xlkx1s_!;vIJTHQ^)j5?hd%@lbXgye+eb|A#FKErZ*CLs-(Zw59<{re
zn8&kCV?~a1E8)B_=0;E2YYbT+Ec)g{F|6}WHo4WRxzi+@%K?@dNiPROwWk#pGSUhR
zMyUgokF!JrBR33wYUS~lDPrp$KEY6_b*wncQ{1C0^oj|6z5VM$$f2-hs&#_rWvBB4
z_Qb%YdV~fquh;L9vt#)H9YNB!;_$~2aO<sRbLn*7hOUCn_djd)e`K1{Gco<2WA>KP
zrtE)fw(pHM6mAhFG+-zmkSLk7d`tC~c#=Z;TrUy{<NwSYmxT)<H}L?}zQu;^O`Ol-
z0IeOW`sg2aqHL<ap{72M?(UE0IDM3|Vqvl%3r2jxy=i=@ce?EmJW=HQKv86xyK@&Y
zB_;=<QN;$+q6rHRcf^OTj?VWVYhxKL#`hazAulB-WG2sZXEio6|9rYzU&khQytsTA
zxx3hY3_ji7Cn{Vb+dmkh>CktRw~|)`L|5IRtdahRA)q&+<uu+W&<V&x7ATQ&WM~z?
zeO^>beTT=4Ap6~!woM%?#hVzj%+xpm+CR81^VHfeIVTMJ=tm<8+zM*GiC@9GG7u2t
z?}|5Q83>_QTyad7*IOKMWUIL8sYIxjIHwve?af)=P|6(i@1$Qf(_r;=h>9bSVx!pd
zo!c@X5(?+*wu5!S+ci<wBklLGVU+m{iP^9WrH;ydbE{c6$UhxWw)Cjf#YsLSsi|Fv
zcqbWhG{wZ%*rjV8z4fyR;rn36RI=s~h6-=Gm@SwK0rrJT2$-Wr3#^k0dZl|AV?q%v
zLw%W&6k3o9NQ@~TO8(u?Mhkh|s~WuTK7gT*&wl_K@iHE(0uv>U?;c*tR*5ck+<x0$
zormbtdXgqnBQrfB%_c)ACJnnZYfJ+yqoHFCchDL>PT$rXl6K9Ys7u!MQjLVfhzh-r
z;w{1OxnNg*(ZyG<{r82=hG9TR(Fys${OuLVZGs{Sq{zJ>ZU&Su?K^#dy18zFRkR|W
zT$6CD-ESF}s1&DZeN_c^mQmPlzi8)BadC+MPhU1npls6GEZ&Q|`OjKE+&oiHo%ZJO
zNONE(y=2?hZ#1x|X|_oc1<Ue46wM<?^?rZ)L<&w6=Mv=AeIX!_g)e{!%o-EHZX4mT
zr_3<ZAFT`GrIa#eWl~wTDTMN}`ed%g)?fjH(&}vyCgMCIN({nU-GQ`o>2f~^S)7b<
zwu$|v%lTjfwCUYklQZWA?dPjM0!Dj?b~OPsP+tcNCmJ-V{0`s%9`OVOWimlMQKPs+
zYVp+jH*m3&<kgK@jDrjZPk^r1A$lRD&{}8F=qh$=GG}{_MqZCN!vYcn`ups^3v{O>
z@v~1p2oPF9BS*OwZv+^H1pM(~C4IRjpCtOT9)A27k4(7e0oi9{Blts5c7L|}cKZ`U
zr%1<?TNMBP)mmu<)3U3h$$>9o6HdG-L1Ds#CqjirR1WzK2L|Ogs0=8@6FG$xTY}Hf
zR{qlQBJ>dVB9uefdoH+f>z#8rBblH3J;L?(h?JP@vjAU^t66avNb&$vHqXFM`!`h?
zEJ51=um?mql)pQ#3BjuTk`8Qxh=lUk`vnd(_+cQu(+@v%xTpwGF59j?flFVA)?SeG
zf~od9_R=F9H0we}q9g8hzmM{Qos<YCAIqK<fEm?lCO#ZiSZpYy{+cO`rz5mDsZ|Yx
zMN1bU)w*TGx;COnvHNJVN~KC<?B!$l?cfC}f9@Ho%b&O^*|}-~9GA3zt$a^rtlqq=
zu^NKO*4*SF8Q4hc!fxe%0VMGTBP^F%$mF2*w#fSYj2~7{8uvI*$piO~{f@<Vy+Rha
z13h5O;jKvOR!NW^zZ<_;98D=QZp4f6TOOm-RMV+6JXH9irllS$<?+;uy0FJPkRl4b
zt$is)GdcsQq}LtKXq%b*!~4rl9=_Ej-@~c5R4jY0_R0oIX}!9o$v1^C<E~Hk`Xolg
zrhwq^a)sc?+aigkgUEWgIswUiCD=Vf()Df8sy}`Foi9YSw32LE2%1eD0Kus1GQzGj
z^bVh^s$Y9$Xu&vvJvX!PQYp1)76vqfzj|*8XrP5Ku||*R{9|y$gGK2QK`5wjc?0wG
zLqp>52IdC^GVn@_g48rRx_G<ScUl$-d;pyA_FDd<;xjTZ{eN`nif*<>__Q+m<_eBh
zP_+N>>FMbHXMwPHaKvY4VTGcVGO{*tG{ygo#QHxQ{Fc=EqOn+Ex<39sf*QmT3PY+M
z&VqUTMJQx%gtaG^FCa%Sj4%W)^w|-0bGg8b;U1qno)u(C6q4KbC=8BgmxH9Y-N!J8
zly}H2eBA+3d-jHsItBt?CN1ohJqo-|EPpx7EyJWBa9D#3c2K@4?aR;Ts};|v>I<Lw
zhixuRBbKWQX|c>VXeC8p#F|x$N7|BAOQgKcPbiku1<&YNkqP3Z&TJa#XMfMHC$_9i
zZW}!Kq1zgd^*l9roT@uiYK7vXuSAundUx8~PGLXrxQdjs(|p|^)Kz9HkHr-ZSY`cB
zk1gRNL2nv@TzHaInloy)z5tCh=}^}Qjv{_z3JEEWHWf0S2@BJd)zKc(cDjmR+CD>8
z@Sc<`B9^Hid_b{kDm2(}aFA{w{Kaomv9ccm>@T}$FXhs400GiynhXip9vFr(WJg9B
zU7@*Zej$D|4Y8l0H2fS<C@JU-VH>93qUK0=?4U~|1_HI!$mzr-fFhOgGBK6hq+P$R
z)n^LS8AKOIQ80485+<7&6~Z5>UEww+=EtxGf>21Mr!0JBMea|K4Y`BP01$_uJ2IfC
zeJnKFswGw`RbEkFfU;mI0Fk<k$xw)7Qg&h>9Yzm)pyt@2_0qt~TR<cNoJ~l;>9u++
z5zWJR4Lq2X15MaAj!a)J3^xhL`l(b6Jo!-Wia`f4E{-7T)c9*_Yv&KvpHDA{50#!D
zx7913&JS0gSJ$64gBHRbotx(x_?(+=Pj{bJ&&kc@mxaBU9`B#M>tEl{9Kjg<*o`yY
zJ!C^3hJk|TU4GRqL+8<f+a1TH&)XC0Ao_0Q-a!gH-nN#qef+rZqPDV-D*R{~+)yg`
z*{s140q}~jFr{0l+5QBzE(|NHp3lWe0xsDmp*IE3XrMO5QQ48`>47Q(w%usMNZn!H
z=c0ga%KM^#CO#&vI<*)c)*|6+-VlJ?ktvABX3x9h%lFBEmn-zX&UWxu#=y$QZQ~a-
zVxKcEV&w&qzk{eyWa6h`E`Lx8ssf0g*p0AoxLv41#1A4HA#B^_k1<7O=nES}F-SHN
zZ?XFyF!W!9nTf=SBdNL!Sz_`THici2X0TY1pqwK@k&&>>X86odK4BV6-TH4Tfu@f5
zmi2ZH`;f|m_vU5b%lueJp~JQX{EbuQFiJ?erCMB6NNE1aMuWrPL~6s2ptA&ziMS!R
zWDJeAky)YR-or)$A(2=O@hAm#MVx#(e|fd2gaNob5eaSBgsfbvmjY_-wgLBx>H=8I
zQTs@OG>x@QvAyC(ZCCcVoX6jhFhU9*Dt(QMmTdzAds)pf*kO?;eg~(*!!#Su;wQz!
zAREt_J=^nDOneJHuWxhK8O2}T%Q`bp$S=u!>X%0(g9^A1tB170nAnM9OtIuAe<nNF
zG+xt+m0*wwV1U+RQ)R$NR#fyvobDLQCJz)E4_4OMHmjKiQ$oxwslPgTXWb0AkbSU4
zy5jmbpv;MUzZ>s&&Y3Wjmk1ZZh8B`(q!uUn2LrLIQp`yAs#Yw#h{%PU6gQHivzS~k
z?<!?ys;-UW&xO>J%K5u#CHImJj*3qbm{!7$N;SE`VzJVLQDYA9TpDA8H5;SFOq!a}
zy?X#-!{~GCrn^M+oOaCFWSSalW?GYF%|P*<OO<I?;-v*(oU3-~`-K}iK9UnuDDMO<
z*qSQEM~z&7?bdrUU9}0Vj%`g3lV*9`GLfb+!Za2_Gj5=jW=}jE{YHKMF6SYywq<1;
zO%LZ|KR2~NPJupBI;8VD!sVhDv`6$y-FD1}yQ?J_?iPT|h_ppz;#bNhfwT|J>NnyI
zbwqWSSa$HiS;c{QXWGo~SRL1<_y#Nb!aVQk_7Kz65SU9vTl)tz!P9;9KPo>H<A1=(
z=vn{o^O-4C@n4-9rTawX)zm;gnX?GUL2RROfj=epR?oMY&M>2SX*^yuKA{+6(k4eV
zPF;{wd=AKgL72BYYG)%<Gy@U0i^GG3)8jtAK`vB4{K&K!m!2I-lW=m#wE2%0HX|Wf
zj)@PCo4D;N0W`@zL2apN_K%$%bH^uLn!Mj0!2eX8gNc^^mv39IWk*x9tIgHh?0>#(
zp3a|(r;UiP{e!iOmz&7Y{|9657-U)0b!(<oY1_7Ko^)oVv(mP0SK791+qP}H(ni<U
zx4(|teJk#Lf1DjF;{4dLPVAUx>@}Y;Cd*|awi#<veX(*LVWR3jHT^iR4(>RoG~Gl-
z>*l(Ok#D@!0lNJg8W&D)IS{6_4%2X`B71r~&|IwgFa}ET@QPxlXoTp-y;vswRy;8}
zw_K2h+@#$dWBc+1^8_6eS))iLCk$$tcZ%pcsph#mauPS)iE;9zQdb(oKI_JiB6Eo&
zqfvGVPL7ro6LR=+C%?Xb0dI_rfKRpz+3*1;-4xS}HYMVTxEQ0qowSdqP1!YW8G6e-
zr}mj`H2I36fm(ayW;RQm=AX@NYU%>uf-3srB=lB>?$83KZb1Ffiu>P$xahvOSo*7~
z^crIqydyzK`0{T%rz8CwGebsUC?7z|Hchnw@=?YB{D&57oBA*A;cO1oOr56sy>llY
zWhX@>Ro7J!4k1iwPCtpkLHrRgs@2r1-Azbs{JTb=%iG)lHiVH1-u6123oqdS`{oAS
zBh_C49Ow$yvp%+E-9sIt;1NXAv{gK)emnCjR$m+~R_sq2is!u5NLC;v1&&oU$N=bN
zH4oh!1riW@Bt(N~C{e7l{l?<+w}8KmD%l5DStO@d%0hL81g~2v4r5_e;N_9_oySFu
ze&|yU10g72xhlL9emKjCHbLRhFaJk#qyR$TW?rojL_LT<1grS=HX~Fc_Ntb|pra+A
z!*xO-l}Nx51~n=6!4NO%;=vo;Yv(Wu9)j6}y3WY94OY%DjDEC(8mkzefZkAGJ#|^9
z3vgH{uwjv2Cq5FS1Cd2MM?4jnFhAaXbaO$4&?AOcacl?DQtVAb{XZo20)HML`oGq-
ze`+h2<1X!ix&QXpR}W4e%WZd1QB$M0#{OB|-rrJI?Ls4^W)nxcCaeGLf+Ze>01h-_
zW9(AGTX-$7D?4lhmsZ+HMN^0KP8s8Z9j~b0)6k^SQ_yh;kkof>0g=UFl<|Ik26|xP
z6I`fnwEnU9^;8UIkoxQyV9G7uAQC?~kPP05m15np-}X!;p|9tF4cS}`?E#K)(N561
zM%mMJaj8Jxko7c7L~4N>%7a8gSOX)Lxebx5Bv^%l;&~uNICfTAluyVz_(cj;H)%|V
zDsn~oGb${FZdGD~aMi6<2;=YEICMw7hu)?Rt0GxwjtdB+EL;_Dk;<2IiCzULeYPT4
zrcsro%mc1-BC7BlIYpRYeLH5cu=0DpxcuBAD?Tru%4T*{%Nj0Zq<oK8Q>j3^4^LTI
z2{5gJq;h&{!NNP_LO2YL985GHQGJL+f4IL4wg*gzpjZ?)Fd{<#PzfERl%&t(r${Gc
z%*Z=N40LQSPn0*>0vwH>HCvb9LUi9u-W)Ar(;X(qrREfAeq$f_Ef26gZtCIe-ec-^
z`2|!jBD<mL60Xz0%&658BoyVItLH=n(L4zIQ=LES_|levYYS<A?}FkPhp(oHGV8tz
ze=ty$%%|wVpgdzVu$IvZCx+vt`t5I>5O&wl^Ra0ETLm*r8xs3v)V^}C_Oy|t-;aFe
zby>Zd@NecdhiR;`!b>$aIi+ylazdB#6=2~A(2#wcP=v+U0VRK1DtLSi;EjkY>*j5f
z@WX}KstRhkoRFW>-*pyibt?t9;ws7PvJ#ktz-^K9hz_w=Jk-`ysH7{PX=&?EJRBR9
zvnA~Y`ILnI&#Ob(UV?k1jaLBV6vBxsb2dpJLzvWM<4P3kUcDwTr2}K-^ZcE(Evtj0
zWqJ;~PohTD_jV;7bLd=raN$oc>C{1haL>B2)h(i|=M~{>FeWM?VI)sK`s9x+qqAXf
z;j}y=FIr?;uT_LT(`P~D4vF$N=pvldn;%Fm^iC*s(R#X52hBZmisxV2Ei4{2X8}|Q
zC||q9z^~gW#i%0np!qt!UY0vNUypn?!Q>#nZGUx8?g$2^PhvQW<+xTD?qngtir(13
zU-yVEG009eAV;zzMrR5fI-k9G{yfP!jywIrf-Oeq=kn^fy>M=Ce4QW0Gz2j*N(plA
zQk;^jg&8&x_75{bqT45>x81syDiQlcTc_$fDUF7i?Cb$k4Svt$mxq7fYQL*!`XJcx
z+`(fw@KOH|er68UURZ+lTO-(jjWalufM|i_-Od(n{eEq}ZFcYVhMLT?k{<fPU=*<3
zX6UL(;r`R5yPf9AU~2jeoOA&O{m=i<f6))j|M_{@f7_q_i+-3=hvJvqH!E=|UvA<l
zbC6=e1_W5Mqn5=d;+ei3bPxiCB~nR;Ei5OTL4(!y(&KOQR|W-U072SD%Z*6XOhYI8
zi!$dcGIqnOqe9^mO^PvQL;evFR|tX-LE5`{*g7>pytm^;B*u-VhZLc<bF;s@7#h1^
z*?2uVyvmq68Vk<A{`7KTdb>S8hujeGb$#{VZEyduzR~|!sBoNK_h}PU7e!2V#K9>b
zxg>^z>PPLi6HG^x41h}Ma|3QJ@KS{bAfE?MU-ZQ?5e;aviEykV7!AqZCk08Th7A~$
z7n>(9BiJ}q-;X~Vx#cXkumgnEV+9kDSQO(`q11E1*2#lp=vdem5`w91=e3Lz+DFNi
zw2z0VC+49!<u4?gHR-E;M?eBopbl;rHh5*`y{ZN9Hn#hlTH3MGp?bGs6$YVNdX>w@
zDHbYIy)%{1Os^_Uhq9>kA%t6XjvBb>?nYP;H^v2(+o(bnQ@Z3L-NEDHyCd>Lhv#=t
z94LXcu@hUFumWVkR@j1m<;SZ}BmhSY@Fba7q?=FtrUn;yLdl%KH^r13<2?outHnCO
z6;oE<9BHxX5@<R@T74l`dmOLNcR&|27qI9E0n#n@;9qf+Q1uLME=Zm*vA?C#rtD#8
zZZun_=CM1kX%(s)f}e)X9IPYxENysbrp%<7klm7C+yt*$q;4=toK40uGVV|936i+<
z9ONa`VaO^9DlG6J7?4#-KHvL<oQPdKR2rCrH|&byx{qa!jQ=|`tx;o@=s7EU=@@%m
zG^aw-+8y-zdBMwZB6+>*a4QM=(IU;<Q8Y+l;4162F7Q=w*vN5!5SvoYz;nQst(J~M
z<^Z{Vq*0O~Os}2(vhj$>b>4^{Jp_~4>UvhzI<;ZCK~^x^mNgIHunuQ_rZ)KZV38~&
za>ZXTv8krR_7^<#bk|&93t)X?Fg7KO<8>gTu@A&P(UPs_I*Lp<Xs}A}Nng&dGSXE(
z%|n4MRZ^Yqk@oRsO1Y>bjVcxZMCrEYHhGEu+_HXT_zV9A|B~!CNBqmFF3@WKua7mH
z9}>*Pp1siujqqg7^=CmgRg}#<@)hkeB^4v{tvYj*F%rz*tKSNUFnC1W*F~#8<pzPS
z0vDAS2iePAS2wyugs6r9gNHv*7sE{^W3BqFd)hyL^t1&$m7Dxh&8f+~0PpJFq<^BF
zUff)|eM6P^wf5p5n^anoQfr|4P08sinZ5;m*0FT?7aLo&YLJ~@vR1O(<zF+=X@2#;
z!PyIAJiBbg@xFuGl}8*lK=?pjIH_>M;Xr1hOg27ARb5M4GT&Blc>=7xYOZa;G%LDn
zQ+8xboVWmtHSBe*R|XOXDao%_=HJ4!Rn{?%f-TivpLlp_?$(cuOdylXm49eo{Pp@N
zd*}rCen{piD1<Ulj;eKRqf8=wxV-cbe}{)>DkKOUl9hQ<6#h}&;wj;6B!=WF)OP+Y
z<HzokPc4AyD@+z{yt#Fk_BQ==`mS?x@VPGc<hLUpC5lT{=sD=J1h2=Z-Q?J&()8=N
zrQ*quhr30iL96(6s<|FT2HOt%J9kN?%Ao}m@UXwLX!h19e&71(f&Z1F?Yo7Lh%r<7
zzUHg&T=T8?c(wL&YHPMy5BOK5Wc}CM2Gf6jj<lt|^IzRypa0PfR{uhOJY~IJCejY<
zMF4I8OU1735ldH-QcQil;fakW<MKE&lBhqCrt@xRb@TY-nBmfb;C?u`@bi0q$2AZN
zg%vVm&VZ7G_L22#fgNbvuo;QMbw~OB{c^X^Fc~n2VmG(y_2TTthOdo4OwDK|7zhC1
zM)T*y(9?@!*t|KoIdgq5ItYZQsjcY#Wyjn(L)|&Q9*$gb_F>-(3I9T|0qXH<>Ty5N
z=P=7e+hd!k?u1C0;7f+O=%c1f&WyUJOI-G{ewD3M(0-!)-`R2Q1~n5y9%;wUO>2m2
zqJ&|g3WJ1U<jKztIdT{I55>miECQ_VpC4v89rX2$jd>L&)qPFB`NhKmZa4{>5qZ%=
zg^{HJ``lG4#yY@0ywJ`zIn;|4DqMurv_7dtF0rde?RFE7*5~B)273_Fa8Hu0jBV2&
z@omy9j=xY@bWJEf*a`L%^1}0j7IefbTtsjYBU6b7ug#Xdb=4Q2+>b(sJJ~#}=#*Dv
zBjX{$h7W6&3x6b;rT6`sd_E>dBC@AQ%OR3O4#N7=RcKD+tWaL7S){ZOz_Fxq&JEFN
zwFyOvDo6&*47Jnh-YT0RtU(drZW}-l#-9fxDf32;Fd#4I*CUJ>k+Yklx31~6lL0Fs
zDHGG%BNhz^A(egmV@~}$%_VfJ&{(IkVX}2Ct?Yb$v(o!MurIL8mWL<R+DECy?!?uH
zE~kwu%Xmm+2{q114nH<zup82=$*#3JH+-s;OaG~D`g48QO!KgbCowHU{p=^h0yiDq
z)M{ATFP$ox?$ygOi>#Gx6-|1YMmnv5piX_^UUY62VwkxgZmr*-=)a%Wjo!%bsgxlo
zhvEE(k^hv9lxa&0ayfh)M^feC8TKKfWC)vhB$?~Y$gK2cCCCbq!_6ij>sZJO8ej!I
zlnSL3`Fy-t4<y*!oQ5@MHzkEpCWK1uxu@|EY|x=3%*+$`6!^xNbUrw;XkTP;wwSB<
z!^+sy8+dS(P44^+#Zf6tis&YPS^d^=gwKQoft0twiGvO{Kogh)T5{r0n5|qyurat^
zw`d#qOsUqIfeN_`nA7Jn8%?Xbh(1$prKUEMUU(Yu$u<zotke4})E!m2M3nay$VFlD
zEJxWoBqk^Q$#k~z=hmHScaRz=5P=Ba6Wy(aVc==1O;Bs}D<r`TnJ_Hxk7slic~c^2
zU^KaYiKLxgeD<l+R7|25BA)bA*X%dhlt3ldp_>;KAUYssg1<KjRH|ddUBY4aB>GT2
zx9aec63P%4>FmRNWC(|-KMJu6E97_0vY|${K{RIpgZ}F%OQ&B!rQ3Jn^_IIs{f_ia
zjYHpVjZfcd9L6|f|D|=AG{*@CEq$WjdoocpQp$FSQn>WY?M1jW!^5|#)*DccpcZ9F
zve`_pkMpf7obQZ(TVu704RzMVWJER7wJX!k+U?;(j`}Y9H$v0Q+{-1w30=qjKdaS0
zQahPg{vTH$;s4nds4~?7(=Uh^yyuf+FkZMiHmze~Dg{oWu0hLgKr|Q{1y*-+J7s`c
zDfgJQ(}#v@74^$STV{pE-*xm7B@qdLX$Xt>a~KzF9B9PE*N6AlMS>;#RoNv!xk0kb
zlKKo~dM~+n@=#^*nFE1M4UA7;bxsXh5}z6-xQYoq(%A+~p|6qBB#s>;Wz;`6_m5;N
zrvKjCf3P_g1%Mi(g53#ouMjAZAtm_Uziz_zk6)0Pk>x+;f{!$$ZE-mMiwm~0Wj;6N
z^g<Exgqj7Tnict#bADLcynF{TKfW|ze}BQV$<d}I8GXQR78ginJc>2#mHBD{NDvcF
za5ZIpc|X3pT>A9`+N@tN{P2!_j~Zzh+1%NIOTOA9GwwbnMZi@$z5=yWuR*tdOS^mA
zWk|M}S3a$<q2>L~M6Gh7D#7<r^=1&cBGsV-Bo8)a0SuWwOgKrHIv<u=BQJiU!@x7!
zLE3cf(_3dcLgK^Z^3ul<-lN%v^7ys+3p{B#;Qj-&sFF5w_$}qpny!9|;4X?~{Im6L
z(&^iKBTWPPvgLIXD>9kwDGxM=Q0soZc?%C)-{j(R-==G4NusLKjzCqTZ6>+<w?zy{
z-A`s$m-NC&lxd9<h0xZ-`LztM6frM}uqI+Vd^FJ?%iJ|g$Lm!!{7t5vH*yg4lgqjv
z#i~%k6oSwkKypRGoULvu5Wczb%|9LVQV0WXRZmm*{6^4Ef6Bj#S#3;fR)-$<>)PY7
zsUa5V>Cxk|hyyl%h72oLcWBgHKl4Cxkmn&dwQX0{8{f83VpM6hZ5bPbhHgBPtIt9j
z>R7pE?0e*dFdm?L_sf7bP-`s(RA#Oagj+X=L_K<K(?oDZGaK0P2oMtv&Z?wC%}IS=
zc9br9TalUBDu@y(h2Gk<TtIun5V^__fx*;#mwP1{5Z7xSd@_k*D@MqWpvwtWat5gj
z6`ta{A?*a1b8+86MNNtrQm18o&O%)b?%I?~HK$_urGp_t#;^SNI-Q&swJ1&06uzA=
ziBw-EMJoLq2xDWGywxeVTMkq-Lky3b2P>i}^JZA8h#NEuIR+gC5+0^wYmVO+9rrJx
zio%SXJN1V`(Lr{*bIPrX(L_XvEvSl&VWNIn&5!$$s$8|mZ-HV0ZaC~d_>4>pM8seF
z&V5a0d<3mCUr+!Sr5Z$i!Br*&jB*TWQ`!Ods}q{-0fd}4vaWnym+a}9R6f%yOI%Sk
zeTbKvRSJ26N0)B`zj#RLPchA)fOht$`!cynF<0N;gOdJqx`d3SwIzx4$=h$AVdDNO
zNq&`h8d`#&fWSrO*U!t9beUI*0WE(Rwq)x;$`2HHiJxK}Uhki`5jS8Q;Lpenm`kNI
zc1)uw^lG5M@GA2m?B>4h6Ex^aBl|ORF_GSAK%x9dymFTVuW^BSKX@ow&{mTk)9d#H
zWc<cV;+%4PD=0HjJX5Mvfx&;IZeKF0$&ew67!P2&k6>zKGIcZHX=@yJ5S?)bd$UB8
z6&2yb`6o8?Z^*geipwEMMJ2y_^3`xZZyw&>F!f@}M0KC9K(3KG(1oq6{jwia)4N^9
zzOLaK)xZiauXLyB{s@P&(r;DvgR}12O@%0&F9~yqCwFm&*DeL6<Q4x>Ni^vcr+>RH
zQYY@*baFnbUpZt#2t_co;J1(Blvgn2mni|x^y&t7@Ap{=>&4M$&cnA!?8qkH|6`;X
zsoxN9@COFKib`gejF1aLIR=9~W;H+dtODpyV|N!QRi%7yppwlqhpT=?KP<(%eQZ0)
zw9A|i(EW>a5(Ueea%m~?rr4?<44$o1dN=5X$Uv;Bp>N>P7ht8YyO7$w7|?I#y1#<L
zJ3D&9K~JIV?IwI+a4{_A78N0%Ot(jPJ{|lU{<J7|mw5#?bAAM>@%7ia;V`Oqd(0Wq
z>yD+-q$76PjR`glFJs49)EQRf;Ko>xrBB{80Ke<<E*Tl1*v9bMe_y=}2rQeW)P3xe
z(gWL0+*O{zc4<mh{|$a&tAzrw4PT)y@!%%^t73;wyf2ObpQ5jlU@tn=c3fy`dk5c$
zY<LP9)bl_Y|14tZ_s<VRR={K-jnja&J{Vm&<FNi>NcBFjgbCi!Q#l!AzB0=OzDMG;
zMe6i3hx>*`Wni3xRroLOD3KL=yRt>hPRMxq#$iITi;CHZ<y1Y+;m9a8Gg!teBi+0G
z?%A>EApb_IN{9DaE|KU7#jRW}lxCr6gPuwDp_6m*{TO`g+jmUf$3OS^?jrWOE?3DY
z6~S<-9F1n7cAfIdb;$_$K7N>6&Q@k<`0Goa3bK4|1T>1cp7>D?qU!FY6(c|vm4#ia
zExf=_4@@h}s}_V5Uo$d}ynkKCF}Pq2e*MO8v$*iT*sS}RGFMV_|3)UK%(jbP5eYa}
z_hbb*RL=5%OkYrC;u<BLy&@X|a_p?nw+}=G^!9IomH1}EQ<e!Wfxp1djOl1&B3y`v
zN4SK<bU*$4n!apiaG}FZW59lofQRh_9I%t-9jjEHs6=!+$<Ga)RGuhbr>&FdwyaIr
zbWxp~I-xVxrsLZ^Z?@8m-{KH6ip#tQwQX}0?gLwjx^zvpV9Y@dT90rse}BeLU$1N!
zsPVEwcTu{&67N_QZ1D$LxXig#M;7B04W44d4rI6EjN#H&ltk3LfhdSS&xx|2$q@EG
z4Mf!-uzO>Al#^@7C|tq%f2C1e7x&F|TNa-xtNrz=M_TYM^p|-C+pHNjYYp$KU8}9o
zdwn%Hw9zk+*H0n5664T?>nc4D2SwoyL>lhuVy${+<4=?RWgK2)gxy)7OK~@5NckCw
z4h+K<ijiB<N>PfD)rTSQZVW58#FKPclc0c%na<!ku0C#k&c|Yzv3?}nO3&@;$ik7s
zy3`)OX!Q@DAb&aPm0T_|2`O~{G%j~HA4f7!Z8C=Er8>m%1s&w{`6oZj_79oD#>DzR
z@;CnjnbM^!Yr7zb+WD?_ITv>`(cPlE4f-1~z7eglkqr{nBQ%{{-2xlf{Pn%FsjE*8
zPcbZ9$-v#^c&hVOst*0<WW1{wbEX9uubZ?gDotJ`<9fiNII1+0JlfoDF(;0k;u|__
zlm8|rd6~jN!d!`g1Ka2*fFC4qQa#NF7cNsSB0=*0#EJJ!L8LfmW@PAW=+${w&WJ_!
z15b_*M)YP_0dXQ<Qykvw`oJhR&rxTRCGltRpNU6<28~5W*G*$&)c4U@kk^os7Z){z
zCaY)_go=^h8`^E&W|Nj1><nlM*ZHpue8QB#_A!*c%Y_|Xi(4^w=5;hV=_gm`Kz2Ra
z&(!z3+-h;I%LSRB*m%HmLmB+c(JKD+?r0Sa%?A)oKggA%F>7HNF*SOmNJz#Tq1a+K
zqy?VtFzq^f1b9bfiE_HH3J%I<Px=EQjI~rRvQfvLNnuIAR|b(N4}x5Xu#j6ajC@Fj
z;&rp3myH8-(VS4Q$(TZSA}$W=uYJ^*V!@HEHPR(<6t@A53)=Pj%5L3SXV@NAmJpRi
zfkmHWslL#h0@!xXN+T9!RbIF5-G#3_r1$Y&VLbB3j5{ufg<-bnJ5#Pt!~`{Jt@!wr
zBm3UiCo!r~mEbiu&gJ5`{)aSIZcFgoqJpIx#_v0P9y)|@rO%qPo=J{=J;0Epb|C*!
zrR@KBrkL3{|6`gZOMT5@gB#g<My=GMym+@9vtKwiXs2`J7CEMfl*O3^Ar>s6vlyUD
zZ=YF{V*`{Z7IXPhtYs{$jxC+T?+wJR8NGV?fcIl3@T}uD?3g55p-+W1eLLESH~gn;
zlAQR$&?$n0p{=#`M4PqC7td!a{+jOyMI5eqxgpYlnOvHm2mAhXY`A%5Kpq+<H>0~g
zfh*_2;pcSqAv72B&OJD48YzKUDG!Hp#*wQ!&8Wn?nMpCDu^FrJ>DQl4<(OsHbOl}4
zI1wT#17b|2>eN_U;Zg6Mt6I&i3~t-<dDj{ndx2KH><}~>al<hQMO4K!K+n0k>YtlL
z^;8*Arq@`8hAqodsnDzw+34lkFWoGS#o{Y3Q(cZUQx{_#n21oj>RafT#hc1yr5|La
za*uY`2L#W}(`wCDTj+)k=IfCjN;C4_){^6+S+BjwF(2{WOhcp}2cBjMNSw;*+O({2
z)4>Aei={fz%SRatu6i!s!ra(mKNX4_IkK9ryc;<1cUG2iT(;M3bPj)LE=_b%#vj})
zR1arHD3kLljsZ#N`P+amJ5t)qCQ!mFl;ekYY7tXANs*^fNuMlpl5w#9#X{0FswVr@
z<Xw>^OBr%Zi<uCLrrvJdH;(To4_70h8WCKxx+=>1ooC@)e$@t0^dZBPAgMmCCQ0*}
zirL8Y_Ufy0?TCd+;e`Vyt@<d^-9wl>4?O+(gXdRH@H*ZItQFpv^k>s#Sq;~vqS_CB
zXW&moxlUStjeRYD8QM{;NlQ0LYT2|Y9<~osB2zvD$63I);nEVB`Zij){$T4LjkeR-
zpSr$U^!M{1Kew*)`h%&^>1EB6(SbX`8v|IVk36XNeiI72$0$rgoy3Wl-H|7bD$gd0
zefr>uidK#_X=8lu??XK;uf9*F%#?lOqq@I0m|N>`SK8fR%WKDR)p!fX3{ty+KxC#@
zn>#4I>88BK!2NaY;#}Za;QYFN!bBFqAd3_yN?zY}DhuMVE^Ez!p`oAj)<51G$?$4p
zO7XL<bz@v><ml`H*_o0(JIUB3MvQN*s8Vj=w;k*A7byt4VV{d4ZdC-<xD(CANjSAd
zic+RJl4K@L7fT}PiUj7*EV7!!Ou!wD45&dK4?b2*?2a?kOLhYN+cGrArMxe#Fy$i0
zN3y?*YZpeaPHV3uF#@6c^3+jzra909Ee+N?amtWpCqL9w;4Iv$T);y^$@Q{z0*UtT
zycmB+{2e2U?=yujf^eg{^SYe}is5R$QK`xLYqO<I-wx-%g40+1JIUK^nw;r(k(8v$
zpY!iaGeL8243Ni#n2m9LzQL~(<kK>nyQjJDZg~IM&i!zs0gM18qNJSjLG<7%SLEPv
zq0&$<=MiPpHjOqknttcZZnym|Z6HWle|X5@XKCB}N-O=abma&e{8kJLyyyPs^_Hng
zT892?V%WTLc;kkFuoFJUCH8KhARR%oytb6i+eT9{tit%d_5S1TOTi~nmAOAjHU%EZ
zKj~TtC;|!gGEr~qBOHC*w%e*_J#WxrZYW});0&UJ^?2D_n$M8Iox;smQ(_n?QK^|$
zVXH|WxoRvc#C#$k-7k0ujW$CY(D0SF-_8@NGv8Spqud#P==l<Nq;MS~a1hN7P192V
zuwL*4czG|)W`TU^TX3@hB-Bo$IJIotHL8YMpQ&d|DF?i?wiv6hfZQDkDI4?U>Tuu9
zWX~Q7OVNHN;?J~ogQFrFQsAT?1dO#7;=*L{QK+%Sn`-eQN*HYfNO1_MF?BA;9<rLh
zo{v4s;HDt|PGq4Gf_+}tI_iFmKsjZ%6rzC~==xaOjkdFpTHu6Fzbpj_5dL5h0MEp4
zJ2RnAfI(F3oxV#+grZKJFbzSNEP$gzm%&e4evw=bdZC{1=gp^{6G_mQT%dJT#8dU|
z9M$H2jKWZ9D;?+i=o1{qGI~B4Rg~%lgAR^7f%{&7L~kWFE2#U%-Nx;X0?)}4A*Oh^
zDGj#?=}TsGkm`cBUm>rADj!x14wW?PZuV&(S?5GT0hT3aMr9%{(HRjOgy8%QY@G4@
zhJu}S6JhAEwQ8eym7N5*q-PL1fZ%>W#1ozs1tab#ml7;zqu=Cet1UM7w!c#|?azCG
z@fAgzouFRyni4F3-N%U3_bc{4p5?Bys|loi$%kE6PeWx&vTMMrw03CrJ(IN_)s`!F
zz)|!&4|UU9tC*%K*}YH$N|S2UrRZ8bx|F;KhpWP|4<m-QQR^Y+Y*s<APm9#(7eX59
z5tGmglv(5(ac!o(nsoXBgjFa9UimFct?`n>#8qnyj>d#cCrt#N>dtoYzTUm8w~kz<
z005p$;d_g|FQU_~N%e}e{O03v$fiy&ypsg$&!&qmw0=HVuDt6v<v2eYxwFOzW(?iW
zHR4}I-aM@i>}8~AqVheS%5wih0An-hZ9@B|?ihO64o*xdvcmwFF79=s^8Rh<w?8x6
zer&h!A}(L8vH-TgZ;`i$8|pPxcWd{~RH6mG&pl}|dz{qK?AOW3AajyXJ5ld8hN_4N
zEx!kcNAbj?C0I=)r;1)5P{XTPVj$OklbA=Um5I^3dN01b4s}Pcec-Z^NB=t_@@KlZ
zfm?)-A;0sFly+}fz^g&RiptD!O=q%|TzE?wX5*ONI%tMa{QYvDrQH#8^wx>S>WXE)
zn>98Ao7Q=rIhrR5X6+oe9tnv4X*C<~*RK_9dc$(jxBVt@a%wz|SSYb`a7dJTh8xj$
z;_oUNct|fO{GjRIom!PLTSnIopl)=T9EIe+|9n1k=Yof;eOcJ=uD#~L8cthpn{<!V
zeP;@PH0Gg^j=JJl=>)|z<=qa`w%euEKTf(uP^yeyaTr56U=yR4Ncq1t*ce|kf)pm?
zwaW+1x|nU!Xb1aeFf-pa=`!h6SlsS9PJ`?~3u!*fhX|5jawN>CNtnK4By=&#rgTW7
zqiRzJ{ZZoID35F`F86(gRTJkQty@1?HR#l(y{crt28VjX6=8>+1MMMK@h^L7N?r95
zL=8QUN_$3N=*pr(EF|4c8Lo70GAhi0$U&py=bZZfMxu&-rIal_)kUUQvc`d*;Wz%9
z*8b(h)wjC(H~BX|PP0T{EC*sd>Vd*A<;&@V973MB{GDSo#{cbBNCYwsRxLyN@6g<c
zAc#_7ZjoAp4g1r3V4FDEi@bUmJcRak=NU%sPWpl8#l?nkbNf;Nai?8z<k$X^m@xHk
zBTn^;69|&=w9C8TM={c&s?-?psDfN2`kCtLto987-kqbGo+&$YrRk6Zdx|h}Zv9dL
z<RG;Z1B{1*JacT_^$|YdSr>YlgrLj!hN|5gEm@iA-F<D<Ybubxf#31i+M&-|CEeY(
z<Civ)XE3^70inXYPcY~GlhmU*ZY3^Q-T-?oGQTUQMj9elN4#Fr!`zkIhplfQe=_9!
ze`(FX83_MN4h%EXf6fL}sc$%JaQ>IpJhd*P{y}WhX_W0Hdl|+@zmXxcWJI6`os6hd
z_?xh5&g+wXHdki-=XgWY5j6^sy}8R@v6<`Eh&fasPq5E7Q_z91z|tqdHHMNLd5tfg
zI0&mVM8ds42AGdjh>N*aEfC0O@TeMhZ)%P?*5gH@jh0h3cbao9_mF8QwODlPXTkbh
z7Z!*rZ%X4UXM|inn3yTA=2-14WyO;JNluJD;%??+8tFS<RIOq-KJ+5sis3zBB;x@5
zO*YmGu-%Mev1ZGz6%`**xssg&udJ5B?NpYvHd#0L1JL063w}AHbVnD3%B)S@6t@qb
z=|<6?a6kCa3`7&RFHY5`qcEy)f|i&nk9Jc)4u@x=jA3GXDmFQn!K{i8D?a+oOf;KK
zHk3zMutwwdXGXE1yAp(iImQpPj1+WmSpg1SNZb;Jk)|4OX{CBo5?h094fsJk?;gm8
zi3@^;uryuDI#1A(n#7YJr{S^j7*w>~dI0Enzl9zUV71cMex89y#5fg<V|nR_Wb)w}
zjH{cqr)KlFUiFMHWngKux1QEom!kz)glwT4Q1F@Aj)MgXIGM!9!Deq8tqwccU$@<R
zJu0WBnb*25C$n&jTqC6Hk|w#I%vp-gDp^Z?paR)`cDR3yQX0piL>6~!y|d@ML6aE?
z8=YK~vqKU+jhBn_S@<Fc5*BNiahKqQ$n*4c;z(o3UJtwDT4#eu$(28KLtQQOWS7tL
zkPaSHmaAFk-CThhmD*}G(ZG!XB?ub~)qK!^=}U&2WT!;wEN&0HXRT}m%%_ezwT`f*
zLY=%q`9TwpF!nWQa4(3VZ<$pfc{PxAb2C^QrZ1a<I?Z{Q36e83jYND)VYWHzIka_=
zD0(E{vXYO{OqBN0b&oI~-^BQA26M<s(sNY<sGgydU2lI25FeH5L^faH02%~%&5~vB
zjRfRM?#E-xb!7UuXyq5EfGmwm5|0JzG=`6Wr0vGPHm5wy^0_44a!+W$5B*WS{C0vK
zF6)TqPxpcFqQZWFo|F&9$-i(~R&AM7nHtE&h0P>d>(KFMPOW37!Gy%9Fj+>VycBCU
z>ZvJhnA9Vf*-LRI0F#buHbDWN030-Uy9UBAUM$s=V7rjnW_08|f#Pf{!QL1$`jGu#
zIO#XZsXO~{+gck_RvKta1`R*#0~rq6NSTUSyAE5o{bdPPdH*(K;D7RNqde*>9%GS6
z;sXxbHAO<T-HZ6BkInJ55xug5C}O29%Pnc+lsC=Ly|*`4Orlw8bM-eyX4tCB>;1}0
zg`Xd0R3AeZZC!P)?VoCOIvx%*^sX>q(*2(IPMM-M&(Po9cv)}U+``a%NA>GwpPL+(
z`km#!gHrm6w_bNN`s!2tf<ak&?X^MyGqMNNof|jXXUMUHD~99op}V_L?Hpu)v}4=)
zK!Vmu04~%K%6P83%2z-if4@||$;9UzUXU7qQdDSp^ZX6}f)VIed}?EsxB7A%Y*zQ8
zSXKX{Rzd$Eb@#r3>V!sR<H+U6E3hhp@c@991Q&cN%9rta^GIQI)Hczf>^I^P&IBJI
z8k%bx>$Vxq+fGD@V)Ia38P0`Is(Uz4C!D6P_nf?2Z(BXAvgCy)_iPWy%*^dh^FaLG
zAjV=W4PX&eUb9L9dd6HwqFME;?@c?Q_}%idr%NQehahz2a2KIzxSLzMo5*miu|flD
zi`jlR;@o&6wx!k^hDP~I%DhovQ~9AEFEBaJ-#52Du--#i+PI4qi1^d$xV=o6;|z*(
zO$s9Os4?jNhY`tP+s)KqhZzlm-jo7<Z#^LbZTB|!rL&y|d^uS$4>N~n4eu#?Vb}vF
zmK9ebq>b80*Y(+vza{Juu&EF}=kg6qUliE<Xh`HI42a}@svu%|aL{L%+b(or(<!O`
zm+*F{#5PqybSL<y@0oulwB$}1WJr>KQKAvmj=`9AYhsTVQ<_B2`JJu;;beq(Y}@ZR
z57z1g(Uz-zzvUfz1>Y95?l!BA32irwJ6AuEl^>jk=Cg=3H>3`s{Fu~^cG3Xh8pZ&A
z*N)IrOhWV9ej@HAYENOcPW1MC1QFcR+)sQ3hqzeM1d#^B6Bh_5uL<H<-dz0n>v`1O
zr&drAWE(ZwEuBTmEjnQ2Yemwd?{MN)kYc4DNkQ{tZXgtc_UPC^)K;IM%3Yx*q39TP
zYp|PN&Y3J2-k+h*4QmJW9?N;HTFMH?lhh9&YQ~6Ml&ZUUJFmxF*E%c9#x)h;rT%_s
zeAY+!h^`x+eX4hDKTAwv2$2*Eo*QeOzJP!FcbEtVGT*fnb4J<RASH*u=VMu9y*u|I
zWMOVB9(Th*1;kxt%{e#tiC~V<{9tWPemxRl!Y-i{_bnG2ig;BJkbvzUJ;;^O!JTbG
zFFL!Xk!wK-&P0*q-=t3K4$SEN1Fr&9PT+<Nz48Yd_e)t#;GyzXU5PE|MMGY@7o@Uy
z1{e_@wgU>o0(neChtE>0NgNgkDEk=Ud-rXjB8xS_F7F5Lmo@AM=th%&57?+aLha2#
zkf0c7zuz&IPE8;(J$kR^ltZwsB2q6NK+k3z5eBmzKGv1na4t%v*owCKwgnO&5rjeq
z7&;jFP(~gCS6E5T4SBU4p{FbL?2izUHrQqZWXHobLl#?Q14N;ElVaASgYcM54J&VZ
zRW2-h2GdNw-7V;uxzkInq1F-L;^wT3@oR;#rF>(Un|i~4s(Xl6I@E8^e~8yR5Mx}B
z45;*Ptbs0uR$Os3bzVqM3N`TJ#p5~>#;d+i4j0}<ofQr^nepE_3J&<+xIrY;z=Nz<
zvZX+;y-G$-(Ru4~JYDZMx^2I7JN}_o*x3G|R+yML*#6UTM^~B?Hh<j!TPHM)GJfHE
z?{h~$2?iz_ZYk&i7F1=xBk)E|O(z%9(hil7l40?gV5fvX1oEY~9IpdUeunfJ0tL#e
z)jqzh?4Li$<pDwaAr<uP6A|+2L4ks()B*zDFMQc0+k0NyU!I3fpWsPwq5QyvBz|B9
zse|Zf6Zmf6)pU1>gf7wx!%$8JC;$r?yB}eU?(UNxBI^8k^nAD_7ghIFPkeeXs|KT;
z3<C>b^seZ?6FG|c0*wS;o2xnhJds!H^A5%8Ool7@t*O7>zaE678jZ@ih&AHy{?NJ1
zi)Qv!Q7|^|W(`CNmYdE@AgIhafJhA-nhzI#aMa9!J&qSOb1w=yvDZ0H)S3OY@F*uF
zNP2e{EyU3Tb;d?yi|BIr^ggJZ6QT83JtjsnDjQF!F%*OUp~^2IZ=YR7tsqY@z5+2{
zC|a&YpJK7y)+Cbx3=eEF7y%sQ0YaMRQ{zPx%nlo@*!+q1DU<TbZrSStZD$UL!{JO)
z(%z@Q%ivJ$icdcRb`Lh)FJv%k%EKDwhsYu#f_SD~7cmu7wo2BnU;h??2#NV-oOgQP
zqS??zgZb%9ma=6T4!mfezI0(RMMN8OgQ-M_lzwdAv7=bNVPCORMX+G2fqT__kz;;5
z_lWOfmHcYaK_%2h>$rMk{^L<Pi}m?a<l%*6>k1b3XlH@9x?6sqm;E8)uUmHDbHxSz
z;#g^ojb7c_VF3HxOr=NCvW!++;?>34zM$W(`y<cg_SVJt<-oh})onWq$9Ex^MWpJ+
z5Q6mS@-d)wU<PY9fHR-KdS16J`jj9U?9rnUB8Vw>b<nL5J9L%_U7L<KTZt=J>Csnz
zN>ROJ(aT4ULsB`G%>cgqMwI};Tf`y{whV)Eo5c+?NI%%TP1}>u0&3S=8qmaXFCtCI
zalfv|HsKb*=zwA2-K}C%wkqzny%Uvm&VV-7IcxRP9;(?9mV1N;I0TiF5~Q3$kbcpu
z^~HIw8O6aC`v5+0xb=%RDSP$ihW0V=m}43*B-&99f^+F+uyZa6RY9`S`5elsMDBHw
z+xuwHS>G|H;txhf+4Eu#kCs(zJfCS@37a<P#ddDbcGw78v?a_8vmS?+OET;Hf!!Vs
zOsKHn59EHXUD2KEG6KkQnL2cq_+39n_(J8PDkHv|$%C+}lPpDfVTmS4)F=^&=q+qr
zh^&-yhk#-*!TlEbNB=!c%5Cd;azew1PmrVqqMgG(t3Pvw|HO)9|7^QZ_?5)tN-+QH
z302Lf=^<7dI+AnwdO>*Tfmv>;yUkz(zvJk}oV1bXZpH2Vr^J)@Yy^oXlj0sT4HSt3
z$M|rupfmnOA0$X{>2t9l^xPMbA18zLJBkpn3B$&%Rpj!!ZSoHv&7G6dy{Yu%bn;tB
zRyvx-2#%^Bk{=o>Qr;_nkY1)e(jLQ<YZj6zerm_rKCaOFO7K1>cx^fg{JOx7!>Spp
z&%qUNx3L~(2MjS3n}~@<J5CY?TViEy(_0QxJ9LZ~hS=Khg3+li+aGajz>K7c=Y{W7
z3Tbi4*|D(uGF1D3q9l1g{iD#aGXKws-~T(qD<{+cYj|a5{crZ?|BU`DY5ubP)rRzU
zU@zfp%2c2`C&o62q5mcQJpC+41S$ABgu9&;nc}{BNTpTtclXY0{y_vS-mmaF?cNL`
zHq#e-PPOcT@$&$~q+QB-q!(Q)mi<;EM6E2ly0k+BfH-PO05+2LPmsL6BWm$lnxv;2
zCFdr-qH{{rgwipjqkZByl43I@5bAUpy67M{U{#8vCqW1_IFTSky5^sQ4OGgNK&^>Q
zpD=wzlwgMC%>I6x>6l=8A%ujrS_+_e{w%1!_&^9ErpRb1{wClOmxXbR88ITS=AVfX
z#VN71mYJ6*o|B3OhEzc>8avK~>F~y74q8Yb?vhe~re!7SagT$Q^TlE*Gl84fJLI*(
z_OZ3Jow6!qNay~{n1DZB*8(@r_-Uq5wvMYDoPbuTh^cJuxNlb)B`;*0Asw49dx0iD
zsE4Wyy4u_nuoTWpeQ+9RUby<IZ;uJx7#_8Xy)>krjyh?~Hn!?GDLcW|u0DtgJv9`+
zfqS^*GVF(AY2k>bY78?Xm1}pH3#B|{kJ%`?evFEmJczPuxS<&ImEecf(@z?am7iu&
zzFvCiQ2QxPNz204ea<`$g_<0MMONug;J}=fPIOcm0-l^G-R!I{MV9)RL5CLL-;gG~
zXLV;+E+&$orWJMrtFV-`SPA)yj*M~H&pCPBCwSeGkxv3$ZKF-h%@B=!3|HG)6`iXM
zyTUxPAy8=5f-zR}X99@8fUim3O$HFb`rlFFIN39D7}vYKa5K<6O+C9hdPeXV$*gf$
z14jV+YtAkH0OLdtF6!yVEs0vv)rq0dyoyqc0s`Mn9G{GtXdOB9>n0qZ_oFr(9oaAS
zRu0dXF}tp}KWktDxZOFIZ~~wQhfel3{)K!!9G$6|Gu@Lh<R=*RB~zb<4!nFH=1;#p
zt$`-14CHKUXZLXBHwM-*Q)oW+bz1C}XFo4?zWu-ZxGQ)*9UShCZOC8B$QfQC4mkL9
z7B<<kwr4lls&tm7t^ZyQobfnZJ3qjawVb5kPByTUTZ2OuJm73C{Vvd4Q@Hrm`P%z|
zjDLVGGOF@fUY^e!<Y3!YKvwW-@p1PH)G+l4NIwFmT+eRHN4yr!23)n&E;()M4(feI
zI0S71Fw}4XohSaVGjj5#u}^RLmCSBc<;?rTZ&8l6Zm&mI`BC5ee4mXRAFLS!^2gRm
zHzqF|Hy1bemKLC(E<sa>buE#>>*K$+gk3@ayZPsNayNXICC~G4zFA%AT&L&19eKI3
zvOIW+<u8F;1A@5T44l<(jK~1g{6on)-g3q$i0v?iS*Ey#eYfr4B-Wm0Ml2-MRs{YI
zBn5yQ;Kvbh@qVAG>yDKBO?Z|Sk=SNy28lI`UY`9q_&YVNy%_sS0rK{vn7T-=p8D61
zVbQye`d4mn0d*zw;Np2cHfJ_gQk{?x(yUpsIO&|QEP!SI2sMM)-1+IebkiG5gYd}y
z8n$0NZ{DFe_haV6RM8Gc3qdv>ZX<Yz>6GH{lmP+V4+~!bFLm1K=^XfhHLoY601Dk`
zx#AHYcHeJQb4zu2iX6yHlekE~*qluyGjYS@Qc42FOp3=PkIo>6`q@t=f|=qFCqhwb
z)@c#JZz(e-KFIOx*8bb^fxUb&aQ@xO%ena65t;6M77_KQ;FmAA9-q&1fo_4!&PR!c
z8BoY#qD6nBR3CnxG<;c%_usnc4F_i(u3rQD_hY)-S^Qf%8(IMz@_Vp!8WG`JxhWqF
z1rjMIYt?YVcV_l+kVwfi+@`|Z`K?)Xv4_B8K*#aT%M5RToVG1+_l{u2Li=r)>_hbs
zN930$AH6<;qjq_Do5Lkma{3Zx=9+CiWp!!E3{vD*1x4YKbCKM*N!8uKxcDORT{@C9
z684mPfC&<TGi{_o=T>r*!Et_eMYPfKCD|Ro1oQVi08=@CMogSp`p!j#IlRrtt$4FX
za9wD)+Kb0fS6LomVztQhAf7vpr*3l9#j_=|?3b}s=5M@gW2*||RzmoKwAx5nCR;S7
z*5gQ6{!g!HUTu=3pQJWOo@jW2-2gnmWN25UC?%aUpeReYYFW`5m<{DT%jM*{su8gE
z+eHo9*fxz+cyz#B@TTY!5i!i)#TTd0Ofa;GjJ2chYAJ}U5%W4N2KbQ#clY6DaOOFs
z=ohFDr@$NIr+2`EOuPKRpWh`T3GxZAb0r83I~4D_2s?7eCkq7#UzaaiPiZyf5{dSu
zu-+Z8!_&Ei-z8ZJdV}@*yMEbIHq7w}xkN(Z2b9m-QWbI27!B?l{jd8sSI{nm>2p5Y
zlRKZ==GI@`<BhOANFhde!;~gf0LMv^w7IwBq%=#khS(Z=QrL+DedA-zp-5i*JAe*3
z&BF-OFOy>CpY_HjhmrP^$}E}g${ihnte6K^e?J2Yxm_@t$L)(@n_&~I>M4T`G*@J^
z@w;*WibfP|X6La*GNlw5ZD!Z8O)|%n$^YL%I)zrd-f3)}?EkxP8Cxf_OsUka|DP6`
zuI^P(9WpLML9AFP>j{MXbzPOBCN(sF?xyvux-YM}<7Q(haBs=L!ukIWZ-7HocMTP~
z4Cw>Gb@zmme*%$aXwNM4;((g!0G*2g8A|NxF7c~d3PeJxT%vZRGD!LDzi8bR#->jx
z82d~+vsPhUf4fC!u-ZioKs>(Z5F<=XX`d!OcU1tx1z|e>i7<#ckIgqrUsbPjSWDXt
z0SX*k>V4+tz*ETui#p9C+E==+8CS#KM_h?lhoAvO&W9ZWCLoHAB;H}mAW%=NnkymE
zZIhA~146u^E#eXYL{<xKnJe<H0Pekk{XNMu*h`e#*gRvTpt!PpqxO&^?85$P&!~Wu
z6Q_;EhO=%U44$S?yIN&hbO*o;uujEud>Z1{v}o~p>Wh<imkrIO?Rb!X#F7;_f*CN|
zQF)~`!K8d56+jO!q7L07DfLM%_lwEjlZ^on$OUO!GYdy5Xsfc(jRtoLu+$>CVzg8c
zoo=p)A3XtLBXr()wksuO)GB}|-Wn+jtwut-W0uJ)?LCbn49)kdf2zo}66B&D)l&7e
zXu9~kGo!3EZ53xIB^95si;DW&i28#Fj$x~N@@OHhX`aNc#<FuFs=$NVv3f>PUzYt!
zPYJfCd1b63frM%3PKydn#Jq|@3BvlqL84C9acPjVkIg~%h9{U?O-;$lAY@fu(e3|2
zaL<$c_7CC5&iF5{9wQsmfBc5~mq^YKMEvV_zhsKrpXw&yQ?;OZoD~_OtD@aUghV@l
zN?@=0Yt&;WW%{Gt3ujY~OIVBF{U>8BZA}|28OR@jg0iRkQI38aMaC~k{U>ZEjv)B1
zy3i!+O?BT<Sq|+?w+8{kwxP^(h&~+Ko?daD5jD;$xP7B~&2|3no)tMF_os38E{7?L
z%%p(iu0$G9b5jA?`dS<^a~@o3R@nU&RN(Z7(6ms!SlW*7mmLVbX;6Vu@tq>pZ)X3r
z<X8(;zhm3})`(yqlRe`Xf~Q;GYG8S^#6%u{3ga6w!G_JkuId=u@>A9TWqk@K+$QOW
zy1Fn?q1vkE7I-2DfPGU}>)W3QY1~0?*v-Wliu|hd#^%<X=0QH158j#TX3Za9(53^8
z^mU4(n-?%?8)1c<d}pXDhUhxrlAYZ#`Lz(jQQbH@yKo0bYuAApi}Hagh$X5wwflH4
z<t*RO+g|qG2Z*5hnKpN+)AT(J6V1ve4D%a=k96al<gnD2$6+?*k!b*PyCGi9PWt98
zqoug`CA9hr<5ZVPHnw3clqbh3{6jCSsxJryE{IuorxpijVgbpIoD|&#J0>I*3<Vki
z*?wMgflL9&wI>#5m(SDd<(Uz01oSiRvFjSzKqSLbJdwHr`#w|=l>~;j`WX$`-#z@e
z-$tK?-81`-VuA@E$7_|6{v~}PMSh`|P=<ntftq%H!9jyg4BO?u@G`$u9Y#5UQ!|P<
zQ`Y1=*cM_@G}K%(>j%4G<t@-G^$_$?Lw+KU5XRvJA%zH6v^V!NvB-ovj&Osb45ZCo
zw>x|z2NyNMG4WPWdXFxa4>QmiTn>7ugTw#SBF`1FM^CCTmgNj4Uji{<ZX&;=tbkKF
zW|0AjMm5@KfnN%N0>diuGY1DRp+6FbmF%863W<sLI8#i@NZGSeu~9c|liWWhfmW1k
zPM%kPxHdMYW^o}+bLyEjlZ2NLZI+N}q_-5y>59vef!x4BV)rDe&${~*nABKuY)*Z8
zZ^#SQ*b;AHHP*QCm+bUHmL!^Q4NDA$jInvIp9_4h39I_$Z*jn?SqSqE@Rs7rkVTxb
z?Keu8?g_qa0wqJhkdBpbFm=ICrYa4}sG+oENNF1z7C-ka67s4IrJl-Fu)>@jA4{my
zm`3LzunHLB?rwGn@FAkG^Q0)na0dCa7qpmDe%YFGnbf2p>A8VMG&g(726o%rHxU`>
z>?aj8r4d%@=H`Poe4(N)-*|o|G{TjR5V-0wy|Mf;6m0PJ@v-3U&7<z7r9*l87A!1M
zYqv1MvZvbQq_)%_YrdAR^p0i!YlEr$r^QdHGmoOL--+6QM{PH}3<a-Jg&4Z;Y`DPj
zSVZv>D{BO-5<PLMLh*?#IH_=WkCUtq;I>0-<bM?#)_)WmE;g3`7+$<ell`w^6Y}R3
z4Ht!K`;zujEhZ@)z;}rFnyiX0i3^Da+TSW$`1Mi^yyitqtANW>g(gM_*71jzr-7wm
zc)7oSc?xT~Ca1(yZ3U=#2+0no@~lCODn3XP6%0f|bAjUG>jS5F^+){DP>{YCzER4~
zn(_0;6LZI~<H`&{o6p03GQMw)Pm^FdRVrItC=YSQo_=<#1^SB<6L5S_r~C8y`dVlL
zuIWAy({JjtGl`T8jzqe11VxJ>oQ%8ojtrIyA7M&<NXh#Z=7gGR=-7ZR6zI`7XSN1N
z2B?(mi<y1BI7zBNsjA#HS53Jrx55^l3!Ixxb9K(qVHbJSO1$6$!d$trLe!yyg@!P{
zRU}Hx<j&F=>FfCkuEXxEKt}#au8$Gl)G*BP!AQ|-&@fJF(xKkl0#Nrb&hk+zA`OX>
ztwLn3KCe1i4s5D)AkDUk(Pw1?DGD7j8)n_<z`_g}#?i$}QCX9j%vohVRCGognH0Dm
zWwN<dmg8Iwsi!T)dQjd#(N?h#npt<jn<xzSjh#j;H3Mxc5`$=T@^h;s*eiJd7j5qt
zB-<Bd50-7)Rj+KnvTfV8ZQHi}%El|(wr#s=s=KGByQhEA|DF$%`+PWYZsd)OjJzk$
zUTf{W+_DSmWWkH3S8nT?_(TY#Y2;8s<uQKgU|=9MVi59Yj!8#nst8?G^Q-?fc2?uv
zAU{(wk|$YY`;4GAC$A?*dVHbF=ZH5aNK8FiCv*`^mPfoG2yZ*2FHU3M@1@+o29AuG
zmVo^^sWm0)2V!r*50>b^n~!5u(F{-ymA0@bI=2D%5zet;r%{7u5)x!Qu&?kKzFF4A
zv5Wm}1xL@j!)>9WDMxR;U7vtQKl5!m+uc<hsdAr$%}Tg<GhvK|BvIp?w<=RpQDGkq
zT}bL~E9oOTkTs#RnIkL!0Dz?Uma(3kT}WF@%1PF(&SpK=T5rjbEPa<R{wQd=(-{sR
z2J(hKfTjL1PF&Pt6rR;t{BWiBD9)`@4-2d>8xXSu?)2?h0fy$v8hopzKj7C-3M{&1
zPmXrUvQR4how}VP6a-DjV1{C7MzYBwkv+89r`sJW;|Y@O?F!5mv)S@~t@&peH1I2X
zeVpuCw_W_AD`h-iyIj5s$ae}qkzQuFH_*3JpBs40oGyo*U@SjHKm?RDga1%Wri}8G
z^n-<}fZ=ViD2x)l<+Gi*0kGJeJA!H=h)=6i7KfE#zEy3rI)cu5fE2*qCidLGFL_&H
zc4K_u&V|F8?MfJSf|+m`yf^M^<b+W@@E~eTuO%DhNV$ZZNUAt|vXPaaRrRySnphK3
zqO?i;^?9CdzJZrw6IA0LwgQ&_QVI1x08R|_46OeNabo+wmCpZX#HmYDI}*1I(K}Z+
zfA~OH^^q_c#6;gUI*|{^<T8|G2tLLYLNX<N5U>Ae`#8x)RhO>HHgl$>*R7(oc<xl8
z+^M2ymoiCWk1|Va0Q6%y13JS92HoOr89~EBTD#V$uw!PPy$36N%&U|XG1><_JTE~z
zP@s)<xhA0u&QTnB_0?K>0lRGhDVkwDAM$n~)V@_Ebw;gSaxqh`A47S+?oAbjs~PB_
z`giAALhVbWg?JIi+*;yE+NMa~9~?<4@hVQVzi~V`&{&rGLl^)x%!oU{$sO3Gi$xMV
zU{=LVQ+O$j(q7fSE{|J?Ps?~}l9k3UER!f33K>I;D-B>0gXkXdN+1oV@oGC3R||{R
zI*3^$tK*29a<C@tt&4^>Js%P>5_~mF7dEouFcO!bhfWeSczYHXLa>{GYLXqdQ=-hD
z{$502V;hpqUoBk{t7qDZiksZiOc;50<7RoJ-}{gDBe>*5OtG#nE+l(J%_=;fnz;IJ
zSv8-RKQpdJFNC(gCnt#9w!ADcr;U(?R8m+iMik{AC2H;XR8P2|CcItj7lz(?LBj)A
z>=ktUlD#H)F`Zon@F;J^<}-TMb?p+dlTRC<M`{b%t$|eRuZHFc-h>gFXvpMGV(oAa
zJFD5G=M|Ff#D2G^J=(-5lz&D&oR`$vV8AwHPSd8x+Spv2YQ~6O^lkv8xWy#N3CE6t
zMJ&{xGS!R<(b)pP5+Ac7aJ-!aJ1+LN@tdV@a(KR2-8_Bt^=*?awZ+H7%e|8M!}ED=
z&h!0P8v4DF-R=4DQz9P923nWj$^H5IkUIZtYsRyY?bGV}owTLb?I0=?r_s>X(;UxD
z{ufJqoUg-nzQOeLfinB+;pL4XJExTQ<L+wvxow%Z#J0PwzXz9JZ2UM@o3HGE-wOtU
z0`?6-s?0<eO292dIEaPV4hN**(JIVK!NSEA{k)9@y!gdjmG9(>U?&Fi)@V=;rC}DH
z5u{3ILe*pZ2hS`z-up7Aj~Y$o*Sow76WByYKwn^7;#C<P=_WJ~y*VzFLomKO5lR6R
zX*G#0-nbD0*S{gHAb??(Alr_Ys=wZ2rf%_RBpsPGQw!MnNtF53a1xD-wvYN4#5;KL
zFVa?gZ%)yN2hjGWMf25wK${m}6j1$v1{hlGmGVwl;9Gnt+B{dD@)nO{a)<ThI}U&(
z=y$TOkV1w<kdacUs2W>7FJ~zFK`Jq8wVKBG*V%_?bF}Sp#36?8YN4taEPlI?Q4%aT
z6JhD&UF5l|Ed?>YAXi%!ARXy@(F7gi*;P_~5oUjV#rl%TZvB<U<N5r!U#Da1D-*0g
zhUzhY*}L5}k9h;q!teBUaq}Obb{}Eq_5U_Vr5aB>b~VMMl%eD0>h^VxwUsPQoz^Xy
zKY!QFom=0|I49-VyL#XjY4>$|JJH(g?9n{j-!Fx~zUIw2{ruS6zViJz-E@uZDKQrV
zZK!r+@_xMn_5c<wbnNGV#AmpX0>Yi107TCZrANz8Mj!`);}@7guz<*K3qC^2&@^kF
zvC3Z+Xb3ie{;{{GVg9#;w$LNI3{SJG{Qo);HN(qy4?053_|KC_6WI$ie}Nr@sf_Kb
zhP1dOe`A7jTD*<Uw>Z!aZBT3xp4Y?0BJUp1OAyg6Bmg1}uCjg1B0)sG<Uh4=TZq7n
za~JM(cgD5boUia+5P4a!x~qp4tJ?^R^?EEdA!bv-e=eT5Z1f-y93GQ6U;qVtqP>*2
zw&6B@y}~es&D(RV;6q?!T?AqNbS>;Tf^Q^-nBkJMm4mgiXDt84is<b5;1h+LQjUO)
z`F_QTKBu3;Kipo#T)Ndc6@2~S#xwK_a%yD77(}nkv*<uw{I=8Y4Rwp(v!3oOVNw)T
z7HPUI-#y|Wg#%|mO5z|OfQxV^Q2u5Dat&12UO*>Spsp1ZTTU+$zQwM@hlMGho8mSC
zhy=rZ7ZlX3E!i^Mvjtb40G9zXTs;o^VL-UcAzsZdq^5YHYa2iq@c`6`pFb<QrY63Q
z7wQ5RVz<43pa_1#D~?904uzil$;Rq1Y!&KBO!|)gw*_Y$-IO9QV7$N#nVy3fl-Vcw
zFboU`(V_|jOw9VBUW-$Lr#e<%E7_-Bu9b|w(ICLkxCFSN+;6%uP+**e8Z?aOtcbEk
zN*9<%$C2W>GY*7Rc}pg!s2ux;F-w1Yru`LX^D6R`EJNYCWXFvmw2Q~*X{$QnSJ%f$
zRAK(Sq-iVuqo0Q=47h?vyvIpo<ADmyCc_>fUyZ+!fhw9Rf@Z9X4hmz7zYHIQ{j^vM
zDk1(19PqLX6{&<>@oEt%RN)GBK%h+(<!55dlV$&||H_HK9_6W83J_aqI>eoDW8}QE
zt^?eI^v7s=U-NYJJo3WX4SqAi^bqn?W7bL)Mjy;swhUd60F=HRiLrR=FM{5SVa>uI
zLPVRn2SQjah`30j4y=FfT8;ZaOfnW=FnQw<PB&NRsLNCx6_^(<@nr^u2m?mnO@p+Y
z)bd%`q>UdC8p9tH8gi3%ndSPJx=L9kDV};*tJAZX)6!-g`tpc}bBzVOOEB^&_hBdl
zrrTCf2ok*hpPxEHX8JG<y%ODQ6#^fF^N6ZvQ0~xc_F$!`OKM^(c5e6SbyoN3x9{6g
z>P^ahHal>^zYiivLJxV^{9rx82Lu78P!qUGoD2?TN3)}uiTrfIe_cSzAZ3;`OPQh0
zR~KjqGJ%}@mjz3+rJ1UHbyprlDkiF`;zLGDbEqe}by0&3FnS@$o)d&o!VzCg-9+rG
zHj92FAC!aAN&WEZ!aVKgL{yN+3-m-)1Wj;;jkhoIB4Yf4ND9`5`t`h<^(+x|lo7bK
znLyJ4<O26|RyiwUq4$EYdkiC?ksi8Z1}!gA9e#e#g1-s~MSMC}xvm9OexHr*Ov5$W
zVftrmeDd+oVm*0gLDs1;BE1QNwo{-H)py_0ax-#5Gt*Y`Sqb>o&=xDqV}odksE*ta
zgZ1G;r{1eV%@Wt|SO)4PAYdb@ii!rlR-uE^lS`(Tn;#)G^-AAu!v1a1pA3FL=R2tV
zXy99@fvS=8=;PY>h|?o)taWXmA+4s!1gz<k-2^F;Q96%c<!JL-joyR8nJ$3>f1plK
zbtxk&e1wF+tE0<uCL(FAEMJWhbuTsAZ+fCJomI95KWq)pfGi5{x_^bWYGJ!-ySZi5
zth;L8+<}o(y1|Tt4w}rqlac=|Y1gHOKoQ%#tfc>d|B_n#M^NgYM4}7~4FBtf>p%8c
zw(BDQYM*5Wyz8+YiwC^`=y5_gGYJclA#c=NRFt4(3H*7B8y?kvXxUb8_FH-fm^2e-
z>iKel4Srt_cW37lwB`>l9c_&*gucIk5Cal554Z#YjTk|gNKmHUEk$cr)ZC``!xd!`
zhVEBa#JIJNceiHp?6q24ZjFGzu4i||TxD#e?jPHi-Xi%Ivjl{IMuwZA!Y0>H&A*^8
z`~o1ay~%I)q0i{O)?%RDBq56}$~ijQast>kAwmd6LI_wztu-*)q?y(!LZ^(IHHi*w
zG7ZM<+s3(6sa9zN1cH~tfRHhFTppdDYQ2)I#*g}n+F#YQpU)pILCUMi$(tEn)RgxJ
z`R(WVCbuI>#?|AOy>QL((!^Ng@mg+o6PjG>T)`0p`C*_Z7ec4{1sS`nGRf(=lnmLF
z+;qyXPr1g4Sf*OR+)A%UbqivW4qG>KSQOPdB=Di^=a6+aU^xVySRfL!_jAJZ1J$!X
zu0MHYqhq(6Wx~MrPh?3wlYqd5+-sjWa99p=$Q;`ypOU|lczy>(0DeJb{so>Wh^IpZ
z6_jB48!nLaaz=++DnuSs?CoGJ!L{!H&Gzfn#2X2nJ8k@3h0^P!KNJ;7<VWlg!<F~r
z3j{l+-5&k@;D=`0pxy4F_C(R2=tXj=`@x`Q-c31+6d^O~W7iPXVm@mGjRWBsAv0f2
zo2^AS#C;AuxW7h82vU57a3C#(Z6_R<S;NW@xYH!0_8Oc}QX_`6ld%~&Am658drxvD
zyHb;Yh8Uy(6z=S0EZ=v(YN03-<$QQki`@+iP8ID5$@X}^$t7h<w_CvxTbUS9-r(!q
z7%_7H7I!qy^hBt4)o5b{1x$?V&OX3@&Flt+2{BKsha@)<9}jef3_~pEhn}8^tH7)I
zM=}Zwy24{lS{*%ku}20j@o8i)t^7Dsb0aR(G1A5031<;UGL_?)CAC}buk<R4|La@^
z4<<IzJw}ZhC{9FE{>MiFF_`Gz7E(?xky%8?NVxeMEV$A#eop#lBwR{{xhMf}GX%sr
zQI`R&dg3Xf-&P{43=JbwmODA*zc&bf7jj4As3WXYW!EF-<Ct$t7yPu8$Dtt8PJ7-a
zJgMXFfp&*kBeb7V;?N%P-~@wS4RT!YIbKmWfWadfq1`>^5f7wqf$I<G`0l=ufl%MX
zug6saW;pNROG9YTL!s=1wZPxzg^O-yO*)QhPm)d_TbHoEvBvbhlOGsb`lHO<^F_Gs
zsv0C*Kec8l&oHmCATBbk(MU!d{c9%Qw79>M61QD=7@if&>cXc;A~5>%swr@vW_DM#
zzQhpkH54j7%)mv*FGw5&bz8(OH5G3K?eXQ{F4x>l=kpDU5_h;z{@qDM{B82hCzm?1
z=SjZX!yha_@*8Y#v<EXIY|>}W<1`4xj>x)%5B*Fra;6qQ22cS~uJRSzzBeF_VZbk@
z5K!cD;N95%79U>FkKrGZShgQW^#3MiXJGnwK<odUqaJ4bYqMCtA3TebjBnCT_h<<y
z5vTzdC^v%wB>Iy1id}(n71NWR*M~f^##jPHSlD_F4UmX`eExlFNOX?=4Waks5|eqD
zZN6cgDoBh300~5Rem#sKk3JG30YDyA`gnHoH8-x;E50fs5;FA0OBMI460QmRRD?I&
zr#+3p%0{a*_shXucpRP_@EM;k3>YT->r3aU@(!^&?nebC>|UP6(c|NZwkbs)c$@}x
zuU7ngh$_-r@&JMehmbNW^rI;wb$4lrij>VH(H*E<{b*J=L_sU(w)(6_|8d?@Q%-!`
z1vl9I;H!0i)5kcYn01DR3k6`}w2_b~n@&!SUxlKZ7sVH;zS#J6?$uV!?!$y}*w(fs
zga{;86Cjs#K_(A`AdF2Cxp%G2r8`mo9$KXaW90D}oQ5#apKp`6$~nHTo|5`KXOR7C
zoEhN8UcTE`tWkB*KE0+gQP=ulwZKf0p(;KPqVQ-T(EtTVq89(%ka7G=gUp<7$eVbf
zBG4o8$)%-nuWewp-}-#;NH3ABtiVr_lkwM_K>nQ_&yXNNN7s9Y_ik*?6&S*`ynGPh
zaw{cK8UV@Tre}|KoVzMLA7%^$ktgr9?xRLz*{9@Y`3<`dS(F)76@*+6d>v|vQl>hN
z)u2e7GT&F(@d!Rz9Ixm^?W}bKOsM@mN+$aaD<RTPU!brs+`UpPnJBT}JiMY?OK^k8
zq~n#+ETrIaVWs8f>)N4AK`P!;Du9@(tb~<=tTioYu&s@7rQ2yq_+=z_$l;#y$|p@2
zerHqud$l8iWBcTQ^r_j|3T!conn4H<mf0jvO{WlI<KSzW!hA7}?c?|!5J-N?d8hd&
zte^a4R7H0exq{RWOO9g5z-^+k#0$pba=75D3GfZZdAZ<90y1FAeSlNAa^GHKnDLkM
zVSg-GJmIMP<2I%qCcQNV?yYcn4_3XHG<jKWrS%)GI4tg`DEC}qnw_AK$T8ERMwn`e
zih|E1X}d+Va4{_l;Vfe~pCUV!7@%FcSCFuj?nmvV^WH@%#d{o#++WH}GjYK@Ud-K|
zW`_<rh1(*^@XAn2Hi?}+*C3^SffT5*y>NvI5Xh!srEk@nGP?@p_RD9pCjp2ZlriG~
z4~n~{GZRdqyxq~(c97P*(|MjLRpt*G9P0EQ?q(~B1>2)U%xH{ypg?W(DEa(mX3T@O
zB$cO&8^BaIgs%s3JvS9vv!BPL+dSpgo9`<<;5T_knsHRCt1FqkP=f2#n7Cqfk0tfm
zvh3E@9n-aKpKGS*wTabZ;;jH5w{66TA^dBuwkx9M(xF1|7vyXrR?!GNo%mE1-_Z_s
z)bnDUDunYkze#li5ApUbTu^b7Z}Mxa>#ukbKw8*5>ShjfR$6N1Ckc@UqXsRa3RTuh
z3rrq>CFH{{Jp^F+!bnXeU6<1@$4*R!?9f&?Xj~H^SO?rC`{~8!>$V3&zr-!&w{J~+
z>8+*{*@vR+pz0_^Sqf$%T+Ho)<z1O^1aSVHMn8#dh*Dtz&t#(M=upoKm)q3F)X}#K
z<F)<Q_@xf*NABLIRolJkR^4^e9!Pa3VSv#_&|QzL1V3PjfR8m&|6acYyT*%6XohN!
z-o=!B!o?>B9V`4j*-Rt68G17JoC}*Rd|!VP8WHdAgettNgg52ofCX(8wn)byI^X@#
zJ_U6*7akB`1?Fc`?XZ!V)+@u~#{7QDMRi&1EGMfsCM6*$d=SiqS_X((l}3vdYE4}0
z64aU89ZnR9#8ZyU0IV2<y!DTW7`3+B`-cl(=@iF*a$sfsFI)5fj4!hP2fp|(u;jnK
z{?B{&q^5Kv&M0En^q-MJ@6oB3kywDb+75x{Ua-2k-sTzpB{-8?#)c@7+OKDyc;(q>
z>k}*~XZ<*nf|3f|?ue)fFU0goY3AesX$zt>dMqC&Hjg<*+i0a!%@hr6`o<#p6<CQ7
z*}fei<F;SnBNuQ>07dKpoG&0O5=9nYY|=&tDucwu5t>P9A-t;ov=_OL*cx5#-}0#)
zX7x<@(NN9(bee;(*fQNQB#i``NG>l+5=fXabL8TrqUKpftg9I$r~UF7WDIxDW~hI3
zF|8X@l0xqLsmf*=MHCjz)#uhq=OyN3wlqqM1iPB1#H)Ynq$-7Pm}@H<XXlpPT#%aq
z12oATY$;lgYrk~4ny%3(XO0vWsgT{}EtqZLW?|{9(1_nu_JHf4;@RWxH2od+20h-n
zo&u1uTIy<IKq7ztC8c1R?yaQfPx7^2(;B&6yI>9Gul(>imr4UmyH)eZEJV3h8y~`I
zyG$LeG9x025PhY;czc6khfFP_64Bx>8iim`%Z%B+Se+r2+CihP&9k1^yXsmiQjh7=
zrB;vWd*wnoh&|NNci^7OMQQAV(_LK9V670l2q_COG51;hn<dSV{{&)JVdfb(pHD~<
z84UUJGCy=l%xZ!>NIiH%$npRYWFDNKZP@tnwwB9pskITa(=3`n>gNjeUNwV;pwR^Q
z4A$~0NC?7iX%B*4*zkd$Wx^hCBeX^$hsST9IBJqq#IBz=7STd%h8k45{?XaY-uk{z
zeRq0$+TFc<y?VGh-_ALH#P{`h8QwhY7%s)X&gDUwfIHkjeJ<tA>i*`t|G}sFKaY<e
z{J{_RrdY-%d!-u)Z~jtRR=ZuTKb$FZc1f0Qb9ehnl`fyM^?1F#dy`T1NN;;NhWh@l
zH_I9{4NV5+q-Meu0@|8R%cNJ79|AzU!o-a}9iv;HB}v=$ZL3A-X||+(0i%X`@A&cN
zt-;ncB=oE=Q)iHr7i30x4mN)D-zGK0(J{<t(<5Yg2XZ#a&v3>O2HL!~1Ucejj$x6-
z4yl~hNH4})#Mf|qBrP!w5G>NDL4YqbX(;@dp=!sIv`~?XW~6?n+}c4&MEBffA;Zzl
zmVEmG^gdF1_M~g=G(BG4?x*qh^VwqIr;#sVUlVGK-9Mrw3u_s05<LI@xv869FCCt~
zzT#)f-s<*Al)1k4^_Ic^{+!$nnAtq_%-iPW^1WqA`Mm!M(8cTF@$UYvuiCn|^=<#Y
zEtTEo{>mtkRr;9pCi8edh0J{dcXqdXyNAZg$(VSR5^u&r0ShJ6&%&?KZ>0CYAcoP5
zMjN4t(<Er&v+>^mY6EqEKKPdfPl9K@EB_7PHgE^{1H2LLe_g=y=K^vExq~=B93hSq
zC5Yw2gT5Q~-ovyMy^1?P1#WBmpvvfyH<J`<#$~R}=ciBRr+Q8S4=|FkMj(073`vE@
zS~o>oa|~4AYAD0lQWbUGj0eGT4UaS;gf7UyazZ3&GypBKPovV{XTQnX8A-NUz8J54
zv~pO`!mgv3yp!r!S`kV~R*ppiG)1#;kFwX77iF~uud7Xwbe2}TJNfZw7@;XaI{`PZ
zPZof9WD2&<pYjAV8PBm6OeJo?++tqf^wI313^o&m+Mq`=r!=sv>6aig(ok$>>*@;_
z&x!FPBUK$pFf|7uc6v~5X`$-I@O%P`oxqC$YvVDMnu-ERhORlypsL#T3m$in-B3ki
z{&+jP!C=#qU)=ue1Hv6t{YACzTQ9Jbt6e%^6pVxC+w*hZn7Y$^2>sB(0#86wBf3U8
zek`UneiQ>hO~qj6&$SU0n?8W@zTm$u1$yttly)CCdj}a+N_`Xd)n#>+*cjq#abV`)
zr}QSIl+U+7<}Fl_BRF*$Mt_7w&dN>@&i2jF<{0=c`uY-2`9*_;u~#JvB=r0zguxU%
zVV2R!C+c{i8BVlF#7)sa_l=_p)n2&5834iL)?a8ah!1!$X|ZAq{2AoULIPp_P%@`7
zjcNyhyz3#=k8ChOj_6b^g+7KFpL;E}6$G?dXdcCUO2#TKpS9hQAEc^^)ka}26#hPL
zvqyM5p(*ynMvG(YrzN4}R~?-tQOWl-sUvl=?F$Gosf(m5x$7dkw&^celIl*xq{a#n
z=cQx^9V__@t`qk}L|!C4+pHeDwTj%ikXz^TeuQYKZ@m2+z*{41vYrhzx7VhPj>2ke
zv+i4S2FvfZO4_?H=`abs_7od(cE&$FTU+e(9oSDDj{WJIG|;i;n>^6*^xZ|mQvhKl
zzuZ*zgT6V_P!2WoTxf1p^6T4SAccg;_nYZI(2HRBOZ`Jk`2Y8y_W$ou<^OMillkK!
zR&(fDWTvD0c=xWQP5r*~W%8KqJpDE*Az&?zGEGgNEPot$6wd@LWim(Gy&{8R+gH$h
zam+qN&5Aszt&Sknk_)GSEQ|zL*`7^)g^eDNbECxFOf=K3XR$4X;!0ClO~NIMa{^I8
z3SHuN$yIluej?;kmNv&FOTtS5y!vZWC+-((7a%lAgwMd$gaMQMMzV17g=vJ)Eb``{
z=`^Z-T{gwlC=jDySDYY>uyWT(nU4T4@GvF*xPPeW6dKNoW$j@F{@GU-;;+;KHda~M
zO9>z*foo^7i0(w$v=*=Y%?T9a=~kgH{pPFjBv$BKR(Kzqi}4EyBG%A|Ywi_eM+p0f
zbE;2%PaEKe1aO-9q}OZzxQG2>pR9^`{!GE2Vnvb8hiY7s+FD|b<L0O*%Le7s_v!R*
zSf**28poq2%$hly9q8E6t)Fj4V&mDk4o*?wplr$nx5RNg;tz;v;%BHij%lQRyE4y8
zvLJf^RS8rFyMbk<l`0bMUd$(7Jvz(>&U>GEr5|78c~SVT9}Y@oNG`+EC1Me$j;tJK
zMBf=W?@N8wajuUa!0;hI&VVaeeEARq`K^-Ki3`z^E0-EUv2@d)BTnr47`w^N?9fXI
zgGUenyAu6fNLje#M*^gaOaViiWy14-#-P^S=ImfX@peGyTcto+YSxx>n?4}*73x~h
z$jQrgZS@y4zvBSVaxj!&C9U)N@G=ME=Ce-Uas~%}{|Smq(6*f#5*68j+18tZqt#Yz
z@ha0k?qkO4W2W8p&^fmY{ER*DUyqDZ-th8|87qg~*O0(4vm^*D<LIVFt0KiflceBv
zaT<;-hi+NE^Aqqi9=P`JZW}aSy)(&Qj}?oknWH<>T=L$5F;h7^B0k<pzfv|apT^SJ
zx^1Sd$O7AsU;MI+WII-3`w58%?N{dwjLM`g36R9R7J7av9G!e@>)^MW)@8)#Km&XM
zF9$x4{s{=+_|JY}|BoNo4~+T0JWL*HNXGv3aqhnUGs6J51hXBxrI80Yvjwz0$2YS!
zRs$u;NV010_w^Ek*Zi1AEq{&zjl!X}61U*RbZ_5H7~EwF@0p8dcmKZrw%ac%kXDKy
zALVv#n}aYSw+mT-xI3h2+eLT%)dysH>iUgg6Tv*b{fA`eS|;|<77eW9jbi0eH{eOP
zr{?=6%>S{dg;C62n=ZAhsFvvU@H0XV0}EUjEqkmNpXY84rSN7Dg&uIZ0dE{pkK;p7
zz)d|o&;9G;wD-Nf0H}vpg1O(q&)bQLG)|?!Fa{Tw0~qrI-WvzN^24bg^>JQm2Si^Z
zpBQft2a|%UhJ`WB^W1F>wJ050tt}5v6%3K^yj*B>9S%q>paQOHj=v`VdJk7TNd6}a
zMk<bp&jlC~e%xFCJ0*`Vv=QS<-J4BZgR-a4gG5i-h%8=!KwM(DO2+-xg!n}!eMbac
zP{dk#UJaGhdRc`Zty)l2hW}-iH^ww4owi<|34~O&@pp`(7KeE11Y@NqQ2Gr8^c%V=
z)JWWnj^HfG*7R}}S<o$nxcoJH7Eul5d5%@`-Yd6wor13&{KUwgNkTw6PGJHvx&T>^
zo`25CHCsC8br*84fdZ`$a?LKH-v?a>xmkRHU_1a{w>w6ea%6w&ucJq*<T@*P2t*QL
zsTdO%s(A?gsZBP`@eBC!LsMI6?6io&1Om&$)DAhd!~%KiKDWU)*<aZCwQnDIJ7=El
z(9-!Zk+8Lka$%Ivw(Vv!v}Vb+u=hveQ;6ud*LMigVKFZpY+Y^~YFoGVF@_BWmxX1?
zBn;vfZyL}9gdQHIh=uJCalE%nSP!&U--OIZYPM;`hadv$3GYua5QO~m=t)AJivS|d
z5Fei~T*f9~o5UeGMXRO;ri+dJ%`+YH0)Pm!a1;~CcJ#Qmk+70m!ldbIouL_R0au4j
z2@25+gi-<v7Ja)06W-#`7#$3lA%tH?-K=buPhA^D(Z1H{*z+;gmHG=7Q5K)4f?NrK
z)iE9(l;+H$S`GXVQ+8`|Z(?h}WN8Gwl)xNC^5XlXQUq8T1Ef;5c>5~`j`NR_9_gnK
zf0i#w=gOw9DWJfHCTBp_37l4xd-&;8gv=dE%<!z?d#4l<3rMg?g%Lr2^U?(nb+RD3
zPIONBXJiNuSCBBzx9}sXk^FT|^k=DjEj}@yK4n8gpU(Ex&pjS7+p~xC&+qK9D=x@D
zMHFc6no!PQALH{k6+_-~=M};!8nt0QbNE{qERSx>dDi@_E~)f;WEqt9xc77UTH<&u
zblKKtOp|zu5%zfsQP#&H4A7%BH&^#4`ao4JNF9fiUxEEVt;6!I$r+7>2)IZ@qn!es
zsqVK$bkfz;#yYuMj>5HrN;dcE*+tg9g9!vs(?}kUm~hq@b@E{y)_oe0o++^kBm>h~
zRj2x29GTn=P8wr0lv+lScV;@T6O;(Eo_7XR%E&%T)9tq7v$J*^O@riM7v-?f0J{a_
z9}FBfQ<2eL8X!4Rf+eGfc*K))ra2!!Tft)#a~P#jhBGSb??9j>U8YRdpb%`1(O6*!
zVSf=5DoY_AC^x@eV);>p`QUHoZWK+6Qz<>=+@m2G9_Og(c!RmVDQYiR4sfna$(fka
z6x+Wz8a}ZO{$cu&ta@X4ySsaxfa9;&H%iY-#22Ws(Y3dkq!)2nvDKO6njqN_(AAXM
zLaGKNP8A2rPECWBfK}fyHRebcI_d(llCr2+O@d<u!G}aY_%wmZnX`o{%EYD1=dX(F
zNFr~zJ$5^}LQCj~X*2+s13Jf!HzVZKU;$zZ&!gd2D(hVPnl=J70lIKuG;1A~{6{lq
z2y%EoaQVBBZV&y%#UXunYmR;{i;^18kb};&<Ljq&TezqzFLJrgAH~5F{&*?LqAPKi
zt--(zHc@1t-|oQWm-8ItUoUFWY&)~gnGAR||6U|7x!ld(q25kZ!!ZBciq|UF>8JO%
zUEQ)BvFm^Mmtg<LZvX#Zd{KU?iT7FmrJA@^$}eq~bWqimW(H<vkQ!{O7+BEI5eN7V
zxSlK8-^YtW()^)TNFgPjU(rZk5NXtL%KMBHga(m6*GBC1b^YOJ7txO}oS%_z{yNOD
zib50zB}A<76o`BCOPu{x|7ewUb6XmOuw}JjbZ_r`dr<V!CF>%`umi%U?Va-z*uPm5
zig{K|q+*4E#Xe!%(=F<*zlo%@pWgs#L*O^JQ!cN5-5T!%hlJzA;WJ&X&CGm|gM%bX
zQY>l`9v;OER4!y{TG-oZq;h)(?=P~<YIyPHHLK;yd%Fkm$oV;i2=C4<(#pxZE2kHp
zNTvqX%n5i(<4?KD<468%(vHVxAlpqhQ-I^<$a!WS9&ehchBc$X)O}3g?!Zs20;S7R
zPhk$XvI3qA%JN5LUug}mYwSBOS`czPGbpzZPO^sqcmz@uMM*}W02VQ3?T$e|hw1}0
zdp<-!7y(pCis&~|gSW;LFJ||V0#73fWq`LMV(U8%{py9*RTwLsB{YZZMW3JCabnJ`
zfjNe&L}bJe<l{b#C)A!*H%O@BvAVwA%3}7_<<s%B)Ez{{a^cl(u`rYhQz6JAg2WMC
z`r(lolQ$UAqmAsFMO=?h<&Pwi%Klav{%;cTt4vk6#e*f`9C2z2>{M>Gx4wewLf@B_
zkN%1_w(<Exd#E(3v)@o^XtqaW1WWUmx(yb}KJL*MLyv{A43*LmwO)<f$~z;h@W2@;
zazK)O5ye`}3uTF!S>+Pm7Xj%EA;dHMXifsO&li@sboE0`4Cg)*y>9$^5-Y}Tnf!eq
zU=?Tl^^+zYP95wM4Sp!}+iN`2ofGX`V_UgSkD{jNOC%Jo*RkbXDHy6FKUKKXf72q^
zh27`Egp{a3!?O4X%dIdy<4`Jy1`#kJ`kPDP9pKn_E!cu>0s=>(O-;+~NkY}s)ufVQ
z3=iO7E1o$QZb+tjmn?kZ^)qv;<3hMBSC@74V~b}OBk||`+#guqplb<GL8wF!f3eDm
ztFw{1q4c<nVYR#jH_^$A9Om!@8@b(f5ScuF4ax?{-d3WYcfMRo&uzXze~%rfugKu?
zz{1#FF$&j}t1jIQh|@q2i4LjToWV%LqtAjg@P(Fqo<!h!Zmt5S8exF+z^L^*vh)iH
z_S?g6BQY`2bzBZv)#q$u4!GTOlyDz8_W3usiY&vmjENP0RcN=>@iZROy)Qa!X7$`y
zMxPQP@g!J4&+}Hi@65F(&#EembX78L&L&-Kt(*F27{hgZ_n&q$d+&E|nK|6hLG7o;
zV{!gk)eu#!X8}ZeE<2LaA1$b*9bpLCg<{KEu1Ya!pm%3#-pmB-I^mp}Ta?s~(WUAs
zG6@w=m0feb#Lb|4EV5LK&z*g|qwHHKh{=&(OBIErDZMD!=&d5QUer8*(;=$Q>-W&V
zJWCcc+BQ>;#h-W@eoXe}4}K5$QupEqeG!P3r#m}a5s<IKtb~UIK0=f4PFeZTcl2WL
zBpP_l1%2(TNSSg-eL?&M-T*(cLaS>gTlHpCYaAh7)L5-z5eJn#nms>I`kHyeT9*k4
zprR3r-7#^sI#P-%7>&~^7&Qo4lC&KCC^1yi#1|x-@aOXxoYL4?btJ|7lT*^Pibwv{
z+v6S{AR-(4pLR-)fAU44|KE2cy;PU}abZO0I##_&1ArXgO&!^C@wfKpevms^(#EKu
zDFvfmk4p|#yolYdBPD%HDI2MthYjFkHBPwhN+%2z6uE*V#{aW2kNjPxK*HlQ&Exy!
z_SoV3com2;P`G5s_d$A#k{p2y8WPYrpc_ZG{C9D3`zu$ru<ISb?gBW6{jm*L67_i`
zyEup0Jhvy!Kt67uvVSt$<l>FZ%J2E~*tgHUJ~~!V<=1oB0@^V<tbL~R8ceoV+kS+$
zrl$XpKYG4;EEJXGOy8n+t2o1jB7!mJ$dE&_OWm~9(Yjd+9DD<-O7gR_3m+!0JU}02
zDUQ3CVVQBli#)IH$eRgTLQZj>)7s*wK=Zc1rxw6TV+au7!a+Tonnkm3luvUL$rNZ|
z6e%X?q$q|ZAA6Gm2BOU0l^eIQ1Gs(TTzx3&50nQ&VgQ^TFB<?ln8zRoAUcm)D#4~2
zf!rRAi_lGDzT@`w>;}ThuzCn7dvChEvF(Lfb&+7eUrzP5n}9fpi$#a%6_*GtX~OG+
zRt?dGdYg^4i{cl+C|*r%Jsgv5vH>JfebS6D`ElGp%jhSJL#tl9^?n2TwOgeTyd|-|
z^P;=L4|bZC>wm5acZ4W#)l)@iKkRUYbkX@|(}q)FC9tH^OE)i8E74TLCeITrxxdQ;
zR-w@*8X=bcq<-phszvH_uytht-RVE0#nHl{?W-`2^>8XtFt@lVX(yLRYZwGjwH1<7
zXd(sXR+mvC`3FKAOL0%`;Ot|5xCBD>Rj~EeB9mZrVQO)YK>?V~(|zgo8;RKgrY`tt
zHb;DO8x;!yw`A;7kM<7UG)a5uGHsqUNa!+-6%#4!H)^rgUaz@5ysm!hZTn#GEq^gO
zb-0Ym)GNzKzX!1{oP<MSZg@($@9Tn{@3-@6f(;XG&HFHf#lx@Wq*`C`Ok25w*%Xw9
z#KW$l$Qj2?(_{kL7{fM|twhExonf+*g+-tuJ(SYN+*fpDe%VP<Oh(^OSbiu`pV+EK
zmgMONIGme6ET(Gh#mU{)4N|ttwd^@Kw0LmEYTBRfsHA6%8}HS=9-uy7JpFRc3aQ_&
zdh{rhOF%~}OAvDNbA&8S;98E7APk$qA*}2qqicVK_M|*eJ-J8xh&28v3;&Lz3eV~}
zQ1Udv@$E${#qH&d?RZ?W=C<eP6C17`eS~O;0vaADXCbhp*$yQ^5s-s4nG3`s_cvX>
zJA}F8Vn_JQk1WLMW?u_O7iB8<ERHO!VqdYiZ>H)}ayupXNF26B@9`VXT;9KJyRm2g
zMkBLECFI^rPQ-FJEXw@rtP(hcn-=(Zr78JYrvUOBjS@UflEMPl7V;gWE&@x%rAjOA
zDqxl>MW~A}VRRhkGa@Zgp$Lh^w2VnRh{&On$epHQ278ylD>|S^RCyxp_Ad#jacfzw
zhvAg=*?p+*BmU_E_v%0FM{Ga!aQ{K{|IcjFe;9%@va<iL)#{U)nw2;|{SkP31Y$=e
zW6y}n|6gkLXM_~$W9sg|KWcR<=jABaRdYk;Y`<H1Sy$h;cBP6;5*q`GUp8~d;d%M3
z6_eOWMXt7XlzY=<MH-{n85*j#Neipkq`$_B-(+E{X<awkNeWaxS~AKiTA+SeOv*!F
zxi~Ze`Bedf#>z3-0!oB+fzq~jM`9a$FMkt@;G=d*b?N?Enlx9D5jb#DPN?P7StK#o
z(xefG;?SuUuXron^DM5u#^TA)f@}IUsWBruF3ykD!y-X>FwG)UarTixLNkfUEWNhW
zL0Vp3xD)d9xS0Y~?Mj1+_6_1C6@OGOa9nI83OK4TxJ^H@*&!sUaJtHoS&}i>Ga^yZ
zxK&*)@pRpkQc-iQQBmxHzoclrYP{%JLH4CJ+|k9UbG+bSf^Hx(KBO%X*F^lB!Ke4J
zg!JZeV1w~aNJSnR9~*y|1rz8Libjf@8UnZw;rqu}TtyoTEMe*A^Qcd6YO*|bZdsSZ
zb7(d%7<a8Js;Y_uE78$XJ*0a$s$+6qAI;Ijs1L_(DV*RNgAb4^nqW$s>7jia5v$5w
z?$Zc1K)EBo<!Vc>tT?ZF2&ziS?Q-GinbFES)4L2FUBTpi$g#N0MQxQW+){fKu640(
z5HUX4>OY8hdjoXY<H75hATnWl2=WdnXcq)*sd>j>z96f*bYpO7uxx^K)m1~BNZ*|R
z8OjQr7Xbv=Js)wlfQxG-nq9fd=4@o^e7*lWq13;5Iymg>g|Y1Wb-JtTT6&GQ{pt37
zr5s{E@<-N}>+|&ClL`ynwG{tq+ehpVy(`)vdE%)6PfrHTRK<9}b+6%e$uHTtQ(?}}
z-P=P$PHyR*_rpov>szZ=@3*7O-yfNL0#^2!06>z%d6K)utrMBl(c}Eo^m)+CfsK4b
za*=RIY3fQS1t&ey0oElXGwP1Y0FY`7gQO*3yMNYf70KPZG0Frw(JAOdh7KZ7-e8O3
zsCabg4DtecR3GwZgpd}{SL&7sz}(RMDf{Ns)Zb9G^HhGt2%Sd$2D+4V0gqTEu)`>(
zz1h^A*ro8x@bJ13<otZJ^&P$lNIg5<-Jj;|i3!c&=JCQXn;!WrIeNS`r9VZn4#}sL
z*oXR)ozvs#^}dJ4HGld#|D(&l`&x!SeLT+n!U?*P*#`dn+U@srWjnimNALFic)7a#
zd|&jA-u`m<-j}ug(f+f*=9je|SG`Of+ivYzgD~ITA)I(RaRne9GZ`dQ|J2@@C7EU6
z+sxLQJygp(GLvKZYXP@`IlvxZjj+U75-k5$GU?yhrB$H+9&LX{{v4AtXD#imgvF<9
zY*ww#4G2<VpqE%ua?eh&ttglU7?`+v)nQI>_wi0#tc~$2V%Lo;<f}n&?%HtA)Ph2l
zqKdSv%9H4cgUDl9O&FvBEZp_cVpdDCa7G9RCJTm@AON<_i=GhTwJ=ft(RTj&A@(*y
zo}rQ2_)%O?!insf?cNE9=?Uz(B>jVtPsr!03f0Vp*!mB9KQXLOQMD{ZGeBn_sA5WQ
zrQ~lC8jd~P`DwxB|K9cuMe1ceWPqoH*C8nDVAK)J2|vK0!QLao4H*L0a-@N}1{}^o
zM`qb;#~9R=gR+6yVcOe23N2`{x}Zc(+7S_IjWiT4+SSj^oj2u)k%fdT-0wA^P1Q$f
zL?j0M77hv`sKXyHxHr6smK<ic@`;i{vtB0Y<&;NaWo(RZi|(n-tWAAS&umKX?~QZ|
zfQ`P;NgGL{+Z2baS?4D~!7FST*%-m?II}rob!HYIF6$N1VK$(1&a6btB!1kG9)*U!
zF)n#;Ti@tavsM-*2-M@|!Z60q^nd{$3RR5~0O?5BY@0_1>2Pd#O%4`S)k%Uvd!JpD
zLfZ>L6i*dZJ!;a#3vuA>JhS%Uq=9f0jrSNY?K&v-m`tCaSOX)Tz5#kf(}uhVMyGo&
ztV|~fL&I`*4kO|q1p>a3>>vvcXDBiA$MS#>gs?Ue>$il)GuwS5K&wI2aWi&fpu4jR
z2I(xh_~O=$7tS=J_udgD>?91r98<I0j;%0J2utrRymBdLz7$n%nSmq|qHoG}4$!A=
zT<hiJ$L;Q0h7tIMp!6Uzd^S+3?7keaS^q)P*f$Q+O<PAXXrtte?1-uSN}t-H{F;8m
zD6zN0n9@-|c6CVI#)#F&h9I;sAc)Yxa+^MmoR*=rMCs<pwk2W^Q9k!)F;Y*~6b_6A
z6Bz}tIrHMGx4I<4tYI8o5kvTv_8VW=UUU1qiu`KaTUQ(9S|035e(&<%hGzJ3s&@;R
zaA4uTJva+yjb>`hK<WP!!0G=j)Pep>tW9`u%%w*0J!;X4+8vC4bYOx!rw_t4BcenB
zacCtFwS|Z^y&s~o{iVJavr1@98dnv+SQ@t#00JUxw~K3ciVtKb7;Fh}qCF0FqSkqF
z+MFzOvxtb5KRH<(U+CTk@~<qF_Y^+MG%H34d^oL2!Q*glD@jQ&u?#{lr_x<;Xc6;9
zMiMLba$HCs(4h<tbKwkkw09$}%N&wQ2668izHb9U1MLbFzTK2dj^m%1r3R7TUx8_<
zTNy7os$U^H9VsCvJGG%@QKn!u6u!X$mGCz*xmaJ5G?3`xw_dR<d=R&G`y=IpjI@YO
zw-!90lk7zqex#h!tTMW3WvEntS?b8d;brZgQ9`TU8cy^Fw$eQZetfYzkCiKVfGjNj
zNO2^eutAYi0_H{?;xvJETe`7Ekx$r_4LyG9PZLTIouTfvisdS--TH*L4s13ttUP&u
zEjyYeaXq;lgoPJ9WM$+_%29NX=<1jULO4`ubf;W?Sn`s9`;#_>r>0>^0-uL&{H=T0
z^^e;*5_ywX?-E8v8?o~`T4fsE7<coa!IVJ?&Y=D}DCWAO0s6t!Q&##rJxB>w+gHlH
zTkkL8{;f&(BL%fc8RAiMhZ;_b*T*Ju&9{$w#f!=CzImUuLN;<AasIDRfA)qOX0h{G
zYIeFMTH)4g)S^tg8)*jRmp*-4#R+RQB_TJL_#nEY8+i#tm&|xz^nMV2r}$5HX~HxU
z4k?sEv8ju9f&(p*XX>z14J{T&$nY^W&O2x1A`C&o#p#3#Jky=@5bPl&xcyFwp;)$&
zkzSX+&n;U5C|XJ6_Amq6S%V7lDNC_cSH?ACM04B40Ee9EzU<c00qpRut`9}$0zHh6
z0W>f4C?y(0-T|F~?>2##g5W{Qfp4$bg6~ve==V2a=)|%?%Y{{zvkNTer<Kd=(3R$)
z*XpvJ)PEhrvjCBWfM}w?)G=UcIIy+sSbFx%fA`G%2LFG!Fu|if{Q}>^31jdN$6XGl
zf2h-DU|?kX-_lu^>i-T@HckMr3Mv!{xh$he`_Y1nbclfi4V&NqdPEn2<nZ%$Hx`wb
zx0q4IFd@4U^26WdI9?xeB;_|X<vF{8R_@zYR%U7jS&f5^G{rZ0)RKeZC`J;2A?z+$
z7;fr*15-#9L)61dg7=;OIo#d8%iBcYzavL`JCv>Dr5RWZi)lLeOIn3Pf8UZXS6HCS
zLycIIv&HM>DEZWDhyH~h1qGC1OY<aKwz3E>XHo14v-mpPL3QpEPSH7(q!gQsU~9wB
zht*A5>bDr+E_V1Pf24zDJ&te|gt9w-7j46{RvS5wHWG^qfp`Q!XaNxqC#Vz-9Jjz{
zBa9JI^SXY*h)CxDXM~(=+#!BkPULl9W=70x&CA1V%j4b-=N)WOEz2jxJvljN?}yMi
z0J6#|Lu4(3&f9mYMXd7m;Bm{2wdY)cp6{hA<4SPlbmGiOc($2+5reTX43^OAfI5d#
z9f+bx?FVp-LnK<a0jLmdtqQrI+@_?^5EyH)Mb-i_^<#3;D3Zohd<v`FS>Nz&Q@R>;
zx_VvT?QVvfxZHu8<T2WT(1Vy<UkyXpS(_}rq8Rr=bk8Js*qdGrOE+Xn<2aIGbL{*r
zp;!(job%RE)1=<x*kPvIq4^-5^tGUiYx||G?U{1UB~$rcivzbrw6Oh#D?{$P3^qb&
zGB|6{HCw&`7S~eYuChnj%1J1SG(0VigRRMq;ur?!vdBz#t5bUo`mO;=MVub-_nV9@
z2cLVKIZ;&Q{0}N+bP=2yEkVpGI>wYJ&)?R+M@2PQpk-Pf{$fCNDLPpPJ~+l;BOl~+
z`Bq+}cIh6PvKn7WAP(PZ>94{xeb=JUf<Wv}G<)Pg^c)y(6;>Zzc5vC3r5AI?CqQG9
z8$HoGcB+I`V1qPx4|7xa$^P=nd&h=aPIVp(e>IQ97D>!6JdLhi^G{lQW!Ak^w=_#l
z4y}zajveNgcTD7x6mh5YSxg+wwT8zUM>m~PwVle-$H|O5QtN?7Ljm5zk5VNWJG>H*
zZIO?45{9sCae1-rXy%l>y1o}SiZ3<Zo0mdQ<*47${PEP;_Z-3Y`_1bYInrUEl*`_(
z_Ir)1Q$z7uCje)hO@bryu#lVCHVzB2Fb`_zgd^n8_uT_)hAz;utKu`?l&cI+M<<<}
z;N;^wO8qU*A{2aq)U{QSWUr8}#}m^QqAyl#aP~@|NaJQ%eIM$Z$QL}bh-CcwCvYgR
z)g=%W41J#OE`-~YTwTl~n^>{CGILp((ttuP)zb+Z(_C8_z#O-aPGHdj0Jrdg_kx;a
zM7IK4N_nSr4!lB$azo<@k%ofcIRN^R*E`B#mC9MHIarEe?MVtZ(0d~@njq)gV^Vf!
z@{Fg<_a+Yfr?)u_C$a?bw)hCMDB*44_QE3xf9v4s6pYNCj{|$6y6?>J;lXKwqXtS5
zW9K@a=Yf{%RYKyh2-fus{dH09T4SwMMj5nn+3)$OYE}IPtQjv@`lpSHo&G=N>Hp5d
zGyL!_|4SZzQ%%~oQ0!lDLpv#wj?zkAzpZ_Vb1~1A{*jgm2>ndLl&D*SUzt%y!?Kto
z1Dk&N49%v*_<TDxHP2_x-hA0B`BR^;G=F?$Y<Z*sTK5re*xBc8V}6Cx^~UcYBaEff
zW`(XRH^#We_PXHJ6woR4;kz2@C2f)I(rR{ZcxQioniD;VB#Al_OCT(xw+eTAHGYF%
zh4T^>Ctg0&Uh{5mv%?#bomL0j=?0sYXc_`ELQbh6%Dx0^I|YZRqjmFmbzyg$$f5;T
z?p=Y33`);?ru;tiBZ_FtSbXNSPPIphnJ89fd<rqsFC?jo#|XJ#WF==1IO@(E;1a#p
z&ajqjMMVi+slZ8J&{6=eN`<4|(ZMBz@w<`Uep6fm>Eu6$E#2QkG6fd<5E4mpf2%GA
z*g+C`H6OvGmxi7rRf2-`=!BdJCD)Who0vcDL(Z<lNn7y4l}vdW9{j#U?(+-ghn2(l
zIM2ayF6pxq$Gy4Yl<P+g*+$t-pV!Onvx5u22PEN<D(`gC-$4uJ|D=TZVF^nj4?89G
zZ|IaTC`F)t(aoua=|m4YTMm~U$t@->rv#Aie&u-q=82!_5RHC)D=|qe#pBk?z_RUQ
z$#6ms76*yjV~Lvh6eIz#U@ey#p;ezD88vBdhK!mtX$`x++N|TUc8WJ{Ip0NxH`laF
zIPGfE>v}buls*HJ6TfR{`E-hwc*S!?Q%;-A(t4pgD6Uw#-9Z2^Q_56piN(q-SMM9J
zeBW55sV3s;wc?72wz?Vj<sC!(wjuqk6^!f7Q%#;7HJLmrr%Kw-&Mu3)B8RVv#fgnc
zcB`X>6RMTeINfx~ilAGre+wvD2-X!V-&*Te;v79+|Hhk&k$SFe-@p@ON?qPdiIG1+
z3;AYkXktFk)p(C0>F|JLbn44LyO^|V-Zx_>VtJwD?~aF9qYV>ZxBoXuv=u436^;QH
z9=ZmDExS!nQCSU08^KCVO)WF+Tv=9OnI1JQMg&+6q0M&34iv;wm=+U}Ji&Y}26(I(
z=3Oe62^^E4<`~^lb5U*E$_ZYDsSMA(@0fc|C;=pPe2_>3+#Mg=s@};Yo^7nTD-CFl
zT`ZB@C$j`@doy%2UR)rt&v4D0_C{NPyW^nG99xh*j^aoeLxYg%JlnX5{WpsS(AIgS
zh{yu5!l=J`pAcaXoZGYllF6;Zr^@2*M~z<&MH(CyCyX9Ob~5V~^v$$se9V`fHU-o%
z2&e+-Y_O@EG^H@9d7W_m#fTh}!vJIp^{TYIzm<=^u5tJByb8QJrul%@`ute=ar#Hd
zX5fjH_(B$A$(YOv_AVUxvDsLnYx!Yw=_66tK)GKORTO3<VZ3>STUv%_0ZU@|b*zyx
zZ*RKdEca*r41cGA@-pL3&=!R(S<TRbXQ3fyn%`O-yr`0v@<dyaeQTT?31*>oz?CVk
zq@z6pISDfLBOpA6z%oHdfE<&ZIGIul$|4l~U#z`zknCHR{#mxoQ?66CZQHh8r)=A{
zZQHhOow85aQ}^wTp0~Si{N~QY#NLsavGd=|^<8_dXMLWI9Ev|8aS&k<Y(H_*Mpf;r
zeVjin-S3d2?R9?tO&DWiXZ*J?_J1xn{yQAwpZ7<)H05J(T423<zTp^;!juYyWPl>R
z#K9Xt)Ov&7V%X1pPH@7ryEL^ke6QDU@*SH|imoGkO~4pDQ_;b%!GvWoiNPM1WyDts
z$jXnw#5QNJ(oU(Y8uJ`)Y8KXyr(?qxjPrWe1tjuwd(;v_1)F(8H1)5x2sCH#rNr|*
zZ{~!Y+ty+4yp2?I7(9uOgM<6jjwU@9u&ie{x7Z{qXpwt%>$fyQYo5(}b$tzZI(em%
z=G4k(4LL;`F)kkE3?tS=%2_T|g%(oV`AUE|h8Ti~TXUMy0!(k>NPIWRq-9%0YLfXQ
z9m$2KiSr3+IUgeN{Lw`zg~D2vSx7(ZwDTfV{&;thqTbH66DvDt>)SOQXip0BZZ{PU
zXhA8>mp-@3SrPScT1ZA6PwqCQCk4|77cLJm2ew%b;7pi}l}eE(N5;6&cPx3D42u<m
zUb9GuFQ46@kua7*QU{|Zo>is0mth#9h-c**G%`8;xC=|x(`m>dk-G2PKz5Yg#)7=Q
zNG&7YJ>AX^q{IWJ<!*K7Er+x%!T^@QVIWB;p4+qlZ(^#HCQlCTfziNaQ9naroF!(B
z>$552Dvs(^im>BMhrU>`h8@UT12ZJXz(<*&LHm-&v@r*UNAS+}(ytw~kq)58QqJ<C
z#|qn6=sgjdg0)a`fpN&d^o3zObI7N|0>%;v_=%K&J?zUfdjW?8=8KpfQ0Y4`+z{}^
z!EK;74H5q6f(!#l`KOy9OV)HjyF;ea&PLCu+vD-#>gJWVXR~Br1s*;|)+Iz0@9TxB
z=IeD~pywKHhpYSR_VWY&FH8pe`^!zrKnw3|XDg?t%hwh8hhF@aTcXB~+U@OzUYOma
zwkAIv-4fH8`8{Qh_tV=4Lk=!kpXbBNo-4+>PkOJ1CEU-CFO1S=tu1|t2~ZPBe@l2(
zV$6m&Hx$E4V0(qYRDMKm6asAr#w|%im0uwtVgpvKUXSrp2XW-p*1E!raCN^_K#mHX
z*4GFFJw#9lsN4E)L3jt2s(s0faYOan^urS89f)E`Tfs00CIJ4pAvke^wQ}F<nZl^e
z<%QPK2DAJ+fCdTBhw%9U)DrZFPvlWs0|gTr7uW-nE{}a=GJ4EEs#XK;N1v{R0%*l#
z^;5W)0K+kF-WRYJKakN?xjfRcZ68z1m>@j3SE`##{xK*+4|;mjf;<$roi%(I0?k>Z
zPvm92+#VaD$s|64LX9kj)T4lmnG=(%0DoG<&=e(*WYEnaH&>@&)uw$n88Q(E28$AX
zwpVwoEfY%Xu$;!xt+uy3H49s`HkV@UEU8re?Ntv@0_Nk6T@B&IsV6@yNq@m~ht9%Q
z-4jcTQA{#e0C9!ROg*O0Eor0z8O+jEc^xx=&1;qs%z918xqtVZq+X29n!8#%Yp<++
zs?pY8Ko6;Qh!M(98sfSbIwtCjyxWh}UlZR3<I^VoawFGQU5QryCUO-Lg-cg+=3GEQ
z9ud7$j{uc)Y`Weq>fJKot_IZ4yqAIoWIX0~D+Ssjn2Y|`2UN-ILxMcGnQ_--5cst>
zvrctRWcc6&r6fIOMxqZ11Xxcnv#+iqjgD%4c#?R_3yXahV7SCurw|4ig?CCFI#i5I
zDtx{;k*nQupD0-M%AcI1VDnHo9Vd{zM+=&hD|*;f!cEFkA=i+*b9-mFD^T|n&eOf$
zI5IEdFBdo&iC<v+{zIzNU1S3KMM_UmNLuV3#*dYuQ;mW)xsX6TB@u)-s_WZmev>6l
zlC7K5p2_J#Q{8R!$hb~Bz8>yp^mbd-hXd`793sxmF<c{0t`rOkcl&M?ZmE|rKyI>>
zo`7{pPqk;i5DE^<d*lO4_`5D6`o@ZoFOh>UBM%bVT(2f(ZsfZqL$1F*;Zt4yTt<b&
z@P56Ma&)<Rx~9+TJfroM-9Z!53Jjud5|qt$_8W$zT;uI@wS!MOy1u*%b98opJ&<?3
z+)wEO&2H^!iuHW+OLjX@k7nMWy<Oj*Kes+Y4$wy4FKwTAKkpwiMsybz!|=xx9F%)J
z-+ssv>A|x*XTHAs`W<H{AEaq&lmv?u2oL1jUvWjl?o4tplVRaRVZny|$LU-FVmvvq
zqF8aP6hWE*Ly!^F7|Op6j$$VX(gYbojNry_;@OEE#Qr*H3;vB2Zz)zSI%cb~bhJ0_
zHHkAHyrLdE)yLG0AEKpx2*^$2>NZ1g%%a!Vf2>Ke!CqKhQ~MkobuB3pB)og<$ozZ0
zUqNv~*vZ7UdrC&Bcg>B{qXRN@l?oC;GlQfBK;9ak`O*|2TVifi9a*{S!gWq)DLdq%
z3X*0mhCY*@4y`B)A3tma=tPt?mMzR5C{^`6mJKzwz_KVRXbhXY{;2<WFs$|m-b{iO
ziHuVhX>^$4i9Vyfh6V~ecP3KF?10LCqY`WhkQuRzJm8j0=5qQWz0nMnFfG9bBe~rZ
zC0Tn_CVa3U6a$!_c0XCVJF~tC3|})Vl(JfomYh{F%r}1o^o7tJux}T+>#7oJ7Shxa
z(7++Zpo~za<i~D?K#C!qCi}qn03PZs-)PMCAef&4dD0~?f9Tn{{OAoVP+Km*A@M<%
zoNvc74e0%K2!0oodMu?C9%O!!Fl+J4nLY8Syiy@xMh6PvIWqyo>JUx(>S{r26rpEA
z|M`AiD0O;=0^;WFQB#&V%`cJGoz>ztucDaoQ7<$<k8Es^)_4C8zaoU(kYI#EsSejA
z*n5dLmN}hpKxnrno*i9%jHdndiC9e!6iI&HB_ay_8~<I#qg{p>Qd=weN-fsbm`x$l
zh2hCw3j0gpJ?i2vBq6hYZK0{@bUXAq1I_?|De4b@{eZ)51BQJ*IjLb2WA7>-!ks&Z
zCnf&9m}79!mdp8fYSICA8_WkT><k>2#D@h$6xL$Hg#Bs|`(+`}n7If-q5AtQ0^ik_
zV{B6x&F^b+f<56HJ-Q~b$E^EIX#4vZJqx?@i<kO#&(w2&{d-B#^A6f+O5&l%%T9gg
z11;smTiVx7(&5KjpyC#r#RO85aoBpjkhOZg1@&#B(2~aYyL?k}sfMK3a%?4mx?p{<
z5$qUNJTtN7zYf}hext=}irEnS_Yc&>nqtkdRs?GT^+86Kk`Xs54ZpJ7(qC1HyYEhP
zQLPWnDZRqN;fEW*cEWwo11xL2ZI^aa{MpR=k%{l1cEpi4cBF1#`JTzEp#!7dPe>j3
z_4>S}c#jOupBELF_4c6&{c~~@zDA*5+{xR$e9zDv_I4<h+9!vPz@gC*zCatI4if)&
zqR2=`|9Ac*bl+<9e|X=$RNwgaCqekjO^qJVD%mCOt(8wRH^y(fJ}7te7vW;OqB<N-
zE?!ST1f^tt;Yodmbh8-1Is-eG>5*M<VeNyS{`nm$%?58)1$+*KesWakROq1<H9>4~
zKavnZf778=Wyg6>Z7*D#x@2UHY(>74*<5#T%<q^C=DeH3$D@k7!-1dKn2Rjauo6Lu
z#B|y=7D*4jFMIHi`Xrdqf9@`iDbH=BLA#5~FO+kw1hi5}SFqN!lo1koH<tKSWG(3K
zMyv}p2&xvj3KM4~3qdwZ0pCd$?T`n}Q@dN^Qb_m8tf07@@`aA;M!|Ih2_qN$7h5D1
zQcRA(`1ZUtBnULeI4V|Mt^*da`wGINTRhzFHRvZMn|@?NP18y0H%I3a)enj!hRW0E
zHSU1+i>Qrb{S1xp5DO?FKqnI<lTFE+%x)GpqIAFs@iSJ!xKI}!NF^3gx>3(fpYOj>
zXZ^t9BG$obHDc^<*)Y5)+2f%ZO%&OemXv3XVs=s%Xr+B~72ogL^s-JD!=o(A?l4NW
z)65Jq-5NBiRS3XVVHja1o*>Ws*0jKb<*>h)zLzK8RAle2kd~T+I_qHZfZMe1fcB2V
z>&UZ05Qv#`FT6~R6F}4pk8rHo`j0^a4nJ~oM-DixaEJ-og0sWXRl*GX(OsZ}QUP0W
z4#H^16QROQAWkPCROHtr1(vLWu%9zu3e!(ZGCd$4DuU`hx9&W-(FA&SKj=#?2$0YO
zQuaQ<Sjq6Sq*94WgBeOcPb)Y3ZK#QF+|nX1IjABXQ}u~%_}0F;gS_#$s4QL@tSplE
z!>Fj2*hJCg`cux)7Q;$P+aRVh0UNKtbH<g9G><0;G0>cb$UuHFF1epjDUT|FZLao6
zj2%sA2(+m_04k*uL&9o4Cnt3p-08UFZR*j*+1+%ltUAnO1B!V$RBvM+skq29PcLHa
z)Iz29;kRSaQ8DF*_dKd1f&|?B?rgmLT6n)-0wb8<UMksXpktBLDaFgMuxWSvgo@TJ
zwCd(lLmmsRaY_~EeoqyryB3Z#c21e;VUFT8O}(h93jK9E+6)N{BQa`D;y0X4jL}Vo
z)U|*6*p19Xt8|)vvpS7tI!%BD+oeV1($Aqtpvv&D;hZ=4>f7c=5-6c&gBVWdz7hH$
zB*SxB>s}tAq(*QQ93M-;2Djk+>4LiGmg7#Y8<;E|FR6JPZ~vge#*H{XH7a5RUy~Fu
zuRZ2!+D9>P38KF|;D%23Bm0@Bi=_L}vwP$Va}Pw}u<$|C)M?}4(^!8DYo`|=Z|f8A
z(<e$`<R(!7omlgW`e4Uq0rBR8JMnb)HxIaXyDSt9Ge0xO{dEHqzrQAnmG^~s1{LJV
z$xE-uS|Yc~aE^=}--`*C3!l+1zT=9!W?&F1w$;}aE;~{{LBLQD;<ecv7Y8WFa;j18
zS5M}V)Kb`w6Ri)}TX2QHaj9pd`#aDpJsrzG9<IJr*R(BQL-D?@*{w?oA5o2bvKCp#
zG;fk+jC;)E6tEFMR&r3s&vI_!^_G@Wppk))!fHw{S0L~+%(?ErrT3ej_nS_EZsk1V
z?M+Qk%{jlT(WT*;qT%r@9^B)z3(Z{18o<nj;e>eskl&j`o1P-+zdw7sh}km3<1q|$
zg4s7hK0vC2`~|wZ$VxB&2Z72MXM&=h5gdt6Z@IlI*1kC3$nDSoj5b0cSs#b=(22W-
z+vC{8fz{;M&VA}d=x3K+cPk?~p%4~-GR{{p%u$}wP|`WS>x&WGSDN5~ztWn7&Qo_c
ztYldTUyt%a+HIGL%u*LiGh1({xlu-j%mBmzfP1!fE#Fkku23i4&ri#P->M~rPrIcT
zA7iHq_1?#Y6XMkouZ7^I2(1ydpYKaxNQ%2ejv9%f3cccZCW(^)6lHN_m-yckU)___
zN~ohuP0H<&<A^wuFy$iz)L}L1cwqo=_&0u_e9ht8zg~xUUnukN?#y3mjDJLvW!C`n
z8>+t}qEfb@?g?Y-^!;S+bn@@>Oe(K!^}*^09I2ly5EXpO>C^2(oya6PZ+C36{c^5e
zk_Z)lu#}vp&xGLoS%C3^Mt#u0>@FCLxo^asm8gh0<l|P*Sl1?$qj(QZW<!oB>S*0v
zmBU_Xem5T0-b{el5)dubmY1Zi{%Bfd-A|91*2F8<+r*VE(?TShioQ4K{&9-|$_BoO
zT#)M+>lmYL!~;Hg4EJzfnA+?MD%`vZJ<W;8_GQ}pV1$H&z`tSKFpKTqL4dQ|#aEFK
zhNMZf$M3r)Rs-93$Fq2xA3+c|wlSP*<cWPI@CY2;OJ818Sb#6b)J(oj5IPq+vhT!)
zpb3!{e@7?R-XSf~B{`koSfymd5&@2$#>0?y{O9zjqfb?RqLZ~O!sYtRCd|6!r|j#k
zR$z?NqQIG#hexlnT5E>h>eUKS8b}L8)(*Jvy{+2FPE~>%W|25!ZnjxA&!`r-`SV7{
zS?;f_vRVNUfKsFS$z_P-mtWeKFyY1d@Ozz74jUFA_}gY7jr~~Mhj4ph4SbpETvSZu
zhIQUzUjDfI`qa>jbs#SRyui;exWD+2XaY9K#1F*E0aRIBS^M0i03dHXImSTL5b#4=
zm;v9^k&VoiP|W{Sf37W#1OiH|F+I=&XRnsYMyu|$l=I5m;&x!KD>>qon<W)T@Z??b
zv~fydKVlJGu68w-&fX7te`rTM&!}r!&*qdLLnf4kHBZqEVYkER`Hgf%3)!6vvCUx{
z{}Q`rP(IfFm|D6F=pMmjmNwbA0h{0-d7Cxy2&lU??0<mMzMDLIKgN#4V@)Wcn5<wT
zMJJ;1{p2k|apT~QlwFQgD&E|QWfEBqbv(;<E^;dXDBbR9UH_)an-~dns!U!tM0^1j
zQM0T4T}s5l@^7`X|7-!UGyG=<0RR7)>u{O+FQEO%^(TslRAKEJtJS{{(<<FXsDv!e
za&&6G1;u9}*Ee*a9v{1NY)QW~BGcB^vT{B<82=iuctTc)$|*_M;FMsmr*C2xUI2|p
zvLuZwgew*(IK|=C^7<M7*F&G@y|o#IK$Sl+3BHE)u=VLW=2TUPxm96weicpT^mGX}
zzwN7`KHVd&<=|#awyHn1@nXI=VsnHpo4dpOW}@Z-J;N*+)P9kvR3LkZRtHFgac9e0
z8Phn!>ExKQA7g+Iwt+da%cxELMLv0XIa$AX{2+pNV!!Z};kN6hSV#NL+^2sTTv+bV
z$o}M~8;1Gk)k2dnW+%1jSl8xu7?;yJsqh&YDGUy>5F!zfB}rT?7A2&7wO+Epk@zQI
zpF+M-*x9k|6LKwuwRIS-cpO43kA}vta*B=kj15i^bzrQB^Zw$t;#tj*9~XKj#<e~)
zl|$Mi{OQAS*u#>nVabO`>(8i@qsaw`CkuG@+WtH?ilcznf_SPhsj-^x&o9rX0nbi?
z)iSnqnEE#tA$^KCGf#{bxck;Wmkz;LxR~cHCnaDH^#P>0-lJYILP%$$05jY+S*z!+
z;PO95@EJt)&Q`tnjGBA>&#6NTSwksTJMsVY!-K^18yd_6T4aonVzAqTQ8o6CBkmT_
z@dEY*3=y~O3;mE~Jp_`Zi8<Qvx=$e64%Cm$@!&ThbmsjiYtaY+git{gOZN*ugyyEb
zP%LqB=w>W%9(b^zfA6IU*}1-4bq>bac)h9f0!xN%%Gc1R!4*mixQl#{ql8*<&Kp#z
zy6CfLn&~?N4r~Deq`6+da`N(j#z-)MaURj8DviZpphOuQ9t5N!^f~1LlkbGu1)o@B
zE3~u_(ndBtNoP<Y5C!6;u`aixY~)6p|MNv7j>esK6g>#_iQ=zU-X4an2*R03&<Bau
zsKQFGwI#)5MdK&9^BL-`7=(=%qLQ_O2u_2E78?fzzSbcue=ut%ml?~*9h^D0_iNdz
zEOWk2WE1`Pam!Jq-9VB)#G+Oi)p_unPA_!kA0H)gY5w3Yi=d^^87#nua^(_xh|WcD
z8VX}%ze;f4g$4b`vEszYkpIbsiY?{T`n0dBibs0+4$}(NQ=6O^uKBcC0)LteC41uv
z;|@o8U`Af>cS;#_q=v==N5pZGyzT}zzy>UhvNx;$!vJ#_G9c#f)BypgugsQ+e%ggG
zJd$ju=l}gzXJBXh|5hevVgIMsqbW`83Y-x{ub!TKyo8kP>X_;)F#THUL=Yr|<R9@6
z++=@9Qd2#mczR!+*q0caC^w8m5gP=1Ynj^_mrTnmDrU-l+|8y-?B^wm4MrueE=FY=
zF|mr2?>{ZZPrheMl)*X#EbLwd6H|!Tm=j{TgIDw){!wLH;iyV3ecV%#U37KF4s(Bq
zt0cx6=xU|hKSTR@bXe1>9t>>nnFMNxVN=I(4V=B~1#_2?w$)FWc*bW<oX2iMDQ*<K
zK^8XM!y;3kYY|^KWtu^Hx<n^Z0WtVTRWdv!V2PYub!48Sa+?leZ)fnxi`twBufTG?
znn$EYTr^`wqg;pw>_<iA4@!WV`6I7TXX;NV*CmTHLww>U=A6X7vgpH__{!1|3o`}_
zp7qkPn|>R022?y)MYAUQVGM3{N}geiv36HGsA9)%Z}L<{sXWb0EJH1|_;go%iMd6K
zG}*;Psbl11jk{eG0E>cE+{$eCI80daEUt?ni8C&4B805Bdj^FTmg|K@kF=+y#gK5#
z<eT_U5N6`lKP%b6WQH(O0QHL_hw-tDS25)J!#Jp<@b*rV7wqasy^)0=gn01c!a=4c
zbo2FQx2(p$jmsGD&a=mt>5zyRC39Hy0Oc_Ei696DB3`;$Nr?juzdbbzJKzbrLAZAG
z4VcPPc~dc9F?sO@%oq>#opout44I_R>q^jpg63i$en3mn*G+;9yy;8VatfwGgk4Y9
z%0Q;^9z@KwHt?F*MDLHH*aepNfF{?|inn|G`F(Z&XNPyIWZ|!BIo_2FEzjq{K<DQj
zJp5NXr`O%{)!Um(4U~nTo7d~X={I^>cka7d?&b3JDCizEYCN!yLcQhyVw(xuhl-mf
zGT^0K@;@cIr{5CYyWO3Vn(p=iA5LGh_=+Cd6s@}KGD#J9&=^lx|4jgaJS{FbKtG8i
zLU2E|p8?pW>Fh01U{CS#=I}uD7LS8M;U<ZzgP<!qQIYjnxHypUR<Gj3set<$#Ezl}
zEDUD|PDAkC*vUd8*WXjb5Cd2c<;;p*m9h+DtGRGM1VjZE?rY<~#+h<B<G(@EGjLEL
zuhKTCZq(~m;uwnc(M+^-#UYn*C6{Vnj;f3qXun9_L^cpp?5TjP$@&~c<fvJf0`T?D
z$GQ<IYV`{2ex+p*DNDgV4j0H{kQSU^f{K9(EJTYCP(Qhc1(h40J0d{wyP4tBNpL4L
zpeNG&R2UDHt`W_<`S!ek%+BJ!1LYKdl1mJh(^Aj#8p#q&WC-wzP_=x=r@M%(BceMP
zcSK0R<wwZl!3T0O!j`eBrQ>^svBz{rfEeDzO*=*OTQrn~3=NSsX*sv5bR!3HKYd@b
zyNtx-3_oL|AVCPFSC&Y=!Y6i#4VSh7XXqgc2m2JFu^k)BoS3Jg@oO2~>Ly!0^mEF2
zGj_8pf+_QW9n5wDs*LW6+qIfLL@ZB~ltk<4jz=Bc;1={la75SwiZ&O6M%X6*6NVT;
zNtXDy!&B<F@<M?-8cdJ&F?WxFV1SRvV>YyXDBQ4zg}6qX*0L%VU^Fb!F3#X09nJ@6
zV(7ewlHlmfNx``CVwP-LV~`aMo7{FTW0SsQZae`+Kv8A-o@$KyXN})A+|IqGvhaHT
zw8rO+Q$?*jR}3D{=l#}c^Zw#?Z{YqhvU91}xa}jYMWlZ?TuRwoXRmJI&LzfHcN5rz
zgWKnoDobbk>z$zM^Du)KJhx+^<;MH#X0L0-`|<b#bcfg5^Xc*PV}2q1>oMi++Q;>7
z>R`rvSZGLh^XJ=3hKuX^hTHi7WA-)u%>^7s_8N{>6jpGMeRoL+y^3KX9T^OqU_ZnR
z5m(F&SN|QvjNvU;hTs3qC#j{=z=$|r@fd4K?wV2~Sb@<B?M=yWyv{Tz(ngl`5`_)(
z{E@w^HjVoZg9X<EC!sBk3t|ujawgvVD=g5ThEkXOrR$=G5E0?F+q;jM*^E5Eze7ai
z(dv!h3|E+?5P6lBMQ=S>#zO|o4|c=qlT+|%DufSKx$}OU0Aabzt`Er1PUU_hyy$BE
zfl{pB{tCF^DXa6Vtq9FoxGs2q{!2@sPCTwGiMvA^A4eumRYz<cF}-;T+C><U=cGhK
zXw`2CUbPm_1AO{rw28WP?lW$d#eo2*Zpm^THK0Nb1)@@nIN#k?3xEI2Ic^nwNCLX6
zX5acW35|nR;V?jfP@oSuEK~w9uU92Xt4uWS<YFpItpKh#dPc2a(*_@(Ky%6iivS;%
zp;^m1VB5F2!>UQWkNItnOwbUX69?2l!0Ed>jhj^ah%GisFyUs9O9lfeE?UrgVw8fp
zrnJDChM2N&B)aHlASrVl0mR8Rm0*BiW7{SVULH&lopo5+s1^Z=XjYVEIdnur`nFQh
zkgH2z1?k?#zt@~?x#X}F+$mbtx(u7FJ9hfwH2xKmx%{Zf&dn?EO8EUsqHgpvh^Z*H
z7-c?DAYB2TD0?5F3nxEC@cvh*4%`e*HanMt>(R~Ro_c>npef`GQZ@;fq-*jG)qfno
z&R}ITb5&fFb*a**_mMzeH~kuuK8!N{raN6KwcWwEv3EO>S36z?mT&LgIC3+<_|<tC
zxk4ilcmNL`h3!%EC1ywc2I_wZUaiCbG(Gh|Mrd*!>hdk+hUcb0nwG=Tg&WUpYj`n@
z)C4zXg&cC1L(|T?45;U(-yK3-{1IL;GwqS_A;61aUWD7A4-R&4v2&3{&osW$9@yer
zo4JsMXqB~aSym4O6j1%V!s`e4D`YIX4`E|?5%CWAxADe}$z-i%PWpu#enWKxTVVuJ
z9W~|fDqi3#m<2k*60a;axTpca{dcK5@Dt5X4f6wY0Igt$8BV@tg{YluMHA2e5)<qX
zfUhY>%B^u2n-a<-wKCvimubkX9Kqo`bNVZ|e)Af^CbB=>)HPYTt;#kul2*WsZT9H}
zlmY67eK8Tn=GU?B6eI2twy)aSCFUS$$Klptv-?bIW5agvCqNC?Sw=W$|F1#^ky@w1
z0>=Q`n#lT`()QoBYCkh{eIXw{<UQBvvdNmUh6NtL$ZExvMrZiy#)twR-*3#5$Ba8J
z&hS7COn3@@f?sv`knxy<xDhu?Jl4b;u=p?Xv>@K^eA`M`A$r*pjb5}w;V;nIf|%bc
zFuf8p|JUZAW6mz0^DFzWy|>oi5M+9gg-zS_+R}^rmgjm<lya#-X4H{Q#1Tj)L>V|!
zDtO%VgqMd<{I^wc_J23C{Vx?5-9M(WT>hsD?Co0xR%l<+bDDi1mJ#Rt#Po-nQ(%x3
z4!Ik|l~m_4W(Q~;q!zc=gy9EyiH&}F>C>6AQ&U}2bM1p$@YJh0eMcKfpgId73(@j^
zUp=UTznVXpP-`<9<@wdw+yui;ROTDL4i9_USIyI&I}Lre`>N2yW=z3(O$2Y>o)%F?
z#kdKkkBcImhWqptwPB)deD51LbR>d*P^ZmB@wxPgu$zkJOa>UKz`<=t%jCggXPUc>
zUbA;1tIHdnRhJx?vKJLxXCnj=WwuvmHnIf{I**NwTVH0g{TL2^s=uZcVuV>(svKGi
zGogS+1VXpGw~I=rZybV3d-K^x$_!&|xs^0f>td*32ZQk-%dqmcjSYETRN+(_O;ma;
z5G$k>J{<_y$0#fgz`>@6If|`+uZzFy<*`hycz+0#q$ZdUk#rDqNRnm8A2qs*;<9g!
zs)mEIb@S36-KL8!#T#vo#43oEXuPUUQ$f$94hy_=Q%Y8s4V!gyA5Q(Vx_%{7D8(-&
zeFov!#xe4KBNjiY`mi8OH_PMgnElxh@J6{~0(_A~SZ@YaDIi`A-I=mpRrRYQow&pw
zVlIapkvA&=w%)c{H!?SrXA*ej70$3p*kl=k8eX(TiV|TG(S<xk@!y;VhxQ6$p3eAt
zF?t=DlrKz+#cO2IDiK0!p<8iea732<{A{*CrYP>P52|l%Hm<hx+}zkrbajaE*2Xw?
z-IsUPyS3$UJ!hLQ$m&JnJfJg~Sf3J{wRvraGqZRrUoItFPz5<&`Istad!}0AerIs?
z&Pc><J{76saN`iWk=|ym42ej%Jl#;Jn7dkX0M=_4_}Ha<nz#H~zuvaFA%E0#`fLwQ
zFEy${Jw%TqbLEoDUL#Q^cFL9jSy;{QAH*>>L^Dg(d|brxC^V0OIq3FJR{m0k+dLV`
zG4`2mL{J~nihA&}emsf($K0<iz^yA#=0(H86dt8NcN1=!pYujYQpP*eLtjz~^dN5S
z=H%Zy`JO_-bFzvXoCoZbR_J42T}`0Ytk&aEp~jk#^nOhtv9+tpxMfEB%9jwXbE@n9
zK^x@lyuY~?l{uG=cQsE)U#9n=TEM8YM=iFK&Bu%si^SQ3s?C{E{m2zp_^9n0Kr9YM
zAcBb3k{ORAJ>2S8dbBi{e-1JH#K7GSD^`i;$uG3Aqb}C-tkX};RuV*6EqM<Orb#X?
zKoN&~zBDKA$FZ+mN8$EzNVZS^(cwKY_9<7C4%_NI5irM2cLt*K*yt5O_wu68!C&u=
zF;^|XD*JIP!Q<0%rA=#q&EWy`gG+|>BX+O47xR3e6B9t=BRltje%8E^r>NrnArHKa
z;21zfAa>r}pb{q2`RjFEA67>8{1Lq{DJyQ-f!Om@t;30vuk95maYBU51)RRp77|u4
z6f{@-`?(Aq3auARKM3GsKEi<>&x_XZ1(*jY68XRHt&H@{f7`XA|4$#Ze^8?Oo^H?l
zR;hl+r$!Rll8TJaacat;xL=V+M^OlA`Zy?h2zoWR-|^`OK1*@m>|{w-GJK?d@oU$L
zat&1PbW}hVt!XRICo}Jx9Vj2JAD^7v5oHK{A%p{l7)bT6Bgx~}ciYJu92ygxgS)qh
zGq#^gxBR++G3bo=uQ|5pJIA`bT5TD=61xoeDPq57CzA4LSa)X6&v^$^6RhnPb9%hk
z9jZ+O_Z%e7ZI#J35KNs$(**jHWDy1mOX*0MhpBD~u$+*M-PS?MwFZkiaOHpsAyN27
z2G20($-1M$>7ss6K+Kf+Q7qB+_)y&*65nELV4v|u%Igx696F1^J~&Cic9)qnc4N+W
z@mp-iaceUuhJhHX>_YmeA8{CQBBPzNMftx$tz45<^Qv+b4u!6hpO+IDJBht((N<CE
zL0gOd9t2U>Pp^F2G^76!$8{RzO+gSUTM&8z;7N*jOzKM8DyxeNxJHaX>XQA5g2p0k
z<in)J6(r)0wHvZ72LDViUMRyFHrk;KulayM&}A$KqK-rAZJ-!cr)#8deVt=L)2X`y
z8IXxJYFP#vkL%dKG|tij&z@bb%9u?Naobo2NXZ-CZ{(nba!&OQY`Q4ZDa5#jEKMJE
zq-mxvOFUlBJ{OOfzibF%@5<{Zcgslj-Z?}eafdht4j^VdQfUmPJm^hJ)`t~@R5WWJ
ztps!Lo&+)jdR{<`;D_tt;8)AO=zgnVs)s?<g~y)_9@hfZ(0St>G&XNC&s`YTa4WKX
zYUeUc`mvobP-hO867-^k$9p0Yn&q;HY{~rMt~}Tsm9Yhy!F_(h@&K9;4MGgkluz^m
zfD02!X+GcP)fwRRZohMdctIExbnpu{srBc)@33)dj`8Qf!;|>TugippHrjwn!DfDz
z<7fI&5aHfS4u1^0agM^h#B-m{ZLX_@saME(gjNazO=eB44!J62?Qp`%+%3_8P6%}E
zs7<SvHoTGy{*Qbu8rpZl(11b=my*+dzDw3+%e*_ZiBItamsdxf6=$dCr?UJr=-bO8
zdm4CKgC6F)TJ9;XfM1UA+(|fd`zQXT1yF>r{J{@$wru<`-1A>8KScUAi?2Z!P>3aM
zipS`C+kHZTc0={rHN*MXf$}4zX{kolx%KNj)Mbi0E7|oVO*|zew+&Sv3+_v7NmpZX
z*;1$I=!Z-yGhKG_OAEpYrR7ChrCnID3P1gK5YWYaxBh^vERo``3ne(Axma2=Y}y2y
z96&e2&Ui#H5@p1vH2o+jK2c>Qxg~#mC>_P_oH5vd_!U`K=cnATLTTTn^x~e1PjO!>
zNS==+(?(^@&Gp5_`vzYn_ACEy!UzN1-=#SW|1YdtH6;HMyMJ4^+GUE_m+Zu}Nf%Cu
zRd7$eCPyJO*A2jdfv|<Cz8$pC6CatIWw3=g{{}{z753UuRn~@utx5u|x<;>M+m@oO
zwnZdN1`tCql1@q1<6|(QCkD(PPp7HupshQ9PI+3`sj)S8u0XPLp+o{8Du%tZX&=9u
zO@4r<GZ|V#4*%lq)>U&sp4>&3Qn%N<>V2RVX}!OPL<aa}BaMQfbQJtyhZ=aherxT6
zOr6|Dq+AjQ1l2HJ+JhgEwlj7ZDEMrgf%iEDUj%(VrF@{ZU$q>~;@fQkKT(Cw<3%U3
zueUT3VsTgi{s^M)wJ(Y-Gzb_xh(N>}iTD@6o;g<5#%iWty0-RMKen*$0CdD}i>2<h
zfnoN&pG^-jp~UZFBQ1n+1~MAvJ|;q$xz5JFZ@ziJi-%UriE2|CB~7%!uYbxRL){kC
z@&J^mHx?@eEgArk`c{YZ5H|l)n|<%bQ3j@75OyYduW{6+sK~d``3PQTcU}Hsx@!<x
zny^p@LW2(h5`;dZKJFx~j-*t2)$(V3(_5KecuO18gg})jH%H{%*Ia;VrqllF+V+;J
z%_#!Cmz`n>MzftliPi3zHy)Dk?&@tc6m0E<+6o`7nC`Hf@$?hTQU$HAt0;9nof3e9
zvA&ESkJQQsL#{#Tl`{DNnh6S7_+4J;L#g1RjU4mf{HjOu*6}EQ^B;cp3tP|=ZPTP2
zTLUy8Ih$mfa_|TwU$S+-AsyTI#fbOIx^G`$U*|8|V1T80V6dT5F~lHWB&~^Z+(ojA
z46wAN;9F+j5Wqknx}g!!&?f0f3xFJ0JLT{`9-@ht)GX_M`EP6Q$XPlCLB$6<h1ALr
z)uADUCQP}KBoK*l`>aXnWw%X5RrqLY-jsM?G8n<H+LWLn!8DFQUVeamle<j45N=8-
zmi79ZF&bIbfW@hOW2NI?5l+_9A6NlR?o*@y<re9*ouamT>)jnvm&6*GY7^~ePESUj
zQe8(Or#^M;M*idx3hEuWsrqEG5Lrj|hQWF5$dxn@6ys*PR$B!=;cJOm{8_-aj)5E(
zu1*HF4|HVq9I@MtVt1EQg=N9|A8IbTtV2q25x+i^q&gxlDzODu%57}(!XnQIjn!h`
z{zw|&8++_cfa0X;o*jg-vGvHvNq5CrOlmQ&Kt@|j?sOj)40v=ix+uPPZlPc;b^H?a
zIr0w2ep1xAv0S_V-ahIqdABRlCka%U{Xs4Fn?c+T34;%pMd&}vJ{F**lI0YYU^mEZ
z!q%H<^R$4*>JWwI1-<p_e_++5-McE+0Di%QAwjhkmauEsj>I*;<Of@*64@W?*J2Qd
z;7GzXXyJSM;+znT_Qk{!dAwQH<Tod`h<oi<674{%q;wpUlqIxVvWkUxZKt%3beSA=
zxth-43G7IF+wg{EG}|_V7IEz;J>@3WywBAyd!K!fvd)7g@@R}>%e9-Ybu!W(*iOa`
zx*ASX(hBAarPbe>HuD}gh>UN5j2;&e@ANtam#?g-IH`0QM66*#H_9EfT4)qLRk^&`
z^t~Qh-D!!MDVM9BJ}8Qg+Lx0f1QURJsPLf3iKlM}zy(+VRCY020&@c3&rQcB++5mQ
z2(E)yXOW?~`pa1gK;WnQZUIA7MSobk-{LjnQ*HcB(4=Ge_n`TIme&6}$>ZB5`yWcG
zzIz4{BrX5tJE>eAuT+a+MVkSFq;}yYR(~;I!VQAVX!r=6>64?_uKiSPIM2=Ig&lBP
zZn`L1B1y7zyqHgtq!gi$KhCo|SoM-iN`9vrfB#M?T{XooxpkD8T{+E8ByYY?Bq_vl
zWnm5)n{?9vW@mmzBvWX@a~Agc(&8_Vje+USvUsZ@Jz4ZRinMYz+C`i!UWLq-?9a@x
z`Y2v~5H<H6MHBWtjiWEV&)#pj>Dd%LVu1`VNg>OHU~yInhIW1A3Y-vyY6%mJzxV<h
z8aVj^E3UAnik&XBC@&(JZ<v2iob_{DjA6Q`iI`<)=qvH4LquHN3zn+PsoRpdtez26
z+*%O}A@y<3HMSkDpQ}H(EQuX{gYeTj@Uq`DG={9{N>?UG5gWFiFth=4mp*<mOq)$u
z(KPF>wglCY{;Z4=9WQDzW15xff#)5TV%5WfC!Z<&8GF!t_J+B%7%=W_h8Py{SRg(4
z@kK<kAmv3w)-cqeI4<~seI}kAwo^CW{=p3)aOvs}UuJ<&4LH9+<#<`5zKs2oo~4L?
zRc<tI-tCwmlpNFf!ze?L%nb1nHLpF^*A{*cw353A?<|nkv|pjgz)TTHj^FdmPF~;I
z^G~AOkcDfuo?6szF2D-;JL$r+k-Ulq)Jl;7*0h_$=TUb4TC4<0Ku8rs9@$<Cs%m0l
zHCT!LR{y2eHl#Q`<{x*IwV=<X?0{0{<ygVquUc41b$QRF<Su|&>!KZhDeLdgc6hf-
z7E=D|6<o{gc)yN}yu2S@<9#~0yq?{DxIA6!z8%VKXMVcAE}WjPFnu=^b~bz7QgvL$
zfbpUaQZPB0M1IUN{~*bLn|x*9_DGi5DPF+)q03#oaChH<xAS?&Bbefy)BU^;_x16G
zV;5i)lxU}H#Ou!Fj@E-4>hM!jU*1s^jSC;Br^5oR=PQs=kZ9~?(@`4Dnip-tn{R{R
zAYJVdSR>JTkr~d*1<qhidO(0*^NJ9hzBg<lazZ1*<PT#X0yqsS30I3UT0I`L$^@`W
zK&|m=rJGp0X3Tp1k~>{QIY`=SknmER^g0&wmX9cf(cESr6(<m7X2vZR^tvwEW-_*6
z9B6g*WTS}x(@*~5v=A+El<8%7Q+an4j2JyY!K}7wZB^ro8Z^P%Q`IV8Bdghjn#p4T
ziW@4)SJJG{*Onf^mOUPI_vib8@4Rq6Vk9I6`7-G3C)FX>N6^Lw#()7aYPziJ<^7+J
zgAUv4KE^B^9-lawnk%2z)a`F$Y~8=O@|NFRdC&LV?A4p64`^O4Z}+dO&w(|=4Boei
z50{;ukF(nnrXZkWM4BwTz7*S?18djHo9m-N@>zH_SNFbYQ;xo4ep>om3_BPDhRibK
zTzWJ56M78#e}l#UJ3XE{_CMBT<hzM3<Dy;)$z-f!DcX}Ewf-eZ$5V@)?|G8&Gd|p2
ziqfiKzfG{N!cxv9J|&8G{(kal$`&6T*iLBlA)lS_?q`T4h@xH>o)H+Wi~(wE@y8i4
zRCe1BBQT$$0B7KHB2<}BZR=dI01vxF*vcRHd_V$3GcbTNzJPF=BXepEejf3?B}67k
z48I<X2JL8j6=5I@EI3042t&`>ciCfv5tcZ(mYO&XWJ_>O^;-?s?dg?W<ZfYF*2Lqg
zPK7%Bh;wtNx{hK&D#aR_4(du&0~PbKqx&Iov!f8Wpn3a@cj7<<5xEmoByC_)*4QeJ
z2X|{DR{7n_fP7{bG;e0O5myHfSY?G@?!HLr$FnK&=oGxbAXx-4xCVT!NZFl92%v5|
zcFtZ;%6(0Tn(@oe9TSP4Jw@r4$Prf-pj^`I#db|_n{ba~`|c>R7sFHg8hE8Be<oIK
z`vzgt<U>QxeD>DX)UD1H+}`0|4en+#$@=v-+;{#?mUX~JQf72Zb|wH!5gC@pvBPOU
zjntXS+Az5>$&P()4<oMRNBAHN^?x&HEI@JCo85|og8HGeArF#VCHJIT%{ixmRv~*t
zC+L3O6BV1C=z}=FK$Z3}9=n}9x@O5d;cL)KUrHZ~p|x%Cuit^%zI8rd@&8&!eE)k@
z|4*3AUkCq<f$<j+V>lZVY7}U$cDm7#sHSTl^LNq*hwfdSGgnPutkz{F<<oo_HoP^6
zLMw-ALcT}58nK@T(v@e2n82i0RlufAf08kyXkf=ez}+$d;?JPVQ+F6hAuU)9(p`mz
z%xtHn*drF>smv=35X5~r3=yFNFZcF9z0vc5{MJF$E4#XnLx7Y8S&>lv3ded=YHBhB
ztI`%KkdUeifG2D#Va#l}R+}Le_gyd#d2k>HUi1k3<_2KaT>zK;hOSkaqHSOQ<cs91
zetk-clzQF}5{e$X)yVOup8Q>;-wt>~{eCdFH+3F;B5g}E_j!p_<ssYfTx2l>=x|mf
zPxJ8YcxVG-QFS9925c=9=`1$A{A{dClp$4{gV|7s!x@wZJI?gmREgG2zkQ&Qvhnl2
zzj`0dHeVE`93!47&Yc~vI{TB|qGSf*1XAGWbZlZPl@cKKjcqr;m5s)5(?pmyw%u5Z
z)4o1Wp5Q{VBUqCDnQgjd%FSM1p)hPh)fP8lV#NY_`uTGJbhj;z4Qt~Pti6YLcrq#w
z2f<)B#}LjmG_M1D0-fCIx2dx+{tb>E8Q)-r?cM}a-PFZoD}NYC{x(^il}MMuBQ1;D
z;<<8E(Q+f^{aN;@!1GP5l0Rp@0X$cb*0WUDYdGH0)5SzNkWb-MZ--kEC^|AP$<21s
zAR$C%c%CQXpvmTz?QAZKp;mh{q1!ICB^DsK;Y>dEA%iBE>1xC%ylSIWNEa)CymjVQ
zir!xJAcklIo2Vzh0porLEaYao)2Db}e}9H~CEk=lSSt-xGhwy1`l3qx7oz`&){-Ta
z-BtL?af?B9mNEUr{3g2x6x;ZG#*kX`$2AnZU3!!yWIe$yxh091^_wNALr{+zt;QZA
z!h%MZ_}eEH;dX27YB;O2@!looRzWw(?hQ@)QJ0P;;7B)BKG}seH1SHl-N4-${Svkn
z#6Sc2%7j64=2C%rZ;Sb*yRFTNz#-UA5S`{(W#&DTT?zMk{OlPtgjefC!nXJNj?{^T
z_6SPFmurQ>FY4L0#v=L8c`_gK#2%(`U92GHs1^;B<^S0EwrkkV$^8Fm@4=p)N&f%S
z9s_z&0BPp?*8)u|keiezJ0e4IKnZWVSm)_7qT!Ob8x4h6-q{dp$#0yInehQQ`T{Nd
zn+%nK`ETGG>`ZL`k$?UZ0h`8m5uFdg>zMM9KCa)y^m=1NLb|YAp(wefQGkp9Whod!
zYD4Vfxtdm--gQMa6giJ;6~u%4crr8XmI`DFfqw-!9(N}OEJYOnh8!jw5NnHwpf(Q(
zau%bNzkt)x;V}y0kEJ^{wd(m@9`WxBdhT~=_=>}8`>S(1hqx@M{H6Dqm!~;Vkl|{|
zxmmp-r}m4V9W<R2%c(N{&+QYGKQzB5BP;cBHfD&sJ_);f*KNLP5zJ5xN2hzV;Af{|
zp?--ZjnPOFT<v2+MW>BoEBMbuP_xg(UhGevIOqWfk%uPllFj0<1%*upK(q2Q;n5~n
z*dayQ`cMU&z8)X2t#WN{m53a^h+m+yzZtqedTJ4X;MmaOPoo3HYr-jUr!(D-_%`bo
zscXl6gxxZDJ{Wd32Tz^%KcO*qYD{90w`K-XkB9Ra3LqQl<rsfa1JJ{&QM_5J$p9NY
zQTRD|RsdPjO4=4-y=T+Ls^$Hn%-4L|lP9cN0;3nN`eyBO@zb}UwfL{Z?U0RN<!X}i
zP`6{-WVr><PxdV`VZ9>(_t70}R@LRAFWsJdqRB8M`?fPyCa03AGeVCc^{mhSRDmdx
zu#_ia1gq$e)Ks)-55lZa3Nut1UkR#rfe6x_8z#ITb#s`%6QB9X-Gj=_TUp#9+|{3)
z-mtNpa6^!~A;Ru<!I3VsVAmc>fmJ549{O=nXKGTUBH)tYKIwh-f+J%JF3H`_T2?|a
zDMC=8%k^;LoyW9Ql5L1TS>gWFTJg)8&WulMy+q+ksjX!oc57UVjWaSOg&oJFoD?f0
z#P#&KruvU6O#L+g^m;^z9JK|VX94q?Rb%s#H2bq1oK0y{a=nVW&CHM%F+2^bB_)+G
zFs=xl+!8egKON#CV$G6WA)=b`(O6@DVwnh|ECu9p<P{9J<jz*O=8%TB`!MUKgZgl)
zCBTZ&U=>z$>&o2Gbo24S=+f5l&`Np7N~M6yHS-@g6dxE%VbkSTUGXz(>>*4}c1^A!
zUGoit=oVNqNj4%JhkVL-m7)Pga6{>*1ZKs4jUa*cE!1mdMN89Tj3DI+lc1%;yX^dR
z`*xJ(W&4_$0Gl1(U!Ls0#0qSJeoLL&8LB$?7}0r5Th63!we=?vQaAz8YzFFEn6>o!
ztqb!vQBiFOwONec<~11d*(xZ)x2?&obId)f8jWkJoi8t_ya?p{5%#$o@&US-U*^I5
z(jr_dLRZ7UhK$rVDHG4`b@jIW8Zf(oFSNQv|GOx_{x>pKdb)pZPxdbYuE<|Kf}0b7
zWfRk$mJuLn-yiN~P7c8Sb>`{-<dB40gM2=oqQu6xmM_K<O>w>8NK9><*O#4c4U}h9
zTxVY0`9lvYhf5=}0GkUSOyIUJyW-+h3E3cmd}_9C*IPPY(j*^0J0)A*eh&!9yMJWr
zeooQ8(S`o$m|Bp&HG7-l^}aQJ=>SI)Ro?DLqL$(@^YmneoBw$A@cvqHM+IlYegC5m
zc)`XJeRG?@CS=Y$l@Brw+2D&`UuOeXgHQ2Jq~QJr+`j)gGo!kF=`Zd+6-lNu(YF$J
z5XeS$B9VBIr@GDKTX{e$mIbsyCeVZxmcAm#sHeObg1XYw6gJ<KQz-kiNBW@X)1!Vk
zr-V5nHwIM>tef>>w_nux9F1dsqYyb~Bmx;&CCWh5an}by2!jVsn+B{q4>pTXpl9M6
zY_>4zVye($SYcXAw38(n@?wgSqdG2Nf($Q`JPA-`y~IY+yGW^8MXDAAc!Jht(|p_B
zrN&~M4suJQQBdAMp1>qFfKkLl#RtZXTEwKw$j7^0y=*R<Y7$Mic0*T1B*nWv1n*T;
zllVOihR3zzJ%()BCBDH`P+6pm@bo)*0)SZ~2`5F(D^N_jRAKA<NK_UWqs~-+uv1!O
zIL#3FTskV97mAl{lXcY6Pnv9^)^CK22nNu=iP+c43-bvXaHs_<qIQsg$^A#mOG)HI
zBiyD59y_X|&ShbYzB~J6?>WQaWclfCe<w3d@i~bgdyh^p1tDF`&lAvEqLoCv2cU)f
z`iShhWKZZP9%o!{BoTO5WUKfs7D!Qu(#=LsT9>2}(#Gx_fc%HnWyy#iszrQ>%&4&$
zc85N=%Q=(!?@BROvg3%a7eL%7O*N5oJso(8a&j=jF&*YsdgoKNr)yXCUS;L;BVujk
zt12S-9e|mscu@(GihqzUZ^NBb5G0#^qSW^devK=1B-h3{nCncbruf3mdOlg$sD4s}
z(;`w%qFB11my$a9p%zdYABe{-0bvMrjO2a~QJ$A&v1P$1{g$37veMfE$H2KBvihAl
zt+Nfy?McT?)N%AGZe;qIng&sSP3EHove)p9+et|YVOaWo6x2KoY?lX#ZlzN`o0K|l
zt-@CR*tn<LO3X?Z)b0b!ereQd@*5z;Ay{!uo)fc}j6TyX(;ldJ6_okh=c5J5G=sgK
zG|Ak-XvJ;}_Gm&Z|0XB%SAJybAbLZlTD6wvgYAcw_sjhP9qkze5zHOz8(>c(-aFGI
zd9b;zZR!y<W-)h*etH8IDL!wy45L(UV%^yYgA{;ZV7f^vAM702hCeOKCS6v>BY(z(
zeTklOd!{JQA1fYFQ?^NrWuZ<>0~wILQMNh(rJ<SwB&s9$pEz-*$a<iao^Km}&Um?L
zp0@uk;L!hb#O=Q^Cz$E}F<WQqzcVM|62<<DIYI1gX8Avu6P%mUSgt?qOu?8VHK(eM
z7Fbb8yoirmS;R@~t*MnwZ107%yYj@x-^UdUZ->RYu(7&`DSt<$gSptsl^?$2mPizB
z=8?d$cK~YL09R!L9Nb_OP{7)%moJ<4#GPSm%YOTUZOd8`o=ppqBf?*oWR}VgU{L67
zcxqXPIZk@J`tvt{`p%cPbVf{h&=da{qMZ{q^7f;d@|mR3*n$hjBjpL+TMH3XwiLIp
zpI%OgWy|i9szjcNZ)A9jBF@6a3=~?56P0ENCptjXNK?Vi#Dejo5AGdaSa*FVbs!ay
zt(q$H?MqJ%^KYa3b~zQZ?$v`aEYZPDn=MKpLN+P6d(o8XLTwk6r1Rp53lu$BH|@!8
zpYq6Px$%A0mK?W%nB<#3Y9>q->evV`CtkX1IbogNaefZK7%ok>N>3;!F>E1Tijx|A
z(^sf`0YO+Y(!dQOOAB)#kYVvu2qlhk|7<SeKepx6n%Y$bX~C#^b%-spHvtvJGF8!<
z$5^`Fh039t0$kf_$nFX8LCbC#Kx-Nm3K6$7cd%QnL3;)eF?cp(PZ+Q=IQr=loU)(8
zn~uUHzL<eg^~clkS-8ry15#u_;AR7)pb!4o8Uq%N-uAp(OJm5)w@m36WjBrDd1iwS
z%O+JZaIQ3M!B5f9Z4wT)ATEpZc^w-W&@%s5mUF6P$28nPhU;pF>CBZ(&F9B~7l+r&
z&CRo;%j@F-Z}!ncPLI#?%GPPc$kWbNkB@dOYa(k2Z}z9->#I}NH|Ey%YKOaMrQjII
zs4m;%)|9rb_uQy4GRe<IxA+@5ak}C7I&e@i;^?03{dBP5vu(;XalL@9ySexEB?DB!
zGf=8#%}uESsiH%(Y6jASSJ<L<kE<4HGEC#7Kj)72$pm>($r+1375s4`;P3InlP$vF
z|Dx=igDmT|EuXfnN@rHuww+mN+qP|1+O}=mwpD4{?)+ZAzVG(CUv%FWYoCY}`<#DH
z#Euzru06(Y&;@C!DF%_hQll9Nq7@oGD)5UM^J<qF4k-^CW1d}_qE#AJ98yLYyNyt5
z+}8m^Q(QxL*5$-L;KK>tZ4^mPb_wcxOm@E7DKktJV-q`a!cnMct}{(REdu;P$naRO
zNb;U1V%e9iMZ^3F0GF%zk$|AN2hy_>YZv2{Hn<W8f)@1WeK0laW2O%(XvS}ZP_++k
z*Y{&Bx~A9d4}Os0`0i~LLU1X#T}&)wTX~xapPc@MiMe-h4~@G41D1_go*6D1OD<xa
zz&RhRu|sRc`7(Gc!sv`t4FFFrIR#6_&bsmN{c)tU{`h1ZIr>OB#SHvPRt=Z^4GEBp
ztV{M`jum+FpQQG7fff)LVrPU9IuQQ81UqJse!&zt!rF7h(33h*xmt0)CF7A{Y~0$y
zin>-X{F~mvLt{UPK_?E<e7pfY`oOl(yoet#4i5c4Mc22WRxr3oVJUqPWFiB|s6>1_
zE<$<wAY_{<>?gSRY7H$Q+-}r{CKwF`k3-oI1iEpBfYKXWe+Q(5aC;7o_R?4{>t}IO
z5nLh&(<G_Fx;5QH3>6t(i$uE)#IwOJBQRJX{)&fX$1mvNC#aam{oPU=qaeU%&dMdw
zoB?AwR*(fPpey(7T;o=#!E&$~)2kH}9ym~Ir~!Tno{ee*h7Y)TFo#fiH69*Gv6YCv
z1=GolL-hXhijM?Q9y|xbZ6bgwukZS4LVb^u?g8=?>vM<zF<d{w545|BAS+w6oakrD
z^+W(3t9o8Du#GNQF}S^MkN-zcn6B@7XO&;mV=Kz8XbiS^U)rSqmi<Z{A+(<aWckyL
z6THV`C08-9>%u&w5eMMu;({+<pW)s-M82-U83A%&Gb@WEh@@i$u=$>gOKtEd@`wL#
zXWdx}v;r-)`?SUpeyaD>P85NG6ojDxz9}~bAbx~JZ@P%^+47adL3IxQ-!8g<i6=pM
zy0vLgE5m4#5T1IqNI+M+8H^hR5jO5(QjX)4O`EOL#Q2v^?A)YbyUTlMVbsy~97nSR
zj|A6ok~l?OSRb)gDV-#t9JFlwXG9AUUGdYWdlJu2IyhNx{V1TftS9%&-5)=OZOvyc
zz9q=dJJOEsH_z8hv)Zm+lj9ckZ_u>)!t3R6(x)ALR>gn8FT6b9v%aCtFA<LJPG9p1
z?swy&K9B+~#npb_66E_{3;p;|-kp%)=&S8bAGYPu!>ey{^X8LHxBEM}bUsAiXVIl2
z?RMAjruGcgQPg}69^G|CDo3^&gBBJJY}ioYujpWr{yh35utd`ZQxFEw|7(FW{L88X
z`PcQIL80rIFOwpra60$y;*5$q_bC~RO#MX>pY@XE8gV6k3Pj=x5d1z?z}YF>w2jJ*
z0gqWmJcJ|BFRHM&E(EngE9giTh>u0YXvNnM9%vAK<V%F-(wPFF^Y{=X&;y=#glnGg
zrqp$TB`aCLft0-@{{8B*&Ez3sV|Dlq_fPyCeWeJH{v|%ga^CDYV8&59I2r<3y1oxH
z_8JN;nh#G>2rH;gt*3BfXMR7dQhn1F5wRIV&9!Uz42#ciQM1(VP0=##Ybhax+?(N}
zx!L$ASZFaoAS=)-J!4$qs;>zd-`W)-R4SBHU222Sr6(*l^&%E~tkKc5X|9`l(9p_*
z4!u;F&a>^c2+D=GtUiElU|wDwS(Hqi<`2q4l0)mPYiBMx9g<ij+g*>$-QpLTvSZ2b
zs@j8w(=EXruJz?RX<`<i?cw3HwJR?A^q9Lg=7`?^u&A`t)N>K7cn$Lfey3bf{4j*9
zA~LXpnAq;H?Ulu(VXEZUY6uYLyF#bc1PN2x>;{KC{tTAA63|zAvJv;U+`lCFMej;V
zEWNYw&VAXKjLGUGeVbhHj%rYIqe6)pzRhLRp}e$~6ohX7zIZGA8gD{!!K#?Z!aYF9
ziAj<<B2!iVCCwT^!gzcTZ@moD1%1VHoa1>D1RB`%HP{;jids#%p+DR7OZIglWqKek
zR}+0}$whaTY?`+N0?ne@#d?hJF`~qH!21Tf4AY~g=a}U+Q8*K|xnI@Onb~06MX0DM
z_eEI`Q#+_y8&TV<)D{mZ&|pIW6*{aD{k3V68qE(mjum98ZGSU=s4;=@`<WQW<dhmF
zWKCXGtu`L1F2k(+%2s3)`Ep`57&%g5R3|>sApYa=kE`I(V`Gu1ar++^u-f0Oe<hKx
z{-12yJ(GjbFMz&pG|J!fjf{-!f8)1g`%kC$zjIurD*bDn?=R%yGX#>F^P`7Hg<_1m
zP^%nDY7B^^9TG{fz#`tZT?T~Du=Dw!GE{JwI+LSLH^w*TQbv~&2hN<gZ5`2oGi<Pl
z+3y+Egvp`Bo)`nDk0bx*(WgFpW;9;s!PwKrU`ix<`HO3Ph7Ql~E}d<F>y>2PFrzt4
z^}L`ko=<*~O{vYnNBa`8&J?gqrF)dOqu15`<v-&G2pFGk=s{?DhUI1`n7*n~uIP6Z
zo`8sd;A7`j1H~SvU7CT9yhn&Tb@`sEh56>mLVEmC6r0RSEPR|{ia&Zzr~|n3jPB1|
zz{<Ylf$&tq1<Hlwcgo!oO#Aqj_5#=JWeP|YOF?RwNN7>n=r`^3c@yJ3Q<*M<<@ZS@
z`Iuvs8#mx!0vO1uKK4rw+S(%?9Hy9D2QclL*3#$AyHgZ(2c5^mPY5W#*vQ>3>P@|@
zA9H=~!zA2ZrX`XRq29tKH-AOx)2=F{;DhCCW&sd{9%2?)|LK$&-m*WW?)=jUw!jwZ
z?`{`Bxm1phLzzr)VqVGO4ZA3#{OL~b6%l?@4g^sdCi`$+#U>9`4%tk|o`nFq;JahZ
za{2brAJUOm+zjPgL9mpf?pq)QC~?TP5i;S}e|q%n!~v<Cj<qV+2iuUYg~A}Mw>iP|
zGiTo{wJ@j$vJvw~&F3j!rV984X>-Ji4eEfg`X^jU=AZ^Tcv?vNf_0_UA20=<HkdLi
ztn-nUNja^A^_Hy4-&Vd?PElDJHkZYf{uSpG7FlB{RaX>53?0Qco;!G@$!59Q87!45
z34|QmYfj^o=~rObHr-)eT7|{aTdydJR9kk@qnssDvyvK!5JsFyJ$egB!R-wNfjjm@
zyvq0cKQsvXZ6ON3r#Uap_MqCjf#P7v(GLNGPv_~8rJB%DF08$C7W_DDO%YSkKV&L#
zKT|74GU!@hW$kp~I811wTKPPhI4^P&iCSRNga}Klhf*3wnyh+KE`HbET2sfUVFWrm
zo$?Mr51Yj1l4O~6hph06@r4Bi3*`Ae4&Ve1U^Qxw|6cus3{ydqH&Az{k8d#)qFWS`
z(4@<AL?&N_W9M|DSlx;scTgQWsTUB9xrR+tTBPQ+#r2Nc#YA&jABAzEi&`-|in=~*
z8dxlM);f7%6>_{6S~XUHiSYAYx}O_dpVHz|{YL4FO!&zODSV>4&9O9qL1Cw`%wy^1
zLzQ|sU7qZCLvE&5BXgTk3giWw#zY39Xy;RS$s9TDCS0|KeQ^n{1;>S7H8^MM(@-=#
z*TZAx-V-yjjbG!8=E1j$%{^E#Jp(holV$&1srWWt{#T01KXm!@EdQ>w;xAqP|4wx}
zUIAhmw~^zfsiCeJ?rXLrn5oNvnhi;gO9xrO>rN0FMiN$2J~wRMBm^gCFk!pf;<6Vz
z-gKLJ^90V$eubwyEPeI4`#iXtv?))A&nCO^M0iGUEB}t1#mBhCKmwTU(Xi@RcS(tP
z(oyl@@_4(3%v}R}DX3M;0{}_!7zcP(Y+&&b0P7x>Q!06qgxhH^e!Ovc8!;@58-Dz-
zo8@WGH)?w$MSm#B>UVM3=cp>^=TF@*k{|ZvVUf|bu(_41l{%YNJ{<wapyzJFY@j1=
z_q(%BPU$r>4D~KH1!E|4g=p`_oiWpAS)l^&Ng2SJ0aJCv=Ww?JCa!`<;K9rJgXH^q
zP>)ldi3dpd+ZLVh5P0NG=qPB+5UWV8VYNx_Ue5~2B%rQ5fF?5s*JtE0qL&pKQuIIz
z5sqpqL{QaMDj%_f0&AHT?z|f#a`OutR+8dv^A}knI(o?1l{VOr(3E80LkMlvI|4Yy
z+*uW9sI9N9E~0i2!^G1nhccu6`4F!FB4W?aJ2C+U7)agaGal$Yrm5Wy3X?$E#JFG0
zIvN7X#un#?3HymeB!0eE@oMDscJ89AG#3vf6m?oyyuqDIHXM8)kUfA|;2itNA7yst
z0j{BTGX@!mKbQFVzQ*jZiVf_hB5xLJ&#$@TO=x`om@1ok<~^~{fJdx)+Jl16JXK?5
z*twNmc@ZsCpo$+0%iKTx)<+JpC#ts>;jG;}l1oH!OAmkibKDlsRf^P!nV49yJ2?87
z%i1MT7tp>{=c3wFxu?zD!L$bmprz<>NXisq4KM(kAT~S<sMjaxZc#Y#q`b;>IkIw^
z&up|JxB$)rqpl$IV#XmhgTa0RQBL@ACC^}gX+7jfzB(eYCQ6H1>|_r;+&j*!SvBt5
zaxfD*1Vt1d?c|Z;@Y@C1i9W--L5vO<Xj)UFc;pD?x?g#Xz=ylM_KP)frPbFw8y5)$
z1jsGH;+U?pOCao$7XmVhRh1DXoE;<qA(|i1^x_36%sfd3DnTcaoF$~RP%;C}LxICd
zE37~UHR|Mka#5T{vA8ivVM)sJ30*~(K4m(uC1o&YSigmOAwV(RZxvDB!d&pOWM{zB
z0&_E3`grW#b=Bn1i5;W0`RtN@N5`X8u5PW|b-=!7NN=d>z_zHZW76oUCU3Lxi;F^Y
zcu^4NNbjCH>e3|$S34_1uzSE<rJ$~=L$;giK)8s~cqipxwS;Q!s<sZIo{$kaH5Y9N
zSW6~hzk~}DDC0KuHiJza9AYMCTW@*X-m&_!cxj(;fotOuPCzSaw6**>S50&q&Z<;g
zFZ+<t)re;0MKhEK{pdWFck}dhj$2-{P87_hR<w;nl>=XUvl-Pv&wxoZa5ouZldz)o
zd9mbkhS2a%xMefR4%PUrLBiQAulccLWL#pDxY_gMn8`|0h2MjybeiK;NCE=CO&&`p
za4u`|syy8X08S2H=x?gkx4rzIs@4BnDfUkXVvJ1xHhk+ZrPvQvB(Lmm2<s-g)?zYG
z4@exw6^&u<H-mNQx%W&nsYZ|V_i)U|6MLccLnfcShkly}EN--Wo6U)1sRV~p0i`9D
zJVFzi7)oRY(9tO@RN4wQj#M2h@%~0fZZ7Ffa1iMrHoGjvekD?9{8p$2LF#2+%j2&k
z6tbT_aPSHz+-Pu$^)9l+vKAlgu)pCw#(~=98tipbFOzymA-xiy8d^2X0D0j$c45=l
zZo{MGn>9ZeULuq+61Qvyeb%T|p!GNH2m1@3H0<&Z{dvrTh&T?{{Nw)US(6{{7lE;w
zHE*z$IE_pDA8uq*@=M<Fq=8j>tAIRWu~Rsj9Y2eJfJvnb3W_GNdq8{BDtD6fQg8@J
z3d)=-#|23(WsehZGBrqyjGb3v)up<X>&=)8<0S}G?N1{HZA@en!I8Y2L5q_~ZDE^{
z>2G1Hk)2v2^-!mVH%pKW#x<LhIkl}Lz$eUt7t4xoLMfDNUI6haeRu_tD4t~xQZ8(S
z_){;eQSTL6q+ce@%sDm(56y1|%Q6b!#eLh2)yNt)=nYylPHX~uEu7lQLq`{MFbqV(
zUcH1tig+Gz(-Uw?LWj-g77X%3>h$-85%WvYnsXrJQ;yLP3WZL!$lCDXMVV$n$=N-F
zKp8Hk^PtBHc_U~_NJYy>z|B7ZJ80($@pU#1229^{X0bh<O!HWmtN}|UMt+3^<O+Ak
zgWd<5Ucl$D2FAJQWqB6B6&E*|;A&I5B@uCEbGNg1er``6*}mRBK3?t|-k<Ju<Gntx
z@@{=T9o{c)e!w5Su|hfsx_i97o!}K$Z_9SKuzP*KazWUVJqX41Cy4;96{j8sQNc7D
zh}mhS?XNf4xu?kZbbflJ$;b-net*B@J%ZT&V)nclg#EPgCCq0XNLs?-9xPc<rT=AC
zUsACwI}%K$nCDui5jF~r-p$QI2PuT#7Tnh)B&XkdpDUl|24WPBDOa3Eg!Dy*B5p7g
zm#dg>IGu*&MZSYnx+bR=Yi`MN!wL^RjVh9&%O~$g35ShPRr?DLLSj%#WGt3`TYRul
z6I!O|hb)Gtf1sAEVK>dP?6otQr*q0u{&6|>;ap0FmRLmDg)Tx^o-Z~P{&jjh4oIG9
zMan3@b6!0DfB-01P14G_x?!E=C4>msLesNfq&@*V84WlX(uBM-pr7KR=C}799?ZDt
z*wGLd#(dYnThX7w9>}r2+9(I#u;oz+1X2EZEJ4ljP0e3%$6Lb~@;asDAdOOf6%e!^
z^@YK}zd`7_-GXfGFg8p_Tm5$rF~X0T-v;#iNVPFE7*MKFbNmRhJcA%CmWH+d^Zq)6
zV7r#*l%AXrpiuE6%0ESt-5)~W?2j@fmIeIh;K;)L#IF5;oFJ;ICzXZPqeB@xpwxLq
z3^R8SuY3vF>*JxJ1Zu__K?qiHgA&j>J+a9$H03zQ(8&ska<}-z(bJOyEfE);j?~kl
zu@Ee3orAzgFk6}wB6bfLn^mKrCNY6oP1c-yW8@X<1CRqF4U!~f=_Cjy9c@kh>vE|&
zh#{br896xkAv)WM>vO~~a_t@c*;>gvh({=zS4sHA80PH5Ca0WEdm#n1{hK8H5tLQX
zI9)GkN`@TN^!rq5ooD3RsPG1z@}SE{{?xx3!b~Ho!>|@zTd{dPeRR_ufc3|@zTFgO
zZ$*V@Z*1~7TF{OxlS|L<(Q5W$TeYyJ$anV*%g|n3ZFO@1o3M3yeiCMAXZyH8*!aAi
z(gGeI-ZNKZ+u-rE7_YqFeFo!Q@@#eU`lKujem47L`8?ljs<?CqUZS9?IC;UjzFgvF
ziwH=`T9bx&^9^fZICE*CXJIp8abYoFabU4vxiO8IF3e&inW+}(r3M>e{kKhs5pI&3
z>Y-lh|FtnvP0~{fHo{E$-i+R^M^ePc=?58PIJfnPq26IGtvR(ijBlit0}20-jZCiV
zQ8VRf<B}(~l{o+c!mO}I02^apE1+n*mtIs8;dj>aM`w3tLW{Ly(^oQ-a`HfE9rO$f
zrD3PQ`^~SAU|*V&iK0LW9$?$odlB0LZfq?Ii6OYqhjpZ35K*!yPgROjn71HP)#EM)
z|I9Rt5CaOWo9pUSw&uttvF(Y0lz#-GoTd~O+jI|X3=BAEk2|%WhE}CDqPn-!mW$^}
zTjB;m3<I?!HdDo;;p_7JfEQ3B-5k8uz|FD@1i~<uLaXxqDv`hEqD$~tviNGRy(Re`
z()_Fl|8NK6VFV*;(SP6;u3wKVSkpTz1ULvOI3EW@A10)&CW*FRFMNlMf{s`1`r!5p
z-O~;HWSvBNDuT#U=Zzvaw!tgX90W_rtyFFqb(>q(l)5%29E)t}>E0;BT?<yKOiQTo
zf!k|U@J85irchP;sI|R?&`qDqevo1|@kc>K>Isee6~#R2D!;kX?W=(tyGrUMO;$?j
z<!-oA8Kp-_2II2U{h(46Wp>amB`3shDP2@2$cTr|bM8aw?FkmQ0=lY8<12c#n3u2$
z=ul?!dPzkD?3FDN!3<>-_BB+pAPJBCa3$^9VMF>WV)$Pbm&9MrWwx!0@Kro=)?QR)
zpS91gqxKe`KU$f6+BQx_d45jHhE$aZsx0JEp3a~;ktMZxnv*4Qph*4*FN7VO3pT<|
zvQll*Tm7%i0T<Pg9y|QFo9a}*{gxthCj2B!5*uEc8<6<MCrEtu+x^{s#ADMGq6@&z
zF!zOh$ps{e-J75lqo5{T1*6a>OoHt9XXXdRC~tM(CE)Ha%G~s&`axEAuZY;Zt6Nzs
z`WF{MWq6QlI10m?{%?)|!Lp2oHHF~R(OX8Ie;m!xn=*Ns6}h1CC3uo%G_`6bB#3Vf
zNF{WgndD?ClhpJA>dK)!bV+ITVGg0U7fSd}L@qSEPQ^*zdY<k~-|aC#MazcVazt>)
z2`f4uxYni^lEt{zyTxTVvy;05Lj6xPCwEtU>C7!85V6QttSWX6hX<n9530*>jPRr8
zugd{X4uj9rfw7(6(ul63dGh{xc_2!s%n^zj+lO0dpiANB+u5np>`;00ujm)bQ|vqf
zH<Utw!@+;knlUl_-5|-p_-}V!2WjFq-$ClWS1+g>#=i9Z(wgq+0FwaRXqV(aFv2gS
zm>-Zv?9HE+lhqlB<O;;b?1#i-1Ab_@yKifZiuNB{faFiB)OdGhYHw~32ob^|{3J``
z5wU|Ij>^Y?^ey>a6mk6d`U|7qGPS}Pj})fwY;X9q@n-h%#;$pyJ-hm;Tyrzh`{ml`
zRU@z<ZZa<nBqLnH(W&zF_Am;~TntW<nGT3PPGa=7EP4=C!r}pZmu(E{;_Rv0rz`yO
zCtL7xXzyzs)a4-@(FT4x9SDRCcc;9EvS}($<(v`AThc@`yjfkTqIr6eB5xH)hsL;A
zy3u3$i*x;WJ+Tv(C`P0MMnBsvVV`k*#z*6%iFgvpFn~cpddKei!pJu7b9-JUe#IyG
z;Fiye9x5<d9|}aEzZmE#xNriH%v!KFzfxKkl(k)0m~t*34+Awwf-aF&(}&4pN5;{*
zrKhHrGd2pnF99Zc0-bLSoqrbe926O7jKJ^lV)aDDHVe$Qy9gCCedtLMaX@58Lkbe)
zF`7gd(%{%s<X-}O10zac<f$=Ez3=f;;!EiU{(vK*RV1>bfi2^X0^V+g$jE#(b{|9_
ze*`RmevMPfL2ZEePBfntFW%B={B%pgr6^i&GdA5;bAK;h13Khn610UQgD$6p7)~Qu
z(?uQyj9rjef~b=p4#9Io##R_6#xiswvgB6qNg;qLA>8x3UO=5gktJIr&>HTgQlkR0
zZ(vdNKwWvgUV(W`wt?A<we7J}bt!=R**((Wx?18m9x%KWRK{qF3Bm>^D398ZEzb9#
z9ZbMXv<g3p8^w!RI)aK<;aq@qS(kvxPLq~D!DLWL>*h#>h~#_kxfRDRmYFqEM2`x^
zna1aP*TWasCr679Bn3sO!T3kUR>jA@v#Q-a2Fu;OeY}f~v$<XBtU^#h_>QmzBWzI5
zqQbk7!f&bYoPS&W$0i4y>VmzftRj+Af2#fR3>>9aWh%R&XYX33rq3dv>bL63=rlLw
z@l5CPRG@*@<wQ?>1n1B@)(k5LlW?&Pky^!HZ(LXE6}4RW@tBzlYRXD<wYYlltclwD
z`45nR=w7u@8+G{=Q`F}jagK9gkU1G465t{mSu(&NRa(_1MoAZHxh4`wxnZ37uby&`
z%%~dpz;#FEI=N!Q-!B8OCqt1&^z1p!I%`n%;FCU)6%PH&{LT*_q;-eV@#t2K_;k$+
z`oQ$`WX({BD9fdT?+WAKQ<^;Pc?)AwkiJ40b4~3ZpunSTZ&v`cT0)t0C!9HTiD!B9
z=j$$I$P6QT9|0LYAI2;PATJGJk<$P#oHN$R-kUH+_VC47%_sUe-E!Xk+fbm*Z<Bq)
zevt)fP4r*a;QK;KO*PU8c~MP@<DMqCY6a6L`j`lQ)#oVtS13128H83cYE|~XhC?Il
zeuGgcoh6PSdV(IgcT3zAFWy`P{y~@<FAm(3MeFWn|MbE77BR+{jB@#KdVBo>RdkWG
zWyF4|fY08kfRFZGuwSa@*{5$uS1!mUim1<efEDEM;+Yc3j;`N&d9KuBSmV0B!i*02
z+LCVb?@lT;Id=uV>4|KDSTT*r<{TjziVu;m8N)0()m73)z?V`>LEuA?dI88oEz~Lg
zjzW42`4jthHInV0$Jc)xw*4Ov;{V@dzL}<Fvqpy!eDy}@xhmfOaMSFvEU{joC9`3L
zoM*K}z^@nDxZa&DE)asu872+ncq(ifp80h1a(!b4;m!x*)q#KTVC-r40O(5)#81z+
zcO?d4S;7az1ZF++jO*3<E&-xOoBh*ee9%1WBa6$(CH>%aQVhxd<;}{D`b+u!b~wt{
z#ilEgmw?WhCrkv4vMS<)OJUVU+v|_U50O~S5cw2IE2TrhE$ePQqjfwg5Ote_<ZND_
z4n7?b#TZEdvTo%WrCAXXJ>!;qqmZU(M-4*iqV$If;<iWRR39OJq$xT*J7N*00I~2$
zOYqfI<4ozJC_NFr=H_%oU{a=Juuvqk1I&TMUVLIK+BNnZoN|yQAcSNvhrIAydDeM1
zO=C%PezYQ>FmgS)p_nqlWFV(~hA=<#VS=P`OYTM_(p68!)@<&{%X)#xh_v`{tWspj
zM1b^j4<m;RC6NUcK_n^Ah~M-}mR!>3CVNyOl=6!<?Rv~@HuA4IoAaV74XzdV@^OvF
z1WWdw)C9BH8-PwIb6GM!8ODR`7^EoEz$z4{P5A&M^8+k%-?vQ}(Ca*GQ+NC+;jB>f
z&Bmx{m8nFKk9sdC@mmMa*gH+5T8(>Ak4I}JPlDi+wRdA49i5{F@zv$2<{DJ*574KR
zw8-F%;d5{-d|NmM8^~ymDo8!DcloqXm(t}`8c(Q?DwJAj*WgsQ<4>(7#A}t<3;Q_q
z<&|W5aUYxH@5(lal@6TR@YTAN+<r4p%ub2v5+6F7Ay)vZ>l>kpYUD#}DF$sm*)_o$
zf3aCi{pm@@IOgeu?41!kt;UsO9}aI7agHj9(7BtmWd53W7Gn-UU8wR8^iP9p2}@rn
zQ(!@hzECO-E8|#fRm^Nestj;Smuw(XdrC|&eT`}Oy#2kocqxm6orjQ|<ag`G(0Lhi
z5;B<)auO+>QGVbcGzH^wi_}KRKivSDTU0yzW6P3_#y-w4C#4b}VhGhZ<8fV~c9yR&
z%Hq_?rZ;Osq-}fY?9V2_QZ6UTD%n*Os*rdc66b(~8Fz<@Ug1|>n*Q*`PG)sBDN1nH
zmR8F%2HSGB@VHEDgL_SP(vK?@8+MaQ>RF9(Yrf9@=JOOlDz>~aV@TsXcOgZj3d}|o
z>uILJ8D)+ypX6VQwP(h^^PSF}hics5U?dsAkW(BGZ6a|wob_j-w6LWH*e@}Dv@0%i
zthUyZvK@YF?k&BF3$`$u0yL+RvElX~tBN^^=;auqYCvOF$gQZ{0gqOnGU$M&-BiZw
zhc^;p>fE545riII%a*LCim2b$_S~BVTeYQaf1gbz!`HW;crti(@EUCAQ=jBy)b&?g
zO>24Q4qZ;9;gL3&L)SR?e@)h9lGkN-SB#(;IC3$%7)ytt&=3(G1a1KzCc46>Ty7yl
z?SoU=$5;ML$D1E!j53|69wc2`_nQ9MBuRduj@DI0w~8kFEqax^1+S3^?7-7`2i=E!
z_`9}_k@eq~ps_>I{%1{NU}pYLZT$bgrgds)L}-4etY3Yfxa|v9#!0Z!LAP6EfFRLX
z@DkH|+ja=5)=f}7Vz7MiY!^F}nMjLSK%4?cG%q;3DrB~m(wruQxBQW(m=%{t8XcSy
zE15?SH%wLllnbCNyp2CK@9D=%FOT%00M)tb3js(l;PI1?k+^)9%3P4CNXxuW5l2NA
zJQQhcBg>RpIMfdhQ*@DIjMcEEXU$`>s};#>IMyHE$oE^{3Z3hTw)6q@obXKURWE<z
zWsl?K(fsatO@H_t9CF72clZZO!Sz&0NUr4nEEXR!K?pm|shihq6{VH2z#v;RoA;Z`
zc;v`Bqmcx_#_}JhXO;(Y6G#f%on$|TRJ3ok^ifK-K$Va27{=p6RT01<TSO(#RCFMb
z<Hc&DOuH7L_?D;KDh?!Kg%ZvtCSBu(tB`df2&2573B{h9OYODpQaHgut%1H3<Y-$8
zl(j>}QNYW(9a$roK*y3Hc*GN~0FPP3?k<#6-5|=Y@~iEm$ihxs*Pp5v4xmDrT6Kt0
zRmH<{{j4$}u@A3G7D-6ZwCUs4E$Te$31AMA9ahoxC4fRy08LG(1`3mykq9e*tAohR
zA%awTfuq3JI8f$8fm~ozq9TL>4rt>a^7Ezr(_b5;&pH|aK4>j}ieH=Fu!EtiT>xKb
zc#a1AA;Rb!Q0l-sBB(F%jx`~jhjs~8o|qDHu9vsM3WwC*yEQ5}WL-Nyi(=t7=)lUB
z+Oz+{*K6JXn_r(5gWC;bi`!E!SIEz!a6SUP_-@_j%l>F;duyAQD{IQ8>wV9z;_H3>
zqg&?a3YX`@_31>5&lTco+K21w<o$yaGh+TFq&+(NQY-Epl@_htwJuBlsi6`DXbE%L
z>rA`D_K5ZJJtU+V{_=h9<1!iktK%;3;psBnM&{Xu03FLKs1X~T9(w5n3*GGpdaaX=
zDV36!UzUOxk02}rfp)p}US35q=<gAgwg(#MK<A3R-nx;+`EwWXgxyA7ogcI}EReNS
zEqEI*f8qXiS^bS%o^-rFUpl$(Pj^1;ays+IgD-YefPO}qmqVnFP#6SNXX^O4KR+MB
zS+?DsC0B2}Tx?cHUw$mxd}6_5PJM@@Uw+;&8Ymn+z4&M2b$PzJKDkBc?bUa0d_2#G
zT;X~<36jF7wW4(#oxx{cTOI2foHvZ#F|yzuk-T6IAIK0BErePQ9PF42G3H$fL5qYU
z25-Q$V%jtA8;?xJq!X|TUIuT#wqo0}?pu#+#<UZ-{B^^{5I2mIKtzZr4$)kQB@aU`
z(o_Va?r)Yo)4u7*Y)n3Zs9+TK?HjXoQ2x|UkQbbA_)L??R}eG9_ioYFT{V&l2kSin
zi<`*em8qa67@KKPhul#5TpqhvM58%5Df}jtV~p6I=GawIoj-hHBA&lgT1@L=!om^o
zQkz(&Tks*e?j14ciRGg72|fIl@jaiHTu+VQs{`)@5kBy7Z8scA<BX62;uyAN+xf_F
zM>pl9u4@bkm%L2+RBJ>C!88V6$*gqmR@5@^;SJ=|YNzsrdSA($$nucxWD%niP^EMy
z$0YDmJrg+oq@rex!ZZ?-_NvxTq#2c1Smoif<Tg;xisH!0126d9Hivg(H5q7^M(w+y
z<##>)Fv9H&VwP9jdBs9+i;2+A0q<zTrxan2j;A8}5lT;JW%ZRFmbhBl0}aH0E@A-@
zh>HzIiy+lm1P95g3v}RLCs*e{Pm7*pU#oF=Ep3=;ocde6b1B~_R<@(zryRboDQ9>H
zk|&}u<LXn>iBZDl*a0f3fCeoDTNoi@Hw2+9&4I2se&<<bI=*ezlXn6}%s6u(S;n?h
z4++>dJ{Vn~t`mu2HE%nE0vXOOJHAMh8K{O$j5=XwA^CYK$$R5z;;}5+&?tx?M7~eh
zK$aIHNQkc@Y5z}&R(%*FBL{FbYHUO0JK|m}(DjDHFtbi#lqm9Zn^fXIDe`;#aWkO{
z9Jx~h6UOMQ29+hQ9C?Br`pEP&=jZtGmTcX2s{NR;vyM8e=OVA$2Hau@MX;_v*S*>q
zP`1b)#@ytTA6H|Tg@r_=?|&6NVWO$XMWI!}LoK4y924WfW1!d1*_gGkq>E*a3U*)&
zHau&T#q?MU;SK8e$w4IHfK8frYk*9E_|c6rk4xDm%{}1r>(+CC$;?2RXvOgD^q4C9
z(j*GW<FO~LLd&9Hc~5j1gO(g(iwVT=)?l%SPe9Ey_J=`XV?6vUuTB^UnaN@#fIK5X
z#4gjlKo=KswS;|870-b{g9COEmx|C$#D)Me`Kd91&X2<XK|T!SAH+5}4?<5pkSw8Z
zqF>%m%QbO&t#M6vdQGprotlES+k71<7wYibkv`CX(CsJwn?dtc80n7Toit9?xbc=4
z(iYW#R}2Y8F8_Fs36LM#&jF_G=iPWAI5<>B4<+RY`ECpHyl1F}4StaYzlI?<y4aIN
z89FHkc>4t%uC3xyC3;Ue6r>r?RJ1swu2~g)HB4+Jz)mM38+kR+q&naSn$h(VbWMTv
z5;WNbSmE{(OwE|h63lZOh&Q`!7Wc0&_u$X<lh-)6R~F+>_pC11{G&U^i!bJ1N1M4;
z`)QU`>Hl^8NvGMa^RxMnt9q?R1y=39Ep}{smVL{S)tGt$6TzwA4A}p;p+7Ph6G4D3
zNE56HvwBH<4Kyl)VNbtrKC&2d0Ti$X3f2e>V~&ooKtorer7PRsljH2i|KGka)!<a!
zUjK7`bnklNKYess=cq{1#ZPs9mC#fEZTIQutY=sv=wMOZ*xEEx755TxD6YgKzM9PR
zisHJuSa{HKm3eSGRm1k?SQH)6<V~yHio!hTXeP4jq+(mX>fmo3*I!$2g+&j#2>2N(
z)y`?bTn8^glmB7Gz+hFeF+lab-J~kYOl*W&4H+fRafLr&kTNa@Jqwuhw~CZ&Ss`{=
zT2dV!v_|_i|DlIA-!&wU`)C&Xz-+0>bYYRu#p8H~OGCSIRqNpn`$WohK8d=gx1^yq
z+V34IVhGPR^(5%Hrq&vU0yL4MX>CAY7ee})EB|K+^W~8S42x9AE(YWAw}h<FtqY2W
z2#FKhcx_qD(o|KrnTZKGyoY1WP?T_pnSlv8^4Aq;ghqyj$FIStO1EDJU`N8G6SXe%
z6IfWR?9}JZ{nl{bAYUEcs=ZC1CI(Aw)^3fNB~~uA$HrSG*pwwN%G8=|i4`aFx5j~Y
zj>h9Lum;zGg39yV=uE=?rH)JWv~_Kv*i)8qX1R%wQ}=+}HYVD4xm?&qB9E9CUA2O(
zCdmoXh#uTOa#-|g1#8?#NbO|b_=RK>`Wrk9H{__lnUR@U{<-1&$BfMWZ}Y(q)Fl5B
zq;wsr9H#&<kME>^3sNBe5~N)A*($Iy{}ECXUTym7LKc^yPDK${t23aox5ZE0dod2)
z?l5&uo;^SGXUX=Npxx48)J-MLVK=<1##BlD_E)V~ev-U7rt(&W8>OO0NP9PTX<4JW
z44SJ|g$ut}z0|0Fymf47elUKWeJ}%TBLXL>lGdk^sHdpUZ7a?_2)Z#VBxrxSs`zyA
zJcm{9^e@7w-ZdamXd9|aQ>Ka_mN8Rkhj~2R!V3BO2#r;^@-Mu9-5e6WF&V^Z`&)G>
z?Om9&TL?KX3WcUOWp!r@k(q<aiQp8Ou2V7sB#2v8mQFLLFU~6*BsZTUB3~hT$VaDu
zM5n}v?plX)!*{3DKaC4&lX~p4bo*>ncsI;8;feqZsR+$6Eki6`rBzNi5Ndd^7@k4r
zcDVL8r}n099C8aQXxe?h@#*J*xIqz)O>m3hwi$caJ3Ml{Z}8g7A^T4J;fl;N^v%Oc
z-Sj<EaJTEGig40!c}GP5oC@=j3P&PE7WWTkARPR%!YcSuZa&`97At}+Ww-Mb%blUP
zy^3&DjZ%1<X&@C?waZnLZb9WStyDd0=K{2e#P)qeuzBELo1G6D5=QYh1q)sB)HLPu
zBeU>KE=OO#s<D>~LQZEz`sHW!#84QRml?-)PN2CGv_H$$Lsm_bt^{Oyw)`rn_~!uK
zQgcKCx#%85DOh4oK@`Q^e44)MZBMMZu}Zrh2zda2_@vx_Qi_^U#Ne0R=`uQsUP5%a
z4N*NdIE_~)|3zo{aKp9t4KF@nu&OFty{%)tJh!tWWw5HjTpKDH8H0H&q7|xqCM%Y+
zHV2gyS-E1sZ#{5=!b)F!<jJdkx*Ya05_%*C14v+nY?!v4pc*BRCO|nJr-k#=GjRsF
zROO)C#Ns68y`rp5br!g_s+NnNL(%w2r&78U^-^q<t5ulCv1Ko_)HP0*6DnRH<8V6d
zQtW!p`@29GP@oqe$3SS?6i;yxe**0p&70(4M#DsbZ{%EwN$jCPEHntH<wky!Cydn4
zuw)ns<jm&m2%+|dF<@n6-)v$+vHb)ZdKgwVRyhJO$%q=Z;@j@3IIUI#=?9PxQETPg
zP~fY$hk3$Pb{V(Nd~0?rD=vp%MU+RPoO@21_5R63=wa`pT3L3nA5ywlTW0`pYra4H
z?lID3T(?7x?b1Czg2PpbR|{~MYsWnXvg15De|J=;x|Ia!oyyr~ap~I!JFAI9mxX8Q
z>vwl!nuZt<1qdqzisc~PA)t#~fv{qx0uh$a=Nnp^%6<VwfQl3u;%H16h|rLLDliA)
z)$MB}fsm-(4K>9`T3e1>!GmoG)KKlXfGw&OA>~>+WkszoqWOcEngfSGto9lg#ZMO%
z5eZD_PiYeYvb`==!8~m>n8=?2*l^5+?2a9+-!=*ei?zy8(&begiThpFjU(hwLVHo5
z=CSk|$@!LH_dR9}?E#NWmZ~UXrSKLjZF{ZihOW)_1zIyZ&hI{#{%Hl3mWxENsU8wi
ztSQu|t;M)3Ph5*MVw~tWiG~Y`dTw%N)G)DN;!zHLh_VmBnho$@?A_0xU2$37e^UTi
z=>Ik^{eOUvQyc%zk3-o!qHK}!O*uG7*Mg&?o-x|dDxyQt5a-@QCvKGnySs6)S`B{4
zA#ST1K0#3<OA@-x>hJIOAQ#DkB*L><xqMx}KHLNc;D-Sv@`cF`m0z$EMp1>Di`@z2
z+W726kxbea-8=APj?fX=-nDCzx&%wBT}StXxVq|s;YGAvQH=M?7g_5tlBV>$+XQdL
zoWpGTF^56XIOy9XeaUL08~7}mNH|J5xA~RLyA=}xsB(^$vf7yhOrDRyCKwbSq>3fC
zakRWy|2}NN&E#Ti+VdK`^n)LhwSC5N?J`C(tzvc;fbEmbT@@CIPhm<(HbI>Qy!EQy
zTK}{mX<)X3&o#M)d~bcl*S3hBN3_5Yh3RL%8z;g_BXhVqs1TG0$w)`3wVE6=khzdb
z+t=DDvi6L;ENLG~p%Np;Wa+#TzuVw`<RlPtk<O2cHmY$lK7o(1Es&8d=3xz@(5BGV
z0<=HFq}jl<LR2Kod1O5Y6f%S`dI?o{^~6{vZdAe*8ariU6DTZ^-O+j%*@|}TB0;bL
zB<s3>kE7v~_W;oJ)N_jrk^Od&hjK|$I_Ap4CqXjKB*evQE3)lz;m8EGi}uN^puWo8
z@*A%FP9u6#U9<h-*qtsM9rL!~;;ZpeWh9bZBgTJ-+Eb0fev56<j<3l^F)p`RxYSXa
zl-4(?6uD?QZ<S#+&++V*ufRP*3X@aCi)RcVC-e~tAEnZ)X%+)fJy;`^PD}S2=H^DE
zm#M4zi0mEgF};q;P9{yyGo?qJd3Ln%7a&&Mp6tF@bK0f}t8qo6s}<9+?@{Y1i&KfR
zyJt<(B9@+$%*|Rd`#`O@=M#r%<8|%Zr_$7ATHlWIz7*ELqaag7xAF|BDaMCILO<4u
zlUp>9l{PVm{mv+Dnn2_Xim!PqgX>_uUc?AV_VEqQ4+N{o6+uLJMv=>q^#*KqBr3L=
zFPp<UttM*f8?+nn6aAT6VJcY}%xBl2#Giy^(bi`*(~l}h;{ho+;aWbL;lQZ&Wgjou
zHmW!Cfm)6!{)16S|M9qmE20;o{{(Dfhm561uf=!)N(Mzjf&roy!-+IPZrP0f^*37n
zR?#5SA;lsx-@Zli@~0C1IfjiJe*q}gSl|UbD7PdCG>fcXN2sxOLax*qn)=e0jkf2B
zm4B%dH07|<!_G)ufVg#%{w6u8?3re_7{6|DmXvp*Zfo*O8SJJ(V=jjx_5;JD!MVl7
ztI5eO)Z!#*)CucIbL->6sD%yhsus@ZXWB)%5E|FI&49_n2~WFr10II?bV%)jhm^-`
zeq$WFPc-@o88qEK0ZRQiK{Y$W`^K*(vNf5ZJDQKNSM?VdOegZszUdrLNA>_3tbPiK
zT}wAVX^qv(bZ3I5S=U>Xmc043vHjy-Qy#4i73*(CVl{=D*9C*gPfu5~gTdXes^?AJ
z;c?6Nc*Z>$6Jw;ftp};!ous2E$#a$uIuwU#VQM;Qsrh7xtWkM(DOR^brF3xB{!lPR
zoCPt$hTZTspP<O!fBbI>8Wa0JcXa=6(cVAd-I)I;$?00rKbm2^w*MnJ{X1S(=pQcD
z0dG_6X98(7;h7%nXJD}}FL>tub~vCbD)|e5U=@d}3B-^BvfEIV1~UrrhDD(~9@DV&
zmz<x}`@oV;Z{`KTzl1%+mETKc%pl%;kx_MZiNj1+U`3$B>;mCzvmx@6N}|GMk)3dP
z>V%d^$HbU|^Hs&A&h;@!4iCwO!i++riS3Mfaf)J^rl!qzt$9Ks2Lec;wu}_<m%+@&
zJ;VIQX`;?$mwvdBX*W*TSjZzK+cXqxKp*<HlnowWCpCl<MaSll=O`Wn&c#KO0V;S7
zl%kO#5s9N#?-K=~8EvU3WUlpamnttSQFAX|rd5EhHC?Ecw6~;DCN7u~A{P$bY3wc-
z62=pk=z5pa#c1U2K4{?4;59CQ!B6mWKkT|!8J5o5LYfjwMLbn%%6~KIq)JsBP!1P3
zCv8Za%4wi2fNG|Pm@#KP^GFOHo&XZUUNrh#4}tWU;+UOK{I(q(P{{hg*rBxDQ65&W
zp#s{*VFY2xKdogSeQ&)M=6u8Hz$X~pQ-6I!*6+Gvhio;JH>83NmnAT6-?cY+^<41!
z_#-HB1f8lEUobQq?s@P6qLbt_jbhSDjy++tB^HhhY!%k_10+IO04a_0If{M9cb0a`
zO1p4aFw0-ZQUsJIhL}3RD1I6fOeoW1k2fHE5@q2+f%ncR{j7YhiQSeZh?;58>__Rl
z+-aO7VWs9oZ_oe~lp?5XgHi3Wy^^#3PVQ>$_`I5KKDj+uZ+5}6rm5!n{Kim!o<-Yy
zIXJwWUT(rSUv*JKz#U8;Oec41c6}xt4o)7VuV3*N*aT5aW<!(1v*y?bW26k&ZrW}q
zpRCq9x+T*1w0U?X(@f9fd4FDRH)FQF8NBXBV86e873yNQ_C^tsux59PRRM=5#wrbq
z%R|VKuJD)a1BxP4r2r($N*~k*nkE&`qeAR~og`uFH$e{y^Uw;@QU;N={>JpHcg?wK
zdA!z=SW^Fm)(0^~UnOu&mQtKe1hO0}?a3M>O6YBmy(-AyGA8wtGS6Q5Tv$bzX>9G-
zP&m|B1fY{bfIPbBr&(TuC4`?8P;$>MzNUeXL9lUJrX?Im>0t!n1b;>?K{m+%8mXyR
zmc}(MX^ZX9rDGhz?~!A1gXM}WL9xLizl%v?B0EpM(8Tbb4Y9=%3X~l=f%@mBFi_4{
zU^)GTpEbey!`B<S&B4%vTrx<66@FIbPMPUP2-74?MBMKk2pw}I1NIM4zJH*&=AXFk
zSc2W8WRUcq!e;B-FHvaZ%te@Q+%Ov|n}ZwAQHQJ}<n+S9cTplAijJvm4mOP;1N4T(
zk<K|vxXaNs+mWy)7;tSt>mDR#;20aqk<7=n9%>1>z@yZ`uUSuSpM2htS8J0!PcDYD
zz1@*nJ6k;7^!z*dctWjeEt|Vj632Lk&|*7sZYa5&-Ciy)-&C3r7`6VQ&(T!d_`Jky
zy}wTC2xO&?FY|SKz3p_GeBHVE^gg+KJ0N}DyeVl9O?=uvc80II+TKW{GfP@7H+OWr
zpWuFZB<VC^7xQcz?5at&bJZtofnx<=1)v4s0!Q?&fN2X}0fz$)^Xcb&Bgp>3(A8l4
zca!7)wmkle(UAfCUyY96cfkI2*~9K*jWEYp;4kz22d~U}#Ac~N6J;MlG2suzY6r(7
zY!?S|KsS8h79|me60H!m9$?PIe5%qjkYjGiow&ex+V=?qm<gP8$%}~2G-QE_G`@R!
zF-tpX!<)$VFC;=HKU<p2D`r2}D*HM@5Dlp9_Q0~M`|n4^3%{|qz;N}2s4jMJVN_;9
zVOrd-YKaUx#IX_V#d~{7E0-b0e!P=8+~(Q&5(rgw5OYg55Tk+-*A}wqr>xMrCIQn<
zV<<&7v5={ND2qcfnVVGbhp&ugD+@w-dV@GAl;O@?PKuor>1m(-Z#p&1C(S&K;$wr6
z0sXT=Tijont#IT~t{si?_#Bd}Z1p<{F0`q!%r&jn=ayP*5lffU_9J<6YM2!2qf;OC
zHS@;t9Ck`LwWhy_K*S%B`w(_Rg*7M#tfNPDA3Pjz5LGRLboZUWOia@_-vCW|y~EFo
zB6PAz2634K`ol}yM)aunXEW;8nQG)zt&@zMrSuvTF@>Ly(Rnw^#B5@=3HpWq*nX5`
z5(^ybnaozXv+2)f3Bhwlpt-XV&aRF9sCeWWJBJwY)|aRxx~WzmO#)8>yW=20NP&4n
zyDjmTUa$qN#ECVPY^>4H|2gX{h+}JNF<E1kHI)n7b(tB0vo`ijypp*xRN+3%)f^0d
zFRHjJ8RYwg$U@C*;>6urq=`l00FP>Nkzof=Nb~zC?HpV-MCp3$7(vvoEUA4fTO3!Z
zm(nOE3M%7X9W>&>jmlLOlonN=H|(x#x#P@$H%#y*Y|&jl0|g%J63By`##YQjio9!q
z1WdCJUPunqBPPk;LMXDI!S$q8a#qS)*2Z8+>UmhRmZ2togiRdc3N7c20RvUH%2^YX
zmitXE7}|AxdWAwR0jKDBovQhUuVS~5oB2F26h9wzpZ+&@L*goOve~@?h27PHsS0Gr
zafL!2Fr>0MzpM<Kk{7DKt+X@MfCNJiy8jb$5RBMfNB^yCa1rJ7ZVh|GFBHFSiD(3W
z2_OH(RtLsfj_Cqp%@3otzO6%&5gNoEyxAk+vqA{-LyAEHW<et_WAp{70_Wic80t%E
z`RvBLNBhj2S2P<?o+9H{WL3jr<37USjS2Z~z7MU;qP14cZNc|~E5!ZZmCX<Ysd{e%
zgu~|(-@kiAck9`pwqMa{aE-J?OFfseQHj8*W3-^IGjW}oK0b#RP8+qKm|j74#JISh
zXxu*Y%~m@_V)hvCnrG}!Zzd2xb_e0k|21=oDlId6=|swlJoxLHQSKe*meHz^9cu2B
zV#{oi%P<d2-d(78aSp{sWnjy!v3w_n`1I`V<G5jM{sTD4l;js$0-x}U%32~Kn|@dl
z1wr^t=e&7yzg+QRweYnJ<yY!75oaaOMe(cW_Ttq=gSn#jyxC&YWx@8M*#Ym<3Wt%p
z<9x$$k=d;Dd|rByaJEc452_aj_p77({mK3D{E6QNa4Wby{66jocZ>(V#^0=HK3qQ>
zFB9h1Df9cx#beI$Sr_H}s|t~AwaBh|G%pkC*D3Y;%*JC*i&>YYysL@}FB2QI`UY%4
zEUF74jJd8X3b<gV0=6PCl_E(Ma`Wj=!J2ZE$*0nJ84s7OEzY2f$Ca=*!EnXoY%Q`l
z(#K6*&JMOiQWbX)g(mBI!Q$AjKA2_qjU0#U;&=G;cohD>X+~LC{&u>i`yb>yj9Y)p
z6u+nIhVSY6J677r&)Ap7nEtx#giqFzL@qakV4RU<CYqLdv~r6Ol|76zfQp!m)NQIY
z`H5fPXw`43m{OI;M@xHilSl{$Fb01_jfdSIa5Sbr1|*ciq}3U>ij{UNJ0@4h8Vy3w
zHS<;5_0H+#ZIhcjN)hOgs1zVjpWBA#@bMYCGv*WcH~WcQ@9oEUG3Ak-A7;mpR1z=c
z^~R<_Af+yvd^hDT)i<m?Nw;HR8i_SB8yd#xSqv~p&B`|zE;^=}6}oB`Z-Xq<CgN$#
zgx0d=eU2Yd4-`Q^31^W`l$z1E!MDO3r@nv4#ST4+<(u4Y)yEeRN82{pF%kXFmJX?J
z)JYS1Y5u`9i#51+mBgK=wJws2&Rx}+Ay3a-&5u=Q96c=o5?FJu1}8eu2VuK$(om?r
zb6TwJ>>fE@IQ296y~W<iddT*!YU`oNJ~Ft<uGRUf06pD^HIDIgu@xJZ!RE;r2dVJ3
zv-B5a9~r*_5Wd`<HCvcpLK)47q%J{koo+K8@GoM7fvOgKqLoyT4A$KCE#r97w+JKM
z{LWsQGyp6+iy85nP0}0e1^dvIWqVMzaSEEydVAzyFTI+TO)!=eMO7km9o1whE6SSj
z!`D4o>8x%25c<0xq#6L5M|4<#r92>FTB0H>M$O4ZTRS@zJA`WyZ4w0|<8!(?#_<wM
zSCoWz1v;OX>D^;-4X!@-j6`385MoAZ@c6T`Ra4%6p;?Y1pCr|iN_=bNF{Im@6w&;&
zvR5U5J|3|5&B9iT^YgtiK%hZ_twlj*wHw`gI_2=bY|8*A>SE=;27QKoW?OViaLS?0
z@FATs5`1w@c%ptR*Rn#Z4Mvy2al<Zw7!@(8R`#qD!#D*L?pt7}q<X?kZ4{Ii^sgGT
zCRjhG&cU|F=t}+s#fo?A-{M`!k^`)F?582#oT)EJr-7YH(Fl#A)65ZX4T31hjl*MU
zzRqqPWN8j`{$G^6V~}NQmo=KU?MmCWZQC|0Z9B8ls<iD&+qP}n{_?!tef#w7bH8|R
z%qR9*F(dZ=Ib%O#4p<E2e6a9jiKGIB>uT^cr0YR}?0i_u)RnYLw_cd~UQ?&TXRc#s
zl0pL^nB6fdg762oLQdFnjs;>_N+e11m8^Wkwy^#w^pADfN@>7!<}3wB9CMFdP6QBU
zSqaj$9LaA`ph_{Ni~d<FnJ066WyXl6f^6LgTetl!5WZ1D0AJ||C_O<9#0^(fu(ltN
zcNG6N4tuq7P3m8xeh63&kdB*|Tgp(eEdFX(p@0Z$Xy+3kA{F$IW$(SBvrFOz)3rtv
z<0a%-5VAwZ#;%Fo`SgOhK?&KS38Pl%px4=jD|ZGJruE-Xb{-Z7*OXmpMPu=Q9^yvO
z60iX<6d6!5uH32c=EeXsB=qZcQynknXgh7_<UQ9%u>H#rf61MHEc@ZxoM^mYsM9h)
zEME(td&G=4<p*C%|1Bvs<f0C%ABeum30o%k=u1Bq$r3%Z6#Yg57qlS2jPmV?M4wj|
zP4y?Bia(^;KMW`~hJWG+GXMVw%F>kh-hZI{wf|TKU>(yjwo}qq)Q96i=`0M!UydnX
z)G{VZK)KsHvT&eQfR+$RkT8S#K|MD=_tX&`E7#vc+1(Qgt?kp5t<&C-7X(H?NI=wc
zPO8s{AWYEfOs@s9WcPSt>+oXoIJ8D3sBVHcBd&xf%>vazS>!VaG-QFrr|aADTQX{k
zc+JS+2EaGYHAP8qN31n!DOflQa-Luk)H0@E7l0~-WyBOf_8FfbfVodg*Z0*PQCqIG
z+YW@z4AojXY);XbZV)G$OFXziaW8;c_z?dAbKk3Jr8)(EhLpExL<qXqQ4m`E85<JE
zBbS)?Jk%Y@lJYKyY2yQPgYIZZ=QK1an7qK+xT0{HHWRMrC7$!5!97$h1O}0!y(YNs
zhbG1Bync03n;{rcktnE|ICJA+q!mo%g42;QQ;;((M!}TbpZk^j>41KCwx#<eI<*gG
z2*9}JtaUpjcq|q0&r0JsogY_zeeS&l7&Jb{bE`n`2I)@t*Yp@@-`j7$1AwJL+0?|R
z5FiHu;3CM~hUNtQ*=n$&JI}Ix+qvUKBbM`7A&$-61cF_~Hxo>3*lDb}b$V0g5Dq=S
z=)EQ5H%puZ1}M3OK%4|YR>l4OG`BD_(^j|O%ka?E2h`VN09w}@!q%0-j9hKYPIc`r
z*B!P3L&%jo=w#&WQ%7yk^;#@s!CyXj>Q1OeAQVDva*JVT675(Rlo|248Zd=V3!rUl
z^hf;y3?eAv3Sb(3&|Cev_($@Tzvk<j7c9G?e*Yv?BRL$7bS!|tTnhP}|7+wiECMdJ
zEAKQ(c;PX%A_ecJAuQIxzuk&W7s*yOC=^6Rj8@>c-3AxR7n0ESiw#$>30pC}_8CJe
zYe{ONO0Z+(jF50RMiRDl2_#wcuQGmCwvfp~B|76y`IZbj{DG9FjOfG#TU*xh!c<mK
zT2><{#TLClx`yULdGXP#usdcoy7$#O8o**WV(!#gg7QBAs38^F4MJA6*&B*ZIc{$q
zE=nmO<cH#Yv%Mq;x$BF?9oZPbp=E}3GY+c45+RjgxR^gN&MP!AF`a2=L2M4A1<b>R
zQ9S=h)K*@S2dA&>wCIB~k7TH0(@-EtHY0+@o|)9Wt`=3BdfSJwy*bzE=p4dx<LvD@
z*01AbZVN=)`Y*-rOZ}Yja+uS=7%>ff7%Be|S=}XV$46W)a~@v<211HDtNP2RAJfXL
zHH*t$g^x^aWBc_hxAm#W<9OHOnxq_&U_GINM!iKe)z!Y^4L*Tok7c8Zn@I^hUv95m
zulCjAGhm=3I61QIg^_dJhPzoCmD`q-#`&^BmyWm6tDIT(Pw4(o-Mf+K<bu6-peKGY
zMZ-Kg4LE|}*Lb<T;REnsoPO^zW41E7$(aG;ll{9M7B%xTYy=5>BjAA`n=z2IEHPhl
z;14R?%r~7nb19@$g0c@NHCqBv0b)p4!#DD#QurpgdS5nh*2C^Umb2`ctn0hKi>mpI
zB(vWuOj1S+2=0_(r;~L^#iX0(8f(YIr9CiEiqT7cfF3v!25>r=`>%((`N=`Cj8l|n
zYLHb|bSXYDp{=i-L@vy6$Ru4OFv#-6Myf|ac)C$;$ZZ!ycdQK_OD;fSlEsKxv}$l1
z=&&t}l)6&A-ldhT+&lm5zkEWBB9NUgk?Bj2)*{NE>-n@9uj+siLq!fqT$~DAdqNY*
ztk)75xE!$d4XGCkFB1-j%*g`C0YKUrh6Gr&67gKc|FBY`+Z3QXoA(2QP@kNzygXAJ
zTdBs&JKFr|(kpRIP-#I4ZMv}8u;9fnd;f<E&m+y52}@oPqln80Y@}NPXp-;|WYm$#
z7zWJP@kM%G&gGrA?RyG|dHvHYXJz<%sM!A<`TUz)nT7G+PDPy1(kjRKCQtHp3C4~@
ziaN(h0@c^F4NBkx)}MtE55mVhAtY1tM$*;1JiSQHu?VR%+MvlG(+@2axAA%FP$r$`
z5j8$2$Ir7Oh4a_MH>;qNj~vt5mkH8}TiWpEkA$?y&ms73Xy(n%lSm2o;o&>;-(GJ4
zY9D2E5SQI8$c9_@y2#ES{n9~--m&{BBYo7ya+4-+7k=<wMN-90FM6f#DT)_v!z>Q0
zM}OE13}CW}`#b2c4LF4{4*d9381HV#3mCvm^^;!R07k7h4b|s(UXhA~g3T#BK4Qxz
zI6>7Qh1kMsf<(sh`<GAHw`@vfby)J0EslYGmXu6>G%f6`W*Mmmh<d?j2gGiX$V3Bx
zmKK3IH_t03-04VFCWo(kmmRhdvD61JG#9W&5HXQZ-O)NRTE_IoOJ8gJqHp7Dv6oTR
z5sKETJB%>59(jF-yQ~R#Ns@Od-ZkpR;EZGf;7=(f;9hs`gkEs=1&J@RRRUL3x;kp3
z@y{X(T+QLAx(_P-?J9;v$s3Z)eCnk}5zRz(z&u%@Jj-6M7%VgUPF#uLKKZ+~aP$Q*
z)pCY|YjwkC3cg!C@?~5o1))t#u$*44oy`p5QF7mOff6QpIl4$CN_14V=sWb<IRYbE
zGaI6lp7X^nBM>$d{wF$@DAPEeLmH(O#;*0!XQ2La6^s`XScVWk2guwb{9v(up){Nu
z`XLAEy4TGKW>3mv!%fJ5jN$r^EFMqxj@PHD*pvH*w<o!5JlQSZ&!J-XkE0)3UuQe_
z6LX<I@c7zkuVD`+PGXDsGCRFx?sj+XH}=}(PFO5P?)pThIz38vv0CR0pS_2i^osv7
z?0fq*?3*hu{+D6jHxZL}H}|j5Yme?OP~r+xLcLP_9lXLNZLbzhJVsmlnqZt(<3345
z?3GMlSO-vPS>|J~@YY{JG2@rhr^B$caI&mU&Pu1vi(fK==<)QRIOp=@u|O%)^{s7j
zaVmDwU}$rKBJ;@uGJcYHo#vJNh!Y)rRWIadwd_zP4sGjbq4m3jY}R>n#62{9JH(fX
zuo*E40bkvR>h?CFG7yV|e?(aaZ~d~(SwBEsvvYiYQ3ePbjfd(;V3-sIvRg|fCe<V2
zcqPDK#=#VY+Af2zv;sp99xGQ<2RUrS820cm5sJ{l5ZTZ5A7AH6H`}tQQs}Br_3|_?
zIvDKW;rQ>+ph-Q^kA!_}6EVXJTg}r@^>SpV8Ms6aXl`xr1R0Ep@DPvTo`Dhwv+m?G
z=YkCh<As}KQU%>cjhC1ve<yHk(&Ih|8Y-LT{jf&1*E;2OVhRWAfZ-S>#2^46blay)
z=J{!AFSs8Y1?u84W<k3U)J?}?92RkMtqsj!BdT1U3TMKT4)O<<(u2VN*r^6rxVOoH
zn0A=8)>UO|kO6);VPWJx<Bq^pUjad#kglT!pKZYf0lHK@I%~$-p{pSdwr$yVyDJ&g
z;Oe&p-cN(D%WdwsZUW?PXFIXVIU7-soy=5SQ{kGqxn_&Z3pncL{Yx&-8vXI{p7d?N
z6dJDsd8o%Hj~Y_KZCJE6z$-<xTEXU94(K&i2qUgJQL$S7_wGYyu95vSB}4}UJtL8b
z<@+3SB?J|$c!NO?6?W)H030}!FRL~Ng1UHM!?BnRWH@rq4>5W=owX=8fkS+xjJ^mz
zHk?JDZml;-w8RcCR$yJHlpT@ZbYZSnf&n~L`Ox7TeoZkC#9+IvJ}SbcKJGPy2&*+<
zpjOQ}bf(uh^p!B%J}I)kA8T6%8>PLSo4%hX#cVgzk;PBhq^r=k%>rXDH|P-(5SZlz
zDuwRBOR{u)+#PS9h}3f@2Y2Wv_op{0awikMzRyw!9_-%G_d`$FydE7+c`r}iUa#*v
z8+$`eY0jUQbg_Kz&yN|!db7&mh6yEy={_%yz?+XePM+>%cSK-W&TdCcW7iou1oH@*
z5GkRa5Jt$&LHq>v5dPzTD{BA$T^0WSyR?B@!2S;Nn(eb$NtvLg9a3l771oVLoH<M@
z5)0RZ=2N}{_6#;p|NMmt=YLLKjG_{kw&8}>sUp(w=FAr%LN!eH%aH(E?Knu9iBykD
z`=Ao0If=_Y2kvYd8vHzc{;Y?XdqKbwnfQ?@v+Y;dW3Q-Tb(=kU0)hbhl}T=Nzpk&r
z%_;sqo44vC;+VS`jqBCtxyob8Bsl4oPFXS52JD=t?eZU2sy9b)I0Z#`I4+V0Px<R*
z4{gkI^!Y3Kkw@d+wH}3$?;v)QB;qqyxGCBirJ=#v3~l-o52P#|82B4jv@9Kd9<h3J
zD3t~(=nR|He05<ffd!^=D=R>b*v<9P(oKKR-c*LZ^ypx>LzP&=9o>;wJN+}!^HGrw
z&PPxSxD^kiVe5Ak*<aVGra00DRHxv<8+o6St`BRGuT&agi|63Av?AU$w+(h(WnHth
zbkuoeKd4>CUZtG<XQIqhw(9v#j7;(0R6df5Fp0URP1@p@qkmrU_rjyRv=naFxMm&G
zo(aKNH7<_5UdMM*{%Ged@W7%(A4`nD;a{7fMzHh5-0d}F+twWg;t0vsy#ZLWf}7>=
zG|+$IpVHK;t-W3ZvS_PYwxOXm5lGGi5j@}VCd<{9{0f-tRYX!p3N|(JFFPu<iC?j!
zu7Gh13}Q32>WE)HqpBeB*Op)9F8$r{Ofn1#3VCX4WMEO!Qrequ9oxFW7&`J|9=(wY
zk4xpLd(OMY<XC-lvwbdBWtQDz+vslXU41gEy6hp-_RVeYDAFgu+t8J8!<&vh<cIB6
zgUbTF&v0j`wrjOf@MYe3%=gin>ggM&3hqeYra}+-Ir{AfIt$nH2*}o$;7m<Y^MK1U
z_`q&z#dXQ*raBq$x~wao^%A2z;?vyq$J*n}__mr;`r)gf&v@Q<!Jy6w{J6<ERr4qB
z|2X%`n*Tb~9+YT2sL;m$qy7_peB!_SI~*y7{}=Sg{a@yIM8Ccpuq%>~{({>G#2~P_
zA=8322p-@2u74E{k6<KTt_@GGN+x9m-d7~?6=d`s!S1R#pKo>RC-cL((nLbp(cFBS
zcUd7Zj0VJ91p#^X*Evu^m!42gBMUAc<ns2~b=M2vVf;tzQoF{Yt;+IZ9fS_EHPd=k
z>vFkqg^9m@d$$q!xX>S=4H5S`!nWESO)cNfjkl1^mRP=^6)POt1B8fWSxa4%TD7^B
zNWwgK%!uXkljFktj~{flkc9sP46yw}1_c8X$G?wU{sjz(qI7<{&zl1;8b1~`ftUl`
zV;m=gfl$prL5C|!NW?{cKF9PcOxUn9C8A&p>;Rk4#%?`5x(i{@5Mp?@1|B{ecsyMa
ziLxUBKnCQUC<(juf1{RV=A5ye>*y!EG^6)BXZPQ0R7Ao?t-lf>&GY<n>RR7ihxHXv
zM($Fo;gR8XJ{maBA3*QN`aldm<jz78j6F~)Kq^3l08@9db|sGYZqTANz5oaFcEdK_
zKi^MiUm1H~Hl3_5%TgO_WhWWe%+1agOw;go4L^7`i|a31j|D31(s2gV5~C6Fa~|h(
zIDcD~5Qf3_I;O@H_(>X?F!dMs2*q9k_LB*d_x`d{rpyxr>~~|<Y1tvVtI+pj#8IdM
z2+s4EmyondF_<A>F;5qmAo>oZB$j^o#VL=K>rBSp+aTD`n*agfK+$UYY{@6)N3|kv
z7?-5c8I~PS6ho~626Bg@5(F1Zs_o$HRfYV%esqr{9uv@v$WY-%R9E>#xDbVfU+<;$
zF^ruRpOwbuNBJEEm;z|C24ImU&tg-vdEk_?YgHk=%iVl0O^gr(+B4*Z!JFWBhxU90
zI6cK-nq(b$uFo(rnx=O^b4R6^e_PI1xYHiwEGlB4ny|d{QyEYkdvF?K)KC4YQ7dyU
z&O)z|6o3jRb+EZyTTRfWbD;{7m%SDYa3IM6E0;I;x$=Z@dXk8dlk5}IJ7?R|F9C<2
zpbBa-%sLPn#09ppkx=I}k<OiLcJIXl+b-h4i)~F8>w&bfu}Xt;`zQzOZvGzRdyhp6
zf5plZxUw6pS9w{})7iQ9`2NIvjz&%8O$m^@-+JEykSf)MGhcTOJ;{q3RYU#4kbH5s
z_}XGt4TfU?MGO@w>#vQP8^2<qM8nH|Z1^g;BoK{do#-d{TowfZ79ZC0Fqt-v5Qm+f
zGY6%xf#RyUOevIna(gms`Z$ZqVkk%#JCKUXG7Bku+$hqRC+amHv05JK2*(rRgyYkJ
z<67(f3nXz);n2U8{cLhGH8h$rc44V1%_G=I-&B}$E)|`Z#6SMF>^VyO>hZj&^$w#q
zvEDc>of)Y(RHl8ErfJ3W1?N|kt|;zWR{B%t(K>j3z-ha4oVV@OYHe;Ht$kWqdVSnu
zZ6#&U!`eMnNsL{vg^bD_``FXXxfWWA??pV}Mc@*NA}Qy3z1NLBkdkeJO{?Tx<(77k
zWT=Mc<`Otd*2D!#$dhIxGfO?^yOCyK3iQimty9%FT6sZTb)iU2RJ&*@s?q@l79@as
zf>xJpG86p8aE7H=N14lhjmxK6j1EUeHhtB8<E~|syUmh&pGNUCi9w<*4rLAe;m|n`
z#|vREU~k#?IPb2gebQXxxLF~GyxPIIyV~&rOF%OY0O;@ImSTTQj~dPyDUh*4KyX*F
zPZGYGd<xODYAGfJC_6mPmqs4~v#y9tx2bX5+O!@4dwPxU<y>;|^5oBftEmyF`)Q5A
z1cBeT8F_8^t2|g`v*{3bJrE0V5~9si&ts?J*O3`yCt9O2=235!kRK*2F$8X;;RUWB
z2l~K3;XW9Rm~$eiDN0JDBMgeG*DL~>|D!wf0Dz$n=DUl=K17LXSdzxUk9uca+eqRu
zV9p^ZFN;j0WHpJqu(m-6@v)%xl*Q~3>kqNycjF3|mZp}1;N=|~1g?}Wd_sXA-a&<|
zOHFE`1t#x0JFjiU!`dQ^Ffk5kUbR}JBJFa?Z4EGCft_)mZ>ztpx1AV43KacG23Ydg
z1Vcl*b&rDp62Suk;ZjtA5C&E~zR+^cwOSi)T`gYvbg&*K1>_tXi<(tc8nwdw^;chm
z0P=RSb@?^nzc@SI(2rQPf&XcQGyZ+~^zY*VO#gP!WyfFkWi9_5bXh3(|0$+CD$#On
zhhM`AE9x0netV?BO_D6+RKXe((Y8;H<YXw+mvzcR$ZQlPXerPcrZvl#*R}{2bhddb
z;ebrrs_qGA{?3(CmQX<*H7ZLSm#`G1;H1E!AR%LUc&bZKee)ckn3pe;WpOFxXzZ2K
zvQHPL2UE6)nV+4vSm7OjNk@?C11C^)>Vu~UTCZb5mQ7;{i@URU)};L4fl(GbfNxhM
z8YBm_#~ufg&`FdhPm2T)qG%1vPK=L)tmYr3HH}P+h$K0Zv9QkVTGECR_4bP+Hj5$<
z#CFmKkedmLZ%Jfq6#rQ&R|K(`NBPjhURp?N^6h}q%<q3kPN`ldNK*!9m4_iT(Zl)<
zHVzjh9D6{4_3F8TdX|b>j@=|zPSs*GNSGq{MUJLj>MaVOp~;59TVkm0WNz;_bS0?L
zxLCMv$!OWL-RLRre`0B&BjwykVP-qgcqv2zwiG0#n%8XlF;`MGfI|<KTv5oYNA0zT
zf={w!Ntm39%kLms+f63&7KxfiI!QaLR9lCDtj$7vk~i&31TJa6C{Ia~yj__K3Eu50
z0`yVSts+vONOw*LG?W)Mv%QalkwqI%X$B)ssI3%8K%nYjG}6l>=t`YhT@n;AqmSWV
z{^ZJlwI2D{ql+Tr58QVOWOqfU&rGuhjRE^PIm0L#*ypO&Z4NNlf_f=*X1nI<-r#b#
z`%P)O@bztRPO-bXy!>WCed{RCa{0a<irc={wL9Cs?)NT=T>)ze+PmIgZd1mdt_t|p
zwSAgDp0_uXj)^v^L96LCDtQf+^as4_Lz+6B>2#aNuw7r~KGr;4-A_KR$3lI#JSv8N
zY_M&uZ*P6Y3Q&8NdxLvoF@oB8p8-hE*9++X^jJd`DZ|6_M2e`zDdWYrk2;t>977Da
zhH-%i*t#_b_YVc(!VFfuNirLc5!@*b8#K!$pAb7(DbU4`x4etu(CjaN=M6KYVgQLP
ziAqpxqZ1DXMPOi#jC~<Supf(<4zD>(6H?|T)GVVI|0P!^#4;r$IKUl)yie*QMX8^x
zJ*62r@q^tB+4DX$r!y%57bx@>dKP_~5CTet0}_N4FSF=KSenswm|0c{W-<l%8&i@q
zzUJm&!#>m0g`Zh-p)kmd2V77&F7eA?h1)V97w8mtq(aIXU>D<uK`H@pDnT|d;0oe-
zgcq~XYP3YOoHVjQQXq9+;sHTKG0SZSkKn<%QN0s@ddQkT8*#H>a#oAS7_koGnWKI%
zHI;y}+-7{ub7S#wQXnwO`_1vNCN!bG5;H>KdEwE)OlY$|kOntrJ9-FtWW+omNHt*+
z{+z^$S^+|>fu1=xG-wPJXhN91$WDO18F@H~O9W28gp5Up)u8&XHQThtxI;tFWyFwo
z7z};C@qnHfz;NCAIe{Q3p+qVOVLAW~XkiAqs+j#OIKU@s>aub2CC%7n20w>zZ%{+g
zmc;<Sm0DC7Gt(s9sT+Th49q>vHY^NSe@JYm=jcSDT;VFdJ#8u2S|fn3C#_nOAp{_w
z;(B>{&Kg5-m$@3K5I<09ak?F>u2_O9p`1!Xhn^;cR;8mHw?V*wuR9UA0AWk$34qbG
zLKm00<1BR)YuM--NnUnVMemS8d1}E^6Pfoc<f=jp;n(nm5jZbJ2e+}`r(qTSe(Mw{
z$*G7<8f}fIf$?F1AjxiM5eDIz=jTPKj+u@4k~bk8_;C^_vGBx5g+JxXE`BqALG`8z
zoe+gV9%x1okmzaB6~At747-tj)I&s+lljXN!&zq-J5mp6&oNJNENP}J*JI*rw&=cq
z&8``eRR$M@^q2l}Rtb_6Q8i$Zu~^ciI<H0QT6SL*B2ipJ+Ec2$j<164#8`*F{1+!p
zQ(JOk!~5ms=xMJ|?a38hXNBrw19wH&m!x8HpoUL`6}vU6T1VHn_3QZ%jeO(aVNC7v
z+g@-9{OPj^*%y|ZE#)fr<J)bJY3vOb;#&axIaVCo^zk+O==;q_yD5g>+W0zcmB<4W
z&+Y2?yoTH|rA7UN-p0Xaw7n@pn6)Xq3W_a=BZwu4Cx|JCE4T&<ErgM7WBw<1CBzi_
zzhCyIaY7H-nTK*Y;3gc*V}zV=(*M<BW-iX<fSs@~mlbltN}rql9dh`u7W{;}`J|8&
zUi#hi@6bbT=A&GWzgo0HPMGOq)7l}2brwEdB{6}O(7<t}Y&iA!rM1V%W#Fwfi71t1
zJ^L56Y=-uS`V&X(I%l_{(&NMYd&hnKg7t|B_~HiXfI`y%)(VSYt%-jW1xv+1qnaH#
zx+5`kG4=YA?P@4^sqiX5a|f6}>fGd^r8<K{*V{exsjNEmg-Z?OS0cd2zmOie1tesS
z{LIxsADem-DLqGX?+uOiuROMq)0eK~RpAZi&)_oItP3lxA>yEam9VlCzx^0!x{!^S
zy1<v6PwJtxqC3KLfdse@gXBA5f#v>T6WiWNXM$D((5S>p4VTrze?g7Xhlcg8mYn)f
zb@xg?1DWy1DwW=%vk-V_oJz}RTW|xu8uor!3uzAQRS}_$L|tLDC^k7&B{4Z*n4R+%
zrIl;cqK4P#++~ZnrUgVZfoiUr(=zU!Xt0&^7>AVtj8u*LgcM&JiRr{yBfcm&PwLc5
zIa+H8oujgV`Z9I^-=mf$s&V5DoLcocPp&)78SG5CWPEi<s%p~cYX&b;oPFT=Ip$$Q
zI#K+R9bB^F@WJ3=YO-ST>p=~OVro#ImAYWOnv~f^G|w0=&walszo#I)w+XF3bD7Nn
zj;?B=wnqe8<<YSR6N+&Ij8Yq>sg0M8PYR%MeyR2q<9HcIWyRI<qpdULQVh-8tK&~Y
zb8QWc^4&li@7;8JM`m4ez>Bg9k0K(J3rzX*{i_JWMGhY{gl(PVF2;c<R~~36U}!R`
zALNJLYe)9b@mZSthyzjg)Qtnp+X9RAI{aI-Zqgc4nka9evC($wy~qqWchN`3dRNwg
z{)E?AuTeJC9nL}dnkxybD`0NyF%z6-*ir)4AsEJcmSLDH>fr(_MNtQmtkl;rhl$qn
z&5l~#ph9OCXCJYk+*HGJkBuc2S7~*w74P|>9cG+1c#*X2oHqVPs%OUew{*fFQ^E{H
z*iz_0;m~UXZ{9jWv?|$-dae{<-l4jl%e909O=R~qYN`v><f8DNIjO9**`NbHCQlKD
zvwXTxg5|fnsHrITgyyAk=g#A{=T+1fdKZ2=dGA)0JeG12a!_a3C)?Q8Bem&kU))x`
zY{n)OD=0%~5KFS`Fg^xhzma<hc242GfOjR<|H&Gco$2qprvFDb$^LIgL0xF-#AdLg
zd^_Lv>_yEte}c5dQrS>F)Q+xM2a`J}Dg#mcN;7Wc^TD+@g~)ahE}F1qhL`Tv{*mo@
z`OwJlQx)zf?}mW)hmp_ICz0s#4Ztmg<Dpb&Gn^0(KIQM1UibE?{H}w(rSEa~7d+r1
z4&a~0gxsCqZuyFCOg~(w|MmP}a46`;S53?H0O0MGN>Je=ETNE=4{p^PSjZCNcZ3*R
z%6xx0E$FZ8>&HOG<+t$V`||UMdy+ozDe6phK|cN2bk!JHKedXjcKzinXE-mD;V(05
zAT0vnY3vX;W=%sBn<$OfbH0%paVDqV>i&M4{p^i1ytFn9og^vX5JlY>zyTUPNPRh!
zfikPqg-4FvI>w-5g?-HpBcy6+FosU*2>PPM#!_0=v(BO@_UQam8HtQGb&!)t%P5`P
z6H82!{zDtcdWU@b^<CmkTZSiN8>BMEio&CfxVg)dyY4vwoRX^WCRgr=D2|nzY`qZ_
zm|$90b{G{uXp_<ICIwpE4??`1Uy57j@hY0GG@owG8{tN4Rg&exD_E!lY&JSfN*Z47
zA^59w<fEKIApA;p3h-ukufK$_Gvog_EcOqm?{$?gixM(#zDri>&{(Za(A2MlnirWC
z)D!;n%Q(D1C-rD}EDS`4f*_tVrIiG<oEIRWl1Q%<Gjm|*p_Wa3q22^LmdloQRanzF
z(X5J^6-2P>1P6nnLyvMPk9R1d2cjjt=+7x_?f0?~L_3oLQ#IB&&1yz7zM-wN)y8yc
zd?67SCWREd9Taf$$9&<FO77uamstQHqHEIXs}|zX#f70jn06{DFX{g2W?M8jsKdd=
zR@C`ZqarX9l&T{G?6rqD>(B@jgi;LJ=k%D_C*FrJJCeB;W6CijNTfh_=2C0KPGkG%
zO&b+=wXK@ytW5Oj(~{~=y9pEkBaV>Vz9rLl##};@T)*ICTTS(yFq-DJ43nw8lk!l6
zq!@%IfY2OF1`o`^1UxHMo+YQ+V-2Rgl^aHqZ)r_cjL3sra<dRo3tlCo&fE}Pp%Epw
zxlbhHUSn{styrr0y$P+#yxx#d+lRrl)i`CO8A(^mj+C+z&9%DMT54+^H&n-C;jZ8k
z0ft%0O7&R)FYArw!em(PfuS+|r#E7rAJG_(8S|q+N!?0RF*18~)miAe(lWjvfQX^y
z`0C;EQ802%V<;|p9CVF|e=g%LgtX1i5>vNU<q1E0-ePn4QQhH;2D{lmzy7#6Ss+7O
z+@&|<8zTlYc9GaA4Flo#lZI%oG$1u2E;`9dAw<q0kutIYdMw6;GYlA+ptyfTut}Wm
zo)S-0%iZp#e?Mey`RaQG4|GCTTUoOq!0~G5KwLsC7D+v044k()c~8^>1o@kFelf5=
zvdFCjMJJF)XjwB@VOe`*^-y80KpEjQ%dD^A*K<5QCzl<IO`RR1e-e&H;zJfrg!&yC
zl<!?y?|p7|D?#vU!$EKe<9C<zDHoU7W>gOi#oT8s2m2wP(?itsYITmbPe$+5c`gJ-
zZbV>cUw%z0J4??mk1-&*wD!I@`8?5`OUums!&RD{{h!cK2KIm7Zc_7aGI55nr~D=z
zeJ%ai;I@<o1m(4f)-YnV<a*m$zRvl+aAAcaaw}xANSuOWhWAc8(_zff?~Dok>rbG?
zp{*lyyG;UNpl>E8d8#ig1~62;9|0tSN{<)j_YElk#`J`5=fg6tcx53<$gQQTDrUlp
z?>_MOVgcqG{`?bhS;af`<#iW>_L$^X=q(J9BYv_YHMcZCgcbYE?gZ8FFX#Qb{v>Qp
zGXSGYE6FaC&-i5qp1}kPk`$Y2<0gYg2UyRh`;Aui!_%={8wRbpaqz3Fx}7yIACA#W
z)GxycCi&0rH%0GGY;d-8P>QV|Ds)DuN$^7a7#szxVbBHFYgYHEr2n&_52yjww&={I
zZTHQK8<w&q!=>>5)gC*!u`7Zck0xl#drER6Vb^!X8+GVB%C_QFBTwL=-dj}IOT7Id
z&00Oq)HZRJ4&C`y!;wILRyTZKNB?^eKrG6DKZh@HVUQ`68t*ugr0hn54{8`u5=CAj
zYF!-NS8UCz0|NbIgI4yQVC7JwL6-ZLH|Ey5Lw$wuU`I72H=bXDFC?su5>=;_a)0-k
zMzcU~amK2sSxVM*|I<qYE$Ll9;8tUuoEw6YswdarRq@5Wzr-sU16RbMY2HRAl}|uT
z_&KbqB*axvNg4OEQtNaGd*rg!iN;oM4p@~419kP4H3qCAeBcvAC2Op;FFXQpg=mH%
zdNGKw)Coz#MFeYuuzivL)Ux5l*U}hIW}*N{JkQUOhq4HFA%@lJollOBRg;y~+t_07
zQQ}nn>Zfu(pkiV%>~0Y=3x*cF1fc&<O-ypoFihDMZwe^g6oBY;%>s`eor}IY<>jU(
z%_Rxl>{Rql7TWuMF~vmYHB&#FkTO?$ng;-40IGAfME&Dst-)1tq{li7q~5Yvd%79K
zH^O*!OY9OYFYl2ku;>%+Jb}<c7A&kSd}>5MeA)$mz&HqmJfJ*;9QM>9+XMLz%0mQs
zK}lrooKqdwZSrnvra@$JCuO7&Koqal<(>go+L<hQ*V-XwEEA3Uy+fyg#>a}n9~|qo
z>{v?pAB+KrYP%_<69{$!-q%XdrWe%pMge}q5{Rm3v6d%IVyt<kPKFiIaOvbBt(ro9
zy>`iBRdivE3QE#r)O*265coAm5kbvoR%Rv*@Mjpm^O+Cu(#x_i?fI3VfDoRj0E2-w
zZR(UJaPJ)TCf$k~p|y!BRi>KG$rL9lG9dk+Fi8Y)4lSOYk_v$z;a{9=fg^X2RgTC&
zSiM5Il!8EZzty}&HiEXaweYK4O{^~*Ha1x)_xsEzq35l)N9)c>`KrgsuWxtqU-T#I
zbaLksKIJbcM=!REYMN(Ht!SEEM3fSi`x8y8vFJn(-ZA&u_O(Vk<XvLB`DZp6ZWGlh
zDI)Rt1rQ(*p78-OX|Yx?y~_%ndLch}94Al#2ff9xDF8_G#)WA`8{-rKqxN$n_sLx2
zWGFjOzJbFr3!Qur{D|~Y%uf~r#~24#D#DN`N7z7sGry+RLj^Ece)t}i%v)C&2=bDI
zAOwEqgLo{4)jYmK^ezC!LD;f?n0Rb#e>d3uzk-L1^nY_=WB9kh!+&KT{5N>`kQ)zT
zYTz23zz<})1TPWw1Lx|SOvDt)6Ee{?Md7@JdlgVDGrIv2CDvN$`BV9uOq5^UKw3&{
zA%lE#Eh=Yymyqt(K&%77<2x^fwHwPxLOHm(E6hBu6)&D#?58+~-C&v5%H1ezv-^}>
zG6m@&#~k8jMNC@7D<d(!Fa#?}eRs0mpOn!AvYuR=j-fNo@SF4vhgYF=LkmN??*;a!
ziNN5}eD-6&2-Wt?ZBnfpbf9M@%{yCbb(^{3u4AC8%&5u-lN9CHTylfsjjEjGn5&o~
zdiYOD32+rVg)))QCU%um-WXyPg1h9RvQ-9fu%%Y7B}JPtSFJLgOr{5Av8NStVj@Qr
zY1@jJ-b8C#1v6RIk);rmv{ENinGl-+9E2cPM-+$()J$?37E32>IT^1P9by*LFBt6A
z1`a%znuQigYa=}jaT#v|{A5lHeK?C!I+%iD+tV`yuY4IFO~oH~ZITu!&JvWx0!}vg
z%*O6z7RR>@MG7g!$)ZHPIBxW>+Sox%q8WQ&KjS0h6rs~ACUaYhMzwW=%aS!$>t?3)
zoaN6clm5W?{Yh5v;y^u>6!E^Ix=$*ZsFhcb7g3?dE`8q3853UcGVHU)I0YVu7u5qB
z<|cCT%Vu<JJPTP#E&c^tU#s3{9kQaXlh9_s%9^FuXT3fS3QXAV4cGmL9a}9nWF<#F
zX6+rPj>`9miDuM^+BD&GsETczl=c2_r9NY<-1o~XgH~^AW8)hV?EHF;4PE)Vi*@~K
zZSivY`V3urs)jxna(8;!JCIAU)0;d03s3ZZ`lai11f$w$ay|IF#Zb4tLWutG?EYM%
z)hnCUx7pV@`&&DC<>T`<wKl%@bW&bzdwu0ghM$KS&{H}#@gOi0ysC0{c|@W7n%L{K
z&Fq(WuhaVeVCNg^+^v9fa=9nR*AXs2cX@?Jx(xm3RM7zAy94guSi$2wkZ$93aeMpm
z(RO7g@LwW7TVHo^JD-l*l<jZDC)#n7EBuf5iW=W7^yi(=y*JmUAIVDWd~ahH@O+&g
zho33TCV3c|83t*8g%>t?=g<NdbMxu%S~O%%F|S$$<`c|Je+L$-Gix$?Fb}R`{J*x+
zf32xijhm*e0`q!iewmvyTQXB)(O~gl{pW7_KQ`3=_4#*3$iLp9Z!^5%AS^~gcq$4_
z&*4;l8c<MCwBGkap1k2IGESbOAJs_PHeh}Bf^d<sgJzQf_O%m5vep^eHT3Zsh?$Fr
zV1krKbY=|@8UtdcVjrlWt(HGc8Twy5wTV{(E?*VIe+VufYR>Xb8H&wCL({aymNxVB
zUug;GttUCsoMa#Aw-)1zl4>{Dv(%w>dFIBS7UU2+#^BhS!K%v-ycTOI%(zC&n~N7G
z>?36K@F4SC2CL%{nmHUdkL{?SyOP=Sql$rsD2}}CL52#Nh@mAp^~81CfZPF~kJ}TM
zc`E%H&#Bkw)=j`J*H`ewvy`iy23Bu&@1E4wxE{VRmbk`mPJbWHO{{3lngBXn>+90*
ziDt|$MA6aD*tUuI5*eR&yusI?fS;>ZHsmR=4{*B6Fybi~Kb}*H|J8a#1H}dwgYEqM
zW?zrcW!fcZUzqR73dy#Nr(8)aJ?(w!7KY0Ygsz6%8qcpiA0$T#_$B+6kH5W{>v?P2
zo$sI0q3Gunyyt)f%({+{+NL;>RertQj89)(v10wZ0Ubz${>i)sbJv1QK+k+_<<c|0
zn?;4k0!+`AbPuaf{qB#igy*W1!xb83?mqUO`;XB}kG`c`-6_J_h~AG-Sv3iWil<)=
zXpqe<S;Rv0>`n&F0$7CV$KYP+cYpHtMr%%Fnk9eop=xyo&BTCU5yFBz34*3PjVDdC
zjTZ)vy%QtVC=EEx1aC_;3!2kr34&74-YNr-ziEYEmkEQZTCC}!qHgLDtw&-pIdmO^
zxNhb6RH<G^jSs0%{n2hCN(?AJQhHONSSBx@S&;<Hu8J*}wI@-JoMUGC8L%_vj!67d
zu!hIj>?CL(Xek|J_bD=PXLjVL<HF>u)LaE9fvAHK50#n5>8{%xyA|8~)hH{ERo75%
z>^^Y>bdannr0Z$JhziL!g`Sn?`_VSrp4D6ivSV~Vr4k;-u9~zQ#U7&oSqbsyRfX&~
z%AH0;i)t44IVSz8g6C?Cr}ft^a?j2W-6frJIS8QkFJZ@yb6N<uH`7=0(8}dVYgedd
zv6BQw{ZRQq4(uLh<7)oCOd3=e7}|s9080Hg8`SSP`TSJ^L~GKdrv{j^yY8z3dY}HG
zs2g9xEuox6Oeu%NLKEJD#X=LEo-8UOm0wh-FgMG;_@DJJ>Foa?S@~-N|5G$$3+R95
zjcVaH@!R^Z1O2N}GV}N!r%|*`jRPSRd1h-|0b{0p*nR#2M`WY^6@^&8z)P-*m^CuK
zEg9H~)KW?T>5{KUH>op|wA6)vpl^;at3NU#bWPTycgmQz1VL<CM)CQLdM&uwNL~D{
zZiy(el5v@Vn%jhV%ScPZ`XI{1gaIGO_sr^^TMwCV^b>7^I_jaGtsmL|f=>`>&ACQM
zd^@CxAeI^dX?Z%~$w(%9z{k313XTA=u^GQsAu^%v$kwe}WN950@-=U*g{p{0$^n*^
zZ+BPLN?E++jVlLqqc{M+9$36WA|Jmtj)^m*FTM&ADq|&Gh`nMZ9ltiILLkQ;Ae==B
ziI|m%<U18Fz95P!d?J3olUOy>&=ll+Cij|pz>U(O(T-y}(0)n0l!+%7m!X9W28H27
z31<}$2kLj~EhHp+C-s9d+C<%wh#H($+zlek`v<89N}sBEJ5^}o>wcDtLZ@mrtImav
zgs?NKu6yL^xv7Mso`SH1<AY;9am&FzQVP$VYSc<{YsqspyFST#3Q&fuZHl)O>#u;W
zjEwRtaA}mhXPd9@?{o*BWi3C##Cx)SF8$dU^AXlTxg?2}qs-U3G)ek!N+TpWyLV2}
zE<*QL9+DtlPq-02RV1k5-pXITws$3|AziLU-6iSi)W9cQofGXQO;tPEF}ZtZQJh#f
zMOO}cl0MZPlMAeSmCODXG@*sLk_b#p<hQ+cym^Kl6TYt9Kdb?2_}Frv;W#j|M$TUV
z2vk)6&;_kUv^$qH;Q5Vb%Y3V*urm+gEzvy#w+y>((D#ncckQNwq-7@%e*;(-2IMeo
zSkCViMV%*zQ@0G~RmFP8_xvp=|A+sP9L%hL%h38Sppc2}-!|-B_`W&+7DxEz5!v5M
z?zyP(9ba~k%yO<KJsM6}bSy%wAXgI^5pp~Jet)e5G=-62lVN%;#Oa1Xd04nBCuDAs
zFuy$rV;!0r%b#Nc$RY=d5UP7m7nguZ2nUhjPc?SW)W-ReBAKX{aeupClaM^QA!`Td
zS~{9|7(4i89UD>&lWRgMMug38TTvoLUf@3KzKaz|k6yfG+}+WMjSN60++&bi*_JPG
z@fdY$wq|08G$XnZOCD=p_f2u@<wJEXmRFm^RbNK@ICBg}w@hl??pEGjowBRdL4A2@
zdzc#yoq6quMTIay#TUmg2_m30V3r_}OLoWLVV<It2r>?zPRF1;6$6YiNuVu2?k7$=
zHSmD-rC{3Bks-IKTWf=uGB!#3Xm74i(OV1+9TM?&^z15?yTtbDzJ@<T46O$daxGL4
zXMtzWjavy>jj5=6uJzE?vx%r%EWbDCZkqoi&h^KB_{L3s$OI?9uNwz*X>eiin^xns
zyN{W9u3kn{1}~KZ<z}F&UP=ffW73CpQG_*@!^X)%Ao3wi=!XPM`G?dTGCYSLfTao#
zR@8{k@6Yid*BCsMUuqxYP#I#kzd02i>x#E<wE1YS5_WfskZV==E~BJ-Q7G(2*uAk>
z<q{ngbeZ1kDyC#}2>AsuY${q`^xQn_h3{4WAJcZ5iL*5{LhEwZKQwlW8NZyh)HxfC
zGdCuT)aCoiq=-qAchyX8uCF_&tB3^iW#~C?c1$p^!W12XwLu{W^fjcTFGHSXl8;v2
zARH$}-wcbrYs)akp{nxTkqGu@B}1Sz4tK>^=jd^Zb^_!eg=a~({edVca);dl?H_F@
z?-V;OXaaS<jZ87D3!{i8&>W$d%%lyQ3t^ey{qN}S@R}(ffRDKkdIT6Twn515iQUKp
zWB|-U#7fsA%cP*%L`IN<{l;PU+ZBdyw7bo3KiNq$@3o2(AZs3C_fh25M)K!^DG4XW
zMYkXuD|soc!5U2h%>B|DZGzz7!0n9G+M1Y9M!UbabPxV-xMJtjAxPch*#M^`NcxD^
z%J%?EOHSqibDn(8lzPm#k4rM)QF8sk=?MOH8j<HWg3D=jn8_&>xmdVuUrbYs6N<CE
zkJimH9PNaPUR21Hb@gV*2OhsgsB|Yy8_kL8n)EDA{+JJx_LccK-<T1Wa!TuQg@BR-
zIB|tM{hmea`UvpNw5h`yZ86Gwo&zK)3>q0;P&KbtiEN<xs#vS<;W}+OxeJ;G;50C%
z;%sy7YV>4oRjJpyll2bM<nt(AwEV!Z0HPJE_@M$2?|FFR8(88|^IL5ubJ{?L2RKSk
z#aeJRHBH4PXjQ|)X-zeCnQNN6Au(mLupRv<@%}J(G{Puw_AXER8R62dh4K>Nuqp3n
z;;};ZKr^B_G&cYuwI!>Gg-$xhVxDr3Eq4;p;M~t=&v@5PPdDJEvA&H@ZF)EuYdS{K
zxRVK-X+c%GEZCL2dN@a}X~c0STbjUB9iAHSRuvV#GQJ0#X)qki%|fA$5*}yLc^U@<
zWouKgI%G~$-=~v-c-nOVe3Ga^PO5wN#A1+Cyx@;3wH*jTBz)`{YUy`d)~Z}!fetS0
zo8g$v6=iS%ch1)mk_)POU)Jys&p4Ri9PoMXL@hEEVctqYS;3S1(!*eLN^y}O<%B#e
zdSeupwh}aOkq3F<p`z12`xz)j{)Xw+N6#9HEVCDvua>&;QFk}S@Ah8EFB4qHnVny@
z&jR9{*`l*#Gl2+0Rwm`9{raO44dJ^9bt;DAOeeC7X3Zr4orD9zHZI#CdC=@+I2ZD{
z2X$H<pXzuY@Wn9d1^;_yGt+<T)Bkr$D<jjt&HX&k9COSNN7;CzQeRb3E#8Vfj~~sa
z97}Y`LJ1~0Cnw?;7|vIo7?P7}Fw&Ko!yr*DYUc-Y;5gcxdzyg&djf>v)#Z2c;_7mH
z03<+y@B<O(d8gHfAs+_{A?8;Gy)&_QdimbsghzNacBMN&&>?#Rb#$f^cs&=lUz>cM
zEP#8|6o7DR#zQ~|+&jgvR>5yg!!P>!)#CF3(EpVW0by&J+9M@g^P(AE4QXp?J^3wL
z=k*_}_w5aJQ!sXiyf#4K|0e4>YWfWpA#XdoYZnGPSlkI~7SnTav+_%Vho&sbSW~b6
z0>EI9vn|NRp2?<2570UY3SK1C3YEC084Hzy^{gxWutpU2D|aFPaCSKj&Y|A+VlhoX
z@_P<LJmEi?g8@xE4LgYp&wyuaf&wG&dzFPi7&@qw9k4K<r+yW(C$+R<9DrC#HLPUS
zp0XA}Q(as{6x-kvBds6_AfL+r!KNp<7)iZSVU*CEPs!(9n5i<J58u>)r|Se%oSu9A
zGr&G99f(qbFFV_y(%~eytP2q1H^y21OcAS<{vEsKTYe!wAv4SBtT}R}f|z;brE}bT
zBda`dCu%Z0x&(N_rs_6Z4dRrc)2DRB=qV_B8Oz>WgK3`{<t~5_<BcSXs3+B%#RoY~
zp=2x=RHNiVHHA`zPEp1Fw_!lMg-E}%-o!yFl1U2UH?qi0GbxsDG^zxz_|3UhBBF_o
zp^*8wsb;RiaD<ft#bDZ)zhlTO$1C>Z1fb?p<1KT-RVZ}hYI>r=%a*;#LLKGe7I(z|
zd5evRN8F4jb9~opw2+?EU$PPtoqplLa)(4&%Ga#0xfgRg&(MVK-Vnwksib!4ZiJyy
zz97s4Xo|<<FH)uAHq^d3^B{rWvZbGk)8UuMKYSRH_~_g`4wcenBRt-$%k#7-14jUo
zh|acyH3A7#M-UUg?)^Jt%XNmaeeBklc)}Hz*|S9J@Mi|oyXzOo#S<7Zz-^T{-x6XM
z)#ii|aw|T7B#yk!7fR8`a(S8^zDwzR!A)uDgW0OqhL{AB2GQJQ>mkXtH;cFLWe)#+
zFWG#gzZG<p#5B?}txh}4b?RHdR$+OuiUPbZ_6O@Ms7cbM9`27nG<k%wyA2p%6KTV%
zHU)s_w8^e`pFG3vTCdMy?$1vWM~8w9n_kgXTE%vIgqFsPE_u?2_hMp{p7%o@H$dEZ
zMNeU3jwv#+&A54==Hbk^buV391udp78GMctQ7Ph8$S>r+>=tWQnk{{NHk$m2r-CcE
z4;N*OK(M<jAFDH+Vi$as;@tY0h%%hPX=P12gi=kdr8<Rp72_SEQ#Ph+RcSatyBzUO
zYSCh=r28(mTX`ZbZ4r~1>E}5`8U4BbN@y-`DFe3k;dj@pYlUO>FpZWWHf9MMHWv>)
zH@(8xn9?mFC~yd>s&zsls}s(?s@g-Pj}^zf!Yy#7Q<K;~UVRL=_E{!m$@2TyIBB|q
zu>+(L{F;wkqV`yF8BRRXIEL*uB_=?t8kV2H?v!<%q)#FMN=~~1+=gfAHEi}OjpWso
zU{39yb2fuq0~B~x8kvT-W=!wSG_ChH4EbL69Ox_8s{k$9wcGOFC!lkpK$yN|^R!%>
zvg+RDA0O|#e2+=gxc_@ZGO@G&+wdp@``_SE4wip6>H1r1`IpO_&wsegi4~A50!0C|
zbldo&(IfFSu)J?`MiQFWU@yb{`h4@5hnXUU=XKIFhlj$@otzvegB###Y|JQ(w`|HP
zi9jZ*bwTJLrtm3OvTWq+k!a*lNPu<DEAF}5Ne<b<nngmb<$9MyDt-3MCuXbwjzEfL
z{Jq;RvRs;;YRyX)Zk0CG*iX2hNNUTPUEi<j^)iF}RHkm^x02h9JxPJ;2=!I#J4)_W
zS;h(KhIz-aKFyg8oG6nOX?vXmvo(I#k`<JARaQ@RuEzy|6z85oE-|-@LN>Egt{^kJ
zq97h0wU}vHn#8*`J)cTyf?$?$^*CRkR^g8=GqSWDpB~OPgwPg=gR-wXytJL3*dM16
zf9ujjl60XXvr5XiEPrtqoz<W;gr%=N)UG8}$9N@(Fu3oQl(2B<wrrVj>4uO;F~z3I
zXth8gbJzwusz`SiX(^YP?kk2{>g0+kE}^r05M^_2Lul0YHkh=%!Rw9&2x?q{b@?jz
ztRs_u<CWj0IdLS4wFti0O%}y)#cHlZF`);@>*6bkrLTOd-5O41bcdR8MzViW5YTV|
z2)x1ug29i8-77BK|0%F+BtOw5UW+Aw`@7$Ae%S%FJkAzgpQUc6S)Ph;3-gzH^w>Pu
zfMq)aBy=Cp9aT<ezdlkqrs`g_MBiV{3$o4Bpfm(KX)M+91|Gi!%Bc&63RSCYV%2h+
zvI%BeR5PSQnFZ_h4mSr^kB_VD%ej-W?>m5;Yw?xOf2u9<KAl}(PjC93o_=hixP{(6
z-Ve=X%j$LIx;wbOK7M`_-xoodxq{A<sIgTQwWu{+r>A`<Q>5v(ix*$bp1{Xe|Cl?;
zxNpne{J7&yc-$(c>um1p0@YR?InTC<N5xVoG;%&FGi7M$7tt98ebO#Y)pB_YGN&Am
z6SjBv2VZpNrr*Cz2K64d)f%YdQY2Ty>(JHDyA~u}9p0_VJqp;+FK_X%H(xoe1OeQe
zl6P&U9tM<*<?%1<hCX1hr3S-92+SP#HZYAVAGn6?EbK!~xwql`_zQXsF9f@~zrKs*
z>x>T7<>mECH>calLSP+xe3o+HYo3dmvJ<?6s?OB)cJun$;CYrVR@-WqE?#=-@{;_{
z6g{c#p58dI%Gu)cdOfCfW8BqyzPUMqzq;bHz4ZJR)je%}U0;?D>CK`<;HNVyc^#Fz
z{<voCwc7ac8|Wh}*eqcCef(XYj2r?k56r=*88lZw9f2pT7BLI;90asSU>=`13@r%E
zK;Rx9DD9sLIPHJc{}k4JV?KX1j3K{k<=0jf1DSC3xk9DF7_NtS<$gppKg?uV6QiXt
zhOzhXEzG2KE4x1$!NOQ?N>utFC!QdZA;h1o2Z*YHG)s+2?<{DoVUUyn@jw&|I5c<e
zol(%-FfNy#k*8q5lx@j!vXccyB<-ND_!Tzr+wKPvbS*T;o@e6<+W8w~%rq40ZBSt1
zxwYtKKV28%7V%ss|9m0N#Sv)Cl<~?xz`Iw>k^TRNv3Cs8E!eU}%eHOX_AcA@F59+k
z+qP}n#xC33#arig_v`N0C;FbXGUCgO_2-NHkz>x8V~%kNO8k^C&12sR!)}jFR+r*&
zUrg*7a=i<G6V*ghX!Xtm-Nk@S3gY$bfN-aW*ag9F`Zo`E6+KIs`r-({xJs8Sh@cM}
zwv!<24kn-sbm2xWpiDgKFhGb!lOCeLJlRxmZmbN0UcIy$Mc5yc2oxs?^={=?s)s+k
zWwOh-!GnWfUfxuh{7itpIwjQZ?1Mb|y`K=bC+k#MbW~T#(Iq7k`@mE2Tha3oV^%x|
zOW6?KphAQxSvMnm@exoo1m4&Ylg-fJ++{X$?Y2T?p~>S~zuM9@bvi^c0cx2icO^)^
z_P|rL2qA3lPOk90HcRUUywLsOXH~WDpFO26PUj_l^$DTccRIKFYi}Gk$QuroOW7lA
zXrT13wCQ3Uq)?qGQ~B^3=6U#N4l6@tty~4d*%zj1*ySX+ObjWFKG#7x!pYRVFD$hk
zKUHnrHHi3=G8cqjPuiC_%?hdw@&>RK-YGd>i=7f4GH2~ckR0OvU~EBAHR6qkLKSQb
zpq2U^h{z&L3tUMV6!l&c0mI`b_@SV78ojs*-y<$KE4@<bw?Rmo1ap&|2tZ1k;uMo2
z=*wvLeoYc!)}VE<L@1r|qo|;_nsk4V#WF!{TiGn>kxzx4Tb-FT`p_cXL8Fu_3eKx|
z(T!Mzj#pFL;+}YSBjT1LvNC&gkr(r@CyA)$;)V@CJCdNU)pZWF16LFsvDKFHq+E-L
ziY@7YpEDv$rMqBtF*QA?t05#%0@&sfNf5N#m$0P*xUH8@=pgY(wc6KxUe6DmDUPfK
z;A#*?e)>S0(E^sCJom;MZ!7rNnPsbP%RJ^~FFd%P78b=u4N7SafBuSF2q!9WiWYw-
zO_CpiNG~EXqO7@%u&@#xLezY<+Zyc9=-7U#f(tMw9&xR5#2TEOQ>pv)yS)hQ8;pMx
z!ct@tIb962psp=}GRy1(Pp7||@mTZrgtV*a9;YpTXT*VIU5%Q|Q~E%n2@zmOZCkzJ
zSL?&&E>+NM<7d#?_aKkf1RLPj(?5R5zjBY)eXPuLy+^#14f!_yecW&CSED7a{^uB@
zjbr^MFKyiad_4*Nt?iH#4m|4LM9}`R#vz9k%_70DPtYHd2tfS*Ek=A2{#<Qn^etj&
z$#+el{=O`Zel>BQG6^^ipeHr@{TZP<ph)%%r}c<bFzm7=rGB<7UXT9(QhI!Sa*Occ
z9byIhBg&wNBe}QuxRH~zZ<p2M#J_I^98jeto@LEpL&h`e^2(BZ)rA{6_*g6rA(UyF
zkKm7QXKz_fG@5uhysbB+#lK~)OFui+_QrW~pc+t(58<XH7Bma&KN$X|NoV*g&U5PE
zsGblOn*=?R`wd1r>tvl9UF_lVBt96w{f<(?&YojlcVZY%prE)OJdT<Ri<*6?OY&n`
zyPKHx>EJ`m@yQf1_(kPdkH@N@t)SZQLXL1cBm`e+SB_9;eonoeaANICZ~8?c520a`
z`uM%W%L_v%83SK;US<9X_^Y}c>EZJrC!%IGyt~vvHT<At!h^G|T$3L^t*fB8;9`;b
zhpt?TLfqZ4A0?=DLtfTz+1!{*@U&g86)6;Flv*nN5OKZ0%yIp|7`SbZ@0#j%6hctW
z@;#V-6-)Xjw!6^LYf7JIFbuyr&W=Afy=dQ5fR2PsHFed%0#oDVvKjSqTETYQs2jy_
zq*rRMthq@Mf^j&D*@vH?tW9)@qiWLnEHA8aHYqgM{<hchaBGC^p1e~ZO={DN8~)vn
z*Np?GhW~SY!WW>!4<+?~Mx`taEG+*sapz?F_n4F5|Lu-;^&{q<De#ZP-6UQ7&S-~0
zN7aHzMTaz{L!4R|DIpI))NixneALyA#K7%9f7E>pIxgMcbB6bYkr`qO2t?Mjtj$Mv
zDn^c>q|a|H4}-=A;6xq?j)RTR1c|1mf4CohNa1iz*NX!k0=Md6viC#L<4<jj#rO86
zfmtQ`AGuI|bUnb7^CzGwQkJyzUS8ZRK5jmPa>50Hbiy@15JzJZwo?k+{7Y5-G0rf>
z_sI9gMa;G|<QTXx|40MQt^?!%3CPVXYy`;9i{2Ffqt1F{EnPS7#Lax>g!HKgiP{(h
zmMjDfVk1^RM90dgvCjl?lr5VPG5AYPvh&zRvM_alfYT-nMpOi3rNY<N1KpY~(Ugl*
z!%cWzwfJ~%dfCgNyKDkX6WmhXmF0tKhMA5G?bsIdo%IY`O%jaG)}G^PG?|jWO4FD}
zwS1x^p(P>L{_r>F)l36yCPw!pJ91&|a!BPtQRqE>LeCj-DzEV(ETUk_&Bt-&gEKak
z1lZ8n@d&ZBJKpAk&Q%DExJ4HrvQ2>vWNZ|c>DKOOt*U}`Q6KRGG)@}#f{mv4okMi3
zgw|NyP6=;ImZMT35a`5E;X?;oRBpjsxV>+A9+Ye)hIy6aQ=-kx^2P9H(FM2~%?9f_
zZ$)n_3toR&^fWm9O!`rRN0>Egpimgc;?@y(HB>$kPD2G=)bb82J+vftfp1{`tmYv^
z#xFN2=2xszxgz~i0);|8uMCor!K<}(?=m)Dyaq9(GR`zA*lEnZN>Dx!?Kp@ZM<Ghs
z8}^d~TQ*ILqY{O!yM1r(A66zpV4(zzlA>svI(dFpQWi=)81boS97NZe0850_yr#ge
zvyqU5ir=DByD6@z?}qSkyy%w$RVaHE?U&>9mj*RO{IGJn<#z$P%f=<OrtY5L`rxGI
z9S5&^6i&2XS>9f7+<$j=9VD`dwAqy<dXc-%wOx5hnS;jbF!g?QH@3~Ty=W4QnD@MZ
zAL0p&Rj|t9zmmE1gXepDJc^HvXSKrqf$ZBK8diE-h387RHI?Zky0eSOhYq%#&Xr^<
z_oQ6L^DYL<UDGx7&U({tjwvbkZu3@+vq1u*jiR9B-yn->$rs5d>h%}&fw)P?G-9cV
z<nLUuL+9>`un7ed;MugpareFcejfQuV6Ynz&F^iMc$Bq})m8%UIej5t+(4Zkon{vb
z+<6_&IP_21#MTXuSHMZM-cQTJdgPiTDYzof9ZYH(g>hj{X9>vyV;m~kFu}_G)5nT6
zsd@B~>%P|GFzo;ASom>s_#gE=CiZ_PFa8V1LUo*s)DQzg$jv9Jqg8(ivpF9xluLi%
zqiU++3RQp=Azj3mN2xh{^<_bW6x?{+IQJ}akg32L3C9h)v!b#vsk&ebNm;Id<z6wE
zjBMP!JikyPU*1USC*dTlVaG{CO=A`bx(>v)X=ApH&~CTM=37nKQrao)h;VI?eZ<Ld
zRb_o3pc`a0VHCl`U!O#QM5*ZV!gk968Y&5g{xoEJGLNzDx(S(nJ#UO1%@Hf&C6ERe
zV*Krt+OIOU797T33S1p^n5G5k7@rF?F!)!Vz1%Q+R?WEz)Rd+mn(zFSe<M*G$R>yt
z(UG_)e`nc+%*1mxj|xti6@~YzGLd6bk^a>$buVtItlH%I3nqL!oAS^0n1S`b?D7A~
zmty!Y;!8|F0-^u+`G23y)0)<{IIW0ZHG2I&T4=W%%wC{T06zX30I*bRqVq$~;38a`
z7CPB>T%_hc-nn~tuFfmwvt}=7fS1dcbevusSuebl>zJlRMoc8SkZd@KNQPa+q+O8P
zmHLKLlER6&)bY~sWx{NSh6)uJhUDam5x`Oks-RQein_tDBIrp-za*SQ??N1qiUSNz
zoXU}4m5#I%3~DM<tk$?ewIcQiy!M?LM$5zjlaY5kPC!vw{7o^YP?CYD??fm8?zgoV
zq7RdSq7ECc(U~2*rHWF+SRq<bl9&w$uXhYHYO^RY@H^VuqPI7VysmflMb3`VkhNEq
z*V6mm1{eh!`VC-Bkhzp;$cDI5315^?92f{)9Sg&2GBc%%<V}r`Iae`<g=e{bWrY)-
z!VvN{1+fOf{yA$AmZl-3IFfPDI3nsNF-ayF${`^sPLSwP_E|3)OV(qgTyEV-PjV~<
zdh};nxrM240K1h|+)WtQZXW+-%T@l@$f4{6aQ;ZvVoiF6N;VXobF~|ku|2E?x}<%`
zk5!1J7-};ZW`g=eUfSQ#KLVt}FvbdsJR6&LY3VEqLjY9=m`j(3nvX&aQiX2ZA1Z(~
zuZolm7krpzTm?$oK~WhJqOML6(&HwgK~~=DG$IGqD#9*~3xyjP)0swC0HJXpnACr&
z0~-fNkbx?~Lb#t2shJ49lB@ErjUrjuB-^5LzD>IseUs4P$xfe!AaKCH838fQ4FF_(
zh+Hl=hi2#F#KatJwx(7sGjaH$Gnf78t5$9Gn@fX7ubMvfqV*hc>wQ=cZI@oRtJuS_
z!L7M#Ev;3^5YzC5P!j`midsW;Ty%@sV2k<Tv0S5zEt0=>r)p}y`0w%Y$H9U_0{L<0
zjqL5sw^)CARRv|#{D5q`Tcz%VRFjC234)zF+pk&=6uR}a2Mj1*5R`g#3NK(nO064r
z8wEGhJ#lc99(5{n8x-M`KTw`1$}})uWGG0DGIbDiMxf6W>DEEyXeU+h24u|$hLyS*
z$XZCvN?XJr_AYV5VGbra2tjU_Pz4ZHydO;aHOmNt@sD5j2q;Go6hYSEJ_36ZHz%Ti
z;8=+Co+wof<3Hw(Gh9Y;%0p;I%nYiP3hB5;i6gQkNyK@5=NWJ;C`rR_hbN+lf5^&U
zD2R*_7qlsOHmlP8K_ZpK^{F%j-NH<SsRxRBh%`88_Vpu81|ZHdf(vl!Bt*ANBo{Nm
z6`Cp#*Z7D}NL113m<`q@`pqo{192z-Cc_S7?ImG!e3I?a%*AV=Y#z*9j!I$+0o|p~
z3(Lsdj9FR>-Vt;&TFWXiqV&>-`F%bhGpvCc9xpm)b9Y+>ODivAj$9;*;MSc7<+@EK
zw6NT%C5bCs(^@#RYUy<80T*dzbaOI>GmmVsAADR-OaYu4+yJj9FQ%)}P;^f&z--lA
zFCvY-N46XBQ&QDVX4QJrsgPr<c#FckbFyn0vG2565-WLZ>+NW1Q01zrTZJuzZG_j6
zmcWo%1F#9BZ=)fi&9chvOGeVMOl`8-?T<&&vH$xdWkdtZ)I6)${=ZF5t_XW`Mb)I5
zG%>-}tkmxMN`GAd37U^iB4*6!w-9(ms8+6k4Ge}U{F*Uph7Af`?qKno6Ou7O0x9<M
zfMI3bQVn@8(5_0^(o;y22?tUkEL}o1q}YWdOv(SudF+YGgFr6PWc4zUp;iXl0QjYo
zw_O*N(LA?BAs=ooRw0JiK+6}Ce=8=?KNI34ho)_np@Gy)7s&RA0k>i-CWyFBq{xCW
z=R4VW2EroG%R5L1q$uY?5oD1rT83y$m%XNGPU|<IU#wa3m%*giUQxr$TiRPt_*_(p
zQyak^2D#u?(GwKq4}ak3u2DP?KAlL3$h=WMR;w3%WMFWgf>AsoQydJd0s*jW!pVRi
zl{7ysK?+f1A0vo00cOy`;iq&4WgWZ(riQSv!0IvNpsmPfmOk%nZ~*Xrh>Ip7z1O@P
z4bV4f^R5)-EGhIf{v$NFAQ@t@hFyl-5>McL+D(kl8}fV@-J6JoCs8QzTN1@aI7Aog
z>6b8^*rSATeYFy6z$gz&1yLLQ@1BAnVka-eev-xRm|EbBdU4hLM+(i_=6sGdOWp_?
zmWq@7zup{WsZJmWRkGS(C=ln;_GAi=v~J{V7Q6170j3o)x9LWJ!A*#)oh4}N7o#`&
zfj;D1X_`sIEJ@<qAPTl+8<sLGDVp->U&2kRn!Z1rt@uI683lQWg({ZQ;e3+|dja`W
z<0k^RX0F(P0y7MJKE?P&I;Sj5VWAM_)1Vf<No|OZ6nY0~y*O2fF7%3q!0H(ti00+9
zBlv}Ln5NQYjjtEgQd?Mo-3-qj))%jwhGa28D~fdxR+!Vkw-A6ETy^h}FwsLwPqCi9
zZ=#3V5dS6ob+k~U0HxG^M}I$Rfrf2%lqcxqFPlU8U-lERtw93ukLV1VexS}764Z5r
z>ECo=7C8OWmN=bdbbQ&nRw)<!>u_uAHzXn-ayMCyzKfIe9vbJy+q=3?vp(Jq_jb7#
z(Z!)#J*$)4C)r=#HL0CBtESynkN4K!mv3LMi#}f7Uh}cFztlb>H_2nK?C^_vUlOS^
zC&N{{f@8F+YiO^u9;(of>L#Z@gKgi*Pk2>~KW}o*$)ahyPFVDw7Qfe(&_^|N@M?TI
zChxoWpIUD73@>#1F2dtjQU8cF-Fu|&jq331K36d1^<0E%^T}U4;9aTEA3!!o;Ir1$
zu8q*^HAg?_;8xS{UqyH{`uaM9vv)$nkFw&NpME{#EHZb0Y(cv;cLo~F!j6Vg9(IXE
z=e{h0AEzeLU*gk!be!D1+p}Tg)%-P7%Vh7Ay*Lm?$grPCo5WInhBt}6H`D$*&|l+X
zeg2{|`D)mY>-c8(J^scR@0D%rn`SubdHE2!edV>`Jagj7GcpKacLy4z<IB&{s$g{=
zcD8kiH@@rx5etQA`GtN(-a`0K!YIZc<>CJfRsWYTij#x&|GQRUWMTc6qp0N{zea5K
zf6$3;({N}WJ^=y%WEM+66`RDPa=hY=meCBQLP)vIeZ6}PENa>(b*?n2$$8W89LLjb
zb+ZVoLD8d_Y1>RB7i<TEV1+rz?qGvVBNmbR(<3Z{^g|Z0^3&#<M9oZOw11~R-9rTZ
z*5%HCoZ{A`;Y7xn5|EI$&sB!;9p)TMW}DF<OJWmtm#6cAmBtXcOaeIKAOV1iLozK+
zw?j+^La(J!fKyf$_Lr&~5^*d;aSpJPh!muhXH(1=uYEIzp0F7616%k5Tm~?KXnIPS
z9i0S%8rBGysviaohRH+7nnH>Af~79+wT-PmAfT6DvA`q{k}QoWRz~Bc-%d2aaAch2
zm?xbTNI)kM5?I_$NY^AIY7Qqw5JV9cv>qNHUX5zR;l+?u7e_7K6bFt@Za(wN+;ktO
z%}pxi5#bP6V^GVDTXWh|KBDfmFDk4BDhw3SI(S$V2?mjcnd%%71}JFM87_&)oFT1V
zxF0=lECI!A1X>P7XhMlXqK~zI*cpOq<_eKwlqMLe3?Raa05L^19=td%Pbf=i%Hr2Y
z5Flz8esD>?TRehMHPIp=M>X*j(nw1V_LzXvh5%VSL(D+{$FHJDviJo?a$-f{QYHUq
zs+p{mBJi|<ID1)tD0?_<o^^OeP|1E(r1>+>;YMX~cv-mQFj;z-vr1Bm`wC$vkF#)S
zqycI~%E+mR4s&>_^N`cWkwhC&2HRo|kEe)iFC+4sl~PNW`;_iAn&ZtQ4^{=_&xCFt
z@5g;<&PaUM5l&yZ!Hf`SteLjs$M=ws%|~m|K^6rmtQw@|;=P;E&F_M}m=ILhhbD`3
z^%`9l09I~)I^<n*AZB4M*i1Tz7RWtd*huYTM%b6XTRNJqUl=Pa>rD+Cq_C^zpKNMZ
z1}9wYM)7Xai*L>u7s_L-sBdX*GzmcmTdlbZs|wW@nfhPioBSskCeK=|Qy$6SqOHkd
zDnu;F9Mi>mhg&P4w!0}hd{6@YnGn7b<C@TA8irmfAx)n>PRO*TLrhoSEIsn;NUejh
z69-M!@hQ>|UOf2ty%zi5j4wMhtp;al8hE#Bs2=14nU5#OaV1Q&C&tNa4}sW&aAUiX
zk0*Z0AAipnPEr>$ul8LQ77f%OXH!wWe4i^aZ5Jx0PbGQQH06d8m&Pm>$wuAcLDYKG
zNv*gmG^+tSWtG3#B;^iuxuz{wCV!vNu|E}(6@|JiHo4~*sjXZr&z**E?Ec=Zz+RM_
z$|$j{C}s_2Ykv_1wS5Ni9g(R-y5LFvhQ=Zu8E;ldYW1Y-bzea$kka`EXj-0<p=&hx
zg9h6%(^4bri!%MGo8c=rT|>9c0(R=~x7X$`{#(>Q{==^X^<sx-xI0hNftMFQ%4Tr<
zdzYi@@@v*&u5P8~^UL>AdlB35&WEOA7jYXXzft!I#$u{&rB;od4{79Acb_lYht3HR
zbhu9toMt#P-L}GN&0iJJ(wo}_RD#aSqI`<dFYhbfIZT;@e(qzJKuNm}jjE*+slSR^
zX^UTK-NUsGU8^?zV|`b<QO+gL5xZ=p$YuDId>38TQ)J^8vo@)2YF64gTXD&?@%?nx
zpvWiX+*#?KigJIbR9<T6Gij}VX&a7eG%8HI>702rH#N<2;YH?)Yk9SUTZ-A<rsIfM
zD@?s3O{sscVl|tX?%Z%-fm)m1XdVr+)G|&>_2-?0@n)oNHwUyT-)JIlD>7&`n5KNw
z|IWdygI@(%+b4eW^O#%O=ac*1K=l37Gl8~y0?xC+_gghdcYl}k&NY}k#+*3A^4pz)
zx0sivn(4Yne!8@sDKZ{k&H7H|+naE&^vL&3=cs#?hJG>dbD=dwh5f*Tf2-Vb_Sk33
z4r^<MUesp)#na4V$7lEI?*{+CK^JxN{(&y%@h7Us%I0`)z-I&bKfwkj*8fUN`rojD
z^FK3eMs}8eS>E^`RMRcCe|Y)UwSw4L^+*5;`sGP%;sl_{%q7y_YhKV;+;nteb5p(@
zF#qx_#|<wam8#$VmBuJC6Nw)>xMyKbav&}BJ1yb0*I1E@geS2?00=ws8)R_~I><<M
zpQJLKm17c1l_NUncYxzN48?2^0>^}X9t)gf0OYWJmXZMFA=F~f^Vo_8Srm#mAo}Rw
zDiB35mfH+TDikGTk|3t8e*@7YW**DTt$ts0PoMI6UN?POlfM+9pF^1xpf-;*uq5*-
zJPL<Dcpwyb(kemW?H>$4P0AVn1t`k?BqlgYP?Wk=vlJ!@HUIsIGyy<#OAbX)dAxQ%
zsU%nb1Zh+pBMKJ4A_M#~LJ$YEXA=mPKRCs(iUeaAMMmn-9|Q(rjs8GjImD*sX?tyW
zP$>%#*8(ArQ)PLP`vRakN%|vphova;Q0)}cQBV1^Oa}pWpxRK8n*PYNCu1Ux0DvCa
z=0Snf0$?1YRF0K&95}{(2zG*Ld;W^2{IJ?I12UL^NyB`0olpIez-9Mv?SADk;Gcc#
zG1Tx(e%*;ky_#?+0g8eLPPZI&4jY9feln(k>{XzW`RsilC!94Yw0=U;R05@`V5|V@
zZhr7%08SoCc{4K%w)5*yb7$Ew-gsr1n4r{cLgs&F0=Pe*PHBYa@X(st1mW5m<NWc<
zR4K{@z-V+00*-F;0bYnm?OyDZE9J$(YytT|3Gg)rC<uT(ptj7h1gj%}B!rv6#Gq{i
zeF;e8>LPQxJ%(fW8VdE)a#b%+X+3+&O*V%u+)v?_p`YKL8c#;`aXuVJ3EoYuRM)ii
z9mA(N*-+W^>N$@yMgd9Lw5(RvXfz8Y-)LbzaZ1(d>AkYAX0=5?E5-`ZS|hwO(J_&q
zD3JLT`X>qKWHpER(U+oBaJqg!dctctn4~F~C_idl{WVtJ<D%*f*K=JwS@Q}%Fx43`
zK_6(Lb;B;TeHqpfg%7qZRq&v7TNGMjg4=esEt@n|*kke_{`hN)<G>{v$q9N?<H^?V
zlQDd!0pF>Bp`pa?)&4E{-h$^lZ)|4uXNxJxejx|laPj$tM4eXD{XU4OG>UC{KQ|NY
zlzd+6HbzGE;47-3@?{{&?~|+BxWIZ{tRwaUwkIlN2a!mYC{*R})-^=NRw+kSAK@Nw
z<}apmf0S$m#eRnDqy;~n)R{dD4_MCV5$Op^5voiR#Q7|71T|LH#96eY5v8x_xEjkl
zX1obc@HXqawysU3^LgDn25AcBPj54j7N6~9FUObsBk&fN{X~o0(VJ7J%~kWro^CFe
zifp%4n5N5KDe-0wz%-vT=Xq-b|Gw@LY@cw=y}zUt{jR$5l~L(^G)kGO6K@!u?X@>Q
z4KoZo%1;xK*~9f>G(i9Rb$XV&^Rw&5MS;4>n)L4^9NQ%_AMShx`o_(xJuBq{>#nff
znR4FL@TNvDuf!*}M9U!Ed5gY6Oyd!3)`sUFqN~5RQ!Fnc$}lq(;*f744*z1}tuYVL
zszPN&vcVS<-zWl+^FUWSx20`vu)z|)LNT|$x|-aKQoD-k$q?Ye?MlGT^FXMrdDPTz
zrTIrk2^q89m5>atDVgk#po+21O&i8!Q{|curmclkc(BlH+a1wxVDCg{seLmfg^ie(
zfR2I)H2A9Iz^+TyCWY;)g2IcUW5ZJ&qRQwi@wv&L+Jhp`)U^uEl(4|MLY%d>prj^J
zr3Y^i%k?gL_aC;O(6oGwQ~YQ1f?#<!XMfz5t9<SOA2UfMi8qCv)Jy0C)Ydz^c6gp1
zr?J&G5OwytB$cHUpJo=z8EX~Oe0NbfHaREfmM*#MY4P8<Jd>Kq>&4#iK-ljjD8zP8
z#b3BoA6ePJG6UbUxV(dw)jksfK0+mU!nYq(-A(xi+s}OXPlLL#Bg3!eOZUAhJ^5z2
zFf>Q6hn?9g;Lul>&W{A@ucj>R-Y^QIXN+ml<FA%uq3O9$bv?BGdavGFd}J>^fz)in
z-%>WD-)!4g!G%N|&qt8;iR*iQwf`lVks4}X)*K|U(7q<$nU?l6S}9k`bd*i(nl8FW
zI`u}Uo=aVd&Th?#DQvWl=e(-Wm|Z9zR8gzxEGqA7G8@ws+Gt(Z>}B?VH|L0~(!^~W
z8`Our6(RD*{dlN&mkzjd`Th=kXZTCc2F@!ap;yCiY*>eR3)PJC!1ov2TZB7b%TdrW
za#QuwD8kGZuG2^9jucUoJ4gBgq!M<@)=6-TYnmK}j||m)_i;;3xKg{j7T8%!Q@~2E
zY9$<Fc_)e$VCBc_507+mNXd0HRz2lW%@wzI^cAiIs$}Y);OEcv;J=t#asJob>R+d<
zUH#zaf8<sG&C@Ty5DsQrC7=szW*b1DUK*1T#LD#I4q7Lxt|0-YshZcC4b2Nsr~Qw9
zI5S=k`*5zstZGrr)$PWTd%mMVapE9kO@KmnWYMBROlP=AeQi1ZPft<CD_|r%j@mhZ
zh-r>^CPn%=g*cQNb7M5D&=;hGAn;|<qSf(}%!I5FiZ~Rpm68kw%0(#hD}G+Z=@2->
zl+Z~+KhYdz0UfGegd^6RRG=5upGm(K3s5x5Ltx<I`uJHU^(aR?iX%brK6xG@h5$<c
z6$|3@1hi;ycnVDU21aQ@@jM*)3zgX<ba4$hWtOdgB!oFOXuLzH&!8zHVZO;ev9MvF
zMtcm{$MR@ZE`V#8YiEEl5$-tBwKj{iJ)Rj2lOc~8W@1ioTsnr|DCf68Wo=L<=>!K^
zU{oGs*d{7b<S?pINhiFG7u|UaH7{WORgo#+!&Pc!z|wX^nwT{5crQSaN^sp^qTD}n
zf&$}N{&@hT{tiHtpzQ&4j0gZH0s!&^q-A5^l&Do)z*G$Mr1y~GR1i3d3GTQ@{*9j-
z&IlB?^K8U8@(6f&JP&w^06=5@LdsxD%x;Ik`To%0TnwD3B=quV6v(a|Xi>4s`u&D@
z7=CEl+QCCXPBD%Iq;^}-<ETPQA?xGqg3N^Y{*PEq0zObR1q8HlOQ+at1g{$)uxHcr
z8UlvNYZPc7&_yvvp#xqduwo1OxOrlId`?jc;%|1($FQ9rLv*{dKAH8C)Xx^2I8~yJ
zXTc2UZ>*m0&zH;14rpIntVMG#98q>@NX-KVVS}uzJjQ7Uc5>R2tJZJQI=ihLQ{v)D
z?vcOz2Z2IED~<v0QmaVE6goCyMpov%qB_U6JQW#<Swt$h>7*+lN}HHvcjKBwk}qOP
zN-2K{jwEqyUfDTw?-XMYvC0;UseS51TAj_2*3+3?lSMy=bmAL($qsV5>AHDsSMft!
z6C4y%d|Hl7G>>#NxYw)ejCEw-o4YlgQaw%hIl(<V**^i%n{+X~Do&{l4hPr4f-0~4
z{<VIIi4Z)l8wMtBEGoB%gulqJA-3NfkSGQgI>@df_*=76O0GsM#2srWz`~7qF(>s7
zb+rt=<JgPZaU-_f-eM?DJG;#?$%fwIrx#Lw3u(1&!VWf8@#C8i*frh5Mt=K+I=oYg
z+DGEo6$DE`vClRfd|wB9j2D!`ecMO@ez7}qetQ90BHZ~l>&iQz-BZhw&hK*Dz2xso
zuXn0iH|<v38tR6dEXDJ+;KZ{j-#jwI9I)o7XPU%@$gV_gL2GQTT_W1NPU&0@jy%Wh
z9fKTo5xVd@X2R^aDc?#hw0HWEDG|d}^0SfY8ynlz64gKXa3APMUYR4!gdyBGf!S#U
zuY%U6Y8$hPzu4)_4`LQ|!_eD(WGx#XDVxv9%*(`Pk}htHo~F;7C;3)szq3Z($5)wn
z&5*omI<G5n+Z||%Ynjlu$@6a-b#@E<ylbraNnLHTwqqwt(Lg^xC@z<F#7x$mPR>t&
zZ8q$=u0P999nrI6Neq)FYEE3pM=N-tib{9Xa9(MHXSWkt>;}(JCgB*b2yQ-IY)KMg
z)k{}$jjpA9OKNM8r*g%gi!LlmkJ*6lN>TPhQX`ek>s4+%nsnI@J+?)GV}C3TS10fw
zO>B3uL$!f5$!12lE9lLmt52<R9!cNsENd;^H!{8My;DQXfM2{l;C1l+9Mm3qc3`}W
zG>j@IcHKp2_XwrqoI3%qIH<9lkbJiE!j8G@KW}^baWjj&;C8=emkxxachM}&&cf5K
z#GaBqP~0q@-8E1-CPvYY=><LJ<`A&LimRQ)7O)vQ@9b^%L$$5M^1Yk^pXEEU)a+qu
z<m9pcy>@)<f*SL=q!E|?OI#}`Qiz!SPUzVCYWE0_;77e^*K7Oculo|5{EoBW0=9)m
zl>F>F9ba%o*E{uL(&$rosh3LRpfe$w;OcGm4PKUpq5n^Sgq8JQyzAuT;QY@b9UCL(
zzk1bq^k0D_`X4|7MB<#-0|erHj&qL3mVg62Uzu>p@%faUPgo;%Xk)xyId;|mm?j0+
zoBlT>>qry35Yu!6V>D@CHwThhur4cHUmihPBNT||gk#0RL<EJV5zDmS@nk=(uj>!^
z2FFam+}r^PG!{PRFbI78Bm<$*Wuvq>t$cuUaiknD@`Ptxm%fk`$N?!bS<C|#iK#v)
zyr#DH0CyaVVX!%oSr#Hf&tMNh+9BR3=x`B28(m)<6Ba5>B9$g8PNHZYE1_S}lYh&t
z02k;1O)7!E0)amoyi0&l8q6&K$s`JU{cIr0ep_FXu$QVIIO$s-6r`?8y&Q8f=bAyV
z0Te6<q@ka00ElLmuy;bBkwPvA?VhM2fj=XGKNX^a#AMT9J^b_lKP?plSsxZuh#O3t
zN}YV)9}S3BL@r4%l2FHFV*!YvQXX-3W}rY0LNo}!yLnx^dQ_-*i^G7YALpo_Jn7Pr
zV0-FT03oWFHA@4S%#S2(xSxPB4u=R7Nz)d=A5hXA|J_eXd$b22p|K7u?WRbKwojoH
z#1K%EEBdZ^g~61e2QBQZD&l}goQ)5ZR*-;-T!4uKeGJkm+8c|6)&Q0udUc*&zh`v}
z3HtJpZRQDwBv1}8#PzORZORkxT_7@6y$=N6kj6jTfAsYYfRm8nSO7sBDhrW-S2_wC
zgoVRD<<|yisXePcIS7{|9!aRCUv_|_Q|3V&0nffG0$M+89lUUK0UG#hZet;`%cIGg
z&qPejdg5?iD~>FAs3Oip>E&=_Ah~HD?W?_M*27JGd!I?zG_w`j+DGs-=}$K1Xa%;G
zWj}rA$}bafkZP`c`JPrnue6JjIljmiL?!mBlodR!GQc1=i6q|r$s~F*l0}k+I7R_@
zwkyi))kK!Lb%<ZrK4PLuEn!CH6uv>7J<!BC#cuxB-lAg0zs>XRbFtTe?;5CameE&y
zZUrUQ1$j2^mxqjo-pM$D`3*|KAI_3pMZcbfOZ*ZuT_<?y!#*DwNB&|RBjNWIeV9Zg
zd_1e0f(%;lhhG{TwL8V(kwv8|Zpp>s*6ZjQGV>#9XJ^I}9X{@IJ8a4OQ7L|7zvlIi
zHhQ|Hxwwl$d6G5@G{|&k8z{zR3@Srp%*Ew0xW1@aY8T5mgPXAAwR**6k{1Hf{_>q+
zna;-G^YJSG;$)}AsY+}Tka-ynynAmc=<r;A0B2KpyLM^qJDyD`t5CG)DnPk>3sWr)
zG8<%&ij*~<X12Z84aF|A(HEb7!Q-u{q~Y(_cyXd^D>a_&D#ifAJmIp-{xP;UDII2?
z7)0~UOz8EMQz|TRlw7eNrK3~cgYSgaNpAhvIFHQ~drV46kUFnmo83$}*<I+;qBfwi
za7_-0BQ{q2kto#7vocrO@{MdjHdA#I!&um0+Kn<<%8I~yk&<jLJFvC5E8$c9hQwb*
zmsWb4L4NX)6*-vMuI#k8^^UcZupIGNC@R)HRrsdinl*apwI4jlFS#NkBQJkXz;kkr
zzPCb8C2n!LdOC&i;&HmSZq3dtHs#OHm9P5CI~TUn5qNJo%wbSl$YN1eq}YFoMYnXJ
zc8&S;@kB>usaA0lb7r{`w<8$S(Gw<3_h9V`;;8u@g>{zP9i;;5#ckihZ`INB0h*cH
zXiL0_z0p5&vdQ0X-C}#$*&ObfAMF{v`4*J({Vv!kU$70{u?E^)cA><Q)Yb$n{=)aN
zb9KB0{qm9aQC(R1g8P|+Y@j-w8ZFDy9erSYFVuT0V=h^lqY%0|ZE>O;Y#Yi?hWaUF
zoq3b+aQNWSNm|H8)1w-t+|>k{Gbd#iaFn%E`P5<`WcKPCV1Iv;_b0)}MO(FrqiVu*
zYLdm`%K0%vHQjY9B8j5iW^$P9Y!UhItxG&+<MU?<$`0!<K80b<;R}Z^=A(%?lXn)!
zk97GC(6P&7_`ybR16F%!%HA(G(eEcv<?t1ZF|fB8D0IIQ$fZCIZ%=ZY1mSAcN?i5r
z&5nxBR(Fp^+dM}PD(fex9y3rwtPgG0?o1@~rPmSc*UpZIm3f4kakkJMs&BeeB-4R%
zuG(Pc)(1~kQ;k>ltaG^bZ&qrqmTg$8!I$pIB12*N=#X_l*U&^`*UoEP*==8d@{Cfz
ze*$9c|E-$%Z;k1U1nkVr|AAOc1T3s<3{3yA{HZ<u$Moj|C)2+vUH-pmJ6`R;adcM=
zq+2alKASDw|94n#uGnn0SZ}s;x?1))&ia~uc^VoP5K-l5r&BjNsPH1HXc!4YO*jRN
zxP*`yy0pwVNk2wy0T};KWr>juxPK#qig9xha$ElH<L7`u?^_!f!_+r3G%<i4ysfq4
z1ja@N2yAPJ3uxk+{th;_GJLxckkC@%1~}>cJ3{yTojtfbG(SEtgC_7-&V$OYFx<aB
zySjO_*WAHn22GR<V8XW)RGkA524E%zY=n#rEJhW8TdB`ihecq0aD07e1d_nq(&)s@
z_QdRul_5-B9Vka;CV;Xv?Uykn_c!p@uU1x*I!BPT&jWi9jP>?4E^f^<j`?p>9RqGp
zcrj>ZM^k{KZw)49CdP%Yi>Z%q3MqJmrWP=Y&&WR@@=W$j_D-MRAG05ZU~FhBc;Cfa
zzAw-8(*WOL&$;<%vW95aWFt^$Gor`ekJpCxw(1M(eH#NmN9xoDNZ+fi`0~oy(EP^0
zig5qT_`qMigat;%uTr_;t%0$Lm7#ISzGvIi@aj8XWt7iS+~3ezjlSk4mR5hvK6?GX
z9jGzX*SIR(*1m5#zB^jKjhR33wO0P%8*Bq^L_J1DNq>WUIexrIzVJ^_a&vEJq?c)7
zaA$shGebeHw*z;breSgdI#pR!0nGTKNk?5q@qH)S9PXc*8D4zo!7uYyf3#X#+{Mz7
zkzJFCGF(u*i1PHOoM~XrIH$Q2D-mH?3;!vgu^H6)ow5gag0_Su7~(;VGNi~5d+c;q
zC)6&tHhfr<<6GR_%EcrM2U~pbD@?!40zOKgV}f@LKt*`09SBN?;pP}_)EF|T+U9t-
z_nIJ7kB(p&!*`K&$d6^NpqI+UU}`_|<u2nJE%2<n7rt}%$&?!OBSJ1wc7Y-U=&S>8
zW3Thvcxeom_L;T;4gk~y7X@7T*^Vmp7g4sq5xfT)j!2wuG%P_Mzv3;)`=9oCX`&!>
zO)2ofj<P;m{ir=Oc&5Qs&Shit>O$z|vzp#kBS+D*rud5Kyod0ddgzi=MmG#tbu6K)
zB9!c7-~_vBJ1$!L0kw{v`TNvk2&<hM%oOPN^7dH3#Im9K{ya~bM2CX*yZEHLZ71z<
z+Igx(rwUcxZa{^rj(sUGni9Lzk&?HtF3mx)URN+C)0}(3f=D(0dDasbEVE^}U<cqo
zNj*TYS&xOUlfZT!pr2^8q0Oc@ja#(hV#cE7f$3yQ+9|yx>B?<k9lUJBkM@JMW6#d*
z=^r`R>#Cx92<d}aH$6j-x4pfR))AqTfyHl7K&!t{-dqqjw%C+e^wjf#ul<A8D7~29
zN3e6;*T3On$w^y?+ie*hi!IV;R$pY^h5P{MRSe<z;w(vYwPcO;Zd3DbdSDcnjj-5#
zG7j#M&L$2g1%&0Usy=S%HI#dA+a*W9R!h<o8kKzV$)DlizlXju+W$s_#2*hggHcO*
zdRy_3B|us$fyfZ^pe)xbE5?7vh_2sw9Y>B}#i|=tWh=JA>lSJWgFqiBqq_5u$;Ny*
zM%Ue}B{`!3EF8CyJ1CE~P00PiQMVcHPeJyf2uj?Irp`56*1Fh){x~9?mfP{+O5!Aw
zcec)n;r(m#kvgPL7yZccIt+A23t9ik_n?WVo%~3uDv64^^&B^F9g6{>VgOr|^6cVe
zO5O-$v&Y1_muPit&*6Ppn>NO9&j0`}mb#B(Z|4TPntN*--V7e!l=b{W#&aG9@t4Y@
zBIZ~Us(CMYPu8%W$XWYc2d<cJneK%z9omu><wbrV)ZXig7lHI4eKi(ICtenH5~Rx1
zel8%}qxefFD{))p&0<{^>UG`1<4xvR!@)%PiJ$UHxE2<#)ePqZ{dH#OWOOBvniVMI
zi5eykx<?5dMrnO~sL>Jww=|L^FRPvP75#*X@KT4YJRjGqmV;&k)6VzN<}<}U1^5i?
zYC@7<)(Izt!a<sAEQfeCp9RBmAiE3in3!QXSX#YJd)u}UAIaREWscnHD(*c9+g(@(
zm05)YKb0VAfn)wNCoa@aJIQMw>med~yKEJ*nr%VybJZP>^^Qt_qPQFqz+Yy`llrUd
z8Up^VNw5#OsgRciwW6@k>}9XN#z<B5N;tuMSJMG`ib0@RW0Vm(L2r?Ssj*vQ)++%{
z0GA{B+IU4l3}n=O!-FeGBTb5Jcy+$p@z6m=+oPC~<%!vmJ5dsSo@CP3BT=Qq9<y}o
zvx#{E&8@4p2!duH!-XqXo~XSaK7>$Qv4FRzRY)8WPBtf;AZabOxw=7y{McfYAy%TE
z%2wM`63;^SRky?rJ?!!$=>ZNHC$zAMI|B4KU1PV+I>eL@?}d|Fb0uTcb@+P8vZV}d
zl^{~^$2wp>ycWpsQ?%t-OIPXRiaDF3#6}p;TZ!vVc0=xA_?!h1D4Uz$L;}CDQSKVN
zf&Q)AGvXjFymUKlyv{#5+EMGMTG&+Ul(GEP+b{|-h7s0rP31fksp$;d4R=#qI8{j1
zqc;je6&SW}M|`BwSbb8XL+L_S*8(uOQTNd2Z&F<#fS()Fe-HZnTXtG33MTtcLxf9;
z?^>XPLYyKj*^n#2*+1f1StlB^W_oNzYOh6(I{^IHmNwgY<}a2lPsZ%G&G1j;Cq=`!
zYNENqp6yJ6xSSFw_FXVx%mp+3t#wH8RX3E;lShr~(RLC^&xqVUr@W~3YpcMe?n7#f
z>^T{gevk52K=~{2iO2FVxXi^d!Wd=8pZa_*6W;jEu-m}$z7yy*2QJvHN_0JlB?cq(
zn7J=00v83*wZ0B#F_DqlaDNl;{P;x0=BoPyaVFzMjOS=aYjypVbnw120bDGtjXYu7
zIcQq5b{}pef&^`?vmh)8{hNSKg@0bdELh-soPJ1LC#nvHdk_Vv(c!U8nf9UbNt1|i
z83B8zd_QXFOPZmvQ?o1I+SJMk2Nd1;(Eny2ua<LUg|xVW`EET_fF=Io=HG(M<wh*m
zJQo$dYKI1c6?v+*7V|w4>lNo;Fx6piz+f#0A+4GlQr2BZ5tL7?p#xhy%Q&##xV&_k
zn{;W8vsp8l2|dU%!zF5D(|9vc<~V-^XzNCpHb#tE76?NnOYl84=GOB(X`p0yR+K!a
zt&F25_kkMpJ7Q<TN5e6?6ExpJ(|Yj!a=_zFq`T9m!~13lpJ*@w#&)*|Be5-4j^REa
zN*wO7wdRPK!sHVqh#gIz{-t@N8MjwA2p&A#*u`9a4Y?IH9xcD~qH&juJTY5Hw{pJ0
znov_?*j<N85N{eJc5eW+xm{&wPF+TwH;%bcKfKpE^&K<+5r>(+xVo3SWl0TUV9IUN
zPPz_n!GmzYMvT?gb0HasmM~}^$L0`Xy+vouvXfe|4M9b_u3}#t^sZ?sBQR%DFgCn7
zF%q`<<|fFH!5arb$)nRP40%suaXUq`!xn=XYP6;%6gu{Nx-THzDahC!V(Pk?*nJlE
zbdtCmHNK+=?fpk0YSlD(b;iX&;MvR6TKSGifp$Uj5&|BTl3d^K2^EME_e`9KDjz1X
zmsfmVM262CCZg)BIaDsb4B~!~fik*=ja(zF-+BzY3rmpy{(LukKmmhqUEiC3#0S5&
zjCnhLqo<8&#gfm%uY5-dJJ(<G_t?U49pPHKxyIx~6!V1=w)$P-2p`J}xs^c4bZro>
zr1tuIF*8}mC}cP%_*$Hmz_i157W_z3xI*exg;-Dg$<g;8Y3Ry1f_y<YgV5TufJvlJ
zw1m$Y?jotL2fXe?0+IR{n@)Ez5KM@QNI8HOvEhC&k+D8Jc1M(shzjSRL`uzi0a1rJ
zkIGIpiX=jv?B5IWDX4&}Ub7f1Ql8iJf0qoyVUYB%ld<FJf)!NCr^@5-thh#ZX+;4Y
z+b`{;VttAUdv3vIg9q8>?u6P5IgknDcp^w9$03-AU;Qb|3xYtJfeaW;Z=~Ke5-F{%
zuFl5jjrqP1uJ_)eJW(%oexFq|-x|CWawtlo#t5d50ZXKTVe5$|_2N%2fv^k$)$-zz
zU?8?d7ua%-6LKe`tOga%?Q8qlu*R7<DjlN-nqjzYe;-MBca<N*n(Lzpz{bcn28z%(
zRq%ZF6hcCMOw31kH>fx7Y+wjAwRr-L?zCIcP%H~J<E0+ahOT<n0Xwe8{XR=vxCz$T
zp8T^MsP2ND^ak?H2gK#zcE#0Fa36W^B&TIPwvwTL9|IIqHyvEwZ;=-xOnvke6E9>d
zs4<`$OU@h69>q{-I|6~CG+-&Y?fQ9A8Zu*}4g4Cm_sZd%_3gJpX!eXg!KGrw1(_Jg
z*&=uCa-+i3vxrjCZ8drtQ-|>T6^OFnLxq70P)taPTC4`V_FK41IA5z%a(7u?P9Z)X
zc~zVmbqFiLYFcTx^)Tcl3aj{=@!YWCzSTv5aS|I(n7);nZ&#ybBInKD0}}OmwyR}=
zkqfF~<Pk7^SI^xmk(Srn8Xa$^zOF7UXkw>~-#51-Rn(n34V;}}mN0xEcN0XJ1gF7G
z&pPp98oqFgEhQ~iKBk1T5DmM<FhFnIqal=M_8}^)ICsL7+VPmxYy2i_IX-?DQ!uvl
z(odtrtI#7f9e>=JK$Z)WWJ}qPvX=)eeNHh<+wa_}YEO+tTjmpMC<jMe3p+i+Sa`Q4
zS>jJm4-SjpF=Hr9dzmF|82e#5q~r(dyJKbhca0Ay32`P=NY3|JF%5teLIJI!J>xc3
z$5WMg*5B@Oae@>rhcb#^bC^^6EYvU_Gj4RqAo1Yb{Z`kl1l(gIx+G+U*jrXU2*_9E
zI2ZA|{CZJX$<@-xF|P_{fxZD)mG+n0+$6{Q;Mf9zQ0rSX2b20}HH&)+E)n=i<A9MA
zUR4p-kJ5)y9r)P<;BwA0nY7=h?r+EF&TBWyw1OfF3RALRUzD=!6WLP@nv1N7O0#t5
zeBaDTXxmzf6D##N7B{&Rr<u=$h0d`((SnJrqTJ>VIgeD7$w(SO1azPeNbJHbdF@0(
zm%f<=1j|GXle@r;lW?s~!%+vTab`22i@!F|4=J<mjs$FwW87dKzr^@+2~=)7FFVay
z+lhHO%eyK-go`recKom5kBzRt(6QDdPHqwhuAhNw0ynLt^T_b_*jYig0PP6LY4Urh
z152nQn@gfi*Io5Wx@lE-w$cud!(<@|O0lMD<=osovqu826zX`y2-_hn%P${Fj$Anz
ziojaaI|-tglY$1S6lREuWBsg<K#(%W=6!X2m_%{BhPKc~KD0lzIrp->n(vvr=y_ZW
zr=OQtVwQH2gjAL@p{@?&BIjmT&+HHa>@%VEzd1O02{iS7$!o7&1RjMsQR>**)95`b
zwp!oDR+O<sUstWRL|OqYAJiWZh%^=Z`5<C76xsFesjDPK^cUwAB~ZLH9m+Yib*rg(
zWKvmRRE<uH*x(m4zS?w^D6re&XQW6nbV_8_NpAAP=kD7n3)MkOK6nuSQpk^!Zqfno
zkhXo&KIjRgDyBgR-XAw)01Sbbz_<fb#adgn8BnqvK2xufzJ=r}f}M_DWM7*pv}l>R
zEH_4(=wgdm@im8}jt54i;?T~c!rG+3A#VF)a0<QdhdxqBu;J{cKKRAj9h2Je7m0%}
zmWaYevsPUH)&-r)mbPsm#W;43u4SD&50~#_!Jidf0=*Li;u75osnh`gCyzqZXOJ5n
zwAV$NjpnVtoI#PXChGiN2j;4}EpdwxHD@7=Jis*9jHYt_clAK#5#i?+7COgUndRK?
z+bLT;kL3Zh))4sBH|al@o$b0zv#ZI`MC&UqT>=NZGq(y893gh*7^F7WUf547QCH4%
zD?}U2li89Tm4@)#(y#l6j68mAPcY8JQGfop+%P@ng47o>8NB@o4^*u@Cshjc>l&3w
z<SHHiIc2TFU}LjDpbF3`7&orkV6%E#HR^$TA+5DMtC6Aw|ID~ce@0Vwg3SLE#W;zv
z*F=Ja!7*H3@e=OEU2(sX`K&_NV9F3N_y#nfhon>WJq5UXzKFKrS*Ni8>YjW48pnDn
z;%uJ3I*op*en^xP<>FQg7!VSa4icKqGDq955W_mjba|9)`D<>Sx@sS7TZ0-T(A7|X
z24v{>J;74W;`GxHsVQh1NjScex;{XH69bX$DZpcGO{YJ)KfMdx$)rga0R|DDOlMwE
z6`JWb#{X0S7oP%2YiTTrIXc%MlS;ZsANmYmrVreg(;Q6B2?+wLkCu=CCUNZ;`OWe5
zDJlVOaE@>&2g3eH90dltVOw}sg32;0%t<s)Lml=EIKZY0y&jzra{$gSZ0ouic4^F^
z302Z#OhSdXBS;Edy}Fdm`1ofyvOGvMrq{0UN^7)T_E=b*7WT)~nLY->i3TLZOtP&H
zG<nsga(}gDvh$BOO+#g-)MYLJH<*C7c0H4asa`$Yr(SAxx#4wIqZ5Lxp=nnIKzP(a
zjTJUtO7x9fqp_n}CzE7H6nQft^j%`owy%2sz9OUV3akHgJTU{WBnAg~Mk_nw;fMb1
zElBEog`ilsOs&>TqnQlZX&+aq^wrd?8OrpXYA7=H+kn9DyrRwlJfa!-aTGmRRlCl^
z?PQ^tta^f{c7jTfd39mQhF57<J$|9Z$SSTvJkm5U$6sc-%k?#Q6RXKbka;U$BNu&u
z(pDRw-6SuEH3r3!6Hcwj3r6~8i|W^6Im860f_<RdF{6M5k&d^WJPuPgJAFj6mC9gt
zkEM{*IrOe|_T2%rZ$_ba{hn$pMcEbRD%mb%w;Z#>Jlz`Z8VPZiTNl?BpW1CKcv#AS
zq@B?Go=3)MinF2gj{|2pGj2nl396;KE2!hAxQtw@3k=-T8DdY&nQ`{<mLbX{*L?Nf
z4D-mP&;T042j(&-Czt$dd&)D=9OSgaL%RBl5<jS2k)-(&Oiqef?1TaewDrKvVqv<a
zF1!H(C;&|bS;F`QIR6|YHH>rYYB9^>joNYppqK2Lhn}=N!Mt>w@S8~<a-rI#@!y8`
z?s>|~y*K9(H)=*2C3HdatY!w0k1vVIX9Sh_U<)DQQLv{~A@|kxKES1n4cNu8AG*F9
zkp6dk8y{`OfQHxsNZtPdJV3+0@-K9>{k20+Vy$+!W50)WlnbqQ1^FA3sZZJB^d)Xh
z;3i-+48x;T*xI)$xq5|p+l*44>oY}j>%okS>82_;;k)xOt<(>SI8r5yzrt`LE2q2q
zzzwXK9c3Fu8+~u|5mW|=;#X&X$kIleZx2(Eo?EAaV`nxQL)JqtE0Q6YNSrkf7FaNU
z3t5ay{!B6T#+C?&Z8nS{jd){X6?sjnu=mMICW^X?<-38i=&m*C%zlKS4?e>JaHEcX
zwzn3agZW~uBEmRxq63GF3zQUnV-5af^2*pT@Mvwt1kR93(N7}KJVuwnDg<#$k!q26
zS?U%I6n3NB6<I{gT~3Ep_pZ`V0g-c-u1<#bpE&+a@(Xh~eQxQ$k9vjNeoT`1SkY<h
zzrZ;!#9((CAr&BUR7fPAb>c}>i{|Qe>%y@YrCGH&Qv=6}f;9YwM93v&&)L18$PV%1
zKx!16Tah3{Bb#N<=lW7vmX|Y`*UG}^Vv{FiyZUhC%Yar^b@~~20Rs*iiQ>8Q`jLzP
zS1oJ-k?r7^gi%$T7IywoLVcV#MD!}5<CL|*O-JcIyMqi|3dAm-$o@oAG+nkv5FdxP
z)gB;!IZ_-@_%Z7_hs@|W`(w^~(Vm|Z^W=o`msqWU-5iv=GL6GXcuXM<^-(FkT703m
z;td?C{w~>NRP<TGn>fknz-$O9L`CMXa0JGlc&sq2Iyo%KCP*#{oD8U`{aqBESn&kJ
zYfY}5F=*Uk|JQfxu`Kpd5tC(wrWAhj_yur~0Y|2=7R95z!zxi@CwRST5<g8SR-qF$
z_Ev+_=#96(CLb^G(e?Zpf?q`1OAuCj42_lCl_M@j2%7uGj<;z56=LzJ<xMI&5aoo&
z@=cd7iqqT{`m$2jPwX^563F|gh4H}0z0V9#!Eua6liC)Q)&sHZwo~5-0RVL_>eC~h
zbAICbss&EgOVi8cK!w@tim)0QiThk?$_dkL|3L66-JEo#YuDGWiY%4NON~L7{q)_L
zm!Oc;S2(nQk06}UkzCZU_~YR4<}=KUe%t^OXZ1kxHQW2hr^IJg8(+Yv?P*BaZ6e-^
zLe89<kxzL0*<~rr6B&Pw<I%eV0yDb+b-VRaGbdZ{K7$T|gJc<i$^O0eu;RGUZ3lZO
zW0PWpfa}A}w4~kmM0qYt=ldt7#c(J-fir4YOj$c58HHeAc&WG*HWiYNB~xF4Ye4jt
zv6F^@jJxLC{{A-|Jk4@Vg#~nrOp+ujn}hD>dPY)sw9^W1)EC~dg||2401b4&CsCGX
zU~|-&!1%K}BY^x(SuUGGU5!?SvJgG<Apge2>!zR&K-Y^hh#V|2AAFavO6BkFJp)uj
z4WQ6=v!yCcOhVFBY=$M4H)L|#!D)hLxHxxm3)BE2j{8|{yzI4evuQiNl8aD8CbZlb
zt!LjK4Zc=;#9b*<lv$%XjMsZ7`DY94%=55-y?NNyYCBCYoiCYO^zm^a2$1PMRZc94
zo*)t2O$>v5aDDE=xY%f6heJH2m_0v3!jH^t>DsB-)_A=aLEGn*PEi{wfQrFA68S!?
zJ1jNr<r$?<$Z#%4i5oi69^Fc+K(-m;7CPPsCLm_2ye?kS?sf6L%w=m8=KqyjAjsap
z(_!9RZ^+G|f|bzd2v*3}q*x-*!C}(4RUFInRQR(M2-j0_u0QCKQK!l2R+r?2rXvq0
zo+Xfqc$Ix3Yg!0VqyV;Pl*RjiC0%OVfvo}t17o~jj}!V6Mzr8b6cxQ>iA6BAjHg5@
zSYh4fTJ*9(7#V#praI|uUerWq=22lj>z@A+Wc=CGyLuAUWh!VYECl)F(ZL;qFp6($
zpcq0-?kCK-^pwtpO(N6{oO2m}nYpv%cy2WGHgN_XH*k3N*I4=)O;_CJBqKmp`ueT6
z!j_;f6TdnCCfe*$%q<&{=n4-ZmeP7OZ!*YEOfaOo>p6kf!l4kdwod5FV51UyO8)Yq
zoRkv+rPa>|2%>}K$ic_k2s(MhA;pwjVB*mJ@%N0r$?fKFVVcak|8=gGk$jHg3Y+QZ
zmM}5>T>2!2wkaI<USoTz>$z!>8b}H`w8koa1$?x#s{B|z0>USo@VJRuIY`qVWW{*x
zmiY$Kcq01BNJXC0gUm*!yUnl2a^L;omyxg%-`8w)=I`}p_AgAQ!gE6=GVE@M@>Uqx
zzWOr6QYYuHNcOh(Uv(edLsLE>XERYuz?`WfK4n-ioYYT4o>i_@-wZj)hGEiEd4`AV
zkr~s*$?Ue+iRuE(SJrNV<6`(pEkd-@29fpDpW?98w~E4#h{nBSMffp6F2Rw4nmT_u
z2~+x9qlz1YQa&gS(+E4H=G;M4d_&~kBMH`ay6O`{a+`?fk<6UCTKb{FI?Dib!m!Ul
zlFT&EI3+-_{YjB3Sq2$E2GU!^87h5h1=|Y;x{Oq?-;cwW6R))PV;1G(ST6%q*vNWR
zV~$_cHaq1S5D)P&Y6Bl$3$>?*z-<B!gw-KpPyGoE3b(;$KEw@eM{(k1h;Q`bwqh9*
z-#91AxAGid>8UQ%Yu#4mfN={a=#U=Qj#jRnjA9Vw-$W*?txJ|o*WHr^)SP8m+dd4o
z))g|JXIJNvwi{nz;%*cz%0=rN9HR*Jb#VGvxNi3FW#400ePH%(2I>vQlIdL&!;rj{
zoNY+7@ZJo!jI<iX-sK<Fc3gufPM|=vPu=>u=%zkkMaSyOXiAitRD5U4;%X6*Kh;@C
z_?*(X@~aTl^yG?<O&>wIXGl$>sT(<8_=Nba#Ro+%2(6gDIrO=M!8=npX#w)FJPThN
z7-L0t<|oqq!!o-(1d!Zeyr7D`xHf1K1xuW>&&@|htV>n0Ne0L;S?ffZ3A)sAH?2{P
zEa1QGN`3sG{dn#j<%Ho>*N_==3CcPkUKru6FkU;W%hpD0Jh(r1qe$D60B=9Pj;uzJ
z64S5^=fE6b!XTz&&6xG%BG7WB8y?x9s=+oT)vC`J-)F{;TrnxQ(O$ZOi|mHCN?YZO
zXI)+qEheL;%?h~~j=dScvGerW!LevfK{=MZ%-NAdZe+J8MQNa#l}ieokUb$B>;rq)
zi<8k1DM|og9yMytf1jT38cMU*j4lz#+>vEa?mOIS0t{S_nkE)5VJttQnm>Hl7SvY-
zZKb6arKFM-O+RcMdDLQZGmN(0!Sft*qXjfkI(<8Ze+>Y?J0OyQusHp=$n#f{1dysI
zhVQNf*i(qE;hTQIdzMn$8R#wNNJa8p9O3lXT!47ngJTQAhdu_d#J9=IXpQ`O)?_qG
zZta25Co8Q+o-H^^?J<}5J#^Q_YKbB&_DPZnR!B&4j+k$*^#R27yE>yKB`{WzphOZw
zE}cMeE*aBkbCY5!-2jp80>%C&fAAybndz`M6)fbU*)b;7&OPm!Niv|9zlq8<0vAR;
z3|iSQV><}idC~R`5=dm(WqB-fkD%nCJzZAwt=_SeQh3kDVx3pTZ36@*GeOp=40!-^
zi8TfJPClThQeZ5RGid7N`woXK!we7q>#-1M1>UR-SDakCpL@|SP-gpT9mpTE?AuMt
zo6GbVoRyZ2JlRB}<#UH&pB0I$>8*hF6rICYN0+*o6TY9Kzv_&F2PkJu1U{LqHFQ=C
z`NgA+0r}Ddu)sQ3U7<0hN4K+%MP#%3t3@cUwe)1aHd}h)Ixnt7_ayRKX6(p`BRF8x
zwKH+#_g3VahGN+Cr<`1E_19g!Db${c_Wq@rgU?M$nqsPTxQ|7y*p3xvWD?Kd3{_m5
zRj4gE6ajibY?dr&UFrAusP5ZiUYm5F5ErY22mCLxL$0^dssf(rZhYZMmzFwCdkXzE
zPJ!$6u$5rHhl$1Zjs04SRmk{XoybhBb2P_0t5R>Dk^z^QuR9?r1wI9h3EDgaMi6?K
z&JtV~M*tOfHE)C^$~D9WVu*EnJvlz=U=usF_#Lbww}*`-K}R(O0wYiz7cbiwocF)H
z@AQ?I6e8}#jHKaAE<Oq7@o})R*jn1ZXXDG6maLO8`H+0%@}8e2S?@Zkz?t+3AjV%=
z9`=^m*FwKBw-xpGCJl;;U20la%!hDEs(>H8>`U<8`Q}!4%5>ptp7bGg6Ri%_s}o>_
z$N|KeV{Fit<XW^mPbjuG%H*1UdW3EbK(1VuO3_K8`EiYN8uWjzWQp?Djh8-vPl#cG
z2ruJ@y#C%4Um%x~^}|F<g=M$4Ag`2NJ8fmP{XvbTYYv$xoD8G^vCFuqQLX(UDw*fI
zc;iDI0$s*_iZ?I}4g-llt1o8MfYw7LOJudsPl?q4zkq<NIP4`>(~pB>PPD54B0|Fw
zE)fA>I>>XoxP>Z^(YV7cpQuWnfUEvJnQ=KQ(%ZNhsx}q&We;nlIphp>pW4<ys}@k>
zHvM8skOQUxmZcs0puX8a!j_n2{Y8h0(X=s{xFlP=?A2+X)VxDEDzLp-3XF!H{wdIS
z*QCQZ7eK4Oxqgo{vmdRTC$mrvadJt3Qqq>=L1zJm_nj%!GPmTC<ry9zDKlRF21ms~
z=A&X0`qw`{B)BkE{nu#jQ{qIysT+p5dj;%Bg9h!r0v1AHz+jt#NXTJ_{JMBF6wCdG
z@3A0Wtl(gz@J-ck*;;e~sM6%~$>RtWnE9D`Hi|iXk-{szpNXVf+#ftxFASIL@kJLS
zOos4@0c||)$u;1tTnuyThw$EEa6Wg|+Jrq^M5ZTE*objjqaZQXfHV}$$&K?Oq8R)1
z`^fE$+z4|0wK>AiaPk%K?^}e<(r1y&bGdn`caZ$Nzn%K2Otdta#2xayzhT(3)#Kwp
zV6PwkG6u!j0!5Z>z@!JdM>lo6jo+NQBV(cilQKIUO+LH`%#m^el)?rs+ZXoOj@!?8
z4zBUQHll}2cKDltM~Lusf3a9z)uA1<XzOseuVmY8OZu=mRfC++1^Yu<urrBUAZ1Ic
z&}rfD9?uf=-NRjc2n{U`L-Rw0PCui|O$p@IFUs4%HrX88hU&$ln7RnOanf{m;$n^o
zW9`P2q6}p4`444eCZI3{J{_}gLp0E~4yghoZn(-0meCeCDSy-!!dw_<DKC#^RhdN&
zk0|tWGy7$CRHLet5uK(0@4OZm>seX7rCZfY%OMQQX#Jg?2tiFP`8ZJgTwue4@Sf|r
zJU{4kEFSO_9GiDyjudJdpq0KPxkg>iRiDU@9)CS|xa#^b1%x9B)5p<g4rwSIZD}Tj
zIvzm&X%BuEqs%3!9Gr2HK+yZJVJ`JLGcfZV!j&_yc79ZN_Hne{UdPQ)^|(eNaX9bW
zQA<wQ#51JQV#R)C+JBYga;*nPDjS|Fa!x8s^^_)&`>1$5IWBIh>@}k7lL0jD<J}<k
z%-Ea*<x%kK*U>x7b>+VXYbr<%u>?NT;us5+rwDY^u3efk=gWH6pRPyIDzm#0Dc7Y<
z;gNmW4FvDXp1^d@2wUr7u%(iUo1;?e#okGx3vB=Ef4L-YxTmVS2x8n@Yo6OkH^w}^
zbA@sd9D5lwy{k^47+1u#A_Jcu{;7h>RG{3zWY#ZDsC>I_trDhrhN->ufI?^;7d=`2
z=%)GchsBb4o{@I$vuz`u((L90tr)TBT$;uB$4c7k?<$M~Q)+^0vK(uX;xM?fTY{3~
zSF@Ily=y><)mhLz_Ft(_#z`97vq68}X!T^m?O8MW`FrVo(UA-a_*x+G-$AR#dRA|j
z6z5@97*8?*%LC#dkUguia;&<t&ClE3Jbc9u<Gq>IsXLqET663pTY7bUd~b=pd2brB
zUIH3rW*g1Dc*f{~&}-D*+#&^b7#Bu}zpxh$NCQZ0tCa^rlFt=a0&FeUTyyH@H!6zI
zgOqe_vh#QXVM~RT+ep^CEBh#(hUt5uUao^_OR4+who-|@Jc}MW+*Q#_qdaQv6w)Yy
z>oLz&5ETzQHgA83yF9>YT7g+(x8Vz8!1q5%m0<8nf+S(!vm((oKPkL`|7HVYq@F)Q
z^pZ7v!pMCM55hBn?0sg!<&iX?rJZW>inhJhR0t?FiDxCPWO(XF9|!xnesgkYdQtnk
zdT3?%!+9Zbe@AwR!TFk}`@NFgyJHsj8#Ktgq8YH&$0yC}hytX<ptxd1<PrVNAGg)s
zs_F=>k)+gKi|-dtB7x7W42h^p|DM#GubZOmknlMZaIj6-P%#xkAH;p?h`v<@?@kZ1
zi%UqV&ZAplk8U_l`EogK>e+h1O}SwPj(58*r3z=jBCbKr<@qyVr{#*^!kU7vQV5s5
z#%XF!cO2#(&*uQ4N<t8XzUjr2MUFTJf=O7F(A+%Z`}|fvF@MyQ5uPFm4q|5lOY~$=
zL1%N|vgbg#uV1PZr(1y+SefRNm>R`of-bQnneE4?ltg)1E&f4iegbl<(Px+KOM)X#
z@f_e7ZKqB5zg4KXrSUABhT&y@-WE*h1c!EZ2vgsEynD%e_Scv=^amP`bU!kLBR~k5
z6}N$p<(d>98nIXGOGv{TE`Ow3X6+A&?dz;DARh72#**pnmbp~@lL+>Evc+e1Q!89|
z`z$)E*{<c|_kPSq4%Yp*!O3qWMG#S53;1J4+O&h)`-#+NnI*M`($3-+$#`W^H*W=c
z*(HIVB&!}lQm@t#U~=1qcIe}^BEs}&Z-xZFsJfAJsdaY3g+&U|f-@Cz-`S<dEtTb<
z2U<c8QnL=djth1kWCQ0`u@tWKOsy#@@bL}!qYQ5??-$>eJUJQ`7im#saU}vPXHew!
zdlBT#;{r*%&k8Oq{zlRcmM$wa?8OiPGAVUwqqRiwn1NFCCE~d`bS|i99W$8xHn{^}
ziq2fWySU6*O)f-$X$9I$txJjse<!a8_(<r=Wod4Rbpepdjro=*E6{IgH5iQxQtJ>1
zkv<KOk254^Q75kY9`Zp!#NWHMr?t||sw~5`$o3+O_erMHkn%5Sex8Xg49z;bmy>X-
z(ej1M3+3aW<}#2EAOO|dRFc#$8y>DshuOU-Hs@i7%MAp!HQ^_I`8gyTk?*b{=j5Qf
z+vgRCtj*kTVXn21Q2}qGvX`&UTAw9Q$XE1D2l2oIYc|XK-<SnMRMmVR$9Bu2lJB|^
zYjFKxfbC^<B!I$=!p*%3v9ys7|H@HZ^wLok6E4uS?ck+vT9gB-3La1J662gs&kHj+
zMytKyPrxK+B>HD7`&_>+^>i#fT=3+2-BgT}gca2A`qQiR<6JT~LGy<m(`km+D^&c5
z0nO|0n6(Qshw0dZK_?M|b%KXqX4a)vYbN>TL9E6dxw#VG{w2<hCLa1OAu%~BtIQNb
zRB^DD*h6Ve{VwABDXn1ev=)E1;`Je`zT5>1SAqMu`jOj+hUnfNTxmL{NdtwdMZPkg
zk5HOWwcvkSMw4SnByi)`RqB9ViUXfsAr~#v!b>l4_~h&ddU3{`znNmAl;LCre&d6L
zKE~;BIjEV2Ck#Amjt)%dxz8-2qc&br0WBbe<)mn^j>V>-$Vc}gl~8OQ={Y;2()7^n
zF?_P26RW7xW>7~Ob5>M2k!orQVKdb<)YYkS028Esr}pWVB^v>#`h6YsUO#TX1CLga
z1(^il6~&OS_JiJ;UuTi$2Yi7uH5&cT3}@hdr%&$19I3T{tXX&<vlS!!O6i?)Td42q
zlImsF7oYqVzGX!y55tLboXB>Xt8D3iL*df}$I$n+{dzut3x|yCDZ1j%9o9^vpOXG_
z@OO9SDj@Bxm*zxF7oc|EW-<=`(BCnAp8TpUp;{|$CYu4x6^pd4v%``7w}=%Z*!Y9v
z>=jBYQ*%!6@6oL-Nq%RQduRS1>LoP;RM+UbpGQ_DbL%?66K532mv*jQIfq|-Jy#_&
zHMKt}zEo)9?byE;Q8s(B)10*3={{kfEv{9;_>ClxQ0p6i**ldHK!=H|q<($z0}Bf<
z*tO9S0-XZ2T0-Q;L0f=_S%RMI_wy}$R~?UwdH8Y}s80tMD80F>7wwL7Qfm-*Ght%x
zH-30wfu=!H=y=gB0=I~7C=KYsG{~AdY=;(!`b&|BYaO&U)=uWUmF3-F!5g|v_(iUl
zrca@^{ZcWO7E>aWtDtVaM7B%UTgjp8ZoAc?yhmVfL+$R_pDp7K#Q6LekQTqz+@#}?
z0!5b9TDlIxVw_ZZ(?ENb`xT~tr)H~k7GG4x+w$24tw1BPZhxHPX5)H?*@;vdp7Mpy
z!Qdf+&{&U_y5?UP^~y`L+fTwF4ZE@OL@`pDNmu_lB~<1pF9-MQ!oykefN~5tgi^GQ
z@B7@W*lf#N3S&M=8%@bU+O#QgF!*gsS+l1J=26;yH4KU#-I1;URSk%IG%-795rqd=
z4y6URX``rqpq+vBkL$5a%H#b`WBwEKc6%MDs<zfSj9=mxK4gZ80`bAT^bR%N*$h`x
zm<vR&xkNW0yzmxS<zl;sFsKA-apN|_ucKN@Z=5cz)b<p2ji~+l1Z=uuUv!DaoAU>R
zbb4zrokMK~UcCY=yUd6=2*SX|y5K4xe0P%FNudv^Ko?ExFCmnKE}VeYg|Ul1N4*Ql
zBw1vjVQ5g`hMPmt14;HxS<;3~TFI(-@2tq<yC!V%ypoA`Shp(wU!;W!JH!6v?7%M&
zC=a*npxG3va&W9X-N2^JYDTCl&Gha{%-T?%+{*M9YT?Aw9d>T8za0;8D)rK^R`=hb
z`M?U^U`n`x@%F05*+9t38H`VL$gv9Ln?^t`5gyImB)6a;-1R1&9h*6#AuA-#6oF}q
zWBH0Gg*`>sq+MUn+M&MvZ`~%e#w3=hP$mx+{N3FyiG1&Tgg*|*e8#1Nbyutkr=}x@
zNC$CI6*XSypB@p`V{I>n6iK6b@hkXeurJwdTW0P3E%ii%noxYw6v38H>|=X-uaI6G
zK{4l$j_S~PFK4RwiA?nQ;AW?YZC8LyGI~ASX5}vE@=81>))8f)wJn)#(a)X<jK7M(
z2C*nu=9$F-zeNx3Qi+>8hL$5Y1||fhs$qI}ThLLXmn0+Y>$9#XR9Xz+(+1d}SB_TC
zfUCTjq!e>DUV6&coDHvcqFf|HiKri7!u(^aBV|fV*oeo#f01eTZzcNg^NmCd#`je`
zL41pLk4>k94d_AXhqK+rfEFE-9C$|3me||sX+)H`GO3<^9i-&S9)+8Vxdvu1^3JVs
zeN)p@po3C)O4dHno2d@<o;y5BE8XQtf+$cFF619TC&np`b;&7z7e`Q7I+JRM6^;%<
z`Md8)6h~tj&gfx4USkL-+ycLw2GjH3B?<0*xjOL{2ze;sK7BRyKLhuOJ~>-i%^qBC
z?)}e+#V6Vm5q*J<1h)&d;Gd1ES$Aw_e|%i9w{>Ro?Y?Ht`d$$1CJb~?ATBlMOGITa
zXu=S7*9T%Ilyi4VYWiMX@(1?Oc&!A`M<Pu7nVNIVH5BI$*q=Og%micD9^C?~#J{{e
zWyvms>se^<w?4nZ7aySL)M4;PuX8HBn}~A9*yA^Oo$RVgUwXx2#3>?_RVlKJZQJW$
z;i=@NRg&#IK9HK`k1GykF_HQ(Zh@8rd$bdz7CD$`S6C%IgxcR;r@heDE|j3BEr#dn
z0lzh9%>#;hEtxpwI9ZjgfLf_;8^fM(#@`t1<>rtDgEmB4e9>O-nH=A+#htyod~Fw;
zwIHhRuG1hDD_V|x8%CF)+CZ*<zf}+*NzgnN(Xl_y+O#GX^iIX;+$*@(Yjh%&<ek9G
zfjgaOb8bNW<(nXEk9MwCA>YnOS@xQRAvb$ooZGT648))!P|DW0l&CMHEfM3xM;B}H
zhRsC$^vSFv(G!vi(dnknb<}1g*bxq5OG25(kxL~coU|DtgU=}(Y_vO@bwnhZ8lYcY
z1&g#^I>UY9K}LmM5I5j_b!ypc(H7ohj@%gOk>8v&ZS1N*F=SWg^TcEIQ#FpcM3Zcp
zciK~(jkmXIcKeojc?jRFDr9)K<a}@4C)7HbsE~JIeD9E{AuQ!V?9LhkY{%(uCfGLV
zADMtYGMoPGin_h2mi0uIIt_*PRAiJBevWb`G}cT4m8%NP<z>o=G+uh{_UdGe1dnLq
zm>s}RWOGzT5NuWR5(ZVBB<}oZZ)_)K3b*oN$zEg4RznPyjk)#X;U7eIa&F6=gYE^6
zV~Y=)MbnF<e20(=Q?}Lim;0iQ?NRk)o)Pr8Gv6(mX`_^PBj0c`v%h&L_GO+#3rZ+H
z|C^$X-}yluPR$z}GIz2LUmX=ltUc=fOP@Oi(x1Ss(KvxhSrjXJH<0R2CG9OZb}Jn+
z%S6ImCZl`_tOm)3T%()c*|udEa`Ja(M3`e^p{SHaUlI?GL(Fq`O288zX1tMQkC?5f
zEJ)C+PN-Cr){Mt`Ka6MCVP#dHG`f5hze-Gi>j?cmoGu&H*kY?3Tp`u#-QYTO;k`*h
z&4bl&0Dsw4Y?&hj3*UQH7m|#%A8u2lkUj9nDd@U^*VBE!llqCD1>G}@uSdcA5g%K-
z{MOEVjFo3Odk4%I4&mh?_NynVb#-AoIJ%Ic1%VLca+0mf@(psP*(?i2DYUxF=>jEq
z(o_>pV!JCg+d7VE0CZ6Fa{djCA`2VgtBk8sE=X}mh5BLSM{^QQ6}nI3Hrv4V>>#fv
z_cLFR-|z&T0%^m=PTHW>U5Qv&wewNGI@z^U{=K`ri9f1}<Xl#`uYS;Z5qfv4<5ll=
zDUrM7P1@9hz{TrAOY;FYxOBV6NS`V>IvDKiM8M3cBjmBI!-x~7Cytlttlev*G|`uz
zrp9LOehJ?1ZfGz7+$Ln=ER%K>SWa+<chR1$;ck*I$2+obdYJ+C#PNjve$IfCEgZf7
z;LLWN?%Hea!5DIt%;Kcs(ubKHBq0<56JlGYahriA<kwUT5%75G&LoF}MAcCj9iNG-
zUPKU1_gV}Aih2=(IBz5>GvBur6?y6&#HvwFKovpyYyyN@Zh_;TDh|i6mV>%bPY{vx
z>-oLX<#Ljy8o!)m%%ynaI&f<_&-zTQinHa)hfu_oy4UrYp})(tz!8H~o<aP=417%@
zy7TTR6%hNXm$o#id}lvT$B`!`vC+r@i%eetG*~L_x^15F)=2$*psLL|2HMn6r3pat
zx`kxe;gj^bxoncS>IE?m8EVo0{<a2Sjy2lVuS~xro15pf9l9<=*O<uBX)DzB*$kQr
zwl#K!G^H@@jEQK-k{g>-_pu=L$SK}~f!lH6H7RrAX=O`MSTy}Y(W?hOU$Tx~NGu5$
z6+|$}%NE819jwv&+GH^ztPzn{&pN=&*mA=f#l_#%6KnB5l1U1`>6{$Vdrxg{MqV!R
zg|gZArlKCmAO$p9Y<=|wCSZ#~nsF`~*|jg$+2y^<>K2%t;1d<cF@B_kGluqib#Fko
zcCW;Wd)0>GnQ?MKnvg3ey_z;Y7Q5fh1magcJ4P=;KcEYWrqnrIM)@nmEQzeM>}nH9
zH__NEW&>u#W?m|;w>&i&b$>Gqnud9Lpd60(x-)24=-AIdnNS8v-*7{pL3j8eC8Xfr
z`GQOmr_7QUjK1yDF=mpg==waLth|`8*cC(^e<X3{=MsBc$7Ua2MKJGf(xcj7e}1Ai
z+vS<0dMn-*eq@+^-_^miIWbOaAQa>c7-cwRRus#845r$0w?E9*L&5vF5PAETe0bg&
z;><K(Wnsw|?;*JE7kT9xcL#XHD?ilw;{MMG>zqsEZeEfDz#RB)LW9`A7LvqR=#mjl
zM!5jkfI%l1>HTSF#oL&9NC-9zGO_Ep`|e|}<UhZooX7`~UZgmCnj6I3-tgKwSzMRS
zau`~p2MWhDZ|g+AsglF2&@7~nc4ps=(zO!*xOa^=0I5MxsjkaJkd)A(L$Bls0;wSz
z8z_J|SnwoJC3Xe)+RHl;_prR%a0~4SRtZ~tI}D~m{I)&^*W$|+n(iPMf3v;;16oRJ
zQi*b{mrJAWXvFn4_HlpG-28#Omc3D|o=7gFx<2Od$n=R*qM<FsX#KjkK1#=u!Yq?b
zZ2&c28f7B*BMh8HOYy2IHII8UVpEw%-PY2lO+2O28e0^N?b{H}`n5S`h|GY42qq(#
zmGBZ<OT|q~>~$}6pgwU>%eR2gc)@+`349^bz)iyXM5$ISqZ{|58~|6E6h%4iaLe{H
ztgG{!SeP#w^f*Erptdc#eP?BY=QB95Oi|XxyqxnvzGKw)C*W*KZyF5s*_NtEDOvWp
z+hc&j+skZiv^K?^w3O3D;<#Jk@CpS4X#yPt-#n_oJC+9=wsxf*XuA)Jjpf|079%x3
z&iWklESHSe&PCpiaLSk<dvMpTZBYm3PRc9g7CHRtM%pU8^0|3%3uy^N7@8zM6j+{n
zAjw34+{#e%It@x~2dL=Mys155Es#=tU>RMRf&vD-8;@_k)pR&OEzg&23@a2TSa!Rr
z1E`K;W#3<$i^1vGu%*Sq1yOeucjX1UxO6>uWgo4V%2Xc56x7@e1XSn=ENOfZV?xX`
zgkt5hKNaq~XVZ8J2i2DtCpGG1S(Y{J1_Q*w?Bf>Loo;?<{sx=4HTgab8sV5kBk<>?
z-(tM*AXkz%?!(QW_sb=5<1r(=#x4v8+DF^A=$Q+p&|=k^r&GjFS4p=LCzj<_Pe-v>
zPDwkTH+iJ-?%S-N|GJ0bqQ^ifan$mUdlItDVUaV6ls2=>a=lzgSuQzd2Hzpz{}agN
z`y*_1q7gbLX?683NW-C`)c1=5G^osLrj+B--iXboXFX)PZBFZR!w{$5AWq)Zz$s_N
zI;RO&XHl6Bi~{8ZW)D<$YljWhcEFE#vPczTilUB#dy#-u#$6e-Iu!t9uwYR*ol0fw
zx^E%MKQJ9{kIaf7xnysrksCGc<7CO7VY`S{J*lv_ewPM=n)ASR^Ae}Uhb@B-9h}f~
zoaQWgMPGO1$IdvCiJ*u?I&^fyWiu8ImYgy1*BL5xZKQkl3`N}CMHR`C;qTu#(jm%H
z`p#ZC7K`B`PO?g@&=Kv1f-byXI=f!u4r8(9uCs?N%5ONf$_z0<WIn$s_C|`77&@{B
zo|&19v6(c8Ai}@5jqfkF@A%-y;@?yvY*L3be0fpJvF3B>^p&tBI+{L<n;jVO>)c+t
z%=fya(49MpbT*`(7)(WOqLAP(8x&WOV0MsX>Q%{<X8}%|7tD^7v##8>BDvp5oLL|C
z-OH3$Jl)YPEutOzbPJ&JnOc?Q<3=sbDinvvjHC!EL3<b+=rrXL`&WYcT(4>vnXXfD
zJagr)2MfAGQ6cmK6&um=p+qjJSQIXZaI{SMKJ!rfmL+g?@7*sjoTk%}R<~j~Rm{^<
zwKVUG?ds})LzNh=C;m-dfn<p}-yJv%&e%(x%ZQ83*Ws@-GuWS-sR<-7QRx&SuphEQ
zsILW{`;}6nHX<#Sv7v6r1KW4w4xw^~DP>lzM2|7%Zedjg!SQ!g0sAj?F`PTGha>oX
zrmh|m6S!1Qn^+xIOvq8Iw+3+WKl$EtFg(wM0`IN#vzPwj@xCX+3h{<ZdRf09&dCl<
z>uDmHAeY3Y%s;Jn$H7Pdcg+X1D79Iyu8RS$lfh8-q}M3i7U!Abu<rP-`_wAk>xlCN
zl`Lfw?x)}t4k0r|X_{^px8w@f0C%Yk;=VF=t(~S{<d*ys4}&^e!>wBlncOi)jrt&D
zGhi@~Ig9rKw83!G6g#3PBi?YJJ1t^8&>JgO@YOojMLTkzf;Vd%S;hyP1Qvrn{7KxP
zqU%+cTyW#sY_HS7dSZpxGrh}1@Y&v|vHPnn@Bz$Ci{`+>fVy?zgkVE_Sm5lHiR@D{
z^7(Z+48Tq>BeMPt>tN9>SPQG-?G@4_Zd>eK2DD}0O#CL#NbT$JYBwAKu5Wo?m}uEk
zWQ@34m0d-5627McicK-@lWRWrg_Jt_P<;LERMl`K;e}O4?k}eXoXDjd0}2T*4t@G|
z-WazV-}D3DQqn&Xzu5hS{$qL`p{beahw2~+J6g)^I*3@rvhf`6nI(N$F$PDqT3IzZ
zs)J?cv)|WE;O8G<BCgk4{<Dy4DnoB529dGP9p0^?mD(C-oAgQx9dBnQsQi&*vD(u_
zhF9IopJ1!CR*d}@;0YYcC)?BA<;H|vNh%B&Z2;4BaqE4unH;`83{WuZG~`^F()_UA
zhInr<jY5p2It=6APaLRir-bpcQqf-AjFv3;8WTFpZFQ^TyQ<<C-BUQ*ty{it;l!je
z0}c?#0o9DWjyAQz1gc?fu6nl{_PLe2E82CVnI7W&qWkl;PvtzrJ$vzCpm`*NK&i_0
znGCv-!%`J+u~%0ibNq^MM#~J>{-@G7$ryojip|IR`7ULh1jq?LYPsT9yOJnS5F2bn
z=BsshVw<QcQ6uhdu=t@yL$xsJYDT;nQ^dSRf#f@z&qEVz7CCO>puaME=$?1D&Ywlu
z82)U@pc{R0)xC-yx(P5mZG@TEjCIo$2j)#3bHb)O$OBd*rOGbY%zNeMF#R|gWx-U3
zD@R;)Eg-m0qf*Qa;KA5;OtURdK?=ymY&?c8K-eY81lOR?*=_@-5QDIioZ4&V$Tj2e
z(<e_y^ln-~)=!71iUUYWW>Oppp)wYCU&|6ZI90>3#xFYt$|HD&*9`18uAkuH0?r1V
zNA;c$dSplf6Po$_WYH6-6ZsK>sSM2`u&q-gt_gBB1i?F1XNyPdi~1TUnKCz8yPF{(
z`t}=-tRM6hO^9hVr~FI_Mqt{|Gw=2lJbm4$rtsw*cLMRYgPL-@>7<gY!fNsbDfUNz
z#yo64CKIPmH%eBWG@QIBD8fa@3{{4%j%{t_j887*5#k@IVasT_q7cK!cNOKo`wL72
zwwD#dXhk$a=#xVV{#g{P9|0jsiKX7cqIOUlw!dSV4dT5kP&L>?uQL(xV_ZIF8W=nO
z%t%e+!}$H(_yYT1T%6OAC<>5l+qP}nwr$(CZQFMDZriqP+qQWpChp6bFQ~VStg4ln
zFP@&cI%-OH!1|2OD-E3&;Sv?7k;YU*GluORu2a{y4Fr4!$RIp=A|v3YcFT<IrG12p
zKy@n0Xo@D9I6?e4omkNnPoAFvtNwBj-6tX4|1yB$2QR6)d<WL-8+f~z@9z)3*q^m;
z+GI^z>DWlMi-bUseDe35cW(t%ww7PH)5z@GWAFS)B9mKx5e3;MUM_(CRVZ<CC88>w
z5-peC5aM4}zn&>mwU>?~q0xn~6ytHPn<WUSj@>8Lj?-k+?%nQ}sw6<eAaGs-LH;n)
zfv7gF&@B@(0SF~O$H;wB+gBdsdt(|LmiFgDLM|qW4|RvS__6mHL|o6{#<&$Z97(%`
zkfzJe1yF3^P}K}-7wiSGQ`TxV&Ya(_@fju-mXDtwLncJA7_FqrH<62iM^S!79Ukv^
z?W{mgkHgUqIBzkqG@Nfj(d*q4u2f=6T^t~}oA*CMYko39&@i17E73Ty{Ar<AfNWSY
zd+<jvaq<&PC4WxdC1$*1(w<8blgEXe6whFZGTKqm8j<@sf7@O6DC3r}b1?voM(J~O
zk8^;(ToZ#ial?JqAGbpCz`liQu=#%tpq;EGzGwAZi5bowI4)w)*m_N!vE;?0C=e$-
z!-VZgzp0BSAZmcrSUv}6R5WVW6+W;25y~>wvbX6T&d;@1a+bD9BG7mbek-T_9$fXI
z8bdVbemDoj`_NgqN^M_4V?6udOQh?I_>O!ZccMY$i&JoCkXC;c<8K|Mw%QowMhIUo
z6L@4sWb_4RlN^Ma8&aERw$a5D4?u2OGz65LQtn|LckaDgsDku(;c3_%&xRzro7M>|
zM;vJ@MX@rU&lTzvS^aUT)yB3sOed>^8Aa(#OfR&_$LjcJIe}nvgEMux!c!3&Qu911
z0(*M|{-xVQX7a@;gDdMRc1)*8e`FW;xNm90uJ6<^XV3+cU{hY3tkDLCo=WZz5C|HG
z6a;yshY3j94U67;@H(DO%vn9az+R&msQ$`mpzi;@b%v0;fSYm<!0tSdUhLrzEu3Tt
z(<Mb_`1yjc-glX_-(_@%R{qOuB^$2_Ek!o-$A{KD1JmLdb>>&$Glhf&t8|$hiNFEG
z)6<u}0J-e8UP@3#Rui~SJP4mJ^#nT~0l?PbcyKOEz-MEvoM*$21nQ@!KA`S!l*|vu
zQVNiblCTH361-njNTF+gI?AN0+Hbrl_)tK0&kxi|lFGjySwc2NMq>H9C4b-BBxiZ8
znf_*5<Rim@3~rWK{nu~A4m1|N`Yl&}?6jw%ONe)jq3N&Dd|dY44;IwXB`E31mK5GR
zC(%|f%%m)EBv2*e_hQL@AI1?wrTD0LS+&cd2AktTB%o}A`S}RYGw$6f%*+-{deC&#
z_U>1!iZ4~$2sMzf>j#Xs6^1V9{PBY~Fc<<Kyjp8mvFvwxVl>e)xzFXz0>M<gtAS&B
z7m)1l{LZmpX@U$#`%7s@V9@Rx?m@zb9Vk)N{<i3@d1hy*7d2r!cNI(|%szg0vyx~?
zGBeauqrl<$IgzEF8!3->{E(8>$ty?HAO=Y2er7{U(U|D&(^0I&nuJO&1Zt76xtE3^
zRZCJ=9PIDcyZdSIxTf2;7;;TAu`nO}64@nIWt@vj1lyD*c@iH7{lO6C=J_PD-yR##
zz7}_vH(J9mTvGrwS!}Jp&6_uS_?&-XcQ%L#U85vLzZ#E4v+VLEun6lo$$dr*Li;Xp
z@N<p{pGbnpE=Vew)g=u{ykyO&!9|wP!=x3D{jWI&kXe_)aeE`Oa(~$e(}*jV@ax?|
zfGwBeLR8g4f<|BuAv+>t_-)?Ywzra~k&bXWU;#zhCgcz}-`+MzUaZ%r@~sqB(&k*3
zI9b%mh+8H^xxtp!K_MSGn6Vkrgi45%)Y0PWvG9na_b9l#F<?7To7haTMw)TCcO?cr
z#P)>xrJNmin=I{t`X?WoaaGw(PvgH!V!gWK?#E1D)>a+4r<N;D=Q~|=<YfYR;MK{t
zzG@T7p$SnrU>vmShO07=T7=#&i8?%2eWK3pRg-$v^u*<I26GL5f7^%s?dG$SWe3CK
z#d}scXzCF>9&AX)-07aooom=d>C`N1)AfjG!qS9g1COy>AeQJZwaBomZ@nuqMrz-s
z*ys!^Mtz$9qd(4Ii@Te-!!7X$RdGml7?ae)1{L3Mm{D10Z~4&JtkTen#vy+OFv+K2
zYJ)>rHpf5}&laoW>;fi8NWY-?Nz%yzhH4P%Lm`<@x055f$4{<s3>-5ZnEp8HyQpZu
zu&WdEtmeTOBNM(V+Mb@lz2-eiZ*{&fgUHOGMc&jN;0=N?=)yJB5<`2ILg*^}N4U4$
zZF1<Bp|}g43b_X&l|k+t+E@zGxcG3zC}g+Hect)#F=F~<=%daVlx2<lEm8V`o9Ex3
zCSD-*)gAtAD2;A)q94OE&V^=P@K<2^F;JcQYQRGc3zN)~(y5UV#D=K=+rLV;zAqf`
zRiQkg#cz;Orz>4=Xn!sT&(L6`8Y6=(QY#6Vr4+-O4HF^e*Mzs3UQ8ljDYbqQ50+*{
zF-SC{%jyzi5IUKzOF{tmSsJ!SAcXAe3SUz#wEjU0)gfUlrfL-1QL(d^3L=-BH3<YE
z`t!csr{{=a7>TYsfV7;<hdhmEh-5eq2+o6)+|lDv1fQAL{uL&I&ne~I2?F|AHu&Y4
zX>gUN$#^V9Wr9<WJu@2!_icydqMI6SF1+H4--<@k9Y=2?1i^cViZv;k=`-(|T1f+B
z?xH%?bWLgwv>A6F)k5Cuybh+sjE|%xlSluoBDV>^>-=OvM(t<tlLH#uSSwL8DNx3a
z_qA%MP6|s(_6GbixoYcrHmRS`^!IgqEC-!^k8g&C0Ntt9=n?jBemq|gMSeJ&mmn*8
zQq>yiYZq^u8p>;ZYSQ{r?l3|vI$p<ppYX8>X-I`#nRvTXAb`nXbdtiXE>COFS4L*v
z_>yMkml%;P3)V6dQVXW$H7xA0!8*U+j}9Q2Pmm|OZ;FGa5@U=lDbVPQ33i>p&npgD
zEiQwsn|u-wRu<i<9smqr=2VU?m{2!qhg}<5{5Qe&+F<^CO7&S0%jEE#9&+zN;;isd
z-H$bCrzr*oD(*s{SZ69G6dO5L;~AT~`>uAaMR#p0UC9C^<y96Rau4qBsOfL+RF*zA
z1Acz@WPOq=OkzS%i|kBOH6QanttE;8K4dnjqsgY8JtQ_7y}&+?O)rm5lx0FVNr2r|
zZ3==8*|E=t3wxiKC&<~S>$Kj&*c2Zho#tc4EPO$lswDvPuz6j<G+xY#jZ_~W(%^M=
zl>%G!N!5`C(_1tkgPp_k=hXEb?XT-OtCDXkwi3nppJb1)wHtn-4eskiESxnGyoG*i
zQxzYWBIH3&b2(2uiGK;!cmG;eyT|{gV?l_hzC>z+%{Oy)GKip3JHFM$%I9cxt7>N4
zA`HY{cl^AWW{i|*a3+Zj;rFguoB<zZiXO~yO3bt=S_MmQt&y~uISzxeN*oHOW{M)`
zP7;Cz3d8AW0=W=Sg}heS>8;AoT)HUFu?9`d<7>rI(_2P!<2{^`P=0cn3x>~s2xzwT
zv`~P($-%Z15dbDW*#i58r=gWC_Z5PsnPq5j`C8+1PWX!%fZbXS&d9l3A$w9|UYvP+
z8`ahw=XbrbJsa0IoaMVFUvx<*0ne(1e;=N$6He<T{iGzVF1nIqlUeNQxX!;ka$FgI
zD>TnK5049$qG@j<@t%rGB^9`6`K$OPPkI+l-Y4DOEo87?&A;n7qn5F~QxV@L1U(WG
z;fr$~1u|i%uj#zS-fg(sz)#vuLpZ+Q)#>LSDI*=@>8g~TuMlhp%!*fV7wqGUQfIij
zblwLX#C=c<BR1bwi7#7sxuj%t3mZEb9$A!S>c6y#Bh>p*^J4VO&%45Atz{gg&@C%q
zSIRa7iXv-tdLrU`&v-uEIR*xw{-ohWDTiH8mYpX+hd*=Kz^ZY!H#!<9S+6wf70RCE
zos_YdIcN2-rsdEp%vGGLPToSsoZccEHQ}DXk8X$DVf8@Wtw^uL+;&RhC%7%p1IiYS
zf=hy@#3=0cE4t)&-hZ#&qv}de1Ix-X{09DNyVKDAVdLrfzIGZDU$UxyQ{I&-M7j%|
z!%mvt?hMPYrn3iLkYPw7>sR5!-KCF#3~aGov`Fp5@a0Z+3#NPNI47=`0r-OVcr8Gr
zsh!57Lh1q7i&49dTbCCz6;l-!S6IZ?mE+$N@T1!!;z^$C1=B3j*}&X1Zur2`{-Fhv
z{#XkqRX!jJ@8K&fsbuVfexvcO_$gOOVq7f`*qhZV?E)=cE@BdXxb{6yIva*I2AnRu
zlEAS2<>(0UL}<JZLV2{GF!SQic7--@#D)oIut3Q_7w|>fb|a*F@-MS<ZJ&KSQhax>
zZ=S<m?msF-YzXt)I|+xB26BiKXoz2^kMFMEH|cXJ)fMhiq2NdxA$#%tm$-XBaeqTQ
zJGyVh4@K3{8G;-7?wec=sER#5Q$qn+X+9LYl<vu}Qb(46wwR0eDz0cH>4pQKbhvm$
zz}NcI;WZZ=PwC@Q;5*G<p|@0Fc7uvU%x+h6aAZyZx{$}5huS&Yy~waAo*5~TEF}rd
zN|lIRDW0|*v&PyJ$8h;;!taiE-Xr#RC3=R04De!6G98K5fQ6~l6km)Yn;yIU{L?G6
zve6i>LQnU?a{D^<ccO7BZ!!%g2eO&P(EDfODM`<dhLD0L<F7J<!6;J1#F;~c$WB*2
zwoy%p*ZMaJ{f?oOyOC7f;jpaSD_!{4=r<!QSs%0w7Cpml4UHtSky%!ori1G68_7Qe
zu<eW=fKf^2#tzH~eH1=7So3^U<?L?f0O%b^74#=KPRsplKM8f}11k&XMBI`zSl&>9
zcn!SpQUby0W5O)DT6ZbPhS3(=?NAJM#q`(oTy!pzxYSCirw3Vv%YEhk4i^l(K!4e`
z(0gPaVz{gf4v@EGtDUO}sG2gG-Wbw*_LkjTZjZHJ=(ZeO2ucwnaa~PET5mJ2&LJF}
zpSmVTye+o&X%ro%;Fss<O#9(vLqj<Jv1Z0r#^~2cu81vj|7U#8`A__Wf3pYR)EfZ<
z9>|cc0w-Z5I>PSt%(>@=cng-R-favGqCpA~F-74Ds9;%F7*9Zg^2c?O7e|kO+kvz7
zW9<sZ1jnyR8LroEaS!Gx-h(FV&;{@7Pg8HSVJjLaSZHMvB`056;=oe{hiu6wGZtAU
zrt2UD+l!Y=hH*r%Wq!+HmCOS=?|B5dEq6_+5<4tg>F0S>W|iV@8-BzLC;X7Twwu$A
z)1MoIw$e}EW+D0jRoPziDGJ~Q-2RoXe)8EFgUn3JdFAi3EF5tyi)P4ywzzW+PycIQ
z^|w_Z8l)8wo84h!VF$(@ML?8?u-8wCvM%v64y~mYYQ#$po|%>iN&*HU09Gc7F!Q7<
z49@I*Shd_JHy~gubk>Z=8N{~Ng{Y{B;guG`$??{ARNM$a3Rbb@t0+W*_=D_i#bO%B
z+{1V{#LXW=5QqhDiZ61W@o>o1kab#RWLMf_xOulMP6@neF1WpnjGDVGyf|i2VE>T~
zG$3yG#S4h=cX`9C@eQsdTc+V#t*ZbPS4arVSpXY#Y%GX_XYSAOH)T7TlncKy%5(p;
zYvUrE8z}ZH!bxsln3&iLMI7@PAi;|syTNnAg`O0zS&>G<zNCB{FZokm#yTD-JK&GX
zsq#|DYUk&;X@cgUH0q3H5hB%VZ*2FSUr%3d?b}(PbD>+k?PO>jeqrX;<k9nBE2A!W
z7nuw6#tXefz#{MnM3jqBONwCmhK(F-TkP#=5htO{Ii4&48NAy7HNWrKaEN7kg)91U
za|-e!Uu7fTCkY2}y0T4?y%=dGXoENF7Xg7)<-bCYfw61duaN3^MK7j(e5LvHVj5Dk
zgR1_yJFG6Z<z$(^*RV0o1L;QL$Ih>+W*EJ20HyTRK(Iif=ClqEqqIBwBkQ3hK_9q3
z`YYJZ!|T>QOGWU=6FkR#jF!E43$Djfkt4idA^Bk%p`ykboee}W+0!qMx9-!%TZChg
zS=I$c9OuURAXw$hOyY}cHP2<R?8_qLHQ~i!$E7cD)3tQAS>)vdao-~X^GcwT@oPjr
zs~q<8xz`!>uWCfbc!R;*RbU{@$@!^;ruvYj>?wR9L7K&Fzi9FjH^)Sde^SezGq<Ee
zCBns1X2^0FgWra7;5SB`yY(HIT+6zp<`Um%wVpfCAy_g*6hOs6)tk6uSF%$jwE!#u
zEe&XBZVz#VE?}}{gLg|R*=#AXm|bb8#}H&2nm`sl&%(IBmqV1yj(9c6Ff-GD_sYo#
z>R2BK#Lk<A;8qLoE{E2*3Pd$VU_dOea~*0cSE3BUJOWb4Lm@KMPNnb6vstqQ>|kAI
z@0K=h0B(4<u%0|=%fuZUU{M(YwCY+(AZ=@AT+O8lJ@mjEB_8xo>BZ&k#&GS#ST6$$
zwD?J$(N*i78naL53>5kh@{9-WC1Rh7|1kF8zTOuDd^Di46dWZ=J24*015b<94$JN|
zs&s<oMi7+D7=!Ds3&{akQ%pd{2NfYgH0WPGdTUh^4!%QlRbl_Q5En7m9Huk0%>m`d
zMAC1u42+WdG$Einn*IW`%VJc?1k5(x2t3~D>+`eKU+hKkOUugKDni}DC`t%72$bg`
zmZ42(tNl`lx!$5Z`AyqW`%dC=0>)-CHGCf(;Z>U1By%3SKG$z*e;;}5KpLWUNnS4;
z5`XUfi7&r1+Xu$^Vq2KEWy)cb%I@?vF48C`^I$s7b7MlCTKBuJ&5v$bh3T5g5Qy7c
z_68f!vqXIgpQ%ND?K^}L^|k(sHBMc(PlyU9117;U06{>$zbxB$&BBUYOWmr}+kBiL
zzq6L_Vs|P=paN?D72>6NUc#U;1WREHJU4h*nrYBD;^J*Xtb740KWc{E%ckqRlnmnK
z<TkHMF6<Uo`S)!}T%dllm-@ByNwuo#ZMoM1f;qaJcn`@y2y>3X0}a?3LAIfWZ^>#8
zjP^zBMd>As9y$?+x-PijJuo+&uV6^|tmW-Sx}(;u_Og`~I~9fN)?v5o_rlMTK;&X%
zfeuXIACuTwyRF}!svL};D5h|=>!D;yAll7Y_<Fd;)Mr_)dT2^!vgk{h=kfA_GO?bD
z+h1+cisfVI!`}XJ{cb$5WsXEkuO=3gc>}X{q1K`T^_EjPzOgJAr|&L3hx|h139{X+
z#C+hmvKwAXmug=@mIut44l=#64S4%i8fog!H!_acf$TI}j~!ysGttUxst^4R$rM{3
z!O}PayzQVM2O1;cC|Aku5D6BRdAbJ+0@_e;)){Zn)z3UhK{j8c^O|^9+fd}h$~`b!
z{NpQth3YoJd8D7&QF}5ym@bLOhZ{h3%ZlI7#KY!z4DX8yv9)LT$+s`&FPG0yiFY91
zX>J}UbsHqb-AUm<=iCF$%^p!#4k2XOdXUR4|5!}Q{cq_}@Q2U$XRNwbnc483rjL;<
zgyS+l*CJ1))o`^tmYa$^4_~r!WuZ(s7BL7Qu{2j4iHtpdedBVt2c-R2t*kLPJjIU3
z4_BYi3t{!r`Lni~!j;qsze4goww4GZOTfa;f}buCa>9ECuH3N={+_Zc1<N{P3Xk3;
z+Lolo!QE~b9|!ehZfsEChEP+2CB5x7+wmWG4Hw*;yY7Ji-!vN!Co}HwQMGEO<;=VG
z8lV6Tib9EMNni9pE`~)87y(AK`g%N}ua4NCi=(4y#b#_38fEa`&^!R&9>P=7c!r??
zF`)R1U<w_$(49uo&6XP#L%hk?wmL6>ihSxp;@BP+g@dMKe$$*2Qw79*ZcBmCd%QAM
z`TdpCs_*t1ux<%guZghvp~9)jZGWtu@ummgib&<jA--%bi-<t}gt9BDEr7IOip;c*
zOQd{!AyDdVhzKSV1{<)F1;BS4lFu}IMrq<`{Z(!Z`9@-Xya}dcwgL5iaraMk2G3|z
znM@J$;h^f8FMaT?MGtOHrQWBq?-Ssx;DLWJFFuj9mvipw;RgVlTCRj{-&jY$XKS^4
zT>>I9A|y#JHNjzHLH(c<vQ1+i-o4OpL|^s`(Io8+SFos%3ea^(HQY-<!^69v|4?sS
z=qab6j=L~`ywpN-2fTXAGQI&FlfkNfZbB4u2cT!B^zQBg7fD?QIO<L1RQ2Or5rh@C
zu;;MSEn)F^_(nXVYm{?c8d+RGby%Z`pxDH*<_r~Cv$v+J&g{YixiYO%OZA>9n}Y4R
zs4qGS9a{Am0?bwpUozPtK}c}D()7CaS679cK{6vu{h7V=OgC!FR}gtlI1%YOkmwpk
zP}^NV=G@n%j!N#;LNO0W^-pc`yeVTpz5C;EEj=oPypDQLrpJ(0k3)^POJ7c4gK&>T
zv+8GbO{;8;7;Ym(!(ow!J0hq?UCHnueceMCKrM4ZqR=Fg83Oran)oeUi`K1aj6xOJ
znk%{%t2Z1Lh#=>y^4L=pL9$>);KB;UaE9m0ai&S`+6))Ye{nH>z|>g%>!dbad03Nf
zakH|)J0}!FX7QSI($xCCc^9j%z;;xHUx-`m!f&=X;(s*Ikw0$3?$Kn8NA-pEUar`C
zeJe1ZUXU2)HaQo#e6>i=K`O3%PeJ}S;xLlzFw(wKPZ}Q-)*-K{UB%TD&veI-S7$i?
zcID`wObL?<0H#ezde}xPkH-he>`X{)B#6@Z**gzr-w|+*opsC-l~_YVCNVT#r&2?$
z0r%e$AD=oTE^CQ)f(1-%lUGs2)0l5R3VF{jA(KowB57w&$CLzIkEeKXy_Uu!img;O
z7QZsH+kSpQD|d|TIz-DgMauj7AnC;z_PFILZK1dZA}6_ZfVd&5plISCeyF*KdxHPR
zQ_MW*Zk2%M`8~h#6_&8P$@st8yNq{zCmOrzk!GMfh&p=;py0;ImNI3|a&qXn|Aj7h
z_3LcX!H^HZepP!C8I>&lIxWe@Oj_$oO~WR~PWLktzd-eOlO%5GgVm1$$!>5Pz&%m=
zt09gqLwuCR<?!9~&{;h0+|mQ0ZGz$@vBZ^vBBg6%89<3tZz>_@6-_r()j1aHa@QJS
zk)7#%jkKh<_FA<9tDSS3sxOYY>cJMGXPVI%_Ms0{6sG+51i$W?@NXZElBjdQn~)KW
zi7ghg!hllX9=FD5FxS^*-g=Ohif;S0hg7k(o--<3LuXbF@Hmr2Ih@k%M|1M9J3OOj
z&XUb|Xuvc!07S3Q!4I>A1}MDmx{QGv<X6`RAx`v(%x4GDzZ??o#F#;WXMzJ0bIb$I
zoxSvN;>ekW7R6r5(s6I1tJCa}93dff3u>>0iCb3Kj*Hh0hax*W98fD^8LOB-F3s=@
z-y4x46CALtB1XBwO8L>G|FW9f=?=dv|KP+LYi_WI(Lc&1Yi?zk`TKKYs^XR1<tHY5
z<mN{Qgv%=VP$daFvW__8Hn^cTKFIYqYtCC~Ka5kV3nP7KwQ_``MBKuu@8qnc>texF
zam)&T2p~AI5h`8O#Sda6ap%8JK9c1?9pz(NMuT9&XBCUp;dS*XMpCQt8was~<9XK<
zNU)fFvs7JhhV+8Zr;75rD%$<C*@LBLqkw7E?U5S$s*N&XAhT|dEe*tDWP>$!$^v1?
zjbhys2Z>wo*vxi*4iMNZy}hop_B%jyi?_cx&H(sl=mX16n(DYaLn%RDsC|w9lT_B+
z{OXZAj-ztFfTAKB&^GM133=b7sk+nqEsU%cuoQw}HqXew@y4}_8EecN#h9cFdq>iI
zU;bwnvo$N<M!kTK-iqOy0}r5MEa=DPrC7KYbgidYVg)M4B{+(&QCkVdo|6^*uFbA^
z{gpzSo_gAOH2$JYcw}VrFBzjBk!k+hjS(y61S-A)Kii_H7A3dq7*gRt$ULw8IF3Cd
zh-_i)hQ^!qGPaJ&EhS9a9NUqx^ervrQI|Bt<?cgmly8o0`{O=DTsj6zC<bqjnA69M
zw~Xj#dLda5z0IZ~{d(JVoKpt08j9e!Fv4Y-QTWR)Tv~e#(LT|v8NkiT`)vB?aKfFN
z*FDFZS<T&9#|W~je=+@8r_g$juZ$3LQ9{yXG99;pq2wbYW1YuAZ<z`r8R$epYrA>y
ze)p@j<=k`U$}&mq_UlUxSu(VHmmjpk08>mNn-uQ~YI;_9{k>d(JbjLebfoT_QbuoV
zY5X+7Ba6;HYEaPIjoigHU&g3M+aUv7TCXGf-2afv-F*P)!)JWSs46$HtZL@L?xtzV
zV@H5xZ5tBed0bS+FRRbAEGsfF{58O+db(-p;(V@~jnr!!X&>}Bh(E9VsGL$Z(5exc
zz~5_S%(a4`O*y#&OQ;m)B=|0@ju)mqLj|VR{t)xtBIyah)J$>o{Bx3^DMNa%;era<
z`P<<TBl?}pc+wf4HFToM%5^TP8VW6mwh21JUYm@KpmY(Fl*p!R-(gL#K^xpLJ}&L*
z46bev)p15^JG!{xJmv0xOFd#2>hgcuTSlWyNT$A2)Ic=bW(Qq*y>$!!{VnH*t#Q~9
zSOi82Vp0ORx+xaxMEPrNa|)Aia)^d(*b1>R+1U;ii2VVq!*VAXD_doDjm?6NNaW2S
zA0eJ#HT$6T)d;8A-ZnwJSqmdVkf5MJQ8rX(nrMTHMNy0@9K8<3fi`{;ubhPGZ{y3#
z59s=SX&`5F*w&yp!+7-ZYD6taSF^_)U`e9D8y7ZOuhCDTqN{&L_5BGvV{hQ4sr8M*
ztO!pS2TbcP(4or7ROe%OYpE4Iw$@7_m4){_1Menaas%2$7qiedqe0U|F@=gXdqVHB
zN5~v7Q>1kL&UaGieO+}8=8$2vh9uzVXbImqQkc?Py0w&wdPj2dHSypGsk^{+Gz>`$
zL``~jEN^$l5;_f)O4o<e*tYrF)r}Yp&RPzCN7K0aC|b)Imbw7LvZ<jEbBmSgtW+P<
z5>V9gC>GIL)_^j*v&mkQ@xP9)$oOwppuVzykJL!ht#d1$|5xOZR?CSUB(fY{rVA`L
zi)Zd3w^pUD5kN588MAwGVy9qIq`3))i9-*~amkg>HnTe|YjvP(^7)4N-@8@SKm}t%
zvCl%$3m_uE&bF%ihC?DVr&-8lx+TOph-K%g@2XhSi=K4M?G<*dU~<o~(2d>Vv>BAT
z&W)q1-s%t2tKVUu@Pzp81_NwYL`658LLA2IHD>TtKpMo;8VlW`FkWivx?`rh4I%BH
zA9geV2YJJOU}m`nOF?C%yoH}uGDAd5r;K)34vOIt{*R@Ez7BvQO(o0<s1>ROS_g<>
zoM29YAyjn|#^uYmhqQZYaL4<^W@`P3@O34&ee<j1_+UHq-&o-60h7@VMek#y*L&Dc
zU1o4!-KNz%+fX25CYbiq1R_S%#9Y^bv5V=<W;#rh?;Yr=Wo_cnCC}$iL-k4ZqeX$a
ztFTkebymS-&WT~rm1pw$J1HW=T|9BO2aU3}IhExN1=$1No5@b2GJQWOlKXGqlmSUk
z`rAt1&1G3o)7*sNeq=c1YsO!a{&93>rDBX5C?jG#L9^><t1OZCVN8%;^(e=e=TXrl
zzT#anvtbITg(L~0StujwWJ4H>ptTG6wAh5U)|%@A1^Qb1mAN4}I=pfZqv7xy!x?_w
zfTx5Pk&CcT$T3lQF6Mt(_}m00PAmq(2mkM{0c_DEg>{4IW}())agxJk3{j_M(B#5`
ztGDz|$qw&B2Pre^XMi5=Apt`h6w3Qy<`wnOrPxzhJI`l&%QfQH&b3A#5Z|0y4rh;V
zV&dfWGW!%|Cs8Sf@oCrF{it_eR$$so!Y+kUn(Nt@f96QHdW;(b*u2~smTc0{=EcjL
zBoEaze0}1=sQ@s>$f-hI7Vlhq&~E;wYl1VSK%y=!U^H&Edy4=Za-(F~jJEQ+r7wG@
zMH-q?N7eP`{BqX6SKo@`ZU;G|5)2@<LL=cb=ioU4PN$y<xSi6?(<&ds`DUM7bz_5`
zo6@%23+DfXZs-JLSHu)F&j6^twMOX_fYzpk(pmbZYMqUI$gogwg-J>NvVWa9+fS$~
zP(0R7pGQbs5+I1XS;ILOn-M3fa`4_pg%?9=_<xms_&~eTR?l9_<7xO&n(1tgH~#9+
zk`I63gf*M7nbhloeIK`mlV$z71T;nB&2}wwYMofL6o>aEx&TUY3h`|LNKwr3d&PX+
zPAfMv;uSi>gxAUTUK_>VT3Pf!t;W6+fBN9Ut=M1ketAa*2&q0gI4Q~Ffe;5lRR<`0
zR&%-GK8gL9x$Jms`F=|++fqRh*6Q2!{6p((S^c73TaLhs2|P`gET1l5iaM{bRyh+5
zwdipWpvz)H#%tXHVPX>&&XqG1da^YC8@fESHo)VG^Hv~t%WFxb?nfegJJJ?nTDCuh
zbd$k5j3_S2HQoliw$pRiM$D+J?GdJpnIf&Jzv#UlCtdfIB)3E5O#Jz-tOONLo~y`T
zrOJ$VpX$c7k-#s-5zm5><sPv^w7H#Njv;a`@%qEpS^ppBq_<9fd*6uQ8%Ba7&Gmp1
zKnaZ&X@bylpF*CHiH#uQd0j=>s*DKT275Gx)guRSDgngXgCOg9x0wDX%IDoQY)Va0
z4<ABv>_w}U{xwpW+PZn~**RuH#5&)4!Y=ha<In+nBaxc~-u*MWQ5Fd!;#*V%YD!l~
z>HLm}Uzzia)y+wZc7jl?piuCM!ko-^<y3=&qNRs+C(4X1e1?VT@UL(SY-2AHIDK4_
z5Dc4+cp=EV!L(21F`WSpvPi^ue{PmhgSyI4T7)ci7><4$RWZ%Vd%!nRo~2BAyoc)&
z_xPAM7-xMef~O-^?MWbmakmr5{j}78<btamC+pw2NG!#u)6V<!E9O@z&<hWsJP%nF
zK&MY`aO29<?~J*xQH=uBI|YIR$YVB?+-9NE1BRT@d=RJKEi*feP*u%U#HJoYfve;s
zxS*fj(=Z-!o5M@niof*nApdi0DeWe1K$o~+X>!u_w&Q~S#5R>WfccZ6C}94{n!i*;
ziw(N(WkGXsT(Eg2R)tNqLQea<y!~SJVt47V1*7h><v#}Hda7-$JikB?TF;=4EcKlQ
zOg|1!2eC_u$tnZ<Yb-nMM7lvFxs{#nsM&5J=8=<fP%#80OxY8DctlpY+D1WM+`;N7
zT9w^=hVJ5v<<@uuLMWBlvCFDmt=1eZWEXp15eGAu6-eKva2eE<A)yWgokzN}Bd3OK
zd=TiDQ2vW&ANZ*0H<DxrzX%Lro02@8dj63(q~2J|Qm}tzHL3q4%+uLj!?lS{c9!}H
z>w~Nzz5a+kHWB4^Gy`(Tw$;_e?Y}pCudB&`S|;?d%Z>u9P+1&e*y9C+khK!nD^{-v
z*iANwO2bea+VG922M2GCIi&tHb98UnvQ17c;|Tq6Z6lP`fQxAUaqImUVE#)U|5NjP
zgT&x6S<|Kli{3Y8t==~oC{*b3ng#g!Mi`<g7h@o`Hm8Zice?zms3ZUPUGFJIyeCrR
z!ZUjo^#xSSHM%6yrtzH!82O@+Eza=pQ8+;zI482HFdBp@#A8*$hg|s<0sXHD$b*P*
zVqotlk~Q>^=EqIT@`)&Nd$98i=2S8szeDRK?(5RKrJD3KjYG8rYBmFxpb#1gYcI4&
ztQfE9efp!jBERQg+?za8c&@_rJU2N{;@z3tsLdicux6zq-s)Zni2Kt^o)@KzD&@tC
z$wyGHwe&VK;Rap_uCW@N&v+*%St)@3)K<HR-xG`rGkC%%DB=9sIml7ioK#YbR{^wV
zl38Xofhv1^T~}cDZulxe8wR>`42dw4v~Wk@f}ziEb@_lPeJw&2UPxtCh3pYA+u7;s
zvO!<BiT-@!X^@-taUiPxi`^wqrvU?c#|ON6t8xj_FpH*aH2SSJkvMyd@wMX#B1vd{
z25}9JCfE>ATG1LqiD5rU=WK$bSz}l$QCma_<&*MXeUfv+@aJ-s&+qn4nH4^$_r&qO
z{t^7=<g>{k2T*eTRNjBj0_3KQ%@9n#T)k$sy~W(I42%TQBg224I5t|FIa(D5#(jb;
zX5Gl>S<$6HVrdN_DNP6{bCbGLK^<nz1QzsVlrPoLrEx`IfK^whR)x4HD?D0ecm1lI
z6M*$x=-jvi`@U}U47S+PWj|U%^UMYZv(a;U`(qhVcU@Tpx|~YEde2`3jA|Z_!A`**
z(?;{hxaeAk+NzwA(n9Fgoul*M(SIw`H;j)Bd%@2ZD{~VPGWuIZlXacg@cXid0<Xe~
zASEc>eRLaR^ql=O=?FN%o)r{&NV>)m^G>}F)ApR{NEwO2NTe{>&F;gxX7&KKa9?<X
zBv9_pyC)hecIfwxf{%)4h1uMeFj!)plt>c15?iM!?MS)KF0|;qZy10+i-&Mhdlrep
z@e{8zIvHk<BT6gs5ywmywGIo8C<s8=f}jkzkgBd(qc7)B#zq*NE_@!-6bkEf2lado
zGHST{)s@z=L8s88PB1JHnfM`0j6?!4%<|CEdC9F(7GksVGgYY;i&+`9a+Q-oSQAe(
zFnPI*)EGhwsSHQ`i(xY<7+rKd%lGT$M+zqayosY{n3^`nWnJf)x%vM}C)B<uIV#>P
zDXP)-5%++vi44W(&C^FtNb-ZbLF|PsZ9876p20y1F`+d=Xk>p&8z6P#u_1?+cuGU3
z53f$nI%UTl;Mn*1M9u+RIc(w)(^K&_*0v9^R{}c?Z}x218y7Z|^|?JoR?C%WWPiG~
zn|TbCJUmHw;B>!hca<=t{{^VeDdB3@7M6PsCD8mS5|o~VF!OK>QP)c56jBrLJQWx!
z4_NS0S|uzh@+8d5Z%8bF>!N|c%Wg=ZziT8t>LnP#Rja6EeV?QcJ<@6j!D6Yptbzbg
z12L|%N9faaVpbd+0=vP3ze|to(9>*0019svnL}YPwQMR)f8hg|QeWxWSIrGZfj+L6
ztDVPw7&cq>>VS!7J7%aSuN!)evps3A5>rDv%eGQIZa%&eg~J1a#H&uTIbT`TY+rbA
zPy2t6@X(^LZLB~<f5bq#cmjD?E9P=Z=n@934`gzy36SAG$+s)W*hUJBI<JzVqCZme
z_ZsdqEbu|(<+S#%;#Jx#47`qv0J$_bjGdOmV(i)-1|6ChNwY4}Puh@QF%cBomw!*-
z>fl|IoU|SQXth4FF`IvH%8cx3R^A|{73h?Prg+g)ly{ngaE`^W&{%j4`u3myYX<TT
zTIy?vm_LAY>@a(Ro<8%v?7i6K{RBuU_egPogBoB1Oq?C(zl5<DSnXgvn1;t?T$-0_
z2{>JwizN?z5#Bd<fYtriE66JJcJl2xs?diG<Qv;^5qHc_0sG<aw$j4;1!+N-AE#Gw
zQ32tqcg9z8R6Gxf5G8t>nsZFdP1PawtN+hQv9;kdfH9Pkf{hSZi+iw(%BiL97b>yr
zTi@;3vBcuQt*@S-o&UCAQV#)Dh=M~@HH74}G59$K^ekR{PRBf6#Waf931A&Nt0N_M
z2`Ya5Ri;9ZS_i}T7a&9yyb1BFIbgYbOLIw;bpQRSC4{E58ztc)>P($=XPD24;HuGy
zz`PQ^hZE!NqYBpDuU){am;{!h8jf-*{3&29<#<kPANtP6#X_zZ+5dq>SWLtmfnv#L
z+F&(EzG<V{0ACe0LQF@WnI?E3lWbpkO4xvbgFf#<MmZ4C!nvbXJ&G(b3U>U<p3?zi
z@4N<0RPiNB@|Pk_1*cma=`4$WbWIVO<&a9TO2qcKo?PXa%7$Gz>ImDaNNYTFkQut{
zws=shCrsj<&XCMy`NjBG_@iPF<Pk<q<Pd`ejU~<xKR9SYGaQbors+VXSsE(5K>xm&
z4Cgb@aw?G4H1D=FlkXsT3k+CvgqTk>2j>s*N5Igx{03FMFgaF4I&7VW9Hzcy``52{
zYWi~(`p<QBR<htJK%_^)>}hR-ZyILHG<Ou&Z2HPi$@U<?L<KW1ZQ3Z(R?>%Y&HpkZ
zYc4;d)^Tp2CGzC#R@B^#0Uh0sSqtCC$2zlpU|V8ILIOV&X38CfFYeys%C!-QH>3a_
z(PLwz1R%LpC7T}Usxafs)5KBqv_(Q>yB(yi3uL8C$}81{5VGCR$b8wt+h9JjR0|)2
z^l-M!Wzt~%r9Lv;qSrkaVxPc+UiAWek-Z8(F8IFzH8-O6)KwwpAqtF-K5_u|iIx*I
z{bUmh{g@7+Hbiwp&d~@K%p}gpKJW@hk#;^&YXr~(G3i3F!aHgBpn^(aik>gJF{~t*
z+{p^ud%v$1uU3<h8WF@9x339m_~`TeF#**ZG@x~jsE5lyZn`ddWFsoQa#tl?zm4QQ
zlF8Ut-3^ZifQ@!t{V*Td{B2i3Rd3#`1M*m#Wp381g9cV1vwyx^3fpk^+Q%(VkZTf;
z|M=!9bFk#`r1D6_3)>%U3)-*(7+3-+G#$MNPF?#vGI!i4fRQSJ1a^rmc+!yHZmL`(
zHMM0rj!j(FgbUqq6<9a>qMyem)rj>epfJmp@9MF;zbcL0X2r(s!>Q(Gf==$nCU>3G
zWD0=AwrY8E)0@Sym!|RBf|<WkwK(BqE%!GOx@c(Zqhhc40-ZtUA#NN}fk0=%ZDvHE
z0K^bH287+l$<1Vx@vtRu>GSqPu+4|sxJm<E0&}fRg1FrKnLkNg_c*+9b{h6#Yc4kS
zbm`V^yq(H3Im6vfk*r_&QA_wqpaZ<XI9b=wuVzHVc{(o<;NAAAUlg57xV(dOXiWBI
z6*XTX#pCgpvFDX(=E%UJJg<^^$WVU;^kpKNOm0eN4HwXPuvXpHMBm50nZsDTWDj_O
zygGxH%W|f2mQv9SN|B58YH69Oy4sKdQrbs}l;PDnE<8H7T!PH4Qu#tjIKry&B|0M?
zs`?z7aseE;SlcOobqSu&1@-n$%_JCsP1Q%l;dSz1j}xzte=9a6fED2n7atC{Z-?#r
zOMZNlV%%5?Y6_cUIlqCK&bWcUMF0HLmTp*qP_i&h;b&B{_N+5PV`^|U92rsGa%J0a
z#OXI9u?GD*XSZVbS>k+}pafNbEjsf(hl(|vgid5-)U)aWW$=?7aL&C5s3xQ7!H_pF
zo>3B{JA|ihVi*UMcKL32vT<G{7&T()JduFxl>WJCfyL;Pc(1!SX;MJ+#}^?ljmhCA
z@%D;4LOS%O=`R#r;0G$EO=aLr^e1~gbxXn$h)DNb*+gad8Y;@$J~di9iY#1>wcDn2
zo`k)L56c=`es$td{`qFT%3d=@mQq%Mdco6R4RVBS(dMN;czgGdv2r-FKjwm{|73@^
zs$CXfOiuj&Lh7<#H&Ktr*&O}c+~K>Rep&SDq$959<pinRM#9x1`{1B3;4(4C8TVyM
zkuj`(E`PE}sLB=?X?{^`@p;$n>2mCvM9*xi<j|nJ7UKVQ7zqb{;i>vrKGS2o^WiDD
ziAWdcU#*#&*u1QqfrOw4OUQdE7j<38zxbC`XRy7K@FKizmt3e)KhAqm_Y+h2fu^00
z0pZk?8~l)|>gxF#SmA>K-+M|x`kRiou#P-A%5AEzbQ)b|?^#kdj3v4M7F-J}24r>s
zamZj95|_l*^J_|5G==tDl#Q%ovD;ggN8NRV;s7@01E32TQG49M2zJCGWtHfB5F~LF
zQ1c;>0@*4)fp5M0FmC$s8;U!muCSM-x!`@d3r_MH;CfyGs!@-6Z`s5VfwWD$j1EGV
zu-;jT#-!U;(E4C5O6}WFWGSXP#P34eW_9eLK5UsY4&s&ubAN{L_oL!rTYt@2TXq&{
zxmcM}o}aBsl$USxaRO7uY>(4v*p+$#LXCn_si;JanSl#J5+hZ>brxax@o7Dq3Wa*F
zx$$b(ZKR1_JUedw*>h}0Yo5j~B+V=J6?|k8eQ{>VrTflAfT`EppfyNAso_Fj-2?R%
zw5`^<Ij=%<{rg60H8GFP<s;t+BVjH5X|BDu*Q}?TWu-kCsog=w2Mz?;M86ZYql2%`
zWv+%-Vn^W>ZwgBlP1Lc<ui_8D3d6lD7)^iDFXDmP<ob0Xv<nZc1b0@F6BS&8n5Rhv
z_*N7bP+kB;guM=ZlY7s{s{NGgG)kCZJs)=2d#`i4N0pQIY%s@@(Apcj?ucl+<fu=e
z_9`2;!jn$eFHfxT@b74euXNZ`?PwFrCtZba*MkJ0lzqSOq^Asd;V1~f8?>sP0sFFu
z;gQ}5h*}Mh896M<8CfTq08$~;oKZCfX^J$g0%WrC-=JG_s`S@%Cy1>z1UC{1vi4pU
zSbq{rSOpVE)UW5L4T0cT7nYR3U*)zWRCK9bhe(+T-%}<Mu2PWHOF29gn=XqrW9yAZ
z?A_(%6@Z+1sVC<zo#oS_eZ)v;hj4|ikU3ovwXB|eG61Qp7n23d*G5JMXShyZF)gf*
zt~khoR{RHPWQg*w47RCYP=k|C+A?$NF2kgoe5#O+7#}QSQNrEu+FL;HJQheoO_Rx&
zaq(P<xCgs^Sl43~_?!VOLJQyX@+;NkV@I5!3X&58jGvE{&EV(40FDP<;QeMheRNw<
z(a+Jq2Audkrs^0jYJW_fV;R7(`C)mF%i5!p5VcNYmgvg8Q5MOZ3ezg!UjK<&t>3Kr
zvb5mRK(@O+-j2-1SU<%{23uO208U8rV7W*6(6~H2*yCZ8D(>U)5S(fK5y|_aFR4E@
zd2(Dtjn49izFhFJgK-OnCV;Z!w{@kQeGjI`uzcmE@4I|*wTCK!rJv$@6&Esf)`GKU
zIgJHTf<5<hC3Z1+Ho+aY_o-wPq?Bi}oQ9dL?oXD!mx?voNw1}mO(${+=iJh{U1=-`
zL*lvXdsTC^q7%uE0Dsf@)M6#rIqVs6nO(ZyvG3oqXIOqc#*MgIq!hYD_X<N6e-Z~N
z29nCItJHn%lGth<SQ9QwbbjMyNa{7*wG+sBqrFLB*|^ywf#~R{x?ky)=pY<<(^khZ
zD74NYG3?WBJN1u27VP#vWUX`$QjIVy0&p2BX=5-VFyr~uECC196Sb?D#py?lGEf#A
z^@^u^s(0zYZxibZ$ZN=ygGlfdyyayewazkO;HCSbA=kqIIF|uC7BwvUT(&Q_GK|+^
zAl)P2x=*DCSu6v2zf57832uKwD0P=;p{8AD+$lu4*{&lCYrYfPBn`?zHfCDYbrXzt
z1DQO{Q9f<(Jy4^)3dPpF2vwJB5nNiG5jv=}npJtpXKQ-8P?Ai9dkLxuAv=OowjAR)
zOTAExp~&w`9q>iXY>+n->_L;BugtC@kCR8fFs$#umDmcVMg{dvOkCf?7YXYqr+_*h
z?4VUf_@v5<CdISu>idgdXxLhYNwq)rH|~S?KkBORGNBLJ9SF@JL4s|3F!JGN=`pQ>
zRok8fp<SOt&O6FRPAM{H#j}~Ux>_S>8f%GwOoJ5HSJ0odgu%K|xCjPeJ<k%&B`Lu$
z7I^?*sfP}IpqDx|o=is4s&$H7xU38*%1a-)Q;Y>&d1YJQQ(`E*dhirL&_Sa0n1t(+
z)N~7}y7cbS-{!n1vNHBT4C;g6kCt~FK4M9PO0CQi^90y=7iD=m5LnS@-6A|V*#^Ls
zx!36}k~NPLXO8otTnBQ<<hgXPH*@)M&hoVCuYrmq+fQ<0RzzmZ{n!WYZ6-baJDC3>
zWNN+bEaTYzRH5qj^fu=1;qFGtzmqsJ5paIfQHBfBnMo{5K<}vFz6=t!o4@Xolqia_
zmDGvEOgv^)IQ@J{w^_UEr|59<KRL9k!A5UU$?fN2bKS54D`oNK^Qm+YD&dVlAKRsx
zgclmm6`G<s^vuUrKL;!gM6|$|JhKW1NAq&lnO%jh;0`cxWb!(4e25`&m}SM9=cIVL
zek7q<tba)lDA4{$AqVl|9Jc?a;$7<Br%g!vnY94@Ja4<9y**aaa5rfO&h;3E3SvCi
z5^xBp(Cu^(CIu;jKE|N`9O1VZTrM=(jH|=&^nAy%Drg7L%91yj)`8nn6rmZeHfsJ<
z9`NFR)RmDqqJi%ltOUxYATkdTgJpd@mKzN!!5bE8*@j9A_qOC!Mui-gAjjrle57_R
z`!ELD9Vr2pY^bhE7WY&*br1GJ+>_uFLm@?};lC78KcI}VLYnXN@qsT4Bc!lBE?d7i
zn+)3M$&z@^O^Jp9$<dJ;<Fm#g<5W}{6c6@8TOzwW4|zAWZ@&ihVZ>kzb6iI@gEGVW
zzXE{505=2WPEKlRlp)~sPbO@ps$7AXm8G3R9>7`PM%1!t8>pNPbX7UV$eLHmCe*<z
zU8R*cW2WI6oAS;xN)t2{c#Wk=2j}_@jxo3q)2IX<Spmarw;C^2!Zk-XT|?683$x{G
z*NBqksIbb4sp*=h;zrA&@^mjUcsfWg_#GWdN{5HOd<MPWgTaORdRLiWTtvK(okw}!
z*}Z1kw3g0^xvQwbX2E6Y7RJq-;~iF^Zj&P+>OY%6^HACf$&N|*I@r%^D}z;ruMH>U
z{^f#B=eA2u<u&e+t){fbX1&L8-q@lqG`Id*NpB7i$_-CNC(K-u9tIf=_`rf21tAf`
zI=zzR+N6JDMI88}e+|=?mM`h0P*8JHc6Y+kY<sU=;Y}(%=_=88Q?!>*G=dL?qT+|p
zrogRszP@i8dAmK{uO#TkhBs!dp}F886k<(WYkk7=MOVB}W4@OGy|nlU1<31MV0clZ
zRZYBrdjrE)f+-b&l~oOM?G@_2b@eqb7m1*2>c$4)to=#H;r3cFxrs+RKctX#D{puU
zk7+`vrWGslIIUZx%ZQpPqmb7YCJtnj`)^Wy8P~sB%<mWO`KWrF<lrh+P1UTnu~{Qz
z5+2r2WnKUA4xQIqdMfX35s!o3CvTFuy~Cg34GQ$4lVw8DTloRl%e;Y(41XH;UIP~s
zPFOSTNjzU;g!7zHQJf`i-|NA<Uuu8LY7&t;1<UtC0)n-r*TaHT8ZYbjK7aP>`H_On
z4xbI3^1MYfA=2GKPUDF85l*>kUGK&l4XBG{h&hjzFy;FLu;o^AdIVk-xTEWICJs|u
zk+2v$ON3*d`^WD}1Nmp9)&WSB&HIkVhWD=#Xwi_P?I>PL5NnT`6vFBgMte<2X0zU2
z-3p#vU*I<sGx!O4B8RZ%Qy(<81IvBn!Xj7d)4q$Ymtu+vavqn3HM!syY`+Bo-9`%G
zco>Zy;BpE<<PpX_1D}2=k1)cPhT`40WACC=7wK)QxgX{oXTIv;_uR*8m};s3an`Ti
z3IkEYx*z508_^h(2<d^713N6}<8&qMZ^od~J=Y_eo|Z=mBK-|lWyHGiQO-NTv$%ug
zc?$!^IjEN#FNo@r^)UPBHJCgdHg*9eNq@xSwkUliJdh5k?JB_^=6Ag*@X4z-(MXu#
z@xg2crVfN1xjh5rbk1@;%9ii?^zujZ&WdgsW@wV;V}2A_*<>(rzhceC*^yq|ae5G_
zkT)_^Ry`7yWb6wioh2LTYR>V0;o4SRh1=Jw=1Kq9T`}V4I4%Y24z4ALJM>)vfCXh#
zg`{4TTT`|szDUe4_ihcP#JrGIQE{Jz{u+&;d+0LbUX^3!m^DsZ!QJ#ZyH8;h{wNg*
zZa82bK>LB-y|4jw8e(>6y8&UI<gOC<6ush&KwLPdp~%$xV?>3MUe-U*PrliZ7p>PA
zPI1@>!Gd$07rMv2XvhUfRA_Ih%v@`hSmx75pOwlKv$rjrgd#*bRm2;XAI>?Tu4NL-
zjVE$5(gCBM9dH3V31t?WFlpMUQKG+oup$Zl{8f-}fI3Nl*E^n5#ACfUQ|S-eB+>w+
z?|8A&Y_;@qApF0s&Y>|E0BP2-d1BkiiEZ1qZQHg_Y}>YN+jjECx-*Nr_~sAvwyIJ6
z3`I%cP?V`N$gcoNDLyh(4Q$W%4d`SCLwXpPKxb25)9Hj=xNaKqi!&(V%l?G-!Oe%)
zVhdAb^iaQ<jt>-d;o2d6J(9RO=KiCe=27Z{^fS-U$lYn_wXQ0$YfelWrNOsb3qrap
zUi|JrKpL+;#{1(G<&UnVV3hiff^79S*REmG)g!dq%}Isb-|K1)`ouWsIvaxfyh)9E
z31vD@VrdtlxJe~CV@@U0C-99~ZTLi<myR4GCogl808%gHi+;bc$TE)K(eRDU?0+Vu
z)!)hcm1yA^<CfeUzSwHG9;&w2+s@>?WOb0o!-I@gH~Bt;?7jNBK&UrtMKYmEBaSs=
zgUEqt&_a|6rN^Xgphfk%;dsJ@o@Ai2(mgaO*g+UF_{D<-2)3T4$vP+6XAouq{dqZ&
z)Qu^PFsd|;gCP-^OMWeHvYy@kRMYS-lu>m&nuW>S`Bbq46mC{mxqiKURYVkYjGyy~
zxoRST-Q=L<xyK>nGMMGG>N3$*qjw346f$i}NU6N`bWTdB&ahmqEP0&S==m&fpIoNa
z#!MB_EbE)s)gxmSZCqIJQXvjY2-$-*g69)x1ODSnU4>lNtJ0F+W<-n&6z0c5rIaGX
zF#M@bhZV37BJ791j<smf{0RfQ<)qf3nw$y*Y4q0aiL_sof3CDpm)4P#H+D8H4TS#A
zhjMK3I%`Oit$y!_!@>_eC5}@BxE|u;<EW^SLGzcnXx9xg3X`d}OzU9|vLADgAEj{<
zrxYo}Iw)90Z-EW#uZ~vDvCDOrEdPBx7<RTS5DiK<+>nQV;$~EZ><$wGM1hui9`Pbn
zP~?v)))^csny-5X0_Jhb@qDohzHdc!`HA&j+NkNV2e!lCJE<KQ7;(YgPQz#;nqG77
z_X$QAj70CU2{v-9X#c?pD~reG8}EaF)p=_6(fu<AUq|eNsmY&8DdE&0S)i^pQ{PY7
zQtJ&uiTl5ASI%<stDI}We0_nzJki23GOYmu%9hc1=N9&CKHzOTU~-9$VQpu4X=E5a
zspCwL`Rm+xBsUZ@8mVx*)XZLSm(c9kKb-0AX8Y1~e1Fb5l9Q<pl|4}-Y`n~&sf{l|
z2Mpd7qWbvz)GZUE3u5{;I#NUfh@;jz*F)M@D0|1XTdZcZRR0?0AXN9K>}?UoXM$$;
zCcB3NjY4ur+P7vC#Pu>GP>S-qMS1Ug#g0Th0N&J;Ym(X8%6cAL;5m(7Z8eKwSo*Bj
zpQI=~>ReJOVT^A+tFyG9J?QU+rr7PZuyVlpGk{*f%uTeYp;(YE1!ox4l5;TQNvn#N
z+_qQplM?tjV@>{0np)$Uvpyat@CCJaw&&g$=sL~!OjA4UB4nJJy=yTB<~(e2h+KHy
zyb3>#X(CCGsW#E|=lDG7tp6LRNnD@*7w8o<;8_@7p*P{xv!VUgL0hudO*Pl8=&cAn
zEfh6|?KVHh%pATKTaqyfaIgD-+1RwH8(eKXbM^(yPrg?2;sV166s~%;+)a8{m~;J&
zzt28uY#qi}kNfGnbOZw)QW}=oB^Xfr6Y05hP*%vvy!5PHp%7Gn5C_&P^q<ZNvQ>kR
zqKvEO7J(F&r@|v^L)qvUSBt&0p|X-3+b>t7b;;jHa6)O!CiCBV0xkKBQqUJWE|ER2
z(q_spdp~9g5kYOT4xr@Gr|dIi<mOyql1wTsXQaV9>CIgJw(-;K+T$$mNGhr-y8Q4i
za7qN%iP3hZp2^UOO$v^abP*0z`GQe^0_vaN*h5@G%QPM|8dPH~E%4Vv4|ty2orZ>g
z4fj)~Fm1u@y28dA?-J6srzMja9jarj1`R2EH;S6m8jyqDOF+wXc4iS)MgRu+ls@>i
zev6|$gbsmq#yL$)$<z+iOx(gKHsshBKdeV@Q|kce3-i-gMCKS9ZU>I#j1<z4uU^%Q
zHF;E_Exqc13+|jn^dru@ae+22tV6%xGjqr9b!}oeH0cmB&t0bS&zoVFoP3;W%15>k
zRRgGM2vzn}ecxiBP~Qq%$>A1i#o#ghKIJZ%E4^saPPp8V#gY1n2c?*38x=t{{UP#u
zQni;a3@#H8Ml848s>k@ktwp#kvcPhGZ#~+GZG$v7g#R)=gK4Eoj7NnE-%Il*dj{>F
zixV-`gE-h&9@P{?zK)`EZAKbq{9^eEa%i5ri|fJW3Nre3Gr--Yy9&Bkvs&~{T1&*7
z@cE_M8jWtNlAYhuZs3M~tATv(>BTNXq`d3fAD%G+BXZUCrF!FOJW^@}9Gsqk=U89%
zukLcUDq01Dv@uHraZjh;%e*Cpmt4z?TVP4F3M2<%6smzg=`w%ki;!NiQG$VqGG15*
zmjm!kR~IX#+&}3Yph476<K@?+t-R&zp=4%{$^xBBgpOoG^nV@^sr5xO_wM5-)+jPG
zUD<lmxJK(smxH`*+dMZF44b-!tQtW<IUeX)ARXPgh$khf`)(x-TocSI<ty#bDVfra
zADG3Cb-?hMqqG0`N#nz6<yQ+c6}kp6Iw#R1XTz65Z$JA1UXHQi_H5DxNlZ|&azBF_
z77h>3y`A}*&6bagDQY`TVf(6`k_!z=g_C5BT&&WzDcmbI3-Wu^RIvUpjUq<vA*2({
zHT=vQo<h5yoME(iNn@J^pGvRfojef`(=bI%A^<In0^J)@1hBnm{e$V_d-gv14t5r}
zFs2|iY=)Bjt7|Dx^r@2J2UpSmldz*AQmd%4Rq;0LLSG>~P%-XksM*RZ+Ehn>IMWi)
z6^~=G3;$J2Nt43+mWDgcdWUmzX{8J<Jfahp+|JrYpDOQhrfh#4@B*k}s9XhqruKU5
zeymS!uDJiArAJCtV3OJpngTtKdvwOEj7qqupZfsLIwr`lbg;)ng@QXw-+P%HVH%B?
zBt0Td(K-(^to@?8pl>J!HWFm4lYp5`D%r1>zB5G+2}xGUlh4`ayu&?U?Aad6B447n
zmj;dPh;WRXek8JOgwT)rsZ+{OmLrutmd62cC8$rzFMGMG6$)jucBm>C(yTQX%qpZ^
zPI1zu+*r>XmjCRL9|#?=-*5<Bg-!;r#)#yKcDl)H$6C8Aw#c$h1P%SA9ob`#;OATE
zVirW%nSP7%T~m(hBFUb>T4^p!(&%B?`aU(x;<|so5p6xBqi<aU^xC=~C}mdWLuJ>O
zlw{rJ#sW3TnN=STr_%S))~wE*<&h(=z%}l+JzBohUO^mRd44P;CEqHj_l5dX?6Ur`
zg@B+<ePH<%VK2y?0zEb9-HkuXL&^4Bv_0^IivG;h2%`Fn7o&<45E5{j@)Sp6ds5R-
zc>>~~toasIZ)TnrIYEZk195%9Nxw>aZWJ+1qnMIBaSgS5xE>yF9D%Stu*BO*ia#$v
z<w&ss&lEWup&4JBL&%iftqJ)aVrE%a5~-LP^nz!%f??FoJj<JdafF1Q-)otNU_s7}
z(U<xbiVtAi<H=|poJ4{8CopoW5A`S|nq>TD8D{NMv;b1in@IWm^OY}FJ?+H3sNu{T
z{6Oq>BVAt$v!giCBodF|6~2R;2YxZIbN)mrX;TZA?*T1S|6^l>ijnEoE6idxI$EZ8
zffMv(lE9-DoA{VFnK)(zAi2Fzf>rt>d=D4HFv32iZYXJ68-Kg#vhg`eFY+ku@cwju
zmJ>tc73BhZhCCkGwp|yntemqRhiT}p7?e-VeE=bO+i_ex`EBeSw0&h-B6+EZ{M;c3
zM1lV|EC2btR0A$UXIxOGQ)1p~JL4jNMA4324(qC6TQ+ZFaobfi{$-5}u=nw(uH~rW
z-#nW_LILQ?Ncu$o2jZRW$=enlugq(JSE-_VFJ`cZK~a4r8d7Jb(s(Db#y?T>HkEC9
zm7B{$2{9h}fo3I5b#*3IZRJo(#$$zdyEOoPi*5a*BV54#Rt;I18Aah4&`gNTcoMWq
z-3ye*!!y=>3|^sjxB`Qnw1*2Cj2&ty!QarJ+Q1RW4*{-!*KV`}Y4gQc-$obk7W45U
zme2TE>UF+7#ilcCp1bL={6(b*TmMyQevj8rp6~019D@Tk_&?_)o%Mg<s&N|r3J_1d
zR%dLZblTK`A|<Q}R`r`c)Ld-Kq({^sIPwb*uZK?o+<uxPa&pMG>3<k$u8G8FCsAr=
zTa7e;$|}!~IPjPHtQh~0T0rb~yl@Q(`Y4%UE~UUKky?9AKQ^Jt?}Sr!O8MK*t76{)
zpP-+vBCT7sj5gN`#17J1&7LNje7w6#?ddWQidfSx`N~D95T*Wlrqo#6ie4@nl_yoa
zd~vmgj|qVbUaw+wKZUk7MKqoNn0rjt!WY)5uK!AY!X6141ebmy4lN=#7Fg}c2Nez;
ze@~Hv=~&#nR&j4QjuDkM*+I6bM3ZtB3b>I(rp9L0>ld%rEaH1nQl7*yF#M6Qo;vKf
zvI)7rP3%JF`Nlt{6m0CX<QS9un<~hbi<JrBw*`o$$rv6E)AO>+hZ>&~lEdqo0wGgc
zVX>|{HaUS(!!S{K!g??k-jX{sUJmoXk)mVOXj8$wkX@z@M61hKNg54mlL4t8zL}kQ
zccj1qM!0ySl?qI#h@_g-nLeuT_E;ZNCS9jhW-Bj|Kve3^I7@*%NTqb{3?=@kS><zu
z+li0$;Q0{i_8XZQa{xsSRXYq_BoR<YS(#Ac4j>HwsOod$51N_{OmjX1EKvtn27rf<
zLtE#VAiK{RON2Lo^(L+AT^gEgq`q@|n$yXIXhJK4iAGCiVsIY7q#8>#!Xopa>`q7;
z9Xk^}v!i>du1<GK@ga`}guVr&scCx~0tvW)BH9mw-Jb{4FMVxIv0F@&K0#B@A^<+=
zWLVu?&=qT`_Y6MnHiVe_GG}h>@OXIsNS!&sSb`?gxlg?(TJ2RW*zw3b%yHfl+MAp%
zZOd)rMzUCns-7dOQ2qob+yfvTcZlz`+j~cQHKH)azkPfHXdwY=f7JdBIgC!3p+UjR
zIbv1x#BBK(ByG+5WZwu0Q{dJ7VNR&0ZU?e_-I~6M1b#=f`K}i|#X~rRWrIi%_nKmw
zw3C;P5}zugKVL)VdfAx$>rKOmK()GpjxYS2Z%1kEwzX4p5Hso46&JFG$q^h7PjV+U
zAS2XM3sXNx67Dkq{0Ug^R?@-irxYM)d6h>z#-0wzj>u->g79yGAKtk9zad$DBxI%b
zHnj(kt}Hdh^5)vx4N{avWIMxqEuh9{rZPMTinxHKuag4i7Rf?zITW@-&`Vn@p3QAB
z1=&TMOz*sEmcL6Y-q3gPPgvsV;zx)YlRvDINP>zKQ%rtt2`C&qIEIN8TO1aDl)g&A
z@2ryDYRKrALDj+S*+n(zC$fMFvSi?R9KeE<n$EM2k(MpXLS*qj)3<EMQ!tJwP~@F(
z+E9}(p=f5fId@dJ*?VFhJcl8&fyGxQ<6~S)HF=cqD2`0t(=DUHp4(~!^`%*=Ccmob
zr<s4kQ6-*^HWAJ~MNnNCPYP94@APa*FgYDJR$pw-YylasD&3e+zS8O89h`SIrZD;G
zHx(9%G2M>_z_3XY(0BSADH6tt*0>fi2dC#x!t8Hv4+2e4!%JtVHklBR!h>JngQjmt
z=9pZ4><k9CN4z=Y@71bjZBTAw3f%<5>GAftEp2;-6yx!!XOlkH-&uNXxvCJoo8qJY
z&1I3|+3irEheFqh0ZeA3d+N;S%yPnWu!e8mY1&rnM>_ypCbPFnbD+tm(n6dA6%XZm
z1Lc;{$jfqE7iDGFK9VV7tqw)ZE<nf%r;I*8;ZbyUK}rlNQPn2W4999XNPCYk1^=<v
zg_W{mv%z4&>%9%n%u&C&XNAM|I%Is~=f$D0qBehMzOsVD`P<p50W{Rf|D~R}W_%8O
z(DQMJa?_1zPy8WipNu37lU1deyvMUW_zHE6t*wx6dS)A4{#_(%C)G2e6L*xKcte#o
zA<KcxXXOGKwKOdi54xzboU>cbn|BAjY(}qJ{DVJ4)3X(j+prVik~6b8SkF~~HcS3>
zD40te5@?|2u9;4YSm;86njZ$7DTaP9MF-^3GZr;F@bo*M78zYb<dW=Ubwoh9J@du>
z^A0E_!c>&clPI7)hbL+_Dbo4>5NbV5xNsX|d&Y|WwuEl0W@%Db1jRtl^8VSjp?KMN
zHjFltBofAr^wg3o(pNt;6^ICCQMF3ODfFD-***!*m_No_hNWfmA;dSrSU3KSvKo+o
z;2$czOlzd|%f+3z4OLlVT;ztEuu}iIPo;2%cGy&G!GCPt4<-?|HN`KMen8yS6b?C>
zq1g3EHj7U1>|h~0Z3)~hWQ8gQXJ7+286qTqx;whsK`n^P2R-_jG6;E@XYD*F6Q)j&
zd)Nnr4SL(ioDf#=bKB98S^M1J?_HeGXA3sk<@QYD-8g%jDi`SMx^B7F@XJxyGtO4K
zzE-F)l|Q<y!T}>8qTFaUp)@!_rUwFzl};BtAOyV)9DJig5mKd~Pk?^OGhhvRl?M!`
zTED;AC(jWM8j&b+-0|JcL3q?6iWP$z?&KV=Mn%s!d>rjtTd<Wz^hhnlrWnz!><RLo
z6$*Rst&8NVHkD4ma5?Gi`_lO5%#>2H=lPVmCybtz-v1n;Z6V?4(Vi|<K3YZqt_c%<
zD@r2Zcji0G{%LipBKo#ys1jsBHj88M3{`O~q@iCX3_8rhCxINKy)S1}B|Sv6*HFvW
z65+m^gk*R%^_^&Q@(<sN7dJ%ebepveMy9bva3Qc$+At~JL4FL{AiqzjLFdlB!I&yQ
zh9W@6bn}4^2}YUulN!ISg2SD<-P`TYq+t|eGlIoYWUJeqUGt@lCFzs)!rjzvkcD2@
zxJm$3K&ro0pR9XwwCO!mp81t8RPSR3%QwCZf;=!U7QXitvOCVg=wnWbH1gn)7d6Q#
zl_zWYYXN4uv;dXz!AmEhHS-c?ynxjD@ZVf+T5Y13vX805#ts_(8uRD}h^EZ4R?n@f
z{;M;_0*$qAU=V-kDK7Y--x|FlvZc-ymv<5LUitS<Wsbg;Fsc6HjNxcHuzTk7lu;sZ
z=Xmn=&}?Jw?~o}fqGiKEW1oOK=wTogU<6HhS{ZIpAM7;g7zV~u;2*T8FmYaiktTKF
zlkJ>$NLq=kPlJcY&5y@L<EynWlw{Z5#(aT)j$tTn%FPf&x{EyyjN^U@=0JTK+aAH9
z9g^DmEny;*m#iUEcI!(()IH<+*8+Jm!x*X@=<Lm2D-I-#Ngr|IaPDiDLp6vUlMtQp
z2y5yG$8xChsFvvQvV<-sz!;}}Yv5ogd&kY%Gh*%Rn-j<%1Oqw*EtQ}wi5SGInS{aq
zBC{l!Wu4TR&N4Bf!0j*}b>NpX3?9Rxd87hLo+Pa8Bx1ZC_Idbm3|-<4&>;^5M`(iM
zHp{5y2!*=rI2T_?Ym34!g4Wu9dApmnoL}fQ<aSa}8iYsX?RUzOUyve%`ib+E<aRs0
z@<NIyG2U;tR|G#+Yej7-nm~jhk4JfOo3fOB?zN=-nd}T%EA}$M-iKEhO0_<niGVfX
ztR$xyj?S0r@uqOZeKvWAV@&Ba;(mo2SWSLx@>0;(=(cUzb&2k`Zbji~uf<TdI_Bi@
zeb#*x#r3xf9uFH#{g%ZWKGjnsw!WDeE2mRV;J<SZWfpz<i?0mISKDdR_zAAxn3r5v
zjR~g(tdX5qvQ5~@)fsD*a7K_2P}Um~siN^t$(R3a_m?gt>o)czR(e--rbW7*pFgOE
zOjhmH`RD&4`nltCSe8nl@xp+5+FJJ{q&_JBBWbNskDJiA>nLBA+2V0+)!MC#Fr{gg
zH(px%1#YfGc-G>3fNK%i#_ZOkTnM?l9>B#xRwECps&*2nF>kJ+%g{j|Ppoa_CDi+m
z5lRJw3+xy}#WI+~h7h7Gvtxp&UDGHbA{?l2S1cVCNUKDT!ctout2XehD&*Z%u^1MN
zYW01o>}xDt5d6&#;DsP9D|)k5ow_Hqf8jQKBVaXI=lK50vcU4=jM3e*4Vw!0AJ6aM
zbrXCk%HTOGAYB8vM-DfE9O>4Xo_AY$(P?5JIQB~yg;pYpH;W0u@)IS2O$uibq5b+v
zb=ApQ6fLP~<<G{t+_WuUdug72rCmR<Pn!Ri4hOs6dA3L%J8+W-_JM!0fVT4-2T_JI
za3q{ZO_=+z-KqJe>=SWyUzG%_<YkE!t@d2wH#xoAZ<r_Mi6<C8Ng#5KsmpA+B;Q7(
zqMy`|3}&|Jo#asaR(Pwk#eVUNvYdx6hmst`w}OYzTRqkOb&wn&J#9Fdpe7D~1vQ&L
zMU&Zq(^|QjS1)c|;ngUq{Uw>Mzy@`+IPms|P3%7IBL+4m`)N#>mD#UdP7YpYS2BE)
zJg^pP|LV<z8*!Hl>^*-T&}SjR!%NOi2*3s_sSuG1FEV;k$Kp6}LIcte$#o9@XM+R_
zVJv*COIc3zDgQz<6mAslkrR&q!K|ujWghR?TGP(x8)+dLpOn_j_Y?__-&+L64<|Q%
zDph`{dZ@28A7_>KuXIzg73QiQU9!V5{Fb*vxe@;uXCT{a-NXw6^C8EJg-->DerK%f
z098m)z}4xyg}bVV)<kcln!G&(>RS`^ESSa?MYrJ}U+yfA^G?6vEb|V+iN6IfgO`kB
z-UApIOvn#^$)Q5-*T3+dev!`~Vf`!$^|tG`NGa7stXZVAoxU{@UGyN2t;0!Q{`w<r
z5a%LD<3fC)56J40)Y6MrLP2&%QysBgI)EPSkWI3wW}?)dko`llo6guw*2Eec#aA<C
zOOZtD72raUxDIKNozZoco-W)T3QiF9RW7UBX*2A`MM5;`MC0i^lPeZwav8nzi*^85
zv+edr0fEpU5BWjj+US~S>VoKsCo&x)gOY}%$}<r>x0unjKP1U;K4q5hXfv+1B|wKx
z056co>vQD55fpn`Qnf)H<rGPX?uMJ#1xv}P=G@3q)OaxjTRH*zpsV?)HD`~qMb&tX
z$P6Mj^)yDMCkR+JjD&|yyegrC7R5#DIcf6G=FS>fmB;x=)6m*vE}~VrR$-(}bM|e-
zKJGf)l{wG=^y@jbt_SbfCH~Bx6m-CFMm6b*BwW58kM+SU-QXF|hT|DjC<V2$O1OEC
zOUdQHl_EI~_41(bnu+3eHv{M`)aUbg;-5*0mkvsh{=~C4cGZ~m&9P55hx<WAF@M%j
zfuGwC($hJ?+A`#hDziCKM3>L1PQXhqWgGs$^lo9|Asm|C4})7Q)1nTt9U#DZ%$lSH
zijammiGyXb<VJqbrceV4hZ6wuA`i1gzFRSjOrhN_1d);bo?v$w|1k*!HPfJ6e~LA5
z;8~x{><D2ymPezaT{ROw+8Q)zvhWRj3z^ebk(+?I3qQ}@PdNT+9U6dHjfq57m+TIt
zWo<v#gSwJ%1n;;h_hSpA$WF=6WxP&!57d(gMJ~y{`VI;gIPd^gVT}dw4Xks}Hmpj4
z`Po(^h{ndA#$Xqs`l3#glR6ePW`eiebdeIGe{%CKj8jr%t-Q0(2^llYb9v9RAJm6U
zSMA$P<u5HI7=T9kpWuttDms<r_e~w`rvbQ<S2()sd!&AKNWK4QT%)s&$td&T>_ZOe
zJ-Sk+VddLYd;6_E;z|zOotAeCctCf!hsvt8Gzi0}NANsVitWW~gu%o6ZjDUf+aiZv
zxnG8jLHdY7`t?iuRI=%{&WAFjaO*AB)}bC+vsdbSib&N5eK9(Oc}3_eyIwb!92U>Z
z68GbYw@BCc`BQ^Ow&*zL&nTuZuvIc$GIrMJ6YS2(W=_P7PlFK~{~pd=NGX_3vPrGu
zji_k7|AmILz2LEx7)CMQ;^N*^&zwYE3iv#8A3slFO7YVs1`EQ!YBmX)1%NYcBdd)R
z6Udx)z4U>u0|NqWlquLGBX_|bc%RV2W#oNk;CiK<M5Z<^-d!&{*{@r=aq;(clS+-8
znGzHtGs$Httfc%ubb1EShbI59$Yx6Q8YDzJn@I%Wrf_N}V8mOdd)(aeSlqDl6Hu4Y
zXL-~4dLDwG`_TM$Y>Vi>RIbIp;Bpu1(C=8~YhpSwqG;x}or{hb)i-0VYMQ**ZvR5v
z6R8oD;2B(2;~n3H%OTm|sHH$_nyB0O2MaQmvT4_uT|+)YCyJNor=T?yaVeB%H!KQo
zQL|mY(U`OReH}`)f}d<ShtY(;9ZGr>pld>ivGa@4seGQ8b5F87xrLL72vmuN*q9`N
z!+xT*nfCsv-df=2WsS}gMLW3)zVyz!%84tB!s+No%S<GDPUv|y2I32rFz*Qqp4?BV
zxsp1><0RjbiCWi;Jf#sG)&^}sM9{^nlw7M>7@yA(@YHw}v!<Tfn;biZ%=>Nza>VJK
z0R_%r95jSc5_1($_+3^|N}3%a1(6a~6b~=TJI7WtgP<N9R363=kQb^GeMzG35rF*?
z_@+ECoN!8a{a6Y+3~Ej#!%Fg^SFWlAZ~0y|i9Gfpml(Bp7Crw%>!PfeSEw%)98gdb
zWxIvXMNZdqPOe)IlH$A$kUQ`#;@;ipdb1MsDHny1czUCX&iI*Ovdt9+Z5mAMhz-8{
zUny<0&6A>@RLMGNJD3jceIe8LYTZPEv)3=Mj<C5fu&#e1f#w!&5JmD>gpcf#^_|B-
zn=jQTzLX>B!HhW*bX77zn?#NxkRhohCmY(AlV(hthp`OsXEcUK8C8PgB9}cN|1^=@
zYbq@UnAfc&yaklV!dlzMK}hcqzx|z60l>kH4{a<p#E0w7E5h9^AszB1Xlq{RHtl=}
zJ5g$crV*ix?>!O3RL%iBNpk18>?e_d&^J)vf?5MBV(IT~+%SxVXzPPAjcR-J6&Ly<
zRJ0>PDkk#MAY&RG?kRPQX3kv>BIgJ^h8@^OYLZ0HA}<_O<778}oZA+T;CX5cGshA>
zbMN{=J`F3uJW3H<e6f3Z%QZ<BOwag%GJ5w>Uc8mOt;R-xs<ES?riJu3uFMRvO$O8F
zzGfSi!(LM*t8_zZGY53=bB!O@%9SOI6f^#AM!6|tvry4dP9d&EdK&yaP9<%(`n#Lh
zXUlxj)yA38&CX2g7>&W|$#GuhIsEK*B;tzGpUy&TFdEmN;G)TX|7zvY1wMeT`$&c7
z%Et0PJ21isD(LD;h?4~Wm@HI<p3C7{!>V}{MD{!?Q#q*>l;g5tWLkShXJ(|UK1wCO
z5kW`V6O6fx6$+^kFWNdc#BKYn2__|8qBckDslQ!S$L4Z8?ugbO?Knd%1Fr3yapA<R
zd=3#w!!1t!CrqXR`3~<OyH=f^Z8@SvuIfv;=CvNLV{ku^JEr_c?Sk%OUyD!3P1g(b
zkA3#wuwXFsB4zL{ql#`&P=kS}y7U_>Up{#-N_(sehm-h_S_t$%NFmJ#KJq`l%n6K9
z6HhhPmqQaU3vch<C*6Quxi;YR*cr)?x$}~jQHxxv+#!)wL6M3n9))<|by&cfC*c06
zlXsbf^>UPCRozJin~hd#V!sUKX4#a#(Zrfs>U3=@y;+a+@6})nJ=i)DDW39%#y9aA
z5Z)$^W;{Qx2%$UM*J&vE&|Ct6@@oa>D>3Gg=uAr`Q;X5T2+WAp>dCV;enBav3RCXz
zNC~-Tt>Vnj{9Y({srS`*1>lu#&1okg149q@Q_?yE*k&TTl`RmjF&4wz6TnD1^TnWP
z+AeA38_7KGZt*ISs)}b9l<K|W8KBC^yTtYAPc~G*LchV{4V$|%E9AE^I*We~^Tl(X
z0`?5K9U7qajIjZ<u+M)frroodmOnkZ8qaI?95##3zN4CLbhhqWA~OL?-5ZIGPJvEL
z;DXKDEG>sHGM!TkGNQWsY~a8J2OxG3u=))33CR!?ToTli+;_r(Y^XC2%bnVWi({!G
z;g;*u>RMwAlExiuYhyR_)GO~gLQ5Q4^)G;lW(ILqtcK%%_N*S*hmvlo*1fFO-dMI(
z3#l<^MW<SWguuD_ttF98cy7W8qW|J-0+VtX`EBLH&dnC=k{^LI&2Sc$<s>|SuWJWY
zJL;Bdy-J$OKnY`RTeAI*D}KvEV4;m0EPRN2yZr+<<Y_!9UMpizif1r9fKOdrb%j2)
z)pM(;;{v~arojp>3jQ~2t{XHE_c`D4VPC}yvpjn{40~?1&w>C7d!-XMEo(1&J}ir{
zSlG++BDkcWKkG;HSo)qW@q79PA^2E(*p);f3PYu>1r7y<yvWZ*nwyHP-oar><i(UL
zDHSYL!rxqo+tL{8d*C?0P+hW(CJR<pfTzV<*Zr~%sjXy(ZCPR54@+1B)e<4&!_z(r
zf`M)QAjh6XD^=lilYDc<U1&ylgn%00ascG*a}*%Zr~<Sr4G&PeDscF#7^I8X1QPKA
z&{&4R@zIQ#QD~6c`6SI6<=z8Wa=PY~O&o6aYyZcT_<#rnbWot4IBcA-v}u9tRhp<D
zu{|D2tA$1(5z#s}l7u?1Z>KkZ71x}jO0b<7XErqcjmhHh>j=-YJHyvBd|e~Ig<xZ{
zbD*x5p<gchu;F9wn=-GNh|MctM%{kW&Uw;b@DMHpW<hr;)M-R9?MlV1u(PssAYi+b
z`aRXrfE+)R!9#zER<IuHA{8i%&ikIJB7;Y<w=w%OJqK9L^4Uqc5_*JcU@Vdzy4F5x
zx^Ple`DuAO0-~m}MW`^vh|p7T{_^b@%;(@>NqY#Ts?`4GXhawAs)}L%<U+_6y2}V$
ziNq>~!<wwLz^f3P!>3GJCV^08;|*PS#V8?(WVZhoHrvfM(IE!j4pjDGd>X6%uzS!B
zmbL>-uH5v}e((}W%n<u~W7rflJK^y{9wdvG%qB6I`caR+nEw}_51o>K`BQe{>ze@$
zupcA)C1~SR4xk)SSi{Q?uZQl6=p6{VVSqMkUr;QvInH}BKX;las6>4syeT&5Hu!-I
zBvHjJ3y-Cuj5P&o7b9PQWnEv!=8?pR>dP8+!tH!czVJc)xm*vux6JJ(G#~f3#%<kA
z;&g0v@Y32+MMbEa`O!{&*M4Rbx!S)<dIW0H{H^x#n@!xueDGA4BVAJoMKzf-ZoicH
zatfjCVuE9{rJ22KBQG7`ih-46*Ajq@0wYJ(&15Sk#Bxod^$1E-D9WrHLpej<R2MPj
zo@-_?=c%*ccbBo>NcX&Stt|>K=>yS7*zmQfHV@H6GaPKluZE#`-+Y?(q02!KB$M0O
z8#g8XghDD$`Sh!tn<V}cZlJF;MT@cNeQv>~)V_?5D-e$PEr8wbpyxKatZ2VFo*0ke
z+sGXsz3QsFY~@vS>etg4KLEslZN?75p|&U<B+(QS|B2d>q97&K*CcB)60qPr5QI;d
z_YpY6-#^yL7%h!QO%zQU6+|fCOAAiIy47W#+=PJyI!R7CDtyU*P+736TIg`f^aA)3
zE3t2l{aKvim^#lHx|L7r6}h?*NQ<GjNok3LLT?v8CT`-XSstkn(dl}~@$j2E4d$hb
zo(I>7t88UlH{&z=3JQ-EH5MdG)7UIH{_3t)Z){aZ5s)E$y*{>owxIIrM3D~`l{y+}
z>Bfmp>Lbv&_;2dFqqNwaUdbFoN`#{#PC);UcH^DRE}X4zEjpzonHHn6=vd>z9(z!a
z|0Fk`+r6B-gC2F{HD}-!`+Y~d)lItml&8b!mC7}jw^T@&R1MQYJ|@0>-FV?h?ckr;
zlUFv4eD<j$`sao$&+c3f|LzbgWDeBp^AA>8OzQxH;CSarRov`YdnX(>!sft#np|O>
z-=x}dX^iIWN~459zLP_dqA$S+={nWDtU)SlX2fnp$9!c#IPsvJq293)SG{zteN!y_
zS%{io=MpyMz66xi!GCWnc~*o2>Zr6D9Qe2||1x&%wRhv;p0TXI+ey+mqheg}0qvN|
z@QYssKY}r6=bJz!8i1>d5)4ovR=F$MBRa^l3Q+|~&^AD+?g^Us@N!r}>fWW?ORoIY
z4P3OvgXj9q`**3Dm=Q4+2<;vxH|J(%ys4A>Sszj`$Z>lY#w~7G@>XSILSH-tQuJMM
z??T##fyjQ0e(T0XD47E`uP%0w1W{D%|GdRqNyt>vBX!vvo}%#yJECndk%D;{o3u4;
z$f#qB4^wJ$#l*r$%zPVRk<;IEm0oqLju&}Wn2I6NhqU=1)9{+bO^}ESf3l>G;DuIY
zCzF~QL?3o?wLLb~5p$W32nFWcqgDWm8mcW{jhx#%)&Mam2ruF%!lIvo;zL?Gf{89z
zrh4ghX{DLWdt4`;>8w!Y3nwQ8nK?MKA5+-T04oT|u+zR}B<RL+!Ovv>wx}a-h5l&g
ze>!u3(>r1hhd8-)ym_>p2B!K*iH{0r|8lhm{sge`Y$xSO8CZru)M89&ndF+-1~ZV6
z8T*fRjcta$AhJ~o`2CD{ap0==Bl?8lLfrtODf}E@K1-n&lnYjBn5RHVB@0d{uaMnu
ztC^vSsWxr@5uxoZz~%gi9SW^<wknog1c+d7)5U7P?$PfwmnEB8);d!z{O+{2k>@<J
zgHx5AE*QBl7c-8;)Pw%?=Fy`#^uI?o;S%UJyf0$q&uX35AQ1$Abe5e(F^oWquI0hd
zpf{D~yH{}U3&k|BVRHL~bED-!K*nIL7$Pt^9nUM?hY9jk>HKZ4K0%gv1dfSHSHvH;
zQ_%o)VmFFf1zZEcnqmokc^Sug)y>1gK*LoRvM@w>=jpVl&-)UO`Z_D3(CW8#U3Zqe
zbVMJr?Q6$>_ddyr&4ICe?S+74fG?IcfL~`Qk1+m<DKl`*4vwVsVT&_lChrO=cg%8j
zq&d_x2f|Ew$kC1(QlT)MM~e1~+-KhEZC%Mc)S-lsHafsk*ee;k9=>Nf>oQXWtpd_w
zEGkrT4I8PcI0bjFxqlTYOF(PxaBdjv<_GoLKrlCM4)KTE10}^|zE<0wvPiRyJHqfk
zx2=QB@Od)wfXHH!h@tx`=khQwFGn=72s~^WTW581dGA~pLYOkk!#7@8dHYt-<M}#1
zuF&VhaoUJKAbAn3+`4X!k_DyI(xlCe3=&a)a=#hDPIYeQ7-Tn~EGj<b52wf5((=zN
zQI0R~gw6C5%``0}m_Y)~u6+J6sVWbohY7M%f{IAhe-~rK$T$+V5Z*}F(YNQ-+sb$%
zb|^>4vU;Ja-)x)%nfyDxej`Vq8=ti@9w(xZ)uaGz6n(g;wh)MYTc1fxjLB`M6E-m*
zI+$$^a;ab~=y6R-RTd~NA@kJ7JS${tLWLPOc&8z!w)#nqB$_<Oo?R&<hO4#7F!cXa
zkThofvGm&V9dogjMwM0MJUb*>&>DqSafnO5_6UVTzWm2qu%@9)mwztRXirT<X&i;1
zFrMNaR=Lf~t^r+1q6G_bxz&-tpB)BH7Ixit#dodXWVtj?AP|kX{Wgrbxln0b$gg$x
zJR*qDxalYW#XBZ#tk>h)X*^S00%{7qC+VzvgzaQh!Dhqkty?98(gd3HUCf;ld=wer
z3ue3+P#I}T+iIOv3f%+-WJvv<DhO_fNbX0Ud(hLcua6?Uo3yDyl-|V1L?viW6~7;V
zoU5x7#fwUf=2a9-sTB3l{*NIsm4nhQU_E_$<PiMQIgxrhOR=fG2v4IDZV|W*Oickb
zy{Kx$b0OYljN|slFf><~mT;oE$ksKtc(?gRx|CrV8I<LNUbu~wf~Ys_U*Zh^w6`w{
z=Omaag51Kyp<wNJ?M_601E^3<J;!atcr1^`{c@zUX4pJl>s&^-v-<r+t$E<*s6e6~
zR}$kfOa^?69IW=V(&zB+L$p@Ptt#SQu%f9;>;F^|8lnTmJ<zAjFow>5T(5`ceP-!_
zDQyNwFQmchw8tEK%v7J%L{z6Lf7M=Y9A8-U_}p0(1n9{*h1A=93yx>*1;Szuwo)=p
zmBVb(f57gGzE|80&1cJx3rL0x=s%lKrECbGNN^iTCu66fet5ycd5Q@u+ZEUJIfkj~
zIt2wifqbwzPD4eQF7j4TkvP?Egx73<S0Lh2ZnyYu(@n^8e~~0>NT!iGPJ71|&92cp
z+1dIydsBDpP2#RBWKp1)Up8CVoez!8+^4<mr?#cz<qmB(zIh}Ux{vm|)yR||3uhlr
zoPTw#fU<hb!do~f7tH4;v+eT`=JR*0%iq}G4lR})4eoQVOXOJ<%6RsK@*AKKuK&w`
zFZ&2tE>!Vi(5mr*93L+phH)AO=pYC)@H1}(Qqzg4K*SO~X)HCf|1OMfT>(2wTYKBZ
zTBK6Q$6Rw=zh-^!R&U4y#5wy_ySqJYyv$bG>$u5mj<-B8VO>gB`lL%|Vc@FKLUUwd
z3==lCDU4-;%t#GwEnV&#rxjRk-BUk@2cPD*Y^vcIIF^0!GP=qBR=sKBJ&IIPXxp~F
z;9te-gBRG13k0L+!bv7LvZQoLSRuT(e3CXU^PQtbh$L@obwJ`1&@gRAJE{nn&);55
zD@ZuYG&zY5XlWpLiP;I18H>z}=;vrQAk)G*tu7&>YH>GX&~Q6s`ms5xV#N{7TIRR$
zVq~Pz*YPHuu)Wj=&dD4kiN(Sx{-O^Jowj&snaVG|wJ^1Pf#kJ*pUaY`@`(aRlLv2d
zUSZA`jxSq$HobgZ0YjOJ`B>-Ka=e6Vi0?t06PGl2HejR5H%IAuhQAs7pb{jT#CUe|
z)9&rB_>Hmp#E6*mQ0C=&`A3C&nE^-HZAkZ_IKN@7$T!Lmdc|Kc%&2tNVnKG;@o*wd
zdO1`oVA5rYzQq9flo)sm@^Df&c#{P!s2rMAY>Q;J3QeaAgC{am$hDv*7afQ^Lo+~#
zV>Oluf3Hz8Jacvf3<PE;(aXO&_+jsj$HN47Ug8bo^csvOO46C9tk{HAkgW2)rty>+
zg7-_g9l5bDp9iYEKAeMQFy~gq2Z5Ey*qV{oQ->@`^>ztO6{m0a_*=&a>D(H#5>4(*
z`*L@FU(-*40+uN{#O5O;mNo`ueF(k&#yOTMR!al5>?^aS&f&o@@0|HP2hE2|kp^E3
zKeB`%p(6^C{KXbJqQwh=A|n&_Hy(pq?>{71$Mg2|<z`fKH;7S}Rf=^Ipol_dZhLUp
zxSnkIHc}PMc9Q#l)OyBq3T19&b98cLVQmU!Ze(v_Y6>(rH8mhGAa7!73OqatFHB`_
zXLM*WAU88H3NK7$ZfA68G9WWDGC4O2FHB`_XLM*YATSCqOl59obZ8(mGB+_XARr(h
zAPO%=X>4?5av(28Y+-a|L}g=dWMv9IJ_>Vma%Ev{3V7OlxC3x*-Lf?t8#_Dp6Wg|(
z9otTx*tTukwr$(Sj?JB9$NAoKPu>6i_k2}ft*TYCNB8JnJ?30hs}~uOf-;?uiJg(D
zxSg#t9TPnxH$d6W#nISQ#LmQ2!O@OR${Fwv3daR70UA34j7-gewr~vp3YIgpF$FM4
zYsm>|Y0&&%91~Nse}_rgn%Mz3|IIdWvH$N3uBMJoKs#Fi<v+=(0FI`HCU&;g9{;Av
z**OD^O#zhuP6_}z0Sp1oj)o?tHinK?06R0le?2fj*~7`%)W!)QWovBbXm96e=xk~N
z0NMhaEldHbw!nW`%Fh3Q|Ilqs80;JYcK>)B0mgPNw$6?~Qz!cWtSe$?@8JkEw{Qkf
z{?C$`k%^J^Uw7vJM_GD+khL{H>0bmVfRd?`siW&Z;D6>9H?#p-d;FJf|J~ca6a)Mt
z*?)!toy39erX~tNXJd<hdUrAXH!S6BXbm(LvNg9h1<)}u{To!ZH8FLx2HKh`*f{~6
z|0OCN6C>k)_*5)_##XkbPEG)}f4fX=P5z^%*gp-K0Bz0xrJAjYp`*$F2mF^?!4PQc
ztm0vB3Sj)7-Tg0{>3{5BhW|+G4$x+#XJlmhC-9%^zgN2dQBK*}(ay?L9cbcg@&C^#
zENtfv@S<bpWCqZ&vatd<nOFf#jO>g)|C?^?;^=5<>-=BG>YrBr+cpFKql~GkyQwkU
z#;Tn$Pq1ZrR)%vuX5Q$us75;xq!^3aNcXs9xWhS+S|i9vZ<T^r8t#t;T|B&S?jv>y
zxvS@!jQ4hDhUlS@One9t997+cF&syQ2OFT82qjlQp&Rbi!r{0)fw0ZZ;!E%jdgS>6
zT8+_wt}zCzkrw^0VmHPdzC=H-g@HI+-~D~q{hmKt89*>TO(sdB1O}}8dU%)bs*h14
zcdW#>*g60llyyh~P}R$+Tmk*#3@m%dv*0L#`~&LMfXN-JZ(0~b6(_cDjsqJ`|5{Z)
znwz%=(d7QE@&qcGqXD_o|8G%=OT&YpoJ?&U>J=$Vj$%4<Ao(y-3vIt$4yYahY|E}v
zQcA8Fp=ju(y)w%YUv*<VOhnXZ>{J@)U}Ixl`Sf2KnA3LF+-c1`ix|~cO6OWklHXLW
zD*ytlM@C&#nY@7I-XiO>N#hy47!i5Q&CG#XhE-;KsYoG#OonRJdU~K#o5%nF?=M<0
z+{@8Ty+0DQlr7U|E0EB%8m1@rF3I&hVq<(Zn~?|-g@){lv2Bh^d$~+N@lQM^uLT`Y
zRha2p?LL~ChqUpodAsNo_##Cu=+8sXa_v`Ay406nDaI9{|E5{yY1IW<ESEWO`|z!b
z7|K@goQfZ;_RSYE+#&4yR9Su@CB8Ztq+*3miy_&{&d#8ZGUrOj+KVg&A~RN=3u?f@
zH0{!IQHMOdQ1(};%)6hUSXeI?e0ADZ^wzPSWP7&%)_&uWk~~~JsY2JMP75RP4jeI7
zH!Q6<lk}eUMOD2W8P+!xrCArDG*(9NG(D)h+d)R*B>9gYr(RjbpLcWBRJ5g@H1)Mh
z>XB~9RvoF*lUMGx`shsB@+KBLY?4`tC|Z4s;T+wR)e?l#pb8EKt{0sz<ddob4}7A$
zwfk|H*lg;@&oV74@bSGE?e`@1C-z|dL$^><fc-@ecl?tV#YDU*%Ol$8G(4X0cIKi>
z@?E0_vgUbUdZ2-bjjYvCokpI{u`57f2!4%@*-fk+=N%h-wPehZn8TH=ScF#L5JyXg
zvTMqa4-n1x@zC7on!4KDq|JeITEQ{BDaFkCp0uxu^s=}OqeVBXARB>}M$KwUq3M<6
z%$gvEgKZM61i@=gR*!Z?n4-+2Xu;=a?%i=WrjQNO85Cw;>-ih0PutG7n2`}HJFA&V
z<5&;vSIsbP6lxTL^)=E-xU>K9U)7r5o%sbmx*jibanB`y#UT7{xrRy50QAcI??pwg
z8Oxcn%rEB!hCUT5H7ONv>x13qCx43+ZQu|5y=swckA`1{_Py4OKeyuI-FIsCeVfvz
z0Nl<{(uWTGccEaF%P!%x9aJSn2~^&do`$5kt~x~206}v~rc?P$?;LB}(8?TMn2pqv
zNfs$$P9{sbnu(gzV55a^@NVr#$JZUb*ZF2AmZweCqvC22QkNze1M2g8<5ei#+vEpJ
z_m$HYpqqu8MAI@oDO;q@ha`5L$52ZY{LPWZxQ|+jawuUDmx5V~%I<J%-WL%{#iyFK
zGpF9Yp1)A&^}$cf$kgGgTfOB6M@fyc2C?3dw1bM77bA@#WE5OHVgtTR!+jNUb3AY(
z`eGRMB0Qz?jC6VjzFSho4;xBVqO*D5$7b|YMU4zb3Uu@jqhfZ2tIP+9!W1#{=$e>@
zYyyt6<>5--<@r_cwe?_NBwH>Ce2A82^z>~%$cMNx&6`Wm``ZM?9YG3W+hEo1SJ}$#
zA}SR~Y4=~KsI#Xh-@J!k;>eg_>Zgk+<})_p9^mw%*U+{7T*{1)hZ%C`_ais{l^-sW
z8~s{l5WB+kl^kx@{B)+W6m3B{CKAcOpKx@7zTU^YElRebgeEXTKiF=1!qC5m&a7PJ
zV<^BGT)D`!5aKiJANmHC<eDVB=2GO;l2H>kc<kDxo5MNQC#1aH55OC{g1K!ZpY7ut
zw0KrP;<Zub-MyZbGeCTpwJ_7pkP9+~*2Kjs7)<c^;~>9v!)1|vELqTx1S^>{S!~?+
zUV!{g2vBfdQHP$p5w-rD-(BrBS3w(V1Y1xTfjc68tnu8clylPha@PYbdXFq@t~g}n
z5S9mulh6cF{2^>Zp10Kh;PsJY;SpLP4t2ApF<D|^%BSD}-TwV|{yGu?u>baM;A%;F
zH<qv1()H!A+CeG3)#gV0DtUTgxnJ$b<xiD#d<?YsaVM1+n|=#ARVrsI(lNli%W^q4
zlt&C@3Jv%UqJQNPMgRP;@RK8m;C!Ga+xN@zi|*d__s%!2Ybb_zr^m<+69rAiWPJEi
z`oc0y8!T(Vf{b9D7f~l1ICNwL@nG#M=!6F=5!s|n(sR)pfK&k|3AO`BTU`4~>|i2_
z<9B(faQN>{fx}Y{9PMI(5})vPg_w?CwR4yzfsU)FK}cl1i8q(XSt$fIhu#M#C@#Mh
znt;(RyDI<8YyKnjT$fX@)XX-giKX}6e+{U7Q{MR<@DfU-HgCFX9<#z$1ulFleXAMv
zTJ}D}X!8gxgoUf!NUA*>`u*^o2`r}|*V%w3TW4J8Na*iMFA$L9WDA3l`YbM7TO$*|
zonJ&_xs{+pv=Dq#t@t2Z$W@a(7&TV6q1@JSTx+M;<Y;hk^q_KcY?4%SF_A>APHGxS
z%$jQfoDw(^c@$^!8*)wHz5j*=HdXw62}m+rYC7qH-B4B9oY{5}2^u*wg+pg<mp6wb
zBPJY)n>j4Ek|{Sm@K-#MjQR}FLQ&3xZ#w(MUWK!-d_07QwE*G0pJ3Eo)3fKpSj-W9
z(H51Z*s;`aRxEvzdq9AP{CAL$eB2<L0IPkG>vnmU$W4^rzkf%cus)V)lCJfk9H;X>
zs3nc&ut8|Tk>U_@a3Q<YF1B`LkS5>GLzzw2ZE9d1FKB%A)S(1|3+5gbke{<gxYYV3
zD;A3{jkA|kAI!HA<_{-e=%E}BM<IWInEdpYC|0M#+rs#K60d?z3x#26{m6dOAmO=S
z5f60J0TJ*1$S~M$$F9-1G3D!>6zhY3;A@jA<l~gAZnRgS&zIHI%2NTq*+Qm@W*bUF
zPhlu(K!_+uBX$mCAjp~DR~xNZ^wvR(pLMWJd%d|$u&<cLfvE}^kJo)ZP}07yHCj5x
z%*5?tjlvb6S%wQzT)O{o)c|8K=41)VaE0^BXgYLwA7opooE(T)t;{&Ty2@5Anr9;f
ze_ASa^%XFkLmSr<S?&#zh5eE%;Jx<Q9<<j`=Az2<ij=mEzp~8%kdDQ^NIU&_PU4sl
z!*FA!v(y=u!fpENSG`&PU2d=F>hqR2QyUQ*-}cu+Z;1X^iiM>4GK3b}*DdXR*^#Qs
zeHAdYwnF?6+1#|Z8KP95Jj(mT@dbE~>QR{Ai&!bGawPr)c|tY&_)Kq}w<Jv1^CaPD
zP(!=u3+m!{swTOXlDbXW!_TWt<Iev*=;$+0{(T-|JSR}zMM7xBIrqN4VG5YUwA|J6
z<P>U(!bTTfw<gTReoR|VS~Zq>w}W$fuf8z+o&K5<cE3Ks391IC5Jw{niA!K`)w&-I
zlsCyfL&R<Mf5}ChnT}o^toCbj(V~-VlxiK3`Wnq8Ag_i#$uJJaI?RFlL$=}1BDHVW
z2sM9OUDz$Dl2<3;aFT(~FWdA0P90u|W$z$y&aMl+zS7-~MsJl2S!PE1%ER+p=KaZ)
zJ!w%oF-M$1-5b3Q<Cj!5Q%|V)(ZaZl*HSr9FAWJUl`HWWkG-Nak)-i*hQHn%J7lhY
z?-4XJKcU=eelL1b!sEn_;%C!+w%#oF{1S7ro=cn+<wRP=#Wgh++*h+#SowW!XYj{a
z8<Z_4rLUd0|D4OYl^bq8x32Xr?k)I$w+>Bo_m<f#77j?`3e@3Dr6~2c{NV>-Y+Y5-
z!lwVzkEg3*$I9bvzoi7noEl#}D;}?)7F*OQZ!Ift2d2pGp!UO*O@G~X){bDv91>cm
zNxeq2^gk=UZu0C3l&-{EwW@bQ&GX-+mPa@Zwv#xlvVh#c*mX;xT%8~DPoQrtWX&Rp
zO2|b8K$W9mhH>z~DKp;Vdn^I7HV)YuwM-(2y;zY=bSGcn-i=UQD)}kq{L~agTM`6h
zxOUi*34EnPVVOMJ$XQSVk&K;D68a90LG=g2bcxlDS8UQ!q!s~qd{&?DC_@<jFJkDh
zKKjCk(Ot*xdPqe)z`;nRREj2pVQ7}%)B^1(rD<~{oQ#FKKhS~-&4%D-N}c=t7+m0^
zcq^IKRRXuCLOV*b(4&~&ny7@Mya)#$a`Ju=1#6|HAxpXSH?{&AEmKWBjQ)|v3#wjD
zdbs8Og8vkJXLV`nlVm*6r=<(f&mc&BW?hJyL3}~N1sZvL2!$r`;6d6T_ckKG1@<7k
zr=t&H#$gC$iuk;bmDiuARIaZ@IJwC)I2(HH61Z|K1k%MCGYCe?z1*zutUM;Tw2QZ?
zHINvmnQ8KevKRavz6|BO8tNCMB#|WGuG*MNRh|phxe;q|Y-L+QvrZ8o*;UOVc<ZAJ
zL|lRsc>jW>ch`-)Y?E3sdMSmhsAGtZy8#+OZ}CMw+ObbQ`BrTPw-~I{fa$^XEYX90
zIxY{L;h<wUJ3J!fagufSv#9@*Dt{=}U<45n>3yfp)1vP8Wp&$SGe>bXrd+rkTRwKg
zx{y(f>tJ{u*Oc{g`;z`Our=X*sRR!)kfshBHmZSe8)U|1!P#w-bAZ;&LO2OT8aS<P
z_ggcSB6TZd9P-7fRpYqHng}1*iq;wBtF%x3R;u2Gh#l|ql{k}FnHueWF5D;b`;EdH
zlZR758b8ezfA9>Md<8s0>RY+<rv%+Ecn&qPt^T>;zI6%cj0oA!(wIG19gKks+F$xk
z2c=`u3OZ;^j0e2(aI#;d&%KQ~BH)}Z)^nm3`^9IFNb9O(g#<uSEb)C<m6CkL!=`7N
z;c7kbb)T*5K8Kz`cd@35&c(fLMDs>^uQBtrIB{%poS_bbC}Lj2ocAOf<Q{r12RSTF
z!gM{_!pL)KbE0pCdo#bOzu4Hc+w0SkTJRgbvAUE-2$l)X9I^s01iq6{Qmc_f;7iEr
zW`VT|gZhx7GG?hNB0z0_TAKnIm^l6yfxc6O@@&P|!3^<HZ|eRY<<S0(B<(vglqG1*
zatkunACVgzsICQ_6zazwU&(YW6D=J+{JhaN(6MHwTQ+HCaXQBgK2sR}nJI9t!bML_
zxeWMoRG;Q_V5rw-ZQ`<=U5&Qaa`RU8#_+9B4X#DJlr;w>y=brIx%3lc;ysfDB``37
zx6RsAlSLcqG->J4NWU3k4l>)nO-iqIef3xH35g6%-c?WWkxu*wVl)$T=gwkJh&a&U
zZp+5iK6+!ItmD;VziDeD2^F$)y)N@eGhbuMM?EU-`0Vh)G@MXL+`E6SAS|EPZusLT
z?~0KB`3=e;qo4~_-hN{y1l(>msg9qgV?P^_#O?*{=d?d7S;=Al%D?BhffLngeZ%e~
zr1I7to}V>5sU{$nc&+alzwoqvKJ+55-a84=jnl5*yt&EZ$hD0IVo6ExA*)$AbB|4U
zj#)?5(uiSZ5n9ZA!@Z5Og?HGJPjW(dZ|hyW*|2CWgWf;HqI?={A(5F6O7VDtks^hy
z)X8-bFco8tqC5D}s|9xw>nvGSJyVS2s1lwG^Jb32MRXnx-IbYs2Q4p^!DT^%gbhjH
z{TXf9xubLkaI-Meb8=E8Hd*Yz2-z+}L2hEwqxMfL3VZrIh`*z<ELg`fxr7CgUHBcU
zkhWzcwG(Fkj8*7Wp_emBSPB@#Pow*GGi8cDrjCO-r|U(C4#Wr6P>SOw{wW#LU7S!F
ziVv61^ehJ_frWRQVwQCR$sIGc6)#xCMWFp<WN-re%YcmvgT3qKljTVA*3Xt^8Z8hH
z$6Hd7RgJtI6b%R442}nXgOs%=&)9xeK7jYMn^bu1=dRNZ7xrx4-JgpK@AdH}sW*(8
z`o*wo3qWe?j)lIqZ+mlnH=QsCn#VhpYA9`9hjcZ`1{0hpj{J6JHl1BFnA}{T4pbOI
z3gR*b_7i`GLGld)d<u*R8AZWL&%0vgW^3@r4e9-dYNZ>}T1~phX^k7)GM5mILf=h1
zTg2v{&;{NeejX7)TE|r7Lax2zW1+s+4y1=;i+>J%KOR$VR6pcl3spjCqfxc&3jfJh
zd(FTyf@iQ~sFC57aJ(!3FAI5MkKdx_l%t*qY%>`IFRxmX$!LV^4GIKLq%5<+L^76Z
zflTi8*D(ca+wA)zje@mBX{KADdf5!+GP+^e`mzIzBaFdfOOv_Y1<(dpm`=du%ZkDz
zcH>d1z=L(yM80fcBOG3W_Bp4-D6aCbXHgbtZTK^>%C^7z+b#!)prp66G~o>}#UOS(
z_2V=73PaEab)rs|BE@b-v+q*lV0iLAouGupid$5X+{t#T82zoJLxL38;gjs=MCzAU
z&Lme$S#&mnm6i<Zsd_MX=rp0`S|!_Em)+dN<6-wakHyvVJmn4kW7f*aF^Uz{{X**r
z`7Z{wUZ@HYWA?V{4ss3TKd{eF6p_y7%xgrrg{$y*leJ|bT;-pMG+Z&Lc%s#^GiB}|
zG=CTez`QFakfql0LW(Z0JYq2E9Knt~*U4xwEPMlkpxQTi@q7Z<WaqJ{ek(`e2&1aj
zI+BHh&)!*DhPNsddrFl0PO5r+OKqJ+Ftq7{+yZm6i8a2+W#ZIyCkr<s=wP8kPM&G9
z<Uh+;rq<8g+inuilVr&(n$!-~-f7v$wC$t!_*$7H$O^M8U|PqKlP9sG(2My3iLB?T
zxrY2MO}v`$53@Ne%dE`^_vuO%Iq1@9jW4_nX<Ja825K4|0!E1BA~;y=ebhH$Xolf!
z8AyLO9vh;?gqbja9y+GmTvGbrj6~WQGGrTDOw8#SG;PNBknjRVEU{D06yIUDk>_Q^
zao<>|h**2XFq%O!*DvPzKEFaO`oFawypFO9e`cOH_T#p81Ox%AA|wYC#fw9Fm|^>J
zZ1Sr)_gN`@3+`xQMv94s9TbU~<ra>p40{N`h<}M>mIZ_ck!Wej(+*tlnP}AzKZY-g
zg1*<l<)?>RoT4H(y<Q_j8UDOZNv2`byO~|$Ig7FVC6wKn+hYk@ZJGdosIjq9T7Cn6
zQh7fxln|+qI$kh{6yx}wrB}fs>!!q2_uO)lW7d98faq>9C&gU>GB0!r&ve$3#$?jC
z=1oM0ADBxc@ApZmADf^%(XTVjn3?D`gBSQjsCHr{Yx>kF{VK`<%gQXq&%k*|FD=}F
zHVO%;g`9I$7FsfD`PQDWVdkPD%QQHFzRZ%ERo%AqDYd_3PjAX=k@D*4Lyjclt%9=1
zBg6@rPdh96VJ}?qJerassdL7+dW;uio4ndzs(wONm7>7yc3vDD`5Kca4wO}tdD&{8
z4ffIgXI?XPgMyHm*T}=swx-<WX5UP*&V(D)Ik;NLpkB+7@C@IDY7P$n0i7*CSP;~G
zZD95Wl-CCg%?}#GkAb~$;h5{EneD?SnsP?a&E5$5{mAQ7`PjJbw2o86cN9}ee~Om)
z$rns?#JP;GadW{Evn1GnG|6emBKYFifFk6EDFQGh6t35bu}IAU4W>4WcPF2{Tx4@l
zyd+#lmmOcEa_z;SI<(^|?_QY*Yc(;0@`umL9=r<9(8mhWD)BSuyc`n=0|}VH+;0e_
zq~&AZYO}K??x;=4X4TBy(y!(A>5@!GqiQ3?5YMvwEErD&u~v#pu(k7lleW1@)Fcn_
zY)T=bl#BO2#GpDexlDtRro$|qB2yjNK=5}01rt`_F2?{J%H7;~H|f|AXNzB$urNH}
zaKn2@6KF>9uObAP2rSXipBnb5u<{R??Y<S4jc|rw`2xmufLF*QUEyZ3dp~?BpqGWD
zgYX@_E^KUIRVQR^A~h&jC~30_DaQ~QyF}E|-4K5h@(AAr>Q;W@1_J1z3tzqM<A7;7
zH@s_@E_0J`od;{nMP5gRNJ(qbj0}Nn?`E;ES%&%u<FM;oIn^#h`FAs%u?xI(2C$<0
zQMzwYBo9)^Y(1H=xO*yiDusw6wXTQCR6O-WOd=^SB@oA!v}$&<?kcp(*aP~}dZ42g
z0ZPvr%JBVhwD4blMlL7+(#8sv9ygI)ez&15^W9_%K_^EtAnyeIOzyJEb@!pQw-#N}
z(l9p}AssA2@6c?JqoGT#MmrK`ayl)~cr(k(^o62%9r!ysI^RD5+YGY7mV&=s%R;Hz
zk#L-p_frt#h94t3OZp8xd1$8JL!tOdbR%iNJ1e5ktlG}iyizQYNR`C|4|Jv2XtgAf
z&@p>^nr`kqTiqR|cAQI&r@7-oizx25X(!Xq3pARd%8%rs+j{)RqZNVTzL_^a30O#f
zMS)lg1ca_9E2pXUjipZIFaclb<hhdF-l|cyi&-Zp?ykwW3eZ^x1ZR_vG1_tQ>7vgB
ze56{Dk<?QeWTmzVN2J_&8trPvf;)UC?QGLB-`qL*BS=|3Y2*yDuYTQt6PXhu-XV-f
zS?huUA8a+x-^kBVjBFNh*Tmg#H%IazzXCtc49L#La#Sc9Rr_O)41dWt{L~K4JszEo
z&<X3$b=AGsAFR>++tt^+TqugEsT&iBxArDNm`pIt8kcFaP6;Ec?(}0`wl3zn_g26#
zsMmwfKC;W2%t|zFJd3`9xA-Rk!P+aA1)H_34D-$I^lenMF;%94Y93O^+yQ%-A4W2-
zEk=b{r;2f0(DNnvChOqp&Z0U33vYpxqa3k0yON!n|8s$xTiKWSApMqbQns|q7GhWB
z!meomeU>M8BPRWGxMExPG3EVNV~S!=gsNZjId}Uhq>OI{Cg#*=UWQ7X1jSi667!oY
z^w64>Xs2kaWw6mFzeRFxYE&mu0R6j;NwLB>%^6k-Ge7m*vGlO|BJn;0CU=7Q>VmTO
z592z<SH61IDq;}6-1p5pK2E)d<S#XRpL){=go>)0C8B&El1U-USP@wyKT`r9L&q!Q
zdSUqd9brHz-aImH3#BOQ1ZLf#uosCs^|C<N-qBLhlsVdsc%T96vUl<WTEYi5R%)T;
zcwGhGth(^S@+;<!5tMlc&>1_9-W&W&uP*`vBq=RfgOUO!Vl;Xj@FV_I*@Gdz3<~t2
zpQ6I>%PXKn_wbnmpyhJbHS}%8r{=p9__GZ@4()q{^gfknaUI3AGkptkK9HHQ;IYOZ
z8cw?8I8y~svM*~1F{S(LT8?1s3DhwEy6s?`1^ubdn!)D#$ul_+@v3HtcsPm<j2+Tt
z(Klr^Zdy%Hk7geD<Q!!AT3*4@<l@>6&w4bn7cf$MifUboQ)rwF$O`K;r!xD?ut^TZ
zs4G=9<-EDY9K>uYNNkXiPq^cI*guSgJKg)`zsxI@OBL)gJQ*W?Aw32nOF6@QgAF2<
z?5z~>@WUXf=(y{g>&jjNOhEL!U8Y5Sh)_$o)Mmdmmzm}`a=D#Y*8ULsC(OV-qfJbC
zh!z+veqolf{5w8DHZ{JiR}J~`NIl^?G6AtmfeEc=sYSHByQWP>AR9eIS{;}yKG;&0
zh<}zcK9Lj0NPF051+T1$Km-I6W-ej24D>`r&CrX1F~b_SWoPGA^8Kb_Z~d`y>Qd9n
zhJ5-<mEZ=H#6Ag)vjU|#Mag$Ahmnbwb(~vl?B{4j*+siIYknchW|hw(n6qJP#<p!y
z+~Rx+hs@ww!n)24UpaXBIjkWUiV<XZ`O-#fNPhnJ<L>JK&ecw8KnY1MUpw)>U66Ka
zp1@e&b<z2Tc;15$PR5_y^hH8cLXq-@Dug5S2!p)5qU)^Nh@n*G@MGC|jw76O{#ejA
zdw+q8mxw3bC>3q9xt&550xbR4cx70xv>oLnJ3-lcRYQWM5@v%<mT}nDIZ*^umagY#
zhRvr&)=cmSoGA76t)N6muR%@#!ywyb4+P2~duW#Fgz0JmuF+9EDhz(D{8#sn?ZTd(
zE)N9Tx-|qa1PA=m`y5&Ej;%>E5sTkCST=Z+3kJce))21UuPh7Ur1BLpS$7pgs3UkE
z0bwO#cvVM+8|3A$jYXQwtS5QCE^!^{o9e&uIVr$`P5XI8ZuMUJ?G)gUH?Cfh44iNy
zO^$SDr+z=$x^RY@1En9B`Iu3#cOuyR;~ey-YNW#aGP|oSd=@^SNL~~jQB~xk^YwH>
zMv(D$<7u>VPNUV~G<X?u&i!t)Cr3vE!WnZH!rm=gJM!vP@8@xBa86Zf=V7dbTk}uq
zmEp-Eyu^_uBh8X2Uvhv%j$xL0iUBi{8To==`Znz8UcZrStlxWbO^PLHXZv_c`bwe2
zxS6!10n69y6j0mZf8TmtXsy{f*)6sNT8yKLsqO@|_GlXR>hU=hX&o}9vVTYnE!iWT
zgT!>QE%c^Vovy(GU=Cru@5is@1awFEs~>RZhrs!J^^C>%IFe<hKn}iqmPGXf$k#%3
zK)fA@=u@;Y?aow!7*>$wkBLaJ-kmjGU%*<ti%;q<%6*LH6K}mYVS3{lG=BH+4=$!e
z_i|-J@HoER^;$uxJ3gK)FPuL4_=vn54j;q^-+X_<If0cI+Wit5Lhe#8j>WbbYbsp*
zOVm8qGyRQw&DdKNbz!)U7c16fZm!-uxn!w}W{DcHu*b^K++kl(^moB^n;?V0y)8XL
z@`8MW>aaB{f{wLJ!mh4sfQpV^#eu3+U=|{r@iVLbsQw-mbLrXmkz7(iMEGlwY(`Nr
z3O$gC@Zkm(9lTf`cTBU|<(sXMx&>R!ykEk)&{apCIebn)k7f;hj~4Iq06##$zZ4nU
z7`q&$mE?3QoQII^T5ztiIy#(q)Ei?eOa_NPasM?tbv6Rdd9-~(-!;t_I}WO0B~IZx
zVke{TtX;;-fya2G7MV(c92UMdy+shWt^^#Ufj{mdDQ|+DMy;=))Mgl0N%NgjlLcyv
z{z<I?8M{$=tBKlv$GN`kN(Rlj-d?DPns)hM>qu>=awm4ZScDA!M(=AnqMIHh`p*RR
zxkYQh*ZJy-R?R6$KM7|Y`6T~oqponLtPY#4{(U*^JC#*2th#+D+5FcS$(S?9J%sWh
zP!e}1U^z+GdU=_ACaVD=F8tS~IulF#G?r1yK+Z99z*$Z`@G~7iFvE~D7B;{k7Y53C
zPF1NKHvK!X?3XVq7&`9B<X9m6d+lfTP8+>vGcnVoG%8n&wByt#$%=$}`w3=KiKw~R
zaW}0d7I5B-^A;&l)A9UtH4`PYGKjXc*tWdGfdqeO7rR+s(zKVK2XE~~zyAOqCPYjn
zJ7V35Z`#E`iHL3XE`XA$%<hM+nO|<os9b&DKvtDk<$ww)oBv1Ksyzn=8o#i@az`&(
z6}X4PDvTq&vrhJ=mg>aX@0qZyY&#|^!5w<Ud?_jRsqM~Wg~A?-8ZOjzmWP%PD`B_l
zsJ+Sc<Qq@&%jrbtEH9_E1xKWHN5n!RJB2E%yCSWX_;2^I`L6ij|JbOPMDRth+E*&3
zhYJWCqIQc7-|R-e+h7NIIr|IVT|?eX_wbcWoJ%xRmyJ_ARpkO%O(jDoZFjQ8;28S;
zcEP6ugm4Jch1+(l9R(5Id1@0E+|I{z{I(dC?%o5Gzl(-yV6FtO{6jIg##?neB;;s@
zgjL2>wvgsluVL#xheMReCqwO=zQ^{t<tCrZFxzV1dL3EPu6vyrhH-JzYF=ltG$Ejq
zc;lO%oEV+#n5mFaJRq|$6ooPwb~%;3xK3|J8GV>T-puwdV4dL}J?_-y2l1-!j3M-+
zMDX&c!Q@3g^9jl;{bf9fSVZPnVsfSiNjUgLVmkRFJ-&9z{h|0|Ip={H6;QG6Q2o}N
zgkHf!O-Ro9X$hsdp@0(lq4FwL<bKZ2gu-myw(iA_wd9#w?92W}no}hBy5JO17ZCAh
zaC{q&>@P#iT{1h<Z+~JHjiffQVyaV&+QNOiH!TV6FzZ?2Z1H17X<*-BAM`e{8JypO
zhOF%Jxv&+ZG_mFW-)aTU(7m(;I&;@|l$|XKLi00c{*i&p9Z;&{eiPstzkSs7H24Lq
z&xJbw>nMqg*JEke1FGhfJU=`ta|KwAnv9x(U2w~$TX2X~53`4y>34!0foYTTaERf{
zD*6*)0vA{>2M6$^V~allQW#21x?$$q&F*EH$L_;RUwZ_J1bL!3f#}B!F7ZyR=zOva
zIhJpKXjOq%gZcD&TC241r3o}9amnDUS9u;W9IQ=xvREV^YOFpBfsK*^%^EIy%vIa)
z;4|<VI($@{z)e(I0`WNu-6B_h^akDF4!h&;78$CMum+a|_#&!FVvd9adoH;9mBOND
zYLQL+=!c?j343>hI*%Ks7Mm9m4ulZbv+~%VNSa0>>UTa=8DFIc-gfV#+D}gjBN@gc
zHQXP24{!xxu%v|*IY1mTCRjcG<p(_w5{zW{H`?xJF=wx$^Lvr8LkV@bRX%Vn8=paM
zeB08`+b}|&<#ARg0JgnC&5=j=swHQ4{uy-*V%d92Zw2q(w<C+ujCFR!MPS?2vr+-V
zAtQPbL5_A=*QkU3PHbQRoOz)#qcdZQbu;6i44f-+gIQEo1uf^u$#qcM=-Vy!(w}u>
zCz_Encm(*hN89FUDN}B)TFx7OXYD?IM;OhWFB#s21|X#ubh->wg4g=c>^0jnn<QsV
z>zjFj6eX7ui|lNzd|&crl=jNBd6TXenlfn5%pCD}NxB@`3?7E~oMSBofPvSkBMsM`
zZp59d@)7{-gtxgYQyuD8f0zioB);~PddrVrh8lg3Ek0yM%<la4CL|dSqE;HH%;mCt
zShEm|oWbKpkco-yBL`#BC}a+?G5c8E?Bi0>2DPd~!{ngwor@JQR#|u?)qm_+7y3L!
zk;4#+8;E#qU4PwIE6V}iB$id!DC0>U6_V}jf;kNp?&!sKc<!Fet&Z%MS+64ZwsG}Y
zEEL1iq>A2Con<ogcDZ+9I+Roo(y5z1k9eyOm$=iXU4QNv1X$koxxXc@WQ_xF!u2pV
z=?T{Z@UyZTW+#loz|0H8uh!)uR{VA`W7=IOG<oEU>oyA~>WOVNW^v=~%%j~Qtv)uJ
zh8cTIOut)&7mADx+w0X1D`|wz;HmMgwc;l3I+}f$IRBcBEZju=!GNY>w3UZJILKW1
zTO~0s{g!9>yt8ZgAb6^&s<@yB$(vx;N4@RO$_V}2TFOYUg<F^Z32Z(#r#!8lgT?GO
z+50r2iE|^-%{@!Vp08zQ@=8gv<5Ef-^lpxQIQgm4K-zowiG7UNgVuHe^wI5S7Q7u{
zC4#L5%Na$E-Ws~WB<x`>T|rjVGNR#;B|0xJ+5Fh{PxWGUz@6cr&38fuu7f_WnioOE
zYm!`)HSUcr4K%uXW_*~5ZRBDaK5D+DswsEUcGb%7&z~Fh?U!M|y^Lnv$UlXsv2@6I
zAC74I`bRXeL_P=DzB;Lsgit&$-|U}=2z+&Bi&mL%0gJLpw%VFZCOc*%wA%<oX3n!s
zOQ2#|$#k9%8le|c-_$6-{Prox!8txd!9U9(%jHWYM;DYyY8z*?Y|=hdY$LM6%dmMa
zt#d5T8&=%svD}ByN>EIzE5NIm={0WyQR1N6hG!5AVu0zw1aimM0Bl@K&(*>L>IL;&
zqE!rT2o<8_`8TiFlG~fRALxBzsRcqbJ1b7g>OO`7o*ZxBE>pe7<W|JI5XXpj)~Wyc
z6JzM__K12et{H~VH;{A=-qaASz<InJJ}-930Ajg^D~BJ>#t%&=E#$@9BT~+jx}WiP
zEOs)N7?up|YXL<3$aPHRG9`f)s=d)Q`{&^~;Bp>DOJoz{N%o~PF<aBya=WrbYK$;R
z|B8rg*E$l)vrPX$NuJefV9f@#_^QwiuO-?j1Z0%P7s8TusNpTdU2uQ(v&#d|28}1M
z;PgRTb4tYaXrkD<uFH*-+3h+?WqH@f>fm{_xm3EBzGTj@^yeJb6^V2+UXbtgJo4V3
zn>qfi(ol@f1t#Kf$yjAs%N_Y{ZN=nKC$1bk0P+;Nd;Ovwvl~Z+UxUVeYUc&6g%L?<
zO0&Lp5?n9|iKhLa9)lvYkF^codYI&cG1_T+9IQSP-mjGfdydj6_O2*U_Vns~gb-Nn
ztbvz&Oi<_VI#`WS=qo2%$g-UZOG!iM)no|6wAH~1a5iVQT8Fk*&as%5HJsVB@)Hi9
zf~QuKFcdx&p0;AEm<j^oJFkX2r(gh*q~jrpm<yOQuQKG?JicJr(Tp9XX_g$f25$;>
zjU^U?S{d#Ar-|{PlS4rv>Wg%yd3{VMds)1vZTWSUm$fC&=O*c02&+fpT?cbtdsN<L
zW`sgN;%L;;u-aemkRvP`-&E8+;owj7REB)NP5^9&QaRNS>O7VT=+C)DxE&@lFq$c4
zk2`_a?X>bPb-&qrBPW^ukaz@{Vn~q7A;<g4Ej<BFanWrw4^#Td^s%^;%n#cD1^$jr
zDSwoV2VKy_7vR7vwx|a>j+GKN5U<rfVnWci37qNX{1=z@HK^`w@?Fo3U*t1X!Evje
zi;|KTMF;yAG%4Hop)+!GhmmpFgWMJ4cTE229Vr=ZLuO5@BF)0Omkl|{47KAA*|*d8
zeQI|f>(~3-yfdLYGRyea$G@K2gMhP!Rh<|Zy3D*8_+I#UI6H3EB-Dnsou~SrBI*1R
zczBo_KWms9JwrY(`xh#=g^dLVptdNB(T{Qy@_0~*JU}5XPWk&x@*zG<DXM5N5M_2^
z`hxv`axC|FYh^qZk-Nc4kwY_KOcRW&ZVM=Ke}O)j;Lwo8*oRUdB5ZCH7?q>osO{p(
zCg)ok7h!-*=?2UZ;?a0S)yCYA#6iTIoX)b|l*L6hrswsQyTnS#w6s04H>>*_!3?SB
zIbMMKz>DiDn3QhRAo3=XX}<rg{LBBWhrsV!4bt+!!`wk!eZB0RUNKjzE%t6^n`<$1
zq{Dn?%a^s85J>S$y(uTQsuq>%IJ_o$)k2C-S>2FI8_U+cT|6$hvgYc@;#?EkJY6Sz
za14YN&c`OZax8eB7eC~;vGyzV@a|vo%2`%K6JYB%1q;y`>gk!?4ijkmvdyLYl^X+7
z1Y_Pu^NP0<1HQ9}oKy4lv}5+$dDYD7>{R{=L=5S#12C{Y1BX!~_G1=PE1|%8?Q6`8
zfOmP%c%@V<Fy6jc6nA8|2F<)_*2obCdE5l+TD1v2{1!x!+#DjgzcP7Mo=jMb0tNh$
zk&BIZ-pwi0tv{s8ZRFOCaFazGt)|F}9cSUc>$q0VHq|G_6#%-O-d5!w4@ZFFv+*Im
z0KiS6nClr<%d;3N9q31qbiuF1o{`trx!oMyrf_3{oY(2j(g%|}2YWmEHs6mX{KYD}
z{3aEQ3MYmxTR!A|B{tvRCl=u*U}d0}v*8r!F!FXnE>IcWRA95`C+>x_jlwImrd=SS
zcdOrjat^&3FZ#fzZn=H1`KT&wq2z>qceRYc-!Ypfg@hl1ey!Q6(Gl#u!Fl<Ofm{3}
zvRY~A#6f$c#c7N1*(!rJYVnJ#k?y3Q6~NKJwPd4T0$F_XG$;`5s3r6&AB_`0>-Zyn
z{66d$zH=miFtMw#czYYm7ZAc*@xgEPYa~Bn%doIQVmfnxVj~hd15&20qnOY>HrL>Z
zn1S|r2@KU8Q2+<8n2vpE;LCU`qS$%W*a@`*O$Zdp>V8SHgLQPU%t!C%@LmtiW8c0V
zgdALfc$3>%)aZAEoM%!csNQ*g<;RAms+amWr%@(xAa664fqpiahP_G5`wF;TQGIyz
zxFn1j93--_`ZuG{@gg&Ru4!g2rokL)qdte~VX|zDuG+PSARK&wGg<f#&B6tiOmRNd
zVVh&A2<?b?#44&kW2=N5-5s5zN#GDaAd4bB#^>#hWN#oO5PVnkvhBl5$*~@Y42Jb(
z*gy0VW)o&DTlzg#O1JNU_>3fQfGFWp$bxV=^y;GaXttd_#g~3B6ZW=O5#yUw^I5vI
zGTT$!b)t^z6Ns70eX%9e(ljO)QLiR=Qub#O3WW|Re;tBL9AfTCznh$<ADY;8{!7$h
z)Sg`SfSNVHT<X!fZ$tf&A@Hd@Hw{ik14h6;(t$q>=h9w?A8FupvQEWu!!4LiVgW+N
z%#o`L8TaRC(32JY*zwFByGch;=CtFt`FF{4-^->37`2kgdDGhSL823$laqamuU$(q
z`G~5FXK4G8sk~7-qpX?P6wZ*{CnR4BxEnYB6V~IWne1bMwyjq=GTu97tiG5t(Ac5?
z%gHtOqh(?~+Jt_@;fi6Vbjlmms;%`Wnw7SI-$3*8IZEJH)Viz@{AU|2Z8%RLw25&D
z7ih(6+uNH%^q<o!1YPDl$57yoywDN=#y(X-#fu>bI?}DSl0@DK4xcvT?#|xv_U#*`
zf%@;{W>3>$WId%Ff5EDE>SJP&i~ea6dk?4;5QuK7Yma+}haJz>f!xy83KXK+il}tZ
zkkV#!DdsZ6qZ4DJKl(8X37G;pzHoAzP-m0Pj6hnCm39?4IPsqUqyJIj@g{8o0kV`W
zA3WzaZza7Wh&L4xCt%r|ue+_l8Z)AaFodC&q#9!`aFWfV%fJ~7kygerZK23TdwTcC
zFrD?QEtBbgf1ruO!8K_9VcrsEXywn;H`DD`CoBNdw}8i$k>HTa+a9e#tTL(BkJ}c1
z8IJq4iGxGod=gAapx&H?WUAB&W>q4UzK*t@jQ?GVvK=5XwU(&DCcyr9Y9Fxdy3v?Y
zEF-{yW9OX%4C*BQOCr-*S+VR4vJA5DHybugUuBoKIG6l1(PhJ+SB`|L*J;`)K}yHT
z_N^Ehm0w$m{wt6{{y@lWj^WDJdXF_V=GMr>*}!sF;+-xxEcbFIn+rh-1mt;&gr5on
z<p&tFc}rziO4Ha{Ag>rbi90phM5KDlXMdr+pXlaahvJt};V-=m<6kOWZvKAvFn{I~
zu9m#St;6pEL&F^*tqPsWyuep*J;hfObPMB^*w4#szzYi@P3a4$Tp{3Ch{H^{_aipz
z20nUc)R|%$?MInQrkpLLB+Uu5^!eo|jsEwTkdVOH@+90J6?KePszN+zD<Y-Vd$_o#
z1MtVY(JB|e3^i!`^2CCqG)?_HyMzRMzjXz7I(XW+%B_Hw7DN;0m2*p6&fBuiwWe2t
zslo{Sd8BR%%gZ^J3<vM7^}o^-k@b^>50Q7jNrNJAA-j1D1r0otJVBu)@hZOF_S9t5
zHDDMr3-k2&c{tHFTnMp8`ok%_EH8e7bQbEzB3NVIjBO)iFeSRDIF1g<<M#SplRmdn
zUBNCj8lH^Nk4qbQZI?5P?zN%Ype>mknr3>ggT16<UrGN#TzIXi0D+s(&MDtRjpPgr
zDZ*e%6DI=hGCaxIeg?OP;aUcC7<nD_dgv%dH`KsIS~kKv0yVliHB%nTdehb5q0Wp0
z12XEu(;1MhHlwYF`@OiEPx?`4BEPz3mqMT;QTrE0iIZnUViME^SdiT1Y!VVC35}Fh
ze-M&SHU;7u@Aci9{&RZ`boJ;zbBI+HsU-vw7;3~>jyBM5{85n^tE^c%91A&sL%A87
zGUrUBKLg^*K$ULl`c}m3{`izy&so-IjY>Fg-RQO<(30H<S^gCLEE?d@ewk&4o5>6m
zeFr+3Q3>QnJ@7}iY?kamcEgGZ?I}LOg?8JGP$Ck*o+T)eLSZ?^hO&vcNL7d|O3wV`
zCoO|?!=$Cw?w^d~2^le>cu$0PBbmto?-mp~2|ftjEbWGm@?H8BJ!z#YJ&l+YGu_WZ
z@={eu;NWkNef;JI7tW^(w`i_w^7(W=3sWd15HJ{OqqE<$5YY%s;37`aX+4FdN7Cbt
z8nM-2ntNwiym-yHV}8;86H-#JRSnmjY7$pJl!eQp)yq}e%Q?eV{8gm||LD=B(+jm|
ztaZiTgMJlj0*l!Evxz36Zo&8DRR<IAgP>;r2Sjh&sM~HlZDR^vg>CLwR<Oh-Cp_5I
zPx(B3gkq4c&r-(mB)Z|XC~u#RdDV`V?$5`fRM)`Zv7eUG&U1Tg8oh5a>>cq~$hNJo
z=Y6}~Pcb#P>w~bBEm}<BXH`Ke$xnVvt$EJB59iIMp2fOraH^^2)VU}G+?@(cuE#-q
zJqlir_}Dt+w|5s}tbC^6p+c&f4(cI8?`}23m9GCrI3@7-{Pt+HJ8#sn_0AO=23(Bh
zPm8X@pvKPcvC?0Nayz6O4XlN$%9gQVDtDX}pT^z`!#)iv&cJ4Be*9r1#)__X3^PIX
ztmU+T`;D7KRnuhtp4Vg<LMU6x<ikaf4^N$TZO(ocZgSc1hS+ajyxxsMN2}ZwbYzPv
z>OkC~ko1D_l3cUo5`LyfYBF5=B+N2_BmaC~wGNkN=<nc1BFb3qMhyOE)*c!W3CBDJ
z!Vj{=03*9)h;Q3!%+#{S2q)?0Ehul=1nPT&w0+7z>uuXFvlOF2gqUnn_=r6*-17W4
zO%Y~RMRP3?Sw^!ovcJWj9krM0dP3_T(rWi%V-2&$pXbYwRjI<`><!<4XiiLrk)bp<
z?p1TL6xff8Vf)3rNQ_84c7L#%6z6LgXe=T3X;p9x9lgj2g!LR$+kYvUQ2fX1fsSpe
zmG|cAoZU2GzmPPer&>R?j~72NvzeeeXiCp}7%N@dz(2$WeW<H#|9ATQC=Rw-o$Vs>
zZ?xK9SCE%6vbyW9X@59*^-dD#D~Dp$ZfuY@%Mg~mJ)Jk`+!n71PwSm=vl%;B5rn|P
zL~N0-1y2I*mZvJX_rruuzDPTKwYXo3h(<2~ZEJmklW<u%yRx^`jR-2WHs7LHU$Ps&
zVvitGY<Y5oqD3LptpEPHq=Qx<D^N*2qHwHbmMbXk@!2LCY?CvgjBY=HjPe7^qj}3~
zl8oP95Y#)Klu($6Aig;GQ~xnOOO^@<LO~R?;qdVH)$_QkmzVcI7^IaVK<WuENJPnV
zk3xL!2mP50BdC~6+k}zSbn@sgmU)rEgA)fxz4owa&J{{Lb0gqyuvy(3vV0hq0|%g!
zn`{v>X|^=%)*4GH#R~qymm;<w)EW~$bIiQ0_?1%Y0Y9(OlZK5I&wEiGwlFNMf$XUW
zSO108`0@$>+j@_dR1%~Z9JZgZH(_lc=W-ggj$b!A^dlRQW--Tv0(k1fc$p&pMtPdQ
zBo#<BMXl0=N`ll6=KiU>0B2Fn<lUdnlT|g~BT&&UmvmCeUfkwP#VIqp?p#rNqK1C<
z@Kl2J_h$8NXl)UJ(!xzvD+)?Q|G?NdnBOoroqXfq!ht?FEUuj~`2*lT@%u@-TYIX=
z06(|l`h(b2$9}+wy=i}L$X^FZ(EX$RakpAKTa_L(XKXEJ6qE&2tjR7Ye5N`ddKrNP
z6(xG_d^aPSDDe)5zphHIw&BulHLiKK$gwy&ur%PN54klZHDa!7s;;b53LpFTVBik)
z=yEBm7}?!|A6&_0HJp@Dm#eZ`=>;K`uFx)<dv)CFZAzum(vBPf?C<@p?&E&Hd2C%h
z=k-zZu5Y@<&HDK!W6zn!bn+QE!N*lLC3s>WxX$%<Q;TBcT3Xlg18DHRfo6?~zBvk7
zLls5IS3z);QGpT1T&mv<O`uP;TwI`|$8nrg>P4gR+bdD1S9}${+feaSsyWG>7epji
zo}fSgsLL(c{DCwoL<cKOO-HpCHMR|gzq?;&5d^33$`zG&XL15G_>C0Ce7@f3(C>bY
z)bjW)Y_{Pc5&S|S<mhWFRFK@c7dTS|=ny2>=x%<<v0#GGjV$0ZKo_ZM57dkWrkXF_
z`-boqs-3k?T1U3AX(pet-!whE-GN=rG9sxN$%&4~jIl~(4DZNfMV2+jLr`z852hy~
zC&PQ@RFb?aO;Z1c;Ri7&(dI}UJ9}B*iiFdzq!_|mQS4j-(<p&NpK$+t1jXxnkh8XS
zP4#M0dq%Jaz3#R!h+_id#TOCAn`B$Vkv)@##0x2;nb=M2(Jbm~6c~4=M{_{HNFraq
z|1`gpl_P4pZ-*tdDz$AQo7|#NSL2Q<(<$2@J=_?bn^o|<SY7UBt<nCY5q^}qqx4&*
zIMo!d9|jX<oBYOi$mtQMv;l|B`Be#y68`!fKZULSFxfLIujT)tcJ#6QO_WiDHEjtJ
z{X>y$8wH;;CwV^Q?lEiV&zD{Q7S-b;M$C8X6?6&>k~S(+kc+0`W}D8g80a4g1sAr@
zMV$A}t(P|Fp9$AIN)y*XPC>~ym3FAJsSq*E2z$#*KMXvIY_Qx2=V`r#2aU{D8xvf+
z2wk?Q4i@EQ^1Mu*s`g0>hf(V#%z$hO&xqG>fgPS6G?DmQ!#=r$tvnIBh6}$lvIs-c
z<Uwu<ydNw0(t}&3GFWwREd&!CE=M3rIUl4F+kU7?4Js!IacZJEc5Zz^J_k^^J0UHE
zcH*+yY&(_+QyV~Ky#Hy%M#H0kbpfuUxCvu6uv)Y^#$itdpp&Z+UbYBjk<Ylk!x6*Z
zfh$06s91-D`CXu`_SR5m5x-mAl44JNlh$vDBm^7b!-Hn8PG+Bykr5l+_qUtQMfvQS
z$wR<H_oa<g97C$mZzz<mTAq=tJKXc06NA-Z$G5W32-UF0;xo$hiPdX3>X9j7*mA~*
zq9c+fai;zC*l5CF@)y9u#L`PC8!b6On<JZq99jX+hC4K8)-iCWU}far7jPdEf#7TP
zu*bEE{xu=2Ws+D}z>ISZiN}akzXm}VAKF-Ihzrx5S46m7LZi)k(A^aAT`{DpXSHZ>
z8w9z5r{AbnTh@K9D1JkR(`j-hqfo(7a3x6YTA944n$&npA}3Mu)W67v86ekSKPBjj
zo>kz1|2e%T_}e@Yzre(|ZcVD*ifgv6_50A=(vYh5da#`N{8$%kPXc+4koE>%<Rsmr
zRdPxwDPA%KR5&tPPp&>O&2!2793vLM#;9BM79~x!z|+SOgN57WMN*FXr}YN){R!S~
zZKk=yAKEK|*g=J9PZ69K#7?$l-`>zUfu0Qh3D-@j3HClf2wFw77u^ZQUg!LC2asj|
z<JSUz>e$X8`17ix<fvOpfuh0E5d$YdR1$khGJ24%a<!zaLi}d7wJni?H2ncRmcg$w
zbNp_lF7eJRV#88>5H_A~PK_ktw*5@XtBIP~Hu0yl)V-k_pR~wdGBSdD4A6QbNLxbN
zyKpLh^x_%jyCPbSfS=uxOgDlrm?JJ`Q|%!`pXXkcd!twqwL!$Nv2dM^tK*S1Hf0S?
z2ihM&xKYPY{Ft}|M{wc?%s~02NE~#8y=%qq{YfG<g2yHuWUH4(%9rD1%Ky{FIW38z
z09m$d+qP}nwr$(CdCInJ+qP}nRnr~OFL%BoH`Y#vLm)@=;ZL}UTQs!LkUQXYlz&KG
zxEG=oz}_%8i8AcliSrw;vQ$*yne`e(Z$8$C+(Rc;RK3B)9z$kB6k%z|>BGVMX%~%l
zOf{PiQMum2GpJmq*Xpp>*KHPix)!IeYNnG{B}H}G9-fBK_!^DXSvs{`*8<JRL-^zC
zKu?R)C1S;1G=Rv@tRsuWD${n<jRnh3vnY`Df8VJuhUgwenwJnx{Lm;?aOYna{@f1@
z7j0LA95~K2No`qCa31CSs<w?w*9?Pa2$M*0(OH`9_N`wA2eF$u7!SXOrv_>@zZ7B;
z+%4IS*MM||RZ})|Jw1vcz{v7Kg_`$Uog$EE<nJU<yflsc9$1@4hE0wF?w$DpqT=*-
z2Jq3BF(4#*nsCQ6snn!Nd?wpp?Y@k7*46BGJ%Bz+4z}<%!({A06XEHS8Ls@6bmKTz
zUD;dL@1_LUw%&`rZK|iRJ#S$TQNf{XPrnd#Op182zu|o_^Wn8v!sHp9f#|WNfb`H{
z*44~3;KtO7Nu<PHDB_sU01007*bSZ=E`%?dR#|@M5SKLYSO2o!5L#0=yIFaO;%q4*
z1`${dt^*uiCbisvT*lll0VC{BzU(8j`8v{$nem9en92fgAx^#`cmNg`Y+tR>8q!u!
z5#W>So*}h$pH5wek}4i*6~YH-SHzqCwZVxU&lfWGF%m6n#Kzl9cqB<kkb<24`I}q0
zBv=Ufpj(da_m-LYjkw*QjI(V^&C=rs>syGs6`*!&JV+{6iIGUC%={WZ4oz_u?^)?Z
zrYqArlqt_y0WQU%;9N%s89f|Ax}qoPr{#+Ij;i=%Nbpe=9Oa+B0(+REUviUTp#so`
zIGau(f9T)Y{gx>-4ox!r?6v8kUUk=p<xuGH?i=aAofQl@HeC#1%c3ZYF1+^Nv(!xH
z&cLxc2Fkq^axq8SKX7FaL@*H+EgSrUnzODpwmMQ*V8eCc%>V%jmCd@<iz~O5L0t->
zc_mK8JFgt7J9ZbRX0Y;au?oB4Ng8}hI2ZHxREh0Hm!q7y5019XY5fOT#vI>9(PB+h
zC9M#rFj=Ak$pa<u{guz(bj)4uy?+sC)(9sW77B|~qq?!(#C)GUap*Kh=^e@@It+qQ
z_nAbNA1pcqhsi$pJc%MW%D``7?Bx8=va7mZd0Ok@wGf|e6?*;hef~!7!q%<sey9<V
z&7lFlVbLz--vA&MBT8w=O$IXUe(h$QD3@i64Le+IX|>bw7pC0CxXbI9*lh(UAGxN0
zGrdJijh$+AuifhU0QV$l0`T_kI#Q>AT}*Zu7?&kc>^obt7wRm{@;@TS=r(K!7WObT
z`2Ox?6)bBjH$s?mJ#^^I(s?<3*9C-o*Dfb>$X5fvh(y01F}K6H?sEMtKj1N@q6~LR
zCdg)Z0OP8{tz8&e%D0-o#<#naCpJvJDx|3UJSLBlwOb%u;JsBQ5ig8->>F1OFbbc)
z`^PO@MaPD*@eg#4Nh>|ug;*gom|gKEXkb}0LepV1?N+oypHwY4s9saPhl38i{JEU6
z7LXcKXeCSLq2$wVoMjIUx&~l@mi0nCcVr(%7u8OowE=h6Q@dLcR@aAD9i|CLe>_rk
zm86iuXk<5_B-CKdY;o75d%aqeT}EYt=p_p!l?n(>lfy`BNqKn_e<lIB!?~V1&;Mx+
zFZkaENbiz_FZQeaH_u^=&f&qEQ`!Tq0~vck1*h22bF0MIs!XASUm=5;E|5qaLE|w)
zBx7(z54JbN-If<E2Cb$fipx*So80%~4=U~!Cq$J-S?S37+;)ic5{S158<div0H(u*
zP8zH~-&l0nHOBbOzDx21A@1xVJkYSLm3z5lDM?^vszmHc(X{25HFoS=q+~)+<sZn5
zi$8Mr^PP@gH7$Hl6{H4^RCSR>>JT7b%ANsQn{nETwf1TsLK@lIJ5I>a7z1VYM%Fle
zuAye!tVR_;{oKj;SNsb{10)}QMuPmSa+k!T^VQb<qG6GE|I`E)vboRuuS>+oU!UW6
z@&u9Or!fF`XL)O7nPSy%0I}p}73_dtX+~?p*Y3Ljf^C`$Zr=Vk=N|HW^tWdM$HL^z
zbt{bg0I!yyim^1=CQ|<x<y})XfwK0OyHZA9t@@g^c)+O45mn}cOw}SujLsYNH<@v&
zi$(J5Prr*K{1@M|Fq&S3@`y@cv1>7^&GGcxZ7rdn2;q`;oFXsh9+c~&^Bu;}2rcPs
z8WQY%!UyOP!#t9#E?VOHK+QhW*UOu8Zr3X{1=~T25YfOAeMupE-S2b}lU}ZZ4zeqj
z85Oj^!smwHMTjyduwP1bq{zi#8jfSDRj$5zDWWpI2ojvu!`5^ydBvp4s5>(K^2k&$
zAi2KNVUUc0mO!C`UM^!bWGWl9<jdkoR}MV!o^I?;+P83+&l>M0Y-~$Z%h~lYf8qAO
zybozprDu%G72_5DVi0UNGV*34?AnKpz9TWKlDsoBQ>UEUK+mz!pyn$Z$O7*dmvHtZ
zK^hrn5W_tfFe3})33v_P`9j*@e*C0<UWR&O<14na7iv$9rW{%CgACJ(5beH&skcDk
z%n^=+Kc;yK9_lRQw2;(|j}2Y-H&AaGa2A5HQrb+8+%Mxw&anC0)^ZKH6J%4oWC-u6
zEkZ1dZFgjFos0H(+YWR+02Dn+D4Ro=t?n_S$4)NCCZEf49ZEzad-Qo)m%|}AU=*Q)
z2vj!~;7KI7p0y-rNK-B(_0JgT51|d=xjT{7;a}Ra5Up_LABef!9gYsR81U+*sbtkl
zKgsIeo5VbM#P%Cvs!TA1ymJ7$NXb8#&zY<h-5%+u^W(RLsL#T<U{nC&^U>6(uNZ0$
z0Ir*M7A|#h=_#0#V-nVx=ZL5S0wZItQ2N~^Z>UeF4r^$>(G}aNyOhbfZv~65s!+{m
z&_h-maX!>IO0`f}h+%~s{0#+Ce=Yg$ZK77N^3!i+(XU+I3Q43qcsdh?G*VIV`AY@;
z_;+Y>?5bnJVFvc7Szn%cQFQp!RZM)Q5Ow`lL-b>ko5>c>7&<Z#mOY{z9Oc_)7a^H>
z>;k<p5gXBE`$T2SabkHe(DO9dH`sKP!36mxcZ#mwH)R5i)*HmEC?X1LLp~t;<GKG!
z+^@9G+;UeLs<c{p48a&X&V9|<aMJqmGA0lPP8n4@`w0VVGHLjZo`%3S7lx;o(a&8s
zfF~ya5oY1a@;XQ%c*ROGr(XK#oow_81;a)8?p35ILNj#v|FY9xcw1<t!}ej~qEWQR
ziJEtT8d7*6F-GJ2E`$elz&N65-3yVaO5W0UW`7BLWR;)hbq0vWIK-BxH7MTj*9;Fd
z1=+i3)cvy*IC!{uQ6?#<>_;fB3=}A`AhY|9_Y69j5VG*XS#u8OtJ+mNp@_Isp8QNM
z7p&gfjAc-IM0~a*`{1mwElljc#<GJ=6FXa@dF-9I%+cr5V4u!-hr`7bCJRQ{da}JH
zvKOovOVZMT_|Ni-q_1XglLe7Nr{&*&ak@1nXuaIAu$GmREy&wY_UarU;($>dx>h#b
zBA{QlwAK&=stWP@KPp_W99-r94cG0c%-y~0RHbf+`usUr#|BCS-6Su|>=U>$$e$~6
zkGA~^zkM4ld4-gMuv|jC^KQp}Iri5obX<@+14eUYjSlEgCmbfmz)S(zS<0q2nN7%q
z$9?DWjO1wG3pO4G*~uAGRrIX49G$Ejs3g1D{sx{g<9y0NkcubEw3Jcu3ZHZ&PvU?C
z)u#&DiXKZwuL668E<52mW!!gnc13F)Iy18&nf+cjfpJj{w#*sT>cc=7n9w#TH_=D1
z>16<GX}JFY)Q>a`jIqA9?1rZe&V#}@Wb1R^)?d`!49L8~%h)Oljw{ThY*b6Gsl@4%
zejktu;Zg2wd1^?uv3HRB1JOy`duKa;+UO_?y?@p(CY?(`Er5?89&eMXRSFhV!P$Jp
zDm*<SMLtKDlNMf1%+0q=8>T$k9fm?~gIv2vVsvpEmg`}Y(3OAp<*y9OOu1SiwW<Ll
zT@->NV6Qo|THOr(st7ZkpZ>J?_74xN`7XkDU8USer0O8-x8@8y8q>G~mywZRo~h_q
zi*xG()yO8rr>SaD8S7ldrK7r>aM+vnsE?UxgIEcNDvdbSj18hUm2gkhx+vryhkgA?
zDQ2{Af?P|ta~^LKUye;k@m1dmr#8jUccm<tU6yg-YkGG4^OW=EpaZl1zuAG=hja4U
zh!SRGJ_~1o++Nq_s6GB}B}))<0yqJNL?Rb%eU%RIl{nzpHOjCYN;{3(FT5vPhhK^x
z_9&^YT1gq6Ew*ybb#rV|Vo(5pb?olIC%rQ_Kh5&<n2(LgXu+5K>dH;NX;-}=qk7GB
zMwXmTF*e>!6#OVjv+T9=ojUt(qT!j5&->3@{|=EU8uS$#^8}}BWOsWZ&>^GsnE>xQ
z6u5`!y>G78!~wQbv$2^S1W37@(Ahe*(@r>P2Xa{I?MiN=hgbvHPeC>oHyFtOZuh<T
z*PO$GvPt`GB0fw|psHkYij$Md9K{0EE$-FfHOb5|%exMXr~KaVJ*tgS`Eg{4`XBh?
zEiKq2h7gE87#sI|DSyJ<V^ZOwS)f>c!c&IQ)G>NeK~H1%?IU&e*IA}NcX#!hjqep)
zNjdNo(LIlT90nLE5W8P8I&uY7oSS=_5aoSraAIn3xQhl7pe8Q}F*!(AN|k*0w@9?X
z)HstwvgZhIOPW=K3k7dl0=o(;zdX28(({$_8|u<^_0^54JfS?rzJhZ0>^$5WkzqM*
zNCavcFg8WO;j!BL!1k$ZJFLlIw|f)ptE#M`qki+*VOj?TXPIw?#G%S0KNa7a2pQF=
z{P7u^_#Xl?3}`GkTAq6$a_~OadQ%Lp=NiEpxCbg4f8GT<>xD9+lX4V08rp~>h~8^Z
zUi$LY*(s6z(L4jWBf`WCsRlm23lRL|QqE{i$_F(&2{~>*3XSd^<uMOw+IK<;LD;l1
zW6BCD#GSjbDtU=~a^`N+%kDNK<5J&T-hz9~po@>!pf&JhDBTua2Ao<@SKO@D3&RU}
zA#xl8wPNo_6I4MKq^nSh8MJv#jH7K+sj`^)K?%NBd>g>`SI#IVJy0PF`$j-<E9^V7
zRDG<~avV2c!SPpzv!LlW`?(dEZPW_L8A<!JkXD?Wu=P{yy$6}vCBCna7tSPB>ROfn
zw?%;S!%saOF43f_Py_rO$$ztgz(8$5v-ncV<IPgNWWXo|!%dE_A^Xe;5JjB$Y}iFk
z3@>Nc04Rf@L&zzAIu65mM@0YmGa}4!NACA!R7J7#A0Mr^gsqU=ZgZLH5DYrqOD^o0
z&5SJjL-a<@Se~PLg{>6^?0&#cgd#7hE+Fac{h@FkQFg=V43ncN9QgW{(<_P8f(85@
zYHkB7hMIBFCQ2X?RcTj>`+|2})Tc5)3D?Hb6TL2a$z$86)f`cr)B&-?fcQf&?H@o1
z1EPm_^K(Pm*(ICRSmD=aWziJh8(1ZYeqPKK?i8EFKizQfDReIsIryIlC7hrtY-t`d
z5xUPF1!%1ipIUW}hKDFBOfQ=dCjMK@*iZ1BeM;S?a!3K&OEl^~qHGa*FbyQ9Si|kY
zCGc?>dI3?87j*l?4!@o2!7d@E@l|7zl*h^Wwj=}O6OvXdeX3XVzOq~J;ka1A$P3J5
z;d}v(*Rm*m*ia-RLlV#oTs-jQ64p_M+5^rKAVS{D>TRn-%`C~Lf2MX9{w8nc%6_XL
z`>*l?g9n?Hr(ww9VehU$&#86XfP3H(moKi<Ub;CEnLmHo!4<JQx!C#HOTP3{gVqP~
zP0BB*Xqo19Y{8X42liXlC=I!!kfMun>t|?4x&~D$!V>$@C;fDo8<~mziwVV<Vp}}W
z>1A=axv`kJS`_*q;-N(qOWw7spsT|LtY-2HS@>s@)tu0TWVh{0rsKn>LSX0$%*hP;
z_|JmbrR!{{>FrP`oSF)*mhMaS|KlTvNErdYe}+mlHG8;ECLMVEGGrw<Qa*=OWQWiX
zDi+6RU;48Eg`zS~ra^Z={1$@q8BvHa4rA0^Ii}vOoUL>(Q!6OPFMc3s;wT3fOC`_B
z-$(4oZ(hTmjlV<Ct8!iIb?m`m{i5RkJ)}Xxoe*mT{k;xYP88jm40&vQCF)DM1G#Eg
zSa^Q7EpcIdY-lOEyoB1;(=yxM6)248A#>cnZ8w$#qiE9o?w)epe%x}OyTVq4mtfC=
zQx8LhTOdS3b)vEMVJSgln|JBjlov8-#<YC3ryVm|`7PjfhJXTE4J__U9Vbl~<Psuk
zD^|Wi<dKevL^@bVkEo7w!m(6uWQ2736z|&^?IruMe<1p$m^U=<HxCby#RX>XaOMPO
z1X{0VN>n(27=00HwY&ZaiA)5^YVOy(MkH$c+om8?ukAZ?dYJyQYGW59TxbT9ER#ud
z@_Jwn+QL{u^ClsX`<R0Di&`<6lsXt!&qsRsv~yjdDql+x3a;Nrw12*C;ni)^jgjr?
zu{v&Pf9#9t6HR9$Xu^PlY;O1g==G#)<XVl(IApI3q0iE$2bWDSiX3q_^jgvnoLPer
z;rf=HSCP^MM}kCCyi<AK1y)c9cCEeX>eknX$SRMt%3xp8VBdx}w1un1WQ4&;{2~QO
zao|)@^(7Lu9fC=*o7nUWVd!+k3M~Uvh#2lnTdiYrWi5v3AHgMe(Zj-N64F=@Lp@Gh
zVY{ALWTPQcO+j&R{Q)q(CM0#ig}c#Pt`iggj0%W{$vx|oxreLnxSpQZjrgo=_Fap`
z?5<g!a4oc$=*v=9H=O#33&hGvE{Ty2GNOZ4bLu@D4e*rUH~thh9{bg0B|oa7{B0$u
zl91WZZEXD<Z87q(kV@e%MTRM6HjPYB`A{5`1=g+s!b-9+{z7*_j#m$X{+Qq5O|7tV
zbBcDgq!(Ap(Q|oR96VCdpu~X-sK@#|K(WKUOD6l56o4;JC&QI7pRtyl;_Y62!a2QW
z1VmB{b5E+#<epP+rq`c+fGEp!$4Gv|Pt62I15~&|xR(j_Lwg@p$Je*X=_U`JpmZOj
zHoeXyV}8FdSJ&r@){``Z4{JGMvDj#P3>cUFchA{ezDh?QqpZ>^4u+PbvJEU}V6NWE
z7{0Tr8wz!;oyiST=Was-cy7$RB`htK>btQXcmj0difa$6l^iI?++3a!gWozTt5%&t
zaUSTrb;;yJG3Ul*^oe^Wl}0*;1c?@B(O=qfcela1h*@eUg-?X~!f1*mjrXhz&QANR
zvw(_Ul>EGf_h$We6g|rv7#uL*YTeaDwU-Q)vSCKz+AG5*KnjM*kN?{Pa|CGmO1i*`
zqjDbK|LxUTg<OSeYi<zWHYh<w!Iq*PBM`H1o>OPPl#bFWwJxLVP4AUbTzB@5nD`&R
zKlJX@uv%s{DKwaP!W^9L-oW7jSn`lK&<E?G-3pY?o&5CCTxo8PxqO-4;z&ep_TcW!
zaLW=^g+#4$=n&r>JTgt&MD_^+GyMYG+cu-2yk0%w6>3|d&3>&LW6OEewm3!svn8C7
zrMR}9v^jj`%)vOxMc8GO#b1z5E*>jjeP1%gQ#fX%5fX|sp(5+yC}L0~TgQvLkLamu
z2wh6>wKh|-=a{6l2yg{@wM<}K0q}!gj279^rJGJzVuDN4ZZ7&E&2`&W(Y99#As4Hl
z-ubTWtaQyVXhQ~K(P64mV@8kJ?3Se@|9@{o0FBxd3<|?7wCWS42_&YQ6VKkxI)Nl#
zaNsFpgTCy)nS3a`EhfBeEM)hCti!wm-wV{#U%Ko#G?j(HO+8!vH+blmNXdC-UnMzR
zg#|`;w-5U$P}IwRPV+m60ix>7_P5h@cxpGU@ypJ+u0xMIQ8g~I2kUv2;Ig01kW2`1
ztJ$0M1bsTuEF&T`LcU|eyuannpYlwCzsOMQJ&of&HXmURE9yo+{4XW-Kq-tdmpww{
zA3=lnSHMTdG$aP?|4epF-?g<?Q))><A@5FbyH)&5i*!kSfv13p>_ym4T+kKgzR$&6
z&Fhnv<a(SZjwIT-@E6XlIu0*4Z#*W=N3_<rGLj==z=iNH&5W?raEYU`fWYBV^^20n
zo~Qbve8W#F!}kNKAum>psI}9AEZ{4{Uyv_&TU`XnKh69Wgib!BUY!!^F5CJr7aYNc
z^Z#AT-xh(L;lx0Bl2xh+m)$4kgKg81tA9#vidXTzh-z!N%fsZ{G5;61wy%0+&vxqh
zEz1K>1tT`#_o_3$&z_tZ7e_QL3_Q9-fl9o`;LRqT7<$hpVF#6nd9Wr6LQ4yssPF4l
z5PAdFcRxRgdLJ8mbt;*ms9kyRcBafX4)$E=jH{>F7jpH-B<-8flu9f43~J0=Lvz7H
zD8QPy*7}5J^F9zQ$3sG=v@Ru_#`hHco~WZ`=Du!xq0;EJ)RXCvx!vvB<UrD})foS(
zNhLh>Kyvy91mHVw<j-4Kf3Z&TWRq`~zPU;1nyhg9PK81A!)MlAz`WpL^O>#fg9k9@
z{3|2JeIlI|#UHwKPUQfFb>}x`6|y$GIrRX|op-p(cyynty^Mc^8t=ecP|gAOhTxD;
zZf<)G_1sClz#D9~Zz9?xD!BNgo=$jAwTjRzFbYcpT%XG3JVBhQao8)0<<9t52IrA1
z0l$f(?Ma)TE*6iaxh+u$)J11QeHtIL>oh?tAAU?V5Ks9MbQ8^Ey|+1XXhTT6)troS
zV3<UM)(Wc3<PEVqBsEa@^z#VwPG4xtSkq71P4^f92NrAi*tR5U&i-%}c^F*N)F({$
zY`>50uRW@Y9Ol%m9Jl#^6owLmXf&^e2h1*G1=d?``i_Ruafhmo0w9h!h*L%mQJR9b
zDLvb_41|R8*nMM9wc>v_*qy9(sQ5J-_{h;fZ2{w$04ju24+`+0!8}3Gh%Q7gtR$!Q
zB!V0IU%e4bY`lUgw=k&#otI*%vdDw*?jfG+DvVP--Vz;(TX!VEZbPhAc2tb^u{qP|
zAK%va$kK*D+`_kUQ^lHA_@rZQudr(clXH%RZXGDV)1hFMNe~sh#Yf{`MQp;@H|tpS
z-rh?P96MJxv##8KjJ}wOZF7!ovGa~1;5`f)i0uN(0TomPwHqUsJKIUuZ_f|ya14^>
zqpSYCuBo2Cl2Z3Rl9#<27;li9QMR}d8mI*hGjPREKi<~fJvKK#d}MeaYR(z53W1{X
zwTgN^>S<Z5?ho+X)_uQo4uo$b_N3NV=)=015GP6jrh9*Xh72Hij53>G$+Ce~1UCjD
z&1#u2Y~6}d%i}Z_Vy%Z7r9@ROYW-qKtzVe9y*h8R!|cHj@w~$Q{Ov3|ZXZn93$~<w
zQ;v3sXDnQjzj-k+C618rfR~HL=+ZNamwu~3Cklxc-AUf^Ltf4`l(}<@)@IX^S)SIM
zt8(R3b5X-jL+CDYv$;Z2H1RoatJ7giMif*W8KkYUQaQpK+}CJX5nDf?^Urb-AkWr{
z3Mf@OEhag)qP=5iHP}-SJjwF(H*u$eY}Imo3}G@^iT{d*pihC;y~{DU?KGJaeoC`T
zD*GWWAGRti!{=pV_Pf&!v@48^8&OToigr)9nGnY`8PT6(hdg{pbkNx2*r!-ZN8CkZ
zGK(NDfmNZUR45}mfQ`+8;)|0Q&S&F-!BVCP#sG;z6`GeN|0GghaaBn2syXMkETq*Y
z-~~+^4q$BaerfGBfuPm1XB)DyGh)nSw(!bvvUnDthdP*S755X<%D;9Z3U**5eB(Ck
zFu<i9aO#;OvaRg<g{<NvFEfk0&A@3Pp+Ls?!aJU7?A%EE9)ad`QaO_S_qsOg!|yAk
z{%}Y(+@d-VIj7x%O}5t9u_QBV5eAQ>aI61cbWSSPBmZmI>)!zdD$ro>;JnvRoGM0#
zO!@Y-B=oRq{!#Dj!kWO1)#w1rq;@A}tj~yqxth0+-PUhrL7;O&d)ZZRSdytz2S=yG
zC=6_*aQ*km7X9x1<XD<0G5FaFe4FIKQOBnwJKl+|dlLsgow;Iq4x8QFVxUHeqI$R?
zYw1TXr_aE4;YxH8V0pLaj603mw?vH=3a?8M3=WQ<3Ngt(5~2#|Nwo2W%)yaWXeHx7
zmv_L1=wN)JO}N5-@mpJ#Wd8%Vq3@rs(6uTE$5}Pv9Ydc*S>x{Yf6=AeqaeYXbDdd2
z%z6X#24Vm06{cSv7F_4|Zhf+v1T+*2I#?Gbc!wBLNm#xTJhdpNB(}5nM`?{vpM<c}
zK-sCj6??1RYsQ{zNQMXKeZHUC<xuAL460(35+rRp#?(5kP6ibiE@uYT&pq00bK-~Z
zpQh8LQFYS%@F0o%{qdH(^wjH}>qODE=f86X6O0)@v{Zx?sGPeK_+K>DChi~#hM~p-
zEF+1r&AlZ=4n075N`F9}<wWVIk0STPhW<-0DI6T7r8Qx`CmUI5iLdBPa(TKT`u@5>
zp9{G2p%yO8Jgw<}TD|{pKwrqKtO{%l>E;OR`gyj?^Aslo2s=eU9~M>;?4oBjfbV@8
z1%U4Hpaje-U0}hL#5~d5>|YC_N^6VS5!(Tn#=y+At`b?vj8U)Vx;BXhSa|2QU)<&U
z;)*K}er)&<_QcB0BY|KgoU?q?gQFk2nH^${Gbd2_4x0eZCqzQxN)xa)li%3)PMqvl
zzA)}?o~GKrzPz%Z!X5(NHt1dghyIM<H+K4bL|AeAM3_%iQcJX`)wMbFfy0RwOXz=m
zJ~iR+ZKDy7jdA9Z9qMj@OM3U-5hwIjMd~3RucSW=Vpto6r(z0iNeiKdS+!%Y8vJ*X
z0@}e_U6VI1N`7lBy75@qSTzl@d`O|5>YrMtet1p2{YUsYK7R>PH8(TRT|&B<u!EV{
z+2`*rlRWF)r1OmW{x|$}w1f>u4T9I(m^ZUblh+O&JjB;n;12`h*;PgT3ONSD25qE^
z2aFWni5h*D`*iIg=)IcF?WWKT6|2Z4>;HH^b9_}(e+LKIsXL9lOq0W*2eKn=kw+Qw
z80k<5scIv5E_`x28vxTvWWe3{X7|=gKbbKIc=cjK7ogn-sNU)=rc=SK(O4H1tOAlx
z*2Urv=RM*EX|`n}d7@E-1;{nvp&WDRnHo>cj?M%pw=NQnF1jUuvtJyn=c+)RCHp=U
z%q0p5G*E;R(}<+wp_ty6=K0}rn%X%#zZ{p?TN+DsSn^$zKNFR-LGb@aMd85Y8PigB
zdXaaB)v$D$N$NJx-Xy&}2>YEtbu;eLJkChWMtz!`)HiMP1Id^;Z1j|V8zrqude`LQ
zFPXLho{ezWVYDUR09in$zs1U%GNv@G^!ts(@nlWXaxX1qM+QJ}{0$Y_iJ?Zb6=YPN
zKb2C#sX@FzRcofcpR%Ra8;+%ZA&|L{PZWh3bSx1<AsrmIo&#uiJ8#p}BX(7iYjxyK
z0!vCdQN1DV+MbaMjVOpv_LQ`*2n}sOIvtRng&x;|6#a}>vR$s3LBI|M>psuo8QN*&
zVLvLBF4WG{n-hyTmU0UB;HO5f%XaqR!h=^t;wKzdHk=z%LH0vcejkX!XDZraHp#gJ
zAiAn@+dTT!%^4$iX7x?|?zb=W-0dyAP~|Y~6mbAd`$3blE8I#0mI)Ac<9;4X^C81S
z`udE>1fci!QOCi{@c~YU?CS09b^dCYo2m*J;)S#Xso2T;UNQ4_N1qA_TAqyAK!pD$
z{=^~pH|Om)4|2`>s3b79t!3ZrPUl~XYlKfcr73BbW;BFFl(i~w!rh1q1Sno8dJ7@k
z{3x`0L~%i6!=R>vN$xs8#Yb3%Jzy!y4mGdhG{mFCpS($3UxGS4@0{5S>>P2;CM_D_
zPOvB~ca*n_k7%|2Y?JG5!o{#%Y;MdM7!-@0AE_-IMR=s)WN2Q}38BIpg!apI6XEy=
z70p{^J|my#Q0YTM6Itx4{)Rr`FU4!ME+5i0r5@M}1&)ww-x>Faq4^i)i;4!xt8&7!
zWK(Yijt@*9LYNU=o&kD#tP)z9s2Yi&HH&I879icmWT59_al836;Or2ZssXS#vI!i!
zDnALH{6?GQGGzha`$k!jCEJbHz{zbmLLj#ev-cdH;we>w20&GR)!DZVginsE)Iey<
zGEEu-5{pF-L+?_n@{(SKF7%_o>|lVS`z!5lgc>SLe0FI_mwBItDUs3ISDcf!blyzV
z*@@-B{7QSvlH?NuFL7&)`{M`;^HpWy(h=+yngcQ5*>pF2E6)Ga8LsDS%0eX*PT4co
z-mHb+XA7X0?~a=k?GY0oNt_RJ2+e=H)&skJC_ipWU;y{*=4SzLekl<+6NMiB8I<?K
zzBjlEM`tc@KY@*mxa7_i6n!c_vFv#D_33Fh=7~s8O+6Y2Q5SRUDs}s>3<t^<Zd1Cj
z3E)IB>gy|RKze;oxB)6S7G?(oG+T$G>Q#uy=^pWJ$g}*!n!nmK8SNQRmqM)KB8$k}
zgh#B$Bq|E{+4Gr7%JvL|N9TYw*7XTpQ6b;W;{=6ZwoB#Q7!SbC39Q7St4AWCfP1pv
zbcd(Nj9WE+-}OiZ)cJtv$H;6Lb9%{Kn{uZc<%sk%5nBD-(R{<OQjyA@dSeGaMQ|PQ
z(8)?gY(q5ca=x2_FquP|Fhr405u<T!<%Ty--S{Cvs_EHLiSagC+JrkG#{#@ZN5WKc
zfBG3WV(?14T4E^Vr;%SKtm;lGPUD5Z%asDR5wSf#*3cCR*r8f}tVgch86bY?{}7CH
zX^??J0fU>rqs;RiaGg>qTmC_5@}S9QZ0!tlUr0l`8^lIBR8cN8IJ6r3Iupq!5$U#&
zv>WS!Et%G@h_r2nmJt5mvdMZ1fi|_9la$3bw?>a&1R<e<>iuSY12hHt$|#?AKH231
zXnT|RB@!IXiE+u7A)r2GtW<GhY^khq$~$Zy6K<f~DW1tcj^lDGd$nRixh<g9Uhn7+
z4gp{AntM)P%G5aFF+8EHJP}X%I9#fOLml+I>@x;&RSHXFA7-y9Ao>lrlBKh4Td(XA
zuF#WnXc2#1#dx#nW2I(zC0&p;UVtqn0daLeHg_~gpUq50Vq*!*3{uOB+O73u%2C5T
zq@YeRk%JzGfg|1G3;2yuc-$OB>m|Wo+IEmZOub*Rj54riFwEf*NgCw&9aw1<0NlW9
ztT%y+r!3Sx@0fj$saK4opi@XQAZV-=+Tf4aZ&Ui(=&@ig7A4b>Xbo7HN=@;_D0=wm
znUBI#{H$deV6zalHvp+b%AL_1w0(WDx}a^ZmQvuY2zTWV4&B!TDCC^SiMY}zqc8sW
z>MNwp$91JPBO}_y*V=-D_2v<N>nud0bp*@Hb6fGPV3mg;!44oz5UnOgJ4+Sv4zo?c
z8G4Q0ZK5l$;_X?Yv=ogjBE=31H9=*L`pCX>JG<93bg;3XxjKjz4!}H{>O-hhTd^ZI
zU-*%5OW>Cly4g9Go_~dZEJ-9nA!l?yW3-Z;`uHqRvHVoJ9^oA8#|N*gvKV|IrPR*A
zr|KHG1dl~6sHHFFm%;bxc10qgU3U83S+V>B(p8xd4vCySy-zC-MA%$eGfF=g&$^VQ
zP4eX~wWh00FNsy+T(wB3g>h;oQFi6A58L9yCX)wNlx*8o1#J9x<eA+){7=qzQ^>3f
zz<dI*i&DVfwNbER*GX-93GU7+3qnbsb+;J_H}Fbujn&|M#yi=`N>Oy2<Kj6+P3B))
z?KZh{J~Z%_^lfPC`+)O*dx@YR#Pgjc$0CBkEq281DD4A4vi9%*cP!&--9=TycKG8A
z%x@+)AiB3197WW;t}%4J^c}g@z+GSmN%@UKb}-}9`KCtSBmK&rHcd!6RZbF?l~M*z
z&+kZUXb2<9mS8%d4%7w7G1Mx$Y2m=%%ezn8c}xMVf7W}Wc6g3NmVQRZ5V-mb50?L}
z#(8!<^prHZpw>Jlxfp<}U+w9=Vzp1oL!{JdxF`Kwv_hm`$Vn}CL`Bt15?D$}Fm?zi
zHNAdYydvxQgcvm6kGlmr{Q=XW<eo3)BDJ&jGr?CZk}8WqW-9RDkXi@<Y+J+dvf@_o
zYRH2>9!1?la<eOaNnWz&82s($lh?+D)%*Q;;-U+HX}vd`ZVLrxgF%AVdmCPuqyBQw
z3Wx2r<Q9^G+~u&rsk-0)Fw;hr<Z^8x&Q4nyB^62Y9X=)f)D23iZ60RxJV!r`$bs!`
zz>xk4&_q*Rz@H?!T2Aq4k}Cl@XKvu=i77E|RDfPSddLNpZiO?s$q3u?_jC$`v#h))
z8=lbr;>G4hRKxs#-d=Qq!IZU)j_#nvR-Ahy3lM$RtXWZ)*Bl7HukC}-TNSo~a(t;x
z4yK4&%kmq=khLKpc|k{%KuQdPGd4w8&qBnZ?N~q_BA%MPX|>(y>!Q`DXEE>%v-Gi<
z2lFD8<Wb~N0Iw4s-uiV{_>y!5`(LyJ6CQ?=tar6gqBB(Bw^ahg3k*QQI2Uc_Ib@Pj
zDql&5MB@(M9SF-W1jGTmLwTUrZVWidmdKZNTW^EpX_Qpg4=h&}PJkP7uyQ+$;cf?`
z|D2R2@>><LgAjKEX_DLV_tNP<<!rGb0UiZ^4y(z0>q3a%)^PgOBa-(27GM+`fDq8V
z`pv9KEPaFy6=k`3_B7<8k*l5fw#daUGc?6ez(vKmz}W*8E9AKS-IXWr=Z8jmx99UK
z-jH|kFHa&S8rb>DgF%N{Kg_%SPQjXek^E1oh6r?ZL}+I(<eiyRely;LBQ#kL7A~0=
zS0%TW8Lnf6=dwr|ccT!fyCfng4|anQI$&JshU1xJdss*Rttam~={AZDPozsg^TvM%
z$g%}!?Y_EzG*EKzJL`loorx-)JtTu*giNw=2$5xfZ6deD;r`avI#2fL9X%YQ^j02#
zRp%F{+8lr&l^{`!sDWd!*AxX<rl=0A@{qf2(<5$#lp+2}&Jk4+|M&fb{|D=hSv+2(
zJ$^2pyaneRf-8m5Q=&1y%?&+*JNv_EYugJ>?(R)}1v}3BqM}pNG_J{tU5xym?``w4
z5Cd=;zC^URNW;j!yNt6Fcge7xVMpi7JMQ`BI-_ZY1Az_JToTL~2E14Cfn<USFg$&>
zp!eBBz2|DJoEeJk?Vas@qZ*4S_+G!|b<_k7%YGV9LS!hsB%;#qpI8zeG|r6>;vc5n
ziCyGHKieghxv`*$aG5__=S~x6PiV`_$7xOP(@_Rp)qdke{)YmxdtR8}ruDUI*^v$~
z8fKnCTmupNVbT)avf6AZy<EG(?(~xs>A+%mPG%3;c!fg%2?Kh5RlI4<L@R+&Y$DcU
zhnVoFR}_{7Cj-DL>hXHu^eKBFXo`f3n=z`dS80S-ugjfFi$1({lC;&2HvkUnn%D@g
z9SS<T`g&zM4Gfn%I4C4J@b!JGuy%fqQ9Pd<@DG)5o+J5pZ)`;?I1`u0o1menM_tWB
zXY;@9-HS!!j%cF9m&pN~6|4bhYxfg9cLJeoXP|OOzApS$gu^~mim1ExqMPFD3L{Wj
z(LCT*F7=KIhkvE1imH2daBu%(!7v^i+!Q=3(*#t;lpo<kXCYdlYEGDNsE|(hza!>!
zksoO0Jlm&c0;=x!X-^S$W8f;=A-xE6ICd>QwpoN=p$c4<dB>r~M88^)ioACPrXR<R
z`PPd;RiN=<tE^3qRBwob4(=}I)8S?1$gFbGcEQfzIo*?7dP%L0*1+?O0syk-do2LP
z1Jf7|Zro>TuS1k{Waw>DL0sSXQKvp#fVB1y6suCw@MnSSg4fJi=m4AD?5LDV`B4sd
z?7kiX370pYxoEGsVX#cz7Yi5Bn2fP!pA7Y(H-w5YQLnEEoRQ-@u+5aznR8vZmDwZd
zTcJ$Cd2f{IqOXsZRxgprgabwXl8mH%FHKj-?1rSV$ObUd>?m=1KpZLJ2BXrb*@7u;
z)P39z_$6D5JoBvZfIQHd^LA}BBqz0kHNq9e+9dSjeQ?-a7Jf)r-ZdS{nP*|4>e0w`
z@ThXXabrs1gJc`Bt{lskp<JGS=8{XQbJmieWEisa@2>ghGyP1};1a8u)X>+QXH<NA
z7Z@UrGcx82zU{TJP4zk2eL<v8IJ4dsur$GTK86g)WF=|SNiS*hHax8H6F@F9TB7WE
zb5({HW0ac6v9PSnj`4`q{2iab!wH}A;(w(I!WWeU)wZ5qan;`%Ol?Q#Ey@ITx(5%j
z_u+Y}SK^$2vy7L8S{SU>A@%E%OVO8<>=Yo4H5k0SzW*rJr3^^Q>aM;=lli{f`BlO1
zDQZ@fGKVvcLrKZ<6t}-QE-l0!2@R^P@S})ffb9L?R~OWvf&lL$*PPdvkLChRLn77>
zvOtYoNgY(?TfA#zqY2{CC!Q6ma`ducY0fBD>i->C*zmmyD8W>J<au6RmVIfK@-m~X
ze;TdrGu_stG^ykHBzcyq8$LxnS8}`H=abMnNr?=(`@lHgQ5PO}z+3zH$%>;J%(>kf
zp>DO@dj>|;bvZJbp$dDZG4se9VH6p!L<#>fPs#q{!!}ZS81K0I$L@dcm<==>J?6{R
zM(z2d>|7M%WZ>a_Ecz)HdX{sQ=`bjxxH;5FmQXu8GYt{I6m#J&f_@M+swfT$PjSP8
zNs<0_u}s9yLdLGm)^RbGqEe8;(s&?P;0QOde#eP2oedtP0L)Oxt|%@}kGF;D>TzZ#
zORr{dAxVH*i2+-m{3;Wq#apg=*jl^4R;g*QESvG{Acw$VYqzo9-aHGzk-uO2XpDq#
zG>UFBCC35#?VJg%jk_VHe?HW3>_ad)iwq~zkS>?jyFD6A9YM0LpgkUQ5ZkPm!r08)
zJ?w6}Nrvt3LcIoSD^(r~qM*{d;QF^@N*z07yJIf4U>#eH0W?kXPq>E1tNe$!mJjpG
zH<x+g5}YDOQ`P|Ve%xYk=d!dG%1a=4H5xW?P)(_L`P;(svLqzOWu?{E4cxt$Se3t5
zpSdue@eXVo+jg1lH(*Q&``ZclI{j6L(35#l7>p#^HY#jEw5%+}c&={+p#>pt(jRR5
z?3%^;^Uvv$^kqCfE-qt$p6}~5T@a;0|H)%a{@&7zP{`ucf>wK{GA9i6FsIIs?no*R
z+$HPahOgcg6A8Rm<L!mIpz7F|h0dhRKiT1aCZ3YmFsScuCPP7$W|ek#0tY0=)mEcQ
zv$9t{@@yaX!crZ;!#v$_s%%Tw%0e%tqp-fU)UEfoxY9x^lLBT4wjbDBwYvnkoZBJS
z69|H@_?fEJp2S!WS|Iu^MR%qiOLN-TK=UZNZeV_OrEM4NX)w(WXVuP1#5ik-!K-UW
zIR`xC<Y0aKZ}swJw|wQETY=HpDJAcX)e(QtrlBW2PdrNQ8g^am?4d^WXE&`m7S!i9
z;N*^?=87e9b%Z>^x@#6J%#@%z^E`f)#jha%cVQ5VH=EX(?|hKu?sukf?&TqE{0MF-
zVdkH;A4a*Bz4H6^Rl!_14JoMepW!zo>F$x$EKB2_w|D?tw_N#Z&=(n@NQ)1A+&{4{
zDLH2dWWaMm(ROkPzqpEK$|;+irIX7TIhs*Qi;kIx#A4TuhCvsw`;?s-9<tpKO5JT1
zEDiViYTfNzRcj{mX@hdI8=C~1Ogm_$2p&!+jZz1F3Uy{mkyjc6QdYP2Jz<H8t8NLT
z%*U$oaKWIg!O8;=9%ZxR&~BiC0s2eaEZSY>xCKq)^Z7TnDax>U3q`oT*FGs}&H5E$
zxXfyPpZ-pXYlhP%y$BZ@>RT+xgbo2FeyU=e>X&e~L-yUcGVl)xJr}ef+!u5lwP!ng
zZJm4vw#=m=5#KIa9#Ft<HNZ7AU+SwP=iD~DPTfCB$volJE|_g0G09#t*C5(=#m385
zHp^50^zS1~vKlh;)gcUW@|1^vrsXN(49-QUEGfe?uSXj40s5x>&*<Sym62*<^-@SS
zRjGm110!h@O`?fFT)HY?+ft$snspNwIbXdGA-nq#eMZR9fh^%;S9KO_Bu^uUm^RD|
z2LaC)t{C=x=fvvR>GQ;~{p6Mtd`WX8KQj+a*x(N^PeR+1x4WJhFSB0@D4*KjB1yBp
z`&%J`pmBdV;6^2BB&UUTe^_z@lsWEbKQt63RI86w4!n%D@J|FW<{k@fkEywsj10c4
z^5oHSmpKzhp}UR|ns|I=djpL2XIaSDZAXDtaEJCo=pT2B`wpi}?{HvL93Ur9?xW9S
z$W7`u@|w^Yzf3)ldZ}2TE*wu$#REZ6Y|&~49q)#<YWAYCUaZSVcS*UmcWXRZfd>VA
zB1hZ{W}xe23qGU?q$=i;iBeL8TJQZbhRYwnor!&fu`XfK`kxr#NGS_-uHmU1oUtZI
z<?6P(BN2$9UVqLrKIw>%A255opNwL}(_hR#7h9epQs*M{3htNJ0*{z1XrNVwWf?&!
zpsOn>6hi0CgARe#r}{3|c--Nd4|y&el_%T<IB@Fs0ZZ_`<EV_d*04v}y)qn*`+eE&
z#&KLmSTGrlB#NOMvW}BO0@-I535>REt=|@W<iEJYP&H!g4pnXNBR}_WD4oH%VXLYq
zGFtxrirk$P(~oTBciL47rl&^1eY~hY_r|(i9Iz6BBUj|c0dG9PcMXkcb2rnQ^_)g3
zSkhrPrg<jd?q@reQE0wUV*#kS479~3MqldPrwnGgA<$hoCS8~jOZ{^oBtmc83)A##
z)p&cH`5^(n0&M5`LYJ0YA7D$*ZZH?_nWp$CIBuo640-tUG&2kd4k>yRL1D)Po>If{
zz`zO)DffH(XvU6>ulGS#p{|x%vaMdwJs)tgS2Iyfdq|g1fsD-eXnr!yT#r1JS10L^
zb&U12n%8SnFhHhD_HAp}XLn0!hBu3nTMV+kvYCT_5DlC?*I*WD-i(^Fim?TLsdK!}
zAth2@k_Q)O7)q2o+v!2pkEjp}Sgh)uQIZC(u)}!pYd=<6LGJ^`IqC-9HHVTHMjcQ`
zdjX9kP~ow+=K^(6bC&5$A3I|*WeF(oW5@!)8(6stWf7I^h+=~PE|??+=+9JO$Y@{e
zsFRkN1ZNz!%4<VIQ5_6tv3`QCZ5*L>gPJEYc@+t5qGK0mA~i~>X>FzNb7b?JD$6*$
zoWC}X!^EpvEqy$9oOeofRtn&ExM6jlp=;M4Mw!6JaZr2KiF`akC7JpMVqUAZmD*)E
zy35>iBr}x9MkhzRC#w*#_#t34!wTSy2H4q>oNUTyTxlvZ?)M;PF!PR$Hkk9pK@H%0
z1Y5HCI9S8@D*-ykg#TmssNK$Q{Psy-!UbRLj#?J8A4?FZBS`yF*m{t$Bj(b=URV#x
z;ZtS1o5$p>;rD;QMd-~ECz2iR=AUBo{-p<(RW4sbs#|H1&A@)8pI{ScRq)5hfxFO5
zHz05osUmYe@oMFO`3Un$A};d2^R8{ZjwFtF5oEGN1YqF>WYQT;`^MK@RV&5XinEd2
z)idXR8)nBtC-i!!rpp;<R43rD$xtcDwYF~^M%O^_tlR-5YWFTKL7eZO$nEnYx~-IX
zEsTzdhX~E;Y2=7uW>_w>HAl4Cgdvt2Z%!<VdOr$UYc|XKuF7jYVZH3oOi>w+i6%dh
z{?VOgtANzs>GQ^%oxX+l_)^eJ0O8<So^ygXFFr#M>{VYNPfT4qSs7f$BOPI)?^0$r
zF(0ai#W)kB&Ea^ww$0dEqt`7eXX4mz>Z8tVjFFN-GfpS*_p<#SB}i_yE-eR7u;t;c
zEqf6m^D7Gz0J@}HD{krG3_yhIn{m=3CKz!i$Cw}(%eM_0t&FG!@6+517)wF3F8y_H
zoQ;gOa0HmTeO!MRdf~ii)XQu-KboKQt<a(M5QeTrghCSx`UJ1@FrJaBk-a*oePGC)
zz2zJ2YISEMYczqgk_J*?2(pIJ3Cx2c_yG`K<4kIGhOd|6tgmi5f>q#WXxcMVt^&K~
z=|AYX8Ek6ux1-yw+L+}Zq1d{G=*SEPPhQ9>K>*8nh5m`ZDF^+6)KPwj94gy~8X$gz
z9Fo+f7)eTW=s*l!nt8)m>VE>p%EmWy%b4MiL<>U+&tII)F9*YPK1fgBxlJ6Z`V$m&
zpR6k~?y;9^!mSfh`tPCnEV;YjAQ)m4vOsE)`2At3LWbFil|3|R+5#-irTno58Z@(C
z=&Kx7IKw)89#6S};=#Uk*uS~K4f|l3Ik$pj%7%7P3mDJm7ToBi(ktsQDGcftHD{)!
zrU!!a^ZSw<gC!WGeT!sw5I{0+ChvTyUh$@Q3IEAdWyLm}fOJ94mp!mVe{-W8BhmxM
z`8+$AGJy-m%5e*$ZJ7<*vaSqy!l?LvQCJMuph=mbB-23}fdX^MY*sFMrneI`*jSxn
zY9K9SwQQEawt<)BVrYYf<GU{;??9uW1=0&$`#a02y{EY*8c9VGXAa%)Nv%T_vo!%1
zU7>0=(+GxUMmgJA6Y^=i53t-VH)oUOYb^J@x|#$@kWCDFC&(MZ0aKnKKt--EG+K-j
zQ9*6oY$U5rW0eF@sb2}TgNje=$Jbb1T=Y!{g#bRAKUF?ZjU7!{ZGy2{>#P3hPx5m^
zhkSa7M_H)ZgBNDOfJea}yV(+l4AI1wzJ+(d29E_YEG8BSIXSk?6l^D-4vWo~iVM_>
zDcfj1V@1YqY}r^6Ii<Qc=9TrY6N<fFI^r}KqJ3D$3Pn$cET=kXonY0_66k+hi~UF@
znh9HxVCsA~*i;f8VX^bK<T`?%toYxl=|?up6v~VquH7R0(6SE!Io$SwzFAB#J?o^a
zeE8Z10KVquaK8GP3zM=H<<?ohQ{)^gtubGbdw>Ry9+!%l2o@o2X0cbgi{0k1hKHn%
z5pkl+LDMo;kq~Dx@cF-!iJ7EXN`Yv)yuaqYo%O;L(ANnyt0jj<ryVUDmh|pys(QM>
z9zf|3PH8JRw%UX%+}Xg`RSfIjl2wF0s*a-^twfAC+bB=LP5||UgIuraU|>g}aS7yS
zD8Qc0uH~N;1~O&-hAx~0zr}5i+ZjMsGJg<rCNv`;<=$n-D1fb`BGlxW&q!h1p}m}l
z_>kf?qM9JQ4Sp_2>~cVBK!`Y1w=Kh0Kx&o93lgxMDMl#jB03$Rr@s##%!kJU@fuls
zrWG)M`;+y|FCO10X0$_e59uJ`750_WE{r09cOC0lgyap)bR-9tsfZ1sAtB;@thCwi
zDRGF&URTC+O}=ZciOp&jf%cpU|C@1v-a4*}`gcNKW-8t)90(x~g2w3u;=eq?)YK$I
z;UM1(Oq5rL24>~9B8)61!hB>li=pvKsFJ`XSY7Dl$(APw4l0kq(-l4wO-w_WqMSJ$
zfi-6;W@yJFnb@zwIBc7o@#9j#uqfT_(QQL_QG(F-FqsMP{tw~v4&5?@Y-sB8dX5`m
z&q`gvYr44lo?CnJPhmMRpv!pYxUBiAV04nV8|)#gm!oDg8^rVW{{i%suX74zZe(+G
za%Ev{3T19&Z(?c+G&eOjATS_rVrmLJJPI#NWo~D5XfYr+GdT(`Ol59obZ9alG&wOa
zH3~0GWo~D5Xfq%%3NK7$ZfA68AT&8OHa8$3ARr(LFGgu>bY*fNFGg%(bY(<kV{c?-
z3O+sxb98cLVQmU{+I+WTaHieXEgTyiJGo=qwr#7!JGRxaZQHi(j;)UEq+_0E?_KXX
z-?!`3sa18=)R<$<vDS}ukrOMa(hHl~|1^`Zw{xLqW?<q0sM@<anV5;%o0=&(+0(0-
zS-9F5I|2Sl!EpmjfhI11pJo<7J2=LFAry>l%>axlqEb5A%C!G?psAVpe`=-e%<TbO
z|7l_B>hM2p+{~Puf%bL)s(%=$0ZwMdruKF=p8tU<*t-Bt%m7sX=?nlm1B?MKPR6EY
zw#H7@0DE)5zqbxh^>lVIvvmeY+nLxqIoLZHyO@~*fOY^EOEZAF9q?bCs>?sYf8che
zjP_0d`+qV{026yxI~OOQnKQ$G^NQL#csc<sEL{Lp|HouuVrHWI*O%r0LCXLTwy^=I
z{0ra=P%(2hb8`D9_}~2{jBSB7p8po_e`foaY=D1s`|nbqvjot?%v1^JVq*Ev?5<}2
z2}`>e+W<|3?JR7}08IZ0soR;FIoSa1%#`e%fiC}&l%ARCzjJDqKoe^_GiPT2`+rhq
zcBcQuDE`klra(K3e`#iCYV2hCzXShHD;WdrT+}=r{x|Qx&i3Cv^Z)VXjsFqW1E9;q
zz{JG-PvJl7e=Pm~qEmHovbQ$V1e&^7{(n1)h}e4oyy=;_SpoEH>}&wee+*_O4kq9K
zKitIC$;r&l<=>a;pHcqDHwXTsi<y~+nF-wbioFSMuvL0ihD$wG-sqH=Ry#4IIIH_e
z_qbKK;~9`fE9j@e3MGjQ++R!jczBWA2b>ZLH?KEYpRLXev4fwo@gc-;)OGtNaGVvM
z?0{-wlw3ijZn#%V$D{HDqBeKSFQHrLk*9NL4JIS{#u%_hI*e!KZp>K%$$madBMG>^
zySuQvodEVSfKYszY?4+93|RNo@HYQtACp$@SczY;%`b3JwjoJCRWFxn1@w<ou<RkP
zg2M=k52#loW)JMXDG^L{+}OTZP8>MHD|N$Y9=;w#)4R9IW2j`#2IS6w=b{qVhI=6e
z+1fg^OET6R<#d)nieaP{x_<i{Py<5PmTi@!lw5NnvCs<#Rn|lP>c)7Oh^W!n-)W$O
zjg58XQ_r?AC+%#xQ`&i!G3u{WF11*syVP#W07C2sCVe#7ykASbMK-6CCesEnqKa4>
znFF<qD=Y-kk-~zRjMeJ(3_$5N(E$MdGkP)H%i(o>01}O~9rI@^kjShWrYH9{$?ZL2
zeS9XHi5L=vmi&vUZI)YisZ3D$Z#)*CB|T7Gg!x<JE}Djytns#atLOyyB26O{z{|jT
z<zG^|*q2`^&K+U+rd{S`-33~#kU4O3|E-P~%3ko4N)W8^%^xz{A>#K~S$-}pu`(H?
zW{p9IDb>os!Dxsw>qf-Zi!2QyJ64_xYQ)Jr<=S#yhdexA_N-Fo(@$6|Vvq~IGG!-r
z<5W+&HPe6Nu>L?r5w4k3q3>I#gPC{>ju@*SmR6iec1QQ3uGx+Z>lcdBtdCF{D=T!8
z9@O3KD64dw{MVn$pseEWyM=lx`eIL-=IRB_NH=7wo^<K)D^FW}bS7PS6RSNA>5OC)
zouTD$j(*BY2|{U51t%l-i{2OVaaDmQ0dd~y-8f8aHqFClnGQAh_)d)OdlJVZN3h|6
zd#D+};k<_@{?VIqBHoPk0ey4|9$#cDbHO$FwowaN`z$a$&`8u)&ib%UD^Ksp4WKjx
zze>;IF5Zs&jsw0@GUi0W>Be3xN~d&ytD{HN_1l;q5Y6;)-`wYxy3*XF%ZYnZ!8x@d
z&BFGcw5N{rvakiCLqDS=7lEEe!)8XQ?VaSpmLQIcV;Zdj!Dm5UkA6v%qROmn$?tFB
z({Vedlnv7v6y{LtwTslJYwuUg#Dtxl)y%AQWPmPTGmIC77KLDQg>)S55^(gaUbEYo
zU*N0n`Jxc_R03ECBIs6Vm<0WWQJMd}pzJ+uHC>kZ<ucFMr)I4otp;whzuo*8V40!|
z{6Vl&Et>7wAaCr@Yr}-P85i&Ir)JNuDeX6a#|28}z)|2f6s&T|HJq-4x}+$9+NaXX
zm@L;#kC+A^WI@GzqL}HEV`CRunZpOOo_ajVDow)0Y(-x)QF9XfbN(B=Tlc}~^^d{p
zT(dLl<A(ZSakVI!YZHtS&DovF3Y7j$^1YSE@<|KO-O^pMX^DZ1JyP#O3a8F<s3i*i
z`cP}!S0hC=l&FYX$-G5vdpI`lix{QiQ$yE<%izu+KsfYj9~CPyb-3!rVCmjTN~^3v
zyf-9mzhe63r&bX%3LZX*5r3xfo*IP(KKM_DVi?ULe3kNybOuNM8#3h&TPk(p(>cF~
zW{gy2tqdkg42+MT#T-hPnfH=~DdHB<H8Bm@gq){K!<BwZb1UGhYr()s_FPi<5FPF4
zshfU~4+&M;H`k!|w+YHy!W6`|!K&@AvgMlvG-{C2?q{f|)5k}@ynA^GWGpbvlLZuu
zX<G?Ta0am}=-Pg6RVK)T42841k?VlU4_B%6ejRg&Z4rh_PWLMTdNVo7wxAqS$z&iZ
z9KDdA&k<jXid`s?DU9$B_UoQ7jPIdSYd6IhN^nLuZgL%j_zZ{pzJW!BCQ0wv6a|fB
zw8V8@`*xY;aL%;}X&;Y$@W!rS9$Tp=hxi5^-er(@T{J}x?<ds^5MLG@th7_)f{dY6
z3GoU>Q+$Cq$Z!2{Iiw$pmJB1oDi+L^>(_qgAiD{_lw6iIp=YnfY(D3<S9&ef(8n6V
z=9Na^4oMzryf!NpoOQlD3_y$CBMX}=4%j$F6oC??v_X`AiQ16otPDT+e5F`<g_lV}
z-EC-17g?F}DLFy6cAw|2A`t+4Z|_EKR%Ewh`N}O_UydssR5F`w?j)~LC+Ajs)lS?2
z)JaE2K+7Mu(uuL@H=w^u73@Si23WRPFJ_1GNTAH10pCFkue@RypC6X~3WO0{_cY{t
z{#pLf-P-{^`6hJ@#SrfdnAu@spvhQF_g^Ys*v4stWi8l{5o~j!nnVK!PD~)4Y<&ft
z@L(mP8&pXKt_A~;YT%^7_5c~ns{qMAScnn?UEXS(0Xx6J;i(1=x3NJ<j`_Mm%*HR<
zIn9zl$JI0-B(vTmnoAU{l>(bX?}8JQm);6Z!RVIUR8jMq{|Z0V<rFM7v(IW{8@%^l
z0V>~A|Lp#H2_@E;GuyU^S!S;S7der+(GGhpd!J^qeE=51!qu)P)t(OR-hXET%PA@J
z)}hJQnC3eY`n%E#1Qod0!(gO8iwjp*$p!J|7SLI5B<T??h2GRFK8WUXHKg`OO*Cz(
zw)C7<+bK6V8yuZHsokBLq%>SjrO>LAnnn_{W?KL!giges#n}SJ+!OfkyU@U<iszSK
zNydv!$6c`N>M9%4Tdtx(Bd2C?7%c6I7Lep5L?dz22j$kX<!1W<%EwYspTBfaRP*4Q
zPUShOaQ9S?hVZfHA$;}{es<UN?D#Sjb4H)HMP(^>EcTlh%N*zK6XGL34-!#~8)Xw>
zw=ZztENv6JiwOku|1l)0k7b^uZ@n+a?Ys+WNuxbz5T0<NJir>9&n~r(tsNPp&A0bd
zW!HEAJurtKG`@1;SOUQfbB6}V&sil}Z2giGk0p@7-N~vC=3fsBfD<(KRE>wDRJ=P#
zetag5)$8!FG&!5Zub|gKVO(50beJ?scq&-H2i<T)#J@W<4z~Ye-)Pd9^7T%J{UI>$
zwLu;7aY9}<+N;#($7W{jr9{wdDO*Lm1*K)6G!!);Oq`<?I}0)p<Rajwi(V{t<ESIR
zHrS@S*4!r4S4`{3T!oC!=P?&3<<Qp}EfZsI>VCdT=?2g)!viTU-MhbRfH4|#wgP3m
z#FhU!6*{~JvL#$j0Yt1;Wtv-AVJ{cUvlWIvDV4tb`Zbk97uOS6?gNsA^O7s*v-;T{
zw9`=Ls?Pk1l(t2%yu}HSiN!fjJNbA@;+zo2bZ4Qr(i@h>YkKyt-l+dBcTjfoeaoA!
zjfjnJd$u$fVvtX<l(JZY&|&|&p}Q+PR9CyJ0*2OBNE{$rnDsV8lp0b*`5ZgF0PoN|
z3-fysD`nIUB_1J<sb?OZ7%cJ@MTmMHC7q0F=ob7yU7b!eq*hZ>x5#<~_;hJK1ilBI
zd<V+E&q7RQ1<SigiLAM1-`Cd70Fzi&+Xh}-!c9>)7$R#nM7cN*X-i2fCerWraL(`5
z=f=C~uPI@7YZF|c8gNQ+v?7prghrRGd(l8e)9h13yw-r1T*T?A=#{~0|29`0dZ|X~
z))DEi(Og1`YUtw(lVI$F9Js&a>jA9Nd&Z4Wb2rt6-BN0Kb&`(983Y1yP50n5;f2@^
zj*@2_`p|33-Tmkc*2$1%=47wDyt}gRk8T`E3o?m05{#NY7<HKP($&m8p%RDl<Fek1
z<v@cpB)C-W#3Ou;iqb^V#*1lzdJCM8+4`La(9HaVa_hOB=t)V>V|z-}rn_u|8J@XC
zmSh9hIBTkjw2Jd98f>_)X78}_yWGy;kJC0NJ1#0ed!K+=*E4H(ynG&gn{B)s@Btq^
z+UV|0^BHVhkj7=GgXu~!ns3E}52Dz*s-*difX5$?m&Hz%M_c}j2~IgReg@XO-a##P
zXuo}QtbH7rBfEpz4^lP)^xN4wf+2HA>6|AG8qw4LF8jGFawt)`k!;qg-wHR+eUn)o
z;x^b#;<CvBasy-6tb}v*e#||BzO|4yizccd7Zm{24u=`X!2^Gr^Bvt`3!1la%GGFO
z5<~37if&*y`vG^Zh3ir&PO#>F|AuHwf`AOy4O=vYuXHRdQ)C}G4N4%Ewf~uf@rT!_
z`h#(*#QKkSY|>(+4k37aR-fM}V;F%v33ON=L*f1Cwo`XKq%uBWf22}6MVrw$G)w6B
zJlzSES#u<utfgiE(2^S6mJl_i&f{(jF7QF3m0agCfyYa!9VJ=#LEL{;OwvhFl#?Gh
zd9R3)t<uVvwcO?#M+u#dxuzb*@KEao%^)W|+-h$=;5U3{b!qCOR6Oygl`GKSC`fa9
zO_+vJVqVe}8hLyOg*I{jUdAZ*CL+HDc0atQqYq))aR}u%$ypy8UjT2ZLSKtWa+6nZ
zHuUNRaQR3Wq>C+P5R8mxsafesbxde+8-GJ*ATdrm)ATP@FZerr8Om8Tlsu#qu@vC8
z+JstNksH>f5qn{5c}r5ePMHANO~W#H^P>wyLXr!3_kwM3+l{<rn_4k?A&so8XN-Zj
z4jRE=`9(3>u}3laR&5Tq5UkRG<;nac*@JO1t_YpssAoJgJR<CQoOO#@)c;AHKNM>;
zf{2Lp{-@5%vTpaHy6vKwv$z^dA>5ulA17i>_-BmUV0a$)Z=0p|MZ+s#Yr^|t2|i>X
zZ5<A5R0Gi#$h7Oci~9!G0G+v|ND_z)a7xpDS38w5bu(lf^2NDT>!`_wm;l&{-Wlbm
zvPbh)s@a8z6Yu+#IGtFT8tri=(kHt6Mrnh^%cUekkY-0Pc#2H13?3o<t=fqyNiPr2
zsX@NkKRevFCJCJpA@^AtvjeM#IdD!VZ|J;VIwqr}htABj&!-3{_eJ*9+n6H?&gE(|
zD`vS@d<u!QrcPc+2qeRn*n?Fo$yYvTdZHb!HUMAq-OTQD>>2bBZ>s2A*x5p~Xk_pn
zvsjIjz@fk$>M)8T;WN&8PqIbsVc>RDz}6;8H=rwwJfpE7{${*0_y7HjgG0BqHYKG4
zzwQ^SPxTYQD#3+QPVj}$ZxTv+C6X9?5n0nbuvTf%5K>ImJXK8;s2f0MTR;mFC-5TJ
zcY;u!t^7KeAu;Mh)8C^S+P|KpdrOY82(4XiNzV2oa-9>+t)P=q^T_innZ9MBrNdW%
zFWMG5*4%8<HqAUv?}*X&H)cR)3Y?or(PL9CBf%{7rv*J2+Ld{mgdA5_qurIloOQhk
zd@EFgTM-{s&3;KQ`m04Q!vwiR&m>_93{2onvo7^y(YhvWT6#3nE@RAoX8X5k>6N~p
z;W7acv9am9`UwHju|Hvqc4F@AY3wmECkEV2*|^3>Zw!=8yk_h-U2P<hQg*KQMIKq^
zYfSm5XQe&AJ${&$Ga9K!H|jFN(pl|#073G$D8=7hP)=DTeW>#GYja`XR<mh!{2V>U
z>4+3gFK9oP!)eKK4#!vi9p^Qim`>{(PA3tykM8i?jPY?bA&KN`eb4y0m(A0GH%0Z%
zafp7LZvDpfbrxr?T{I9|T9O}G!`g*sY{F~ICaRWJ94m{+a{3$YZJa&4!;WH-3&Lm1
z;QY;&Rc8tG?miaf^XDcKxy7I~uNN2@QrL2xLKh)(G1e%Cqd$X2a3_i0qIK01<w%Y?
z(eW@}<~UqL=fTiznb~*H(qb807BooMkR<-!(Uw2AR2~2xRu%>>F6zW4%Rex}c1uu@
z8(0iz{Zq;!UcUDd?`W*^HnGgEVL{~Q{s(GgZ5c`JM470uO1)|f3Z@B*zXl1?=)c{~
znB$LV;$Y6`dl8}o34k?J5_pM!OUCpUCRB#v!xb~V%E3ut;oX0;$T@@Lj+xj=6fEE&
z(8>QaI);@uVyDLB=(_%7J(Rlfx1*gx55&jyky2*UplAn0$Hg&+<0V)pW9!K?ahOs3
z#dp$8CbEjU?fi!uXD09V@A<jU+IW-n8)i-YLfDlhAhq?6rJ=50dvkp^y$C0|=R37}
zC|zEMOf~5`Gn^Q%;#Ouhy?ry7!fc=(R2V`E;u0p#qd<mH@--uT3XCW@Wx;aKyK?46
zYw*W4+1-bFr90AUO}gkwjXT^Dw=k_z-*r5D#KzyydA=S2UQr=Br&QHK?wz9};l9@n
zr28Ywe=dDLK67qVKjc9RbwX+5&uY14f#a|Cnt>$*uVAUrpT?IW@oob0mWm{vyJBZl
zqh1K?(;0*>uR2o6=tLY1N`#MOtTVyHvR12s%pUbuF$Jqz9D5^;LbXL{W}Be~*^K3~
z`eE6Iasx~wOu^!dles<x&_<V7&cLOMiozrglTqrx{WZ5l{%nz-xO{}|v(AZ8+~r|U
zVyw`*@TcOHZO?mKuKS3fWH&Rk;SDgwAojfV<J0*{L(oQbV$N2g#qK|6-lfOE@D+VK
zL5Yf$H>o3elI_(o`&-F|geY;sCppfDH7_n*NG}(&=xv26tr#^^4Pb6DXhSV@O18Q#
zx_L;(!|r$=imT^%%Nqj5Y*dqDlq;(Hh1U}DUyN$K(Uc;_9PBh56&fgh;G7*RBc07!
z)QIwkRN?a_>&ij6DLxTvxnWZCMr&ke%HBd~|1kLl^RApgo?6QXDYmryfXS?P2s`#v
zC#%Ic{|yL&YTw|)_x;5#H-}BVs~UwXf~H>UL>>-4b8BT4-l|mWC0Xh>sqXzPy?GkJ
z*rpG11I*1P(fXp0jnmMdEL@MEhlLI~exl7%{48Vry>{x+cAa>ZBu8%9q_MyHPRCBJ
z>kz%e-^wgWUYKPG(>jivJc$#9QOqAmY%@p0J>-93>fJ<ekj-gTW@ApYM_;PUNuN$<
za_(bH*MjCeP}Au6YlK)Kf|J$3S91e~b{O7{k!-i|$QV5)%#;!Iz$xAKg31?nB+}lP
zG26s)V%ET@X(PUeln*dsg_Cls{0_T?JSQuG_r^+1%+@20*$k4oc0R}d`4wu}|E+uP
zeVAQ{nt9gPkJsAqD+pK>AvK^ZQ5@340^65kn_tbf$42E>a7!CAQcOJTs7%75Fn>sG
z+(QUPA}^X*_A4}qR7XdVZs45XRHugIA$&m$^t}cyKRw*?1P!_A^$Hov81*V8nU>w)
zdS;RLG{#O|IJ+~q#|pICECK#NYkj%2{2Kna@@`-#AyO%IykHP1#_2uFpn_G-U4^^u
zspUAwy#0<4(Zh09nx_I}PWS|#d6$Z?7SYFG@go}j(=KSUc2;UJezoGhQL-YP^lHAP
zQ6>_<W)U8mgU+QCZhXnc4QC96@JDA(5RS0vEUU#`x7sRAXRO}wC!e2x?-9)(ln|`p
z<CwLq!ZC9{=uW-g3*6{p!HFS;gopIOCH3J+E&CFBMn81ttK@Z_X8%y1d1ccFR-!9%
zB96x*K|B6De!mZTw`3W`Q_@GE$e@*$9@X`B!SrC>@S<uN4816)3nxc`-2M3-9~VTs
zKjT{6Jud(Oiv!C+J8MtGRpRFV)z>e^;rDrU!fC<o!!V0M)km9yT)xZ3r@-aizxQ?L
z!%FlNmOyVb8e;wt63<-cAQ2;fc+L^qN`Bjq^EaRFC985hqa2V^Z3V;vW$c5(^`s0#
zQ-)Xv0fR!bpN?SCb;#CVIbk?u->z$FtBtIqu)&3an{w8@6ulH)i`^+u_JEdr(PNJv
z1My;oV`U-h>|-sty+_1*$B_!Q`-V^?n7!jDkSsI|Tt7+mG9m4t4zU;ao!2FN)*0{Y
z-C7wQm41sDPZ+QsY>?^mLNUMC`gzJ{f(SZ9v8j<K-o*ddqsXgNA{cy3;r=&QvD~Oa
z_BzEbwvisSq^oTxVq*CgUW6zuW2~)N6TzOKD|CRKUh~)ZFC|VE78mEs5sa)JUW=y&
zSBcK)&7Q{jrnjXz;m+H2bf6aWbexE?NDaNV94YwcjIDJ~TF()*J*6Oq+We1t;A&U~
zc7(J2$z(sfE>7Hfh>UbWsaSHDWHV`TX}_X$$q}Xh=*2n7v};W|O~bbzAeily)D40j
zO?x?ZV?3%!%SJOKrA*@2%VFGOlaR>~!c*uy|3vSEj^xdT-j<S3_Mak93j#b@j@)XL
za{!U{W<f!w=7QT0@>8|ybQrfX2vOO!l>ST7S$oC2NxOINhH3X9dJOZ@+*);-XgZVd
zfvbh{pvbgd+Kb=rIo4og7cp(EZR9Q8?k>t{&8KRT`~AlFxulLT78RVKOpr<>wf5tb
zzl=4?$)+l%qgsIw)xzD$sHr}>4#=yz0W21VgE-At1CQsAikp#T-8iTnxu;i9l9X!k
z$J!hQTri&shXq?i3F3&^U9%y^Ow_nr7R*;A?&mTBQaen9L<?6uOK(Jdb0@|Jy?Klq
zLY4ZCw>8D#k`B+y={*B_2YsD+MH`1?gCb9U7<ahbbX4ZLIkmluOGv)FeL8_j-yyoT
z1j$Z+&-({z)>huH!EdzIcqhMp@`haLU+O(R*iHTs`SD(g{^4E}>jp!xr#)pc&Vs>r
zd>im)Y+sMJcsc#yD29jpkrQiM>|}xmmOGIl2MTo8vCYcRf$o4}RfI?%YIR+&t_21@
zj^^{KN6%8_xi1jm-eX)!4?40d)CckIJ%OxmO0g*Rj`clhj+(CU*I{jLP|UHYG3icm
zZD-Ss5WctBAJ$I^$xh3|^um=O1#wYkvz(7&HKNjKG8P2p7vi`EL~c|Iwa96`nHQ9e
zp-zNDgd2QqcSE#?A84-+rNr3EiaB#M5iUhxVHLs52(ehBBNE^-*Xr&?TR~E*2nGD9
zeFDNT&PENaXl*^E?cCWcCNB&rgdv-;YC7MXj!cpnGlU|-Boxk2@(R4ES0jPo3R|2%
z|NLp*OM!V_xB<b)g5LBZ+{-PxEzR^y+uyS!I<3;jLWbLu`eyc>R#<AM_PPHvDCvJz
z^NyzM4?&fCLCoDp954rwCGWoMDa1Ya%FY?J)PVI#@GE>MxH^t;r+o)=%Dh`!j>r<Q
z*f|9PvLX@AueZ|mmQpCe;&v^gs`*Kjd~>DD&(V5)#{y|AONg2U?pLd6vr=A#g;u1H
zMnC3<%DEH2`nm1fzJW*?{8mx;X9<1IDraW3jDemeIfrbkU`xND{!yS}ZX5A!5_el_
zJ>7tcLaXIDycfgK*x)K!CQMYo5@ISM%2rde(w*oT$@p*|i{7`6<=cGsOoZ}Vey}JJ
zZYba0l}h7y&Yjbe(>rtnT~n$J(HSV=EZeL1#t3orvDZjo<U%o%k!XuW<E%^?s(%8H
z3G=<Y>+zGCIug&%>14co&P+Yf*I)M6N+qi2f?j00>pM@G<3VtqqwM+Q7(J%u)^(fv
zRXEuW+O%SHg4Y%h-GHT4v5vGaMmARG?``fq<OMbAKDhQG{>U&+-wsb9@(;SeKGloS
zE~-Sy%boh_M|=S*+vO{PwtBok%7Aq%y=xm5AOHt*`MKtYR~FqjuvD_^?6>4qC4P<c
z;CuCET7JKg05KarlZ>Df=SpzrDGy+}v%KMZPfB^W;gAzNFX&x6-k%A60#X~w6znx)
zmHYko;XgWa;k-@15CZ2bsYMIa@78d<y>$(r-x&I%igYuXZVcFnjurk^OA?3fJx+CW
zwv2|s%TyOxrVN4dUz*TBb${RM`-)w6)YFN$(qNt4&s;67l&zh^=3J1v)QR2Aqrw&o
z&@XQt8X5N^QOELW!7CgZ05eIDT<G7g#S-$4Pv644A_K#miE<!(!9F+N#MI%y*P}lm
zPBZ0?A!<COsV&OOs;AOR97vQrlE2AI4<crGF)|z>5A>%bWE?s_HfLkPb0CEy`g+vW
ziLay}mp?gOsb_y6xAm>((bp{31O{zo&yV9JTUK52Y#?#Cz4J)*aPO(>Fcyu16%{?*
z)|>@xXi|@O4CY7{R|+NXMz@01bEuzg3Ga3=RSd+P%=81Z8t7ObVYy1@Uai|PP~V-g
z;QgL_4OFzc;8r~*d6>YfpX?cZf5`C~-62#luJhay?1PK^Kv9xj@ED!~d2f(IoGdiG
z?qJWO6?E39+6U2S`Xj=7vuU4M4KU;V{f!W0SurPq^rETysv!Ey5nRW%d-qRE1xzfj
zAD*3ib|aG@ma)j>lYskKad6~oOr8W#PFeP4Y3wT!S%}Ol3Z72=A~1Za&sPTS`3QZs
z=YH_wfi}N)K3M7q3{H`b2xR<BHXiY<fsexvDv>XnD<c1>i&Dd=P%@W%yu8?GRf$1j
zFLNHk6MYZS8Jy;$M_X}?N-G6hZ}KhA?u(1@@V)Egx`!UYl%L;5InI{4xu=-~EWr42
zolI&(yh*&lC)d3%Oy41FJg(77a?NukPA8&Z9gpOfg4n>?%JcOu8C%J6s_8eeg+HNW
zWc+p3iUjqTY=!&p+lp43@d2NaK_u7bzmc}dvm#Rg_3G4b+{mICtY22`VaLufV!EKQ
zd{k+n4NnL?0TS+}Y2H4`<dZ*+^IWWv<aq9jI{d&<?o)p(**1T4*wNBRNUkzYBT%?+
zyjD>$w_<8JGNV7|$hYB3GHyiRPX0~awc1nWN<t<vFWIHoE0qShE-LxVX|O_{ro_T_
zchXF_i+10m2@_oxxkxv+wxlfbol8(%?l0p_#3r`Dmj7c|V2}I7QY>3DGmZleXk>RY
zC~v3b@UetMPziEXXr?{3Ue(rY<Re+FYvEj5zbzCJ>esDnR`?wXOI~Ici~b9r>l#C>
zeW!Vv8H4d^7GlI=knL9vEq7gs=PQw_EJisSM{kXmE(uqxA4h2{X#m$0x@$mmi51O^
z3W35xLdJzNzjmDOY*kcm;rs8-uFS&RZkj5*E$O>{v(TXqA3<vS_uY9>vqjI{6TTWf
zy87<{>W4lySf*5U)h5cd+`rMa;(YX)btx%Z@w*-hn<K;WTPIYT>Lr<U^D*dS?`#!j
zXZ+ibF1G+;K%Kw5yz_C4ZDt_^Va(74-Ak3}4@Iw!q8bO5R<<q0Kf{a}QENI|2bG5I
z^V+Q6mn6~pjGA_f`-<n&2h|M!9=wIj;wAl!Vyu`1Z|$2I^l9;fxSq?BnxDo82-A1@
zt8?}2hx7~G$CkKA!%ExVa9|EyJ#+=^_}n;F@0POx4Wq4R11w0Na`USR4?Pofka~Og
z9v>(`%O+n4`u@zdIe`lZkD&*<`ATXa^KwHD%g#)j440Q4(dd**5>WZ$L$q$!Q8#1<
za-8K}Mi4ET#)(FGe>L=eyt{p}C<_cD&PTyN34r_prF22W;~{mgyAa8K_H2Uu#gD7)
z{#JSZ1xJNeq#X2@OZVG{s3iKxGF2{35oQfw8x8l<P=VSnV7&E?noFVqog_Vk?ruRY
zk}PmluykuYgt1r-ibwb}ty+R<9i1$1oI&<kS1;A>EC;8fR>lo15L7a?5?OabibRL8
z*DFCz+=OGA9GN5p@;bJSEWuqBCT99266El7|IL*8<o!uZfQpClH}bZy%2UEL`Wmsd
zSD-jjm#r{B+WxvFut^3!a-W$mOs!p~O<q<*DwW9PL|4V4jsr%`!*p;uS#uug*nCX<
z-mA->!bVM#;5PoqL%Yu#j*A}Rws*IGJ%62wpvk0IcPNHll=~)on>U4R<ULyM*g@}3
zR$A=dnSWl~eXU}gR0Sg)$wX)HTA7sZvZf;syQi725mTnfGzV`2Dw)|ooVVxlY5`Yj
z0ja$IBD$@fgjI6R3wGp7VobM80AwdlVARL<=f<o>JSgDf!b$=tX5!~NNJ9i_+zRTC
zrIjT%hBz9vIIQvP6LN@c>z9gRp2A1x@FtSn`j%0;h_oH~(l5d0jsi_aP>nw{mI;>j
zwDMP!5umc2VAk@Mo>}|O7;<I8^Iu!%12AJz)-f-Km$QXICpHC1vt{{aPJA3tAsX+D
zP;hFDAcdP!E(JU{875xkQprtQvxx}eTk_DBi@L3zkn{VodDd@Lc~v#M80-DHoZ})=
zvRjXj?I^n7igLvjGAJn`Swq^uCnrZOpaZih%9HYhP#BsC6P~ky-}seP4!@lQ<G1(%
z2|Ev)mlQJb`Lph)lmS=LXCaD)#I|~kpjhdrLalx30-8wW#Da<thMTz514=CEHAAz@
zc<jR954Dym)UP~_l_(YBg(7z<3n<fR*k+LRmVMA!TFCyCq7u?uPIu@5eAm3nlfXtQ
zs%f3!XkF&km2`R<u%|>CLL1Pe+sPpcud}l}vqEk63;u94c_%$(M~4y8ea&7Iawq3T
zv0rY+N-SQnoSyUw2d{~u3$S_SDU1_sT?M-n8HO6*&{CASC@YmKXIi<PxN_LE3M<Dw
za+Pgwf?7^`eRQ9>AhK{<el&L?JEq;SWis|{T|0Eb*hEW)?jF5-4;f+IjBdsn%+`So
zHhGF0M@I0otNU8oz>Ll%tlr9_m(%^92G^%^%igdzaes-d3XJl$Q9{jG+aApq8PU99
z#jTu)u>9>auA4inp+FS%-J=WQGo>9>wqf7F5+TfMMcfL|I9#QX*mg}e5|W{=lVv~V
z=H71{yyH3;O5#?`Q6Y`hjhNzOH-tf}5YKs{4q5|&LD#toTtiZ{6@ucFBEYm2PBsGP
zmu^Ffns`ifO9n(27U(y`k??>Rl?NET9KpfZJae{FN^QWs7v*YVaCYA=J)fz+6(*4I
z>m=`QrkXPRzNrHqObuyK<PAHw1Rr(pbM6S4dUgDE#dJ;g_hO^x_H<%xok67zt`PP}
z!~F3JEZaT#i_rM*Lp43uYySslba00$Kj%6$*(~T9`oZ5SG|8+1dWZ#klw^xulr};(
z-=j0bD92+0{I9=nFQIc+*s_0T(>g+=yE5Q34{?lvZ<u`I-Y)#^=P*$;vWyT4FW!MZ
zW|mAR%8wKZr?-jGNkvVsP@$iJy1mR&ws3e%M|kF>Kb?G`#I`;K1dN5*6fO$!E29QO
zCr6!pms;vV+j;<Set50ealvx77R+f8Sf7+<D(5=9<&C7}CsjKZkCtX;jVi!BQ*UN+
z`N)%&6+epZ7W@$bl;Y1J<F!zUu}xrAl@-<t>w^!Mx0vyz8|=J!P)C5nU%|_;xEp#J
zsaNcH>2_RI;#Ie4c#UD|t?NY6_z66*#ScELIsd{~Wzl|TL-$Pyadsa7d-WM#{8^Qg
zP*yd4XDe80K|eYlgR9@C$|3t_bxUb%JFFP@`^S(q=w6b*4gP&-3DxdvexDhkU+hn1
zmlz&B6^0=ljl(ueyoKW9F!@Xb84nkD+6rrkO~#3Q5w<JP(NI0j+E{{@k`M<j*ucz|
zyufZUU#y?>SZU=MbrYa~Prp_EcE7a3Anm`ig~b!<I|QkZ3sW*e49ST1<`BkwPKd{g
zjl2R1<x6ufqQ{tjE~Ucw2ysA;kEh+g6pnZ0NY9}y@wth74hC||SqoneZ%3G|!QPx8
zIa~aX1@|2vSpNEbQ>|D$GkhoO<npqTY+#mp(JLay5u}*BJM-x!s$YDIWMY+;U~2A0
zWYXvz@49oy)GoA@G0n;*KpyRgg}0a0B%;)@qaC@S-%NIUWi>-s82!Yd_-Xk5Zrh-D
z$~@z)5Kbh=35leGv++kqC7U>41J>;W1(58o%l>^+5kL@Rc)T0@!wcav9$H!Xyw_Rv
z;<KT=Z>>F02;~cZw`r_A2mZqGu-)UT0MK*fTz)Xg(c;@ezhpn}<Y2Vrs{tWW;JOky
z+yr)@UuSmOZRJI;jfLd&z$6RrK@<4C^(0rTi;9dXn8-ttWT-1+*Oaf`$UGHEQaoP0
zoZerM6eUBdo~e1(Eg%5C-=x^Fl3NJ5l@Ie}#t+$VEhSpengP!=3UIuEVpEF^5bZiI
zl<`A(x!D7qQdh#{yj&ybSAiz1OogRmxdz3nu^hND@Z%Qxq9=<G2Q|pK?zkYAH{tCy
z@fQjc>+u<x-*FtXXGB1{`m`8=WugSPQSY}1r$~Q+HkPHTgOiSQrK8&(VqxT+<u}e{
zq>he<XMD_KTrv+de)jwBour^fG<v66U1P7|f~(R=TH6>vFeO_9*LL%!ZL^W-I!r)O
z*j#mKE6xRz9CSTYCG+J-GoGwj%{tf}<4@*&%m|=^nqOUkFx_qs7u1X?Mc5X4W3?Bu
znZSn#L%D%KTC-CwB3(+ec=KmC`fOE(!ue&5rt{=?(u;)0Ex3vH!OLy;gmRgtW1^Vz
z7J){Kt#7D=&W-g57Yh8CoN`gG8}DZ}fSZ}1i9Th6)~ekaR4C<g>tZ9Mt#|ht_7eb`
zz055>+STY<9X?vcZLTbfK3wGwfnEIOC-|-=5iFd?DdvMe7wCsmSMDA>iERljk7Vb!
z6W8duu(hGB6^mPT;Ne+xuv0U}NpBt)ZW<V4e`?tdOB<YnL~zK~VJWuk;x9i^7H2Y;
zZRgpqGs7J+uCs<Aq0h`y`6Cjr1l8(FL}w1px#Vrh@k8udev3-@6iqQ_b5qzJINjL#
zGKfna#_KlXB2hhT>tFTwxTCr5_n?J!{w)YU$=RF!#x-UWph;0<sF1sp6DY=(KfTz%
zolwBsgeL4<Jy`g`iJZn(dhz%q-dh!6st=h~&j798fdjs)Nc59);2TEb&~D*VI5!tE
zYgjE8YDR_+aPlKZ7(4I@Te4^>J-2I)V}JlD637+MGr9{0CI3@eekb}qv#yH|jP_mP
zCJFNoMLIs$)a+(t#dN8%^^hGq9}&9ndU0OwGBr%0A+j!;NR3=OSCHDDzF|)7XCzKy
z_Z}ugPb}sR@IXDalkrYtMdQGl>R&VAAR%#pI5YlE-omXmWF5sX8(2noOn$98X%yg+
z@lW6xbWp9;Du472?Dw(?*5RMKXllHZrcJEV&jzvvyq~@VppAJTSRy}ZJ#G+$J7Q4w
znV4UWw{alexAb*2)2^zOtFzaVALT{L^IlCk$>*-ICuYyLQ>pJ~uf4-nhR1E~7rl$&
z6W_f6E(p<%n77i?ozspl7k?4`71>uovk~7BV<9thF|~Kxf-@_tUH@8Km}A1l71f?E
z-UO*aTgmf^GF1c)!Brq7KCiH@-Xon`a(ZG2#yktq3(v9zFFfkRbUm<*u>W+^cG51B
zH57cB71~rSQF+YB#I1wN{8JFbeGDx<h`npll(OzrSRebydBx%!<Uz>e5jFfb&(%zX
z7Z_@Hn=#O=i@UBaNq*71E{su*%!=!&UWR+y2a`5Noutzw7)o9XuSuCwA84z>Vpopd
z(`tX6`HFv(6opDi70-Wu*56FPAcE!Hu_yr*Jc#RRt%G~fHfQ_<pW5`7gi*tIx9yFH
z4||ILWC5}#((jCVXe^)~OTp?v6L|ERnrXU#XrdRl{Rj0PFZBIeMmeK}Upylur2$Xh
z2S0&{l@z2yQmBEZM8+mxNGvqSj$x+PBJq}_qAr=-un%lC!`pjpcclzo0v&SOrRExN
zqw0!Ab&${r-&7tBXLkAnNAW;=o`yzv)JW}onCG`TaArqh$@7i__hr=AV2Ma;;A>s7
z<Hk$q9GsCvfOmA@PQ>^x{$~iz?5moVbM`OQn&+`x7+X_ZM3M+nB7~Bv=k<$&r;^=H
zQR^P^z-+-Q(WJUlF2$YfYo@oL6anNUm)}(p66Ut%<_uIv_7rYby)l<HJvrWsNf<b}
zY>x9YNK6Aw=qoq`$~b`~YZq7U6t!P2nT^b=^p}Ae1vMzwdY)xgdX<W2Z%Zwtoxz1O
zERQ!co^>|EgcFhUb4B`W7trf*KT@VN(VIHCsWbX7v@E=Tu(=ltKEsoui1o||L(p{r
zc#}b9F6WDX_8$RLSJFN=YP4nra;RlT88;2`I0?>JbD2cI7gC?PWA|>V-)l#qNm}&&
zVg+uXBQF|3!{rVWWOTs)?rEBmX^2c;a89oCH>7;Gd(MZq2+eSMmaLlbN=vnHj?Jz@
zvIx7T{YIqG^blPO1aw8^YVY;w^sD!#gN18u%birgA|-CoLD!ZF-*u@Utg%1<UgRe#
z$nUqlrSPc*d<VTS*y2>3EmOM^UF+MlUA88mF5XroF0mH!_Y7Mmgc0b+3en7zw`s~y
zGbq<2Yik+RVy1Aj?2~V$uNnj6X$|90X()zl{e^h`=+Yi}>k!$ka{4$@RBwHNwGG(R
z(e5o4Fdj?I4ML#)E?degT{6oJOtCwE1*%v<41fpWOBHsFWG}VpG=)nX>W*iU4o+$7
z?-RrBc<WvB_}sfugM!v7^rwMo66vi3aJd9{CRO1J9W8tg`!8*=sWNBoSRoR+Btx`3
zIM|DDdoCYhrN01pX^yG~JvL!c36_==vn)w)mopMR;8{t+OnLSqHtGgGdZ#rx{(dCP
z&wiZb(Wj4<tz+})*+^5DRz-)DS>DtLoUIgPrTD0(`;p`+jwA-l%<Prm-ci{$D-t#X
zTV^HH8<0NX7?n26(p$w3@Eqk^Y!4%svmH|j&8w+`ElNFP+P6UoRUA0>hX*%17GLY6
zQiqiE@a4DiG8gFiagn}g&lRp9nliF}$t#(~V!}SzxQa*0?Kd-vM|DTHsNn@Un^Jc8
zfB|%O>*UJI9@xNM)KuU07MKa`E-FXRi)qC^bVW<ZkHI4yZ^76S;ZyALlU7JAW5R8E
zrGi2|$Weuxfj)=Dqo9VZR*(dt+T{t~)4M5VSP$9DH_hy<L<A~i$#*1ae?xMp3u$Zy
z45icC+4DR2#%yX-2Sbw2lvjS)S*^L{G&Rk=$0_KQm3uxbDZ`roYI6?05F?=18)1M;
ztyxi&ez2yO-7<Em9<sxyR#`qq6zF{qz0&!;<!?^(1h3f!6x@1wKy-|uC~Z}!@YiCG
zEp_m<hOPYpwtU*(1a$doAvTlilreky^-lZNEp3%><)uI2<?)3v1)c{Jw#Sc{s#8Pm
z3?a?VIL%%I8C{ryd%HZ_0nt(BGPTmYULGBEsQqlMQM&-Ql4~|$LUft3?2Wo`-Mlx5
zX&`NXhg!XeajyTFLfsSJ&S7w$$+O;{T;LlwA_y{-gmUq68P!fDHV+L^(>QCwln)V<
z)DzGKVRjkHV+o~{CKNR%rrN9x{~$7JGHApJ?ou?+nP5N<fZn#{4ngvUy1-fPl3K1z
zygAi1yu3vfx%X`sWo*X4j3j#XoL@v3+|0t-FU_N-EV6&>;C1P~j^@kAWTvSF{TgTJ
zUeWNktw;?Y5Fg${X7I~-`3ZB4v>F>KJ`gz+YKT3c0`vhkS^kKOpm1y;SmYZ3dqyT;
zya94VOdarX#pyv3CE8CZ6Az<@$9@IOxb2pkDRR)FO2)Sz!RyU>4?P_9S@LJH5`!*z
zpM-Tc+)GGsp%%(_^^};UHc@T<1$qX^?->8J(Ov5_ZU_Drgd!DD+3!;WB%f8nFNPQU
zN6#-9V#Z4wT#H5ZjQ~b4n=`iTisBaLQaEJ?R}<DW`0wg8dx8Qbe80p_WA_!7sQ&VC
zBn0UH*oT!-!GS4f{y{=RF1FIvo8&2H!h?eXX-4H<D+}I0<y18q5kqvZWW)-k0WxPf
zusED&O4aM#6Q)vZOFbHF&Rd(#7O)$TzRq5s$5E)WCf={fy~Stg(-;UjieXSWt5ZpP
zxw7UH8L^ZJ3@w8<NM>?kJL=$pHfmS#pxs7c&y8Bw`@<$6p|LIhTe9TP4|#zFmFiFX
zuU0A(F2>hMch5&?D9+RQHL20dDSL(-WWOWL6wkiD7;InS0n*@@S1oHiVl(6xea9oZ
z)G()QqQE8+riH?=zL1^RE$wL{spFUc$OMdg!{h1gmIPPSDFw(z4?`9a;&yYfG@rlo
zBw3c9NFU#BRG_>d)5_Ov-J?l@Zms-DJW4o<_h9nFlYI^zRzBbvwy9PJEP&xxw5nxO
zJA;rnF_`ZgVhdI;#rJ46=NafJQC8@q){q~Xpc~@|5tzi3Wlj7-Sx{@9{JYU{^YSW<
z08S`G_c=Bn)P&Br+x|YAZp{Trhy8hBYF14iMTHb+k`8bMnPZex;l^yn8Br*0k*2T-
z*rX)7T*T>h-QhlR4zymIxu4VRyrvTyBwT;$2Il#jopl2+ruK{CSeAcc+HMFajNI{+
zgc9)vDwfk1quQLE3e2IaDKQn}JGBS4{!RNK@TqDW2@ohF##xY(%Q^bMPuqFIevs)j
z;N2A$iPqoe_r06l@k1!a=CrgB^#1*NcraVG@*s?aMX^~$`|h*vRNWL?3S;DWY8il=
zz;p4a5`?bJ5B7UT*rq^=L<#g4MFOQF#QKmicoEp1d>=jE!507XN2gpZQ5i3CPu4i#
zPn_YFq#t{2s-J8s!yDFtAiS%IEU4%6K4$s$Z34?E!pMV!s<2B-oriDblT6Pauw;tV
z7PtYfcsPW>%J8WK?&f3x!4<{&mX7XnE5N+SOTjp@%kw0yt{Jcqh75CdDdq^+n4Xgl
zp-y}(^p_tvM?ooJ5JXXZe0#wWhy_z`JY19OGF8O9Z4|IcLCE%D__l|N`kFMJO7{z(
zJoG!#2?9DyWKUu;B4qC?hgEIB0A=8HEUC~4c&ftLjyG|3x->dfLXV@7FyVUY;pZ_5
zS3)+0m&h5==0M);jT|TcNJ1|ll18fNsH_hnySUy=tOL`Xdrwz>i;N*^5PBz&*gGav
z7xs}D=2Y28U6_*$-J~v+wgBX<CWy<|xtt3{m@E<^@3apcXHdFA`?<atc4X0E=4^z*
zb?I0|z*r3nnp1gO_q21hc^92+Ug1`i$G;FM42bUH9$RJ%rm-?2PRe=2-3ciaMb(GQ
zikkGafl-jD#9KvF#Nb9H$vkMiQDmM+W6nug&AQ~_6hjc2Lh-I%;t@g|7tkFzNx;3K
zL!}&la@U_J@&CM@9Zwk1Zhz6c%PLlMp~8Q*Wl?l@wAgFZcuCqfy^M-EBoXz%q|1;!
zr=HblZ^EHFuguW=RLq&|3i+U|Sg<?ZPt<0{EAF_u&Ze*Rvn7k_yHTpFP20PXg5poS
zc_jR@TA@3+^!{KlK9)p5asS$(H4+sjEwlNpwbAj!e)!KL-n3=7H`ng4nJrV9N6`z%
zj1Rl%Z6maCL~C4)2Okn;UJd2ILFwx@p9BPEDk{53J?e-4!nGTwdCp4TIVcqOFNaTW
zj00E7M93Ve;;`Fe*P+CUyIQ5N0s3XhrsO$rFdrLDw7qLgiB;z*#Ut~rLQ4N_7&gI>
z^kSVQb=VJX<ClcKW%C;^JVeIl44fD4uLZ{z1<?Tr&9EP}W@GnG_P^NMJ|a=Vvy`xk
zhb34c8A`U$d6h^zyD&D>Mxh=jnB4-3wL&7IJ*|GQ)cC~T8VhW9aGkE4hj8&rA7eD?
zhu=OK;e(9cpV*-g{jTS~Kg2jN4|@jAOFODZsj6qjRuCz$)kQd9wY<|C4Vw6ebtPVS
zUZV1f6Zctr9yY1gmYS`TPTx7Tqi`tBWN0QnAiS|GM(iL_4Q+9=ky$eN_&gsu(cGDO
zYCY0s$pAS)hd;IWqL+i=G4t;2Bx3ZyB@_*B@&KM_phaN;M?KYMkt`{$cC^&O`?P)b
zBOJXSBM{`7Wt}vK2&s%vn-A<(GRw1H4uECul`3*ao~o<&_&Oa&p7baf)>U7U(c&{f
zB=G5Xc!B+xKUpjUWTK(Lip3vTBdR<Km|y$QvVu9*syL}-q)eCFUlbO<YQB7!i4<Sl
zs-Nh{nh6PuL{0-y4zn@3nrI!<%Gw2nxbqACRLuzXI%NQ0jiqHz<>sL^U&*r|m<Y{w
zBPGHpn*F{mnvph}m0OfCpsmJg9HqQvlU~n`KX<(`>2XQr@m9vjxCC;`<COxaYGz{Y
z#txWDyKT(PYMRGpA;_M7b+C>B7+#xhLjt4Z@O1xZ=4HJJk*y3aXl=`a$8>}{#*TNH
zng&+ZZVOyN&mN$7xE>X3dGuX{Ve*Z7ZiNgm7M?*Qhj<{&fJZpEl~(2=OX_8v`F6m2
zq-{^O9!-N4)2P7*ShZ$ytA+&9n+Ea(dy*&OeyeV&cfUTguUc>W8Dll{G=%L*!v}u(
zNq`fk1`T3hTP4Qc(L*01v_em+;QekzE!D0{TT0NQSrfWU3i>><!FJ(>G8&K*hfrIT
z#rTXkq~5NPxYp|cMP`xNlb}_vbuT}cUK8>uIZh{d$I(Au)DM`J$G1Df8Il0w>0*6X
zN2!*+*ngYIac{+=&BesPG-~H^;63-xxnM^j1PLq*$7cMv1bds^r1XwzZHP)rEbgEl
z|NQKfaz}(SxY)T$`+I@nCc1d|_>MTiKD~IUd*uE;=6gNUb(Mx0#OV<0xBF1+4+Y>=
zI5<HyJ2|1UU@M{XmFNc}o8B5q=hT=%)T}<p93|r0jNAY4-`!j2n0!t%)#ELlt`_x#
z5cj1Bb_cGka|}XjY-_If<_(e<a#N9y#(DmrtS{(26z5E>D%w7Zzd4huF_;x#9%vM%
zk;h~MH!K~d`7V5P^C(8<7u94zl_9CyPP8^vHiMoPdBe<MoTJ#y3`H!uD&TSisfxg@
zc!DW)0|R?V8zi^e7I*vI1B#<8pI&TQO8Iav>Ma>J)h0HrUbwe$j1}ThZ&@RyfynUP
zKff6hS%%V}ib#+x{yV<?$@fdD&sG-&%PhJSzWhNxfXs*~zkx*&@dB9JK^!FtUR~-r
za6leRoU9;sFo9jfR^v5Zg%Y~<1M(^j!t0nl5MvG`f<HHEmR4J~xI#sYF&70DdT2bv
z#f0p@^L;wMQy+m=4zXn_o!KVY;mv8z#<y>#TT-25$G{kzAj%}ADdrZK+nrzdW8}nr
z6@2yV2f6R*PiPzuV-#0m6bJ>Izz-;BG+Og=DDC-EP+U5D{7^}16z)}F8;DF9<DKA-
zwMx;o3Yjev_eqItHC2qGw(-~xut%rT`9>}TskpGWa$+(^C0B-GI+)muLi~X|Ey8|^
zb#-n(C?#ZR(6@6BCry_9W@%Kanh`E#qyRWfKE6cDxY)^OkdkJxOyg<PY$ENLXmo>0
z&v5^a8JlWzxcvk5eoc~!X>v@*Z7F+SO&3d@F;B8;=!`_c9>l&+?~?^r8DWLx+dp(4
z(_$bxuX}}cAFXJemv4V=@A|9)OV0tE)VtLd{MO|W&*6k<6D*ZT*4s2bX>o#DXUs@E
zqM|OYJFiR9n~pz}N&flO67Xk|kb18u{0j7S#C&#qp4uQja#{zG4=Ef+N|^s$yAIZt
zjpT)?-gDOUT}h{@kB<ydInFCnlXA^lMwDS!|Ez)`5hpOCb3M5f{^-g6YclG}Dqf(l
z5m^rWy=!l{z8(Q#jv*zS#t7wwl}JZ86hZLmrgv9VpmMtLd~}A07vhgD7^(CtO!*_n
z-#H?Bnq~@VYUA#<_VoY7#W!t<0w}q*ZQHipyKURHZQHhO+qP}nw(q&K=H<*U)I%z%
zN^<uca-7{gPt&4aib}l5n27h%+>@^K;gMZ$zDm2^Okl>J0VK9i*>4is7t2kCsQnoQ
z#OGpy#Xbw1%BBfG(E5-sX!FiueLPe%?v3&t0;yd28#+={-08MB#Gr?Nxe7&yTjDzB
zSPsO_=tyoj;_kv2izUAh7Xp&Lbd<Q|L8ce}_I}|mIG~nyEPNndbB6hh9@3T&>zLH_
zK>^tq_M?(>W{A+ir#~}y`zYNhVRhq+(BU8s&;-__NHypHo0!;Fh>B+$Qa+vfj|!=o
zF`SJ2qxPRKdEV>{k4)6LPgIruGX$pi+2@oB)SAp#A-@?<2r0_UoCe$3>nI_K+p+MU
zjZib&pO>_KS7B$cGk;Yvgl7SErCX)h;IBiYKEJ}yS+ozmTxiUW-X5lghNh@iF;;!O
z)GjAT6#ml(`-Q$$7N?qkN+vX2FEu7PT&cUKEG|)U#dGsyV|_B%$!37tvMpP-o2n0z
zE@%|*Pu1E}t#qo*gB)y)LlhKVaL}G(rjjn$R6e4;iVLaqYHu={l0}YdG=&yHHyJ9f
zX~aK9io}NK5uB|j)v`gcnY)F8kg6^T909I6K;`SB72ei2rZQU|f&{ogZX*ydf};~$
zjyvv`j}S5xg!?2$6P}RGn*g-V47nmx!zjMcy6{%kgmn}^aGC;uWgA{02^@U`8GEc_
z+<%CSI<O{Sb4j->vy5vopuUeH|K5uCa~_qKWy?Oe<1(&|Gg6}s9)$PyV5ho*15@`r
zZOUu2Whn+bd){+6<P=3j$Nuaxm$LJw^kE){;-BFdW3HUtW34H8?5qU91Y(bCQ0<>Z
zQsQxsL?o9&N+tKA^9_D+YKL|BWPjoemR#!G2=7V3T>MSH*vx)KeVwPh@t{ORkR|4H
z4QuHF&~=^>!2j&P6xXfZhK@(AL|9pJ;U6oX9N_6ow-b}W!R$SSz5OC77-xa_DB`11
z9@;%dbX6>;gS|K13U8g1V6BCSjQF8nZC<5aXgVEQoH{4C@-%9Y?&Zybo-NCN#N#2`
z?I1v0!qyGxkjV7J<j@h}(>50TP&>M=6sN#!8f}~tOSacy`!HjZI0DZLVKP>>ZT~nH
zg(D5E)WFo6IUKjH(C8q^)Q>N?*jn+Pr0u7}Ar^VNFdQ_W`@sS)CU7qEGjV3xRzdi#
z`553M7S}9UaZ&T;20{&o1{f^Jg`vH94?XKw+{1Df@kBvp7@z`ftWelTW5Dxe8|Dbb
zXr+ceCq7U*(UGIc4y1MCr2W-Ux1rTXXwnR-rG3wte$R|9^WweE`?|u~03OCXE|3uc
z@pV2}Pt@gve_t;Fr^rA6ZnNuupkg6)4`&DCQ~i$}8xC7Vw<}(F`&~MYgwieam>ja+
z7PU0{&<lGpZFP}NZ+f&giWkitS^pqil6JYcliAXy;d7okBtwt^??Z6%Q6%P{&b{oD
z9BwBjePqL{Z6^?*1mdveYaTYvb}`{yd{5;LH<$O4m8EQRorHoB0qbQBz05J)$gVS#
zj0u853WupCBw`%7tOj+3R+{Z_7v<lDM#OCSC_KbS9Bq<}*Fl}OBf^jWl?8Z0>=%0=
z%gg|$w45HYmP8w`FZuUFmtYFj4TB6wTEnvl%*r(+-k}?ClxOi4dtT-cGz|&GL!!RF
zJwi)Df$7^Y_d8{`$AOr|L|O78-Kraaq(I6c^+W@P{)vDa*xqj7GI$mncf%L;V(9mR
z@)`k+pBE|-5OKh3#smac#zIU|{sZq;L#%T+>B&j$VXqa%WA^4Z4#z;JY@GfN;zu>V
zY=6d654>XY54s0VZoJWKfTf>|>G;%+{plvujGHe}6N4Tjb2VBJ@zyZ;eP9(%ob>A4
zUY`>&j|cB7i<&dPRQ5dj@#vq|a#Of56~T!{B)%$mT;<djnk~+EU;TpdWs``nyMGEO
z#PqcXcwaTwga$HEV402q3pMgSJXN%Wgn#)TMFL$-?2h;Mt^)PBHvwo{&VF6iWfA$Z
ztd*FFE^Dc*bEVQVVeoR>jrzXoqrc&gI;(Xf1xXu-=36sCr<PAQ`Xsd%deY7i9GMmH
zG)xi~F~>0))#I`+86hXfw=&$yRHPz@slsA8p#aQYlEKM0FB}`Nxt6H~j_DY8G6=2C
zbx?<#0~gQko!|Eb$0bo_Qw7yWTZbpYO^Q{BgW-a{3ibC|$>x`;A*&AzRf!X|_r(xP
zHSVTXKsEDz+e9gb=z;urj79-<w)n~8$qts94cd?T5^ljeC=XHYs<g$6SY*UFH*6!<
zrd>Mzwdxvh<poD;Lu<{LgytZ}XaSs1Ktj4TX`#@-f8}DUFf!G0D%>-Q*S#PiLwsPq
zq!@xN$3X%j{Ve>F&`Z9`POx5Q)Y&Q~P(YlIC@JbMU+Vo*qYqiA(@032+%FMKvo#}W
zQD+R@94&WKVZc9o7H*SqG6LcGnNE;(S@&%!5QewTj&}djl4Jf&aZ{EVD3&N)z=RZ9
zl343S5>ZmoIbzrkrwBqo@DxR)8fylBR<)wLpu@L#VdZ7ypC{>b61UG;@-Ex-jd{HT
z;cNjLh+M@8mSSD(5xm2)<t(`C(d60eyfoh=I%ckuv(5d}ImwL*z+lSUD?uoH2>8&W
z;NS*rMFK~6mQr7=uqqm`2Fe{5pU;|8SEcpoGfO1T!I)~?GZZ5XFp=2ZbH|lTYjlMJ
z^UK@Rpx~ES&C2XXoOqOxjfh*6kdn8T>LGV@RAm60)lnpapPL5d3<C-nn~`fT$Z4Y3
z=lH0=Ifgb6V$1G}S$Dc}X$0a8&WA^I-xw(dNVaM>>74TZ4?cLaj{=`lK_T{0@rDTn
zxPZ8&H<Li5s(stPR9VEsvoINW$eC8F{-UA9ChqQs&A$*ck~J7_n{Y*3;rpen()6_B
zri0UjhCID^{YR=YT3-2no;n4adMjCo`hl7BRDK{=qxd50EiHQcqcvSODK$c;^;X20
z?tC8gIh6b`0J9TnLBboG>eKir#G^)xFb1noSuryk-nq4kS7jW^YmU9;ud@fLIahM~
zZmg^7W_yk!es|!N^qJF$vL;Ek-CQvJZIXp*%X;x1ss3HZ&sE0zRFTB=Q;rz)^I{~5
zR6oJB3VXbEk4Pfn@fm6#M`$TBMGiRSUIYGc=O{@uJ=e=EWD^ETPYqlZr}1ZRZ<?tB
z7W=DtL(5_jBdQOe1>73GAI7WOk|#>K{{EUClG?5bKxuKa=5?c#m$}Yp=ZlhW%NN7N
z_c0_ZWtanDDe=Ngp&7@ojyF<E_$hCiP89_Jr2<j)Y0$zctY*8?1bx4g#CtkQ&{m<6
zc}HGy`v@Q)UKQyB4))%mcEb&5i-Sq?*shZ5Y!9SbKkflS2H=;yh(Je7&=>`YvIL|d
zp-*j#6xW$)CkeycJfNRAe1Xe%s2PC8mD_{*)~O~{9lLFW`w0So99ZtpSEE-H=)zjW
zsEN3zmO;3?S7CBIp1FIJL&|rqT;f?NL{xC>ny0Q&!>krKX3m`Hl{$SdVtUrAe+*jd
zHkEg<a_bQB&Xi_$8fku2X}J2Y6E;;@FNNy?AaPI?b?r|u`Re@K4{#zkTc7Py%|5v?
zTM;rILnk9^?~HBN*U__{H3E7qptKpHZFk3QSY8D|sx$N2)4sD_7MT{{Te}#K!!?}h
zRbuai=XK8kd}2V8bTyN_d2+}?z1T8HTZO;Cy&ghWPo-=f{3)WzT8w8?N_+x=a}NT4
z<o60$8s`jH5Gp#QB2&*CHJaQ>3a|#L>^&}TUKg{q3(>$@oi8I#z}EzGr%Wta9r{_Z
zsY96VX%!Y6eH6Z63s}NF%dNvMkg<HtP6FVDsbRD%Efs|9$?Vug=;Tc--|k&>TW){p
zgvIAX-Nn3%<Y9e3`j5(6GXBh4GuMA+JRxw-z0kHSsJHF6TOK*QN1(mUX0rzq0CdiQ
zTq7McS)yJz;YQBgH1!KC`F@+8aX3FUOT@aCc0!4K#L0)Ce?Y6^p<}>G_0*y;ah&=>
z41bBhLNg|y+0N(2vK3Woe5A(S`KkpTHVV5zF9~T2c^|$O51%N;d6zZ9Q@fzTAro^c
z-xS*_f-SpBOx;~8#js4B1^4;bRcHXp!DZTP+a?g$t_O&ynNe{NvMm8-5}HY_o&AGp
zPd9)m2x?eR&a%z!tCRZ-Co~R>C5_$un2yGg^u}c}wx}};7oWehKe=d<pZK)PbsXd<
z>w(i-yjW@(xzfpvLAqfu>)UB~-MlIp+<&aK{VE$M{H<j4JGee04U_kReRQI{N#<kY
zT*nNAXQXHEN4HtQMT6z&O+!P5GLdLOfkuhn5L1cGzr9k^`_0!#8#<pe@FrF<>_Um_
z%aX!8$7&9CUcfAK0{%u45R;#n2>_n9?K#^WeDn#TIHcI#DCzjDJisKfO3p*acnesM
zkGoC}{ifsN#+;NhHE($*#>U2|%ipzHBj8s0w`Nl)cWqpWrN!ge$NYFJ9fyYi2gVE=
zd5~p!^o~zIHGDS$#J#5|OG-$&0_7ygqcC6%Z?#tyA|{SbZ)^$gM{jpi<NEjMTN_>;
zl+Fo@$%*C?SHZ+!a8z8(qx=M^K_UFY$Rbb+Gp#7W2a7;U?QOjS*)4(CTA3!UgUa{e
ztLJo4U|8X<a@pyI6zoZm7Itb}&CH;HMnaZniis<?L--t+DuYsd&%y5l4n-}eahpRt
zwlBbS!y|fE)b^v>PP8c7TwJNj{*&r7FJM)}#np+i$Lv-I$V(jeG^-7Z1j(`M<Y9b;
z!Q^N*DpD_3+Nm}64(>Sa!^pdH2f5^pWTYIf6L6Wy7Z&Hmqp5>2S~D38spVorO4JR@
zJ2^R<_IM;RP{tzJ+rPkh^3G)7%5bVi;jaj4Xb#A*7Q;{q1savc#Jgl$wvmc_ZKlm?
zQUcqGumUYKD_s^Bw&Ve!DFI!cUlZ3la}=PbD|;D39)ILr`w+=ujQg_y*}sX$OvSgz
zJEdRkp;=tMBH=(7?$^H^@nuInulGVY41nSLrN%C5#7F+oDc`J^#=mmu{-x5dC-v4o
zl{R<AmDXdIm9)V>=sl3F%h7V-n_e&0AJYCQhH<HgL4Z6T_3aVQu`qda-3lX%ObO|i
zG{AEYelDm8@?@S6Pg-Dwwf7b~3!NX@=G$$sFK>P2=!C%2&r!+)vMpvbFJqRJk1TT5
z$uiN545SjJ00Tn}U#ln4RU0s_4;?)I<)~%^D?FjMZQQp?4R6>@=TCZ9Pl~qS(BueA
z1;s?^0qxFMcuU{T>R&+eA9lANB;^KNaD95F0n~y5YF+3mT2&Sjtoj2o!+C)tH0K*(
zy7#SdF%DXseA2a>ZTqA*M+o7Hj$uA>qs3%djKdWhX5CexbYK~x-aZ^5Slp^Wy|C_O
zLk2KI<%l5tP;5OK#h61n5Uz9K;7gHJ=~?1;XY+;9TD`Y+EoFsH>G(uo+(jED0oMk^
zg$)9se#NIf1k4o;jFRyW-Zn0dkAj@{GsTm1HjgAbT=rmF&tZKSh8)J8y!9Z5k5D@%
zKma~~5pz;7t?1_(m$(dcgP%iVAk<PUP30vulrO137#%Jmzqg>QFsz%D#>~R0l)#4n
z2LvrUFs!S#aoF@Cp1quV`k?d=^EJ$K<|!rk-YKPK0sthRG`(iK7MVBX*YIQ0W;l(8
zydYh8f7f`t-_A{~YQF8^oyAlfy5#|%>p}Ys#Ui}x)$g54)nHW5g(ChCw$D3OL_yFI
zbN}$SIuIELu5NZozk(e4tzY8BXU1HQ_d%UwS_1n$8g(JLt#+`d0pG1?+w^!G?3ut&
zX>b)Q(K2GQT*j25chsLfwJ=(b1dR5x6JWSvf%A`uE;*O*9D;S3%k)sLHLB}AMmI`3
zvl47Mi8gt22Qjp{l$!2^F$YwxBTRH7yNSEH><&jR?!5~I4#r$@9ic6x!JwkZW^}RW
zx&3r$sWH*9xrYg!IVV~5FaiHzsck}RUdkHo>>7_AE|)gTe<dInqD;O8#}>xPeB_LK
z_Ew}K50$fnd~_D?JDw+ukCB_YMW)4MMsP2FcLY?Uo!H3a3$iz}^o1J*Gi$S?qh@`n
zOkC0|U4g!dWo6L)K7<E8f4wo1Bgv|cmCaq^1|)lr{Gvo#oO@T*KEum#s;C+V-_H!+
zj*Kr~+3fjr!${rFIm(FG+FMM31PKz>#3m=JY$8oowwRjdnd62PIbhWWH0r=_A5ya^
zC$_2<nersOCVJIef-QVlSx~dUpi+X<buxY52yG(C&GZIIOPinR!f!Kb`LEFK>TV!P
z(3(NETSK;oNWh3i&cKz==Ps1<ymPdM(a-%LGZOziP!&QSn-?q!U;0-=*HIpkD}=1I
z%$Da$egu>X%VmtMkYm!0GreJuX)bNcbQT9o+=_wll9APl*rD-m9Gj@CgDOo0$Ai;8
z)YaomfPKi_qyt%`P(6}S?t899C{TEVuPuKaxD>C5>(S9NNq-VWO!x$9%W{&D3LvRt
zH-t!?e0f^_--SiG06l0g+^Xk0iynJ^BV5#Lp;Ac<=OPS;$~MbvDagzVKh!wn)`|Wv
z2sVUVA(O%Q^*U8L!)F-mFnI#G9aX$Ay{+%Xf1jcF8;_{$cy%rw_7f7t!8^mt<rlB_
zEzb&5O-_45YQAk+ok`q7GSySSl@4W^yZ237O>T5U9c9P6FQ&m2#3!s%AYiEq5Oyb1
zcbiS7$)J#yt!X>8`@j@K6W=o~uwQy4_^mv#&aJKOzbE@-f2d-6RXmwy1w6hm2)<$7
z=V<u}CHnmSC7qeyh1!5DhwdA}@d|-_IW-LKseg{sIA1~tTj9P}EZq<Sujhv$Pzn7l
za)sLg$ck5l-P)(6qz#gG<rD(tkRlQC9L}UN(c8}rrOeOwz${p49XjGT;;cENe4~9$
zNfv~n&;K>bI-x?amE}078}`JJwk1JjTg=K(<`K0VEI$5*5xO?AtY*kw<-zNA_gCcs
zhon-F4Iw|rV!g&o`e)GUO3uV4&o!4Lf{w})pq>iLJ4aqErM!hq!EXu%GJq3Y_-~E)
zWzzCjXZoNg4>lR_qdd062{5Sd;6+`$p6|y3RC@GHXhGcmsSVf4X^OfTzj&i|*q%p1
zWj&!UB>Jt)7s_FyT-=p3)j$_KICCs|wJQf4D7hEL#zev+0v)1n&q`H0wtjGP?R~W@
z!sV2H%4%K+%vLn+?xiQ?6`21RUi|t&st=m(*W$P2<U?4Nf9a0}M2de6`CLT)&3psH
z1xer7RhLzhib@IB$C57qv^H+Sq2}Be;5#_-#>SOe2m1imbmgLqW4a2er1LO<5$1`B
zDl%H2)17k&Ql!J51>rce;7dOW`@ehQ7YwF8xC8#`B&nvLZHA4me6)`^!U)<Vk2nI$
ztHKluX^UvB=2EumHY$4C>7dl7P6DCF4IZ6;CT~`R_+tvjOv04RH0j2QMq(R|C+2<Q
zUjnu1dzOUMZwEG+E>QWAA^OMV{>BTU%SU*{vq9c>>-VAE9I<9RHy2cy_1lVD*X=uQ
zz`KkCu*>KWW;5qVh(&8gS?i^C{P*ifgmJ5D4mkV`cSqB`OZ=6=k@BFca(|pXZS=m&
zd&`2V#fd}8TTTgj=Qlzv{`8pnABgDsM;9JQ5v{o=@3CTZuEE4>@0l(B7U0RR3Hx42
zv|?xSx)y2~&s%obD5|Fkv;;dm%_bEwmIwLXq18mldI)0VC7OYvAFcN66nk>hbiFF+
zERX`XuzmZKJ}f&$XWd9IPZd-g@2JV!dbUl<CPV=?g=UX`g}{k8#8n{=_l%n#x-Nmh
zCGg?<I59<j0B!(K!iV|aioOj^N{D)jbW8s3l`oP%j5+z+N--O)iVF`d-$kht1;7A|
z%^$F$T2*VC&d`?tu(6I7$pI)^qTsh(&epHH%Lzu2mZ55418!yjG{=-@Dj$|Jt3NTe
zHMEaS_woa;38#D|rmlu9S*wrh&F-jGTS|HH2%_Sax)DdaH=jxcSCQ-i5p7e5A|wXm
zlHV*)ch-;NENp4Al>^O6+$3Z>$grjauICjr9E~@+Ylv=DJbyq9L@RP`e{nB-gI2Q{
zuJ>XQk@6u_6jwIH_wS1~$uxFKObOW*)xJCix5b+$yW+Y^Rr?)_D)F8|cf^I#!Qpop
z`+wVClJ2&9Kp_xR)9z?kBXbcB5b%Q3@$Rx|jjBLYqWK3z13EW2sz@vm>+m!w`|XMy
z)NI7Fh&t+0acf?jbRmwjw<8M^8eLt3BNc((7^Q-(V$Ai@XjV=&1L%L5KOGq)|JX|}
zoxe|0mFAKmNM-&pM7(rIpeb!yN52T8n!B>VNJ0{|>H0on2}DkeV8Kc<V$V5-6G|BI
zK6jN%0(!Eg?)y4rGBe3Cr5JcmJ^6v<bQ@p^w3PI;aHT0$DB#nlPyCT+=4c-0ODSfL
zyoF9=N2s}%70CVc)uL`6rpAW_hen6ToSI4wp^^K*oR0+R4QaSLy?jfI=z~o?D$Y!4
z$@XlD;qMkG=qIL=6X>lyPa~dCZnJgT_$+0@DO^+Ir{FUw0KZ9Cc4arT^XMcy0T|E$
zSC7G<#Zdq(UWT#i^qi`Y)>~i$Tx1U%-J%+DT}R9Y?+6HYYUB%2wO;d8ssXv>Jsvb7
zH|LG~Q@)2V!L7*4DZtm2WBVlwpgx8RZQ*+g1flomVZlO9<q#ia!Dh;0DS$3!REpA}
z;OvGe^dHtsH;AX-;H~e{R~V$}oVB7NqP}%?U9XyKL@((GsGV1O{-nu)>|EY2>gP~W
z`Whhzp%{PhEO_5E-}IAWbTsPcowU!=k23R|6Ya;|W)F&LcXPzD;$G#wfL}v=@xrwU
zS5gHoq$%On{n5v)A)qOd%~DU0dWS;+J(`h?9OHsQ1@nBHd%zAYFDu}gE}q!$tF}n_
zu|?+>64^V7+eu9w0V!hOa+<A~5yG43_MyMW{4yS|b@`25Tno#N_6I@08Tse8k<iIv
zW`$*M$jeDne;Zt9-aucp)wi+KsT9WNE15jfn%O<myJnYCzJX|aPN<x@?SqZ-oM&3s
z=uU)nGzYwRHaiX*{)V6SlW&c2B}AoR_WW+5F^1bZml5`qaKY~GoDXv_71hJ<fqR+1
z9NKpqe8<$UMuQpf4+modRfYlLe2DYGInp*t<uB}!Bdc1tYifLUjpE0ItSm?sd7kPh
zRFma6IjRu9A8$eEtt3<`CGKG4o)feHMk7OKhk3I-;G!YORm;J_+gwSsBGbZ@Wbxj_
zADhcl$Q45kSuRfxrWl~>r<7qliFSAsVKa{K0YaQej##rRBw#9dFNqkO79k*%FZ-!d
zx`zH3WL#{4MGcIbNLYMy44nEG2g0fQOB+mOdG*`gm*iy@p(7Wi+3tPgG7lox(OY-T
zQVHvktA%P^%-{;*0-TfW5%;jcH3x(Vo9aG&PFhoi1c;f;wuQ95?=KYWUVNM&8{Uto
z8yM6;0=Qm3bJ>o#;F*aIh%u}1(FPSLVHZUP-v2Gl;bo@-vT3p_SH$|U2H!gh(4<R)
zJB*QsnrUZk!}%pMG5a>q&KROkF;Icrl@Z`Uvl-<wvqTj85x_mQ2cQI(HfssJHobzS
z$Z-4CJ0aVgl099bAQ_Mh%9&>#zL=Vqig&4M%AKTLrSk8!Yx$ObaYm=zHEB<jVrirk
zsm9t91H2|^uUo~z9m*s!yCCnPIG{T-NYKrn#5WS$+$B_SI_GPWf!UQe>5h1EU|4d}
z<ZkoG295X?NtDw|95V=3NDVwlAv7>H3t(<i$COtpNgQkxMewVT*4I(b0|zfj*9yZ0
zhvXV3!T?epr9r4kbW>CWDS-%%{{h3#NBRfth)1k3<1AbWyqd>yY=uNrr!!)C`1F_2
zwnFtITvG7&G7`@8eORU3Fvs72D`eJ`wF-e^jboqx9z5FFEvb+BCOVnhN3>)>0FV2r
zT3V2kCs0Cv?>b?gUwh}~6#`pQ2PL%w?C9#$_y*#Qvphvdut14PNYv~;qA4z&vUd{C
zbRra{Wtw(XY7gWc$EqEpf$);kJW5OFZ`A?I@O?$$zZcFw;%5VqdzS3Vdd5&mm3pU9
zu<I$rV5YCD<Cn~8kv!uEW?{W_wG^xz2u}Kz_zkQ~W@%3#MRE>|%#SacsTDDVVKlPL
zTS!s@;#Z4<I*=yd*6weFOjh?hP)$w_6?N-jT$KHh)b6uOW4xlcY<s9G;jEMCH+B6p
zR>w(iDk?!;o2N%%y)1r*)LZbcva|V<v5+i&yMe-(%-Qvki_YI%+^YF6tsNe<eTNrn
zS3UlP{G{q19Nyt_O-OO$LsfJWX97tk8JsOL`e5_QJI;T3RNS8b1Qm5CAz-6R#m_o8
zbER3?ZJUT2rfgT%I18KN-A<vK9iJxmx%yVLu-o`Z&?BDvmS$7!lCC?}1VO`34L;Z&
zR2(%Vm)vGy6M+uWem5dRbvlOnh4|IdJ;N*Sn4US`r&^RvZCf$KoEgs(vhp-FKKTtu
z?Eo!SSMb@k5ja=&AAG_wFt@>MLr`6IisW2?OUbn3X1VN=L+GSio*VoPvdM(5t`Rec
zKvy6=w5x;37!cGY;MBn%v2Gnu-Zkma5Jynv3&t2JJN-X9x^b)_VtV4?Y+kZZxtf3M
z^J1fuA}*sFDmb~NK_-^V*wfu<N-4C8K`@aNPEDZ9iTeWCo?NNfNxu7a88hV*j7f(e
z=4p0qazV**Ug^==Q1<<)Za#PC!PQ15Ah%St)>nblMpqCcUPo34vx-YHWuNt2ATfqK
z$PkL9n{}q}C5K0x)rpp9QZfxE=d=L_0vJ5fImvkKY%L4TQyjv%@K}wFt%3aYLE}E3
za*NAd@0_^tzH#1Fz<5KolAcIwNKB=3eX4ZlOvJ2Marv3ZW02M421okR9h@gF!mTi)
zAnX6WQ(g=Zk~d)O`0L3$2ud1BiOudd-du;CPW*~^@rk57opM$WKLFTtsa5FYW@|~B
z{`l-a@xR*jw!wV6s#Sh~+ob@9Lnii?{lUQaaRXU)`aQfJ166My$09IwTufp75h}`N
z>pi<Ul|Ra++}txBZT?8kVe$#a6j^?()dbK{>`v)UlPH(EQ(^J?16aErJDlf}P|XT(
ztQ59UWn1(9N#%U{2ah1}`wCe2dqwgaeZ^KfG-Y`{ZmS-DDMEczK+;j-8~EWxPPaju
z_Z$H|tMM&0HwMTm$Lo<!FiaaUh%n`loCt8$Ma6hHf?kb(450)kLuXm!%RNnN#8;%3
zG2DS}gRt+XCe524$CGUS`v*YS-&sT+r_Eh!`Nh$uwi%$B2^VJHhqJ}->7=vVd{Aee
zNwkNURg#W%?M`R<NT)l~D=q~~n6=R=Mcjj3--qu=-zj*q#|vhNQ6PtvgnHK4Cq!3%
zw%kCR8WB_89UrE+nyCJ1ftAsScw0s6(_5+{W#403JLLoIb)%IZ=w=Or5R^m!_$TTc
z=EWuECABeQb^v2QoWHIhr6OzOCzkPo9V({BGfeeG0+|@ki*cZMF|ye>>-p-p!}iX3
z_DlELoN0zJyc_iU7b+CoH!t3HId!nfCNa)uU8>UIdcp&ig#7V4zbFL=fqbWe_XrZ2
zBnXKQ(-5hZJniTbnDP>Zi?%YEKh4)~Z-T6s$AN+!B?a6N)~Zps%R(3a%U(-V+m-Xk
zTfZ_0ObbI$eWaJrf6%RSCc|&5oBl=FF@ppj78cN#or*r?Ua3L8IdFbb?@XNiRA4$Z
z)cHOxcI+ljq00r|>t_N)F;Wn}LsQ-qN1m4stM+mb)h901{}5YX7$EQ-5oh7+QPU%4
zMj>m&rRW~&pwNJ`<NlB~O5o_=`J-{1(xV=SwSn>~KkDdnoFB9A*OW{-Kb+Rx!>NRt
zVYpw9qw^_|&D<W8Mgb|dvASAx*7TdzS99srQgpHxJc>w030(sbvNCu3Lw^kOPLJh8
zo>>1tVcbX~trvLkZ?VGD6oh8}{=0l}B)H}{1SCV*t<ZOlY}>s&XSPH29kdhF4rD?B
zFw2_%vC4$zx91?*zdWy;o+TY#n|`{=Mbh<G?+vg(oWqwin*nQzW!?)yG@s}<Y;4q!
zDo!e*_#tU^%I<m;%Iaa9hi`fu!tb41MT!!frChX{1zMWn`6A#;@w}I69a#FfcKW9=
z#G{Qf)8EUADh?~T0xL@Dd`j(5R^mgd4I_J&SL{9p{>pug9Qwr|a4K2;_p9%Yd<y(v
zY|?nbjl@>;2gYB|5Og1!w&OH!j9=vj4QO3G^5HUoo35j=e9@XdP1nx;>mmgR`%ekn
zeWRLFWFL)1)p_gLD$Ra``T^1BS9Q$vc54p8LLaWMHgIX9g}poZX&@(rkaa-D6m-3(
zd55N-h>9{Rc4fyjG(be?Pn!AR??LSpaZAri%}^+s1#Ny$1vN2n&KOFO^tk`8i&l=y
z+(ya?QXsZR$PWhDnnmr`8PsvId0~Cj%`3^^IgSc~cx?J>(2BK>oTshqP*x#1MKVs&
zmDTOcZkcl}_v25K7MuL#3qfLNLI!`MdVeN7yCks36S4!0uWid=<h&WXUlx=t?;(n%
z3k_2S7If&~Zii;BPsoJOWBMs=V5Bt=J%!koJRw~z0PhC`=UQ2!#s}`xV5D&9>_M`0
zUqJGL!dY&jJo|Ueol+-@jq2Xq7zG(+Ry>75Y-CUvRC1QquV7P3ib|~Cyy#Vh%jZ+D
z6}CK_0{u1i6618=R2fM%s_DY-NwuMyxoo@<6p^kie1yO|fLt7Uy<I)3(ODnoG`I~(
zVG03h(u5OT$6zY;<;U5kA097-Kl>tVt*9Dt&DIA+t%*>f(|V*t=LGUN8{VU`6jy{Q
z8LSDHB`UA+GC1WL?%EOLreiv?Af9rdV*6;v^TfTs8B`L=m%5*r!WT64d<+Purd<Dr
z<Q~wezWsj2iptKMLVHoeEnkZ>Nq{<4z7bDVfaGn0S%xVwjbsjyC>nNDNG#1Al_@}s
zy8O#v9XMfk)XoE;9747I1w`5N606{wK#)RHh~b2b*Yb|kxfaR&r6m>fmaSl1s|}Mr
ztc_}Jck2qhhO1x!2ZDL^wtK43WW*l|m7}Nr!eb<{?^Fg6>We8nYVts@b>?NMGUtUH
zDM=u<TbP_Fh9freOU`&74z{7;fWPX4KTh|*P38!!+de9+F*CyoaQzOX{l*Z&R>!4*
z*vSSjT9P?@_EG=NLNwgjk4FP3BnW0A4Hf$+P>}YSThx=3gK6Gw%WUa!{Y2L_FUH^^
z&8*HRjNo-hD3jzeUu`Y9)R4I_t3f)Z)%<V-NnT%GXd<c<cq_TH_HuoF+*EaMS<4>2
zqk@`7NO!sx^OKy@2<baxvY!c<^_jrZ<~U8*x6QtffaLf^EDgosd&96JEQsBRmDKtQ
zd)5YUT^kQ(onIvZzpQTd=<FW-WD>-DV#(?|#l!`|zX9u6azWJU$FMN>1}^G>*sY;i
zm+K8-9KEKnhD`GgmW^$85Uq3|&NNONMkD965g*J$hqaWCFHQ2acSxsmi?Hz@VBRku
zkkGb3b_VaEs&g9V)82|n=|O(doDr=v{)odnHc^b5{&p!)OFJiTcA7z4m9oaXst)%h
ze*GWRLo@U}c&Zy)xximKc6@>B!0XT}LE3>H8UeX<SiLmphOKTWKwuy54_m`jkX0kh
z{TTu3)RJDPu5_G=2s<(D;8*REgN8oHmI#X)743>k|4mSoIzH6{2=ThbgIyWn>9Rjl
zz#tJ1?T|$ZZ2UWlA>m~mX6gWydjm-NsR>i;{Rf&7p?+ZbNul`?Dj+06@e%5^8%2%N
zZaNJAbV`WWyc|a!n2<ed`k=pE6Dc^tpv3Z7ZLHQZLi+*`?HME#^vJHcGoOZLgz3eO
zO^|;BmmE+nr~n#1K;^J`*iiBC_Qku}<J2qHqDn?ya+hkDNb0wkZ(-qJ=XHBp&C^nv
zm;zPU8H*Ht)3NQX?=x-96}+_O-In)%r&?Hx&jInOOF}6MwY;ddy<4WvX<E;Myv}ig
zbuNj23vU&c{)tlOkuT0{e|HtCR_;l(PP|iD$5V0Zlt9zJ@U0VXzlFyFaZ|r9btUkc
zBZR6@qM2?ui4)za{n`uvYX2lv{Aan4igE?7wJtKwq9cmO_a0fYYBlXfWodZ49lIR|
z;=$S}Gh_m<3>3Dmx8mN3e!<N>s}#CN)z4@uQfbMKoTSV9*sW#f?R0H3q(<OK|1Occ
z^FJk<Ztvk8?pm^-?!9(9ttcQ6p>3~5P(GxHI;*3b-I<9lm6=WK`!~eal53;s^R$21
z8d1hS=kG!gF{$Wc$bk0Lkrh<cr8<~4!Iu5o7a2UR*mvOxld!4y_i1oj=4uV9di!*w
zOug4{p~qDz*RTIxkEoTbay=b+soTRJ`Iue2YhyLkr5*<)2p<Z`<MRnZwWlO3U5wc-
zVhS&=TYo<Ad7ZUEBSmn%rlRb3QPqpF@opmFM@gDxuAT1G7zfiNM(NJ}qO;M&)c7_6
zNG~WED%RB~apvIia)!B1Cdo~Rb^f^qxZIrp(W;B%=}D1I%5#yN1C``;wNfYa6(m=9
zB>sjGmD872l<JA;>xdlp*@b)s5*tb>K)$&>VB`!idhuU)w?EeZk<cZBg8wo!#@Wn8
zV*;#7cyY8D+PL1JisTr2=9&~RaaW)Yz&q;7m><Mzb16-^qyt^rpuZ#|13GSUtw0&g
zrIi8E;Z&_)r28VZMEkuDuCUx_0iPSNkL}v~Ck1oM^S8(O)8rFHve_aw89;4pG78mJ
z>F@78LS|;pPK^kRO<%Uiznc^n-<bAIO?g3+VJ$E~%O41u<+dNkvZn`<Ev((pEzbdP
zfWT(z?sc8D-vOeVzx~jOyOWsb9o)P+<Sg$y<tDFK<gBL5M%t!X6fHW|-fp5zgwQoP
zM}4ROe#Skw@%s91rtg6H5dlu5?oP2gh*uwX)AE|QYWeL;S(-L5@wW+L%L+-^1D#3_
zHY~vaEuvD&5Se4a=lO9j+Or<WjA}pT?y$m6^K5dhvch#Kl%j%&1xbeirCCnDiJDPX
zKFMVI8&=XH=otd|pJq5dKqCwM8|LOeARBZsgeRi4ZF2{MXU!j)WxeFV)q<@aaF5D0
zoK{x84l@jwm4gi&>ic<n9iCxx9xDoH6;MfMBWW%+WTD#@YE!|f+L&06oHWVMc%j1A
zH%9oWOKnHIijnhA!0wJ{2H=-nj(ecYi3t6`u(EeEU7mOc6Lg43LQ@iXIT~`t!R+V!
zZL}l_Moy8m#u03?8vEOXhf`L5SCJkgQM{19E->k&NeK%i?Mk-9JQ?c+tMF}_%B+*9
zS6L$KedZgQzuF2@PUc~7<#$%l;_N<)i#c>tZypfwovnpDuPALs!8}z*;P^!kfgmXs
z7c6;4Z3oT0<DxoixHbq}Fcrw-8G+2r-p|N}Ksso){!4p$D=hBCSaHRxDW&XZf4)^e
zE%4&$%TiuBum_kFKp{CxXYTw2S}D-WuETou2ehlN$Pke)QcrpgA)#%Ph7V*79Z*B)
z>QTr*v9xEB<@s}i$%H_?N`HB<`TXgi!WLn}G9Sjo?tZ}Z_MpXAEQvAkxy<lI=AFpl
z_d?l6v}U0Xp*p_bPaQAT@f{X2sTgw%JH+-coMc<tZ0j>VvOgqdR2lgA0M0O?2n!1M
zJ3rY6OHn6H?*+k!<o0!Z5qh%ievstLOP|VX;<07uPjtpEdxr(^e&wKIkf!C@?;Rng
z6*u#g5LIoak<``}X8i)#MDUG+gX>If2-kpz6^Q$83bdy<LT(z<h<46l13tZcO6MxD
z#1~WztaKM78WqArI^tz+vFf78Y8|vuHxSm(;<4Pr6{r10k0fRCNmL=YkaKQ+tuDD(
z@EL<c)-yHt>b@2(z;L9`-8+KHGT4t%tfe|TZzv<5{($la>2iav1DJ1sY_5YE{zF*p
zBr`CqSV8Y>zFLP~0~MT`j^I*sC^>0-<~=^<%ZmkTx|x`aq9SxhMG9u@C`;O3i~9HP
zTutYj^c9kBjd;4XDNAqFg%TDY6_;-ln!9suqN0g0Vs>>E_&k}u0i*#1m5eO{HNnqC
z$inm}RvhbwUeID0)LYTs2Htv@j2&nqEL|ek@U=|iDo4PPholU9VdwshRJ*Gg%i|4d
z!*~se8!{i?i77Ue9En)U0~*`UfyjV5`8jkCU&V;LD>Ty%nOhGNx&d@(J#DCpMixp(
zC?a;>W%j*26=apQU&lo59>B}e{Edrh!Em}bU9DV7T39Syh5-=E8VtuLtoIW~+=`3v
zX{2cXFDwO};t!Tfpg-8To(qtDEu1M29QY}4lx0g(cIZxjSBKlyvKt^ebP6y8SAx1U
zU$nl&XSTxGjk$*h_6;$3?`POFlhz%hLI*UmJ6!fm<V~M-SGO_`oCP2+&NC-Y%WLBp
z!_W5Aq8uVEMsUh%^d{>MHCA>`1|TVX`g9M}I^Eni;Kkd_@GtxnjlA;F0~NgE|A*nO
zvY|CeppjswisQF6*gK%VBy6q0U+uXifsP%@fq-Ow!}`8qYB5C8fO-$jYDzQN1y}Uk
z8(#(`dd-<2PC6@)96IDa)U^!Qaljtd*}>6e672f~6LW6!Q+BrkyGU=WGbVa~*7d#$
z)nbL{e%5gV@+zZ>t$WsQHn-9$0|^;|E?=|eQ*TWChpMUr7kQYPt^whkU`CB*vNp;x
z%Y6f<-!}ifh$MHf$@cwmBshj~@1|MMsM<>=9LWwaRLOX2K}@F46Y6*ThL_+%dvbA2
z6@1J7ZRn=wN!5vJ>CsSBq;{Wr(kPT>=szJ(@U^Ju{WFi-(Q9X4-lA_Oz*{EF;B)kx
znFjuY9Jpbsf#83!TQ<_sg$lBjYYL5HjFZ$5m*ONjA0DwTyh(Z?Rvh%83dj=z0BdeI
zy*YyK!U|D2YK6sDLEF>Szt^}&>F66EUqLpG4#r^+*o&5~{xe*so?r^FBfwLy0+0R%
z+AHZ|<o*MHM#EMW%afwXv!pK@=EDH;G3MG07qmVhKZ!%RX^Wp07PGThmgmg4)nhl0
zA@1A{;1r%jO<qYT8hRITIIIyo>>|qHrwHst*UkyqU26Dj!4aX=Bd6PSr_L!7zc_XB
z(tUMnW-12%=936RkQ(}X@tE(l9J_i=^Z8^uTIk0n{>;3R*AoaHF~S&fQ%r1u74Xgg
zK#pSgOAjL}!u3YHQU5yDZ2u)V@ON`493P84lHY)I@C2#Y0|AwtAOo@~)KMrH7=ec`
z-H;SSlRH@*OH)Y-F6A#Y4D}fA2t#61YMz6E2n$6E5iD4JqwJDQ3l`;zud!4-xV~_Z
zlc|*7kh7S;%~?MH!dIB-(75>9_v@QMrq&YhC~Bo}=cVa&a!9iei)WCAQl|Ays!^uu
z!BiPEU%RvY8lQs{#4<MWn^T@66dxYuTOEP(;_ZpLPH_s|-uee9$3s>H(CM8M)VMPB
zJ7bbFf{LhlIb)EnyBFa7hCz$p*c94!c!3mMd4Uy-a`jsdxyIXNc8};;QK^GHD2m$_
zlw?p?EMppS<LacdztOT%nI|~F3_agn`Ofa>cf8@l&=l(4?dR71v^MbAHRt+yQ1vw<
zny46`L>KJ96sP(~38$<o*9%NGrf2wuT#3F+8ZH4(KvvvrHpe{z`jmTn`gn*5;X(2A
zj7#4RmZ?!~AcKDqJaX+_bV)nz&#VitxpiX|j6@Vg=`i?bO$aPjUDp*q3hOJ={q)C3
zm8~^boRYw_F>u?yDJDZm?b4B@j`{Etldso~UWxIkplqXs&FScNSk)f__hez`^&icY
z1<2jLRw^YufgCzqP39fAbV&y+@g;I+^$8a=-6?Vs+K><43tMB~${crsfL4rq55Hpo
zpFqfXu98@d6=D<bCJKbGrru1xThrcKZ^EjnF|DfvRn&SC3BE;}F<Jpw?|tx(qhGNi
zkBR>oqprM`@`YACqtG4Po-b4iZIgo1hY~_?2yX1?iX;6D=@@sg^_U4pdI_M`@k0<W
zd$a^6qA&CZBq)v*yp7L}1l0NGdN!Oq9YW4dzy&O$n-XZ+sqIJa0+qFG5s=1TqgLhq
zZP6HwR@|g1oXVx3E0MWj+=}r?^w}H=o&G;ut<^62jsy6ohj{$$xG@L_5^ya0e|zT#
z0IKQZ^e#*G?P9ZlI9)q*PKj=zdJX8XTvB)jWSsVENzXV;e;8XpN;?I}`mMe2xh=15
zaSyqLMwr0CVbeDQEMl}oW&t3k_WH!sDmXlM0=u!~`%`k&D*fXS58B!_)T&AB8Gs&j
zyW7G5ifyYqsg*xto>6BlV+TD>|AN*f6Y)7CKKz@gdlL8UTk*?_LDej?{uu#60hoY`
zM(t`OOM%xh#7CR#q}&8Ky_MAjYj9Ql0@K`;c?`e!9tpc(_ouTE>5m3q$_==KJSSA4
zI7=p}<7mnDqb(*6WYU@*#2qHucL*qfB-J_>(i)GUp$Gy)R?AC7ByvnrvKa;QQJ4_@
z#?D+{U)mcs5w&}chqLJa@GZiot)hq|*4Y0woe{P5ZP1J&S4-XKO?WKeAoi9wtlxEH
zrf9Z1s5rl30DOsVwW1Xh^t<=76AXyyw*o2YU<J$L*%tp1`y89OPEWA6Q93KSTLGms
znyeH7J@OkWpObN?x(SMUw8QmFo*U)Me7%yELqFz=k?mzuUaxRA2ckD`t@#TCbQbc-
zd1CgnS^6k1Tg>vk(z$%+JNe_-ddDnJrfAke(l`#DByQ4k+$gThkD})B4(pA1{gCU~
zxz9&@pq+}ZMt!@wSk=W-$Cv@cNf-#vGZAajJFxY<&+?eiZG6nM))S*=1#bMVQow|8
zv@vDg=&f&MB!x$V3*ujz8e*xUCgg849TP8x4x^lBGv=fDBsIV{EjPP7)_dQD1HX_~
zT=SQsP4}dQcuWN5N)caOxr$g>>`ZcB8$s-nA}j7Cfa0FPzv&Gh+#a_hF;jo)Ol4HK
zcz%U?#p9dLF>J;7YocrI&+M$%GBwnN+61~Q21(n^NZAP!NYInc*y2S}oV6RvzkH4>
zk7-RqY7wDe_S-rhgMRo(<O^dwra|^nhfa}CJGe^DsJ5<=JQo8~$cChn!9j^#LK=s5
z8;pF*D;ic;szsNmhqh{#%0R*6Kck4v-RxZvc6t`=rRxib^P*DZk`n+CTxb-mm$RYl
z-t~yqJB8-6>PI!chVg{{W}DhODnY#{URi~CC3wB7bYpTO@>3_8ekImLv1*hK*SWNS
zuQM-KMzZy~y#z0(@{StKv1Uz9(#-6&<Z8;~(qyl8A|4nm_Q%FGl;fHjRZ9VdGbdWb
zyQ~^`wsQp16*Qa7^cV}2_izu_uY^-Pe(n3-4`8-K>5lbRJ#PZDwya~Z_GxA+puY8O
zM5i6iS^%>|zcqh=l?NI2>ubc=ac)J9^Xd&uNOQ{bHrO)lxM~KPY}ytMH44#85ingo
zOXCT@)t`;L+Oi*qABRb_!_px~#4k?WEk+25rG@n%$*=!PD8NExYaK~1U~}l<cf&Gt
zB;f(p5@2>(24#v^G!eroYU8tLgTjhEUslMZ)OtYAEBB;fV#M>@o^4ffA|5=l2dNQR
zW5Y(ylsS!jDf3yNBH5!0^KF^;xY^#~m12F)*t<=2?{+Tp^LbCn9s0|u4<X@bySi$K
z<5l7VI}A!Y?O8~UT~)SGt?ixi0nlmREW_7MYa37}-4E~gE{o_TW#U-mbHXR#sJh~2
z$O{DC^L1gnNwYT`B}jeg!>bSfyL#PKJ--lj74-EqP`>Sqyw>46S}*q<TNni20|QpI
zO|J5Epl~KlaKiCd+A`H8Rc+%};ZMBKe8<K$kJ(h8GCD%dP+$@?Kfx)Z8zgUpg8v5N
zD;evV%QzX0%<ZabTE(o_M1k>vWyoO1vn->0P>-E40>zw>o^r87YE=bBx+nxkz+Mm_
zt09cBPk)Hq+02a~(_5d%|Bg&V3dKSX2O9MVh<@LTz*G9b<l<X_K2a=o^*od2#oV>q
z4I*I$(N!l4kcaj2M;@u0bH9$H5?#j5W~tddBdHHjir0h<W)z1^u$0Pa7+I)pk4vwK
zUifFSwVE1>g%PKN+vU}?=crR;M2PdOA`>Y(cyyO)b!EUNN}2{s>Nhu_3rDzA6D(;Z
zKHKXN-#e?(GaHceu*o5O;c@dO^f;z&slE<fd+EACh26M8Of0up7C9T!m(GheR)YS`
z<EPlRqKCep-K2`s&YFfI!+8D5GTGVX#sAsXw`JUc6w|CmUSs4f;}WDeCs(jyZQqYT
z@YZ~n?Q+uq9ih#m=;p$weD4(+B_Tc}FM|!pXLK((v;XBmggf`)X@xO^Tm*EhtohJ9
zqDYOJ@0&X(cMMKK#AdeP?QzN>=tV^Yf~w~&+@P$IQ((G%|CEH8nL@q-N3(T}ZMjf0
zcykpBqOh-7iOmg?fCXm71^dN^aEUDkv+yj@%juk*QgTJ|=8LKf7S1f9TFE#b^kh<{
z&tv6sY#jgfDz3nwCdhQhieM<F^Ka_ci!!lw?DbfKLyH*y`4rIsnGwt!7fI&qv~YY8
zUm(@mrbC}~MzQ@Z;xqT7Ukas*u?ap$g*?+crG@|l_yjA8g_(?2B<a&q>jrq0fnpHC
zuD=k?zBU7GIBtfCem=$$r=C7XPvmzls6K@y$%SqKe$l{uTB<b*B6BfA`DGso9HXM9
z!;_BWqe{U^d3=OxHA4U^&mThBtX%Jdx%=)YVq@58KT*q^V}CFkZ6DZTIY;aVwmR{)
zi`XzuNk|SwiG83y8WW}7TyBasZOQNc8}*=2Ec4}9C;g0BZ70R7N~ia&2|mRKrUW>c
z4QTtQO1p|*Mfr!JX|jv=@wR284nx)Ei0c2;Ewc)7cbU2q7PE3l#lTEb5->=7X%^Jq
zt0<ZA;;M_;Vw`GNWp;7jC{xuE?Z2)}xnzC2Pk@h3Afqynz$pjqqp_n>SBM{6i)1Hs
z%!@KfcO7-2UzhMooY6YCIve_|j1GF35%hOMJis!WjVgGz{UBvd<wEU)aeLdx!AmT3
zgSM1-%akgfM<Z0E?#4TWXsHFw9Y|(fU|W1dJJsHt7VVK=p91~2J@&+>S<*7&<6sGv
zl#MvjKzSOV91Y6+-?hz|DW^QHLzR+bO4B(9^xs4aSy^E=b~k(rSz<l$@coa`pkG@a
zW;&qlF80U2nkI3CI?Nb1CZBMaw6%Am!`AGRPSRw1FnJoTistV<<Q&wf?7}af!&;4#
zOE(I#?P=9=FEEl9&k6V|%zePb8wVHmHz1#;baGNQk#|SuQAeG9+c30dydXu!ddxeY
z9UE-iLJPDh=n<(pW(lMYwqg;CLKLd<)bRkI#}ggOVAJw>Nv)8Q6z!}2tr{=24*x^J
zOK1oP6n?MJ3_fyhkb+#dLyP5aiHwFRpcaxuh-Sg`$de7x8Xd4lXugi)k{1-wXLX1D
z*d@!`WAfWNULGkT$Cu7yBMh;*hMatia%Ffb!Vsv$OU_myHlrZ_ZmZ__FRK+X7-gBN
zzlGlA9L{0tnhjx0T|oY9BIrm{kux$gdnZIS6fihVGc&Y7_W*0BB!Ko$t{p9<U+boM
zcA-w4oJa1iCw|I>CTuqE&^q;2Lv(f?5qB}W=?p>14w}FwRPA5VU%#$EPe?NH!DwM&
zaWT>2Aek^j{6l7WD~uxP0tgzblJK)wHa)c_23M;gO3F~b>r1dRruf55m|HCE?M%$`
z*6X`%7&XTLk&Vnq@!+$Grh-0`1-J~I8~9Q;JYJf86wgLouYbTOMW?}b!rvkd{0+xc
zHvf;Sb5Ifn(3SMqwr$(CZQHhO+qP}{AKSKV&z@6TyH)qTL#mQay1#T<lfiEHCD>P2
zTSW)sg|@D0hL%ilx2zf}OcQL#QP}%7s-zZxC7`7N4bSZ%uFyq7A>CZVqj~N@Ai-qR
zAv^}Gc2o%Gt|#?X438!C0g<Ee$7I(+ZyFA3c5XL1GDsgC!QuKQ+%i1GSeiAVrC1iu
zKSQQcNL>)#B-LYh2LyJ8L>gs_Ru|$JH3=L~f)f$2!ClF(u3$+sB|hSl0=AZEghl2<
z*qsnJId&y_=0wnx5=kteQdKrr$Nz?!h}INkNu2QDL!HlQj`1i7W0@X%ybd1)F)C;I
zNg_^79$~fYJabZe)e1t4BX?Q<;8cZg@X_w469H;0R1gOjk3{S9Z_k*9=|T}rks!x~
z3J`ei@!Z>2)d|hN%A0CZS$<OXAwc>C)BE-T#N5w;ln&(eJs85(<vzNY1DSru*Z*=}
z8kTS`OTE({(vzgszg;@rfz{t%qv0%$T%#J!$o1fi^tKW}?c^2iwE@S1I*_I*4`!^`
z0CFWoSY7N>7o%raw$u@l=FEt<pHNq#c&wd1kCM0~KoIvZ%H`1oTjPwB+5CGe{>Pwt
z8l>LfP+1du&HE~_G%`>wA!OTU81LU6<Uo2sh)O*4fnh+qYWBPLlJ6BZm<L8OHoc0t
zNt69xBPYhnT4m04psxnIcOO??gAhp}k^}l@DO6~O)%=MHX1xLLb5zrOufl;G6m#W0
zaIR3P%q-O8*^g0!fk0ZH><*x1Z9mtGxh1Rxojjcxfxib?$)3kaQm*_;&JUG;wVxAL
z5oSm~IKR6I7W1l&IP9`sz;J0OPtDIk!^hWh0v_<+4_B7KGa-(c`wT0e8mSa7zu|ge
zNr>k{3i{kD^ZjkXKOh2l7pz4>Pl(=^;Ckcv4MW{oA!SJo6lotwyfg+MTP)}#B&W{A
z_VnHNZ>hV9q8im+4<gMrQ%D;_JDDUNt2ED!1XqZad7mX_R*|N$ZmeNvoqX2uFQcEi
zY+yAw+Z&yYl&n{pZP0~}@fPDl9>1{AnoptLUoGDW&8vy!{dV`q*xQ=PVNQqOMP2Qs
z*>V@^$G*~X8^3j%#iT9J1*KN}=A){wlEbLj5=-%g(NL(lH=edFd4fsj?s$BSwyRq)
zB{Iyi6-e*JaML{K#hlN};OH<2npXZS#Clhaj49z(T2nFxAkhNRRTB(b<BJw=WP}tS
ztnYi5fm0$lPK>rQ4NQhkY*H{>tK7tPGj~|<O>G0mTNL2TGE4|WWKk`TB7=-_an$T)
zVqqm)^k}2K`eFnj>vV(snuxH3Mv><!cZlAxe>*UO$NK(;)2e?-5&EClM4?9>yv!5T
zWguJ8YhZ5<4O#1vs~rBi?_Z^Ok~tzDd!-@!K`HwJ_@mX0!@1s03_QtMkMwf|)(B^;
zFY9G*L3+rvQzX;Qj9A@qBv}cDqbpIdplxQKy+37#2}d|ZwFppN)%`g)EeY<C9zgF9
zl2KF|;^D|Cif@T*XJXqv=2SL$*opMb5fI-ufb)@pWkkuvF=y-vNT|WcRpETNU55Xl
zqwEOkx`3K`H5LS@ODoCdk@PjzmjN~iolE7C3ogi6mK?E&NziThrxRDPRG}!vUsT?3
z$d#D{WfuNs!*C_$N|><yVkZr1Kb4b!dUtqAz?7Ht4rsYlq;`39!ER-K7WGx!=jWnL
z6)x+OuZu0WDe}FI12lczU}Vu0#N%Xlq%N>!=#ulV*XoY8Y~QXix3bEiXDoCKLBb3j
zGTr?KF18f>affNkN2iQ!^l<chJ-#Pn-a|E$w9-!h_T)&aFtVw}wH0jL+%I$`gc|+u
zA}H}-syLP9cWeb`bK8>UCAFG^r+AM=PSs4VmB2yM!PzxSV#X{-(bFxB^$H}Waw?2M
zRvJNkli0zD3pp9QeoZpUMa<W(lucEiOKNtQU%NI%;GvEIr>=B<2Xxoy0KXPo;Jx`a
z%JWZu@U?bcVI7V!!@`DFFkp#zlBNOMZ;e$0L|6;oWcyz|yIQyQJ=3vJe}ZSw2%05K
z8*$=j4j$EDj$O!THW1pXNXXx;uD#6&L49-aS=TeR-(?kuh?La;gYkkc1Z(?8LLxJ#
zS;%F2B*Zz0WmkyxFT3{pKMU;TNd3=Ixw=eH;P5s!3OZH-z>3PxaTFfr@@*$LVDp+;
z5G5Q%0=c8MCl#{XPvy-ccV66F2=lx7-_s;Gnc1RztySaEkzMUR25Bk6SXY1*004BK
zfRuFA%IVc~e=x0v95T~CKxq+A3hBQVsLZ9OIjTbN8=w!F^2RpN(*XC7C6?nPHrb3~
zpxiQ*-|>E^CVtU=6{v*r1ejOE=&$h|wCDIV>Qv&RbAyoY;5h9}J#+kCds{T~8+pVB
z_@<=cS>D*g-eghStgXK0c)l$4a@xPME-;;*#Tsj-!KC~(mp-5svJ(;{+YW)Q1S}I=
z&MLz_WL|8dH+i^tPV}-kS?m#82omu*y(dIBMMv_SgKj1^Xg{(jtqu8r?2qR_K~q7;
z7fx?@GinlL#fn+o<0ravgGg^YL~^r&=;i4X<#F*5hpAu0(j3SrLb##$8!0{p@8skF
z?P=H$_}x4e)iC`AxYG|~OxC~-(08S0n{nv+C74QaqrpJ6XWHh&<;1YtYqBpMqQ+Ox
z+%0k$ZFyM%9l`%s)ddu^>(eM=%$bt)H-p?lqqt1|WQE27UF60B?`sq+H%7j9-(kFe
z!hhI!%Mr;E?kqdcG%OgY{o4Vnr6W^(YWWJzJ4a~wmn_F!JEy+?$Y3o3RR#PzWlrHt
z7>#DI?B6W-2~>jTTvF6F9_t8<d30ZzX8xh<Tq<VyCo<mo?7y~PGE#p`6#|=I>Rl@N
z)p8A1M}!8&RkmL10wK;sh_dJ*fytbJ?`@*wn@vGX2_d^~v)xXA=w^>Pe1KDI@vvHI
zHY(A&TH=Nk^JyExRdokQDe(Q;^<s&kA0Fc+XL6k1DsBw1jLm*YldHMk@0eiNEw)_O
z3A~^Di2vdiFl*lbNsoDqM>J$8_s!@hU|Y|fM*ru2kRxb$XPu?tCLYQHNWD=z!OH!~
z-LJ|+dpNtz>A98s9ypY0Bc$;17zA#P>-C<*Fsq>k0pBcszB+iCqcYVXQyY{zfX@&+
zAJ;a+wI{~O2kH^MW!@e?#AW+5f99LuC}y?8q??9?0~!9@wJwxIp0e8Z(Q5t3l+Siy
z9L{!LmHZ^49}Ef*nYnFnTuuEHkv5I~e+^O9jBsBdLSob@b<!FF^V7B+G2`G%H%1%g
z9pZaUE`4?EQI*XW2`GCw(hbVE^2%V5Xx_fFWRsaW_HZ>wKOsDkCBMCI(Tudn7?Z?G
z-*opLt7i5igZ{7kCBvdq)-{V%3bHc-edX{5&a+xG&R3f~lrJb&XFQqqBC#v@ELvvb
z4qgz}1>6wPQ$To>f1J<Eh>#f(W>m3bLHW{=#>B-YT{~z-RU{rUw9EGBfWN>lnFHfL
znN5*CPtpz(=TI)Uv^#rXc*YIyYA#)_4*1j8PrSm<KK2Tl_;JV>`GQV&S03W!PeQd7
z3fR&9EG_>gA@n^WS-$^K263)}4;_|Qs^rh0-zTVXQe1gsu7KQr+&ehkef}XfQoT)X
z_BJ07?At!$)v-mEtD1?TWwE}99>J0mA4P;<^1s<71mQZ-Y1V*vX$zh+V<&QlBLd+2
z$0v7lM<@y(m=3Kc<LrHN!?_Fa=+1uAFvytV*V6Wo>G&YVANJHQ^{cmFGdw;AyL%b=
zo=mFhoajha>Fg<R&o{&H&R;J4bR8ZYPx>G0SxDAlp!<?T(u<G0j929MMQ!|H!e$YI
zV#`A4&m)Evi4ze_?kg$!Tsb;=j~p_IX+qqE#5aS<Fsp>T8<Q#S2%8YBksyP4W{2i9
zFWu<$M(1=$NFBHF{cyspJm;9hqKg;P#Xz~Ay-)jv(A<CrGJ1GxGj+)>bZUI+)xB!#
z=U}w2k-B0Kj{okTHW0myF8)}fw$+X(<-Ix)h8rVY|8^N0NBkwI*nGvE#Zw;?Laj>U
z2@WIe$It|U4zkvZE*RU{3|0NQNFiMQu)?ru;Djg%I`QNlg~QtXbt$F&jkL}g252|6
zdt3mdU;#7g8a9j+o3|P=XPw^c3soMFS>dV=BPwhx#);s)3_^g&b5-eN<5O2tVXZ&2
zN%+q>ks4!(CY!eZh+G&^98UQ_Hs#Mf@;OgaSdjtSQo_|kO{b-gHv{I2c4Ga}KQi3o
z%p)VS0cScDVj~OK$+Q#K@1CKao5f4x2>cI9+{AuMIZ&)K6%&e$oU8GS&E0)3o;W-W
zvquMKE`iq@y_Q0=|N3z$yh`DMO;gb>J5wlUH|)cZl-_!QCmJm~4H~>eX%NkkyItr=
zkis-&-0+llEdBOo9x5&=S+c~r!oONum+yg43^LSI%h<qp_Ebxb2GrGhcYHq3>=_t2
zN4arC0*I1F5&u?^;XU-@M-`S5T8LKVvO;G>FEB@kpgy)DU_o1UIqZo)ap7p)0Ks<d
zHzZz4=J{k$af2+HXXt4~Y`BXLESWGmp2yzh7if)a&`BR?${%Cx9SY$6pTV>0O4M|l
z%;tbo5Yca9W1J>RyxBWh3{#p*QS83?d-Ng<Bkqqz2l79^0_VcWnCB`#t+Rh$exm9A
zzJ-#gq!DC2cE4SmCl$i}tsfTRcRQvIU({f4RDvCGC8r)&Vo{rK;wF}<m@~nx#)+9_
z(Y&s%y{TgMWf{>wWnrOlM-lMtTjX`dl}j_nXGnZ;QG65rnQI6S<}qC@DZA@KtVM@v
z6L>9rx=&2~{Ta}M3;x-|+>Myo<ty2rw*q#2Gt@9<uvXOwHpkO%NY8}8Cyz{wW5iY^
ztjyq=Hap~bsZLp-KkvpTDhh~BCt7bt4*`L!T=ov6K&(=^9%OKPV2QXm+WiknHa$pk
z%!$H<I_;O!((9KLhyZn53kN*{-2<xGIu4L6Koa5cguZ*sR#u9d_d0FzssIU@kke{I
zB>qPMVzUU)V}&%orKWjA9aD>OgLfswj>AOuba}9<9~5bTEADcD#ytS=k{a5UNk;&~
zjYk+d3GLO*+G%VAxo^R!3Zuu28sO|lM2OMl6~+dL<;2@|82uBA6jh=BbZ)WKDJ(xV
z(FYwwm9p=;SXlo+uJ6yy%#wS^um@%#zC@HF9K!;MTrflc(C$Q;v4zjD@M#qM{|7sC
z`e`w!;4GeU@HJyQTvW!UT^^dhxwWuQ2v8A+FinEzXF!w#{NTRS@ZIhRKCG*J3w4SY
z2S^xrOs@MiV0L}$I}znJz<9Ibyvf^Vo%PvHma9;2d$<Sv3AiMVW>212B2?FrA@91f
z8M|N%CAVjx`hz4~1d!P7wt0shmO!#=IMtivGOU5oTs)lXZ9&ttq0&JY<q~(g@K9%R
z6KgRFy{VNGwyyL<t?BcqmQBMRQl-L=THjaFWr-?zQ#r^Oi3L$aA>c?~D|$oAgNb1-
zE*hWp%aJxX!b{|4iOd(Ca9I1SlTnORIf+_BMY8GoCJGq&iSNWkW!db-oJg5ieFF?=
zc;npWq;)n{Zqyxj!pMC&-`{)YuLJvJYQ@%sun6HE+s`s}UQQ#Dj)!7;Uz+EK%V}!o
z?5;tt$MF~g<TsKa!XD#n)6|JcZi#IZ=g38IOrGO2!ol6H9E>P0q`Q7I(MaH6ADo`;
z4vEb%uyu$pLJE4Lic^s*Anxueq)-CwnNG%$H;u*Y*Zeu%pHZZ2q7j01L^uQ{iF^+`
z_RH)uWVD1?5GMXBkN04gcvf1=EPaCrci;uxQe)6rURUMx*{9j#ieX&T>z_g0s|<tE
zLWEC&%Vd6@pW)LBuYJnewb2-o-(Q`i;lZaxa0b_?RRuR+?LB=CDxQqqAxK*5YT7Z|
z1e}v~4a1j<T@ao*-5{!v&<Ohv@zj+qS|&*L_E6VC2Tl4fYWFxe0~*uwnGndu6lMD8
zR<IRHF}{PcrLVYJ#sr<zlZsHozA3fHu_Q-)1CNSro?^~(^UbN!DWAx#-aVh^naSEC
zW5${pl#;_SWdZ^dN1FK))|g+tDlt!8ULW1#t@LR_@?^NXNr*9u5XmU}_lRz3DH^Cb
zezQ|nvQLi&5$E{3oBJI&<!hPxzvit9LYcgpLki8Q#~hkm#)o(98<Z(a&Ij((N_lcV
zsG@3fNponz4LFGw$-}lfX?$VIdG(0;&OdgB5F<;xx|}0u;&~<5<cl*yp}jxcrs8v~
zQn)%r0%8E3r#VVh{(Ub$*KawJk>@n(n5*mw0AXdUouwS9tRFH%m*Itn==ygQ6_Jvb
zt9Aj8>Y%m$$X%MLR`$Y?8pDtA=O1(8gJm)O@U|*p{{7x!{^_|LBSN;!;>pJ)%wPjf
zn-iBOU!i@o9tpaIr}!6^ifrCCjyGtg-|!wx2Xzd*`R}MhWUbTtW<Tq##!9IxWldY+
zb-ZCt+(ZUg3QYaN-FUAD#@Mfq#^B@pWR|rD=G@2tj#d~rRU$$@5@auV-jM9>+{J#>
zfkGXSv5{)FCP9hnp_xw>7{N>!*?RSE=RT$EG!-wZ=z+&XhMe9eAaCKV_BM~3yPa5#
z{`kw0`*K6`ZnzON@kQU5k@vh~;+V*Sj>OPa+KG;GaobxB?KT4C3xlawdDOawg&k&A
zeRZ|AoZcX8>ogav^?hS;Mg+^ykEpW{aX&YC!OA1ffNZmM!nQaIQk(f7`hOSe`AmAF
zs(Cm+?LsW8apGwgS3{1GFM;i*_R=}{ov+dMV@wD)4;`UlU=SH8;sv2KVwU!|ljaJv
zJ}7(?e!s>ZPwJT%joFD5Llvw@-=KCWoW2mH=E<rm6bZK~UBg}<dr^o~0$5ak<XjWb
z{|Jnl|Hfy6Y`W=8KBplPgZaTu|1pqmcX56Ril5?ckNYxaj0A<GiD~-Yr_?rL;7;X#
zS+ay&mzuptNd}m>PT=VVmqvoLgdBp+W*w4S0<SxTk2^tqkUeK#!(g!OMQ11xkLJ>x
z^^a>?=)|7o48fR|9tN3@!+4EFoTxb$D;?Zyk&~B3KdAwa#rc#>0SZs>l^nwa3+|jU
zt<_@b?uRo&{3Po0C}3CNJUO+qzCRoIF|OR|1Vu$DH7L>+n62&(SGZhofz`1?4-M_C
zVMOW9mRN=s6V1j3diEe+SYX-u=({*<o0nOfz<j<UEsB>8Tmu1Otl~xc1=r|=fI;C?
zdvgm#(DMJ@RRpWT%YwdHFjZZ{+y2shmyk1P#?nL<Tznbuj^}zLsh#y+T6psB!B3&S
zUtyI>!fh6iWn6smXW;Y|&~oukk;gCg$>cBX6SU9D<d*WC)#u(i7*DNg+q2mXeMKgK
zoNF(&UF+@y_2zYvlWH|mwCKWbQfV+EKyOT{1_p3f_0_V25x2GgAX1)vT5;641T68=
z>z-*JC|?t*0%p&;1y5%1K^*tRphFJS_OWcNz30fY#2)3Q8@UY*Rtd)6YlV~xkZzGK
z+(tDqd*Mvy|MiTFv;}L^8R_k^$5%p1s<cREHt~^=PM7Wc^EzQ|*nAV%ZPOOkT8rLt
zMqQrOW|W|6$PmtBEsd}XMjE(CWj5`{5q&>4^?M@QkWD;!!nD)3#N|MRS#Q+7r{<TF
zU98nxtstDVWT|ExN^c$)dtS3ln@92^{<-O6wZkb;nH1O?s=svo-J!fh6Ejk%$5g@>
z9IWP-Ok|Mx-k7|MUVggKf3CySy1BcD4Qn-GQAY#>oneb2H##h=ktivbxZH_+`qD2^
zE5|KR$&j*Qv+Jg9)A_+f82tcOdY0`OMhRH=+oip?)fym2K}Kn!gEAq~eu<aebt_R3
z*IrPBL+$6U?4lobU)`%<GPJURmc!!H@$>{ExLAXD@7zDLH@?bGvO{zXW2m$!;aGDV
zvy+Z_z{uljsv`7T{8OA05IBXW#xqJ@a~tT0oeKE%%>+l6KC9kM2V7MS%_>A33cw0T
zZR`oy&B*3oDezag>M$I)FQaYv-SIj4Uj5cVLWX>Cu!pf=DA<kDzv2^SCQ9f@<bTja
z$2DbQum>;1h?<6w=LbisfNOxybBg3BCm{9zeR6haRUh95xsihoXV2Y0GYmq9VcNT;
zfo+sRM6`H5^Ouq2A&JhLRLdT)jNY0M!Pqb<Wo2|7r7^)G2~RNsc-o8k)u8DJgl1u;
zy_7?9Jigb~1N+wt7IPl?ush@wMy1MT=pNX8a;#ngDTYjM-faDt+p>%?zDCvyG}4P>
zqw0WI!kp;u_$xPKtRf+DU%s~%0o6g-(c)ANq}l95^U~P5$=15pB&qc}*~o}hMe}~d
zVy`&QX|lC<6k|2*R++1VcEST5e%jHYKa^PI|3EA!?Zroi0~y>RvCzNh!g36W!Ej_8
z2W|ZwX}Ot6Cf7YU5P-!xBtaL~Wni2|{-z8Rwr4lbaW&+`@J$SCDkqC4zCAC3NirP<
zD=YkxXabCGZgZTVcl|VJN9_=og|E*is*D~^Rcl26YWdfKHna*1C^msE^kG9WY5A%Q
z(j&lJOr-TsIi>tKAmvgGRmVkWpC8QDQwmOXyl>Q$5?FAaxuU;*7vAh?FE0Jg`siUo
zU~ZKr59X9#L6>~}pZdVQK_L{3va~WradHS4{oeogbfs4HNei^Ipu3-GMxSGa*eUBa
z^%u9u{Dp1r5^1Lc-Ih*1`vc?j7K`>}npJ-msKzGA$*;B8yJeIoD4O{EIghfU!K*z(
ztzZ2r6(x%SC~j&l?v`gPxNUucR%mGfNCYn;A!}-#yp`0Uy2IW<S@?w#4hAZ7spWqd
znQhnk)&xCn2-aCao6e)Eb9U~VGgQuxu#iHB6?I|<B3!T#M`$N`Tz8fchPr6#ym!LG
z|Edmc>)=;tVvmnmCD$)i{M4t2DjZgPJ?FvB5IZ=);dV^<07;*v>t$n`yzk@mSV9Z$
zvgeo=0w!Y~LE5ip<c|&{R}iY8jiUs?-<(>NKylL;Fv1}GhN=fqzx>!h`Uarq(*01f
z4Ysj~lFI~y55a;_3veMONBg-G8s1q~wSvbs?<P9o#qU|6y!(9ZPzC?qRl(55LoRuQ
zHnUIEJdpuKn{Qe+!d7y4kq}3<agn3j0iK#+QZme<oo^eoULCE004iWeIuWof<2JMl
zip~Gn3Z19*G^0n?U+fClZ>-Ou(s&pHiD{dM3<WYAXkr>Wp^JSrC~Z%sxzG57V~@j`
z-dswXRpsqNHVeF3VgexS27L3pC^|UC8ugJf0v#k<u5|b>Yg0|b>46|&0BZkCA$f2=
zkj7s?J<pWYGI%%25Dl18cr316YDcf1@y(tLJ5+jkl<jSYlM5y_Kaj*fUcv)p+mh6!
zSQgrK*4V*XvlnLXS-JD2s&0Z0yF7woUV9p9@1D}k=5NyV+w3j4d?7+l`zf|sf0J3N
z`KEHcg-#28Y8kJ$qaW`Y3HN#njQVk5kqCA-0OH26*Beb(K|$kD2#~}cI;JAN604xz
zi3L`Q{NVzFo7etc+yH~aQtK7aLmC~P!?_uhcB1TyD2wMr-12Z#q;)Q?xPD?79h3Vz
zH=IES8}td0Z{P=Qtb4J}@Ku2a=kv%*HOFJa7Fkg5xN_qnu*x}05MB7V%b!>EJ(<T8
zmSnZDeGH{cOX;E?itQ5nE`1+fY{W+U2@<6b`lXs$KmeeirKwC?aCZ`EbxRjeJH3lH
z|DYtH@aFvDAo%+G23gm5af6_Nf)j@oa{#-kI0`ORy~68#B9>0;k7t1h3pD);tvHU4
ze;}@(?U!_u7F2B0^$7%r{VZ6vDl}o<?1_o_$=WE?WHHMTo&b||cLY_5ZP}dnMaMPc
z&*`zyTv87#;G_#?_&_i^eP>XY%7{HbpzPBoOnC~8F5Nj~*fNh=KhV`rw0vpXP<Dl^
z_k_%J5Rh=opuv7*$V7RY&CrBL(=t=DGtwZEcVk&XJm5<Pyb;K=STWHsVd2x6-^TR7
z_7J6aeY9+A%z#TYBkqV<30&`w1;(r#FPhtTjcuR;)_jngw^fg_*k)*@EKilugp8Ca
zC#Zgqtv{X=r`%{o7M8wdRvdu+p?pdg<o&R4B2b;n-J~>4J$R=`c<Qf6VX5y#lYq%Y
zg@jI*I2EFFyT9fsSnahZ?^hBSAKnQ5Hv(@;pyow;y+GD;NwUh@%>c?f2kG=*f#phO
zV4E1ya!3J=y={1x{MF{-^R$OZEH^(xj*qg}RyjOgcUX_gVu~1lv<mxgRth3>4|1*|
zF^J`)qO|?qpDpW+&M(0NiTQi*C*?x1>CHOVHvpj(^qo!YM?X1*5=F$GCe^(7phI|(
z6)qr*k{q{!(RAKd2J*3}4@dk~a_>t1dL}St%}<TS*)Zrs6)C*Tv8Ecsg|Tgin85sA
z1O&~TV96q-M(hIc6H~0mqQ8Z2dyl*^v201_U2VIYCr=&g1&0x$F8_sKvJ4;?w=7%4
zs|MzXktN?+R#^PuL=SZDh7oed7~j?uddA43PF}PH40<hooU<h3Nk@<FR<!cz_qC&E
z4^kdj;w{3c6Y0qe`JfhPdDbU^$s<;Dzg#S5#@D6|rdjvMElRT*SdAnT<R~G|>Ug`f
zHvN@x6fmW{0Hp>@mx(g!9G$bP3F%K(a3OVPVe&Ow5rccH=cbqj6P!73N57bMT&tHQ
z*W&*2bYW1eb8x_!8#XB~%LPd_Q%x3KaFTV$aV#kR*<XP3jh!kX-}pig3j;4+rKA;0
zaztL50Peuxarn$Dil)rLPERAU%#Bg?>`(0<!Ry3K+$i613E5#_T((59C}ev3Ns;EX
zfEYT2tm~}lBmDLOg_;6_wR8@{boO|5`H|kFT3>994e%wJnysB5;(}vaj}e%R62PI_
z&gJ?SQv&|CRs$Vu($NlMwj38duvOQ7etoE1zkqv5C8`eKRR?-qIe@J#{xipga2Kg^
zPLa*KoH+~>rtbh0O8yNzd#TUq^Y5W#jcexuyKx@bLk$}6<!LwWo1#vsLt)?3&I7t0
zliL{8|K5{)Bl&e1QoRFcuABX(coxc>c)|X&AR_;WOg0FF#Ks#N29k*`N=lnRmPaz;
zf@~gLs$<pt7*nib`bR4ziVm#N=rj-vP*I!7;d`5c@N3}&JyOxO4q7Sw&w|`X`xgWz
zJtDj4pP3(cmuANw5NSH*#ZzTPdz}>bUv$1jMaTaJBA`0?@l5-&5hNhe%BhQ5m;1-;
zuWHl<!j^Q*xL}dkTMUC$ABa4-I|cy<L`p#R8WTtby(26i0ED}EvlzZgyD!0x6+EJ{
zW`^jf1yAN3gIQfsH;>!v)!&1D%3C+Rzl@iZ<ezW&Goy}=SVgxW5Rr1HE9VaQvFIw%
zEBRU<##6WSI@Q#t3a%@$s!U%gA_^&lwZ)6#beVrPn2dtV2O5Q&2)k?{16<s`g$6`8
zj!8e0w`@7{24e3*k5(?3I5D=*>4fhco1FWkDeKC&KW)`fc015lC$iTXQRE2*NRy=0
zeKqiJ&}yL5vP!c6Pyl2=o4+q*8V{Ugb?eS>yzumT)sXf#CndpB&E&2p;)w{7AmduD
z%|=uPqSGw<ez!9`l7rB)V3kABaAfXrz<7~BP5OS@?HKKyagTT7Qg`@)hQ0XCzuB0j
z-o5k4R<~yx-i-UbYyCLjz6QAHOu)_J1D4mFSSCk&THsr<QcH2p0^H=UGU5MkqrU!?
ze=^j35PwRObeE=`e@Iob^-ZqDGA&88NEYfIojj52FZqowdG(`}(B(`f7T6q{5Qw)*
zL*6bO>V|%A{+&S6bL(4)jQQc9Qpt<{2Tx-L6@$6)plDkI=N1LAKOYS375A%H5GUxq
z@LK@&5~7VL#i3lcK$n&;QTXTKt&y0Tt6?4%plY&i82H9XJxbOUVTS2Sz1Wmp*01m&
z@Rf0EJnvC-G7qEn)Q02eijfxwl*fHsZFeDD4={2R3biC|lbT*g9!o~WRt;27nn`@a
za5grxRv!=l&6AZtb6DkR*zwl!0*cNT+IN2mpLge=NKG`h5CtWb5Rv&nFB;;D;REt8
zm#_X~@wJG_)WqTs&nX?p{v+f`tRqP*&C9qdVAlP5k4Ryk#$6zkA@3%P5Vc_#N(C%t
z+c%dWM<)A86Qd0^R|@7s0Zv!j2J&mYzmBY@j@uE;OX+QhQ`NPKKEz^ihET%7wzUOy
z9K~ysQhJCCmKk?Hu83Rf`}9XoWq$9$xHoyG@LZ*j453B9)<k*9d_J5bRHBY~6@nLQ
z1YLQBf$<B&!l|s-DeIXr^pDx{P>$#EgTnTFt?$T5M_k0>0$K^73_@TuGn~ZSNAcz_
z5cY3v*0XDt8s9gHOM^7=_IJ6`g^D~vlGK^5J0=dhK#lAXv7FOSmo9VTPv1?g3y<G(
z?TB5rf4f4ch?ZH)oHmjM+?*XnJezIKocIKqxqXWQ`i*{3JVL2{=C{3RIkwsvk1Qf)
zt+?!P0cE}98FRGIU-6s}t7YsuBy!#i5Vqr86wNF=e>-RTYdg=FWzH;{FZ-Ur1uK3)
z;3T$bC%hJeJiSn<)8#g}2e1y=5@%)EA46gYVCFt9=>^2Gf?kO7O66io+oq-p{=WgH
zh{Qg9D|pH`%Q~r;Z7sJwbWSl@;lnb5MM{&i#dw*GHgq%&Ze|?e%TF|fDdq4_%SVf`
zd<q&6;7=>nG+ZR<ByJBB<vd$T-(DIqdXp0DnyLS^W%+bl9gb3q&aMUg!u|hui7suy
zB&OWw7Lxa^{$ezph?Ae<mqMMcYFA`0Mw$uQ;LZ9)!1Q+bWuIioBLvSw-QJ2-aqisr
z4K0E;<nHauGeL8FM-EZeB@_D~z;D?On94lsCVMA+FLsD8nOt~dVvEVduFa3Jw2I?H
z%kwNyZS9O7?l#X(?9{ZQh33y0>->S2*o<Tgh#x-mC-yu#nCyQ}K5VZ~utf{u6ie+y
zCYUa<X?d#u>$<(%&WT`NL3EU6+GS#?Rv#hHT>}PG<t{HhWF!ta5DS3S1?SiTzLGQY
zv>O3Cf;crk-JWRh<ut&AeW)PO86X~zl{FrTXWds&QRWQNU3pBJEA8+i4MpR&@Yh)$
zADq^Di0jZYvX4~m1;r92a^J}U=1*$7s?2?rV@K$3xH%|$*lg3sBW}m85~fJ)`54{>
zp2OAMFw=)l?WT?r;7tiGPsaKcS3^p!Bkcqj4XMW_?bLAMRl~;D0HTa#p`Xuz^{imA
z#-OTv(MR$i+0$FZoTpTaKFBHooKQ2eE&l)~RbcXDo7#W2dok|xI77=pHSSSN{xXzI
zV2;@JIbM`1#{%wn83WN*xzt96L^XgYHa0sJxKgCO>0SmqlZ^N?3p$;(RYmc9OU0~p
zZ@rQ}HZNzrfOT4sR33>zk}lv|CR5BvCrctyKkhI@>Fn@8v0aq31>!%yKYhfgWSw@w
z?(xYeLk%-N7G*cFDwviDkX)=$#ERML6H*nxXAsMXIo$mIWteUGx0b@+YSV#1;o?B?
z>%AS3_#np)81W_Yf_A_#<A$`JV4s|Kr~t|W!&f;2`GsiFUp(`aljMXscFx+l9U`s|
zZti`wdf^Z1m{PO~4EE=8dJe(H>>J4<6tNl?e!XH-7QHy#PmPHFMuQHlDg^K^>LtN#
z;KcBNY-me_ZQ5(LYW2I^hyca=8BA?FXZAPz)dQTnP+e5K((67kzY>?JD7rMqNpoNl
zl6u#;Dpz$I>x8DTv5@BoDL|qwX{fC~2=!-9VI=c76K?*zqjEyTn1a(9zxCxuF=}{$
z<RRzcZL%41Ox>*Xg8)fFu?oo4B|>MI!pfAyUcm6SZk*dV;IzTe`no3>sW7=lZsY+^
zD%Zw6D&M@DE7wTaeAoBpE823iCqFBfz<8kSF1Dpemjh=LVincg+9mZ!h)s}Y0<q~p
zX|?X<XV>6kTx7S>-i6*FK+zTRn4BeIBG)I{TW&7!8ilKWPY7PpZq2%<h^41uZ_S&^
z9GPG}Efco<6ZciwKwRJiR58(2<^!0ucE{Y1q0+f4U(vD$xqweHpDtaB?yR5HXlT&F
zqamkCe)8ZQoL(z&a>TmYTUbzvKp9<Lz>yGNQQRGonuqL@^YdT((@*eCHMMRvPo?Yj
zh)-TP>O!nd=bLszYp>2Z9R!>cY9GLr%Fk53py}!;myjY?=Z2W{?GspHZxNc^iX#Mn
zHny?>eo*w(>g6Y{Gca&O=Ox=aq`dg4d9>_~QwB?=XEhqyuGIOf?y%uHF8l%#|GB5~
z&oWJQl#ipZX24Ky?a0GP+-!$sT$64=U!%6s=_YmC_$+yFLGZN&daK2&q1-t0j3D%<
zpaWC-NEj1GW_Rf!Pi~drssie)zMJs`+Ayr_iNH0LtYIo2$W8&~VzT%a%lCyRY)K^*
zDDn1Pn+5-}%{i+S*X0dPZ0svltpb^~v`EYY{inoTOw;eZyMEje*Na^Z>dVHIzv$aH
z$&n}&|LDgYjj0MPYJ$~0KeN-3GHnonyfQOmW5bKta@XrEw(saFwfDnx8nx>A#2^B?
z{?tmORWv%_mKq9q4G5OJt`?n<uU~rvk$%;jliUhm8KI^boVFb!k4B6OY$%m2D-6Fc
zJ*!7J<i0VCdlFBlS+QlYquj^+p6S_s>I*3os=VOT48tD44OeTwgh7J)`QK4KZh2Vc
z54`LG`h4{2+bNc)2lXXsV_d;G>Lb~}uj@6Ufp^xH8KS)_A>1rxA{*#xEVjM++VM@h
zPSLBdf=DB_slt*H?mWuYjZ`#C6npJm>LMt9Y*w(&svtg-Ce*)Ry9|B}iU;X227wT`
zHn+zjLm8IB%wi0g^!wnf4AcP_4`|4zVva4uLk;)uKY)@{X_YP5a2YQy_2!PxuN1Pi
zJ7x<ME)W4KII$7mWIsBL4quHr=F}2WTkwjX&w-l>)FU!N&{G%|RE$V(Zz)(GJnBG=
z1fHD;wkt6Q%r4pq=eBDX0GNs0bx^dOV71bxoAIz~dxmP2czVSMPBREjkFm=uT<Ak>
zv~1imn1#7@4}6Q%ttm0ngy13xRZeXq%K4$6Jq{gb=6be`Rz*^F^1A0i`b6RY>+$@U
zgehBV)38EiB5&XVUBU;nhZ_rfal+ZSjWdTG-|SwRtbN$NM#&I}f;*-x__jpeVC@v_
zWL}$Z#=8>()G?mC#(E$=+bU$<=6dyipdi=?1@7SviTZkD5Runa&RxsiOPtk=0qJ;L
z$LQ_r0mv<P%Qa~e&uZzS(Rh$oeR_^Yvic=wdHkH^OD|>><|PmYtdxmG=O%FRI!5VI
z5LDp0`{QXLe6L~A3$M5^G%fKl+$!H`Ij7eBB(xGIR?+tU4td(lBIn;?_SClN;nNCU
zvwU5PPa@(<nJ3-enx(H2iay%{jZwbN<0LEGHVm^@ESP8pMZ!s+NaF(Ac%6!ULmc0r
z{blJjPj@T(Z|4D!MHiF9Cf!ww7Od*|1?3Z%J{FP+gk+9-Ka?SB-m{v~nM^6%L&fCB
z_{hsnVA}J6PI9bZw3<lol2if#a10+2HU{Q21}GZsDQ-urm2vtE9a)+HnHRe_S?*Kp
z7z;9DsK4$XT5nO7dzZZgNv48FFyS0I{(*TYF9{ED`B-G<N~s9Qn(xX^rt1H!vFw~#
zdi%EO(&hiXI+v~Q$58nLu_@q3(1E3dEm=yJ$U$$zyj<pp^X@I{t$c-^jEHpy*gq@h
zd-FF8-u6lo;83W(1g>wAAL)@H$a6~QesIzsGf(Vrc9-`!^=8(*Ry?2A9iFt=*%DkZ
z3&WdG&ygxu4}p^8{u}h<b%S1#pP9HL8&TWZn)3Bx*87w)O0f+iu!NIp((62!rKHnj
zNNPVHbCMgC6srS@_<Xls^Evs_Eld3+R@dMTqcbf}ga+Beq&GWA4bF%GG@_~JFZop+
z$q%mIN1y6&{fMHV($aW+aRM;7U|ljTfmPjOI|&OcFv4r-p5w@PMF9zKsIT(kvU<kT
zJ9EDst{avgCEZLEJmV(wQaCq9hdr94!>3$4Kv5Ik_k>wa%1dYYykzK6mvChK!NTTz
zIzEf@G?^HXm<ab&DREhrC-~-5XbuOge8C=8jcpwi<{7L{iR>$&Dfl+biH~RFEAh+D
zfv1(K?lDWv#gO+@!`t4n#o?3a@|kqhg(8&&7agO_{2F>uPBg@Y7@^c>nxkyBpY9Sv
zN<V2Ge@ilX@HjeX<e*DDUA@kY09^$V=m>NCfPB~gFXDXW%)`Maq8Ts)eTGHsyc>tK
z9_gzkvJ7P$HlN=C6zmH5OB;qNO|h2(v{><B<1L_7qY4f_!W7kC@{_3^&H%(E7M?)-
z?A_3H;Ffl13uGT&iK_&7ZWb4~2gQ=X?u#bcmUKb*lM}hgu9=@rdttv>S%)?r?0o*O
zk$f+=2tmQ1`J1HXfGx~Si2Eb*_<mTTe}%>xR0?Wu8Rj%1!Fbpj{ZaZj#$+<NXJc2X
zr=Ti@2ds9Q7=%ib)eIi%pnqvbt^mK>oUxI?oP-0{o@8g#SW}fOu&EQh?1hiTT^+K^
zp|`NR>(=;g;cX&gEq!XmDG&=ACPDTK%1pxnHky^k$u(nSLEZ9lq|k(MUlD%H%%cI0
z`vxsroZ48n+Jz)G-h6;0oCBB!2H!@1ea39VC51Mg8+f}SnR}+l2Z`6Gi^kGCHJMmy
zKfUu??g%sQ5IhYVKj-JgwG==S1`*7D`7*;32RYExe@@qaMW}_KSozy!KkPoM33%9Z
z1e1Y?GEd*wHD9mQORzcT>yHZ_oKlVt8Af;InPnu6{vfJ^@~;r=$?G|{rEP|n6Y_IL
zOTIHZWKF#F3wBlwrm$yV(&$CU9~`?)py~Jr&}X|G8adu$Cv~3xa`J6s7~C7dPt1IE
zV-vLA!><UwPSQ8Meh*Z^qhu>D`m*cy??1|{xT6uq;?S<UaL*f@P0g~<lcz2)U}1ks
z30)$yOrLacld6h!=x{pYSq^l!k*MI}WLIe)3lLLh<ydXIo##W#nmR#!Qk{{q*%i(B
z`8tcc94uTE<?v8mz(y>PL^3-a$*b?T9iE{T2B2k<kVT%`46e;zSJ&CM(;Z_zQSzky
zKv63IvtPMrPmUJjN?|?HsXWlhY^bW!4;T^b)4OY7*U9b8u|_iQ$@E;lwO7nR43noJ
z@8t~WFp|j)e&t!9hO5VAR_jb-VachlBxXo+@)$?Bsr-V3vox*hnBHTRfV-f~P~B^F
zzaZccL0_1e7_-?RLrz(r3dsSCB5fV)i(;k#y5|NcFl71Zl$~+Qg)y0D)p)YY#Fg2e
zd|k%<_o<v<r8ye<R!_jC$+c}qt)+jWx6scTLegf=&>ZuZ;?53C@dU<q5tc5;mF9)=
zdmRq;2x;H=UNs3`tNNC3eV@bH($pUn=+Uc-{LDjLU#4Xm@e(HHk-fF9eHC_h&;XA>
zL3%T&x9WEe7_Bes`9~O2Al4`0^Pm>t*+ORFg!0>8P{>^JGxqdP*@8^NYf7myXEp}f
zS)ajjQfqkQ8TY$2>?%d%aC>YucXN@F(d4cM=~KCdn~Pm%rEJ+?R$HefF6vTq%6RFj
z++DoUa}3{2(4XR)yzT}B_6{O$?xtoR+H3yu_+Of?U^?0teeGgA;6Pu$8=kzqLEglX
zu`(B~q>F^h>^<zQdM(p!a(FK3mZg7>rHR=n3kbG&%y#z+Dc?dZ<Ty!qQXN>JZ<=aP
zl9r*o-nX;Abc|tls0=$tVpG~k_9J(nW9kEnqGDEX$^dnwiDo#NcrbS?S5I%MVuCkV
z0x;-4tk5~Gi{T2SEAF2Nbl}$KAB4nh(^3N+cOLxL$#202Ini${xhf1FtR(mk5e7Y7
z$cn(B9qK}p{M|evcbDCRS5F&GlfEQoS1aC`<g8(m=Wz5aaJ(u%rc;u<^<Zx3OIyHZ
zp~B89@_obW=YGsvUfi#-Zgy%z4791hyT!6~l|D$Xj<cjD?Cb@IPXwsDO4?Y3HK%Vk
zS=Viof<py;E;1V8qNtbdis9o-tq{x4YLM>Vp99lFy1UR+;acC>WHrLtabzg4%$fmi
zzh&w@funev-N0nqPc*e7kZ6nz9MWG$3$2=KNWAtsCl_|yf7u1VDmR>gQ0XL@>mGE5
zFx4GUe4YS;u1u?mO+cp20ZWjTgp%{%;?9X~<8@W4er9uR%2`S3u%BICyX}P6K2NMZ
zN0!x{`9NgGb&{>}8^&)J?shf=(bfg0h|fA#!^J9;LBHcP7KRGytt}3Iqan}V(x$1c
zQ<ST(<uS@gNq^-QJpQgmo8ITd%%AO7;18=a)W^9n^<*(z>{EnSFw_YFe0PE$Y38-W
z3jq{`usB;{@r^;4KOZQ*Z>FIdIrCHkT=VhHVACGKD<TJq6Q3ecphRP$MyVJ9Khg7g
z0Saf-jzXoCn?@{s*H#Ai|B)QC>A%MKDG@sV4QGT=%%i2(c^|!0!=OI6=dS70#a}Ny
zV8=BRYkg7F>R9#B=Jmf@9Om#d_d`VIkHJ06v-UD$v4_I8(=28Wj(%*4-Pdr^*p0?*
zN@?^*AYo{rAx0Dj4`t<qJ4~+`Jyg5nQ2U{x@`sbkp0yyTMyn{W=spS&-xe&q5}o*{
zg?~_Nv1hwjhN)X9C{_^kd-j0{GwrOel5bZ@*MAS3GMf)->W`#cGvK;KpU^jAV2Nee
zgv4bHOFeqB>SCf`$b`KNhMLt0)k<!&hrT#PGs6W6R8n8B``4##zsF!*C*A-bWs(#q
z>2`UptMQV7X--RZaRSFzvE)h|_2dBd0%vsvQSzrLCsggwDbd&#Fk$&c1wQ)W@(Qmx
zS|9`5)noo7swZFQl3NZRzU;4q2`WSD=`|r!3-={)pRha8a$8jafdJmA6U+hSLwslG
zUgvQfGg4+#HeWuVHO&SR>d{l6gFkH|#zG5&u0;RtGtUcD8Lc>(Rt6jG>R0_kU_o?=
zQ+7sn_>JUeuQi^eLmPTtOqy=WK^?)cI0!;V)LSpd5x9O^$p)8JSVjyH4|Xctmux4|
zI}{(F%iEQOax1=wNOx2c_$g#GDz!j#8nV}7+t?!ff>m?a15tWQ;vMrb&^_M0W3j=F
zL|r{b1q>!l<*IU5pjxda8g7Q<d<TqZE1xm_M4#$iJ_Xz`8$_^!-fXMq?CBIr7nV{6
z#yURS<;k*b+NiP#5QL70CwnsRc%qNg>k6{x@bFv<mkMIVF3+4d#$D{z``amw>}_a-
z)Q!iEWQEUtG5fo0ssLGjCoVjaG2dwx1`s)-@cGHn)Y?YzN0Gx>2LcTVStP5Z;OaL!
z(fl_D4qa%5NI?_m;n#nbr>JJoS{8;X5bu!J+I5DAQ{dJT=OnP9vKm~V%fbkktU&4+
zA}(~T(bYY*VV^RJ%W2W*I%b^9`}|Q;mXsb!I%Lu<Q;Ors`JbsEG<0GAaX2G<JU_le
z9(|CuH@u7imiAy6>)F2)y!Z?H&36Yjqpwd5@yic-gQp{LnrI#u5IRsRX=~&s+?ae-
z29Y)o7O?wIIam?gzWV7oq=d*FFPdfvf^lvb@CEkbc`lqRe?oOhygJl;Go<kg6Sc$F
z?nyA_K&K;gLpHE~Lm0jHsdFEvu9$yd1w<I&2PuJ5JLwn~SZ<)h(BVq{uKX{EW6_56
z)q{b{V>%pi5Tf>?)?t46Y!44=W(KmFp(i5YuZHosxB9#K#k|MzTK~waYSGT~X%!4h
zah`-qDxqF56PYB398H8@aBP@q-T`e#nIX3*Q}^W7Y%a3`v^T+rnG(J)b>nx2vwJ%P
zDdV7G@&5gQR4oj^;S<Tfn8;C~`t%!WB@8Xa7X}G7dAceA%PO25_s)r!=)`S2$->wy
zejpJvFuCzQV~fOcEzUgn<%}5|LVjmdPgG_FQS|oKISY~F#CP<ToXqo+;05n^8o0yt
zV9J@7!yz?rdA{Fli;_?_FKbz|Q6X5{adFe?^6Wht_UKNtAEphm3BR;0+$P@GowwL7
zbiYbq3j<>N-}=Ix!y0c6kr6<*D=JG>uq!>qsgqFipoLP3Y}G|sK<6!xtK<-ye3}RM
z`a;wLBs2=kiY(9N(K2&pL=zd^?RHNC@=SK_pyyuT^a3^wki7?YBpwFuTf=;n8IZR^
z6<}{e)OO<nPcebO7_k)<gw=Xhn!dH{B-=Q|x+^G!WZ{(iR6k)$DzOxTV)MlHy4oFG
zVw_e?16R|`X&k7{05A=#_ihq4I*$qMR?d@Bg}X18!PX4Km+MY4$~D3Fk;i?gqUsS$
z(GkmFveiUg`zPKoz{2*HL{A@B=WIHy>XL0jqS`c{7H$$M+cmKJ<lN5vQbE40d?AsR
z(>U?Co(u&Y=MZI?=J=zb$^eg)L#KjzKZ1IxWUlX%m5V=5#ve<xNsewvJ(kc@>%o4K
zW#9~Pyry`CYj;F648Wa%-3GZ*#-^D!0)ph{BGY#}GBS@z7i?89G-tQNgTt632v=AZ
z&ybCl`qobaqWg7w->8LWZIwBGMX_wlIf>V#F98=NBd@%MqlvItJ#(glz0i{P97{Yp
z26y8)=5$tzllDu(T8X?B3>zz&E{<S*PCbj8Z^9n<?J}?UJ2NFidmFH{7m6dvaq})L
zG{&&JTj2~St5cW{O#8A}XgfS<S>4n@?bPdX9^XV`ZPVwp)|||M8X^wQ#8Zbill2$g
zj)%W6Hw5tTPG(WLEG!1NgvSQy>r&W(aREH*#NuPcWKNPzG3Tm<)<1(x+<`5ALvnTm
zu)-`rzqA^%@OZLVh#L_Mf)8j_$5OK#1hOrFnemr)?$h_^Xe5&hctdXz!aVJsNi{~r
zlzOmDm^Z9RC2ddfLz(aPmMv>58b9GZ?A(AuTB|muCNzo`kUfsH{?<!>H0g+H`j_|!
z2SLWXkW$Y?)`V3i^?0x5XqU66RvW&3g#^#W{GIRvRcsR}X9U0O_XiFGWmQB~tc+&}
z{c|<ATx7r<t;ZyO9AXVxYNF;T6{(>D4^QUaqBW-0pfIAR6_5L{)@)jjC?zn=Mp785
zf3<i=B@J4g<=ICl0VN&{g_Gk7#c)Iz*DY+zndoV#`2Ndue@Y&HM+*;D8dWxr0ZQ$D
zlcsTg+KU1WjEUdTU7PE-HmJ>UL@cW-#*9V7%G@&j6vd$)(7<`UR0%4;59t=@>^O>o
z{c1*uOb=<@sl#Tz($wijyN?B3kY(%K<mwt6=i=3sF_=d(@PU$FIBI}vfWkC<!O?B_
zzR=}c>^V+|!Cv2nL@4JGf&M^ukfNk)kXYAB+_Zh$aUWyWL$P(zqFNx2k0$?gAaw1Z
z11%0@;h+D`B^n*SSDivdq?`7}<sUg+4IX9f2tMrU!^P5767=$*v<{twT|YL##$m~s
zRlC~PVHpc#RMS33S*}kqkc?iXEKG_sSHaxCu6l00Qn{b!#|D~m_=wC`aLO(6IS?{I
zPk(pyq5M8rS47hQd%NU7*CS>(c<EHi1SX<b35)7`JQ64s-aF^#CnDEU0)5G~eDpl|
z6geW3WU<Sw%Ou=T3atVA1JrF)D&OddYvEKHLxbWzM*5tqysa^Bjt9{UMvz2gTQfVI
z@2>{Ve8q!*PkBYjMNGK8h5?4<oKl98L<iDY%64u;o9rfE9nhaNr<eX<0Xqs?A+A#P
z;*FfX9w+0BdMBEx*YFWQb$%ch6JL@@v~pgjaqzeOyLc^zl8L*xD1xXkuRcAHwVAw>
zXcGoPdCoJB77Srr7jphu&UhBMlg{<1(f9#`KS{aH-05H^F~P7xw71Q=X<aCdpQ*P1
zl;JR#cVT`cK@#Tei*za7Z=F1DO+!|0*?XwE<OBmWSe(y;)`v!!#VX=Q$p23ORRF60
zf6FQR?Tr-RFcn1I36(N_h0T8_%ryMqY~#+(a|~#=W+Qcp{j{oCXHkAE56=OJaCQ86
zFN$<j6yV}pGQCBFUq?ekS)0KyGkcpc?p+8gZL_6)0hz28jS6LMWOH<KWnpa!Wo~3|
zVrmLBH#IpRFd%PYY6?6&3NK7$ZfA68F(5KCGcXD-Ol59obZ9alF)%YYH3~0GWo~D5
zXfq%%3NK7$ZfA68ATcpEGch0_ARr(LFGgu>bY*fNFGg%(bY(<kV{c?-3O+sxb98cL
zVQmU{+KsjaR216x21++bOEZ9U4qZwkDBUds3^3Ht-QA#cH_{!_ARvl}bV+xYbUwV-
z-@SPMZ>{$hYr&lT?fCXS-#!B^m69r}m<iMfEDg1BWCgNu2m&PJrL=$?01gf=HVzI>
zG+J6Uh@%zwe|9ul4Y0if1ZpGr$V0*&403$1NrD_79Oa=lfLBgd03a6t$SVls6Xf6k
zaB^@6{Lc|;F9?tXIYUeU@@#-tP#drV8m$D>*3BMbX72b<=YJmo491KApnw1$%U|ID
zF>A0r#291)kOw)MgRLJb8iT9=s!(GH*wO8e6b!=Vj*hm1?CdTsE^Huc2R5j^*$YM%
zfD6RY9H0Vr0NXo*O#r`I1}K26!M|H$L!$+#nL`}@22`P@jxHd3FyO&p1u+KOI6Q<n
z*_eRs0T0yys<LtbMO(1V-_CM>JFozL4+j8b1OAilcl57B5Szb(LB_^VYg>?w8^p#8
zU<$DU0~Dp@*c@FQSpXm#lV6S?D+lO9JjfXYu>u)AIQ&H&1dtX}27n$q{N0~}u|34r
z(SgkYV)d&>_Frip#w=xHA_29w2HQ9~p#9<}39$zoKg`{Y{rBZs+CW`wJpOkwh1i&w
z{%XU-$(CK+24d#~mX-YN^I$^z*JcKG1n_VOaPsl;0>E|vu&c2-`>z6OZnogRK;SR)
zLl0gawoqGu=|dY}FNi7l;SZXJ1IQT+aI|*<dwDzt{yd=pfdCVTu_M3;YzDDG`#1Z8
z8EpFZ`QiTUA+7*jjtBVx0UW=6{`sW;pfD4tjg{NK?tfh|yZS2?8BH<f-!1<Gii<;C
z0UoTJ+yGWi0bULOkei1az{kf4@cJ{65(x5}$fICc8&fDi;BUSUgZdw5=iee=_`M;F
zfIo97Kp&(F1~B}qbUh9p4&#Sk!2jQR|10GGP3FI{{C_I_zl@}vtgQaR8UA+oKRC!5
zV&(SR<3YMkjt^@f4}Dk$oBz6Mg8$xKd9Vq@$@;&jtRv`Q9mH(Rto|7!#6cS33N}%K
zI2xP(t(d><>c4i)3St9Rf;vEcy)FP&AP2{Pu!jXRwtRR&93Eu(7X*ITpFc}V*%(7j
zek~d&4=(^@Zx3=qd${I@6A!=x_^^>CVAsEd3}9!ofjT~f03O=&0+>SW(SF@2FDHQA
z$R1=2esFsL{<3j%{I*$vO&!G{W@i5&oV<?^qsIsj|6|1QA3N{kT-Fbd;y(e8Q#soH
zTjFoCCD`$Ijeil2|3Z(-|0nEU0uOqBj67<S=g)kPIB@*~{H5PNNC5v2_K3xQ10T`g
z`4f7~<3A}M596UNQ!6Oc{{J-0$@LrgzbphE`g+)Au$eu`$^_zIYXx%q2jl1b7kl9F
zFF)YJb=W?9AK3i9^M7Z-`6uu%!tn@rJR%_fpTNHeFZW~Qkw_jFe=NkunI4S=$oJd+
zSMz_c<@yiuh}L7|555BbK^_t2`2%{y_zwn;DgTk|G3ST>7ufAVCcj>}$BcnM&VTHG
zumW;Fwm$+M{GA`J?%^qLfc*8+{6hYC6pWqh9~e0P^>9D9{qOwOSMkH5gN@OqC!oec
zffhA^trt~dq%N#$gU?54*EEwESv{uh+nlbjkP{dy)BP9h&%_eGe8iqzkzzO%ouRz-
z*lDUm4rz{4Zn^fndGl6fc&!C(YKZXD_xv5P+Fo*0GFCOwb<bNnPYpjy_$JtnSG3u7
zPW)I(xp)^YJuh8rdrN){1W&9fuUGQEM!zW;VU1Fc)bo2kMw?@lF-}B{<j6{n!i0P1
z`ef|%Y!o-^`#t5Wx6Ei>yHQ*P9y*Jh(U;@yb7g9r4xPl$h;@j_;ZJb~hUq-SHxpky
zo%ARwn9m!diZ->qe<`ASrsn5p>6A{-L#T+M1OOg?(`Bm=|015SD-Pv||AslB0Yy1{
zDdd-T?2@%ZdRitSZ}J+M3_h4nPg~7{#<>jB`U{{HNs5QB`K9+l4hFM7mu_!CP%2*}
zS8oQPAf1y!j=uOR>MJk~dj8;@^Yupt+~rU>zE?1^Skj8@h?Y8x5^#l<esS)lDNK9?
zT29U`sXdsH@27(=%;+W}%G7W6@tGydHN^^g7Z=8ADfY-$`^J?rT!r#z_3bloMDrJ6
zk?*6kaB^NECsZu@?yY&7UcKTD+~J?P^l*-smOsh8@VQuSq)ziL+dzD`?UL$Tw%j)s
zaz_BH-?{7Ak%fW5uzsawNrGyL5RE^+)yo4Ukj#t<ohSOF`&AQaYqes)uy-ItFyi9B
zB&>O>s5Og?T105;tDOo6MEXJKlG8W2BpL^AoK#FNOEA0_f7OAT>(=O3-OD&{=p>tm
zBbG3XYzEz>MLxAjT#KQhy&+<hQ>v`sBn<sv^#08*k$#em2FuvhIUZR+n}zNeflSR<
ztwxaiI}0QJf^UjGQ{IlJI^t`o3>TY_r2^A*)>glIT_x0)`HWq@7tayipdCalx~mMg
zuhaKaT8O>C(vwD-IQc5;el8PKT;=vGDfgK@B1RqQG#h)htHn2dt~6}q(;xdwF9O_O
zkgT^o5ja;EPIqRVId;xlWV}&R%addxZ;_Sr%4)D9t6DDlVdOgb;{1%<3CpW3wt{X;
zaB%0jmHs%zZptuJu(7c&WIJCfE}|Ur<Ncm)jF>^<_uDhf&_3NH_9tsnD&K@YAE|6b
z3B_Ykf8u56Ev|h{Dk4Gk?S+t;Ib-`c#!FDs0jhwFaxvvxr?9{1)AhsP3<f+(;ThP_
zs5DYf=3qvzaivdRL-0B2imOd1678&wZx(1)WU<U^8aBJvq|d;z*Q}wcwFu;y(-?u2
zHme&0o*o6pl%b|>&QaVP71tkk^qmz$M2_Ynba20i^()ZG?pM&m>Iuxce7WupQmTNr
zjrn<p*lf7n1}Wb~U|8k+s(?rg>uy?*dt4bQ*4qD8SLChFW&gcblFulg!Vkt?R>xDu
ze1lRcSq6jIck;SV762n{A9mDCi*(y}I!M2|&uw{R&!XF7OH6Q9bZGxHUY6+1??Wu;
zEu$!hyEI+9eR}Eq)uvo+-%L&1Y}&xIIZzQlU$YE72)mu@^_<y8EyU??b1Dw}g7oZC
zp(!YKGT{QZtdR2Dt_V1#%Jy|4mpP`D<8zpg@Y4Bv{e}VO%!&zXWS#T<g6uCCX_b`M
z&)YOL%(b7;soZ&#Q*rwIm~Gvp-%+sVCI3<#V-%Hbng1q$Xr_301!b%1jxg(KXEDjQ
zr5*ZC(NJRGCxRuedJn8a`PXx?zHiR1-%R(s1kI^+h`}4d_J@Dt&?yr4Sk2GDG?`dy
zL*jgOmRPa+;u^rGGKvy5__~ZQatR@MGA0&fc<O1F8n)2qs?FVUW?k-B@t;gEWC<N8
zcHG`VeA%^<03P@tO%XWN^A@^Eu7slD`6FGqwY|&IvqJ-VkH{M(MvZ|w?;LTKG$4nL
zKDpX;)VKsEK3M3v=qTU3%~d5YnkRAA={UtKixiGucKww<bAJnM<<B=*;+TWYd<=}}
z!Ph~q1h08h;&1Feb}Qc=;#tiJKCM74{9<UgH|BN#$@=+5F|QqgZwr4}{xwwxPZfj1
zHj$}tzZ3^bi;T;>&dv+7y7xWbWRMC7m4TEB5du_eLj35WGYct#zTeB}wpn8sBm+1&
z8V`#OO)VwJk@0o?vMLg1U5O*-=%2d=cWL^_92cWQywM!6-iPB7+?Kh7Tt-FZ)hgz<
z`D$!ega$#<#$E->EfS4h(Y{L{_851YONNt%A-<s&M5~wr>dxMEA7nhy#k84l2m)s=
z>9b0%u_YcSryD!cy-AeleU`bBA*(MrmHer+gg!SW@a^EindHk5Riqx}XCuD0u%sp9
zfgQ-a!uJ~&Ro=*l)RJ2rX5*Pa{6#|O7y>&p|IZ4)r5Cf>8EyV>BwFdr72k)ptIE$t
zg4)B{y97<2mTKY1cg^l%<@ht;zV~)S@^+2v**q~WVlFY1F@C%1j8C1;JOod*ZnhE)
zPi}=xBt&buIoO0r;~$yoyi1Mh|AF?0eDOyFv3I-^!3rKeL7!=x6ZE5qB(G}>h1Orl
z?zGmf=Y3i;YoJQckx+$fPIV_kO^5o-o`zcyISE3!vfUEN)eq-#ClB>NS%`#PLWsNc
zaf@N5rJ_q|Db7>NW#M-^x0M#?3gR{T0R{MnY;GSQr<z>d*Qh)mp^=}TQN!Q~k#0?D
z-{_Y{8)j2MfJYuTHPt$SRW?sNf5ManYN)a}SMtvGF0=K4te<(V(Be<O7#>;PV^s5z
z=-5acv!u>@%SDO50<95cQc<5|iJssMT@@y!;r_^tVi@Tv`RT9+7Ahz;;rN5XLq3p~
zMld8PxFS{GWn(_`G)+RRBr}2DLO<vE_4RA+P{b4ReJQc-=lRuxf?wqo$3UM(m^HM%
z=&lqRvCNTMgQl69!inB};}uBK39r)c&u?iB4e+cqr0D-Hf=?LX=zJu-nd>q&XwS*`
zX^5OMjADX)m30MP(Ycxjz89xU)Xk?j!e0QO-%K-f4?8*h`SK^Mt9P&m%mEeZ*F5(x
z!a6Xg&}cSCqdsxdW|BkYhDGnp{*Sp;v?FCIp`Lme;*i%&3bcpbJLFiU9&bkIjHWY9
z?S;u!?cw~Nk%CNBV&nRHu4-n`!jW<^I|Kk|=rokkADT|=hpaSrd<k2dH93e8Z-soT
zt}|;zWTF(FHhL`f?XAO`lV*LAO`4^#g47*sy4A)#okn?fZ;!ViT<1uf77o#Z1$Afd
zeaA=Q%o4t>O?(IL`;qGJ`z<U+b~R`ivv_fmATn}IABlX1sH%$vUi#;`V3kbH)Vm2Q
zZA|Gz8C9Qs0tngXw<O`6#DWPg@wHb@ZUdW-U)I99@4<w5XzIYI0(Dn=8c?;Ui*N$f
zX8H;0uLMmxK2%eIo8&Y7C-*A*E1l&K)X)owaT7Z9a8`X|iQU*?wq}2BYuHUdHQjZL
zm%Vwss5Vp72%DQ~73xt@q~4W8kE3hYro`?HTK_y@T}55`nw=05MY{S_Dx_{KvRoi`
zoUmG%V<nmb*NcTcmpZ;$b9Ttes1+kW3)fVIm_MVRpI9SOgg{MRPe}bu*P2|CPwy_e
zg+6pT$c#L<#~x|_`v_CM=S4@W1!}5UJcgf87KE9>&2$lO-&{u$1_S2M#IGr_nd6Xp
zZ@mS>DKEo><*kF#Opz(0*}iCqKio^j*Y2kYXuhlKolU5!xdw;vrz7uJC{@NyV|N?w
zq(17URov%5V5_K|i`A393JI<q=qewnjIIexky*l+v<F|hbSCucZ$Mn|CWx2$$bd?1
zx79Qz$qFVj^<O(h%J}^As>D5$ZnhH-#=iKww2C=7eSYDg)+?#c*L=V8<<Mkl*CXz-
z9A$)fMbnYsI6&SOFE|@A<HvxPbIz>C@)XlnN1wm3qdT{$+{M`HjTFi5)J^AyLEo2f
zG^Cf4W24nw9~*GD%Tu50YxChUrAJTL2PoeGs5F&lM+5D9PfO0sU>8l~1t>a4ZAh3}
zlxhHV?PSq#>5Jh;3tWMvTjI^G2v5M+7-ynD6>~)k#ET$8l%Pw!^W<;hZpJQH$?k+1
zSfBbW0NCLA`q5FDNs-vzE!;ZP1oZ?IZ+zRW0V#j7C!4E<1Z-uNK7F!r6^q8)X2KI&
zEl*c}h;KJv>g?!sz=;!3lm_vb`)AlU1V15`%F7NUFfqXEbt(B64mS;{Z?p#F3L&3z
zVh~Y_^lqc+8@{O~EY<-KA@eIy!0K%cp135ZGy0JQ2+#^-<2?PDW)i5%>oZ2I6(!;o
z9Lc<rDfdOY=+$=vH&Nha|0WhCV28onL%oLNW(ZtEY+bs?BF%2HnoJ%(MoulmM<Vjk
z0wZ0+%l|Bq6K&lJO|j0(h(TJL|JzHPtA*-4?kcukOt2~z3QUcDk#Kd*JA5NikAW}w
zVwA9Ks-c))OkYGoJJ00uKIQNxvCnOE#CP}NDaurU#g_6z5c*CAX|M$?_Sewr_|OYF
zdF!8$X|Hw;8K^ju+t+Q0Al5xqxR}rwksY3P2i6z58e}MmZ4Hj!FL;-&Uu3?{phR?X
ze5ZQ<)=(#4@ui9(g<66c*RD(#_fp0r{V@2o3Z`Mnte<5Db7Sc|@zOjtj)mt^6W=U!
zu|-Bf3CRvOEEeaL*$@)<h|k{EVxGq(b>s$|0?HBc+^d6`DYmVnpNQ}$i$jzzp}J>o
zc+S;t*<>?RTbX0~1h5G#2G#52OqfIK<4m36TfY|ccv~yyRGU(CJG?xSA2b?b+*Z-%
zT<UpI;l^na_8j3uz#wsp`jdnsgdFhR&eY9u6AXW_D3dUgNd4+M)oDyD+pb%dzLI(L
zQ-z74(qSOpfSqsm!MeRRp4om)m3(xaO5ZYj1Qzag7uv}ituIr9t1az2_D=L-3${*R
zJm#+?hT6z4?1BNB$67C2c})&N&)apd-(hwf5td?sG1J%ZLhe!g<4=#S($$(Sa=+{y
zya@^VXkie+x;{*=Ym;x|MxdyZC#bl3HzuMU9be+}KDM39T}av<v{ZtP@ZKwy{S9Us
z>*A1P7Cp3vicE1;nLeVwT5j-6G#R6Kij7L$b?Dgp30TGAx%%_L0!h2|>eqz}@8~j}
z1h%dbzS*S>N(u`u0O~6vaF*nZ+Ss^SkM4Z}jaTmwX=hyZODDsG5Gu;Mdf+h*pSG=m
z_8O8^+=8<(f9T8zjz@J{y^U@^Gw{2#nRLg-qWa2if1@5*2sd81&&}9tuGsia9$DU`
z=Nop@`a7Sy35*>e_h}uw)^%8;1E`9KgL)r9taPWWzV8%lpR&~O&MF#ST;WN0wtW_V
z#_(K~_J+5;n?QtlM6bxUvH5gnBGpwOr7;@5Ho`_^kYXfu+m9>beB9GnFS{r0JcwCI
ziFyrJi44td<5Yw(w}l4F(>w4Ca_<|^fjnH(7S18=C+6WJWaC=`TaED*;p<7Hd=q#&
zrxMy{NU&>BwZ2UQid~B(f-F>x?{3xcgCw3Y&6w-H^!rE)%N@0;7<|%B!<_FHm&i*!
z^g>026-SWfi<TU1hju1GZtNGyFq6`97TtNIo3e_E5b0dC?d*KNP_wUE#{NUq9c_a7
z#8830kOV!C{lN0W`&M&WXMt0CM?&N|bqaC(g1Q3b-UPaUE9raN-m!uoV=Vld5o<e1
z4QD&jP2Z)n{F383S(WppG+T!xrE1)Vc^oon*wm0Df!Hb>fpg(y)q>a0=De5#?3uR8
zHRF9U)6Nim=PT5WI@0y@rHY&#kUhIa<Sd~z`O%6M`D1sGKIy2iaMy%y4E$m4z@<w9
zlYs6keivj^whWR(BVX$DD%#g51`wX0Zh<Zd&vSk9R~xHvPq}k!9W8ppyWdL0Cw~r!
z_lx+3)>BS0_L3638Mekw4WK~qqdCS$T@a-L5y(>p8y!nka~zd-rNNCLtBX|X0hWxk
z^GDluLe6Ca@~!)mPT6q#3|)zjnWyml<2dbf1Q9G&stdOqHRXxgs$Q1}ed4eDR4A5?
z;jl|*ccTZjD{79`us~D>&_XJ>EdAcWJhM6dc}hf&-~BvNiiW8hO;bYyqb*x@QAwG9
zhsgJ2<re<M*rE>vW1O4F$Qtz{PQEqd+Gj1bzkQfZ$*@L~<J&Mj=g(k{^@&-SLFa21
zD>i*_gjLFrzt0}|<s5l0RfTp^l6%?&_1D_~=><!bZz&DqgKI41sfLzIv=Pa&PDY>`
z8@)S~nGh!;6XJP9%x{8@vm%XEyV`}S(O1_w?3!9ugo;}+SQYwR9z_<hXr0xjNeQ8E
z!fZX`*S4SQD5mu>959#M8t%xTqBZkidZ(rjQ-tuDn4^4#>zF9j*Z2Xf8y^eYP3O$T
z8Ioie)YR=}$9H^*rU{5=sK?xO&(3;%flxUSwb_MJ!lgqMLH^<-TJC+sg61R$OZvS~
zg}+NK$NZ{sN9`uLSYlQDtpDz}SveeIzZNNw&`K4Cu>F*(0RmPELNVIx1c=43HWn_d
z2PZjLAYX5{;bRHjdK$?XTy|5$b7!AH)?@lDNoi~|qvuw{pRX0k)=^#h-GyR=U+)!a
z8K4b$V=dFgXv<u(_mNSQQr(5$eR7iizFV^L5%%Yk8XsNCZ@rA@vQgi~V_4zD=!~L0
z#2q@>wh(9MHAA{TddOtVyK#DBw~p_>FaO>(%QkUo8{6J+4}12VJ+9|tGAB9Y4J%OM
zNTO@ETUsi@xzB5`{u9XOZRCC<?-vc+!R<3t!p%B9FHOhNwY&FGeAXvt)1K;*;qxhe
zg(Kb&Z->&8{ra>pevtl(p?9O8<+H<#kQTStR%XW#J%2)w0?1~J7F8%y!c$;;T}Dq%
z!g)h>GukqEKY-1(eK#DcPxK5+!6r*$!rGHwj($Uq(Jq1>W(~e=WB$Y#Cv%pBN4aGh
z5#}jTy@f}Kn-u#Fjp(rVTow)#H~9KBAskqY6u6lElc*^rq%I;W)KMOEZuJ!%M$7n~
zbzh{*nAj2*Ye|VWrPYVukn;XJH&_{lHuLj`%bZoisiBs>UJVg8Rps5I!v@L_JeZ>O
z_$q1RRAIB=?N{T=<r+lx7cs*l`sU{|OAC$*4oY0-P{a4PPk8g@xnf{h*Gk$jFKQ_{
z+NSzL=J~i=<?iGUk`{@9OfQ)HBwyC^_{v~wjnI8^`0$p<<ulXKMN-bYk6%}XNViuV
z7|Nq)l*Fgo0p?IeN80dLF)iJw-9%2qFlSivY=SpOCyAjy0s=7a>A)*%4ipZRQu&|O
zxIhC5OeuFt+iA}6Oa+$uC_HB{lGnWJQ6goo>s*T-KWH0=Dy4oF)k8GyJG1XCK8+St
zHocEm$_t=SaMevcLq8ZQwmI>^QJS0qlGA7+4>=d4iSf^O@aCAf$9>3<+dFc6xo^ay
zbd%x2!j{n%Lu1y0R2md{HzX}M-MJ5c$bBp5geR_XE(#)3TuatBAdB#3nHj0_Ca-vN
z-VFXx6o1C5m~=t$q`CW^#{JpWxMBeCoywaiSnNTPq7;wjriPq+ycu`oEnM6t=2#w$
z-gX=MAnI>PP7HbjVm>=$wZ^J0M(O3z>ZpPY{53(wh<=#pSq}3;zB<73rgYW{V(wL^
z?uvS=#NnivK5%Nwro_0NCfP}i46C%RmXuj=stwk3sc5Z#Y>q^6Ov%vdly7VySxtjY
z|7fL2`5DHTP6}EZOEku0ckJj5*5qgBZM|EBXas!I_n&PVM-6%wJdLxS8thPoZVjb=
zL&1H4kbgy~hb6*{TgNr&D5r+8#k)+|ql8Okgd2^=rNxC~pX9dy-57Pj;a%!fM=q$4
z^z!4lzxElbniZ0sgwlP`Vz-*TVRYxUh@^=MBkStU(hKjy{-Vv4S`u(zz*T8q>0b<Y
zh*GNij%79^uJLJ1^b~9?j4Fb=6|!xfpuTA(L0X!S65v8pZt?Rgd7QON%<Z><J6t8_
zFbdYkSCwS*F2;=;-fbpQ2Yhfsk|td>%cB$2?7~4UBaVq+be@*rxBXSEm5Xv66(jgW
z@0H)VOy(<m2x`OHJkBF+J;GyfDQmNBzWRJl%sX&QQIbE4eQ@L8fZ$U<a#`q~V8eFP
z-{-gTR_@HQTt0#8V|3PY!)9CGoBlKI3U~4hMoR~bf#E`Z?AMxAW!_oeQ>E$yr26f+
z<0jKb5WDk74bk%}5u}Eqqw&&Sh#4W-Q~J^D=)0#2*i)r&kJ*_WqU_oexG@Y)YMIT{
zWoSCQQ9EW7l)eGYmPx3zWk--S*5f0++8F>Vn&rNBZ1VaRQsiL)s<RiXr@%3?t=-z%
z_43QZ2}tc2ea)&pR+HzZa(nc{)iRfukH2rBGfn0RpGB08q>2VU2(^}YcEfJ*O(3zi
zojS~Li+OeM$jyLcX^!_^&@#-OHma=E+iauAz=IDl@a8_x%eKc;pQ6&C$V53!^`Npc
zmN-ba4ojk`N2F@=GQ3G6Q5fTs!o}s$r-E&&9uS^7fs?tK(nl_<5y)w0EXpt5*=FuZ
z!Pb8NgG~e_HV=YC5|EoTO<h#SbY=QxG2fh016Q@c@H-G%ZDCMwzz)?`<iqaDfGqgk
zMB`zuh7Backn&R==FIP?J&zX)U|pCj>4@<+I+f$})$pLBx$<4smn;>a(FhT3d~c*G
z0|=ciN|aU&$Gi*mdu{to-lm)vUx+9lTfv3E*tb*ebE#}V9)q(!uXwgY)R9E}P7-Tb
z;XolS+5J8T)yTJQx-C-xbe(FW35!d=paF{sH#Kws_Tx5eZsnzj1%xox2IiznL;350
zI(q}fmDQ({rV&gPR(1z(+pk7}gPyOJl&^#+C*AaGRy?0>RR*3-wY%|XtZpFczn<5L
z?vt=CmM8NuT+E(ap-oR;cY8XW0U->dXeYWHs<5lA(wY&U=}+>$?&n?|_Idld0rh_W
zeCYeM`78Z>oNb<?Ugnj%o^Cr|cLxJkC?0n}V3HK$%wW}a7^YC+xdX>r(<MHXrT)*c
z@HA;0zEaf}wWqSXo~G-p%%99(tvk27+L5GIQ+Rhkr4(Lm;l`Jiy;>CT<rCp=n3U*;
zS}sKHM~PwRShWCbH)@gmNa;Sc1DUQ0_|IgJa?6V^vpLDVFGQOo=_IwLgkgtYB0t4t
zzI~n>4&Fu3oRW7oI;)G!B6$%(jX9QS>N>HaFWjlVW8jj>6gqJeLZV>f8YK)bsvR37
zCWEVpS>&dt`n;I7k#!hLRE>hR$|PLlkeM4a;@Of6cPkO!^|{iesRI#<mwZD_s|(!$
ze~59^QKCaz!7E(dO_Tjt4V6b@$35mqI@WkvLp67(C4}yY3->@`SsTs#ixfe+13%W_
zTTeUgRwecdNx7!Y0`U7Br`WKI2}@U|&hPg^t>y7YbE-EVi_U1L^AL_SR}<-zdE3)h
zV=41>WM6YOxt6FbyGbg=jvVRR-6`qDP@ku8aF&?Q2VmJ=l=g5_S<z?+chPCG5nbl{
zL@s!6lv*`5Veu**<fhLkF0_3m5gw~bJp_@p?M$yCc$OC<&BoL)nrI4s=bOM7*)GZP
zC~W7?k?1Y1%c7}T2v99mlkS#bl`8BGF|bqq^r<#s4uw^ag6Icn=%zwhW$*`=AB}c=
zWwxd};-rNwV=e_&E_Jd{x1PYUWCMw}HYGzo4SS!9TB-b_Z?`P05Kq+JmTrYq>V)xc
z4a7B7;25``sSQY2`+vueY3*G??MR}wN59{4G@=UI=HL3PBS&yvXNwXPGpe;(iiaAZ
zG6)gKMDvswpSgc6X4*@xz*UT$Q<Esbwxrv(ZQHhO+qP}n*loLew{6?DZTs$+h!Zgr
za~|$n)i0=8D>J|J+^LT^+b|LZ8A@A0v20oVyMWGBLKC4no@G)#Y)|;0cYoM_2tOXM
zX;9BAEv>zl*}oplbylZvFCLpr#~gplf`FVmOaLl&*s4h#AE;c)H`5PzDJpo`L`P5)
zr6M*a!hRQWT{6*>A3^;CzW3U;97C$=R%Y~2AAhm7TaZxig!%DLW0hiIE`4>6gl=Bs
z>mjtJ*OV&Uy$9W;Q-}9Q=9S3Xooaz1<%C*{k5*1y;90mh*<_{R={hZ|%);-ckAuq=
zR;y3w)oka2MZ-=+ZlI+`T&jgyvi56d60H24rfLfNCdlg!QGI>qF>+*rtv*eTp4B&+
zEPNTkS_}IIv17Z6{etF9=1Z^D)1T7l|AI-sdqUMqgO}{WLfP`Mwo)8S!M<pe{^O9K
zH?lwb0VO-CMqOUrAaFo(;`9T3MU-xF3U7@rtQum+fdsS#4;{Flxw8niTuR9yb&Y(d
zj0UVP%yApBu1~u+plaQ}>2N;l(`bS`DbYaMJr&<3Ggd!ZJP|K!Z>E!LXs#_jhSrr;
z6LJ}IrBKax)7;<ZcF`kRr6s=0BZFAjfFd2nzBgL3r4YgOuTt?&^~pKfi}5wZi_kL^
zM}MhV9ji**F5Sh6mW-pA_pi#CUyv~hJ`3_Bhk>!j)pxRtRkUw~L6W)dzdWi4DA@Dn
zo14;+Eys2;istVKI^qNFvjO3PNSY)}M0b%I1z)z@-M{nYd<V9l#A}>&7p&IU@p%)n
z(S!A-jut8E-U_Kwr|rmHE(qx%+BGzCAfH#pC40SR9>%Hud%0Y~8G?kUb~{$}vRd3I
zm38AH@SzU#I}@a8T-_M26cJd0lte5WOw{U$0-{(y-jgp%7pC`0Kpi7U2F625!<CEL
z+Z%|fRq#V6*XkQH;iKDe5EZ|5)H<@K3izt3>%na4<on_ad5mAfW)3~0E;}!M^ACY(
zEI<>*8iTfQPAaVg*Ns7kjb?hwE#1@uwU&}#U0kuW2BZuOp)|@-WmR_eP>)_@_eS1K
zw*5Mq6&HKtpcZq<qAhDwjfl;V3v6=KGhy0{)Qb;rhPWg_Zc;ns$2zOc%?Gj>e-X0U
z)ReTCX-3%F+=1W+)?tS${}xikh$~2AbA1^}Vv+(Rudxl7mmQjK{rGwC@$S6buG8N%
z>m|}}itj$92SBzp`lErH(EujK(n6MxQ&8cIaiGo+NhqH?u75zZC7jS-saKG~O+j0v
zUIhP*Co<h2wg&x;KY`9?JMS0n1bgJB@TxH^wjw<H2P@)yAPgTUf<f^*BD*<A^|aEF
zPI0;&FNJprK>I2$$PHzze`J;7HZRij8O14BL=R_+*133;i4H^^H<otkbL6xv2xqYj
zDS2#(c`0&Co?J#916k1=X^$E0JrzYX&ci!=c55WtKex=NS~nErpcQ~a&I`~t30=^f
zDA*<&`Jw0zg&Eqt`MS{dXYU!8JvGM<MEMe=FQ?;I=rlsrvI|WWV1CrkW6b~t@(qx(
zYB3}*D<B7HBbwdqo%We4TWad;9!5dZ&a0v0_R6r#!O=RJ9B7_CF9$PXKHa*YvhhY|
z-n?L9j`^FUy4f^Ca5tLJ<D6nNJj?D(91;ghrzOnZCPLbvxyieC3dJ%V3)f=})Z<)~
z04?tm7u6<vr|x-kOnA8tNNmZ~D<YsbaEni@;HEAId!z<q@S#eKL?CAN3KaGk{!rdi
zX4HOL=-RU`I+M1Anw><OilEw&zo~6SbV;us9ZiXNJU?ppDEhzQ6bx2m$>;5L7hYC2
z`sl<}=x!(%o5f6}uNsumLhvvIC#JMLGsL1iEh6W!g3IOj<O3W!5RTM;L9{%%8^%Up
zGZh-lAk?orSH?L1B~dr^71uj^5jq{0(jYL}QDQ~1MLI#JJHMnaOR*u&L+_TVksRGH
z{j;K1a62D4MR;U3jmW*Cn+U$`NNfK(`J1gCFfEEl37eCz3ow$Nyas-rN9$DcK{sd$
zT@+DhkEz>q1Q<2^s`ZEA)&zW;_`zv&MRmcs(zq!@L(zhkxXHZH<wkm^BKGz^RBD7a
zqzCGQB$-PAFiXW(2x*ygjiPqGMA16mx~o(5C|<B;c{1$Jd8$1UMbazey{8c0nFc+D
zt<y_dce%ejeTwS?)mB3!z`n3=t~Vc@+o+51r2|0Oh^qCV!hy~yNZpdmL@sNLlH&L)
zt6C`xOuL64xLjdf<?~tl=j%8I$X{P12JUJYjV089VPkI=#-EkhI|c?>#wib+=s=;^
z_C6H%zPCi^g)y!Es-qj~x8{?#=GQ>WDSao%haMKD78uw}9`|xZKi%Lq=;dVi(CbbU
zN`og0QYWWqIcL<)7{mt4?vg*H-wGWdbb2L%&X9(9FuH<E@)5|{+dIt1(SJBa+s$hy
zapUqLRa-Va3o;c^zf$XmBLukIT>C^=*t>`p1?|4Y7<3})nz7AAvs&-1%P56-kNeR4
zfVWlr#B~aNpwzf9yK@Y3^|wc@;Ye8)y>cQjOYWIZxCD<ljiB5R`GyPnlD};^hfu1q
z44SJ1{GUrT<A%PC{9XpAv|;)mWjdX$TQx#2pqoTr253;RtTdGu_L}}!TX9M0!MXT-
zh>nwe`qhL5$KLH@5NabG1{%QlAEcPH*0O1`K#!?EY2^`snXjl=TbWiSd;1a6CmsN0
zL~Z+7y5nGw4-H05Yb~y)F{(bK=XIE-QC4}w0sN#)aRL1PZF48x_yfMS`;O^*{tY3}
zf_6$KdyRI!dd($Terp}HZ85+^+_R6ml)Oyl6IO;N+cD6G4u|;QjV2-=jU;=-inS|8
z5ulp;@*ubJZ~BEQziAO#9f0U8^-YiPI;y!Y5V&#&{S#pAkEi>_VVPkbIPWdR!s)e6
zNFovKO!)(i_jG~P1X4Ep5~rjqY&n)RDgQAM{MGs30+Minxy>rww$o+FiwBV`({5(h
zi$K~?@tBMpn}Welpul^dfCbe;nLGI7gr`#THo;_HZydirOx2rK{{GxFf*nmLoV4OH
zxwrZ#*3Lv~NyWi|@%d8KLX!nvHnLEg(_)<`9tjfrl-Y&SAq1Ra%*mrhBIeTAa6N$z
z>kQvj1`5f;e~TSS(Bo68@AtDWI6Zf^fn{Ph>|*_|1Gw9zVRl&NWoeYrVnOuG`EK@{
zThBVAMjgF}ypt0PIShn-=Hd75G{9;k`E+?Qyx5wIfUW9cw{(rSnPf4c-Ng$#hACq%
ztV^&+^wBv56}i>%RC>O8)lTBeV;TmD-8e$%N^yTQ5lOa;&M`+S_R+{4!df*}rRIJ_
zN2qKjEND)FLa4B>yVP=n5RqT{Sxq<o+h)E~PN!yE@et2yyCL01^0PKybrE<~DJUIT
z2#xcc$*+-V2#ZJ=Cy^vIM%FwTCEI2RN-0U)bMk^A7icM}yX$Xu@V@47KKAB{1vnIJ
zK<dZDrn{p^JGxg>w-lu5UJGD-7K~`WAyY2V1sudUoP2$Df7xKH_XZCfl~ONl&ynN!
zeIQr*w%zMaEgW36kdoC2J&VOW>w$WtDxA8wR&f=?LeE*J%?9`e)`v|v1N#7>D|CH9
zC~cNL>zLB4BG_bTJ6oEsI2(tXxnv5;lp*c)1A6%=z3447aj=YyO{`zGX<aTx|8%m0
zR*a=!q~ta%kh*tM%VH0IxBE#^mc(wnDb@BqjX4>QC%a&ol!Esvvj-rq2vyGN5@DsV
zq>h75r_p^#KMYObg~>vZMMl*|{1$%;1%n7x2d8&pg_y1G?)=Pdy3GccPN<w4p{*3L
zDoq_X^F@8yAt}>#Ab$bvM7Ii!Kp<n8q2aPJiU@nuW`Y%#n?ed!0`><o@uWT#<$a{P
ze$zi>jN<XqJToXD9Dk}68|Ok$E?|ZU#~R}2Ecg7ZDCTh<kNhFCY+NH7LMB*kBCA6b
zx!Z)&4~oHc2?Gk!=Mg2jn$fH7SDZf#MmCjaP>W{x@Z?Bz`g_OFd;2>f+Rnb?C-OHg
zdJU`^vv6=W8mq<EanquOdh61<5raM-H}q$a)Cu;~Nil3Nt#z(&03?@xK1!_AJ#5D{
zqeM`wX@j4C&4aK-|Lk|VMBHUNJK9CJgJOOp!H?DYl=&S4G!4t-5sQpdW5%DiQi|Yj
zLk@Og)cKbvo&E5Npy6i&0XlGiNX!SSNwK3}!1^Z+ssCp5W@G#>MsG$IMz;SryO{{s
z7+9J9bNj!2-s}vFod3t>{lD>fyMwDFZ!WMcgybuUaw!H1BTAfgzf$3rgdq^5kdP!I
zMG<sM-l|qmbq7l%QV~cXmIR|vQc2MR=_|~3zi$2NUw*2sNbO{Ln!QZz%r>rK&t$pB
zzM)<xx-qbl08PT90Z0O-y1KIV6A(a<&;S5I0&;N1?rW&8?fL<fQNqLq5+f@81}caR
z5@>YEg9Z%)yCy>9uK=qPXkb8~B?CoE2MP!fB1k;g6ClxosswuUT?cjm3DA-Ng#kKH
z5i9Okv=t$v_0Ifj7qz2+N1&mk^gC7OQkaJg3lk=U$fpCj1abbC9S>~-icL`9fVupR
zN7`bN)1Z-(QkYv@R0cQ*Llh{)J@&W(;N90C&nJWd7WVGD3PdLcekQQ{%U=Wuz!5ki
z=P-2}fVcp72`M-j;1N!Z10`&5M+uMc!oY&i%QYvV3|<}|MCcdn>IZxS@M{ByKm+ij
z-p;>{9=aEyKaN0wavl~$(6tXCo)5f*3bv%ah}h?4cnd%X@d-^Zgc>|_D$GORpdNvX
zcUR>Kp&&aC0MRf1X3htL0xNFF2!zpgIUEy&ZhE&_9ragTtfRdjK?7#5kRO!;3vBqd
zvvn_T&{dFt7m?3j8_pr5dvF-z1zsVD45IB5IHPm|A7%&it#uYL3^Xc;WGIwWB#^<K
zU<&$z#G^i-o*n9i8luPH(ERREkWf(o{0oDDzYQ4X7x=DUfX6_9<^ynQ_LLv^w+X<2
zVGa*&80gv{&H-d^clq8fW6&Pk|Hq;YeGVw33?Boc<L&Em9BP)qn`B`5bN5bPoW{aM
z{<=1aayKvW7ga?`$P3W-Ei^RXOQ?{50E2`G5M`AA?YnVC7vf7E|2MHJ(m?=-K1%l~
zdbhmS`{hj-=!PF^AMmF!0}n}F7bD?wSEdytGARFz{_Z#PwpZ!LpQX3_p_l2qms`nu
z@8W}c<{SI&cZkr=LH(8=g2xr#;GStt^k6vP&AuG-s@2Wa@Z}EN?iY7?Na)T<L4@Og
zavB~D5g_vGWsuRzfj;9~mjmtMDT}T5n+!e3Ih5GIJrUz=7&UknDDa0mOdWFx)IZkX
zZtDv-$lxw%@P}&*8aoVqng3HkR3PyM{Q?;Y6<BD_B7z;j|II@h<qUXaatZI`IeJ?_
z5D)|@S|0Gg`~{ecU_Kx^7!4T+ic4?H^v$*p-0xFBKyWphj1K2l1NH+TcBJ(&n(R&O
zr^O!<AM7zm5HJ~8qTJ=Lv#Z0TVF^(q`R_ym%(&V{K|8{)LLe(<i?-{M8Dl>Mx3$z~
z`-Ta+#`H$Jto-;lgZa&1+^iF<cTMXUAAje@AL35kt)cgyl6#Rn;52RIlvoh?DW#OP
z;;8v;5O>X@@{RH1YuwIQ4octO%A2o@vvE=NLb6)ag~HVhJ<X&N;aMWfkK<&*mhswI
z4^fx&{kkoAyLv(C`831|FU-~^Q>}Oa$Xrwv=<aV{<|W-~T*W{wym+-9=MbX`)sjyQ
z6%ji1j2WKp7v|PAW9HzIQMnv7!^NJ@VIKJj!{(iq#fO*DW!alzH%IbKl5mSV^esec
zR;w_%f8X--UMx%g9OMzXr8(+!j{wGNnQ-f_T<XuEi$YYi!fRgE9nOZk$3VBr(qRXA
zhF9EekoCreV|qW=>^dVJ93jZ8-%6PRJ+<Sa*@N{vX9%y2!TmqHjkCB+c4Q`mi6=`&
zH0mtGm|Hjoa<nu)%fPv6c22T?Z=7$B2fRP|I1(GnxNB;pbd+yU6(cbISZoPR$JmSY
zCm+)O-OJLl@KSAJnpw1U+-)7Oo|h_Hh1EP$53_be{_eO7{BUGZf*Q+xv?-!x;s_JW
z=R8nbe^OF^!_%fT%;pu0)Z2+Z+!pv~<sCx<G7J!xG<dovyKzZo04@g?SU!-*-R-8c
z<=Qx*Sov(q3R19pn0be9KT-JxphaXCo!hG@aQGHMrq|7rr`}cmI5EYNyckA!3r;oM
zE;jWAl_KZ&+pxSDRmY#ld;+Z@9};V&pu1u_7Q2s8WsLXlUQi;!mLMZ|zj9tVsr^gL
zyeOA;#p)KVhW+B)>0CVhd@Ds$3i>$88xKaeYM-@j^gAKZrjwWR?giM;9oJPCHXnZq
zDMz)p)quWX_Ykw<t;tqO*jjeb`kG_I#ohC%_t2bH*Q{!A`gtrEN4hGp+2>XxtEn|h
z*wa!Y$2jZ;@5#k;3u-G!+a@q0xmz_2In)i03)o~S7b6BsDBmuD_c5P8^Cr%YLslJ_
zkPL|}z}{E|G2`NE4r|!fI5%R246IyUzE3*zUOGgc)enkA0`7_TT53)eip87zFDfxQ
z(s8t1AJ-y1_7;b2K@_Y26B4mTkw|Admozt({RD#>-1;U~T0Tmr=*L)9g?1rM*r+1c
zpeKUR>^1((?K4--*<#Sjq_%zQFvD;a{e*tRU7(>?x(i@dtT^2vMRPc1j-;^M*`B$M
z#V!kP26TOCzXC3?8O<}zPFI|blG@SeawX`FU5Pu%%}h8Qj@{$8&eIVD=DrN-QPgK(
zv(u;zckqvK(gGAW&tr$RP<jBB7~bvek3f&zvp=Uo|G2IBd3wkQmoB#Y%_*0q%MM3f
z=`v3GQ=8wFicN=*6BsVTEN(e6=-$tF3uKvjhTw@9zk6CP`@V+Ri)G_T0Q>#KS~o>5
zWbTqo-PMgdXP-;kzNG)Jt@^IUPEU?6!Og0Dh2z#3{~@cud}OT<g&>9&d8G*yVT&x@
zLeJRr+#vsz?|s8#nx2CmrjyI`ho$^I&7-e{o6Vztf2}UZ%#|dvj7cSEZOmbcXZ;f_
zRtBw9ZH!ZMx5JdNJTt2CTdG>p&3P;+e#;}qK+kfR*_8e03(6v*ap@M)>Lcvi9G~0^
zv0dWN*yKtsmFO<20a)kDYDWYcfy1Xa@d`AQt?PqHIkkhKW!i;6$qkvP+S)jz_U+9M
zYGI<b>Bii%DmH+D`ps!p-pCl@I*=aP#t1v`=ng1ga;sufG0r)eFik>Xz(s(KB5=Xy
zfD+6)?36Fl>yo%nU-+qlvh0Wy5u+V;(@o7FgiVC6fWX|9bN93YAvFa~LmPdAGhOHK
z2U^b0!k~A@!1$k5VPWrviB+f9!)^%q4KA|nX`g-WE?3<|yh1;qS!mK|JC&0iq~N9{
zCH>%S?b*9%VkB>}qcU8O4>MyeN&Y9MkB6sxgL;k5UjYz{O;Jk1uG6XKtZegv?Dw!K
z@>p-iRNf&yJeE3(Z=kg0H%=yd(~bG<quRmtatw?O!xC$+x*BeD7B|yQ1)597o9L9-
z#3-DS+nLp=Y;NiYvkk7eFif7hO0rYB4;o$s4NHgR5MB*T)7BR?!H>VnnF2^HGbLvI
zxS$>%%tsVA<r01ogkZ7=@2oW?>t?MyM@-M<!F-wPTigRBNn1(a;`!^rZy4zPAx^bW
z^WD_;V$yLJJN_i_)dzX2?dD#;?c3;xk2ukSbHRe<UYyI0=@$VPStr}qT#hx*;qqRG
zvP#si%;^CghP9(3STw#7^USwXbQzNJv+wdU5k&VgY9ve|8NOqyQMek0sXbj2G1gGo
z%hab*W>QN2_CRXo(5MJ0Bq=wz^Ic7h)nU>H6jrbd?-b>-5=S5Xf>V>egz>3c*b3L@
z>dz9w!hgfkA3Zkc)=BFRPP?h>i&s0nP9N+m^xl`n>J}nsx-UHt*pITDWmfSlmzf)I
z6Cdhd89N1p9%#-SnS<dhRRLKl@Nrl?&1rEkOd(w@&q7B%IJEtEsLP~W{PpzeW@Q)5
z#{GZISctV}O4Y~(g2O>=|Kh8US`l>Q{my0d%Z<t_*hwsfo<I)<RtKIce=|FJ&FAg6
zW|k!JZ+Ti73D-HR^YzrW{)Xi`y;mmQ{CK~ud{!wtjjxt_hrti1*H(WaoP6*3Y*sUC
zn+hAzMj3lgxpSYI1#2Do{2G`XVlIB9$Q3qC3kP%yMx~L7`#MXVcJtlD!#vrA-B>q)
zziXkvK>olXtB54IE9pfn`*aur^UdHtzJe%qZTXBCxrB@0;ZWm?h}AO}aP08jUQc{m
zNS-P!GQ61LanGko;!L;FTOVD=&~KrLd7jH3#GI2Hma1k!Ku<s6zLzPHaTQWhv#uys
z!K@Ns72C9q(Kv#Z)m4CqkFEu&=EzF=H1s2K@AVTLR+=d1EGjD6Gu!Ix^qtKpUAXU7
zcwHR>nUj7#BPg#sZ57|Cbl!)g`A=4gx>|3qdmC;l$9FNK0O6WAiLNePd7kw2nQwT1
ztu;S2Jx?@P5$e2r;@|g&f*1l|5;M}Y%`{uDyp-sUSdj?YF$bQIeM3~i%D;R+<v3zv
zq+YNRUo8tEm!GA#PDDc~#wTnSs&wmdQQ2>zED~Mf=ek{@*7q=z0AGmsPCK35YoRLb
z%XzXyj8*0S8T%79JH&&J=ZO~EzQMVUg#2`_0l+J+COOnnWOW~tSN_zrKe^M~oS!7H
zRh5#e;lc@dJTgEbUa;T{g)7uQ);6mYEA`~v7ugQ419&mTStPs%ZbU#GV@p6_$iB{A
z#F^L`rAOl4li~f{sq{?A-z+^jS(=@Tl0GoJot(a5b8M&B2<6NM8s%T(XVJUMvmz(#
z7Cwj|cEeU{B&dS6h!^~T|EBlD8x*wUL#BAIG9Zh{-u}uNz1E{D|FPoK!4G^cAtXs{
z)nm&4>_uGE;5FYIAkyvLo*LhI;#(59k=`4r*i1cxAM?7gLD$M94XK2%L(ry=jP2lO
zrZhC}O*3Ni?pUU0>B{IxrB~{2d~C-89yQiZlizK=-H41m08XD7oQ`h6i$81m#tG+R
zs<g&?hui2@$GOQlbW7Ryz^W$fdHA6Ai>Tc}AYkeU`qq%Q1rB$q)sZBpLiJc2)b48h
zw)EN@AWzlAH1&+|#KE84nc;U)z@&+_+uBMKOOyPHJ*$8tP>ZjUU^Jlxw|oj|g~ezG
z!t=;}0u>@DRsR@9eJ7(1hftP2Ijwx`QUNgjJj*8hMgOhy<tm24=TGbF>nfnzGWij@
zI0}oG9yLYA(H^p`@l^mc@-b47{?RVrLxFir7W^Y<G4fq*KZ7c5<ubxddt1A+A7BH@
zh$*LzXN;3%^$5+U)d4D?+c!qlvm9<>7+m!@1c+OSZRn=0=$<mU?atLMBtLD*&icsg
zJ+c@vUq@4Rsn74_{Bu5p5Z3f8p<yGoxhpX4kBLWLH|FF~$K#UIbg?dyj5*Fn<>E6M
z`eZq$kY|L)jrv-gdc$-GZYAYJ)$yIlaoN|9`1&!fTH82y+hwZhCSC1VSUDdU!x7gg
z4m?-qoDYazB%yLeJ$xG?H*5hTI#%VNja1VbX-=m|D;s=KDR-!2{jeG=6!6{zeLU4N
zzAg=wq{zOQkOsvSiEyYwEle|vE~TSAtS{4r3r@CB@Ks4HLroYtmOkQ^XSOMlPK~wU
z*MlTwmFj&jx;d%a*66Sw3)_;>OykUZlk-n#1D|nF_{qPjwCWNPbx!I_OLzwAC{>_J
z&r%hxMQLGkb$4hn)9E<aGrkQc@AtDa)4crRzi>fE4R^rMj>*y2k$XC_BrjDAoFPM8
zB$CMZ(^!nQgC<jvm;do0T6x)8&Pguqs@*C5iNF>a6axl#(|*lojcm99^%}Er-(6FJ
zYnr4@Kf}9Gi1nI1TMlV-wCEIX4u+yRilrtBD#f5wqt-pOfF*QHSXX(hyJ=Nb;KDL#
z4&n|C?`A(R_E5(DR=+oj51I+$eRypml#FaHrz;1d?sFo!DxQ>(DeT$QUvKXnk=BDE
zRqm`Lo*$;QWAnmt_LG1m%_o-Ka}t0BYQq%0<CB}P2qP8u#ZrRc#9XrDuNmrd)e`cQ
z8ZO^rJ?qXeRNcOUo!SsLv70xt3N|x=Ut{anr<)d6UoT}N_<8sXXhotX0%7E@<Z@FT
zNy(3_?(Vad;`S7#8TB9`sg`7X`w+h)3l=A{16KdmeaJ`7a=bH5;N+%J@wCy`9H7`}
z7l`Ej3;fKSV6}08z$w_D4dl_@!!QD^hiUn-^)2;fU_2aM%v1@vWIhViEuFig_K2LO
z$&5kh-fD0DzzKv8-FzAUTH^ci5|br@rx@CD+7m~+;@!b2*DSDc(a=6;+Pk-0a9&Kc
ziLHX8XOU_U3Wi1A^r>t6Q~G{(WL-K(WBh(y6>2ziZjQZ;c9dEX@CrExUbZ7k78Xbc
zE#mmbp49OJae{6SyeSmyt37oacbeIAaiTXm9c$C}?$PD7RS|wXjr)}dSVE*AS%+mJ
zmH#(!q$6hb-rUSS;;$1L`f74{DvoUlvmR&mx7M^vq=X`WdM6Lc<48eNviAuHYcW%e
zqjW?xNi069WI^|_KBi%76Sa@>OW;LZzn2mf=4)wGj7zMusk;B*{a5kHL?S|1+_UBP
zSENymak<xlHgD06ghl+^+w2Lv1y<nPI_drEM0`zZ>H}KQ3e~xZ`Me`>3+1MZX6=pX
zmXKPb?YS4jiOqF0i302;`ul+*3divtc&Mw)#tUo5<x-3LckIl%X|wt4v1+_pb0#x$
zyVaZ9KbL<Rj?fVf8h#;YH<`jf4M<krO(qYeD?CxfA?d&88#oV_LMesxkHpiH8}Xrs
zVg@yhxVkFztFCU;H3|Zl4|niF*!_G&b_fLF?fx9p$KJRXe@nXc#wC8UY;3SyZ-RkV
z%#5U1g;cZGi}`8HT$PX|1(GwjG`A`1>k4sHvzKjeuO!nc<I~0;X^nW|iw7cfT*M3~
zs&?6x>calEL~h6n;*gbhcb0i;R6R{7+bt-X6esf1+jHlKmE0O$<>a(>_uNmqqcZ=p
z+n4At&j&}V4P9eFmrYAV@8L9~^ldR|d%x@P25ZVPYCH54!23}jIhFRRYVtx`P5#bq
zsyeUFa6+Reg7=k?!pZCPt!)s{qAp{7TBzJdU07)&GbOHGZpk7%vIi<A9zbA9Z8gx$
zGSJ;5eLs1MZ<f~DpW5^e;>3`A!odt<S!H+sdG06z;UGN`h4kgthJtIi+Ugn%Wq3$J
zNuC#~>x2+#e(bMvtcsXzMJF}8MbbVcI$;j{tfs4F8KdX(aCk=~l>;#g6m<($@sP=#
z2klVOhK|A}iU>P_x62f4m)6fJ!ya<WPB+xNwFSyr1J?l(u}Xo8@BN-ZW9*s_8=l9T
zTeb3~r1~SukM@xF2EFCb>1`54Wp#eE^=h}tTymwHhZ41z3+dVrst;Hup88^qr<T$#
z;!9KH?fZd{BU78tFJwmF1_twz3R1{-g!jE!YAO<XV~7Jy8oXPtDiYZiT{mXxhV~o-
zbWl$D`{$#Al_uSMn#&Mjf}RAtDQZ8!8XMy!+I>Dz&{|ddP<8!t{#afl5>i+~8+k#f
zn^7b242mv_B=?s$uyfDh%T|^!TR2z~H_SdfjPHO_f<tN6MY>y2<?Ct2QA(`%s-P>~
zDTM>-o1L&Crf!>PJ(PydFHOAtA^zp?^DnAfR^LIv#Zh<auy@bB5dJM0@{@LHS0<&j
zDV(BQT`pnxaVRy#7RSKlm~WLDza|L&OnSBhll=wTr?~OnGMxkwJNKp~4U|Fg+A~jG
zZhVJ3s4*q9Jki`bs+Oq>t1!4^LR3!frR!G11|~T9*I(-CqM&Dyoj&ZnE`$fVG6TkQ
zLvYbQ^!}tO;ZYnOtt%NBT(8(Oz43qkFkEJ9o@P+0_;mf1$+`Wv9;;PyS@tp646`A|
zq+d12uYZGv{0%pBY+V#X_u9}&)|dXPUuRx-<F|ZP?7E<^lsNj3vdMX}Wfz*Rw}Os6
z?E}vB2cqyCtewq9b){!`EQ8G(Ddi<dJQ*(c?_<MgB_o(^*+aUE+$xSD{}7>Q@CTH+
z-qKn+l41p)4A1%LMX~o&VpCb$-d%xQasW@cTiWmULP}!I18R`?!41P<4A6%WW~UGo
z3Y7LX1G`ovE8Ai#tSRO<9Z~-6@P6vvbLPkBq41;FNY?_OIHavBSC)PLpR7RSF<qTR
z&AHcpz%~{73Qg!;HpfagS?SGLNMobf*;$4g!;z92x%i<uW$aC^NF61!SE5d>g~LGU
zxLN26m#gDK;sJ<edUHKh4n#lu_h|`vsRK#AZ~eHTOVLLYvIl3L$-^+WY+`cdv9qv%
zCf0U|`YLJN?jIh@qYuac^jWWIUeR5s9W0F}S4sN)Fh(Z`rYXyA3)M0U9SM`m?aMJk
z!1EhId+q|Um{rcs7InIN>=VaH)CV}+<oGn7b7Uk~cPq_JtCwXI1m8JBYX|mUg9F6)
z>v;1DFtG0D%PoF-{0NdLCX!VRbnt0V{=&taI)Sjpq{zz0jfq}Xl||b2jyX(+7r!~j
z^Lwz$Z9}VE1VI<EIjRld(k6OZi>Tkq?Pf{i0^mD47xFq)tGd%OmA7cdE`6Kt()AM{
z1Bx4sXD{D)v&FQv{TyW3L%6uV8I6RuQ{#W_Bhy)w6z)owot4X#<4}XCSD(I+n`bmD
zN>^1rKreAxKZ(ZeDkV}8L==%j93Bw#N`o0_OfS4hb6=i!8s;T=)66|><p;fKP}u-C
zkQ$I)4}KPCf(j{mspkhKi3F|)C;kLu{!L!@)-f(IMu6Hl;H_T$`STT6CnyL<*2f>}
z!-v-CRyw6IBO#gnP)y#}oM%;KKD3ZGpK!|+E>6xQTEgHv!RV5EHN|~rL2M5@faiDI
z7=;JaJ<`eArXM#+?al@Bho=eRi0a}5%9>vv_xGGm0dLuVY=iBymf=clRbk((&N`_g
z{Vvz><&pX`o6C)ZOo^L#OQQxxiRz|%o;t+;4z{iARu^#lm0WQ(HIu+JQ^$Va6<@4s
zkcaN_{msTn?p|Aj{`ju!{+MkZf&f*cR60Awuu>7f9jI1ccV{fg;nGn+;K&udC`=VK
ztQxmCp}{(0jW~uRb2-GIOyl(*Vjs|0+5yqU%kw-Y1vCxu_zhA}Qs{H0#f6#e01OzD
zpv|X2^j!P9@$P0ehp)ix<~Y0ln^jzd7Jy$zWx}<k3rdS5bDSt|+tDd&1Nbza=7YmZ
zW@r+pCgh=ZI8+*wTMDs>$syZmyR|dCqD=?W*{`N#Z4#*h#?WtgzJY|R!dOm53MeHp
zZ4lb3G2YeRBU6U2v>9?by73^-6XAM#mX<Fi(W>HXIMCA$=cgF1uh^JxAJ<bU>`EhF
zLBjIQ1TO{6`x5y`JQ{{HZ@2^Jt$O-e5i`NVivaxDQ5t=s`>po7#63Lib2JZRg<Fje
z)TdD_F>&u>-3u7gK)ks27s4>NF0`#JUZ?TQqqh7cb-*_AtoD9Vl8x@p>2u%>AkppY
z88Xz`Ymj|C=5U)ra{T=Ve1jrO@4sq%|5UsG59Y+k!p`v@Rqvn1$HKt;pWFYV@iB8U
z{I4}W7f_X@?G-+WbP8A`;3Oqx7kBsYBLLtSIA#VISfC_&LZGOH8U_f6AeM!K0GITp
zg?twu<F418*PmZ&-`zjWe$F$EhgYXp=J;@Gz{Y-}<RQQ!uz=$C!5{%j0H(Uc2nYZm
zP|yH@LPIz>WC|!S58z}c48a7v1rTjZ{u~s<ga#;J%78@-ejN}T0+Y~S0Fb}}MoI*Z
zlmrL_ARtmd*AW6zfGT}E3giK`00Pi};T}lN6vSKK2@2@yC?^@c=>T@(cmNVgO4+yS
z90K#OK_Og0`vAHO6Q~Daob!k<z!=0h1qdm-{7Md!ZG{RqN<l$<e0)F#?DBvh3!-u1
z@O!|b?LaXExd_GVLy&jL3<B_1VDHm10U5vs2T;#nvDn1fihKqnAOU~|3gQr3$jJMk
z?m+~A8H0c$f4BgA-66hO8)xjn`TFuY0Run;{`t$uKg@|JzSJ4k<`MT#;KDuy_vrx4
zDPVx+{;7Kc?}cvw25pP-;tW&Bfsx?e!37L|fk*b$j|~F;6cPf&5Wa=ymJuUe#66(g
zg$(#9p?qb6IZswPkeh~NVS~gyko^23JflE54(W8}?cL7sg%JDf0s6DeD6}O%$|K3?
z-2|07l!v2>C?Cf|gAx6(p8yF0iUbPew-qx08~_Hqadklb$k-p8!o8b<e1|Z?1OaU!
z-T;`xKuRDiVFtYu!`*>+3j<ITqmxkV@x%UZAqEQoke3rf?EyFj4n+Ru-~tP0``<E7
ziwpSzstLrr2mtKO?)Iv&VR&*9?lJWGcm7_~<=6MrNB*l}w;%a!prj=90R;Ra5&-B;
zpg;ilpk#8KB09^ngz1~(H~kW?CfmXUO!(5JJ^AxfyY?9m(Dn-h4!@Vz{)B+BECjIi
zi`LGnfN~b`9`G}(^o#oNtNb=g^P4sH+eyGmk8zcr`I-Lx%NV?!&(O;chcWL}zyO>d
z-4_G+)vG&A*c+~fegN|9^l4v31{H$|BH9LhrN0XCtP=bolns+lA-=k6^&Xq`lg)&I
z0>>!eA>_|T4d5qZ{WBScNjn;zQ8tK7Jg!d$VdnJXSxmKsxc}rl6e2PVU_gR&B_3nR
zWJV4J0bpo04|(t@Jp>>K$N>}z321Nz3Yd*3agc|b3IYg%n_w^+jxcq38uMZK6$||f
z{kz@Y3u}}C5D|`Es_0MrZ_C(WjEDHfj}#H`lm82Y-hbp8?&<jh^wU2y4;^_g_Pu}=
zcC!1aPY(`q0PRSUizUt>70zD{W6$GR<uMxW?<(LHV!ZrC3h&>C(lCB`Mb__}`w%k<
zJw`ly+HKovmaf|FPa)H&17@?LiYT&7j?I=0uD3>t1zW?0AP=|T;!C%t;<Gv~GBjpi
z_!rIN9MJ;p1ECD=GZ;-gE<AlmHkfAuIS<Zt*FEPfvuEsHb+Flq(KZSCiNU|iGTZg+
zQZY^^e2S5a%?E`ad7n3mSN~xiUz`|aJ+Dr)3CS}uxaA-e8u*4DikB1jG0)qW=xAMN
zR?a^!4)PRSbQi%qU_mSj?OH@Yao=XO*gi*wP?`;MH(K*Y8wgwI{8*t#!$a0P)O}Zl
z*@E<xlOX$24*gilNz@TG6*oyXucND<E)7YQ@C#v}Y5uC3W!1xE<>G4(^swpZuG6}Y
zBD3VuOv2W>&ogyY=XR9t=Wq=DnY^17KV3U`N`KOQg1O@@;HIKOYj|o#qJ+7-Pq^sA
zz9~bPrQ%<}!<#?7tq)jy41{M#oy!=278M}ymGo|-{u9Gp#2M%{4<uFX_MDYIZYMY3
zU12RmH$3uOm)pmqzExe|ma5t+4{<yvt9Ii<{l*6<y}*UE*oy^*wyjk|ia1~CYP2x9
ziG-_tcUq64uwti&OK}C;_<QZPzrv)aFHW<<a!`5RM|dT*M&suv_F@jN<ZlGrIB{8Q
zNl_9KsrE%3MB`$&Ql3T04!OR}(4h0L6&uvJ-Tjihs(MWK{e_2Tb<WuSO-p1FdzH2<
zP<wmi)w+}aWEC_%UsEX*xcYvIUD=CzW_cfFmSpOH+ea%Q6pR0NIV57G%>w`a2+?#K
z>uU(Hu`SqXt*)|t?D`LC@B*%Ja&lMaPqfB0^5FMmbzWH7I<P^>N}VW|72ralV+F@w
z5xJGf<AkO`P|cgR4C>R24NqnK;Y!PqsHb_lu>m!C{MrTGATmRO)UCcWu>AZCYGb{!
zg!Bt99?r}DB>&O@OlhGSo6IuTW1eOB%m^gSOEhVl7f{r9i$!~pU~6=v*~W0+WoOv^
zZBq9IqcipVW1Pd;1`YDC2-lFm7Sz=Z3wHkcbTB7|vCeUSJ!H<>9J$jFtToMIqASao
z2u*h*76dd{(^<3iD-YOKuZODyhqJVfj)6RsaZ@L>fHkxzSkTP912}$^>!A)h3(SJA
zgK4vj53!8hli-a(qxpAKvK;OWm63hZ_8RSRrtviVaHAMfFg5T!1cjJPe$TpP$e1BC
za3nJRapdooZCfi(T$qe|R$cA+$bAZ)tRW0tpEq02AG)7j9>NbOvsg&nd^tVTX0zZ)
zePB&Q1dL_E8_H+Nu{0KsOcsnPT$=?YZ0YCJ_a9l5O=);6uw+xgDw~`vNY2LM<2=7E
z5{a3^pMys2B3ymH|3vgeuve*eY}cFl?rd(mAEPhnTF2GH^nsQp0qfQ%Z`(-kW%YpE
z0@Q<cm>;f?Uk6~1*~IU#uo-%-8)utmk`1P905xv7?0R`>y<05vTY(1@rQT9kn6n}>
zGrvze?8eNWMA@J(fX8KYQy?$07?_9W`15)zPv*&VS}Y5-azdTo#`2z<5frc9WbL6G
zIC!2Y+~B)bFN^0*oQ8T7e~Og{&-m|9Q?im<sa)x44R))W<k*^!3a=*NgGHeoaX&QY
z&2v1=`F{9(4SUU!VyjM2>UVt|Igh?NbM%at$}2x;5qrla=QcapdiD1;{mj3ahYC3p
z>Zid%!mHty?Kd2AM@)-T6EOF;;gM>FM%|ikR`(UFP3naFQhvUi?>V|dh>M?x7wB2#
z(Ld^k*dUy}fC^-8mvoh0k~`JK>X5eLvifLMSk1`}jH8eb9N_WSACA&4J_3-jHWWH=
z;kK3E2f~{({+^6&KG3-SCKHX@yEO3XNK$Y4rpNw4uk|ulJp`owKncjA7v$jDi8Hoa
z@|e3&r_JWI>CoQY;zF^S8YPlGhmzVh!{rE2roX?*Sue9iaycTSCxPbrwU`ZIb;Kx#
zI+kM0qI|%hK8L6s&$o0Fp`4%4jCzM2&15)l2#}{G*>b-mr@h;=dXuyXsn?*dA0P#a
zAw!gG)#jG5Xji|QdbQ)A&;zvwCvuUzq8Js|%h=UxGm^0(deolZZbO5n9IyCm%LP`{
zpaFsUB$(A6^sm?BlV)}Y`jc^2R@aSWbvXw^<+jV8(>(O><)mZh*Ew5@fe#TB##d`#
zrIb`1tUrfES&Bj32u6$j&@+nLdW#vj<aBZyHpjo)T0~G;T6t|LA8q&FYKmzLWzAxt
z9~bJLx8GOP+?>_G5Y0EM<@2u7P>wS*vi}+W_HVYb(U^D-$!nEm7=y`Q>jzES2K4El
z#XAmfJq-Zhea4|XkFScHAm*A3hTxGPz%}o&yq%`JjT>7xe$~})B!oxaCn<LxQO_Vc
z`_^D@iDH*NoQ`;krX__8v%x7DFQ(Rdey<}kOwr7q0KXNYlsZMBLX&Jxn<`$W0aEQ1
zEjPtLS=)OV?du_u)%`D1qejg$WDNGHl4@4x4Z}7Tzqn>MVBR#Qm+_CuV@i!71d-2{
zTf}@3r}4q+zb~%(gaqS@FZ*t}WsAu!Z?eBbYC=H^U(Y}lj$%19{$5$+t*>(uNf`q?
zE*Y)4QqI-29c?6?rEpJC%$kl1$1~_0GE?x}QqdHGvb`SyVp$fhi}}3*zmbJR+v2xl
zM|<yv8hEE(&ymL$BB-lN0BP?}dsZ0c(s&Fi*bbudIb;x|L2|I(h@2*jO_a#eF79vC
zRw!|%(8kE0kT@F0Z!5IZ1p-leC1)~9?!&V{YDDiN$wF{^vmUsB<p)8yyb!=S)J0PI
z$2gMY+Emn!MX5-ZdXDawK=}GzP?MytMw^_9%MCy4u_jhfBQ~Au9VpKihxFe1!flzT
zl7_4)!rTO(nvnE?!t^^i$CNFN!kp*4JqEu!m4t9cBEs)NarAQVcO|EXn}@tlC?^E0
zyR+Oy*z{-sbn14I_sczIe*BeX_VA_+>s!zMBCiMk;@^VQnq<l*Mi=uY=7|7Gk`MH5
zRvtAaP8CdN)xiG({cYYlB!R4N3iMViYM-=jGZ8L+ldbb68s!_9?0n-U**k_))HdhC
z^r3R<_8T*;^~qB9x-?tXF^T71sh>XW$xPWuv4`sqQc9&-yn!;B%X|Y@*14BP%J$&k
z&4v#FT7(1n6~>llZW;{SAC}0Q34a|9>jRHMj#j=s<i{y&=su5?%o=Eqn%*l?b{Yq2
z@MaV7w`Z+1Q0{UDC75k8r%s7DLPdNiq6~jXtgniMPC8Rk`J54TwLg;JR^O{^KI$4M
z`CqbHca@9&W+MBT-iF69R})0t_?p?ibzR48@IKKjg_!aC4XVLA9GRroS>vlASWX!}
z$kpj-^Auz!tWvo8OhRd9_{|%?7bQDwE&wAfX)umUXkmEsSYnzwyySQYz;3Fr+F#yL
z+1ZP5SDVJ@5XV(DXu*B<6bt~gpDYnY5&Iu=kroYCMu>L>H@*{%x&q`%Y6<=>Ct16I
zZN2Oi`{qpoVe)h&P(ic?^N-FijF%7+9?2;Z)auN>vt)?rXw{6O<o{Dv7#ht#8MY03
zV1KR=r(3$^JbPgGVcPmw&^v3H_lW)scw4HQ8@Bk-%9FfvpOJbqkS@sHdyPJFs%o$9
z1AhGg&2%q;tsfl5=kmWGoU6b=0Y>^Al7v)GOwiN=HpPVio%Fx%+W&){16uBrM8n-6
z(&d8D=d8VlSI19)i!E}xnr!j!DJ>-pQb+<{;+3dN+GImMxfOSvy5gO^yiz~F8NVB;
zR~+fTFj8?pbS8AYS`ZWd#M+0)@{<*A(EjazuSR+@2DKZaWL9*c<5R8%844jksIn4C
zgYp|;i+>?*urE%_9YBye)ZZmRt}iz6Y6OyJw=!DF<t59}=GdfE(*@8L<#*#_yhDYY
zj~FX}`w}zoQzYdrn&FXiolHuS)j01wTiPGq7)OV-Zb?L<{z6um5V&Y5D&L7D+0d1f
z(5VtUfB3thj9^ZF(hdFkYNT%^+>=GYS=B%q`D5}A53!*@iAL-uiOongbH>cx%Y<ja
z>Ko&Jy9#bAej-}Wg)1i2A|obcuof=6>+Z&hoR+DSTG&xH>4?7KW>)TI3_+2{l9Hsp
z3rWX??xTQEzs8w5iN#|7Yh$h)x|q#a4BpLTXQ4P6S66@MmD2@ZqN%VUfwIL+#$Y<$
zR>uHYMkiRh)$rK^ng?r3Ok%y#t06{tv44M=(64NyKX$l2y|}g^I@P8?s>(l6crKL?
z*J|#^-KJYygTmTN4tDOb%QdN#t07JEY6JYzKgY|MR6s)S5mH9_;Co`}x*urJ#S+NK
zi(X;74PSwZQmZx$9BQx9LwY-)(jIt$UjlwnVGhG<$3uDJ_c9k!Wk-LR=8`t09M63>
zj#{UHA|sBOVlxkjaBc1(c$LU8zt5<HLDh@IeFm?@c4BKd#SYNSZxr^l!+>@{jz~8z
zHJ|f?@q;tc&95{qH>LxB&Z>jz;=GEyj5f;F8!`m)&@?ME6Ku?)B!zyv8k-XBF`NbF
zdv>K^|Em`wZwiHB^qpe#?sJK~LU~EiLbhLiY&9>z*3C8=W?W$9d5!ky*-^cql2<yN
zgJA44$GCU1FY#@=n&qn&mdI6WA^o#F8XuFW?q{GFP#aj{b-$L+(nw;*Qv_>wk5EI3
zCTkNCWFfJzHGseld@m_hh$Cs{tsz?%H4<Fjr9GLuBylm?(?~gF4P@)I7<_nq+@6Qu
zLy;BcD&bt3vYV=zC+F=A;5l)~(^PnpXQ#iYOPtqQeI8L_P~D$N<Q%)2(G(1QA7(76
z5N1}%Yu2R4RwV%Kju#}2s`>qem0t29?orWzqxtCjdZ&cP?X%L2^?Mne@0@TaB=vDp
z_78w+b1_WWD!tmC?aCVtA*M+)B}*Sn7guR*Max}zXe%isM$Uov<lbn-ayxQit%LmX
z!f#m5?Tb&{ji}1$fRNo*ta_Q_-N>|L_Cm+_i30e?Cro->l~095m24~HCaJ>kb8&OU
zH}puUL#B}BE~^s3uTo{$U^HW_j^SS|YbxEVAP^r?)yeZV=4Yf}n`vJ#ruV+U8mxUO
z)2A1mtj3J#oe}7z)w1YKJYsN5yf)zM!=37+{>=njTQIwgq!76-`!oK*n)%9AF@0xV
zc4C^8$NO{xF}%_VT%I=~H#nzP(_>W22d`0gb|dB_Pp9)jgX>~S9ZDx=>R=;S>H^1@
z<x2cn#y(wR@L9N@sS>kw&jZF#S)%u0bNyO>9UbboY40FE4K<POVGpZ#(0R+N2{an>
zdj4+%?e+nAi3fii0&jB3#_K?Xiz!-G^h2zjik2if%NOy${;15{F(hp0dRf6t-XI<R
z3KpbI>2KT93Dmh8a{j-<7Ck#<%d@PYciq2pk>ra<iCRAx>VE1Q=P%70FTU*^>E$_I
zT>X;IC}+!@oKg&SDMwa}J@%II9*ReGPzsbh8#g+lbYr-A?$empMCTJTE^14Wwncxd
zysh{xxpm5FeMxAwg7o(myVnCc3nQP_RB4dh)O@-OJh3mH$%<|5O1|?gE!?d$29P{2
zsgoz-i7P_hEXmD)iiEN`1saGh*v`rLKPd50XO(y{bmz2hJDxNbte%=zSqtvDa*ExM
zeBZ9&kgtvYf__HDOWQwqe=}5-VE5F!^K=Y8d^@nVE7xM<Zy_yv#;({nR>MKLnA#c2
zKdEaLvV3p7ES2>$<^CIxrhO{{x1Zx?L31$k;lY>x?omoi=GfVI*CB^u!vwb3w`i1)
znisKGE_B$G4c}Zrn7za1YzDtNQR+P>3~Qvq4{bHz+ubNnPuM>iS`2RjG4B#iKkEnM
zodpb9QU>Aq_;l`6IppFuVl81_N%yo^|KP<FxT4}6zaNnz_}!ZIO*>_LeH;A_uASmj
z8Xa4$lD24Wy}O2Sct6ZKx;0-P1H7n;;@G}WdU8ACd-2XcJn!UTb=cI<|8%5_@cLQ4
z-AgCt20Me?${0WTc%TKvQutjoQz{ny&J|3ur~BPaB$w%_T5$?jFkPphnfp06pRotC
z&)l%?oH=rKpe42<G+265DHSKK<fZdDCYWt<ofQBJs+e`LZ1eeuc<E%Muu#yo4lk-Y
zWK;tB|8XtqDOXX<zP1jKDo%{aN4QXGZT$SKIK8M!cnnYagSMNT{TR(0=8#s)ub=y<
zB-h=bu<e!OR3*N*iN2E7@7Ch2#XNGru3bu-ISGl$e&>cimB8o-Cr@c-k@%2SE$Uxx
zO;DUsn(&=inBF--ccN=KWnkp_S=Y(zV#-BdD+;D}TRmB3?0dqs!DYt}t5)R7q`ZoG
zdZ=4=U+h=pr}}R$*U-LFT#noH4!Oo!mHat*dg6P;{&h$-N&%|4V+<jp<@FYJR}#Dd
zjE`xr`i`<QZrXYl(Yt*vqE8sMWk#*3??ZonMmrJ|u1ig1<ZaR=`?p`hCYf<bXV*I+
z=VgXS8i_W<wp`-FYUfU{NK$rIzJ8nX%GrNdb7R2_j|2=C;Yuwhy~{!r816V^%nVyC
zUu!lzJ>HY9l@Z>@1Gn7mY?jvuf!B=rES`hMv9q6bSSoBnf&@9{JGs!mHDMr}J`vtL
z^R?Ys*=L`)r@1z3Roe{dW0H3iL7Z-_UU3FrF1EivcSLhk6pA*krhbCaW`$=F`|O2p
zXd26su%(u%yk4li)J;af&~BOmq&C^PV`-!N6Mq=a7Rj+%QL)rUHU%Nm!`$9)Y#qZs
zG>ZfH3fXZ)9NAN+{ttps9`oJ(l`)h=vI7CHbiK>w)kMC<VI%wHht%xrYWgm^#22V`
zov7PkZ|#E-XQ^gDlP;>{>aNn(0n~o}M$0IRWSzEj_q<Nr4qc(g-1&&z?POd>a%%|`
zY#~-zvkpX70qGtpTlU@g;@z#b#FXv$Q3^%mhe3nld!*=#MJxF8HqgXyI2B%IGaJfu
zg4{e>zErpv+!^Y6+_*Gf3xc;z5{em{3&Hfm6-r`ci{pA9V%oAGFxX#?ZGG0L+V>6Y
zKid~ED)^4eWvn`q!WgPVICI9>I<jgTHP8up=siLrg!?~?oimhZO@OW2wr!iYZQHhO
z+qP}nwr$_GZTsD1-kRVKCaiUWiYho8-)?-0I8?pOwj@iUZ)U1C+g)HtW5Szfx0P86
z(auf;KkyS|<pY3qDWR?<t}`s3(hQxf{YUiZ-h6A;0BVSUt0dy(W6SdUU<rCV7U?u_
zPg%L@j<%=@K_R|5Jje79+aZn}fhx4=3DqDqQkWP|s$X?mzdxgCeYuZ~cB8{4)5AVh
zwZ@H{;exz3Yp1?#06y@(+s(GVlOG!<r59UQJZ~V9+NAx4Hi07Mut=aUM$KVKxm|}A
z4Ff}b<O9%;de`y`#EpN|88tSKV%WNbUpXnuOJ}BM(Pa(4rdJZKatd*ocb?8&=*7ml
zQQAh$#93xQuq&Z8+iBHiygjZ{zNBuk30Jg4dDL*eoaQ2%Q#rOM<<QwxLx-)~?I)+k
zj?Hp2+~XoX_v-alDRz*@ze;vJ&@=zjjoTZToJtq5>KjKtijIYj+EAkkSue46fT-v;
zG3dfM%hD`=-^&t-YQ0ujWa)Grv533@tPAqoPWI4%q0RT~62=$VCR#+3xT&~gyO1Kk
z03jv$4D+u{GF-O(82fd`TQ-z1EmqQO`91it?q7zCw=h@qKJ23>M<p6WI6FA^>dlj@
z14iFy7`A+$x(f83g8Rzp%u$yHFY_Mo$dO`>CHp6a<zI#!V=83aLO0Jr3S{^~%t+GK
zTzTf`6>0=a>-*~wbuUQ)75kYr=?>%q9p1E5u#Nk!4R_d3rw+^_B_}sdVp)b+fFlq^
z=PICKC=maqi`59dFwYBona1XDPw!s$<Dtt*t4TO<kd94@YC_xHXe&|vGcvlksnxNd
zL0D*t`A84-QKcp39Q3g8n;yB=Aw#d$lFJq8a_$i|f`u;(Bo<`2lFggD1Kawjr%gS&
zgs0#BoDCgd9?XWJBG=?FMlaKcY3P~Kt=v$XYpgU^6P=pYLI{2fGcCrv{57b)+N_Pg
z)fCx+zBAUTzT37+cJdr)=jb0@P!_m^Aq)IHp4*YNj+oZt{-R`iRHmi3qySe!NUV=;
zbd*Vjrd#tuOntX3d+NZUPyRy-Jt(QzHqfM2EQCOpjt<2xCZ<s!sj?|UVyn>-rlL{q
zVW>EbyWQPdz;RpGaQLF8|FYmQ=u!=Hty|L0%42EGvIz%@XPH(D;x(V8oD^5jnpg(-
zwvjQl^bKn*AT~j(DCBu4&&Z}Jvv`lU3LMkurVt<O*qZ#8X^<*g5ArW^NBK8`Tl~kd
zGaioEr6ZsbRA#-*F`Z~c*uG-msptedeAW*0z48_tU4-ng$+*WVjerL%yrMmrWSSxe
z#Pm_sh4lhMag><j-U`BUNfI=55#}2lit6y?sA~<2L^XeIa;BZV8z*h4Nhe)k&0>z(
z;l<A3TtLNSiTcs^h1Y7xhWejwkPfn%v7EbCp$SbD(Th?O>@#j@`tjb9Z4dWwE_uI7
zfqM+Uh@9zL*7w>1cAX+G2Aa2t9vU_^Q6p@TloZU#lq98vfMc@2sM<v6Np**bi(f!7
zZe8a8v{hOE%T{G)`u}A~CIa?93~c{pt8%b$F#n%z)pk(jB_C|ISn-JB<YHipVo;Fc
zt}YUA7>sDx)&+u`#cfhtpha3*LRcUq#YI8_g8c5K&C}kz&-dRskKHDxRv$h!y|vSi
zAJ<Xgx{4u66WAt@iJ^RPSzS6jME~f_SVVew|FE#UsIV~jv<M0G=wSb!phKqfpgjl%
z7^(+7EGBydgor`o0P*BPEEp(x*8l+bcfjsIqHaMVZf*cxJiPs1iorc10F8X=b*%hR
zQ29^+0SIj}%=9F19BZpE;gR<_HGnEqY=1R%^|J{(C;wz(2)I@tK|o9XV%%{=)CDLA
zz_b49euCn$KJ9wmHX))M;r5R0t*v(dSO_|dE!FI6$bQ)9W?=Im1cG+zajd&4V}H(h
z#HTVAI&CulT3n3#G=6Z?AZOq~0RR*LK&^m$h~x<9Wh@Zz{6b*oWMx1sS%NdY$x1&;
zZon@KIRCV$S35VqYCjGjAwO*3nri)6S^jcx3`^*GKp}nrSPvDtVy*{B0Em{K5{O^|
zeG0F9Y6KWS%|aC4GB}WNDGNaVa;V?su=>EDoOlR25R8W^(HlKmd7IQA&8dDITmg9!
zpjS2DMt$~is1-N(8@?eI|G}+%-CvFk|AN&&eUtUB4hO42fjK(-l8t`O@(4G4>9i1#
z2z!WVC@6cN{@H*8Zq!^idVtP6+5CMWeFn;~yt=Xs<o>XgF!Au~V8I{#&w>Ct0stgi
zpht(#yMcar_8uO9tNPR={Yb{pp$9*QxS&DSKb%D;@Zep*(H`<ohoE;~p5NayC?gD3
z7$^t#aUZvcPEu79RTvfaUzG251<fqexO*cL<S=_EXh(o|4-Zg4!NDK^yuU9+kRZPk
zhke#5`F|Mv@AS?KKc4cEb$;#rpXUPW0e`Nga6=90Ao@Sx7i8Q)+^gm7x_{@Ke{CLq
z#~$=_f9iyPccYJV2;z4gSNl+Y_#E?zF%Is<fJd~^Ni3n4^~-hqf9fg$&+1lGfHi`2
zaDLS)gFy;g1)waW3|Yu;mFRNI{Py*Oxz}-xYam56H@-(Q`|Nx4KEVKm1Tq;w-qv7)
z8V~m0@i8qYs&VAoAb><>c8L%y!+vnp;4DF#ep;-(fdT;taNwT$MP{&EkO1883tJbl
zj()PH0c`0gU?P|R^DUnM*ZcR{`@p8i0rtQ+IR*@F2+Yuz5#$GV+4=aO!iEfN{bKIz
zFhYprX@Q7w<c*p>u}_Z?LF~1F_J2X|3IN#Wp+EpP2!5eYet~~C=_|QylTYe;ez1{!
z!S-4M`t|SpJP^=+f_^pu@#+0HU)E4z0)M*@1eEa*2p-Vs6UuwOweV390xhrlQbO_y
z<<qa}Ie$89et?UdaR=nlhxt}1`auIhxC%<|+$_2&j3_Qdfq#@%=s}D>KLG>%0T6#z
z=mFZ0D?u5omcQSrkNpgJL{zRpz>D7A;*6Bp+2=oA6Tfl*Cs~1_4{q{uf1^-T>Fn7j
zZ*Q+f7*F_4ch?vpj^G*xbMitoBEUBKU@g0<Qwv=Wx4!&a8vi2VSuLZJ`5XVFUz(If
z=F~Er)7Tj3yxX&jIMuQk8?I($e&0DRZx>T9`Eh74f=7_8wbA)ZXRy6}H&3Ano52<}
z;v!2z>XGwCE)|kCGmJ8rN-(7NK%!lo?0P<y?1_OJUWVM^^&`qV%L?a7$3!Pi-mDAx
znF!CdN8&Is5k~p~u*teCs4QZC2K+$b_z-TRNxW|xZL@RQm|L9tnEL5K{Yvyo%BAR=
zKtPgKlrj0d*na_#YG{`_5(C9(Xzpp3vQVZ5G|n>yd%V|%%s)IU`}$dJ0rW}r)O1Gw
z`oyFW!_u`d1ZXr|-5euhJ;uOn-)_CZXB`Y(=PH*mNHVKv6y?B|;HlgTIeIdTp9843
z;O5z`l<JR=qKj6ZLBl4j(*wWH^hWb_H-}5c9Ue2wAO&$_zkO+H$|DP&vRE6{+dpzc
z@(Yd#7gji{<?Q&bn7rL1*V@&6_8*7FA?2HG%G_>-b_&HAM29sSC;c>pUY~pQbUDOG
z0-Y`me_S~Rj$@_7)yq1yreCE~>Ro4B!_2J0fi9pY4museFn)Xmdvp%tyQs9yE3VvP
zMu14Y%OE1gv*V;L_G*+P;7p)Gn<Ghio{T*{EWO?1YSesV6lOkuYHsf`&KT0}k8OFg
zn^JM$(qGX;u4p7D$Qcd5dNr_ZE{Z1oj0rYPaV+vZd(TKn>Si9T3+ssaZE{xeO)@4|
z!g>Kh%3WnY>9G0n2yXrq@@0}eZ@>A<aY-$vb{T%JaEvnkvv32)0fgiUi@OO$^F>T6
z@@TZ@+VRW(x~>KT-{Lh`uz$vmS$PwFI1lYkSa<1ukDqr=YiOd~`Nx~ksi*t(D9*%b
ztV09SOYf?o{cbjazxCgL0Q725pl9aD0N;QK3z`IQ`^}}_t0;juT&Dys-Fv{#AW2BU
z!nQs5m#vI!4+6ZKt<`3JMdfwCvV!b-i4SmE!Rg)DWMHu?O@WoO8iio;=uq;)S`f{$
zMP^`5uq`KZseTW%+0Oo>**A8el%^krOiAtYN=R_FPc3eo2%c|44fS#NG0FQ0SLM*9
zxhp!yui`L=_`1$yx9386koZa29#lz2*V>GIhRX78K3FLnr^Aa;faskpqR1=1Rb-)(
zhmWq9f}xEYFuyP;+1&0a)Xaej2aZPia76FdFHWm<9Yk4V7>$E1u(Y#_mq;7gGiB<Y
z`yOppw5@5u_gNtSvWUrIhL^g;dduM2Cm8!^hZ^Hkl4W$(rDdHT_xxOErNYPiDfNOl
zvQ->DWpIgtvbqkCFzGJvR~C0U$~33?3ix?^k@reZAz0+|R{uFpvXXl7?G|>kIUrsb
zp2lj}7r7@F_`V!D#wCL}-o;K_q$F#GXV7UeNj!mKWNI<Kwx@(FC7H=HhL49gLu%#X
zCWvcn4zQM>)UnJgNLwlGm?`@QDWis^^5m@7)VTxi)EiqAf%&+-ggLjMTKVf<5uNJG
zFAsW%>&sASImNZN<8~d1$vkV-n_Fb1XtFq-v?Zkk10&&n=dF=Ms`>tWN3OTlo<aT{
z?E>TKIbw-q*5ZX!1u1*gBp_G|f=@ZjYw<wjy^0EfOai@p2s$16o;2vfW|g<m91Kv-
z-5e<hoW;5xcyUMV9$eZh&4dWD3$55s-5$C2;w|IoY1wp$%4sV~W7X#V;!V6uO=}~=
zY2og+dU~Pr>-{HiE>z@X*~p{julL@}GPZ;gb*C95t2`B<St|b|y_@o&Z7!#IxB|`>
zO<MK}@NgQ9U@nct;E^t^lD7kYo~#i$!k|O1RX2ZJi=V>^J7*h@36K>m!{_+=I8KL$
zr=jE*Od?9tLb$mp9Mr`@td175kyP2*_SG}+L0(~<D$Z=&T5aNI+a<$M#Hcm-c6)lk
zdGMhmWOgy6WE7&`q+5uu<dV)<g>+N}rj-#yZaW1j<MHT6|M}%1>YoKZiVWtI>HA1B
zTxt7Lu?kfCWk}99iyqsH$7nUoUQY{vzX`a=2VNVq4u1e@-VZB10X_OhO+$VLN6+~6
z$?I9RpRBbEPayF=1|mqA;1lC5#YiyRst`~9+UQ1?^e`(@<5U4NbX9rq*Bv`M)KRB?
z6k^i3FRYSks(%uBbT#=6xx~t@1I`7pXLjo{?hHvJOS;}2auN*WKKQdaX&<R`jW-)o
z4lODPhz15w_G33IpNaPABz*}mShmY+Ic42T88waDe6nondo^zzc3cZurAFq?$Xd>x
zfOWrmn{O@q0F>4(@D@DgEGIZcp}5Bm4XDuDmU}*`K<38uaAn0cV)54C2PMoFsR<jG
zQ=!SWmDA+r{4s7S@f~uVJ$ZYZm>egrxD~Q8kR8R<MsjLN7{^D;PHb7{hJR}Cm2b*6
ziCu}#OCxfT1cQdCAx*V}nP3oC2aW*>@}kFChP{Z{!5yFD*-d?^klAr4|6UI2DfH>@
zWMDQOWg8P74x)2;G}>h>N2PU4wg*4B#(>E3?9=;5R<vpgsb->>BaRy8A~YR%UI-$N
zO=|$<aPEn4d{$jpvOB#Ut$M!oIjZF@Bi$961>NmY04WKZwwheNZm-0*Wu$=k*h9TO
z#5X)O&RTsukx&@!jgmgP+o?3&Hebzq??ft)R|G_al5Z&ZL-4E}dYX<Ou0ze>-jK|L
z3&*GojMCuBdnHi`$x--pq+4y1{+g|P?SXi-jYH=ZR{KjyzW=I%*8yA>3Q>v9sU}mV
z2wYWw%5Kndt44X6S^Nhn>mRo?J~mTf1m9!Syz}QKT%pDLul|hvOQ$wMIZNqyX*uxB
zi~->j;dTiTeHMbO1WY=iW%ntCe?pk+O_`sLM_eyR`{O{6{oGQG*|*P5PRPM(+Y4Uj
zxxx?6^G7VOo}wycPL!CnNhrmk>|yN5PD}AF;>)o0_A+de#>?@INUwFor4;M{P3S@+
zeldjD%g~r<vhMM~Rn-%m@~p(x^`mC{4bQ-b{e;1U7SHDkwQigun6T=z_~nR>?{5^C
zj4Z`f^>uO1@q4?%Clp$^Q(oHVzlj;KalS+y**%9ym$Nk>b?3&k3h4~?!?~`9S8kj0
zv7vRnA<g+xhHK(dj+MG^cwQ9W^57iqsB<NUrAD1d$4TQ;R$@Ykz04ShDXi<AssW`H
zct4MqpO=$G#5@!roDnt?qU}>dipFpoT2{ZJ6^PRNlE3zzZae5QY{4-x;bQ1_drUCV
z9a7Bi`>>m9^epe49^mF8r>osm9Pe<)$Ha&)QoYCT?91|aa*<3SY|T}1*+bYyFa9N9
znq1Pk0!r{HyMN{|(d3Wa9*I-9-95Q<oeZRT;BJ{^-L?6#DL&P0ISF<qDkO7zx{tJG
z94jvowT&h{QK&$A^!T|w)<oIkdm=q|<&U(+Qs4NsIw3y8OjI<npWqdYc?R#N``x`1
z;56wQRdeUPtPi0T?^}6RsIIzxLaDaT@}98Oxli@7W5pNf>bHaNa^tq^au#f)NNyye
zzKc0K1@pVzoEweEpy}QXw4G@J-E&*khu36&Q=B#EpC-A0Y2O)F-2lJyd9omT!gb7K
z{PLo*B<vbC$DNmNGi(82?Pl<*ZG=1T)(kt%^I~Q|T(vdfw8$y@$XDb#hr6$*-a*1m
zz<`69dS~`V@JZEH%BB|(-W{*)r_$t=;Qzos9+4BbFso@1&39m!Z}GdkDktZFWcqgG
z$7qO^B?%qNtLGTI%xE_L*1X@>E*I1L-Dje0$d<9>{A@^#8X0tap*^U3F>a}qBKbdD
z)6la`Q#lC&{zM7B3vE`+{w>p3wAjSP{8S`8hF9Ip*^R}8N9NV^Xyd+b#^Va4YbJ30
zihqqMof$g!LM@Xsb@^}vnOogtuyEW59T|0y(y2wuyZ?J(Ra-<c%jvUCkAh+6R2~H7
zHrDR_kSTT>NAf3Co+j31+)rGpxw2Lw4jREreZxdZ&fygC)7}dc9qlz0Th*@pr3p=S
z;{9FiJoimi<~ot*Vp=kmazm<a1onKFi!7iQP9j!t91_MBlkC|9rH#HRSB6V|df0&X
zwS8LIZ2+9i@D?q#`PBjS8NYyd<r2zkh``$}A`_=LSz2g=Vwn!o2iy6fZwrfbPUx}%
zsIb0%(a<zqYJH~A0{lSTwwh!CVhwMHfep@+PY?5&TD;>dKP#WJ+}FN~OK&v6TtxnS
z?T-7l1U%}H_rN1_C$_I1xyJ+Cvq5!punjhq=ou!Yvq)uH*#YROHxd5!O6C%eU+x`t
zB;}Et^uz@IYj~MH=(br)@ZO@RJDZXwMtubHB(K4)?*ij%PyS%Q4^k_tU04g`Ehm(^
z*6na?yt9T?FB9!;zeQ;*@mk~5b7CY37WJju@iDJVTZZ_=kYq`3yRk^nG}nuhfs>sf
zlXts{Xo$VF(PFZzlE^Y;fI&Q!8<V%^WR6qL!4qVma#CjgBM~juphHt|^xE08?1v>y
zlx?6x$oBv<^nDJK^59MO3@>7p<;AAANU|BdsqE}AW7e9+Z!F2hsD6GN!(&_b<g$+D
z>WD8J_S^ZEUoL*G!i)%4w%vh5@*RhAqcLpEdN#8PE>U&ufu0Aisa8#cfz~}m0U33v
z&0@jC@sWX8bt`Mmy0v(UgUWQTF7`+@HaY<rx)qCxMM1L;vkqf7GC|P??k%0N!HRam
zZba00QcpBRx}HnuANppv_M}LBNhqaNDV>(e=-ZPO)d}HdSyp34%JP?3EH^!}oOzlG
zLm*0;Q(nJAJ&94m*1>S7+?Z?ZBg5i{B8NS%#x%6$ovNt*r#Dj*+SbPgq1UO8+DT|-
zK4f~5tdCPO=mJ`}CabnIN)3mju*RhNLy`Ft72HhK#3Fa6DjDnD&ND3&riQABkPzqC
zCP@^RfDj2X*KxJj38$Ww)uq!ir@RKZ)qV|Fm6&vPT|9gw%Py=CjoDaBuXxk}MdV`V
zs1}QIQS*Upxvqr@&YiUb6wW*gQ>Iu6iF;tiB$+-#Ry8??Z=o#Fo^#+OUEG3fMcXtH
z-3s;$?<@^&gfowCPDwsvskzFhLf337xPZmD7yE=f$f;WDri0tU$?kIJvEpcW9Q&6G
zO-@11gkwj;B()X4N-7$S_wQ>x1aLjNw~XXh_}cP2j`c%883Vcr;SY^o;dUqT+RFAU
zvtEiEm~Mup($p6czEPp%imY^B6o!0cB2pJ`BrKWa2#YrI2uz|ciXVOVdOS&}l#_~t
zGEtar8v#teOCO>@uivZoob#cKe)oE?!SLKC-vQ{wj>y}<_j3FJo}mii)#XnrlN*<A
z=6)OTXL-pUs|P}A=u~fEan=$SnHjnhM-WTp&zV<%M<(R7pg0NPM@Lj(S$X93Qpi#S
z1?Pe&`k&Ao&4ntFoZ=(fwU_<FEITH8D^kuBV6BeBW6Xc@eWk~6TmAeTJDN&!X`nj=
zW@zao`S8nEACYS+JkYkmWN8)t)S1|pEeLCBySJ{%C*f<T4`_JY<MUdmlobb6I5ajj
z)$aB@l*sZ0A<*LPbn1(jwGG@enf~R8@Fc_*MZeA3Q@v*dh$x1Upb}9=PnFRH7qPuA
z87*b9%{0%H?8eAk)u5u&wzfe_89y3$Z5_Z}U+l3^AD(e~Z03taK^z7*%xZd65tI|j
zUZgj!j)%A0W_yq!H)`q6*3Uh9b{-9&R*y}03duivRM-a(5f418{M}SzrZI9lz|fLp
zHb`hyLbsTwL<zD9c7)=3lkSXen}{SKdV0A=9Avs4U=B0=L#3OxS=pQSv`>%EApv$m
z8iM?e!{p_2;Zr*Oa?IM{GL<x?su?w`PMbUoWGA0r0fSliZMt4V$s5Pg%T-RoS4y}L
zJA)lPs-&ol(Q-vje~$dnT>Gs<TS$CscVKU_fT6D$*Jr&xlJxGH&<4?ij8Nm%spp^#
ze}#`nDMVw_PtcoWj~ohhOv+1I=9$$4Xu>evScYtxkX+fRx;G#s$rSf}x@zu;WU+xG
ziYxiS_gAg(YBg@bIAm41Hn}c(UZ_4gan6s|F^D4}`?`3cY~tC{6mmkUA20U#uN4wh
zU5+x7vGWF*xR|j>?8}T5o@ic#WR}H<c{16Z->q?4@0=d?8IVoS)Ovg2vRu7amQig0
z=!XSk%qK<4{dS~?KYSTO2AZ4QeyJwyIK7r(t(CsLTVv-B9PUEz_NdNMIMLMMtQViF
zrrXqv+k38@wy*fjX+ObVu|?-J73?^Z)nEmx({GxUcp81hC?zQModGelmAMW>(9r30
zc~&Q_TDhm}^7E-A*n5brSVwjWn+HsYs7;d3<Uc^+qx4?bvu_Cnu*Yq*9xo<@B0M`V
zBQAQga7JbhPGDd$6Q#<;SB~>VF?jJSjWYuv!8qfpzGU2{4<oA4Yyw`2`@9!Eq2+?@
z_$XASd}(Tgy1NiLLtpm#YTkYCA$xB_`YdDQQ^>BI*<kc|!w6frD!${3>OPDls*Me2
zrM(#{(80+O!OXgr5Y^_kARW)?abZ*U!t+%$k704}%;`3=-fb>Fl+#K^#A|a@7sCWE
zPUE8&#A=~IKnMDwrtWioJO{a>uXSE%@S*mRIR{ETnW~b>Ae6?5|L)m3cCDIqBpIR8
z!&%HQ^)7OeVIwT^O!dQB6TZp4aJ_XBL$s;)UUCE@WB!o}&4=v(xrZm*fh|PC2=;NW
zU6W3WL#uR-fv+}gkd6c5KSP9pqV!aLQJ_9n|F?WSCa(61iHL4h!|s@dj(1XQ=Td`-
zG^wlQXzgMVoIWd1xy9E6(V0^Xjn$h%c#*79u+(3A1=TX4-;(0Y+(Nl1hkf&B=h>$N
zeS-SeCu}xaK61C^V5dJ*H1yIKRR(w$SB+bHu1aRR^tj3(_nlO6YU~u#Et;8|XY{}f
zgOR@*H4n!WM6tT1<b_uKGmx~$aD;&eW|%8(=+AK7L1im9N#E1<pUWEE{jVRX0x<>W
z{z>Zhw~nyp5_!tt+zp=6V+^H67UvHN$gm|&_(P?xsnJ?U%~$XzdYFn{bQ-!ap&jF=
zgS@vokBdSI&%e4^Gu;y2<A0Jy&sWM3m$Yjby+*`}bryBR&VGEo;7=9ORq<ytvYM~m
zw=_55bMR|6Fk$eg2ryH}PIC)qSSuv)n8_UR8M{K=hH~~u;Vj)_64ev82~ONPz9cH*
za|`T?XGhX>$uxg~TWxH~uLGuz-`z}L;NLHF?3IrB?}5^AdLCE8575bJl-Y|}XGV4v
z`Rp`8|FMfXxoTGfxW^lAxxj}N6+H~Q6blh)KWlo74~P`*Sro`;q9FcuYV;ALZI1Xh
z!Y!U=k)SCR`6tvTIDSxT!JhfiA!HOUW}Rl4BjF#xe!lSr1k@`{<&(0>AC9#}w3l6B
z<*w0ctICwUs74%%PP0)*IXM}TFn)F!dI`cK#o(2$<mmPC?f4Lp0^llYCd7#!%qF*E
zMu=;Tuz?b_Uo3a{O-?r#A`?)e`B8vTL)^80`kT{+I4C+(Wb~P-?p?h-S`TvZrct9W
zQGSdVeB>2WsbxA#rt_F^OmDT);9BE)J;jO$n=B5_Ojt^%k#L0i-Y31&?sOPZxjdT)
zJ1r{;i*KFFY1SR_K7i)sMdrlP$JY<fHB5C03LWE8J8)=j0LS1yrt061Orv9gjyQJ+
zAVRLvu0Am5sH(O{!yH$<_QXkCoh7$uI}~e_OE=!E`V#ybS-RSBpX$>onGU5z)t9<Y
z6>9q=>Au=FxF^-r8ka}G=@e|Q?aa$hSS`WJqR^l}?I?jTqiHdHw(U5pG#h(>(Gm_k
z=DImBm<u*(W=Na+p^1CSMDoAHQfpe6)ZY8YAI?1$`VOcQzxZN^ePdIyq@AL~y?r+~
zD5VSS3y)-`6FIEF(^0;G@*fu6AG8Fl*#$*84!z|W+h^GIocwKe<06ea<_Xm_@Dx+L
zsJFFOZw67k)4ZN1fjR}oJT6aeZ-2H@*>b>W*UN(@bIc)~^&($17b|PFX9w$+Hkei!
zruuf0puj@oad&^Qq`h&Rp-n`=Qv9=RnBd)eJVzk=2Be@-h<0Q;oR!jI$cLWZTU;@^
zLSVsyRmGs$gO-eCmh)LcTI97(82?&?f$w9}&GdKZtP%B#xx^};DEG?1pcdR!%1_VF
zo@$~18_x)im3g`0El^0pNqy{gCV7miWWll)zN4e4dFp@$%+4e^%7l42;DjR5j=}Ht
z&k&NCn{A=vLvS#s=8lo;$`_;EZRDDFi1dJ{6Qo>9Lmcna&k!aY{x~%jl=sZk;k@8D
zb+u?WS1nt8%4+tqePvjXKu?c5<8dw;+`r@(yxMKCyF~WhmO~ERK=CgI1uz-v0Pc9I
zV#mNB%jHEyMkkgnTAsISA#OynP<UJ<%(jx;`3`MDLbpYQpGk+*>qC!(Oq_zR9Uvx~
zsnX15i~A)y;cnf>%_0U}7JY5(QJ>_p<R;R^H;czT+V3C2=x*vVvC`)zPEdlm7}q1M
z0bMuf0;v7YIpFb;?_%@nI2>vlH;tj<59v`*o(SS538s|I#J@+ptk5<~lg}5&`@x>;
zP}%V});?+}Y?Nr^XAT$^nF-GGh)`!)=_fa;rS9!X?1L}+J-0g3Zn5F*KSqUwur=!$
zaLpJ&b)I@Cz37Y|7pmHQLuk;ZJ>D^){yb#ko;h8vKf<E*NC!0n3fB1!zLovjMo;RA
z;^=EVb?CoZJZee2$6YfLaeIIqDL*oM8hbG3YKdzBq=BGkxX|=0fKAZGYZGkVdRZ&T
zZDPXSy>FmeRP%g99H>9F7=MHlFfx?5*285#G`;A`1WxGwD)iEr5n0>EP<k_%4ZS`x
z@X^g68Gv;ua9YP!|Fpm-+(=VUFK8V;i%wI{zB;qK>Y=gg9_6w>(xIce9Wn2=8j)Pd
zHuDB^!_ud8U=s*W8EI#B$6RM{RSDHnv|Lv^;l4O8faN*`<h_~lrMh<LRz5v)-9FiD
zqiLU1y<mRMZg$)_Tk(lzkqf;_<-Bt6LkbYLr8O~WrYRd{&<i2JI6n0Ni|#DbT<g)y
z-AkGar(he#Ip#g@O6t2Ko-Wh~(0;3TUg)O8kk256Z`_WtYy#7mFK9{Dave(PDW6A3
z0FyUiGrgi0Tt+SQU{nJ=;D4(S1c)J-ci_znlx)wvsOzzmW-^u!k$W%P<KVoE8$!H&
ztFO@Jb*IUtonE2wG4#qhcGYI0_|U_ITjDS(A)Hgo29_Fk^q>x$lH9$^R}d{W8+m(u
zyROy<fpqyoVJZ$H*Y$lOwd+K5S(-Mp95=c6v}5cu^K|{Xy2z7gW`Ztf)%BnzieyWM
zC2T8&HX9EVr%h{NA}Kj-ENeX>+hk>wh~mE>Pdbj{Nwyn-@a13+S{*prBjvWK<?FMQ
zit-rNMxNsPX4%ptd7jSX>^m3_Eybj$t^(R!)_o|i`6PN`AoZdvHOkK|ULkg9*yKcR
zJe+tPLosQrO3p7;axthYGu*3VOyC#oERO#861AJ}|119ImC55#y9#{P46gX^MYcj?
zOW1--LkE;#xHK4M(|XX{kDIcHgfXaZ7a<L~PPpRt`3M%-8FN*wsEuL)n*V9RtHdDs
zhkLDBBDU|ss{UFr{zqXa>uI0O3!QU~c-PUE*pU5)Bon|AUJFot&NT*|8N`#GRIe19
zIWkP){3MG$3W<3vwIS^=pN!JQbMdC4bIoh1Ls$9)DMN)6eqLx?zqTv?Vj+qvSr;zA
zX4E0Z<~0R|JQ)O2Uv_J3FQJ=u+o=qzrDfYdxrf~q&TL*Rzg9LV_VY?6Tcco{;4*d5
zQ{O!rK8eL=%L}iuqTIFd$&$5C^MB+RV62B;-|ZmUPYY|AsySKW$U@m@EnOm@b`|5>
zCV&K1ux8C*8dK2eMUm$3ehea;6{S-{-+imn@vkbJ%T{b(N*RgW4T;RTK&Huozt00E
zFQX-e>ACdjL~b!(ZyCjVl@Yb&1JdhSD{&ci{6d(=BcdPkj^hs8s?X$>NhKxFeJoSX
z`+Le-tPVs9V(O@eqNrrWcbaKuUZF&}r8EPPMjf8qJ&W4ir^NbuO-9r;H|Gd?Be4BN
zmEK%0ywL8=l)6W0)WbyY1t4{*{cI?hU2h7WIe0qfee(&<XV%L*TgBJCdo!T0)Io|;
z(qT(-?af4^#n!POi$K}LE&P;vi*U*__dylGtX<<rG{8)&X_}JU#H}hDWAm693U&?7
zxYksDtDU;&0Rp(#%zW=#s=ZCk7VdWcdh6b^;Mv-M<JG`g-7ai8;8c0x-+>#vR*dtG
z#bIb~X@+|_HB`x*w!)p+!F@z$bB&sqA|$q3bcGL;E%LgnvS0I7>Clr^uY4(}nU&i9
za06vm>qNre=K?>YYLnxQj6T<^bwxm3ZxsVXJZp4LW6g{~KZ!o{>tz{%%@^baxv<&o
zusS^E7M2)k&r$n;9saz<@VYuPGqO)lVcaV{q`sbY+r8?GX4WuKcbmVuYNuh0#>vCl
zC~=dWMIhyX``xgMVSVefEf$x9bh8St^;%f=-H}W9%xiHAR^v$yd$OZBe+-zo*v}8_
zgq7Kwatd*GW2vnx&#m~EsTd)%D~Dzcj%Q{yrtR*ykd5(<GKdSlM94zh<;IDVfxq0y
z)>;|IhxdFcPM4zn<ej<|$Hb$UlwB^jjuxk-4#lSkC|i?-(%C)EFww5e==caXjH|4h
z`q{zvmkF6RQJ_`U-L7UgJq2bb%fXV){sSQ??mAxoFY8fFLQv4aqipEJ2tx(Fb*gNH
zK(Kw}pzTnyu;cf#*_w}5oQz!=EEWfJttd2v4#7V6=g`&!yeZwpGg1;_(_P1^t={W6
zZ8Zbs$=mJqm}}gv6vlN#5IB<-f)iWgE*|^p>4;Ho&m|{{*{3@AN~$zUB}S3_8$(hQ
zJ#vNeIr`+`C=blv3*OdpqU>LFXPO_?#rh)$N~PN@#Eu2Vw^$)>^)^)*CNVD?@f0|h
z2RCF=&_w97G(wGl#;JfUFM~%%CA9<SX-I?1%ChqzD%tIJ{0F9%Q3Lrsu)J0YS+slL
z6k9qix{I{XgPu6`rkOtYJ=%Sf)Yx+Lm0a^~!s5*n(gnz|l^(f0ZjAfqQ_~RS^FL=+
z@_4CBBJ#NxQi?sEH-;`jm(&~uS~S+APrfCLMu>b+b}=0Xmrz!;sd<-6EQbY>0&c^s
z#`q?O6m^&LOl+_!QY4igzALYHpl8(_n0}=kO{xNPlnNpwi}Q@dH9Z_LBhi*s1f93v
zZv)nZ=?U%@LTZN=!ApujkMQaFQ|9wwY|w<VvBzzkg+q_>D}}gUsq>OF9LF&boR%>k
z^eV0o`K~$vALP&xVPXm~-`NplM0)vwwca`EQFv9%CS@~Va=dC3!RB)Z=nOS8>UcYR
zd2W4q^~55~>zu}Gz!zx74`fP^Fc-pVut!}qjjFRETaU52l6vdw-XeP>wtSi;DkvO`
z_e?n%d1aEuj-Y*@?)#Z!%FR0Bm@A3f*<FO#(g%vsQF;VWJbMkL8%U?$@8`b&s)o#5
z|CwB4`>*60(;t@q$*M6CurRVS{Lkrs13_3BSXutB$u%EkNcLo#%~p|CLE{+mPg~O~
zD~%@O-8Ky7pow$?w2if3<2l*J5*kRt$Scw=`BdwO@I)T1-@D%3&!64T?_EFcF~&LV
zOAnn>pPjgV6J#ey7xq!mf*^rI(*8sO(tePOIL6?EZB}{t^e||!$Pg=id3yj5B2j@w
z79gNZbr6)e;ry0-_h6*_AQ7NYc|HIdc>e)OeGk%bAOOGsf?;85At1O5(P=<rTml4u
zR3487{{a9T$RS`l2sV;T(F+g_`0Gf(ekQ-SeULi85pM)_97ITjA7mX60>dFj^JIJg
zvGzrA3&ZZOdT};4ZUPz@WWeTl2sRjfLHnS{ylDW=?-22yrb&n(u|Wj?yWh&6+Yw<t
zxHyqcgu8!)68<4Xb%5JgfLcrqpfv<@5VT05)%N^&hCR*(N-VLf%h)k5N(qVWP8X>~
zSef`?{q16XVG5K%RQltSki_B70}&EH7aRakdtC^O#C4#3Xd(BU_<v$4VSVnuKd`^z
zID4TXg>v!t7yuEW1rYWih`DEOq>W(i28_e)<cxyWCV>O{GY0C#=y?75fXKAC{iqR~
z6QKuy7+^8PwE@5kwZqcvV4`=z2$#R8LqZAWIR0X92Xgk>2*XbO9tHWn!rF$^^fdq0
zUx45uTtBgOvd1g26>up)Zklf)sfTs#k0!9%lLVLZg86&8<$cRq(wD0H?h@Q_fQ~=#
zPm6^33!?LLg~{ZaA+o~}-?U=-BXE9;KrSML-@kR9@_>HKLh$e;v*mEFMUNn!!k{CT
z1U`uN$i%s)aqUA2!;A7hXys&aLxhRLQSxlT#r*~H{f23_5*Xm0AkiEBLaPz|%rAd)
zfCEB*oluF|%A&+8va~SOHNwBspgl6HV?qt~$d?6;j{zvqpr4BfcbE%Fp+Wp3I{k^g
z<(iM-#3c$D-3=l4MEb!I=isHB9A}P>04ZmX*YHz`2ui-wNb3N^F~3bH`2i@K`e;ND
z2u$28zdx_YgbDX)zq)@4pNun<u)o@5wm_qvK(7Rbv3&*)&mUj?#(sm~VPT``5eNWN
zNtx;>q`v0;Q2{*S4d?sH1|igG6jaOb5sB)48|CT%f%c>P7oy_~^e5ZwOpVPB@)}ND
zo-H@vK`O<ingj{*;&kmOFLXTj|K?(Bi}H?mJ&lfvu6eNL(-{l4uN~9)6+UVUVuG}v
zsBIPsCOkZtv<-FYJJ#(_9`2_M|7{hK`D#>`BdHK|9aQqkORe>ko4(;!osY>-gHN-e
zZMV*vU6!+g<I2RZl+9n;yo>CO16x`5T6@A|Q-$#yfSGA<fDf*Byc%CM&caVo5-hKb
zG}*iHf$Jy;?yhuQ!RIZ}AB*D(Lw8l=@xxkbHOI@eth7<$6^)g~)bq2Na4nc<7hNX0
z#I;^E+)`q93Z<?vAxnO7B4#(F9ZC#E#x0P=va@1#wdR2Pf#i668(OOyT35L0eStOV
zBiiverlp<R-vFR8nrE;9ASs@@Y8mA+mMK|yY|$NhDZ0-w3A_~e`+WM!k+H!Lam?3&
zJ-?asXsMT2>rkNU$mOF2ymntvt+neW9{XFcE4~PrV@IUbeXZPDkhH((d1|X>4`#n>
zzXoE4cF$wd*u>2^%&$sZFCAo0d-BZ_f(xSRK3bVq0i!H-P+zb_tt#%a!##o<tK!r4
zH4Eh%G&_ytLX7P^7bVlRIY{QvNO>XiqWQr#Mi+3jYVdjesqBOXl}F_S2f|)OiylPp
zZ522!T8$>Uf_0@zsdkZE$-TySdP{`OQ*Npj?6lhmn?e8~%1Hb!#ik31&FA(U@W~fh
z;!182ZRNYmAEZ<(jVb<7xv$<IvfZVVPArK|?&EvmdouJQLM}koK-aljlFeu2^kIBZ
z*t$b;0>4$i3jQY@TtO!>#|=&Q?;nue^coA_>_&sb?fk#1MKCU0TylEl=b}GG`8`2D
z)k*x)YNC&z-zGJYtavTOkej}&e}(sqXrvqmw2?jUg-8i}W9s~ii#WMbrCLQCL&Q#V
zTOpF!nz}ucW~nLJj7Gg+h#Bda_h|IG5S1T#8^$})<t{*k5IE=HY%i1&cBS0ZF3-QW
z+kOjGXFl3Hntz$TQ#-F@PD~@H@K0k%Bf8Y4#Q{h0-II+$1dYqks{eix&j8Qb{%{mS
zboAJ3w4SDKLb}k4g=Emi`BL|IiUvN8k_m}atUKaMs49M~(xa#L+#+Ty-AgUW?Z}px
z6xLu0*SJiTRFWRMQua3Xiczs`8KHN2yo=G=i14vMxH=~(o|6H#Ckz-I2>isFJ(nV;
zga`$rj(-Jl)u<LDoo_IA7IdqI0beAzXezce+TuEKHzvq2aAXm!Xf-xBe=Upr-Pj(h
zQe&(@F~CjI!+(y9=iAQusQ;AY)b713+S-nOxz5TpmK#;Yw2vipZC&EtCpv%^LUFkS
zP1}_-NPfWIvfV9j4PXFadBtMHO1!{e-p;?jOX$Szessy{+iT1N+`ndq_X!fnn8QAy
ztU6HSGqU_(tXZ2?QGylqi~Ci?tVtp1&q+U@O%_b_V(F+&-6G!G4EbUwy?yL&ZQnA&
z{UE#T%0HCiiiG<%Xyg*}7_XwCrkrm*jX6#x8T|t>s>OP1Gno#=EiZrmwH_Pf<z?+t
z*kG%E$ccJO4Nt0PuPa;;#qw8zIaS@YPF+8mQb;ez_%>Q<-5Bjn@x9qD4sx8)lnb;8
z1yFzwX(=>FnTM;pFR^a7@g$`}C!U=cS$049c`EL&c^7%*o$81ce|U`n;hy5Rjm+mB
zq_fOo8FxyXi$hUXX3d()R}{|HxYMHc0#`o8RjpOJT&J#_5#_tVS@}|uv|aKrRIYbs
zHSV97!(6%vrL$QXCs~9(D>o;ny}w=4tn;e7=2B-h8YijqpKpw!%o4COZx3de%fOmf
zSe>S#w-6%e=0mB#B2Am>P$$r^y&EmGcrjTGGJJ)+e{j)q3!+)1*Kp@U!KQOZo^?7M
zNl|Ir%P_OIDN&S)=B7SBd}`Jol=FfzaSuh<LV&AY(D_<DKK!tw&vQEdaRLOed$yTu
zM*Y6tvLtBePJ20nPuE6wyve|Ug$TzKy`L{~3_tMiB}Z*LJ?p`Wouvtssv?<AEF*&<
zjYynapje7B-s2+>uf*dBxE>nA;g0OR%DGD`85F?qks(p9@)w2ArkY(rsS<b2Ur!#d
zT?|^FR^v;cn`B6`Ty(f1A)3qQ>VuO(KG}wfD78bg=KMp<qzuxcnfjDMXK@=zt!8ik
z1_!Za0uDHTOW7|OZL&k&R8~)&(6WmnUW8jfz7{+dNYj*ld5FO%es4e9ST-pNaUj@b
z{F@>HKRlhoGA<y2T^3=|s-6?Nwbhvn9&((VuNnbTxawXt{fmFYlV)gdfM$S(@B5`6
z{XVtnh2Hfp;iXzlsIy_Bu<M}#ADteS_;CCJzA3EHLp|_W0n~u8F|il2u^Wi7QF2L=
z$z|6?jH%WkQjsUn0>jgU@m!g?luYno7g>nPuHI*lB=68wr0r>z>(KoWe+RX{#g#hk
zcu8Z``(5xB5}v4Gn4K^xVj%u8m`Y|-l7aEc6jCcFpN35Smr+LT#3$9yH%E6c29sqw
z<R8|e%Y6SZ(<}y@#wI6m|FO*rK$gQG(N4dG*^`NQOgZSGFC7Z#4=<^y<vmw1zw%%+
zTSC&w_~SqO<b}#dsw4?m-I(8wj@kL5oYtOASHZv0x$Vsckd4#>;^B|=N97&pOMlPx
za$c%yNiE0VoZWMnu4h@~8^j}rDhlC-86PiR@@v6q_m<bI4uYH&@9(EAdd@U_F8fkp
zVsPt4hkRwzAxJ9TvRrBFSO=1(!I?0(5k(e@nzGz5vD7V-jzomVCHxJv#e@NLf*pI3
zZLC*p2WcArLlJrua7z;T%(ffQSEZ{q?`S9Aq@+eKGsugYkxZc<Tk(T*ikp0OXFI?p
zGj1;(VsIPB72eN=0y>Z2qByEf4L!z-<Ywt-C{sMK9SY$`XGQ)1M_`;y9w|sX`QbO&
z#fBzycx5{K_(<4FcE?hx)%|BygzCVo<{k2^jbEy?Z9sf%)AyyU1*{GU_2h}&#(wG7
zPX}=EdWlH`mlyLIKB)N>($?H<@?t$LX6uD7WY@Y3dVie8{)SVe#+)hMyTS-J)*I(y
zh12#(*I@g0LRJ%f(;r6<Gt<^SZwHqN@n{=8(pW^Mp#w@X_!TMagTWMQ`af~(qGTV-
zslGB&6*Ps%sHcwmR3_7}<KafQZB;a7n@GWDI;miFjBfGxQ-5jvz1}|JsW2>hM%G8<
zmKRg;%h);&V#Y6AQ?oZKb_Gf&^H|VX2G(|>LMKm`^uF$YqP6%}Sx4x81MU{-kV%10
zCK#<xls>B<1NIC%s-R_2w^@QxGw^(EqI}ujnnQ+4@gdVC(YLT}{iC7Ay?l$bU2wv>
z>7jhY%{p5PhU1-C@OpR_4QoFG6+{nw1Y$N|%9tawWC9m@OAA9Z+}m^&Na~Af;{{D^
zZTha#9|E}Nj-uRuL~im#uw*%wQ#pv~^Vr=QC>KP?U(@OTwC^tXC@D#C+24zW_RF$L
z91Ny2<Y<5llsibq<njNTg3z9`dn%N>J&dNTH(j1=H&)LWa;y<=4i08>AI9UcZu=QE
zj(e8yoZFwWlk{pS7gFSkq77_J3VrWk9UDD3-HQ$AqABR{kdsW=KBmzNJ@h~7`~#*_
zton%bQD~Nb@qJjGPLWsCmAC(Lse*w&&YOI$(ecq}L@wb|%2qP<XtdJld~G?qZ?WR3
z;UKZ2_^L-Ys+qYp0W@^7-}G}dsb(={PS?lHcUn!ZKUS{%+ZoYxkG}RJvxwW#+_>Da
zb-d~yTa}^l8S~F~tfUVWm2l$fF}mI;m!Q_Y{*`R$`AUq__Ae!|QWW5r#6#&k!}n=Z
zZ$ZJ&vS4_4Ce!oQ3q$uDg7)6WTU!ISFg|D|@oE<>z+7>vHp9$|z3$9_dsK7{eOzir
z=Pj{J{hU!*tIlqsLFhF)t3mkv@DIKEn|9GEmANSsa^(hj3RcDT@9kG?rs+n{XEX|?
zTi+VYmcswr)75}jJ*u+dL?2f?lOMG3>ZiA=!oF6^JZv?%`sl@PqRPU1li;k*?pvbo
zP@JIqm-tKgjG__a+m`~@P)vZbgPt3ZeHr2!ZLGlo)G5+|btSlsjEA@5J<j42=-ctU
zOg?IvQ5zpjEc&VEBo$rQtlgKoxP-c;c`uzM-AyQo`l|5m%9)}aTbrX)QRq6p$tQxn
zgtL+_*v0)P4syGSC-K<otP55VS;1cia#ek-!QX_pv}22sq_$tRQB~1CTR^&v1_JB#
zAX$b$1Ph(fsH<C6^*X`!e!dj?xl^HA+8o3_Iq||{HziJ$Po9ZDyI=8s4Jzhk_*-?1
zTZ#PMb66GJ#v8#=EvfPJkqj}Ub%sv4#=In%2$;kqSg%3RXi0_flusYGB&Z$)^{#yt
zA)^Rbh(D^v&)?#C1@*TG&)N*P=lccxirYU`nthXihbCGj@~DVUZa-;vm8XbBFTMY(
zmv<=>+HH%SjwRAtSurlNJ2jmg1Wg!e$Kc&tC&!4s!)6GB`o2vjq_aANDscEdZJ6K$
zYUo`W?rQPv&9pdDpLE(XAN^E$#8f@Y=@t9N9J1E#y*lw{dVO}A>&P?BB+=fGx;toi
zyVjIM>Ickmt;0Tma7!!WBgm%vyfD+YgD5C)4-RD<Q@}6I7vaJz{nn}rJR%QKDVHC8
zKQN`(5;zHp`n4%ptpU{u|K<Nath4ZZ$j4i37)C%2>eSmnwiu%c$fw%9?~Y*pPQWs6
z17NSny%#Hk+*M=p8kN7LXgYh6*+@?vV2cw}AB>b<YzupsQKqNhM&=W1(|EA-`%#;M
z8!D0M3k}HudvnTw;E?W@a`Eh2_kV0v)=7;%0N>Cs8zSjU$_AdPFEJ+!nCn^NfLGh$
z{us>~vl_FxF?PLKjr_UTIzn|23WfRIRKgyc6^KCobWOx5l^Lw&gcfpWLK`>#G*;_r
zk$WiAHW=#r8+ElQD*R58)ikjx<;anFsa?F}X0X;WQ@5IOU4pj40Hreau3(p}!-LU>
zeg>tIJEKA$k9NAN5I<S(mH^{>I=A*y#ZgsiDC>SWB>EO`?Q*h!)o^s~$oUITs2HgM
zh7Y727Z*m!@@c_pjPrtdH&;`gwUa}<47+n%!LnSZK$}j4L4q^4r1Mex#>3a~i~Ot|
zM}=Lb5tVQ`KFxv0*=-QDJg>S$EJBzQ8XPeT&aA9lk-AFaT{(*9)Mb3h1@~ctgwwyp
zpt2BCpUpI)mw<mRL|#@6LxxJi8-En+D=OY9%U%S|FmFbxut0NJQL_?V`~kX|KUV}?
z+0bTY_Zhu=r8;XG_`B*%=ey$jG#SBNVh$oh@6xp8vIr2O_3FN{np{R1AV+6Vkz$J#
zd%TGjUgm}mzlDb6Ub$KOo-C%nSRd$$WOy-mw0umvCNt}ROuDR7+HgF`ZOHvt?;XXw
zrTcSr8@Ipk;@KjnpvJZg?jjYyLA-pgC>L}T>G$$F7;a~gW!sZScQF7a#r^@`g5%b#
zo>`n7zfd(mh>f>jtx9vD);RoV?)?^x&VmQA_O~rc0@fmg#U3x5*4m4kF25nV-Yyr8
zb_kkD{;XC`epvp1t1RcQu%U0^_S_sNYlgLn0!d$N@C5WDDZ$vzFZ?QT|LT9rh8+JT
z8!|KeudPrP_Wv%tva&J%p9`-p%HZlsR*PT)pdcxQ{Ux}WixT{rXF8sBR2Yf;g@lll
zPgIh(1dAvj|Grl)mQeRZ5`+US7P}RX<F9Xfx!vCUW}i+>Om{Qidf!gFfq2N!4zK)M
z-3*396G8!(@+a`8T*kHnAteGt2)7O@J_iA1y@#Mc3>h&Oh=72|4-x|um8S=w0rww3
zsK*hO#ES<CC?4mxB1Qtd5FGWNfJ1-)n8N8h>(>W}4MYSa4~B*kBKMds33C_)+Ozt-
z8<J@E8`(m12g-m%_#p3M;6Dj9Vl<}*h_x?GS{^>=zN0`VK}0|bgap>42!=*uAno8E
ziS+`=`I40Q;+evb5*|SCpYJpI{hcu83kVnENC^KY6%O_zy8DL<19Xn6`qKdg^$RT=
z;gGXHfng*@8A2R&iW4D%CnP`J-NoAv#FNy!qf209K*iXTNp|kdBMHkD)`R5B0^|b#
zhWQtalOp6YfZV7nGXjM5cNIdAkBK4zpmU=*O-m00QNX~+1QL%w8TT=3umXgQB_7ru
zL3AWjrt&xY?<=qq`HujXYI6EBL-|C9A^~Ec2@&820qEyUh_j=KI*P<U{-P2KDqsNp
zKgP}>NV8~5qiM5Jm9}lG(zb2ewr$(CZQEw0ZU0@-5gqXcZ_u+FaRzsA2lvFc_qUdS
zu=}q#k8Gr4P9%Hta*T{LZJ02zoMr){d<HtdhNuJ;mw>|qbFhQrn+B?5P5b40MyRof
zGd~!Rk%M_%@{V<s<-9KnBH#empSt(6_<3`qa%zo<qifLEwc9=dc{S9{OZs`h;2Cnk
zF~Tw<oiUmKz`g*Ai$V@;8&xEj5ZH_&b9*~t@4~&-92G(k{l3^w00$1_o7OZ!c;Ln`
zR*w2{E*}06f6FM=p0J?ct&i2ezaOy6s?nO1OZp6N^Gpjb4CvM&H?V<&gXLe9kQ7Dt
zKfITx=Xxk1)MD~SOy@D?7<_bBi1#lBgODOA1LF9#5XS?@vvZ%501BGeDX*q>*6=w@
zOb;oZ*`8STR3?T1CZ0*8;75o$&UUb8Z6UA-^7p}xwZD<cp9JP8dN4m+*Uz)0-<ZJT
zd!OVX<Qs(wk<a0<{>=z}!s?q`J91ixAi;|p`E~s}Z!$fRJ>L90<kLxUOfhGDzt6f+
z(LI^m9#r@T6({cW`27-411M+>g;+(oL+*nB87>(&OG-<EM6I3p_$L1)I(Rg3>Hcs}
zh<L5{GVv@(z=O=tnLtP*egBh_{G@EiXk}_uwi<({M`~fAN?B1c^qK|Zsf{)(<6sb@
zY~&q=Z*5bzr~BlAeJ4`B6d_7*^CE0rfSP~#T&f$v)zH~_eR7c!EL)WK{>^{1%Q9>J
z>r{jrPu{Y4tasI|&d%pFa_3y)(dkV!iPg?*!a~jKW2fB|KpO@6mp*+(v@p=jwhqSY
zG?DM<LoP`6ZlrVkj9l%$@v1x&Y%8U(q3SGA?NLB@5x>!=A%j$AlKKXGUu$R!a(W9<
zXH9x`#R@OoqS88NLJU4O<F{|Kp;0X>B}#*|RhUm3jsu$H_b8?wfc9=%8I8l8awIYq
z6s1}g)5whbY0DC<o6Lm^@VKqb|AXQyw>2<)iMdyNSej}DaLa2V%l~oWQv|L#xn<o{
zFcBI3E0rTE6GpuCNOJmZJoG?<fX-&UJ;iz>IlX#TJn7s`N8Xva<zN8)m8xg4LUwNY
zXFAdsO@!IG@nw11XKc_V2em9TK3o-mN-$wQmaLE0oz1cNz(m(CAUf10A5@IZheX;J
z(+}&@bYuFjrcu(~QM*K$>GTpPLi8(E%{I@jEtGpcORE5PQBZ+{?V3$$t9@Z}9`Rf%
z4&Al|hK%KDZ(Wj$+kg_#LSdvAU$ksFn&(lVDoN(fNd6QvARA$rARD0A5gQ_KSt0jX
zas^{=i`XULu$Ur`o-8on851wj(Zs#Y&7O!Zp@62tE-xV_Pa}C1k+;;IG83`alb1yJ
z2cOo*fQFK_%3|~$Cdr)oTLE}X(^zz^<!pWi5k}4n$NDM2B~{nCN?2+<&az9=FYMOD
z<vrrE)iz}^1~*b3BU2(2M+T!ZCAZWR0!wV*ZZ(<9^xLCGO^S!=Uhk??Wp)I3C?V}v
zBe^9bb6yk>qt1_Y<jLZbxNL@JL>>>9!seiS?{ZoR_EY=Y80Fq}X#FAany2klkH+g1
zcDbkNl0>m{FrhdtOAq#kSw$fpp$ZpQT;~hbZu$CGE%n)n$+~h{$B-hq?xC-jk1Eff
zmF!>o+JuFqfyjefcKwq%)Dvz8SLjN>bMjjJ?j~HmPTYk>Ov&o&Oqgu=DZVJD_Ur9$
zl>*MaQAD(SrjKb$B|Crjm+#Xz?lJJk5|tR)c$UQiadfHaJhXHyVt6X9!nm&%<>)c1
z>SZ*Kd?|1UiAeC!9rQ25M0Zde%?H^y5KurdH4;E0;bL?a5;UGMa%?8;RPi|6pGUZE
zVo6iAWv6;SOfb6@6SfoEy<k)67kRU$%Zd+fTZRl^6t~lAn-GqqFQ0qNF7BO@PmyIj
z{g$+{o?@Z<dZ+UhOR+simS>6`ziU2qB<w1(|6-K>>*Ar7@;o8K#<|?0sX;m7**<wF
zdOj?5>6I4t+05O2*+xd7kujcZ$3hCT-9TWeDtWkq7^;sbM|E%W#p_WYbT#|}W$fCB
z{4|=?ELG2$DjYqgR{5DkmmX`vV|UNXs#tWH6{;0;xj9YlShbd;pZfdrj8njvkeI^6
z%7}cuf8xHHvICZjn7<tga$zh++7^m4^<<^?EyL1xyyL`Yf_3f-+wxgZFvgp2t!pcN
z1kCz0%VzW~eT#d1G17=$@n%X}hk-w8(Mm!!X~z<x%{LCKX%q3e%~vg=b%15vGCtr=
z*Q!7-tZzV8_Zr=|T;IgNZey^hIeFP??`Ci@TPl?}VH+DwCl~_R!eu)uwUO`$Yy0$c
z(TZ3UYX39IY$NCWc43go`nvORiqaNrlg>_sf%GT$JI_y()xuXr#kO7w1%7Y+!2~BB
z0lHEywKRM7T1~U*>!^n){GT7l!GK-GYV_@pRIqk8{_6)Ln`jVY&?$kxx=&k48Ch#_
zaF+qJfp@s!?O$jf0g@}p@|?wwv-e_>(hK`BF}bMF@Mj`&yLoe@Y%maNjwVI>ar^ga
z#VU>Tq&N7zX~b(0%UEuTw2SB8))2Qe{e-eK7O#H@Kn>$}t@4yynI&0I0FJg%Qk9~!
znR^%cH_P=>XU<3L9f3{WoedbZtL?Oa>Ep#VYfXP@?_!-?5S*;ovUO9~4l1f;vO>kx
zMM#J(><)g~)4a5_cDuNRodk(7b0;f3F4mYNuK7H4u~>!x7>uNl0B6Y;QYrbQ<`8}6
zN1ip~1cQ3ahPeQ@@dnXJbI(x@OSz?@2_H?XF1TdR0C${O;`3%+^JbfO%?>tyHUG7m
z?XMZ%rv(u^k|%|U-8w0TDIp1>n&q*ti#f<Di09`39`uX9Ad+S)X@+2#;MflK_>_CU
z=~ACN^P-#PoABn9YaDN>_uG7xfl@<J%h|hSDmjzViSH^@(=umI_V>TW6qEF}akD@n
zgen?99IVn5$wpiWm$w|JftU42Y?#eF-N?usadNWT@WDr2Pt=M!g;w*n);QFvjw;k#
zEWRs9UuQ+mZE>@mTx9oaq411)ylsCfhm2`!Nny!q5Gdi&ev_noRrI0opC&LN{mxtB
z<(hLK%hXa5p4Ls=b12N$BM(j^$V(IN+DfJIy(qcmMeC3?o0omCZSyIM$Ynjr+HUkV
zW;7W}K5yO%w{+z3@xEr-a#P<JGN=*Ff>^6a<5@s-_MUd$l|cAdV^`smTU3?pN!{qS
zfiQef&AwlThf5vNqnlFnQYy7wdFgNE?C|tD_LX||4dV-LDV{xB%&PlpoAcDO#m$ca
z^t}K1_Jv4;3ET)YQnKb$BFbk{;-)l>_Iy_{9B~`$=`2~OZp$^ht-OYvMa#-axtY?V
zzk{d6)P-`pv_kjJ8+(_hq1x<hkVOsEyA@$~m5A9Q6DHaj>G7kdF8;huCE-?A_tERG
zH*WS~So3PSQQ`-%(9v7(WU;`b^#mf2<1|Q+O6j^Y*qD3`TNW%TaIg-2@)Z9iQxKKW
zHusqoHxxZ>qPToHZ{|l(&vQBzzS5)ws|sGvf{?ZLEh{&_nPHJ#8GoRB1lsXOP3P~K
za<GiQU7Urm(qwh<t%+W}6^b30)}SgQ(X1gZ^Pxu5BK^k?bWxtEG1L?v+mEA(%j{2+
zwcPrErbGFa#7qy%a--na$YELEN!+H;`|&bwK$;pAfm*A-$ij%+ul-6bAaFJv>XNS(
z(Phvz&uj*AYE0TXS*VhQfHJpLS;=6|Ptez|$vc3oBak69^}wkJ-@#@ud;A?T)V<}w
z)=6!qdG>6i#mXE`TobYV2q(XHKI0_SZSu>OvT2b`?})VB=FFVx()}D8UdaxbY0d1~
zUC--@T0vU(qxpN$QvKy-@}EKZ*$UBmYn8tafficXXi56uhw?d4M=r$z)h}-WPx;A<
zCXYW4pKGsDQG*P*^V!-sq}R=ykie?jTEp}(Yu_N#FYuG9&$B`MOOPDaU6}DaBrIql
zr87#8VUim)=v#752tLmm0zS%zBn8!a`enj*bM<t`L6SKA@#!3OMcfd7cJ$NP>3U^z
z@~^TyOM94F&)@`>Bsce{4A#WD<KK6WcE|N?)XOCus7+p%_PNv56R918k1n;Mgvw7P
zDz^n!!)1?S{e@5HpkgsX&lfuirWU<CP8{d!@XfyXFCTfn!=q=4s#{Lz^i#Ty<Ya2Y
z_d9FKtdeHL%M*;t)J2)|Bq4P%m$6miv{A5!g}F+0JccjR;%O2W7~VegZO~S}wR(9g
znwECa=n9G>tU#AsJ-8<suBqUTdPeU%Np=afk@3Sz0}Sd@HAXMlzvqHRbRH=p)`nDW
ziZtvG$p<MR#q|=}$2O`}-Eqt~YN2DHDOXu|sB`2nOU)mRn|cSKXZWGvsI}=?g9B11
z0cX08cVja%9JBV5c610PM_fl&C$XCaO+3ur3318j-)PAL?~17n7u4KcB?3stG;Uca
z*b_x8ZI~N*GFOYgc^Mq*uJ&VSP1Px#fFnZ#+qTsU`8lMt-$cXOC%cu>sg^lj2{U5R
z6bY{OuR6xuG}|I|EHFfhMh)@ISiQfMW|SOufkp)$;dX~y@vU~(JI_WR=oK(<>`j=j
zmGvw?2r){{g6s(X!1bXBSAB4;m70JqHeV&-&`Q|xAhf|gCh>Qe|LwK4#ip#%acMSc
zr7{Rnlr7%T1>j6h5Q&WsmcS6M52yCDw7@2(rB<p8p5C>Fs=?ee`gp@Jrb%`dVCAze
zofRe_JA@6U|G{zXV(vDoD|ntS8x}9hxc?#*+zM9fa(%Ds9pSZDHle-f8gZjy{)7VD
ztThI!eaY!a|HY8}=X&E@F+Kq8LofF?2(qoqCNp7GQ-~up9hYcRGAj66n+j$1?_GJD
z&1nv0C{;4IXT+`{k8LWod?X>THhsPTJhvV%&6Sc+#!-L^?kG^I){j?;g0X-_b?(kx
zNE|qw^GMe<)YsqYK#b>w+l)%oUe_7@UycG{WA+QTyMID9cmWNIT(P&%e2^JZ2Dv(9
z7d*-Ek`dH=h3SbTct0=z;M=zr(lWObH9FW1AGvNT6y8FuHbMHFghSfL3=76``&TUj
zml2=`raflX)O%na5u-2%eAt*y(T{ay4SnMd1vOiiCZSDAl;U9*abt1RE11V~I0L>D
zrA&q<r@}P7IDluv1WGdQ^E*_@F+DV=pVpn_1Yuv63Vo2+Hi{Nsaq9;BYwAdC4*pz+
zuSWQATZ@*OFNcZFONv>#ueH7x_O_3yrvDM=tVOd-g>>D{>XO9w4_THb>T3=wZA)oI
z#(DQw8lf-s+VNVVwTGWWJ;RAx3OtkRxg%5by%OKSQgqTu$9|Dvz7M_?>|T%RxhegL
zfwsTp_Q|K`CnR8}u<S8kO-Keqt!<5~yrZ@Guw+PcFi%DFq1^Z~INYX!sU|HB-Y#2p
zpUTt38%eIECCPk22+ggO;AR8;+&~}3`i((v&90}&{9jSm!jUP;GkXZy16}2<kuaO<
zBiS~`JVt2YuV$nQ_zNlEOXXZF%UecDaB#^@<+FMZ4{w@vFnuW3mGy?cuX#nf_ge*)
z^0x1TxcaI)zHy5Z=k~qwY&M`CM5euNZxeWy|6+p~Vizf4#*ozV-brb3orhh@?>hHs
zoe$TjttR>*T#x>bg;2`{#89H*3LC>BsYy&Fgs9M?%P+5XP+5i@3xt3_36YRI@n)6!
z(u2OPZSOgsqIrFCy-th~JV{#UCo6S^6J#pm$1GjhvT@bEgcik^tP2(0c6Y#^A5hxX
zO<W%_2ltYt(=C<cd)J2*PR6$uk&BUlOn_k(LuZ<XcHbE=^ti2xyV42NX&QKsH#Q`r
zlXO(s0~$ZEyF(+$j<={G(Vw#X7)G@}Dl>#d>mCX#rWdhEID4%f5VRw{k7$!~RERIR
zLn>EGYM~$&Xp6fFVv60d!kJUcTs&V#TTuqSR_8u;tY5sjRh9I{zgP(fEj1dGPo;uW
zvT2h<*41_=h0uimEs8R(Wj&DedscL{zkL6MrEu1}m3jxEvg=$GL2MFV|E7!`oEMaG
zV)R`eD#4jL{}sX{jaUO-v%WjUhF{x49ec6n-#VjHfCp`wnhEvds&M_6SE%Q&oGoo{
z7M~#6+?M2ak5d|Wo-lTs^HZAApZE;qxp%UAuij|{P~;!2)Gl#@4~Kx87+>g;4SjQb
zjv{bd{*p;g<6MA+#Vif=2>~hPux-aL;DNP`(*QW_T-}x<zQv5=-*BzPB@Jt>)y8)4
z)Ea{bd!5Sp1Hlv0>`p!)I$(#!k*JK1p%p@;nHy)(%U%cBRO?N0?hM^iTw%>19=6Lv
zQy<y+$e90XH81jJOKssx%R49aJ8ug?Ac9`RQ3p60T9TH!b~iZWvj#C=nneoVFf&%m
z-R1pyn5UDT-p5yfdUkt;7;n^yD?;pprnH?Z)V|1?Oe!xA!04)WUj%o{2AmcW3B7j<
zt;<^6Sv-@hi&-rxSG`PlML+VZysy92{PI;|$Rz}UhI*nT?c%_mHutc}%RmuY%~u0+
zzUTXFNrW&x71CS`AH*o3?5}I{+ow%&6D<J>Dk&)*X-3#8S9Yb+pHhL<DJ^bqVWaFe
zd!Ohg@#-%^%@gP?@Q&ACkXAnpxLcGPJpWc=Ia9$!r_4%in2vT7clar_zM{)Z$1=pE
z4i{A}vu3CyOxF<qa<RIC3R&U?StxNbD$iO&V>n-Pv8e_s(=e}M3S+!t@z0(mJ6F2P
z$atYgM6mjUBR)MGOiOpn#>QSJib?Bzg4cu?nrGIMqhjwQf$#RXL;qCbDEXEFbUsz6
zLQYlF><oFk$V9!}rnlF8ecB^$P3ujzk35|oN4fZ~iw^LmyBjU`U$Ds@RYvuq#38Vw
z`Vy*|ve?aTIY>0AqJG2&vebI4P#mn0E!Ks$zI?egXdM9w?+krY^?HcvJZ#i2u~4ti
z#yxZ=qX$KAaVGeFp-u>jxoakG?;+$Tws*0SjD4@4nN2jF9a}_gMc__iUnRZ-t8=%k
z9J>m}eW=w0$*WZ8NXU;6j|a}&Lce7T?au%}KcFO;|1I5SXZ&yJHY?+QTJktK{wvOr
zos)y@|GO0H;tZ~i^0V3MPf!AqNW_Fy%+}c@)aew>2oV&+h*d1eg&Kd89FGc&hC~8>
zjtVU(@JG0__rPQB^)>tTvsP!V_0;#l_i${q*s7n}F0MJKl0b*Cn}9DLOaS95Y8k=^
z$76+29}EWs2&92QK0ZB01{ogQXkT27lmUVYSYB`lcvBxANE{UiM8aQ~6l@p>L?9n%
zbzXgI2p4W>)4v|*U%U;i2SEC#YV+a%1Fa7l+^-$}TBBUi!HoxspYOBxLR{{1$(3*B
zW<&9R^lNkiuJ5=`sD;A!=Uom+*F%nY5F9dy{S8F^FBs1U3mc3F(n0*4e%cAKpJxZ$
z3wZI29_}ZU2RhZ;1^AiK+{6kFUGp%=d-;b|@uI@_bNm6WjEe!Plvks`ithhd*Krev
z#a>>)#a=@W5|p@m^z!hyAJFD~i#Pwx50(W3?9zWXBoD<G%^n(sNN|e)1iGyt4~2sb
zaseMuSerEoDj>K(XaCCId~gHw8|%t52Qvt)UxyG1(TjqO7u~Jj72f|3af2sDH3YUK
z#(6BGfBpdl98d(r8C3-k&VIO+6$1p2U}vY1MLTFbg1VF!$?y<(<TIcR0W`F$6O=*T
zZrTNpo5B_0h5~FIG7AG6CL|O<IWUBK1M^AFjo4bnz_5u2>;gb}?WA5;f4Q^5x><`2
z2UtM;4*B!;buaSM@X_~v@Y{`I@W5|IYy>pG9nN*Y^}!PwcufoHZ4Ld0UaB4z=io8!
zy5-z&>uXAB7cJBVqK+ev;ZBcXTL=6GzGW@Mh#(dSk^w~P6o327a~P_f8g*-qUm2*I
zfZ(prIW*Qjyy=ql-Y<+f;KwTAsel0y?Dw~xh=9&Dpv%!360kf2P!|%@5DfY3yS<J}
zX`}|T@Y~@8Hb4gY9Ru-+WHbjw4GsDwu->pG)KjN6hTK1Q8ab@514Mx+4dlvWet1d!
z3j_Qv2$@3cj`fRP7_Q$Aa^wqC+v%t+jOYgx1o*9h!i@vu9OeTSZLknvKnArR7NDd3
z6Iu_Q({S{IJ!mTIuXA+;7##-wqQe7>6IjF7`c0|xE`oRI00ay{fP7JZ#|+>_@oo<s
z)Tyt*h`s5}uJ!9R0;R&IPU}Ia6Pb%0>-jVI@8KeRW<iOA^~V7Nbbh7MxViyO=>2Op
zKlK~{POY8mPJ_Qc^7DcA`G<apgMJbknnZ_sfYAl6x?FjK`SbjKw7&UYL<E7);qQ*n
zz(L%>g++i5)h`MI1Aze+(DwpG4Q<b-q$9OEeSi7wfKe~tLy+dAi3k%7Wa*if80*X(
z`4Olz;6A0GVHTmFs<pcTw6(MIS9gT>7Ua7;k0Nmu6`9wmdP@5=r}?;qR)yD!gHpi$
zm2yU1%QI#*H9o2Q8%dBY>{FubDA?R<sEpX#pa&c<8PF?Ox!3r5T?BEV%plc$t$%IO
zG5Hw0q5z(P4$G7K+McV=X5^yVUB^vuJNvB=?{zXWf=C-(>FqcO%}}`A#2e0Srn#UB
zS}nlgzP@_KB!q1*K~J@2;NzCD91Pvm(5+Z9Qq*lFOca?)4dIU5*swPneEmoJk{qs%
ziw?#5%x72p&TdkEcbcA8X3cw(P<dz*N{@=T3kj!gu8C~sDddA+r&L%xmT2A06#T=$
zqSl;LSnFY3{*f}A^-y$69Qv)1)mCL1S|?I3<;V?Npmy^%s)$H#^6i;+%4m8qJAxyJ
z!$e(<mj<cuzkjZ$l)1|uW`HR1hi}tEVeU^7hX}nPg^k%L8~USPf}Q@bSqdLQUXR9K
z+dGu<EAWj!<nw<aJSv>C+@VMog6jy0lhAzfkvujU?|R&XuR1)}${y9jLFfzINqLad
z7sZy|(XDWgXJqx7_cGdkGSjxwUbr(O$opjJOpwXUAxtR2D{#$A>$e0X?4!jxyH4q-
z_sU+eW5RTqvId_1Ueq;noQW?44i_#zdKp30?eUHVi<N}j3v8Q!ogH4XQ<FN?#II=@
z#S3DI(3P3mXui;e?@b`=wkKUzpK|qS+KS+DD7QoX%Q^`}p@ehfmy;{_wBL~Igh7Kd
z30oD-Z|9`@y0PsMR`9_`?a*AkSG=c6Xg`<u(?}s$?|tXaoxzd?Ca8G8IS`%2qD3fV
z$>S=Tt|I68;Jp=FOOf4I^T70dhUw*a?|QU}VcLPr#YA|>UMo2$@_X)u7klasXO%FB
z!B1wn%0VP&RyV#Z(gpvtF>tPYdx^qZYh5X2iN#It=uN_Nuz0*l{c4@a4-tHK$qBbR
zo$&7`7vH7{1uZd3#ApUX(&;@m_s!Jh(kEtS*iY?4jK#pW)#IA>zB1DLn)|_F0Lamv
zjiS|`m+Y;qTDO+S(o0%rw4$lD3zE4aJNqtLR5Odj-)j>4Jxy}mTH*?mbVY>YEeo_I
zlwu~2<H#O^RE3>f;d-6xOS{WovKJ1&7#60AE-L)K!gNAUmRKTOAoW)x_gW<bNTtqX
zm*HCvfNsn*ab80_%_Q{<kI&%@S<+yRq3<%2hMb^h?l<1K%A@nPKO_i$FP9CXKrp}W
zdfjvxUsu~b(fE#Zy*wg=@_|FH&B*JzXsl#7|4P+Au{>%3=Ir7*soLj+fLe>Q6RF0>
zFI=xVycNQr4$!$Utd0)of4`fJlWSxhzL@OS^2IUQWTJQ96Bpj8Ghg^OCCEYgmjsUw
zfYGK?OTwD$`pGcNIu;>z%%HCT;WK$c+^v~Sf8g{8Jlu!)CNiruAe6~b??Uc$m)_KD
zRHQZ*$m#Wc&!RhY>nW1#$oBC$CyJ^L<$;cWJ(+e>%A(xWpnTIh9zCM+vJWJe;=5r8
z`SH4HQVmCf_d&Mrj=<#)TGx6@SavZJ-ZSXs$Qrvs%+%=zw3S$=krxS_9L$fG@vZXz
z$>;5xA7@~mn-RNA7<39FlMGVK+9^^ByvUUrhb2wBFvD6wG#n)1_iCF$%~JvyHRVD$
ztpw9%E426+2yI~me&Z1;bghO}RnbDF8O%TC8NvJx3NY(=$7q_=tCsB`IG638ct7F5
zNgHp_oUef`2+wL%ZjB2@TCVrt9x5&rV}xXGM9E|?yeBG<8+$s{_4~Fbs{juzDleDS
zfaIZ*8n&gZPU7qXtki!|r*bIj(ww{9|4GvNU>0esp=eb398+mFAl&5SKMs~1;cR~Y
zxVn5dS%UzTt{uKGH#|=2(Qp`xGaT`wvf=(trb_gRa07iXplJXiJ7rLF;g^cMe=u`D
zNb50=Mf-V)9ci*#E<?l&RbmR7kXdD)+`OEbEy_t-QI1M6mGZJ!9wh)Cos58(2Hu#P
zgHg$Ah@@b}m_(IS{^tL{Bvgplt2WLAXyBpC><vt8p}f>6M40j%u2+BdLiOe_UwAvV
zBJDMZ@fV8Jx-1MokoXU#VChy1j#JerW;519-pFo1A^B;G&XV>f!6fQ;2Sc)4l1}24
zVo_zRBOJ%=WFgvY>-$h^-#+?tPO%Ud(8FU8Ce?xFCrfU-Qe;RedDnK0JvkM8mo=&9
z3GGT9F?YaH26O7ldPueNcl{Guh1*WF`tuNb;eL)CI3Fa;z_y%{vfJ`v{p;C^aam-<
z=D~E{&6sqSuSBmtJ(PqW<(#54y>o|>;vuXA$=;jRd^0BiZ%T~5tbj@Jb4W+_Np6v@
zp;hL>DEXCAyb)9;^|1A5t9~})JHBtVFK0fy92&Kr&!s$uvaM1EYa>s1DS@-ih$!0J
z(pq)L@(-t03<nw5TKNbs$_0Dt&LEw6G=pb8Y7G+njYHV-AxNb6<aPCfQdPp8BHX?>
zj!vf>w!5-H*kjN|DX9WDppl`^>(-b`r=}}j$Xw|!ee?Zm(Yr(je{p+h)gTM070CZV
zNc;R^TSvVuzc`YA|6YSnEn#(t=^bI|2ZE*Ua_p0HpSw};i7S#a|0v8Z>DT?8kK0nn
znUl5!n}Ln1(2{rJ!E%200Q_^6)55KH+^(za<zf|}9Rl0&5tqc9_14G1m(gY}vw){C
zL6I_Bd1+4Yd)50D2<?gsL6T5L4y}YI2Y<2djZ50|;+Cz!$O8iAB=}*q3C~geG2-I>
z#twggbB{K14gLp?K>q>vmmi)^$R*U@*m%!%VlXY6#tq`ghlyGq_Xl~A3MD^n1{U)x
z=5UpnNaa=Us<Dnz{sYJ~>@_7qosxpI1g^|KV;Fqx&XN%#xQ!8M6>*Xn0TV1sz_dA8
zcBN;Sw)+gdr&4(4jLX)W%!>vkgz)Me3<{6L=Qci!*eQI3+t@-7AGxi&!#K^it9=-9
z_=12;F0`wJ>sxgk3C&ioy%kQvggnyi;|`)Z2Lw!&&<mbebQ#S0PMH#Nj`ZB=PbJ>N
zWpkR&lz4&UjP4wv-H@nL=2K4EV$ts&d9*)V5mrM@0t0;s^Qro@8FqRv84(NPrCoKo
zni1KUP&(JSQ-c<Q1&Ce5)4yno;4i+o8f5C9NjTOhZ3kE9z)9WSH5dJIkBl4?M-IBs
zLp@vF!{WKRZNr^|IGT>H>B$%Gb0133ekBTPd+zy-O!Y}L|KkGIwz1d@1A5uDYxqux
zCE(QZWFBZ9w@YmrbMZ`RXL<fU#mK02<m(1W?@hn*5%q3PAQ?cu*zVS<y$Uk(4+rIF
z@J29RJtM4=sW;k`Ku9l!2qjN-4RT-IgCmFqaj#Qj1$`cMpuUpW;>kR5;dUW?m@!sA
z38SpNwfoRI7v5vhe7!tG_eN{;7PWN+27EQTpN+)IF$mHnUpWn=#YI@t`QMc$h*k`g
zig$n^=QHc_n;T*vAJ@@!qMMXH?J)mEP;~GJ;AW1vj5k?xJE%Pl=nyUB)Di7om34jS
z8Y2E8K!OwTnh&7&TQ+hVM03dJL7E_D*=7)<!@G+t5-$0EIgkc{6O7MNYaFowbXba0
z$|*Qi$%Y9`C%T_SsE-G$wj5oInmr)ajVF5|(H*o^<Ao?6CVMUnF!{sJ=GL++sgLX1
zfTWCHcRH4_;x}OUf^G0y!PSPVOnvO1D>e{yeS~(~QqIO7byPwiOh8NC!huHE?pm6R
z>KiUlN2O9;lB1FWePt^{VlB@jz5X$+7^Sc%xkDaI58{+o>}k<qx9H4&|IrpMg^X>9
z6WrcmVRf49is>`<+G$btAME(YF^DSqYU~n!l1&*J8o1wmRh&r-r2A6<h9xPeXAW!F
zo1FD{1>U;dbG6W`#d=lz5J99UlY{L#x2N%R@^e)<|DGrs1jGJ4mV%+I(4q2{%okN?
zX(sJpSXXv<%SKYCJy;?!q$*xkwDd8<72$C@XiJpbA7OG*`165F+9bMpUgNR_nbo$t
z?Jq0Q_uR_N!p{G+5IL^jm`sIMwd;}*l{j?2W(<**d9Q5$vapv)@~_nWbR^kBI_{Oc
zz1b&jIMZkcz3qE@OK#6fW!gJK;WyYq71oGpgE71h9@}F-<MB2mH0&TGsAy;fT}DPr
zkG`Ivcw_>@*?Vjyv7h2wMuZOF&PkSc_py)E6J*Qsq|G0ijA_fc?gFq7-5n&r%H4wE
z98b!Kz}8|}-V5E*q_xdbTU1jD^C1e7F6XBnbV_O0M(y?&>LFzRjf~=4;Agp=1Ez~*
zjr#zL!{(oN#oR{o4$oxUvdhR!ve6|>eWWqw&w)=D1%M$5i<Rx*YcdJFP9<<GT|(#1
z)Qpv!?-`ABu_6&?r8CzJ<LK4bl?f#YUX$hIn_?+d`$UO#jQ!gX4%tvDU@d|;goNI_
zKfVAN$bm$|lWU9|+YYfE2D+QLZnTxSE@+nYD>y0`kL)KcE1n$0Z#5Xcq=DO!H3O$O
zIVaJ2S>8+?(5%}TFR-EQIM3?_m3ro3R@JF#_>~e;THsaWO>~aIN?Ds>qakuA?ITs&
zR^x1G%~R4ZRxSJx+gDR7{ed8XWk<VT`j)Nc(r4oE00UWR3k+eLGZjNrl&f(boHdbL
zpe&}se3^D>gy2E%x9*yl|H2bgW7RB#hgv))Va+_+#n=-&t%7Iph-z9nrI4_1Ru{R+
z=>gimMhkpw*F?K)$LuJ~H{5j;qqz{(X;02SaB93TH9ogOM=*Nk%dar+j`>})dy`)B
z-G-#O*sR6&050t<xMra%$cbaCwauuAgY8pmVQtrkl?q3a9`);3!(;a3vtT-Y7^KWR
zWeeA?BGdsLwKNs?pE+GAdQbYb2Z_v@%6(&LFo$a$O<X|Xq2Sb~N-yEWb1>}eZ5Wkn
zkNx0y4xOv<21+>ws2yQzFLHYKaq%9~G~!|;-tZ}jF5L30srJT!x|PAPGTQ-{N%Z9^
zw0am>T3PnjT>#wDpGtFMdkbB$C%x9HSc-F}rvci04sOELC4K+99O)=5Ls*N|+*rb-
zAILa?6v`;L+WGQsQW)%_Vl$uAqtnKL)ZWxNC5x}&<AP+p?0Yc{u@23p&-S7(pQ@+V
z5O@#cG;;R*okl*CZ`WWXQ2&xI$8uzVN3LHN=F*+pOuvSJU}9y~Yorv!+-HjRr7N#X
ziU{?4X*95@XKl3X_Yd}ip(qScRNRG$80p<#>ebx+Jmcnt<`0+hoe}ljcO$Gg2ffpP
z1o9o{s2hhN$a`O*8y>wi5kA=|S+fXYO{Buxq&I%Bog4JX5^3yN#HiMetX_B5PapgO
zi=~6y@oTrfHEC{Il{|~s<xyUmmia^M9?tnp_9n^%Y)W0v<rz7}4)9GshNE;hW;KEl
zl%pwpzRRY+sElUk@ozN?F&dJbge5E<7s9ekb)j3)gvB&En(vL(?%ZS8W(tx449#Xx
zLj+&?LLg<Ncf_0mPI{?*^Xw;vjSih-9qg~<u|MDv;tq&TauuF?Q{}`Swcw(o_SzwC
z4fY2eIs527%#Rv(gg&lbp&HfZp59u7z?dx@+6uM#vcXFdOSZavO?3GcAB`HD^;M92
z?P2B0aNaPD64ePfjud*^-JpTm&Si1A+bk|_?*u)OcD`o5j9dA2S$vWPdLN`bNjD>9
z`E02<*m`)|_eD5GZK+$2<DepNsL?91TGv~>lv5#6ySzOkJ^F-xkS^I%lCz@qeU!_2
ztubc`YSCA6+%~yzt`B!H-dVdy{jP>8R$fnym4UAuGM|6eS|6vNcaeRVS_;<4p*UD)
z!rFP%!(_|jEGNw$9;T{e4cx-p113{{97oUM7RoTd1b>iE&*_v6&>K$=GQ$dr%PlK0
zNAQoEuV0n4;VMT@bQD@^z6rI7Il)N8P|&`aG8Kwp!XJc#bkA2p%OljflP|r}6tPYh
zE$oUhazoIw`fs4=I1du39htF3Z*!$H+MxQz$c-bsxPGJd7<8Mo;k)|EIg_b@2zX$%
zjS8SY#rSUfx4rAmtgVQty*kvXcd@?ZFn&AK*pCiO?CG*u%7MW}j7+OCZeX~H?}72;
z()uUON8yl-Zg*FgS6TYXd_F!*-U}7({0RO%1}E(4Zg$=>y9vnR#aC5X&Q*;Z*Qp7?
zc*=&`@VH7tR#Xji(1SWC{2~JNrtZF=YW~F&CtK8s6<;lXL>zSuGdLL#7ly;56Yy)L
z8bJ}wCTS%ALW&bO#YC*Eqt-aBL_;f{rpjT-lc0tce;L-OSY0}-FLLasn<aSTymXg0
zD`gKTq<J|Jhnbu;j0OJTX*-dgVtUg`eGgjNf9^-=ok$^=(B}<D@U#=sY>mZ9sqluW
zVx32~^Uo0vI5uyNlDU(gzy$U459{&xX#~^QTLvWf6s*GMgM#h#fz>@zH@M)VutJ8)
z#3RUskSBYR4KtIlqQZ&WTfi0SN5f0SVsKeTw1SL%R*1p&2r|Td4tAaGiR8Us<L`<R
zpX0`!%vxN0)ak=bl|GAdl$1^zOR9X=1ica@mp{$bGCwUIOxE4nUp`DXya?Nsf-PS+
z7g8*a9uc9B^%wz1#u)t{tM;?>Y~(b^Iky&8LJZuF@t0vZy^zmcb!j<=yh5*=Yb*9a
z(7NW5)8So7K;E-*rE7HCmj&$lJq-{3oyYMUOnKI0+0O4O3DyJK?Ii|JP)bL&eXrC-
zk7~S2HZBF-uYRnR>z(FwEBmV1N2maV=uOkW^E6(|b}0bUs?VGlZxh*?(?l1MDuejT
zPh-Iq4VHQPS#E1HBW8y>pSIg*uR%?BDGTicvZe)wh67_EEj4d0>npSVXgIA7nxITK
zhb+D!5amBS={H4r1MNcNxp#y@mh+3?_8IGJZG>r*r`O|{R3qDKPGi#caw~^#AvHJy
zWL4jX*-@_G?3a7(`SW_hj@xoOQL|@5YN%UPnP1nvsHHjWA!jSEG{>VyRw?{r6*4c>
z!?ivVTy64Ytfhyq{=8M0`s=C^G5d?)*v?7&DJEZp7rk7R492fy*`(jj>P4k{Z=5_r
zarf6##ox(0lq<~=UTD=9*f0AsGTB33LJ7h3jf98VDR}Jp2T7A?mE+>E4J6=f6Yuh)
z_AL`6OnT)b9_;tHk_}^<(@(e~!@~z``Rj9SFgMF!T+?o>)S4cH?V<>Mm>>9NEQ-~;
z9%`r!v@*A<i~Bv@p^NH7$KIltGMFA*F$ZB@Jucbj%qM-uL>OTV-W-_BCX1ujcES67
zo!9DO?CpKc#lrG(WcSGNJW>prHOYZnkMZn<P4%qRKy@*q;DSSj07`#%E%uauzbg-i
ze`dJqTIP@Bsu$z)eT*p@nGsb;!8kx~ICIBl>Oby>wVVuiulRa4_*FD*<IleEd>kp4
z-2PQRr#9<ZX>?T##*&F^32Mu$O1|XvV)$bTQY^VpFRWQ{Q?@oy3Qm$+51)k-?35Pk
z9@8JXIi_9h@qM(g@(^QL>u@bmalBCvF0C*9%ZX-*jdNw$k9+z~d@c{KH1vwkjm2ld
zOlSxNIDyHJw5-kqU)=)4Fx+Y=*X(%JJ>4-Jml}%4Jv$QLSIqW_9%OeX<WZ4^_%J|;
z1DSH-R=KuTbs1xFd;-P0#r^6EG4P*+g0i2qR|!zQO-Kfk;x;}-;TUI>-<NW}cF@U&
z8@|x+>a6qiWVQL>DOGNYA;}UVoYlDvgTt-7SG4O?Lf~BB!nxhzD4dz_CBL2~j2?cL
z$$ZFERPJ3<{pc31RIeHpH+OPVlrfo9bA8GDa^q3&+*e+w@lea1VMw*v><jOhH5UmA
zqvjT`f6RE^WA>k|Te7L>TOa;nSPsDO8@uoJ(RxajGSH_EYid2zwl+Io=t%9FEhM?+
z*^l1+V(MbcC>+B4bxqZ|SuvrIYtsr|le>Z#<N{ZEjr(DdYCnwlFD$JqXN>Cmpo@2i
zzmuq$;o039LFLB$rK-h<9)W9mX#5&@7l*r4YU@_(-wW=HDNBDN`k_>A@+|6|tNQxu
zCZ3`A_-i4!*{8h_N)P$>XQ8|CgmF@WxW;fh2sadaXGp;f`Wh;mq-6s57UEkE>LUqI
zgQKsGyR7_=(_)1qg6&Ip@D(j){KK}FkYv*M{H){m*9A7Wg#Hyu8=cz%%=Csz#*+r|
zr)Sjm>XTcKR$XS6n`dQ#2$-pH#41U9w-BhQr@;^C4tR1m<4KO-ZS=MIrJk!9p&Lu|
z<x;-sGO2IhNG`L66K`z4pNrcF_$}Rn*ZGa=TtsRQhP0qXn{-4)iCz>+N7w?dh_uFi
zRUV1OiLSRCGL#@5k9Wb94z2;E67^^2_^n#!WDL!WNVxy5utH6VkWwe@0y%HA%JrwX
z{WNXd8M|qSxYU{h8o9IFtR!4ff9~vjM}<qjCY?BT*zmG6Gv`VFptjC;?_tMYYA+{#
zQ|hBpmjgGoo+s_`(G&jW<<mDp57b?van3SG(Svoy9IVFq$|fr9?-cbP5nVSZqOKT2
z4;RBs{yE^1#(KV{wP{>PBHyc+&tRrsaa@UO#F_KapcpPCHj^5cM@$@5dqwVMlZ?W$
zLa3D&#9eaZ0aJq*3O_TGAjG8Xwb%ToOD5Bca;$WfCSLTN)dA0q?V}b=rPr`f4WAz`
zd_w!x*+`r$*FV41x`UFy^=sc05>X1;Ueq{e+TKI(m-iU7B)kE;E5?M4<KO#f)(QA4
z`$LFDMkXk~F$bih!Z;oIrNng5n|o`=+J;{xT}isr1}M{Q<hNTiwdCqKzYDRzv^47(
z2L?XHvnq++YB~2#OnM`@wa<zz>RX2H=C$qc+uu}-d!}APEnL`F1`aRkAO?n04*!Mi
z;VZGiA`RnfJyeC+%3WqbZkzEWeB^SU7GbiGki`<619fl_#CpX~y}i6Q71+*FiDeX~
zo+va-V4d~J^;`>)oU871)*{Z&=C(p_Ds)XkrXfks<(_9T%=8DV@z6bHW-bxAN4KZ+
zqqB=TiW!|yh3OeX?a~-TU!dP;6Zor{-9G^sn%?8lvP0}zfIu-7`Xa>@g{F!Y<+CIH
z+H=C9ZaQUZOA=(jepJV;6(8!tjmZP&?Vm)ft&40TTEUf{Izci3*8|lg@#f3pQG@#I
zs~}27;0R+_6(SKfIeU(dTpC|>u*1Dedst9%PO_OZ=x=<#5^PKKhO73LT5G7exFipw
zopTz47^~~u`eQqeC)Fz>qin5BiW)uqfnK7DF`8T#|5`G{xwWt1a7?xr|3KD>RW<sm
zCDqT}>`{M|B{xdsRd*5S4LiJq2V~A<G!&W<8Zl*2Bntrk!$Pvqe;w8TBT^CA8Ct^d
z@cbX6ijjbkgY`d(XpF3EjEw(x_@4kPMm7!xrvDw*!_bRbSUa0I63~lU8#tQ?n;6*{
zo51k#!8kcPni$x?xNkPMnkZ{;wX#NX5AX^c+|&->26lB}+}zw?Z3kM}!~c6g(hiyu
zd+9yR^_lhY)2mW#sYJZ4nqxfGG!}l4bPgO95JF+huR*~@;U#7l5LKP%UrmTeko_lH
z+Su#|?w6h$ABPbWCA-o$+&2bpMq{*RI)Jd}$N=rd;`Ax@6=q;^97q5jma&$>;maNc
z5_0pejz|h<?8<>g@n77hCqQ&|E@Wo0|FX|o&(v7Yl=mry0=?eGrJ0`2`B~O4DEd@7
zmup4h6`1TFhdH}2Hvwj-XN3@7m|6rq_OHbkg#{v=N7pwuwAHtP0m=kPmRSa-k}Qmr
z1acUuqk523e!>q)$-chwM*s|os7q*wWTO)hQIyfZFesFOhgVTh0^U{F8ou}KL}BBV
ze(nJ0kH~N2I;zSV@)|<A$q@&0yg=zdbYuvz=zg0&80-&-rv83PpW>`8K$~AHKmryw
z$Hrqb{reXel6EIoXTpfi3`K2CKT&y&9U;X1XlMH1Z|z*^>RDf;(M>h?VcD`LWPTWL
zO%y<wu*4`@VoCfGhW&jxZ1{@4BtGjoKWXb>NV0yGngCn`dlts`IJOgrSiirM<5N>W
z)^Ie|Invhf)-ik_IdIw8xWFHN8+^TFqo{r;EWl9Q+#Fx$grD?jLx9$I3>U)vu^bC6
zFHh|DpVvstbuMntFQF_zw+(TAO;Jcu($OB83{WZhDw*Fp_m2iM?AE^yR-j{|VWH%t
zGZTK0nG`wfm;6}j?!?S_-5LKgn22#ZaCY<+5E-2+(9u|Y<R%s@sEswBI``mrg52<-
z5IK2%5cd!u!-`rfzpCaG(2}LOj+OPJu)DqyD>$+`xVs^&c<fimFKrKK&B)*Y{tZm_
z0))a&>=ep>{gau>Z(7QKf{5A8d+qNEqy%auhGGY+loSV>T~-tj1>HX}ura*#m=_BW
zwwJ~MGJ+*8ixUK_)7bl|q&2gGgS7v!`7ykwUHXaoKh(uYJ~v3k1hlT+02!euaX@h9
zFO3Pv{{WiieyY;an_61317e{6`bmB4IY6+XuigNV-}R6!$^6<CCxLOT;i>^sW(O3e
z2d0KJb_d4BKW+7XTckN~e5QKytJ9muMhZVn6<-^&tN2cz4ln9ooSi_<($!ahzn{M)
zh^e(u#KeFu_bkZpnZ7U%5U&CA;0)oDR^nBb^Iv%x@6^)Et5D|F238RIrbdvzIXHf|
z4>9_}fiPoZ_a~hjmBz?y|74;9f%eeOpA&=YpI#lqGk|mTe{0CN$^r$A<1g)nUlwfv
zh++}*wgE(R2tYvteG6f8cYeqy#3=p)$N2H^{rdNaZ^+i)h;IucCwu?sEB*SPh;Iro
zZ;0b?`t_fX#^4MbU!aYkzq5aU`8l`%*i3%PuAi_!Wz7$Mv8jEo@JD?uKY7fr^wr15
zK+Q}FO$<NoyI;h@tN*^ipiI=g(QyXf6@H%I_tJYDgfU>Z1~88P3#8LyvB5XK(C9th
zKO!aMJctpy>ecTU>)Yq+yJd$j^ic0@ZUdBo(SeyUP)ipG>mE#_lr1#*BOF~iMKIA}
zAx~BRp-O5*h{i9=a%>2FVuz6JUGZ%1Dos!qDmh_)^<Lto?YmYMJ=oy8bs<ri)5EHo
zpx$bq_%i5KLfM{0EJc@;K*)8Xt~x3GLY#_RuPEJxtcbNY1t_1%lV*Bl_xbk>&wZ?8
z&Yx<X)y`DNb4`}9<=`9rn2Xln=&gV@mKETOr=M^<Y?FE7c4^+KJp4dTDsYpvosd>Y
zES0Cf9u@KMm(qy8m`kanK4Y$usS3EBxQ>OpdMdOvI0s<B?$cKGGpX+#O#N7~d%uNH
z9cn35naTri#x(}MkJ%I5Ix>t;UVgs*@~J45u)dAKJ>%~`mkTD;(Oca;7jb4q#{pmU
zJqq<s-@F;b{sNyT2Jo?tT5J`^8g25CJqW-0TV4~v;i{^cT$8V5lX~+R96pPc`b%_@
z$|FFkN}L~>3N(*gJ+f&-)1$F?IJXrm^X8nTnu1$V1pu=nNGhAbTYJMuY|ANWrvK?~
zQXzXA8`AdLx%JGAiTyoK8@#@c|6Tds-PZHz`Qj^}tjDs8KEC|>x0Skw-a)>XMY6(?
zXw>lX!E#1YFfR<b>?9YZLPFm8g0aG{Z3fs1ek>V`3po+Df&34gb*Lu<$+8sWHf2p#
z*g{_Mp|f5|2<Mh5uOdThBEQR-P%%<L&2_L#t7)1rGY33bq<Z@Ilb`E~KtD)SKG$&F
z@}8qabN3|!)5wA#SJ$n4&HXJ{IFkpp%)Qj#TYve#S=E>&N)BedOZq6~PENykrcAH(
zS=fWgBEC4FyeXw}I~9YQ+B!{Fd0B?6msTz2!yYSzJBBNk51A~4<uVIskjzwf{>h43
zLpqcxhbCetDy+JbA#BW;Qe=D|D+gI(fiWad<!$R4&?w*jAY%$fJJ^NooJ*r@83fY4
zcr<SAFUDq2&sWo#SC(E8VZOv_NQ}~zHx&o`oX?i9Yfb_y^k1LGK{S$8pJneOBxst3
zp-JW%#o^TkUT?QuY!rP~9zdTYYyW+-CJ$M=|D{Ul$-*g#>*NRx)aQpx?RO+G*))&V
z4z|S=43y?^OMh9)<z6y2@sAc~PTcb?&{e7FT*4l;y&@oZmu-Y{zAj?wi}qqON^WdS
zae%kFFmrO9P*rndVy=7gq`i3ABNF%v=$O%ctf0Q9c;1^nV&5CDc)B3z0Sf~Dnnj@g
zNf%p)miM7T#A7Pxvi!o3SGPLXsrs<Js_2~ckTDP>BHY1qQas{#jc!k}K8`9#AD2ue
zFd$^<=hQK^6^H|clPGL*=BhigCmM7no-64{;Zf^Zox@D%{crJ*o@*+?9y9gsJg=@F
z)FvcXcJ3Siss?WHuF@D-dwH&!`<iFTDuieFGA>d$ozylpG~mgX@8?CJN2ye&GX1i*
z$4NeL*r&`x7#G8KmVIE)&)zb(Wr@BVtDnaKapoTUL%u?CBsV};%VaLCqC&F$%Uw56
zL$a*0WN^&IN68+fA<n?yrm?qM3a+PSwGZM9>mi7!mQ$s>q3Z4TqSnrXTLeM&1FQ=`
zAT!MLxGJe2Sb3`A@@&p4eF~bjHM_{^tECyY?`!u&X8EQ_5Q|#99rj&}^~u7<QTqnP
zdaj!_Q@i=gg(zivlwj<%+BGnO#B2G!^sPIA24m(ov{~llC(a#h>bpGONN^}p!Qu!_
z))?{H=<N6j8QnS)YIP|T8U+p(9LbN-PPfXS<Yxem%{7g2E}pnGYXLpF8*p$q&ZsTY
zxs%Ax>P}stoN&7)af!^#CEVQt@T{{MC!*)J4RJ#W!P_N-2?)qyoobY}-7_uBDWc8c
zfFk{o?5I5vn#3W_4ND@q#o}Mj%~NtG2J_e9OF|I@H+}Y225<W5&lUr;0gPK(km4d|
zftCh7mJHD|kH_G~!{cJN#El4-NGHx5Jh~SF26c|7;arMA8!L{&sGxm%!j&-EG<Vn{
z{`{P^Y%6NA7{ABZGk6J4_2d?xn>vxV4AGz)8Jsdn=EGYY8HRuNXQgTt-M4r9<Xew*
z9CBW4wX6f~qF82VttZ{ul*Tzk+45eog9de$0x`s_7I-AjGS#~;O&#${(60o%Gec+6
z(h^5pGp^Efo`}XzYx=_+i*tC(FL|S>&$4>;l{+)Z?r+D4KSSxe%azIrt?~)NCw79#
zmcJ}Q2FE}6;i;iA&zB)LCqfF~%wqc_U4CCM^`%}jfeMwu;=&%hu@gIVo7&R+>OiP=
zI*PRqbgIQ}K0C+#V}5)>z<{(TxR28IeaLYHtoN7ei191V!`Ec08ddxcUnXPP$3W~A
zHJU!|09<u=>2gYrD6w95>=VZpnVT7gZyMSH7w$2-dF%UX)rVHj+ZQ|zG6apUY{Ox-
z-vyj&wsenRInxEJ9i;WmH$veNZgmgh5euQ2K+TiZ_$U_o*wiGt24-hvJ~d|yS1ByI
z`6%XLw<D>Nhi5gHg4MY09f5B5-^Q6hjX#V|y6b`oY37D+Br)3rw6z96z~8BJrl339
zNG}p(CfFK*9Lh4>itVJ~4LkLAr9gk~YsJg@lazF12qGUW%yha7vyBTFePR~>Ci$ml
zGAy1dl`wo&;TPsgQ9%9=W9JYn4AVsEYumPMyw|pE+qP}nwr$(CZCmptliB@?NqX5y
zCzW)ks`}J()PEn1Mo=;NT6h8*;1vcLM&sD46{$lDui#D-oY<H^oR#JhDf?;kiu9>D
zY<hfL>TV<%UVJ2qnjczJ23U>uqIg7I%nbkI{#ud)C5>}l_PVL_?blzgi$$srNssD_
zzxSTxz=Y^Vra=!mDmku#Vhel{51w(@-oINg?X(%|fTWDJ_Zet{KORw-xl*kJXJzJm
zijv_PtA9MgmrN&+TG$6^vUf89iG;ncD~hLaO;U%kyqdc)1-4lT45$<IlyKpGuH;xA
z%rC$!U~J{4iSb#fayS^cuRAlmtF_!YEuCwxmyu149$r_5e#9?`p<*=%4#%W7y88U>
zkN{!PWaw>NdWX+q!npC2V;}}MXDwAgHLj>~%X!ijmDqq^FCnILhtSg(Dt^&iGw#-Q
znm01Lhe`5o*!5dBPqw1s)z%i4h5IB<^|=k1Il7Bmj1JI=a%0{oweTCNUdA{8>X-F~
zMTTE@QdLanzJb$tx{{MX-ih(8w&N71@Y`rC5}WeWDcCZEGwd9bdQ+bF9n%jk5BUk!
zEK1@-=GkvZx7Yeu3{In9XdFUDKE%1*a&5N7ohLjKhdNL>vd{skn$lX_UdP&@%kn<>
zCl>DnP2vTHFg6lrh{DPWiQ&^O`!;aqkidsElLM^G3N0++<MZa;0oZ<hV25}Y#Z3Mi
zK%<7x1l(|Ya`1#;hs^=df>iu%K~Q5MD5eB#Xp@Gl6pPY4l}*ZQ?_%bq!#`~Vc2wfJ
z2Rd4{YF&a>d!GS&;8Jg{2Q})ri58WIXI6>v%7Hy!#DtC63HUVO6R4EvP}OqAyIu=*
zsg!7NrMQiNm}gaxMJtb>#5V!4mMbsV+fIM&<3`+S2D#ykjL;~rfDoRw4RtN9@%T<)
z@m|=pqem&AgV`@fz8ZRe1ia3R*DBxQE9mO99tg7n2(#V0`~>b79>^gbb}Rf*<qB~5
zh34f>ei5c=nAoH1{y^>+AeGKF=xk(A)$Yi}N^Nydtp(1T$q1AEsJ!H<8$~z)v5t`q
zh`0ceXaFfCoo07yfxnvv*<Z3j0Ty{Mam%SO$691WhSKaaZ*HYNvg8C{R*o*0-bZ}D
z`Semqn%Y9;zG9z4BMo-GrYIL*f^R9xfM+&*J**PF9acdu*!#BJf+`rW?^XF4)o47b
znbw0tX4Q?n=Q9ijWwlWDRvOsL(|SQx$vqJ9tv6~S!xtk;(wHPxe-8Dw_?C?A5~ddI
z;}PpFH4&0^K}!&?J|Ucoy&Fz*R72s(`7$9&_7KY8XBWIDb^a<jCs*FZZjjjRQkiKP
zG=qM49qj1Q%>+-0&yCHz;QZ?vFy=!ubxd;s)`c2Q9~Xqg{Cam4Hf`n#W;NL&)euCY
zS;^c72cb`Psg0B=eKVJ5qOM~O6cN!z6iL9It>9n@%inSdrZ=J3nG9HG?FsW<gOZO&
zAt1|6<f9~P`+AwMqjCt>)9h^{`wtSiz?vswP~0EYsV9Isp1t~J#cU@M_h6GC?`_+)
zU8^_gNv#&g!)YjF111>x^a;|ufK2KeqUhj<L+|L?Btu-x8HWpCrNzMoE_)Pz%5x}3
zKdh>!f{i9hcAS_^R3rTFRRsOZylUS`eh$11GPLUorq)`~=(e=DtTF|Q;he`Ja!Mwp
z4PzVA^jP^t{&MtM%SpWnVek;FOAD-hq`sX~OD5Pno6uX%)oF1tpedEfg*X+zjl7(|
zh=G0VP4q5A>ddX7BahH1_c?i4W^)5Ei$6W%@ws=TP;xskvfrsw#YnecJJ{}2F%^!v
z-ICJleK2KS_!DZ6cwS8SahKs9@BDcE${u2fckS+&)D5YdEg>;z-Dgrh$gBdr0cxR=
z!mE?|bp2qE2&M&zyDw&ES_YJe+FQPK79VVa1cqqFi}ZN~73aMoE$v*32&eMgE_JI!
zmvnW3-S^$fVJ*oWe+(`(RfRd^OP9N3&%$0m1cDsqOfW`z=t2gBr`j&cg-w}!{@s!*
z;#Hy`=^qq3yxeJn2W^^pO?fqHKo*;$WZZ*fc`}9y`&})9XA$B}=%Uu0yhhVO`S6hl
zlPzQ%lQiwQTac;V-0pDB4fnOjjrp5zM<q^}Z4qkyNdJL=Ar#?Wq$60U*j*dcP$47>
zb$ZDFnD2lCjDRBxs5HE0rNq&(KIeSGxlP(kcnAWgBO3W2FHG5dfXe%M9r4Oj(O_Of
zZRW2gzwh@iC$Z`a1}2v~5$JD*1e_g?6mc>F-}!NjK=Cx0r=qO`VdvuKQ(H9%ogO*|
zT=-r$=~51wD!21@-JVR3E~%M#p~`79X*}WxtOr9GOD)%PIi8UETl!NQWEorS^7dPd
za^OF!QalpFwjfC3hz5L(nNUJ$Nb>%pmvLrp2?!S^2w1irOZA9PyBIo%0m(Zit0Ol&
z-Q57L?d_X|)-9aZD~K{o#d-rT6R$+dK@u6vlr42o<kN=-r>WyPLLH<d2qORv1Z}ry
z+#-P*wW)n#UAuGPkGvWw@|@rj4M@d$7YGF+p6e>|o!<$k8ohI-2T3O~@UE%+R)Y&O
z&EnAG>mG2}bL+5RNfrE|;$U43ojLRD4Qm4+XL=cxDS3MjXO~d$y1{{AvQ4aB(FUwu
z^0I~v&Ju589M!{=+?7ZPyKh;vKsmLzL__}ClmjyVM=7SVi5T|5KNA3J>shtW6l=d)
z*;1k;DSi5~`Emn|lHvCa%_W<drVha2mo-i(1bx+sc!PGo*8AQ?&opcb)Pdp7eSyY3
zKwT4lSxfpf3g1ITt(}m2tC2X$;PojEADLOxEg|>RJ2A6HJxvuksvV{tVC$WA(?0@9
zfri5;U?6c-5sb%hs+<L5JfZRrinq`3?)6KbeSQDIf{oSE_Tji_*)$ER)LQ)YO>MZl
zmJ(EnLhp-eP9Ha@=yE)6aN9%Qfa}`!Ezv^X@TG19{otWup#%>6xmrLBMFveZ3ZG@{
znhyNogEu>c(mkp3?i|>wHpy>@472)h8w36jDc6@l*O6ilbRncU*yU-IeJbb;6VHAZ
z-{_%b!J__!)xk01)sU9Qii^$<^^dg91Q&2RGQ;INBHc$<Ua<VS*=95q0c-tJHaPru
z|LIJJE>W<?c)543`lo#f9c$fqZ=~<%u6F<7n^Mc`;j-4Pk@<c!A6vNJ;-Il%nC4`K
zZ7P<^Kk5|gFg!za7c}3}G6riXjzD>@Blx#EDCoB;BU(@b`JRmLttazd+g(T%gZ!q*
zCiwXJrFu?Cw?vnlOK%K@JXJxg1$tk#^4(PmO?%=-Mw6w?3H~*h50qew7opaHi&9GY
z%#97I=+bY+AM0z;!{zzVL1dsh>dx3m0iuwcQyncdsje12Da%o4;NF`W#?fU`>s}cz
z)3OTLMU#yoEXvi=>tSrtQezGz<Ub>M;t1jvG1Kq~lfZwb`vkeRj}N$?$LI<hT-Bjx
zq6dTlJrj{%B&OKCV1v9bJGG|sJY}4fhf{74RCv?soYKPUb(B~2ho_4i2&XzSHH5Ou
z<U)pLT@q6u%go0|U)N2U>S%ba&nb>O71VcYmj~DXKH);Y5tOA|5h&RmC+RZTA>-lQ
z_S|z4y%9{s_L=wg4EpJix`_D!bj@*RbEA~i$J7)1wB;1&^=U0$$B|<}A<8=xZh6ob
z*AKIEsg&8^5I<kgsiIe~J}+!5e{S6Z`-@3=A&+_()PVMGk{PSbryBs&Ms<_YNJ0>D
zlACl%laS`rIF;(a1Tm~v>t#SCZ>}g1ieFO1R?=4yqH49Y?DZ2Y;v8O-Ffh7}p4TN0
zxoKSj-JWlfsVRwidm|jJ(mI8X=9-VwgP(uMof(oS0su=``bn-F3gs-nJg8}Ynwn_Z
zJ%<$RF`Q30e~St9Ky3CfvX=t?VQ+FULYBp~X)ElvSM}&FlVDiTIV#%dZ?3hYo6oma
z#aL>~`aKt*`{e+43Y^N)ysjs%dM_b=NnB{MZq8<qkNVus7x-AAG{#q&v_S*0DFpqd
zEXuE*Aah0tLeG)!BRNNin#{I$a;eS=Sql0d$xa5Nghvg}K_aUttK0$2TnB?rTsQrK
zki@cxoR$y<O`+2_y39BfFV?$ud9%cw<iKWV^ALvY2kd%diFnLnJ=GOCS75K>A)x3c
zxW51D)I-$)^JJ6Z?YgsgXGm4tmd`PxG9^*c$>ZpZ3u0()L69fK$ULo)<=_%CU1rT<
ze1C4MeJOb{?Xh<Ctp&N2jT7X607Z_Q)I@X2=jglOZqaDikD8g;A+BLdfjW)R*caTQ
zGW_!C&V{hz4@%?$mx7~$lQO4V?HJ<^)leLU0{L~;ctrU=n4NV~Fkx;SD~_dhnEE|a
z(BwZ5$9swocT`!l7r+RosldkC)(7F;gk#BmJw7c(%JlB5t!9Cj&>Y6pLmbm-g|#si
z`o^FwcpKiAQ4e}tLbjKan`JEDw}|B4%8-*7-0i9H!MIYrB_AF>Vrr>~{MVJh;1w?~
z+uz6Kyz}CXEr9Dc-Z4Zt>sSQkP9$kYn2MiA3*cQ7xP^eHz6dM#<%^B;Q*HeuoIdh=
zcd0}`!9LB&SMA5u8|ea2yH<%iNX`k80zU4Nws{`>IR%78+c%UIK!>6n*@^fll`Z{Y
zFM#oMBMAsx*XizkUfQ`zOvSIF5gHU$_<;CgzhY@IwZ7Ve))<r_!onS>6DjK`)`=2#
zg401=Gls?hVv*-5aw&Rv`A3k2@)be{33@+#O2=0_nlPx&_c7-_6N6ka^`N|pnD)(P
z1+f&?(aDy0;D0G+I72Ajn~bHYx3)}3n$7woV1#HzV9A4U<v>UGaKu>W14S4j$#6Ij
zM#P{Jas28|SBm&n2tZhp&*swed#vsFXD!3f4o_a{4J8kKadM{e;Dx-LjwJz+6I9ku
zy+^FF9enB{#8^#gpLYG$(Ytf&--Gi;<!1CLeD{@#A#W^pNQoUw<vDoX%*uWR!0ZT*
zeVdyiA}PTzY&V!|E2N?qP10p$j)jf;5aLrKBcXs)VR&#3L`br~YmYp-pjOZ1iLFpx
zL4pT035VGB(eH(Mz_Gkt2T#h+Nw4dj$-Gm3_w?lUHnl;E<Mp5PyJ8m#^A#nI#JX|T
zaY|<`^T9exG{4wum%#6Ix30748M$9`UKu?t=1f%lpzLEMMqGzf?&&Tf<q1U603-nt
z#Y;||*<jt?1wwpT279_(DLkM2AN|H5gpSc1hP@aABi%$+`QvpvpW~1<xTf!jHUo(O
zUiE+Q)2-q@^$k&S-_v{psI6a;NJ3jv!&#3-HGjaw!Y0L~;z3m0E<BZ^^}9tL5E45@
z?gF%DS$6IqN?AgCxbF4@YYOC6A6QVRE+JoA#KAq?x9hx6oBReTZJ6E@a%v;8i0v#5
z9|do=v-C=^UioDwvtB5?_<EwiVUE)s)AOo2y7adPs{2NjT20~i8QAn`f!O2?bwyZm
zyBR-S|H=)v$CcH}%4Tqc>X50kqHsSpqpn>klo8J4EVj~!H<n$AjI9C6d3!acL?Y=^
zn47jJF`4M}>GCZ>NFTs8UeU}u%){|l)W|aV6Rh=mhTes@$f0I|iwO*UV4-YKIq<72
zW*UiRz>3PeWpa>J|5fKVUa2I~t>Jq2Q{Bz%rLUjP(P*N1n*G-7wh*31{2(VtMD*t4
zS$GCPIj&f-^rJEWZE%yRdrvGpiV=s*6li6K)3vu3GY08pMIY5!3_3l6k+P}b8JmTo
z3{PK=WA(EQ>d$~U^u&*JP|vAIz@~TSJqXK4F}WlL)Sv(?Uk`rdr`<>QiDI3dA-*?`
zO;Np5_jELj+d}(yW!J^QmI0UydCt!93*(`@Cy&RBJa5`DV6Y_Liw#X7xH+tjCr=s9
z{wwUT`0$jix@u7U*N>J@-aL{2+(Gw{4%TgI#sWM)?_OoAeZfOK#}F!&FZZ{J9-&MC
z<th&LO-008fmH`90Qyc|>Ym+4R9V(xxb{6UkWqbdC7+ieqHvQNT8*_E8|@o}J5$3`
zrsg{vC-`hXyiU>!4p%Q$!PL#r)zQcD?NJ!m;i(0U!!<&312LuB+z5(ic^1JcoY*h&
zv#Q(Iq{uelh6Wj@Djk3@qh5hq5Y9Gx`yuTTWS}?sNc;16!WFFI&p$+X`JyJvDLgug
z#pBT5@_baazcR5r%3a@)+}-e0sE}A#K!xXjYI}Q*j=~OKX<EyqrOkSu>?>%Rl7>W7
zunKZhJZ`-iyfk3OSn5_<0x=*lFl7+_lh*FhpANYXgx2cBKo(&fk@VLnIKQ0oU^aS3
zhuLE$ElntGk4Z8WQH$y3)ST>M)VgrStxh(YSNpgwG%E4gKvMFV*ss>&gM32HEoRD}
zT;ZVMXs@sK1c}U=-q8FbrXC{3!!NAq3M#c2Y?KkvACp0%<Rce5o(4@v!0@&IE8ChP
z&pS^!tbuvookCPGS54AY_ajmlBHW11bD(B9j)Nj113Enm_`LQRyMei@6O9b57cV$8
zfS8lT^qE0sxq~4W`2{zFr*%D-OxN!XfJL=c&d)_El#B}cI)%#a5LM{HBTC|t5|aV`
z=<a;c>z5N@r|}&=dLo0gZ$I$l@i1O(I#2g1-!ll}j-p-2$<-s1l!3MPU#8Y9#VqJL
zJ-N2aV$T~Pu1tP-E&`!5e3(3PFZv9jD`+QYk0&thmz$FLqKS6JEyKnk|GJsl*;h$K
z%ro3W`X>t0>*%)rI>h%eeet}^b?uW!?~^z(yFwrrHwwO|V92nj4~KR(7iI81fd~_k
zPt%dFzJAJxAW-Xf_NG6JTb`22u2zzyF|Il6s*4m#HTFva1-#KGi~FP!YK)zOvJUHO
zxQviOpZR4<z6|p>krEK{FW4^C(G@p9eN5-MDWNrVc2LUyV_x!VBXs`>_^M5@53Kwr
z(7bZF?(~c!l&ArZQ>wdnTsQZp&}0-d!a(N1<R@J;o5Im9FYrXjEyk=5VXw%6g!{rQ
z450T)BE@4^pW9Gid9ik-=ZJv9FllIJiWs;~Hf4Y_2uMm#t+Nm3&$a&_sQ`un3CtXD
zOYTdMrL~kKNa8fyjpOih$8v=_U;(<g?3&adN!Ug#!iK1Br8)l|u_Q)Gp4ESv<?N9z
z7&&ZLmV;OdH3$%#I?#AY#45_rr`b5Sp#<xdw9Y1fhm|E1k^4G<Y$|V8R_Bj43lyU?
zc1*1yrj?C&7BV)tbBlI)TN~kQnY(zZ)vMl>-YYyD;}o}mG+lbG7gsH#Sc(Q9KrALk
zCmYscHUB^@@B`5G9_;PXpD1N8i;=X)b;~HO<rZ;Fx30mwZ8(T3^O8F*N(n7E2O~5Z
z*+<t~G==#QT2f&ZU6m<SDY2cex$m6$B9NysNxM$AJ$*ThJ!}0Tzmxf^ODqqgVkMF9
zciNm<yl+KYqba_!E=I%zg6^-K!6SEZk+PB4_Xc<9+e7A_E6fM$GWW>w^VP0RZ~jUe
zJ+ZEMgWed8*X=|*_nm7lTlE$HBISUs7d0n=>y|rZ)96iLII8uo0A_gLDWfxH6dE5n
z^LD7Z@W%aRqpo6i)^<aQgVmGRk2O@Y$Q2EFQ;D3Fyv*A%Xi^e*?w=s8A601g$eX^|
zb8BAetekZ;!f%YcAP@~R;k+}mR=H=n=30l?K+cTNKUYklMxwK3&H{guJI{mQ@%N-8
ziPNrsa5!1kphV~e2=t|(L<Qf`zfzDzDhm6QlxoeFIf8v=5#SF<6J*yA++ZeQ;n&VH
zBa=BAlx%2`BZ(Gl#CT{6OS(4eYrPA$IYUB5%s?_cz7j~_EwawQ_>Yz)Dw&0QF-Cj%
zh(DWTSv>R8RprTEfeUF4P03tZEDWtL*nXsGZ0+#@TyzePJXnc6J#c(8A=I`J6)HVb
z)H6b|TEc4v7FY#@F?yUVMGO~;DJHhdO?7I5qct^@l;GAy4R2VgGL*cPWN0;vLUplp
zkNlj}uH$u-kQ*?r#HZi!Tk1pSD58T>R$SLCE;rky7d}UHkTW#92wS4MbYNSMuQ-`1
zn0Sgf6=<V{^m>_C(P8NuuGJ4(VZxWxbt@-@_Z1<wSv(vLBlcKDmuSn0{2oGR;s6_c
z^ye!DRr$FiDUm-Rf<~)+BC7b{!MViMa7V4aLGz1z&TGj>gbBxTOhqETZQWwZ^xDfP
zjtfX%$$uu!n;&Wfs+Rx>kE+msh)srmENOw-qjr!AU@t&%P`5vG=uX7PdSq!=4kdMH
zjAfS(kHc+E)2J&u62v+7+WY`9Y8^yA9O=Dt3wvd{7^n#o#EG{;ZtxRs?idu&UTftU
ztm)8kTZ3z}yC?mI8$eYs8v6IvQsL4mnhluVR?WIUqjsQM%@_#E5!0Us`5p145a47q
zv+U!JUqy{nCzbhgYQ_m6{}8Rx?Wqahh%7S9QITC;#caa#nmwlMvZq@%T0xtAe8C;*
zMa}zdGha!QlA#ls_Q$s9j1NB8bWv1gY<$SD59Iyr(e!3{g(^II{E&&HfxDL15g}-^
zAeR2Q!+y>57=A#aEGDi@M-S2Qn{XwlTLB0gO!t=~xJJ$&&)C-(TYjcjllxvzOSUh_
zI^Y<h`NRws$<IR;R$>DUFwEzEXllv^%;gY((e@hXm#9zoB{kgj`*M@MNU?~yv8fJP
zFTHtEMSz&G3R0s-YyP5>&5t}l&&ug|K@4zj^2Un++MK0Cfl?ONx7)=i-Lu=(@9$cG
zZ1m3E=ncxOm#+FgfeXMDO5qLA<7Bbm*S2Y)?B!Ku3Ro@nr~RXZQPMyT@HeLXbI#{<
z;XToMHFb7N$hEy0*==}x&_)#ZvGBF{k0Y;XAc=VXgJm>bG%WDP%94}nl^*;%ETok$
z5y43PxTB9;RaKN!&>{87#m?2%IFth#&w@7X7?l$#$j$<I+Us?kU8fs0W6F+FueV+7
z18<>qp6-Zoh;mOOw*gUi!}YAh@Nd7({5EoR@-qNUm;*$RAkBa7SZYOhRvl@II|T?l
zs^IUx3w1QLSu=U+AMO?R`*gyTz&8|*D&Gibhrd}i&!2Yj-#8LBASn^*e-FbLLMp?I
zedx?A-H&IEW#_AzV0(iLonAp1y^4l+Ld6lUkh1RDp;cNQ4_#Q)@ovnqvCTw3xe8K-
z8n7zDam3tupeXab^o{y*ZM3G$Yt^x?J#ZY*nPWL*I{^cu0y~(bAgDy67UToThVC7+
zHf=pTA3IZX@u7pdq+`)RlIjzX3a?u^<%`G>G85FQBk|RL#Mwgp|9TV_A-xwbiN=T|
zd3@$<^1oPR)H9}--S5uL79wj)?O9e?*P;zFT%V)Ek|qS_aF=kpl^p>|AdOUKv?@iG
znZ@nd-OaTx-&c6aKwo9y#JMue;Z?`Z#K$s~1bam%&xh=sOCPe~3{Hdt7?)Q3JB|pB
z|4uwi2tk?bWx#}Ig$d_>L%2}hXiYLO3W0WrpE9{zLA4}1{X3Jc49=@w8&-p$-ETIS
zdEaF=Is69~sR5=-gtdn55+ll!JF4y{f|4f&k~ghS-`>FYofZH(z*+M>rj^J7%_f1*
zoa(46OZFY0((-i_&^7wi>Igr2Ec@E4p6WcYF^NcKxCS`Lz9y%$|ARl(6=c?Qfm26m
zSzf>E?oqA`3<+W=6gm<*l52b>okpo2j)?xFokxY&bb^qLmG*e4bL-aHF$9+g_ikB+
zOw1y#kn;@Ocu@`>Z?8r4`dV8MI#I;kIb;ukC!b;yY<JiWv&uKhSULx}155KGdR@|g
zFlc8Sw7Nyclcb7;pCdHgXL3T|FXBdBo>__IG7I7{GvV4f#O$_<HP`V(eP0VG+Gs;U
z+o$Q#Ekr~kS7od;Nu=-b+P1;9{z?M}FPzyfiJWq#%9CvRNB0PVUnES>RJV6PY5pZl
zU)OmL0ZkHCpXCtbms*918?kfYcS!An6>;%d+l>UN<J#k{hm&?C5zPU8cd?(hMWAk5
zYkI@X=7uiS7)V|qv5Jt2&2$4Xb@W@vixbF75<1av|Def?Ulf-Wrj#4OIX@7#=5kCD
zhb-PM&}rA_#Z}xlY#}4KFgeuL<)B!ag+0dj^pZTuM1#mTu#uHtDP02ilBQ<i(>`oR
zq^WrL(KD<pODz`OB3w_uR{%~o05L489YlL~#-3<%>Daz)aHnj7)%%6j8_edL%0xe5
z>^j)yhl<_vj$6n1Ad*}$7+WfZv;&1!0S;-2r5gmNf2E~ROYg8i$Zx>4c5MATzC+Sb
zx%Gn)rP-P1f*SJ=c%vO-I`k)BJgGz`8<RKB@LSf{#$p+>)un>yog<1ol1E>xGxWuV
zFRRvUvD(AFf3D@yO)4=&BU%)SErc0K+liqwC{*?k4~8ZNPNVPRE&IQE&I}GzTNnJk
z30MocWhge54(;@&nl2f+Ebm=88o`^*qET2YAaN!qxOrZb1*!rifPrMw`i)#)6n$Dc
zh|`*2+c4;`9nlh%v3%UOE!t)Bk;x94Xtsjgyg^`F?)^Nc+ehO_XR8<aQx9!pJd|z{
z29>*-31|8+<_(|esiGXHj}mY9?4IJeL%y@5Rr5xblwXIk{okT2n%f};W{uAD_A=$W
z4w=Q&57E+j?Gy4oKa1HC82_K7v4Z2`x%Tw<ejP=HRGn!IKC@nP8Jg&yuvptwyBDs+
zAF4@#Babr`2MON(Nf~qh{z0w+ObY^Iwb%V_4uG2;%Z>Q}Nu|T~IjxhOmaQ(rHHL{X
zNbkbZLWL5*(~~Ed>>X!G>}Ir$@KSi&ROWFSIZNp)tWLjxG?#k|pVJfG$-338>VFnT
zff4YOGZI{er7!NpYT~L=Mf{j0W`Ax*G+ldQZ=06Kvns7<VJbMIR_ML=dE4$h$|2T$
zsR%B#m{Eav2`Jok!dP|n0}W{=3y?l=fulL^0r|lE4U}j2Tdy~#SM2P3l>&?_p1r-v
zWq;X%8}E>>XBD~knuBpL`gQC+#q&pHSWHc*ObS8@uw$;~M{FY1{hI8(YC#mcfh?q)
zQB>?Vl}wlrzJD~=(BY?9#Zq|~#3ia5u-vT|GY3EFZOJwN)rIm+yhDI=(}ye{9SyB^
zrB}DTn*xPgg59xm@Xr@uBar{)Uv(~nqd(Vq0@cmAvErPA!t}fE#CuA0SOX+s)_6Y-
zY0^g&j-kG5+=lww-N{QN`;9VDefU2ps64Fev2#ynC%5`_a3{%2tBx-<9X@NB2hopu
zuDcauMu9U^7mQiJvhub69$=Q{-#YCHW-7S{U?z9G%s%4j-~PSv737?1$Zl!9)xYj&
z@MNVusmnFIV#wLKiRDGUEtQdc;<b^VWs{_j6q|3e_FNF9&#lEs4h7go22Pekr(rRh
z_m7`<QEY;NaB|cBGt7ZeS$)ek)<~1l>QJ1lmJ(hx);DrsE_N!Z8_2KR)jw}W?~otV
z7mDX~qGIxDpDI-mCYkf77Q`jCuEx{ag=ydi1*!eD#>1Q}f}n4BmdUz%$^XDnv7ps3
zU3BPw<MvoA$eFVQ_80B0!CK`z!{jsayCS8yLaE@J$2m5m1N7Fg6O0n>7sli!aOj;C
z>bt944CrcduFg?;1C$bGRzbi+QfvVmD}keLC^HM#eg{)s)Ti@7?BpI|LHPy(>rXr&
z;-J!MBJ9hoso->UGr|uTTl){--6PwCQN;qMI$H+QEE-dlcx9<3GeuDGigMCCoU5cQ
zHH_%w#A4!bE<=`QXdb$ObO)4jTaD-B(%yGe#G%#*j;KjBJ`>2}v1Ket>)1wOQ9$sj
zz|mAAG%=Ombnlp(&sfhVdi=OFh>#jcpp!T75BZg3L;rg;);UaG5I29m;<B<gFTh>u
zHT?bFk`%s|o8kC$;82oNX=gRZ@gRO-o|FLhl^X0xD}nf?8)3a0Et9HQpfkDcU7a)v
zi`Hwyj(v_L%Tkqf61bCHTiztR7gfVu?i9<33G+30%RhLCQmh8$L(2T1+h(ol0KJ4;
zoDENkXL$9IbNA!4QN}^+bQVHxdq{CYY-5*KBMN^7hU<tmgQZ_RVuMB2NUv3NeA-Jg
zDgFW$-t_p>X4NK+SOxUQ{wDFmHv0g0;vQ!UVaIYUY_}Y#0Rx!%E66~d7hm-)(?ot}
z+3rp+5FTh&@VmCujx+ncEo^OY(p_|K$!KsTP!ml={46rLDLOi1Kh7zOwcVgE75#L+
zt|dk255m*2gFok;>>O~c79pUSzm3QyPt;eT-}=}A3rnQ$VV@E__oEsJrFN6z*t7R=
zvPs#XWMN|YnbjfwnPJU53dPF2QX6N<q_4On8Jl=XetmI6P@UfSptiS_fT+$-lWvS0
zamesp+vhL8sznd3pY-B}ty}OkZ4mM@6wgxzqBr^m*Rie<5$I^+-%ley8+ai`O+KHy
zb6MxKURduBh=YXv4bI%0DT6p<;6dc46nhG5*<ixujo>fK9KZYqtxB5pXLuvuu#h6X
zv`Q7kkOfL~W3`P~29&C^yfN6bS=-ma<!KplYOkz!De8NohcMF0lG6a^xHpMG2&g`q
z5&B5b=9kpAe%c-?7yGc%l^xI|0XFidi<kRRxd2jgC2VjJ!^9jY#6vPkmXgUk{IJc#
z&tWY0dKD|kqPGZIP>b28O-DmP<A!@Wm^Qc6Kk`SuKKHL=Rw%$ncwQ;zd10QP#vhIM
z%N$Rm@XpTSA6$1`yX0CgcoQ!T9dP#{6=C0RNFRj8+-U%q^<8X%8d}-G=uATQ8D2<o
z&i-GODrR?Yf6RQAYw_&91kkW}ibv3jlaUL5qopA!9JDCs%a?7bB>YyOn#<mOe%l5C
zHwlc6Boq?mr;{czA36=L1Q$V)>k>q#c<l=o(ARH@El$ffmQ%2)xX)>9mp4jlv0_Re
zx<SbCB2Xcsve+F*x@nz@Ys;J{ff6d;TBm7KU-(XyMY3>I<7cT6pU7T2XU9vPPyWf>
zXLCIjTprX~qey}^S8s~oF}?EgD+W=-Mr90BrC+BvG}t5UvKI~Z8RsFPYyxlnd2`Tr
zNg=NDQ9h@kEHe`4p4Dnzi(~(H*OT+Gl>dGf3SRZaMU^C?ux%@3uhT(7i@Qy0cp2`E
zeFi3EU>(ub0$cOFVIH=_D?Wm&<2x#yR<MWhFz|XnWBeLy&}ht?dO3b0?OYb*vjXYB
z!L*xu=0ez?<ODm#Gcz+JEevz^2`*di1gxrOd5-s-mK?ubaJ4N>DOAYk!(gq~od6c_
z2m6m)VE*k|bU_wzb1KnK)#0u}<_~0r9j)37!&kd@a2JqmC$5hIspIV5BIBsJCQA>a
z^+{Xr&#+c$CWZx|B`uz;$;~y+I2USa(77d}0O3#YQL5D&xKCi8*V5@MY&~F&(yP1D
z5p6%zUfD!E@jPH9^ZYZsgliE=vKTe#4A9S*O4Jv^F-^Bxzng4b>h<FxstITIj5A8y
zjq(cZ>+if1Rd#a%z)ZFl@g`bE;MmCvj3_5D12=lgxMguy*_$?boFTSEZG|6J#cdFf
z!5o0~W7=Y>=8b1=RhZiy*CL|P0P6U}t0CMgIn5$ZC{DOO;+TFQQ(@0H*nmRW5K5K)
zD9F<~SBb;{H{Ea)%-^k|7K!S{^ykw#x|$_Lv?*>P@Pl56-1TDN$~#Xi9erXUFuRdK
z?Tu0#8dg`p8ZF0@M`&n?2~tog)#)#~im;;W=c@k20TG;(__N1+&d}F^F~^)l(d8^i
z^~C&l=a;QU?2((Y+Yy+3ps!NO3vY93wPU_-er%$~tJqFcg-~w!%{DQiGsG41IJJIn
z7~CJPf7%s$aY{u%JMAR??5|b#lleAAwuO$Oa3Ky!wviBHRQGAqBNBNubv!YvFzf~<
zD6Yg?0J6@8npL6dup0TEVQog543!MRx9QBb{9JRercUs3csJd-ls|v7ajDDty9Hok
z_S+VD7R{DxNi8XpfhC@HFVuN6;-N+otcSdl$)0gz$$Tm;O5cx;)BizAaxj;BiIFY)
z{ycS*K=hdxOXGgR!<z22SWLx_g5<C<oVPhYbIR#sP<i2HLr2z4x{jQ+$3Tihx#~_m
zeBjkF^S&yK(T^%oby2MrqPSw}X^bQ?N~w0eVx;WhxA;%<eVN)8_YLjeJH$HtdDhbS
ztSzAMW{_Z<96$DnL~kDuh_z36VAIxfZC1U8S*LZL=h!F0hux*Ce*Aop@aUBb!id7f
z?zoid7&_;1ROeoeIypF@*H=l(s@~Yd7)S8Qi``2>zG=@#V|H>6G;W|1u2el|`bu|P
zvFZ&O7M%9^8miWur?3cXFf`j+h`Ec%TiIxap2T>o0KcEK^SW7{ed~N$9j6;lB-s@=
zlK?5%+L6@?Q%^Ow6?<n5g9l3gyw#-_y*+Il=r9%20U0V;!9OlqDM2N2zDaWRwkYFH
z5vY&pz&&`PIlsss#jRbH9R9@@L*-yy8y77R<iiY)7vecFaj;%7iIcZu<u}7#?!9~z
z%{b@9U&yuAe}r;0rG-0)fkAFlxs8~GgNH)$Fo(rFv_OJi<=X8eFOfdbrpYE3coWT>
zg)~1py<~79!9<d26i2Gq`ud7WU}u;u)sKUz9VKo|HY>(cnbTIl4I;nB#_ChhkiWbt
z#~c|yQH}~Wv(@y=p^E6|7^=>s_*s}p8p&|zh-U`zbCA}A#xx({MRvI)xCmskK*7{T
z2%o*G5Ye1;6sLeCz0%-E*fpx>#<r)c<>YR|gI9o;^DYliX>z<BA?HlZ(BiOajekTq
zj{~mEoS(?LyTnzp%#W`AyY^ny3?x>@9L$V}z$FOajxo!NDN$j?olf@wcJ%++)8IE}
zn$;=nnudL9T+kv;Xr^3d4pfA87zI$6X~}9!?w=MH;l;DPFX(ZhAe&BW2GBl)@y>~^
zDrJ=&FYNU+CS-BXWZsA%w)_j-s72pxlOr@%QHde|Fr#)>`vZt|TSf<3l~nW({-fw}
zglP^779}_tyra`l@80>m2vQCe{xQST?@phV?t)E~bL>c;QK=eb49bqFHX~C*jQ?V%
z4|10(D$O0yZ0hw?^?5vBJUmlA1XjV(C2QTp=)5F%!lqRpw^ecE55`W8vI>e&G^Rub
z>m5--Fzj)s^(&E5<I{m(1%5g#Nl8RZXXM1IVM&(>BF>{gjW_f1j6zBaN{_Q8!x$#8
z5Cp%Hen%3+4G$NCDL;vvl;jeujpzhTuS4Yd7W8CcvEltN_4XQdldfsOJD7z~biwkh
z2Js~f<6o3IHDNH6YCUQ|oPi)#^zo#GPlSaNk9tskR!cq`SL(eE(~PhIhkv{L?A)Wn
zo}S<0m~UvQhKeHlx<<?9B<C>e=&x!Y?6`58Z?m)1+^xhHIBmV1G(H!WJepdGE%7>c
zwy*;UjevZ<Ak$`lE0Joy|K`R;?^^*>+7UtyFLl@TEv<!c@&Zz-xwINooOu#dl1vLo
zfV$O?XLek>O$<%EW~1J3*DW5H1rZyuiwNH|*$G{Zr~8$7uFfnzR=iJd>ww3|hE}Ye
zEOC8blhX=1%?_8<1kY&tQ{PhYn`x<C9Yj^VayK#9xdOMf-eAh9AmS!uE!0R*0%$GV
zy}9dpIdQsI!IK}_zFUIfgGl80`(P3ED&M)0=Df&2BjAawQg=-|t5BoRdeZ!>{|rx#
zqrA_eMt2b0D8ul1CXDSKKMUk0s!OdoglOgAS}+2>uh&cpibG$_H&R~^#=cDGqrQf$
zJv(9+xyNIhZVw??CmIKFYU1U|UNN4H!3mHtSM!Sk@_VE*uuKrXQ74lVJql&e!Nreo
z$5<8B>Co&`up!_m<NG12Qu!%bBDNSCsKsl8WFmo?Jqv#J%9noY2$FSilZhz`y&7du
z?|x_@|Ncd0dB+RORBj9s5R$;;A6~bYo|46cGMiAMv1wm#=gk!=rj1Om-MJI%MThml
zf;)x>8$U#w|9Zq&3or{CQQ(Cxp%$<j*v=zuh?C796Uvi}Lf)d0oKNhknypqy(csb$
zoRJ`3?ILy`RS%i@$a|X)hN7)@o3{ko4khw4RP|3?IPp_iYjG2*O{js{picKxrZ9Qp
zRvAIO@2CfkORHQkgw1H0)50<hh|bgd?zuuGVnf{|P-oOEp(|quo!$1MK5MobuwCA=
zk-xHSi*>Y)it<(5QZ?c@(2uf?8V&fF5bgWAAKXX|-c$`(ld#w{1GgJ1zAfwA4tFBN
z>i$bWcM*jb{gn-Pjxgop&S<+%TRX<qbS9+4nqe2EXFo@pPtP$6#i!&T%!7KI`Qlk0
zy7SjhFGH(P$9i{jj7^eMCgG~@vn39Ej3$$_fG25e8{JeAOd%}fm-A^x?c+%oYoUA!
zr(_;*4I8^6`u=YeLCS7KZuY%~aVZgI>eEBk{R**QZB$q#DszI)!f}kN!>w#s&%&|!
z0$Bx_<t>grmn#N$K5x}zj+*ag51{5u$Hv(*%eW85!w!vVzR&q7aJdQ^vs?vKl+&s<
zeCaI~yzfpntdk3>T0kfo+NF2>O3DUf4VWjD&{2V*O4Rd6WECl63Nic)Mu-KNG4w}S
zWxQ=8yO!()(!o*~$HApa$-ymEQiOzXb(0racfqP>CW*1)7ELQcmgK)2eX^<7D?6v(
zcD1=2?0h_C6Wnp+y-`dB5n9XhO}Fa0#^9s=Be`=`zNyMYpAfa1ZJ2s6+SuZI5yZ!U
z^;V=ro<i-Zx>Y=wg&jJS=I=iVZ>U$26nB#|^;F|-KuAGw!hL<kZVSEwVyy;7EJleR
zv-9jtL!L!t55aTR{$(*gOcZt@o5!z%#vEmNt*D(cQ}8^w+Ds1+kUUWy*V>j{x<@ho
zhVyw9#J^<fP#OF6$|(>mdoz*aE{0SArrZ@UW<r^bD_2SF_ujUgvQFQKOCfAcMa=q!
zo#=9=`K_Z>G#v8=hAY%iTL`MwqanSu(>A}JJWz2zhUT%}bigbX^GX67^;I3B;Z#Xz
zjZ*YEtds(Ixiek(9WxoFd-duzJ3zsinnk$L)p$Bq_@{iMF3Mbiwly+rR|G3b?zKK-
zO%P3nC5(aBDZ-BH29$8n3nmBg9+H#Ib0tEWlC0ao1c-0jb{~(_EJ0T+lI{Fv*4-Eg
zQ)6jhqm=uy|7?=7RS2;-VyekCeY8UrxgC>H{uX($Y<y1R=yRkgAp*C+bBcyBQP*Bh
zuoVt&o_2w;;<~zSP-%ND>fd~IU`+|HQU|?JSl!&L@z%A9B(9`JA)9W4%Q!g%N}L!d
zxt}>;vtz$+)+?EBL3#uWgM5L2AEsHiD1vJst;1)OEP&6=+YRcs#^d>7bzkL3A`-M-
zsAuw5Fgk`?X$W78+b~xqA=a;+&6sfT>Y<uBLFC#bb@>gXuI6`;!?&{$&~PV1G=?TM
zcj*4<J=$A&ar(uR`ZqE>O&&u6N-b(O8u-{I2Z6?JDMuoJNg^!}W7AGfJnOuj5>Yf9
z)=F-=aK4olJkG(H2_0iw{`KRV?n1n+r`E?XFQ+<i3uj;?`>KpO@(H2eqxsnh^KJQq
z`axNt2A#lJK8=<feh&c9QhQ(utg4eYTOIA}FkF9gI^KGq*i{`SZq@WO&BJFi2vl|w
zp;7Vg8Oqaf>Uu-l)w@;z!*1*#26OQiSpdPef;*|J333u59CY~}s!w2@3=mdfo~gfQ
zp4xRIE>JyDMDjaJq$Jc-ep^eT4oZsy-I%p$o|$jYvsLIp%J?Q;QQ*{jI$zu|1)IYM
zP8?>z=x^z7RBM=@Zl^H8QxM!uQcfPlbr|&B4k&1!B6}Gfrdy<poOO^x-UD%3WDPh^
z@5})$N)e~+xrqh_%GdQ-Z`A7t?B<ut?`RGB&^zRuzScy=rhZ+l9*X?t1{A71+oDy{
z3e=J3*95uFo~??3_Q7Mh=mK3YdvdZ88<P)-2hi5xmm&^*?CAfe)dInzA}fjd<0;-3
zW{%_lU82)6`xC*QQMhGH!wP{cZlJaGWAH$HDS6Aeuo(`W$h&9|9O~j(x(WCdR1~+7
zVzSld-+M@UDemo>0g!3fr@74ls(N;E)g2sKgQh0DwI0X_IDz=f%=76(KSDdt=RW?|
z@_6jn7PBNU%l8ry@Ee}m_Rj}TCbhnTd!qcS$9FsK=I;+k5uoNzFVU?f$%8x=2W{e4
z4gpye5z7_fV)G5bX=M8bI1uaiI5RkYa-R=MdiAR3lY6Dlv)zXd%@EHdAFsVK?w5o1
zvsmf7o|2>_U~<a*D@^xvmnIKo$>M*qAz1$x8-k6I<^RDTm<X7d7&-r^j{_4k0~5#p
zCmW)=Ln*mLtDZEw^=8{*t5xS>T5_w^=D&rVqFQ3pwCI=P^!0k<(3SUj7sfN%2E=SS
z9NnVJhM7>vloQ{OQxMUfQ^Wj!xeza5orM7{EddQ7IW#%-IW3I)bs4as(bB^U!~GkB
zv#QJe8^PH5wuUeWtc*Ym%)sgG?d|&G07R~@c(G`vr?UW)eHh9tEX{u=ALJnV=MQ<g
zeHpT7rmX;vefYIxtFEtXWUaG)guivjDKgRkv!`SLO3+rw{^5azrODkO{)&RSs00Rv
zR)^;Lu=37~&8-YT<C+*5>;E$30i|zn09XGU05CQ-u(ZCHGLe7uXb8#u0sAL*hsO`?
zG9$VYB7XSjH2r%bief^NmiFxYLy-HY7y4H}?qITS0MJ@RSLsIo5Z<lWTfWlbdy01U
zAhP;;`}8ASWoKnYW6@q`{rgl<MMnOD%9=U?p7>gFXns2ZT}%ClV{W3ge=RQb_Vm*G
zbj^UFwaEROBD((D0V4x2iYEKI3Wn!r^#|JI*8H9p8kH1Mv*E{>56Y35j>&<gmBqpN
z6Y>%DGw1wUCD;GA*b~sx(=yh#;rb`~BL{~zD@(Ro`ph6b?PrC>`B!6OY5+ZSKUXQf
zk+}(&j`25Sdu{la$E$`s<_9kPF*hs!Lu#OJZDw@?$iT)JVi4cv^s9x$|KV50^z+A+
z@P{w{{U^NtNB;QxpVdR3)pMWr;a}g|N1sqsm%7#z_Imnv7RdV?I~W1<`&q)rULdSB
zO*PN2+`%7@;?6N2$8Vl|aW?EPoZ$HCbsa|ZbM&7<vprIqJv*anvppLVcsiE)U(RBm
ztBRHpvX~XLR#KnkFPAs~E;8|v-%~}p7KX;gFSBdwpMAzU)}Py@ALwTqM^IsWb8sTr
z%-?U}FLv3xnui%G{8!)duNb+b>*(LpJ$z(+eGXTDRBS{HCjZ#r`19M~&&klp;OUcJ
z`8z$rl3&gVOg1j<HsGe;)^FB_+|%FDn;v=29~iNzm9g3PTrC9`_J618)csk!kJ%g_
zp1-TD_}P8$mHs__-xX%1w`XAteCswaA40ZD&rRCbj!k3dpsnCtn*rg%HqcusQqn!w
zwOl<(!S9_^F8-B67ue+<{9fLx^S1qQCztYK-Hp|ko=dTI(DgqN7GO~b?-Jb|=5s<B
z&(j9Ywadr7c4{Q<04^DtDRXURseTq{eF6sXEsY#PxI77-$Q^2)$BBMg$L(7suVRGl
z#w^1GE9yD>3gkL}K1wrkGqU|#bOYzL50^^92A3H%^1sM(luf?AF1UMNH+}~FJ)KOI
z2c(TVi=E??Y5EZ&e5X)+g<-i<Ug857CQ!?f^Dj^RYX)kYu3K^PcRuxZx`h~xSe$%P
zI>?IXV@SN=O|*UJXM_=W*HEnA5H86dI5&J2By_%rz@A~q7nnSDr-cm<<Jj;hGYW#x
zqT^+CNHyj4NMxMlntWKxCPTRc)Ar|^Hq2Y^tqy)K(gnMGiT$b{adp|pz^3&FThjx*
zC*<aTS-}PL#Q4$?xI?+rdy#VGU%@0RHUS1{ob)RXNS1tUen!~Uf$KgJ@XLLmbQgOw
zYEa^mY0>zy?!|pLJ9RMv<4<U~x1xo-1>#|w=I?pxta#5p6DspSaV56HYS}i<>W<#A
z%l@8PgVato`4%%^(;JpJj#EaYZl9od&Eg^P1w92Z$(>@6Gi4JEK9>VJ>=do^H!oVc
z0Wbn19J~A7<9=hxhG3B>?TmSP*Ovb=Y&d;A<jbBZw6NeglTrBI!NhJ%T9@Q0m?N)F
zB7;8D96DO}OO0Bf;H3=DF47>|32k~>Kr?OP|1ahlYrz=dXvbES%!1yaLI-BD$K=7G
z9D+A)M3Uaoqc_or2vmA#6>ImX-G35xEtiz6jWHiBF;ft#vw+m~pa~%lrGJgcJ%{9)
z4nVFF@+riZP4dTm;#i`=>Cu*8pmH%^6{<Ueij*JYLLb?_m$ftrbA7iyCm8fEY)5$0
zH8nBi^2SIE4YPBYB~b_8zN}4S5$(}fk4d(3HnWk7pX@wo@6Ws)dj3pakVVdMMj#R9
z!iTr}ZZWll%zU+nl<Da^z{Y?Qu*Ko}na7;H8ol}jfR-FkxiUP+m^yKF`*ElOnhOOo
z4l#%#-G;=L$!}+6ntG5J<2*|5Edk%PbpJ4@6*$vgK^U6L6=AQhpL9&IvYw?%v^(S?
zW1A4N%QDDWB8LA-SqiVFtLS<oaT{z5kI#$Z&|H~*K&%6h;GUnx2w)Zq-vn^$K$_nH
zJ87`awqF~%VtU`?&sdxcBK*b$Cq9j2P5Vy38Vj@EQ0te;1tCn?@!SFokP#k=C>Ta_
z(~f4&3oY%x3?J1dZ33;?js*d*R-uN(;voKqyS4JuCBQ`A)kUup>ZA!4oD=S(gmfU`
zqY31*Ns(ZQ6h~`6^ExLda)S#@l0jSVX?W2-<)<9fw3^8hJ^j|IIUm8<F~NRvL}t*I
z0sqk<u5T0=FQ@rg;g!DR@AvGKmMO^L<#R`4LCYz}L4o#^WQdj;qy}Y1MRCA8$KjNq
zQ$m>OsM%HN%Db7t*LiNZ#x8$XR@`I~r8C3ic9)v@I70x6M6RR*X|T`N{8Ce~WZBi5
zpmy}-VDIBAh+%sVw&p52y*NF7_a`B{-$RsmXoBo!Vu{`R5GcWU1ivO#1bhTq-X85|
z4s$BZ!LH=8ugrS56yzLjz%r#R6trq^*@jll3~R+;S&aBrworVbF#0!YsyrXt8+~sh
z>zuG#zNHzC<ax=_c<M!6xVrdo_N1_$3a`p$8=Hi-&7cb@0LtB^=4ewR+_P<BxZ980
zrMp0JC+}P2=-P_9cGjU4*6~fm!sTVSEW{Km5{gd{#Nbd1Tc-a2a3HMub$Qr*L^l8`
zQRY^km^pa7py<t?`Ccww%res&5;P^+tgf(sGoY}G=jdjY-vgbyJo1`S$1P}?&b_Ez
zL!NS%J6f5J{j?-sRA8vOzt8-!qXd`LtnvtQWA;^wnfBg2AlOfANdL59$rB`9Sl=?M
zCqMm3<T)St`<?!7)__!E!^g-bojhl|Hb|PyjMQA@l!Pl%15uLsQt-apnxsp;UqESz
zNaEbqeGxy<<u?9|0Vg${${pzt_8o-GM+Sq>t1<&80A;ZoP>3{Ov@0XEZV;5N{~0q)
z@zqpXxqt>cp$e~=7qbc~0Ufd)qF)xpH_12?T_&vP>Wtt{e*FsPO~Ydk3d}R}7ovYz
zPU~`owmA5#&-|<2xMz6SilD!Q;Y_AN=Ah@t7_uO1SqRpZ|H%T*%qVL|vP8hpdL0@S
zZ&j<|SDl-#N3yV=EwI*0v&}Wh0+#^%H~%TmBE!WJhUrd3-b(s4C3%=;)+A#9mh_W~
z`ijrUS(Ngh>+a|`0@B0_0*|@MFK<vkmkVUawUuCc5!a}A;w-cxb<nx4grSZx#DF8W
zJKXcN^0TRt6`D-PIS<zwly46|>n*yhdK{BT2Jg)x7_Q4JfA9KKNB5o!iS<%eB+KJ&
zlO(<t_f{p;GEtwzOC#<sFD)U}_ouVaU2S~ph>N$8X}^CoteV$7rrkX0G|Jo?rHHf`
z^&K{H&tADYm$SEwa{jlEC^Bb~ac~fr#%K;|Dt*v`ak;b}#;U|@2M_9=Sm;!r)xrM(
zH$ce0tx~v1$lU~!ytE$Ls9Jn6kqw;N#8a55n1>jFEzBzgW^HCKaI|mlnNw!#^?$Z1
zPN{i}oWdH!l`4JpZNOcXB<XyjKMe$^4J<=529?Nic|Cg(qg0Z0`mevEk@$~R$g^OX
zd1n>JIl)m!llPwu@Hi*giM}^w<z%^j58;E1LK6AaH3?BMWMD{s)+w<yvQP>#+O9tc
zk+BUNz+AuzC4wjcM%k_PeQK`xQ+sfHqnR)*z};IrPFM~ilT7jEQ<PdkR|RSj?d9v(
z80KL<Zf6lQNT*E2qL42e8|qAhj-b^Rz_ZY#Md}6)ln;{9bs(%lsqGCy%El*b0)HcB
z<Ng@%x3^`UbnY9C(UU=rSnEuwrBd;*vyHA|5q5?@Znim9i1Iv~I5aiX@Z1tMaVSuB
z`BM?s5R1>Fj1cD#HCbj`()rPR*VO^bQ-5g35|LlXVs%Z4>7(!t)DihIMwnw24sR(@
zNsFtkfE$F_UqiwH6ovGrQHjZM4Sk=uZDs&c-?2|EF%JI|G)Nkx2|G|ZuWRk+KEZ3+
z5zDWK-T>}gvgOeMyJss+@YRIC#YVPfH3(u;A?Vazd*)8zZ2|3s*!d$wbq#NSm72qb
z)a_h@K~2Jn?nXC=byAs_&+*#Jj8G{N!S73dND!LTdJQ?)uec&kzpQd&F@pSz9Bx;N
z9`cl&Dfm;^s4L(|?ur3ra>q@)<hB+cT=^Hskz&HKKQ+fcg#A7Ja6q96SF+-N8mZ$j
z!b4e|yb9`q95?#3wm0{GQj?k!oq(u+f#oH<f!RG+D5+${;Rkfu=9#w8uh-(2ujH?$
z_FPmarAVC2;j<;Ay&Hnmq9kn}3wm05;6HQg*fFE?pHUYfiBG<2o4Rs1C4ol_g6R-~
z2E3ruH$?ftaH~gn*35b?$8^?v51d@fzcOVV>?I7{1fj@4Y$NL~ioSHXc>D-xa2B{7
zO&=_Ohw*TA{d^ZX9-``{wqk7&^R<;k!`C>aBaH;u`d^JiyX7Fdt6mRCz5Wd0hv}4Z
z0m}`3h_6i(?Fti3V$KYvRF_OI7D3YQH>|QrAj>MYsS#AtXvTNnmi{IA2Nqtkn>kjp
z9Mv~ok>3c->rliKd*12mmyGT0*)~wC%e*LeyH$)Tf>z7IPW?&gOvJf&;_m2EVw=ok
z*ygDx5!kF(E*S_8c<L41bMk2yo(V%h;>8@wlY&YmLg5FuRvJj*KvM-XbVx`(W>?eT
zivp?+?E0!|Dz&wQ5xS2tjay2z3{Cb{82&Q`+i#YF52A%!(4g9_CZ*C=uj*(m(Pe4#
z=aUyt9Y?RP3It#(s_(@i5@IDwGuktY5e3{Wh;Epv2SNCT@uAOAEuy$xlO9b&VC`=8
zZTG9Bg9yBHTn$uAEw^G@AcgT~H`R;6I~Y2CDZ8R>9#i?nMlvmygo&JmrLZ{`n?KaU
zm1FWk@o8FUAVU2=MRQ^fjzhR4adP8ENxwxGtWh~gxC&cGkqd7Go-S==oti^TzAIrY
zkKCfD(-otV@m`nreyrN$dLmsa`GM+9=m{ZaT(!{%AQ{0H7m{^2v8to?6f(B5@B1kv
zvq2n@`Yu+gB}<_Mo9EryG%A83@W#wOfKzaL|H6D|U89qaK66?SXJxpISW)1r#pwA-
zrPN^KOz8QJkZqh0gZQMJ-ikr20#0rkmqxzWGFHJxSfFYdbe!f%PQMY)q?WkSOC>XN
zZrs+?w~6~}#>D2a5%rPZSM8^h3v+`N91TA5SqiUktT=Wr_-s@H<rcZ&2Aob=!>+B8
z2oz7to%>X@d7K8ydg8$W)$H(o)BYKGP3N@jK00U~OvdpCMT+`tT={DOop}34NlbD7
z#D(UHj&gL4))Hk<wsp|B9n`(6H~1@a;xnc<h>WoJ94#!U;3G}7y|P{?<BP6NJgD-`
zP?>6*xk3GtlDIm(Q}Ps-a|mhTlB7}&X}e(#5b@NGYAM01EO+3%r8_w`MVgB@pDV)i
zzAd4O;)IIMQNNgrKI~An5JWsVG9<}c(w%v$sIz$Bi+x*UZ9X?Q8<3=>!Eh&{THj2g
zQgg3X9=GtqE)Fm_IO8uZV{kBhz06wErsRueu~<KQJ`%T)Snq01?TZz7B62YY9S^_D
z>_`QKb3xWkediyORVj8wV}{oD4VK`nDfi=kV=LG4D5`x_9BIT}+0R;_+O@;T&(ynh
zP#dF(farI81^sr}Q%pkq&Vt`(I$HKem7<cw5G!D;A-6jNgJvEvcTj@oOrmgwbbWz}
z=*8TiErHpV`OG`YbiG7&Pdi>8+3A<p61}2C6}ncA@ks;VM2NqJu3siN5aG$dl-n>b
z+vM)?Aq4iLk2k{xK}0Q{SDR6;C0UaftLHCh2@(qf1AxN$%$g2dMZD^16`9m!ebVvw
z(O)ReA|Uv(GGFj0wB5WtuRcnW(sJ)0e&xB%a9Gx*n-r%6oKAa5H&nE<8PN#uLoTVe
z$JeC6l49|T4Re~h>cL!`^P30(=Q4i*W*_KDIRnQ};*{mORk5~P3JuzJgv|~35Zf1M
zEmG43hKdeQP!@luCU`i%|DX^Y`J?8UWo1Y^pK?_tp3!xlr5$N)|76oQm_gfhUl@N>
zS?qNNS{UbviQ*hVAY8^HeumnSLkh!m6iOqUh4ap`62-Y6P;M=mC}utYODN&9rWCHG
z2Oa$rBobI`6k>O0QH8k}lPSY$Nh7)Z#w5Q&s>l699gL$R|C#6DIZ+HKL1n*+)=n<M
zGwO6)<ce7GE}DVVG-9RQlh^FhUTTt5CpQK)Z;7QlK`13m_G`6Y$$?$W&3V}PVNDqH
z3uv%%G1c4t*ItG&l9nia%rMw873{8T6TFJ-TEg*s0#3L}SZ8XvBNrN=?{co%eP*5o
z_OeV>Xt*+;<-+}2tZu}FON#PInlwc<g(b4u)E7tgCy_gS%eH>|pEUTNdkYh!)WsfV
z29iN&UwMofftbjwX+GAe{Dm()IXVUtg~y!=Zm_JjUF{in&(V!p3MD<cz&`pPz=;T{
zQ!&=S9)@I?C23}qOoO2lKvjYt)+`i~yQK(0b(elP{8f%ydEIFN7dBWUU(^+nwN6oY
zd)mb6;BPo!kLjZ^_o7gjLH1-XSJB{D4L1)s!NLK->D@LR%=xQ6LcX;3d4tmOo_8-|
z`Xp1yzi;L-eFzc-!&eg33|d--Kj_wsKyl<?t-q<d*lY}cXC*1I+~$M&^AEyx#ZRe7
zdsN5_aKjncYN}QupjWJ$tkjh7vs_TK*i$x$@5|c*=k4t5hbNB^Iltck1vl!!P?Gl^
z;h}$uuzJ@bW2BZ?_{oB6b7{U~S8iEp{SmseL`Io;jTN`RWEY|N9SB@dYt_erSbLy_
zwG?+jb!OfdRaW@Ez;@`8a$ymw7-gA3fqh{dV2=gqrEG?kyA8*iq3flP1gE{@+tW++
zar|VLG#yZOGn{#>C32wo1SpA8sHu^!z=F^H#dDX4Fq>hAU4`desc(&ILIZkiZ1n6t
zV7`p3UfkY2!n|NLa|iLdl(@q2vKh6!c7_NeC-Wo<^mbal9}c;^qN+d>kVx4P6Y?A$
zULL9noy=+|CgYGiF{N0Akx6$Fp5I~G+H+P%d~)1v$H8O(X)#z6wl1Nw=;Q5l!;x$4
zk*Zepy1ibadDdbQyXkJO-T^Mbno=!_I|Z5|m?q_W!Ffy1Ks&Y4UvaAhB`-!BY=399
zAuU?K7L0$yCDZk#51S~Jdqt-IZcYPCf)=P&ga93a^=BDb)8>?Q0ji``*3#U+XL(I0
zEtJ`kY@LJ%^D95zv{0WqR?jC2_S(dof>R^x-E;OW8~J)?cUejScp|Ovh|pz4-%8Tu
zXcDJQzUYp71{TRaut@^y3r#ePF+O%3ZPRPuy9QB#%f)1tEIbR}-mt#nA`cf=9DWj3
zr?haAe#tNqm=F5*hU9K7GZ6+R+p7Rn$}Xe|h#8ue=K4EF0~DW>sGHz=>|K7~1mEVN
zdGTAz1m~RJ*hs1{(lq=jjUl-@4A%+>NAm#Gu9pL(l2flC6~=&8IKRH~3q#**b@8&n
zLV_C|?jEaG^<#)_fNLyY)F6cq=qfP(&{%;lyYEhJr@x7d6A2M_@bD}cjQFBhZ6x^3
zJpIJD05y`BQEsM5$730wUd>Wy(~HC>hKZp{-!J$vYspJ(_w|b(Pqx5W#BU9ZZo!b2
z+vb`ulj4@diUNBgOBz9DFvw6(U1*(SKcyOu3JJQJ`X#g5t@KVHB3gh5>Y_JWObN1Z
zVL~c|suQ19wV0RBZ;k0iQerw!VU_ABCm23|H3cA`5OwA*ZyuT&H2m&!-{W7#|9xXb
zG^{+T4Eo*s(6f<$zMqT~2qPik$AI_>L<b}wea8q4TX&s5*YC7RU21*Qh%={#Tg@C)
zNx-wqEd-2wG-CfwvY6Z|@-<dZ!BL#IhfE#z-rhA!l0);O;d_|uDdx1+VR!J@xGwG3
znXm_$q)ma0+<4SgV9{l7%p!>##&F4i&w#1k<(3<8z%Z#(5Q&&9v0crDMj#y>5?mCk
zE~$AF%%qgYU(s{=?iVD%xR)zFiUVR0C5m@<BI*n8!OqaUy12c|pcPTCZ0Q+OIlKQ`
zD2(A@+2S0<_q$fg!e_0GJ@bnCtSsBP{~G}o4rj3?>rnSi_Shq=i52URHtic2e^$%+
zY4vg;BuwJGmY4=#6$G_6-^iHDh9Wtdqi{DHnD)k*XAfWfM>gaD+Ptug^W<fb0Nb6{
z0J|zq$AQWO_(!(xnB*qXwhqkOH$>Rj`xsQ&&WJL_<*eoyq+IhW+vX>41k2QGpU(9V
z1zNwy#zhBb)@^4>V_|N5mn5LI_5rKm3Ab>qz-ibmC=WAwph1^~kPMXd1&oJj`O7ND
z6|tV+Sp|biwyuE1H<!nG!l*@?kZDjk-m0*AQ^+g5nh%d%EqfUf6(J4>i<JPfP@UKT
zxQ<H^X6ft)_T#5>REjsz>$4U?jCHxOH@WZjhkC=-;|H8xnts0|LS}b_Ux*b7jUh1Y
zJrnI%`<E_C@Q&#KHeO2wr)%eX{or&Oh3r+u1PL@jO?0PjhuL1!*Pdos55D|>V+!+L
zEAKBdZ&0*XLyncn?T&&}o(C_P%6!1MF`;xt+2@s!58d?Fxmu(cVbTJmaNwJSdzyh(
zoT&k%w$C0XA1S@<Xx5Jif-d4g)HEL2J=X(2EDXZ_{qo3{N*B8)7F8@>e~GN%3CJAq
zse!H5#uER1;s`KeD(5xBK~sjG&}s*60v?bY_QM8+g+9*CfH%qYiM}5i1LEc6i`5jt
zh*QQc_)|qQXg9#21HY82bjR5+x50k7_6R4<$DnsqI62MKgUgyvt774(AF4^8*nZ4h
zL>2jec52k1E2K4oFrJO}@ws^`%$~8KxIvH%puxUSN@H1&-Yo{0fV3s{vc8z+v2v3B
z)@12O%xT`H7b!l;+$dg<+<L9_5+<_h9P5#nzEWcCe2$KhUY}rhNF>I}n$~Oj7L%c6
zn(`tYM3QIy!E3CH|B*Y^`7m@W>t@wFZkKPiPwe!uo<lA7MB*a0J}yATtLaEc$~s3>
zFvTpT*=%a?_IF@frR0<{Wy9+h`r<CT7~ayqUN9)x^DsKL<LMk6Admt*iRIU(=+0B|
zCbW(zc=L^fh#X5Go)w#3&C^VSnV>$!9cx%D`hTH5_PJx^r;sT8z#Xkf1y{eA!DnAr
zr+6-q*w8PStv+HO-7u^;rN1(*ax6T>uNY!0j`&Uv`ku;kNeYtR=ym-wW)4o-dv1zk
zJ)U@;l!A!ntpNEGTKzH-#|1$jMU2?Ix?CW|^^J-LPU+&1HUY7@{Y>*T%sY9^8n3p!
zjcpLkKw%)T9AS*Z&$ox{q-}m5)_zBrYrkx!Ykjd5ER(jewVIgzl2ee}#|OurGMcxO
z;bR9!#9W`xDv~GY90<>~HNZ?yR}^SfW0+*ob$<}4g~Kfd@6O)}e>Y%Zq9dh`;2@a(
zyB%mq<t;Bvp6QYUE@n>YV=Q$r!JfwUz86~T0(mMB(#di1k;DBXd_Zq0c8o+(`(;fm
zEr3a66rK-9N$1B@7<0#w1yPl|%XC4m*enMTGOF7A!#>osGOI3lPK*E`42ZM!oL5(<
z==k{j1VO*D6+Dy<?Yy-j5v%Fh4*8scRgG#Gq&?A@j(!e4Lm-><InmKF^}6$GKGg_=
zH>@K1F{GrbMFU=PvMlIBME~?Kt8<_7z-pwkSf=@rM%m{OQ(+~he5E0UhG$Q}e$OuN
zvY8aCw-ZyvK3E{6-ew3(oAc@B)!J1+2{IX>f<w~`>0)xK-{*GxI(!6b5zeO#*y|>#
z)EFLjReQy(HENEDYUjAw(X8+dB0pbj*?Kz)<6S}b-QBo7h1Hm<0nFY^tc}l`x)tP-
zX=1?m^^_RYKXiLqsoN(q#B6@=A*J?UXH-yYl?~8b7J1!a*;?PtQKG;WH1T+*WxQ7x
z6YgrRH<aP!@25TV>w$kf5Lc1R{FxX*+-*BTY=9~=7H_>{bYc?%FV$P?S(^d4_I`nh
zDyMry!Fmfxj=J?p`*zY~0B8g8VsRq^$Ue?xr~%_POj%ntih+3B1{3Nc6QjDbX<?|g
zHMAB^3dV(KqREWg2zrFPTl^H*hahlOh|knxNA;RYYf3}=xHz_sYl|aZWOHzI%|#R+
zBiZUAyma<9EkaV-uEn)Os>5asjL}K3)nQ6ZmewHlA<xjI@^o$Lta{ZMVNzP6B4mG0
z|3@|MF#2@wYMq2=v>oec`IR_hNhGcFm$UC!Rv|066oTMjKWBN)nxvczvcjFN06U{f
z-d@#F)F>59Nr2L>^F*4mUQ!#`oB#FgAxa_}Lv-KnN|^f^rjF0%#lr;nW`I;+0|nYa
zHkz$i|KQ>g7zkN*nC_aDRIC`W%}H7gOHQGI(of-K-1J1nDbQc7>#Z<-O;L^>uk$I8
z0kaWOVRBz18+SSKX|Ds>NF^*1(&|9-_n7SE=W2Xjs-<dKc7sCx@bkQ2AcPaD6SVM~
zDT60?zhyD%`r+6|g4XIw8y8nE<_}^w%5oX<QkQg;BR{f?9-e>S@unw9LN>dQtAwS;
zWhghEQ>T|&JnFus0OumH_3_2Wk)S^_A`#p7M$4|TW+Z5KXhcg&9aD%iw^K}%PRz(h
z+XhQUu>T%=sS^kK%t3#v1T8bH-n`w49IU<S=3C^mrSl%Bir`D0r)6!aV*tlH^HleT
z4QTOD0(mg~qe56``cB`PKgPtPe3|uTlHI~oufeCT`UJ*RkXG@9n6b#q6;UR4)_Ulk
z9tPfB%NTga8T7i$O;PG23;2<&=>^RxGk}YTl-d;qUDed^%^6T|zuzcGoN=?B0?End
z!R~103W+wFcQdH1e~2;(Pyf3I*zYhLbHDHfU*-^YqxkVp!lNrz@y$<h#c}}?m`V@>
zK7zcTV0PncUGb>Eb3sMQo5OTc-i<M5UWUTB$s6%aXs!5>u*6M}G)2jYt^PwJ`xb24
z$e+F`zu6jH4;HJiDG=K=13zcdI>eE2-(ZV@pUZ*B_4u`<_t$a+o-Gfd&k)awgA0)I
z=(}}a;$sXpok<Hhtci-A4!g4}MK4d`b(JZqTSoz@>D3;P<tcvsdjRe`S)3f$q_Zj|
zgsq&YE>sYlnu!#I?LkuV5>&B1OpeX>+1~}+(<MM2{DuM{3^D81CV|8?c5fZG!M~b`
zaOMS%h$n)f2AV>FZ=>y3aYK>lM&Kfre6EqyOrr|r!c3lP<*`Hf1ou`ho5|^;`i{x4
z(`429xFzMU(dJ`rr=b{{b(@h&g{LpPm5zuCGdup&na<XMb+asqgHS7EWonU*RbF!{
z*{7*?0Sl7cvt5H|1Rvrq#^ZV1;^r<u`=s@se25ddEWL{Zdp_SN*bwY{LjnowWYUTA
zCeKBuj|+Fl#UP_s5wdj3TGlr*{xZv!-Z4n(!f*A;eMr<*4LL(YEj`(xwh`9VK};J4
zy`{C9J;5JktAQlPA&V`e{<xa1)v{v-dq{Upnb)>Nj<Oi!5U5Ev3%z-x3IFXwk2DQa
z5jG;2nQzQM|6OS@^R3*kept`7C^fmLK%Pyk-7#`S#<{wrGU8;w{u4x?%uNK&%G!-6
z=6;!-dzq}ffUD|mNWIXNsf%#7T4k}!7@d&~-&s)ShAW>S8w-n<wWT`@YDsJaC7X_w
zwoC=b#Z0&HuB4Al9Jg_`Xr3vJU?A}Lj}AYEg=;}&mDLF<WRQ(om|yH+VW9_W&<H^Z
zjX+~yU!6dE^FR*RO`@9ObFe?q8oclYinGvTYt3E>c%VV#ed(%x`|?KVjwKrSo0mY!
zBEp=!+X2k!#|vHarvMR-fjHa=?#xqqGY4kd7kQn0w`FDm9mJu?h-i9bcCW=k>oZW^
za1XF{(Up2H+SCJB2^ny`&v_NPqGG;FIWiorC0l$}sSGc6$(jW>rNWBV5sxIpHFmw1
zs-nV@3=GN1EHCIZ>72$>SSID&&+wvP^%!Blvj~I0kJHxSxojCF3J-G^Js%k_2b8`o
zg2as5wW?%U=w?udh(bnmr$$o&-%Y~n!NiC)YBum2D<6!iD7de)?MflNuzI<7+jnF^
zr2%Tcw5=+xqh;EwCdHP|*i{~2=;)!^%mJ~9kk{esl<t9k2eS{UVDBLHablA1_+2eZ
z5?O~)((ym<tZ8Vyeb;OL{Vl)L+`x+}^^q(hbt&dpr_k~#g(+j%;#+zIgoa2jgWE@a
zlrE*OnoStwVu2Ao!nQf#Yp4O6&E)I3j`B0w4Dwj|YxZi7`XXgGY)ooaJt!IBe)qSr
z#qAi;l@?E*)yh`d%Hza1KFsu@3JN9B4ogJ*PVW9Iv7LH`AObTe8P&Xy9zJ+q^?zS%
zE?G?OE-Yq{z^=VUGvQu;UmesA4`TjN3B>}kvARX%lpkmdqGP6@(ct}>L=KA;12tz;
z+;A4Z5dSRt#;xo709OHyXfMyzB{?wY5X*nhIP`t&VMY3|n0aSe*4MiX_VG25PHTzC
ze53QE*>CK=a#7QesNSi~y2@u2iCEHOCzQ5b5%t^{I+#>op+D7hadD~HJBh>bbe>HL
zso~mYG-A3f#9fULx0^F`gs1bOq+g+1LIp7Lw|>s;cQp8Mpv_G2Yvh*?OXiYfV;(5<
z*I9DG77mE#3%*NGQ4Fiz!K<Ob{n%EYj2Lt-qwUZBjZ7Zz+F=|PftIu09xNPh=7pnb
zSP_OC0TE7H=p%xwslg;Gu=d%XBo24{4x}ZLAF9G&r&MB{_m!Mo{l0X?usU&0nO-NX
zZLkG?F`A0s6P=aU;`sw!wihxa92`#7lb^V{8$+}^pw4U>rr^F*_(tK3Cx{-;iVREj
z+WOZ(F5^dsj_%u-Z{sopVq5g9Q9rml*0A?)9I4Ogvuasd3W^r}QTyc!ZG^pnWz9`O
zcBQe^F$^v!fn#wrpNsy7hB1;BN4b31a5(vqEX_QfluchC`8-<MY=OZDhad&5H?{WF
zRBCdxLVlVE-3pzHCA2b^iqSzWca{th1jQqF1cO>uV4b@!hzg6`*A+rs##o0AL__mJ
z#`dT^z8;;(vXU7DUQ#1IU1sXF*fA&l()3xH1@KlX(q$<kO}V@CHDLwKyqHv9v^Y*|
z1Qr^Q-10JJ0SX*b5mrJsK!X8^`OhIC7u2w$T(p5_S;c6%ej^Hj&>{9N)`YL%QYH;l
zJ0L=_u~@^rDtCp|i=7_M$&dYZS4sAg-x>U?{fnno@shH}`rEDgtQ2K+`zr>E=427|
zaxdK<i2i%l4p9ZsB?B6>B_-OCJk=^N(wTqw$AtToc*|koha}3O29w?5jzcsjB?{~y
zRI{!b2zZYtAh~7G+*T!33WjwIT&DuvTU|un>PDYV;10p9iTKxo2AW4bhL}+#ySSqM
zc;s9xQxuh9{dRkW+mCkm=vgv|(1ttq-L<DvC-gi%dTX_~m+D2v{&cv2EzHDR_r_5g
zpkUv8*l+8t7ks&ga|qNQ6_-g_Sp4`PN9uBlH6DzJZM?0Ie*aE3sQGOyn*MW)klxy}
zh<-S?2t(-Sg+v@C%W-BlbJxdGKNZPGuCui%UE1IU8-&88Sg(zSLr%krDHX`X?4VD2
z;L0j``s1qS@3!+H%<do`;_4b@X(k<6OTw};j{|E85(9?qLciGbXsAwzx5PQ*LmHW8
z4k;Pi<)3|}YSZx747JEImF!g;LLLV9>~MP0_-a;)wwv4ppOfY*D+A#qI4#gM51_WI
zT2I&kG)DyX6h-Z8#T9##C1)~FYW$jFX(T^@qx;<L&)`PSPOY$RxvWyCevJ9m2-h#a
zJnH-i;Y>^PZ4)UM=&bpvajDUXGJFpuuA#4sfUK_Amrg_4xdN0`<CCh5XcwpD0?7h_
zEF&X2b(_Yjti6W_&&Y7?2Mt<b`YeS<8|oX={OR@w%AIBjQ7?sOu)m^&(~>k&A7qQN
zh^e()H(A6@b9=o<Q}8ON&_7LdO>D`{H3Ee1hi-67%R3vYFFyGaBYOvd*JXMf#50{D
zz0I}B^S+@XEC{s1Ms_*Y>1$=oUVK>1-h~Xx<6@SL-pJlm?sDJE^9wSP`<uo6AnH7)
zEY%O_rn7OzTolM8cRa3Dlk=Bf?mBW7wykAz$cjfzaS4YwXB8eY$So6c{Im(O9|P8`
zVkwElHc#=uZi0k7HjeFty7J2D^B{1Ho<a!@0E(2vIB`b*MhJr1EKWY+Tb(7d%|JrS
zS|KUhUWRMfUCPg@xp^yryZexJ!1&+-_pR*5CtL<{60c>FTwrX`H6E!l%wo)BCFA+b
zA<P%R{KFtZEEJe%xG!7TM)AO&Gu1JPzicpyDUiBi)4BLT#%xk8IbW*bVt-YjwE9o8
z&x_Og$CJQ{a2E+*I&w`baSFGa$RBsw&_%NRc%TmbW_i-e9T@ci&W&(Ygk3nj(XL=6
z&yh0GF_Sx<TS#=$K;uMQZU%IsmJyX+BU_A=TpxDVb5p4v_<MnCsepC2X4{;2Tn`+_
z;aS}eb88wbbp5pRp?ce7z$}bSkyb5A^2lVkz3f)kENAgeJ?Nj!ibO(R<&um>z{oQ2
z++75Hrj@zCm{lBKgcxTi{+NZchk2Qqpq$kuWIBY=>ROa^VYft@c@s@Gcn*(VzTh$T
za@@uu)5N1@iF%FD?Pl4Hgk39WyC58gA`sz*k0cV56+;p;!I+TPfpacX&1~5(?RKP8
zgb|}hm`99$kJ^p7hWHDn;gds4lF5#$tW;!43b`Cd4~hlI*{Sv&OY|R>)2yr%mu}|H
zq>}J3bzrAB5=GzJpPj&4TQ|UD2jV&+{MU6_o&>GYo^>k)&TY06&8LW#Rl<HhAE?o!
zzt?UM6|U*-{%|#x{06#=5f6~iw#7ejX!@Q@?3%K$ZX`;_W3xjbR%(y0;7$)i4BXWb
zCdLJ+FX728P8Oj*r%->8?vzisQ0=$~ZR;v~?-LtIMwdaTmbV^WvK0?j!kqmk)$BKh
zr#K7)s0ZY3&DjTP$z;KFp#m$A`%2xnN0-U9&4gcLLBW%R<l4K+4OE?AtR`d>gT&!@
z)$fOB`a(vOI%?jMNOGN(R_?TcLTeXJ=E-O7{M?x){FKi&SjQ+>&!{e7RfrSyPyUM5
zcQKT%`5we^aw?`_-jNSNgqws~5$+@I8m8Uw?Bz7C-toS1s6JprmWt61pqEV0te#|m
zHlTf-x>qpIFITWVz{wu93iKtMTDQ%QOga}Inc(se4)`2=G&IQjsd$C?$SQev$E3vH
z$>CK510j=%;8;HD+n>3vWY4*b#oT59?`qap#98ob%|Ap5%{8>)Spb*%MEMp}JxgTz
z>v6g_BCC4HBA>aY6d#0KpU-kR_!<ocKL`CwAwtPV7E(E<@;C!(7UNPY#EnwWR+aSa
z{-|K}!ASj-Ztf2w&SGiTbytD$W^NWMnlfE@hv?dD)7UP84+wkqv`N@&nFgC=<#S{h
zmd##s>adhA^3qeXdx_}^8>KoZ;Kxp-sMM>rpquX%={~X!M&se<lyEGoo+X;{IPy-%
zRkcO95xE7~7l1L}J;Smt#VeIJHP+7Tw4O9Qb;4Vg^b;F|k`=WWwc>j}$1wcB7J@(I
zq*n&%Q?CzhO(6+OCoN)>F$k!oFGl7NMh)HEvZi1hcW7msT1+U}!F5pUedZfk_M!sh
zF?jXHb*nhuWMR+DbF}t0WT{{JQZRad+NU*V69T+b42NEkp!P87fT0N_h36#H!QtEb
zQ``Xh7gjalUAfN2PokaM*s^L=oSz_k7L8^aSS)2k^Ixz$$5}y23!J_`69b5&o!EE%
z(1MspQ$Nwqz!QxL&$4EYj{zSsqb^fFyDznI48pglTggb1nKvEinXmY2_;Kf1Nbxnc
zWcwV@T(=>@c%iX{p1god_caz`q7%r>Di3sd7KXo8^P8DbxEq<A-RBR<NO(y_OiXX`
z#;X$P`B9->v7ZvC&NXXp(vDgV$CTe891tu>F25Cy4L-FMY5hY?F17e!excU;2qBH!
zfYU3h!8W$Fpi()H0=K?VN#GAjqJ0Du?3JKv;hjuC8j{UiU%9Se*FTWUqjQmU-t5-Q
z3CGK};q-IV{<7%JTy6Iq;CpE`y*E|7Vwv8-AO59MM?>)t5yQ4o-J#2s2B-k8N!Pa~
zjTKd1yQmq+yJXq2W)<*Ma8LT&J{TqB@7qEzCq|9Lydr#3_)}%zRfSwsa+S|ptu@Rv
zo}};d?uD?oP?!gU9uz{IKY2Lv0IFqqEs{{}MzJVBUxZ5EE{+|%NG#Xea9MkVPv)%M
zPa32|)t$y_d<Y0PW~ly5nA?%t)|F^ZR?{6PQ(+mrn^2{^kZ0&e>qRCYDt9UdtL{4N
z&45sM)2EPWE>vlfLDA>@MT|SGI^+s4B2&Kp^3(7RRw@*6qDA}}fX=j;)7h)7<r=pq
zUEUmQQL;6!#-1`3P|v90iGpuZVGppz2D8<(ZTbB?AQp(LQj6<DTON=iU_~s=hz-oM
zG4lpUCyIsy=%(W+T7gdDsXk@cJx2f3+6)JjG8r4C<3&7NZnAU0r0A`6>^N5%u7(!I
zr4aML_m?`0h5WdD0aF`&OJw&`#$d#{n;-&xq;>%srS(}#S-<))!Iw_Tp85_`WZu&}
z(@wU}J(5$Ar?d;>37Rz1ey6jC22oKQ$HNP4|BQY3z_N@jp6FFtPW;$jXgoGF8zd|V
zIMvtBhkrQ;`x4Pl(Xm}oRuy>DFs($<CNY)dGCJJ8Ljt}Y)6QV{VzkA{fp74{Ps}~?
z%q!F3xz+e!9Q&km5f_y8M#p(lQ?*gVIO=~?>c+DsCBYik1KN+<jE(sEmQxk-5zSbj
zGZOStYJ`1UYpR>e2N^SG(f3)FRX#yGjiMSNgS5y%eCI6PxRk0fuL&lc?Ok^3otV69
zGwGo~^5faK3)%$lq(k&}pPXu$_-R<rD0=Iq?}*J`=q%%_mmYXI00_CIT!RUkfcd7Q
z$xpNjaZKo5&oT?m%1@I@x`d?;o8MDnc)>Gjc5>CJjcnAsaL6SD7s-IBOo=7rRXb3g
z%k->DwtPXQ+_GFk!surM&5SIdbs)s>Ps`}SMWJ9`?U+~&6tAjX*xm#6?}6wCFQ%?K
zO3iWYIs8c5B@<~`z)}=?y>`C60jn42pZvx-Bpf+lcj;v@1(*geZt_om{$B71$9yCN
znYh5W6eyk1B#X~m<z8cr<cy$(UA192Vd(zT>{hT<cif&n4wilS6u2t&NAv>^m>jPi
z1uyf8gBzazypVMoI#Y96i5xmC8KO+lDJ!ZI;4Z?Uk3FKz-Ws@b%`18>w@E2<iV~ny
z-GV%4K7qy4pOf)rJ1^!hPegD=LT}02WeQ?xoHO@wG>1gUqygg_IvpuSeP}dbgyE&2
zBcmTCW5dLz-$g_mdRAf^gT8X(ve$vtA32`9#l|1r7W#VND~!qlnz_k35_g`jt|8Pk
zm%}9cz3cA}6UNxjc(FoP1FeXH1k&p5_It)?Cr74xc#mIJ@~9>h>^|z6?-sW>X0?u<
z-VuQ)#gZPB9U^`}hC(=}J7CnTtngf{f#Fv=MnGCWt#s*YCvFjSQ*(>baP<94XRqQl
z&Y!JyQa5y%SyOsN1>i}-V~qlDa&i<w&8U9M>8z!ZMM2C0x@dvxb*&96V;(X~|L;Xv
zcebcXyYG9L_FuWllA=P_P4Cdge60AcDX0gMGNl7Wx}#}Clm<R|FBAtj`bgdbk@9$*
zcR|kEXEDapoH&#V887l5_n*icK^LKSUzq0!H89B6ZSc{LDG%$*gI1vJ_su<;G*2*v
zA3id275pg1$c~Ea_tftWe?V+zlH%4_uSr`O-s<grAiZ4SraFswt!xnPaWhjtZ#^PK
z(mt#xUv}CUG*U|2RcgN}{QL5K+%V&iJ|ui8#{8>6L^J5M1P62-e~#Qo)y<#DesWXQ
zdfHDz5=zk_0>!5nx9+NWo>dHr{Si75ZM9#YmU}es6~-H3IzMdsMd~!+F^`Ab^Tfz)
zj{T~64H!~uRL4+COSZ&58tc*RkESm>F32=*m9HbQWHs-96=RAh*naJDx<GMPZ4XXn
zT9=1M+jjCWzQCtGtj8V$c!6t7-E_>7o&NR;QJ|(GeGDx+j`ZP!V8*v;hg)%7pKM^l
zsbl~p_CXOL^8#s~{H7j!cQ-HlFgUJZ04`Hu8i`pdYh4q_I9Cr6j0KW(Zbde-3avtd
zm&y$5ufg0P<qjduaN0$qfeSS{JXYg`Y3InSZ%S?0(~bk`PWIp*jc7s9WAc2p={qTm
z@U;+(vM8}02et{!WQa5qBwhO-f3X?s?%&oDAy5xxVjzr9m|})PSa9F|iA!$$4o{*(
zbQ}?^Pv0I*)fnObEcaF4Dq5EF5<bw<H7FJC+NfzW-vyh?1ID}#R=ZW&s3Df{U7R>2
z#tDBE>jaI0#8bxR9ZBH}nj#Y^NN|ByHPRnAxp6Tip|(}xbylAD-N)TGY0*UA>VKDn
z^b#fO%l_n76RfN-(jvGY(*{CPOo_HYLnNfdDO4o?;YI%nxs)TRo#zGkmZfEuj%+CR
z^-e_68)pifLqa2lLpc>e5|L`!+K%oisoNkm4(uPnzNu^>tq^m&VFpVmdW&vcWu)Rm
z*uzHBu$kFkFIAhb2-Z|#Qb41YH{1b#&Smf)SF(Vqt&`5RJ1R&buV@k~ggPAdyrv}k
z-H2}Fr;VID>d&*x4V$3{>Os$1s)shC$s21ay0J<A(-0`b?KHxn-~OH6)A;?l#^k!S
zv))tV7br)U5Y2D6QtZ=&(~%lR0k1<hYKY2=?%21y-=5hNN_+e-=6Fbx!FAepD~x@4
z#i@Rq>0NGrLb2agQ>!}8Vt+`Zf7+A<oK^O@g0fN-ax(gIi_e7d0+&tQuOANy+Dv!u
zICw5DK(ywI%5#ACC(MjmDps%(-k%aKyN&F#(36I84nzTq_|ZhVD+^+-pkJw5_mfxt
z?j@!3m;r|!iKV%iTgI*_@dsvohzM9y9u5iMryLGcNUtxRE&n>7TSUR7UDwxtQPZY+
z3Y5}ZfA>y(o;*v_k3slR2`zQKgv+yTDT)roKA8~NLvGcKvcW_{VzW(s*3^GMzS}mm
zDAfE5@ptiJ?l@p~Z2cBBd7kV~gz97q6w~oXCkJ~Z(SavRE4jDb7L9*vfgHUg=rX2<
z0FhKUb3}Y*&O#ug6A_<H1~}t%tU*V!G-0W|U-u3567*BtG@UZCDLtw1bDGEaX;b9`
zi0cW*bKjYIwp^lFTK_kfkx}L|!Q<E_U*ANP3Hd3RUum}-T|A?dg`sC!(M%BN9Xwr;
z^f6Gy%fNfay-e&;uW~P%<kV$=h)y2L*GY@E{p6@mo47LeKvD(v9e393-;^uND2UY?
zW^=laAFOVj)NlxS5{j89^?r#WgwXY%)=rENpl6X$z(Qc58|^HbWFe$?N_OU%HNh%j
zRQgyJaGhsFA|ADAJZG}H7~rH!gA;t2A(k=S*!H>&tOxGkJQ>jqG)?ck@1_+{XO58t
z+qv%#;&l2;WQBOMvhouqR5S=E2S`J_m#ybC8S6h5RA{rfnLFk({I)<gWz;obH4Z7}
zPQxt`kqZg2h7j|he3M3SS#Il1E~cIGJ1Smvfobcdnk!N8{&;$v6?MFw{C4fkjKBM_
zLet7uFi;OWxj9jYkD=%XKi4=mrerw9p)|if0Cg@MAT0=Hf<m?_+M3=Q9Il%>_G=wH
z2xD%2kk80eMu5V%gJF!FUX<G#zTfFjRBJ9L&U5r3>q4hf@Ty<aZx6d2AWX;a-2?@;
zgf)p@9Z(KW1zDzrRz5e>?{*IJYn_wE5&7s(r<zfC;SuiZUgl=@>J#SJ-wiyiwRdY$
z#3Xa6zTia0!CFH`dcyY}kQ>A$)t^Q5*n-d^MO%9~uZQN?xZr~uMheu&=Ug~S4y{M}
zNu`TPJo{zh(?eWzAD?!!n-?y#-mAV9Hi9*0zQCjfjqN{Jg0N(bZ^K)VN3{8z5kw48
z>6XzCn|00gi^_p*CVoL8y=G5iYu6V`=_%_tZj-~-P2TYOOX@hI{&H&NDxO+m^{6y*
zsHw0Vq=dy2F#j%&LaRhvv%OxzX|!McfW?ED&_kTD`bpwDo3BA&5co%wCPG6dR^b}e
z!q%+)<}@vvJ}W&d%pik)vR?{_;$tG8=)nAEpd0W6yU)<}QODNUg1(>J_bDCyfRc}{
zRt3PG$`7*c*d$ckZoK2o1Wn*H(nE~rUS5oix@7p^h<;C#Fj6u6NV4O3d6Zyg!c0iR
zO?vT~-8bA5*uL&U-j;+ven%66DhS_0H<P-BF~zC(4fH4S{p@xfjDqFM+{Kj!mA!^Z
zX0N`K5+Gs3*n_cT!qAH_!VPV3Q!}Fre_`@*eO4`G7PSPCr~zM}*JX)^Dw#95+v{OM
zJ*!a$(sP`Dgj~r{bho&;4yf~88%P%&Bb{L`gz@<Te~nzd%E{=VTj4TKzH06XQWG9L
zM)2Da&QMOd(P5><PkGtL<<eeGC@#rZH#4S2g-jpF(9iFipp_N(S}rRDB@-MJ(HuIb
zB`<<<PUr(asPlln8%4duGGiRj=)M4S7Jdt2Xl>D<ywP5(2J+F3vy~D7!tULKu!yU$
zw6(S(E3&kU8oQKWKZ5LrezD2Fx#urr6huG`CWjOFwa+<)y?G_Z%JKac5$qqniX0Wh
zwoaTGWA>W3V=Ae(LQlzM%NG}EDR851!{ABNiSe>nBv3E9n&(gGc-B*PU5?|wWrM?s
z99mo2XD4GPH`6)$ba(oZogYAWT4FO)=%_|517T;`ra>`Jxkd$ugFcE1rk}4c9n@<Z
zH;ZrfeW*CI5Ssw@RhU<Yy?eJvj$O8GVN5pYU0V9^flt&Omu6a4GppM+{-%oxH6){{
z0|NW9G^2<f9WU?2cPJyjv|E~p)GCDGrL;}7#}4m1?ilUlecU_QgM2i_5&5%e4}AkE
zDn^4QgEG}9o4t-F)j8llh48-8RtS-M@6H?)cN-@Pbx*aLnZGC#ax{F8WKHq<rx^_M
zM>3owe@H=(F3)?F?@{51y3LMRHa#DDvG(ibm)A|^b&0ta#0(Q&JMyLqG&f@}s@0xU
zE6QT;V9W!jNBAgrl<UPitL7-ZcfI;1-n8lON4N1j^i8KZC5iYy7CFS&fm<kv=;mwG
z!G?45-fWf6jyuRl#1tD^(OrYQUJ|p%zIT&SZ0Ax|3Hc(uOFYMeonuXjjOh5uz~8BO
z(tvTPk&4COIf5hh+p&X%3}iJ>v#zhM=lJkYla%Mi32L~<kfyDHoWi5+w0X$_HzZRC
zydWCxnuqPU&n*ru(RvGb>SM}0u4uM$3#Mg@nS!s#97!(Nd>Pc$6<*bI@ZZ!f+CzmY
zdYy~y-FIJAgRK?t__ol3k?TB;Fz~zZN{ot=!80`vx>sr0+Bs2X@T>23hCvJIt3~}6
z`TomBnGSMAP%<Uhca<jOUomW*>e>MLCDnyypaOK#6vP+QPFyuX7cl#F2M|GM93A=z
z7py+6--@<*uMsVTlqcQ=mLx_|``4u=F(A9;{ly?VW#DmIZLntA-0fPq86!eVD+qR6
z5rz*e+brEnuKV`eKm4ll7a@gN;h;D6nk1$jESN_4p!B}n%uRvf@z+X!b=ao$_1qG!
zj(Gud?^)TsS1@(zuoJ*NC+8YcTd0{)eJdsWDB|&WoeO5EgxBtRP7XTeq@%Dj>Rb&o
z*B%>-;&M6KJw9tE^s~T*4Zp6=H0hl*3AuIulw6`%^lgyy`$=GG8FR=_#ShJ-lrS8Y
zk#L^d^TCI7n=x_cIv-wyZ3iEw2_v?Rnpl*-kRVt6Qa??Q55h$q3<pZ2`~L;>7z^im
z5oKR6JO+;3^b^@+`}R&a5p;E4{8NKi-5&b969AIrGCMT!_yMhB4OuSUZ7FL5BfR0Q
z{Jmr<`ccGmkDFKTuXoB#sX<Xq9#F!=v1qPNZr{GB^=d21NKb2BFpr3VfN`YHiZXte
z@6w<3+)9_*(#osYxy1qJ$dq^wd%-*(_aI75_dhDjy!ydLpj!Kj^j*2!?_t2~DaM8<
z#kMdB+zFpJmdE2rJgN>vv;ub+P%8+JIO++W8P*z=f1SDrXBXea$Jqp3yjANq)L<p+
za^Q=(Bh8WaT^RW1DhZ@xGmF?5$A_pq`Cf_+hpzg$atGx7r>E8|laBzf+E!$UX*F>b
z)Tt>XD7iL*gnF5PHQKnBS>+_D+lO$rwz&O6w9=OOS3wjwjz&uzQ`_GNvklf%q{s0C
z#WTn>Gn_cK%A+*p%j+PGS{m5c9-kB&##AcOXW|--0jc?6WXhNX2mg3D)@DQYMInM<
zuxNImy1q{;C7i%}1bKY3b<no}pFe+bb<hSnjI+?5)6~F+|5+sP@oV$T%_T5rQtntH
z4j>cMz7vxxx)7S9lO;KvS=I(=rQ5^S2x0{ffZoi~N{dn1B&V_)y!JU=GHd28kSBQD
zFba*P5Z5S7k{nP8aPtyA+v}|!+@Cqk?qEC~b*^)ITF|?*w@gp|b~t=6F+7~=p6HMw
zgBBJlAQycGJ8GW&#Qr4;;NK2wJCRjN6wfj!kfSN~rr4TrW_B$yAIin@@hUZYrx($U
zFqW(~#m)S5fswf-s=rD1E2Go1he^qEZb7S^va6O_EZA|vf_fmWVP8ymdMl;pz%3X?
z*Iex*1c#2q6Pq25W#iG}Pz&c18t}H;fB;=O-PLwWPK|keLh&1!J+KMmd5p||Vfss(
zB&EPDBSx`b?4l71t3ZznU$(MpAQH8X%Cyy8@`;hc%uP#sGBch|_JWzFQk>@nSDF5l
znpkMBzM*|0$QYN)LBk8Q0@_VKXu8Qbw_3jzhW3xfO34;qDT#qpXrRu7dgP1u>9}hX
z!rnPMip5|Yv`L{|HJNWwOr9sWuJ_Iwyo9D>+vMZ^Z7mwR&b1%{Ah%xkgnl^Mtn&l8
z|5YpYC>)PN+`}U=)9EqjpH56HG%4b4hu)D5l^@_Eb$<htipzS$%Bz+mbn7NfdaZN{
zE>n$uwR^_YS$av+0xjU#%o-N&7%`fuPM_~b+Qs<cyWerC0$-4kM*qaG!sHdYqGt(|
zt<29Rm(3**ux|97B&6T!0HbFhPwLZeot?m7b@2EvsBcvKgCYxxAU;rHX!OWj_|dwo
zAVdnuR%jV%)Zupq&-T!9dm<|k!Hz|$q^RIkb3PmN;E8~W2x^6mW(e~H<G~jKf^Wmx
zyzlp^8rOtMOFE4HY>%Ubo4tmYcj~|^Qqe8|Ru1BQ{U?u5;y>f;gMZ{3^0_657LC%1
z$xPnwdi7Ngu<T#8N3WIeox##iA{nP58K^1i;Zj|VHk01W`JvyD+vW$R^c<UwBLX^E
zN6P%3wxs`cF=w~q=q1~E?@~LD{A_{O7qC<0@kYUHr+aAbaTI7;{)Pg-hQy@IBr9i%
zx}U1v9ArMC3MpQ>eUT6KkXI4}a+M}vv1~SVMo$Y=tecf;6lQ=Be1%=X>5YCszWrMp
z6LDHw(ChpGH|h5P#RU0?JX9{9Ks`3LjvKq!!X*rj?x4v^x|YoQSclsn)}Wa;7LXN&
z7#I2oOYO9k*T5ueYT+~LyUNH)HJ;*7mA&|tf!!q&4<bV{NU!^K0n2WVvNz{rxtW}t
z<Q0-pNdn3!U?g9{S^JSN1yS}MUYps8Vp%AKvsS_R;lB2RJLB+!knf&O9PrH%oHZV!
z4>w17ScP%>Ix((hEkhA*ARpWeV+LtI=Nqk}?;>jwtq~5BWt9oFHwL)umdL+l5t(!o
z8AAA~#e%63c<e#5%K1IK=ov)noTfcc#j3mIkKhrvFL*~ZT>8ZUA~BG)dGn+rHx+eE
z$OG8Awb(i_Il3oPq~vOlZT_8c5w>oyge$_w!oPa``ZwNEKAq#E^`AiqGC-!xJ4o#|
zYmP!9<S1*?pMtH?KnIMPF4LX)(IAr$TvI$imFwj7#W>uB(so;eYY54N*k{lY3xhF1
z54D(B<JVWat(WtR3{n1%KDQ<HcqGD-XvfccY@rT-S$xGwDp`RS&^E>^)gf%b^&!6$
zi5WV9nD!&A@P^<{S8jaYbta=CSdgZRcFQZwf}}#s2PR-k8B!T_UBqA~8owMfUs_NL
zKup>#1|*L#NIOqQ)CH}CI2siVL3Y9>HKbIaE9J}5!J8#n>wgQz{1c=xR_j0Mz950E
zi3GOC&-=Kq{nAeb9(s)sdJN%@^3O`tmWi6&?M%`si5lc#w)<(4C-tk(G~i>xUhHn7
z{v@Kzzt#QN%pZp`res<tSB$l^c||ZxCrp~jr$7su)jA|L>6rh^Sx4h&>%wep;tq6C
zIUWuO!gVbkrCk%?0Kz_=-1=>k2+K?Dg<ll1NsX9w>dz@-JJ`&A$|n~6M1=}!uI>nw
zVJT)tCRpj+y5ch%p@cwKGf?MVP*^2lUeBvcV!}DFc~Opu*Pf&N9xb$6dq(`4MwW+t
zckZr6SCTIPBGl?CPECfPmYHNDe=tWeZXxHZrfa0c>m#1*X32yRBrgDC3tPf>2F|8J
zL6Q?xxiiM0YmVnB=a3Nj>nIjqP97wulWiJ-Cosvss(xDU?6aRWEOA=4Xh*@_-ko;u
zSXkC0XV9KxjRNlo$T~(*H#^9ErpUxh3rUfU3o!!dj-SfPUmb^+h_+hDE^kjo!F?0Z
z&4aVDpOVu8@xa#Ti%Z!D>5O-I>rJ*SHIR@(atH1wMdH(9H7$BQ22Mc~W&x;P6nW9+
zA8FWOG0&--H6Wr&0DXeTTY(fP_SRlpN=*x2wnrMw-PEBsQQPFrRq8ZVCor{#OsxrP
zIb8zqztOyi91af_9Q7v8xVt7IX5$t^$5mY1*~^{2*Neggy6ZKC+2^M!lF+_bWn$9&
zXLQo8EOmZg@+`<UXUtB-sX3;-d2$>(Z09}Jm~^s<+!S;~0%yYoVkL`?{vYs<_`nUO
za&BQ}bCsYM1wJQBItTw<;|JGC%ijL*h~o;@E~U_KH4+|oZ#hqs5MJxn)k!`fnJ~9C
zdQ`iEPb#?fl7=0)IA<q>0A)2@tWSD>nPLO?efQwifXAfI<09+s;q-iV@VXP{>*`2E
zph$m^^Scl8CDG(eXH^R?O{!q66cpMMtt)8ti!3zyl1g~QSrd39#yEi5IB>I$K>*fz
zQcC?bIxQjo7tN#063IzK{hGIb%WFz<KcmRO&dw&xT5lQb)T9m%5y|N#>CToG=N@k~
zc(uN))Hy#t<`bZdol|sXQMYYVv2EKnDyrB{Dz<Igwr$&X^2N4o+o$en?VNib{)hXx
z+uqM>w>9S&y`Q1rA?8ye&&|jdeRUaUbx#sPzo}mvzhEZon>`567D!rn2Z;s$<|_3X
zGi#?D9-hQO=VwI-Y_7x*Rkh238?|Zo{C5W)B1WT>=AaIyh&-)hP=55hFv|`|ZJ-eB
z><Rd_?;bs-gz@R9p3ED@z}I8m!`|GH_+X8c*I(I2F~|YE4AsSmKf#smwGAc(%tC+^
zJth-tU-qGw9+6F9I2HWmcnY+-QtL;BZNio&H~_uvi~ay{7+uGAQi4uU_*P2>LjSFk
zZ{l?$aBw^DW1qYu3{3}56`{FY8>fI(QO@W7Jq&i9+p^kKPl_CZxRco&&3G~DL3*dC
zR%;preyBp=s4JcW4U!q0jTu^tit8QEsx5){W@;4viseQdMhuG>!pCet19UsV61?6;
z1_76@=55vGMtkckpcESxi%$A$A7NRf>F0nIaRnlWm5+6(z)nT%C&Z{LAQJVLs`n<#
zde3ofm2i*N!X7w;+Wm_&nX#5ruV9~{hWpc2dZhkCf>Tyg{+${h-}|J1_*&wgq`0O|
zf3oQY*w#Xb8;kwsXK!4Vr`|@Z>A!)sFf17!Px(deElyp2+kgzs?}#3k^AKDG_7~5A
zOE!#~)7VXNTbzX{tVk*)D#)RY>Do=~Oq@PC<-dHlm6S)pjBqgm-!;%SCLxtKe%o`1
zqx64v{Gjhrlct>`8X{jU<#lbP$Q_hgJLx_U%WbRZh!!ueA>M6EgNa>2dwEdULiMu@
z1h>43_e%C7b_9l-?nzTZV`y<JYv?yO^L&~0J$%KU{UTLYVJyY)TmhNvXse~4sXbeB
zj|5*17-G0o+bA>HLdbQ*6#N8727;^nk9_{NeK65BYl}6#*Fl3ww9j#lb8#E?;7~S(
zA-<?Xea^4f%ir-==yauI%ObDB9B^`xu}G)JkOICqF3$CRP;^B%xe}C<S4^=9Dcdq}
zf%|rH2$+{UvM_Q@V~nc$;cuiynhIiqzxp){2?<e%0$sj(zpU!|db4;L^?}snbD2f(
zW4XJyty8!f&RG+#CNL=(Nwn}}%3QU!1FCNgl-ff&$`mqaxH*LG%yy2zl(A}K20R2I
zOuiP;cdk8_xi>x(%k@~lZb!DX$g~9%d)|s~YOAq!)ZjHR>6b2Gl8rT>;NA$v=zfRe
zNlR8|%TelB5$^fW<|^sJ9FNvtg7L%MMXewM5o*y&-+L%S6?m{lZxQ>9j~90j@J0Hr
z+>Lu8@E{_rsDMBak$_Nr%q;v(gfzp;lSHSNl107GAj17uB{$C3vw}yVv@T2Jl^;8o
zU{rQ7jJYGT)?U7T{`+b}fLArpLZ;QiUIV1-Zk`RVl439v$=tE2duqQa*AM19_u{wu
z6yBM#sad2&TfZYmWWLKAF5{io?)h+|+mGH7nY@Z3m0LzdrJdo>kte6gaN%6n=jnWR
zqEiYL6m9PQcA_gft-zv{6A)LgLPtS(Gl|_GCHk&n`&<p`lm4x>^_9FhTY4MqSsO<1
z^W0*^h_#yb$Lh?t`8Br(<s7kq%5*`OfeSTYnyY6*7Ext5=C_eDu^?5mqRXR3HB)f{
zXSYUqe&nS+N1&G5m#_QFkkpJ!4!Y#W2}A&vEol+(5IIqdstl$SvZR@hzv&f7mzHNd
z4LPn)z4*{m8EKP{RwGKD?eGu@W=wX-#J<1l8oB^AK55~j=Syjowz2!xRWdjzJgGSb
zVQS@?B&AJo_b?3vB?F?$2hBjQh`zX$C`!yEf&AzJMd`nP+^XxECA$?|jD4V<+ruf`
zwtXGK5@ez(oN8oha&N;G*YP4v3NmT{Y9vrE2v+TDzA8Xzy)?cMk?Qsmc0U3QFgoSy
z>R!D0OCU)xMm}?<c$dqGTWgEq9_SiUK#8z@yzn~ZnUt_u+9&Gj_?T^!ZmZow$*1g_
zf0Y(Y5zmQG4W7@5vP+c3{dC7#2PrZ`MD#@+J!L!US}I3<nw0QnB#GBXg<mLUGfNgt
z8?~ijjX3BQ#T(0VR;UEsZ-SOr&`Zep-q_fO|M=^^$gl*shqaumqopTpbw>%aqM8-5
z!nZnj#a+9ZENmVAdz8W~zlz%R`F0Fmt(#j=1f~stF}QRq7fP)j7z8TjN7PBr@OuwN
zoVh98YQU;9JlY$x`@0O0Hf#E?q?!a1NWjK*KnO6`k{znD7EVc(#Yek|<&Dg2jM`QL
zPrQ3U<C=fX#)lJw^`9g6+By+O5K^OtF(xk!q#!RoWRDA3;H)Fsx@b}Yl9P^Se=ef5
zX^(RbtBvHSwYl*XxZc9kqayo<M1|SkZ_l;8e)E1sX)-rjbR<TcE!UCrI5fPyHG`da
zox3t7*??7G(ap6hY4<LM##^(iR%bj-Wk6`xl3@d`_qriRBtnC&=w!6TJ~A~qoO2C(
zDI^C-Ye~E0ZY5j!%{gGP;1{-Tt50Hf55blMT%9BC!4;cx!_JLQx|eRKmC!HhtDoxV
zle;Q<|DMUq@&Y#R(Pc-!4TH*%x)ri|s<*O50SneezXSu>kvI{l!OSFPZ56aJcMNKJ
z1OS1yzUwSaNI(eu9)XzP0OB+#=B0zz3jHD%*F>`CVvri8dIBfXlUqRD@`-LO0X*=A
zFSY^KKUZIe&CUGMIjHGNS>>|u;FQz642=ThEtYs8Y92P`3Q|KbBX<bW=kVsOH*cF#
zf?>?q(mzRe66zl6l}3>WCXCF&G@@KY#h2)ec_=xGc*whpi-f(N=Y(7V%GWcGaE^^7
zo#c;61_jQnR_)*F;^gm18tuX0!uM?=?(VX+``Omhhvm)+5wNbxaacLyltrZlB1h2A
zTHdev3&wV%tFMvOHD|@|j#~ZNT^m87akCGL_v*`XGzO9%s}mw@Hx7R*O)(fQ0beX;
zgvyDwskauf&#a@od)(X6Sri)mTbve~N8e7b#oePM43Q+vn&{%;b0ZX?fo)apS36H_
zmwAr0&$(^MJ~PN+?NmDP5HJ`~pQ*Wb{Rd?eNbM<>%LnYbRLWqM%82kYf`8{~_-%(=
zS<vpvK11a`KB2<Bs#$AHnO64fYYq2y`p*{7y_b5V)c1+cd}cRm(jq^EvjS%zaL%c+
z-taqRLj0@BXSSurZ?Wos!;s=pt0Kv_T&FBtPzeE|>L`z?C`bE#{%^hL(^yDWhm{a{
ztp-Xk{`h)Y?Y9^rN4Y9&Za`jY4?`oDLHxt$5*(}$Iu`p_2P{;(?}qeBUpB!BRlts<
z)?R(k;vV~SFM6^7m_2BFy)a{*M++Iy(kfxE@Hpr=3(=FAO-Lu~y4%zFWY??oCbSQw
zit?R&pmXJr-r0M@`eu05gLm4$zTR90ygNL&G&u986wT<Rz48ED2<$;^Q_$?I=Z$ij
zV21Z>zL57ia&u6Yt)2H#9+qjYnc?<nu7;cp<S`w;3fO?uTmi0qWy8QxEHt)^>zq#o
zTer7<4*CP6Pz<>2Hdu$R3J*`x)$}3G^zT0j*{Hcitkrv<o&fM-P8hH#DyHDpAnU-j
zc0hRW&w4-q`I7_2Wi`gP-$3V=Z4N5@*Pu=+iI$eizeZx4^5AXDo1;{<dZQOvU<sp8
zX;EKH&HAls2q2jFV0b>-V7E_CZYc$Xx}?Li<jhh5%+)!e#TcOfR6`yfoIjZb^u=W1
z>)A|CN+}gvm(<-pD@@9e&iB}EG_yyWs21YSw)qEDt<R=?77k(1UI@(w7`+OHvDj}5
zqRoEB5cEDu04m4MH{=@M%6*nNRCJi+zD|+Ojd4i0AkKX~??O5#=O_;0Pre69glM`|
z&KM!@{Z*tcMGESPc#!}7iYp`4w~hj;<+C;uNS%qF48_mnQhhjoZe-VQf~wwyJ<ey?
z#G8=B;?mADh(^Q~jywv1FZ6B#QiMtA#6bwRN3<hnEFZeSDgN6+JaR@OSHh`Fh{b;0
zw?&rFvYgQG%gA0a2K_n0jG7=&um@qO9T*FHqY|W*I;0>5w_vZ(Bd3GZTb}~-Kn^^8
z_m>*%&rk>XyH?T_5+~G*Q&D?jn-scIA;onr&yP`KjGLzRG?{zi<H?`(@ECr)ZmW(S
zk^;hKgCfnUCPwP?bhDt-yF!CM|5Bs)>nb&*g|uBS(_^n@rHcWe$Kpoq9V+MG@ifbZ
zW^U6I#OHqgmx0%;XY*0{{^!ffm$M026XzR5rrSj8h~$6P&e4Z<Zv<+=+t$iBro{kc
zXr@s9A@Y5(ZPP7y{(=4R@(MU`#!q&*flFkW#FJg-EZKs|nzUUecCoBeH!p9Y77xl3
zTuI-BlX^Rb$P`tZU%J8niUBplzv`+2oVjY=g3Jv5f)E6`^f&__NQOf1??5sVyHMHU
z1pxwS2@4PgZ}Dv8;_O){Ndv+g*<ZxkEU8jYu?&y$Ff5ze9wVv~=s|Nbw&|&Wq&<v%
z_1chUk}adsF()a2lIdH}k$C3b;Eb7fveIIQ0HtdvoA1IBpZP?lFZ$tWAT$;qN$m5k
z&%dmdrr48SCT8aB&<m|Rp^gLcB?|IBFahbrCQ^7?l)Q6LL(&m^U1^eZSnj=csYm_i
zSQcj~VpcIo7j=-OoL(U{d+ot(C_JXA>#v^Ul1CmLi+!pZ*ZXBC?h={&dr?A5Tzz9D
zrwCrnWZ^Bc=E-Ori|huOL1B$kkoHEDm;h53+`-NpY4IPwEM7hMU_91|5-AyV!-_BB
z#EgMW8!OZl)*N57bDnaJ;u>pc!i#F=U1f>OP#G)&GX;et*(bO5BW8Hvloi}Br96Tz
zBfHzp{+jb3y;G2(A{};j-|e8Wt1;lifO^E}Pv#(6S2&lSTN6IR2c)|I-f);Nh=i<A
zZOo5?w@eD%b>wA7UK_p>)y<`hL{QRNt*eBv$gykwjH`u|S!bLyE$fyC&Oa|p2i1rL
zqz$TqPty(|1HN^b6cVo#@LI5|?!vWz^9+Yk=$_ix;=%}O&nqvqZ`xyw&$ugmZCB&E
zNK(+{O)MNwU+3pdMU~q9Ea3V&K!xQ^R<nm{kQ(&g-wc7J@M|(cSC6q)d{MO!fPyHe
zxT_>E;2W5+p(Y?f4|t1kE9^%3++tn0DlswsAb{qIq&eGhvET*IB_L%Jg3k9p(xy57
zBW;?6@&D4MIa%1b{^yB>@PA(raxrnS{Woix5Sl^M+{($=fsjGeO5e#?*x1n4$QYWB
z58Bbm!C2oK+HJks#aJ0<y@hR<N>K|YX#e{9TBl099>_lqs6YR&eXGH6zPLSu>Sn%x
z<|bZ5=U^f4PWwyrL$~hD7Qq-m$FaKoIH6n2temw;os2I3Sb&RiC_OnbS_m?)u((to
z96VD!Adk9AQNFIYpb-o67W9s<v^d!XID;!N5Ak1SF%MAy!3-+ymj+N<8&Fn2K){Uo
zuX%Y01d6!kXcoV{@X+}5^rM}W9eN<g*xvY3aZCU<aGvM;9yF*B-lad{2}Q(@a~?Y)
zq+QD|;3@$<LIDl!L4kkkpFYCc5|B8EKYp@uB4%xAfr98jXmMbOz#z=PRe7>Nx4IC1
zV5kOA@7mZ=SK(5~;EW)BxPIt)rHpx<V6%H1L%?=G+-re)b)`E%h_HHBM&R~7=l(3H
zKmmXKv}t`|+!{0me;^&~0{Gll`Ss<Du;tTOCI_GncVXSeD~L#5Qe{nXSWb4@_46n&
zfto$;n}RgjKIzX8o`oNRxHJ6;8U0vhx87Xo9SQr8K^=iI-G3kX5Ifodf5q<sg9ujr
zRQrMLf;O=`Hh_ZY*HL6pT<`q*I@|9@wL3i!Wc5If^)Ys(m(?#zU!V?c!%?=TGz)!N
zuDx0??&Pv}MFhzJ49w2q)40{i{uRd&jQh(RaquQv-wRpa1S;5%7br!b^5oJ&{-iVs
zDCB1=Y3<ul=o=yGm7jU{2jKQ&JNR`H^}Rv%-Mf9_t2oi_*Kc9++{Z8I*Smw|2fh0u
z1oHYd???OHi^<}NxO(6l@m^Yrf6GwbJN6y^HB1Lp?~?&lNtn5-tBNQMe+b{S2*jni
z3RE@OC%5%(xZ?j8IJ=J!^47e<7aLmd=y3NH;&~n1NFU;D%V)cnkZ|ZecRd&UNsOyA
zAg7@ri~Zr}v*Y`2$!#6@WCb5ypzlYjs1&;Sb7hYsF$r}MZ+CcnZ048Vkx#<lSpN(h
z;^DFP3%}8q(9qjWV?}x+3II<J&_B}O59Rj${qb;<c*k#MM1(FsvEx~IWd&aErP1t*
z<uiy*FDomD^6mJM3;px)X?o@_;F>r5kS<XcY?o&}KB{~oSx~jDk2N^$A{(?#7&?%#
z^49KiG|Rq|G;7?CywDjz*s^Z>2O%0Cr5E>lB{_}#eFLih(ORM}Fe~+@42#FX9{n`q
zEu8~+RR)?#$@Pi28{W5qmZ)d%V0P;-*771VRhtm`&+_C|kcZpYsn}J<MW~$Pp5KvS
zycV9&Y=AaQsAl?$Wh%Y?$9+JLXI^b<=@n+VVRjT33wX5ZY82pPkfB1@6hW!95jzs1
z2uIE5ue&p?<jUIwFAt_cYCK8@vGfEh`F^l_<n4S8q?g@%5^JP?@j4SO+qotM<aRx`
z=`xh(+Ha>F%iK77h~?B#qwzS?(Nw{F%9%f8uQ8`~12*_fWf3)n^S~(-?{9;V&#emq
zqS;-yFu3MQk`^Y1c97V|2NY>b?;m5b)*pn4Vpn*HC;E#lLn#|b9uR5~-8vba3QHy9
zAyZa~L29OW9<<V&K_UCaOQIC`8Wje%L!{Pk<F*Za`Ibl?TCQG|4EEuCF4eZJ`5cD<
zGCcL3scOS6j91=raoc)2*-k@Bgsl9h7g5HWQ9|#s#vycz=*zr^+@oXh{R<8BcOk!z
z2`?yjAz<F=jwQ;pZqalEX5(PSh>AqQj>EijT1+R+pxQk`)AN^?hAJIu%6C`S5G%Ec
zeuT?Kgl4HY*fSbPbb8`)CNMHGb_$HzgZR?Z=m>Q#u_~u^@dy>qS<h4iA}gJKi^CJd
z8w_Ge2TXG>j_9^>yk0winx);&;$QGcS`tMqD{6_xmemDg&QHF3p?kVQ(RFo&5(CA<
z%EU#HC#ux|t)t?d8lJUyB=MTb5b~qm^MN(GgxfVQi9w2UP0RXPxLp4YJz_`~izeNS
zgi@QwFwipdB4W&eYv_ts73K!ELSj+3?X{8mDKB^}(KqQg)n6)uU_%L0<uY#N7XWb&
zWD7Boe;vN|D!vt7i-vZ&8&iHMtbA3x?!0CnjHtHV7wPp<$v&NTtwUK3@ZPbg>^rF<
zyu{IhVe|2de!$;Skb`VuU@wRpbl4fG?G)3bYcp-soDv~EhHP0&7)cp_zAJy8#-P<y
zBUeTBYgc)NWpO;wOFH5zX1ot`uMJ+?rP(3E3LWniwj-M<`@8q^JYVT=nq`9V%b(qH
zStEeyXBcid#&2FdkpkWEmRduF!2ohfU43V5I;FSXy;;SUaK;>(d9ShDjQU_J6xTjb
zd^&WayW$NqQ7pOZjpJ)`E7@$-zwEMsK<s_?tHJUX)exZ^7K*a@_Ce*hmJ0U{?7@+V
z6}y*02kYZSBNORwUNq)YTL0v*<g|<&4-7i{34uRdJ!G11XdK|&iW^R@&v4G&?{z4{
zY`Y$GC|pCnYKB;Ks7(vBa+dx+${bbMbKqYd(%bfb8tf%?NF~_ctYFAq?Jc8Q5{}h%
zm*><`nO0i-ek;M68!(v%?G0x;Hghp~IeWx-FZ{L9#Qt?Q2MIL-FDkh?RXL*0A{dWN
zf0z9IjAhUm+kJzZ&d}@Z!N74TW`At*+7J@51maQyRdtFmZ4ui=fD(+$H(0mo(_^CT
z+dBd*lQwE({yk5}tUUrY?2;+~!?01!K&{qYLUk+(R*{IGXL>p0*ptS%MAx5xkb}_W
zoLJL|I+51CQ8{y#5QY6^Aj^$8q{PZJ;H}$`2Y4wtfLRH0=^LlfFVCXC$c&(V-r~$E
z5Xszg=p7%WA&@Cfd0@C+cOR>jhZT3iCT1c>7wg%<9mRpFee__shN#`SH^6&sPu@MD
z+pIe<k~7sp6&3d$FT7XJ$(}~+HcTF<8_q`6Jo{(A_&bp5#^M6FEY#_a$cvQmdYqDJ
zubC4r4Ez^PNkTl3hK|`@8UdT$N&1~IM{cXxg(v$ylnShRBk&3(kuIBuZ^Z&rasH^Q
z^2+6Slv*FRNzwHWBcExQguo-+ZBU;e&)Dd~n4A^J(4VQH&_xtr1x`Y+WOiWTw-*a@
zOf(&?HY+YDNPW*~x#{(3a+#@sBkgru2@vnKNqNPX*>Yqd9soiYy`#&{M{rH^lzdv@
z$-*f9R6o2mX#^=dbAyexGo&6b7pQ+ep;$O#Vozfq%#LQqFJ8XeXj5$2B)Jx0ZaWR>
z%Q%?P8iVe<yL(S)`F%ifo<JKj^H*|En(m#HBuR|t-}Z@91E@}}b>+lgFSqS!%&i2e
zDKT(mY~v~zTJf2V$>*GArs>FQg9e=S!g!@FM9RZib^QdC62x8jBn$wzeL^f8jcmZE
zX4rJY8ItBz7m(@sVy}nO-l51tUe0Y5AV~E(eXzx7eEzGYId$Tg>|j}=G2~i`&j>?1
zRhSpn0UB9F+f6OPat}RjP*H}q)D#vg!?4$*arcd`IR&qFJ?4e6xij<M&b`TcAa_n!
zG^P@M1KhjPIi#`^X>0(NSe+(}CP9n=7-LVNY`tg;)O+A@`vd_zUmyUwjoMkCAIcKC
z$9q2Yq9YZE$CpHJy@!JQ9&WdVHCp}B0sYm8(<P@)m<7|B8uw$;W7lF(^}|2on-Tr&
z`6P?7`FvFRzPN7kY8pwq5)<PiXJjswXi6RN@nmo`aK>D!F122HM_Avkc>AH#7*A7l
zS}CDZ4d+GfIl;j5@HeToRc2*N&Lwc5k|2cuKC=IL&zL9z{xSj}`H>3;<vtfSwe?1V
zZ(|Aw7wPCuZ0<lB^eiYMHmIC?I_eL1Xl0rVW87CvDn!yb>-^{eZ;!j;OZ?xgKQFtL
z6NUq3O2!qZ>Gp6X(xnw|U{~_gVsxK_n~6Of(6z>-AH80FUQ{VBoURhCSr2^bLoBU6
zJ!Lb$sbwO<W3NEJ;$k>UmU%XL=Bms0>Uu1BzAKv7H^}`_g}@xc<G~1eIG;-^)Tzv0
zL;l>m94I`ab7V!+Z7{IIy_V~%S744Ewql+j*g<2^11Y1bm`Nct1qpa5I7lz1`Ht&G
zbh-37ieXKp2aeqTjo(Dt(`iQn!T*e2Edu3&z0hQ_h{MJRD0}B@zz6H<48Y&G=#~#j
zJ9+d#!PRSV@(>JpNf$1pDk1)^RuwL<+;jm%EueKT94pccY?dfX9%)?;>S)9xZiK|;
zC2WBB!zsC;ih&HiMuO>#Go6G6e#ga-+<HHG(~jnW@If64BbQzBNBm+obG7U|zh76B
zuxo8tW6f&@9M5{YNpwj}ZKmE!i2fGgSEz5jM^fD#+8dq-iJjm@?!i64fIOsx)+^&y
z^0$s1!M!oUoC1hm3L@Pm;sCE-%q@=cwL%0-Fx)aW@Jcz8d@!2I)J(F{wQj^@8-(1G
zx1h|%K(F<GT(4x#Grt^dxV=(jw1QrZoM2#)Fhx&pUOr)~YjXr-l)E<2Ksk?!o}v9I
zwbo~$ZX$lbxqH0i^4rcH==joV)awR1$FI_OQIwe0#l68Sci*U@l`l-b#rJP5k|f^+
z;Aq%tHf7%+sdw*3(UTD1P{Jra0wDDuZ6)_C;|&~P;LVDAK<T7`&dwfg806nesid3%
zL!nKVqS!`K4muee5ENmXbg(>7!!Oc_P-5wBU#FwG_FW;tuPb~6H1#I@000YidbNtB
z4S}fi_av=VLM%Z2wf<aYfaE-b$Mbn=7{p9pq**dXhId#%Ae}I282$o{Da_-+4R5g3
zV2;w^I<fe^6}Q8iBS*VhLT3Fet8T;ltVIHpwS!NxG(_v&Kj*P^Mim&~$_CZm;bbx>
zH_V59j|jJXhtVUBTD4)*p|d5zc`w5A%xzA*@in0L<<JLc24r#69-C%3=0$4N?;4dp
zuf1wkxFrA$>H=;KT%W_FrX1Z|AwV-qjTmxOaI{{?E=WOB<UEQJlLZ3KT3loqxTR2%
zyK#f*M()BtNp9>Z+iurvp_nd;(;h*f#8pf=I;WrA>ZcNTsR#Ax8<dF>U8F6de>YBi
z-uir)E!JI|7@}M+EC1GD@@xlDfToZJwaQGY8BKvpwRR<VAqs`Dk2}pHevps0toaz5
zaZrNxML0cV%7&klHg8F}GSy>vt#RcSkY|>)Wi}gk5w}gg!57+Ub+lHb9wJ_BlV-p|
zcxK<3TLT^(#xkYxu-x$C@swl&>!iCIfj|+A;=%p_vCl3jLtxmvxS}*}Cqo8I#2&;Q
z2_WBUxx?xcqc+K7sH!dEnPk2Bq14?scvo&7xd4>QClHj}h)4b&mqFud#3xeZkYlIU
zGqEL?G<ckvm5?UgvZ$!ph0?0~Q?z}Zv<`gJc$lE&cybrr?z5qMM4u&ANcK#0(3)Ri
zy`D&E#NeM!J;@YU)K-vE%{}mO&qH}eCcUImyHVFpR&<!&x08sc;f<1qEITpwRNwh(
z9q0Bp)7XYXW!=(k9fue4X)T}*`;N8mhAk_$HFCB@di=QiBp*_Mc?M@>ivY(Wo&ir2
zlCU8coO{9o(1TX9A?Iy1hbNoA=#y3~m7#r&TDj?o$0Qou53o2m_*Q&K=VsN6jdHm{
z*#hi(`W7pS;@}QjDjys}P_21ei|l)Hz)LptT7qPp$Svd6K5g#QTH_KmqUzFm2E6AM
z(bY7lnQ;%fx!DE|i42+_kX9JU`nI+#Q|+i0Mo>brzT{H8xbg9W2f3l1Cjg!dcV<Wj
zfo0XBciF_j^yg>;Hk5KVBsL^-G&V_~3c{|H@{&mEn11|GzG%sAF=gtFS*Gy81-r*!
zxK_~cc+@cPTuNH&tk39I#XYsOTvOWO{SyUXic$<up|BA+jce_!+I@dOrOF$VBvniY
z>?h%5d#{&^CImZY7-4CpcNVJtG0p%csej=f(kOW>|84~f-{0wg4Ge#H4q4=Sg}C2H
z!Pfrbys=T$*GYh-Oa;k|sDOm%B`;gdz~)i1<)o=Gmog};*ojgj+R^_V=ez1DA4Z~t
z;N?U+$+k6g>b!=cKu4_{PKZRChSt_OK55<Vxg~8dQZ#PSVaH@PjaYsBO^n$)!^Mk~
zK0(<hWjw)@l%s5cWXO%=o_(S7vX@q2@GJGxF=7dg%Il(hUk2?U{O_$m<r`iJCbOra
z7*yoR6|Y(9%s2qzg|xAA?aqOWHM;{T|J>A`R=H}?jqSqf;V+$h##T(J>@q}wpIlAv
zlq3qg-|N*&(ADLQ^k8o#!kqI7_Ma3)mGhG!94HfN6)5GEu03U?zqFPGwCXlEB)_Mt
z+OAGD`8XIn1eiI@E)MkW_GgllB~T?@CXqFSza<fShaQ3372u4~A-<|1#uhVQXgAC+
zg9CYrZ_CSmm2Mc-*B35srIJtYx<_13O5kF(u~0P66}C?3`o4>oI<{`c+Meb>`&wxp
zrG%rgLFcv_izi{WpEI88X0xN8#*gEwx7%LODl3ISD<c;<vFSLOXic6K@xY^NH7d0M
z8diJUh!lc9RoCr!7#g3SoTBtw^^w*hVdiffY0h~d3T|*Bwlc!?k6U%C%jQB2X{2VR
z4sIFIU^Uvy8*yf*<0gH*SlQg9TGD>DHlN>$+)Tw05QNeeY9Tx2?Y&!xwa06~2MQuy
zqB%mxM02Saj1`#$mW{;=bf7IxG6}YQtoB*zCiAZ$vzUlVO~=M7bpG6Z5M%igY%`$R
z@KVW}Xwh%)oh_)*e@B^SXx`g5Iot9GAJA-<&`|P3l*mKWCc_%i8|8iIa8Dlvl)2<D
z2|;Z2vpHXiS2pr(KO)s*3a1b3V-f}e;@^`WCR;t!YXV!J#L5O<Ty_88=`x@9L;aYY
zq+*zCTWa2AuCcw!n$>=6wcsj!O*}`GVZ_km7gspiH9)uZO(}j_e%RyA7qkJuvwUF!
zJU;gXxK$DPmrIb0y4caDkZgdp{an_yCQcVj5{yP-;iB;*0tB?jji`QAS-Yrd$s!b!
zUVj7CvqVh1CsFqeSK)(=K0t?MmqVr)*g0B)h0tmSs+sV1^gWx6kZrZCta)seLQvI*
zF#sBGQlyfgsZc-n?V4s!(qq1kuuicy8l+vwZ~s*I6t~ikP&+|(PLL5r9U3*Sf><w+
zXW@;4Dx&a<5#*){2)EAbHc|G>CdB#&vVn!KTTtls&hrYaMBpGw{WHRVBX>ez+P2+7
z)S8U^CP^R$>V|=eDrxjH5DUq%+BDc*iW^Y6!i!**^qyC&mByhlgDs~*0pE+QT#!#A
zBG<^oYk!%Ha;=<QZxoprLa3VyWBqW~%{jP-5pjrkX!XLD<t1!#LD-yesEClsHJ&b+
zQ(2ViQy8nH!T1<@d82hPE1E~MBc;-qX^bzuNzCVHcVeu#llr@C+K9(ja2FM+<2k&5
z7i51fX+Y^8@{z0|bi~uvd1y4K{SHzB%c|elEZb`0fqukoVg9eyou@SNocSQ7dN`Ma
z`gP4=AHvLSEgG?(wi4DLd=tHmxN?(vY-%mV_Jo?0BtjcL+?!xMju(xP8O7Zs>NDvz
zNf2E$B#$$TE&nb!FJ||9HEAR3D8l^0pC;A8S>1d!Ui1AvwEtT^8!*EnrBhQJZKeb4
zt-w|*DhD!8X>Fa^B1n2h@V98uy0a{V?8eUW{1ck*s!<26$mJG=YBtLhZ<Z-%a_uRY
zBPhUbnXnBRq*x@>si$~I0SNxx<Wy=QEvJ>C?|LMKai*s6c?w+OBsAP<k9d1f;}Bf6
z_A?K$QD^k!5Ei4NKszPi(UDz`ZrYww*zW8=+#mk)i_wFl<xT%L=5zp11)_Wb`#<t+
z&w6C~jR@l|BXj-ZBpCki1ntA$<v;Is=<y4v%d1OY?Bt#v+7LSN?@>e5$J-!*l>6^!
z#?L$r4C=J;!Vls(8g84;piIsnLW)u&VKY5=Pxiqp2C3-IN}L2{cC39OCuQQ-=-5)v
zY~;3lxx;I_O3ixWcn?UiTH!E?ZiIJUzudj;v{7yaX)M`9AF_~3_$zRK#aP~+4I$xW
z=q1-iq3YvG1y+|~Q1dCO)u^nvqR5CnP5AIB1{zzr4&ZrKk>cM=V~_HP91giN2c$P9
zVn=`-y>VM_6h;UIDovREn<_@veUvJpEN7pE?IILzWiOh{aI0}IzNb+P%D~3>rQ0X`
z19X7KsRI!YmlXsX*cOgx?xxe_O&&O?h6yN~LBPS6M{0+^)}zbzJfStixE=tu_@r)K
zRyVhf<D;A}qAwnDe3*7HGzr_m-)4@o*#lxgiEZ!131*jsbpJ7_s#8&6z4uP4LPv1P
zKeQeUzJaN!F>WPckDk@uks<^k>Wg)XhUQFZ82B=YG4`Niq^K|7bGU+4Fo>CUwvq}4
z{IyYWe;$IaVsbguT1)I`-ALVxc7u(SZ(Tt45eO0p28U@gIbHh^k(~Zqw(F%r_8U7H
z#EjNf>~4j!o%#Ht88qG0kAlJ1y#}oQHV|J<2ApdT&=CC&og>m<qH)N^O$bDd7FA=u
z=`Ng5UaHxGY;My+T8cGAlM_kmWr3)w7i~f!)KO@})A!-`W>pNX*qJ_J*2$7UAr8CY
zOHJTPoE7R^8@M?;r*;++DWUZiY!?45{lzPjn%biy>Qeqy67{2gNNAJs#~}>Ni)X<J
z*WSH~uTs_@-h$f2TIl6q<q8YL$0*{sj+<lJ3ICe7THC3f!$=&KLmjCwJsG>0PgJyF
z0K_;_p%6TrJGB%hqNXG?{yq%tqIH4*qFq=9yT&0(=E?h`C-RPx2r($@&6Hr`=~`$Q
z>&Zs^&|6_UCSWK7)8*b=PUm3PZ_zyiHOE_k@ARBUM>{!eSnu4DH_H68oFa7|xp<_O
zHjRikC<jTd1oXbM%tZIX60><0hn|a!UU7KCc5{K$*BJ56JGB6p(Y-T&CX3sykdb8~
ze$AW>Tp(!|zKt+t2FKgDxzBpQElU8pRF%L!YDD1~1TQ!iOj<mk!-i&a1ozIur7~E~
z@IdAEn2g)~<)FgqO4!VUdDQ5Pu1fy~cQ2J0My0~`DsqX$Roubgo)V>1JsJXn@X3ye
zA*!-(o6av+OPQz+9_B>|VJ_Q7X{MpCdyW>`0@H~C**QmwG)6>IH^a>{e<)-<cckq?
z`w;@Fvn_<^Fhb4Z&aVl)3&;hgxG3k&&F7NqC`mFBCn?}ETgvl223Ajoh2^Ent;eWF
zF9-ix?ZmAUyGR8$0-ydP-qc5A^qejvBb=961``}KFwHXWk_z`&%&z@<b~JKvZ6PQZ
zvnMMxnsiYt1C`6wEI~$U`5`GOjE%Q1bci*kFpV{HCCMAP7A2972SEl>ovw>Lew+y0
zJ*`BpA^W+0l-F!@fS?t8EmhY5L4GM@jl?El7%GrV`+dSB`FNslf6r%)(^RQ#8%(D~
zq7UiAqk!#3)ux48EnE=nMWNtI&5}S1j2|LbtW@c=T<Y5Me0^6hET=^N?a7|}cTR^<
zfdlvfXWX2wVs6I<F)J8SomZXNBwgzSsdp09sDcn_0=*v{?)<niyy!<(obbTe>g6Pu
zim9CE@$>@Jeb3iD7|fg&YIs-%bn$#zrK3W;jtXy4H4pdEQ{YzYx{Q_tb<m`=(*}|S
zYipn@-#{q3RV22pU0KD(LUV!DG!T#YEfl;{Sn-LT%usz$;BLqmtx810K`oln-tN#j
zC0n`!!`ZeSz36D<B||nxzRrtnuzJ~z4;NVvpttqkbScQ)dF~Yc^=kP$;+BD|8va4a
z8_JYV(awxDX7XNSURXu|;)wPi;5F~Sa=XBZPJ1Ros`yvdo&;|2RtkglZaE&V_dO5K
z_c;*3gjke;V*p&Neo#~zjO>)jgGv4CdS7UqjL`CM=R$GFpKT69vOTs!LoBZF*Xpf5
zV4&O^XLCz92B0er_q@FtUB9NHPw_>RO|P5MyKnaCu@)qqL4<vLqFYBbcJSRx_`gK!
z%CGBfvqQI8%h}U12^m<8bTdhIgNV?RL~f|6*f$$=_A3^N7fAaU&){uEk(dv$|7vat
zg~##}cSJqa2dFa%Sd1@|!8tc1_C%)1oI*IF(bLOqSnqf0Pe!I=Zsn&+yz{EDmL1W)
zY%ylE#1v`Co}2W8?yuj7c&RC{CL#b+83s^2!*?56$t3Y_ta_Y;c5QpKuquF`<u<^*
z9h_+wA;8CkxO*fZVRGg3KWj`cd32U~rjh#Y^AA|+U&Gi;8GeO0=?E?*s7oJriNz@h
zOWJ}O(rIsbKxFclOzj7%#upoRSvQ_bgh#}}+zJ6BA$l~q$F1ifjXTJA9~3zPxIKP`
zMZ;+TWuDtcgwO+}Q`jBf`=<O3NLrt~7iI@95M6LbJ~q4?2-!|nM>-VYtg4Ra6b*V)
zQbkOb$}bZQcPT~f-#&gaGv)+Dr&SA9Yc+15);t?`E90rcDm&zR`lX+wA+iPVx=E_;
z)KiT^SDjTS!HvIfWj5fD1vG_&72_S-7W;-{Kys9-`YQO)qopEsqmj&u7??%yeEn#d
z#hJytT6ha1QW3b=ulsK%#vCjf%Z7KNfIe|?@SzXK<j8#R<P44_k}xbOZ=mo=*E7a-
zgEp<1W;?N9l_)Uw-S-bc@-^7ejGCovJCo|4mmcy<#J>NgYVE1uTRUi9C+D!1-aQux
z(r{{K6lfgNRf_HK5n+evFOtJZzoD$v><R4CVEp4&Gt|QG!7^MA<=Y9O4)tzYGTtcY
zDC&8js6yA<M05^CS=D9(aAY+l1jB%V154C^+s$E$s5xaX?6U6_6UO7Fm!81}6~=3r
zSV_((3$e;WZk?!i5Ppk8GG{`xU(Ax-zQn0CoO)9_!P}ImF0$G7-I-O^7-Ns0XmESU
zYfFxfn}gHdaQV#>G_?6F-Zj!vQc$F_2adkKx5p~&FakpM>T5>(F-~fEmJk2(U)b7X
z_S&$-B!fu}?A?9$;XiZta<b*P>C!i9m}2DJS??G7-#N_h!XM?ZR2zK)u6<-tww_dZ
z{pYV3+8-+pe)B1Gv?2>DZYLpIZx9agAm`8&?UOl2sC`88BftJ#e-p3`8e`VnS%5vj
z3TQ-7Q7(6}@<#d7Q3o-*@Vo;5D)nhh^zit_ac%(lnHZGuh_EIJziSC<?Jd=|oE5t?
z{ZeDi6Q4oDjHE?@6^T@=uJz$6Qvy8VEXb9zNAin*p@ey{1)7c9jbBdNfaCslS1pZE
z)B-=h;If%hyqGIzcxqtpee8yiC795nu!@rN?hBFVrmF2-+`$D&t`KH6!Lu%gv0f?s
zPber2t}$2+p4f}x8bgaZzDCgD1M@W`$e=0$gmm$ybJQo#M|7Nsa`>MlPq+v|k60~M
zQPVWS3oTn4rj!YyES^jH=;WZtQ`}EsIA#!AE4*dwOU*`N_PU;WFdOy+D{B#@+Ajr8
z_gL9J;UK~V&YP9=D#(t?OMrgbNf?)TQXr1TwfJ?zX-^a1?oo!vQ1Z9*S8VcdI@kU9
zta;LEm#F6ZX5niYlIU5jlfw9>Q-;!)Xth1yo&tfnnc6VSs#1OlXc41t<7u1uC9d-3
zF~RC%yfoEZ#$+gqd{-&Cb~UJJABb+}uGxvqyYE&`%>@p~+sf{vf`&<>#QM}U(O`1c
zVP(aSwF_*#erAf*dZ9WZ6#kc(&6{=;#5gt<!MW%P5;GhJ$>+v|Cd!NTpeMFH%W3B7
z4gmt9Lh#ab={}@;qPSI-H#+_TOHEwz_vW-O;9F_Pj-qRPhH8P<H&FMF%gSsD<mFsE
z56fCOBT1^8C8Ot`6n0$gHQLi@4k`m`3I#B)j&8uPPGd+=Izf1^nk81^W+I8!qB}Bf
z^B)g6ZuDhR=`$w~Nb`8;$&|A*E9+>fpc@aEP<}W37f~&ydh6B-V`)?!DY+?l80L=k
z5ob?@G;im~l?~{QZmWYg%H#O-2kNpG)iEaWaS_@x#rqZS2Z}4BX}PZsG6@@rn~?W?
zS?Or>e8x@5T6842w(Qt{y*3m)4dUR&G+<ufnGyRD`(Aa0CL>wK=Kff1%y>ys?keKs
z6(tE*qWq>386LDw(jM!U$5bnmmL3GVY{Tt4s6yQLny*36YzKiM<odUE*5{yog(r%S
z@)@97pnU0S<Z+{^aT9W~<F)4t8mO+2+Zl2tt=l2^2Sc;Yp(dE!hJxYVpI~H>iFTEZ
z3knb`j_0pnL3ae*@KL;_)srQQ5n(Ag?jDTUxOwlnAt%T1{lqI#(wdWwY$PVdc-9kc
zC=T5Zz5x0JU491y7HLQf8eX|VU2;oKid=N$uO<|zxoM)WKRVBWs~KND?Ax{0Pn<u!
zK%pTCpZ_sq#PuIDMojGglPdaWv}a@G{9nQS{~I!5W@lk!|Nl1H8^)Pv*Zm9rUwMq1
zq+ZcCPPANA&~LDJTKylhJ$sMO4v*{gQ1%<!&L=05sp+v!JM+p&IH!^Dy`)osXbbp=
zK?}RAlk{Vh7Lc)sin})rjW>y-ra8BWrlohEE(;ATR(f4veNSS0b#z!yA`^zd(kdjO
zffb~l2{@y@gTp}FuOREIO}t5&sVtx*U#2oMGqazGdpXFyzAt`WUsLXk%yl5ZPfr`B
z=GwYC&IadK@M~{QnUNK&9Wx7fTAC&nNYK&YKNP4hFs1J}HHZ)P1}4v`=C8wSxT;CG
zU-QQyBkVCF%0MWZ5PnO(us|U#?4rj%Nyp+p7*exHc?fn*EbOd5JX5Pvv#Zwx7<7Hp
ztNrVnb9Xm9eVa&n=XtgVkF5~0EkKSqLNum@zL1~InXBJYfa(5c+n}d^ntDceO)&~F
zAqYnvioTtydt!s1{)I8=owMA{m}B29pnu->c+r-S>7S&>+8c&R-P**S_Gps6vDK-w
zTrKbsAQDG09vbhL-u9n>HLkQDslm0m&5_4m%voUEe^(K_sH!MDyq|&3;qOwO9~FvS
z-*TOu>FM#Wowb(#cG@4y#*C<v|F_fL5u5vK%G$yP<e$@CF}{we9*mLcJ9G=McFpfi
zKLY>59(JG0CiIjV)DvGFoy^k*SOpCcTwXr2l>|Ea$sT|Fxe@;MCA|AV@&CyMeEi65
z{?yF<G=%^BT%Y{F(;ZtGlU3SUf8LPx`Q|VInfLLBMos;oSKC_K9DVyn{M^J>jQj}R
z|3H%L75%iL1+bU+9_baCej*gM7pApm*2j0XC+1`I{liDu=6*NRY+z#X{7v7${8@U~
zQ3XCt!^HZ@yh%-O&o7?hk2d{6NUe{a^F8@iEAUNc2?IoyK}ybl-k6^E+TZ&&yS^XU
zyri!GP>PH!pZdao3nC#mID@#&Gu7AsI!)v9`P_1tlm?PbA<+2=jrsv*`daxbx--_l
z377ZQY9(6Y8~H;p;~U-dgA-L2onJl2Unw#su`_ad1N@O2;=36#4G-_@{N+0HL-)=<
z`a|z*XvkqGl(s~sDReCmxfsP-ianW?4}~<<gHJx2PuRKtt_$Ygqtdd)SDw$4b>u;?
zB=`uYq@`PZ9ck*43z4uGO2Uh&in=v6qUL+sC0^$75XUjLv1Shch2<2nw=k_j!dr_`
z)Y{Lwim3wcN`Z>%9KlC;Te|McDWJ`-^p?<e_*M|lw?^cKQ?fXm=Ss*1l)reGo1I9J
z^_yP-HVPgA&SfXy1%G$mXjCB4{=(3MA{&W}yZwUe_acsue{Xx;NDt-^lT%qrONqOd
z$WnZo%@6ypEhvAi(QQS=3Z$6p*L*-Kk>!GvAY-Bw*GsD@3ha}%#2<j8uU`FwiFFY!
z&56KI-aDH1N|v=yr$g9}8`9U2JxeJ%MNCF5_6b|FBde<}jm1k#9s8+S#w?i0DLP3F
zDJP9GPq7R<ffSwZ*l2sYECKtI6v!A`1B+ZK`go30Hlw$Nbh!;Sb`eY~mRA0ZlX5jS
zfJ-~rg*>M%N}9t3A*VmNjPeX;%iu%Jqp$PZ(s-0S;ILj`dsZi}tJ{Pl7OXjLEL%z&
zqdAmqz_wxHlgnDdACgrA88Q?&fej{8H&xt=9w&B+g?YRFX(c~885`M@+|Q&;uZbq0
zIT5AA6jB|S5&f2keMX_5VjnR~pK&MjG@$cuj!$D12WU)MXmS)<MOO#>;bJ=KVNan2
zhjb<SS0pvJ0!1%Ho9<BkPF1UAG`BQ<mmvyW?8qY<gogxVaG__Z2TJu=C}fjavnQ|W
zH?p;zZmi3S%NYcV3Mup`d<8%u;gx$l!c}#A1={Ll3ey2-CT9eXoh#T1H6Hw$6bUCy
zmDhR=1?sNcYo-7pV>$bXD&eY1X~#-3zgP1yifFqxoEykvo87y1bZWcS_+bR)XFbI{
zxO`LXGqd4k7nBLE3#^dKiQk3W9fFSmf>`2Vx9<p%ovx@6W#;!8wq;I1<u&|xz+dH9
zR`UWbNLM}bWTA?(5nSIPy{qy4&4<3y+_}FVt#WJQ>WVq&9y?xMOGaj)2Hlgirb)aU
zAT2!+N!GTdBv})r(#?scc>ZuXF;U~PmZtBoL$4FSwwE4C{KlG*oG-k`nTeDOtX7yh
zyQXDLPnx+ThX3#gMP*YroDPb`+j&{Xtyum@bn+J{VGTEOImw7|Iw#c8a@(?3{L7p@
z=oYt7VxBt3Q)KHX_g5Em`ILz3f;^}mJ+<?C!I3vQ*$d}h#xR>EdNZwav7&t+OC(_>
zb!`AXGGE4>FK?@ACt9gLeU4LKF9jlS8aSh-6;PDuU^rSX;0g>&SJB<v7HYSCRs~9j
zt>~akdg`Xl6GV$mPc${BB{4Rw&;IkxzS4Wr08=efyt-cN{yS;NfyJGZRsq1cCfFCK
zVWlY~Ivly%S;Wa`+4IWu91R&0c3#`Cg*Jq6{s5nWFc>I_N33_dL@Vd^$E$Z@uwq~X
zX7HKvPmGc{PPbW>my*z#umX6k+W}nIF&;}{qr;f{{MCz)qoxS9H+)5MV$&w%-G&gy
z3?5H4x;orNSw>kKu3v}|)^>-K#3$$>7Z;?p5j5HZ{A`Cnbn?LTuP=>iQ&O_FsGwni
zz}<ZMhSny_{&`DA!~HGTZj^B6QZ@FvC1z-CNP^^U*sSw*(rXKo1CVw(;WGsj44$}A
zU&#)vG)BOcj1Mm&Fmo8HC3jS^h7V67KW#OYIRRj@ZfcK0O!N=hFDQ78oUuvcZ8WfQ
zGk}dBHS$D%vZ^O#4Z%kG#HPSTY2Kp5aIF<0!>(A}K;*;M{W>Di@07wUyzhkqTEQ7u
zfk^A3zx8~-`8fJwWH;c$$Q3%hplK^UPQg0YVKjd`oMw(F&=9fD@{ae8sF}ad3YBt~
z!x|7{*?6^=JF9*kGmzk1za&|Vu=L&9bGI9bx{YC`6X-HVk}!h&5znS_bPhvo;^-VV
za_RGY4zhNK#}?B{#9nYeITnD6-UxnNkctb+I@+IZT73X}gK|L6nQqF<2OT|_*9NkA
z8Y<MkKz}~3u)|Y9&awVYQF8VCWaB!7YF?d(3fxKeBg~j%AyI$Rct`@d|CIC8_m{T5
zp^4E*#}tupC!@9>I~Li}oz*7?IwbN+iRoaEuDqHk4&Z;tYzUsAKhPRoXOc}pIqL9J
zkR;N@<@ybbkwI+uqnWf2UxOU{DWwB@izat^Vvl&|wQdmp-f5)j5)WcuU_V6Dj#lly
zrKXE5{MysQ0IvtEf(sGNw_3AR7pw+NWR(I#&_V*Y1`OSy**Z~!6d7Wo5d~bXPY`7f
z56g$HW|$<euyo&-9qQi1?~r-;?gaGcun5TjL@IcFjjB_5syXJ99F|%X5*1YKIA)1I
zC3m1ijO(am!{{rj<Z)D?h#u2aKAa{W85L5pXotYl&=Jb@_$&2{=t+|dtH;)41bh+%
zzMBW&(@0{}w4LQSfrituKz1zZM{6q7meK&19hbsNN#93tz4?1SNu7~8;5wgA_3s>4
zfFlHY<Xzx8J1o7a#_nrTMpc%1VFOX+O($}=ZScO+YjIQ9{NbKaySxy}PihI^LQ!LT
z&x7_8C|3|fTfb4YqPN=NuTin_{Da`n1m@V^Pcyk!Q~#}7PjynfZw8TqN3k<1e!bB9
z_n<NH>;!!b^Ene<<i~D;GO+G!w!Cz%_B`|j6PjgGyB5v&-o$`;w&C$k6Kl;vYs&>J
z^%pN`oq=3z`AKb<yE95I`avdL%my{gwB$b8^NLJ1Sv)ROQqa{j*4JpTatikBX?C?L
z9Y_DfNHanSuh&+am3fGLUy7pSjLll`hvGT?Q)5mIqb)3#fjK1IoGMZg$((pZeDDKA
z?Pa05y<4G`c4C!hdNB6pHPYd&Tcv(F`9EfptXxVJ7v<Tmc;`>EWR`8elXlgiDH#K%
zXBN3k$!=Z9EsftJp9Z`5GIfe2Rq<)Gz!^QJ7DdUTQr&0rt9@qUu;cQr&1bYHIFp*>
zwvHk7{#eZ9$*<+nR?NV{&%kli8{2W*cR=5lG`2N?;p(Qn*O#QeAdUGObX?PLHj06G
zI(Eg6O*o}|{laqSDO)qpiVPK$7NR^66&x&hG0<u!v%_ko_X8I#>?iGWmC4DkmWmdD
z9ul?i18s%NKNMGIA-+ud6^c!meW%wP7Q)j;L1MA)qH}3p2KZ7SiL`hTY*-|RJT(#Y
zM(A2O{3WTuPwJ!+W3MlxK9DK?Eke1$@0ir_$H|2$^Sj~XdQCjD(t!umL3X8oXuwe5
zo?T;1SkiK0A>GnLg+AdDHRVSKP|Q%%X?$y;&1xOCTwZqpf2w6`zjkH=<G^21HbVrX
z_6>q7xUwYE-keELZ?w&k<Qz49JuxDRk4-q&T^*QuzHi@cs<!z=yA>J5po#`!FgAMJ
zC(r6aP@5=hAGnh-bWh*!t<!b<VLX4%n%N%OpT5kJ`(FSyK*_)QAU;nGg+ljyqLIhX
zHr1Uxq76HWuD%o)PRU`~Y8eI`nN=!gx;pSGpLmW7_D<VpQ7boOj&U^ShsF#(@Fr;T
zlXqF)r<$8hvE~<eptu3bPwRs=<EDtKshK<RUPK^X*(tr+0il`VeVdLvIvWj@6gi0=
z6M@^3B&>HW@KBesM2UmAaV%}jebUK(hrw$&)`89H7G5WgH$F9ET!s>Vk1pbj(!_U3
z;Glh9CDT;s0Df%8U_?g~!0fPvsuXq(3OG^-&Z-AfO1AyE?^;l68JzVX`|wuSiz+j+
z>dQe}phR=Jn%^GCxaL!v*@Z^u-oWP%ymNUKFzFg^U;Lem$Ht!}INT(5Taa9seO9wE
z@<njv&Ip3d?kVk05o8p#8sH(`#;xLy>wU=S^pJU#3|xv@edaBNT~qdQwdFO}#T``q
zxm-1Y%XX}s5}lekKlG)ehicZfBU6b(%X=zX$yHsDah*<(7Nt%vapH`ASubi#4}mzN
zEmB`uH!F(nh=|Z|0=9$4_^-0k)L#m;p2x4&vE$Zbt8C=iD-qH$`%$~z)H-Z`tML}5
z3jpLd`<D^?iM`{D8`+rc+F_lrA88UZ1VOfs`W%xlJ+Ncp^HSUy-N(yaWcS8rCf(Vt
zJW%m;e{lE-jj**Z^q?pgrvu~P3!L39!52~y9`YQ$^gw%=G{=(&%<?8VZUg7cz=9>s
zY&)IAa-MOYBF>vklsbcQ^$ex&+};)33)8cls0>`#zj=C$5!snwanD%HVMNA0(Ym|%
za$Zl5t^DR2fo}P%DmL6e0`9u91!%uyLa?y8t?Yg=^p>tkl<HNXVQWO4k<%Dekl0Hg
z1SWopWP_V_iKjt7Ct{sMP)j8TnEKtcQLXOsTyM>?FSB_SQoVdc`wYR>D)u;deX#&@
z-mf@q$sp|;*6NT-Nol>pV<x;tc~||ybofIEW8$lMAbp`d{^CsLvy45*Uh<~B(>SY4
zkS?#e)S7V{L$OhHzA&_j-&XRnr3dd3tgn-KDkBLNH+JhTjdgM8;0U}zl|fzR0sdiI
zn-lgro`2wdgXUtvbc8+ORzIhB7bFySg!LWobtDON6d6(5vun#!t;M|x3f)&QHe+<a
zC!djW$2u$7Ev9MnF~1IJ_%b+cLe$4r{TnCcIOVe+2RJ7E`q9agv+++6@OF>MPm$R+
z9z|sDvGMoWZ$(j-=C&Dp_Je{Xb%N3d`e?qt9~o&ak`$4db*Bw?vN<#F>k0aJF3@vV
zflTTZYn;vCWVuXp&`Hs~{&1lLGC@SVz(RgG<qGJeqC7{+n?rDLdmZAje@^0pA4L&<
zub^KC+u7^e4W4;5yP#~FW<o<Od(@%H1|y!~y}Lc-Fi4{ql*&c(Qjf!MaNUr%@KcGG
znIK8MYC}+Av}JBA<z5A&is~IE3mI88?_ztOj{S)o;g9uVTC$4|9J#(%sZUY7MNh$t
zxZ3AKme2hb^WN?b831~iVB7{g-yf|)t-?`7@l&Yl%}_-+Dt7!%Mwy-pgi@9K_(^2v
zF+m`D5VbK`Y+*?JNq?xz%i1XlLmd|QX*n{DG<29nxq5^VhV79o$`(0caPY2eFYxl}
zi4i*V)G7R>Sp~_ErAFplD}JCkX{Q-}c?bs0<JZr;(NsHPG8P9bi_V+{KNm@&J3b)>
z4gxJ;mhSGFr<Q)mwzw&%<=<Z_atL*j9)yn;6i)3IMMkIIo-UL}dli6}`ICPu8R8=Q
z9>_b&q2g|y+E+nYZIT>UX9uo|ozV)xul4<ns%wcdFbOCf+)@!@oHDFT+U=<^SdGkm
zU^qQT8Lt!3jea^!nDsSvbr5NQXl*j->vqTNH61%!l5W?WPIniFy86lp#!Rq_%Tjta
z=ciLdA;~nj&s)e+%|d#|`|-;%PD$+HP3vxvfqW8r?RQZYqX$D^TB{EUN|UHo<IiUM
zo16U+Zf&|oVhf%ln-?lIV43JEi4s>>t^BNI?xrTqGYv0<aP7gLyMAvymmBtP-aqRE
z=Q+c2cS%r$uIBH}y6>5b72ZGLnQ<i|mcJQhB`Nu3>=Q*MNb>j}kw0loZ>9sS4nwv-
z`GFLUH&=&*;;VFU>Ggtdy}txEhq8sKHdFpg%FgX?&!%$;L0$(3E;w{XL1+h@y67+v
zPj#~qX?*uW$R>+COd@%D$45X3ZE$5Le|oa%YvhkimuG%~@q00RmjxvEfoK2X8+yi-
z*UfEwg&2$*FN|1WM>DT27gok>Sao(bo+nxZXkEl5{5an^ucJPCXi?I#5p8hBLoD8Q
znUoG(12~(n!)V8=g}CfHQzF2B@THG=q^3YlBhn`ev|+WJPH*Q+B#()fGy2y*zmgnt
zfjw4GvTW&ZsWPyz<IuejYmBnoaV=*DY7UvZi7y!jW(#Fb64t^8rTFfP6KHOPf^{Jz
z3UedkE5hgZmBgwcth(Z5WNS@`MejJK4@2)jRJV>X=7qL)$jnn?OD(Kfiig<go4GDh
zx0r5XZyk3Wi70qI=AN@Y$XbF>=}V0ZJ{x|(4c-_eG`n84$PbI-vSp)$zJ{c7-F&J3
zqD<ES)zl8YnjUkU{?bI}sPGA?{R{>alE)N7nS1uSe!o!4CK?|KxjeITi^UgNILzkk
z;P=rys+k}ry{Y}m)YSU<i&50#EL?V<jJ*<jmBypOkcp>GM{ZRNzT64-o9tJ#rx&oG
zz}#fSy(knw8O^dg<3W4Ehx0T<6}G|R2IMN1gn3_mJGi5GXfJpT**`{uaO?Flm$-_A
z*o|c!IS<!rS_JP}5$Z8>n$gFcCtj6#=uh=WI=s)5l0X7B!(`{iwEflz3~C*R-m=!=
ziaGDCvsoK6xye(v`p`?_xJp5ji=Z1lb864jk}kp~&D?gOXfbjfk1r~^Oe>`pUA0(c
zPi>77^Vj^>r1ke8->KvA_(jEa1>OMS?_x6JKJ4LpE=TFoCDq+=s<p>vTo6CRe%bZV
zZE1Tv?gr%@V3lnRc{zzkKkj|qCc^E)g5lrEUQz8g-iPFiVx6b)%Faew!xxU1>Eo-V
zw@7x<D6KRwfk*RXH7Bl`E)(3w^k7n^qAD}AJ&axAeq(Qy#a@@OUv>4qRhL1v?eXJ}
z?)dQDN;Dyj9T;yK=#-mC{_w^OD`2j!I>*g4ua;URL-IWAcHGQRu)rp^+x&^AU?f9}
z*{V#J!7zrfmDX-wQgJa_rRqAoVzECW7$=8?Fxyl7ot{CGI^}ZCSlr0TlHGP<)$+uC
z2fR&{cI16vBfB_mqaWy=v9>=}n|!&_D>8WazDkUtCMr|FtB{UQJ&%S<U)hz4qQoeo
zMRElr>xkm>&BW@T3g6RkNV%v=%Ck;r7b}7O^abSr4H<fxA{*W4gaT7J2{`b}KEQp|
zr8b2|tj45J+RQF*<NmNj)`mH1dhd(!8kvg)67|B1Ay}&)KFVtbk=6+e?(6nBB5zQ9
zf2S$N^iko|417YP%Z2d28*^XC_sqPpA!0>znaDr+<*HAC_xbywpVO%<VwCNXb`Q*N
zy^G%jR%7o6D04~8XP_me|A87Wl*}GJFlbz<LIy<8kaqkk=3aY!L=gt_vLcaTT?qdA
zV=#psVtIN}1*-6s#4<uJ_(an9F+YRK53V^Dajyi0qod^$$)BY`d%Jp?`*U}6yIv8*
zWSc?kgF=m}oYC7e5)U%PqipJ*Q|JD@nkvWa$}P)v@sc+(A>4_8MRuo_8~YiZ^$Iy@
zXI=4F%A9Li)idv$qtT%|K1%51FSW4NEo|IUjXDKN{0C(~;>kesXN$H^!-hOw$D^V_
zFb0KV;~W1CszPHcG#nke?9<rLvVsv$Z7&^eZ{pr~UHHv=x&}xxdsm=FG<p8-La@G*
z!V6|C@+ZdV5_Qc(x>LL*-0CP(Z%C_yO^gP|;vZR8rZ5YE($5KqyG&;$$Te^xRSs~r
z$tTz|nkn}WTJN-TXyz_y%zt7f5=XH?brpM!3*rD2kuq16XrY1|S?HCoYOS#Y_&ClN
zXdGA%qVc&CSbvtKaDZeudpE89YH&~n90d`R8!eI94-)8B*b3u$`K&9};=EdI4EjDa
zpxQc|>HZix!gu-WrGTc5M7WPCicn_QnJT9nJkYur5EIaLDY+|~r#GVPJTTZMZ8?6#
zX(l$1scaJ7o+4rtTYmwDdvo<nj1s=6vkJ?b)akyp=NaT?#9+|mNl)m3!kKj`IH$m3
z2|;fg+jUWVe;MWw+FsZqA-iY<WQo!b&q|0ob2i&Po_s|0Z;~1(Z@<~7SDl8zWAQ<y
zp?RArd$aP0Ou4?{77S`t+BhJQ&BSvs(Mv+dGA_p=&~F<w*`UVe;uln`!RA2ndHj)w
z*gp8eoJ+o#3oFd9wUH3m&YuM$BiWJKlZ64L>revfArhAE`|Q_{+X&q^et^wVHHB*E
zqY^b?<MxLS^86u3zNkHS>d|Dm^X#^#+u^OPL2caQCD+j&(<-&2{Y}H)2?FP}%!;<V
z8+;SvJ$?;Y-|+LfQWVxi47=b8N2@e_VUh6Fc`|(cz{jEqw_`9hopY6dHr}V_bpd%m
zJuAz$7ZsXmVv(k@NV^*j=)4r*N0s0iFm1UV4AIIwW5vH3QUIIYSs4<BPH9@30TQeh
zbq)lRZPfV=r#s+9@~&|z6WG``_Ni2#)o8HFW&;LI)BZF1H$|5Ci34_C{q2DJuQ98z
zP)3KYLd8DFnh)7r_J;ydey1ZLqGYz6Q@<DY=0pKmy@^4f&U|h~p9->Ms^89GTaI+~
zu+Lut*!p-w7a=JCZTgGf?78v>3p*<$lBpr=qeU@l%49PjM`CjPrKfH2;Y^%wfXRVu
z>X}UiPus^+aP<eXozR=olFSj+#WrmHySH(|)L$6S7daEtgCs99zwAC!yaWl_0pqlB
zye=Wvfj@iMff!BG2H;gH9=EIpAqiL%xigmch39Sa!}R?b{u~D661B=T48>WxKK^!E
zN{Sgoy5T8RpIGogiT>Jz_D>80>-B-~<&%izFR^-E8Q$Yxmy3b?O~uLGNL$At1r!h<
z{)8w%pCCfjsBuzG4)YnQ4V<?y;n5cl9u?v&zvbS%?^7BS^srE&+}+5&dg%!sriA4N
zJfSrk^Rw@TV>nUpR6y`0`xq+V^B61v9N8lOBY<@A(?+mDt*Ff${%0;9z$Vc9PomuQ
z5VOEf#u%T(LA0ZgGMWP+ts|*cd0!$o;(CDMQ<@6<ydIpJA5>)D`;D#BX2+`TGl;%Z
zrUcN0U((SH!?}Y8eXw$TD$P*zwOVzP1;+ElgzsxP$c{opRuj<OWfCKD+Yz*L$fF0s
zfk8%W+d+64*9Vs`KRX$2)^^c`!+Rt4q9dv>wt(}mj^$o{08=~k)khz*ZkWRqs`w5V
zib&EDA~j*_a&k<<a%O$(e1IvJloX-xurzG(K6#=?NZJugEU{0VSgB)-NB#Jrtu{g4
zfIZt$2EXtR1rWCP=N~h-(-yhBl{dlQ-<p>qKI@13XO*(KV^zi$p~rdJwV?e~_I4!f
z-DaLb)!#<!^w7kn2!F(yiVi%m!^B+$-Y~YWC8E00+!koQy$i0O9I?qIYntFafbwRq
z*i(RcBm6LCg^!em<lI{uq8<-`$IrE1x9W5aCnE=8<OVYt_-$8{nVi0ZImSNM@--dn
zxOP$Neg}T&e<0I8+5D;FxW+{5%-`u81S?2iMTebpp*7-x@g#eU3`bTCLKH9Dxl7xO
zbqlr_sBp#cKIap7>+_&!Og1`Ex!j@x2F+cF{dCzxXgyyPEX(QMye4$j+!;W2ivX9b
z6#WW7GSiK^^6=0G#2csD#+Th93$Gfykw#y8vWY{ZM3wYooDqUSdZK3N;#PtsVfL45
z|0Jl&3OF)ao_dZ_SRfTU!daD@v}R`(^OY*R!^G24X;Ps^t?r-CJ=cgNPR6jy-V{jg
zU0A`~PaYUWY+C1RN9ui;m0;Ad7NPfwPy>Mh?r#O*f(jZ^-{6juzb-0j7<isBwQeoi
z-FPD6k&DXSFHpp+Bg>S8w0w+#6FCKrc%wqbvG4d?1N12FFT`lut?Ar)8Xg`Js`@@>
zGnk$i3h1Z5d7*I2Bbca>S$7hEO>!yE#u++zo%G^^E-D%FgQ@WvvN&bpe`WCt^G?&C
zrlt<)Q?8+Lq|CV9*rN8X^19aV`k+Fm`FUfnkd5LA3V&xp@1$>FLTP@;8%ES-1-MoT
z62TM~hzLc^C~vZ^;@~eBbAVDJA}LRVAiF93ut3{W%&qV3H|Q><RkR~M>`>CUKJ8*F
zaI}(vyZo$V?`U=@gc6&begAWR4J7`e-<@hDO2uAFSS<eV*5i^i{w{s0ZMHuwRXV{6
zJ_r_V!|z<uA)^LRuj%O)rjjkD>!YToV^_@yGe@*<?F{oB<zIQih&1(XP`q?dr{o~s
zrsY?hm-xA}2T(n16Q}~wh&U0dJawSsE$p?+L@EbvXVP2k&2qkAHuw=aYg!b{%lx?~
z5qA4jtJ4mbLF=^QAMXmNW)E|9U`L_6BEzp@e?v?9K2f9j!@hS%u5GExyLSA}^P0<!
zeOeNeG2!Q^x0lAtiv#6umg=+yXC)@<Y@`EJoIM#IRO1->1;2y<)l$6ZQ;XMs#%!9?
z#vffQo=Vph7=A44r=`s4TWN+K+J+XF(L68|(+wU*7(+dGcRefCkzrP_Z@gtS2Urp?
zYM`U`*EFN$eVU*h2;f*hX%hx94brYd?TqH$=ZAGRlr1e@_062F^?d>M2yC<bz%5Kf
z7-Mu9k4!<kI2s8E2Hy=dX*1|yX}5~b=?aBW$a$vN$I(ObQI;k$o%AB@HQdd5aIho(
zwzN=d(i;pmmrckIS5S2;YSt(qZ)(0Iul<1d;;HwO?DW(aGjh56qk*)w=upFoyCm+p
z`}B!4kiJd9c>rGuSMRtI@W<XgT1CWr)JXJfzEEHeC^WL+2QWu%NV$sWi@1Alf0v&I
z&HxcuRq593LD0M&^&!LFK6jZ&!=E!ad0Gp#khB@U)<ZHhW}N;M!&gH$;Ek?(J{r_C
zX(YYesqM%4U?UX;{pks)C)aD@dB=Q`Le=+79`x8#JM5J)S=h(#8)bZ-OyX#7AO5J(
zu24IbC#Ot}u5K~#Gidd%xtt$x+w@#dSd24fP%}lm<yK<3c>wv%>e)_yEU0Hl9WJ>+
zpQ2a0^CwkfJUt`>&Z~RGdUZ>DmBtQL2YY1^lzK75rs3_(nV$xtm}4mrlj5x;!1SW4
zHw5<s{74>OzlSz*3uERB-UhIEvqwVY6tg9t*aqt%3Xuvf{YoUb(RqZ+ieW5&ZfOL&
z@Jz>I)mn#bw)!U)YlBG^7}sz?o1#NhT(wB^&}!9e8h9J9(Xd*nc8pA<qiWmx(h|8T
zOaHyHOAJ^k^49t@F^!m7slJH1Upb!P2D`vzw#(0}cdI21RVY%0AtY~2VdQl;Bfp?f
zUvUr~m@lXw@t%=u%oK+%T$BB`em135lY`fh>?em4&BP;?<x>#$IRXi{QZkeEAw^yn
zgp<H-p@gI;2^{hfi;ab28V+Q#(qm4A1eDGSaHuBX1>Jd`oomC1$^u=R6_`#XP>MEw
ztLf9RcbbWN&*RSL`MGvOGxWc>)k0y3u_2Qs%+}=VMBVElGa9~oA!az#)L?o3RII8n
zd1o{I5_D8|zA3f|r9&n-n~}JaU#A3WuzQLP2?*JL!;v!NX;P2jcfG%H;O4oEfd74v
z<lsP(;Z@R#nmL``p_o`rRIPT*e$d`WuqP$31qUHI)UOkd876p@fg`>PMHkhY_^axs
z5-8M}_dGscD!k$#C9{y7W<}X;cZolGn#EwejJ5x1O{9(wDHe^FZPl<l*mOkeXJK;R
zL(|K5Ng@=QA*8klT_df8z!Zslf@AZ~+ZyL}GKYTERlYeIs5zCnEg(CBVBFgs+PGec
zbD>D`shlY}c(g4?ccXKH<!!#35teIssG#!v;qDFOXg+o8dYkik*?MupKE!QsxA6uN
zMURX9Sb>#!6?6;-?E(adDE+%`OZEEXI5H;!(p_0TCkqU_V2&4R=+g<Ar2!LLK`kc}
zsgl=RbnVII?2lP+ZP!>&cI@sAi7g6Ow#5&p`=rkE*<kD<SK_otH*%u=eLErXPxv$Q
zeMTaGZEMH6Xf~gMCX+$gjXd19JJdpURF=sef_J-D{La%Wn`e~k$0L8U^m7OE;I8k2
zbl(@SJ!v&6ilYbMM6lVQQkj6%KGSP!H6OzO&gDRV0%zm?^exOg5xW7h$}{)oj3n01
zBYWAxNVfrS@;?5lj2T*E?#*!+u!)AAgdj<glYow?Jm%ZQszA}G@9Hu=qmj+r6j^%6
zH_9tV^*cLsH_t-UF^Z1Vw2mgU5W275p0EqpJ^M;a8???|%U*FZc%?8f>C^t-&2lRq
zb8y9D&t$-U87C+lOduQiPW*Aak_@^L&Ozbx5@bPK!@oXYb;)qDxXxJ3UB2&n+M*Yt
zmgFB*wjZ0tC@0uH$WYh1+PMA42HcUk<GtWp?^es$u*N;&f)!i}Cw+gjhTCaYyvQy<
z#hWZJl(-zQ9BGa13;(KaS~Qo^Z&;lYH>)x|;Xm)`bSm8ELMhAzXsf~-@m4<TMDUo~
zulmkrx{D`;mf^n|$nE&u8+yFfv0c@n4b{6TeOu?iyyJzHf*iV4ay7pO1~83jjK!2b
z1q>Ho@}_mTkik?$!i;~GDZ+i5YUwxbALKoslBpfKm&EP*DT^q&j--7BAGY%vfd&|F
zas>w_RN@}|i8bN=Wn3B<Hsch!-Iw6%psrNca!CE)gMtm~6@HYAXzV7k`tBtz_06ts
zsoD~Uz8+gVoIFH>!#c89JAwQervse<*^i++W^VcNo3vPU>$eHP;`~n!3_1&RgX!ec
z2-wY>v}b?bsT81o;J7+jz)_m$Ak|XCTQI9Vo=D9q9AItEOGWF4MQB3OvHesZx}@>D
z@F+vaxkItGi8suxugO!H7)gsq+_`290-)ft43WC#fO0ON3yjRHrx&Ml+SJY=KmcF`
zN&jGnj+RHF@$S;HvP?>V8|rr@x8FNU)_Z#T(kx1m5h_<>ZzSBYiHkKCQU9I1J`MD+
zLP{fbFoDygs`~{)G1pB>hB=7xu?%*u`#{zB+5cOud)&aM%G(DPnuFPm3u8|{KK-~|
zH5-K+F@9YtvBEnh9F2)4f`LAy3ahDXH1@~ZumDvO-_dka-%CXN2<OHRTt|y{wMN&Q
z=r~{!?SvdD!TBFdgyt7l&m|TLZeEQNOIEyB_ML7KMV!G&SV>4b>z{J_G{OuD*+pa5
zLOGN;)ux&wZFQSqwZ+h0IM7v{H`=+xK&PwY&=tbljoy6&oF?$d4{I~E<m)bu@i4SP
zD3j{>x`*sLwV%8YhaB%KxFne}A8w8<q%A>K5#DB+2vaVE0hwquYZ%m;@($J)HcB>p
z-8{)fytBLQZ_lo$#r&@WZj;^KHh&6^@|ShI@Rcj}oKgAb#y@x@X76t{AjLSe;{u5c
z7Jm*N#ki<8b=@D#yxq0l4)qYS2VSm1cOtu;W?~+(5$m<%N7c4g%mzv4_A{<3@OgQ`
zS7_S(%G}axzTmv9?06tU6+P3Ey%-GDhtiM)r7gNLn2~}2Q+gG~oS}2t8{F+K+oPAE
z25;<rq0*=az0TL^VllkzTBP?79iCrG?1uJyIPHXRfDS^;ZoGXkL_oHBkIk&R8l&En
zy=XPYs(~4zZWRH0dLaZ>BK^HKAn*gn>_^tqK6-Cxe?l81>JE1K1o#W%L7(Q4gzt0w
z_!syFG_2xyJCXeO@uyJ@=7mMfk_=xlHX|#(lYsU$=RzY9hc{^Zc<-5;kEp1-P^*$2
zIprHN9}PipUcC@;8rd`J>33JPl>4uzgiQLJIk~6i^bsdvpO@kd*f^*d-d)raqrPT!
z@fxVC#_CK74L09%Z;mdK#Chm^GO{9I3aSDi17Aj?4f0J8*OAO1OKpe1mJ1Fx$hsPp
z7b^%(N`9f^>-yl~7RT_>L@nL65Xe%^+EM7EH&q8372lazv26!fQm`VOgGE=yB(lTH
z6?AO++u;=zP&v>{$W~ZYj9IZXzm#&?{Cx^dVOS#D{pUg%9~xQk)f_+eDO^JT_B&3D
z!c47>&Q&>7i`?6!4Qxqn4(T=P7_slAX1i_S0LJU~niw>#V@X;6uxj9*TS<wZ+52_o
zIHq^`%+5|+q3ZI9G8I4_&~+f765<w?m@oS>+Gcx=2rVKzKAO_`pstQAS%-`>C4p3(
zAPNQaLOCTVip)yW?`fKDR3i;?Qwhzderg>VP7a!2FzmyR3nTstldtb{RXva^&JYEZ
zLvT<RTXKSyUSbIxvs>*&4u6oSviXG6cE?4Ng}46TxlwJf&d!7j)7!PdR4Y>M4TOb6
zMG&h?7>LDGUF~Hlt%~S`8zORI+cFFwBv;=Xd5Cj6mo2!0i0Yz?n}B55L_ZhRim+B<
z@8)j$4j?#;YpIY{ytwx_p3Wt{MU-0;bHE~%)0P!VA?@W1!V4$WMtFhKPO%$M@990c
zS+pajyT^Jo|MD^K4`G>Kk5CHE$?HOHKgAu?@S6zp2HN$KfXykLT&5li-yp)fcwAOo
zpX*Vc1aIniC5dk(mL_4~T!J4}@muH(%>!d3B_m~N3s^jd#dfp;ToR?Q8IOWf+uYiA
zN07F<jOU!9D-#rS8e?}JAqgoXV;N+B6}6@zEIxIBtfn{}4yEv{K;l6WxED)Aax79v
zB?}Bge@nwy)$VNIWm!y7!W@(NE60EAG|M-N{~~FWV4kn7p`q4+UV={e8mymZmhyYD
zOx#7HIJ=C^{4-fKf%r~4(p$ZFE8+5Owhd)m*Zxz9fbMY0i}IKTC0p0LI)2G&OFXdi
z;p;c4R_-LVq&c)NsFHZo5h^;E7!^Mk4J!{%rq3bqQgCK&@MhO9WS~gZuTp2tSiu(d
zPCy|LJ}`LS{zREg@WN4&+w{+%V>P14H{g_{bH8zN&p*UO7r-W741RGZ^0<%t#y>Qm
z4qgR6^o-xkrfRUlQK9eS7Ovh~nReK0BZX*qn7Lkfft||KQA+vS<L#KSV_2jNxT-&o
zifYJnfJ%y*;>-JUKVRqEEy4-gs~f&GTRKFzgn8u$I${fBi9HxHKW+A?yCPxI6{Tf!
zS7~kK1Eo{2{p_!L1<Bv-Li@fOq1E5z!#%0l;ysR_?a;|vjexd^e0Ro#S1%@}(5$GX
zZ^E+imvj3Vv3uVMrv!^)=~%{EmQbbpeI_&^FY(Ds#EID(Inh!~04`ydT4uqXvWo4Y
z5ZxL`Dg0V+&&saUegyF9_5f?=*lH;?*_U8@TdFj_i;Od3&#P36bl5#J8#B`ieOgJ+
z{G3V^|AKwV`T$ZC=<q-YSKEZA%phF{D{N9LfJC<v$5q^nfZD##uzMO%iuKT&$|LR<
zw)qj)bl;-7^7%tnMbx&C*Gk9br#AW{x4~th*_G%(m9>#Ws;;fWXmld5V+M;7t@;~5
z#^-~#Vci?As5^w%OZ}_3%f53<zMwPUJ?X0{OTA39d;fO4tJ&>L596@|_?gfoIGkl#
zxT|-m4-Q&W1G8mN+Omm^_+-yU{G@uXT@AoH&Fezap0A>xa4MSYs0M_5b;g-*rA_u_
zThW562Nx97&08rJHX;?oHcRuG6_mzsHYeGB!jXu6*gIo!8<aETP=ZX@or@TPtM}6y
zQ3}xf`J#Ptb5Xk|&rE_IKBR6dYHU~|u7i=pa4&1RxNzH%ak_a1{pN7M1+`O+%n5Ou
zX6f}tj+vcL<Z{<2k{eLOFIFV=V)~4&;q+P3uolA*c@~)gCtNESm<AjW1%tBi0q6)F
zz;2lik`tHk`~9gJSf@OZ)fDx;r~k5`n3ZrRxgu_Ukc<Dw&NUPc8y>no9?yvm()p#H
zAv$H~Ge4*8ixVI0{YcWcG)?{(s_-~^vM5M2hS}=&)Cn{ttnmQI$b2PQ{NQqMFJMKC
z15Qg7YJPG#U(wOc3x%co(PB{dZj0Bgfkj|leaa`rIRJlTJ$O0z6@{K2sR;dX3RT!9
zATn(mt+hYqiq52m4_+2QPnp`-YIF0$Xl(bCM}J<DX4Y<VHKrw2A95;hq6^=4K0i*>
zi7_*Fm0I<VbLqK5YO9>})kvGbyAQ=Fc2E)aIe&t%kHFd0jO);_hfH^43KkGU{q}Wu
zq-%ShE!Y3V(}O>k;uzR!9cT)(%yW6jv-E_@e_J;4Yy_-?w?ly2n#|L2rj^crgKa%8
z<uT%-B-8Rx2A#GXtRp<&)a)8zkh9;4uOT;SwhPLoCE34d8sf8gnezGj+!csB6?z<~
zmT^>UHEo5zr6c%wFa7(oo&-UIa}<r>fpC)$oqcm>-#U=Q-)<)M_}9}o_RzBulidfj
zYTX~6iHL;n_Xb(pLfK{U;X$a=UL<2jQ#nSS$GM2Gb<oW0G`)RvIo3(!o&}QoH6-pN
z#KZYZYK<mt^``WI*k|e{HMN#i_M7Q*+bFa<f6W??g3HAuMbOdQ?L0mMFTIy>3fT$%
ziTtDfcc!4bbkBB>?Z6h)qF7ug(&*nsjN;sV0#}p9(8PD>fQ|%MBXBnR$j7qfJqm+n
zbks?!;{`3t03uRJnwUeQL;6rU1$Da%OQ^ftdUi4S^-{Z%FkZ5au5IOS;$6ru&u}b!
z73hd=OeNX`9>eW<*~F7^`I}6TtWQ$-S<JM%r}_^t4tZI*?784g&t4e~kB}L#$BgJM
z&|Ng|)G>iU6TFEw6mD9Tj)GeUx%nYkT$6q*kg(bPhiDEMIB%!Rroo>3aa+<K?_j_)
zL1t0sJ=8WJ2#-hvf_-5lzn+DgyOB`rW;Ke`;R{Z`KlixP-#7gG$AM<m2Qs0UckAQn
z<5H5$<!4I&;73SR8uI(x8a&A^sT&N^dJuPe+;>R8lPod0<&IZ8MFiTO#ke4JPdsFL
z8INv%z5gG6i|}C9>7)BVQ^7MjQrYSK_R=pqy5qs{cQC9B@s5WghdHVyMM9uq(|pO)
zG>vbhF3QVOm3oP1@w$rkfHvfMuG{T)RvPGds_ph6uO{ll&%lkdmDA&0H=EhgNy4_S
zi@;IBIx(BSZjW7zzMl1|S4BHzp5m6j!OQ?=lJ8Un1t@NpiAa2DG(xPWN$aYS`f-PZ
z1ghe$Kn}=H*(RF-tLnLNJPhpC0PzwAo>T$oFH{m^P8fE?hS2MnJ~dsLfDxby;TJ5z
zzk5eMG>ii^#YMz_eyc6OacxM>k0XFfuQKt+GH`Fy>=Kk3Kwr#Eeci&Khih9!<E2D%
zk|Pamqtng@1>07*X7)Z-?~8Kvp~tJ+yK)pQtkg;*UcvtMG+h}keWUL3vj?^$RJ%5d
z%8~$pg(~|vOgRVTMHb^P;G+xfd7r05y2qNn9R29TsW`f^;C#n(|1F!|*S(G*l9lr;
z#4t)oj0tS&A#`=0dAI$L&4sO!c2WFgCI;*`gj`3X=$UYYHkV3bNq?pbwiTnp1>zX%
z2Nwu)|CF4qfGLrr!8Q2Y<8mz%jj7N~7;h<@kn0e%#Ve@gp)fgeerV@8pCO+8OJw9%
ztP$=p!$?(DfaX2{M@2IShTO-drR(*zB-96kG_W+ry(gZ75a!&&R^T%>*PL$$m%Ih<
zKzC6TWau?6q4R)Q1Dj(BJe(pg1az{Z46-xN!5vBRo)t?{jQDTSSh4{pkx;HtlBk0i
zp6yOGjR59d9nM1c&ZAuq)z9Bf>M!o=?tW-f@425U@SqpFLrEaYTNrfDCQb=ua2!u?
zOsXVJ+a&4e0cqsH31ulfd8daPy(4AuEQmuwa6n=RZmOrd4s!fYFbpezRf3G&c;3tP
ze4EW(9-xzkOKq=rHPqB{pTj90@Krn35E+_uuhda+wM?4Z)LE+4d5K9Xtk3lk17s?L
z_~T>>?W>23r1Ycl1LI|?PeG7st{kkrD<zO0&^wh>*#vJ$Hq+ho0VbC9ZG!xP7NMTA
z0u=bmg6BP3Dz}jjYM9}rqEe8UgA>kOkL)4UT6wwW6W$>jMu$eR_f)?`$^?x*ebb_?
zn1cPyf88#39wdiTDMpJ9_(0bV#dVYkL^#1e{`qqsEMU(=K3=lXQ0t$yK0vywcce`}
zzeUgti&`XfGhi1PCAl`rW*_IZ-tDW6Q&or%ut^<e)kxS*-F6Ln1=+{K$&HV<wPYu>
zBzM2ZZUV_KU^Zzcig}aY%CdwV=><_J^<~e}0FmV$nkk5dgCm@Ygt(*^GRj|O$9M5I
zKa`=4_{;rfMc7LP&}MhU-Ab9_eYS8sW$|bvBtO;~1g2$-;)z$@B16jA6{>EDp`%?C
zYy-`Wcip2or>#KAi_RMow28CCoDUU3fAwW5XSIzo_j%Cjcb<w<b54XHI4PgMQ-dW&
zbG~cOQ3lD*g3=&B(k2U#9dIBHc^lGcsm^5#dp9beu+i3JvG7z^>Yim+1>5YTFvyt0
zd9i5TUkaYaeRT)hyY{lZ0a6aKav%I0<U?yV84C-(bMLV+|8Ni4cktdZ5872u-?mH*
zZjiR%+2Mb^!V^l!6;iYE(!Pi21TE5i!G49u(7El)wa9P=8mh$76hwFR`5r$qB)*FE
z!sCA^QlY0FfPN`5^%*&P`ZFcQ#r7!B?A<Ce+fFu@HtkEER6aQYKiF_j9v;ec&xzwn
zjk#OG>9OD~>Gm8ty-I{Vr?4V%2{j=2&8X33J|vVXmb7t1st!e@E&#l$U3+OaWtbfC
zI|2w-KvTjRp`Q-hv9pqR04&VPDAwFproa}hDC#E%v;OO1kFPCW>;A~xqDQC!Ya<^~
z>sKs!!|T%<nqFOkql&78YdHxhI7CMC))h^IRx(@@92fIYw!q92rz`tox8~>3JyX(v
zOLFr1#QE90GZrWQ^SSF2N>nlhc)4+h#%E4f)|m1vMIOwzx;-r|H=or!pMxjWkDCcy
z_{|qNjl?6kvUh#7madX0xO>trDpo4Yy~v;z3#ou>u~_4N_~V3!qG(6#9Ak$aI@9w2
z%mv@kEUwA-szF4%G4L`Ge$N~k`INW&C8j6uD?Kc(J-)A`Q8r_T$Bjd0G{ddLx%BkQ
z<mR+^Dy7t)MTq<Ky{s*4StFh^iA?#O9a|}bd!v1;c$3GP+kBZ6a=kZ?SsuYl*k9n1
zV6PsCcS~WCr*^8%ABSsf!|>PoxKKb=KMR6t46sd9hib+Gi8qjkrpo40Y6TT81gB3~
zd8tw)XA%Xj5+Hx&#C$)bU{psa%TmCwV*c3lij7Trm5Z~ed+(D8HD(QKV0sTc*ld#T
zdk#>SIcQ!l6x>13+Ibf~0hGxFZ35Azzajm}LnWt|k?e>xtJ-A%_DDqTj+fang+^cC
z3_hIJ)Gq#ed27q^8S+~0+L3%;6vnr;(KX#K?Yn(!JT$88KE7uNRaCMeDz|pHG=7F?
zwCx|QEgMe4-40Z^i#9kyh9fDW|DdGu?*bh)Ct5;}b=By@Fq5F(Bs{pZQEFM5`HF-<
z(~s5lP;rWi+RsFj)PK_a3F3}@aES@>%5oPQ-INJIN<{jbx=`?eEH(N;y7XM|IEo@r
zA5>Nise^fhgrl;hP3uUb5UtzXBAO2)GcI7uZh9k}x#R2aokK#e7z11~I^!~}R|-vW
z^n(5cKEy*->DSh1p10n<?K}ie31#_m|40|v>gst>0ONv#C5Kte{9rb%hM|Zan*HZ4
zALBP|km5sS?AYks)xob!c5Iiyw_O8mg1q1N4ETe%mJ$5GY&5LZu766KUp|8POd#w0
z&0)27Ry5fJqTH<|)3DU5g;UaJ`&Hr8O+&0#+`XtMW%6k#njvJ%PUR3UOddGeUgjVS
z_P$r~)sC<^-XJ=^{1{;5+QI=3aT#Ehx%!jMw4z=j?z%x??|^=ux;lbK_s9syZ|V38
z8br3h33}em*%gHtvp^d@rQkoy@@A>lAZTbE#Y$PJpz6mvVtaA^H#f7-i{P3<mUtK2
z2Zo})@UuK%AcUi&BN5nP;PyjM<U58;g;as$3Cc}^E4hL)%-#(hp>F{uUle4yB~~CY
z<%obgjuB<bmgYp8xwUfPN$+UPkCui@)#V{vIJ*>v;3s<x!fEIW-$<DSJj63jx@j~1
z9|}vROK)A%Pd0LmjnHm-pQ(n!%BH<-<~F=J^I1^>&e^B+ee~B4#Ftf(-TM1)*H90Q
zWm@3X@mN9rJKt@wSiOkxy`qW0RZM6T0kw5<$I`3><sEF%zd#CT!uAUiGSR800!~rJ
zJ?4CG(o))4cae%z7CjM_qwd|~AT)d9Mj<_4c3)b{S24ER=l$2>ZIq^tM`bP16?-D&
zh)r{AE|7HXx|Y{OEV%OC1QTECLwbvDy&{@e{8+kEA$qX`yMN>}LAnS@EQ4Rim!^82
zOM4YrXkE39<xxUL$HvN{6t5-k&`zP!i2<2TJ^yszo{jp0cPMVY!?uJFc_kosLz-CJ
z@-6~Hqj0#UEXxy35qNzY<IROGyGjw6S9Yt5Exqwjg}pBk<-o;{lAJY=ADZ*gK)uQ{
zl}n?@%P}|$JGPv}(Y<a;m{?HEZi2F!A}XwH5KApKZe4!_lo@!<YIgN#B>s8m&oG>_
z9XyGZ^@`N`n>Hz+kh`2+;Z!x#-<kl&tCw)h)U`>2TRRz+Q#IY16b?B0xtK`<cRF`&
zxc**nko!HbCRaA<r`BBW_v3?=qN9gp_G$6#@qMvh;ajS;*luO@RBHnueh=ctHGzoA
zo7S%W&c2`GW=$pJg^=~0DsM=`RE3XR+@r^*yo=&h-A3yLuPSt|#8$i`5$oAu=JF4K
z7=P}QZ#G}EH5g~{%l5&zbp-naLr!H<N|qgl5PqGu{c=ZrhP)7tn=h#5*hbBW=)2G1
zXvp+M4f~SlJxnOjo+O=PKQ`Vc@(&1|ho^UN&0Sj^57hHE$%jT-lFmUt0k9z$C!@j<
za*!h@fW6~{QDJ0R{bH0MslIrP^!*m&SmA!t<nRMffimc0pPL|Mcz*$BB2d6aInqIh
z#sXNrtA-i^F}ZhDJ~@>G@1i>O%IHi0uA(=#S@Sd&$3mo|KgPt|Ar1FbOlQ~dNr1$s
zVLe_i=vys2-KPpWZ@Unr6*%H4<CddFV`MhbQK%~2_M;xHa2RywTLj};n$T(%?2?!t
z(bnNMfCB0u0J(k`G{q**wYzS@1kGTOHf<KI9yz@Wl)-G@zyzQwo9p(U`$~f2e>N(D
zmOST`ISrUqL;5GEuc|%BFd_)cGH#94+wgC`*J0hG;$aKki?e&^QPVwz>dI4ZZ>I@B
zj_Rjyn7R|6yRV*|&WS8^tMYz!jz_<N<a_tO4XyNIwmzl5yRR{FZN7=RmGKBLg#=3}
zZ5X0=UrdCt)hTEOkM>hA#%1BqpvRmV{uxjwaG_4zZyjgZha`43+}c~9EXMM5_iUmE
z8VD~x)K48qW&j>my&F+cTEYRzzVnQ=0_g|{zU^PRLx|uHp14sVo<I*+9@9V~!`_C+
zk6aU4@A^_=@h=g2;-FsTR>V+Fbk;)>USQMAWT-`Ow<H4me*wKELfRQPKexcy3kAP>
zB<>B7&Q*JHpP0_JPY?o`sprhHX_>CB6**;|LzU>7fh=cvFX?C@+!^e-3vB!(MOsZ~
z>)9vl)3xB38EeUL*yRUBV{J7_<)y<sJ{^0UiY!{1IHB{9=a?~`=-<2(xR)RO;cJ=e
zKr%0Lv<B^0lPDU$j-}d0B=@-WfBhjkIVb%f^4<p2iIv#c$;qc4veUawy7^_l&mdsC
zB6?JsjbD`s5~Q8lA~NmiG(ZbJ0!2#zRw6{Gn(m4@oAip2C~aKmb&WeC8cNOwLc4!w
z$ry;&yub87*S^Xr)x`E3O@_*KdvRd4BnXbBLb@N&kj@LDs`2d6d>`GkCVAG@A`0lK
zYd@)Ccn!Bw*0m8JtcdXfAUJeKU^~r8-EyGM!gNU>x?X~`Ra8vg5-{xq%E`{DK%Rap
zP@Cl>D?DzJCyqeI&q^XaN_Ta|vY8f;72(o~Xu!ekaDPL9N>KN0-Q=7G9pY5aA@ct-
zS#?+wJRf6G(t&F$Fgbe(mn}&oN*5Ka2JbVQWVq*9ZulI1UI|O8rbz7+E+98%34ez`
z<U{ac;bguV>$2i1=}iJ@(ss!|`>pik0GJ|@78*nEi_+x^P=cN8NUlca-~>O5ig=Vy
zL3>NuK$#)oS0*{Lr~NET1HF>Kb@#g-POFW*POd1B1Dziye9B1Lv#z->j>0NGkvnAQ
zwQ)|R1+R)Tkb}-IQ2>o>-RfUQn7{g;L59uXWq=31OTZu|lFu~*8C!#ksXAj6b&y$u
z=ucTF3*ZazGOCjG<q?_$ElHB6kC|38UP7oR7E1pOuF$->M2VC|qoq~3dFfU?O6v*u
zu)Nn}KxpYyzfDPpDSm6H_|eZ0-5QU9ANk{eWsJbH-2o`@Wl#Px6Y<kw8S|7bg?He#
zDG1fZ?@9N+7_^9SS{4aEaHO+XttC*jtbivNgXG8zf|oXGVc9zFzs1Wed@a*pTu>OF
zOLrmQFFK1uNMSycx-LdejUh;k3nz<>pL`Eesxc8|L$Kl#v0o8Okf|jHC|m_hwI#dw
zNb$E@Vl=#&PC!WjKdsBd9(K2D13-VAw5ULktX>(4G|8YYFnnZK($T~`Oj~WZSx&9D
zkTrOGsMQ7dtw75k?6tj0ubiXOHNtX$qa1ny^mwz3fgvLq)N35=2cM5e^j$$FAcxML
zJ}GU+Sqs^wql>egpemo~w;yf5#B64pzGC58!Mm_=VZ)>{9Kw`DId)$ow!rVZE4<>&
zH`Y#_!e%+K>^C_qnp^rOdL-CjHR!6>u7|ofHDlBP$1)(E@A7<>H(}t{SWms0z1W5)
z<D`GoZsbbd<{q^a-WB70SzZx`lTTs1@S{(g=8i>tk%ehL-A#vEe!%-e)KfQVt1bP8
z?a6M&v2xQa%B<I~);|Y2>$dU~a@co)QX^w+ZYNoP97-BizFGFmW?G96l76Zi$P2}u
zH;^YX+%xxZP^>Or{4*=-=J+sZ5Ti?@&TM)4`DgcB0jK5{b3L@Fjr$ylyX<<$l%Wg{
zPG9qqwg4keP@{FEn%1!ZONKmBcJLZ|7<0zuquZ%1Hzipkv8x84cDHijF<QLSN(jHX
z|7L$#ORu5pl6Uq|7r?~RH_Uimv(5b@AS2!YPcrD<%#v%{n#x_Jy323yh)}gpy>feJ
zU}w178vF!S1dK?0tv#)MVj0S6sd8Cwq^lYOBicG&Yp>w#k`;3DbuD9=7Yd8ikfV}g
z2)t3;rql8Tf6+VMG%$c$4W1i_Sc*ku7g3kpoZwK+vM^@=cZU?H*F1aqf!i%r(2Q!J
zXTA7zxuZf#!Vh@ZghOtIDgv_l&Woj2aytAx+Hd}1;*dz$EEJ$Ses|DcLet}MFhaj`
z^RdLsTQOBoFT(?X!3+MWq1tzNNU@XULf$#q7}cMTNP#8uPQ(}|O`7(An-Y}zZr=x#
z8JPM7FJ@RQY^argj^lZe_08QpQztJi=?pG#d$|+C`Qzo%cH+#wVzjK{Yq0Y1jeqk3
zwZX-@pQeq;IJ4q{%-9LjpdD<p-SR7Ic1Br@eX$;DHxx5wIOwZfR}1G#FA>IXQ1|wX
zjhwm8w0lq$Ldd4BYPC^DNi|k0>bX>jSYr>zT4np}D1y!z507<;3wa86e}GsU+-XAe
z3%%I&8knMKpIm0~zmP?zt11Dh*4EIFYT_d~SlM<~b8x*b^hc-U#6OMp(r%3$i^21W
z`(Es)U$#m~x<-&Ro`Rb2xaqIi#-ouNL%3g{5giE-m|~rxRH((qe`b4zViaRshi>wA
z)uB8OWY<gX)wG$>GFF!!N0v(%#E=C6Bk-`g^^Y#633al7gn>Ig&oW<NKR|>AIbPp$
z|6=<)JR1}wE7T^4kJ20;k|igw!|UxHq!B6z{m`#z-g4<$UuafjxhkWnPGR`>v^p4Q
zDKzk3r+6*sa`RmTw)Hj#QC|U8Maw5-m}}C<YwK(@NaXWLzAFDLHU#Rff8OjW0-%-%
zZ=2-mzz6@Tg!Cg)0Fr%jXzK=s&_kMoowlBx>ghsBj})<xU1v5>Fm!?4%N0+_<id#J
zSEDx_7?Lh0QC0!q$5G4=l>mkl{Q+zew<qE3vGM%j)-8~uNHukfq--Kn`FQ2H>`gs=
z8_kULiiWqm262YGL+z^1>|y>ZQ{l1Sy<X{OiW^4b)kKH4yXx5v<?nCTx{}P)JSG-m
z&mr8Ir&NgrfQ35s)(uK4%rb=><Ud1zmqkaqFYeMPV=ABHwSFi~>Z0ll&<Z`+={%fJ
zSd9;KKvkm;)T!+8QAe2$WPnX0`vLPm@PbC*!A`}PR6jcErbJm|ez11T<EYogg8R8<
zo}-B!oSn%SVIHD3P{o<=jF2rkrok@kVH&H$W@qv+x7-RghSZiD-vjz;Y{Sa=0WbVw
z_=_G-`>S;TG1@Tim5@~@JFoIBj#^E4LYFo<o0!+!vVgNHzTi2s{fDC)?1Y4~O|^vT
zp41ofxy>Q$z*I1mg8<HS^`bK^ntT*THIgrhj+;1nxOUz3S8AF0Q8D7f_&2Nwe#o;1
zSup2(L);{8ZZ5)KVKAZ~L&@X{1&s}f?5XKBp*Ng2OM#S5&YF%|${MjY2hSFz_}=_M
zTIVNva)2j>tj+6}J)9k7K7Wu?=_M0U-yHt&KgqF~&<rb(Pq;dOs@qrebl#>?@bHo{
z7B!f(4KKX%1q`D~kM>=yT-|_)SdC^OR1Gff)*GrALZ}!rD!1G%+9bIu+;K&x1it`K
zz5dVGe)`4g{NC@z1t}*_@AyT3bpk4kn-zo(eAXvs4jdJjh)2A&u)DGeg}n+0L@e~V
zUOz`tny$iRK*XE2kz9}+i<uQD%pOU{&TA`^lQQHF#z<N=)J9jy3n<?7Sf{xJ6irK6
zH5@>5IFHX?I^n;Ac_yF~KCu7qA^*BZFm#x&lkn_5iuNvb7y@JFg{L&vVQ?R!Spx)y
zuY^s+BQ(Kk-s!Zrp)qK5M9Hr;;Rd4h0G3q*cgx3>LV9Oo!f|2QE8kvYVFK<0cco@y
z6_Zl5pnl$%i&4hLsB_LDWOLuinROc1ty!c*#279_qBO*q*=G!jv?oKl0mQSxV|sgY
z?q>)QQKtEzRXw>XZ9UFCymX1`m^>MwFlAGAbwvP0pHW4vLR$X^Fr}4+6*CK|KWlm~
z%8UXd^{es0sn<JD;41m18OGj{1R_$h=a4)FL|WuWRv^c($el;G+>cpey*(}1lWgRQ
z7{5eQb@O$K5Mw6NO;7Hb8<#KRDhPmYxhu4%SG}oM$a3;QcAa=6q%=4qdothUix>>C
z(?yR~oH<2si=Zf0L4F8Owms{8ZJE{Vz=A!KJ>XnC^1SF{<ZPb}h9o2voCF$_UrXdS
z=B@kV|GAYK9PvV`dcTVX(uk71n!)_yI-_51R-!=lw%FcAphx$Uk=N&jb6_n6^2)`k
zkL{|Z?b5k7YJ`)wM9j#_T0=buU&5YmRR$!@E7aal8pT`Yp9Sy(l?!RBK|!1H-`^b?
z+DFk+3^(Y>dX78nUKuLP9~7F8heS2QhOvGmECjWBy1XOL6K?RF3blto+{C(=OrSHl
zq`)M;7K%&};Jlhp{KF3*n#%M#R3JA*d|)T@rn*(}TyOr3Z}qL@GyxDwBy9xVlb=;y
z^E?5lPG!1V1E{BEt!-=5QZQl3y;?aOo@gKYC?k#T2y(@%bT|8C*oS2`aFpFsuP{s$
z9pGczw*4O4wr$(CZQHhO+qP}{^lOtQZPJ_G&HRB$_GGX1Flpb&U`;|y0nIiTSOg|E
zB0m4JTF-Mv<q!D5B|FmJ>R?HaMkwHG6vFr}4_-oy{+R!lfgvY^s@WN&TfEXvLa~Vt
zy2Us(L^eNfo`)AY{fvjDQcabXJAdYxxWF_$dn>9BjhY)2T196B0ulIj6w=-bb@rBz
zhGYGUqy=6dXS|!3G~(DJJp=Go6pt}k@hbi0^-uEGn5h&KgQ<ry;N@P+Z>V_A*184J
zJ?kHde;!u^3wy3ATM6H{#ha%l_oSi>t=eHb#leA_OADPGH(T_vGLb<tIr&2TSwxq`
zXY{OkE`%&2rV#g77MnCHYhO+X=$H4Y>j#Q1UBHl7@EVcyu3?iz7DWTWUf?pldv1NX
z!V}bO25pav!eZ1bqDhVffi{l-;$n7qG#-L_-nUY2NL{8~3oZy17u~%P_d!bMI2URV
zVm1Zus8(j2^LKpCWx11@U+W8^G#>3w92zzP7C|4e5IB?}d{pQV@rv3YJ<00;>;y>P
z*D%VG%=v6`*R2Ny3JoyThtjY%k1We%E`E9%YtG#`CGIVsGDEWX#~J91aWi=VR0be{
zZ00yft;y|E^4q^3v!jrOei=Tc|C*LQr5q9-%je7*WkqWi8eOBQutL8ymT0pRk^!Fd
zP=V-ma%`){)^u{D_<H{jIW2m+-Y3l5;ChnBv<6>3LyR54QpihJE}b-O6AI9naAojT
zB8X<@G~bNr^6u$8M`UD5q{if2XX;?-2s$$066AUN`Sk1u#mQ!UBOD3f7N7w;<EN@|
z5gij)MpIazC0ZWM-S$_}S_@K$FiO47QMuiApnX=SDrJ?!lb+=0IMWj!U00D^x9B|G
z02|1cBhiFJ-v6i2Sf>7cMq#1EQ>5OTFi}ur)-Y2V@m91qP{9AKjEc|~L@h?hR)Djv
zd#Fr{ctNOqc}*rxdcy3eD@wMS&3|f=?1_x*#ItsQ-#jumsHdT)Y>62md?N2@P>4uX
zhY47$O}o^OtAZYJmGL9`{ce1ddEDvb3`<li4TDHzvnmVy{!{5y-V1I~y#76OzNi_|
zSd^u0ym3mNPFXJXapb`;jOMy@F)4pNz0y)$|KemdvP60<xV(+G9AmJ@sJcbbIAipg
z2Ps}nHaX~5dk<W+*Bc0QG2Mzl7bo6hwr`SK?ra1k?Hs{T9AZaSGo#~fI~^H9xx&Cv
zjM9h7E~Ku*?HI0W&Bb)bdk@UV=PF$J=esV6xoQLAtfiN$WhOvHbl<W@Cw*Ph^n$av
zviE1o_|@@l$C)z=f2ncK?S7KnfbbxY39#V<6Q`r!ch5_Prl{heRL8B57WL;KqF{*7
zhp&Zyb58(SMK<pbe_Eqyqv5}0gTpH2D{nc;Pp)yhDCKlAclgX^yDyy9QGmO)yn84a
zNd6)J-;8FTHQ;lBV(;y_fu>>|{2ekKhxvhHBbcbkkOlX;0bN>YOD5>?)vXzPwc1CT
z-+@v#rJU0!i7D$M%m-+MInbDP&OQswUP-6g6*Ze#|A_f^@0d_@)bkQJSDFh49T3hv
zW!{deYCI=@Y|rtMJ$gspa}Xv1dV4SK%JT;!L=$+i98_x#_B-5xT2*^`Rj=`-26shA
zWU<4*GA#5E&!;&B&4)PJG>!sIve~<(eQV=^OLYSasiqSo;lzXKqPHdnVEs!&iLFAs
zfo~17UV8}+MaFZV9*K5)BkTEor;pOVYp%}(2meD?DQ<z6`@p+w<+N3Qb`vw3tfX2I
zqP2~o-GFKu=9}O1m%ti$fYja6I?u(}hD5>TIak4v)vI^KZ@MvewktL7Qpb*x+C{cM
zYD{FDQc-g$rTioG@+!=I;;_IJ2J=&f2mnUkpsZCex|jd3>h^%V;^pQi8ctICUutLC
z%mZoNgk0qbuS;#lK%TVDik+1c#smkwiTkvks<D@&wWBzI)81@T-{Gz!PVGjm?7$k$
z>t;qNJZuc~T10$uMc(W;=M2=c^jdwNW%UMG1?Rfodr|_^2eEpN6ev~k1!a*w&r)f9
zp^OLKa&-XSs;#dm`NtCE7yPRWFvr`|9~2Kz{-FV1bfXD0KvYGVs_P~<_9nV8XDxBH
zw@b#g0rgWyjzV`b_=2NXP7ng3yhl0tTbJN;7pT=UWXg=+cp_mNU+@yaXm-SoOH`=u
zGJp_oiaKb0LuoPU>lM`?pe>Z@c^=5s>g!hKC-H4Kitfp<l@Mkv(<drKTG&1x&tat@
znD5OYj&H1-MS$(?_kG?awO_LH$W2w0Yg>$Pw$E>*d~-=YRDhKik`O)-w>sxvj3@&L
zKH6h*;973Uf4D0QP1a2ajwOAzV+x#>i7}w3x-os>f~N<?QLuP^x}ZrZD^a%qqnd=i
zYBKNW&QpKIb9^pOnnt;cu0KLwj+MakhZJ8=Lz3uA4%c1%0!pa!qUEvy7)pDPBM9(%
zwe#b;*|i8vsEanj>e!>K#sK2%a~6xNw5R$H{h;d9c#dHW;cpxyp&3AFSj9FcL^>)K
zRPfcF6uB3~8eG5pWOkttlQYwIGh67tuObytE<QK+L;kvDo&5S6-;6mPL&)vo*}Lc6
zMBK3wV1$Z*W!#qM&1*Gr4{7>6T(0KW;Dq9Y<Z*d!s2SaaY3zQPt$n-V<l?{@lUrvb
zY(gd&vL@A&m|GwAUS^PcFkrGp;uDA73Nv0{t`}48vQH&n;Y+fVVFxh9y=8>)Lwevb
zJk?5%MhgvR<n~y}5w}*^&1v?u?o2{OX50)d_i7gESV_8uKlN1ldJI>KQP#PhwDUgf
z=DI0yD1`=CS(Qy%m9hm?Ogej6R@U^BPz%t?EtrRr6Y(NOgQ}52$Yr}lpa>X%po78+
zl2+45rS2$2w($iss!bu0n*jD#-U3|TL>Ada_5V1Cf5H9{@4JovKu(UXN8g;GCIffF
z1W%vpp6rx1!yc#n;P!C#{IhAi#T~>BbW|`;d&EBZGZz~G5cNyjjU!XM^mkJgDY1(0
zibHnNY<4`KhZUel+=`XlITh#f_coKO1gr#%%@4!)wQ-zCVtwHH7h>1^WY61Rj*rRC
z9$Afxq{cQ0e6RD^#zOLNI)H*_YkDwCj5gRDA&m^>F;guoeis$#(bSrH5*S&Wyj8Xr
z%4>d-x5k4Y&<Bo&L#2&rYbJkwYTx@`5xCA*U`d(|$Ii49lS0I?;}jiavnu4r)l}5<
z+J&;v&GehO$Pcx(1)FXzbPkDgu{RyN-ed9B*xxGx-!+{)nZ`d|gq<f8uOQ3w6~VVw
zvu{>pMI|20KY_Up7UD<K%-TyR?0ObVS2rASW)hHHFrvm~PMJcUb^~-hU+4;f;8z8D
zcUCM}a5!{%f1Cj7Nmhp^NTa16x`!E3=oz<OGro0@`{9Qjq|ZheE*QhZk2SWXZ|#JN
zrDFm|oNd9=55}smV2CCbxLfx@sgvL$qpJ}YJ}Z~puJ$r@4=Sz1p1zY_*hLexVHD6X
z7cOZCg`TWf2M7)(b*B3**mlsP<;CpLSs`jm+t#KFS7H;IHeG|40ZfE3!GLR@C5oyJ
zl=I0Ws^BkAWHF5jw6J&N-+I`go{4+DrHSJoH(OA1uO79MYS^2*YkaE;q{#<MUl(G%
zpZ#b<KXXq4I{eOSUWaOPQ1~eL^_V~8Vi6;GL&81$O$k0%^I6pNJAJPMT>ZR&<E~IE
zqCp8THYJ{)H2*g~v8LSS_2-Y&e)7tYCO-0GCZ&A;*Nmi{*>aWp&CjaS*d34>?()`H
zC79>)1d#_03`taj7fdEXt6P(A%SP~XKdovA^#6*eKGwj7x@z6TO{&i=tb98VfF4-*
z`vK%2e@e>C)RKfRClLjIYCSLm7E_ADhDKYGf&h=hMf4(fTGO2M!L~#xP9Qztw}{aD
zVY@9Fg|@G}Zd(m(RFKa3Q8v-mS1zm4J8KUS?)<Wl<gXv|9X8!*Y2pJ}<KY=vkJc8v
z)i{U^)RyNUW0g`^phoMPJUDUn5H_qw{@^O1K7XQP>C96cywtzOvo668&K|!50$!v=
zWMIEF2T`Z%s8Pz{{ecN)bCAQYDt*Ru>oUinZxZ(1|42FcfTH?og-(#G{rk)RtYD?d
zxd7r`lTpN*D%$kRptOH@MJ`dOxlLWoaQyeT4EtH=>F9=$wA~UcwBPi%!+2|V;vkt#
z;VV~ZS!Fk)z0Ik`DO7+hf`r_EJL(A^c^Qs$(Z6#$`sHj6jguJE?8w`qH*2X<uBdw4
zT+OEFs;;;g5-QIjE_I#Gzhoyb-Bu0=Q?X~k0a3W&3WktlUm)Vp%pI%@&zQv)L!N@8
zI)1Wh1zQ*%2jHs39g?G0&VjKrWt3^H@UzU;VPaX3u-A_G;k(gD?I&U{T5l$`<&pi4
z&rBu@+}D{)UgQhp=pUuVgT>~2PwPi<$-s8TT28gD?}e<pnEPk7C%V8nX&7$aq=+JQ
zED05SxmTobB`ISy$0UQ_d!`ECo1d_^*21(2hkIL?`-!7pwRo1eqSxg3MqFjo(@TT{
zO4vwZOD?VXLqfa-IW&<YCd!-LekkAE{N6UD#oh<lzDZm}fierRbYUMJpJVwh_{h=q
zdkIxe0N@dt9ER>z!bAc4gaa&XK;F-=aMCjHbwV<$jT>u#WMH2H_ZZWKfUYn=E;n$a
zd`W&jpfp6Sy_xz-)C|Smu;dto!X%R{$QX!!e`mE&(<poZI?jPySkJ6raZun)LDusU
z$PDBHIO8`1Crhki)vC=-x~vhyzSWMI02k7DQLvAx>*AS;eV;;TEOn2Vi_&rmKMSl_
zc_LLKnZQYl(GPON$Q*djJuRhDWq5>v7B)y|;pfmWP@`6L9s+k8@;N35%5`Ohi`3Mz
zZVl--ku_oYBJzV{%o1%`xg824O6qr4(d$wOA@DbaB`s?x6}N`uUqC!HMPSFdvHdWG
zswIJTMYZ^(_)AxK<pIKyNE4N(h5Cyj_RbJ<eQI=z{_6Oi7^M~Wy2-T{l91pl+bLhh
zNZ7q{sExT0MU}ox4MheOLLx_w0lJG&2w4zNlqJ+_KoRza7`u}9``n&%^}2%oYj?XG
z!B1QANw!KecC!O+k(xAwtp{1$YbZHi*sLD2JH6YrV;KZC`{r3da#8B3+Sd=HWn!rL
zaa>TS4C$f56Rw>|4y6FB{FBih3Z+YmDD2d!fDzkotVEUMNH=gpVg7>>&T>C(S%Kp*
z;~0xIFxtV;%40?^dIF#*QY&>$Gzouq!mp1I<?`@?mvSx%zqA{&Og~9Oaiq!g07+#(
z5ykKtw;<`q>bvXsx%THKhUsoA8l^B+o4Hj5pOq%B_p^Fx)GmWEeRjAzD2>Cy>01ga
z)QC<d66J|1H_@SygQT7EAu=#kBTa)!1)u2{&2`d3YqhuDdrJSHdYb+#v)fp_Gz~0Z
zQIuggH+H_oiV|FeMZZ1qbC?CwDgk+Q3CvLoD*;qeALM6<k^r`YM({%7Y7o7|ZCb8h
zr2SM<=04h@F-YD#Lm6;hVNVjI-8$)zSeDf{5*r^^fuQcdOA|kYxP>}lGCN$lAY1J)
zU&36>(|OaaV(_6$Yx~+vUX^QXZh^>j9bV>|vK-6b?%1$f5!yH6Fk_){LC_q*0H2+|
zH~VGcL^)K$59_sG9!5V$vP1$5q!lV|LrPhAq58$u0M$n8RPQFtyfJ>%Xg~B)bITEn
zzA$`&;mF+Dlu!K1Zs@xWmy8IWU$&B$cWVU}VH(AW*oRX+P!E6&w-gJFq?&a<T@%=x
z{nB;~pk+SBERD%GBy1XgwI2=9V^*1^$qFRE#cIC!mai7zy2uhp(W2=#wU2|(3gJ1P
zLZ3?0*StM}!=N4V@wCO<e`TEYrdzL=C<&J<0|Sp-?sbp5e~<?`>v~y`2-H}W5hVhZ
zC}8Eb&#M_BkmVypgS>VKH;Sdhq}8d=Y~mur=ifT4s0nxjsC-M<JZB`tVu02^!dVO1
zt@}r_LNblPVtHWKb!Pfjgacr8ADnH031*);Y-9SH7g~+y<@(72m?{?42y;_-R0hL9
z28%Y~l-{@lA<eozwh6$G^0JMk@Hv+$hxJeto3<j~)0k8HqA&(^?FIhKH_%shGHAjX
zE-!wj5F`(yboWC4KBx$aienD$dkn7F+)%8-WvzJ4A!c%`po~T-XIu}#>u_POZHZkU
zUYX$R5?oQ+u#zeQIY^&ba&*+UO_-sP<1kBB8|ocSt~CfHc5szUN80Yz6j0!;&>f0<
zC#)#s09`4{31ll<8t^tp>hJi*w28&D&2gH&wgR%3XjPDOJ|qg~dw?gxOvxzu=Hnzg
z|K6VfpCCoVyseDNlq=5MnjJNxj~~OrQ;-JHWrnlreM@2_7&1!c5<-v(fxaP@=_utC
zr&&~x-oyS|DfHdSCI{_DN*&>+9RuqG(9Y^~fyboHuo2beS61@MmXSgGX6{?6H9}IB
zqVGCGikBXbm<PA3Q!AehBhbZ5irK&@)<I*!<ul|NV<o(npto?@Hh7!NV)}vJCci(n
z)LVy9)vIiWQ6>D$eO&`-)5?^#;VK~34f|&n9NIy%Hza>gEIiTS`h`$t$P6;*->S~a
zP)g%Px~Wd<nQWFv(xp2&(NvyIJ<f;1lzI92-|ye>YMqbE;`q9ef>*W#!96ntsnYt1
zEW;x+nqzJ{hrgc7Z^O`c(%<&;2NK)J4?(}Rs{!lE=`QL+k=F}uwP%36v!Ua}yD4s)
zUhQBv`e|s*czBLQHQH=mE~J}xR-4ciIWO@8JMV#nI2EE_Sx#?2fPS0Mb97J04QTw)
z1)Gf|+vXtep13V)g(g+hL?!_eAEYu@cA)n;htk*V;9gHH#m6rAbN3vmF2w_y-XUob
zjS7JxgRYU1r65Vj2SYx7PusqTUUk>uSMNxr4B1$8Q+ZZ*!sOTaif`(g1k<cIVm8GX
z%Pv~3vzmS7b|DkxCq1)r>u*S8?LNICzb}^@+`RXrg)P<m>Tv&Tbcfo(f9xb8SmMl{
zmSt-iKg4uK!G<k_@t~|Q;nz~WTDS0E)(AJy&bm&_<5UJE7^MZraPAjhKv{u>!thLk
z73xH9i#*$-*kgS|VoPz#jVVq}KhtGW9&P%gxE1F5@1AgHHiAOoLGWVSG@egmohE#A
zOPion#tCWRe+#!=2FnKd|7f6+K$Jpvc<&J0>ocsv(Td#+1_11TMse_*bM>&7zoO}j
zEj)qHz)NSFcd>f-X?~2GvU9G>U=Osx8=(>PY3OY?B5ua$U~;g@wD_8An`_E!k!5yi
z*NjSy*Xxoe7()2aCH)3{^&}EK1JEm^+xmCI!)dSVH0<)ETXMFoxeG-8ZN2&qkwR;i
zxCq{!%eqhFm@YfQZrung^kvd$um!lVVOtnnw#(nOXK0H@Ckl>LL$YX|a-UPu;s&{J
zdu2s(VRgrBcbd*#dcW5!%EBK@`&<;JqH6K50mN=<;*hV<)-VDQtofX^m6cd&@=~~l
z3W#+;C)-(UMw_}YzL#3~+A#eGq{bf}*7l~cUh+zK>SMZ|5K*vLlR4v#|3yuj^dAKd
zXGwz<{>Gy!3Qq>VDVS{(4(~kmyDAVgMn0c)QLL0&o)WXe<bf@G69XGu`X>BYg2nP=
zuADI=Ja$Hn=`b8xdkLxt0}sKuD3Z$_zIt$P_Q=*eW$a0__F&Q0%R#A;^NC1gle<Z}
z4vG0&q{1e>(b}0Ltuty(I!B;}*qy)%r^AX!AG$YnTiMZeidkT~+MgyUMS}r7_cUYK
zP+A;Mxz@#sOUMsn82&qZj$kmf%a*=hpXn6Z5!EI1FNVS)M(6{K=y0qp4Z;}7)&l^h
zCEg1CMb^QChT$Val_C}7{66tQ>tZ74?4wzPJ3S|HZ!+~`JJIWo?fhqP=7`j}&x|El
zAh5Pv;|JG)*HO;4*#&~B52U3%kBIT9{w_(}+RdKi63-on+TorkM!ElnEmiI~;f3%$
zj#%nh=@qza<%_%M0yVkqgpbNR(#T&JN!pkBMX4adfl~6}N1BVUBt2k<r>ik<gj#77
zrCjm#nvmo7gXfc)2&ZtZ*h&L$_&~!Q=?w`cM!tE$RjJ)^o!^eozfQ_UvQ+P(#67r%
z_~(j=Nx?~Gln|=o3PBCt%BawpK;BjyRvcT>VkQml)*SRqq7**OaWg)LQ*hjOY^%)H
z4R^og+4cR<kYc}uA4cMWI~s|}_X!PwmOnPyOcSw-^4ux(E9)Hm21wE!60)d!r^_wK
z5@{}PV#9y{hql-(i7lMWjHu)l?j?q6z&jWd(xr3@val|<SbtphgIjCBglG+yW~>R)
zF;M0c^ia!z*I8>)VjgIy=1;<(&=Fh9k?;?yWsaMUiY?iY;|wM%Wa!2<vUR*8A?I~w
zu*f>(%Pay709KNNvRfv!%wUE*WmlyE@)ZpNUdypb{?9E|92Kk_@ObkazY=fyfXlVU
zjIr~XNRMaNm0$~#s(Gs)8}&CX5MGK;GrMpi+tm!O>@^e=b~d`0XP{W6bUxcJ>$|DL
zOc~$}5tN=bI%~yjc);&<zw^a#3urdbk60D`M^To1!|BV?Wluu$wW!DBsc~NDLCt+S
zaafDYUX%X3s9Dpng?9mU;X@j&wjcoAWcAVyaP0qO=dl0#f7&@r4F89n^B<nW_`k3J
zmz~4J#PR=O=OmlxG?8Yv#!OBp{io%$+GM4)>SUz+ucMl#9?#v@+b&+)mzv+(+6J}h
zaJ2eAqndxOyv4y#lg)u-ucD=fudZ~>(atd104k&|phDQ#Lda5Hn3|+sT0;K|&!McS
zPpT-0D(T4yFNmT<=igZdAuzOr(KiHQaQvTm4p8t7E_V*i46QZ*iXR6>MTN!h)MFfU
z|M(MMd|$$B>ZvP0V;{V2DeEg-YpE-pALVZzaEgo!ARSsM0FxB9l7D@7U}<qSyT72e
zEGmJmrP-yiIk3DvOJf5w*tj}Y)+R@mBGB}84gl+)E<p6njWo^QMkdpa9-V-hKVkpI
z!1DT|U3x`WL&Xpu0&M@l^2F-m#>X8~|01UTt+DypV;eZ#D}WT25w=>AAJn@QThpJK
z0G#ZrACL8)q2C0`b&gJUOb!jMHh^CfRe1PMmn=H8P@113m)36x;H#-0kIZ$9&fmsj
zUwt35p8~p&6qlKQR7KbnUq`g`e_@hYIM&cyKc_#^|HR~PHO$)CRM*o#6+qbjiRJD2
ziQ&=ZjSu=6y<5M|C%rF_Z@VrwXZ!kxZ?)#1zV6>9Fe!91C2S^64AIg)ju@TaQ`WZT
zpn^X$Ws;leYrtrjzt*gF&TsHj)63^NFoka>rewb;2IhvgrU#Jp^vr?zsg1608OZ#1
zKc$RYKiT6yz=%IFkYD_Sx4*phKb@k#8-%|<p1Hpq&FPg<(b@UcXFUP`zjh-4Mt?s#
z%&xzot@+i((Wk%DzdfE6nLnN5zZB8Mk$XF8kelhhNr5SevB81<54-+Fjp@aqW#NIv
zk#!(_Q}d@bwZF!y&5U%;4fV~8-<4;*8<2TMMuu-T>5;+3_08M7)23f+#ulcJyO+O|
z3%!-4smV!+C@aN3{5Pta#+kj3MKfcgKeyn)Jqde-2WJM*?>gtp##gexj`uJ*@$p^Y
zeVM88QLucIL;YVbfD8=4Y?|rbzqIW?BE#>xmg<72q!Muctbo{nfCu^uzCM`0<6k*o
z{<pq`@J-CEtZyc+NxD#1GQY7KKG?th(OOwqKb7A#*n79HzlPtn{j5ySbgZFj+AVA+
zaIKQRN&AYCY0O-dh1)w55L`G13JZD4Y54Niy9YV=y_0H&_j1SrtL)QH%O`D4w|DO3
zLQbrkk$S>A36`GOuLCLRjkLIKk<|fN9~ALAHK06)0=yfaCW<ED(vi`8*Jieg@?y1b
zzyP+Lfol+#CuvOn>r{My(Z^%|eY0#;{IK1KWr$!!Eqh;qROiocVP1AtM($-N@LjiX
zsUI%5%;YVECHj+WhK-S>{tLOF2lV$WY^(1m1t6a?PYtcmH5X}~hL6}d%yV0eX^_{}
zM#ma^Yw{+fJ$jBVUdQhZ?Z}Tlo~&ZKb6G^e+_|8U*}SGC*V;_*8KEByZzZo@;G0>(
zfp4P9Ll>^P1cCs~{M>!69xv`cdGgEq)^^@w>)d5n(`!WX0vdL5WwshAFnqZZ`;oDZ
z6bbZyHH9I^csnU|YTzg0jRNr;Jhj?<z_5!kTR^qzcD$tuz{i;tqD)!h%CeRCeeiO3
zvg+B7F}5v5iI~r}@0u$eAp|Z&@qFF*)%)c;Lq;N`#;$j&dom{`QTsxkDhzehqzECR
zLNEu);@!3;WsV`kXRROWjMP`?d-lk?$|a;8wqWJ~P$FHUdVD-4-l6eB;6{M>rcZdt
z-b7GA7=Y04$Xf3Bb4>aPKcW4IZ)PuRR3^SY<c^=#SdM=VG2o>#RR<<4HRV%t<m{-0
z?{=$QFJwHT^t%ret71x>n?gY-!G#S!OhFT^%fuj9&7D8GS^N{O?R8>d*kRK{8<vJ6
z$mA}^7?j*RDkWanE7&5fQD`t8{%&?lbXnbpb|?v=w<w5Q>8!vy9ZC`r7eIH1@RzMS
zDHex0*G>L=D1v_QR~$Zu`b;Va&w)OtzI9<obE`F4cHP)sqbj4301<YPy+caFF!!p_
zsrNze8Ou8F-#FnYH$y^g#Ou-ZwYB)Elx*tI$2%`?@~XmxU?h}Dw1n+JO&2(eBQ2@!
zMWR=0s_466zrZl1j9x#LthI_z%=6%>$X2F9UB}obXdFh@9ew)Pl@jZ(Zd}A2(?rW%
z?yLyw`xX+ExX!L-va&LL)xObt)|jO?=LjdgV?IturWs<<{m|iIQ|rAiTujaZS5gt1
zA0+V!6mKEN&D}Fl(U<<4xq``levj3m0nZ((Rz8oe7pEJCAnF$V<hXZVnytx~vQ*C2
z&7np7=b_h$*dX1L{;0G)U#>Iek16zyTK+tHBfOaa_{-5mnM41zc&D8j^n!nGZJleR
zg$A5yg58cqVzscyOt0=I(L=I!-oO>;T5x85x~|sz0yIX+D6Q+F4EvkpEY<kW&$Co_
zm*(*HZt((4E}P_zW+=bABps@gb<g+M@PxqQr)lv`>TdU)2*IoygU9+(@JOD1J}63<
zJ3DEM%ktZNP`+eg(}D}2pr{x%yuEb+v-6&|R&*X41;+;54WDaLtM&;rPn1DoDyu-i
z9f1vp;zCup3t2^E7`2001U|(9u+-UvsHRnid%%(aRvqbF4H)5}xVsMT_0{J1=Ml+z
zSm0ql1l=|77gpthZ1dK<CB(BY^^nA#UXCh>3*!?BWCU_4v-$3_vkwbXQDDZoDyMDi
z@3$#rz<S8~VwJynS7mc&NhQ^@szazq@%C_MTs8<`#izv%oiwFg!%!ya%~GHp3fCG-
zxo6}ZnIORqBt=+ooGs5KMZk@)^*IlB(^qnI5?Mz)V>TSft^}|ypvpHaHpJeUQbQc;
zFm$N@UcxbV97)JaE6tH&F2b~?89t#Wmt*S`)U~nt(W}*UwM#}?J$AlkYp(<VLtY*U
zr}P2%+E^?ZgpEw<J>=&)e`W)a&tvaIn69_K^kY{Wka<mlXTiTzl!CND=@Lm{cUO!|
ztw07KK|XKO8tOwRwccGx!#=)%cXi>}@w8lz3^k$si#~7VSUl&hBR6r<T-uxGj8J=!
zn7ZA0;Ntlt{9y*-O61;l$kYQa$B<rdkTNgOM7h+=i!xg87mWu+0t>oQvsBu<png&*
z45;>T*%Y6#b@)Vs<2KH>Y3HP4^01qV=ysy`e7;Gse)@OYewY8XLM{gzBvw?d{W7Fa
zrtAURn=IW8*llR&asRgYS`e|oXbAg=kbX{`GLMVQ6h&)5SqA1eyL`WXY;t;cN_6hB
zjh>=ozJfOr+R-ELq8NhS3>$iVqnGZs7@}bM?{;3_>dsKqS0#%B<@m<O8KMwYTxDTg
z4|hohirJ7$aXdw&QMjE*I8pk^69P7cm5y>^fi~%@{e;-5x?xc#o8^oKf}T1IlG2_y
zo^_4<U9(TpJ9q^Cnb~Z5ifh_Peji?{3iN$U5N?gPJ>~MrqHT2WtQ`G~8RGNN*()dH
z3-HdnvU3gv2P|(9R;KG8<)`3j{s<%#`lrVzf{yv3!M*N9qIY|I_Ib&#Cfv8vc9)KL
zkfGh4krv78(&k&ykY?8YAR#P!4T&^Fd8rgGAI;X{H2tW`q!kCkW-?`#tOqe03N9}W
zA|<w2wI&eM+f#PMhB)+HH}*i>(D0_P27bOEz?)`JTL1(#nQVbRr&+y8YQHQ=$MCqZ
zN0JSB9NV2E;BHURf~Vs-h;kbFW1HDasIYyKROYYVa$jF;^&BaaPevqsfWcX=<KfA;
z=DY1l<+q5^p&_&@4s2EsJb}uABR>ZW*RbXZuY*JkR_A-$@I@5-cdzX~vl(&g_OCQU
zO8aT6aQPjt62Zc$fPfHVc)T0sR}nuxh5J_|6-*840j|fx5xWr_{RMs=96#Ojm+(t>
z9eV`3)ASBapNe9wIn5oLrs?~3KF=j{Tod}d@eD?{l0Is9wP>eww5;7^(+h#ZUG2nj
z|DoeZ6anNJVJ0rVr6Gb(WGH|b;TIiS9Zn3kdZL~$d0|Vugcx$|@hTEj39(;c-2h-l
z*=kgo>|8=f;II%nyNav9u75rjii1TK28MpRE9<0ZJOznk9nqZG+A;qAUMdO{6yGi<
zw}^{&Mm`d-K!?QWe$L9LlH<z(Kl&8Yy8|D;PL;`Im$Jku{|!AXztVJp{wPr&lQkpl
z&O^Zj39KNpt>v2aA3_f(N;a|oQ7S9Q=s^W8ipv>Gw$dX3XpgxBqzi|t({?|^?!YT8
z>wm=m^_z#WaKO*V{(~@*`dM|dutWnb&RbN=ijQ3Q$QBBBI1%c%d|X*cn$pPa;3bK*
z1ufoQEPYP=kw8I0ENfZV2WT<)4O>-kc2gv0H2vFN%mJr;jxZ8Aq;my<(3Vtr$%BMw
z=p6@-JN4n)KSvef2<4``*5bhF&Hk9D1?Dt>wyIZ>L!*Rn5v1>e@SVdZC!@YqA;?Ag
zPWyeo_w3F-q<1ywb-pOUOd&Xed;x_52uZ4o{&q|{Dv`t~@Y=+Fgpx6b5hq5HakV!8
zz@5fsd@T{(435@S9B@vWe91oIjh;gs4WPoUXVyUP(irfuvq#ZC=v}uM?PN1jlhlUf
z0lG@H#9XvjljA3RNW<TIxC2PRsArFD0VU0IlSu*bp)%rH+_7;%pudas*p!8uvCt5r
zTG+VE%*)g&vG&{b))QMd3>4y-$2N3a!>+io@7?uaMj5_E5368Z4S{#Q*zovKCUq<7
z{`s_YEEd!>(r4o`)&hOJhvq6QN1@*!+1T5iUr?2N<zskGpF6rN;IO0e<2}(fC0FgE
zk`q9C@*A&&O-VS*12Fgxv;coZDG|NJvT4Lq$08QPA^=;tHwuzI)_$>_{TY0UL`neT
zh?_RnURHEf&pkx{{j!V>B5$D#h!dUk<Aq|~@-lFEH9I;mAb{h{S>@a@cOI4b*8sKd
z@ZgUH5=EZesnRznC<qEk!p1O4@5k21!>|zE!ZIIlcKNHZPQn&-5NE&|Qv``2z!AZ@
z%1Ihm;wlKjsOnj3d)as_D(Irb(R3mE0xC+uiM~N(drGPzKnon%?a5n0T_h$$OsW1D
z9M~C9TjQ?5#?2`+{pBcXf_YLi+qhRZKF1CrXd(jlRK<Pvt;4JbW2;ACcfc#@@DNrh
zG9T1Xc9LGcdb!8(*rr&+$Sf&O-y6+yDH_{Bz4;OYoDA8F{^o`cF4Z3(qCpjSg$1ch
zZwpcoMTItTg-%Lo#k_JP5y@zTc^{n9<nB>2xlJA=OKs2YFSH&Nf^5TW@MmSwzI)qh
zwgiemk#J5@JUl$eh!qonkE?YSp43r?vCm2Xf<CnC7+i7~Ymjr0Y0~AD1z6ndkBlG|
z;d?p&wIF7`*voBgvUMH^W2TY|lvBu(XI|se_8o(0Yc$toh<AVrsxUvwYq3M4HC|s)
z{ca}ftO^hP7QC?lNNE)*{s?{<D>`hTFVM{X?Vwt#f#0CfX46&}FX0(@0rf-HBnDos
zV)@*tz^>AMv#ZC2b^3x!OB+m0lurO1mk{wVC%g1O0JTAi$P~@(lv%*FzN<_7i!#!E
z`BfAhA-=}fYpODsI3Js?It1-}HXC#%X^bT&S*lqYw8`ertZlrWL{Wt`y;p0?qkF5f
zKf~z3#w$o_Y*WzFFHl)oa)9<p=8?bSz5|7jUX_$C*?3R1x->@bm#sv^>7Z419nC-5
zH=?XnR3$J4z0WahA&jl`YuF5V8y*1i8Q#e|$7MBL`^-y*F{@V_8yB3&gJ746?Cb@~
zx<qwaar~J|T#V%oh39gPr6yzV9_&I>zM7;ehD<H9)*i=mD4^2N$Shh(4wO#aBO?ni
zJsj{~damp?#VkmZb*0J<23<MwQ7ariLzGTlZ%iOEuGc&`HmWXwe;RG}n<l;R$m-|J
z@${j(&;1`$_~{h2adj%pNBrcq9vs+z>CktL0OyXeTB1kqqPB))h$eP4=Om+-#PWlG
z(p7=9gLxA1&TcQwB+-YrW@1xhuc~;)(DR*wDUEL$D5jb$C>#7pK2?LNSGl@XNM)7O
zmEixZiC&QGUQI!vl+Fz<#!mbgh6s-b4i8E82T6ccV2(Ba%!2^lHV;7lG^>d<l(#%Y
zsXh}Rslg^8$!|r;Tetk6Q*aOjK?!dZcP~cgmn|vdL*r6^)JuPus$7oU`nL0bP_mX8
zOk6w&b%1cWZP=(w_<1klGw|JF?Aj7ga2h!w4!6#SQfH~~mTdSq2Whf;&8VvoLP=(1
zqN-z%srHV)W1N~>{dX+%8%=YU28xEX2eh%|08ZB8F@u93r`}>}R8nG~ftcTG6AF1a
zo~M2#XjMgd<X?Fn<o4xgq!Ptt#Z!;7+mLPBqA{kV3=-BwvN#4UGg}Z)hS!!rHSf{P
zlf~Uy%9WJ6bHvnk2ICQ0BnF*pf7IL>A5dXVM9d0tYO;`Vh3}=YNv5nbldI&#{s!Hv
zLlj3qM#ugOdP3O={CZyuy;DVYPE30$>&~5Mg|Y<VSNPI8`4o$Qb#@jcA(RokL%muo
zAFtH2G_-iKNTV(Cay#7LqPLt-O$$=U$+s48CcZZp(V#HE)40T@5bME{F^ZU3_M0E-
zbJ_P^?OeQ+pK<YOQs=A;t^YRT->mmN$*b<AC(hZ<Nhv9v+AvG*fjD6pv<oqM%Ixgf
z=``mZ8S51`c1!jUisSi6+SLI$+__pHeUi|FJh3qP@nXGRWRVEp!v*T%Hz~(EhEQHN
zZS2nx?!M_1+6+X*TjRl>n(*{P5@15<zM-6CvDV+|!gko^4RO3l4nVVs)p!)35RKkg
zazF)<(&rJfvf(F3ZwRh2ma8v1xvVk|+;;eb{G((e9JHp{*KS6~^8NA`IaIz}rGT*I
zVbvD<$C69qWcd8$#candcz&&r-VH?Hj&6G<WoUL=`ue^xTMHyJv=Jjke_qU>>=$&x
z{iJ(^<V*ZSsR5XiciEvB*v@y8@UrX%U~Rr)S_tHOv_yHH6YYEcrXc8il*ac-LdSvr
zP`TAx*=)u~%l7nK7Ud$0R4RgM?ZdsfCN@7~)UBW)B!!UaK8BZ2;W2;T4>M2YoLPL)
z80;zufKOc<Rv-%3Y4bY1c{a)}g7UGa>tO~<8wpE|sm9=!R8r^D&qTBM&;G$K;W+_`
zp4|n#Owa{@QJ|BdQ=yRVIy6b{cAskly4PqL#ZS5rN=|(=HPWZmcGFK@=ns?4Pww}k
z2k$`U=$Ozdi;S7m5(gQ+HUvcNpK*S0(lMPK9(A9?F|6dN>^@vBk^kB;bS;UKX34$$
zdxqgTi9|58Cr<R_9v+d7!wsZfCJ|%`;kH>3-BG*L<FB`pRxFamjTn7Vb2@^M(O*8x
zmxL!8OYADF2|n>yvXwsrYc0w=zUeerRd>ADA+pRdyptOq^N(`0;qd$-*pg=)q@1in
zJz<oZD~Zai5HuVwUG@ZBII-!=Gp5f?EQ3n_O6E%$ZT2f|)F)fCfQ-JV%rE385S@x_
zq$IV>7=8`|D{zB*Dd5umNP;?g@rB$s#x}K0;%s%LY-Y=UD@{_Ych*MD1Sk)ZTobLF
zda1kbw^UUA;KRCXMWkIIV)*f2I$z}PjV(vc#I44xeJoWw;ldA2?M#W@&Mx&kFdB-C
z^Tn(T@WR>Xr$kzFbqsels_XM!X9=PgK$@Gco?Jl3+0dA}fvTy87swvv2miH@`26xC
zqpa|c^y{RY0i5e)&d}k5p_zIsj|x(-yj)VYBdn+n4OLAU!Pa~$1NGsxvE*fcG9F}+
zfFc=E8P$#KtbCLG$XF8fdC8H~BV41MmTTlTbYp@}b#|jVLAyW)qz+68l7*l8%3MQY
z^rguD5ekzIEXxu+%kKGZ;lM~NT?rd^WF<8S?35X=pFf77fnSvr+2MVW75fnMo;gxl
zBKxaTd2vkeC$b~oQ#`blM-JIXJdJ`L<)(H;xG47+J7=|~`0a3M+}8!8!kA7G?Eesa
zpUQiJ&;`~uEKgbg+eHfVcPRrZKIcetIQRKyOWOD&iy=|WkKl+q7EClb`B^<sq5oL-
z(;@^>h1$I(&w<kxZJdhI@?D0{yCjx;3K>HG6oUb7v1VBT($%(MSjehQ+tC9vsludV
zj@;%2G#=?q**bU*zmlg66cwY1{BT_zV2m0J!T!qZ-8MzNW%-N_rqmQ)CE(mx{0*+2
zt16UnEC>tL^4k3_%>;_F(ynH)amN!6s~>IAVb%RR&}RH+?PqCLDPlvi0V=WBJZ1N^
ziF3h%coI57=5qS=i9dM3#@)yy9XD=@RS>^}Tm-{45}dX=s@qKlNe?pt)Gci@NGpxZ
zDxp{vp>$3-CyELd=sjidY9`l7G$ivtEvu?@DD<$b-BeA-*P#70I!YxR4nCKk?iSVy
z?$x<Jl*pl5=^8w#6lKCMBq#JIZN*yb_1!!J>A0}2*vm)}9E1^>Cf_H5t!FhgPlKuB
z+03;-iXN9|lVqyw5rY|J230$F=kX$<qJ|KE?vrtRk_|cfSKuZ4G2Q*W<tiJ;oa1Vz
zG}STaGKYV{1yr_ZaJj<N=+<vm#X^DtM1@C#uWg2+LrJhY3ee61!hCoxj8Do(Nq1!S
zgYaQXPKuV&r9XWXaDS(4ZdQoVX4M;}CaAA9{!io*6EQ9NmmsYeE2d*2zC2+xCJ>8j
ze_)_Yg|wtQ!)8@lh}stc=`=vCya+KCJJj;$OAe>8xF6WX4<5(UObXHqX(~%*W}lfJ
zii%Wr{&|mPX1J7%HKOd_IMKZtv3r)Dt(-xZR9eG)r<X1uL7dx%BFXF$4l0zGj@W^N
zu_%V%@z}2eLRo1%9O%B)4<<e9Y*DpMT-CYUbw2-aew;iP?EvjH&$g{EZox1oSziys
zT%a-nGR6Ko*$Wi>sX@;1MY2eA-+ggudpUtuNHz(3K*0$$T}NEm+r=GS)OcYm=Qck{
zyShE|)8_M|(CG0K7`c9+WN1P$QK0Tx{3+^z+q1|`4U}FBO|Zv)eB9<E`SB2qZ|Yn~
z9n$O@V!Ydo-V>0Z*5RaKi0<v)H*CQ9<Bl;&co);CH>Cj5n#L7SCe+>cTvkD$)rD?-
ztvlo!bM5NTPK3IeZ{UNr$bc7BotR`!|Ht2XRzuNn6;RZWhjZ!T9roobVgv2Ugv~`;
z(|zE`zUr7%F7M@oE%k!)kG-xY%_qW>9R0>$QM!&!s2Bm_+PFl50!_<uBPSPzwnEB5
zLn#C-3>;0cdLinTh;rRl`BT-n1kv^iGG-TkEr*>HVV+ztTO-==Lk{u$j>ne9GL_2+
zuG?I6{D44cn$WtFfc?D!)hOk1E^-SW@<a1S+1wg0!`8HtL+&O1h~un%&Vh+LbtGsj
zyH0y#`ko1O%)P;aHUz~cIzZ7<!cI%;rgSo>1sBqb#941(Z1G&U&<bz!@s1@-FsXQ>
zsaz)%9ee|hQ`mTeGVO{RemyZN$c2^|T0q84$r}9#^$zbAY=}+m4rTi;MY+B|pbA<F
z+*8-ujjIl>980*TmXx7%1u)HaXUkb8#30~@v~loy1$yu5k{8->$t?TC{q8T5a^!Jc
zJwhv(L5nP^92ZvYUf@G{)xAg;Z)QcnV@~=sQD<@a2U1KM2naEsNNLNJB*G-k{&k#M
zXODcsR)?w9)A44wtX2t=uoiH3-Hb?rZ8~#|%H%o1IBvOKI-z3---=)frL_^xitc)b
zO9n0au~s<4-SMoLrS#xg_E(o6{jsWWoO^?L>cli>!d72vpVq>wBYdboaw9<$S_diG
zfd_mk8nfrX^$Bb`6fiW<zG&5=Dk`H{6{fM$V+aY@fnvr*Hja05?TBfqadzSvG#f<+
zc7ylam%4Sq*-(}@U{<<9>NF+!{BQ<JF;nml+|#imAFt95_yTlpRk9PPGS6xV?bw3B
zs4=XT^zv%Lg<fIYyJYQVfz&Z$Q2!a1&$(I1e=1<#Na3m;oQ<qNyob=LeWkE8_LqTb
z#3<3Wh4Jfi8GtJxc%03KINU~fNG`kI-r_c{gb2Y3m4h&MpO2r$rRvfrwh0`6TaIiv
z`wKA?sXO{A5G#Tg>rOV~53-*i<UFr2X3dg{SWMski}eia?tPJ%yE2Tqf7Bzm+@+4`
z{<*a%)3}njBYx*p)=j(9CG=D6=|4~4<gOcDv@aB))``s%eL-v$XLNBth&BnFdBEmo
zv@JEbR0h)~4m5D5M_SLQEL+)PI-hQeWSfTnDjP|$WxFqF&GXi!L05Rdm^WLN=*;}B
z`cs)lE)6JlM;h!Gi0QGbo5XaE#SIvz<uvu*7+bVk4E2I!D<qvZ^M?$WR!4vKZr{Hv
z(0B8zxbdbOH-T;6Kq*}Znw_oJ#)eTd3CW#`ni468x{uk!1#(P5`U^^&rb$1JO>X@n
zWHj2Gua;ds9=Vy#iwp%BwM%VqUF=FC+=G1+#&H{SlwZO&FRxp6tqn|*_zEpzkp7N8
zRgf57JId^)Q5)b6i?4)7W$yO8=M`OoiDYc*$AW2a>FPVrDe7G+8@@}+xX_`wt}1&U
z13s&ytkh0+Gnl-7<f@P;C$W)>qRg#wmMUY98Dg3i5W;HP9nJO2StSLLWFkRz@au7L
zY4ArchHXXcU?Hov=RNfg7B!c!@WCOIfPTx5Bg=_t$$_Vv891EV{<M2Ci@rJC)dgJ|
z3RkXQ*6~S|d90IJ!JfbzfnmZAOBMpYjwd2ED-LgkFw;c-9+uF>IJc$38zCTtiS6Sc
zF@Uv(wT{Tt63bR}f_rn8=%#ZqbEx{<jke0-mSC=Om0y2qV34S6^8!mhcDDnB>6=)@
zS8B?1p)exK%CXiB0Qr1?7pN&Eh$OUB?TF|9@d53H&saI&&`|Zino)v*3$EO4;}Y|v
zF17+jrZW>Bz?x&5SVkP1TP&(viGvY`DBprcfh~v&WMU+By7R&{1Y48lEz`L)QGlOn
z$(qa(H=&fcALdZkJqx?Bk*GfFKVI|J5(({~b=r4)wVr70lgeHO1VoJY6ciFo<?fCb
zJ&Q6D$i$z_GC{Z2BF5z)=KUBa=Rqrv!D%Z<oAH{c*9Fkb{`c60sYP+!^5;YPX6G>k
zvbRlX!SsRg8*xT&W6-m}Dp34P#BFs!pvVWlF!wILvxO%(@CgL<wRqK)TnA&0B98vF
zJaBI|Ssrt7not8XfaUhhUBq7JQxVF_En$Q<Ln+|`(f3n$Ei!MdQNY8+j$gF*IUtp6
z!MJcu3%N;0U|IdrY>H4WfU(Jl7NYh{m^pz8blC7rHTrs~qRCUT*Sm5e+rn$`^id)v
zV8)C%kT`Y;&Tg)olSdPW9nbtoYMZ#i0fH_53n@7}>4HO~hH*<G_+u3FIei~hwP<pK
zLYhd~{?gDI8zKxp?I$QH(lVU<LP@U`WU(Wbqvk?l`r!onPM(KbbmU~5!(9v=d+!p|
zcrAnHBJv`5u7#Q2<a!kKRimSf-t_?q9+P41t7>Ws2-@ejjIvvp2X5sV_W5VL4p)!K
zq<qoU%8{q>N<@a4v~6FZ3RVU=cm?6*i0w#GO|mx1d|VX$h+wuLt%wDPUk;p7LN#{!
zfr^M)4wJ=&XY1)h0=vPO-Atj8iX?&R+4dJ_1f?ZF#5gQI{Pcv*FD=HwjGLBB4B!^Q
zZ-iTj=NtBNyE2n0js)Fg-r!;t0LRKp7#*Zu2(N?upQ`b@-fpR<*^{r<%BDIU8@Ry`
z-JZmmHKs^LXDZ*TbJKA?wSbev&ZxXl)md`{D>dvI=3_<jm^5P^_rw0ABpt5O&~|on
z)c#sqrqNo$zSaWnUeL~EY?~Z&kyve@tb5tvt&N^2;UlXcQy0R54Ifz-^2uAx;LC^5
z5S#Yu6_Hz1X%>hONsZh|-|j^F5((gVkZDG?{ukopk+KU7Er+De^={Rn+i02}Cje2{
z96WJonsrKWzY;!3(L013citXmH<e)9!}o_dR$wjYzH*<xwBrv`AGdiDPftR#G+aPJ
zQUNcQ{ndeuXaOLEh+3jMj3TF8JxcMSi#mJ9W~|)5&qcEzA>=c#C%<lDIkhZ{dant<
zuMU@7kPSD}YxfSa;!;7+o`NegBWx0P?wj82?7`Nr_&JcN-&FG&oIDVG#ILtZwpANW
zj^`X!6^EM8BF7~<yxU$(AKV%8Qk;$=C_khBjs>BL#8wuoD2}Ib-Gqm}4Uqz;IOwdm
zG}XHu&NYV<>AsAcP;i&;9<`u&CkS7H*PC2gF#Lw9m>!PNLnH}jHcT0TL-$0Qsv`dF
zA*XncY!)(M8OJ%Oi-6xLhmB)+3#$F$Dm8Po@u8^BCr?%KAyyloHzP()+@_4CXV{Pp
zz5^Yl${`rBohEKbYRDMEFQdLBVseEG^WmGtq^U87ir;7skL2Jky1I>{K3#4nt1O4@
zK}@#=XevU-M>+t09&!YhSyIHdt~$5iwrqs>mw&c+UMYC3&#w}{_YxkZ7x}IMR}-J&
zgua}&^+koX*8;n)tGWdXQ*1#OqmhYofFwcUd`@pS`sRapV~*w741<OcAiknia(uL;
z^&=lK)-a*totyTf9U{uXIp1BG!2Y}RzT??;iGwsVwYTCPeA*A!u)D)8$PBHqKV!V{
z$z40uRS{c(GE(Tii>v^5wKZ*&Dw`o3tN01gdmV3bk*Y9(`^8`T@&r91&n~mixU&T<
z-<pcze7VC0H<$Q?eQ4%belaKze_$8RzzGW{B{C|lCBz3QxN;;M>mC;0D5*@3lWI5{
zo_`~O8Ngcr<v8alO0Txjxgb3ny_>Qt%U*k--2_{S%7my!N6L}YkyDX|nIk{DkS_0)
z5`G5nQ-_CBw8QOtS>-cJX=P+&R!!cQB+(q15|*N%Iw*lPkS6(YsN#VtXjMbFV&Is=
zn{A3brmM36IM9nx>JL^-qadJc!Zd~iDKE2_X0s!o)fmoHp1v%LP1XugZ3lgDlocyR
zX~-R?VitUDAc8|1({Z920=oGJHD5|ygR$5E{Fh)AX{OWZ!h&&Z-xwHUQlQ$viHkC>
zkA<?pAKFU;04HL7>YNFA7cE0TgrIngtKV>}>$TLFg6mn>-PqVpc47?Y@B^hyoS0Vj
z7KS?Psg7+onZb4K5hu-9jt<BOP<+}?6Sdm_Gxv{^;C=mS8V9WFwpEYpLzy=>E}f0A
z(g?swJ$aG+a^RHrUoixS9x?6uFz>A#5EopfU^wrIu=uiu2eL#)`4{X`UY&&}#==NZ
z@MGG{5(VR0?BeTltuM@BRfsOc%Krd?Kz_eppL@{E47_gOQ-Z2Z9I6`glV_gd@c<aL
z1^gfaMtE`>NA7}V<ccQ~aAYUZ5n05(#uQC%UJ!^NOKU$cw?VA*AfhbQ^nG=Cb~f#4
zkexnZaV^B<fURQS8v@c>o08}Vcc9*^Jm_^Y=P4f2kIUNkX)sg#vD-;}d8o7E4No52
ztU9~GL{Lmb-0nSKQgKGvYiX(iH-he=?T?NK_G|4cD_pvTvht-nUBc}jT`mT5s965i
z=ylT=rvhEz?o$V1YN3gX%}!32CDF5!lLJ+}loq^tQ*M4NWhIF_lbObbdi2%ifgC8Y
zIJtMNRZ$~w%HV!k21%paS*29J<Ml=n!*iOgtefZI#2Ob$SSzPyMe&0m9{GVeF0nym
zukn*3lZf|z=)R|7q^+1@qUdQL%Zy&?Wy};8EfhKg&%!a>v{`NiBCAeX;R4b7AOy5I
zWAgt9Mp5#BoO&A=Q=wt^u(q}y9{IL-pB1lIj-YRU_$hj~K3@NaEp;-N=_?sT_#-7>
z?cS@*!MWk4)p6nXrLHn+IWW<V<4Q%{>#FT4ZIJ`0*T|N<PNZ!iboO!D!x603&)HR3
z1lWw02*l~;(4C!tgEp@l?*QT>>+6D;hEF0U#0mvWHwsdBt$@Tq$;+z366tMFK7Y&3
z5H$!Ges&8RCY+RrkJh%QFI|G!UnSZG*F@~Lu!O!7a|AkmDTAP~n4oYpk}S&KcV#l!
zVa~nt_0cvAYBn1`##J9{fwA*p$6yRT>ZdoA(!JJi_G}$gdG60i-Q(w06e(-`aS|7_
z&BC8?ouRTdNO$&;9;DVY<g3o)bk`OC6o~SqFkdPy;GQ^bKGN5cjIzn&&ZPQZQ-{`~
zH54uYRb_Z5ZppuVg|r$GgHJI#wPDdkpCv+OUZMvQ<FMmWiNjC<U7fB}_B1zC|Ax}5
zE`XPDH&uM;0VhXR9Fii{605P@@I0ItP76#^dUaI1(L+<`u;mqto6rtKZVyuYLDb9<
z|I_s(<K9p%y@-XR4rh_#lpVpIVncTi-a05FTf?R1ut;ds1P{7trR0qJqNJlQ@Sfwe
za6yW$?{R3v{aLNju*dOAtQR@pXGyz~8uA?2Bn)S_VL(`J>=zHJ&E3YLNkQllTIqDC
zror2$cL=wXY?Q_4Sbp$m_-wg2ZP5<lUap7$MqUCcKlaK8VHlJzueV9gh=X!0)mcHb
zy3Jk<A&n?r;(Qo`<f;?!B-mc}1=3{vtvINUVAW3l<&p%r!}j8dG5$jM7ouEFIM-@(
zI$qv8+D9H;r%LzSh;bHPWfpGwptLx7KuVJn6yG~fHMF%`LSCidlhP2PPYlw10!Fn4
z0iu&j*uD+c%>|Al%XRK}+qiirP7JVYvu$~E#%>8D0h+<`c|UlTIT)lp4+76s(}+4H
zL(-IU$7VwC68A(vr?$rwD_t(<Ce8}RvlfjRPKpblS3}FIU-o!oVB*#PjAFDUi;G9S
zZMs;>w8ZX7(77BB$!2k2)t9Ph;XW<JBNDjV%l17*Wpjf*z?*|*B2CWulZsZXstIZw
zh~}^t+LNgjMSn9#>BaSHRsWLN=!Z{wv$$cmxG+3hx#sCB+Wl<qFt4-g^GUaF{$x;;
zz-3bsI?%E+gFHF8BpwUBhVB4tOyw7-S^;F}m5~fVEpI?2K@w?6ih>om913OTOd6nQ
zY|RYUopP-_8%x>hE^Wmr%y+Ty7g;)ZrvcL<o<*HLwM&AuH*POTi#_v2<WLBzg(@oy
zwKs@WD-Cw&nFKRm_J<nr$@i-$fBHAxY(+9zxX_zATFvW`TP@#JxJ}xCXm;#?1GCu2
zlHT^usZ3u4TQpDT_^adznnGV*(rAAAsFR-Kayy0Hl6QfsIR~q5vvc6N%3zdRNh#V}
z9^BCW>|w<?v2-pSUMXe!E(dhfpOT1fVwV%7Rq2A=5A>s_%v`E%O;_G0j&ap<W4imR
zl2t0o%Y?<ypk;}<4<npJ5>II4W!Bra&adF_ho=bV&1<BCV@ans8KUtz@}S2vtk_o2
zhlB{%RzpW$5lH!$6RgcimZRb2oC*Ky>)tQ^2>by6gOLT^I~VoEV*>A12+he6nF>ny
zA5sr8jjkHwQ%_n?7I-3_#7PhXuPSE1P2>EQ%l<o^cIZoj!GuVpu~eGAhz;s_j&9HU
ziir_4mNM-xLXS>xF4Jm)I~CiPazt?4rStEBiQYS>fra8}Xe*5h{eS-6D2zY5zRQ`O
zu}_z1Ob}%pR$pY9wvK;Sl_wy2ri%U%-7dmza$HR4K8$ZqT-dihB!fk_Dcqt;!16is
zmaBFi{zo~^WHwg>G_{0QCRRSkCoSOi#{#5nY7Vzz3KfHlb5I<;Td(l3zHuIg{dBy}
zr9o(A^56vUdBWh}VP-r%!jLDY&XIrqdY0e$kKs$OWt9c@U<Q36GDe|VMpWuP?@>KI
zjE3+qee&?@j^I@8ZhE6&PCD)mP-&hwS33fZ`H3#*+YMOzqdV8LJ?H%ZD4yio-EL6}
z@Kj6NIp#1B8Du@dOU=U}TORd2$&}KnH^oUCuHc2mg%tyoC?RSa{<CZ5toI6<P}^8S
zO3Ge)mQ(TH7Vf{bNIC**N%kT1a>ok?DkNmrkPj)a-WeCc)h6w!rj@=^B0GAoD4Oxj
z{N*BBH%xAzCg{q)mU!@SbYQYwlV3q7>UOoD;f3nve?n&A-~U8NN4#?<3ib&f>&^d?
z0J}0bL9S~`k0W=GF(jdu#ttf)tz#JlCrOzo&!+VxzQbrX5BCq1-;>S=ZG$}?qyTjv
zLZBPxTn-eF+%6Ne5{=RvI3a+-9Z;D5!4hF0fk*fh+>OT@NI=mBG%F*WVcG?X*td!;
zqtI$)CAE#PQAQ7VKwsT1kM(0e#4j5}1KRX)(hW~B)dE*sa)!Edr6BE;v_-PmVMGyd
zoiM5I*l;2s3Ab6iu}P<dMIzvA?$zADQ{2eU!>K@c2D&LVuPq~5NK`5geJkI-L~(z8
z|BxXllG6JTQiVU_6;$82Ie-_rmO6YIIGQmY=CJ-y?0`ZU_v!vPWEs@Tdudb=CKwb!
z8Y&bD(uE;Y4;M@C74XCBZA!yaEb3aKKo3^B&m=c`q1fzsHOuF&A;jp+;?77F_!_^2
zrrl60HoX#dMHC(lZc{=cEqR}Knu}oFHRR888LXk7Q=8BBV8w<D7R7pi)svdhk_#_T
zH@zdib)3D;yWuVpa54kNG-m7m8vJxmzmAxS<EBgoO1Mbdr1^6Vb2C?{af@leJ%y6)
zphdRifPpn~-D|uRbxr-mlpvL_6=D)ufs+ua3u{UrsMpEQ)`Np}Mia7C=V=4|5Qr-m
z68<B6>#B%fEhVzcHSb!DdIk)rK29*0i!#+Uzu&r4{)oM>8iF6T<3JMBoT<~n-~NUf
z@y~=2`?|C1fJp_9hH+zJn<94Ck!2>FR&Yc$LhFhQ@kDbA`6hfYQfws~{<Xs(6{y3D
z8B}O*MxXmJ5msRnz7R2fstNWTm18|~=!o6FKV(-l`a(2xs5}N7&=+rrdCeS;)_oAv
zSs#oc%Pd=$hr&vM_HJK+^3E-`j@uJ$ufbq|pwMHDX90YW!Fapg25YDraIg(e%=F!>
zulq66z(k94CrM^9hN;2IFuG(|Q;v_Pn__6Q#+|gdA|v6VU^Fr?#N56DJnY1_8||e=
zNNExh{Yk7A!a@)uY>Ou!A7Hj*YuwJBsgOB01|smAmJ`1+zZkSDYB)x*i`GBAqx{^o
z_;Dahi#+dm(-kw}E?b?iR`CwiU|axl&COA(Vbz-&{d-|*P9F9p_3&ZiO+cAsy0&N?
zEghENkV{5J&2D3);*G7`-+2whsRr88qQdpz?xF!!cgqmEdK-%Da>(s1MUalle>1gX
z{Xs@Z<p^4(?JPu9!9SU3r@j9?Ojb1tQLM+!ZmvZlz8)k;X8om2i}`!0m3YcZ&`$<V
zWieX`0vSWAh8?*Uyo&ez$bLyw2(67?+MKA1lGEO>3HndnL$Zs7Dd|S`tD`5}+tOn_
z!T#)F+K!#5AFW=!WdJWJny9VafK0IsSWoWnFo@GJ(Pbo*B_bqHS777IM9SQO1)mDp
z)*5wovzkq*bc$MOXb9o2@}{ZDpzbG9o&D0lDGmz0FqlYOjf{5WAj)VMQaf9a7JiO}
z&KY5?Vuv4RQ4-IGb7(s6=mY&Cjc~e>wiBs^#>mDI=ii!vf^4DV;y^9hYMn)^8F%X2
zAag$&i$S#F<Uk`0Zjss{5RZ+LEd3#iNwBsUdw(v0FwItM1rgbh4v-0dmK?r5u5*34
z5MjTZ&8{-fWBjcvYOdals!7YCTd5ro>DpQs6Z0#Zh<wB(f%($vClJ+1KJc?DR<tX$
zfqM0{jbZbbgv?Rkg?|gR@$zp+@=8I9^f*a2u)WTh=?o(6zRknfnie{e%=%5nlQPo)
zzY^U&z%~3r(bAM6OxVbPrPV4AVG1XM(4ol+J$}7Hpr<)T!vUDhN<7aGp*{37u^FP-
z?3VAiO6HI>{p%!1m7G{MqOtfWbu~Ds&^SagCUX|C0VKoGZthk~WJg9mh;B?+RPzDB
zAovxJ)yhMy(k`G;zC2Mg8x`e&Xcg-O^FdzLaF?rlcI!k%Oo#J<-%Gkx%zkPY0SX34
zbY3)pgtbt3)4OK7)86p#nulp*weKB2l*M?0WnWw}+=o**=;nWuTbLrJKC2J)dGs}&
zM6;z(DkV-{uVBniBCXYg)%kMP^3tL++^Y}~dgl=lUILr>(j+V4O9AX4H$pwIpN)&l
z8haFJRDbRPek)7{%Ncm6EY!5NVuyDf%JWdwPK(689L_TT%FFJy;-4~{2}iWf#I?3~
zn7pkEvQlr=)_Fob)&Z;6W?|m2BvkJ#)R#Bw?=hqq|M^(IsB}+#!}8pLUxi5}*eL~8
zhgE~*ZkWUr#2^P3!K!e7PPeyhX4Vwn%NAjHw;Sw}T+F9V^EL;fqOa@|jYL(xKis?E
zWgz!ZXCtu*CBRZmMIKbv^tp7-xOO7EID?ZAt-tZqHyQ;D+Lo-tysc<hO<`UG9%4cf
z$$_;~LQ|o_EZ6JXU&voRU|pf=7sZ9U2I5{A$T8voOJs;;N^>rXMO_L6(+19b?MlSX
zSuf5!)QyEXn|f@r2^Vkh6Vg~0{z0;-T-fe2%p~7YpC|k<FN5IiBv3LGrZ~U^6hBI%
z1zHo1pu2#Lz7ymw)#pK>^+IT5Tc4~Vt`|F>lA(#08MhD`3b;AoN)kpvN#Q+H-3NUR
z#2^5JvjIdJt}C$EryiFW{}^wtp2DEBcx}cCm%uCcNn1a5VjunzugKuT=)|YlyIdiP
zXcK<gY@=ko>!BMsi=1|fafoLY`gOatU_Jj*_ajYV&)BOh;?03O<BD5ycM!OSm7!Bp
zg4WB40J+4R{}GS2-x&4n3)R4}&dg|NMbOdHz_^`Lw7O8y^7wXWl>*p+=ID<}!gZ)R
zOYO*PJe_`2ekv$<kX@AYdkWW;XMVs{luuX5geM88dT))vDW;HD0$JJMP8`?pEhyt!
zzAVyL>yj7B*VF}k4yYkj>6@AG$I7FgQ*sRLkk*F+#ygEPW1k!(hZQi`q~82+6jU=4
zK{$STW8VaNq9)av#RQ+r$TXZK70?mCX5Ax|5XVIeu{8?9qXLD<i9!f7CYD}GNnfSx
zR|%5KcVIW8*iS*26?Qk5jj5c<KvtcpB=Q#ltnuFb${Eo28Hg{a>e^p)3m!>5jU<eA
zIE4*iVuh7uwHEyF$x}5C@oCipun`%zO4TsYPoS|48~!}tTOpxWd<U5l_3K8w2`dj~
z*y4qhwOujF+@oN9qu?1775kzhf6y)0F95w{u?7akVu9<Msv|iJlls)W&Yyf$<3*BN
z*hVAnLZ!XPV@6R|SzfCqUg!<2zvvVNnTb0C72@%ZzM0UjP3<VsrF3ER1g4>oiL|$o
zm_t+s$etHNbF@D+O)MGhhh!nudVRJT6;Yi=jjwf<Q|zT(pQ>(TR8>h7S6!}r*Ywc>
z<R-*AGosFHwg8)HBcnD~ZGj=y_g0(rd=O;5nY9uz?0M+|pOP87;TAm|*d#)a*jth$
z;=00j?HdSRtiJ*s9GeC~kn^<&xpYqW=F~i;TAH^;df?QPg78STD!&owAYw}p=9ebh
zYBEqedI5#VIBvg#(i=Vr=f1>O&<DU;B%6xWIJz_4WBiI=;s$d!ZK=d@JYHkF`pZI*
zuN(ffPj_9vI2vwCS{uv0Wm<gS!?eA|9Ldfon}qn~!#AOi+GngiWWmX?P{tT8`5skX
zzMVZxG4H9WFuAUd=znT(zA-iXU|{Q>EoCB#c*itnoqpQUO}c@d9=ske8`<<hsq^6|
zFcQeWxK%L5x8T{<Rc&t547J?S+~}5j<URPyZv(*Rrm&vB^otSWB=!yD-CaBT>Sub`
zK+4Rkv&O5K83bZPMHNi=_95`B%TsiJe|MalS8-=IZI*pfz>9-n<|*loHS}Ei8%=6O
ztm)xUf6sP65_Vg<55VmlX6#FMEWt{t7o*8iHec~PnhWbTNjrO<)?hg4+h_;CJXw#d
z2pZc}NMBrJX!g04g?xAgT*$`-jeWOHMd}ghQZ6d1KZ%V~E(b~mqN>SUUHAEc;Vurf
zqZdjp>(8B)6(?Rm@6@)bjneK%>}-PLm*?m0>(|bTgW#^9qcGjoTA(Kn$;+MRKd8UI
z6!Eqy-nLr$e4lIPFbD@v=+M#H^EvDvh+!gWTrzZhR=)<wv?cNiavzxT43j!~&xe)C
zyiPe&kI);3=b)H9XyHBMW}gCy#+BsxwaMY<Er?ni!5DSmKno^exL^iH8Y39WW(;=!
zwmWP9p559|{If0ysJ3nqViS$HT1dkT)QoRq*ml4^d~J<(@jqQCp^G&U-Z13%IvrDf
z-)XorwHn7eEl1H4j@)O8Qm9Wkw3p$z;x!xr9-3Hlgg!ew#1F_@XaZxGdH4x8UQB@^
zw?`c$Wea?vRB-bk#i+W7g<Uz25RV7k0dNsTK5)l>)>Nn+qO>{{??-Y1%?gCD<eQxQ
z&efYh)_wJP4l7aTAwJ(CnW#=@x<o(`xRSaF2KTR2=qI1&9ZzveOLp>cJ5gX+pE{>H
z9-Zx-sPS^{3Fn<p*$B83d~0muH<02)^2=<#>5oxrKOxBUVofN0?x?JR+xTU%r-tfy
z`W&8q0_b}L|N87gYvNUbg2pr8{E&1%B3<epib~1Hwn<YuVYNGt8#JRToY^f{lM4m%
zRW!z7TUGf@PP`7YnXu43+22_8w4^V7bl*?A9GO$fC_kwS^Fy+*=nMT&SkZr)B!!Dk
zIP~c`SA=m-9;#HM8@_$O;8->1jo}U&^r+;#t|weQBZO5x);+KYGds9q(xm8_^*L=m
z5!@mpQ<EoIkA2J;yzoZu%Oo#Eypumrp+|oY?4l{$B7CuU$TC1rW@j23(ZVgu)}*dn
z5m9-*32DN2_nrMppge3{V!SH?chK)ace$Rc-?%1*aDj1_YhAoI$y2wQp~DC!qG56B
zYo7q<;8f5>Cjc#rH=+JDNn&|^W<BHxYJrP<Dt@kOy>M@hXb3%ewtv?{-M%#h_|}DI
zy8!2XujbMYyNCP81<ui7rhH&@&2>5cG;MRa{VU|QlS)*=mG9!7*=T#^8(21dQ+$eY
z1RRFC<nnUvbvl{s))zjeInkqlGw*Wr8<P<2d+M|;yP5*e8SEiCif-j9`vWKbSobqf
zy0EL1<Ep!jxYzKLJ2tmL@JIjQmjdtHfn=dXRm$K9xqf@c`;_rF{(=V*uwtX0et7PK
z7V!7h9NXCG6Q_DCcLXgTWrOjmN_oeMzoW%kDqLjxx6NI9KQ*BIR@pf*k3gB;DvR~|
zq%_Z>kr+5K1C#^K6b_5R$4;0d_Ba;=@jXlg+}@sIYxZ?*Jpd=GG3NeT^XxUhxuh4>
z{^AA7pqa2OyG{(HaLA)Ib9{Cv(?;_i<Wnb?s+U?-j(&$n#d0F?OYh?558Cjau&}gg
zXly$e5Y#N=t(tRC_0CE&u(w`+s7ST-4Jfp(YaaD_4|qEB1a^hH)z}2y_V1~M$en9X
z-G|qHTmL^KlSPP3sZ*6|shkZ@)!&3#?es`ieg;J|+5`YNaMBV?eAe_YpC-W(T$-&Z
zsHjqe_Y+l&9pDbX03%D9G(SYGO07|vdc|P$L3oWOPG)8JZ1?qUW>XTvI<4jCDyQI~
z^k5=@r)KmVp;DKsHi(HgpH_jVekW+SVF-fkHcK2H8R&7LH&Ad`Ez_yp*6OnLR9QDQ
zZ21{GbbulGJXM>faZq5rspnX!lFRIc{ihh0@AlR&d$|;_62M7`B*-2f+x+M&gD{Q(
z(a<V>J~EW7v<_+c38*?Oizakkr+%DTw*i*%DyeVAS6*cH<96=&0EAm9%nz@qCl*Kt
zHH5OVB?2gT@yYay_S1P=0zv<^$S3&y^;;}hkMHVMO~6me*k3z77v@ms%4Q86rgZu~
z*<I1RY|BhlSF1SoX(vcbO#HS`*jR!Vbiq)5>)GF;_^cr~ejC|T$EF&Tr>V1sw~90=
zPIfveM#qZob%TlCvFD%&kURK1gNFD8KR5ZS)oOwfaff57192&&as2`Zk_yRtmg}z_
ztKv6Sima`Ql|4I{hvKQJ2eQ<07ZubaL!}f%;Yb!ojN>K<RI97Mo>t94HxIe*ZKc5y
z9MwR}&#b8A;fXdSskC}681$CRT$f3rP~{=8L0vfxMbw#WxEDw1gqNERcMcS!*ytza
zUJs|qG5QDWl|jBB_Fxv$WV%UbJ<~PtGs!RlG20CB>M04R=I^(b+;v4x#%zE(D~R&4
zq4?Xjd6N?UzW`ktqUA?CS?yJ2@(F8buNpgwUZ$yv)7kW+?6^qQv;o`UF58-P$|7s6
z4sKOvA^JIHBF~8W-<-tTEt0vsk)I|<GWmO5674_Ho2dr8j9H;q*<Z(Du)jC0&lDSn
z>T&qwMR!0dkDCWMwOJUyk4F~F`wp$x<5QsQuP%O)&mp~)bb*;KKNy1+VFh>few{W+
z6EF&!F?#?7<i0GI_-^i3fsX{bI!i5iqh*Dch82HgJoQI^9;(w{^cc>EDsXi-`YX5u
z#K{_e|9@q>KnYJHqcb@MA5IOF9RBrjM8XH7`$`svVO45*=RR5B0%|^#?XUlQooWGn
zQVZ?qN#^|b>Hd9byc=hsAmFdflhyZ?JsEyajz|cIh0d2zyWWMvo-dtTXjwfPV7It7
z;@P0J3ioa6h%_Dg4{R-BzxLj8Ljyp5)LazMxYgVGFB}h=SESkz7bzR`KBUTKGJ0ZQ
z>%~FM<oK5g8<#r<Q?-B@zxo`{Rhg4XJh=rBwc)@N0Q6CD7b|uz9^N2CDB^mpWEyTs
zQIbB@>uIY19D~uv6ymkgjqv4AkWqf=h|A3w@Z82EzjHJ+W>?~*|ATRuYU%;1$n!G@
zLv|0$FTl&?4+1iq+)=T8D5uDkgw}-O@~*%%b+UHeI9D+yX4Q_l;I{d0IpbUQPE8=;
z)kn~`+2*Rv$T{(-IqenRl}$uAjTAg}BPid`v&xjDAr3OIMqsk!X2H=61X2~lG^&fK
z_zFsR!X5Kytw?abWuELe4CL{vZCz>Y<Q#9cJYr?67q0|!nW@LgIg0GVVS~uB9;BPl
zDwbg-J&jXHV_xnd<CXAKnHx&Gxw--p0`9QM=@QyKZOPSCYaaBV38jRqPY#h{LPX6F
z+ujxFFAi<UuFX&WMwgQ7<o=J@Dy<{e`SV@onYnHsdFNuh0LCAgt1pAKAwisy2W0bm
z%Ifk;b|Q(mkMC2f<|J_C(-sfFKkra*9S%wpvSPlb^3A)e-0%2i7RV)!C|=h^86@63
zD6IY%922bIj<gCHl7UJWbFYoO0nDm^xhzoz%LHLn-I8tXVe5UZ{Lz8aaM_cj4MFBa
zR<j~*E}Wj%z>`%bcSPVuVt%BfW0;ejyw=|XMn&PK0mvku_>Yma*rwX6b&J>feX*mM
zkK;Ks<!i`7fh7+Fs_5;P^roSu_L?<a!X43EN=eB6Sb-0Jp%CT)tIyQ<e|HiCxzWH*
z#@bsr$Lh)Qdp3oE$r&Zp;r)Ue8gc^SE4?W@>Mhbw8-Va-5j$|=H1*n{vqum~dBE2x
zk^vxfZO_s3jjE<Y$Jvw0qGgugWn!2M{%J1p`Lqfu6f|N#l)QVUl@l)ja`LD$XVwSO
z`C@XLMcakWN_b{Y80eL)=9Taa<L8`R=tjN!+$2KEmE9zKg}?>dl=RS9h-&AWNPloc
zKfp9Y!4>rb8W?WlY_k@^k4<kMU;JMDDvs01S&-pcjJFnAA0dpc|Ez&Wkf@_t*Jc<$
zli{*X6Y>amzsa%lk{+NAIyH`uv+QO&xpL%XkDgAe0j{T~W6lGY2PM&v)~e(QotIiJ
zLjLS&6@au?xV}OeTz+v~N9O=la>FVi54{;DA_@Cr^+N+H;ZS@V`gq4twK@nGrq3WE
zhoo8}?U}Gh!M1FTZNXdr>Sw@L@*^|_{eYctu8<c~>U*ucro+R-s5aR&G$axZvqek}
zk)=Yhc+T{bRmo5?`)UiS-v2C&YL3c!WOkn1aY7VrwWr`Lshn8%ftn0B+-0XV$&J~Z
zwo0BHEqZM$<kGTf*kafrAgb<Lm{O-pjHFJQr&1bh?H|>;3{y)w+}kxag_C(b3L8W1
z1mbdJ9aD|MEn3y$cEDn4JipQPE>RmxKZlLpWxs|ysIDcQD{kBABlVF}xat-1SLcpL
z7!#O&kPQX^5$Zakd$$kLC}_!Qr;>O7*AYJthx;IiN?uXtHd46)^#kgY2?%bS^%+At
z7m1o6&uIN+*DaqjoH@^F2p+x$V-F)St95wh4a6CP5x0|O!=RH{2n)f9kzNewW5W=2
zpCa2rW8_%goA&YG$oSG4rKE49&$^D~2t#IK$_`_B`Wqo-q6A{HBw3@Rl**!r{A2GA
z^g==(+Ks7Rfua@?S6}eIHiG5=E`goRrt!xgic@QXhbo4m!X+t+N8T!gDkV$Hkx+hc
zD@zPXBs|3#Vp(xZM(?<J6Dtp-WrUUh`KlKQ<dy0k`B*lW-(XVjWHr2NL#AK)W|+!&
ziM`<TaqLbF&b79msRrBPYpvj>Pq&;^^6IAgiT40*pYyO;z7JZ|RSQ+$Z6wgL=+ovP
zagVhA{MF8UnuRV&KqR*RdDAbY@OCVM0MjLJ?%5`TWCrQQ`}cDnhK5xW4W;>V%~73H
zMr7hGRfd>q?L`FF9`9koQ%c`p0RXcjziM6vs31=w_B9=yEKKQNDeA{2xGHfuvnZGI
z(ls17pWoiFAG)kvYh9N7lzBW8WS=f8TY~?lFCxG`-3*8<0sFp}`3}9s(o4amWIM26
zDRb~9a#Zm3aTGw0Qas8Jim`GzZ56g@O_AtRW{PkM2IrcK=DF-N+pIHAywzDp_Ypa=
z>rbBHR6O;8>ltiM5^p(+pXezOn$U1UylWF2dF3YS#e$LxkrlE%s!RG9Ia9`)Zf#`N
zDi+~9MC>37GcMyFa@kWyI}>m9JLLbp9jDhjKNK#h-4Acc6QCVL6sb5&CaqPs1V)&u
z2a~04=R~9Y+PZC~^^5ZRKC(BSb0qTc<SEKf;=D_{)+^ta6cc<8`rohneaO}dCi1y!
z3?rqqe0_5rPvKJxJsjjp3%>d8AehIraXzrx@K8RJ8KbCIb?*mEuTLAY@7AIkX56u5
z0Zz<?_l?Ql?4KN8G#p4*!-!ob>%4(gK0wq(^m|NHSDSM>uq@I`KfIDBxqpK!?a8=6
zYdW;eH+t(hHR(HW?$|bP9CMn+DoF~v)T^SSB}_7NUc!?OkHs+Th=}K1VS;v3EDtZ=
zi`htI#;iu=cO?)_<63T7!!f6edhw3qVtESlJs8vCeeV^XQ=Eb>baqNoF!{)mmTKIZ
z!v=fwY0o?D2l3t-y?$kn>2(X*D78oiQv-AxN8*uJUXah#0jcXnR>#28J<vi<Pu#Ti
zgsr5w7^o{JigUcsL^tbT9^#>VJx117kVvMk2Y&k&%Mg5~_O)MF%7D_Dt`H4k{V`Jy
zx(L1kz;K({OI-tTAMhfgX8s7lBvhs!-1Hhn&><4*1@r`-U{_1TvmHayz8F`{k8F{)
z0fz`Kr^MZo9x$7{AvQ+Q4*Tu-RxUujzpVT8qfr=W=NQR3aQ0|UB2A>~{sc$IEJr78
zYH>fNjt6Cl#2)S%4g)_^+*gU-nR)^yf6#+6Y73PsR!tf&#+(b(;s2w59%{T+4CK7n
z13CuZe9@xHPYT%R^~bj-Z&8c}UWlhsUag(T>@bhk=8j*y47qh-a(wh>Wf=JmeWC9|
zuA_d(4qbBB3U5;uoOs&lR~jk_Jo0-#P4Zf=8@4|j?*an}Res7#_|5hr>70I>SdQLj
z{^cGp#<q+?v$uzcUu&t;MQx~Dptr(qXjxkH+E*{de^FtA>q4ViHpy+v;IP^P*otOV
z**@u<Hk!&0NKn{{L^h0L`pZjWuqJLC!Wb&}J&iV#TP$sQ$)P9+RZ?Mgk%`Q}P<-o>
z_M2z6QK$2{Bs(8#mkK*hx+k$J2wiLMm}KxO&j2x#niOXXt>W3io}w^UDScf>;Ia*f
zH3=KMcxdHLm6clA3zh;5nnD3Dnk(TU6-!9#4<gWI10R`$LX>g@@Q@*T(VG<fua}Bp
z2S6x8?n^@UuglH$r%SP42iA+H&l!^YEc;F(@_{rljzIODbtSx0L%qzoprLnu0H1!w
z;7Aj0Tv?hYzSZ5+1`p$23e}fn{%#r`AFC59<U%PhO57+#9g<T&)6_VhYEMcwM91lp
zV!NDXVx!<P7I0WG9(qow3`e_PYp(P)$eNSKW?Ll8f}IxV)3Qn_?oeMx?S%9}i{Xjb
zzS$2Vvo=(BUNqK|w&fe1C1R4-T2e=<g+v_wM%X6~(lilPn2Fu+aoCZEuKUWT`-YgA
z6Eww_z>oEO1tTU_Q@S8?D~k)=rI!YVmtAL=mkfH>AyoVhhfH}|H(0<ixsfz^-wk81
zPxGXVl%L|H=+^b~oTd;UrqT<@nLnsn$E2{N3@sRtX#nL#)45U;SylL2`7R`R9#jBg
zYR<!*x~G{IMW8oyc7RYK!z;I812`7ey9~3jFY;sGJtcN*lK*$1ICYv`6>=q`ZheJO
z{>v!6FGs#g4V@Ta=<HY73R!#i*v!JQ7ew1l?6$H6;Hel0*^vVO)IPax@N@KdB&gI7
zlAb(C<K-Vt7+I^Zl}h(8Oy5hNZ3OKpY@Ifad?EmPl5+Zsc1yH#0**ezeNi*C`|O%w
z^GfyIRjLaYXUe5$UrANkE5S-#?D4AkXIV8n6k*2}NUJP69<b{8VL0P_n|r8_kOZ8^
z;E-(MOXo!T!BQ**s9ku?xS(#Ci^4hiZ6v0Ggm(qa`)Iev@?HW=9_#BHPJ7B2JINcg
zQ1?-Rdi~Lf5VH`gHcgUf?thkXT84r8*%8;kyyR})U%b$wWNI?V1J2;#5_aFm6&rL^
zNcy<oO<D33dM({(fYB8;o^%79^Vu<*nB$BFp7*i|$lVJ9JJekfGNTYG_Y*nr(A#{_
z_TDdX{f^vP=KFl6SFS$QBWb@!ei9J!P7$9g-69RCacfUp1EgW2-EPi)=b?C>1I%-)
zax}rTa>{B0TOd@9*IlWpN^}K6EVC;hISd%9ig=29GND7{u7k&{N>LMqIS%V+-Q*ay
z<Kzc0;6@ris>VI*^mkx6oz5jv&y~w0pilwqe0H3Io1poeoj|g^>&^d$V6c<4!SH@c
zQgjkYY2)b;#aixKu?rjuN8#4yR$`%_l_DE8>hb(~V`^lRBAZe*KiWuj3J8j_BFv!<
za#|oK_YH7p92T>O+N5siFfHo%<7Mc?S+=R^O_`H#XlwJiyXi#k^={L$#0tT1bk^%&
z>e{flZh({v7-7l<a9)iDS)7!l5<>%aX7EgU;zL!NBM(cVN5rLRfwS3F1ie!%xtJ+k
z&Tgng<61RJNEKyDOY@n}vVs_u0*jMoq-xu$Gq)noTDs6}Gh+z&B{r<?QNMg=#4$<|
zUfwg{y51Vwe0ig-8AYM{ntsFeCP&udvF&b+RaGQxhLO1E@XDCu36^TCYvxKr=b%pR
z22G`7Q+l8}w*3uTIP%%B8J(YxH{wY}XJEB5p4dL{x+)N{Q|J=Hx?4MSbcl;OvCwa@
zj|YnPq$>9M$@wX{x}djt?frPP#KDx=bETEmx^&NdwlNsm$B4irFIjy`wz4BmpD!pg
zuVmHde3%zCXC>kpK)1bT{BuyGco<6TR7|Ll(<SI(bo+aXqlRy?N9k*|ni}CGZfR70
z+x{e<HD%AWYxpEO#3$C-FoM$Pz$9i8oq+y)>TuwM`T1tw+-7WPd*muv4%Js_(IDos
zUhx&tngZ@;biE?!gNu!&l2L%HvZi0MsyAsGkR{-vbg=$S#_=bb^eB1_UbuGuN2)v>
z3!mg|;tQ%Eb$Igtl&@e}Wg>Y5(plRx@J>IKF_6&()9xqyC-cDO<<mfx@7;U%;#;NP
z1BYxUm;xlC#pc$&ysRuZGMq0%{3eWV;_%~}4=T$x1%|`<B6Cu5S7lCpSkPVQ@>KMS
z);^@s&k0G6@v)Oi<M!EY=_h3ZJJ_>7b8Jqh^GTB;+aJ6arG&@Ur?2`(PXeS6x#T{s
zo!ng?)cgf*bQ@oZ;1T$KEqzBH_`FIeG!<htHQk0DKUn>Yptp%qz5HOwXo8-cZjb)|
z6U73^rX$<;qkGyzQynbus}9QVqM{2_?gh+iL(a#Z@bzzrTC~>7X=?D_WJhxDa_w$_
z$YhD3+3=ck<$ap|x{m;vorVvbM+}fwwCAR{wW?H!^d0`>VrLvZOMFzauG~oA|F4Uy
z`m_soC1NG*?A+z&)<{3Z0mQmG6uq+|8N_mS9nhZ+!{3sCAdhLk?6@;322vjQj)rk*
z$f^JVFL1TRQjPr(WI43_LlKEDxd?<C0bqf0h^RfoU%pN(w6aJlqcov8m}#i%TXf4H
zM^){M&?WxsBo-uuud!iq<?4sT(x?hos)SdxS64f+744%JIvEY|_rPhtBSmJcb3P*z
z&oq=s1!esR^s#bbHug4NVkJudfBLe?X_d0#fV<v<xRz_h6OW-1-6$VI9G0bD_J-zN
znaOWlf^jqa0@*;eUWe8eJgS+YCVX{ER1^RG@4x)FDSvzhA}&z}pszjz>h{;o8<isV
z7FAIHh6f<mZxpx?{CRQ6&b&8fy{?`zno)@eM%9M1n+dp+JllmRLF!Dg6q5U2`~PJa
zsg$1fN@n>^(7}EQ7|%9TgxzSHRQD7W(}c9iTfu(<pirK}flg^_;72@f8|BD4!_nY@
z1>;8&bnTfyd+*3sAaL|ytpKUMJ%n5p7%!*eA9{KW^mhzd_|^rm9bGd?w)kj@kwT)a
zKEqp0`)nSK2*b&%-z6azDk2J8{dNF^VXWh<z)DS+|H(9Ob$46$D%44hS!g~7hKlR3
zXDtf%T-ePIoibo^>gXZ1A)dA0d8#{9AcIT`>##0><=2m>G^h-!AH3%<Nx4l2M&NdC
z2?-*8>#qGg)1aS@IkUR?cW^8w$I;@v-~xF^L8~fjg0Yj-XFg|W)hLTCWX6^2sop>_
z1q&|iV;fgBHg1$aKIwD2mn>Z$F8pVQ&8EPoDv64p(*fGa<hZqC?T`N!-2zZ*VI@Mp
zC1B*5J#(K>*@j+JPGWJD%6v>vd~zTp0`_A_H+sQP5$KwPTk)Ef+l9*sC6L4-mZt^K
z6=4ta^&7ADcSERs<u$X~o!HO>aRLxWtA`X-NsdhrTC{)`m`HZ=T_+YG$wUwtw!d{@
zm&ZJ{mV+vxjdKdY)g4-RT)<GzZSVeJ4SY+lcQIn(gtq!Gi*YwU<+lqzsfiH++%W-G
z-TthvNiYyIpw$93_YVq*Xw!lA>f+i#I1lPX5{VVGr<WwmgOCr{Ek*2da`pntK~S<d
zou%;KiaF1^fLNNJx*DQ2rt2@_{AoNAN@fWQ3Xl@8oPjzcski5d4Y~)oGN2QLx=D&Q
z)%gz13F%7PI|Ro~j#6rQn>6b6IF3%~eqp$#`YQyFU=7CJC?e)<)J*Ib;5_&c=cz#@
z4KBl8(#ws%<ui)1<{Jcoi^B4}W#%w%$fV{!o$LdH-Mmh+oK4J+So+P}=@+)#cx?PD
zVSN5gt6bzOa@YnOPsjqu**FgKvw@}%@RY#|zxfz1DD}jg>V-LwsYyF3zM9b_m9_A=
zi@l)5JB~i?guo#@Y47Cdwx(Kyar**H0IKij7H1|k0>TBO;tWhG1_lT*j3uWsJjq*J
zAmPHZ-mzqXb{4vOj;Tcu`m-U7CeGNRF+I>GRK8E|ie}0U77f>jkuB*&S?69*oJnJl
zT<O!Ai?1tMafhHvuSCJIGbh@RjE&=U^@s^Dl3hMs!_KgC0Y7O(e_~N&>O<0eA`C%4
z^owu#)Jy%tt733=t5RzYU`Z7xPh9&2xn5D_hT>+=tiFqojH;al!3yRd#~+345b5?a
zWXj>`JYmyH2-}_f0qc9nd8ut$^TjCylYt-lHH6PCeeG3u8AAkI<jy-y29gNh*<N}=
z!5C>4-ZPwnLfQeGzl+dD#6+=9(cQ^kB+KWli<)ZYVM0dN5FzF#@Oxpxx$m!FGW$q;
zwugfWa=&3IrY08jiXJ$*9qU0sGapBaC})Fc;y^&)2oPaol+l{5V!FNUNKhgpdnjd)
z14KD$BS|1!zD4Ctr<Y@TK5wJDeky;lC_tszufx<?b0P)UdL3PV<Osc`CHt8macr(A
zC<B~hU<iI0dzvF#S8crxa%9hxJyC)R%=Rx>%|N6h0lG%vMa}h)8o|>2zrN{5_H9I9
zgi{b<9FT>|23p!45o@4%G>Fq~5&3?!eWN2i4et*7?+apgyK0B#(u$(NA>v~Y`@tW>
z4p1}(;j3C-OR%qWN>lUjR1maT6cFkZZzi5}PCxm>5kf~S!{F%{2fj{h%HT;Y?h|nz
zv)QeI-9WXb*yENrB;Q+5nl6cM9h-KN!D{HSh`fR`?^1X&uRVnzEByN91T5OZ`IY5t
zazs^2+Yu!HPlHiIkxaHJ^vC$dVTMQfvzzq~9Hcx1cRN@!zJk~v`GK}qBC5~S<5mVi
zRNP@Ose+;3K?OcKpOStN^gQt>n-5_?UOVyYdq-M96<1N0N80n&i*OlDI`GR)a_=sj
z1BP@C6AV%(*4vifkEgcvFKGb)`Pb9X%iXSg(05~#$X}Vm`(l`lVV8O$#hivnc{G0Q
z0sc<~Xxa?TY3FzmfW__SJj~p*{Gp>Qj<0J)ASqxV+bNAB_!?DF{i>v<P&6K4%4XQz
z2}#kBeT*Jvs~q~)!50iP?T&4}8M9DN2pS_&ydzv1$jizk!GoI6iSY}?*MI4tLA&&i
z%;5T+`w!UpM6YcV1^PSm;&GTJL^kkrjeo@aC}3;~9o#f+gkQSse5GWB-XOW76RCIi
zU8wq?H&!L`3@;LujjdoFYWAWz<4i_~1`uo3Id%TY-`St2#dC@duw$YN09Sb!OINBd
zCQw{MSZ9GYtC+|chbp)@g+FL*%}AkIJ**Y$BY+n;x*kt?*-?GGM=h+g?J7KTKe^PS
zz}JA)s-xkA(#dKPg%<f^sS<eZMWv+JH|t}#R`Aoo@Q%~!`N4tnI6!=3g>j&jYDQi8
zo7A6##)XQ?*DLxs_=-_cyz-19#V&3qRG+1z(ayL&{Kc8mu&P?NzM6#wtj5Y(-w?J8
zm<g;uF$|CjWo~41baG{3Z3<;>WN%_>3N$x3GaxV^Z(?c+JUj|7Ol59obZ9XkF*Y(V
z3NK7$ZfA68G9WTAFfud>FHB`_XLM*YATSCqOl59obZ8(lFgP|cARr(hAPO%=X>4?5
zav(28Y+-a|L}g=dWMv9IJ_>Vma%Ev{3V7PAd1G{CUAJaz+c~jq+qTUUyJDv*wry8z
ztAeW7wr!*9{l30^d))3{-S@{iW3P#4;+b=eJ;tUWQC4FRF?TezkaTo#V_;!q<^^b|
zYAKu8*_oIF9pxEh+)V5Mf5jYd6cl1E7A9^$M+b2eHw#{XmW4S$+`<gN$_ilN;o*U!
z0Ejs{dAR_stla?Ae=%t2=;;4b@{b9?)aySqe?47+Rt^BlzYiW3c8*T=77lKIA^x8o
z)hsLkZq^n6OQ4+vKuk$lM@CT!KrN-H0g$qAuyFZXgtEJ-9ncIQ4>YrIaJ8TTSUS1@
z?EZZKm^nI_1OF+_mGLi@h%3MZ;Ob;y2K?)0;b~^!^pA=j;AG)q4|H|?`wRfO0<2t2
z9Nhl)z|9c=bTG4XH~%Mrzj8~*e<kGP;`rCX{;%#Yn6jg*o2!`%(8&$(H>$F@<iGQD
zvo>-2C$=l_uNvTJ`Pb6i(aimylK!Rp3-ed&W&(6@1-Mywy8RQ&)B<1*bak>b@%kJ4
zFPM`H@Ly@Ty8<1o{xbu5fQyBdiHo_Ng{$jdn7`Qnbmu?o1pIH+n>abydHu`V@n2T|
zD+i#PtA(8<BOD9s-<W1@e`8w#9pIS$85J1^OGf|;^S|Zh?oR)q^RRIFSA(eknHieD
zNtl>BI@oyu%q=Y8m=qn|{)Pll|G%-!_}@qJe*?w;Ed>9!(EI;K?tkUze_P`J-+lhC
z(30+Uc8VtU7XNYr{Kp*wkO%y8$pFp%4|5ZHpq<zM&EdaVYFYd{OH~UicRLf8|Em5s
zyoiI<Uq%>s7&-ne1-eQCJuS?Yfo^8j0810QzrFgGT*JZK!o?2gVDXpJf3*|9z{1S@
zUv%o$Kr>s1f11znZ<U3E`F~CSFHQeSo=H(#PEJ>w?*F`3|7EBAmwz{PFDDDY{|us~
z;AsB84*#HtiaL4%d>B}`m;nrI%p8B$?(gF9u(SC6Un2esh~+=e3MOtYKu>@kGb1xI
z3*hhnfBgRQWbj{LBpl2f&HtGgH8&Fn^S?v(zXboN&D>pF{&M@Tz5iSEf1Lj{?iLoF
z7G`iOi;iY|p*HDR8E$phdBfx4dINnd;KN}~WsT~^g@HAWy#eXr+dNgq@3mo!O_RKz
z?OqF~P9OWSv<IVoSawvssTP+71b$>R{e`F-l$~632f<8+Wr!JPTG2jN^YU8|x}40K
z2Rq{`s)nWSP=u46Y%VZYPc(sK9)*Dz<d05B{$^<nahP2wO(1A>MW<(!Nykr*R6}DU
z!@qiOAp7_6>2l*>DfmpV{Zmd!lHI(FUG5vrJ|SN{xSS{9Uwd%Vt;64=s^w1gym5kW
z)V2iLB<LOorBtpZOH;ce+s+cTG!@>@+s$<FXjkQsn1Iy<dw+m@8E$FS62n?u)YIe}
z_xBco5svaW;u+aAga`1s;M*dFNnvh!+SvPC5iOsiMj;x9wBi8x0_}%xD3v5caEyB$
zmwOLU!JcT|G%|NryxUUbK)C4Es$}Z#(!>vQ4BNIvU!KK$?g(JpBlf#uSP$Q{G!BOG
znAVQiKx}U+k69gkQhhqYuWzsd4O9-2jZwZE3QsWy=$^C-qOe>P)j{Rt7{h)u8D7Aj
zqM-yc_*@%$+jQ6S+^VHa5Gqzg_0Oji(iD#Y-ER(5Zlwlk>|{juV!$@JRBN-t)C|2$
z=Y$R)+YllN$F5DCT{Fdt4pyE(mWtiLt#fZy|9%jl_;V>Y0v*AbwrXn4;=SwVBR{R?
zWO0;34)DOZ8+9Qg+)?C~xmp9!w-*JHM%mz(5vlpcfTA(I)j&W&c+(4<hOp4j*?_@-
zlh<wX_9qwaSATh04;0OeN17-Ns#uT)M-MJH<RIAzeVD!-pBHTSEOI}DGuFA~eImIX
z5ST~mw(`*A=Hhi|E5Xd8T0JHL^zaj4-^fo^mDY_U*l}_+&<kwQk#UWdC85+aZP`{S
zLh1klJeD6&qv-f?Q@+1vu5=pC00KJYRE3j5NSzNqG=A)U?ABtQ?C8#BTWob5dEDP@
z54W~iffl?Dl<vdb0-Q<Yc;_WsR0!IxCn*JKN>x`-JKBVoNS<4t4ouIGc5pUy#Si<}
zv?xsG^H^k{V$n2v_^B|^SLwu`7Z|t;S?}D%E33;_%k6kwfOu9NJ;Yx7q-n_^zl_33
z_)ygGxF!%HBjol|KtGbxqd2X)O~L5HzuwBFal;)V^tb(dc<&n)GMKkEqOvnMj=08W
zP8FXDUPZ%g4O>CS38-z3f23tIk-Ju!GOT&HM6)z<&|)4obx_O*8Q|)@;bBLCawm4h
zogY1rF8rY2EqBg^xh0<S1H00zq&0#z&{}@3exYJcph(W{BIG6M0Znc@ZG>A_(Fey?
z)9M>{YK~OKarVo2;O6E^=2>-ylfv2w(9ZRS=W4#`;>I&es76+6zU9BeGf~#3K}O9R
zG!)YXLgRus>2oxein`zq;(HRy^;t_lUH98@u}k^aKX4(=%x2)!Fc@=rjt<m)Q1}qm
zct}9wSAUqEh~ir5X+MUJa(hyJn;CQXLzJx`VrW^s9?`zlG4KD~n3#hC3_J#jUZzXO
zylVv%zqWhA6vm7tC6Uk&wH))a5GsPeVjlAWu9V8kSqc+$L%~mB{X_4T)6(T9dfJzB
z_IHw4&`V~=h9UU7SnkzPQexqBO^)==2m{%<yjHI#CdI5DH@Jse$BniZ0e#vodA6z&
zgQ}Q=0!sIy_i&H$tw56yA|tx#s6*Bm2kA<xJ-g5Ew^HDwD;nP%@a-hnXci72*sfH3
zravG?gnCi5%j5m$xdn)#-6JqP6(7{$!wP<0nmur{{c@upx}CrC=A-0^B-EXKRzgDO
z0f`hcQt+78cXr1(Alq(T0f+wb>YECi^g1MKY&)TnO21=lme4hhOih>8_U=qMTX$+=
zYZYN>9^+U-IN5X;_P+F*`nKk-a^caO(vNGV9-j`ul1IckqhRRkLN(&(HAG2R9bA(L
zWyDmn0D9}}R>I7uro`8wFrA#JWfOVi<U88N|48WcqoCFw%B+@v=~}IyD^E-gCMoH?
z$h1qRBIQ$^xiP!>Ji*o(^s8v_z=pS&A{2+PT%wQ(i}6M92o|uXc|f`K=-B=adx>`~
zYKi-4B-}LYh9UNeUDp2|!f6q*^;|$v&Bc>HgW#QuB!HRlsNZIktz?!ursHW9@M}B1
zlqWQ~u>|P~O3#9dUv1s(ejv^4Dg|arZ}Ecu3rjsRd6op&<dY)NYsyeeVKbS=_qOro
zRuBdiUnWSrra~xJ<V;!Ce*Y}+Jx_Bq;09!>Bqh~iA9;_g6w3EYkpu+7k4?%Op~{PU
zg=CKBh7h_dU&njN*2Zt@`C|NW94{L%r4gLR^}rwA!NR#sl-~tu<f_GGq~x|CuhOb~
zXG_4Ucus1Ro1xj0kzWwVx+O};O7gbV>r3*;_))SvCwal?m|{GD+zKewDFr5{tfx+<
zOaE5XN>8Sk(JNcmw3czTpX}DfbEe^MR*(EH!7!lg!V`D%ShtOE^mM@g+Hm6O;u_n=
z$(gltvn!Z6J|hebG*SsBk|nEb<mJHuTnyWKNlxTSfYAme$>SOY8on>OU4lj))qET2
z<q+THmUy&0D*gG4D25lm@c}XsPM&d2ab6}A*D*z8^mgW^_=h9r6v_}`=d{o6%%=>e
zT8VT8kyvYc^m9nxZ7j`W4C>DcTE>jyGE~0~5A+Vr^_FkKCT>!r<<`oy33(!musG(2
zLg9)iA7ac!9$EL%W;2{wl#Z|G<?lY0YUnt5WFBMM0Pygwsu*d#Rx}17H;|TrV^*4|
z_-a;M;}VGP2~8@dZ(&Ki$6t1NQ7jjn;)LRX_;Hc)``WVN#$e{~rXvR~8L@DkrFl>s
z^He}{06Rd$zjh3aDodCldls6*re$0Y<n_j+9bp$#Rpu+1dW<c{4IF(<mWx#-<*$^j
z1^^sXP$iXFt9f9yt4;u1*wzehTKdgqveQ@o^jikj&w2LX;LXp~`e>WOX3n9@NLf(r
z)t}+_KF?S<2)|iERASgS`Bw%H(@xQc$YQIW?K8kI@&kMED-?o1j*UrmTeZ1#sdar7
z;(0lJBO(*^aJs@x;BjC?iVnG<l>46?tyNYN6Jlkbsup+aAr++_;pna-tM8h)+riS=
z44Te;T_h1#`bcuW!Iam24AW<FNagZl1c(28M#3MXo^^xOQt@uUJ1S=tSPBlepF=|H
zw*56BGJU377OqN17q1isqvydS?W@?*&)O*<6}6CM>?lYO=-%<>{bezOGk~_~hEs6v
z0w?66EyhunCPZi3&EqW~=+(u3$<3LdTh7(i*CK^Qq|ZuU4p7AaF(8xN=gP1fmzJ6W
z;MSA){Mqok18ds=akHBGfm${G2kK0>Sy|et8cgC%P70#YvA_BTKW>+@Cb>qp{rj8)
zmcg6CFxHptl)t2-aW+*2MnHKVO5U5L_KwgV>!Z%}Os;g)N8y8>So5iBzme|MW(V_T
z=QF$zu_er^Rp>r~rvl@T9yc=fsc@;5-@c2ly5LRBWuLc5DPpyeCGfi+hiEEK8VFu#
zPxa{EpC);VZbV9}XGKCoEi(;9jPeD*?M_*t0EUi}6a`%=AQC~QjqpYNX(T>;dSTW=
z9>lFJwJeN))s+I%ax75OC$tj?k_^hVKrc7Yiu@-g2r*L>NU8k3he&Nefuva={2!#o
zM>cb_N`h%BWVsHPK!hKM9=Gd=vTDkzo}hfL)Amp%zMtxhZV9y^)6Y-a#X&`K3Ae*0
zM_oOxU;{aPU>=lSP=n3&?J`RM@ani!nIm-NfI8UXc@ix}#Q0M&$tImBuvgkZ1sYVo
z2em>vJ)1u5)5FUF-n;Uw&v^5Nb7X4NlU=nM6qsg-!v*_1L^Mzjq~=3%K5EcUYh|{P
zon?jveCmniF^QXe&hK(ZR#a|6iI%;EXx*YwRvmP<lgH3sPN!=IoJcbnyTDQSFsG;@
zS~WY6j@VGQhoR6LW{({AMcjy*6F?O{J{Vdp|D{A)R%zyELX{Wnu88;~uXG#RiN<yd
ztH_K|P~HoG0`&fwRBd9+!;XmG{N%&Ga{lV8$M-SvF>$KyT0+NT#~)r6{0#E)<Y!tw
zdS6L&fI#ISyXefdPPeF1Lm<B6)EW2zgD}$P=ipOiv5UBF-2?nn<ok#}`<$(Jm&b6O
zM)m$!m`3L0BU{J0Oy+8BOhY^lZf<klH%#@QJex|^Y=icBziW70qgbskXliV$f=UZf
z`P8@xmyEdAKV#i2o3-sHotj9L(7`>AF2pgwEX#Cxq+}`75qD<r>Ev(gdt<kq&{Blc
zVG5C;w;V&lJlG5(sfb(|->F<0L>4#T!4ImxIBlgMeCV$`f2?1S|A}ROF3@gDev<v*
z2d$6ghq#9wS;QMu^=M01;B{Za-3RQ;IPLfoogBZ)?SyGovS1-^UQwWSXNHB<5|oy3
ztEzOi&YZ9_g1_2v$Vw%0C<<80gle6+?x3@Cro39mmjX7v^DS`7Qg4cetN^{n_}&S^
zd7zbCUp`FWJv*b`6WI`c@MZJc)4F+Ev!++E85@12k4&TLB5OpMybUvAOPRj-5EY}{
z13CO#4<W|TPW`G8oeRLp1^z@zP~KnD^^ExN+&`k})dBU}w3cR&V%bUk15yMliYsKe
zY)7H)D=*DGB3;i>+-BELFLd?emgZWC;Z4w3<lj=hX*HHH+CckfJnwq>G}Gs6{t<7M
z$nT$XCp8ho>Ltd7N8aw);f=o#${2PdK~u+ZGD$Y`GZRo^5f9Mha>_Hy>78hbN_|9@
zAShTyL_oA{$r<l)y|y*8;h?fno~Fn(?b_1-<l@p~u=q|DF+MElc+bgNgx7$JIoOUx
zkBcAT(pk^l7QkxmF-K0wu#x&}nC(Z&kf&2`;v?D2<0{D&)qs_!6Y9^j!>S=eYrqMQ
zgD%pGoYDlXmhia0!Ux)53bjOi+w@oVojRKAJY_5)6NV;)Q2{wBQ!p^JUXh|`SC@mJ
z|K(J6){2GVL!<*Er+=9blVoOifBi`HfD-wn0BSIDdxO4aRE$3p)6A9Tpdv}KNtxns
zkk$XNdg6fRojq`E&IQ)u#3bXvok9#rI<!fw@QdUO4V#PW!KKLkru8_;Mn=jSJbcy<
z`_$uZS`(BuiQ*o4IwN8_Q}n>86zbiY!q0mu@q69&=v?G)^o2W|1eo~YIlAkVi*N*_
z3oxe4^hag_)%os`=8b9>_-Il*l>om<a_Kk~r1;#KrQzCNr<(WkLRk)^$vfRF<6L;i
zCrEv3c?==-<%`KJI1!OPlkTWj%rmtV?&2eGTGa-V;55Qn<MBJqVpXtF8M3`8y@7Al
z#;1WqvV0w5!?a;_uY83`;i8cwzJooQP^`PB=3m6o&>>JVqE%aTY!IE<`p;REd^y&H
zzv*4)Yc)SeM!MYMDpDzJr7JQRx^7{?=Bm3JlrkhKgz_+Y$Wp%_s%kp8?I;<Q2cL!f
zi111$35z)VjBcDk%!am^mZn1m^bL{+rt9d=*BLlZxn1|SIy`{BUZ9mbge+S>NyHD<
zK4FqWqQB94nS!R-)_qyK3a>*$Oo;ZpgXL8#DfD<gaYO}5{a$T>Qc%M*+l|Wf@i1pt
z^dUp3*K+I62>ICrnO`8t5_bSI^DG!x$iO)r7hiEu|C>x@x63^?#^P4f9{Q7V==OYZ
z`SX}U=Q;{A5fB!Ptc|%ic-#Uv&`C7sNSP1O!K2;~K7qZJ)B<T(Zmnu?^v&b?B1SYW
z8V{VskNg|HU|uj!sC8@FHdV=C=dTTEn~wMK(o?9z;;<z<@DW5Er{V8sH9R~c22_R=
zg#>H#xLkEV?C|A{I~%j1$l0F3zol9mV5!zldre@u$^OwttS*<RqzPbJdn7!tMDEN2
zc&^41-3S?h5P>v?)gBA@VH$sqjM|}sMsh&~-(VO-w2Rne7Zga=h}kOkKZzP+1bB6T
z%0YfhqF)Mnzt=P$t7IhbQZozOdYQwDmVOufNyzfHhVk1ko1--pGYNmp(C3UEltO$N
z>&=Iu(sbC19R>KJ2YEtptApb!8}RELqq6OUQc35W*+V7=1Siez55y6DNn~D)s4<&U
z^mREarHbTy^amreq=Gg}90C0V6k`$rm6P7jH_8@}jz(X8RRdeJ(Jm3*OEMs-a<+`1
z)6S^^4*G6U)N&@TK*|eDc~KD$QfL*TT*>8O>~0h$Yq@AJft{wY<fa{_b3YkhCwn1v
zH>Ythh+m1f+!H4gM7bMG#y~rgJO@-i^INCd7J^wwyim|7ITF9e%dwfFj(TMQb+)oe
zz}?w&3xOT4Io8a$+#$6kjWQulcilBd2)8bW{s$ya!Vb+o`?1&Z0`AKyOfPaSW@VhH
zcbPWmm09MiX<kYzo9OV~c%K;4ISqC>S*Y_*onKh!{o@NC#EKvL@~Se`8OKGaddXi?
z0)eO5U8+9e)}kV@q=vCxP#(6;KTnD6Bj0S<z^pwF>Twl8s~>qol(x-u_B#-kW=R9|
zHMlTH?CO#E4DkCBWFxw#XZD01WNErB5hpnN5DJA6Aqsr+>!rIkLk|UxzxOY`B!>M$
zUk<s8y+<edSwQr)a3%TyM2Nlz1(z-Q6VNGo%2_mu8-+wPv%e*Zvtzr$)AIBF;6Qof
zsxgK{TOEcrgSomsSP69P)>Vni89f)w{f3#I_hEy>zz{t1Q-uRE?{F1qs%g)I$+Ql$
z4Pn}bGuesZM}8qzd2Alpx=CnV%QX;apT{)T&NIfUXv(J{U>D!%1J}NWa_vz5!C|+l
z&cOBlkprHH9THJAYSe7N7F-0Ha@cL^M_qIH-P0%CS=dO-MjB7mQ8Vzd8grf!w$ST%
zZyp$zXbA2fyh8_u55}&PbGW=)ULXG3?DJl;%n$={TagAk<Gq5c_V24GaK6VF0@j(g
z>f$Zi>A6^+Mi}3}`<j{O2EAcQlF}F+Ri~J6uDBXcFTXX$#%0(baRqy~voT@}?~1bv
za<{-e;A<qTQSHkQgr@s=x$cQi7IXq-vXDG)A1X#QH-8fhv3I-BaM(#FPZR?D1mO=H
zg%|#+L5lNxwbwC)6d9n`o)$56aWPm=8AhjIf;=ILfDs`xGx&v?A9>u8>=&`fXQt7m
zf2(hX`#OEnQE6W;V_fqTM@O3QBOd{_Rk`Jj?|t#>*~<tae{cd9yB$hZ+i+v&KKCnU
zR0Hj50#JkZinRVdL3~RXb)vM^u5wnf?gbsOPTJd6S&BO7=vs?gDb?II{lOxaP$5mL
ztCU0j*99gDyMnyV0{50dr|qio0Qru9L3XboA{9jo62JD-59-)FwhJk%U$;;|;i1K#
zrsr>P=!@J_vq<)9RGZ})Y{oy~*lhY6fM@6RH0lLa<Hgl0!(ie)n{MRKOz92ynY?VJ
zk54Gq`AO_omWMg0O<yT)+3X=CbqR-j*}c;kJL{_W-vO45t!no>vB_?~3`RT4MpZ2{
z2G*+hR-8v@SRd-uR@#E$c*3HGCd9~H{Z~=~W~a4x%$K7y4__v0Oy1g*2h+K{qu5~E
zi<vTer9XhP)K8@6=!+8q(&OWJ%2>bp$tbjYlBO2+JsA?igk}4Z<sR$ZB|Cn}>sh85
z*fDX|KF8$^QF?2TaP)OQSagEuH)CSMpOfF9`s3=Mu{f7zF^C=`PGi`bY?#TbrRHu;
z41lRCS^7V$Gc&7kYH2?}vLB{d>#bp)Trq%^Jz!CHboBxenr|1k6Vu@K1&QA@g9<`M
z++P}xe7t94rtACO8a7JcoW?lK)oK#gNUA|DC<b?s$@O&18!{_=b9KP4Esx~1#_~-*
zc`xZN?{K#S?VfuZ%J(_r??cLUocFz8OI}Vb5v>kIa3n3LnWu{D%KK_yX1X<Rd;~5N
zMQERXxD^XQa_tcYv0Wl+^A<I^7hs;~Fp(IEEP$o3G%P9C$graubSEva*s;kXlOJHF
z2ePz#8-0bA?gOkS(Ln>-a)$05Uq~DY#O_7pe9>CaNmGLEK~Q<w6CSpGJM@e8<dVhx
zD*$6rqt!u$uZ@vGdd`CRa5O7dpK>>@o{i)9`fzSd9%A5{6GI(op>Ruf@zZkN;|7NV
z3=iL_c=F87zmyJ5KJ?Vuk#ye#x~ibdU-APQCG-dmlWcB%gv>1{(sSL3$cEMLavMcX
zq$R7<OP0e!cCS6{0u-MIN(x1f<v$dD!EK!)(4ZHsE$3qad%5r}-Ko;y;V#g?%+R1O
zJ?~Lp%3#=&`!dqsu<HeWxXyqGu?3%L6Lo;o3%>CrQ653ALw$Ld$S%d<O4B@#5ODn#
zbaScfEwGCmji#YrK%^j<_A1ijp|O`fHdt%%=_c9#fL+Pm4w#PBZ2tj8*X+X8*$8RF
zdE(EL={>C6L@yKTjnYLU%+h4ZaoJaJ)h%GGLQ}VLah`^He+G&!c&~b^-2NNIF)K9_
z6RDz5+-7{<Xi*ValAJ0$oV|HAvF6^w8`U1SCBA~P<j)|Nsz{Of;|#W(Rc?t0F@dQH
z+9D+9zS9RbDuPX{2El8L9fnSJI8{5wS+$<?^o60hHE#8z^7G~#LW4D45jg(oHCCt*
zT-{F@&!oVe1jdNkaQHjiO!eh?op?WjcmLq;OLFp9KS1b2mhX=7rA$tC98HgSaovQI
zX_AMK`I(b#h0ro*X9Nld>n|td)zDx-3XAIuH;nEJ32&C_HdJ^eI3n#v*KO2Zp&Tq{
zT3<N0_tN{PAQB*#*y0<%Gc-Yll=Bjw@%jD6dXLUMQVgA?ZE+Ea$^1bWnXgXzh*528
zCYJBd7{^xz%NHN;nF5}mPcgCY*RI6_8-o%dUUa5Q0|LSRJJA@1w1(CR!^bGE=QZ~r
zQ4UQtZfQi!6k(eMYx2Oz1&U`w;21M#zW@aOx|opL_)k@c8mVF{DDVDCT_Pun4}R9n
zW-Sv69R_J;SOlNphcRwR(lFfjoRbhgXE@E9LX0f6o(epD93>Gxg8GKBggrg{=+z>g
zmo7mq{e8B^Me2uMi*+11lYQaYKYUJ>7*lm7k(i0_n(^7mz5E13<0u{wJ>)Z#*%s`^
z^gIv&(;dNF*&_Igs;C}t))G1e1L*x8?W2LqT1c@nu0_f#G7=~H{djrO{vzBjQ`w|Y
znDeedvyoI00o424bR>eNU;USk)VP_vW6ovXwW;gk31n*MQI|ifDU^?ylgJToaevX<
z&ujW(xZBGHKs-@@-p+6i<MGB~<h~iCu!CD6>`Da&6BoE!0MFk-s@Eyvz1)Wad;;a|
z)K&LndA6%Cgd4OsaVB0;-)=Bdq?W-Hdm%Or-+baFO{`|B+)`m08b!d)YDhoxF7KNB
zDn&onKXz_IiVS$D5vOszvm7!$7gSH$Jq?Hoci=&CrBh=^<}&C`lz2IXfv2Bx$Y;57
zYx=H8eXQ~{ma?KclEwnpW&0H;qIbisV5wu-v`?w_`A(IVF`6|}))V*aZzO4sh~QEv
zA*^CIt`R#3kqJgAu;rw7#_#dw0+~3{h}cmr_mVY)*_GCynfojYEaHyxsM_6}@xN^I
z_+yF0oEqtwe;r7=rrt!M-%zTagGu)-pLG#})Q7c@OovPUUL1oPph)^DEBMFM^*o3@
z;9asJbnpq**E^>i@=|Df=N^uOg}kyM2L<<=40Za!RJblo%TjaD=UW4-M2FxkGn%=l
zOy$6L3g}i2H0&WdmA<q~<HNi=dNn_b)|r%y^Zu6nkZXnHxy>TsAsm3uSaelY2rFz8
z>(Tf6>1P$qN~@&NuLmc~fYJmH3H;CGbPn|9o$-?HSsoLU#^g=cr3x*))aB#USe$Uz
zh{)|ZI(<M**9&-_(ipD{n=)9a*zjq`6JLK*hZ?nx9%~+8YKP8}1~0;Q9#s=es%csz
zp6MzhnW1<e7D7XM*CI`I=&-qH&?-)$xBK;@5fN=Wn#*52cC9X0#O<Z$O9JWL@{8qs
zj`2XUt4mEr6sP8y3GD-0(gcJh$HsP6$3%1dbEgz`H#8rnmGp8eqMAl&RqSZ8$Wmxs
z6%r1iCj&$)22CJuF0%2pkjMYR+jPr6LbkNRsU@PUioH&^p+vLgE*ydTjZB8?2=R=i
zu{;>uM`$I8;)I4p4H2`arTsOhoA*VCXL$=(lj)Iitd5?gu3Q^2GiAqY`R=OIb-DXi
zSXEsFWTVH?2@(3`vI^O>7p|Agk9v})37L4&49kl~h<5wVW$jWqHvCFqO|tcuWpC+L
z@*(xJFsazjU2Zx~ig4^B_|fWBE*&n53h`i+prCi1cI>6cJJ7K#3#En8KS}R$YF{Si
zq#oGhy1kR%;WI<_12xYSp{w5R4#~zq%;)I+{V#}=YtV_~Xu+kA;L?iCsDjs$PJU$~
z(-YeM<UFarZepMK{OR+K7zkX|_e8c{doil3nVWU-FbZF;a&_AeJ7Jnt<gu+JGN!2|
z`=C0<rjH({?+jG2J0<lBO2Sh~<_fu{DW|PysX8cK`AMMeS`Dcngb`(PR?6POLnIYg
z0u#(wF=mZW-LJ9SN<5I^<=&*LRk}kg=MvoAueMVVF-9&27jD6%OY}?KaelLsHn<RW
zSH#sr_n1pI#K|xxq43RKre&oZt|K24`8usg`6x9F#VQexF#u<gQ9Pf6AZRz@+wxES
zsrMF$F6YKs;jKnqYdz6BP20eix1}EuclJ)fo&iX|RvheSGcbsdi&A?O0OOF<{=4=e
zF$4rzMC`0I93<|K32Mf=Tst;Q`-gqYdtHLPCjkLs>}%5_=xtqVr``lIzq=-Yyz6W{
zhk^a6X(q(G83JjaLAK7#fcoEdMDStw{QUfZLsk-ptmlK=^rfvX#}z>k_rF*D6BTu<
zcN?m2V8>mP)=a@LtOj#L4T&xq=|1>Tki0G26AOLPKJ?(;iDGR(n!yY*lGO<_q(cm<
z+)K8Ui*%jbx9R6|bQ`Nz#`N{<n(Y2iq}q7q72f%|Q{Bn!s9{c71i7<by6W>=>29wR
z)~x7B;dRW6o%OP7bwP13;uSlS;C)9%_*^Byn1U+1?>5rxH~E1HT%SRe{!NI2$+uNi
zGAp6`a&|?TGuHL|tsrN4ZUYW;lY+nuo&G6kJAKODtOfUd@pm}tw5>Kz|5D*0jX>Go
zF(qRCh>jV$fZmfUK=1iP%G%XWp<d2`C!<^hAMRZn$*x9h%679hx$dEf-PTJhlWH{h
z40)iUsV79??vcCuJ%6#+1k0D5v!cV+o|x0LNhk10B5qCOyT}G<JK~p8f%PDopr!l9
ztv$_S{YNvy8*M<%{|ao!!`%TlWv(-!lB*hI6J_ivtX@gh7P=M>fyw`}Bt)X?t8$C*
z>>|(nGd76mDh!4|FQvLKmFz7PswxGg>YG*9jx9qj)M|l$1R>>*iTyWYlx@Q_FQ-87
zC_tc6JChnS9?LYg`!1gzCk9+^F;?lLztS-q3bAh+k#8Zr%6gFz_Ot=w?MhH+o;FB5
z!hXl&4NN4kuvJ(a-j}+?a$=OJKH+CK(U*qNeKKG4t;~aUP~CgJ0~ZKP$bKgX=hC}D
z5<@7DnA}XWdj!=)3+qi8CjO|i7ry3wXvgOt>5^q!q&2Lvej^CK&g%T$&t|K!kBu<v
zGpatn>~oFPTL$DQO>4ZF2-AT-c#YWBsaBpE#ifrWZOvUqni7tMUl1QTE0U$ch5dKk
z)9~N5!C4Ub-m$i}TIlIFuk~Hdot1k}H!Yp_Pz=}$gTtht+iGV)257RcG(U+>%$_&#
zJ#V@&=y1f1vs#|nqg)`i^}E&Ik4#7N5L<MQr-`L31{E5p+StT__JLCp9A#2#C>0Bn
zT)U3~`0Q_&gf;C`gkAO^ds@aM{B*j&l=uu~9Nius`-%{g2fv_dpA`1pS>+I}_Jg7T
zxJC%xP4KOzH=g;I3b{q;#x@|Y9j_nMBnEwwo47{YkKsm9<oIxl&W1g}44GsB`7C+c
zU1Oi4<|N>J)!=~|B(^&S8}QJgKZFurgJAF**G2e4YN$BSdLBCWk|{f5Fc}@2qW*_F
zUA6PLis7=HFg3KAN!_*<eJxD&Uqs}PBws8uQfj}PL-e0Mp?_b}kSZ_G@V(c%b5|D4
zdhfUrMl0(<0|VB>1u+W|kUILN!l}+7u)m?vFqpqAaVXO2+NdVn3E^vZfrp!fAt`LK
z*O2in<-~-Sz>>lpyfy*34qkj+c*OgdX6~bR<2X6jSbVw~4rmVwLc}5Ke8sVjK1{y9
zjWhzHx`d1C#-=~7QpGw^gIw*e2{Rl)-$A8g4+hb}`jK0Z_Xr`O-nIfRTq%HrPTlp<
zVWzPxsB>t=vdEuBYYTjL#Yu=$_|RhX>m6LlAq$z^!uFMDhRSnO<j8JTd>nRAb|js0
zZN<>LM}%8<4MPb%;%o<~+JPz8x*OMdTnrsSNmW2!N>UusDmWw3w{Z>6@Nyh<jU-G-
zre2vt^-gmn<HNE~T}@y-Tp5NU2|Nlve*X0VDWtaKA}4uXLzJ)fGTN5F$R6lLGkrL!
z+HsZE56Q^o#oPI4uYP`y?hcK7zH}D{0%EjMiSl>{N)D=qs%eh`*M;mEqs6PU3HiDX
zDnniJlXr0yyK|>Ru<sA|ycX@<k4#ND1c!?~?oW8Q9-RhtwX(43Ayqfdyf#;I9|v_e
z?}3HXywE4M1DP}=Mk*4Q+rfS-4|plv%WtP%bS=z0+Lfq|ZjQY_K?Udmbo=BY#JU>P
zbm&*5@<=n06z_?!FcBJzdNZ#1wxI(Tw~iz}9OE!O#O(CF4zb2*rRK#$jZ{%Uwrl=x
zk$6Ff6W?<z5eK;~T8{>_nOKrnyDjlq7$QihBl6XTX$u<F_=xhp>Eifuam~$8U&e@e
z)4rjKgse(5rB8<zWii?I+cV^ew@_@$y7L96#W@fG#Z7$E)gxp3#xa8Pe62DP+2_EX
zu@Bvzp&nX+dQ2Ru7(AQd%sV7zR10SoOX_Hc@Aog&zz`p?b0bE4e}63xmJt>^P6L%8
zgS^u$e|zLpE4nqkio((33WgSBM15BqFcx2sABPzd)zon}CF6g#OmLLM5x-Sn=e`>6
zhp!8!Q-ml_v^#0O1OidW%~ZUJIZ?#>R=DYEu`40kz@BWIPY})F8HT||G@E^LxsNSk
zh|BcMSuR>Kjj<WubT*Iuc(N>Dk@XGPb6d4R=9ZYRE3qLd(Hf|e;(pP?otnFsd=h)?
zQfypV1Fg^<%%OE-T6RfWR*{ONNUh3IoqV#jWX1DG>uDBu#_90#LpY{t+jpwCYg{6{
zp8X=)#e$dwoITT1S548<H*E=^d;_dL@+$TY*+H&F-X8SAtnL=Eq}5`tI;_IWb#dvS
zPR$*(4aJBdoT+ax{7MvBfm&ijO-T<^Ys!=ysCT<$v>E$a3-RR6EO4XayGx@vbJva3
zGI4H$Kw#O^rKT0McC)W+W0WcEgy-w{(z7x0CVh*Wa_+98py>VULI5tziwsro^xl<;
zTvSW4#}z3_|F%f05#95_n*<9RQg9s2;461>96zx?k2djY^c77^hh4pUOo3L#sycTQ
zF)e$z`18yVSMVs;B?T*>(97vUGkw0m(m*F2)JV>f&6ZPr>i2!}bOOyT7+yP<D#6Vu
zkjxd?)0vw!05o24vy}-cUVGnV^@4d_0(B(8qM1caRo=Ded1W15I-5f!vIwx`BPkwc
zpu<B_FyeXjQ-%LC4Xr3rZ(^T3pURr3WE~r^Ef~omff;waZAE5xgD)IGQ6(!bV-je2
zRC2y;^0rB*QC<g5X!R$wDk2J9mhD@h1ace*yU|X1P1xiJWZ_xFlIyqn{bBErWV0P9
zwg(6~x3HO+>nU*4=~7SN2dSr|6sO#KO_ExtyjR@YjRH>Lu-cXC)<U;YIc4?LNuT$p
z-(a9;!Zc6KWVMho9IRHZ*_ovQp~V?y$k_lU;Psv|@EbjTnR}5mW^O94HRLNa9zi&G
zGZho;qtIXo{pDLNiK79Fai{AVmA2SbATteGHL8s-i%r`v<m<*|$2E$BN&cCg`A7}_
zO**h*{r)Zw)%GI`sbK$)c&|8-tEZ|L#yumq=TAS3)+V_gWn<U?U8Syda(u0Oumps7
zT=p1&v1nL*PbRO%)Iu8i@6G<F-!D?`g79WFw>~YkVNvMH(D8NC(tr5KDjR#!kZ?z4
zF-S(Iy25{+_+#c{o77rIXE;!pxZB1h^6tq6ci^nXQP9ew_lGh?o#~SQ&o5+pe8t4+
z`g2aH$p@>85I0&ZC1gvJC@>(>rExIAWqh-15b6HSg)2X*xK>dJ7qBun0H+2h-7*HK
zD)*g$o-bo_c+A*1JHHs5ZrT|Y$UHn!Bw?fg?)H@Ocr*DI0*N5o%zg}8{MO1cr57eq
z*Sf`+hV=Pn+k|`~Qr&N?<UazfD@YYOs8!q0gP3okLY8X~^sXEeG}8b9HCn7y_m#jH
z$mlbewb4BK&K*l%SkMFS=_ujHVR#1)WcQ!;>NfedtE#RGhm~M=K1`O8xkg8Y69PQs
zp-7Zm&_{gfg+CSokq3%@Pn+)xDj)mf>m-Xn2;(1rN_9@fi1X5Z?|z@}-VOhLGVZ*-
zP(}=ncHr;fXkU8jFpXlyEHS*siEAjCW<y`Dgr*+q+f0T#%cm!6tKXFb=RE!S+ULhm
zw)q=1d?fj`$-#b6OR6GhbnQo-0a8zfsoAxA^H)*$?a$vHGJ7|8gw@8DFk%g^buUZc
z8N%d6v*)<9_6SvcSLpcCzlzxPJkeg58ISuGJFbymnG4QjDsp9^RW>BRT7)S&3u~__
zS9Y`xweA4XuhIK6Uq$Vrn-FNG*NTq4e@vza5<`#28qOLn4ygD={=9qk>@$XYGe{z2
z@mH3|^W4hF@rgqX6o30ObLHNYxFx=w#;7W6KYLm;ChJT3X@|Y$h!DTW1cEKv>V=J=
ze);`R2U6$@6oP4C&P<*+3>M?^o1Km#z8y->?YerfZzl8eI66=9R^pSGA!@;KQ}4@Z
z^Wm|EU;qLq$C6&4o)SCd@bC^J3_8z5x7g!{Gp?>j(j^3roFsohsabOaCVIgMO~sTD
zxu<G}=7O1Dej!A%23uEm2T=urbkdYq9l-S$9wSP2@jBaO;^4zzyiRoU^cU~#cyyF0
zQEvV64vg~ko@W->APvHdTF^pc<gUvtypW|&8FXm%{-f-q7#?*t2Dn7xxys`>50;!o
zzMZ3S?cIu;%u|p;>JNifEL_Ok;M?Oj%}~<GyCFiisFll?m&V-=%<=ReI<;4#&2-I!
zvSQ4>Jc?KpPlrg+4`w06w@zZSf*=lms2SZJxQx$t2T&h$4TuSTsmON|E7YBSH-G!d
z(J#SIhM!>}Ztx3+oX}ST$y$mFUT^GFksjRFTfX1;+ups=sGE!}>Ra-eo@Q(fH&Rb+
zpP3T@zYJ`v!5q_|Q{0(#pnqK&7B1|jekX;Znpc&~%g2I%L;)zJRwrtVs+Iw-JpCHN
zX6x8<I#On9<dG_88fZ{zL>B?`bwQUbA-E7+SDD@`lrYf+{G@+u;NU31QAf(Onr3r+
zKGx*>L4{6GZ=@SEp6)KYsYS&~(}tM%G<-9<UUX%i;H@vv$Oiawq&(t<=*`QW@J(^U
zZGa<clCIbB7iZgAc%Vq_INR#RumY_=B_Y3|t{W<QJ?KokS!oUAHP7C}xGi$B3mpS%
ziRG(|lY(gXW|1z{PiCmc1B8WbCh+v#5d#%Lr5mKG=jgQIY5e(U%$elY;<MuY$=zgs
zC6l9btEoFx7p%J<cf?rFKefmPzGZY)HXhzVoufUH(MdhEt4PS?=RO_qpXX2bqWDo~
zQ;sF{O-{sCrMAa#%`6B)%ZLe2{Latesgh@6wB#t`_5HC&+ES!8*-YbJvpU@z!YoOI
z>C6H-+F^#XHJ?&6z4E%DCX{8+FCw2l^kJ!vYNJl*Z%jj2l{a>)X9NEOW$sMrD{NkP
z%>SIO)rylU9@8bIu1Sj}>W-xSA{`uhjkD|yKAG;ih|=}!`Z?d01=_krF>xYZ>K(JL
zsbCyRhMpyFY#P(#=&M-WB0cderlj%Wu35FPth#jwB~gyy)5^6Id$VVsl*afR({`Ot
zZa;Mqn~U)(;w2(8(_B=fSiOJi?5il6Y02*_iH)>)mbA$b%{K$o3@^PYtJsq15NsU?
zn+yvjmj`yxU`)8t_lfcw+7!j~WYG0FJXQZ>{h|U%BWq(+Mc~L%N3djzo4*I-SNGT1
zG?+d4LURdvoBS{;xb3Bn?N(2;>(LN3L~LA0U@cb1>#<ayo>WgjUu?-S%kU=oW3zVd
za2OED|6P|qYci9uCy;?iIZx$nIU6?1wYu{5wu@x<j1YWTu3*C`LAU!6S+77iJ>$tM
zekVy=^gcVHt1mUsC^cDrx<B_RKlNx@);RvyiRV|H$Bhl|em*wX{@V9pCSMVierk^-
zZ#a(lsp!cHmZI2hURykvWhh8fzTNl{a=&mQIWI#Nu-{cCquKS?;a&*|)v!Q&5U$%V
z-X!CtmKr)w!#e3aC<un_@_qZl+J}q;ufFv>;J5DSwNYm~rity5gsz{XWKjf@Q#cWD
zAs<OSf_kv8;1XOn%8b|b_1~Y1)(eH>k)9*@ohnoIfG{L@%@v3`0)hE3eDs(}%fX#C
zs?8z$B6OF`NAshyXJue3Um!6YAa+O<l732YFKzI0iTEAu7(NB3%erN5m}{gxH7qk^
zX}Ep*tI%9ZVno9X^^)cr>}`<ife#t1X#<y1KRG%}st;>!ANpblLBl`Fj%1@%s&~>V
zcRHYS#*_6eVZ!l=<g2}n8i-K>C6=9=IPh3bV-dr8ZGE-B7bJL|c0(iGk8wEZ(}T~x
ziLaAzjTBnV>1f~xk1%SRiMR&PQZJ&?x^M0@7>uZ8Zq7C&kJw=BK#tGeaSwY(=tYow
z>Xr?y%D|#fe?#5iFK7VkiNf^I#~j`}=Zv6enRWYR%c`sb>_ypnWM)aXU>D+J!at?C
zPi!%XS&@_FTc$lUym3Jj8-%#T_Yr`=hhe4BavtE!&<=ZqG}wFB0u&CjZ4+`#&bae9
zS2EL?DT#DMb5GXcb;Mjgg(f3GV9cxReS~6f>NzWQ`)#ZjHh+aF42&>BEoXIz4Q5!s
zeQnqozC|0xoST$vJ@PlS@$G!{no@lDuf@#g=M3g4l}#6!AL><pcm$|CWL|h@`9Npf
zGM_#AnVYr^Oxa%!A3~5337X8fL|Fyh#s5L$-Pi`B&spp#UrBe9j5+7v?lTX3ZQ#7M
zy*Jd^h=rCniLA;w)BfEeQ0V8_k5`yOs?$%`e7)B_kFC;p1Y4;6ryOu5t^<pBMobz!
z7{2+3p;dA+Kk|5>ia+#@I3us5dj$;8!b{`~4UsDUe6VN;u=I(AL7gt`MgFr`k~~Lb
zN{=9+z3$yy6ZSaQWPBK_PMn&Y76_W>st)2;zc=ih)}<ULnZD<1*=VdDY_0MVX;Ty3
z&fjTcM*P(ZoE5p^gN6@hcIace=UR2j`8B-C(5c}?vqTl(eT|@W%Q8@ewLPc$q}?gg
zS%PS_BO@*$%cO+E!uJi+9ayA2nULIto2nSJxM0biN4`#Df2ibE{$`&d2nB8r#?oc)
z=~_@MI#{wrn4rqHdbdnrQs3Zn2}w-1{^OD{^5_$aR3p2h0p}3=+4zds953PRmx)FR
zK4;G9VDiAOKLA`TZb>JDDo8QX0U8GN2=zOY$@LPuj=z09J7a{OIBhOcA7nhSEBqB_
z6w)bO{;VheF<W`*c$!EvTdQ+8d7M~;ld34EDOGI(QiUPfb1(jZCH_#nZ$n1(4g#Ay
zv9l;}mbzldJ2xp_XvkkBXREqiT98LdMqG9!#st%*6QUgI9(p(paixSQFw*K9!I$Ef
zk-Clw2|CV8eJS)oW7ASgkULzK&K<UmzAp2dr@{tA8s08Rlcp9l$z>;PcqxHb%mWX)
zxBBqNkJi9f2a^+uN4QoqkXRCr<^AOkhhAfEq__{HZEd~j%qQ5)tl$cv9er#8a0bJ$
z!@g4koB|rY+}`q&Bd-gYlk6&t0H@usYz;i+ESukS)#QivX4dF6hB_4(zLr?Xw2Tn?
z11^>QYmlULH$u1)Ng`_YHL%}Xazm;m9&wCZ%eG?LE&8=i>#%%QiE5sfS=`*W9jhs}
z0X`@v`3MtkxhlD(%jVp-dVeSxn;omsk<G0;2RLB-Ka3_?6>VH{<8s;+^w~N&$D4=m
z^HP=CDG5E~6+HD@`F?$WIS}dD8M7Y8M!b8U0{Mxg^{1?Os(+it@@Jj*vQNx4Mi)%~
zOnV?kr8<c-@Z`cE^(kV-uPTYb$VZ8kAp-?LuYi=}q@c@oP_-AF5!_T%GQvI?Ej5CY
zpGpQW{o1sZzih7c-P_d-9qmiiG-wb$%%Ng~`hFcEj|D1Ab4cmtVkYJ)I#nWw%72})
zlLEa*ss@aO3=@HCzGU0rW=i&h!8v76K~F>*W`_>CIcXOE4^x7E%%j&dafMsAlk1hN
zc7?$k*^Wft7?NFl-Ja*K>%inh>ym4B9R-M?R?Ea6F*lNlcMjX1);sfEF=uoY1x;<s
zYH=2@=#XSIMy?6A@Y<(g(O~5^#1LOaK6ZUSIo>&Ac%0AcA&W-W8|YdHST|r>*cK|M
z6jEX@y|CIIA5H9t)`YY4?V?eMcO+$nA<Qc%%BjuR>*T;un4?J*bN~PZgH1XA*zew;
zoJy0$%j9G@p)xsBWc4cPF{~`nNc6cv#W-q9F=a~L49PolE~Lit5xkV7segc%sf!)2
z`QQl6i$|s52j?WMs<0by_!zOa^jCF|7QYRg*d^E+A(_xHutzU0S2ntdm9-qBKaX1{
z9*+-uUO$_&WZg{i+33cqgl;1l4-#)~;@!j2@-PeJ7qvx=$1HQq*^eR0XqXd&&Z?oy
zN}ZY~T@6P{(6qsp4rw|NOZF=CRnDi$<T<x5lcqAO`F-7+8qc^3X(^|eH0cf=zRkhe
zh}?u6-7B~R)vwz+`TRytX5AL)(K(6i{zk~U#<yNL$Z@u`bK7)OOo~x)Q9<CncZ-l2
z(y-KKA}NW9H;O%FMap;nZFCMyJf{(n(rT|?TaF#-hJ3Q~^V~Z5!#e}rPEG2zz4X4F
zu4W~Qze8_*U<hF|{wj&}%M#m2Y5JX;@F@-?w4=N!N!7{-vABWAd(<@WQVU9k==Pi@
zXlB2jF)i88S*InAHe?R9B-4Za!T`#{YMZdsJ*t>Ayy*x74?Txz&AjGdlc%Zh&L#jE
z{g!;IXHpvTHG)z<mj!b;W)pXV>Lan!?sS<M+?YYBmrGyS1MT>0b|rm2?{P%P!jQ*r
zrLRjE9OyQhltp{ZxbVb)U|%nfW*l`Ur1ToHoV-+Oi<7oyHzcL2P^c?q>s*Np8h%A+
z7aYM|z=%gC<t$K|p4fy~V$6G6P}G86>X|+@JV1s113tPvS~RjF%_i@h((g0yen`+q
zz`5m}bn>`cW*SpkEoql4N8bmH_S{J-x1VwgVQ}u2=}!~4sA2ysr1zf-gAC;3oLy6F
z=gm$R0Qjd?-~q}_;4uGm%$}w0h)6N{0`+;9gSxv3xwif{E;6?IA+TQ>{zZTw7O$=S
zGA%|7bcRtjMo2c{bB*QV`St<3_-;|sP5#^)^P)w`ru^osBgf3`EHOX`i>Cc=2sD0l
z`c6zWYCa7*o=_l`<wIc(CCO<gF|H1YY%4Zg9aphi6z*Z1wVG(dCkJ{~L2I^6Rx#^%
z6=XP@kvT}A*{c0^&TGqjh<?;?m|gqm8gpm{5B2XM>;-ep+80f<(XQj&jg&|JaT~Z8
zHL7|peO*X@-qn2mo>_{oF&*$iM_krW(Lx&U?pc}ywUI>8M5v+gh=A7=hSr_WNb1P!
zlcPGK{^ipvzVWNBt7LZB$rzh_$Cb1$t6k_l@M;Z=iWXK{wh(H}Q+OhFhQsE)y-=&+
zj{S&yZ7{#iW!xT;#SuHyo3hK!uu!EAPPU3p9Y2%nb~8i=1gW-YJjS*hmL7{5?LYA2
z$WPZRvpH(Df=77t=RjoziVH@ilr>76zvDssrh|g^JAhUGrya%;G858E-%Q9KE#2i2
zTKE0op7}Ene3MW)UjA{eC`gK5!8tF^vnFEL+h9M7-uV&LY6Tpsytg9I#}|pcrdE#1
zED$*&=3-Bx;&mGY9DH_7<w8KIOu|>FtY_eMOIEAtxsNfM?%SX_-#8-5^y$m{cT1f>
zyi?!CqA52!CXp9@zp?BNY|M!^iognLRi!ZRhlKO^K{R`UfF*cu)HJ`>>2rmLum#;B
z`+r0>*RM(rTz{1@_ZOoIM6citA{~yDoG;Sb?a!8nvrlqblV%R)I1U{!mp^%JIL%cr
zna2s^pxdR4x5=WKiwX&*PMR|BT5<C?ZbRw52rkB<B{TXh@p+^<$K+75p>EnpifV6V
z6t|8}EW%7O4BwkwV?hS<K|Q4Vamda@t0+Mu{63In%`hs+{Lp5!iFY-e!KBRe(u5kC
zAlR0~7wxwWwQ=81sgKfN3a_oujHK0l{rrMv3uUAV^?+3HSv7`G%B5$Pl=o<7Y4>b8
zi?_TguT>qnl#%$e-V;_n$5_E)pBdhsTh3Z}>>uODS2>XVW90hU%32LY{u(=!b|rkV
z{+j>xhs>EMLPCPmMVpBMww^BvdSgt*VA1@7d15FjSbwQN;HG#3#ZOmiWpjShsd47f
z?*35Y1m=6rc(7&gO#Dm^?7ez5iv*`DN~Jr3j+;)%AHCqxw`{1ts~<LLOYhZ+`0|6H
zczk&|Bb{S%=Nq~`6`(zf$>poo((x!OTbF;!AcnRbZ=@E5(<GpGwh%b(4*3p4W({K)
z=2$Hi^e&HI${t99q-P30NF(sqDA-wFnz;EYF#O6fuHk2YuqLk@XDQ=}nj#t|`VBtO
zlb^O7i`90H_kei)XC0<$_=9vE=49p<50XRqg3s?PHO$R~l{DU0gwBO>qrmsQgLh!>
zY{VLM;g)$XG;BPtaAy99HF<u*{|J5%f$yhVS4H1;JW=&x41%+YTm43VC?EyRKV{GL
z>o;n{L~zI}Bx4&A><X&uw3eV=s&Vi~9P8_jet-g(jRs`)lKT!BLf>9Pmr9geL4`6T
zB&x`pbNC(|D4T=kRu~z(oADG}(|x`Tdf68h=&K!*u2;1OmmOY>h8J^!s|n_Psa${j
zY-+=(4r6kQ<J124^a%%sei*BDp$l2~+;c4JMwbbOSV)P4Ay8P>$!TZq0NnQ<OB6lF
zstEoYE74Gka{dK1A&OEcjgx$>np{+l9<30He(l;KE^i|*V-X{;3?Ok+DEMSgG%mD|
zO&<tcl0~_L465?SoA=y+s}4bxfaor9LSVN?*I3)L3Gnmtn{f7wFa#!b7CjYg`*8oh
zEaDiB!NZx#UK9k$U^NV93AFIKcK1P>`c>Q0>F&I5??p<@00w%)TE?8TuBljGp7_x(
z{}GsF&=C}k=&WN6vdYz|GWHtA?|B>oD03*m6emcJSSt=LO^*(G6ay#&cZg|VqddG=
zo{*)E5aHP&Bi0jfY}kpEO|BdT8K&06YL(xgQbGv;L|cY!6fZ?+ZT>#&t{V`u%s0fq
zPAKi!Co9}+)0jp{BGhwDG-u3#Qjk~Kr7Eor#|8MJeJ_&-|0(xXd*J6e7(ds;05AiX
z%f0lZpt8xp7Y1c^FfxkVt~EVA&kCO~!?=G&MJRHP6foSy^yICwZXx$)V+qzjNy(4+
z%*0%2vy=Y-23RtlTjMXu!6N%e*e1$q7~7Sjl8fofok11nN1_c|)o535ds64u$%qUJ
zQZmFm7cWaM%`lqemSD$h5e6_%XEtLE1WD6Zb^0}4-Fbrk&9{Oi9i~UAh536BScuXn
zUy(u7_m)s7@oIC8T32@%M_xW*?%A#GwF&M&+12AEuc}aH&tGkv7FOmZpz|hv@P{|t
zyN7KC75b7q@=Yt}y62(h0;+D}5A(79)%03Yl^uR5bE8*R^JzuieB^(jTuOHNsI?qh
z8dT&VkE5=OVNJP$aiMp*%T;<baa5AYJkwdpA)gUt_or?~Cm(>!80xvg=J9NcA65sn
zgX3a6Qjx7mTl&kwz?P2Ddeb#FHHbxfD;D<W6--wE*ZQXOGO5>(2rR+oNnxEYu)-R$
z^PsBoG{wQ@tA#=BB^ucVGJj`PIzMfnr24xtGN5fnUtRWH4bR|@gK#7B$8xw}t_-l&
z7}gX{aCn&Y;fjbnFkyit)@HN2{8V5wC{n#vPo-Vx0^;lL+rGgNH(f_WTt7B4tioh?
z^G6gVV6}(W4>PD}e;tCOE0XJ@31Z<WEc>X#ch#_mq+DyQ|1K-T!_QDZsX%=1qF)`1
zJGk^0b_Z)w!dfkG+02BL1pGBLUQ^IjZMDbn<gP=MTss$i65HxIrYVGK93VZRbSLq{
z1V+L#OcAjn*_#lCZRhBS8-H9<;^+5QF;b;@&W1$W;jGb7SvuxD!3#Ij2KG>fN|$*?
zat17ceCuS>09V1oD{Xv!LS}Yem)$jdXjvF_ASSrAxKb`T4>>TZN>HG@s*PyYhx9f9
z%IRDdL1oqY-S<dsK#9CQ{9w5SC0Yw8^ZQy@9L)b_pLPL90zB3(ii7Wx)3KEDj$(cw
z16if2m(IB@h*VO=3)30OL0e7JDTJ1?-n$FAx@rj0=;s9@r0uNq-NK{(W3&9<5t>*r
zN`+YWBmkaMaiI-;H&A+MHI@NhO1kWzG8ofJUao>@2YL!3q<HTN6dGkcn0q4T=}J58
zRf=P9Sjbe#jcy_O+O=hZTr0C#^&o6N6d17}X0>I)q?&{GzKCeTl+cAQho`9Kfqg1R
z{aVTQ;_^dCY&+Yn_q<i<r&T+zAxTCptAH;?=~sxM0cD^1@HpQ2fG{xIiX4!p7g<)t
zP61o}Z$Fr#o1+L8U^-`1$D0atCJgv(dIVm|A<`~F2caCIx`91kUbb4i+{@KxQof=5
z;PuC&R<eeI{?3(5m7k1nZ>p3r985o}@Bcn?6NKJCVYtl_AT(k=29H&%P-anX8;3p5
zu2OB!wy=wlkqj7`1CAIFBz1V(gfYJMfpv|4NX`$;$=ogCP;|sc8NyzJgzjqPxW&9z
z(~6f_&gBV8^GV`Zgcl@%I@u_Bn8q_@^0nfb(xD&3iA)ZBjYBZ{LT&<SQ!?3~gVr|)
z<Vbn))EXMZ;P!%VkOOMc%or$LJWbJWK|~~s7XCvl$3QtLCJqWq7F9Wv&zhoe8aItX
z8Jrae4!flzD>E^xLEtYAX0nhmr=R1YJPUFfWW+>8_AgOP|6iK_?Sc-*KfGYp|K9&q
zVx==&aaN|X4Iof7jUrmVKjfZ3lQDuGI6Q`j6_OAFteIkCiXRQDr%gwV0L4~+SH9Yj
z2(=%H%fk10MSVT9C~&HnO->Y(Ku&6@w%O(%K?7_b0u|3I_jMEMhGsSnFlHxhohe`h
zhxRY!K5I^#PEKGq=usKXEW<iOl;=@7aPMVS^b?r#L_S0;m5CPd6lj7ANl!{@E{w;@
z)CH{xhpb#FqUL0<`jo|#(^>Pln!Aey%v!{hQ%1=eBv25K^D2~qrtMovIyny~(VbLU
z0)ferR@49Oh`*22o_QA{U-(e&Ag2n)3yP8nE%h9I8)EDr0r&QjG~Z7?bk8Sl?&>77
zX?wb!-fA0xD&=T_t@wDWOu$r_a(Ls!uUjfv^P(Y5uj4+Ct@Vm#<#!1(GS)qh;9M}R
zwdH|`n?_Z=lOAL)*{<z=GxiErB_Wh})^ipj)j;Q*c4?%KSn7GuW817e4uUq}<kXA&
z<*{`s8nlP+k>XjmDhMRDqH6B}Gd)+tK}c&esB6mtz{z+X_&z}UcuqokysAG3xyay#
zn1l#u)1RdRX)X|MHCH~h*5mL+_459~T`sKN|6;z&=sz0JCvR_#V2Qj=LbxG^(?^<m
z!mDa=kQ-K$*Bo0`?`sfcZ#)cg7<{0JJ6&Wj!fkZ@p@rNPDHCp3moV#`a?CCIDG$@t
zuOwTCO#gf3VUO2qDT5lKn66Ba&oc|4O5;<ok0Ojf3i5K2n*(k`X1(_?QJ<n?zxtQ7
zFFgv>7Kf@P6x5(_GO?;))pscFrvt8Q{~Abxf+3^&9Pfr<%pWOuTMeVpIj|#{<)Bs~
zxEf;tv#3{VU>yZRK4p<y;k+=sSM7WD0WhBT=oPSFC5nr%0XP|o5w*H}5R6ADsvs`%
zzM*1kojL6ntly|dACe-pNy{$+zow_|B0-s3wz&+XNLVWN{pOU9+o<G(!jm?$88oe1
z!@<s`Xd)WfW(delp%r~FgO={ibmGuhgnw2oFu$-hZf!6~xez8nB7-YM&;%O4tFH`T
zRKON%qDp782?SRR<oN8EpYaIvmsE<jGZin5M;&-Ww-+jU$kZL8G{@9|bN2zDn>#0@
z<j8lhCzj&~VQ~QJ31Ay0sS%x(s3>-I1$OJVb{Htfvu8+3iWcOiNx-hWf?we6Rz&I~
zq@*`Lhpp+7`{|&^?tK%?6Sn$G*Vi(s-iNQ7M4^kSWw(vLe2_uhrs2qVP?Z;L&mxUA
z2LGQLKBr$U2=&Fl@v-jH#XO%#_`e*fR~zo|Xyt_c)))Dj=dky)E1?S@$F#JFwDEbi
zg&ft+QfC820bEy_kQeaNsSM_7?7dY`U2WFy8z4Y%g1fsD+%32TcbA2`yOZF-WnsbH
z-95OwySw{YZ+CzD>$5v`zV51Xu`f1Nvudu%tXdcAH=h3(&yZ)o^Fn14KBw@j1*pIN
zV-|UZwu#~@W$an5L~agM<|(>za$&qDwnobneB=0f2px=zEg|{sp6i&P1RZt(p{Lbl
zpe1Q!iX*IB6kbwEaJ`&obtc4*OCBo<0*o2UpVG7M)C9)}4;K1MDSm9~b8N0G11|#6
z7-ARsdyPHiYWrKLP^8_cM+2JQ>T!_R@{JF-1n0SCf;3(aI~*gL%^hw$Fmn*#;U^P%
z0AGXSTs0MyJ;J{>lxJq#Mb1huUnz}HBd?BYeLNhKmtl~Z7*jwm#tK<Td@NU`()c!+
zSOaSOd#bJe{xH1o39q`M9sH}Bv)s5!lkWC40qiXzEILlgx38dn(sQt2nV__+V}3cG
zAil!0`O%(e(n4J=_yB6c5Q85>k=^HvsZT$JRd7{<uTv>H%)8b<#h^rHBKu1<)TtR>
z?_ruW5ZxYfVLaF&3x5b-3=2g!6>-q+S5bom$9XR|`0d1yu=njVy1iU5qX0R7g8;Da
z3aIR7MeNC-2^y$Ork&<X-2*8h9;G%&R8rNumnOqb715xiq$;Ay6ZS%c4ffErZ<5;d
zshlkT^VMX38zGfM%63(=>4j|cX-YqIK)mZsNtLSBMcHk*{NOJJ1vECr@>!+4oqQLR
z&HSd=NDpEU1^VjEn2w=?T!+BE)!9ySqw3I0*S18RGsPP{rrA#`a;W}KRf|rO_3n^r
z&}Ux(6NHay$<V@{fju)ag3BS@%ZCHR2M~h!p9~9VkJj$j<mDmXaI+?qlFJ$gek~%)
zPWwsj4(dEX1AMQh>;&_sd=$!#WVj*aG(^ElYSXM<zmr&bq<<Z!C~&bBiO0-mCDN)c
z4kxhjrea+@Vvm&t(_BG4YqNuvAasg>D~SVKd=hJS63_66t+1!?&G*T_1kdZYSz%~*
z&^Ivj{2|aRHm8p{L*g)Lv1ZZLN=U)HZP(=>c6MO}myt;Z|4o<U7yKwm=YyFqXDK<D
zm<xYd$FU7rTxi>X_>?>xPm<IxV;ZbEgtk6(JyGsFSyj`vmjvy38#A0dp6KUYE3rxi
zTarSx4J?k8612}LDT_j0JtzlW$|+FvhR9DJ>zqy6>T)ci!<U)}rw10>5Kfa(?TJRY
zqo=WANH6V@HHCg{S-|O7S6Ee`Buo9Kz^q^%s3O0lkHhfXfUi9Yo!>-IWD6E6l~qo6
zosplQ*)G_rb_ciLS-VF9@4h5=>_Fe)om(7ptP^`Bt(_T1Z|+e#HE$SlGY>y8*Fhwg
zV{~QjJNcfkM`E}z0h+M$NTx2n9|9jGBU8~Z=3}r>LIQuAzhw4>zb@<jt2;x_s0{CP
zE`plU)@)d)%}+?$35-n^DJ)HhDnbvW5s_ioHk7>Ri;AFPkMd={{Rr3+&xujwW5F~g
zLE{$aNm;{XUCd42R=P5_+IO_i>9gA*;2-d!IgIb)<LQ)PJfB(82d?cSUQ!G-BtC7g
z(W!^an&!b*m9|Uy{@BX{a;8wZ?KkUq6A-T$zkDZkX!M7*N{fQ7adO(abQGdRCFxSM
z6GDo6ABx17;>;zo6|An%jh24Uw(QFSMAs`n-G;M#jL{t&H#M~IP_EI?&A$mRhC}3t
z&-7W!Q^rZBlekUen6`m5*3W&kHkX&uTx4E_SS3JQ*gZk(@lbXS_ENf`;STn7lv903
za{g&3^@B9ETL9fT2EPBooL#Q?)^7QdWo!$%$VIvZ2FDSKB)%VBwyz$=uyhM+j9y}<
zcTc#n3_4KXH2=|trBKrgxW4I_L?lAndFF|6xPfWI(WS{B38qZ?)3f<|hNU;ylvxT_
z<|CoJXdLQ8BjZ?5TPfh)Y!~4pXP^$`*F;T8xXimLX&zQ^q&7lm7SGOgam!(`dCGi`
zPLf(zGMpDGM$TNb$*lCo$zs*b7t&vl^=vL@sH%Gn`}@K{pRuQEOfYpLvtr!f?@Pjn
zg^R=XR!WV$9VR=W|2X%N?O*0T5^;Xv_-mLW=<k1?=a`_<WwXqQ+;&Xc82aH*h|OFS
z1sw}YVi4S6p}sIs)HaxOvItD;i}7Pj2jlt(Cmh{f41pDI^x8=4J(-fLulI6(Iia3#
zs`uQHXJf>C5uEw!{TXm&mCeYNFWBG>MVRnud3HGGi9K?aedMU!$08%7=Y={pGvv(e
zv?ucRos3Wysiy<&MB#pXQ@lA(itmoNKF^2u?$B$cphj{!X_Y*;C3E1Icht71KY72$
zE8^65p@>AsJHFg(Qox`YuYXe^WnA-EeF5H^F9UgqT;n@)U8BY-VTcck>k*_bYK`)1
zcG@ncw`)N%Lu!L<5*IKd<Tbfe)#8#*w{>J1_wgy7u3gSM8L_9N`MvO9p|&k!eBgt>
z6y<T<Q4jxF7?<y2dVuDOf#Lb;xb1#c;>~hp62qO|SPO$p%O)EcZP~oVZ2d7A{Ht#{
zwz>P_dSlw}S?d?gqf$*u+CWoKhj@{~Zw0Exn7;0*4>?)Ry;b_8V_FOg>QtRn=G6?H
zg`I^oXfiNjk@N`T-#Ind^jTM?{HlmiM;81h117(a8YZ_iRrZal+*@);ez0d-swF-l
z^fQoVJO<|^UFrPgLlo7co9@^K#(>4M5$h?fKkU2(s^J6YDJK;_b!W7<DEHh5iR+b#
zvu8XCVRP#SDu*1vkB?NkYB(_NVWu+gCB^cwYF*j<n*EMo<kleFe__J{x(Kms{XoS(
z*7f=kkM+hfc*9FsU14%d(%hpEtXkt(^5JFJA|2LP(_YPS6a#$K^hO7z>b_BJ=*Qz-
z73u=3(_y#!6G~NY0UjE@VCU$^j6cSo;I?YUP*Q2<G)S()oK0tjxmk`((nAQ0#umH6
z<A9r!l4D$zJ2H5K*e$@Fi2Yv4v4vwYz|J8VU`t}`;*1P%;2O)~#65_;mcA3Q0^E_g
zCja{*k3nqDwjKEPBw*KlEX$4CV2TGhzzuYJNe<Z3dMiLl^yP1L8zj_i-HG0xkbYyW
z*#w$Xyl?Z3{(R{&cT<bP)9fm27%|u6B0l)qDfB((?e*rgd>ReO)Pr!MDD3)%Dd=9U
z-<>xnod2f0sg36H*fQ+}Y8l~d@sGyH{;$T!{)O}Jw)pQDqq3}R4ij4IHEnCC)S-XB
z3za`jSkPk($QCV%XV+9e0_haNKmE>rKAykUosufys|}snWM;{@x7y!ZzMhjmvH`sI
zrK2&roQ#Q>yxw>&dU9}RsHMx-rxTQwS$n^7^MBvZE~$4xsd0+4UNUWV-jUV9AFpex
zev|)lH&IK7R!VAGaaYihC$%&?H|RBuZ|Tc@_G+%&-a#vVoYdjs+h2L^4U;)<%FX?%
zme+mRSMV5+_+$`Tb+`nbeU=+v9@F5wl+KhinWpD~dTrV8oL?3y9cUQp!j~tysyw|L
znK3(vfVFkH%(*z7zv{YXLB7X$w#|uPJF{5E;frIz!+I(-!k@lWRKMRYph<={rBxwn
zeQ(3eaXfHk-1kx<;+fRZeyeKcp|&Es3*X;fdh)Fv`nW7tJ6uJfly7<N-Ak+5u%Y8y
zPNzbS=lz7@a9x8`ZO`BoU#`;vx*-}0LvOkMh7B4zJQ);9O;$HJ3WPup0b+}AnDAI|
zyn_zv$D`E3CmlqF76a8CW0WsNED%5iv4jsbN|hfNJ~5^rOG$R);N>MFMNLIGZhTVj
z`&4e&|GN}o4yqGx_0^Hn1m1V_*y7<JfEVHdGhUYVVTEx}DYTR8nRX+9`6@^mNn|JD
zJ0s?ARMA$Xiyvk>V;J`5FkZia%q}E*?I_0IA?^nK%mo~SvHOV~nR*^#5i+WXOJi8Z
z8l&0`uR6vR<Qwllwo$Hlf=B&uulPA|G01e#hm`?gkbz`g6T^wUr@_RKn4~1w66F}y
z^5DZ$e9hDtE`phe$QR`~y>!pB>dL2lBQ5|npUKk7hrXug7VPr#j}&wKE5)2&SpI#A
zL6VXth}?FKzATcvBiL#QPM2U}4o3PoK~|ho+%V+tY%vH0w=MG4{%qP`q<lL;V8w@I
zYr5vSvAZ{YP2sY7akRaoTT=1HtKC00Fq?Er-P}|ZdgpNSdUN2t-*P;A4Y=h+Jnff1
zs@I%9_7PZxylS>ZN*dGC@TE+6vtEDC*N<-YzFgl%P$|y7&ZgIuzjRvYi~F%Xf4VrZ
zOc*-MDtBJhACvUz^l9{cwg6n?6<1?2*4BsjJoK&R%f06DH}zW`HoVURT&X|E*;-Op
zPM>Yi^p4OONLkcY+tYV2y&DE%Xj3`P_Xi2*Uv0i)z_C5!Fk}>5rt_|f))Z7-Z@2p@
zHM~=B)mkwWOm~T7Oielzad*I_3AaZt=cv`Z4;|&)@4oKL0)24rCPU7y&6l6-mpdNK
zvTqgXQ%$iS+n={ne<seaCbYn6uLe~#$%UEJGMTYE(j>zMqXkF6Nm7j<Fu`g<<H53T
z)NTeK1j9!FDURVIz$1PMcJ1gZFDPC|_~nQqh>g;-VbOl+ia^Igz(QaF4{0iaOViP*
z6KKf!h^(3;i`%^m!DMylo&ZJ}WT`%b)N8*`qEtTLiERW8bhJgZYLa^i@Wj<0aNa4D
zUXuof9h2LQN*d-%Hex#W;7EbgBmkr)GL^0wj*NQ%keclOT}=XJ{r*xDN05@B{!L9P
z{-!4EMiGzffd99eEXOiuU}!Oi_-L`@KRQv$hIty}_D^V3zM*--&1DAYRr)j+Kpx9y
z7wErY-&ViDuYsPTX5IpN0qI<<n58V@dyvO6g-Om7AVopo>j%~ZVp4MT<ir7$^aoQo
zV_96d2C>I~!~v`3CC7eGOOEya1a7WMj!m3{6vZi)rIlw8yNPQsWl#F=kLbt&C-mfi
z3+X}Z7e$AV3`K{0U6AHT52ieT6ovj>31(P;P*(~hC!lNGfckNIB-JrL<t3WYJZc9?
zZnQ;lXX2Asn}OWwZx|s;a71CeYeImnC-#6H-E@neoC>hudCj+#L9Pcz=O+7p<QDRW
zp1kZ09JbXl<R1-@^Ir{-{Xg;T{*RKAm(d;hfi{82p?lX|v)8NnvvI7uHpb1XB^mL8
zx72T(^$%oP6JG>QnAmlmiDuH{r_itJ<yufFzIa?qs~P&7I<D8=+L_CqOcWN>7N%J=
zw@%h@H%$t+dgD@hF4aDdX_fW9^m*eIY(_RW-|oqaScm5GeP5(x;TgSKxfc)j681Q*
zD?DAJ-!*+v&nQjywbnl0nr~Qxl|2vRDFS+(3=biicvh}S4b`cm_bt^}zm7MM)4Dvv
zvBBc8HM=6w*J*o}eB}Yst%^v6>JW<&ZwjBlLY;{zbO{;~F~iZ&5iwt|oNQLo=2Gow
zOHj|V31yZ4C|=*)sKzxFm7^ez<7Vl^Hc^IQ;(>(Q2)LMPDvu{C?NSJP+@KoegpTr4
ziP)eqjTqt9xBf6PxhfE<)PFJGS{@S3@SEj>uh3#`!1!w{uk$BhO$Rsw7OA|%Ez{Sr
z?+*_Cyr9OP51HTRcC9|&XLM2yVH;5zp3w@0aO!8mC*QHCB^%&<fo6$dT-?CM?S|#r
zjyOJ`rFehFyAAS`pYZH%qufYNzQ=Kg&x8-ujo9Qu-g}H;hx-v{kO2ug&W{VS_L#Gv
zG4$LuhZ`xdh22TxaALcWqs+Tw+<RU`)b~cj?)7ueveM=&WUsM9+aDq3`d5hm(*Wh4
z)THgd)MR%e_aAB!W*iuNN$k7s&9bdr>@nCiXY^va*9K6x;rCwdFUQsM0(#9IwFpEf
z72Vq3I=hD1*c?^0;k_~mc&tC3Jsm%WgQAnux$&cVk3vw~&<?i~n$4%p?Q8Oysr)qW
z^B~<t;QV~z+pkruY_u<;^|IQbi`YV)f%NRgZvRrv^XztiY?o5-NI#c+voapaw7ES>
z<5Y;`M5TG(Bfsy1*G_e<VTli$mR-_MI>jXpXJ3e+Hva5$Kv7kstIz#KdiM5|PVHuu
zKxD*htvwy!>xFe^RI6DeIx(HOe;B#4DjHr;bgd-NTD**OfatwEeCTvBQRq~wTQYXu
z(;V}-eReRrwLRH}<mSWUUYBdx6Sc&&<+W7jvxHE=uL&AET9U6J?sM?=I4Utb{5G`u
zvD~J6UXikoTN9f#jT~Jj807r)2a};CLSaFcK_Ein+AZ02g$P55$;CNCiGhhJWa&FF
z(O4C^i*-&GL*?ux;zS^%==VcKKt+ULgY~S;29K7PFRiyof`2NR%tzR~BEYJ@sUGq}
z?PjAp^4#J)nWdh$+=!@x7;EByuGgmY>1so$+~#(eAwMUV>$^p1+#_S2CR+(_Hb5XZ
zkfRZD3IR|ow2U-iUbBp1T{{jEq|&={bXjzUFt;v~gH#0ublpdAbndLiu%|Kid!?~Z
zM44Dm=!;UGlBe$xq*@PRxrUcviFA}<PkTLq6y^Oldqj4P{wwtzMyUW~hyLqLCg>(O
zBj6S=GyKHj#;EQq=QWr4In#M>Lc4MdF9jqk*}8Kn-x!Zsop2q;5Osk&@X>M(a{i!*
zLSk%Tmkh8%Dd*6yOUXqlPsxQU?hxWi$tBVe2h0Wu%wMs}&K?;cmGSS7YT|)a4e`J(
z#=onIht%H#6wt>l>9O91mt2%Y%KxP(_y4UZy-BR%SI`A}I@AtkA9L223T$?QvQWXG
zi4kA!AsX9clsMB4%LEH(#-@EP&KaYo-Gni{4<a4C9k-5(e!P8Xkh-<`lQD8J{i`vu
zvoQbLu3h>+E6VBr5TL|Lm9*7{#%!9H@}4h_jW#4jX3nh6JzaZJ>z*#8hYUR(-dEq-
z*i@Hu<6HCaX=p7!T^&DA>4uaqEl959A6u=mPp6YsEFakwp9bIh##TM&zh>#<+drH7
zKH(PU>R6|b_|`p7XXCvbR$dn<JigWAmFc}Ly<zvX-;<ZFr5<k3wP#_-k?@m#uH8W5
ziIAP=hjnJma|&)Pn?Cyv7P)q^lAElvw8m!52joH5x;dq$Kgv%Xv%M*evYA4*T^krA
zm_IGQPLyVL^iQ?NSu5(`vAyG+aWPwJYkZZlqg(+#z3G2el(4NssatKjU0KN-%0aJT
zlsP}TxLec8KGd<7Pq@Z<uC5kYJS%I6cOXv`t-7+(dcR%{Kwn->!GVmz;cE;{)GTi`
z`GqfsaYmfiQ;wVyV@K=)3E`Sks87Tklgx#_7lZYHeP_3Tu88KqkQ;v-OY|-2D`@19
z-6BE_31wW2WIFan#Fyfa$>{-Hc0z8O+DTF<a~o8BUUu=OKnU?p8ZkR`7BN%&O7;(X
z2WP{Z@ysu56(yTFa98l1nM&+s8`qUY^7!EUCl!8h{l8Ha`QU%DkhB9y<<TzrJ1KuT
zb7}RqQC|FT=wJAe{45nijJQAvk^ee^Zi*E;(gnZm$nMr3QYk#fxy}6)Huz2~&bPUS
zdXr&y4#P>Y*Cs~LL)nvsY2=0&<j$WvD!gT!4Crc92g-w;a*t8l?A!(@7Z}*NhL+Rg
zd*I#Y*UhK>Ek?=i2paCdUCV$wXY5h-ruZYs%>N29D;Mj(4YGCIUy(L)$oUKGIkBQg
zR&unL%BLT_LPG}2#XnTrDDKEaVWF^N*x!Bl2qb1z5QU-vYuFtSaf222ZJ8xcAAMd%
z7NKjMZOI<e)(862WWLfr!#dr=W(h_v-UYX%X`W>pO^?c*I4$yNT*?hwR_H79VU%qu
z9@)aI_2ff8+?YW6ikxLWT`eUn?Njl`ug-=M?4=-7XA13wz4@6BYj#v#q&V6@XqB-$
z2OHJuGN~JMnVo4{m_P}f1f8`q85YN{+BwK;EZ+O~J!>rs)lKg&Nh@J^Ef_DSq+LJk
zK-!A=Q{M*5`fAn`d9Smo3F1b0YV@N4t;KoXkE_5jc!8JYTq`*GKY2oTq~~Cl_Ncn>
zZXsLauOsqAf4XPQ`Zi^6IT7R<8sr(86Y`D$$_+>QT+3d#aTp_Y=67AvH4yPJ9e2KO
zY3(F$(e4B#la7DaIR)N8UIlD%UBz;K)%|Q6<G6+8E_-sege?9$K%L0;2kK71nQ-Rd
z1|i$vhVKvFhk=&ZuJ`UMSj!x$ueMxs4g{NY-S(}*0whj2QLmz3?N6u;4S{N8y$2d2
z2eK{vj)wC9(Gv^%HL8fco_RZ4!Va>tRnS;Q+lW=d2A~bS9kCR`{j2p6RLROX{o8|l
z&$QO&E5t_`ywE@L^W|UpVdrB1_xS-Wgj-C=ZTo6&NC`_2eS~EGu*8cnKRK<m5-jS}
zBxE>HB)@$MeS5yeMJ-k3Lzc~0A8EaWbG$!x=9yG!Xh6OFy_m-5n$%t!cXEG*wz|5j
z<8E%n!>gsWXu-=TP&MC9+u)*qzeoXcdM`G9r`7Z7((qLKP+&hRH24SbpS-QUrN_%E
z!SOHZ9c_PWKt;Ur<io>19B|wdyhAsRzZtmOZ??=#cH05+=&!13%<D%uvbISXOe?Hp
z+myT&hn{9Lpv0;;b7`+1Xt1C0$F#okBGoU}hB9a}EZ&UbXN11RvGRYNPHM0AUiGPb
zt^l1&bLYT~PjAdbP^Ma14Xrjg`rc|um-fcF9eMrcUM(}#a?w0gUcEnEUi(`9S~(ee
zjjn~!M^kNmdn)6QSuGH;GGqUUhm^OS00F<n1}4TSn3ejSFXn?_*abBObmfOEnBSZq
zW{{hp(fqpm6y1oN{k}G9BU^m=AdwXEp*VX&)xWC=oJqt(M&|*(+pyBe>8B((J`7e7
zfK1=aX)>G%*sm=Xv>5@;gw%tbi;EiKc!U=dx(RP#%nP3+`QeBi*m8U60l2<w9`<c{
zJAzk(%eDU<6PrVuM36~(mw*9vIWl0`y9SPc!0^Mc*fBlY;SeJ@Hx+Udau1_-Kg96d
zI!gViJxF?;W1voAokJY|@EIfMx!uMPvw)uh(PC&Jrf$a>=wH!_jq;Rm$-JVPa-XoZ
zvJJUaH}FH=nQS)G-dF7aSBa6F*;D%g%S{-M*sowln#xR~C`oxDaW5mVMXCsK+<&~n
z1$=(xxQoy7^UKtihdh-F>G1$PZbY9$vWL9sjik9Q#&qb82zZ|q(F<Bsj*f%jUyLyC
zRwvoZQ&f!J=XWslw3?{u$j+9w&nGkKF`@#RLEC$M%hI1;q3urtd;e$&|9Wr~5eEy`
zzi$aDe^~-r>wj5-iv+kzknz6AR;9H{@vJT}^QuHl7Bb@N{qZHK+v42^9D(Rez)9E2
ztDBCz@~ZVHnb*$AD6rX4FGcds(GmXa?5dobP=<#WS1HW{$j1k=gpxT^kR{OQ)s0^0
z50=y~Wm>XSw^N#5Pu6AxcrU!RJn|JGwr#|nHZ3B_@@#IjbM&k^w5OTY_uE*E&_9|y
zDNWbO%kVmkS!bIam-2M>P$%m;RUQ^v&DT2ne(tG{!T2<Il#jzedar|jVxWKdbj_OF
zQ@rXdlv;h5md=atTA=iRETi|dbR+xRb!HpCm~#8*wt`!z&#TU$VeL|%ef!qRg~k3l
zwvll4=FTe9Tt%d!qLwyouax35?UPKj>oqu4tqql5_MRqf*5|^Zh$_A7=oL((2~`A7
zpU4cmVyM+=Tv)`^5L{Wy`75AL2nQ}4br%)`+B{&8OoWtSLL@s>Q`@zwSqyT}OIi7K
z@dophS6d%j+=$v8o~{THl9#O19}on$@!e&kli6T?zq&>XE?rS#K2D~WnI8+_)7r^-
zzYV{=Mfmoc+SYOdW?v|X6AldNjto^_Qu_dP3}T)|e_zNnYMb{ro+hMDU^j7i&=IVg
zoF)E%`*zdz1U}jY36SG+R|k<^-gaEUh&}9<g#Oj$T|$5j?o(c_<%mCO_XWP9Jfxfi
zpNO3>Hz@+&Bsn-|+;`5;oC-od*~B!1m>&JWhl!XSWoQ9npD2EqLnqfF&cj}JnoY-?
z-0?_QZ=0EseKApf4w}0rg&*UQiu@d5)xWaDby|6R&aQLAmU)Y=*yU5#_awtB+^4QA
zGn@{ta{>JF+*j^XRpW7TUKuwl`;v@a8MUAXOZ$?4{p;R+DtJ7O!8_y2*>yZjW`WET
zuSv~udXrxJbKbUfH;;$+ZL#C}GV~;0{aJPo>U{0q69HihscKWS=kI|Igb3ZC4aUps
z!f$^x8rFZ64=yIA|Hx>Txsh9s)ha<*w=OmTGDQY>h+Ja`t{Vv1i;ME8-W{0lCL;An
zkN43vmTA^MMIvVHns|8dLR>R__{Y3De4|*dMZ@$i8wKQbfNbmap<8F=tFHSxc=GE;
zCq}+Hd_L#`1>W~q@>}ihW`zBlH(cZb5v$8h(etIzal&5s{lEbJtvBj7o@18*S8~Ak
z{^lub1j~<|g3X)6Sbl9x7&PC20QOhZg6&Ge4%k~!sT9B41&(L@eol;QICSQYSWoPJ
zpXQ6t<N9h;^iSD5I_(Dy%jA5OkC(4*VJ)GTCVF&9hHtd=1oG27_AlBazB1mxkV}=7
zcS(ABz6_+$hEF)c7$pe@+tS6!d++V5qRj7h{aRzw&&05@>tXKB&-?^P%M5bMmuwi)
z)M|!Ji=|<lcfMuj16mn+<xS2*39?U<9tve)c~jJ?A%08XRA9;<7f}2lH$J#^fNg+%
zT>Ax<m0{r?4qXXVN!$rGG4mmzP7~|HOy`%458wDm@->aIW|*{l0>z6IwX+-IsTmT{
zztD_L6dDi*skB1&(uLH_EK6KeX6e&HS$vr2VffI<_5oUMtV-};tMXt%Rhk}k1I#Ux
z{3dyY1u2z0OQ=#~+Rr0<EEQ_z6Tc<SH`3CYe6U-b6gF!nwH({j{GRp<i?Z^9mDh%J
zhL8!(>?mb2^$(0i)Hc+6q-tnMLAF$Takia`nrm>_WYWLvSYx9!>*)=*w9A<DS1P4;
zVD`R|h+qCxpHV0cFh^$Ogl@JR)b;Duw2$D{bc|?A<%&1XN-Yz9go~nHca(XgKZoRK
z+jH|z9+&b@%Wj|2uQ13MY@;^K8zc*<iv?p&X@oVN3^OVg(Zm*c<jwiW(ZeD&1i1DX
zv3Yot!1`TzioQ9b;qN<`inXH*tS|@;JFn=cL)UA;0n%qCLC7C-LU9ax;~y<c<ImKt
zB*|Dc*-BrgUeOK#%A`iN7}~+1#Wmd_uA9SEFQSdgSUDSG&p)>AET}m>>Wp%K^U;sq
zjKI6BN_5bSvS+XvW!4!$7}!vH-@|4sLR|lr@Otii*TJ#)-ahhnuX5MXu=u`a^8U>I
zEb#l}J*)aXaO@3<;#nYm@txTCJ(S|PgV{YXzjags<N~t2HEmNjx8bh3r%uHQpQE>-
zxsir2uVvKsy!q+eh}&-7UiQsL%U%kUs~-{k(W?KDOJ`yG`^kv^PA=U#w#p_D6jZnh
z^f`ZzhyN(DL@q30J|zJ!`wLk`r;A@vT3@Rl-u-!rA=e@SY`f*1=Vg6!ak?miU3(U*
zED>#Np`?o`qW(OAqGnO9DqF-H={w$I?wutkcAr!&(>g|rFl>vR9kB+F9k6g<f>t;9
zMf8P>sbd2$b5LJ$TuAWMGB~0zl(D-W(W5X>x_g~0=ob=e+cyV<!C1v36x&!^mHp)d
zNnTBVw=eUb21gH|EzYCJsO`E=u!0kuZ3x;MTW>+%!5(Ie-3y?+n_CGOfcz01lJofr
z_PS#ZsEnxDaA5s`+#L}Q_g-}VJ^KiGE0k|7WmuLpmsrlni@C3A9$iFfkTy1A3a|Q=
zpOWfPz`C?B?X$NW-cz+bT%OH2Ed_o0w>uT=*gRIjGtp`ll)I85BIGwFfmBaIE2jQB
ztUXMi=eY2ecdw?o9(u-wZT|2^&US_u!~kGc@{hmR9}=#ttp9$h%35*Qri2N3^_g}>
zG}lRpIX@SPm`oUjI-GyztD39y3GqA%ma+cNjt&A`#kwJIP8LT~&TI7*Z-M~rddg2*
z&|)1~w&J8jRBJ;~bQo!>5eNsH_<{kzoLAac)-ujc!K6N)nt6-HhMhKyTPg$7Rf@9B
z(i;xQcabFV2V7;H)#MG<Vvld!`oH;#v+IFc1{U5oq0!Cp0{R}aM`}2HO#P)sB_@;X
z3nKE}lF$eMlZ2@$8N&mZyr18dih7we&ggM?;AQx=pTJVv$~9y2SzoHVe3&o`E;51$
z#R3$bWRgIJ^4s8~8Rr|M3E6OA&=5v#TSpSY#D!|0+R)_`*zQ<VE6t?TL5rdv^)0r<
zkh<9PDg$ZOZUxpgcEM0uB9n*@!Vlj(V(I~sh(#nxLPGjIU>S171mQlS0M`o&DFd&r
z7LaMYYlZCU4bWS4PbwhED~D`^Phek~!t42laT&g^ZxhAFH4NALV;Yqhz_%IM*Ee+B
zn~Hl5aDhKe1idLh?9#SVbuHhdut&McJ&?T_P;rc$!|bAT2D<`fcPG_z`>naYhwi=+
zF6|t<EE@k2DSP0#;g2x<A#cg{pO~*wmknr70F<{}6Un_1G$Pu_4<qrH%*CrOOEs&T
z|45;cs~_Vp^5(<GGNGcz6~r68-s;X$4-nu-^wDGoT;u?#`LivdwEQZ3Gc;}4htmuT
zyKZbhkjGEk+|%$|DdA;dI<}49=O1O_x!Z;<f9vk`kQv(0Oz{P!-S@;FPW6_wtZCnQ
zl^fc>w12&-I<k$rvD0hJdLTfHzk#z-W7|{tIIbv#xsa;J!B#)%9zGeSF0;pwKz({3
z4xCKK3<3>0_0gVeC|u=bdbB(DD7@dBayzaA4Xhh2Z*aYO*r8vuhN=Ie#{6zGYK<RK
z-v+jXJhg1~6$Z25;K`R~mIwpM^JNR~oc?>v9N05WDoXR6vL9axt(zq^hhf{YH80by
zujgj~kdc%P29FJ9Ra{J?(%v5)A{C+tuJE@3x{{!HE|-b8g(B@6u(ScOsyzN|eDgcG
zVXt&I&x{@kqDTsn990I2YLMFJt#~^*ulnjCyeRT(1p!3Pw}52Jyky*(X|&`1(h8B@
zLP99wxiCFW*KoH!uNg^oaoXM!E-ez%QG5%E(iBwFaqKx+b(S3Mxe?XxO_-6+alkzh
z&}r}^gZ8~EN&W3;_X*jAefz)c5iLr?xx<bvX4H&yPDJf>WSg`(M0~=YvSsReG`g<`
zQmiBeB4*r*ox<Dz7a_hn1BR7ztI8fwPGFZuDp2jZQp>f;q5>X<(`o&)&tuPUjcQES
ze<bfui8}WGU>W+~b`xBLD`|t!P-6ajH$hSI*IzLdkL}5XO&!Wt+kTeowFm(lP|lh{
zW0qirtUYO~FMU~BuUYcVab1*yLsUywt{j^ny+oU>JcF>;aXpcT<NLCgmQ|WZW6p-P
zQo9}vhuEut+3woW9?M7Gt-IOR45!uSt{c=R$=<fP9-cB^Jrv{c5_-;?q~metW!_Jq
zty8jr=Ax^+kE{D;IE{Ty%8sRJ#Yo`2U=AdCqAE!ul@D4OW*;Qhyn)cJOY8&ensmK4
z!5HHr?_EP@zYdl3n%ZsFdQVl*-CZ%O@z6eD(<l9M5T_Aq<3E#^lj-1a*712g08UuZ
zR^(gJx}M$T9p8_pg<~i(^cmJ@h?N($mlef0a7I;8_YHCJ(<n$gl=^d=BxCM;3#qKW
zndFxJovonjm_a6k_;rMR&EWS>nr=r1G9@&&@9g@M{C)jF*4>B+J^r7LNtI=y;MuBy
zmd1rY%e>W2)Eym-Fn{2;(7c~`9)UF}Vq(0SF_Pv9EMFEN7#@m}9mtG%5iBM4Z=+Y<
zazg03NAR1BDJkF>#h#4scaBsT{#G5SFuL>Ilrx~zJIRtCeY+LPk$Syh=3=`C{+0MJ
zeoMG#DVP^+k!Tcb<J@1VB#-Fv5lTXIg!ZN36Z7rghR2E6O2xTmS{l$N2=P?Jz>!(_
zX2!zOv3xN7ps^S6NH&OtBh%khI#Ak5hy!oil4>t|Q1IFDZ0Y`YD{eoG(^6-!8}3~s
zQo?Q136gelG=g@a5d`KKkp5+*(<H?*(NU%mabpNppn(J^$Zy|h<_2Fr-_Ux)w`%3X
z{4lrGJ{Q@-PouL;9RB`-6M2qbS|EFaIger>_(#zHkj!Ie{|~lZmAh>Iwr%!TfzzqK
zArJ8ax-cw^1fI)(ZJU8W|A_edoX+7Q<=TSsb)4}p&`(#qJeOpvi}@NQy%_syaF_lD
zy7&JAo$KAM++p^?#SK~BTkC8^V`XpIaCv5*2X>8pcRA0&Xy!do^{mdOoOt4M1LN6z
z?(rI8-67>o+$XQ1aa**D-E<;hZJ4eu0Ud!E>uaK!w)oiZ&PuzdFms`1I6Hk#tWJNC
zPS6V$^*7S@E4Kjdzp#ni@A-kN=F51?W>v`301qzUv+JD;rYzy}(5>?|jZH=U_ca}}
zLw%v&W=&`bJc?{b_XBS`y*JW&gy(u|Jd~^z3HM6%3AoMPhfm(uN56vOrK=XK5`S72
z`>y>r>I5LvN#;6nDfS>g4<V~25@;5{e8$;j+n4PAN@Q4n8AME?611-^kxylTY@PMX
zJ1i1rIS3{pP%!BBGYtiaNF?1p4z`7{<vKd5rH#!ANtKb>?zy)-ACcNo_b<Xz>)e}t
zXD=NQbq`b+?l_jr(E7YfH%KeKF+Vke9Ql-M<@o+&vg=2@<XaW^<*-}lkr&ESwBwy=
z?gyos4lXz4qD+2@f<0YF1R!%_X~z|>rH5W^co5u7xZ@atRyXGfrq>~p>p{Vk(G7ri
zY@>^+yoebo#~XgnppnvKvnvQ8n1kiIC6Yff(_}1JJ~m>r%NRK9Gct62q3^al8)4Zk
z5Si>a-!4A4$|Scj4MpF(Txsi7SUPUPZ)U*HZk8Q|pvmc2f8iLIXi}yz#=OZbNK5;&
z5@rYf*t_{_)A7RIcUXFTdUnqiYF%?>zS*8IGok!6rqB0{Ad--;lz?NBFn88S_79|U
zf`a+~{XBId_WysDboAOKT3H#6|Ia18z4<TF!+9E#+MVKF_OrunY{KeWf!Zv*+-gl0
z+I#}v_uJ`RZIBxt>lfH>?Zms5^SwHhZ6cn0Cx+T)M**MQdQaE>TGBmjpI~0ew(f2(
z<c`;qJ}!bv))8NKpBz`&k%U*JsRyfUid*@F?zvK-Y?$$9@s9J40zX{Ba{Zk8RXj|r
z{z5O$`ho16e=Ly4I?V`|9>efSID9|*UEjoY@}{UlTaUwb>#jJgwFOQC5jH3yQCgvi
z6g@|m5Q&m9e?6_4;aq^5l685ggnD@>y+!2kN#uI(pE$R*q{(e17+8om+{GhIC?w|s
zaqcn#=RhT*uJL4+PO&pVRly~XLIF);ihItqBZ0ojg&9j7hVvD-(k#=^sGHp&bZGPM
zpo@p#XpFIanfX1}PU}+I)KnGT8>N%ri^zBDr(FaW?o*reb1L)nVH&FHh6Kfl0>F=C
zExmmSyWjx?jg1q_{IXv}70E1lx9_OGr^akdyEo7L++kNx8OP`{QDFN1l%<g8^OOus
z^&Dff>sSBsKJ=V2ef&#@Ogl>#=xDQm(r?g_B^}xnD3CyqBd~1`g&<}x8|jX9Y6RMj
z=A(u~ps=E<O-Wg|zA=BHbL{tuHHJr^*w&siQe8tr2x-t6ZKSxF{?6rk>S?@&u>FzS
zdPBGs;kiF_=t=tynqg!BieaY89}X~v;bXH=J;FB3MWGFCXA+#L^;wQ6lT(|O7i#`k
zOZO?mhP?yJ%hF!&xL=3x@1Xlc_TMAUO;bF<2?`p-{1Nm&WYYfACH=p==DfR@e)gaR
z6|^6MB)gT?v&toH`5R6@aRg)Vo<P2fRjNyiVn7DrUrV}mxexCd$aiT6E$Os<akxv4
z|L|SZ)I_y1qRY8fc|f3h@vQt90D*o+R(@?=WqmY_0BM_cb*kOOeDp3vd#}lngnaP*
z7;x`7<!tM1{TJ!rna9((kY1jY-;pte?JgMSk(`%TV4P|5$P8DN-93YDiBqB6neh?u
zk3ghDUxJJOjdUgT%NpJ(_?T8tLb<a>Yeef>_3wwMZFq98Jf1B-@{zoEpPJp#=P|Xu
zo{!Z!RT3H3YKINs7vkJ{b_j4Yzs}7MJn!F+!Hin&zQ$YchS=ye`>eM%n-T9S(y^zR
z?4%xLx`I&0>;D_-NFdZdTW%nrUI%<K`9v#?R5l&-De@}yn#_O%nR(f*8~LA{mv8@Y
zUV?^%h`OIXp`-rH7E5%D1bHxZ=V3qCI5=E>E>x9VjkgyN#Lzk$Oh(_aD?VLwx0e#r
zwf*>*KA18tvOMM13SLhD=%kIFEEsX8kus8ree^;cd7=ri?t9`keT#DmbG|aie<@m2
zLnKf+Du^B|YMR%E07Q@<SMZLfu3?^%Y=Ylz0Udi1nGl~L^|(jypNej<Ua_ND+?c{F
zEk=6G2m{@*&LlP5m<&WVM@L?4a^$xGDJ*+&WJYdG*!$IlHCMd6SZgk9+o>1wq<brM
z=Rch}B2XV{h2SpPPjCrojvZ5wG(o<LW}#6a#@~FGEEA2t`z{mz<hwlJd9AEpt=v1+
zJ*IUvVBFW)tPBA1EP0GCGdsk$#gR%vLCy=}wNd&XN&i#M?>}^2{?8!Y^8b?le@O?C
zo|s_N@<-DDkkjM(Pe@mlwO!^!1EnYIkiez_En}<6ijm>J=?Nk(W5`x6l#nuOKvWt3
zz9WA3;X3D|w2lb`60@D#n^OH~dOY5L9g4m-QqR+?dqS7HHo@*Le)M#ev9VclZ^Y|%
zdL!t@>OWqc=D2lC34OLxZ5Y+@<TxGIBYCvJ(Pc1dtmdbj-E<ddQ>u(=_Wtx<p4fA=
zYsZa^|E{0a<+G5fM{?{K5LR>4Q~Imgx7%T9e44|g-J)#!D@Ww0mtUMxH3pl7_w0Rb
zX1^#|k@zQ>2$gLg&f)Jqp&}6<`|c|W9648H<sB7OEM95Xc+->A6d%bo-qrgnPYM9e
z&q7+QuD)CZdf-k}=~fLE@bXnABOA|W$TLrlX|}nc^G4qi`$G~CuCe3>R*@4|Z}4KJ
zj?0RleA0r<U+l(nwXYtWb}@dI;QJ_MTE`9Le_qBz<=*1-Pct7yWsuAD?!}h>3RG_T
z(Rji7*<KO7QG?Yi7r`U`Kp{~-8JrJC!s-jn4^BBj@!?@Hpfde}wK(c$N|n&qC~59^
zC#u7yToK9W*bHfskk=CVoZH!li*=iy5l))I=^;Lr&WKj-1SHNwjZkY$XMQYEa3oG+
z$xrbSMHu$#Mn#l>g#AV)o3#PE{N;Q5K>K^#_o2ggAjEhKbR3BT1d@GEOkmYWFuw?a
zQyUlsQsNfdPFMYe1*b@K8!P0vr~W`)FSR@6_a3d)u5e;oia_6<c8{%^f)Gx8M>?kG
zEa$S#P}rxk-GV$w2CN>}AwE7%>U)_kNv<3mk1y_1F_Qg2cwnhf$3)ES2^^o}sFxX)
zW(IVJzE3zm<Z*7B6R;-42LLK*1z_iNNMq;BSz~Q$*vHo}|DN8BC<1gPb4J~NUc&1j
z+nH1tGle*ha$F+<%ro`#HzzXRCIPF?%K`A|<j3QcHiJ(r7t(X`<C37818NOlxw=w;
z(EHe1%wE7N*QEvkY?S<%?qV#ylqFAdE=7_4nDs(MYV2YQ)G>4u%fc_I>p&5;6H(W?
z69MRy9#fx}2Gxv%s`UP<8UL$59Mp0IdaczHR5R{AnDX?GN2TO|QZ4DRjrmxXhq@hj
zzPcTmI?!97kAJlwT|Y0NCPo(}jpKRGf~LC<h?gfu`VVTdLm`>b<BEh|!YPrir0UUx
z6Im=`+2K42cFJfmWme{3{$Q^&@k!k3o=>s-ZMJ?5H~-3j5sHd9-4JNoNGHvUdf1y6
zhX`2lfwc|2vgiL5#=t#V@h+r)lOz8IjfRfq^M_S|gXy0)%az^jOo$le4K0+MtYI1D
zh?tp}{%RI<aC9Q#;^Ks5lrgaZIGGW#F){z6ed&L0B|O!(bX;byc{|hViat%N67p*t
zbK*wbpX+UEbV=<sX91Y~sCO^cB5#s?I$mFe6!iZ^Chagi%{6WR37o7_{bY9$Qc^Tk
z+r*th)`rF4;^fP~=7@WLaA(->pc(a_6Fml^CXfTnE5^Zem4+-D`T($`I&MGMapfqd
zuRn`~4Mk0dwy3gMMh_DQYqJ#?Sh%LB-ArMDzo@>DqN^JBJBqR;*q6t+WM}996gFTf
z=R~l>ZoshoGTu0BPaO*o<}%jl#&L>L0&5YKvyw<b|JH|7IzXLEX^>3CIzqe|;_|z}
zDJ9@AEB{J{I<Qp`U;?Rwn!-SLRc>q=l-I(|5;9k6W!lu!!aZI}{8sxLIAo7V&k}m&
zpE?Gt`RyNqz8fr}ia<v;k4GI4?`C6e0Rg@Vq{5C5qyrsovyHgOQOU{hl8X@nL#tql
z5rVF3nI)r}H6mm&p$4m6LJhARj>qFi!DU|{7@)1WF_sEWmL8akrl<&BOa0A5dk_!d
z<0pTs6@PFA$KY^f*mG{=8pSNU^zC1CbCnHFOyFby!H=EMs+I7Oxpgj`*vUlm7#|T?
z9O80PvoIG@`2E=In7|H<4Zx9km=MK=g(N2Cvx%_={B?*79Ep@ays!NFjb@gq1*Sv`
zoWZbDQxbp~F>fx>w$D-kV$4R*jhETekb?OK``ajp&vC_JAAr^G2@dNIuh{Fa7K1l^
z5Yj~Jh}(cfd<hJkQn=<6-tY4lIZGd$Y==xO6*b?3J>>kxeXwjoGTsF(v<_8_rwsK3
zB`^xv^x?bo&Vk=XA(wKV@NA$p>;+a<;2MoX*>0EXy2%Nz(Rjb@0JbO4n}xdIn)KHC
z?)4se{Y4QAA@q@a{MuFhC3YBZ9H}aqRW!KoQ#d~c8%AcwY0TMbe4INQAI;aK7g{<g
zx;^z3R^!l9+^;S3+PJyee0^K>n|bpDtTr|y&84HoYpn#)es!d4CTESCA-@xF2P)K8
zuG!_cO?Bc-5#WwgnsbMBOc|P41T4wev_CnLeqJQMYW%r*>#`H@>%Ar{r^ZxNP1bse
z{g>{vc16*9Y4zm3HBV^({#)tw%XHVcn4iO!UHHT$e6V6O?~moF%Q@g%NnFB}J^UMM
zc9=wnXC$>x1b%elc*L*2`<j)R#HgJ=+7It3r-m?WEdPYR{pNs}6C?`Dq*O@#;iG`(
zhCdZ`6lpG*T&%~>Z7a@xJ@npaqNmt{N7{aRF$Rz4>Z}AIs{j>r9CuijZ#vV?#W^!X
znp}QP5F2@5WAJ`)ELdH5_!vY;=k_1%0fPE|saO&hR$Tpkc-)*M;C4TL;)sFKXb2k8
zo)m})V9Y^+eL?cYYVsNtUg&D!`aQ{-7aT8m3^W~o7xJY#eE^*AAkb~gg(Z7gHQN#k
zsJqJhy?G(!L?$`konJQKo6YacnxQKP(kc>X1WkEnFv55@=aEF;n0T~N9dK_(U)4NJ
z8Q~=m6nPbWT)j0BW}!!DI^Npg>Qd(vdJ6h3X~I2}W7tp|q`DgGSqMid-H0XCsl}_=
z*py!z<hmT6?VGhaBE<4E6-!5)*)4|51H-NzF#0i_z0Lt=<5K}2jJ&hXAYU#tb#1}2
zodd0yP;U56vzlgo{F0zzzQe7N(d37ldArX6ow;kBTOAXZ96j+U`mmLvIir=`I_Q3|
zR;+fIx%%S_PD0i&TAJzGHWi<;5%0gEa5yHL>NBp4!;OHG;k(IihsX|9pqZY(#m%vP
zJHZFQB*rbVle>q)4&3$B6G`<QmDsn}`?@seRn}LRFwxv-L*$d`5vqDQlo;V0#vOd4
zka=2+s7ZdKk*)ldZ38b53udSv9i>df9PG}ktWd;<FTvThR4VczVD<E~$q$Kch4ND>
zk8i-**lMdx7zh~sZ&ZfDf`_G<5nHkOBRao%j3jATC9BG{#2zup$n?Vc-IPh6M8(eS
z3AU+bJ3mfW(M*0y<j-iJwsFjD37u(FT0t7>E<AH((G6}U%AQ^m&Ay2c)iCI2Wa%Uj
z8Wc8(Vik0HVZ*^Ke5MPOp_s6({xm65@hcev4xO0!3hQG(3=IFLz<P-882y9aMKkUz
zuo51MYTq)QGYAJ@caMuUE~<!ZSf;pOdwG+I?*?k@JUmUPe2F66Vtx_-fX+tFAUWq`
zG$<i4P~f@pyT9b-C+Rt^VodL8Tj;NvK{B@#5sz2o^~Cm!PxPDk!qF9jucL<*YyI?t
ziE#{l5`SO|i#xyvoQ9P(>pWF^b6x#iKKv1DfQ?`KYg*M#lKwF-?Sq9cX_`=GrD;wF
zjF(9ks4X&t@kMq|>rtzA(fd_aaj)uv*~{5KdH~mV8j??$3ih1#vM;b_ALAjviv<60
z9LEUT1|zJ0UHF34l^c(_2sRUZGA**#zLdyd8!_Dv@dzW1rm>vt%ZHI8JUu~U6vblj
zl#$K$kX5*-O>S<YLWguLI(kOIKo#o+7zP3z?rx<ZeB3OmDG3AUisyKeuG%_7L3X(4
z`Ooq=bgIo&!RiePUoQpu49<RRUzmG0BEC+9xx1+na`|}}>EM<wtS{^Lb+f8QRD78#
zG!U8j;Uv-JYd>SAQFS-j6LK3oJ3U#*SeQ72V!IXG4;4Ehavto$Z1iPazn|IVYoOVd
zYT<ID?Wg-gkD2u;*k1_9vS;MT<8V5)it7|ev|$w6<WDo|Yh=jR?l*D1#OLqOJI;GZ
z(L8Zh`IfYJm(UohodM}JX)=A${R)VgPNXbMpY-;2M^O))s&_ZZjLdPiBTM@&zv%X~
zk2-81*H?wRr7M>1qyS~6x_1LBd0Zox+0Mj0^zh7Z70!pDuSAopzNRT>4vRsSAbM5F
znwz=_AS2ym+Iew5oV7sQ-3{jsElgjUy)~xZ=HD9gXApdQcIFD=86S(5mtPa$BwUyC
zoXxTC;7*T1iXI5$cfyIEuaS9L8yMIgUmOoT@C{Y$`gLGZ1f>T+0<>AMj$c9C6#`$r
z&CYYobUt?9?q556yC;Y$hel@~-=Dp}Z=dJaClvNQWQFgbF4&*teGcI>DL4^#>Gv@_
zS05bO%uO8|QpilGac3(7bW}^bXx#avBMdDf%;c92c-+>y^0eY8XKG}sls=u-<5eCx
z+I~Y^L~kkftyn;(i`A()Hdg`1cImNcm$51ST(bCHt+(ihP`ns{u=qW4(M`DuCM|y2
zg0--Gubo{~75gIKxZ9EMtY&Zn^!-L1t5N&2rTyg9)M?D)OTbave3iaE--t!}7W`WS
z{I<(l!;Tt(1;AbK>)=~y{<dvfMCKe3;h|nF5Z~+lRTb9j^76JPnlKUV!K+=&+wX&}
zxAs{<yU17$c$w4<;9A)?dckO7lf|W5w_5E4*d|Ip=Hbj&0sUH>*H3*=l^Pc#vKo%f
z%1wYc{58dz-laA)40krGVe<3>Wk*|Yz3jIvY-xzIEiYE|S)No~IL$c0>&4~P0J%zr
z%6vnhL$hY)Ljug<bmLaWS)))I(xQFN<i1cr#bVk<PNB}t^!*7$i#}n}i*w$XR(ZH)
z1<tFQE(h)?ctR`|v`$A-N8IAc@y+Y7kWv#JV3tC<*eCQ5aJ%oh7{Az;@8e;A*Yg~T
zM9}7O<-cz!C}bKM^L5)3zLty>LyT-*W0Ls9Y91?=^#?IJFB_R6(+8)VtL>*JkI81M
zhQ08o$}>isYe@?NCbObNn+7ZD5qwQJ;oO}dMJq-fv*oP-EH?k8S$*CNcI|JLLPZ3l
zCCBDm2lF+9Suazy{$pMSPDDRdNzHZUPQ{j_s6J@Pdh<(0G!o-$nO|4yTGXW$F@Alr
zpPa7NyB`8mM4BB|y-j^Zq%9TmK1Y%6KXGB+yACfX7&z}t*5lIyjKp{n!oCY95Ckz;
zY4D`r)YOX_j^WwXZv!Ksta~*Bqb33*rgW3{c<4_?AD3$g4GaLsbCAzzl?*HPCmrSe
zvUSS``Ltd0fhQJzAHQnHcFLKLT$U72Vp0<MUx(Lm?pBKgP`_Z!VAUam{WcPqC224F
zEX<W6Gm!Zj2I@I`3DpC@8}&G}6MJ<96NiLDA)_#T7u%LZ8hUI!Q@OqjilEZQ)62Qm
z|A;?F`iZS)xvMn;zP(bxor`L*u$kfVnvPJ^OQyRIT;aRyw1>Lrd!Rl`DY3V@;;5YY
zV9&KH<@p*vC;dBKM}P_;QYQ(+z;11aPg*DXkj`Xib)&pyT%sA}nVgJvYI4?WzWE-i
z#MiNffxO*QG-AEhmy4}a$(mM#GiWU{X-|PR-_(Ti`~+h@TIOtbmXEX5(`l*cV-_yD
zdOdX>PAbtykC+oDjlRoi+wNa7KxKjfwgz}}Nye3^mF+&4dxMwLVo;$Ox3j;Ps9=r0
zBy9s2MPupIs~EJ9#TXy6EbYwF1&6wJ@{N;b@upfWKA5nj%*Ke?ha{CtYIg)AC2olH
z4Ew;yffotj#(lheUcxatokdUcW2O9m7<&ikN}neIG`8(zV%xTDXJXsQ#I|kQwrx9^
z*w(}^^V{A3?s;$D?%tEHPBp5!x~k9pI@NuvZs0*UyD$U9a^;T?X^8d2PaU_CGuI&7
zax!%@J3HKN?v~s+(r<q`w$p-+R`=Q#0b1Q4^R(d&*R_F0W;h|!7*qC+9N-cNipQVV
z3SbHOa|LzYCwu8?KFXSTz1|Ep;M3u~J3W0i;=lk{4Aa50<@xm*hlmwdi^p-bRH<Oc
zX+|xdXXAzvYBcXwxKPQN_1pnpsOws-HfYECy6k+P96@xo&aFDAwNI%!)fMEz)$P8?
zPpI|uSNC6_AbGp_INUvSR`@6{V`0-9y{7j1mNi-PH(CQQc<TdO&^Xvd?pqQLV9Q%$
zI;r{TTF4KefYvsiWf2bTFN|E{b(__^T5(gaRH|FzGr7^;0NVK{23NRr&-Y6P&Mj#@
zJf3e&ov2waCTzVpb@6F$`<qX1#B8kN&Ep4{6`zoQVBN)O@I=ru5vM?AFRayovte1S
zHm!^mR^h(aUdZ?bZRLDftuB0h;I%3ZdOw~$jx|J7)8J(bt+RjOS5K)mj|bs<i!^LL
zy%6ZEfedpA)v4JjpX)+3agiX^@MUKHdY@H2$+8Q_Pql|Ai!ZUGTn`hTgDU%#g{74S
zQsnTgynncU@{$rVC9}R>nvbe6lb~R`StGVWz%q&l*e}{;k^Dg&dMu!d0@ze9SR{{Z
zN9RA@AB?=H-WCex0qNUW>d)N&%3VtEObnx)tVbXbb0EGd1ayN$x+Y>5PA_71>c^H#
zWwhf-!FDSddPp-8h%38CkjB)9B>$#Xc&0p(8sqUerKBR@PsV^?+&L>z+NURC19XCr
zWe}R@8_**U14xLRTTe>8OXoHpduL9?M6Uh%>v<u$IDxjWIU9}c!u*!mU0+2*za1@U
z>>b})B6hKWqVFh{XNLa|7lzWS#qXNtvIhy7UpB~xXgi8lC34ntIso)*=e^MV^c6QH
zkd?#u(X{M+)JZb8$Md7G-VPJVG_bc2{C}PtWMw)b2)tm1{ajgl@~wlAzp%|fn4cQ6
zr;i|hV*<QPFfR60z^%Tu^Yx^?>=-HSv`T(q8?`^Uye~i~Z;?f6H!%J2phl<N#l(i4
z2s*3Jtof;NJ$?83T#6g)9HPI3R4>cXyaBoPK7~_*&afjp=;9iD?bfrlyH4vkb*afQ
zL34QV3BRgKXVsQftJBT3GC}v5yteqlnBh*Ty)787!hO5tuIcTwuq#1*U#L{vD1NzE
z^;_biPw8z*Dx?UGb1acWx*FX;kxHG?#nk*oM?8eAE8y$^D*NDyd+QM?cffE2+D@!J
zRx7*Q0X4=`IO{6m-MCtM<V1#&^M|H*sakE5n32YtUDtrl$h!_Dla10_$`I7mQz#)e
z-l5B%1?&Xqq%&pbIGO7BV?6zX=#UuPYOdr`P5r<UPPG>ot7}A2R&E|`leMD^5ZJfm
zM8i7R5!M}|Kr1)mx!pg}oejoqCS1|a5cFC+sOLAdF2A>|N8a%K%sw&-qeWZ;t2V+E
zptU1x!hfjRbrQ;EC9f=9`0`-w?7fu%o5_86c5~C)DL&gvMMh1QF#bVI=X*n!{8>4i
z(^cZLZVUFsB0?H;XWB#7htK!gAeVmf)wb@5bg1O~YMjZ(<@1D{C=E)Q*gYbtj1a--
z*9uE@QwE*MgS>uj*duW>;7QmYVfVy2D;U_2qwFcP>WGx%yZbf0H-pEyi$l7ikMzas
zd@iAEiG&>bW?J?KzvoYbTm^y0y`Actod1qJG9vB3T1_vAeHuY^Z2C2a1Y+o$HYaqL
z1{y1^#wwWUZbT^$*)6_{(T|xiAjQD1eb8yv(5VRf842y!wwa)M09?B#gGPbK<Q+p2
zRS-f#1e$(gKv^qaY${i!P8%5{?#<U3C#LgfVLP}>0#VB=CW#v-%YpFsX^sZk#(<Fy
z?F+sz=<F*Xe$JkVk>|=!oXTkl_ZSKPGR`mdAC7<sIyX+o_iC>sBcBN%3>@g}PuPBc
zGKk6tsWPwm;3MJuDB*43P_ZfdR^4d#Y1*)lGM<zv=kKSZ1Q@IP230`Sg2uaw>*H!8
zZrNKYjsdCikY4#O@sw_VAIn2l?B$0h5lHBUs<+LQw?--j5dbH(QsN#j+RG|!no=T<
zt+EGvheVSLdM^(}b&M$kf%iZFyx{ldR_#9^WT#bjx!2yvmK0qspn&?q<lq&Fa1_lz
zS>O7HQhar|Ok2AI8cm7V{;emp?-u^Q@ddKe|JRE=|C29}{XczyjEs!`{tJiy-4|G-
zDHB0F2HUk#Gs^dql<)y|!k^)ffgxRl?QjSvyva|Ej5H0yFHb7QqO;JGN{i<M{qz*n
zv}*|{ibCmP!KNxJ)YWE9qY8B^OH|xgtJyQvvPIYoPJs2-9!Vf<!wt%}DF07lWE;tg
zUL&SpE!0@qlf`^l#UZ+;7h@$dwVodT7ytCH;G+kytUC6dFeV9WEh^Y|G`6}Vn_L!`
z{U=agt%~WrX0)$;>Z-gy_$`eBdS(|Xqf2xVHAv5(fm1$YVtds2pWr(6W$j@8d?v#q
zhVl`XwfgV@$f-Lp@`tS;R^7hU&>-FouMNGq*cSH0;<#~i1g?YF7*`ebZ3dU4bX04w
zOE0w<=cDNSuMJlu`GR7fA(ec)yP~Y;q5hJVCC49ZJ<O+ADr}VC^-+|+VTbkwm%JPg
zHdb4m7k>%y5R-p&cRxn2#&sM*uo9NxCPl5Xp$w5oX!g|nDR<!;EI2<JgMB+nc+oY8
z!9uT#J{#_?idJT`QRm8!%F~*`QySi)=}C(#BiddS;k@HMSB@S(>Oo(*p|885zQ9_3
z_3GBLT6dOUqgd7mWqmH%5wmub>qsvt|C8-JnB7{Wb$A7RNu4b&SYEwb4fUK_e_amo
z`1UJ7B=6pxc<A-1R!ng+x1jmO#Yu`Me%)~%(YAdgjVJ!otY?O(4B`Oms{3{L#WtI6
zoCkB2N~WzL7F?mIA(pFPU1L;xKpST)VM%g_#sP?mM{iVVsMCRyG6NlIv-|DpX_xNn
zV`b!W0G{t-e?Mg-Was04MX%fD*ZuhnKJF!jdNaN){^!l<wO(_N>P^;-C7lf2I{*>@
zfv=V`!20#e#^u{}kI1+Vrx|e(aS?M0b&52d_#@Dcd2KL!>eY`#5UgINeSego(^1#h
z!Fbg|&(b@*GEEbTh^~AfNWpMQaB;q}IZ%pmas|;a`k@8=z0Q!A@I+#n;0m%QKt}A*
zekj^Yf0UF^q!sZ_-63%CkT(fCfJ(K(S^wbp7$V<~NG{*v;yqLfDST>Xl8;19WMP1!
zzK+5(ViI>^aXd(W3F)#icle3#B(%IytrD*kXj19UxkI$#=4de{tY#77lm|$~D7v?j
z6s*Eb+Ue$VT(5xjaLOB~6j6f<kr^hNm4U^Fw`Nt^vGrD%VEXOGt&x)cogq6Z7uLz~
zcEvF~;Ij|oaJQ<2G@50P8%6Ktu<65q1YtxUxCcCgYK4T5nvl;p;s7!=)I(FW;!7~H
zQPVpXjWFQ5B{Kkkwc2svOtU3VcEt`zBYD@pNZNOjN<Rt?Hc|*WEl%w!uG49vvD&1X
z(9l}UFyRydmHZUnv&`$n?KzQ617;d2XkQ|VTbrm9?O9uu@*wB9aD=0Uj=+0SDSYX1
z=`;P<iJw|p0m{Zw3fXCriIa(&7aO(fsm%RI6tSfW376kVUZo~}EcU$xGi4Sr-qIt(
z-*fj*@=pdpLe_{M@*c=<(3{kYrtNd6uz2qfbcx#|9gFW$;FXo0W*-$}YRA=+<As9l
z{-)moR07~?Mm&4d^)R6@i-ScV^9wLj`t&FQBy67TY5Uco+^KC98^B=)$(IHs4%3aW
zzw|{TlA!EEwQbHUi%*w7K;<)F184wp<66g-0U@o)?ffnuE?NAu2m_vD8FSYkVx){f
zp&p425Q&S(UeqiBno=_4l40d#;;qld{_OD@^Nm<(y(lyZrzw0`T;txW<fv8GvD{{*
zv3_C6O8oIp=b!yszgxF+8pspLKTk99w~Fbu;m>Mi?~#3pz1w9Z6B*G5=VG!KRR<^8
z$Q_4PZ!UblijCv%jVBQ4-n^#p6dRaSsn8prVu~0P#aZ920#mc}uBsxd%^qVa=D6!s
zjB1+5{XPcH?4Q=&L}fJK3fVd`;dA!y7t}I8ps&2u&3|#M_w0bhf?(x#o0lJ-A=j}t
zID~h>BR?~Jc<=GT^S~#XTk}Z0@qN8M9?SJSi1~b--lx=jNJnqqj^w-+HP+tzTz*@~
z>Hd7LquZ!^--sr<quU<ZgU{jneEYM!Nv8&ehjGIp1R2EV^*-}+LlOd5KMRtiCl*Bi
zrUvFP=!fgc58xTnCZ_($v8>o_g21qw&3>yJE`3k_RG44m@_HEIG%%RQqhTE~X~@=-
zVLQLav+xC7Z9LkGsHNV`6lfD1bb}qdins@hsO|NOIMT*7w8y2i_bS$eg3X9sH)5U(
zf?{RB1-%0hO#2-q#esix3aE?qGrN&k0qeWVmU1HRVdiz?KvmxP(qo=c0K>yf<k`uk
zO%h}%9@!WaN?%A8!}P@a3zSzYx>#&Yydd1~hB@!%%p2%nS9R#|ikqQf7G|d)(U6TO
z4nDING0KczCNaW|TT<9(wftsQFeWEyK8E2=Qv3y>2kA~ua>NE<Ps-gC8=`37#*YE>
zOOuOx5st3OaVM5OY`yabxAXeL%Ul_jtwjh*l}gekh4ZpST5vhxuU>?!_-sa)A;HJC
z4t8|GV5V;iEqKl!pNb1j?P@x@E$?B3tP6P9SS2fZ^jM)zHzlORl%s{L_cggBgN2r)
zg$Z5y@|gOzx!g3m{4B-&%?5G2&e4JpybkCVU`I>BPD6qr2|4cA&)V{X-CAhCYW&>l
zHQeq>(r!H|SRrXe;SqILTehTt625@4rr<0+zGQ*v-_Q!25O*#2>04G;(#)6w=euM5
z;zQ?RnX<>YJnH2#XBHQ{$Ll_EL&pm^!}~i00k{R)ngtt4niv=%t6w9(%&6BgquV^#
znkR>O@eSHwl2nq%+og<CWs#d;ij>EJgXx<ym*Cc<E@Q_Wubp{MR@#Wj&Te?c9>BqQ
zFyY6g;<kzyuPzr|Ffh&>!Lq_5BV{4=!XuUO5l9EhBq)Ig?p1|bQwx>YuhH;#r6ZU!
zZzM6n51Wqc@CceR*FP1xh@UbAip!(Es^=iTz{MMA+y1-V(f2;*KQDj&kGmu0f1Q*v
zu(SQ2cSk{~8|4?Hu))_WN{i5e6u4&~1N;yNlz|9$K#p~EV1huNE`oS`iY=GPMr+-g
zXU!iHMe}-}p7z)Ui~ZJiX}Lam{9=%=)T4z0RDqR#zjQJ{tWmRd0(3R;v3{O%ll)cz
zyGa4*lAf39S6AWCo=6{$tI2Z#MMMWsMUv1I?!q?(EFbSDozuqB2gK7Ptpuv_Ble{Y
z^5+Fq{+WB1i<Vl$S$?3NyY>yFYE;S9JE-GRDq)TMked9-jl}~&ArE+y6U8fh>S&QP
zu!s>+tz>din|_j&Nz(o9Ek{JpR2iG+&0I%X{-mkOZt`oQ(BKs3uh&gRL+X#knHP%3
ze&nJQ7RoVeLU1LyCtp+_n+$4a76c+^ZY+{kBD|zXI7pSrAsDL*))f&@(ToQsA?c(P
zX#;3?t;DxK^;#G9uF}RH<^+SOwPrt8b;1o?*@n*oSuT?O5ZC;ZIZri@MyJCeL8e9G
zF348nchMotNHPB_N<6>pS%M7vls49>ehkiRP^4LHY<_In+5aJ#^gIb?e(XuJ#wovX
z#Y2K~ZJ9sH<1{IeEb9VbJVA?7`cQ(5OXVAYRz6CQb-)x0->KU!5{Ed|yUt1$C3Mkw
z339H9t&kNacy+tYk9A|2DMuuhpo)`$`)-qE;nXHlXW_6vK%O4NCdLbCvi%vVP!Fco
zX%S3Z*lnWF$p^i@#4Lb~A|6>>n_Z3?k|Bz-TB;IRtA}i7#Zk)V=<)u1IWO^ff7;L4
zQJz;q`FyU)`7E{f8S#00d~5OfaD#sO7<#t-^1=tMiShY(;_IT+D#Z7B83N7#oCik0
zxp7|4^fU8)#9J)^cL2LM2fo+`T;)bk{K>}e4Ui9P11d&<Nw5jLao3v+bPm3H1YGM$
zVA9J9bE7{tMdwb%4m1_Nu@vtrC<kd0g|h;2APLkn+~)OuZ@c63rBC@eQBxD+==n$P
z>ocbJ697S};S{VTwY_6eTH|&y*cU$VEd00;8wQ`C%98*VUYd&7QC{8$;5ER1A;5o9
zK;Qx30UWdl95gM!|BMIWLqOnE!0#xx_x9hb(*uBB08dl`zH3EUAfEYe2>Ty$`H2Wq
z(R;IEvtfQV0$d6Pzx|}H>F64iaosS&Z?7oP7G*(DX6nlljFe^Z0vWru^enl0EAz}+
zKdBJ)QnAabIc-%XDqY8i(3OvcD#sE~X$P<gjEIl)w~(Km>#fRp>+rPwt-IIksuku5
zOi{S^IEPk~TNTh%q`6iap2G~{O~B4jHZ<dF<ZB0?IRShR#b<zo{|<10muenyOyBzi
z;H0<#9nH_m{|i7B{(|C$2paDkk3AUNqz0GuGqHe=2XGD863Plp9KAOx7E6&eIB_IE
za<m7n9DPdRR*U&f`zl|9p~1cyC+j@UVj>eufK@fAK!pdf(@e2qe7M}9u^jQF>P#g_
zlzFZ^+gF#lOvTtxWoV5DpzCjksnFtG;MoXL=jzQ!Gq)|^wTR-Pe0?IUD|mjo^yKj1
zD6>>NUtOV0i_oax)@@*OkDo_NsgcYll2}}TivDq6w_TNUGZJp-c&N7ZSNEUKzzK+m
zB8Wd;7Eh()6=0{q{ncQL$^{Crto}-$5-3U<ivF=_Y1Zv`vbn0eSCqz2M4;ZnTg&=o
zXzE%hX5zqF?14YZz=TmGAdhufZFtUA7yvD*$qx@gM<ZSYDyk`e?YAUFRa1*)zVvN$
z$pB^ZCGsj>>b1A!uHO*MQq|8=#fK@{4Jiniu9BAA03Y#Gg}pYuzW{$BlRK9fZBmH1
zdcZuY&f8Em=kZEz9yERlZv=Z}D|*2phj%~zv@KJ_79TEEWG*N3=aMF8DL0Cw6e;ik
zQiY$SddI;;iLwCkD3B~uATd;?z*4!JPE<6N98OboDz*>t#K3&O?pAx>zUeA?3vN*|
zVNNUXZpVLK1dF#XM#(x2`wb#=`lUREp%O02QvI1vOqBXbfML5J`+6$N=Q*TSL&S)d
zoFBxac$v0v%AB2)Z~ARQZ^aqO;ReNrO8ZB263jl~1gwp)iEFvk>7(sB&!Ln{W}D>v
z3?4=6G$lvz;cmso8Zv*+U3m+=u~oJ3Cl;VCuddSBzvv^#!>hy=H71+X+|>6PkWCyP
zb5jn-W3gc`qEp#dpzv5^7uZh4L2j2rX!Du9&S;dAk4cM4D>5^QQGUB9F6m%sK$*G-
z`+XQ#mKkYWWTruc@cu?tGs~{P$uS4<P(viGCpMfqKb9(*i!Ng!J%*vgiMdd_TZT-s
zFHFHXw*cavl>4Q2uW-vMuzF24s;*l8D+4oyFIH(na%r>-i~~0iHx=*nk%&zD-fz}R
zVoi=$djMbq^hRb>s(&ch_gUxe%ay&^awDf;&)VnggN|PtpJNxsf%ALjrmR`)B|jd(
zf!<6OQC4AM@qA_Z)KK#Qmpy-{;-!v3j)=*!4+Gb_|Gu4n;VVNY<Na^n-~aVa`~Ud<
z=J?n9n~{y<--Dj~@7|3l&5dZB7DTTVwTA*S--Gf!9eNn{I%AS%&>swYA-xcy=grKk
zTbRZss^MR}Di;&A>=&opVOo{E&TgXv)z(!-zd28+LC*;9dU%~rB*40fERBA03aNr+
zf=7eXI<b0Su1>!Y0H1V5ao6TSCuZOFy8Mu+AEM2ZPFpP^XCHQk&^>~24pof}oj`&_
zd4Yo+1vYX+Nr=o9!VV-2hRYJ7^-T`L(hpzN=g%9D(HB<?a5sQa1a1{|&<0jIgOH9z
ztbq+eAsryn4}#=mBxqo`Ax||`B^~8VGqxUOu#foRa-eX7k$o6Xrp^sjnIMuJW01%_
zN5_$f2o2GXq9G9Q4Fm(m9AJal>LI!Nr&A{ri<nuNnyQ&-nVJR*CKn5*L1MZ0-BCc+
z?4nNo0wZF%0}v<e?|>HZ+ya8+5DweOWTT@)umN|WdFl*!fL|y4@kOx_ioh7#6O?{Q
zjeWAYajjsr@vIVz-vE;8CkgH-1pw*QY(ZqLqmglPl2b59Ov*Ji1Rmefoh&fbsdhZK
z>byI88yLE)Kq?T`X+DnVPZAxZy<SO=Se|g<EHQn2%<Y8VBFnv6ae=2U+xGT&FVn})
z!r3=rf3!uybudzr{!oJ-p=&6~CXm6n^cCdZ1P6C6#5KjFI!wCEjgJ#8guVz1Zx^I$
z$1D6Ki#^6N0F~^IK~Jc>5Nw<!p|cRSmqtJ&nn;XKz`79DMSx_!7s(}mtYoLapB<&$
ztLl#&#M8GwT~pKJdAmE_ddWlg@oawC$?<xLioB%v?%;WPb4;1y1d0AA*`*!j>puOW
zYM@Sf|MR$lHH+fwh8_E?2-r?apbg-q4fuc+tLgiTNe5sHDp$dmXcuc1&9@N<0&A8H
z@D}J~-kKksIoBMpT^-;RmA?Wo5Ap=6H;II+>mbe!R-JmSntJVwBqorGUyFK;md0Js
zta=X3vQd<%m-vleVjG?0_rre0)~ypQ##ieUU$H`54^B451;@^ivpJ<?04>i7))Ku8
zcDj=Qy*_9k#T{CP?~sW81ypV~IT{8UdQjhnZy_DjLMOSjy@I?p;39gExh{&bHp+o2
zz$0Mpk%+*Mh(IGysgBT8gg?~E&vw|*b(Y0W@>)CCZ(*$oEz%*Nk@GYFg+aePy+F{G
zzr$GwEP9Vb0E3i%4pHQ0`OB2>@1m&NLz$KEFHp#{gW7P3q-%h#Izee?`cqT`Sd{Qj
zP|P1{1Egd!5=g^Xf@l%UUQRTh2^CNL>a;Q)-Z$pbHN7`JuaVS~4y<FHQ?{hW)iOmg
zq(*p$@?WyxN1jFXH35}xGx<Id_O-*N7lvJRfc@fh287WEoi*pT0(u3Sr3HwFn3_h8
zm_|OIZr~ME1y;R2PrJ1Xto6{-%gW6c8EO~Fl8Q>oz{N-7p;+cQMfHDDY9JfRCmU)f
z>jDCST{Tf}ExeMj<ljm#_|N*qV)Lg$^@qZfNvK>_nFcZuwbbE0YyJwEg$s!}wBkB_
zQ2C5u78yh61@#U9cEZ(Sb1Qu;&kg3-hwnpLj<jKm1?is6sA37VB<O|Iac0Ld1(*fy
z_+t7fdn$-^=3;X!rPSeBs9ZEdKYe{H$reC~jA3ROp&;|o8UTxh)ME2edf2pY3t(BK
z2SY6&Y=Q~@MZusKeyRg57Uvb4*V4nTqw-(=u=0R_hX1h*7FY<VgIrq+R;do$31;<V
z`fUr<OrUj&StlqIS-%Y_u&sF6K+E2V_#Z>4{9Q(79H8OJDDhf=%2L55Ay6VAQ0Au&
zsR+N2cO9_#x6GDZt(R`qy;i0pM>>`ZH+X1Ka48pYDNkHr(_!&ZXebtGzSY=Hp1T$=
zI2SLt6~FA;PNMTuy|GZfrBKaR7T5rJ7Twnc&`#K_xlq2buy#J)za|>t1b2nzwD7cX
z|Ek<!x1^mUOG+jwqom19BF{#m2BFa}ordBc-sWAepprp;oAG|duw^*=Nrt&+veG)`
zC$f20vit94#_C;HvNS4OY&6!TR>d8Lr~CLQUet?T7up`$MK;<yR`_p9&oy-2JexdU
zBkrEuw(XhQUbnB$1G}AknQ!Q=r+YV?<OPFiyOb8JcRqSVgL*R`1>8|4vxecm#IS7r
zHJzhd`&ZibsmMN6VQ|Q)h*pr290%j_q{u4f8A5q5Z8+f~p^S&}_Zd=h8rBY^d^&q;
zI20oGNVg;!bktzb(Q&av>cR#l;WD}=Ok_>wMep>mcCPr%_%k23a`jxxstmf_uqu(e
zxMx4TSyg_lY0}FCOXp5@7;wyt9j&(~Th%nE$g{dBdccjNd6xUvk9AKQO-;>yBpONK
z8DBB(60mmz9ZxYTs~N#cm#X$rVRLH*^50QfhrOikM%&KAVkdQorx^Lj3`Iub1WK~h
zBRQ$eN_?jZzeB2oWFq3Sl8dOlA$LrH6yq>!5|;2!I&qMsLJBT&lBb!>1*!0&uSc91
z5l~NTi*OpcRIfwrQ9`|U0QZfqGf;O|*5>ne=o%cHtQB9f8K1Pw=jtwv`M93>>fGXa
z@aZH5NTFdv%|3?D!FPFk-pT55k1b40kWPA_G3`uTn09Q=_T=o|xb$Ji;pMe?24Ya(
zP%+=R+<s=v`miEWU8s2t3Hb~igBN*Us?Rt#=k2K+dfxcNHN72Rj)g&uxPAlI9V{K{
zb#gh$baZp`cy8u<^1*4LV+hjckvHzTdLG%@0<Si@|3&k5zV+EMa(Vlpzn!KY@oD(w
z-tl#L^WxO=wtMRKr?aDD=ks!Qri1zYXbwe!?%^v?d-<~D^7dobY5nj6Rm$<!!^ejc
z$CXfQyJxLo;UeR2qnq2K6X(=}<?Rk0?>#9L*ujhXcgVf29q-j_2UNr*KyBj-4pXF2
zKw2oLrhL}Xdg3o5+_$$+H`lM-n=+rzi6<Q&?%UvrHq6l_u21M0-a{f}L)p@(ht>_B
z&Fw8IK{SG8$db=LHXUEj3uWy6=G@sIKH1I3TlzC>SI|VZ*8OE!3rdM4d-;r1&>eyu
z@#18u+SL+@Mp!-Oi%PU|`Ygsbpa(^md_dlt-7{4c8M^&K{JP)+pvttj;j5WmO~Lm5
z_I4K?$42JPBRlLz7>DZe45tYC?baiyQw%i*cgq+)b2MYR7}`iFX0I@L6BANk#m}c0
z5gf%yl6Qo_Nn{?PP5g)`Bln90r)(m7KoSB&w@izfG91MZu~O(GbWV=cJ=PMJ0nDE4
zJ(?q_GYlKWX<A(z_F=^*tlk0H4f`g4ItCb%ogFPW<~4<jXpOfrqm6%bY7t<v{^d5_
z4k{8~ZzEtXtkF_>UvGmtRjD;U6OOR|1);Yv$##_4D$Ha80h{3<y~f7oq$(oJCg_}B
zlr?aBjr%>9<O%B7$P7Yzzd{B~!Bc%<wt5RknX9XZAN2Vff7ArO^YIp6!z2V@q%T{h
zA2}eWwTd&HA0^496@ja4nruCY7&X`kV2+z?`A<&>4uaAYeUf~$nratmYGXZ5PPB13
z31=yaHB65`F?sm-DFPt`%^cMC-K7|LYY~iNG&j6;Vuv|$dIBHzq0erh1fYqD)(7SW
z8a-|U#a#;Lek34Qp}|U^VX|K25HH1e!L<i@t&*1p=GeaP_m!Z-ZS_Me!&0)l!c@NE
zRD?w`3O-%AMnU33e)^JYjVlM*2KC3`<H^X%{ujV1BA)htx4~xqKOP@4&@(aoYlqFu
z#Q5*Al>YY}c97Os^<gXQ){2^2T8fAWkpy7Wr$0f2Re-LOzo>i!jJPY*%lnPI_ewF?
z&Qt(C3of1lj?oUGNVFYF*>0%vEkaBtGTwunA@@N*>F+$8cr3iLIjmj3XpD;AVKJ!*
zG6Z7OG{5Fbf;kvKPcPyLv-c1)a`B6ZM5qZ9V-gD<sl+J#n^2^_qw$4C6C+k2;+KU@
z5sF2NIKf&#v6&><vpXQ3^p^vnPWGRUBKgML$6*`lEC;7D;D1i_4WIP{Rp#{PT7BHu
z`r=j6V*%#K>J`)#&|6salAV?g{fYafJl<PEsce%fqEc10j*N>EFx3xE0;CuhAQ{;n
z+!#Qqe}o3N7m7%ON8D^91s)QJNP~~F_L>yHwcn0_?u)X8QVjFE-Eq=@k)!pn*pgg3
zY_=uHH!IRDJOpE(rcemoko!=Gzl1lb0JI*l{ZOJsE9b<lXkDE*{dO~Pk2J04B;m>(
z0&+ok>mjz5jpnR61c{!T)T_|<cD*o=SsEdNm$gn`#hTK9NGs6YxET{8*+f1cOZ_0-
zsj0MLD48Lk9$5)QOa_*Qlqo#fsY%(osj!0N5ILq@(?9K`pgI|N4aSJnN>SZ$Z1uUY
z!dP-#PiZ4YgYeZvrz$XT%xS<-f}Kl-hA0`UGN4+u6(FF00r&^ItcAM89Xo^!4So3*
zfbH@UWnXIgl}_cWn%V?A&m3#vZ@c6`%<yVaBW@I(s9NjS(h7U(RJCU!2iDi{>R}WH
z+qoqSF(rIw7Rafir6pq1*Si6Dz0aGHoUhx7nh`zk=j)uW&j~f1ec*N!XE+~TUSk+c
z_cZ?EVTTl!gj=0rAqm$fTF4U;)(yFDvJu;_KTjEYUuS<7Qt*AWeGkl(nAxoTrQ!Q^
zl2ud#^t$=caP6u|k;#A&RiaEGks+$1xl*l^+<1GnR07&2dbMca;(gglazV}yQBV*R
z&V!{YeVjLaQEtHCopP4lh(<y%Y%kt0ZM_GE+>l4wFc7T=ND#DfUFrL<L1z@0(i)^t
z74G7Hc4acccEY~1BG^ak_$J0^jlhnAZI9a8P?6qC`LR*1jrC%ep|?-08hKYD3N!|P
zC$JpBmEMoK)+%iGERY5i4o|>{+w;VX6O=U(ZibChe?My#O$7=)K-g*=GQvX&JxI-h
z5q5x*9VT_e0!J8Yqt(aquh@qp4L#VyP!J?XCj+$A#?8f$IZ6!s9W~|`DGJgAI~pN%
z*$Rjm#r7oiE&oRT?J~F1e$`9eZyEm(NJ8Ra7HNW3PMe?Qcj-u5WI>Q;g*NuH6>tN7
zgYwQA-d3z>_|SxfR;=%bfwQ(Db?Y!wQ&8)%R6E92j6LmhqS1FojdeoVeZaQ#%P|!j
zV1}&AnR3gS5RCSf#)eZvvx!kGwm@s@_oT6Xz)1ZhlHWsJ{q%1>Vhl(Wz*cPOkd{8u
z_h&VdV@t~~JQGDjm~+|4(sKvnwBbj)Sq?T&7{kd0+D_lu0NTdlqK&vEokm6>5;THs
z82r}lx8|zcc@x<lSEG043GTNQ5fM4~S#ndJb6C=?!0(bwbN9iFWKs#a_*sOGK1^w1
z-EB20#xh}5{U;rDF&TBm@PzD~H~W~kSFLSXwg+#eeop}s?pJYuLUxMakC3NX8Kyk`
z!-Uwc1taSUWZS?YLu$zgkJK%KSMW!l8r|g+AY}J&Sd>%=756+VcRt=M1x`vS_^O{U
zPj24`puzqWNk&){+x6;dXe{n175Ck;Uj6OS2LZONH<bdsnfTr8no+bK@gJ<>&256d
z6?)WV`WL8}PS}`I@Xo7Ja3@7NVB0Y=2k{z}BYwiZ753x55TZeLe!)1o9yW#?+zG5F
ze(op%i(&oODsqAK?5r8`3j9mg1-<UB{e&UY@}R!ea5}u_l={6s{8*DHTM6{MXNP0!
zM6b}cwQEPe;YP5SGB9V59$urdkAniNS?CiAOTt#!r1<r9M-+xky#&<;73-0X*(5m_
zb0-;_DRY`0IJXJ9z5_kpR@h9Rn-d-j_<r1qdGL?FX2puxOJT#y2(jU2g~#PaQ*$7U
zBRocDtoOWBVI`oW+)tK50p=8OT<~u}!WAP2Q$AuBw5j5daH+RoGzA#FVv&=uOl9!i
z{N_|Ivzr2LJ0uc~6UBC{9cxv8pT&V`OFMmRB)m8kJffEF7mXb|=63&-9Le^?_`M3m
ze^;q~E9>)#fGE_wQ^!>3rJQ)y9vr!Iu9n;F-kA`|0nUcV-T~PYJ5Zxs&!yx0oS<<k
zSnm@DmyMv#!@_$&0dCd!U8lIyI7iLoV8x^KLp?aMlM_fo%aib*MWVdDc(ipF9$x?)
z=D;5{jmv1XvIbrm`L5g2?<eeClO6&)`fpN^gC_#Cw0)Z<G{uk2rztH?&(udGC2q$Q
zOBn=F_}psr`&)yJp32RU4E>Pb#8o9(0nW&$C^k;Vbj$`lcE=QP2VS^QCMZJ>5$D4N
zqbZ@?_!vt8wiMZ=1ZasLl2_QA&4&df!^kxUAmTu#%+Ru@-TLN~wpOZ7+0ds0TFVGM
zMnCM65YlCII;<Z`?wKCgbM5ysNZr?kKhs6bsf}PQuM0Rz^9U_s$RR_Adv0B&pN?t^
zOFNXum1Y0I>K4Jk?^Rr=RBWy<NW*(I6?&F66wMm1&2`LX%}v07xG*kUdWO!OP6-gC
z6|uG379~!ZEfm!hugiY=acO&F=5T3y^ht7tAsxB`U1TxFGZFihsZz9CUe}-q-7&OF
z2O5;*y-bvGN&Dr%e8o(bJkk3W6&RN*kbM3`nJCvT+VyT4K^lsya^xgp-Ii_Q+!13R
z^6~VE4|x>UV`q7!Y&CV{+H=skrw@ekk)wL%#ptOvdFANOVyTYb&L=Xt56s7$@}SN2
z)A`k|f*Wdr<QN%JLh`h%=%fnNtnk1Ot?#%+IXwVEyDO|*?p>hk-5`QdtDCa<f>yK%
zY68v*Xotx?*bOOsKUx*Wge^CEbU&<i-d7{=a{$*obAHD=dtM@I*VeW0_7wp<&5!6M
zu5snaFS86kj+?#>kLdpS!_{PKar5l+=yOcpEr6u6{dU>W)zB*S`RoTzx?HP^DB)S9
z@70oi&bsq#D-ITrz<&U9*&)!8#P;=aX$ca&@ii*KYD?7Z)A>p*Uxvd0y+bDSr@Q(_
z+1K|G6T|!C(6-0xZH3S0Yv~LUuc#}ZT=ta(=$bP2c}b@Sa(^k*)x|y&Gszn(aJNEd
zXIa8Zt)4mTQZ<>Z&YXQMcVNHcx(=MEcrvt75*HCu@6<8C)}Z2Xofzy>&afxU&j+pp
zFX6gJjo>LuW`R)tcLR!hSOSIoIdL#h5@{U;y$%0WLSq%s991lS71_o*O0h^SYm|wK
zmGyYSDk;VHQ<i|iMc>o*l=|zd3uws2S2Wo|Y@L)L)**F}f7~7ky+WC+4z;~gCHD_5
z|96kailk9S8UvNPN&u*Cc1nKvXXg($<P31?1a_^mPL*Ezmrqa7)*7BG-|(u%Vhx+a
zr_G((abng=>u*;3EtD9PJ6{r@7iQrLgsbU^qzuiZhZ2!h62?eu9Xc3;29-rmxxP$<
z?%cx$2Z!`DKzGFhKSc4yh`(JVX^4h?$8BpiAk9#re8VSSg#e@b%-=oS3fnQ=waFON
zf2I$jO@O0KVy@f)=<)u#rtxt6U7*GG%)|L;q%jx>k~Ub2J(mA?KK@EOTo+%pODyJj
z-6Lkx$0wDkT1S<^92T*4C2fEaohSh_5XTv&&tCudg`05wJrInTn`Yr!)&fB@3SkR-
z1Td0b`k^=llbk~>tIz1scTNmXRBENMD(Q$-RE?}bf{rP9`^TgO;^9lR=>&CxZB3t%
zgEu~kS^AC8MOpPRYa}Z7qCknvA9yXp^I3{_#LW`fC}#LV-`XvrV|38$aAPKfZZ@8z
zP5|9tEZ!UQx5iWwMnw;OCRNr3X!uqygs8$WkkAl)_!v_h=FmaLXr#fNESiKAHfk*G
z_BB8oKr$Y%di_rYQbSuW#O%RkEmDS38*3!3Q`U<VOf+lc(1U>$L$s_ZuSh?pjQij2
z8rmAe65*B;%o6sV2*(t50SGO5E&I10IlcRLKV&WUr8*%=d8INu_6jPvp>9PKFoXVR
z7xVTO!|%KSvUC&V*?Yfn8=$PW00q!Lta}Q3B_B8=5%q$v6CKCoz2UjjO^@_*j@T42
zhQ;ATI`gx->XA*HNKC1NA7tkaUTREjZ)tr`ou?tpSrY_mP+x8`YN!sRE}S7+Nxcu_
z81n+bYN9H7#CgY{Wt=)%MD)diEH_GVv$u#|(jJm6846SsZA7wfPfS{C-78ByM>`U=
z+P;#o(k99<%-Glx2b#J<QxD%yN3k>;Y>B3SsMDE#=|$87NcJdV594Yq(-3EX$Lms$
z@1D3BKGzs&Lu`u4<Bi|-d}HiQ%;O<u@uNUm9MEa9g?e>_VXI#K>vj1m9_TQYMPoEN
z4X<_~<oXi;7cG;FQNPc94+)K%)DrD&KlSE$o;R98pypzqmxlD1wCW9w%T)@a!Qd*H
zo)l`6Mt?HAO{+hZZ&gJo(cI$g!W}!ClW^q&u*-{<r;ni{b2im$jJzpkq~xRaI(LA)
zi7sy+&;8}5tEHEwBEk@g<RhFtQnBO5CpX5|$N5Kbb4T||7%tXSS{G|COEGV~^+(sk
zpc@wRcV?YB;J24<Q>#WtY@;o+k!RF^5K*q78gXvsb^U0ma0l0VcNZn5tD#x8v!iO8
zZR)KplqxLGL2oa1cGmSq=Qd}aZR2g`b#eCs8{JZ^l+)022hqwE<AK@L6y4GcS*fn^
z>%2K{+1C2o#hv=!rn%?C22ppFv@*k+A-v@o(?W`SiUl^OBzNs|{3Z1^%(Lc9G}IsM
zQg%^WsWW$8C0Ajd3=dK5Td8_G=+RNd3g+jc3#H4>1?I};TV{NIMN^}nTIwFEuB=DA
z4+>lFZ5{PD6fcghYRTDrt}WK0Sm-p>SZ`puDtEJHJ#erwXGcR<PkAwC%l8+{N%~9Q
zP+K*MM{_BuqMXNUsD1r)-7qCuTOB2fHt%kKttz5HD#Nu&2z_A4x~p+l)0!vCMM2Fw
zQnsVC|9qIYsuF9aN>BT6vK3uur_^&@w&J`9R!z%n@R)U$^6GAkuxgJshkE`SJx!x1
zHg!I3n7HO)YF~}<Fw~{BUV*!PgnU@E7?LB;SpKwJWx>s&jy7JfV4~A~nfk3}xI?i8
zdEB}=V$+=1-(L5%-Cm7tjlW=`xiK<$F1is+Y>UE6$6H)yi*h9veCL4QF{DI$KK!|V
zQ`F7#n6J1T#cI74*=P>UbDlw#vJ}#-&_d=?gRdGL6<l%TT(Fyexa~S3hq4mIN>k))
zn3J5dR0ZvP+ST5r>+?}P!}b)Pnz|eXmhQe~mg~-84R1``Y13$a7{#Jgb6=ciyLed_
zYjmovIkI&2E!uF-a@3X{4XeED&UD29Rf+AfYlvEL!^XFPsZ8yDFRDtp>J3$li_7W_
zk<?&YuP?17E$G=zor`8P^A7JIKO3B2J5RJdC~jRTxjAzmK2UUWW;?Wl9M#%fw76x&
zvC@2p1etJhdHY_1^I?`{HqR8)`dh2l7%8M{ay&Ii<Q#H5Qd(^5jWpa9QqayXUF}=e
zc3QMsTfu>~HjU@iwT%znXC%eW-)R=_+vkd-Y$G+$&YI<%zu*zfeS!YJP8Ak<rvDmL
zUf#jh$l1`?fq>l3$VAy#orZyynVy!BnF5MVz}d;n)`47`iH-I9Vbi9kr&sz5oc;p+
zzrf}%(57epe(33i{{rK`z~x_<)nCl^FR=RytpC;U59L3M{)1fkUv2-ewx*y{Hg~cz
z{yz^d`j_7MFZhSI;a}{Z0nGnm|H$x9gX7=ge+;VlxA2e5CjY`5{)PP~BV!6WC1(Q*
zV?!r$3OXSNV|}Opmtv&<68$6if6hbwZ{2^4^p8&d5#^s=|55#qTK-d)Mt@2Ek=2lb
zPRiKb)z-ntk^I{h|BPvFYa^`hWK2LV%*ja4$V$(`z`?*w|DEJc^Sg(Pt<nE~qq2j(
zot?1}0iCG6m7_5foxHM$I<2U)m6d_Mjm=-<E1Q`+5`6#t+ajPQP&IaN{2q>giI#zx
zfrW#Sm4%6hk(Tv8R>wDiu{wdYxr;Huw<L^o!o~*X`Zg2<mUjA1W{$@C1TOzh&q~Wg
z`@NQU|2aOlHL!qU=3x1sF1@jTUv&Go=pgKDEdTQ8F)@9|6#3`#zkBqOv}9s*T41+k
zYFa0p5MxSh!Gitr*H~o2Nv+WZ(2$Q=81`0!O%-#$x=}=;^^USDEU*9Ik;I3@lZVew
zWx7La9SLelbBDGbCWFT{iy9-i@6UrNuo6}Wlw%(jPBtrKUA~EObKRho1!qjMz?2DD
zp}w#@3JlRau^ucnlH~uWIie|Id|j?D>KD?j(mJDnmBb1``}VpOPUH%kQ3#14-)rG>
zVp)?6<_Te?u(#VwWiVrN4vx;>&CZRhwwiaeFJT13F_g**_m?)o<uGqsT+28;5Ojy`
z_sD7?1B0Wss4Cl{%Q<7QK6gIGh8*ZN)6#>hwCIX|S!&rjtM9xG|ES>JW=D*GgWEH~
zyG&)x#@lOE$mPy;L}IK!vxg5FAFxXP0a@7{=l3I=6JQf;Eh$kUZ&8IBngXt!;s?kN
zOC-7=e!*smaMrT~X~R-7iuFOUffj*Ak@P~d2>bYanqrCY=2T;dKon~QFoDK6vj~>@
zcPc}%x*}HbIkXO#(Aa#M=rpoIGpRf?t-0jV)1G`9k=xks+CzHfP$i16aDnh<Wk)n?
zM(IRN+^u{w<ck6=6l-%gA@RB7kxpdAM&mlKSdkJZyo1y&4eQW+niQ;~!F=rqNd)V*
zB|n<5=FPVgQal1&w;v)}kk*Z;yUOH&8OJlWHa*>MmkvE%Uuz%NeA{}uyqz9zN0_%v
z9-=<$r4gX|ZzIiyqQ<I|(&JX4m?i3Upo}V&>fmXVtcQK?yGKtIYCT;Z?6ufVAG=3G
z5ulq!E)0rU#X9w@1WK3VjfTVK(iE`u!X{<zGf-D={g$Pt6w2rA{^c?#GO+4R`DKA=
z8fmah6F-|n(^S$R83XnXDV_mpu%lnI0pY^kKrAYGC{hqtjx5hp%d)M&d6p7z+t-S*
z?;-a92ud6W9s3>FXRF+RarU@F2%RP~4KSHhs0Ff&!;9L7GXORn;98_RO7)j4)B=@i
zWIGIorBO40Ov+b5x>k!<7gy|{eK)3eo6XhW)SK3ms0^OW)ofDB45a#{QMGalR+nwn
z2dn{Ta|VpHx&l>ZhM@D|G`@2<AsUshde15iJ%OZQ)fkYIRiwDOof7S#u9e3CikQ{A
zk!VMT7h98yQn^ZAfomsiLhXq<9w9O3#j?C*`)1cU5ogM`z<~>mG9^bdLot@94@(bW
z=T39uVxD2yg4o@2Y;uDWGFN)6gCZuGqZ>~u*D2F04T+u%aSjc)24?_ndcd{Fb`TnV
zPs)E_wpgdkpg3FxmZ?PjWQcd@s4-XzaMK&EMXtkPxO}M^W+arR5#f8rNDN9t`8U>K
zW-vUJyHB{AZrKW&v{dtu6{;QbnNduo5DkV$C==h4SLh67B8G5mb0HEZ-*<OGHI)uS
zp;YC$AWV#aF;O)cnPf~QN5`DxOfCCqqGX~qAcJ7@4Om^t1HzK5P9UcqjD+OOipj7&
z1QR&cW|btqS+1OluIGSMog>+}w4!2U;o4NCa$H)ZmI=kk@((%ZvnXnt>IIj>{GP6s
zY8{HY`Uz06F@(yrMElnkbH&=$Cq$fL%Gj^s6LTyb0xKk&7LH=c1d2QZqLJXzO;n=a
z8ajzpph*Zl^(`|<F@|lycH;5lVhT&mEiolUoa4yXBt>M>FAkrYMPC|Z{azz)ST<mG
z=Ny|BE&{Dg389C+WiCs;X_U?V#`H{KtAyX$N?x|3^4*yE&8n4+7!+kP!IV!V-EfFG
zx4cfREYjHIvJ!@LX2ZF_X=+2D{CF?FG<6c_aAd<klGoU(B`rZtbA^V4<TuykU?5FS
zTsxq1Jc%1f$r~~OhdzYj(n_9Z5)y%AQGo^3@bl<GgFy*avjh`nx46M8g+`JTMQJtP
zX#DMR@aY8z&Ci%D2-=_woGSHTJR(4Ch5vrR+=@qye<8A+g|G|DkqTphnaBooKi;IX
zibHT20ivsTt~OcaC!^N$$1(s*qN+TFBvY)ad>aNeBxlj(ipyPtIatxqf!Ks}WSZgl
z_iAaQlGd0(m6dA6VlO*v<HQ!XzSqux)rLH1W7LX@W${?@BCp-7qoEoj;Gix!=9J>*
zDvefuTIz!=o!lx=3an&Qhpj*OwIJZ2Qnfj}Fb57X@ky;VV`;rquJRpw6;}josn!A!
z2o*l4WS^Z@^@r`lexSSgjcS<!=p)OyJCJfrHuIHy-ckN!blUzE7;0u9$2ByNy4N_)
zPz!*?!MOBzrDdxH+ldDdn_O#|{t^`1x99N$R~7>IUe~Us4v*5IdU$raF*xm+e&Y0E
zsdW@kMm{PW2CbANLpn7Hyh0%PPb3jgjn>#}{eachvld{ZaGbpU*qxT3Gl`UKPzr&Z
zDHu5iOGP%a-&BRht#)dKJaoL)xx2qp@`{3So<`#ji&^-FPnjx%;-^8p%!rg)=!Qpb
z0c^m{Ur?^=I2bSP={UWR1#4R*pv-RfwD3uz={lndg1qS|28rg7%$N303ol$Xix4<L
z89M2CNGZ4`<o=v_S;pl4$EaJIST#(e73jbo$1GY8^&oF~a4_vcqa7|52hJ{CmT`@s
zG@3@%i13=i*tbqP3|md1>-jj1;z#v?bU+B@L_dygBv_l2rlB3<o7ovx9i*_}7oxRF
zlh5VrE%P+1sggR-G_(ShS_&)L84(<=lIN;R>*1y_RW>lo7+XXlN@=2v)cew)=JG`r
z(d_34%_HvD3`(N|Guw)|8x$ihL@?&9^1w`W8|4F}*sU1Q`6-;|1e>%?**eh5i;j?6
zxbr!Y9QM$7E@t+5nN99S(t7S0_@dJJ+4s(wDYRo=_T8{gQu(G^TS35Lam3l3h0}tU
zLH2vO$tE3@G?`}y;PIJrP45p>?(_R%<ztAAW_}xRIEL(U#M|$|Tb6_gRfa52nu`Cp
z%?}zM)t?h|oX|$2VIG*I;(7q^k7<Nhii?f0mr6Tssb?&SLSv!Tbg~8%nv=Gp@qvv%
zONUY6G}}!${9O#4_gBi~w0v1(9JnclfV6KWxI=yYnr2A*yL-D=BP%J0SIkGR=4ZEh
zQ4K{^yFRs_?fY<d2O4`Bt|VtvPpBW{hcz|c@v3W7jDo~u^~ghEXM@vsbUWSe?;9Z<
z&+z$&hjgyjc(m6!B89~r5Ul;-D}u4eHsjst&*UD|NhvjWv@_d!pZ71WuC`a+U2lZJ
zy`E8H`otk|IKil}fp!)1R|pTK2YRly-tF%m?q7E+dfT7RcP~BH-kqI2K96@Pbi>HF
zbw^uUGmL@|jxK>~h$dpGTWf1olN|sS{`&>d@qTb4iMUEG(nCw=YI=mr3gZRQ+fSU~
z*jk&bd$csw-L0?lb<x{8V#M^CLnR{46%|@KTjoobt0g~_ZVD<400$JdPR>Nt*xK7?
z$}{GGR26tnPJXhdV{dJq9Eq~I>N;|14spKj+Yy)ZhiGkWo*k*Wy6U=eapK(xUn+=!
z4a{wwU8%aZw0AtWZ_M4smWW{h2)ib%$(XREZ8=|v(&JrYX#<XQfw=^%!J4q8ZrQ8)
zk3PkEl=wa7pHm+1l-;T=3ks9PBd>++4f2}56Le~ypAo1w=tP?MbY*=Mxjd<@3AXZ}
zz&o#<ew}Ho>Y{T%IiMVnSJtO?MUxBj&9iJ*>6}Xbq^0KwJt4>sxY-)hb2arGi&8!%
zDx0Zx2(4E;)WUj89-d?P12@zywANu)C>gjyhMzZ!ZwXR8%f|LJ6_r4A3mfhguxDJB
zY4+6}afaWJCsqYF&c-hL4M?73!tT#pGR{8tg4XfrX0$78EiN*!OvR2i5__c44OC=Q
zVdc&a87(m_oQ=AtUX10Q5)_`)E?*gTOT4rn9{Rd1AERQt_V@RypQ<9gs2Wl(bM9<D
z$B2|y&a{5Ksh)_jeFV-yv{-#K!FoWy9{4R4sEQqR5V46Jz2UxjkC@Ww@2%|vuAIFU
z?aZ9ICWyAqc!3`b=eQ+&J;+Wqy~3UCpcwT4O@3DSdfS={zYJDX1&1uPDEBBX%CSV*
zrM&TuN#|&<Z7tzw*V#6C>y_!bW9{l5zt)kx?JI%3EuY8y`3^us_sOo;1Ia>IEsH6}
zVRTy|_u;k}x+m<M&oL+GFzO|nmy<HReIS~EZOtO`QMnk}IZ83@61^w){>Y?EwCU2p
zO72Znk(1-SKJB;Sacc`6?nOq=r_Iveg*7okeLJuw-|1U`-;T8#noqQD+r5$i0P%I_
zKsObxBl^~}^Krdp-0?B7HL!Kw!nd>esU!MZ7pdp8FJr^j+sEzWZ0bGYSnuAgN^jxp
zdhXEH)~9{*^Zs*h>T#y#fcVer=ga*@jnCKR=NrLbOUt!h=TDOEU$=YU47*Pmt|@Fj
zsxwwGE+6>7$=Ux+w$bnZ?Eh<GseikVtgdKm0!1fbV`S_`phf>3XOW3j2Z~O~+{5@E
z9EwhjK#PHZiGbm|R?*hh>2K}dsIH2CiA-$&7XP{n{uYG^w7BVo+1VMH1zDMdSww~C
zMMW3{S%n1{ggMyh*;twWpVqzuEQ)8#mkg4V<cwq_BbgzJf&wy<L2?iYBZ%Z7=OmIu
zG6*<iL`jkdNg^;XAd-eC$dL08U+cZQ@9y5W@BR1fov*+Ces%iPsWaUb&N)?Gt*)u5
z%<=zb0ZRJ&u8-p4a<bBY(XZo~^$Wa(Hk}G+n+|Y97XQjZY@WC8oGTaqI0?#^D^6v>
z<=k@+r&lMmRe^NsDo32s+!1;+<1$Uj7ed`6Zaw{umAlZ9hFa%=?HFDyF7aVx2<R92
z?3MCbK57l-Z{=+~aS#A}B;82K4fq9rpP!tZ&zKgu@XY2fX?P)9^m1L>&F#Rh4p)9=
z##23W!OUf*U7X3kfL=n3@MpK;dbsdVG}Hu3&V(r5TAZ|#+JyM4wOnOLdM&XLoXLbF
z+Zy*pSp#>nZN+4lzqvdSz4_<U?b}U+dW(#T(cgDp2AH2u-sxAN`w6!-A^2(oVu|90
z;J9}!)7_P~CO@4?f$m~+*Bxcskhn&=)v~_q)SbfrY(udb8C1(o-l-~!7rh9K&2|tL
z-Ol^6_+^WvIgwrvHjz#ai%^|f@lUz;o>Pj-zc6;a<dnsA;+(XML__HFx`dorhm+c0
zu1Cq$VtLl$zwCs{V)ZRb<%E*#VzGsde08$Mb&bqj@HnXRFkWP`h-QY+xOcfV;rcAH
z%|=J>60yN)e2GUE<XS^AcZr|C&36f+3LJFJC$KAc#USN#qp%}S7li{cGdrRmRTnp%
z-3+zb2#>p|WU={dC=V+@q4c9?MZ-S+)@;HNCAFk?x<zEMG8g4%L!x(y0|${~$m5HS
zPq7zUNKdiB@+hwfNHx>2A-vf|)|hOqe9=NWr{e!eyz4n}ubOGV5MJOyX-u|Mu4*A|
zRDtf|sCLI&E_P{pY+hDi?z4RAR>WOjF-$cvHT&sKAeqKT@`g?$S>h`0dFd$v9cxzC
zXjTZ58DZHoVL}ocDp%lf&?a>~m&JBkz#;ACX>tuu_*E`$DP3C8`(olvHPg7EU5m>D
z<HTS8sMvrZceBeaW4mCoB|$_5!SjjJDy9y@<wlnaqr?yw*RBlY%WL!H5XND{wZw8S
zjR{P_2A+PKvsVqAtm%eNfiztUp~NT!i8W(o_*^YvQkP#7S#qJYR7-RKC*jwwe6pPo
zKr34`*?2jXnY3EPUFQj+DyBk1?vE}JM(iOrX3`HTqMlBORWX$s!ausaF#;I>6Y;E}
zzV+IBhu1lpd_7devie;^pVduE(z$XP?IQwo^$g^*{3a1uqjGaV1I^Datf`3~mJ4Fv
z*uqjFH{82#JG+^Ue0^eqCuGCD8CeX$u7=zBzMfXz$J}{@4hZN~uwfJaP}24)R8EZk
zrz*R>jq)KJr@u=rhQfptw-zs>Tf&4~$VR#{)Vr4LPB-rqfqB8v^+_&!!g3>{QKz5@
z3A1bcB`XRiV1Xzn^ecp78}8*?Qt5+ooWcoM#LJ05LvZs{zQ)WW@VsoqD<jBiNq}24
z^Wm?x4oWL!B6IMJagcD*1b&odlq7^sWNNHasfnUW{k18QoK=kQ%;rXA<n7v<$DKB^
zxS0z~vyqcAgDdCOSb-`~0pV~%${THXnTy00k<TEk?wxK;Z>Hx_%nDvK+wffEv3D>N
zdb^w3_tkX!m5lA2x-WJzc!rR?w}tQcf+#2A^t5ro!R5RUb<@(q+BI?SocB}Exz(4J
z!nBt8&z|V8W90EAh&EoNx1m=WEg<WL`1E5kql}w9Rzzu(<bZ=vWV&Jap<854l1NZ3
zC4sf7%Xv?$v&o9ON8EgvVePF?K@#j$IyU%~$elONN|m~9cGu!9m7$+%fm$AgJZ8UA
zG?M}DIH^Jb&o|L_kXd8b%PCw1u=%Q?)MnTm1nao8AvPXSJL1@rzVS%*%`~)1iOFKt
zO$ooxU}k90#D_&UjHJD)3_KGIwUTNgnw2M7Gnfhf!G<1&<3r?{CL$LnGpIhHVkdPV
zj23s?+pv8Kj+mEpa1lJ@OKvkh=7(6@r41nVxXAzryM>*5(3|P2{Wk7HqP$M9FSh6+
znWCQC^?6^B)x85Ii<M*(nb!s5Eg8zcP(3Y^HTl8Kid<$1FG{3#RE1x{E!H1Cy=pVb
ztZP2Di{RiUv3{aWa(&CJ-i$Sq>Suy*sD-u;g%+C2dOQYolY5BK^zB%=#mBQO@$MX|
zumnnbAPcj#w|4fG_o3Z5k<8sYr5`27B)c=I+!I{y|LP1dZ<%Q5kv`L%${UM4x%1k9
z`;hTR`WQy9PyT1ZX`<u@nUZHDwsXq6sZJ{@+KpCyE#WE~Dck}|MEP32(rRy0mIahL
zZlZNhJsDXz`}^f&)5o%dZ&pRE^lQrIkAY4fRC0eO;=U_Vau|H+Sy?>NFI_NlFL1sp
z?P>><Xf&3fA-E(nb(waxhjHc@>sK!r?Dz70Jr)h+G|QisdXmV{!7XZVw!Mt#EKJc~
zeW-oCNmA$DFPcK`tn+l`9x!wrSC43{NVL~TmUIuAY=$+D?yvUIe~|}SpM4npu!3@r
z@eXR>fUXNiVFWG@`-%$L>AJa0aD=QmO7cx<EdMmoXVwnd9S{XG<1UxoY?z0ecE{l|
zqLzFz9Q7#ZwwLk=98r7YPB_XaXvUx%D$mp~(|^j@KxkI}pd%(~4nk>G9taJnoFRk`
z9<pk~ju9q43>P1MDZCwH58|3+4X=ktl!#Wvc$Y13<>>qxnO9I-<AipDs4XKi=z<@`
zO@3X_GC)0S${4e%PL2Xv{I-ft`t?ygVnNFor4%#rIeRQqMhhN<dj9y3maPog%0>o-
z9$_#OC&DEiK{V(*^o7YGU8`&EgFH>R5NMOjKXrQM+FQeBcm{TP>2>?3T$)FgWFuod
zlyB`)ci5-&_(~LT5?0n!^<1pRXwd#$dcPEF_B94a2+LP4TaT?;x%J)!TDe#mC4df6
z6##tE_(Adt;d82mN^#Q7GsdKM7G9{hSosNGWvKg+uij5Z5j_K+Ps{cIiN>x{WcyQa
zWpOe|bkc>$RNi#srK!|uTmFfpOozg!oq*s<J8+&LCJ1)9pw}bch_|0M$Q}l7Fjd|^
z9AJOA)pwo0u9e^VnE;TbS9&bD<90K8-<@maGGsVkXDj?#j439~$Br}j2GQ$|H`6t(
z2}WG2HD?W8MbA2OE)tQ@^O_<O_<kb?v>WwsE4f{gX)28(C@*%267*ne<U~X0+y3JL
zFaB$(PaRiz&N0Iyq3W0n^<WP%rf$v?#f`ct>R&=yTwDh2JY(#EjbHrvK5CqGdVJwZ
z*;px87)O3KI46X6$Tx4aHH<V)WP$Tjr#p_UG=#D@EGka0tu_w8=eyBG=@#Ce6EFj-
zH`5(!GDTfaVZ)!!u=Hz&E97f7G|Fr4ZJ`20<Wnn#OXagShVLC^+ym9hnhh8_fpg`K
z5@DCi&+nEia8Jn0>^7Zu75by6YaS;WZA>)ce&5hlJj#V#@_}wv4L|gbn)vY;l|LPP
z>?+md^jKyX9c^%47G;}D^Emh3ksa)E4a9=(2c*sg!Q@Leh94ef7~}`-6!RM<$Dd?!
zjUh!#j$^A3+OU=b!(6E&G_37R2C?w*u@m@GWL9^pV^4~S$k?Z-3Z=6Jrm5RjrmS{n
zK(dz{XDJ|5VE%zS^~kHw><Qk_&5HDL4UUXqmj|euWHoaqvF{zmgE8M?q73>z>9yb<
z>JkF?a9K^bL7l+N>76KRoVVpDS(_NU8Tq#)O|%7IRtkq*A+N4ii@mW|T+Lj2iA{2*
zJ1*Q1@>n{ZO7DGUJN=v-Ts)FA(1Ipp+k36cY$FQUHbEQC=hf>Uu<rof`fGXluZRWm
zQH2?6ip_hhZ9sFSc?q|f%9f{Vjg9(kKX!A@Tx6`@V!6+%bD{xiOrMt{=mW;rNHEOK
z@ODd16tKbSnn7QwRb12oFvfR&LDn5yB7PeH6Ue#~hsw_cjf>>7;pe*y?j=dZ*+nQ^
zUpybT17-x+m%NTd&!F#z4g!PU8sWqt07mTcdRxr1KKh!txBT4a6*E}CdzOr%v`PF+
zIBXcXw`=u9-u=OzLph(f;%6GPiNq@JDl!MTw=rJlUj2}n8sStf>fMUDC-urvknj?=
zyc482nd0aoF8qLX=RC!vuL71X_Da5eskkml4ve-j+&I;mt4k6EZ%64OY~$#_%m#hm
zufm;6Mv!-AsAbfe#V0(Mv#%t~n#EUnH)vH$$yIe;QCxa{q_&Ua0K3=e*4fA5fTO{E
z(ZSL!EZFW39$9Z_Y@DuJY2rqg>S7y~igafo_i%&hs@6i7!Lx6;zNKY1f{nXaXJI<9
zWv%aboy;CIcPLDF^2<;G#=tk^Bf&3Ax2=b1y-H6dyhXh)Rsw)JEZ`GSc2DxlYgh(H
zK2wa*&j-)#;&{L{vz&=N9rsxC<(Mq_r8Vne_ncV<zMqvKn@=oeUAH3qMpNXWTRygy
z^e@yY4%G;Lc3ME_UyhPZb!Bjjcd!O$L}GrMj)h8kMGm#H-WT8X?yg~*5DQsL3+u54
zUo1LX-l(wVuDs70e3e{ebGtH{Y+-w6fo_WUvo(0*6n%1R$GjP#R7+3Zy^Y*!L{VBE
zcLCl(beGafmpV-Hs@{jJrCHGz3H=y9Ys{v2V{O5AV<?d0EEh*6hj1k9^Kkg~0{N5h
z1DMxU`??Qfbx6QixN%qFFI5>o!Z{RcYo0uP+s#~$E;q=D&o1Ci@Ihp}ML=wqOWPQ3
zU(yi{iA(Cd(&3|kB9xaewnuXc3;IdPu-i3Xk&GCg#|$b7i3tyWZ!?sa?2X0UYj(zB
zPDysixcrsDs+UM&GFyIxR7>b-Z@+YJqPTm){fnttq~F!&ViKL+vk)n}cpEw4`?<yn
zztiiI4FpE-hO>j|$%Gm01{i{RA225DS-ZMzw020f)Y>Pg+hvEy*_m&eyUNP+Hk1rI
zc@@|HiV`0uvrAx@NtZ|Y?|@(8M)%4}-F++0mAGfUsj?A>3I}@WI(3!MIxtLf*xS;s
zRJ3;`aMNPr;w>n~dyUh54;5JfJ7DhMW;&JN-Lu>N^q>ckejpBG{hTQeIxJkLALpb7
z@jBS)V%g~#tod9+J}Tq0B@AHR<RrIAVOvWjwdv-f8WqkW38_KH4U-MXABR>|0R_y{
z;I-tdbCXRhnQ_7vYx2W)!|dk<AK&UTtcM%9=m3o<uPRy3loZ&vFxCzs#y+G>DjcC%
zi)YlOvT-5fQG_xOQx&hpG3sWT!m9wUb%<(od;(1K&bw2@+2FOZM$OFI)=_o^>Dwc%
zjqVC>uo$MY<U!f--i`zW$KA?P6od-}xxc8gxEN)zqrq!67pbR`lDxQmi#RzCfLEp@
zm~^|I0`BJ`i$!D?67xB}7^NwOFAia>?j&o<kiSNcXws)L<DM)!Zd`ce$9b0u5^Q%`
zHW5+)F7mxxs|!7ZC~rbu$Y%kXlqz~Lyb<u4m0~=UvtLNeM`E{ESo_x1783J4*eY{T
z1u&nFCO|at2@Dy(Y-n!7b72+S%w*(KwXx+k?W804KiIc*EDxZ!TI{=!Ly+JNvt6>f
z;fP5wnW$#*2l+M<q%`)v_czSW><}Y4-vR1eYi^9_>&wLn*t5Co7?D6m-hMIFs2dVP
ziHi?m&(=29KQIzWxQ@B{nr}>>10#b(D~+?bYq#?oDtG|+dtJ(U_v^)QYwD+`DjV{c
z83cr(U)q)0{S7nx9V0JXj4=u(5h&<w8N(?B2s}9EPMVNj=-Rd3Xt@eNMDoCe865|J
z;@CraD6Ae@o~&ts%t-s}-`uFO2?G>;LkB+rf3J5rEFP>i?Xm&(ZpSAk1?gFO)Fg`t
zXC<y1ZunaTYlufOxQ{xWZXu;FP08xFIM(U~yn`+?OtkV%!-ihY$Pf={BP(HFf6OC~
z9tJaAZXUXRlusNQv&|5!SA1jns-=BPs5t+ix3XtMMB5S$Vn!25ph+dZzBau(R2{Vn
zL&KL5Uk2<ll<Osw0QzTM@5=*Qpc~ELsK?7%9j)9Pao4p3Q98x{a9d#8(s;nTZ~anh
z<E(2$OGwA@mn3>cTqoW%Hs6hLZ_g?gP`l(yrDT64xVJKAgiX&91XXIGw4W5P%iyUO
z&T<&oY(}0DQdtiD93H%K)rrr?Oq*!Mo3;-<xo-(#UK=*`xzD?F&(rSMzhcn~q4L($
z>s|fFHzi+$qaFRkW;~b2QEZn@;^<a!W@5$rmigH+1Q|orPmkqJYsAkbybXgT&2kkT
z?UvFx>eYX$&lKMF4uZ;-YJoSD0~L!z*6&$b`Lf3U4B`DyMEatv8l*^1rX<Gx^If&*
zc8s@I*Tu$H<rt7TmQJC!2qs^v2&LBFWy=u4Q_C{h`2l&<(Jn4u6Jl7)G}+~7lHKyQ
zgUNC-Ty88g+gfNd;x_E_GW5j{f*G4=r8qJa8KJfD#x^E(-O1(1Vd%u-=2o;igoV6x
z8&gq)vJt~S-)G2)sE2TQi(Z|r120#En@(r$`IWj@<>}0QRVe!E#}E+@4Aby*WZTb^
zqOa0T*fNWBvyp@lK5q{w?d5~Z_Esm<nWpH1&BpvOs`!XXG1AF8?sQe?pa*d#;k|>&
zjj@fpB#YqjDBIqNPfI*-n4LZM)kz2A)>Y%4O3afbFy!MlSSzNshz1597JrfD?qw9;
zo=X0D)s&MgdT;JKl;;d#n?tpnz+gq-soXB*jl!6`;OqzZjDHw<Cf4AOU?<evpG$7%
zKC83Op<5RBZYO!!S&F<2ywd`4&JkRmWmxN#W$g7S-IfWav?P+)UJTm{Qi*ylk&qvf
zL2C7b=CWUG8xszs!44M3eM(7UPzmpj>*OMK#$HniD6m4^0@wiDFHp=ocp<Zhouao!
zB}Zb(<dJm_^YVAF&z19`T@Rp*(w5l#>qT>R4P4dU8xmKPK(^z)RiBId8&^r5hum;k
z!(J;jhP_@I?E8mC8&|t1jSf!qw1Hke&6AyWU1z;H&=hk#>nNcm?7R;uGU@%d0Xc+^
z^!SUzy4oZ153qnu!ljX)>LB3|O))GI@JZQ~&5OMk-R;t~q#5tFM*=yRGP;RP@X<DY
z8&|W*chN2;Bzdk=jpTQ_$){emR8SP7c&CVGtQCB5Y*FsAShrQ7=<me`Y*U2iTaR3~
zIc13f%C+O60;$%5FpJJy@Gh!&Q8p)R96~S**L2S1zv>HJU-OF!3Wwl}lx&~rX~G?v
zaE-wyR_<}$-xqji!)6vtV}d9a+&8ZFQ`gsq&ox+vwRWk^$`X1Dor6gK#Q45bO0;S#
zv1S)ro$hCp3Abo?l;yF}JZ&oueaX@9?6iR?L(!ud#HVxp=`Qn4><8ZWBs&_6jP03d
zr6c@0D-go^4j(~Bm6xUKA37(!>>e*^(U;~WN+LK&iV?MVD}gUeH-2QrPOf&oew2|g
zW}Eh~I4_maGSi<N%6o=zO5-R_U^wFc#?N@Su{19=m<6Z~E$oU<BwM(Tasuec6hcLd
z@fdl&@#Eh;DaotIPY@yOavO`Cx~I3Az;<8BXMQH=ECKjJucSipM+T*I|2*X+qw<6&
zVf8R?qh#;35RT#@Mu&m-gn@TO%kq2#5vzg_JLED5U~bFupAQx=gGroGZokW`*^2S@
zSqx1~P#=l`^i<YX!hAt|IZ!X&yBL+cma@x+No-)hLuBxvsJN}u<ce)>Slx|Q)E>vJ
zsvp#{RI;bzS&tA|*KJFQ8{*QN5#D&H#D4B6>}V}A6t4N|p;E~jLY|#Yg`6%-%Zu)|
zt`Z3IWqqkDS5s%pNXpmI39*knfL*gz30d=0hEKrO=;^iKOjAT?uwJv$q+5tvE$q{F
zP@TKvX2cfEY021bj$9V&v%yGBa&OCAE&Z`-j%r4HQfC+PAfK=p#o2_V1s-9))T~ta
zWAI0SAxy*Fa<UYeHeyE?BZhm8^o#Nq_m!-kFr%xxZ#Sj*kMDeG$%at4m(BYH)1$Yi
z`eoIU7oi)LCP`6*yF__ismRX~^kjOAHx>2T(9^baB)&I17Q8p+8Wfr4_fH>dsKb0t
zYBUdDZUh0J$k?ClFGB`gtl1(CEPh6K(R@gE0=Ekt<(<zsZxK-VuLgj3R;vI*{%5cx
zrWtl}=Fullo`pBOvQ{?Fd@VBRCpCoh=mic758iy8Hkr)3Xj>C9A%h*bQ=JGD-3zy3
z<CR2bi+{@&Kgu5Pz<D}UbH6S5+U){gNAXpzI8Dxg2VsV<vVtA8Y${(2VWT8(pd@do
zq+j_P^f%uK1-ot;J6_pT5b;N`=IjB<ZunT%#UfIHR_i;dN3S}HalDoBpd(9BXK2?E
zI38R&?&FDn&ldN|5r=1s*Ze6Hh2{Qc?3*v$Z&|4)zBvMO3Bb`j|3En}qc}2lHi2>e
zK_1mlh$L|E`bhdoylMXi94l~K_czVoR?OF4KY)u!<#01mqjov0!l}rd;-g;3@Q<$J
zXSw%7;ju-W17u|>QJXRV%j?-!r%^PT+#md<kDwF2fhn6ZbS3j1=Irro2hV(mY8KNF
z)*5gzXZ;$^J+iyMh0;3f7pR6tk>2qgT~JkDV$Qqc+7DGfX8wSBbqUC{@8CQqi}kn{
z2wb8@0m6G^_$PMX*@okg5liup7O1H2GbuX%<K<iJfat&H7@I55v*S2|?(!`$>;lCQ
z;Qzh6{{}!9SSMY-@XGrWoCexV6x(_v>7gqBsP<N)a4f}}RR8PMM$NQ8eh$a)AiZ!D
zwh4}o4Wn@M&01=sE*BHrXEv;WV^d3YrwfP~o7*r0L4Cma?Q);O4X|O~jK+mfCUx7q
z$@B+#^1c)j5wbP^6XNGvqy?N_2;DFcaK{RVP>FOoG~xX71PI|SC0m!BFOJ^tDIEG;
z>L*>#d`adzDR(=AbL1S@(){#)s%8=X9r%ev5Na9hoLt?%FhzHC>0RE{I0)hR&s_h`
z{*R!qeJEUlVo;I)py00n;XOdAxk2!SsBL4@BW0XfesIBqPzn4&+J7~;jsJ5#A*fgd
zqrY*RtY?Xt?8bB<u6!X$OCg~@C$87;E={?=J8(T3|5bQoEfF0YOE?sm=gEK*n*hEi
z5cdbd(UWk3DWYs^N!QS-TCxT>`P2>IBODsX=9y~whgjccvp<7(g5_xtfa^)PNcnqG
zjQ0QU5%DNSx4}-<zSen+EZl63+U4+6#;XE%S%8ndbZGe>6=;M~<7@v<!0Vh$wNQH4
zi6n|4`|~8G*_-F|e5Z$tTRVYYyr?IJ`agiJD8REOGFVcLCFEgC^VR>);8uH2_}Cfp
z9BoC#+>=PnsBO=mQm7I2-v!iGT<X8@A;mr6uX{3m_2pCj%Vuo7njdk6_o=Pj85q5#
zo=W;qLh7G^LkFWo>;1Nl^KW~8L9mk=C_k5()KkPe%~ZOF{}mX|D!d<GDyg2z^N>x-
zZ*)a9mGJKXyp*lz+yJe-*lX-DdR}}#ikI=Nz4PnB{{m2>ZmEyoFC0kZrFg68yk9U7
z{yzYNev+QsfupnO{?{+OSrRjt4-{)8cHsY_nkusy+Ez<(47Z8tX@qKCV^GP(&pYLu
z+y)OXKAR}wXe~0ZMAJa1|IP2si6#*vqF<!8h!TR_tnTvun?W9>0HNRR3i8FGSfo;n
z;@hRo>+<?N339WQK(6MCJDDo_Y+*;Y7ieFK@AnFN!<i^t{w2VSfA{6!3ttMmCvJ9w
z!q=`E3Gc;=dSjTlUH%ymE^x5mgdzF$i&YYFq>29`NIhV}cgY4jcIEdDj~4wE&FR=Q
zhW+%P<+7d4nr_ad6tKz{(D-xGHl<78;v#B<9A2^~@&6BcmVbL**s?^?3t}vx#SqdY
z_zl?`WeZX{(e<OL63kY8RPwm=)|4XMJWZs=3xXEmXb3Ja)!9Vmvmo+2$ce?)4Fst4
z76cT-*LHE9bPM^Cd2}%|;1K(=!D>`Oaqx+0`SlA%@#jWy6Zr21sxFaGTfI6B>!H8Z
z&asCdEBcQluC+A-$C<{y6LQ&|YMrbD_5@8fT3`JI)lfqGe<gTzfI~sN7DD$o-Tqz4
zzY63;^$TjH9Fj7Y*2juE!vR@}{JRCC`yA-Y51WH{Tcg;yfBx@3&-9w$#oJ(IbQ+it
zd<EtrBYwNRon|J4Xd7JDh=;X|4eGb990qgFG&T_#X?_>pR9hjrwIna$Y*QrX)*zOU
za0n4OoJy9sZ$UOEBo#ue3ulxiSzA!F2vKb!%3P4J2(5sSRl~`eZZt03nGO96q1cAw
z`Vu8Ch$}|A33$~j9G~3y4@0xQ3R!q16M?BT=?51E`w5}`a+_B{nK;@^_J#!g<)p6+
z896E>RE!)#DhIJl)3HAY;{JsT=&h^DeJs<OQ>O`{{&K2UM<0TX3PUQ@uuW?&oeW9(
z%jy2eHB&jB1I9RpMYoH}#=ZKTyTkb<Lvse38`Wuks75Uf?44Dow#@vaC(y|NjOl+b
z!3CZCiuvIqu&{bknTe`@0v++kD53v%Vn(T=T7UA{?r#>ilS&D7)0!?PR7ptVT>X5=
zNV^WUXp7TL&Ot%7G{XpfwB_mM=MOuqCzWqCOhr4KQ0bRG4py;6m8PSvn#V#Zr%D0`
zKPBs`l^ohJZ83y5*&h4qFRDiVtHGBW1!_@oG%7%mAFNU?<2H@$JGjno{;KYWmPk6Z
z|0H?N(F&5KvKkI2rVvzozgy7#uLs&+!+lqR+xaYh?mPolNc=WJ<tTrE-b$c$pY73C
zJ<H#@RMYUXVLMc%`{UfR{0^rw5Ln+3j&!JHG3wMd!LP6(tBmyiTND-Vudm;+2=RJ#
zKWk?*G2DxeV3Err>?;Jnx?8%5T;`9r4k6s`29_#NEO07@_3h&2bxZn^i!SiAhJD|q
zYYpWr2;qCMMlYcMP9+6LlbH55sQYb`CGcwpK3a;Wwn{rK>Y?Xp=ik$e6$MD8@hPVV
z0tbs$F*O!x63?@t(0|YOuK{TXT=;Rg0tayR7|R096hcio(fxw{m?c;mCM8)T^I(DJ
zkTF(NqUb&GJSk(`M}B{*XS_x0p5M5h#^DK+b4t+!_r6}b9ls)IP_q6!BI{JUomS~e
zmd5J-+af+=N85@w;6QuvhBReO9<ZY=TJG`T>EMG+o|HII;SX*7rAtEbb?0e<?GEBE
z^Nqi7#y@{!=6U4PSTdyN>mnfo>~9nHGG0rQb)jxLi@Qi<m*IHG<`ISOs_)NNz+fZH
zM*9t)V&oeO<x!|VqvW~fK_O~M8G*BVygu{<p|&e{KE#Hw-9_6wHag$%<#lw-bY>K8
zYyyA8QCt?QSN}+CWIiimhU@M!;!0D6i;w5j5C=rm<IOd)J1Yhxvo-66xo0I2l<Wnr
z-9vXD1<vTk?+MgB{d~Kc{tI;T6N43_!j{0>tWQO<YEP)LVz=VqzdELeTpqCmc)a@7
zGptdRWNslKk1hCXs3$K!Y@9ALemKNTUGn*x;m}Mx$BZ};6@K=<L#~K@+@i0PgIYM%
zYWyAdUa+cR`X?U}t$G#B07RD^HGY(63q@g|?=GbprgB|8VxRLHm}qq^Hcy?OJ*q82
zADXl-9kE}y%uKWzpqa-0RNxwXqV9(yL^Teg9`uRl2hR0Dw<rgF;+90!FzagychxX2
zQ^=TZtS+m&1*OI8Ra$izq+Dui{w%eVWXYr&)T4-H(YR^nIo@C@Jpla#0FT&rP7(X#
zFZGqOp!oKl<24%8HHY@@(+GX%(qQ0%s79K>6JD9EQJDC6&HD%LXIU|o+-dnQa||<n
z8N2byTxw`w-V00K7SJ9(OxA~<rNr!)e_JVafJE#IeqAYLkKqTYVV?dndQxgNkTrX#
zM75fpIq=X+{nr5Wi<9TLk4BoYGq23UTQ{B0wD>_mIvRdAoju1ZnG0K)B%6ORa}5ev
zXJwbZH<P*qN^R#YYF(0SeyUudJr0QdCB!vY5s;Og#NPsx{F+}md4GJ*X>T<xpN;lQ
zh5=N*o%?0{{`i{ab5ec~D>E%&!v46E=JOW<AXamBSi!~N(?=_b`{Q*LDfWYU`LvN#
zY57UZ)MtRKruVDUVd1w|udO0EJjeTlc>(4}j|hP)G4DwLHBFGXY5B^x2C-JlDB^zN
z0Oy#qko|F>nsTP0_{^{hFMvu)w0aUlr_SGD{t-_NV_D=tIk-bx6`~J~cD@<0Kco4K
zBMrrCMp$@ZgdiYVZH&32%HLs1i&n!p$50Th%ET?rs9{diratOJ+xU;U(~@fva>-Y}
z$1M>>?$-r9msQ716v-YJ25vsMC75OfR7L<|1c+9Fst8oR=wB=m`#PHR9BIjR{#0rh
zY=i^RDmS(<XPVWm7V^SCdo*vvz8u00Ab#@UOxvNKi`G|G^d%|`G&*EZ!@T+_5V3!t
zF~XTf0e_;#57IhhQp1eykwxscaSiG<q+K8GMDOF}y}PjIN8K(0&LhVID*PycxDKM#
z;{2UAq^mQUcbp~Ck|{nNr4|Ne_gCv0K)LR~pJ%<h=y}zzn)B`=!Y%Jb58rYF+MMfG
zRzh^p*1|grfN3fH-V)#P;zKRF*F33d`M#V*Nc*8lg%_z8*+5oFDpNiI(4wrKM8NBP
z8a<7)`~;uzTMn=Hf3w~yy12E<XAk*XisMh}9l)2m=erBkJHF*PWA_IJPzh)?is}(m
z;z=Uk7{D(8dak2^xj(2^>;5NW6X&0d{POoP1HR{bZ)1MTUO4z0r=|vGfWMsb(Tl>1
z9-c>mY^!`rBTHWeK;ryyw-dGEmlbRqmkdh}iew4)X#HLFS;PkQa&CS>lw4@I2E7a0
zrzz7+kr~wMslNLfDAs`Xu0E72NoQRh6T8f{GehyIWE1!n9KX!-AzPe5J=q@hJY}{x
z<CUUgTlyZM^hZz_AwP)fNWW;4b(Qay2Y;M@PrTSq$LOL0M&NJzJ1c8|<|)3c`7PV<
z4|ee{iqSg2mQv|)SEDnqW}-8&5z$5}j|~z;_m_~AebJLxZPAlmXK9#99|uo$=CK@)
zi=Q!fO69DsFfn&Znn5S#RBpOn9j7EzZubx5L2QUgDUUgn#>d;=KJSlHx#=A|2^244
zeLDX=NTewD>3qlO+nARnen2d?Xb$F~$6cfZ+v!5b9GAA|UoXZSmu$YKxcd?0(6Rr_
z40F2C;kJ00@3x3`Z`dVkrc*xUzP-{>6zPG{AYo^Ewy1FRF46}y*L~935t+PVe(DrF
z*A2Q>BVo^VUIcc<FHHU<wx64_LPlw4Hn0-QFl=sc?4(RpC7yFrB`YR{#e=$+V;)}|
zdZI!)HyL%kv}Trn(R{j+)zsfX{yZp{W@+q+YG_tdxpAl?ugH2^mL$#heF{#IWp;Io
zt^lh*UXl6q+izph$C;Fwn>C!(lPDF7_u%IF=?v5%seSdZtznI4`IFT&_~1ly#e$)(
z&6dUEHUn}YW2Z_pW4TY&o-UTFkGlsa1j>gDb+amR&WxMo#SNPVVUTq2qF`CKp>AYF
z4zaP^YOSZYuAqB5cvi5i$x!!Kg+*@ChS@6e)1ZKFdd1i1*GA3!PfI2Zt3_@So37G+
z_${D9C-SwCoOGq9i^(c2(a3KKvk`!SOcP`*s}-Ju(QAM3a{!meu05t15wK4O_Y0Q2
zGSscAu()KA6~9X+gx}{SXtBDX8$D>eiew*w+5A?*IMesH0DuAr<9V`Lp5tQF+|nat
zmJa?XSoRjsV{gN7HBI8T#0s52WQD;GML;Y-)&|HmK%RDhBTqCCseQrD*h9k3^Lu&g
zzeNy-LYu`X&NxU8fVFNwoV}at9~PqmQT4EiT7t=Lc@qj|Yi`%DZ?*Uh-FK$Q6|DJy
zb&<7%f1l>&U<rdm@RGX4Wy$##ZvAbZr5gyG|6d$Of%9lS^VqYmp$*sUV0FKUe||CF
c>xGTy3tvxQxxSo?oQ$LtF%OTX-d*DV1>G^@!~g&Q

literal 0
HcmV?d00001

diff --git a/docs/2_39/functions-reference/array_operations.html b/docs/2_39/functions-reference/array_operations.html
new file mode 100644
index 000000000..dc0b99ca2
--- /dev/null
+++ b/docs/2_39/functions-reference/array_operations.html
@@ -0,0 +1,1507 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Array Operations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/matrix_operations.html" rel="next">
+<link href="../functions-reference/complex-valued_basic_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Array Operations">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/array_operations.html">Array Operations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#array-operations" id="toc-array-operations" class="nav-link active" data-scroll-target="#array-operations">Array Operations</a>
+  <ul class="collapse">
+  <li><a href="#array-reductions" id="toc-array-reductions" class="nav-link" data-scroll-target="#array-reductions">Reductions</a>
+  <ul class="collapse">
+  <li><a href="#minimum-and-maximum" id="toc-minimum-and-maximum" class="nav-link" data-scroll-target="#minimum-and-maximum">Minimum and maximum</a></li>
+  <li><a href="#sum-product-and-log-sum-of-exp" id="toc-sum-product-and-log-sum-of-exp" class="nav-link" data-scroll-target="#sum-product-and-log-sum-of-exp">Sum, product, and log sum of exp</a></li>
+  <li><a href="#sample-mean-variance-and-standard-deviation" id="toc-sample-mean-variance-and-standard-deviation" class="nav-link" data-scroll-target="#sample-mean-variance-and-standard-deviation">Sample mean, variance, and standard deviation</a></li>
+  <li><a href="#norms" id="toc-norms" class="nav-link" data-scroll-target="#norms">Norms</a></li>
+  <li><a href="#euclidean-distance-and-squared-distance" id="toc-euclidean-distance-and-squared-distance" class="nav-link" data-scroll-target="#euclidean-distance-and-squared-distance">Euclidean distance and squared distance</a></li>
+  <li><a href="#quantile" id="toc-quantile" class="nav-link" data-scroll-target="#quantile">Quantile</a></li>
+  </ul></li>
+  <li><a href="#array-size-and-dimension-function" id="toc-array-size-and-dimension-function" class="nav-link" data-scroll-target="#array-size-and-dimension-function">Array size and dimension function</a></li>
+  <li><a href="#array-broadcasting" id="toc-array-broadcasting" class="nav-link" data-scroll-target="#array-broadcasting">Array broadcasting</a></li>
+  <li><a href="#array-concatenation" id="toc-array-concatenation" class="nav-link" data-scroll-target="#array-concatenation">Array concatenation</a></li>
+  <li><a href="#sorting-functions" id="toc-sorting-functions" class="nav-link" data-scroll-target="#sorting-functions">Sorting functions</a></li>
+  <li><a href="#reversing-functions" id="toc-reversing-functions" class="nav-link" data-scroll-target="#reversing-functions">Reversing functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/array_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/array_operations.html">Array Operations</a></li></ol></nav></header>
+
+
+
+
+<section id="array-operations" class="level1">
+<h1>Array Operations</h1>
+<section id="array-reductions" class="level2">
+<h2 class="anchored" data-anchor-id="array-reductions">Reductions</h2>
+<p>The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).</p>
+<section id="minimum-and-maximum" class="level3">
+<h3 class="anchored" data-anchor-id="minimum-and-maximum">Minimum and maximum</h3>
+<a id="index-entry-c4f3e810f70c09849013a04b92b38f9e93067e2f" class="index-entry"></a> <!-- real; min; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#min" class="unlink" title="Jump to index entry"><code>min</code></a></strong><code>(array[] real x)</code><br> The minimum value in x, or <span class="math inline">\(+\infty\)</span> if x is size 0.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6f45b7ae2a16ec959cb22d041d7b50c43dada819" class="index-entry"></a> <!-- int; min; (array[] int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#min" class="unlink" title="Jump to index entry"><code>min</code></a></strong><code>(array[] int x)</code><br> The minimum value in x, or error if x is size 0.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-1dc0a9e5bd6f78b5d2904cba8c6375fa264e6990" class="index-entry"></a> <!-- real; max; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#max" class="unlink" title="Jump to index entry"><code>max</code></a></strong><code>(array[] real x)</code><br> The maximum value in x, or <span class="math inline">\(-\infty\)</span> if x is size 0.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b77840f6a6da285fe070a2db533be9470af54dfc" class="index-entry"></a> <!-- int; max; (array[] int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#max" class="unlink" title="Jump to index entry"><code>max</code></a></strong><code>(array[] int x)</code><br> The maximum value in x, or error if x is size 0.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="sum-product-and-log-sum-of-exp" class="level3">
+<h3 class="anchored" data-anchor-id="sum-product-and-log-sum-of-exp">Sum, product, and log sum of exp</h3>
+<a id="index-entry-d97ce537eae8d48ac3ddc98520d0c5eaafd6fe27" class="index-entry"></a> <!-- int; sum; (array[] int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(array[] int x)</code><br> The sum of the elements in x, or 0 if the array is empty.</p>
+<small><i>Available since 2.1</i></small>
+<a id="index-entry-acfd44849111bfa6024f821401db39f1181f4d14" class="index-entry"></a> <!-- real; sum; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(array[] real x)</code><br> The sum of the elements in x; see definition above.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-e7ecc93be79d76e228adf2a4a2f9cdd908a7238b" class="index-entry"></a> <!-- complex; sum; (array[] complex x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(array[] complex x)</code><br> The sum of the elements in x; see definition above.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-936503ecb56cc2f9aa31395e3b80d5dd901f14ab" class="index-entry"></a> <!-- real; prod; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(array[] real x)</code><br> The product of the elements in x, or 1 if x is size 0.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-69d45ac78302f42fc87a23263c5230e4ee093a6e" class="index-entry"></a> <!-- real; prod; (array[] int x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(array[] int x)</code><br> The product of the elements in x, <span class="math display">\[\begin{equation*}
+\text{product}(x) = \begin{cases}
+\prod_{n=1}^N x_n &amp; \text{if} N &gt; 0 \\[4pt] 1 &amp; \text{if} N = 0
+\end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7b26d9d0a4894d9b4686f4847de2d728f050eeca" class="index-entry"></a> <!-- real; log_sum_exp; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_sum_exp" class="unlink" title="Jump to index entry"><code>log_sum_exp</code></a></strong><code>(array[] real x)</code><br> The natural logarithm of the sum of the exponentials of the elements in x, or <span class="math inline">\(-\infty\)</span> if the array is empty.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="sample-mean-variance-and-standard-deviation" class="level3">
+<h3 class="anchored" data-anchor-id="sample-mean-variance-and-standard-deviation">Sample mean, variance, and standard deviation</h3>
+<p>The sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> The sample deviation is defined as the square root of the sample deviation, but is not unbiased.</p>
+<a id="index-entry-bb1b7fccc6b17cde8f9e75a947464b97bc4082d3" class="index-entry"></a> <!-- real; mean; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#mean" class="unlink" title="Jump to index entry"><code>mean</code></a></strong><code>(array[] real x)</code><br> The sample mean of the elements in x. For an array <span class="math inline">\(x\)</span> of size <span class="math inline">\(N &gt;
+0\)</span>, <span class="math display">\[\begin{equation*}
+\text{mean}(x) \ = \ \bar{x} \ = \ \frac{1}{N} \sum_{n=1}^N
+x_n.
+\end{equation*}\]</span> It is an error to the call the mean function with an array of size <span class="math inline">\(0\)</span>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-92ca3dfea614e505cf3d32942458a9c948f45e92" class="index-entry"></a> <!-- real; variance; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#variance" class="unlink" title="Jump to index entry"><code>variance</code></a></strong><code>(array[] real x)</code><br> The sample variance of the elements in x. For <span class="math inline">\(N &gt; 0\)</span>, <span class="math display">\[\begin{equation*}
+\text{variance}(x) \ = \ \begin{cases} \frac{1}{N-1} \sum_{n=1}^N (x_n
+- \bar{x})^2 &amp; \text{if } N &gt; 1 \\[4pt] 0 &amp; \text{if } N = 1
+\end{cases}
+\end{equation*}\]</span> It is an error to call the <code>variance</code> function with an array of size 0.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2ff81571d634ef5353d30cfebadc9561bf519ee4" class="index-entry"></a> <!-- real; sd; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sd" class="unlink" title="Jump to index entry"><code>sd</code></a></strong><code>(array[] real x)</code><br> The sample standard deviation of elements in x. <span class="math display">\[\begin{equation*}
+\text{sd}(x) =
+\begin{cases} \sqrt{\, \text{variance}(x)} &amp; \text{if } N &gt; 1 \\[4pt]
+0 &amp; \text{if } N = 0 \end{cases}
+\end{equation*}\]</span> It is an error to call the <code>sd</code> function with an array of size 0.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="norms" class="level3">
+<h3 class="anchored" data-anchor-id="norms">Norms</h3>
+<a id="index-entry-64a22469a1ca08b874dfd1f7dba4e60a02311a4d" class="index-entry"></a> <!-- real; norm1; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm1" class="unlink" title="Jump to index entry"><code>norm1</code></a></strong><code>(vector x)</code><br> The L1 norm of x, defined by <span class="math display">\[\begin{equation*}
+\text{norm1}(x) \ = \ \textstyle \sum_{n=1}^N (|x_n|)
+\end{equation*}\]</span> where <code>N</code> is the size of x.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-3a384b9bbb192b688afb1c9e95a490200da65412" class="index-entry"></a> <!-- real; norm1; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm1" class="unlink" title="Jump to index entry"><code>norm1</code></a></strong><code>(row_vector x)</code><br> The L1 norm of x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d29d5686fa464cc82765e365c73f1ebb70954ffa" class="index-entry"></a> <!-- real; norm1; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm1" class="unlink" title="Jump to index entry"><code>norm1</code></a></strong><code>(array[] real x)</code><br> The L1 norm of x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-bd1aec552c96bca3230c994fb47fa42e54632e5e" class="index-entry"></a> <!-- real; norm2; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm2" class="unlink" title="Jump to index entry"><code>norm2</code></a></strong><code>(vector x)</code><br> The L2 norm of x, defined by <span class="math display">\[\begin{equation*}
+\text{norm2}(x) \ = \ \sqrt{\textstyle \sum_{n=1}^N (x_n)^2}
+\end{equation*}\]</span> where <code>N</code> is the size of x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-15be5905b3ac40ce77299341b9e7434d6e5a721e" class="index-entry"></a> <!-- real; norm2; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm2" class="unlink" title="Jump to index entry"><code>norm2</code></a></strong><code>(row_vector x)</code><br> The L2 norm of x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-95748dc7d816217c82a900a5cefd801980794a10" class="index-entry"></a> <!-- real; norm2; (array[] real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm2" class="unlink" title="Jump to index entry"><code>norm2</code></a></strong><code>(array[] real x)</code><br> The L2 norm of x</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="euclidean-distance-and-squared-distance" class="level3">
+<h3 class="anchored" data-anchor-id="euclidean-distance-and-squared-distance">Euclidean distance and squared distance</h3>
+<a id="index-entry-702855e7be40e4fcfbe31b5f08538c5fbfba210e" class="index-entry"></a> <!-- real; distance; (vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#distance" class="unlink" title="Jump to index entry"><code>distance</code></a></strong><code>(vector x, vector y)</code><br> The Euclidean distance between x and y, defined by <span class="math display">\[\begin{equation*}
+\text{distance}(x,y) \ = \ \sqrt{\textstyle \sum_{n=1}^N (x_n - y_n)^2}
+\end{equation*}\]</span> where <code>N</code> is the size of x and y. It is an error to call <code>distance</code> with arguments of unequal size.</p>
+<small><i>Available since 2.2</i></small>
+<a id="index-entry-d68901fc0a3ff9e987854900a185ee2713e5c520" class="index-entry"></a> <!-- real; distance; (vector x, row_vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#distance" class="unlink" title="Jump to index entry"><code>distance</code></a></strong><code>(vector x, row_vector y)</code><br> The Euclidean distance between x and y</p>
+<small><i>Available since 2.2</i></small>
+<a id="index-entry-a9f701404655eb686e31e66d38918d7e08cd9f3b" class="index-entry"></a> <!-- real; distance; (row_vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#distance" class="unlink" title="Jump to index entry"><code>distance</code></a></strong><code>(row_vector x, vector y)</code><br> The Euclidean distance between x and y</p>
+<small><i>Available since 2.2</i></small>
+<a id="index-entry-c6112c85adc9432ce6b005aff57babec0dbee55e" class="index-entry"></a> <!-- real; distance; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#distance" class="unlink" title="Jump to index entry"><code>distance</code></a></strong><code>(row_vector x, row_vector y)</code><br> The Euclidean distance between x and y</p>
+<small><i>Available since 2.2</i></small>
+<a id="index-entry-47ca27c5ccd71b8bebcadb9723ea604c50120857" class="index-entry"></a> <!-- real; squared_distance; (vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#squared_distance" class="unlink" title="Jump to index entry"><code>squared_distance</code></a></strong><code>(vector x, vector y)</code><br> The squared Euclidean distance between x and y, defined by <span class="math display">\[\begin{equation*}
+\mathrm{squared\_distance}(x,y) \ = \ \text{distance}(x,y)^2 \ = \ \textstyle \sum_{n=1}^N (x_n - y_n)^2,
+\end{equation*}\]</span> where <code>N</code> is the size of x and y. It is an error to call <code>squared_distance</code> with arguments of unequal size.</p>
+<small><i>Available since 2.7</i></small>
+<a id="index-entry-e2cb205544293c316ec3e5356c4d1dbebfafb159" class="index-entry"></a> <!-- real; squared_distance; (vector x, row_vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#squared_distance" class="unlink" title="Jump to index entry"><code>squared_distance</code></a></strong><code>(vector x, row_vector y)</code><br> The squared Euclidean distance between x and y</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-8887aa905b17d368550a915063d1bb8fbf4ccf59" class="index-entry"></a> <!-- real; squared_distance; (row_vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#squared_distance" class="unlink" title="Jump to index entry"><code>squared_distance</code></a></strong><code>(row_vector x, vector y)</code><br> The squared Euclidean distance between x and y</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-794d03938547aad86944e55cbb1f64f9ea22dee5" class="index-entry"></a> <!-- real; squared_distance; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#squared_distance" class="unlink" title="Jump to index entry"><code>squared_distance</code></a></strong><code>(row_vector x, row_vector y)</code><br> The Euclidean distance between x and y</p>
+<small><i>Available since 2.26</i></small>
+</section>
+<section id="quantile" class="level3">
+<h3 class="anchored" data-anchor-id="quantile">Quantile</h3>
+<p>Produces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.</p>
+<p>Implements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).</p>
+<a id="index-entry-8c7df96194aa20f4691df231ed7577942daf6d6f" class="index-entry"></a> <!-- real; quantile; (data array[] real x, data real p); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#quantile" class="unlink" title="Jump to index entry"><code>quantile</code></a></strong><code>(data array[] real x, data real p)</code><br> The p-th quantile of x</p>
+<small><i>Available since 2.27</i></small>
+<a id="index-entry-b0471485d6b0fd5b43b349cbe4e386d6c38f5064" class="index-entry"></a> <!-- array[] real; quantile; (data array[] real x, data array[] real p); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#quantile" class="unlink" title="Jump to index entry"><code>quantile</code></a></strong><code>(data array[] real x, data array[] real p)</code><br> An array containing the quantiles of x given by the array of probabilities p</p>
+<small><i>Available since 2.27</i></small>
+</section>
+</section>
+<section id="array-size-and-dimension-function" class="level2">
+<h2 class="anchored" data-anchor-id="array-size-and-dimension-function">Array size and dimension function</h2>
+<p>The size of an array or matrix can be obtained using the <code>dims()</code> function. The <code>dims()</code> function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">7</span>, <span class="dv">8</span>, <span class="dv">9</span>] <span class="dt">real</span> x;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">7</span>] <span class="dt">matrix</span>[<span class="dv">8</span>, <span class="dv">9</span>] y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>then calling <code>dims(x)</code> or <code>dims(y)</code> returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.</p>
+<p>The <code>size()</code> function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[M, N] <span class="dt">real</span> a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>the size of <code>a</code> is <code>M</code>.</p>
+<p>The function <code>num_elements</code>, on the other hand, measures all of the elements, so that the array <code>a</code> above has <span class="math inline">\(M \times N\)</span> elements.</p>
+<p>The specialized functions <code>rows()</code> and <code>cols()</code> should be used to extract the dimensions of vectors and matrices.</p>
+<a id="index-entry-15460652371e3354bdc590e3632fc540ab76b525" class="index-entry"></a> <!-- array[] int; dims; (T x); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#dims" class="unlink" title="Jump to index entry"><code>dims</code></a></strong><code>(T x)</code><br> Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a7f3950f11e27f31d9607b74a4cc9d5201f6f5ba" class="index-entry"></a> <!-- int; num_elements; (array[] T x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(array[] T x)</code><br> Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if <code>x</code> is of type <code>array[4, 3] real</code> then <code>num_elements(x)</code> is 12, and if <code>y</code> is declared as <code>array[5] matrix[3, 4] y</code>, then <code>size(y)</code> evaluates to 60.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-c12761833dc86f2313582b4fe440b0b82d4c8cb6" class="index-entry"></a> <!-- int; size; (array[] T x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(array[] T x)</code><br> Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if <code>x</code> is of type <code>array[4, 3] real</code> then <code>size(x)</code> is 4.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="array-broadcasting" class="level2">
+<h2 class="anchored" data-anchor-id="array-broadcasting">Array broadcasting</h2>
+<p>The following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.</p>
+<a id="index-entry-b16f7217e0f04da666e3544e1f8e27e04e429973" class="index-entry"></a> <!-- array[] T; rep_array; (T x, int n); -->
+<p></p>
+<p><code>array[] T</code> <strong><a href="../functions-reference/functions_index.html#rep_array" class="unlink" title="Jump to index entry"><code>rep_array</code></a></strong><code>(T x, int n)</code><br> Return the n array with every entry assigned to x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a5563d5b3ee4bd63fb6cbbef35e86bd940d587df" class="index-entry"></a> <!-- array[,] T; rep_array; (T x, int m, int n); -->
+<p></p>
+<p><code>array [,] T</code> <strong><a href="../functions-reference/functions_index.html#rep_array" class="unlink" title="Jump to index entry"><code>rep_array</code></a></strong><code>(T x, int m, int n)</code><br> Return the m by n array with every entry assigned to x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a2816f987a767657e3323203e069be5323f40964" class="index-entry"></a> <!-- array[,,] T; rep_array; (T x, int k, int m, int n); -->
+<p></p>
+<p><code>array[,,] T</code> <strong><a href="../functions-reference/functions_index.html#rep_array" class="unlink" title="Jump to index entry"><code>rep_array</code></a></strong><code>(T x, int k, int m, int n)</code><br> Return the k by m by n array with every entry assigned to x.</p>
+<small><i>Available since 2.0</i></small>
+<p>For example, <code>rep_array(1.0,5)</code> produces a real array (type <code>array[] real</code>) of size 5 with all values set to 1.0. On the other hand, <code>rep_array(1,5)</code> produces an integer array (type <code>array[] int</code>) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">real</span> y;</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">int</span> x;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a> x = rep_array(<span class="dv">1</span>, <span class="dv">5</span>);     <span class="co">// ok</span></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a> y = rep_array(<span class="fl">1.0</span>, <span class="dv">5</span>);   <span class="co">// ok</span></span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a> x = rep_array(<span class="fl">1.0</span>, <span class="dv">5</span>);   <span class="co">// illegal</span></span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a> y = rep_array(<span class="dv">1</span>, <span class="dv">5</span>);     <span class="co">// illegal</span></span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a> x = y;                  <span class="co">// illegal</span></span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a> y = x;                  <span class="co">// illegal</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the value being repeated <code>v</code> is a vector (i.e., <code>T</code> is <code>vector</code>), then <code>rep_array(v, 27)</code> is a size 27 array consisting of 27 copies of the vector <code>v</code>.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] v;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">vector</span>[<span class="dv">5</span>] a;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a> a = rep_array(v, <span class="dv">3</span>);  <span class="co">// fill a with copies of v</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a> a[<span class="dv">2</span>, <span class="dv">4</span>] = <span class="fl">9.0</span>;        <span class="co">// v[4], a[1, 4], a[3, 4] unchanged</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the <code>rep_array</code> functions is called. For instance, consider the following legal code snippet.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">6</span>] <span class="dt">real</span> a;</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>, <span class="dv">5</span>, <span class="dv">6</span>] <span class="dt">real</span> b;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a> b = rep_array(a, <span class="dv">3</span>, <span class="dv">4</span>); <span class="co">//  make (3 x 4) copies of a</span></span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a> b[<span class="dv">1</span>, <span class="dv">1</span>, <span class="dv">1</span>, <span class="dv">1</span>] = <span class="fl">27.9</span>;    <span class="co">//  a[1, 1] unchanged</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After the assignment to <code>b</code>, the value for <code>b[j, k, m, n]</code> is equal to <code>a[m, n]</code> where it is defined, for <code>j</code> in <code>1:3</code>, <code>k</code> in <code>1:4</code>, <code>m</code> in <code>1:5</code>, and <code>n</code> in <code>1:6</code>.</p>
+</section>
+<section id="array-concatenation" class="level2">
+<h2 class="anchored" data-anchor-id="array-concatenation">Array concatenation</h2>
+<a id="index-entry-a92d8a1e904382ddb4b800a330feedcb3c04f2c2" class="index-entry"></a> <!-- T; append_array; (T x, T y); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#append_array" class="unlink" title="Jump to index entry"><code>append_array</code></a></strong><code>(T x, T y)</code><br> Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.</p>
+<small><i>Available since 2.18</i></small>
+<p>For example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">1</span>, <span class="dv">7</span>] <span class="dt">matrix</span>[<span class="dv">4</span>, <span class="dv">6</span>] x1;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">1</span>, <span class="dv">7</span>] <span class="dt">matrix</span>[<span class="dv">4</span>, <span class="dv">6</span>] x2;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">1</span>, <span class="dv">7</span>] <span class="dt">matrix</span>[<span class="dv">4</span>, <span class="dv">6</span>] x3;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a> x3 = append_array(x1, x2);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="sorting-functions" class="level2">
+<h2 class="anchored" data-anchor-id="sorting-functions">Sorting functions</h2>
+<p>Sorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if <code>v</code> is declared as a real array of size 3, with values <span class="math display">\[\begin{equation*}
+\text{v} = (1, -10.3,
+20.987),
+\end{equation*}\]</span> then the various sort routines produce <span class="math display">\[\begin{eqnarray*}
+\mathrm{sort\_asc(v)} &amp; = &amp;  (-10.3,1,20.987) \\[4pt]
+\mathrm{sort\_desc(v)} &amp; = &amp;  (20.987,1,-10.3) \\[4pt]
+\mathrm{sort\_indices\_asc(v)} &amp; = &amp;  (2,1,3) \\[4pt]
+\mathrm{sort\_indices\_desc(v)} &amp; = &amp;  (3,1,2)
+\end{eqnarray*}\]</span></p>
+<a id="index-entry-29411da2ef662a3af3793ab20c89f7e7b677bd0e" class="index-entry"></a> <!-- array[] real; sort_asc; (array[] real v); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#sort_asc" class="unlink" title="Jump to index entry"><code>sort_asc</code></a></strong><code>(array[] real v)</code><br> Sort the elements of v in ascending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6e75416be841609073a75454c55e2fbc19f46f47" class="index-entry"></a> <!-- array[] int; sort_asc; (array[] int v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_asc" class="unlink" title="Jump to index entry"><code>sort_asc</code></a></strong><code>(array[] int v)</code><br> Sort the elements of v in ascending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-fce023d8773b92438ce6a759565ce44deab3681c" class="index-entry"></a> <!-- array[] real; sort_desc; (array[] real v); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#sort_desc" class="unlink" title="Jump to index entry"><code>sort_desc</code></a></strong><code>(array[] real v)</code><br> Sort the elements of v in descending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-355460b3ae8892bf699932f697801d5f961e04e8" class="index-entry"></a> <!-- array[] int; sort_desc; (array[] int v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_desc" class="unlink" title="Jump to index entry"><code>sort_desc</code></a></strong><code>(array[] int v)</code><br> Sort the elements of v in descending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-88b08c67456c5ae147bac2ccabf5a2435eb84def" class="index-entry"></a> <!-- array[] int; sort_indices_asc; (array[] real v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_asc" class="unlink" title="Jump to index entry"><code>sort_indices_asc</code></a></strong><code>(array[] real v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in ascending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-22998db91982cee6e0c3b1a618078f76b085d02c" class="index-entry"></a> <!-- array[] int; sort_indices_asc; (array[] int v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_asc" class="unlink" title="Jump to index entry"><code>sort_indices_asc</code></a></strong><code>(array[] int v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in ascending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-fa29a352fe066806f7fafbebc5a2510bab51596f" class="index-entry"></a> <!-- array[] int; sort_indices_desc; (array[] real v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_desc" class="unlink" title="Jump to index entry"><code>sort_indices_desc</code></a></strong><code>(array[] real v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in descending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-f8faa56dc58a61ee15ff8a3545ecc6aee8b3110d" class="index-entry"></a> <!-- array[] int; sort_indices_desc; (array[] int v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_desc" class="unlink" title="Jump to index entry"><code>sort_indices_desc</code></a></strong><code>(array[] int v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in descending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-a28838301370f881f9ec571b8676f50b783b629c" class="index-entry"></a> <!-- int; rank; (array[] real v, int s); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rank" class="unlink" title="Jump to index entry"><code>rank</code></a></strong><code>(array[] real v, int s)</code><br> Number of components of v less than v[s]</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a6d0770a4ab5bc51a44d3626199ee7de7658377e" class="index-entry"></a> <!-- int; rank; (array[] int v, int s); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rank" class="unlink" title="Jump to index entry"><code>rank</code></a></strong><code>(array[] int v, int s)</code><br> Number of components of v less than v[s]</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="reversing-functions" class="level2">
+<h2 class="anchored" data-anchor-id="reversing-functions">Reversing functions</h2>
+<p>Stan provides functions to create a new array by reversing the order of elements in an existing array. For example, if <code>v</code> is declared as a real array of size 3, with values <span class="math display">\[\begin{equation*}
+\text{v} = (1,\, -10.3,\, 20.987),
+\end{equation*}\]</span> then <span class="math display">\[\begin{equation*}
+\mathrm{reverse(v)} = (20.987,\, -10.3,\, 1).
+\end{equation*}\]</span></p>
+<a id="index-entry-9759b3b50ddba20806c98d04d90c234afd5b230e" class="index-entry"></a> <!-- array[] T; reverse; (array[] T v); -->
+<p></p>
+<p><code>array[] T</code> <strong><a href="../functions-reference/functions_index.html#reverse" class="unlink" title="Jump to index entry"><code>reverse</code></a></strong><code>(array[] T v)</code><br> Return a new array containing the elements of the argument in reverse order.</p>
+<small><i>Available since 2.23</i></small>
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Dividing by <span class="math inline">\(N\)</span> rather than <span class="math inline">\((N-1)\)</span> produces a maximum likelihood estimate of variance, which is biased to underestimate variance.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/complex-valued_basic_functions.html" class="pagination-link" aria-label="Complex-Valued Basic Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Complex-Valued Basic Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/matrix_operations.html" class="pagination-link" aria-label="Matrix Operations">
+        <span class="nav-page-text">Matrix Operations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/array_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/binary_distributions.html b/docs/2_39/functions-reference/binary_distributions.html
new file mode 100644
index 000000000..18a7b19d6
--- /dev/null
+++ b/docs/2_39/functions-reference/binary_distributions.html
@@ -0,0 +1,1331 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Binary Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/bounded_discrete_distributions.html" rel="next">
+<link href="../functions-reference/conventions_for_probability_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Binary Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Binary Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#binary-distributions" id="toc-binary-distributions" class="nav-link active" data-scroll-target="#binary-distributions">Binary Distributions</a>
+  <ul class="collapse">
+  <li><a href="#bernoulli-distribution" id="toc-bernoulli-distribution" class="nav-link" data-scroll-target="#bernoulli-distribution">Bernoulli distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function" id="toc-probability-mass-function" class="nav-link" data-scroll-target="#probability-mass-function">Probability mass function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan Functions</a></li>
+  </ul></li>
+  <li><a href="#bernoulli-logit-distribution" id="toc-bernoulli-logit-distribution" class="nav-link" data-scroll-target="#bernoulli-logit-distribution">Bernoulli distribution, logit parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-1" id="toc-probability-mass-function-1" class="nav-link" data-scroll-target="#probability-mass-function-1">Probability mass function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan Functions</a></li>
+  </ul></li>
+  <li><a href="#bernoulli-logit-glm" id="toc-bernoulli-logit-glm" class="nav-link" data-scroll-target="#bernoulli-logit-glm">Bernoulli-logit generalized linear model (Logistic Regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-2" id="toc-probability-mass-function-2" class="nav-link" data-scroll-target="#probability-mass-function-2">Probability mass function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan Functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/binary_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Binary Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="binary-distributions" class="level1">
+<h1>Binary Distributions</h1>
+<p>Binary probability distributions have support on <span class="math inline">\(\{0,1\}\)</span>, where 1 represents the value true and 0 the value false.</p>
+<section id="bernoulli-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="bernoulli-distribution">Bernoulli distribution</h2>
+<section id="probability-mass-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function">Probability mass function</h3>
+<p>If <span class="math inline">\(\theta \in [0,1]\)</span>, then for <span class="math inline">\(y \in \{0,1\}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Bernoulli}(y~|~\theta) = \left\{ \begin{array}{ll} \theta &amp;
+\text{if } y = 1, \text{ and} \\ 1 - \theta &amp; \text{if } y = 0.
+\end{array} \right.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#bernoulli" class="unlink" title="Jump to index entry"><code>bernoulli</code></a></strong><code>(theta)</code></p>
+<p>Increment target log probability density with <code>bernoulli_lupmf(y | theta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7bb1d0fe6798b5a1780314a5829d10cf070cb584" class="index-entry"></a> <!-- real; bernoulli ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan Functions</h3>
+<a id="index-entry-d5704b5b3fb92b6acbe402518995306b417633a6" class="index-entry"></a> <!-- real; bernoulli_lpmf; (ints y | reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_lpmf</code></a></strong><code>(ints y | reals theta)</code><br> The log Bernoulli probability mass of y given chance of success <code>theta</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-d71dcd384cd23a024264d9e21555dafc50336c88" class="index-entry"></a> <!-- real; bernoulli_lupmf; (ints y | reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_lupmf</code></a></strong><code>(ints y | reals theta)</code><br> The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-fba3b9784cb145bce67dc4b2b5b7e6beedcde682" class="index-entry"></a> <!-- real; bernoulli_cdf; (ints y | reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_cdf" class="unlink" title="Jump to index entry"><code>bernoulli_cdf</code></a></strong><code>(ints y | reals theta)</code><br> The Bernoulli cumulative distribution function of y given chance of success <code>theta</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-aee9b81290c031013a18766f7c73a0d661732bc1" class="index-entry"></a> <!-- real; bernoulli_lcdf; (ints y | reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_lcdf" class="unlink" title="Jump to index entry"><code>bernoulli_lcdf</code></a></strong><code>(ints y | reals theta)</code><br> The log of the Bernoulli cumulative distribution function of y given chance of success <code>theta</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-8669190b9a16302c87a7a759518c13b507ba01d5" class="index-entry"></a> <!-- real; bernoulli_lccdf; (ints y | reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_lccdf" class="unlink" title="Jump to index entry"><code>bernoulli_lccdf</code></a></strong><code>(ints y | reals theta)</code><br> The log of the Bernoulli complementary cumulative distribution function of y given chance of success <code>theta</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-8e32579888cc9b06d6c81f2c372fc7b94641ce47" class="index-entry"></a> <!-- R; bernoulli_rng; (reals theta); -->
+<p></p>
+<p><code>ints</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_rng" class="unlink" title="Jump to index entry"><code>bernoulli_rng</code></a></strong><code>(reals theta)</code><br> Generate a Bernoulli variate with chance of success <code>theta</code> or an array of Bernoulli variates given an array of <code>thetas</code> of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="bernoulli-logit-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="bernoulli-logit-distribution">Bernoulli distribution, logit parameterization</h2>
+<p>Stan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.</p>
+<section id="probability-mass-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-1">Probability mass function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}\)</span>, then for <span class="math inline">\(y \in \{0,1\}\)</span>, <span class="math display">\[\begin{equation*}
+\text{BernoulliLogit}(y~|~\alpha) = \text{Bernoulli}(y |
+\text{logit}^{-1}(\alpha)) = \left\{ \begin{array}{ll}
+\text{logit}^{-1}(\alpha) &amp; \text{if } y = 1, \text{ and} \\ 1 -
+\text{logit}^{-1}(\alpha) &amp;  \text{if } y = 0. \end{array} \right.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit" class="unlink" title="Jump to index entry"><code>bernoulli_logit</code></a></strong><code>(alpha)</code></p>
+<p>Increment target log probability density with <code>bernoulli_logit_lupmf(y | alpha)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-778ebacd22f9f00e5c91bc26ac55e9ef4530423a" class="index-entry"></a> <!-- real; bernoulli_logit ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan Functions</h3>
+<a id="index-entry-cad8c604c21adb32fa4d2b5789b7a3888ebaf554" class="index-entry"></a> <!-- real; bernoulli_logit_lpmf; (ints y | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_lpmf</code></a></strong><code>(ints y | reals alpha)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha)</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-74fd84a52ce20d7b2b2914acecea456438e780a5" class="index-entry"></a> <!-- real; bernoulli_logit_lupmf; (ints y | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_lupmf</code></a></strong><code>(ints y | reals alpha)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha)</code> dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-275ab72ec22b30d37f9bc02991be25a6bd23f49a" class="index-entry"></a> <!-- R; bernoulli_logit_rng; (reals alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_rng" class="unlink" title="Jump to index entry"><code>bernoulli_logit_rng</code></a></strong><code>(reals alpha)</code><br> Generate a Bernoulli variate with chance of success <span class="math inline">\(\text{logit}^{-1}(\alpha)\)</span>; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="bernoulli-logit-glm" class="level2">
+<h2 class="anchored" data-anchor-id="bernoulli-logit-glm">Bernoulli-logit generalized linear model (Logistic Regression)</h2>
+<p>Stan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e.&nbsp;a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.</p>
+<section id="probability-mass-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-2">Probability mass function</h3>
+<p>If <span class="math inline">\(x\in \mathbb{R}^{n\cdot m}, \alpha \in \mathbb{R}^n, \beta\in
+\mathbb{R}^m\)</span>, then for <span class="math inline">\(y \in {\{0,1\}}^n\)</span>, <span class="math display">\[\begin{align*}
+&amp;\text{BernoulliLogitGLM}(y~|~x, \alpha, \beta) = \prod_{1\leq i \leq
+n}\text{Bernoulli}(y_i~|~\text{logit}^{-1}(\alpha_i + x_i\cdot
+\beta))\\ &amp;= \prod_{1\leq i \leq n} \left\{ \begin{array}{ll}
+\text{logit}^{-1}(\alpha_i + \sum_{1\leq j\leq m}x_{ij}\cdot \beta_j)
+&amp; \text{if } y_i = 1, \text{ and} \\ 1 - \text{logit}^{-1}(\alpha_i +
+\sum_{1\leq j\leq m}x_{ij}\cdot \beta_j) &amp; \text{if } y_i = 0.
+\end{array} \right. \end{align*}\]</span></p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm</code></a></strong><code>(x, alpha, beta)</code></p>
+<p>Increment target log probability density with <code>bernoulli_logit_glm_lupmf(y | x, alpha, beta)</code>.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-cc775222ff8767d0ddb5f23a1f67b77eac3ef7f1" class="index-entry"></a> <!-- real; bernoulli_logit_glm ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan Functions</h3>
+<a id="index-entry-b84bfdcb0924789b419a611e996c6185d4c9ec71" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lpmf; (int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lpmf</code></a></strong><code>(int y | matrix x, real alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-58e5cd13a2585a62b0a4305dfb70b2276419bd31" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lupmf; (int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lupmf</code></a></strong><code>(int y | matrix x, real alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8595b1aadcd41c4266263ddfa5529f3381f42c02" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lpmf; (int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lpmf</code></a></strong><code>(int y | matrix x, vector alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-09934df75edb200a53b9e3f261e389a5c908bb0e" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lupmf; (int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lupmf</code></a></strong><code>(int y | matrix x, vector alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-e6ed6ee376ada4610e458b39eb40199e201a1752" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lpmf; (array[] int y | row_vector x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, real alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-75fe439a6c3fdf76668cd7afcb6fcc6d7d2aee4e" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lupmf; (array[] int y | row_vector x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, real alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-616a93c521e26ed8b747183681c8f77775ffad47" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lpmf; (array[] int y | row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-06917600c21e8a8c04a19d2e00412d5adc619330" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lupmf; (array[] int y | row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-3c4e9f4b1e229095991ef7d66e815550e6189de8" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lpmf; (array[] int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, real alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-171d0ba3be3d570d3167c621b3a6e7bac0f34b49" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lupmf; (array[] int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, real alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-5ba40a905c796c1bcd54602e2cdde64d36486f54" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lpmf; (array[] int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-75e3b277f495a675db21144dfd54403c5b61e4b1" class="index-entry"></a> <!-- real; bernoulli_logit_glm_lupmf; (array[] int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, vector beta)</code><br> The log Bernoulli probability mass of y given chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-1caea52d2e1f6a4419889e2a233b20178f262a89" class="index-entry"></a> <!-- array[] int; bernoulli_logit_glm_rng; (matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_rng" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_rng</code></a></strong><code>(matrix x, vector alpha, vector beta)</code><br> Generate an array of Bernoulli variates with chances of success <code>inv_logit(alpha + x * beta)</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-486921f4f12ed8fec65ef02c934ac90a947400a0" class="index-entry"></a> <!-- array[] int; bernoulli_logit_glm_rng; (row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#bernoulli_logit_glm_rng" class="unlink" title="Jump to index entry"><code>bernoulli_logit_glm_rng</code></a></strong><code>(row_vector x, vector alpha, vector beta)</code><br> Generate an array of Bernoulli variates with chances of success <code>inv_logit(alpha + x * beta)</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.29</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/conventions_for_probability_functions.html" class="pagination-link" aria-label="Conventions for Probability Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Conventions for Probability Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/bounded_discrete_distributions.html" class="pagination-link" aria-label="Bounded Discrete Distributions">
+        <span class="nav-page-text">Bounded Discrete Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/binary_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/bounded_continuous_distributions.html b/docs/2_39/functions-reference/bounded_continuous_distributions.html
new file mode 100644
index 000000000..eda015ead
--- /dev/null
+++ b/docs/2_39/functions-reference/bounded_continuous_distributions.html
@@ -0,0 +1,1197 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Bounded Continuous Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/distributions_over_unbounded_vectors.html" rel="next">
+<link href="../functions-reference/circular_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Bounded Continuous Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/bounded_continuous_distributions.html">Bounded Continuous Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#bounded-continuous-distributions" id="toc-bounded-continuous-distributions" class="nav-link active" data-scroll-target="#bounded-continuous-distributions">Bounded Continuous Distributions</a>
+  <ul class="collapse">
+  <li><a href="#uniform-distribution" id="toc-uniform-distribution" class="nav-link" data-scroll-target="#uniform-distribution">Uniform distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/bounded_continuous_distributions.html">Bounded Continuous Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="bounded-continuous-distributions" class="level1">
+<h1>Bounded Continuous Distributions</h1>
+<p>The bounded continuous probabilities have support on a finite interval of real numbers.</p>
+<section id="uniform-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="uniform-distribution">Uniform distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}\)</span> and <span class="math inline">\(\beta \in (\alpha,\infty)\)</span>, then for <span class="math inline">\(y \in [\alpha,\beta]\)</span>, <span class="math display">\[\begin{equation*} \text{Uniform}(y|\alpha,\beta) =
+\frac{1}{\beta - \alpha} . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#uniform" class="unlink" title="Jump to index entry"><code>uniform</code></a></strong><code>(alpha, beta)</code></p>
+<p>Increment target log probability density with <code>uniform_lupdf(y | alpha, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-69818db9c270c04d0dc488ff67849b0468ace0a2" class="index-entry"></a> <!-- real; uniform ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-cb6789cd6142d930347246410f7f33f97c3a02d6" class="index-entry"></a> <!-- real; uniform_lpdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#uniform_lpdf" class="unlink" title="Jump to index entry"><code>uniform_lpdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the uniform density of y given lower bound alpha and upper bound beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-7de2ecbf83919f5a16933fa6cc4876a9cced78c9" class="index-entry"></a> <!-- real; uniform_lupdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#uniform_lupdf" class="unlink" title="Jump to index entry"><code>uniform_lupdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-900d3b829004efdfb52e8a87bb4e3db6f8dc7e2f" class="index-entry"></a> <!-- real; uniform_cdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#uniform_cdf" class="unlink" title="Jump to index entry"><code>uniform_cdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The uniform cumulative distribution function of y given lower bound alpha and upper bound beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2d8ca4cf434451528427871a1554072069737db0" class="index-entry"></a> <!-- real; uniform_lcdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#uniform_lcdf" class="unlink" title="Jump to index entry"><code>uniform_lcdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-82f75d6a35e833738553f28093a1965fde067385" class="index-entry"></a> <!-- real; uniform_lccdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#uniform_lccdf" class="unlink" title="Jump to index entry"><code>uniform_lccdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-bbdbbdc16c0990d4e4cc5876ee605cf3c8bcb6be" class="index-entry"></a> <!-- R; uniform_rng; (reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#uniform_rng" class="unlink" title="Jump to index entry"><code>uniform_rng</code></a></strong><code>(reals alpha, reals beta)</code><br> Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/circular_distributions.html" class="pagination-link" aria-label="Circular Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Circular Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="pagination-link" aria-label="Distributions over Unbounded Vectors">
+        <span class="nav-page-text">Distributions over Unbounded Vectors</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/bounded_discrete_distributions.html b/docs/2_39/functions-reference/bounded_discrete_distributions.html
new file mode 100644
index 000000000..36da4d4c7
--- /dev/null
+++ b/docs/2_39/functions-reference/bounded_discrete_distributions.html
@@ -0,0 +1,1768 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Bounded Discrete Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/unbounded_discrete_distributions.html" rel="next">
+<link href="../functions-reference/binary_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Bounded Discrete Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/bounded_discrete_distributions.html">Bounded Discrete Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#bounded-discrete-distributions" id="toc-bounded-discrete-distributions" class="nav-link active" data-scroll-target="#bounded-discrete-distributions">Bounded Discrete Distributions</a>
+  <ul class="collapse">
+  <li><a href="#binomial-distribution" id="toc-binomial-distribution" class="nav-link" data-scroll-target="#binomial-distribution">Binomial distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function" id="toc-probability-mass-function" class="nav-link" data-scroll-target="#probability-mass-function">Probability mass function</a></li>
+  <li><a href="#log-probability-mass-function" id="toc-log-probability-mass-function" class="nav-link" data-scroll-target="#log-probability-mass-function">Log probability mass function</a></li>
+  <li><a href="#gradient-of-log-probability-mass-function" id="toc-gradient-of-log-probability-mass-function" class="nav-link" data-scroll-target="#gradient-of-log-probability-mass-function">Gradient of log probability mass function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#binomial-distribution-logit-parameterization" id="toc-binomial-distribution-logit-parameterization" class="nav-link" data-scroll-target="#binomial-distribution-logit-parameterization">Binomial distribution, logit parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-1" id="toc-probability-mass-function-1" class="nav-link" data-scroll-target="#probability-mass-function-1">Probability mass function</a></li>
+  <li><a href="#log-probability-mass-function-1" id="toc-log-probability-mass-function-1" class="nav-link" data-scroll-target="#log-probability-mass-function-1">Log probability mass function</a></li>
+  <li><a href="#gradient-of-log-probability-mass-function-1" id="toc-gradient-of-log-probability-mass-function-1" class="nav-link" data-scroll-target="#gradient-of-log-probability-mass-function-1">Gradient of log probability mass function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#binomial-logit-glm" id="toc-binomial-logit-glm" class="nav-link" data-scroll-target="#binomial-logit-glm">Binomial-logit generalized linear model (Logistic Regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-2" id="toc-probability-mass-function-2" class="nav-link" data-scroll-target="#probability-mass-function-2">Probability mass function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan Functions</a></li>
+  </ul></li>
+  <li><a href="#beta-binomial-distribution" id="toc-beta-binomial-distribution" class="nav-link" data-scroll-target="#beta-binomial-distribution">Beta-binomial distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-3" id="toc-probability-mass-function-3" class="nav-link" data-scroll-target="#probability-mass-function-3">Probability mass function</a></li>
+  <li><a href="#distribution-statement-3" id="toc-distribution-statement-3" class="nav-link" data-scroll-target="#distribution-statement-3">Distribution statement</a></li>
+  <li><a href="#stan-functions-3" id="toc-stan-functions-3" class="nav-link" data-scroll-target="#stan-functions-3">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#hypergeometric-distribution" id="toc-hypergeometric-distribution" class="nav-link" data-scroll-target="#hypergeometric-distribution">Hypergeometric distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-4" id="toc-probability-mass-function-4" class="nav-link" data-scroll-target="#probability-mass-function-4">Probability mass function</a></li>
+  <li><a href="#distribution-statement-4" id="toc-distribution-statement-4" class="nav-link" data-scroll-target="#distribution-statement-4">Distribution statement</a></li>
+  <li><a href="#stan-functions-4" id="toc-stan-functions-4" class="nav-link" data-scroll-target="#stan-functions-4">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#categorical-distribution" id="toc-categorical-distribution" class="nav-link" data-scroll-target="#categorical-distribution">Categorical distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-functions" id="toc-probability-mass-functions" class="nav-link" data-scroll-target="#probability-mass-functions">Probability mass functions</a></li>
+  <li><a href="#distribution-statement-5" id="toc-distribution-statement-5" class="nav-link" data-scroll-target="#distribution-statement-5">Distribution statement</a></li>
+  <li><a href="#distribution-statement-6" id="toc-distribution-statement-6" class="nav-link" data-scroll-target="#distribution-statement-6">Distribution statement</a></li>
+  <li><a href="#stan-functions-5" id="toc-stan-functions-5" class="nav-link" data-scroll-target="#stan-functions-5">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#categorical-logit-glm" id="toc-categorical-logit-glm" class="nav-link" data-scroll-target="#categorical-logit-glm">Categorical logit generalized linear model (softmax regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-functions-1" id="toc-probability-mass-functions-1" class="nav-link" data-scroll-target="#probability-mass-functions-1">Probability mass functions</a></li>
+  <li><a href="#distribution-statement-7" id="toc-distribution-statement-7" class="nav-link" data-scroll-target="#distribution-statement-7">Distribution statement</a></li>
+  <li><a href="#stan-functions-6" id="toc-stan-functions-6" class="nav-link" data-scroll-target="#stan-functions-6">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#discrete-range-distribution" id="toc-discrete-range-distribution" class="nav-link" data-scroll-target="#discrete-range-distribution">Discrete range distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-functions-2" id="toc-probability-mass-functions-2" class="nav-link" data-scroll-target="#probability-mass-functions-2">Probability mass functions</a></li>
+  <li><a href="#distribution-statement-8" id="toc-distribution-statement-8" class="nav-link" data-scroll-target="#distribution-statement-8">Distribution statement</a></li>
+  <li><a href="#stan-functions-7" id="toc-stan-functions-7" class="nav-link" data-scroll-target="#stan-functions-7">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#ordered-logistic-distribution" id="toc-ordered-logistic-distribution" class="nav-link" data-scroll-target="#ordered-logistic-distribution">Ordered logistic distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-5" id="toc-probability-mass-function-5" class="nav-link" data-scroll-target="#probability-mass-function-5">Probability mass function</a></li>
+  <li><a href="#distribution-statement-9" id="toc-distribution-statement-9" class="nav-link" data-scroll-target="#distribution-statement-9">Distribution statement</a></li>
+  <li><a href="#stan-functions-8" id="toc-stan-functions-8" class="nav-link" data-scroll-target="#stan-functions-8">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#ordered-logistic-generalized-linear-model-ordinal-regression" id="toc-ordered-logistic-generalized-linear-model-ordinal-regression" class="nav-link" data-scroll-target="#ordered-logistic-generalized-linear-model-ordinal-regression">Ordered logistic generalized linear model (ordinal regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-6" id="toc-probability-mass-function-6" class="nav-link" data-scroll-target="#probability-mass-function-6">Probability mass function</a></li>
+  <li><a href="#distribution-statement-10" id="toc-distribution-statement-10" class="nav-link" data-scroll-target="#distribution-statement-10">Distribution statement</a></li>
+  <li><a href="#stan-functions-9" id="toc-stan-functions-9" class="nav-link" data-scroll-target="#stan-functions-9">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#ordered-probit-distribution" id="toc-ordered-probit-distribution" class="nav-link" data-scroll-target="#ordered-probit-distribution">Ordered probit distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-7" id="toc-probability-mass-function-7" class="nav-link" data-scroll-target="#probability-mass-function-7">Probability mass function</a></li>
+  <li><a href="#distribution-statement-11" id="toc-distribution-statement-11" class="nav-link" data-scroll-target="#distribution-statement-11">Distribution statement</a></li>
+  <li><a href="#stan-functions-10" id="toc-stan-functions-10" class="nav-link" data-scroll-target="#stan-functions-10">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/bounded_discrete_distributions.html">Bounded Discrete Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="bounded-discrete-distributions" class="level1">
+<h1>Bounded Discrete Distributions</h1>
+<p>Bounded discrete probability functions have support on <span class="math inline">\(\{ 0, \ldots,
+N \}\)</span> for some upper bound <span class="math inline">\(N\)</span>.</p>
+<section id="binomial-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="binomial-distribution">Binomial distribution</h2>
+<section id="probability-mass-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function">Probability mass function</h3>
+<p>Suppose <span class="math inline">\(N \in \mathbb{N}\)</span> and <span class="math inline">\(\theta \in [0,1]\)</span>, and <span class="math inline">\(n \in
+\{0,\ldots,N\}\)</span>. <span class="math display">\[\begin{equation*} \text{Binomial}(n~|~N,\theta) = \binom{N}{n}
+\theta^n (1 - \theta)^{N - n}. \end{equation*}\]</span></p>
+</section>
+<section id="log-probability-mass-function" class="level3">
+<h3 class="anchored" data-anchor-id="log-probability-mass-function">Log probability mass function</h3>
+<p><span class="math display">\[\begin{eqnarray*} \log \text{Binomial}(n~|~N,\theta) &amp; = &amp; \log
+\Gamma(N+1) - \log \Gamma(n + 1) - \log \Gamma(N- n + 1) \\[4pt] &amp; &amp; {
+} + n \log \theta + (N - n) \log (1 - \theta), \end{eqnarray*}\]</span></p>
+</section>
+<section id="gradient-of-log-probability-mass-function" class="level3">
+<h3 class="anchored" data-anchor-id="gradient-of-log-probability-mass-function">Gradient of log probability mass function</h3>
+<p><span class="math display">\[\begin{equation*} \frac{\partial}{\partial \theta} \log \text{Binomial}(n~|~N,\theta)
+= \frac{n}{\theta} - \frac{N - n}{1 - \theta} \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#binomial" class="unlink" title="Jump to index entry"><code>binomial</code></a></strong><code>(N, theta)</code></p>
+<p>Increment target log probability density with <code>binomial_lupmf(n | N, theta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7076d8b3677d5489e492bb3f6f923f9a62c33372" class="index-entry"></a> <!-- real ; binomial ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-8798e65d04763014c36aa7b9c6509c71c538efb6" class="index-entry"></a> <!-- real; binomial_lpmf; (ints n | ints N, reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_lpmf" class="unlink" title="Jump to index entry"><code>binomial_lpmf</code></a></strong><code>(ints n | ints N, reals theta)</code><br> The log binomial probability mass of n successes in N trials given chance of success theta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-d9a18d8e413d00481fdf0a47e523a5232870fc25" class="index-entry"></a> <!-- real; binomial_lupmf; (ints n | ints N, reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_lupmf" class="unlink" title="Jump to index entry"><code>binomial_lupmf</code></a></strong><code>(ints n | ints N, reals theta)</code><br> The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-304f7e96d0cb2c64c6ff84e3b67d56b0ff58e1cc" class="index-entry"></a> <!-- real; binomial_cdf; (ints n | ints N, reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_cdf" class="unlink" title="Jump to index entry"><code>binomial_cdf</code></a></strong><code>(ints n | ints N, reals theta)</code><br> The binomial cumulative distribution function of n successes in N trials given chance of success theta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0414522d06188bdb5122e227051e36f457a2a1c6" class="index-entry"></a> <!-- real; binomial_lcdf; (ints n | ints N, reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_lcdf" class="unlink" title="Jump to index entry"><code>binomial_lcdf</code></a></strong><code>(ints n | ints N, reals theta)</code><br> The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-51c3d20caad1f513ce40fc1c8af17ec9c1d82eca" class="index-entry"></a> <!-- real; binomial_lccdf; (ints n | ints N, reals theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_lccdf" class="unlink" title="Jump to index entry"><code>binomial_lccdf</code></a></strong><code>(ints n | ints N, reals theta)</code><br> The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-15ce73a3a3caa43cbbe65c4e58495cf18ed8886e" class="index-entry"></a> <!-- R; binomial_rng; (ints N, reals theta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#binomial_rng" class="unlink" title="Jump to index entry"><code>binomial_rng</code></a></strong><code>(ints N, reals theta)</code><br> Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="binomial-distribution-logit-parameterization" class="level2">
+<h2 class="anchored" data-anchor-id="binomial-distribution-logit-parameterization">Binomial distribution, logit parameterization</h2>
+<p>Stan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.</p>
+<section id="probability-mass-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-1">Probability mass function</h3>
+<p>Suppose <span class="math inline">\(N \in \mathbb{N}\)</span>, <span class="math inline">\(\alpha \in \mathbb{R}\)</span>, and <span class="math inline">\(n \in
+\{0,\ldots,N\}\)</span>. Then <span class="math display">\[\begin{eqnarray*}
+\text{BinomialLogit}(n~|~N,\alpha) &amp; = &amp;
+\text{Binomial}(n~|~N,\text{logit}^{-1}(\alpha)) \\[6pt] &amp; = &amp;
+\binom{N}{n} \left( \text{logit}^{-1}(\alpha) \right)^{n}  \left( 1 -
+\text{logit}^{-1}(\alpha) \right)^{N - n}.  \end{eqnarray*}\]</span></p>
+</section>
+<section id="log-probability-mass-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="log-probability-mass-function-1">Log probability mass function</h3>
+<p><span class="math display">\[\begin{eqnarray*} \log \text{BinomialLogit}(n~|~N,\alpha) &amp; = &amp; \log
+\Gamma(N+1) - \log \Gamma(n + 1) - \log \Gamma(N- n + 1) \\[4pt]   &amp; &amp;
+{ } + n \log \text{logit}^{-1}(\alpha) + (N - n) \log \left( 1 -
+\text{logit}^{-1}(\alpha) \right), \end{eqnarray*}\]</span></p>
+</section>
+<section id="gradient-of-log-probability-mass-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="gradient-of-log-probability-mass-function-1">Gradient of log probability mass function</h3>
+<p><span class="math display">\[\begin{equation*} \frac{\partial}{\partial \alpha} \log
+\text{BinomialLogit}(n~|~N,\alpha) =
+\frac{n}{\text{logit}^{-1}(-\alpha)} - \frac{N -
+n}{\text{logit}^{-1}(\alpha)} \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit" class="unlink" title="Jump to index entry"><code>binomial_logit</code></a></strong><code>(N, alpha)</code></p>
+<p>Increment target log probability density with <code>binomial_logit_lupmf(n | N, alpha)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0c90bd9f55ca58f075077bde8a8c7fa4b260b346" class="index-entry"></a> <!-- real; binomial_logit ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-2231c87f532ddd0458a862cf4cc6e80ca618fc99" class="index-entry"></a> <!-- real; binomial_logit_lpmf; (ints n | ints N, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_lpmf</code></a></strong><code>(ints n | ints N, reals alpha)</code><br> The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-2fe1708373b5f137c68a8ac756400a2be58856fd" class="index-entry"></a> <!-- real; binomial_logit_lupmf; (ints n | ints N, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_lupmf</code></a></strong><code>(ints n | ints N, reals alpha)</code><br> The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="binomial-logit-glm" class="level2">
+<h2 class="anchored" data-anchor-id="binomial-logit-glm">Binomial-logit generalized linear model (Logistic Regression)</h2>
+<p>Stan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.</p>
+<section id="probability-mass-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-2">Probability mass function</h3>
+<p>Suppose <span class="math inline">\(N \in \mathbb{N}\)</span>, <span class="math inline">\(x\in \mathbb{R}^{n\cdot m}, \alpha \in \mathbb{R}^n, \beta \in \mathbb{R}^m\)</span>, and <span class="math inline">\(n \in
+\{0,\ldots,N\}\)</span>. Then <span class="math display">\[\begin{align*}
+  &amp;\text{BinomialLogitGLM}(n~|~N, x, \alpha, \beta) = \text{Binomial}(n~|~N,\text{logit}^{-1}(\alpha_i + x_i \cdot \beta)) \\
+  &amp;= \binom{N}{n} \left( \text{logit}^{-1}(\alpha_i + \sum_{1\leq j\leq m}x_{ij}\cdot \beta_j) \right)^{n}  \left( 1 - \text{logit}^{-1}(\alpha_i + \sum_{1\leq j\leq m}x_{ij}\cdot \beta_j) \right)^{N - n}.
+\end{align*}\]</span></p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm" class="unlink" title="Jump to index entry"><code>binomial_logit_glm</code></a></strong><code>(N, x, alpha, beta)</code></p>
+<p>Increment target log probability density with <code>binomial_logit_glm_lupmf(n | N, x, alpha, beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-c8c31102f7b5abf104dd462dd52e20ad4ee53d24" class="index-entry"></a> <!-- real; binomial_logit_glm ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan Functions</h3>
+<a id="index-entry-3da9351d4d4f1762941c69259fcfddf39dd65eae" class="index-entry"></a> <!-- real; binomial_logit_glm_lpmf; (int n | int N, matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lpmf</code></a></strong><code>(int n | int N, matrix x, real alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-de178183d4185e13aa59ffee44cc49a8ba2dc7e9" class="index-entry"></a> <!-- real; binomial_logit_glm_lupmf; (int n | int N, matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lupmf</code></a></strong><code>(int n | int N, matrix x, real alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-a7a84a22f1a74964827e7e3c3c2ecb0233e5cf37" class="index-entry"></a> <!-- real; binomial_logit_glm_lpmf; (int n | int N, matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lpmf</code></a></strong><code>(int n | int N, matrix x, vector alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-588f750b18cf5987a93c0022ecc43c232efc3e24" class="index-entry"></a> <!-- real; binomial_logit_glm_lupmf; (int n | int N, matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lupmf</code></a></strong><code>(int n | int N, matrix x, vector alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-e1b1936f95d710de4ad0682cbf085a530908d39a" class="index-entry"></a> <!-- real; binomial_logit_glm_lpmf; (array[] int n | array[] int N, row_vector x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lpmf</code></a></strong><code>(array[] int n | array[] int N, row_vector x, real alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-428627db76fccf914ceaefa4cd5884bd131556b9" class="index-entry"></a> <!-- real; binomial_logit_glm_lupmf; (array[] int n | array[] int N, row_vector x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lupmf</code></a></strong><code>(array[] int n | array[] int N, row_vector x, real alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-e38992f8d67e6ef5ce4e08abdfb07fe5ff1d3c9a" class="index-entry"></a> <!-- real; binomial_logit_glm_lpmf; (array[] int n | array[] int N, row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lpmf</code></a></strong><code>(array[] int n | array[] int N, row_vector x, vector alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-7d6e3169c60256f12c03411a7509e0370a1ee157" class="index-entry"></a> <!-- real; binomial_logit_glm_lupmf; (array[] int n | array[] int N, row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lupmf</code></a></strong><code>(array[] int n | array[] int N, row_vector x, vector alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-3f85893b1fea02de309015929662b0b20beb9309" class="index-entry"></a> <!-- real; binomial_logit_glm_lpmf; (array[] int n | array[] int N, matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lpmf</code></a></strong><code>(array[] int n | array[] int N, matrix x, real alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-f45f2f92b4005c55ad8f5cdd2285c9708fefe806" class="index-entry"></a> <!-- real; binomial_logit_glm_lupmf; (array[] int n | array[] int N, matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lupmf</code></a></strong><code>(array[] int n | array[] int N, matrix x, real alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-1882ad2cd7ca87f3d1050d5be18f2adcc431d794" class="index-entry"></a> <!-- real; binomial_logit_glm_lpmf; (array[] int n | array[] int N, matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lpmf</code></a></strong><code>(array[] int n | array[] int N, matrix x, vector alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-9889fa94aac823796e6949b79108363bfbbbaa9d" class="index-entry"></a> <!-- real; binomial_logit_glm_lupmf; (array[] int n | array[] int N, matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binomial_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>binomial_logit_glm_lupmf</code></a></strong><code>(array[] int n | array[] int N, matrix x, vector alpha, vector beta)</code><br> The log binomial probability mass of n given N trials and chance of success <code>inv_logit(alpha + x * beta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+</section>
+</section>
+<section id="beta-binomial-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="beta-binomial-distribution">Beta-binomial distribution</h2>
+<section id="probability-mass-function-3" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-3">Probability mass function</h3>
+<p>If <span class="math inline">\(N \in \mathbb{N}\)</span>, <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, and <span class="math inline">\(\beta \in
+\mathbb{R}^+\)</span>, then for <span class="math inline">\(n \in {0,\ldots,N}\)</span>, <span class="math display">\[\begin{equation*}
+\text{BetaBinomial}(n~|~N,\alpha,\beta) = \binom{N}{n}
+\frac{\mathrm{B}(n+\alpha, N -n +   \beta)}{\mathrm{B}(\alpha,\beta)},
+\end{equation*}\]</span> where the beta function <span class="math inline">\(\mathrm{B}(u,v)\)</span> is defined for <span class="math inline">\(u \in
+\mathbb{R}^+\)</span> and <span class="math inline">\(v \in \mathbb{R}^+\)</span> by <span class="math display">\[\begin{equation*} \mathrm{B}(u,v) =
+\frac{\Gamma(u) \ \Gamma(v)}{\Gamma(u + v)}. \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-3" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-3">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial" class="unlink" title="Jump to index entry"><code>beta_binomial</code></a></strong><code>(N, alpha, beta)</code></p>
+<p>Increment target log probability density with <code>beta_binomial_lupmf(n | N, alpha, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-989d2a39dee7b5799bfbc8bc1274f0307b4894a2" class="index-entry"></a> <!-- real; beta_binomial ~; -->
+<p></p>
+</section>
+<section id="stan-functions-3" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
+<a id="index-entry-5e63ca67ce5a7886091457e81ed9ad8abad1af53" class="index-entry"></a> <!-- real; beta_binomial_lpmf; (ints n | ints N, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial_lpmf" class="unlink" title="Jump to index entry"><code>beta_binomial_lpmf</code></a></strong><code>(ints n | ints N, reals alpha, reals beta)</code><br> The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-91a92506bc4098570edad18162c4925e53b060cb" class="index-entry"></a> <!-- real; beta_binomial_lupmf; (ints n | ints N, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial_lupmf" class="unlink" title="Jump to index entry"><code>beta_binomial_lupmf</code></a></strong><code>(ints n | ints N, reals alpha, reals beta)</code><br> The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-ba9143f242e640009de45aa086f5082832c63ac0" class="index-entry"></a> <!-- real; beta_binomial_cdf; (ints n | ints N, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial_cdf" class="unlink" title="Jump to index entry"><code>beta_binomial_cdf</code></a></strong><code>(ints n | ints N, reals alpha, reals beta)</code><br> The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-85c6b546de98c04c4454a1521ecc3fad803b7994" class="index-entry"></a> <!-- real; beta_binomial_lcdf; (ints n | ints N, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial_lcdf" class="unlink" title="Jump to index entry"><code>beta_binomial_lcdf</code></a></strong><code>(ints n | ints N, reals alpha, reals beta)</code><br> The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-865ef09530f64c02214e607ccdee381fca79831c" class="index-entry"></a> <!-- real; beta_binomial_lccdf; (ints n | ints N, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial_lccdf" class="unlink" title="Jump to index entry"><code>beta_binomial_lccdf</code></a></strong><code>(ints n | ints N, reals alpha, reals beta)</code><br> The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-2103ccd2a4395de2cf0bc035105ff78f46f83671" class="index-entry"></a> <!-- R; beta_binomial_rng; (ints N, reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#beta_binomial_rng" class="unlink" title="Jump to index entry"><code>beta_binomial_rng</code></a></strong><code>(ints N, reals alpha, reals beta)</code><br> Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="hypergeometric-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="hypergeometric-distribution">Hypergeometric distribution</h2>
+<section id="probability-mass-function-4" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-4">Probability mass function</h3>
+<p>If <span class="math inline">\(a \in \mathbb{N}\)</span>, <span class="math inline">\(b \in \mathbb{N}\)</span>, and <span class="math inline">\(N \in
+\{0,\ldots,a+b\}\)</span>, then for <span class="math inline">\(n \in \{\max(0,N-b),\ldots,\min(a,N)\}\)</span>, <span class="math display">\[\begin{equation*} \text{Hypergeometric}(n~|~N,a,b) = \frac{\normalsize{\binom{a}{n}
+\binom{b}{N - n}}}      {\normalsize{\binom{a + b}{N}}}. \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-4" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-4">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric" class="unlink" title="Jump to index entry"><code>hypergeometric</code></a></strong><code>(N, a, b)</code></p>
+<p>Increment target log probability density with <code>hypergeometric_lupmf(n | N, a, b)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2fc02e8425e812e91f8ef47c9f53bd538c248cbd" class="index-entry"></a> <!-- real; hypergeometric ~; -->
+<p></p>
+</section>
+<section id="stan-functions-4" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-4">Stan functions</h3>
+<a id="index-entry-74678ac15a9ad9dc38759c7f81421cd6abe7f8e8" class="index-entry"></a> <!-- real; hypergeometric_lpmf; (int n | int N, int a, int b); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_lpmf" class="unlink" title="Jump to index entry"><code>hypergeometric_lpmf</code></a></strong><code>(int n | int N, int a, int b)</code><br> The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-bd50c4360d03f67c2f750c84c2b4b7f53ade58e9" class="index-entry"></a> <!-- real; hypergeometric_lupmf; (int n | int N, int a, int b); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_lupmf" class="unlink" title="Jump to index entry"><code>hypergeometric_lupmf</code></a></strong><code>(int n | int N, int a, int b)</code><br> The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-b7d868252dbebe75b731f601d6af7252a228c9bb" class="index-entry"></a> <!-- int; hypergeometric_rng; (int N, int a, int2 b); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_rng" class="unlink" title="Jump to index entry"><code>hypergeometric_rng</code></a></strong><code>(int N, int a, int b)</code><br> Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="categorical-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="categorical-distribution">Categorical distribution</h2>
+<section id="probability-mass-functions" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-functions">Probability mass functions</h3>
+<p>If <span class="math inline">\(N \in \mathbb{N}\)</span>, <span class="math inline">\(N &gt; 0\)</span>, and if <span class="math inline">\(\theta \in \mathbb{R}^N\)</span> forms an <span class="math inline">\(N\)</span>-simplex (i.e., has nonnegative entries summing to one), then for <span class="math inline">\(y \in \{1,\ldots,N\}\)</span>, <span class="math display">\[\begin{equation*} \text{Categorical}(y~|~\theta) =
+\theta_y. \end{equation*}\]</span> In addition, Stan provides a log-odds scaled categorical distribution, <span class="math display">\[\begin{equation*} \text{CategoricalLogit}(y~|~\beta) =
+\text{Categorical}(y~|~\text{softmax}(\beta)). \end{equation*}\]</span> See <a href="../functions-reference/matrix_operations.html#softmax">the definition of softmax</a> for the definition of the softmax function.</p>
+</section>
+<section id="distribution-statement-5" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-5">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#categorical" class="unlink" title="Jump to index entry"><code>categorical</code></a></strong><code>(theta)</code></p>
+<p>Increment target log probability density with <code>categorical_lupmf(y | theta)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-bd19bcb6308a78b3569b39d80a5050df0380eae5" class="index-entry"></a> <!-- real; categorical ~; -->
+<p></p>
+</section>
+<section id="distribution-statement-6" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-6">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit" class="unlink" title="Jump to index entry"><code>categorical_logit</code></a></strong><code>(beta)</code></p>
+<p>Increment target log probability density with <code>categorical_logit_lupmf(y | beta)</code>.</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-2ffcb74c4ba70a5424f59a10d068421051f4963c" class="index-entry"></a> <!-- real; categorical_logit ~; -->
+<p></p>
+</section>
+<section id="stan-functions-5" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-5">Stan functions</h3>
+<p>All of the categorical distributions are vectorized so that the outcome y can be a single integer (type <code>int</code>) or an array of integers (type <code>array[] int</code>).</p>
+<a id="index-entry-9dfc05c1a7f9f9cf620187547781111da6ad08ab" class="index-entry"></a> <!-- real; categorical_lpmf; (ints y | vector theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_lpmf" class="unlink" title="Jump to index entry"><code>categorical_lpmf</code></a></strong><code>(ints y | vector theta)</code><br> The log categorical probability mass function with outcome(s) y in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-83c502e2e302199ed3e04bed34c4bef07808ba9d" class="index-entry"></a> <!-- real; categorical_lupmf; (ints y | vector theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_lupmf" class="unlink" title="Jump to index entry"><code>categorical_lupmf</code></a></strong><code>(ints y | vector theta)</code><br> The log categorical probability mass function with outcome(s) y in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-da54bfd2ecc17ee462038470f94a63c7be030e08" class="index-entry"></a> <!-- real; categorical_logit_lpmf; (ints y | vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_lpmf" class="unlink" title="Jump to index entry"><code>categorical_logit_lpmf</code></a></strong><code>(ints y | vector beta)</code><br> The log categorical probability mass function with outcome(s) y in <span class="math inline">\(1:N\)</span> given log-odds of outcomes beta.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-07422d7054ee8fa6c1816736064c38c694563486" class="index-entry"></a> <!-- real; categorical_logit_lupmf; (ints y | vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_lupmf" class="unlink" title="Jump to index entry"><code>categorical_logit_lupmf</code></a></strong><code>(ints y | vector beta)</code><br> The log categorical probability mass function with outcome(s) y in <span class="math inline">\(1:N\)</span> given log-odds of outcomes beta dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-e829b16f14961a0a24d2dd1a7895dcda4ce79bd1" class="index-entry"></a> <!-- int; categorical_rng; (vector theta); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#categorical_rng" class="unlink" title="Jump to index entry"><code>categorical_rng</code></a></strong><code>(vector theta)</code><br> Generate a categorical variate with <span class="math inline">\(N\)</span>-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2da030b7d2ce945503ab8ef37e97ee5afc78305e" class="index-entry"></a> <!-- int; categorical_logit_rng; (vector beta); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_rng" class="unlink" title="Jump to index entry"><code>categorical_logit_rng</code></a></strong><code>(vector beta)</code><br> Generate a categorical variate with outcome in range <span class="math inline">\(1:N\)</span> from log-odds vector beta; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.16</i></small>
+</section>
+</section>
+<section id="categorical-logit-glm" class="level2">
+<h2 class="anchored" data-anchor-id="categorical-logit-glm">Categorical logit generalized linear model (softmax regression)</h2>
+<p>Stan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e.&nbsp;a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.</p>
+<p>Note that the implementation does not put any restrictions on the coefficient matrix <span class="math inline">\(\beta\)</span>. It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the <a href="https://mc-stan.org/users/documentation/">Stan User’s Guide</a>.</p>
+<section id="probability-mass-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-functions-1">Probability mass functions</h3>
+<p>If <span class="math inline">\(N,M,K \in \mathbb{N}\)</span>, <span class="math inline">\(N,M,K &gt; 0\)</span>, and if <span class="math inline">\(x\in \mathbb{R}^{M\times K}, \alpha \in \mathbb{R}^N, \beta\in \mathbb{R}^{K\cdot N}\)</span>, then for <span class="math inline">\(y \in \{1,\ldots,N\}^M\)</span>, <span class="math display">\[\begin{equation*}
+\begin{split}
+\text{CategoricalLogitGLM}(y~|~x,\alpha,\beta)
+&amp; =  \prod_{1\leq i \leq M}\text{CategoricalLogit}(y_i~|~\alpha+x_i\cdot\beta)  \\[8pt]
+&amp; =  \prod_{1\leq i \leq M}\text{Categorical}(y_i~|~softmax(\alpha+x_i\cdot\beta)).
+\end{split}
+\end{equation*}\]</span> See <a href="../functions-reference/matrix_operations.html#softmax">the definition of softmax</a> for the definition of the softmax function.</p>
+</section>
+<section id="distribution-statement-7" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-7">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm" class="unlink" title="Jump to index entry"><code>categorical_logit_glm</code></a></strong><code>(x, alpha, beta)</code></p>
+<p>Increment target log probability density with <code>categorical_logit_glm_lupmf(y | x, alpha, beta)</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-9d4042f47493597a30307ede3c9743ad09549f4d" class="index-entry"></a> <!-- real; categorical_logit_glm ~; -->
+<p></p>
+</section>
+<section id="stan-functions-6" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-6">Stan functions</h3>
+<a id="index-entry-4e1f97ef06a17d796b0a7cc5ae22003b03f95401" class="index-entry"></a> <!-- real; categorical_logit_glm_lpmf; (int y | row_vector x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lpmf</code></a></strong><code>(int y | row_vector x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcome <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-3e5402103e17dd48bb2625e6a2b76c48782a1a72" class="index-entry"></a> <!-- real; categorical_logit_glm_lupmf; (int y | row_vector x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lupmf</code></a></strong><code>(int y | row_vector x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcome <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-b41f55e8856f7805a26e7a2ee183d028df87f769" class="index-entry"></a> <!-- real; categorical_logit_glm_lpmf; (int y | matrix x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lpmf</code></a></strong><code>(int y | matrix x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcomes <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-866be3d0f99507bddabbbe0652cce251858d2c37" class="index-entry"></a> <!-- real; categorical_logit_glm_lupmf; (int y | matrix x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lupmf</code></a></strong><code>(int y | matrix x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcomes <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8f756acaa498908962c53d1a6b0af17598fd8524" class="index-entry"></a> <!-- real; categorical_logit_glm_lpmf; (array[] int y | row_vector x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcomes <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-518462f77e8ba90187fbc7302cb54236a0c8b333" class="index-entry"></a> <!-- real; categorical_logit_glm_lupmf; (array[] int y | row_vector x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcomes <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-512bbe1dc00cc3ef23d8272ec1c0da56105f1c42" class="index-entry"></a> <!-- real; categorical_logit_glm_lpmf; (array[] int y | matrix x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lpmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcomes <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-09da9504e8676f6b35327c85d13771d63f77b83b" class="index-entry"></a> <!-- real; categorical_logit_glm_lupmf; (array[] int y | matrix x, vector alpha, matrix beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#categorical_logit_glm_lupmf" class="unlink" title="Jump to index entry"><code>categorical_logit_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, matrix beta)</code><br> The log categorical probability mass function with outcomes <code>y</code> in <span class="math inline">\(1:N\)</span> given <span class="math inline">\(N\)</span>-vector of log-odds of outcomes <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="discrete-range-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="discrete-range-distribution">Discrete range distribution</h2>
+<section id="probability-mass-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-functions-2">Probability mass functions</h3>
+<p>If <span class="math inline">\(l, u \in \mathbb{Z}\)</span> are lower and upper bounds (<span class="math inline">\(l \le u\)</span>), then for any integer <span class="math inline">\(y \in \{l,\ldots,u\}\)</span>, <span class="math display">\[\begin{equation*} \text{DiscreteRange}(y ~|~ l, u) =
+\frac{1}{u - l + 1}. \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-8" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-8">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#discrete_range" class="unlink" title="Jump to index entry"><code>discrete_range</code></a></strong><code>(l, u)</code></p>
+<p>Increment the target log probability density with <code>discrete_range_lupmf(y | l, u)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-081a4566d9bb14c20636b30012e90be941b53978" class="index-entry"></a> <!-- real; discrete_range ~; -->
+<p></p>
+</section>
+<section id="stan-functions-7" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
+<p>All of the discrete range distributions are vectorized so that the outcome <code>y</code> and the bounds <code>l, u</code> can be a single integer (type <code>int</code>) or an array of integers (type <code>array[] int</code>).</p>
+<a id="index-entry-4aeab4cc5d644e56b9409bff2224f5e620c7ce53" class="index-entry"></a> <!-- real; discrete_range_lpmf; (ints y | ints l, ints u); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#discrete_range_lpmf" class="unlink" title="Jump to index entry"><code>discrete_range_lpmf</code></a></strong><code>(ints y | ints l, ints u)</code><br> The log probability mass function with outcome(s) y in <span class="math inline">\(l:u\)</span>.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-32514432b1689ab58c658ff3b435675a6e798cef" class="index-entry"></a> <!-- real; discrete_range_lupmf; (ints y | ints l, ints u); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#discrete_range_lupmf" class="unlink" title="Jump to index entry"><code>discrete_range_lupmf</code></a></strong><code>(ints y | ints l, ints u)</code><br> The log probability mass function with outcome(s) y in <span class="math inline">\(l:u\)</span> dropping constant additive terms.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-92d375a0144eecf4c3bac183f06cde0a573ae0ef" class="index-entry"></a> <!-- real; discrete_range_cdf; (ints y | ints l, ints u); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#discrete_range_cdf" class="unlink" title="Jump to index entry"><code>discrete_range_cdf</code></a></strong><code>(ints y | ints l, ints u)</code><br> The discrete range cumulative distribution function for the given y, lower and upper bounds.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-be12560a3a7c198ea965f08a78dab04d83742159" class="index-entry"></a> <!-- real; discrete_range_lcdf; (ints y | ints l, ints u); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#discrete_range_lcdf" class="unlink" title="Jump to index entry"><code>discrete_range_lcdf</code></a></strong><code>(ints y | ints l, ints u)</code><br> The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-3dc029db4d1f37e783107470ba6665c647f5a202" class="index-entry"></a> <!-- real; discrete_range_lccdf; (ints y | ints l, ints u); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#discrete_range_lccdf" class="unlink" title="Jump to index entry"><code>discrete_range_lccdf</code></a></strong><code>(ints y | ints l, ints u)</code><br> The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-a4c6bdebab12a3547ca7c13ac62e456d0b74c9dc" class="index-entry"></a> <!-- ints; discrete_range_rng; (ints l, ints u); -->
+<p></p>
+<p><code>ints</code> <strong><a href="../functions-reference/functions_index.html#discrete_range_rng" class="unlink" title="Jump to index entry"><code>discrete_range_rng</code></a></strong><code>(ints l, ints u)</code><br> Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.26</i></small>
+</section>
+</section>
+<section id="ordered-logistic-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="ordered-logistic-distribution">Ordered logistic distribution</h2>
+<section id="probability-mass-function-5" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-5">Probability mass function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span> with <span class="math inline">\(K &gt; 2\)</span>, <span class="math inline">\(c \in \mathbb{R}^{K-1}\)</span> such that <span class="math inline">\(c_k &lt; c_{k+1}\)</span> for <span class="math inline">\(k \in \{1,\ldots,K-2\}\)</span>, and <span class="math inline">\(\eta \in
+\mathbb{R}\)</span>, then for <span class="math inline">\(k \in \{1,\ldots,K\}\)</span>, <span class="math display">\[\begin{equation*}
+\text{OrderedLogistic}(k~|~\eta,c) = \left\{ \begin{array}{ll} 1 -
+\text{logit}^{-1}(\eta - c_1)  &amp;  \text{if } k = 1, \\[4pt]
+\text{logit}^{-1}(\eta - c_{k-1}) - \text{logit}^{-1}(\eta - c_{k})  &amp;
+\text{if } 1 &lt; k &lt; K, \text{and} \\[4pt] \text{logit}^{-1}(\eta -
+c_{K-1}) - 0  &amp;  \text{if } k = K. \end{array} \right. \end{equation*}\]</span> The <span class="math inline">\(k=K\)</span> case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the <span class="math inline">\(k=1\)</span> and <span class="math inline">\(k=K\)</span> edge cases can be subsumed into the general definition by setting <span class="math inline">\(c_0 = -\infty\)</span> and <span class="math inline">\(c_K = +\infty\)</span> with <span class="math inline">\(\text{logit}^{-1}(-\infty) = 0\)</span> and <span class="math inline">\(\text{logit}^{-1}(\infty) = 1\)</span>.</p>
+</section>
+<section id="distribution-statement-9" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-9">Distribution statement</h3>
+<p><code>k ~</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic" class="unlink" title="Jump to index entry"><code>ordered_logistic</code></a></strong><code>(eta, c)</code></p>
+<p>Increment target log probability density with <code>ordered_logistic_lupmf(k | eta, c)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2bffcaec141463bcf3650a49ab910f0e9b02c0b0" class="index-entry"></a> <!-- real; ordered_logistic ~; -->
+<p></p>
+</section>
+<section id="stan-functions-8" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-8">Stan functions</h3>
+<a id="index-entry-8ec1ec8642d86b6ff109ab22859c8bc55de49be0" class="index-entry"></a> <!-- real; ordered_logistic_lpmf; (ints k | vector eta, vectors c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_lpmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_lpmf</code></a></strong><code>(ints k | vector eta, vectors c)</code><br> The log ordered logistic probability mass of k given linear predictors <code>eta</code>, and cutpoints <code>c</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-56bd6927e7b4b6a5fd660bda74057a8ad1712bb5" class="index-entry"></a> <!-- real; ordered_logistic_lupmf; (ints k | vector eta, vectors c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_lupmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_lupmf</code></a></strong><code>(ints k | vector eta, vectors c)</code><br> The log ordered logistic probability mass of k given linear predictors <code>eta</code>, and cutpoints <code>c</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-27848db7ec1c230a757f0fab71e6bfa2bd6d155e" class="index-entry"></a> <!-- int; ordered_logistic_rng; (real eta, vector c); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_rng" class="unlink" title="Jump to index entry"><code>ordered_logistic_rng</code></a></strong><code>(real eta, vector c)</code><br> Generate an ordered logistic variate with linear predictor <code>eta</code> and cutpoints <code>c</code>; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="ordered-logistic-generalized-linear-model-ordinal-regression" class="level2">
+<h2 class="anchored" data-anchor-id="ordered-logistic-generalized-linear-model-ordinal-regression">Ordered logistic generalized linear model (ordinal regression)</h2>
+<section id="probability-mass-function-6" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-6">Probability mass function</h3>
+<p>If <span class="math inline">\(N,M,K \in \mathbb{N}\)</span> with <span class="math inline">\(N, M &gt; 0\)</span>, <span class="math inline">\(K &gt; 2\)</span>, <span class="math inline">\(c \in \mathbb{R}^{K-1}\)</span> such that <span class="math inline">\(c_k &lt; c_{k+1}\)</span> for <span class="math inline">\(k \in \{1,\ldots,K-2\}\)</span>, and <span class="math inline">\(x\in \mathbb{R}^{N\times M}, \beta\in \mathbb{R}^M\)</span>, then for <span class="math inline">\(y \in \{1,\ldots,K\}^N\)</span>, <span class="math display">\[\begin{equation*}
+\begin{split}
+\\
+&amp; \text{OrderedLogisticGLM}(y~|~x,\beta,c) \\[8pt]
+&amp; = \prod_{1\leq i \leq N}\text{OrderedLogistic}(y_i~|~x_i\cdot \beta,c) \\
+&amp; = \prod_{1\leq i \leq N}  \left\{ \begin{array}{ll}
+1 - \text{logit}^{-1}(x_i\cdot \beta - c_1)  &amp;  \text{if } y = 1, \\[4pt]
+\text{logit}^{-1}(x_i\cdot \beta - c_{y-1}) - \text{logit}^{-1}(x_i\cdot \beta - c_{y}) &amp; \text{if } 1 &lt; y &lt; K, \text{and} \\[4pt]
+\text{logit}^{-1}(x_i\cdot \beta - c_{K-1}) - 0  &amp;  \text{if } y = K.
+\end{array} \right.
+\end{split}
+\end{equation*}\]</span> The <span class="math inline">\(k=K\)</span> case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the <span class="math inline">\(y=1\)</span> and <span class="math inline">\(y=K\)</span> edge cases can be subsumed into the general definition by setting <span class="math inline">\(c_0 = -\infty\)</span> and <span class="math inline">\(c_K = +\infty\)</span> with <span class="math inline">\(\text{logit}^{-1}(-\infty) = 0\)</span> and <span class="math inline">\(\text{logit}^{-1}(\infty) = 1\)</span>.</p>
+</section>
+<section id="distribution-statement-10" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-10">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm</code></a></strong><code>(x, beta, c)</code></p>
+<p>Increment target log probability density with <code>ordered_logistic_lupmf(y | x, beta, c)</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-2bffcaec141463bcf3650a49ab910f0e9b02c0b0" class="index-entry"></a> <!-- real; ordered_logistic ~; -->
+<p></p>
+</section>
+<section id="stan-functions-9" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-9">Stan functions</h3>
+<a id="index-entry-2c635de14c16f079f1741deefe7257e6a42810dc" class="index-entry"></a> <!-- real; ordered_logistic_glm_lpmf; (int y | row_vector x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lpmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lpmf</code></a></strong><code>(int y | row_vector x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-5c21721436d27b4570a4356fcf87939c6695e2bc" class="index-entry"></a> <!-- real; ordered_logistic_glm_lupmf; (int y | row_vector x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lupmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lupmf</code></a></strong><code>(int y | row_vector x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c dropping constant additive terms. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-c11116cef97e5808f28ca2124ff702866ce897e4" class="index-entry"></a> <!-- real; ordered_logistic_glm_lpmf; (int y | matrix x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lpmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lpmf</code></a></strong><code>(int y | matrix x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-757d0591bda8f231fba81cb4faa1d71ba85deb4f" class="index-entry"></a> <!-- real; ordered_logistic_glm_lupmf; (int y | matrix x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lupmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lupmf</code></a></strong><code>(int y | matrix x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c dropping constant additive terms. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-13f1725d36ca9cd5b34e0f7533be511bf7a2ec70" class="index-entry"></a> <!-- real; ordered_logistic_glm_lpmf; (array[] int y | row_vector x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lpmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-1c7b2fb2eee1fbd25742cd97c0396297b00db969" class="index-entry"></a> <!-- real; ordered_logistic_glm_lupmf; (array[] int y | row_vector x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lupmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c dropping constant additive terms. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-c3713f9c51b422f753ef46eab1b6000b57a85f83" class="index-entry"></a> <!-- real; ordered_logistic_glm_lpmf; (array[] int y | matrix x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lpmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c.&nbsp;The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-55686d3ab941df81acce76e8e1f880716336638b" class="index-entry"></a> <!-- real; ordered_logistic_glm_lupmf; (array[] int y | matrix x, vector beta, vector c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_logistic_glm_lupmf" class="unlink" title="Jump to index entry"><code>ordered_logistic_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, vector beta, vector c)</code><br> The log ordered logistic probability mass of y, given linear predictors <code>x * beta</code>, and cutpoints c dropping constant additive terms. The cutpoints <code>c</code> must be ordered.</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="ordered-probit-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="ordered-probit-distribution">Ordered probit distribution</h2>
+<section id="probability-mass-function-7" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-7">Probability mass function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span> with <span class="math inline">\(K &gt; 2\)</span>, <span class="math inline">\(c \in \mathbb{R}^{K-1}\)</span> such that <span class="math inline">\(c_k &lt; c_{k+1}\)</span> for <span class="math inline">\(k \in \{1,\ldots,K-2\}\)</span>, and <span class="math inline">\(\eta \in
+\mathbb{R}\)</span>, then for <span class="math inline">\(k \in \{1,\ldots,K\}\)</span>, <span class="math display">\[\begin{equation*}
+\text{OrderedProbit}(k~|~\eta,c) = \left\{ \begin{array}{ll} 1 -
+\Phi(\eta - c_1) &amp; \text{if } k = 1, \\[4pt] \Phi(\eta - c_{k-1}) -
+\Phi(\eta - c_{k})  &amp; \text{if } 1 &lt; k &lt; K, \text{and} \\[4pt]
+\Phi(\eta - c_{K-1}) - 0 &amp; \text{if } k = K. \end{array} \right. \end{equation*}\]</span> The <span class="math inline">\(k=K\)</span> case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the <span class="math inline">\(k=1\)</span> and <span class="math inline">\(k=K\)</span> edge cases can be subsumed into the general definition by setting <span class="math inline">\(c_0 =
+-\infty\)</span> and <span class="math inline">\(c_K = +\infty\)</span> with <span class="math inline">\(\Phi(-\infty) = 0\)</span> and <span class="math inline">\(\Phi(\infty) = 1\)</span>.</p>
+</section>
+<section id="distribution-statement-11" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-11">Distribution statement</h3>
+<p><code>k ~</code> <strong><a href="../functions-reference/functions_index.html#ordered_probit" class="unlink" title="Jump to index entry"><code>ordered_probit</code></a></strong><code>(eta, c)</code></p>
+<p>Increment target log probability density with <code>ordered_probit_lupmf(k | eta, c)</code>.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-4a767db733c5f8854e1a70dae8c94ad673adf650" class="index-entry"></a> <!-- real; ordered_probit ~; -->
+<p></p>
+</section>
+<section id="stan-functions-10" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-10">Stan functions</h3>
+<a id="index-entry-e9015f46859715b88bc9f7f4bd6b7f1f9ffc51c4" class="index-entry"></a> <!-- real; ordered_probit_lpmf; (ints k | vector eta, vectors c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_probit_lpmf" class="unlink" title="Jump to index entry"><code>ordered_probit_lpmf</code></a></strong><code>(ints k | vector eta, vectors c)</code><br> The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-3cd2703fe784d2f29b24fb4f3d897ea2b5264da2" class="index-entry"></a> <!-- real; ordered_probit_lupmf; (ints k | vector eta, vectors c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_probit_lupmf" class="unlink" title="Jump to index entry"><code>ordered_probit_lupmf</code></a></strong><code>(ints k | vector eta, vectors c)</code><br> The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-bb0439ad56fbb41b1ca9f56fcb28b32383d9f6b8" class="index-entry"></a> <!-- real; ordered_probit_lpmf; (ints k | real eta, vectors c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_probit_lpmf" class="unlink" title="Jump to index entry"><code>ordered_probit_lpmf</code></a></strong><code>(ints k | real eta, vectors c)</code><br> The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-2099a3d4968bfa03c46d810ff23a8b1279db770d" class="index-entry"></a> <!-- real; ordered_probit_lupmf; (ints k | real eta, vectors c); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ordered_probit_lupmf" class="unlink" title="Jump to index entry"><code>ordered_probit_lupmf</code></a></strong><code>(ints k | real eta, vectors c)</code><br> The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-ce1eb98cf9c98edbf6d662f465f30c9816c1d6fe" class="index-entry"></a> <!-- int; ordered_probit_rng; (real eta, vector c); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#ordered_probit_rng" class="unlink" title="Jump to index entry"><code>ordered_probit_rng</code></a></strong><code>(real eta, vector c)</code><br> Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/binary_distributions.html" class="pagination-link" aria-label="Binary Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Binary Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/unbounded_discrete_distributions.html" class="pagination-link" aria-label="Unbounded Discrete Distributions">
+        <span class="nav-page-text">Unbounded Discrete Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/circular_distributions.html b/docs/2_39/functions-reference/circular_distributions.html
new file mode 100644
index 000000000..a2e8bc372
--- /dev/null
+++ b/docs/2_39/functions-reference/circular_distributions.html
@@ -0,0 +1,1248 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Circular Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/bounded_continuous_distributions.html" rel="next">
+<link href="../functions-reference/continuous_distributions_on_0_1.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Circular Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/circular_distributions.html">Circular Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#circular-distributions" id="toc-circular-distributions" class="nav-link active" data-scroll-target="#circular-distributions">Circular Distributions</a>
+  <ul class="collapse">
+  <li><a href="#von-mises-distribution" id="toc-von-mises-distribution" class="nav-link" data-scroll-target="#von-mises-distribution">Von Mises distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  <li><a href="#numerical-stability" id="toc-numerical-stability" class="nav-link" data-scroll-target="#numerical-stability">Numerical stability</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/circular_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/circular_distributions.html">Circular Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="circular-distributions" class="level1">
+<h1>Circular Distributions</h1>
+<p>Circular distributions are defined for finite values y in any interval of length <span class="math inline">\(2\pi\)</span>.</p>
+<section id="von-mises-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="von-mises-distribution">Von Mises distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\kappa \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in
+\mathbb{R}\)</span>, <span class="math display">\[\begin{equation*} \text{VonMises}(y|\mu,\kappa) =
+\frac{\exp(\kappa\cos(y-\mu))}{2\pi I_0(\kappa)} \!. \end{equation*}\]</span> In order for this density to properly normalize, <span class="math inline">\(y\)</span> must be restricted to some interval <span class="math inline">\((c, c + 2\pi)\)</span> of length <span class="math inline">\(2 \pi\)</span>, because <span class="math display">\[\begin{equation*} \int_{c}^{c +
+2\pi} \text{VonMises}(y|\mu,\kappa) dy = 1. \end{equation*}\]</span> Similarly, if <span class="math inline">\(\mu\)</span> is a parameter, it will typically be restricted to the same range as <span class="math inline">\(y\)</span>.</p>
+<p>If <span class="math inline">\(\kappa &gt; 0\)</span>, a von Mises distribution with its <span class="math inline">\(2 \pi\)</span> interval of support centered around its location <span class="math inline">\(\mu\)</span> will have a single mode at <span class="math inline">\(\mu\)</span>; for example, restricting <span class="math inline">\(y\)</span> to <span class="math inline">\((-\pi,\pi)\)</span> and taking <span class="math inline">\(\mu = 0\)</span> leads to a single local optimum at the mode <span class="math inline">\(\mu\)</span>. If the location <span class="math inline">\(\mu\)</span> is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean <span class="math inline">\(\mu\)</span>.</p>
+<p>For <span class="math inline">\(\kappa = 0\)</span>, the Von Mises distribution corresponds to the circular uniform distribution with density <span class="math inline">\(1 / (2 \pi)\)</span> (independently of the values of <span class="math inline">\(y\)</span> or <span class="math inline">\(\mu\)</span>).</p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#von_mises" class="unlink" title="Jump to index entry"><code>von_mises</code></a></strong><code>(mu, kappa)</code></p>
+<p>Increment target log probability density with <code>von_mises_lupdf(y | mu, kappa)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f221b5198520eda13cdf889d916d8a03cdfa8be2" class="index-entry"></a> <!-- real; von_mises ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-5da3c7b18a3059fdead5df590cb8ec3f315ab761" class="index-entry"></a> <!-- real; von_mises_lpdf; (reals y | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#von_mises_lpdf" class="unlink" title="Jump to index entry"><code>von_mises_lpdf</code></a></strong><code>(reals y | reals mu, reals kappa)</code><br> The log of the von mises density of y given location mu and scale kappa.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-22013d6e68ae63899ad52c563e8bd7509807f9e5" class="index-entry"></a> <!-- real; von_mises_lupdf; (reals y | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#von_mises_lupdf" class="unlink" title="Jump to index entry"><code>von_mises_lupdf</code></a></strong><code>(reals y | reals mu, reals kappa)</code><br> The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-338b3220a2284559e37acf449e5797e8228eccec" class="index-entry"></a> <!-- real; von_mises_cdf; (reals y | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#von_mises_cdf" class="unlink" title="Jump to index entry"><code>von_mises_cdf</code></a></strong><code>(reals y | reals mu, reals kappa)</code><br> The von mises cumulative distribution function of y given location mu and scale kappa.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-007efd59371f452dd7592246dd40178529aafdbf" class="index-entry"></a> <!-- real; von_mises_lcdf; (reals y | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#von_mises_lcdf" class="unlink" title="Jump to index entry"><code>von_mises_lcdf</code></a></strong><code>(reals y | reals mu, reals kappa)</code><br> The log of the von mises cumulative distribution function of y given location mu and scale kappa.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-b597e729ddd78b9e20def4bc010de60d211a0e78" class="index-entry"></a> <!-- real; von_mises_lccdf; (reals y | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#von_mises_lccdf" class="unlink" title="Jump to index entry"><code>von_mises_lccdf</code></a></strong><code>(reals y | reals mu, reals kappa)</code><br> The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-4deed6b0f3f1d7212c757bed3a7ab75376f4463c" class="index-entry"></a> <!-- R; von_mises_rng; (reals mu, reals kappa); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#von_mises_rng" class="unlink" title="Jump to index entry"><code>von_mises_rng</code></a></strong><code>(reals mu, reals kappa)</code><br> Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval <span class="math inline">\([(\mu \mod 2\pi)-\pi,(\mu \mod
+2\pi)+\pi]\)</span>); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+<section id="numerical-stability" class="level3">
+<h3 class="anchored" data-anchor-id="numerical-stability">Numerical stability</h3>
+<p>Evaluating the Von Mises distribution for <span class="math inline">\(\kappa &gt; 100\)</span> is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as <span class="math inline">\(\kappa \rightarrow \infty\)</span>, <span class="math display">\[\begin{equation*}
+\text{VonMises}(y|\mu,\kappa) \rightarrow \text{Normal}(\mu, \sqrt{1 /
+\kappa}). \end{equation*}\]</span> The workaround is to replace <code>y ~ von_mises(mu,kappa)</code> with</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (kappa &lt; <span class="dv">100</span>) {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  y ~ von_mises(mu, kappa);</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sqrt(<span class="dv">1</span> / kappa));</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/continuous_distributions_on_0_1.html" class="pagination-link" aria-label="Continuous Distributions on [0, 1]">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Continuous Distributions on [0, 1]</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/bounded_continuous_distributions.html" class="pagination-link" aria-label="Bounded Continuous Distributions">
+        <span class="nav-page-text">Bounded Continuous Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/circular_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/complex-valued_basic_functions.html b/docs/2_39/functions-reference/complex-valued_basic_functions.html
new file mode 100644
index 000000000..70ee5f82b
--- /dev/null
+++ b/docs/2_39/functions-reference/complex-valued_basic_functions.html
@@ -0,0 +1,1544 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Complex-Valued Basic Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/array_operations.html" rel="next">
+<link href="../functions-reference/real-valued_basic_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Complex-Valued Basic Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/complex-valued_basic_functions.html">Complex-Valued Basic Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#complex-valued-basic-functions" id="toc-complex-valued-basic-functions" class="nav-link active" data-scroll-target="#complex-valued-basic-functions">Complex-Valued Basic Functions</a>
+  <ul class="collapse">
+  <li><a href="#complex-assign-promote" id="toc-complex-assign-promote" class="nav-link" data-scroll-target="#complex-assign-promote">Complex assignment and promotion</a>
+  <ul class="collapse">
+  <li><a href="#complex-function-arguments" id="toc-complex-function-arguments" class="nav-link" data-scroll-target="#complex-function-arguments">Complex function arguments</a></li>
+  </ul></li>
+  <li><a href="#complex-constructors" id="toc-complex-constructors" class="nav-link" data-scroll-target="#complex-constructors">Complex constructors and accessors</a>
+  <ul class="collapse">
+  <li><a href="#complex-constructors-1" id="toc-complex-constructors-1" class="nav-link" data-scroll-target="#complex-constructors-1">Complex constructors</a></li>
+  <li><a href="#complex-accessors" id="toc-complex-accessors" class="nav-link" data-scroll-target="#complex-accessors">Complex accessors</a></li>
+  </ul></li>
+  <li><a href="#complex-arithmetic" id="toc-complex-arithmetic" class="nav-link" data-scroll-target="#complex-arithmetic">Complex arithmetic operators</a>
+  <ul class="collapse">
+  <li><a href="#unary-operators" id="toc-unary-operators" class="nav-link" data-scroll-target="#unary-operators">Unary operators</a></li>
+  <li><a href="#binary-operators" id="toc-binary-operators" class="nav-link" data-scroll-target="#binary-operators">Binary operators</a></li>
+  </ul></li>
+  <li><a href="#complex-comparison" id="toc-complex-comparison" class="nav-link" data-scroll-target="#complex-comparison">Complex comparison operators</a></li>
+  <li><a href="#complex-assignment" id="toc-complex-assignment" class="nav-link" data-scroll-target="#complex-assignment">Complex (compound) assignment operators</a></li>
+  <li><a href="#complex-special" id="toc-complex-special" class="nav-link" data-scroll-target="#complex-special">Complex special functions</a></li>
+  <li><a href="#complex-exponential" id="toc-complex-exponential" class="nav-link" data-scroll-target="#complex-exponential">Complex exponential and power functions</a></li>
+  <li><a href="#complex-trig" id="toc-complex-trig" class="nav-link" data-scroll-target="#complex-trig">Complex trigonometric functions</a></li>
+  <li><a href="#complex-h-trig" id="toc-complex-h-trig" class="nav-link" data-scroll-target="#complex-h-trig">Complex hyperbolic trigonometric functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/complex-valued_basic_functions.html">Complex-Valued Basic Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="complex-valued-basic-functions" class="level1">
+<h1>Complex-Valued Basic Functions</h1>
+<p>This chapter describes built-in functions that operate on complex numbers, either as an argument type or a return type. This includes the arithmetic operators generalized to complex numbers.</p>
+<section id="complex-assign-promote" class="level2">
+<h2 class="anchored" data-anchor-id="complex-assign-promote">Complex assignment and promotion</h2>
+<p>Just as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n = <span class="dv">5</span>;       <span class="co">// n = 5</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = a;      <span class="co">// x = 5.0</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z1 = n;  <span class="co">// z = 5.0 + 0.0i</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z2 = x;  <span class="co">// z = 5.0 + 0.0i</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="complex-function-arguments" class="level3">
+<h3 class="anchored" data-anchor-id="complex-function-arguments">Complex function arguments</h3>
+<p>Function arguments of type <code>int</code> or <code>real</code> may be promoted to type <code>complex</code>. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if <code>z</code> is complex, then <code>pow(z, 2)</code> will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where <code>z + 2</code> will be legal and produce a complex result. Functions such as <code>arg</code> and <code>conj</code> that are only available for complex numbers can accept integer or real arguments, promoting them to <code>complex</code> before applying the function.</p>
+</section>
+</section>
+<section id="complex-constructors" class="level2">
+<h2 class="anchored" data-anchor-id="complex-constructors">Complex constructors and accessors</h2>
+<section id="complex-constructors-1" class="level3">
+<h3 class="anchored" data-anchor-id="complex-constructors-1">Complex constructors</h3>
+<p>Variables and constants of type <code>complex</code> are constructed from zero, one, or two real numbers.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z1 = to_complex();        <span class="co">// z1 = 0.0 + 0.0i</span></span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> re = -<span class="fl">2.9</span>;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z2 = to_complex(re);      <span class="co">// z2 = -2.9 + 0.0i</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> im = <span class="fl">1.3</span>;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z3 = to_complex(re, im);  <span class="co">// z3 = -2.9 + 1.3i</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<a id="index-entry-a10094b116ddff99cf1ba20f448c331b23a70d51" class="index-entry"></a> <!-- complex; to_complex; (); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#to_complex" class="unlink" title="Jump to index entry"><code>to_complex</code></a></strong><code>()</code><br> Return complex number with real part 0.0 and imaginary part 0.0.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-8cc4a1be479f01313e354510097189d0ab95da1a" class="index-entry"></a> <!-- complex; to_complex; (real re); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#to_complex" class="unlink" title="Jump to index entry"><code>to_complex</code></a></strong><code>(real re)</code><br> Return complex number with real part <code>re</code> and imaginary part 0.0.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-982b3b69d3e75c559ef3cada7b1212eecc087fac" class="index-entry"></a> <!-- complex; to_complex; (real re, real im); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#to_complex" class="unlink" title="Jump to index entry"><code>to_complex</code></a></strong><code>(real re, real im)</code><br> Return complex number with real part <code>re</code> and imaginary part <code>im</code>.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-e003c1a960f7ece5344e1cd438e4a3ce92e225e9" class="index-entry"></a> <!-- Z; to_complex; (T1 re, T2 im); -->
+<p></p>
+<p><code>Z</code> <strong><a href="../functions-reference/functions_index.html#to_complex" class="unlink" title="Jump to index entry"><code>to_complex</code></a></strong><code>(T1 re, T2 im)</code><br> Vectorized implementation of the <code>to_complex</code> function.</p>
+<p><code>T1</code> and <code>T2</code> can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="complex-accessors" class="level3">
+<h3 class="anchored" data-anchor-id="complex-accessors">Complex accessors</h3>
+<p>Given a complex number, its real and imaginary parts can be extracted with the following functions.</p>
+<a id="index-entry-3c18c19d9aa231a8f1a921c7f6c289ef225ae34f" class="index-entry"></a> <!-- real; get_real; (complex z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#get_real" class="unlink" title="Jump to index entry"><code>get_real</code></a></strong><code>(complex z)</code><br> Return the real part of the complex number <code>z</code>.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-15202c172bf8397f0f3a6ecd5c16fe489df20273" class="index-entry"></a> <!-- real; get_imag; (complex z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#get_imag" class="unlink" title="Jump to index entry"><code>get_imag</code></a></strong><code>(complex z)</code><br> Return the imaginary part of the complex number <code>z</code>.</p>
+<small><i>Available since 2.28</i></small>
+</section>
+</section>
+<section id="complex-arithmetic" class="level2">
+<h2 class="anchored" data-anchor-id="complex-arithmetic">Complex arithmetic operators</h2>
+<p>The arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type <code>complex</code>. If there are two arguments and only one is of type <code>complex</code>, then the other will be promoted to type <code>complex</code> before performing the operation.</p>
+<section id="unary-operators" class="level3">
+<h3 class="anchored" data-anchor-id="unary-operators">Unary operators</h3>
+<a id="index-entry-5fcb0e5cdd3aa28c383cc6808af8d0523f581e11" class="index-entry"></a> <!-- complex; operator+; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex z)</code><br> Return the complex argument <code>z</code>, <span class="math display">\[\begin{equation*} +z = z. \end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-34804cfa841db4ee3b8175eb5bbb0a124b639049" class="index-entry"></a> <!-- complex; operator-; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex z)</code><br> Return the negation of the complex argument <code>z</code>, which for <span class="math inline">\(z = x + yi\)</span> is <span class="math display">\[\begin{equation*} -z = -x - yi. \end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f" class="index-entry"></a> <!-- T; operator-; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(T x)</code><br> Vectorized version of <code>operator-</code>. If <code>T x</code> is a (possibly nested) array of complex numbers, <code>-x</code> is the same shape array where each individual value is negated.</p>
+<small><i>Available since 2.31</i></small>
+</section>
+<section id="binary-operators" class="level3">
+<h3 class="anchored" data-anchor-id="binary-operators">Binary operators</h3>
+<a id="index-entry-b67896f1b7c21199de738af1835d35fe36635012" class="index-entry"></a> <!-- complex; operator+; (complex x, complex y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex x, complex y)</code><br> Return the sum of x and y, <span class="math display">\[\begin{equation*} (x + y) = \text{operator+}(x, y) = x + y. \end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-7aaca5ac69e98f6d2bc4db949541e8c69d18e694" class="index-entry"></a> <!-- complex; operator-; (complex x, complex y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex x, complex y)</code><br> Return the difference between x and y, <span class="math display">\[\begin{equation*} (x - y) =
+\text{operator-}(x, y) = x - y. \end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-e3a6425a4fd372a6a28b66eb58c06245046f87ae" class="index-entry"></a> <!-- complex; operator*; (complex x, complex y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex x, complex y)</code><br> Return the product of x and y, <span class="math display">\[\begin{equation*} (x \, * \, y) = \text{operator*}(x, y) = x
+\times y. \end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-2982c4088224c83c22a60e3c585c997d23773024" class="index-entry"></a> <!-- complex; operator/; (complex x, complex y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(complex x, complex y)</code><br> Return the quotient of x and y, <span class="math display">\[\begin{equation*} (x / y) = \text{operator/}(x,y) =
+\frac{x}{y} \end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-25f7fc457739ce80dced6697c7730e60913ababa" class="index-entry"></a> <!-- complex; operator^; (complex x, complex y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator^" class="unlink" title="Jump to index entry"><code>operator^</code></a></strong><code>(complex x, complex y)</code><br> Return x raised to the power of y, <span class="math display">\[\begin{equation*}
+(x^\mathrm{\wedge}y)= \text{operator}^\mathrm{\wedge}(x,y)
+= \textrm{exp}(y \, \log(x)).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+</section>
+</section>
+<section id="complex-comparison" class="level2">
+<h2 class="anchored" data-anchor-id="complex-comparison">Complex comparison operators</h2>
+<p>Complex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>z1 == z2</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is equivalent to</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>get_real(z1) == get_real(z2) &amp;&amp; get_imag(z1) == get_imag(z2)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As with other complex functions, if one of the arguments is of type <code>real</code> or <code>int</code>, it will be promoted to type <code>complex</code> before comparison. For example, if <code>z</code> is of type <code>complex</code>, then <code>z == 0</code> will be true if <code>z</code> has real component equal to 0.0 and complex component equal to 0.0.</p>
+<p><em>Warning:</em> As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.</p>
+<a id="index-entry-4250f55203843166af55fdbb17eefad6ec3d54ef" class="index-entry"></a> <!-- int; operator==; (complex x, complex y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator==" class="unlink" title="Jump to index entry"><code>operator==</code></a></strong><code>(complex x, complex y)</code><br> Return 1 if x is equal to y and 0 otherwise, <span class="math display">\[\begin{equation*}
+(x \,\text{==}\, y)
+\ = \ \text{operator==}(x,y)
+\ = \ \begin{cases} 1 &amp; \text{if $x = y$}, \ \text{and} \\ 0 &amp; \text{otherwise.}
+\end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-9d820ba333127706253b1578dc13615c5898dfe4" class="index-entry"></a> <!-- int; operator!=; (complex x, complex y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator!=" class="unlink" title="Jump to index entry"><code>operator!=</code></a></strong><code>(complex x, complex y)</code><br> Return 1 if x is not equal to y and 0 otherwise, <span class="math display">\[\begin{equation*}
+(x \,\text{!=}\, y)
+\ = \ \text{operator!=}(x,y)
+\ = \ \begin{cases} 1 &amp; \text{if $x \neq y$}, \ \text{and} \\ 0 &amp;
+\text{otherwise.} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+</section>
+<section id="complex-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="complex-assignment">Complex (compound) assignment operators</h2>
+<p>The assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.</p>
+<p>Assignment of complex numbers works elementwise. If an expression of type <code>int</code> or <code>real</code> is assigned to a complex number, it will be promoted before assignment as if calling <code>to_complex()</code>, so that the imaginary component is 0.0.</p>
+<a id="index-entry-90376a831758ab6b86587380c1139d0f1d298b88" class="index-entry"></a> <!-- void; operator=; (complex x, complex y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator=" class="unlink" title="Jump to index entry"><code>operator=</code></a></strong><code>(complex x, complex y)</code><br> <code>y = x;</code> assigns a (copy of) the value of <code>y</code> to <code>x</code>.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-cc6740802b1a8a2be0ba358ef3799d66e260a341" class="index-entry"></a> <!-- void; operator+=; (complex x, complex y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator+=" class="unlink" title="Jump to index entry"><code>operator+=</code></a></strong><code>(complex x, complex y)</code><br> <code>x += y;</code> is equivalent to <code>x = x + y;</code>.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-7bc55fdd3a17228fab1be65f974b778974c9fd12" class="index-entry"></a> <!-- void; operator-=; (complex x, complex y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator-=" class="unlink" title="Jump to index entry"><code>operator-=</code></a></strong><code>(complex x, complex y)</code><br> <code>x -= y;</code> is equivalent to <code>x = x - y;</code>.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-0c45a357fc47d9ccf6a7fb7780ee7d1ada6cc2bd" class="index-entry"></a> <!-- void; operator*=; (complex x, complex y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator\*=" class="unlink" title="Jump to index entry"><code>operator*=</code></a></strong><code>(complex x, complex y)</code><br> <code>x *= y;</code> is equivalent to <code>x = x * y;</code>.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-f6845827ad2fc15504afc3931ae66a1aa1c8c0d8" class="index-entry"></a> <!-- void; operator/=; (complex x, complex y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator/=" class="unlink" title="Jump to index entry"><code>operator/=</code></a></strong><code>(complex x, complex y)</code><br> <code>x /= y;</code> is equivalent to <code>x = x / y;</code>.</p>
+<small><i>Available since 2.28</i></small>
+</section>
+<section id="complex-special" class="level2">
+<h2 class="anchored" data-anchor-id="complex-special">Complex special functions</h2>
+<p>The following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.</p>
+<a id="index-entry-00d0fd704b516cbdb1c9d8cfdceb177a76481cc4" class="index-entry"></a> <!-- real; abs; (complex z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#abs" class="unlink" title="Jump to index entry"><code>abs</code></a></strong><code>(complex z)</code><br> Return the absolute value of z, also known as the modulus or magnitude, which for <span class="math inline">\(z = x + yi\)</span> is <span class="math display">\[\begin{equation*}
+\textrm{abs}(z) = \sqrt{x^2 + y^2}.
+\end{equation*}\]</span></p>
+<p>This function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a <code>complex_vector[n]</code> input will return a <code>vector[n]</code> output, with each element transformed by the above equation.</p>
+<small><i>Available since 2.28, vectorized in 2.30</i></small>
+<a id="index-entry-14bed65ae47bba6447a5f2bb816ef9fe09bb8dd2" class="index-entry"></a> <!-- real; arg; (complex z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#arg" class="unlink" title="Jump to index entry"><code>arg</code></a></strong><code>(complex z)</code><br> Return the phase angle (in radians) of z, which for <span class="math inline">\(z = x + yi\)</span> is <span class="math display">\[\begin{equation*}
+\textrm{arg}(z) = \textrm{atan2}(y, x) = \textrm{atan}(y / x).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-8e1a8cccaab89f6f630e0b34e2292689c9aa46a3" class="index-entry"></a> <!-- real; norm; (complex z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#norm" class="unlink" title="Jump to index entry"><code>norm</code></a></strong><code>(complex z)</code><br> Return the Euclidean norm of z, which is its absolute value squared, and which for <span class="math inline">\(z = x + yi\)</span> is <span class="math display">\[\begin{equation*}
+\textrm{norm}(z) = \textrm{abs}^2(z) = x^2 + y^2.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-bd2ae18bb245caf34d1b2f12b3883c47d8aa7e0a" class="index-entry"></a> <!-- complex; conj; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#conj" class="unlink" title="Jump to index entry"><code>conj</code></a></strong><code>(complex z)</code><br> Return the complex conjugate of z, which negates the imaginary component, so that if <span class="math inline">\(z = x + yi\)</span>, <span class="math display">\[\begin{equation*}
+\textrm{conj}(z) = x - yi.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-17692d922273aaba0c5317617aab6585dda28bfa" class="index-entry"></a> <!-- Z; conj; (Z z); -->
+<p></p>
+<p><code>Z</code> <strong><a href="../functions-reference/functions_index.html#conj" class="unlink" title="Jump to index entry"><code>conj</code></a></strong><code>(Z z)</code><br> Vectorized version of <code>conj</code>. This will apply the <code>conj</code> function to each element of a complex array, vector, or matrix.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-8b9fe1df9b2e6d862d634979cd5a62478d612669" class="index-entry"></a> <!-- complex; proj; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#proj" class="unlink" title="Jump to index entry"><code>proj</code></a></strong><code>(complex z)</code><br> Return the projection of <code>z</code> onto the Riemann sphere, which for <span class="math inline">\(z = x
++ yi\)</span> is <span class="math display">\[\begin{equation*}
+\textrm{proj}(z)
+= \begin{cases}
+    z &amp; \textrm{if} \ z \ \textrm{is finite, and} \\
+    0 + \textrm{sign}(y)i &amp; \textrm{otherwise,}
+\end{cases}
+\end{equation*}\]</span> where <span class="math inline">\(\textrm{sign}(y)\)</span> is -1 if <span class="math inline">\(y\)</span> is negative and 1 otherwise.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-267e629612ebf97b806525573c76f11e23005e4a" class="index-entry"></a> <!-- complex; polar; (real r, real theta); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#polar" class="unlink" title="Jump to index entry"><code>polar</code></a></strong><code>(real r, real theta)</code><br> Return the complex number with magnitude (absolute value) <code>r</code> and phase angle <code>theta</code>.</p>
+<small><i>Available since 2.28</i></small>
+</section>
+<section id="complex-exponential" class="level2">
+<h2 class="anchored" data-anchor-id="complex-exponential">Complex exponential and power functions</h2>
+<p>The exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.</p>
+<a id="index-entry-3c31357925206d6237b540fd4e33930d37dccb79" class="index-entry"></a> <!-- complex; exp; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#exp" class="unlink" title="Jump to index entry"><code>exp</code></a></strong><code>(complex z)</code><br> Return the complex natural exponential of <code>z</code>, which for <span class="math inline">\(z = x + yi\)</span> is <span class="math display">\[\begin{equation*}
+\exp z = \exp(x) \textrm{cis}(y) = \exp(x) (\cos(y) + i \sin(y)).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-2ae2007325e8a3d31124d85afaa37799bb6c8ea6" class="index-entry"></a> <!-- complex; log; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#log" class="unlink" title="Jump to index entry"><code>log</code></a></strong><code>(complex z)</code><br> Return the complex natural logarithm of <code>z</code>, which for <span class="math inline">\(z = \textrm{polar}(r,
+\theta)\)</span> is <span class="math display">\[\begin{equation*}
+\log z = \log r + \theta i.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-4d3907a677d8f4d4f89b1edc324ffe5fa2bfbafe" class="index-entry"></a> <!-- complex; log10; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#log10" class="unlink" title="Jump to index entry"><code>log10</code></a></strong><code>(complex z)</code><br> Return the complex common logarithm of <code>z</code>, <span class="math display">\[\begin{equation*}
+\log_{10} z = \frac{\log z}{\log 10}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-a687c481b458dcd50e13b27c0423402b6332fc08" class="index-entry"></a> <!-- complex; pow; (complex x, complex y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#pow" class="unlink" title="Jump to index entry"><code>pow</code></a></strong><code>(complex x, complex y)</code><br> Return x raised to the power of y, <span class="math display">\[\begin{equation*}
+\text{pow}(x,y) = \textrm{exp}(y \, \log(x)).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-24c507386fa0ba9857a2bc0d50a9e2553e9ceba6" class="index-entry"></a> <!-- Z; pow; (T1 x, T2 y); -->
+<p></p>
+<p><code>Z</code> <strong><a href="../functions-reference/functions_index.html#pow" class="unlink" title="Jump to index entry"><code>pow</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>pow</code> function</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1e8198cd9adf62979a9a07f6d4548f72a1ada06d" class="index-entry"></a> <!-- complex; sqrt; (complex x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sqrt" class="unlink" title="Jump to index entry"><code>sqrt</code></a></strong><code>(complex x)</code><br> Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.</p>
+<small><i>Available since 2.28</i></small>
+</section>
+<section id="complex-trig" class="level2">
+<h2 class="anchored" data-anchor-id="complex-trig">Complex trigonometric functions</h2>
+<p>The standard trigonometric functions are supported for complex numbers.</p>
+<a id="index-entry-3d86aa681cf64514f405ada706ad9969e5f33080" class="index-entry"></a> <!-- complex; cos; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#cos" class="unlink" title="Jump to index entry"><code>cos</code></a></strong><code>(complex z)</code><br> Return the complex cosine of z, which is <span class="math display">\[\begin{equation*}
+\cos(z)
+= \textrm{cosh}(z \, i)
+= \frac{\displaystyle \exp(z \, i) + \exp(-z \, i)}
+       {\displaystyle 2}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-13f23f848a721cdfa87c2d1f4b4067cb9df6911e" class="index-entry"></a> <!-- complex; sin; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sin" class="unlink" title="Jump to index entry"><code>sin</code></a></strong><code>(complex z)</code><br> Return the complex sine of z, <span class="math display">\[\begin{equation*}
+\sin(z)
+= -\textrm{sinh}(z \, i) \, i
+= \frac{\displaystyle \exp(z \, i) - \exp(-z \, i)}
+       {\displaystyle 2 \, i}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-e54110489bc4cabaef154e4257634208c6acce5f" class="index-entry"></a> <!-- complex; tan; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#tan" class="unlink" title="Jump to index entry"><code>tan</code></a></strong><code>(complex z)</code><br> Return the complex tangent of z, <span class="math display">\[\begin{equation*}
+\tan(z)
+= -\textrm{tanh}(z \, i) \, i
+= \frac{(\exp(-z \, i) - \exp(z \, i)) \, i}
+       {\exp(-z \, i) + \exp(z \, i)}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-a3994349ad6faefbd24c6ab7f4c46819b77b7593" class="index-entry"></a> <!-- complex; acos; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#acos" class="unlink" title="Jump to index entry"><code>acos</code></a></strong><code>(complex z)</code><br> Return the complex arc (inverse) cosine of z, <span class="math display">\[\begin{equation*}
+\textrm{acos}(z)
+= \frac{1}{2} \pi + \log (z \, i + \sqrt{1 - z^2}) \, i.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-e8dbabd64d9392bddc505ca75aaf4aa1ef314c3d" class="index-entry"></a> <!-- complex; asin; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#asin" class="unlink" title="Jump to index entry"><code>asin</code></a></strong><code>(complex z)</code><br> Return the complex arc (inverse) sine of z, <span class="math display">\[\begin{equation*}
+\text{asin}(z)
+= -\log(z \, i + \sqrt{1 - z^2}) \, i.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-f60565512ea3debd990f3ae7ee57978b45210673" class="index-entry"></a> <!-- complex; atan; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#atan" class="unlink" title="Jump to index entry"><code>atan</code></a></strong><code>(complex z)</code><br> Return the complex arc (inverse) tangent of z, <span class="math display">\[\begin{equation*}
+\text{atan}(z)
+= - \frac{1}{2} (\log(1 - z \, i) - \log(1 + z \, i)) \, i.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+</section>
+<section id="complex-h-trig" class="level2">
+<h2 class="anchored" data-anchor-id="complex-h-trig">Complex hyperbolic trigonometric functions</h2>
+<p>The standard hyperbolic trigonometric functions are supported for complex numbers.</p>
+<a id="index-entry-9b0fa15f5eebadc83a3678cb57fd8a1f4d089b94" class="index-entry"></a> <!-- complex; cosh; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#cosh" class="unlink" title="Jump to index entry"><code>cosh</code></a></strong><code>(complex z)</code><br> Return the complex hyperbolic cosine of z, <span class="math display">\[\begin{equation*}
+\textrm{cosh}(z)
+= \frac{\exp(z) + \exp(-z)}
+       {2}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-8511329268543fd1404cfddba3d7bf1867ed3665" class="index-entry"></a> <!-- complex; sinh; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sinh" class="unlink" title="Jump to index entry"><code>sinh</code></a></strong><code>(complex z)</code><br> Return the complex hyperbolic sine of z, <span class="math display">\[\begin{equation*}
+\textrm{sinh}(z)
+= \frac{\displaystyle \exp(z) - \exp(-z)}
+       {\displaystyle 2}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-23b49be15163071c1ebe75f992e21daaba8d915f" class="index-entry"></a> <!-- complex; tanh; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#tanh" class="unlink" title="Jump to index entry"><code>tanh</code></a></strong><code>(complex z)</code><br> Return the complex hyperbolic tangent of z, <span class="math display">\[\begin{equation*}
+\textrm{tanh}(z)
+\ = \ \frac{\textrm{sinh}(z)}
+           {\textrm{cosh}(z)}
+\ = \ \frac{\displaystyle \exp(z) - \exp(-z)}
+           {\displaystyle \exp(z) + \exp(-z)}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-5d36ca029281c0668f4125ee4d7400c52fcff71c" class="index-entry"></a> <!-- complex; acosh; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#acosh" class="unlink" title="Jump to index entry"><code>acosh</code></a></strong><code>(complex z)</code><br> Return the complex hyperbolic arc (inverse) cosine of z, <span class="math display">\[\begin{equation*}
+\textrm{acosh}(z)
+= \log(z + \sqrt{(z + 1)(z - 1)}).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-8fd580d63363a17a791a49b5b536b475a5bef834" class="index-entry"></a> <!-- complex; asinh; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#asinh" class="unlink" title="Jump to index entry"><code>asinh</code></a></strong><code>(complex z)</code><br> Return the complex hyperbolic arc (inverse) sine of z, <span class="math display">\[\begin{equation*}
+\textrm{asinh}(z)
+= \log(z + \sqrt{1 + z^2}).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-51283fe4d2aef2c21fe4abe006b88ed2a9ab8b92" class="index-entry"></a> <!-- complex; atanh; (complex z); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#atanh" class="unlink" title="Jump to index entry"><code>atanh</code></a></strong><code>(complex z)</code><br> Return the complex hyperbolic arc (inverse) tangent of z, <span class="math display">\[\begin{equation*}
+\textrm{atanh}(z)
+= \frac{\log(1 + z) - \log(1 - z)}
+       {2}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.28</i></small>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/real-valued_basic_functions.html" class="pagination-link" aria-label="Real-Valued Basic Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Real-Valued Basic Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/array_operations.html" class="pagination-link" aria-label="Array Operations">
+        <span class="nav-page-text">Array Operations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/complex_matrix_operations.html b/docs/2_39/functions-reference/complex_matrix_operations.html
new file mode 100644
index 000000000..35eaadeb8
--- /dev/null
+++ b/docs/2_39/functions-reference/complex_matrix_operations.html
@@ -0,0 +1,2106 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Complex Matrix Operations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/sparse_matrix_operations.html" rel="next">
+<link href="../functions-reference/matrix_operations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Complex Matrix Operations">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/complex_matrix_operations.html">Complex Matrix Operations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#complex_matrix-operations" id="toc-complex_matrix-operations" class="nav-link active" data-scroll-target="#complex_matrix-operations">Complex Matrix Operations</a>
+  <ul class="collapse">
+  <li><a href="#complex-promotion" id="toc-complex-promotion" class="nav-link" data-scroll-target="#complex-promotion">Complex promotion</a>
+  <ul class="collapse">
+  <li><a href="#promotion-of-complex-arguments" id="toc-promotion-of-complex-arguments" class="nav-link" data-scroll-target="#promotion-of-complex-arguments">Promotion of complex arguments</a></li>
+  <li><a href="#signature-selection" id="toc-signature-selection" class="nav-link" data-scroll-target="#signature-selection">Signature selection</a></li>
+  <li><a href="#signatures-for-complex-functions" id="toc-signatures-for-complex-functions" class="nav-link" data-scroll-target="#signatures-for-complex-functions">Signatures for complex functions</a></li>
+  <li><a href="#generic-functions-work-for-complex-containers" id="toc-generic-functions-work-for-complex-containers" class="nav-link" data-scroll-target="#generic-functions-work-for-complex-containers">Generic functions work for complex containers</a></li>
+  </ul></li>
+  <li><a href="#integer-valued-complex-matrix-size-functions" id="toc-integer-valued-complex-matrix-size-functions" class="nav-link" data-scroll-target="#integer-valued-complex-matrix-size-functions">Integer-valued complex matrix size functions</a></li>
+  <li><a href="#complex-matrix-arithmetic-operators" id="toc-complex-matrix-arithmetic-operators" class="nav-link" data-scroll-target="#complex-matrix-arithmetic-operators">Complex matrix arithmetic operators</a>
+  <ul class="collapse">
+  <li><a href="#negation-prefix-operators" id="toc-negation-prefix-operators" class="nav-link" data-scroll-target="#negation-prefix-operators">Negation prefix operators</a></li>
+  <li><a href="#infix-complex_matrix-operators" id="toc-infix-complex_matrix-operators" class="nav-link" data-scroll-target="#infix-complex_matrix-operators">Infix complex_matrix operators</a></li>
+  <li><a href="#broadcast-infix-operators" id="toc-broadcast-infix-operators" class="nav-link" data-scroll-target="#broadcast-infix-operators">Broadcast infix operators</a></li>
+  </ul></li>
+  <li><a href="#complex-transposition-operator" id="toc-complex-transposition-operator" class="nav-link" data-scroll-target="#complex-transposition-operator">Complex Transposition Operator</a></li>
+  <li><a href="#complex-elementwise-functions" id="toc-complex-elementwise-functions" class="nav-link" data-scroll-target="#complex-elementwise-functions">Complex elementwise functions</a></li>
+  <li><a href="#dot-products-and-specialized-products-for-complex-matrices" id="toc-dot-products-and-specialized-products-for-complex-matrices" class="nav-link" data-scroll-target="#dot-products-and-specialized-products-for-complex-matrices">Dot products and specialized products for complex matrices</a>
+  <ul class="collapse">
+  <li><a href="#specialized-products" id="toc-specialized-products" class="nav-link" data-scroll-target="#specialized-products">Specialized products</a></li>
+  </ul></li>
+  <li><a href="#complex-reductions" id="toc-complex-reductions" class="nav-link" data-scroll-target="#complex-reductions">Complex reductions</a>
+  <ul class="collapse">
+  <li><a href="#sums-and-products" id="toc-sums-and-products" class="nav-link" data-scroll-target="#sums-and-products">Sums and products</a></li>
+  </ul></li>
+  <li><a href="#vectorized-accessor-functions" id="toc-vectorized-accessor-functions" class="nav-link" data-scroll-target="#vectorized-accessor-functions">Vectorized accessor functions</a>
+  <ul class="collapse">
+  <li><a href="#type-demotion" id="toc-type-demotion" class="nav-link" data-scroll-target="#type-demotion">Type “demotion”</a></li>
+  <li><a href="#real-and-imaginary-component-accessor-functions" id="toc-real-and-imaginary-component-accessor-functions" class="nav-link" data-scroll-target="#real-and-imaginary-component-accessor-functions">Real and imaginary component accessor functions</a></li>
+  </ul></li>
+  <li><a href="#complex-broadcast-functions" id="toc-complex-broadcast-functions" class="nav-link" data-scroll-target="#complex-broadcast-functions">Complex broadcast functions</a>
+  <ul class="collapse">
+  <li><a href="#symmetrization" id="toc-symmetrization" class="nav-link" data-scroll-target="#symmetrization">Symmetrization</a></li>
+  </ul></li>
+  <li><a href="#diagonal-complex-matrix-functions" id="toc-diagonal-complex-matrix-functions" class="nav-link" data-scroll-target="#diagonal-complex-matrix-functions">Diagonal complex matrix functions</a></li>
+  <li><a href="#slicing-and-blocking-functions-for-complex-matrices" id="toc-slicing-and-blocking-functions-for-complex-matrices" class="nav-link" data-scroll-target="#slicing-and-blocking-functions-for-complex-matrices">Slicing and blocking functions for complex matrices</a>
+  <ul class="collapse">
+  <li><a href="#columns-and-rows" id="toc-columns-and-rows" class="nav-link" data-scroll-target="#columns-and-rows">Columns and rows</a></li>
+  <li><a href="#block-operations" id="toc-block-operations" class="nav-link" data-scroll-target="#block-operations">Block operations</a></li>
+  </ul></li>
+  <li><a href="#complex-matrix-concatenation" id="toc-complex-matrix-concatenation" class="nav-link" data-scroll-target="#complex-matrix-concatenation">Complex matrix concatenation</a>
+  <ul class="collapse">
+  <li><a href="#horizontal-concatenation" id="toc-horizontal-concatenation" class="nav-link" data-scroll-target="#horizontal-concatenation">Horizontal concatenation</a></li>
+  <li><a href="#vertical-concatenation" id="toc-vertical-concatenation" class="nav-link" data-scroll-target="#vertical-concatenation">Vertical concatenation</a></li>
+  </ul></li>
+  <li><a href="#complex-special-matrix-functions" id="toc-complex-special-matrix-functions" class="nav-link" data-scroll-target="#complex-special-matrix-functions">Complex special matrix functions</a>
+  <ul class="collapse">
+  <li><a href="#fast-fourier-transforms" id="toc-fast-fourier-transforms" class="nav-link" data-scroll-target="#fast-fourier-transforms">Fast Fourier transforms</a></li>
+  <li><a href="#cumulative-sums" id="toc-cumulative-sums" class="nav-link" data-scroll-target="#cumulative-sums">Cumulative sums</a></li>
+  </ul></li>
+  <li><a href="#complex-linear-algebra-functions" id="toc-complex-linear-algebra-functions" class="nav-link" data-scroll-target="#complex-linear-algebra-functions">Complex linear algebra functions</a>
+  <ul class="collapse">
+  <li><a href="#complex-matrix-division-operators-and-functions" id="toc-complex-matrix-division-operators-and-functions" class="nav-link" data-scroll-target="#complex-matrix-division-operators-and-functions">Complex matrix division operators and functions</a></li>
+  <li><a href="#linear-algebra-functions" id="toc-linear-algebra-functions" class="nav-link" data-scroll-target="#linear-algebra-functions">Linear algebra functions</a></li>
+  </ul></li>
+  <li><a href="#reverse-functions-for-complex-matrices" id="toc-reverse-functions-for-complex-matrices" class="nav-link" data-scroll-target="#reverse-functions-for-complex-matrices">Reverse functions for complex matrices</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/complex_matrix_operations.html">Complex Matrix Operations</a></li></ol></nav></header>
+
+
+
+
+<section id="complex_matrix-operations" class="level1">
+<h1>Complex Matrix Operations</h1>
+<section id="complex-promotion" class="level2">
+<h2 class="anchored" data-anchor-id="complex-promotion">Complex promotion</h2>
+<p>This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.</p>
+<section id="promotion-of-complex-arguments" class="level3">
+<h3 class="anchored" data-anchor-id="promotion-of-complex-arguments">Promotion of complex arguments</h3>
+<p>If an expression <code>e</code> can be assigned to a variable of type <code>T</code>, then it can be used as an argument to a function that is specified to take arguments of type <code>T</code>. For instance, <code>sqrt(real)</code> is specified to take a <code>real</code> argument, but an integer expression such as <code>2 + 2</code> of type <code>int</code> can be passed to <code>sqrt</code>, so that <code>sqrt(2 + 2)</code> is well defined. This works by promoting the integer expression <code>2 + 2</code> to be of <code>real</code> type.</p>
+<p>The rules for promotion in Stan are simple:</p>
+<ul>
+<li><code>int</code> may be promoted to <code>real</code>,</li>
+<li><code>real</code> may be promoted to <code>complex</code>,</li>
+<li><code>vector</code> can be promoted to <code>complex_vector</code>,</li>
+<li><code>row_vector</code> can be promoted to <code>complex_row_vector</code>,</li>
+<li><code>matrix</code> can be promoted to <code>complex_matrix</code>,</li>
+<li>if <code>T</code> can be promoted to <code>U</code> and <code>U</code> can be promoted to <code>V</code>, then <code>T</code> can be promoted to <code>V</code> (transitive), and</li>
+<li>if <code>T</code> can be promoted to <code>U</code>, then <code>T[]</code> can be promoted to <code>U[]</code> (covariant).</li>
+</ul>
+</section>
+<section id="signature-selection" class="level3">
+<h3 class="anchored" data-anchor-id="signature-selection">Signature selection</h3>
+<p>When a function is called, the definition requiring the fewest number of promotions is used. For example, when calling <code>vector + vector</code>, the real-valued signature is used. When calling any of <code>complex_vector + vector</code>, <code>vector + complex_vector</code>, or <code>complex_vector + complex_vector</code>, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.</p>
+</section>
+<section id="signatures-for-complex-functions" class="level3">
+<h3 class="anchored" data-anchor-id="signatures-for-complex-functions">Signatures for complex functions</h3>
+<p>Complex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, <code>complex operator+(complex_vector, complex_vector)</code>. Through promotion, <code>operator+</code> may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures <code>operator+(complex_vector, vector)</code> and <code>operator+(vector, complex_vector)</code>.</p>
+</section>
+<section id="generic-functions-work-for-complex-containers" class="level3">
+<h3 class="anchored" data-anchor-id="generic-functions-work-for-complex-containers">Generic functions work for complex containers</h3>
+<p>Generic functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions <code>append_array</code>, <code>dims</code>, <code>head</code>, <code>num_elements</code>, <code>rep_array</code>, <code>reverse</code>, <code>segment</code>, <code>size</code>, and <code>tail</code>.</p>
+</section>
+</section>
+<section id="integer-valued-complex-matrix-size-functions" class="level2">
+<h2 class="anchored" data-anchor-id="integer-valued-complex-matrix-size-functions">Integer-valued complex matrix size functions</h2>
+<a id="index-entry-5f67a5c5dc6caa692838121eb1be3a0294ad191f" class="index-entry"></a> <!-- int; num_elements; (complex_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(complex_vector x)</code><br> The total number of elements in the vector x (same as function <code>rows</code>)</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-18b5a5e8da0b97e5db61bc2558be5861bb0e60c9" class="index-entry"></a> <!-- int; num_elements; (complex_row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(complex_row_vector x)</code><br> The total number of elements in the vector x (same as function <code>cols</code>)</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ea9bb96c32abb986ac353999ee38effe98c2999e" class="index-entry"></a> <!-- int; num_elements; (complex_matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(complex_matrix x)</code><br> The total number of elements in the matrix x. For example, if <code>x</code> is a <span class="math inline">\(5 \times 3\)</span> matrix, then <code>num_elements(x)</code> is 15</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-8005f2c9cb8700f40b7a2aa479a6d62cc1ab309d" class="index-entry"></a> <!-- int; rows; (complex_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rows" class="unlink" title="Jump to index entry"><code>rows</code></a></strong><code>(complex_vector x)</code><br> The number of rows in the vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-e14fb4048c19b782910ef083ea77519d2136be69" class="index-entry"></a> <!-- int; rows; (complex_row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rows" class="unlink" title="Jump to index entry"><code>rows</code></a></strong><code>(complex_row_vector x)</code><br> The number of rows in the row vector x, namely 1</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-9ebbc38c2cab2bb85fc80a2aace93188ecfcf794" class="index-entry"></a> <!-- int; rows; (complex_matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rows" class="unlink" title="Jump to index entry"><code>rows</code></a></strong><code>(complex_matrix x)</code><br> The number of rows in the matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-2926894bc112a2f76fc88ec0d1b66ca72f6907ab" class="index-entry"></a> <!-- int; cols; (complex_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#cols" class="unlink" title="Jump to index entry"><code>cols</code></a></strong><code>(complex_vector x)</code><br> The number of columns in the vector x, namely 1</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a3afb9b9b85c96fcbb2ceccffe8220cdf6930ee0" class="index-entry"></a> <!-- int; cols; (complex_row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#cols" class="unlink" title="Jump to index entry"><code>cols</code></a></strong><code>(complex_row_vector x)</code><br> The number of columns in the row vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-22eb74ccaa0e3cbd1875a10b1d25e448aaca1078" class="index-entry"></a> <!-- int; cols; (complex_matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#cols" class="unlink" title="Jump to index entry"><code>cols</code></a></strong><code>(complex_matrix x)</code><br> The number of columns in the matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-4a14d6b215559d7664aabfafde37e9153206b209" class="index-entry"></a> <!-- int; size; (complex_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(complex_vector x)</code><br> The size of <code>x</code>, i.e., the number of elements</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-dd6512340d46b06fe64606e36622bf23ed255eab" class="index-entry"></a> <!-- int; size; (complex_row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(complex_row_vector x)</code><br> The size of <code>x</code>, i.e., the number of elements</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-69098855a034f282c0294e1c6c21cded79ec8a3c" class="index-entry"></a> <!-- int; size; (matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(matrix x)</code><br> The size of the matrix <code>x</code>. For example, if <code>x</code> is a <span class="math inline">\(5 \times 3\)</span> matrix, then <code>size(x)</code> is 15.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="complex-matrix-arithmetic-operators" class="level2">
+<h2 class="anchored" data-anchor-id="complex-matrix-arithmetic-operators">Complex matrix arithmetic operators</h2>
+<p>Stan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.</p>
+<section id="negation-prefix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="negation-prefix-operators">Negation prefix operators</h3>
+<a id="index-entry-e5e977ad448b9356d9822a744773c212eb665b80" class="index-entry"></a> <!-- complex_vector; operator-; (complex_vector x); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_vector x)</code><br> The negation of the vector x.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-4849f1558fb4d3addc5296dec7dc076fc33e2ac8" class="index-entry"></a> <!-- complex_row_vector; operator-; (complex_row_vector x); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_row_vector x)</code><br> The negation of the row vector x.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c67e9b27b61ec87f6671f881e99e0215abca731a" class="index-entry"></a> <!-- complex_matrix; operator-; (complex_matrix x); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_matrix x)</code><br> The negation of the matrix x.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f" class="index-entry"></a> <!-- T; operator-; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(T x)</code><br> Vectorized version of <code>operator-</code>. If <code>T x</code> is a (possibly nested) array of matrix types, <code>-x</code> is the same shape array where each individual value is negated.</p>
+<small><i>Available since 2.31</i></small>
+</section>
+<section id="infix-complex_matrix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="infix-complex_matrix-operators">Infix complex_matrix operators</h3>
+<a id="index-entry-c4da2f5e129a64de4d83c12e2a8d2d79d7cd1619" class="index-entry"></a> <!-- complex_vector; operator+; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The sum of the vectors x and y.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-e4a7e3cd9cefa9964eb063b241f963f7b27007e7" class="index-entry"></a> <!-- complex_row_vector; operator+; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The sum of the row vectors x and y.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1c6e89ffde910406a3f594d11872988647733d8f" class="index-entry"></a> <!-- complex_matrix; operator+; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The sum of the matrices x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-2b25bbe909c531da7ca053dec04d2a0d66767407" class="index-entry"></a> <!-- complex_vector; operator-; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The difference between the vectors x and y.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c159327226df72cad0b808e6e72271542904ae9b" class="index-entry"></a> <!-- complex_row_vector; operator-; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The difference between the row vectors x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-4b7f3f2a15c41213894ee287a6fd4532d8ca59fb" class="index-entry"></a> <!-- complex_matrix; operator-; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The difference between the matrices x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a79cd86a9e357f4daa3c2a357da7fe74a1591449" class="index-entry"></a> <!-- complex_vector; operator*; (complex x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex x, complex_vector y)</code><br> The product of the scalar x and vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-96bd6d9926f3cf86d4abdf08767a6078cd45be14" class="index-entry"></a> <!-- complex_row_vector; operator*; (complex x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex x, complex_row_vector y)</code><br> The product of the scalar x and the row vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a69bac506948c37d27beb39e793ff1339a67bf85" class="index-entry"></a> <!-- complex_matrix; operator*; (complex x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex x, complex_matrix y)</code><br> The product of the scalar x and the matrix y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-23a93f16e6035cc32b84ecf97c32845e9008702c" class="index-entry"></a> <!-- complex_vector; operator*; (complex_vector x, complex y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_vector x, complex y)</code><br> The product of the scalar y and vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6036495abafc183b219e271873881cec3f6e46a3" class="index-entry"></a> <!-- complex_matrix; operator*; (complex_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_vector x, complex_row_vector y)</code><br> The product of the vector x and row vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-989f9db43310c8b2278983062fe9699e275f0f4b" class="index-entry"></a> <!-- complex_row_vector; operator*; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_row_vector x, complex y)</code><br> The product of the scalar y and row vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6a24a75ba519ef262ca78b4f7e1077124ff4e67b" class="index-entry"></a> <!-- complex; operator*; (complex_row_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_row_vector x, complex_vector y)</code><br> The product of the row vector x and vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-eb9d524d09ed2def387f6d32e50434f1d04bb45a" class="index-entry"></a> <!-- complex_row_vector; operator*; (complex_row_vector x, complex_matrix y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_row_vector x, complex_matrix y)</code><br> The product of the row vector x and matrix y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ec01c1a12f2dee572064680e87666ff43581e98e" class="index-entry"></a> <!-- complex_matrix; operator*; (complex_matrix x, complex y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_matrix x, complex y)</code><br> The product of the scalar y and matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-7e59bb7495322cfae3a60e7cd858c8ea8ad00be4" class="index-entry"></a> <!-- complex_vector; operator*; (complex_matrix x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_matrix x, complex_vector y)</code><br> The product of the matrix x and vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-e6c145519223a7e32befa818a711dd8470e7264d" class="index-entry"></a> <!-- complex_matrix; operator*; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The product of the matrices x and y</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="broadcast-infix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="broadcast-infix-operators">Broadcast infix operators</h3>
+<a id="index-entry-e2a4a5c6172fcff3b822f22f17e4012616aa798b" class="index-entry"></a> <!-- complex_vector; operator+; (complex_vector x, complex y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex_vector x, complex y)</code><br> The result of adding y to every entry in the vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5a80c22fc9ab785c199e3ea9ba3903429f3308f9" class="index-entry"></a> <!-- complex_vector; operator+; (complex x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex x, complex_vector y)</code><br> The result of adding x to every entry in the vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-83570c79a3f99ef98736dec5190a738977bf1986" class="index-entry"></a> <!-- complex_row_vector; operator+; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex_row_vector x, complex y)</code><br> The result of adding y to every entry in the row vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-cf423b623dd5296c4bed0996e11793cfff99b12b" class="index-entry"></a> <!-- complex_row_vector; operator+; (complex x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex x, complex_row_vector y)</code><br> The result of adding x to every entry in the row vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-0b1ee3eff7624a28d75d9cbdf065d8cb936c401a" class="index-entry"></a> <!-- complex_matrix; operator+; (complex_matrix x, complex y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex_matrix x, complex y)</code><br> The result of adding y to every entry in the matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-3cbf469cad59cf318ccf139f90a016fc4bfe3db1" class="index-entry"></a> <!-- complex_matrix; operator+; (complex x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(complex x, complex_matrix y)</code><br> The result of adding x to every entry in the matrix y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-750993752e1971b5caa4b40e151244295537237d" class="index-entry"></a> <!-- complex_vector; operator-; (complex_vector x, complex y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_vector x, complex y)</code><br> The result of subtracting y from every entry in the vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-149f85a65fcb71890f2740b37f3d8e99d171c293" class="index-entry"></a> <!-- complex_vector; operator-; (complex x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex x, complex_vector y)</code><br> The result of adding x to every entry in the negation of the vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-23a25b349111e68acc8ccbf019f247a0857b9f9a" class="index-entry"></a> <!-- complex_row_vector; operator-; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_row_vector x, complex y)</code><br> The result of subtracting y from every entry in the row vector x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-be634502437e0887fcb531772138f252af5cd41b" class="index-entry"></a> <!-- complex_row_vector; operator-; (complex x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex x, complex_row_vector y)</code><br> The result of adding x to every entry in the negation of the row vector y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-19c0fa6f4a31be826fff56b079a14d03456e94c6" class="index-entry"></a> <!-- complex_matrix; operator-; (complex_matrix x, complex y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex_matrix x, complex y)</code><br> The result of subtracting y from every entry in the matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a49311f2dd1b5d25aa68b099771fc781e15783c3" class="index-entry"></a> <!-- complex_matrix; operator-; (complex x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(complex x, complex_matrix y)</code><br> The result of adding x to every entry in negation of the matrix y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-0042e2fa5c0ff5c4695b13b4a482a8b37f1ab450" class="index-entry"></a> <!-- complex_vector; operator/; (complex_vector x, complex y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(complex_vector x, complex y)</code><br> The result of dividing each entry in the vector x by y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-05c2018d6b4d3473de214d967f936575bbe2ce9e" class="index-entry"></a> <!-- complex_row_vector; operator/; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(complex_row_vector x, complex y)</code><br> The result of dividing each entry in the row vector x by y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-0e4a06f38977dd429031b5e8cbd2c34d97c11b6d" class="index-entry"></a> <!-- complex_matrix; operator/; (complex_matrix x, complex y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(complex_matrix x, complex y)</code><br> The result of dividing each entry in the matrix x by y</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="complex-transposition-operator" class="level2">
+<h2 class="anchored" data-anchor-id="complex-transposition-operator">Complex Transposition Operator</h2>
+<p>Complex complex_matrix transposition is represented using a postfix operator.</p>
+<a id="index-entry-091a4926b279f9df738ff9dcd356e2436c3f9485" class="index-entry"></a> <!-- complex_matrix; operator'; (complex_matrix x); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator'" class="unlink" title="Jump to index entry"><code>operator'</code></a></strong><code>(complex_matrix x)</code><br> The transpose of the matrix <code>x</code>, written as <code>x'</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-7b294083ea64b206b4b02446640de44855355156" class="index-entry"></a> <!-- complex_row_vector; operator'; (complex_vector x); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator'" class="unlink" title="Jump to index entry"><code>operator'</code></a></strong><code>(complex_vector x)</code><br> The transpose of the vector <code>x</code>, written as <code>x'</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ef45f4afc752618674b3095733d0dad1150a39fd" class="index-entry"></a> <!-- complex_vector; operator'; (complex_row_vector x); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator'" class="unlink" title="Jump to index entry"><code>operator'</code></a></strong><code>(complex_row_vector x)</code><br> The transpose of the row vector <code>x</code>, written as <code>x'</code></p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="complex-elementwise-functions" class="level2">
+<h2 class="anchored" data-anchor-id="complex-elementwise-functions">Complex elementwise functions</h2>
+<p>As in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.</p>
+<a id="index-entry-955eb57e60395b2be45f5af21989a3aff5f5db70" class="index-entry"></a> <!-- complex_vector; operator.*; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.\*" class="unlink" title="Jump to index entry"><code>operator.*</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The elementwise product of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-8225aa3a59b24f94b5efa4533b49050ef65e64ca" class="index-entry"></a> <!-- complex_row_vector; operator.*; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.\*" class="unlink" title="Jump to index entry"><code>operator.*</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The elementwise product of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-8eb7c34b5da648cda9b6f992d4218d023c05d1da" class="index-entry"></a> <!-- complex_matrix; operator.*; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.\*" class="unlink" title="Jump to index entry"><code>operator.*</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The elementwise product of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-0f3d53dc1ff728a9ee7518e0ad5992c1fedcb7ff" class="index-entry"></a> <!-- complex_vector; operator./; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-97bb359024d4f5bab6603c1d02519ed1e7641737" class="index-entry"></a> <!-- complex_vector; operator./; (complex x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex x, complex_vector y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-17e7fc9ed1291f3c3a332e54b65762dcac73f699" class="index-entry"></a> <!-- complex_vector; operator./; (complex_vector x, complex y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex_vector x, complex y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d6ea602ecffe4e62bb465b28956200dec419c158" class="index-entry"></a> <!-- complex_row_vector; operator./; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5380455bf8d01c0d94b7994157ecd464dff285ab" class="index-entry"></a> <!-- complex_row_vector; operator./; (complex x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex x, complex_row_vector y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-f83946686d65ec544820b3c6401ce2e3000cc1a9" class="index-entry"></a> <!-- complex_row_vector; operator./; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex_row_vector x, complex y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a68b100764978dc37adb03c4ac341b3369fa83af" class="index-entry"></a> <!-- complex_matrix; operator./; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-55e666acb9e555e6085cb69b1bf90e25b210d14b" class="index-entry"></a> <!-- complex_matrix; operator./; (complex x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex x, complex_matrix y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-f5684bfa80a80f64afb4be43b829a06f2d3201fa" class="index-entry"></a> <!-- complex_matrix; operator./; (complex_matrix x, complex y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(complex_matrix x, complex y)</code><br> The elementwise quotient of <code>x</code> and <code>y</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-4d24b6cb15adf28b25dc40a155bf4c76980d37d0" class="index-entry"></a> <!-- complex_vector; operator.^; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a7f9b0419a421d43341556c8b8823bb6737fa14c" class="index-entry"></a> <!-- complex_vector; operator.^; (complex_vector x, complex y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex_vector x, complex y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-af111e3482f1da8bfba4fa5bae01a0084ebebca3" class="index-entry"></a> <!-- complex_vector; operator.^; (complex x, complex_vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex x, complex_vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ba84dc5cb8e20d287fb783ebc26ecf24df38f043" class="index-entry"></a> <!-- complex_row_vector; operator.^; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-875148aa0d5deecb6b31fecb24608a167d02fa95" class="index-entry"></a> <!-- complex_row_vector; operator.^; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex_row_vector x, complex y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-fa71ce75b9e1f602439d979ae6e9a6161effce6d" class="index-entry"></a> <!-- complex_row_vector; operator.^; (complex x, complex_row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex x, complex_row_vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-091231072e09466bffb7f4aad16c86348a965b44" class="index-entry"></a> <!-- complex_matrix; operator.^; ( complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>( complex_matrix x, complex_matrix y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1e6130ad7edcc189ada93ab46995b79b280ada4d" class="index-entry"></a> <!-- complex_matrix; operator.^; ( complex_matrix x, complex y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>( complex_matrix x, complex y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-fd8c739930b5a8f56920738e91cec159c5e50783" class="index-entry"></a> <!-- complex_matrix; operator.^; (complex x, complex_matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(complex x, complex_matrix y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="dot-products-and-specialized-products-for-complex-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="dot-products-and-specialized-products-for-complex-matrices">Dot products and specialized products for complex matrices</h2>
+<a id="index-entry-bc037b34a4fdc977a944e39e9277f8e6c700bd5b" class="index-entry"></a> <!-- complex; dot_product; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-dd62ebed52c5e3210bb8a0155e1d5dfa70497439" class="index-entry"></a> <!-- complex; dot_product; (complex_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(complex_vector x, complex_row_vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d932a1eed6d525ce1077543b39b6834836923949" class="index-entry"></a> <!-- complex; dot_product; (complex_row_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(complex_row_vector x, complex_vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-13cdb97ea7e259e98791362de8a261d17dab4e1c" class="index-entry"></a> <!-- complex; dot_product; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-73c167059a6fb3ee21b48776d84d7c14cd6b6b9a" class="index-entry"></a> <!-- complex_row_vector; columns_dot_product; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_product" class="unlink" title="Jump to index entry"><code>columns_dot_product</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The dot product of the columns of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-aee28160ebf0573187fba8d1b9917b490dfd47cd" class="index-entry"></a> <!-- complex_row_vector; columns_dot_product; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_product" class="unlink" title="Jump to index entry"><code>columns_dot_product</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The dot product of the columns of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c0a3895052c73d4b7553b42dfbcf4ceee17d6ebd" class="index-entry"></a> <!-- complex_row_vector; columns_dot_product; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_product" class="unlink" title="Jump to index entry"><code>columns_dot_product</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The dot product of the columns of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-9869bbad286feb6e3252d28a43fac51f950044c7" class="index-entry"></a> <!-- complex_vector; rows_dot_product; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_product" class="unlink" title="Jump to index entry"><code>rows_dot_product</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> The dot product of the rows of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-734a199664218183d672f594e0ba3c472067cf38" class="index-entry"></a> <!-- complex_vector; rows_dot_product; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_product" class="unlink" title="Jump to index entry"><code>rows_dot_product</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> The dot product of the rows of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-74d6233826ba207dc78ce8fa7c02a1177299542d" class="index-entry"></a> <!-- complex_vector; rows_dot_product; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_product" class="unlink" title="Jump to index entry"><code>rows_dot_product</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> The dot product of the rows of x and y</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1ae5feb968f94e219b01c1a6943eb1532161e16f" class="index-entry"></a> <!-- complex; dot_self; (complex_vector x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#dot_self" class="unlink" title="Jump to index entry"><code>dot_self</code></a></strong><code>(complex_vector x)</code><br> The dot product of the vector x with itself</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d0945f42b843344d7b22f5d088d229c5f5ba64bf" class="index-entry"></a> <!-- complex; dot_self; (complex_row_vector x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#dot_self" class="unlink" title="Jump to index entry"><code>dot_self</code></a></strong><code>(complex_row_vector x)</code><br> The dot product of the row vector x with itself</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-25320510319f7efaed97fc7327bcb3481a1fd055" class="index-entry"></a> <!-- complex_row_vector; columns_dot_self; (complex_vector x); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_self" class="unlink" title="Jump to index entry"><code>columns_dot_self</code></a></strong><code>(complex_vector x)</code><br> The dot product of the columns of x with themselves</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5a3c0ca6ac56df87695d6d53c5ed41bd076db1bf" class="index-entry"></a> <!-- complex_row_vector; columns_dot_self; (complex_row_vector x); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_self" class="unlink" title="Jump to index entry"><code>columns_dot_self</code></a></strong><code>(complex_row_vector x)</code><br> The dot product of the columns of x with themselves</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-499a93465efa4feea0949bffdec099ac446865be" class="index-entry"></a> <!-- complex_row_vector; columns_dot_self; (complex_matrix x); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_self" class="unlink" title="Jump to index entry"><code>columns_dot_self</code></a></strong><code>(complex_matrix x)</code><br> The dot product of the columns of x with themselves</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-9089a74ceffaf17dbabc8404c8575d1f4d6f92a9" class="index-entry"></a> <!-- complex_vector; rows_dot_self; (complex_vector x); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_self" class="unlink" title="Jump to index entry"><code>rows_dot_self</code></a></strong><code>(complex_vector x)</code><br> The dot product of the rows of x with themselves</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-f485e7669d290a4148d1373c4705206fdef737e3" class="index-entry"></a> <!-- complex_vector; rows_dot_self; (complex_row_vector x); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_self" class="unlink" title="Jump to index entry"><code>rows_dot_self</code></a></strong><code>(complex_row_vector x)</code><br> The dot product of the rows of x with themselves</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-771452c2e576061539abb7bece2cec0051b51eb7" class="index-entry"></a> <!-- complex_vector; rows_dot_self; (complex_matrix x); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_self" class="unlink" title="Jump to index entry"><code>rows_dot_self</code></a></strong><code>(complex_matrix x)</code><br> The dot product of the rows of x with themselves</p>
+<small><i>Available since 2.30</i></small>
+<section id="specialized-products" class="level3">
+<h3 class="anchored" data-anchor-id="specialized-products">Specialized products</h3>
+<a id="index-entry-7907eb3baed066d578933e3832fe8366c02116eb" class="index-entry"></a> <!-- complex_matrix; diag_pre_multiply; (complex_vector v, complex_matrix m); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_pre_multiply" class="unlink" title="Jump to index entry"><code>diag_pre_multiply</code></a></strong><code>(complex_vector v, complex_matrix m)</code><br> Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., <code>diag_matrix(v) * m</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-29b61ea8d42d42252259e35dbe0c427c485ad9e8" class="index-entry"></a> <!-- complex_matrix; diag_pre_multiply; (complex_row_vector v, complex_matrix m); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_pre_multiply" class="unlink" title="Jump to index entry"><code>diag_pre_multiply</code></a></strong><code>(complex_row_vector v, complex_matrix m)</code><br> Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., <code>diag_matrix(rv) * m</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-b8e527d5b7adcfe81f12d70984ba81d895332835" class="index-entry"></a> <!-- complex_matrix; diag_post_multiply; (complex_matrix m, complex_vector v); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_post_multiply" class="unlink" title="Jump to index entry"><code>diag_post_multiply</code></a></strong><code>(complex_matrix m, complex_vector v)</code><br> Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., <code>m * diag_matrix(v)</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-82629e9f9045325a28b4c0610df36226bc6c4562" class="index-entry"></a> <!-- complex_matrix; diag_post_multiply; (complex_matrix m, complex_row_vector v); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_post_multiply" class="unlink" title="Jump to index entry"><code>diag_post_multiply</code></a></strong><code>(complex_matrix m, complex_row_vector v)</code><br> Return the product of the matrix <code>m</code> and the diagonal matrix formed from the the row vector <code>rv</code>, i.e., <code>m * diag_matrix(rv)</code>.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="complex-reductions" class="level2">
+<h2 class="anchored" data-anchor-id="complex-reductions">Complex reductions</h2>
+<section id="sums-and-products" class="level3">
+<h3 class="anchored" data-anchor-id="sums-and-products">Sums and products</h3>
+<a id="index-entry-3cf6379c6f6f87df7a312af029238c24af388e45" class="index-entry"></a> <!-- complex; sum; (complex_vector x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(complex_vector x)</code><br> The sum of the values in x, or 0 if x is empty</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-bbcdf1952ec801328e03cb3637166b7a9f2f3cff" class="index-entry"></a> <!-- complex; sum; (complex_row_vector x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(complex_row_vector x)</code><br> The sum of the values in x, or 0 if x is empty</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-3cba7a8e0e6f9daf41a8ac994c85688f887c228b" class="index-entry"></a> <!-- complex; sum; (complex_matrix x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(complex_matrix x)</code><br> The sum of the values in x, or 0 if x is empty</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-12d6f5e8068b2b8308d32394e6debc2fa118e082" class="index-entry"></a> <!-- complex; prod; (complex_vector x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(complex_vector x)</code><br> The product of the values in x, or 1 if x is empty</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5353a09f94f67878ef023d0c25bdf91d78c1cfaa" class="index-entry"></a> <!-- complex; prod; (complex_row_vector x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(complex_row_vector x)</code><br> The product of the values in x, or 1 if x is empty</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-2c4545b2f6cd6bbf4e42a4d446d210160e082d42" class="index-entry"></a> <!-- complex; prod; (complex_matrix x); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(complex_matrix x)</code><br> The product of the values in x, or 1 if x is empty</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="vectorized-accessor-functions" class="level2">
+<h2 class="anchored" data-anchor-id="vectorized-accessor-functions">Vectorized accessor functions</h2>
+<p>Much like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.</p>
+<section id="type-demotion" class="level3">
+<h3 class="anchored" data-anchor-id="type-demotion">Type “demotion”</h3>
+<p>These functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a <code>real</code> type. For example, <code>get_real(complex_matrix M)</code> yields a <code>matrix</code> containing the real component of each value in <code>M</code>.</p>
+<p>The following table contains examples of what this notation can mean:</p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;">Type <code>T</code></th>
+<th style="text-align: left;">Type <code>T_demoted</code></th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><code>complex</code></td>
+<td style="text-align: left;"><code>real</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_vector</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex_row_vector</code></td>
+<td style="text-align: left;"><code>row_vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>array[] complex</code></td>
+<td style="text-align: left;"><code>array[] real</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>array[,,] complex</code></td>
+<td style="text-align: left;"><code>array[,,] real</code></td>
+</tr>
+</tbody>
+</table>
+</section>
+<section id="real-and-imaginary-component-accessor-functions" class="level3">
+<h3 class="anchored" data-anchor-id="real-and-imaginary-component-accessor-functions">Real and imaginary component accessor functions</h3>
+<a id="index-entry-2562277862b92c4536cb33d80bfed98e8a586d81" class="index-entry"></a> <!-- T; get_real; (T x); -->
+<p></p>
+<p><code>T_demoted</code> <strong><a href="../functions-reference/functions_index.html#get_real" class="unlink" title="Jump to index entry"><code>get_real</code></a></strong><code>(T x)</code><br> Given an object of complex type <code>T</code>, return the same shape object but of type <code>real</code> by getting the real component of each element of <code>x</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-48dc2caf35c9102abf142e71d6755de12af13e0c" class="index-entry"></a> <!-- T; get_imag; (T x); -->
+<p></p>
+<p><code>T_demoted</code> <strong><a href="../functions-reference/functions_index.html#get_imag" class="unlink" title="Jump to index entry"><code>get_imag</code></a></strong><code>(T x)</code><br> Given an object of complex type <code>T</code>, return the same shape object but of type <code>real</code> by getting the imaginary component of each element of <code>x</code>.</p>
+<small><i>Available since 2.30</i></small>
+<p>For example, given the Stan declaration</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex_vector</span>[<span class="dv">2</span>] z = [<span class="dv">3</span>+<span class="fl">4i</span>, <span class="dv">5</span>+<span class="fl">6i</span>]';</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A call <code>get_real(z)</code> will yield the vector <code>[3, 5]'</code>, and a call <code>get_imag(z)</code> will yield the vector <code>[4, 6]'</code>.</p>
+</section>
+</section>
+<section id="complex-broadcast-functions" class="level2">
+<h2 class="anchored" data-anchor-id="complex-broadcast-functions">Complex broadcast functions</h2>
+<p>The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.</p>
+<a id="index-entry-d9c905ecc938069ed009528bdab94da0534015b1" class="index-entry"></a> <!-- complex_vector; rep_vector; (complex z, int m); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#rep_vector" class="unlink" title="Jump to index entry"><code>rep_vector</code></a></strong><code>(complex z, int m)</code><br> Return the size m (column) vector consisting of copies of z.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5bf4e064db435eff0410b72666643037d22a395c" class="index-entry"></a> <!-- complex_row_vector; rep_row_vector; (complex z, int n); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#rep_row_vector" class="unlink" title="Jump to index entry"><code>rep_row_vector</code></a></strong><code>(complex z, int n)</code><br> Return the size n row vector consisting of copies of z.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-24c948121a2ee087e060ad1257d8719e1b9aa2c3" class="index-entry"></a> <!-- complex_matrix; rep_matrix; (complex z, int m, int n); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#rep_matrix" class="unlink" title="Jump to index entry"><code>rep_matrix</code></a></strong><code>(complex z, int m, int n)</code><br> Return the m by n matrix consisting of copies of z.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-680d23b80116edb9e6752b03801cf43a42571d6a" class="index-entry"></a> <!-- complex_matrix; rep_matrix; (complex_vector v, int n); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#rep_matrix" class="unlink" title="Jump to index entry"><code>rep_matrix</code></a></strong><code>(complex_vector v, int n)</code><br> Return the m by n matrix consisting of n copies of the (column) vector v of size m.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-e106eee17c0bdf0c73908ed222c19f595e566330" class="index-entry"></a> <!-- complex_matrix; rep_matrix; (complex_row_vector rv, int m); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#rep_matrix" class="unlink" title="Jump to index entry"><code>rep_matrix</code></a></strong><code>(complex_row_vector rv, int m)</code><br> Return the m by n matrix consisting of m copies of the row vector rv of size n.</p>
+<small><i>Available since 2.30</i></small>
+<section id="symmetrization" class="level3">
+<h3 class="anchored" data-anchor-id="symmetrization">Symmetrization</h3>
+<a id="index-entry-99c3b184331464981edec14b1ad54d7ed2f94251" class="index-entry"></a> <!-- complex_matrix; symmetrize_from_lower_tri; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#symmetrize_from_lower_tri" class="unlink" title="Jump to index entry"><code>symmetrize_from_lower_tri</code></a></strong><code>(complex_matrix A)</code><br> Construct a symmetric matrix from the lower triangle of A.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="diagonal-complex-matrix-functions" class="level2">
+<h2 class="anchored" data-anchor-id="diagonal-complex-matrix-functions">Diagonal complex matrix functions</h2>
+<a id="index-entry-ebc82f3647c3357e9ff2d0c3de2f5996ac435a22" class="index-entry"></a> <!-- complex_matrix; add_diag; (complex_matrix m, complex_row_vector d); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#add_diag" class="unlink" title="Jump to index entry"><code>add_diag</code></a></strong><code>(complex_matrix m, complex_row_vector d)</code><br> Add row_vector <code>d</code> to the diagonal of matrix <code>m</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c75a27fd3abf5459a2683dc6261be6c0045b71d6" class="index-entry"></a> <!-- complex_matrix; add_diag; (complex_matrix m, complex_vector d); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#add_diag" class="unlink" title="Jump to index entry"><code>add_diag</code></a></strong><code>(complex_matrix m, complex_vector d)</code><br> Add vector <code>d</code> to the diagonal of matrix <code>m</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1bd31f2687a12296c53d88aedd4fe7b91be42e6b" class="index-entry"></a> <!-- complex_matrix; add_diag; (complex_matrix m, complex_real d); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#add_diag" class="unlink" title="Jump to index entry"><code>add_diag</code></a></strong><code>(complex_matrix m, complex d)</code><br> Add scalar <code>d</code> to every diagonal element of matrix <code>m</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5d28ae699b02999166db3502cec1f2c9b74d1c34" class="index-entry"></a> <!-- complex_vector; diagonal; (complex_matrix x); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#diagonal" class="unlink" title="Jump to index entry"><code>diagonal</code></a></strong><code>(complex_matrix x)</code><br> The diagonal of the matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-06edc5538b52d1a11f2a6f45be9ca1e0683bbb1e" class="index-entry"></a> <!-- complex_matrix; diag_matrix; (complex_vector x); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_matrix" class="unlink" title="Jump to index entry"><code>diag_matrix</code></a></strong><code>(complex_vector x)</code><br> The diagonal matrix with diagonal x</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="slicing-and-blocking-functions-for-complex-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="slicing-and-blocking-functions-for-complex-matrices">Slicing and blocking functions for complex matrices</h2>
+<p>Stan provides several functions for generating slices or blocks or diagonal entries for matrices.</p>
+<section id="columns-and-rows" class="level3">
+<h3 class="anchored" data-anchor-id="columns-and-rows">Columns and rows</h3>
+<a id="index-entry-3a0cf3be0b3793d63afbee36001fbeed1f83f08f" class="index-entry"></a> <!-- complex_vector; col; (complex_matrix x, int n); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#col" class="unlink" title="Jump to index entry"><code>col</code></a></strong><code>(complex_matrix x, int n)</code><br> The n-th column of matrix x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-f8cb43eb2a385abfd8880fb1c741f30d7c691414" class="index-entry"></a> <!-- complex_row_vector; row; (complex_matrix x, int m); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#row" class="unlink" title="Jump to index entry"><code>row</code></a></strong><code>(complex_matrix x, int m)</code><br> The m-th row of matrix x</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="block-operations" class="level3">
+<h3 class="anchored" data-anchor-id="block-operations">Block operations</h3>
+<section id="matrix-slicing-operations" class="level4">
+<h4 class="anchored" data-anchor-id="matrix-slicing-operations">Matrix slicing operations</h4>
+<a id="index-entry-2d1266e93b2287bbb26f4015769ebc779ef29f81" class="index-entry"></a> <!-- complex_matrix; block; (complex_matrix x, int i, int j, int n_rows, int n_cols); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#block" class="unlink" title="Jump to index entry"><code>block</code></a></strong><code>(complex_matrix x, int i, int j, int n_rows, int n_cols)</code><br> Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-177f046c456c0586f89f6731e2de2124009c4498" class="index-entry"></a> <!-- complex_vector; sub_col; (complex_matrix x, int i, int j, int n_rows); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#sub_col" class="unlink" title="Jump to index entry"><code>sub_col</code></a></strong><code>(complex_matrix x, int i, int j, int n_rows)</code><br> Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-07fb8f1cd8ba380cf781e1871c50fba8f8c31164" class="index-entry"></a> <!-- complex_row_vector; sub_row; (complex_matrix x, int i, int j, int n_cols); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#sub_row" class="unlink" title="Jump to index entry"><code>sub_row</code></a></strong><code>(complex_matrix x, int i, int j, int n_cols)</code><br> Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="vector-slicing-operations." class="level4">
+<h4 class="anchored" data-anchor-id="vector-slicing-operations.">Vector slicing operations.</h4>
+<a id="index-entry-98d4004db4f7912b486209fbaa83ee186a0b4a0f" class="index-entry"></a> <!-- complex_vector; head; (complex_vector v, int n); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#head" class="unlink" title="Jump to index entry"><code>head</code></a></strong><code>(complex_vector v, int n)</code><br> Return the vector consisting of the first n elements of v.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-9b30a21ff61d6bb263aeae6fe0f8eeaad19b6091" class="index-entry"></a> <!-- complex_row_vector; head; (complex_row_vector rv, int n); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#head" class="unlink" title="Jump to index entry"><code>head</code></a></strong><code>(complex_row_vector rv, int n)</code><br> Return the row vector consisting of the first n elements of rv.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5072d264c287300a4f02919b52048769218f5864" class="index-entry"></a> <!-- complex_vector; tail; (complex_vector v, int n); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#tail" class="unlink" title="Jump to index entry"><code>tail</code></a></strong><code>(complex_vector v, int n)</code><br> Return the vector consisting of the last n elements of v.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-104f1db13950e7c221c9ce66c329f9ebb415fbe2" class="index-entry"></a> <!-- complex_row_vector; tail; (complex_row_vector rv, int n); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#tail" class="unlink" title="Jump to index entry"><code>tail</code></a></strong><code>(complex_row_vector rv, int n)</code><br> Return the row vector consisting of the last n elements of rv.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-96d8e2e13f7a5c50b4fef83a87ac1dcd395ab08f" class="index-entry"></a> <!-- complex_vector; segment; (complex_vector v, int i, int n); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#segment" class="unlink" title="Jump to index entry"><code>segment</code></a></strong><code>(complex_vector v, int i, int n)</code><br> Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-b59ddb42d7b01c4f637d83f1654f9b63e6f1d9cd" class="index-entry"></a> <!-- complex_row_vector; segment; (complex_row_vector rv, int i, int n); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#segment" class="unlink" title="Jump to index entry"><code>segment</code></a></strong><code>(complex_row_vector rv, int i, int n)</code><br> Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+</section>
+<section id="complex-matrix-concatenation" class="level2">
+<h2 class="anchored" data-anchor-id="complex-matrix-concatenation">Complex matrix concatenation</h2>
+<section id="horizontal-concatenation" class="level3">
+<h3 class="anchored" data-anchor-id="horizontal-concatenation">Horizontal concatenation</h3>
+<a id="index-entry-ffc1159a99f60f1f9ed57275f62abef2a8565af2" class="index-entry"></a> <!-- complex_matrix; append_col; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> Combine matrices x and y by column. The matrices must have the same number of rows.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-3cdea5abd8dcc4f4aa3a63ec4b57e3beb1e548d1" class="index-entry"></a> <!-- complex_matrix; append_col; (complex_matrix x, complex_vector y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex_matrix x, complex_vector y)</code><br> Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-925808fc8edc656698d2fa5a95c71e4392b64386" class="index-entry"></a> <!-- complex_matrix; append_col; (complex_vector x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex_vector x, complex_matrix y)</code><br> Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a7e682a73b7b4c9c5e762266afcff98336e95763" class="index-entry"></a> <!-- complex_matrix; append_col; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> Combine vectors x and y by column. The vectors must have the same number of rows.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-e38bc5278f6ed1cde644188c8eb17161f960454d" class="index-entry"></a> <!-- complex_row_vector; append_col; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6d9a3c740a41d03e1694a8957fc0686db16578c7" class="index-entry"></a> <!-- complex_row_vector; append_col; (complex x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex x, complex_row_vector y)</code><br> Append x to the front of y, returning another row vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-435f75dc018a8af24be3ee5ac716757fc6ed0919" class="index-entry"></a> <!-- complex_row_vector; append_col; (complex_row_vector x, complex y); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(complex_row_vector x, complex y)</code><br> Append y to the end of x, returning another row vector.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="vertical-concatenation" class="level3">
+<h3 class="anchored" data-anchor-id="vertical-concatenation">Vertical concatenation</h3>
+<a id="index-entry-ca4bcfd187ccd1d2f4f0816dfc7f2b31f95ff4b2" class="index-entry"></a> <!-- complex_matrix; append_row; (complex_matrix x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex_matrix x, complex_matrix y)</code><br> Combine matrices x and y by row. The matrices must have the same number of columns.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a0c350c64c953d5aa3a9de44bc62448e1b90233d" class="index-entry"></a> <!-- complex_matrix; append_row; (complex_matrix x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex_matrix x, complex_row_vector y)</code><br> Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-cbbd35ff37b9d41a2c66dbb60739e4b40a3b2c92" class="index-entry"></a> <!-- complex_matrix; append_row; (complex_row_vector x, complex_matrix y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex_row_vector x, complex_matrix y)</code><br> Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d3f8a5fb9c0e574f1b8fc41e58b9ff4d768b1316" class="index-entry"></a> <!-- complex_matrix; append_row; (complex_row_vector x, complex_row_vector y); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex_row_vector x, complex_row_vector y)</code><br> Combine row vectors x and y by row. The row vectors must have the same number of columns.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-101152cf279176e6d126b58acb513a382b3076db" class="index-entry"></a> <!-- complex_vector; append_row; (complex_vector x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex_vector x, complex_vector y)</code><br> Concatenate vectors x and y of any size into another vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-a9fd96b44e573f1bb61f0cf8266f4bdc7a805dd3" class="index-entry"></a> <!-- complex_vector; append_row; (complex x, complex_vector y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex x, complex_vector y)</code><br> Append x to the top of y, returning another vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-8140228bb8f902ae61c5e8eba43e81e6d32de7bd" class="index-entry"></a> <!-- complex_vector; append_row; (complex_vector x, complex y); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(complex_vector x, complex y)</code><br> Append y to the bottom of x, returning another vector.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="complex-special-matrix-functions" class="level2">
+<h2 class="anchored" data-anchor-id="complex-special-matrix-functions">Complex special matrix functions</h2>
+<section id="fast-fourier-transforms" class="level3">
+<h3 class="anchored" data-anchor-id="fast-fourier-transforms">Fast Fourier transforms</h3>
+<p>Stan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors <code>u</code> and <code>v</code>,</p>
+<pre><code>    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v</code></pre>
+<p>and in the 2-dimensional case for complex matrices <code>A</code> and <code>B</code>,</p>
+<pre><code>    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B</code></pre>
+<p>Although the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.</p>
+<a id="index-entry-109377c74fb51b412a7ba92432ed83655dc5cca8" class="index-entry"></a> <!-- complex_vector; fft; (complex_vector v); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#fft" class="unlink" title="Jump to index entry"><code>fft</code></a></strong><code>(complex_vector v)</code><br> Return the discrete Fourier transform of the specified complex vector <code>v</code>. If <span class="math inline">\(v \in \mathbb{C}^N\)</span> is a complex vector with <span class="math inline">\(N\)</span> elements and <span class="math inline">\(u =
+\textrm{fft}(v)\)</span>, then <span class="math display">\[\begin{equation*}
+u_n = \sum_{m &lt; n}
+v_m \cdot
+\exp\left(\frac{-n \cdot m \cdot 2 \cdot \pi \cdot \sqrt{-1}}{N}\right).
+\end{equation*}\]</span></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-37fccc5b3423b32e66112f0b9f406c043fa4d0e2" class="index-entry"></a> <!-- complex_matrix; fft2; (complex_matrix m); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#fft2" class="unlink" title="Jump to index entry"><code>fft2</code></a></strong><code>(complex_matrix m)</code><br> Return the 2D discrete Fourier transform of the specified complex matrix <code>m</code>. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-27a0956f66e4b8b0fc804abec9e5245baa2706aa" class="index-entry"></a> <!-- complex_vector; inv_fft; (complex_vector u); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#inv_fft" class="unlink" title="Jump to index entry"><code>inv_fft</code></a></strong><code>(complex_vector u)</code><br> Return the inverse of the discrete Fourier transform of the specified complex vector <code>u</code>. The inverse FFT (this function) is scaled so that <code>fft(inv_fft(u)) == u</code>. If <span class="math inline">\(u \in \mathbb{C}^N\)</span> is a complex vector with <span class="math inline">\(N\)</span> elements and <span class="math inline">\(v = \textrm{fft}^{-1}(u)\)</span>, then <span class="math display">\[\begin{equation*}
+v_n =  \frac{1}{N} \sum_{m &lt; n}
+u_m \cdot
+\exp\left(\frac{n \cdot m \cdot 2 \cdot \pi \cdot \sqrt{-1}}{N}\right).
+\end{equation*}\]</span> This only differs from the FFT by the sign inside the exponential and the scaling. The <span class="math inline">\(\frac{1}{N}\)</span> scaling ensures that <code>fft(inv_fft(u)) == u</code> and <code>inv_fft(fft(v)) == v</code> for complex vectors <code>u</code> and <code>v</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-81586d35e7a520465df734fc44af4a3ad069f695" class="index-entry"></a> <!-- complex_matrix; inv_fft2; (complex_matrix m); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#inv_fft2" class="unlink" title="Jump to index entry"><code>inv_fft2</code></a></strong><code>(complex_matrix m)</code><br> Return the inverse of the 2D discrete Fourier transform of the specified complex matrix <code>m</code>. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures <code>fft2(inv_fft2(A)) == A</code> and <code>inv_fft2(fft2(B)) == B</code>.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="cumulative-sums" class="level3">
+<h3 class="anchored" data-anchor-id="cumulative-sums">Cumulative sums</h3>
+<p>The cumulative sum of a sequence <span class="math inline">\(x_1,\ldots,x_N\)</span> is the sequence <span class="math inline">\(y_1,\ldots,y_N\)</span>, where <span class="math display">\[\begin{equation*} y_n = \sum_{m = 1}^{n} x_m. \end{equation*}\]</span></p>
+<a id="index-entry-d6b4157abb8f1f7140f20c5b9182ad2f7d69fe37" class="index-entry"></a> <!-- array[] complex; cumulative_sum; (array[] complex x); -->
+<p></p>
+<p><code>array[] complex</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(array[] complex x)</code><br> The cumulative sum of x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-45545f14726c28e6b63c8ec09e06e53b33e4ca02" class="index-entry"></a> <!-- complex_vector; cumulative_sum; (complex_vector v); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(complex_vector v)</code><br> The cumulative sum of v</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c104f0e3e10596f559a8a023c84f92f1b919bc0f" class="index-entry"></a> <!-- complex_row_vector; cumulative_sum; (complex_row_vector rv); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(complex_row_vector rv)</code><br> The cumulative sum of rv</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="complex-linear-algebra-functions" class="level2">
+<h2 class="anchored" data-anchor-id="complex-linear-algebra-functions">Complex linear algebra functions</h2>
+<section id="complex-matrix-division-operators-and-functions" class="level3">
+<h3 class="anchored" data-anchor-id="complex-matrix-division-operators-and-functions">Complex matrix division operators and functions</h3>
+<p>In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.</p>
+<section id="complex-matrix-division-operators" class="level4">
+<h4 class="anchored" data-anchor-id="complex-matrix-division-operators">Complex matrix division operators</h4>
+<a id="index-entry-f47bdbfb8220a45ba54119c3805497e1d4596752" class="index-entry"></a> <!-- complex_row_vector; operator/; (complex_row_vector b, complex_matrix A); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(complex_row_vector b, complex_matrix A)</code><br> The right division of b by A; equivalently <code>b * inverse(A)</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-469c065cfa735cf1880ee348a7459ff469411426" class="index-entry"></a> <!-- complex_matrix; operator/; (complex_matrix B, complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(complex_matrix B, complex_matrix A)</code><br> The right division of B by A; equivalently <code>B * inverse(A)</code></p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="linear-algebra-functions" class="level3">
+<h3 class="anchored" data-anchor-id="linear-algebra-functions">Linear algebra functions</h3>
+<section id="trace" class="level4">
+<h4 class="anchored" data-anchor-id="trace">Trace</h4>
+<a id="index-entry-c811524b370c422ec42cb96ee9cedc298094e51a" class="index-entry"></a> <!-- complex; trace; (complex_matrix A); -->
+<p></p>
+<p><code>complex</code> <strong><a href="../functions-reference/functions_index.html#trace" class="unlink" title="Jump to index entry"><code>trace</code></a></strong><code>(complex_matrix A)</code><br> The trace of A, or 0 if A is empty; A is not required to be diagonal</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="eigendecomposition" class="level4">
+<h4 class="anchored" data-anchor-id="eigendecomposition">Eigendecomposition</h4>
+<a id="index-entry-a3edb93c2a79b350e0e1ca880e3ada6deb6a30de" class="index-entry"></a> <!-- complex_vector; eigenvalues; (complex_matrix A); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#eigenvalues" class="unlink" title="Jump to index entry"><code>eigenvalues</code></a></strong><code>(complex_matrix A)</code><br> The complex-valued vector of eigenvalues of the matrix <code>A</code>. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.</p>
+<small><i>Available since 2.32</i></small>
+<a id="index-entry-bc7b059733d8c7e8becdef5196cfe34a63233de0" class="index-entry"></a> <!-- complex_matrix; eigenvectors; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#eigenvectors" class="unlink" title="Jump to index entry"><code>eigenvectors</code></a></strong><code>(complex_matrix A)</code><br> The matrix with the complex-valued (column) eigenvectors of the matrix <code>A</code> in the same order as returned by the function <code>eigenvalues</code></p>
+<small><i>Available since 2.32</i></small>
+<a id="index-entry-cfb26916a57582e2e9ee4f68ed99c821756d3789" class="index-entry"></a> <!-- tuple(complex_matrix, complex_vector); eigendecompose; (complex_matrix A); -->
+<p></p>
+<p><code>tuple(complex_matrix, complex_vector)</code> <strong><a href="../functions-reference/functions_index.html#eigendecompose" class="unlink" title="Jump to index entry"><code>eigendecompose</code></a></strong><code>(complex_matrix A)</code><br> Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix <code>A</code>. This function is equivalent to <code>(eigenvectors(A), eigenvalues(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+<a id="index-entry-e271860ae502125255d5f4092244d884f2ce9692" class="index-entry"></a> <!-- complex_vector; eigenvalues_sym; (complex_matrix A); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#eigenvalues_sym" class="unlink" title="Jump to index entry"><code>eigenvalues_sym</code></a></strong><code>(complex_matrix A)</code><br> The vector of eigenvalues of a symmetric matrix <code>A</code> in ascending order</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-67a04481d07493cc21dae09ba3cfde2b4e974951" class="index-entry"></a> <!-- complex_matrix; eigenvectors_sym; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#eigenvectors_sym" class="unlink" title="Jump to index entry"><code>eigenvectors_sym</code></a></strong><code>(complex_matrix A)</code><br> The matrix with the (column) eigenvectors of symmetric matrix <code>A</code> in the same order as returned by the function <code>eigenvalues_sym</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-2e32adea8a000a491b0d0d5e736075c377a776b6" class="index-entry"></a> <!-- tuple(complex_matrix, complex_vector); eigendecompose_sym; (complex_matrix A); -->
+<p></p>
+<p><code>tuple(complex_matrix, complex_vector)</code> <strong><a href="../functions-reference/functions_index.html#eigendecompose_sym" class="unlink" title="Jump to index entry"><code>eigendecompose_sym</code></a></strong><code>(complex_matrix A)</code><br> Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix <code>A</code>. This function is equivalent to <code>(eigenvectors_sym(A), eigenvalues_sym(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+<p>Because multiplying an eigenvector by <span class="math inline">\(-1\)</span> results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by <span class="math inline">\(-1\)</span>.</p>
+<p>The condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).</p>
+</section>
+<section id="singular-value-decomposition" class="level4">
+<h4 class="anchored" data-anchor-id="singular-value-decomposition">Singular value decomposition</h4>
+<p>The matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, <span class="math display">\[\begin{equation*} A = U D V^T. \end{equation*}\]</span> The matrices of singular vectors here are thin. That is for an <span class="math inline">\(N\)</span> by <span class="math inline">\(P\)</span> input A, <span class="math inline">\(M = min(N, P)\)</span>, U is size <span class="math inline">\(N\)</span> by <span class="math inline">\(M\)</span> and V is size <span class="math inline">\(P\)</span> by <span class="math inline">\(M\)</span>.</p>
+<a id="index-entry-491e5da4f54a933c48db0826ddc9f9bff1b864f2" class="index-entry"></a> <!-- vector; singular_values; (complex_matrix A); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#singular_values" class="unlink" title="Jump to index entry"><code>singular_values</code></a></strong><code>(complex_matrix A)</code><br> The singular values of <code>A</code> in descending order</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-66a9fbf01c3972f54a4e86c8766100b5aa78a187" class="index-entry"></a> <!-- complex_matrix; svd_U; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#svd_U" class="unlink" title="Jump to index entry"><code>svd_U</code></a></strong><code>(complex_matrix A)</code><br> The left-singular vectors of <code>A</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ae2d55818ccca5b294b06be1082ae607075dba37" class="index-entry"></a> <!-- complex_matrix; svd_V; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#svd_V" class="unlink" title="Jump to index entry"><code>svd_V</code></a></strong><code>(complex_matrix A)</code><br> The right-singular vectors of <code>A</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-250461905ab5c8f3351131c12e8ea6f1b3e68795" class="index-entry"></a> <!-- tuple(complex_matrix, vector, complex_matrix); svd; (complex_matrix A); -->
+<p></p>
+<p><code>tuple(complex_matrix, vector, complex_matrix)</code> <strong><a href="../functions-reference/functions_index.html#svd" class="unlink" title="Jump to index entry"><code>svd</code></a></strong><code>(complex_matrix A)</code><br> Returns a tuple containing the left-singular vectors of <code>A</code>, the singular values of <code>A</code> in descending order, and the right-singular values of <code>A</code>. This function is equivalent to <code>(svd_U(A), singular_values(A), svd_V(A))</code> but with a lower computational cost due to the shared work between the different components.</p>
+<small><i>Available since 2.33</i></small>
+</section>
+<section id="complex-schur-decomposition" class="level4">
+<h4 class="anchored" data-anchor-id="complex-schur-decomposition">Complex Schur Decomposition</h4>
+<p>The complex Schur decomposition of a square matrix <span class="math inline">\(A\)</span> produces a complex unitary matrix <span class="math inline">\(U\)</span> and a complex upper-triangular Schur form matrix <span class="math inline">\(T\)</span> such that <span class="math display">\[A = U \cdot T \cdot U^{-1}\]</span></p>
+<p>Since <span class="math inline">\(U\)</span> is unitary, its inverse is also its conjugate transpose, <span class="math inline">\(U^{-1} = U^*\)</span>, <span class="math inline">\(U^*(i, j) = \mathrm{conj}(U(j, i))\)</span></p>
+<a id="index-entry-f43e2c8a9893e96b7c09a618d7b6f69350bfc460" class="index-entry"></a> <!-- complex_matrix; complex_schur_decompose_t; (matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#complex_schur_decompose_t" class="unlink" title="Jump to index entry"><code>complex_schur_decompose_t</code></a></strong><code>(matrix A)</code><br> Compute the upper-triangular Schur form matrix of the complex Schur decomposition of <code>A</code>.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-1728988d586fc721b7a1d7d21d1aa3252bb03f94" class="index-entry"></a> <!-- complex_matrix; complex_schur_decompose_t; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#complex_schur_decompose_t" class="unlink" title="Jump to index entry"><code>complex_schur_decompose_t</code></a></strong><code>(complex_matrix A)</code><br> Compute the upper-triangular Schur form matrix of the complex Schur decomposition of <code>A</code>.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-b169076e82ada507c5c30f36fe4a5bda58af9e0a" class="index-entry"></a> <!-- complex_matrix; complex_schur_decompose_u; (matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#complex_schur_decompose_u" class="unlink" title="Jump to index entry"><code>complex_schur_decompose_u</code></a></strong><code>(matrix A)</code><br> Compute the unitary matrix of the complex Schur decomposition of <code>A</code>.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-fdc477180371e8366e3a46dc88f3875f088907c8" class="index-entry"></a> <!-- complex_matrix; complex_schur_decompose_u; (complex_matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#complex_schur_decompose_u" class="unlink" title="Jump to index entry"><code>complex_schur_decompose_u</code></a></strong><code>(complex_matrix A)</code><br> Compute the unitary matrix of the complex Schur decomposition of <code>A</code>.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-9c45b8917b1de6f500415512b1d06896cd5f7b5c" class="index-entry"></a> <!-- tuple(complex_matrix, complex_matrix); complex_schur_decompose; (matrix A); -->
+<p></p>
+<p><code>tuple(complex_matrix, complex_matrix)</code> <strong><a href="../functions-reference/functions_index.html#complex_schur_decompose" class="unlink" title="Jump to index entry"><code>complex_schur_decompose</code></a></strong><code>(matrix A)</code><br> Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of <code>A</code>. This function is equivalent to <code>(complex_schur_decompose_u(A), complex_schur_decompose_t(A))</code> but with a lower computational cost due to the shared work between the two results. This overload is equivalent to <code>complex_schur_decompose(to_complex(A,0))</code> but is more efficient.</p>
+<small><i>Available since 2.33</i></small>
+<a id="index-entry-400ef877200b32483b4f5a7223c3231666d153b7" class="index-entry"></a> <!-- tuple(complex_matrix, complex_matrix); complex_schur_decompose; (complex_matrix A); -->
+<p></p>
+<p><code>tuple(complex_matrix, complex_matrix)</code> <strong><a href="../functions-reference/functions_index.html#complex_schur_decompose" class="unlink" title="Jump to index entry"><code>complex_schur_decompose</code></a></strong><code>(complex_matrix A)</code><br> Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of <code>A</code>. This function is equivalent to <code>(complex_schur_decompose_u(A), complex_schur_decompose_t(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+</section>
+</section>
+</section>
+<section id="reverse-functions-for-complex-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="reverse-functions-for-complex-matrices">Reverse functions for complex matrices</h2>
+<a id="index-entry-4967c86d60f1814cc9c1b26860a5b0a2d875cfa6" class="index-entry"></a> <!-- complex_vector; reverse; (complex_vector v); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#reverse" class="unlink" title="Jump to index entry"><code>reverse</code></a></strong><code>(complex_vector v)</code><br> Return a new vector containing the elements of the argument in reverse order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6518a921f55ed48810b7eddb6fdeebc850a60c27" class="index-entry"></a> <!-- complex_row_vector; reverse; (complex_row_vector v); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#reverse" class="unlink" title="Jump to index entry"><code>reverse</code></a></strong><code>(complex_row_vector v)</code><br> Return a new row vector containing the elements of the argument in reverse order.</p>
+<small><i>Available since 2.30</i></small>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/matrix_operations.html" class="pagination-link" aria-label="Matrix Operations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Matrix Operations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/sparse_matrix_operations.html" class="pagination-link" aria-label="Sparse Matrix Operations">
+        <span class="nav-page-text">Sparse Matrix Operations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/compound_arithmetic_and_assignment.html b/docs/2_39/functions-reference/compound_arithmetic_and_assignment.html
new file mode 100644
index 000000000..1b6ce2cfe
--- /dev/null
+++ b/docs/2_39/functions-reference/compound_arithmetic_and_assignment.html
@@ -0,0 +1,1177 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Compound Arithmetic and Assignment</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/higher-order_functions.html" rel="next">
+<link href="../functions-reference/mixed_operations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Compound Arithmetic and Assignment">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/compound_arithmetic_and_assignment.html">Compound Arithmetic and Assignment</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#compound-arithmetic-and-assignment" id="toc-compound-arithmetic-and-assignment" class="nav-link active" data-scroll-target="#compound-arithmetic-and-assignment">Compound Arithmetic and Assignment</a>
+  <ul class="collapse">
+  <li><a href="#compound-addition-and-assignment" id="toc-compound-addition-and-assignment" class="nav-link" data-scroll-target="#compound-addition-and-assignment">Compound addition and assignment</a></li>
+  <li><a href="#compound-subtraction-and-assignment" id="toc-compound-subtraction-and-assignment" class="nav-link" data-scroll-target="#compound-subtraction-and-assignment">Compound subtraction and assignment</a></li>
+  <li><a href="#compound-multiplication-and-assignment" id="toc-compound-multiplication-and-assignment" class="nav-link" data-scroll-target="#compound-multiplication-and-assignment">Compound multiplication and assignment</a></li>
+  <li><a href="#compound-division-and-assignment" id="toc-compound-division-and-assignment" class="nav-link" data-scroll-target="#compound-division-and-assignment">Compound division and assignment</a></li>
+  <li><a href="#compound-elementwise-multiplication-and-assignment" id="toc-compound-elementwise-multiplication-and-assignment" class="nav-link" data-scroll-target="#compound-elementwise-multiplication-and-assignment">Compound elementwise multiplication and assignment</a></li>
+  <li><a href="#compound-elementwise-division-and-assignment" id="toc-compound-elementwise-division-and-assignment" class="nav-link" data-scroll-target="#compound-elementwise-division-and-assignment">Compound elementwise division and assignment</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/compound_arithmetic_and_assignment.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/compound_arithmetic_and_assignment.html">Compound Arithmetic and Assignment</a></li></ol></nav></header>
+
+
+
+
+<section id="compound-arithmetic-and-assignment" class="level1">
+<h1>Compound Arithmetic and Assignment</h1>
+<p>Compound arithmetic and assignment statements combine an arithmetic operation and assignment, replacing a statement such as</p>
+<pre><code> x = x op y;</code></pre>
+<p>with the more compact compound form</p>
+<pre><code> x op= y;</code></pre>
+<p>For example, <code>x = x + 1;</code> may be replaced with <code>x += 1;</code>. This works for all types that support arithmetic, including the scalar types <code>int</code>, <code>real</code>, <code>complex</code>, the real matrix types <code>vector</code>, <code>row_vector</code>, and <code>matrix</code>, and the complex matrix types, <code>complex_vector</code>, <code>complex_row_vector</code>, and <code>complex_matrix</code>.</p>
+<section id="compound-addition-and-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="compound-addition-and-assignment">Compound addition and assignment</h2>
+<p>Compound addition and assignment works wherever the corresponding addition and assignment would be well formed.</p>
+<a id="index-entry-967b2a8e9bdbd6f8a1fc5a6deede7134f075a813" class="index-entry"></a> <!-- void; operator+=; (T x, U y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator+=" class="unlink" title="Jump to index entry"><code>operator+=</code></a></strong><code>(T x, U y)</code><br> <code>x += y</code> is equivalent to <code>x = x + y</code>. Defined for all types <code>T</code> and <code>U</code> where <code>T = T + U</code> is well formed.</p>
+<small><i>Available since 2.17, complex signatures added in 2.30</i></small>
+</section>
+<section id="compound-subtraction-and-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="compound-subtraction-and-assignment">Compound subtraction and assignment</h2>
+<p>Compound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.</p>
+<a id="index-entry-e3beea1794e1e8704ae6177a2392b78fccebd4d4" class="index-entry"></a> <!-- void; operator-=; (T x, U y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator-=" class="unlink" title="Jump to index entry"><code>operator-=</code></a></strong><code>(T x, U y)</code><br> <code>x -= y</code> is equivalent to <code>x = x - y</code>. Defined for all types <code>T</code> and <code>U</code> where <code>T = T - U</code> is well formed.</p>
+<small><i>Available since 2.17, complex signatures added in 2.30</i></small>
+</section>
+<section id="compound-multiplication-and-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="compound-multiplication-and-assignment">Compound multiplication and assignment</h2>
+<p>Compound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.</p>
+<a id="index-entry-518f24c806ec47af7c7aa48506a1cd57ef3676df" class="index-entry"></a> <!-- void; operator*=; (T x, U y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator\*=" class="unlink" title="Jump to index entry"><code>operator*=</code></a></strong><code>(T x, U y)</code><br> <code>x *= y</code> is equivalent to <code>x = x * y</code>. Defined for all types <code>T</code> and <code>U</code> where <code>T = T * U</code> is well formed.</p>
+<small><i>Available since 2.17, complex signatures added in 2.30</i></small>
+</section>
+<section id="compound-division-and-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="compound-division-and-assignment">Compound division and assignment</h2>
+<p>Compound division and assignment works wherever the corresponding division and assignment would be well formed.</p>
+<a id="index-entry-eee88771fc7180e557937b705c365f97c600b293" class="index-entry"></a> <!-- void; operator/=; (T x, U y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator/=" class="unlink" title="Jump to index entry"><code>operator/=</code></a></strong><code>(T x, U y)</code><br> <code>x /= y</code> is equivalent to <code>x = x / y</code>. Defined for all types <code>T</code> and <code>U</code> where <code>T = T / U</code> is well formed.</p>
+<small><i>Available since 2.17, complex signatures added in 2.30</i></small>
+</section>
+<section id="compound-elementwise-multiplication-and-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="compound-elementwise-multiplication-and-assignment">Compound elementwise multiplication and assignment</h2>
+<p>Compound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.</p>
+<a id="index-entry-549243a6af6a91c078444182753a8e8f435862e3" class="index-entry"></a> <!-- void; operator.*=; (T x, U y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator.\*=" class="unlink" title="Jump to index entry"><code>operator.*=</code></a></strong><code>(T x, U y)</code><br> <code>x .*= y</code> is equivalent to <code>x = x .* y</code>. Defined for all types <code>T</code> and <code>U</code> where <code>T = T .* U</code> is well formed.</p>
+<small><i>Available since 2.17, complex signatures added in 2.30</i></small>
+</section>
+<section id="compound-elementwise-division-and-assignment" class="level2">
+<h2 class="anchored" data-anchor-id="compound-elementwise-division-and-assignment">Compound elementwise division and assignment</h2>
+<p>Compound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.</p>
+<a id="index-entry-afa1bfd51ef434b4843fe93b63d0cd3807ac9c7b" class="index-entry"></a> <!-- void; operator./=; (T x, U y); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#operator./=" class="unlink" title="Jump to index entry"><code>operator./=</code></a></strong><code>(T x, U y)</code><br> <code>x ./= y</code> is equivalent to <code>x = x ./ y</code>. Defined for all types <code>T</code> and <code>U</code> where <code>T = T ./ U</code> is well formed.</p>
+<small><i>Available since 2.17, complex signatures added in 2.30</i></small>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/mixed_operations.html" class="pagination-link" aria-label="Mixed Operations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Mixed Operations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/higher-order_functions.html" class="pagination-link" aria-label="Higher-Order Functions">
+        <span class="nav-page-text">Higher-Order Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/compound_arithmetic_and_assignment.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/continuous_distributions_on_0_1.html b/docs/2_39/functions-reference/continuous_distributions_on_0_1.html
new file mode 100644
index 000000000..df8c1baeb
--- /dev/null
+++ b/docs/2_39/functions-reference/continuous_distributions_on_0_1.html
@@ -0,0 +1,1248 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Continuous Distributions on [0, 1]</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/circular_distributions.html" rel="next">
+<link href="../functions-reference/positive_lower-bounded_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Continuous Distributions on [0, 1]">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/continuous_distributions_on_0_1.html">Continuous Distributions on [0, 1]</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#continuous-distributions-on-0-1" id="toc-continuous-distributions-on-0-1" class="nav-link active" data-scroll-target="#continuous-distributions-on-0-1">Continuous Distributions on [0, 1]</a>
+  <ul class="collapse">
+  <li><a href="#beta-distribution" id="toc-beta-distribution" class="nav-link" data-scroll-target="#beta-distribution">Beta distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#beta-proportion-distribution" id="toc-beta-proportion-distribution" class="nav-link" data-scroll-target="#beta-proportion-distribution">Beta proportion distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/continuous_distributions_on_0_1.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/continuous_distributions_on_0_1.html">Continuous Distributions on [0, 1]</a></li></ol></nav></header>
+
+
+
+
+<section id="continuous-distributions-on-0-1" class="level1">
+<h1>Continuous Distributions on [0, 1]</h1>
+<p>The continuous distributions with outcomes in the interval <span class="math inline">\([0,1]\)</span> are used to characterized bounded quantities, including probabilities.</p>
+<section id="beta-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="beta-distribution">Beta distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span> and <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(\theta \in (0,1)\)</span>, <span class="math display">\[\begin{equation*} \text{Beta}(\theta|\alpha,\beta) =
+\frac{1}{\mathrm{B}(\alpha,\beta)} \, \theta^{\alpha - 1} \, (1 -
+\theta)^{\beta - 1} , \end{equation*}\]</span> where the beta function <span class="math inline">\(\mathrm{B}()\)</span> is as defined in section <a href="../functions-reference/real-valued_basic_functions.html#betafun">combinatorial functions</a>.</p>
+<p><em><strong>Warning:</strong></em> If <span class="math inline">\(\theta = 0\)</span> or <span class="math inline">\(\theta = 1\)</span>, then the probability is 0 and the log probability is <span class="math inline">\(-\infty\)</span>. Similarly, the distribution requires strictly positive parameters, <span class="math inline">\(\alpha, \beta &gt;
+0\)</span>.</p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>theta ~</code> <strong><a href="../functions-reference/functions_index.html#beta" class="unlink" title="Jump to index entry"><code>beta</code></a></strong><code>(alpha, beta)</code></p>
+<p>Increment target log probability density with <code>beta_lupdf(theta | alpha, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-ed6b0e2a1365933147e892aab802b3da10b2bb6d" class="index-entry"></a> <!-- real; beta ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-93d2282b7bada6cf97c5977413fb03d509c396ea" class="index-entry"></a> <!-- real; beta_lpdf; (reals theta | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_lpdf" class="unlink" title="Jump to index entry"><code>beta_lpdf</code></a></strong><code>(reals theta | reals alpha, reals beta)</code><br> The log of the beta density of <code>theta</code> in <span class="math inline">\([0,1]\)</span> given positive prior successes (plus one) alpha and prior failures (plus one) beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-340181b236cc79820c87cfca7dc7e8d9ef5bdf6e" class="index-entry"></a> <!-- real; beta_lupdf; (reals theta | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_lupdf" class="unlink" title="Jump to index entry"><code>beta_lupdf</code></a></strong><code>(reals theta | reals alpha, reals beta)</code><br> The log of the beta density of <code>theta</code> in <span class="math inline">\([0,1]\)</span> given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-096320ad72e962315021c5f04c98350b8ff52744" class="index-entry"></a> <!-- real; beta_cdf; (reals theta | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_cdf" class="unlink" title="Jump to index entry"><code>beta_cdf</code></a></strong><code>(reals theta | reals alpha, reals beta)</code><br> The beta cumulative distribution function of <code>theta</code> in <span class="math inline">\([0,1]\)</span> given positive prior successes (plus one) alpha and prior failures (plus one) beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a3067ae5b48a826e8ad6971507a45b59a9e242ed" class="index-entry"></a> <!-- real; beta_lcdf; (reals theta | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_lcdf" class="unlink" title="Jump to index entry"><code>beta_lcdf</code></a></strong><code>(reals theta | reals alpha, reals beta)</code><br> The log of the beta cumulative distribution function of <code>theta</code> in <span class="math inline">\([0,1]\)</span> given positive prior successes (plus one) alpha and prior failures (plus one) beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-70ce7e460a07bc188c0e0eb7ab6c0ed1d212d523" class="index-entry"></a> <!-- real; beta_lccdf; (reals theta | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_lccdf" class="unlink" title="Jump to index entry"><code>beta_lccdf</code></a></strong><code>(reals theta | reals alpha, reals beta)</code><br> The log of the beta complementary cumulative distribution function of <code>theta</code> in <span class="math inline">\([0,1]\)</span> given positive prior successes (plus one) alpha and prior failures (plus one) beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-447854f88734efbef65bc89140a035fe0f16cc74" class="index-entry"></a> <!-- R; beta_rng; (reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#beta_rng" class="unlink" title="Jump to index entry"><code>beta_rng</code></a></strong><code>(reals alpha, reals beta)</code><br> Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="beta-proportion-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="beta-proportion-distribution">Beta proportion distribution</h2>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in (0, 1)\)</span> and <span class="math inline">\(\kappa \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(\theta
+\in (0,1)\)</span>, <span class="math display">\[\begin{equation*} \mathrm{Beta\_Proportion}(\theta|\mu,\kappa) =
+\frac{1}{\mathrm{B}(\mu \kappa, (1 - \mu) \kappa)} \,
+\theta^{\mu\kappa - 1} \, (1 - \theta)^{(1 - \mu)\kappa- 1} , \end{equation*}\]</span> where the beta function <span class="math inline">\(\mathrm{B}()\)</span> is as defined in section <a href="../functions-reference/real-valued_basic_functions.html#betafun">combinatorial functions</a>.</p>
+<p><em><strong>Warning:</strong></em> If <span class="math inline">\(\theta = 0\)</span> or <span class="math inline">\(\theta = 1\)</span>, then the probability is 0 and the log probability is <span class="math inline">\(-\infty\)</span>. Similarly, the distribution requires <span class="math inline">\(\mu \in (0, 1)\)</span> and strictly positive parameter, <span class="math inline">\(\kappa &gt; 0\)</span>.</p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>theta ~</code> <strong><a href="../functions-reference/functions_index.html#beta_proportion" class="unlink" title="Jump to index entry"><code>beta_proportion</code></a></strong><code>(mu, kappa)</code></p>
+<p>Increment target log probability density with <code>beta_proportion_lupdf(theta | mu, kappa)</code>.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-9e3ba228a8bb07d871e13efa26340614b125cd7f" class="index-entry"></a> <!-- real; beta_proportion ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-508527134a33175b28ae22d5ff494a10fb777b4e" class="index-entry"></a> <!-- real; beta_proportion_lpdf; (reals theta | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_proportion_lpdf" class="unlink" title="Jump to index entry"><code>beta_proportion_lpdf</code></a></strong><code>(reals theta | reals mu, reals kappa)</code><br> The log of the beta_proportion density of <code>theta</code> in <span class="math inline">\((0,1)\)</span> given mean mu and precision kappa</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-813c2f603806061c22e0d21c42f757aa47c7a3e7" class="index-entry"></a> <!-- real; beta_proportion_lupdf; (reals theta | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_proportion_lupdf" class="unlink" title="Jump to index entry"><code>beta_proportion_lupdf</code></a></strong><code>(reals theta | reals mu, reals kappa)</code><br> The log of the beta_proportion density of <code>theta</code> in <span class="math inline">\((0,1)\)</span> given mean mu and precision kappa dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-61a129c59e7facb5ab538ca66fb2f4f1750bf7e1" class="index-entry"></a> <!-- real; beta_proportion_lcdf; (reals theta | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_proportion_lcdf" class="unlink" title="Jump to index entry"><code>beta_proportion_lcdf</code></a></strong><code>(reals theta | reals mu, reals kappa)</code><br> The log of the beta_proportion cumulative distribution function of <code>theta</code> in <span class="math inline">\((0,1)\)</span> given mean mu and precision kappa</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-8c5423cb82f066b9e55e6fdeefcbb2355698308f" class="index-entry"></a> <!-- real; beta_proportion_lccdf; (reals theta | reals mu, reals kappa); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_proportion_lccdf" class="unlink" title="Jump to index entry"><code>beta_proportion_lccdf</code></a></strong><code>(reals theta | reals mu, reals kappa)</code><br> The log of the beta_proportion complementary cumulative distribution function of <code>theta</code> in <span class="math inline">\((0,1)\)</span> given mean mu and precision kappa</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-55d15275007130c8e57a41bf24370311b3581959" class="index-entry"></a> <!-- R; beta_proportion_rng; (reals mu, reals kappa); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#beta_proportion_rng" class="unlink" title="Jump to index entry"><code>beta_proportion_rng</code></a></strong><code>(reals mu, reals kappa)</code><br> Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/positive_lower-bounded_distributions.html" class="pagination-link" aria-label="Positive Lower-Bounded Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Positive Lower-Bounded Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/circular_distributions.html" class="pagination-link" aria-label="Circular Distributions">
+        <span class="nav-page-text">Circular Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/continuous_distributions_on_0_1.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/conventions_for_probability_functions.html b/docs/2_39/functions-reference/conventions_for_probability_functions.html
new file mode 100644
index 000000000..892c0238d
--- /dev/null
+++ b/docs/2_39/functions-reference/conventions_for_probability_functions.html
@@ -0,0 +1,1379 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Conventions for Probability Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/binary_distributions.html" rel="next">
+<link href="../functions-reference/removed_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Conventions for Probability Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/conventions_for_probability_functions.html">Conventions for Probability Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#conventions-for-probability-functions" id="toc-conventions-for-probability-functions" class="nav-link active" data-scroll-target="#conventions-for-probability-functions">Conventions for Probability Functions</a>
+  <ul class="collapse">
+  <li><a href="#suffix-marks-type-of-function" id="toc-suffix-marks-type-of-function" class="nav-link" data-scroll-target="#suffix-marks-type-of-function">Suffix marks type of function</a></li>
+  <li><a href="#argument-order-and-the-vertical-bar" id="toc-argument-order-and-the-vertical-bar" class="nav-link" data-scroll-target="#argument-order-and-the-vertical-bar">Argument order and the vertical bar</a></li>
+  <li><a href="#sampling-notation" id="toc-sampling-notation" class="nav-link" data-scroll-target="#sampling-notation">Sampling notation</a></li>
+  <li><a href="#finite-inputs" id="toc-finite-inputs" class="nav-link" data-scroll-target="#finite-inputs">Finite inputs</a></li>
+  <li><a href="#boundary-conditions" id="toc-boundary-conditions" class="nav-link" data-scroll-target="#boundary-conditions">Boundary conditions</a></li>
+  <li><a href="#distributions-prng" id="toc-distributions-prng" class="nav-link" data-scroll-target="#distributions-prng">Pseudorandom number generators</a>
+  <ul class="collapse">
+  <li><a href="#restricted-to-transformed-data-and-generated-quantities" id="toc-restricted-to-transformed-data-and-generated-quantities" class="nav-link" data-scroll-target="#restricted-to-transformed-data-and-generated-quantities">Restricted to transformed data and generated quantities</a></li>
+  <li><a href="#limited-vectorization" id="toc-limited-vectorization" class="nav-link" data-scroll-target="#limited-vectorization">Limited vectorization</a></li>
+  </ul></li>
+  <li><a href="#cumulative-distribution-functions" id="toc-cumulative-distribution-functions" class="nav-link" data-scroll-target="#cumulative-distribution-functions">Cumulative distribution functions</a></li>
+  <li><a href="#vectorization" id="toc-vectorization" class="nav-link" data-scroll-target="#vectorization">Vectorization</a>
+  <ul class="collapse">
+  <li><a href="#prob-vectorization" id="toc-prob-vectorization" class="nav-link" data-scroll-target="#prob-vectorization">Vectorized function signatures</a></li>
+  <li><a href="#evaluating-vectorized-log-probability-functions" id="toc-evaluating-vectorized-log-probability-functions" class="nav-link" data-scroll-target="#evaluating-vectorized-log-probability-functions">Evaluating vectorized log probability functions</a></li>
+  <li><a href="#prng-vectorization" id="toc-prng-vectorization" class="nav-link" data-scroll-target="#prng-vectorization">Evaluating vectorized PRNG functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/conventions_for_probability_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/conventions_for_probability_functions.html">Conventions for Probability Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="conventions-for-probability-functions" class="level1">
+<h1>Conventions for Probability Functions</h1>
+<p>Functions associated with distributions are set up to follow the same naming conventions for both built-in distributions and for user-defined distributions.</p>
+<section id="suffix-marks-type-of-function" class="level2">
+<h2 class="anchored" data-anchor-id="suffix-marks-type-of-function">Suffix marks type of function</h2>
+<p>The suffix is determined by the type of function according to the following table.</p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 73%">
+<col style="width: 14%">
+<col style="width: 11%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: left;">function</th>
+<th style="text-align: left;">outcome</th>
+<th style="text-align: left;">suffix</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;">log probability mass function</td>
+<td style="text-align: left;">discrete</td>
+<td style="text-align: left;"><code>_lpmf</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">log probability density function</td>
+<td style="text-align: left;">continuous</td>
+<td style="text-align: left;"><code>_lpdf</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">log cumulative distribution function</td>
+<td style="text-align: left;">any</td>
+<td style="text-align: left;"><code>_lcdf</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">log complementary cumulative distribution function</td>
+<td style="text-align: left;">any</td>
+<td style="text-align: left;"><code>_lccdf</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">random number generator</td>
+<td style="text-align: left;">any</td>
+<td style="text-align: left;"><code>_rng</code></td>
+</tr>
+</tbody>
+</table>
+<p>For example, <code>normal_lpdf</code> is the log of the normal probability density function (pdf) and <code>bernoulli_lpmf</code> is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, <code>normal_lcdf</code> and <code>bernoulli_lcdf</code>.</p>
+</section>
+<section id="argument-order-and-the-vertical-bar" class="level2">
+<h2 class="anchored" data-anchor-id="argument-order-and-the-vertical-bar">Argument order and the vertical bar</h2>
+<p>Each probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, <code>normal_lpdf(y | mu, sigma)</code> returns the value of mathematical formula <span class="math inline">\(\log \text{Normal}(y \, | \, \mu, \sigma)\)</span>. Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., <code>normal_lcdf(y_low | mu, sigma)</code></p>
+</section>
+<section id="sampling-notation" class="level2">
+<h2 class="anchored" data-anchor-id="sampling-notation">Sampling notation</h2>
+<p>The notation</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>provides the same (proportional) contribution to the model log density as the explicit target density increment,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a> <span class="kw">target +=</span> normal_lpdf(y | mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (<code>~</code>) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.</p>
+</section>
+<section id="finite-inputs" class="level2">
+<h2 class="anchored" data-anchor-id="finite-inputs">Finite inputs</h2>
+<p>All of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.</p>
+</section>
+<section id="boundary-conditions" class="level2">
+<h2 class="anchored" data-anchor-id="boundary-conditions">Boundary conditions</h2>
+<p>Many distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter <span class="math inline">\(\sigma &gt; 0\)</span>. If <span class="math inline">\(\sigma = 0\)</span>, the probability function will throw an exception.</p>
+<p>This is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.</p>
+</section>
+<section id="distributions-prng" class="level2">
+<h2 class="anchored" data-anchor-id="distributions-prng">Pseudorandom number generators</h2>
+<p>For most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix <code>_rng</code>. For example, the function <code>normal_rng(real, real)</code> accepts two real arguments, an unconstrained location <span class="math inline">\(\mu\)</span> and positive scale <span class="math inline">\(\sigma &gt; 0\)</span>, and returns an unconstrained pseudorandom value drawn from <span class="math inline">\(\text{Normal}(\mu,\sigma)\)</span>. There are also vectorized forms of random number generators which return more than one random variate at a time.</p>
+<section id="restricted-to-transformed-data-and-generated-quantities" class="level3">
+<h3 class="anchored" data-anchor-id="restricted-to-transformed-data-and-generated-quantities">Restricted to transformed data and generated quantities</h3>
+<p>Unlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.</p>
+</section>
+<section id="limited-vectorization" class="level3">
+<h3 class="anchored" data-anchor-id="limited-vectorization">Limited vectorization</h3>
+<p>Unlike the probability functions, only some of the PRNG functions are vectorized.</p>
+</section>
+</section>
+<section id="cumulative-distribution-functions" class="level2">
+<h2 class="anchored" data-anchor-id="cumulative-distribution-functions">Cumulative distribution functions</h2>
+<p>For most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.</p>
+<p>For a univariate random variable <span class="math inline">\(Y\)</span> with probability function <span class="math inline">\(p_Y(y \, | \, \theta)\)</span>, the cumulative distribution function (CDF) <span class="math inline">\(F_Y\)</span> is defined by <span class="math display">\[\begin{equation*}
+F_Y(y) \ = \ \text{Pr}[Y \le y] \ = \ \int_{-\infty}^y p(y\, | \, \theta) \ \text{d}y.
+\end{equation*}\]</span> The complementary cumulative distribution function (CCDF) is defined as <span class="math display">\[\begin{equation*}
+\text{Pr}[Y &gt; y] \ = \ 1 - F_Y(y).
+\end{equation*}\]</span> The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly <span class="math inline">\(10^{-300}\)</span>), but not numbers very close to 1 (the closest you can get is roughly <span class="math inline">\(1 - 10^{-15}\)</span>).</p>
+<p>In Stan, there is a cumulative distribution function for each probability function. For instance, <code>normal_cdf(y | mu, sigma)</code> is defined by <span class="math display">\[\begin{equation*}
+\int_{-\infty}^y \text{Normal}(y \, | \, \mu, \sigma) \ \text{d}y.
+\end{equation*}\]</span> There are also log forms of the CDF and CCDF for most univariate distributions. For example, <code>normal_lcdf(y | mu, sigma)</code> is defined by <span class="math display">\[\begin{equation*}
+\log \left( \int_{-\infty}^y \text{Normal}(y \, | \, \mu, \sigma) \   \text{d}y \right)
+\end{equation*}\]</span> and <code>normal_lccdf(y | mu, sigma)</code> is defined by <span class="math display">\[\begin{equation*}
+\log \left( 1 - \int_{-\infty}^y \text{Normal}(y \, | \, \mu, \sigma) \   \text{d}y \right).
+\end{equation*}\]</span></p>
+</section>
+<section id="vectorization" class="level2">
+<h2 class="anchored" data-anchor-id="vectorization">Vectorization</h2>
+<p>Stan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a> for more details.</p>
+<p>In all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.</p>
+<p>Stan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.</p>
+<p>Stan also overloads some scalar functions, such as <code>log</code> and <code>exp</code>, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.</p>
+<section id="prob-vectorization" class="level3">
+<h3 class="anchored" data-anchor-id="prob-vectorization">Vectorized function signatures</h3>
+<section id="vectorized-scalar-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="vectorized-scalar-arguments">Vectorized scalar arguments</h4>
+<p>The normal probability function is specified with the signature</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a> normal_lpdf(reals | reals, reals);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The pseudotype <code>reals</code> is used to indicate that an argument position may be vectorized. Argument positions declared as <code>reals</code> may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use <code>normal_lpdf(row_vector | vector, real)</code> as long as the vector and row vector have the same size.</p>
+</section>
+<section id="vectorized-vector-and-row-vector-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="vectorized-vector-and-row-vector-arguments">Vectorized vector and row vector arguments</h4>
+<p>The multivariate normal distribution accepting vector or array of vector arguments is written as</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a> multi_normal_lpdf(vectors | vectors, <span class="dt">matrix</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.</p>
+</section>
+<section id="vectorized-integer-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="vectorized-integer-arguments">Vectorized integer arguments</h4>
+<p>The pseudotype <code>ints</code> is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.</p>
+</section>
+</section>
+<section id="evaluating-vectorized-log-probability-functions" class="level3">
+<h3 class="anchored" data-anchor-id="evaluating-vectorized-log-probability-functions">Evaluating vectorized log probability functions</h3>
+<p>The result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely <code>real</code> or <code>int</code>, is repeated. For instance, if <code>y</code> is a vector of size <code>N</code>, <code>mu</code> is a vector of size <code>N</code>, and <code>sigma</code> is a scalar, then</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a> ll = normal_lpdf(y | mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is just a more efficient way to write</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a> ll = <span class="dv">0</span>;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>   ll = ll + normal_lpdf(y[n] | mu[n], sigma);</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a> }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With the same arguments, the vectorized sampling statement</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a> y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>has the same effect on the total log probability as</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>   y[n] ~ normal(mu[n], sigma);</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a> }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="prng-vectorization" class="level3">
+<h3 class="anchored" data-anchor-id="prng-vectorization">Evaluating vectorized PRNG functions</h3>
+<p>Some PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes <code>reals</code> or <code>ints</code>. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">3</span>] mu = <span class="co">// ...</span></span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span> x = normal_rng(mu, <span class="dv">3</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="argument-types" class="level4">
+<h4 class="anchored" data-anchor-id="argument-types">Argument types</h4>
+<p>In the case of PRNG functions, arguments marked <code>ints</code> may be integers or integer arrays, whereas arguments marked <code>reals</code> may be integers or reals, integer or real arrays, vectors, or row vectors.</p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 12%">
+<col style="width: 87%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: left;">pseudotype</th>
+<th style="text-align: left;">allowable PRNG arguments</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><code>ints</code></td>
+<td style="text-align: left;"><code>int</code>, <code>array[] int</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>reals</code></td>
+<td style="text-align: left;"><code>int</code>, <code>array[] int</code>, <code>real</code>, <code>array[] real</code>, <code>vector</code>, <code>row_vector</code></td>
+</tr>
+</tbody>
+</table>
+</section>
+<section id="dimension-matching" class="level4">
+<h4 class="anchored" data-anchor-id="dimension-matching">Dimension matching</h4>
+<p>In general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the <code>normal_rng</code> function may be called with one vector argument and one real array argument as long as they have the same number of elements.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">3</span>] mu = <span class="co">// ...</span></span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span> sigma = <span class="co">// ...</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span> x = normal_rng(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="return-type" class="level4">
+<h4 class="anchored" data-anchor-id="return-type">Return type</h4>
+<p>The result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (<code>array[] real</code>) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return <code>ints</code> and continuous distributions return <code>reals</code>, each of appropriate size. The symbol <code>R</code> denotes such a return type.</p>
+
+
+</section>
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/removed_functions.html" class="pagination-link" aria-label="Removed Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Removed Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/binary_distributions.html" class="pagination-link" aria-label="Binary Distributions">
+        <span class="nav-page-text">Binary Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/conventions_for_probability_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/correlation_matrix_distributions.html b/docs/2_39/functions-reference/correlation_matrix_distributions.html
new file mode 100644
index 000000000..2d9c7541a
--- /dev/null
+++ b/docs/2_39/functions-reference/correlation_matrix_distributions.html
@@ -0,0 +1,1308 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Correlation Matrix Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/covariance_matrix_distributions.html" rel="next">
+<link href="../functions-reference/simplex_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Correlation Matrix Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/correlation_matrix_distributions.html">Correlation Matrix Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#correlation-matrix-distributions" id="toc-correlation-matrix-distributions" class="nav-link active" data-scroll-target="#correlation-matrix-distributions">Correlation Matrix Distributions</a>
+  <ul class="collapse">
+  <li><a href="#lkj-correlation" id="toc-lkj-correlation" class="nav-link" data-scroll-target="#lkj-correlation">LKJ correlation distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#cholesky-lkj-correlation-distribution" id="toc-cholesky-lkj-correlation-distribution" class="nav-link" data-scroll-target="#cholesky-lkj-correlation-distribution">Cholesky LKJ correlation distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/correlation_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/correlation_matrix_distributions.html">Correlation Matrix Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="correlation-matrix-distributions" class="level1">
+<h1>Correlation Matrix Distributions</h1>
+<p>The correlation matrix distributions have support on the (Cholesky factors of) correlation matrices. A Cholesky factor <span class="math inline">\(L\)</span> for a <span class="math inline">\(K
+\times K\)</span> correlation matrix <span class="math inline">\(\Sigma\)</span> of dimension <span class="math inline">\(K\)</span> has rows of unit length so that the diagonal of <span class="math inline">\(L L^{\top}\)</span> is the unit <span class="math inline">\(K\)</span>-vector. Even though models are usually conceptualized in terms of correlation matrices, it is better to operationalize them in terms of their Cholesky factors. If you are interested in the posterior distribution of the correlations, you can recover them in the generated quantities block via</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> <span class="kw">generated quantities</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>   <span class="dt">corr_matrix</span>[K] Sigma;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>   Sigma = multiply_lower_tri_self_transpose(L);</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a> }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="lkj-correlation" class="level2">
+<h2 class="anchored" data-anchor-id="lkj-correlation">LKJ correlation distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>For <span class="math inline">\(\eta &gt; 0\)</span>, if <span class="math inline">\(\Sigma\)</span> a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then <span class="math display">\[\begin{equation*}
+\text{LkjCorr}(\Sigma|\eta) \propto \det \left( \Sigma \right)^{(\eta
+- 1)}. \end{equation*}\]</span> The expectation is the identity matrix for any positive value of the shape parameter <span class="math inline">\(\eta\)</span>, which can be interpreted like the shape parameter of a symmetric beta distribution:</p>
+<ul>
+<li><p>if <span class="math inline">\(\eta = 1\)</span>, then the density is uniform over correlation matrices of order <span class="math inline">\(K\)</span>;</p></li>
+<li><p>if <span class="math inline">\(\eta &gt; 1\)</span>, the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger <span class="math inline">\(\eta\)</span>; and</p></li>
+<li><p>for <span class="math inline">\(0 &lt; \eta &lt; 1\)</span>, the density has a trough at the identity matrix.</p></li>
+<li><p>if <span class="math inline">\(\eta\)</span> were an unknown parameter, the Jeffreys prior is proportional to <span class="math inline">\(\sqrt{2\sum_{k=1}^{K-1}\left(
+\psi_1\left(\eta+\frac{K-k-1}{2}\right) -   2\psi_1\left(2\eta+K-k-1
+\right)\right)}\)</span>, where <span class="math inline">\(\psi_1()\)</span> is the trigamma function</p></li>
+</ul>
+<p>See <span class="citation" data-cites="LewandowskiKurowickaJoe:2009">(<a href="#ref-LewandowskiKurowickaJoe:2009" role="doc-biblioref">Lewandowski, Kurowicka, and Joe 2009</a>)</span> for definitions. However, it is much better computationally to work directly with the Cholesky factor of <span class="math inline">\(\Sigma\)</span>, so this distribution should never be explicitly used in practice.</p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr" class="unlink" title="Jump to index entry"><code>lkj_corr</code></a></strong><code>(eta)</code></p>
+<p>Increment target log probability density with <code>lkj_corr_lupdf(y | eta)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-beb0c10c5b7b26c3295ff6f47561e58ffc211a45" class="index-entry"></a> <!-- real; lkj_corr ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-34f4437ba5f424e835889403cf07a8d4f63921f2" class="index-entry"></a> <!-- real; lkj_corr_lpdf; (matrix y | real eta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_lpdf" class="unlink" title="Jump to index entry"><code>lkj_corr_lpdf</code></a></strong><code>(matrix y | real eta)</code><br> The log of the LKJ density for the correlation matrix y given nonnegative shape eta. <code>lkj_corr_cholesky_lpdf</code> is faster, more numerically stable, uses less memory, and should be preferred to this.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-61643c9cb87b1ed68a761b62034abf5253ca8f38" class="index-entry"></a> <!-- real; lkj_corr_lupdf; (matrix y | real eta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_lupdf" class="unlink" title="Jump to index entry"><code>lkj_corr_lupdf</code></a></strong><code>(matrix y | real eta)</code><br> The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. <code>lkj_corr_cholesky_lupdf</code> is faster, more numerically stable, uses less memory, and should be preferred to this.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-6af330a1989d2c53be55e7707c77a9e616e1eba2" class="index-entry"></a> <!-- matrix; lkj_corr_rng; (int K, real eta); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_rng" class="unlink" title="Jump to index entry"><code>lkj_corr_rng</code></a></strong><code>(int K, real eta)</code><br> Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="cholesky-lkj-correlation-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="cholesky-lkj-correlation-distribution">Cholesky LKJ correlation distribution</h2>
+<p>Stan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if <code>L</code> is a Cholesky factor of a correlation matrix, then</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a> L ~ lkj_corr_cholesky(<span class="fl">2.0</span>); <span class="co"># implies L * L' ~ lkj_corr(2.0);</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because Stan requires models to have support on all valid constrained parameters, <code>L</code> will almost always<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a> <span class="kw">parameters</span> {   <span class="dt">cholesky_factor_corr</span>[K] L;   <span class="co"># rather than corr_matrix[K] Sigma;   // ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>For <span class="math inline">\(\eta &gt; 0\)</span>, if <span class="math inline">\(L\)</span> is a <span class="math inline">\(K \times K\)</span> lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then <span class="math display">\[\begin{equation*} \text{LkjCholesky}(L|\eta)
+\propto \left|J\right|\det(L L^\top)^{(\eta - 1)} = \prod_{k=2}^K
+L_{kk}^{K-k+2\eta-2}. \end{equation*}\]</span> See the previous section for details on interpreting the shape parameter <span class="math inline">\(\eta\)</span>. Note that even if <span class="math inline">\(\eta=1\)</span>, it is still essential to evaluate the density function because the density of <span class="math inline">\(L\)</span> is not constant, regardless of the value of <span class="math inline">\(\eta\)</span>, even though the density of <span class="math inline">\(LL^\top\)</span> is constant iff <span class="math inline">\(\eta=1\)</span>.</p>
+<p>A lower triangular <span class="math inline">\(L\)</span> is a Cholesky factor for a correlation matrix if and only if <span class="math inline">\(L_{k,k} &gt; 0\)</span> for <span class="math inline">\(k \in 1{:}K\)</span> and each row <span class="math inline">\(L_k\)</span> has unit Euclidean length.</p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>L ~</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_cholesky" class="unlink" title="Jump to index entry"><code>lkj_corr_cholesky</code></a></strong><code>(eta)</code></p>
+<p>Increment target log probability density with <code>lkj_corr_cholesky_lupdf(L | eta)</code>.</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-d3519d9e1f129c4f006c23bcd5995f869bcfb56d" class="index-entry"></a> <!-- real; lkj_corr_cholesky ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-03f62c1ae419e1804760f00f477c89ad5911c97a" class="index-entry"></a> <!-- real; lkj_corr_cholesky_lpdf; (matrix L | real eta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>lkj_corr_cholesky_lpdf</code></a></strong><code>(matrix L | real eta)</code><br> The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-d612c55f03ba91b1722cbfc3ccd917a5fd209b0b" class="index-entry"></a> <!-- real; lkj_corr_cholesky_lupdf; (matrix L | real eta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>lkj_corr_cholesky_lupdf</code></a></strong><code>(matrix L | real eta)</code><br> The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8efcbe3b1b5dfa560b74b1a1642f6f855f32c26f" class="index-entry"></a> <!-- matrix; lkj_corr_cholesky_rng; (int K, real eta); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#lkj_corr_cholesky_rng" class="unlink" title="Jump to index entry"><code>lkj_corr_cholesky_rng</code></a></strong><code>(int K, real eta)</code><br> Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.4</i></small>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-LewandowskiKurowickaJoe:2009" class="csl-entry" role="listitem">
+Lewandowski, Daniel, Dorota Kurowicka, and Harry Joe. 2009. <span>“Generating Random Correlation Matrices Based on Vines and Extended Onion Method.”</span> <em>Journal of Multivariate Analysis</em> 100: 1989–2001.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>It is possible to build up a valid <code>L</code> within Stan, but that would then require Jacobian adjustments to imply the intended posterior.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/simplex_distributions.html" class="pagination-link" aria-label="Simplex Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Simplex Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/covariance_matrix_distributions.html" class="pagination-link" aria-label="Covariance Matrix Distributions">
+        <span class="nav-page-text">Covariance Matrix Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/correlation_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/covariance_matrix_distributions.html b/docs/2_39/functions-reference/covariance_matrix_distributions.html
new file mode 100644
index 000000000..26470d87e
--- /dev/null
+++ b/docs/2_39/functions-reference/covariance_matrix_distributions.html
@@ -0,0 +1,1312 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Covariance Matrix Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/hidden_markov_models.html" rel="next">
+<link href="../functions-reference/correlation_matrix_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Covariance Matrix Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/covariance_matrix_distributions.html">Covariance Matrix Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#covariance-matrix-distributions" id="toc-covariance-matrix-distributions" class="nav-link active" data-scroll-target="#covariance-matrix-distributions">Covariance Matrix Distributions</a>
+  <ul class="collapse">
+  <li><a href="#wishart-distribution" id="toc-wishart-distribution" class="nav-link" data-scroll-target="#wishart-distribution">Wishart distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#wishart-cholesky-distribution" id="toc-wishart-cholesky-distribution" class="nav-link" data-scroll-target="#wishart-cholesky-distribution">Wishart distribution, Cholesky Parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#inverse-wishart-distribution" id="toc-inverse-wishart-distribution" class="nav-link" data-scroll-target="#inverse-wishart-distribution">Inverse Wishart distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-2" id="toc-probability-density-function-2" class="nav-link" data-scroll-target="#probability-density-function-2">Probability density function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#inverse-wishart-cholesky-distribution" id="toc-inverse-wishart-cholesky-distribution" class="nav-link" data-scroll-target="#inverse-wishart-cholesky-distribution">Inverse Wishart distribution, Cholesky Parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-3" id="toc-probability-density-function-3" class="nav-link" data-scroll-target="#probability-density-function-3">Probability density function</a></li>
+  <li><a href="#stan-functions-3" id="toc-stan-functions-3" class="nav-link" data-scroll-target="#stan-functions-3">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/covariance_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/covariance_matrix_distributions.html">Covariance Matrix Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="covariance-matrix-distributions" class="level1">
+<h1>Covariance Matrix Distributions</h1>
+<p>The covariance matrix distributions have support on symmetric, positive-definite <span class="math inline">\(K \times K\)</span> matrices or their Cholesky factors (square, lower triangular matrices with positive diagonal elements).</p>
+<section id="wishart-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="wishart-distribution">Wishart distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\nu \in (K-1,\infty)\)</span>, and <span class="math inline">\(S \in
+\mathbb{R}^{K \times K}\)</span> is symmetric and positive definite, then for symmetric and positive-definite <span class="math inline">\(W \in \mathbb{R}^{K \times K}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Wishart}(W \mid \nu,S) = \frac{1}{2^{\nu K / 2}}
+\ \frac{1}{\Gamma_K \! \left( \frac{\nu}{2} \right)}
+\ \left| S \right|^{-\nu/2} \ \left| W \right|^{(\nu - K - 1)/2}
+\ \exp \! \left(- \frac{1}{2} \ \text{tr}\left( S^{-1} W \right) \right) \! ,
+\end{equation*}\]</span> where <span class="math inline">\(\text{tr}()\)</span> is the matrix trace function, and <span class="math inline">\(\Gamma_K()\)</span> is the multivariate Gamma function, <span class="math display">\[\begin{equation*}
+\Gamma_K(x) = \frac{1}{\pi^{K(K-1)/4}} \ \prod_{k=1}^K \Gamma \left( x + \frac{1 - k}{2} \right) \!.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>W ~</code> <strong><a href="../functions-reference/functions_index.html#wishart" class="unlink" title="Jump to index entry"><code>wishart</code></a></strong><code>(nu, Sigma)</code></p>
+<p>Increment target log probability density with <code>wishart_lupdf(W | nu, Sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-5f58cfc02fd2621376aab535950c5b180adee3b0" class="index-entry"></a> <!-- real; wishart ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-9ebed125213e04734c320d50523b729c926de41e" class="index-entry"></a> <!-- real; wishart_lpdf; (matrix W | real nu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wishart_lpdf" class="unlink" title="Jump to index entry"><code>wishart_lpdf</code></a></strong><code>(matrix W | real nu, matrix Sigma)</code><br> Return the log of the Wishart density for symmetric and positive-definite matrix <code>W</code> given degrees of freedom <code>nu</code> and symmetric and positive-definite scale matrix <code>Sigma</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-96892afabfd06300ee33a9c56e7aa11ef798d28c" class="index-entry"></a> <!-- real; wishart_lupdf; (matrix W | real nu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wishart_lupdf" class="unlink" title="Jump to index entry"><code>wishart_lupdf</code></a></strong><code>(matrix W | real nu, matrix Sigma)</code><br> Return the log of the Wishart density for symmetric and positive-definite matrix <code>W</code> given degrees of freedom <code>nu</code> and symmetric and positive-definite scale matrix <code>Sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-83370c265d9c1947eb72bc6d0aeab2570102b443" class="index-entry"></a> <!-- matrix; wishart_rng; (real nu, matrix Sigma); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#wishart_rng" class="unlink" title="Jump to index entry"><code>wishart_rng</code></a></strong><code>(real nu, matrix Sigma)</code><br> Generate a Wishart variate with degrees of freedom <code>nu</code> and symmetric and positive-definite scale matrix <code>Sigma</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="wishart-cholesky-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="wishart-cholesky-distribution">Wishart distribution, Cholesky Parameterization</h2>
+<p>The Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If <span class="math inline">\(S\)</span> and <span class="math inline">\(W\)</span> are positive definite matrices with Cholesky factors <span class="math inline">\(L_S\)</span> and <span class="math inline">\(L_W\)</span> (i.e., <span class="math inline">\(S = L_S L_S^{\top}\)</span> and <span class="math inline">\(W = L_W L_W^{\top}\)</span>), then the Cholesky parameterization is defined so that <span class="math display">\[\begin{equation*}
+L_W \sim \textrm{WishartCholesky}(\nu, L_S)
+\end{equation*}\]</span> if and only if <span class="math display">\[\begin{equation*}
+W \sim \textrm{Wishart}(\nu, S).
+\end{equation*}\]</span></p>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\nu \in (K-1, \infty)\)</span>, and <span class="math inline">\(L_S, L_W \in \mathbb{R}^{K \times K}\)</span> are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is <span class="math display">\[\begin{equation*}
+\text{WishartCholesky}(L_W \mid \nu,L_S)
+= \text{Wishart}(L_W L_W^{\top} \mid \nu,L_S L_S^{\top}) \, \left| J_{f^{-1}} \right|,
+\end{equation*}\]</span> where <span class="math inline">\(J_{f^{-1}}\)</span> is the Jacobian of the (inverse) transform of the variate, <span class="math inline">\(f^{-1}(L_W) = L_W L_W^{\top}\)</span>. The log absolute determinant is <span class="math display">\[\begin{equation*}
+\log \left| J_{f^{-1}} \right|
+= K \log(2) + \sum_{k=1}^K (K - k + 1) \log {(L_W)_{k,\, k}}.
+\end{equation*}\]</span></p>
+<p>The probability functions will raise errors if <span class="math inline">\(\nu \leq K - 1\)</span> or if <span class="math inline">\(L_S\)</span> and <span class="math inline">\(L_W\)</span> are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.</p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-73fb720e23b1747c1083dedea45393a2c31d1a97" class="index-entry"></a> <!-- real; wishart_cholesky_lpdf; (matrix L_W | real nu, matrix L_S); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wishart_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>wishart_cholesky_lpdf</code></a></strong><code>(matrix L_W | real nu, matrix L_S)</code><br> Return the log of the Wishart density for lower-triangular Cholesky factor <code>L_W</code> given degrees of freedom <code>nu</code> and lower-triangular Cholesky factor of the scale matrix <code>L_S</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-f87979d321f7d0bb07867f16b505ba45fd7a7376" class="index-entry"></a> <!-- real; wishart_cholesky_lupdf; (matrix L_W | real nu, matrix L_S); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wishart_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>wishart_cholesky_lupdf</code></a></strong><code>(matrix L_W | real nu, matrix L_S)</code><br> Return the log of the Wishart density for lower-triangular Cholesky factor of <code>L_W</code> given degrees of freedom <code>nu</code> and lower-triangular Cholesky factor of the scale matrix <code>L_S</code> dropping constant additive terms.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-311a69fbe7214ccc71cc4a5a9b6293967b467a38" class="index-entry"></a> <!-- matrix; wishart_cholesky_rng; (real nu, matrix L_S); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#wishart_cholesky_rng" class="unlink" title="Jump to index entry"><code>wishart_cholesky_rng</code></a></strong><code>(real nu, matrix L_S)</code><br> Generate the Cholesky factor of a Wishart variate with degrees of freedom <code>nu</code> and lower-triangular Cholesky factor of the scale matrix <code>L_S</code>; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="inverse-wishart-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="inverse-wishart-distribution">Inverse Wishart distribution</h2>
+<section id="probability-density-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-2">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\nu \in (K-1,\infty)\)</span>, and <span class="math inline">\(S \in
+\mathbb{R}^{K \times   K}\)</span> is symmetric and positive definite, then for symmetric and positive-definite <span class="math inline">\(W \in \mathbb{R}^{K \times K}\)</span>, <span class="math display">\[\begin{equation*}
+\text{InvWishart}(W \mid \nu,S) = \frac{1}{2^{\nu K / 2}} \ \frac{1}{\Gamma_K \! \left( \frac{\nu}{2} \right)}
+\ \left| S \right|^{\nu/2} \ \left| W \right|^{-(\nu + K + 1)/2}
+\ \exp \! \left( - \frac{1}{2} \ \text{tr}(SW^{-1}) \right) \! .
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>W ~</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart" class="unlink" title="Jump to index entry"><code>inv_wishart</code></a></strong><code>(nu, Sigma)</code></p>
+<p>Increment target log probability density with <code>inv_wishart_lupdf(W | nu, Sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-3961ee487e240f180388c9629267c086f03aceed" class="index-entry"></a> <!-- real; inv_wishart ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-38e0ab3788169c92def514827eaeff93db4d3af3" class="index-entry"></a> <!-- real; inv_wishart_lpdf; (matrix W | real nu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart_lpdf" class="unlink" title="Jump to index entry"><code>inv_wishart_lpdf</code></a></strong><code>(matrix W | real nu, matrix Sigma)</code><br> Return the log of the inverse Wishart density for symmetric and positive-definite matrix <code>W</code> given degrees of freedom <code>nu</code> and symmetric and positive-definite scale matrix <code>Sigma</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-2e69b51d6833e168c6a6d9646e93f6fb883609bd" class="index-entry"></a> <!-- real; inv_wishart_lupdf; (matrix W | real nu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart_lupdf" class="unlink" title="Jump to index entry"><code>inv_wishart_lupdf</code></a></strong><code>(matrix W | real nu, matrix Sigma)</code><br> Return the log of the inverse Wishart density for symmetric and positive-definite matrix <code>W</code> given degrees of freedom <code>nu</code> and symmetric and positive-definite scale matrix <code>Sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-ed1a3712028dbe64ab7bebe658f72ad4d101d3cf" class="index-entry"></a> <!-- matrix; inv_wishart_rng; (real nu, matrix Sigma); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart_rng" class="unlink" title="Jump to index entry"><code>inv_wishart_rng</code></a></strong><code>(real nu, matrix Sigma)</code><br> Generate an inverse Wishart variate with degrees of freedom <code>nu</code> and symmetric and positive-definite scale matrix <code>Sigma</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="inverse-wishart-cholesky-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="inverse-wishart-cholesky-distribution">Inverse Wishart distribution, Cholesky Parameterization</h2>
+<p>The Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If <span class="math inline">\(S\)</span> and <span class="math inline">\(W\)</span> are positive definite matrices with Cholesky factors <span class="math inline">\(L_S\)</span> and <span class="math inline">\(L_W\)</span> (i.e., <span class="math inline">\(S = L_S L_S^{\top}\)</span> and <span class="math inline">\(W = L_W L_W^{\top}\)</span>), then the Cholesky parameterization is defined so that <span class="math display">\[\begin{equation*}
+L_W \sim \textrm{InvWishartCholesky}(\nu, L_S)
+\end{equation*}\]</span> if and only if <span class="math display">\[\begin{equation*}
+W \sim \textrm{InvWishart}(\nu, S).
+\end{equation*}\]</span></p>
+<section id="probability-density-function-3" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-3">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\nu \in (K-1, \infty)\)</span>, and <span class="math inline">\(L_S, L_W \in
+\mathbb{R}^{K \times K}\)</span> are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is <span class="math display">\[\begin{equation*}
+\text{InvWishartCholesky}(L_W \mid \nu,L_S) =
+\text{InvWishart}(L_WL_W^{\top} \mid \nu, L_S L_S^{\top}) \, \left| J_{f^{-1}} \right|,
+\end{equation*}\]</span> where <span class="math inline">\(J_{f^{-1}}\)</span> is the Jacobian of the (inverse) transform of the variate, <span class="math inline">\(f^{-1}(L_W) = L_W L_W^{\top}\)</span>. The log absolute determinant is <span class="math display">\[\begin{equation*}
+\log \left| J_{f^{-1}} \right|
+= K \log(2) + \sum_{k=1}^K (K - k + 1) \log {(L_W)_{k,\, k}}.
+\end{equation*}\]</span></p>
+<p>The probability functions will raise errors if <span class="math inline">\(\nu \leq K - 1\)</span> or if <span class="math inline">\(L_S\)</span> and <span class="math inline">\(L_W\)</span> are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.</p>
+</section>
+<section id="stan-functions-3" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
+<a id="index-entry-0d09dfe80fda861a5757201ffaddddac82e00aa5" class="index-entry"></a> <!-- real; inv_wishart_cholesky_lpdf; (matrix L_W | real nu, matrix L_S); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>inv_wishart_cholesky_lpdf</code></a></strong><code>(matrix L_W | real nu, matrix L_S)</code><br> Return the log of the inverse Wishart density for lower-triangular Cholesky factor <code>L_W</code> given degrees of freedom <code>nu</code> and lower-triangular Cholesky factor of the scale matrix <code>L_S</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-eaf70d1f046e96371485e6dd22f5b0fe44e33fdf" class="index-entry"></a> <!-- real; inv_wishart_cholesky_lupdf; (matrix L_W | real nu, matrix L_S); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>inv_wishart_cholesky_lupdf</code></a></strong><code>(matrix L_W | real nu, matrix L_S)</code><br> Return the log of the inverse Wishart density for lower-triangular Cholesky factor of <code>L_W</code> given degrees of freedom <code>nu</code> and lower-triangular Cholesky factor of the scale matrix <code>L_S</code> dropping constant additive terms.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-327529ac254bd3b80d8b9c48d711700d3ed47fa3" class="index-entry"></a> <!-- matrix; inv_wishart_cholesky_rng; (real nu, matrix L_S); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#inv_wishart_cholesky_rng" class="unlink" title="Jump to index entry"><code>inv_wishart_cholesky_rng</code></a></strong><code>(real nu, matrix L_S)</code><br> Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom <code>nu</code> and lower-triangular Cholesky factor of the scale matrix <code>L_S</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.30</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/correlation_matrix_distributions.html" class="pagination-link" aria-label="Correlation Matrix Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Correlation Matrix Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/hidden_markov_models.html" class="pagination-link" aria-label="Hidden Markov Models">
+        <span class="nav-page-text">Hidden Markov Models</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/covariance_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/deprecated_functions.html b/docs/2_39/functions-reference/deprecated_functions.html
new file mode 100644
index 000000000..1a9def5ee
--- /dev/null
+++ b/docs/2_39/functions-reference/deprecated_functions.html
@@ -0,0 +1,1357 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Deprecated Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/removed_functions.html" rel="next">
+<link href="../functions-reference/transform_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Deprecated Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/deprecated_functions.html">Deprecated Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#deprecated-functions" id="toc-deprecated-functions" class="nav-link active" data-scroll-target="#deprecated-functions">Deprecated Functions</a>
+  <ul class="collapse">
+  <li><a href="#integer-division-with-operator" id="toc-integer-division-with-operator" class="nav-link" data-scroll-target="#integer-division-with-operator">Integer division with <code>operator/</code></a></li>
+  <li><a href="#functions-old-ode-solver" id="toc-functions-old-ode-solver" class="nav-link" data-scroll-target="#functions-old-ode-solver">integrate_ode_rk45, integrate_ode_adams, integrate_ode_bdf ODE Integrators</a>
+  <ul class="collapse">
+  <li><a href="#specifying-an-ordinary-differential-equation-as-a-function" id="toc-specifying-an-ordinary-differential-equation-as-a-function" class="nav-link" data-scroll-target="#specifying-an-ordinary-differential-equation-as-a-function">Specifying an ordinary differential equation as a function</a></li>
+  <li><a href="#non-stiff-solver" id="toc-non-stiff-solver" class="nav-link" data-scroll-target="#non-stiff-solver">Non-stiff solver</a></li>
+  <li><a href="#stiff-solver" id="toc-stiff-solver" class="nav-link" data-scroll-target="#stiff-solver">Stiff solver</a></li>
+  <li><a href="#arguments-to-the-ode-solvers" id="toc-arguments-to-the-ode-solvers" class="nav-link" data-scroll-target="#arguments-to-the-ode-solvers">Arguments to the ODE solvers</a></li>
+  </ul></li>
+  <li><a href="#functions-old-algebra-solver" id="toc-functions-old-algebra-solver" class="nav-link" data-scroll-target="#functions-old-algebra-solver">algebra_solver, algebra_solver_newton algebraic solvers</a>
+  <ul class="collapse">
+  <li><a href="#specifying-an-algebraic-equation-as-a-function" id="toc-specifying-an-algebraic-equation-as-a-function" class="nav-link" data-scroll-target="#specifying-an-algebraic-equation-as-a-function">Specifying an algebraic equation as a function</a></li>
+  <li><a href="#call-to-the-algebraic-solver" id="toc-call-to-the-algebraic-solver" class="nav-link" data-scroll-target="#call-to-the-algebraic-solver">Call to the algebraic solver</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/deprecated_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/deprecated_functions.html">Deprecated Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="deprecated-functions" class="level1">
+<h1>Deprecated Functions</h1>
+<p>This appendix lists currently deprecated functionality along with how to replace it.</p>
+<p>Starting in Stan 2.29, deprecated functions with drop in replacements (such as the renaming of <code>get_lp</code> or <code>multiply_log</code>) will be removed 3 versions later e.g., functions deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in <a href="../functions-reference/removed_functions.html">Removed Functions</a>. The Stan compiler can <a href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html#stanc-pretty-printing">automatically update</a> these on the behalf of the user for the entire deprecation window and at least one version following the removal.</p>
+<section id="integer-division-with-operator" class="level2">
+<h2 class="anchored" data-anchor-id="integer-division-with-operator">Integer division with <code>operator/</code></h2>
+<p><em>Deprecated</em>: Using <code>/</code> with two integer arguments is interpreted as integer floor division, such that</p>
+<p><span class="math display">\[ 1 / 2 = 0 \]</span></p>
+<p>This is deprecated due to its confusion with real-valued division, where</p>
+<p><span class="math display">\[ 1.0 / 2.0 = 0.5 \]</span></p>
+<p><em>Replacement</em>: Use the integer division operator <code>operator%/%</code> instead.</p>
+</section>
+<section id="functions-old-ode-solver" class="level2">
+<h2 class="anchored" data-anchor-id="functions-old-ode-solver">integrate_ode_rk45, integrate_ode_adams, integrate_ode_bdf ODE Integrators</h2>
+<p>These ODE integrator functions have been replaced by those described in <a href="../functions-reference/higher-order_functions.html#functions-ode-solver">Ordinary Differential Equation (ODE) Solvers</a>.</p>
+<section id="specifying-an-ordinary-differential-equation-as-a-function" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-an-ordinary-differential-equation-as-a-function">Specifying an ordinary differential equation as a function</h3>
+<p>A system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[] <span class="dt">real</span> ode(<span class="dt">real</span> time, <span class="dt">array</span>[] <span class="dt">real</span> state, <span class="dt">array</span>[] <span class="dt">real</span> theta,</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>                 <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="dt">array</span>[] <span class="dt">int</span> x_i);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.</p>
+<p>The arguments to this function are:</p>
+<ul>
+<li><p><em><code>time</code></em>, the time to evaluate the ODE system</p></li>
+<li><p><em><code>state</code></em>, the state of the ODE system at the time specified</p></li>
+<li><p><em><code>theta</code></em>, parameter values used to evaluate the ODE system</p></li>
+<li><p><em><code>x_r</code></em>, data values used to evaluate the ODE system</p></li>
+<li><p><em><code>x_i</code></em>, integer data values used to evaluate the ODE system.</p></li>
+</ul>
+<p>The ODE system function separates parameter values, <em><code>theta</code></em>, from data values, <em><code>x_r</code></em>, for efficiency in computing the gradients of the ODE.</p>
+</section>
+<section id="non-stiff-solver" class="level3">
+<h3 class="anchored" data-anchor-id="non-stiff-solver">Non-stiff solver</h3>
+<a id="index-entry-fddefd52462ff3a8c14cca8da72bb5ef1596fbe0" class="index-entry"></a> <!-- array[,] real; integrate_ode_rk45; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode_rk45" class="unlink" title="Jump to index entry"><code>integrate_ode_rk45</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i)</code><br> Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.</p>
+<small><i>Available since 2.10, deprecated in 2.24</i></small>
+<a id="index-entry-ed3282ca9353c591d894371c5da7f6c427ec52ff" class="index-entry"></a> <!-- array[,] real; integrate_ode_rk45; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode_rk45" class="unlink" title="Jump to index entry"><code>integrate_ode_rk45</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps)</code><br> Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.</p>
+<small><i>Available since 2.10, deprecated in 2.24</i></small>
+<a id="index-entry-f2263dbf460fa8a8e16c36bec1723e7cb60e8ff7" class="index-entry"></a> <!-- array[,] real; integrate_ode; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode" class="unlink" title="Jump to index entry"><code>integrate_ode</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i)</code><br> Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.</p>
+<small><i>Available since 2.10, deprecated in 2.24</i></small>
+<a id="index-entry-0243ee75a588fbb73c6e50121f71844a6cbfb6e6" class="index-entry"></a> <!-- array[,] real; integrate_ode_adams; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode_adams" class="unlink" title="Jump to index entry"><code>integrate_ode_adams</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i)</code><br> Solves the ODE system for the times provided using the Adams-Moulton method.</p>
+<small><i>Available since 2.23, deprecated in 2.24</i></small>
+<a id="index-entry-f382cd6903765b51897e2aa1b1eaa55cf63fd4d6" class="index-entry"></a> <!-- array[,] real; integrate_ode_adams; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode_adams" class="unlink" title="Jump to index entry"><code>integrate_ode_adams</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps)</code><br> Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.</p>
+<small><i>Available since 2.23, deprecated in 2.24</i></small>
+</section>
+<section id="stiff-solver" class="level3">
+<h3 class="anchored" data-anchor-id="stiff-solver">Stiff solver</h3>
+<a id="index-entry-0cb2d79940f9e180689b48555a1a25401db9b41a" class="index-entry"></a> <!-- array[,] real; integrate_ode_bdf; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode_bdf" class="unlink" title="Jump to index entry"><code>integrate_ode_bdf</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i)</code><br> Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.</p>
+<small><i>Available since 2.10, deprecated in 2.24</i></small>
+<a id="index-entry-3766842394a3f0b7eb3333fb202227f261c1f535" class="index-entry"></a> <!-- array[,] real; integrate_ode_bdf; (function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#integrate_ode_bdf" class="unlink" title="Jump to index entry"><code>integrate_ode_bdf</code></a></strong><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps)</code><br> Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.</p>
+<small><i>Available since 2.10, deprecated in 2.24</i></small>
+</section>
+<section id="arguments-to-the-ode-solvers" class="level3">
+<h3 class="anchored" data-anchor-id="arguments-to-the-ode-solvers">Arguments to the ODE solvers</h3>
+<p>The arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.</p>
+<ul>
+<li><em><code>ode</code></em>: function literal referring to a function specifying the system of differential equations with signature:</li>
+</ul>
+<pre><code>(real, array[] real, array[] real, data array[] real, data array[] int):array[] real</code></pre>
+<p>The arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,</p>
+<ul>
+<li><p><em><code>initial_state</code></em>: initial state, type <code>array[] real</code>,</p></li>
+<li><p><em><code>initial_time</code></em>: initial time, type <code>int</code> or <code>real</code>,</p></li>
+<li><p><em><code>times</code></em>: solution times, type <code>array[] real</code>,</p></li>
+<li><p><em><code>theta</code></em>: parameters, type <code>array[] real</code>,</p></li>
+<li><p><code>data</code> <em><code>x_r</code></em>: real data, type <code>array[] real</code>, data only, and</p></li>
+<li><p><code>data</code> <em><code>x_i</code></em>: integer data, type <code>array[] int</code>, data only.</p></li>
+</ul>
+<p>For more fine-grained control of the ODE solvers, these parameters can also be provided:</p>
+<ul>
+<li><p><code>data</code> <em><code>rel_tol</code></em>: relative tolerance for the ODE solver, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>abs_tol</code></em>: absolute tolerance for the ODE solver, type <code>real</code>, data only, and</p></li>
+<li><p><code>data</code> <em><code>max_num_steps</code></em>: maximum number of steps to take in the ODE solver, type <code>int</code>, data only.</p></li>
+</ul>
+<section id="return-values" class="level4">
+<h4 class="anchored" data-anchor-id="return-values">Return values</h4>
+<p>The return value for the ODE solvers is an array of type <code>array[,] real</code>, with values consisting of solutions at the specified times.</p>
+</section>
+<section id="sizes-and-parallel-arrays" class="level4">
+<h4 class="anchored" data-anchor-id="sizes-and-parallel-arrays">Sizes and parallel arrays</h4>
+<p>The sizes must match, and in particular, the following groups are of the same size:</p>
+<ul>
+<li><p>state variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,</p></li>
+<li><p>solution times and number of rows of the return value of the solver,</p></li>
+<li><p>parameters, real data and integer data passed to the solver will be passed to the system function</p></li>
+</ul>
+</section>
+</section>
+</section>
+<section id="functions-old-algebra-solver" class="level2">
+<h2 class="anchored" data-anchor-id="functions-old-algebra-solver">algebra_solver, algebra_solver_newton algebraic solvers</h2>
+<p>These algebraic solver functions have been replaced by those described in <a href="../functions-reference/higher-order_functions.html#functions-algebraic-solver">Algebraic Equation Solvers</a>..</p>
+<section id="specifying-an-algebraic-equation-as-a-function" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-an-algebraic-equation-as-a-function">Specifying an algebraic equation as a function</h3>
+<p>An algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span> algebra_system(<span class="dt">vector</span> y, <span class="dt">vector</span> theta,</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>                              <span class="kw">data</span> <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="dt">array</span>[] <span class="dt">int</span> x_i)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.</p>
+<p>The argument of this function are:</p>
+<ul>
+<li><p><em><code>y</code></em>, the unknowns we wish to solve for</p></li>
+<li><p><em><code>theta</code></em>, parameter values used to evaluate the algebraic system</p></li>
+<li><p><em><code>x_r</code></em>, data values used to evaluate the algebraic system</p></li>
+<li><p><em><code>x_i</code></em>, integer data used to evaluate the algebraic system</p></li>
+</ul>
+<p>The algebraic system function separates parameter values, <em><code>theta</code></em>, from data values, <em><code>x_r</code></em>, for efficiency in propagating the derivatives through the algebraic system.</p>
+</section>
+<section id="call-to-the-algebraic-solver" class="level3">
+<h3 class="anchored" data-anchor-id="call-to-the-algebraic-solver">Call to the algebraic solver</h3>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#algebra_solver" class="unlink" title="Jump to index entry"><code>algebra_solver</code></a></strong><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i)</code><br> Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.</p>
+<small><i>Available since 2.17, deprecated in 2.31</i></small>
+<a id="index-entry-717206dad47a7c1194d4b531baf6ffdd3c2d028a" class="index-entry"></a> <!-- vector; algebra_solver; (function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#algebra_solver" class="unlink" title="Jump to index entry"><code>algebra_solver</code></a></strong><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps)</code><br> Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.</p>
+<small><i>Available since 2.17, deprecated in 2.31</i></small>
+<p><em>Note:</em> In future releases, the function <code>algebra_solver</code> will be deprecated and replaced with <code>algebra_solver_powell</code>.</p>
+<a id="index-entry-c32bcab203b8a5c04135db0647b03444d7effefb" class="index-entry"></a> <!-- vector; algebra_solver_newton; (function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#algebra_solver_newton" class="unlink" title="Jump to index entry"><code>algebra_solver_newton</code></a></strong><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i)</code><br> Solves the algebraic system, given an initial guess, using Newton’s method.</p>
+<small><i>Available since 2.24, deprecated in 2.31</i></small>
+<a id="index-entry-e7af0c43c395b76b7182b00996444a8b8377cac4" class="index-entry"></a> <!-- vector; algebra_solver_newton; (function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#algebra_solver_newton" class="unlink" title="Jump to index entry"><code>algebra_solver_newton</code></a></strong><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps)</code><br> Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.</p>
+<small><i>Available since 2.24, deprecated in 2.31</i></small>
+<section id="arguments-to-the-algebraic-solver" class="level4">
+<h4 class="anchored" data-anchor-id="arguments-to-the-algebraic-solver">Arguments to the algebraic solver</h4>
+<p>The arguments to the algebraic solvers are as follows:</p>
+<ul>
+<li><p><em><code>algebra_system</code></em>: function literal referring to a function specifying the system of algebraic equations with signature <code>(vector, vector, array[] real, array[] int):vector</code>. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,</p></li>
+<li><p><em><code>y_guess</code></em>: initial guess for the solution, type <code>vector</code>,</p></li>
+<li><p><em><code>theta</code></em>: parameters only, type <code>vector</code>,</p></li>
+<li><p><em><code>x_r</code></em>: real data only, type <code>array[] real</code>, and</p></li>
+<li><p><em><code>x_i</code></em>: integer data only, type <code>array[] int</code>.</p></li>
+</ul>
+<p>For more fine-grained control of the algebraic solver, these parameters can also be provided:</p>
+<ul>
+<li><p><em><code>rel_tol</code></em>: relative tolerance for the algebraic solver, type <code>real</code>, data only,</p></li>
+<li><p><em><code>function_tol</code></em>: function tolerance for the algebraic solver, type <code>real</code>, data only,</p></li>
+<li><p><em><code>max_num_steps</code></em>: maximum number of steps to take in the algebraic solver, type <code>int</code>, data only.</p></li>
+</ul>
+</section>
+<section id="return-value" class="level4">
+<h4 class="anchored" data-anchor-id="return-value">Return value</h4>
+<p>The return value for the algebraic solver is an object of type <code>vector</code>, with values which, when plugged in as <code>y</code> make the algebraic function go to 0.</p>
+</section>
+<section id="sizes-and-parallel-arrays-1" class="level4">
+<h4 class="anchored" data-anchor-id="sizes-and-parallel-arrays-1">Sizes and parallel arrays</h4>
+<p>Certain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.</p>
+<p>The parameters, real data, and integer data will be passed from the solver directly to the system function.</p>
+
+
+</section>
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/transform_functions.html" class="pagination-link" aria-label="Variable Transformation Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Variable Transformation Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/removed_functions.html" class="pagination-link" aria-label="Removed Functions">
+        <span class="nav-page-text">Removed Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/deprecated_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/distributions_over_unbounded_vectors.html b/docs/2_39/functions-reference/distributions_over_unbounded_vectors.html
new file mode 100644
index 000000000..aa66e4846
--- /dev/null
+++ b/docs/2_39/functions-reference/distributions_over_unbounded_vectors.html
@@ -0,0 +1,1624 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Distributions over Unbounded Vectors</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/simplex_distributions.html" rel="next">
+<link href="../functions-reference/bounded_continuous_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Distributions over Unbounded Vectors">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/distributions_over_unbounded_vectors.html">Distributions over Unbounded Vectors</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#distributions-over-unbounded-vectors" id="toc-distributions-over-unbounded-vectors" class="nav-link active" data-scroll-target="#distributions-over-unbounded-vectors">Distributions over Unbounded Vectors</a>
+  <ul class="collapse">
+  <li><a href="#multivariate-normal-distribution" id="toc-multivariate-normal-distribution" class="nav-link" data-scroll-target="#multivariate-normal-distribution">Multivariate normal distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multivariate-normal-distribution-precision-parameterization" id="toc-multivariate-normal-distribution-precision-parameterization" class="nav-link" data-scroll-target="#multivariate-normal-distribution-precision-parameterization">Multivariate normal distribution, precision parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multi-normal-cholesky-fun" id="toc-multi-normal-cholesky-fun" class="nav-link" data-scroll-target="#multi-normal-cholesky-fun">Multivariate normal distribution, Cholesky parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-2" id="toc-probability-density-function-2" class="nav-link" data-scroll-target="#probability-density-function-2">Probability density function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multivariate-gaussian-process-distribution" id="toc-multivariate-gaussian-process-distribution" class="nav-link" data-scroll-target="#multivariate-gaussian-process-distribution">Multivariate Gaussian process distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-3" id="toc-probability-density-function-3" class="nav-link" data-scroll-target="#probability-density-function-3">Probability density function</a></li>
+  <li><a href="#distribution-statement-3" id="toc-distribution-statement-3" class="nav-link" data-scroll-target="#distribution-statement-3">Distribution statement</a></li>
+  <li><a href="#stan-functions-3" id="toc-stan-functions-3" class="nav-link" data-scroll-target="#stan-functions-3">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multivariate-gaussian-process-distribution-cholesky-parameterization" id="toc-multivariate-gaussian-process-distribution-cholesky-parameterization" class="nav-link" data-scroll-target="#multivariate-gaussian-process-distribution-cholesky-parameterization">Multivariate Gaussian process distribution, Cholesky parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-4" id="toc-probability-density-function-4" class="nav-link" data-scroll-target="#probability-density-function-4">Probability density function</a></li>
+  <li><a href="#distribution-statement-4" id="toc-distribution-statement-4" class="nav-link" data-scroll-target="#distribution-statement-4">Distribution statement</a></li>
+  <li><a href="#stan-functions-4" id="toc-stan-functions-4" class="nav-link" data-scroll-target="#stan-functions-4">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multivariate-student-t-distribution" id="toc-multivariate-student-t-distribution" class="nav-link" data-scroll-target="#multivariate-student-t-distribution">Multivariate Student-t distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-5" id="toc-probability-density-function-5" class="nav-link" data-scroll-target="#probability-density-function-5">Probability density function</a></li>
+  <li><a href="#distribution-statement-5" id="toc-distribution-statement-5" class="nav-link" data-scroll-target="#distribution-statement-5">Distribution statement</a></li>
+  <li><a href="#stan-functions-5" id="toc-stan-functions-5" class="nav-link" data-scroll-target="#stan-functions-5">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multi-student-t-cholesky-fun" id="toc-multi-student-t-cholesky-fun" class="nav-link" data-scroll-target="#multi-student-t-cholesky-fun">Multivariate Student-t distribution, Cholesky parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-6" id="toc-probability-density-function-6" class="nav-link" data-scroll-target="#probability-density-function-6">Probability density function</a></li>
+  <li><a href="#distribution-statement-6" id="toc-distribution-statement-6" class="nav-link" data-scroll-target="#distribution-statement-6">Distribution statement</a></li>
+  <li><a href="#stan-functions-6" id="toc-stan-functions-6" class="nav-link" data-scroll-target="#stan-functions-6">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#gaussian-dynamic-linear-models" id="toc-gaussian-dynamic-linear-models" class="nav-link" data-scroll-target="#gaussian-dynamic-linear-models">Gaussian dynamic linear models</a>
+  <ul class="collapse">
+  <li><a href="#distribution-statement-7" id="toc-distribution-statement-7" class="nav-link" data-scroll-target="#distribution-statement-7">Distribution statement</a></li>
+  <li><a href="#stan-functions-7" id="toc-stan-functions-7" class="nav-link" data-scroll-target="#stan-functions-7">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/distributions_over_unbounded_vectors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/distributions_over_unbounded_vectors.html">Distributions over Unbounded Vectors</a></li></ol></nav></header>
+
+
+
+
+<section id="distributions-over-unbounded-vectors" class="level1">
+<h1>Distributions over Unbounded Vectors</h1>
+<p>The unbounded vector probability distributions have support on all of <span class="math inline">\(\mathbb{R}^K\)</span> for some fixed <span class="math inline">\(K\)</span>.</p>
+<section id="multivariate-normal-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-normal-distribution">Multivariate normal distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\mu \in \mathbb{R}^K\)</span>, and <span class="math inline">\(\Sigma \in
+\mathbb{R}^{K \times   K}\)</span> is symmetric and positive definite, then for <span class="math inline">\(y \in \mathbb{R}^K\)</span>, <span class="math display">\[\begin{equation*}
+\text{MultiNormal}(y|\mu,\Sigma) =
+\frac{1}{\left( 2 \pi \right)^{K/2}} \ \frac{1}{\sqrt{|\Sigma|}}
+\  \exp \! \left( \! - \frac{1}{2} (y - \mu)^{\top} \, \Sigma^{-1} \, (y - \mu) \right) \! ,
+\end{equation*}\]</span> where <span class="math inline">\(|\Sigma|\)</span> is the absolute determinant of <span class="math inline">\(\Sigma\)</span>.</p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_normal" class="unlink" title="Jump to index entry"><code>multi_normal</code></a></strong><code>(mu, Sigma)</code></p>
+<p>Increment target log probability density with <code>multi_normal_lupdf(y | mu, Sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-55bcd96ac570a9cd0612aa8c9853df2ef3ca7750" class="index-entry"></a> <!-- real; multi_normal ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<p>The multivariate normal probability function is overloaded to allow the variate vector <span class="math inline">\(y\)</span> and location vector <span class="math inline">\(\mu\)</span> to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a> for a description of vectorization.</p>
+<a id="index-entry-ea7d4337cebac7fdbfa928b24a3626208823e80e" class="index-entry"></a> <!-- real; multi_normal_lpdf; (vectors y | vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_lpdf</code></a></strong><code>(vectors y | vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-7786848b839eeda0bfa728760464ac4a8369550d" class="index-entry"></a> <!-- real; multi_normal_lupdf; (vectors y | vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_lupdf</code></a></strong><code>(vectors y | vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-c1e8d116401a5e7f66812da95e1285dc2a3b111a" class="index-entry"></a> <!-- real; multi_normal_lpdf; (vectors y | row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_lpdf</code></a></strong><code>(vectors y | row_vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-a0fb7f94e06645589741da551fba4a55a0a33e44" class="index-entry"></a> <!-- real; multi_normal_lupdf; (vectors y | row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_lupdf</code></a></strong><code>(vectors y | row_vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-fe635d231edc36e95bda6936e3dc7f58cf04d7b9" class="index-entry"></a> <!-- real; multi_normal_lpdf; (row_vectors y | vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_lpdf</code></a></strong><code>(row_vectors y | vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-69f99f5b6cefce126adc27f7b20728705d7e8a53" class="index-entry"></a> <!-- real; multi_normal_lupdf; (row_vectors y | vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_lupdf</code></a></strong><code>(row_vectors y | vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-7d157b823f0ae33310094712caff4096b644a282" class="index-entry"></a> <!-- real; multi_normal_lpdf; (row_vectors y | row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_lpdf</code></a></strong><code>(row_vectors y | row_vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-86c5dd8e6cf2c484c3c76fffcadc99f786b7265b" class="index-entry"></a> <!-- real; multi_normal_lupdf; (row_vectors y | row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_lupdf</code></a></strong><code>(row_vectors y | row_vectors mu, matrix Sigma)</code><br> The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<p>Although there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call <code>multi_normal_cholesky_rng</code>; see section <a href="#multi-normal-cholesky-fun">multi-variate normal, cholesky parameterization</a>.</p>
+<a id="index-entry-d83a51f846b5823f2e419d6189b27cd98643a584" class="index-entry"></a> <!-- vector; multi_normal_rng; (vector mu, matrix Sigma); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_rng" class="unlink" title="Jump to index entry"><code>multi_normal_rng</code></a></strong><code>(vector mu, matrix Sigma)</code><br> Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-70771ae7e3296e2a09378f81d64a16d0ffdebe49" class="index-entry"></a> <!-- vector; multi_normal_rng; (row_vector mu, matrix Sigma); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_rng" class="unlink" title="Jump to index entry"><code>multi_normal_rng</code></a></strong><code>(row_vector mu, matrix Sigma)</code><br> Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-ba20f2e8644cf2fff8761565c61ae10f1160df3a" class="index-entry"></a> <!-- vectors; multi_normal_rng; (vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_rng" class="unlink" title="Jump to index entry"><code>multi_normal_rng</code></a></strong><code>(vectors mu, matrix Sigma)</code><br> Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-214830eeea24ebd3c375e92411ceab1101a15b6c" class="index-entry"></a> <!-- vectors; multi_normal_rng; (row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_rng" class="unlink" title="Jump to index entry"><code>multi_normal_rng</code></a></strong><code>(row_vectors mu, matrix Sigma)</code><br> Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="multivariate-normal-distribution-precision-parameterization" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-normal-distribution-precision-parameterization">Multivariate normal distribution, precision parameterization</h2>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\mu \in \mathbb{R}^K\)</span>, and <span class="math inline">\(\Omega \in
+\mathbb{R}^{K \times   K}\)</span> is symmetric and positive definite, then for <span class="math inline">\(y \in \mathbb{R}^K\)</span>, <span class="math display">\[\begin{equation*} \text{MultiNormalPrecision}(y|\mu,\Omega)
+= \text{MultiNormal}(y|\mu,\Omega^{-1}) \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec" class="unlink" title="Jump to index entry"><code>multi_normal_prec</code></a></strong><code>(mu, Omega)</code></p>
+<p>Increment target log probability density with <code>multi_normal_prec_lupdf(y | mu, Omega)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-541ca6516f2a9c57d99c4d0b57a2b0a1df8a105d" class="index-entry"></a> <!-- real; multi_normal_prec ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-6a00a4a10d9e84eaa51e885395294f4b4e1272dd" class="index-entry"></a> <!-- real; multi_normal_prec_lpdf; (vectors y | vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lpdf</code></a></strong><code>(vectors y | vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-e4db099421dc8d5cb5ccaf22226497142e4ad656" class="index-entry"></a> <!-- real; multi_normal_prec_lupdf; (vectors y | vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lupdf</code></a></strong><code>(vectors y | vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-dcfb9bff5f659aa499401ffd77ffd7bbaa784936" class="index-entry"></a> <!-- real; multi_normal_prec_lpdf; (vectors y | row_vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lpdf</code></a></strong><code>(vectors y | row_vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-ec0990b5c058b76b441e3132b8e4986f8fc3b9a6" class="index-entry"></a> <!-- real; multi_normal_prec_lupdf; (vectors y | row_vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lupdf</code></a></strong><code>(vectors y | row_vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8efc2d6f42f7fd5eeb961e3959b617201d45a58c" class="index-entry"></a> <!-- real; multi_normal_prec_lpdf; (row_vectors y | vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lpdf</code></a></strong><code>(row_vectors y | vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-314ed87058998a1934a7cd80a00abca432768d68" class="index-entry"></a> <!-- real; multi_normal_prec_lupdf; (row_vectors y | vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lupdf</code></a></strong><code>(row_vectors y | vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-eb92ca14c43c32d999ac0b10e2160f641a23f265" class="index-entry"></a> <!-- real; multi_normal_prec_lpdf; (row_vectors y | row_vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lpdf</code></a></strong><code>(row_vectors y | row_vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-9cf50bfea31a9655df9c07627982295a3a884afa" class="index-entry"></a> <!-- real; multi_normal_prec_lupdf; (row_vectors y | row_vectors mu, matrix Omega); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_prec_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_prec_lupdf</code></a></strong><code>(row_vectors y | row_vectors mu, matrix Omega)</code><br> The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="multi-normal-cholesky-fun" class="level2">
+<h2 class="anchored" data-anchor-id="multi-normal-cholesky-fun">Multivariate normal distribution, Cholesky parameterization</h2>
+<section id="probability-density-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-2">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\mu \in \mathbb{R}^K\)</span>, and <span class="math inline">\(L \in
+\mathbb{R}^{K \times K}\)</span> is lower triangular and such that <span class="math inline">\(LL^{\top}\)</span> is positive definite, then for <span class="math inline">\(y \in \mathbb{R}^K\)</span>, <span class="math display">\[\begin{equation*}
+\text{MultiNormalCholesky}(y|\mu,L) =
+\text{MultiNormal}(y|\mu,LL^{\top}). \end{equation*}\]</span> If <span class="math inline">\(L\)</span> is lower triangular and <span class="math inline">\(LL^{top}\)</span> is a <span class="math inline">\(K \times K\)</span> positive definite matrix, then <span class="math inline">\(L_{k,k}\)</span> must be strictly positive for <span class="math inline">\(k \in 1{:}K\)</span>. If an <span class="math inline">\(L\)</span> is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.</p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky</code></a></strong><code>(mu, L)</code></p>
+<p>Increment target log probability density with <code>multi_normal_cholesky_lupdf(y | mu, L)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-478582106e708a75c36c3c557d30196407f15133" class="index-entry"></a> <!-- real; multi_normal_cholesky ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-b197f45c690e4b7dd4ebe29d9f75790be4acf177" class="index-entry"></a> <!-- real; multi_normal_cholesky_lpdf; (vectors y | vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lpdf</code></a></strong><code>(vectors y | vectors mu, matrix L)</code><br> The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-2ac54d7aebe6a22693ee5a9016fef469246a746c" class="index-entry"></a> <!-- real; multi_normal_cholesky_lupdf; (vectors y | vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lupdf</code></a></strong><code>(vectors y | vectors mu, matrix L)</code><br> The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-efdc316743c9c9f1a6ee782650494ee45cb045cd" class="index-entry"></a> <!-- real; multi_normal_cholesky_lpdf; (vectors y | row_vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lpdf</code></a></strong><code>(vectors y | row_vectors mu, matrix L)</code><br> The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-a04f4fcdecc170b474051229edff4d2358b9c619" class="index-entry"></a> <!-- real; multi_normal_cholesky_lupdf; (vectors y | row_vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lupdf</code></a></strong><code>(vectors y | row_vectors mu, matrix L)</code><br> The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-69e5b64a11f77c97cd0fffeaee869c5358969237" class="index-entry"></a> <!-- real; multi_normal_cholesky_lpdf; (row_vectors y | vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lpdf</code></a></strong><code>(row_vectors y | vectors mu, matrix L)</code><br> The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-d3682df812ba6f35ad608d62b19f6c8ca23601f7" class="index-entry"></a> <!-- real; multi_normal_cholesky_lupdf; (row_vectors y | vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lupdf</code></a></strong><code>(row_vectors y | vectors mu, matrix L)</code><br> The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-d5709f43c49fc57a5ef58e94e176962d2bff379a" class="index-entry"></a> <!-- real; multi_normal_cholesky_lpdf; (row_vectors y | row_vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lpdf</code></a></strong><code>(row_vectors y | row_vectors mu, matrix L)</code><br> The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-1606261a4f25b4e68b2bf4b1c8bf2513e62cdf5c" class="index-entry"></a> <!-- real; multi_normal_cholesky_lupdf; (row_vectors y | row_vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_lupdf</code></a></strong><code>(row_vectors y | row_vectors mu, matrix L)</code><br> The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-b2936af169906302b9686d4673d8c5a8da710efb" class="index-entry"></a> <!-- vector; multi_normal_cholesky_rng; (vector mu, matrix L); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_rng</code></a></strong><code>(vector mu, matrix L)</code><br> Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-83a370eb75f07ac047e03e4dde9d64cf42e4c432" class="index-entry"></a> <!-- vector; multi_normal_cholesky_rng; (row_vector mu, matrix L); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_rng</code></a></strong><code>(row_vector mu, matrix L)</code><br> Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-7e60f3380a18c3f7bbcc383bfa7fcd15c14c054b" class="index-entry"></a> <!-- vectors; multi_normal_cholesky_rng; (vectors mu, matrix L); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_rng</code></a></strong><code>(vectors mu, matrix L)</code><br> Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-e6934af3e7cb8c68c3aa044307129bbaf17ffabf" class="index-entry"></a> <!-- vectors; multi_normal_cholesky_rng; (row_vectors mu, matrix L); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#multi_normal_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_normal_cholesky_rng</code></a></strong><code>(row_vectors mu, matrix L)</code><br> Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="multivariate-gaussian-process-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-gaussian-process-distribution">Multivariate Gaussian process distribution</h2>
+<section id="probability-density-function-3" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-3">Probability density function</h3>
+<p>If <span class="math inline">\(K,N \in \mathbb{N}\)</span>, <span class="math inline">\(\Sigma \in \mathbb{R}^{N \times N}\)</span> is symmetric, positive definite kernel matrix and <span class="math inline">\(w \in \mathbb{R}^{K}\)</span> is a vector of positive inverse scales, then for <span class="math inline">\(y \in \mathbb{R}^{K
+\times N}\)</span>, <span class="math display">\[\begin{equation*} \text{MultiGP}(y|\Sigma,w) = \prod_{i=1}^{K}
+\text{MultiNormal}(y_i|0,w_i^{-1} \Sigma), \end{equation*}\]</span> where <span class="math inline">\(y_i\)</span> is the <span class="math inline">\(i\)</span>th row of <span class="math inline">\(y\)</span>. This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.</p>
+</section>
+<section id="distribution-statement-3" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-3">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_gp" class="unlink" title="Jump to index entry"><code>multi_gp</code></a></strong><code>(Sigma, w)</code></p>
+<p>Increment target log probability density with <code>multi_gp_lupdf(y | Sigma, w)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-457f8d9e9ebaf6012b892a8241fe77b322473f49" class="index-entry"></a> <!-- real; multi_gp ~; -->
+<p></p>
+</section>
+<section id="stan-functions-3" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
+<a id="index-entry-fdd1f01264b88aa3f7a0cae0db463d1f87d5e0e8" class="index-entry"></a> <!-- real; multi_gp_lpdf; (matrix y | matrix Sigma, vector w); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_gp_lpdf" class="unlink" title="Jump to index entry"><code>multi_gp_lpdf</code></a></strong><code>(matrix y | matrix Sigma, vector w)</code><br> The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-ead480cd44ceccdf956822e08476ca3b5ca14a76" class="index-entry"></a> <!-- real; multi_gp_lupdf; (matrix y | matrix Sigma, vector w); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_gp_lupdf" class="unlink" title="Jump to index entry"><code>multi_gp_lupdf</code></a></strong><code>(matrix y | matrix Sigma, vector w)</code><br> The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="multivariate-gaussian-process-distribution-cholesky-parameterization" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-gaussian-process-distribution-cholesky-parameterization">Multivariate Gaussian process distribution, Cholesky parameterization</h2>
+<section id="probability-density-function-4" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-4">Probability density function</h3>
+<p>If <span class="math inline">\(K,N \in \mathbb{N}\)</span>, <span class="math inline">\(L \in \mathbb{R}^{N \times N}\)</span> is lower triangular and such that <span class="math inline">\(LL^{\top}\)</span> is positive definite kernel matrix (implying <span class="math inline">\(L_{n,n} &gt; 0\)</span> for <span class="math inline">\(n \in 1{:}N\)</span>), and <span class="math inline">\(w \in
+\mathbb{R}^{K}\)</span> is a vector of positive inverse scales, then for <span class="math inline">\(y
+\in \mathbb{R}^{K \times N}\)</span>, <span class="math display">\[\begin{equation*} \text{MultiGPCholesky}(y \, | \ L,w)
+= \prod_{i=1}^{K} \text{MultiNormal}(y_i|0,w_i^{-1} LL^{\top}), \end{equation*}\]</span> where <span class="math inline">\(y_i\)</span> is the <span class="math inline">\(i\)</span>th row of <span class="math inline">\(y\)</span>. This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than <span class="math inline">\(\text{MultiGP}()\)</span>. Note that this function does not take into account the mean prediction.</p>
+</section>
+<section id="distribution-statement-4" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-4">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_gp_cholesky" class="unlink" title="Jump to index entry"><code>multi_gp_cholesky</code></a></strong><code>(L, w)</code></p>
+<p>Increment target log probability density with <code>multi_gp_cholesky_lupdf(y | L, w)</code>.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-5a77e4209bc50ea87ef1dbcce3bad7b10b094a6b" class="index-entry"></a> <!-- real; multi_gp_cholesky ~; -->
+<p></p>
+</section>
+<section id="stan-functions-4" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-4">Stan functions</h3>
+<a id="index-entry-2903bb116c036e35d3cb81e008dcad2b7a6e6756" class="index-entry"></a> <!-- real; multi_gp_cholesky_lpdf; (matrix y | matrix L, vector w); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_gp_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>multi_gp_cholesky_lpdf</code></a></strong><code>(matrix y | matrix L, vector w)</code><br> The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-df18f8bd56610fdd6df3192990ee5cb001142817" class="index-entry"></a> <!-- real; multi_gp_cholesky_lupdf; (matrix y | matrix L, vector w); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_gp_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>multi_gp_cholesky_lupdf</code></a></strong><code>(matrix y | matrix L, vector w)</code><br> The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="multivariate-student-t-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-student-t-distribution">Multivariate Student-t distribution</h2>
+<section id="probability-density-function-5" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-5">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\nu \in \mathbb{R}^+\)</span>, <span class="math inline">\(\mu \in \mathbb{R}^K\)</span>, and <span class="math inline">\(\Sigma \in \mathbb{R}^{K \times K}\)</span> is symmetric and positive definite, then for <span class="math inline">\(y \in \mathbb{R}^K\)</span>, <span class="math display">\[\begin{equation*} \begin{array}{l}
+\text{MultiStudentT}(y\,|\,\nu,\,\mu,\,\Sigma) \\  =
+\frac{1}{\pi^{K/2}} \ \frac{1}{\nu^{K/2}} \ \frac{\Gamma\!\left((\nu +
+K)/2\right)}      {\Gamma(\nu/2)} \ \frac{1}{\sqrt{\left| \Sigma
+\right|}} \ \left( 1 + \frac{1}{\nu} \, \left(y - \mu\right)^{\top} \,
+\Sigma^{-1} \, \left(y - \mu\right) \right)^{-(\nu + K)/2} \! .
+\end{array} \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-5" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-5">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t" class="unlink" title="Jump to index entry"><code>multi_student_t</code></a></strong><code>(nu, mu, Sigma)</code></p>
+<p>Increment target log probability density with <code>multi_student_t_lupdf(y | nu, mu, Sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6f17c5611de64710688848607c0edd592e244fe1" class="index-entry"></a> <!-- real; multi_student_t ~; -->
+<p></p>
+</section>
+<section id="stan-functions-5" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-5">Stan functions</h3>
+<a id="index-entry-69dbd18bfe0fa38bad7531c187ffd5ddbc68655e" class="index-entry"></a> <!-- real; multi_student_t_lpdf; (vectors y | real nu, vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lpdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lpdf</code></a></strong><code>(vectors y | real nu, vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-ed2f0cb9d8bc8294e4c924451d7f556f82d5157c" class="index-entry"></a> <!-- real; multi_student_t_lupdf; (vectors y | real nu, vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lupdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lupdf</code></a></strong><code>(vectors y | real nu, vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-97bab7be0967824f68beadb09ab9fc5152e4b605" class="index-entry"></a> <!-- real; multi_student_t_lpdf; (vectors y | real nu, row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lpdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lpdf</code></a></strong><code>(vectors y | real nu, row_vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-74a88cd182b98cb2259da79ec60775f5d1f01084" class="index-entry"></a> <!-- real; multi_student_t_lupdf; (vectors y | real nu, row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lupdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lupdf</code></a></strong><code>(vectors y | real nu, row_vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-b0e0974e59e75e19f6b6c4aa22d8395d7ef922de" class="index-entry"></a> <!-- real; multi_student_t_lpdf; (row_vectors y | real nu, vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lpdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lpdf</code></a></strong><code>(row_vectors y | real nu, vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-7b3d937ea2e301a8d2455c2939ef7fc5c815d5ca" class="index-entry"></a> <!-- real; multi_student_t_lupdf; (row_vectors y | real nu, vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lupdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lupdf</code></a></strong><code>(row_vectors y | real nu, vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-5b6e5c7b1a5d8704f96553d2ccd4fb053bdb4917" class="index-entry"></a> <!-- real; multi_student_t_lpdf; (row_vectors y | real nu, row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lpdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lpdf</code></a></strong><code>(row_vectors y | real nu, row_vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-d92dadb3604f6521420ca422b762d25b60b94dbf" class="index-entry"></a> <!-- real; multi_student_t_lupdf; (row_vectors y | real nu, row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_lupdf" class="unlink" title="Jump to index entry"><code>multi_student_t_lupdf</code></a></strong><code>(row_vectors y | real nu, row_vectors mu, matrix Sigma)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-e2a99afdcb2cc3ec5ffc0367d7d241f2b511a575" class="index-entry"></a> <!-- vector; multi_student_t_rng; (real nu, vector mu, matrix Sigma); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_rng</code></a></strong><code>(real nu, vector mu, matrix Sigma)</code><br> Generate a multivariate Student-<span class="math inline">\(t\)</span> variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-18f756f159784f68dc00bac1706c5e31f805daf3" class="index-entry"></a> <!-- vector; multi_student_t_rng; (real nu, row_vector mu, matrix Sigma); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_rng</code></a></strong><code>(real nu, row_vector mu, matrix Sigma)</code><br> Generate a multivariate Student-<span class="math inline">\(t\)</span> variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-d65fa98b0f3f647ffd2cbe3f445ab17be93735dc" class="index-entry"></a> <!-- vectors; multi_student_t_rng; (real nu, vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_rng</code></a></strong><code>(real nu, vectors mu, matrix Sigma)</code><br> Generate an array of multivariate Student-<span class="math inline">\(t\)</span> variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-f05b8a3480e2360d1224acb270aca05c90674ba1" class="index-entry"></a> <!-- vectors; multi_student_t_rng; (real nu, row_vectors mu, matrix Sigma); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_rng</code></a></strong><code>(real nu, row_vectors mu, matrix Sigma)</code><br> Generate an array of multivariate Student-<span class="math inline">\(t\)</span> variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="multi-student-t-cholesky-fun" class="level2">
+<h2 class="anchored" data-anchor-id="multi-student-t-cholesky-fun">Multivariate Student-t distribution, Cholesky parameterization</h2>
+<section id="probability-density-function-6" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-6">Probability density function</h3>
+<p>Let <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(\nu \in \mathbb{R}^+\)</span>, <span class="math inline">\(\mu \in \mathbb{R}^K\)</span>, and <span class="math inline">\(L\)</span> a <span class="math inline">\(K \times K\)</span> lower-triangular matrix with strictly positive, finite diagonal then <span class="math display">\[\begin{equation*}
+\begin{array}{l}
+\text{MultiStudentTCholesky}(y\,\mid \nu,\,\mu,\,L) \\  =
+\frac{1}{\pi^{K/2}} \ \frac{1}{\nu^{K/2}} \ \frac{\Gamma\!\left((\nu +
+K)/2\right)}      {\Gamma(\nu/2)} \ \frac{1}{\left| L
+\right|} \ \left( 1 + \frac{1}{\nu} \, \left(y - \mu\right)^{\top} \,
+L^{-T}L^{-1} \, \left(y - \mu\right) \right)^{-(\nu + K)/2} \! .
+\end{array}
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-6" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-6">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_cholesky" class="unlink" title="Jump to index entry"><code>multi_student_t_cholesky</code></a></strong><code>(nu, mu, L)</code></p>
+<p>Increment target log probability density with <code>multi_student_t_cholesky_lupdf(y | nu, mu, L)</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-190447ef796a8d04715897b349d1433608a8a0e3" class="index-entry"></a> <!-- real; multi_student_t_cholesky ~; -->
+<p></p>
+</section>
+<section id="stan-functions-6" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-6">Stan functions</h3>
+<a id="index-entry-79368cb327eb8f4f3056a3e71077b92df2406595" class="index-entry"></a> <!-- real; multi_student_t_cholesky_lpdf; (vectors y | real nu, vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_cholesky_lpdf" class="unlink" title="Jump to index entry"><code>multi_student_t_cholesky_lpdf</code></a></strong><code>(vectors y | real nu, vectors mu, matrix L)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of vector or array of vectors <code>y</code> given degrees of freedom <code>nu</code>, location vector or array of vectors <code>mu</code>, and Cholesky factor of the scale matrix <code>L</code>. For a definition of the arguments compatible with the <code>vectors</code> type, see the <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">probability vectorization section</a>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-4b504cf934bd0448a4e3174a7a7b34f46cfa7d8c" class="index-entry"></a> <!-- real; multi_student_t_cholesky_lupdf; (vectors y | real nu, vectors mu, matrix L); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_cholesky_lupdf" class="unlink" title="Jump to index entry"><code>multi_student_t_cholesky_lupdf</code></a></strong><code>(vectors y | real nu, vectors mu, matrix L)</code><br> The log of the multivariate Student-<span class="math inline">\(t\)</span> density of vector or vector array <code>y</code> given degrees of freedom <code>nu</code>, location vector or vector array <code>mu</code>, and Cholesky factor of the scale matrix <code>L</code>, dropping constant additive terms. For a definition of arguments compatible with the <code>vectors</code> type, see the <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">probability vectorization section</a>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-37c54942aa5cd7ed73303ca868eeb65b156be282" class="index-entry"></a> <!-- vector; multi_student_cholesky_t_rng; (real nu, vector mu, matrix L); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_cholesky_rng</code></a></strong><code>(real nu, vector mu, matrix L)</code><br> Generate a multivariate Student-<span class="math inline">\(t\)</span> variate with degrees of freedom <code>nu</code>, location <code>mu</code>, and Cholesky factor of the scale matrix <code>L</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-da90fd4020ba98291c4e0c6ae45340c32e83bdc6" class="index-entry"></a> <!-- array[] vector; multi_student_t_cholesky_rng; (real nu, array[] vector mu, matrix L); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_cholesky_rng</code></a></strong><code>(real nu, array[] vector mu, matrix L)</code><br> Generate a multivariate Student-<span class="math inline">\(t\)</span> variate with degrees of freedom <code>nu</code>, location array <code>mu</code>, and Cholesky factor of the scale matrix <code>L</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-48a1c071b3760543bf2019dc6c37df56b26a25b4" class="index-entry"></a> <!-- array[] vector; multi_student_t_cholesky_rng; (real nu, array[] row_vector mu, matrix L); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#multi_student_t_cholesky_rng" class="unlink" title="Jump to index entry"><code>multi_student_t_cholesky_rng</code></a></strong><code>(real nu, array[] row_vector mu, matrix L)</code><br> Generate an array of multivariate Student-<span class="math inline">\(t\)</span> variate with degrees of freedom <code>nu</code>, location array <code>mu</code>, and Cholesky factor of the scale matrix <code>L</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="gaussian-dynamic-linear-models" class="level2">
+<h2 class="anchored" data-anchor-id="gaussian-dynamic-linear-models">Gaussian dynamic linear models</h2>
+<p>A Gaussian Dynamic Linear model is defined as follows, For <span class="math inline">\(t \in 1,
+\dots, T\)</span>, <span class="math display">\[\begin{equation*}   \begin{aligned}[t]     y_{t} &amp;\sim N(F' \theta_{t}, V)
+\\     \theta_{t} &amp;\sim N(G \theta_{t - 1}, W) \\     \theta_{0} &amp;\sim
+N(m_{0}, C_{0})   \end{aligned} \end{equation*}\]</span> where <span class="math inline">\(y\)</span> is <span class="math inline">\(n \times T\)</span> matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (<span class="math inline">\(p(y | F, G, V, W, m_{0}, C_{0})\)</span>). This log-density is a system that is calculated using the Kalman Filter. If <span class="math inline">\(V\)</span> is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used <span class="citation" data-cites="DurbinKoopman:2001">(<a href="#ref-DurbinKoopman:2001" role="doc-biblioref">Durbin and Koopman 2001, sec. 6.4</a>)</span>.</p>
+<section id="distribution-statement-7" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-7">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#gaussian_dlm_obs" class="unlink" title="Jump to index entry"><code>gaussian_dlm_obs</code></a></strong><code>(F, G, V, W, m0, C0)</code></p>
+<p>Increment target log probability density with <code>gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4fd242bf2dcd5be49146d526d8fbe39878489a73" class="index-entry"></a> <!-- real; gaussian_dlm_obs ~; -->
+<p></p>
+</section>
+<section id="stan-functions-7" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
+<p>The following two functions differ in the type of their V, the first taking a full observation covariance matrix V&nbsp;and the second a vector V&nbsp;representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.</p>
+<a id="index-entry-18a51c9f7a0d2635dc8adf29e702d2a918af65ed" class="index-entry"></a> <!-- real; gaussian_dlm_obs_lpdf; (matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gaussian_dlm_obs_lpdf" class="unlink" title="Jump to index entry"><code>gaussian_dlm_obs_lpdf</code></a></strong><code>(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0)</code><br> The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-c55d98e7e02d60b3177dcb12f3f8a4d71d7b7770" class="index-entry"></a> <!-- real; gaussian_dlm_obs_lupdf; (matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gaussian_dlm_obs_lupdf" class="unlink" title="Jump to index entry"><code>gaussian_dlm_obs_lupdf</code></a></strong><code>(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0)</code><br> The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8cca79a125123f82ff2e6982bc58099c4b4f2db4" class="index-entry"></a> <!-- real; gaussian_dlm_obs_lpdf; (matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gaussian_dlm_obs_lpdf" class="unlink" title="Jump to index entry"><code>gaussian_dlm_obs_lpdf</code></a></strong><code>(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0)</code><br> The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-14842eb0078230e2e500d5478fb688e1fd8c7fd3" class="index-entry"></a> <!-- real; gaussian_dlm_obs_lupdf; (matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gaussian_dlm_obs_lupdf" class="unlink" title="Jump to index entry"><code>gaussian_dlm_obs_lupdf</code></a></strong><code>(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0)</code><br> The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-DurbinKoopman:2001" class="csl-entry" role="listitem">
+Durbin, J., and S. J. Koopman. 2001. <em>Time Series Analysis by State Space Methods</em>. New York: Oxford University Press.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/bounded_continuous_distributions.html" class="pagination-link" aria-label="Bounded Continuous Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Bounded Continuous Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/simplex_distributions.html" class="pagination-link" aria-label="Simplex Distributions">
+        <span class="nav-page-text">Simplex Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/distributions_over_unbounded_vectors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/embedded_laplace.html b/docs/2_39/functions-reference/embedded_laplace.html
new file mode 100644
index 000000000..e3e23970e
--- /dev/null
+++ b/docs/2_39/functions-reference/embedded_laplace.html
@@ -0,0 +1,1517 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Embedded Laplace Approximation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/mathematical_functions.html" rel="next">
+<link href="../functions-reference/hidden_markov_models.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Embedded Laplace Approximation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/embedded_laplace.html">Embedded Laplace Approximation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#embedded-laplace-approximation" id="toc-embedded-laplace-approximation" class="nav-link active" data-scroll-target="#embedded-laplace-approximation">Embedded Laplace Approximation</a>
+  <ul class="collapse">
+  <li><a href="#approximating-the-log-marginal-likelihood-log-py-mid-phi" id="toc-approximating-the-log-marginal-likelihood-log-py-mid-phi" class="nav-link" data-scroll-target="#approximating-the-log-marginal-likelihood-log-py-mid-phi">Approximating the log marginal likelihood <span class="math inline">\(\log p(y \mid \phi)\)</span></a></li>
+  <li><a href="#laplace-likelihood_spec" id="toc-laplace-likelihood_spec" class="nav-link" data-scroll-target="#laplace-likelihood_spec">Specifying the log likelihood function</a></li>
+  <li><a href="#specifying-the-covariance-function" id="toc-specifying-the-covariance-function" class="nav-link" data-scroll-target="#specifying-the-covariance-function">Specifying the covariance function</a></li>
+  <li><a href="#control-parameters" id="toc-control-parameters" class="nav-link" data-scroll-target="#control-parameters">Control parameters</a></li>
+  <li><a href="#sample-from-the-approximate-conditional-hatptheta-mid-y-phi" id="toc-sample-from-the-approximate-conditional-hatptheta-mid-y-phi" class="nav-link" data-scroll-target="#sample-from-the-approximate-conditional-hatptheta-mid-y-phi">Sample from the approximate conditional <span class="math inline">\(\hat{p}(\theta \mid y, \phi)\)</span></a></li>
+  <li><a href="#built-in-laplace-marginal-likelihood-functions" id="toc-built-in-laplace-marginal-likelihood-functions" class="nav-link" data-scroll-target="#built-in-laplace-marginal-likelihood-functions">Built-in Laplace marginal likelihood functions</a>
+  <ul class="collapse">
+  <li><a href="#poisson-with-log-link" id="toc-poisson-with-log-link" class="nav-link" data-scroll-target="#poisson-with-log-link">Poisson with log link</a></li>
+  <li><a href="#negative-binomial-with-log-link" id="toc-negative-binomial-with-log-link" class="nav-link" data-scroll-target="#negative-binomial-with-log-link">Negative Binomial with log link</a></li>
+  <li><a href="#bernoulli-with-logit-link" id="toc-bernoulli-with-logit-link" class="nav-link" data-scroll-target="#bernoulli-with-logit-link">Bernoulli with logit link</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/embedded_laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/embedded_laplace.html">Embedded Laplace Approximation</a></li></ol></nav></header>
+
+
+
+
+<section id="embedded-laplace-approximation" class="level1">
+<h1>Embedded Laplace Approximation</h1>
+<p>The embedded Laplace approximation can be used to approximate certain marginal and conditional distributions that arise in latent Gaussian models. Embedded Laplace replaces explicit sampling of (high-dimensional) Gaussian latent variables with a local Gaussian approximation. In doing so, it marginalizes out the latent Gaussian variables. Inference can then be performed on the remaining, often low-dimensional, parameters. The embedded Laplace approximation in Stan is best suited for latent Gaussian models when jointly sampling over all model parameters is expensive and the conditional posterior of the Gaussian latent variables is reasonably close to Gaussian.</p>
+<p>For observed data <span class="math inline">\(y\)</span>, latent Gaussian variables <span class="math inline">\(\theta\)</span>, and hyperparameters <span class="math inline">\(\phi\)</span>, a latent Gaussian model observes the following hierarchical structure: <span class="math display">\[\begin{eqnarray}
+  \phi &amp;\sim&amp; p(\phi), \\
+  \theta &amp;\sim&amp; \text{MultiNormal}(0, K(\phi)), \\
+  y &amp;\sim&amp; p(y \mid \theta, \phi).
+\end{eqnarray}\]</span> In this formulation, <span class="math inline">\(p(y \mid \theta, \phi)\)</span> is the data model that specifies how observations are generated conditional on <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span>. <span class="math inline">\(K(\phi)\)</span> denotes the prior covariance matrix for the latent Gaussian variables <span class="math inline">\(\theta\)</span> and is parameterized by <span class="math inline">\(\phi\)</span>. The prior on <span class="math inline">\(\theta\)</span> is centered at 0, however an offset can always be added when specifying the data model <span class="math inline">\(p(y \mid \theta, \phi)\)</span>.</p>
+<p>Conditioning on observations <span class="math inline">\(y\)</span> we obtain the joint posterior <span class="math inline">\(p(\phi, \theta \mid y) \propto p(y \mid \theta, \phi) p(\theta |
+\phi) p(\phi)\)</span>, where <span class="math inline">\(p(y \mid \theta, \phi)\)</span> as function of <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span> is the likelihood function. To sample from the joint posterior, we can either use a standard method, such as Markov chain Monte Carlo, or we can follow a two-step procedure:</p>
+<ol type="1">
+<li>sample from the <em>marginal posterior</em> <span class="math inline">\(p(\phi \mid y)\)</span>,</li>
+<li>sample from the <em>conditional posterior</em> <span class="math inline">\(p(\theta \mid y, \phi)\)</span>.</li>
+</ol>
+<p>In the above procedure, neither the marginal posterior nor the conditional posterior are typically available in closed form and so they must be approximated. The marginal posterior can be written as <span class="math inline">\(p(\phi \mid y) \propto p(y \mid \phi) p(\phi)\)</span>, where <span class="math inline">\(p(y \mid \phi) = \int p(y \mid \phi, \theta) p(\theta) \text{d}\theta\)</span> is called the marginal likelihood. The Laplace method approximates <span class="math inline">\(p(y \mid \phi, \theta) p(\theta)\)</span> with a normal distribution centered at the mode, <span class="math display">\[
+  \theta^* = \underset{\theta}{\text{argmax}} \ \log p(\theta \mid y, \phi),
+\]</span> and <span class="math inline">\(\theta^*\)</span> is obtained using a numerical optimizer. The resulting Gaussian integral can be evaluated analytically to obtain an approximation to the log marginal likelihood <span class="math inline">\(\log \hat p(y \mid \phi) \approx \log p(y \mid \phi)\)</span>. Specifically: <span class="math display">\[
+  \hat p(y \mid \phi) = \frac{p(\theta^* \mid \phi) p(y \mid \theta^*, \phi)}{\hat p (\theta^* \mid \phi, y)}.
+\]</span></p>
+<p>Combining this marginal likelihood with the prior in the <code>model</code> block, we can then sample from the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span> using one of Stan’s algorithms. The marginal posterior is lower dimensional and likely to have a simpler geometry leading to more efficient inference. On the other hand each marginal likelihood computation is more costly, and the combined change in efficiency depends on the application.</p>
+<p>To obtain posterior draws for <span class="math inline">\(\theta\)</span>, we sample from the normal approximation to <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in <code>generated quantities</code>. The process of iteratively sampling from <span class="math inline">\(p(\phi \mid y)\)</span> (say, with MCMC) and then <span class="math inline">\(p(\theta \mid y, \phi)\)</span> produces posterior draws from the joint posterior <span class="math inline">\(p(\theta, \phi \mid y)\)</span>.</p>
+<p>The Laplace approximation is especially useful if <span class="math inline">\(p(y \mid \phi, \theta)\)</span> as function of <span class="math inline">\(\theta\)</span> is log-concave, e.g., in case of Poisson, binomial, negative-binomial, and Bernoulli. (The likelihood of normal model is also log concave, however when the likelihood is normal, marginalization can be performed exactly and does not required an approximation.) Stan’s embedded Laplace approximation is restricted to the case where the prior <span class="math inline">\(p(\theta \mid \phi)\)</span> is multivariate normal. Furthermore, the likelihood <span class="math inline">\(p(y \mid \phi, \theta)\)</span> must be computed using only operations which support higher-order derivatives (see section <a href="#laplace_likelihood_spec">specifying the likelihood function</a>).</p>
+<p>The Laplace approximation can also be useful in generated quantities to marginalize out latent variables even if the sampling had been done using the full joint posterior.</p>
+<section id="approximating-the-log-marginal-likelihood-log-py-mid-phi" class="level2">
+<h2 class="anchored" data-anchor-id="approximating-the-log-marginal-likelihood-log-py-mid-phi">Approximating the log marginal likelihood <span class="math inline">\(\log p(y \mid \phi)\)</span></h2>
+<p>In the <code>model</code> block, we increment <code>target</code> with <code>laplace_marginal</code>, a function that approximates the log marginal likelihood <span class="math inline">\(\log p(y \mid \phi)\)</span>. The signature of the function is:</p>
+<a id="index-entry-7ffad9885f7c8f209fffbfae2182cbc3ac432b59" class="index-entry"></a> <!-- real; laplace_marginal; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal" class="unlink" title="Jump to index entry"><code>laplace_marginal</code></a></strong><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code></p>
+<p>which returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span>.</p>
+<p>The embedded Laplace functions accept two functors whose user defined arguments are passed in as tuples to <code>laplace_marginal</code>.</p>
+<ol type="1">
+<li><code>likelihood_function</code> - user-specified log likelihood whose first argument is the vector of latent Gaussian variables <span class="math inline">\(\theta\)</span>. The subsequent arguments are user defined.</li>
+</ol>
+<ul>
+<li><code>real likelihood_function(vector theta, likelihood_arguments_1, likelihood_arguments_2, ...)</code>.</li>
+</ul>
+<ol start="2" type="1">
+<li><code>likelihood_arguments</code> - A tuple of arguments whose internal members are be passed to the log likelihood function. This tuple does NOT include the latent variable <span class="math inline">\(\theta\)</span>.</li>
+<li><code>hessian_block_size</code> - the block size of the Hessian of the log likelihood, <span class="math inline">\(\partial^2 \log p(y \mid \theta, \phi) / \partial \theta^2\)</span>.</li>
+<li><code>covariance_function</code> - A function that returns the covariance matrix of the multivariate normal prior on <span class="math inline">\(\theta\)</span>.</li>
+</ol>
+<ul>
+<li><code>matrix covariance_function(covariance_argument_1, covariance_argument_2, ...)</code>.</li>
+</ul>
+<ol start="5" type="1">
+<li><code>covariance_arguments</code> A tuple of the arguments whose internal members will be passed to the the covariance function.</li>
+</ol>
+<small><i>Available since 2.39</i></small>
+<p>Below we go over each argument in more detail.</p>
+</section>
+<section id="laplace-likelihood_spec" class="level2">
+<h2 class="anchored" data-anchor-id="laplace-likelihood_spec">Specifying the log likelihood function</h2>
+<p>The first step to use the embedded Laplace approximation is to write down a function in the <code>functions</code> block which returns the log likelihood <span class="math inline">\(\log p(y \mid \theta, \phi)\)</span>.</p>
+<p>There are a few constraints on this function:</p>
+<ol type="1">
+<li><p>The function return type must be <code>real</code>.</p></li>
+<li><p>The first argument must be the latent Gaussian variable <span class="math inline">\(\theta\)</span> and must have type <code>vector</code>.</p></li>
+<li><p>The operations in the function must support higher-order automatic differentiation (AD). Most functions in Stan support higher-order AD. The exceptions are functions with specialized calls for reverse-mode AD, and these are higher-order functions (algebraic solvers, differential equation solvers, and integrators), the marginalization function for hidden Markov models (HMM) function, and the embedded Laplace approximation itself.</p></li>
+</ol>
+<p>The base signature of the function is</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> likelihood_function(<span class="dt">vector</span> theta, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>...</code> represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments <code>...</code> and each argument can be passed as data or parameter.</p>
+<p>The tuple after <code>likelihood_function</code> contains the arguments that get passed to <code>likelihood_function</code> <em>excluding <span class="math inline">\(\theta\)</span></em>. For instance, if a user defined likelihood uses a real and a matrix, the likelihood function’s signature would first have a vector and then a real and matrix argument.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> likelihood_fun(<span class="dt">vector</span> theta, <span class="dt">real</span> a, <span class="dt">matrix</span> X)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The call to the laplace marginal would start with this likelihood and tuple holding the other likelihood arguments. We do not need to pass <code>theta</code>, since it is marginalized out and therefore does not appear explicitly as a passed parameter.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> val = laplace_marginal(likelihood_fun, (a, X), hessian_block_size, ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the <code>likelihood_function</code> has only one argument, the tuple syntax is <code>(a, )</code>.</p>
+<p>As always, users should use parameter arguments only when necessary in order to speed up differentiation. In general, we recommend marking data only arguments with the keyword <code>data</code>, for example,</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> likelihood_function(<span class="dt">vector</span> theta, <span class="kw">data</span> <span class="dt">vector</span> x, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In addition to the likelihood function, users must specify the block size of the Hessian, <span class="math inline">\(\partial^2 \log p(y \mid \theta, \phi) / \partial \theta^2\)</span>. The Hessian is often block diagonal and this structure can be taken advantage of for fast computation. For example, if <span class="math inline">\(y_i\)</span> only depends on <span class="math inline">\(\theta_i\)</span>, then the Hessian is diagonal and <code>hessian_block_size=1</code>,</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> val = laplace_marginal(likelihood_fun, (a, X), <span class="dv">1</span>, ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>On the other hand, if the Hessian is not block diagonal, we can always set <code>hessian_block_size=n</code> where <span class="math inline">\(n\)</span> is the size of <span class="math inline">\(\theta\)</span>.</p>
+</section>
+<section id="specifying-the-covariance-function" class="level2">
+<h2 class="anchored" data-anchor-id="specifying-the-covariance-function">Specifying the covariance function</h2>
+<p>The argument <code>covariance_function</code> returns the prior covariance matrix <span class="math inline">\(K\)</span>. The signature for this function is the same as a standard stan function. It’s return type must be a matrix of size <span class="math inline">\(n \times n\)</span> where <span class="math inline">\(n\)</span> is the size of <span class="math inline">\(\theta\)</span>.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> covariance_function(...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>...</code> represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments <code>...</code> and each argument can be passed as data or parameter. The variables <span class="math inline">\(\phi\)</span> is implicitly defined as the collection of all non-data arguments passed to <code>likelihood_function</code> (excluding <span class="math inline">\(\theta\)</span>) and <code>covariance_function</code>.</p>
+<p>The tuple after <code>covariance_function</code> contains the arguments that get passed to <code>covariance_function</code>. For instance, if a user defined covariance function uses two vectors</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> cov_fun(<span class="dt">real</span> b, <span class="dt">matrix</span> Z)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>the call to the Laplace marginal would include the covariance function and a tuple holding the covariance function arguments.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> val = laplace_marginal(likelihood_fun, (a, X), cov_fun, (b, Z), ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the <code>covariance_function</code> has only one argument, the tuple syntax is <code>(b, )</code>.</p>
+</section>
+<section id="control-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="control-parameters">Control parameters</h2>
+<p>It also possible to specify control parameters, which can help improve the optimization that underlies the Laplace approximation, using <code>laplace_marginal_tol</code> with the following signature:</p>
+<a id="index-entry-9d40f0ee623d8ad92f054b2d83550d7a8ddf4b56" class="index-entry"></a> <!-- real; laplace_marginal_tol; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol</code></a></strong><code>(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...),  tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>The final argument, tolerances, is a tuple with the following elements</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">vector</span> theta_init, <span class="dt">real</span> tol, <span class="dt">int</span> max_steps, <span class="dt">int</span> solver,</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> max_steps_linesearch, <span class="dt">int</span> allow_fallback)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> and allows the user to tune the control parameters of the approximation.</p>
+<ul>
+<li><p><code>theta_init</code>: the initial guess for a Newton solver when finding the mode of <span class="math inline">\(p(\theta \mid y, \phi)\)</span>. By default, it is a zero-vector.</p></li>
+<li><p><code>tol</code>: the tolerance <span class="math inline">\(\epsilon\)</span> of the optimizer. Specifically, the optimizer stops when <span class="math inline">\(||\nabla \log p(\theta \mid y, \phi)|| \le \epsilon\)</span>. By default, the value is <span class="math inline">\(\epsilon \approx 1.49 \times 10^{-8}\)</span>, which is the square-root of machine precision.</p></li>
+<li><p><code>max_num_steps</code>: the maximum number of steps taken by the optimizer before it gives up (in which case the Metropolis proposal gets rejected). The default is 500 steps.</p></li>
+<li><p><code>solver</code>: choice of Newton solver. The optimizer underlying the Laplace approximation does one of three matrix decompositions to compute a Newton step. The problem determines which decomposition is numerically stable. By default (<code>solver=1</code>), the solver attempts a Cholesky decomposition of the negative Hessian of the log likelihood, <span class="math inline">\(- \partial^2 \log p(y \mid \theta, \phi) / \partial^2 \theta\)</span>. This operation is legal if the negative Hessian is positive-definite, which will always be true when the likelihood as function of <span class="math inline">\(\theta\)</span> is log concave. If <code>solver=2</code>, the solver makes a Cholesky decomposition of the covariance matrix <span class="math inline">\(K(\phi)\)</span>. Since a covariance matrix is always positive-definite, computing its Cholesky decomposition is always a legal operation, at least in theory. In practice, we may not be able to compute the Cholesky decomposition of the negative Hessian nor of the covariance matrix, either because it does not exist or because of numerical issues. In that case, we can use <code>solver=3</code> which uses a more expensive but less specialized approach to compute a Newton step.</p></li>
+<li><p><code>max_steps_linesearch</code>: maximum number of steps in linesearch. The linesearch adjusts to step size to ensure that a Newton step leads to an increase in the objective function (i.e., <span class="math inline">\(f(\theta) = p(\theta \mid \phi, y)\)</span>). If a standard Newton step does not improve the objective function, the step is adjusted iteratively until the objective function increases or the maximum number of steps in the linesearch is reached. By default, <code>max_steps_linesearch=1000</code>. Setting <code>max_steps_linesearch=0</code> results in no linesearch.</p></li>
+<li><p><code>allow_fallback</code>: If user set solver fails, this flag determines whether to fallback to the next solver. For example, if the user specifies <code>solver=1</code> but the Cholesky decomposition of the negative Hessian <span class="math inline">\(- \partial^2 \log p(y \mid \theta, \phi) / \partial^2 \theta\)</span> fails, the optimizer will try <code>solver=2</code> instead. By default, <code>allow_fallback = 1</code> (TRUE).</p></li>
+</ul>
+<small><i>Available since 2.39</i></small>
+<p>The embedded Laplace approximation’s options have a helper callable <code>generate_laplace_options(int theta_size)</code> that will generate the tuple for the user. This can be useful for quickly setting up the control parameters in the <code>transformed data</code> block to reuse within the model.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">vector</span>[theta_size], <span class="dt">real</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>) laplace_ops = generate_laplace_options(theta_size);</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a><span class="co">// Modify solver type</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>laplace_ops<span class="fl">.5</span> = <span class="dv">2</span>;</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a><span class="co">// Turn off fallthrough</span></span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>laplace_ops<span class="fl">.7</span> = <span class="dv">0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<a id="index-entry-60827bf41179724b5e6e8a0fc73da3515b44fc46" class="index-entry"></a> <!-- tuple(vector, real, int, int, int, int); generate_laplace_options; (int dimension); -->
+<p></p>
+<p><code>tuple(vector, real, int, int, int, int)</code> <strong><a href="../functions-reference/functions_index.html#generate_laplace_options" class="unlink" title="Jump to index entry"><code>generate_laplace_options</code></a></strong><code>(int dimension)</code><br></p>
+<p>Create a default laplace options tuple for a <code>theta_init</code> of size <code>dimension</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-d1c421fc2e179180c0a9a45ea4733480792940bc" class="index-entry"></a> <!-- tuple(vector, real, int, int, int, int); generate_laplace_options; (vector theta_init); -->
+<p></p>
+<p><code>tuple(vector, real, int, int, int, int)</code> <strong><a href="../functions-reference/functions_index.html#generate_laplace_options" class="unlink" title="Jump to index entry"><code>generate_laplace_options</code></a></strong><code>(vector theta_init)</code><br></p>
+<p>Create a default Laplace options tuple containing <code>theta_init</code>.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="sample-from-the-approximate-conditional-hatptheta-mid-y-phi" class="level2">
+<h2 class="anchored" data-anchor-id="sample-from-the-approximate-conditional-hatptheta-mid-y-phi">Sample from the approximate conditional <span class="math inline">\(\hat{p}(\theta \mid y, \phi)\)</span></h2>
+<p>In <code>generated quantities</code>, it is possible to sample from the Laplace approximation of <span class="math inline">\(p(\theta \mid \phi, y)\)</span> using <code>laplace_latent_rng</code>. The signature for <code>laplace_latent_rng</code> follows closely the signature for <code>laplace_marginal</code>:</p>
+<a id="index-entry-60d4efd35a460871aff97a758f5d366faee26465" class="index-entry"></a> <!-- vector; laplace_latent_rng; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_rng</code></a></strong><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Samples from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span>.</p>
+<small><i>Available since 2.39</i></small>
+<p>Once again, it is possible to specify control parameters:</p>
+<a id="index-entry-eb06251dd7ccd5f9b7882b7e6950fa99cc0c2d57" class="index-entry"></a> <!-- vector; laplace_latent_rng_tol; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><code>laplace_latent_tol_rng</code></strong><code>(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...), tuple(vector, real, int, int, int, int) tolerances)</code><br> Samples from the approximate conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="built-in-laplace-marginal-likelihood-functions" class="level2">
+<h2 class="anchored" data-anchor-id="built-in-laplace-marginal-likelihood-functions">Built-in Laplace marginal likelihood functions</h2>
+<p>Stan provides convenient wrappers for the embedded Laplace approximation when applied to latent Gaussian models with certain likelihoods arising from some common data models. With this wrapper, the likelihood is pre-specified and does not need to be specified by the user. The selection of supported likelihoods is currently narrow and expected to grow. The wrappers exist for the user’s convenience but are not more computationally efficient than specifying log likelihoods in the <code>functions</code> block.</p>
+<section id="poisson-with-log-link" class="level3">
+<h3 class="anchored" data-anchor-id="poisson-with-log-link">Poisson with log link</h3>
+<p>Given count data, with each observed count <span class="math inline">\(y_i\)</span> associated with a group <span class="math inline">\(g(i)\)</span> and a corresponding latent variable <span class="math inline">\(\theta_{g(i)}\)</span>, and a Poisson model, the likelihood is <span class="math display">\[
+p(y \mid \theta, \phi) = \prod_i\text{Poisson} (y_i \mid \exp(\theta_{g(i)} + m_{g(i)})),
+\]</span> where <span class="math inline">\(m_{g(i)}\)</span> acts as an offset for <span class="math inline">\(\theta_{g(i)}\)</span>. This can also be interpreted as a prior mean on <span class="math inline">\(\theta_{g(i)}\)</span>. The arguments required to compute this likelihood are:</p>
+<ul>
+<li><code>y</code>: an array of counts.</li>
+<li><code>y_index</code>: an array whose <span class="math inline">\(i^\text{th}\)</span> element indicates to which group the <span class="math inline">\(i^\text{th}\)</span> observation belongs to.</li>
+<li><code>m</code>: a vector of offsets or prior means for <span class="math inline">\(\theta\)</span>.</li>
+</ul>
+<a id="index-entry-b2c6946b65561039eb8bf3999fb59d38a6336e10" class="index-entry"></a> <!-- real; laplace_marginal_poisson_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_poisson_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_poisson_log</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-4a7ecca812a308ea648bb31996559cba79738d83" class="index-entry"></a> <!-- real; laplace_marginal_tol_poisson_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_poisson_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_poisson_log</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_tol_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covaraince_arguments, tolerances)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<p>The signatures for the embedded Laplace approximation function with a Poisson likelihood are</p>
+<a id="index-entry-ca1a086a83293f252043ab8b127ec6c9ef16deec" class="index-entry"></a> <!-- real; laplace_marginal_poisson_log_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments; -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_poisson_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_poisson_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-51cad53596dd0d74c0fe99b84a3c975cbec3866a" class="index-entry"></a> <!-- real; laplace_marginal_tol_poisson_log_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_poisson_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_poisson_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-d0ecc660f5107f7409bd76ba059c5ea82ec81ba0" class="index-entry"></a> <!-- real; laplace_marginal_poisson_log_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments; -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_poisson_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_poisson_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-23e9ed91fd3f98c314ddbce86a77005fdf8ed5f3" class="index-entry"></a> <!-- real; laplace_marginal_tol_poisson_log_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_poisson_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_poisson_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-a6bf063f06956735a484fd5e960352833373ec15" class="index-entry"></a> <!-- vector; laplace_latent_poisson_log_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p> <code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_poisson_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_poisson_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br> Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-8fec77181949b3ca47b1916246cd0384394f87bc" class="index-entry"></a> <!-- vector; laplace_latent_tol_poisson_log_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_tol_poisson_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_tol_poisson_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="negative-binomial-with-log-link" class="level3">
+<h3 class="anchored" data-anchor-id="negative-binomial-with-log-link">Negative Binomial with log link</h3>
+<p>The negative Binomial distribution generalizes the Poisson distribution by introducing the dispersion parameter <span class="math inline">\(\eta\)</span>. The corresponding likelihood is then <span class="math display">\[
+p(y \mid \theta, \phi) = \prod_i\text{NegBinomial2} (y_i \mid \exp(\theta_{g(i)} + m_{g(i)}), \eta).
+\]</span> Here we use the alternative parameterization implemented in Stan, meaning that <span class="math display">\[
+\mathbb E(y_i) = \exp (\theta_{g(i)} + m_{g(i)}), \\
+\text{Var}(y_i) = \mathbb E(y_i) + \frac{(\mathbb E(y_i))^2}{\eta}.
+\]</span> The arguments for the likelihood function are:</p>
+<ul>
+<li><code>y</code>: the observed counts</li>
+<li><code>y_index</code>: an array whose <span class="math inline">\(i^\text{th}\)</span> element indicates to which group the <span class="math inline">\(i^\text{th}\)</span> observation belongs to.</li>
+<li><code>eta</code>: the overdispersion parameter.</li>
+<li><code>m</code>: a vector of offsets or prior means for <span class="math inline">\(\theta\)</span>.</li>
+</ul>
+<a id="index-entry-7f807083cfb7347ece100da5dfe719186a678d02" class="index-entry"></a> <!-- real; laplace_marginal_neg_binomial_2_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_neg_binomial_2_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_neg_binomial_2_log</code></a></strong><code>(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-40016f7d5d96b9d999ca42f446720cf3f3cd5769" class="index-entry"></a> <!-- real; laplace_marginal_tol_neg_binomial_2_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_neg_binomial_2_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_neg_binomial_2_log</code></a></strong><code>(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_tol_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<p>The function signatures for the embedded Laplace approximation with a negative Binomial likelihood are</p>
+<a id="index-entry-b74e5e572a36211903f16f5deed7158384c21355" class="index-entry"></a> <!-- real; laplace_marginal_neg_binomial_2_log_lpmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_neg_binomial_2_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_neg_binomial_2_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-cefee8b5275ca255f121386fb14c9354cf95594c" class="index-entry"></a> <!-- real; laplace_marginal_tol_neg_binomial_2_log_lpmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_neg_binomial_2_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_neg_binomial_2_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-5d9163f94a110e178c54b2bd192e9aded134cd3a" class="index-entry"></a> <!-- real; laplace_marginal_neg_binomial_2_log_lupmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_neg_binomial_2_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_neg_binomial_2_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, function covariance_function, data int hessian_block_size, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-efb046fcbadbcaaa59a274ab3c5b1ed15f23add9" class="index-entry"></a> <!-- real; laplace_marginal_tol_neg_binomial_2_log_lupmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_neg_binomial_2_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_neg_binomial_2_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-79fe43b1435d4b3821038d2cd934d8c136e8808c" class="index-entry"></a> <!-- vector; laplace_latent_neg_binomial_2_log_rng; (array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_neg_binomial_2_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_neg_binomial_2_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative binomial distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-259ff7a39109d52d1194cf72a125b72127ddd2b1" class="index-entry"></a> <!-- vector; laplace_latent_tol_neg_binomial_2_log_rng; (array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_tol_neg_binomial_2_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_tol_neg_binomial_2_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative binomial distribution with a log link and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="bernoulli-with-logit-link" class="level3">
+<h3 class="anchored" data-anchor-id="bernoulli-with-logit-link">Bernoulli with logit link</h3>
+<p>Given binary outcome <span class="math inline">\(y_i \in \{0, 1\}\)</span> and Bernoulli model, the likelihood is <span class="math display">\[
+p(y \mid \theta, \phi) = \prod_i\text{Bernoulli} (y_i \mid \text{logit}^{-1}(\theta_{g(i)} + m_{g(i)})).
+\]</span> The arguments of the likelihood function are:</p>
+<ul>
+<li><code>y</code>: the observed counts</li>
+<li><code>y_index</code>: an array whose <span class="math inline">\(i^\text{th}\)</span> element indicates to which group the <span class="math inline">\(i^\text{th}\)</span> observation belongs to.</li>
+<li><code>m</code>: a vector of offsets or prior means for <span class="math inline">\(\theta\)</span>.</li>
+</ul>
+<a id="index-entry-1d93ab799518d0aac88e63c01f9655f36c7cbeb6" class="index-entry"></a> <!-- real; laplace_marginal_bernoulli_logit ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_bernoulli_logit" class="unlink" title="Jump to index entry"><code>laplace_marginal_bernoulli_logit</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-92d43f7c3643c7d85966bbfc0b2a71546facb37c" class="index-entry"></a> <!-- real; laplace_marginal_tol_bernoulli_logit ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_bernoulli_logit" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_bernoulli_logit</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_tol_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<p>The function signatures for the embedded Laplace approximation with a Bernoulli likelihood are</p>
+<a id="index-entry-fd129c664f37d1cdd280abaf7ee82bfaf9c8e54e" class="index-entry"></a> <!-- real; laplace_marginal_bernoulli_logit_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_bernoulli_logit_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_bernoulli_logit_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-5126f7d7829072e478ea2f7257d226219b43e1d4" class="index-entry"></a> <!-- real; laplace_marginal_tol_bernoulli_logit_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_bernoulli_logit_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_bernoulli_logit_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link and allows the user to tune the control parameters.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-53e5e8065cf24c3e72a435d5fccf4d14a587b245" class="index-entry"></a> <!-- real; laplace_marginal_bernoulli_logit_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_bernoulli_logit_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_bernoulli_logit_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-7d93f40162970812a81fc61df49dcb53f993e293" class="index-entry"></a> <!-- real; laplace_marginal_tol_bernoulli_logit_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_bernoulli_logit_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_bernoulli_logit_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link and allows the user to tune the control parameters, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-59f1ae97dbd720c3dd7cd15f92e7e3c6f93ae22e" class="index-entry"></a> <!-- vector; laplace_latent_bernoulli_logit_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_bernoulli_logit_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_bernoulli_logit_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Bernoulli distribution with a logit link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-4566ad7c6cd5f0ee057539a6e2a0bea28554f082" class="index-entry"></a> <!-- vector; laplace_latent_tol_bernoulli_logit_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_tol_bernoulli_logit_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_tol_bernoulli_logit_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Bernoulli distribution with a logit link, and lets the user tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+<!-- ## Draw approximate samples for out-of-sample latent variables. -->
+<!-- In many applications, it is of interest to draw latent variables for -->
+<!-- in-sample and out-of-sample predictions. We respectively denote these latent -->
+<!-- variables $\theta$ and $\theta^*$. In a latent Gaussian model,  -->
+<!-- $(\theta, \theta^*)$ jointly follow a prior multivariate normal distribution: -->
+<!-- $$ -->
+<!--   \theta, \theta^* \sim \text{MultiNormal}(0, {\bf K}(\phi)), -->
+<!-- $$ -->
+<!-- where $\bf K$ designates the joint covariance matrix over $\theta, \theta^*$. -->
+<!-- We can break $\bf K$ into three components, -->
+<!-- $$ -->
+<!-- {\bf K} = \begin{bmatrix} -->
+<!--   K & \\ -->
+<!--   K^* & K^{**} -->
+<!-- \end{bmatrix}, -->
+<!-- $$ -->
+<!-- where $K$ is the prior covariance matrix for $\theta$, $K^{**}$ the prior -->
+<!-- covariance matrix for $\theta^*$, and $K^*$ the covariance matrix between -->
+<!-- $\theta$ and $\theta^*$. -->
+<!-- Stan supports the case where $\theta$ is associated with an in-sample -->
+<!-- covariate $X$ and $\theta^*$ with an out-of-sample covariate $X^*$. -->
+<!-- Furthermore, the covariance function is written in such a way that -->
+<!-- $$ -->
+<!-- K = f(..., X, X), \ \ K^{**} = f(..., X^*, X^*), \ \ K^* = f(..., X, X^*), -->
+<!-- $$ -->
+<!-- as is typically the case in Gaussian process models. -->
+<!-- The -->
+<!-- function `laplace_latent_rng` samples from the Laplace approximation -->
+<!-- and admits nearly the same arguments as `laplace_marginal`. A key difference -->
+<!-- is that  -->
+<!-- ``` -->
+<!-- vector laplace_latent_rng(function likelihood_function, tuple(...), vector theta_0,  -->
+<a id="index-entry-770ccdd326df6b55b9b82f28548a93c005c11324" class="index-entry"></a> <!--                           function covariance_function, tuple(...)); -->
+<!-- ``` -->
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/hidden_markov_models.html" class="pagination-link" aria-label="Hidden Markov Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Hidden Markov Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/mathematical_functions.html" class="pagination-link" aria-label="Mathematical Functions">
+        <span class="nav-page-text">Mathematical Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/embedded_laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/functions_index.html b/docs/2_39/functions-reference/functions_index.html
new file mode 100644
index 000000000..cf99a8514
--- /dev/null
+++ b/docs/2_39/functions-reference/functions_index.html
@@ -0,0 +1,7656 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Alphabetical Index</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/mathematical_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Alphabetical Index">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Appendix</a></li><li class="breadcrumb-item"><a href="../functions-reference/functions_index.html">Alphabetical Index</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#alphabetical-index" id="toc-alphabetical-index" class="nav-link active" data-scroll-target="#alphabetical-index">Alphabetical Index</a>
+  <ul class="collapse">
+  <li><a href="#a" id="toc-a" class="nav-link" data-scroll-target="#a">A</a></li>
+  <li><a href="#b" id="toc-b" class="nav-link" data-scroll-target="#b">B</a></li>
+  <li><a href="#c" id="toc-c" class="nav-link" data-scroll-target="#c">C</a></li>
+  <li><a href="#d" id="toc-d" class="nav-link" data-scroll-target="#d">D</a></li>
+  <li><a href="#e" id="toc-e" class="nav-link" data-scroll-target="#e">E</a></li>
+  <li><a href="#f" id="toc-f" class="nav-link" data-scroll-target="#f">F</a></li>
+  <li><a href="#g" id="toc-g" class="nav-link" data-scroll-target="#g">G</a></li>
+  <li><a href="#h" id="toc-h" class="nav-link" data-scroll-target="#h">H</a></li>
+  <li><a href="#i" id="toc-i" class="nav-link" data-scroll-target="#i">I</a></li>
+  <li><a href="#l" id="toc-l" class="nav-link" data-scroll-target="#l">L</a></li>
+  <li><a href="#m" id="toc-m" class="nav-link" data-scroll-target="#m">M</a></li>
+  <li><a href="#n" id="toc-n" class="nav-link" data-scroll-target="#n">N</a></li>
+  <li><a href="#o" id="toc-o" class="nav-link" data-scroll-target="#o">O</a></li>
+  <li><a href="#p" id="toc-p" class="nav-link" data-scroll-target="#p">P</a></li>
+  <li><a href="#q" id="toc-q" class="nav-link" data-scroll-target="#q">Q</a></li>
+  <li><a href="#r" id="toc-r" class="nav-link" data-scroll-target="#r">R</a></li>
+  <li><a href="#s" id="toc-s" class="nav-link" data-scroll-target="#s">S</a></li>
+  <li><a href="#t" id="toc-t" class="nav-link" data-scroll-target="#t">T</a></li>
+  <li><a href="#u" id="toc-u" class="nav-link" data-scroll-target="#u">U</a></li>
+  <li><a href="#v" id="toc-v" class="nav-link" data-scroll-target="#v">V</a></li>
+  <li><a href="#w" id="toc-w" class="nav-link" data-scroll-target="#w">W</a></li>
+  <li><a href="#y" id="toc-y" class="nav-link" data-scroll-target="#y">Y</a></li>
+  <li><a href="#z" id="toc-z" class="nav-link" data-scroll-target="#z">Z</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/functions_index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Appendix</a></li><li class="breadcrumb-item"><a href="../functions-reference/functions_index.html">Alphabetical Index</a></li></ol></nav></header>
+
+
+
+
+<!-- This file is generated by gen_index.py. Do not edit it directly. -->
+<section id="alphabetical-index" class="level1">
+<h1>Alphabetical Index</h1>
+<section id="a" class="level2">
+<h2 class="anchored" data-anchor-id="a">A</h2>
+<p><a id="abs" href="#abs" class="anchored unlink"><strong>abs</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-00d0fd704b516cbdb1c9d8cfdceb177a76481cc4"><code>(complex z) : real</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-c197d7c04ea9bf23cf252f2092111b3c5aadd02d"><code>(T x) : T</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-c197d7c04ea9bf23cf252f2092111b3c5aadd02d"><code>(T x) : T</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="acos" href="#acos" class="anchored unlink"><strong>acos</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-a3994349ad6faefbd24c6ab7f4c46819b77b7593"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-c928df245747cbd0d0fdb91212e73e8c3d5ace77"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="acosh" href="#acosh" class="anchored unlink"><strong>acosh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-5d36ca029281c0668f4125ee4d7400c52fcff71c"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-228328eeafc82944f3bdf9162ddbad97efddff0c"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="add_diag" href="#add_diag" class="anchored unlink"><strong>add_diag</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-1bd31f2687a12296c53d88aedd4fe7b91be42e6b"><code>(complex_matrix m, complex_real d) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ebc82f3647c3357e9ff2d0c3de2f5996ac435a22"><code>(complex_matrix m, complex_row_vector d) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c75a27fd3abf5459a2683dc6261be6c0045b71d6"><code>(complex_matrix m, complex_vector d) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6c3894d5b0c4aa6fcae69f294bd4a39e91e3731b"><code>(matrix m, real d) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6c34e1496216d483a0c9876713871db5b36de2b1"><code>(matrix m, row_vector d) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fd1fe55afe2b42fa662d88b9e836e27f40aed6ed"><code>(matrix m, vector d) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="algebra_solver" href="#algebra_solver" class="anchored unlink"><strong>algebra_solver</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-717206dad47a7c1194d4b531baf6ffdd3c2d028a"><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="algebra_solver_newton" href="#algebra_solver_newton" class="anchored unlink"><strong>algebra_solver_newton</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-c32bcab203b8a5c04135db0647b03444d7effefb"><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-e7af0c43c395b76b7182b00996444a8b8377cac4"><code>(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="append_array" href="#append_array" class="anchored unlink"><strong>append_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a92d8a1e904382ddb4b800a330feedcb3c04f2c2"><code>(T x, T y) : T</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="append_col" href="#append_col" class="anchored unlink"><strong>append_col</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-6d9a3c740a41d03e1694a8957fc0686db16578c7"><code>(complex x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ffc1159a99f60f1f9ed57275f62abef2a8565af2"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-3cdea5abd8dcc4f4aa3a63ec4b57e3beb1e548d1"><code>(complex_matrix x, complex_vector y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-435f75dc018a8af24be3ee5ac716757fc6ed0919"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e38bc5278f6ed1cde644188c8eb17161f960454d"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-925808fc8edc656698d2fa5a95c71e4392b64386"><code>(complex_vector x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a7e682a73b7b4c9c5e762266afcff98336e95763"><code>(complex_vector x, complex_vector y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9e081ce392909a0df24bb36f03bfa797bed61afc"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-147e13862145f964e1cbb7fcc77a67089a7e5e34"><code>(matrix x, vector y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c2a28748bc8f89b100ba5a447e6a37085a0c322f"><code>(real x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3bee320543341ea29edfb9d04aca5cc49f8e0b95"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-21151b3cbeecb129181ef54f899636e705e64a59"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-864b816280933455ecf6e923c13d45129a29b0fd"><code>(vector x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-58eedcbe1e2b7f299a093b3640fbf33c5ed668a7"><code>(vector x, vector y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="append_row" href="#append_row" class="anchored unlink"><strong>append_row</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a9fd96b44e573f1bb61f0cf8266f4bdc7a805dd3"><code>(complex x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ca4bcfd187ccd1d2f4f0816dfc7f2b31f95ff4b2"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a0c350c64c953d5aa3a9de44bc62448e1b90233d"><code>(complex_matrix x, complex_row_vector y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-cbbd35ff37b9d41a2c66dbb60739e4b40a3b2c92"><code>(complex_row_vector x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-d3f8a5fb9c0e574f1b8fc41e58b9ff4d768b1316"><code>(complex_row_vector x, complex_row_vector y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-8140228bb8f902ae61c5e8eba43e81e6d32de7bd"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-101152cf279176e6d126b58acb513a382b3076db"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-135048f5cff906d2164940e5b4b34af8be6d9972"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1f14ca719bab4b3d6eff9ae371f10a5fac48b85c"><code>(matrix x, row_vector y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-be0707e02550507f8665c2fa85d337275029a3c5"><code>(real x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-54b2d5adfa559ab3f257ab5acd8f7f57c4fafbf9"><code>(row_vector x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6f6580337b50792274a5b4740260362421589d97"><code>(row_vector x, row_vector y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-626869953f524fd6120faadc1ed9dcdb8f710167"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f406895b07b910f47d6bd89b7d603c711582be11"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="arg" href="#arg" class="anchored unlink"><strong>arg</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-14bed65ae47bba6447a5f2bb816ef9fe09bb8dd2"><code>(complex z) : real</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="asin" href="#asin" class="anchored unlink"><strong>asin</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-e8dbabd64d9392bddc505ca75aaf4aa1ef314c3d"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f2bf9ebf3337e4f590b7f30ebe4aaeb04d2ca9a3"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="asinh" href="#asinh" class="anchored unlink"><strong>asinh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-8fd580d63363a17a791a49b5b536b475a5bef834"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-a267074b1b9acb770b679aeaca1e79f326fecd09"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="atan" href="#atan" class="anchored unlink"><strong>atan</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-f60565512ea3debd990f3ae7ee57978b45210673"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4ff6fd29f63ba53b4b854f16fb4122d25f817257"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="atan2" href="#atan2" class="anchored unlink"><strong>atan2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-bf8aeecad9e7aec2135da1c0e68b6c6e380eb65c"><code>(T y, T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="atanh" href="#atanh" class="anchored unlink"><strong>atanh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-51283fe4d2aef2c21fe4abe006b88ed2a9ab8b92"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1a2910acee55dabfad98224fee341fccc8af7cb0"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="b" class="level2">
+<h2 class="anchored" data-anchor-id="b">B</h2>
+<p><a id="bernoulli" href="#bernoulli" class="anchored unlink"><strong>bernoulli</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-7bb1d0fe6798b5a1780314a5829d10cf070cb584">distribution statement</a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_cdf" href="#bernoulli_cdf" class="anchored unlink"><strong>bernoulli_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-fba3b9784cb145bce67dc4b2b5b7e6beedcde682"><code>(ints y | reals theta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_lccdf" href="#bernoulli_lccdf" class="anchored unlink"><strong>bernoulli_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-8669190b9a16302c87a7a759518c13b507ba01d5"><code>(ints y | reals theta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_lcdf" href="#bernoulli_lcdf" class="anchored unlink"><strong>bernoulli_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-aee9b81290c031013a18766f7c73a0d661732bc1"><code>(ints y | reals theta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit" href="#bernoulli_logit" class="anchored unlink"><strong>bernoulli_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-778ebacd22f9f00e5c91bc26ac55e9ef4530423a">distribution statement</a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_glm" href="#bernoulli_logit_glm" class="anchored unlink"><strong>bernoulli_logit_glm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-cc775222ff8767d0ddb5f23a1f67b77eac3ef7f1">distribution statement</a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_glm_lpmf" href="#bernoulli_logit_glm_lpmf" class="anchored unlink"><strong>bernoulli_logit_glm_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-3c4e9f4b1e229095991ef7d66e815550e6189de8"><code>(array[] int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-5ba40a905c796c1bcd54602e2cdde64d36486f54"><code>(array[] int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-e6ed6ee376ada4610e458b39eb40199e201a1752"><code>(array[] int y | row_vector x, real alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-616a93c521e26ed8b747183681c8f77775ffad47"><code>(array[] int y | row_vector x, vector alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-b84bfdcb0924789b419a611e996c6185d4c9ec71"><code>(int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-8595b1aadcd41c4266263ddfa5529f3381f42c02"><code>(int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_glm_lupmf" href="#bernoulli_logit_glm_lupmf" class="anchored unlink"><strong>bernoulli_logit_glm_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-171d0ba3be3d570d3167c621b3a6e7bac0f34b49"><code>(array[] int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-75e3b277f495a675db21144dfd54403c5b61e4b1"><code>(array[] int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-75fe439a6c3fdf76668cd7afcb6fcc6d7d2aee4e"><code>(array[] int y | row_vector x, real alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-06917600c21e8a8c04a19d2e00412d5adc619330"><code>(array[] int y | row_vector x, vector alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-58e5cd13a2585a62b0a4305dfb70b2276419bd31"><code>(int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-09934df75edb200a53b9e3f261e389a5c908bb0e"><code>(int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_glm_rng" href="#bernoulli_logit_glm_rng" class="anchored unlink"><strong>bernoulli_logit_glm_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-1caea52d2e1f6a4419889e2a233b20178f262a89"><code>(matrix x, vector alpha, vector beta) : array[] int</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-486921f4f12ed8fec65ef02c934ac90a947400a0"><code>(row_vector x, vector alpha, vector beta) : array[] int</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_lpmf" href="#bernoulli_logit_lpmf" class="anchored unlink"><strong>bernoulli_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-cad8c604c21adb32fa4d2b5789b7a3888ebaf554"><code>(ints y | reals alpha) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_lupmf" href="#bernoulli_logit_lupmf" class="anchored unlink"><strong>bernoulli_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-74fd84a52ce20d7b2b2914acecea456438e780a5"><code>(ints y | reals alpha) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_logit_rng" href="#bernoulli_logit_rng" class="anchored unlink"><strong>bernoulli_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-275ab72ec22b30d37f9bc02991be25a6bd23f49a"><code>(reals alpha) : R</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_lpmf" href="#bernoulli_lpmf" class="anchored unlink"><strong>bernoulli_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-d5704b5b3fb92b6acbe402518995306b417633a6"><code>(ints y | reals theta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_lupmf" href="#bernoulli_lupmf" class="anchored unlink"><strong>bernoulli_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-d71dcd384cd23a024264d9e21555dafc50336c88"><code>(ints y | reals theta) : real</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bernoulli_rng" href="#bernoulli_rng" class="anchored unlink"><strong>bernoulli_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/binary_distributions.html#index-entry-8e32579888cc9b06d6c81f2c372fc7b94641ce47"><code>(reals theta) : R</code></a> <span class="detail">(binary_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="bessel_first_kind" href="#bessel_first_kind" class="anchored unlink"><strong>bessel_first_kind</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-3bc21536b0609c1753010054298a02498c21c610"><code>(int v, real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-88cb374bd5bc83af305e30102a59f6e38cd04165"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="bessel_second_kind" href="#bessel_second_kind" class="anchored unlink"><strong>bessel_second_kind</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-a76f46f2143eed84250e51f4f9fa21abf7ff8085"><code>(int v, real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7d09ab04cf9e5ad369f0475cf89c0dcf0e6cf137"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta" href="#beta" class="anchored unlink"><strong>beta</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-bbec535fbcce7c57b2dd2cc189383f0c180038ed"><code>(real alpha, real beta) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-faea3c6dc569671c76ab57211c00ebac2394742c"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-ed6b0e2a1365933147e892aab802b3da10b2bb6d">distribution statement</a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial" href="#beta_binomial" class="anchored unlink"><strong>beta_binomial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-989d2a39dee7b5799bfbc8bc1274f0307b4894a2">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial_cdf" href="#beta_binomial_cdf" class="anchored unlink"><strong>beta_binomial_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-ba9143f242e640009de45aa086f5082832c63ac0"><code>(ints n | ints N, reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial_lccdf" href="#beta_binomial_lccdf" class="anchored unlink"><strong>beta_binomial_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-865ef09530f64c02214e607ccdee381fca79831c"><code>(ints n | ints N, reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial_lcdf" href="#beta_binomial_lcdf" class="anchored unlink"><strong>beta_binomial_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-85c6b546de98c04c4454a1521ecc3fad803b7994"><code>(ints n | ints N, reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial_lpmf" href="#beta_binomial_lpmf" class="anchored unlink"><strong>beta_binomial_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-5e63ca67ce5a7886091457e81ed9ad8abad1af53"><code>(ints n | ints N, reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial_lupmf" href="#beta_binomial_lupmf" class="anchored unlink"><strong>beta_binomial_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-91a92506bc4098570edad18162c4925e53b060cb"><code>(ints n | ints N, reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_binomial_rng" href="#beta_binomial_rng" class="anchored unlink"><strong>beta_binomial_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2103ccd2a4395de2cf0bc035105ff78f46f83671"><code>(ints N, reals alpha, reals beta) : R</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_cdf" href="#beta_cdf" class="anchored unlink"><strong>beta_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-096320ad72e962315021c5f04c98350b8ff52744"><code>(reals theta | reals alpha, reals beta) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_lccdf" href="#beta_lccdf" class="anchored unlink"><strong>beta_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-70ce7e460a07bc188c0e0eb7ab6c0ed1d212d523"><code>(reals theta | reals alpha, reals beta) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_lcdf" href="#beta_lcdf" class="anchored unlink"><strong>beta_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-a3067ae5b48a826e8ad6971507a45b59a9e242ed"><code>(reals theta | reals alpha, reals beta) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_lpdf" href="#beta_lpdf" class="anchored unlink"><strong>beta_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-93d2282b7bada6cf97c5977413fb03d509c396ea"><code>(reals theta | reals alpha, reals beta) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_lupdf" href="#beta_lupdf" class="anchored unlink"><strong>beta_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-340181b236cc79820c87cfca7dc7e8d9ef5bdf6e"><code>(reals theta | reals alpha, reals beta) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial" href="#beta_neg_binomial" class="anchored unlink"><strong>beta_neg_binomial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-0c7465aa1beceb6e7e303af36b60e2b847fc562a">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial_cdf" href="#beta_neg_binomial_cdf" class="anchored unlink"><strong>beta_neg_binomial_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-ecce69045f7901c0a69ccc67aaae5ddda48304b2"><code>(ints n | reals r, reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial_lccdf" href="#beta_neg_binomial_lccdf" class="anchored unlink"><strong>beta_neg_binomial_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-50c63a123857b7480722a4aeb3d884c95317cdc3"><code>(ints n | reals r, reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial_lcdf" href="#beta_neg_binomial_lcdf" class="anchored unlink"><strong>beta_neg_binomial_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-fbab2aa3f292996672306979423648761d0ab98f"><code>(ints n | reals r, reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial_lpmf" href="#beta_neg_binomial_lpmf" class="anchored unlink"><strong>beta_neg_binomial_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-9d835cd7ed67218478d70e1db49912bf2d710a1e"><code>(ints n | reals r, reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial_lupmf" href="#beta_neg_binomial_lupmf" class="anchored unlink"><strong>beta_neg_binomial_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-1bd0b7fec4d1f3e315d4328c1844d15641058615"><code>(ints n | reals r, reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_neg_binomial_rng" href="#beta_neg_binomial_rng" class="anchored unlink"><strong>beta_neg_binomial_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-8c97955cf7e95ec459f434cad32f31a963d196cd"><code>(reals r, reals alpha, reals beta) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_proportion" href="#beta_proportion" class="anchored unlink"><strong>beta_proportion</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-9e3ba228a8bb07d871e13efa26340614b125cd7f">distribution statement</a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_proportion_lccdf" href="#beta_proportion_lccdf" class="anchored unlink"><strong>beta_proportion_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-8c5423cb82f066b9e55e6fdeefcbb2355698308f"><code>(reals theta | reals mu, reals kappa) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_proportion_lcdf" href="#beta_proportion_lcdf" class="anchored unlink"><strong>beta_proportion_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-61a129c59e7facb5ab538ca66fb2f4f1750bf7e1"><code>(reals theta | reals mu, reals kappa) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_proportion_lpdf" href="#beta_proportion_lpdf" class="anchored unlink"><strong>beta_proportion_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-508527134a33175b28ae22d5ff494a10fb777b4e"><code>(reals theta | reals mu, reals kappa) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_proportion_lupdf" href="#beta_proportion_lupdf" class="anchored unlink"><strong>beta_proportion_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-813c2f603806061c22e0d21c42f757aa47c7a3e7"><code>(reals theta | reals mu, reals kappa) : real</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_proportion_rng" href="#beta_proportion_rng" class="anchored unlink"><strong>beta_proportion_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-55d15275007130c8e57a41bf24370311b3581959"><code>(reals mu, reals kappa) : R</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="beta_rng" href="#beta_rng" class="anchored unlink"><strong>beta_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/continuous_distributions_on_0_1.html#index-entry-447854f88734efbef65bc89140a035fe0f16cc74"><code>(reals alpha, reals beta) : R</code></a> <span class="detail">(continuous_distributions_on_0_1.html)</span>
+</div></li>
+</ul>
+<p><a id="binary_log_loss" href="#binary_log_loss" class="anchored unlink"><strong>binary_log_loss</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-44fff3495ce5cd36ebde4587e802cb7bdc795794"><code>(int y, real y_hat) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-05d751372242ac11c1c7356d09fb0341276b897c"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial" href="#binomial" class="anchored unlink"><strong>binomial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-04d7179640d2a656dc12c9b1c5ed98a04620ef81">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_cdf" href="#binomial_cdf" class="anchored unlink"><strong>binomial_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-304f7e96d0cb2c64c6ff84e3b67d56b0ff58e1cc"><code>(ints n | ints N, reals theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_lccdf" href="#binomial_lccdf" class="anchored unlink"><strong>binomial_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-51c3d20caad1f513ce40fc1c8af17ec9c1d82eca"><code>(ints n | ints N, reals theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_lcdf" href="#binomial_lcdf" class="anchored unlink"><strong>binomial_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-0414522d06188bdb5122e227051e36f457a2a1c6"><code>(ints n | ints N, reals theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_logit" href="#binomial_logit" class="anchored unlink"><strong>binomial_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-0c90bd9f55ca58f075077bde8a8c7fa4b260b346">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_logit_glm" href="#binomial_logit_glm" class="anchored unlink"><strong>binomial_logit_glm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-c8c31102f7b5abf104dd462dd52e20ad4ee53d24">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_logit_glm_lpmf" href="#binomial_logit_glm_lpmf" class="anchored unlink"><strong>binomial_logit_glm_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-3f85893b1fea02de309015929662b0b20beb9309"><code>(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-1882ad2cd7ca87f3d1050d5be18f2adcc431d794"><code>(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-e1b1936f95d710de4ad0682cbf085a530908d39a"><code>(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-e38992f8d67e6ef5ce4e08abdfb07fe5ff1d3c9a"><code>(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-3da9351d4d4f1762941c69259fcfddf39dd65eae"><code>(int n | int N, matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-a7a84a22f1a74964827e7e3c3c2ecb0233e5cf37"><code>(int n | int N, matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_logit_glm_lupmf" href="#binomial_logit_glm_lupmf" class="anchored unlink"><strong>binomial_logit_glm_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-f45f2f92b4005c55ad8f5cdd2285c9708fefe806"><code>(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-9889fa94aac823796e6949b79108363bfbbbaa9d"><code>(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-428627db76fccf914ceaefa4cd5884bd131556b9"><code>(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-7d6e3169c60256f12c03411a7509e0370a1ee157"><code>(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-de178183d4185e13aa59ffee44cc49a8ba2dc7e9"><code>(int n | int N, matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-588f750b18cf5987a93c0022ecc43c232efc3e24"><code>(int n | int N, matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_logit_lpmf" href="#binomial_logit_lpmf" class="anchored unlink"><strong>binomial_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2231c87f532ddd0458a862cf4cc6e80ca618fc99"><code>(ints n | ints N, reals alpha) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_logit_lupmf" href="#binomial_logit_lupmf" class="anchored unlink"><strong>binomial_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2fe1708373b5f137c68a8ac756400a2be58856fd"><code>(ints n | ints N, reals alpha) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_lpmf" href="#binomial_lpmf" class="anchored unlink"><strong>binomial_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-8798e65d04763014c36aa7b9c6509c71c538efb6"><code>(ints n | ints N, reals theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_lupmf" href="#binomial_lupmf" class="anchored unlink"><strong>binomial_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-d9a18d8e413d00481fdf0a47e523a5232870fc25"><code>(ints n | ints N, reals theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="binomial_rng" href="#binomial_rng" class="anchored unlink"><strong>binomial_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-15ce73a3a3caa43cbbe65c4e58495cf18ed8886e"><code>(ints N, reals theta) : R</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="block" href="#block" class="anchored unlink"><strong>block</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-2d1266e93b2287bbb26f4015769ebc779ef29f81"><code>(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ed12d70d85024ac1bb87c1c1c5971659fb5f2cc1"><code>(matrix x, int i, int j, int n_rows, int n_cols) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="c" class="level2">
+<h2 class="anchored" data-anchor-id="c">C</h2>
+<p><a id="categorical" href="#categorical" class="anchored unlink"><strong>categorical</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-bd19bcb6308a78b3569b39d80a5050df0380eae5">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit" href="#categorical_logit" class="anchored unlink"><strong>categorical_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2ffcb74c4ba70a5424f59a10d068421051f4963c">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit_glm" href="#categorical_logit_glm" class="anchored unlink"><strong>categorical_logit_glm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-9d4042f47493597a30307ede3c9743ad09549f4d">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit_glm_lpmf" href="#categorical_logit_glm_lpmf" class="anchored unlink"><strong>categorical_logit_glm_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-512bbe1dc00cc3ef23d8272ec1c0da56105f1c42"><code>(array[] int y | matrix x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-8f756acaa498908962c53d1a6b0af17598fd8524"><code>(array[] int y | row_vector x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-b41f55e8856f7805a26e7a2ee183d028df87f769"><code>(int y | matrix x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-4e1f97ef06a17d796b0a7cc5ae22003b03f95401"><code>(int y | row_vector x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit_glm_lupmf" href="#categorical_logit_glm_lupmf" class="anchored unlink"><strong>categorical_logit_glm_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-09da9504e8676f6b35327c85d13771d63f77b83b"><code>(array[] int y | matrix x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-518462f77e8ba90187fbc7302cb54236a0c8b333"><code>(array[] int y | row_vector x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-866be3d0f99507bddabbbe0652cce251858d2c37"><code>(int y | matrix x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-3e5402103e17dd48bb2625e6a2b76c48782a1a72"><code>(int y | row_vector x, vector alpha, matrix beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit_lpmf" href="#categorical_logit_lpmf" class="anchored unlink"><strong>categorical_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-da54bfd2ecc17ee462038470f94a63c7be030e08"><code>(ints y | vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit_lupmf" href="#categorical_logit_lupmf" class="anchored unlink"><strong>categorical_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-07422d7054ee8fa6c1816736064c38c694563486"><code>(ints y | vector beta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_logit_rng" href="#categorical_logit_rng" class="anchored unlink"><strong>categorical_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2da030b7d2ce945503ab8ef37e97ee5afc78305e"><code>(vector beta) : int</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_lpmf" href="#categorical_lpmf" class="anchored unlink"><strong>categorical_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-9dfc05c1a7f9f9cf620187547781111da6ad08ab"><code>(ints y | vector theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_lupmf" href="#categorical_lupmf" class="anchored unlink"><strong>categorical_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-83c502e2e302199ed3e04bed34c4bef07808ba9d"><code>(ints y | vector theta) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="categorical_rng" href="#categorical_rng" class="anchored unlink"><strong>categorical_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-e829b16f14961a0a24d2dd1a7895dcda4ce79bd1"><code>(vector theta) : int</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy" href="#cauchy" class="anchored unlink"><strong>cauchy</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4e9a3b08fddc23e306b5a3b0a4f199124cb7cd90">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy_cdf" href="#cauchy_cdf" class="anchored unlink"><strong>cauchy_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f636bb8d19bdc544f177d76c3e93b11a8947a332"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy_lccdf" href="#cauchy_lccdf" class="anchored unlink"><strong>cauchy_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-252fde132b94e629570349d740a553001ba151d7"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy_lcdf" href="#cauchy_lcdf" class="anchored unlink"><strong>cauchy_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-1470e53e1ef3170600faa1e10925e17becd26f46"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy_lpdf" href="#cauchy_lpdf" class="anchored unlink"><strong>cauchy_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-365b2742adf729c8087f3eb28841c38baec29082"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy_lupdf" href="#cauchy_lupdf" class="anchored unlink"><strong>cauchy_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-70ded17908d0eaf198cca69f74a0bec65a4ec897"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cauchy_rng" href="#cauchy_rng" class="anchored unlink"><strong>cauchy_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-56d08e840245967636a7f56a3b26cf0ad54161f8"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="cbrt" href="#cbrt" class="anchored unlink"><strong>cbrt</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-e5c3ef31b3dfc2b73c3572239e82a2ae8629df79"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ceil" href="#ceil" class="anchored unlink"><strong>ceil</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-3989753a963097381576ca0373b3190681b0bbe8"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square" href="#chi_square" class="anchored unlink"><strong>chi_square</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-a8bd60e9ff005e8522863542e28a64336a42e65c">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square_cdf" href="#chi_square_cdf" class="anchored unlink"><strong>chi_square_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-face5eba84183b72541fdba1f3b6ea4a1949e9b5"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square_lccdf" href="#chi_square_lccdf" class="anchored unlink"><strong>chi_square_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-a010e1cf6b0229303867a826777cc2a9b6c3bff9"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square_lcdf" href="#chi_square_lcdf" class="anchored unlink"><strong>chi_square_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-90d49c25cb5546006ff8802e5928c09d11c03dc8"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square_lpdf" href="#chi_square_lpdf" class="anchored unlink"><strong>chi_square_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-6c7045c1cb405776f7ed13ebbb5da4b27240ea15"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square_lupdf" href="#chi_square_lupdf" class="anchored unlink"><strong>chi_square_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-9cccc8fcca1c8dd1b172176bd72c946a84d611eb"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chi_square_rng" href="#chi_square_rng" class="anchored unlink"><strong>chi_square_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-ced9b4410298033c3bdae65b66331f4a6d973c89"><code>(reals nu) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="chol2inv" href="#chol2inv" class="anchored unlink"><strong>chol2inv</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0fbf0fe61f7df06da84a357eb40e9c719ff806ac"><code>(matrix L) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_decompose" href="#cholesky_decompose" class="anchored unlink"><strong>cholesky_decompose</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a7eaa217876baa0054d78dfff05f5685ef6261bb"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_factor_corr_constrain" href="#cholesky_factor_corr_constrain" class="anchored unlink"><strong>cholesky_factor_corr_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-13178bf49e47f4d98c973800329f95f8fd49bfa1"><code>(vectors y, int K) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_factor_corr_jacobian" href="#cholesky_factor_corr_jacobian" class="anchored unlink"><strong>cholesky_factor_corr_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-1e720292cb09512b6fdca84e4cc39a087f43ec0a"><code>(vectors y, int K) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_factor_corr_unconstrain" href="#cholesky_factor_corr_unconstrain" class="anchored unlink"><strong>cholesky_factor_corr_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-98b8008a7ecf84ed0131fe5e84c2a239b3739d30"><code>(matrices x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_factor_cov_constrain" href="#cholesky_factor_cov_constrain" class="anchored unlink"><strong>cholesky_factor_cov_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-0cddde7e0aa5cc9213f393acebb3b13b68b7bfa8"><code>(vectors y, int M, int N) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_factor_cov_jacobian" href="#cholesky_factor_cov_jacobian" class="anchored unlink"><strong>cholesky_factor_cov_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-fd1e2249d8d133c4bf9d8e6b44f67c9c69314da3"><code>(vectors y, int M, int N) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cholesky_factor_cov_unconstrain" href="#cholesky_factor_cov_unconstrain" class="anchored unlink"><strong>cholesky_factor_cov_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-74c3b28f5b00bf51f6299872cce7f2b1a18bf3ac"><code>(matrices x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="choose" href="#choose" class="anchored unlink"><strong>choose</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-fd4d9124ce89bf38a7ca1601e28f14947fcae3ba"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-ed9d9ccf90b8976a4cc2b40babee7c60d4807087"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="col" href="#col" class="anchored unlink"><strong>col</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-3a0cf3be0b3793d63afbee36001fbeed1f83f08f"><code>(complex_matrix x, int n) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fe3819da259646140d101538903c86c5bc1b49e9"><code>(matrix x, int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="cols" href="#cols" class="anchored unlink"><strong>cols</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-22eb74ccaa0e3cbd1875a10b1d25e448aaca1078"><code>(complex_matrix x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a3afb9b9b85c96fcbb2ceccffe8220cdf6930ee0"><code>(complex_row_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-2926894bc112a2f76fc88ec0d1b66ca72f6907ab"><code>(complex_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f399f6ffab24708be4ff867cbaef541158fc00bf"><code>(matrix x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-417aeef6881030f640a4ea077e02335739bf0e3e"><code>(row_vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8113ac540f4cc3aa182c3afcfa5d1665778f7fd9"><code>(vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="columns_dot_product" href="#columns_dot_product" class="anchored unlink"><strong>columns_dot_product</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c0a3895052c73d4b7553b42dfbcf4ceee17d6ebd"><code>(complex_matrix x, complex_matrix y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-aee28160ebf0573187fba8d1b9917b490dfd47cd"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-73c167059a6fb3ee21b48776d84d7c14cd6b6b9a"><code>(complex_vector x, complex_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-758593f0bee7c0421df7fc503237288851879118"><code>(matrix x, matrix y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4758c18c6242ed86d28834ea3c356de1d5e22b86"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7b3a4d9f9236657fd6d5970b8840e4a2e7e176bf"><code>(vector x, vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="columns_dot_self" href="#columns_dot_self" class="anchored unlink"><strong>columns_dot_self</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-499a93465efa4feea0949bffdec099ac446865be"><code>(complex_matrix x) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5a3c0ca6ac56df87695d6d53c5ed41bd076db1bf"><code>(complex_row_vector x) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-25320510319f7efaed97fc7327bcb3481a1fd055"><code>(complex_vector x) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-664db2ea209dd84f3f62d813817a0be019ff3afb"><code>(matrix x) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-258f64fdc22d8a6e1012cfa375d94222962c3add"><code>(row_vector x) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f0c5a660d69d25748bdafa9cda98ee0cfa190cbe"><code>(vector x) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="complex_schur_decompose" href="#complex_schur_decompose" class="anchored unlink"><strong>complex_schur_decompose</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-400ef877200b32483b4f5a7223c3231666d153b7"><code>(complex_matrix A) : tuple(complex_matrix, complex_matrix)</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-9c45b8917b1de6f500415512b1d06896cd5f7b5c"><code>(matrix A) : tuple(complex_matrix, complex_matrix)</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="complex_schur_decompose_t" href="#complex_schur_decompose_t" class="anchored unlink"><strong>complex_schur_decompose_t</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-1728988d586fc721b7a1d7d21d1aa3252bb03f94"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-f43e2c8a9893e96b7c09a618d7b6f69350bfc460"><code>(matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="complex_schur_decompose_u" href="#complex_schur_decompose_u" class="anchored unlink"><strong>complex_schur_decompose_u</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-fdc477180371e8366e3a46dc88f3875f088907c8"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-b169076e82ada507c5c30f36fe4a5bda58af9e0a"><code>(matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="conj" href="#conj" class="anchored unlink"><strong>conj</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-bd2ae18bb245caf34d1b2f12b3883c47d8aa7e0a"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-17692d922273aaba0c5317617aab6585dda28bfa"><code>(Z z) : Z</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="corr_matrix_constrain" href="#corr_matrix_constrain" class="anchored unlink"><strong>corr_matrix_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-594161bab69b742477ec84f4a399816cc0d939a1"><code>(vectors y, int K) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="corr_matrix_jacobian" href="#corr_matrix_jacobian" class="anchored unlink"><strong>corr_matrix_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-a3ce311c60e3c6821b434df3f73f83c5e037a03b"><code>(vectors y, int K) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="corr_matrix_unconstrain" href="#corr_matrix_unconstrain" class="anchored unlink"><strong>corr_matrix_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-0fcba057e29e1ae091121cb62708675880f50c39"><code>(matrices x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cos" href="#cos" class="anchored unlink"><strong>cos</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-3d86aa681cf64514f405ada706ad9969e5f33080"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-c587dc7908ecfe46cdd09800844ad9ea90e8463b"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cosh" href="#cosh" class="anchored unlink"><strong>cosh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-9b0fa15f5eebadc83a3678cb57fd8a1f4d089b94"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1afe4d43c0f158ff3446b0459e002ad8343ee24e"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cov_exp_quad" href="#cov_exp_quad" class="anchored unlink"><strong>cov_exp_quad</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/removed_functions.html#index-entry-1ba90fc13ad6ba28ca73fc3b94b0a65af09669f8"><code>(array[] real x, real alpha, real rho) : matrix</code></a> <span class="detail">(removed_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/removed_functions.html#index-entry-95bfeb84294b276027c24d3f8856560234423e7c"><code>(array[] real x1, array[] real x2, real alpha, real rho) : matrix</code></a> <span class="detail">(removed_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/removed_functions.html#index-entry-92099ecf9309d8f643e455cb6e3f79c985d35cb1"><code>(row_vectors x, real alpha, real rho) : matrix</code></a> <span class="detail">(removed_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/removed_functions.html#index-entry-c05741ae8e35d6b11a03c0098ac00b99d5a5c845"><code>(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix</code></a> <span class="detail">(removed_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/removed_functions.html#index-entry-d271045bb5d0614ff3a2f0e3d8d279d134cf0304"><code>(vectors x, real alpha, real rho) : matrix</code></a> <span class="detail">(removed_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/removed_functions.html#index-entry-70b15de7e29e4e64f05ca3fc552da349d4a66c1f"><code>(vectors x1, vectors x2, real alpha, real rho) : matrix</code></a> <span class="detail">(removed_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cov_matrix_constrain" href="#cov_matrix_constrain" class="anchored unlink"><strong>cov_matrix_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-6dc98b07eda2e6dfde0020cf98dae68b590aad28"><code>(vectors y, int K) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cov_matrix_jacobian" href="#cov_matrix_jacobian" class="anchored unlink"><strong>cov_matrix_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-cad2be5fc2ee16a8419b70db0422629b843de6c3"><code>(vectors y, int K) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="cov_matrix_unconstrain" href="#cov_matrix_unconstrain" class="anchored unlink"><strong>cov_matrix_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-00d19b2a52626d58b6b07f2d27fdd2dbad3c12c0"><code>(matrices x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="crossprod" href="#crossprod" class="anchored unlink"><strong>crossprod</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c5372c9d5715e11b5d87a465d6d671d608ef9513"><code>(matrix x) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="csr_extract" href="#csr_extract" class="anchored unlink"><strong>csr_extract</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/sparse_matrix_operations.html#index-entry-32fea6602ecbd4c1ea15d25bb2a6b2d2fcf58e52"><code>(matrix a) : tuple(vector, array[] int, array[] int)</code></a> <span class="detail">(sparse_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="csr_extract_u" href="#csr_extract_u" class="anchored unlink"><strong>csr_extract_u</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/sparse_matrix_operations.html#index-entry-1f750e4b73b9d17d5eadd184840443bab5de2dda"><code>(matrix a) : array[] int</code></a> <span class="detail">(sparse_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="csr_extract_v" href="#csr_extract_v" class="anchored unlink"><strong>csr_extract_v</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/sparse_matrix_operations.html#index-entry-2569af86a359f94d59f18b21a0b4a15d7ad13f5f"><code>(matrix a) : array[] int</code></a> <span class="detail">(sparse_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="csr_extract_w" href="#csr_extract_w" class="anchored unlink"><strong>csr_extract_w</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/sparse_matrix_operations.html#index-entry-fa08d7d6e24c28fcab7d2276501e5cbf1b12e09d"><code>(matrix a) : vector</code></a> <span class="detail">(sparse_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="csr_matrix_times_vector" href="#csr_matrix_times_vector" class="anchored unlink"><strong>csr_matrix_times_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/sparse_matrix_operations.html#index-entry-a5e3d826290e2d8bff8cf1e1019bd2656ad43d2b"><code>(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector</code></a> <span class="detail">(sparse_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="csr_to_dense_matrix" href="#csr_to_dense_matrix" class="anchored unlink"><strong>csr_to_dense_matrix</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/sparse_matrix_operations.html#index-entry-b8ce33cfed90d588aaa17f7c95c10b4206863d15"><code>(int m, int n, vector w, array[] int v, array[] int u) : matrix</code></a> <span class="detail">(sparse_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="cumulative_sum" href="#cumulative_sum" class="anchored unlink"><strong>cumulative_sum</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-d6b4157abb8f1f7140f20c5b9182ad2f7d69fe37"><code>(array[] complex x) : array[] complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-092f8a13b9e06f8346cb61b8fc6a6fbacb3b5cc5"><code>(array[] int x) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1e965776faac7516107384178cafd382f87f909a"><code>(array[] real x) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c104f0e3e10596f559a8a023c84f92f1b919bc0f"><code>(complex_row_vector rv) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-45545f14726c28e6b63c8ec09e06e53b33e4ca02"><code>(complex_vector v) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-951e3c465251f531a70f0dce6772057bf114685c"><code>(row_vector rv) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7c00c2971c29f98a54f16c9468d5b4a6eb529ff8"><code>(vector v) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="d" class="level2">
+<h2 class="anchored" data-anchor-id="d">D</h2>
+<p><a id="dae" href="#dae" class="anchored unlink"><strong>dae</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-c47e85b0ae801c7df44337e2c2b7ffc5c84037bf"><code>(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="dae_tol" href="#dae_tol" class="anchored unlink"><strong>dae_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-52f7e7ef9a4a58eb94353bf2cb98511deb94aac9"><code>(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="determinant" href="#determinant" class="anchored unlink"><strong>determinant</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2f0c697d96cb41a09f35f183f179f15b1ee0f3fa"><code>(matrix A) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="diag_matrix" href="#diag_matrix" class="anchored unlink"><strong>diag_matrix</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-06edc5538b52d1a11f2a6f45be9ca1e0683bbb1e"><code>(complex_vector x) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-74845a2c2d1e1321f2bfdf69d4c7f321c856f555"><code>(vector x) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="diag_post_multiply" href="#diag_post_multiply" class="anchored unlink"><strong>diag_post_multiply</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-82629e9f9045325a28b4c0610df36226bc6c4562"><code>(complex_matrix m, complex_row_vector v) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-b8e527d5b7adcfe81f12d70984ba81d895332835"><code>(complex_matrix m, complex_vector v) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fedf6c86ebf0663043f5e986e077701e6ae185f4"><code>(matrix m, row_vector rv) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2ff483d0f2d35ce6055a030912e1909c110453dc"><code>(matrix m, vector v) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="diag_pre_multiply" href="#diag_pre_multiply" class="anchored unlink"><strong>diag_pre_multiply</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-29b61ea8d42d42252259e35dbe0c427c485ad9e8"><code>(complex_row_vector v, complex_matrix m) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-7907eb3baed066d578933e3832fe8366c02116eb"><code>(complex_vector v, complex_matrix m) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-87b1033ce509f3990ef4516c29c0b9b08d67c37b"><code>(row_vector rv, matrix m) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0fc4d71e7bc087ce4f9dec77e1df8df248f41bac"><code>(vector v, matrix m) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="diagonal" href="#diagonal" class="anchored unlink"><strong>diagonal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5d28ae699b02999166db3502cec1f2c9b74d1c34"><code>(complex_matrix x) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6d4ef3406b8dabab1d53b29bb53ae58d983d1404"><code>(matrix x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="digamma" href="#digamma" class="anchored unlink"><strong>digamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-2ce1d8a0cdd9878fe64366c88a2381d6e702ff7d"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="dims" href="#dims" class="anchored unlink"><strong>dims</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-15460652371e3354bdc590e3632fc540ab76b525"><code>(T x) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet" href="#dirichlet" class="anchored unlink"><strong>dirichlet</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/simplex_distributions.html#index-entry-2b1866f861130d47e076c909b2da014d36b08615">distribution statement</a> <span class="detail">(simplex_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_lpdf" href="#dirichlet_lpdf" class="anchored unlink"><strong>dirichlet_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/simplex_distributions.html#index-entry-912dbafe0e9fce423e5413cf88d46731b610742d"><code>(vectors theta | vectors alpha) : real</code></a> <span class="detail">(simplex_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_lupdf" href="#dirichlet_lupdf" class="anchored unlink"><strong>dirichlet_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/simplex_distributions.html#index-entry-d287f450cfac7ef63c57f921d82484ef44abdfa0"><code>(vectors theta | vectors alpha) : real</code></a> <span class="detail">(simplex_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_multinomial" href="#dirichlet_multinomial" class="anchored unlink"><strong>dirichlet_multinomial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-3f5317d36179306fa3d125648b669c29f0ae4632">distribution statement</a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_multinomial_lpmf" href="#dirichlet_multinomial_lpmf" class="anchored unlink"><strong>dirichlet_multinomial_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-fdad801bbd6eb934645dbde5c01f271aa4e741b4"><code>(array[] int y | vector alpha) : real</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_multinomial_lupmf" href="#dirichlet_multinomial_lupmf" class="anchored unlink"><strong>dirichlet_multinomial_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-6b8a21266747ae90543cbd60108ee12b4d08d5e3"><code>(array[] int y | vector alpha) : real</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_multinomial_rng" href="#dirichlet_multinomial_rng" class="anchored unlink"><strong>dirichlet_multinomial_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-bdd1be0680f9ae13a49c9c4e6af87b4f9c7359df"><code>(vector alpha, int N) : array[] int</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="dirichlet_rng" href="#dirichlet_rng" class="anchored unlink"><strong>dirichlet_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/simplex_distributions.html#index-entry-055ed033fb2096aa99f68c19fe9c3a956577f3b1"><code>(vector alpha) : vector</code></a> <span class="detail">(simplex_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range" href="#discrete_range" class="anchored unlink"><strong>discrete_range</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-081a4566d9bb14c20636b30012e90be941b53978">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range_cdf" href="#discrete_range_cdf" class="anchored unlink"><strong>discrete_range_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-92d375a0144eecf4c3bac183f06cde0a573ae0ef"><code>(ints y | ints l, ints u) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range_lccdf" href="#discrete_range_lccdf" class="anchored unlink"><strong>discrete_range_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-3dc029db4d1f37e783107470ba6665c647f5a202"><code>(ints y | ints l, ints u) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range_lcdf" href="#discrete_range_lcdf" class="anchored unlink"><strong>discrete_range_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-be12560a3a7c198ea965f08a78dab04d83742159"><code>(ints y | ints l, ints u) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range_lpmf" href="#discrete_range_lpmf" class="anchored unlink"><strong>discrete_range_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-4aeab4cc5d644e56b9409bff2224f5e620c7ce53"><code>(ints y | ints l, ints u) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range_lupmf" href="#discrete_range_lupmf" class="anchored unlink"><strong>discrete_range_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-32514432b1689ab58c658ff3b435675a6e798cef"><code>(ints y | ints l, ints u) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="discrete_range_rng" href="#discrete_range_rng" class="anchored unlink"><strong>discrete_range_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-a4c6bdebab12a3547ca7c13ac62e456d0b74c9dc"><code>(ints l, ints u) : ints</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="distance" href="#distance" class="anchored unlink"><strong>distance</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-c6112c85adc9432ce6b005aff57babec0dbee55e"><code>(row_vector x, row_vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a9f701404655eb686e31e66d38918d7e08cd9f3b"><code>(row_vector x, vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-d68901fc0a3ff9e987854900a185ee2713e5c520"><code>(vector x, row_vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-702855e7be40e4fcfbe31b5f08538c5fbfba210e"><code>(vector x, vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="dot_product" href="#dot_product" class="anchored unlink"><strong>dot_product</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-13cdb97ea7e259e98791362de8a261d17dab4e1c"><code>(complex_row_vector x, complex_row_vector y) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-d932a1eed6d525ce1077543b39b6834836923949"><code>(complex_row_vector x, complex_vector y) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-dd62ebed52c5e3210bb8a0155e1d5dfa70497439"><code>(complex_vector x, complex_row_vector y) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-bc037b34a4fdc977a944e39e9277f8e6c700bd5b"><code>(complex_vector x, complex_vector y) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5fa1a043fd0dcebec0cb14fd09c1a965c33d37fe"><code>(row_vector x, row_vector y) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5624c84e1a33f2cf540a3d8a2994205918d9915c"><code>(row_vector x, vector y) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4ce3d3846b8c242984aca259fd01579e1b2179af"><code>(vector x, row_vector y) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6f3cb76e0af87d28bceef319e85e75976645662e"><code>(vector x, vector y) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="dot_self" href="#dot_self" class="anchored unlink"><strong>dot_self</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-d0945f42b843344d7b22f5d088d229c5f5ba64bf"><code>(complex_row_vector x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-1ae5feb968f94e219b01c1a6943eb1532161e16f"><code>(complex_vector x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-806eb76bdef36d3ee3a9a009027593ad32d755f1"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1ae4fbadff5f8d76241c1cfc33e2c36db7decc73"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential" href="#double_exponential" class="anchored unlink"><strong>double_exponential</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-b3a44920aadcdd73a61301af6bfb0ba2be51abf8">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential_cdf" href="#double_exponential_cdf" class="anchored unlink"><strong>double_exponential_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-8339bce2e4f464e36cd18cd462828f0b740d2657"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential_lccdf" href="#double_exponential_lccdf" class="anchored unlink"><strong>double_exponential_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-14eaa0ccbccd40ad1a210cba1462140349b9c18d"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential_lcdf" href="#double_exponential_lcdf" class="anchored unlink"><strong>double_exponential_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-a77ae5c4ace742352452353b5faa44105bc79f55"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential_lpdf" href="#double_exponential_lpdf" class="anchored unlink"><strong>double_exponential_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f0eddd37cb58304271349b38fc0dc4de2aeb94f4"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential_lupdf" href="#double_exponential_lupdf" class="anchored unlink"><strong>double_exponential_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-32427ca40b710f689e9f6f5f77ca5f91afc0ae09"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="double_exponential_rng" href="#double_exponential_rng" class="anchored unlink"><strong>double_exponential_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-5489a23aaef4c1953d0e22e808a68a6fc37ab1d8"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="e" class="level2">
+<h2 class="anchored" data-anchor-id="e">E</h2>
+<p><a id="e" href="#e" class="anchored unlink"><strong>e</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-c9e1664eab4cc021df10c6902738cce047b7d8f8"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="eigendecompose" href="#eigendecompose" class="anchored unlink"><strong>eigendecompose</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-cfb26916a57582e2e9ee4f68ed99c821756d3789"><code>(complex_matrix A) : tuple(complex_matrix, complex_vector)</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-080c58f8bf67678c670ce7454674192000c074d1"><code>(matrix A) : tuple(complex_matrix, complex_vector)</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="eigendecompose_sym" href="#eigendecompose_sym" class="anchored unlink"><strong>eigendecompose_sym</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-2e32adea8a000a491b0d0d5e736075c377a776b6"><code>(complex_matrix A) : tuple(complex_matrix, complex_vector)</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7bd20778cc4eb845435d980b2323a26d021c94a0"><code>(matrix A) : tuple(matrix, vector)</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="eigenvalues" href="#eigenvalues" class="anchored unlink"><strong>eigenvalues</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a3edb93c2a79b350e0e1ca880e3ada6deb6a30de"><code>(complex_matrix A) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-db72fae33eaa940967d1488e046f1d41ad7bd7b7"><code>(matrix A) : complex_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="eigenvalues_sym" href="#eigenvalues_sym" class="anchored unlink"><strong>eigenvalues_sym</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e271860ae502125255d5f4092244d884f2ce9692"><code>(complex_matrix A) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-09bb39cb077a959df51666f18ac7319ea53f8748"><code>(matrix A) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="eigenvectors" href="#eigenvectors" class="anchored unlink"><strong>eigenvectors</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-bc7b059733d8c7e8becdef5196cfe34a63233de0"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6c608598725ec7c838453abdeb440c875eafe1a5"><code>(matrix A) : complex_matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="eigenvectors_sym" href="#eigenvectors_sym" class="anchored unlink"><strong>eigenvectors_sym</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-67a04481d07493cc21dae09ba3cfde2b4e974951"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fb1919fccf8d3e19664c486e0cb67c876593f5a1"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="erf" href="#erf" class="anchored unlink"><strong>erf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-3ed55c0864c7839ec52fa38be9974f0e43bd5a50"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="erfc" href="#erfc" class="anchored unlink"><strong>erfc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7dc68f18ec6e0c591f011d5ff8f4075637b9dd5c"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp" href="#exp" class="anchored unlink"><strong>exp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-3c31357925206d6237b540fd4e33930d37dccb79"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f32e064beb0f00e975210d246d3f80431ea9b85f"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp2" href="#exp2" class="anchored unlink"><strong>exp2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7db9b662f6846f90e9edf891fac2a64701c213d2"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal" href="#exp_mod_normal" class="anchored unlink"><strong>exp_mod_normal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-b426ea4ba5c486f0dee2d11bcd8f019e1914e368">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal_cdf" href="#exp_mod_normal_cdf" class="anchored unlink"><strong>exp_mod_normal_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-e60633c944fc492d9aabe37e92180634dc3beb1f"><code>(reals y | reals mu, reals sigma, reals lambda) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal_lccdf" href="#exp_mod_normal_lccdf" class="anchored unlink"><strong>exp_mod_normal_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-ee7f0016f4ebc38e3cf0d9590e2a47ae35d6d131"><code>(reals y | reals mu, reals sigma, reals lambda) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal_lcdf" href="#exp_mod_normal_lcdf" class="anchored unlink"><strong>exp_mod_normal_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-06dbb9069af1f1382e22ed5f13335630182ae084"><code>(reals y | reals mu, reals sigma, reals lambda) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal_lpdf" href="#exp_mod_normal_lpdf" class="anchored unlink"><strong>exp_mod_normal_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-bb7d02b1917d231421d8c9d936fb60acd1dee69c"><code>(reals y | reals mu, reals sigma, reals lambda) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal_lupdf" href="#exp_mod_normal_lupdf" class="anchored unlink"><strong>exp_mod_normal_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-8933df6537efc956060be8137a18b3593c5c4833"><code>(reals y | reals mu, reals sigma, reals lambda) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exp_mod_normal_rng" href="#exp_mod_normal_rng" class="anchored unlink"><strong>exp_mod_normal_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-673b303c582c3796fa693d9eee79c861cc6e4b3f"><code>(reals mu, reals sigma, reals lambda) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="expm1" href="#expm1" class="anchored unlink"><strong>expm1</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-159b5bb9f4c8cbaad0c033c64d1cd44196240c8d"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential" href="#exponential" class="anchored unlink"><strong>exponential</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-659974bc00c9d41d7a429ae37133a99bb8b957a4">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential_cdf" href="#exponential_cdf" class="anchored unlink"><strong>exponential_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-ad4cd5c7ba3c05b884083c95b23a472cffbc1a00"><code>(reals y | reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential_lccdf" href="#exponential_lccdf" class="anchored unlink"><strong>exponential_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-7c9705bfaff832ebfe92cc4dcc784a77c1e1bd16"><code>(reals y | reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential_lcdf" href="#exponential_lcdf" class="anchored unlink"><strong>exponential_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-0677eab24885c7279772fbf058f8cb58eb12d88a"><code>(reals y | reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential_lpdf" href="#exponential_lpdf" class="anchored unlink"><strong>exponential_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-5338844e89484284ae797d5d7f2b46148d8e8d2c"><code>(reals y | reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential_lupdf" href="#exponential_lupdf" class="anchored unlink"><strong>exponential_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-fdea6af1db60962c86c57847a99b985330592f02"><code>(reals y | reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="exponential_rng" href="#exponential_rng" class="anchored unlink"><strong>exponential_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-2fefb0b09b4b70f993267af296d92862bc9c900d"><code>(reals beta) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="f" class="level2">
+<h2 class="anchored" data-anchor-id="f">F</h2>
+<p><a id="falling_factorial" href="#falling_factorial" class="anchored unlink"><strong>falling_factorial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-501ae7bd47545a3ffa57a4c67cc0ffb27c69f93a"><code>(real x, real n) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d534a0845142902e1168c6b016843f301f8826d6"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fatal_error" href="#fatal_error" class="anchored unlink"><strong>fatal_error</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/void_functions.html#index-entry-36f600d7cd1daea1b94b57949bed5e7914fd9442"><code>(T1 x1,..., TN xN) : void</code></a> <span class="detail">(void_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fdim" href="#fdim" class="anchored unlink"><strong>fdim</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8e4c91cd9725a9e73f23c330bf345dbc813f4a44"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-e5223bd7c30433273feb0a16c3c3950552c1ebce"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fft" href="#fft" class="anchored unlink"><strong>fft</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-109377c74fb51b412a7ba92432ed83655dc5cca8"><code>(complex_vector v) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="fft2" href="#fft2" class="anchored unlink"><strong>fft2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-37fccc5b3423b32e66112f0b9f406c043fa4d0e2"><code>(complex_matrix m) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="floor" href="#floor" class="anchored unlink"><strong>floor</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f4309f63f497c18607ceca16a255a6c8c3722153"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fma" href="#fma" class="anchored unlink"><strong>fma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-086b6805da630957962521a7c4c5eb1c108313a2"><code>(real x, real y, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fmax" href="#fmax" class="anchored unlink"><strong>fmax</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-84f91b685bc8ac919063009145d0bda6df79898e"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-9619ac6ed9fc6345c116fdaa93c6eb7f4fd18d6c"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fmin" href="#fmin" class="anchored unlink"><strong>fmin</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1c3fab0d71dfe4827399431840361247caf51f6c"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-745a8f98349953b6346521de02a0f66c0e837597"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="fmod" href="#fmod" class="anchored unlink"><strong>fmod</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-fed64eefc7fd2673daac1482490fb6dfc5e37647"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-206cb5e9da51b6d1255b47d019273ab50a84687c"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet" href="#frechet" class="anchored unlink"><strong>frechet</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-033a6fd0ca894c93b58f6808ff8fe3a63d1bb978">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet_cdf" href="#frechet_cdf" class="anchored unlink"><strong>frechet_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-7b1c306049c2aeb7399a3d6df5ea3c13265def52"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet_lccdf" href="#frechet_lccdf" class="anchored unlink"><strong>frechet_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-81c59a2210fa846a77bc5ceeb512c5ab2403c36d"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet_lcdf" href="#frechet_lcdf" class="anchored unlink"><strong>frechet_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-2e6b14eab76b40cbf5f51add826d06a805ae9757"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet_lpdf" href="#frechet_lpdf" class="anchored unlink"><strong>frechet_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-5b4868d7aafac4f3559c062d04dca87352e5dc4e"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet_lupdf" href="#frechet_lupdf" class="anchored unlink"><strong>frechet_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-99b375230fb5a80dcd8e6c9889b9da0b20cb1901"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="frechet_rng" href="#frechet_rng" class="anchored unlink"><strong>frechet_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-d8574e1b5b0094a19eaec6f9bbc2b458ee6c80d2"><code>(reals alpha, reals sigma) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="g" class="level2">
+<h2 class="anchored" data-anchor-id="g">G</h2>
+<p><a id="gamma" href="#gamma" class="anchored unlink"><strong>gamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-91a7d59d68ed36bf3b8d400643b7211047f9c616">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_cdf" href="#gamma_cdf" class="anchored unlink"><strong>gamma_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-f4d7f090f9393c8fcb25cb5a0a93e6fbfeebbcc3"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_lccdf" href="#gamma_lccdf" class="anchored unlink"><strong>gamma_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-7529e81d26b2ce290a96efca419a000b84883388"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_lcdf" href="#gamma_lcdf" class="anchored unlink"><strong>gamma_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-6a62c558a39dc345eeafa81df49ccbb5e63cff39"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_lpdf" href="#gamma_lpdf" class="anchored unlink"><strong>gamma_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-76500ac07efdc8ff316f8c5d83738088c623861d"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_lupdf" href="#gamma_lupdf" class="anchored unlink"><strong>gamma_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-4631a12c69e80e1a603f5e8455debc9086e93879"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_p" href="#gamma_p" class="anchored unlink"><strong>gamma_p</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-6a3cb9f6d55852cff80a67883177778548b712d0"><code>(real a, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-07d2b4d046ed3e94ab2394782dbfb848ce338126"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_q" href="#gamma_q" class="anchored unlink"><strong>gamma_q</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-5cd1f2570dd2f81dc1a8e6f344ba953142a29865"><code>(real a, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4949830b870b509235b2289a32a4e00f6e52e309"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="gamma_rng" href="#gamma_rng" class="anchored unlink"><strong>gamma_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-f7478aca325aa138c901aac6630912e23ca8a779"><code>(reals alpha, reals beta) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gaussian_dlm_obs" href="#gaussian_dlm_obs" class="anchored unlink"><strong>gaussian_dlm_obs</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-4fd242bf2dcd5be49146d526d8fbe39878489a73">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="gaussian_dlm_obs_lpdf" href="#gaussian_dlm_obs_lpdf" class="anchored unlink"><strong>gaussian_dlm_obs_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-18a51c9f7a0d2635dc8adf29e702d2a918af65ed"><code>(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-8cca79a125123f82ff2e6982bc58099c4b4f2db4"><code>(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="gaussian_dlm_obs_lupdf" href="#gaussian_dlm_obs_lupdf" class="anchored unlink"><strong>gaussian_dlm_obs_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-c55d98e7e02d60b3177dcb12f3f8a4d71d7b7770"><code>(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-14842eb0078230e2e500d5478fb688e1fd8c7fd3"><code>(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="generalized_inverse" href="#generalized_inverse" class="anchored unlink"><strong>generalized_inverse</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-49a205d1f503afa341f4ec2f12b8d9484c5bc3f0"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="generate_laplace_options" href="#generate_laplace_options" class="anchored unlink"><strong>generate_laplace_options</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-60827bf41179724b5e6e8a0fc73da3515b44fc46"><code>(int dimension) : tuple(vector, real, int, int, int, int)</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-d1c421fc2e179180c0a9a45ea4733480792940bc"><code>(vector theta_init) : tuple(vector, real, int, int, int, int)</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="get_imag" href="#get_imag" class="anchored unlink"><strong>get_imag</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-15202c172bf8397f0f3a6ecd5c16fe489df20273"><code>(complex z) : real</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-48dc2caf35c9102abf142e71d6755de12af13e0c"><code>(T x) : T</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="get_real" href="#get_real" class="anchored unlink"><strong>get_real</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-3c18c19d9aa231a8f1a921c7f6c289ef225ae34f"><code>(complex z) : real</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-2562277862b92c4536cb33d80bfed98e8a586d81"><code>(T x) : T</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gp_dot_prod_cov" href="#gp_dot_prod_cov" class="anchored unlink"><strong>gp_dot_prod_cov</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7d386a186ed2caf01b35da180fe71bf966425b74"><code>(array[] real x, real sigma) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0e85dcc0b82afffce83d044f3699f82c43cda4fc"><code>(array[] real x1, array[] real x2, real sigma) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b77643cdb29d01e2d4846872277e01f4d9e4a17c"><code>(vectors x, real sigma) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5929a214570377c8b9d1fe93d53acf0d146a7686"><code>(vectors x1, vectors x2, real sigma) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gp_exp_quad_cov" href="#gp_exp_quad_cov" class="anchored unlink"><strong>gp_exp_quad_cov</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a59f2c545ffe0d99724685b78624ca3c6e6687dd"><code>(array[] real x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-efeafc090e77d1a04aa32be00cc0f1f697dd613c"><code>(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2a5cce0c9c4a586fd14ce028b6ece6593c0863c7"><code>(vectors x, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f9f7059ee7dbdb3c7ec60022629df94df4366eda"><code>(vectors x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0171054e36a54ff590742f7f3dfb18db315818ed"><code>(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gp_exponential_cov" href="#gp_exponential_cov" class="anchored unlink"><strong>gp_exponential_cov</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-825dd9547256f5dcb295bf954db6f374cf924944"><code>(array[] real x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-178aa0d85f72dea02342080bc2c8d9e0236570c1"><code>(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4f6f2b3dfcd6826bfc4a157531179eaf4da21e17"><code>(vectors x, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-467147dd6252c7d33347630a0101aae12325838e"><code>(vectors x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-78a300ec67a8ef9d073a915b1112c99d5439ebf2"><code>(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9299d6de2d2e6c2e7be0930c480e0cf59f93e1d8"><code>(vectors x1, vectors x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gp_matern23_cov" href="#gp_matern23_cov" class="anchored unlink"><strong>gp_matern23_cov</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c67bd8178298aaea402200c34fc4e58b5f716374"><code>(array[] real x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-77c38f1e81ceac337fd3248ff042bb819cbe5e6f"><code>(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-df0ecce0aa7b3af55d9c7e160b4e1c96caa7b1eb"><code>(vectors x, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4049f44098827dd5df10dfe260933523d87ff79a"><code>(vectors x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-34469df0872fa77c6199a495b6bedc9803d186a4"><code>(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c0967d2d9deec4cab7150c95aa49dc8633d2f6af"><code>(vectors x1, vectors x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gp_matern52_cov" href="#gp_matern52_cov" class="anchored unlink"><strong>gp_matern52_cov</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-48da2b928754ae4e47279653b6b2dc8a1849fd3e"><code>(array[] real x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5e89dd9a9be3e6c08de8184961c03cf9bc6e1aa4"><code>(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-21a29eaabce12eae5d070311d157793c941e98bd"><code>(vectors x, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-e58aa2e1105325707d54de56f9bf8f8d0400e801"><code>(vectors x, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f40fe74b859bd5fdf393b580e35eeba5ee0c15da"><code>(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ba526a205e3145ede4f46b6211f8e2cc4a6580f7"><code>(vectors x1, vectors x2, real sigma, real length_scale) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gp_periodic_cov" href="#gp_periodic_cov" class="anchored unlink"><strong>gp_periodic_cov</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6707fa8fb889db9572c791a5ec6e7ead0310ba3f"><code>(array[] real x, real sigma, real length_scale, real period) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7f2049d9d9a6b2a24a7765d9f5af9b6df966a8b4"><code>(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c89ef06585db1d6db2d29210cae24aada2fa7d9f"><code>(vectors x, real sigma, real length_scale, real period) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3d85d317bf98b3cbd503a360059bfb1121a22a98"><code>(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel" href="#gumbel" class="anchored unlink"><strong>gumbel</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-846f4cd18ce2c11c64adc5430feb805c8548af12">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel_cdf" href="#gumbel_cdf" class="anchored unlink"><strong>gumbel_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-d75196df1f351627464af18ef549a68bbb63bf5e"><code>(reals y | reals mu, reals beta) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel_lccdf" href="#gumbel_lccdf" class="anchored unlink"><strong>gumbel_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-c56b96f1fa2edfa9edfa06313030df780f89753c"><code>(reals y | reals mu, reals beta) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel_lcdf" href="#gumbel_lcdf" class="anchored unlink"><strong>gumbel_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4cc1f4c7bfa6eedc97fbb6074ec3965cff5be24f"><code>(reals y | reals mu, reals beta) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel_lpdf" href="#gumbel_lpdf" class="anchored unlink"><strong>gumbel_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-93b41c8b64bc3c223c99733bf33da8c3ec9ac492"><code>(reals y | reals mu, reals beta) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel_lupdf" href="#gumbel_lupdf" class="anchored unlink"><strong>gumbel_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-b9841ea9c5279d1940a06429efc4126c8f1735d9"><code>(reals y | reals mu, reals beta) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="gumbel_rng" href="#gumbel_rng" class="anchored unlink"><strong>gumbel_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f0072bf29736a2f234a30ab11de070e3194bd4e3"><code>(reals mu, reals beta) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="h" class="level2">
+<h2 class="anchored" data-anchor-id="h">H</h2>
+<p><a id="head" href="#head" class="anchored unlink"><strong>head</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-57085692c2de5cb3a0be6af59e4d4db1087270b9"><code>(array[] T sv, int n) : array[] T</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-9b30a21ff61d6bb263aeae6fe0f8eeaad19b6091"><code>(complex_row_vector rv, int n) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-98d4004db4f7912b486209fbaa83ee186a0b4a0f"><code>(complex_vector v, int n) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3107ded9497fc42a02b0573d3453f8cfb6302a41"><code>(row_vector rv, int n) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8422dde5d63a6c756136d01ceeae0de6de1b00cb"><code>(vector v, int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="hmm_hidden_state_prob" href="#hmm_hidden_state_prob" class="anchored unlink"><strong>hmm_hidden_state_prob</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/hidden_markov_models.html#index-entry-2ff49049acfb95386aefa57a16ea987ee4b7b68e"><code>(matrix log_omega, matrix Gamma, vector rho) : matrix</code></a> <span class="detail">(hidden_markov_models.html)</span>
+</div></li>
+</ul>
+<p><a id="hmm_latent_rng" href="#hmm_latent_rng" class="anchored unlink"><strong>hmm_latent_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/hidden_markov_models.html#index-entry-48084fc56672a13c812780b444bee6bc9b630cc2"><code>(matrix log_omega, matrix Gamma, vector rho) : array[] int</code></a> <span class="detail">(hidden_markov_models.html)</span>
+</div></li>
+</ul>
+<p><a id="hmm_marginal" href="#hmm_marginal" class="anchored unlink"><strong>hmm_marginal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/hidden_markov_models.html#index-entry-6569db76a2047e7ba00db34ba2b0b2e000bd4eaa"><code>(matrix log_omega, matrix Gamma, vector rho) : real</code></a> <span class="detail">(hidden_markov_models.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric" href="#hypergeometric" class="anchored unlink"><strong>hypergeometric</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2fc02e8425e812e91f8ef47c9f53bd538c248cbd">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_1F0" href="#hypergeometric_1F0" class="anchored unlink"><strong>hypergeometric_1F0</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-e29b2c1e8db395e7d15862eb2f1f1e63ac58b10f"><code>(real a, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_2F1" href="#hypergeometric_2F1" class="anchored unlink"><strong>hypergeometric_2F1</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4ebcc0750c9365f9fd711bbd07b560cd76a3c016"><code>(real a1, real a2, real b1, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_3F2" href="#hypergeometric_3F2" class="anchored unlink"><strong>hypergeometric_3F2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f578817f233389fabf92d2db8bcfc1878fec4eca"><code>(T1 a, T2 b, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_lpmf" href="#hypergeometric_lpmf" class="anchored unlink"><strong>hypergeometric_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-74678ac15a9ad9dc38759c7f81421cd6abe7f8e8"><code>(int n | int N, int a, int b) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_lupmf" href="#hypergeometric_lupmf" class="anchored unlink"><strong>hypergeometric_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-bd50c4360d03f67c2f750c84c2b4b7f53ade58e9"><code>(int n | int N, int a, int b) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_pFq" href="#hypergeometric_pFq" class="anchored unlink"><strong>hypergeometric_pFq</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-2177c85f2269dea084da1c4e5b2a6f52c3337579"><code>(T1 a, T2 b, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypergeometric_rng" href="#hypergeometric_rng" class="anchored unlink"><strong>hypergeometric_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-b7d868252dbebe75b731f601d6af7252a228c9bb"><code>(int N, int a, int2 b) : int</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="hypot" href="#hypot" class="anchored unlink"><strong>hypot</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-91e267ebdf31823d577cac8796d2fc05ffc4123b"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d55f4ca2bc092698d4e852dae4516f76ef5e8738"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="i" class="level2">
+<h2 class="anchored" data-anchor-id="i">I</h2>
+<p><a id="identity_matrix" href="#identity_matrix" class="anchored unlink"><strong>identity_matrix</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7751d8abe876b03fe020df0799dd498b2be022a8"><code>(int k) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="inc_beta" href="#inc_beta" class="anchored unlink"><strong>inc_beta</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8af921f70494e8dac40b7e567b55041a92982782"><code>(real alpha, real beta, real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="int_step" href="#int_step" class="anchored unlink"><strong>int_step</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-317ccd335f85a10d3c7902b15dd6af1f6ef5d16c"><code>(int x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-ee3b76044455776615f3a8a6106dccb8ad9a235d"><code>(real x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="integrate_1d" href="#integrate_1d" class="anchored unlink"><strong>integrate_1d</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-39f0cbb60146855f8f98ba56502efcbfd04df2aa"><code>(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-029333bb86d68fe853789eea2568f760cec8eff3"><code>(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="integrate_ode" href="#integrate_ode" class="anchored unlink"><strong>integrate_ode</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-f2263dbf460fa8a8e16c36bec1723e7cb60e8ff7"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="integrate_ode_adams" href="#integrate_ode_adams" class="anchored unlink"><strong>integrate_ode_adams</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-0243ee75a588fbb73c6e50121f71844a6cbfb6e6"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-f382cd6903765b51897e2aa1b1eaa55cf63fd4d6"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="integrate_ode_bdf" href="#integrate_ode_bdf" class="anchored unlink"><strong>integrate_ode_bdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-0cb2d79940f9e180689b48555a1a25401db9b41a"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-3766842394a3f0b7eb3333fb202227f261c1f535"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="integrate_ode_rk45" href="#integrate_ode_rk45" class="anchored unlink"><strong>integrate_ode_rk45</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-fddefd52462ff3a8c14cca8da72bb5ef1596fbe0"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/deprecated_functions.html#index-entry-ed3282ca9353c591d894371c5da7f6c427ec52ff"><code>(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real</code></a> <span class="detail">(deprecated_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv" href="#inv" class="anchored unlink"><strong>inv</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7e3cb8bb5e02b0568495e3495ab81c8ab671dd96"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square" href="#inv_chi_square" class="anchored unlink"><strong>inv_chi_square</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-2c55516e33a8c93de05a426a4331b0fd2c73665c">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square_cdf" href="#inv_chi_square_cdf" class="anchored unlink"><strong>inv_chi_square_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-a5b53196a941f74078d28a76d43875006ecaa749"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square_lccdf" href="#inv_chi_square_lccdf" class="anchored unlink"><strong>inv_chi_square_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-5fb795e9994cfaedf511409172a6457e8df06bff"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square_lcdf" href="#inv_chi_square_lcdf" class="anchored unlink"><strong>inv_chi_square_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-19a25046f012bc203f47cddbd716160311344693"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square_lpdf" href="#inv_chi_square_lpdf" class="anchored unlink"><strong>inv_chi_square_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-c49192fa8c8d7405bca27fccb88443c3a31b82f9"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square_lupdf" href="#inv_chi_square_lupdf" class="anchored unlink"><strong>inv_chi_square_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-a5086fcc0bf03bde2c07057468cd23a2936131b7"><code>(reals y | reals nu) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_chi_square_rng" href="#inv_chi_square_rng" class="anchored unlink"><strong>inv_chi_square_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-b2c4491a4b62bfff6789788b555b1de2f50ad833"><code>(reals nu) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_cloglog" href="#inv_cloglog" class="anchored unlink"><strong>inv_cloglog</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-40b628501a54a1c1fc2eed89973daabf330a9c2c"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_erfc" href="#inv_erfc" class="anchored unlink"><strong>inv_erfc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-436991c650707e4567f4627cf8c7216e20bd1d68"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_fft" href="#inv_fft" class="anchored unlink"><strong>inv_fft</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-27a0956f66e4b8b0fc804abec9e5245baa2706aa"><code>(complex_vector u) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_fft2" href="#inv_fft2" class="anchored unlink"><strong>inv_fft2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-81586d35e7a520465df734fc44af4a3ad069f695"><code>(complex_matrix m) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma" href="#inv_gamma" class="anchored unlink"><strong>inv_gamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-0a75a7a89dfd9fb418dfcd098f5f987d27c95da2">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma_cdf" href="#inv_gamma_cdf" class="anchored unlink"><strong>inv_gamma_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-8b7859ab9bdd43c5cb4a55298dbada0eb100e006"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma_lccdf" href="#inv_gamma_lccdf" class="anchored unlink"><strong>inv_gamma_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-779e4be4cd1ce9f8858d662138e8af3a0a221e5e"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma_lcdf" href="#inv_gamma_lcdf" class="anchored unlink"><strong>inv_gamma_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-24ddd0a34b2809b8f5afa3c540f1223c8cf683f8"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma_lpdf" href="#inv_gamma_lpdf" class="anchored unlink"><strong>inv_gamma_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-7c937cfca51ac916d72ff10b1184a8d6d39ee650"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma_lupdf" href="#inv_gamma_lupdf" class="anchored unlink"><strong>inv_gamma_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-c799b2a02063b44bf348b1e09d1a2bdbb8fefbd5"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_gamma_rng" href="#inv_gamma_rng" class="anchored unlink"><strong>inv_gamma_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-460487d2c9969e63b7c9f8ff329c0295b8807239"><code>(reals alpha, reals beta) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_inc_beta" href="#inv_inc_beta" class="anchored unlink"><strong>inv_inc_beta</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-75728a2fe483e42a84c2fb5562469cb0fcd56250"><code>(real alpha, real beta, real p) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_logit" href="#inv_logit" class="anchored unlink"><strong>inv_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-587aa4ba61e421c4ca21bad5c94b5b919e16d34d"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_Phi" href="#inv_Phi" class="anchored unlink"><strong>inv_Phi</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f0054cc2edcc0949c41276325f6db59b8e0c4e15"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_sqrt" href="#inv_sqrt" class="anchored unlink"><strong>inv_sqrt</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4dcb1d28ce2c1de0b6668a4095511ceeadb29fd0"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_square" href="#inv_square" class="anchored unlink"><strong>inv_square</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b4f13710242aa2542c07afc496e1966099678f35"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart" href="#inv_wishart" class="anchored unlink"><strong>inv_wishart</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-3961ee487e240f180388c9629267c086f03aceed">distribution statement</a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart_cholesky_lpdf" href="#inv_wishart_cholesky_lpdf" class="anchored unlink"><strong>inv_wishart_cholesky_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-0d09dfe80fda861a5757201ffaddddac82e00aa5"><code>(matrix L_W | real nu, matrix L_S) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart_cholesky_lupdf" href="#inv_wishart_cholesky_lupdf" class="anchored unlink"><strong>inv_wishart_cholesky_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-eaf70d1f046e96371485e6dd22f5b0fe44e33fdf"><code>(matrix L_W | real nu, matrix L_S) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart_cholesky_rng" href="#inv_wishart_cholesky_rng" class="anchored unlink"><strong>inv_wishart_cholesky_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-327529ac254bd3b80d8b9c48d711700d3ed47fa3"><code>(real nu, matrix L_S) : matrix</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart_lpdf" href="#inv_wishart_lpdf" class="anchored unlink"><strong>inv_wishart_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-38e0ab3788169c92def514827eaeff93db4d3af3"><code>(matrix W | real nu, matrix Sigma) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart_lupdf" href="#inv_wishart_lupdf" class="anchored unlink"><strong>inv_wishart_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-2e69b51d6833e168c6a6d9646e93f6fb883609bd"><code>(matrix W | real nu, matrix Sigma) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inv_wishart_rng" href="#inv_wishart_rng" class="anchored unlink"><strong>inv_wishart_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-ed1a3712028dbe64ab7bebe658f72ad4d101d3cf"><code>(real nu, matrix Sigma) : matrix</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="inverse" href="#inverse" class="anchored unlink"><strong>inverse</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4f71689bce62cc64928a17fe2768a2a243320ff1"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="inverse_spd" href="#inverse_spd" class="anchored unlink"><strong>inverse_spd</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-cf78ebfad5c78b789158eb47797d8c00862001b3"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="is_inf" href="#is_inf" class="anchored unlink"><strong>is_inf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7240fd89c6199124fb25731ffcdb1459029b04df"><code>(real x) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="is_nan" href="#is_nan" class="anchored unlink"><strong>is_nan</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d9e6d18b33b36e488e49f87c4efd470c39e6a5b1"><code>(real x) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="l" class="level2">
+<h2 class="anchored" data-anchor-id="l">L</h2>
+<p><a id="lambert_w0" href="#lambert_w0" class="anchored unlink"><strong>lambert_w0</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8762bfe8eb64d40ae8bf46eb733dc054da100f51"><code>(reals x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lambert_wm1" href="#lambert_wm1" class="anchored unlink"><strong>lambert_wm1</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-ab5ca07ba7ba53020939ca3ba7c6e64ccf05cf19"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_bernoulli_logit_rng" href="#laplace_latent_bernoulli_logit_rng" class="anchored unlink"><strong>laplace_latent_bernoulli_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-59f1ae97dbd720c3dd7cd15f92e7e3c6f93ae22e"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_neg_binomial_2_log_rng" href="#laplace_latent_neg_binomial_2_log_rng" class="anchored unlink"><strong>laplace_latent_neg_binomial_2_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-79fe43b1435d4b3821038d2cd934d8c136e8808c"><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_poisson_log_rng" href="#laplace_latent_poisson_log_rng" class="anchored unlink"><strong>laplace_latent_poisson_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-a6bf063f06956735a484fd5e960352833373ec15"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_rng" href="#laplace_latent_rng" class="anchored unlink"><strong>laplace_latent_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-60d4efd35a460871aff97a758f5d366faee26465"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_rng_tol" href="#laplace_latent_rng_tol" class="anchored unlink"><strong>laplace_latent_rng_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-eb06251dd7ccd5f9b7882b7e6950fa99cc0c2d57"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_tol_bernoulli_logit_rng" href="#laplace_latent_tol_bernoulli_logit_rng" class="anchored unlink"><strong>laplace_latent_tol_bernoulli_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-4566ad7c6cd5f0ee057539a6e2a0bea28554f082"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_tol_neg_binomial_2_log_rng" href="#laplace_latent_tol_neg_binomial_2_log_rng" class="anchored unlink"><strong>laplace_latent_tol_neg_binomial_2_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-259ff7a39109d52d1194cf72a125b72127ddd2b1"><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_tol_poisson_log_rng" href="#laplace_latent_tol_poisson_log_rng" class="anchored unlink"><strong>laplace_latent_tol_poisson_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-8fec77181949b3ca47b1916246cd0384394f87bc"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal" href="#laplace_marginal" class="anchored unlink"><strong>laplace_marginal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-7ffad9885f7c8f209fffbfae2182cbc3ac432b59"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_bernoulli_logit" href="#laplace_marginal_bernoulli_logit" class="anchored unlink"><strong>laplace_marginal_bernoulli_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-1d93ab799518d0aac88e63c01f9655f36c7cbeb6">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_bernoulli_logit_lpmf" href="#laplace_marginal_bernoulli_logit_lpmf" class="anchored unlink"><strong>laplace_marginal_bernoulli_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-fd129c664f37d1cdd280abaf7ee82bfaf9c8e54e"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_bernoulli_logit_lupmf" href="#laplace_marginal_bernoulli_logit_lupmf" class="anchored unlink"><strong>laplace_marginal_bernoulli_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-53e5e8065cf24c3e72a435d5fccf4d14a587b245"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_neg_binomial_2_log" href="#laplace_marginal_neg_binomial_2_log" class="anchored unlink"><strong>laplace_marginal_neg_binomial_2_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-7f807083cfb7347ece100da5dfe719186a678d02">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_neg_binomial_2_log_lpmf" href="#laplace_marginal_neg_binomial_2_log_lpmf" class="anchored unlink"><strong>laplace_marginal_neg_binomial_2_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-b74e5e572a36211903f16f5deed7158384c21355"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_neg_binomial_2_log_lupmf" href="#laplace_marginal_neg_binomial_2_log_lupmf" class="anchored unlink"><strong>laplace_marginal_neg_binomial_2_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-5d9163f94a110e178c54b2bd192e9aded134cd3a"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_poisson_log" href="#laplace_marginal_poisson_log" class="anchored unlink"><strong>laplace_marginal_poisson_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-b2c6946b65561039eb8bf3999fb59d38a6336e10">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_poisson_log_lpmf" href="#laplace_marginal_poisson_log_lpmf" class="anchored unlink"><strong>laplace_marginal_poisson_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-ca1a086a83293f252043ab8b127ec6c9ef16deec"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_poisson_log_lupmf" href="#laplace_marginal_poisson_log_lupmf" class="anchored unlink"><strong>laplace_marginal_poisson_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-d0ecc660f5107f7409bd76ba059c5ea82ec81ba0"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol" href="#laplace_marginal_tol" class="anchored unlink"><strong>laplace_marginal_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-9d40f0ee623d8ad92f054b2d83550d7a8ddf4b56"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_bernoulli_logit" href="#laplace_marginal_tol_bernoulli_logit" class="anchored unlink"><strong>laplace_marginal_tol_bernoulli_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-92d43f7c3643c7d85966bbfc0b2a71546facb37c">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_bernoulli_logit_lpmf" href="#laplace_marginal_tol_bernoulli_logit_lpmf" class="anchored unlink"><strong>laplace_marginal_tol_bernoulli_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-5126f7d7829072e478ea2f7257d226219b43e1d4"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_bernoulli_logit_lupmf" href="#laplace_marginal_tol_bernoulli_logit_lupmf" class="anchored unlink"><strong>laplace_marginal_tol_bernoulli_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-7d93f40162970812a81fc61df49dcb53f993e293"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_neg_binomial_2_log" href="#laplace_marginal_tol_neg_binomial_2_log" class="anchored unlink"><strong>laplace_marginal_tol_neg_binomial_2_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-40016f7d5d96b9d999ca42f446720cf3f3cd5769">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_neg_binomial_2_log_lpmf" href="#laplace_marginal_tol_neg_binomial_2_log_lpmf" class="anchored unlink"><strong>laplace_marginal_tol_neg_binomial_2_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-cefee8b5275ca255f121386fb14c9354cf95594c"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_neg_binomial_2_log_lupmf" href="#laplace_marginal_tol_neg_binomial_2_log_lupmf" class="anchored unlink"><strong>laplace_marginal_tol_neg_binomial_2_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-efb046fcbadbcaaa59a274ab3c5b1ed15f23add9"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_poisson_log" href="#laplace_marginal_tol_poisson_log" class="anchored unlink"><strong>laplace_marginal_tol_poisson_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-4a7ecca812a308ea648bb31996559cba79738d83">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_poisson_log_lpmf" href="#laplace_marginal_tol_poisson_log_lpmf" class="anchored unlink"><strong>laplace_marginal_tol_poisson_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-51cad53596dd0d74c0fe99b84a3c975cbec3866a"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_poisson_log_lupmf" href="#laplace_marginal_tol_poisson_log_lupmf" class="anchored unlink"><strong>laplace_marginal_tol_poisson_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-23e9ed91fd3f98c314ddbce86a77005fdf8ed5f3"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="lbeta" href="#lbeta" class="anchored unlink"><strong>lbeta</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-def48992e0f8724381904fba78466b0f4a0d14bd"><code>(real alpha, real beta) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-ffad8a6ffd89011bffb5c4e0eb7817d550b4d438"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lchoose" href="#lchoose" class="anchored unlink"><strong>lchoose</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8db53df4cf96e86f2466a1a67a046f1e92e24a41"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-57668b872958a1c748381739b2639e2c592b1635"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ldexp" href="#ldexp" class="anchored unlink"><strong>ldexp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-87e15a5115cbb3320b6c75601022db5ce42d567f"><code>(real x, int y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-04b6e7e9f36a90c1074fe831b302e5a839cded46"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lgamma" href="#lgamma" class="anchored unlink"><strong>lgamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-51737f6b8d61ebaae578b09d29899a9f72cc8dff"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="linspaced_array" href="#linspaced_array" class="anchored unlink"><strong>linspaced_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-71ca78d469339eb375a0b3e4fd656056c1f4bc37"><code>(int n, data real lower, data real upper) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="linspaced_int_array" href="#linspaced_int_array" class="anchored unlink"><strong>linspaced_int_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8ca31c5441d89c38eb6df32f4e84fa99f2542229"><code>(int n, int lower, int upper) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="linspaced_row_vector" href="#linspaced_row_vector" class="anchored unlink"><strong>linspaced_row_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-acc0b0d5ab9d4a99bd5c019f6b645424790aa4e9"><code>(int n, data real lower, data real upper) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="linspaced_vector" href="#linspaced_vector" class="anchored unlink"><strong>linspaced_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3f3107a87c63d9fe79a69329003d27f4b107d02e"><code>(int n, data real lower, data real upper) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr" href="#lkj_corr" class="anchored unlink"><strong>lkj_corr</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-beb0c10c5b7b26c3295ff6f47561e58ffc211a45">distribution statement</a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_cholesky" href="#lkj_corr_cholesky" class="anchored unlink"><strong>lkj_corr_cholesky</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-d3519d9e1f129c4f006c23bcd5995f869bcfb56d">distribution statement</a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_cholesky_lpdf" href="#lkj_corr_cholesky_lpdf" class="anchored unlink"><strong>lkj_corr_cholesky_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-03f62c1ae419e1804760f00f477c89ad5911c97a"><code>(matrix L | real eta) : real</code></a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_cholesky_lupdf" href="#lkj_corr_cholesky_lupdf" class="anchored unlink"><strong>lkj_corr_cholesky_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-d612c55f03ba91b1722cbfc3ccd917a5fd209b0b"><code>(matrix L | real eta) : real</code></a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_cholesky_rng" href="#lkj_corr_cholesky_rng" class="anchored unlink"><strong>lkj_corr_cholesky_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-8efcbe3b1b5dfa560b74b1a1642f6f855f32c26f"><code>(int K, real eta) : matrix</code></a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_lpdf" href="#lkj_corr_lpdf" class="anchored unlink"><strong>lkj_corr_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-34f4437ba5f424e835889403cf07a8d4f63921f2"><code>(matrix y | real eta) : real</code></a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_lupdf" href="#lkj_corr_lupdf" class="anchored unlink"><strong>lkj_corr_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-61643c9cb87b1ed68a761b62034abf5253ca8f38"><code>(matrix y | real eta) : real</code></a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lkj_corr_rng" href="#lkj_corr_rng" class="anchored unlink"><strong>lkj_corr_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/correlation_matrix_distributions.html#index-entry-6af330a1989d2c53be55e7707c77a9e616e1eba2"><code>(int K, real eta) : matrix</code></a> <span class="detail">(correlation_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lmgamma" href="#lmgamma" class="anchored unlink"><strong>lmgamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d6d5bb15c2402be65d4a671ea3ee4462f2f96da9"><code>(int n, real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f8c2c4db4a7befdf0727ddb48d54d9de4cf442a1"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lmultiply" href="#lmultiply" class="anchored unlink"><strong>lmultiply</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d6659cf1923baf938d87ae82cf8fecf0665eb99d"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-5789074da7a4d0fb0e2acfe2db85408ed582f1a5"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log" href="#log" class="anchored unlink"><strong>log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-2ae2007325e8a3d31124d85afaa37799bb6c8ea6"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-656b9ebaf2d3c913d48b33e37256a7fd908c2c74"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log10" href="#log10" class="anchored unlink"><strong>log10</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-58b920bd8ecd5758726663c1663723631b6e1040"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-4d3907a677d8f4d4f89b1edc324ffe5fa2bfbafe"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-ad4651b8f653b4d8e1e6e50131371c8be7a35444"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log1m" href="#log1m" class="anchored unlink"><strong>log1m</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-44e5298adea287581561adf0bc2875d063d97713"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log1m_exp" href="#log1m_exp" class="anchored unlink"><strong>log1m_exp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d3d38c55ec9be3639895f7ff75fac4b6b765c29a"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log1m_inv_logit" href="#log1m_inv_logit" class="anchored unlink"><strong>log1m_inv_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-235d149b04f04823876ae89ac37a083db5908ce8"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log1p" href="#log1p" class="anchored unlink"><strong>log1p</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-95ced8098d5ed0668cffa0d1be3921eed95f1d76"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log1p_exp" href="#log1p_exp" class="anchored unlink"><strong>log1p_exp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4f41ed63ba44a1e663dcc15f1dddbea84a2e1fde"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log2" href="#log2" class="anchored unlink"><strong>log2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8e60dd0524fa17b2bb9c1df45ec2cb91f313d8f6"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8aa277afaded318050922a7e576d5531918c4fa3"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_determinant" href="#log_determinant" class="anchored unlink"><strong>log_determinant</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9e3fc21034e716f9152396c5bc91933e51a5adea"><code>(matrix A) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="log_diff_exp" href="#log_diff_exp" class="anchored unlink"><strong>log_diff_exp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-db5adfe0722d967066a4801ba7088570feda3d4f"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-a68b0818c49fbd4972dfb2a14fea67387662718e"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_falling_factorial" href="#log_falling_factorial" class="anchored unlink"><strong>log_falling_factorial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-914b5210fcade0135a4887b4af7ce01cf07f84bf"><code>(real x, real n) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_inv_logit" href="#log_inv_logit" class="anchored unlink"><strong>log_inv_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-35e5ab34a9cdca5c2c6b281a69b7bff27a2a200b"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_inv_logit_diff" href="#log_inv_logit_diff" class="anchored unlink"><strong>log_inv_logit_diff</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-10df9e89f06a7af185332206207ab46c15793245"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_mix" href="#log_mix" class="anchored unlink"><strong>log_mix</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-569bcea73548d03bb5791681fcc60ee056c259b3"><code>(real theta, real lp1, real lp2) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-83690ccdb1ecc891d26739b85cb08f317eece075"><code>(T1 thetas, T2 lps) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_modified_bessel_first_kind" href="#log_modified_bessel_first_kind" class="anchored unlink"><strong>log_modified_bessel_first_kind</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-832bf9903539f39025b4823c5ad6ff48bf17ccad"><code>(real v, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-f338c7d47a8fdaa9b97a44fadfdd5fb71ceccb3b"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_rising_factorial" href="#log_rising_factorial" class="anchored unlink"><strong>log_rising_factorial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-72c705188e9fa2abc1fb72e724383889c9c3bc26"><code>(real x, real n) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-67ec81c237fb98951165722160d48a4e93bbf167"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="log_softmax" href="#log_softmax" class="anchored unlink"><strong>log_softmax</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-49224e65079937569f0952ce5740c03bd8dcb98a"><code>(vector x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="log_sum_exp" href="#log_sum_exp" class="anchored unlink"><strong>log_sum_exp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-7b26d9d0a4894d9b4686f4847de2d728f050eeca"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-54afbd4baa0c2ccf7b5392bdff0c6c87fce8222c"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-41527e620b092f6ca6efaca55f839d7ff1404f5b"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-da09d53c7797409929bfa868dfa8b30d767c1d2b"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-12d207b1c215fc3ea15553df653cae3cbb14fbdf"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic" href="#logistic" class="anchored unlink"><strong>logistic</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-83ebb5b2c1c5976cb145e5c49c438dcedef9d5cc">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic_cdf" href="#logistic_cdf" class="anchored unlink"><strong>logistic_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4883524cdc0ee6a2baed3cdf4c56a2221fc3c5e4"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic_lccdf" href="#logistic_lccdf" class="anchored unlink"><strong>logistic_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-6cce712fc62f4073bbd1140a1dacdff12e0f72ce"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic_lcdf" href="#logistic_lcdf" class="anchored unlink"><strong>logistic_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-ae238a73fddb4ade983ee9f6ff91cdd7dcc57968"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic_lpdf" href="#logistic_lpdf" class="anchored unlink"><strong>logistic_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-defc2f03147e4fc4d6ea7092ba14bfa3e211169e"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic_lupdf" href="#logistic_lupdf" class="anchored unlink"><strong>logistic_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-77bd7fb60be125c2b18d3f44f85bc656a0b9b919"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logistic_rng" href="#logistic_rng" class="anchored unlink"><strong>logistic_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-9327491d09f9d375a4734c67a3727ca525f940f5"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="logit" href="#logit" class="anchored unlink"><strong>logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b473710a74faf02bff89cd1f779befb3517d0ba6"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="loglogistic" href="#loglogistic" class="anchored unlink"><strong>loglogistic</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-34bd0bbc32687356d26bf7bc8adb2ed107ee7c41">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="loglogistic_cdf" href="#loglogistic_cdf" class="anchored unlink"><strong>loglogistic_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-b4193e37a39fc626b93a8d5b938675e6a5f682ef"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="loglogistic_lpdf" href="#loglogistic_lpdf" class="anchored unlink"><strong>loglogistic_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-54448e697522e5d4ca36ca4b7acca93cda712054"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="loglogistic_rng" href="#loglogistic_rng" class="anchored unlink"><strong>loglogistic_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-d1b23b045cd0f829704373f8a06d3f077646b0d0"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal" href="#lognormal" class="anchored unlink"><strong>lognormal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-96de647c4a4c68e79cbd1680d0976df411556c5f">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal_cdf" href="#lognormal_cdf" class="anchored unlink"><strong>lognormal_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-765ee6d534fa4f5fb5e2d621a21456071b4c8458"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal_lccdf" href="#lognormal_lccdf" class="anchored unlink"><strong>lognormal_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-64e46338c2669c1f1b399608c9d1c654bc02710b"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal_lcdf" href="#lognormal_lcdf" class="anchored unlink"><strong>lognormal_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-93c2a2ecc235e787a00b95dad3075afbbf4a21db"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal_lpdf" href="#lognormal_lpdf" class="anchored unlink"><strong>lognormal_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-ce67c5b635189c70cda3beeadc6fd4238ba2b535"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal_lupdf" href="#lognormal_lupdf" class="anchored unlink"><strong>lognormal_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-3d3d0388e0be101f05b8314d8bed921fef2a999b"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lognormal_rng" href="#lognormal_rng" class="anchored unlink"><strong>lognormal_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-ea103d5de1264a7df987cedd3cb455d3302557af"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="lower_bound_constrain" href="#lower_bound_constrain" class="anchored unlink"><strong>lower_bound_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-e152c4494a196c9bd932faad21a3555c712d102c"><code>(reals y, reals lb) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lower_bound_jacobian" href="#lower_bound_jacobian" class="anchored unlink"><strong>lower_bound_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-dda4a6ffc09ae6ffec4218c092003fea973c174e"><code>(reals y, reals lb) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lower_bound_unconstrain" href="#lower_bound_unconstrain" class="anchored unlink"><strong>lower_bound_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-d178c95f782831ea0616fdc58b25d16826f98164"><code>(reals x, reals lb) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lower_upper_bound_constrain" href="#lower_upper_bound_constrain" class="anchored unlink"><strong>lower_upper_bound_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-d44d27ade0aed4fc8aa29d480aa1cbd767281f72"><code>(reals y, reals lb, reals ub) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lower_upper_bound_jacobian" href="#lower_upper_bound_jacobian" class="anchored unlink"><strong>lower_upper_bound_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-fa7585cac2eb147a254907e9aacffb346393c41e"><code>(reals y, reals lb, reals ub) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="lower_upper_bound_unconstrain" href="#lower_upper_bound_unconstrain" class="anchored unlink"><strong>lower_upper_bound_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-da54a8c2ddfb181210e40b0e47d122f16713af86"><code>(reals x, reals lb, reals ub) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="m" class="level2">
+<h2 class="anchored" data-anchor-id="m">M</h2>
+<p><a id="machine_precision" href="#machine_precision" class="anchored unlink"><strong>machine_precision</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b5f5a8656ff1f34e85d53f1ed43ff61eef17279b"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="map_rect" href="#map_rect" class="anchored unlink"><strong>map_rect</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-6286c3d1b323b131fcba8e6c7123559540554abb"><code>(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="matrix_exp" href="#matrix_exp" class="anchored unlink"><strong>matrix_exp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8aafa05928b6e41fc186f3ab4600c607bd5dd27c"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="matrix_exp_multiply" href="#matrix_exp_multiply" class="anchored unlink"><strong>matrix_exp_multiply</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-07146ab787c63991cae5bac69256c64bc7dc73e5"><code>(matrix A, matrix B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="matrix_power" href="#matrix_power" class="anchored unlink"><strong>matrix_power</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4343283f37a884cc983cd269f8a680db1f8f7a9a"><code>(matrix A, int B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="max" href="#max" class="anchored unlink"><strong>max</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-b77840f6a6da285fe070a2db533be9470af54dfc"><code>(array[] int x) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-1dc0a9e5bd6f78b5d2904cba8c6375fa264e6990"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-fbc1a906ce513495836edb5c1cd222c564bcc768"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3f0b6dd3ba52477b4fef8985ad1e0a330b31afaf"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-260d6b034f7c7e6500278a93262c8d5af5ebe636"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f2cd3ae46c31e2abdd35a92b4d10506f3c51f553"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="mdivide_left_spd" href="#mdivide_left_spd" class="anchored unlink"><strong>mdivide_left_spd</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-cf5f3991e5dd2accb21ce017e9f96521b0735e68"><code>(matrix A, matrix B) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0faeb825bd118874deb217de05883995b606c1e7"><code>(matrix A, vector b) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="mdivide_left_tri_low" href="#mdivide_left_tri_low" class="anchored unlink"><strong>mdivide_left_tri_low</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-871e5638d23fa382b4f449a8fa0a5fa1c9556ecd"><code>(matrix A, matrix B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-713d2983956ecba7e2e36a9e0614def9ea574bc6"><code>(matrix A, vector b) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="mdivide_right_spd" href="#mdivide_right_spd" class="anchored unlink"><strong>mdivide_right_spd</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-835081fc5b867bbda5d6b16d5849e6ecbe5f952c"><code>(matrix B, matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-d0585dccea5c1361c5123a20b0b892e33c4b3ad7"><code>(row_vector b, matrix A) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="mdivide_right_tri_low" href="#mdivide_right_tri_low" class="anchored unlink"><strong>mdivide_right_tri_low</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-36a5e85b63e1d37ab7beac6c411e72d20254fcad"><code>(matrix B, matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-44c350e5b90595515ab70568c03a3283d78a20c7"><code>(row_vector b, matrix A) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="mean" href="#mean" class="anchored unlink"><strong>mean</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-bb1b7fccc6b17cde8f9e75a947464b97bc4082d3"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fedafc3d15883df02af3a5cc3c738594e4aac9e9"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-793945619c88050a3af516f403737ac131fdc6a3"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-895445a6a0170ef1bb8d7d7d9ed5eacdb328b24e"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="min" href="#min" class="anchored unlink"><strong>min</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-6f45b7ae2a16ec959cb22d041d7b50c43dada819"><code>(array[] int x) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-c4f3e810f70c09849013a04b92b38f9e93067e2f"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-82a2d72a544284ee41b99d4842427ee71f7025b9"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f8c78579caf9c583f2f55067b99104c0e7132124"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9aecdccfa0bdd71b2c969de650b8e72bad49df26"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1930adf5f047ff599b49aa2b5dd87c8fabe47d92"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="modified_bessel_first_kind" href="#modified_bessel_first_kind" class="anchored unlink"><strong>modified_bessel_first_kind</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1b49c405644022c418024b3fe2f764d081114dc5"><code>(int v, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-63276374ee38edd44205674f5f32b4dcf306f671"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="modified_bessel_second_kind" href="#modified_bessel_second_kind" class="anchored unlink"><strong>modified_bessel_second_kind</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-e174dc75cd0e499de88270d4616de1ca6cb81235"><code>(int v, real z) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-6340b47f5e4a5642b6c080962ca50fde4d6ea5d9"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_gp" href="#multi_gp" class="anchored unlink"><strong>multi_gp</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-457f8d9e9ebaf6012b892a8241fe77b322473f49">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_gp_cholesky" href="#multi_gp_cholesky" class="anchored unlink"><strong>multi_gp_cholesky</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-5a77e4209bc50ea87ef1dbcce3bad7b10b094a6b">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_gp_cholesky_lpdf" href="#multi_gp_cholesky_lpdf" class="anchored unlink"><strong>multi_gp_cholesky_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-2903bb116c036e35d3cb81e008dcad2b7a6e6756"><code>(matrix y | matrix L, vector w) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_gp_cholesky_lupdf" href="#multi_gp_cholesky_lupdf" class="anchored unlink"><strong>multi_gp_cholesky_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-df18f8bd56610fdd6df3192990ee5cb001142817"><code>(matrix y | matrix L, vector w) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_gp_lpdf" href="#multi_gp_lpdf" class="anchored unlink"><strong>multi_gp_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-fdd1f01264b88aa3f7a0cae0db463d1f87d5e0e8"><code>(matrix y | matrix Sigma, vector w) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_gp_lupdf" href="#multi_gp_lupdf" class="anchored unlink"><strong>multi_gp_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-ead480cd44ceccdf956822e08476ca3b5ca14a76"><code>(matrix y | matrix Sigma, vector w) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal" href="#multi_normal" class="anchored unlink"><strong>multi_normal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-55bcd96ac570a9cd0612aa8c9853df2ef3ca7750">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_cholesky" href="#multi_normal_cholesky" class="anchored unlink"><strong>multi_normal_cholesky</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-478582106e708a75c36c3c557d30196407f15133">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_cholesky_lpdf" href="#multi_normal_cholesky_lpdf" class="anchored unlink"><strong>multi_normal_cholesky_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-d5709f43c49fc57a5ef58e94e176962d2bff379a"><code>(row_vectors y | row_vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-69e5b64a11f77c97cd0fffeaee869c5358969237"><code>(row_vectors y | vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-efdc316743c9c9f1a6ee782650494ee45cb045cd"><code>(vectors y | row_vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-b197f45c690e4b7dd4ebe29d9f75790be4acf177"><code>(vectors y | vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_cholesky_lupdf" href="#multi_normal_cholesky_lupdf" class="anchored unlink"><strong>multi_normal_cholesky_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-1606261a4f25b4e68b2bf4b1c8bf2513e62cdf5c"><code>(row_vectors y | row_vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-d3682df812ba6f35ad608d62b19f6c8ca23601f7"><code>(row_vectors y | vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-a04f4fcdecc170b474051229edff4d2358b9c619"><code>(vectors y | row_vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-2ac54d7aebe6a22693ee5a9016fef469246a746c"><code>(vectors y | vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_cholesky_rng" href="#multi_normal_cholesky_rng" class="anchored unlink"><strong>multi_normal_cholesky_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-83a370eb75f07ac047e03e4dde9d64cf42e4c432"><code>(row_vector mu, matrix L) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-e6934af3e7cb8c68c3aa044307129bbaf17ffabf"><code>(row_vectors mu, matrix L) : vectors</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-b2936af169906302b9686d4673d8c5a8da710efb"><code>(vector mu, matrix L) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-7e60f3380a18c3f7bbcc383bfa7fcd15c14c054b"><code>(vectors mu, matrix L) : vectors</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_lpdf" href="#multi_normal_lpdf" class="anchored unlink"><strong>multi_normal_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-7d157b823f0ae33310094712caff4096b644a282"><code>(row_vectors y | row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-fe635d231edc36e95bda6936e3dc7f58cf04d7b9"><code>(row_vectors y | vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-c1e8d116401a5e7f66812da95e1285dc2a3b111a"><code>(vectors y | row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-ea7d4337cebac7fdbfa928b24a3626208823e80e"><code>(vectors y | vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_lupdf" href="#multi_normal_lupdf" class="anchored unlink"><strong>multi_normal_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-86c5dd8e6cf2c484c3c76fffcadc99f786b7265b"><code>(row_vectors y | row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-69f99f5b6cefce126adc27f7b20728705d7e8a53"><code>(row_vectors y | vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-a0fb7f94e06645589741da551fba4a55a0a33e44"><code>(vectors y | row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-7786848b839eeda0bfa728760464ac4a8369550d"><code>(vectors y | vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_prec" href="#multi_normal_prec" class="anchored unlink"><strong>multi_normal_prec</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-541ca6516f2a9c57d99c4d0b57a2b0a1df8a105d">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_prec_lpdf" href="#multi_normal_prec_lpdf" class="anchored unlink"><strong>multi_normal_prec_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-eb92ca14c43c32d999ac0b10e2160f641a23f265"><code>(row_vectors y | row_vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-8efc2d6f42f7fd5eeb961e3959b617201d45a58c"><code>(row_vectors y | vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-dcfb9bff5f659aa499401ffd77ffd7bbaa784936"><code>(vectors y | row_vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-6a00a4a10d9e84eaa51e885395294f4b4e1272dd"><code>(vectors y | vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_prec_lupdf" href="#multi_normal_prec_lupdf" class="anchored unlink"><strong>multi_normal_prec_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-9cf50bfea31a9655df9c07627982295a3a884afa"><code>(row_vectors y | row_vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-314ed87058998a1934a7cd80a00abca432768d68"><code>(row_vectors y | vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-ec0990b5c058b76b441e3132b8e4986f8fc3b9a6"><code>(vectors y | row_vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-e4db099421dc8d5cb5ccaf22226497142e4ad656"><code>(vectors y | vectors mu, matrix Omega) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_normal_rng" href="#multi_normal_rng" class="anchored unlink"><strong>multi_normal_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-70771ae7e3296e2a09378f81d64a16d0ffdebe49"><code>(row_vector mu, matrix Sigma) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-214830eeea24ebd3c375e92411ceab1101a15b6c"><code>(row_vectors mu, matrix Sigma) : vectors</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-d83a51f846b5823f2e419d6189b27cd98643a584"><code>(vector mu, matrix Sigma) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-ba20f2e8644cf2fff8761565c61ae10f1160df3a"><code>(vectors mu, matrix Sigma) : vectors</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_cholesky_t_rng" href="#multi_student_cholesky_t_rng" class="anchored unlink"><strong>multi_student_cholesky_t_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-37c54942aa5cd7ed73303ca868eeb65b156be282"><code>(real nu, vector mu, matrix L) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t" href="#multi_student_t" class="anchored unlink"><strong>multi_student_t</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-6f17c5611de64710688848607c0edd592e244fe1">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_cholesky" href="#multi_student_t_cholesky" class="anchored unlink"><strong>multi_student_t_cholesky</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-190447ef796a8d04715897b349d1433608a8a0e3">distribution statement</a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_cholesky_lpdf" href="#multi_student_t_cholesky_lpdf" class="anchored unlink"><strong>multi_student_t_cholesky_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-79368cb327eb8f4f3056a3e71077b92df2406595"><code>(vectors y | real nu, vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_cholesky_lupdf" href="#multi_student_t_cholesky_lupdf" class="anchored unlink"><strong>multi_student_t_cholesky_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-4b504cf934bd0448a4e3174a7a7b34f46cfa7d8c"><code>(vectors y | real nu, vectors mu, matrix L) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_cholesky_rng" href="#multi_student_t_cholesky_rng" class="anchored unlink"><strong>multi_student_t_cholesky_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-48a1c071b3760543bf2019dc6c37df56b26a25b4"><code>(real nu, array[] row_vector mu, matrix L) : array[] vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-da90fd4020ba98291c4e0c6ae45340c32e83bdc6"><code>(real nu, array[] vector mu, matrix L) : array[] vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_lpdf" href="#multi_student_t_lpdf" class="anchored unlink"><strong>multi_student_t_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-5b6e5c7b1a5d8704f96553d2ccd4fb053bdb4917"><code>(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-b0e0974e59e75e19f6b6c4aa22d8395d7ef922de"><code>(row_vectors y | real nu, vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-97bab7be0967824f68beadb09ab9fc5152e4b605"><code>(vectors y | real nu, row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-69dbd18bfe0fa38bad7531c187ffd5ddbc68655e"><code>(vectors y | real nu, vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_lupdf" href="#multi_student_t_lupdf" class="anchored unlink"><strong>multi_student_t_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-d92dadb3604f6521420ca422b762d25b60b94dbf"><code>(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-7b3d937ea2e301a8d2455c2939ef7fc5c815d5ca"><code>(row_vectors y | real nu, vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-74a88cd182b98cb2259da79ec60775f5d1f01084"><code>(vectors y | real nu, row_vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-ed2f0cb9d8bc8294e4c924451d7f556f82d5157c"><code>(vectors y | real nu, vectors mu, matrix Sigma) : real</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multi_student_t_rng" href="#multi_student_t_rng" class="anchored unlink"><strong>multi_student_t_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-18f756f159784f68dc00bac1706c5e31f805daf3"><code>(real nu, row_vector mu, matrix Sigma) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-f05b8a3480e2360d1224acb270aca05c90674ba1"><code>(real nu, row_vectors mu, matrix Sigma) : vectors</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-e2a99afdcb2cc3ec5ffc0367d7d241f2b511a575"><code>(real nu, vector mu, matrix Sigma) : vector</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/distributions_over_unbounded_vectors.html#index-entry-d65fa98b0f3f647ffd2cbe3f445ab17be93735dc"><code>(real nu, vectors mu, matrix Sigma) : vectors</code></a> <span class="detail">(distributions_over_unbounded_vectors.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial" href="#multinomial" class="anchored unlink"><strong>multinomial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-e60b9005f8bb7c315f7c9c28123c0e6cf5a64f66">distribution statement</a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_logit" href="#multinomial_logit" class="anchored unlink"><strong>multinomial_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-00ac83839e3f3aac2fa9ff06c7cfb47d659ef796">distribution statement</a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_logit_lpmf" href="#multinomial_logit_lpmf" class="anchored unlink"><strong>multinomial_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-dac1445017dc2f0105939d8fd00f7654057c9a95"><code>(array[] int y | vector gamma) : real</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_logit_lupmf" href="#multinomial_logit_lupmf" class="anchored unlink"><strong>multinomial_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-b2a96795c168ea1f40d7be5db3359edbf96cc95f"><code>(array[] int y | vector gamma) : real</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_logit_rng" href="#multinomial_logit_rng" class="anchored unlink"><strong>multinomial_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-1d2c47ee6f88d4e1c284b4647634a23d3b215792"><code>(vector gamma, int N) : array[] int</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_lpmf" href="#multinomial_lpmf" class="anchored unlink"><strong>multinomial_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-2a1a28ece722b9157657d45c59162d74bb3757df"><code>(array[] int y | vector theta) : real</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_lupmf" href="#multinomial_lupmf" class="anchored unlink"><strong>multinomial_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-43f3a751ea3d648a3e23143c525562a1f20e99a3"><code>(array[] int y | vector theta) : real</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multinomial_rng" href="#multinomial_rng" class="anchored unlink"><strong>multinomial_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/multivariate_discrete_distributions.html#index-entry-be7dbffff355abebf7628e28e92ff3e0fc11e115"><code>(vector theta, int N) : array[] int</code></a> <span class="detail">(multivariate_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="multiply_lower_tri_self_transpose" href="#multiply_lower_tri_self_transpose" class="anchored unlink"><strong>multiply_lower_tri_self_transpose</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-074a26dceab0e18f9117c1bd04f0818b1c303de2"><code>(matrix x) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="n" class="level2">
+<h2 class="anchored" data-anchor-id="n">N</h2>
+<p><a id="neg_binomial" href="#neg_binomial" class="anchored unlink"><strong>neg_binomial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-62400243069f91416bfa36c834b3635bebcbd43e">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2" href="#neg_binomial_2" class="anchored unlink"><strong>neg_binomial_2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-ea4ba68ab4a32b0e2125adc6e5304e54972de2a9">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_cdf" href="#neg_binomial_2_cdf" class="anchored unlink"><strong>neg_binomial_2_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-a5768bbcd476f90cbae7573d0978531f6ad5a9fa"><code>(ints n | reals mu, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_lccdf" href="#neg_binomial_2_lccdf" class="anchored unlink"><strong>neg_binomial_2_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-bfa4fa4a4278e2a259bbd9bae0faca3d4e99b1eb"><code>(ints n | reals mu, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_lcdf" href="#neg_binomial_2_lcdf" class="anchored unlink"><strong>neg_binomial_2_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-7bb63856ad5d268ddb11337ad2e4e53f7e54a273"><code>(ints n | reals mu, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log" href="#neg_binomial_2_log" class="anchored unlink"><strong>neg_binomial_2_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-92d907cea1dc522c9b36cf1caa3f7c6e8360fe7b">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log_glm" href="#neg_binomial_2_log_glm" class="anchored unlink"><strong>neg_binomial_2_log_glm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-28895026932258af73faa16c7dec6a131af227bb">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log_glm_lpmf" href="#neg_binomial_2_log_glm_lpmf" class="anchored unlink"><strong>neg_binomial_2_log_glm_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-145c118038ba258504502a0ca981d7c12532bc51"><code>(array[] int y | matrix x, real alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-f5314e218cc2c6e5fe4e6301645ec2c0fdafbc11"><code>(array[] int y | matrix x, vector alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-16a3cf3cfe06586324838df97a33fbb2c2efeef7"><code>(array[] int y | row_vector x, real alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-30a171234617b31ebd3414cb4a4f82c1b62b3d34"><code>(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-7fe4a0797bb0b9fced5c0f2d75c06e1e479ea248"><code>(int y | matrix x, real alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-91d1747bb7a9e5a3edc9ed5ec4c377f643eac2ca"><code>(int y | matrix x, vector alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log_glm_lupmf" href="#neg_binomial_2_log_glm_lupmf" class="anchored unlink"><strong>neg_binomial_2_log_glm_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-c5e5488af5e4e66825ea2b163862a7d35f6dd124"><code>(array[] int y | matrix x, real alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-7346fdbc1e0f782c1d3ab15e9f444e03bc33b644"><code>(array[] int y | matrix x, vector alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-c6a6b66b9c19ab430c255ed8047ad014e1f6414c"><code>(array[] int y | row_vector x, real alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-e736537c82f56050b3aaf75e43b6d50a38477dd9"><code>(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-09ef2622134a433c0f9c3ad0643e20255979e98a"><code>(int y | matrix x, real alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-d01dd7632fe36e888d0067b62cae9fd79230af8d"><code>(int y | matrix x, vector alpha, vector beta, real phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log_lpmf" href="#neg_binomial_2_log_lpmf" class="anchored unlink"><strong>neg_binomial_2_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-e583555ff09b9ca6a30c94843f0e429145fe29ad"><code>(ints n | reals eta, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log_lupmf" href="#neg_binomial_2_log_lupmf" class="anchored unlink"><strong>neg_binomial_2_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-79456eb164dc3e83dad8b862ae24a86882f2d536"><code>(ints n | reals eta, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_log_rng" href="#neg_binomial_2_log_rng" class="anchored unlink"><strong>neg_binomial_2_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-81dfe168f8aedf6dbf3f3ff8025936b2cf1052a1"><code>(reals eta, reals phi) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_lpmf" href="#neg_binomial_2_lpmf" class="anchored unlink"><strong>neg_binomial_2_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-711f077d61fb2160d3c8b33015b15af366056763"><code>(ints n | reals mu, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_lupmf" href="#neg_binomial_2_lupmf" class="anchored unlink"><strong>neg_binomial_2_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-3232cdce0c2d7dce167ba0946aafbbbe5cc4a28f"><code>(ints n | reals mu, reals phi) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_2_rng" href="#neg_binomial_2_rng" class="anchored unlink"><strong>neg_binomial_2_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-eabf098a3499209e5f0959ac998ed4fcf53b0a51"><code>(reals mu, reals phi) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_cdf" href="#neg_binomial_cdf" class="anchored unlink"><strong>neg_binomial_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-de9079d9e962df30ca27f6860eae99efcc654189"><code>(ints n | reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_lccdf" href="#neg_binomial_lccdf" class="anchored unlink"><strong>neg_binomial_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-8c72c425ce2f6cbc3dad23f6abb0dd9278749e84"><code>(ints n | reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_lcdf" href="#neg_binomial_lcdf" class="anchored unlink"><strong>neg_binomial_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-83ea716052404e4f13e9f1c9f52092361d064615"><code>(ints n | reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_lpmf" href="#neg_binomial_lpmf" class="anchored unlink"><strong>neg_binomial_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-2a11a12aef7b6b441ea425acd9d5cb21aa6e66c0"><code>(ints n | reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_lupmf" href="#neg_binomial_lupmf" class="anchored unlink"><strong>neg_binomial_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-93c924df2baea99a71249bcad641d8f75c6e41cd"><code>(ints n | reals alpha, reals beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="neg_binomial_rng" href="#neg_binomial_rng" class="anchored unlink"><strong>neg_binomial_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-4c2098ee820ccf0732443eb5d87c0d6d839cce4d"><code>(reals alpha, reals beta) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="negative_infinity" href="#negative_infinity" class="anchored unlink"><strong>negative_infinity</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4fcafd8237c585340253854dec13739165d456bd"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="norm" href="#norm" class="anchored unlink"><strong>norm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-8e1a8cccaab89f6f630e0b34e2292689c9aa46a3"><code>(complex z) : real</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="norm1" href="#norm1" class="anchored unlink"><strong>norm1</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-d29d5686fa464cc82765e365c73f1ebb70954ffa"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-3a384b9bbb192b688afb1c9e95a490200da65412"><code>(row_vector x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-64a22469a1ca08b874dfd1f7dba4e60a02311a4d"><code>(vector x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="norm2" href="#norm2" class="anchored unlink"><strong>norm2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-95748dc7d816217c82a900a5cefd801980794a10"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-15be5905b3ac40ce77299341b9e7434d6e5a721e"><code>(row_vector x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-bd1aec552c96bca3230c994fb47fa42e54632e5e"><code>(vector x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="normal" href="#normal" class="anchored unlink"><strong>normal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-72458a5e0e1a905dc04f11d1b05e4136543f6142">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_cdf" href="#normal_cdf" class="anchored unlink"><strong>normal_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-03470cc37c6a6c27433e9854162ed863ac635e68"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_id_glm" href="#normal_id_glm" class="anchored unlink"><strong>normal_id_glm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f2e910491c0d85196d2d372dcaffcc4cc22572ce">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_id_glm_lpdf" href="#normal_id_glm_lpdf" class="anchored unlink"><strong>normal_id_glm_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4436deedaa4adcaf697328f4ab6228f12bf213be"><code>(real y | matrix x, real alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-90d8f9b890b2588b36ed285cfb031e89bf170935"><code>(real y | matrix x, real alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4fc2715c69e219915d6372ded62c513ec2dad353"><code>(real y | matrix x, vector alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-7991a0ad137f72771bee6ae8bfe2a09dba063c5a"><code>(real y | matrix x, vector alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-5bce6923d1e27b6dfdb1112d0e26e6f2fccf3704"><code>(vector y | matrix x, real alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-1428e7cc017ba5e7e160484c9051cfb458221d42"><code>(vector y | matrix x, real alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-c8b35bcd76ae895df16171156e9ddc2b50977858"><code>(vector y | matrix x, vector alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-bc577f1feffdbfa17e7321f5cd733afccd7e822f"><code>(vector y | matrix x, vector alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-1d3d0f48ac6b8ae87eab8640482747cc1d3bcc65"><code>(vector y | row_vector x, real alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-d67cdb934c8fdbe441e7968e31d3d42bdc6bb46f"><code>(vector y | row_vector x, vector alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_id_glm_lupdf" href="#normal_id_glm_lupdf" class="anchored unlink"><strong>normal_id_glm_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-ef0f44ddfb9e65d97bec95df9bc84e88a5262ebc"><code>(real y | matrix x, real alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-2604f4a479cd2758a8273ebc888a4f35d707bcc3"><code>(real y | matrix x, real alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-baaf096f4f58b53d3142ade26076839f36733a39"><code>(real y | matrix x, vector alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-d7b73dfd16de89ad209b2a04905cdae1169f761c"><code>(real y | matrix x, vector alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-b410b1737bb33ec4400eae94d05bb37faea0c0e9"><code>(vector y | matrix x, real alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-e4b62bbce38476467b6d5d8347d048688b1b22fe"><code>(vector y | matrix x, real alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-348b5a5c6bc0a9340114bd7744603300d84580bb"><code>(vector y | matrix x, vector alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-6c41e1877f2f156d78215c8b4f25b5795223a0d3"><code>(vector y | matrix x, vector alpha, vector beta, vector sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-1aefc00a65f6adeb0cba0329b66f968d6caf1427"><code>(vector y | row_vector x, real alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-29d66019b47c2be1df4e8cb54f85a3459f41d7be"><code>(vector y | row_vector x, vector alpha, vector beta, real sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_lccdf" href="#normal_lccdf" class="anchored unlink"><strong>normal_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-0e0917cae8b192a1adca7f9d708d48f291d953c0"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_lcdf" href="#normal_lcdf" class="anchored unlink"><strong>normal_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-0c8b83888fa71a0e469f8d91bc4d13c2ee3adb7a"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_lpdf" href="#normal_lpdf" class="anchored unlink"><strong>normal_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-7fb57a9b998de709bd4e5f0f69713d26e2ee62ae"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_lupdf" href="#normal_lupdf" class="anchored unlink"><strong>normal_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-429c956476257b277125a0f2bddd560aaa7416a6"><code>(reals y | reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="normal_rng" href="#normal_rng" class="anchored unlink"><strong>normal_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-a2ef357923cb90482b3d3a4e7a66b0a04b56f3fe"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="not_a_number" href="#not_a_number" class="anchored unlink"><strong>not_a_number</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-00757653dd74e4f55dcd63fbdfb21a9e5005469c"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="num_elements" href="#num_elements" class="anchored unlink"><strong>num_elements</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a7f3950f11e27f31d9607b74a4cc9d5201f6f5ba"><code>(array[] T x) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ea9bb96c32abb986ac353999ee38effe98c2999e"><code>(complex_matrix x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-18b5a5e8da0b97e5db61bc2558be5861bb0e60c9"><code>(complex_row_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5f67a5c5dc6caa692838121eb1be3a0294ad191f"><code>(complex_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-77dd48e13c18141a06bb49ceb67e01b76c4e3fae"><code>(matrix x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-25d1da3aed0f07d9c6a36b34c0d238d180ebc272"><code>(row_vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9173ff057d7657581e5abf426a39637cb2339c11"><code>(vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="o" class="level2">
+<h2 class="anchored" data-anchor-id="o">O</h2>
+<p><a id="ode_adams" href="#ode_adams" class="anchored unlink"><strong>ode_adams</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-0cbdc3972873453988201c42896556c64240fd04"><code>(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_adams_tol" href="#ode_adams_tol" class="anchored unlink"><strong>ode_adams_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-6272f555abc93dcbbf8700fbeafca26f4b6f686e"><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_adjoint_tol_ctl" href="#ode_adjoint_tol_ctl" class="anchored unlink"><strong>ode_adjoint_tol_ctl</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-95a49804904581f863b2d682d1d17408cc0371d8"><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_bdf" href="#ode_bdf" class="anchored unlink"><strong>ode_bdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-302230625fcf25f300cf21fa4b0f34e9494d3a53"><code>(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_bdf_tol" href="#ode_bdf_tol" class="anchored unlink"><strong>ode_bdf_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-a1ea8ce1da0b3666b341644d924730abb2caf7af"><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_ckrk" href="#ode_ckrk" class="anchored unlink"><strong>ode_ckrk</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-02a83fd8c68af0422903498868b0e1c9c2542659"><code>(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_ckrk_tol" href="#ode_ckrk_tol" class="anchored unlink"><strong>ode_ckrk_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-25e82b14c8715c7f6c5fd13961ffa60016433854"><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_rk45" href="#ode_rk45" class="anchored unlink"><strong>ode_rk45</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-4a595847f7a33bc641756482fe60187677e31fa1"><code>(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ode_rk45_tol" href="#ode_rk45_tol" class="anchored unlink"><strong>ode_rk45_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-1fd0eab9e49603406a50ad94820da2fb10dec2bc"><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="offset_multiplier_constrain" href="#offset_multiplier_constrain" class="anchored unlink"><strong>offset_multiplier_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-c55eb604debca345fb51ad9d71cb36b6f20bb60e"><code>(reals y, reals offset, reals mult) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="offset_multiplier_jacobian" href="#offset_multiplier_jacobian" class="anchored unlink"><strong>offset_multiplier_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-2448386d5029a2ea68a25d2eac851459f98eb968"><code>(reals y, reals offset, reals mult) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="offset_multiplier_unconstrain" href="#offset_multiplier_unconstrain" class="anchored unlink"><strong>offset_multiplier_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-6ed536361121ffb486071705925b0107a779e401"><code>(reals x, reals offset, reals mult) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="one_hot_array" href="#one_hot_array" class="anchored unlink"><strong>one_hot_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-268289da1aa3a477c94f93562ea8dec0d09ca486"><code>(int n, int k) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="one_hot_int_array" href="#one_hot_int_array" class="anchored unlink"><strong>one_hot_int_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0beac7fcfa02cd2f93d1869abaf3555fa0711440"><code>(int n, int k) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="one_hot_row_vector" href="#one_hot_row_vector" class="anchored unlink"><strong>one_hot_row_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0a0220657a7b045ad8375607c32be8215c237455"><code>(int n, int k) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="one_hot_vector" href="#one_hot_vector" class="anchored unlink"><strong>one_hot_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ad02bb0cabeceb6ad40303c9ba844e75c5fab300"><code>(int K, int k) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="ones_array" href="#ones_array" class="anchored unlink"><strong>ones_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0fafc0a373e0e8f0bbaeee5c35eba8570f9149d9"><code>(int n) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="ones_int_array" href="#ones_int_array" class="anchored unlink"><strong>ones_int_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9105f9deb3c88299539acb48f505c2ecb81107f4"><code>(int n) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="ones_row_vector" href="#ones_row_vector" class="anchored unlink"><strong>ones_row_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a8497a47aca811a739e649d1ce295d1629665664"><code>(int n) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="ones_vector" href="#ones_vector" class="anchored unlink"><strong>ones_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f291f3a346f24dfea23941e9dc9e908a2dfbe31d"><code>(int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator!" href="#operator!" class="anchored unlink"><strong>operator!</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-079682d0e7e0678530187f5fc5e2b993f1b5b1cc"><code>(int x) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-211d31dd129d2ce61f68b4264b9318aa39a096d7"><code>(real x) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator!=" href="#operator!=" class="anchored unlink"><strong>operator!=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-9d820ba333127706253b1578dc13615c5898dfe4"><code>(complex x, complex y) : int</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-34c680d818a4a8758b819f7b8a3b11f1d1f094b4"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-3cc230ef52c7125f7fd034984257fb3b2e13bbcc"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator%" href="#operator%" class="anchored unlink"><strong>operator%</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-1a37a12af3bb33938a4e3fab2375e8f57617354e"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator%/%" href="#operator%/%" class="anchored unlink"><strong>operator%/%</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-5c1d4a2c0b351939f98b1c7a20a25f07b08366ba"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator&amp;&amp;" href="#operator&amp;&amp;" class="anchored unlink"><strong>operator&amp;&amp;</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-705fd1857d26667bd15e9cbd54c1a605ce286b17"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7fab4d7139fe3c56c90269284ac8fb2a27aa9c3b"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p>&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;<strong>operator’</strong>:</p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-091a4926b279f9df738ff9dcd356e2436c3f9485"><code>(complex_matrix x) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ef45f4afc752618674b3095733d0dad1150a39fd"><code>(complex_row_vector x) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-7b294083ea64b206b4b02446640de44855355156"><code>(complex_vector x) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-794bf86d8a081a75147a013481f07740310bb0d8"><code>(matrix x) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-e3432686a771bad98175ab4a594d8156ab915492"><code>(row_vector x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5b05ea20fccc5c44353c7f5813ec3ce694b07cb6"><code>(vector x) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator\*" href="#operator\*" class="anchored unlink"><strong>operator*</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-e3a6425a4fd372a6a28b66eb58c06245046f87ae"><code>(complex x, complex y) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a69bac506948c37d27beb39e793ff1339a67bf85"><code>(complex x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-96bd6d9926f3cf86d4abdf08767a6078cd45be14"><code>(complex x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a79cd86a9e357f4daa3c2a357da7fe74a1591449"><code>(complex x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ec01c1a12f2dee572064680e87666ff43581e98e"><code>(complex_matrix x, complex y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e6c145519223a7e32befa818a711dd8470e7264d"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-7e59bb7495322cfae3a60e7cd858c8ea8ad00be4"><code>(complex_matrix x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-989f9db43310c8b2278983062fe9699e275f0f4b"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-eb9d524d09ed2def387f6d32e50434f1d04bb45a"><code>(complex_row_vector x, complex_matrix y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-6a24a75ba519ef262ca78b4f7e1077124ff4e67b"><code>(complex_row_vector x, complex_vector y) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-23a93f16e6035cc32b84ecf97c32845e9008702c"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-6036495abafc183b219e271873881cec3f6e46a3"><code>(complex_vector x, complex_row_vector y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-74f2ac23b7142dcb99d72c2d668ea84d112f43ef"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a87089afcc4f8cc865a9447f93284dd4a7903eaf"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7c526c1ddbd012a188345676d03d7d8e52a2f4af"><code>(matrix x, real y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c61d79dac022af07a0f553fb494511099410cdac"><code>(matrix x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-662e0b2447d3b372fc219293e38410d49bad1b7b"><code>(real x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-08f55cee893fdef254457be66b7a82c1d942ad0e"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-84a0cd07fdb794ec75d40dfa3c1638586ce59f1a"><code>(real x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-af34e073da12073d94a55fde49701a8f65786b67"><code>(real x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0d741e9840edb32e9575051107190f0632ec5dd3"><code>(row_vector x, matrix y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-852b4c73c95c85062105d4a058743e342d831c9a"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c64d64dd72ff726168c121c89601f1e1fa602276"><code>(row_vector x, vector y) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-29c74f1694c2364957ad8d79e09cfb3ad47e775b"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3669aca452e27a8b48f0358b3d3103bc4d0d9530"><code>(vector x, row_vector y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator\*=" href="#operator\*=" class="anchored unlink"><strong>operator*=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-0c45a357fc47d9ccf6a7fb7780ee7d1ada6cc2bd"><code>(complex x, complex y) : void</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/compound_arithmetic_and_assignment.html#index-entry-518f24c806ec47af7c7aa48506a1cd57ef3676df"><code>(T x, U y) : void</code></a> <span class="detail">(compound_arithmetic_and_assignment.html)</span>
+</div></li>
+</ul>
+<p><a id="operator+" href="#operator+" class="anchored unlink"><strong>operator+</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-b67896f1b7c21199de738af1835d35fe36635012"><code>(complex x, complex y) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-3cbf469cad59cf318ccf139f90a016fc4bfe3db1"><code>(complex x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-cf423b623dd5296c4bed0996e11793cfff99b12b"><code>(complex x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5a80c22fc9ab785c199e3ea9ba3903429f3308f9"><code>(complex x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-5fcb0e5cdd3aa28c383cc6808af8d0523f581e11"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-0b1ee3eff7624a28d75d9cbdf065d8cb936c401a"><code>(complex_matrix x, complex y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-1c6e89ffde910406a3f594d11872988647733d8f"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-83570c79a3f99ef98736dec5190a738977bf1986"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e4a7e3cd9cefa9964eb063b241f963f7b27007e7"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e2a4a5c6172fcff3b822f22f17e4012616aa798b"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c4da2f5e129a64de4d83c12e2a8d2d79d7cd1619"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-1b4db535db6b20e4b8af55f41af6ee4d6b5df6bd"><code>(int x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-2e81f507ecb4144548330dac4d8bbc43686cd3e2"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-718549a6ddfc2c775f69b8bc7a28fc7c053a7fbe"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-77931799ac0a444f9530777cce44cb987abf32cb"><code>(matrix x, real y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-54245a5adfc60f65ced33cbdf4571aeb7536afc3"><code>(real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0465a61949555ef4ef44514ccba8174500deeefa"><code>(real x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4b2af0abc1a2c8338794b49b801ff4d63be2f574"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fd064b77a97b1ec9168687d5a25cde11da015591"><code>(real x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f9354975ff6bc2974a4d1661855cea524ca49b50"><code>(real x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-26044cd5506e5c4fc2579fc6466dd3f81d6408ce"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-33e5719f2f10f7ff91219c35213757ec24c6b306"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-925b4d95d12aaad5ae3a3d5a5b8d06d9f36c3b9b"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1a0c872aa1e302e3fd5a2a167b97e3ca6719d10c"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator+=" href="#operator+=" class="anchored unlink"><strong>operator+=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-cc6740802b1a8a2be0ba358ef3799d66e260a341"><code>(complex x, complex y) : void</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/compound_arithmetic_and_assignment.html#index-entry-967b2a8e9bdbd6f8a1fc5a6deede7134f075a813"><code>(T x, U y) : void</code></a> <span class="detail">(compound_arithmetic_and_assignment.html)</span>
+</div></li>
+</ul>
+<p><a id="operator-" href="#operator-" class="anchored unlink"><strong>operator-</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-7aaca5ac69e98f6d2bc4db949541e8c69d18e694"><code>(complex x, complex y) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a49311f2dd1b5d25aa68b099771fc781e15783c3"><code>(complex x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-be634502437e0887fcb531772138f252af5cd41b"><code>(complex x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-149f85a65fcb71890f2740b37f3d8e99d171c293"><code>(complex x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-34804cfa841db4ee3b8175eb5bbb0a124b639049"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c67e9b27b61ec87f6671f881e99e0215abca731a"><code>(complex_matrix x) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-19c0fa6f4a31be826fff56b079a14d03456e94c6"><code>(complex_matrix x, complex y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-4b7f3f2a15c41213894ee287a6fd4532d8ca59fb"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-4849f1558fb4d3addc5296dec7dc076fc33e2ac8"><code>(complex_row_vector x) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-23a25b349111e68acc8ccbf019f247a0857b9f9a"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c159327226df72cad0b808e6e72271542904ae9b"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e5e977ad448b9356d9822a744773c212eb665b80"><code>(complex_vector x) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-750993752e1971b5caa4b40e151244295537237d"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-2b25bbe909c531da7ca053dec04d2a0d66767407"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-ef76e937371aef8a99cfd4a9964fd54af23d7344"><code>(int x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-63c9680853b8c13d6503afb15b5628911298a42a"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6ca73a0303dd24f82990cf099176d76ea4f50d3b"><code>(matrix x) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b9ba1d02d3a73b7e8733514fb6b3a1c59414ee37"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-88db741d085202e8824acf87fdf5652fbc176914"><code>(matrix x, real y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-2055033f59d0333f7b514dd496153b272c1b274c"><code>(real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-aa6b0f0e0b80d8bcc4d95c623445e021142a9439"><code>(real x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-739fd04c545b6d65855bc20e05617df813db0598"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6002a07176762b3c481c6761e8e06b1294e8d625"><code>(real x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-76b3ba43bb249ac182c9f4fea095ea81ede76ebd"><code>(real x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7fa245ebd36b8195ee47b5cf42158c3936625d93"><code>(row_vector x) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0e7a932586415d9249383de3efe664eb105f4a4f"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f9f5c0ef9c18e80c24aa789a90b7d1253096fa3a"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f"><code>(T x) : T</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f"><code>(T x) : T</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f"><code>(T x) : T</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f"><code>(T x) : T</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f"><code>(T x) : T</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8339baafb56b84ffc02ddc234752ab1f7a46e1e3"><code>(vector x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0dea3575356afb28b43da9a9985361d5ba7e1c84"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8b1168c1cebe948416436c1c61cb7d7210949fae"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator-=" href="#operator-=" class="anchored unlink"><strong>operator-=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-7bc55fdd3a17228fab1be65f974b778974c9fd12"><code>(complex x, complex y) : void</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/compound_arithmetic_and_assignment.html#index-entry-e3beea1794e1e8704ae6177a2392b78fccebd4d4"><code>(T x, U y) : void</code></a> <span class="detail">(compound_arithmetic_and_assignment.html)</span>
+</div></li>
+</ul>
+<p><a id="operator.\*" href="#operator.\*" class="anchored unlink"><strong>operator.*</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-8eb7c34b5da648cda9b6f992d4218d023c05d1da"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-8225aa3a59b24f94b5efa4533b49050ef65e64ca"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-955eb57e60395b2be45f5af21989a3aff5f5db70"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6dec717a6b41050c89af60556ec7baae6bb8bb75"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-296aa89afe5256f9cb944c3ce9458197bd833313"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f6b9c926ae8c20a429454fbfd788a05c447ca21f"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator.\*=" href="#operator.\*=" class="anchored unlink"><strong>operator.*=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/compound_arithmetic_and_assignment.html#index-entry-549243a6af6a91c078444182753a8e8f435862e3"><code>(T x, U y) : void</code></a> <span class="detail">(compound_arithmetic_and_assignment.html)</span>
+</div></li>
+</ul>
+<p><a id="operator./" href="#operator./" class="anchored unlink"><strong>operator./</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-55e666acb9e555e6085cb69b1bf90e25b210d14b"><code>(complex x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5380455bf8d01c0d94b7994157ecd464dff285ab"><code>(complex x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-97bb359024d4f5bab6603c1d02519ed1e7641737"><code>(complex x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-f5684bfa80a80f64afb4be43b829a06f2d3201fa"><code>(complex_matrix x, complex y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a68b100764978dc37adb03c4ac341b3369fa83af"><code>(complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-f83946686d65ec544820b3c6401ce2e3000cc1a9"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-d6ea602ecffe4e62bb465b28956200dec419c158"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-17e7fc9ed1291f3c3a332e54b65762dcac73f699"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-0f3d53dc1ff728a9ee7518e0ad5992c1fedcb7ff"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-63e3151ce39b028bb15225d8305e8ae59d1d2b9c"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-33b751b1c6fcaa79a1f249d90774c58df3a3146f"><code>(matrix x, real y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ed3f9fc9cd17f66b01b08112c1a1c0eb063dfbc7"><code>(real x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c8cde8c39dee0b427f9770e8e63ece827cebac04"><code>(real x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1800a4c4e6451dd14c0759f41c3b260e10010627"><code>(real x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-dcdaa0d007eeecd45a7d5ac2cf47b91d52c07fc8"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c9b71dbef4490527107a1dcfcd38f77b374c74d5"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-66caea638b41f99f47c90fbf9f08a165f7bc312b"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-65a45dfeaa613598d736e8e631a8900b78e3c9ff"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator./=" href="#operator./=" class="anchored unlink"><strong>operator./=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/compound_arithmetic_and_assignment.html#index-entry-afa1bfd51ef434b4843fe93b63d0cd3807ac9c7b"><code>(T x, U y) : void</code></a> <span class="detail">(compound_arithmetic_and_assignment.html)</span>
+</div></li>
+</ul>
+<p><a id="operator.^" href="#operator.^" class="anchored unlink"><strong>operator.^</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-1e6130ad7edcc189ada93ab46995b79b280ada4d"><code>( complex_matrix x, complex y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-091231072e09466bffb7f4aad16c86348a965b44"><code>( complex_matrix x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-fd8c739930b5a8f56920738e91cec159c5e50783"><code>(complex x, complex_matrix y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-fa71ce75b9e1f602439d979ae6e9a6161effce6d"><code>(complex x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-af111e3482f1da8bfba4fa5bae01a0084ebebca3"><code>(complex x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-875148aa0d5deecb6b31fecb24608a167d02fa95"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ba84dc5cb8e20d287fb783ebc26ecf24df38f043"><code>(complex_row_vector x, complex_row_vector y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-a7f9b0419a421d43341556c8b8823bb6737fa14c"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-4d24b6cb15adf28b25dc40a155bf4c76980d37d0"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ee33dd887a5da3fafc96e59aa3cd75c6d5051cea"><code>(matrix x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1d5f88db2e08d98185f4bdad36ccdeb289d1a58c"><code>(matrix x, real y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8955da8055e2b2a58a20eac3964a35cdf330b786"><code>(real x, matrix y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ea70cfe8f041b8cdd5f8cfcc99356a42ca783deb"><code>(real x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-396441e397b74d06863454c78d4b5af50dda330f"><code>(real x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-257383daeef20f4a85bd2e23e4f3c9c2ecfb013d"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a4fdb1f768269d25cc5aab5dafbe0535c73ee226"><code>(row_vector x, row_vector y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9f9fe3db8b2a57c2c414d4efe2eb5aa93b6b86e4"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-71f690034947993173f537ecaa9c8ee8a6e8c1e8"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator/" href="#operator/" class="anchored unlink"><strong>operator/</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-2982c4088224c83c22a60e3c585c997d23773024"><code>(complex x, complex y) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-469c065cfa735cf1880ee348a7459ff469411426"><code>(complex_matrix B, complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-0e4a06f38977dd429031b5e8cbd2c34d97c11b6d"><code>(complex_matrix x, complex y) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-f47bdbfb8220a45ba54119c3805497e1d4596752"><code>(complex_row_vector b, complex_matrix A) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-05c2018d6b4d3473de214d967f936575bbe2ce9e"><code>(complex_row_vector x, complex y) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-0042e2fa5c0ff5c4695b13b4a482a8b37f1ab450"><code>(complex_vector x, complex y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-4ae4285a5c320ab57f323608bdfa3c237c8e33e8"><code>(int x, int y) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-aa2855d27dc938ec16cafecff875185e51ba660d"><code>(matrix B, matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-dc385d6a823b14919aaa86e64b873846b31b422b"><code>(matrix x, real y) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b28e30b8930bb68726d2de15d2857875cb73fcfb"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c8f636c70157ca03f436fa1fa6ad66005ee280c7"><code>(row_vector b, matrix A) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6ad75a35e6fb68d75c09604b3945b20cc0dabb1e"><code>(row_vector x, real y) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-31c5a97eb5a6aa036118bd60ae51ce6f8d86c465"><code>(vector x, real y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator/=" href="#operator/=" class="anchored unlink"><strong>operator/=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-f6845827ad2fc15504afc3931ae66a1aa1c8c0d8"><code>(complex x, complex y) : void</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/compound_arithmetic_and_assignment.html#index-entry-eee88771fc7180e557937b705c365f97c600b293"><code>(T x, U y) : void</code></a> <span class="detail">(compound_arithmetic_and_assignment.html)</span>
+</div></li>
+</ul>
+<p><a id="operator<" href="#operator<" class="anchored unlink"><strong>operator&lt;</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-eb4fbef11b05d04b29b1d9b813f5e07945bf9ad8"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-45d9721208dde1bade759d23819a2a80bbde7bb0"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator<=" href="#operator<=" class="anchored unlink"><strong>operator&lt;=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-ae3e3c382d593269517c04ea38e93c34b1938ed3"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b2b6ee942a7836d609c24691ae5327be085eabf8"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator=" href="#operator=" class="anchored unlink"><strong>operator=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-90376a831758ab6b86587380c1139d0f1d298b88"><code>(complex x, complex y) : void</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator==" href="#operator==" class="anchored unlink"><strong>operator==</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-4250f55203843166af55fdbb17eefad6ec3d54ef"><code>(complex x, complex y) : int</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-2ef075eb2521760dca18f52bea00eda6609a70d7"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-876a6664927a13415aef76d12a52894d6047d958"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator>" href="#operator>" class="anchored unlink"><strong>operator&gt;</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-7897823400bd03a18304e92dea188037789e2fc0"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-47bdb3c70b9ec7cdf0f976442ca728e5cd85ea39"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator>=" href="#operator>=" class="anchored unlink"><strong>operator&gt;=</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-9d2656ef1df47087d225d2cfd011938c49677d09"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-29ba70695594992fa689387dd1f623e794aa8ef2"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator\\" href="#operator\\" class="anchored unlink"><strong>operator\</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-1a08f375cfc4067e97bfed98fae8490e8b6f0698"><code>(matrix A, matrix B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fc754907294117fa562e6fcaf1b8afcf0d2efabd"><code>(matrix A, vector b) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="operator^" href="#operator^" class="anchored unlink"><strong>operator^</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-25f7fc457739ce80dced6697c7730e60913ababa"><code>(complex x, complex y) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-329b197f72b50051c6de252c7dd804a96a399051"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="operator||" href="#operator||" class="anchored unlink"><strong>operator||</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8fbd92b4b30a50912cd8d6602927ee93cf00c392"><code>(int x, int y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-d1b0b32e522a6d270bdccaa3b56f1dfaef2dfa1f"><code>(real x, real y) : int</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_constrain" href="#ordered_constrain" class="anchored unlink"><strong>ordered_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-ef08c8144bb3789c27d35f7a7c30cd57f8a9a573"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_jacobian" href="#ordered_jacobian" class="anchored unlink"><strong>ordered_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-6372fc8ee07f534b79aa45d152097a421af0a938"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_logistic" href="#ordered_logistic" class="anchored unlink"><strong>ordered_logistic</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2bffcaec141463bcf3650a49ab910f0e9b02c0b0">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_logistic_glm_lpmf" href="#ordered_logistic_glm_lpmf" class="anchored unlink"><strong>ordered_logistic_glm_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-c3713f9c51b422f753ef46eab1b6000b57a85f83"><code>(array[] int y | matrix x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-13f1725d36ca9cd5b34e0f7533be511bf7a2ec70"><code>(array[] int y | row_vector x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-c11116cef97e5808f28ca2124ff702866ce897e4"><code>(int y | matrix x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2c635de14c16f079f1741deefe7257e6a42810dc"><code>(int y | row_vector x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_logistic_glm_lupmf" href="#ordered_logistic_glm_lupmf" class="anchored unlink"><strong>ordered_logistic_glm_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-55686d3ab941df81acce76e8e1f880716336638b"><code>(array[] int y | matrix x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-1c7b2fb2eee1fbd25742cd97c0396297b00db969"><code>(array[] int y | row_vector x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-757d0591bda8f231fba81cb4faa1d71ba85deb4f"><code>(int y | matrix x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-5c21721436d27b4570a4356fcf87939c6695e2bc"><code>(int y | row_vector x, vector beta, vector c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_logistic_lpmf" href="#ordered_logistic_lpmf" class="anchored unlink"><strong>ordered_logistic_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-8ec1ec8642d86b6ff109ab22859c8bc55de49be0"><code>(ints k | vector eta, vectors c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_logistic_lupmf" href="#ordered_logistic_lupmf" class="anchored unlink"><strong>ordered_logistic_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-56bd6927e7b4b6a5fd660bda74057a8ad1712bb5"><code>(ints k | vector eta, vectors c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_logistic_rng" href="#ordered_logistic_rng" class="anchored unlink"><strong>ordered_logistic_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-27848db7ec1c230a757f0fab71e6bfa2bd6d155e"><code>(real eta, vector c) : int</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_probit" href="#ordered_probit" class="anchored unlink"><strong>ordered_probit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-4a767db733c5f8854e1a70dae8c94ad673adf650">distribution statement</a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_probit_lpmf" href="#ordered_probit_lpmf" class="anchored unlink"><strong>ordered_probit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-bb0439ad56fbb41b1ca9f56fcb28b32383d9f6b8"><code>(ints k | real eta, vectors c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-e9015f46859715b88bc9f7f4bd6b7f1f9ffc51c4"><code>(ints k | vector eta, vectors c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_probit_lupmf" href="#ordered_probit_lupmf" class="anchored unlink"><strong>ordered_probit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-2099a3d4968bfa03c46d810ff23a8b1279db770d"><code>(ints k | real eta, vectors c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-3cd2703fe784d2f29b24fb4f3d897ea2b5264da2"><code>(ints k | vector eta, vectors c) : real</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_probit_rng" href="#ordered_probit_rng" class="anchored unlink"><strong>ordered_probit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_discrete_distributions.html#index-entry-ce1eb98cf9c98edbf6d662f465f30c9816c1d6fe"><code>(real eta, vector c) : int</code></a> <span class="detail">(bounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="ordered_unconstrain" href="#ordered_unconstrain" class="anchored unlink"><strong>ordered_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-66bebbbf4a16907781a03d7910e05157b57546ae"><code>(vectors x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="owens_t" href="#owens_t" class="anchored unlink"><strong>owens_t</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1ebaab3f9c8786de6e45b245eb0557156a1b2a77"><code>(real h, real a) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-8ac2b39d27b678ab373ba570f18b633093c6511d"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="p" class="level2">
+<h2 class="anchored" data-anchor-id="p">P</h2>
+<p><a id="pareto" href="#pareto" class="anchored unlink"><strong>pareto</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-3868686bd088fa3182f600afc3a7982dfadfd439">distribution statement</a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_cdf" href="#pareto_cdf" class="anchored unlink"><strong>pareto_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-4abdf5cdce078735156bd724266dc2ea7f8b1d55"><code>(reals y | reals y_min, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_lccdf" href="#pareto_lccdf" class="anchored unlink"><strong>pareto_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-579987e1abda763d062fecdf84a43f9c43e7b2fd"><code>(reals y | reals y_min, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_lcdf" href="#pareto_lcdf" class="anchored unlink"><strong>pareto_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-4fb849c817175501195c6219da4b7f1cecad0b01"><code>(reals y | reals y_min, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_lpdf" href="#pareto_lpdf" class="anchored unlink"><strong>pareto_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-2b559a321ccbd5c4df2c32eafefcb7ed61d8497f"><code>(reals y | reals y_min, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_lupdf" href="#pareto_lupdf" class="anchored unlink"><strong>pareto_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-9908f14844c4a9e7fe15c5add847cd1baf8fa56c"><code>(reals y | reals y_min, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_rng" href="#pareto_rng" class="anchored unlink"><strong>pareto_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-33472469dac2a620b0f461365535f1f605f15104"><code>(reals y_min, reals alpha) : R</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2" href="#pareto_type_2" class="anchored unlink"><strong>pareto_type_2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-ae502c44f119dc495e9e3cb809d28da32af6d702">distribution statement</a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2_cdf" href="#pareto_type_2_cdf" class="anchored unlink"><strong>pareto_type_2_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-faf0b6f41480a1c429a34e564f63e9004c93fab1"><code>(reals y | reals mu, reals lambda, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2_lccdf" href="#pareto_type_2_lccdf" class="anchored unlink"><strong>pareto_type_2_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-1df9330c369a563524ac6a3256c1a18befe4f9f8"><code>(reals y | reals mu, reals lambda, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2_lcdf" href="#pareto_type_2_lcdf" class="anchored unlink"><strong>pareto_type_2_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-207a8840717cf656dcedb5daad63e6d01e8af9be"><code>(reals y | reals mu, reals lambda, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2_lpdf" href="#pareto_type_2_lpdf" class="anchored unlink"><strong>pareto_type_2_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-a01b65ffac6ae18cc37fa6af7c644e99276c1787"><code>(reals y | reals mu, reals lambda, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2_lupdf" href="#pareto_type_2_lupdf" class="anchored unlink"><strong>pareto_type_2_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-62575702409a828f934e569a6d98dffa197f2b93"><code>(reals y | reals mu, reals lambda, reals alpha) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="pareto_type_2_rng" href="#pareto_type_2_rng" class="anchored unlink"><strong>pareto_type_2_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-53cbc03223fcc002f296906a7751b5808b49507a"><code>(reals mu, reals lambda, reals alpha) : R</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="Phi" href="#Phi" class="anchored unlink"><strong>Phi</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-2a50a935e0b3ed75c3a19e2bee25f3a01c7799f7"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="Phi_approx" href="#Phi_approx" class="anchored unlink"><strong>Phi_approx</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1b36b9b4f834de053d1a484782736f41af7b107e"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="pi" href="#pi" class="anchored unlink"><strong>pi</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-53d994dc279b92ee9a153eb72068c45e7587230e"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson" href="#poisson" class="anchored unlink"><strong>poisson</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-149c7a2f2ae0aede0ae6b19854be4a72ba136d10">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_cdf" href="#poisson_cdf" class="anchored unlink"><strong>poisson_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-b9228882bc1807740bfbe2278fe3865895ddc14f"><code>(ints n | reals lambda) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_lccdf" href="#poisson_lccdf" class="anchored unlink"><strong>poisson_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-f630914d24a4ab0a6d95f950289e5a89655ee0b5"><code>(ints n | reals lambda) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_lcdf" href="#poisson_lcdf" class="anchored unlink"><strong>poisson_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-c53c3f02b2a02ee52cf555667086bbe0d63d9893"><code>(ints n | reals lambda) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log" href="#poisson_log" class="anchored unlink"><strong>poisson_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-48321c3e1d3764b7424c6aa9000bfb2c7c135d49">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log_glm" href="#poisson_log_glm" class="anchored unlink"><strong>poisson_log_glm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-6a7861af018d1e99d1f24782d56214c5b92e7586">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log_glm_lpmf" href="#poisson_log_glm_lpmf" class="anchored unlink"><strong>poisson_log_glm_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-9ccc16b1cd683cb83b9f375ecf6d419b4b28ad2e"><code>(array[] int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-eba869933862786def5f0691fd7751eeb0de388f"><code>(array[] int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-5edf084bfc5e8e7f2942c123d2485b9947832b4a"><code>(array[] int y | row_vector x, real alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-83fff7e4697319fe19b71c0eecbdf5ac8049b7cb"><code>(array[] int y | row_vector x, vector alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-63439a9c4647ac7c084fdb2e91704175532de973"><code>(int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-809af19433732f5658706fbd90b7e3489497b084"><code>(int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log_glm_lupmf" href="#poisson_log_glm_lupmf" class="anchored unlink"><strong>poisson_log_glm_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-a22ac3bb5bf0da63d28e08cceb394ea47b6f355b"><code>(array[] int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-edf1d3e6f8239fe43a077fe8e6dfdb7735799c66"><code>(array[] int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-279c01365489bac4622e7dcfcc8075202555568c"><code>(array[] int y | row_vector x, real alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-f817a209ecaeb8259b7e13294695ad91e1bac17b"><code>(array[] int y | row_vector x, vector alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-f122468454f01edf31a51909ec8967c9de864cf3"><code>(int y | matrix x, real alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-8754addaa08a378fc56a498e81745574e870f6fb"><code>(int y | matrix x, vector alpha, vector beta) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log_lpmf" href="#poisson_log_lpmf" class="anchored unlink"><strong>poisson_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-0a17b71120ddbfe9f78828a8db8ac327d7dfe433"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log_lupmf" href="#poisson_log_lupmf" class="anchored unlink"><strong>poisson_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-6b7b04889612003d401cf29b1505e899a5ac3bdf"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_log_rng" href="#poisson_log_rng" class="anchored unlink"><strong>poisson_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-2698e8369a4abee3bff347589811f0b59008cd62"><code>(reals alpha) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_lpmf" href="#poisson_lpmf" class="anchored unlink"><strong>poisson_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-468e745a522407843e4e4e0b5d50391126cfe3d6"><code>(ints n | reals lambda) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_lupmf" href="#poisson_lupmf" class="anchored unlink"><strong>poisson_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-ec57a3b0e25d5e9dd5cd37c27f9557e8123a4b4a"><code>(ints n | reals lambda) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="poisson_rng" href="#poisson_rng" class="anchored unlink"><strong>poisson_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-08d9cd65c9ed1ce8f26288c3f3c41ad143ddd486"><code>(reals lambda) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="polar" href="#polar" class="anchored unlink"><strong>polar</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-267e629612ebf97b806525573c76f11e23005e4a"><code>(real r, real theta) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="positive_infinity" href="#positive_infinity" class="anchored unlink"><strong>positive_infinity</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-21afff517501f022247efe89c81a255a74533422"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="positive_ordered_constrain" href="#positive_ordered_constrain" class="anchored unlink"><strong>positive_ordered_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-a5bd3aeff6187e7710c5c9aede0aa53b0bf05605"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="positive_ordered_jacobian" href="#positive_ordered_jacobian" class="anchored unlink"><strong>positive_ordered_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-72ae38a185968521d676ce568e4dd9dfb0ec3211"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="positive_ordered_unconstrain" href="#positive_ordered_unconstrain" class="anchored unlink"><strong>positive_ordered_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-ea0b4a03eff5caa57b2d154df97f2c8a63b820e5"><code>(vectors x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="pow" href="#pow" class="anchored unlink"><strong>pow</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-a687c481b458dcd50e13b27c0423402b6332fc08"><code>(complex x, complex y) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-0882f3a0b5f515684dd2a29821a274a09ada9426"><code>(real x, real y) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-c9ec850adf117cff8e4f8da2a8287b92a152b481"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-24c507386fa0ba9857a2bc0d50a9e2553e9ceba6"><code>(T1 x, T2 y) : Z</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="print" href="#print" class="anchored unlink"><strong>print</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/void_functions.html#index-entry-61af0108923d685571278dd77b2b31e38e67397a"><code>(T1 x1,..., TN xN) : void</code></a> <span class="detail">(void_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="prod" href="#prod" class="anchored unlink"><strong>prod</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-69d45ac78302f42fc87a23263c5230e4ee093a6e"><code>(array[] int x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-936503ecb56cc2f9aa31395e3b80d5dd901f14ab"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-2c4545b2f6cd6bbf4e42a4d446d210160e082d42"><code>(complex_matrix x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5353a09f94f67878ef023d0c25bdf91d78c1cfaa"><code>(complex_row_vector x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-12d6f5e8068b2b8308d32394e6debc2fa118e082"><code>(complex_vector x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b45dad30f80e5937e91c940a770331aae131753c"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c081c52af0fe73a003a0a406a8378cde350a34bf"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9713d871d72b91892fb1be528d008fcb62d40ba7"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="proj" href="#proj" class="anchored unlink"><strong>proj</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-8b9fe1df9b2e6d862d634979cd5a62478d612669"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="q" class="level2">
+<h2 class="anchored" data-anchor-id="q">Q</h2>
+<p><a id="qr" href="#qr" class="anchored unlink"><strong>qr</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0f3b3496deb99ba4839dce3aea55e6de916877e0"><code>(matrix A) : tuple(matrix, matrix)</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="qr_Q" href="#qr_Q" class="anchored unlink"><strong>qr_Q</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ecf3d80487f50dec1b5d8f7309ae1b12124339ee"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="qr_R" href="#qr_R" class="anchored unlink"><strong>qr_R</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b8d888170b3dfed1c077267a9a01cb4496d276f5"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="qr_thin" href="#qr_thin" class="anchored unlink"><strong>qr_thin</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-01eb10e298ec5e0b2570e847ea477501b645f229"><code>(matrix A) : tuple(matrix, matrix)</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="qr_thin_Q" href="#qr_thin_Q" class="anchored unlink"><strong>qr_thin_Q</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-00071108ad7f8a1ecce0c8659b622ad0c7e244b7"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="qr_thin_R" href="#qr_thin_R" class="anchored unlink"><strong>qr_thin_R</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-854d13c8d7fed0092f63b68aed49a27b3f7fd886"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="quad_form" href="#quad_form" class="anchored unlink"><strong>quad_form</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0cac546f1a4dfcfd8676f091dc0f7ab8cf808e4d"><code>(matrix A, matrix B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7d865d93151dbb42928e39384a33278c4695d490"><code>(matrix A, vector B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="quad_form_diag" href="#quad_form_diag" class="anchored unlink"><strong>quad_form_diag</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4c9654c4a3b4068858d9351459b026a697f02b8e"><code>(matrix m, row_vector rv) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7d5c708ffbdc829d13ee747d32cdeb5f6b5c5976"><code>(matrix m, vector v) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="quad_form_sym" href="#quad_form_sym" class="anchored unlink"><strong>quad_form_sym</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-cb6d97ec303631d01e1fbf4737567987878d14b2"><code>(matrix A, matrix B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-362d5553d7ff140b2163b35fed3667b17ad4c3d3"><code>(matrix A, vector B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="quantile" href="#quantile" class="anchored unlink"><strong>quantile</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-b0471485d6b0fd5b43b349cbe4e386d6c38f5064"><code>(data array[] real x, data array[] real p) : array[] real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-8c7df96194aa20f4691df231ed7577942daf6d6f"><code>(data array[] real x, data real p) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-ba154b25a2265003899559c90fe541c3a07a82e0"><code>(data row_vector x, data array[] real p) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5ca7a0e8fda0d0ac93828d2154479406eead9a61"><code>(data row_vector x, data real p) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-49b22dc2be74bb430f6e2faee8a2f6985280f397"><code>(data vector x, data array[] real p) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5bcb3bb3e5ae7f0a50cf00b2fb37cc35759014da"><code>(data vector x, data real p) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="r" class="level2">
+<h2 class="anchored" data-anchor-id="r">R</h2>
+<p><a id="rank" href="#rank" class="anchored unlink"><strong>rank</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a6d0770a4ab5bc51a44d3626199ee7de7658377e"><code>(array[] int v, int s) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a28838301370f881f9ec571b8676f50b783b629c"><code>(array[] real v, int s) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9dc76dd605fabb78a31bc22a4cbc3bba0371758d"><code>(row_vector v, int s) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a2612df2dc4076d90d32d7205a10905c38bcb241"><code>(vector v, int s) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh" href="#rayleigh" class="anchored unlink"><strong>rayleigh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-21e80622249b881fc41605bf036fc4acba1d2e27">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh_cdf" href="#rayleigh_cdf" class="anchored unlink"><strong>rayleigh_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-1a411f1904a7749524e9155b6695826fb8243d66"><code>(real y | real sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh_lccdf" href="#rayleigh_lccdf" class="anchored unlink"><strong>rayleigh_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-9d06d218d58232df14b1c128f0c42df1c1babaa8"><code>(real y | real sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh_lcdf" href="#rayleigh_lcdf" class="anchored unlink"><strong>rayleigh_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-4782b996228bdd1d006a2f0c6a7406b741855120"><code>(real y | real sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh_lpdf" href="#rayleigh_lpdf" class="anchored unlink"><strong>rayleigh_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-b04b39e49861434e1dc33eaab453c14e28da9f01"><code>(reals y | reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh_lupdf" href="#rayleigh_lupdf" class="anchored unlink"><strong>rayleigh_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-9778ca15ec9fc0fc7856485732abc302454e60c5"><code>(reals y | reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="rayleigh_rng" href="#rayleigh_rng" class="anchored unlink"><strong>rayleigh_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-97387dc941bc4a4869eee5819e8687df2f4ee96b"><code>(reals sigma) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="reduce_sum" href="#reduce_sum" class="anchored unlink"><strong>reduce_sum</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-a5981e88d5cfd9835be7c6fbd0f9b671cce9bad9"><code>(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="reduce_sum_static" href="#reduce_sum_static" class="anchored unlink"><strong>reduce_sum_static</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-ef3aada570b019e2390174fe86359b5af6427541"><code>(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="reject" href="#reject" class="anchored unlink"><strong>reject</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/void_functions.html#index-entry-4724be86187ac6f6f6706c7edabfc36317b4a586"><code>(T1 x1,..., TN xN) : void</code></a> <span class="detail">(void_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="rep_array" href="#rep_array" class="anchored unlink"><strong>rep_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a2816f987a767657e3323203e069be5323f40964"><code>(T x, int k, int m, int n) : array[,,] T</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-a5563d5b3ee4bd63fb6cbbef35e86bd940d587df"><code>(T x, int m, int n) : array[,] T</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-b16f7217e0f04da666e3544e1f8e27e04e429973"><code>(T x, int n) : array[] T</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rep_matrix" href="#rep_matrix" class="anchored unlink"><strong>rep_matrix</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-24c948121a2ee087e060ad1257d8719e1b9aa2c3"><code>(complex z, int m, int n) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e106eee17c0bdf0c73908ed222c19f595e566330"><code>(complex_row_vector rv, int m) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-680d23b80116edb9e6752b03801cf43a42571d6a"><code>(complex_vector v, int n) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f6dac444f76cc32475bbb31cceff748d10649662"><code>(real x, int m, int n) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a54769495396ea1427a26c419009fcf4b6b69cbf"><code>(row_vector rv, int m) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9088cad16f7504b9acf856be4fc670b42cbd9eaa"><code>(vector v, int n) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rep_row_vector" href="#rep_row_vector" class="anchored unlink"><strong>rep_row_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5bf4e064db435eff0410b72666643037d22a395c"><code>(complex z, int n) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-15a03522d48822ef641c68071772fb40994377ff"><code>(real x, int n) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rep_vector" href="#rep_vector" class="anchored unlink"><strong>rep_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-d9c905ecc938069ed009528bdab94da0534015b1"><code>(complex z, int m) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-946d7c6f889b9fa740b16d2261aea52649b26b16"><code>(real x, int m) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="reverse" href="#reverse" class="anchored unlink"><strong>reverse</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-9759b3b50ddba20806c98d04d90c234afd5b230e"><code>(array[] T v) : array[] T</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-6518a921f55ed48810b7eddb6fdeebc850a60c27"><code>(complex_row_vector v) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-4967c86d60f1814cc9c1b26860a5b0a2d875cfa6"><code>(complex_vector v) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6b1f33889b3d5f2093ad7854f2692c91b8adcd02"><code>(row_vector v) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-568dc0f51c2d88dc59f98ff0b5cf0080469ad789"><code>(vector v) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rising_factorial" href="#rising_factorial" class="anchored unlink"><strong>rising_factorial</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-740279c8bcb940f23acd2a981763223a674105b8"><code>(real x, int n) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1908eb05437a542da8fa4551ec25c60d068905cb"><code>(T1 x, T2 y) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="round" href="#round" class="anchored unlink"><strong>round</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b0ef648badd54cfdfec4e07e6a94362748815ccc"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="row" href="#row" class="anchored unlink"><strong>row</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-f8cb43eb2a385abfd8880fb1c741f30d7c691414"><code>(complex_matrix x, int m) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-5ad5162727cc2ef4908de930d86a6f86f1716ff8"><code>(matrix x, int m) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rows" href="#rows" class="anchored unlink"><strong>rows</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-9ebbc38c2cab2bb85fc80a2aace93188ecfcf794"><code>(complex_matrix x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-e14fb4048c19b782910ef083ea77519d2136be69"><code>(complex_row_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-8005f2c9cb8700f40b7a2aa479a6d62cc1ab309d"><code>(complex_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-65ffc1cace11a1dd2675fb84a331101e1ea8b626"><code>(matrix x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-578ea4c98303c7c7498cb69c853a18e09d5c44fd"><code>(row_vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c6697d2988eb22bf8fd56b05ffcde6779627ebf8"><code>(vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rows_dot_product" href="#rows_dot_product" class="anchored unlink"><strong>rows_dot_product</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-74d6233826ba207dc78ce8fa7c02a1177299542d"><code>(complex_matrix x, complex_matrix y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-734a199664218183d672f594e0ba3c472067cf38"><code>(complex_row_vector x, complex_row_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-9869bbad286feb6e3252d28a43fac51f950044c7"><code>(complex_vector x, complex_vector y) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-cf6ee59a255f43dac007f4dd3ef01a1119343cb0"><code>(matrix x, matrix y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9bb7b3a8ee495648abc1b0eff2ecebe52c99be25"><code>(row_vector x, row_vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-de1fe07c43494dac868152a24e25fcf81955732e"><code>(vector x, vector y) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="rows_dot_self" href="#rows_dot_self" class="anchored unlink"><strong>rows_dot_self</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-771452c2e576061539abb7bece2cec0051b51eb7"><code>(complex_matrix x) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-f485e7669d290a4148d1373c4705206fdef737e3"><code>(complex_row_vector x) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-9089a74ceffaf17dbabc8404c8575d1f4d6f92a9"><code>(complex_vector x) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-947ee99ec791f7bead5c155ced2ef89e5ac7e8b7"><code>(matrix x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b624c89d6c4d2e43359ff78346c37076b91730f1"><code>(row_vector x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a451434f24e97a590c47f3f0e691af26a61e8705"><code>(vector x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="s" class="level2">
+<h2 class="anchored" data-anchor-id="s">S</h2>
+<p><a id="scale_matrix_exp_multiply" href="#scale_matrix_exp_multiply" class="anchored unlink"><strong>scale_matrix_exp_multiply</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-09a3d2d3107a8732b5360be5accb8fcce6bd2987"><code>(real t, matrix A, matrix B) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square" href="#scaled_inv_chi_square" class="anchored unlink"><strong>scaled_inv_chi_square</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-d782005ee920606bf8bdd43885ae449421dce0f6">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square_cdf" href="#scaled_inv_chi_square_cdf" class="anchored unlink"><strong>scaled_inv_chi_square_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-be017a6dd0ea4e5a40dfbe2ed649d0837f935724"><code>(reals y | reals nu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square_lccdf" href="#scaled_inv_chi_square_lccdf" class="anchored unlink"><strong>scaled_inv_chi_square_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-5f64b2f0bae60e336dba14a7efe683097fc8c8c8"><code>(reals y | reals nu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square_lcdf" href="#scaled_inv_chi_square_lcdf" class="anchored unlink"><strong>scaled_inv_chi_square_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-e14593aac1cd14f7cf744da7a520e81ea74f2923"><code>(reals y | reals nu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square_lpdf" href="#scaled_inv_chi_square_lpdf" class="anchored unlink"><strong>scaled_inv_chi_square_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-9df549db2fb37eff9dfe90a6a22d14745f8c5fe5"><code>(reals y | reals nu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square_lupdf" href="#scaled_inv_chi_square_lupdf" class="anchored unlink"><strong>scaled_inv_chi_square_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-8b7dfa52248b073d01e53ceabcc698299d321803"><code>(reals y | reals nu, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="scaled_inv_chi_square_rng" href="#scaled_inv_chi_square_rng" class="anchored unlink"><strong>scaled_inv_chi_square_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-f0ccb76a8d7070c50483a9191a0177d55dc7a4a8"><code>(reals nu, reals sigma) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="sd" href="#sd" class="anchored unlink"><strong>sd</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-2ff81571d634ef5353d30cfebadc9561bf519ee4"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8af54c041853b0f940ca2f60a845500498fdcf2b"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0182a4edcf541473e053aff58f5a7d2eafa83574"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-41a7ec70ca2889f5e84fd426dfa08137bfaba28a"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="segment" href="#segment" class="anchored unlink"><strong>segment</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2374a68547cdedbbdfbb1b7212dd1ccaf48d84b3"><code>(array[] T sv, int i, int n) : array[] T</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-b59ddb42d7b01c4f637d83f1654f9b63e6f1d9cd"><code>(complex_row_vector rv, int i, int n) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-96d8e2e13f7a5c50b4fef83a87ac1dcd395ab08f"><code>(complex_vector v, int i, int n) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-26587b215542055b9fe52316209a88041c119b87"><code>(row_vector rv, int i, int n) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-71cb7ac02f1e74f4b6fff77caa9b5ef0bd15346a"><code>(vector v, int i, int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="simplex_constrain" href="#simplex_constrain" class="anchored unlink"><strong>simplex_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-055048ff4bc60482699857e7993258efc99d45f0"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="simplex_jacobian" href="#simplex_jacobian" class="anchored unlink"><strong>simplex_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-446c0a0a855c1f82fe5f0df51dddf43419abd15f"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="simplex_unconstrain" href="#simplex_unconstrain" class="anchored unlink"><strong>simplex_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-ffa2ad29148676e3139f6d81a3fba15199481f30"><code>(vectors x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="sin" href="#sin" class="anchored unlink"><strong>sin</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-13f23f848a721cdfa87c2d1f4b4067cb9df6911e"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-318c6c59e1d99978abb1432132810c6f097ed01f"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="singular_values" href="#singular_values" class="anchored unlink"><strong>singular_values</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-491e5da4f54a933c48db0826ddc9f9bff1b864f2"><code>(complex_matrix A) : vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-561216aa62b3975fc9a1a4e4d2158ed263b307f2"><code>(matrix A) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sinh" href="#sinh" class="anchored unlink"><strong>sinh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-8511329268543fd1404cfddba3d7bf1867ed3665"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1532402d69ee65eaf786341825da10fa8941a092"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="size" href="#size" class="anchored unlink"><strong>size</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-c12761833dc86f2313582b4fe440b0b82d4c8cb6"><code>(array[] T x) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-dd6512340d46b06fe64606e36622bf23ed255eab"><code>(complex_row_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-4a14d6b215559d7664aabfafde37e9153206b209"><code>(complex_vector x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-cc587d014dd8cd99986939af8ad1854a3c3425f0"><code>(int x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-69098855a034f282c0294e1c6c21cded79ec8a3c"><code>(matrix x) : int</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-69098855a034f282c0294e1c6c21cded79ec8a3c"><code>(matrix x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-f9c96384668fa7a7880511c2c0ad2558be24e0a4"><code>(real x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f449c3a25c74efefc2c1462cf914a52a4050d575"><code>(row_vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9b99cb12fadc8373556db600c949cc44b43fe907"><code>(vector x) : int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential" href="#skew_double_exponential" class="anchored unlink"><strong>skew_double_exponential</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-c133dee4da92b9a21386219b4a5fbb6b5ced9f23">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential_cdf" href="#skew_double_exponential_cdf" class="anchored unlink"><strong>skew_double_exponential_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-bbc59f7f6b6b9b62a78cc86780d31dbc261ef032"><code>(reals y | reals mu, reals sigma, reals tau) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential_lccdf" href="#skew_double_exponential_lccdf" class="anchored unlink"><strong>skew_double_exponential_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-3b13bcb4e64d73e98c4ff5ebfa01691c2ba8cb43"><code>(reals y | reals mu, reals sigma, reals tau) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential_lcdf" href="#skew_double_exponential_lcdf" class="anchored unlink"><strong>skew_double_exponential_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4d7480cfff426af8ef69e62abd1a90b246aa386d"><code>(reals y | reals mu, reals sigma, reals tau) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential_lpdf" href="#skew_double_exponential_lpdf" class="anchored unlink"><strong>skew_double_exponential_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-903ddf1695b27de2b4e5bcdfd95493d2b2c40105"><code>(reals y | reals mu, reals sigma, reals tau) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential_lupdf" href="#skew_double_exponential_lupdf" class="anchored unlink"><strong>skew_double_exponential_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-344e5af2a56736dcd0d5a8df8fa7f5204b2392b4"><code>(reals y | reals mu, reals sigma, reals tau) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_double_exponential_rng" href="#skew_double_exponential_rng" class="anchored unlink"><strong>skew_double_exponential_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-5ba4f22bc09b1f9b4aaef277e2148dd77a3dd8b6"><code>(reals mu, reals sigma) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal" href="#skew_normal" class="anchored unlink"><strong>skew_normal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-70ae09d4a5fd934163b857b8af6509a78d7717df">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal_cdf" href="#skew_normal_cdf" class="anchored unlink"><strong>skew_normal_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4e1fe342b2c2a2f8efd1b736d66c0d918bcd6699"><code>(reals y | reals xi, reals omega, reals alpha) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal_lccdf" href="#skew_normal_lccdf" class="anchored unlink"><strong>skew_normal_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f317d012b6071ab55a6345c89527fb421e4607ec"><code>(reals y | reals xi, reals omega, reals alpha) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal_lcdf" href="#skew_normal_lcdf" class="anchored unlink"><strong>skew_normal_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-55d8df839becff01c2c535a00f18ae07d745f20f"><code>(reals y | reals xi, reals omega, reals alpha) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal_lpdf" href="#skew_normal_lpdf" class="anchored unlink"><strong>skew_normal_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f14e52bfc7ddf652898086b26b652a779adfffbd"><code>(reals y | reals xi, reals omega, reals alpha) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal_lupdf" href="#skew_normal_lupdf" class="anchored unlink"><strong>skew_normal_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-57aff3449f2e94559d54277820ea0537a5b2d44f"><code>(reals y | reals xi, reals omega, reals alpha) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="skew_normal_rng" href="#skew_normal_rng" class="anchored unlink"><strong>skew_normal_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-18eec472b4732674d456a7bf36148e16322c3a0d"><code>(reals xi, reals omega, real alpha) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="softmax" href="#softmax" class="anchored unlink"><strong>softmax</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9267f9986949f0269013001d932bef84c805fa45"><code>(vector x) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="solve_newton" href="#solve_newton" class="anchored unlink"><strong>solve_newton</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-85f9ba73adda0bd2add83f06a31c5cd63ccb63fa"><code>(function algebra_system, vector y_guess, ...) : vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="solve_newton_tol" href="#solve_newton_tol" class="anchored unlink"><strong>solve_newton_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-0283f73334357f2f1b816fdd22197b93bdf02321"><code>(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="solve_powell" href="#solve_powell" class="anchored unlink"><strong>solve_powell</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-cb1a93d1015ecd4883af1f04902daf27984a760d"><code>(function algebra_system, vector y_guess, ...) : vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="solve_powell_tol" href="#solve_powell_tol" class="anchored unlink"><strong>solve_powell_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/higher-order_functions.html#index-entry-6d887fe07d5519082bcb6c789283ce3909e661d7"><code>(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector</code></a> <span class="detail">(higher-order_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="sort_asc" href="#sort_asc" class="anchored unlink"><strong>sort_asc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-6e75416be841609073a75454c55e2fbc19f46f47"><code>(array[] int v) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-29411da2ef662a3af3793ab20c89f7e7b677bd0e"><code>(array[] real v) : array[] real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9bc88f386e57fb9d4e7a2e0fd914d4525b7271c8"><code>(row_vector v) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-3f7cd093f7a97b178afa3f28332faee31600a6c9"><code>(vector v) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sort_desc" href="#sort_desc" class="anchored unlink"><strong>sort_desc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-355460b3ae8892bf699932f697801d5f961e04e8"><code>(array[] int v) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-fce023d8773b92438ce6a759565ce44deab3681c"><code>(array[] real v) : array[] real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-455584fc2e149f835e4f434734d42537044906a0"><code>(row_vector v) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9e567cdae4513e4fd1de2f40bda8578202f4b75f"><code>(vector v) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sort_indices_asc" href="#sort_indices_asc" class="anchored unlink"><strong>sort_indices_asc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-22998db91982cee6e0c3b1a618078f76b085d02c"><code>(array[] int v) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-88b08c67456c5ae147bac2ccabf5a2435eb84def"><code>(array[] real v) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-162c46d9f5b0bdb7329ba9d5db3e248a017bf548"><code>(row_vector v) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f37b48f43d2c6eaaeaa6b012dfae4912e9ac6869"><code>(vector v) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sort_indices_desc" href="#sort_indices_desc" class="anchored unlink"><strong>sort_indices_desc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-f8faa56dc58a61ee15ff8a3545ecc6aee8b3110d"><code>(array[] int v) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-fa29a352fe066806f7fafbebc5a2510bab51596f"><code>(array[] real v) : array[] int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-29d0b54b37e8b0e7e36145bd3ad202aa5d5c7362"><code>(row_vector v) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7660a9ad94113629502c6f1a3796f2e1e4033ce0"><code>(vector v) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sqrt" href="#sqrt" class="anchored unlink"><strong>sqrt</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-1e8198cd9adf62979a9a07f6d4548f72a1ada06d"><code>(complex x) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b3bbd4cc788019bca7f70823777f48ce5f43e4b4"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="sqrt2" href="#sqrt2" class="anchored unlink"><strong>sqrt2</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-1ded3db8572ff7287e4d61f28f486873fd4b76a1"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="square" href="#square" class="anchored unlink"><strong>square</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b9f6f7932609a3f72301693abb282aa60845a210"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="squared_distance" href="#squared_distance" class="anchored unlink"><strong>squared_distance</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-794d03938547aad86944e55cbb1f64f9ea22dee5"><code>(row_vector x, row_vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-8887aa905b17d368550a915063d1bb8fbf4ccf59"><code>(row_vector x, vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-e2cb205544293c316ec3e5356c4d1dbebfafb159"><code>(vector x, row_vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-47ca27c5ccd71b8bebcadb9723ea604c50120857"><code>(vector x, vector y) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal" href="#std_normal" class="anchored unlink"><strong>std_normal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-8fd9b26fa3b56249d7987f7189e234de82c38812">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_cdf" href="#std_normal_cdf" class="anchored unlink"><strong>std_normal_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-d682f4986b7887882cc42788a87dd1d67950e7cc"><code>(reals y) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_lccdf" href="#std_normal_lccdf" class="anchored unlink"><strong>std_normal_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-bf6a81086184bebf728088395f30c5755cd8562a"><code>(reals y) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_lcdf" href="#std_normal_lcdf" class="anchored unlink"><strong>std_normal_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-925bd18ef58df9b2bf3bf0dc86898c8633dc2373"><code>(reals y) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_log_qf" href="#std_normal_log_qf" class="anchored unlink"><strong>std_normal_log_qf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-fe25ac0f35036372c19c2da7bb28d37f9a80cf2c"><code>(T x) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_lpdf" href="#std_normal_lpdf" class="anchored unlink"><strong>std_normal_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-3ae6425bfbb49379f4d0565899a18873e6b23747"><code>(reals y) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_lupdf" href="#std_normal_lupdf" class="anchored unlink"><strong>std_normal_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-813f27102ca53e25d42a3ef9c42bd919f4390d6d"><code>(reals y) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_qf" href="#std_normal_qf" class="anchored unlink"><strong>std_normal_qf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-9bac63544b561d2822dd4d977b4e0ce0b78f183b"><code>(T x) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="std_normal_rng" href="#std_normal_rng" class="anchored unlink"><strong>std_normal_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-4d08a0371e8f4ad5a8ac20add31783b3a31ed310"><code>() : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="step" href="#step" class="anchored unlink"><strong>step</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-b5b8e03f66b31507dac3535af6f2d8f615e4a3af"><code>(real x) : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="stochastic_column_constrain" href="#stochastic_column_constrain" class="anchored unlink"><strong>stochastic_column_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-4dbd1cdc02cc83f417b72d0ce15257683589e66c"><code>(matrices y) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="stochastic_column_jacobian" href="#stochastic_column_jacobian" class="anchored unlink"><strong>stochastic_column_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-a21e88d79c0bb4cc8424754bfac7f0bfada0be42"><code>(matrices y) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="stochastic_column_unconstrain" href="#stochastic_column_unconstrain" class="anchored unlink"><strong>stochastic_column_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-a04a2804c70fb0fe2219a7d11244837772f7d404"><code>(matrices x) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="stochastic_row_constrain" href="#stochastic_row_constrain" class="anchored unlink"><strong>stochastic_row_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-e2ce12f9782fbeb39c70aa822e3b8a5d4ef4baf8"><code>(matrices y) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="stochastic_row_jacobian" href="#stochastic_row_jacobian" class="anchored unlink"><strong>stochastic_row_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-b4c7ad92b82702ad1209da8fa540ff52110b0bf0"><code>(matrices y) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="stochastic_row_unconstrain" href="#stochastic_row_unconstrain" class="anchored unlink"><strong>stochastic_row_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-4bf40767700b7e2fc07e1045d565df3dc93422ee"><code>(matrices x) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t" href="#student_t" class="anchored unlink"><strong>student_t</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-2483b90edc6356f6ed17c209456b226b1bc0099b">distribution statement</a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t_cdf" href="#student_t_cdf" class="anchored unlink"><strong>student_t_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-c14ffe34acf237ad6f14abea4b420b8d1ddc91c0"><code>(reals y | reals nu, reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t_lccdf" href="#student_t_lccdf" class="anchored unlink"><strong>student_t_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-f066450f95e0c6d86c44d8a4ecd7c462c0349ca9"><code>(reals y | reals nu, reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t_lcdf" href="#student_t_lcdf" class="anchored unlink"><strong>student_t_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-c66e35d8ec8239f600bd048ba98bb327054d7e3e"><code>(reals y | reals nu, reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t_lpdf" href="#student_t_lpdf" class="anchored unlink"><strong>student_t_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-b9eec4ab31579ecbd1b7330def17803b0e0fc7a4"><code>(reals y | reals nu, reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t_lupdf" href="#student_t_lupdf" class="anchored unlink"><strong>student_t_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-324d76557438cfbe15364f652d1fde26b764ad5e"><code>(reals y | reals nu, reals mu, reals sigma) : real</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="student_t_rng" href="#student_t_rng" class="anchored unlink"><strong>student_t_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_continuous_distributions.html#index-entry-66f1428f578865e4e4d3a0fc3c845d58bbfbdfb0"><code>(reals nu, reals mu, reals sigma) : R</code></a> <span class="detail">(unbounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="sub_col" href="#sub_col" class="anchored unlink"><strong>sub_col</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-177f046c456c0586f89f6731e2de2124009c4498"><code>(complex_matrix x, int i, int j, int n_rows) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-fb1126ac4a4410b2b2264dab2a17005f57546883"><code>(matrix x, int i, int j, int n_rows) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sub_row" href="#sub_row" class="anchored unlink"><strong>sub_row</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-07fb8f1cd8ba380cf781e1871c50fba8f8c31164"><code>(complex_matrix x, int i, int j, int n_cols) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f6b0581323605859a322af2e4cfb0eac1d6a67f8"><code>(matrix x, int i, int j, int n_cols) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sum" href="#sum" class="anchored unlink"><strong>sum</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-e7ecc93be79d76e228adf2a4a2f9cdd908a7238b"><code>(array[] complex x) : complex</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-d97ce537eae8d48ac3ddc98520d0c5eaafd6fe27"><code>(array[] int x) : int</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-acfd44849111bfa6024f821401db39f1181f4d14"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-3cba7a8e0e6f9daf41a8ac994c85688f887c228b"><code>(complex_matrix x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-bbcdf1952ec801328e03cb3637166b7a9f2f3cff"><code>(complex_row_vector x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-3cf6379c6f6f87df7a312af029238c24af388e45"><code>(complex_vector x) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7a966705e2b5c7762ea80adedd66384f947e0d59"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-6e609b116c423a798df7d214f82a296fb3c45baa"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a3848926d02b63042a5cca7f5f2114a98d5a3b54"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="sum_to_zero_constrain" href="#sum_to_zero_constrain" class="anchored unlink"><strong>sum_to_zero_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-1c9a600c3c9e18bf8b358b6e0e1a196de3adb95d"><code>(matrices y) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-6c0d205ee6463bcaa0673c8d9607d2ca20bf2860"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="sum_to_zero_jacobian" href="#sum_to_zero_jacobian" class="anchored unlink"><strong>sum_to_zero_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-5fbdab6082a06ae172bb0aa3d1c27d00d6acce40"><code>(matrices y) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-aeaf42b6a6102bf306ab44b353d64e75c52f5867"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="sum_to_zero_unconstrain" href="#sum_to_zero_unconstrain" class="anchored unlink"><strong>sum_to_zero_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-05440d06c001dcf3039cebf913897b51b2ee23e5"><code>(matrices x) : matrices</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-43bd9130545904d1b8cea836c9621a69cd06f361"><code>(vectors x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="svd" href="#svd" class="anchored unlink"><strong>svd</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-250461905ab5c8f3351131c12e8ea6f1b3e68795"><code>(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix)</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2c5c96a002835c270f25ef8c2e2cebacf7a03c02"><code>(matrix A) : tuple(matrix, vector, matrix)</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="svd_U" href="#svd_U" class="anchored unlink"><strong>svd_U</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-66a9fbf01c3972f54a4e86c8766100b5aa78a187"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-11a4ca04c6b3043104d40033be0b59dca0812809"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="svd_V" href="#svd_V" class="anchored unlink"><strong>svd_V</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-ae2d55818ccca5b294b06be1082ae607075dba37"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-93d86ec7f54351b8ea720e400f10a534c1b87ee0"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="symmetrize_from_lower_tri" href="#symmetrize_from_lower_tri" class="anchored unlink"><strong>symmetrize_from_lower_tri</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-99c3b184331464981edec14b1ad54d7ed2f94251"><code>(complex_matrix A) : complex_matrix</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-e4997fec1a21248abde86960359ea25d95244d9b"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="t" class="level2">
+<h2 class="anchored" data-anchor-id="t">T</h2>
+<p><a id="tail" href="#tail" class="anchored unlink"><strong>tail</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4cdc4c7bbfd1b9012474960694df39fd3ac0bd73"><code>(array[] T sv, int n) : array[] T</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-104f1db13950e7c221c9ce66c329f9ebb415fbe2"><code>(complex_row_vector rv, int n) : complex_row_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-5072d264c287300a4f02919b52048769218f5864"><code>(complex_vector v, int n) : complex_vector</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-aa91264e50992aae5e2a411f1e27c3136e23148a"><code>(row_vector rv, int n) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-c59d1ebbba5b4cd5cf3ff22130cb426b46df662a"><code>(vector v, int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="tan" href="#tan" class="anchored unlink"><strong>tan</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-e54110489bc4cabaef154e4257634208c6acce5f"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-02f62dc1271129383c9e257d1a6dea18851a7411"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="tanh" href="#tanh" class="anchored unlink"><strong>tanh</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-23b49be15163071c1ebe75f992e21daaba8d915f"><code>(complex z) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4f8f8565db3a7961f780e532e82cd5db1659bdc9"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="target" href="#target" class="anchored unlink"><strong>target</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-138fdd54bc9b599fe8f41f4558f1be1aa3ace172"><code>() : real</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="tcrossprod" href="#tcrossprod" class="anchored unlink"><strong>tcrossprod</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-0debd0686ed364955c554407a48932c2cc0e1ecf"><code>(matrix x) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="tgamma" href="#tgamma" class="anchored unlink"><strong>tgamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-86c34f22971aeaafc52aaea7501baa231b8ce5e4"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="to_array_1d" href="#to_array_1d" class="anchored unlink"><strong>to_array_1d</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-a283222b01f625bc6705e111e10bdc2759e3d24c"><code>(array[...] complex a) : array[] complex</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-febcf6672aef91647fe55778211ffd168082990f"><code>(array[...] int a) : array[] int</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-d4bca2f4496bd91a3f19415933ba8d5dee724894"><code>(array[...] real a) : array[] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-841e76b5eb14b885bfb8ea725edd4531171d3dc7"><code>(complex_matrix m) : array[] complex</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-9dbf7ca0cd05cc0137132abe8de599edc8db6fb6"><code>(complex_row_vector v) : array[] complex</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-242195f746365e4ab07756d8dfc2d627d760131c"><code>(complex_vector v) : array[] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-89bbd3b92786cfeacbc58a9b8e1815eb4704fe94"><code>(matrix m) : array[] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-d5e3525dd3b1532ec70529a13582b2ac5684b7a4"><code>(row_vector v) : array[] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-2f5e845bb7c1046346206e70df45afa871d3bbca"><code>(vector v) : array[] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="to_array_2d" href="#to_array_2d" class="anchored unlink"><strong>to_array_2d</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-67b1d2f656b711d5b1f916e3e04e26038c9d38d6"><code>(complex_matrix m) : array[,] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-1523c6dc72269a2b99daf2481a6f15790faf2dd7"><code>(matrix m) : array[,] real</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="to_complex" href="#to_complex" class="anchored unlink"><strong>to_complex</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-a10094b116ddff99cf1ba20f448c331b23a70d51"><code>() : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-8cc4a1be479f01313e354510097189d0ab95da1a"><code>(real re) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-982b3b69d3e75c559ef3cada7b1212eecc087fac"><code>(real re, real im) : complex</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/complex-valued_basic_functions.html#index-entry-e003c1a960f7ece5344e1cd438e4a3ce92e225e9"><code>(T1 re, T2 im) : Z</code></a> <span class="detail">(complex-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="to_int" href="#to_int" class="anchored unlink"><strong>to_int</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/integer-valued_basic_functions.html#index-entry-04b9411b8a7b904eccd2c168519bf4ccca638602"><code>(data real x) : int</code></a> <span class="detail">(integer-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="to_matrix" href="#to_matrix" class="anchored unlink"><strong>to_matrix</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-e92b8a0a6976052fa8b22b2d2678e9daa1dc22b0"><code>(array[,] complex a ) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-3e8783df8a1f5a3a9252432f08e410ff98dcc3cc"><code>(array[,] int a) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-57abafe2911be5f079350cd935db3790df0e435b"><code>(array[,] real a) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-bf764d9c03e5342181199ab42cf29145c1ad5854"><code>(array[] complex a, int m, int n) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-cae4771e239eee8858e814def2ab98579b41b5eb"><code>(array[] complex a, int m, int n, int col_major) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-30e7bab8b93cd06e7e722a69f87795853a321d6f"><code>(array[] complex_row_vector vs) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-e84dbcb742d813d14126e7e47ce17fad4d98c4ce"><code>(array[] int a, int m, int n) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-2114a4d1eb26abd7a5a168f852fa2824b04264a7"><code>(array[] int a, int m, int n, int col_major) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-437db9cf0cf2250154a98c0bdad40ed9b384a755"><code>(array[] real a, int m, int n) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-49a6288ee0b5970768e1b5eb800c235cdc531dcb"><code>(array[] real a, int m, int n, int col_major) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-43ba4671c0aedffb35391977e1454635f31140b1"><code>(array[] row_vector vs) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-e90eb83d724da49ec11f03ab61860acf90355da5"><code>(complex_matrix A, int m, int n, int col_major) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-1adffb84d0b4c5a1ad67ca2dce769cbdf687f577"><code>(complex_matrix m) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-847d5e8f061de6894025068cf5f2eb0ba8215a31"><code>(complex_matrix M, int m, int n) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-80cf0f8d6dea99916e4e16a9a5c04829bb454065"><code>(complex_row_vector v) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-0b65adda0355db4bb294b2accb6ac1294603c298"><code>(complex_row_vector v, int m, int n) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-23ae52701cdc53e4b1252d837f1fd68b9cf3c092"><code>(complex_row_vector v, int m, int n, int col_major) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-aef6a0f55fed76a89b125774992391483f616f15"><code>(complex_vector v) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-b6f6e03fe842ef9ce3d6ca9aabb56537f122d502"><code>(complex_vector v, int m, int n) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-38aaa00ecbf298c2b6ea8664028ae6e1db65995b"><code>(complex_vector v, int m, int n, int col_major) : complex_matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-ce9fd0a6767674968523193f7115d495489cf449"><code>(matrix A, int m, int n, int col_major) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-98af15f087742b3c48602bac340d541ab6538cc6"><code>(matrix m) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-c513be95c486e67bd8b313bb220e6184fea9292a"><code>(matrix M, int m, int n) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-792df8d223c6b836eee60bf39ca9d7d3f3e159b7"><code>(row_vector v) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-ed0fbc154447855f068f1a47cdf0f67c288a21d0"><code>(row_vector v, int m, int n) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-7f27f7aa2cd34abf4dc96d66caac862b24766c48"><code>(row_vector v, int m, int n, int col_major) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-50fdb634f6d81eb2f8f0a2b2aba97a29cb575e69"><code>(vector v) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-aed7ba7d91d4311c9ff956befae058ab2c5a56d5"><code>(vector v, int m, int n) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-0cf7e9303e3c6015089d4d776f68673d9becfe06"><code>(vector v, int m, int n, int col_major) : matrix</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="to_row_vector" href="#to_row_vector" class="anchored unlink"><strong>to_row_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-a7fd1cdc7869d1b6354acbb6c4d601343e80bc32"><code>(array[] complex a) : complex_row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-dfad2435683430f7d34faf59c9ef36aaeef30c0f"><code>(array[] int a) : row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-cea7d9d7492857fd3d8ca4b00cbb08cd09beb181"><code>(array[] real a) : row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-48b04a9a3d0a2ba72b4564f6c56c7787d7d85a3d"><code>(complex_matrix m) : complex_row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-c2aadf6729da7ef673b6e4729cd3640e7e8ff075"><code>(complex_row_vector v) : complex_row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-91aa719d0ab39507c033078fbe737e6992bf641b"><code>(complex_vector v) : complex_row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-2233e067e71fece1a5091dbf29c83f891c0c53ac"><code>(matrix m) : row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-dd3935354cbe9c750365fa9d199935099c0fccd5"><code>(row_vector v) : row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-d10d7dd47bea28e9f6be7cb1fee79ba98a6291e2"><code>(vector v) : row_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="to_vector" href="#to_vector" class="anchored unlink"><strong>to_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-b7e01d700955b123c6a7d94113e75dae2182ca9d"><code>(array[] complex a) : complex_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-32661d5b8760ecddd2d8777793db7ece329e55ec"><code>(array[] int a) : vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-c912c1c0f3589db9d1a5d2bffb2b3b697c20f3d7"><code>(array[] real a) : vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-038b0de9dc724055453f09b0cbcbd18ed0b245f5"><code>(complex_matrix m) : complex_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-5976d67e9306d7472825e4d06e32e82b5c9ed3b0"><code>(complex_row_vector v) : complex_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-d05604f9063eb5e5f46e231e2dedd8a2be6250c3"><code>(complex_vector v) : complex_vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-6f0903591f2ae71c76fb2f1c0a3330e7a39749c3"><code>(matrix m) : vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-3c628a0f6a20eda39461617826fd33046b2dc1c3"><code>(row_vector v) : vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/mixed_operations.html#index-entry-5eaf76d379576fc126f8cb1a1ddde01467fd11b4"><code>(vector v) : vector</code></a> <span class="detail">(mixed_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="trace" href="#trace" class="anchored unlink"><strong>trace</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/complex_matrix_operations.html#index-entry-c811524b370c422ec42cb96ee9cedc298094e51a"><code>(complex_matrix A) : complex</code></a> <span class="detail">(complex_matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-932fc0bd0a556886a853004f8ad6bc3432760b62"><code>(matrix A) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="trace_dot" href="#trace_dot" class="anchored unlink"><strong>trace_dot</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b067f1dc33e4e60522d490bc635d4305d0e9e7fb"><code>(matrix A, matrix B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="trace_gen_quad_form" href="#trace_gen_quad_form" class="anchored unlink"><strong>trace_gen_quad_form</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-725c7c07b916a247f645ce441212b4a82778b096"><code>(matrix D ,matrix A, matrix B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="trace_quad_form" href="#trace_quad_form" class="anchored unlink"><strong>trace_quad_form</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9e673dba39e3d7e6297d26eda1aaa01ff737cd89"><code>(matrix A, matrix B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-4102366e38a8c1eb82e650b1867da9a37385278a"><code>(matrix A, vector B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="trigamma" href="#trigamma" class="anchored unlink"><strong>trigamma</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-af71c258411af390fa86b5cee39043dc0f7aa2aa"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="trunc" href="#trunc" class="anchored unlink"><strong>trunc</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/real-valued_basic_functions.html#index-entry-4a8fc8e985608887eaa81d790ca2ee2e93e9c16e"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="u" class="level2">
+<h2 class="anchored" data-anchor-id="u">U</h2>
+<p><a id="uniform" href="#uniform" class="anchored unlink"><strong>uniform</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-69818db9c270c04d0dc488ff67849b0468ace0a2">distribution statement</a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_cdf" href="#uniform_cdf" class="anchored unlink"><strong>uniform_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-900d3b829004efdfb52e8a87bb4e3db6f8dc7e2f"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_lccdf" href="#uniform_lccdf" class="anchored unlink"><strong>uniform_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-82f75d6a35e833738553f28093a1965fde067385"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_lcdf" href="#uniform_lcdf" class="anchored unlink"><strong>uniform_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-2d8ca4cf434451528427871a1554072069737db0"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_lpdf" href="#uniform_lpdf" class="anchored unlink"><strong>uniform_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-cb6789cd6142d930347246410f7f33f97c3a02d6"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_lupdf" href="#uniform_lupdf" class="anchored unlink"><strong>uniform_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-7de2ecbf83919f5a16933fa6cc4876a9cced78c9"><code>(reals y | reals alpha, reals beta) : real</code></a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_rng" href="#uniform_rng" class="anchored unlink"><strong>uniform_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/bounded_continuous_distributions.html#index-entry-bbdbbdc16c0990d4e4cc5876ee605cf3c8bcb6be"><code>(reals alpha, reals beta) : R</code></a> <span class="detail">(bounded_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="uniform_simplex" href="#uniform_simplex" class="anchored unlink"><strong>uniform_simplex</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2dec8337a5bdb9528da444f890d895306c7de4e2"><code>(int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="unit_vectors_constrain" href="#unit_vectors_constrain" class="anchored unlink"><strong>unit_vectors_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-3e7ebedc4498436d0a2336970080bf855c52eaa7"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="unit_vectors_jacobian" href="#unit_vectors_jacobian" class="anchored unlink"><strong>unit_vectors_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-dad75a5d7cfa2f707b43c61f4208876d4f1ff189"><code>(vectors y) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="unit_vectors_unconstrain" href="#unit_vectors_unconstrain" class="anchored unlink"><strong>unit_vectors_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-8030393eb30e7d856eced92680f3c9e8454c19d2"><code>(vectors x) : vectors</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="upper_bound_constrain" href="#upper_bound_constrain" class="anchored unlink"><strong>upper_bound_constrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-762c88752de52b0a7df91820157122b1ddd4e9f1"><code>(reals y, reals ub) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="upper_bound_jacobian" href="#upper_bound_jacobian" class="anchored unlink"><strong>upper_bound_jacobian</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-f2ff19db2443c6d37b95266808e87959db60e3cd"><code>(reals x, reals ub) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+<p><a id="upper_bound_unconstrain" href="#upper_bound_unconstrain" class="anchored unlink"><strong>upper_bound_unconstrain</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/transform_functions.html#index-entry-2addb4ba29c3510e0935adaeb229c3bb34d49798"><code>(reals x, reals ub) : reals</code></a> <span class="detail">(transform_functions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="v" class="level2">
+<h2 class="anchored" data-anchor-id="v">V</h2>
+<p><a id="variance" href="#variance" class="anchored unlink"><strong>variance</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/array_operations.html#index-entry-92ca3dfea614e505cf3d32942458a9c948f45e92"><code>(array[] real x) : real</code></a> <span class="detail">(array_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-a59da01b6e381f01f0c9a08137440eea6455a2a2"><code>(matrix x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-f595237eba9511d08bff024bd377900cc94e0c62"><code>(row_vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-500efa24724b539cd9b32e37fb5895b0cd48f91c"><code>(vector x) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises" href="#von_mises" class="anchored unlink"><strong>von_mises</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-f221b5198520eda13cdf889d916d8a03cdfa8be2">distribution statement</a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises_cdf" href="#von_mises_cdf" class="anchored unlink"><strong>von_mises_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-338b3220a2284559e37acf449e5797e8228eccec"><code>(reals y | reals mu, reals kappa) : real</code></a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises_lccdf" href="#von_mises_lccdf" class="anchored unlink"><strong>von_mises_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-b597e729ddd78b9e20def4bc010de60d211a0e78"><code>(reals y | reals mu, reals kappa) : real</code></a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises_lcdf" href="#von_mises_lcdf" class="anchored unlink"><strong>von_mises_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-007efd59371f452dd7592246dd40178529aafdbf"><code>(reals y | reals mu, reals kappa) : real</code></a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises_lpdf" href="#von_mises_lpdf" class="anchored unlink"><strong>von_mises_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-5da3c7b18a3059fdead5df590cb8ec3f315ab761"><code>(reals y | reals mu, reals kappa) : real</code></a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises_lupdf" href="#von_mises_lupdf" class="anchored unlink"><strong>von_mises_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-22013d6e68ae63899ad52c563e8bd7509807f9e5"><code>(reals y | reals mu, reals kappa) : real</code></a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="von_mises_rng" href="#von_mises_rng" class="anchored unlink"><strong>von_mises_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/circular_distributions.html#index-entry-4deed6b0f3f1d7212c757bed3a7ab75376f4463c"><code>(reals mu, reals kappa) : R</code></a> <span class="detail">(circular_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="w" class="level2">
+<h2 class="anchored" data-anchor-id="w">W</h2>
+<p><a id="weibull" href="#weibull" class="anchored unlink"><strong>weibull</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-d1783373dc225fca3334f2daff416bfaf76d19fb">distribution statement</a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="weibull_cdf" href="#weibull_cdf" class="anchored unlink"><strong>weibull_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-bcc6a2519ae687354f1d256427ad98df385010f2"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="weibull_lccdf" href="#weibull_lccdf" class="anchored unlink"><strong>weibull_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-1a56c82ab70897619b18238a305b239f0d09fb01"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="weibull_lcdf" href="#weibull_lcdf" class="anchored unlink"><strong>weibull_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-90d29f1f2935caa157a84fa85aff411976e008ff"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="weibull_lpdf" href="#weibull_lpdf" class="anchored unlink"><strong>weibull_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-3797583b218047be3359306086c47d43b91a7cad"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="weibull_lupdf" href="#weibull_lupdf" class="anchored unlink"><strong>weibull_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-4d40f28dddfa369222ba458b9ff111fdb7747664"><code>(reals y | reals alpha, reals sigma) : real</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="weibull_rng" href="#weibull_rng" class="anchored unlink"><strong>weibull_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_continuous_distributions.html#index-entry-e84fd16978c29fe385bfa374bc3850c6a37a5347"><code>(reals alpha, reals sigma) : R</code></a> <span class="detail">(positive_continuous_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wiener" href="#wiener" class="anchored unlink"><strong>wiener</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-a8609b3c51ac8487f51a5e8a445e626015dad49f">distribution statement</a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wiener_lccdf_unnorm" href="#wiener_lccdf_unnorm" class="anchored unlink"><strong>wiener_lccdf_unnorm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-fd7dd400c38dd551f61ba9778cad3aefbb675a5e"><code>(real y, real alpha, real tau, real beta, real delta) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-d195ea5f4d8eba49c56313a0623fd5deb56d02fd"><code>(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wiener_lcdf_unnorm" href="#wiener_lcdf_unnorm" class="anchored unlink"><strong>wiener_lcdf_unnorm</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-ffe70852994a9eadec88b671cd728e98feffb122"><code>(real y, real alpha, real tau, real beta, real delta) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-59df6a6c983bf3f79c62ff8031367686b0ffa70a"><code>(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wiener_lpdf" href="#wiener_lpdf" class="anchored unlink"><strong>wiener_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-dd5110522b6a5fb9baac9dab11ad3dca296d7a24"><code>(real y | real alpha, real tau, real beta, real delta, real var_delta) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-60f0354d0a8198be7994a9f5d5cab49f31a26aa1"><code>(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-7c21d286a5cd6e11195df3046653f79a72c95e63"><code>(reals y | reals alpha, reals tau, reals beta, reals delta) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wiener_lupdf" href="#wiener_lupdf" class="anchored unlink"><strong>wiener_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-27fbddb7c175dbc42b3a77639443abb89b1c93ba"><code>(real y | real alpha, real tau, real beta, real delta, real var_delta) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-330634d3f68975f72410dd323545f5801a409119"><code>(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/positive_lower-bounded_distributions.html#index-entry-9511656e972acbc5b9054dabc90abaa4214250d3"><code>(reals y | reals alpha, reals tau, reals beta, reals delta) : real</code></a> <span class="detail">(positive_lower-bounded_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart" href="#wishart" class="anchored unlink"><strong>wishart</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-5f58cfc02fd2621376aab535950c5b180adee3b0">distribution statement</a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart_cholesky_lpdf" href="#wishart_cholesky_lpdf" class="anchored unlink"><strong>wishart_cholesky_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-73fb720e23b1747c1083dedea45393a2c31d1a97"><code>(matrix L_W | real nu, matrix L_S) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart_cholesky_lupdf" href="#wishart_cholesky_lupdf" class="anchored unlink"><strong>wishart_cholesky_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-f87979d321f7d0bb07867f16b505ba45fd7a7376"><code>(matrix L_W | real nu, matrix L_S) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart_cholesky_rng" href="#wishart_cholesky_rng" class="anchored unlink"><strong>wishart_cholesky_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-311a69fbe7214ccc71cc4a5a9b6293967b467a38"><code>(real nu, matrix L_S) : matrix</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart_lpdf" href="#wishart_lpdf" class="anchored unlink"><strong>wishart_lpdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-9ebed125213e04734c320d50523b729c926de41e"><code>(matrix W | real nu, matrix Sigma) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart_lupdf" href="#wishart_lupdf" class="anchored unlink"><strong>wishart_lupdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-96892afabfd06300ee33a9c56e7aa11ef798d28c"><code>(matrix W | real nu, matrix Sigma) : real</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="wishart_rng" href="#wishart_rng" class="anchored unlink"><strong>wishart_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/covariance_matrix_distributions.html#index-entry-83370c265d9c1947eb72bc6d0aeab2570102b443"><code>(real nu, matrix Sigma) : matrix</code></a> <span class="detail">(covariance_matrix_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="y" class="level2">
+<h2 class="anchored" data-anchor-id="y">Y</h2>
+<p><a id="yule_simon" href="#yule_simon" class="anchored unlink"><strong>yule_simon</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-0a6380458dc845b883a913fb2227d90dee08322f">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_cdf" href="#yule_simon_cdf" class="anchored unlink"><strong>yule_simon_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-be440f4170e8f674bbbf23f2e94e8803e94cec24"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lccdf" href="#yule_simon_lccdf" class="anchored unlink"><strong>yule_simon_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-3def60eb9f4fdb6c157d57d9c12a79c1be38a69e"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lcdf" href="#yule_simon_lcdf" class="anchored unlink"><strong>yule_simon_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-17da2af212f09a989d004314b9af2fe44bbe25a8"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lpmf" href="#yule_simon_lpmf" class="anchored unlink"><strong>yule_simon_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-0fa5c053a53c4eb74a8936030170ca24edf5997d"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lupmf" href="#yule_simon_lupmf" class="anchored unlink"><strong>yule_simon_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-ab47efb6b4bbb3f4bdaa849c799ee6e10bbe6036"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_rng" href="#yule_simon_rng" class="anchored unlink"><strong>yule_simon_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-fb10a835dac9bc498feff76a66b8bf9737013d89"><code>(reals alpha) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+</section>
+<section id="z" class="level2">
+<h2 class="anchored" data-anchor-id="z">Z</h2>
+<p><a id="zeros_array" href="#zeros_array" class="anchored unlink"><strong>zeros_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-9d79405416a0d21e4f87a6517060a784d4b5d2b7"><code>(int n) : array[] real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="zeros_int_array" href="#zeros_int_array" class="anchored unlink"><strong>zeros_int_array</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-2c7602e1d0977230b64541b764981fca1bf1ad21"><code>(int n) : array[] int</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="zeros_row_vector" href="#zeros_row_vector" class="anchored unlink"><strong>zeros_row_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-7b23944569e7b31c1e4f17cc78de7996783733f1"><code>(int n) : row_vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+<p><a id="zeros_vector" href="#zeros_vector" class="anchored unlink"><strong>zeros_vector</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-8b17549ebfa64ad40fc338a638fa00df789cc57d"><code>(int n) : vector</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/mathematical_functions.html" class="pagination-link" aria-label="Mathematical Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Mathematical Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/functions_index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/hidden_markov_models.html b/docs/2_39/functions-reference/hidden_markov_models.html
new file mode 100644
index 000000000..4b055c0fc
--- /dev/null
+++ b/docs/2_39/functions-reference/hidden_markov_models.html
@@ -0,0 +1,1178 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Hidden Markov Models</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/embedded_laplace.html" rel="next">
+<link href="../functions-reference/covariance_matrix_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Hidden Markov Models">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Hidden Markov Models</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#hidden-markov-models" id="toc-hidden-markov-models" class="nav-link active" data-scroll-target="#hidden-markov-models">Hidden Markov Models</a>
+  <ul class="collapse">
+  <li><a href="#hmm-stan-functions" id="toc-hmm-stan-functions" class="nav-link" data-scroll-target="#hmm-stan-functions">Stan functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/hidden_markov_models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Hidden Markov Models</a></li></ol></nav></header>
+
+
+
+
+<section id="hidden-markov-models" class="level1">
+<h1>Hidden Markov Models</h1>
+<p>An elementary first-order Hidden Markov model is a probabilistic model over <span class="math inline">\(N\)</span> observations, <span class="math inline">\(y_n\)</span>, and <span class="math inline">\(N\)</span> hidden states, <span class="math inline">\(x_n\)</span>, which can be fully defined by the conditional distributions <span class="math inline">\(p(y_n \mid x_n, \phi)\)</span> and <span class="math inline">\(p(x_n \mid x_{n - 1}, \phi)\)</span>. Here we make the dependency on additional model parameters, <span class="math inline">\(\phi\)</span>, explicit. When <span class="math inline">\(x\)</span> is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate <span class="math inline">\(x\)</span> out.</p>
+<p>When each state <span class="math inline">\(x\)</span> takes a value over a discrete and finite set, say <span class="math inline">\(\{1, 2, ..., K\}\)</span>, we can take advantage of the dependency structure to marginalize <span class="math inline">\(x\)</span> and compute <span class="math inline">\(p(y \mid \phi)\)</span>. We start by defining the conditional observational distribution, stored in a <span class="math inline">\(K \times N\)</span> matrix <span class="math inline">\(\omega\)</span> with <span class="math display">\[
+\omega_{kn} = p(y_n \mid x_n = k, \phi).
+\]</span> Next, we introduce the <span class="math inline">\(K \times K\)</span> transition matrix, <span class="math inline">\(\Gamma\)</span>, with <span class="math display">\[
+  \Gamma_{ij} = p(x_n = j \mid x_{n - 1} = i, \phi).
+\]</span> Each row defines a probability distribution and must therefore be a simplex (i.e.&nbsp;its components must add to 1). Currently, Stan only supports stationary transitions where a single transition matrix is used for all transitions. Finally we define the initial state <span class="math inline">\(K\)</span>-vector <span class="math inline">\(\rho\)</span>, with <span class="math display">\[
+  \rho_k = p(x_0 = k \mid \phi).
+\]</span></p>
+<p>The Stan functions that support this type of model are special in that the user does not explicitly pass <span class="math inline">\(y\)</span> and <span class="math inline">\(\phi\)</span> as arguments. Instead, the user passes <span class="math inline">\(\log \omega\)</span>, <span class="math inline">\(\Gamma\)</span>, and <span class="math inline">\(\rho\)</span>, which in turn depend on <span class="math inline">\(y\)</span> and <span class="math inline">\(\phi\)</span>.</p>
+<section id="hmm-stan-functions" class="level2">
+<h2 class="anchored" data-anchor-id="hmm-stan-functions">Stan functions</h2>
+<a id="index-entry-6569db76a2047e7ba00db34ba2b0b2e000bd4eaa" class="index-entry"></a> <!-- real; hmm_marginal; (matrix log_omega, matrix Gamma, vector rho); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hmm_marginal" class="unlink" title="Jump to index entry"><code>hmm_marginal</code></a></strong><code>(matrix log_omega, matrix Gamma, vector rho)</code><br> Returns the log probability density of <span class="math inline">\(y\)</span>, with <span class="math inline">\(x_n\)</span> integrated out at each iteration.</p>
+<small><i>Available since 2.24</i></small>
+<p>The arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.</p>
+<ul>
+<li><p><em><code>log_omega</code></em>: <span class="math inline">\(\log \omega_{kn} = \log p(y_n \mid x_n = k, \phi)\)</span>, log density of each output,</p></li>
+<li><p><em><code>Gamma</code></em>: <span class="math inline">\(\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \phi)\)</span>, the transition matrix,</p></li>
+<li><p><em><code>rho</code></em>: <span class="math inline">\(\rho_k = p(x_0 = k \mid \phi)\)</span>, the initial state probability.</p></li>
+</ul>
+<a id="index-entry-48084fc56672a13c812780b444bee6bc9b630cc2" class="index-entry"></a> <!-- array[] int; hmm_latent_rng; (matrix log_omega, matrix Gamma, vector rho); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#hmm_latent_rng" class="unlink" title="Jump to index entry"><code>hmm_latent_rng</code></a></strong><code>(matrix log_omega, matrix Gamma, vector rho)</code><br> Returns a length <span class="math inline">\(N\)</span> array of integers over <span class="math inline">\(\{1, ..., K\}\)</span>, sampled from the joint posterior distribution of the hidden states, <span class="math inline">\(p(x \mid \phi, y)\)</span>. May be only used in transformed data and generated quantities.</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-2ff49049acfb95386aefa57a16ea987ee4b7b68e" class="index-entry"></a> <!-- matrix; hmm_hidden_state_prob; (matrix log_omega, matrix Gamma, vector rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#hmm_hidden_state_prob" class="unlink" title="Jump to index entry"><code>hmm_hidden_state_prob</code></a></strong><code>(matrix log_omega, matrix Gamma, vector rho)</code><br> Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a <span class="math inline">\(K \times N\)</span> matrix. <!-- The marginal posterior probabilities of each hidden state value, --> <!-- $p(x_n = k \mid \omega, \Gamma, \rho)$ for all iterations  --> <!-- $n \in \{1, ..., N \}$, stored in a matrix. --> The <span class="math inline">\(n^\mathrm{th}\)</span> column is a simplex of probabilities for the <span class="math inline">\(n^\mathrm{th}\)</span> variable. Moreover, let <span class="math inline">\(A\)</span> be the output. Then <span class="math inline">\(A_{ij} = p(x_j = i \mid \phi, y)\)</span>. <!-- $A_{ij} = p(x_j = i \mid \omega, \Gamma, \rho)$. --> This function may only be used in transformed data and generated quantities.</p>
+<small><i>Available since 2.24</i></small>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/covariance_matrix_distributions.html" class="pagination-link" aria-label="Covariance Matrix Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Covariance Matrix Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/embedded_laplace.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <span class="nav-page-text">Embedded Laplace Approximation</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/hidden_markov_models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/higher-order_functions.html b/docs/2_39/functions-reference/higher-order_functions.html
new file mode 100644
index 000000000..1b73b613a
--- /dev/null
+++ b/docs/2_39/functions-reference/higher-order_functions.html
@@ -0,0 +1,1670 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Higher-Order Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/transform_functions.html" rel="next">
+<link href="../functions-reference/compound_arithmetic_and_assignment.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Higher-Order Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/higher-order_functions.html">Higher-Order Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#higher-order-functions" id="toc-higher-order-functions" class="nav-link active" data-scroll-target="#higher-order-functions">Higher-Order Functions</a>
+  <ul class="collapse">
+  <li><a href="#functions-algebraic-solver" id="toc-functions-algebraic-solver" class="nav-link" data-scroll-target="#functions-algebraic-solver">Algebraic equation solvers</a>
+  <ul class="collapse">
+  <li><a href="#specifying-an-algebraic-equation-as-a-function" id="toc-specifying-an-algebraic-equation-as-a-function" class="nav-link" data-scroll-target="#specifying-an-algebraic-equation-as-a-function">Specifying an algebraic equation as a function</a></li>
+  <li><a href="#call-to-the-algebraic-solver" id="toc-call-to-the-algebraic-solver" class="nav-link" data-scroll-target="#call-to-the-algebraic-solver">Call to the algebraic solver</a></li>
+  </ul></li>
+  <li><a href="#functions-ode-solver" id="toc-functions-ode-solver" class="nav-link" data-scroll-target="#functions-ode-solver">Ordinary differential equation (ODE) solvers</a>
+  <ul class="collapse">
+  <li><a href="#non-stiff-solver" id="toc-non-stiff-solver" class="nav-link" data-scroll-target="#non-stiff-solver">Non-stiff solver</a></li>
+  <li><a href="#stiff-solver" id="toc-stiff-solver" class="nav-link" data-scroll-target="#stiff-solver">Stiff solver</a></li>
+  <li><a href="#adjoint-solver" id="toc-adjoint-solver" class="nav-link" data-scroll-target="#adjoint-solver">Adjoint solver</a></li>
+  <li><a href="#ode-system-function" id="toc-ode-system-function" class="nav-link" data-scroll-target="#ode-system-function">ODE system function</a></li>
+  <li><a href="#forward-sensitivity-solver" id="toc-forward-sensitivity-solver" class="nav-link" data-scroll-target="#forward-sensitivity-solver">Arguments to the ODE solvers</a></li>
+  <li><a href="#adjoint-sensitivity-solver" id="toc-adjoint-sensitivity-solver" class="nav-link" data-scroll-target="#adjoint-sensitivity-solver">Arguments to the adjoint ODE solver</a></li>
+  </ul></li>
+  <li><a href="#functions-dae-solver" id="toc-functions-dae-solver" class="nav-link" data-scroll-target="#functions-dae-solver">Differential-Algebraic equation (DAE) solver</a>
+  <ul class="collapse">
+  <li><a href="#the-dae-solver" id="toc-the-dae-solver" class="nav-link" data-scroll-target="#the-dae-solver">The DAE solver</a></li>
+  <li><a href="#dae-system-function" id="toc-dae-system-function" class="nav-link" data-scroll-target="#dae-system-function">DAE system function</a></li>
+  <li><a href="#dae-solver" id="toc-dae-solver" class="nav-link" data-scroll-target="#dae-solver">Arguments to the DAE solver</a></li>
+  </ul></li>
+  <li><a href="#functions-1d-integrator" id="toc-functions-1d-integrator" class="nav-link" data-scroll-target="#functions-1d-integrator">1D integrator</a>
+  <ul class="collapse">
+  <li><a href="#specifying-an-integrand-as-a-function" id="toc-specifying-an-integrand-as-a-function" class="nav-link" data-scroll-target="#specifying-an-integrand-as-a-function">Specifying an integrand as a function</a></li>
+  <li><a href="#call-to-the-1d-integrator" id="toc-call-to-the-1d-integrator" class="nav-link" data-scroll-target="#call-to-the-1d-integrator">Call to the 1D integrator</a></li>
+  </ul></li>
+  <li><a href="#functions-reduce" id="toc-functions-reduce" class="nav-link" data-scroll-target="#functions-reduce">Reduce-sum function</a>
+  <ul class="collapse">
+  <li><a href="#specifying-the-reduce-sum-function" id="toc-specifying-the-reduce-sum-function" class="nav-link" data-scroll-target="#specifying-the-reduce-sum-function">Specifying the reduce-sum function</a></li>
+  <li><a href="#functions-partial-sum" id="toc-functions-partial-sum" class="nav-link" data-scroll-target="#functions-partial-sum">The partial sum function</a></li>
+  </ul></li>
+  <li><a href="#functions-map" id="toc-functions-map" class="nav-link" data-scroll-target="#functions-map">Map-rect function</a>
+  <ul class="collapse">
+  <li><a href="#specifying-the-mapped-function" id="toc-specifying-the-mapped-function" class="nav-link" data-scroll-target="#specifying-the-mapped-function">Specifying the mapped function</a></li>
+  <li><a href="#rectangular-map" id="toc-rectangular-map" class="nav-link" data-scroll-target="#rectangular-map">Rectangular map</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/higher-order_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/higher-order_functions.html">Higher-Order Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="higher-order-functions" class="level1">
+<h1>Higher-Order Functions</h1>
+<p>Stan provides a few higher-order functions that act on other functions. In all cases, the function arguments to the higher-order functions are defined as functions within the Stan language and passed by name to the higher-order functions.</p>
+<section id="functions-algebraic-solver" class="level2">
+<h2 class="anchored" data-anchor-id="functions-algebraic-solver">Algebraic equation solvers</h2>
+<p>Stan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.</p>
+<p>An algebraic solver is a higher-order function, i.e.&nbsp;it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section <a href="#functions-ode-solver">Ordinary Differential Equation (ODE) Solvers</a> and <a href="#functions-dae-solver">Differential Algebraic Equation (DAE) solver</a>). Ordinary Stan functions do not allow functions as arguments.</p>
+<section id="specifying-an-algebraic-equation-as-a-function" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-an-algebraic-equation-as-a-function">Specifying an algebraic equation as a function</h3>
+<p>An algebraic system is specified as an ordinary function in Stan within the function block. The function must return a <code>vector</code> and takes in, as its first argument, the unknowns <span class="math inline">\(y\)</span> we wish to solve for, also passed as a <code>vector</code>. This argument is followed by additional arguments as specified by the user; we call such arguments <em>variadic arguments</em> and denote them <code>...</code>. The signature of the algebraic system is then:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span> algebra_system (<span class="dt">vector</span> y, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword <code>data</code>. In the below example, the last variadic argument, <span class="math inline">\(x\)</span>, is restricted to being data:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> algebra_system (<span class="dt">vector</span> y, <span class="dt">vector</span> theta, <span class="kw">data</span> <span class="dt">vector</span> x)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Distinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.</p>
+</section>
+<section id="call-to-the-algebraic-solver" class="level3">
+<h3 class="anchored" data-anchor-id="call-to-the-algebraic-solver">Call to the algebraic solver</h3>
+<a id="index-entry-85f9ba73adda0bd2add83f06a31c5cd63ccb63fa" class="index-entry"></a> <!-- vector; solve_newton; (function algebra_system, vector y_guess, ...); -->
+<p> <code>vector</code> <strong><a href="../functions-reference/functions_index.html#solve_newton" class="unlink" title="Jump to index entry"><code>solve_newton</code></a></strong><code>(function algebra_system, vector y_guess, ...)</code><br> Solves the algebraic system, given an initial guess, using Newton’s method.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-0283f73334357f2f1b816fdd22197b93bdf02321" class="index-entry"></a> <!-- vector; solve_newton_tol; (function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...); -->
+<p> <code>vector</code> <strong><a href="../functions-reference/functions_index.html#solve_newton_tol" class="unlink" title="Jump to index entry"><code>solve_newton_tol</code></a></strong><code>(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...)</code><br> Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-cb1a93d1015ecd4883af1f04902daf27984a760d" class="index-entry"></a> <!-- vector; solve_powell; (function algebra_system, vector y_guess, ...); -->
+<p> <code>vector</code> <strong><a href="../functions-reference/functions_index.html#solve_powell" class="unlink" title="Jump to index entry"><code>solve_powell</code></a></strong><code>(function algebra_system, vector y_guess, ...)</code><br> Solves the algebraic system, given an initial guess, using Powell’s hybrid method.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-6d887fe07d5519082bcb6c789283ce3909e661d7" class="index-entry"></a> <!-- vector; solve_powell_tol; (function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...); -->
+<p> <code>vector</code> <strong><a href="../functions-reference/functions_index.html#solve_powell_tol" class="unlink" title="Jump to index entry"><code>solve_powell_tol</code></a></strong><code>(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...)</code><br> Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.</p>
+<small><i>Available since 2.31</i></small>
+<section id="arguments-to-the-algebraic-solver" class="level4">
+<h4 class="anchored" data-anchor-id="arguments-to-the-algebraic-solver">Arguments to the algebraic solver</h4>
+<p>The arguments to the algebraic solvers are as follows:</p>
+<ul>
+<li><p><em><code>algebra_system</code></em>: function literal referring to a function specifying the system of algebraic equations with signature <code>(vector, ...):vector</code>. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,</p></li>
+<li><p><em><code>y_guess</code></em>: initial guess for the solution, type <code>vector</code>,</p></li>
+<li><p><em><code>...</code></em>: variadic arguments.</p></li>
+</ul>
+<p>The algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:</p>
+<ul>
+<li><p><em><code>scaling_step</code></em>: <em>for the Newton solver only</em>, the scaled-step stopping tolerance, type <code>real</code>, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is <span class="math inline">\(10^{-3}\)</span>.</p></li>
+<li><p><em><code>rel_tol</code></em>: <em>for the Powell solver only</em>, the relative tolerance, type <code>real</code>, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is <span class="math inline">\(10^{-10}\)</span>.</p></li>
+<li><p><em><code>function_tol</code></em>: function tolerance for the algebraic solver, type <code>real</code>, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is <span class="math inline">\(10^{-6}\)</span>.</p></li>
+<li><p><em><code>max_num_steps</code></em>: maximum number of steps to take in the algebraic solver, type <code>int</code>, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is <span class="math inline">\(200\)</span>.</p></li>
+</ul>
+<p>The difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.</p>
+</section>
+<section id="return-value" class="level4">
+<h4 class="anchored" data-anchor-id="return-value">Return value</h4>
+<p>The return value for the algebraic solver is an object of type <code>vector</code>, with values which, when plugged in as <code>y</code> make the algebraic function go to 0 (approximately, within the specified function tolerance).</p>
+</section>
+<section id="sizes-and-parallel-arrays" class="level4">
+<h4 class="anchored" data-anchor-id="sizes-and-parallel-arrays">Sizes and parallel arrays</h4>
+<p>Certain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.</p>
+</section>
+<section id="algorithmic-details" class="level4">
+<h4 class="anchored" data-anchor-id="algorithmic-details">Algorithmic details</h4>
+<p>Stan offers two methods to solve algebraic equations. <code>solve_newton</code> and <code>solve_newton_tol</code> use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite <span class="citation" data-cites="Hindmarsh:2005">(<a href="#ref-Hindmarsh:2005" role="doc-biblioref">Hindmarsh et al. 2005</a>)</span>. For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.</p>
+<p><code>solve_powell</code> and <code>solve_powell_tol</code> are based on the Powell hybrid method <span class="citation" data-cites="Powell:1970">(<a href="#ref-Powell:1970" role="doc-biblioref">Powell 1970</a>)</span>, which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library <span class="citation" data-cites="Eigen:2013">(<a href="#ref-Eigen:2013" role="doc-biblioref">Guennebaud, Jacob, et al. 2010</a>)</span>. This solver is in turn based on the algorithm developed for the package MINPACK-1 <span class="citation" data-cites="minpack:1980">(<a href="#ref-minpack:1980" role="doc-biblioref">Jorge J. More 1980</a>)</span>.</p>
+<p>For both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see <span class="citation" data-cites="Gaebler:2021">(<a href="#ref-Gaebler:2021" role="doc-biblioref">Gaebler 2021</a>)</span> and <span class="citation" data-cites="Margossian:2022">(<a href="#ref-Margossian:2022" role="doc-biblioref">Margossian and Betancourt 2022</a>)</span>.</p>
+</section>
+</section>
+</section>
+<section id="functions-ode-solver" class="level2">
+<h2 class="anchored" data-anchor-id="functions-ode-solver">Ordinary differential equation (ODE) solvers</h2>
+<p>Stan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).</p>
+<p>Solving an initial value ODE means given a set of differential equations <span class="math inline">\(y'(t, \theta) = f(t, y, \theta)\)</span> and initial conditions <span class="math inline">\(y(t_0, \theta)\)</span>, solving for <span class="math inline">\(y\)</span> at a sequence of times <span class="math inline">\(t_0 &lt; t_1 &lt; t_2, \cdots &lt; t_n\)</span>. <span class="math inline">\(f(t, y, \theta)\)</span> is referred to here as the ODE system function.</p>
+<p><span class="math inline">\(f(t, y, \theta)\)</span> will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.</p>
+<p>To make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as <code>...</code>. The types of the arguments represented by <code>...</code> in the ODE solve function call must match the types of the arguments represented by <code>...</code> in the user-supplied system function.</p>
+<section id="non-stiff-solver" class="level3">
+<h3 class="anchored" data-anchor-id="non-stiff-solver">Non-stiff solver</h3>
+<a id="index-entry-4a595847f7a33bc641756482fe60187677e31fa1" class="index-entry"></a> <!-- array[] vector; ode_rk45; (function ode, vector initial_state, real initial_time, array[] real times, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_rk45" class="unlink" title="Jump to index entry"><code>ode_rk45</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, ...)</code><br> Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-1fd0eab9e49603406a50ad94820da2fb10dec2bc" class="index-entry"></a> <!-- array[] vector; ode_rk45_tol; (function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_rk45_tol" class="unlink" title="Jump to index entry"><code>ode_rk45_tol</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...)</code><br> Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-02a83fd8c68af0422903498868b0e1c9c2542659" class="index-entry"></a> <!-- array[] vector; ode_ckrk; (function ode, vector initial_state, real initial_time, array[] real times, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_ckrk" class="unlink" title="Jump to index entry"><code>ode_ckrk</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, ...)</code><br> Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.</p>
+<small><i>Available since 2.27</i></small>
+<a id="index-entry-25e82b14c8715c7f6c5fd13961ffa60016433854" class="index-entry"></a> <!-- array[] vector; ode_ckrk_tol; (function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_ckrk_tol" class="unlink" title="Jump to index entry"><code>ode_ckrk_tol</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...)</code><br> Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.</p>
+<small><i>Available since 2.27</i></small>
+<a id="index-entry-0cbdc3972873453988201c42896556c64240fd04" class="index-entry"></a> <!-- array[] vector; ode_adams; (function ode, vector initial_state, real initial_time, array[] real times, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_adams" class="unlink" title="Jump to index entry"><code>ode_adams</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, ...)</code><br> Solves the ODE system for the times provided using the Adams-Moulton method.</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-6272f555abc93dcbbf8700fbeafca26f4b6f686e" class="index-entry"></a> <!-- array[] vector; ode_adams_tol; (function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_adams_tol" class="unlink" title="Jump to index entry"><code>ode_adams_tol</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...)</code><br> Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.</p>
+<small><i>Available since 2.24</i></small>
+</section>
+<section id="stiff-solver" class="level3">
+<h3 class="anchored" data-anchor-id="stiff-solver">Stiff solver</h3>
+<a id="index-entry-302230625fcf25f300cf21fa4b0f34e9494d3a53" class="index-entry"></a> <!-- array[] vector; ode_bdf; (function ode, vector initial_state, real initial_time, array[] real times, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_bdf" class="unlink" title="Jump to index entry"><code>ode_bdf</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, ...)</code><br> Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-a1ea8ce1da0b3666b341644d924730abb2caf7af" class="index-entry"></a> <!-- array[] vector; ode_bdf_tol; (function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_bdf_tol" class="unlink" title="Jump to index entry"><code>ode_bdf_tol</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...)</code><br> Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.</p>
+<small><i>Available since 2.24</i></small>
+</section>
+<section id="adjoint-solver" class="level3">
+<h3 class="anchored" data-anchor-id="adjoint-solver">Adjoint solver</h3>
+<a id="index-entry-95a49804904581f863b2d682d1d17408cc0371d8" class="index-entry"></a> <!-- array[] vector; ode_adjoint_tol_ctl; (function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#ode_adjoint_tol_ctl" class="unlink" title="Jump to index entry"><code>ode_adjoint_tol_ctl</code></a></strong><code>(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)</code><br></p>
+<p>Solves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.</p>
+<small><i>Available since 2.27</i></small>
+</section>
+<section id="ode-system-function" class="level3">
+<h3 class="anchored" data-anchor-id="ode-system-function">ODE system function</h3>
+<p>The first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a <code>vector</code> return type, and the first two arguments must be a <code>real</code> and <code>vector</code> in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span> ode(<span class="dt">real</span> time, <span class="dt">vector</span> state, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.</p>
+<p>The arguments to this function are:</p>
+<ul>
+<li><p><em><code>time</code></em>, the time to evaluate the ODE system</p></li>
+<li><p><em><code>state</code></em>, the state of the ODE system at the time specified</p></li>
+<li><p><em><code>...</code></em>, sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the <code>...</code> arguments of the ODE solve function call.</p></li>
+</ul>
+</section>
+<section id="forward-sensitivity-solver" class="level3">
+<h3 class="anchored" data-anchor-id="forward-sensitivity-solver">Arguments to the ODE solvers</h3>
+<p>The arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see <a href="#adjoint-sensitivity-solver">Arguments to the adjoint ODE solver</a>.</p>
+<ul>
+<li><p><em><code>ode</code></em>: ODE system function,</p></li>
+<li><p><em><code>initial_state</code></em>: initial state, type <code>vector</code>,</p></li>
+<li><p><em><code>initial_time</code></em>: initial time, type <code>real</code>,</p></li>
+<li><p><em><code>times</code></em>: solution times, type <code>array[] real</code>,</p></li>
+<li><p><em><code>...</code></em>: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the <code>...</code> arguments of the ODE system function.</p></li>
+</ul>
+<p>For the versions of the ode solver functions ending in <code>_tol</code>, these three parameters must be provided after <code>times</code> and before the <code>...</code> arguments:</p>
+<ul>
+<li><p><code>data</code> <em><code>rel_tol</code></em>: relative tolerance for the ODE solver, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>abs_tol</code></em>: absolute tolerance for the ODE solver, type <code>real</code>, data only, and</p></li>
+<li><p><em><code>max_num_steps</code></em>: maximum number of steps to take between output times in the ODE solver, type <code>int</code>, data only.</p></li>
+</ul>
+<p>Because the tolerances are <code>data</code> arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.</p>
+</section>
+<section id="adjoint-sensitivity-solver" class="level3">
+<h3 class="anchored" data-anchor-id="adjoint-sensitivity-solver">Arguments to the adjoint ODE solver</h3>
+<p>The arguments to the adjoint ODE solver are different from those for the other functions (for those see <a href="#forward-sensitivity-solver">Arguments to the ODE solvers</a>).</p>
+<ul>
+<li><p><em><code>ode</code></em>: ODE system function,</p></li>
+<li><p><em><code>initial_state</code></em>: initial state, type <code>vector</code>,</p></li>
+<li><p><em><code>initial_time</code></em>: initial time, type <code>real</code>,</p></li>
+<li><p><em><code>times</code></em>: solution times, type <code>array[] real</code>,</p></li>
+<li><p><code>data</code> <em><code>rel_tol_forward</code></em>: Relative tolerance for forward solve, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>abs_tol_forward</code></em>: Absolute tolerance vector for each state for forward solve, type <code>vector</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>rel_tol_backward</code></em>: Relative tolerance for backward solve, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>abs_tol_backward</code></em>: Absolute tolerance vector for each state for backward solve, type <code>vector</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>rel_tol_quadrature</code></em>: Relative tolerance for backward quadrature, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>abs_tol_quadrature</code></em>: Absolute tolerance for backward quadrature, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>max_num_steps</code></em>: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type <code>int</code>, data only,</p></li>
+<li><p><em><code>num_steps_between_checkpoints</code></em>: number of steps between checkpointing forward solution, type <code>int</code>, data only,</p></li>
+<li><p><em><code>interpolation_polynomial</code></em>: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type <code>int</code>, data only,</p></li>
+<li><p><em><code>solver_forward</code></em>: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type <code>int</code>, data only,</p></li>
+<li><p><em><code>solver_backward</code></em>: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type <code>int</code>, data only.</p></li>
+<li><p><em><code>...</code></em>: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the <code>...</code> arguments of the ODE system function.</p></li>
+</ul>
+<p>Because the tolerances are <code>data</code> arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.</p>
+<section id="return-values" class="level4">
+<h4 class="anchored" data-anchor-id="return-values">Return values</h4>
+<p>The return value for the ODE solvers is an array of vectors (type <code>array[] vector</code>), one vector representing the state of the system at every time in specified in the <code>times</code> argument.</p>
+</section>
+<section id="array-and-vector-sizes" class="level4">
+<h4 class="anchored" data-anchor-id="array-and-vector-sizes">Array and vector sizes</h4>
+<p>The sizes must match, and in particular, the following groups are of the same size:</p>
+<ul>
+<li><p>state variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,</p></li>
+<li><p>number of solution times and number of vectors in the output.</p></li>
+</ul>
+</section>
+</section>
+</section>
+<section id="functions-dae-solver" class="level2">
+<h2 class="anchored" data-anchor-id="functions-dae-solver">Differential-Algebraic equation (DAE) solver</h2>
+<p>Stan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 <span class="citation" data-cites="serban_user:2021">(<a href="#ref-serban_user:2021" role="doc-biblioref">Serban et al. 2021</a>)</span>.</p>
+<p>Solving an initial value DAE means given a set of residual functions <span class="math inline">\(r(y'(t, \theta), y(t, \theta), t)\)</span> and initial conditions <span class="math inline">\((y(t_0, \theta), y'(t_0, \theta))\)</span>, solving for <span class="math inline">\(y\)</span> at a sequence of times <span class="math inline">\(t_0 &lt; t_1 \leq t_2, \cdots \leq t_n\)</span>. The residual function <span class="math inline">\(r(y', y, t, \theta)\)</span> will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.</p>
+<p>Similar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as <code>...</code>, and the types of these arguments, also represented by <code>...</code> in the DAE solver call, must match the types of the arguments represented by <code>...</code> in the user-supplied system function.</p>
+<section id="the-dae-solver" class="level3">
+<h3 class="anchored" data-anchor-id="the-dae-solver">The DAE solver</h3>
+<a id="index-entry-c47e85b0ae801c7df44337e2c2b7ffc5c84037bf" class="index-entry"></a> <!-- array[] vector; dae; (function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#dae" class="unlink" title="Jump to index entry"><code>dae</code></a></strong><code>(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...)</code><br> Solves the DAE system using the backward differentiation formula (BDF) method <span class="citation" data-cites="serban_user:2021">(<a href="#ref-serban_user:2021" role="doc-biblioref">Serban et al. 2021</a>)</span>.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-52f7e7ef9a4a58eb94353bf2cb98511deb94aac9" class="index-entry"></a> <!-- array[] vector; dae_tol; (function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...); -->
+<p></p>
+<p><code>array[] vector</code> <strong><a href="../functions-reference/functions_index.html#dae_tol" class="unlink" title="Jump to index entry"><code>dae_tol</code></a></strong><code>(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...)</code><br> Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.</p>
+<small><i>Available since 2.29</i></small>
+</section>
+<section id="dae-system-function" class="level3">
+<h3 class="anchored" data-anchor-id="dae-system-function">DAE system function</h3>
+<p>The first argument to the DAE solver is the DAE residual function. The DAE residual function must have a <code>vector</code> return type, and the first three arguments must be a <code>real</code>, <code>vector</code>, and <code>vector</code>, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> residual(<span class="dt">real</span> time, <span class="dt">vector</span> state, <span class="dt">vector</span> state_derivative, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.</p>
+<p>The arguments to this function are:</p>
+<ul>
+<li><p><em><code>time</code></em>, the time to evaluate the DAE system</p></li>
+<li><p><em><code>state</code></em>, the state of the DAE system at the time specified</p></li>
+<li><p><em><code>state_derivative</code></em>, the time derivatives of the state of the DAE system at the time specified</p></li>
+<li><p><em><code>...</code></em>, sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the <code>...</code> arguments of the DAE solve function call.</p></li>
+</ul>
+</section>
+<section id="dae-solver" class="level3">
+<h3 class="anchored" data-anchor-id="dae-solver">Arguments to the DAE solver</h3>
+<p>The arguments to the DAE solver are</p>
+<ul>
+<li><p><em><code>residual</code></em>: DAE residual function,</p></li>
+<li><p><em><code>initial_state</code></em>: initial state, type <code>vector</code>,</p></li>
+<li><p><em><code>initial_state_derivative</code></em>: time derivative of the initial state, type <code>vector</code>,</p></li>
+<li><p><em><code>initial_time</code></em>: initial time, type <code>data real</code>,</p></li>
+<li><p><em><code>times</code></em>: solution times, type <code>data array[] real</code>,</p></li>
+<li><p><em><code>...</code></em>: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the <code>...</code> arguments of the DAE residual function.</p></li>
+</ul>
+<p>For <code>dae_tol</code>, the following three parameters must be provided after <code>times</code> and before the <code>...</code> arguments:</p>
+<ul>
+<li><p><code>data</code> <em><code>rel_tol</code></em>: relative tolerance for the DAE solver, type <code>real</code>, data only,</p></li>
+<li><p><code>data</code> <em><code>abs_tol</code></em>: absolute tolerance for the DAE solver, type <code>real</code>, data only, and</p></li>
+<li><p><em><code>max_num_steps</code></em>: maximum number of steps to take between output times in the DAE solver, type <code>int</code>, data only.</p></li>
+</ul>
+<p>Because the tolerances are <code>data</code> arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.</p>
+<section id="consistency-of-the-initial-conditions" class="level4">
+<h4 class="anchored" data-anchor-id="consistency-of-the-initial-conditions">Consistency of the initial conditions</h4>
+<p>The user is responsible to ensure the residual function becomes zero at the initial time, <code>t0</code>, when the arguments <code>initial_state</code> and <code>initial_state_derivative</code> are introduced as <code>state</code> and <code>state_derivative</code>, respectively.</p>
+</section>
+<section id="return-values-1" class="level4">
+<h4 class="anchored" data-anchor-id="return-values-1">Return values</h4>
+<p>The return value for the DAE solvers is an array of vectors (type <code>array[] vector</code>), one vector representing the state of the system at every time specified in the <code>times</code> argument.</p>
+</section>
+<section id="array-and-vector-sizes-1" class="level4">
+<h4 class="anchored" data-anchor-id="array-and-vector-sizes-1">Array and vector sizes</h4>
+<p>The sizes must match, and in particular, the following groups are of the same size:</p>
+<ul>
+<li><p>state variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,</p></li>
+<li><p>number of solution times and number of vectors in the output.</p></li>
+</ul>
+</section>
+</section>
+</section>
+<section id="functions-1d-integrator" class="level2">
+<h2 class="anchored" data-anchor-id="functions-1d-integrator">1D integrator</h2>
+<p>Stan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.</p>
+<p>It operates similarly to the <a href="#functions-algebraic-solver">algebraic solver</a> and the <a href="#functions-ode-solver">ordinary differential equations solver</a> in that it allows as an argument a function.</p>
+<p>Like both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.</p>
+<section id="specifying-an-integrand-as-a-function" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-an-integrand-as-a-function">Specifying an integrand as a function</h3>
+<p>Performing a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> integrand(<span class="dt">real</span> x, <span class="dt">real</span> xc, <span class="dt">array</span>[] <span class="dt">real</span> theta,</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="dt">array</span>[] <span class="dt">int</span> x_i)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function should return the value of the integrand evaluated at the point x.</p>
+<p>The argument of this function are:</p>
+<ul>
+<li><p><em><code>x</code></em>, the independent variable being integrated over</p></li>
+<li><p><em><code>xc</code></em>, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section <a href="#precision-loss">Precision Loss</a>).</p></li>
+<li><p><em><code>theta</code></em>, parameter values used to evaluate the integral</p></li>
+<li><p><em><code>x_r</code></em>, data values used to evaluate the integral</p></li>
+<li><p><em><code>x_i</code></em>, integer data used to evaluate the integral</p></li>
+</ul>
+<p>Like algebraic solver and the differential equations solver, the 1D integrator separates parameter values, <code>theta</code>, from data values, <code>x_r</code>.</p>
+</section>
+<section id="call-to-the-1d-integrator" class="level3">
+<h3 class="anchored" data-anchor-id="call-to-the-1d-integrator">Call to the 1D integrator</h3>
+<a id="index-entry-39f0cbb60146855f8f98ba56502efcbfd04df2aa" class="index-entry"></a> <!-- real; integrate_1d; (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#integrate_1d" class="unlink" title="Jump to index entry"><code>integrate_1d</code></a></strong> <code>(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i)</code><br> Integrates the integrand from a to b.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-029333bb86d68fe853789eea2568f760cec8eff3" class="index-entry"></a> <!-- real; integrate_1d; (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#integrate_1d" class="unlink" title="Jump to index entry"><code>integrate_1d</code></a></strong> <code>(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance)</code><br> Integrates the integrand from a to b with the given relative tolerance.</p>
+<small><i>Available since 2.23</i></small>
+<section id="arguments-to-the-1d-integrator" class="level4">
+<h4 class="anchored" data-anchor-id="arguments-to-the-1d-integrator">Arguments to the 1D integrator</h4>
+<p>The arguments to the 1D integrator are as follows:</p>
+<ul>
+<li><em><code>integrand</code></em>: function literal referring to a function specifying the integrand with signature <code>(real, real, array[] real, array[] real, array[] int):real</code> The arguments represent
+<ul>
+<li><ol type="1">
+<li>where integrand is evaluated,</li>
+</ol></li>
+<li><ol start="2" type="1">
+<li>distance from evaluation point to integration limit for definite integrals,</li>
+</ol></li>
+<li><ol start="3" type="1">
+<li>parameters,</li>
+</ol></li>
+<li><ol start="4" type="1">
+<li>real data</li>
+</ol></li>
+<li><ol start="5" type="1">
+<li>integer data, and the return value is the integrand evaluated at the given point,</li>
+</ol></li>
+</ul></li>
+<li><em><code>a</code></em>: left limit of integration, may be negative infinity, type <code>real</code>,</li>
+<li><em><code>b</code></em>: right limit of integration, may be positive infinity, type <code>real</code>,</li>
+<li><em><code>theta</code></em>: parameters only, type <code>array[] real</code>,</li>
+<li><em><code>x_r</code></em>: real data only, type <code>array[] real</code>,</li>
+<li><em><code>x_i</code></em>: integer data only, type <code>array[] int</code>.</li>
+</ul>
+<p>A <code>relative_tolerance</code> argument can optionally be provided for more control over the algorithm:</p>
+<ul>
+<li><em><code>relative_tolerance</code></em>: relative tolerance for the 1d integrator, type <code>real</code>, data only.</li>
+</ul>
+</section>
+<section id="return-value-1" class="level4">
+<h4 class="anchored" data-anchor-id="return-value-1">Return value</h4>
+<p>The return value for the 1D integrator is a <code>real</code>, the value of the integral.</p>
+</section>
+<section id="zero-crossing" class="level4">
+<h4 class="anchored" data-anchor-id="zero-crossing">Zero-crossing integrals</h4>
+<p>For numeric stability, integrals on the (possibly infinite) interval <span class="math inline">\((a, b)\)</span> that cross zero are split into two integrals, one from <span class="math inline">\((a, 0)\)</span> and one from <span class="math inline">\((0, b)\)</span>. Each integral is separately integrated to the given <code>relative_tolerance</code>.</p>
+</section>
+<section id="precision-loss" class="level4">
+<h4 class="anchored" data-anchor-id="precision-loss">Precision loss near limits of integration in definite integrals</h4>
+<p>When integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, <code>1.0 - x</code>). <code>xc</code> (as passed to the integrand) is a high-precision version of the distance between <code>x</code> and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, <code>xc</code> will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, <code>xc</code> is set to <code>NaN</code>.</p>
+</section>
+<section id="algorithmic-details-1" class="level4">
+<h4 class="anchored" data-anchor-id="algorithmic-details-1">Algorithmic details</h4>
+<p>Internally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in <span class="citation" data-cites="Takahasi:1974">(<a href="#ref-Takahasi:1974" role="doc-biblioref">Takahasi and Mori 1974</a>)</span>, <span class="citation" data-cites="Mori:1978">(<a href="#ref-Mori:1978" role="doc-biblioref">Mori 1978</a>)</span>, <span class="citation" data-cites="Bailey:2005">(<a href="#ref-Bailey:2005" role="doc-biblioref">Bailey, Jeyabalan, and Li 2005</a>)</span>, and <span class="citation" data-cites="Tanaka:2009">(<a href="#ref-Tanaka:2009" role="doc-biblioref">Tanaka et al. 2009</a>)</span>.</p>
+<p>The gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.</p>
+</section>
+</section>
+</section>
+<section id="functions-reduce" class="level2">
+<h2 class="anchored" data-anchor-id="functions-reduce">Reduce-sum function</h2>
+<p>Stan provides a higher-order reduce function for summation. A function which returns a scalar <code>g: U -&gt; real</code> is mapped to every element of a list of type <code>array[] U</code>, <code>{ x1, x2, ... }</code> and all the results are accumulated,</p>
+<p><code>g(x1) + g(x2) + ...</code></p>
+<p>For efficiency reasons the reduce function doesn’t work with the element-wise evaluated function <code>g</code> itself, but instead works through evaluating partial sums, <code>f: array[] U -&gt; real</code>, where:</p>
+<pre><code>f({ x1 }) = g(x1)
+f({ x1, x2 }) = g(x1) + g(x2)
+f({ x1, x2, ... }) = g(x1) + g(x2) + ...</code></pre>
+<p>Mathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:</p>
+<ul>
+<li><code>reduce_sum</code>: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.</li>
+<li><code>reduce_sum_static</code>: Compute the same sum as <code>reduce_sum</code>, but partition the input in the same way for given data set (in <code>reduce_sum</code> this partitioning might change depending on computer load). This should result in stable numerical evaluations.</li>
+</ul>
+<section id="specifying-the-reduce-sum-function" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-the-reduce-sum-function">Specifying the reduce-sum function</h3>
+<p>The higher-order reduce function takes a partial sum function <code>f</code>, an array argument <code>x</code> (with one array element for each term in the sum), a recommended <code>grainsize</code>, and a set of shared arguments. This representation allows parallelization of the resultant sum.</p>
+<a id="index-entry-a5981e88d5cfd9835be7c6fbd0f9b671cce9bad9" class="index-entry"></a> <!-- real; reduce_sum; (F f, array[] T x, int grainsize, T1 s1, T2 s2, ...); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#reduce_sum" class="unlink" title="Jump to index entry"><code>reduce_sum</code></a></strong><code>(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)</code><br></p>
+<a id="index-entry-ef3aada570b019e2390174fe86359b5af6427541" class="index-entry"></a> <!-- real; reduce_sum_static; (F f, array[] T x, int grainsize, T1 s1, T2 s2, ...); -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#reduce_sum_static" class="unlink" title="Jump to index entry"><code>reduce_sum_static</code></a></strong><code>(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)</code><br></p>
+<p>Returns the equivalent of <code>f(x, 1, size(x), s1, s2, ...)</code>, but computes the result in parallel by breaking the array <code>x</code> into independent partial sums. <code>s1, s2, ...</code> are shared between all terms in the sum.</p>
+<small><i>Available since 2.23</i></small>
+<ul>
+<li><em><code>f</code></em>: function literal referring to a function specifying the partial sum operation. Refer to the <a href="#functions-partial-sum">partial sum function</a>.</li>
+<li><em><code>x</code></em>: array of <code>T</code>, one for each term of the reduction, <code>T</code> can be any type,</li>
+<li><em><code>grainsize</code></em>: For <code>reduce_sum</code>, <code>grainsize</code> is the recommended size of the partial sum (<code>grainsize = 1</code> means pick totally automatically). For <code>reduce_sum_static</code>, <code>grainsize</code> determines the maximum size of the partial sums, type <code>int</code>,</li>
+<li><em><code>s1</code></em>: first (optional) shared argument, type <code>T1</code>, where <code>T1</code> can be any type</li>
+<li><em><code>s2</code></em>: second (optional) shared argument, type <code>T2</code>, where <code>T2</code> can be any type,</li>
+<li><em><code>...</code></em>: remainder of shared arguments, each of which can be any type.</li>
+</ul>
+</section>
+<section id="functions-partial-sum" class="level3">
+<h3 class="anchored" data-anchor-id="functions-partial-sum">The partial sum function</h3>
+<p>The partial sum function must have the following signature where the type <code>T</code>, and the types of all the shared arguments (<code>T1</code>, <code>T2</code>, …) match those of the original <code>reduce_sum</code> (<code>reduce_sum_static</code>) call.</p>
+<pre><code>(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real</code></pre>
+<p>The partial sum function returns the sum of the <code>start</code> to <code>end</code> terms (inclusive) of the overall calculations. The arguments to the partial sum function are:</p>
+<ul>
+<li><p><em><code>x_subset</code></em>, the subset of <code>x</code> a given partial sum is responsible for computing, type <code>array[] T</code>, where <code>T</code> matches the type of <code>x</code> in <code>reduce_sum</code> (<code>reduce_sum_static</code>)</p></li>
+<li><p><em><code>start</code></em>, the index of the first term of the partial sum, type <code>int</code></p></li>
+<li><p><em><code>end</code></em>, the index of the last term of the partial sum (inclusive), type <code>int</code></p></li>
+<li><p><em><code>s1</code></em>, first shared argument, type <code>T1</code>, matching type of <code>s1</code> in <code>reduce_sum</code> (<code>reduce_sum_static</code>)</p></li>
+<li><p><em><code>s2</code></em>, second shared argument, type <code>T2</code>, matching type of <code>s2</code> in <code>reduce_sum</code> (<code>reduce_sum_static</code>)</p></li>
+<li><p><em><code>...</code></em>, remainder of shared arguments, with types matching those in <code>reduce_sum</code> (<code>reduce_sum_static</code>)</p></li>
+</ul>
+</section>
+</section>
+<section id="functions-map" class="level2">
+<h2 class="anchored" data-anchor-id="functions-map">Map-rect function</h2>
+<p>Stan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.</p>
+<section id="specifying-the-mapped-function" class="level3">
+<h3 class="anchored" data-anchor-id="specifying-the-mapped-function">Specifying the mapped function</h3>
+<p>The function being mapped must have a signature identical to that of the function <code>f</code> in the following declaration.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span> f(<span class="dt">vector</span> phi, <span class="dt">vector</span> theta,</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>          <span class="kw">data</span> <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="kw">data</span> <span class="dt">array</span>[] <span class="dt">int</span> x_i);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:</p>
+<ul>
+<li><p><em><code>phi</code></em>, the sequence of parameters shared across shards</p></li>
+<li><p><em><code>theta</code></em>, the sequence of parameters specific to this shard</p></li>
+<li><p><em><code>x_r</code></em>, sequence of real-valued data</p></li>
+<li><p><em><code>x_i</code></em>, sequence of integer data</p></li>
+</ul>
+<p>All input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.</p>
+</section>
+<section id="rectangular-map" class="level3">
+<h3 class="anchored" data-anchor-id="rectangular-map">Rectangular map</h3>
+<p>The rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.</p>
+<a id="index-entry-6286c3d1b323b131fcba8e6c7123559540554abb" class="index-entry"></a> <!-- vector; map_rect; (F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#map_rect" class="unlink" title="Jump to index entry"><code>map_rect</code></a></strong><code>(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i)</code><br> Return the concatenation of the results of applying the function f, of type <code>(vector, vector, array[] real, array[] int):vector</code> elementwise, i.e., <code>f(phi, theta[n], x_r[n], x_i[n])</code> for each <code>n</code> in <code>1:N</code>, where <code>N</code> is the size of the parallel arrays of job-specific/local parameters <code>theta</code>, real data <code>x_r</code>, and integer data <code>x_r</code>. The shared/global parameters <code>phi</code> are passed to each invocation of <code>f</code>.</p>
+<small><i>Available since 2.18</i></small>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Bailey:2005" class="csl-entry" role="listitem">
+Bailey, David H., Karthik Jeyabalan, and Xiaoye S. Li. 2005. <span>“A Comparison of Three High-Precision Quadrature Schemes.”</span> <em>Experiment. Math.</em> 14 (3): 317–29. <a href="https://projecteuclid.org:443/euclid.em/1128371757">https://projecteuclid.org:443/euclid.em/1128371757</a>.
+</div>
+<div id="ref-Gaebler:2021" class="csl-entry" role="listitem">
+Gaebler, Johann D. 2021. <span>“Autodiff for Implicit Functions in Stan.”</span> <a href="https://www.jgaeb.com/2021/09/13/implicit-autodiff.html#fn:7">https://www.jgaeb.com/2021/09/13/implicit-autodiff.html#fn:7</a>.
+</div>
+<div id="ref-Eigen:2013" class="csl-entry" role="listitem">
+Guennebaud, Gaël, Benoît Jacob, et al. 2010. <span>“Eigen V3.”</span> http://eigen.tuxfamily.org.
+</div>
+<div id="ref-Hindmarsh:2005" class="csl-entry" role="listitem">
+Hindmarsh, Alan C, Peter N Brown, Keith E Grant, Steven L Lee, Radu Serban, Dan E Shumaker, and Carol S Woodward. 2005. <span>“<span>SUNDIALS</span>: Suite of Nonlinear and Differential/Algebraic Equation Solvers.”</span> <em>ACM Transactions on Mathematical Software (TOMS)</em> 31 (3): 363–96.
+</div>
+<div id="ref-minpack:1980" class="csl-entry" role="listitem">
+Jorge J. More, Kenneth E. Hillstrom, Burton S. Garbow. 1980. <em>User Guide for MINPACK-1</em>. 9700 South Cass Avenue, Argonne, Illinois 60439: Argonne National Laboratory.
+</div>
+<div id="ref-Margossian:2022" class="csl-entry" role="listitem">
+Margossian, Charles C, and Michael Betancourt. 2022. <span>“Efficient Automatic Differentiation of Implicit Functions.”</span> <em>Preprint. arXiv:2112.14217.</em>
+</div>
+<div id="ref-Mori:1978" class="csl-entry" role="listitem">
+Mori, Masatake. 1978. <span>“An IMT-Type Double Exponential Formula for Numerical Integration.”</span> <em>Publications of the Research Institute for Mathematical Sciences</em> 14 (3): 713–29. <a href="https://doi.org/10.2977/prims/1195188835">https://doi.org/10.2977/prims/1195188835</a>.
+</div>
+<div id="ref-Powell:1970" class="csl-entry" role="listitem">
+Powell, Michael J. D. 1970. <span>“A Hybrid Method for Nonlinear Equations.”</span> In <em>Numerical Methods for Nonlinear Algebraic Equations</em>, edited by P. Rabinowitz. Gordon; Breach.
+</div>
+<div id="ref-serban_user:2021" class="csl-entry" role="listitem">
+Serban, Radu, Cosmin Petra, Alan C. Hindmarsh, Cody J. Balos, David J. Gardner, Daniel R. Reynolds, and Carol S. Woodward. 2021. <span>“User <span>Documentation</span> for <span>IDAS</span> V5.0.0.”</span> Lawrence Livermore National Laboratory.
+</div>
+<div id="ref-Takahasi:1974" class="csl-entry" role="listitem">
+Takahasi, Hidetosi, and Masatake Mori. 1974. <span>“Double Exponential Formulas for Numerical Integration.”</span> <em>Publications of the Research Institute for Mathematical Sciences</em> 9 (3): 721–41. <a href="https://doi.org/10.2977/prims/1195192451">https://doi.org/10.2977/prims/1195192451</a>.
+</div>
+<div id="ref-Tanaka:2009" class="csl-entry" role="listitem">
+Tanaka, Ken’ichiro, Masaaki Sugihara, Kazuo Murota, and Masatake Mori. 2009. <span>“Function Classes for Double Exponential Integration Formulas.”</span> <em>Numerische Mathematik</em> 111 (4): 631–55. <a href="https://doi.org/10.1007/s00211-008-0195-1">https://doi.org/10.1007/s00211-008-0195-1</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="pagination-link" aria-label="Compound Arithmetic and Assignment">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Compound Arithmetic and Assignment</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/transform_functions.html" class="pagination-link" aria-label="Variable Transformation Functions">
+        <span class="nav-page-text">Variable Transformation Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/higher-order_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/img/logo_tm.png b/docs/2_39/functions-reference/img/logo_tm.png
new file mode 100644
index 0000000000000000000000000000000000000000..48c9769c7de2c7c8585b52a5fdee686aaa47c8b5
GIT binary patch
literal 21065
zcmZtt1yr0r@F)zki%W59ad$6Xq`2$iwz#{yJH@TI7I$|RDJ|~qQrz7?e*b&ld%yRb
zclYdhCdo)9n<tsfOd^#OBtId2L4tyU`XntSrUC^8(ERU*0QXUYYqKl-@c=lhNQyvJ
zj}sn#6qHOgrOo8!q3Azg1SlW?6AJo2l#ee|xFr<K|G-dC)E`eMDCj)E|830!{y(h%
z%{=J;3;%~Wr3x+c!9>dvr0Jq5FUMzWZ_8|GVsB*1>|yKh9|F`j555o3*3`w2)Wg=s
z&Y90cfc$?j_&(tOs#(ZM{|CjzT7X<rUWruH-pQ1di<y;~m0S>sl$7+FlZhFhikQUz
zCjTf2kXyL8IPkHsxVyVEyK^wxJDIbv@$&Mru(GqTvon2QFgbhLxfpsd**R1EFDCyF
zA2CyBV<$@o7fX9P(*O7x8ri$L2#}NiC+Pot{#QR;EY1F}Bs=H-t=5NvEdRBzuraf;
z{2$&Qq~HFl<x_OBH2skLAAdo%Z~ue&|El}na=x+rC;b1fGyhA||Ec{@RS@YL%l{cR
zL8Q<6Tkud&!cfv;-$5RL(`<MTLh<^C3iBkVRK~x&`-D<(jIzH0ay&dqGyJU7;XB0c
z=l+m$k8iE;lK-IkG0p;ZP=4bF7n&?B?t%%-Wm8SH5>uQPSKp5~flJFNT%#xBC+hwi
zejWTXGkUMbOUoH&M^;cWVgU%~0CHiN2q+9v;QvP<4;%0c1OYK)T!i&n6SRDCks9A3
zLfu!YAY$*`l5297u1ONMcf2q`MRD1T;o>h!KilKYUbSSem2e#W-KMr5GOmf`!kP^#
z*{PJMX-Rc;JBaIQ>v((U-jl6dt0|rkBh`eFg1|+)gv?z_e1}}%M!WnP{FlCJ7tVHy
zioa?Y+NeM<;I>GLwDHN7E8a$750ZY<n^c(2sW36-NIh>geF>K|Fnr6$V>I2MmneEa
zDB8pG9&nYY^eI6i0NbYMwl(HdPfs)5r4x(Urh(TfV~n+U4(T<MF@6>dhefX{zot`a
z;i>&qGf{35+kP9T4(<+GdmvQ8Y|&w&;M&S`Ktfy`9U%XTKwfG9T9{3{x}LCVs1+Z}
zei@}kCed4fu$M1Gb`gSLzsTTk!zbcGH_vL8%73jdFv0RkAU@W0KJN512vcD$*CF`O
z)2c8xh<@)lG*nrQX?*aM%g?j$XSVYCHetkhdmd8$)o@}VWJ(CtO4SIg5raA}sdZ{E
zP;2r%czBTdGDz-2pZLIvy7n|=<QeUitdTFHK7R82nM(7cNu#KoGUfU|Zysd7BqNM1
zT~6oB;_ve%A*!#Yq-oUXM@rX&(Me$mP=@I%cP?EC_PN{-Qq`O4+KR`<T`Gd@V0k<K
zk|KtV=&&0Xvv3$MW!^InKjN|e-m45ARuKX`sg?D_nIv31zPx~(X<o|LQDkK@!7Abm
z_$GAeNS9H<L5PH+n4Z$_Jcm1)g?oR!ahd4vk3@!xEP+K}&NbgIWR_$rJ%$%^H3nU^
z)j#kFDa-sd;Ox0VSAO~wHMEEwYZU6Wel(_GGE)M^hws(l|0&9#x3)rJuPAs{E2y;e
zyM}ibN_1)?`AO6A0FcaLc#a|y5O%_2*`+Y?zCn?;PmG+$(+)C)IT$LbN4=EHauq+8
z^G3_?Ggt>MkXUU7OO-u|vfV5ojJU}m5t1YVYtTL~U;IhDYy>jhbYZ5$6oH|4@TM!?
z?=DjhX+~eRe4Bjq^=J-6=~2{}834K$n}i9&Zq)dvBWdqOZ$tIC3Zbw+py(F!@>5F+
z{B2h8bxfyNhS=r^P`QP}lKvI%0V=jt91!w7UB2k>%GvuIEJu&K4jL%cwNh$xe+)_o
zMCp164~Z*BZ#~c7$)L*FEa~S~KV`8L1-d=V{(#>}reEF3!aYmIvsO`)LG&pVY{iwn
z_f-<pLh@I>cXNLxl%}Mn@~d^ESPvs38@Vr_`?#IBm#uow3D^?#@<Eah5wH2bE;tz;
zW%TC4W;cjRckT;}<SDw!Ssgy%Z3q!BOVacc(6P-M6xvsb_sb*7PDw+`OLfFu%Db4x
z_wVK|s8XuHpacubsA0q+W8sX*&(p&wg<EI*7Fp_?UE)aUuzHt$!zqqU?Xzj_a~BK*
zd~DrcsmM2J0>y3w({m7yv5Zc+ztsi=O*}*@H;+g`hQp@sgXXJ5-_>+x4SK4#Wv!Qa
zne=Xn$n>@Xg@FnY^26ZVB7!1<vmVBhzoJ}xm8<}jS|n&UjhB&)Zktj*!R;DvtB3jw
zw<yHTSPY`1aA7BG9%}RYl*Ig<)V*jt_dlxiDcejRMC*x(WF-4puX}|-G;ZqsdNj)x
zF=}u{ppZ0}0sPZhMvt?5ynR6_<*v>UYEdqgT6W=^UlApObgf>Boc?vKh?re}6+I&Y
zFku{~M4lh~+Ts0Mc=oBNhD+ELsf3a4K*aU_-bXvO#<H2V=cm<ymyJg<Zk?b!5}1KF
zKOa8+R3Zw!+vgEyi(>*Id@pZAI!$}&$Mo~Prj6eq&CK<uBbbiHnc>KWxEI=vE0_Uh
z6fBM<4d|VsXky(g)e@f{2^!F^7dQvkL-W|*Ui5JMOEaizvGZibhvi8H!j{a;8YKO_
zFIR>B>bVuANg}EL=HDFrK|n)TI#acz_Nm{5F*S-D_??WEIa>|whok<}<X)q3wQjF8
zWLWHK3Jp1DRT*+_-J}#?{WD&Ljf@mL3!R%oR~4&ZuE9L<w=P^bq_#W}+a_S`@uXdT
z|GX+iov1Q=7Am@6Br?DX0UQ(DeEIs|+SB8){$&oK`1^M$$kUP~!+q7)R6=P_^AY9N
ztmf7`EONbHRzM?68x?9B!}hS~9PRod)w#ZM5RXc&pUWtgHTT52>X0DsG6JHXu0q<!
z`ijnK2tsk$4+^E&-2D90L3uiFl_CROS6n19=(dhyK=Zg{@1@&~;YZU`L~OlvfUp^1
zZ?YeX0TljMxB~y!_^CepXL+ByRh%ytVOC<4u*aqv-#YBp0=6Y~N)E_aXGdgd2{4od
z0EWe}^895WbJHEv7T+oklZ%`gvrZ;UR$ILt`TY-3LDL2cIIB<MpF5RizGKvM_KItP
z6!J!7aFEnA%OyylO$w%I`}bRpcN(09)I3#1nThEZ_1lHzOnh5diZ3dVcrsJl?toDW
zWf{WQXWZYkzFnEKr3c2ney*#JINpqTw--_G_vdIHAd`(W)I^}uM9Xd`_kS*}VPMQ9
zp}uE83puFGKy+zQDJf9lq-~O?HrDxR(!nn1?Dr$XGeqWNuG;;H{sHHn%)3!}4D)hc
zA_g<7*c;-0q<J%tlQOnh34p@SVYP^a=0{cV6W{J#tqE2ylb|Pu_12broj7_)Ye~H<
zR)^2${q)IK?eoWI7@+`opodh3N*FAPy%y~1!Y->rPF6L#Mu<PRNkzXbPS{uJq+N(s
z95%2pJcRr?@UkkNM?he3bqcFhQH(8tn~}NAzzO&NeSNU9*xO)f2o#}!jYiMWYw<j4
zp;^3}Hj6iZ*f{P;V8;YZe7qjcn)OP*0+LiwBzDjP6w2yA6Gjh?Vg7WVUD5_kcR`Pi
zj+%Oymq>S!w>o}xg|lsmZqA=7p0yi(NV8_UWM>4*xPGCZE216aDi+zE7I%bpa<goP
zWk(>M`XFg~d~bWi|28jOQE$?kQi|Y@AoYNZtzLd1nH_2eGM+p%AVf}@9j0n!XJc^p
zO>Z39Hw#hJcv`#(7g~MM-cojJ$bR-9Go|@3nGc-uF=f<Rl?LBB2|0Tc)jZp>9sM+3
z0y@*w=(77n#Iu6*#@kFhS^P)L_fC95msBprdd2geoj~~@$s#dge7KjT)|1s|)|~HR
zKfhuWgWtQ&<X@`i7X?}YTbqA62FnH^<de6HLZ7N;%1V`njKt_Mo>;b%E?7%tUT`wm
z`f=$P$$zcn(&}cql`=l{=6m%jGs}nx>jNXH;b!Vvt?7=iL6zJ5RNOH5mC^2E1{id9
zx)_dloBCRxH*zM47_FeP6!OGifHae)#bJlLpv?d;hQNNV=)a9T0={n6_Dg+Rx{C~p
z`$Ba1FRq*?bhg`<Ta$*fVUrjU+&L&2+bZR+0F{EixBWsw8`KR4JDqqQhws5oDIwjv
zwcVmQdTByuBS;U;oj3)eaT%18Ujon6%MnBbFwzRG`lk#CiX5A*ODI488#2$<sQ47=
zo5pT~bFwkNuehu~-d%37x9mE1*Cliq1@+%dALM%nJz1m1TjZvZ%<_}5n1SuYb9p(_
z$ry1?I7sY&c6Vj@e+SfadPu@)nOs7n$G&lhSAukRgpl+~`~FM`{rUP@T`gtKkwwD4
z+ll3umN$*uuK*EjsdE^;Gmh0$vkizwrNao!fe8T;1s^||qoMd&^yExWCXgT!Cf>Zf
zwEB6@ZHZA<VxMqWbE7@|${?b(4kVT6M=4n$^I0L!dKCw~FF^yw4<`D2Z<E#eomyt>
zMiEP1yZa;{%<o04*HS6h7j<S=xRXaUvUY9adDwN>x4MlWoF7LlEt?p6#P8EKlVc-)
z{#YWzqZsCc4c0C^hs|mD;s`BwrMD*OX5rouBlMKTJ3$ycri|lH;e(&fgix#W+o;y*
z_5OtYOK(~ZU~c2|fA6YwGF?UcmrMbF1SN{08ub;Zw^L?Rk<8&(kqp>9CB$ukqW>lD
zNR%4A^hX?*iCbz0=`M{y<A}YIj>_8#9G}V`XmbA}VA0HrPw**MU~_l0$DyhKI}Aqb
zB<RiAe%2`fU*7(RJGtcKAu-t+|E=DOOZp^={<rcV<=jsmX&kI|vNtba!~g|KwSC#D
zN|3xk_1r(}`JmTxw5ttT+E9Wfq@3Ne%m9y<+z@CX=kR!y(@gcnf1ACJV~EgK>$674
zoMm8=$J)*O7nQg4{xb1hN@shR0Toc){Cnqjo?lVNul?WfUF~TPbnyUiT#?t}j<45O
z9jKOjdd^69WlaPuYYuYv0Nd+=6rE9i-ocaiRS1exI27f~73e%qT)i6i?_m}Zx5GLy
z6Q_<U(p{(`Qi%<-vyUtRw$&l-HDkr1+ml>YX$CiTT18@Twz-nauV~>illbxJX}hKj
z<yyrmX1S;4b!EpabxYSZn+zYkxR?Fy9e>m_I=0vs5a&`j8SYc`!>=b^n-@&$h`oz)
zuQ3k{!c!L?@!`1&K{ZqAI<&bU!+@xiKxq#5Y<gvhb|v#B$`ZjJ=x38a1bNL`|ChYY
z?5g4yD<&wa&;Hb;zf}g)k6sK>>bLy_6>=klH(?fQFrOxTnJ%1?nZDm5-PNISZ`770
zcUjWJ0Zh+fe#)FXRcW)n><-f}pD0q-f<_NvMir>DGn2EgPY-EkZ}jH=TL(jp1qJLj
zR!q!V-NF+y5D~0h$xs1|_}_KPO51Y#q`5%)rdiT6SoO|Fv&+nL^o!;TRUEzswtULd
ze=G|{n$`tqH5L|+R>@EXFx6$mNYh~)R(=MBoIE|oFge0B=-VHwoVxU1;_j%cRIj@k
zJ6uPaR+t+j^hQepVt)T+_@Ds6Op1)!mdU+B&-Z)-vEQiI&@(Mc3a&c6|H9l)dN>!8
z7Kg!fh>JjWL;#b0f-Wu9kiw=6=&!S}^O;|@wJOZ#gN=Tx7HwAJjsC*dW>sa~<XSL&
z7128v9R{t-ml0G+JC-qg9|r~RmbLOJX|9<d=qOp!vBQ|4dR8S{_szAVxn?**yqGTv
z0t8im?Jhs<MsYv<d=V&T?u)zP*_E+ePLF)2SaD^*@(qr=3{i=4m>quys4ED!>7(mN
z6;z`{4%8;={oKLD=6%L*m7u|ibT{>JWU`~1-`6?O<zz1_P70zL3V=cB<qhoD!Y2`a
zXiy;sW>ECzEkTOt{%Y|STgB_)2o)|MBNrmY@mR^~1k68&Y*wQuX=wZ~u0~|6*A$%u
zfQm<97Goby+Ig)U)>H;H_qVi%B#U%SyN5by6$W55zhQ#^YG!m^0r5<yPym7*#pI{N
z*!N{h(S%OOcL`0IPA$n&k6dc)h~e$t$z=wbW!GWst#IMcDg`2b$h0$3^qd8Vn|f<o
zlh={buU7-99AvU&P2j;_q2boW_tI@}HSa^kfK}e$!1O-_@>Rh@*;Yu*fSdfdgp(7T
zaZi0(Zh4flAR?yXx1AYkh5XcDhx15zU`=N4COIKy9%3aaR0;j8FxYM+l4&8I^c^LI
zmCRev=O{^Dm5VT?yQIn>`TR+1393cWoYW2dV>2R;5f+$P8X7bouR#>=5H*7kTf_^J
z?Umc4MPJSp&nVcrR@umd_%M{10znS=Ibw$)R0KEJnA}%FzDgGl5N+X(s6-VQGc*_v
zM+`cR;s=$sWwL;0KZ6yY0<a-1CF&D8<Z7-W$t^pg#+G%)Nt23iwFtV^ar&>6hD@@e
zDJDQ36YN0Qy~dvNL)QfH*UR@7@QN*Bg;X2;%;2lT&)-<rAE0QO%-mN1YQIk54|mv(
zPv4usB>$N*spN<J(ZQ+x!)sGJj^Xng*<-?D-34MW5dk{Ds@Bu{Qn1}K5eT`tQsF5e
zxnd<2?Et*46!4xcI?PEL;o?Y%9^tM{^iDJy1}6^@tZ(1^_3G*)#Y%vO=*sllA?SX*
zpj0Daa$#2DGXkrdtm<CsA$#(!8U{qUb|g*q`RKv9Q8ZA)ciZ*XF*R_ND`no;^Ha=+
zL8WU>;rND9eStK5VWEV5auvBRV6ZNeP`}&Rkp{kDzyL&Jez<f5v3N>O#kNcZ4fcj$
zquxc*YV-&iI==lX>rUD+l9zEp*NostiSPg>e>i|HGJCfkPM#cP!oPx6gPPI{HR7Fa
zpQ#TVw|H)FHTixdTcfeM1Ry*}fI%aX$rU~w1NW;@Zz^d$x2ApfW452<afR>QWdbp!
zqTxK;1felVGym%W0E;$uBK0?WsxewCB^oU4m*`$An`eGH4YXa~Z;Wq<=Rc)~=Ok^C
zvk|gHdSf|GahRY?nFxn779)mh9yZwef0URcUPc??o?N96sB~A&Ei}g8(#yK*v7QtK
zr4+N$haiOT0%+)`kK6*L%|O2%_~v7fa5fmU0XE~g1aYz>j}O@1?<F+`4%_>MSH3Ev
zL4&+i+}ey;VQ?k~3>ZCXHpe272<RUc!Vy0p7P$gMtGd7+P9YD4%R>CZgbGg7?*Yo{
ztv2lT#?f9?_?pQ3^-RQynr7l)=ZYZ{20qE6C)D{xkAOdn3@PeK6A4)m?Id;V`KW0M
zv1swKA##jN79guZ#q~ASAW0QZQ(B-#eIUq|ts!l>GAr3yaovxo&8E>HZAx5BI0+`C
zuW~@rbAjhDF4fWzIv9>a@cuoy-T&<{&wymvhSc?lh-G`Q=cXb+pE+@j@o8_^S3=g-
z)H^AZuAtM~QdTqPOSHFzKn{a9cPJHuRdO@J<_<66P~FtDSy<dX%|_hLuBxr%#Y%#O
z(H*0&<aX2w?ud5BiHrkEg@;e3cw*P;CWVIN50kF)LK3N{UW5q8X^I~-L(*~im^is3
zQ+zl{%20aq77N~B*GFg9hR|cl;=wl(eFkeO(Ixr93WLA){Xg3^{l7>rWXT8*xci!7
z6v`th4vK7m<npjc&IjeXJUUsdOqh|IU}&-?Go^0nc)yH}&9y&NjqF`Vr_4>i^gm~w
zFRM!ot5x$Ksw@N4nUg`WKp3wU8S&8k2B}v|J7XF#d7{0Cnkp#&yL~VBg+JwBhLxiC
z63)`Iywa`oMKgZ)(i1d%D~m6G^Von!(`DFIN`@&>OV1Ra);qpiR+c#wq^yOrkNg6g
zKEn!iwr>W#6#5i4zctLRyr#tgT0QG+)(Y3d5>3j<HihAR*dUZX($UJ1?LIP>*N!fs
zQAHcKyrNt6k}M<7O0lsCV?kAiAwv{pSi1q|WheqibI}rsPPYT$vG5AQ56Mw#3X)Uo
zOyWJ$g%4V^(`OY~z#8#Ps>a4PMolg3O2!_eI>x(eE8pxTxccOhF}noCBi>*D@wX(U
zVacyfvPat9{!YbM;YMspaq5<8I=k$L6?+o8g8Y5>*^kqw`?EJGEj8Emnzg8tLNdcm
zEVJvmJub=jo^&)A2jryG*#nwowKAdN^nx0jtz}B}$(=Od!M%n{y6m`ZoSv&U`g>ke
z<yuK?;mg*jvq7|;LM#sa7%9XaH|6ME3Q`|QYu0$_h5uMYUxYSnCYz8uG$t?s#_a*W
zfluB@nvl24!k|!{XwM+I)2IidKVCl`bT+Y}N}f~N%`TLGMyV+!CsQcRKz7bv?^79a
z03Iyy+CcSM`Kk*CQs88h5Z0gHp|Bhc{%;Do<k7<SrAp#IN^0CN8G-D-$dP}cRN1!l
zlGLSt_&kL<=-g{`(v(~V79;$euouO{P`ik@vAP|^g!RaSu|XlzFd_UP-q;_hU)cy*
zaLzZqEDf8Irfp_H-TFfwRzbYaU70MER~pp;jEf<tJN-qvuj2kku9p%4on%G2ip2%5
z_Ie6WB9Ta@^xMf%t;IYEL6M9qocFbB5H0^COWxtuQ3GCRmD<*S>C`PRRQ}7Z(GpPU
zz@r~CNJTZ@X(sO}{3gW<kf14*79jqC;ty5eSFWFDdTnl#XgE_tgf}#VU_4+%C~OV0
z87UWMQIQ|8TG9u_VtTFR#sNCOVZ8)fA0{M-Fpc(y=cR@-+T#p+kL@q~h&Kv44gdK2
z&VQYdB3*A4<sppA2$mNa*#WB(k|CM6E&eJG70Gv6rqEgv<AUs-{+7&_vJrb&ou%K*
zzt|}MhjMzVfZ)|H{XmP9=rUS(!Q)W6N5#G2!4F3k&oWEcl>gqlhq=dZu18OcXL|%m
za_TR*?Mn%s(PhaE^g}MMpQ~sD)q>mYsl`tu*~Sj`AB}m`%V&^H86jqQ?#p5Vr@JTZ
z3Ikk5^E)9PUOHF<!;TUELyPT$*yyaGRe8b4pFT2Ty*j`tz6DlD)B3;j;r+Cr!SqNx
zHyq*fH!S@4Ua(i{`@&3$0OIC6iBbr`E>ZP9DRoYW4mK0n>EFoM2Z_9TXxvA)!N045
znTlwgnL<a;SPl_k@GlVGBTjb(^IC_2vY~^_dzb;IDT-1tw4L&j0<Qqkho1#vo)d9S
zPZD6U_OE+Q$NOh@d-S<Xnn|c<WP3-*M)#hBeL-!aYq^Dm0*Ddb3NYNfF;O7RUwtA1
zVH=?%*i_``D8RbJYr)EWGlQcmEeMxOedw;h=6{rU+liVl4F$V_0qIkCQ7-<cfAN1k
zA44ry60^@3lB2@_<;_y}tvLlhVT@|&m&VC7>G0Xa7@<ALwCng0O?%&pz#X!!p-MbR
zURw0@ob4(U!C!R2P)JNAsf27`(_42($ZyWdBI=xvUh?U-DdL_iC7mps&{FMvvFOR=
zfFjp!i$eDazfd-yoqgR-ATj7O68LFh3C2V8(!c{?eE8%ub~37<7&=o%X<)7EY6-fS
z(yu(J9i&6r%sA68*+|Uq^pKixgGF}=g{nfXeEcUr-(ordJK%~{B_OFSbw9tpebL;H
zlQLxG8+B(q@LJt4a91hIx#6W6`C7{#Uabxtn4vD%kH_0d9NhK1=GyOH`W1I(VgRpG
z;QDP}3uT;O_PSL)F2Yfz?#ofQRX`UHnJW}c;Q&3|$GJckF-F=gM$hs3C3wAY;+QOm
zhwca!br);r6G+ihg-v^Y30fd~?zdyUl58gLOTe>j4yb%>NU>=OE-?`SMub#Qo2CyB
zDXmH{X#uHuZrJ&m@dx{j%w|CXjVRV#hIA;+AjNyNRV+)u+JhC9I~~p)W+XNv@P}Dz
z?@D{8dhOgP;T$xTKO#cz0&IU5al&OwMnH}8wx|pQO>MGVHgKCOctoX`rvzYJ4D5>#
z_C$^e*&CtqxqXy6PYLSUx$PVE{egP5jIv4nlOIO7a$Uu^yz)HY-YpsOt0u0mT3`@t
zXARv&hg45J-n+m!`B{y=^^0MmoS|Z`4g70R%n}yYCaspik6RfC^&oHc9#6veze2!1
zyDp0&uSyMg37L(6iDwc&b);N>cl`1Z1_$H^)fKar^sR>${yeMm`*==ITod_pK;nAE
z%HC|y0RaYrD6k)^XrxZ-igppT!#}~KFTTP+J|hv4m1n9(*0iIvZVu%mKnb*NCJBOu
zj|&>xUK1`KTLqXJ*oKyqbcddlnTzx|stQjh6r@xcmy*h3Iyi)<Cq1be^m*UPLTLW{
z^~Dj}7r<&Y0a;Wo47a0tG(e>sH3vKFIw0~4LRIAcU0#AK&fOOs9G@Vw#?)@)@ez)i
zS4?;L`491{tISTPrC4sy%jJm1mxe^u(bh4CT+gmD*6AygAwXwiL?AY$7kn&dUvw^E
zND#~AR8K+uP+^bmi@akbD|mS+UeF83ifDd8ixG%G9&jn<A?sqk>fXY1c2KqCt6hty
zAZ2i4`jbNHD4$!o$@9!e7q_C(smg}6|JoX@cezhM$+Hs^ZzY>6uN;a`R)>S|(M2HH
z|DE?H*A3Ss<|j-{ABa{sBML+TKuP*LE+3<PLDUQ1^>?8p2sYeT(Du&)fyl(Z(I+TR
z(QmnEyK>IqIB$2ChMG;LLdm#9;@L;<JN~3j2;w?dhj68<ube1ZYy{I`J4sltMDRh$
z)pH1dGjO!hY%jJ=iC%`pCH%{=<JSF`xqE|j#n76^dP058E-bgxZz1o#M_Nf%c6I>;
zL2-FNrseM(6T!P=2V7%PGf%T^$y%x~%QgXZT3(92Z!ejnKAPv^lUnXD1GT;46sy^n
zubOHNIK&1`X!ABt#9yqfd{{^$G+e&ULm=`b!g4ZvDCvHvMW_T=1IuOf<mlgku0)lJ
zYl|I`5MG@N-R`b_g<$J$w-sLK&o5^-1~q(~@9p*sTYpnF`q4U2^O~%LZi^(k2hH!o
z134zb2Ts0!Nk|`{1g>K_Xv@*@`?*&SsR6sRYg>BMd8|n2G|nyEN9eXuACgX56E-%G
zWef+)?Hz2I7IVo#uJYDuuqOu3s0iZx7W0Hsf2KVe3Dl;j3`OKgsUYmIrnEPdk~K(N
zY*Tdz8Q$DZ0yu<!hv2sgf)Ee8F)U6S`>*^@1Mh_%;g-L!4fv;1<y+kQ-f&T(eF?-p
zwTY1LpTkRz2>6X?GT6|z2=k6zxkjBWi>2GW0f=@G6}5j)U3M_MK7X&nQun)o)#O|D
zt&ua<_?n6%x^qbp|IGpaCf5sBySddLm5Mx14Y<l%!9C^O1a?dvt(ZgVQbV`gFPP%A
zKHKcpSu=C&&D~f-JFi?v&1M?&H9%M3om6b8JBc`^3PXKT<V8+|ANfjwuaH&BR8BK#
zXI|)yRj&cNV&k~&oYm5-X+I}6-j)SHF;e>LC1ZrxTX;?s($m!d%a=(ZSJxg$9)^l)
z=RvZ~Uavfr6By-&U5!%cJ^qQFuSf(qW_IApTD!pSN#=#Vc}LQSIvpPC>jwxX<AFP7
zwq4ReyBF&EJS-dA&M!X{A3!IF0X-}i>StWiDngBkO`c~4Ohm61$Ln(Sj9tpkB4%3T
z`0(Ni56=;?PYewVr1I}+FWCg@<Se2Z_`m7>QP=fM)z&j-tQxC6&`as)cA;Auy`1UA
zPs?K_lm0?}K>=8J`YZOV^y1Y^gdK-;H|qC`j-O!wnu*FDQTO0C=kd8tEF?GQmGbJh
zR#)ghvfKzsoj{r9t8jaVx;~48@F=c8>Va>Q1^xMy@t={g>ijmaZLp171(-fUW&F&a
zxxapEc%PtPuF6K$y!wOt@Pv#4j}fxTus&SY<g<Dh(7ydzou)}djcxn%$1#s1y>T8p
zky{RN;ESHDnMS*HjwJH@;jRdS>Q!QIP(BJuHQ;l=xvch9xlm=;6BdrXzWh=r+2oWG
zvSf>WNIgr7$Z|v-X&9Q33P)#K@)bH@5aB@h>Ogu^8KRx#(`tknss+I(!2B1-vsCIX
z$P{cKbO+EJ%Ql>sx&1n1*dTlxs8^ka&-JGlkqzTIpt(_5kwiR^(rsLTS@EV6%wHu;
zys5uVWp2VD)lU}G<L7}-{JbK$2~tjGV9Cr|vSCu&Bh{A_T7jVxmH58=2QhxupH=*7
z+9u9Q(IJ}X$un`eFvzx{&g4t*ZOG~DPPBxB$9BKy#o{#y(0#h~w#+5ydd#p%<aDd6
zTEj&J>8z0z(EtmAPeecnC{2x^nzS<}Ueo<Xuh{@yK%Sb8_K76Ece8^;MybtDY<Ldd
zPR8ui#>%11i|E~qpYqcexL1dk(4z1CmP4>q`H~Jxw(7(`u$Ec{b8c&^ztJzcRm@0L
zv_5rPhY=8Ly$jUAr7(m+3X14~1&Uu}qjgz~U9zm-Yf&1l3KLb-sAv!iZGmnyQVk?4
z360Qsl8YiONN!sV$8xQ|?}m6CZMpTO59OuKtK7MFa2M_H9BH*UQ%)%Wbqv)P6*w#y
ze`Rfia=1iVz4M8_M<T#T3ES64`=1gc%~xSIEy`O0=J7Dn7DyNZuP7wR3#zte|J=3H
zF9;fKblrYah_!`tR(%K!M=5!m7SVAT-QuDS)IoN0;Uj6TQAZv_s)}hUUPg-`(hkb6
z(vyo(JAFkhzw<?tUJ@mB?)yquLG&GBCbGhmQ7J$-yjOlZc-YsWR?FtKj}F!H`}A-@
z^2S|<=CM6&@KXT8X$F04yFU4%+`o`Re0W3lDqQ5~@lOgA+j5C30tu`1nnW{(|A?`S
zAPFGRCj0`Uuar!YN1p%~Hvbm!bn>=4CUlj%Lh3nx|MqsK+lvGPC4M)oa&A=pJOV4k
z#WtKFMa~){E)nB7h|wq3TDGiNIm$=)zK=&M={~SUn6`f`d^EsB{Yrv^pR3xnt7W3+
zHX5YWIU17*xd@bn_QekG7#y?7eMwO6e2UCq+U5OaS3^=eafRo|P;&$LY5H-`td8R~
z?OG^QZIjK|dCMWnE&$vp&wgX!4JRXFgst6V$z8abeJ|wp{JXWhsU}y_mpu|dcyNav
z`@wq9UW^gM^RJfJx;?D`iu_QVmnW?03}0aN>p<cYS}1^HBJjZvyIJ(u_e<%^$J$0?
zvZ_-{C<XD7&0@p<>5V8xlXP6$@~4j~_u|{)^OlaqR5u#YE$7xWqg6NwOqRmGyRwa{
zAXliuJtjfn`OE_F>UM>t%{aW-R<fFnc(?c5v4h8TIN0v2E}bSTL&o+}EFf@Y3aSNZ
z{%@O(?MaypA8K`}{0>X^mqfUTQ%_oPLI7c56$>vyhkAEZ#Q@j#fVuYoS;Pfg$C~VN
zRnT7Vs#&Z<fG?MUjWN5~7Eblp)6=Nj=f1=M1XfjFFwM!+0#fPXBNY$PU9Alb=Me~)
z#dgl5>B5^w7Nq*<`T9ee58rJah~Hp=h;O{A!04riYdXses5=%}TF^1v?_lXz?X)o9
zI=w5_NOaB(LvZ*PrXmcJ2#*>pt?`1C7~Z;M#g&(MXvBn3+uYQSOMmiuz=ZzYj5d@8
z@Kt8*NUlW`1`}Gx!wA8y0%eMg)$DDLmHu3Xb&v|K#i?9g4Mw|y9z1cxtvA&n7XMjh
zOFV$>U$g33ECG0C%?y==5@IcXbio$S?vl)4l@w9WLQFy@P7J8aj*nQMbJ#^k;d4eU
zipxOvMDJ@H#N4s_q0)j|8JiH4@+=5Wgwet>?@eZ~hmg#RAw|+~gFmIvq^d_hz`+YX
za;NM3sgf1d+gAn=f?TaR+tg}X(*jpUiDgT$^s&$reoE}>BB<7A|GVM>H#&waob)jn
zt3E0(zOGwW=dojVDy(!dc;b^gRDd0OQ^OFj&+IYWK^_EhT~AypjYtjZ&X-2f_w_Z+
zee0a|Q{=Fm?jVwSw(+$u#dbK13Qkkhs=iPu$1Z6F*-mp+)e%p8D5RovR|t*jX~v{K
z7xJy%6;KA{>e8Fs;Pb{^tdrZ-Y9AJl)CsJsrnD2e>B&Ba<v0W6I^GoUzgnZhdd{ot
z|D~+xHw!IX;qD7`Ww4}5my-+DjL|8a4Gwah=IbH7le;g!`AvHubSszcHpY(6>&`Y%
zZ0;r`pGy%CfS<W`{boI>>&XbzCLosIO?)o#Yq8Ugkv+hW2egPT-5QB&F8{f#PIFMI
zS?@tUin=e1PFtZ#P>PaqWf1ONo<e;V^22*l6oiHw|CZaSfDL9ePe7${CQulw3L2c$
z8QQDN8b*o<h{$819)QdK3GEtw-!|VqgG(LAHa9TGgR4__ENf9A0HAPUaf8L44q3&Z
zrR0N4?yzoNAq&E72+omTokF@~>zdllK*)R<F@S`~)&!%}U6&>FNrbfOVApiFZR4N3
zZjv;zq;D6!*5u7mx6S?Ja~G%HUn}jeia2w0kBu!04xnQOsKCXAR7!J@adARtx$$di
zPg!PD=gm!hXP4dTmB&AJFsMyW{u-emlhWow&#VN*<@c_4>CeO|N!xLY5I%C451g_+
z37V-XIVYngJ-mtkI3^#PT3Ux?h<ZEX8?scPBRTfaDtF=g3pE}><NKIZTdU<^x+?vA
z!vL%6@1M!sYhzRNk9z*<4OFjuzCM#FFfJ#bRCfj*#h!#eQNTl%*URabq|$p4P2{r>
zMaMIcg#*ul;dG*i5)6vN@28#rW}>f3h4BP6z_u-pS6)Kd>GRz?@!yg8>bUa?aF7<}
zm-3r%<2#8FB0h14XLmrQREDFt@X*4rx*@Z4vS@_+y9vM@tfFEdlo`dWuG)S0hrS<@
z!!kff3`G7xEm&V@wJ~oDw-<C(XFJf@<OHn-E4h4_3Ou{xh7WDX(tqIsTT18)Q1Ckj
ze>mYh3eYd|K-*Ws;+~f&@l^`|Q5LSLKH+bBMRH#J7>X5U(ry|vm2|Mi3cvdzhABE0
zxK>@<)IXskCc`CmFWN}F{7vmt8S)!nlS=>VpP5h>-4z(h7*Epe*{HNNBR8$a+d=rr
zJ|^h88f>al^itjj=#PEKaPsrMwd4V7A~udqQ^`LMP^|<Jf0X|ZbFcRJ*l;vJ-Kod#
z1L^9tHLI}S(?*SUm!T`l$oDKVM>I-x$H=Tyl!hkFe~<9EMk&7@0$huqca<UAy^ntP
zX_He9)TBma5mL|FRLR!P;$_zSM`4BGmOrb9`QHv=<<V2pVIr4;)tD_kB8Re1@=9<$
zLAAEH)Z4#2+hb!Tp#ah~;kytDU6%beHw%&IW>g$otk|{bX6~a7z`&%!)IUzv!>i2R
z2xMhi*l^*7vB>sm->XtE8Ylmq)b9IMjy4Q2&M+4&-q(n<bad<A5sF6m`S~}jZGL43
zeziaO^E{4eSkL+@ET<WyUR759-g%V|7Y2<DTUC};$!#?*_F}dF5nw)a5z=ZFHJ|7=
z3}Lt6@j_3r#h}x{S;ZxH`+f__<{{=+gPjiPL7?Uk`d60H)lO@0rQ)Y!3_sx2o*s)8
zcm^|2T}ICCPeH%@R2a_6B@!3PwAL_V4vUCiP+)kKB0JSVstwUDJ-`>x<X%lOzY^;!
zgk@lsiVUI$)%FYM$qhvcy94_sxX7&Yjydtu9Nh|NT%ltZxQH1<2BY=;8*yaZb))>6
zfAVh)oYhRaTE`aLig|50O3Y6Z1ncA2WDyUX;L|OxBTVzg!4xFT<o9^U^UmP-NO+J(
zu&vBnq40@amk_u8buF;kQ1~Np*Y_zdksZcJzO%F?(Oa!HZ^vL;>M^tnM|eLXfE#8Y
z$9M@p+n{qmOiooRk^x=bc~Dtf{NE>Gp(_N%xb|(|cwWDD5{((Eb#$=8!><;fF8ak~
z+%>aRzJFWmNO{g;y&Ax<()3&(+f)4gM#QwE`9fEQ)Du1jL>8(?ga{(mkL19i>zl+Q
ze$BVTp?3dsHHdmPxZ`vC%+!poyUC4~0?;TetPx{Jm!$+6k1IJX@i{FfVC6_5-~kIO
zi|(&tUO%MgylPgz3t=EMtqU}%vH_txkhRA?%sPPynR$`EuMuUgfw88sbva?#b&Rh0
zHK+I1cf7}ULt2{$|Gfr!_2U>@Z~{ZYQga?q*EK6z5)p9n@ej63Lx;EP2e2uM%dJrB
zE|RYHjE=Q(qEgi$YOZK^j>~@TqKU^|yZ=|~`|<V)44Hysuq*FZ=}J~%+6NnP>?HeM
zE4p{l;n0O5a@Wpg$y#7Cm?sPi9c`)jSY;*zsyC0yYZPHeC||vxhB)=`CiyWCI00L4
zoZxnkF}Kr3Z38EC0Zde3bJZH=$b4LWliTPN_vBp~a9w|Fx6G+3#BXNuQTZzeka;>^
z3<e}gN=-%Nt}Xv@SRX$%TPmoxt<=b_%*UjqXQKkCg}ZjqWDw=v%=HS=w21*UI+Hha
zX8;F^!jT9T!NQWhKSm~+QLA*Eop-gH$!ni(j?c+ifIsb5EQng{u(5QBo1+fPs}+&e
zP2^(D;H~%`KI`F4@nb0Q2d2-Wpw|R8j6W~cGp&<2_9MJX{Giw??;Xp)L=JyesV>m2
z!LHQbUj&jS1gYf}mP+j*x{x&3)>H(w7rgR(cOyM%;V_pAs^gdlP}swe)cqt8nXhod
z?^z5U$#gm2aSGnJ)g#;^1s+kOXQm=<b3Tl=tu=FOgd5`aD}IvC;DzNYw0KdA<gfoV
z_87>$=T!_YWIuAxtVkrn<u5#xvM*I3zQsED_e4eRAGz1w#1ldq6cJ72-jpLZ#pVDB
zayASfI(=wd@0<Kvlh(Vm1b!GMK<v!e#e`kExDP6rHWTd26rn@In4||_kQQ%QBrr%t
zEyl?Ww9XlMJp_(`J8Zu;YB$k#wVRH|><wwGb4^qbwRF{k#hsH1J-=IW#0ROlZzQ<@
zFeg9D3BeZBxi+@pa?BDw0YWr_)TepP<wNSQe)qQ5G(PU<xs9{Ig(<^z$tJolL6!5n
z)_7&6U10<|kb)y3gDWFBz*ji$04(ON^(NTpW#bAnv*P}hcUM3q%SRM_R$y`Q@7tm6
zl0F&+_^X%O)CLlt4x_N4XJ@hOV_$4Q!~+#De1Ui!TE6wNFhgW)aB!$E7Ka-kM$>zG
zEDfP|y#7oTQ@T_Xz0&?|=tTZeU#GTjpMe7R&O?=Uj|{jaBRqYmmya{V+>U9c>z-yU
zi)*ZV5IlDH2~P8|N29mb53x3|Qe)7r=SI_QQfH4o<iZPOK1C~Wy&r0o13M`Pi5mEY
zoHPu5z>Q}q$*iq%`RJ8D|55C-b`zwD_=f~V=j)$6hu;r2wLDPGd4}@RpMjX>6+U!J
zG_c;SQLn*=TP%<+Umr`<Jt`oCrk4^?B|OftORV>EzQrJV(AxH#s);aAQc@BE|L#sT
zDzZmE-E1tJ0}a49ecg`YIv+9+mtK#9ylI}}u;9(kH(occ59T$9k=7&f*r>bQE~BM%
z^r$o9il~SX`rPcMu=;8KR;)c=8psdn+j|=j9Y>R<p1+yOZu{GveE5og5?B0dsE^u!
zKCe@Z6ch0PMdF4p9jIE}2`jOu2Xs9AS#N|d$ODaJ(7A=Z*WQCG)ix~iINOIQU;8>s
z;p=~G86k|lVAm`VEZTxi(rn+H9E`BJT8l4;f9%zi#e@3cZwZmm<;0b0|LRWu?R^s%
z!Ik}Vt*3)6Rpw%A0FzPq($syG<+r|122}1fWx;KncfV?#K`uQ1xV&*ApV#yYlfL_z
z5XaVW<7dZSL@XDif5d6?-=Jy-k-%deu_)a~`Vl<8eEbcVVhHfU(r^g*(({%hj%zT?
zf6FpC!tgxE+cN83AKf*ou=J4<a3Db{NI~$Mu@jiVFp~F@shD3}-npW6Pco4sh^hYV
zD@pzIYDY*F1NGifPS!WcDRI}Km7Xj4wy2C(mD;)N7wIe#3Rnik4P`nE8B#WHbV_}v
zY=&OB!+lt=-61c3N>&ES#A+HmUJgee1yv>%*f?@%#zh1Uc#7*WpKJ8;h1{X)$g3`|
zY7m9(4J5OG4j^}+2h@qkc@Tn5UI+_4m&o|2aZ{Oj35eUh=+jZU;OiJ@VLhU@QL~Uh
zzwM{I?3nB=Vm4ZQBIg#2UhBbn747;Ai}ObRkbSG-udo2*lSu)ZWLzKqW>)Zt2fm#E
z@ZEWhG9r%cX7u16g1lW@Ao)&y41TlijzBkq23e3l5-y94<Avnq5A4c3ssHgx^qK!{
z6NReIOoy?Yn(1ObB|SUUTovZf(TBdKFwR`+*#ao1O!(F6U3K|euixF^uTS4*uSb3)
zTg3HN+63$~L$%uDitZihd^YKN`aQC+ik~sLs#&tKdF-n9_}UB>2WUCkxZ1+6E~?>o
zeROzJAaJTRb7V{vUry@&E#r(g)Oz0B)TGPJocw-&7GMA3ZjIN?f8>)CM7`x&NT%0r
zKw9p<OFkabc|u;=dO`^hUg%fNiuJ14I7dUkW0xCrLGMro>e`iX)a^ucp3aYjt_=U8
zO4u-0ele1dZaRpP68YNs@y0@4@}c;8;P(}984JwTZZJxEneBK+TxmJs6e{&o!=ZWJ
zN&*-)m8<mkzd-pd`(qr)kfX1|use?J3{re%Bm{sFl<M3P=y^T@)zP6U=gKi%=oU)=
zhDFgIYz*5sM2eaxZ{CzJ%Lwf`2&D<VQiR5O?5U8?uZ~>2o92VLI<yjcLiCuh(+|fJ
zpxswWH|A1X7eNw>GQfMT-(~MoHZ&I4IWK7P_9M)M5YrFN@&A?vQN6~`;uO~3qsRC4
z4ifc5rqiH0PaE}UW_oYA&Va6>03)j-Os<-}e}aA$Il2<=2(DhB2N>q(EagI2hJXko
zYJVI5)g~6JE(HnucI-?z#mApQ_MUq7(+I7@@;|))rs#kw^oYShb?0#QRu@X)9@2p{
z{`rBe>K|Wu@^)Vnum_KX+I{WXHS4}u2z-R;MG%q_gkltE1czi%8UZc>7znU|7<4Gi
zebN=zZ~%O7G#t_j59>)zhE0R6O3Y9LLuCmMZPY#bKYz?1QBbV^nKm)w6K6YRp$0-q
z(d01!7f|R}>d0Xn>C-F^KkS<wT0NNpaP%5!=|oAj;nnS2i)^08Ufg)C;>Ut;z6k9#
z1!=IM-@iH`NPP6X3BA`g5j^D(fE)&(%-C&^_Oab)Hwqh0Ingj^drTCf3VYo#xF0Um
z74N#D(_}nRYP#Gi#UBxPKqD9kr#84y)+gU20$eWo<7!uU@^tfku2SP&tL#kLoYf`&
zcic|#Uu2#9oI7|RoAdO-n^!jF_59)A0#^<1v>2py?}U<`!2<@4YP)C-{I6aeOAb=e
zQj>8N2mJQRNw?U|1MB1`46O9uPwf;BOm=)~s{gRiF3P!Hv%UV-EuJMi`rNye9r9zu
zCj@w$<Z^^`RL<p>g6C2*MxoA$NcYN3dJtqAex<x?pYfTrfI|aXI<;y~9R!3F?F`Gl
zk&}LeN|C{QXGPiE*kCiRq>nr>-7D7=3S4Z)4Q7<IYKog`n7~n<ZtJ~iPXYHNlv?&B
zCQY)q!6RL2w&o3K|8{^InjSL^z)(J>e=?x2$%;#&9%(Nmh5PiW5oKq<4|&q^nwonL
zF+{jTG;z%f=Ze+{5*1|cCGcL6D3MMjs7emZr(|XPyy*if-~VJpO0Hcd5J?u;bN)|B
zv)oO(U$on=uJ*la<?1)X)E99^O)?8(f*#naMUp>JY6UB$qumGy3-tM*>o6!3DW%ca
z16h>Z@#u}=d1FHTB}DA7NgAg>(`N0H5g-HPdM;|+bD=x&eeF^`{a>3GpD4^nZb$Nw
z4N5gRc7>*8aWQBW>Y~q`FzO;(<pROvcl{cJ{YZ8J=tq=Rt;i;I$BqPCyTDiM3edfB
zn^x<a!3hc}cEx;h`GEji<ZaC>O7qS<%S|KR>a*4hEVmB0EU}6HjXTb-PG^ec&WDQN
zyQw5D0HarX^d&c1MomuJk`1vY;2sTV^SgHmwFFlu&-dr$#ABvE%b2Q5j+8R8y36;>
zPj;*ad9(kN>_p_B_2iijqI+v8;mpx?9mb*H@clI#q%|wg&*0ICfhKif)dW+rXc?n-
zIw`5bj>zg?!G}Uu8YGp$vMF{bDI|h01JkpY_^nSUF?C+l%vw0=73xX&<SSZ(TNE)n
zPrf*G>YO*mZiF`e?w0GGG9v}8unb=uTJ_zXV<4V};XjOTKGx1&0RUF@JKph|`4*Qm
zxreJt^0SXNFMWajDbKhe4k|3T(nN-yjMNAf)O2i*5XSSp4lFKe=VLpPB$=6oZOMS&
zRi4kMQaB~^s+_MNwnxA4D<EG?DRQUcIGWdn9$---4$MD6YH>9ml4V%$ay_%PQylhT
zrNym`f7JZ^TodBbT0AInX=HjPDvv|+HWj+_UddSZnKXRsKmn)G-xF5)=V_m|0Iy|J
zPil}6Lj+%5fg^m>T^RKWGyrG2fAD)(ON}4X`|l`fkAC<^T5EjtgSP`AX!G}(Nvlt3
zlrI|GdByFY)+#NeE1tYL1ya}Qj+$5Tf~hQ(a4IckIcu;9Y3=ungXV<GX}-e1ZFb{w
zTuk{(;3}vWS8%6C3AvPiWbqPs0kD&uN-b_nVA`p%VJ_lc7vV>o!iSrPVz_M)meIeb
zv7%O7<9S7%7CMCKiS_O1X_oD9ObX%P9Lf2-EL>IgbPz+T;=%CrbBa?c0%8?at7Iy-
zY=+Wq%f;5gWxBBOc{;TP(>m*;{`U?7x^#l7)6@XR9Tdo7nJNe1mnnN%zdtMYJ2G68
z`tgK}s{v7)25fMfit8qc-`jAp=XEmuhHBeK#%Cxg|5)~Wf@xGY0!^pGEe}-{I=Y>z
zQA3YqDX3(Vfr_5;eIjZbGoiA(eDgvJwioT7!K%b{J?aWoV(Y_};5BU4!-{EFsG%_F
ziLX$%X)mjn_j+{+Jm~O3Ku9ZdJU9`;VFDSJMJ>I@udyN-VT`Ev2!}qw(yQyDi!xMV
zn-E16G;9~~zjwKuZHOqlOZl@bbY{HEdm@h_i1(K%d?d0g9wz(Iv=rl}TE#$o1*=GK
zV&6AFKp(xoTTt|B?4YLg!m~Za=s;hR)l`1AMdKR1=@65Myo#$~xV7yA3v(@+s*58L
z%$s|Uoj4?%lrp7KZ3L-PG|IYiEBi6&(5G9(Re$MJz4O<EkZMF+Ko*w6SC=DZq!cHd
zQ<O4T2AI~j+o0!I&a09EbD3c*^tjhTRmIr+sJ~?x4~ry6Ol2unu0-Q6{?WKc3l?Q(
z)xWD^l>Vwj5dyU?UlRN`_|EV1Z5;o6%$t{{nt$i~>%bimL~`9@++k-a@_x>@)!PFh
zK>zbkk*C!HLQ9OhbJljHkDYmEM<_Uv+Ur6iwz*waAu^0z)T+PsK9v#iaHE{w&&n2e
zD99qX8b+h9RD{@dgU?rXfispq`H1pHUjA#OHR#e{x#dFeANtn2n;M7DR;B-^@Yvk>
zw<M=q4AVuE1=-D$#wC_F)`4L)D7CSgvm=~GG*BbNp#N7|@FIF0|Ixt5wT$f;AqAz|
z{>X9%`7vBnZz~1**>hGHFw>9XFl&wHwQ4`d>D&z+zI#PKJkp)ea^>0rd5tcW3tLl$
zTxR3vzWP$y#_Pet?hEnk7$h~Gmx^)<(;_<vDZMwF$FRbU^qSh3p5gl+t`rQ1bet68
zPePU2Vcs@uzoJ&e)Z%uVa0P|d|BPr?$w6l_z@LxLZVk-hXQg8fyC4Z(#eF1C`leff
zF!p*%1N#+xb~k>+y}=*zEIQbvU}79Z3&#Zhv>t7SwNkDY(Vu&jtMN3QDTq=(#8P{C
zy_qXS%E5!EgoM%)KMwL(xuic8B8BYXvx>*$pZ%76TRV`sz2+$o3Mf0Mr3k1Ct<m@O
zUhlB7N<YG`3~=7DhOHs8-R7=9OrEYs@CRXxz3-}k;6596Mrc-MVa7ch`Sx7MSDh+l
zFsG25!8{E%psjas-a9olsj7*>=sEChAK8z&6z!@t_4QU8Z{0D!C=VL<Xom4ICGD_-
za8ut(jhK2Oezv0xDc{c(s0_MCU6I(iJ6o&-Cv`SajmI8~GK4-at64&jl<|5xJ>%#4
zqWDkIxzL96&Q$dPs(p^1%N_m+1tGZ#X_q`ZzJTPErYL%zNl1u3Wd$e?5s`i_`M2Tw
zmeVVn)B`g|uQ@eap}+1l7U{OXqD>TsbEevdr&gJc2~in%Q{75kW-ZE9`lW|kUaL+A
zRBOGh_gYE9*z3G~<!AQz|J8Ax(Qr6X8&`uBMDHac#Ol$@#_B@!)uToiB^HZBThVI>
z5+$tOgJ311too1MiC&_Y)oXUWoA>kkex7^g%$#%YoSA3t^E0ldsjxr+Tuj@u?hMt)
zi5w&9!Sc_$RoFBApX$fym#L(6dNEfpiO{1A-uvOUqnF&Me^w4gv;J68B=~&1U!9KJ
ztnhKkuT-lrt?iI)*`0!a1zOrUWsbT9p|P`y?CEYYLl-d6A)|D`9%TL3u@T_JP0<Sv
z?}5}m7C>r2NsEY&<MYH#HAlmfE3{|!@x&ZRa$LV1uI3h)?4g^WZCgy?1kM(1AcGbA
zf==}MstrmUk81&PLKm5xxx6d+dp!s`m7Cae4DP!yi?o{sG2|9aO?<h5fad*3<E5(t
z2W-T%@=473wG2vikQ}d+C^Z|pF?g=<traEp(L*{3!F$hNi(A1n%4>Cd*4xZNfEfs!
z<-v_N|5eAU@Da=vHqdfP&^hxp&PSuX!fHL#UROoG^q?~M;|qKLP?nn_U5#R(pmDWV
zB6k40SFa0(;y&fndP%Sg!v5@O+``vCI;xHW%3Pof#oFDbKs|oLW`amJEmYiv4W@sy
z6ln1~*ykHlR0P$;8)?XD8j!c5`vfAyi+F)6xlcok>;D?};39P;c{J3cPjMHwi~j!N
z@7j%%_lEe*@EQ49ry)&8@r3w7_5}a|KXs`HW|v~Ou}G~a6Iw6n{fK4`84-RK34n7=
zT2H;IUAsGQP-0CYgn(JBb^|l-b2>@q+>za;zoAvrMg-%TyyKqud_ph$J!i>u>2Pb}
z&GnA&*>jNW7eIcaN6DVNo<(6#nSP`raG|(+dYmt<bDh^}>YujvH2kBBf7zm&d6DD>
zP}FrCP7famM2HHjIV--LL5*U4^He25>)8BwlZuB`wT<7OPbkw=oDf-*tcl&l1Pu1U
z5c;mM<qSc|4j<FN_H>nd-X~(ju#MHh@QYL@`YLcRDxOumK*@Armh9MSyQlEZA>rBM
zs!4Oql%mo`R#+z)pV$s0iq-%mG*Od*6YpXK379Tt?Kg*G^OoK!eK4`CaU~YWx49VX
zB^#@|IS{W8NUb%DWKFktRUFq`;<#-&Cwa`no2T_=`N>>@UZo;uswF=GFZ=9ys)RFU
z5vpiZk!1U^GLTq+>o`~-vHH2I`_Tm47JOPlS;xVddG&+iIG*LRJGa&ole%D>Q5;O9
zKr5=bz+4)k=6?7EIR+K>$RQ&q7SJ~}sY#4o7h^qI|Bg}@aXLs|+i%s+@08S1IOOa?
z)5R*zf|Sl(bHmIuQ<w%quUE$dD;t&=9)`|7F&q07^(dS6+(3{sTT<nnDK}ZD;jznd
z@r#}<A4r+KAnv^6BEZtxHYb}{(6u;D9YAeP+v)tQm`wmC`5xthF9;78;yC-_(cwI>
z<ZCsVLcI)EF366>KaoIR^%rk5x=a4_%TAz4i$<p|Dz3&w1S2tESL@N0{HXPSg?dtT
zpW^XCfy>~erQ643QX*aO6GVvEK2jia1N-=Be9~Ci3}_bKTrJn0H-VBOjE*gRgtpKH
z`ip=BO~9)yOe)2{|4uv~gK_5z;;&XFwPoBj`j{^j_D|Dgjdz)#0698+O?=w-_M&bm
zKW(r``?U}NC05E!xF75gy?j+8_P5O#xdCFW5$v=;#p$-3qz`3IBD+XT_$FMgGQ3;Q
z%d9@Uz>DBN?q>2|kJg6XF%PqKTBG8ITVD5+oXQW5Z+g#x#d5r~+$;lVN3~ZHlHu;t
zf&;g(0Hnm=LPOAA)qu@{!qmP)9@2uV0{PB23C9iDAZ7AhiJ6cUv4+kN?|rOP13lJw
zPdp!aTfcRxA?thOVT7C|w%Xdx#l<?3#mo}OZ4O$6UJ_?nV|U4mV+?Bv1`8Mg;yS5N
z=U~1*Apk!${=e4SWA*7-&!)WAbJR6Sgh8IgZcKA*o*!%zBAHcO=Q`#M7Rm+tJP7yZ
zO@YgIG-VeXd?<OTmZzQrr%zl83e`1IY4jWq3+TWT`)6v=j{{rWe`m4CEFZWRA<M+^
z?8LB}f;wlZ=kC}EmFEqcQk2bASToNVj%uR~2a^`{c34(c`Y(_UWNPehqKhC$ls}FW
zJKgmfW|t-A%}0?$RDbT3xW|UEkZEqO$0=E2Pce?)^~wy|Z2MIK7aDVnpccEQGdoBG
ztEs)HeAq5G!Kx~k1(nnNIKQ1J(U=XOMtchUK>c{XPul4$e!`<?h`7&R{>4A7<p_?a
z>qI}(babltb{noGQm>(!5tH-hWb#|yol{8SvqrnM4w(RRA_UG)6%z&{dTKwrTd>=E
zjC*QtsJZ=ta3}`1`1JHvW(nSe`pz}4mPVX~<D4x<q}y-aW)h|OqN9;p&qDjjZq~TL
zfTm2=)syxOs;&oOdDFglhkDxoTrh`3YYC$(c(*?jVM14Jw{LF#isQ3zUwu|f)~h`H
zPoTVVoPs%}qf2g0FF9%gC!dkv`=WsRwL4L1J_qb{=C!S-*E{=;BAYhzb`hpx+>FHW
z*Rg=$7zqMi(Civd+v{_y(@v*l!N<2LbX8Qok(sD4RZO3;$GZ}$OaojhGZ3W~<d>lc
z$o&`cOlcS>0LqVWfC2pL8&m0Jn<@NcOBXvh=h-$*HT45>5<{Kgt)j^ik|oRfZiM8u
z3cm=ync|Mk*?7qZUS5wXk7z^)gHk3n{uHgI1d5Ct_OfZG=Gvy*+Ke8cnf5ZozrSxl
z_(^IitNKN}b8;+KX9$A|F4~x8RNQ+_bvo^FBZZ3Q_nT!k2h`2TVX1(#V!C3LjYL(&
zK^VUxdl8i?xRA)Oa8=Vv{^d^$>L-T91Mb$%ODyopR4}l7WmhWuG+l}{KE2Ki7)MGJ
z8TP#IwV3)9T5@WhGw~0pnNKG!lut*{eb^}SIU!B1SElr-nyxT(6~acXAA?KT+p2QH
znuQavXQZ;RszY6oE&c9=njT|Lsq|5(q&b%*0$66o{XjHa^$~>7S^8_MJS?~)(f}Vs
zI?7{8#dE@D2D3RV<(xWX`ouAWj*$enC=?s=tAaTQm`p}fS%yNEElOFz{l*+6hguKG
zr`hR0CtDp|_0BdpEar_B?{UNvo;_vM>!HT~iffpjrlYkqcK^0ehxww`Xl<j(jKL+)
zw8(3$mBiNXPJGNOw!12qa9Qe>=%~=`Z6GsgKCI*N+$m(S^m3$-LeX($M~Aw|wiUqx
z{Sr9hkwsmaY2DSaMC&M=759h$MC#DpflwD%u<+Jfy5Qj32`j*na)5Ej%g)m;zZK+V
zErNdhNYq1v_;Mc;w9$g&8;d4}(c6x<3{EWaUPQ1?I=%=cww&k$|Gt{xwd#$w8Fka<
zN4WK|r*m<(9R2<aT`ySqP3zI1|M0)<uKFK5w2--jztVe$xtBPC!EWE5A!jyg)G46>
z83dt$@E&1F`pRQloAxFo?OY#4^`)kCeTkZjc#pj_|BuCAq|lbePL2o!*>#l<(M*3&
zkKbSxFIT;fpt#(uIV`pt6qQ30E41IY5t-*%%v<f~rM0@&DTJy+M?r}I=}air@^%%@
z*D^fTg2rv~X86UifX|mDA9c5c)4Cs&GRr%vZ{al5M_}**8*|<MX^{9|r-9b*d2j!_
zn6T?A3xpA@hi=>#{ve~#(;pR=M#*RUTE@9>(Q}TxXV3C}Ary^*>OqG$e^Yr0UzOwC
ztkB}$9N)YbX#PAB?4~lEZ-&X+9rUKXcOylN^smqvB2pF2=zcOgJ9`*lyBBdv74%3T
z37O<8cP-qYn>kOlq3Fq!ek&*HPirl(*5b4xKxspfBg;5X<$28}w1_;q4U}Tl?alZj
zz3!~_j3z(w-ZKJnr;CG25t-wQ@t=+ixxzy79N14T_KB#fyj?@ia3(@`?byinr-Tkw
zZ6L1?$bRF7?$wo<70X!6=zhDu*XIL$Ea|()`Nr81eMU@~hxA(>>Jy7=En^+R50?3_
zb;IbY(1Y-ho@sgL+)}F~dbP1}e!@5PG_}FyiVaC$s!5`Rjva;k$Fz=Fn}w;j`hNdz
zt%W^m`bk25*k5^R&T?<!C`)fF%x^1d=)k{=F<>bnxsj}<{_-6+b`-Xn0&h-A<s@=3
z{=hk^?EmXzV^}#XxfYwe`PatIshy|Ku#!DPGV7s$!&8HvIO-V5dXvJXYkz3zp$J-N
zh@YZ?8<MjP8w%=X6HOYvCErRR9rS;O^^k`4Vn?fMKuirTFOlf8F%le`CpfIQq{{M6
zycDo;{ViYvWF*sKQ*}x;#jZd?U!$bslPcIoJ*vI%fu`%3*Z%uED;s2R+D|(|b0MIi
z*mRIqrv^fa>savq%~QNS2iP8a6L3CRRyE|FgoGbJl(Q`bevY4E@ojTC`x~KSwun7P
zcQ2E0@|0|*<fm)Fmb9Hop!lzIaeOhmbdEk4GHdnwX43ZMa=VxR-iWg!rh%5nIXVAG
z1{)+f-5HugB)>hn6G=XURvSt{LD;whwBD8ck;h?n+Qf|x=(9t_ZB!3#q62cz|7?x9
zj%}*Fk=x2`%oUyA<r*;2Vdv{LHhM9<MoIE^%6_c%0&NN1rOQn2a&}3qdC*pqp%NLb
z`*p;}7D^pN`LSS1?nO=)B0_ug2RmT!`u9%YP793aKb--2-K=3r%toV_c}DX<K>-_5
zB75~kJbQt}B3B}SGx8yUk?4w6W#rvA)b7<KBbA^Fsi#qfcGQ@B6T-O><u7{8hS`p?
zDJ1yhB@aoNe;jQGSsHL@Y%=?f5xF;t?pb$#9ewJxC#Q#Q`&a0!B12o)jsEl>w2rPu
z^|U$KqjKgiXNEQ8Wzr;;Mbo9Y_~T~s-KTCUqyj1>d0ppW$f0ePrTeknCDhPpdB4=7
z`&ZUm4$|3FTjZRxMyVl#3*W|!_b3#e{64$juk~`D=V#uIpTSjsP$E!4d^p7St&eh!
zhn-===?`A<oGGWC#_f-Pc^^2xFEJy$t7%NbkW23L;T~(ow^pB&8RH%j@XfS_XH*>l
z1IVNC{Gakc3NJezu=#=rWo+QaU+dr8vU>+qo~SSa8jUD=QsKrbB4wKQR<Xs>zehS;
zE*bN`IZ#GH*c@SaIvL-09*HGox%f=0?-Yw<&*i#T<uBKtV>=eh7a+wQ2%Fveg{+zO
zDCm(Sy-kGompv|TS4Lo@<aAfI>AIK}J`Ij2#?|57{%-iB^NB`*N054sC!#wSIw-|7
z<>u(R&B|<WNtvX567Z-pUP^{){@_ThRn0<#x%g`jcZzpANm($KHl~a-_U_Ym{~-~>
zvzFmWPw@nC=AvKaWA7;M0Y1;!MLevY<}}6g8>vE#j0G)GJAX{t4MV#%)q5P6T(q_&
z9{l^Tu0n05{P&Ot43K(%)K%?<2(V=D#ctn93;ycnRb{?*NDz>l{AJDnTp45_*VB(!
ztyFQwJ^{JMjoaOa^u~Kgxl&<e_q%?z(PWRkJevZ=TXa@?m4OfsaT7``JgPTY&2}HR
zG(3>{0(`DO-z+XPMugD`Vk&zjNWPw2A{N_&C!A6%_t50)x*R0vUILz@vNXxBH3#0G
zQJ;MW6V|(!-g8z3v6~%GZi?=DZ&~-U>+<b&*YC0E`WH<68nBczi1l$6A_>l2>ev<X
zH^MMd<t#FPU<W7i{$D)4|DV+Nzaym7;dfNZPC%~4;k{8Dml5x|nyzY%vMv0709UFk
Au>b%7

literal 0
HcmV?d00001

diff --git a/docs/2_39/functions-reference/index.html b/docs/2_39/functions-reference/index.html
new file mode 100644
index 000000000..820f6d50b
--- /dev/null
+++ b/docs/2_39/functions-reference/index.html
@@ -0,0 +1,1142 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Stan Functions Reference</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/void_functions.html" rel="next">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Functions Reference – Stan Docs">
+<meta property="og:description" content="Version 2.39">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed slimcontent">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/index.html">Stan Functions Reference</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar zindex-bottom">
+    </div>
+<!-- main -->
+<main class="content page-columns page-full" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">Stan Functions Reference</h1>
+<p class="subtitle lead">Version 2.39</p>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    
+  
+    
+  </div>
+  
+
+
+</header>
+
+
+<div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
+<p>This is the reference for the functions defined in the <a href="https://github.com/stan-dev/math">Stan math library</a> and available in the <a href="https://mc-stan.org/">Stan</a> programming language.</p>
+<p>For more information the Stan language and inference engines and how to use Stan for Bayesian inference, see</p>
+<ul>
+<li><p>the <a href="https://mc-stan.org/docs/stan-users-guide/index.html"><strong>Stan User’s Guide</strong></a>. The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:</p></li>
+<li><p>the <a href="https://mc-stan.org/docs/reference-manual/index.html"><strong>Stan Reference Manual</strong></a>. Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.</p></li>
+</ul>
+<p><a href="https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf">Download the pdf version of this manual</a>.</p>
+<section id="copyright-and-trademark" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
+<ul>
+<li><p>Copyright 2011–2025, Stan Development Team and their assignees.</p></li>
+<li><p>The Stan name and logo are registered trademarks of NumFOCUS.</p></li>
+</ul>
+</section>
+<section id="licensing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
+<ul>
+<li><p><em>Text content:</em> <a href="https://creativecommons.org/licenses/by-nd/4.0/legalcode">CC-BY ND 4.0 license</a></p></li>
+<li><p><em>Computer code:</em> <a href="https://opensource.org/licenses/BSD-3-Clause">BSD 3-clause license</a></p></li>
+<li><p><em>Logo:</em> <a href="https://mc-stan.org/about/#copyright-and-trademark">Stan logo usage guidelines</a></p></li>
+</ul>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/void_functions.html" class="pagination-link" aria-label="Void Functions">
+        <span class="nav-page-text">Void Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/integer-valued_basic_functions.html b/docs/2_39/functions-reference/integer-valued_basic_functions.html
new file mode 100644
index 000000000..4788a0d0f
--- /dev/null
+++ b/docs/2_39/functions-reference/integer-valued_basic_functions.html
@@ -0,0 +1,1274 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Integer-Valued Basic Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/real-valued_basic_functions.html" rel="next">
+<link href="../functions-reference/void_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Integer-Valued Basic Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/integer-valued_basic_functions.html">Integer-Valued Basic Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#integer-valued-basic-functions" id="toc-integer-valued-basic-functions" class="nav-link active" data-scroll-target="#integer-valued-basic-functions">Integer-Valued Basic Functions</a>
+  <ul class="collapse">
+  <li><a href="#int-arithmetic" id="toc-int-arithmetic" class="nav-link" data-scroll-target="#int-arithmetic">Integer-valued arithmetic operators</a>
+  <ul class="collapse">
+  <li><a href="#binary-infix-operators" id="toc-binary-infix-operators" class="nav-link" data-scroll-target="#binary-infix-operators">Binary infix operators</a></li>
+  <li><a href="#unary-prefix-operators" id="toc-unary-prefix-operators" class="nav-link" data-scroll-target="#unary-prefix-operators">Unary prefix operators</a></li>
+  </ul></li>
+  <li><a href="#absolute-functions" id="toc-absolute-functions" class="nav-link" data-scroll-target="#absolute-functions">Absolute functions</a></li>
+  <li><a href="#bound-functions" id="toc-bound-functions" class="nav-link" data-scroll-target="#bound-functions">Bound functions</a></li>
+  <li><a href="#size-functions" id="toc-size-functions" class="nav-link" data-scroll-target="#size-functions">Size functions</a></li>
+  <li><a href="#casting-functions" id="toc-casting-functions" class="nav-link" data-scroll-target="#casting-functions">Casting functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/integer-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/integer-valued_basic_functions.html">Integer-Valued Basic Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="integer-valued-basic-functions" class="level1">
+<h1>Integer-Valued Basic Functions</h1>
+<p>This chapter describes Stan’s built-in function that take various types of arguments and return integer values.</p>
+<section id="int-arithmetic" class="level2">
+<h2 class="anchored" data-anchor-id="int-arithmetic">Integer-valued arithmetic operators</h2>
+<p>Stan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in <code>2 + 2</code>, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in <code>2.0 + 2</code> and <code>2 + 2.0</code>, then the integer is promoted to a floating point value and floating-point arithmetic is used.</p>
+<p>Integer arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.</p>
+<p>Secondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, <code>1 / 2</code> evaluates to 0 and <code>5 / 3</code> evaluates to 1.</p>
+<p>If one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have <code>1 / 2</code> and <code>-1 / 2</code> evaluate to 0, <code>-7 / 2</code> evaluate to -3, and <code>7 / 2</code> evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.</p>
+<p>Unlike floating point division, where <code>1.0 / 0.0</code> produces the special positive infinite value, integer division by zero, as in <code>1 / 0</code>, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.</p>
+<section id="binary-infix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="binary-infix-operators">Binary infix operators</h3>
+<p>Operators are described using the C++ syntax. For instance, the binary operator of addition, written <code>X + Y</code>, would have the Stan signature <code>int operator+(int, int)</code> indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.</p>
+<a id="index-entry-2e81f507ecb4144548330dac4d8bbc43686cd3e2" class="index-entry"></a> <!-- int; operator+; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(int x, int y)</code><br> The sum of the addends x and y <span class="math display">\[\begin{equation*} \text{operator+}(x,y) = (x + y) \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-63c9680853b8c13d6503afb15b5628911298a42a" class="index-entry"></a> <!-- int; operator-; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(int x, int y)</code><br> The difference between the minuend x and subtrahend y <span class="math display">\[\begin{equation*}
+\text{operator-}(x,y) = (x - y) \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-74f2ac23b7142dcb99d72c2d668ea84d112f43ef" class="index-entry"></a> <!-- int; operator*; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(int x, int y)</code><br> The product of the factors x and y <span class="math display">\[\begin{equation*} \text{operator*}(x,y) = (x
+\times y) \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4ae4285a5c320ab57f323608bdfa3c237c8e33e8" class="index-entry"></a> <!-- int; operator/; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(int x, int y)</code><br> The integer quotient of the dividend x and divisor y <span class="math display">\[\begin{equation*}
+\text{operator/}(x,y) = \begin{cases} \lfloor x / y \rfloor &amp; \text{if
+} x / y \geq 0 \\ - \lfloor \text{floor}(-x / y) \rfloor &amp; \text{if }
+x / y &lt; 0. \end{cases} \end{equation*}\]</span> <strong>deprecated;</strong> - use <code>operator%/%</code> instead.</p>
+<small><i>Available since 2.0, deprecated in 2.24</i></small>
+<a id="index-entry-5c1d4a2c0b351939f98b1c7a20a25f07b08366ba" class="index-entry"></a> <!-- int; operator%/%; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator%/%" class="unlink" title="Jump to index entry"><code>operator%/%</code></a></strong><code>(int x, int y)</code><br> The integer quotient of the dividend x and divisor y <span class="math display">\[\begin{equation*}
+\text{operator\%/\%}(x,y) = \begin{cases} \lfloor x / y \rfloor &amp; \text{if
+} x / y \geq 0 \\ - \lfloor \text{floor}(-x / y) \rfloor &amp; \text{if }
+x / y &lt; 0. \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-1a37a12af3bb33938a4e3fab2375e8f57617354e" class="index-entry"></a> <!-- int; operator%; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator%" class="unlink" title="Jump to index entry"><code>operator%</code></a></strong><code>(int x, int y)</code><br> x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. <span class="math display">\[\begin{equation*} \mathrm{operator\%}(x, y) \ =
+\ x \ \text{mod} \ y \ = \ x - y * \lfloor x / y \rfloor \end{equation*}\]</span></p>
+<small><i>Available since 2.13</i></small>
+</section>
+<section id="unary-prefix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="unary-prefix-operators">Unary prefix operators</h3>
+<a id="index-entry-ef76e937371aef8a99cfd4a9964fd54af23d7344" class="index-entry"></a> <!-- int; operator-; (int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(int x)</code><br> The negation of the subtrahend x <span class="math display">\[\begin{equation*} \text{operator-}(x) = -x \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f" class="index-entry"></a> <!-- T; operator-; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(T x)</code><br> Vectorized version of <code>operator-</code>. If <code>T x</code> is a (possibly nested) array of integers, <code>-x</code> is the same shape array where each individual integer is negated.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-1b4db535db6b20e4b8af55f41af6ee4d6b5df6bd" class="index-entry"></a> <!-- int; operator+; (int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(int x)</code><br> This is a no-op. <span class="math display">\[\begin{equation*} \text{operator+}(x) = x \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="absolute-functions" class="level2">
+<h2 class="anchored" data-anchor-id="absolute-functions">Absolute functions</h2>
+<a id="index-entry-c197d7c04ea9bf23cf252f2092111b3c5aadd02d" class="index-entry"></a> <!-- T; abs; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#abs" class="unlink" title="Jump to index entry"><code>abs</code></a></strong><code>(T x)</code><br> The absolute value of x.</p>
+<p>This function works elementwise over containers such as vectors. Given a type <code>T</code> which is <code>int</code>, or an array of <code>int</code>s, <code>abs</code> returns the same type where each element has had its absolute value taken.</p>
+<small><i>Available since 2.0, vectorized in 2.30</i></small>
+<a id="index-entry-317ccd335f85a10d3c7902b15dd6af1f6ef5d16c" class="index-entry"></a> <!-- int; int_step; (int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#int_step" class="unlink" title="Jump to index entry"><code>int_step</code></a></strong><code>(int x)</code><br></p>
+<a id="index-entry-ee3b76044455776615f3a8a6106dccb8ad9a235d" class="index-entry"></a> <!-- int; int_step; (real x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#int_step" class="unlink" title="Jump to index entry"><code>int_step</code></a></strong><code>(real x)</code><br> Return the step function of x as an integer, <span class="math display">\[\begin{equation*} \mathrm{int\_step}(x)
+= \begin{cases} 1 &amp; \text{if } x &gt; 0 \\ 0 &amp; \text{if } x \leq 0 \text{
+or } x \text{ is } NaN \end{cases} \end{equation*}\]</span> <em><strong>Warning:</strong></em> <code>int_step(0)</code> and <code>int_step(NaN)</code> return 0 whereas <code>step(0)</code> and <code>step(NaN)</code> return 1.</p>
+<p>See the warning in section <a href="../functions-reference/real-valued_basic_functions.html#step-functions">step functions</a> about the dangers of step functions applied to anything other than data.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="bound-functions" class="level2">
+<h2 class="anchored" data-anchor-id="bound-functions">Bound functions</h2>
+<a id="index-entry-82a2d72a544284ee41b99d4842427ee71f7025b9" class="index-entry"></a> <!-- int; min; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#min" class="unlink" title="Jump to index entry"><code>min</code></a></strong><code>(int x, int y)</code><br> Return the minimum of x and y. <span class="math display">\[\begin{equation*} \text{min}(x, y) = \begin{cases} x &amp;
+\text{if } x &lt; y\\ y &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-fbc1a906ce513495836edb5c1cd222c564bcc768" class="index-entry"></a> <!-- int; max; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#max" class="unlink" title="Jump to index entry"><code>max</code></a></strong><code>(int x, int y)</code><br> Return the maximum of x and y. <span class="math display">\[\begin{equation*} \text{max}(x, y) = \begin{cases} x &amp;
+\text{if } x &gt; y\\ y &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="size-functions" class="level2">
+<h2 class="anchored" data-anchor-id="size-functions">Size functions</h2>
+<a id="index-entry-cc587d014dd8cd99986939af8ad1854a3c3425f0" class="index-entry"></a> <!-- int; size; (int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(int x)</code><br></p>
+<a id="index-entry-f9c96384668fa7a7880511c2c0ad2558be24e0a4" class="index-entry"></a> <!-- int; size; (real x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(real x)</code><br></p>
+<p>Return the size of <code>x</code> which for scalar-valued <code>x</code> is 1</p>
+<small><i>Available since 2.26</i></small>
+</section>
+<section id="casting-functions" class="level2">
+<h2 class="anchored" data-anchor-id="casting-functions">Casting functions</h2>
+<p>It is possible to cast real numbers to integers as long as the real value is <code>data</code>. See <a href="https://mc-stan.org/docs/reference-manual/user-functions.html#data-only-qualifiers">data only qualifiers</a> in the Stan Reference Manual.</p>
+<a id="index-entry-04b9411b8a7b904eccd2c168519bf4ccca638602" class="index-entry"></a> <!-- int; to_int; (data real x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#to_int" class="unlink" title="Jump to index entry"><code>to_int</code></a></strong><code>(data real x)</code><br></p>
+<p>Return the value <code>x</code> truncated to an integer. This will throw an error if the value of <code>x</code> is too big to represent as a 32-bit signed integer.</p>
+<p>This is similar to <code>trunc</code> (see <a href="../functions-reference/real-valued_basic_functions.html#rounding-functions">Rounding functions</a>) but the return type is of type <code>int</code>. For example, <code>to_int(3.9)</code> is <code>3</code>, and <code>to_int(-3.9)</code> is <code>-3</code>.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-04b9411b8a7b904eccd2c168519bf4ccca638602" class="index-entry"></a> <!-- int; to_int; (data real x); -->
+<p></p>
+<p><code>I</code> <strong><a href="../functions-reference/functions_index.html#to_int" class="unlink" title="Jump to index entry"><code>to_int</code></a></strong><code>(data T x)</code><br></p>
+<p>The vectorized version of <code>to_int</code>. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.</p>
+<small><i>Available since 2.31</i></small>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/void_functions.html" class="pagination-link" aria-label="Void Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Void Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/real-valued_basic_functions.html" class="pagination-link" aria-label="Real-Valued Basic Functions">
+        <span class="nav-page-text">Real-Valued Basic Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/integer-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/mathematical_functions.html b/docs/2_39/functions-reference/mathematical_functions.html
new file mode 100644
index 000000000..d81ac39a9
--- /dev/null
+++ b/docs/2_39/functions-reference/mathematical_functions.html
@@ -0,0 +1,1178 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Mathematical Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/functions_index.html" rel="next">
+<link href="../functions-reference/embedded_laplace.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Mathematical Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Appendix</a></li><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Mathematical Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#math-functions" id="toc-math-functions" class="nav-link active" data-scroll-target="#math-functions">Mathematical Functions</a>
+  <ul class="collapse">
+  <li><a href="#beta-appendix" id="toc-beta-appendix" class="nav-link" data-scroll-target="#beta-appendix">Beta</a></li>
+  <li><a href="#inc-beta-appendix" id="toc-inc-beta-appendix" class="nav-link" data-scroll-target="#inc-beta-appendix">Incomplete beta</a></li>
+  <li><a href="#gamma-appendix" id="toc-gamma-appendix" class="nav-link" data-scroll-target="#gamma-appendix">Gamma</a></li>
+  <li><a href="#digamma-appendix" id="toc-digamma-appendix" class="nav-link" data-scroll-target="#digamma-appendix">Digamma</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mathematical_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Appendix</a></li><li class="breadcrumb-item"><a href="../functions-reference/mathematical_functions.html">Mathematical Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="math-functions" class="level1">
+<h1>Mathematical Functions</h1>
+<p>This appendix provides the definition of several mathematical functions used throughout the manual.</p>
+<section id="beta-appendix" class="level2">
+<h2 class="anchored" data-anchor-id="beta-appendix">Beta</h2>
+<p>The beta function, <span class="math inline">\(\text{B}(a, b)\)</span>, computes the normalizing constant for the beta distribution, and is defined for <span class="math inline">\(a &gt; 0\)</span> and <span class="math inline">\(b
+&gt; 0\)</span> by <span class="math display">\[\begin{equation*} \text{B}(a,b) \ = \ \int_0^1 u^{a - 1} (1 - u)^{b - 1} \,
+du \ = \ \frac{\Gamma(a) \, \Gamma(b)}{\Gamma(a+b)} \, , \end{equation*}\]</span> where <span class="math inline">\(\Gamma(x)\)</span> is the <a href="#gamma-appendix">Gamma function</a>.</p>
+</section>
+<section id="inc-beta-appendix" class="level2">
+<h2 class="anchored" data-anchor-id="inc-beta-appendix">Incomplete beta</h2>
+<p>The incomplete beta function, <span class="math inline">\(\text{B}(x; a, b)\)</span>, is defined for <span class="math inline">\(x
+\in [0, 1]\)</span> and <span class="math inline">\(a, b \geq 0\)</span> such that <span class="math inline">\(a + b \neq 0\)</span> by <span class="math display">\[\begin{equation*}
+\text{B}(x; \, a, b) \ = \ \int_0^x u^{a - 1} \, (1 - u)^{b - 1} \,
+du, \end{equation*}\]</span> where <span class="math inline">\(\text{B}(a, b)\)</span> is the beta function defined in <a href="#beta-appendix">appendix</a>. If <span class="math inline">\(x = 1\)</span>, the incomplete beta function reduces to the beta function, <span class="math inline">\(\text{B}(1; a, b) = \text{B}(a, b)\)</span>.</p>
+<p>The regularized incomplete beta function divides the incomplete beta function by the beta function, <span class="math display">\[\begin{equation*} I_x(a, b) \ = \ \frac{\text{B}(x; \,
+a, b)}{B(a, b)} \, . \end{equation*}\]</span></p>
+</section>
+<section id="gamma-appendix" class="level2">
+<h2 class="anchored" data-anchor-id="gamma-appendix">Gamma</h2>
+<p>The gamma function, <span class="math inline">\(\Gamma(x)\)</span>, is the generalization of the factorial function to continuous variables, defined so that for positive integers <span class="math inline">\(n\)</span>, <span class="math display">\[\begin{equation*} \Gamma(n+1) = n! \end{equation*}\]</span> Generalizing to all positive numbers and non-integer negative numbers, <span class="math display">\[\begin{equation*} \Gamma(x) =
+\int_0^{\infty} u^{x - 1} \exp(-u) \, du. \end{equation*}\]</span></p>
+</section>
+<section id="digamma-appendix" class="level2">
+<h2 class="anchored" data-anchor-id="digamma-appendix">Digamma</h2>
+<p>The digamma function <span class="math inline">\(\Psi\)</span> is the derivative of the <span class="math inline">\(\log \Gamma\)</span> function, <span class="math display">\[\begin{equation*}
+\Psi(u) \ = \ \frac{d}{d u} \log \Gamma(u) \ = \ \frac{1}{\Gamma(u)} \ \frac{d}{d u} \Gamma(u).
+\end{equation*}\]</span></p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/embedded_laplace.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Embedded Laplace Approximation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/functions_index.html" class="pagination-link" aria-label="Alphabetical Index">
+        <span class="nav-page-text">Alphabetical Index</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mathematical_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/matrix_operations.html b/docs/2_39/functions-reference/matrix_operations.html
new file mode 100644
index 000000000..40457a087
--- /dev/null
+++ b/docs/2_39/functions-reference/matrix_operations.html
@@ -0,0 +1,2681 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Matrix Operations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/complex_matrix_operations.html" rel="next">
+<link href="../functions-reference/array_operations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Matrix Operations">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/matrix_operations.html">Matrix Operations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#matrix-operations" id="toc-matrix-operations" class="nav-link active" data-scroll-target="#matrix-operations">Matrix Operations</a>
+  <ul class="collapse">
+  <li><a href="#integer-valued-matrix-size-functions" id="toc-integer-valued-matrix-size-functions" class="nav-link" data-scroll-target="#integer-valued-matrix-size-functions">Integer-valued matrix size functions</a></li>
+  <li><a href="#matrix-arithmetic-operators" id="toc-matrix-arithmetic-operators" class="nav-link" data-scroll-target="#matrix-arithmetic-operators">Matrix arithmetic operators</a>
+  <ul class="collapse">
+  <li><a href="#negation-prefix-operators" id="toc-negation-prefix-operators" class="nav-link" data-scroll-target="#negation-prefix-operators">Negation prefix operators</a></li>
+  <li><a href="#infix-matrix-operators" id="toc-infix-matrix-operators" class="nav-link" data-scroll-target="#infix-matrix-operators">Infix matrix operators</a></li>
+  <li><a href="#broadcast-infix-operators" id="toc-broadcast-infix-operators" class="nav-link" data-scroll-target="#broadcast-infix-operators">Broadcast infix operators</a></li>
+  </ul></li>
+  <li><a href="#transposition-operator" id="toc-transposition-operator" class="nav-link" data-scroll-target="#transposition-operator">Transposition operator</a></li>
+  <li><a href="#elementwise-functions" id="toc-elementwise-functions" class="nav-link" data-scroll-target="#elementwise-functions">Elementwise functions</a></li>
+  <li><a href="#dot-products-and-specialized-products" id="toc-dot-products-and-specialized-products" class="nav-link" data-scroll-target="#dot-products-and-specialized-products">Dot products and specialized products</a>
+  <ul class="collapse">
+  <li><a href="#specialized-products" id="toc-specialized-products" class="nav-link" data-scroll-target="#specialized-products">Specialized products</a></li>
+  </ul></li>
+  <li><a href="#reductions" id="toc-reductions" class="nav-link" data-scroll-target="#reductions">Reductions</a>
+  <ul class="collapse">
+  <li><a href="#log-sum-of-exponents" id="toc-log-sum-of-exponents" class="nav-link" data-scroll-target="#log-sum-of-exponents">Log sum of exponents</a></li>
+  <li><a href="#minimum-and-maximum" id="toc-minimum-and-maximum" class="nav-link" data-scroll-target="#minimum-and-maximum">Minimum and maximum</a></li>
+  <li><a href="#sums-and-products" id="toc-sums-and-products" class="nav-link" data-scroll-target="#sums-and-products">Sums and products</a></li>
+  <li><a href="#sample-moments" id="toc-sample-moments" class="nav-link" data-scroll-target="#sample-moments">Sample moments</a></li>
+  <li><a href="#quantile" id="toc-quantile" class="nav-link" data-scroll-target="#quantile">Quantile</a></li>
+  </ul></li>
+  <li><a href="#matrix-broadcast" id="toc-matrix-broadcast" class="nav-link" data-scroll-target="#matrix-broadcast">Broadcast functions</a>
+  <ul class="collapse">
+  <li><a href="#symmetrization" id="toc-symmetrization" class="nav-link" data-scroll-target="#symmetrization">Symmetrization</a></li>
+  </ul></li>
+  <li><a href="#diagonal-matrix-functions" id="toc-diagonal-matrix-functions" class="nav-link" data-scroll-target="#diagonal-matrix-functions">Diagonal matrix functions</a></li>
+  <li><a href="#container-construction" id="toc-container-construction" class="nav-link" data-scroll-target="#container-construction">Container construction functions</a></li>
+  <li><a href="#slicing-and-blocking-functions" id="toc-slicing-and-blocking-functions" class="nav-link" data-scroll-target="#slicing-and-blocking-functions">Slicing and blocking functions</a>
+  <ul class="collapse">
+  <li><a href="#columns-and-rows" id="toc-columns-and-rows" class="nav-link" data-scroll-target="#columns-and-rows">Columns and rows</a></li>
+  <li><a href="#block-operations" id="toc-block-operations" class="nav-link" data-scroll-target="#block-operations">Block operations</a></li>
+  </ul></li>
+  <li><a href="#matrix-concatenation" id="toc-matrix-concatenation" class="nav-link" data-scroll-target="#matrix-concatenation">Matrix and vector concatenation</a></li>
+  <li><a href="#softmax" id="toc-softmax" class="nav-link" data-scroll-target="#softmax">Special matrix functions</a>
+  <ul class="collapse">
+  <li><a href="#softmax-1" id="toc-softmax-1" class="nav-link" data-scroll-target="#softmax-1">Softmax</a></li>
+  <li><a href="#cumulative-sums" id="toc-cumulative-sums" class="nav-link" data-scroll-target="#cumulative-sums">Cumulative sums</a></li>
+  </ul></li>
+  <li><a href="#gaussian-process-covariance-functions" id="toc-gaussian-process-covariance-functions" class="nav-link" data-scroll-target="#gaussian-process-covariance-functions">Gaussian Process Covariance Functions</a>
+  <ul class="collapse">
+  <li><a href="#exponentiated-quadratic-kernel" id="toc-exponentiated-quadratic-kernel" class="nav-link" data-scroll-target="#exponentiated-quadratic-kernel">Exponentiated quadratic kernel</a></li>
+  <li><a href="#dot-product-kernel" id="toc-dot-product-kernel" class="nav-link" data-scroll-target="#dot-product-kernel">Dot product kernel</a></li>
+  <li><a href="#exponential-kernel" id="toc-exponential-kernel" class="nav-link" data-scroll-target="#exponential-kernel">Exponential kernel</a></li>
+  <li><a href="#matern-32-kernel" id="toc-matern-32-kernel" class="nav-link" data-scroll-target="#matern-32-kernel">Matern 3/2 kernel</a></li>
+  <li><a href="#matern-52-kernel" id="toc-matern-52-kernel" class="nav-link" data-scroll-target="#matern-52-kernel">Matern 5/2 kernel</a></li>
+  <li><a href="#periodic-kernel" id="toc-periodic-kernel" class="nav-link" data-scroll-target="#periodic-kernel">Periodic kernel</a></li>
+  </ul></li>
+  <li><a href="#linear-algebra-functions-and-solvers" id="toc-linear-algebra-functions-and-solvers" class="nav-link" data-scroll-target="#linear-algebra-functions-and-solvers">Linear algebra functions and solvers</a>
+  <ul class="collapse">
+  <li><a href="#matrix-division-operators-and-functions" id="toc-matrix-division-operators-and-functions" class="nav-link" data-scroll-target="#matrix-division-operators-and-functions">Matrix division operators and functions</a></li>
+  <li><a href="#symmetric-positive-definite-matrix-division-functions" id="toc-symmetric-positive-definite-matrix-division-functions" class="nav-link" data-scroll-target="#symmetric-positive-definite-matrix-division-functions">Symmetric positive-definite matrix division functions</a></li>
+  <li><a href="#matrix-exponential" id="toc-matrix-exponential" class="nav-link" data-scroll-target="#matrix-exponential">Matrix exponential</a></li>
+  <li><a href="#matrix-power" id="toc-matrix-power" class="nav-link" data-scroll-target="#matrix-power">Matrix power</a></li>
+  <li><a href="#linear-algebra-functions" id="toc-linear-algebra-functions" class="nav-link" data-scroll-target="#linear-algebra-functions">Linear algebra functions</a></li>
+  </ul></li>
+  <li><a href="#sort-functions" id="toc-sort-functions" class="nav-link" data-scroll-target="#sort-functions">Sort functions</a></li>
+  <li><a href="#reverse-functions" id="toc-reverse-functions" class="nav-link" data-scroll-target="#reverse-functions">Reverse functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/matrix_operations.html">Matrix Operations</a></li></ol></nav></header>
+
+
+
+
+<section id="matrix-operations" class="level1">
+<h1>Matrix Operations</h1>
+<section id="integer-valued-matrix-size-functions" class="level2">
+<h2 class="anchored" data-anchor-id="integer-valued-matrix-size-functions">Integer-valued matrix size functions</h2>
+<a id="index-entry-9173ff057d7657581e5abf426a39637cb2339c11" class="index-entry"></a> <!-- int; num_elements; (vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(vector x)</code><br> The total number of elements in the vector x (same as function <code>rows</code>)</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-25d1da3aed0f07d9c6a36b34c0d238d180ebc272" class="index-entry"></a> <!-- int; num_elements; (row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(row_vector x)</code><br> The total number of elements in the vector x (same as function <code>cols</code>)</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-77dd48e13c18141a06bb49ceb67e01b76c4e3fae" class="index-entry"></a> <!-- int; num_elements; (matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#num_elements" class="unlink" title="Jump to index entry"><code>num_elements</code></a></strong><code>(matrix x)</code><br> The total number of elements in the matrix x. For example, if <code>x</code> is a <span class="math inline">\(5 \times 3\)</span> matrix, then <code>num_elements(x)</code> is 15</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-c6697d2988eb22bf8fd56b05ffcde6779627ebf8" class="index-entry"></a> <!-- int; rows; (vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rows" class="unlink" title="Jump to index entry"><code>rows</code></a></strong><code>(vector x)</code><br> The number of rows in the vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-578ea4c98303c7c7498cb69c853a18e09d5c44fd" class="index-entry"></a> <!-- int; rows; (row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rows" class="unlink" title="Jump to index entry"><code>rows</code></a></strong><code>(row_vector x)</code><br> The number of rows in the row vector x, namely 1</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-65ffc1cace11a1dd2675fb84a331101e1ea8b626" class="index-entry"></a> <!-- int; rows; (matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rows" class="unlink" title="Jump to index entry"><code>rows</code></a></strong><code>(matrix x)</code><br> The number of rows in the matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-8113ac540f4cc3aa182c3afcfa5d1665778f7fd9" class="index-entry"></a> <!-- int; cols; (vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#cols" class="unlink" title="Jump to index entry"><code>cols</code></a></strong><code>(vector x)</code><br> The number of columns in the vector x, namely 1</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-417aeef6881030f640a4ea077e02335739bf0e3e" class="index-entry"></a> <!-- int; cols; (row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#cols" class="unlink" title="Jump to index entry"><code>cols</code></a></strong><code>(row_vector x)</code><br> The number of columns in the row vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f399f6ffab24708be4ff867cbaef541158fc00bf" class="index-entry"></a> <!-- int; cols; (matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#cols" class="unlink" title="Jump to index entry"><code>cols</code></a></strong><code>(matrix x)</code><br> The number of columns in the matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9b99cb12fadc8373556db600c949cc44b43fe907" class="index-entry"></a> <!-- int; size; (vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(vector x)</code><br> The size of <code>x</code>, i.e., the number of elements</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-f449c3a25c74efefc2c1462cf914a52a4050d575" class="index-entry"></a> <!-- int; size; (row_vector x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(row_vector x)</code><br> The size of <code>x</code>, i.e., the number of elements</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-69098855a034f282c0294e1c6c21cded79ec8a3c" class="index-entry"></a> <!-- int; size; (matrix x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#size" class="unlink" title="Jump to index entry"><code>size</code></a></strong><code>(matrix x)</code><br> The size of the matrix <code>x</code>. For example, if <code>x</code> is a <span class="math inline">\(5 \times 3\)</span> matrix, then <code>size(x)</code> is 15</p>
+<small><i>Available since 2.26</i></small>
+</section>
+<section id="matrix-arithmetic-operators" class="level2">
+<h2 class="anchored" data-anchor-id="matrix-arithmetic-operators">Matrix arithmetic operators</h2>
+<p>Stan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.</p>
+<section id="negation-prefix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="negation-prefix-operators">Negation prefix operators</h3>
+<a id="index-entry-8339baafb56b84ffc02ddc234752ab1f7a46e1e3" class="index-entry"></a> <!-- vector; operator-; (vector x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(vector x)</code><br> The negation of the vector x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7fa245ebd36b8195ee47b5cf42158c3936625d93" class="index-entry"></a> <!-- row_vector; operator-; (row_vector x); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(row_vector x)</code><br> The negation of the row vector x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6ca73a0303dd24f82990cf099176d76ea4f50d3b" class="index-entry"></a> <!-- matrix; operator-; (matrix x); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(matrix x)</code><br> The negation of the matrix x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f" class="index-entry"></a> <!-- T; operator-; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(T x)</code><br> Vectorized version of <code>operator-</code>. If <code>T x</code> is a (possibly nested) array of matrix types, <code>-x</code> is the same shape array where each individual value is negated.</p>
+<small><i>Available since 2.31</i></small>
+</section>
+<section id="infix-matrix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="infix-matrix-operators">Infix matrix operators</h3>
+<a id="index-entry-1a0c872aa1e302e3fd5a2a167b97e3ca6719d10c" class="index-entry"></a> <!-- vector; operator+; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(vector x, vector y)</code><br> The sum of the vectors x and y.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-33e5719f2f10f7ff91219c35213757ec24c6b306" class="index-entry"></a> <!-- row_vector; operator+; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(row_vector x, row_vector y)</code><br> The sum of the row vectors x and y.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-718549a6ddfc2c775f69b8bc7a28fc7c053a7fbe" class="index-entry"></a> <!-- matrix; operator+; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(matrix x, matrix y)</code><br> The sum of the matrices x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-8b1168c1cebe948416436c1c61cb7d7210949fae" class="index-entry"></a> <!-- vector; operator-; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(vector x, vector y)</code><br> The difference between the vectors x and y.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f9f5c0ef9c18e80c24aa789a90b7d1253096fa3a" class="index-entry"></a> <!-- row_vector; operator-; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(row_vector x, row_vector y)</code><br> The difference between the row vectors x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b9ba1d02d3a73b7e8733514fb6b3a1c59414ee37" class="index-entry"></a> <!-- matrix; operator-; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(matrix x, matrix y)</code><br> The difference between the matrices x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-af34e073da12073d94a55fde49701a8f65786b67" class="index-entry"></a> <!-- vector; operator*; (real x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(real x, vector y)</code><br> The product of the scalar x and vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-84a0cd07fdb794ec75d40dfa3c1638586ce59f1a" class="index-entry"></a> <!-- row_vector; operator*; (real x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(real x, row_vector y)</code><br> The product of the scalar x and the row vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-662e0b2447d3b372fc219293e38410d49bad1b7b" class="index-entry"></a> <!-- matrix; operator*; (real x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(real x, matrix y)</code><br> The product of the scalar x and the matrix y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-29c74f1694c2364957ad8d79e09cfb3ad47e775b" class="index-entry"></a> <!-- vector; operator*; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(vector x, real y)</code><br> The product of the scalar y and vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-3669aca452e27a8b48f0358b3d3103bc4d0d9530" class="index-entry"></a> <!-- matrix; operator*; (vector x, row_vector y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(vector x, row_vector y)</code><br> The product of the vector x and row vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-852b4c73c95c85062105d4a058743e342d831c9a" class="index-entry"></a> <!-- row_vector; operator*; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(row_vector x, real y)</code><br> The product of the scalar y and row vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c64d64dd72ff726168c121c89601f1e1fa602276" class="index-entry"></a> <!-- real; operator*; (row_vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(row_vector x, vector y)</code><br> The product of the row vector x and vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0d741e9840edb32e9575051107190f0632ec5dd3" class="index-entry"></a> <!-- row_vector; operator*; (row_vector x, matrix y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(row_vector x, matrix y)</code><br> The product of the row vector x and matrix y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7c526c1ddbd012a188345676d03d7d8e52a2f4af" class="index-entry"></a> <!-- matrix; operator*; (matrix x, real y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(matrix x, real y)</code><br> The product of the scalar y and matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c61d79dac022af07a0f553fb494511099410cdac" class="index-entry"></a> <!-- vector; operator*; (matrix x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(matrix x, vector y)</code><br> The product of the matrix x and vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a87089afcc4f8cc865a9447f93284dd4a7903eaf" class="index-entry"></a> <!-- matrix; operator*; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(matrix x, matrix y)</code><br> The product of the matrices x and y</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="broadcast-infix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="broadcast-infix-operators">Broadcast infix operators</h3>
+<a id="index-entry-925b4d95d12aaad5ae3a3d5a5b8d06d9f36c3b9b" class="index-entry"></a> <!-- vector; operator+; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(vector x, real y)</code><br> The result of adding y to every entry in the vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f9354975ff6bc2974a4d1661855cea524ca49b50" class="index-entry"></a> <!-- vector; operator+; (real x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(real x, vector y)</code><br> The result of adding x to every entry in the vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-26044cd5506e5c4fc2579fc6466dd3f81d6408ce" class="index-entry"></a> <!-- row_vector; operator+; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(row_vector x, real y)</code><br> The result of adding y to every entry in the row vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-fd064b77a97b1ec9168687d5a25cde11da015591" class="index-entry"></a> <!-- row_vector; operator+; (real x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(real x, row_vector y)</code><br> The result of adding x to every entry in the row vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-77931799ac0a444f9530777cce44cb987abf32cb" class="index-entry"></a> <!-- matrix; operator+; (matrix x, real y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(matrix x, real y)</code><br> The result of adding y to every entry in the matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0465a61949555ef4ef44514ccba8174500deeefa" class="index-entry"></a> <!-- matrix; operator+; (real x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(real x, matrix y)</code><br> The result of adding x to every entry in the matrix y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0dea3575356afb28b43da9a9985361d5ba7e1c84" class="index-entry"></a> <!-- vector; operator-; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(vector x, real y)</code><br> The result of subtracting y from every entry in the vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-76b3ba43bb249ac182c9f4fea095ea81ede76ebd" class="index-entry"></a> <!-- vector; operator-; (real x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(real x, vector y)</code><br> The result of adding x to every entry in the negation of the vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0e7a932586415d9249383de3efe664eb105f4a4f" class="index-entry"></a> <!-- row_vector; operator-; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(row_vector x, real y)</code><br> The result of subtracting y from every entry in the row vector x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6002a07176762b3c481c6761e8e06b1294e8d625" class="index-entry"></a> <!-- row_vector; operator-; (real x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(real x, row_vector y)</code><br> The result of adding x to every entry in the negation of the row vector y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-88db741d085202e8824acf87fdf5652fbc176914" class="index-entry"></a> <!-- matrix; operator-; (matrix x, real y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(matrix x, real y)</code><br> The result of subtracting y from every entry in the matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-aa6b0f0e0b80d8bcc4d95c623445e021142a9439" class="index-entry"></a> <!-- matrix; operator-; (real x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(real x, matrix y)</code><br> The result of adding x to every entry in negation of the matrix y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-31c5a97eb5a6aa036118bd60ae51ce6f8d86c465" class="index-entry"></a> <!-- vector; operator/; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(vector x, real y)</code><br> The result of dividing each entry in the vector x by y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6ad75a35e6fb68d75c09604b3945b20cc0dabb1e" class="index-entry"></a> <!-- row_vector; operator/; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(row_vector x, real y)</code><br> The result of dividing each entry in the row vector x by y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-dc385d6a823b14919aaa86e64b873846b31b422b" class="index-entry"></a> <!-- matrix; operator/; (matrix x, real y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(matrix x, real y)</code><br> The result of dividing each entry in the matrix x by y</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="transposition-operator" class="level2">
+<h2 class="anchored" data-anchor-id="transposition-operator">Transposition operator</h2>
+<p>Matrix transposition is represented using a postfix operator.</p>
+<a id="index-entry-794bf86d8a081a75147a013481f07740310bb0d8" class="index-entry"></a> <!-- matrix; operator'; (matrix x); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator'" class="unlink" title="Jump to index entry"><code>operator'</code></a></strong><code>(matrix x)</code><br> The transpose of the matrix x, written as <code>x'</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-5b05ea20fccc5c44353c7f5813ec3ce694b07cb6" class="index-entry"></a> <!-- row_vector; operator'; (vector x); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator'" class="unlink" title="Jump to index entry"><code>operator'</code></a></strong><code>(vector x)</code><br> The transpose of the vector x, written as <code>x'</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-e3432686a771bad98175ab4a594d8156ab915492" class="index-entry"></a> <!-- vector; operator'; (row_vector x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator'" class="unlink" title="Jump to index entry"><code>operator'</code></a></strong><code>(row_vector x)</code><br> The transpose of the row vector x, written as <code>x'</code></p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="elementwise-functions" class="level2">
+<h2 class="anchored" data-anchor-id="elementwise-functions">Elementwise functions</h2>
+<p>Elementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section <a href="../functions-reference/real-valued_basic_functions.html#fun-vectorization">function vectorization</a> for the general cases.</p>
+<a id="index-entry-f6b9c926ae8c20a429454fbfd788a05c447ca21f" class="index-entry"></a> <!-- vector; operator.*; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.\*" class="unlink" title="Jump to index entry"><code>operator.*</code></a></strong><code>(vector x, vector y)</code><br> The elementwise product of y and x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-296aa89afe5256f9cb944c3ce9458197bd833313" class="index-entry"></a> <!-- row_vector; operator.*; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.\*" class="unlink" title="Jump to index entry"><code>operator.*</code></a></strong><code>(row_vector x, row_vector y)</code><br> The elementwise product of y and x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6dec717a6b41050c89af60556ec7baae6bb8bb75" class="index-entry"></a> <!-- matrix; operator.*; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.\*" class="unlink" title="Jump to index entry"><code>operator.*</code></a></strong><code>(matrix x, matrix y)</code><br> The elementwise product of y and x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-65a45dfeaa613598d736e8e631a8900b78e3c9ff" class="index-entry"></a> <!-- vector; operator./; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(vector x, vector y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-66caea638b41f99f47c90fbf9f08a165f7bc312b" class="index-entry"></a> <!-- vector; operator./; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(vector x, real y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-1800a4c4e6451dd14c0759f41c3b260e10010627" class="index-entry"></a> <!-- vector; operator./; (real x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(real x, vector y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-c9b71dbef4490527107a1dcfcd38f77b374c74d5" class="index-entry"></a> <!-- row_vector; operator./; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(row_vector x, row_vector y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-dcdaa0d007eeecd45a7d5ac2cf47b91d52c07fc8" class="index-entry"></a> <!-- row_vector; operator./; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(row_vector x, real y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-c8cde8c39dee0b427f9770e8e63ece827cebac04" class="index-entry"></a> <!-- row_vector; operator./; (real x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(real x, row_vector y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-63e3151ce39b028bb15225d8305e8ae59d1d2b9c" class="index-entry"></a> <!-- matrix; operator./; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(matrix x, matrix y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-33b751b1c6fcaa79a1f249d90774c58df3a3146f" class="index-entry"></a> <!-- matrix; operator./; (matrix x, real y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(matrix x, real y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-ed3f9fc9cd17f66b01b08112c1a1c0eb063dfbc7" class="index-entry"></a> <!-- matrix; operator./; (real x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator./" class="unlink" title="Jump to index entry"><code>operator./</code></a></strong><code>(real x, matrix y)</code><br> The elementwise quotient of y and x</p>
+<small><i>Available since 2.4</i></small>
+<a id="index-entry-71f690034947993173f537ecaa9c8ee8a6e8c1e8" class="index-entry"></a> <!-- vector; operator.^; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(vector x, vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-9f9fe3db8b2a57c2c414d4efe2eb5aa93b6b86e4" class="index-entry"></a> <!-- vector; operator.^; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(vector x, real y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-396441e397b74d06863454c78d4b5af50dda330f" class="index-entry"></a> <!-- vector; operator.^; (real x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(real x, vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-a4fdb1f768269d25cc5aab5dafbe0535c73ee226" class="index-entry"></a> <!-- row_vector; operator.^; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(row_vector x, row_vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-257383daeef20f4a85bd2e23e4f3c9c2ecfb013d" class="index-entry"></a> <!-- row_vector; operator.^; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(row_vector x, real y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-ea70cfe8f041b8cdd5f8cfcc99356a42ca783deb" class="index-entry"></a> <!-- row_vector; operator.^; (real x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(real x, row_vector y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-ee33dd887a5da3fafc96e59aa3cd75c6d5051cea" class="index-entry"></a> <!-- matrix; operator.^; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(matrix x, matrix y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-1d5f88db2e08d98185f4bdad36ccdeb289d1a58c" class="index-entry"></a> <!-- matrix; operator.^; (matrix x, real y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(matrix x, real y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-8955da8055e2b2a58a20eac3964a35cdf330b786" class="index-entry"></a> <!-- matrix; operator.^; (real x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator.^" class="unlink" title="Jump to index entry"><code>operator.^</code></a></strong><code>(real x, matrix y)</code><br> The elementwise power of y and x</p>
+<small><i>Available since 2.24</i></small>
+</section>
+<section id="dot-products-and-specialized-products" class="level2">
+<h2 class="anchored" data-anchor-id="dot-products-and-specialized-products">Dot products and specialized products</h2>
+<a id="index-entry-6f3cb76e0af87d28bceef319e85e75976645662e" class="index-entry"></a> <!-- real; dot_product; (vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(vector x, vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4ce3d3846b8c242984aca259fd01579e1b2179af" class="index-entry"></a> <!-- real; dot_product; (vector x, row_vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(vector x, row_vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-5624c84e1a33f2cf540a3d8a2994205918d9915c" class="index-entry"></a> <!-- real; dot_product; (row_vector x, vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(row_vector x, vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-5fa1a043fd0dcebec0cb14fd09c1a965c33d37fe" class="index-entry"></a> <!-- real; dot_product; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dot_product" class="unlink" title="Jump to index entry"><code>dot_product</code></a></strong><code>(row_vector x, row_vector y)</code><br> The dot product of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7b3a4d9f9236657fd6d5970b8840e4a2e7e176bf" class="index-entry"></a> <!-- row_vector; columns_dot_product; (vector x, vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_product" class="unlink" title="Jump to index entry"><code>columns_dot_product</code></a></strong><code>(vector x, vector y)</code><br> The dot product of the columns of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4758c18c6242ed86d28834ea3c356de1d5e22b86" class="index-entry"></a> <!-- row_vector; columns_dot_product; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_product" class="unlink" title="Jump to index entry"><code>columns_dot_product</code></a></strong><code>(row_vector x, row_vector y)</code><br> The dot product of the columns of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-758593f0bee7c0421df7fc503237288851879118" class="index-entry"></a> <!-- row_vector; columns_dot_product; (matrix x, matrix y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_product" class="unlink" title="Jump to index entry"><code>columns_dot_product</code></a></strong><code>(matrix x, matrix y)</code><br> The dot product of the columns of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-de1fe07c43494dac868152a24e25fcf81955732e" class="index-entry"></a> <!-- vector; rows_dot_product; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_product" class="unlink" title="Jump to index entry"><code>rows_dot_product</code></a></strong><code>(vector x, vector y)</code><br> The dot product of the rows of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9bb7b3a8ee495648abc1b0eff2ecebe52c99be25" class="index-entry"></a> <!-- vector; rows_dot_product; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_product" class="unlink" title="Jump to index entry"><code>rows_dot_product</code></a></strong><code>(row_vector x, row_vector y)</code><br> The dot product of the rows of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-cf6ee59a255f43dac007f4dd3ef01a1119343cb0" class="index-entry"></a> <!-- vector; rows_dot_product; (matrix x, matrix y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_product" class="unlink" title="Jump to index entry"><code>rows_dot_product</code></a></strong><code>(matrix x, matrix y)</code><br> The dot product of the rows of x and y</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-1ae4fbadff5f8d76241c1cfc33e2c36db7decc73" class="index-entry"></a> <!-- real; dot_self; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dot_self" class="unlink" title="Jump to index entry"><code>dot_self</code></a></strong><code>(vector x)</code><br> The dot product of the vector x with itself</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-806eb76bdef36d3ee3a9a009027593ad32d755f1" class="index-entry"></a> <!-- real; dot_self; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dot_self" class="unlink" title="Jump to index entry"><code>dot_self</code></a></strong><code>(row_vector x)</code><br> The dot product of the row vector x with itself</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f0c5a660d69d25748bdafa9cda98ee0cfa190cbe" class="index-entry"></a> <!-- row_vector; columns_dot_self; (vector x); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_self" class="unlink" title="Jump to index entry"><code>columns_dot_self</code></a></strong><code>(vector x)</code><br> The dot product of the columns of x with themselves</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-258f64fdc22d8a6e1012cfa375d94222962c3add" class="index-entry"></a> <!-- row_vector; columns_dot_self; (row_vector x); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_self" class="unlink" title="Jump to index entry"><code>columns_dot_self</code></a></strong><code>(row_vector x)</code><br> The dot product of the columns of x with themselves</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-664db2ea209dd84f3f62d813817a0be019ff3afb" class="index-entry"></a> <!-- row_vector; columns_dot_self; (matrix x); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#columns_dot_self" class="unlink" title="Jump to index entry"><code>columns_dot_self</code></a></strong><code>(matrix x)</code><br> The dot product of the columns of x with themselves</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a451434f24e97a590c47f3f0e691af26a61e8705" class="index-entry"></a> <!-- vector; rows_dot_self; (vector x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_self" class="unlink" title="Jump to index entry"><code>rows_dot_self</code></a></strong><code>(vector x)</code><br> The dot product of the rows of x with themselves</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b624c89d6c4d2e43359ff78346c37076b91730f1" class="index-entry"></a> <!-- vector; rows_dot_self; (row_vector x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_self" class="unlink" title="Jump to index entry"><code>rows_dot_self</code></a></strong><code>(row_vector x)</code><br> The dot product of the rows of x with themselves</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-947ee99ec791f7bead5c155ced2ef89e5ac7e8b7" class="index-entry"></a> <!-- vector; rows_dot_self; (matrix x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rows_dot_self" class="unlink" title="Jump to index entry"><code>rows_dot_self</code></a></strong><code>(matrix x)</code><br> The dot product of the rows of x with themselves</p>
+<small><i>Available since 2.0</i></small>
+<section id="specialized-products" class="level3">
+<h3 class="anchored" data-anchor-id="specialized-products">Specialized products</h3>
+<a id="index-entry-0debd0686ed364955c554407a48932c2cc0e1ecf" class="index-entry"></a> <!-- matrix; tcrossprod; (matrix x); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#tcrossprod" class="unlink" title="Jump to index entry"><code>tcrossprod</code></a></strong><code>(matrix x)</code><br> The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix <span class="math inline">\(\text{x}\,\text{x}^{\top}\)</span>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c5372c9d5715e11b5d87a465d6d671d608ef9513" class="index-entry"></a> <!-- matrix; crossprod; (matrix x); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#crossprod" class="unlink" title="Jump to index entry"><code>crossprod</code></a></strong><code>(matrix x)</code><br> The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix <span class="math inline">\(\text{x}^{\top}\,\text{x}\)</span>.</p>
+<small><i>Available since 2.0</i></small>
+<p>The following functions all provide shorthand forms for common expressions, which are also much more efficient.</p>
+<a id="index-entry-0cac546f1a4dfcfd8676f091dc0f7ab8cf808e4d" class="index-entry"></a> <!-- matrix; quad_form; (matrix A, matrix B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#quad_form" class="unlink" title="Jump to index entry"><code>quad_form</code></a></strong><code>(matrix A, matrix B)</code><br> The quadratic form, i.e., <code>B' * A * B</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7d865d93151dbb42928e39384a33278c4695d490" class="index-entry"></a> <!-- real; quad_form; (matrix A, vector B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#quad_form" class="unlink" title="Jump to index entry"><code>quad_form</code></a></strong><code>(matrix A, vector B)</code><br> The quadratic form, i.e., <code>B' * A * B</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7d5c708ffbdc829d13ee747d32cdeb5f6b5c5976" class="index-entry"></a> <!-- matrix; quad_form_diag; (matrix m, vector v); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#quad_form_diag" class="unlink" title="Jump to index entry"><code>quad_form_diag</code></a></strong><code>(matrix m, vector v)</code><br> The quadratic form using the column vector v as a diagonal matrix, i.e., <code>diag_matrix(v) * m * diag_matrix(v)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-4c9654c4a3b4068858d9351459b026a697f02b8e" class="index-entry"></a> <!-- matrix; quad_form_diag; (matrix m, row_vector rv); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#quad_form_diag" class="unlink" title="Jump to index entry"><code>quad_form_diag</code></a></strong><code>(matrix m, row_vector rv)</code><br> The quadratic form using the row vector rv as a diagonal matrix, i.e., <code>diag_matrix(rv) * m * diag_matrix(rv)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-cb6d97ec303631d01e1fbf4737567987878d14b2" class="index-entry"></a> <!-- matrix; quad_form_sym; (matrix A, matrix B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#quad_form_sym" class="unlink" title="Jump to index entry"><code>quad_form_sym</code></a></strong><code>(matrix A, matrix B)</code><br> Similarly to quad_form, gives <code>B' * A * B</code>, but additionally checks if A is symmetric and ensures that the result is also symmetric.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-362d5553d7ff140b2163b35fed3667b17ad4c3d3" class="index-entry"></a> <!-- real; quad_form_sym; (matrix A, vector B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#quad_form_sym" class="unlink" title="Jump to index entry"><code>quad_form_sym</code></a></strong><code>(matrix A, vector B)</code><br> Similarly to quad_form, gives <code>B' * A * B</code>, but additionally checks if A is symmetric and ensures that the result is also symmetric.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-b067f1dc33e4e60522d490bc635d4305d0e9e7fb" class="index-entry"></a> <!-- real; trace_dot; (matrix A, matrix B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace_dot" class="unlink" title="Jump to index entry"><code>trace_dot</code></a></strong><code>(matrix A, matrix B)</code><br> The trace of the matrix product, i.e., <code>trace(A * B)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-9e673dba39e3d7e6297d26eda1aaa01ff737cd89" class="index-entry"></a> <!-- real; trace_quad_form; (matrix A, matrix B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace_quad_form" class="unlink" title="Jump to index entry"><code>trace_quad_form</code></a></strong><code>(matrix A, matrix B)</code><br> The trace of the quadratic form, i.e., <code>trace(B' * A * B)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4102366e38a8c1eb82e650b1867da9a37385278a" class="index-entry"></a> <!-- real; trace_quad_form; (matrix A, vector B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace_quad_form" class="unlink" title="Jump to index entry"><code>trace_quad_form</code></a></strong><code>(matrix A, vector B)</code><br> The trace of the quadratic form, i.e., <code>trace(B' * A * B)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-725c7c07b916a247f645ce441212b4a82778b096" class="index-entry"></a> <!-- real; trace_gen_quad_form; (matrix D ,matrix A, matrix B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace_gen_quad_form" class="unlink" title="Jump to index entry"><code>trace_gen_quad_form</code></a></strong><code>(matrix D, matrix A, matrix B)</code><br> The trace of a generalized quadratic form, i.e., <code>trace(D * B' * A * B).</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-074a26dceab0e18f9117c1bd04f0818b1c303de2" class="index-entry"></a> <!-- matrix; multiply_lower_tri_self_transpose; (matrix x); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#multiply_lower_tri_self_transpose" class="unlink" title="Jump to index entry"><code>multiply_lower_tri_self_transpose</code></a></strong><code>(matrix x)</code><br> The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if <code>L</code> is a matrix of the same dimensions as x with <code>L(m,n)</code> equal to <code>x(m,n)</code> for <span class="math inline">\(\text{n}
+\leq \text{m}\)</span> and <code>L(m,n)</code> equal to 0 if <span class="math inline">\(\text{n} &gt; \text{m}\)</span>, the result is the symmetric matrix <span class="math inline">\(\text{L}\,\text{L}^{\top}\)</span>. This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0fc4d71e7bc087ce4f9dec77e1df8df248f41bac" class="index-entry"></a> <!-- matrix; diag_pre_multiply; (vector v, matrix m); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_pre_multiply" class="unlink" title="Jump to index entry"><code>diag_pre_multiply</code></a></strong><code>(vector v, matrix m)</code><br> Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., <code>diag_matrix(v) * m</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-87b1033ce509f3990ef4516c29c0b9b08d67c37b" class="index-entry"></a> <!-- matrix; diag_pre_multiply; (row_vector rv, matrix m); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_pre_multiply" class="unlink" title="Jump to index entry"><code>diag_pre_multiply</code></a></strong><code>(row_vector rv, matrix m)</code><br> Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., <code>diag_matrix(rv) * m</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2ff483d0f2d35ce6055a030912e1909c110453dc" class="index-entry"></a> <!-- matrix; diag_post_multiply; (matrix m, vector v); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_post_multiply" class="unlink" title="Jump to index entry"><code>diag_post_multiply</code></a></strong><code>(matrix m, vector v)</code><br> Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., <code>m * diag_matrix(v)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-fedf6c86ebf0663043f5e986e077701e6ae185f4" class="index-entry"></a> <!-- matrix; diag_post_multiply; (matrix m, row_vector rv); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_post_multiply" class="unlink" title="Jump to index entry"><code>diag_post_multiply</code></a></strong><code>(matrix m, row_vector rv)</code><br> Return the product of the matrix <code>m</code> and the diagonal matrix formed from the the row vector <code>rv</code>, i.e., <code>m * diag_matrix(rv)</code>.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="reductions" class="level2">
+<h2 class="anchored" data-anchor-id="reductions">Reductions</h2>
+<section id="log-sum-of-exponents" class="level3">
+<h3 class="anchored" data-anchor-id="log-sum-of-exponents">Log sum of exponents</h3>
+<a id="index-entry-12d207b1c215fc3ea15553df653cae3cbb14fbdf" class="index-entry"></a> <!-- real; log_sum_exp; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_sum_exp" class="unlink" title="Jump to index entry"><code>log_sum_exp</code></a></strong><code>(vector x)</code><br> The natural logarithm of the sum of the exponentials of the elements in x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-41527e620b092f6ca6efaca55f839d7ff1404f5b" class="index-entry"></a> <!-- real; log_sum_exp; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_sum_exp" class="unlink" title="Jump to index entry"><code>log_sum_exp</code></a></strong><code>(row_vector x)</code><br> The natural logarithm of the sum of the exponentials of the elements in x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-54afbd4baa0c2ccf7b5392bdff0c6c87fce8222c" class="index-entry"></a> <!-- real; log_sum_exp; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_sum_exp" class="unlink" title="Jump to index entry"><code>log_sum_exp</code></a></strong><code>(matrix x)</code><br> The natural logarithm of the sum of the exponentials of the elements in x</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="minimum-and-maximum" class="level3">
+<h3 class="anchored" data-anchor-id="minimum-and-maximum">Minimum and maximum</h3>
+<a id="index-entry-1930adf5f047ff599b49aa2b5dd87c8fabe47d92" class="index-entry"></a> <!-- real; min; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#min" class="unlink" title="Jump to index entry"><code>min</code></a></strong><code>(vector x)</code><br> The minimum value in x, or <span class="math inline">\(+\infty\)</span> if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9aecdccfa0bdd71b2c969de650b8e72bad49df26" class="index-entry"></a> <!-- real; min; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#min" class="unlink" title="Jump to index entry"><code>min</code></a></strong><code>(row_vector x)</code><br> The minimum value in x, or <span class="math inline">\(+\infty\)</span> if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f8c78579caf9c583f2f55067b99104c0e7132124" class="index-entry"></a> <!-- real; min; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#min" class="unlink" title="Jump to index entry"><code>min</code></a></strong><code>(matrix x)</code><br> The minimum value in x, or <span class="math inline">\(+\infty\)</span> if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f2cd3ae46c31e2abdd35a92b4d10506f3c51f553" class="index-entry"></a> <!-- real; max; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#max" class="unlink" title="Jump to index entry"><code>max</code></a></strong><code>(vector x)</code><br> The maximum value in x, or <span class="math inline">\(-\infty\)</span> if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-260d6b034f7c7e6500278a93262c8d5af5ebe636" class="index-entry"></a> <!-- real; max; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#max" class="unlink" title="Jump to index entry"><code>max</code></a></strong><code>(row_vector x)</code><br> The maximum value in x, or <span class="math inline">\(-\infty\)</span> if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-3f0b6dd3ba52477b4fef8985ad1e0a330b31afaf" class="index-entry"></a> <!-- real; max; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#max" class="unlink" title="Jump to index entry"><code>max</code></a></strong><code>(matrix x)</code><br> The maximum value in x, or <span class="math inline">\(-\infty\)</span> if x is empty</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="sums-and-products" class="level3">
+<h3 class="anchored" data-anchor-id="sums-and-products">Sums and products</h3>
+<a id="index-entry-a3848926d02b63042a5cca7f5f2114a98d5a3b54" class="index-entry"></a> <!-- real; sum; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(vector x)</code><br> The sum of the values in x, or 0 if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6e609b116c423a798df7d214f82a296fb3c45baa" class="index-entry"></a> <!-- real; sum; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(row_vector x)</code><br> The sum of the values in x, or 0 if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7a966705e2b5c7762ea80adedd66384f947e0d59" class="index-entry"></a> <!-- real; sum; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sum" class="unlink" title="Jump to index entry"><code>sum</code></a></strong><code>(matrix x)</code><br> The sum of the values in x, or 0 if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9713d871d72b91892fb1be528d008fcb62d40ba7" class="index-entry"></a> <!-- real; prod; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(vector x)</code><br> The product of the values in x, or 1 if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c081c52af0fe73a003a0a406a8378cde350a34bf" class="index-entry"></a> <!-- real; prod; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(row_vector x)</code><br> The product of the values in x, or 1 if x is empty</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b45dad30f80e5937e91c940a770331aae131753c" class="index-entry"></a> <!-- real; prod; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#prod" class="unlink" title="Jump to index entry"><code>prod</code></a></strong><code>(matrix x)</code><br> The product of the values in x, or 1 if x is empty</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="sample-moments" class="level3">
+<h3 class="anchored" data-anchor-id="sample-moments">Sample moments</h3>
+<p>Full definitions are provided for sample moments in section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a>.</p>
+<a id="index-entry-895445a6a0170ef1bb8d7d7d9ed5eacdb328b24e" class="index-entry"></a> <!-- real; mean; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#mean" class="unlink" title="Jump to index entry"><code>mean</code></a></strong><code>(vector x)</code><br> The sample mean of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-793945619c88050a3af516f403737ac131fdc6a3" class="index-entry"></a> <!-- real; mean; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#mean" class="unlink" title="Jump to index entry"><code>mean</code></a></strong><code>(row_vector x)</code><br> The sample mean of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-fedafc3d15883df02af3a5cc3c738594e4aac9e9" class="index-entry"></a> <!-- real; mean; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#mean" class="unlink" title="Jump to index entry"><code>mean</code></a></strong><code>(matrix x)</code><br> The sample mean of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-500efa24724b539cd9b32e37fb5895b0cd48f91c" class="index-entry"></a> <!-- real; variance; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#variance" class="unlink" title="Jump to index entry"><code>variance</code></a></strong><code>(vector x)</code><br> The sample variance of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f595237eba9511d08bff024bd377900cc94e0c62" class="index-entry"></a> <!-- real; variance; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#variance" class="unlink" title="Jump to index entry"><code>variance</code></a></strong><code>(row_vector x)</code><br> The sample variance of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a59da01b6e381f01f0c9a08137440eea6455a2a2" class="index-entry"></a> <!-- real; variance; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#variance" class="unlink" title="Jump to index entry"><code>variance</code></a></strong><code>(matrix x)</code><br> The sample variance of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-41a7ec70ca2889f5e84fd426dfa08137bfaba28a" class="index-entry"></a> <!-- real; sd; (vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sd" class="unlink" title="Jump to index entry"><code>sd</code></a></strong><code>(vector x)</code><br> The sample standard deviation of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0182a4edcf541473e053aff58f5a7d2eafa83574" class="index-entry"></a> <!-- real; sd; (row_vector x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sd" class="unlink" title="Jump to index entry"><code>sd</code></a></strong><code>(row_vector x)</code><br> The sample standard deviation of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-8af54c041853b0f940ca2f60a845500498fdcf2b" class="index-entry"></a> <!-- real; sd; (matrix x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sd" class="unlink" title="Jump to index entry"><code>sd</code></a></strong><code>(matrix x)</code><br> The sample standard deviation of the values in x; see section <a href="../functions-reference/array_operations.html#array-reductions">array reductions</a> for details.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="quantile" class="level3">
+<h3 class="anchored" data-anchor-id="quantile">Quantile</h3>
+<p>Produces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.</p>
+<p>Implements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).</p>
+<a id="index-entry-5bcb3bb3e5ae7f0a50cf00b2fb37cc35759014da" class="index-entry"></a> <!-- real; quantile; (data vector x, data real p); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#quantile" class="unlink" title="Jump to index entry"><code>quantile</code></a></strong><code>(data vector x, data real p)</code><br> The p-th quantile of x</p>
+<small><i>Available since 2.27</i></small>
+<a id="index-entry-49b22dc2be74bb430f6e2faee8a2f6985280f397" class="index-entry"></a> <!-- array[] real; quantile; (data vector x, data array[] real p); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#quantile" class="unlink" title="Jump to index entry"><code>quantile</code></a></strong><code>(data vector x, data array[] real p)</code><br> An array containing the quantiles of x given by the array of probabilities p</p>
+<small><i>Available since 2.27</i></small>
+<a id="index-entry-5ca7a0e8fda0d0ac93828d2154479406eead9a61" class="index-entry"></a> <!-- real; quantile; (data row_vector x, data real p); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#quantile" class="unlink" title="Jump to index entry"><code>quantile</code></a></strong><code>(data row_vector x, data real p)</code><br> The p-th quantile of x</p>
+<small><i>Available since 2.27</i></small>
+<a id="index-entry-ba154b25a2265003899559c90fe541c3a07a82e0" class="index-entry"></a> <!-- array[] real; quantile; (data row_vector x, data array[] real p); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#quantile" class="unlink" title="Jump to index entry"><code>quantile</code></a></strong><code>(data row_vector x, data array[] real p)</code><br> An array containing the quantiles of x given by the array of probabilities p</p>
+<small><i>Available since 2.27</i></small>
+</section>
+</section>
+<section id="matrix-broadcast" class="level2">
+<h2 class="anchored" data-anchor-id="matrix-broadcast">Broadcast functions</h2>
+<p>The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.</p>
+<a id="index-entry-946d7c6f889b9fa740b16d2261aea52649b26b16" class="index-entry"></a> <!-- vector; rep_vector; (real x, int m); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#rep_vector" class="unlink" title="Jump to index entry"><code>rep_vector</code></a></strong><code>(real x, int m)</code><br> Return the size m (column) vector consisting of copies of x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-15a03522d48822ef641c68071772fb40994377ff" class="index-entry"></a> <!-- row_vector; rep_row_vector; (real x, int n); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#rep_row_vector" class="unlink" title="Jump to index entry"><code>rep_row_vector</code></a></strong><code>(real x, int n)</code><br> Return the size n row vector consisting of copies of x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f6dac444f76cc32475bbb31cceff748d10649662" class="index-entry"></a> <!-- matrix; rep_matrix; (real x, int m, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#rep_matrix" class="unlink" title="Jump to index entry"><code>rep_matrix</code></a></strong><code>(real x, int m, int n)</code><br> Return the m by n matrix consisting of copies of x.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9088cad16f7504b9acf856be4fc670b42cbd9eaa" class="index-entry"></a> <!-- matrix; rep_matrix; (vector v, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#rep_matrix" class="unlink" title="Jump to index entry"><code>rep_matrix</code></a></strong><code>(vector v, int n)</code><br> Return the m by n matrix consisting of n copies of the (column) vector v of size m.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a54769495396ea1427a26c419009fcf4b6b69cbf" class="index-entry"></a> <!-- matrix; rep_matrix; (row_vector rv, int m); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#rep_matrix" class="unlink" title="Jump to index entry"><code>rep_matrix</code></a></strong><code>(row_vector rv, int m)</code><br> Return the m by n matrix consisting of m copies of the row vector rv of size n.</p>
+<small><i>Available since 2.0</i></small>
+<p>Unlike the situation with array broadcasting (see section <a href="../functions-reference/array_operations.html#array-broadcasting">array broadcasting</a>), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">3</span>] x;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a> x = rep_vector(<span class="dv">1</span>, <span class="dv">3</span>);</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a> x = rep_vector(<span class="fl">1.0</span>, <span class="dv">3</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There are no integer vector or matrix types, so integer values are automatically promoted.</p>
+<section id="symmetrization" class="level3">
+<h3 class="anchored" data-anchor-id="symmetrization">Symmetrization</h3>
+<a id="index-entry-e4997fec1a21248abde86960359ea25d95244d9b" class="index-entry"></a> <!-- matrix; symmetrize_from_lower_tri; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#symmetrize_from_lower_tri" class="unlink" title="Jump to index entry"><code>symmetrize_from_lower_tri</code></a></strong><code>(matrix A)</code><br></p>
+<p>Construct a symmetric matrix from the lower triangle of A.</p>
+<small><i>Available since 2.26</i></small>
+</section>
+</section>
+<section id="diagonal-matrix-functions" class="level2">
+<h2 class="anchored" data-anchor-id="diagonal-matrix-functions">Diagonal matrix functions</h2>
+<a id="index-entry-6c34e1496216d483a0c9876713871db5b36de2b1" class="index-entry"></a> <!-- matrix; add_diag; (matrix m, row_vector d); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#add_diag" class="unlink" title="Jump to index entry"><code>add_diag</code></a></strong><code>(matrix m, row_vector d)</code><br> Add row_vector <code>d</code> to the diagonal of matrix <code>m</code>.</p>
+<small><i>Available since 2.21</i></small>
+<a id="index-entry-fd1fe55afe2b42fa662d88b9e836e27f40aed6ed" class="index-entry"></a> <!-- matrix; add_diag; (matrix m, vector d); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#add_diag" class="unlink" title="Jump to index entry"><code>add_diag</code></a></strong><code>(matrix m, vector d)</code><br> Add vector <code>d</code> to the diagonal of matrix <code>m</code>.</p>
+<small><i>Available since 2.21</i></small>
+<a id="index-entry-6c3894d5b0c4aa6fcae69f294bd4a39e91e3731b" class="index-entry"></a> <!-- matrix; add_diag; (matrix m, real d); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#add_diag" class="unlink" title="Jump to index entry"><code>add_diag</code></a></strong><code>(matrix m, real d)</code><br> Add scalar <code>d</code> to every diagonal element of matrix <code>m</code>.</p>
+<small><i>Available since 2.21</i></small>
+<a id="index-entry-6d4ef3406b8dabab1d53b29bb53ae58d983d1404" class="index-entry"></a> <!-- vector; diagonal; (matrix x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#diagonal" class="unlink" title="Jump to index entry"><code>diagonal</code></a></strong><code>(matrix x)</code><br> The diagonal of the matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-74845a2c2d1e1321f2bfdf69d4c7f321c856f555" class="index-entry"></a> <!-- matrix; diag_matrix; (vector x); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#diag_matrix" class="unlink" title="Jump to index entry"><code>diag_matrix</code></a></strong><code>(vector x)</code><br> The diagonal matrix with diagonal x</p>
+<small><i>Available since 2.0</i></small>
+<p>Although the <code>diag_matrix</code> function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a> y ~ multi_normal(mu, diag_matrix(square(sigma)));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>it is much more efficient to just use a univariate normal, which produces the same density,</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a> y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Rather than writing <code>m * diag_matrix(v)</code> where <code>m</code> is a matrix and <code>v</code> is a vector, it is much more efficient to write <code>diag_post_multiply(m, v)</code> (and similarly for pre-multiplication). By the same token, it is better to use <code>quad_form_diag(m, v)</code> rather than <code>quad_form(m, diag_matrix(v))</code>.</p>
+<a id="index-entry-7751d8abe876b03fe020df0799dd498b2be022a8" class="index-entry"></a> <!-- matrix; identity_matrix; (int k); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#identity_matrix" class="unlink" title="Jump to index entry"><code>identity_matrix</code></a></strong><code>(int k)</code><br> Create an identity matrix of size <span class="math inline">\(k \times k\)</span></p>
+<small><i>Available since 2.26</i></small>
+</section>
+<section id="container-construction" class="level2">
+<h2 class="anchored" data-anchor-id="container-construction">Container construction functions</h2>
+<a id="index-entry-71ca78d469339eb375a0b3e4fd656056c1f4bc37" class="index-entry"></a> <!-- array[] real; linspaced_array; (int n, data real lower, data real upper); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#linspaced_array" class="unlink" title="Jump to index entry"><code>linspaced_array</code></a></strong><code>(int n, data real lower, data real upper)</code><br> Create a real array of length <code>n</code> of equidistantly-spaced elements between <code>lower</code> and <code>upper</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-8ca31c5441d89c38eb6df32f4e84fa99f2542229" class="index-entry"></a> <!-- array[] real; linspaced_int_array; (int n, int lower, int upper); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#linspaced_int_array" class="unlink" title="Jump to index entry"><code>linspaced_int_array</code></a></strong><code>(int n, int lower, int upper)</code><br> Create a regularly spaced, increasing integer array of length <code>n</code> between <code>lower</code> and <code>upper</code>, inclusively. If <code>(upper - lower) / (n - 1)</code> is less than one, repeat each output <code>(n - 1) / (upper - lower)</code> times. If neither <code>(upper - lower) / (n - 1)</code> or <code>(n - 1) / (upper - lower)</code> are integers, <code>upper</code> is reduced until one of these is true.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-3f3107a87c63d9fe79a69329003d27f4b107d02e" class="index-entry"></a> <!-- vector; linspaced_vector; (int n, data real lower, data real upper); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#linspaced_vector" class="unlink" title="Jump to index entry"><code>linspaced_vector</code></a></strong><code>(int n, data real lower, data real upper)</code><br> Create an <code>n</code>-dimensional vector of equidistantly-spaced elements between <code>lower</code> and <code>upper</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-acc0b0d5ab9d4a99bd5c019f6b645424790aa4e9" class="index-entry"></a> <!-- row_vector; linspaced_row_vector; (int n, data real lower, data real upper); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#linspaced_row_vector" class="unlink" title="Jump to index entry"><code>linspaced_row_vector</code></a></strong><code>(int n, data real lower, data real upper)</code><br> Create an <code>n</code>-dimensional row-vector of equidistantly-spaced elements between <code>lower</code> and <code>upper</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-0beac7fcfa02cd2f93d1869abaf3555fa0711440" class="index-entry"></a> <!-- array[] int; one_hot_int_array; (int n, int k); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#one_hot_int_array" class="unlink" title="Jump to index entry"><code>one_hot_int_array</code></a></strong><code>(int n, int k)</code><br> Create a one-hot encoded int array of length <code>n</code> with <code>array[k] = 1</code></p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-268289da1aa3a477c94f93562ea8dec0d09ca486" class="index-entry"></a> <!-- array[] real; one_hot_array; (int n, int k); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#one_hot_array" class="unlink" title="Jump to index entry"><code>one_hot_array</code></a></strong><code>(int n, int k)</code><br> Create a one-hot encoded real array of length <code>n</code> with <code>array[k] = 1</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-ad02bb0cabeceb6ad40303c9ba844e75c5fab300" class="index-entry"></a> <!-- vector; one_hot_vector; (int K, int k); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#one_hot_vector" class="unlink" title="Jump to index entry"><code>one_hot_vector</code></a></strong><code>(int n, int k)</code><br> Create an <code>n</code>-dimensional one-hot encoded vector with <code>vector[k] = 1</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-0a0220657a7b045ad8375607c32be8215c237455" class="index-entry"></a> <!-- row_vector; one_hot_row_vector; (int n, int k); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#one_hot_row_vector" class="unlink" title="Jump to index entry"><code>one_hot_row_vector</code></a></strong><code>(int n, int k)</code><br> Create an <code>n</code>-dimensional one-hot encoded row-vector with <code>row_vector[k] = 1</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-9105f9deb3c88299539acb48f505c2ecb81107f4" class="index-entry"></a> <!-- array[] int; ones_int_array; (int n); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#ones_int_array" class="unlink" title="Jump to index entry"><code>ones_int_array</code></a></strong><code>(int n)</code><br> Create an int array of length <code>n</code> of all ones</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-0fafc0a373e0e8f0bbaeee5c35eba8570f9149d9" class="index-entry"></a> <!-- array[] real; ones_array; (int n); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#ones_array" class="unlink" title="Jump to index entry"><code>ones_array</code></a></strong><code>(int n)</code><br> Create a real array of length <code>n</code> of all ones</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-f291f3a346f24dfea23941e9dc9e908a2dfbe31d" class="index-entry"></a> <!-- vector; ones_vector; (int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#ones_vector" class="unlink" title="Jump to index entry"><code>ones_vector</code></a></strong><code>(int n)</code><br> Create an <code>n</code>-dimensional vector of all ones</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-a8497a47aca811a739e649d1ce295d1629665664" class="index-entry"></a> <!-- row_vector; ones_row_vector; (int n); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#ones_row_vector" class="unlink" title="Jump to index entry"><code>ones_row_vector</code></a></strong><code>(int n)</code><br> Create an <code>n</code>-dimensional row-vector of all ones</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-2c7602e1d0977230b64541b764981fca1bf1ad21" class="index-entry"></a> <!-- array[] int; zeros_int_array; (int n); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#zeros_int_array" class="unlink" title="Jump to index entry"><code>zeros_int_array</code></a></strong><code>(int n)</code><br> Create an int array of length <code>n</code> of all zeros</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-9d79405416a0d21e4f87a6517060a784d4b5d2b7" class="index-entry"></a> <!-- array[] real; zeros_array; (int n); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#zeros_array" class="unlink" title="Jump to index entry"><code>zeros_array</code></a></strong><code>(int n)</code><br> Create a real array of length <code>n</code> of all zeros</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-8b17549ebfa64ad40fc338a638fa00df789cc57d" class="index-entry"></a> <!-- vector; zeros_vector; (int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#zeros_vector" class="unlink" title="Jump to index entry"><code>zeros_vector</code></a></strong><code>(int n)</code><br> Create an <code>n</code>-dimensional vector of all zeros</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-7b23944569e7b31c1e4f17cc78de7996783733f1" class="index-entry"></a> <!-- row_vector; zeros_row_vector; (int n); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#zeros_row_vector" class="unlink" title="Jump to index entry"><code>zeros_row_vector</code></a></strong><code>(int n)</code><br> Create an <code>n</code>-dimensional row-vector of all zeros</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-2dec8337a5bdb9528da444f890d895306c7de4e2" class="index-entry"></a> <!-- vector; uniform_simplex; (int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#uniform_simplex" class="unlink" title="Jump to index entry"><code>uniform_simplex</code></a></strong><code>(int n)</code><br> Create an <code>n</code>-dimensional simplex with elements <code>vector[i] = 1 / n</code> for all <span class="math inline">\(i \in 1, \dots, n\)</span></p>
+<small><i>Available since 2.24</i></small>
+</section>
+<section id="slicing-and-blocking-functions" class="level2">
+<h2 class="anchored" data-anchor-id="slicing-and-blocking-functions">Slicing and blocking functions</h2>
+<p>Stan provides several functions for generating slices or blocks or diagonal entries for matrices.</p>
+<section id="columns-and-rows" class="level3">
+<h3 class="anchored" data-anchor-id="columns-and-rows">Columns and rows</h3>
+<a id="index-entry-fe3819da259646140d101538903c86c5bc1b49e9" class="index-entry"></a> <!-- vector; col; (matrix x, int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#col" class="unlink" title="Jump to index entry"><code>col</code></a></strong><code>(matrix x, int n)</code><br> The n-th column of matrix x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-5ad5162727cc2ef4908de930d86a6f86f1716ff8" class="index-entry"></a> <!-- row_vector; row; (matrix x, int m); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#row" class="unlink" title="Jump to index entry"><code>row</code></a></strong><code>(matrix x, int m)</code><br> The m-th row of matrix x</p>
+<small><i>Available since 2.0</i></small>
+<p>The <code>row</code> function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation <code>x[m]</code> is just an alternative way of writing <code>row(x,m)</code>. The <code>col</code> function may <strong>not</strong>, be used as an lvalue, nor is there an indexing based shorthand for it.</p>
+</section>
+<section id="block-operations" class="level3">
+<h3 class="anchored" data-anchor-id="block-operations">Block operations</h3>
+<section id="matrix-slicing-operations" class="level4">
+<h4 class="anchored" data-anchor-id="matrix-slicing-operations">Matrix slicing operations</h4>
+<p>Block operations may be used to extract a sub-block of a matrix.</p>
+<a id="index-entry-ed12d70d85024ac1bb87c1c1c5971659fb5f2cc1" class="index-entry"></a> <!-- matrix; block; (matrix x, int i, int j, int n_rows, int n_cols); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#block" class="unlink" title="Jump to index entry"><code>block</code></a></strong><code>(matrix x, int i, int j, int n_rows, int n_cols)</code><br> Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.</p>
+<small><i>Available since 2.0</i></small>
+<p>The sub-row and sub-column operations may be used to extract a slice of row or column from a matrix</p>
+<a id="index-entry-fb1126ac4a4410b2b2264dab2a17005f57546883" class="index-entry"></a> <!-- vector; sub_col; (matrix x, int i, int j, int n_rows); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#sub_col" class="unlink" title="Jump to index entry"><code>sub_col</code></a></strong><code>(matrix x, int i, int j, int n_rows)</code><br> Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f6b0581323605859a322af2e4cfb0eac1d6a67f8" class="index-entry"></a> <!-- row_vector; sub_row; (matrix x, int i, int j, int n_cols); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#sub_row" class="unlink" title="Jump to index entry"><code>sub_row</code></a></strong><code>(matrix x, int i, int j, int n_cols)</code><br> Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="vector-and-array-slicing-operations" class="level4">
+<h4 class="anchored" data-anchor-id="vector-and-array-slicing-operations">Vector and array slicing operations</h4>
+<p>The head operation extracts the first <span class="math inline">\(n\)</span> elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.</p>
+<a id="index-entry-8422dde5d63a6c756136d01ceeae0de6de1b00cb" class="index-entry"></a> <!-- vector; head; (vector v, int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#head" class="unlink" title="Jump to index entry"><code>head</code></a></strong><code>(vector v, int n)</code><br> Return the vector consisting of the first n elements of v.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-3107ded9497fc42a02b0573d3453f8cfb6302a41" class="index-entry"></a> <!-- row_vector; head; (row_vector rv, int n); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#head" class="unlink" title="Jump to index entry"><code>head</code></a></strong><code>(row_vector rv, int n)</code><br> Return the row vector consisting of the first n elements of rv.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-57085692c2de5cb3a0be6af59e4d4db1087270b9" class="index-entry"></a> <!-- array[] T; head; (array[] T sv, int n); -->
+<p></p>
+<p><code>array[] T</code> <strong><a href="../functions-reference/functions_index.html#head" class="unlink" title="Jump to index entry"><code>head</code></a></strong><code>(array[] T sv, int n)</code><br> Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements <code>T</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c59d1ebbba5b4cd5cf3ff22130cb426b46df662a" class="index-entry"></a> <!-- vector; tail; (vector v, int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#tail" class="unlink" title="Jump to index entry"><code>tail</code></a></strong><code>(vector v, int n)</code><br> Return the vector consisting of the last n elements of v.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-aa91264e50992aae5e2a411f1e27c3136e23148a" class="index-entry"></a> <!-- row_vector; tail; (row_vector rv, int n); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#tail" class="unlink" title="Jump to index entry"><code>tail</code></a></strong><code>(row_vector rv, int n)</code><br> Return the row vector consisting of the last n elements of rv.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4cdc4c7bbfd1b9012474960694df39fd3ac0bd73" class="index-entry"></a> <!-- array[] T; tail; (array[] T sv, int n); -->
+<p></p>
+<p><code>array[] T</code> <strong><a href="../functions-reference/functions_index.html#tail" class="unlink" title="Jump to index entry"><code>tail</code></a></strong><code>(array[] T sv, int n)</code><br> Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements <code>T</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-71cb7ac02f1e74f4b6fff77caa9b5ef0bd15346a" class="index-entry"></a> <!-- vector; segment; (vector v, int i, int n); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#segment" class="unlink" title="Jump to index entry"><code>segment</code></a></strong><code>(vector v, int i, int n)</code><br> Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-26587b215542055b9fe52316209a88041c119b87" class="index-entry"></a> <!-- row_vector; segment; (row_vector rv, int i, int n); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#segment" class="unlink" title="Jump to index entry"><code>segment</code></a></strong><code>(row_vector rv, int i, int n)</code><br> Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.</p>
+<small><i>Available since 2.10</i></small>
+<a id="index-entry-2374a68547cdedbbdfbb1b7212dd1ccaf48d84b3" class="index-entry"></a> <!-- array[] T; segment; (array[] T sv, int i, int n); -->
+<p></p>
+<p><code>array[] T</code> <strong><a href="../functions-reference/functions_index.html#segment" class="unlink" title="Jump to index entry"><code>segment</code></a></strong><code>(array[] T sv, int i, int n)</code><br> Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements <code>T</code>.</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+</section>
+<section id="matrix-concatenation" class="level2">
+<h2 class="anchored" data-anchor-id="matrix-concatenation">Matrix and vector concatenation</h2>
+<p>Stan’s matrix and vector concatenation operations <code>append_col</code> and <code>append_row</code> are like the operations <code>cbind</code> and <code>rbind</code> in R.</p>
+<section id="horizontal-concatenation" class="level4">
+<h4 class="anchored" data-anchor-id="horizontal-concatenation">Horizontal concatenation</h4>
+<a id="index-entry-9e081ce392909a0df24bb36f03bfa797bed61afc" class="index-entry"></a> <!-- matrix; append_col; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(matrix x, matrix y)</code><br> Combine matrices x and y by column. The matrices must have the same number of rows.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-147e13862145f964e1cbb7fcc77a67089a7e5e34" class="index-entry"></a> <!-- matrix; append_col; (matrix x, vector y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(matrix x, vector y)</code><br> Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-864b816280933455ecf6e923c13d45129a29b0fd" class="index-entry"></a> <!-- matrix; append_col; (vector x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(vector x, matrix y)</code><br> Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-58eedcbe1e2b7f299a093b3640fbf33c5ed668a7" class="index-entry"></a> <!-- matrix; append_col; (vector x, vector y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(vector x, vector y)</code><br> Combine vectors x and y by column. The vectors must have the same number of rows.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-21151b3cbeecb129181ef54f899636e705e64a59" class="index-entry"></a> <!-- row_vector; append_col; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(row_vector x, row_vector y)</code><br> Combine row vectors x and y of any size into another row vector by appending y to the end of x.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-c2a28748bc8f89b100ba5a447e6a37085a0c322f" class="index-entry"></a> <!-- row_vector; append_col; (real x, row_vector y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(real x, row_vector y)</code><br> Append x to the front of y, returning another row vector.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-3bee320543341ea29edfb9d04aca5cc49f8e0b95" class="index-entry"></a> <!-- row_vector; append_col; (row_vector x, real y); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#append_col" class="unlink" title="Jump to index entry"><code>append_col</code></a></strong><code>(row_vector x, real y)</code><br> Append y to the end of x, returning another row vector.</p>
+<small><i>Available since 2.12</i></small>
+</section>
+<section id="vertical-concatenation" class="level4">
+<h4 class="anchored" data-anchor-id="vertical-concatenation">Vertical concatenation</h4>
+<a id="index-entry-135048f5cff906d2164940e5b4b34af8be6d9972" class="index-entry"></a> <!-- matrix; append_row; (matrix x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(matrix x, matrix y)</code><br> Combine matrices x and y by row. The matrices must have the same number of columns.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-1f14ca719bab4b3d6eff9ae371f10a5fac48b85c" class="index-entry"></a> <!-- matrix; append_row; (matrix x, row_vector y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(matrix x, row_vector y)</code><br> Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-54b2d5adfa559ab3f257ab5acd8f7f57c4fafbf9" class="index-entry"></a> <!-- matrix; append_row; (row_vector x, matrix y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(row_vector x, matrix y)</code><br> Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-6f6580337b50792274a5b4740260362421589d97" class="index-entry"></a> <!-- matrix; append_row; (row_vector x, row_vector y); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(row_vector x, row_vector y)</code><br> Combine row vectors x and y by row. The row vectors must have the same number of columns.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-f406895b07b910f47d6bd89b7d603c711582be11" class="index-entry"></a> <!-- vector; append_row; (vector x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(vector x, vector y)</code><br> Concatenate vectors x and y of any size into another vector.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-be0707e02550507f8665c2fa85d337275029a3c5" class="index-entry"></a> <!-- vector; append_row; (real x, vector y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(real x, vector y)</code><br> Append x to the top of y, returning another vector.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-626869953f524fd6120faadc1ed9dcdb8f710167" class="index-entry"></a> <!-- vector; append_row; (vector x, real y); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#append_row" class="unlink" title="Jump to index entry"><code>append_row</code></a></strong><code>(vector x, real y)</code><br> Append y to the bottom of x, returning another vector.</p>
+<small><i>Available since 2.12</i></small>
+</section>
+</section>
+<section id="softmax" class="level2">
+<h2 class="anchored" data-anchor-id="softmax">Special matrix functions</h2>
+<section id="softmax-1" class="level3">
+<h3 class="anchored" data-anchor-id="softmax-1">Softmax</h3>
+<p>The softmax function maps<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> <span class="math inline">\(y \in \mathbb{R}^K\)</span> to the <span class="math inline">\(K\)</span>-simplex by <span class="math display">\[\begin{equation*} \text{softmax}(y)  = \frac{\exp(y)}
+{\sum_{k=1}^K \exp(y_k)}, \end{equation*}\]</span> where <span class="math inline">\(\exp(y)\)</span> is the componentwise exponentiation of <span class="math inline">\(y\)</span>. Softmax is usually calculated on the log scale, <span class="math display">\[\begin{eqnarray*} \log \text{softmax}(y) &amp; = &amp; \ y - \log \sum_{k=1}^K
+\exp(y_k) \\[4pt] &amp; = &amp; y - \mathrm{log\_sum\_exp}(y). \end{eqnarray*}\]</span> where the vector <span class="math inline">\(y\)</span> minus the scalar <span class="math inline">\(\mathrm{log\_sum\_exp}(y)\)</span> subtracts the scalar from each component of <span class="math inline">\(y\)</span>.</p>
+<p>Stan provides the following functions for softmax and its log.</p>
+<a id="index-entry-9267f9986949f0269013001d932bef84c805fa45" class="index-entry"></a> <!-- vector; softmax; (vector x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#softmax" class="unlink" title="Jump to index entry"><code>softmax</code></a></strong><code>(vector x)</code><br> The softmax of x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-49224e65079937569f0952ce5740c03bd8dcb98a" class="index-entry"></a> <!-- vector; log_softmax; (vector x); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#log_softmax" class="unlink" title="Jump to index entry"><code>log_softmax</code></a></strong><code>(vector x)</code><br> The natural logarithm of the softmax of x</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="cumulative-sums" class="level3">
+<h3 class="anchored" data-anchor-id="cumulative-sums">Cumulative sums</h3>
+<p>The cumulative sum of a sequence <span class="math inline">\(x_1,\ldots,x_N\)</span> is the sequence <span class="math inline">\(y_1,\ldots,y_N\)</span>, where <span class="math display">\[\begin{equation*} y_n = \sum_{m = 1}^{n} x_m. \end{equation*}\]</span></p>
+<a id="index-entry-092f8a13b9e06f8346cb61b8fc6a6fbacb3b5cc5" class="index-entry"></a> <!-- array[] int; cumulative_sum; (array[] int x); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(array[] int x)</code><br> The cumulative sum of x</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1e965776faac7516107384178cafd382f87f909a" class="index-entry"></a> <!-- array[] real; cumulative_sum; (array[] real x); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(array[] real x)</code><br> The cumulative sum of x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7c00c2971c29f98a54f16c9468d5b4a6eb529ff8" class="index-entry"></a> <!-- vector; cumulative_sum; (vector v); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(vector v)</code><br> The cumulative sum of v</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-951e3c465251f531a70f0dce6772057bf114685c" class="index-entry"></a> <!-- row_vector; cumulative_sum; (row_vector rv); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#cumulative_sum" class="unlink" title="Jump to index entry"><code>cumulative_sum</code></a></strong><code>(row_vector rv)</code><br> The cumulative sum of rv</p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="gaussian-process-covariance-functions" class="level2">
+<h2 class="anchored" data-anchor-id="gaussian-process-covariance-functions">Gaussian Process Covariance Functions</h2>
+<p>The Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.</p>
+<p>For one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by <span class="math inline">\(K_{ij} = k(x_i, x_j)\)</span> (where <span class="math inline">\(x_i\)</span> is the <span class="math inline">\(i^{th}\)</span> element of the array <span class="math inline">\(x\)</span>) and the cross-covariance is given by <span class="math inline">\(K_{ij} = k(x_i, y_j)\)</span>.</p>
+<p>For multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by <span class="math inline">\(K_{ij} = k(\mathbf{x}_i, \mathbf{x}_j)\)</span> (where <span class="math inline">\(\mathbf{x}_i\)</span> is the <span class="math inline">\(i^{th}\)</span> vector in the array <span class="math inline">\(x\)</span>) and the cross-covariance is given by <span class="math inline">\(K_{ij} = k(\mathbf{x}_i, \mathbf{y}_j)\)</span>.</p>
+<section id="exponentiated-quadratic-kernel" class="level3">
+<h3 class="anchored" data-anchor-id="exponentiated-quadratic-kernel">Exponentiated quadratic kernel</h3>
+<p>With magnitude <span class="math inline">\(\sigma\)</span> and length scale <span class="math inline">\(l\)</span>, the exponentiated quadratic kernel is:</p>
+<p><span class="math display">\[
+k(\mathbf{x}_i, \mathbf{x}_j) = \sigma^2 \exp \left( -\frac{|\mathbf{x}_i - \mathbf{x}_j|^2}{2l^2} \right)
+\]</span></p>
+<a id="index-entry-a59f2c545ffe0d99724685b78624ca3c6e6687dd" class="index-entry"></a> <!-- matrix; gp_exp_quad_cov; (array[] real x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exp_quad_cov" class="unlink" title="Jump to index entry"><code>gp_exp_quad_cov</code></a></strong><code>(array[] real x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with exponentiated quadratic kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-efeafc090e77d1a04aa32be00cc0f1f697dd613c" class="index-entry"></a> <!-- matrix; gp_exp_quad_cov; (array[] real x1, array[] real x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exp_quad_cov" class="unlink" title="Jump to index entry"><code>gp_exp_quad_cov</code></a></strong><code>(array[] real x1, array[] real x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with exponentiated quadratic kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-f9f7059ee7dbdb3c7ec60022629df94df4366eda" class="index-entry"></a> <!-- matrix; gp_exp_quad_cov; (vectors x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exp_quad_cov" class="unlink" title="Jump to index entry"><code>gp_exp_quad_cov</code></a></strong><code>(vectors x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with exponentiated quadratic kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-2a5cce0c9c4a586fd14ce028b6ece6593c0863c7" class="index-entry"></a> <!-- matrix; gp_exp_quad_cov; (vectors x, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exp_quad_cov" class="unlink" title="Jump to index entry"><code>gp_exp_quad_cov</code></a></strong><code>(vectors x, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-2a5cce0c9c4a586fd14ce028b6ece6593c0863c7" class="index-entry"></a> <!-- matrix; gp_exp_quad_cov; (vectors x, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exp_quad_cov" class="unlink" title="Jump to index entry"><code>gp_exp_quad_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with exponentiated quadratic kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-0171054e36a54ff590742f7f3dfb18db315818ed" class="index-entry"></a> <!-- matrix; gp_exp_quad_cov; (vectors x1, vectors x2, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exp_quad_cov" class="unlink" title="Jump to index entry"><code>gp_exp_quad_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+</section>
+<section id="dot-product-kernel" class="level3">
+<h3 class="anchored" data-anchor-id="dot-product-kernel">Dot product kernel</h3>
+<p>With bias <span class="math inline">\(\sigma_0\)</span> the dot product kernel is:</p>
+<p><span class="math display">\[
+k(\mathbf{x}_i, \mathbf{x}_j) = \sigma_0^2 + \mathbf{x}_i^T \mathbf{x}_j
+\]</span></p>
+<a id="index-entry-7d386a186ed2caf01b35da180fe71bf966425b74" class="index-entry"></a> <!-- matrix; gp_dot_prod_cov; (array[] real x, real sigma); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_dot_prod_cov" class="unlink" title="Jump to index entry"><code>gp_dot_prod_cov</code></a></strong><code>(array[] real x, real sigma)</code><br></p>
+<p>Gaussian process covariance with dot product kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-0e85dcc0b82afffce83d044f3699f82c43cda4fc" class="index-entry"></a> <!-- matrix; gp_dot_prod_cov; (array[] real x1, array[] real x2, real sigma); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_dot_prod_cov" class="unlink" title="Jump to index entry"><code>gp_dot_prod_cov</code></a></strong><code>(array[] real x1, array[] real x2, real sigma)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with dot product kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-b77643cdb29d01e2d4846872277e01f4d9e4a17c" class="index-entry"></a> <!-- matrix; gp_dot_prod_cov; (vectors x, real sigma); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_dot_prod_cov" class="unlink" title="Jump to index entry"><code>gp_dot_prod_cov</code></a></strong><code>(vectors x, real sigma)</code><br></p>
+<p>Gaussian process covariance with dot product kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-5929a214570377c8b9d1fe93d53acf0d146a7686" class="index-entry"></a> <!-- matrix; gp_dot_prod_cov; (vectors x1, vectors x2, real sigma); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_dot_prod_cov" class="unlink" title="Jump to index entry"><code>gp_dot_prod_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with dot product kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+</section>
+<section id="exponential-kernel" class="level3">
+<h3 class="anchored" data-anchor-id="exponential-kernel">Exponential kernel</h3>
+<p>With magnitude <span class="math inline">\(\sigma\)</span> and length scale <span class="math inline">\(l\)</span>, the exponential kernel is:</p>
+<p><span class="math display">\[
+k(\mathbf{x}_i, \mathbf{x}_j) = \sigma^2 \exp \left( -\frac{|\mathbf{x}_i - \mathbf{x}_j|}{l} \right)
+\]</span></p>
+<a id="index-entry-825dd9547256f5dcb295bf954db6f374cf924944" class="index-entry"></a> <!-- matrix; gp_exponential_cov; (array[] real x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exponential_cov" class="unlink" title="Jump to index entry"><code>gp_exponential_cov</code></a></strong><code>(array[] real x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with exponential kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-178aa0d85f72dea02342080bc2c8d9e0236570c1" class="index-entry"></a> <!-- matrix; gp_exponential_cov; (array[] real x1, array[] real x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exponential_cov" class="unlink" title="Jump to index entry"><code>gp_exponential_cov</code></a></strong><code>(array[] real x1, array[] real x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with exponential kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-467147dd6252c7d33347630a0101aae12325838e" class="index-entry"></a> <!-- matrix; gp_exponential_cov; (vectors x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exponential_cov" class="unlink" title="Jump to index entry"><code>gp_exponential_cov</code></a></strong><code>(vectors x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with exponential kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-4f6f2b3dfcd6826bfc4a157531179eaf4da21e17" class="index-entry"></a> <!-- matrix; gp_exponential_cov; (vectors x, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exponential_cov" class="unlink" title="Jump to index entry"><code>gp_exponential_cov</code></a></strong><code>(vectors x, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-9299d6de2d2e6c2e7be0930c480e0cf59f93e1d8" class="index-entry"></a> <!-- matrix; gp_exponential_cov; (vectors x1, vectors x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exponential_cov" class="unlink" title="Jump to index entry"><code>gp_exponential_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with exponential kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-78a300ec67a8ef9d073a915b1112c99d5439ebf2" class="index-entry"></a> <!-- matrix; gp_exponential_cov; (vectors x1, vectors x2, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_exponential_cov" class="unlink" title="Jump to index entry"><code>gp_exponential_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with exponential kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+</section>
+<section id="matern-32-kernel" class="level3">
+<h3 class="anchored" data-anchor-id="matern-32-kernel">Matern 3/2 kernel</h3>
+<p>With magnitude <span class="math inline">\(\sigma\)</span> and length scale <span class="math inline">\(l\)</span>, the Matern 3/2 kernel is:</p>
+<p><span class="math display">\[
+k(\mathbf{x}_i, \mathbf{x}_j) = \sigma^2 \left( 1 + \frac{\sqrt{3}|\mathbf{x}_i - \mathbf{x}_j|}{l} \right) \exp \left( -\frac{\sqrt{3}|\mathbf{x}_i - \mathbf{x}_j|}{l} \right)
+\]</span></p>
+<a id="index-entry-c67bd8178298aaea402200c34fc4e58b5f716374" class="index-entry"></a> <!-- matrix; gp_matern23_cov; (array[] real x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><code>gp_matern32_cov</code></strong><code>(array[] real x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with Matern 3/2 kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-77c38f1e81ceac337fd3248ff042bb819cbe5e6f" class="index-entry"></a> <!-- matrix; gp_matern23_cov; (array[] real x1, array[] real x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><code>gp_matern32_cov</code></strong><code>(array[] real x1, array[] real x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with Matern 3/2 kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-4049f44098827dd5df10dfe260933523d87ff79a" class="index-entry"></a> <!-- matrix; gp_matern23_cov; (vectors x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><code>gp_matern32_cov</code></strong><code>(vectors x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with Matern 3/2 kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-df0ecce0aa7b3af55d9c7e160b4e1c96caa7b1eb" class="index-entry"></a> <!-- matrix; gp_matern23_cov; (vectors x, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><code>gp_matern32_cov</code></strong><code>(vectors x, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-c0967d2d9deec4cab7150c95aa49dc8633d2f6af" class="index-entry"></a> <!-- matrix; gp_matern23_cov; (vectors x1, vectors x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><code>gp_matern32_cov</code></strong><code>(vectors x1, vectors x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with Matern 3/2 kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-34469df0872fa77c6199a495b6bedc9803d186a4" class="index-entry"></a> <!-- matrix; gp_matern23_cov; (vectors x1, vectors x2, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><code>gp_matern32_cov</code></strong><code>(vectors x1, vectors x2, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+</section>
+<section id="matern-52-kernel" class="level3">
+<h3 class="anchored" data-anchor-id="matern-52-kernel">Matern 5/2 kernel</h3>
+<p>With magnitude <span class="math inline">\(\sigma\)</span> and length scale <span class="math inline">\(l\)</span>, the Matern 5/2 kernel is:</p>
+<p><span class="math display">\[
+k(\mathbf{x}_i, \mathbf{x}_j) = \sigma^2 \left( 1 + \frac{\sqrt{5}|\mathbf{x}_i - \mathbf{x}_j|}{l} + \frac{5 |\mathbf{x}_i - \mathbf{x}_j|^2}{3l^2} \right)
+\exp \left( -\frac{\sqrt{5} |\mathbf{x}_i - \mathbf{x}_j|}{l} \right)
+\]</span></p>
+<a id="index-entry-48da2b928754ae4e47279653b6b2dc8a1849fd3e" class="index-entry"></a> <!-- matrix; gp_matern52_cov; (array[] real x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_matern52_cov" class="unlink" title="Jump to index entry"><code>gp_matern52_cov</code></a></strong><code>(array[] real x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with Matern 5/2 kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-5e89dd9a9be3e6c08de8184961c03cf9bc6e1aa4" class="index-entry"></a> <!-- matrix; gp_matern52_cov; (array[] real x1, array[] real x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_matern52_cov" class="unlink" title="Jump to index entry"><code>gp_matern52_cov</code></a></strong><code>(array[] real x1, array[] real x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with Matern 5/2 kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-e58aa2e1105325707d54de56f9bf8f8d0400e801" class="index-entry"></a> <!-- matrix; gp_matern52_cov; (vectors x, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_matern52_cov" class="unlink" title="Jump to index entry"><code>gp_matern52_cov</code></a></strong><code>(vectors x, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process covariance with Matern 5/2 kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-21a29eaabce12eae5d070311d157793c941e98bd" class="index-entry"></a> <!-- matrix; gp_matern52_cov; (vectors x, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_matern52_cov" class="unlink" title="Jump to index entry"><code>gp_matern52_cov</code></a></strong><code>(vectors x, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-ba526a205e3145ede4f46b6211f8e2cc4a6580f7" class="index-entry"></a> <!-- matrix; gp_matern52_cov; (vectors x1, vectors x2, real sigma, real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_matern52_cov" class="unlink" title="Jump to index entry"><code>gp_matern52_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with Matern 5/2 kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-f40fe74b859bd5fdf393b580e35eeba5ee0c15da" class="index-entry"></a> <!-- matrix; gp_matern52_cov; (vectors x1, vectors x2, real sigma, array[] real length_scale); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_matern52_cov" class="unlink" title="Jump to index entry"><code>gp_matern52_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, array[] real length_scale)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+</section>
+<section id="periodic-kernel" class="level3">
+<h3 class="anchored" data-anchor-id="periodic-kernel">Periodic kernel</h3>
+<p>With magnitude <span class="math inline">\(\sigma\)</span>, length scale <span class="math inline">\(l\)</span>, and period <span class="math inline">\(p\)</span>, the periodic kernel is:</p>
+<p><span class="math display">\[
+k(\mathbf{x}_i, \mathbf{x}_j) = \sigma^2 \exp \left(-\frac{2 \sin^2 \left( \pi \frac{|\mathbf{x}_i - \mathbf{x}_j|}{p} \right) }{l^2} \right)
+\]</span></p>
+<a id="index-entry-6707fa8fb889db9572c791a5ec6e7ead0310ba3f" class="index-entry"></a> <!-- matrix; gp_periodic_cov; (array[] real x, real sigma, real length_scale, real period); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_periodic_cov" class="unlink" title="Jump to index entry"><code>gp_periodic_cov</code></a></strong><code>(array[] real x, real sigma, real length_scale, real period)</code><br></p>
+<p>Gaussian process covariance with periodic kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-7f2049d9d9a6b2a24a7765d9f5af9b6df966a8b4" class="index-entry"></a> <!-- matrix; gp_periodic_cov; (array[] real x1, array[] real x2, real sigma, real length_scale, real period); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_periodic_cov" class="unlink" title="Jump to index entry"><code>gp_periodic_cov</code></a></strong><code>(array[] real x1, array[] real x2, real sigma, real length_scale, real period)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with periodic kernel in one dimension.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-c89ef06585db1d6db2d29210cae24aada2fa7d9f" class="index-entry"></a> <!-- matrix; gp_periodic_cov; (vectors x, real sigma, real length_scale, real period); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_periodic_cov" class="unlink" title="Jump to index entry"><code>gp_periodic_cov</code></a></strong><code>(vectors x, real sigma, real length_scale, real period)</code><br></p>
+<p>Gaussian process covariance with periodic kernel in multiple dimensions.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-3d85d317bf98b3cbd503a360059bfb1121a22a98" class="index-entry"></a> <!-- matrix; gp_periodic_cov; (vectors x1, vectors x2, real sigma, real length_scale, real period); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#gp_periodic_cov" class="unlink" title="Jump to index entry"><code>gp_periodic_cov</code></a></strong><code>(vectors x1, vectors x2, real sigma, real length_scale, real period)</code><br></p>
+<p>Gaussian process cross-covariance of <code>x1</code> and <code>x2</code> with periodic kernel in multiple dimensions with a length scale for each dimension.</p>
+<small><i>Available since 2.20</i></small>
+</section>
+</section>
+<section id="linear-algebra-functions-and-solvers" class="level2">
+<h2 class="anchored" data-anchor-id="linear-algebra-functions-and-solvers">Linear algebra functions and solvers</h2>
+<section id="matrix-division-operators-and-functions" class="level3">
+<h3 class="anchored" data-anchor-id="matrix-division-operators-and-functions">Matrix division operators and functions</h3>
+<p>In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.</p>
+<section id="matrix-division-operators" class="level4">
+<h4 class="anchored" data-anchor-id="matrix-division-operators">Matrix division operators</h4>
+<a id="index-entry-c8f636c70157ca03f436fa1fa6ad66005ee280c7" class="index-entry"></a> <!-- row_vector; operator/; (row_vector b, matrix A); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(row_vector b, matrix A)</code><br> The right division of b by A; equivalently <code>b * inverse(A)</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-aa2855d27dc938ec16cafecff875185e51ba660d" class="index-entry"></a> <!-- matrix; operator/; (matrix B, matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(matrix B, matrix A)</code><br> The right division of B by A; equivalently <code>B * inverse(A)</code></p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-fc754907294117fa562e6fcaf1b8afcf0d2efabd" class="index-entry"></a> <!-- vector; operator\; (matrix A, vector b); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#operator\\" class="unlink" title="Jump to index entry"><code>operator\</code></a></strong><code>(matrix A, vector b)</code><br> The left division of A by b; equivalently <code>inverse(A) * b</code></p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-1a08f375cfc4067e97bfed98fae8490e8b6f0698" class="index-entry"></a> <!-- matrix; operator\; (matrix A, matrix B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#operator\\" class="unlink" title="Jump to index entry"><code>operator\</code></a></strong><code>(matrix A, matrix B)</code><br> The left division of A by B; equivalently <code>inverse(A) * B</code></p>
+<small><i>Available since 2.18</i></small>
+</section>
+<section id="lower-triangular-matrix-division-functions" class="level4">
+<h4 class="anchored" data-anchor-id="lower-triangular-matrix-division-functions">Lower-triangular matrix division functions</h4>
+<p>There are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix <span class="math inline">\(\text{tri}(A)\)</span> is used in the definitions and defined by <span class="math display">\[\begin{equation*} \text{tri}(A)[m,n] = \left\{
+\begin{array}{ll} A[m,n] &amp; \text{if } m \geq n, \text{ and} \\[4pt] 0
+&amp; \text{otherwise}. \end{array} \right. \end{equation*}\]</span> When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.</p>
+<a id="index-entry-713d2983956ecba7e2e36a9e0614def9ea574bc6" class="index-entry"></a> <!-- vector; mdivide_left_tri_low; (matrix A, vector b); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#mdivide_left_tri_low" class="unlink" title="Jump to index entry"><code>mdivide_left_tri_low</code></a></strong><code>(matrix A, vector b)</code><br> The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form <code>inverse(tri(A)) * b</code>, where <code>tri(A)</code> is the lower-triangular portion of A with the above-diagonal entries set to zero.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-871e5638d23fa382b4f449a8fa0a5fa1c9556ecd" class="index-entry"></a> <!-- matrix; mdivide_left_tri_low; (matrix A, matrix B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#mdivide_left_tri_low" class="unlink" title="Jump to index entry"><code>mdivide_left_tri_low</code></a></strong><code>(matrix A, matrix B)</code><br> The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form <code>inverse(tri(A)) * B</code>, where <code>tri(A)</code> is the lower-triangular portion of A with the above-diagonal entries set to zero.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-44c350e5b90595515ab70568c03a3283d78a20c7" class="index-entry"></a> <!-- row_vector; mdivide_right_tri_low; (row_vector b, matrix A); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#mdivide_right_tri_low" class="unlink" title="Jump to index entry"><code>mdivide_right_tri_low</code></a></strong><code>(row_vector b, matrix A)</code><br> The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form <code>b * inverse(tri(A))</code>, where <code>tri(A)</code> is the lower-triangular portion of A with the above-diagonal entries set to zero.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-36a5e85b63e1d37ab7beac6c411e72d20254fcad" class="index-entry"></a> <!-- matrix; mdivide_right_tri_low; (matrix B, matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#mdivide_right_tri_low" class="unlink" title="Jump to index entry"><code>mdivide_right_tri_low</code></a></strong><code>(matrix B, matrix A)</code><br> The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form <code>B * inverse(tri(A))</code>, where <code>tri(A)</code> is the lower-triangular portion of A with the above-diagonal entries set to zero.</p>
+<small><i>Available since 2.5</i></small>
+</section>
+</section>
+<section id="symmetric-positive-definite-matrix-division-functions" class="level3">
+<h3 class="anchored" data-anchor-id="symmetric-positive-definite-matrix-division-functions">Symmetric positive-definite matrix division functions</h3>
+<p>There are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.</p>
+<a id="index-entry-0faeb825bd118874deb217de05883995b606c1e7" class="index-entry"></a> <!-- matrix; mdivide_left_spd; (matrix A, vector b); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#mdivide_left_spd" class="unlink" title="Jump to index entry"><code>mdivide_left_spd</code></a></strong><code>(matrix A, vector b)</code><br> The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form <code>inverse(A) * b</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-cf5f3991e5dd2accb21ce017e9f96521b0735e68" class="index-entry"></a> <!-- vector; mdivide_left_spd; (matrix A, matrix B); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#mdivide_left_spd" class="unlink" title="Jump to index entry"><code>mdivide_left_spd</code></a></strong><code>(matrix A, matrix B)</code><br> The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form <code>inverse(A) * B</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-d0585dccea5c1361c5123a20b0b892e33c4b3ad7" class="index-entry"></a> <!-- row_vector; mdivide_right_spd; (row_vector b, matrix A); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#mdivide_right_spd" class="unlink" title="Jump to index entry"><code>mdivide_right_spd</code></a></strong><code>(row_vector b, matrix A)</code><br> The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form <code>b *inverse(A)</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-835081fc5b867bbda5d6b16d5849e6ecbe5f952c" class="index-entry"></a> <!-- matrix; mdivide_right_spd; (matrix B, matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#mdivide_right_spd" class="unlink" title="Jump to index entry"><code>mdivide_right_spd</code></a></strong><code>(matrix B, matrix A)</code><br> The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form <code>B * inverse(A)</code>.</p>
+<small><i>Available since 2.12</i></small>
+</section>
+<section id="matrix-exponential" class="level3">
+<h3 class="anchored" data-anchor-id="matrix-exponential">Matrix exponential</h3>
+<p>The exponential of the matrix <span class="math inline">\(A\)</span> is formally defined by the convergent power series: <span class="math display">\[\begin{equation*} e^A = \sum_{n=0}^{\infty} \dfrac{A^n}{n!}
+\end{equation*}\]</span></p>
+<a id="index-entry-8aafa05928b6e41fc186f3ab4600c607bd5dd27c" class="index-entry"></a> <!-- matrix; matrix_exp; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#matrix_exp" class="unlink" title="Jump to index entry"><code>matrix_exp</code></a></strong><code>(matrix A)</code><br> The matrix exponential of A</p>
+<small><i>Available since 2.13</i></small>
+<a id="index-entry-07146ab787c63991cae5bac69256c64bc7dc73e5" class="index-entry"></a> <!-- matrix; matrix_exp_multiply; (matrix A, matrix B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#matrix_exp_multiply" class="unlink" title="Jump to index entry"><code>matrix_exp_multiply</code></a></strong><code>(matrix A, matrix B)</code><br> The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form <code>matrix_exp(A) * B</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-09a3d2d3107a8732b5360be5accb8fcce6bd2987" class="index-entry"></a> <!-- matrix; scale_matrix_exp_multiply; (real t, matrix A, matrix B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#scale_matrix_exp_multiply" class="unlink" title="Jump to index entry"><code>scale_matrix_exp_multiply</code></a></strong><code>(real t, matrix A, matrix B)</code><br> The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form <code>matrix_exp(t * A) * B</code>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+<section id="matrix-power" class="level3">
+<h3 class="anchored" data-anchor-id="matrix-power">Matrix power</h3>
+<p>Returns the nth power of the specific matrix: <span class="math display">\[\begin{equation*} M^n = M_1 * ... * M_n \end{equation*}\]</span></p>
+<a id="index-entry-4343283f37a884cc983cd269f8a680db1f8f7a9a" class="index-entry"></a> <!-- matrix; matrix_power; (matrix A, int B); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#matrix_power" class="unlink" title="Jump to index entry"><code>matrix_power</code></a></strong><code>(matrix A, int B)</code><br> Matrix A raised to the power B.</p>
+<small><i>Available since 2.24</i></small>
+</section>
+<section id="linear-algebra-functions" class="level3">
+<h3 class="anchored" data-anchor-id="linear-algebra-functions">Linear algebra functions</h3>
+<section id="trace" class="level4">
+<h4 class="anchored" data-anchor-id="trace">Trace</h4>
+<a id="index-entry-932fc0bd0a556886a853004f8ad6bc3432760b62" class="index-entry"></a> <!-- real; trace; (matrix A); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace" class="unlink" title="Jump to index entry"><code>trace</code></a></strong><code>(matrix A)</code><br> The trace of A, or 0 if A is empty; A is not required to be diagonal</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="determinants" class="level4">
+<h4 class="anchored" data-anchor-id="determinants">Determinants</h4>
+<a id="index-entry-2f0c697d96cb41a09f35f183f179f15b1ee0f3fa" class="index-entry"></a> <!-- real; determinant; (matrix A); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#determinant" class="unlink" title="Jump to index entry"><code>determinant</code></a></strong><code>(matrix A)</code><br> The determinant of A</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9e3fc21034e716f9152396c5bc91933e51a5adea" class="index-entry"></a> <!-- real; log_determinant; (matrix A); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_determinant" class="unlink" title="Jump to index entry"><code>log_determinant</code></a></strong><code>(matrix A)</code><br> The log of the absolute value of the determinant of A</p>
+<small><i>Available since 2.0</i></small>
+<p><code>real</code> <strong><code>log_determinant_spd</code></strong><code>(matrix A)</code><br> The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+<section id="inverses" class="level4">
+<h4 class="anchored" data-anchor-id="inverses">Inverses</h4>
+<p>It is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix <code>m</code> and post-multiplying by a vector or matrix <code>a</code>, as in <code>inverse(m) * a</code>, it is better to code this using matrix division, as in <code>m \ a</code>. The pre-multiplication case is similar, with <code>b * inverse(m)</code> being more efficiently coded as as <code>b / m</code>. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.</p>
+<p><em><strong>Warning:</strong></em> The function <code>inv(m)</code> is the elementwise inverse function, which returns <code>1 / m[i, j]</code> for each element.</p>
+<a id="index-entry-4f71689bce62cc64928a17fe2768a2a243320ff1" class="index-entry"></a> <!-- matrix; inverse; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#inverse" class="unlink" title="Jump to index entry"><code>inverse</code></a></strong><code>(matrix A)</code><br> Compute the inverse of A</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-cf78ebfad5c78b789158eb47797d8c00862001b3" class="index-entry"></a> <!-- matrix; inverse_spd; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#inverse_spd" class="unlink" title="Jump to index entry"><code>inverse_spd</code></a></strong><code>(matrix A)</code><br> Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0fbf0fe61f7df06da84a357eb40e9c719ff806ac" class="index-entry"></a> <!-- matrix; chol2inv; (matrix L); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#chol2inv" class="unlink" title="Jump to index entry"><code>chol2inv</code></a></strong><code>(matrix L)</code><br> Compute the inverse of the matrix whose cholesky factorization is L. That is, for <span class="math inline">\(A = L L^T\)</span>, return <span class="math inline">\(A^{-1}\)</span>.</p>
+<small><i>Available since 2.26</i></small>
+</section>
+<section id="generalized-inverse" class="level4">
+<h4 class="anchored" data-anchor-id="generalized-inverse">Generalized Inverse</h4>
+<p>The generalized inverse <span class="math inline">\(M^+\)</span> of a matrix <span class="math inline">\(M\)</span> is a matrix that satisfies <span class="math inline">\(M M^+ M = M\)</span>. For an invertible, square matrix <span class="math inline">\(M\)</span>, <span class="math inline">\(M^+\)</span> is equivalent to <span class="math inline">\(M^{-1}\)</span>. The dimensions of <span class="math inline">\(M^+\)</span> are equivalent to the dimensions of <span class="math inline">\(M^T\)</span>. The generalized inverse exists for any matrix, so the <span class="math inline">\(M\)</span> may be singular or less than full rank.</p>
+<p>Even though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank <span class="citation" data-cites="GolubPereyra:1973">(<a href="#ref-GolubPereyra:1973" role="doc-biblioref">Golub and Pereyra 1973</a>)</span>, meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix <span class="math inline">\(A\)</span> as a function of <span class="math inline">\(\epsilon\)</span>:</p>
+<p><span class="math display">\[
+A = \left(
+    \begin{array}{cccc}
+    1 + \epsilon &amp; 2 &amp; 1 \\
+    2 &amp; 4 &amp; 2
+    \end{array}
+    \right)
+\]</span></p>
+<p>When <span class="math inline">\(\epsilon = 0\)</span>, <span class="math inline">\(A\)</span> is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If <span class="math inline">\(\epsilon \neq 0\)</span>, the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at <span class="math inline">\(\epsilon = 0\)</span>, and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.</p>
+<a id="index-entry-49a205d1f503afa341f4ec2f12b8d9484c5bc3f0" class="index-entry"></a> <!-- matrix; generalized_inverse; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#generalized_inverse" class="unlink" title="Jump to index entry"><code>generalized_inverse</code></a></strong><code>(matrix A)</code><br> The generalized inverse of A</p>
+<small><i>Available since 2.26</i></small>
+</section>
+<section id="eigendecomposition" class="level4">
+<h4 class="anchored" data-anchor-id="eigendecomposition">Eigendecomposition</h4>
+<a id="index-entry-db72fae33eaa940967d1488e046f1d41ad7bd7b7" class="index-entry"></a> <!-- complex_vector; eigenvalues; (matrix A); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#eigenvalues" class="unlink" title="Jump to index entry"><code>eigenvalues</code></a></strong><code>(matrix A)</code><br> The complex-valued vector of eigenvalues of the matrix <code>A</code>. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6c608598725ec7c838453abdeb440c875eafe1a5" class="index-entry"></a> <!-- complex_matrix; eigenvectors; (matrix A); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#eigenvectors" class="unlink" title="Jump to index entry"><code>eigenvectors</code></a></strong><code>(matrix A)</code><br> The matrix with the complex-valued (column) eigenvectors of the matrix <code>A</code> in the same order as returned by the function <code>eigenvalues</code></p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-080c58f8bf67678c670ce7454674192000c074d1" class="index-entry"></a> <!-- tuple(complex_matrix, complex_vector); eigendecompose; (matrix A); -->
+<p></p>
+<p><code>tuple(complex_matrix, complex_vector)</code> <strong><a href="../functions-reference/functions_index.html#eigendecompose" class="unlink" title="Jump to index entry"><code>eigendecompose</code></a></strong><code>(matrix A)</code><br> Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix <code>A</code>. This function is equivalent to <code>(eigenvectors(A), eigenvalues(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+<a id="index-entry-09bb39cb077a959df51666f18ac7319ea53f8748" class="index-entry"></a> <!-- vector; eigenvalues_sym; (matrix A); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#eigenvalues_sym" class="unlink" title="Jump to index entry"><code>eigenvalues_sym</code></a></strong><code>(matrix A)</code><br> The vector of eigenvalues of a symmetric matrix <code>A</code> in ascending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-fb1919fccf8d3e19664c486e0cb67c876593f5a1" class="index-entry"></a> <!-- matrix; eigenvectors_sym; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#eigenvectors_sym" class="unlink" title="Jump to index entry"><code>eigenvectors_sym</code></a></strong><code>(matrix A)</code><br> The matrix with the (column) eigenvectors of symmetric matrix <code>A</code> in the same order as returned by the function <code>eigenvalues_sym</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7bd20778cc4eb845435d980b2323a26d021c94a0" class="index-entry"></a> <!-- tuple(matrix, vector); eigendecompose_sym; (matrix A); -->
+<p></p>
+<p><code>tuple(matrix, vector)</code> <strong><a href="../functions-reference/functions_index.html#eigendecompose_sym" class="unlink" title="Jump to index entry"><code>eigendecompose_sym</code></a></strong><code>(matrix A)</code><br> Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix <code>A</code>. This function is equivalent to <code>(eigenvectors_sym(A), eigenvalues_sym(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+<p>Because multiplying an eigenvector by <span class="math inline">\(-1\)</span> results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by <span class="math inline">\(-1\)</span>.</p>
+<p>The condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).</p>
+</section>
+<section id="QR-decomposition" class="level4">
+<h4 class="anchored" data-anchor-id="QR-decomposition">QR decomposition</h4>
+<a id="index-entry-00071108ad7f8a1ecce0c8659b622ad0c7e244b7" class="index-entry"></a> <!-- matrix; qr_thin_Q; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#qr_thin_Q" class="unlink" title="Jump to index entry"><code>qr_thin_Q</code></a></strong><code>(matrix A)</code><br> The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-854d13c8d7fed0092f63b68aed49a27b3f7fd886" class="index-entry"></a> <!-- matrix; qr_thin_R; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#qr_thin_R" class="unlink" title="Jump to index entry"><code>qr_thin_R</code></a></strong><code>(matrix A)</code><br> The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-01eb10e298ec5e0b2570e847ea477501b645f229" class="index-entry"></a> <!-- tuple(matrix, matrix); qr_thin; (matrix A); -->
+<p></p>
+<p><code>tuple(matrix, matrix)</code> <strong><a href="../functions-reference/functions_index.html#qr_thin" class="unlink" title="Jump to index entry"><code>qr_thin</code></a></strong><code>(matrix A)</code><br> Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to <code>(qr_thin_Q(A), qr_thin_R(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+<a id="index-entry-ecf3d80487f50dec1b5d8f7309ae1b12124339ee" class="index-entry"></a> <!-- matrix; qr_Q; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#qr_Q" class="unlink" title="Jump to index entry"><code>qr_Q</code></a></strong><code>(matrix A)</code><br> The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-b8d888170b3dfed1c077267a9a01cb4496d276f5" class="index-entry"></a> <!-- matrix; qr_R; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#qr_R" class="unlink" title="Jump to index entry"><code>qr_R</code></a></strong><code>(matrix A)</code><br> The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-0f3b3496deb99ba4839dce3aea55e6de916877e0" class="index-entry"></a> <!-- tuple(matrix, matrix); qr; (matrix A); -->
+<p> <code>tuple(matrix, matrix)</code> <strong><a href="../functions-reference/functions_index.html#qr" class="unlink" title="Jump to index entry"><code>qr</code></a></strong><code>(matrix A)</code><br> Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthogonal matrix in the fat QR decomposition and the second element (“R”) is upper trapezoidal. This function is equivalent to <code>(qr_Q(A), qr_R(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<small><i>Available since 2.33</i></small>
+<p>The thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as <span class="math display">\[\begin{equation*} A = Q \, R. \end{equation*}\]</span> Multiplying a column of an orthogonal matrix by <span class="math inline">\(-1\)</span> still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by <span class="math inline">\(-1\)</span> without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.</p>
+</section>
+<section id="cholesky-decomposition" class="level4">
+<h4 class="anchored" data-anchor-id="cholesky-decomposition">Cholesky decomposition</h4>
+<p>Every symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If <span class="math inline">\(\Sigma\)</span> is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector <span class="math inline">\(L\)</span> such that <span class="math display">\[\begin{equation*} \Sigma = L \, L^{\top}. \end{equation*}\]</span></p>
+<a id="index-entry-a7eaa217876baa0054d78dfff05f5685ef6261bb" class="index-entry"></a> <!-- matrix; cholesky_decompose; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cholesky_decompose" class="unlink" title="Jump to index entry"><code>cholesky_decompose</code></a></strong><code>(matrix A)</code><br> The lower-triangular Cholesky factor of the symmetric positive-definite matrix A</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="singular-value-decomposition" class="level4">
+<h4 class="anchored" data-anchor-id="singular-value-decomposition">Singular value decomposition</h4>
+<p>The matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, <span class="math display">\[\begin{equation*} A = U D V^T. \end{equation*}\]</span> The matrices of singular vectors here are thin. That is for an <span class="math inline">\(N\)</span> by <span class="math inline">\(P\)</span> input A, <span class="math inline">\(M = min(N, P)\)</span>, U is size <span class="math inline">\(N\)</span> by <span class="math inline">\(M\)</span> and V is size <span class="math inline">\(P\)</span> by <span class="math inline">\(M\)</span>.</p>
+<a id="index-entry-561216aa62b3975fc9a1a4e4d2158ed263b307f2" class="index-entry"></a> <!-- vector; singular_values; (matrix A); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#singular_values" class="unlink" title="Jump to index entry"><code>singular_values</code></a></strong><code>(matrix A)</code><br> The singular values of <code>A</code> in descending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-11a4ca04c6b3043104d40033be0b59dca0812809" class="index-entry"></a> <!-- matrix; svd_U; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#svd_U" class="unlink" title="Jump to index entry"><code>svd_U</code></a></strong><code>(matrix A)</code><br> The left-singular vectors of <code>A</code></p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-93d86ec7f54351b8ea720e400f10a534c1b87ee0" class="index-entry"></a> <!-- matrix; svd_V; (matrix A); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#svd_V" class="unlink" title="Jump to index entry"><code>svd_V</code></a></strong><code>(matrix A)</code><br> The right-singular vectors of <code>A</code></p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-2c5c96a002835c270f25ef8c2e2cebacf7a03c02" class="index-entry"></a> <!-- tuple(matrix, vector, matrix); svd; (matrix A); -->
+<p></p>
+<p><code>tuple(matrix, vector, matrix)</code> <strong><a href="../functions-reference/functions_index.html#svd" class="unlink" title="Jump to index entry"><code>svd</code></a></strong><code>(matrix A)</code><br> Returns a tuple containing the left-singular vectors of <code>A</code>, the singular values of <code>A</code> in descending order, and the right-singular values of <code>A</code>. This function is equivalent to <code>(svd_U(A), singular_values(A), svd_V(A))</code> but with a lower computational cost due to the shared work between the different components.</p>
+<small><i>Available since 2.33</i></small>
+</section>
+</section>
+</section>
+<section id="sort-functions" class="level2">
+<h2 class="anchored" data-anchor-id="sort-functions">Sort functions</h2>
+<p>See the <a href="../functions-reference/array_operations.html#sorting-functions">sorting functions section</a> for examples of how the functions work.</p>
+<a id="index-entry-3f7cd093f7a97b178afa3f28332faee31600a6c9" class="index-entry"></a> <!-- vector; sort_asc; (vector v); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#sort_asc" class="unlink" title="Jump to index entry"><code>sort_asc</code></a></strong><code>(vector v)</code><br> Sort the elements of v in ascending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9bc88f386e57fb9d4e7a2e0fd914d4525b7271c8" class="index-entry"></a> <!-- row_vector; sort_asc; (row_vector v); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#sort_asc" class="unlink" title="Jump to index entry"><code>sort_asc</code></a></strong><code>(row_vector v)</code><br> Sort the elements of v in ascending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9e567cdae4513e4fd1de2f40bda8578202f4b75f" class="index-entry"></a> <!-- vector; sort_desc; (vector v); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#sort_desc" class="unlink" title="Jump to index entry"><code>sort_desc</code></a></strong><code>(vector v)</code><br> Sort the elements of v in descending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-455584fc2e149f835e4f434734d42537044906a0" class="index-entry"></a> <!-- row_vector; sort_desc; (row_vector v); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#sort_desc" class="unlink" title="Jump to index entry"><code>sort_desc</code></a></strong><code>(row_vector v)</code><br> Sort the elements of v in descending order</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f37b48f43d2c6eaaeaa6b012dfae4912e9ac6869" class="index-entry"></a> <!-- array[] int; sort_indices_asc; (vector v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_asc" class="unlink" title="Jump to index entry"><code>sort_indices_asc</code></a></strong><code>(vector v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in ascending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-162c46d9f5b0bdb7329ba9d5db3e248a017bf548" class="index-entry"></a> <!-- array[] int; sort_indices_asc; (row_vector v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_asc" class="unlink" title="Jump to index entry"><code>sort_indices_asc</code></a></strong><code>(row_vector v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in ascending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-7660a9ad94113629502c6f1a3796f2e1e4033ce0" class="index-entry"></a> <!-- array[] int; sort_indices_desc; (vector v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_desc" class="unlink" title="Jump to index entry"><code>sort_indices_desc</code></a></strong><code>(vector v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in descending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-29d0b54b37e8b0e7e36145bd3ad202aa5d5c7362" class="index-entry"></a> <!-- array[] int; sort_indices_desc; (row_vector v); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#sort_indices_desc" class="unlink" title="Jump to index entry"><code>sort_indices_desc</code></a></strong><code>(row_vector v)</code><br> Return an array of indices between 1 and the size of v, sorted to index v in descending order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-a2612df2dc4076d90d32d7205a10905c38bcb241" class="index-entry"></a> <!-- int; rank; (vector v, int s); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rank" class="unlink" title="Jump to index entry"><code>rank</code></a></strong><code>(vector v, int s)</code><br> Number of components of v less than v[s]</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9dc76dd605fabb78a31bc22a4cbc3bba0371758d" class="index-entry"></a> <!-- int; rank; (row_vector v, int s); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#rank" class="unlink" title="Jump to index entry"><code>rank</code></a></strong><code>(row_vector v, int s)</code><br> Number of components of v less than v[s]</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="reverse-functions" class="level2">
+<h2 class="anchored" data-anchor-id="reverse-functions">Reverse functions</h2>
+<a id="index-entry-568dc0f51c2d88dc59f98ff0b5cf0080469ad789" class="index-entry"></a> <!-- vector; reverse; (vector v); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#reverse" class="unlink" title="Jump to index entry"><code>reverse</code></a></strong><code>(vector v)</code><br> Return a new vector containing the elements of the argument in reverse order.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-6b1f33889b3d5f2093ad7854f2692c91b8adcd02" class="index-entry"></a> <!-- row_vector; reverse; (row_vector v); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#reverse" class="unlink" title="Jump to index entry"><code>reverse</code></a></strong><code>(row_vector v)</code><br> Return a new row vector containing the elements of the argument in reverse order.</p>
+<small><i>Available since 2.23</i></small>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GolubPereyra:1973" class="csl-entry" role="listitem">
+Golub, G. H., and V. Pereyra. 1973. <span>“The Differentiation of Pseudo-Inverses and Nonlinear Least Squares Problems Whose Variables Separate.”</span> <em>SIAM Journal on Numerical Analysis</em> 10 (2): 413–32. <a href="https://doi.org/10.1137/0710036">https://doi.org/10.1137/0710036</a>.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The softmax function is so called because in the limit as <span class="math inline">\(y_n \rightarrow \infty\)</span> with <span class="math inline">\(y_m\)</span> for <span class="math inline">\(m \neq n\)</span> held constant, the result tends toward the “one-hot” vector <span class="math inline">\(\theta\)</span> with <span class="math inline">\(\theta_n
+= 1\)</span> and <span class="math inline">\(\theta_m = 0\)</span> for <span class="math inline">\(m \neq n\)</span>, thus providing a “soft” version of the maximum function.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/array_operations.html" class="pagination-link" aria-label="Array Operations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Array Operations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/complex_matrix_operations.html" class="pagination-link" aria-label="Complex Matrix Operations">
+        <span class="nav-page-text">Complex Matrix Operations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/mixed_operations.html b/docs/2_39/functions-reference/mixed_operations.html
new file mode 100644
index 000000000..af4d23e2a
--- /dev/null
+++ b/docs/2_39/functions-reference/mixed_operations.html
@@ -0,0 +1,1378 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Mixed Operations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/compound_arithmetic_and_assignment.html" rel="next">
+<link href="../functions-reference/sparse_matrix_operations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Mixed Operations">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/mixed_operations.html">Mixed Operations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#mixed-operations" id="toc-mixed-operations" class="nav-link active" data-scroll-target="#mixed-operations">Mixed Operations</a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mixed_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/mixed_operations.html">Mixed Operations</a></li></ol></nav></header>
+
+
+
+
+<section id="mixed-operations" class="level1">
+<h1>Mixed Operations</h1>
+<p>These functions perform conversions between Stan containers matrix, vector, row vector and arrays.</p>
+<a id="index-entry-98af15f087742b3c48602bac340d541ab6538cc6" class="index-entry"></a> <!-- matrix; to_matrix; (matrix m); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(matrix m)</code><br> Return the matrix <code>m</code> itself.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-1adffb84d0b4c5a1ad67ca2dce769cbdf687f577" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_matrix m); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_matrix m)</code><br> Return the matrix <code>m</code> itself.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-50fdb634f6d81eb2f8f0a2b2aba97a29cb575e69" class="index-entry"></a> <!-- matrix; to_matrix; (vector v); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(vector v)</code><br> Convert the column vector <code>v</code> to a <code>size(v)</code> by 1 matrix.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-aef6a0f55fed76a89b125774992391483f616f15" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_vector v); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_vector v)</code><br> Convert the column vector <code>v</code> to a <code>size(v)</code> by 1 matrix.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-792df8d223c6b836eee60bf39ca9d7d3f3e159b7" class="index-entry"></a> <!-- matrix; to_matrix; (row_vector v); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(row_vector v)</code><br> Convert the row vector <code>v</code> to a 1 by <code>size(v)</code> matrix.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-80cf0f8d6dea99916e4e16a9a5c04829bb454065" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_row_vector v); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_row_vector v)</code><br> Convert the row vector <code>v</code> to a 1 by <code>size(v)</code> matrix.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c513be95c486e67bd8b313bb220e6184fea9292a" class="index-entry"></a> <!-- matrix; to_matrix; (matrix M, int m, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(matrix M, int m, int n)</code><br> Convert a matrix <code>A</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-847d5e8f061de6894025068cf5f2eb0ba8215a31" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_matrix M, int m, int n); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_matrix M, int m, int n)</code><br> Convert a matrix <code>A</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-aed7ba7d91d4311c9ff956befae058ab2c5a56d5" class="index-entry"></a> <!-- matrix; to_matrix; (vector v, int m, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(vector v, int m, int n)</code><br> Convert a vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-b6f6e03fe842ef9ce3d6ca9aabb56537f122d502" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_vector v, int m, int n); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_vector v, int m, int n)</code><br> Convert a vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ed0fbc154447855f068f1a47cdf0f67c288a21d0" class="index-entry"></a> <!-- matrix; to_matrix; (row_vector v, int m, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(row_vector v, int m, int n)</code><br> Convert a row_vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-0b65adda0355db4bb294b2accb6ac1294603c298" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_row_vector v, int m, int n); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_row_vector v, int m, int n)</code><br> Convert a row vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-ce9fd0a6767674968523193f7115d495489cf449" class="index-entry"></a> <!-- matrix; to_matrix; (matrix A, int m, int n, int col_major); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(matrix A, int m, int n, int col_major)</code><br> Convert a matrix <code>A</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-e90eb83d724da49ec11f03ab61860acf90355da5" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_matrix A, int m, int n, int col_major); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_matrix A, int m, int n, int col_major)</code><br> Convert a matrix <code>A</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-0cf7e9303e3c6015089d4d776f68673d9becfe06" class="index-entry"></a> <!-- matrix; to_matrix; (vector v, int m, int n, int col_major); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(vector v, int m, int n, int col_major)</code><br> Convert a vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-38aaa00ecbf298c2b6ea8664028ae6e1db65995b" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_vector v, int m, int n, int col_major); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_vector v, int m, int n, int col_major)</code><br> Convert a vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-7f27f7aa2cd34abf4dc96d66caac862b24766c48" class="index-entry"></a> <!-- matrix; to_matrix; (row_vector v, int m, int n, int col_major); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(row_vector v, int m, int n, int col_major)</code><br> Convert a row vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-23ae52701cdc53e4b1252d837f1fd68b9cf3c092" class="index-entry"></a> <!-- complex_matrix; to_matrix; (complex_row_vector v, int m, int n, int col_major); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(complex_row_vector v, int m, int n, int col_major)</code><br> Convert a row vector <code>v</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-437db9cf0cf2250154a98c0bdad40ed9b384a755" class="index-entry"></a> <!-- matrix; to_matrix; (array[] real a, int m, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] real a, int m, int n)</code><br> Convert a one-dimensional array <code>a</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-e84dbcb742d813d14126e7e47ce17fad4d98c4ce" class="index-entry"></a> <!-- matrix; to_matrix; (array[] int a, int m, int n); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] int a, int m, int n)</code><br> Convert a one-dimensional array <code>a</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-bf764d9c03e5342181199ab42cf29145c1ad5854" class="index-entry"></a> <!-- complex_matrix; to_matrix; (array[] complex a, int m, int n); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] complex a, int m, int n)</code><br> Convert a one-dimensional array <code>a</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-49a6288ee0b5970768e1b5eb800c235cdc531dcb" class="index-entry"></a> <!-- matrix; to_matrix; (array[] real a, int m, int n, int col_major); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] real a, int m, int n, int col_major)</code><br> Convert a one-dimensional array <code>a</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-2114a4d1eb26abd7a5a168f852fa2824b04264a7" class="index-entry"></a> <!-- matrix; to_matrix; (array[] int a, int m, int n, int col_major); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] int a, int m, int n, int col_major)</code><br> Convert a one-dimensional array <code>a</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-cae4771e239eee8858e814def2ab98579b41b5eb" class="index-entry"></a> <!-- complex_matrix; to_matrix; (array[] complex a, int m, int n, int col_major); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] complex a, int m, int n, int col_major)</code><br> Convert a one-dimensional array <code>a</code> to a matrix with <code>m</code> rows and <code>n</code> columns filled in row-major order if <code>col_major</code> equals 0 (otherwise, they get filled in column-major order).</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-43ba4671c0aedffb35391977e1454635f31140b1" class="index-entry"></a> <!-- matrix; to_matrix; (array[] row_vector vs); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] row_vector vs)</code><br> Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-30e7bab8b93cd06e7e722a69f87795853a321d6f" class="index-entry"></a> <!-- complex_matrix; to_matrix; (array[] complex_row_vector vs); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[] complex_row_vector vs)</code><br> Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-57abafe2911be5f079350cd935db3790df0e435b" class="index-entry"></a> <!-- matrix; to_matrix; (array[,] real a); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[,] real a)</code><br> Convert the two dimensional array <code>a</code> to a matrix with the same dimensions and indexing order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-3e8783df8a1f5a3a9252432f08e410ff98dcc3cc" class="index-entry"></a> <!-- matrix; to_matrix; (array[,] int a); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[,] int a)</code><br> Convert the two dimensional array <code>a</code> to a matrix with the same dimensions and indexing order. If any of the dimensions of <code>a</code> are zero, the result will be a <span class="math inline">\(0 \times 0\)</span> matrix.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-e92b8a0a6976052fa8b22b2d2678e9daa1dc22b0" class="index-entry"></a> <!-- complex_matrix; to_matrix; (array[,] complex a ); -->
+<p></p>
+<p><code>complex_matrix</code> <strong><a href="../functions-reference/functions_index.html#to_matrix" class="unlink" title="Jump to index entry"><code>to_matrix</code></a></strong><code>(array[,] complex a )</code><br> Convert the two dimensional array <code>a</code> to a matrix with the same dimensions and indexing order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6f0903591f2ae71c76fb2f1c0a3330e7a39749c3" class="index-entry"></a> <!-- vector; to_vector; (matrix m); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(matrix m)</code><br> Convert the matrix <code>m</code> to a column vector in column-major order.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-038b0de9dc724055453f09b0cbcbd18ed0b245f5" class="index-entry"></a> <!-- complex_vector; to_vector; (complex_matrix m); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(complex_matrix m)</code><br> Convert the matrix <code>m</code> to a column vector in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-5eaf76d379576fc126f8cb1a1ddde01467fd11b4" class="index-entry"></a> <!-- vector; to_vector; (vector v); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(vector v)</code><br> Return the column vector <code>v</code> itself.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-d05604f9063eb5e5f46e231e2dedd8a2be6250c3" class="index-entry"></a> <!-- complex_vector; to_vector; (complex_vector v); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(complex_vector v)</code><br> Return the column vector <code>v</code> itself.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-3c628a0f6a20eda39461617826fd33046b2dc1c3" class="index-entry"></a> <!-- vector; to_vector; (row_vector v); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(row_vector v)</code><br> Convert the row vector <code>v</code> to a column vector.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-5976d67e9306d7472825e4d06e32e82b5c9ed3b0" class="index-entry"></a> <!-- complex_vector; to_vector; (complex_row_vector v); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(complex_row_vector v)</code><br> Convert the row vector <code>v</code> to a column vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-c912c1c0f3589db9d1a5d2bffb2b3b697c20f3d7" class="index-entry"></a> <!-- vector; to_vector; (array[] real a); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(array[] real a)</code><br> Convert the one-dimensional array <code>a</code> to a column vector.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-32661d5b8760ecddd2d8777793db7ece329e55ec" class="index-entry"></a> <!-- vector; to_vector; (array[] int a); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(array[] int a)</code><br> Convert the one-dimensional integer array <code>a</code> to a column vector.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-b7e01d700955b123c6a7d94113e75dae2182ca9d" class="index-entry"></a> <!-- complex_vector; to_vector; (array[] complex a); -->
+<p></p>
+<p><code>complex_vector</code> <strong><a href="../functions-reference/functions_index.html#to_vector" class="unlink" title="Jump to index entry"><code>to_vector</code></a></strong><code>(array[] complex a)</code><br> Convert the one-dimensional complex array <code>a</code> to a column vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-2233e067e71fece1a5091dbf29c83f891c0c53ac" class="index-entry"></a> <!-- row_vector; to_row_vector; (matrix m); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(matrix m)</code><br> Convert the matrix <code>m</code> to a row vector in column-major order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-48b04a9a3d0a2ba72b4564f6c56c7787d7d85a3d" class="index-entry"></a> <!-- complex_row_vector; to_row_vector; (complex_matrix m); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(complex_matrix m)</code><br> Convert the matrix <code>m</code> to a row vector in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d10d7dd47bea28e9f6be7cb1fee79ba98a6291e2" class="index-entry"></a> <!-- row_vector; to_row_vector; (vector v); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(vector v)</code><br> Convert the column vector <code>v</code> to a row vector.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-91aa719d0ab39507c033078fbe737e6992bf641b" class="index-entry"></a> <!-- complex_row_vector; to_row_vector; (complex_vector v); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(complex_vector v)</code><br> Convert the column vector <code>v</code> to a row vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-dd3935354cbe9c750365fa9d199935099c0fccd5" class="index-entry"></a> <!-- row_vector; to_row_vector; (row_vector v); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(row_vector v)</code><br> Return the row vector <code>v</code> itself.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-c2aadf6729da7ef673b6e4729cd3640e7e8ff075" class="index-entry"></a> <!-- complex_row_vector; to_row_vector; (complex_row_vector v); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(complex_row_vector v)</code><br> Return the row vector <code>v</code> itself.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-cea7d9d7492857fd3d8ca4b00cbb08cd09beb181" class="index-entry"></a> <!-- row_vector; to_row_vector; (array[] real a); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(array[] real a)</code><br> Convert the one-dimensional array <code>a</code> to a row vector.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-dfad2435683430f7d34faf59c9ef36aaeef30c0f" class="index-entry"></a> <!-- row_vector; to_row_vector; (array[] int a); -->
+<p></p>
+<p><code>row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(array[] int a)</code><br> Convert the one-dimensional array <code>a</code> to a row vector.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-a7fd1cdc7869d1b6354acbb6c4d601343e80bc32" class="index-entry"></a> <!-- complex_row_vector; to_row_vector; (array[] complex a); -->
+<p></p>
+<p><code>complex_row_vector</code> <strong><a href="../functions-reference/functions_index.html#to_row_vector" class="unlink" title="Jump to index entry"><code>to_row_vector</code></a></strong><code>(array[] complex a)</code><br> Convert the one-dimensional complex array <code>a</code> to a row vector.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-1523c6dc72269a2b99daf2481a6f15790faf2dd7" class="index-entry"></a> <!-- array[,] real; to_array_2d; (matrix m); -->
+<p></p>
+<p><code>array[,] real</code> <strong><a href="../functions-reference/functions_index.html#to_array_2d" class="unlink" title="Jump to index entry"><code>to_array_2d</code></a></strong><code>(matrix m)</code><br> Convert the matrix <code>m</code> to a two dimensional array with the same dimensions and indexing order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-67b1d2f656b711d5b1f916e3e04e26038c9d38d6" class="index-entry"></a> <!-- array[,] real; to_array_2d; (complex_matrix m); -->
+<p></p>
+<p><code>array[,] complex</code> <strong><a href="../functions-reference/functions_index.html#to_array_2d" class="unlink" title="Jump to index entry"><code>to_array_2d</code></a></strong><code>(complex_matrix m)</code><br> Convert the matrix <code>m</code> to a two dimensional array with the same dimensions and indexing order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-2f5e845bb7c1046346206e70df45afa871d3bbca" class="index-entry"></a> <!-- array[] real; to_array_1d; (vector v); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(vector v)</code><br> Convert the column vector <code>v</code> to a one-dimensional array.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-242195f746365e4ab07756d8dfc2d627d760131c" class="index-entry"></a> <!-- array[] real; to_array_1d; (complex_vector v); -->
+<p></p>
+<p><code>array[] complex</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(complex_vector v)</code><br> Convert the column vector <code>v</code> to a one-dimensional array.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d5e3525dd3b1532ec70529a13582b2ac5684b7a4" class="index-entry"></a> <!-- array[] real; to_array_1d; (row_vector v); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(row_vector v)</code><br> Convert the row vector <code>v</code> to a one-dimensional array.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-9dbf7ca0cd05cc0137132abe8de599edc8db6fb6" class="index-entry"></a> <!-- array[] complex; to_array_1d; (complex_row_vector v); -->
+<p></p>
+<p><code>array[] complex</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(complex_row_vector v)</code><br> Convert the row vector <code>v</code> to a one-dimensional array.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-89bbd3b92786cfeacbc58a9b8e1815eb4704fe94" class="index-entry"></a> <!-- array[] real; to_array_1d; (matrix m); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(matrix m)</code><br> Convert the matrix <code>m</code> to a one-dimensional array in column-major order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-841e76b5eb14b885bfb8ea725edd4531171d3dc7" class="index-entry"></a> <!-- array[] complex; to_array_1d; (complex_matrix m); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(complex_matrix m)</code><br> Convert the matrix <code>m</code> to a one-dimensional array in column-major order.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-d4bca2f4496bd91a3f19415933ba8d5dee724894" class="index-entry"></a> <!-- array[] real; to_array_1d; (array[...] real a); -->
+<p></p>
+<p><code>array[] real</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(array[...] real a)</code><br> Convert the array <code>a</code> (of any dimension up to 10) to a one-dimensional array in row-major order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-febcf6672aef91647fe55778211ffd168082990f" class="index-entry"></a> <!-- array[] int; to_array_1d; (array[...] int a); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(array[...] int a)</code><br> Convert the array <code>a</code> (of any dimension up to 10) to a one-dimensional array in row-major order.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-a283222b01f625bc6705e111e10bdc2759e3d24c" class="index-entry"></a> <!-- array[] complex; to_array_1d; (array[...] complex a); -->
+<p></p>
+<p><code>array[] complex</code> <strong><a href="../functions-reference/functions_index.html#to_array_1d" class="unlink" title="Jump to index entry"><code>to_array_1d</code></a></strong><code>(array[...] complex a)</code><br> Convert the array <code>a</code> (of any dimension up to 10) to a one-dimensional array in row-major order.</p>
+<small><i>Available since 2.30</i></small>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/sparse_matrix_operations.html" class="pagination-link" aria-label="Sparse Matrix Operations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Sparse Matrix Operations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="pagination-link" aria-label="Compound Arithmetic and Assignment">
+        <span class="nav-page-text">Compound Arithmetic and Assignment</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mixed_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/multivariate_discrete_distributions.html b/docs/2_39/functions-reference/multivariate_discrete_distributions.html
new file mode 100644
index 000000000..007d82fcc
--- /dev/null
+++ b/docs/2_39/functions-reference/multivariate_discrete_distributions.html
@@ -0,0 +1,1273 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Multivariate Discrete Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/unbounded_continuous_distributions.html" rel="next">
+<link href="../functions-reference/unbounded_discrete_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Multivariate Discrete Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/multivariate_discrete_distributions.html">Multivariate Discrete Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#multivariate-discrete-distributions" id="toc-multivariate-discrete-distributions" class="nav-link active" data-scroll-target="#multivariate-discrete-distributions">Multivariate Discrete Distributions</a>
+  <ul class="collapse">
+  <li><a href="#multinomial-distribution" id="toc-multinomial-distribution" class="nav-link" data-scroll-target="#multinomial-distribution">Multinomial distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function" id="toc-probability-mass-function" class="nav-link" data-scroll-target="#probability-mass-function">Probability mass function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#multinomial-distribution-logit-parameterization" id="toc-multinomial-distribution-logit-parameterization" class="nav-link" data-scroll-target="#multinomial-distribution-logit-parameterization">Multinomial distribution, logit parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-1" id="toc-probability-mass-function-1" class="nav-link" data-scroll-target="#probability-mass-function-1">Probability mass function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#dirichlet-multinomial-distribution" id="toc-dirichlet-multinomial-distribution" class="nav-link" data-scroll-target="#dirichlet-multinomial-distribution">Dirichlet-multinomial distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-2" id="toc-probability-mass-function-2" class="nav-link" data-scroll-target="#probability-mass-function-2">Probability mass function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/multivariate_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/multivariate_discrete_distributions.html">Multivariate Discrete Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="multivariate-discrete-distributions" class="level1">
+<h1>Multivariate Discrete Distributions</h1>
+<p>The multivariate discrete distributions are over multiple integer values, which are expressed in Stan as arrays.</p>
+<section id="multinomial-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="multinomial-distribution">Multinomial distribution</h2>
+<section id="probability-mass-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function">Probability mass function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(N \in \mathbb{N}\)</span>, and <span class="math inline">\(\theta \in \text{$K$-simplex}\)</span>, then for <span class="math inline">\(y \in \mathbb{N}^K\)</span> such that <span class="math inline">\(\sum_{k=1}^K y_k = N\)</span>, <span class="math display">\[\begin{equation*}
+\text{Multinomial}(y|\theta) = \binom{N}{y_1,\ldots,y_K} \prod_{k=1}^K \theta_k^{y_k},
+\end{equation*}\]</span> where the multinomial coefficient is defined by <span class="math display">\[\begin{equation*}
+\binom{N}{y_1,\ldots,y_k} = \frac{N!}{\prod_{k=1}^K y_k!}.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multinomial" class="unlink" title="Jump to index entry"><code>multinomial</code></a></strong><code>(theta)</code></p>
+<p>Increment target log probability density with <code>multinomial_lupmf(y | theta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-e60b9005f8bb7c315f7c9c28123c0e6cf5a64f66" class="index-entry"></a> <!-- real; multinomial ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-2a1a28ece722b9157657d45c59162d74bb3757df" class="index-entry"></a> <!-- real; multinomial_lpmf; (array[] int y | vector theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multinomial_lpmf" class="unlink" title="Jump to index entry"><code>multinomial_lpmf</code></a></strong><code>(array[] int y | vector theta)</code><br> The log multinomial probability mass function with outcome array <code>y</code> of size <span class="math inline">\(K\)</span> given the <span class="math inline">\(K\)</span>-simplex distribution parameter theta and (implicit) total count <code>N = sum(y)</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-43f3a751ea3d648a3e23143c525562a1f20e99a3" class="index-entry"></a> <!-- real; multinomial_lupmf; (array[] int y | vector theta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multinomial_lupmf" class="unlink" title="Jump to index entry"><code>multinomial_lupmf</code></a></strong><code>(array[] int y | vector theta)</code><br> The log multinomial probability mass function with outcome array <code>y</code> of size <span class="math inline">\(K\)</span> given the <span class="math inline">\(K\)</span>-simplex distribution parameter theta and (implicit) total count <code>N = sum(y)</code> dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-be7dbffff355abebf7628e28e92ff3e0fc11e115" class="index-entry"></a> <!-- array[] int; multinomial_rng; (vector theta, int N); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#multinomial_rng" class="unlink" title="Jump to index entry"><code>multinomial_rng</code></a></strong><code>(vector theta, int N)</code><br> Generate a multinomial variate with simplex distribution parameter theta and total count <span class="math inline">\(N\)</span>; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.8</i></small>
+</section>
+</section>
+<section id="multinomial-distribution-logit-parameterization" class="level2">
+<h2 class="anchored" data-anchor-id="multinomial-distribution-logit-parameterization">Multinomial distribution, logit parameterization</h2>
+<p>Stan also provides a version of the multinomial probability mass function distribution with the <span class="math inline">\(\text{$K$-simplex}\)</span> for the event count probabilities per category given on the unconstrained logistic scale.</p>
+<section id="probability-mass-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-1">Probability mass function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(N \in \mathbb{N}\)</span>, and <span class="math inline">\(\text{softmax}(\theta) \in \text{$K$-simplex}\)</span>, then for <span class="math inline">\(y \in \mathbb{N}^K\)</span> such that <span class="math inline">\(\sum_{k=1}^K y_k = N\)</span>, <span class="math display">\[\begin{equation*}
+\begin{split}
+\text{MultinomialLogit}(y \mid \gamma) &amp; = \text{Multinomial}(y \mid \text{softmax}(\gamma)) \\
+&amp; = \binom{N}{y_1,\ldots,y_K} \prod_{k=1}^K [\text{softmax}(\gamma_k)]^{y_k},
+\end{split}
+\end{equation*}\]</span> where the multinomial coefficient is defined by <span class="math display">\[\begin{equation*}
+\binom{N}{y_1,\ldots,y_k} = \frac{N!}{\prod_{k=1}^K y_k!}.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#multinomial_logit" class="unlink" title="Jump to index entry"><code>multinomial_logit</code></a></strong><code>(gamma)</code></p>
+<p>Increment target log probability density with <code>multinomial_logit_lupmf(y | gamma)</code>.</p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-00ac83839e3f3aac2fa9ff06c7cfb47d659ef796" class="index-entry"></a> <!-- real; multinomial_logit ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-dac1445017dc2f0105939d8fd00f7654057c9a95" class="index-entry"></a> <!-- real; multinomial_logit_lpmf; (array[] int y | vector gamma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multinomial_logit_lpmf" class="unlink" title="Jump to index entry"><code>multinomial_logit_lpmf</code></a></strong><code>(array[] int y | vector gamma)</code><br> The log multinomial probability mass function with outcome array <code>y</code> of size <span class="math inline">\(K\)</span> given the log <span class="math inline">\(K\)</span>-simplex distribution parameter <span class="math inline">\(\gamma\)</span> and (implicit) total count <code>N = sum(y)</code></p>
+<small><i>Available since 2.24</i></small>
+<a id="index-entry-b2a96795c168ea1f40d7be5db3359edbf96cc95f" class="index-entry"></a> <!-- real; multinomial_logit_lupmf; (array[] int y | vector gamma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#multinomial_logit_lupmf" class="unlink" title="Jump to index entry"><code>multinomial_logit_lupmf</code></a></strong><code>(array[] int y | vector gamma)</code><br> The log multinomial probability mass function with outcome array <code>y</code> of size <span class="math inline">\(K\)</span> given the log <span class="math inline">\(K\)</span>-simplex distribution parameter <span class="math inline">\(\gamma\)</span> and (implicit) total count <code>N = sum(y)</code> dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-1d2c47ee6f88d4e1c284b4647634a23d3b215792" class="index-entry"></a> <!-- array[] int; multinomial_logit_rng; (vector gamma, int N); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#multinomial_logit_rng" class="unlink" title="Jump to index entry"><code>multinomial_logit_rng</code></a></strong><code>(vector gamma, int N)</code><br> Generate a variate from a multinomial distribution with probabilities <code>softmax(gamma)</code> and total count <code>N</code>; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.24</i></small>
+</section>
+</section>
+<section id="dirichlet-multinomial-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="dirichlet-multinomial-distribution">Dirichlet-multinomial distribution</h2>
+<p>Stan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.</p>
+<section id="probability-mass-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-2">Probability mass function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span>, <span class="math inline">\(N \in \mathbb{N}\)</span>, and <span class="math inline">\(\alpha \in
+\mathbb{R}_{+}^K\)</span>, then for <span class="math inline">\(y \in \mathbb{N}^K\)</span> such that <span class="math inline">\(\sum_{k=1}^K y_k = N\)</span>, the PMF of the Dirichlet-multinomial distribution is defined as <span class="math display">\[\begin{equation*}
+\text{DirMult}(y|\theta) =
+\frac{\Gamma(\alpha_0)\Gamma(N+1)}{\Gamma(N+\alpha_0)} \prod_{k=1}^K \frac{\Gamma(y_k + \alpha_k)}{\Gamma(\alpha_k)\Gamma(y_k+1)},
+\end{equation*}\]</span> where <span class="math inline">\(\alpha_0\)</span> is defined as <span class="math inline">\(\alpha_0 = \sum_{k=1}^K \alpha_k\)</span>.</p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_multinomial" class="unlink" title="Jump to index entry"><code>dirichlet_multinomial</code></a></strong><code>(alpha)</code></p>
+<p>Increment target log probability density with <code>dirichlet_multinomial_lupmf(y | alpha)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-3f5317d36179306fa3d125648b669c29f0ae4632" class="index-entry"></a> <!-- real; dirichlet_multinomial ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-fdad801bbd6eb934645dbde5c01f271aa4e741b4" class="index-entry"></a> <!-- real; dirichlet_multinomial_lpmf; (array[] int y | vector alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_multinomial_lpmf" class="unlink" title="Jump to index entry"><code>dirichlet_multinomial_lpmf</code></a></strong><code>(array[] int y | vector alpha)</code><br> The log multinomial probability mass function with outcome array <code>y</code> with <span class="math inline">\(K\)</span> elements given the positive <span class="math inline">\(K\)</span>-vector distribution parameter <code>alpha</code> and (implicit) total count <code>N = sum(y)</code>.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-6b8a21266747ae90543cbd60108ee12b4d08d5e3" class="index-entry"></a> <!-- real; dirichlet_multinomial_lupmf; (array[] int y | vector alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_multinomial_lupmf" class="unlink" title="Jump to index entry"><code>dirichlet_multinomial_lupmf</code></a></strong><code>(array[] int y | vector alpha)</code><br> The log multinomial probability mass function with outcome array <code>y</code> with <span class="math inline">\(K\)</span> elements, given the positive <span class="math inline">\(K\)</span>-vector distribution parameter <code>alpha</code> and (implicit) total count <code>N = sum(y)</code> dropping constant additive terms.</p>
+<small><i>Available since 2.34</i></small>
+<a id="index-entry-bdd1be0680f9ae13a49c9c4e6af87b4f9c7359df" class="index-entry"></a> <!-- array[] int; dirichlet_multinomial_rng; (vector alpha, int N); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_multinomial_rng" class="unlink" title="Jump to index entry"><code>dirichlet_multinomial_rng</code></a></strong><code>(vector alpha, int N)</code><br> Generate a multinomial variate with positive vector distribution parameter <code>alpha</code> and total count <code>N</code>; may only be used in transformed data and generated quantities blocks. This is equivalent to <code>multinomial_rng(dirichlet_rng(alpha), N)</code>.</p>
+<small><i>Available since 2.34</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/unbounded_discrete_distributions.html" class="pagination-link" aria-label="Unbounded Discrete Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Unbounded Discrete Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/unbounded_continuous_distributions.html" class="pagination-link" aria-label="Unbounded Continuous Distributions">
+        <span class="nav-page-text">Unbounded Continuous Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/multivariate_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/positive_continuous_distributions.html b/docs/2_39/functions-reference/positive_continuous_distributions.html
new file mode 100644
index 000000000..6c27f5814
--- /dev/null
+++ b/docs/2_39/functions-reference/positive_continuous_distributions.html
@@ -0,0 +1,1699 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Positive Continuous Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/positive_lower-bounded_distributions.html" rel="next">
+<link href="../functions-reference/unbounded_continuous_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Positive Continuous Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/positive_continuous_distributions.html">Positive Continuous Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#positive-continuous-distributions" id="toc-positive-continuous-distributions" class="nav-link active" data-scroll-target="#positive-continuous-distributions">Positive Continuous Distributions</a>
+  <ul class="collapse">
+  <li><a href="#lognormal" id="toc-lognormal" class="nav-link" data-scroll-target="#lognormal">Lognormal distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#chi-square-distribution" id="toc-chi-square-distribution" class="nav-link" data-scroll-target="#chi-square-distribution">Chi-square distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#inverse-chi-square-distribution" id="toc-inverse-chi-square-distribution" class="nav-link" data-scroll-target="#inverse-chi-square-distribution">Inverse chi-square distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-2" id="toc-probability-density-function-2" class="nav-link" data-scroll-target="#probability-density-function-2">Probability density function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#scaled-inverse-chi-square-distribution" id="toc-scaled-inverse-chi-square-distribution" class="nav-link" data-scroll-target="#scaled-inverse-chi-square-distribution">Scaled inverse chi-square distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-3" id="toc-probability-density-function-3" class="nav-link" data-scroll-target="#probability-density-function-3">Probability density function</a></li>
+  <li><a href="#distribution-statement-3" id="toc-distribution-statement-3" class="nav-link" data-scroll-target="#distribution-statement-3">Distribution statement</a></li>
+  <li><a href="#stan-functions-3" id="toc-stan-functions-3" class="nav-link" data-scroll-target="#stan-functions-3">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#exponential-distribution" id="toc-exponential-distribution" class="nav-link" data-scroll-target="#exponential-distribution">Exponential distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-4" id="toc-probability-density-function-4" class="nav-link" data-scroll-target="#probability-density-function-4">Probability density function</a></li>
+  <li><a href="#distribution-statement-4" id="toc-distribution-statement-4" class="nav-link" data-scroll-target="#distribution-statement-4">Distribution statement</a></li>
+  <li><a href="#stan-functions-4" id="toc-stan-functions-4" class="nav-link" data-scroll-target="#stan-functions-4">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#gamma-distribution" id="toc-gamma-distribution" class="nav-link" data-scroll-target="#gamma-distribution">Gamma distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-5" id="toc-probability-density-function-5" class="nav-link" data-scroll-target="#probability-density-function-5">Probability density function</a></li>
+  <li><a href="#distribution-statement-5" id="toc-distribution-statement-5" class="nav-link" data-scroll-target="#distribution-statement-5">Distribution statement</a></li>
+  <li><a href="#stan-functions-5" id="toc-stan-functions-5" class="nav-link" data-scroll-target="#stan-functions-5">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#inverse-gamma-distribution" id="toc-inverse-gamma-distribution" class="nav-link" data-scroll-target="#inverse-gamma-distribution">Inverse gamma Distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-6" id="toc-probability-density-function-6" class="nav-link" data-scroll-target="#probability-density-function-6">Probability density function</a></li>
+  <li><a href="#distribution-statement-6" id="toc-distribution-statement-6" class="nav-link" data-scroll-target="#distribution-statement-6">Distribution statement</a></li>
+  <li><a href="#stan-functions-6" id="toc-stan-functions-6" class="nav-link" data-scroll-target="#stan-functions-6">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#weibull-distribution" id="toc-weibull-distribution" class="nav-link" data-scroll-target="#weibull-distribution">Weibull distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-7" id="toc-probability-density-function-7" class="nav-link" data-scroll-target="#probability-density-function-7">Probability density function</a></li>
+  <li><a href="#distribution-statement-7" id="toc-distribution-statement-7" class="nav-link" data-scroll-target="#distribution-statement-7">Distribution statement</a></li>
+  <li><a href="#stan-functions-7" id="toc-stan-functions-7" class="nav-link" data-scroll-target="#stan-functions-7">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#frechet-distribution" id="toc-frechet-distribution" class="nav-link" data-scroll-target="#frechet-distribution">Frechet distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-8" id="toc-probability-density-function-8" class="nav-link" data-scroll-target="#probability-density-function-8">Probability density function</a></li>
+  <li><a href="#distribution-statement-8" id="toc-distribution-statement-8" class="nav-link" data-scroll-target="#distribution-statement-8">Distribution statement</a></li>
+  <li><a href="#stan-functions-8" id="toc-stan-functions-8" class="nav-link" data-scroll-target="#stan-functions-8">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#rayleigh-distribution" id="toc-rayleigh-distribution" class="nav-link" data-scroll-target="#rayleigh-distribution">Rayleigh distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-9" id="toc-probability-density-function-9" class="nav-link" data-scroll-target="#probability-density-function-9">Probability density function</a></li>
+  <li><a href="#distribution-statement-9" id="toc-distribution-statement-9" class="nav-link" data-scroll-target="#distribution-statement-9">Distribution statement</a></li>
+  <li><a href="#stan-functions-9" id="toc-stan-functions-9" class="nav-link" data-scroll-target="#stan-functions-9">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#log-logistic-distribution" id="toc-log-logistic-distribution" class="nav-link" data-scroll-target="#log-logistic-distribution">Log-logistic distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-10" id="toc-probability-density-function-10" class="nav-link" data-scroll-target="#probability-density-function-10">Probability density function</a></li>
+  <li><a href="#distribution-statement-10" id="toc-distribution-statement-10" class="nav-link" data-scroll-target="#distribution-statement-10">Distribution statement</a></li>
+  <li><a href="#stan-functions-10" id="toc-stan-functions-10" class="nav-link" data-scroll-target="#stan-functions-10">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/positive_continuous_distributions.html">Positive Continuous Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="positive-continuous-distributions" class="level1">
+<h1>Positive Continuous Distributions</h1>
+<p>The positive continuous probability functions have support on the positive real numbers.</p>
+<section id="lognormal" class="level2">
+<h2 class="anchored" data-anchor-id="lognormal">Lognormal distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in
+\mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*} \text{LogNormal}(y|\mu,\sigma) = \frac{1}{\sqrt{2
+\pi} \ \sigma} \, \frac{1}{y} \ \exp \! \left(        - \, \frac{1}{2}
+\, \left( \frac{\log y - \mu}{\sigma} \right)^2      \right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#lognormal" class="unlink" title="Jump to index entry"><code>lognormal</code></a></strong><code>(mu, sigma)</code></p>
+<p>Increment target log probability density with <code>lognormal_lupdf(y | mu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-96de647c4a4c68e79cbd1680d0976df411556c5f" class="index-entry"></a> <!-- real; lognormal ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-ce67c5b635189c70cda3beeadc6fd4238ba2b535" class="index-entry"></a> <!-- real; lognormal_lpdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lognormal_lpdf" class="unlink" title="Jump to index entry"><code>lognormal_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the lognormal density of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-3d3d0388e0be101f05b8314d8bed921fef2a999b" class="index-entry"></a> <!-- real; lognormal_lupdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lognormal_lupdf" class="unlink" title="Jump to index entry"><code>lognormal_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-765ee6d534fa4f5fb5e2d621a21456071b4c8458" class="index-entry"></a> <!-- real; lognormal_cdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lognormal_cdf" class="unlink" title="Jump to index entry"><code>lognormal_cdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The cumulative lognormal distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-93c2a2ecc235e787a00b95dad3075afbbf4a21db" class="index-entry"></a> <!-- real; lognormal_lcdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lognormal_lcdf" class="unlink" title="Jump to index entry"><code>lognormal_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the lognormal cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-64e46338c2669c1f1b399608c9d1c654bc02710b" class="index-entry"></a> <!-- real; lognormal_lccdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lognormal_lccdf" class="unlink" title="Jump to index entry"><code>lognormal_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-ea103d5de1264a7df987cedd3cb455d3302557af" class="index-entry"></a> <!-- R; lognormal_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lognormal_rng" class="unlink" title="Jump to index entry"><code>lognormal_rng</code></a></strong><code>(reals mu, reals sigma)</code><br> Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.22</i></small>
+</section>
+</section>
+<section id="chi-square-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="chi-square-distribution">Chi-square distribution</h2>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>If <span class="math inline">\(\nu \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*}
+\text{ChiSquare}(y|\nu) = \frac{2^{-\nu/2}}     {\Gamma(\nu / 2)} \,
+y^{\nu/2 - 1} \, \exp \! \left( -\, \frac{1}{2} \, y \right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#chi_square" class="unlink" title="Jump to index entry"><code>chi_square</code></a></strong><code>(nu)</code></p>
+<p>Increment target log probability density with <code>chi_square_lupdf(y | nu)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a8bd60e9ff005e8522863542e28a64336a42e65c" class="index-entry"></a> <!-- real; chi_square ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-6c7045c1cb405776f7ed13ebbb5da4b27240ea15" class="index-entry"></a> <!-- real; chi_square_lpdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#chi_square_lpdf" class="unlink" title="Jump to index entry"><code>chi_square_lpdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the Chi-square density of y given degrees of freedom nu</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-9cccc8fcca1c8dd1b172176bd72c946a84d611eb" class="index-entry"></a> <!-- real; chi_square_lupdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#chi_square_lupdf" class="unlink" title="Jump to index entry"><code>chi_square_lupdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-face5eba84183b72541fdba1f3b6ea4a1949e9b5" class="index-entry"></a> <!-- real; chi_square_cdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#chi_square_cdf" class="unlink" title="Jump to index entry"><code>chi_square_cdf</code></a></strong><code>(reals y | reals nu)</code><br> The Chi-square cumulative distribution function of y given degrees of freedom nu</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-90d49c25cb5546006ff8802e5928c09d11c03dc8" class="index-entry"></a> <!-- real; chi_square_lcdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#chi_square_lcdf" class="unlink" title="Jump to index entry"><code>chi_square_lcdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the Chi-square cumulative distribution function of y given degrees of freedom nu</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-a010e1cf6b0229303867a826777cc2a9b6c3bff9" class="index-entry"></a> <!-- real; chi_square_lccdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#chi_square_lccdf" class="unlink" title="Jump to index entry"><code>chi_square_lccdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-ced9b4410298033c3bdae65b66331f4a6d973c89" class="index-entry"></a> <!-- R; chi_square_rng; (reals nu); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#chi_square_rng" class="unlink" title="Jump to index entry"><code>chi_square_rng</code></a></strong><code>(reals nu)</code><br> Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="inverse-chi-square-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="inverse-chi-square-distribution">Inverse chi-square distribution</h2>
+<section id="probability-density-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-2">Probability density function</h3>
+<p>If <span class="math inline">\(\nu \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*}
+\text{InvChiSquare}(y \, | \, \nu) = \frac{2^{-\nu/2}}    {\Gamma(\nu
+/ 2)} \, y^{-\nu/2 - 1} \, \exp\! \left( \! - \, \frac{1}{2} \,
+\frac{1}{y} \right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square" class="unlink" title="Jump to index entry"><code>inv_chi_square</code></a></strong><code>(nu)</code></p>
+<p>Increment target log probability density with <code>inv_chi_square_lupdf(y | nu)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2c55516e33a8c93de05a426a4331b0fd2c73665c" class="index-entry"></a> <!-- real; inv_chi_square ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-c49192fa8c8d7405bca27fccb88443c3a31b82f9" class="index-entry"></a> <!-- real; inv_chi_square_lpdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square_lpdf" class="unlink" title="Jump to index entry"><code>inv_chi_square_lpdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the inverse Chi-square density of y given degrees of freedom nu</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-a5086fcc0bf03bde2c07057468cd23a2936131b7" class="index-entry"></a> <!-- real; inv_chi_square_lupdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square_lupdf" class="unlink" title="Jump to index entry"><code>inv_chi_square_lupdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-a5b53196a941f74078d28a76d43875006ecaa749" class="index-entry"></a> <!-- real; inv_chi_square_cdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square_cdf" class="unlink" title="Jump to index entry"><code>inv_chi_square_cdf</code></a></strong><code>(reals y | reals nu)</code><br> The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-19a25046f012bc203f47cddbd716160311344693" class="index-entry"></a> <!-- real; inv_chi_square_lcdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square_lcdf" class="unlink" title="Jump to index entry"><code>inv_chi_square_lcdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-5fb795e9994cfaedf511409172a6457e8df06bff" class="index-entry"></a> <!-- real; inv_chi_square_lccdf; (reals y | reals nu); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square_lccdf" class="unlink" title="Jump to index entry"><code>inv_chi_square_lccdf</code></a></strong><code>(reals y | reals nu)</code><br> The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-b2c4491a4b62bfff6789788b555b1de2f50ad833" class="index-entry"></a> <!-- R; inv_chi_square_rng; (reals nu); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_chi_square_rng" class="unlink" title="Jump to index entry"><code>inv_chi_square_rng</code></a></strong><code>(reals nu)</code><br> Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="scaled-inverse-chi-square-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="scaled-inverse-chi-square-distribution">Scaled inverse chi-square distribution</h2>
+<section id="probability-density-function-3" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-3">Probability density function</h3>
+<p>If <span class="math inline">\(\nu \in \mathbb{R}^+\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y
+\in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*} \text{ScaledInvChiSquare}(y|\nu,\sigma) =
+\frac{(\nu / 2)^{\nu/2}}      {\Gamma(\nu / 2)} \, \sigma^{\nu} \,
+y^{-(\nu/2 + 1)} \, \exp \! \left( \!    - \, \frac{1}{2} \, \nu \,
+\sigma^2 \, \frac{1}{y} \right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-3" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-3">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square</code></a></strong><code>(nu, sigma)</code></p>
+<p>Increment target log probability density with <code>scaled_inv_chi_square_lupdf(y | nu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-d782005ee920606bf8bdd43885ae449421dce0f6" class="index-entry"></a> <!-- real; scaled_inv_chi_square ~; -->
+<p></p>
+</section>
+<section id="stan-functions-3" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
+<a id="index-entry-9df549db2fb37eff9dfe90a6a22d14745f8c5fe5" class="index-entry"></a> <!-- real; scaled_inv_chi_square_lpdf; (reals y | reals nu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square_lpdf" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square_lpdf</code></a></strong><code>(reals y | reals nu, reals sigma)</code><br> The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-8b7dfa52248b073d01e53ceabcc698299d321803" class="index-entry"></a> <!-- real; scaled_inv_chi_square_lupdf; (reals y | reals nu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square_lupdf" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square_lupdf</code></a></strong><code>(reals y | reals nu, reals sigma)</code><br> The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-be017a6dd0ea4e5a40dfbe2ed649d0837f935724" class="index-entry"></a> <!-- real; scaled_inv_chi_square_cdf; (reals y | reals nu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square_cdf" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square_cdf</code></a></strong><code>(reals y | reals nu, reals sigma)</code><br> The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-e14593aac1cd14f7cf744da7a520e81ea74f2923" class="index-entry"></a> <!-- real; scaled_inv_chi_square_lcdf; (reals y | reals nu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square_lcdf" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square_lcdf</code></a></strong><code>(reals y | reals nu, reals sigma)</code><br> The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-5f64b2f0bae60e336dba14a7efe683097fc8c8c8" class="index-entry"></a> <!-- real; scaled_inv_chi_square_lccdf; (reals y | reals nu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square_lccdf" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square_lccdf</code></a></strong><code>(reals y | reals nu, reals sigma)</code><br> The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-f0ccb76a8d7070c50483a9191a0177d55dc7a4a8" class="index-entry"></a> <!-- R; scaled_inv_chi_square_rng; (reals nu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#scaled_inv_chi_square_rng" class="unlink" title="Jump to index entry"><code>scaled_inv_chi_square_rng</code></a></strong><code>(reals nu, reals sigma)</code><br> Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="exponential-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="exponential-distribution">Exponential distribution</h2>
+<section id="probability-density-function-4" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-4">Probability density function</h3>
+<p>If inverse scale (rate) <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*}
+\text{Exponential}(y|\beta) = \beta \, \exp ( - \beta \, y ) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-4" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-4">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#exponential" class="unlink" title="Jump to index entry"><code>exponential</code></a></strong><code>(beta)</code></p>
+<p>Increment target log probability density with <code>exponential_lupdf(y | beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-659974bc00c9d41d7a429ae37133a99bb8b957a4" class="index-entry"></a> <!-- real; exponential ~; -->
+<p></p>
+</section>
+<section id="stan-functions-4" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-4">Stan functions</h3>
+<a id="index-entry-5338844e89484284ae797d5d7f2b46148d8e8d2c" class="index-entry"></a> <!-- real; exponential_lpdf; (reals y | reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exponential_lpdf" class="unlink" title="Jump to index entry"><code>exponential_lpdf</code></a></strong><code>(reals y | reals beta)</code><br> The log of the exponential density of y given inverse scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-fdea6af1db60962c86c57847a99b985330592f02" class="index-entry"></a> <!-- real; exponential_lupdf; (reals y | reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exponential_lupdf" class="unlink" title="Jump to index entry"><code>exponential_lupdf</code></a></strong><code>(reals y | reals beta)</code><br> The log of the exponential density of y given inverse scale beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-ad4cd5c7ba3c05b884083c95b23a472cffbc1a00" class="index-entry"></a> <!-- real; exponential_cdf; (reals y | reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exponential_cdf" class="unlink" title="Jump to index entry"><code>exponential_cdf</code></a></strong><code>(reals y | reals beta)</code><br> The exponential cumulative distribution function of y given inverse scale beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0677eab24885c7279772fbf058f8cb58eb12d88a" class="index-entry"></a> <!-- real; exponential_lcdf; (reals y | reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exponential_lcdf" class="unlink" title="Jump to index entry"><code>exponential_lcdf</code></a></strong><code>(reals y | reals beta)</code><br> The log of the exponential cumulative distribution function of y given inverse scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-7c9705bfaff832ebfe92cc4dcc784a77c1e1bd16" class="index-entry"></a> <!-- real; exponential_lccdf; (reals y | reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exponential_lccdf" class="unlink" title="Jump to index entry"><code>exponential_lccdf</code></a></strong><code>(reals y | reals beta)</code><br> The log of the exponential complementary cumulative distribution function of y given inverse scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-2fefb0b09b4b70f993267af296d92862bc9c900d" class="index-entry"></a> <!-- R; exponential_rng; (reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#exponential_rng" class="unlink" title="Jump to index entry"><code>exponential_rng</code></a></strong><code>(reals beta)</code><br> Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="gamma-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="gamma-distribution">Gamma distribution</h2>
+<section id="probability-density-function-5" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-5">Probability density function</h3>
+<p>If the shape parameter <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span> and the rate (or inverse scale) parameter <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y
+\in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*} \text{Gamma}(y|\alpha,\beta) =
+\frac{\beta^{\alpha}}      {\Gamma(\alpha)} \, y^{\alpha - 1}
+\exp(-\beta \, y) . \end{equation*}\]</span></p>
+<p>Under the shape and rate formulation of the Gamma distribution, <span class="math inline">\(\mathbb{E}[y] = \alpha / \beta\)</span> and <span class="math inline">\(\textrm{var}[y] = \alpha / \beta^2\)</span>.</p>
+</section>
+<section id="distribution-statement-5" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-5">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#gamma" class="unlink" title="Jump to index entry"><code>gamma</code></a></strong><code>(alpha, beta)</code></p>
+<p>Increment target log probability density with <code>gamma_lupdf(y | alpha, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-91a7d59d68ed36bf3b8d400643b7211047f9c616" class="index-entry"></a> <!-- real; gamma ~; -->
+<p></p>
+</section>
+<section id="stan-functions-5" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-5">Stan functions</h3>
+<a id="index-entry-76500ac07efdc8ff316f8c5d83738088c623861d" class="index-entry"></a> <!-- real; gamma_lpdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_lpdf" class="unlink" title="Jump to index entry"><code>gamma_lpdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the gamma density of y given shape alpha and inverse scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-4631a12c69e80e1a603f5e8455debc9086e93879" class="index-entry"></a> <!-- real; gamma_lupdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_lupdf" class="unlink" title="Jump to index entry"><code>gamma_lupdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-f4d7f090f9393c8fcb25cb5a0a93e6fbfeebbcc3" class="index-entry"></a> <!-- real; gamma_cdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_cdf" class="unlink" title="Jump to index entry"><code>gamma_cdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The cumulative gamma distribution function of y given shape alpha and inverse scale beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-6a62c558a39dc345eeafa81df49ccbb5e63cff39" class="index-entry"></a> <!-- real; gamma_lcdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_lcdf" class="unlink" title="Jump to index entry"><code>gamma_lcdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-7529e81d26b2ce290a96efca419a000b84883388" class="index-entry"></a> <!-- real; gamma_lccdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_lccdf" class="unlink" title="Jump to index entry"><code>gamma_lccdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-f7478aca325aa138c901aac6630912e23ca8a779" class="index-entry"></a> <!-- R; gamma_rng; (reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#gamma_rng" class="unlink" title="Jump to index entry"><code>gamma_rng</code></a></strong><code>(reals alpha, reals beta)</code><br> Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="inverse-gamma-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="inverse-gamma-distribution">Inverse gamma Distribution</h2>
+<section id="probability-density-function-6" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-6">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span> and <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y
+\in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*} \text{InvGamma}(y|\alpha,\beta) =
+\frac{\beta^{\alpha}}      {\Gamma(\alpha)} \ y^{-(\alpha + 1)} \,
+\exp \! \left( \! - \beta \, \frac{1}{y} \right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-6" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-6">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma" class="unlink" title="Jump to index entry"><code>inv_gamma</code></a></strong><code>(alpha, beta)</code></p>
+<p>Increment target log probability density with <code>inv_gamma_lupdf(y | alpha, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0a75a7a89dfd9fb418dfcd098f5f987d27c95da2" class="index-entry"></a> <!-- real; inv_gamma ~; -->
+<p></p>
+</section>
+<section id="stan-functions-6" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-6">Stan functions</h3>
+<a id="index-entry-7c937cfca51ac916d72ff10b1184a8d6d39ee650" class="index-entry"></a> <!-- real; inv_gamma_lpdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma_lpdf" class="unlink" title="Jump to index entry"><code>inv_gamma_lpdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the inverse gamma density of y given shape alpha and scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-c799b2a02063b44bf348b1e09d1a2bdbb8fefbd5" class="index-entry"></a> <!-- real; inv_gamma_lupdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma_lupdf" class="unlink" title="Jump to index entry"><code>inv_gamma_lupdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8b7859ab9bdd43c5cb4a55298dbada0eb100e006" class="index-entry"></a> <!-- real; inv_gamma_cdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma_cdf" class="unlink" title="Jump to index entry"><code>inv_gamma_cdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The inverse gamma cumulative distribution function of y given shape alpha and scale beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-24ddd0a34b2809b8f5afa3c540f1223c8cf683f8" class="index-entry"></a> <!-- real; inv_gamma_lcdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma_lcdf" class="unlink" title="Jump to index entry"><code>inv_gamma_lcdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-779e4be4cd1ce9f8858d662138e8af3a0a221e5e" class="index-entry"></a> <!-- real; inv_gamma_lccdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma_lccdf" class="unlink" title="Jump to index entry"><code>inv_gamma_lccdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-460487d2c9969e63b7c9f8ff329c0295b8807239" class="index-entry"></a> <!-- R; inv_gamma_rng; (reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_gamma_rng" class="unlink" title="Jump to index entry"><code>inv_gamma_rng</code></a></strong><code>(reals alpha, reals beta)</code><br> Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="weibull-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="weibull-distribution">Weibull distribution</h2>
+<section id="probability-density-function-7" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-7">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in [0,\infty)\)</span>, <span class="math display">\[\begin{equation*} \text{Weibull}(y|\alpha,\sigma) =
+\frac{\alpha}{\sigma} \, \left( \frac{y}{\sigma} \right)^{\alpha - 1}
+\, \exp \! \left( \! - \left( \frac{y}{\sigma} \right)^{\alpha}
+\right) . \end{equation*}\]</span></p>
+<p>Note that if <span class="math inline">\(Y \propto \text{Weibull}(\alpha,\sigma)\)</span>, then <span class="math inline">\(Y^{-1}
+\propto \text{Frechet}(\alpha,\sigma^{-1})\)</span>.</p>
+</section>
+<section id="distribution-statement-7" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-7">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#weibull" class="unlink" title="Jump to index entry"><code>weibull</code></a></strong><code>(alpha, sigma)</code></p>
+<p>Increment target log probability density with <code>weibull_lupdf(y | alpha, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-d1783373dc225fca3334f2daff416bfaf76d19fb" class="index-entry"></a> <!-- real; weibull ~; -->
+<p></p>
+</section>
+<section id="stan-functions-7" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
+<a id="index-entry-3797583b218047be3359306086c47d43b91a7cad" class="index-entry"></a> <!-- real; weibull_lpdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#weibull_lpdf" class="unlink" title="Jump to index entry"><code>weibull_lpdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Weibull density of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-4d40f28dddfa369222ba458b9ff111fdb7747664" class="index-entry"></a> <!-- real; weibull_lupdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#weibull_lupdf" class="unlink" title="Jump to index entry"><code>weibull_lupdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-bcc6a2519ae687354f1d256427ad98df385010f2" class="index-entry"></a> <!-- real; weibull_cdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#weibull_cdf" class="unlink" title="Jump to index entry"><code>weibull_cdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The Weibull cumulative distribution function of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-90d29f1f2935caa157a84fa85aff411976e008ff" class="index-entry"></a> <!-- real; weibull_lcdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#weibull_lcdf" class="unlink" title="Jump to index entry"><code>weibull_lcdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-1a56c82ab70897619b18238a305b239f0d09fb01" class="index-entry"></a> <!-- real; weibull_lccdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#weibull_lccdf" class="unlink" title="Jump to index entry"><code>weibull_lccdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-e84fd16978c29fe385bfa374bc3850c6a37a5347" class="index-entry"></a> <!-- R; weibull_rng; (reals alpha, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#weibull_rng" class="unlink" title="Jump to index entry"><code>weibull_rng</code></a></strong><code>(reals alpha, reals sigma)</code><br> Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="frechet-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="frechet-distribution">Frechet distribution</h2>
+<section id="probability-density-function-8" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-8">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*} \text{Frechet}(y|\alpha,\sigma) =
+\frac{\alpha}{\sigma} \, \left( \frac{y}{\sigma} \right)^{-\alpha - 1}
+\, \exp \! \left( \! - \left( \frac{y}{\sigma} \right)^{-\alpha}
+\right) . \end{equation*}\]</span></p>
+<p>Note that if <span class="math inline">\(Y \propto \text{Frechet}(\alpha,\sigma)\)</span>, then <span class="math inline">\(Y^{-1}
+\propto \text{Weibull}(\alpha,\sigma^{-1})\)</span>.</p>
+</section>
+<section id="distribution-statement-8" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-8">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#frechet" class="unlink" title="Jump to index entry"><code>frechet</code></a></strong><code>(alpha, sigma)</code></p>
+<p>Increment target log probability density with <code>frechet_lupdf(y | alpha, sigma)</code>.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-033a6fd0ca894c93b58f6808ff8fe3a63d1bb978" class="index-entry"></a> <!-- real; frechet ~; -->
+<p></p>
+</section>
+<section id="stan-functions-8" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-8">Stan functions</h3>
+<a id="index-entry-5b4868d7aafac4f3559c062d04dca87352e5dc4e" class="index-entry"></a> <!-- real; frechet_lpdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#frechet_lpdf" class="unlink" title="Jump to index entry"><code>frechet_lpdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Frechet density of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-99b375230fb5a80dcd8e6c9889b9da0b20cb1901" class="index-entry"></a> <!-- real; frechet_lupdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#frechet_lupdf" class="unlink" title="Jump to index entry"><code>frechet_lupdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-7b1c306049c2aeb7399a3d6df5ea3c13265def52" class="index-entry"></a> <!-- real; frechet_cdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#frechet_cdf" class="unlink" title="Jump to index entry"><code>frechet_cdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The Frechet cumulative distribution function of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-2e6b14eab76b40cbf5f51add826d06a805ae9757" class="index-entry"></a> <!-- real; frechet_lcdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#frechet_lcdf" class="unlink" title="Jump to index entry"><code>frechet_lcdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-81c59a2210fa846a77bc5ceeb512c5ab2403c36d" class="index-entry"></a> <!-- real; frechet_lccdf; (reals y | reals alpha, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#frechet_lccdf" class="unlink" title="Jump to index entry"><code>frechet_lccdf</code></a></strong><code>(reals y | reals alpha, reals sigma)</code><br> The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-d8574e1b5b0094a19eaec6f9bbc2b458ee6c80d2" class="index-entry"></a> <!-- R; frechet_rng; (reals alpha, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#frechet_rng" class="unlink" title="Jump to index entry"><code>frechet_rng</code></a></strong><code>(reals alpha, reals sigma)</code><br> Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="rayleigh-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="rayleigh-distribution">Rayleigh distribution</h2>
+<section id="probability-density-function-9" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-9">Probability density function</h3>
+<p>If <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in [0,\infty)\)</span>, <span class="math display">\[\begin{equation*}
+\text{Rayleigh}(y|\sigma) = \frac{y}{\sigma^2} \exp(-y^2 / 2\sigma^2)
+\!. \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-9" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-9">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#rayleigh" class="unlink" title="Jump to index entry"><code>rayleigh</code></a></strong><code>(sigma)</code></p>
+<p>Increment target log probability density with <code>rayleigh_lupdf(y | sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-21e80622249b881fc41605bf036fc4acba1d2e27" class="index-entry"></a> <!-- real; rayleigh ~; -->
+<p></p>
+</section>
+<section id="stan-functions-9" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-9">Stan functions</h3>
+<a id="index-entry-b04b39e49861434e1dc33eaab453c14e28da9f01" class="index-entry"></a> <!-- real; rayleigh_lpdf; (reals y | reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#rayleigh_lpdf" class="unlink" title="Jump to index entry"><code>rayleigh_lpdf</code></a></strong><code>(reals y | reals sigma)</code><br> The log of the Rayleigh density of y given scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-9778ca15ec9fc0fc7856485732abc302454e60c5" class="index-entry"></a> <!-- real; rayleigh_lupdf; (reals y | reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#rayleigh_lupdf" class="unlink" title="Jump to index entry"><code>rayleigh_lupdf</code></a></strong><code>(reals y | reals sigma)</code><br> The log of the Rayleigh density of y given scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-1a411f1904a7749524e9155b6695826fb8243d66" class="index-entry"></a> <!-- real; rayleigh_cdf; (real y | real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#rayleigh_cdf" class="unlink" title="Jump to index entry"><code>rayleigh_cdf</code></a></strong><code>(real y | real sigma)</code><br> The Rayleigh cumulative distribution of y given scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4782b996228bdd1d006a2f0c6a7406b741855120" class="index-entry"></a> <!-- real; rayleigh_lcdf; (real y | real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#rayleigh_lcdf" class="unlink" title="Jump to index entry"><code>rayleigh_lcdf</code></a></strong><code>(real y | real sigma)</code><br> The log of the Rayleigh cumulative distribution of y given scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-9d06d218d58232df14b1c128f0c42df1c1babaa8" class="index-entry"></a> <!-- real; rayleigh_lccdf; (real y | real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#rayleigh_lccdf" class="unlink" title="Jump to index entry"><code>rayleigh_lccdf</code></a></strong><code>(real y | real sigma)</code><br> The log of the Rayleigh complementary cumulative distribution of y given scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-97387dc941bc4a4869eee5819e8687df2f4ee96b" class="index-entry"></a> <!-- R; rayleigh_rng; (reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#rayleigh_rng" class="unlink" title="Jump to index entry"><code>rayleigh_rng</code></a></strong><code>(reals sigma)</code><br> Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="log-logistic-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="log-logistic-distribution">Log-logistic distribution</h2>
+<section id="probability-density-function-10" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-10">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha, \beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^+\)</span>, <span class="math display">\[\begin{equation*}
+\text{Log-Logistic}(y|\alpha,\beta) =
+\frac{\ \left(\frac{\beta}{\alpha}\right) \left(\frac{y}{\alpha}\right)^{\beta-1}\ }{\left(1 + \left(\frac{y}{\alpha}\right)^\beta\right)^2} .
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-10" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-10">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#loglogistic" class="unlink" title="Jump to index entry"><code>loglogistic</code></a></strong><code>(alpha, beta)</code></p>
+<p>Increment target log probability density with unnormalized version of <code>loglogistic_lpdf(y | alpha, beta)</code></p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-34bd0bbc32687356d26bf7bc8adb2ed107ee7c41" class="index-entry"></a> <!-- real; loglogistic ~; -->
+<p></p>
+</section>
+<section id="stan-functions-10" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-10">Stan functions</h3>
+<a id="index-entry-54448e697522e5d4ca36ca4b7acca93cda712054" class="index-entry"></a> <!-- real; loglogistic_lpdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#loglogistic_lpdf" class="unlink" title="Jump to index entry"><code>loglogistic_lpdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log of the log-logistic density of y given scale alpha and shape beta</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-b4193e37a39fc626b93a8d5b938675e6a5f682ef" class="index-entry"></a> <!-- real; loglogistic_cdf; (reals y | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#loglogistic_cdf" class="unlink" title="Jump to index entry"><code>loglogistic_cdf</code></a></strong><code>(reals y | reals alpha, reals beta)</code><br> The log-logistic cumulative distribution function of y given scale alpha and shape beta</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-d1b23b045cd0f829704373f8a06d3f077646b0d0" class="index-entry"></a> <!-- R; loglogistic_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#loglogistic_rng" class="unlink" title="Jump to index entry"><code>loglogistic_rng</code></a></strong><code>(reals alpha, reals beta)</code><br> Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.29</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/unbounded_continuous_distributions.html" class="pagination-link" aria-label="Unbounded Continuous Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Unbounded Continuous Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/positive_lower-bounded_distributions.html" class="pagination-link" aria-label="Positive Lower-Bounded Distributions">
+        <span class="nav-page-text">Positive Lower-Bounded Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/positive_lower-bounded_distributions.html b/docs/2_39/functions-reference/positive_lower-bounded_distributions.html
new file mode 100644
index 000000000..aa3c8aca2
--- /dev/null
+++ b/docs/2_39/functions-reference/positive_lower-bounded_distributions.html
@@ -0,0 +1,1452 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Positive Lower-Bounded Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/continuous_distributions_on_0_1.html" rel="next">
+<link href="../functions-reference/positive_continuous_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Positive Lower-Bounded Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/positive_lower-bounded_distributions.html">Positive Lower-Bounded Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#positive-lower-bounded-distributions" id="toc-positive-lower-bounded-distributions" class="nav-link active" data-scroll-target="#positive-lower-bounded-distributions">Positive Lower-Bounded Distributions</a>
+  <ul class="collapse">
+  <li><a href="#pareto-distribution" id="toc-pareto-distribution" class="nav-link" data-scroll-target="#pareto-distribution">Pareto distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#pareto-type-2-distribution" id="toc-pareto-type-2-distribution" class="nav-link" data-scroll-target="#pareto-type-2-distribution">Pareto type 2 distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#wiener-first-passage-time-distribution" id="toc-wiener-first-passage-time-distribution" class="nav-link" data-scroll-target="#wiener-first-passage-time-distribution">Wiener First Passage Time Distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-2" id="toc-probability-density-function-2" class="nav-link" data-scroll-target="#probability-density-function-2">Probability density function</a></li>
+  <li><a href="#cumulative-distribution-function" id="toc-cumulative-distribution-function" class="nav-link" data-scroll-target="#cumulative-distribution-function">Cumulative distribution function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  <li><a href="#boundaries" id="toc-boundaries" class="nav-link" data-scroll-target="#boundaries">Boundaries</a></li>
+  <li><a href="#vectorization" id="toc-vectorization" class="nav-link" data-scroll-target="#vectorization">Vectorization</a></li>
+  <li><a href="#tolerance-tuning" id="toc-tolerance-tuning" class="nav-link" data-scroll-target="#tolerance-tuning">Tolerance tuning</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_lower-bounded_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/positive_lower-bounded_distributions.html">Positive Lower-Bounded Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="positive-lower-bounded-distributions" class="level1">
+<h1>Positive Lower-Bounded Distributions</h1>
+<p>The positive lower-bounded probabilities have support on real values above some positive minimum value.</p>
+<section id="pareto-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="pareto-distribution">Pareto distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(y_{\text{min}} \in \mathbb{R}^+\)</span> and <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^+\)</span> with <span class="math inline">\(y \geq y_{\text{min}}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Pareto}(y|y_{\text{min}},\alpha) = \frac{\displaystyle \alpha\,y_{\text{min}}^\alpha}{\displaystyle y^{\alpha+1}}.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#pareto" class="unlink" title="Jump to index entry"><code>pareto</code></a></strong><code>(y_min, alpha)</code></p>
+<p>Increment target log probability density with <code>pareto_lupdf(y | y_min, alpha)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-3868686bd088fa3182f600afc3a7982dfadfd439" class="index-entry"></a> <!-- real; pareto ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-2b559a321ccbd5c4df2c32eafefcb7ed61d8497f" class="index-entry"></a> <!-- real; pareto_lpdf; (reals y | reals y_min, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_lpdf" class="unlink" title="Jump to index entry"><code>pareto_lpdf</code></a></strong><code>(reals y | reals y_min, reals alpha)</code><br> The log of the Pareto density of y given positive minimum value y_min and shape alpha</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-9908f14844c4a9e7fe15c5add847cd1baf8fa56c" class="index-entry"></a> <!-- real; pareto_lupdf; (reals y | reals y_min, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_lupdf" class="unlink" title="Jump to index entry"><code>pareto_lupdf</code></a></strong><code>(reals y | reals y_min, reals alpha)</code><br> The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-4abdf5cdce078735156bd724266dc2ea7f8b1d55" class="index-entry"></a> <!-- real; pareto_cdf; (reals y | reals y_min, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_cdf" class="unlink" title="Jump to index entry"><code>pareto_cdf</code></a></strong><code>(reals y | reals y_min, reals alpha)</code><br> The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4fb849c817175501195c6219da4b7f1cecad0b01" class="index-entry"></a> <!-- real; pareto_lcdf; (reals y | reals y_min, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_lcdf" class="unlink" title="Jump to index entry"><code>pareto_lcdf</code></a></strong><code>(reals y | reals y_min, reals alpha)</code><br> The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-579987e1abda763d062fecdf84a43f9c43e7b2fd" class="index-entry"></a> <!-- real; pareto_lccdf; (reals y | reals y_min, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_lccdf" class="unlink" title="Jump to index entry"><code>pareto_lccdf</code></a></strong><code>(reals y | reals y_min, reals alpha)</code><br> The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-33472469dac2a620b0f461365535f1f605f15104" class="index-entry"></a> <!-- R; pareto_rng; (reals y_min, reals alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#pareto_rng" class="unlink" title="Jump to index entry"><code>pareto_rng</code></a></strong><code>(reals y_min, reals alpha)</code><br> Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="pareto-type-2-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="pareto-type-2-distribution">Pareto type 2 distribution</h2>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span>, <span class="math inline">\(\lambda \in \mathbb{R}^+\)</span>, and <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \geq \mu\)</span>, <span class="math display">\[\begin{equation*}
+\mathrm{Pareto\_Type\_2}(y|\mu,\lambda,\alpha) =
+\ \frac{\alpha}{\lambda} \, \left( 1+\frac{y-\mu}{\lambda} \right)^{-(\alpha+1)} \! .
+\end{equation*}\]</span></p>
+<p>Note that the Lomax distribution is a Pareto Type 2 distribution with <span class="math inline">\(\mu=0\)</span>.</p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2" class="unlink" title="Jump to index entry"><code>pareto_type_2</code></a></strong><code>(mu, lambda, alpha)</code></p>
+<p>Increment target log probability density with <code>pareto_type_2_lupdf(y | mu, lambda, alpha)</code>.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-ae502c44f119dc495e9e3cb809d28da32af6d702" class="index-entry"></a> <!-- real; pareto_type_2 ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-a01b65ffac6ae18cc37fa6af7c644e99276c1787" class="index-entry"></a> <!-- real; pareto_type_2_lpdf; (reals y | reals mu, reals lambda, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2_lpdf" class="unlink" title="Jump to index entry"><code>pareto_type_2_lpdf</code></a></strong><code>(reals y | reals mu, reals lambda, reals alpha)</code><br> The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-62575702409a828f934e569a6d98dffa197f2b93" class="index-entry"></a> <!-- real; pareto_type_2_lupdf; (reals y | reals mu, reals lambda, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2_lupdf" class="unlink" title="Jump to index entry"><code>pareto_type_2_lupdf</code></a></strong><code>(reals y | reals mu, reals lambda, reals alpha)</code><br> The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-faf0b6f41480a1c429a34e564f63e9004c93fab1" class="index-entry"></a> <!-- real; pareto_type_2_cdf; (reals y | reals mu, reals lambda, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2_cdf" class="unlink" title="Jump to index entry"><code>pareto_type_2_cdf</code></a></strong><code>(reals y | reals mu, reals lambda, reals alpha)</code><br> The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-207a8840717cf656dcedb5daad63e6d01e8af9be" class="index-entry"></a> <!-- real; pareto_type_2_lcdf; (reals y | reals mu, reals lambda, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2_lcdf" class="unlink" title="Jump to index entry"><code>pareto_type_2_lcdf</code></a></strong><code>(reals y | reals mu, reals lambda, reals alpha)</code><br> The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-1df9330c369a563524ac6a3256c1a18befe4f9f8" class="index-entry"></a> <!-- real; pareto_type_2_lccdf; (reals y | reals mu, reals lambda, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2_lccdf" class="unlink" title="Jump to index entry"><code>pareto_type_2_lccdf</code></a></strong><code>(reals y | reals mu, reals lambda, reals alpha)</code><br> The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-53cbc03223fcc002f296906a7751b5808b49507a" class="index-entry"></a> <!-- R; pareto_type_2_rng; (reals mu, reals lambda, reals alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#pareto_type_2_rng" class="unlink" title="Jump to index entry"><code>pareto_type_2_rng</code></a></strong><code>(reals mu, reals lambda, reals alpha)</code><br> Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="wiener-first-passage-time-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="wiener-first-passage-time-distribution">Wiener First Passage Time Distribution</h2>
+<p>For an extended explanation of how to use the <code>wiener_lpdf</code> and <code>wiener_l[c]cdf_unnorm</code> functions, see <span class="citation" data-cites="Henrich2024">Henrich et al. (<a href="#ref-Henrich2024" role="doc-biblioref">2024</a>)</span>.</p>
+<section id="probability-density-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-2">Probability density function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, <span class="math inline">\(\tau \in \mathbb{R}^+\)</span>, <span class="math inline">\(\beta \in (0, 1)\)</span>, <span class="math inline">\(\delta \in \mathbb{R}\)</span>, <span class="math inline">\(s_{\delta} \in \mathbb{R}^{\geq 0}\)</span>, <span class="math inline">\(s_{\beta} \in [0, 1)\)</span>, and <span class="math inline">\(s_{\tau} \in \mathbb{R}^{\geq 0}\)</span> then for <span class="math inline">\(y &gt; \tau\)</span>,</p>
+<p><span class="math display">\[\begin{equation*}
+\begin{split}
+&amp;\text{Wiener}(y\mid \alpha,\tau,\beta,\delta,s_{\delta},s_{\beta},s_{\tau}) =
+\\
+&amp;\frac{1}{s_{\tau}}\int_{\tau}^{\tau+s_{\tau}}\frac{1}{s_{\beta}}\int_{\beta-\frac{1}{2}s_{\beta}}^{\beta+\frac{1}{2}s_{\beta}}\int_{-\infty}^{\infty} p_3(y-{\tau_0}\mid \alpha,\nu,\omega)
+\\
+&amp;\times \frac{1}{\sqrt{2\pi s_{\delta}^2}}\exp\Bigl(-\frac{(\nu-\delta)^2}{2s_{\delta}^2}\Bigr) \,d\nu \,d\omega \,d{\tau_0}=
+\\
+&amp;\frac{1}{s_{\tau}}\int_{\tau}^{\tau+s_{\tau}}\frac{1}{s_{\beta}}\int_{\beta-\frac{1}{2}s_{\beta}}^{\beta+\frac{1}{2}s_{\beta}} M\times p_3(y-{\tau_0}\mid \alpha,\nu,\omega) \,d\omega \,d{\tau_0},
+\end{split}
+\end{equation*}\]</span></p>
+<p>where <span class="math inline">\(p()\)</span> denotes the density function, and <span class="math inline">\(M\)</span> and <span class="math inline">\(p_3()\)</span> are defined, by using <span class="math inline">\(t:=y-{\tau_0}\)</span>, as</p>
+<p><span class="math display">\[\begin{equation*}
+M \coloneqq \frac{1}{\sqrt{1+s_{\delta}^2t}}\exp\Bigl(\alpha{\delta}\omega+\frac{\delta^2t}{2}+\frac{s_{\delta}^2\alpha^2\omega^2-2\alpha{\delta}\omega-\delta^2t}{2(1+s_{\delta}^2t)}\Bigr)\text{ and}
+\end{equation*}\]</span></p>
+<p><span class="math display">\[\begin{equation*}
+p_3(t\mid \alpha,\delta,\beta) \coloneqq \frac{1}{\alpha^2}\exp\Bigl(-\alpha\delta\beta-\frac{\delta^2t}{2}\Bigr)f(\frac{t}{\alpha^2}\mid 0,1,\beta),
+\end{equation*}\]</span></p>
+<p>where <span class="math inline">\(f(t^*=\frac{t}{\alpha^2}\mid0,1,\beta)\)</span> can be specified in two ways:</p>
+<p><span class="math display">\[\begin{equation*}
+f_l(t^*\mid 0,1,\beta) = \sum_{k=1}^\infty k\pi \exp\Bigl(-\frac{k^2\pi^2t^*}{2}\Bigr)\sin(k\pi \beta)\text{ and}
+\end{equation*}\]</span></p>
+<p><span class="math display">\[\begin{equation*}
+f_s(t^*\mid0,1,\beta) = \sum_{k=-\infty}^\infty \frac{1}{\sqrt{2\pi(t^*)^3}}(\beta+2k) \exp\Bigl(-\frac{(\beta+2k)^2}{2t^*}\Bigr).
+\end{equation*}\]</span></p>
+<p>Which of these is used in the computations depends on which expression requires the smaller number of components <span class="math inline">\(k\)</span> to guarantee a pre-specified precision</p>
+<p>In the case where <span class="math inline">\(s_{\delta}\)</span>, <span class="math inline">\(s_{\beta}\)</span>, and <span class="math inline">\(s_{\tau}\)</span> are all <span class="math inline">\(0\)</span>, this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): <span class="math display">\[\begin{equation*}
+\text{Wiener}(y|\alpha, \tau, \beta, \delta) =
+\frac{\alpha}{(y-\tau)^{3/2}} \exp \! \left(- \delta \alpha \beta -
+\frac{\delta^2(y-\tau)}{2}\right) \sum_{k = - \infty}^{\infty} (2k +
+\beta) \phi \! \left(\frac{(2k + \beta)\alpha }{\sqrt{y - \tau}}\right),
+\end{equation*}\]</span> where <span class="math inline">\(\phi(x)\)</span> denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): <span class="math display">\[\begin{equation*}
+\text{Wiener}(y|\alpha, \tau, \beta, \delta) =
+\frac{\pi}{\alpha^2} \exp \! \left(- \delta \alpha \beta -
+\frac{\delta^2(y-\tau)}{2}\right) \sum_{k = 1}^{\infty} k \exp \! \left(-\frac{k^2\pi^2(y-\tau)};
+{2\alpha^2}\right) \sin \!(k\pi\beta)
+\end{equation*}\]</span> see <span class="citation" data-cites="Feller1968">(<a href="#ref-Feller1968" role="doc-biblioref">Feller 1968</a>)</span>, <span class="citation" data-cites="NavarroFuss2009">(<a href="#ref-NavarroFuss2009" role="doc-biblioref">Navarro and Fuss 2009</a>)</span>.</p>
+</section>
+<section id="cumulative-distribution-function" class="level3">
+<h3 class="anchored" data-anchor-id="cumulative-distribution-function">Cumulative distribution function</h3>
+<p>For the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.</p>
+<p>Let <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(\tau\)</span>, <span class="math inline">\(\beta\)</span>, <span class="math inline">\(\delta\)</span>, <span class="math inline">\(s_{\delta}\)</span>, <span class="math inline">\(s_{\beta}\)</span>, <span class="math inline">\(s_{\tau}\)</span> and <span class="math inline">\(y\)</span> be as above.</p>
+<p>The formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in <span class="math inline">\(\tau\)</span> for the time being) at the upper boundary is stated as follows:</p>
+<p><span class="math display">\[\begin{equation}
+F(y|\alpha, \beta, \delta) = P(\alpha, \beta, \delta) -
+    \exp\left(\delta\alpha(1-\beta)-\frac{\delta^2 y}{2}\right)F_l(y|\alpha,\beta,\delta),
+\end{equation}\]</span> where <span class="math inline">\(P(\alpha,\beta,\delta)\)</span> is the probability to hit the upper boundary, defined as</p>
+<p><span class="math display">\[\begin{equation}
+P(\alpha, \beta, \delta) =
+\begin{cases}
+    \frac{1-\exp(2\delta \alpha \beta)}{\exp(-2\delta \alpha(1-\beta)) - \exp(2\delta \alpha \beta)}, &amp; \text{for } \delta\neq 0 \\
+    \beta, &amp; \text{for } \delta=0,
+\end{cases}
+\end{equation}\]</span></p>
+<p>and</p>
+<p><span class="math display">\[\begin{equation}
+F_l(y|\alpha, \beta, \delta) =
+    \frac{2\pi}{\alpha^2}\sum_{k=1}^{\infty}{\frac{k\sin{k\pi(1-\beta)}}{\delta^2+(k\pi)^2/\alpha^2}}\exp(-\frac{k^2\pi^2y}{2\alpha^2}).
+\end{equation}\]</span></p>
+<p>The formula for the small-time cdf at the upper boundary is stated as follows:</p>
+<p><span class="math display">\[\begin{equation}
+F(y|\alpha,\beta,\delta) = \exp\left(\delta \alpha(1-\beta) -\frac{\delta^2y}{2}\right)F_s(y|\alpha, \beta,\delta),
+\end{equation}\]</span> where</p>
+<p><span class="math display">\[\begin{equation}
+F_s(y|\alpha,\beta,\delta) = \sum_{k=0}^{\infty}(-1)^k\phi\left(\frac{\alpha(k+\beta^{*}_k)}
+    {\sqrt{y}} \right) \times \left( R \left(\frac{\alpha(k+\beta^{*}_k)+\delta y}{\sqrt{y}} \right)
+        + R \left(\frac{\alpha(k+\beta^*_k)-\delta y}{\sqrt{y}} \right)\right),
+\end{equation}\]</span></p>
+<p>where <span class="math inline">\(\beta^*_k=(1-\beta)\)</span> for <span class="math inline">\(k\)</span> even, <span class="math inline">\(\beta^*_k=\beta\)</span> for <span class="math inline">\(k\)</span> odd, and <span class="math inline">\(R\)</span> is Mill’s ratio.</p>
+<p>The cdf for the lower boundary is <span class="math inline">\(F(y|\alpha,1-\beta,-\delta)\)</span></p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#wiener" class="unlink" title="Jump to index entry"><code>wiener</code></a></strong><code>(alpha, tau, beta, delta)</code></p>
+<p>Increment target log probability density with <code>wiener_lupdf(y | alpha, tau, beta, delta)</code>.</p>
+<small><i>Available since 2.7</i></small>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#wiener" class="unlink" title="Jump to index entry"><code>wiener</code></a></strong><code>(alpha, tau, beta, delta, var_delta)</code> Increment target log probability density with <code>wiener_lupdf(y | alpha, tau, beta, delta, var_delta)</code>.</p>
+<small><i>Available since 2.35</i></small>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#wiener" class="unlink" title="Jump to index entry"><code>wiener</code></a></strong><code>(alpha, tau, beta, delta, var_delta, var_beta, var_tau)</code> Increment target log probability density with <code>wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau)</code>.</p>
+<small><i>Available since 2.35</i></small>
+<a id="index-entry-a8609b3c51ac8487f51a5e8a445e626015dad49f" class="index-entry"></a> <!-- real; wiener ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-7c21d286a5cd6e11195df3046653f79a72c95e63" class="index-entry"></a> <!-- real; wiener_lpdf; (reals y | reals alpha, reals tau, reals beta, reals delta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lpdf" class="unlink" title="Jump to index entry"><code>wiener_lpdf</code></a></strong><code>(reals y | reals alpha, reals tau, reals beta, reals delta)</code><br> The log of the Wiener first passage time density of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, and drift rate <code>delta</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-dd5110522b6a5fb9baac9dab11ad3dca296d7a24" class="index-entry"></a> <!-- real; wiener_lpdf; (real y | real alpha, real tau, real beta, real delta, real var_delta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lpdf" class="unlink" title="Jump to index entry"><code>wiener_lpdf</code></a></strong><code>(real y | real alpha, real tau, real beta, real delta, real var_delta)</code><br> The log of the Wiener first passage time density of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, drift rate <code>delta</code>, and inter-trial drift rate variability <code>var_delta</code>.</p>
+<p>Setting <code>var_delta</code> to <code>0</code> recovers the 4-parameter signature above.</p>
+<small><i>Available since 2.35</i></small>
+<a id="index-entry-60f0354d0a8198be7994a9f5d5cab49f31a26aa1" class="index-entry"></a> <!-- real; wiener_lpdf; (real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lpdf" class="unlink" title="Jump to index entry"><code>wiener_lpdf</code></a></strong><code>(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)</code><br> The log of the Wiener first passage time density of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, drift rate <code>delta</code>, inter-trial drift rate variability <code>var_delta</code>, inter-trial variability of the starting point (bias) <code>var_beta</code>, and inter-trial variability of the non-decision time <code>var_tau</code>.</p>
+<p>Setting <code>var_delta</code>, <code>var_beta</code>, and <code>var_tau</code> to <code>0</code> recovers the 4-parameter signature above.</p>
+<small><i>Available since 2.35</i></small>
+<a id="index-entry-9511656e972acbc5b9054dabc90abaa4214250d3" class="index-entry"></a> <!-- real; wiener_lupdf; (reals y | reals alpha, reals tau, reals beta, reals delta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lupdf" class="unlink" title="Jump to index entry"><code>wiener_lupdf</code></a></strong><code>(reals y | reals alpha, reals tau, reals beta, reals delta)</code><br> The log of the Wiener first passage time density of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, and drift rate <code>delta</code>, dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-27fbddb7c175dbc42b3a77639443abb89b1c93ba" class="index-entry"></a> <!-- real; wiener_lupdf; (real y | real alpha, real tau, real beta, real delta, real var_delta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lupdf" class="unlink" title="Jump to index entry"><code>wiener_lupdf</code></a></strong><code>(real y | real alpha, real tau, real beta, real delta, real var_delta)</code><br> The log of the Wiener first passage time density of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, drift rate <code>delta</code>, and inter-trial drift rate variability <code>var_delta</code>, dropping constant additive terms.</p>
+<p>Setting <code>var_delta</code> to <code>0</code> recovers the 4-parameter signature above.</p>
+<small><i>Available since 2.35</i></small>
+<a id="index-entry-330634d3f68975f72410dd323545f5801a409119" class="index-entry"></a> <!-- real; wiener_lupdf; (real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lupdf" class="unlink" title="Jump to index entry"><code>wiener_lupdf</code></a></strong><code>(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)</code><br> The log of the Wiener first passage time density of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, drift rate <code>delta</code>, inter-trial drift rate variability <code>var_delta</code>, inter-trial variability of the starting point (bias) <code>var_beta</code>, and inter-trial variability of the non-decision time <code>var_tau</code>, dropping constant additive terms.</p>
+<p>Setting <code>var_delta</code>, <code>var_beta</code>, and <code>var_tau</code> to <code>0</code> recovers the 4-parameter signature above.</p>
+<small><i>Available since 2.35</i></small>
+<hr>
+<p><strong>Note:</strong> The <code>lcdf</code> and <code>lccdf</code> functions for the <code>wiener</code> distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper <a href="#boundaries">boundary</a>.</p>
+<p>Similarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical <span class="math inline">\(1 - \textrm{cdf}\)</span>.</p>
+<a id="index-entry-ffe70852994a9eadec88b671cd728e98feffb122" class="index-entry"></a> <!-- real; wiener_lcdf_unnorm; (real y, real alpha, real tau, real beta, real delta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lcdf_unnorm" class="unlink" title="Jump to index entry"><code>wiener_lcdf_unnorm</code></a></strong><code>(real y, real alpha, real tau, real beta, real delta)</code><br></p>
+<p>The log of the cumulative distribution function (cdf) of the Wiener distribution of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, and drift rate <code>delta</code>.</p>
+<small><i>Available since 2.38</i></small>
+<a id="index-entry-fd7dd400c38dd551f61ba9778cad3aefbb675a5e" class="index-entry"></a> <!-- real; wiener_lccdf_unnorm; (real y, real alpha, real tau, real beta, real delta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lccdf_unnorm" class="unlink" title="Jump to index entry"><code>wiener_lccdf_unnorm</code></a></strong><code>(real y, real alpha, real tau, real beta, real delta)</code><br></p>
+<p>The log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, and drift rate <code>delta</code>.</p>
+<small><i>Available since 2.38</i></small>
+<a id="index-entry-59df6a6c983bf3f79c62ff8031367686b0ffa70a" class="index-entry"></a> <!-- real; wiener_lcdf_unnorm; (real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lcdf_unnorm" class="unlink" title="Jump to index entry"><code>wiener_lcdf_unnorm</code></a></strong><code>(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)</code><br></p>
+<p>The log of the cumulative distribution function (cdf) of the Wiener distribution of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, drift rate <code>delta</code>, inter-trial drift rate variability <code>var_delta</code>, inter-trial variability of the starting point (bias) <code>var_beta</code>, and inter-trial variability of the non-decision time <code>var_tau</code>.</p>
+<small><i>Available since 2.38</i></small>
+<a id="index-entry-d195ea5f4d8eba49c56313a0623fd5deb56d02fd" class="index-entry"></a> <!-- real; wiener_lccdf_unnorm; (real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#wiener_lccdf_unnorm" class="unlink" title="Jump to index entry"><code>wiener_lccdf_unnorm</code></a></strong><code>(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)</code><br></p>
+<p>The log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of <code>y</code> given boundary separation <code>alpha</code>, non-decision time <code>tau</code>, starting point <code>beta</code>, drift rate <code>delta</code>, inter-trial drift rate variability <code>var_delta</code>, inter-trial variability of the starting point (bias) <code>var_beta</code>, and inter-trial variability of the non-decision time <code>var_tau</code>.</p>
+<small><i>Available since 2.38</i></small>
+</section>
+<section id="boundaries" class="level3">
+<h3 class="anchored" data-anchor-id="boundaries">Boundaries</h3>
+<p>Stan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use <span class="math display">\[\begin{equation*}
+\text{Wiener}(y | \alpha, \tau, 1 - \beta, - \delta)
+\end{equation*}\]</span> For more details, see the appendix of <span class="citation" data-cites="Vandekerckhove-Wabersich:2014">Vandekerckhove and Wabersich (<a href="#ref-Vandekerckhove-Wabersich:2014" role="doc-biblioref">2014</a>)</span>.</p>
+</section>
+<section id="vectorization" class="level3">
+<h3 class="anchored" data-anchor-id="vectorization">Vectorization</h3>
+<p>The 5- and 7-argument forms of the <code>wiener</code> distribution functions (listed above as recieving only <code>real</code>) are implemented in such a way where they can be fully <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized</a>, but currently only versions that accept all <code>real</code> and all <code>vector</code> arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by <a href="https://github.com/stan-dev/stanc3/issues/new">opening an issue</a>.</p>
+</section>
+<section id="tolerance-tuning" class="level3">
+<h3 class="anchored" data-anchor-id="tolerance-tuning">Tolerance tuning</h3>
+<p>The 5- and 7-argument forms of the <code>wiener</code> distribution functions can also accept an additional <code>data real</code> argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to <code>1e-4</code> for the density and <code>1e-8</code> for the cdf functions.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Feller1968" class="csl-entry" role="listitem">
+Feller, William. 1968. <em>An Introduction to Probability Theory and Its Applications</em>. Vol. 1. 3. Wiley, New York.
+</div>
+<div id="ref-Henrich2024" class="csl-entry" role="listitem">
+Henrich, Franziska, Raphael Hartmann, Valentin Pratz, Andreas Voss, and Karl Christoph Klauer. 2024. <span>“The Seven-Parameter Diffusion Model: An Implementation in Stan for Bayesian Analyses.”</span> <em>Behavior Research Methods</em> 56 (4): 3102–16. <a href="https://doi.org/10.3758/s13428-023-02179-1">https://doi.org/10.3758/s13428-023-02179-1</a>.
+</div>
+<div id="ref-NavarroFuss2009" class="csl-entry" role="listitem">
+Navarro, Danielle J, and Ian G Fuss. 2009. <span>“Fast and Accurate Calculations for First-Passage Times in <span>W</span>iener Diffusion Models.”</span> <em>Journal of Mathematical Psychology</em> 53 (4): 222–30.
+</div>
+<div id="ref-Vandekerckhove-Wabersich:2014" class="csl-entry" role="listitem">
+Vandekerckhove, Joachim, and Dominik Wabersich. 2014. <span>“The <span>R</span><span>W</span>iener Package: An <span>R</span> Package Providing Distribution Functions for the <span>W</span>iener Diffusion Model.”</span> <em>The R Journal</em> 6/1. <a href="http://journal.r-project.org/archive/2014-1/vandekerckhove-wabersich.pdf">http://journal.r-project.org/archive/2014-1/vandekerckhove-wabersich.pdf</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/positive_continuous_distributions.html" class="pagination-link" aria-label="Positive Continuous Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Positive Continuous Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/continuous_distributions_on_0_1.html" class="pagination-link" aria-label="Continuous Distributions on [0, 1]">
+        <span class="nav-page-text">Continuous Distributions on [0, 1]</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_lower-bounded_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/real-valued_basic_functions.html b/docs/2_39/functions-reference/real-valued_basic_functions.html
new file mode 100644
index 000000000..96f0b6ebe
--- /dev/null
+++ b/docs/2_39/functions-reference/real-valued_basic_functions.html
@@ -0,0 +1,2260 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Real-Valued Basic Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/complex-valued_basic_functions.html" rel="next">
+<link href="../functions-reference/integer-valued_basic_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Real-Valued Basic Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/real-valued_basic_functions.html">Real-Valued Basic Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#real-valued-basic-functions" id="toc-real-valued-basic-functions" class="nav-link active" data-scroll-target="#real-valued-basic-functions">Real-Valued Basic Functions</a>
+  <ul class="collapse">
+  <li><a href="#fun-vectorization" id="toc-fun-vectorization" class="nav-link" data-scroll-target="#fun-vectorization">Vectorization of real-valued functions</a>
+  <ul class="collapse">
+  <li><a href="#unary-function-vectorization" id="toc-unary-function-vectorization" class="nav-link" data-scroll-target="#unary-function-vectorization">Unary function vectorization</a></li>
+  <li><a href="#binary-function-vectorization" id="toc-binary-function-vectorization" class="nav-link" data-scroll-target="#binary-function-vectorization">Binary function vectorization</a></li>
+  </ul></li>
+  <li><a href="#built-in-constants" id="toc-built-in-constants" class="nav-link" data-scroll-target="#built-in-constants">Mathematical constants</a></li>
+  <li><a href="#special-values" id="toc-special-values" class="nav-link" data-scroll-target="#special-values">Special values</a></li>
+  <li><a href="#get-log-prob" id="toc-get-log-prob" class="nav-link" data-scroll-target="#get-log-prob">Log probability function</a></li>
+  <li><a href="#logical-functions" id="toc-logical-functions" class="nav-link" data-scroll-target="#logical-functions">Logical functions</a>
+  <ul class="collapse">
+  <li><a href="#comparison-operators" id="toc-comparison-operators" class="nav-link" data-scroll-target="#comparison-operators">Comparison operators</a></li>
+  <li><a href="#boolean-operators" id="toc-boolean-operators" class="nav-link" data-scroll-target="#boolean-operators">Boolean operators</a></li>
+  <li><a href="#logical-functions-1" id="toc-logical-functions-1" class="nav-link" data-scroll-target="#logical-functions-1">Logical functions</a></li>
+  </ul></li>
+  <li><a href="#real-valued-arithmetic-operators" id="toc-real-valued-arithmetic-operators" class="nav-link" data-scroll-target="#real-valued-arithmetic-operators">Real-valued arithmetic operators</a>
+  <ul class="collapse">
+  <li><a href="#binary-infix-operators" id="toc-binary-infix-operators" class="nav-link" data-scroll-target="#binary-infix-operators">Binary infix operators</a></li>
+  <li><a href="#unary-prefix-operators" id="toc-unary-prefix-operators" class="nav-link" data-scroll-target="#unary-prefix-operators">Unary prefix operators</a></li>
+  </ul></li>
+  <li><a href="#step-functions" id="toc-step-functions" class="nav-link" data-scroll-target="#step-functions">Step-like functions</a>
+  <ul class="collapse">
+  <li><a href="#absolute-value-functions" id="toc-absolute-value-functions" class="nav-link" data-scroll-target="#absolute-value-functions">Absolute value functions</a></li>
+  <li><a href="#bounds-functions" id="toc-bounds-functions" class="nav-link" data-scroll-target="#bounds-functions">Bounds functions</a></li>
+  <li><a href="#arithmetic-functions" id="toc-arithmetic-functions" class="nav-link" data-scroll-target="#arithmetic-functions">Arithmetic functions</a></li>
+  <li><a href="#rounding-functions" id="toc-rounding-functions" class="nav-link" data-scroll-target="#rounding-functions">Rounding functions</a></li>
+  </ul></li>
+  <li><a href="#power-and-logarithm-functions" id="toc-power-and-logarithm-functions" class="nav-link" data-scroll-target="#power-and-logarithm-functions">Power and logarithm functions</a></li>
+  <li><a href="#trigonometric-functions" id="toc-trigonometric-functions" class="nav-link" data-scroll-target="#trigonometric-functions">Trigonometric functions</a></li>
+  <li><a href="#hyperbolic-trigonometric-functions" id="toc-hyperbolic-trigonometric-functions" class="nav-link" data-scroll-target="#hyperbolic-trigonometric-functions">Hyperbolic trigonometric functions</a></li>
+  <li><a href="#link-functions" id="toc-link-functions" class="nav-link" data-scroll-target="#link-functions">Link functions</a></li>
+  <li><a href="#Phi-function" id="toc-Phi-function" class="nav-link" data-scroll-target="#Phi-function">Probability-related functions</a>
+  <ul class="collapse">
+  <li><a href="#normal-cumulative-distribution-functions" id="toc-normal-cumulative-distribution-functions" class="nav-link" data-scroll-target="#normal-cumulative-distribution-functions">Normal cumulative distribution functions</a></li>
+  <li><a href="#other-probability-related-functions" id="toc-other-probability-related-functions" class="nav-link" data-scroll-target="#other-probability-related-functions">Other probability-related functions</a></li>
+  </ul></li>
+  <li><a href="#betafun" id="toc-betafun" class="nav-link" data-scroll-target="#betafun">Combinatorial functions</a></li>
+  <li><a href="#composed-functions" id="toc-composed-functions" class="nav-link" data-scroll-target="#composed-functions">Composed functions</a></li>
+  <li><a href="#special-functions" id="toc-special-functions" class="nav-link" data-scroll-target="#special-functions">Special functions</a></li>
+  <li><a href="#hypergeometric-functions" id="toc-hypergeometric-functions" class="nav-link" data-scroll-target="#hypergeometric-functions">Hypergeometric Functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/real-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/real-valued_basic_functions.html">Real-Valued Basic Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="real-valued-basic-functions" class="level1">
+<h1>Real-Valued Basic Functions</h1>
+<p>This chapter describes built-in functions that take zero or more real or integer arguments and return real values.</p>
+<section id="fun-vectorization" class="level2">
+<h2 class="anchored" data-anchor-id="fun-vectorization">Vectorization of real-valued functions</h2>
+<p>Although listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.</p>
+<section id="unary-function-vectorization" class="level3">
+<h3 class="anchored" data-anchor-id="unary-function-vectorization">Unary function vectorization</h3>
+<p>Many of Stan’s unary functions can be applied to any argument type. For example, the exponential function, <code>exp</code>, can be applied to <code>real</code> arguments or arrays of <code>real</code> arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.</p>
+<section id="real-and-real-array-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="real-and-real-array-arguments">Real and real array arguments</h4>
+<p>When applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like <code>exp()</code> are defined elementwise. For example,</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> <span class="co">// declare some variables for arguments</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> x0;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">real</span> x1;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">7</span>] <span class="dt">real</span> x2;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a> <span class="co">// ...</span></span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a> <span class="co">// declare some variables for results</span></span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> y0;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">real</span> y1;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">7</span>] <span class="dt">real</span> y2;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a> <span class="co">// ...</span></span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a> <span class="co">// calculate and assign results</span></span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a> y0 = exp(x0);</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a> y1 = exp(x1);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a> y2 = exp(x2);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When <code>exp</code> is applied to an array, it applies elementwise. For example, the statement above,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a> y2 = exp(x2);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>produces the same result for <code>y2</code> as the explicit loop</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">4</span>) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">7</span>) {</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>    y2[i, j] = exp(x2[i, j]);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="vector-and-matrix-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="vector-and-matrix-arguments">Vector and matrix arguments</h4>
+<p>Vectorized functions also apply elementwise to vectors and matrices. For example,</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">7</span>] xrv;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">20</span>] xm;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">7</span>] yrv;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">20</span>] ym;</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a> yv = exp(xv);</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a> yrv = exp(xrv);</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a> ym = exp(xm);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Arrays of vectors and matrices work the same way. For example,</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">matrix</span>[<span class="dv">17</span>, <span class="dv">93</span>] u;</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">matrix</span>[<span class="dv">17</span>, <span class="dv">93</span>] z;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a> z = exp(u);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After this has been executed, <code>z[i, j, k]</code> will be equal to <code>exp(u[i, j, k])</code>.</p>
+</section>
+<section id="integer-and-integer-array-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="integer-and-integer-array-arguments">Integer and integer array arguments</h4>
+<p>Integer arguments are promoted to real values in vectorized unary functions. Thus if <code>n</code> is of type <code>int</code>, <code>exp(n)</code> is of type <code>real</code>. Arrays work the same way, so that if <code>n2</code> is a one dimensional array of integers, then <code>exp(n2)</code> will be a one-dimensional array of reals with the same number of elements as <code>n2</code>. For example,</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">23</span>] <span class="dt">int</span> n1;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">23</span>] <span class="dt">real</span> z1;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a> z1 = exp(n1);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It would be illegal to try to assign <code>exp(n1)</code> to an array of integers; the return type is a real array.</p>
+</section>
+</section>
+<section id="binary-function-vectorization" class="level3">
+<h3 class="anchored" data-anchor-id="binary-function-vectorization">Binary function vectorization</h3>
+<p>Like the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.</p>
+<section id="scalar-and-scalar-array-arguments" class="level4">
+<h4 class="anchored" data-anchor-id="scalar-and-scalar-array-arguments">Scalar and scalar array arguments</h4>
+<p>When applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like <code>pow()</code> are defined elementwise. For example,</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a> <span class="co">// declare some variables for arguments</span></span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> x00;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> x01;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">real</span> x10;</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>]<span class="dt">real</span> x11;</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">7</span>] <span class="dt">real</span> x20;</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">7</span>] <span class="dt">real</span> x21;</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a> <span class="co">// ...</span></span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a> <span class="co">// declare some variables for results</span></span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> y0;</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">real</span> y1;</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">7</span>] <span class="dt">real</span> y2;</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a> <span class="co">// ...</span></span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a> <span class="co">// calculate and assign results</span></span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a> y0 = pow(x00, x01);</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a> y1 = pow(x10, x11);</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a> y2 = pow(x20, x21);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When <code>pow</code> is applied to two arrays, it applies elementwise. For example, the statement above,</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a> y2 = pow(x20, x21);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>produces the same result for <code>y2</code> as the explicit loop</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">4</span>) {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">7</span>) {</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>    y2[i, j] = pow(x20[i, j], x21[i, j]);</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Alternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>y2 = pow(x20, x00);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>produces the same result for <code>y2</code> as the explicit loop:</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">4</span>) {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">7</span>) {</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>    y2[i, j] = pow(x20[i, j], x00);</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="vector-and-matrix-arguments-1" class="level4">
+<h4 class="anchored" data-anchor-id="vector-and-matrix-arguments-1">Vector and matrix arguments</h4>
+<p>Vectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> x00;</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv00;</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv01;</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">7</span>] xrv;</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">20</span>] xm;</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv;</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">7</span>] yrv;</span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">20</span>] ym;</span>
+<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a> yv = pow(xv00, xv01);</span>
+<span id="cb12-12"><a href="#cb12-12" aria-hidden="true" tabindex="-1"></a> yrv = pow(xrv, x00);</span>
+<span id="cb12-13"><a href="#cb12-13" aria-hidden="true" tabindex="-1"></a> ym = pow(x00, xm);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Arrays of vectors and matrices work the same way. For example,</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">matrix</span>[<span class="dv">17</span>, <span class="dv">93</span>] u;</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">matrix</span>[<span class="dv">17</span>, <span class="dv">93</span>] z;</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a> z = pow(u, x00);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After this has been executed, <code>z[i, j, k]</code> will be equal to <code>pow(u[i, j, k], x00)</code>.</p>
+</section>
+<section id="input-return-types" class="level4">
+<h4 class="anchored" data-anchor-id="input-return-types">Input &amp; return types</h4>
+<p>Vectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv;</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">7</span>] xrv;</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">20</span>] xm;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv = pow(xv, xv)</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">7</span>] yrv = pow(xrv, xrv)</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">20</span>] = pow(xm, xm)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>But the following statements are not:</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv;</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">7</span>] xv2;</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">row_vector</span>[<span class="dv">5</span>] xrv;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a> <span class="co">// Cannot mix different types</span></span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv = pow(xv, xrv)</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a> <span class="co">// Cannot mix different sizes of the same type</span></span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv = pow(xv, xv2)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>While the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., <code>bessel_first_kind</code>). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv;</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">5</span>, <span class="dv">5</span>] xm;</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">int</span> xi;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">5</span>] <span class="dt">int</span> xii;</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv = bessel_first_kind(xi, xv);</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">5</span>, <span class="dv">5</span>] ym = bessel_first_kind(xii, xm);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Whereas these are not:</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] xv;</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">5</span>, <span class="dv">5</span>] xm;</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a> <span class="dt">array</span>[<span class="dv">7</span>] <span class="dt">int</span> xi;</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a> <span class="co">// Dimensions of containers do not match</span></span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span>[<span class="dv">5</span>] yv = bessel_first_kind(xi, xv);</span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a> <span class="co">// Function requires first argument be an integer type</span></span>
+<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a> <span class="dt">matrix</span>[<span class="dv">5</span>, <span class="dv">5</span>] ym = bessel_first_kind(xm, xm);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+</section>
+<section id="built-in-constants" class="level2">
+<h2 class="anchored" data-anchor-id="built-in-constants">Mathematical constants</h2>
+<p>Constants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant <span class="math inline">\(\pi\)</span> must be written in a Stan program as <code>pi()</code>.</p>
+<a id="index-entry-53d994dc279b92ee9a153eb72068c45e7587230e" class="index-entry"></a> <!-- real; pi; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pi" class="unlink" title="Jump to index entry"><code>pi</code></a></strong><code>()</code><br> <span class="math inline">\(\pi\)</span>, the ratio of a circle’s circumference to its diameter</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c9e1664eab4cc021df10c6902738cce047b7d8f8" class="index-entry"></a> <!-- real; e; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#e" class="unlink" title="Jump to index entry"><code>e</code></a></strong><code>()</code><br> <span class="math inline">\(e\)</span>, the base of the natural logarithm</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-1ded3db8572ff7287e4d61f28f486873fd4b76a1" class="index-entry"></a> <!-- real; sqrt2; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#sqrt2" class="unlink" title="Jump to index entry"><code>sqrt2</code></a></strong><code>()</code><br> The square root of 2</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-8e60dd0524fa17b2bb9c1df45ec2cb91f313d8f6" class="index-entry"></a> <!-- real; log2; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log2" class="unlink" title="Jump to index entry"><code>log2</code></a></strong><code>()</code><br> The natural logarithm of 2</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-58b920bd8ecd5758726663c1663723631b6e1040" class="index-entry"></a> <!-- real; log10; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log10" class="unlink" title="Jump to index entry"><code>log10</code></a></strong><code>()</code><br> The natural logarithm of 10</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="special-values" class="level2">
+<h2 class="anchored" data-anchor-id="special-values">Special values</h2>
+<a id="index-entry-00757653dd74e4f55dcd63fbdfb21a9e5005469c" class="index-entry"></a> <!-- real; not_a_number; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#not_a_number" class="unlink" title="Jump to index entry"><code>not_a_number</code></a></strong><code>()</code><br> Not-a-number, a special non-finite real value returned to signal an error</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-21afff517501f022247efe89c81a255a74533422" class="index-entry"></a> <!-- real; positive_infinity; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#positive_infinity" class="unlink" title="Jump to index entry"><code>positive_infinity</code></a></strong><code>()</code><br> Positive infinity, a special non-finite real value larger than all finite numbers</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4fcafd8237c585340253854dec13739165d456bd" class="index-entry"></a> <!-- real; negative_infinity; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#negative_infinity" class="unlink" title="Jump to index entry"><code>negative_infinity</code></a></strong><code>()</code><br> Negative infinity, a special non-finite real value smaller than all finite numbers</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b5f5a8656ff1f34e85d53f1ed43ff61eef17279b" class="index-entry"></a> <!-- real; machine_precision; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#machine_precision" class="unlink" title="Jump to index entry"><code>machine_precision</code></a></strong><code>()</code><br> The smallest number <span class="math inline">\(x\)</span> such that <span class="math inline">\((x + 1) \neq 1\)</span> in floating-point arithmetic on the current hardware platform</p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="get-log-prob" class="level2">
+<h2 class="anchored" data-anchor-id="get-log-prob">Log probability function</h2>
+<p>The basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (<code>~</code>) and log probability increment statement (<code>increment_log_prob</code>) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.</p>
+<p>Stan provides a special built-in function <code>target()</code> that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.</p>
+<a id="index-entry-138fdd54bc9b599fe8f41f4558f1be1aa3ace172" class="index-entry"></a> <!-- real; target; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#target" class="unlink" title="Jump to index entry"><code>target</code></a></strong><code>()</code><br> Return the current value of the log probability accumulator.</p>
+<small><i>Available since 2.10</i></small>
+<p><code>target</code> acts like a function ending in <code>_lp</code>, meaning that it may only may only be used in the model block.</p>
+</section>
+<section id="logical-functions" class="level2">
+<h2 class="anchored" data-anchor-id="logical-functions">Logical functions</h2>
+<p>Like C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.</p>
+<section id="comparison-operators" class="level3">
+<h3 class="anchored" data-anchor-id="comparison-operators">Comparison operators</h3>
+<p>All comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.</p>
+<a id="index-entry-eb4fbef11b05d04b29b1d9b813f5e07945bf9ad8" class="index-entry"></a> <!-- int; operator<; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator<" class="unlink" title="Jump to index entry"><code>operator&lt;</code></a></strong><code>(int x, int y)</code><br></p>
+<a id="index-entry-45d9721208dde1bade759d23819a2a80bbde7bb0" class="index-entry"></a> <!-- int; operator<; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator<" class="unlink" title="Jump to index entry"><code>operator&lt;</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is less than y and 0 otherwise. <span class="math display">\[\begin{equation*} \text{operator&lt;}(x,y)
+= \begin{cases} 1 &amp; \text{if $x &lt; y$} \\ 0 &amp; \text{otherwise}
+\end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-ae3e3c382d593269517c04ea38e93c34b1938ed3" class="index-entry"></a> <!-- int; operator<=; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator<=" class="unlink" title="Jump to index entry"><code>operator&lt;=</code></a></strong><code>(int x, int y)</code><br></p>
+<a id="index-entry-b2b6ee942a7836d609c24691ae5327be085eabf8" class="index-entry"></a> <!-- int; operator<=; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator<=" class="unlink" title="Jump to index entry"><code>operator&lt;=</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is less than or equal y and 0 otherwise. <span class="math display">\[\begin{equation*}
+\text{operator&lt;=}(x,y) = \begin{cases} 1 &amp; \text{if $x \leq y$} \\ 0 &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7897823400bd03a18304e92dea188037789e2fc0" class="index-entry"></a> <!-- int; operator>; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator>" class="unlink" title="Jump to index entry"><code>operator&gt;</code></a></strong><code>(int x, int y)</code><br></p>
+<a id="index-entry-47bdb3c70b9ec7cdf0f976442ca728e5cd85ea39" class="index-entry"></a> <!-- int; operator>; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator>" class="unlink" title="Jump to index entry"><code>operator&gt;</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is greater than y and 0 otherwise. <span class="math display">\[\begin{equation*}
+\text{operator&gt;}(x,y) = \begin{cases} 1 &amp; \text{if $x &gt; y$} \\ 0 &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9d2656ef1df47087d225d2cfd011938c49677d09" class="index-entry"></a> <!-- int; operator>=; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator>=" class="unlink" title="Jump to index entry"><code>operator&gt;=</code></a></strong><code>(int x, int y)</code><br></p>
+<a id="index-entry-29ba70695594992fa689387dd1f623e794aa8ef2" class="index-entry"></a> <!-- int; operator>=; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator>=" class="unlink" title="Jump to index entry"><code>operator&gt;=</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is greater than or equal to y and 0 otherwise. <span class="math display">\[\begin{equation*}
+\text{operator&gt;=}(x,y) = \begin{cases} 1 &amp; \text{if $x \geq y$} \\ 0 &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2ef075eb2521760dca18f52bea00eda6609a70d7" class="index-entry"></a> <!-- int; operator==; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator==" class="unlink" title="Jump to index entry"><code>operator==</code></a></strong><code>(int x, int y)</code><br></p>
+<a id="index-entry-876a6664927a13415aef76d12a52894d6047d958" class="index-entry"></a> <!-- int; operator==; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator==" class="unlink" title="Jump to index entry"><code>operator==</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is equal to y and 0 otherwise. <span class="math display">\[\begin{equation*}
+\text{operator==}(x,y) = \begin{cases} 1 &amp; \text{if $x = y$} \\ 0 &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-34c680d818a4a8758b819f7b8a3b11f1d1f094b4" class="index-entry"></a> <!-- int; operator!=; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator!=" class="unlink" title="Jump to index entry"><code>operator!=</code></a></strong><code>(int x, int y)</code><br></p>
+<a id="index-entry-3cc230ef52c7125f7fd034984257fb3b2e13bbcc" class="index-entry"></a> <!-- int; operator!=; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator!=" class="unlink" title="Jump to index entry"><code>operator!=</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is not equal to y and 0 otherwise. <span class="math display">\[\begin{equation*}
+\text{operator!=}(x,y) = \begin{cases} 1 &amp; \text{if $x \neq y$} \\ 0 &amp;
+\text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+</section>
+<section id="boolean-operators" class="level3">
+<h3 class="anchored" data-anchor-id="boolean-operators">Boolean operators</h3>
+<p>Boolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as <code>!a &amp;&amp; b</code> is interpreted as <code>(!a) &amp;&amp; b</code>, and <code>a &lt; b || c &gt;= d &amp;&amp; e != f</code> as <code>(a &lt; b) || (((c &gt;= d) &amp;&amp; (e != f)))</code>.</p>
+<a id="index-entry-079682d0e7e0678530187f5fc5e2b993f1b5b1cc" class="index-entry"></a> <!-- int; operator!; (int x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator!" class="unlink" title="Jump to index entry"><code>operator!</code></a></strong><code>(int x)</code><br> Return 1 if x is zero and 0 otherwise. <span class="math display">\[\begin{equation*} \text{operator!}(x) =
+\begin{cases} 0 &amp; \text{if $x \neq 0$} \\ 1 &amp; \text{if $x = 0$}
+\end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-211d31dd129d2ce61f68b4264b9318aa39a096d7" class="index-entry"></a> <!-- int; operator!; (real x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator!" class="unlink" title="Jump to index entry"><code>operator!</code></a></strong><code>(real x)</code><br> Return 1 if x is zero and 0 otherwise. <span class="math display">\[\begin{equation*} \text{operator!}(x) =
+\begin{cases} 0 &amp; \text{if $x \neq 0.0$} \\ 1 &amp; \text{if $x = 0.0$}
+\end{cases} \end{equation*}\]</span> <strong>deprecated;</strong> - use <code>operator==</code> instead.</p>
+<small><i>Available since 2.0, deprecated in 2.31</i></small>
+<a id="index-entry-705fd1857d26667bd15e9cbd54c1a605ce286b17" class="index-entry"></a> <!-- int; operator&&; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator&amp;&amp;" class="unlink" title="Jump to index entry"><code>operator&amp;&amp;</code></a></strong><code>(int x, int y)</code><br></p>
+<p>Return 1 if x is unequal to 0 and y is unequal to 0. <span class="math display">\[\begin{equation*}
+\mathrm{operator\&amp;\&amp;}(x,y) = \begin{cases} 1 &amp; \text{if $x \neq 0$}
+\text{ and } y \neq 0\\ 0 &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7fab4d7139fe3c56c90269284ac8fb2a27aa9c3b" class="index-entry"></a> <!-- int; operator&&; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator&amp;&amp;" class="unlink" title="Jump to index entry"><code>operator&amp;&amp;</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is unequal to 0.0 and y is unequal to 0.0. <span class="math display">\[\begin{equation*}
+\mathrm{operator\&amp;\&amp;}(x,y) = \begin{cases} 1 &amp; \text{if $x \neq 0.0$}
+\text{ and } y \neq 0.0\\ 0 &amp; \text{otherwise} \end{cases} \end{equation*}\]</span> <strong>deprecated</strong></p>
+<small><i>Available since 2.0, deprecated in 2.31</i></small>
+<a id="index-entry-8fbd92b4b30a50912cd8d6602927ee93cf00c392" class="index-entry"></a> <!-- int; operator||; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator||" class="unlink" title="Jump to index entry"><code>operator||</code></a></strong><code>(int x, int y)</code><br> Return 1 if x is unequal to 0 or y is unequal to 0. <span class="math display">\[\begin{equation*}
+\text{operator||}(x,y) = \begin{cases} 1 &amp; \text{if $x \neq 0$}
+\textrm{ or } y \neq 0\\ 0 &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-d1b0b32e522a6d270bdccaa3b56f1dfaef2dfa1f" class="index-entry"></a> <!-- int; operator||; (real x, real y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#operator||" class="unlink" title="Jump to index entry"><code>operator||</code></a></strong><code>(real x, real y)</code><br> Return 1 if x is unequal to 0.0 or y is unequal to 0.0. <span class="math display">\[\begin{equation*}
+\text{operator||}(x,y) = \begin{cases} 1 &amp; \text{if $x \neq 0.0$}
+\textrm{ or } y \neq 0.0\\ 0 &amp; \text{otherwise} \end{cases} \end{equation*}\]</span> <strong>deprecated</strong></p>
+<small><i>Available since 2.0, deprecated in 2.31</i></small>
+<section id="boolean-operator-short-circuiting" class="level4">
+<h4 class="anchored" data-anchor-id="boolean-operator-short-circuiting">Boolean operator short circuiting</h4>
+<p>Like in C++, the boolean operators <code>&amp;&amp;</code> and <code>||</code> and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating <code>a || b</code>, if <code>a</code> evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression <code>b</code>. Similarly, evaluating <code>a &amp;&amp; b</code> first evaluates <code>a</code>, and if the result is zero, returns 0 without evaluating <code>b</code>.</p>
+</section>
+</section>
+<section id="logical-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="logical-functions-1">Logical functions</h3>
+<p>The logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.</p>
+<a id="index-entry-b5b8e03f66b31507dac3535af6f2d8f615e4a3af" class="index-entry"></a> <!-- real; step; (real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#step" class="unlink" title="Jump to index entry"><code>step</code></a></strong><code>(real x)</code><br> Return 1 if x is positive and 0 otherwise. <span class="math display">\[\begin{equation*} \text{step}(x) =
+\begin{cases} 0 &amp; \text{if } x &lt; 0 \\ 1 &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span> <em><strong>Warning:</strong></em> <code>int_step(0)</code> and <code>int_step(NaN)</code> return 0 whereas <code>step(0)</code> and <code>step(NaN)</code> return 1.</p>
+<p>The step function is often used in BUGS to perform conditional operations. For instance, <code>step(a-b)</code> evaluates to 1 if <code>a</code> is greater than <code>b</code> and evaluates to 0 otherwise. <code>step</code> is a step-like functions; see the warning in section <a href="#step-functions">step functions</a> applied to expressions dependent on parameters.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-7240fd89c6199124fb25731ffcdb1459029b04df" class="index-entry"></a> <!-- int; is_inf; (real x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#is_inf" class="unlink" title="Jump to index entry"><code>is_inf</code></a></strong><code>(real x)</code><br> Return 1 if x is infinite (positive or negative) and 0 otherwise.</p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-d9e6d18b33b36e488e49f87c4efd470c39e6a5b1" class="index-entry"></a> <!-- int; is_nan; (real x); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#is_nan" class="unlink" title="Jump to index entry"><code>is_nan</code></a></strong><code>(real x)</code><br> Return 1 if x is NaN and 0 otherwise.</p>
+<small><i>Available since 2.5</i></small>
+<p>Care must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section <a href="#step-functions">step-like functions</a> for details.</p>
+</section>
+</section>
+<section id="real-valued-arithmetic-operators" class="level2">
+<h2 class="anchored" data-anchor-id="real-valued-arithmetic-operators">Real-valued arithmetic operators</h2>
+<p>The arithmetic operators are presented using C++ notation. For instance <code>operator+(x,y)</code> refers to the binary addition operator and <code>operator-(x)</code> to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as <code>x + y</code> and <code>-x</code>, respectively.</p>
+<section id="binary-infix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="binary-infix-operators">Binary infix operators</h3>
+<a id="index-entry-4b2af0abc1a2c8338794b49b801ff4d63be2f574" class="index-entry"></a> <!-- real; operator+; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(real x, real y)</code><br> Return the sum of x and y. <span class="math display">\[\begin{equation*} (x + y) = \text{operator+}(x,y) = x+y \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-739fd04c545b6d65855bc20e05617df813db0598" class="index-entry"></a> <!-- real; operator-; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(real x, real y)</code><br> Return the difference between x and y. <span class="math display">\[\begin{equation*} (x - y) =
+\text{operator-}(x,y) = x - y \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-08f55cee893fdef254457be66b7a82c1d942ad0e" class="index-entry"></a> <!-- real; operator*; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator\*" class="unlink" title="Jump to index entry"><code>operator*</code></a></strong><code>(real x, real y)</code><br> Return the product of x and y. <span class="math display">\[\begin{equation*} (x * y) = \text{operator*}(x,y) = xy
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b28e30b8930bb68726d2de15d2857875cb73fcfb" class="index-entry"></a> <!-- real; operator/; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator/" class="unlink" title="Jump to index entry"><code>operator/</code></a></strong><code>(real x, real y)</code><br> Return the quotient of x and y. <span class="math display">\[\begin{equation*} (x / y) = \text{operator/}(x,y) =
+\frac{x}{y} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-329b197f72b50051c6de252c7dd804a96a399051" class="index-entry"></a> <!-- real; operator^; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator^" class="unlink" title="Jump to index entry"><code>operator^</code></a></strong><code>(real x, real y)</code><br> Return x raised to the power of y. <span class="math display">\[\begin{equation*} (x^\mathrm{\wedge}y) =
+\text{operator}^\mathrm{\wedge}(x,y) = x^y \end{equation*}\]</span></p>
+<small><i>Available since 2.5</i></small>
+</section>
+<section id="unary-prefix-operators" class="level3">
+<h3 class="anchored" data-anchor-id="unary-prefix-operators">Unary prefix operators</h3>
+<a id="index-entry-2055033f59d0333f7b514dd496153b272c1b274c" class="index-entry"></a> <!-- real; operator-; (real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(real x)</code><br> Return the negation of the subtrahend x. <span class="math display">\[\begin{equation*} \text{operator-}(x) = (-x)
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-12e466dce5df3d58d3ad7f47d7637bc66c0c3b0f" class="index-entry"></a> <!-- T; operator-; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#operator-" class="unlink" title="Jump to index entry"><code>operator-</code></a></strong><code>(T x)</code><br> Vectorized version of <code>operator-</code>. If <code>T x</code> is a (possibly nested) array of reals, <code>-x</code> is the same shape array where each individual number is negated.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-54245a5adfc60f65ced33cbdf4571aeb7536afc3" class="index-entry"></a> <!-- real; operator+; (real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#operator+" class="unlink" title="Jump to index entry"><code>operator+</code></a></strong><code>(real x)</code><br> Return the value of x. <span class="math display">\[\begin{equation*} \text{operator+}(x) = x \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+</section>
+</section>
+<section id="step-functions" class="level2">
+<h2 class="anchored" data-anchor-id="step-functions">Step-like functions</h2>
+<p><em><strong>Warning:</strong></em> <em>These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.</em></p>
+<section id="absolute-value-functions" class="level3">
+<h3 class="anchored" data-anchor-id="absolute-value-functions">Absolute value functions</h3>
+<a id="index-entry-c197d7c04ea9bf23cf252f2092111b3c5aadd02d" class="index-entry"></a> <!-- T; abs; (T x); -->
+<p></p>
+<p><code>T</code> <strong><a href="../functions-reference/functions_index.html#abs" class="unlink" title="Jump to index entry"><code>abs</code></a></strong><code>(T x)</code><br> The absolute value of x.</p>
+<p>This function works elementwise over containers such as vectors. Given a type <code>T</code> which is <code>real</code> <code>vector</code>, <code>row_vector</code>, <code>matrix</code>, or an array of those types, <code>abs</code> returns the same type where each element has had its absolute value taken.</p>
+<small><i>Available since 2.0, vectorized in 2.30</i></small>
+<a id="index-entry-8e4c91cd9725a9e73f23c330bf345dbc813f4a44" class="index-entry"></a> <!-- real; fdim; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#fdim" class="unlink" title="Jump to index entry"><code>fdim</code></a></strong><code>(real x, real y)</code><br> Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. <span class="math display">\[\begin{equation*} \text{fdim}(x,y) = \begin{cases} x-y &amp;
+\text{if } x \geq y \\ 0 &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-e5223bd7c30433273feb0a16c3c3950552c1ebce" class="index-entry"></a> <!-- R; fdim; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#fdim" class="unlink" title="Jump to index entry"><code>fdim</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>fdim</code> function</p>
+<small><i>Available since 2.25</i></small>
+</section>
+<section id="bounds-functions" class="level3">
+<h3 class="anchored" data-anchor-id="bounds-functions">Bounds functions</h3>
+<a id="index-entry-1c3fab0d71dfe4827399431840361247caf51f6c" class="index-entry"></a> <!-- real; fmin; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#fmin" class="unlink" title="Jump to index entry"><code>fmin</code></a></strong><code>(real x, real y)</code><br> Return the minimum of x and y; see warning above. <span class="math display">\[\begin{equation*} \text{fmin}(x,y) = \begin{cases} x &amp;
+\text{if } x \leq y \\ y &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-745a8f98349953b6346521de02a0f66c0e837597" class="index-entry"></a> <!-- R; fmin; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#fmin" class="unlink" title="Jump to index entry"><code>fmin</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>fmin</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-84f91b685bc8ac919063009145d0bda6df79898e" class="index-entry"></a> <!-- real; fmax; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#fmax" class="unlink" title="Jump to index entry"><code>fmax</code></a></strong><code>(real x, real y)</code><br> Return the maximum of x and y; see warning above. <span class="math display">\[\begin{equation*} \text{fmax}(x,y) = \begin{cases} x &amp;
+\text{if } x \geq y \\ y &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-9619ac6ed9fc6345c116fdaa93c6eb7f4fd18d6c" class="index-entry"></a> <!-- R; fmax; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#fmax" class="unlink" title="Jump to index entry"><code>fmax</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>fmax</code> function</p>
+<small><i>Available since 2.25</i></small>
+</section>
+<section id="arithmetic-functions" class="level3">
+<h3 class="anchored" data-anchor-id="arithmetic-functions">Arithmetic functions</h3>
+<a id="index-entry-fed64eefc7fd2673daac1482490fb6dfc5e37647" class="index-entry"></a> <!-- real; fmod; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#fmod" class="unlink" title="Jump to index entry"><code>fmod</code></a></strong><code>(real x, real y)</code><br> Return the real value remainder after dividing x by y; see warning above. <span class="math display">\[\begin{equation*} \text{fmod}(x,y) = x - \left\lfloor \frac{x}{y} \right\rfloor \, y \end{equation*}\]</span> The operator <span class="math inline">\(\lfloor u \rfloor\)</span> is the floor operation; see below.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-206cb5e9da51b6d1255b47d019273ab50a84687c" class="index-entry"></a> <!-- R; fmod; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#fmod" class="unlink" title="Jump to index entry"><code>fmod</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>fmod</code> function</p>
+<small><i>Available since 2.25</i></small>
+</section>
+<section id="rounding-functions" class="level3">
+<h3 class="anchored" data-anchor-id="rounding-functions">Rounding functions</h3>
+<p><em><strong>Warning:</strong></em> Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.</p>
+<p>The rounding functions cannot be used as indices to arrays because they return real values. For operations over <code>data</code> or in the <code>generated quantities</code> block, the <a href="../functions-reference/integer-valued_basic_functions.html#casting-functions"><code>to_int()</code> function</a> can be used.</p>
+<a id="index-entry-f4309f63f497c18607ceca16a255a6c8c3722153" class="index-entry"></a> <!-- R; floor; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#floor" class="unlink" title="Jump to index entry"><code>floor</code></a></strong><code>(T x)</code><br> The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section <a href="#step-functions">step-like functions</a></p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-3989753a963097381576ca0373b3190681b0bbe8" class="index-entry"></a> <!-- R; ceil; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#ceil" class="unlink" title="Jump to index entry"><code>ceil</code></a></strong><code>(T x)</code><br> The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section <a href="#step-functions">step-like functions</a></p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-b0ef648badd54cfdfec4e07e6a94362748815ccc" class="index-entry"></a> <!-- R; round; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#round" class="unlink" title="Jump to index entry"><code>round</code></a></strong><code>(T x)</code><br> The nearest integer to x, converted to a real value; see warning at start of section <a href="#step-functions">step-like functions</a></p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-4a8fc8e985608887eaa81d790ca2ee2e93e9c16e" class="index-entry"></a> <!-- R; trunc; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#trunc" class="unlink" title="Jump to index entry"><code>trunc</code></a></strong><code>(T x)</code><br> The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section <a href="#step-functions">step-like functions</a></p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+</section>
+</section>
+<section id="power-and-logarithm-functions" class="level2">
+<h2 class="anchored" data-anchor-id="power-and-logarithm-functions">Power and logarithm functions</h2>
+<a id="index-entry-b3bbd4cc788019bca7f70823777f48ce5f43e4b4" class="index-entry"></a> <!-- R; sqrt; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#sqrt" class="unlink" title="Jump to index entry"><code>sqrt</code></a></strong><code>(T x)</code><br> The square root of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-e5c3ef31b3dfc2b73c3572239e82a2ae8629df79" class="index-entry"></a> <!-- R; cbrt; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#cbrt" class="unlink" title="Jump to index entry"><code>cbrt</code></a></strong><code>(T x)</code><br> The cube root of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-b9f6f7932609a3f72301693abb282aa60845a210" class="index-entry"></a> <!-- R; square; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#square" class="unlink" title="Jump to index entry"><code>square</code></a></strong><code>(T x)</code><br> The square of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-f32e064beb0f00e975210d246d3f80431ea9b85f" class="index-entry"></a> <!-- R; exp; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#exp" class="unlink" title="Jump to index entry"><code>exp</code></a></strong><code>(T x)</code><br> The natural exponential of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-7db9b662f6846f90e9edf891fac2a64701c213d2" class="index-entry"></a> <!-- R; exp2; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#exp2" class="unlink" title="Jump to index entry"><code>exp2</code></a></strong><code>(T x)</code><br> The base-2 exponential of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-656b9ebaf2d3c913d48b33e37256a7fd908c2c74" class="index-entry"></a> <!-- R; log; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log" class="unlink" title="Jump to index entry"><code>log</code></a></strong><code>(T x)</code><br> The natural logarithm of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-8aa277afaded318050922a7e576d5531918c4fa3" class="index-entry"></a> <!-- R; log2; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log2" class="unlink" title="Jump to index entry"><code>log2</code></a></strong><code>(T x)</code><br> The base-2 logarithm of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-ad4651b8f653b4d8e1e6e50131371c8be7a35444" class="index-entry"></a> <!-- R; log10; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log10" class="unlink" title="Jump to index entry"><code>log10</code></a></strong><code>(T x)</code><br> The base-10 logarithm of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-0882f3a0b5f515684dd2a29821a274a09ada9426" class="index-entry"></a> <!-- real; pow; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#pow" class="unlink" title="Jump to index entry"><code>pow</code></a></strong><code>(real x, real y)</code><br> Return x raised to the power of y. <span class="math display">\[\begin{equation*} \text{pow}(x,y) = x^y \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c9ec850adf117cff8e4f8da2a8287b92a152b481" class="index-entry"></a> <!-- R; pow; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#pow" class="unlink" title="Jump to index entry"><code>pow</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>pow</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-7e3cb8bb5e02b0568495e3495ab81c8ab671dd96" class="index-entry"></a> <!-- R; inv; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv" class="unlink" title="Jump to index entry"><code>inv</code></a></strong><code>(T x)</code><br> The inverse of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-4dcb1d28ce2c1de0b6668a4095511ceeadb29fd0" class="index-entry"></a> <!-- R; inv_sqrt; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_sqrt" class="unlink" title="Jump to index entry"><code>inv_sqrt</code></a></strong><code>(T x)</code><br> The inverse of the square root of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-b4f13710242aa2542c07afc496e1966099678f35" class="index-entry"></a> <!-- R; inv_square; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_square" class="unlink" title="Jump to index entry"><code>inv_square</code></a></strong><code>(T x)</code><br> The inverse of the square of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+</section>
+<section id="trigonometric-functions" class="level2">
+<h2 class="anchored" data-anchor-id="trigonometric-functions">Trigonometric functions</h2>
+<a id="index-entry-91e267ebdf31823d577cac8796d2fc05ffc4123b" class="index-entry"></a> <!-- real; hypot; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypot" class="unlink" title="Jump to index entry"><code>hypot</code></a></strong><code>(real x, real y)</code><br> Return the length of the hypotenuse of a right triangle with sides of length x and y. <span class="math display">\[\begin{equation*} \text{hypot}(x,y) = \begin{cases} \sqrt{x^2+y^2} &amp;
+\text{if } x,y\geq 0 \\ \textrm{NaN} &amp; \text{otherwise} \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-d55f4ca2bc092698d4e852dae4516f76ef5e8738" class="index-entry"></a> <!-- R; hypot; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#hypot" class="unlink" title="Jump to index entry"><code>hypot</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>hypot</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-c587dc7908ecfe46cdd09800844ad9ea90e8463b" class="index-entry"></a> <!-- R; cos; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#cos" class="unlink" title="Jump to index entry"><code>cos</code></a></strong><code>(T x)</code><br> The cosine of the angle x (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-318c6c59e1d99978abb1432132810c6f097ed01f" class="index-entry"></a> <!-- R; sin; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#sin" class="unlink" title="Jump to index entry"><code>sin</code></a></strong><code>(T x)</code><br> The sine of the angle x (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-02f62dc1271129383c9e257d1a6dea18851a7411" class="index-entry"></a> <!-- R; tan; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#tan" class="unlink" title="Jump to index entry"><code>tan</code></a></strong><code>(T x)</code><br> The tangent of the angle x (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-c928df245747cbd0d0fdb91212e73e8c3d5ace77" class="index-entry"></a> <!-- R; acos; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#acos" class="unlink" title="Jump to index entry"><code>acos</code></a></strong><code>(T x)</code><br> The principal arc (inverse) cosine (in radians) of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-f2bf9ebf3337e4f590b7f30ebe4aaeb04d2ca9a3" class="index-entry"></a> <!-- R; asin; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#asin" class="unlink" title="Jump to index entry"><code>asin</code></a></strong><code>(T x)</code><br> The principal arc (inverse) sine (in radians) of x</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4ff6fd29f63ba53b4b854f16fb4122d25f817257" class="index-entry"></a> <!-- R; atan; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#atan" class="unlink" title="Jump to index entry"><code>atan</code></a></strong><code>(T x)</code><br> The principal arc (inverse) tangent (in radians) of x, with values from <span class="math inline">\(-\pi/2\)</span> to <span class="math inline">\(\pi/2\)</span></p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-bf8aeecad9e7aec2135da1c0e68b6c6e380eb65c" class="index-entry"></a> <!-- R; atan2; (T y, T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#atan2" class="unlink" title="Jump to index entry"><code>atan2</code></a></strong><code>(T y, T x)</code><br> Return the principal arc (inverse) tangent (in radians) of y divided by x, <span class="math display">\[\begin{equation*} \text{atan2}(y, x) = \arctan\left(\frac{y}{x}\right) \end{equation*}\]</span></p>
+<small><i>Available since 2.0, vectorized in 2.34</i></small>
+</section>
+<section id="hyperbolic-trigonometric-functions" class="level2">
+<h2 class="anchored" data-anchor-id="hyperbolic-trigonometric-functions">Hyperbolic trigonometric functions</h2>
+<a id="index-entry-1afe4d43c0f158ff3446b0459e002ad8343ee24e" class="index-entry"></a> <!-- R; cosh; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#cosh" class="unlink" title="Jump to index entry"><code>cosh</code></a></strong><code>(T x)</code><br> The hyperbolic cosine of x (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-1532402d69ee65eaf786341825da10fa8941a092" class="index-entry"></a> <!-- R; sinh; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#sinh" class="unlink" title="Jump to index entry"><code>sinh</code></a></strong><code>(T x)</code><br> The hyperbolic sine of x (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-4f8f8565db3a7961f780e532e82cd5db1659bdc9" class="index-entry"></a> <!-- R; tanh; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#tanh" class="unlink" title="Jump to index entry"><code>tanh</code></a></strong><code>(T x)</code><br> The hyperbolic tangent of x (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-228328eeafc82944f3bdf9162ddbad97efddff0c" class="index-entry"></a> <!-- R; acosh; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#acosh" class="unlink" title="Jump to index entry"><code>acosh</code></a></strong><code>(T x)</code><br> The inverse hyperbolic cosine (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-a267074b1b9acb770b679aeaca1e79f326fecd09" class="index-entry"></a> <!-- R; asinh; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#asinh" class="unlink" title="Jump to index entry"><code>asinh</code></a></strong><code>(T x)</code><br> The inverse hyperbolic cosine (in radians)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-1a2910acee55dabfad98224fee341fccc8af7cb0" class="index-entry"></a> <!-- R; atanh; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#atanh" class="unlink" title="Jump to index entry"><code>atanh</code></a></strong><code>(T x)</code><br> The inverse hyperbolic tangent (in radians) of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+</section>
+<section id="link-functions" class="level2">
+<h2 class="anchored" data-anchor-id="link-functions">Link functions</h2>
+<p>The following functions are commonly used as link functions in generalized linear models. The function <span class="math inline">\(\Phi\)</span> is also commonly used as a link function (see section <a href="#Phi-function">probability-related functions</a>).</p>
+<a id="index-entry-b473710a74faf02bff89cd1f779befb3517d0ba6" class="index-entry"></a> <!-- R; logit; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#logit" class="unlink" title="Jump to index entry"><code>logit</code></a></strong><code>(T x)</code><br> The log odds, or logit, function applied to x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-587aa4ba61e421c4ca21bad5c94b5b919e16d34d" class="index-entry"></a> <!-- R; inv_logit; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_logit" class="unlink" title="Jump to index entry"><code>inv_logit</code></a></strong><code>(T x)</code><br> The logistic sigmoid function applied to x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-40b628501a54a1c1fc2eed89973daabf330a9c2c" class="index-entry"></a> <!-- R; inv_cloglog; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_cloglog" class="unlink" title="Jump to index entry"><code>inv_cloglog</code></a></strong><code>(T x)</code><br> The inverse of the complementary log-log function applied to x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+</section>
+<section id="Phi-function" class="level2">
+<h2 class="anchored" data-anchor-id="Phi-function">Probability-related functions</h2>
+<section id="normal-cumulative-distribution-functions" class="level3">
+<h3 class="anchored" data-anchor-id="normal-cumulative-distribution-functions">Normal cumulative distribution functions</h3>
+<p>The error function <code>erf</code> is related to the standard normal cumulative distribution function <span class="math inline">\(\Phi\)</span> by scaling. See section <a href="../functions-reference/unbounded_continuous_distributions.html#normal-distribution">normal distribution</a> for the general normal cumulative distribution function (and its complement).</p>
+<a id="index-entry-3ed55c0864c7839ec52fa38be9974f0e43bd5a50" class="index-entry"></a> <!-- R; erf; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#erf" class="unlink" title="Jump to index entry"><code>erf</code></a></strong><code>(T x)</code><br> The error function, also known as the Gauss error function, of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-7dc68f18ec6e0c591f011d5ff8f4075637b9dd5c" class="index-entry"></a> <!-- R; erfc; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#erfc" class="unlink" title="Jump to index entry"><code>erfc</code></a></strong><code>(T x)</code><br> The complementary error function of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-436991c650707e4567f4627cf8c7216e20bd1d68" class="index-entry"></a> <!-- R; inv_erfc; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_erfc" class="unlink" title="Jump to index entry"><code>inv_erfc</code></a></strong><code>(T x)</code><br> The inverse of the complementary error function of x</p>
+<small><i>Available since 2.29, vectorized in 2.29</i></small>
+<a id="index-entry-2a50a935e0b3ed75c3a19e2bee25f3a01c7799f7" class="index-entry"></a> <!-- R; Phi; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#Phi" class="unlink" title="Jump to index entry"><code>Phi</code></a></strong><code>(T x)</code><br> The standard normal cumulative distribution function of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-f0054cc2edcc0949c41276325f6db59b8e0c4e15" class="index-entry"></a> <!-- R; inv_Phi; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#inv_Phi" class="unlink" title="Jump to index entry"><code>inv_Phi</code></a></strong><code>(T x)</code><br> Return the value of the inverse standard normal cdf <span class="math inline">\(\Phi^{-1}\)</span> at the specified quantile <code>x</code>. The details of the algorithm can be found in <span class="citation" data-cites="Wichura:1988">(<a href="#ref-Wichura:1988" role="doc-biblioref">Wichura 1988</a>)</span>. Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-1b36b9b4f834de053d1a484782736f41af7b107e" class="index-entry"></a> <!-- R; Phi_approx; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#Phi_approx" class="unlink" title="Jump to index entry"><code>Phi_approx</code></a></strong><code>(T x)</code><br> The fast approximation of the unit (may replace <code>Phi</code> for probit regression with maximum absolute error of 0.00014, see <span class="citation" data-cites="BowlingEtAl:2009">(<a href="#ref-BowlingEtAl:2009" role="doc-biblioref">Bowling et al. 2009</a>)</span> for details)</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+</section>
+<section id="other-probability-related-functions" class="level3">
+<h3 class="anchored" data-anchor-id="other-probability-related-functions">Other probability-related functions</h3>
+<a id="index-entry-44fff3495ce5cd36ebde4587e802cb7bdc795794" class="index-entry"></a> <!-- real; binary_log_loss; (int y, real y_hat); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#binary_log_loss" class="unlink" title="Jump to index entry"><code>binary_log_loss</code></a></strong><code>(int y, real y_hat)</code><br> Return the log loss function for for predicting <span class="math inline">\(\hat{y} \in [0,1]\)</span> for boolean outcome <span class="math inline">\(y \in \{0,1\}\)</span>. <span class="math display">\[\begin{equation*}
+\mathrm{binary\_log\_loss}(y,\hat{y}) = \begin{cases} -\log \hat{y} &amp;
+\text{if } y = 1\\ -\log (1 - \hat{y}) &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-05d751372242ac11c1c7356d09fb0341276b897c" class="index-entry"></a> <!-- R; binary_log_loss; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#binary_log_loss" class="unlink" title="Jump to index entry"><code>binary_log_loss</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>binary_log_loss</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-1ebaab3f9c8786de6e45b245eb0557156a1b2a77" class="index-entry"></a> <!-- real; owens_t; (real h, real a); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#owens_t" class="unlink" title="Jump to index entry"><code>owens_t</code></a></strong><code>(real h, real a)</code><br> Return the Owen’s T function for the probability of the event <span class="math inline">\(X &gt; h\)</span> and <span class="math inline">\(0&lt;Y&lt;aX\)</span> where X and Y are independent standard normal random variables. <span class="math display">\[\begin{equation*} \mathrm{owens\_t}(h,a) = \frac{1}{2\pi} \int_0^a
+\frac{\exp(-\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \end{equation*}\]</span></p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8ac2b39d27b678ab373ba570f18b633093c6511d" class="index-entry"></a> <!-- R; owens_t; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#owens_t" class="unlink" title="Jump to index entry"><code>owens_t</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>owens_t</code> function</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="betafun" class="level2">
+<h2 class="anchored" data-anchor-id="betafun">Combinatorial functions</h2>
+<a id="index-entry-bbec535fbcce7c57b2dd2cc189383f0c180038ed" class="index-entry"></a> <!-- real; beta; (real alpha, real beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta" class="unlink" title="Jump to index entry"><code>beta</code></a></strong><code>(real alpha, real beta)</code><br> Return the beta function applied to alpha and beta. The beta function, <span class="math inline">\(\text{B}(\alpha,\beta)\)</span>, computes the normalizing constant for the beta distribution, and is defined for <span class="math inline">\(\alpha &gt; 0\)</span> and <span class="math inline">\(\beta &gt; 0\)</span>. See section <a href="../functions-reference/mathematical_functions.html#beta-appendix">appendix</a> for definition of <span class="math inline">\(\text{B}(\alpha, \beta)\)</span>.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-faea3c6dc569671c76ab57211c00ebac2394742c" class="index-entry"></a> <!-- R; beta; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#beta" class="unlink" title="Jump to index entry"><code>beta</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>beta</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8af921f70494e8dac40b7e567b55041a92982782" class="index-entry"></a> <!-- real; inc_beta; (real alpha, real beta, real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inc_beta" class="unlink" title="Jump to index entry"><code>inc_beta</code></a></strong><code>(real alpha, real beta, real x)</code><br> Return the regularized incomplete beta function up to x applied to alpha and beta. See section <a href="../functions-reference/mathematical_functions.html#inc-beta-appendix">appendix</a> for a definition.</p>
+<small><i>Available since 2.10</i></small>
+<a id="index-entry-75728a2fe483e42a84c2fb5562469cb0fcd56250" class="index-entry"></a> <!-- real; inv_inc_beta; (real alpha, real beta, real p); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#inv_inc_beta" class="unlink" title="Jump to index entry"><code>inv_inc_beta</code></a></strong><code>(real alpha, real beta, real p)</code><br> Return the inverse of the regularized incomplete beta function. The return value <code>x</code> is the value that solves <code>p = inc_beta(alpha, beta, x)</code>. See section <a href="../functions-reference/mathematical_functions.html#inc-beta-appendix">appendix</a> for a definition of the <code>inc_beta</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-def48992e0f8724381904fba78466b0f4a0d14bd" class="index-entry"></a> <!-- real; lbeta; (real alpha, real beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lbeta" class="unlink" title="Jump to index entry"><code>lbeta</code></a></strong><code>(real alpha, real beta)</code><br> Return the natural logarithm of the beta function applied to alpha and beta. The beta function, <span class="math inline">\(\text{B}(\alpha,\beta)\)</span>, computes the normalizing constant for the beta distribution, and is defined for <span class="math inline">\(\alpha &gt; 0\)</span> and <span class="math inline">\(\beta &gt; 0\)</span>. <span class="math display">\[\begin{equation*}
+\text{lbeta}(\alpha,\beta) = \log \Gamma(\alpha) + \log \Gamma(\beta) - \log \Gamma(\alpha+\beta)
+\end{equation*}\]</span> See section <a href="../functions-reference/mathematical_functions.html#beta-appendix">appendix</a> for definition of <span class="math inline">\(\text{B}(\alpha, \beta)\)</span>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-ffad8a6ffd89011bffb5c4e0eb7817d550b4d438" class="index-entry"></a> <!-- R; lbeta; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lbeta" class="unlink" title="Jump to index entry"><code>lbeta</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>lbeta</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-86c34f22971aeaafc52aaea7501baa231b8ce5e4" class="index-entry"></a> <!-- R; tgamma; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#tgamma" class="unlink" title="Jump to index entry"><code>tgamma</code></a></strong><code>(T x)</code><br> The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that <span class="math inline">\(\Gamma(n+1) = n!\)</span>. See for a full definition of <span class="math inline">\(\Gamma(x)\)</span>. The function is defined for positive numbers and non-integral negative numbers,</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-51737f6b8d61ebaae578b09d29899a9f72cc8dff" class="index-entry"></a> <!-- R; lgamma; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lgamma" class="unlink" title="Jump to index entry"><code>lgamma</code></a></strong><code>(T x)</code><br> The natural logarithm of the gamma function applied to x,</p>
+<small><i>Available since 2.0, vectorized in 2.15</i></small>
+<a id="index-entry-2ce1d8a0cdd9878fe64366c88a2381d6e702ff7d" class="index-entry"></a> <!-- R; digamma; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#digamma" class="unlink" title="Jump to index entry"><code>digamma</code></a></strong><code>(T x)</code><br> The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-af71c258411af390fa86b5cee39043dc0f7aa2aa" class="index-entry"></a> <!-- R; trigamma; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#trigamma" class="unlink" title="Jump to index entry"><code>trigamma</code></a></strong><code>(T x)</code><br> The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-d6d5bb15c2402be65d4a671ea3ee4462f2f96da9" class="index-entry"></a> <!-- real; lmgamma; (int n, real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lmgamma" class="unlink" title="Jump to index entry"><code>lmgamma</code></a></strong><code>(int n, real x)</code><br> Return the natural logarithm of the multivariate gamma function <span class="math inline">\(\Gamma_n\)</span> with n dimensions applied to x. <span class="math display">\[\begin{equation*}
+\text{lmgamma}(n,x) =
+\begin{cases} \frac{n(n-1)}{4} \log \pi + \sum_{j=1}^n \log \Gamma\left(x + \frac{1 - j}{2}\right)
+&amp; \text{if } x\not\in \{\dots,-3,-2,-1,0\}\\ \textrm{error} &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-f8c2c4db4a7befdf0727ddb48d54d9de4cf442a1" class="index-entry"></a> <!-- R; lmgamma; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lmgamma" class="unlink" title="Jump to index entry"><code>lmgamma</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>lmgamma</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-6a3cb9f6d55852cff80a67883177778548b712d0" class="index-entry"></a> <!-- real; gamma_p; (real a, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_p" class="unlink" title="Jump to index entry"><code>gamma_p</code></a></strong><code>(real a, real z)</code><br> Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. <span class="math display">\[\begin{equation*}
+\mathrm{gamma\_p}(a,z) =
+\begin{cases} \frac{1}{\Gamma(a)}\int_0^zt^{a-1}e^{-t}dt &amp;
+\text{if } a &gt; 0, z \geq 0 \\ \textrm{error} &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-07d2b4d046ed3e94ab2394782dbfb848ce338126" class="index-entry"></a> <!-- R; gamma_p; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#gamma_p" class="unlink" title="Jump to index entry"><code>gamma_p</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>gamma_p</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-5cd1f2570dd2f81dc1a8e6f344ba953142a29865" class="index-entry"></a> <!-- real; gamma_q; (real a, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gamma_q" class="unlink" title="Jump to index entry"><code>gamma_q</code></a></strong><code>(real a, real z)</code><br> Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. <span class="math display">\[\begin{equation*}
+\mathrm{gamma\_q}(a,z) =
+\begin{cases} \frac{1}{\Gamma(a)}\int_z^\infty t^{a-1}e^{-t}dt &amp;
+\text{if } a &gt; 0, z \geq 0 \\[6pt] \textrm{error} &amp; \text{otherwise}
+\end{cases}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4949830b870b509235b2289a32a4e00f6e52e309" class="index-entry"></a> <!-- R; gamma_q; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#gamma_q" class="unlink" title="Jump to index entry"><code>gamma_q</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>gamma_q</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-fd4d9124ce89bf38a7ca1601e28f14947fcae3ba" class="index-entry"></a> <!-- int; choose; (int x, int y); -->
+<p></p>
+<p><code>int</code> <strong><a href="../functions-reference/functions_index.html#choose" class="unlink" title="Jump to index entry"><code>choose</code></a></strong><code>(int x, int y)</code><br> Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as <span class="math inline">\(\binom{x}{y}\)</span> and pronounced “x choose y.” In its the antilog of the <code>lchoose</code> function but returns an integer rather than a real number with no non-zero decimal places. For <span class="math inline">\(0 \leq y \leq x\)</span>, the binomial coefficient function can be defined via the factorial function <span class="math display">\[\begin{equation*}
+\text{choose}(x,y) = \frac{x!}{\left(y!\right)\left(x - y\right)!}.
+\end{equation*}\]</span></p>
+<small><i>Available since 2.14</i></small>
+<a id="index-entry-ed9d9ccf90b8976a4cc2b40babee7c60d4807087" class="index-entry"></a> <!-- R; choose; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#choose" class="unlink" title="Jump to index entry"><code>choose</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>choose</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-3bc21536b0609c1753010054298a02498c21c610" class="index-entry"></a> <!-- real; bessel_first_kind; (int v, real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bessel_first_kind" class="unlink" title="Jump to index entry"><code>bessel_first_kind</code></a></strong><code>(int v, real x)</code><br> Return the Bessel function of the first kind with order v applied to x. <span class="math display">\[\begin{equation*}
+\mathrm{bessel\_first\_kind}(v,x) = J_v(x),
+\end{equation*}\]</span> where <span class="math display">\[\begin{equation*}
+J_v(x)=\left(\frac{1}{2}x\right)^v \sum_{k=0}^\infty
+\frac{\left(-\frac{1}{4}x^2\right)^k}{k!\, \Gamma(v+k+1)}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-88cb374bd5bc83af305e30102a59f6e38cd04165" class="index-entry"></a> <!-- R; bessel_first_kind; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#bessel_first_kind" class="unlink" title="Jump to index entry"><code>bessel_first_kind</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>bessel_first_kind</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-a76f46f2143eed84250e51f4f9fa21abf7ff8085" class="index-entry"></a> <!-- real; bessel_second_kind; (int v, real x); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#bessel_second_kind" class="unlink" title="Jump to index entry"><code>bessel_second_kind</code></a></strong><code>(int v, real x)</code><br> Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For <span class="math inline">\(x,v &gt; 0\)</span>, <span class="math display">\[\begin{equation*}
+\mathrm{bessel\_second\_kind}(v,x) =
+\begin{cases} Y_v(x) &amp; \text{if } x &gt; 0 \\ \textrm{error} &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span> where <span class="math display">\[\begin{equation*}
+Y_v(x)=\frac{J_v(x)\cos(v\pi)-J_{-v}(x)}{\sin(v\pi)}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.5</i></small>
+<a id="index-entry-7d09ab04cf9e5ad369f0475cf89c0dcf0e6cf137" class="index-entry"></a> <!-- R; bessel_second_kind; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#bessel_second_kind" class="unlink" title="Jump to index entry"><code>bessel_second_kind</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>bessel_second_kind</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-1b49c405644022c418024b3fe2f764d081114dc5" class="index-entry"></a> <!-- real; modified_bessel_first_kind; (int v, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#modified_bessel_first_kind" class="unlink" title="Jump to index entry"><code>modified_bessel_first_kind</code></a></strong><code>(int v, real z)</code><br> Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. <span class="math display">\[\begin{equation*}
+\mathrm{modified\_bessel\_first\_kind}(v,z) = I_v(z)
+\end{equation*}\]</span> where <span class="math display">\[\begin{equation*}
+{I_v}(z) = \left(\frac{1}{2}z\right)^v\sum_{k=0}^\infty \frac{\left(\frac{1}{4}z^2\right)^k}{k!\Gamma(v+k+1)}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.1</i></small>
+<a id="index-entry-63276374ee38edd44205674f5f32b4dcf306f671" class="index-entry"></a> <!-- R; modified_bessel_first_kind; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#modified_bessel_first_kind" class="unlink" title="Jump to index entry"><code>modified_bessel_first_kind</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>modified_bessel_first_kind</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-832bf9903539f39025b4823c5ad6ff48bf17ccad" class="index-entry"></a> <!-- real; log_modified_bessel_first_kind; (real v, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_modified_bessel_first_kind" class="unlink" title="Jump to index entry"><code>log_modified_bessel_first_kind</code></a></strong><code>(real v, real z)</code><br> Return the log of the modified Bessel function of the first kind. v does not have to be an integer.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-f338c7d47a8fdaa9b97a44fadfdd5fb71ceccb3b" class="index-entry"></a> <!-- R; log_modified_bessel_first_kind; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_modified_bessel_first_kind" class="unlink" title="Jump to index entry"><code>log_modified_bessel_first_kind</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>log_modified_bessel_first_kind</code> function</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-e174dc75cd0e499de88270d4616de1ca6cb81235" class="index-entry"></a> <!-- real; modified_bessel_second_kind; (int v, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#modified_bessel_second_kind" class="unlink" title="Jump to index entry"><code>modified_bessel_second_kind</code></a></strong><code>(int v, real z)</code><br> Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. <span class="math display">\[\begin{equation*}
+\mathrm{modified\_bessel\_second\_kind}(v,z) =
+\begin{cases} K_v(z) &amp;  \text{if } z &gt; 0 \\ \textrm{error} &amp; \text{if } z \leq 0 \end{cases}
+\end{equation*}\]</span> where <span class="math display">\[\begin{equation*} {K_v}(z) = \frac{\pi}{2}\cdot\frac{I_{-v}(z) - I_{v}(z)}{\sin(v\pi)}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.1</i></small>
+<a id="index-entry-6340b47f5e4a5642b6c080962ca50fde4d6ea5d9" class="index-entry"></a> <!-- R; modified_bessel_second_kind; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#modified_bessel_second_kind" class="unlink" title="Jump to index entry"><code>modified_bessel_second_kind</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>modified_bessel_second_kind</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-501ae7bd47545a3ffa57a4c67cc0ffb27c69f93a" class="index-entry"></a> <!-- real; falling_factorial; (real x, real n); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#falling_factorial" class="unlink" title="Jump to index entry"><code>falling_factorial</code></a></strong><code>(real x, real n)</code><br> Return the falling factorial of x with power n defined for positive x and real n. <span class="math display">\[\begin{equation*}
+\mathrm{falling\_factorial}(x,n) =
+\begin{cases} (x)_n &amp; \text{if } x &gt; 0 \\ \textrm{error} &amp; \text{if } x \leq 0 \end{cases}
+\end{equation*}\]</span> where <span class="math display">\[\begin{equation*}
+(x)_n=\frac{\Gamma(x+1)}{\Gamma(x-n+1)}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-d534a0845142902e1168c6b016843f301f8826d6" class="index-entry"></a> <!-- R; falling_factorial; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#falling_factorial" class="unlink" title="Jump to index entry"><code>falling_factorial</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>falling_factorial</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8db53df4cf96e86f2466a1a67a046f1e92e24a41" class="index-entry"></a> <!-- real; lchoose; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lchoose" class="unlink" title="Jump to index entry"><code>lchoose</code></a></strong><code>(real x, real y)</code><br> Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as <span class="math inline">\(\binom{x}{y}\)</span> and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For <span class="math inline">\(0 \leq y \leq x\)</span>, <span class="math display">\[\begin{equation*} \mathrm{binomial\_coefficient\_log}(x,y) =
+\log\Gamma(x+1) - \log\Gamma(y+1) - \log\Gamma(x-y+1). \end{equation*}\]</span></p>
+<small><i>Available since 2.10</i></small>
+<a id="index-entry-57668b872958a1c748381739b2639e2c592b1635" class="index-entry"></a> <!-- R; lchoose; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lchoose" class="unlink" title="Jump to index entry"><code>lchoose</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>lchoose</code> function</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-914b5210fcade0135a4887b4af7ce01cf07f84bf" class="index-entry"></a> <!-- real; log_falling_factorial; (real x, real n); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_falling_factorial" class="unlink" title="Jump to index entry"><code>log_falling_factorial</code></a></strong><code>(real x, real n)</code><br> Return the log of the falling factorial of x with power n defined for positive x and real n.&nbsp;<span class="math display">\[\begin{equation*} \mathrm{log\_falling\_factorial}(x,n) =
+\begin{cases} \log (x)_n &amp; \text{if } x &gt; 0 \\ \textrm{error} &amp;
+\text{if } x \leq 0 \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-740279c8bcb940f23acd2a981763223a674105b8" class="index-entry"></a> <!-- real; rising_factorial; (real x, int n); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#rising_factorial" class="unlink" title="Jump to index entry"><code>rising_factorial</code></a></strong><code>(real x, int n)</code><br> Return the rising factorial of x with power n defined for positive x and integer n. <span class="math display">\[\begin{equation*}
+\mathrm{rising\_factorial}(x,n) = \begin{cases} x^{(n)} &amp; \text{if } x &gt; 0 \\ \textrm{error} &amp; \text{if } x \leq 0 \end{cases}
+\end{equation*}\]</span> where <span class="math display">\[\begin{equation*} x^{(n)}=\frac{\Gamma(x+n)}{\Gamma(x)} \end{equation*}\]</span></p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-1908eb05437a542da8fa4551ec25c60d068905cb" class="index-entry"></a> <!-- R; rising_factorial; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#rising_factorial" class="unlink" title="Jump to index entry"><code>rising_factorial</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>rising_factorial</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-72c705188e9fa2abc1fb72e724383889c9c3bc26" class="index-entry"></a> <!-- real; log_rising_factorial; (real x, real n); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_rising_factorial" class="unlink" title="Jump to index entry"><code>log_rising_factorial</code></a></strong><code>(real x, real n)</code><br> Return the log of the rising factorial of x with power n defined for positive x and real n.&nbsp;<span class="math display">\[\begin{equation*} \mathrm{log\_rising\_factorial}(x,n) =
+\begin{cases} \log x^{(n)} &amp; \text{if } x &gt; 0 \\ \textrm{error} &amp;
+\text{if } x \leq 0 \end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-67ec81c237fb98951165722160d48a4e93bbf167" class="index-entry"></a> <!-- R; log_rising_factorial; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_rising_factorial" class="unlink" title="Jump to index entry"><code>log_rising_factorial</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>log_rising_factorial</code> function</p>
+<small><i>Available since 2.25</i></small>
+</section>
+<section id="composed-functions" class="level2">
+<h2 class="anchored" data-anchor-id="composed-functions">Composed functions</h2>
+<p>The functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.</p>
+<a id="index-entry-159b5bb9f4c8cbaad0c033c64d1cd44196240c8d" class="index-entry"></a> <!-- R; expm1; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#expm1" class="unlink" title="Jump to index entry"><code>expm1</code></a></strong><code>(T x)</code><br> The natural exponential of x minus 1</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-086b6805da630957962521a7c4c5eb1c108313a2" class="index-entry"></a> <!-- real; fma; (real x, real y, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#fma" class="unlink" title="Jump to index entry"><code>fma</code></a></strong><code>(real x, real y, real z)</code><br> Return z plus the result of x multiplied by y. <span class="math display">\[\begin{equation*} \text{fma}(x,y,z) =
+(x \times y) + z \end{equation*}\]</span></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-87e15a5115cbb3320b6c75601022db5ce42d567f" class="index-entry"></a> <!-- real; ldexp; (real x, int y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#ldexp" class="unlink" title="Jump to index entry"><code>ldexp</code></a></strong><code>(real x, int y)</code><br> Return the product of x and two raised to the y power. <span class="math display">\[\begin{equation*}
+\text{ldexp}(x,y) = x 2^y  \end{equation*}\]</span></p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-04b6e7e9f36a90c1074fe831b302e5a839cded46" class="index-entry"></a> <!-- R; ldexp; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#ldexp" class="unlink" title="Jump to index entry"><code>ldexp</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>ldexp</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-d6659cf1923baf938d87ae82cf8fecf0665eb99d" class="index-entry"></a> <!-- real; lmultiply; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#lmultiply" class="unlink" title="Jump to index entry"><code>lmultiply</code></a></strong><code>(real x, real y)</code><br> Return the product of x and the natural logarithm of y. <span class="math display">\[\begin{equation*}
+\text{lmultiply}(x,y) = \begin{cases} 0 &amp; \text{if } x = y = 0 \\ x
+\log y &amp; \text{if } x, y \neq 0 \\ \text{NaN} &amp; \text{otherwise}
+\end{cases} \end{equation*}\]</span></p>
+<small><i>Available since 2.10</i></small>
+<a id="index-entry-5789074da7a4d0fb0e2acfe2db85408ed582f1a5" class="index-entry"></a> <!-- R; lmultiply; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lmultiply" class="unlink" title="Jump to index entry"><code>lmultiply</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>lmultiply</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-95ced8098d5ed0668cffa0d1be3921eed95f1d76" class="index-entry"></a> <!-- R; log1p; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log1p" class="unlink" title="Jump to index entry"><code>log1p</code></a></strong><code>(T x)</code><br> The natural logarithm of 1 plus x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-44e5298adea287581561adf0bc2875d063d97713" class="index-entry"></a> <!-- R; log1m; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log1m" class="unlink" title="Jump to index entry"><code>log1m</code></a></strong><code>(T x)</code><br> The natural logarithm of 1 minus x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-4f41ed63ba44a1e663dcc15f1dddbea84a2e1fde" class="index-entry"></a> <!-- R; log1p_exp; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log1p_exp" class="unlink" title="Jump to index entry"><code>log1p_exp</code></a></strong><code>(T x)</code><br> The natural logarithm of one plus the natural exponentiation of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-d3d38c55ec9be3639895f7ff75fac4b6b765c29a" class="index-entry"></a> <!-- R; log1m_exp; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log1m_exp" class="unlink" title="Jump to index entry"><code>log1m_exp</code></a></strong><code>(T x)</code><br> The logarithm of one minus the natural exponentiation of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-db5adfe0722d967066a4801ba7088570feda3d4f" class="index-entry"></a> <!-- real; log_diff_exp; (real x, real y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_diff_exp" class="unlink" title="Jump to index entry"><code>log_diff_exp</code></a></strong><code>(real x, real y)</code><br> Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. <span class="math display">\[\begin{equation*}
+\mathrm{log\_diff\_exp}(x,y) = \begin{cases} \log(\exp(x)-\exp(y)) &amp;
+\text{if } +\infty &gt; x \ge y \\[6pt]
+\textrm{NaN} &amp; \text{otherwise} \end{cases}
+\end{equation*}\]</span></p>
+<p>When x is equal to y, <code>log_diff_exp(x, y)</code> returns <span class="math inline">\(-\infty\)</span>, consistent with <code>log(0)</code> returning <span class="math inline">\(-\infty\)</span>. This includes the case in which x and y are both equal to <span class="math inline">\(-\infty\)</span>, which corresponds to <code>log(0 - 0)</code> because <code>exp(negative_infinity())</code> returns 0.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a68b0818c49fbd4972dfb2a14fea67387662718e" class="index-entry"></a> <!-- R; log_diff_exp; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_diff_exp" class="unlink" title="Jump to index entry"><code>log_diff_exp</code></a></strong><code>(T1 x, T2 y)</code><br> Vectorized implementation of the <code>log_diff_exp</code> function</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-569bcea73548d03bb5791681fcc60ee056c259b3" class="index-entry"></a> <!-- real; log_mix; (real theta, real lp1, real lp2); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#log_mix" class="unlink" title="Jump to index entry"><code>log_mix</code></a></strong><code>(real theta, real lp1, real lp2)</code><br> Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by <span class="math display">\[\begin{eqnarray*}
+\mathrm{log\_mix}(\theta, \lambda_1, \lambda_2) &amp; = &amp; \log \!\left(
+\theta \exp(\lambda_1) + \left( 1 - \theta \right) \exp(\lambda_2)
+\right) \\[3pt] &amp; = &amp; \mathrm{log\_sum\_exp}\!\left(\log(\theta) +
+\lambda_1, \ \log(1 - \theta) + \lambda_2\right). \end{eqnarray*}\]</span></p>
+<small><i>Available since 2.6</i></small>
+<a id="index-entry-83690ccdb1ecc891d26739b85cb08f317eece075" class="index-entry"></a> <!-- R; log_mix; (T1 thetas, T2 lps); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_mix" class="unlink" title="Jump to index entry"><code>log_mix</code></a></strong><code>(T1 thetas, T2 lps)</code><br></p>
+<p>Calculates the log mixture density given <code>thetas</code>, mixing proportions which should be between 0 and 1 and sum to 1, and <code>lps</code>, log densities. The <code>lps</code> variable must be either a 1-d container of the same length as <code>thetas</code>, or an array of such.</p>
+<p><span class="math display">\[\begin{eqnarray*}
+\mathrm{log\_mix}(\theta, \lambda)
+&amp; = &amp; \log \!\left( \sum_{n=1}^N \theta_n * \exp(\lambda_n) \right) \\[3pt]
+&amp; = &amp; \mathrm{log\_sum\_exp}\!\left(\log(\theta) + \lambda\right).
+\end{eqnarray*}\]</span></p>
+<p>This is a generalization of the above signature of three arguments to more than two densities. For example, <code>log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2})</code>.</p>
+<small><i>Available since 2.26</i></small>
+<a id="index-entry-da09d53c7797409929bfa868dfa8b30d767c1d2b" class="index-entry"></a> <!-- R; log_sum_exp; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_sum_exp" class="unlink" title="Jump to index entry"><code>log_sum_exp</code></a></strong><code>(T1 x, T2 y)</code><br> Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. <span class="math display">\[\begin{equation*}
+\mathrm{log\_sum\_exp}(x,y) = \log(\exp(x)+\exp(y)) \end{equation*}\]</span></p>
+<small><i>Available since 2.0, vectorized in 2.33</i></small>
+<a id="index-entry-35e5ab34a9cdca5c2c6b281a69b7bff27a2a200b" class="index-entry"></a> <!-- R; log_inv_logit; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_inv_logit" class="unlink" title="Jump to index entry"><code>log_inv_logit</code></a></strong><code>(T x)</code><br> The natural logarithm of the inverse logit function of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+<a id="index-entry-10df9e89f06a7af185332206207ab46c15793245" class="index-entry"></a> <!-- R; log_inv_logit_diff; (T1 x, T2 y); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log_inv_logit_diff" class="unlink" title="Jump to index entry"><code>log_inv_logit_diff</code></a></strong><code>(T1 x, T2 y)</code><br> The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-235d149b04f04823876ae89ac37a083db5908ce8" class="index-entry"></a> <!-- R; log1m_inv_logit; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#log1m_inv_logit" class="unlink" title="Jump to index entry"><code>log1m_inv_logit</code></a></strong><code>(T x)</code><br> The natural logarithm of 1 minus the inverse logit function of x</p>
+<small><i>Available since 2.0, vectorized in 2.13</i></small>
+</section>
+<section id="special-functions" class="level2">
+<h2 class="anchored" data-anchor-id="special-functions">Special functions</h2>
+<a id="index-entry-8762bfe8eb64d40ae8bf46eb733dc054da100f51" class="index-entry"></a> <!-- R; lambert_w0; (reals x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lambert_w0" class="unlink" title="Jump to index entry"><code>lambert_w0</code></a></strong><code>(T x)</code><br> Implementation of the <span class="math inline">\(W_0\)</span> branch of the Lambert W function, i.e., solution to the function <span class="math inline">\(W_0(x) \exp^{ W_0(x)} = x\)</span></p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-ab5ca07ba7ba53020939ca3ba7c6e64ccf05cf19" class="index-entry"></a> <!-- R; lambert_wm1; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#lambert_wm1" class="unlink" title="Jump to index entry"><code>lambert_wm1</code></a></strong><code>(T x)</code><br> Implementation of the <span class="math inline">\(W_{-1}\)</span> branch of the Lambert W function, i.e., solution to the function <span class="math inline">\(W_{-1}(x) \exp^{W_{-1}(x)} = x\)</span></p>
+<small><i>Available since 2.25</i></small>
+</section>
+<section id="hypergeometric-functions" class="level2">
+<h2 class="anchored" data-anchor-id="hypergeometric-functions">Hypergeometric Functions</h2>
+<p>Hypergeometric functions refer to a power series of the form <span class="math display">\[\begin{equation*}
+_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \sum_{n=0}^\infty \frac{(a_1)_n\cdot\cdot\cdot(a_p)_n}{(b_1)_n\cdot\cdot\cdot(b_q)_n} \frac{z^n}{n!}
+\end{equation*}\]</span> where <span class="math inline">\((a)_n\)</span> is the Pochhammer symbol defined as <span class="math inline">\((a)_n = \frac{\Gamma(a+n)}{\Gamma(a)}\)</span>.</p>
+<p>The gradients of the hypergeometric function are given by: <span class="math display">\[\begin{equation*}
+\frac{\partial }{\partial a_1} =
+  \sum_{k=0}^{\infty}{
+    \frac
+      {\psi\left(k+a_1\right)\left(\prod_{j=1}^p\left(a_j\right)_k\right)z^k}
+      {k!\prod_{j=1}^q\left(b_j\right)_k}}
+    - \psi\left(a_1\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)
+\end{equation*}\]</span> <span class="math display">\[\begin{equation*}
+\frac{\partial }{\partial b_1} =
+  \psi\left(b_1\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -
+  \sum_{k=0}^{\infty}{
+    \frac
+      {\psi\left(k+b_1\right)\left(\prod_{j=1}^p\left(a_j\right)_k\right)z^k}
+      {k!\prod_{j=1}^q\left(b_j\right)_k}}
+\end{equation*}\]</span> <span class="math display">\[\begin{equation*}
+  \frac{\partial }{\partial z} =
+  \frac{\prod_{j=1}^{p}a_j}{\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)
+\end{equation*}\]</span></p>
+<p>Stan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.</p>
+<a id="index-entry-e29b2c1e8db395e7d15862eb2f1f1e63ac58b10f" class="index-entry"></a> <!-- real; hypergeometric_1F0; (real a, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_1F0" class="unlink" title="Jump to index entry"><code>hypergeometric_1F0</code></a></strong><code>(real a, real z)</code><br> Special case of the hypergeometric function with <span class="math inline">\(p=1\)</span> and <span class="math inline">\(q=0\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-4ebcc0750c9365f9fd711bbd07b560cd76a3c016" class="index-entry"></a> <!-- real; hypergeometric_2F1; (real a1, real a2, real b1, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_2F1" class="unlink" title="Jump to index entry"><code>hypergeometric_2F1</code></a></strong><code>(real a1, real a2, real b1, real z)</code><br> Special case of the hypergeometric function with <span class="math inline">\(p=2\)</span> and <span class="math inline">\(q=1\)</span>. If the function does not meet convergence criteria for given inputs, the function will attempt to apply <a href="https://mathworld.wolfram.com/EulersHypergeometricTransformations.html">Euler’s transformation</a> to improve convergence: <span class="math display">\[\begin{equation*}
+{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \frac{z}{z-1})\cdot(1-z)^{-a_2}
+\end{equation*}\]</span></p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-f578817f233389fabf92d2db8bcfc1878fec4eca" class="index-entry"></a> <!-- real; hypergeometric_3F2; (T1 a, T2 b, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_3F2" class="unlink" title="Jump to index entry"><code>hypergeometric_3F2</code></a></strong><code>(T1 a, T2 b, real z)</code><br> Special case of the hypergeometric function with <span class="math inline">\(p=3\)</span> and <span class="math inline">\(q=2\)</span>, where a and b are vectors of length 3 and 2, respectively.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-2177c85f2269dea084da1c4e5b2a6f52c3337579" class="index-entry"></a> <!-- real; hypergeometric_pFq; (T1 a, T2 b, real z); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#hypergeometric_pFq" class="unlink" title="Jump to index entry"><code>hypergeometric_pFq</code></a></strong><code>(T1 a, T2 b, real z)</code><br> Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.</p>
+<small><i>Available since 2.37</i></small>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-BowlingEtAl:2009" class="csl-entry" role="listitem">
+Bowling, Shannon R., Mohammad T. Khasawneh, Sittichai Kaewkuekool, and Byung Rae Cho. 2009. <span>“A Logistic Approximation to the Cumulative Normal Distribution.”</span> <em>Journal of Industrial Engineering and Management</em> 2 (1): 114–27.
+</div>
+<div id="ref-Wichura:1988" class="csl-entry" role="listitem">
+Wichura, Michael J. 1988. <span>“Algorithm <span>AS</span> 241: The Percentage Points of the Normal Distribution.”</span> <em>Journal of the Royal Statistical Society. Series C (Applied Statistics)</em> 37 (3): 477–84. <a href="http://www.jstor.org/stable/2347330">http://www.jstor.org/stable/2347330</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/integer-valued_basic_functions.html" class="pagination-link" aria-label="Integer-Valued Basic Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Integer-Valued Basic Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/complex-valued_basic_functions.html" class="pagination-link" aria-label="Complex-Valued Basic Functions">
+        <span class="nav-page-text">Complex-Valued Basic Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/real-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/references.html b/docs/2_39/functions-reference/references.html
new file mode 100644
index 000000000..8c2cd1b2a
--- /dev/null
+++ b/docs/2_39/functions-reference/references.html
@@ -0,0 +1,821 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>references</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Docs">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#references" id="toc-references" class="nav-link active" data-scroll-target="#references"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
+
+
+
+
+<section id="references" class="level1 unnumbered">
+<h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></h1>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/removed_functions.html b/docs/2_39/functions-reference/removed_functions.html
new file mode 100644
index 000000000..869196b5c
--- /dev/null
+++ b/docs/2_39/functions-reference/removed_functions.html
@@ -0,0 +1,1209 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Removed Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/conventions_for_probability_functions.html" rel="next">
+<link href="../functions-reference/deprecated_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Removed Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/removed_functions.html">Removed Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#removed-functions" id="toc-removed-functions" class="nav-link active" data-scroll-target="#removed-functions">Removed Functions</a>
+  <ul class="collapse">
+  <li><a href="#multiply_log-and-binomial_coefficient_log-functions" id="toc-multiply_log-and-binomial_coefficient_log-functions" class="nav-link" data-scroll-target="#multiply_log-and-binomial_coefficient_log-functions"><code>multiply_log</code> and <code>binomial_coefficient_log</code> functions</a></li>
+  <li><a href="#get_lp-function" id="toc-get_lp-function" class="nav-link" data-scroll-target="#get_lp-function"><code>get_lp()</code> function</a></li>
+  <li><a href="#fabs-function" id="toc-fabs-function" class="nav-link" data-scroll-target="#fabs-function"><code>fabs</code> function</a></li>
+  <li><a href="#cov_exp_quad" id="toc-cov_exp_quad" class="nav-link" data-scroll-target="#cov_exp_quad">Exponentiated quadratic covariance functions</a></li>
+  <li><a href="#real-arguments-to-logical-operators-operator-operator-and-operator" id="toc-real-arguments-to-logical-operators-operator-operator-and-operator" class="nav-link" data-scroll-target="#real-arguments-to-logical-operators-operator-operator-and-operator">Real arguments to logical operators <code>operator&amp;&amp;</code>, <code>operator||</code>, and <code>operator!</code></a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/removed_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/removed_functions.html">Removed Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="removed-functions" class="level1">
+<h1>Removed Functions</h1>
+<p>Functions which once existed in the Stan language and have since been replaced or removed are listed here.</p>
+<section id="multiply_log-and-binomial_coefficient_log-functions" class="level2">
+<h2 class="anchored" data-anchor-id="multiply_log-and-binomial_coefficient_log-functions"><code>multiply_log</code> and <code>binomial_coefficient_log</code> functions</h2>
+<p><em>Removed</em>: Currently two non-conforming functions ending in suffix <code>_log</code>.</p>
+<p><em>Replacement</em>: Replace <code>multiply_log(...)</code> with <code>lmultiply(...)</code>. Replace <code>binomial_coefficient_log(...)</code> with <code>lchoose(...)</code>.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="get_lp-function" class="level2">
+<h2 class="anchored" data-anchor-id="get_lp-function"><code>get_lp()</code> function</h2>
+<p><em>Removed</em>: The built-in no-argument function <code>get_lp()</code> is deprecated.</p>
+<p><em>Replacement</em>: Use the no-argument function <code>target()</code> instead.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="fabs-function" class="level2">
+<h2 class="anchored" data-anchor-id="fabs-function"><code>fabs</code> function</h2>
+<p><em>Removed</em>: The unary function <code>fabs</code> is deprecated.</p>
+<p><em>Replacement</em>: Use the unary function <code>abs</code> instead. Note that the return type for <code>abs</code> is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="cov_exp_quad" class="level2">
+<h2 class="anchored" data-anchor-id="cov_exp_quad">Exponentiated quadratic covariance functions</h2>
+<p>These covariance functions have been replaced by those described in <a href="../functions-reference/matrix_operations.html#gaussian-process-covariance-functions">Gaussian Process Covariance Functions</a></p>
+<p>With magnitude <span class="math inline">\(\alpha\)</span> and length scale <span class="math inline">\(l\)</span>, the exponentiated quadratic kernel is:</p>
+<p><span class="math display">\[
+k(x_i, x_j) = \alpha^2 \exp \left(-\dfrac{1}{2\rho^2} \sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \right)
+\]</span></p>
+<a id="index-entry-92099ecf9309d8f643e455cb6e3f79c985d35cb1" class="index-entry"></a> <!-- matrix; cov_exp_quad; (row_vectors x, real alpha, real rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cov_exp_quad" class="unlink" title="Jump to index entry"><code>cov_exp_quad</code></a></strong><code>(row_vectors x, real alpha, real rho)</code><br> The covariance matrix with an exponentiated quadratic kernel of x.</p>
+<small><i>Available since 2.16, deprecated since 2.20, removed in in 2.33</i></small>
+<a id="index-entry-d271045bb5d0614ff3a2f0e3d8d279d134cf0304" class="index-entry"></a> <!-- matrix; cov_exp_quad; (vectors x, real alpha, real rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cov_exp_quad" class="unlink" title="Jump to index entry"><code>cov_exp_quad</code></a></strong><code>(vectors x, real alpha, real rho)</code><br> The covariance matrix with an exponentiated quadratic kernel of x.</p>
+<small><i>Available since 2.16, deprecated since 2.20, removed in in 2.33</i></small>
+<a id="index-entry-1ba90fc13ad6ba28ca73fc3b94b0a65af09669f8" class="index-entry"></a> <!-- matrix; cov_exp_quad; (array[] real x, real alpha, real rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cov_exp_quad" class="unlink" title="Jump to index entry"><code>cov_exp_quad</code></a></strong><code>(array[] real x, real alpha, real rho)</code><br> The covariance matrix with an exponentiated quadratic kernel of x.</p>
+<small><i>Available since 2.16, deprecated since 2.20, removed in in 2.33</i></small>
+<a id="index-entry-c05741ae8e35d6b11a03c0098ac00b99d5a5c845" class="index-entry"></a> <!-- matrix; cov_exp_quad; (row_vectors x1, row_vectors x2, real alpha, real rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cov_exp_quad" class="unlink" title="Jump to index entry"><code>cov_exp_quad</code></a></strong><code>(row_vectors x1, row_vectors x2, real alpha, real rho)</code><br> The covariance matrix with an exponentiated quadratic kernel of x1 and x2.</p>
+<small><i>Available since 2.18, deprecated since 2.20, removed in in 2.33</i></small>
+<a id="index-entry-70b15de7e29e4e64f05ca3fc552da349d4a66c1f" class="index-entry"></a> <!-- matrix; cov_exp_quad; (vectors x1, vectors x2, real alpha, real rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cov_exp_quad" class="unlink" title="Jump to index entry"><code>cov_exp_quad</code></a></strong><code>(vectors x1, vectors x2, real alpha, real rho)</code><br> The covariance matrix with an exponentiated quadratic kernel of x1 and x2.</p>
+<small><i>Available since 2.18, deprecated since 2.20, removed in in 2.33</i></small>
+<a id="index-entry-95bfeb84294b276027c24d3f8856560234423e7c" class="index-entry"></a> <!-- matrix; cov_exp_quad; (array[] real x1, array[] real x2, real alpha, real rho); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#cov_exp_quad" class="unlink" title="Jump to index entry"><code>cov_exp_quad</code></a></strong><code>(array[] real x1, array[] real x2, real alpha, real rho)</code><br> The covariance matrix with an exponentiated quadratic kernel of x1 and x2.</p>
+<small><i>Available since 2.18, deprecated since 2.20, removed in in 2.33</i></small>
+</section>
+<section id="real-arguments-to-logical-operators-operator-operator-and-operator" class="level2">
+<h2 class="anchored" data-anchor-id="real-arguments-to-logical-operators-operator-operator-and-operator">Real arguments to logical operators <code>operator&amp;&amp;</code>, <code>operator||</code>, and <code>operator!</code></h2>
+<p><em>Removed</em>: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.</p>
+<p><em>Replacement</em>: Explicit <code>x != 0</code> comparison is preferred instead.</p>
+<p><em>Removed In</em>: Stan 2.34</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/deprecated_functions.html" class="pagination-link" aria-label="Deprecated Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Deprecated Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/conventions_for_probability_functions.html" class="pagination-link" aria-label="Conventions for Probability Functions">
+        <span class="nav-page-text">Conventions for Probability Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/removed_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/simplex_distributions.html b/docs/2_39/functions-reference/simplex_distributions.html
new file mode 100644
index 000000000..8d7aafd24
--- /dev/null
+++ b/docs/2_39/functions-reference/simplex_distributions.html
@@ -0,0 +1,1256 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Simplex Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/correlation_matrix_distributions.html" rel="next">
+<link href="../functions-reference/distributions_over_unbounded_vectors.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Simplex Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/simplex_distributions.html">Simplex Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#simplex-distributions" id="toc-simplex-distributions" class="nav-link active" data-scroll-target="#simplex-distributions">Simplex Distributions</a>
+  <ul class="collapse">
+  <li><a href="#dirichlet-distribution" id="toc-dirichlet-distribution" class="nav-link" data-scroll-target="#dirichlet-distribution">Dirichlet distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#meaning-of-dirichlet-parameters" id="toc-meaning-of-dirichlet-parameters" class="nav-link" data-scroll-target="#meaning-of-dirichlet-parameters">Meaning of Dirichlet parameters</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/simplex_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/simplex_distributions.html">Simplex Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="simplex-distributions" class="level1">
+<h1>Simplex Distributions</h1>
+<p>The simplex probabilities have support on the unit <span class="math inline">\(K\)</span>-simplex for a specified <span class="math inline">\(K\)</span>. A <span class="math inline">\(K\)</span>-dimensional vector <span class="math inline">\(\theta\)</span> is a unit <span class="math inline">\(K\)</span>-simplex if <span class="math inline">\(\theta_k \geq 0\)</span> for <span class="math inline">\(k \in \{1,\ldots,K\}\)</span> and <span class="math inline">\(\sum_{k = 1}^K \theta_k = 1\)</span>.</p>
+<section id="dirichlet-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="dirichlet-distribution">Dirichlet distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(K \in \mathbb{N}\)</span> and <span class="math inline">\(\alpha \in (\mathbb{R}^+)^{K}\)</span>, then for <span class="math inline">\(\theta \in \text{$K$-simplex}\)</span>,</p>
+<p><span class="math display">\[\begin{equation*}
+\text{Dirichlet}(\theta|\alpha) =
+\frac{\Gamma \! \left( \sum_{k=1}^K \alpha_k \right)}{\prod_{k=1}^K \Gamma(\alpha_k)}  \prod_{k=1}^K \theta_k^{\alpha_k -1}
+\end{equation*}\]</span></p>
+<p><em><strong>Warning:</strong></em> If any of the components of <span class="math inline">\(\theta\)</span> satisfies <span class="math inline">\(\theta_i = 0\)</span> or <span class="math inline">\(\theta_i = 1\)</span>, then the probability is <span class="math inline">\(0\)</span> and the log probability is <span class="math inline">\(-\infty\)</span>. Similarly, the distribution requires strictly positive parameters, with <span class="math inline">\(\alpha_i &gt; 0\)</span> for each <span class="math inline">\(i\)</span>.</p>
+</section>
+<section id="meaning-of-dirichlet-parameters" class="level3">
+<h3 class="anchored" data-anchor-id="meaning-of-dirichlet-parameters">Meaning of Dirichlet parameters</h3>
+<p>A symmetric Dirichlet prior is <span class="math inline">\([\alpha, \ldots, \alpha]^{\top}\)</span>. To code this in Stan,</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a> <span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>   <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>   <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a> }</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a> <span class="kw">generated quantities</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>   <span class="dt">vector</span>[K] theta = dirichlet_rng(rep_vector(alpha, K));</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a> }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Taking <span class="math inline">\(K = 10\)</span>, here are the first five draws for <span class="math inline">\(\alpha = 1\)</span>. For <span class="math inline">\(\alpha = 1\)</span>, the distribution is uniform over simplexes.</p>
+<pre><code> 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05
+ 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06
+ 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03
+ 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18
+ 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01</code></pre>
+<p>That does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) <span class="math inline">\(1/K\)</span>. When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on <span class="math inline">\(10\)</span>-simplexes, the probability that a component is greater than the mean of <span class="math inline">\(1/10\)</span> is only 39%. Most of the posterior marginal probability mass for each component is in the interval <span class="math inline">\((0, 0.1)\)</span>.</p>
+<p>When the <span class="math inline">\(\alpha\)</span> value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for <span class="math inline">\(\alpha = 0.001\)</span>.</p>
+<pre><code> 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279
+ 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000
+ 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112
+ 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000
+ 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000</code></pre>
+<p>Each row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.</p>
+<p>As <span class="math inline">\(\alpha\)</span> increases, the draws become increasingly uniform. For <span class="math inline">\(\alpha = 1000\)</span>,</p>
+<pre><code> 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10
+ 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10
+ 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10
+ 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10
+ 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10</code></pre>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>theta ~</code> <strong><a href="../functions-reference/functions_index.html#dirichlet" class="unlink" title="Jump to index entry"><code>dirichlet</code></a></strong><code>(alpha)</code></p>
+<p>Increment target log probability density with <code>dirichlet_lupdf(theta | alpha)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2b1866f861130d47e076c909b2da014d36b08615" class="index-entry"></a> <!-- real; dirichlet ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<p>The Dirichlet probability functions are overloaded to allow the simplex <span class="math inline">\(\theta\)</span> and prior counts (plus one) <span class="math inline">\(\alpha\)</span> to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a> for a description of vectorization.</p>
+<a id="index-entry-912dbafe0e9fce423e5413cf88d46731b610742d" class="index-entry"></a> <!-- real; dirichlet_lpdf; (vectors theta | vectors alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_lpdf" class="unlink" title="Jump to index entry"><code>dirichlet_lpdf</code></a></strong><code>(vectors theta | vectors alpha)</code><br> The log of the Dirichlet density for simplex(es) <code>theta</code> given prior counts (plus one) <code>alpha</code></p>
+<small><i>Available since 2.12, vectorized in 2.21</i></small>
+<a id="index-entry-d287f450cfac7ef63c57f921d82484ef44abdfa0" class="index-entry"></a> <!-- real; dirichlet_lupdf; (vectors theta | vectors alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_lupdf" class="unlink" title="Jump to index entry"><code>dirichlet_lupdf</code></a></strong><code>(vectors theta | vectors alpha)</code><br> The log of the Dirichlet density for simplex(es) <code>theta</code> given prior counts (plus one) <code>alpha</code> dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-055ed033fb2096aa99f68c19fe9c3a956577f3b1" class="index-entry"></a> <!-- vector; dirichlet_rng; (vector alpha); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#dirichlet_rng" class="unlink" title="Jump to index entry"><code>dirichlet_rng</code></a></strong><code>(vector alpha)</code><br> Generate a Dirichlet variate with prior counts (plus one) <code>alpha</code>; may only be used in transformed data and generated quantities blocks</p>
+<small><i>Available since 2.0</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="pagination-link" aria-label="Distributions over Unbounded Vectors">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Distributions over Unbounded Vectors</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/correlation_matrix_distributions.html" class="pagination-link" aria-label="Correlation Matrix Distributions">
+        <span class="nav-page-text">Correlation Matrix Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/simplex_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/sparse_matrix_operations.html b/docs/2_39/functions-reference/sparse_matrix_operations.html
new file mode 100644
index 000000000..e7596edd0
--- /dev/null
+++ b/docs/2_39/functions-reference/sparse_matrix_operations.html
@@ -0,0 +1,1217 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Sparse Matrix Operations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/mixed_operations.html" rel="next">
+<link href="../functions-reference/complex_matrix_operations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Sparse Matrix Operations">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/sparse_matrix_operations.html">Sparse Matrix Operations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#sparse-matrices" id="toc-sparse-matrices" class="nav-link active" data-scroll-target="#sparse-matrices">Sparse Matrix Operations</a>
+  <ul class="collapse">
+  <li><a href="#CSR" id="toc-CSR" class="nav-link" data-scroll-target="#CSR">Compressed row storage</a></li>
+  <li><a href="#conversion-functions" id="toc-conversion-functions" class="nav-link" data-scroll-target="#conversion-functions">Conversion functions</a>
+  <ul class="collapse">
+  <li><a href="#dense-to-sparse-conversion" id="toc-dense-to-sparse-conversion" class="nav-link" data-scroll-target="#dense-to-sparse-conversion">Dense to sparse conversion</a></li>
+  <li><a href="#sparse-to-dense-conversion" id="toc-sparse-to-dense-conversion" class="nav-link" data-scroll-target="#sparse-to-dense-conversion">Sparse to dense conversion</a></li>
+  </ul></li>
+  <li><a href="#sparse-matrix-arithmetic" id="toc-sparse-matrix-arithmetic" class="nav-link" data-scroll-target="#sparse-matrix-arithmetic">Sparse matrix arithmetic</a>
+  <ul class="collapse">
+  <li><a href="#sparse-matrix-multiplication" id="toc-sparse-matrix-multiplication" class="nav-link" data-scroll-target="#sparse-matrix-multiplication">Sparse matrix multiplication</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/sparse_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/sparse_matrix_operations.html">Sparse Matrix Operations</a></li></ol></nav></header>
+
+
+
+
+<section id="sparse-matrices" class="level1">
+<h1>Sparse Matrix Operations</h1>
+<p>For sparse matrices, for which many elements are zero, it is more efficient to use specialized representations to save memory and speed up matrix arithmetic (including derivative calculations). Given Stan’s implementation, there is substantial space (memory) savings by using sparse matrices. Because of the ease of optimizing dense matrix operations, speed improvements only arise at 90% or even greater sparsity; below that level, dense matrices are faster but use more memory.</p>
+<p>Because of this speedup and space savings, it may even be useful to read in a dense matrix and convert it to a sparse matrix before multiplying it by a vector. This chapter covers a very specific form of sparsity consisting of a sparse matrix multiplied by a dense vector.</p>
+<section id="CSR" class="level2">
+<h2 class="anchored" data-anchor-id="CSR">Compressed row storage</h2>
+<p>Sparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix <span class="math display">\[\begin{equation*}
+A = \begin{bmatrix} 19 &amp; 27 &amp; 0 &amp; 0 \\ 0 &amp; 0 &amp; 0 &amp; 0 \\ 0 &amp; 0 &amp; 0 &amp; 52 \\ 81 &amp; 0 &amp; 95 &amp; 33 \end{bmatrix}
+\end{equation*}\]</span> is translated into a vector of the non-zero real values, read by row from the matrix <span class="math inline">\(A\)</span>, <span class="math display">\[\begin{equation*}
+w(A) = \begin{bmatrix} 19 &amp; 27 &amp; 52 &amp; 81 &amp; 95 &amp; 33 \end{bmatrix}^{\top} \! \! \! ,
+\end{equation*}\]</span> an array of integer column indices for the values, <span class="math display">\[\begin{equation*}
+v(A) = \begin{bmatrix} 1 &amp; 2 &amp; 4 &amp; 1 &amp; 3 &amp; 4 \end{bmatrix} \! ,
+\end{equation*}\]</span> and an array of integer indices indicating where in <span class="math inline">\(w(A)\)</span> a given row’s values start, <span class="math display">\[\begin{equation*} u(A) = \begin{bmatrix} 1 &amp; 3 &amp; 3 &amp; 4 &amp; 7 \end{bmatrix} \! ,
+\end{equation*}\]</span> with a padded value at the end to guarantee that <span class="math display">\[\begin{equation*} u(A)[n+1] - u(A)[n] \end{equation*}\]</span> is the number of non-zero elements in row <span class="math inline">\(n\)</span> of the matrix (here <span class="math inline">\(2\)</span>, <span class="math inline">\(0\)</span>, <span class="math inline">\(1\)</span>, and <span class="math inline">\(3\)</span>). Note that because the second row has no non-zero elements both the second and third elements of <span class="math inline">\(u(A)\)</span> correspond to the third element of <span class="math inline">\(w(A)\)</span>, which is <span class="math inline">\(52\)</span>. The values <span class="math inline">\((w(A), \, v(A), \, u(A))\)</span> are sufficient to reconstruct <span class="math inline">\(A\)</span>.</p>
+<p>The values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly <span class="math inline">\(12 K + M\)</span> bytes plus a small constant overhead, which is often considerably fewer bytes than the <span class="math inline">\(M \times
+N\)</span> required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.</p>
+</section>
+<section id="conversion-functions" class="level2">
+<h2 class="anchored" data-anchor-id="conversion-functions">Conversion functions</h2>
+<p>Conversion functions between dense and sparse matrices are provided.</p>
+<section id="dense-to-sparse-conversion" class="level3">
+<h3 class="anchored" data-anchor-id="dense-to-sparse-conversion">Dense to sparse conversion</h3>
+<p>Converting a dense matrix <span class="math inline">\(m\)</span> to a sparse representation produces a vector <span class="math inline">\(w\)</span> and two integer arrays, <span class="math inline">\(u\)</span> and <span class="math inline">\(v\)</span>.</p>
+<a id="index-entry-fa08d7d6e24c28fcab7d2276501e5cbf1b12e09d" class="index-entry"></a> <!-- vector; csr_extract_w; (matrix a); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#csr_extract_w" class="unlink" title="Jump to index entry"><code>csr_extract_w</code></a></strong><code>(matrix a)</code><br> Return non-zero values in matrix a; see section <a href="#CSR">compressed row storage</a>.</p>
+<small><i>Available since 2.8</i></small>
+<a id="index-entry-2569af86a359f94d59f18b21a0b4a15d7ad13f5f" class="index-entry"></a> <!-- array[] int; csr_extract_v; (matrix a); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#csr_extract_v" class="unlink" title="Jump to index entry"><code>csr_extract_v</code></a></strong><code>(matrix a)</code><br> Return column indices for values in <code>csr_extract_w(a)</code>; see <a href="#CSR">compressed row storage</a>.</p>
+<small><i>Available since 2.8</i></small>
+<a id="index-entry-1f750e4b73b9d17d5eadd184840443bab5de2dda" class="index-entry"></a> <!-- array[] int; csr_extract_u; (matrix a); -->
+<p></p>
+<p><code>array[] int</code> <strong><a href="../functions-reference/functions_index.html#csr_extract_u" class="unlink" title="Jump to index entry"><code>csr_extract_u</code></a></strong><code>(matrix a)</code><br> Return array of row starting indices for entries in <code>csr_extract_w(a)</code> followed by the size of <code>csr_extract_w(a)</code> plus one; see section <a href="#CSR">compressed row storage</a>.</p>
+<small><i>Available since 2.8</i></small>
+<a id="index-entry-32fea6602ecbd4c1ea15d25bb2a6b2d2fcf58e52" class="index-entry"></a> <!-- tuple(vector, array[] int, array[] int); csr_extract; (matrix a); -->
+<p></p>
+<p><code>tuple(vector, array[] int, array[] int)</code> <strong><a href="../functions-reference/functions_index.html#csr_extract" class="unlink" title="Jump to index entry"><code>csr_extract</code></a></strong><code>(matrix a)</code><br> Return all three components of the CSR representation of the matrix <code>a</code>; see section <a href="#CSR">compressed row storage</a>. This function is equivalent to <code>(csr_extract_w(a), csr_extract_v(a), csr_extract_u(a))</code>.</p>
+<small><i>Available since 2.33</i></small>
+</section>
+<section id="sparse-to-dense-conversion" class="level3">
+<h3 class="anchored" data-anchor-id="sparse-to-dense-conversion">Sparse to dense conversion</h3>
+<p>To convert a sparse matrix representation to a dense matrix, there is a single function.</p>
+<a id="index-entry-b8ce33cfed90d588aaa17f7c95c10b4206863d15" class="index-entry"></a> <!-- matrix; csr_to_dense_matrix; (int m, int n, vector w, array[] int v, array[] int u); -->
+<p></p>
+<p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#csr_to_dense_matrix" class="unlink" title="Jump to index entry"><code>csr_to_dense_matrix</code></a></strong><code>(int m, int n, vector w, array[] int v, array[] int u)</code><br> Return dense <span class="math inline">\(\text{m} \times \text{n}\)</span> matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section <a href="#CSR">compressed row storage</a> for more details.</p>
+<small><i>Available since 2.10</i></small>
+</section>
+</section>
+<section id="sparse-matrix-arithmetic" class="level2">
+<h2 class="anchored" data-anchor-id="sparse-matrix-arithmetic">Sparse matrix arithmetic</h2>
+<section id="sparse-matrix-multiplication" class="level3">
+<h3 class="anchored" data-anchor-id="sparse-matrix-multiplication">Sparse matrix multiplication</h3>
+<p>The only supported operation is the multiplication of a sparse matrix <span class="math inline">\(A\)</span> and a dense vector <span class="math inline">\(b\)</span> to produce a dense vector <span class="math inline">\(A\,b\)</span>. Multiplying a dense row vector <span class="math inline">\(b\)</span> and a sparse matrix <span class="math inline">\(A\)</span> can be coded using transposition as <span class="math display">\[\begin{equation*}
+b \, A = (A^{\top} \, b^{\top})^{\top},
+\end{equation*}\]</span> but care must be taken to represent <span class="math inline">\(A^{\top}\)</span> rather than <span class="math inline">\(A\)</span> as a sparse matrix.</p>
+<a id="index-entry-a5e3d826290e2d8bff8cf1e1019bd2656ad43d2b" class="index-entry"></a> <!-- vector; csr_matrix_times_vector; (int m, int n, vector w, array[] int v, array[] int u, vector b); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#csr_matrix_times_vector" class="unlink" title="Jump to index entry"><code>csr_matrix_times_vector</code></a></strong><code>(int m, int n, vector w, array[] int v, array[] int u, vector b)</code><br> Multiply the <span class="math inline">\(\text{m} \times \text{n}\)</span> matrix represented by values w, column indices v, and row start indices u by the vector b; see <a href="#CSR">compressed row storage</a>.</p>
+<small><i>Available since 2.18</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/complex_matrix_operations.html" class="pagination-link" aria-label="Complex Matrix Operations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Complex Matrix Operations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/mixed_operations.html" class="pagination-link" aria-label="Mixed Operations">
+        <span class="nav-page-text">Mixed Operations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/sparse_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/transform_functions.html b/docs/2_39/functions-reference/transform_functions.html
new file mode 100644
index 000000000..7311ee935
--- /dev/null
+++ b/docs/2_39/functions-reference/transform_functions.html
@@ -0,0 +1,1469 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Variable Transformation Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/deprecated_functions.html" rel="next">
+<link href="../functions-reference/higher-order_functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Variable Transformation Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/transform_functions.html">Variable Transformation Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#variable-transformation-functions" id="toc-variable-transformation-functions" class="nav-link active" data-scroll-target="#variable-transformation-functions">Variable Transformation Functions</a>
+  <ul class="collapse">
+  <li><a href="#transforms-for-scalars" id="toc-transforms-for-scalars" class="nav-link" data-scroll-target="#transforms-for-scalars">Transforms for scalars</a>
+  <ul class="collapse">
+  <li><a href="#lower-bounds" id="toc-lower-bounds" class="nav-link" data-scroll-target="#lower-bounds">Lower bounds</a></li>
+  <li><a href="#upper-bounds" id="toc-upper-bounds" class="nav-link" data-scroll-target="#upper-bounds">Upper bounds</a></li>
+  <li><a href="#upper-and-lower-bounds" id="toc-upper-and-lower-bounds" class="nav-link" data-scroll-target="#upper-and-lower-bounds">Upper and lower bounds</a></li>
+  <li><a href="#affine-transforms" id="toc-affine-transforms" class="nav-link" data-scroll-target="#affine-transforms">Affine transforms</a></li>
+  </ul></li>
+  <li><a href="#transforms-for-constrained-vectors" id="toc-transforms-for-constrained-vectors" class="nav-link" data-scroll-target="#transforms-for-constrained-vectors">Transforms for constrained vectors</a>
+  <ul class="collapse">
+  <li><a href="#ordered-vectors" id="toc-ordered-vectors" class="nav-link" data-scroll-target="#ordered-vectors">Ordered vectors</a></li>
+  <li><a href="#positive-order-vectors" id="toc-positive-order-vectors" class="nav-link" data-scroll-target="#positive-order-vectors">Positive order vectors</a></li>
+  <li><a href="#simplexes" id="toc-simplexes" class="nav-link" data-scroll-target="#simplexes">Simplexes</a></li>
+  <li><a href="#sum-to-zero-vectors" id="toc-sum-to-zero-vectors" class="nav-link" data-scroll-target="#sum-to-zero-vectors">Sum-to-zero vectors</a></li>
+  <li><a href="#unit-vectors" id="toc-unit-vectors" class="nav-link" data-scroll-target="#unit-vectors">Unit vectors</a></li>
+  </ul></li>
+  <li><a href="#transforms-for-constrained-matrices" id="toc-transforms-for-constrained-matrices" class="nav-link" data-scroll-target="#transforms-for-constrained-matrices">Transforms for constrained matrices</a>
+  <ul class="collapse">
+  <li><a href="#cholesky-factors-of-correlation-matrices" id="toc-cholesky-factors-of-correlation-matrices" class="nav-link" data-scroll-target="#cholesky-factors-of-correlation-matrices">Cholesky factors of correlation matrices</a></li>
+  <li><a href="#cholesky-factors-of-covariance-matrices" id="toc-cholesky-factors-of-covariance-matrices" class="nav-link" data-scroll-target="#cholesky-factors-of-covariance-matrices">Cholesky factors of covariance matrices</a></li>
+  <li><a href="#correlation-matrices" id="toc-correlation-matrices" class="nav-link" data-scroll-target="#correlation-matrices">Correlation matrices</a></li>
+  <li><a href="#covariance-matrices" id="toc-covariance-matrices" class="nav-link" data-scroll-target="#covariance-matrices">Covariance matrices</a></li>
+  <li><a href="#column-stochastic-matrices" id="toc-column-stochastic-matrices" class="nav-link" data-scroll-target="#column-stochastic-matrices">Column-stochastic matrices</a></li>
+  <li><a href="#row-stochastic-matrices" id="toc-row-stochastic-matrices" class="nav-link" data-scroll-target="#row-stochastic-matrices">Row-stochastic matrices</a></li>
+  <li><a href="#sum-to-zero-matrices" id="toc-sum-to-zero-matrices" class="nav-link" data-scroll-target="#sum-to-zero-matrices">Sum-to-zero matrices</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/transform_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/transform_functions.html">Variable Transformation Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="variable-transformation-functions" class="level1">
+<h1>Variable Transformation Functions</h1>
+<p>Variable transformation functions provide implementations of the built-in constraining and unconstraining transforms defined in <a href="https://mc-stan.org/docs/reference-manual/transforms.html">Stan Reference Manual</a>.</p>
+<p>For each of the built-in variable transforms there are three functions named after the transform with differing suffixes. A <code>_unconstrain</code> function that maps from the constrained space back to free variables (the “transform”), A <code>_constrain</code> function that maps from free variables to constrained variables (the “inverse transform”), and a <a href="https://mc-stan.org/docs/reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment"><code>_jacobian</code> function</a>, which computes the same value as the <code>_constrain</code> function while also incrementing the Jacobian accumulator with the log Jacobian determinant.</p>
+<p>For this page, variables named <code>y</code> are unconstrained, while variables named <code>x</code> are in the constrained space. The unconstraining functions will reject if their input does not satisfy the declared constraint.</p>
+<section id="transforms-for-scalars" class="level2">
+<h2 class="anchored" data-anchor-id="transforms-for-scalars">Transforms for scalars</h2>
+<p>These transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.</p>
+<p>The functions are all overloaded to apply to containers elementwise. If the <code>y</code> argument is a container, the others must be either scalars or containers of exactly the same size.</p>
+<section id="lower-bounds" class="level3">
+<h3 class="anchored" data-anchor-id="lower-bounds">Lower bounds</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#lower-bound-transform.section">Lower Bounded Scalar section</a>.</p>
+<a id="index-entry-e152c4494a196c9bd932faad21a3555c712d102c" class="index-entry"></a> <!-- reals; lower_bound_constrain; (reals y, reals lb); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#lower_bound_constrain" class="unlink" title="Jump to index entry"><code>lower_bound_constrain</code></a></strong><code>(reals y, reals lb)</code><br> Takes a value <code>y</code> and lower bound <code>lb</code> and returns the corresponding value which is greater than <code>lb</code> (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-dda4a6ffc09ae6ffec4218c092003fea973c174e" class="index-entry"></a> <!-- reals; lower_bound_jacobian; (reals y, reals lb); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#lower_bound_jacobian" class="unlink" title="Jump to index entry"><code>lower_bound_jacobian</code></a></strong><code>(reals y, reals lb)</code><br> Takes a value <code>y</code> and lower bound <code>lb</code> and returns the corresponding value which is greater than <code>lb</code> (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-d178c95f782831ea0616fdc58b25d16826f98164" class="index-entry"></a> <!-- reals; lower_bound_unconstrain; (reals x, reals lb); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#lower_bound_unconstrain" class="unlink" title="Jump to index entry"><code>lower_bound_unconstrain</code></a></strong><code>(reals x, reals lb)</code><br> Takes a value <code>x</code> which is greater than <code>lb</code> and returns the corresponding unconstrained value.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="upper-bounds" class="level3">
+<h3 class="anchored" data-anchor-id="upper-bounds">Upper bounds</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#upper-bounded-scalar">Upper Bounded Scalar section</a>.</p>
+<a id="index-entry-762c88752de52b0a7df91820157122b1ddd4e9f1" class="index-entry"></a> <!-- reals; upper_bound_constrain; (reals y, reals ub); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#upper_bound_constrain" class="unlink" title="Jump to index entry"><code>upper_bound_constrain</code></a></strong><code>(reals y, reals ub)</code><br> Takes a value <code>y</code> and upper bound <code>ub</code> and returns the corresponding value which is less than <code>ub</code> (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-f2ff19db2443c6d37b95266808e87959db60e3cd" class="index-entry"></a> <!-- reals; upper_bound_jacobian; (reals x, reals ub); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#upper_bound_jacobian" class="unlink" title="Jump to index entry"><code>upper_bound_jacobian</code></a></strong><code>(reals x, reals ub)</code><br> Takes a value <code>y</code> and upper bound <code>ub</code> and returns the corresponding value which is less than <code>ub</code> (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-2addb4ba29c3510e0935adaeb229c3bb34d49798" class="index-entry"></a> <!-- reals; upper_bound_unconstrain; (reals x, reals ub); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#upper_bound_unconstrain" class="unlink" title="Jump to index entry"><code>upper_bound_unconstrain</code></a></strong><code>(reals x, reals ub)</code><br> Takes a value <code>x</code> which is less than <code>ub</code> and returns the corresponding unconstrained value.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="upper-and-lower-bounds" class="level3">
+<h3 class="anchored" data-anchor-id="upper-and-lower-bounds">Upper and lower bounds</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#logit-transform-jacobian.section">Lower and Upper Bounded Scalar section</a>.</p>
+<a id="index-entry-d44d27ade0aed4fc8aa29d480aa1cbd767281f72" class="index-entry"></a> <!-- reals; lower_upper_bound_constrain; (reals y, reals lb, reals ub); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#lower_upper_bound_constrain" class="unlink" title="Jump to index entry"><code>lower_upper_bound_constrain</code></a></strong><code>(reals y, reals lb, reals ub)</code><br> Takes a value <code>y</code>, lower bound <code>lb</code>, and upper bound <code>ub</code> and returns the corresponding value which is bounded between <code>lb</code> and <code>ub</code> (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-fa7585cac2eb147a254907e9aacffb346393c41e" class="index-entry"></a> <!-- reals; lower_upper_bound_jacobian; (reals y, reals lb, reals ub); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#lower_upper_bound_jacobian" class="unlink" title="Jump to index entry"><code>lower_upper_bound_jacobian</code></a></strong><code>(reals y, reals lb, reals ub)</code><br> Takes a value <code>y</code>, lower bound <code>lb</code>, and upper bound <code>ub</code> and returns the corresponding value which is bounded between <code>lb</code> and <code>ub</code> (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-da54a8c2ddfb181210e40b0e47d122f16713af86" class="index-entry"></a> <!-- reals; lower_upper_bound_unconstrain; (reals x, reals lb, reals ub); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#lower_upper_bound_unconstrain" class="unlink" title="Jump to index entry"><code>lower_upper_bound_unconstrain</code></a></strong><code>(reals x, reals lb, reals ub)</code><br> Takes a value <code>x</code> which is bounded between <code>lb</code> and <code>ub</code> and returns returns the corresponding unconstrained value.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="affine-transforms" class="level3">
+<h3 class="anchored" data-anchor-id="affine-transforms">Affine transforms</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#affinely-transformed-scalar">Affinely Transformed Scalar section</a>.</p>
+<a id="index-entry-c55eb604debca345fb51ad9d71cb36b6f20bb60e" class="index-entry"></a> <!-- reals; offset_multiplier_constrain; (reals y, reals offset, reals mult); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#offset_multiplier_constrain" class="unlink" title="Jump to index entry"><code>offset_multiplier_constrain</code></a></strong><code>(reals y, reals offset, reals mult)</code><br> Takes a value <code>y</code>, shift <code>offset</code>, and scale <code>mult</code> and returns a rescaled and shifted value.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-2448386d5029a2ea68a25d2eac851459f98eb968" class="index-entry"></a> <!-- reals; offset_multiplier_jacobian; (reals y, reals offset, reals mult); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#offset_multiplier_jacobian" class="unlink" title="Jump to index entry"><code>offset_multiplier_jacobian</code></a></strong><code>(reals y, reals offset, reals mult)</code><br> Takes a value <code>y</code>, shift <code>offset</code>, and scale <code>mult</code> and returns a rescaled and shifted value.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-6ed536361121ffb486071705925b0107a779e401" class="index-entry"></a> <!-- reals; offset_multiplier_unconstrain; (reals x, reals offset, reals mult); -->
+<p></p>
+<p><code>reals</code> <strong><a href="../functions-reference/functions_index.html#offset_multiplier_unconstrain" class="unlink" title="Jump to index entry"><code>offset_multiplier_unconstrain</code></a></strong><code>(reals x, reals offset, reals mult)</code><br> Takes a value <code>x</code>, shift <code>offset</code>, and scale <code>mult</code> and a value which has been un-scaled and un-shifted.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+</section>
+<section id="transforms-for-constrained-vectors" class="level2">
+<h2 class="anchored" data-anchor-id="transforms-for-constrained-vectors">Transforms for constrained vectors</h2>
+<p>These functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.</p>
+<p>Where <code>vectors</code> is used, this indicates that either a <code>vector</code> or a (possibly multidimensional) array of <code>vector</code>s may be provided. The array will be processed element by element.</p>
+<section id="ordered-vectors" class="level3">
+<h3 class="anchored" data-anchor-id="ordered-vectors">Ordered vectors</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#ordered-vector">Ordered Vector section</a>.</p>
+<a id="index-entry-ef08c8144bb3789c27d35f7a7c30cd57f8a9a573" class="index-entry"></a> <!-- vectors; ordered_constrain; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#ordered_constrain" class="unlink" title="Jump to index entry"><code>ordered_constrain</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector with elements in ascending order.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-6372fc8ee07f534b79aa45d152097a421af0a938" class="index-entry"></a> <!-- vectors; ordered_jacobian; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#ordered_jacobian" class="unlink" title="Jump to index entry"><code>ordered_jacobian</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector with elements in ascending order.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-66bebbbf4a16907781a03d7910e05157b57546ae" class="index-entry"></a> <!-- vectors; ordered_unconstrain; (vectors x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#ordered_unconstrain" class="unlink" title="Jump to index entry"><code>ordered_unconstrain</code></a></strong><code>(vectors x)</code><br> Takes an ordered vector <code>x</code> and returns the corresponding free vector.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="positive-order-vectors" class="level3">
+<h3 class="anchored" data-anchor-id="positive-order-vectors">Positive order vectors</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#positive-ordered-vector">Positive Ordered Vector section</a>.</p>
+<a id="index-entry-a5bd3aeff6187e7710c5c9aede0aa53b0bf05605" class="index-entry"></a> <!-- vectors; positive_ordered_constrain; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#positive_ordered_constrain" class="unlink" title="Jump to index entry"><code>positive_ordered_constrain</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector with positive elements in ascending order.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-72ae38a185968521d676ce568e4dd9dfb0ec3211" class="index-entry"></a> <!-- vectors; positive_ordered_jacobian; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#positive_ordered_jacobian" class="unlink" title="Jump to index entry"><code>positive_ordered_jacobian</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector with positive elements in ascending order.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-ea0b4a03eff5caa57b2d154df97f2c8a63b820e5" class="index-entry"></a> <!-- vectors; positive_ordered_unconstrain; (vectors x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#positive_ordered_unconstrain" class="unlink" title="Jump to index entry"><code>positive_ordered_unconstrain</code></a></strong><code>(vectors x)</code><br> Takes an ordered vector <code>x</code> with positive entries and returns the corresponding free vector.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="simplexes" class="level3">
+<h3 class="anchored" data-anchor-id="simplexes">Simplexes</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#simplex-transform.section">Unit Simplex section</a>.</p>
+<a id="index-entry-055048ff4bc60482699857e7993258efc99d45f0" class="index-entry"></a> <!-- vectors; simplex_constrain; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#simplex_constrain" class="unlink" title="Jump to index entry"><code>simplex_constrain</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a simplex (a vector such that each element is between <code>0</code> and <code>1</code>, and the sum of the elements is <code>1</code>, up to rounding errors).</p>
+<p>This returned vector will have one extra element compared to the input <code>y</code>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-446c0a0a855c1f82fe5f0df51dddf43419abd15f" class="index-entry"></a> <!-- vectors; simplex_jacobian; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#simplex_jacobian" class="unlink" title="Jump to index entry"><code>simplex_jacobian</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a simplex (a vector such that each element is between <code>0</code> and <code>1</code>, and the sum of the elements is <code>1</code>, up to rounding errors).</p>
+<p>This returned vector will have one extra element compared to the input <code>y</code>.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-ffa2ad29148676e3139f6d81a3fba15199481f30" class="index-entry"></a> <!-- vectors; simplex_unconstrain; (vectors x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#simplex_unconstrain" class="unlink" title="Jump to index entry"><code>simplex_unconstrain</code></a></strong><code>(vectors x)</code><br> Takes a simplex <code>x</code> and returns the corresponding free vector.</p>
+<p>This returned vector will have one fewer elements compared to the input <code>x</code>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="sum-to-zero-vectors" class="level3">
+<h3 class="anchored" data-anchor-id="sum-to-zero-vectors">Sum-to-zero vectors</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#sum-to-zero-vector">Zero Sum Vector section</a>.</p>
+<a id="index-entry-6c0d205ee6463bcaa0673c8d9607d2ca20bf2860" class="index-entry"></a> <!-- vectors; sum_to_zero_constrain; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#sum_to_zero_constrain" class="unlink" title="Jump to index entry"><code>sum_to_zero_constrain</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector such that the elements sum to <code>0</code>.</p>
+<p>This returned vector will have one extra element compared to the input <code>y</code>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-aeaf42b6a6102bf306ab44b353d64e75c52f5867" class="index-entry"></a> <!-- vectors; sum_to_zero_jacobian; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#sum_to_zero_jacobian" class="unlink" title="Jump to index entry"><code>sum_to_zero_jacobian</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector such that the elements sum to <code>0</code>.</p>
+<p>The returned vector will have one extra element compared to the input <code>y</code>.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-43bd9130545904d1b8cea836c9621a69cd06f361" class="index-entry"></a> <!-- vectors; sum_to_zero_unconstrain; (vectors x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#sum_to_zero_unconstrain" class="unlink" title="Jump to index entry"><code>sum_to_zero_unconstrain</code></a></strong><code>(vectors x)</code><br> Takes a vector <code>x</code> with elements that sum to <code>0</code> and returns the corresponding free vector.</p>
+<p>This returned vector will have one fewer elements compared to the input <code>x</code>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="unit-vectors" class="level3">
+<h3 class="anchored" data-anchor-id="unit-vectors">Unit vectors</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#unit-vector.section">Unit Vector section</a>.</p>
+<a id="index-entry-3e7ebedc4498436d0a2336970080bf855c52eaa7" class="index-entry"></a> <!-- vectors; unit_vectors_constrain; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#unit_vectors_constrain" class="unlink" title="Jump to index entry"><code>unit_vectors_constrain</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector with unit length, i.e., <code>norm2(unit_vectors_constrain(y)) == 1</code> for any <code>y</code> that has a positive and finite norm itself (if <code>y</code> does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-dad75a5d7cfa2f707b43c61f4208876d4f1ff189" class="index-entry"></a> <!-- vectors; unit_vectors_jacobian; (vectors y); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#unit_vectors_jacobian" class="unlink" title="Jump to index entry"><code>unit_vectors_jacobian</code></a></strong><code>(vectors y)</code><br> Takes a free vector <code>y</code> and returns a vector with unit length. This function rejects if given a vector of all zeros.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-8030393eb30e7d856eced92680f3c9e8454c19d2" class="index-entry"></a> <!-- vectors; unit_vectors_unconstrain; (vectors x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#unit_vectors_unconstrain" class="unlink" title="Jump to index entry"><code>unit_vectors_unconstrain</code></a></strong><code>(vectors x)</code><br> Takes a vector <code>x</code> of unit length and returns the corresponding free vector.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+</section>
+<section id="transforms-for-constrained-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="transforms-for-constrained-matrices">Transforms for constrained matrices</h2>
+<p>Similarly to the above, <code>vectors</code> means a <code>vector</code> or array thereof, and <code>matrices</code> means a <code>matrix</code> or array thereof.</p>
+<section id="cholesky-factors-of-correlation-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="cholesky-factors-of-correlation-matrices">Cholesky factors of correlation matrices</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#cholesky-factors-of-correlation-matrices">Cholesky Factors of Correlation Matrices section</a>.</p>
+<a id="index-entry-13178bf49e47f4d98c973800329f95f8fd49bfa1" class="index-entry"></a> <!-- matrices; cholesky_factor_corr_constrain; (vectors y, int K); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#cholesky_factor_corr_constrain" class="unlink" title="Jump to index entry"><code>cholesky_factor_corr_constrain</code></a></strong><code>(vectors y, int K)</code><br> Takes a vector <code>y</code> and integer <code>K</code>, where <code>length(y) == choose(K, 2)</code>, and returns a <code>K</code> by <code>K</code> Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-1e720292cb09512b6fdca84e4cc39a087f43ec0a" class="index-entry"></a> <!-- matrices; cholesky_factor_corr_jacobian; (vectors y, int K); -->
+<p> Takes a vector <code>y</code> and integer <code>K</code>, where <code>length(y) == choose(K, 2)</code>, and returns a <code>K</code> by <code>K</code> Cholesky factor of a correlation matrix.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment. <code>matrices</code> <strong><a href="../functions-reference/functions_index.html#cholesky_factor_corr_jacobian" class="unlink" title="Jump to index entry"><code>cholesky_factor_corr_jacobian</code></a></strong><code>(vectors y, int K)</code><br></p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-98b8008a7ecf84ed0131fe5e84c2a239b3739d30" class="index-entry"></a> <!-- vectors; cholesky_factor_corr_unconstrain; (matrices x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#cholesky_factor_corr_unconstrain" class="unlink" title="Jump to index entry"><code>cholesky_factor_corr_unconstrain</code></a></strong><code>(matrices x)</code><br> Takes <code>x</code>, a (<span class="math inline">\(K \times K\)</span>) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="cholesky-factors-of-covariance-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="cholesky-factors-of-covariance-matrices">Cholesky factors of covariance matrices</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#cholesky-factors-of-covariance-matrices">Cholesky Factors of Covariance Matrices section</a>.</p>
+<a id="index-entry-0cddde7e0aa5cc9213f393acebb3b13b68b7bfa8" class="index-entry"></a> <!-- matrices; cholesky_factor_cov_constrain; (vectors y, int M, int N); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#cholesky_factor_cov_constrain" class="unlink" title="Jump to index entry"><code>cholesky_factor_cov_constrain</code></a></strong><code>(vectors y, int M, int N)</code><br> Takes a free vector <code>y</code> and integers <code>M</code> and <code>N</code> and returns the <code>M</code> by <code>N</code> Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix <span class="math inline">\(L\)</span>, with a strictly positive diagonal, such that <span class="math inline">\(L^T L\)</span> is positive definite.</p>
+<p>Note that <code>y</code> must have length <code>N + choose(N, 2) + (M - N) * N</code>, and <code>M</code> must be greater than or equal to <code>N</code>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-fd1e2249d8d133c4bf9d8e6b44f67c9c69314da3" class="index-entry"></a> <!-- matrices; cholesky_factor_cov_jacobian; (vectors y, int M, int N); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#cholesky_factor_cov_jacobian" class="unlink" title="Jump to index entry"><code>cholesky_factor_cov_jacobian</code></a></strong><code>(vectors y, int M, int N)</code><br> Takes a free vector <code>y</code> and integers <code>M</code> and <code>N</code> and returns the <code>M</code> by <code>N</code> Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix <span class="math inline">\(L\)</span>, with a strictly positive diagonal, such that <span class="math inline">\(L^T L\)</span> is positive definite.</p>
+<p>Note that <code>y</code> must have length <code>N + choose(N, 2) + (M - N) * N</code>, and <code>M</code> must be greater than or equal to <code>N</code>.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-74c3b28f5b00bf51f6299872cce7f2b1a18bf3ac" class="index-entry"></a> <!-- vectors; cholesky_factor_cov_unconstrain; (matrices x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#cholesky_factor_cov_unconstrain" class="unlink" title="Jump to index entry"><code>cholesky_factor_cov_unconstrain</code></a></strong><code>(matrices x)</code><br> Takes a <span class="math inline">\(M \times N\)</span> matrix <code>x</code> which is a Cholesky factor of a covariance matrix (a matrix <span class="math inline">\(L\)</span> such that <span class="math inline">\(L^T L\)</span> is positive definite) and returns the corresponding free vector of length <span class="math inline">\(N + \binom{N}{2} + (M - N)N\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="correlation-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="correlation-matrices">Correlation matrices</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#correlation-matrix-transform.section">Correlation Matrices section</a>.</p>
+<a id="index-entry-594161bab69b742477ec84f4a399816cc0d939a1" class="index-entry"></a> <!-- matrices; corr_matrix_constrain; (vectors y, int K); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#corr_matrix_constrain" class="unlink" title="Jump to index entry"><code>corr_matrix_constrain</code></a></strong><code>(vectors y, int K)</code><br> Takes a vector <code>y</code> and integer <code>K</code>, where <code>length(y) == choose(K, 2)</code>, and returns a <code>K</code> by <code>K</code> correlation matrix (a positive definite matrix with a unit diagonal).</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-a3ce311c60e3c6821b434df3f73f83c5e037a03b" class="index-entry"></a> <!-- matrices; corr_matrix_jacobian; (vectors y, int K); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#corr_matrix_jacobian" class="unlink" title="Jump to index entry"><code>corr_matrix_jacobian</code></a></strong><code>(vectors y, int K)</code><br> Takes a vector <code>y</code> and integer <code>K</code>, where <code>length(y) == choose(K, 2)</code>, and returns a <code>K</code> by <code>K</code> correlation matrix (a positive definite matrix with a unit diagonal).</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-0fcba057e29e1ae091121cb62708675880f50c39" class="index-entry"></a> <!-- vectors; corr_matrix_unconstrain; (matrices x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#corr_matrix_unconstrain" class="unlink" title="Jump to index entry"><code>corr_matrix_unconstrain</code></a></strong><code>(matrices x)</code><br> Takes a <span class="math inline">\(K \times K\)</span> matrix <code>x</code> which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size <span class="math inline">\(\binom{K}{2}\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="covariance-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="covariance-matrices">Covariance matrices</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#covariance-matrices">Covariance Matrices section</a>.</p>
+<a id="index-entry-6dc98b07eda2e6dfde0020cf98dae68b590aad28" class="index-entry"></a> <!-- matrices; cov_matrix_constrain; (vectors y, int K); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#cov_matrix_constrain" class="unlink" title="Jump to index entry"><code>cov_matrix_constrain</code></a></strong><code>(vectors y, int K)</code><br> Takes a vector <code>y</code> and integer <code>K</code>, where <code>length(y) == K + choose(K, 2)</code>, and returns a <code>K</code> by <code>K</code> covariance matrix (a positive definite matrix).</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-cad2be5fc2ee16a8419b70db0422629b843de6c3" class="index-entry"></a> <!-- matrices; cov_matrix_jacobian; (vectors y, int K); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#cov_matrix_jacobian" class="unlink" title="Jump to index entry"><code>cov_matrix_jacobian</code></a></strong><code>(vectors y, int K)</code><br> Takes a vector <code>y</code> and integer <code>K</code>, where <code>length(y) == K + choose(K, 2)</code>, and returns a <code>K</code> by <code>K</code> covariance matrix (a positive definite matrix).</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-00d19b2a52626d58b6b07f2d27fdd2dbad3c12c0" class="index-entry"></a> <!-- vectors; cov_matrix_unconstrain; (matrices x); -->
+<p></p>
+<p><code>vectors</code> <strong><a href="../functions-reference/functions_index.html#cov_matrix_unconstrain" class="unlink" title="Jump to index entry"><code>cov_matrix_unconstrain</code></a></strong><code>(matrices x)</code><br> Takes a <span class="math inline">\(K \times K\)</span> positive definite matrix <code>x</code> and returns the corresponding free vector of size <span class="math inline">\(K + \binom{K}{2}\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="column-stochastic-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="column-stochastic-matrices">Column-stochastic matrices</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#stochastic-matrix-transform.section">Stochastic Matrix section</a> for column (left) stochastic matrices.</p>
+<a id="index-entry-4dbd1cdc02cc83f417b72d0ce15257683589e66c" class="index-entry"></a> <!-- matrices; stochastic_column_constrain; (matrices y); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#stochastic_column_constrain" class="unlink" title="Jump to index entry"><code>stochastic_column_constrain</code></a></strong><code>(matrices y)</code><br> Takes a free matrix <code>y</code> of size <span class="math inline">\(N \times M\)</span> and returns a left stochastic matrix (a matrix where each column is a simplex) of size <span class="math inline">\(N+1 \times M\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-a21e88d79c0bb4cc8424754bfac7f0bfada0be42" class="index-entry"></a> <!-- matrices; stochastic_column_jacobian; (matrices y); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#stochastic_column_jacobian" class="unlink" title="Jump to index entry"><code>stochastic_column_jacobian</code></a></strong><code>(matrices y)</code><br> Takes a free matrix <code>y</code> of size <span class="math inline">\(N \times M\)</span> and returns a left stochastic matrix (a matrix where each column is a simplex) of size <span class="math inline">\(N+1 \times M\)</span>.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-a04a2804c70fb0fe2219a7d11244837772f7d404" class="index-entry"></a> <!-- matrices; stochastic_column_unconstrain; (matrices x); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#stochastic_column_unconstrain" class="unlink" title="Jump to index entry"><code>stochastic_column_unconstrain</code></a></strong><code>(matrices x)</code><br> Takes a left stochastic matrix <code>x</code> of size <span class="math inline">\(N+1 \times M\)</span> and returns the corresponding free matrix of size <span class="math inline">\(N \times M\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="row-stochastic-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="row-stochastic-matrices">Row-stochastic matrices</h3>
+<p>These functions perform the transform and inverse transform described in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#stochastic-matrix-transform.section">Stochastic Matrix section</a> for row (right) stochastic matrices.</p>
+<a id="index-entry-e2ce12f9782fbeb39c70aa822e3b8a5d4ef4baf8" class="index-entry"></a> <!-- matrices; stochastic_row_constrain; (matrices y); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#stochastic_row_constrain" class="unlink" title="Jump to index entry"><code>stochastic_row_constrain</code></a></strong><code>(matrices y)</code><br> Takes a free matrix <code>y</code> of size <span class="math inline">\(N \times M\)</span> and returns a right stochastic matrix (a matrix where each row is a simplex) of size <span class="math inline">\(N \times M+1\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-b4c7ad92b82702ad1209da8fa540ff52110b0bf0" class="index-entry"></a> <!-- matrices; stochastic_row_jacobian; (matrices y); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#stochastic_row_jacobian" class="unlink" title="Jump to index entry"><code>stochastic_row_jacobian</code></a></strong><code>(matrices y)</code><br> Takes a free matrix <code>y</code> of size <span class="math inline">\(N \times M\)</span> and returns a right stochastic matrix (a matrix where each row is a simplex) of size <span class="math inline">\(N \times M+1\)</span>.</p>
+<p>This function also increments the Jacobian accumulator with the corresponding change of variables adjustment.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-4bf40767700b7e2fc07e1045d565df3dc93422ee" class="index-entry"></a> <!-- matrices; stochastic_row_unconstrain; (matrices x); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#stochastic_row_unconstrain" class="unlink" title="Jump to index entry"><code>stochastic_row_unconstrain</code></a></strong><code>(matrices x)</code><br> Takes a right stochastic matrix <code>x</code> of size <span class="math inline">\(N \times M+1\)</span> and returns the corresponding free matrix of size <span class="math inline">\(N \times M\)</span>.</p>
+<small><i>Available since 2.37</i></small>
+</section>
+<section id="sum-to-zero-matrices" class="level3">
+<h3 class="anchored" data-anchor-id="sum-to-zero-matrices">Sum-to-zero matrices</h3>
+<p>The sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the <a href="https://mc-stan.org/docs/reference-manual/transforms.html#sum-to-zero-matrix-transform">sum-to-zero matrix section of the <em>Reference Manual</em></a>.</p>
+<a id="index-entry-1c9a600c3c9e18bf8b358b6e0e1a196de3adb95d" class="index-entry"></a> <!-- matrices; sum_to_zero_constrain; (matrices y); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#sum_to_zero_constrain" class="unlink" title="Jump to index entry"><code>sum_to_zero_constrain</code></a></strong><code>(matrices y)</code><br> The constraining function maps an unconstrained <code>N x M</code> matrix to an <code>(N + 1) x (M + 1)</code> matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.</p>
+<p>This returned matrix will have one extra row and column compared to the input <code>y</code>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-5fbdab6082a06ae172bb0aa3d1c27d00d6acce40" class="index-entry"></a> <!-- matrices; sum_to_zero_jacobian; (matrices y); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#sum_to_zero_jacobian" class="unlink" title="Jump to index entry"><code>sum_to_zero_jacobian</code></a></strong><code>(matrices y)</code><br> The constraining function maps an unconstrained <code>N x M</code> matrix to an <code>(N + 1) x (M + 1)</code> matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to <code>sum_to_zero_constrain</code>.</p>
+<p>This returned matrix will have one extra row and column compared to the input <code>y</code>.</p>
+<small><i>Available since 2.37</i></small>
+<a id="index-entry-05440d06c001dcf3039cebf913897b51b2ee23e5" class="index-entry"></a> <!-- matrices; sum_to_zero_unconstrain; (matrices x); -->
+<p></p>
+<p><code>matrices</code> <strong><a href="../functions-reference/functions_index.html#sum_to_zero_unconstrain" class="unlink" title="Jump to index entry"><code>sum_to_zero_unconstrain</code></a></strong><code>(matrices x)</code><br> This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.</p>
+<small><i>Available since 2.37</i></small>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/higher-order_functions.html" class="pagination-link" aria-label="Higher-Order Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Higher-Order Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/deprecated_functions.html" class="pagination-link" aria-label="Deprecated Functions">
+        <span class="nav-page-text">Deprecated Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/transform_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/unbounded_continuous_distributions.html b/docs/2_39/functions-reference/unbounded_continuous_distributions.html
new file mode 100644
index 000000000..807dfed4c
--- /dev/null
+++ b/docs/2_39/functions-reference/unbounded_continuous_distributions.html
@@ -0,0 +1,1805 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Unbounded Continuous Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/positive_continuous_distributions.html" rel="next">
+<link href="../functions-reference/multivariate_discrete_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Unbounded Continuous Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Unbounded Continuous Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#unbounded-continuous-distributions" id="toc-unbounded-continuous-distributions" class="nav-link active" data-scroll-target="#unbounded-continuous-distributions">Unbounded Continuous Distributions</a>
+  <ul class="collapse">
+  <li><a href="#normal-distribution" id="toc-normal-distribution" class="nav-link" data-scroll-target="#normal-distribution">Normal distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function" id="toc-probability-density-function" class="nav-link" data-scroll-target="#probability-density-function">Probability density function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  <li><a href="#standard-normal-distribution" id="toc-standard-normal-distribution" class="nav-link" data-scroll-target="#standard-normal-distribution">Standard normal distribution</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#normal-id-glm" id="toc-normal-id-glm" class="nav-link" data-scroll-target="#normal-id-glm">Normal-id generalized linear model (linear regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-distribution-function" id="toc-probability-distribution-function" class="nav-link" data-scroll-target="#probability-distribution-function">Probability distribution function</a></li>
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#exponentially-modified-normal-distribution" id="toc-exponentially-modified-normal-distribution" class="nav-link" data-scroll-target="#exponentially-modified-normal-distribution">Exponentially modified normal distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-1" id="toc-probability-density-function-1" class="nav-link" data-scroll-target="#probability-density-function-1">Probability density function</a></li>
+  <li><a href="#distribution-statement-3" id="toc-distribution-statement-3" class="nav-link" data-scroll-target="#distribution-statement-3">Distribution statement</a></li>
+  <li><a href="#stan-functions-3" id="toc-stan-functions-3" class="nav-link" data-scroll-target="#stan-functions-3">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#skew-normal-distribution" id="toc-skew-normal-distribution" class="nav-link" data-scroll-target="#skew-normal-distribution">Skew normal distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-2" id="toc-probability-density-function-2" class="nav-link" data-scroll-target="#probability-density-function-2">Probability density function</a></li>
+  <li><a href="#distribution-statement-4" id="toc-distribution-statement-4" class="nav-link" data-scroll-target="#distribution-statement-4">Distribution statement</a></li>
+  <li><a href="#stan-functions-4" id="toc-stan-functions-4" class="nav-link" data-scroll-target="#stan-functions-4">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#student-t-distribution" id="toc-student-t-distribution" class="nav-link" data-scroll-target="#student-t-distribution">Student-t distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-3" id="toc-probability-density-function-3" class="nav-link" data-scroll-target="#probability-density-function-3">Probability density function</a></li>
+  <li><a href="#distribution-statement-5" id="toc-distribution-statement-5" class="nav-link" data-scroll-target="#distribution-statement-5">Distribution statement</a></li>
+  <li><a href="#stan-functions-5" id="toc-stan-functions-5" class="nav-link" data-scroll-target="#stan-functions-5">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#cauchy-distribution" id="toc-cauchy-distribution" class="nav-link" data-scroll-target="#cauchy-distribution">Cauchy distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-4" id="toc-probability-density-function-4" class="nav-link" data-scroll-target="#probability-density-function-4">Probability density function</a></li>
+  <li><a href="#distribution-statement-6" id="toc-distribution-statement-6" class="nav-link" data-scroll-target="#distribution-statement-6">Distribution statement</a></li>
+  <li><a href="#stan-functions-6" id="toc-stan-functions-6" class="nav-link" data-scroll-target="#stan-functions-6">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#double-exponential-laplace-distribution" id="toc-double-exponential-laplace-distribution" class="nav-link" data-scroll-target="#double-exponential-laplace-distribution">Double exponential (Laplace) distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-5" id="toc-probability-density-function-5" class="nav-link" data-scroll-target="#probability-density-function-5">Probability density function</a></li>
+  <li><a href="#distribution-statement-7" id="toc-distribution-statement-7" class="nav-link" data-scroll-target="#distribution-statement-7">Distribution statement</a></li>
+  <li><a href="#stan-functions-7" id="toc-stan-functions-7" class="nav-link" data-scroll-target="#stan-functions-7">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#logistic-distribution" id="toc-logistic-distribution" class="nav-link" data-scroll-target="#logistic-distribution">Logistic distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-6" id="toc-probability-density-function-6" class="nav-link" data-scroll-target="#probability-density-function-6">Probability density function</a></li>
+  <li><a href="#distribution-statement-8" id="toc-distribution-statement-8" class="nav-link" data-scroll-target="#distribution-statement-8">Distribution statement</a></li>
+  <li><a href="#stan-functions-8" id="toc-stan-functions-8" class="nav-link" data-scroll-target="#stan-functions-8">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#gumbel-distribution" id="toc-gumbel-distribution" class="nav-link" data-scroll-target="#gumbel-distribution">Gumbel distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-7" id="toc-probability-density-function-7" class="nav-link" data-scroll-target="#probability-density-function-7">Probability density function</a></li>
+  <li><a href="#distribution-statement-9" id="toc-distribution-statement-9" class="nav-link" data-scroll-target="#distribution-statement-9">Distribution statement</a></li>
+  <li><a href="#stan-functions-9" id="toc-stan-functions-9" class="nav-link" data-scroll-target="#stan-functions-9">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#skew-double-exponential-distribution" id="toc-skew-double-exponential-distribution" class="nav-link" data-scroll-target="#skew-double-exponential-distribution">Skew double exponential distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-density-function-8" id="toc-probability-density-function-8" class="nav-link" data-scroll-target="#probability-density-function-8">Probability density function</a></li>
+  <li><a href="#distribution-statement-10" id="toc-distribution-statement-10" class="nav-link" data-scroll-target="#distribution-statement-10">Distribution statement</a></li>
+  <li><a href="#stan-functions-10" id="toc-stan-functions-10" class="nav-link" data-scroll-target="#stan-functions-10">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Continuous Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/unbounded_continuous_distributions.html">Unbounded Continuous Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="unbounded-continuous-distributions" class="level1">
+<h1>Unbounded Continuous Distributions</h1>
+<p>The unbounded univariate continuous probability distributions have support on all real numbers.</p>
+<section id="normal-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="normal-distribution">Normal distribution</h2>
+<section id="probability-density-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in
+\mathbb{R}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Normal}(y|\mu,\sigma) =
+\frac{1}{\sqrt{2 \pi} \ \sigma} \exp\left( - \, \frac{1}{2} \left(\frac{y - \mu}{\sigma} \right)^2 \right) \!.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#normal" class="unlink" title="Jump to index entry"><code>normal</code></a></strong><code>(mu, sigma)</code></p>
+<p>Increment target log probability density with <code>normal_lupdf(y | mu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-72458a5e0e1a905dc04f11d1b05e4136543f6142" class="index-entry"></a> <!-- real; normal ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-7fb57a9b998de709bd4e5f0f69713d26e2ee62ae" class="index-entry"></a> <!-- real; normal_lpdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_lpdf" class="unlink" title="Jump to index entry"><code>normal_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the normal density of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-429c956476257b277125a0f2bddd560aaa7416a6" class="index-entry"></a> <!-- real; normal_lupdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_lupdf" class="unlink" title="Jump to index entry"><code>normal_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the normal density of y given location mu and scale sigma dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-03470cc37c6a6c27433e9854162ed863ac635e68" class="index-entry"></a> <!-- real; normal_cdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_cdf" class="unlink" title="Jump to index entry"><code>normal_cdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for <span class="math inline">\(\frac{{y}-{\mu}}{{\sigma}}\)</span> below -37.5 and overflow to 1 for <span class="math inline">\(\frac{{y}-{\mu}}{{\sigma}}\)</span> above 8.25; the function <code>Phi_approx</code> is more robust in the tails, but must be scaled and translated for anything other than a standard normal.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-0c8b83888fa71a0e469f8d91bc4d13c2ee3adb7a" class="index-entry"></a> <!-- real; normal_lcdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_lcdf" class="unlink" title="Jump to index entry"><code>normal_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to <span class="math inline">\(-\infty\)</span> for <span class="math inline">\(\frac{{y}-{\mu}}{{\sigma}}\)</span> below -37.5 and overflow to 0 for <span class="math inline">\(\frac{{y}-{\mu}}{{\sigma}}\)</span> above 8.25; <code>log(Phi_approx(...))</code> is more robust in the tails, but must be scaled and translated for anything other than a standard normal.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-0e0917cae8b192a1adca7f9d708d48f291d953c0" class="index-entry"></a> <!-- real; normal_lccdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_lccdf" class="unlink" title="Jump to index entry"><code>normal_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for <span class="math inline">\(\frac{{y}-{\mu}}{{\sigma}}\)</span> below -37.5 and underflow to <span class="math inline">\(-\infty\)</span> for <span class="math inline">\(\frac{{y}-{\mu}}{{\sigma}}\)</span> above 8.25; <code>log1m(Phi_approx(...))</code> is more robust in the tails, but must be scaled and translated for anything other than a standard normal.</p>
+<small><i>Available since 2.15</i></small>
+<a id="index-entry-a2ef357923cb90482b3d3a4e7a66b0a04b56f3fe" class="index-entry"></a> <!-- R; normal_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#normal_rng" class="unlink" title="Jump to index entry"><code>normal_rng</code></a></strong><code>(reals mu, reals sigma)</code><br> Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+<section id="standard-normal-distribution" class="level3">
+<h3 class="anchored" data-anchor-id="standard-normal-distribution">Standard normal distribution</h3>
+<p>The standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. <span class="math display">\[\begin{equation*}
+\text{StdNormal}(y) \ = \ \text{Normal}(y \mid 0, 1) \ = \ \frac{1}{\sqrt{2 \pi}} \, \exp \left( \frac{-y^2}{2} \right)\!.
+\end{equation*}\]</span> Up to a proportion on the log scale, where Stan computes, <span class="math display">\[\begin{equation*}
+\log \text{Normal}(y \mid 0, 1) \ = \ \frac{-y^2}{2} + \text{const}.
+\end{equation*}\]</span> With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location <span class="math inline">\(0\)</span> and scale <span class="math inline">\(1\)</span>.</p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#std_normal" class="unlink" title="Jump to index entry"><code>std_normal</code></a></strong><code>()</code></p>
+<p>Increment target log probability density with <code>std_normal_lupdf(y)</code>.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-8fd9b26fa3b56249d7987f7189e234de82c38812" class="index-entry"></a> <!-- real; std_normal ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-3ae6425bfbb49379f4d0565899a18873e6b23747" class="index-entry"></a> <!-- real; std_normal_lpdf; (reals y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#std_normal_lpdf" class="unlink" title="Jump to index entry"><code>std_normal_lpdf</code></a></strong><code>(reals y)</code><br> The standard normal (location zero, scale one) log probability density of y.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-813f27102ca53e25d42a3ef9c42bd919f4390d6d" class="index-entry"></a> <!-- real; std_normal_lupdf; (reals y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#std_normal_lupdf" class="unlink" title="Jump to index entry"><code>std_normal_lupdf</code></a></strong><code>(reals y)</code><br> The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-d682f4986b7887882cc42788a87dd1d67950e7cc" class="index-entry"></a> <!-- real; std_normal_cdf; (reals y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#std_normal_cdf" class="unlink" title="Jump to index entry"><code>std_normal_cdf</code></a></strong><code>(reals y)</code><br> The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for <span class="math inline">\(y\)</span> below -37.5 and overflow to 1 for <span class="math inline">\(y\)</span> above 8.25; the function <code>Phi_approx</code> is more robust in the tails.</p>
+<small><i>Available since 2.21</i></small>
+<a id="index-entry-925bd18ef58df9b2bf3bf0dc86898c8633dc2373" class="index-entry"></a> <!-- real; std_normal_lcdf; (reals y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#std_normal_lcdf" class="unlink" title="Jump to index entry"><code>std_normal_lcdf</code></a></strong><code>(reals y)</code><br> The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to <span class="math inline">\(-\infty\)</span> for <span class="math inline">\(y\)</span> below -37.5 and overflow to 0 for <span class="math inline">\(y\)</span> above 8.25; <code>log(Phi_approx(...))</code> is more robust in the tails.</p>
+<small><i>Available since 2.21</i></small>
+<a id="index-entry-bf6a81086184bebf728088395f30c5755cd8562a" class="index-entry"></a> <!-- real; std_normal_lccdf; (reals y); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#std_normal_lccdf" class="unlink" title="Jump to index entry"><code>std_normal_lccdf</code></a></strong><code>(reals y)</code><br> The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for <span class="math inline">\(y\)</span> below -37.5 and underflow to <span class="math inline">\(-\infty\)</span> for <span class="math inline">\(y\)</span> above 8.25; <code>log1m(Phi_approx(...))</code> is more robust in the tails.</p>
+<small><i>Available since 2.21</i></small>
+<a id="index-entry-9bac63544b561d2822dd4d977b4e0ce0b78f183b" class="index-entry"></a> <!-- R; std_normal_qf; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#std_normal_qf" class="unlink" title="Jump to index entry"><code>std_normal_qf</code></a></strong><code>(T x)</code><br> Returns the value of the inverse standard normal cdf <span class="math inline">\(\Phi^{-1}\)</span> at the specified quantile <code>x</code>. The <code>std_normal_qf</code> is equivalent to the <code>inv_Phi</code> function.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-fe25ac0f35036372c19c2da7bb28d37f9a80cf2c" class="index-entry"></a> <!-- R; std_normal_log_qf; (T x); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#std_normal_log_qf" class="unlink" title="Jump to index entry"><code>std_normal_log_qf</code></a></strong><code>(T x)</code><br> Return the value of the inverse standard normal cdf <span class="math inline">\(\Phi^{-1}\)</span> evaluated at the log of the specified quantile <code>x</code>. This function is equivalent to <code>std_normal_qf(exp(x))</code> but is more numerically stable.</p>
+<small><i>Available since 2.31</i></small>
+<a id="index-entry-4d08a0371e8f4ad5a8ac20add31783b3a31ed310" class="index-entry"></a> <!-- real; std_normal_rng; (); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#std_normal_rng" class="unlink" title="Jump to index entry"><code>std_normal_rng</code></a></strong><code>()</code><br> Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.</p>
+<small><i>Available since 2.21</i></small>
+</section>
+</section>
+<section id="normal-id-glm" class="level2">
+<h2 class="anchored" data-anchor-id="normal-id-glm">Normal-id generalized linear model (linear regression)</h2>
+<p>Stan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e.&nbsp;a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.</p>
+<section id="probability-distribution-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-distribution-function">Probability distribution function</h3>
+<p>If <span class="math inline">\(x\in \mathbb{R}^{n\cdot m}, \alpha \in \mathbb{R}^n, \beta\in
+\mathbb{R}^m, \sigma\in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}^n\)</span>, <span class="math display">\[\begin{equation*} \text{NormalIdGLM}(y|x, \alpha, \beta, \sigma) = \prod_{1\leq i
+\leq n}\text{Normal}(y_i|\alpha_i + x_i\cdot \beta, \sigma). \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm" class="unlink" title="Jump to index entry"><code>normal_id_glm</code></a></strong><code>(x, alpha, beta, sigma)</code></p>
+<p>Increment target log probability density with <code>normal_id_glm_lupdf(y | x, alpha, beta, sigma)</code>.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-f2e910491c0d85196d2d372dcaffcc4cc22572ce" class="index-entry"></a> <!-- real; normal_id_glm ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-4436deedaa4adcaf697328f4ab6228f12bf213be" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (real y | matrix x, real alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(real y | matrix x, real alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-ef0f44ddfb9e65d97bec95df9bc84e88a5262ebc" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (real y | matrix x, real alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(real y | matrix x, real alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-4fc2715c69e219915d6372ded62c513ec2dad353" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (real y | matrix x, vector alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(real y | matrix x, vector alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-baaf096f4f58b53d3142ade26076839f36733a39" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (real y | matrix x, vector alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(real y | matrix x, vector alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-90d8f9b890b2588b36ed285cfb031e89bf170935" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (real y | matrix x, real alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(real y | matrix x, real alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-2604f4a479cd2758a8273ebc888a4f35d707bcc3" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (real y | matrix x, real alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(real y | matrix x, real alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-7991a0ad137f72771bee6ae8bfe2a09dba063c5a" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (real y | matrix x, vector alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(real y | matrix x, vector alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-d7b73dfd16de89ad209b2a04905cdae1169f761c" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (real y | matrix x, vector alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(real y | matrix x, vector alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-1d3d0f48ac6b8ae87eab8640482747cc1d3bcc65" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (vector y | row_vector x, real alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(vector y | row_vector x, real alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-1aefc00a65f6adeb0cba0329b66f968d6caf1427" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (vector y | row_vector x, real alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(vector y | row_vector x, real alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-d67cdb934c8fdbe441e7968e31d3d42bdc6bb46f" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (vector y | row_vector x, vector alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(vector y | row_vector x, vector alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-29d66019b47c2be1df4e8cb54f85a3459f41d7be" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (vector y | row_vector x, vector alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(vector y | row_vector x, vector alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.29</i></small>
+<a id="index-entry-5bce6923d1e27b6dfdb1112d0e26e6f2fccf3704" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (vector y | matrix x, real alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(vector y | matrix x, real alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-b410b1737bb33ec4400eae94d05bb37faea0c0e9" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (vector y | matrix x, real alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(vector y | matrix x, real alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-c8b35bcd76ae895df16171156e9ddc2b50977858" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (vector y | matrix x, vector alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(vector y | matrix x, vector alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-348b5a5c6bc0a9340114bd7744603300d84580bb" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (vector y | matrix x, vector alpha, vector beta, real sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(vector y | matrix x, vector alpha, vector beta, real sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-1428e7cc017ba5e7e160484c9051cfb458221d42" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (vector y | matrix x, real alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(vector y | matrix x, real alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-e4b62bbce38476467b6d5d8347d048688b1b22fe" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (vector y | matrix x, real alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(vector y | matrix x, real alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-bc577f1feffdbfa17e7321f5cd733afccd7e822f" class="index-entry"></a> <!-- real; normal_id_glm_lpdf; (vector y | matrix x, vector alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lpdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lpdf</code></a></strong><code>(vector y | matrix x, vector alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code>.</p>
+<small><i>Available since 2.30</i></small>
+<a id="index-entry-6c41e1877f2f156d78215c8b4f25b5795223a0d3" class="index-entry"></a> <!-- real; normal_id_glm_lupdf; (vector y | matrix x, vector alpha, vector beta, vector sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#normal_id_glm_lupdf" class="unlink" title="Jump to index entry"><code>normal_id_glm_lupdf</code></a></strong><code>(vector y | matrix x, vector alpha, vector beta, vector sigma)</code><br> The log normal probability density of <code>y</code> given location <code>alpha + x * beta</code> and scale <code>sigma</code> dropping constant additive terms.</p>
+<small><i>Available since 2.30</i></small>
+</section>
+</section>
+<section id="exponentially-modified-normal-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="exponentially-modified-normal-distribution">Exponentially modified normal distribution</h2>
+<p>Exponentially modified Gaussian describes the distribution of <span class="math inline">\(Z = X + Y\)</span> when <span class="math inline">\(X\)</span> and <span class="math inline">\(Y\)</span> are independent and <span class="math inline">\(X\)</span> is normally distributed (with mean <span class="math inline">\(\mu\)</span> and standard deviation <span class="math inline">\(\sigma\)</span>) and <span class="math inline">\(Y\)</span> is exponentially distributed (with rate <span class="math inline">\(\lambda\)</span>).</p>
+<section id="probability-density-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-1">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span>, <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, and <span class="math inline">\(\lambda \in
+\mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{equation*}
+\text{ExpModNormal}(y|\mu,\sigma,\lambda) = \frac{\lambda}{2} \ \exp
+\left(\frac{\lambda}{2} \left(2\mu + \lambda \sigma^2 -
+2y\right)\right) \text{erfc}\left(\frac{\mu + \lambda\sigma^2 -
+y}{\sqrt{2}\sigma}\right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-3" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-3">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal" class="unlink" title="Jump to index entry"><code>exp_mod_normal</code></a></strong><code>(mu, sigma, lambda)</code></p>
+<p>Increment target log probability density with <code>exp_mod_normal_lupdf(y | mu, sigma, lambda)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b426ea4ba5c486f0dee2d11bcd8f019e1914e368" class="index-entry"></a> <!-- real; exp_mod_normal ~; -->
+<p></p>
+</section>
+<section id="stan-functions-3" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
+<a id="index-entry-bb7d02b1917d231421d8c9d936fb60acd1dee69c" class="index-entry"></a> <!-- real; exp_mod_normal_lpdf; (reals y | reals mu, reals sigma, reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal_lpdf" class="unlink" title="Jump to index entry"><code>exp_mod_normal_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals lambda)</code><br> The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-8933df6537efc956060be8137a18b3593c5c4833" class="index-entry"></a> <!-- real; exp_mod_normal_lupdf; (reals y | reals mu, reals sigma, reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal_lupdf" class="unlink" title="Jump to index entry"><code>exp_mod_normal_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals lambda)</code><br> The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-e60633c944fc492d9aabe37e92180634dc3beb1f" class="index-entry"></a> <!-- real; exp_mod_normal_cdf; (reals y | reals mu, reals sigma, reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal_cdf" class="unlink" title="Jump to index entry"><code>exp_mod_normal_cdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals lambda)</code><br> The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-06dbb9069af1f1382e22ed5f13335630182ae084" class="index-entry"></a> <!-- real; exp_mod_normal_lcdf; (reals y | reals mu, reals sigma, reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal_lcdf" class="unlink" title="Jump to index entry"><code>exp_mod_normal_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals lambda)</code><br> The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-ee7f0016f4ebc38e3cf0d9590e2a47ae35d6d131" class="index-entry"></a> <!-- real; exp_mod_normal_lccdf; (reals y | reals mu, reals sigma, reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal_lccdf" class="unlink" title="Jump to index entry"><code>exp_mod_normal_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals lambda)</code><br> The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-673b303c582c3796fa693d9eee79c861cc6e4b3f" class="index-entry"></a> <!-- R; exp_mod_normal_rng; (reals mu, reals sigma, reals lambda); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#exp_mod_normal_rng" class="unlink" title="Jump to index entry"><code>exp_mod_normal_rng</code></a></strong><code>(reals mu, reals sigma, reals lambda)</code><br> Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="skew-normal-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="skew-normal-distribution">Skew normal distribution</h2>
+<section id="probability-density-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-2">Probability density function</h3>
+<p>If <span class="math inline">\(\xi \in \mathbb{R}\)</span>, <span class="math inline">\(\omega \in \mathbb{R}^+\)</span>, and <span class="math inline">\(\alpha \in
+\mathbb{R}\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{equation*} \text{SkewNormal}(y \mid
+\xi, \omega, \alpha) = \frac{1}{\omega\sqrt{2\pi}} \ \exp\left( - \,
+\frac{1}{2}            \left(  \frac{y - \xi}{\omega} \right)^2
+\right) \ \left(1 + \text{erf}\left( \alpha\left(\frac{y -
+\xi}{\omega\sqrt{2}}\right)\right)\right) . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-4" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-4">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#skew_normal" class="unlink" title="Jump to index entry"><code>skew_normal</code></a></strong><code>(xi, omega, alpha)</code></p>
+<p>Increment target log probability density with <code>skew_normal_lupdf(y | xi, omega, alpha)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-70ae09d4a5fd934163b857b8af6509a78d7717df" class="index-entry"></a> <!-- real; skew_normal ~; -->
+<p></p>
+</section>
+<section id="stan-functions-4" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-4">Stan functions</h3>
+<a id="index-entry-f14e52bfc7ddf652898086b26b652a779adfffbd" class="index-entry"></a> <!-- real; skew_normal_lpdf; (reals y | reals xi, reals omega, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_normal_lpdf" class="unlink" title="Jump to index entry"><code>skew_normal_lpdf</code></a></strong><code>(reals y | reals xi, reals omega, reals alpha)</code><br> The log of the skew normal density of y given location xi, scale omega, and shape alpha</p>
+<small><i>Available since 2.16</i></small>
+<a id="index-entry-57aff3449f2e94559d54277820ea0537a5b2d44f" class="index-entry"></a> <!-- real; skew_normal_lupdf; (reals y | reals xi, reals omega, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_normal_lupdf" class="unlink" title="Jump to index entry"><code>skew_normal_lupdf</code></a></strong><code>(reals y | reals xi, reals omega, reals alpha)</code><br> The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-4e1fe342b2c2a2f8efd1b736d66c0d918bcd6699" class="index-entry"></a> <!-- real; skew_normal_cdf; (reals y | reals xi, reals omega, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_normal_cdf" class="unlink" title="Jump to index entry"><code>skew_normal_cdf</code></a></strong><code>(reals y | reals xi, reals omega, reals alpha)</code><br> The skew normal distribution function of y given location xi, scale omega, and shape alpha</p>
+<small><i>Available since 2.16</i></small>
+<a id="index-entry-55d8df839becff01c2c535a00f18ae07d745f20f" class="index-entry"></a> <!-- real; skew_normal_lcdf; (reals y | reals xi, reals omega, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_normal_lcdf" class="unlink" title="Jump to index entry"><code>skew_normal_lcdf</code></a></strong><code>(reals y | reals xi, reals omega, reals alpha)</code><br> The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-f317d012b6071ab55a6345c89527fb421e4607ec" class="index-entry"></a> <!-- real; skew_normal_lccdf; (reals y | reals xi, reals omega, reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_normal_lccdf" class="unlink" title="Jump to index entry"><code>skew_normal_lccdf</code></a></strong><code>(reals y | reals xi, reals omega, reals alpha)</code><br> The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-18eec472b4732674d456a7bf36148e16322c3a0d" class="index-entry"></a> <!-- R; skew_normal_rng; (reals xi, reals omega, real alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#skew_normal_rng" class="unlink" title="Jump to index entry"><code>skew_normal_rng</code></a></strong><code>(reals xi, reals omega, real alpha)</code><br> Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="student-t-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="student-t-distribution">Student-t distribution</h2>
+<section id="probability-density-function-3" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-3">Probability density function</h3>
+<p>If <span class="math inline">\(\nu \in \mathbb{R}^+\)</span>, <span class="math inline">\(\mu \in \mathbb{R}\)</span>, and <span class="math inline">\(\sigma \in
+\mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{equation*}
+\text{StudentT}(y|\nu,\mu,\sigma) = \frac{\Gamma\left((\nu +
+1)/2\right)}      {\Gamma(\nu/2)} \ \frac{1}{\sqrt{\nu \pi} \ \sigma}
+\ \left( 1 + \frac{1}{\nu} \left(\frac{y - \mu}{\sigma}\right)^2
+\right)^{-(\nu + 1)/2} \! . \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-5" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-5">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#student_t" class="unlink" title="Jump to index entry"><code>student_t</code></a></strong><code>(nu, mu, sigma)</code></p>
+<p>Increment target log probability density with <code>student_t_lupdf(y | nu, mu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-2483b90edc6356f6ed17c209456b226b1bc0099b" class="index-entry"></a> <!-- real; student_t ~; -->
+<p></p>
+</section>
+<section id="stan-functions-5" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-5">Stan functions</h3>
+<a id="index-entry-b9eec4ab31579ecbd1b7330def17803b0e0fc7a4" class="index-entry"></a> <!-- real; student_t_lpdf; (reals y | reals nu, reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#student_t_lpdf" class="unlink" title="Jump to index entry"><code>student_t_lpdf</code></a></strong><code>(reals y | reals nu, reals mu, reals sigma)</code><br> The log of the Student-<span class="math inline">\(t\)</span> density of y given degrees of freedom nu, location mu, and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-324d76557438cfbe15364f652d1fde26b764ad5e" class="index-entry"></a> <!-- real; student_t_lupdf; (reals y | reals nu, reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#student_t_lupdf" class="unlink" title="Jump to index entry"><code>student_t_lupdf</code></a></strong><code>(reals y | reals nu, reals mu, reals sigma)</code><br> The log of the Student-<span class="math inline">\(t\)</span> density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-c14ffe34acf237ad6f14abea4b420b8d1ddc91c0" class="index-entry"></a> <!-- real; student_t_cdf; (reals y | reals nu, reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#student_t_cdf" class="unlink" title="Jump to index entry"><code>student_t_cdf</code></a></strong><code>(reals y | reals nu, reals mu, reals sigma)</code><br> The Student-<span class="math inline">\(t\)</span> cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c66e35d8ec8239f600bd048ba98bb327054d7e3e" class="index-entry"></a> <!-- real; student_t_lcdf; (reals y | reals nu, reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#student_t_lcdf" class="unlink" title="Jump to index entry"><code>student_t_lcdf</code></a></strong><code>(reals y | reals nu, reals mu, reals sigma)</code><br> The log of the Student-<span class="math inline">\(t\)</span> cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-f066450f95e0c6d86c44d8a4ecd7c462c0349ca9" class="index-entry"></a> <!-- real; student_t_lccdf; (reals y | reals nu, reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#student_t_lccdf" class="unlink" title="Jump to index entry"><code>student_t_lccdf</code></a></strong><code>(reals y | reals nu, reals mu, reals sigma)</code><br> The log of the Student-<span class="math inline">\(t\)</span> complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-66f1428f578865e4e4d3a0fc3c845d58bbfbdfb0" class="index-entry"></a> <!-- R; student_t_rng; (reals nu, reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#student_t_rng" class="unlink" title="Jump to index entry"><code>student_t_rng</code></a></strong><code>(reals nu, reals mu, reals sigma)</code><br> Generate a Student-<span class="math inline">\(t\)</span> variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="cauchy-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="cauchy-distribution">Cauchy distribution</h2>
+<section id="probability-density-function-4" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-4">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Cauchy}(y|\mu,\sigma) = \frac{1}{\pi \sigma} \ \frac{1}{1 + \left((y - \mu)/\sigma\right)^2} .
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-6" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-6">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#cauchy" class="unlink" title="Jump to index entry"><code>cauchy</code></a></strong><code>(mu, sigma)</code></p>
+<p>Increment target log probability density with <code>cauchy_lupdf(y | mu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4e9a3b08fddc23e306b5a3b0a4f199124cb7cd90" class="index-entry"></a> <!-- real; cauchy ~; -->
+<p></p>
+</section>
+<section id="stan-functions-6" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-6">Stan functions</h3>
+<a id="index-entry-365b2742adf729c8087f3eb28841c38baec29082" class="index-entry"></a> <!-- real; cauchy_lpdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#cauchy_lpdf" class="unlink" title="Jump to index entry"><code>cauchy_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the Cauchy density of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-70ded17908d0eaf198cca69f74a0bec65a4ec897" class="index-entry"></a> <!-- real; cauchy_lupdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#cauchy_lupdf" class="unlink" title="Jump to index entry"><code>cauchy_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-f636bb8d19bdc544f177d76c3e93b11a8947a332" class="index-entry"></a> <!-- real; cauchy_cdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#cauchy_cdf" class="unlink" title="Jump to index entry"><code>cauchy_cdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The Cauchy cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-1470e53e1ef3170600faa1e10925e17becd26f46" class="index-entry"></a> <!-- real; cauchy_lcdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#cauchy_lcdf" class="unlink" title="Jump to index entry"><code>cauchy_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the Cauchy cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-252fde132b94e629570349d740a553001ba151d7" class="index-entry"></a> <!-- real; cauchy_lccdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#cauchy_lccdf" class="unlink" title="Jump to index entry"><code>cauchy_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-56d08e840245967636a7f56a3b26cf0ad54161f8" class="index-entry"></a> <!-- R; cauchy_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#cauchy_rng" class="unlink" title="Jump to index entry"><code>cauchy_rng</code></a></strong><code>(reals mu, reals sigma)</code><br> Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="double-exponential-laplace-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="double-exponential-laplace-distribution">Double exponential (Laplace) distribution</h2>
+<section id="probability-density-function-5" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-5">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in
+\mathbb{R}\)</span>, <span class="math display">\[\begin{equation*} \text{DoubleExponential}(y|\mu,\sigma) =
+\frac{1}{2\sigma}   \exp \left( - \, \frac{|y - \mu|}{\sigma} \right)
+. \end{equation*}\]</span> Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section <a href="../functions-reference/positive_continuous_distributions.html#exponential-distribution">exponential distribution</a>), which is parameterized in terms of inverse scale.</p>
+<p>The double-exponential distribution can be defined as a compound exponential-normal distribution <span class="citation" data-cites="Ding:18">(<a href="#ref-Ding:18" role="doc-biblioref">Ding and Blitzstein 2018</a>)</span>. Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write <span class="math display">\[\begin{equation*} \alpha \sim
+\mathsf{Exponential}\left( \frac{1}{2 \sigma^2} \right) \end{equation*}\]</span> and <span class="math display">\[\begin{equation*} \beta \mid
+\alpha \sim \mathsf{Normal}(\mu, \sqrt{\alpha}), \end{equation*}\]</span> then <span class="math display">\[\begin{equation*} \beta \sim
+\mathsf{DoubleExponential}(\mu, \sigma ). \end{equation*}\]</span> This may be used to code a non-centered parameterization by taking <span class="math display">\[\begin{equation*} \beta^{\text{raw}} \sim
+\mathsf{Normal}(0, 1) \end{equation*}\]</span> and defining <span class="math display">\[\begin{equation*} \beta = \mu + \sqrt{\alpha} \,
+\beta^{\text{raw}}. \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-7" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-7">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#double_exponential" class="unlink" title="Jump to index entry"><code>double_exponential</code></a></strong><code>(mu, sigma)</code></p>
+<p>Increment target log probability density with <code>double_exponential_lupdf(y | mu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-b3a44920aadcdd73a61301af6bfb0ba2be51abf8" class="index-entry"></a> <!-- real; double_exponential ~; -->
+<p></p>
+</section>
+<section id="stan-functions-7" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
+<a id="index-entry-f0eddd37cb58304271349b38fc0dc4de2aeb94f4" class="index-entry"></a> <!-- real; double_exponential_lpdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#double_exponential_lpdf" class="unlink" title="Jump to index entry"><code>double_exponential_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the double exponential density of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-32427ca40b710f689e9f6f5f77ca5f91afc0ae09" class="index-entry"></a> <!-- real; double_exponential_lupdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#double_exponential_lupdf" class="unlink" title="Jump to index entry"><code>double_exponential_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-8339bce2e4f464e36cd18cd462828f0b740d2657" class="index-entry"></a> <!-- real; double_exponential_cdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#double_exponential_cdf" class="unlink" title="Jump to index entry"><code>double_exponential_cdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The double exponential cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-a77ae5c4ace742352452353b5faa44105bc79f55" class="index-entry"></a> <!-- real; double_exponential_lcdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#double_exponential_lcdf" class="unlink" title="Jump to index entry"><code>double_exponential_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the double exponential cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-14eaa0ccbccd40ad1a210cba1462140349b9c18d" class="index-entry"></a> <!-- real; double_exponential_lccdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#double_exponential_lccdf" class="unlink" title="Jump to index entry"><code>double_exponential_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-5489a23aaef4c1953d0e22e808a68a6fc37ab1d8" class="index-entry"></a> <!-- R; double_exponential_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#double_exponential_rng" class="unlink" title="Jump to index entry"><code>double_exponential_rng</code></a></strong><code>(reals mu, reals sigma)</code><br> Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="logistic-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="logistic-distribution">Logistic distribution</h2>
+<section id="probability-density-function-6" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-6">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Logistic}(y|\mu,\sigma) = \frac{1}{\sigma} \ \exp\!\left( - \, \frac{y - \mu}{\sigma} \right)
+\ \left(1 + \exp \!\left( - \, \frac{y - \mu}{\sigma} \right) \right)^{\!-2} \! .
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-8" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-8">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#logistic" class="unlink" title="Jump to index entry"><code>logistic</code></a></strong><code>(mu, sigma)</code></p>
+<p>Increment target log probability density with <code>logistic_lupdf(y | mu, sigma)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-83ebb5b2c1c5976cb145e5c49c438dcedef9d5cc" class="index-entry"></a> <!-- real; logistic ~; -->
+<p></p>
+</section>
+<section id="stan-functions-8" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-8">Stan functions</h3>
+<a id="index-entry-defc2f03147e4fc4d6ea7092ba14bfa3e211169e" class="index-entry"></a> <!-- real; logistic_lpdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#logistic_lpdf" class="unlink" title="Jump to index entry"><code>logistic_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the logistic density of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-77bd7fb60be125c2b18d3f44f85bc656a0b9b919" class="index-entry"></a> <!-- real; logistic_lupdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#logistic_lupdf" class="unlink" title="Jump to index entry"><code>logistic_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the logistic density of y given location mu and scale sigma dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-4883524cdc0ee6a2baed3cdf4c56a2221fc3c5e4" class="index-entry"></a> <!-- real; logistic_cdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#logistic_cdf" class="unlink" title="Jump to index entry"><code>logistic_cdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The logistic cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-ae238a73fddb4ade983ee9f6ff91cdd7dcc57968" class="index-entry"></a> <!-- real; logistic_lcdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#logistic_lcdf" class="unlink" title="Jump to index entry"><code>logistic_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the logistic cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-6cce712fc62f4073bbd1140a1dacdff12e0f72ce" class="index-entry"></a> <!-- real; logistic_lccdf; (reals y | reals mu, reals sigma); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#logistic_lccdf" class="unlink" title="Jump to index entry"><code>logistic_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma)</code><br> The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-9327491d09f9d375a4734c67a3727ca525f940f5" class="index-entry"></a> <!-- R; logistic_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#logistic_rng" class="unlink" title="Jump to index entry"><code>logistic_rng</code></a></strong><code>(reals mu, reals sigma)</code><br> Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="gumbel-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="gumbel-distribution">Gumbel distribution</h2>
+<section id="probability-density-function-7" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-7">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span> and <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Gumbel}(y|\mu,\beta) = \frac{1}{\beta} \ \exp\left(-\frac{y-\mu}{\beta}-\exp\left(-\frac{y-\mu}{\beta}\right)\right) .
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-9" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-9">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#gumbel" class="unlink" title="Jump to index entry"><code>gumbel</code></a></strong><code>(mu, beta)</code></p>
+<p>Increment target log probability density with <code>gumbel_lupdf(y | mu, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-846f4cd18ce2c11c64adc5430feb805c8548af12" class="index-entry"></a> <!-- real; gumbel ~; -->
+<p></p>
+</section>
+<section id="stan-functions-9" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-9">Stan functions</h3>
+<a id="index-entry-93b41c8b64bc3c223c99733bf33da8c3ec9ac492" class="index-entry"></a> <!-- real; gumbel_lpdf; (reals y | reals mu, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gumbel_lpdf" class="unlink" title="Jump to index entry"><code>gumbel_lpdf</code></a></strong><code>(reals y | reals mu, reals beta)</code><br> The log of the gumbel density of y given location mu and scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-b9841ea9c5279d1940a06429efc4126c8f1735d9" class="index-entry"></a> <!-- real; gumbel_lupdf; (reals y | reals mu, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gumbel_lupdf" class="unlink" title="Jump to index entry"><code>gumbel_lupdf</code></a></strong><code>(reals y | reals mu, reals beta)</code><br> The log of the gumbel density of y given location mu and scale beta dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-d75196df1f351627464af18ef549a68bbb63bf5e" class="index-entry"></a> <!-- real; gumbel_cdf; (reals y | reals mu, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gumbel_cdf" class="unlink" title="Jump to index entry"><code>gumbel_cdf</code></a></strong><code>(reals y | reals mu, reals beta)</code><br> The gumbel cumulative distribution function of y given location mu and scale beta</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-4cc1f4c7bfa6eedc97fbb6074ec3965cff5be24f" class="index-entry"></a> <!-- real; gumbel_lcdf; (reals y | reals mu, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gumbel_lcdf" class="unlink" title="Jump to index entry"><code>gumbel_lcdf</code></a></strong><code>(reals y | reals mu, reals beta)</code><br> The log of the gumbel cumulative distribution function of y given location mu and scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-c56b96f1fa2edfa9edfa06313030df780f89753c" class="index-entry"></a> <!-- real; gumbel_lccdf; (reals y | reals mu, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#gumbel_lccdf" class="unlink" title="Jump to index entry"><code>gumbel_lccdf</code></a></strong><code>(reals y | reals mu, reals beta)</code><br> The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-f0072bf29736a2f234a30ab11de070e3194bd4e3" class="index-entry"></a> <!-- R; gumbel_rng; (reals mu, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#gumbel_rng" class="unlink" title="Jump to index entry"><code>gumbel_rng</code></a></strong><code>(reals mu, reals beta)</code><br> Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="skew-double-exponential-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="skew-double-exponential-distribution">Skew double exponential distribution</h2>
+<section id="probability-density-function-8" class="level3">
+<h3 class="anchored" data-anchor-id="probability-density-function-8">Probability density function</h3>
+<p>If <span class="math inline">\(\mu \in \mathbb{R}\)</span>, <span class="math inline">\(\sigma \in \mathbb{R}^+\)</span> and <span class="math inline">\(\tau \in [0, 1]\)</span>, then for <span class="math inline">\(y \in \mathbb{R}\)</span>, <span class="math display">\[\begin{aligned}
+&amp; \text{SkewDoubleExponential}  (y|\mu,\sigma, \tau) = \\
+&amp; \qquad \qquad \frac{2 \tau (1 - \tau) }{\sigma} \exp \left[ - \frac{2}{\sigma} \left[ \left(1 - \tau \right) I(y &lt; \mu)  (\mu - y) + \tau I(y &gt; \mu)(y-\mu)  \right]    \right]
+\end{aligned}\]</span></p>
+</section>
+<section id="distribution-statement-10" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-10">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential" class="unlink" title="Jump to index entry"><code>skew_double_exponential</code></a></strong><code>(mu, sigma, tau)</code></p>
+<p>Increment target log probability density with <code>skew_double_exponential(y | mu, sigma, tau)</code></p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-c133dee4da92b9a21386219b4a5fbb6b5ced9f23" class="index-entry"></a> <!-- real; skew_double_exponential ~; -->
+<p></p>
+</section>
+<section id="stan-functions-10" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-10">Stan functions</h3>
+<a id="index-entry-903ddf1695b27de2b4e5bcdfd95493d2b2c40105" class="index-entry"></a> <!-- real; skew_double_exponential_lpdf; (reals y | reals mu, reals sigma, reals tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential_lpdf" class="unlink" title="Jump to index entry"><code>skew_double_exponential_lpdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals tau)</code><br> The log of the skew double exponential density of y given location mu, scale sigma and skewness tau</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-344e5af2a56736dcd0d5a8df8fa7f5204b2392b4" class="index-entry"></a> <!-- real; skew_double_exponential_lupdf; (reals y | reals mu, reals sigma, reals tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential_lupdf" class="unlink" title="Jump to index entry"><code>skew_double_exponential_lupdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals tau)</code><br> The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-bbc59f7f6b6b9b62a78cc86780d31dbc261ef032" class="index-entry"></a> <!-- real; skew_double_exponential_cdf; (reals y | reals mu, reals sigma, reals tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential_cdf" class="unlink" title="Jump to index entry"><code>skew_double_exponential_cdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals tau)</code><br> The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-4d7480cfff426af8ef69e62abd1a90b246aa386d" class="index-entry"></a> <!-- real; skew_double_exponential_lcdf; (reals y | reals mu, reals sigma, reals tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential_lcdf" class="unlink" title="Jump to index entry"><code>skew_double_exponential_lcdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals tau)</code><br> The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-3b13bcb4e64d73e98c4ff5ebfa01691c2ba8cb43" class="index-entry"></a> <!-- real; skew_double_exponential_lccdf; (reals y | reals mu, reals sigma, reals tau); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential_lccdf" class="unlink" title="Jump to index entry"><code>skew_double_exponential_lccdf</code></a></strong><code>(reals y | reals mu, reals sigma, reals tau)</code><br> The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau</p>
+<small><i>Available since 2.28</i></small>
+<a id="index-entry-5ba4f22bc09b1f9b4aaef277e2148dd77a3dd8b6" class="index-entry"></a> <!-- R; skew_double_exponential_rng; (reals mu, reals sigma); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#skew_double_exponential_rng" class="unlink" title="Jump to index entry"><code>skew_double_exponential_rng</code></a></strong><code>(reals mu, reals sigma, reals tau)</code><br> Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prng-vectorization">vectorized PRNG functions</a>.</p>
+<small><i>Available since 2.28</i></small>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Ding:18" class="csl-entry" role="listitem">
+Ding, Peng, and Joseph K. Blitzstein. 2018. <span>“On the Gaussian Mixture Representation of the Laplace Distribution.”</span> <em>The American Statistician</em> 72 (2): 172–74. <a href="https://doi.org/10.1080/00031305.2017.1291448">https://doi.org/10.1080/00031305.2017.1291448</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/multivariate_discrete_distributions.html" class="pagination-link" aria-label="Multivariate Discrete Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Multivariate Discrete Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/positive_continuous_distributions.html" class="pagination-link" aria-label="Positive Continuous Distributions">
+        <span class="nav-page-text">Positive Continuous Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/unbounded_discrete_distributions.html b/docs/2_39/functions-reference/unbounded_discrete_distributions.html
new file mode 100644
index 000000000..d06a2783e
--- /dev/null
+++ b/docs/2_39/functions-reference/unbounded_discrete_distributions.html
@@ -0,0 +1,1662 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Unbounded Discrete Distributions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/multivariate_discrete_distributions.html" rel="next">
+<link href="../functions-reference/bounded_discrete_distributions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Unbounded Discrete Distributions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/unbounded_discrete_distributions.html">Unbounded Discrete Distributions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#unbounded-discrete-distributions" id="toc-unbounded-discrete-distributions" class="nav-link active" data-scroll-target="#unbounded-discrete-distributions">Unbounded Discrete Distributions</a>
+  <ul class="collapse">
+  <li><a href="#negative-binomial-distribution" id="toc-negative-binomial-distribution" class="nav-link" data-scroll-target="#negative-binomial-distribution">Negative binomial distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function" id="toc-probability-mass-function" class="nav-link" data-scroll-target="#probability-mass-function">Probability mass function</a></li>
+  <li><a href="#distribution-statement" id="toc-distribution-statement" class="nav-link" data-scroll-target="#distribution-statement">Distribution statement</a></li>
+  <li><a href="#stan-functions" id="toc-stan-functions" class="nav-link" data-scroll-target="#stan-functions">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#nbalt" id="toc-nbalt" class="nav-link" data-scroll-target="#nbalt">Negative binomial distribution (alternative parameterization)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-1" id="toc-probability-mass-function-1" class="nav-link" data-scroll-target="#probability-mass-function-1">Probability mass function</a></li>
+  <li><a href="#distribution-statement-1" id="toc-distribution-statement-1" class="nav-link" data-scroll-target="#distribution-statement-1">Distribution statement</a></li>
+  <li><a href="#stan-functions-1" id="toc-stan-functions-1" class="nav-link" data-scroll-target="#stan-functions-1">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#neg-binom-2-log" id="toc-neg-binom-2-log" class="nav-link" data-scroll-target="#neg-binom-2-log">Negative binomial distribution (log alternative parameterization)</a>
+  <ul class="collapse">
+  <li><a href="#distribution-statement-2" id="toc-distribution-statement-2" class="nav-link" data-scroll-target="#distribution-statement-2">Distribution statement</a></li>
+  <li><a href="#stan-functions-2" id="toc-stan-functions-2" class="nav-link" data-scroll-target="#stan-functions-2">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#neg-binom-2-log-glm" id="toc-neg-binom-2-log-glm" class="nav-link" data-scroll-target="#neg-binom-2-log-glm">Negative-binomial-2-log generalized linear model (negative binomial regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-2" id="toc-probability-mass-function-2" class="nav-link" data-scroll-target="#probability-mass-function-2">Probability mass function</a></li>
+  <li><a href="#distribution-statement-3" id="toc-distribution-statement-3" class="nav-link" data-scroll-target="#distribution-statement-3">Distribution statement</a></li>
+  <li><a href="#stan-functions-3" id="toc-stan-functions-3" class="nav-link" data-scroll-target="#stan-functions-3">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#poisson" id="toc-poisson" class="nav-link" data-scroll-target="#poisson">Poisson distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-3" id="toc-probability-mass-function-3" class="nav-link" data-scroll-target="#probability-mass-function-3">Probability mass function</a></li>
+  <li><a href="#distribution-statement-4" id="toc-distribution-statement-4" class="nav-link" data-scroll-target="#distribution-statement-4">Distribution statement</a></li>
+  <li><a href="#stan-functions-4" id="toc-stan-functions-4" class="nav-link" data-scroll-target="#stan-functions-4">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#poisson-distribution-log-parameterization" id="toc-poisson-distribution-log-parameterization" class="nav-link" data-scroll-target="#poisson-distribution-log-parameterization">Poisson distribution, log parameterization</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-4" id="toc-probability-mass-function-4" class="nav-link" data-scroll-target="#probability-mass-function-4">Probability mass function</a></li>
+  <li><a href="#distribution-statement-5" id="toc-distribution-statement-5" class="nav-link" data-scroll-target="#distribution-statement-5">Distribution statement</a></li>
+  <li><a href="#stan-functions-5" id="toc-stan-functions-5" class="nav-link" data-scroll-target="#stan-functions-5">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#poisson-log-glm" id="toc-poisson-log-glm" class="nav-link" data-scroll-target="#poisson-log-glm">Poisson-log generalized linear model (Poisson regression)</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-5" id="toc-probability-mass-function-5" class="nav-link" data-scroll-target="#probability-mass-function-5">Probability mass function</a></li>
+  <li><a href="#distribution-statement-6" id="toc-distribution-statement-6" class="nav-link" data-scroll-target="#distribution-statement-6">Distribution statement</a></li>
+  <li><a href="#stan-functions-6" id="toc-stan-functions-6" class="nav-link" data-scroll-target="#stan-functions-6">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#beta-neg-binomial" id="toc-beta-neg-binomial" class="nav-link" data-scroll-target="#beta-neg-binomial">Beta negative binomial distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-6" id="toc-probability-mass-function-6" class="nav-link" data-scroll-target="#probability-mass-function-6">Probability mass function</a></li>
+  <li><a href="#distribution-statement-7" id="toc-distribution-statement-7" class="nav-link" data-scroll-target="#distribution-statement-7">Distribution statement</a></li>
+  <li><a href="#stan-functions-7" id="toc-stan-functions-7" class="nav-link" data-scroll-target="#stan-functions-7">Stan functions</a></li>
+  </ul></li>
+  <li><a href="#yule-simon" id="toc-yule-simon" class="nav-link" data-scroll-target="#yule-simon">Yule Simon distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-7" id="toc-probability-mass-function-7" class="nav-link" data-scroll-target="#probability-mass-function-7">Probability mass function</a></li>
+  <li><a href="#distribution-statement-8" id="toc-distribution-statement-8" class="nav-link" data-scroll-target="#distribution-statement-8">Distribution statement</a></li>
+  <li><a href="#stan-functions-8" id="toc-stan-functions-8" class="nav-link" data-scroll-target="#stan-functions-8">Stan functions</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/binary_distributions.html">Discrete Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/unbounded_discrete_distributions.html">Unbounded Discrete Distributions</a></li></ol></nav></header>
+
+
+
+
+<section id="unbounded-discrete-distributions" class="level1">
+<h1>Unbounded Discrete Distributions</h1>
+<p>The unbounded discrete distributions have support over the natural numbers (i.e., the non-negative integers).</p>
+<section id="negative-binomial-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="negative-binomial-distribution">Negative binomial distribution</h2>
+<p>For the negative binomial distribution Stan uses the parameterization described in <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span>. For alternative parameterizations, see section <a href="#neg-binom-2-log">negative binomial glm</a>.</p>
+<section id="probability-mass-function" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function">Probability mass function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span> and <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(n
+\in \mathbb{N}\)</span>, <span class="math display">\[\begin{equation*} \text{NegBinomial}(n~|~\alpha,\beta) = \binom{n +
+\alpha - 1}{\alpha - 1} \, \left( \frac{\beta}{\beta+1}
+\right)^{\!\alpha} \, \left( \frac{1}{\beta + 1} \right)^{\!n} \!. \end{equation*}\]</span></p>
+<p>The mean and variance of a random variable <span class="math inline">\(n \sim
+\text{NegBinomial}(\alpha,\beta)\)</span> are given by <span class="math display">\[\begin{equation*} \mathbb{E}[n] =
+\frac{\alpha}{\beta} \ \ \text{ and } \ \ \text{Var}[n] =
+\frac{\alpha}{\beta^2} (\beta + 1). \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial" class="unlink" title="Jump to index entry"><code>neg_binomial</code></a></strong><code>(alpha, beta)</code></p>
+<p>Increment target log probability density with <code>neg_binomial_lupmf(n | alpha, beta)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-62400243069f91416bfa36c834b3635bebcbd43e" class="index-entry"></a> <!-- real; neg_binomial ~; -->
+<p></p>
+</section>
+<section id="stan-functions" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
+<a id="index-entry-2a11a12aef7b6b441ea425acd9d5cb21aa6e66c0" class="index-entry"></a> <!-- real; neg_binomial_lpmf; (ints n | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_lpmf</code></a></strong><code>(ints n | reals alpha, reals beta)</code><br> The log negative binomial probability mass of <code>n</code> given shape <code>alpha</code> and inverse scale <code>beta</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-93c924df2baea99a71249bcad641d8f75c6e41cd" class="index-entry"></a> <!-- real; neg_binomial_lupmf; (ints n | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_lupmf</code></a></strong><code>(ints n | reals alpha, reals beta)</code><br> The log negative binomial probability mass of <code>n</code> given shape <code>alpha</code> and inverse scale <code>beta</code> dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-de9079d9e962df30ca27f6860eae99efcc654189" class="index-entry"></a> <!-- real; neg_binomial_cdf; (ints n | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_cdf" class="unlink" title="Jump to index entry"><code>neg_binomial_cdf</code></a></strong><code>(ints n | reals alpha, reals beta)</code><br> The negative binomial cumulative distribution function of <code>n</code> given shape <code>alpha</code> and inverse scale <code>beta</code></p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-83ea716052404e4f13e9f1c9f52092361d064615" class="index-entry"></a> <!-- real; neg_binomial_lcdf; (ints n | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_lcdf" class="unlink" title="Jump to index entry"><code>neg_binomial_lcdf</code></a></strong><code>(ints n | reals alpha, reals beta)</code><br> The log of the negative binomial cumulative distribution function of <code>n</code> given shape <code>alpha</code> and inverse scale <code>beta</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-8c72c425ce2f6cbc3dad23f6abb0dd9278749e84" class="index-entry"></a> <!-- real; neg_binomial_lccdf; (ints n | reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_lccdf" class="unlink" title="Jump to index entry"><code>neg_binomial_lccdf</code></a></strong><code>(ints n | reals alpha, reals beta)</code><br> The log of the negative binomial complementary cumulative distribution function of <code>n</code> given shape <code>alpha</code> and inverse scale <code>beta</code></p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-4c2098ee820ccf0732443eb5d87c0d6d839cce4d" class="index-entry"></a> <!-- R; neg_binomial_rng; (reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_rng" class="unlink" title="Jump to index entry"><code>neg_binomial_rng</code></a></strong><code>(reals alpha, reals beta)</code><br> Generate a negative binomial variate with shape <code>alpha</code> and inverse scale <code>beta</code>; may only be used in transformed data and generated quantities blocks. <code>alpha</code> <span class="math inline">\(/\)</span> <code>beta</code> must be less than <span class="math inline">\(2 ^ {29}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="nbalt" class="level2">
+<h2 class="anchored" data-anchor-id="nbalt">Negative binomial distribution (alternative parameterization)</h2>
+<p>Stan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section <a href="../functions-reference/real-valued_basic_functions.html#betafun">combinatorial functions</a>, below, provides a second alternative parameterization directly in terms of the log mean.</p>
+<section id="probability-mass-function-1" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-1">Probability mass function</h3>
+<p>The first parameterization is for <span class="math inline">\(\mu \in \mathbb{R}^+\)</span> and <span class="math inline">\(\phi \in
+\mathbb{R}^+\)</span>, which for <span class="math inline">\(n \in \mathbb{N}\)</span> is defined as <span class="math display">\[\begin{equation*}
+\text{NegBinomial2}(n \, | \, \mu, \phi)
+= \binom{n + \phi - 1}{n} \, \left( \frac{\mu}{\mu+\phi} \right)^{\!n} \, \left( \frac{\phi}{\mu+\phi} \right)^{\!\phi} \!.
+\end{equation*}\]</span></p>
+<p>The mean and variance of a random variable <span class="math inline">\(n \sim
+\text{NegBinomial2}(n~|~\mu,\phi)\)</span> are <span class="math display">\[\begin{equation*}
+\mathbb{E}[n] = \mu \ \ \ \text{ and } \ \ \ \text{Var}[n] = \mu + \frac{\mu^2}{\phi}.
+\end{equation*}\]</span> Recall that <span class="math inline">\(\text{Poisson}(\mu)\)</span> has variance <span class="math inline">\(\mu\)</span>, so <span class="math inline">\(\mu^2 / \phi &gt; 0\)</span> is the additional variance of the negative binomial above that of the Poisson with mean <span class="math inline">\(\mu\)</span>. So the inverse of parameter <span class="math inline">\(\phi\)</span> controls the overdispersion, scaled by the square of the mean, <span class="math inline">\(\mu^2\)</span>.</p>
+</section>
+<section id="distribution-statement-1" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-1">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2" class="unlink" title="Jump to index entry"><code>neg_binomial_2</code></a></strong><code>(mu, phi)</code></p>
+<p>Increment target log probability density with <code>neg_binomial_2_lupmf(n | mu, phi)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-ea4ba68ab4a32b0e2125adc6e5304e54972de2a9" class="index-entry"></a> <!-- real; neg_binomial_2 ~; -->
+<p></p>
+</section>
+<section id="stan-functions-1" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
+<a id="index-entry-711f077d61fb2160d3c8b33015b15af366056763" class="index-entry"></a> <!-- real; neg_binomial_2_lpmf; (ints n | reals mu, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_lpmf</code></a></strong><code>(ints n | reals mu, reals phi)</code><br> The log negative binomial probability mass of <code>n</code> given location <code>mu</code> and precision <code>phi</code>.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-3232cdce0c2d7dce167ba0946aafbbbe5cc4a28f" class="index-entry"></a> <!-- real; neg_binomial_2_lupmf; (ints n | reals mu, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_lupmf</code></a></strong><code>(ints n | reals mu, reals phi)</code><br> The log negative binomial probability mass of <code>n</code> given location <code>mu</code> and precision <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-a5768bbcd476f90cbae7573d0978531f6ad5a9fa" class="index-entry"></a> <!-- real; neg_binomial_2_cdf; (ints n | reals mu, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_cdf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_cdf</code></a></strong><code>(ints n | reals mu, reals phi)</code><br> The negative binomial cumulative distribution function of <code>n</code> given location <code>mu</code> and precision <code>phi</code>.</p>
+<small><i>Available since 2.6</i></small>
+<a id="index-entry-7bb63856ad5d268ddb11337ad2e4e53f7e54a273" class="index-entry"></a> <!-- real; neg_binomial_2_lcdf; (ints n | reals mu, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_lcdf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_lcdf</code></a></strong><code>(ints n | reals mu, reals phi)</code><br> The log of the negative binomial cumulative distribution function of <code>n</code> given location <code>mu</code> and precision <code>phi</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-bfa4fa4a4278e2a259bbd9bae0faca3d4e99b1eb" class="index-entry"></a> <!-- real; neg_binomial_2_lccdf; (ints n | reals mu, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_lccdf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_lccdf</code></a></strong><code>(ints n | reals mu, reals phi)</code><br> The log of the negative binomial complementary cumulative distribution function of <code>n</code> given location <code>mu</code> and precision <code>phi</code>.</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-eabf098a3499209e5f0959ac998ed4fcf53b0a51" class="index-entry"></a> <!-- R; neg_binomial_2_rng; (reals mu, reals phi); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_rng" class="unlink" title="Jump to index entry"><code>neg_binomial_2_rng</code></a></strong><code>(reals mu, reals phi)</code><br> Generate a negative binomial variate with location <code>mu</code> and precision <code>phi</code>; may only be used in transformed data and generated quantities blocks. <code>mu</code> must be less than <span class="math inline">\(2 ^ {29}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="neg-binom-2-log" class="level2">
+<h2 class="anchored" data-anchor-id="neg-binom-2-log">Negative binomial distribution (log alternative parameterization)</h2>
+<p>Related to the parameterization in section <a href="#nbalt">negative binomial, alternative parameterization</a>, the following parameterization uses a log mean parameter <span class="math inline">\(\eta = \log(\mu)\)</span>, defined for <span class="math inline">\(\eta \in \mathbb{R}\)</span>, <span class="math inline">\(\phi \in \mathbb{R}^+\)</span>, so that for <span class="math inline">\(n \in
+\mathbb{N}\)</span>, <span class="math display">\[\begin{equation*} \text{NegBinomial2Log}(n \, | \, \eta, \phi) =
+\text{NegBinomial2}(n | \exp(\eta), \phi). \end{equation*}\]</span> This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.</p>
+<section id="distribution-statement-2" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-2">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log</code></a></strong><code>(eta, phi)</code></p>
+<p>Increment target log probability density with <code>neg_binomial_2_log_lupmf(n | eta, phi)</code>.</p>
+<small><i>Available since 2.3</i></small>
+<a id="index-entry-92d907cea1dc522c9b36cf1caa3f7c6e8360fe7b" class="index-entry"></a> <!-- real; neg_binomial_2_log ~; -->
+<p></p>
+</section>
+<section id="stan-functions-2" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
+<a id="index-entry-e583555ff09b9ca6a30c94843f0e429145fe29ad" class="index-entry"></a> <!-- real; neg_binomial_2_log_lpmf; (ints n | reals eta, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_lpmf</code></a></strong><code>(ints n | reals eta, reals phi)</code><br> The log negative binomial probability mass of <code>n</code> given log-location <code>eta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.20</i></small>
+<a id="index-entry-79456eb164dc3e83dad8b862ae24a86882f2d536" class="index-entry"></a> <!-- real; neg_binomial_2_log_lupmf; (ints n | reals eta, reals phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_lupmf</code></a></strong><code>(ints n | reals eta, reals phi)</code><br> The log negative binomial probability mass of <code>n</code> given log-location <code>eta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-81dfe168f8aedf6dbf3f3ff8025936b2cf1052a1" class="index-entry"></a> <!-- R; neg_binomial_2_log_rng; (reals eta, reals phi); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_rng" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_rng</code></a></strong><code>(reals eta, reals phi)</code><br> Generate a negative binomial variate with log-location <code>eta</code> and inverse overdispersion control <code>phi</code>; may only be used in transformed data and generated quantities blocks. <code>eta</code> must be less than <span class="math inline">\(29 \log 2\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="neg-binom-2-log-glm" class="level2">
+<h2 class="anchored" data-anchor-id="neg-binom-2-log-glm">Negative-binomial-2-log generalized linear model (negative binomial regression)</h2>
+<p>Stan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e.&nbsp;a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.</p>
+<section id="probability-mass-function-2" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-2">Probability mass function</h3>
+<p>If <span class="math inline">\(x\in \mathbb{R}^{n\cdot m}, \alpha \in \mathbb{R}^n, \beta\in
+\mathbb{R}^m, \phi\in \mathbb{R}^+\)</span>, then for <span class="math inline">\(y \in \mathbb{N}^n\)</span>, <span class="math display">\[\begin{equation*}
+\text{NegBinomial2LogGLM}(y~|~x, \alpha, \beta, \phi) = \prod_{1\leq i
+\leq n}\text{NegBinomial2}(y_i~|~\exp(\alpha_i + x_i\cdot \beta),
+\phi). \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-3" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-3">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm</code></a></strong><code>(x, alpha, beta, phi)</code></p>
+<p>Increment target log probability density with <code>neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi)</code>.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-28895026932258af73faa16c7dec6a131af227bb" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm ~; -->
+<p></p>
+</section>
+<section id="stan-functions-3" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
+<a id="index-entry-7fe4a0797bb0b9fced5c0f2d75c06e1e479ea248" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lpmf; (int y | matrix x, real alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lpmf</code></a></strong><code>(int y | matrix x, real alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-09ef2622134a433c0f9c3ad0643e20255979e98a" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lupmf; (int y | matrix x, real alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lupmf</code></a></strong><code>(int y | matrix x, real alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-91d1747bb7a9e5a3edc9ed5ec4c377f643eac2ca" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lpmf; (int y | matrix x, vector alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lpmf</code></a></strong><code>(int y | matrix x, vector alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-d01dd7632fe36e888d0067b62cae9fd79230af8d" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lupmf; (int y | matrix x, vector alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lupmf</code></a></strong><code>(int y | matrix x, vector alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-16a3cf3cfe06586324838df97a33fbb2c2efeef7" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lpmf; (array[] int y | row_vector x, real alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, real alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-c6a6b66b9c19ab430c255ed8047ad014e1f6414c" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lupmf; (array[] int y | row_vector x, real alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, real alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-30a171234617b31ebd3414cb4a4f82c1b62b3d34" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lpmf; (array[] int y | row_vector x, vector alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-e736537c82f56050b3aaf75e43b6d50a38477dd9" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lupmf; (array[] int y | row_vector x, vector alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-145c118038ba258504502a0ca981d7c12532bc51" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lpmf; (array[] int y | matrix x, real alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, real alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-c5e5488af5e4e66825ea2b163862a7d35f6dd124" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lupmf; (array[] int y | matrix x, real alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, real alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-f5314e218cc2c6e5fe4e6301645ec2c0fdafbc11" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lpmf; (array[] int y | matrix x, vector alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-7346fdbc1e0f782c1d3ab15e9f444e03bc33b644" class="index-entry"></a> <!-- real; neg_binomial_2_log_glm_lupmf; (array[] int y | matrix x, vector alpha, vector beta, real phi); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#neg_binomial_2_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>neg_binomial_2_log_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, vector beta, real phi)</code><br> The log negative binomial probability mass of <code>y</code> given log-location <code>alpha + x * beta</code> and inverse overdispersion parameter <code>phi</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="poisson" class="level2">
+<h2 class="anchored" data-anchor-id="poisson">Poisson distribution</h2>
+<section id="probability-mass-function-3" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-3">Probability mass function</h3>
+<p>If <span class="math inline">\(\lambda \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(n \in \mathbb{N}\)</span>, <span class="math display">\[\begin{equation*}
+\text{Poisson}(n|\lambda) = \frac{1}{n!} \, \lambda^n \,
+\exp(-\lambda). \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-4" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-4">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#poisson" class="unlink" title="Jump to index entry"><code>poisson</code></a></strong><code>(lambda)</code></p>
+<p>Increment target log probability density with <code>poisson_lupmf(n | lambda)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-149c7a2f2ae0aede0ae6b19854be4a72ba136d10" class="index-entry"></a> <!-- real; poisson ~; -->
+<p></p>
+</section>
+<section id="stan-functions-4" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-4">Stan functions</h3>
+<a id="index-entry-468e745a522407843e4e4e0b5d50391126cfe3d6" class="index-entry"></a> <!-- real; poisson_lpmf; (ints n | reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_lpmf" class="unlink" title="Jump to index entry"><code>poisson_lpmf</code></a></strong><code>(ints n | reals lambda)</code><br> The log Poisson probability mass of n given rate lambda</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-ec57a3b0e25d5e9dd5cd37c27f9557e8123a4b4a" class="index-entry"></a> <!-- real; poisson_lupmf; (ints n | reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_lupmf" class="unlink" title="Jump to index entry"><code>poisson_lupmf</code></a></strong><code>(ints n | reals lambda)</code><br> The log Poisson probability mass of n given rate lambda dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-b9228882bc1807740bfbe2278fe3865895ddc14f" class="index-entry"></a> <!-- real; poisson_cdf; (ints n | reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_cdf" class="unlink" title="Jump to index entry"><code>poisson_cdf</code></a></strong><code>(ints n | reals lambda)</code><br> The Poisson cumulative distribution function of n given rate lambda</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-c53c3f02b2a02ee52cf555667086bbe0d63d9893" class="index-entry"></a> <!-- real; poisson_lcdf; (ints n | reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_lcdf" class="unlink" title="Jump to index entry"><code>poisson_lcdf</code></a></strong><code>(ints n | reals lambda)</code><br> The log of the Poisson cumulative distribution function of n given rate lambda</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-f630914d24a4ab0a6d95f950289e5a89655ee0b5" class="index-entry"></a> <!-- real; poisson_lccdf; (ints n | reals lambda); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_lccdf" class="unlink" title="Jump to index entry"><code>poisson_lccdf</code></a></strong><code>(ints n | reals lambda)</code><br> The log of the Poisson complementary cumulative distribution function of n given rate lambda</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-08d9cd65c9ed1ce8f26288c3f3c41ad143ddd486" class="index-entry"></a> <!-- R; poisson_rng; (reals lambda); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#poisson_rng" class="unlink" title="Jump to index entry"><code>poisson_rng</code></a></strong><code>(reals lambda)</code><br> Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than <span class="math inline">\(2^{30}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="poisson-distribution-log-parameterization" class="level2">
+<h2 class="anchored" data-anchor-id="poisson-distribution-log-parameterization">Poisson distribution, log parameterization</h2>
+<p>Stan also provides a parameterization of the Poisson using the log rate <span class="math inline">\(\alpha = \log \lambda\)</span> as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.</p>
+<section id="probability-mass-function-4" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-4">Probability mass function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}\)</span>, then for <span class="math inline">\(n \in \mathbb{N}\)</span>, <span class="math display">\[\begin{equation*}
+\text{PoissonLog}(n|\alpha) = \frac{1}{n!} \, \exp \left(n\alpha -
+\exp(\alpha) \right). \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-5" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-5">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#poisson_log" class="unlink" title="Jump to index entry"><code>poisson_log</code></a></strong><code>(alpha)</code></p>
+<p>Increment target log probability density with <code>poisson_log_lupmf(n | alpha)</code>.</p>
+<small><i>Available since 2.0</i></small>
+<a id="index-entry-48321c3e1d3764b7424c6aa9000bfb2c7c135d49" class="index-entry"></a> <!-- real; poisson_log ~; -->
+<p></p>
+</section>
+<section id="stan-functions-5" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-5">Stan functions</h3>
+<a id="index-entry-0a17b71120ddbfe9f78828a8db8ac327d7dfe433" class="index-entry"></a> <!-- real; poisson_log_lpmf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_lpmf</code></a></strong><code>(ints n | reals alpha)</code><br> The log Poisson probability mass of n given log rate alpha</p>
+<small><i>Available since 2.12</i></small>
+<a id="index-entry-6b7b04889612003d401cf29b1505e899a5ac3bdf" class="index-entry"></a> <!-- real; poisson_log_lupmf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_lupmf</code></a></strong><code>(ints n | reals alpha)</code><br> The log Poisson probability mass of n given log rate alpha dropping constant additive terms</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-2698e8369a4abee3bff347589811f0b59008cd62" class="index-entry"></a> <!-- R; poisson_log_rng; (reals alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_rng" class="unlink" title="Jump to index entry"><code>poisson_log_rng</code></a></strong><code>(reals alpha)</code><br> Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than <span class="math inline">\(30 \log 2\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.18</i></small>
+</section>
+</section>
+<section id="poisson-log-glm" class="level2">
+<h2 class="anchored" data-anchor-id="poisson-log-glm">Poisson-log generalized linear model (Poisson regression)</h2>
+<p>Stan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e.&nbsp;a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.</p>
+<section id="probability-mass-function-5" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-5">Probability mass function</h3>
+<p>If <span class="math inline">\(x\in \mathbb{R}^{n\cdot m}, \alpha \in \mathbb{R}^n, \beta\in
+\mathbb{R}^m\)</span>, then for <span class="math inline">\(y \in \mathbb{N}^n\)</span>, <span class="math display">\[\begin{equation*}
+\text{PoissonLogGLM}(y|x, \alpha, \beta) = \prod_{1\leq i \leq
+n}\text{Poisson}(y_i|\exp(\alpha_i + x_i\cdot \beta)). \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-6" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-6">Distribution statement</h3>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm" class="unlink" title="Jump to index entry"><code>poisson_log_glm</code></a></strong><code>(x, alpha, beta)</code></p>
+<p>Increment target log probability density with <code>poisson_log_glm_lupmf(y | x, alpha, beta)</code>.</p>
+<small><i>Available since 2.19</i></small>
+<a id="index-entry-6a7861af018d1e99d1f24782d56214c5b92e7586" class="index-entry"></a> <!-- real; poisson_log_glm ~; -->
+<p></p>
+</section>
+<section id="stan-functions-6" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-6">Stan functions</h3>
+<a id="index-entry-63439a9c4647ac7c084fdb2e91704175532de973" class="index-entry"></a> <!-- real; poisson_log_glm_lpmf; (int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lpmf</code></a></strong><code>(int y | matrix x, real alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-f122468454f01edf31a51909ec8967c9de864cf3" class="index-entry"></a> <!-- real; poisson_log_glm_lupmf; (int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lupmf</code></a></strong><code>(int y | matrix x, real alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-809af19433732f5658706fbd90b7e3489497b084" class="index-entry"></a> <!-- real; poisson_log_glm_lpmf; (int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lpmf</code></a></strong><code>(int y | matrix x, vector alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-8754addaa08a378fc56a498e81745574e870f6fb" class="index-entry"></a> <!-- real; poisson_log_glm_lupmf; (int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lupmf</code></a></strong><code>(int y | matrix x, vector alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-5edf084bfc5e8e7f2942c123d2485b9947832b4a" class="index-entry"></a> <!-- real; poisson_log_glm_lpmf; (array[] int y | row_vector x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, real alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-279c01365489bac4622e7dcfcc8075202555568c" class="index-entry"></a> <!-- real; poisson_log_glm_lupmf; (array[] int y | row_vector x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, real alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-83fff7e4697319fe19b71c0eecbdf5ac8049b7cb" class="index-entry"></a> <!-- real; poisson_log_glm_lpmf; (array[] int y | row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lpmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.23</i></small>
+<a id="index-entry-f817a209ecaeb8259b7e13294695ad91e1bac17b" class="index-entry"></a> <!-- real; poisson_log_glm_lupmf; (array[] int y | row_vector x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lupmf</code></a></strong><code>(array[] int y | row_vector x, vector alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-9ccc16b1cd683cb83b9f375ecf6d419b4b28ad2e" class="index-entry"></a> <!-- real; poisson_log_glm_lpmf; (array[] int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, real alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-a22ac3bb5bf0da63d28e08cceb394ea47b6f355b" class="index-entry"></a> <!-- real; poisson_log_glm_lupmf; (array[] int y | matrix x, real alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, real alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+<a id="index-entry-eba869933862786def5f0691fd7751eeb0de388f" class="index-entry"></a> <!-- real; poisson_log_glm_lpmf; (array[] int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lpmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lpmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code>.</p>
+<small><i>Available since 2.18</i></small>
+<a id="index-entry-edf1d3e6f8239fe43a077fe8e6dfdb7735799c66" class="index-entry"></a> <!-- real; poisson_log_glm_lupmf; (array[] int y | matrix x, vector alpha, vector beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#poisson_log_glm_lupmf" class="unlink" title="Jump to index entry"><code>poisson_log_glm_lupmf</code></a></strong><code>(array[] int y | matrix x, vector alpha, vector beta)</code><br> The log Poisson probability mass of <code>y</code> given the log-rate <code>alpha + x * beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.25</i></small>
+</section>
+</section>
+<section id="beta-neg-binomial" class="level2">
+<h2 class="anchored" data-anchor-id="beta-neg-binomial">Beta negative binomial distribution</h2>
+<section id="probability-mass-function-6" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-6">Probability mass function</h3>
+<p>If <span class="math inline">\(r \in \mathbb{R}^+\)</span>, <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, and <span class="math inline">\(\beta \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(n \in \mathbb{N}\)</span>, <span class="math display">\[\begin{equation*}
+\text{BetaNegBinomial}(n|r,\alpha,\beta) = \frac {\Gamma (n+r )}{n!\;\Gamma (r )}
+\frac {\mathrm {B} (\beta+n,\alpha +r )}{\mathrm {B} (\beta,\alpha )}. \end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-7" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-7">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial" class="unlink" title="Jump to index entry"><code>beta_neg_binomial</code></a></strong><code>(r,alpha,beta)</code></p>
+<p>Increment target log probability density with <code>beta_neg_binomial_lupmf(n | r, alpha, beta)</code>.</p>
+<small><i>Available since 2.36</i></small>
+<a id="index-entry-0c7465aa1beceb6e7e303af36b60e2b847fc562a" class="index-entry"></a> <!-- real; beta_neg_binomial ~; -->
+<p></p>
+</section>
+<section id="stan-functions-7" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
+<a id="index-entry-9d835cd7ed67218478d70e1db49912bf2d710a1e" class="index-entry"></a> <!-- real; beta_neg_binomial_lpmf; (ints n | reals r, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_lpmf" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_lpmf</code></a></strong><code>(ints n | reals r, reals alpha, reals beta)</code><br> The log beta negative binomial probability mass of <code>n</code> given parameters <code>r</code>, <code>alpha</code> and <code>beta</code>.</p>
+<small><i>Available since 2.36</i></small>
+<a id="index-entry-1bd0b7fec4d1f3e315d4328c1844d15641058615" class="index-entry"></a> <!-- real; beta_neg_binomial_lupmf; (ints n | reals r, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_lupmf" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_lupmf</code></a></strong><code>(ints n | reals r, reals alpha, reals beta)</code><br> The log beta negative binomial probability mass of <code>n</code> given parameters <code>r</code>, <code>alpha</code> and <code>beta</code> dropping constant additive terms.</p>
+<small><i>Available since 2.36</i></small>
+<a id="index-entry-ecce69045f7901c0a69ccc67aaae5ddda48304b2" class="index-entry"></a> <!-- real; beta_neg_binomial_cdf; (ints n | reals r, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_cdf" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_cdf</code></a></strong><code>(ints n | reals r, reals alpha, reals beta)</code><br> The beta negative binomial cumulative distribution function of <code>n</code> given parameters <code>r</code>, <code>alpha</code> and <code>beta</code>.</p>
+<small><i>Available since 2.36</i></small>
+<a id="index-entry-fbab2aa3f292996672306979423648761d0ab98f" class="index-entry"></a> <!-- real; beta_neg_binomial_lcdf; (ints n | reals r, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_lcdf" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_lcdf</code></a></strong><code>(ints n | reals r, reals alpha, reals beta)</code><br> The log of the beta negative binomial cumulative distribution function of <code>n</code> given parameters <code>r</code>, <code>alpha</code> and <code>beta</code>.</p>
+<small><i>Available since 2.36</i></small>
+<a id="index-entry-50c63a123857b7480722a4aeb3d884c95317cdc3" class="index-entry"></a> <!-- real; beta_neg_binomial_lccdf; (ints n | reals r, reals alpha, reals beta); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_lccdf" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_lccdf</code></a></strong><code>(ints n | reals r, reals alpha, reals beta)</code><br> The log of the beta negative binomial complementary cumulative distribution function of <code>n</code> given parameters <code>r</code>, <code>alpha</code> and <code>beta</code>.</p>
+<small><i>Available since 2.36</i></small>
+<a id="index-entry-8c97955cf7e95ec459f434cad32f31a963d196cd" class="index-entry"></a> <!-- R; beta_neg_binomial_rng; (reals r, reals alpha, reals beta); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_rng" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_rng</code></a></strong><code>(reals r, reals alpha, reals beta)</code><br> Generate a beta negative binomial variate with parameters <code>r</code>, <code>alpha</code> and <code>beta</code>; may only be used in transformed data and generated quantities blocks. <code>r</code> <span class="math inline">\(\cdot\)</span> <code>beta</code> <span class="math inline">\(/\)</span> (<code>alpha</code><span class="math inline">\(-1\)</span>) must be less than <span class="math inline">\(2 ^ {29}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.36</i></small>
+</section>
+</section>
+<section id="yule-simon" class="level2">
+<h2 class="anchored" data-anchor-id="yule-simon">Yule Simon distribution</h2>
+<section id="probability-mass-function-7" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-7">Probability mass function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(n \in \mathbb{N}^+=\{1,2,...\}\)</span>, <span class="math display">\[\begin{equation*}
+\text{YuleSimon}(n|\alpha) = \alpha \, \mathrm{B}(\alpha + 1, n) = \alpha \, \frac{\Gamma(n) \, \Gamma(\alpha + 1)}{\Gamma(n + \alpha + 1)}.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-8" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-8">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#yule_simon" class="unlink" title="Jump to index entry"><code>yule_simon</code></a></strong><code>(alpha)</code></p>
+<p>Increment target log probability density with <code>yule_simon_lupmf(n | alpha)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-0a6380458dc845b883a913fb2227d90dee08322f" class="index-entry"></a> <!-- real; yule_simon ~; -->
+<p></p>
+</section>
+<section id="stan-functions-8" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-8">Stan functions</h3>
+<a id="index-entry-0fa5c053a53c4eb74a8936030170ca24edf5997d" class="index-entry"></a> <!-- real; yule_simon_lpmf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lpmf" class="unlink" title="Jump to index entry"><code>yule_simon_lpmf</code></a></strong><code>(ints n | reals alpha)</code><br> The log Yule Simon probability mass of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-ab47efb6b4bbb3f4bdaa849c799ee6e10bbe6036" class="index-entry"></a> <!-- real; yule_simon_lupmf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lupmf" class="unlink" title="Jump to index entry"><code>yule_simon_lupmf</code></a></strong><code>(ints n | reals alpha)</code><br> The log Yule Simon probability mass of <code>n</code> given parameter <code>alpha</code> dropping constant additive terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-be440f4170e8f674bbbf23f2e94e8803e94cec24" class="index-entry"></a> <!-- real; yule_simon_cdf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_cdf" class="unlink" title="Jump to index entry"><code>yule_simon_cdf</code></a></strong><code>(ints n | reals alpha)</code><br> The Yule Simon cumulative distribution function of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-17da2af212f09a989d004314b9af2fe44bbe25a8" class="index-entry"></a> <!-- real; yule_simon_lcdf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lcdf" class="unlink" title="Jump to index entry"><code>yule_simon_lcdf</code></a></strong><code>(ints n | reals alpha)</code><br> The log of the Yule Simon cumulative distribution function of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-3def60eb9f4fdb6c157d57d9c12a79c1be38a69e" class="index-entry"></a> <!-- real; yule_simon_lccdf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lccdf" class="unlink" title="Jump to index entry"><code>yule_simon_lccdf</code></a></strong><code>(ints n | reals alpha)</code><br> The log of the Yule Simon complementary cumulative distribution function of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-fb10a835dac9bc498feff76a66b8bf9737013d89" class="index-entry"></a> <!-- R; yule_simon_rng; (reals alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_rng" class="unlink" title="Jump to index entry"><code>yule_simon_rng</code></a></strong><code>(reals alpha)</code><br> Generate a Yule Simon variate with parameter <code>alpha</code>; may only be used in transformed data and generated quantities blocks. <code>alpha</code> <span class="math inline">\(/\)</span> (<code>alpha</code><span class="math inline">\(-1\)</span>) must be less than <span class="math inline">\(2 ^ {29}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.39</i></small>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/bounded_discrete_distributions.html" class="pagination-link" aria-label="Bounded Discrete Distributions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Bounded Discrete Distributions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/multivariate_discrete_distributions.html" class="pagination-link" aria-label="Multivariate Discrete Distributions">
+        <span class="nav-page-text">Multivariate Discrete Distributions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/functions-reference/void_functions.html b/docs/2_39/functions-reference/void_functions.html
new file mode 100644
index 000000000..3cf06665a
--- /dev/null
+++ b/docs/2_39/functions-reference/void_functions.html
@@ -0,0 +1,1148 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Void Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/integer-valued_basic_functions.html" rel="next">
+<link href="../functions-reference/index.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Void Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Void Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#void-functions" id="toc-void-functions" class="nav-link active" data-scroll-target="#void-functions">Void Functions</a>
+  <ul class="collapse">
+  <li><a href="#print-statement" id="toc-print-statement" class="nav-link" data-scroll-target="#print-statement">Print statement</a></li>
+  <li><a href="#reject-statement" id="toc-reject-statement" class="nav-link" data-scroll-target="#reject-statement">Reject statement</a></li>
+  <li><a href="#fatal-error-statement" id="toc-fatal-error-statement" class="nav-link" data-scroll-target="#fatal-error-statement">Fatal error statement</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/void_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Built-in Functions</a></li><li class="breadcrumb-item"><a href="../functions-reference/void_functions.html">Void Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="void-functions" class="level1">
+<h1>Void Functions</h1>
+<p>Stan supports a few special statements for printing or for signaling an issue with the program.</p>
+<p>Although <code>print</code>, <code>reject</code>, and <code>fatal_error</code> appear to have the syntax of functions, they are actually special kinds of statements with slightly different form and behavior than other functions. First, they are the constructs that allow a variable number of arguments. Second, they are the the only constructs to accept string literals (e.g., <code>"hello world"</code>) as arguments. Third, they have no effect on the log density function and operate solely through side effects.</p>
+<p>The special keyword <code>void</code> is used for their return type because they behave like variadic functions with void return type, even though they are special kinds of statements built in to the language.</p>
+<section id="print-statement" class="level2">
+<h2 class="anchored" data-anchor-id="print-statement">Print statement</h2>
+<p>Printing has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).</p>
+<a id="index-entry-61af0108923d685571278dd77b2b31e38e67397a" class="index-entry"></a> <!-- void; print; (T1 x1,..., TN xN); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#print" class="unlink" title="Jump to index entry"><code>print</code></a></strong><code>(T1 x1,..., TN xN)</code><br> Print the values denoted by the arguments <code>x1</code> through <code>xN</code> on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal <code>'\n'</code>) is inserted at the end of the printed line. The types <code>T1</code> through <code>TN</code> can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).</p>
+<small><i>Available since 2.1</i></small>
+</section>
+<section id="reject-statement" class="level2">
+<h2 class="anchored" data-anchor-id="reject-statement">Reject statement</h2>
+<p>The reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).</p>
+<a id="index-entry-4724be86187ac6f6f6706c7edabfc36317b4a586" class="index-entry"></a> <!-- void; reject; (T1 x1,..., TN xN); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#reject" class="unlink" title="Jump to index entry"><code>reject</code></a></strong><code>(T1 x1,..., TN xN)</code><br> Reject the current iteration and print the values denoted by the arguments <code>x1</code> through <code>xN</code> on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal <code>'\n'</code>) is inserted at the end of the printed line. The types <code>T1</code> through <code>TN</code> can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).</p>
+<small><i>Available since 2.18</i></small>
+</section>
+<section id="fatal-error-statement" class="level2">
+<h2 class="anchored" data-anchor-id="fatal-error-statement">Fatal error statement</h2>
+<p>The fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.</p>
+<a id="index-entry-36f600d7cd1daea1b94b57949bed5e7914fd9442" class="index-entry"></a> <!-- void; fatal_error; (T1 x1,..., TN xN); -->
+<p></p>
+<p><code>void</code> <strong><a href="../functions-reference/functions_index.html#fatal_error" class="unlink" title="Jump to index entry"><code>fatal_error</code></a></strong><code>(T1 x1,..., TN xN)</code><br> Print the values denoted by the arguments <code>x1</code> through <code>xN</code> on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal <code>'\n'</code>) is inserted at the end of the printed line. The types <code>T1</code> through <code>TN</code> can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).</p>
+<small><i>Available since 2.35</i></small>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/index.html" class="pagination-link" aria-label="Stan Functions Reference">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Stan Functions Reference</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/integer-valued_basic_functions.html" class="pagination-link" aria-label="Integer-Valued Basic Functions">
+        <span class="nav-page-text">Integer-Valued Basic Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/void_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/img/logo_tm.png b/docs/2_39/img/logo_tm.png
new file mode 100644
index 0000000000000000000000000000000000000000..48c9769c7de2c7c8585b52a5fdee686aaa47c8b5
GIT binary patch
literal 21065
zcmZtt1yr0r@F)zki%W59ad$6Xq`2$iwz#{yJH@TI7I$|RDJ|~qQrz7?e*b&ld%yRb
zclYdhCdo)9n<tsfOd^#OBtId2L4tyU`XntSrUC^8(ERU*0QXUYYqKl-@c=lhNQyvJ
zj}sn#6qHOgrOo8!q3Azg1SlW?6AJo2l#ee|xFr<K|G-dC)E`eMDCj)E|830!{y(h%
z%{=J;3;%~Wr3x+c!9>dvr0Jq5FUMzWZ_8|GVsB*1>|yKh9|F`j555o3*3`w2)Wg=s
z&Y90cfc$?j_&(tOs#(ZM{|CjzT7X<rUWruH-pQ1di<y;~m0S>sl$7+FlZhFhikQUz
zCjTf2kXyL8IPkHsxVyVEyK^wxJDIbv@$&Mru(GqTvon2QFgbhLxfpsd**R1EFDCyF
zA2CyBV<$@o7fX9P(*O7x8ri$L2#}NiC+Pot{#QR;EY1F}Bs=H-t=5NvEdRBzuraf;
z{2$&Qq~HFl<x_OBH2skLAAdo%Z~ue&|El}na=x+rC;b1fGyhA||Ec{@RS@YL%l{cR
zL8Q<6Tkud&!cfv;-$5RL(`<MTLh<^C3iBkVRK~x&`-D<(jIzH0ay&dqGyJU7;XB0c
z=l+m$k8iE;lK-IkG0p;ZP=4bF7n&?B?t%%-Wm8SH5>uQPSKp5~flJFNT%#xBC+hwi
zejWTXGkUMbOUoH&M^;cWVgU%~0CHiN2q+9v;QvP<4;%0c1OYK)T!i&n6SRDCks9A3
zLfu!YAY$*`l5297u1ONMcf2q`MRD1T;o>h!KilKYUbSSem2e#W-KMr5GOmf`!kP^#
z*{PJMX-Rc;JBaIQ>v((U-jl6dt0|rkBh`eFg1|+)gv?z_e1}}%M!WnP{FlCJ7tVHy
zioa?Y+NeM<;I>GLwDHN7E8a$750ZY<n^c(2sW36-NIh>geF>K|Fnr6$V>I2MmneEa
zDB8pG9&nYY^eI6i0NbYMwl(HdPfs)5r4x(Urh(TfV~n+U4(T<MF@6>dhefX{zot`a
z;i>&qGf{35+kP9T4(<+GdmvQ8Y|&w&;M&S`Ktfy`9U%XTKwfG9T9{3{x}LCVs1+Z}
zei@}kCed4fu$M1Gb`gSLzsTTk!zbcGH_vL8%73jdFv0RkAU@W0KJN512vcD$*CF`O
z)2c8xh<@)lG*nrQX?*aM%g?j$XSVYCHetkhdmd8$)o@}VWJ(CtO4SIg5raA}sdZ{E
zP;2r%czBTdGDz-2pZLIvy7n|=<QeUitdTFHK7R82nM(7cNu#KoGUfU|Zysd7BqNM1
zT~6oB;_ve%A*!#Yq-oUXM@rX&(Me$mP=@I%cP?EC_PN{-Qq`O4+KR`<T`Gd@V0k<K
zk|KtV=&&0Xvv3$MW!^InKjN|e-m45ARuKX`sg?D_nIv31zPx~(X<o|LQDkK@!7Abm
z_$GAeNS9H<L5PH+n4Z$_Jcm1)g?oR!ahd4vk3@!xEP+K}&NbgIWR_$rJ%$%^H3nU^
z)j#kFDa-sd;Ox0VSAO~wHMEEwYZU6Wel(_GGE)M^hws(l|0&9#x3)rJuPAs{E2y;e
zyM}ibN_1)?`AO6A0FcaLc#a|y5O%_2*`+Y?zCn?;PmG+$(+)C)IT$LbN4=EHauq+8
z^G3_?Ggt>MkXUU7OO-u|vfV5ojJU}m5t1YVYtTL~U;IhDYy>jhbYZ5$6oH|4@TM!?
z?=DjhX+~eRe4Bjq^=J-6=~2{}834K$n}i9&Zq)dvBWdqOZ$tIC3Zbw+py(F!@>5F+
z{B2h8bxfyNhS=r^P`QP}lKvI%0V=jt91!w7UB2k>%GvuIEJu&K4jL%cwNh$xe+)_o
zMCp164~Z*BZ#~c7$)L*FEa~S~KV`8L1-d=V{(#>}reEF3!aYmIvsO`)LG&pVY{iwn
z_f-<pLh@I>cXNLxl%}Mn@~d^ESPvs38@Vr_`?#IBm#uow3D^?#@<Eah5wH2bE;tz;
zW%TC4W;cjRckT;}<SDw!Ssgy%Z3q!BOVacc(6P-M6xvsb_sb*7PDw+`OLfFu%Db4x
z_wVK|s8XuHpacubsA0q+W8sX*&(p&wg<EI*7Fp_?UE)aUuzHt$!zqqU?Xzj_a~BK*
zd~DrcsmM2J0>y3w({m7yv5Zc+ztsi=O*}*@H;+g`hQp@sgXXJ5-_>+x4SK4#Wv!Qa
zne=Xn$n>@Xg@FnY^26ZVB7!1<vmVBhzoJ}xm8<}jS|n&UjhB&)Zktj*!R;DvtB3jw
zw<yHTSPY`1aA7BG9%}RYl*Ig<)V*jt_dlxiDcejRMC*x(WF-4puX}|-G;ZqsdNj)x
zF=}u{ppZ0}0sPZhMvt?5ynR6_<*v>UYEdqgT6W=^UlApObgf>Boc?vKh?re}6+I&Y
zFku{~M4lh~+Ts0Mc=oBNhD+ELsf3a4K*aU_-bXvO#<H2V=cm<ymyJg<Zk?b!5}1KF
zKOa8+R3Zw!+vgEyi(>*Id@pZAI!$}&$Mo~Prj6eq&CK<uBbbiHnc>KWxEI=vE0_Uh
z6fBM<4d|VsXky(g)e@f{2^!F^7dQvkL-W|*Ui5JMOEaizvGZibhvi8H!j{a;8YKO_
zFIR>B>bVuANg}EL=HDFrK|n)TI#acz_Nm{5F*S-D_??WEIa>|whok<}<X)q3wQjF8
zWLWHK3Jp1DRT*+_-J}#?{WD&Ljf@mL3!R%oR~4&ZuE9L<w=P^bq_#W}+a_S`@uXdT
z|GX+iov1Q=7Am@6Br?DX0UQ(DeEIs|+SB8){$&oK`1^M$$kUP~!+q7)R6=P_^AY9N
ztmf7`EONbHRzM?68x?9B!}hS~9PRod)w#ZM5RXc&pUWtgHTT52>X0DsG6JHXu0q<!
z`ijnK2tsk$4+^E&-2D90L3uiFl_CROS6n19=(dhyK=Zg{@1@&~;YZU`L~OlvfUp^1
zZ?YeX0TljMxB~y!_^CepXL+ByRh%ytVOC<4u*aqv-#YBp0=6Y~N)E_aXGdgd2{4od
z0EWe}^895WbJHEv7T+oklZ%`gvrZ;UR$ILt`TY-3LDL2cIIB<MpF5RizGKvM_KItP
z6!J!7aFEnA%OyylO$w%I`}bRpcN(09)I3#1nThEZ_1lHzOnh5diZ3dVcrsJl?toDW
zWf{WQXWZYkzFnEKr3c2ney*#JINpqTw--_G_vdIHAd`(W)I^}uM9Xd`_kS*}VPMQ9
zp}uE83puFGKy+zQDJf9lq-~O?HrDxR(!nn1?Dr$XGeqWNuG;;H{sHHn%)3!}4D)hc
zA_g<7*c;-0q<J%tlQOnh34p@SVYP^a=0{cV6W{J#tqE2ylb|Pu_12broj7_)Ye~H<
zR)^2${q)IK?eoWI7@+`opodh3N*FAPy%y~1!Y->rPF6L#Mu<PRNkzXbPS{uJq+N(s
z95%2pJcRr?@UkkNM?he3bqcFhQH(8tn~}NAzzO&NeSNU9*xO)f2o#}!jYiMWYw<j4
zp;^3}Hj6iZ*f{P;V8;YZe7qjcn)OP*0+LiwBzDjP6w2yA6Gjh?Vg7WVUD5_kcR`Pi
zj+%Oymq>S!w>o}xg|lsmZqA=7p0yi(NV8_UWM>4*xPGCZE216aDi+zE7I%bpa<goP
zWk(>M`XFg~d~bWi|28jOQE$?kQi|Y@AoYNZtzLd1nH_2eGM+p%AVf}@9j0n!XJc^p
zO>Z39Hw#hJcv`#(7g~MM-cojJ$bR-9Go|@3nGc-uF=f<Rl?LBB2|0Tc)jZp>9sM+3
z0y@*w=(77n#Iu6*#@kFhS^P)L_fC95msBprdd2geoj~~@$s#dge7KjT)|1s|)|~HR
zKfhuWgWtQ&<X@`i7X?}YTbqA62FnH^<de6HLZ7N;%1V`njKt_Mo>;b%E?7%tUT`wm
z`f=$P$$zcn(&}cql`=l{=6m%jGs}nx>jNXH;b!Vvt?7=iL6zJ5RNOH5mC^2E1{id9
zx)_dloBCRxH*zM47_FeP6!OGifHae)#bJlLpv?d;hQNNV=)a9T0={n6_Dg+Rx{C~p
z`$Ba1FRq*?bhg`<Ta$*fVUrjU+&L&2+bZR+0F{EixBWsw8`KR4JDqqQhws5oDIwjv
zwcVmQdTByuBS;U;oj3)eaT%18Ujon6%MnBbFwzRG`lk#CiX5A*ODI488#2$<sQ47=
zo5pT~bFwkNuehu~-d%37x9mE1*Cliq1@+%dALM%nJz1m1TjZvZ%<_}5n1SuYb9p(_
z$ry1?I7sY&c6Vj@e+SfadPu@)nOs7n$G&lhSAukRgpl+~`~FM`{rUP@T`gtKkwwD4
z+ll3umN$*uuK*EjsdE^;Gmh0$vkizwrNao!fe8T;1s^||qoMd&^yExWCXgT!Cf>Zf
zwEB6@ZHZA<VxMqWbE7@|${?b(4kVT6M=4n$^I0L!dKCw~FF^yw4<`D2Z<E#eomyt>
zMiEP1yZa;{%<o04*HS6h7j<S=xRXaUvUY9adDwN>x4MlWoF7LlEt?p6#P8EKlVc-)
z{#YWzqZsCc4c0C^hs|mD;s`BwrMD*OX5rouBlMKTJ3$ycri|lH;e(&fgix#W+o;y*
z_5OtYOK(~ZU~c2|fA6YwGF?UcmrMbF1SN{08ub;Zw^L?Rk<8&(kqp>9CB$ukqW>lD
zNR%4A^hX?*iCbz0=`M{y<A}YIj>_8#9G}V`XmbA}VA0HrPw**MU~_l0$DyhKI}Aqb
zB<RiAe%2`fU*7(RJGtcKAu-t+|E=DOOZp^={<rcV<=jsmX&kI|vNtba!~g|KwSC#D
zN|3xk_1r(}`JmTxw5ttT+E9Wfq@3Ne%m9y<+z@CX=kR!y(@gcnf1ACJV~EgK>$674
zoMm8=$J)*O7nQg4{xb1hN@shR0Toc){Cnqjo?lVNul?WfUF~TPbnyUiT#?t}j<45O
z9jKOjdd^69WlaPuYYuYv0Nd+=6rE9i-ocaiRS1exI27f~73e%qT)i6i?_m}Zx5GLy
z6Q_<U(p{(`Qi%<-vyUtRw$&l-HDkr1+ml>YX$CiTT18@Twz-nauV~>illbxJX}hKj
z<yyrmX1S;4b!EpabxYSZn+zYkxR?Fy9e>m_I=0vs5a&`j8SYc`!>=b^n-@&$h`oz)
zuQ3k{!c!L?@!`1&K{ZqAI<&bU!+@xiKxq#5Y<gvhb|v#B$`ZjJ=x38a1bNL`|ChYY
z?5g4yD<&wa&;Hb;zf}g)k6sK>>bLy_6>=klH(?fQFrOxTnJ%1?nZDm5-PNISZ`770
zcUjWJ0Zh+fe#)FXRcW)n><-f}pD0q-f<_NvMir>DGn2EgPY-EkZ}jH=TL(jp1qJLj
zR!q!V-NF+y5D~0h$xs1|_}_KPO51Y#q`5%)rdiT6SoO|Fv&+nL^o!;TRUEzswtULd
ze=G|{n$`tqH5L|+R>@EXFx6$mNYh~)R(=MBoIE|oFge0B=-VHwoVxU1;_j%cRIj@k
zJ6uPaR+t+j^hQepVt)T+_@Ds6Op1)!mdU+B&-Z)-vEQiI&@(Mc3a&c6|H9l)dN>!8
z7Kg!fh>JjWL;#b0f-Wu9kiw=6=&!S}^O;|@wJOZ#gN=Tx7HwAJjsC*dW>sa~<XSL&
z7128v9R{t-ml0G+JC-qg9|r~RmbLOJX|9<d=qOp!vBQ|4dR8S{_szAVxn?**yqGTv
z0t8im?Jhs<MsYv<d=V&T?u)zP*_E+ePLF)2SaD^*@(qr=3{i=4m>quys4ED!>7(mN
z6;z`{4%8;={oKLD=6%L*m7u|ibT{>JWU`~1-`6?O<zz1_P70zL3V=cB<qhoD!Y2`a
zXiy;sW>ECzEkTOt{%Y|STgB_)2o)|MBNrmY@mR^~1k68&Y*wQuX=wZ~u0~|6*A$%u
zfQm<97Goby+Ig)U)>H;H_qVi%B#U%SyN5by6$W55zhQ#^YG!m^0r5<yPym7*#pI{N
z*!N{h(S%OOcL`0IPA$n&k6dc)h~e$t$z=wbW!GWst#IMcDg`2b$h0$3^qd8Vn|f<o
zlh={buU7-99AvU&P2j;_q2boW_tI@}HSa^kfK}e$!1O-_@>Rh@*;Yu*fSdfdgp(7T
zaZi0(Zh4flAR?yXx1AYkh5XcDhx15zU`=N4COIKy9%3aaR0;j8FxYM+l4&8I^c^LI
zmCRev=O{^Dm5VT?yQIn>`TR+1393cWoYW2dV>2R;5f+$P8X7bouR#>=5H*7kTf_^J
z?Umc4MPJSp&nVcrR@umd_%M{10znS=Ibw$)R0KEJnA}%FzDgGl5N+X(s6-VQGc*_v
zM+`cR;s=$sWwL;0KZ6yY0<a-1CF&D8<Z7-W$t^pg#+G%)Nt23iwFtV^ar&>6hD@@e
zDJDQ36YN0Qy~dvNL)QfH*UR@7@QN*Bg;X2;%;2lT&)-<rAE0QO%-mN1YQIk54|mv(
zPv4usB>$N*spN<J(ZQ+x!)sGJj^Xng*<-?D-34MW5dk{Ds@Bu{Qn1}K5eT`tQsF5e
zxnd<2?Et*46!4xcI?PEL;o?Y%9^tM{^iDJy1}6^@tZ(1^_3G*)#Y%vO=*sllA?SX*
zpj0Daa$#2DGXkrdtm<CsA$#(!8U{qUb|g*q`RKv9Q8ZA)ciZ*XF*R_ND`no;^Ha=+
zL8WU>;rND9eStK5VWEV5auvBRV6ZNeP`}&Rkp{kDzyL&Jez<f5v3N>O#kNcZ4fcj$
zquxc*YV-&iI==lX>rUD+l9zEp*NostiSPg>e>i|HGJCfkPM#cP!oPx6gPPI{HR7Fa
zpQ#TVw|H)FHTixdTcfeM1Ry*}fI%aX$rU~w1NW;@Zz^d$x2ApfW452<afR>QWdbp!
zqTxK;1felVGym%W0E;$uBK0?WsxewCB^oU4m*`$An`eGH4YXa~Z;Wq<=Rc)~=Ok^C
zvk|gHdSf|GahRY?nFxn779)mh9yZwef0URcUPc??o?N96sB~A&Ei}g8(#yK*v7QtK
zr4+N$haiOT0%+)`kK6*L%|O2%_~v7fa5fmU0XE~g1aYz>j}O@1?<F+`4%_>MSH3Ev
zL4&+i+}ey;VQ?k~3>ZCXHpe272<RUc!Vy0p7P$gMtGd7+P9YD4%R>CZgbGg7?*Yo{
ztv2lT#?f9?_?pQ3^-RQynr7l)=ZYZ{20qE6C)D{xkAOdn3@PeK6A4)m?Id;V`KW0M
zv1swKA##jN79guZ#q~ASAW0QZQ(B-#eIUq|ts!l>GAr3yaovxo&8E>HZAx5BI0+`C
zuW~@rbAjhDF4fWzIv9>a@cuoy-T&<{&wymvhSc?lh-G`Q=cXb+pE+@j@o8_^S3=g-
z)H^AZuAtM~QdTqPOSHFzKn{a9cPJHuRdO@J<_<66P~FtDSy<dX%|_hLuBxr%#Y%#O
z(H*0&<aX2w?ud5BiHrkEg@;e3cw*P;CWVIN50kF)LK3N{UW5q8X^I~-L(*~im^is3
zQ+zl{%20aq77N~B*GFg9hR|cl;=wl(eFkeO(Ixr93WLA){Xg3^{l7>rWXT8*xci!7
z6v`th4vK7m<npjc&IjeXJUUsdOqh|IU}&-?Go^0nc)yH}&9y&NjqF`Vr_4>i^gm~w
zFRM!ot5x$Ksw@N4nUg`WKp3wU8S&8k2B}v|J7XF#d7{0Cnkp#&yL~VBg+JwBhLxiC
z63)`Iywa`oMKgZ)(i1d%D~m6G^Von!(`DFIN`@&>OV1Ra);qpiR+c#wq^yOrkNg6g
zKEn!iwr>W#6#5i4zctLRyr#tgT0QG+)(Y3d5>3j<HihAR*dUZX($UJ1?LIP>*N!fs
zQAHcKyrNt6k}M<7O0lsCV?kAiAwv{pSi1q|WheqibI}rsPPYT$vG5AQ56Mw#3X)Uo
zOyWJ$g%4V^(`OY~z#8#Ps>a4PMolg3O2!_eI>x(eE8pxTxccOhF}noCBi>*D@wX(U
zVacyfvPat9{!YbM;YMspaq5<8I=k$L6?+o8g8Y5>*^kqw`?EJGEj8Emnzg8tLNdcm
zEVJvmJub=jo^&)A2jryG*#nwowKAdN^nx0jtz}B}$(=Od!M%n{y6m`ZoSv&U`g>ke
z<yuK?;mg*jvq7|;LM#sa7%9XaH|6ME3Q`|QYu0$_h5uMYUxYSnCYz8uG$t?s#_a*W
zfluB@nvl24!k|!{XwM+I)2IidKVCl`bT+Y}N}f~N%`TLGMyV+!CsQcRKz7bv?^79a
z03Iyy+CcSM`Kk*CQs88h5Z0gHp|Bhc{%;Do<k7<SrAp#IN^0CN8G-D-$dP}cRN1!l
zlGLSt_&kL<=-g{`(v(~V79;$euouO{P`ik@vAP|^g!RaSu|XlzFd_UP-q;_hU)cy*
zaLzZqEDf8Irfp_H-TFfwRzbYaU70MER~pp;jEf<tJN-qvuj2kku9p%4on%G2ip2%5
z_Ie6WB9Ta@^xMf%t;IYEL6M9qocFbB5H0^COWxtuQ3GCRmD<*S>C`PRRQ}7Z(GpPU
zz@r~CNJTZ@X(sO}{3gW<kf14*79jqC;ty5eSFWFDdTnl#XgE_tgf}#VU_4+%C~OV0
z87UWMQIQ|8TG9u_VtTFR#sNCOVZ8)fA0{M-Fpc(y=cR@-+T#p+kL@q~h&Kv44gdK2
z&VQYdB3*A4<sppA2$mNa*#WB(k|CM6E&eJG70Gv6rqEgv<AUs-{+7&_vJrb&ou%K*
zzt|}MhjMzVfZ)|H{XmP9=rUS(!Q)W6N5#G2!4F3k&oWEcl>gqlhq=dZu18OcXL|%m
za_TR*?Mn%s(PhaE^g}MMpQ~sD)q>mYsl`tu*~Sj`AB}m`%V&^H86jqQ?#p5Vr@JTZ
z3Ikk5^E)9PUOHF<!;TUELyPT$*yyaGRe8b4pFT2Ty*j`tz6DlD)B3;j;r+Cr!SqNx
zHyq*fH!S@4Ua(i{`@&3$0OIC6iBbr`E>ZP9DRoYW4mK0n>EFoM2Z_9TXxvA)!N045
znTlwgnL<a;SPl_k@GlVGBTjb(^IC_2vY~^_dzb;IDT-1tw4L&j0<Qqkho1#vo)d9S
zPZD6U_OE+Q$NOh@d-S<Xnn|c<WP3-*M)#hBeL-!aYq^Dm0*Ddb3NYNfF;O7RUwtA1
zVH=?%*i_``D8RbJYr)EWGlQcmEeMxOedw;h=6{rU+liVl4F$V_0qIkCQ7-<cfAN1k
zA44ry60^@3lB2@_<;_y}tvLlhVT@|&m&VC7>G0Xa7@<ALwCng0O?%&pz#X!!p-MbR
zURw0@ob4(U!C!R2P)JNAsf27`(_42($ZyWdBI=xvUh?U-DdL_iC7mps&{FMvvFOR=
zfFjp!i$eDazfd-yoqgR-ATj7O68LFh3C2V8(!c{?eE8%ub~37<7&=o%X<)7EY6-fS
z(yu(J9i&6r%sA68*+|Uq^pKixgGF}=g{nfXeEcUr-(ordJK%~{B_OFSbw9tpebL;H
zlQLxG8+B(q@LJt4a91hIx#6W6`C7{#Uabxtn4vD%kH_0d9NhK1=GyOH`W1I(VgRpG
z;QDP}3uT;O_PSL)F2Yfz?#ofQRX`UHnJW}c;Q&3|$GJckF-F=gM$hs3C3wAY;+QOm
zhwca!br);r6G+ihg-v^Y30fd~?zdyUl58gLOTe>j4yb%>NU>=OE-?`SMub#Qo2CyB
zDXmH{X#uHuZrJ&m@dx{j%w|CXjVRV#hIA;+AjNyNRV+)u+JhC9I~~p)W+XNv@P}Dz
z?@D{8dhOgP;T$xTKO#cz0&IU5al&OwMnH}8wx|pQO>MGVHgKCOctoX`rvzYJ4D5>#
z_C$^e*&CtqxqXy6PYLSUx$PVE{egP5jIv4nlOIO7a$Uu^yz)HY-YpsOt0u0mT3`@t
zXARv&hg45J-n+m!`B{y=^^0MmoS|Z`4g70R%n}yYCaspik6RfC^&oHc9#6veze2!1
zyDp0&uSyMg37L(6iDwc&b);N>cl`1Z1_$H^)fKar^sR>${yeMm`*==ITod_pK;nAE
z%HC|y0RaYrD6k)^XrxZ-igppT!#}~KFTTP+J|hv4m1n9(*0iIvZVu%mKnb*NCJBOu
zj|&>xUK1`KTLqXJ*oKyqbcddlnTzx|stQjh6r@xcmy*h3Iyi)<Cq1be^m*UPLTLW{
z^~Dj}7r<&Y0a;Wo47a0tG(e>sH3vKFIw0~4LRIAcU0#AK&fOOs9G@Vw#?)@)@ez)i
zS4?;L`491{tISTPrC4sy%jJm1mxe^u(bh4CT+gmD*6AygAwXwiL?AY$7kn&dUvw^E
zND#~AR8K+uP+^bmi@akbD|mS+UeF83ifDd8ixG%G9&jn<A?sqk>fXY1c2KqCt6hty
zAZ2i4`jbNHD4$!o$@9!e7q_C(smg}6|JoX@cezhM$+Hs^ZzY>6uN;a`R)>S|(M2HH
z|DE?H*A3Ss<|j-{ABa{sBML+TKuP*LE+3<PLDUQ1^>?8p2sYeT(Du&)fyl(Z(I+TR
z(QmnEyK>IqIB$2ChMG;LLdm#9;@L;<JN~3j2;w?dhj68<ube1ZYy{I`J4sltMDRh$
z)pH1dGjO!hY%jJ=iC%`pCH%{=<JSF`xqE|j#n76^dP058E-bgxZz1o#M_Nf%c6I>;
zL2-FNrseM(6T!P=2V7%PGf%T^$y%x~%QgXZT3(92Z!ejnKAPv^lUnXD1GT;46sy^n
zubOHNIK&1`X!ABt#9yqfd{{^$G+e&ULm=`b!g4ZvDCvHvMW_T=1IuOf<mlgku0)lJ
zYl|I`5MG@N-R`b_g<$J$w-sLK&o5^-1~q(~@9p*sTYpnF`q4U2^O~%LZi^(k2hH!o
z134zb2Ts0!Nk|`{1g>K_Xv@*@`?*&SsR6sRYg>BMd8|n2G|nyEN9eXuACgX56E-%G
zWef+)?Hz2I7IVo#uJYDuuqOu3s0iZx7W0Hsf2KVe3Dl;j3`OKgsUYmIrnEPdk~K(N
zY*Tdz8Q$DZ0yu<!hv2sgf)Ee8F)U6S`>*^@1Mh_%;g-L!4fv;1<y+kQ-f&T(eF?-p
zwTY1LpTkRz2>6X?GT6|z2=k6zxkjBWi>2GW0f=@G6}5j)U3M_MK7X&nQun)o)#O|D
zt&ua<_?n6%x^qbp|IGpaCf5sBySddLm5Mx14Y<l%!9C^O1a?dvt(ZgVQbV`gFPP%A
zKHKcpSu=C&&D~f-JFi?v&1M?&H9%M3om6b8JBc`^3PXKT<V8+|ANfjwuaH&BR8BK#
zXI|)yRj&cNV&k~&oYm5-X+I}6-j)SHF;e>LC1ZrxTX;?s($m!d%a=(ZSJxg$9)^l)
z=RvZ~Uavfr6By-&U5!%cJ^qQFuSf(qW_IApTD!pSN#=#Vc}LQSIvpPC>jwxX<AFP7
zwq4ReyBF&EJS-dA&M!X{A3!IF0X-}i>StWiDngBkO`c~4Ohm61$Ln(Sj9tpkB4%3T
z`0(Ni56=;?PYewVr1I}+FWCg@<Se2Z_`m7>QP=fM)z&j-tQxC6&`as)cA;Auy`1UA
zPs?K_lm0?}K>=8J`YZOV^y1Y^gdK-;H|qC`j-O!wnu*FDQTO0C=kd8tEF?GQmGbJh
zR#)ghvfKzsoj{r9t8jaVx;~48@F=c8>Va>Q1^xMy@t={g>ijmaZLp171(-fUW&F&a
zxxapEc%PtPuF6K$y!wOt@Pv#4j}fxTus&SY<g<Dh(7ydzou)}djcxn%$1#s1y>T8p
zky{RN;ESHDnMS*HjwJH@;jRdS>Q!QIP(BJuHQ;l=xvch9xlm=;6BdrXzWh=r+2oWG
zvSf>WNIgr7$Z|v-X&9Q33P)#K@)bH@5aB@h>Ogu^8KRx#(`tknss+I(!2B1-vsCIX
z$P{cKbO+EJ%Ql>sx&1n1*dTlxs8^ka&-JGlkqzTIpt(_5kwiR^(rsLTS@EV6%wHu;
zys5uVWp2VD)lU}G<L7}-{JbK$2~tjGV9Cr|vSCu&Bh{A_T7jVxmH58=2QhxupH=*7
z+9u9Q(IJ}X$un`eFvzx{&g4t*ZOG~DPPBxB$9BKy#o{#y(0#h~w#+5ydd#p%<aDd6
zTEj&J>8z0z(EtmAPeecnC{2x^nzS<}Ueo<Xuh{@yK%Sb8_K76Ece8^;MybtDY<Ldd
zPR8ui#>%11i|E~qpYqcexL1dk(4z1CmP4>q`H~Jxw(7(`u$Ec{b8c&^ztJzcRm@0L
zv_5rPhY=8Ly$jUAr7(m+3X14~1&Uu}qjgz~U9zm-Yf&1l3KLb-sAv!iZGmnyQVk?4
z360Qsl8YiONN!sV$8xQ|?}m6CZMpTO59OuKtK7MFa2M_H9BH*UQ%)%Wbqv)P6*w#y
ze`Rfia=1iVz4M8_M<T#T3ES64`=1gc%~xSIEy`O0=J7Dn7DyNZuP7wR3#zte|J=3H
zF9;fKblrYah_!`tR(%K!M=5!m7SVAT-QuDS)IoN0;Uj6TQAZv_s)}hUUPg-`(hkb6
z(vyo(JAFkhzw<?tUJ@mB?)yquLG&GBCbGhmQ7J$-yjOlZc-YsWR?FtKj}F!H`}A-@
z^2S|<=CM6&@KXT8X$F04yFU4%+`o`Re0W3lDqQ5~@lOgA+j5C30tu`1nnW{(|A?`S
zAPFGRCj0`Uuar!YN1p%~Hvbm!bn>=4CUlj%Lh3nx|MqsK+lvGPC4M)oa&A=pJOV4k
z#WtKFMa~){E)nB7h|wq3TDGiNIm$=)zK=&M={~SUn6`f`d^EsB{Yrv^pR3xnt7W3+
zHX5YWIU17*xd@bn_QekG7#y?7eMwO6e2UCq+U5OaS3^=eafRo|P;&$LY5H-`td8R~
z?OG^QZIjK|dCMWnE&$vp&wgX!4JRXFgst6V$z8abeJ|wp{JXWhsU}y_mpu|dcyNav
z`@wq9UW^gM^RJfJx;?D`iu_QVmnW?03}0aN>p<cYS}1^HBJjZvyIJ(u_e<%^$J$0?
zvZ_-{C<XD7&0@p<>5V8xlXP6$@~4j~_u|{)^OlaqR5u#YE$7xWqg6NwOqRmGyRwa{
zAXliuJtjfn`OE_F>UM>t%{aW-R<fFnc(?c5v4h8TIN0v2E}bSTL&o+}EFf@Y3aSNZ
z{%@O(?MaypA8K`}{0>X^mqfUTQ%_oPLI7c56$>vyhkAEZ#Q@j#fVuYoS;Pfg$C~VN
zRnT7Vs#&Z<fG?MUjWN5~7Eblp)6=Nj=f1=M1XfjFFwM!+0#fPXBNY$PU9Alb=Me~)
z#dgl5>B5^w7Nq*<`T9ee58rJah~Hp=h;O{A!04riYdXses5=%}TF^1v?_lXz?X)o9
zI=w5_NOaB(LvZ*PrXmcJ2#*>pt?`1C7~Z;M#g&(MXvBn3+uYQSOMmiuz=ZzYj5d@8
z@Kt8*NUlW`1`}Gx!wA8y0%eMg)$DDLmHu3Xb&v|K#i?9g4Mw|y9z1cxtvA&n7XMjh
zOFV$>U$g33ECG0C%?y==5@IcXbio$S?vl)4l@w9WLQFy@P7J8aj*nQMbJ#^k;d4eU
zipxOvMDJ@H#N4s_q0)j|8JiH4@+=5Wgwet>?@eZ~hmg#RAw|+~gFmIvq^d_hz`+YX
za;NM3sgf1d+gAn=f?TaR+tg}X(*jpUiDgT$^s&$reoE}>BB<7A|GVM>H#&waob)jn
zt3E0(zOGwW=dojVDy(!dc;b^gRDd0OQ^OFj&+IYWK^_EhT~AypjYtjZ&X-2f_w_Z+
zee0a|Q{=Fm?jVwSw(+$u#dbK13Qkkhs=iPu$1Z6F*-mp+)e%p8D5RovR|t*jX~v{K
z7xJy%6;KA{>e8Fs;Pb{^tdrZ-Y9AJl)CsJsrnD2e>B&Ba<v0W6I^GoUzgnZhdd{ot
z|D~+xHw!IX;qD7`Ww4}5my-+DjL|8a4Gwah=IbH7le;g!`AvHubSszcHpY(6>&`Y%
zZ0;r`pGy%CfS<W`{boI>>&XbzCLosIO?)o#Yq8Ugkv+hW2egPT-5QB&F8{f#PIFMI
zS?@tUin=e1PFtZ#P>PaqWf1ONo<e;V^22*l6oiHw|CZaSfDL9ePe7${CQulw3L2c$
z8QQDN8b*o<h{$819)QdK3GEtw-!|VqgG(LAHa9TGgR4__ENf9A0HAPUaf8L44q3&Z
zrR0N4?yzoNAq&E72+omTokF@~>zdllK*)R<F@S`~)&!%}U6&>FNrbfOVApiFZR4N3
zZjv;zq;D6!*5u7mx6S?Ja~G%HUn}jeia2w0kBu!04xnQOsKCXAR7!J@adARtx$$di
zPg!PD=gm!hXP4dTmB&AJFsMyW{u-emlhWow&#VN*<@c_4>CeO|N!xLY5I%C451g_+
z37V-XIVYngJ-mtkI3^#PT3Ux?h<ZEX8?scPBRTfaDtF=g3pE}><NKIZTdU<^x+?vA
z!vL%6@1M!sYhzRNk9z*<4OFjuzCM#FFfJ#bRCfj*#h!#eQNTl%*URabq|$p4P2{r>
zMaMIcg#*ul;dG*i5)6vN@28#rW}>f3h4BP6z_u-pS6)Kd>GRz?@!yg8>bUa?aF7<}
zm-3r%<2#8FB0h14XLmrQREDFt@X*4rx*@Z4vS@_+y9vM@tfFEdlo`dWuG)S0hrS<@
z!!kff3`G7xEm&V@wJ~oDw-<C(XFJf@<OHn-E4h4_3Ou{xh7WDX(tqIsTT18)Q1Ckj
ze>mYh3eYd|K-*Ws;+~f&@l^`|Q5LSLKH+bBMRH#J7>X5U(ry|vm2|Mi3cvdzhABE0
zxK>@<)IXskCc`CmFWN}F{7vmt8S)!nlS=>VpP5h>-4z(h7*Epe*{HNNBR8$a+d=rr
zJ|^h88f>al^itjj=#PEKaPsrMwd4V7A~udqQ^`LMP^|<Jf0X|ZbFcRJ*l;vJ-Kod#
z1L^9tHLI}S(?*SUm!T`l$oDKVM>I-x$H=Tyl!hkFe~<9EMk&7@0$huqca<UAy^ntP
zX_He9)TBma5mL|FRLR!P;$_zSM`4BGmOrb9`QHv=<<V2pVIr4;)tD_kB8Re1@=9<$
zLAAEH)Z4#2+hb!Tp#ah~;kytDU6%beHw%&IW>g$otk|{bX6~a7z`&%!)IUzv!>i2R
z2xMhi*l^*7vB>sm->XtE8Ylmq)b9IMjy4Q2&M+4&-q(n<bad<A5sF6m`S~}jZGL43
zeziaO^E{4eSkL+@ET<WyUR759-g%V|7Y2<DTUC};$!#?*_F}dF5nw)a5z=ZFHJ|7=
z3}Lt6@j_3r#h}x{S;ZxH`+f__<{{=+gPjiPL7?Uk`d60H)lO@0rQ)Y!3_sx2o*s)8
zcm^|2T}ICCPeH%@R2a_6B@!3PwAL_V4vUCiP+)kKB0JSVstwUDJ-`>x<X%lOzY^;!
zgk@lsiVUI$)%FYM$qhvcy94_sxX7&Yjydtu9Nh|NT%ltZxQH1<2BY=;8*yaZb))>6
zfAVh)oYhRaTE`aLig|50O3Y6Z1ncA2WDyUX;L|OxBTVzg!4xFT<o9^U^UmP-NO+J(
zu&vBnq40@amk_u8buF;kQ1~Np*Y_zdksZcJzO%F?(Oa!HZ^vL;>M^tnM|eLXfE#8Y
z$9M@p+n{qmOiooRk^x=bc~Dtf{NE>Gp(_N%xb|(|cwWDD5{((Eb#$=8!><;fF8ak~
z+%>aRzJFWmNO{g;y&Ax<()3&(+f)4gM#QwE`9fEQ)Du1jL>8(?ga{(mkL19i>zl+Q
ze$BVTp?3dsHHdmPxZ`vC%+!poyUC4~0?;TetPx{Jm!$+6k1IJX@i{FfVC6_5-~kIO
zi|(&tUO%MgylPgz3t=EMtqU}%vH_txkhRA?%sPPynR$`EuMuUgfw88sbva?#b&Rh0
zHK+I1cf7}ULt2{$|Gfr!_2U>@Z~{ZYQga?q*EK6z5)p9n@ej63Lx;EP2e2uM%dJrB
zE|RYHjE=Q(qEgi$YOZK^j>~@TqKU^|yZ=|~`|<V)44Hysuq*FZ=}J~%+6NnP>?HeM
zE4p{l;n0O5a@Wpg$y#7Cm?sPi9c`)jSY;*zsyC0yYZPHeC||vxhB)=`CiyWCI00L4
zoZxnkF}Kr3Z38EC0Zde3bJZH=$b4LWliTPN_vBp~a9w|Fx6G+3#BXNuQTZzeka;>^
z3<e}gN=-%Nt}Xv@SRX$%TPmoxt<=b_%*UjqXQKkCg}ZjqWDw=v%=HS=w21*UI+Hha
zX8;F^!jT9T!NQWhKSm~+QLA*Eop-gH$!ni(j?c+ifIsb5EQng{u(5QBo1+fPs}+&e
zP2^(D;H~%`KI`F4@nb0Q2d2-Wpw|R8j6W~cGp&<2_9MJX{Giw??;Xp)L=JyesV>m2
z!LHQbUj&jS1gYf}mP+j*x{x&3)>H(w7rgR(cOyM%;V_pAs^gdlP}swe)cqt8nXhod
z?^z5U$#gm2aSGnJ)g#;^1s+kOXQm=<b3Tl=tu=FOgd5`aD}IvC;DzNYw0KdA<gfoV
z_87>$=T!_YWIuAxtVkrn<u5#xvM*I3zQsED_e4eRAGz1w#1ldq6cJ72-jpLZ#pVDB
zayASfI(=wd@0<Kvlh(Vm1b!GMK<v!e#e`kExDP6rHWTd26rn@In4||_kQQ%QBrr%t
zEyl?Ww9XlMJp_(`J8Zu;YB$k#wVRH|><wwGb4^qbwRF{k#hsH1J-=IW#0ROlZzQ<@
zFeg9D3BeZBxi+@pa?BDw0YWr_)TepP<wNSQe)qQ5G(PU<xs9{Ig(<^z$tJolL6!5n
z)_7&6U10<|kb)y3gDWFBz*ji$04(ON^(NTpW#bAnv*P}hcUM3q%SRM_R$y`Q@7tm6
zl0F&+_^X%O)CLlt4x_N4XJ@hOV_$4Q!~+#De1Ui!TE6wNFhgW)aB!$E7Ka-kM$>zG
zEDfP|y#7oTQ@T_Xz0&?|=tTZeU#GTjpMe7R&O?=Uj|{jaBRqYmmya{V+>U9c>z-yU
zi)*ZV5IlDH2~P8|N29mb53x3|Qe)7r=SI_QQfH4o<iZPOK1C~Wy&r0o13M`Pi5mEY
zoHPu5z>Q}q$*iq%`RJ8D|55C-b`zwD_=f~V=j)$6hu;r2wLDPGd4}@RpMjX>6+U!J
zG_c;SQLn*=TP%<+Umr`<Jt`oCrk4^?B|OftORV>EzQrJV(AxH#s);aAQc@BE|L#sT
zDzZmE-E1tJ0}a49ecg`YIv+9+mtK#9ylI}}u;9(kH(occ59T$9k=7&f*r>bQE~BM%
z^r$o9il~SX`rPcMu=;8KR;)c=8psdn+j|=j9Y>R<p1+yOZu{GveE5og5?B0dsE^u!
zKCe@Z6ch0PMdF4p9jIE}2`jOu2Xs9AS#N|d$ODaJ(7A=Z*WQCG)ix~iINOIQU;8>s
z;p=~G86k|lVAm`VEZTxi(rn+H9E`BJT8l4;f9%zi#e@3cZwZmm<;0b0|LRWu?R^s%
z!Ik}Vt*3)6Rpw%A0FzPq($syG<+r|122}1fWx;KncfV?#K`uQ1xV&*ApV#yYlfL_z
z5XaVW<7dZSL@XDif5d6?-=Jy-k-%deu_)a~`Vl<8eEbcVVhHfU(r^g*(({%hj%zT?
zf6FpC!tgxE+cN83AKf*ou=J4<a3Db{NI~$Mu@jiVFp~F@shD3}-npW6Pco4sh^hYV
zD@pzIYDY*F1NGifPS!WcDRI}Km7Xj4wy2C(mD;)N7wIe#3Rnik4P`nE8B#WHbV_}v
zY=&OB!+lt=-61c3N>&ES#A+HmUJgee1yv>%*f?@%#zh1Uc#7*WpKJ8;h1{X)$g3`|
zY7m9(4J5OG4j^}+2h@qkc@Tn5UI+_4m&o|2aZ{Oj35eUh=+jZU;OiJ@VLhU@QL~Uh
zzwM{I?3nB=Vm4ZQBIg#2UhBbn747;Ai}ObRkbSG-udo2*lSu)ZWLzKqW>)Zt2fm#E
z@ZEWhG9r%cX7u16g1lW@Ao)&y41TlijzBkq23e3l5-y94<Avnq5A4c3ssHgx^qK!{
z6NReIOoy?Yn(1ObB|SUUTovZf(TBdKFwR`+*#ao1O!(F6U3K|euixF^uTS4*uSb3)
zTg3HN+63$~L$%uDitZihd^YKN`aQC+ik~sLs#&tKdF-n9_}UB>2WUCkxZ1+6E~?>o
zeROzJAaJTRb7V{vUry@&E#r(g)Oz0B)TGPJocw-&7GMA3ZjIN?f8>)CM7`x&NT%0r
zKw9p<OFkabc|u;=dO`^hUg%fNiuJ14I7dUkW0xCrLGMro>e`iX)a^ucp3aYjt_=U8
zO4u-0ele1dZaRpP68YNs@y0@4@}c;8;P(}984JwTZZJxEneBK+TxmJs6e{&o!=ZWJ
zN&*-)m8<mkzd-pd`(qr)kfX1|use?J3{re%Bm{sFl<M3P=y^T@)zP6U=gKi%=oU)=
zhDFgIYz*5sM2eaxZ{CzJ%Lwf`2&D<VQiR5O?5U8?uZ~>2o92VLI<yjcLiCuh(+|fJ
zpxswWH|A1X7eNw>GQfMT-(~MoHZ&I4IWK7P_9M)M5YrFN@&A?vQN6~`;uO~3qsRC4
z4ifc5rqiH0PaE}UW_oYA&Va6>03)j-Os<-}e}aA$Il2<=2(DhB2N>q(EagI2hJXko
zYJVI5)g~6JE(HnucI-?z#mApQ_MUq7(+I7@@;|))rs#kw^oYShb?0#QRu@X)9@2p{
z{`rBe>K|Wu@^)Vnum_KX+I{WXHS4}u2z-R;MG%q_gkltE1czi%8UZc>7znU|7<4Gi
zebN=zZ~%O7G#t_j59>)zhE0R6O3Y9LLuCmMZPY#bKYz?1QBbV^nKm)w6K6YRp$0-q
z(d01!7f|R}>d0Xn>C-F^KkS<wT0NNpaP%5!=|oAj;nnS2i)^08Ufg)C;>Ut;z6k9#
z1!=IM-@iH`NPP6X3BA`g5j^D(fE)&(%-C&^_Oab)Hwqh0Ingj^drTCf3VYo#xF0Um
z74N#D(_}nRYP#Gi#UBxPKqD9kr#84y)+gU20$eWo<7!uU@^tfku2SP&tL#kLoYf`&
zcic|#Uu2#9oI7|RoAdO-n^!jF_59)A0#^<1v>2py?}U<`!2<@4YP)C-{I6aeOAb=e
zQj>8N2mJQRNw?U|1MB1`46O9uPwf;BOm=)~s{gRiF3P!Hv%UV-EuJMi`rNye9r9zu
zCj@w$<Z^^`RL<p>g6C2*MxoA$NcYN3dJtqAex<x?pYfTrfI|aXI<;y~9R!3F?F`Gl
zk&}LeN|C{QXGPiE*kCiRq>nr>-7D7=3S4Z)4Q7<IYKog`n7~n<ZtJ~iPXYHNlv?&B
zCQY)q!6RL2w&o3K|8{^InjSL^z)(J>e=?x2$%;#&9%(Nmh5PiW5oKq<4|&q^nwonL
zF+{jTG;z%f=Ze+{5*1|cCGcL6D3MMjs7emZr(|XPyy*if-~VJpO0Hcd5J?u;bN)|B
zv)oO(U$on=uJ*la<?1)X)E99^O)?8(f*#naMUp>JY6UB$qumGy3-tM*>o6!3DW%ca
z16h>Z@#u}=d1FHTB}DA7NgAg>(`N0H5g-HPdM;|+bD=x&eeF^`{a>3GpD4^nZb$Nw
z4N5gRc7>*8aWQBW>Y~q`FzO;(<pROvcl{cJ{YZ8J=tq=Rt;i;I$BqPCyTDiM3edfB
zn^x<a!3hc}cEx;h`GEji<ZaC>O7qS<%S|KR>a*4hEVmB0EU}6HjXTb-PG^ec&WDQN
zyQw5D0HarX^d&c1MomuJk`1vY;2sTV^SgHmwFFlu&-dr$#ABvE%b2Q5j+8R8y36;>
zPj;*ad9(kN>_p_B_2iijqI+v8;mpx?9mb*H@clI#q%|wg&*0ICfhKif)dW+rXc?n-
zIw`5bj>zg?!G}Uu8YGp$vMF{bDI|h01JkpY_^nSUF?C+l%vw0=73xX&<SSZ(TNE)n
zPrf*G>YO*mZiF`e?w0GGG9v}8unb=uTJ_zXV<4V};XjOTKGx1&0RUF@JKph|`4*Qm
zxreJt^0SXNFMWajDbKhe4k|3T(nN-yjMNAf)O2i*5XSSp4lFKe=VLpPB$=6oZOMS&
zRi4kMQaB~^s+_MNwnxA4D<EG?DRQUcIGWdn9$---4$MD6YH>9ml4V%$ay_%PQylhT
zrNym`f7JZ^TodBbT0AInX=HjPDvv|+HWj+_UddSZnKXRsKmn)G-xF5)=V_m|0Iy|J
zPil}6Lj+%5fg^m>T^RKWGyrG2fAD)(ON}4X`|l`fkAC<^T5EjtgSP`AX!G}(Nvlt3
zlrI|GdByFY)+#NeE1tYL1ya}Qj+$5Tf~hQ(a4IckIcu;9Y3=ungXV<GX}-e1ZFb{w
zTuk{(;3}vWS8%6C3AvPiWbqPs0kD&uN-b_nVA`p%VJ_lc7vV>o!iSrPVz_M)meIeb
zv7%O7<9S7%7CMCKiS_O1X_oD9ObX%P9Lf2-EL>IgbPz+T;=%CrbBa?c0%8?at7Iy-
zY=+Wq%f;5gWxBBOc{;TP(>m*;{`U?7x^#l7)6@XR9Tdo7nJNe1mnnN%zdtMYJ2G68
z`tgK}s{v7)25fMfit8qc-`jAp=XEmuhHBeK#%Cxg|5)~Wf@xGY0!^pGEe}-{I=Y>z
zQA3YqDX3(Vfr_5;eIjZbGoiA(eDgvJwioT7!K%b{J?aWoV(Y_};5BU4!-{EFsG%_F
ziLX$%X)mjn_j+{+Jm~O3Ku9ZdJU9`;VFDSJMJ>I@udyN-VT`Ev2!}qw(yQyDi!xMV
zn-E16G;9~~zjwKuZHOqlOZl@bbY{HEdm@h_i1(K%d?d0g9wz(Iv=rl}TE#$o1*=GK
zV&6AFKp(xoTTt|B?4YLg!m~Za=s;hR)l`1AMdKR1=@65Myo#$~xV7yA3v(@+s*58L
z%$s|Uoj4?%lrp7KZ3L-PG|IYiEBi6&(5G9(Re$MJz4O<EkZMF+Ko*w6SC=DZq!cHd
zQ<O4T2AI~j+o0!I&a09EbD3c*^tjhTRmIr+sJ~?x4~ry6Ol2unu0-Q6{?WKc3l?Q(
z)xWD^l>Vwj5dyU?UlRN`_|EV1Z5;o6%$t{{nt$i~>%bimL~`9@++k-a@_x>@)!PFh
zK>zbkk*C!HLQ9OhbJljHkDYmEM<_Uv+Ur6iwz*waAu^0z)T+PsK9v#iaHE{w&&n2e
zD99qX8b+h9RD{@dgU?rXfispq`H1pHUjA#OHR#e{x#dFeANtn2n;M7DR;B-^@Yvk>
zw<M=q4AVuE1=-D$#wC_F)`4L)D7CSgvm=~GG*BbNp#N7|@FIF0|Ixt5wT$f;AqAz|
z{>X9%`7vBnZz~1**>hGHFw>9XFl&wHwQ4`d>D&z+zI#PKJkp)ea^>0rd5tcW3tLl$
zTxR3vzWP$y#_Pet?hEnk7$h~Gmx^)<(;_<vDZMwF$FRbU^qSh3p5gl+t`rQ1bet68
zPePU2Vcs@uzoJ&e)Z%uVa0P|d|BPr?$w6l_z@LxLZVk-hXQg8fyC4Z(#eF1C`leff
zF!p*%1N#+xb~k>+y}=*zEIQbvU}79Z3&#Zhv>t7SwNkDY(Vu&jtMN3QDTq=(#8P{C
zy_qXS%E5!EgoM%)KMwL(xuic8B8BYXvx>*$pZ%76TRV`sz2+$o3Mf0Mr3k1Ct<m@O
zUhlB7N<YG`3~=7DhOHs8-R7=9OrEYs@CRXxz3-}k;6596Mrc-MVa7ch`Sx7MSDh+l
zFsG25!8{E%psjas-a9olsj7*>=sEChAK8z&6z!@t_4QU8Z{0D!C=VL<Xom4ICGD_-
za8ut(jhK2Oezv0xDc{c(s0_MCU6I(iJ6o&-Cv`SajmI8~GK4-at64&jl<|5xJ>%#4
zqWDkIxzL96&Q$dPs(p^1%N_m+1tGZ#X_q`ZzJTPErYL%zNl1u3Wd$e?5s`i_`M2Tw
zmeVVn)B`g|uQ@eap}+1l7U{OXqD>TsbEevdr&gJc2~in%Q{75kW-ZE9`lW|kUaL+A
zRBOGh_gYE9*z3G~<!AQz|J8Ax(Qr6X8&`uBMDHac#Ol$@#_B@!)uToiB^HZBThVI>
z5+$tOgJ311too1MiC&_Y)oXUWoA>kkex7^g%$#%YoSA3t^E0ldsjxr+Tuj@u?hMt)
zi5w&9!Sc_$RoFBApX$fym#L(6dNEfpiO{1A-uvOUqnF&Me^w4gv;J68B=~&1U!9KJ
ztnhKkuT-lrt?iI)*`0!a1zOrUWsbT9p|P`y?CEYYLl-d6A)|D`9%TL3u@T_JP0<Sv
z?}5}m7C>r2NsEY&<MYH#HAlmfE3{|!@x&ZRa$LV1uI3h)?4g^WZCgy?1kM(1AcGbA
zf==}MstrmUk81&PLKm5xxx6d+dp!s`m7Cae4DP!yi?o{sG2|9aO?<h5fad*3<E5(t
z2W-T%@=473wG2vikQ}d+C^Z|pF?g=<traEp(L*{3!F$hNi(A1n%4>Cd*4xZNfEfs!
z<-v_N|5eAU@Da=vHqdfP&^hxp&PSuX!fHL#UROoG^q?~M;|qKLP?nn_U5#R(pmDWV
zB6k40SFa0(;y&fndP%Sg!v5@O+``vCI;xHW%3Pof#oFDbKs|oLW`amJEmYiv4W@sy
z6ln1~*ykHlR0P$;8)?XD8j!c5`vfAyi+F)6xlcok>;D?};39P;c{J3cPjMHwi~j!N
z@7j%%_lEe*@EQ49ry)&8@r3w7_5}a|KXs`HW|v~Ou}G~a6Iw6n{fK4`84-RK34n7=
zT2H;IUAsGQP-0CYgn(JBb^|l-b2>@q+>za;zoAvrMg-%TyyKqud_ph$J!i>u>2Pb}
z&GnA&*>jNW7eIcaN6DVNo<(6#nSP`raG|(+dYmt<bDh^}>YujvH2kBBf7zm&d6DD>
zP}FrCP7famM2HHjIV--LL5*U4^He25>)8BwlZuB`wT<7OPbkw=oDf-*tcl&l1Pu1U
z5c;mM<qSc|4j<FN_H>nd-X~(ju#MHh@QYL@`YLcRDxOumK*@Armh9MSyQlEZA>rBM
zs!4Oql%mo`R#+z)pV$s0iq-%mG*Od*6YpXK379Tt?Kg*G^OoK!eK4`CaU~YWx49VX
zB^#@|IS{W8NUb%DWKFktRUFq`;<#-&Cwa`no2T_=`N>>@UZo;uswF=GFZ=9ys)RFU
z5vpiZk!1U^GLTq+>o`~-vHH2I`_Tm47JOPlS;xVddG&+iIG*LRJGa&ole%D>Q5;O9
zKr5=bz+4)k=6?7EIR+K>$RQ&q7SJ~}sY#4o7h^qI|Bg}@aXLs|+i%s+@08S1IOOa?
z)5R*zf|Sl(bHmIuQ<w%quUE$dD;t&=9)`|7F&q07^(dS6+(3{sTT<nnDK}ZD;jznd
z@r#}<A4r+KAnv^6BEZtxHYb}{(6u;D9YAeP+v)tQm`wmC`5xthF9;78;yC-_(cwI>
z<ZCsVLcI)EF366>KaoIR^%rk5x=a4_%TAz4i$<p|Dz3&w1S2tESL@N0{HXPSg?dtT
zpW^XCfy>~erQ643QX*aO6GVvEK2jia1N-=Be9~Ci3}_bKTrJn0H-VBOjE*gRgtpKH
z`ip=BO~9)yOe)2{|4uv~gK_5z;;&XFwPoBj`j{^j_D|Dgjdz)#0698+O?=w-_M&bm
zKW(r``?U}NC05E!xF75gy?j+8_P5O#xdCFW5$v=;#p$-3qz`3IBD+XT_$FMgGQ3;Q
z%d9@Uz>DBN?q>2|kJg6XF%PqKTBG8ITVD5+oXQW5Z+g#x#d5r~+$;lVN3~ZHlHu;t
zf&;g(0Hnm=LPOAA)qu@{!qmP)9@2uV0{PB23C9iDAZ7AhiJ6cUv4+kN?|rOP13lJw
zPdp!aTfcRxA?thOVT7C|w%Xdx#l<?3#mo}OZ4O$6UJ_?nV|U4mV+?Bv1`8Mg;yS5N
z=U~1*Apk!${=e4SWA*7-&!)WAbJR6Sgh8IgZcKA*o*!%zBAHcO=Q`#M7Rm+tJP7yZ
zO@YgIG-VeXd?<OTmZzQrr%zl83e`1IY4jWq3+TWT`)6v=j{{rWe`m4CEFZWRA<M+^
z?8LB}f;wlZ=kC}EmFEqcQk2bASToNVj%uR~2a^`{c34(c`Y(_UWNPehqKhC$ls}FW
zJKgmfW|t-A%}0?$RDbT3xW|UEkZEqO$0=E2Pce?)^~wy|Z2MIK7aDVnpccEQGdoBG
ztEs)HeAq5G!Kx~k1(nnNIKQ1J(U=XOMtchUK>c{XPul4$e!`<?h`7&R{>4A7<p_?a
z>qI}(babltb{noGQm>(!5tH-hWb#|yol{8SvqrnM4w(RRA_UG)6%z&{dTKwrTd>=E
zjC*QtsJZ=ta3}`1`1JHvW(nSe`pz}4mPVX~<D4x<q}y-aW)h|OqN9;p&qDjjZq~TL
zfTm2=)syxOs;&oOdDFglhkDxoTrh`3YYC$(c(*?jVM14Jw{LF#isQ3zUwu|f)~h`H
zPoTVVoPs%}qf2g0FF9%gC!dkv`=WsRwL4L1J_qb{=C!S-*E{=;BAYhzb`hpx+>FHW
z*Rg=$7zqMi(Civd+v{_y(@v*l!N<2LbX8Qok(sD4RZO3;$GZ}$OaojhGZ3W~<d>lc
z$o&`cOlcS>0LqVWfC2pL8&m0Jn<@NcOBXvh=h-$*HT45>5<{Kgt)j^ik|oRfZiM8u
z3cm=ync|Mk*?7qZUS5wXk7z^)gHk3n{uHgI1d5Ct_OfZG=Gvy*+Ke8cnf5ZozrSxl
z_(^IitNKN}b8;+KX9$A|F4~x8RNQ+_bvo^FBZZ3Q_nT!k2h`2TVX1(#V!C3LjYL(&
zK^VUxdl8i?xRA)Oa8=Vv{^d^$>L-T91Mb$%ODyopR4}l7WmhWuG+l}{KE2Ki7)MGJ
z8TP#IwV3)9T5@WhGw~0pnNKG!lut*{eb^}SIU!B1SElr-nyxT(6~acXAA?KT+p2QH
znuQavXQZ;RszY6oE&c9=njT|Lsq|5(q&b%*0$66o{XjHa^$~>7S^8_MJS?~)(f}Vs
zI?7{8#dE@D2D3RV<(xWX`ouAWj*$enC=?s=tAaTQm`p}fS%yNEElOFz{l*+6hguKG
zr`hR0CtDp|_0BdpEar_B?{UNvo;_vM>!HT~iffpjrlYkqcK^0ehxww`Xl<j(jKL+)
zw8(3$mBiNXPJGNOw!12qa9Qe>=%~=`Z6GsgKCI*N+$m(S^m3$-LeX($M~Aw|wiUqx
z{Sr9hkwsmaY2DSaMC&M=759h$MC#DpflwD%u<+Jfy5Qj32`j*na)5Ej%g)m;zZK+V
zErNdhNYq1v_;Mc;w9$g&8;d4}(c6x<3{EWaUPQ1?I=%=cww&k$|Gt{xwd#$w8Fka<
zN4WK|r*m<(9R2<aT`ySqP3zI1|M0)<uKFK5w2--jztVe$xtBPC!EWE5A!jyg)G46>
z83dt$@E&1F`pRQloAxFo?OY#4^`)kCeTkZjc#pj_|BuCAq|lbePL2o!*>#l<(M*3&
zkKbSxFIT;fpt#(uIV`pt6qQ30E41IY5t-*%%v<f~rM0@&DTJy+M?r}I=}air@^%%@
z*D^fTg2rv~X86UifX|mDA9c5c)4Cs&GRr%vZ{al5M_}**8*|<MX^{9|r-9b*d2j!_
zn6T?A3xpA@hi=>#{ve~#(;pR=M#*RUTE@9>(Q}TxXV3C}Ary^*>OqG$e^Yr0UzOwC
ztkB}$9N)YbX#PAB?4~lEZ-&X+9rUKXcOylN^smqvB2pF2=zcOgJ9`*lyBBdv74%3T
z37O<8cP-qYn>kOlq3Fq!ek&*HPirl(*5b4xKxspfBg;5X<$28}w1_;q4U}Tl?alZj
zz3!~_j3z(w-ZKJnr;CG25t-wQ@t=+ixxzy79N14T_KB#fyj?@ia3(@`?byinr-Tkw
zZ6L1?$bRF7?$wo<70X!6=zhDu*XIL$Ea|()`Nr81eMU@~hxA(>>Jy7=En^+R50?3_
zb;IbY(1Y-ho@sgL+)}F~dbP1}e!@5PG_}FyiVaC$s!5`Rjva;k$Fz=Fn}w;j`hNdz
zt%W^m`bk25*k5^R&T?<!C`)fF%x^1d=)k{=F<>bnxsj}<{_-6+b`-Xn0&h-A<s@=3
z{=hk^?EmXzV^}#XxfYwe`PatIshy|Ku#!DPGV7s$!&8HvIO-V5dXvJXYkz3zp$J-N
zh@YZ?8<MjP8w%=X6HOYvCErRR9rS;O^^k`4Vn?fMKuirTFOlf8F%le`CpfIQq{{M6
zycDo;{ViYvWF*sKQ*}x;#jZd?U!$bslPcIoJ*vI%fu`%3*Z%uED;s2R+D|(|b0MIi
z*mRIqrv^fa>savq%~QNS2iP8a6L3CRRyE|FgoGbJl(Q`bevY4E@ojTC`x~KSwun7P
zcQ2E0@|0|*<fm)Fmb9Hop!lzIaeOhmbdEk4GHdnwX43ZMa=VxR-iWg!rh%5nIXVAG
z1{)+f-5HugB)>hn6G=XURvSt{LD;whwBD8ck;h?n+Qf|x=(9t_ZB!3#q62cz|7?x9
zj%}*Fk=x2`%oUyA<r*;2Vdv{LHhM9<MoIE^%6_c%0&NN1rOQn2a&}3qdC*pqp%NLb
z`*p;}7D^pN`LSS1?nO=)B0_ug2RmT!`u9%YP793aKb--2-K=3r%toV_c}DX<K>-_5
zB75~kJbQt}B3B}SGx8yUk?4w6W#rvA)b7<KBbA^Fsi#qfcGQ@B6T-O><u7{8hS`p?
zDJ1yhB@aoNe;jQGSsHL@Y%=?f5xF;t?pb$#9ewJxC#Q#Q`&a0!B12o)jsEl>w2rPu
z^|U$KqjKgiXNEQ8Wzr;;Mbo9Y_~T~s-KTCUqyj1>d0ppW$f0ePrTeknCDhPpdB4=7
z`&ZUm4$|3FTjZRxMyVl#3*W|!_b3#e{64$juk~`D=V#uIpTSjsP$E!4d^p7St&eh!
zhn-===?`A<oGGWC#_f-Pc^^2xFEJy$t7%NbkW23L;T~(ow^pB&8RH%j@XfS_XH*>l
z1IVNC{Gakc3NJezu=#=rWo+QaU+dr8vU>+qo~SSa8jUD=QsKrbB4wKQR<Xs>zehS;
zE*bN`IZ#GH*c@SaIvL-09*HGox%f=0?-Yw<&*i#T<uBKtV>=eh7a+wQ2%Fveg{+zO
zDCm(Sy-kGompv|TS4Lo@<aAfI>AIK}J`Ij2#?|57{%-iB^NB`*N054sC!#wSIw-|7
z<>u(R&B|<WNtvX567Z-pUP^{){@_ThRn0<#x%g`jcZzpANm($KHl~a-_U_Ym{~-~>
zvzFmWPw@nC=AvKaWA7;M0Y1;!MLevY<}}6g8>vE#j0G)GJAX{t4MV#%)q5P6T(q_&
z9{l^Tu0n05{P&Ot43K(%)K%?<2(V=D#ctn93;ycnRb{?*NDz>l{AJDnTp45_*VB(!
ztyFQwJ^{JMjoaOa^u~Kgxl&<e_q%?z(PWRkJevZ=TXa@?m4OfsaT7``JgPTY&2}HR
zG(3>{0(`DO-z+XPMugD`Vk&zjNWPw2A{N_&C!A6%_t50)x*R0vUILz@vNXxBH3#0G
zQJ;MW6V|(!-g8z3v6~%GZi?=DZ&~-U>+<b&*YC0E`WH<68nBczi1l$6A_>l2>ev<X
zH^MMd<t#FPU<W7i{$D)4|DV+Nzaym7;dfNZPC%~4;k{8Dml5x|nyzY%vMv0709UFk
Au>b%7

literal 0
HcmV?d00001

diff --git a/docs/2_39/img/ulam-fermiac.png b/docs/2_39/img/ulam-fermiac.png
new file mode 100644
index 0000000000000000000000000000000000000000..6eecc6afebc0cc5b72e92925b9ab208b7a6b1af9
GIT binary patch
literal 370448
zcmV)AK*Ya^P)<h;3K|Lk000e1NJLTq00JQZ00JBc1^@s6AoG6o0003QX+uL$Nkc;*
zP;zf(X>4Tx04R}TU|>>6EGWof5@2A+%_}Jia(7aQh>TKTf5^bbAju%cAkUzmm|R@o
z7!csY00b}^6@0tKzzCu*>w?U}l*bFWQc{bPf%G3B&Q2~UDq&z?%mA{L%S#GC>>eOn
zBq<xrUI1bj0UZcrZvnB>A?y=iH4ye4u$qt{XCV6nkj<9_W;228;!B6H1wd@5y&xdy
zn3A8AssQr6hM}dUg(iu9AP7oQD@s7LbACZ(QD%BZiGm|Aq*4_;^OE&+6pRcE%^4U#
z;SN##IS?G4jACaHVhDf`n<K%%;J=iCf#nZE%%qorVSx?<1OHTnn3^F2gMdB*!@fDm
zrA1{>V;Gr?85juh0cPMze1^tA=l}o!32;bRa{vGy!T<ms!U1m3bv^(9KmbWZK~#7F
z?7d5y+_<u?S67v;t?t$9wkO*odvMOo|Np=Eu?NTYbN2RHySvqr>gw<PA_4@NWF|AS
zRFYN~s+a%*aYFzEf*?qq{%`;5e|tDt5l+E5J3Dc_0)L8fdUAGha;oFVaC$n{V3b!G
zrN)@x^<p0$qRJIl*|z){ys1C24sd<k-=Dy(XwQU0sCcAbhi4>s03(w+8J<7LL++tr
zD*I4A`_etgCikk(v>i&HiW}dqj?iF_gK90JbNX1vugO@EK(7tmR?sz+hjS;ZDcvVj
z8TSx3vWlf6XOUG*J&8j0Vdn?5)!&*fBqXzXSm_ky^xYBW(V3`9)~jp1kb@$j$^FB9
zq#uMUsb?3q#!>Q1_L<vWtgvUt(K<*4^~#hT_C7y9%-*m*J))o47TDpya32*uh<c-s
zXS)P5`URFF9(Rz+pN+xmb)XBJ^0S^~j46Aa-i<-4bDRAVQuj#eIaO<9lT9IIi)p-x
znbn=xat?0^){jRvUGgf~(Dm>>&p7O|CFWb;K7S6J?~!?$HthAZZs?z7CH|rk&+!+s
zz+}LnaR~m+vV!#*1ePEZ`X1+-P0#4;PTj#{s8jW-5|pKFeBx!giuZ{d&wYpz_i$QK
znivNaX8ZtCoXHf?@JLYYIBKA5SnCrhcT6Cm@D?y8mMTS=kXew&iIHSQ8ap>BIZtd(
z1cuP6c|9pk=o*8S3A;0h#{V?URFAa9J#Zar6Lc$VlkE9pnFs0}?$}tz_2NVSgLTh~
zq}Bjwh3plL?P)n9rC1m&`f|bcjzh}cWR2iZn~A~n0$3d|+OX@M$jOd5l{C^kZ83{e
z9XoZN;w1QKqH43Jk#?|@XD8%f)2$<%cEX417?aN4;IQq=ru6HietoQw-U}z~)WIs6
z(FwnARghISeo9&<BoEv;6H$Z7c(kF-pei-5V*_mk0GVU7L3R|N(-bzQI5s!cVUP>w
zS=mBiBIvL74$)#tAGRoiP;b&W9u`RQXt|BoAqu;#4<a-XHaj#x8P-S+)p4?cigO!}
zCh$T(ME+6d^&tW~nyOcUnl4$Px^j2<w30*TjpAamfY*30;T8M~$J09RBW~13_8~j~
z_t^<qVKg@;9j-r$jw=gw(tb4q{5lGaygO%h+`{i+-ALW&p*tkNx5+$&!4WjGI#<2L
zH|iYvjb=}9CMWzsYNV{ErA7C#L{~?kxT=gYY;%lyV+r#-twz84&^5?-RRsD{x<i^1
zvI9n)qFz0poGt}=V3h(<I(!X!dY|}*-x6smA={Zwc9*P`7&~I%VH+#lIP9P3`R_}8
z46S4xL*8g7op3ZBApu0SRl`rY6SDO+hb4EQdWP$%kHSP`FV;PticB0Vgf!>N=f!|i
z3cv^Y;zycW=H{9FNIMVS!3GmC;h6DIvP;ikw%_kDwEIYUc0$%zi;k2HO2f=GPV2nY
z>8A0d!#DgX9DRD)O0Us#igQc{p7Po^-|4+!FUt6@Gj(6yTkw&!A#V%*wCEVx&FTcV
zS{XrBLRxAn|FQCA{-F!7;xN{*$_n=w*}(%DaOtgBHAc45&PE#CI*yTDw1e!B8StUr
z!-`1AO=XJwiQ`_!Jw+cuT9T$L7RRG7YLOi`DUSNZz;ac}dQ7!mMHxtwdW@Yep{TrI
z#dUnP6f|%vH`1tN)QOexF^<Th4yC}B*t(TYTkvL!q<bzP&K5r=@YxAjVNDvdH7Jdm
z4GXL~Np;}DrQ`MGMN^(FWrU@ZKT5yv->PG)#nK6}1aE}jN201xs}0k9)uP}v36ig)
zaQ2A@ew$QP7UN;HNc7q&b5E_zCX3aOV8Ntcat6{fjc5#NtcY;Zc>QA&Z6MeqRp(mR
z6Zs<UcyG4Hxc+;H>QFlY9dv!NMvnCqkAgXEo0>3-d6m}$n()Wawpf>_r;2A=DwU~m
z+x062m+D?LQhb(YR;bDz!)8a>2z{$g2Qhk@8az88YZ^xlNynNVb|BPY-RXo=P3A#3
z8{cJ}4Dx-4pFIf*|ELaBOulm5*`e8IqSDkxg*QR_4zkpQNst<_L3QXG+3JQnE!3!Y
z)jnt+qmt&RAKP(qIa6IK$Jx}EdnLz7F23P_%rNuDK<we0F?kNKl$|CyLVQR|MvA6R
zke|w+f~`G5W}MXr3J#%%5r4%FK2>|FN|hU4jXHFe!|L0pihBd>kq~X^vj^22YH2aO
z00g&{RmuSUlntbV$Ly&tl>3P(#d<2G;m;}XM6jyMq2i`}<4``k=$@UBT^6F84oBxB
znMZZtYUs)rE@ZQzI&XFALRx}X`F9qVUFOu@RyK#f7cARqFPeV=<eGS=kHmfHsztYr
z79G-aEnGud;WB7*TJdwNmN3Bib5uW#Y2Ahzt%wZB4eyVfs3xdZn%bgFTm5WWz9#?F
zeY4kTSN(e2)#>Q)yB&_gTg|vboX!(n(XM2iTUGO2*fZrv$e!W02zqux7EKccb@G*S
z5za<6Y@V{&$?%Z-ZFX}f^c)OHa9k=2P4@e>hWaT@i`p_CQ&2JP;@OD>U69)pno+Lf
z(G1e?`@-ckO|d6QFe{rJM7*repiR%XPj91LTLwRls>E8l#1rYZak8|X<4#+<y<1`2
zcG=K?#TMc4o(WLL=TS-6pjru_0@NY-OxU%J!G6hw3bi_9l%)lrtV>2s2WY&oq?o!?
zx#7SXW+W)DvPsdmmo3}PO+2ma7`4BmHPD}9u5P0$Rf_`Loj&EgGsdiz?`d45_K&~F
z)$b-=l}zLHm=j7=(lTTA^hx3FtmPgod^i6d<Kooop*?6mO`Vo>az%wOt#BI@%o2ot
zik_Eq6f`@T2e_`op?g2dIzNF$wv~{Lpw4?<ow#6FBKe=|>6mCR{U#n<+ERR@84}<d
zhZvd+PIdWS=fYVXi;rG)W<kCJk-pZ`3^EO@hf}A_pP!$PbXzSKM9J>!RkW>YEm0*i
zG3EqZO2d=%V6H(%$M4>BMU;HAo|$try(%+n@wq&|?9n2PUABbQ)_HHi`yYgViZcBB
zv4CVC{r%!TvK}Nr)FIq^@DB*yoxk<mqCGmsOu}8BF@Qh6iV%2T2wCAy)stQ7rCojR
zD)dewZwfP?J2?V4>koIq%%<hXMCi3i4QvEW9QR$(`h|WS!ri@IYK4rnXRF`Exp2??
zLOYy^us3z!CXsaDA3Ak3X!}0mm%)i9G|}|)aOd*&WQ1%OH5%0#H5~3)qIU>+voDuR
zj9t;)5idHM@}Ane@^f{eG7mS0zAy~Qvs|Yt&E})6Wy9<Xr6>7ONP}0~5rc82@yvlb
zkjS|33^GFhBTD#{+L^BTg=~c6nYn6P%SPXcj-gSJ29sWoEOF$%tyI+m*-$i|Qo<HM
zhSDyxV#?EpoAN!^73raJE0C8_nx2W$@*V$+$?3h!O(m}E6r%_BY4E1`DeV`?yGgqW
zS_B=TmzAAj1nbME>BKi!yrbI$QcHI@1*>?<j{LT=Is8LlZt$uk`bl?B0LiTBfW5L_
zODDrV@)6lO+dhQgLX2XaJpns-lFxLNE3PVLiR;p*VK=SUIG-nPOv8<|Pc&Qgz)2cS
z_^Tdble<0>4V4`rp+8fU+>i;!by=v=6<$X$2v2+z$^%H*wzxSzwg4;)!OvBBtG9q-
z+a{Cb>HF!>CeLOo(5wQShtQ;s!NzS^&?!^6*rNW7Qw1SF*FwQ8RFdY<%Pa9JhnF}+
za3-tb#Kiv~%0$pmV+CR-W|mbrujF(&5)QGIG=5|GO(y144E6c_wnDG#oOagR#vk;K
zx(xZ2>ha|4P5C+B<21s}g!_p5$Dkqfbq*B8R7SxaN9SrFIMo=wNS8E=<&E1{Omy{n
zz8&)w+u>`7reoTh&J(FlkLs`oI}eXl87WdaEe+V)bZoS#PfT{q6<IPnfM_BS{Lpne
z4>p6=2U#N2n`*y~Rjb=N^bIUK4XS^<P&8N2I;>hzD2JiH`j1Zw$Z~2o__RL;aPsbO
z#4(^p9$ZcGs_w<Ph9Fh`+*Q}l<W=hS3ABoE$1AAK&w_$)>c~|t_VenU9E`JvOR?5!
z1uGi#CuoJsYpmfzJ+ax@*@X_p63&VaSFm0Ik@InTcYAWDzgQ6&oI|%YxX@663@we0
z{-cc5Exiwg6ct1n6#0t5ZGE8pV`vW)xl9%(jBPu#g3O!nQ3qMzVmjfwNobpCC|*<e
zKKaMsv5&m(5ck;$`3@WH7?npSY%!`6W$r{QoTrM>rS#~P-A8`aCu#aCkC~w+u%MaH
z?IZHL#+AKo7uBChznfcKtt_2Xk;d84K=8^XVZJ4VkZ4*8INH>$WPFziSFkk$D7lhg
zy`Vx@9mB0Ft?tQ@7&s|)7EXz`fQaUvm3Z+w)omK&Lu(q;SDu9t9FQ~7-RssFw`0!q
zEHQWwS`lGU#8pK{C!h%#D#?Jd3dli*j)3$wQrZ=1g+cTDdyoaNtgYKR4CW21$L}F*
ztGzZ`?+|X!G`lVE{Wu|AN84SkF-BayROcF({ZVsu^7>Tc#AO0XV;p`0e;I2oUg>vz
zqICED*bdIs?=Y)rawru$9D`&7%gI5k(6+`S9;LxniseG_8nIi(+$=emFj#qyfg1gs
z5AdlPOvFxj8r6jFZ2-|Lu+e=^`lsLv^#Mnr{<Olu)qkxh+!;Z<S!WXOGv5Fv!Mc-p
z--<7+?J;OY#Lhs(tjX}tNALVm#5=8I=wKBE88<h#C%3n^qQ8~SVSn2HlvM}C7y39J
zLLc(wBzJe`rZq2l3D?&j?R?zdzc{(P6b*Dz23F^ICR&xb*Rlz`%~!lGw6BEKAJ>t-
zlKsU_7WHul!%@O8`Y)$LStpTzn6DI$S<Nu**V0{Wz%4V2&$ixX!i_PbXtKTPT&L?*
zh^o<(pTkpH;Ja}`j_Lcy&>qtm$fZMnE&2QCZXI=^qVJhbRPx`l-m|(f&@c&6@ouo8
zPBpMJx-2&;KMFlwKsJ9H{btco_A=iZC}((oQIIRHAI*}qeAH3-G)R2;HZ<UISH4ut
zL`rXQPf0vuz=ni=j6)fLy7AAg47XEk6nr?niDI=I6a59ZIkX~iaiLo(!Utw2M3XuW
z$k}Z;BoFe8%n}^0qA1pF8SuF+b0!CZTRq@d7wK_+E|*BohxxE7;!T!;Ks`{kYL5eS
zhCSrRoN6hBbjTZ4L#()egqQW4b3Ot^2#9K7lxfg!Iq|hMP808KqG;h^4A{!R9cC-W
zbKwpPR6nVB2l~l(<Am(0$hTPTlSZBT(36Kct!>D;`med_;fjExa4nA$QG@xRdT(5X
z%qh18wnT4Tp+mn6P2U1kOr1z4ZY|Re#leGm_dLoD#uM!zDR@n7GAb|T2gQ;#Wxkfr
zkxGP#B%c+Ij+ZA^>Ga?yL8J_}GD6|8-+@bf%&f^-W!%y@IgMc@Y{>)-6E6{{j9Hxt
zAtv<~FJFi;D-0J}csM_~yt<OJaCY+Q#ie<jU9!Ie*(Y)yv`<6M15N~cJ1*pG;Bds#
zFo6SK7BSH0T&ouF#z`TBGb0Sq-re1u{QmKyb!!_{h2vEc>Fb>w)JeE<gM<&~hm)JT
zPr{PEl1qJS1q9eM(6ca?n^lRx;MYc-qCZVub+bMPNSmHR2e3Gp#J0nj8Ay3N=wMYw
zT<5WIvY>dl{`ARg6YAx?BxxJtQT`ZVLBGdf{jKuAa@c*X1)rUeFW8dR5X1+E$2u8G
zflikiROl=W)5Ot%0{JMtgT)j7#z5cfUed84Cxv2Kf;t`=LPGEx@hHO`$9%-{=`ykk
zhH2fjtP~S(b&R1a4wZCDKkioMfrWw*?ZBg~kyj6MJA@oBt*iVJCqt7xw>fZNUdS0?
zg1>lirTp2+rSicXC*<tp?p98SRy1%zUR>%nipp^)a0bHJ3Ej}=R2xrGr5<ziJ`(V>
z;q>%Yj>OHS($K~2BlLqBHHqPYLsJ`BB}DE>@DBtbIS8o*oUFiso-mZAWl~O*es8}+
zZN4|n_tLi-Z!e-}WGrCQ=@eKye|$Gi$l40cbp!6<ewCoX2EA*t-&cnm?)D|yj*m{l
zfOp$T-lfsChM7b4KBldD<$FHXsUH}yE65d_ur#L*N~<n}X2+y2EEi6G>Iw=l^~Xjf
zCs5Nq$y-?ZL&AE`KlqG)QLV4<xb(~BSb3lpUNMj!adjW&Yt@tr?#pSz=s?Y~Mn2WO
z3vZ?H-U@W))`vEGV#nIuDIDy1E*x&T{Qd8LcZ?3cm6GrBz&Rn0LqeNy+-Uy?haC~c
zuiv~rdGq$o$(?Si+{+QUzPs`5mIwBY=r8)Oo^)Y*hIwf3h#ZNVo8bHI&8eP(xjXsv
z>E`6-;ritI=2|Nus*`YIWk<ToF+$h-^INZyTx$=Did!fLCmav<vQTaJn1<sD#}C~^
zGUZV5^Bsn2{-;*G**woj+nw}+LIwBvJi)8;+RX^#<mq$035q`R)b;$AkcUaBNN3A!
z_4;m{kZqONR#wA7#oKV_%DVApu6R4FPLPVXW6zhaY{+@y8EB`=hE1p}vxo<9MIG)8
zr*9aPHFKnaN;a%^jM^QedUEJ(!*YKKIEM9Cqg9=#4^?xZGEpIeS`i5#5j)w{yOJLG
zGZT{lWX#U2e6ZrdZJP_Pg1mnHTB{j)L7bSuul93XvH?`<bKIhdRgQ2h>>$ye+7ug)
z=He)D%S4XF{e?Hx%I}E*Fx&l%l@5XKyg^krS`Z{1>eNEVd)?;YXikVW%JKw^-7~5z
zZR~`>+J}Cf;K%Aaj!Y(_$eETj&4BuSrEG|=r1IC4{gtZxy^5u&=;t0Fc30xNaYD+o
z-Cd;_3>l*h0yD)sK{Z<KMEtf8JOM1I<WqX4UIu;T^Sy6ttI~I7rGZW1(|lCBYj=2b
z^_|+rd6m1LqztE_$^6umrDw$*`zh)-ooB^?I`G5?qle~V2XC@e>P&k)DECt*G9yp0
z&<YQBEI^fwG|}JRvr55)j+1~bAM!TK*~y#NZ%$skcyaQ>kMB=jUcEf|;oUpkPPsh!
z`7eK#<H2nf(TnC@D+H#r@x@xVHoWqo6$`201&149(6V=0x%lvlRwz)xR8B+=eIZb=
z6C$KLp01EDq{Wq!+nams3Hj`mkk6k#o_zlF`Q-ZZb<pbf;!?MTF0=_($ZW)|fU`qC
zil=hK)Jmb3;32M19aER+cS@yQpefx($WM_$_Mvhh3}YY4;cz<irv2bxn{Thh$iAJu
zeV(TcrYj=F+f`H>k9ZuYdGR`s{0~9)-8dnKwnlZ;iudL7wa&SH^*jillX(&vq$MT@
zPmrAzJ)`TUDzuiUHGuS?9|*=JJywjb=1}yjWR7`!&$*f~%OT4VUz4x~f2|yxYC>jU
zuG==G!Hty)CJ5PHkN5hP1e3MOnar8UU&{%((x%h*KfFJA{p!`p`yYSweva2KUzzvq
z8MjLG*sbJvwL`a52wubB@m>AjYxU&*Tv|$|n+UKQOo-@ZnIy9ZRY$U(az-v#>Ciq9
z<ecB=_REc)wu##;wQn|fYZXK*Fu^B;6dZ8`v$mWNGQHzlc&3x`eqoxN5H`nhd-znV
zEYIOM3$*4<Hq{l%_B**}KZ)CRfxVh`C-Y~7wHEl7cS6=0;!EK%khMW^Um*;(`_k;d
z9|q4Ih+nff2ZyN&ZP4k;$Mva|kHVDoY{0>aObUdGr!rzzQ+W+g;08gPc-)@HwCjdY
z^s$hcl69bi`<TV1O*HYdiouEol@6yvlQ;V_<XNx@v{p?Z!$h}Qxwz#iTg|EH;9D(P
zIbag!>6DkRUV6pjryu|1^xyyXx05$--kkjO(@(-bJNbP5Nh>cn9G}|;(=#~<+JkDl
z`*CrNQpo2P3Lehct$4FHM09LWMJBQ@az)3Jyjm+916FM89%=H&u@D=T-Dnl%N>AY2
zNdM1T0lB$mPl)!0i0`f5KT@}6!XC0qZhrMURkS+8Z6aI|WWb$OqORA)ss}3{tSae;
z&9i=6t6=APS4!xnvL_Bl^%Hc>^XG}x-VoR7p$3iKJD|`GOXqjdRe5VOI!4ybF!GNP
z#B-|OW3c_!dDzY?ZTRejeCw_GN7rv<M-DS_x-%&Cr+uHE81oSa)JGYzwx~MMe4IU2
zc-jHG`A-ui0ZU?j1{7}+l@sC#7iS;??+rOQy|cr?iU<2UI9LI>)Fl5(s~+Uvy?c9d
z_44ZEg;ql-i;aam@xlZh&O|(Zn@_sPxga1rHi`0zcl442qTf3?Fn1R?O{^?fD?1m^
zxc#MKR3n$wl}Bd1rv2gc^5o@9?MIP(ZWG;HhXaChLYmtx-UN&Dkq!hCS3TxidikYh
zvI=}F7!L;oyUMZQg=a$Xj#TW0EuX^y7N9eBfgC-epp_<WVM5d33yGfB0?IS^kuC7u
zI3f0|4yb%})B{NW5F|e7W=_<)Qur2(Py@qA7wJ3;UvMm{J{1?Qe!%L}RlX`N`yI>x
zDV{o4iJL=Ou*Mwf28^qFd<phrFv;HNofz2m(ziae0-{OtTrd8*)MK+$RnLGp#+En`
z#4j#hoV<GdO3uc+leaovztWQ{+H{H&!aG1NaXiF_TQWE+*II30&&j9XSpm8AlPNr@
za{c)x+Eog%qQSlq@Y#fVp>M)mNmlGR2`7chaC-&ov8utU%AJ;GSaraOz^><PFqM;V
z{sO&O(YchqSKg-rJt3SS$-n-rm66YSV&?P7$4{TE7fuOw;YvB&2cD{l&8|E-gGGAB
z0(;=Z;2>Q?OTBJl{|Wj5=lB}o&~dbMJ+7A)m;K{*t}mVU6PB%>YKyr*xl8);b{@~u
z%<*}K9dCi-oseoTdP$|{^rG`rBhQt8qcC^=L-nT4)yXG3+DVo59+|FazEPX(R%=?j
z-4tVtj^2jp@=jL1;xYB=+_X)Q<tuwQOnH+}SbGF>z<Fq}pdCsY6$Qzm<7H@_Ix+bt
zs9{4i@EcMCp6AD8()A?hH0g5QLDfpsznCavVnqyya08D?8W2yiypRJR2jucfuy}um
zZ_AwPX%?`B^YZ0OPpDT~O~Aq6X%`-y#rb$6N94uT3r)~ky^s^bN(B1IyRc)y>(npu
zQCxI8JJaS<=6a=>?BIXykIZp<Csz+-=d<&0d|I7!%LDv}Q$1qqM}ft_PrC%KSgjE@
zZ3KUoNVLxc-XfF3BIiZ0a1a2=W;k7{Kl@)URA1Am#I}^}Zsa(zcLjUdcDAY5>sA)L
zaL&S6vSy;=EwI9sW5?^}@9*VoVGrsGtzv>e=1FWj_{SAg-NFX(mEgc1{hCSeill!B
z6lF01ymy8P=}!C=2uHDmSV!3f!__G_;#-*(Q(@s&zSmPQz|*tN1TIPBIoh7jwlp#(
z!UuUUrcJjAm?_H!^xV=l=JXNcMYU3&8O{TV_%al_gC=w<J~$0O&XZ`YG1WL1jCB?s
zzd_23^|K+M18_^X-b|Qv`aU8Wqi>15wAR4ZIt*7namC~@pzl1CFc8&Y8J+R=k~ttV
z$n8?yQYI(Kz~QizPclCac^JLlmEDutU<F`AQJSN(VU;e$#bPT7tY>V|+TX^~#z~B-
zbnq(9gXAVdi74H~8L%_e+Fmq5FCMiC6h?3h_?RD);mtJ@6^@0K6rON3Y1M1cSxInE
zF<0q#nk;X4zlQLO+oc={@8jTYrNV!yO{2AX!TUO1$l+jB;uVgCPUs7*e*FCN&nJKS
z)1PuBL6%89dCvz9hTlOVV93F#U{&JGw?dG6E*`98uwuY!#En)W?xfS5bhKXSM1=!G
z{VhP0ktWZo2X(*Htq;=Y7uTjg(<|Y9`$P@_%&OgZT{~aF^;0>L`SGJxJlGSWqgE-b
z<CRu%UdP5-k6n_L=GD_QA1wo2z0p;)=>OpZ8(wcue*ax>W0fvAII*I`_i1DkIdi}M
z`m0xbup#zDW~?Zor<_08mnV=)m+R}OA39-^nta`!`9rJYY}v8I$<pDMu(><%LtggA
zp53jZR-xom{bveyv4q!NsYKr^nmf1FGC+6luknfR=JZdlXV`AoCOS<e;(ReJHkl2k
znk^|{bJh#6M{L;}V`1r1vZd!NPqi6dvwq$0JDjNNJXMG0<kaq#9*T`ndptOE@lXMD
z^x%kel%S!(k&QAD*Z9-(G)fZpD}l$agrx9)<%n_bu)v_@))V=@OcH-zvR=_+XnQ_i
zHBBY3VP6^2eLfW2gamx;Z?)7s{#53%(l({+;>Zaf!_`<*PIy)-=xd&k#hP?EwjLR>
zA41P$rgZ>al<S&b6Ce%-PKTWkoD8jEcoLRVLR@h6PB4j<Gs4OT6Dw`Om40mytpSJ{
zGPzPOuQo^?D<bFGb?V1_r8l=s(3Q=gc0x3fqbid-L9Z&y0XXgNlM#6glEgVN3-94U
zdsb8*CSRcNwN3m%J{+FFWy7r}bk&WGuNN9@@6tK4Q2kJD9Lp0R^&u+0;Nl#yLQ!=#
zZ`FakCv2*r3DEmW-~}8y_@~_g+;jVh0tNFDUe>sPkq78hB1frd?VL0ofrn$lDxG@{
z0!*Jo+amDBBDP^aZCO``QnAz>ibxgoY1Qkc4f^bc_86q8+n20C=~gzJ${+QWQlykB
z<Z0e;A=B4ahG!>ag`pU94h_{53@p2XF&KO#?{JB`r2MOv%X+&_yu0$FwvNeeO0=RW
zrz9)#^ZHlciz$hLX|rc=HT_bLz=F#Y9!z*V?ZGt^w**dW5*8j;!A!bZqd2*jQ$fG{
z;q6<mT5xLxr-B$K;UddfFlN$>m68kZ_rMv!3AvHHfBy5Ib}o4JI&co;@^xGE!7;kl
z{)%#tkO@vrwp8g>D>LXHPs%Wcg|mb{I4(l>tr6^i{eza<Jkn0}!YUWsrXdy$S6Ww+
zO>^qTh@h+j=*o`0A*xFS<e8WFurg75L(1;h`&PGJYL5piG^`dNhw~JtgFqXw;=yex
z922iL3HOuqK$lP2XL71GK^6gr4jSyr&J|jguM@);rELqvZDj&@hT>8A(v`gHB0kBe
zP$=<#2Z&_qV?&J!L(gP1D*3vZIulC$(zDkp`bc6ZvQ>M8r%PMCIUQH3@Cd3`f9d;)
zrBB*^5wBNOG}Na0X<tNoZB2Vw&rZnHiqpIXf`bE%41G2B!Hm8F;SoCXQla(Taay<P
z#uZbpbo?iBU-qTUyIY1a#%R;Z?3EwyD+$s#*{MGLD7sUW41!s0U?U~LFBnswJJ+nu
z`#tJ5hOfLe;nl>~uEd1{ZW|PsH0iTCE|k_R*_PlvF$#v`QOl`3qRYgM0|IY0pfbT)
zW3f4tZm>JoW>8iss#}#4;#d_cC*&bN6bBwUu_RpIl_6jBF`%NS5P7>P#*uv71Ks#=
ze20T#n+Q*Vx3of!vw=D5b_q1_#@QL1kl52fJ+Y}B4=R!YJajn6MyV8e>>769n>cc|
zglFeV4#_k`{kW|aw|uI+Y*UU8@Zyc){1uF3K$MW<*Qtm_6b74-?gLza6hF#hib+_Z
zhd7GAYDhDu4yD|X4G6y+gVKgaVIuDr`872M()aDLpVY_4t}56v7&Da{!4XSY9ZNW=
z-Ue;2_KM0w|LlbHMq5Y^-FX3F3%D9-3(7ZHbhO?_Fc=Ruy?KJ^xcmq~J?lw&os7qm
z?ah-D9U>dOZ|b1*`Y}=7+<dab!JZCJep(H{0bx=k&ruEtlN<-D43zU!3KKHH(il8+
zF!B059ijzp$3qETEEbzNc?kxnlm>tHiM-HLG@iIMDc<tJ52$sk1eiC^p%)aA&8iC%
zUpXPQ5~5;HCE#$&B_4sTNj753<u(R?)Dc_=^H%n~UgO>e>Mu0$idVt2s(}*$f6A~)
zIU&@Q14rOiPusA1;@d9BNce;7b}t&+Hf>}lL^$vw*ufK?nAx7TVJZb3_--R&;DSL*
zJ+7q_`czvanQYHHV-<XxMeXMHLWH44!zN||BPzWV1`=Zws+%+*NjX7W;`hbiu1Mov
zdNzO)^n-%5L;;S7kLC5XZI@L~<FJB9Xj+xg5?+?BS})<5$%NLho_7J3V$tihInQCI
z1%@+nC+Y?QjijU5IUB9zlb|+y`BqHgZ1DQS0>WCGW*MxwS$3eY2A0WS01PsK4e$dc
zO~{4;G`d?m^%Jl1n{us<l;hTbZ(nFa<JL*_yP8ld_T=XYIVQmu!iNWY6qv}F)R^Fz
z0HI-mu6JeBtp?4Ua&lOFh!v2Fi<^+GH<8}l@Cx{H6i`Et+UoIGgoGnNU6DwcVplEB
zk#II#V*Qb4!e;N{D<?_1@k)0bFN!_53Lm{(cggJo&Ij^%w@17LO^y*0xA5&eXeFd<
zME%X5r(m?X6no;d@MH^i0Urm2xCPN%3s!qhD6{^GZCBA%z2RRnzyn?}dvw4*lQRO&
zEwO}vgR}Vi=il8Pa%))pkW)fw9`HpW;o_v=K9&QESsqmPurx>t00~~bbi!H4oij61
zg=p|s`zgJq3SgV4S1CNzx8PICFP~r1d6ncxRJ*ZCg?2e3rC(XHr(dKkqNR8u8?;!T
zTq1Z(_)#tKn2PMjs~LVjGGkZcIad8E?*XFY*BZ35QNcR2KW0)!Yj*2X1F*BpEU+uz
zm87)?9$j;EggMv^XkFRi$&jTfLn?SPA*{O2S4nSP>|{fS1B)BM<cbr*l{EV`2u!5h
zJ}HNSNsY;lTP+OSb|Qoylf6koEEBxSYvQBNo^n-QL3~&#VI>1uv0@ODvT3YvXf0Ad
zs*5aiLp}jrcqa$bW>iVPI1WVqBRfp;LeEeWH~=2&rT$u}2wz01!UNq4pKR2HdLkc6
zc%uXRWbV(orNph5;9W5e2;a5gaa~q_Zm(k%h{nOL;1RGhY2Ti~w&IN(ub?0+<rh9Q
zC4)4o;skOF&5tcBT{XX#jfHF;+}gRlyFR%U0j)vZeY=NpsbX!E*1wrL4ya}Lb@6GJ
z!93W8R#5N{q)}?<fC$Dn5ITjFBsV}ZH2_+jzK-I(A_r)o{_toaIzzKnUHC?42^(hf
z4<bwcgcbzT&%g)b%@nn<jN+^O(>WoV`urgwtxI){fFa81*ysdb%Ssdd_EST(Y*$wM
zQ4YgNvy=wNDUSeb!l!tL<Tr$M@+lk2zl8kaSy!*a2-?+INK9(zqIpgZgeE&V0)E0p
zP6%b56g6S_HRQgf5PK4WZfqAjA~4c`LjS{IVS=n!%z_{%*2}z7FQ=lOnu(PLrq0j{
zS@q=23w^YX$8(|K>ikwb(UD2odr$;JHtz&;&CEu8yqXbb1v=FV3wH5s6ze2=$dT}C
z+Z9_6a|yNmE2gn35w;6!Y7dEcqUXas4ny2B3FiPR=_G~jF7*VF_O$S*=G_%5BYOHq
z`$}-Ud`m_=*c$`<lQ!UT`$kNg^Cy5F$&J_2*D6Z25j4QCi}}(HIxfWXJU6e}X5t48
zFLwL%QMQzfTdg|9hyUaNiH5hxqAz;em{?kTA^=le^DlwaFp^;q;k1O|5;+RRZ3KLR
zjP;o4f>>-a22XM=+Y<Y0YtMgwvZ82BP=I~nX#?y-Tezt}S)jFH^4^$xfY(tx6p!#G
zw1w7q7how)dWxUyc1#&yNvCfWeyrXj`mSN8a5vsjz4h@eIMQ(+_U>Q@Q+r?;iyL@3
zHT;axj?8UmODPjUFJzSI3l2&bZyXRNYDU-S7fzaIyyHO=t-n<w1RM}ekZ<*3ur_e0
zANeT`O>jIV!<BMPe&CpA#-4=5*%LAS4_e`~?C<wxD1Vzzb$GKYPK2(^FGSBR3LFqt
zLc$Tj$zTNlJ4r94d0GUzc;ZHrK3CkjeQ+<H@(N<bKojTbnKqM3)+xIH6}ZYmCw!5(
zVhcL1*ze^q@Wc;ta5z|Tz`^ijtu%76;6{KaK;aZ`vT}vwOt6xH6GGkF)T~marP93|
z5uo0nD%(&m_R`=uaNuMRvx>nAi3FUSV|(Es^OIx>C>)%gaC|7^AfV3`c9ngp2STx{
zZ6{f*tN_Oe`R&86%8TembwA~49X7whh5aM3%3~{vhigL2EgPPaLYawaJjY@ii1d?D
z#8YHLk@Yn~r?`?sRmSjMvuJVpybaVB?U8mb+ylQybxZVs)oE(wwpdNEjA$*?#tw97
zg|b?|B2uuvoOF*vo?u+@9Mv#+2gdwwF9VG6&aHWxKQwgIczarcfrY`SV$wSU_%!4i
z+#Sq@tDqf-GsS4SLfaWiK<Gc}nhSVyVC)5aMta(0Y?>Hm<~SV!iy?Sweq;>#xIGtC
zJVd5hsZvOAn>i7jO0Jy`;c&}^zX$JcU>=W2SF@KV%e#DXrEV4Y>KZytj!a@+^3O?*
z$7cC>{V%`$W?}wXu5dlkst?|ByFn8xs}|7m^;_OY!fV-IYGvameF+!m0tbYB8Ge*j
zc$TC5h4xtR+I3{&EYL6N8!ewc>SbP%5w~YzFGlrws?Vg0o#91a2-|4F7Y{oLq78>9
z>VzY5>9?zfg9gpC|Kk{mrTLntYzYpI0B>5gL!rD^W)x#kN0M2$RSb|u*qu$Zazo7H
zN}F+UV6X?PPj6%^9Kc`XNN~#u=j5a6%t{Gu!@)`lxYQFT>OwHieH=*YOAP#42l-o-
zaeJs7E}n{_9+$H7xz0Ck(NMv_c2U$x?ML10hzQ(5cZy;MvLbh2&14iX#Mcg;O4zK+
zO0&peoB6N=pT{0S4<5^;mCbgz>_@Ts#3rRpogZyHo3wGk!848cxQ}FCv5ks%(iTmf
zC(Vw9mw@6&{Op7@E6pb+wwp)V0xkM#_${cTlj>1A-!@@u%1Ze)xIK=Sbl23lC+gFi
zv;$;FHHsNfE88d0j6P+k{1lXY7r^^SHE!^!-%w<vj*7S*(+uR9f;|u@#Raw}Y!uWa
z$jR@CRg;d>riuLt<$y3y<6PW;qrik%SMhBS7oQs@X9gE0Fwro;;1u{3?fMwqE4`w<
z-HP!F$l&C}J450fAKq;0Rfd@G`9{sPJ`_i|;MM6-ZRcwyL>7)0&V|$HtgC!Bls;&+
z1if%VOwY=PaC!eptd7`H!hvyF5q_}(;XNZNfo~14tXF^ue*xR!q)UK>iu|7Iog~O)
zPX=((@WMH{xX`UHY$b<{cZ^_n-V=h&;SHW|<&Xx39;pis5GzqUfkS+zHsFPFgd0{{
zSgAn|ANoPIXeCxXLb{N4iZuWHIVpDuM+6dx6G##SBBzD#;=rOPv5oUJe^O1W5a}Kw
z@Fd%GU4l1+-HAi}kal-#w}?#)h|&EhCc!p(x0x9*+otL}J427-jyRZo%CpapIQKl1
z)~f2Lc&82rBRrLygB4sLlqC<gkO}ppx)+bKcg4L_8jUNVw)E8(9+J*ZatP@+MKjZP
z!k%SSn^-Cj`9t-OlQpOpWgIVNr+U-=LEgQH7j7wi!NjY$cPunB8ph^GuH2ceV)}}S
zgvrki2Yrk(O>n;I&q<Dn5xRPGmz58k0AietcR&2#x2gU~uWDz7<E`?+$1|HII5{n{
zP2ONo9pgz7D(hQ2lJ{A!W#=}=jXqSzG{>7#38XPA^}(rXAFe|muJm;<!K06ZkgFMR
zjIc%9^xBmLGQT)9vLJSp4LR2ycn>(XxaZ|&@jj1OF<|9@6%i(U%CVy;lqT?dej!>_
z8=b6sRQlxMb=+2ZsgL>b<jxztMDF_b+N(#r%Y{{uUw{4O<Tt&#{r3+a^m_N-?d<v1
z46<1LP#x`ns9ra6WL^k{OmNYUjj61{&|c_hrzPu*okZq(!UxBM@#IdoV(xX^s?OA@
z0`@TT1h~FR112=o4M3DX^{@3E2~9;uFl9>Lg13qk;HhHz+o7^*J!)vLLpec5KkY-=
zviid`U%1gng4$J4g*;E0XJSn&xnS=yMad$5UI{5nFU3>iF5xY}%h;=WVkgVDWEOa?
z0gt=e8j5cdr#kJ1c8diYn0o+V+A;2uQzl!Arl^mUmqaGfyvp@excUqItNNJJOgw@y
z0dbYh1T{>onbrUSoEe4z$dec~mvcaf7qM%VLXHM|6j<?KzrwqBKg24Co)lp}2rCcp
z=F&O3t|mNnycpD&!1?|OGBo*VwFNo;c$=O!;g7&XS|*QrNOySf1z-5`m(B;y2Ks;x
z+!Lk*0>>HfJ5p3{U$xFO5F2BWVcSq|UW^7W>o1OVOGbM7#cEpRumi$g3c=dNfoOeu
zNyPVB5n*gU<Ix-^(n(UDGGtxe$fCl*xx<;0;%*!C_d8m2+oqgBzBOfMQ1b4CPc^i|
z#B~LZv%))Ju=~wz*fv~E)l1G5t37tHFapH%8CH<2UBVQN0(b?(9!>|aFB;K&)eKs7
zVUq1izV9R_6>SkG*O7HGBlo={84cv6m-;bb+6)05&;x0T^?W;KVxjqL4f6A)|C&%w
zXw}K~t|BMTW@p=<&6vlWIsqHw?Jeg>f-l-XvTAlLR9QBp)-4AnXI4ZAO!A&!Js~rp
z$_deNrz>wJFIFOON|?NGF0Q!3&L?Wvr@-V^ZO;A?Ru$gqb?)4H`O{zi;yoa5<%}Sc
zJ&t_l{<mL#)fN6*IVbO!S3QyP)CdzV4hjy#tyUN$vhlvs$JtI_vTDHu%l?y?#O=7K
zu1uGvf58>KHn#EwU-Z$NSb=+mhv82DY|`|5LIn3Ji8*48bOcB$cm;%5`&6(aP7CsY
z6BsP*&=BJ^rJY93oA8atFaPOW@mhLl+5dqJat4?EaY}IHWaqp1c-@U|L9z1l<NNoz
z9rd@9|M-vpIQftN`9J;I`+xl7AGT@V4mfRWPJR964b`>Y==95f{o=hO-sEbnqP;I=
zv-<}=2q*`U6_M@|6DJ2fu{rHb`g;oQn_Coaw-LAC99o3zW5PFQ>#EN}ADFGLXlw=g
zQD=;@Ob_ddS0%e{x(ffhLF$b&O49%szZzm%BbO!0wCp**(*o6jkLi@XkmAEBFr<Z1
z8RBYS2>UeoCk+TM#f}vm6``B5q|OA4V<8_TyKczxWN1;k>NYainy}D>0fR#mmk;{3
zf+urL>~-7aR+F-6n9OzX^<10~Z$y<7@=_DwmA<UY9tQ%GBa`8+ZZl9{;GPiWP@p3`
zLSJ!}NF0|_oD*eL1L5dN5N+h7P1)GV`B;x6gF`t^O57Ua&8|$SdYgdYI2$PIA(85+
zqIC<$6P0WLULULraF>N0k?GY&Vrqfy2wDap&39aQ3J1GUU-o5SKCf7?R3J6H(jy?Y
zxYvx2t#PW5gG}mPP8n%dV(zqG<VyQ#;Dh6M^En(7HtKrMhnzvL4oMXpN3^oDD4sEy
z$>Gs$9)I|c+g~ytg8V9a>BS1o-6<<tqRcwEenNC#5Ds}(M^sbP$}moWLjY=e=f>#-
zJz-Q@J)y`Z>QLEbK^P4-ZfvKC6(~~@HS7pfIcuDKV1$sFzFp_amtY%oC1aw;68#i@
z`rrTC|6V#z%Qo_yY=oL<e4=%E7fKNr`ryO`C|C^;#yQ}pWKBt}i;OVpB4rdI+hE9t
zFyWQ<AXK>MZRD=22nNlP?5Qr>@~(gBAfDEt2#~FG*{WhD6<^VUhc1Ojx(n4FBU>bH
zhjvgh20m7!qwR}H)d&5s5fet)l?mm64SPq1o<s3AOUa8k+PB&jm?x<y-Tzl+^_6*2
zhL}N%mKSJ%QvJx?yUydbS*rvr;NyuxZ5I=+uB2Hp(8N*>1CqhH*4_%*ggqYY19<cL
zt@{U)6ZUZ5QSAOBJSIIq{;A2F{!BTzo<LPb{%7s=c%!TCziQ><QgZpEJbmu-Cv9A1
zw$pwBoRc@YEyHbvAG88e8&k9Mc6RBFRMuh~i2g$TsBb-yV;iC`hyMJSt#F3DAc87S
z`x-}h2s2%4uaYxkb@-|-B5)zUb#A$Ddu4-i3ebQEM?{VVj)7Nn<dA43!{>x}!H!!-
zB!ngt=tp}s>|o#^fS_AGZYQnUkas)ChU|0s@bP!;^Z0PWD#(BS=l^%|fB%pFp~r83
zQ~E#s$S?065eGeKqxXWyQM}UL9KxrMzp4D%xNLm=^yymXAC8?Km)31FoE)6CxOG+a
zrA~4IFKK&wP~we)h2rQ)tp0DOP(lPlHq;(H%_d+1JxZ{BI!@$~iR@`W`>4sZ<Qvbs
zE>WLReFXwKf{_mkv`U@n7GWO>zv3#}u$bi9UfJJ+vmj|(@HrpmbSR#qIL`^g;F)}U
z?g)rJb%4{KmwmqAef%%G5<<mCp?Yv@6qPc8Rw%<39!9Qo;XKDXEl?)gI4O2gZX)zN
zNW;7M8@qr#A_XHPWA68}ntqCiB2#77iY{Ii7ae&Gz>2DiQbPBP1Lq*FzAYUfA^W*c
zcAO5unRLmQ1Un#_U~xE@v~VcAZ$T>`I2-4htUOFH(WWEADgivWs%Kva`!MXJTWrck
zyp^ybp$g6eP@i&Hs)*YLsA|3{kIkc;nz4TaopF%tREgO7rw&eQ@?9Mev=m&VAM+*(
zJ3+!q%Nb4uv;dlosV9W!ty2BpyPj%~EV;*a@DLr)Gv)0BYGnw!^QiM{z2oKM>rdKy
z^UBWN2W?m-+&_e)$;t?A#ci|-O+Nt`{8cQ6g3Yz@?h)i8FX}@qM9B`l<j9i6o?43{
z9SD(fHX&6)95fM=e-3LbfZ+*ySg%lfGV@sYss32Sm8Y+^kZDKyDV$cmF41!w_m*|^
zb69&hA#{8~#d8dMm`@Z{HU^l!WdHQY^OGKB4B+<Y3WkorQiR(x^#%moW#D)4M9;<v
z57C%&I1PEvL??438yR#8Y}K<Ye9Dw|8R>;kJc+?#Df!@l?2fp2kJ_OtcM`cA%PLMO
zR^?0f7@77PLmn+B{ayRyA96N@J+6H57ejB7lzeQ+;XMLMTwkkCYl1uHevHBev2St2
zyE|U#Y8iZ-2ZAR~@neO`%l?|kc)Mq<lz_~B24w!*zx~_ETkZd-CsICt)VnNjSh!82
zavXz8K1`<x(<>p`8^Q_={sjDiVK1Q$c2fr0)^;wVybCI4?+Ao}t1>2W<!gn63ER$-
z>S`QNRDNuLL-DCg5wC=%p75t!;$^pC^2V89uT6YtPX_OTb1LT#yRhm(K$e|AR#gP!
z?BqORwFk$Ky*Kb>HRH2x%dq!`JvaZ<3JH();*{{NmVf^9pW?*rFYRej{#15H_ZNC|
zDkI0u4XZDDGpk}dlv>H6ZHD%-WSyfqE{xR|rwD+ldljR9r4!{*qOyNKrXWkgLtff1
zL3WuQMeq|cp<SvPs#Z&v3R4@<cL=sgBnx#EPw6Iwt-S^o(T6EOr#@`uCB!Q4X*SZT
zoMox9J@+-VdpRK+g1=!v8Nv-OJuyaW5P(Y|35sqB?HRslfnhM&V&6@&RgkS5`@%lG
zigSmwnzl!*boIE7-@!?EdiY95QcrTI_tGchq~?km98Fw0gX=|MOmMtcf<Spq;3gCd
z4Pz3_a<1?-S^ABqTK!<+B=1`<ny0+aKs4uimxrHB(cTPCY7z=A{Lsg8I{kdEg|a|V
zvs4hp5U2qkX>&~){=gHW#-FtCd*1Z7q0bF`_fDnkHbao_1FJ0wb2M%0L?^0Sx{z1h
zQCu#R_G;K-5~E2rXC~ibk`37R@mhOvxK;Az?I+vY?-r3jHV#91Eyn{oHYXD!4}IVP
zyVxwtYKUZMb?kzDApTRk3zx+j2V(Fe3Rw3GUqZzsP~CNlqT_!NjsW66L?~-krMFbP
zP_av&Ub0j6O?QICfox2gIY;W5+5bQXv_%_GCcF$XYz=K-FaPfKo{&0mmf3A7k)aqs
z4h_BRTKOnmUTKk0D=xU@$+|vKW#$+Z20eOxXoQh=Zlr+rc*T3)*gSUB-}13j)};o%
zCX2Gvv+p!Z`t>$zJXR)(XT?R<=Sf$-4T?gH{ynm0k?$23Z-`vqtpAt^_Ua%KE3|_G
zdY>}VHhSteW<`DqD~u(0m2ZoE8-j_Qw&yP=oLfHIF4*bWwH=FlJsyj5aP{J)<5%25
zVOd`*Av`I<TQ)I=)rOZ291pIfiziM8Z}#Ni4X8i9|IvF1c$8MCZc}W3#cdinDm<OR
zZ5m|Ya9o`8eVV8Z4hkDrt1o!Zh9*VoQ~NiDddFl*YgT_b(-R%Aa$AcjSN3KtIS2`7
zYtRW;iuNsvx+guzSefd0Gj8;u?|1M=SMN~~lAb~GN&=7YVk?KV8R^oUdYJAGd^vX7
zi{baT6cIdCN8i?xo!@Bn<Be8C`0mV)dJ`=U$lL$-|MB}o;Q3jO2pd<~B+F_El4@Vg
zg;uG6-`?^h6t`ovN+OxIzwo^6ShDWbZ={I45pgU%#EF2ehK)^;g_fpr3aiplbj+`1
zb_~kyz^#Vb#nL9ab|l|jvbq}eavd={=`4tL81nAzLM&U2@T&G@FXe5ogd4X6Qr6jV
zL#E1Sc_%OP6))=yOz9q0yo(3r4Y%W+>{<lfzwAm#t7Wr<5?zC|@}&g0eTy#PJ;R#?
zzJ$42^ED}%pkS_W4mPD-$}jP3_&xJJ-4GrjffaE>7&J6z<Rr+OIr$9x1@EP>o;t}j
z(cmPon!tp^GP?%>5w=0W+=c-NQXC1K3N|sm5N*6)L;N&h!kgO-@bi-$;$KVp?B4)}
z%s4mL-+;rBw@P%*koEz+%D9!{kOLrkF~tz0D-$ts4PK7Pi-<!^A{I<Yk<K#Hfpb7G
z@1(QCK)TbzMd+vmZ9$B5tKq^b$-;tZxgnh5evk$?+s1g~ANb-3UFtRqw`8uh0TsLP
zbP`S_?Su~8V&Qh$wO;f_=MYCTLMM{eFB-=7#m3%PtN4t4L#)6EwT^5=^BAS!SC!7v
zFBX*?;y+Bt2J2d4RCS6g^yz|4V|RclQ2AMSF!Zm=Yei(K8Ne&tDZid}M*3>dUQWnp
zpl}_tLVaaPbMoro#A>S(J<e+Njzg@0xF00pIIV|ar7z_G9x0TpVpsW!3r_6wP<qhV
z=)mCE<5hjC9GX59ZpD4szG>w5eERe@UZG=5EZgw*DjW43*+$9*=rLPrtkJZ-^D<GS
zqoeJP4A_qqi#uHrb4AO0CF+SCuL?+!iw8ZSf<vN7>$O%V{7wit21xe-5A;AEoR~a(
z9%yo71?0WnRq^xBe~yVwPulRF3qFL$mHE|Gu7JFJZHI#u5LQfHX-|kZieBKP$brzy
zv3L^0PwvR+;L02ar8cV~ulj1~<`o2~hI57K>UIq38W)|TaT7#)$N6AD;#J!9tjr!U
zvOP8>4f#=10#CVqahdAH=F^KCCU)tMgEf5uObl3ca9w$=yHb8yR(0YbhkMccc9;dJ
zRB-ZcHANR%<s!US8(<%7`Cb4@K()X7L9h`kA@B8uT5iq!pa0kY;n&{tayjTZ7%MJc
z<Yr#lhFdmrQ||BeeH!^Jw8gphr0}$n`?lqZNOB{IQs@$qEuL8n<GkWF1_kEfzv3O|
z7(y(d@h<(cB-;TT9p8OWtz3615`D$L0ZyB2+8LW*k8G@lXHbKO(@b^<gUy<~%?LFz
zQFhHf5qQDD)Rg-_@l{l*!nH$!jygy;x<Qe}6g_gWaHF6)?nc2%z&^63EURKY{1|>c
zpDk&uvxkM2Id!^Dc~*mt+M+r{a@@e_+hT@y!(Z0ui<M04_c(&G6?QUDyn}@fmK!(?
zsm68&baYQaw2t_iYfLV8dMuU+$$`}lO)PEj2%z#jN~#G*lTw@5l=lFEBS2r#fiuF&
z0>RD>Fdd#_r5U3D06+jqL_t&tg+e!y2yGUPJs+_WqRC5B9KUuzL?5qFj}58y5;3ow
zXkzXie&Zudj6HcFEA=Ud3T@gz25#y{(xLjcm75q{xOFvF=&NJ#^LJtt7uSm6kxk)2
zH4GD_yzj&*3^lqBn35qB0yy4Bp_GSYqvMPyhfLYZw@vaEnA^rgqEo%F%M0znxs#K}
zW7Dig@i;c`OnIdh66|+<t(!x78U|-XriD<4c<WF??JdVgNt`Djb$dq12oOQ%pae*S
ziX#`lf&sRe9)4iD?s_t5IiO1kIFRReLzYzz+Ygx`)zzfzBT)7A)?^)8B#T28O!L5j
zVe2Z3bnz}|aTwy6k^;H&4Ve3;E{e|@!fEB(;L$2H=Hh`TLCD!h)|gXwu1sV#z8&0h
zGT(uxEZ65Bfl1Zf0}zrYJYAlR+3XCCjz&8z&l<Gy!kzG5U(o^6dhWqz)Ygo?d(aM2
zx2Mb!lQfHzn6g0R`||A^r==6UlWkOGkbwq$@PsND-{zQ>^?2LT*pF|SFfsYi3JCZ4
z`Rc4!4Ky)vCH+!QYOuqw^${_Jmd7}G3g*^t?$o5i1TDpxXmK*Yw=*N2JQedvc%MJ<
zCEM8ZU|iu}Xd=7t+e-Bau2x3ui0~M$Mv7CdfG`=p*7s0&r1zDczJM?C%lYtAIg-uO
zEd`@5xPRaTr7jP;$_LKvn6-UnY;{&0Ae}-5Tocekv{0Kv+N(&*R&`ZJ8@sNu4=P{a
zMA%g*iillLVLs@!?Ghmj9)VSYTV0aG>*#$;ORF6>r+iJ9HdoteFNfV5p|JwOlS!}N
z=!qp(pCps^efRdAzr_1N?-JqLGynMKKkS(JZMfo&EwMR=?aX7v()&)bX|FRmFb+}-
zS5`<mT>FG~_RONPFEJq3A+V4BwXMWdtTQN~f5QcAV3&?_Me(i2)Pnla2j~y3pp{N*
zx>{vh!2;a+b(K^-M@}2uuv7Ns!g1SYS%WDNeKTk*C}p{Sf;copPc}-2x+ybiH7c)+
z`^^@9r3?7`-HsU~RBAa+1$KK_%PjhmU6lg2@1|uefIS?%w0-j|3s!LQEl`gw#%y?x
zo!BLJg-+9YAEV+DuaV8-EZa2rL4qd|$xcAUwg)@i0}5$HJa=KcA#@w>ZDl*;+jI;j
zatfFL*+|ODhA(H+uI(H^j;^SE+lD219{tq(&VB(uf(yR-2lx+~e0a>ae;qjF91H{;
z;4#U15|hKh?H%9ZNoRydcm4Wt@xlpVlDyDFhod6~mR0i1Tf96|6IV!BeV+#W)nN%&
zaNA@xqXE%_Il2?52Qqp&l_xGdaFwri7|9kZxbYqkKm9`Wl(s*fW!~2bxI3C$#0cI8
zuZYOjFK`AWm)mHBcY0O6I>5=t>rW@ApU-@&#)CS#!d;BwHVwCg^tM;r8<mAqD3q*E
z$m^s+UZG2?#@0ECNOv@ae_*`Nib$`;f#RN=Oub=QX4%*%L#9$2gl=EDRq))xxsgtJ
z4K$P}mpoeon|GGM#pMH4W9&q5r$2NQK~sFk1hcvQo{lLa%(B`9hKi~xbUH>S?I!wO
zkP$C<L!J5+BTVO0#x@)bInpbU10`_EORHEz)X`z2?*g}~gFV|BoW`Rc6ouE|9SyuZ
zNp!G%lD@i6d%}x0EV~sSTZ`GR3&1$0xE;#s<fu8taIArHC8T0=CNl!8mKWsF2Z9J0
zJ+DaHHnUuN+*yQ0Agds}d!if(Ugyn=r|MS83tbJjD_u=0HxJq?qE(TzkDtvGC*xYH
zC%^sno938sF3^(~T$K}2FC(K)^&Sp1=36mLP?vg@I{Q0V?#J2S-66!UH2L9_T=MF6
z@Pt!OX2irME@>xZa=uJ0;d4oPl|9SKb&gF9Q+U9qyb9MiI1mR0-kU;zPaTbwV0=t&
z8nggnr-jvwAu!d@JFz?KGqS193r^EJwVO*p2cUmj2$^;d<RY-TqGK9(A_se3u_uQo
zUX*7K&cFZrzn}cqFaIT0XMW=<UnHiD+5Zt6QCUq3$J2bJ5^qnvQoHfNyy{cPL!~%l
ziHA`3w_O0IDI(gm%Rsp<WwD7GeS>x=S|u_a>e*<R?)wZ@pFTFIuj_^kk%+Q#?^>bz
zC5UyN6pPV{h=>hWW16z^y+HA3w9)9vTNEZ5WG0uUA9SYe`K64as{WE4s_C6fn*NpV
z<GofwMpjHNRk4auigYR0X!;8D8XTA|d5}9@{3{mRr*$cWMotfuO~*MM+)#~p4r6Z^
zlAErPFpsUCZQ^akDZX%{{>9q^+tylI{_KzVz{b>>Zf#ivwvn*7`J5*}NQ2T!6z1Xa
z0-Hyau%O#S!h#+P#CwAoWPEFc$;%Uk;&kwQouSU`r%h;6wFmf2IJb8^!mCvfE$g$1
zmD@L7O_2sTp}rLZ0(sHlK=4iqp19zRr~VqQ91Y%3>OAGzAj!%HkKQtQfzRp(n^WQ0
z_C+Mm=9JM->afK8LB5XA!bHxyRQQWwrmV*{zZslGetkjjJWIElY@3cfZtCAaQo@hG
zGQcE(Ysb?kx>ZJ9S%aW3ejB9|;JUrdN3MwK5Vue$4w`{Nd2B8Kn})fa!&i-QCf;hV
z2(s8X%<Y=%&oGQH?ZU@4SGi>&voR$FN5TM&{P4>NU@UA*0y{)vKsqpGcF7GEvnwJK
z-iJ2zgw2y?ZL(bwZDb+2DMp!j$~45TIbn`?ll!D43So};09bLKB~9b(<%E=e<@K}^
zchWlw55exXt6E}K%?9*InmJcYGMk)ASKO!~5%)ipp6Jo&6Q7fn;8^KeKF#9CiLjOR
z3|Yl=<k=<={W#{zif$B7c(=)6gtTX-c&2Tt?aNlw1uZ?<!5TarGZ%Jy3LenAmNwA+
z4zXV>NTTai0v3dgaD<ySDyn^AI~3X%z<VNiQpKOV*XjtbPG=J0sfKgC590EYr!;sA
zDjQF=ze6h~rnLj2l@L8m!P7Jkx^f1uITl*QsHbB7^wUo#fBExYPF`!J!%wzIQ6|;b
zZ(sXqnLoY%NpP)%y#FDb5IHNJ2vt9NttT~2;_O+$PW|h|r*-yKH^7N)7A!ikd-Q+E
zqs_-u<XS%jX^+4;73&~LL|byCo~~o1JHFK~SP_E9R6e)TY$ZPq6a8qcb?EFpC&I+B
z<gPe;SyO~xTM6ZIS^r)ELzWywR;FIO__Mwz^V+xT_|^<<^7-?plV7wt^6~eNC!g;0
zax(#GpYwc;``vABTsC%Bp<PMT(}5d=oHmy3qFYg3wuG-k+Pmxp+-cPv`O|SAslM$<
z6=?N~N=^WMZKDf<79VsRD4VUOHj-_Et#FktZULP9CT`(W-(BKe)K%_$%CKgs3-H1z
zo<5f-D1B3QE2m5WKHtUDF=e!h!`RBj|Jd+ZSt<yAGnNvh4Z|=Pvoj+24Ogj12LUuk
zhfc3WUL6j)ZOF~?sot%@r;Iy()1rOz8@enpnAmg)?eXEOHQXT@KFhpzoaQ@tFBfxi
zGBI)ZHVm5zaVF|<SWSAd4<jDK^lFDX>Ajo}_K)!Hh**-3RVf@0Hi_b7ywnmt4hT<K
zaEs-&uB3TknKwYHkMQjhRz6<q+a-Ji?bRDz%dW}JD<s@bP>>AA!pFJgmEo7j$fE|A
ziXrLK<W~pyCo+Ru4RC<^H2nMgYW_lOj{QDFJ+Aa3;6q@SR@bJufRzT3V8l#z=LxKE
z0>ceSys)-l`b#0aI-hpGk%P&tF;+q-d;k7}HmqLCan%=j<;e2X%*hEK%%j=pSK-{U
zf%Qy#er&UppCer;wc>YLr4=L_DDFoa&_5QSdib*a%hF?O%|Yj2&6?Ry-5DD2fJE_`
zBeeqh;e71Mr|EG`2E?PDjv3hOPN7h*)NYCa1G>?mA%oN=!hR6i;13>hbs{oI6pxz3
zE8iq>Vj3^T3*!OX)&Pxod%}8bM{(&!T|1hp-&F77J#FN|?^qn3#0D$ufDThV4i~zf
zGE^%|tfw1es=sV8noTe!c(fe-(+-Fxn7jP)Die-Y-xXrh=(!vcCMzqc`~SSA`-b<4
zXyRezB%Z9%Bb%C-_$Ev2-?*^D!9EY14;+zKayqyjgA;-i@mfv?w_bQT7Aq$>Bha!}
z#L_t@oQhkVyte{c_L!93H3|DX1E2F|OX$zrB{a8YHVF80G`r+1@t%Vh|1~_lCnnnl
zUZ-j|vN7UR9IFStEG#+Y4~@(%k0Dn>;8f<55#;K@(8hmwAg#8S6T(UsUs!(s{-d49
zPkNK;XMIuk)A`3-VWCgymH*O>6*j)Se08c9q4^ymDWK%^N^VPUJ3+S5^<_KyMvwZh
znl3ZWo1!0wX!x#?mI!CyK7Yp4N0}3T(@?OcVd+tbm7lMuvG%$hQ+%1I2~38h+N5qM
zfKaU&LHRnt3+1U|YoeDb5jf8Y)<p%Or6B<~f@<VSW!+SA%xCGN-Uu^OjAmIHJtD5(
z)<`K|c{>~6y9_!GyiI7~M+lf>K#(WzYJa2c`&0Tfp6D?!L5uKv1(ZB*fJ0}ljHh)Q
z$abeBX8h`O<*DM15bEag_8Ni(er4*BNN1QT5~_$)+{X@>O`{)TCiVybG!e*|pb}uB
z8n-r#DL9)^{pHhJy_Wm-qtdr3yVVsdi}QK`mHYtB_nJ894_Y0$)9n|*uXWXn)AC6h
zQ`sNz=_6Oa1oaLkGYR6UP*z9igRF?K`Vo%EYhOik8wF>AM?bGLsa@)kTR+N)t|)45
z{Ir_8Px`4Gfw{5<Em5<3s8=K``bY4=VWQ6bIz)r{an(K_<wXf?*rA6E0vosB+5J+&
zxU`6o@*%+~=nr7jzuFNV)&iR3Ntv}%8@P>=4{bu;K|IP?!I602X&JQ*vWy2!{fP`W
z#u_guZMG^qKj*fbzA^J!FHh4uT|R5&gIBkI{GeMgpFaDAW;h-=rrhp99xEiDSuK-(
zz~UT5zo=`RveE8t&wCf3pocErH-cKHx$!nCIR3TnW6&s9y=Yp!x}Y2(E7QYm84Msj
zte=&EBdsumW<L=dz3@~)=r(jrWtQNuFDGybX9TxA>9fRlqOa?M`Z1Uc6>&<OwTd+t
z>U?N@tsxEep#B`COQ8hqCfvp?q&`kQjm~8N;UDt=eJsE^_BlW361*YEu`iW^`XDDH
z#rY&hFnJD6T*Nx9IC{%s)xcOC`s%DE0#SsXv|DSV)|gx5rMr8*36f4}8MEEj-b-)@
z4^i;R#sNTw8-=tO5!UPCxHmY(9}y2ELG8RxPYpJQ1`o=ih<~e(ZSo5luBVdZruN;W
z*$y%N4%IjHvC$BIxO~Zs$#PnUtQd9E;pUT^vwq;Pc*$0Lnocj*wfy-jtQ4p{`12QV
z#mEkbHkRJ$xVif%$KrQSPSC{Ha<v+vwMs2KC@%*~JZ|rGzh4tN6Auo?$B!T6yx@%7
zN^X3W)lMi4q5~WPw^s%y<c+@v^HNX4a2w{W9{J=kTu-b_uA*~wFtXC*`!=NvM}&!#
zkEx+6b_v{wNoS9zB)fm`rEj%U&Ae4^rZJqC>>Aw7TM~vrFy)!EN=0lM=ZBwd=Pmf~
zExp~}(M$Reu!Z%^yk)4II^*1MtIN20=L!yr>9L8*z$FB4Php<;@3e;pM}#&(WB(Pp
z98#ztP3xP-_{W8IJlIG~d*giEYUPaY`TY9JuO}b0C*;HLzsu>kb(=@MSY6}xjQWp!
zRUD2p?J>d;x#0aT>X$g91Up{T+3&m&0R1TN4}Bx>0L=#D&6cD=6isl*0`uvVWfPRJ
zuk^QSSjgv+2u+;wGVO@>=_uTphI0se2@qhHwO-S~vOs77bQ{WEZtDr2ae#GvfnmA{
zs@b+5GmTv9$sMfWDRFnRP|%Q;%YE_F6q&=DgEhu25iC@*#w&{KfHWYyI)0k(dh{W!
zln2lOa*z|!v|E|zR9i{0iN*=!kb^>1p;lq&VGUN~MU%!!(kfNFq}dX@BYl!2dK{|D
zp>$QTW9W~eE%}GW)e8=z-|QnDpB>ez39O@V>lrS_@HT(-OK!EW0&>fXs{FF5c&dR(
zj;mN`PPOvEWW!)`etxZqhDo27U}-XBA`>pPW#YiWps$c-LLrZ1@q=FV%?Htb(q>fR
zckkZ$b_{z&C<nJ~V20Ti^Uc<;RVsUG<}p=ouDXTTpRXeX;Xihm)3=vl8w(J&?zIgV
zxHaxHUbG7|6e>{eLx$sTD2Vbl$*6!ooDk!{K!b7cF=Y_!`jxyJw|&S(uM53LgeREZ
z`ekOndy^_U@rG4zbY|bsEfaV+w>Z!AD}=|(<pw9D)kpN!P5RQLu^Q0TZ<n5rT`nEa
zCoE>oc1Q>uE-caiIUH|+vZK023v;k2oEp6&GD4LfxDwJd=n};ekkbv&5*_w_Nlpxy
zw#sLsKFCytFxNs#AXTsG*C}7p0|osC4CIV89_Na7eS94KsBY_p9*?<UhvQj335ROR
z1BFfd$OaoV2=j=;q-kq1ui`?ANlyKWm8qgpKgE&Ys+g-*;y7Z30a{Od4>y`H^jUU0
zA97Ca;<^E6g@eEq?YZ`66m!1a!p21$ji2Or@HEWNKmTke<cA;L`<4v*IXK|!t6y_M
zb0)-1E1-6ph>aURhLfz0ihaoTF!Fg`A^{9f#)Sv&F(GAcV_~_{q1tAtK65-=mn@)-
z&7-T2ivZfGuq2W#FcES~m+@-+2TAH=f?m5yH#SZH^_P-UK6by-Z8X~LogB{3dV$%^
zt==)CC67-ZSy_vBhuC2i@xu!_9Fm3Of{MJ0#NYbib`f0kL>A*MeNwFB08@D2yYG{D
z;t(RO0Uc;}NSw0yZh~iw?G`AbPy6KpT;RThe$yzcEMa@Cgv<>>DBj@oS_On5)Vb2c
zCCB~1iW8)<RARa>1qr}v&?6s{L!&}rz!?TlIiP(RD3mwz717EVUbaTVFVPG}qmVLc
zY#EgunAwoA2cKcM-h(Jnso62IZJLRB)@s3W_<^JrQI3Q&W;EgEex0yvTKYc_e=1!t
zZbML1O^=g6d1=Pe3ao<MT<g6ax=+tzw!GuR&bEnI_74W9dbPI}@U{QqTu#b^_lRI!
zCKVB+1EEK8wJgufhhxI~Ja~kYPs9K9?|(n}+h6~B^4Guq?c@iog1me8gEybD2ZTK#
z<$#2vKdCD!D7b#AqivIQ%>thrOv|RMGR5O_42`G37McS0%E!1fa2`_gY&<Sx5%ulA
z9b;ZvMIykb*+VeKl7gMF3ZeQ%`^H&9+Vn<(L-YY!yx~&DvOdnp#2T0j5Rrou{G^RC
ziuxie(4>~zkIlsFGs5A-`JjLCgv<x+5y6=_Ir+p*K^^Kd*Xkp3hA*|}h|d79v6c6a
z)MM5+pFan#c+<ydbIii6MSfl5)F-qUeqFCti2?qcI>G`+*wcIKju73GMw9mqFU8P}
zPN-0QGsZCTQS8FN=za$%U3~S^F^ICvj-DhiLd7IJpVpZY#gFn^Ca8FDG@A5Ik!6vA
z&2QC8Ndr4RjeHK)3Rfh0deS<25FC6J<ODyD6H?Ss%*xRFGC5NAF<obB+I>}D!f$+A
zLYSkphTyUn6{wR_Xl`DCJjDrzJjHAh<ndZImNK}H>z&L%)E*G_fjnqM1Sf+No>C6d
z{CNYSQ>1mTUk*mU2ZTp+c|zs|&WBb?xD8WpO`Ql9*O`EFrJ!XB$1vg<{26f8DDbP#
zFVN0&TmdZ|XHYjp6|U>yDRe{yK5fP(-%9(NH)<1q4LIA_M5IA-qW~cj0Qv>ah|4IN
z4Mj}hQo0C-a#B3Wqz3il)mPf*!@)*mHW|kz=%hmr{rQO*>C4&&^z@f<zHv%;(x(E|
zQkV00dVsl_5bX@5V<DGeh>r|nSzL#NcqY&hKN8jOjuiQ|$dA>(PP3!+pJ}$zZKf#2
z1qQ$LJ3^$LY>ji_lOtifqVjSIue_fcoDjnaSIU9sbe-%)ctFhH8z(@|JsAmP*}M3T
z^cV*H;5hJY^@5+kM|qUR*0L+Jps0&A;Kit=nDhefF7Tmr$@gHnqCbWfgNOYOt-zsl
zTPiM=(Z86CS|;@z8p2M7vcnJ_bQ<B^j{=iPAohY>-+a~;>nCkU{cM~0KE3$kP%u$I
z#8s;8k=0@^NiYj(o@BHMjHe%95)Oz~Lf*bPd8e0+{pHVp_I?odhOj~sD<7c)FTZl0
z6H6fm3jJ2TiwlMtHL+O(0|44Gj^R~c(u5jkvKj$=gx{WD*r|iebVQd5PJN=*p#<=b
z(l)0YJmA!oe7(w@R_N)Kbx;dbBJ!RE7;E5@D+wxb+Aw9rXpUDf!la>@by50CZ?vU<
zaN7k+x(aQ=iU_ZDfA#vM-=g~R)s>!<!6DI8Prlu!+kV`hfw6A8Xeq?E1n=*3OGbN2
zN=}Tm+#VFJ`kTdDh-}8#l))|D8?6c)qZEk&-vjsvXP=Ci!LP|fdi*t5ii#-Y+v;h*
zeWX3(jx7*okBcAQUaV#<+vZqNeW<lfvB|z(<3M!_`0Cp+g|$6K2DmA!wR<C1!u@kC
z!|md>mLjuk5Qu%QX*A0N;{<hww@c>^b(ZN4U`gD8)@%yiP@^M-?-IJT{42?JdweC?
zvd64s=@flu2?6*(Ns2bClTyj(c{tN@LO9qH!hnxc;;T`vNOf@K>Cdm5oqmvQ-Z%(w
zB4~1mgnq+uKtC^^>#Ep~<Z4p#J3;g%R`T!;K1{-5%VFxqsPA|I90aNig0w=bYg4E(
zfv0tAI8jt9mzxxgbpJ#zqfeI(GaSoq6@${ZQzNOm!&kPW{#6I;;D&0*k(j7pL%6<u
zr4d}h>r$1*&_CexTUy&R)FktFYqZJqRL@{rMeC(@$^9Md_wd1o2)*i+Px`>vNAx3B
z*Wxx!$i)Wny&B!M@g5L4z<Tlu#i|Y)^;79MxCR1+)}?#!sIoD|-Z?4JE(`|&X}q$b
z3T#hnzPE^Hf{6vn&edOC9JXym6Oy5itALK<_Mm;DKUK?I*m0N=FXqLlB!ZJY5@t1^
z84Oa72k^+y4-!_sKO3@g=+omhk;W~BxGIT~hzd@1hO{xJ`NnYn1gzkmL-?jF)xma@
zA^lixAnz}zqu8iVSGFrzSH8RO=F1n_VIFw`jfwQpJJ3YRwrupT0fRQ{2feau#eqqk
znW&V*(Eyl@4lO4h_vG2La;xvW@D2y|gxv8dcLsVMbM?t+O7{<wYn20PzeQDicn1i9
zD?A507Ckn}75JrIX2yF!cmn3HS^@bnpNM&*myz)rb{rn|2>2ZlJRBi;$nZoK-259F
zs2WUeLO>F*b`_JjQU4lJ68J4dRH_TrP9vyryb4r!MU%#8D<Fc`?HC*o+ksfNWO^>1
zIDcWIVi<9(XklZ{>pYAJZBRo;ik60_g8<QQ)DG?!1AaI}@%Skkv6+-WFW`&2d~opf
zOTC;<`+|6Z8MojF_rjz9(06zui&eE3pJRjSoo>~gOTTl`qvs3VlDR!o9~F+WxgtU@
zL)*{=+0LS?{vrrg$t<K^gwQguj{S9;BDdOY!qB=+VBceQseY^K(%NAa_6{WaSv9R%
zn|VqUz#Jy+fZk1a+9{nEucSf4*rSLsW%X;@eFDO`Ne$U>@;TBg@u-!Lic^QimV_&k
zI#fxoLltG&aFVnFF>D+=U?i7IHc(dqEPsh<!Ly-Y7(>M>*3U?fWC5BXx>Ytvi>iU<
z1Qiu+)}@uKQ^mfCPgYJ`IU&3eRR?=Q?mWl?jln)NKu*Xs89623cc1TXi1)_;<7C$4
zA^(RZd7gM-F9;twtJkyh4hb9&oQiNPLXY4ba+=&y8Wrr|h`Ux+kmEcn0^dQXT8&ig
z&UePKHb{0zVUq+TKDgAnIbqN$@7q$?h>$BKYKksIW(4X=ofM<e6%*5kYM>s{fk8VD
zL1|=*s2msG;K3>h?}WkW<nT&}Zt0!rac}GFIzbCv>_<A6!{L`=={8{e!ACT{6(d~!
zC5JzUjLdLQC{suY`wTS`D8L)o`vAfUNioVyxTGwjB;Gdk3H3LWEv+j)M73(HLuhAc
zdR5wd6*BswK5Z@lIE*@!l3M+AOwbOeLGfd7ES7ZOmO*BB(kpvhXcZQlCesoTG?qoa
zlNM>$Ku~Zg959-mv-%c($8e9N2k#CaSrhSYmw@UtFEvdAbkMV7voi_RY;|^01_JLA
z7^$+74m8*CFttO96+ec-9?Ev89y)um?q>-g(%}{v8P-T{CGXcLqjj`yL($%A#p8|@
z5bXiE&38C(D~7=1xjas3A)5010x-#WxaYoq3MZ9LfELa}rf`^OUh*`IUdjG<Jq^PO
z$iM&l|DrdczCC%by&&FKB5bbK{V_C_?U`is=eF_;E9M#a_#K^xrznW)iI7`$PM#>S
zHk6A;j4FaLNeNbQq>CeHF^SdrP;oo=I@GmEObs9OLC-pQ$WvClD+Z0=JoOT7g<c=?
zRC&RSQ+B5wO5ag;IRV%axi@}8mF%O*-CW@viHL>EN)(Q##adrD3rPw`zoH?!+iX!W
z^S2JrDBP*<_*Lw363mPzZq!!1c#h4->?!(PD<Gfrn)c7a{h%~*KYe0<2W^fctXqP&
z>JJy6#o}J?7I`Vh`2OVT6>TcES~0uVKBHUpS?s`;kTc#h!iV|9qxOclZ<}KC$15PH
zoO-*in`rAG;zLWvpGdH6m}8UPChf*A(=MZLGWdod&2|Uq#M4jM=@9}KJ)w&Fw|QpE
zTfoo|Q6{-Kv?3UtL)JsM{4<}64`<72Cj>FbCU%ej2ArXT3EB!1=dmKo67xC!ScYRU
zR9VF!_URzpgd{@>77G(+DH<7U38{nka-b#juz)<>?^lprA}ijzvIi5Utyi}hf~Lp?
z>v&NCgE`&!WA0)y(j>t713-Z^2PdwVIt>S3#NZi`G}pc05Ght<=Yn^GXy3<6O+vgM
zgm1yHBElr;!5{@0UsTW_$GlSqu_`HsBOCmFAh9Oa#|4lWCw`pmemce}--S~!LJkSG
zne?MB*skYc1NDH-^ax+Yy;>t|c%=;B(3K-Y+VT^QgW_=!-2|&Z8c4PIvTSoqvOZm$
zIz7ddrl{!_8R$iu(-+u~TB{+kkHkutLDR=hk(><oNmfMYU;1!>Vqn}(V)YE?sICK@
z6=YP0Q3ei=AKVPZ*78bd#kPPHQ>=gRfe=VPhex)+)TU+Zk(kL9PPGWM9HP<2)G4PC
zhf!uELp`xkQqVFq_>gg`%a0iDktv=PL)v(Y{~(cqK94u1Qt4=DhuHHPR1sE(V>xy9
z4M=;ZdX1Kd?Nmw@gNg?>>M>PjiFg8^RAExOJA0^HI3K98m2w{wgRXb&RyHxHb4{G9
ziF!o$5i<#B8BsB<!XN(;0y(m1O+QmyBQy{Uk@%GHk}0lY*$=|gEUbPoF?lnp%Hzow
zoD;1w)ZP!i;ZaPK;yM++Oe(~&0rjO;J>KX;W<Pi@2yaLI%U}MYTQgV5Ta!tyZusMK
z-mIvvu4<BD!sKIjeS4SFK~+#6J<-v6il$zL?Nt#Jl8UxFb&yVW2r{iwhk+-y0zcwj
zt_$}3sR`daS*^h4&Pbz7m+Z<@43;0aB~3;102jQ9jYnM|QlNf>8d`BWC_9&32>g;~
z9|^H#%W?FC&TS9bGOiM1HG#H9bWjm*1{_90I0NByG9(!+T2|BOn|FH3rC#0tR=4AB
z-@iZk^x?zFhhKj?xz>m6&h(ub-k^%3;my(F2~E6NHQrO>J}HGh>wO{g*K2*#2y5wq
zZmpUqsqH1M89Gx(4O+cj1!a9LV5v8-V*~Z`)wAoJ{9#JIwN6P1dyT->x5}U6eJxN;
zSt}ydXK5GgTxUq&VSmm=|1XM)CSgy=z5yv{P+CEc%<nXyI>uNG$J8yX9IUqrN$I_;
zHTa84S3%FPU~A)J6*x?G4QPG$TB|X5Ob%Z*P_MpWk+!0)BH^_0Bt^n28a<yS9^ecC
z`(T@L3jDQOUHP7ChTsug91pI5*%!duPzg+gOoHLKU?GNWO&(H)$?Sn;+esa)8G1s`
zVV#FO+B)aLi3OIb0yBkVYa-HzwzYYg<M3mLf-Fq8F~^Re>9Awbl!CYmrCjoI)qo{^
z(b{nVE<1z6F=82}3!7y*Qoti*ab)ys`pordK<)$FDyty5ZUza&4}+Ti;XXh=(BY?E
zlE!*jAiW?Y*G5~YX=@x4oEtl@Xd`~UNTM7hg~~_9zq3D`rnBX%fiM~|BZD>ttaNHv
z#Rw=A(tnunSe338Uoc#W*+8oX=n?3kAP&g;J0a?z#j;NFnmBqDjl)U#NP&}TBji9Z
zl{L;Q*Df12ID{#7=s8!vO-}QWI?3ltzbte|wYCcPK(!TPd*Pw7X*;He+D*M(%A{1p
zo?6@iCS@zLo@^6$o|jG@%K?cKX%JQub(~(CSBN~tz`l=oO}nl})i0SB__X?|Ry^+R
zW1slQcEw&m{?6EBs+9|#fME0BmEx-xdK%)jZo{w{RVyMKd`ko;#LMiu`ex<j#VOy4
z(N!(HbntdebiNSXsa{*?zM5l$#bzg9qV<MWCRolx<v}}B-n?uEQNtH5s+ZGUCn~#^
zZEQPbL~?L|XD31|gL^S`PO$0uTkVAXn)ptU(zRD?$^$#F66DnhWXKkT0guX2OX2tz
zM=&-cpR-fB2_B?C3z%w{Tk6#71CDR8X$9l*TzyG@I1CP=^J~~?W8IFq;?|s=jC%9-
z^~vwQztZiwm;U<lo%$hqg%Y6%-=m=pr>xc~{Xr`z{;YudEAsTPjVPqIH-{i?rUF)C
z=BL>fna|;=EYQq;q(J)P!C|6A^%{6Ot)H<IlG5g6`b0>Elozvum3u<Mi5)B%{q38O
z>nqczjk)V*3qc97jKZsUj;{$+pqR^yZK!kPj|w|R_%XB$cH8M)x`eU>Ut=4Y+<o6a
zLo+ne5${5(n8HlE!^MOQEt4{n6_b$Pe5zHDxb*@)&NZ|+2Kr4I!-*e&n4q#y*aX8w
z5Ldn01CXm9<SBz0=PipajwKF>OmIjg&EelcU}__1@-WKe2TjJ>Oc2&)ONowImmG5m
zWkYEpfwzbYC3T_>8Cps((mlIM7w)PUK4mxLRJw?Jrl=x#=%8QYgV-Y_f^pS$_$n^u
zmJerzGX5v$0<+<EP2^BPCCYf8ker804ml(|dh1m)IY4(h;?_=ZF)c*!F+DtRu;Mln
zeaZb+d#Ksq>dz3MzpIgT$(m84X`Ns_4x(uqX9*xe#;&*10f-h0l;O7+S4>Ht?l2Qv
zx|BF$^qG0`fzdQov>99P)^1Z|<vadeat86SK&9Zm{I+OUMHP3f2ar-jCNY3TU#)ha
z;DXLp!71`sasXFxc)_kyFk%-ry(R`ol0$_XFYzPojc^(qr|@1rf;@O`iNV<wnv6u`
z5k<elVHe@b@~)a$<*-AYKs_6O?!5=hq|>BX(%BH(b?q}lx(LDs3>=3BU+MJ~-G@;$
zzzl`zrvJLf-4G^n9`tE@vk*LN(+xp_d2$qS2Q^>v)qfiAy%Lh@+ay7h!+zsXJG$vh
zgJqFfxv1ZjoB{NH(A^O3y<cDRZJ6tmkNV8|wcc`hqckfNI3jHQ@KY+fCBk(%7u2+Q
zeR`cwwKI8Ka6_I6M7Cj5>xbWYD&`_qKwiD{9*`gOT^LqIa7K8-g&PBymTPe_6&+X2
zx~;D?PROMuCdzNrPi`M{#j6Z$bEfx7U`IbaqPU&kX$fe-<!ak54fO|4IpZo_`6RJ?
ztVF$4=nDXhh?Wjz!pNDvl0zJ02{q!hpdGy7azL0(wLXD!21e?Novn@Af=M^dx4Cn!
z@u+%2s{>mQ?qqY6y3i^@tUhJER91F5(|cSl&*E`u_Jx3FXFzr7WwJ*Hs({^}++2X5
zexjF?f&b~_Cztt&81du9WSr|a>KpW5Rx(%xV_(Si`yaic!5)uae*H!CcPCfs1i$gb
z3~xjguGl)z_7|6|o}Hb1{74)tcl6WW)!)v~*~fGet8ZFW;VaB+d__-=Sn1I!i`oT+
zp%Yjt@zj@WquYdZjR;l0%nGk;Tzfx^Ec`}Nq%BqnvKxk&I4wvDB1X5nJ3VFQ;^@z&
zF>iQ4B|Ia*Tmb5$BiIB@6Wb@C1qt2`$gA{BtaHW1!%rYhh}<t!vT+As4sXgU#rh4*
z)j@&H(ayjP0(~@p(4NJ@0zpfbLmmJ*?J%-Ujbyw5$EnnXwi&O4m_i24hMX!L3@NV=
zX*WVPdMha*1?c0#O`N`{r3<G4o)b@QfuY|VFS_qGJGeEmN|25rEr>kd1K5M#4sl(v
zn6f-p3J{l^ij7?eWO}4bx4xSLj!tIZbJw7(ku6SU^T{8P1xJ1(6(uQD5pTm}puF=&
z)mV1-J0LUx;Y{!)RbJbDq0NWqI&dPr0aE;W)zOIC1024-roId|vMHF>fb-JlrpkyJ
zI8`UE`nd(eW=v1snvs}qT6Oz6O=>bJdfKe(GKg&U0Iq`ppZZYuXdI^of`U^~9h5eV
ztbFlgP+bm%wT$5sqZJ>uqZ|hZ_f5g|@AE)-#<E_535mMI<Pj66-=ZpqgvYd*l<Ia`
zFif_+&NjrLC#!e}Vc(FQX0`c)-f@Er%gR2JoE)V|nh02gb_D`p%?mIt&hr~A`n)TK
z4pW+z?|Jp<fxWA!BJ(SrRc!q#Lo9%C=Xv#*)-6d2zQL{v6vNSN=X7YZS+9;@O|k^8
zP<IWxtvDqMqy>4P^k~7aAe%%Ncv=Rll^A45(_r>ve8P>1%4{J>zx+>oT}bu7#YyWW
zO9sVETSnq6MWu4_u6PN58D5;0F}_4eUG0}g`z2(2S)Su|KTg&{8!Ysjxh`z^#$Vud
zoPwryjm5rdUXtAB0SeIf!#)jz$QP9WoJbn1S-FUpk_8Cd7KFlC=K;PZ8x@9WB$;!D
zLw{bSz`;=dmA(;z)A3gC2I1p$uU~5~h^~BJX>SPM55Ylc-=GO*zEYJ!;=>9Ihj>c1
zPE9F+H^I7uB=H>@>kxJE#0qX6Rd&l#zpm~zh!~$A+b+kFtc1(9PtJ|}s6@$#cEHxw
z#yYts<kJ?E_z=vKsbcUO4Y*w}Pr*)E3P>Il8z0;W-;UHt=vfu@+fTbLGQ}f55#3O5
z9MOtBL?jv>2_5(kQJ2`eW0_h(W5w!Ht6Q&h{EnW&^&S=+O0_o*CVb-{O=<{+v#KEh
zH`+6DuH!<hDCgRj!m5jVr`TE^&W>hT+e;<Lx;bUFMCAj^Cr^2v!y9T!(D(biid61C
zK>f8+b*iPzVU_cdRZcyy%Ur1WW1}qlnGHHBtcoo#9N`|G2`T(edr9&v{U#y`JgqPN
z#9z@3meC2TT?$4An}&jMrL`DRsvQue=P(P_o2_6yUtrOxd+6HqyzKSVOPHOeU2+Ng
zQ1};Qy;Y@0le>cv#SDanMqdcnR`tF}p}VXU$gaC7qm*<>WCqOv4!OP_o~XP6!in4W
z_Fe@1MgrXDr@An3`rzs|pNNU4Vzex;rEzXg#8WU+Zjmnumew;zKji^gU(!w2s8`We
z4JzN);+si#qn#%f)yK~Y3GR>3DXu3%Tp#e0rs`KQbs}w>IWypr-5rv0hrvo-p+!D=
zOr+%nRAEbWSR^AGx(b)cj8zbB=ka*9Ta~(JeQ{d)P2hHP(hgCtWGKd@z{gefei7ep
z5ma*ALmacHHaMk}rO)KmHe)7B&ikft2y4IaE1pK9q*Q5G>k`37za>s_IUxPKLJf^;
zSrX=I2og&<>x8YrLJvZiZn0XSZqrD3+bkx2#W#5q9U6+&I+HJ8_h8b=tOuhXmsQr2
z#u!#Y0=t%tjL`j*^HB}Bao;L&>*V*zopKtRlG^~8nBoDJigVy8>0-WWy@;@*V1_>Q
zp<rCN=)vv~u~<B2zlip27L1jjvyFS2(E85qzDR}kX>>90NxBcs)M(=2Q@&TmdY}p=
zQnas<(}i3G;6&Q*W6+hN-)_l@2>Ud6Uq^+JuHvqH8wHAa0ErBjadpZNbfyuV^X-M6
zh~Z^queB2LS}P&1UMYV0TKDO7TSk)&4hn&#e0advdniN;z4!dEf+0AoFm3><q9dG;
zoU8?6MFqWlOGq^AQNbRJ(H^H2f6ZWhI|W)Er4?ug*E9245^^2AFNbm&dwQ6#`=P#=
z=T32O?FyCLO&|D7XvM)P_%hJi_Vi$dl976OiNoc|qcq3?n!aM4+-c0g?Q^bg3egv?
zv;xBF7%#Eo3Li(ppA1l6<eWt6y;e8i;d4|hN?dSLxt(bGufBVvr*h8qT^VlgoL}%3
z)$CuPIvpnTNkd(5D+kr#fvki_lo`r14ClXkyloLWhAxG*Zya*sY@#pA`eo;2#rfrS
z<aAIw;gAu)iwY&;0g){rc5ZvcX_0{wG6-(NOcDv)*Sjk=QnlXzhfhyow8MT<rC^7>
zU?a^t_A_c*mh0IWZCibt6B3jnm4tFeibp3i(Mr2$o0cn$tg6aV+bO-7bU%&0gN>7g
zQSrERckWQSInEugE;)y3*czNgVioK%-Zt8<J+`4OmmMg4g<Q-6_voxs4JchO+?JxE
znddo`v6+v_Kmw-Tcc+SZclnW;lPe|$GzKy+k>|u0C*xq(c5+H4yfGaIr;hK(cWMR|
z6Dos#n5f<dg45yeeTbNWHcvfM2T$?#T!9Q65iOf@iK_6x&P3`QS*QgC769)pfv;9y
z81!&rh&}meQZx2cpgGU~xXvPi?%0qjH#Wr+xb<o!03=y@dt4%o&Z#%JrtM%osqsWh
zXt;@Bx=gEehVoL57~AL8jznFIdVY}(iSSM(8*PZ74M;`Xg2cq>cJj(tteVkR+Jq`}
z_?^lzz&W6g)@$rrpG2JPA^t2QSs77z1teBOV)JQf5wS4G!FHnz11PTZ%jZkfa7+Li
zcw?w)HwLWCabBNsyAivj^HDnVJY8?e^+h%`Hn_+R6T7Tk^0OI8+c8DEu`OIQ5+8)%
zRoS8t5CVM>Wlce`7;Bc48-)d+yi0dUw`-nr1-=N+x2wJyfj0@317UU|>XmRuR)Mdm
zj39P2p0R?p9T_}ig`QUp0`A;JYpn54$sn#{RIxGoljWq20zXmD5^OtGhBy%2!Fuq1
z>E7!5QMx^tbq)-1b$qydCb|lkqCeT2ZN?@nbS}q+^fWc)HB&wX?US#5XPT+3RLj?S
zGb*}oi!gZTYE$Op3X=CY+~`e}@ian>Y@JPn=i4Y=ae%STB&}d@uG*y#h!vE$KON`U
zxo~*`g4K|hdQ#!)<>mj+-kWgAjU#!Up-|||t17FiyLWot&dhuJ{lC=go1X2O?mAMb
zE3W<h{syCi4EIpDOPR{d5+dlh?1BM=89|U_bp%HQ=fbBG@)k|-a-I=;uDUGmJ0ILH
zl7lN`d+FV~dRj-$3F)wMGHi>o6`q@3*#p8J1IsmPK5^p{JBH=lp3<o>KPDbWl*c?L
zG|aJa%}P=mVenY8T|8n9tx8pP<w>48)7ztIbeujN_$f0zhpyYiJ?@QjPOvJV0@gF?
zOo%izq~0=__sEk)#)4C1H#tH))k5DmhTZ;kHLM=b%|{J+iJhxw{T-nYfQVz%{ULc$
zXNn_(D;JbWT|=o<w;1z!iCMkK%yH@1Dq(D+GOdHMN#oY@8h)z4aGy_B81)<UU<|~h
zAN~{F<bk7{O@)Zkl(l55qS~lE=;9&Vq_5P0VxPvlr3@dN0~u*G-(2_z)a3Umu!$H;
zBVi;M4{0A+={i&<paT8a-ywIAft$SIIE7=1KP>i;H=dXHLnsiJm5`nkx^5Xw1A36S
z+Z?$!{gGH`0KE^-$tqH`<cmZiG_mfcov{0<-4*Fh$t=*F%7+weS}|o8DW|fT;!-~s
zJ~-`Zu`^l^Mhl>v4+!-DLgi~e0B{dTo$5Q@ONxZ)>W(2@7c4-+OqswxyaUDTA+$%d
z`-@<cIeb8s2Kk-ENqR1r`3bv|-eK(|?PI#exgFy~PkEq#^)${uz`N6+(v^1x1bR4T
zIzT$S8ku|_#29CyZo}XlU=Wly1`OE5l8_BYly7WsXhF9gwT~rBaaAKj30hA^MD8Nu
z2kqpO+Zh@u2<My*wr%@xP_B;;(2$w@H^h>c2gi*%q25~waVE$&f8nAdX-vn4T<A+9
z_+2Ty_~uD3VJWPV(uX}xivQ*Iixm^vDnq%{^QSW~@yZ7VMWh{ew&Ov$e1At8N?le#
zsLw5v;duJa5FL|R(W=!f=a<?PUAJg1=udg2Z?(QlC}U?oYk=nF+6;<!p=BLyQ_O@_
zTh)fGBLulBhn5O{`VIM+$5eM*vWUoXOD=hSBFQka=`ZD9$$lKiEwqkeT0R75zP<NE
zRzfTS4!Foby5SVcc6484d~KP1RK6L*GfBtcOtAWJk&~)a=Vymo7}cICP3s!EHl{8U
z$l%cE3%`<J=Dtcd{4x^Pl6DJ%)6P!gEIbCqE;sgwZ1C^$0roRS@LsilEy=EGUETuj
z>bvCFct70_QUUE_co)~xc*}P`??G=vcH6saebdtIAi1aU(%RCy2mDUeEL!NKfYiET
zjW;TXlV5enT(@FyKE`)qas`>oDk`BD$AYAjzKtv+SiJfavZjJ5>U0{iGEuMN_6Zr^
z*WlHzPeJHVXp};8(5|W8-_0LP<fEZ-l_h&vCU&?>On7pwM{*wvjFEC*g!6KJXT(-k
z$ho~zR1=$GX~=gfIyE_Wi8M-&d#DYbcA|XqWQZ~hIZE#%FDzk8OJ3FlUw}=xLkq}N
zD<z<hfB^~-va5RY5HIeP2|vqn`-8I74hI1;_J?G%zXt~-D-wbMmzA$L9pj71rg;a5
zPRQ<YUEc%3ArGAo^iyB?-RWYHCv+W~eJ(`0lsZEEi!gH6X(G<ZAdxM>tCM3vOJZkK
zi_#c$1E`~L=*;Qy5OblX0z1U<Amh1^s48cW20b<2u6(o{OL-Gk0D|>1|7FZ{q2no`
z5I%JJ4d4ev@NLqA{x~&~>$>eN;~JyLE$)QB!>z)8>d5kXc50tUJFI=Rd&`F2R&{gP
zB-0{a>bpf?&M+%NpDUw42ZYM-nCeX0o*p)pUWUzSEtu#ZP2eMNGclZuKOv5!QH_P6
zon1jDwpH6Ebjw?xg8)%cT5vksG&lGQ+y@f!8D{FvNfarXZpRyXah}Flj!2wCPRM>X
z-P;yU_%Ut+zqpm~edXzry>!|L*8U9)$uKhB3h~=8*=&g8;buc$$L$+2y*;z+(}1^a
z$$k{&;E;b_B7~#zh|joFSO$zs)sn?Bm*#C4IPpLz<2_F|I4W8zo0r#}lTfw`p09h%
zW4fyVUBhy5)0LKwx-R;ZaqE1hX<6@euvzUWS>8IH;ZIFVD~EKKza6z0Z*FC%Q~&(m
z8?3S$tQ-pwItl8n)t6hP^q9e4`y@|Tr)ZR?e{dGF8se=r4#$TpXB>Hg$`9U+H<da5
zl(EL4ydq_I-2EB#%Gd2s2m_HDPvg|s7K+rr<7@mwyI?g$&ND@#$^xV8CI*6}15<!<
z`7}YogY804C9i@2L;h9yaha7iWjl0`@>zz|NxKSb4R7$GXrU9hg3)QZ6ZRC{fw`|>
zkRF{gAU)?z7V%Qm8ZbG3bhrmrFh+ZWOGWBN-}^4UJ7Hhh9iV*$cRFwYSzglbutvUD
zMwW^#+*0v56SF!?4#lj}P2DED?!;r;uqF2M;eVQL*vNk;I|>e!cXLPmaso~RbSZ1t
zkB#NUQIp^xw#e5f;kLY)iddKtk#X;2?ux<96)Ol{ZSv>aM(1Ez-Zq)r1>+vOkPy+S
z2rV9V(V%z><|&WjKJkr}I3Um26XL!Q%kqe>M#qy|-mpo$$bo^t5s8En<?PSW{FCAM
zFq!GJ#1MtQlP^BVW8m0Td4thH2xO85!FI40<uPw{?GRb5QuryXb5q67kZgk*nc=4_
zKe;?B296v0W2}Uf;avf2v@lEP=%~L82>XYl+ajX<S@|xVY0G=umcfzq4X01oJLHoy
zl(MQ-u=|{l@Y_gc)~;2jt;z6T#<Z0MBTc2N4<w2?q*NE?hGAN*0svC2Ng_k6@`WRn
zPZAVm;Z+nDITUUV!hVVX-D(T^8G4T*puLpu3L;<FPW%_0&2Rz@T?W4ED)SM)%{aZ4
zzOZxC^@D16`>w2U)@%>Kd0y0We1n3lR{GyjUQz5ToE7W}Vq_+|UMneH<E~n=4|hl{
zPVn}@Jg{uPS{{hJ1%p`2w5r-L=|Qy{Nh|0FW<IFl`{`_Cx*vE`yBgG+x?{5&ITqdQ
zRV2hW@Xc;H0XHXb*C#$ia|Lub(>!JwN?r(WXqTAvTfBu?pfgc>Md)OlkA`JZ2GWxO
zV*M9fOTg{NVe8b99{<4mo`>uMNsJ@H=2GwRyE5VxuJEq9=;-)TRbQ$l{kT0@&c|@6
zH~Ka5=!{?$A7dvT->}L-o%>eegn;YF>uJAVMrPU6+7=r?Ya|WFDZK-*Qr<-|&;xD-
zb8Kb^(@sXCmx1iuIzHG2z|_BC#_zIJ?D+|A+qF6NfL4(#_9}4?Mya(`x5PC@(6#b%
zsMWe<Q%e(-E@VN3uMgWXUigGXz9Yl<d3^Pl4;$*RGD?u&aglZ)N~h7rkh_E~IAM23
zdUdqPF}Dqt`8OzSs~?uXOZH)S4Yg%H@{2PdoMl_M#{U$tZ5Z@hPYeG|`9XFYJxBKf
zw|+vWGY6?0S<&=LNXMPyU=EJ6hlGaPp}R441WAVMt~NDmRt4M>4U&Mw!PU{C-2!P{
zSN$j}IP^7m?4vE{(jFvk*dbg~`fRbX^#E!iEUUJRnq5H|oI?$50{6uy;+7;^N$+E{
ztCOXJWf|bRs*gzB)y@71zGYU4yHFEdDON|}ux!Gtad%WckOFNtRn9b!X)jq323(Z3
zrk@F$u^z`j=fHJ!z}bOOfJT-9V{)wzdAf|w^(tOSmeO@VxQcZnYHq*afYehE@oF*#
zH6ETb?%U`P!{RxoW2v)&(E)(dQAY$3coxwo7c6(v<CXQo8=#Y=b@rj`3qpA;=c)O@
zdoGh+U<chneILo=$_Cy0kb1&w{hE)y(QV`yGTlex!F6fme&`<A`)l`=HnDZwx9rj$
z8lw0f8DF}c^xw{junS%3stu_=73FLGAM1qBr_!h)NdG`YJi~6xfImc2it5wiF*`od
zr;rUia1vEzs-$sx*hl=K)b?=R_WK@?hf@C#+}!symlZ}|^vq=|9d4Q&I+XR8j>^Xc
zW1V3`k`3O6<U3g3bQaxof#Zt8`UibizK>Q<M|~?Uh@gm-57;gA8+&w424a^@akFbh
zER)TSQ=q$kLS2$QAuI(WVY&3TGTldOOJkdFv9{&M<dteoXJ~Lkq@>>zex__>J^*My
zm%o8`^v?i0wJkXrwjKBYIciMQrpI|uGvz%1`xAl8ot!RK8|zRd{*KfD*-Dx7N5jr6
z?;F{jv<qbz^c{&f8GiT0pa^j$NVo!`0nun&<hDuW`)XJoIypWD>}?+(19szK<LkUE
zA^aT2-cs@KkzS1u!f_+t4YYAMG(^zNz5!(Dy#(I&5&j(K$G*zO$nma`g~u-;u{AC#
zl+gWCxt*Xr7tM1_mn~3(hUMbVunWvRiTPZb(=&?ajWQ~}N6T5|OYO2JCr`r!o;t=@
zJ`KPM7nzJ3%QSN3#JYdSG30GKV#Y}CAL`XHAp&NE002M$Nkl<ZWpO3U5IK^XekgRT
zxv}>gK8`Wr2TBR7(Br6$vhwFz487%*^=VhN>|PxL(z-5%PjZ})>^Y<#4~F<*Drr0p
z*)n1dgC1SeY~?Y99&cNn94ensf5zk<;rA3NQ1{)oJ^O@RRzh|osld_?%W2T4rMcFh
zCsV_XQ*(6N-Ib9VnN7!dAB^R;%45gH(qCeT{)f0ag$YCS&f1K?O%1SafK8=)k^rr6
z2hoN09+bK0KDeg47e>C`z%~qTw=%n4>3q{kJ{d-WhIi>dYa}>9F6pUk*2q#1>LTD4
z*1VO(v+`#!YCk4OYCk!7?ys5oE%+q7Sn(=g5jB)|G(2f*(v||o1II%aIhjdR+-1Uq
zU^Jj1(D}$4rI_*6ieFp}y{uA7NSq3utD#318%MaVG%m(M4LDb0pvRT7G`&4jPh`l;
zE9bg;o=F@`$6{%?6sT4t7FhX4`bpTO@@}FgxG_K$OIUErupzrp6DB<F(n^W-eKbhJ
zQ{Hr%cE^Iw=|)}RM(lX0m=e&JhG$NXk(A7Yqh|YLG^6P?j4_9L$*L9^m;1e)ms>LQ
zTRX2!q{fwUNQ}dCR~?LyWcXNSohxJV5FpGdHcMcXaMYjqfGNP}+ZMTCN4=R%7}9J?
zPs}QW$ISepq|$(g0zV8+?HH=ZooA0=GTM^)1Wd=uL2TF@YO**V;=H2UgjR(RA>30n
zYWE-ovwryUG$eK-Ev1j#2~Ws?g+-<{z*wmz%mj<Q7fuEn6_GOFTGc76@}?EAQ9oLc
zC}&r}k79^6kHI9>TiS39&%vfhSU424Q0tn9$D!+R;ZGwXP0I#Jwc*u9I8nBoz)44h
zmByO!be?z7Q0-2?;T0oNRkBo$o}?DU4YAzUr|kB!Ud3GU?qpY12XO-IstrzmUI{5P
zn<p6q;`P*<dzS)uOd)ij)ys+{W9>0$uxO|8)SpD6!xmx(X{yIlm#%~@P4i2TM$;X@
zH}-|(h2X!2k7<rML}yOc6mqE7!28p|#!cDCVctn~4$=T{Ebiw@)kgTOjvQWf@)`7c
zibMj1pQ~5i7Inp@RmFKclvz60i2(H0>x2t7wsL^?h}$u_i+cM67JQKG)jNy=w`k7a
z<m&gy_0w9xV6SMdbd_xAx(pDH+j9Hm1}D>tW}EWZ6&9Ge>XP;jT}6jVtMfsbW4p(<
zt;k*!TOD0ekPu~RucFh8DvQvEI<|jQHJ8)=gq=U-lvIByue6QbINKqHl@9X?TzY;O
z@P?<3bc}*yEL9Z;u2<~_VG}5`yc8?<3b~pGCUk8h#i<jPe8Z*v11Aoi^{hUDNMs)(
z8?QeX-yoi^x}QT-wa7y~Kutey4g<h;g%;`XRDT@yfe7>lICc9>X7WS&&?;io0w=Dq
zD_<9E9)9}l1#;e=y=PviTQ}}C;?@k~(^V3~<@?L%;p&nX*y(M^li(^Djmen0piTDA
zxwm+vPoFAZB$+Niig%3@;ck30+M*rdfBum12-`;v{sSRRIZ$_OvE!sytY%zCSw`B1
znSHhkH8;#U3(f#Y&j8+cA_@}-62f`jQoAEJ`qF}w^3ybSK$Qv~#_2507UsVEt{EE&
zgJ;gsL4eyarXtEh(*f>ZP9FzZCS6a)nIJep9=K);Be)B(B_V*Y1uy-P41@#LFzG4z
z{Kz)AqbQ&SUnRTrE+u($bIHIzCh8{dBZIIt?G1v$0;gJh9kM9dtheE32d8dRzRfEC
zq8Di9s%GfngESB#C%rPlEPXd7=Y_dK7@!?vY*OL*4~%u`5CKN<s7@6>0bSsC1WWfq
zjWBoYV#M!21$ccTHUZ<NKb38Zl*M3q(xpQ+-k>dRhHrj$u=4;cQjH~1$&P&Usp8@{
z+LehpN|(2{JuryGnUq7rGs_<IB<BwqqIPID+4Q>^-<Qo~Z+-QVxEhR__^+;*s4jg3
zScAbH4^~7j^3_tPgXE#a9=F_bc{JVvDM6i0MV6N``x^JOZy0k>D#)Braw2OR(a2XI
zyWi4}?P?k2C5aqEeU7IHC8Bi^P<bgvrgIbMZAc^(!S|J5Z27Y~x4C+vbZi@FN1rpz
zR8|IWBP?}29+i#q<<^3LmPv=W#f4o$+CFe2vfv)#^g+J(D5*hVg&FB!q=FwIdS(r!
z^qI8c91$-@2GFoBqM_Nei%!#=V`ZrRbsP$>-bxssdXl$evv3?oi!LH>FaVbd1*#cf
z+h;!oao`sCKuQ%A{tx)#@EL!1qOE%1CzZ(dNAGvJ+_{(pL=h)3?;@pS4N<vunPB$V
zGTskRO7ZBw6$s?&oLrS2k-006$i*s>EwpyGW0s;%oCw!(lFf2$dtlTl-kmVx-KqRM
z1*Jx0Rkl4MpCr}L)q=Z+ZrF|Tn$#BPC6S4Rfxj4wo?!w}lVv7`#i5}InNez*-@|sH
z{Zl6TgmD5sHKbi2KeP=CoI0BH)UqscOm$th;EsY^9h=r5(gd$qb3fjd5U>0PjnI~z
zf9SpUUNqg1kc{y8)&;N{WG6GrIMk4bhDyWXYvC<<`s``0zTLd4JYMC-pka74;9lG!
zjrN_ZU{b8sbX$8&H|@HucE(B#SB(^2a&-*a*;{WFK|s0l#Evv5SFUPX#a`{?ab#r%
zCSlXJ-PhM+1>u_dThdp=p$UyRNW{`^S)*%VD2twmmD}K(Q8lFXGPTlA)gU4X5Al86
z*$~4E&p2`^u_^G9XY(E9>09cCN71E~pW4>vo~^LBbdUbY<n6hxcF-ruyLupybP3@Z
zB-&9W`3<2;t$s!8)xY8gkHwh?=S4TPNS7n>xYA*p-K+FCD{eF+`PSs<1GpjSil^dv
z+#4F8<9mUB+DhAeYCuD$q6TlPyHTi8qr{ndpAW^b#&@NK?UZ9ms@7BHG3Ih1G9Esn
zz0xqmju1P?{FLDcCnPOJmbo+M^nNkd5SzwKiVHTt8;Jcxe@Uft14lZMKQv~pZn3+p
z<1XLEtNM{i<PW-Y+Pzm_jK&p>4THuFxlxzv$&BLrFgJv@*_;{M!q1TZP&-=A2b^RF
zwb3+ao+s_B%T!swVcO(2h*4%?U0VjXrcP@LZu~~x0guse5R^eHII;3_GaMXm$+)~9
zAb%M*g$@+;G3rwrKwW}x9(_gb?ShOr9IAlfh&%(JWGzIEd*>ZXc~WNRB@ly`FpfW0
zCyaB1neC}n00qe^05q+ac3pg<2WiU7_8SW;0=R8RHN>VfiAvi5UdA}db~@q+ke?^p
zI5tBb`SglrwNf^_b8k?V`qGwUrjl-7$)n*fUG{I|8b@!ztu~b_P`{&iiiSj+i@MNH
z`{6KnEl8!Pj2X)!8?cTpou0HOKLRE}Y;A{Nbmo^I?*u<x)>gx@6dXey+c1iZ_S8<$
zlkn*dpo#4b{~SL#M<%c)I2TO>e{snpoDe4)<!_alF^wv!Vu^)H80D1#woxglexaaY
zauPW~7LW>F!i2Y!{mhAUd1RgBLWXX8<W6&Ha*cR^D*773GIh36$Nts;6mK*GdBbbQ
zXdO<2U7opGoZ(2femjo*u7ZFd&MjBYncyc{8xkMP_VV!x5E)xnHeOyGmhw@s%{Sab
z;+gJ4;I}X<-mTXG)O81WV5|%USA87E#Bx?Na&MRc(Ykk!hAR^F6V#y>3dP~b*KX;P
z7~nx)$QW7-M?VphZ`9<aVzv75gb$V#4&#wUk{DzMBmJ7IX4>}HCnZ>+$SMTox&wK7
zV6?m1G0uDHvj+njzVIwEkiKFM!Q01gYs`Sxa_(Z<Lb<eTYU?m%5*g*j$+{PWM0O}x
zr@ZQcv@{=OFG8iSTruJ3(0c8B_N<8Gz70ZnOVcv)?OeH!quzU=2_AvAWwi&`=xoSm
zt$bW<QJt6ICa=M#++a)Vetfe0&<5YZg6w|B=86vn`ZSY44AFu+ctlsEs~o)CLL0z=
zFFqurHeWn;!^>fAc<D=h_^?)gTv4hQ<sqZ;LU6&<*W*xse%t6uy-08b(<vbl9A&gF
zG8Sao4UyUeTTYBvYInZAbX@Hi4+c#OQp*!7T_ulV(Ity4OTm$e?wcdgJ`H#j*Q9g`
zJB?aKX1|U&A2UG|ylBe<1A{~^K2YvNQlODv_vw!7ajDoKc|-tlp=<w;Z~5)7xyzRG
z`~a&3kNP8=kim7$>JE!pkZt8kQQJ!L8)tqG#4n@BHtucbD71rYtDTngXt>oT8@D@h
zDWGl+sqcaXsS1YA7J^N;L640Wd>ED)g~BZO(F9xAV-bHk_*tSky^rI)^{Y<VhrKC-
zm?)vSk8mp26+0H3?88wh2#guzYdG=C09&qHDQ!}j0PIMP7&Ircg(!H<PC~J}CZxx|
zpyWI(70*fN#@j|Qh8i7et;`@~a>4*3X!toP0wHFCU-XS5(FMkfEc#hB;&jkbJRF1>
zW=1V^C&8LvmGcfPbyB^NAUg6<;%-aYMR~*@jzj9C^MD}iG4MN*;XG82j*uy`4**-E
zl!{D)mUMl2cJ#YCPHm#yk2o_iYizRzZ7gTYvIYvL(GJrU5`z<5iW+)UFL~jXLvZu{
z5K8_^-~Amxqhd>chTm7WuWZISlwNZ^1ohLDM>iAc{jzjv$Bkf%Ej!e5>Qp%A;cPn_
z;}BL;|M*xA9&+^9hC>pp;g4`awhU~4G3i){h(h^}y%UC4Pt@spZ-Q`x3b>;ojok6R
z_HF3gpM_A<x*zbK+ChM*mzbTLJG2hs;URBNl1>5>RCLieHaD8u)4rxBY^P`13ySjW
zn*osRd^~0wy*nWd#tZyJJMspnpc;0dd{or(+xVR>cI|573tShR5Qn~eI1;9~dCzTY
z_+=IaDfN@A3oYuFRi2rL<udDjND_TR$RWQxJ!Lt6u1Y9Y*%S>AneLzPEhDNa((Y%!
zj=<FQo<gs1{U%Mu7r1$Is&7xt!E3pZ$Xhb528hJPa&AA=6A!ZEYT11xUdcvygyU*m
zQTny={ba|J^QX~CMv!vWV{#xAO>b-1v*OwokI+XxF2!wh9FyGc7`JrPfg!6Kz`Jsw
ztnSlLoIDj{x%9~EP&ipBoBfm9Bfy)t-jJAl0&>bBV_lgoBsPQ|{4dGJL6W8uInG+k
z-zzzzQNhbk+6oPLOYI4H+jW_zeVpgQHO<~3sRI6#gTp=uTnW6$udhfotC2bzZWoPT
z$7+R?=&xQfsAgD50LeB}*Sc)oy8`6um5*;<@e~oMYJB(>k;pY<D|#RP)!;>M2%FeJ
z=@Do5E`B+tQ+fECABpa>2_(q{Zd-9e{t%Cckg`)w4MCmoH~D1+gmScbhfB_1OIrq&
zPueq1QetGbFcBK|Nzt)yKPt4&>}UTVoADB|Wc-gjJMnh>Qn|~v{JWAR^-)g9LWKk@
zPY$iBg@y%Yy5eNYKMkUt(fzZR!)QqNRqMZ-6x?p(J^}-OUsWr*1_W|6ouRAFiaUxv
ztjf7D<H~dy^m~Q)l@-#ysC2Z`i8GF5cg$#xFb7pX>u68&Th|>6l5~hplqrm1JjiGk
z0-!BlWuA^0Hc&J`Ivr_%Xqd1}9FnvB7>Mks&?yoK8o3(ROUk4rl2f8}B|{={EhVnH
zQ3;5ljkaQjAF-(p+NKap+-dYbbe~leif=Sv81Qopr!?cR5Nn8Y`^0fM48ON(B-bRH
zn|L5oLlAfH@D_;e?foZJ-B@Mlg_mC11N^RC1?PBC=Hv#Cq@^9l17rwf;+Jo9nw;!X
zp1!iGyrY7VNFSxa`qqlVke7n&;&hGnRDt!ou#o5?U?rh`bpn#HE^SCpRa?fdRno|L
zG18V*NX?QCU2+|l+g11jXcGy!4pKBE?o1W!WbOQ~JigodZG(iEb=>t`HY>>VzH)H}
zrw0Ep<7V`c@hsEvq>geza*|Ix=6u@A*|M96Ccw{}%!);K%&D>jpVryNUe=^+*1=4=
zC)v4dNuw<FWJuhyNpF-duzM1rEe&Mpan>R^5PPoLk<uVp?sQRMCASk(`fjQllUTza
z8?WetJm))|X2W+IG^nSBt=_v`;j$=ZP*;Cb*=`K)xG%CcoO;^IY#Au7P8dwLuC00O
zevmxn0zbUQ7zGT2?-zqu;FnkeQ_z&uJ6h<w{Bsq}pHGo^)IEKkis6g8PoD7_cUM24
zcr&id^DYOUV&Sn}4ZBz2%73NdeE&X<i>oN=yjMh|X%l&SM9BxfIO!KFJVx+J&8KXl
zld0T6s6JOus=(9L8|Ap>T@}Lk6pXQ(Ufr8=_4Zxbwd}Q^&hyv9RT2;Pv3NY=Ng&_i
zdM$r=wbRjiJ|fegA6ZC82e#dE$C<Q=yG6$0^^tY}>z)eJm!psUdU13HEn&0wg0y=*
zvcIG}4RDH8oPnH!v=g{E6)>wa*fCIl3gR|g<?R)btTvE$A;aK40mZKk$5j@1<dCAF
z`WpQrul;%DyFqxAJN3dj9yxq-YF3%hr<m_+*(GgqP6<hw@n4j~3I6DkG3<&?IXP9n
z(2or28F`6<Tm;CBko;(Om)Z7QSuF*X*y3)gKbTd#Uvo%Dj0T{KEQ;9&(MvTh{k1a$
z>7!55<cW7yb)CatS)&vI{I}009hc1=7ZVE6G(;RBDSnP}LK=CFql3Ai)7uL3JN^xo
zcndm4dktJ64&beb{KT1-F|j{Oq8VH8KtX5F(2<6TC1b$|(wq@S@c^}N1F-?yWSiLj
z7$OVj9xVGIfAZ3nqvKN;+IE99LMKY1wA<#1V4KGYaR)xwB7@yHUG{)jdgGcId`Y+l
z(UWZ%kRmg4o(Oj_RGjP;v*J4K=dx6W$Dkj1)9R`HOj;=yf7hlM5*(V$TcVWU2uzu9
z8V@h@5kR-F#VYD0Z#In%Cm=@OVt54>JYkDBbTIbWaH`?P=%pRQxaXDLF=!xw7vlyI
zKm1oVOiIN$%lp9Q77lcTx27Kl&Zpy%A@Wey%>v}8AkD^2EG-U$8-cA<&Ny0>mr*V!
z%XFk2I7^?EW+GeacR$@c=vH5r^J%gAgjrxaibP0Kg@rKyC{69cTelVY%=0Id<I#S}
zEhsEm8;ym+7%zYlER~{Rfgs&7bX@=r|IA=bawa!p&ahp62)64F#n3tx^$^Kp%4gpD
znzfT|hHMV|%I3QL6dU5ySgulT?hoVA&~lyA<`0LXoRDRb+0x5^Wk7HEdSY}q$5YBi
zu`cL%AXPNriI&xd0hg1oaXv$wQ@7g*|D5!StE|eC6P)x$3CmLGrd*T`AyXG+lg7p=
zZtowWO}Qhbd1j(hvLRD1d`RCbFRzO41c6-&mzH7h8o4Sv@>h9UrWv)%lqW)t$}Uea
zKfNS8s*j37JA{_h{rI+@;BStRPuY}S<!zq@n~?kbl-V9vLcGv+1>|wQT<X3JPB*SJ
zwY!qGze0($`NAv8t$&|Hu^+wAa8><6e6Vt54k1@VG%n)KdTh~Bj_JSU$(ctyLE$Zl
zHbz<TfR-x?h)H?z*6159uYIfr<DReVGJeMULXezV&Pbx?FJ9zpzG(oZ^2^FX%frfr
zq4%%2Vqm+oY9s;R%xlP9A;=p$G4L0>lf=CY?^qde^~Sn)@7{6LlQ`l%d;RR}IV&M)
zH+dgNHk}XrxkBUBJ$d5}wov)xnavi^HEoLwwL$?Dc)hXJ`mD?V>xzQ;_sF6iv%+9G
z{nBvFW2K8CpWj+NP|Kz76i!KHX=lMY9R){9J98OW%B?PyoY)AF$5G?c)tBfMnXg%)
zxnd7c+Nn->J8s^S0*-P&;d?S2l&@Fy!e6?shPlQj6M{G%fJP4)gCU;k)sJuyuuF@C
zElSBCIcgP109!T)5*6Qp*mNEktGzL_ABeFA+g#+YD?gB{KAGf(Sic=u)U0U4HqyrE
zX(m{Dpk`Ni2S!a%A61pUqPL;C9pZ7R&}_0ng`hMeigOJRDAj(iKt<|CR+~bo9<DJm
zYVJh4v-3{X{ROgU_AAI)kX6N6#&vEW&A4-VL+5j3Tbz!Sp-%OkG!Ax4my>5#b$_Zx
z>VTO!EWIA8duccLh&$`pkgV~J$}9$z{x?D=mO(Yc{`6_ZP)<XORJr4S_u_Z^!e9-t
z&MCm1lB0wuY(@tsqECHzyW-};$4uQw=?20YDG_C`n>p^fpeA|GXPuf`HX7^HX7UF}
z2PA<gSLA_l^zAh1<n2|kpqjm|5idrPdX0ew=G?F80+0#FdVxAv9)3DK1t;AL4BO><
zjMH!c;((CWV6LvNwN@=JG(%h7z^0sZEShrqs43FcS6+Ff6=x)`MTbV+)=);#%J{d4
zJ^~wh6iCAmd5CRm8M4BSzM$tM8mB;D(%t42e8Ui{17*m^;w;AjsEIfdEsl_nRy&UB
zEu|7CsKX@e6FF6GRE_K?Qy<U|6k}*-peV5_5#uN{#c4c}B8^~u+DM}MFJfhE=EI12
ztXEyv%GCKIoGEz^{!M2cPyA#+fO`JcaO<3qlW6P8e~{d^{iQuiNvZQYVJZ{25CXN+
z_*8fY|FnX1Kd9jz@k$PPe(G?t6B0r8Dow#B#hD}P1nM$xheF-KHBL_y4cBp>K%GA-
zmf4QgQK&oMK|oT_i<~JMol*_5p*xfZq?(pyAVLl{K@BN%wxPY%QFH@r$|v?cVS76u
z#mX*W<qw1UNm{un?;e`#C(x6UTX~zvo$}GnJLQe=fVi9AG`by~Ur2O64W5nH^^Ir#
z4(<TjmUlh|8I4~&;_anR*njbi6$PA+r>ulrLimbrz0~_0Fhx2l8muWOi|=j#iK}r8
zuQGz~^B7k*T>0l-JjQPrWmgQ`i=ffJ=KUZ(l6sDn$_vZ9>d$Hl(+9UUG@i=opDP5e
z9{5cgZ*9mjE;PIhIwu{ReB6T&1I$W=yuJE&<>C#i74^;&Y1ImZdqsT0rJaC<jxr-(
zuZlj?2~i%(v4lKs@+4#R`StU_Dalo?;%#HS7bN>HH2&bcXJ3oAJ~BC`U1w)cX_vg~
ztH0-G&z|$n6zj&yhz8x2mG``_LmF9~QK#spLv%Fsgd!>+jJJ*ayL#`tMRY){@(22~
z<r%s(_8ly7#MCHsXv}C2yjf9FccO*bsMGZdsX@{{O42AvexSvomfSLy)TNGTRr5=~
z>-hGbARRT^=UZ<t`9_iSZCm#8s3-CkK4siUt5!?OQMOAu$gky*yI1pZZkRD@{!xNR
z(I`utdUNVw3dvHGr@y9hv|MV4Y^#dXGi%nT(>Vom#TzY}pW0@!=6}nl3VF?Ws)lB)
z+3ARK$I&MZkn8y@V@%@Rj)_$+G^i>^e2)a9&wD1>!b$sRCuG)N%e0ejj}WV(9Y%^b
zS<ChmttkBRGDzG)r*q6qy>x=WD{>3*D%D8H@XnxXq>eK?7OWz4v1dRTp(=}{4s-`g
z5BNEKg8cWQj6vBj$`u)iBV*Th+<?c#-Hu~$gq7z9_(sM9bi1xjUiFM|z#!E@UZbqR
zfQM!yo4hjpRIl#H(z`0fJ{(u{w|qm!GE5i;#=4c3LJb=RS|g6-sJDnBjY}wM)yrn!
z3N)M}gL^1+64GWtaYO{_4W7wML7h;J#F0D32ieMi=lRjlsSoA}S3x{sgFk3wLZeJ)
zM!q@&ftxMfgfA_fCFS%L^w*^OwdYwOq20^J<DdcQnh<dia=Qt<^ULakbDXD4n#u)i
zoDUF8=e7m)eO_g+1jby0Uhgm&3RjneIBu>?$#4W~^1(vr`;j+<rUaeI1KKDtX?f%p
z3+1`(91^9sg_^|7K&wVa@ug&WJa~ca1Rz5Dw+-?{ilQlmdSQIvj>sbN$WMn@r*0jj
zH`BD9V?>FUOhXop0<mo~{*JokbDdKDrb3jvzwHjzZOxt5-ziY-7K*v2@>40V1sn0W
zE_o2gMpZ_~?uVp2+6hS;RqPZt*`+3v&&X|I=7A^30=p)?(E4CYc91`7sy*1*Lsfuo
z&!G-T{h_wndDNu?7Xl<&iPOZl;<YK)0Eo_Ykzko%<QuleBb`Zd;D#OPgQA7f!+u%^
z4SkSiw`%|%glj_g#EZJ_fw`~hgS<Dg&a|A^4IOh)jE->`9zVs9o}E48doI#{&+QAJ
z7`Q%X=|L2Y!Ak#Qey1^yHD18@)mtYmtTy?&jSzzlx`zBMdnF`(ee+cwqR0w|1{B)Y
zIt<{(@beT356WQBF+We(3&P5UD<+N#_l4MJy)Oa~3#tqnd@HYlw7eC<6}Aq*qeqvy
zvi9ow`uc)$_H=M)*e#0_HuTm;+d9eDdqZg-&O}Bjw3R`pLx(|ppZp0OWdTPqq*HrD
z-epBYhvL0oeub}B{c$RgSz1@DmV~CVR14~tIt{MxOL@;~QmszdHdIXG>sA`8Md}D#
z<!Rl#eEBk~Elp1PhW_fWk}9LCR605Da!Uz)A)5|}doprc3;zBQ(Rr?ZRohzC=Fx=F
z671XbrGkPe9ogfimA^SPQjqA}Rh}$$wSLvDeam0~-aR4eB6=U=cwtCF`V-l?D)JYP
zQ4BEJ!x?t}PlJ*eI_eTvK%%z|VR30-=r*j(H|R^%LlQeV9s>PoVQSQ_Y3`l?x?8)b
z43v3wZbt!S44`UPWgJIczJvY96eN3WhyFs&4k0rXJFkc+P50G9^nS5Yfke~W#)cGB
za{N|vcWsRvbIWt3pyVJlaBx?d70vrI5R#V16y|KD`86?D2*0mvqSHzBpoBr+(h#W2
zI;Kc?rrNri5oQ3jPNnFZCPUd?@xRqD(UDL(*h=*00QCxK-GMWL=_VA$dw**1APv0h
zWG~5jQ0+w;h2hpC$ucIUHk2z#%t(I-+jUCk(Rq@Uo*PBFZZ$+hf`pDs^-V8@o1dBH
zs!DlhndkHXT;X=TG7a2S{Y3mw`2?~gGa-|+RraHM8+0NbpYv!b4n!P-(Z9lI(CuvT
zz{nvn`ziVvv=)s{U9a)g2?;Lvrt7M{-nVhVhso^qtauE{WoR_A$|zL`f?J~_%=oFg
z#OemS;MKo(RBvwT^Y4)_xXMZ$1QkuAt0SOHaiYM_%7Qdo&uJ?K8eWi;TN$D|+f7v-
zQKih4nJ$k*L!D;X6_VfyI}E6|jO4i(X<_A0TtA&qC(0u(j)n|#SBwPzdGdx;fqdkT
zTPy7Zc{YpWOX1X`bUosnm>#mD1jv-!%5Hi3AdE_0%d-kI9EoxiDN_LcBXx);V^M-Q
zO|$QER}1U6+6$s&fXx`;R!mlqcud#ZTQzQ!ka40ML&tFjhchHIiv6ud68q-{=Jb&B
z6l~;IJY^UJF{oqAQ!fnv<6JLt{H9QPdxJ<~g^&>DPYj`TA}(mTs2J^9QQ21pz^X-L
z@^Hcmu`+>M0Ry|rw{}*A8%l@kYBcP?o7%4Ao)lykN6GZMRd*4Wes>-rfKy$_NHHA>
zB&uqmCM@UJx^q2QGEoG4U6&$%CQXHG(H_<N#a+`~D19KIECu3KNrx1zL~EHTT$LAw
zC#}H~Wfi7;nlP%T>39bYz=BO8DiF!EP{eg&3YC(Sd37>}g{5F42))p2T;tGesN>>H
zdR)N`GSFh8fJFLs5n|AU`6NzYg-@R;Si=*KmId(+j-|qgGmaVlTABeV%+$$y^pEH>
zym>=pu4M2R!`yOBbr~g_@=;yUsB4;c#>5KvxJ=p&$_<me!S^tDB0ux}9i(qwzKC&q
z%^nJ#ui$I3It}U`LxBz)r}<V4SKZ!vi2kW9twB$sF7tE9OPout^4$+&+LZ~P+E54Y
zD3`W|S_j}Mt1;quNVnT<KDP>xQ^TuS?UjodO=0`Wcs%Ly?8WFMa9us{>OKYz2f(yX
zVqD(vz7J&T;MaP+f`39RJ&m~Gk<JRSa@hNc<cTApjJG%d8dBSlwg4r*hbts01I`eB
z>Lj;l(1rU&Ab8Ebm3dr6ui&`CA>WIO>$9ggR_Z^xhFso90`KOl{L0gh0jui>iyXzh
z#;$#zzJQjo`jGhr4rosW?nQa`o~LVQW8P>AT;`}Wo`%?-v`NQ|(IErtqE7qEy(sh-
zFjn0#RvaCoWHT>}u2`M&Dq3B*V)Xv%ecl1``t|FxSFc{>DWh6F^1UXt0t1maC+egQ
z*s2rg>O_0?>#*EDsy;ird5R3K60!L@PuC!;^^809eetw(X2yJ+ZJ8)EVaZ#$CuEEc
z*ge%p6@}JO7(3uz-J)r`V4L*F)38->bk`;7(YMu}9Kv&}1Pd*zbGd2v*zLR^D@-qZ
zmP++sxsInQ;Oxg5yAFV&v&I+R{x<Z5lnhcb#$<9JI~SSGxEesA>8jvb%oe%h(9e%0
zv|-(-UG)c34&=`jT7~=gr(Lva5sySYHnk6$vIVe|x+^f>l^ypWNw*vzXz_jZP3s=?
zeN`xYHpo(mN52O)<xDnxH5ZWXu#N}bL!@Z~zlT+k_IE?~5Nwx+?*W_-@oO4CTRh~9
zF^6eZbr@9nU>qm*(<k%{PBpfuPRg9fjunBGa(A-V$=yx^#!7<^uEyVq)t6`Kh`f8t
z?HGp9vwXOYa*UyR$x^*mjQ=tXBq>+6I66!MS;c_s;(!b>=JBFfJxU!{NkhlYpO2m+
zuZD+q><p_82@Z+|mhxMj4Upcmxm4^M<t6Bp7_#Y=`Zz=>K~5(q90eVL`9)=}O62y4
z4uEYDo;Z6_D8oZJyMUQ_dEao`DY79;LJ&$1wnJK(9D&ujk!JJg*fK-fJz{aPl0#mD
zZr+W$G4Aw<I?1ESu5@*PjP)s3H`5`9a7mxoA0<okbh#$&;FcN*WT|mtSyClpbfG>B
zjvsX^zc`HJwve~XYQ>D(MU#BOr(L$qV+ryj$afTLS4_%%NG`-Cv`_<SQ~R6G()`xI
z@oiPtV(-)4zOuU<R#2;qR4M4Pj%O0g8x}``(TK@X7XA$1=>-<@Lwx5JX*E42#Q_~B
zq%j{759$vA{RP2{^YO4mzs_bIZqKG>nW#s3PpD=y;ShIO=LhH~jhe@n=(fCN*p_ec
zerfrwynUtWZ3Rz3uZC|PbGyK+YK%`#XlC?WB$#8M3R%OkPEvJ|BPXrShkGg9=Ydh-
z{T*3#z~Oj<QP<%5-i&PE#OTDZOB{mzv_y19KD>M?W?Es?G17R|73jlG3Y`O2Bs#Rb
z8dr^8RwIz*&6~Gn2$3`8$ZY8MeAIh5T;1_1A47=T%Hc-Or}7wlvidQtQ0SoObSSM?
ztcl4RvUfuLB8zfq{0*l084C-f!81fB;I<>uvE}IvaR#pM=!gR`=w15W5M*{=imM-S
zMtB;@CwdIwnUp{FHGKkn8U2zkJ+A{Y{#=<#T42JW+goVj*NkFoMm=&?GI@W@XWt9*
zW+QK$9{B2nL@(-k=Oi_cG*kRRDH4!bx5HWsC%^&!B`OV6&(-hGt`qxZ3UNB7eD~XO
z+hwW~b~nDJ3X?WxLwf^pCtI`_&BOVqaccf(C*%(8UuVI-F)=H4)H?wf?sVR-sMWT>
zjy4!~It8OF6OU;N868KCFR|_o%`^0eEFwf6;&}x|-dXY$fUw1>(t$hC>hjMScRST*
zX}lT`7X$<ERs9VcHy>Yc8-a;mBk60)H3qGVd*w$tn$Gml=p=cG?`nv*3fhQ=jxgc~
zQwe_wIvSLCWz7mrsZoimcK9u!8)yMIvZ51OmR4}7KaU3w<E`Ot<E_J>W8f9HWjX;m
z4KdP}-jc%pvz%{xTy+u7UA5<2f%&FXwJ7X+d1^#4=<irQa{1p|DX#j2hlX3e@X~Q{
z^&)xb+bYY|fpyUZ>9iD@(L1=tEw^dm3}vWG$Jt*YE{D#at30MHe@EW*6)Pb+Dsgy-
z<IpLEf*2AP4cLHVf&BqL3zTET&OecPAom0=OymW;+^W%YC34vGk21Nsr0y<pkZLug
z9Mo&}iPV!tj6dL{t2>f5MMIRWXsWkeL2}|*x+|U_T8c06XVJjaC-PtVuDZ-&SA9k2
ztw2X+DRWq643gM{DwocvV;pDrZs3-zU?FUxaB3}df%r#B425l)+X`KZ8*J%~rRG<W
z$OO9#j+(d>-0}`iIGJ=U80oJ2n0V$PojJdxha5X;isMp$KmhAM1akoOXDN<+avEo&
z<4OaUr!21dQY?np&5#;LtXy4*4nufIKqvWm7{*XId2nc;bp%{t(fGxf4?<2a8pRpj
zA>roO$H<V!h<iN9z^n7IPb_qXuI(e8#mJyABWTn#ayl2bZ+qLsyumn5$WvDee7{KD
zBC$-_&b@Vm4$>ahgKlt+V)ShfmhqAota#f<xjTOG;`!MN-q`BC1e@@L6%utSFC7tM
z`y#h#kfGmz8YkB_aHJmC^pE{9g3d=Dwy|%YL|*FNd)q}A+YcRKlwG(E)$hvdP;iyW
ze+<2NtoD@dfgX-X_BsKa@iStPqkF*eIIk4glVKsFietBmQ6ps6H_9O&9n&~-@4Qvy
z(>isVEb>SH&_MUmhk6kY8Q1ix?`>m;<rW-e{(T9*EDhZ06T66GS@|$c(O8_=2|x95
z6*>Yl;C3)0<=8e<r1NerKPL8sqO~~GKpf?S$Z+u7L;ROqbU^A+>}kG&SY<x4U7wQH
ziJ~~EWism#Gto&M`4)4YWG1kWO7^6>C6e-Oq4_%}K>@X_Gs?E?r;vZ<cVD_?QHGq{
znaD8{c@rs<I)%b`C0|cp;FyrV;>eY##79TenB4W7`554F!zNFgjQ*6SPJoAo*Q4^T
z3}Dc`#epnd^~Sg;2ZwLTlwbCQ3|fkyafz@PeET9gr9L+~$mf=XfQ0M;8HP7-pc-qe
zv!6oa2Cft}^u9>UZ^%69eHs`=>)N<`TgKZ5xnlR0)ACX<R~<?=;g!!@9=Gm~h+wv-
zoQZZq`Xh7dLLF`UM<=Jt(`TctEx$5F4&suJwu7T#Hx@3sP_UuD>~SIOEiqRy&T-7D
z?+|&wD0lYk_!@mzr1Ho$u%WYqv%$7ieHd<79ZErR{EV^E*=j@QWM}`C&b$?q4ZF(k
z!3vY(!ME<Jhkia`=hDct_)1aHat!ht)}!=5)WAD(2LOMg#ohj&yLlUZ*Jwfr%U5-)
zO!`?-`+PAy2gmd5$jo(z)1`!*0w#wHMn^dzcSikTj=!v$J%gb1z_0(^2Lnzef11#g
zJ{7%(OC`q|L~Gvr29}f85*Q_GjHTwm33<$X=5&Y*9{2G{T}JGN?I)0LL-CVjXL(KM
z9CLyv?f8m;kQ4H2EPeFTtM+mxtPrRE#BUvH`t6b2vch=#eh-}yAVf(Jww;iR3muSh
zLgXWT*X%Vs8d9%>!-?W2d>cj;XV1zYmys~W11J+03~u-WKa8TxbCnI<7-Wsxxbltz
zqEi$Z#w{A;KOg%?GA7^?BS@?hra^CK#1#e8K1q^W3B1drUQR~2e2_sy?(Lw~t!>rm
za{q|$;mGO`{Jm{sU7QhHKaHX2nKEzRENy^(RzF+;;gF8bPXDD-Z9G97ewO~ysr03E
zIwP*YT<~rfm0Y)F3P+yy$0Hxt=GN1V8;8kH$}ZY6;$I!7jp0RAoGv?Tbehl+gP&PA
zbKtn|L`T#WsBCPG^I=~>R|l2)T4j;Y2$Jd$vV*oVApfyJ478NrV{w1o@W<<F-QKw)
z)-@l>IbN4w5>XQIGuPRrbLhI9qjQDnsX0An8h#79$`8Dibx|IeAL)dw$o~2=lL8f9
zh0H~*^p3lMnP@t5WUL7}HQ2eE@o_4tUl-nHlI#B1YO>3h#)-E*qn65_KcTVA>7Et(
zGP3q(3^Yd453OlndZ}N-lnLH;^w9&rcEy-rs4>9T+~3!5U$PPPIW&EF)(Z}={(WLa
zgIkaJQmN5`o_7Q^dbvWE1&?yN3UU#n&8>)yf;U-Uhg^{B@Z^@)tVgyM)E3vuhEBwl
zj>kB}*YT3ShT8B57=w;b<RUh>)inn6rN$E7p&M@x8GGe!nU9<P_~VaxLdbGAfnIy7
zfp=l(bm%<D`}xb~wF<-POKwBu@xibuz1Xf1bZ+8EW!0#SKbopH&&9E@9qLSZeGR^2
zuIGqSHT1fD)<L^(ePrc`Rfv>YLA_xf%fr`8`RYel>9K-R6GP$Jw-#hoC<vqn-332*
zwepc~4$X2~H<MbBMk3wLkGG8+L`#rw&7!mXcX=-eUlPAY4cN_u*;;lwcM`1}35Q~a
zqh*8|9ZNNN##$<hsGpbk%{w7Oe>8psHKZ1Zc4b{T$JUNBb1ptMy)Zczaln595l-a%
zS<nt@eID5dBx_ZF4wa1|70HVKZesUJv}XCb7R(hp>)&1}_pyA84hpuT+akfGJtQ!}
z_!JuDa%$qZ0OReAC#-<z;AnV@NKwi}fc{)1a3iXQB$J41@U9Syjh^y46r{AQAnfUj
z7#=KH<k3+utz{c!i(E1QKkwU+CiS+jZ%4TzVZ8>vLtum(v^uY0*YUXJog1$2XuQ>r
zy7UzeuAH3nx_e*FrSp?bu(TBhO}bL>iWG`FoRD@#0;oI?kHdy@=dCoIJY}>U%G2+<
z$-a)k!FK5|p_Fo*T!9#llv_E{9&kxi`=PZ9bGjs!c63$Jj&#*QF7(s@sr-kQWH=(r
z_?2IhofptD!%BuKs`BEqtQrRt!oqXNW)6eeQdIeE=~DeA@l$JHbWkF@Y;=W@W!s(^
z;B@Usb-^rWmw{|@?T}EC!eWp)Xz*#Yj5#t#IU!AGmgC{+nrt2p|7jqj<k9M+qp-OE
zT@6KdT%3nyKaCilhsu~(Y_}n?6J8H2%g9@9oIx>goLEi`&9Ct@Nq2_0DR1E#WR}me
zhF{fxgnUisQ^zql6VVoNQ7J9akexr~b^(RYo<8U59<#wx|0OS%QXe`BZJ_NBugd$=
z-!VdzMV}b!JOQF3gF$@$?0JqS-0sLrxHP6%VvQa|vKSIWj4^i4#53St@L@NNX>i7s
zyW(&ACHki^_KWFu5X|dzcqo(RPh*@saXS{$i$g-0!FJs6G>>B}yueML+cp6xcYj36
zXkV|S80VG>_12kw&c@XDZ`tgM!}0dbo2)9;NA0}bF=Te*su~!1oe1Uiap$n}2gc(}
zNZQ+87rfHEp2R5~$zR<Wt5<1spZf*hkQutL+{bt?;h{e2(OoyVdaI{yiv=GV2`Kj5
z7lIihZxUt*ZtWE+D`t_~O|S)Gy<kG3c>v){Y+LLtU&-%^nEQXcP3QPZFNzREoicP)
ztg4QZr66hP%LLqMocN;49Pt}RUn=QnlBk9}hP*ATs^9*vYEyS7hmA6ipp)~b;Rl(D
z7#)*Hj+8k#A`+Hj&mJ;Mo>~ayd>Li(GRagExYU^(18a$8EG}M9%kpq1qy^|021keD
zdNkU{AoR)QNp(l`I|XvGuufoett!*Mrl6`lwoTl2!8UIjyKIYV-Gf3;tTXV^5moqL
z9vi$Sd9rLx;3tSRD!agVLr6;^>k!z52n~U6xy-Ez-_^myevUU{xdu+-xll3zU1>Rj
z6Z4b(2Egio#F5CmGrZatM#dmw{NfNuE#*a{_iKoo13L0e0B_w6aFoe*2m_tj2{9j8
z!OO<VLC3OoPJ&CmjrMA_V8DA@Mx5rISND^*9ZpiqBqo1wJah=8tvuJf@l;%&vXPEX
zLY$z$i=*~JD7bMRfH#D%zm-i{R;x7L>Q~*uOP#ylL_KudY`<Z4xXeq+3AtB3Y6CC*
zfn4=;3JwtbO9W)I&vSWdY@4F$5~+Dq3)0S1q+I?SuC@{6JDQc^$fxsw%+-%IhQa_w
zVyD>)52<umYAG1{Yds1g-B74;5lSJ}_+R4DJ0~eZRn9$d5|Kt4gzV;`YXKn{evIeP
zV+*hWIoEE|J&q<KjTHJ5xaA&{i{)5<xDz5Sf{n2{QGHmlPK-&&hoOBu)M#vPqG<&f
z&lB5@Q4mJ<6Uck2!jvHFPnV%5bo<<y$)^(Rz@9L7EZ~8`DzI|9S~n>?;iq{A6`tlJ
zhx!<^WXqvZN?r32hOjC#Kgv@%ezs+&Yip)@S~{x^pt>nh*>8DOdODFW;g(!wcpme9
z1~0;|uP~fAA9XcN|6oX5DRJE*_+Z;)4d^*9I`eAwo3Fm+-5Af#zW@IF?EUyX@weZ8
zd-n3{uVR!nn(E{|##p1N7Bz4Z>b)qwTO*q{y?w*e8}?1E?14@g{nJ^g1fMY??b~rR
ztc>QRD~<HAFqCKIgF0`SSbpXB8ZeT#Oy}XnS1&_D_#6YS_S2NZbCvd&v(~qDbVA~Y
z(O!e_%Bgek?%7j}y*jQJg?SrB9q3Gm;|tKjpS1mAy2p(&*`_##@}O-V7o;uE+KDS1
z+~R4S+1|8G{?vIp$j!9M;>K2G$>!FqYM}?c6Hm7=jRfSOGUSn#Vbm=0xzgdPLUcpr
z#n{{8*@MJEtH)+b=hBF?Mjt<=?LNZ%nAfY@=NFf?a^sxPj<N7GvLg4LC;_rq(vNcY
z<In;{mpaqZQ1-Mh!2F0SR7fN_NLS%M(r>BsXK%6Nbzg`rw0ox22%Ui?faf2*p5N-c
zQA@F<;!4MJ=a{gS$usXLC*&ArzZlq7ZpR8Xrx!sj^n{x$t84-JH6<Jz45kbbClUvw
zWuts)iun=d$@mUj1Dz5#kCs6iLpHh3TJH3Au%5#9XT_n7-InlWQqFA^+F9S-*g1LF
zenJ_yb5=t7kzFV9KKaut&^*txlQqe^cf(sJPcgDD_zul8Zoz!@)mPb&`jkB(J`qxv
z(2&Q_-7rz;*qr+)C`Q;n4ae@#bO{Q#A-|HXf_>(yg6+T<+dq$W6fEa3Z-~5=tlv>@
zdC2>835i}oP&N&|bqkTe0;lX4Zp%F*Z@EpOjQt@yWr?o5^`s0#*S9zq8few4GjPtk
zMPyJ<|G4)>X9Qgi*&gYP06)tfx`V#yc6z#<I_tJWJ~|=FGUHor8+{T*+PlrP-o6>S
zK)$pYqHPdXN>C5Nj1U=Z2l<R5=q~UIW-|vy&S4pi(7H-nT-l+=)Bbckx>J7=@LyWW
zNTlEDhw3UA*Vqhv0EHKS;04f(2@O;wm8)e7$d{j=Sp(^i^hx(f`p9DYh<I!DTENDv
zKJC(vi5H3xA?+aH|F`di?9rJg&o*Q#S&Grjr_=nF)z6csjkcnLlatM}?WX52Ph4HI
zPe(J(ed#UzhF!Fpk>8xUb{9K8E?u!JPAcVl`fmgFvuACMdjJf#7)OoWlc%&KZA~#h
z7!P#s`x$Iu9;JQEiUr5Txi9PT>ePPY(9nSU=#i@_=QtYAvhwl$AO4UhVgBoX{l63c
z%fI|f9zo4FTo$FdR|5kHKfYj++cg);5qi+nA>mL8<-X!E*f<9<ZWN4=F(6Lh%XzU(
zgO(KoVx5DGA8?h~RRC$l_|neMg-eeKsM$oH%gJ9h?fRJN7?aY@mtnPn(pDCD8f*0H
zwe=;SkLSj*(78~~LBrcquGS<Lo$}NxH;3hhmMNL>6S&bv(`g&{z7bUi!<7x&Y26&s
zPT*Vtaqq@^Hp?onFEvwmG_ind&Xrr*(VsO@i$A3cWsZE(gr8|Q*a|QGY?56l20i+c
zS<9@9uA=#--lurJ&%Kr8cZ+al{i5?NpAf*cqdj7&??Fd##FbVA8*L@Zmy`$JtP+it
z5ckkzfrlSN{1Sc%8gMw-sUz5n>CY}O?&^-au7EQjwqw0PR&k#X2RgrL2qr7ZdoRYn
z<D8JKperdJkO}7pz%3w)xJ${=2}nfjmR%W~5BA7XMF1wz2uE}L#0dN3*jM%uY<u7k
zhjK;=sUBD#LHcOa&%yI(ArBxZJBJ3^)rP06RA}@x7&Qvr4}cSbp>ooG`_4^{Y|zYr
zbIKo6MpR!8i4K5!MRZtP?Rfse$8dQZmq&5^?uxf&eAL%flHRy#C#EOuPmMwhgM}PQ
zLaZ^<QFuh1Pnejdq(B6tDe^UT`94p7q%9a5jlLkXQ_#=@Uqjq4MAJ~44*}H@$Rrdl
zIf?Y+v}5D}U!f)~0BnN?#fg!(%5-E>npk7YuE`8pd67d-9TO2b(kJABBbm_hORq!M
zby@MGE$T}-Y^SSB(E*&q<zELyJ@=+lS8@WskaoINM*fk(4WrWGC*lnuTZ|=M&@Gwd
zR?e`Ny>dw<!WW<P9o?6&GUF)7lMe0mjG`rR5AY=2h9ZRRd*!n{KaPc34Ux92;*2ii
zn}VT>fe_3gP045deIfGS2k-&>p9r<}z9UatODR=^YNE&F)@D9nJH#VXjotB)=tG`^
zSWEvG#8$g^xDNO%kddC@x9Q+gxR}~b+#!18N{HV49vTPRI9E6rxenPC9U!zFr`!CF
zz{65ej*lQ1JlQBnnLlIDhlgqJ%jW-pa3ONsPff(f;WTuXrS~czEoiX)F?c<x*c#Jb
zoo_=XQ3@R+Q3;c*dg=TATlmkJugO+HAJG1dpxw@nH`PnFb+!~7k?5uSm*ZU>|Lwp1
z_uMY|-S2)EgX~A<{Km_xSARcy&66De_HX~j)$RXx_Se6>!4cqNbnFpHpu_5$D4sE#
zHyFI{fB(nq{rG?W$N$Lfm+$}Z$Ls-lKAv!SibJAa9B(>0ekVjDc;!<ls7pGx7d0j{
zkgg)U{_!<-=KAdQTOM(x-~9+%ZYOAD83WIrzsT)~7h`27=ZmVR-qQualrG0xR#W^u
zy?Y<z-75~>F7nY~p9In|W5+~{uTQXOIA6c6r(~Ys=(yh?@88H%7dl};z~_cNG&)H>
z0i%)6i??X!v*#DA4(W=33*cj(I#TX9BuKSfVNqtk=C&6y7&{JJsQ?*ypp71K3n(2x
zVH&6ir1Q@iVlm&_sW+SAM8M-6x>M(G-@Zv)dp{z2cs}9QnWVCp0^V6cQXbOcDTjQO
zrC&<6`|P>w$9W6y3-)s;#5I$cD@f{s;*yWKtDn5Ob^nm%$ai&RJI~Hu{rHaT-zifo
z7nCy=;cLAz=-@&j{T7+%!@5$f)uX3;?<md(ly$D3=>+F^mU8LH&y|(zpOF@Nw(;k2
z99fl%z9X@q%*FnIHbNXm3R?ADZtPe|L+8Q=r{i8f7<von2v<<Yw4vl9Kgv^xd}$^b
zhm0Ysd}K=YDm(7i^EH%?Lu+W{h}P8eq(A%D5V<D(?~J3h)#kv>@TLphDo>t7%^?|n
z{6J!Y&M#q%RNf2JHfSfqa-IhAz8`V6obH3)*6rG~t@%KHZt({|4^+3_Z2lo?aZdP&
z%DSxw=smo+2_544Axh$%h&!LjbNUd=DCzXm<q+9_q;e;z?6+Xq`vSx5n>wF8dy&UT
z)vV6Md+OXUsiWY#J#<{Ei~9uWG1K{qHCn<xfo4`epfkT0L<8apsu}R;P(4cC>Jie_
z%n^A5sm9OElNfppXv-F7jByjkO1ki0CPe2KGUw+SaaVB8S=D)mWAm0rQR|iD-fnPJ
zL}yA`PcZbJW0>6-y%;(KdF0hLvRz+scCvxDoOU-m%A+MV-nC5j)rO)a)smxHo90Iy
zML@Z8gi58QSc)IIGJ<H#mBfQu_<_u1OYLwRZux$TdsX5*phL?&`c|8ckc$b93ES_A
zM8%XV3_1XDK#sq&jgA8qlNB0_g{(_1<Caw%Gm0oJd6}`W9GSusqeJo0RO&pG^JCp0
z)C4UGPZI={yC+0uaO^iISG&?S8WCEJ1brKxP`uJB^60}l(^fa{Ee&B-p#fKgpCK~|
zy|fcC><b4t4HBY5q&`{nrqpp$Jtmg<flif-)lJ^hXv)s%rJovSpqe!ZI|uU~ZgHxK
z-T(kV07*naRC+d%TdV>s=T?==`{7QA1ipkLHPDkv_wsS*_Fr}7ad?NI+XHq&>9ldD
z#mlXO)r4?}+HvJI@vd>Am7jpun#<3Vvwiib?31(Kv-07sk|!^?O5`5ASMk<a?w$g#
zApidNzn}g2|NNhNqD6<olanKcOv8+QH-E_+KtUP~0(5-SDp#dCIC<ZPD<CNHlCRPl
za?XRNJpScL2v#EU)J!ZpQ&OCuYn+C9!sC)xqPqg}memPYDBg2*@0idqYQUa8tM^Oj
zkmObZZHYq!-M74><1O&D0#fvJYN87q5Zmij_SM&v`KT*68bqI1(S3Oj?@UrpSh?}~
zomtBljGaGX{{nEi?J?x=y&aZ4<|BaDI6&uq1sFM1S6;#vrMbeOVJsb}lfHt*bcffY
zu?M9c3LS@AUttc;Ep?C_^LF)zD=)t9!#9%pM%Jt_UA-X<IhZR$oC5Uylv`t-L(!G8
zsx#uOQsk;o`r?30!+q+cx}{cU{FaPCXCra|)ACElvXUDZ@)|$$x%Fd=P&zg`jXFZ!
z^6`(BK^1540vQ>T?_3SKrXE@IZKKpX>~yYdXX#dX>M00VQRaAjOrJfuWKW5EYZ$v}
zfNcy1WtW$ZWj&pgTUDbh6-WHo<_fKkBaj;u9c9&}^QXiu#woa;8K-sF{StBLqPaYy
zBf7TVGo|d6nvqqN>EisX;5@fiVVhr7R_2QnvPUVOt|(m#xx1b9FdnDkJ&gJns86c_
zPd^V0utq=oA~0TlIE*v?>^X-{fzE@E$QpaNYT!N#4QnQV`bk<NlIdN;!f<fnh$Fz2
zbvp%V1?3uvJ0U`qL5VyVsd`nsA2eg5BMy>|f`;NsXN3JB?{R3HurOocl|2v`HTgNo
zTs%c4^4<>60eSm|+d0s1Q)b=PsFeYwdCv-o4S9dbS9AGO{#y)S3>*30H^Ieyc%I={
z=mBU*^JEb6mwBtU#~?$`$<U^e2%b%E8h#M2bVfZus?#G*3katD>2ljDe##{%%z#r4
zofKC*;vmva;r*&_;Rvs6wc=6Eh7LgIkz~{bke~2F--ewG$u4DCA<=0f#HmPQZHy(h
zP6u)LN;c9k+Mw)cnfxfP+Jef8Os|Ys&!NLJ@H=uWh0Y+6h@awYVjP2ZBFi>Gy0+Jb
z-gy*X*gZHK5+9$vE7ox|KL-C}a8;i8>73(4{vdSAHDc!$c_kjm4U+5(H|jr?SVuV_
z8B9ZfW1|{tt7aWb;#Ub@Iy=3)+FL!o@ikr4C|+ZKOh2wEeK<GEKc{ERxqME~Fu(XT
zldOvuJ~HZU5?3r<{qW<Ne}DPQe`X)Y#sBqR;&gn&dp$JbzU!mEfAYf*Kg7^w^&#(Q
zfEdF@GnBIic<%M&@!V(Z193B|d|at;f_E2ykG)=VMQ=V=%aZWwco<SwIUb$Aj<evG
zTwlMfw3|b9G~TeUL1)6f9gY$ELf((zlhHJ0s!A&tdm^~~;u9e5t7x9etz+e?$6I*l
zXuWy+BQif_j|}fQA@9ao@vm=c1;;%x-rl+8%3A~O@~x{Sk1>#5(YvR@59s;W=rvdH
z=iFPqz2){1s}wf*kyrIN3eGdx6GEa^5R5Ii9(}RWa&zHI1Xt>;Y@i@ro;Yy52g|mn
zE$B|?Lnrdxo42)ZhW1)_2|rgWo<6PB8E=u<MYeTzKu{k611z8T6gn`jY+aRp$CRkt
zb!(@17JNCSO)s~2&>7w!4aT8`UeMiuOL{B-k)~9F$m2@V6ZXpZbWAzCu3{lGPO^J7
z&c{lW!j)sEPH2Bdv$BUn(PFH<a21YyF4Bq<JRDQ?+w)4F!{(tzpYaQQ-_J6rNw@-a
zDFeUpteDdS@}r^}3#XB&Rjbo*b?`oF6~@8+iNaQ==`Xsb%d5Za2N+hTB)!0m-m6|6
zLIs$#eSBg#$_Y7<#}~BEpn+u^ZFp~o%AC)*8}M^_L*j$k2jTYM{UEhp4?yeOlhhBJ
zy?y;^tT0@iy<{WmS1-TfR?LgD-!XB!7a}hW<93Rh{M@vfiP8=kOY)xckF$|$te`N7
zyUL=0?@z^R5DkyUlOSwW(B@Sjb^`Hs!qp|2(J|gSsoO5T?wmSq&uD-(==DLifj>xz
zp9xAs1wB_y<e`J(hSqG5q})HIupnP!k#%S^oUV{O(t*O@<@+tHq`1nF$*;t(RUq(n
zE~Fvtn5UkhAghC>ft99n9%2z=+-)-kk`@eHXkdVQ1;M==hKmc^3EY*pND4RYN}G^A
z?IIp+a|PrI2hH0G-j2!UUDE2@cfY96Yi<pdJnF_<K?-A|bdrh(ze!U_tP@iEX5v&J
zNErcXsb}}*j8zx%1{N;@<72hCl|>$s3F<0;IEYzsK`gYBy-AKI_b9nRRA(g42f72m
z{~QvLlzBs~Oey$)Q9f{rmaBN|N#gce92y-N=r?W%RHH!#PbCu)1Gfs+2|HlAZbhpL
zKQb=;d<1123CH$H;Um~C9djaU<+be<B(kR+T|MG)oe`9-x|=DA?OYwwjrci-Ni4LM
z1eGJ*r+88Q^83j(;G|Qh`3a=cd2_EFNh2rt20!vlbVlBZqQ*MEpwy6^;EqXdCzfq~
z(CSI*7}6ht{V8aEh-d0)qpKiJwHgx*)xZDyzvp`<fBoCv&R+7A$Qv9H-)`ztGuFSv
zDfx;o-+pb~>o*t*t`xbIaIP^N!S*=Ozrb0k$8vc|8Bdz@lLvm2#Ci=wzPBR(T*Yd@
zG=4rdTDK5lJfR&YL`MY2Cl1Bn)%fRJ>Bc~W0za;#U|ijNsv-6pEw)iQ4wk;%(38C~
z$6+AW+3||`?fbWp@#_5*PS>-uZ?3<IQGEG=w&*b+u8z<(@2<$`<2CG#L;31(bqH-L
z#1YaBwRw9Z-q*E!J?sVI^LS+L)oG_ghryMW8tH&UKRO$=m!h1k)>F4J4qKiaVnyU#
zZ7P05U%A<|oQ}FZBX1puI_F*aQ1|6nv_oX0$_c4{v+w$7zSf)0Z9J!OR)M7LTTkWr
z_4BWikHd(=ll>nU$|pE>?(Yz{z8B*YSF9HC)Rn7HKB*H|fH9$rbyQcEA<<#2eEI+#
zOUE~}y6UC_^3~Vhl=D1RLhRcf?@hCN%XZlh%AOTDJ8{f6CVOVGeib(P)YCp1x&nJ>
zM*KT_6&{KV9E)C2#}O#M4A7{x#(D%Ay5H(M<JD<@=so$VODxa$kd$!znYT;KCv5u6
zvQ#Z}#46mS;|K18osj0$aTp)*Bi<s<5qMufKaB=H*l!Hb)tLG;Z0}Op4A$m3kBvEf
zm%xWhPR0MAjy}Npu@f_g6Rjt*#>CB@-g40}{PnMYP5CQ6Kc5L2dOblpnftnTjatlb
zPZ;i>&`?od2YJd2nFug+{;h)(zc3mwR}}bzp0`{y7}pra8x9>1jeO7{vCwI9Lxy@h
zcEIa=&?f14E5@@L>C&EdJZw`R8uz*q$7z74FI97(J;R_~a9iXWebtwJ2M%KUxUKpU
z?(q!<nAH#IDW~n|BXVavRb6zT?CoSgr-mE#2J+c>DxxUnrK#b~QL8-ij-F_vADnZO
zrtuArp<0*Q3phVMDr-<S9la2xEY1h3Kdx%jrdQ!`7J&2c_LXD9KKF1e_}H*`NRs;W
z4g76uo~%J&AE(wSc;}<Y;MkuAS5(pv(r}2uP50aE!E-EtL}faAF4jbkl%y5;5uxS*
z4Svh3!xyK<aR@yR>+XauGZtn72)xQ}80)6=3_GE%l52{qsxO&C8fYD-e^GgQ+uu>p
zc-;XVTRx|^aC0^1kwJ>4ns%kDaGWY@ngig$P6(tNwTVvpQBawF2@lqQC*9tY{Q(qq
zxK6aA-$uOT2c6V9Sg8-c9=H$Sa|mRg+}-_m>Slg(`VjH^%0Cvrb#)()J#|jZKK0@<
zwLv5HBhJU)Sq1s?pZ}b4uWWrn#+8p}eEs#?Z`d0Go=%4M*}vjDDqeN_4iD{w{p6cY
zHB^4}_Bs0w+}NtYvVWV8a50R^&LK^&{w*hoEOom@XTp^e58rPQXGB98M+9T1<x(Dz
z2z?tynX*^J_ncrneaDGyaYdw7LCOhHPI2483V$WV0_N?eI9Zmx;ALydnSBM6l_A!c
z%HLHO<m4r8aX`4bPy28JGPw=CSf;srbk2U2q$6^EbT0h!zkYPcsx0L=O`u@Zp{aur
zhYTl6hXWmDvo1QbJawdX99#v;<IFlKw>lYT?5ioAU%4WKF4dL#_HpIN4SgW80ug5d
zUF6YdXy?|N?dQlV_Z63*lc$3bM+_!9pZ1F@66#7h^DYkZj-4ld1kpYl{Z?`HVHKOM
zmfIftt{=~i4&bHE3?V_%SN;1!J>R{eW9<`C$|%1cAHv&4^En0qZvh3S<xx?Dexq$N
z=DU7Zkkp-WOG~_lJTCqi+~YXife!1>#NS!M;F&{Djm|YW`eUPi`?A0aFC%8io{i*q
z5`7M)F5r|*?Hi<Jy9zZP_VhVW$2lP@vR^VkhX!iuIDwM`Orx?L6&YwX2p6itT0RfT
zIlTa$kaz%{X$+sh;Fp79J3Ly&Whj}@(LtHPsuw03See{?S4j4T;A9wUcr+5X7z+v6
zQ^M!!Q={OjgVZEQ&?pF@;RK<0ry|+lN>e971Njz5LWjVoR%+Q_9J$o%-OH$zBNc-~
zjmFYRL<21YH<l_-UL%f?%@w&Ms16)XP1>#}h47Z?U}*5=t>g0S-P5eZ_<D04CO4ky
zTsdZMd`y?LBrh=TSH9Zb$9D}BMk^uhEimAi9P2gbYD#{^mn9=J;jau?9YTke>E$Vp
zjzeT+lWK0sv?Bsv^S0Sdq}S%7$666`kBE6+Q!gDKkv9MK&2U1v{iK}PXMz&crR~!Z
zX-B7=5R5ywI$Fn0=(pFG!E+O?EmFUcpV?ZPS&70KdCaYs?D=4?h+iGfxDcU(%^d9{
zG1}@%&S-P%QurA!+0vVPcOG--2t7h}b!9ti%rqlaS_J$}T78p-Md>$VYzN<|4cjiR
zKHB9-0eOJN3hYZ}zXnGCM2~9Ial7=;gfcqB-V??CE1XJAkWC6&TablFI0R=Dt?)<Y
zC?{k)@b(crT(+kXhk<`HHGOiP(x5t-Pt;o`fq^>)U?yAAcPKI-Ge-I(&*%ec;GH6S
z<IH34jxGlMReAGn0>73Rxz($O_v&3;uw8B<eePRBG2Aa%{dmri7nyNe;_I)!&K{9B
zZ(i4Qz~nBD4ntjWLxYJQ3xGYXtKf4Dw#E$OTlC8y=wM{xh>>-EQ1i7y;#Iu2gnr=N
zB)*@-r(j&o$SnkTmVr4x`wH3IJMj3c;?*x2d&Fj14bKbqujsfuLH3S+e8uxWym?iF
ztvtecXlUcKi~(@TJ4t+^M~C4#j*!M*ohYXU*tGGp=i@_sjK%DG!H8OJIvXTuw@nLP
zbmA>6?=5)6SGp*d4@VE<EKaY`7<UlXH@kXHxFI)>7mpQ!OJz6hssZi2yz<>C?yYcT
z1SbRdzjI4X{H%Py{|$ND`wBgaYq`PZ>wwq>R}x&wiW6YF{IV?$i2Hebzlcwd#9_4$
z3lp8%{q#S7wo^yymfJ!)AHH4IHa)-b$(-ULPeU9YZe{t{^(9WtB{I8$l9e5ts@{ZL
zj+m<%<;)hgPGjlc&jBd6GAct}TMsW~=@l>geK{mReFFM0s1Nnr{gA@>$3_yQ(P5YB
zk~WCD{I4RVZutA0p;tso2IhnoXUzAr-doOq%y6XB5*%aCVb8a6V+ts?1!TL>InYow
zqxBY6LF(q8qnwZ#@k{!F8aSZ%2Vn>IV+r(-7#tmVWAH69WaLmjO#-X*0{E~*Pe=!$
zb7CH>{Z&?Y61p0ZPot|xjDrq{dmw)I-FI<7{`J57m$Pq4|I<JH^Vtgw{_no4mzQaf
zoz&ecqA~F5HCM2lwRzN4L&L2YUr^@KeHo=vQyCBYxK~1$B+N2d=oq}?b?R=2{qe_F
zXFt69;q31}{5`7?u1F}eef8~dKHPNp94F*kZW(DbU$TL;uHfr7icX8ik1@dI*tpN-
zZ-4tc2CnK{*^uYuWf}Q4K;lVLSK%G6{T5!wL&I&k`Y`|1l@6P57_Go#ta$=O7~37=
zijH)`e9@MwMqM%Nw%4Z9^0bTPgcx;K7c$Z2(NzEl^7ZzLMmWY<105r+p)O;s%sL^q
z#}y=TOdJ3B>J^VN>x9&!yFc<cwrzTaLmApQemQd;>&WSTAB}b#D+NH%yu6LJ=FPfn
zHh%u1-k^KoE7Gy!`B<E|w7bqJV^3dX=5PaZ+K+=FK_6XS2gd@8`k+z=%63BlrMai1
zE<mE6*-Bt&%6tsx1k(1$^JibDpW2}i&#`Npw;|Lg-rM@(xhYj&n<BF*+gGs%_D+#1
zxKnsU!ABt<!RM2MZWU%jEUUO`rxlR~PkSrVy&+UW!$?Vo#PU(rRUHezI6(1Bn_58>
zE^}N6fBE^zG|+1Q$rzkSIFl2&+lgSCKau<j?GVjF<tyYPGt!(-f+Nvu(mobxz&jCZ
zXq=#PMN8bbTE6+_o3nrY*MB|x=3AYRe@gi`tc3UmPQxpVS_~~$!S+obUODH*X1*7M
zeIwZypoP-<5ZsiQ$0{L{X(E%pX$&c}bqf5hiH?JZe7s8cmXBA{%BrK}D#v#`x#PZ;
z-~ayiF`6%t&-OgS_((gOC0Px4jiJ>zOHW*#2lqhe*!=io-L`r6uAGqC-@q*d40QGi
z;21o8=H8O`;h|Bq-Cw_ZNjXjr&C2Z_WusgJ>MDV&2N$=jY&d1{LNnpik()*}f;w_K
z6U(>;^7cu+i{r_qZ!u-B1-Es2bl6vYi>1@z*WrpsIU=qi#gStVjklk?jinJ+2kQ3q
zYn>CE5A>?jqQfUW(}s*k9El6|ZRpTt1<Mi2f8Hm96LR&u9{Y9G=E)OR44js|-BK%3
zZSZ^L!Icm{C|F}eU&7u|Cz|~V#a9QxvDBr+q{Yjskr@<&C<Nn537&J)Iz>&f4)MjA
z<G98tD>BkS4%^yk<2Az&|EM2jS@!vyNU7Y@6On(dX!cumX|=$PokQ1+@?HOL0QbOU
zHAGq)+?jmR1Gi|C$-SE$l#a|7CuFCOpCHxU+P(M_u=uTzbi(V|vSz(sRm<#QZ)Icy
zskbj^%TtV~6R20g->}N?$3Okk+1KBA3x<iD3Ej!umt4Kz%eWezzM|Bq^+Yc2Q=AXq
zC8EK1|Aa=V&;H)i5Tju!eEo}I#;Cfv)U-!r!C{cEaJ`ZCPspVc@)b^p@<$$T0gZl$
zai@VXFwk~2;)YkN>m2;<ci%zhS#I}u%SYOV#UNt@O>4NbuZ22S5WFoASoMQU?wxoJ
zT<weTWo)z+TyF=I@ht;>s{_KsrEEGd{A2|~JxEg-727k<*T1_8am!o1g!h)#xOPJy
z=*WE@!+EKF66MT9hofAYI$Uu+xQ*faT-?-Yo9yGfqzzc{i$<-?L=M^t2pWUNSe;R>
z=v94=Kl}3vhpF&c;m~PeJg67PMLlBdEfSnh==m6~bktw14wchet72+j9mvO(BK7B(
zS+Af8m7uZ+fmb^#`(tJl5V?QpPk;JT=-Ix$9px&K`Y@!QsY@Hl8+)6n^&w2xnf^IK
z>-kh(;{TRHdff4Ms97~6R!w%i8MeAORK7y~K<0}R@&N^XboHJN{b-sGB+w_j6Tlp{
z`MK^(`a}&d*~VdDqJGZ1FEpwe(sK-!&coL@9<G4+UJ&0EqH*x{jdd?yvPys<@(MbR
z2)LQFq3L~pR|fnR&2w&JxH6I@eU{Vx%T@50^wZ1sHUAp4>`fV0_X|A?g2v0^8=lUQ
z=AZr<CxqJ}FL0i^&dCL{nwjbfc?@p&^wvhccFUfD3)=b}ZT$W3&(D10tA~6v+O7y_
zH1nj123s1`dCSBV4mW$sOB<-Mb*u`b9LoRfIk$`8?>9;wJ*pLr8&(r;Sq;fXRN!@v
zYQ@6W#k(rP6*;j>z^Iqo6}5r(maorB|1mEqQ-0;}aAikFLT4sV0)dO#s8X-gQQoT>
zIw5|X&v&JGiz&C>q!S$RDlNT~5tEWx4qooM1ZBy&Syj=4Pb$@uDj4u=1l18Khw(|i
z-J{^HX(Jo&A~84)bO?1=Am~2<zH!oYKHLveFpe*8?F9na?BDvM1iU5Z<IVmn8^O5J
zZb-g3`@<jp0H6ABpL>~fJk?R(-b+7&XZv(I6{?&k<r+jodG`}t?6OPhQ27FIOkzLl
zV<`M)VC`!|kI-$yma*VKm_wn7xC{HrcC`;mIePB^9vCZk;JTm&zquW)q;}a1%7*{!
zF#;>v_FD!Km&WH0MQ7!xPog1oci^2i%pG&5^2CBke@9cG9U3BoCx45V$-yhCO>n9B
zml8f^@>wRjsm=S^)jSf9@sJ(kC+fjTK3A?d3H^C@&kLQ5KmPHL80znt#JSfGoUcrG
zsojm5Iu~}%w=8pi$9|E&{^hT1a(tcslIB-zYV|hB*SzlBl?$2`LoSlTqgN3${N2YH
zzZz2xWpH)D7o)lA;&Qq;Ue)?Yu<6thYoxN*LMBqv!1CwHO!yAN>?6J!9$$9Ws{%i8
zJK`04PJVdho|D=?p%K)P$ZCxaO{i(ydA#>+ZEW<(i|heGPH%6xAyVh<9Xyo#TUL%<
z;<#O3zkrU$R|kPrBlvsEqOSI<ofHssDuD567xr%`w|MH|;<5Vo)(F#{&BMvCrOsRV
zZq8WY$kQ-Mk5!P|lF<oquM6MT!3lZut{zi%Pl&qEIf>Jvg91~>m47k7@b2-JP7BX^
z>PT4@#}aA$lm9q%<D5R>yF1xni#`pP?|okgFR;_ug`mz<>819F*dBvzR`&+SlJI}{
z;~%1ffB*Bp<0!tZcb_2dV?PLpbMb_|Dbjk2bD=Y&Q~2_$7jdA!|Nf7R<v-9)`>I~3
zrW0bYeey7<51o;`rrmM{A+&>%3_pFYcARoM40!2vd32ewnxpKqUeQ9-Lfa*(VEn32
z)e%RHvR-WpZK|5bf&^^%f#je6s&ECj<=M!iytBb46B)Yjh*)xtA60Tv5O2n%EpsUO
zY@<M<y)%G~?=a{Hm5Trz-QJ)YkO?%j=&EjilSnLHgenAA5`7K9o-j+kr}|FOV7DSh
z*^2fmbBDsuRiG64`XmWoinQpNCact(176h--2644Tq)f9Q}Cw_w!I`r;ogqC4`HgC
zPn>9fbck*G%?A3+-k95Nr>4&-&n+L?ztutVmY=AYc0x2N8exs34u`L4|KlJ2c=pYA
z-{O$Cr{g?^?Dm4&As5_Y;Z}yY-lowBF`c)ka@C)2t>8TT&UcG&t3+e$MD8SC#*MdV
z!p}1Ic+9g^-KNn1`F)dmxs^_VMq6jZTPQj}%BZo|De^+w_InPfMPH7k9$jLWhkGLA
z?NzT=?tWwD&6^)NUT06lD>lEr#jw5A3Bl3Oa8|%!un(Xc!%yP+`YpVikl5>?Q$Xtd
zrBBUV$1!q=UkC1nCp*Nk9_-91*VnA%;3QmJaN7nQM({W}BnjEOqNC$}4RGxbgY7P-
ztKL(h&UC(A!P2ODdqF274qfDRrNyUW*qG`{2u@9&DuK5vJ$dH}x0ck6D<$1G^0I$2
zci@y7o7c#zV0G$|Jt6S6&(*(h8a(x^?rRL`H1Xt<4gxDWpq*Xvq)t}d=r<}#KbBvQ
zC#2+~c2lR`%Hes&s@L~85jwv<C1ragc#E#Cu3S-hgwE^n?APq~G3bPR^Yyoj6QU!c
zQ|gBcU%&n_?GP?)r5(mSUZicF#tV{%td*wE&FH}j8#O}4U#G>L(mkHEnQZ_I-hzf)
z5-E^$95_~sg!S7>u3D*EG`z?XU;w1;xXigS<kk-C0{Tj(#`CUxPfGcNJLB(xJ5Ur^
zJZDcs=Se2ox1y~EcA?d4ds9O}Lx|V?6pSdnaD^)U7qlALVG4ueImW;3j|@W(?uRzS
zIVNqbXa8?LWLeNww$?w?;sKXa`Jp%;qIp8S9qZEasi6(;sraYjW#R$3<Au%vUEXQo
z0l0<EeTgw)H19r<?(i5i4wc`>;!s`AFJ#x)$@g16U#}D5<o|4}f;b_&?C#aHA4RJt
zEXE7UDEF%P#mnbsPsPK?eEs!TdF8vCHoc9}#;s+jCGQ&Aw#8{DlLCB%*U7oP?bom~
zqL42eQZZC+7JY=Vb7e*x4eJ%@A8{<aZK46r?Fz2Uy|trp5#OV(nt8ef7#$NG2Azqk
zuc1fVUj6Wjr)K_!1At+C{V1y>8ZR9R4eDb)DIcSZ%uYytiPu-2zvoe2_dvXG(<^%H
zO|%+#58IzxA1a$YA6&|3HO4EdynGD<bA>bE3e7uKE8NrZif{Oc>um~+i+a`Y`x<<8
zbVc9THruN%vf2@T*U^cOAiL6|;eGE%{-EtsFuf;2+OB?;QFT?p4-J;XBQI~M=|HyA
z@!}=7zA)~&os#_;?%D86FJnFCY3P*`sIwsLAAhX8?-BRO8oHr$E-m{dea>nYZEPdz
zz8G=s8&}f)`H%m6_RUw{u;TWY$Z5Z;<8zD*=u7{)Vy4VK$@T4b-()YyQ{Hm>zyIxj
zL9fP{n{(SS?ORE?9o2(##MT);=3G<G`@(X*Z-8=ZZVzY6HJyI+F_ignz6~7}JBLkn
zu6j`Ve$gL9;uN4mtJ{u;ZLp-?9kQ)CEnnw3I$p5~mmNX$dd~Nj{vn)@U8{eEsZ&pY
zRp<Y5+I^=vp71?V$H{*O=1B4HG<}{_w&^J^-_2?AaL(NZFf^1}h`Q~;RXYa56^Cw@
zJe&wL)*3vGK&~Ds(?Gk|;u%+~`C2JP%&Xqq+lQ_+nofj^{AIxWDBe6*JE>-cgThI?
zfOtiZXOXlk2O0_=J5`ofT;Xb1uSqLs)*;Ay_2|T=QS}8|NS3^9)5tl22M<_fbRzo8
zU!KRfD~GR&cRAcT--fEqRTnNiUi*SB>XhipkM_wA_~&feeaEVZw`wl0c(hl!fP2gl
zUcQzdV@(6%FiArmah`;YbAb~@e8XeIGMARO#$2M;JgVDVHR5<Fv(A$69$6XDImt?h
zPXM{8RtHCUELTUaM6u}uU8#$VA#|Z5psrle(gASA$uZ$3Wmmp*!n{gXZ>>8WELVrr
zyQ>q*bIaRVZEIF&s4r*5l_ScPAE!(G%1h?<d;J=dB63W+IrYU0AG^NIMrSuXKf&R<
z&KHSAEW?~t7y8o`H1{`^Bh4)#>VLpFl)m)hfFPH$+Gb@jRu9Fw@}@0Mo%JUM$nL^O
z8#R3^898e{bt+o50~A3(2lfL)>GXlX6If|xD)^a`-49!~vnso8Us=~37mrpuF76QB
z6d>RiC*(Noet~|V2Ar@~Vaj*i24V$%a;Ev*9sMEMy!X-1=>h)cG}#Aww@kbk2=1d}
za5P43c)aq}p)hF3-QXyXTUH)kbCvC~_TT^dx7>ztrt}K*`|p3B_zyfT`@7%$KCe1=
zgK95xYPdUS$XsHNp|#yyDf@z|7*Ve5b#PpHsXYy~>f%!#dE$fezi?Ho^Wp2c-526(
z<uz!&$Hdq@DcKy0LCO_1SNXQjD`TC9tDDQT&y@=ek`93H5Bbmk_zzxF&psF4Bl7?L
z&;OZQHtrem#b9}bIXb&h7sPL{;Y1h*ujJ3q-bx=nSZa9U_+S)u8vKwSE`!A&;K)24
z&Wr8V8L>_>A;m2aznV+A`%rvITO1dhhP-?W8W&I9U-Gi##^C3661wqrkjE=t#^#3G
z|NP5;#PM*&BP${FnX4>_8sq$!cYEl}*mtf9eaA**<^TGd@1nb}-G4(L>AbzfvGQpo
zb?u6beeSr_G0aK={T0VXCkh@v{`f;2pX(oRj`(};s*$>*UK*E|wd(Tq%dd;34hIT~
zj@;ymPPrKs9=_uJ`Fn4%d_#NQ^4=I}>qPmf03A<O$DsM*>le8N=xsp9@ih+XyLYe8
z{@4HhGY<0ey!XbHGiC6w-}#C2Xj|%*HwqqdB0%^Xcj$8wtW~hQ>y`?Eq)+HNa}I-)
znlKG+hpOYZ{sY*~L`cbpk&+bvRzwy|6BC`L_c`L0_Y~e~0&R(Y_({xf(@dX9c<7a0
zT(Rku%2nY5Gq>Xcr=76wvqqou|1eI7<0>j9VLD++kQ?bl{s25z7s9c>gV4#8K3;-j
zCH|@4?`_E^KH;0!`jk4{vWD;YFy!GHd6oO(QzIIidMvT8xXako`Cd<BaC-v>!SAoQ
zxle=V^1e@?cm=8><C6WeaXZJ=5Dm_-dS{O?R2@8_bwu3b<kfe*8={;64W=6%z11KO
z4bL0+`Hqj*^2GW0Ghh4lwvb=?)quLs#Gs?$Ed;NImB*)P7AHhQ2b@lbhi`(-MoA{L
z|DU}#YmX$y(KV$ADfZkeE9><1oav{}%{=G-|DBn+>7}}JsmzFsT~f^Zejv>0CU?oD
zSTeJ^DVZCM1VI3#5ipQOb8wRm*EM|k<=0}u(D^azGvv4Cs+E|qLu$jSa<2>dxln$b
zh3_|0vXdz|U3gjpXyNE~TLw1j_BCFoC^9Xfe8kD0d#9+_^&Kg1C*03YYM%>@l{0wJ
ztKqs9ryE`{HT9h791R$!McQknH$@}w&K)-v(+Dh&O19xghI()^oi-N#YINuYQGWNB
zBI~&A+H~KI@nIh+rG6I!(XR@>W$~*#>d7DrV}Ug;@aNRg#R-d4-^dO8Sq?Z5*5H+s
zC4E!IMa;zDS3^hdj7H=sINxAA>2(n@UL@e}>FMH#7mmUjvaXwX{`A?=#UmCIQbzuF
zg}N8W2^~^qOYaWm9TG~Jl>)7Jv6#9ViS?~QaYbThH6dPid{yRma~T@>Y1(f}SeJ{C
zEKFqLIp}Ac$s52aJK{6Uf$?smL%>;>+LT@ozF1DzgZ{oC;4$)DSgji)znvC<ZX+&8
zPnW-ed@eIxcpB^F{A*IA4VUe01LMtaA13*AUu%&!ah<;Jy3=oRZr0azHICE?ZA;yQ
zPRs2p(`EOa9o#Lin_Jg=260CIdiAxo&~`A_`=M*y>5pGnAo$DEUygqG;fHt+OxF-_
z!!Tdv{pu?gr7}N%A8(2Z&Tn}b(4_?51SzF`SFZrd(6<~^-o=hn8%!Ob#GKO5*y$ri
z2nxRE^z}aEO$dw$>r+YQ2@!godXTYfgAU?f-qjANF(^>%8cvq=hyeIef&U7hEIL6R
zcrZ}H^8OTFkIs0b1BIrv&W=@N&>N3}QwCh2U^TL&%{pB8`sx)jJNqn3Qdo^TJtOx~
zl<u2>=7wgsL<j2y-{3(Z*Hi<B0wQw?MT1KDDvx-Sq+qOlF<(RN89U0lXw&s99)5(-
zjmR|MqI5Nu_&;?~EK1!qJoX{mrf4cJL>d5@Y6zL`R`+^}G%%fF(fbwe)tD9%Pklp2
z!wJ<n?t8rqmX~ggi5Nk6E2Kppodc0}pIU8e4XxA-Y;Eu*E{wuc#}Nz;recKgtt$<w
zC%=A{w_qIGJ?u9cRgdtLJOoC(*7@)M_HSAD;;y?+$=TM@ChuNru<C6K4r&sS{ug_z
zy9nDp9hQ4;q%GGCFwVQu?KA$H7mRf~Py#yqqwl2B#$n?V*19V+%QLdDNR{9Gn$lr0
zWIEvnXLt1t!8a%~gKCi8Hpru)^?(lrwoA5cx2=buwq=7bx~YJ@ZCJP6`g+FuU(gd`
zM;#)VWo+Ee&0jiY8fg~QZQ{X3!MUQt@pLCSyu<{V4s^SECoJzB-AloHbnruo>bkqR
z*X26D&*snZHT<>L_8BTuiA3?BV0``bZWa-8^P}+GrBj9C!0wcYMu7%_$~JGEpj<3_
z56^?%g%~p`D%X|ZO4r-r&{+c~13Zc#ixMMGJ1`tjze+6qhorm>0Gx^|hjEU42ZXbO
zl?ubBUJn(DX;Hf1%UcwTh0^6s;&h7Mm1j&7_-I~6NO^8rMy{O*KfqbPW%UAN9R!p)
zv}SxAyL{ed+63n6fCUFeJvy$NDFXooi~RZ?jf^8_6@8K;7xpqkIa}Tu_(&c+KX~i9
zl$wrm%0$hbcD5W1<{FSJGQ~48w~5Aw?W5<YUrCJ(h)nKcP&jRILgmU@BO-<kQy*T(
z3u9YwH~{;hJ!zJIJz-Ae_+6koZafCxwyQK*<PP7w(GU<`890^awSB>$v7&t3Cq+RT
z_b;;HB%WN1gu)BoM5IwKhQ?6dlA>?DWh%)juQn>mvqKs%fQMG!d~(sSbhpxyFOT#;
zYD_(;+Y7EJ9OVOj2J<VH%Yqr{BPFLTm^ot=U+_ss-w-zMHKEH%VyH^lQp!n7`U=Ue
zv=zwC%R`yjS@``Et5+#^2q}iSkk?Hx-emT9&fh@ZBsOE1YnbDOua;OJihFdXL&1Nq
zaJey%Mb0~jOoQ!6zfDqE(Q|{`2Fo9;jGetZJfydr*HY;8g}bTDl6K2T<L;Zv8!uth
zvCz_Wa$eGIQKw~n(sy1TRTHm)(-^wYt-^u(4NtJI`Pyo|%KGx?`|rO$`h~R<uAj)k
zleb<Fr$u}_!SBpGW}g2su6?ISR7xT%T<`<VJo$5u*TOS(WxWPU(&-YpQh906xFf0t
zPz({32?|by`27#xXUZe^;p69@ehOX<0E4pBh;VwL<*LGpA@b}O-h_FYr+Qud9fVWi
zx|T$udb@aAAx00pIC*f!Yo(5W?@-8h!$&<oOJk-NUa(;H#ml277;M%V?+0%i`6M6D
z0#gb@hk&`-SiHtASSuk8-w^UC{ox}NE*=<-3gzajmwH?DO!;v?*M9u`i*51C@Q6H%
zv8cx&mB2^p^2}?(>bn!BV>CKc=o(p%c(chi%(Rx?Bw(dmgG@u{HFC@jrYdNW(FQKS
zeG9=Z5EUNXpSD)!T&6FvnL^%_ddr?6Z(r4`#2T^&_gaaOr||+U2K$J-XZ+Ei!SL|)
z@LC&khqXFi#voA^cj%)|K`Ha_g3<FOpBK>EaZI_l$o_{PehO`Rv`lxMi2b9DRSl8$
zc-R*c<dJk(2t5d*(#NlP34QMLE<dj4Q?QL<bNV{I;9h5N9qPm}>h-ktpzJ7i`=kTt
z-G_W7ntfnbr+59MS?aPLW>Roe;xmAEPCO+6vOpKUx{B>w4QRzXl}+9D$%1S{)-zrE
zu+ZcT9ETP99Y8Jo5t-wANfQVr1G^O13w?dYd*GPc@cN1_L;O~4CCiuCK7g4^WiGRH
zs}%9JES5T;XNfnx(&zwq0Ih36#Gixmr2^~ezuf-jsc`FUn8g!<0`P0L9_~Y-qH;Zn
z3MGaE$}FA~o_sYBR1h;S;VJb=Hdjed`J<R9bIfS+E`~!XH4Oq4O#(c+iBvoy6wJOC
zPC1}hejSF2-kB(Cl@o=ep?OqJ;`-(1zcFp`a||Ap@E`&nSi{2gV_q?qpE7=!2#7Z$
zgoX?aAxXcJw_Nm)q=tGb_dy?c%j$g*wl^D7K2B$xvaqy*6dJ%MPX5j@Y~m4-PiXZu
z-nXZGSZ-Xa#PtU01##`iQw%DnV}51vtkWfab4SBLnMC$f5k-q?KY?gfdORTT>Ygi3
zVLjsYSHA?TktF{bBl2GhMT<tNN0#)zxMby@!5Ib3Z{+-zbPW}~Ai_4hi(0+2G`{S;
z8X2BL2k>q##^oj`%8_KB$hslm+z`%1vl<h5Lx#OU3b)=G_b;hM!Fo%eG=0EzDR<ey
z)w(rC^_WTXImAWTG@)J(3Z%<;A0s&@(*}8S@)9?14vjPX^abaLP{2-`3+;(ZZ{go~
z1Y%t7fq$RmjWlk9#a=99YiZmZSBs}GjVp!6|B^>UW2~jK$+>xb8}DqDT{u`G8@Bom
zzIiQwkP`Cj{ZRrPT_TBl7$C2j!9uy%1NbSD?G+*Z7M01xZOh)Gx~Ygac?PCAO-*B%
z^Nd&O*6+IPR_1{hsjcgzOylhHPBg%Wo~hhUxE+a>JHv>0i=yu7gSPg4{>#(?%G5Vh
z+(hfe^Ot$uQFp)V4cvpmDGZzyQD}NU9z8^Xt9ZbPK0+Np9|}#az3?He{6~31L_Y~@
z{HP@CJ}OGR6vdWGrlwuuxd6boU!>uOpMK06I;X7L@GTxzl|})%d!~z6UAN*Q(P!)d
zB5%t0EuJ0UtT^(aoU|w=>a8d5yLf$2#?lP5@UBhKrLRFIjzzgMP33gVTMRAJ-G4*_
zNJabk%g^JHpm>Zk7r<)VxDG?buR=Efp2Ig`G(LQzLtMTQqA??s295M+)V!da>qCBo
z?|=K(e>?i?U;kRuFBo4jKA<s$hBlRcUzb(JeoW9s!(SsK4UND2!~e+Q-S7VVmne7F
z^JqwDM0Q(6X3*u7&r90dAWpqeM=tP11O>x2%0L-rY6NeO=TiQsM{l0IIy!wr8@}P2
zJN#+P=poXOa4N_8Z1-J3M~R*jah>Tc(|F-e5!y@|8p`!CYmhYb&f!-f_zL&QgOltR
zqDL)jLXfS7bI}YPKq{(KE&(ed0+;Ne<~pBqldC)uY|IfXrvX*?8j?&v`YIUu!mLrg
zW#71Rw%njEn{Fw?O?8o}xEsfsk1br+UWn*gpuGUzAMb@g%ppOUE1-#(MP5rJ0<A|K
z3%*Af@?9`Y?YX2;tup?*H$u9fEB{5*iS$*UZ5-ll9TJ_IYmwWhjx=?jy50{~-_igd
z4C9AK)HL?_HZ^cFT~Iy-hfm!M+T3CMduAXppSYKvv$+;<@k9i!dYj$`pPZdiQ3*9k
z%YZm&+w=?j>=^u$S2>o`RUYpJ%6%>`Zl`Lj$H7?Fy08ye017E6{M!A~yzl{D{U;5t
zS0P`UMR{JJq{LkwjWR;=gl<{xrR@Akyr|mTQ!O9s?bOIFa#KDY8gIx<qQVc#a`3b{
z*OR@p3g*ul1Gi}IT9%}X@afbEo)JD~=eiK}zKYeU5O;ok#l~iOKJ<=cleVsx^6I}1
z9i=+bC8~~o@{^7}%G=>}857k8dPVTYfG~2Cz))%bsCJU3!1AhflrO4Xn)NOTTKMcA
zwu+)D4^i4D%DPfZXoJRXfygL1MFV9`Dw<s{f^ox<{(<r8NB(RhuT9sRmUTmEH~VtY
z9O0({3YQdI6~ey&%4?i+5}z_1xJ>I1QTN6A93<<+J=X(Pd<kiTOVN?}M?X!>rKJZy
zEdwvdEx5T83uZ?XjKUxGbhEc(K5X5e1{-Z)H}5}tyf>VoPSr^Mj@0DM?Ntf4{7ogJ
z#cztjvbQW+T-~zpP3c_0&$TY|!|>8_Ux8mr-v_rYx9hqF`XRQ+4{CPYO&N4WAYGms
znXs0-VNl+>?OkKwldg*56<&{1KDc&@@>7}Cp@HIh46Tt`RI1zFCuR?ys$CPJd*3Hf
z%bcMYRk}_E+(V%ne439E<Egg(s>OZUiCR&k%Zg9bKJCI&DLoAZo?yZ0j;TJ?`|TK~
z4^F{p`pd6*trj`t>!-lFPlf9xtkbu7^olrD^AwMWUIyPvIoII8aB$INo~EJQbrQPt
z^~xC9v+)+(FId+gJrD1)g*xJ9BX9sPT!z7<k!CZ9+l86#5Au*{4maDn|G=kpUmtc4
z5sf3C{(A_8>9^p)t#E#99pv+;k#O<=FUwoXo#9P@&tIOgmkF;9{|6orzoqiyPv1vo
z(yF{PA}n9iKzfzz7d0*B)1rBq#$uhxzy9mLXYue)Km8nfeZ$E1`wXvviu*Bdn|N(p
z8OMWQ1A&B^(&y|G%J2nm0_iQPU0C_3A4Z}^z?ZBMvb~g}hC%Ih0$;R?GS({;1A;fD
zYU;&%u5^s`L?+as=jDy~F(zFjqgTjj85`m;o2cn&avu@f@(sqBE}C!n{*nFiD?VW7
zB3tRqo;UObW%-7_FAW+(nnrL2OnPFSiLXQ7gGyEUwgQA(zJq+n=rbV&Drbf<W#4tR
zUA5~12YT<F+%x0hQ_4;Azj1Zhs_|yOlDgsIW@UxOnRn|@(~=h7fkfVO&s28kEVff$
zT8=ipb)cW6^9=;{*`mXmMvw;n{_Z^?cBUoF@nk5NPVBhnkvyW6@W=XXu6`4$9r-d~
z|0cEHyi#fH0q;HNj*&6EAaSw#C}01(hJnb)0dDhDUxT;{2=m!_&eD||nDenWHlf<K
zWE}NwR7EBa#z&orqewFuGPduofmMT{z#VL?CW|GV-T;4lXQb`IS8SOMcyRA+HmGu%
z!Y=_!g9_NPZo%$LE|RA~<+OlX)j!Z+K=EdM0~MB6$ohDt^>wW&s3(278bKC3sJmpN
zQR9=iXT2U80oFbRc}|kjU@Z>BxIW@ME9^8<v~L{lP-OmP?L=)?-UfmDrN~>9olFtL
zsl23H)#QIBFUas|-8_v2jXI~Z?qcAOh=JiIZpaM-OhxUqmy<ynmS@DB(#qN#aYz2h
z#8<3G@U>wLTBq(TpMjeFKwdx87?l~Y%`(zs2F=nQ&yu`Ohjix%ehN~jbb5F)v>~U#
z=emz<tOnkErw8NMZw;C5mwi>+@@K!auh{P+C-6ja)NSy0pwIOj9}QMQPVVZb37eA#
z-c6Di*R_zePHvDQYg2q&ZzT^Rw`kpz_=Cl+;G&uCGn`G)$nc8N*UG_uy$Qw}`;=}H
zmw*38$VL+X9s5KQ(EHI{ZW(Omx*Z_r>kk-j1y#!S10uc-)fkZd?7h<&Doqued`go8
zwV|(-i|IbSnwQ0wC?0p*^oi5AVY03Rye^(}t8zUd_xpZUkrI4IBMNbhyqWR=#-|)~
z$S?v>P%2i6D0!cVU6Y~GjUv@pVjYDG67L+}%Tu~a(l=V7V8DHs1)%@*&;KiV-+udT
zUMv1L_KtYLMqaPjNi%TZdBo;idR5$Q^-K0RaqWPwg!)xm71xhH{lGer=STngFaH``
zmXAUXH~c`Ow4UDe?H=DcIjt`Yv&KS?jh-Tvtz}fW$KdtDYyIRe0*wzB<(}SQ%>?p0
zJFc}I?kS>@RIy6W5C8cC>s-F)J1ReAFA{n1RcJjH6ieBQ3p_@wPq{~1=}|dBN!zwG
z3Lb>|vaY~SS)6OIoJN*BXpq^Kwy_?O+Bx?DYjD2c&8N@PM)m={BX{nB9e4-tytPBy
zJj=ItULPG(pWY5VY~uQy#jhG<E@HNx7-$$Lo;9#s^HJU}4K@f8tNdUJX@qG?DQAQ5
zdX*kOW_MiV^DQ0^+suzH%7<RCv>|z}716NN^KpLW*M9N(_4*m&Vm_88c*^}prK#P%
zF9CDlTHYliq!3RSa4TqFmFqw-;R=2*Y~Am`rsjcilrk}OdTreuEUU-RGBp)s&YFUT
z{^Oh{G3_LJ62Z>=+b3$AT6&gY$Lrkk_sq9EJ9Bh~OEqsAtGdiKK;`<of!l`YD@$Jo
z)37%<r|K*4!d~;}z!VZ1yw6nEuV_UYbjTK9Q9S@B_%((5v--|impJTtQs;*i7*PA^
zIULQVXKa_b61P43fZn|iw>9;Ea%DXN9#Btktl*Y3(w900%yKRM5w>Xc&K7okhc_<L
zeJU2lMVQ`XPD`jbH7aVY2MPp*qnAM3K2iD#t#Mz0bd82^1{Jq7sbK7u8X0-wCCzkS
zq4aD&iQH<Lz#_v8tmK!hfJ`A1Dx`kjJb?!Pq(wU4NvnJk>P20Y0EJyMprItp+i(Aj
z5%LJ-$fjhvZYlftF|QP(a8<ItVjRT}9u@f+yu8N!@BSE1o;*XTv8a)$fl_8noe+01
z=`phD@0EOkG4svW-(u9Tbv-iDV~_#NW<Vgu<zZA`{3$37mAUd#VQO?ZWuWK6y;}4%
z#0vuNC1VU9w_ev!fk%xTJrd&4gHk=E<Y9$1rhtc!?kT$%VVvEKRhkr?27=xU+uD!n
zy?W*L{k&2PpBe|gp`yp=4e#gXV}jxZN4C>PscR(WHCWEBYk75#x0Up~V2o)HNmYyx
zU(F>yo+A2Lga96Ti%p`SfFuxioemTyZj#2z34FT<+Wz<jJKQ?`qu0Yi_fA<agv=j5
zsl~4vs&8In*g>B>8{}6ytRquOq+6}Qb03k5*9u1&31INLgK>x!)}TC$2S+1q*elKx
z_CT$#w~Efyo8fiZePm4eTjdV8kmT5bE7<BktHeu}%GE>&7`$I9NBX6-OL*1^1DVYy
zZ?pU-&~Aeyc~1>SQraJ{lPr=sBy$g4sc_(`awl0ep12KNH&3IRX$#00@H+n{XK`I}
zz1{@DF4#f09c-qxBvS<hu_N15b!*fUjk$H(1WAgdIRVvm+o7tc8)1`5?gS%j-(YFf
zQP^7jjz`WT?Ix#?eUkz!*o;T3v_mg5m`uQ19n$o{|5iF_TQZu;8?AOoWGiPa9|x(D
z`!=?@uEeLoY6~G33+&WRVr$%uxz9cDwenwR6_@3xD{0_sdh4mFuB@*Y?D-anPty-j
zFkhg|KI8i-S(Bh^0yuY2mBTN5vjRiKr%AupB7VILfByD6lp%@>WtB-`2&gChdh;fV
zM`bK+4*uYb*Mzz<4Fn#QY}!Z}u_4y%ZE*S^3qJ$T+ZxobLQ#o{D^J+OA5&IEs7J)}
zt2Y`OC?j}tGqtb3{szOsjkny*)9P4L03RMITaQe00OOZvH8$Sz5joc~yngLA;BPY3
z<Lk-KpR<!G=hwVB^!V}V(N}yj-d$t=_y7IBkN*7KpHr9LF}Z_xgOwuCtC6O+2yX`)
zooRcWL7Rc<$Y*KD8x%GdzKJ6LZVDHAF+6^K&e|6~So8XohKYtEd$HVO$JYeDWpl<N
zRSX9`N+;}&dThPKFZ{S31M;0a<z2ZrbKQxaEDBP;(^r{`r#{E+7cZVd<K3edy#A|k
z>G6VHN1w7cjBV}l0{ULC7m8~@Y!}yOXlPz&;9zvTb#W-(oRd2*j?VAF(^tMS?6en#
zN7fNxJkbyC;L%8>(g7a+fs-M}M>`R+LH_l%VMjj<;Zt~ico+Gy=7Tq~{^_6oIeqso
z_{H(~Np1Lc@1C+Rug$Y(&(kMt&ln%zw(n@bzESWtfT1*K^;7G&uafK$!bW?|_nKK9
zq&ak(0IbI1I*8WLaD9&MuB`e6&fYPq<gDr@rMgUDyeIt&ZG9qNFJzaSQ63bRKzGnv
zh0^C^OquhhiYUoLMom^Y>df>;W}wACTBWU`c^dd^8$+f+bIxV49N4<<hx3gr9aG3-
zeU@QvA5(grU!5<*q|oK2N^<>FnHK&P(=k#p1usLvly;No3f)bLUV^;=uCVJ5hc79l
z=w3c{AhtESb*l74PP**4-cfdf{CkMXVdS}HW|7Z2<963qO7aGEHN7*w+?L?T4K(AL
z<5!XGAknZ6xE)AW(K!%n$Q+}zebUaCPEjNokgb=2T!o=~UHA71-!0L_-odzDkMu06
zis9^lL0$a@N)4ctfU)of+-4D4VanP7qms)hUJdKdA9Q1gyu&C`9)QFMk*ydt-YCH3
z>6%!kK%m(LwGMQ87_z^H{L3!|#bMCEQCWD@htYf;x8x#B-oD_;Td#(4c!8m!6kRXj
zpy^tKOy@9V;gpWvlqW2b)cX>TQeHt#D_R{P65u599MH1_qa2Zq{7UyYR95mO9U6cd
zVKMYj{^qHWUoov_+(R4&IgJ9z01cnyUF4}*bpQ+=7wb?)C6qkcVZ7AZ35`URyEv^=
z<4)t_B}Rzfr?H-R&FCL0_AKs={lix20DeG$zmWlbhIJXXO{QL?uVs+4^-4qao$?kb
zngPXGW2(zYvwtaySoM13a7rjPF2Mi*KmbWZK~x9h&Ns9S8jiyC8jwC)QCfhuqtUDG
zlmM4RbQtwZ#}elFl30-~xRK{$<mE8o3s=@Dj}zpGj|chUHvkiTf;XH|>Ghg+KITDk
zG$!Yp1?#ky$oSBrO%~N!b7~(XjMJf*J^`E*t%GTLaChacVHp#4HTJ%~aVXNu)hzA1
zE9;hNyBH*GXY#JZ!>cX(eGj~ybx+pcL-M8mv8Qu<P#Ns0qviQQ>uRL8IJazCvH2wL
z$fbet4(0Nkt*(7q)svuS<Pi$b*F>FqQQ<xu(>49XB0is#E$>sOVo=#TCY||B7|y2`
zWw5+saI{}C$rGig*F-wf8&NW)JSYK<JJ6u=H_fMVm7hxSHBb6kFA>85<)~3XXpe|%
z6`bRjCY9w&JPCR){QipumV7zYpy8zPWqH%}tmI7)c+RxOq>TKAie8>?pi5lJ%<^yl
z{4EL?Zw&Fzd9&ny{qsNLDf;T@|NO^)Jo=0;(mr+y0=Z=U!%&RMD@Ujvw?1jEooOSp
zUH=MGX))21G8#SK{pBxNv!S;{BT-)DS=u9`ArH~Zt6m-767nm^uV4EWUDiebbByfY
zuvbd;gKCF<`m}7)RYNp~-U+{9qfuqhJM$A?G5+z#pCXSWqR;h~xrW3wDvuws281@$
zI8z2rn|PL9&$es&BD7eChM;w-`pQ$5ZpngO3=9_$54l;F{lF<D+bJFr)(Tx<e4O0D
z`vaV`zT-*W`t1YH@#47V=^Pm-H`~42Kx1_FA}iB`A;-Di$~gSD_}RdV>vE8mPL==V
zV#Mcsaxu1RpO7k3_fE-bQ_`5Y-Y*2-gr@q$gfs0`hGVB%rY2=O<qve;G{m(;FP@N%
z&{t*Vj5)pn`y`h)s6QQQCW^94@s4z-=xg9DSFSJPSDGI)_U*X%^u-zyujn83cV=O5
zRjk<2ZqV621^+XxAEmzLx6ck9W3cUkQU#`RQfa+K$>!_07!~i{vDSlfPIss%qP#U8
zIFzHbmjUA7SVNyrP(B?C?>9!AW^uj7UEjQr!7+H%8yupk9_-wI^+C!FK_tg~4(Z@O
zK}Dq@B)#!quyIz#-55ecbyJV)7Caz@iZpb_x)peh(ncwrfk(8`(@M%pvW~?J4)3p4
zqIvz)x><N?&;ZZ~sSo(EZGB!D<>P&%OJf#&Qg12O;$^V0ttw<LJdR4KwUwm&YY6E9
zQGR+`#O)g*F`j%~b;vc!j=1gOw2@!(y@Q+{u?4)IAs5-Y&gA6ud1U7lMcUuZ*Fq7-
zB8Hz^$G2hD(_M{Q4~)i!XVWxL^w8*yvCN}KE=1)cfV`rt9I+-sZ#E|*9y|)qCZ}=A
z0P<oaGLA*?oOHe0qGtKI^%x=Wap5%0=`cjtOT~`@re9c}(1_8E%d+b)eNgXKJYrTs
z=r%}P*^gi>h9KunYUh6Fb80K36Aeo(N#lMPW3%_xIw+mcx%^y)S(mx$I`{96dUj%6
z5-ruzGO^=V)%3^B8%79B?%;DY$HLvx*_V`?M0Qkgli~~PHXa9-OM#@NJx69i4UR^;
z#(hfuwy8lTz=3xlU*oTgOI5Drx5%z4z6}47Q{K1QOKrW9L-I*`-+B%a!zEfqp-bt|
z(}utycy9`m`i639B_$4@y6sXbANxzZ0$ls_Y)|^)muEb2vo?fjg(yFi%o~)TPn0SW
z-#X~vE}X*n%Byu~6a+785L7O2RF+Z3rHGwxmZZ1TEq(WxuCVv#NsIKXE8*<5`Q;&j
zR%zCnpq_lZj8}Mn`I#-`*{=Mr|M?%aeq(rcoVL(I;F}&Q)G{DgT*`H=^I*466jP?7
zP)>R`VlZedP^ZE2dSNu4Ov_H4_l~pwgvQF3>`fwGW#sWs|MY*c9^vuPpa1;lq>HbN
zF&5^^zg`{Sjd$pF$e19n*6kS(WfTL3_V8nQ4_UAC-FJUE`r-@UIPv>0=yr_-dDUA3
zZVzcycHgs)h+ouI_8KGq@DJZ%Ae`dGC=bSuKm0ASD5C^pjkYydRvPb7Ul}2KL40GV
z@N39rjnJ3ExE=cG(f{}hyRQOMdyU-36Y@9)O6>sa+gB`nrj0`%{ABjbHpZ59swe#=
zwc9r<Qld~E_5=C_#^f73Iqi*l>3$%{$@McTKo>l->Llxb#Gu!}H8RhCWjlM<uOI^#
zr|P}ZSP>~`^W=UGPX@mQWc6)?7#HTff-Fo}x_@29xm{E1i5SpwXtqZ`EP5H-P37o6
zq$Ac_hfo1Zu^J)~FYJ(0XkD}<ZD%jZe~%HSW7ceqCcczjjkJb7^m=LZLknC3H+RIT
zPi}S4ps-`v(egi-8mFCAK&Ve@DX;)sE|8SV$)f}r=KaAtQdk%rWQ$d0G}0Flee6iu
zWp^b_OSjs`5PR2=k&1!02iB%2?*B3Psuq=(3i|cyS4a8$ISD>FyU@&a4~oyd7sRWQ
zGOv_Jedol%T_8OwQNAc@J5;6)JeyC5F=0n5@ZoOvNq4|7>Iy>v=J=WJ@S3wvvVIv>
z8vU@IuJ$}+^F;oL>6f<{La3#J^~smRZ`@ecG8!Bzuoxbs8C1AV5oB6}CwRRVPJ4Lm
zMrdfw+Dwb{F?j|Tn;OI^++$upb?W6SrefNAV7|C>592M+01>v01oI8HyK?tff`T3)
zPxjKP%)#+Mj|bVVIT630PM_Hx)!w#EDO9J>O!LEndNP#9JG8#}uG{$TyKhsA8@2Vi
z9|=nP8R5V2mSO4Facl9ehUlY5eC^k%IQYu?7ii0xA@KD=)l9iX)}`F-XU(hzstYLl
zxk+&uJj+`=Q+Pq_KYle>`aQ~U9ItGXr(y9H+1xwBFr!nMZ>aWm0qygwU9s&;2I3x-
zA)kC%*!n1A{El&Oc-=;!_--C(oXKElz6+0Stm#Cs5qE`=T)My!URiC!R(T7n1Gwrk
z9odEVnlz1zNN0#>4ZbU{K>gjOZ(a6xeZ8+8b`Z^D(}vMaxlP)RGPg_;=ZC|)p|!~Y
z_a5{aQ1;|?>G_-vxScRxjrf3C;jPqF;*QdP)TEN-W&Uc={xDo!hwHlh_n~4LP_IC;
zvwk{cKW&+>gj_ptmnTNI16PsgwpVHSv@3+l%g0ZZlV7CuEfZhejA9T2|5=dAKz<Q7
zx}FUT3Gio;sQF{ScMx^}Zbc{gDx^0YPUUE&Xh7s}9RP~K*JgdX)gz-q(F$?xh`W30
z*-&W<_X3ZOO6M6<4?Y3Ae}zVao`KXuIqP*(F`xE(!DrSzFwJ6`(+kQ)1+G^_WosGp
zl$B<TMui)p=^1#!G)y+k0#2HxqvbAKm+v&`kV^#O?^aOO@pe+bGV+i&y&CsWL{He;
zB#URG;7g{~scZ$eKHEw>uFJ5jZwmR18rNCKw~J`?sQl?qU$bW80bZk8^Pv%<(tpDa
zv6;ev55Lm;x4-?2(NSNceegg<UJHQb%kmmY@!VMdkkzp}!P=<zq;0TBB%U|;a5FpU
zg~86`EN=UcNyFG+4M>a-=yIWK7R%~!8-3#L8M~g6hrLHR)*SIC{2RV))ZaR?{WmhA
z?y5NTN}3dukyNtY5FC)3|1?XZq-`+=tFM&p^clu@llg8KAhnN5Lk$cIu91E<$}YVB
zJlblQBU^Bn(z;!@p#FFlHxb#gCS+;Ct<qZ;NMe0`GvTZ|&do~ifrg~QyP|Er%-yi3
zJjuDCgQsDt@X}#T-pX&d)tLe_k7B8v8dq?Cgv-HvNn#Kj{`M0e*cRG;u)}h$tLsD6
za~L(-P<v?Wfw<2O>M9-;_Z!^YdLC4~dHaEB6ZeQv!MRzKuV^~3_YRz4kDpSWR_}1d
z9|nbs=~O&=U&_F!e4p>C2l7i<$5~%7EoD|llIeL1WGG?l_X%8uqJf~Y_e}_u&<P$J
zjS;5~TyQGwZ-DhxWR=om7S?+2sO0jxXn1jYsTQ;9J@^cztw+TDLo_&?iqHr%?(~P#
z4jOOTNz(F&EyJ^GCtjjna}w^XXoT<}ALFq8gMuA2Wt1iX#jK)@rvU>oirM{4;LSr>
zw6T-DMO5-*-N-TXW#<?nuBkA%k(fs3oxfq&!2k1?OvTXv8j*S^&ZhlDY-8I#yUHTZ
z3@)_vJ-krLSfXrew^KKqb9Nn&dxnJY!Gn2eWBGugEXYNWWOINaa`9Y7i-v3rA?Q}-
z%HP+>m7{$?<57NX2Mw#xLWTC@moNS3-m9918M2w>E{@Q1Xj=sOhCfxsRLBJTyb|Ph
z%Fp!#`Cicp?~U<v*PL-AtrBc$=}yvsZQ2{+#&!O|3d9H*{Rh%#NpZDx=}j+pTZsp=
zTLZ?y;5Pu29*&8uZp22izV7rsd&3(RxPeZZt8+{py$enCZOU%b5zx`Gy4WA}oI%D8
zuo`4~@c9&c?W*fi+KRr9LD!`^M<q3AZ)#MDsl+qyj<>-Dsw!~%s=I+Y9dmx}TL-)y
zf&z0Wa^7DBn5hHgtpQ+Mw(3BkpiJzuDgoC3r0=T>?UOm^$tW_TZGcE~9~wc@nzv@a
z;i5<DR{3kxywxaC0b?jgi^@tvO3#7U_svJS`c0emAhh!G?4gmNQggaR55Z$R0~$Gx
z9z5cjd~oP>`TR52naCeX5yclp2%P7ZbNJK%n0cT~@=5a<!`n=LuL?>eI0huri!xRb
zzd5eGPV`3Um2w)*P3JaiVCs2I8nLRN-KmyyU~0Emzk&=YB`@0SJ_cNjkZLL9xfuxD
z989I(hy43mtjb>+U8|Al6=W@AZ=gvS-(oSG+|x8|ZaKSvNZZA?J+MZUR~__|^c$~B
ze!c+{kDOjF+Q})eOif9TMlp;@hkV=b<iY8q$OD<gpr+kt&(vDx-EMte=ilbK_-&py
z7?gRXZs(c)p~6PiH^*pMq@RLwyqmA5gc_#PM}eDH_y;U8)`Zw%n^xQ3fZK4t4Zj;Y
zQ~z|ibe1hLxv2l}xHR@I;s-AAB-+>9h$A|SX{w&w&D}84tHg)A5qFV%Rhk3QZNqhR
zUzdN|dUuuESJ%GuT{zw+?Yer>(zO3t-d?NBV}uM>wz$sD_{=-XOhv7``-Ja;yn5vt
zfbuZ7zs9j150wC$M$PrB{0&iRD3&OL^iB+A`l(OmcT_|!5yV)Dv4KGn?+eOB#h*7H
zFoGVPJv=(2?o#37R|RGY=g%>moKASfreRLmxacy|FuoB2U)fwscRqD$$n<&&4WC?N
zOgwQBCTI7!&;#PzF!kCgubZOqRK(IK3mQ4?9a1*lQ#~Cep9W<Sxk6WzHrFI1FY=fE
zaU_u+IV!Fy#UjCjB6b00?YU7(*LunbDTCttyf!*3h3xloUeRWHhYFV;DU3G?5;Xjz
z*|yai6!T~pgM-9Dztd@w?eti$1#&t>L!s951iSQcB!E9_d$`DgVyZHPv+!Hj^=l?$
z*6@Z{w=$7$=jil;Xh;${`jQRvT%`Kt=U>Gmls9rPvfQ3MGSWrGoSv^;Ym{k(&eJf}
zzob*1dJ1b0H%RLChzvu-$hg5NZ7V#eXY4gHZ&I#9-|e8}44Z54BAZGpeAnR$@HIcu
zr))ZSkLI=Ds%l{-Rd2vT-hA2R|1jr?=j&*$$Fc}&1|95k+7iUp#SDKy{C&#K=qEhY
z0q0EO=5uWF915x<aEoN1-jtY-C9z7)@(aWoY=dA_CjPBN2p9?7*;d(qs7rXs!b^#6
zzKpJ}z5EbizG>I^uBy|#EEt_fmNkMauG3Tw*1}ADu8XYmo8C4M0uLVXsW?>JelGr{
zZ&zRl)VD}#ptp>NQy>~4DnXT+Z#>*l0gS$zI=PJJ!*i6fo)eU)?ZaVRKApRRNYRfe
zf?{t)p#h@eCnXD1kJv0TYfDlKMS%5i&57m3>EJK@D(2@bO4Oiw%9nKY7^sko-ZBW}
z&C!!0lvTN7=rFzF0!xhz`SWR8Plz58;qz)L^@K*QvtGnA@Ya1Xl%Dymh&*?k3ipP<
zMTP@JaVEeUfAx)(G6uyjzw#xI1{H2AUzt+Az8Rxo^zb3C%kp}xvLU(X=MAHp=FrPf
zuN3>{5z@#Zeh;VRUca`SMx^ZzZgA=`((9BBvXqI3hRq1VhXobcCNZ+;Yl={Z%C+j0
zHp|&AR8i>w*dApl*#|DNIM-{}@ua<IC(BB!ea>}FuQ5cFjcwlYm!D4=imSO!;?t$=
znlPp1dI%_Y<rE*$g@wg4fmP7)CG$3skhdX51EleWmO<|ZqALOJa$;xx?$0+NeHbCd
z({)T=mw%OHs&m|gw5za@w?Y;CrmfP0Wu$L6J~`b3*JT1fx-)lT?&l!d{8SNs6-r%<
zs-}w#_}@)z-=MQE{WgqWhwHlh+thiJGO2&Z&CqmZfF?i20f-&sRaVdP%C1k-FQOQn
zj_E$=!2ghaAq)=k2JvM&1qCS}YJ@iph|?!skD#%k+uU_W!r7fQa8#&mtcb^8Zx-({
zTa>520xa4*Y5T<OpO0hU0$X6pr~u9<d!`pO7+ja&REg6}uW-+|GSa{hJ_;4ZTJ-Ww
zoVfUFiblGe@)0&`LBdy&BQY9G9x;ZXqeIETH;C#<+d7T+^#-y^C<kHsh+s?{fh^gK
zoML=Xr{{QL^iJWuaZQZ)O&5>rA(UTz0hcuuV;_x7$H4DfzE300XTLm+oSf#+pvddL
zw2j}TQGUwEDG`m2H*eVf{$9OhqLHT&V17@FXe8RM;?>LIRGz(nn&?y&3$M@#k=eIH
z51I5v{u9s5yJzL;TT5clD0OYjgGV(z<~L(Zb63{%p(9_vzTnNM7uDX@8~lYwTpn}i
z{1858yiZ&iP<~U>CMMby372XbB$Hl3@p?#U13L!#wMWECw`<!Jxjt4Um9-%{h@@>}
zgm$Kez!X?2ypE^q>;CN^ZX&OpujuX*y2I1$n}f19q3jCzNk_L;z#4C==!+SFX`>X;
z>X0^h!4|g5<@)AYZmJojWpe{=B_?f^^2fcDzrK5aqdUS@v8~fOGYwO`TW)tdeOU3Q
zPfT?O`?xMsVN;(9%s%21waV++v!`UfJ9@!)RAlP&C!f<7OM%?uE48khQ1Q9Ikx0sB
zssKECEHZFvd@#*&k)2aryjTlH(d@}TKK3mV=KU!ztqyL>rz@~+?DeI1NB;$POb~e(
zZJqTbmc{rG#{<5dV&3_~QvPpvT9!`vG6>UN78SLJQwu6F6>sW`p)+2AMbWx@tj0!o
z1xG#g>pdxNLf_w!cU2O3NQWK?r-M$E(Huy(518+{j`AK9zQ*cV!%gFZKF;BEm38Iy
zY0?cE5sn&ZKRf__o2I-%8hVT#fc^T+@9Hp}ra`9Hg+%e@J3{cH!PbU_`;Ta7*!CXo
z9_zG|Z=Jm6$UY$`|94;baXSqZwzTIHqft@d)vIFrowJ#oT-68mxb}NB(B_5~@s<qU
zmx^Sj17Iq``|;$kU&=6irAvcWBT>F>Hw{e?#=B;p%~TX^s+`hZp5fK&9Mqwgh(x4T
zLF$(<;FjwfL~MtCqh%@9m5^Y5!j!ha5m9dPVlbZA7IDd{sS={fh@M2GG4DZK#rFTE
z7&vSbp)&8n*^7zf#VE-jLZ${${afmf8Iau}p;lX618D~<iH!3#@V}Srrn_&FrO%Dd
zxSLKp-G1R!&Sg0)bGIv``!THmKf>k0?=5g&OZR~t>=spT6$uZcF5^wtVjKzmiD=m(
zY*nv)Ol2c28HfjPlp+e#MVzjwaDaZw*4HX%zcg!C^l%Nu=TCV01U5UZ;?c0|$xv#&
zUr6?XKnlJ&p#nI<c(MF3<@kxQV*kCvHt*7;A#uS*XXj@q29#}-<iHst1K3R2m<9bB
z27b2QvmOhT4a9|;6Y5ayze7Puo8`hE_4_7`Qz?1f6GfyEqA}un5OEk6Hg9~Ww8c&N
zR${?JxqLMi_!D3|t&AGVlW$u00WLz9X~uFkZ5(~QK3hm-?Noz&s@ya#tjBdLzB=ox
z!WUoHZmOU2n(@!S{G2rr8V*j;Jp1{_OtHE3eH$OgOwZhdFLB1h03CiwS=!v>t9eDz
zX$N<P{rof62l18=Mw$IVqve~gzFGE2ao>^q$j~-@6GIeU;{`#+V9uXApORKnYosUw
zne^AH_)}^&<vaS7<$Uc|UTquug|8Y1HhrYZiu(+s*VnjxHP=O{$}))H5r)ERuB*26
z5BCP?khTgzQ@eHMr#Q=WLK&<rn>v%WA$8-OXak^#e=jPPymF0f#T4U45f)vM=S%~H
zRa;d#>1+<(i;&-O;e(8j4gG&yaWv(d_U;t9o8m^7j=SM}69eGN{0(H_(u*eiQ}#!V
zTJMJlJ9vFxZnnycf6yw$o!94?g<WaMALY#tj)~^J*}IMz--n3LG(}-kcYi;_y7N?y
za&wDopQtrP-lF{8-XpGJmI!`SHY#kNK=TyZ3l`@J01UiEz^5T2>zt@FLst~FLR86y
zcNVrPRmW9-vRIL+gbNm(@E$}dA+SobMn!hHByQO#D9(lySMdm^kwL{QfhYw`U70q*
z;1O_<IpxHul8b9!I)h0^Pbr1u%ZrtW6Pc*oq$fOD*DyZhy$|USE@#WmK|CCW`mM~w
z$B!qC8hs0XD@AzIG$8b1w0A_su5ogPQRS4G%GfnC&!7G+Wx^|D+(n+ov!@3}o^j2D
zru_^<p%juzU%B(Pi+u5?@v7&;H-}uSp?5-K#C{+>$~M6_Y?`;wTYR=r5*`MBRbca*
z2XOMF$vVMG+t34;dXZLZsNX7MzTN=aO~X>-Q8^d_Yni27m6?Af!y}Xf_ge&Ojp<`w
zsQ=JZw3ti7O(}!gWJ|lD9T#<Vo8MWgU0Y)x_=?a}FSK>m{aX3n8%@Wm<W7Qq8)@$(
z<)B2qr&UimGxTL9E~m`!zx((YV7dVv01R2B%z5r6J@5s<JZ_*UmrX*)w$5>@Z>4wv
z=G1jN`L$cQ(FV(OLpN4w+!cz;T{+jM1*dpdxVx^dF=t@<dd950Yhd)QVmMG~5X8~=
zEf+Y`#~>HwBH#!+aj)oV*mwuL$iM?;;cWjs@XgPf&&Idpcfx$O4g<L5)2C_UdEzBm
zC1c&*W8(3nUp`Ix5i})skx#TM3&4Q&Ed^oSU&A#6uJ3p)PD0*1VLy>$)*e`EV40TD
zTR{f}feMB^>olLkIyEXXHGnJM*Gu)dsPwWAgvt`QC`t>VaAYYS2QHn`vF~^P7hen$
zpUeR~<{KlH_mGM7SK|a_&~N0ZL^ZHx*x;m0_m3gTDy+=R<B0r7?K;Af^uu@RAg|k2
z&ZRvDA2{?X$y=Yp7d#3lZGH&O+tZH<Iu%h~o5E3RWsb5rnQxM~tE#(+e#4tWE>8XK
zyKh)%%qQ*nNTJ5%vu7`(tnC9`XFie6OFq*1_%5$3Q=faL=vn#VvoF%ldO)&Y$o=|Y
zokoGvFqX+Tg%HdeJV&+*jD)}1FNB0>tD^UU-CN^D0!(RHKAtN31fcd87v0tr8u`fY
z9_v=x%knF0K0r=iK_<voqsX52j&I61?Q@sDhj~2e1R+Rm3(pxdsnxS}rbBaWe=u}g
zre}_{WZdA#k)y}$6ibKLfuO3~WLV!wYa`IiUiYD__JzPJOO2KNVEc+EJ^i~0xDxVm
zreq(}CWaQ9ZDNS*q3Nzq0)$S_g+JtFEC`Z@8E_k90dGCnWuwbzTl>%wx(o@SfNRh4
zw=<0Voq4OoonO&i!Q`^rpwGKtu1niP?sdhMeCuu;q2oYey37_LE_SQV9fF-3V?pu>
zA~x{c$m6C@%^ml=E9^_(1-mP4lwV`tx}NKp<Be~kz1tw&T3BSdbaAikv;&8AY}M2F
z=JSMmNv3?t4m{#pZW7aWl{uuYrGm?m=F1$Fj{|TN2-m5bIx}FHPrRpFR0w6bI>40z
z_iY%ZlD$D>1o$=x7;b{c{Upc;o|c!_84@5ShKO}anEWFVzk+Ir0!QvD3KcJnKqCLZ
zTi1|Y=x*9munL!KiMuJ5Tfnz+nU#@vtk*PI$du2h)3n+65Ut*z1j<RJJOv-;#${g#
z&Vr15#UuW$F!)G0@i}ebVSG6#Q6!ea#6ZgQ2?D#r<L3f5xYfwZdI{=NflJl{yjfU-
zOwaJz@z3t@g2{)+$^4gWI_IL+w-^pqWS`K$aS^f$PBm!0`s(YWufO~{28rLFX(Qww
zAE?tiVbxF!EyF`Ij6>;GM&blB9vS^O)#k~QR|+o*HZlZ&RQW2ehK3{UC%=d_268}B
z<KP++m|C)X)HEPb>Qd%ubK%8JuzqnH(r$^>&zv6U0XKmh9)*?=cX)}=Rup!j?lVPu
zr|v%wI8TQbxr7YaT*r5bZ=2MSmi~AN=ajIJI1!j=XEburJ1&r(wyVC9N_Ws_>A5{K
z)^t0Qm-}=_xu&e!zOS!3pJY(D3S!QKvJFdf#sTqL7u?_uiXmI$4z%f3BdsBKrFGta
ziSrmBR$}>WSsye>gq=GbeAS8PPXF5yd-K|rwNH@V;Kt|f-tAzTPwBN)dVR*;r|;7{
zEJh?VujWoNNx8npzOj4;B_dAbn6Jmgape@RNWoX-e%GWFx_EW{<2^u0S&?voR?+Z}
zS(21YsX)J8hFT|3?X+&Mw38Za1Zgr&<12ZieM8TXN5kgn82EZht8(&{0Bqvqw^yb3
zEw~2rTs#b!aNRUS{!9IGDGzCuaTZryLj=Nc>PNXC$0%`e7K|*G_elBRF`=!b?*T@}
z7hgO^u6{+g7N)*ny^n3-&axVs8YHlhc&2#pfVgJHg{XJ%M#b}joL$hX_rX`R^}bkl
z7MODOTQOPdfyc~r*RB+wQlo()-yWvR)8I8c7n2&gF=R1Zd~>F}J}e0472%_IFOSad
z`6+>FOAU=vX~Sc7#`lC=F#FP1bMgFSIwpn{RVzS)m9Bz0otnPj^$xgIvA^TXYMW-T
z+OZ|Hkxklth&2M71=)oUJ=+(ArJXi4(PLtOm_aK4=cB%5J>vx2Z71}{4hOFZvEOlP
zF8%LTguQoXY^fn%1vW0PqeZX-K`)`YZMjQ`K1M<h1{3b4cj`}5?x%@14c69lm^P<>
z8cIH4Ewn}WvrUY5Umw>!U^^iDbg2hG_xUm>PlwK51eKpfc5-2!LrVV<Nl+2qVKljL
z-)@5^>Yxh9;7ocvFg`t#o^0Xt07xp)3ymj}JTxSdM}8X>$>qlaBN%&UIq+s`MxKH1
zr{#6oo|=-jg$8kqW1!8cSn#)kb!}=M@s_I~;9SC>G4SwV1x`9;?Bd*mYwD0@0i;PI
zFL(y7L8m;+OV+MXiH5`lQzoaUO3VFSYO8xUVbeoln|i)LVQWD6Ds^PSdJ-3c>Qy;6
z-6;;_mvtE^em6~%Hl?G<Q8ZPXX*(JuFC<19b0Wki6n}~!2WiSdd|5vPqGR_hK`wgY
z97M%ZQ*(OSpy8wnXB`Y49^u}ekG(-8gyP~{he1tMY-SO<2Kz5}t%OU;#rhnN@J;Tf
zB&uw4Ii(45(@^!;>o-9gBthKv>$FiPE@TX#_djhgUyY02e<-K|lJKh$?#oS6+@P*q
zc%(G)<40OLB+g}@JAFClj@-a6>4!w#ljcLJ*ay=C<~qMQ$$6r0hBwc_^R{6#4RCV|
zaotfYZ8eHJz=x)(Et=k2cB!3&iPK}Ln|k-*U+4kTLd$y$p1sLep5nzLE#S`d4A)tk
zfbrpXWL}^0Ee_tYaE-=;2hu#INjRTVE?x=RPlXnRBp>47I>rQe<m+6og>|49lkXEa
z*D4)rcP0kY0KMb4M|fj~H)S+3PTujv>*iX<Rn~-K7b#NbDfLS`nU!vyF?k}itQ^F$
zfD|~>gqBxb{5g}hcs`_`d}%bOfQ`F`B5A~}+g)OFJ@}3RWjSz#2I=xJO`HQ?5-Krp
zfe2qv1<nM|h*N(96fS&-Lq$Ef12JI2;W@}Q=qTSiEE4UvK9sHYS1EB|$Uk8nhvjnL
zfX}y>gwq)5S4*Spp&_h+Uzy2D?hG`9>pB3=8jLm?ye@g$C&&wJ^l#xL*z&`GmA%5b
z;Y~xx9e90>ScB;F*d1#5RuO&g=%ud$*Y{$+V&_z+0rQp&c>VITYnT4^x4#`d<<;`n
zZ(oPDQ@(lBq3?5<urG95ME0afo<tV?UFx>yV7qPM->w{Fw_Ii-It{5+GD3@P<#I?a
zIm`6dT=>v<`hQv@B}#W&L8Wznu8suAO1aZTZy&n-55S<^dst9NLLOY;GTIf7mjSLz
zkp|oBAR1<%F@Rl%|BsMAI#5WR<G&-?XUUISY(w2lbH^R1fapOj<vVXv^03&ha-G&^
zG0pMAFkcTfaLRM|R!gkXO!sXApET=j1B?Y=9BeXJGjOPk+oRCsXPHV!@Owt1T{zfS
z7l)^ug-G&HhQP%&FASqg4T7p(RI+t1f0UPIj}cN=5LZGyL5;Yl6?}*gK6ndm`E5L&
zIW){di?GlbLj>R+q%9{Up+=Z>UBU{~p)_&T4Q$L0(xiDQyCCdByU!{J=_p!;gruRU
z$yn5Fc1jxrei&DGXk+D#AEeSXYHZgi255$cVB)MObTSl|!BPYj)1x+7f)BY(Iwz2X
zwj`MdMou1r8EQ?8>SXXT%0}*_xqhV<umaL~m}DA<8XlhATy0kv;9FUv+O4$zzStuH
zg@q7v;%6;8<vT!vpyO+}J=l|+zF^Ek-VB~OR+l6icC~2b+~)UCY$rYNUsp*nQ=R-e
z*q!-dT-M(sP`n3#ND(e;V42cIZ+ZagqT+{&<UQ#7$VJ1!JeZF9B7>>!9&F!-<Mazh
z#}5#;=#zeMHf<eW*EB1-gGsQHO7n8(869aNX?{pcy%XHQbzA`$g=IB))K!_R_@sUZ
zf49DNb1Pr>bb~ULK)tczXVWFqCu`nL;35V~G?Wx?LU~Ada{?veKAI<YQF?%2asWg@
z4Gm&vypD?&NkL^%M~TS&K71-ZJ?71op&SKlxJ6#1iK|2UhPWVb_ISi=0e`;b)ci#L
zz*=6pWVt5gG{AgH_lewcK4ITsszCV24`rURqtei1T%mv`Tq<Xd&_;UtjAw92m!7G3
zn2?KUb<KswK;9I>ok3w+%C`{OILq4tT!`McN|OIpTm@_Uyv3_f1*(mYd3(eFRN$H8
z3SW2dJPE&eu!!d^k3tq$@ykzmgC1!&zTnE59nvYIQSla<E}VjM-4B0Dpq%X}{<H=8
z=N`z~#j_d|zImlA@*Noah4q2|)K|pu%DH)*aNqy%L$>Gt`RCdnML6m4&;JhcW|;d#
zrv*%!15j`Q{;&Ll7`ECBSwv1Hq{LfNIk)^GGfH;mEJM=PIL+3pmG&C2-YMx-!yC}L
ze?8WA`=xWt^?aoGjvZ6?*MvOYFT8L0x%R!dc4h2Jo00vF<GX~uEA4m8!WL;W?H$j6
zI|ed<(^{RkRrr95UFA&k?w#}I_yK4S0rBQ_jL2D@raF>Np1dU=4Hre6y&o#QydQc&
zN`andoj^VPiYjuM2uW=Nan(FSbcTwHI$aww>O4PV>u@$Ri&8^(Iwcg<sln23WKd5(
zIQ7UUMuU7=Q$8JE2{IMFawbhh8a{<p_SCInl;@UD(ZkSay5@<q5o-QXHmLnEOtS3w
z;IQ)hH-Z{+FdpNI^O9b)$3rs87<ck2Oj$H;ih^^zR({!EMJZWnKpSTvskfF2g((;8
zV(QyY8e<w;2Kh|9E|b<jsdl7WbE;XneIo*h71pDEHs>YXJnD*^sawOP2sho#qQ(3U
z_67g+jJOUYU(4+UskB*gTb@(7E+2aOg7XIRl=$D#Z3)amGrkg<;x;MmZpmD5;=BV~
zi0vSdhk>)*X~8*gb)BorD)WhcPx8vnA(hEQI5`A(4-7Z+dj$nLq}2m_pMy6v3SiI`
zn}ixoEAi&P*7+NT*-8zsOOLce6>)uDlesT_4fipw=em2aGi^S)-iB>a=REP*aEF{N
z8+?{`AohAk!KF^&q)V6qzXSZ*q$rqC-X}?eix90p{5x`txQf_0dI$QqY{7lT)9T%L
z0(e^@^xzqA0VHLz^Csoo;7THUEk*730B)6fGe{M2(OZvfSs}s2t8d-!0d3k(#joG6
z?*nT<^n^Tmh&P4!U3bvb0La@hXgd_RIDC3iHaVcrDFxGob-xKB@$k5hhP*b<#`~#0
zZ&*O<D?A^5lHQ#-BeU$40%{g^TJA0|&Q&`tV;%F6c2Yis)MH)aNaFZWp-PjovYofj
z8XU-y{2Ya>Y?ahjbxKPG`6-^fR$fvNuZ<`))lyw~LQ;;Ocs?kXX%CD5*u)s9eM|I8
zoFRkY1r`>Fi_{;`(RkM3hH|k(Mg!X)TBeo_oNq;?v8#-T%!`3XAG4p>$6TXg`d~3~
ziEP+$_vAiCC~Zn9)_**sjb0sn|NVdF%^yEns5j1iLZr#@D6~o_K^)T5VR^bHq0=pA
z+r3Yy;+uNj2iGSp3)>lMAid>pOGP%MMSdYD7o>ZUpPbB3A4sY)6b?)ck_Z>$x=L5R
zo~^lknhs$M(qOTqk89KWda1<38-@kP0p)5{+W~-(25|!3zB8>iMLDS2HPDXg$L&{;
zJMxq9o;Sgoj8Aqw(>=v^vN`2{GWvE^Bprj6#fX~Ic9r=UX)TYwW7l;|G_A94bL~@n
zj|E%pWBn@{Gzc_SdM8S^Ij=!GDnIh`MlL*J&2SO#MQRzmseh`3GUs0dAxf(#ahOR=
zMTjQ6@Vg^;LC(+bQa36TrRsE$*Qa=8YTpPjf?uP=kSA%e$OGx`#+w{IQ)C?ne7mt(
zyd1&<S6&^a42Dc%grE@p5S$9wMUsXSjW2nT$nnH(G#(Mp(3YokV1T+b`Uq=CnTiHx
z=<}o?l6G#8$Dg=FHW%j7v<#k}$;%);V4~-RQN95jHKxocEXv5pYcIryAptRWH2_nu
zMx4f<+57=7+j>ndgbmv2N7C9b;?hV6evu1{>!^4!Ze@itu(roAdS*|EI4zUuD2y@p
zK~Y)7z@d!OG=8kkMbocYKrN1-4$ttkFUU7Rw&zf(bIO0Mc;pSswxjV5UiYP<RZeMp
zJ8te@6Gqx7<|*>DsAY0NVu$s0r_bjF35iSmkb*Ok4;w$plggnuCA!3LXo(4KyAh_0
zTSYSvEG2Cf|GUYTj70XOZ$th5uC{rc=(7hB*7=n>!Tm;24Ug(74y+j->yDA`ac#cz
zd#-`?!w?&s)-&$V>uxXWH!QNebZ1xoA-@^O2d?mkrtzE&pEx(4o_$5vujxAIzkc~T
z^S6F?#JrgV;~qnFC}o~V8Kk&zOL35y{!K+ukUmZWKbAvcbZ8t5g@iZ6w=RI>Xq<eL
z<m^peAAP_Z9xA1?hqYtpDeFuIXmDmAwWu^ekk9@X;}z9jb0A)W(+<i*!$cZ>dHO8-
zqo~~cFq(X7{Cvp{q<Nc$Jxj7d7G)zJU{aMb)T6<n5cH7Dp^X*OmNpzPi#KI>L^KE#
zj~lJUP*cGY>?2i{8kIGF(olgjyg1z=jBk+KfmXjeW8HV}mHh8$M2ZWX7VYwdmxK~|
z(z6EWKI?xXBiEiFD-6Ik;?CW_#dZRxNZl_)eD1Jo9g2b*3uCg-meDz=6&G~bzx0AQ
zz2ae7^%IStJJ~qR_I${9avt1!0DSGH`twhJJJS2{^y$;AU9#<p?|Li7{bvR%RUyCT
zOIPzuLh{ydOMzQOtl^Is>^mOYv>!X|Nm9$K+pt=A1DpQSWoHpA{c{5we8F02!9A7f
zx;7<7w;V85Gl9=wH;z}oKdcyU$25lr6u1HCPHZ~_w-fHb#LKk5FM$VE{2PMp<Z#`?
ze?$6Gwe<8rw4=@Cp%3B1lkfxfCLK8SO+UdkU3YX}dgp(T^Wa%EDG$v@4F=raEep>a
zy6WA^y*wf+fttQid8;&X8212YOd1EY;P)yk8De~t5UXC}!cdHdD(O(&6wVk0D7adu
z;NZwd*G~DYJKKxna$u1wij}Dv4Ff~-U|f9ObL#l8O5}L#{Oaq!`8b_`)<qrJDx9eY
z>-6c`4Y1CEH_uPSJJpmYai7LH;)StZDpF?BuA-Ly`f4rpS#c`grY+UZKfsin3f_2m
z55Pq<@{pksKW3q4qZT$Ymxj>}r@qpR@pZ0Y2Ifqyzz3#@D8Oeb=cX;!;6$d@Wn10x
z8#VBvoc+rtX*wEMKcanPn!MlDz(_vVA`MRC)!qeZV^qGUa3Z}#l$|opqGrxjzlf-s
zHp@B{WN#rCr0NyXz_I;YH2s#1%q%OOT0~31R1g$5`l(uPUHnFM{Vst=f4D@OO^1BT
z7@>vi7fA!Mg=V33JYe)G(=A`82-tS^uY>XQsZQ{0HiZI3`3>|)n_)juoavaEjH_y8
z@JaS8m8;;l^XgXEPP@(3JSawNA!kL?uC&`!_IoR{r$eu}XKbpJD8JtU4+}j_i}cOB
z7#}xOgKXt5i8qg2s(YYF`b@*kQC}*NX$4bJq~h1Jp)z~T6RG_5o#t8Zh~FT2!H$&f
zy6Jt`u7301^5s(3SonIR#z>xuc|wlo5^<>v{Wb~aM9O6)y!^yK@Lnxn2+z?m0(_#?
z*l`!kGjRJ|8rP*5{5FXDMLc-@pw_WGM!CX+yL-yVBEz(i-heU&G#c2zJfXf#Qd2B|
z<<^vAcM^32uNORB`(*C6<?asp@Uw^6=<CU6PpU4a1Z;E5$gdyyQ(#^fTNp+}q|2~H
zlK@lbyc)06`Q&qGJOw8aX)MnJ^$m|hG7F>(M7;^*snAc^UR@8y!*`EE!|^+K1?>gj
z^2weqKF(Q<Ycx__$w+#=5AU*J%Nbj}TPD*ncvfO0P{%O_fZi%Y3<%qlw`*7&6GP=L
z@yG$WOaB;%*~g@enqw9e%THktR~dwUjk%3k`;GOU@<!CD>p<{eDX04n*(x3ypa1#_
zuiQ&KR=?z}E#)sQ_JPo3ACz{Pn}T^ATJB3)!%i>_`(Nd(5zumxXGM{=v>_EG&8!N}
zuD4vyu)=p(WJ*FLKANHM6;AP1W|gwcTrR1U3*3UZ$Pnix*oOJYXYc$QRB;p78=$>4
z=%DBKM%JzQ*<bL+ET(s+yY5GNRgUF9MR1_9u0C654_=?5{tsBk(niqVicj9wi}|cl
z=|)+gm{rbw^dNuklc@^OCsGtZnnN)SdNVD-=*995#xX4qGB}KDf;hFI^2{J;WgOyB
z@x4V!X&7W7DOrAw9)hzrKzQBvngiz-uGer)h6t*e95LPTQ;mKX??DDSVe?kSm^Oe|
z>2{+oUqkk_PhUy4{`<K8Rfc}JO$FzUqbflayg{RFvxdPzwCY_3NM!3kExSpJV7S=P
zXEGf_oe|BF&y{WmrWG8F#i61OAJ*+Phf_v5;KQj6Wf?j#mh{>j=j|fav~VpRh~@}Y
z62noUX!XgpjKvt5dP0;dW%QglW!Zn05++6|MuzEk`B0$fqp?DBm8&xJqfxEbkUn_X
z@FaO2Sx}7Z@30O>al}+5ZY-W1_{=N17^n$$D`}B65gQgc*5&fs=?%AJR74WrHw+5i
z7tQaayb|$%t7?oGBWbIJLVMPLS<;qLNx7Um0UA|k6Hv>r`fR~=T3vnm-<s$<wBGJ$
zhr&;Ax}A=>5kE#y#@>MYLfUwL_P%1jpY)a7_gTy9^xv{`Ny{>D4M}D|o_Ogd`7P=E
zP>md3RHX2Cic(bp#%sZ#oM{Em(*UhPcEI;rBd!Z^Y9ikhK{+X+H`ypkLxK5x6(IS}
z8)rvLh*oJwVakAO2q=_AkUTBt$sA?lQ@I|KcjW8tb@6DtA#Yx%PF?Tf@T;xakcx)5
zf6ls)`|KIv^au*djnIk@D=M$VInxT>%jZ1VtH8771cO7Ps@5Qky-?um>94=0PFxG%
zxqI}PUytQuZr^?P=X_Pxw`=SdnjrBS;T?MClYOp9qudF{G1DQ+SR<{T>Rn7*bw>8&
zby%(yWFHwJN^K{fA3Y$pgGN~lj)70SdZJwSp|?mcj~}0NJ;*uj;hRhM?(q>o-Wbsv
zbl0)X_USPQ+@VurCkLJoILwFtG(I@oJ@uV(j7;HHqCyyDT`!=4=@#o`y+zh`9coW|
z>7~$s(Lm51bbW|rZA<^`FV!x?&>}6yBz@){M$<hO9BWW1N8=)KV=@gzy&<0E1xf>_
zw=Lcx$jJxxTF}#ZkS@8GgKhBMAq>7ZmQRjg#d^A98mlNt`w_Jt#0n8$9w(;wp|slv
z7964r?oqlM$m&S)av7J_b**SWwEmmIorFM2%{S0ZksJ!Y)NyEm32tXU9zybTrM%X{
zZM-{?PO=zqEu$`Wlk>Xlm2_T0jKuaaw}q&BR`;)<8Bed_W(#&y0<g4&KH_N`;0lym
zf&ngXwyevpuD$fI?qqeoG9U5{9tOP5k^(z^X4g4e*3kKVU8(ikd2?Fgt2XQ#DmW(G
zYn(Ga;~v7yhTI&tuH~9K=koHIvM4lz%GAY=ep^KYWGOdr2rHzD!na18j!_|;pb*j`
zDg_jH=H*GTd=#(lc!K5i7PQg{^%6vO=I{p$r_gDsP#onnj0K+b9jHv-Je|V_{YHfY
z)Y02JC_CHDX`t`{CMrLU7r|(G5#+<^oqpR#4@((4O#5KysNhovG^%`Sac3#;^o>EU
z2Ei;F;|W8C#Rn<2D@PH7GDgmGd$`E6{z?falfs~p_#8QrK}(VD=4zg8D~$xFN|25F
zrf2|AEXE+-sb7y?#2|dcG{$K*C>v8mS-2_=@hiu|myZTg-VhHDb-Ets9n(26qTso%
zi{kJEZ1ZE?((Scd%1DfH`;R56=W#(WUAbhTD|Of|;fWlL4vi;)d^79Vw~{o3G%)zH
z&EgqTuGTX`_p^9Js)EiE(I!5}BIr3`#P`9Iy$92Tq}nD4D?BMX(k|gKUAp(tG}W*-
z>R32ug3@}F{ajy<Ju|2!ZG(cFp=T)LA(C@p?%X2VG<*Lj4`|nw`-;w>6RsAw(kB4(
zT4?);m`6!so3JZwle--g#~gxF;yhOEAza{Rz85dyjHTlG5=RM7D`A%eE|<DZN~pR`
z!6L8sOSlaNWz>z9^3uZLlVZ+;wmeBDVN*+nhTQ#|v_oUnMQaeVOsZMQ)DD`wRk~^P
zeFh6V-*#{inL~G`nMbjzkVG54IIB>E@o<3RCml$oDIGxjs1ww#ze~UA?z_-ewbkI(
zF!Dyab@>B62k|IQl^A6`LZ5YyBlV9w^BpJ+4$cPCw!-SV^ZP8Ue5uSj%rY@;fgQNK
zPA#4}ih?d(=cUYCeigT#GY)xDLY9e)IqOv@nN(ir8}L8<_~X&fe4pb_fBN?5Pw|TI
zB#HqNx+v>s-yg7E!_U2o=S=SfvUTqV>vZ6`gQE1a?s`eQPwaR7lrKpRU?p3pT71ec
z4@|xl0=#cXI6V^Y2Y6IQKJmZ7V`3ZLXW^=N??XrG#^4Z-bi$b2GVStm^R^F9(()eu
z&4rIg;#U!}w&Ij&m>449b{a}PJX|O92BWN}rYM^F^N}|_CK?-*NgG>^^)4)^fHvWL
zvUmLj(@T0&pzng$fK|wPZL+wP#x>tXv@sMlJXou7jQ2<RczA1>_x#rv7#z=$FUBCV
z6E9zBOkEtk{LdeOIX`;x`D50%+&lX6^Ut%s;<G2U{dqh)&@Dg8rMyz~D-2BcV#%(g
zv|+qZ%1*EmsvnwEiQebNd?O`<2hRJariWtG0FeIaZO5POUPd4WAnmB(;qma{qodEB
zd?s+-ws|PO$W(Y~dCuCMm$YH}6foj6Zk{(t`VZF{#)j0H*yW#$Ae7$<l=XG%4Ii7_
zk`>pH*^YFQYAmU{Q}}h6NQ(6CbIBKO@Z#HK@_w6O-4@Nbi@b4E@#Eo-w|vo4mlJK|
zrC(XM1q$3E2M50iWoUkzcfDmIlwWRHym8{E?gGxa*d1x6k5J?K24@X`V!I}+<fk+v
z(N+2umLa3Y-F!5R{U~q86rxV{oRmv#Ht@}W2gfhhad6R=(cuO5K}v)IYl#$_8|XA*
zXAam+?x&4O=T(ingPJuv5!|K39R;sUS|M0nrl@I?*!H69n*;!FA>Gr4T_}@rn;&Gh
zg>R~JAG(kZT*}>0<}=Q#2Ihua>R8p|>$*Q;kf=akqr}TVz;%w|cZx#;;T;P)9c=y6
zV019ax(9IOE45VOn-vbc6+Ej@>`J8&9TdftwHPR;7zQj>Jn<kehO1A0!0V;($N<k^
z=Ym!Tk*p`d2+&~Yxb(aJqv(HK<*8kz?9`EpT{&M+&K)^>DyWr4lnixhplD1`AoWJ6
zQpP<ij`&8La2eR4^(?QoYQ&VntnZ;{l+|>G%2#?d9#j+?&jTEZe;6D!6=nPLCw}U`
zcmLkeW6DaqUZ9KTlzCTP9jDRGe|dX!^89W*R(>5fZ3eH8A3lzU!6($F^$u;~gMiO6
z@A!#V#i^kF8~0APTD9Hqd`PFV<A_W!OzM?ud8+A|d?1hfd;rhBLEW~o{5#d=+M5R~
z2zF6$8Ni?d&aLY8Qfc%>Hf{VP2mhBK-TWPLlkdF;>`wz>`ltMSgwXXzUZc*`OZ;Pn
z^c%Qymt47`<@qiCHeBDX_V<=j5d=9KOjERo8VMVS?dHE7<%gH5iB`n6)kMq1CcSv6
zu6Q{58FGVbPCX>PkaGy|E|`I!DBxlb!iBrh8n>r-7ZHc2?IG`a1gjfO`rn&}6*2qp
zt<!rjkmw8uJE&h^`X(6XC4EO3ldj6|ysopa*Qh$bZ@hhQyV7^1&G^l~4rYcot^#)t
z4VANpibF-B7evKzA8&=U=84n-8DLA%DSH*L1K&AsQq0}Jc+=xv7b^G|B2iFP#|g%b
zPw1wrgcsA0!q#d1D#AA$%B`neRK^+~(!Hj?6>z>)0^WLB)XRl2gC{~`MFT`Q>GPcR
z6~ZRX@3sKP?w$RHP!v1fix^SJ2MY*(r7!v!+KEI8;&ZJD)boD_Z-;fc#shOLJjjnU
zxqhM+;vz-r^@&+I+{XyXS8s{qE@ABs-W^;bez)Z{9-<fQ?V>7w&QzN8I2~o1U3^))
z-QFiIKI_n^xx+%%Jhj4W8AEzE{2zbB(@tQu&G=o6jwq1OTeMqV<4jq~NAkr()N+|p
zqRx`jX*^gOArBuu7}I>fLYIqWT@TaiUNZe>dLa_o5-JVDrHEV7f3K0Ia~Cfkn&yba
z`wU|^QzG^|+t1%RC4Q@)9c^(_Jn0s5p%u<0f0h3J#Lt!PkJmmS3ThITgW6W;%Z1vl
z+=DKb8|xtGZJ;8riP+9N<SP77yxmZj%Ix8bTRfoLmY0egl3=O68P@O%H>+N{q^}E1
z*N4CaTtJUOzI-gabY4IlxJZ~_FLwbmr!X#+?J8&RJaCR84hB6SR9%!Bo!-IP(D|dt
zjZTZ4#R)Hc?I7*b{B^iHzsvRcCPZ9UW-EWtIHyU&OvjvO^Y}zAZ-y*Pr8RTOpr>L|
zp6^cRvpjVN7O@^@5h>SEt}56&wKf1)pNxG~^8pG|1!_AK_Tt(nZG%tB_93vx9q)P-
z4H2hcvXGH$4H2hjdMZVHJr$tWB~RiQRGC(#64#2zt5ZDknM3-i&o<Lr(B2)_E~rp7
zl;lHYt`Vc+Kj#(McNZEL%7(%@RYcfaqpZ!(dKd2Ll2yG8l(&VBz~c_Cz7(P&u@E#U
z<9m3B-m(7U<;(jqSbV+pEj!Kr`pma}ULXDP%P;XD=@A}{BDVklKmbWZK~$2Ll;M1)
zY%yT;sGMUE7)H~T0T5o78#7~5^U$(z{81JfVNM0Ouvp&RYefUiHc((OL<Zk6P||KS
z9U6lGPC&80Q?Cf#8jtf+3_r?79yG5DaPzfX_6BM1R2pHCV@Us~Ki$u}?{{UL-q(km
zyG-MXEJy!3r2I}8+F%a$u7qyT^<^uXlv|(&)a_0S1>&6WHulc<x;>$tw$wQI*q7dL
zo!_`S{bQZ$19quxs*>n*^?~Z9+us(c<!)V=q7Fs|ADMQ97W!5e4b~lbEByu+A=yD~
z7vCdsD{Cj<jtUQe%YZgoajD}Hqe4j^8Q?U#U`rYFRVY}Hm|Cy`$LJSe&IgxIV+Lh-
zNsi`QWbo1XeeQJ22kn6Ky2A|9X`UBFYQ0|1nd|s~k@qX%#Iu8jL&~kITwgcO>+=F9
zZA~BN7#C#D^Xsrtij{Pqyj^aHHVjX-CZUXwGDK>DqB!+Z_;$oS)|=d8t%gCFfh6Oc
zi&Le`eMB@!^3)vTW9<FGn=kHl(o-<)-B3|#G+4)d3=oyG22YeUidk8|VdFCCZX+ZH
zgH@v-Ew3>X<w}~0H2ZWgKQ5e#Hu)G18WHEcW{b{fOXh=oOi83BD$8!C;7OoP@k?9Q
z&(!+dd~McriL>ikkH*jGF?@i_H+t}!U|4Ak-O1)_E+~GD@g=OXab3+L*VjCH6uD_M
z$%AV_?3eLAFa}X|=xmZaE0cJDGG3Hh-)A7@-{0Bsn>*6@#_1Cb6FpHjSk@|l024E^
z5yo=zC~dYyP0jE<BIN}?7Ss!*vFF|~wQme~hy0}}vfw9uVBnMX_X;3r6!Wx9(#T8y
zqpq8VrTi!>?*?gq-_WCB@C@QTNq@$_C2fhohYeHh`o8t2JbdirL_V9x!L#ta;5$T&
zf=l^^AA~LGUU%MvYk+;U)7C4xo9;2{8#4VMwTtVv_-<RS2_NMb{+34B693Y1su+2l
z{l4`#9kU2ljl5OG$m{C6_!f+bD;zVNyw;y^<7j|f4s6R^bP0Q)7nyb0E*9*a7=)V(
zU>pSY4FKkeCa8$#$#heCZ>$HF8DO2Jk~Yo3-y6}7q4kr&Ty9@{jc<m#DsP7Cc=Mlf
zgXhj$pH;LTtu#~|{S;g2m1n-vb1QFy&FApciXv&IQ(sgv02(41g;fVn_EB_9Ij~;9
zH5eKoD%a49yv`_#0Nv*-it`Om%POf?W6zIgeB?||#;;sw5$iAtd@`3l4?Q7|P{xm+
zJc$?N$)ks<%MbCnXgSj?Z{eAGPlu9JQ5Szap)>7~H6AEbjfK;9D$p`ctjDKznU4W2
z)5t`#R|DdleKq9w9k@+z#acVImfS2smYSP7^;CGbMUt?N@fE|%KCBF|2G}@E+`fM8
zTQM3a7I}^gEa{s~uO7ZU`tpk}Lc4~6ZK#o?0g(0(7e}Uf;I)I|Hr&COvai~12Fr$L
z429<1v#;pJ5HsBeE0EeJ+`Gi|zMrc9iHn1OWh*Vxt`X<@6x&sN-NsJyScgHJmRb6i
zXqwnSZ2we!+iCMvm)!?<UH*(~9bcF2>%d(>^BT{tYuneLB<HYoor_J1veWaH&M_Cd
zVZ7^3$GNI(^O(L+fY+9_GfrXqYRdS?(Jdpyx|)}dT<<4GZ>K2c9BAw;h<t8@#78rH
z%+?qJ-gMX9LSOP{8t29VAj8-KSRzg57v(ri-MS%7FaY47%AJz?MEIR-3petr`bsb@
zVMHR|P6Dv&>;XsIvCbzR0yDw(lW#xm4|Lln;kQ$^uEgn4vT?B((8$matw}@EvsGx%
zMe}eedj}1n$+`&p=$rA)@Rn%>5l;?^<X#&koJ%>wmrwU9ObTbCE6UlM`-wSmkpYmp
zRMIgxFhEcymUW82Jmv6!H3IHls$x=M=1CoV7by7*paV`1haQioOg*?5^=Ho*0nhP-
zbeoA!<)vpsTy2DW_Sxt05a=a{F(ppmp1*vaYdtpBpZw5XZRAF0t!VWqoC9NBP9<dO
zhp8Zy`CazXXt~DsgKX|H<>1>wu9-0SHj7g`Swu_uTBlLQoPA9rt;Yo0rPsiuuOMEp
zg%Iv8uPa;UUCNOGEv#7qr+w_f%Lhkae)(l&W14(fuQC?Ay@T<lV6=2F&o-l}>K+*x
zG4P}tBSG0}D8vYH8VdY+Fl_#Fj3#BEe9iX=#vm@bU99V&;c5Ph^A{->PY8J94RO16
z-@M_k`VF$7jD0~_NM_Mm1V%m;z4F)l&aid=gD7m1-uZ#a53-=Xb=Z|ALBRvOCAu#(
zY_4ab*GLb&ZgQIF^?9nhQA(LqH7%Wfqq9b4uvlReA%;!7NilT^gJ<jBKrXq7Y^QT&
z2mB2}86lyDpL8||SsL~?plDB(AEF=aLH$vaqfn=|?VDt-G}#k_6Zj0a+_YObMo+SM
zR>^bjd<PIi-voW$AQd<ez-&^DHIB+P!I+_M`b4+jlAFE*4CmC>d{IGgCVxUF8ez;A
zfk9$~>#~))$?xmV@3ZN~IT}}=lWzXZ=mghg7I<Mt*rH>qYbzbtD~%^{b$tWJTvtsm
z)OrQqAbE!Z^A$>!z6OfZ7w%ti_o0&vD2MSR?$ff&x`j8-=P^_O3{0w|>>U<tUbxqX
z$`ibPvDM-6z6OY%5bALX<D7+xK3QAeIdaOIEEpudzUdRcZ|}J7<Hw(V%6bG}>2y)2
zPtUG5@C_ULoYOof;B{K!5wC`3+K06iDq-J@dCQ0NoLX>r{0UqdEv_>W?lZP=*8`$L
zwg32GwP(DU@|<ape#Nzi)x0G_-L^wS3f;!<QpPh&Y17{FBCZQ$NMeivalCr{GS}L5
z((NJtnR-$&V`!<cRrr4WR}aivKKAGOkyEB;9>AaSj~9WrWloXNV|nIqyLt_=S6_UY
zjm|Wn<Wpnt;zF+uZD2b=r~O7lQ`$UwN-A&RP`3+VLr2zzxG=apBYt==Whf@?y>)*X
zdnF6M4Wkif+uU_u6Yys<G+5U_a>1?Zda}kwV+VOEEBjaIlSY-3&LRqzS~YHW_KKi<
z$93XdCpo{De=N5ZR`AIe_ec-ys!ZcpU`Bpp?(<U5s{E3EP^9=d2=oI3l?aBMz0ngQ
zZDNBicr!;?%wFG?&01E=bs<ap8Ex$Ac!PbcGe$@=w6L^A%*U$uzwy>h+;W&ZK&M@G
z_MR%Y`Uc7;wM|&Zw{N6$TXFDLE4sz!AX8z}E4WS~KlNV#4LIQUof<5)&jW$+^>rhe
z)2~V!d0mH>#`}7k;|A{B9oK7`2FQF`mEL4E&3!g!iqGZdcq35uD!Uvg2R%0)c>^MA
z0hq66*HFD4x_P?&YWSh;bp29a{;Yr_NmgLG2Z~Mu!wj$UHVW5HktEOF;NCfoe9}Yx
z^)*&q_IN{Fd^!#lq~4Kbr%}Epd!M55R)E{LOq^!PtG`UA=!LP)yq|&s&Z1KAYt*PT
zHCUX2Id%O6B-?f#;=90bk)+cz45brxP(`UaMWex>_ofF`W=z8CS>FfbnIw&d9+aJa
z$=V3yZCxcOV#q}Uz&3EYNdqv(m~y}<aqkqpDz=OKWbFxr=_z?@8(}19ELo=pr0Yan
z`0Lcn<2O$N=LU8$c$hkI(Qu~9RQMDEj4=9vbs~>?g%;czM|!A$Gp_Lyek^BVY?zXT
zeZY3MO^=5$VLhe^+eaHVwFsDXHZINuRR{5uOA&s~YyTOjv%&^h+A4A+b(1qd8kV--
z89Ke;2GAOMeZ3PGiNb}3#yisX!T-L~ZL5W&6{(jb9o}c4lx)EPzS-ZTpr$MmxJp8&
z&X_)K{Dk|Iz>js0_+1!JwCf%Xx|P;>ho0Z7FywIq-8yvL-hlZlcjY#|+;Wwab`X*P
z@9$AFbYI=5J%CnK+ps-Y%rHgZ_HPtRNZph#Zi<KIEQ3`S1=Ms65c%|VJ%dkZ2I+3G
z71n{pM9Ypo(dQJJeynTGiwviMLC`IFxb8Nd%L%vP%BxZL3fH<!K~BFr=ndXZFT_iM
zOZcYxI`0x)qmQbT4BPVC&>dN%FBZ&NR`xkDsDOPF;th(#4Yp2rN{y1I&1SEQEmKNo
zP(;PT1-tD)OfSj_hie+rE2twIhoK~o-5%V23SR&8QJkUZ^aQy7#Z%URxDMoRKmQz$
z#WSX1Vl1JQ^Ia7bqSG)v1@i^#BEG}}@Wtm}96fo=re5UHxKL3~-_m1wH(vYxhwrmK
z;jjPk*T8-G<yYC*>`&i*8%2ML!I5bci~u?(|M2n9>7M(c@6EYy*#tFWtMwk<vGSOw
zdd}zWHb&4eP;T-KCeq8J(60ci*!2i#d}v^~eYw+D8a)b8<?q=o+V6wo0dKxYhu`>l
z{@k}^j*ou${)fC)?JLXf@#4bOhb+of;g?6~ey$&!)!Q=~GSYRzIu>e3ySbIWG`Tm5
zhDg6jq^u1wqM)S~y3)S1orfD+l^29-Jv;PAy)Sx@Vgy3GZK~HzFIHZy_7!V9LM?+B
zv(zEH#*+QP@Gce}yxT{-7Zf^&qW3zzts5dGcKle*qr-sj`>7jpG-uV*uzSv3b{R`%
z{8O2Ro08r;+H^LeED+vYjM&6)=e@^y;d2)bc%N%4*iDM9Nuzn2C$vmztf;i`bX_|Z
z=m+~f384_BC0@Y}<;^CggWAse1E0NJ!q+|?!s^6twMIej>Dc>FU6<aCe;wS-@|*T?
z=Hy`90|m_6|1LMF9T(W@cNOo!Aass0jeU2QFn1U74F@yN3$9IeVms6K$bKj63SVDd
zMP2YHg9hGn-kU$s*DKPML}WPSHQKIk5&Cq)E2Dl|Rb_imIzV_t;l#KRJI9P0q;jy~
zf_%%2N(IdkW#v$04s}IQWfF%gVf?i$Cydc}K~S)7hho<I;o({Zjn&1_0lZ)Rb$^ZC
z7)zz=x(Jo5!Lg>NS@e{gQ=bYpil4Iz)4e>Nu$a{~BCcOJdCZ$8Lpe(m?t_rmUOz<=
zyBXZWm4HGfk;}I{*=KwcNY?a-U{n-;We8~uMKQ}4`ThMA4FM1L`|!~75*aeBb?nqv
zlsiTZ>D6W!5G?f6v!k)%bcSAxtP6p*M`Ho2?WiG?GE9*uYH4$Ut>&V??fu9bD;gon
zTcaaPQ~~)4Z%cWmp)fLS!x$u<$6*D+ibEc2TFI%U;<*hFhy6lVWkc7XX;cFZB}#=6
z*S~T%^trBf{@Mg{7$qwSd<h5OnQ_eV#^2{T@4e$|q%DXoCN{Ms8D3JXdD~=vnwah1
zH5q{(+sv`WewM%~&j(%_uT$lHqTdrO^cs8?a(RUdVnSn6m5U5ox28)0+_HnE4KpWs
zyV^5vPSL4Jwh-;?BWWy~i{Mk@Z9|D}Q+kDJG5#cCz=qCTR>A_a!IGG8qHDeibyuj8
z5hYKy%H8-tROR@!PX8@}dF&Xp)kY}>fI%2xcL*y=Enor(T5WBTN1oa#mt6aSK!m<d
zQLddYPx_vIveZ4%u%x9VKclRCbUkLR;Zi4fOi!G;10O{{kWKZ>IGWx8wVc*<EEyxR
zlpA>aT3S`WDy*Iku!v8yjWb_4<w-sK?U0Y$`gHHpzaKQqldJXk4`a50PxTk$Rb08Y
zKG$;eNAWCg2n{@m>k;rt{y9(0t_yjA2Sg>Npj4vuR!waS{?%7sWl`o|{_+<-PWN55
zn>Wo3vNSw0tu$ntdZ}Aa!53_ME^fE4SLy%Dzx+$qRQ&w&-`HmS$<aUjAOBc1qSWu)
zad%H;q5<JlMHxzO-qd2#bM^~SmeLjx1D9T)Ob9&O=L!E!)}_2Tda7rKX(spjc=qcv
z_<UYG#{+_g0x!i^UwwIW|7-ccAc7BN_4dBgGSxm`<MDA#jhmKfoQY4`e7oi)#*1$P
zy}ZvGG+aNzpwhU~$nwy0qzv??Xy~NR3Xcaw1K|~3icAfG*O0zyJ3e?+*VS`TD)m==
z8K!Cf^B65|t<M(rLNgdRz4j|@sL`c-HJGwpJ8MGP5NVG|)_^F#+(%&WUqs*uK5>#@
zox&_Z!OzzLtNeI6XLw-^D|k~Uj00rg=rT#&{PcaYN%?TGEGQ0J)sGh1q}z)=EqoM&
zeJ|dmJv1LeVac}e?SqdLchu1`>GoOHi{w5WvIptVa;kuYE~jnqDMC!~+^7ppbS$Xp
zw2cA;AQubZjp(~tj5k7cJK&{*w4^~JQ9Ic&k%NZanSC$?O0h8*8}SxKXI+JF1v?j+
z<I-Eh!<=e-An@cdXQzy8b_VA;e_#9(y_f4w$M?n<hp6-vcq3R_2A*z{mb0(~e*=@N
z?Vp0!PFr7Dc4v9<Tjx|y-*An0F4t*m{J9=DcINF$6rN)rP6ue$XehV5$usabFFpr5
zB*Y`KMxtE(16V^MJ&o_X%Uc$(TD5=F(FRA>2vA?v43Izbowq|Ucrs;zr@#%ZRIsjb
zFwKg@r7@vW)${S$XP+H?jzQtuFdk1(xNd^xRE9=@;~A$In$mLTJ{Q#c=+R@I(52qJ
zL*B$7a1p4=`4w-TJVIIf-5I?iwcBeAnK8C9e3m!GDUq`ar))Gr;0tBx8!`qaUUFs`
z#?cwm9?Dn)>sO2p*SPrYo}ZsSW!mW}ykY=AhsKlN)Hy$Yf-&NzWwr3udQ?j8Wb00&
zu4Pf-yY?eqE7peihRGW~{%3o<y8DVY;|&=+L~`lCt~@k4`yLUW26);K`WH_HGS}$S
zfU$+Lo<t#v*fzC~Y&9I0l=or7daABBcDuBLKChQ}Wb`)qR*zF{Jv~#NmW|R(e^D#(
zZylj^RA!Q4uDVhRd|zI}^m(A}%Q}yKza+}*p&{}B{(W#<SC`~KcuvqZQ<pn5^iG_$
zu&%@VQS-eI92Go>A)ke;L8?t%J#e(^;B$+?k3{2%9afU6ldWorF+y&Xlz9v3E5Pe7
zbA#QaLro}eU+X}kKsn+GkQc8Bv2ZK8T&J$3LI7zn9oV-GwT}`qo>36riIeoHd+e-I
z(u0d1#W|x*fh3I7OPL##;<&T4>n1(1uq`D3rNUJ{m#w@n=TO9(({hBsS((R3^y6ZQ
z3pZf;Y?*am+*jAG^mSSBErI%=A|c)x2K1as5tn64T@9cEPis2*dWc}bVtVRV7KE8r
zWpa+v@oCk=JmFn5XkT$r<i!b3|1m{)8fFtMg9<HpP>8w5P(onNtI5{pUJ$h%I#~23
zusY4!H6l(>)=rCPkWd}#Kv39}cRk2kjF2qObU`Xcfm0?LBGz>$`-+@qy~h{4lBxmn
z_19l>d~@{ew|_eN(>LG5n~}Fi;7etA?!r#_l4uBUleAOn|HuFNKWl9Q-jE-D_~Ga&
z#(?V;{=fg{|BWH=%P&9kalb#svl5R1$~*=ebU(PydXsw(vt_(*jA%gA{v5RzhP#_8
z!_3t~ca#LxE3+3EApeTj#kW}g{XhSD^b3Z_v!_3&pIYWm@bLie$9;B7{o;$yjz0VR
zvrNJ0C3?+$^W@zV42gS3-~8#D7-2u-Rnee%`s?#dd$^Xzdf&ZdPZa2qcfAr1*-J!u
zq-`-U%5$~pmlVP<sE_zX=E5j5!ae1u;RyY<;ki~7p*X_`<IB9YzljTFF&a5EcJxHa
z?|t6r(Xi97(ctp6T-#Q-Y8Yp$YD%Y#)flzVr15FDO}=%KHV64JtkY+hj_-q=)3>HJ
zyWn+aXbKf8+AOwB?RAK5mLDUU3cA`p2a7Vh8M-a#l4p$tTNwvjL>BzC{*H8@bC&Zg
zmn?!!zNs8@<^lEH7Nk~OrF2Lhc;@(RQFpoKq|cSUw}u@VqklKTo1cSo2jc3o1E2Z3
zV7k$&QMS3bTw*&PH4sJb8T5r>C;jH9_5ml5_Ny-|Ec_3c)G;dZ#8_8_gP`Dzs$d!J
zh%s1s8u`Q$sxz7hmT{^Ke@V~?U5vH?6M`BQFy?z&h@_ZAT@kFxa1B@i-S9U!Cjz>T
z)G3XfH_^LQ#Wnd&`&OA<7js>!{HAG6HZ23rs6X`(&os<CrcscVf4f4;l1FP54uN*?
zr%_VwVY5+W7$3HqRa&=7(<fpTqi=e=#&B?oL}SFop+SoR)GOfhfnE>~UrX&*Q}v2C
z&7cQF-0~5srRV==?@hbwxRG>S&Vxvewq?7jPjz+QwNC&4pL4(VP*t8~jTFc8JZ~U!
z=iVYIl5$CQ+aPmiCWb&n00;yELwXJR+WMgdz{Rq=T2#D9pS35R;)!D6^g+%Z0iKqL
z^RvAuRVr5N@WlqARl^0G;wtW2+qK1a-2=vwr$}I7$Rbu63-$VSzInr!W4Ryk9T=xk
z+-hql>okmtnH_jKaksaEXLy%zls6>jHb>-Zu@T=))eyEIxbn=@7M^Y0*kXr|N`UJ=
z?AYaZ6i=SD+@vhQRYgUf8!ulEmY3q-oKh-BGnRYat87(LElXjWr&H0bSP(O7Ud%r`
zVn+Vft17DoEXdR*YgGPBtE;+J&)5Era$bsancq@`*^A{iceA>adNdDv-}7zqeIKlW
zR&Hfrpk5ST)c@tr3g+>9AK0tI8(QD5j(L|5p-wpaGCcI7x}~}LJh`Qp$@e`EbKE;e
zlT?OVj)dOf+7ifFu2XNf6d`8FWuL`cQuC9=QkK?m5f1(0ouERmwEP6=Cjjml&BXKe
zKMJ`Lhbw8#@KU!3$Xq%?*STE<20tLfJyG}xO^f)!mts{A*lrxL8mhS~os6Ry^X!J9
zaPbcTWP*xY=~;%4V3G&^7|aieH7}cd-7R&vGs@7Eh{nM2^hV=qMpl@aBB*Pk2j2=S
zVe7LbtdEf@1GwP1$ybGh>WEl;?UcNWf@#hXh6-vA)iP#kEa5swoacPGRF_eoyx`e%
zowu+i9{8*auJw4=UxjaYF97rN#+4(n$1}HgoN2rYYtqPIoTKK(J03D%gxQh$Ts$uN
zI!*2)o_Oe{c!SQ$q0$M<Z-dxj*vZEXFKn)iLglhNAAcFW0#p1Z7j*ccT*|^>r!~Tb
z&2x8_)xp*nO<3aQRD*fFVkeAi1^(+l|8wmn_rLwm-_lV!rQjJDfBO3CC=6Obw1Bvy
z<pEPL53m*-Fuii%V$y+o<CYyVyl72veTeHX%u5A81>!TTC^=IECB!L%Z@&93J8{1I
z_B(da{E)ir`jH3hym>gLetx+45xk4HR!*lER6;VnftK*(@q;WV-K0IZCHIFPzA0-3
z))o~Gagg>m+*KB~w|Bw4@c;7Zmup|UwgnoU4vNx4R&G1CR!PtP-f7msSY|3?Ro96`
zC9=tSocyWZwOOar_Y4j?f3Kv4kK4e|IVB=54i5bSZ*3p;JQdfhsNj6|_(^v3IQ3KY
z!>J&STR(o4^3BP&;APpk<3)Jx&T&VR<sfgII&-^p)_x!iQWsNI+)B;5Ae9^5{SvJA
zSXFpfS7uoEkR{>tsJqttv>EVE9^>AA*7?dCU^?QTnASEaPTZBPfyrHGJDa<}3jPDe
zO87kP^n68}#&L>wi-+I^Y99ZubIMD^u?YtM3wJhwCq?&;OMd{}gMxR$+Y-A7Fr&rW
zHXmP{yW%f?R<MTZ>vE*LN9ryI%wtW@@9=%_aDy%6NiGo$hvz&(Z+i)7mA@@_%3m2d
zk39$ssjP>yGThwVCT>H$B=ZsU4m3P!uYaedJZ3U>;d|RB0N;wv1$?nDMpi{zl*R(&
z1lq{pYr6^BA+V_|K1^{ti-DT{6}=i}DJ>mob~jRinh-EWQU!o<iQvDLU+rTSzvCu;
z0*p8%9D-oZ^m&FeJZ6unjuIUwynbFp(q|5-bDYPVKQB_=?kSDb{AFn4@eb3kfNA_r
zdqtF6d0&z5rF4>SH{`icnCMA?%IS(13zf*EaW6xZw<S26%aDa%luf)+*JijC*$Rpp
zooi8WDj?s5VG6;~wD7c2*byG01lZ}i71srpc7DE>6~x1_Hlxs7ur@+|hAz?<HuBg`
zumn?YlWpZoi;8KqwAe9=x95b==dzXb?b<e-Y!<dAoG`L*`>~@H@m9f6uy3Am0NOE&
zV%gHWNjo@4(XoS8!EuXmb>c;v_$f3!$<)QRDhclL(aK?7;?_ddyPXZcOpSQ(Sy_c#
z6QW|LlCR;>a>{QkhtSHK7LPi2#Raw~Bx#@W$I$A*!lO|((jOk^DNSVTWm$Nchkx}>
zhiKmV44x<Lcs7nsg)iliH-7MrOc>!)DUnXsdE{mGQ`F5oZt^HhuaVc%^L6Pv{&jJ#
z3;W>o0xy0fn_^|v=VEzyIEyi1BmX|kjm(ySj7$pyLaxf=D*q*5@4N^-IgE$sAU^Bv
z1TxSneL<c*Gf*FIH@aoyccYlUu1M;kGV(6Y5tIw%Y{b#r5kd+g|EgirY`YU5C1;&i
zAlTl{&iLdv_n1^I0JqVJNPY%JJO?Iev@ST3Q?NQWteLpoG;(l6OdN%#4LVkoL?R>p
zGqM72&FI0KQxjvg*iq1l1C|T~U{0+BN&B(LFTxb2;vYdp*oJElVBTiZnn@vEzvRW~
zmlEn~3P%VE+D-D?<WuUO!pWm7ypo9F^2B?}%R<3PpChsy>(0Aj6_Vhqe@tikOJY=g
zxFp~PUcu2$i7Rg0eYZ2wH1DpC?pVyrzmu^pAZY+n{#7?+LE&V_lwu;9IVHCw>2%Wb
z`t=c(kC$tI|J(m$8pP8~wG=$yybl+M{-1yOa~7Ze<uCsl>ws2}96`?X#V#MgbIpOB
zB$aLLLdzLAyUSx51(AuCwc4##{LU1Lbc3K)fc=9*?sYcEYm^2N&*&Zn<ed8af}JdW
z!{#S&^I$dU+Qfokdpvh{4Bfq!Ku7(4UDtNy38!}YKm71LYarglx?_zcMe%E6d+mUe
zT~$CHV$IQ_<2sZ_51(M!d4Qr}n4Q1;terl`$0#;<Ji$t_2kl?7E6DXMKmGJ<?G5yF
zyOkEpNmuQ%j&XNKjhAPLIDKXNoGcPi&?t9^vz^LQ6&!i&bNETv4r@}P6hUjY1iL79
zShHg3AkVIo(hJ7H!yn)CF9WY}Q+nbVGw(cm_ewh&cfrJOlxM&a)-E!oUuO5foW1|z
zkK+3tI<L($gkKxukH8P$z%6=;E{Urut(C3g%rWXp*!OwZ$9%pD@Au=7hK3+f*H}<d
z@RsT}UQ?Uv_Ku%%CM9SI#w>5a^`$rQ@=D@Tei#X%e6*=q)BLW-Fq>h_MF<%LP;jLK
z&~bKB;3SIh1gIZ(pA}}t*Sirf;i7*$@%CW=(rd{PmVbW+EE87ZrI;V<MNHeh^E}7P
z=c%wpie^*;xZ+OAU);@McGWSH_=Br=FQ*N3c0T&9=NQCz-RQ?O1#J=C4!qX>kOOtd
zue7T4v_hnn;}8qQwEN}oK2{HoEvEy`+7CKVzo%k7)m3Lxb+IfSPy`JQVlw)tld5~Y
zcZUgLq00_c3&bus_!VBKaAG+CcBV9dQ-`DpmbfmyC2uV+K_O;D735OFo!T(H3P{$E
z059AGZ`;w<c71S9g<#LG;6m@t4zxK<L8o6zzMU=#4{c2)*G1i`^iHX`V054Ri3){x
z@-lDBt6iqmh=-OCYlPD*S@Y558JOhZXh1eXW3C3k-GPT!riFPoZmtjwZ^krBcx(EZ
zOQ{Fu%=4CyDP8i)9a!cNR4;=waGHY5d)06CeTw`GKYVCDMer|-)b9~o7zyEhFCN@N
z38_)XDu8BXz+O_=q)L*?b~*D-`%k}vf(_;_&PK0rvH{ICb{o2<SZXO*6&A^-7?ROS
z1dHYdu(*hjL5Nup%?=u<qqQ^UUv;Gjup<ItrkP@oId*pwJm5?MN1M)#*BJxs6yQ(m
zgZQ{Q-))1(d=J7A?6Vh~Y^eaonvyS<GLlqKX|WI|1-C-A7Q0nBMJWLv<)zQ`2XXXy
zVpcEwDpdJTr=kBj^M|fq*<cQHVVRktf3%0PI9;9HR1d`8Xc94VO<|Wob*L4g%fI~+
z*E;V7oDQFXyYXZ5y)7MWb<{r{7-e@tA)GK`-^5QPB>B;ad5;C+irgfz;EO7BV?jtq
z#Z(2wPo1hWDqg&#Tcr@)VPSmNN4SpUumAq<S^M$-{oB8%gH{<aKRe(|`B7OcbJw(F
zeuVt&^jmqberDYXaO>ME-2GS@ZAa^XxCDRFIhDhZpDCXQtb^E1-S&GoF3NPBgC~=^
zWmtIDTm5&oSggw3lR;Hrq({raDc@GHlwZHD>6d3eKTlq8gSziJ2^4wKJbZAt_8B{R
zwCuRk=MW{!buC^hTDnr$l>tu2Tm-jGED=xm{>}+jq9b_YB3gHwJbU)@+CTpB5Ax^S
z5G**cX31#F3Rth9L3p-dr?TAjvjrS+)*57uO!-p=mV?f8)xeaf@EHQig5pBq9Z3pL
z%+Jeu)b-K3f42KrLkgr2Z|n-jzyo+HVsYT-la_|YRSQup5Bkk=$azJ^o%v0V!I8IF
zn&fF-`04tdmp^`fQyl2Jn2I_uPd%V~x0b8P-RbJeW@WF8x~i?-@1Dn}NGbLTIGc`)
z?1oTuiV-pJS6PW0w{-M`WkDNv=cL0U1c86&2$7z~i4SrmoD!T+`N?Pyfqehqe&)zO
z<ItpCMrsO57YVvZPUk$Yg53G(nC?&ZBao1m5!gp;d9+igIce_B)yJ6`S8(PB1dBce
z%LtoHID;_LNeEZP@x@M%_`1LlPQ8ATUun9`iIicDT;GWsbP2m6<s!`$u)C36nf@|-
zR{SE^2mCoaVcIF>Ldg`%slGJ~322_1U<r|`bJSi)r!Eq&eEf}X-pw;sH#>99|8$a`
z45~8Xc^*y&*ilCDHeWkf@3?goisD{?Qx9vQYbiT)eY~r5WPvx93p+z|*FTE!C<~uk
z5ldaPOuR-LwCpnNXFB_07Lj3hs5Z*fAu9FS<|)Jz@0d0z3_p&S`mcqMcR%Fj38-2B
zfdb&ygk9^h@5lCH4XL#tDml_8-l#39I>DEGdpn~v$ZOY;h_`K1Yt@VA_o57F@rZ>I
zzgUW(quMBz5EiRCWh!pr6S#?=F27wKyw8g>dEyr<4e$(;&MZ15p9vQJah<q8x9@as
zcm-&9lcwcUl`H=Gachfk6c2E0oK4wy`>en5uW|`9y;r+_9(-u;KkolKI4~>ei6ACU
zFY}(lXAe}>)UbDJatTb}bj9h4G)w(keRSmGI*BwdNRN=rh(dla|4c{ke6B0-(=pg-
z^M6VlIlGI{mSe}UKp<cl9qEK?lQ!+-6e#dcE5#<){p<#Ky!JUGGzI0CUw@e$1osaf
z)SS04xsHx%G?|fKa6|wzT71qK3yzAOJ>%>JtR&xj`)x*Qj@HDf%FFz{^wA<yB57y4
zg#XY`Dwr5GLgT^sBui?3A`p4=?#I0R>g%U)5%jbAS4!0alhDm;=osG3i*&-^f0+w+
zt`Ms7AdY++etX^cPTCFNi@(==t%@D69gCezjv}}7^}G+_>Fg}J!|43%IQ*0To)W3(
zzz3EcxnW-YRaWby>k>SO%es;22I~H2pM4(d+b-v3SmIhEq7c#HWsxXx!e`+%0>tyb
z6$%w)pSAAQ;ag@webZ0nWGt8zSN)FJ>0;+B9@3dDva|`?hPjk-gK|LqKifG?curG=
zS812rU6*0Jm#oz&rKpZkKiIz?yzkww1+|}j_B6Y5p5$ayr(?9N*sfWBfx>jm+7byf
zUvc$$dlTyeF7$!3ikg;*zy9^#)}FC$MTPP?mKP613q2?s^27Gw`5anuUXtd;%Y%H4
z*Y!T`PJ&n}XSAFsfac*qhJ@@?vAjy5lP6C7q>K?IDs_GlH(QO#%RJi|M>kjYI!s!0
zReok)apN}zAYG~nBRFl0a!DN1lE$m8AJ${(@Jtf`7i{89kVo3kYPvvL{P;f5il_Mo
zzeSW=Kh5d7b>by#ZlC)1oZ=n6Q<kfaq4d1CSA$pHyYjU661=XIb$_>Nb)rvmzbigg
zeubgDYlf;gWJNv-hSmqS!a7Qfu#h}hpwP7e=9RnoE5J@{PzWs}`P-eLu=D*5VETbA
z;XY#bfhOV{9nA>skVShnZ|%ZQI(tWGnAY8btYu|$1M@gAEfzogrNFmS@xUraeJ&F8
zASlh+;wN3w7r_QjF*zg5bQesoWEv@NlCq#h+>=YZ>KWM>eRgh^9Z5g>ZK+z4dtv@A
z8ms=M^IG*^N;&wD3NZ`lJo`Qiy7?n@z<Q9nK&s$DXKG$%TV+LgjrtRY3*9M}l%t6f
zaYjUkc^CHLeTpjx{+!CtLa>W@-zfyY+NzRmn04cX&QtInzLolr0ZifBxQR~i_5o(}
z;aS#26#}OVqd;Jd(=un9$XOcZM<<*uy!dHBFpUSeRs6zRuyMlnPZ+|MN8qZ}QU&JB
z4lsDvBGJGg4m;1w3bM;$RIDGa0m-f&I&=pcb!<5_L)%oR$;b3Mah6t&kpsMk$bml*
z!#|Wb7pdOcKhCZp={m%M<SrwhjiYjK&WHBg`s%cjgM_S8!76gfhxPzTZpBmEwH?0J
zD3zD6n^6BFi-@Z$T)`3k2|sxZd@IKCuY9}I4R{ITIdBV`_R=0ysYNu%iY?CimsoiF
zI-V1!pRGJP)4k7ioKEw;_<w}|HaO7b+SpRQ^#Qx~c+XvJd+y#DeI)}bSXZ1I)U`Rh
zf7<g0;Nj=3Up;?4MtrVlS3y8gU^--X*KLXqVzzdk+)<&z!RF`}Fcpq3vM$5(5a{@M
zS0E{fGfG2XXpXf}s=(|c!2IgyeRd-V@2juA<oXIz_&#QEj>tw4ao$__F-gBw06Ivb
z^Ux5D0JbC9VRy<Yf_!Js>6!DjC!Ff&$jfhfIB)%u?__xQ903)QQc9{*SkGt-<;ahK
z&5Aqyp7~B4j{-%;THXj}r+9dDU5mo|lR4{P>S19ZhPrf`^|<^y>S{VsLpp8OVZ1tR
zm2Jpef}y|p3=z8&cGH)@0FGtob4Hs37vJsZEK}=LjBV7vi8eB^1M<BsRn`$Pl8m;5
zG%f|7H5}9>RSfHn|BmGCY+ZY_Lr1O?AJ;Z`sMila2(PSchS@P{H8|w}G4n~DVM2Zi
z18_DhJ42}N9+DMnBF`!iy*TrP{Cw6*XD3<*kJXxyKGn08kqLiJ2tDDb_Vu?=?tl1&
zqH$C_Oqs&B*PIfnJD?q*V$q)VIFsXW?W?C>rA<BL8#k^4+1ssiIP6x@+toOd&YyJb
zz|tCJ+dJk!H1j)UZA<d8P~f3kmTjY2JpJ~KZT=OfsDASeTZ~y4>s5HJWvlQuJlFDM
zyVu%zzC)fcOy1tZPuQNH;xt&Nlh6Leno70~GE|1NcB<6a)>mEe!pN&3*S~k3#wzw&
z%dP>ENTZEd@VdEQUWGBrYlKbj390nKdA$E|`)zZe>rqL{riEHe`QLb9YSkwab-e;k
zY3z9zwNBtG6}0NxEZjJsz%?GX%^HGT&w)uHg2A;Iu|}|hexFg!lgAHHLRh!aN(k^t
zfyDunC;~{6u7b<Yek+g^LJEo3ubs9iOGG%SDI6tc7e&Jbac&b-pgCG|ZeArt1;Yld
z5VW(gK(oM<jzVX6EHsHm;WDD#;2<44lI&tYf$-VUu3HC<n?s6#bK${FjP6`qsAWWD
zq(9#+>_xdJSMt@@Vui8cE#;nn#pAET-4&0nDwn12ux3a%F5?<@S(=-0LpUlPfzO?S
z8f*ok%;!=lI}+beL5v%`j$-H#fpQE2`ukVdnl2;xf!sOkC{zTZWD=#z+37@&a(CP9
z8Ru+R=S-^ta>CSxM}J%IoK6t3mIDv+@}yJiWoU?%u#^DbtFD%^Ar8k_QM{i=8Ns@O
z5?=n)KH7(E!*)}Lnw1iwUtjZ)zDf*<!+4crNm4^Q<p7T=7kB`^sn6hITeR-0(7Tw^
zPTxhLZYP)XvHnOHmp5q&UBc$os};J9Gb}CSRRc5gsBqhbU)I&S;trlH;-xG-^exI2
zN`g2ES9-PhxEsmK!O7-kDf8BaY?lWU*M4aYv3xAmcEOpKENXWfGMD8V`1ltOcreK#
z887c~A)YvWpGVj4J+PLdNb8Eb3T#JPN^cpoy98DTU3Z^BxGvhYVee9LZS+3^=Nw3Z
zDj&+T%cFX*Tb>g_4IeG7z)S}wt(4GI`N`9mHkME66z6xvL6RK_-rKwvi^OBj1$e^9
zNejrMoEXVBCJ;Ce?lS^Xc&i}Lfxmk7BFcx~YS0uLAqt$=uinHO^5c)+Bh=rlefQnB
z5l#xILrzR|`sHg*r2HBSh<SKIq=HYztITM!^<6<DpZ&(e;tY@Ca%3FCGDldUJPJdV
zK?`$F#Psr9gCBqVA*=cQTI%!XKQn3t-_ynUBto?U32TdK9Tk}!PlAoaRX`Wb-lYR^
zZ;C?ZQ~lmjV9CpQy$!)DFuL<d7<5HxAf4KL+)Di0kS>H;d|8~EU<(AecFg8)-SSk+
z)CoHLOn;>WNN1;r<w11p{Bt*r78J_GIusNMFl=jkOfzZ^;oXjIn+UFNPyumGk)7oK
z{Oxa17OXFJobJNPt`wBxTD0m84=sbWZPzIdNA6w%mNqQ|Do{C43>c2sqezP%m+!*%
zF0Dr@C*bT<h1NqYDTaCZ0XFOL1(pysDYxlX9VAT%EDgDGmVLFG$TbuyOb^%%b05pc
z{reA5xLDxL)Qk#+w65nXy(}W7jz?jK51tL;@%dx|FP)mwQbHWw%`RC&o2}GxQIXS{
z<RHb}FCLDjLh+yfd69EisvNQQpv-9{@%uF%qV`rRNubcqP?Qc;kjUS`h~*LGjeNSE
zo`Zw+!INT-Tn~cvRRz(q7Wa1Yrk#1u-`jun-n^UpTaGtzl}Ffhd2}n=%%$z|zS@o^
z|1kI(cr#d6KCa|>WxPuR7W}(3_z%x_IA9&nbzQZ@y{0lY5U2dQ@$MKwcJ-~f!azvP
zXpLsHA7THxoE7knmC)^h9@g~a@n<a1dy++ZpEJFoU~nb8X5B2#!_uL^(1LKz0a6OL
zEvyh)KrE0`I(A1!hllm$N(HtfOii~6RE4YgDAY4YjZoJqe7itP;z<f>6*9)ft8?en
zOUfvNLj`Y%v{qs)Ed3nmDje-36jt6%_l5<BT3zfY9HC~xVCKBRAG7zUSIN?xJS!oK
zaj6`_iSRL!Bgd+PbM!vWGa27B>u_&gFXqozV4Jca-ZJN}h<pd{=Ha{(5lq%1kguJ&
zXgX?_&R8@CKDG6h)=2LdJc(eZ+?|)@ufN-=&#@M`UIX<cFr-LJf_2pTrG+e`Z{D>^
zxWmM$MW;+W=G@Hf7)vQQPM!694HXG5@ed8w)$pN8r7aw9fhk^D^8rq_0Xt9ONvq07
zD<eMJ4!k;^X;r#x2PKcl#r7u=)W4JvI4A_M2gU-T1=A@{Erjc|J1q}vYz;r{%r|oI
z8z})Q?zU&YBcpY~(_}LZBi{1R?er=x;3Q9lRb^KSpH|M`kAfi0Ds8cJl79^%&Pztn
z_RPopj5Bq-;GHs{9OSd%(%jBkE<W<iwO*$z1Q(Z%lkz8W-vif*^8qd;2(9gk-o<<T
z<&6)scpZQB-n6Cw2%NAJ&l{v)h4;t%uag5)xmw1vIv9V-xwdZH1b6D1^}@@>A4g-?
zwM8TiKR&;%pATt(PGhJTJV60@`t&JZ34O}_i&#bW_YV+gB_Mh|$U2M86i_ZM)!pOJ
zBlGWVv<fG;=6OP+AJI|(DLh=C5v7K_>ae8R`|rpm4puvTA}j}(F)fp}3sr%zju3U!
zoAVlom-0p;_^EMw7|Tv!<|s?Wqw`Ve(FGqA6u<WA0#=2aosTei9Rf~RRS&8zn2$<F
z!Oy7FC<cB4chh|5e4C?rnSaCVuz7r++rJfq)$p6Vn@8u9XW??SUpg1!oY3h|h_Mm~
zH=VI@@v~F39MFR8qUtYjwVYT!m50+g;+lHFsNUr~U8ptPLkGyzo^?cp!j9h~(_Keo
zU46wp(=6oR;>vV{Ly0i1SM$V5C>_GmFH;`ikq#4DV(}0c({qm!qcUQ>^G?2A4cB|o
z0-eyS9=^f$aR0I%DOM44ti{>cX~r5U^VbTpn<<!Dfa;DMm=d040HH<`Y^7fXgsB@(
z|3wCP5Sj|#34;mqKV$eN&m2hP-2niw<cII#>ATNS%AnnWiGz<W2N#Pw2obmBFVZvD
z3)Q6|S8y8<$X;beo;7cjF_w$)bzKI&`yBp*^Dp6=24Qu6t8iES8=uv9ALIQ)@*eE`
z7-;_Q&4c=9^{tk<AoEshLMmBR;98t$CGQHo3VM;rui{Dc?uepi3ixrZhgd26P}ieJ
z-ph=uiB$pYC`e(WOT(kqdN)g_a^kOW;7?QS#x4RLxSma*GLh}H&krM@G;4b$4KRdT
zYe5_($#aFz==A4#k{}~2?#40y6FwqWosu~RV&bS&5Imk;1h=*J5SmoB?z02O(QbX}
z`z_NmjD8WQn?L^;Vfuy-jtMh!&(OrTxy&)NnP+Wblya|MzncrE`SYj1T>j-E^C&bb
z7W(mSJ#z<$9kU<fv-8n84lq31B0K$12osK<Q7705t9oIFWk>FJQdDBHlO^%1<1|kd
z0Qur{fxENpP&NM-ytRBk#`a*HRT0<UI@=4L{p5i-w-TZftP_qjs#FCQ_wG=QZ=KUB
z(+Wryw9?V*#78TEbufwqbxrF<g&HnLTmxd~tukdt?_dFD@h{odQy07n$961Ug`4$u
ztREq4bCd1AC?mm<gihJme$JT+(mL{rWA4TK%=KNQ%pd^ToVsdDj<~y?Mq26%y;@n^
zQRnoP->Ko$Rru%Nz;Nk~vPM2<TCLogOCI{mTXB(&DE{V+zx2!%5KxH=-p0SByQ(cj
zo?HRzu6Z@M&NKA2CwKi<{bs>`JpKkaP^~J3xrA+|W0rj0_hMb<nk*l~5@Lh+wYo+P
zadamX6+rh{B&wiwtE`t_52a-QkCD(z7t|?8BkAc4^yflg$*vPdQ2IofK+w6fM)T^{
z&LKkTEYl)>Bj(`V&%`R{_FP>Wo<c_f2@o4&8Y+2d680;d#^EjTiIs3ihK6efR{jLl
z#1Li)u{H^tUIFd;7V&f~f|%zU7??l4XYGoXku%n!)ENW_Uq;SRHiT)`Ce)FabY{9}
z@~ohb^()Uq^i>(kkbjbsmv@C;d@Er)hcQj3(9zhT_NZEa>0FN@K{}^ktaJEL!o`E!
zgjQv4*<0>?HNH_G=m4YWP=3~V6#?t19jMcHcA$?~3niXWAmkx=m10ia8FjPQ8kIGT
zZ@h43h6+Wt`Vz;y?Cj0U^p1qY1Hgf;Qes%IZRk;ccBL2(=M<opfiq4-b@ZR<lPDg9
zd-o<jS*UG?Z=1u~kak3V<Rv_nDRJ2%KWX;4vOo#J>e$xCC_ICQS~2Cp$q9>f$^U2Q
zixMC`!)fLD>8GDErTG%&&HRKbowhZ3pcSO<WgYRkEh2(R*@P!59!`;I=~3|+YeIl$
zIjR6z9@1l&E;1rdD?edP)sciKQ<UM|epjO3<~8_^@_8p6{$%}!`#p1j`fEK~t!lTn
zgea6!&jzrzh#(S*y<OABs~vwHjjvFMxpQ6;wxp_{iw30WkT?n&?k>EH<JpRCgHJX*
z3sd1?bR(}e1cgc#_|eb|>+enk7gBBQl!>^u8Uq_4EyfX62rvbM+hzatFMr91&z&j?
zBtOfo5EQloahndnEq&rOHEPo9jH9yyF<;w!ga?9I0b(I2B;92Y{E4f;&a-L3`vPl<
zn4JxtC|rvM@(>Re7Mgj~5okC&Zb8H7455sd7}P&79T9GFU;@8_$5k2l)b8OUK52f)
z$&pxX6so&>PSYIGA#G%;<{66?!&8(j`MeD;DN2U8W+TmoW5b)@3wM0L#hAj;2}hTA
z@f`H0!%c`u<d+{2TZ{^guK;%>Zawn7VHSA>e}GurM*eyPC*RF;yf=(^%zhHXcq4Ad
z>$J0|H-41%PTxF{mWfUwL6`5R<ZA~iAN+`!t}Fs77>w%A?BwM`rLT^>ruAgC>2&5<
z^;yMoXp^!^x(M5&{-09UT)?@%carJ51#gR2PUX2(okRow>hx92#npD<LRPO6I#HDw
zUGpr~3F@B*+%>-9Sk^k%X#|ygu~vW6$tV4{ou1@u9A!lO3<vfmb<{R>z)8dIEIG^q
z)UCA#hkWh;tB-!xt?&{YUFd3h<&)_mQ_|L4gbE*3+zM&pD6&{!>H=<d`nWANN)3ME
zri?j8aeKpp^i09R0C(T)AFv*Usi6J{oq6O-y(~U;eVOO^WG&XP5Q)pq4hwuy0JV_l
zHa22*1cy@Cj^5TzI|mzm#)$F6(MxU=T|Scr{L0Vwa%6LbGwzC$M!f>3!nzz+V&>Jn
zbyvmFU-`E#kBQgulW3N%{rjGDGaNep{N8c<J`eNxDx3^kO{2UXl*>LM2*8--;Br5-
z*K~)v#4mJ#QVcA3GZ>wq<M;dOy>;#7q@<ehBlt}3geB3yaTTDL#F*gy9jKSkTX`+#
z0g1~=2M>)q*kyQ7SdQr1jkoAl#U2X5AY>Kk$wD$1Uv;pEw_p_<1CweHV2}z0BdCWd
zAlFoYI)cP>qaz}%$o5Fi7frDW(JellC2TK+0b<uo^4{Cp%vSz-ifN512)FaTsB=At
zoc#S!S;$;0^S4gJe8z5-zw*^k1(%)N7OnP_bsCu>L6DGpeIYkfBxam{B!*O6;>6v*
zlTc*pj5zn%mYWfi6&_^kis(bY@XlX1j!ms8;OVK2=AUKyWX0?YB*1}~hDm`b8Oq_6
zcvx-tLy1zERMD!=sIV-dZ(Q?SUwe4(F!zIleMVWQ2=aSTLO`6+?K4*VKTF3YEq)o;
zxpPH{0=p<L>9$lyMmbeAAQtZy%yc@ag@m6Sgkj<*PqGd|KQJ;(g+sXAK})k`6M6^_
zABBa>vv>T>D^6HE3(xY_1sDB|H{ATW{Z{YA+Ja>s1#LCf``<fP^YlvjLsvS%#v3qg
zg*ZkbEX7MD1YS@^(pT#|nEr_3YUf@H7F8r0wjG{v>m6$yaouRr?M2C<a^M<<_)%xX
z!%kJF;_YPzi8t4RZ+4hYF9_Q<a*A^C5<fe7zpUE_cL_(DH)y}&k~JcX__e@<kEAar
zo^{!Bq}6Y?l=7l7u`H7aMlnNq#?o+~_Ul06i!VM;{n+2Tmv(O)uUHUNLQsrSg5)Xg
zZn+nKol29+0Z0X5l?X$AMhP*j>vzhDy6@Dg)}xcRD0cj5O>rCWqlb^NqOz+Fi)yD4
zH}h)+H1L42=g0Hlt-E=4`9iRYk!ge#2Li(OP&OBks|?u|lxgx(;<{`lL$A~){JU)x
zpxgZhvOnz#aNgvJSLNOj#vLRQ7TgM#DF%KGGCUt~r|=1c5GrlrjdV#n;6ilTb?y?5
zTY@>u%djp7pW7w=cnJjHST~ewlbb!3U*$E0uRI7-bSgjSC4EqCyyx)+MR?<})VT-$
z;^o9NqX*6m2i`}lK%S(Mc+B$}&yo4!B%P)WJl>IwK}Qik;!b(3<YC$(s^YHCWJ2f}
z2*1NR;NDmXxdiuHdF9yT(?uh&!T7h~)C2hYQ7&DPU7bw=h$rEGX550TuXQqt*yIRl
z-W7&c<4Gy8L!^`P@+<)Fo;>OlhoeUY#L+m;*%Kh0R&f4WA)>~+yGwl5ez-p3&wu(l
zqm-IYuI?q)3k3cKbG4hqkARd8@URDpAP`Tj6F5z_;lIew-FUB0t}t>{7E4PO4O*VA
zxcA!w1S&BA06+jqL_t&%SWHk1d=H06x5KDaMQME(4#J;Jb~@<VT$|2Wo+y}1Wk*Z)
z!GKo}H&UhwQDLrdL-(tO7G;VM-i7UUOY^ofazyQj`N1<45EtdW!R-7qyyu`KwilX}
zTa<Ga1Nf0Jc?NHu+x1gA9Q=iA9`e@4*Jmq4aga}r4hipf<A?xrK#aeS6(Y1sC(k`a
z(fN4q#yt0H|Et1Q<KE;wG=W3Yc9S&oIMSB#nS1iwI7+`Re6X&{FRpVH2>i2~DdkMP
zgg3@Z+T6=85qVdcc<ovZJY0BfXQ;xUl5d*ucI2gH&oJwWR+EHV_eQ+hkrT}PNmZ-w
z2&pWnxbLKs|I^ohO1>vd>jgJ)=X&cIB$U6?elGHS_9N36yj!1ala?>hQkR@&jP(ZP
zRJ_H>I-2?m^Rzhl>>3Zl{WzmS+^wrx3y$8nlZDQkHl-7+^*LK}iEA5<Vt~?+whWC?
zDBy7xh2v>_>-W>AUzSCsc>zwk>{LO<ykZTAxci|#6|Jnd^22y+!;d3#Y4IxAW0{t;
zU-wwoG#?bd3)(*ljXXO-enr^<u4fk|9sa2k<6g{CF+>3Ze`(Yujt}^Khy$(3S!dHG
z+b0DD>g2tY&&bm?Eq}P;K-wmcmdmf*U54wbU)L(B@iKg=Zd^m;M^i%NwVk|@w)#Xn
z#u6eCzk>=J&O468DhtRP&M`o+kTjMEB6b6m@(8Fg=jsR}mJ3E<u6J;&tkVe!l~*Vo
zjvQaDJFm-}vRglJGA?FPOw68A_yD1i#iI(dfmiaQk;&+D{E<P$bOj>#87~*_^ZI4{
zrmXMjqLq@SL#YX*0Kt#M7kr3Imf1Nguz4d6eKJqN+mt8pgVnC_Ph{d5m*4yuut7(X
zKcXe?2?2epgF*YD(#U61h*Q&1u=mz64P)&%fG3*TH6aVkZJVbkmnvIiYi<$zl~3WB
zfMSjj%88?VY<#)9LqaP}@eeq}3_7M2cwhLZo>-Ca6Hc7Z^|buk>Akm`Keyip+=Xo3
zvm9;(?KbH`<KTsLLl@<k06RXBa9Sp%%d?7dbx6gZ@SJp6-{4_(cIv_~g&m`nAp8NK
z>0<?;rqDSG$N1j0rfJ18Y9ND?eiteNOcU+xQ3t7`r&tgC`g0Tt@H4K{4R(|-UwSUh
zan8Arj^~@t*EVRArjsTuIWpEXo8S@Ud*Eta%Tx?uU|kkT@~$sKTlf3}hEqe~Q%yVh
zjTPAkw6qoTSHWsa!_<Z$OlygFLU7TkqDH!VST5@X<C@Y{*6395OtbfDFY7dvIVwuZ
zBX^@|<&dXySBln98q-ECIq;|4TE3~EoMY<D9o~qOVIc%yJjM0h(QH3g0B_Kv-%9qC
z@Fu@0=z`PH=dhwkhp}_tirM*k|HJ)uIWRBTcgoWGfon_3JnnTAe>5e8c9_Q4O`sc>
zcN=}X=*m#1@UTAThhW!Z7|+p<Y07--sOC5p4~2lEjE7h~w0t~b0jZ-8Eg+vg`aC0?
zm@N@Vu2ff0Jm<q!np(f(!)}g-G>0lI{H%A5PuGINC^0w$nd<zqW<+5v-a)-^k_LrG
z?sdasyc@#f6kdJ%Bq%n}Hl#}hA|l0S4o*5To}B{W&NbWNo@~^)1E;68$WQZnc&K(Z
zLG;N^gJ~PX;hR%eDjQ<qx|1#D&OI|>n+^jNAo*`GN{oOTaN6AM=**78T|)Y~op+B9
zbiLq6do4`zP8@bvAE1zqVgQe_7KR+-(h-1y0#uLDneA)Uz^#um-U3duPygyb?O5T%
ztQ_nNb%DiQVLq=*3H*pZ<~_&zpOUrobEV@NG*9`Ec@~u7;o1bti-Hb+ta}au<e?qB
zTS-p{Px%mdK^-{bJPhl}#i%dxU7E6}6n{Hu>%P`S{k539<$Q`WE-eHO6l_;o13bl3
zW_T=llpLlPs?B#97VNLTei~XlBgY+Co?qhuf37*w5@;uG8;?bh4qmvWpio~iGJ<Wa
zEwO;mwpBpRnTpxuMAKLrs3X>oQhKTmot)IJn1g%wW3Su*$85!gZj~ivfP)d!i?hzJ
zS8H8S`Bo-8eCU^~8R)=+SWp<`=u{M>-vi3LyQ@e9qKIH&aZQTx<)2eGnZAK%(ilYw
zJ{MdTu43)YnjPlKSOx;EFcespefW&b@;U9*wlQlTn$pQ6w0E2F-W^1cy~9tYF9%P+
zCUr3!683=S?k)XR6`{l%aegm1dEym*P#*2NURe+3ZuxB<^68(&HP)|D{;5v`|A`*Z
zlm_Fj(smUEA594nwK=~V=_;XF_2e;cG$`+y_Z5a)`=Xj=7$IawRsuDx6M<M;-VWGC
z=xK$YJ@G)X`m*Jt`;TK4Ipq{kSCQ}W?FbvOLM4mA6b4MqxSh~d>jxfc#I-$yQCT4d
z3ngMS@S>-x&2r_L`}wS*)#<(SGT!S3r6YnWf`rGEg`eT|yXZh_4!$~E1*P#7Y?^&5
zE?&amQpy@fA>46Yh@?<dpeW=toAZ+RPO~7*(0Nn`eV#dx*<p@)Z}$}b@Tu}6w(hzv
zg#~k>Z{EU-JS20sjdn%!o{x;-*HEQZ`t0NcP<U2F?_8S?)|XC~yYabP)aG56I;CGH
z{+RcHLl|+Fd^`O-t$xrQ$8*!~dq(#7saO~-d`=wd2{GGpwyx5tB?GPqDjk-*<jRNC
zD0kNox!84+Q*iAF9nI$}y{#M)X3dK7O<ARb*3T&q7G=ARD@+HzrR0=HGUGiRF4HL5
zDw2QZ{fZK;nVwwg=>8=qhl=IP2QRXoBMJ?jtZnpY6D4;;WrP^4r79Zed`aij;^vmt
z3Bm*mxz-02<xC;*?3yK+<l%a<qG+jU8~L_H9m;}Kln@mRFKP5L+-ncZsLH(GgHf@%
zhn#VMqe5oA&0xiHK`GcgX6h38vUys{c*{Ds`)6tkp4%SRv2?oDz%*W^=)-67ZMabi
zoj#Q~aQB*&Ny`x2Td9});;aumjjR1k-p@SV+L>PWAutNU@93AhW8F(x+>~tNe`?n_
z^xe9Iy9~2;>Vtif={oIuX!&SLNHuB&ksX|kM_~o}jG&qb35*6Wu@ZvY)RgQfOhH2Y
zUiV;_){h)@&VnzGQ}-}2FAx3l+bf=S8UaDa2pAL4xVQ81IbR>KYF=Sy8t2qsvBp7x
zv(4zydF6dZF5>9%%AO?J77&Gz^e9j~&bi1y5c6=()PxHxF3yX!riq)%q>3@4R04$W
zwa!`)&EC$jw;9%vrr(l*2u$RTco0^=e@myK@Og{XBKV>>NvCHHh?gJP)74@{MxZ<L
zPhnc4Lh)1)0YoZFSro_@|7b1zZaL3H$*5?U*XfvUF^-DHJq~I+X1=?=iQ|ZjJSZpU
z<sD@o9but@XS?A!ofWh+tvLt5yu`y?L<Ia|1>;=~Mfl4*;z*l3SceBuIE>G`Y4gsr
zbt-%u>4m2=j;`M7xjecY#(frar$_P+&fM?fg2ljjN-xamg$KezC-i6f1uqZv%C=AH
ziM+7<*SU<__^43UnyfzVTqS`hl;Z}AJdY@^Q`+U0$_Ko3EtPlEdr!Tf(+qqNvffvi
zAH7o%L6JeZcbe|_gwC6|d=Z5_hE>guwp1haTU_x-8K4zSnjAzpIQZi8FMxBBPa_CV
zOUP?Z^40S8=FO|MBP<JA80_S8ZU*w>@cu!pA&(wBp+2(PgtqSj(ns7~;JMEtYwM=Y
zx@=pw3e^rG+N}A97uETz_(-3%rf9!lD#Z`eae6Y=RTsV1loUL1{RZ)>PP<sNc$`Dr
zz`vA5;M4ZtCJ2o7jcLQRFIo__f@l#@uBdz{YkEDKIOi-LcqQ&CE5>&a<+J&EPun7H
z*S=UhB+~TyL*vpYHX*nB5q4WreD1sg1%Bn7d}MBMyWp={M9TNK^P9Sm`fz)kO<Q%T
z%Ag@Mf0lUIf!r@kGYkIbKT?jt-SWiE@zlTT6AuYr!q>F;c3nw?YkoeO5+afcG}=-|
zuT9pRxW+RC@RNeF5F<dW<O+;xaK;s$TLk?A=3^F<y87Hplde|@QgWc<_Af#Jc-sn5
z{}5O*QDsKKw)fhvftFA*t@F<c5j%gciwoE#gT*cVA{{3q%`8r(y-Qp2;qLc6jHf^@
zLW;UBtht{NdYe@IQ-}&iIvo`vR;^l~CkV=XY|Mj_=rpn&CXA35e0YowK%Si=M6<0>
zVUUlV>B|)*2MqfT!h!+ma^g=uUace{-=MvZrG_3mIcQQ?s#ut>ol~VV8}kunRm3vI
zqxiw&Sg-Vlw;8>GU$Qohjxe2_f_&Nxtb#+g8cT}s6Klks(vSB5Sb-<>f}GHbvUn_<
z=kIrU?mYW`TNsz6(@(lq!|?0TuT~3#<yY#K(Y?rDnsf1!dQCank)<5CX9pVPZzpDb
zr(dJa5r=gmx~>a($&XM++cDfXGpA1YY#D9R9yZU}Ld~_|lnkT&a>0bhe&-Q7S|Jr@
zl^GRjKc2VgZYyZp#(GdzJ6bPEO}$0uz3GB&J72V#!vnr#O<R8a8Ee0=&^>;@o$au7
zsG!^5`4$d%(*I}L$Rp$J#E<xnlUP~bVo9VfXK(=?I$2+BVJ*r{6a?gr0~DzbCHdh^
zYXu21NCNLzD+)khEt6fmXr;obEO!&AqeyEjL}@31k?i29a!5r4_+0ZgWZnc;UPiZE
zX3iX!eP)3-eI$?UWHYbf7EgHKzw|!hEU@ZSf84`?E~6Pge>I$G-1^>#&2a7d)OBkr
z`%csMPJ6}uqbVV2p!U)<1v@c2cwOV!LR4WPaqblHuD_{H-G)XZ%iJkx{RV{VME=6|
z*?jPgoh<IEaBev!V8CbL9St045hz*)#KWA0yUo;s_naNT=*F#d`JRUHDi!aJ4r>IO
z(-Or6;io`yWaJ1=7#SJG>|8~W0*b$qrHhNix>-Tiqn%0#u?QAGh^L&`#<H-rhw?_U
zZSZr>e`otOBc~HAH!mYpok~g;(5hnKs8n9PVZ^k(g?V`6P5gJ!I7+Ef!Eq`^WdPb@
zf=H%}IQf)19!^JDPIWS*=}N$8u-Bqe7fKp<CQP!W7hZkAw{`9@`rOBa?ZRFc<ax;_
zX%8Qv!(B;EhyBC{^z`e4pN!`vuHIE(QWv1j^s(^BM_kACUJBCSHSq{6q3H8vj&RfI
z`hHooPvJM=BYrddEr0296m1+kswfdC0;M?Hkus9UUyDVRG37^{@JqPabpkx=*EZ#x
zdO#K@q<5OZwGo~}BQLj3QD(7}L{WeV&d_Z-2q3<BZ_)9dYf%e(sjuW|n+RTED{0BY
z>UH5X3V$l>7G_z-n1<nGQE77LhvjMqn>wZShO}<+b)ALyYT3+3;gUb=#F!Fs0P^$;
z>Kk?K=~rJ<?+&si%riKgPL+M<CwAOtY^!Ar7&~QH$i!BAcz17DdE~G1CYPlRP2?Tr
zz<9Gug+&EJ3y^ia+wI%)+9sQ|8t__W!E#e!F?|#Z__L0Ts82W42l*nSjEe@RB}5B|
zys-?$S$@3WW0=DAQwsWdxS{a%6Cc^1A|Ri6U5O0Q?-z8!R7Op1<ez-!xEP_srpUj|
zhhKt&a~LX!(>pv$B^|7>uEh1-@K5Br-lrZ7JQ|zEv)_%Q>om7P`Qs@eU=?D5Y4aV1
z*MCMTK1+(%D2i%a{AxPw&dwrqKH3JJ!;<JQHNWOu1B7Qov3dGe>W&%f@KK7AiDD5*
zJeV6$5fZv-rwVaugoN<GU;GuAjPHCK$!bnG9U#7`UIY1h_IG$BohgkcW;lQH5Qy<L
zWU6~fN{2!;k9t2x@a234aNm4ur>wvOBVZ`BIngXTfFhj47sbPM9vfI@A~2C33en26
zgtMItzM-#1S)K{uB1Q0UMLt5uv^o_7zaW}XnF=1>R=l>1sK_~A{?M<8M)|l$8L5QS
zbb(M5o(faTPjmDU+c7s$MpR0Kl?p8byt*-RpL5YTUS;iC4hy0m`=@dBeplDDuBO+2
z4wvWd2+S>trDceJ34!VaebyO=6xCs+EL8_cmvUd+-Eyu4ZzM2jIrBChJ*DYv1GW`M
z#dr@TWRwn3Q{%f+S826wjlV9#DlyBtvg@oY<f?0x!c&2>_CwnEv%E{WV(@^{Wc%OC
z?i=pZvpiD=^IaC|m+k}wLEVRcE@%gq*OYVkw7}mqiHukAlbnfDO3J`I@8B=s6(=O)
zU(i<F%5RF4g|wMB{OdN5wq<67g*NF6?24;}C2iONL<SdCwn--q7C^hu)GwFZJG}3g
zM7ho@l#cw;MrV-?OZd!hdhGn5>o@B@6!%&Bo~F2pU(F@MeZDH*N9E-V|0&;5u8r^9
zKlQ!KeG0hz;o~VGj)>?iGeV=GO{3Pq;Zr)a>KqlO{6m>^XPpEduFLSfjy>L{lk<Bn
z_t!rE{B!1h`R)lF-ZlzIt<kWdD;<SVqd_}yr$!<KB1#pW3M3-g2a{)}TzE*n6n_qD
z0oQx$d{2dur}Mmy+#RtwvT?+6$VflR5cAnNPsC9u02USPogyN$626M&bO#icLM4g=
z=(NC=AI8{mbPDd<ehiP)S_Ygichspc%XcaeY?)hTl;n=%>|}5yKIZXKPVj>#ZNg1{
z;AtMx7NHCtZ<*7!Q}Zkp`QQh;B5;Njw;WU)p8eo~WqintG%6lnKK(jM$DgoBeZ`3M
z>#zP43sW!hQ%On11pheFxe!=bfB*YG^8B~|{;#Y7`1bqnGjH$N4sqaPTCGj6r^--%
zakVgvGD2D5QEA5SQl7Jr_)Fj~<$#r(?+TN2mtr&g4HJHsQxphjU3WB0*@wTjD|qI!
z<$n&pv~<)|V%2p&ZkUCrr1LaX`6kaYV(+?3PLh!etTO3AJK6Ni3p_8FBJ_R+e5xVs
z!9%%9L8wlddTYB9XJA*ow!Dx5;tXt`ErSD;Q|qM4e%4V~M;ZMev4h2Wd35|H^|~xS
zs6ki~+>zq0E@j9CQ<KLavBC7r5o<;OMt-bsffi7bw4|3`)>qreE7q8ZmtnfpE%;>{
z6FJ{~P8*U(z;e1rzRIWCCQM@f!w+B-&!xdSYh5TFq@6)ulny_B2Yo8rIY&f(fY&zd
zdwZkSnbjh35l30)V89c;UVVNHP4Kvte~%f+AON@$5n2Wv+$NVgKHiF^538e=Q_8X#
z2kdTteK))LbKd>IDT9A+ZsI#Uo0muPXzmx@*f?;bYUcS<|IBf;Uu)dO(Xhs~@A}#L
z=jpDa;o~VG1H)qR93=w$M{wA1spTPR63Ge#qAR3aBay|92!wn07+s?PI1+N9m<I@D
zCAlErK&}YFMUKrB#AP8c8dK$1OzSub5(RXx^mpN9!boF7^poK(lvKD75X*sx?kLoO
z4;oV*d0wD_zXh3F(LNqWg7{4UIv(lG-^<0Hns2p;n5VSs>;P0Mv>LhKaD&CMT0zob
z5zl*+kU@9T6X7DypvjJ)-lc<%2Kq#y%R_!Rb#iPcMrUD&^Y^<ue2|Y0#ck1UZ+5E8
za#3N*2#*e<T`W%W5Dz;^)8x+10~L)u%;qvEcyqVX_b!=+;k_F}`w351_}0CAuK(?H
z#`4+pw|3xYJ*a}cb?QoB`s<v0AUyO@!SigsTz1H<Q1k-C@Xa(v)Qz}8iSnI3toUr1
zSni3IU&*o5g>PEwht@RwRBRK|_z4zbb-W{KZjw`@`UPFVi8@MS4X=IF*`}(4v!kQ4
zrk*YEPbT57Fl!*w?NQj~&(sf7H>aB{_T1!fGIwkZ+FkFZqUIKN&kXTd=aeG9!6WZ^
zHUntP%{;VBl(MFxmHdfUJe6u94jLLpwH)~&ZUhtAx~#15auKS^%0wS+RoRstJ>(_4
zDjy%}Iq&A97s`uigG3>cd@8p@$$S5_zwx`xci64(v;1zI_KyTh9O!bfSXZ54m-(;8
z`{?gBfEAZ^hTjx&%X;?lln_{I!-=&QY|H9F$4G;6^x|kLojV<=4L^JJ5sW*ST5UMS
zQvrFxE*LwlfBEyDm=8W&`}3dwoE;=8BT;0KYBo~8rI2X`@;uK;iltNt%cIxBvY1jQ
z)<yUjt}vI^HHu^x3ZsAh6!a1HjNXNP%#M+>*R`lm!Qxg^O~NV%cED-yqreuTt|}K;
z@c>Vh=G3;ihB=aqLUklMggpyv73$fY!03z_@d!7qMGD$e&TDuJPH&msP*9&^3f|)E
z2XvCqvb)n0NoEbd;T0(4S^io+=L%`=T0^S5NKBrs=xBj)N5)UjevBZ^=gBFT&s_&%
zx%}y0GFP9SNsg@HiH%8kesjl>VF`2SKtb6hj%!HV?O<8EW<kEaKoQX*q7sx*FmQIE
zNrWwn2;pWF_wd?6o4LG6;EKOdx@`L1aaQ5Zz`ghJL(+=Jgv%_a_M2d*<EMP!i+*;e
z)_qSTl{b#CGx|nBw2V(UXvTPRtC1aIM(pNE`dFBN>#hu~dG6%!+R&mx9Lvv6-f5E5
zHLMRA6tL^c9X8vPN8u?nD->3fD9R2RXdhS;v{1T8_4JIjAK-R&RQQuTc)AwJT```*
z>$HO(OB7F=sP)C|t>F<lShs!F66o%+<VsxEmv|EF#raRNf_mJ_Mkzd|k&ZffmHNaY
z)`b&_{1|`ak$kClT2-`qN~<Q?@LhUvPPy<KoaHgCEfxz;I@Ri7oAi`eclb!#bnXdC
zhzfdNu4(aGRvVn%lk!X1k(TtQ@Y^fXsSu_$D}xE>iIZ3zI7|5OEkAx+fnW8XBHZ+W
z{HmN63Hn{0c<I_tT=Y@4{g&NC&mD-kGHvQ20I%Rpr~6c#v#yB4gvTX#w(=ocR^xSD
zFzkKqqbVVmHvm{{MfFN`rL&~bYX()}P$0V6U16Z$P%(M*=s^~rxefOKYlw5dwa8UC
zg0t9`i!RXZ?8Dg+yn;~L^YH6v%{&=AS3+u0T?uvtr6VhMuDB~j!EJuA%m{RXS%OfQ
z%h=WXlHW+viHjzb1a`@IN~=~5;W{PcgM2bhK5T~^Qi<Rv9eWfz1fB;7DPH5aMbIJ4
zR74zQO`Tc<oPG#d;AH&_@i~@>tZG4Ijp$TvN+FORF7owbYMB-gzI?Is5wFUXyyTU6
z;={jIBY7JWG#(xlCGRZfBd#ccgv*=$$khPVWXxZaPxH|{`*|tSB`@>1^XE&VUg_0&
z_5I3-R|MQ7-8_!vDsT0(Q+79)9h1s%lnkDAt$ZYI07xN@v=xI#%Y48}7*Qfrj0*pY
zC*(zWFC5SPN(9sERD8&lAn*8*zOTSEU-DCF76*Flq&-2gKgQ4LlG;v8{X$vM+Lkph
zqzO7e#VDEQ4!1tK;PvgXRtrCFSoE|KqNUEfovsZozz~+dxlcbnj1X?7aX=KMCu>{u
zOv0cp#TmzNmo;GzKj9~L={#i$TBRz-ziUAZeWmaVU%8Tt1@0<#Vnhp(g``)Zak{N|
zpR$xT^D}SY>w-s~CiKSJ;x^DG0#Bo?7NIwPn&8ZmnE$oTJpXVD4h(*@a@PF&&2g#A
zb)`OaeQdfGgkS#j@stodPddm5B5<)m=~_5u1cs3CL{1eD&+f1@^7K;I96Wvc1=fX~
znEJIi?4-kDaOFrN^TB>wMxikpHVVNm^Q^jTp>2+Er&(9aO=$hd&m8Q;9nn|q6bWO+
zP%t=R5g!*@io5F>JpTQ~_*A*`&BndYmphGj|MA0guG=aK!HZFyStX-_>8q~Gu)`H>
z=B$ZU=M;%~9y|IXY%}xWSc_1=u+OaRuQ7^3fbW5i@tikzr27-2FCDKaF7+X;vO?63
z3OlRQVH)-$aSG3NuGEjkVer#6H)UO^^9>^SiF^;skBfr+=$CoCX0-Pfn0~@sCGmg}
zo=S@4>)gHR>~PG>lOcl_;VPexP%KXv*}nertML2Tk3Xh<xaLKL<bv|o1uvL(K^mVe
z^QuQ($cgq*e=I3+s1mz<)-8}$QwE;&NoSW2`B+j?>rwZfjqxA<_y4A|{`=qmo;5<>
zVMVeo<k)lT(m>36w-X{HJIllfQ5^A$#gFo`?#kb?ijmhUa^|nHNk_Y|toc5a8am>(
zbP^$D?%6hec<t4zmkG-^Z>ZNVU%p5O?Rp98MixjzpbCZG$EkM}Sl5zNnL8!xZkRpl
zrpm8trczJL%NB&y!jApkA^ymNdk$!<i#%(=SjWO?f!V(Oqk>Z(l~XdS@~|y=(y*6<
zBYCH5eoB{Dc9EgD`&}CO>5erOsI1fC`33Y@kG-^9^^P%>M=iVV^l@j5@NK*D&9hO2
z^O`jBxxc4FLB+bDg+pu>I8jgVYlX1ztT=h*$Flh#N1Kb2?%{%iJfHa)WQ=*e{@}bv
ze!sO_mrwIYel55+sF^C%ub(>UX<AmXH7xW|&%EEnc^G%%*=UzsY7jo{O|*=JxGR{b
zUNrCuEYVw<{~U&pQvft-k#_V}DS<WMM*F9s+ZZ(mKl$tlof_u-C?6h=?8oPJ=-dyo
zD9jFCL7^gIrFK*#uq;FcS^u;UWJIoIqF5rj6D`>$NdApO0yMKaSEM7%IuYI&&p+|i
zpT9EGdK@}}8Vi5|YvZiGopJ(B7YNO*m#?dXg(gkfT1IMqTun^)A_TN3I^G4JAM%n{
z7GnKX*2Yl?!b8)Ox~H_rJ?3+*LN5MQX>(L|c7jrZ5PyqsRDp0&Z8{kEqfpPb&5X`4
zeM=x$@B$gXENU7R49lA(f58UK(piPC3UbQAZHHQ_JmAbzB!eBWGIfJOvZbYkb0a)T
zytY8wQB`NM!I>1E2a?K5J}M;&X{WU`wJUs6-v}cSuxC!Dc@39_G2WLN3Ao6^xr=Z#
zZiii#&IB{~dF4k~i(K_BS7plcfN7l0x1W=4z?@+CE`8n`AJR_Y)^j_1!{DKJJ347B
z1+Px}G&fK56p!aGUSx6aYdZY2u>o8DdYLXf07mMiu*aoUOrCeAQMC|+@Dp~yvn9`*
zgmNh(th)uP?*<juL?i1TsI#%aQD4((rye?$=vI7ar43oXt(T=7m2zX)5IcdWrA|df
zdUm#1I2*iCs6~w*+v>?FvV->LP8*f${r%k>S#Ns?ZILInEfy68KgMSCwV@L{Ef@dd
zq*2h|#m~eu>vg6}2%Y{2x7@s>(e_qyO1xBR>I!VCcGg&tjh~mubT2%~2miE$a4non
z<V|Mk!c?BgFFx&WP=|XLXT!YT#gW!hH+80XgfU=E5P9$Cz6S@Q_Pz0*+KX_2-Qj)j
z^qq&%+I5_~&*_~#cU;pqEbsk1h0h*3e1b21gT`63TRk?+r9_P<u&2D1BGk)L7-@q0
zz$eeEX&TptJMw8b@1*JY!s#?V4!C_EuXXD*eV@fQf@$XY<#`1U<F(te^DvdHNKyGL
z@Oalfk$J4vPlsyf!?ndHvxOZ2iJ&~q!9+kA#JC<HogQ}=Gde1B>clG}BtY)(c>>|i
z+F$?mU-O~4zx?I@WMp^1sMxf2YAPL$daShF_+z3aAK%?slQj`EbQ`-0i2L~6D(XCL
zEW{xt?0~?!+o9^JGlv7x#UEHFFTbr~n8HdJQ4GMrKfkxKaexwnP?LT~DegMZLZaaR
z`KO-(!xLRqK<;tkqW2valL|pNUh)WGRGqi@D0oX~5rL?7_oaAjpo}S0{UDuA^EE+9
zFKo|9@f#P86zzom_K$Bf2mNa7Vo~Vz!e03l;SX(&)>JqYFm_Pp>!;%N`|`^#1OL^J
z9PFfm1;4clc_xc-RR-Oy;29>)v3potPJu*;@hioy9}&0AtAjhqrzrzRjg~DZKBTNI
zH*r(}wd}w9{=0N8*@a_A%%yAL6)so@;>!=u@dN+FG|UVWG{wq$AG#D+#Lt&MiCZ}&
zO64=s>tozN>3c=XZ|CcM63_!*BXl}>Yn*ia(3(1+AD7OKKYY#`KO-;K5?Ef+W4LP{
z!mB4w(l$Jt%Q|!qCB$zcsYH7J_S<jL9-6nij<!K2il1eTvy;yqxaOfXYkRaK%R$B8
zy0EoDS|VEgyysE@CCr7})Ld}mPMayqh{{uyLD)zF>j7V(Tsyx<UdV*!uj&JKZ&?#!
z+wrUzJ7eoj@j=d+lf0=8y*hS25;rUg)<>-r8F;`i*O!^UYbWIEK2}rLk38m64yN%C
zGu!pio3|+kEpl2^a;^=QyN!L@KlIX8r9Y=NclkI?NhNWbCDT0`Cn3(ZDBW6LQo^9^
zyKTYsEw(Z5nTCQE;fk+!I22{$l-);`zh$k$(o<`dS1CDUma?!$6;}p#!r-uJ@~=03
zXJM1y9UjatmZpHLa<#prPLfC8%`f$R<k1!z>$1-JqO(rvGll!C|9}-ZJganf`{+B2
zdzVhqUgm^pUAmDYuf~5)!;R8j5=ov2Z`FkZ@Iz~E(>afmbb1UP`6!gh8ajB{bT#ki
zIayW{zXD($YPR*>zLrOD5w^I=vyqR<bse|w6KyLr1|Q-FrU`)~F47&Ggs(s8wkBYD
zT<5?mjgg`kh<{o5Ntz%FAiRrF89mTeMQjv)jjC!Dzm^V#hLzF9rF(Q>kNM!7Ru4^+
z*;O&71CRx&Y?&8TdJcFQA@N)a5CUNcsNiHr&H4!&vZM4c8(i$b!w6RQ3)Dg?gz=XV
z_({aL<=Gc~2FoT6`7XUt;-D984uD0U0`M$f#YI>l%;m!|-@n-6#8o?=5|~zy%F=?6
zjx=4K`6s_xPa!OEEQ^VP$4q59*UIR~4!wn&%7K?WcF|`>lVd~{oCfbS{rX;IN<Q)g
zr=R>uToAQW=#jBm%Tq^+p386#b9i>dP<DRU>($Fzqmqw~p_pCJLBSDlsVScBEQC{v
zpE)j^a3|{sc|-A}9CQViN%|KP^AwgWpo}m*Wawos1u#qN`+MA_F$NsJl@ok`5kU(7
z+S(AX9nbL2yYrmkm@wirpXGIg{YV!jXS4(3TBo86(iv|O*D`U6!Va*NkA@W`XXM2(
zFu}J!l^{o}x>*y+(@xd1V&Xa43Q3Cbg`Y3^gE7H{Xs-G3GFhGM2`uWKZM4fG$~5&=
zyt9=!9Wwl~japyG+jdhzR^TL+=HNnyJpH+XLjSUe7_aKLgOl1W3)HnUzAxhu@#6UA
zbt!Pxb=yaj0ordLD5jrySeC};Ke_SaKhB4Gk)M%s!OL=zU-C{Fz+eJ}N_><>;w*kT
zc{b6|@fl3XBjb0P$Twiex|H++pj<@iwoaTs)UD=w1<c!EXWZMw|4<TsE9osQ`EpgM
zHP7A&zoJYorQ5h&rXyX$U5dX7bN2pD{sZSZoa$8wFt5Hw22752)CH)n;A-`9z^$M9
zInwf;#)FV>Znqks!m6ex6o^R0G(^qu^puqqNc*RW-O=)ATyE)n%t+4DIiG&<MLrU>
zKP(<SnmNb1V#7W=tFKWCL!{(^jooPlN7QzvIf;;aOrGGUVsMIZvm;R8){YH2xN*v<
zqa&Sg1iR>Dd4Y^x5N4-X5sl10{79#eXb3Xm3LA}t^%)9K7HJ{?#lscx7kuthA>?$4
zb{3;La#awKGy=xaPdgQfO-5mk6rIAb&}6IYQRgna2o42Dt$J&_)>>#0ah}=5qry0%
zLwd$2Q-RyciNaPF`Uhb1G@E?M6GWY(Cjz@+o`pXu2kwG-LisrzW%)fodHlD3`zzn<
zIp74_uh+i$<_E6t^HH*6g)k5muDe^#-Gd`mD)VqBj<DS&W4)3W`b$^-c(G_EH(pXd
zWr&@H>3qgX#}qc=&XZ4YW)GE87LTW;3}^nG#(3u0@7(9lB0Rw|4sqKu(hX5O#cOr3
zt?9fhm()c&GUIYh3XWwaO#Eznw!2J4Fq+??9cA<bY|BSD_rH5c`(Pwa`KQhjCT#Qf
zSKPZESdaB^1aIB7ZQ0sAO;v-{2B+^P0aI)3B0hq*3b18BTq>%0*l}tBwIj~DCSbY)
z#<~y-AV3aTLv^xC9Jbd=yI;8GOjAir;h$pF;2~j|MkOE?I2BbEoNjQDBjGCTP(*a|
zb4uwg-!IZxr(F|bn^G~;a;oC|1u{p=jMKojm#l4qkJ+(8J&nazJaDn@fQNkN=6`-N
zbws8muMEdpgOx;i<kZqDKFsH{C&J3pSOj5|0|sl6a8)>j8_Nf6B9;)8Hq*(+3{cEh
zR_MAS)!VIpm+I!NQd%Ze??U@^X;XODMW|{7(TOz2sfzu*Ib09?e^cn-(Um00YvRbm
ziFcOMl?m<^&@!ycY=Y8dh{rt72_GZQ@R@L&(S>bvo^%3T52oOXi4kd|=rpa~y1?PR
z8m5e>PN72m#T<%cvlGv@RXY6#<KuD<F^Q^tJm4d8DkgrV(zOhMO~X_GirWS2AoPq7
zAj1jwJeLK*M(k)`!IKded7fzrp#kGDO<#Dn2ir>~^lH@wubM&h9!l2~!3X@)_$E&Y
z=EVz_24qLTrJ}P=hib>MaZ3IhPSdamlYCtruiM^M#ESRmFr<%6nupR?ikzKIg1Ff^
z@d{nVV@pC8q$8VrvJ)|_xH#pNH7f{HJ1XNVjOFWzxEcqg$uMb%T;r}K$q}bhEbge$
zge=oN)l`A#2kFl2NQ}oHMGGwQQ~|QHB3~~%3d18nDbLW%*LTAw5yjaN5v6F0mp_*A
z7+oe-qKv1N8~B={QmuroyaxP=-(|1!c=PAPB|nHQ>xR#(Ztgd5TaA+Uj%Q)F!!LD!
z_tZDS+7eLKfZ=aiF3Um;Qhwl6<yn09KrNMd51ex<MP={!!g^?lkf%K7FM<VFT2F-O
zdzYCw*q$tV4+aAp@UfPtMh8k^p0u`oU#sq6W9A?;wqmPn5Rklxm(jd!kmvXlrcx?G
z+q-O6<(n}(Oz~c@{&<J?Q`!(g6YOMYTjgIEdnymej40J25G2SAew?8&X>l~N_!}qd
zj5767U&-Fztp%-H3=W*8RQXV$%X%5hS=<BDJiX-6pxt+1r;K=(wp>x(YAQvg!s!<;
z%gZtno^2&<8O}(H<zjeSKw<@<eZ>lbf69$GsUMOVM*N0;!b905&&v7KbgVvioOkXG
z=beD-Jz|Z#KFU8WbV2bQvzYO0&c8XdiI4Ed=ZT+p<kuW_@)XV-j}c)`+x)q4Fj{ak
zZR0|Ho}Lo{8IX!SfQE#m8XmdfLGNnYcV)Y7hlUzwXM(?n7uo4}nLLe#yA94or8aPP
zxj5IKlTIH#W&!Hsn2r6Ex6>~%(UZnq6?T+Im`rW%rj-=G;u;gJMBut%&Ks|d)@%Em
zkEKDlLPh#r_7ah8kBMA5=Mb-S^rQP$E?(v2iK7<4q#*21W1hxGTBlK@aR}SvCvgc=
zn7CFXU+Cq|*T^#>-Ngzby`Hh*S{IcUowP)860n6vW-4%5{HY@6Lfhn1!Y?L&^OFFg
z=X?j~+h)Y4aK3kVkGX0VhHCkNR=+~pJc_l2@{&ijOT{=UDm5x|8f%jg-I-^hfY-C1
zYNyHp(`T;n5Uz{n?6%~N@?L#PQtSj&VzmCavfmM%`MPlR?5uV)*!gTU;iS<;;6t+`
z)oimQ-|m1bM*SKuaffG~!-wYlB6i1H4T*m@jE2KKyHFO%m-5j&MFV}8xp`Yk+yUP-
zbC-DFyHxFX*IAzomtWQgm1}vR@@Abd%uD}PJZz6y>w_{c-;MhO3zgxIA3e$lJm-gy
zPAi1|)~PI}6gldE?Lo_o%4LthkARgrM_GB7XX!jC=Y<1r>nSjMiq!$f2IrFqQ&{2~
zt19uNIh5*0xC0UEm-KovunLokuXW7!qXowlM32j~q<EN?flJyCcxHMh>X8Z>3NmfN
z^(aIOZ^Zv0Q@$!?50E)KEGR#`@4BW^#<hyZT0%VQrpmrBbv7>-t*6Di#sw<695V0$
ze{oHDp?GL*aTknM(C0rtC%q{=hh;ZwQpCZvWb#tQ<qRv3@Qzp)qr#H2Qs9R?(O=%`
zQf}11l!Jub!3{RvA?5p%h}UH<zJy(sw=}l%X%$MtUG?uYH@QzE`I|)kCF9I!y9?MT
zxz{8NZ{#icG@n0&PnTu2D4hzO%4<g3<Tn5~z!Z=6&z<<07ZF$M3YRj6H6fJ+ubtea
zyjj_Y7tf9WGBVN&A!C@M3**{h2(;o4%Lff-=a5}3T0XcQbI90}Cp86=Ep<uD_ESd*
zHpF3eJ#uw?1Ywy{Ju+FRY1INzK`8-4v$k;~aE#v#R$|3PL8uca$*2oj33s&b*JUr>
zR_Be^%z?l1-air>lwcw?Bd5rOSJzFL86{QaWyT5(*Sus=E<#E(@hhfX+zz`9UA6PX
zcy$!@sW9va$jdbHHv(OYg-S?1n#CMC>G>(dLXx5#q|;A1($e(If<lx#7v(v9qM*Nb
zfKuXO(Fk@I6uXGGdX$VP$Fl?*UwTi7lj$uy6`C&WQ;Bk2PZl{sV-yyIc?3G(hU^aK
zgkk+~s>~zd0TKo50!1*}Bqy#}G?b@#f@2?S=DRN3l|SLLH+>d&(?+qh@bu@sQs(vA
zpFkS$N;$3Qoku60-iceecSzmEVeVGn=V7TIBTxN}+n05@J2e#${iEa)-!H&cXU7zS
z$_EOB3%gY+8+Y$fF6s2_@VbmUf9X(BNf~$E-22@{l;&s0?OF=!;005NQCzKLlr<a<
zxZn(q!qlgf5$ZJWPKR24r|eqcHEn&<MVHiF>4UJW2QjYOj7@82uOjT1ZzBhg5Awja
zZdXplx6qSUrWWbUbCf!j&I8g^fV&ioC0@d|O+4UJ4_aGYzhqo3B!#aEabGGQ)M@kJ
z0!3Uy^j^=Ffmgf8A`}!AD5q2W*qqPGHl0Va_d;3~##loFUl~?PSf-IE`>qGMo1a-;
zR^bBq7C+J?LEP`+lFGEjluJR#v#Z=?Vc{n5avurj&zGgS3jXiT-|~cL>&az&(*MqG
zUM4d*T{o|S+c`RYEKaem`aFXt&bErKi#0jFT5raCNbE3#fI^{P(<x>$N8wo=sfd}z
z&PeNq3r)Y^+b;RO3nQ{eT-E3m6rziOLg3g*opG|KmD|N*Di$tkw6Sb^$|q4hdqRIZ
zCxxY*S}7wnVzePkw3iJ$!dryII7pX55#kYm#(Qi0#OY|29`TtDU3r_tz;5O@lc#hD
zKVfCZigzps`QuqUqJ$vSv)BI((=%9lP8?|gd4KPisVGK8@allo`7kHn0&(={wVCq|
z6j(K?)WYFzA18L?UuKWO#^R+gn|?>cc0dX~SN_W%w?WHGVOy3eA$C@#QBe0szr_Eg
z-`NNr@SXCZ8qu-TLR3euDlOt;XXdm{reW*|P(p|{_|<pv!m48DCgJ8OJ}rpNUnfrb
z=g=ywEKxXBwytoU&b!yV(REn+_4_KQq{TDkDjXvWd^+wt?N!emce>l?gu{KwnFoH!
zcO@Tjm3~vG@LSJ2Uh@hj7LT}Z6TU@<)@9gu#}WmP-HBx#umje09TukT@T~C@SHe}!
zvrC9)!(&x}|FxEhQG4o&fVgxa54nJBFi)F{^+Bb_@5@}U3&uFH=J2>h8y1Ik>Q(5Z
zgVu@~<rc+OWy-do<>3B9<4_-z576o5ZWP;fyP}8sgmUh8q1-OtmJKUmo$cXyI`v(r
z`&d4-%<doTLj&|^)d0U(8;M_2omH1ayo!f7;NBO}<1+Q)PMN_AWs&%*;H!kV2E^$Y
z+fH~1UELm4p7hgonX8ypo2)jV|ElY{u<HLQ-^HojD$esT<E(~vx<Bq8#(`BnH?-8C
zOX^XF-}Y|#t(M{H`}<%929dhfC?iWcy)UKVL{@{Z#AAejK#)*7UWKHmn<{|PNwGrz
z^DjO}3HgErV4o*k1RMc*g!73V8;R}wlE#Iaz;JY(28mhL7Ot}4D8p^0juafpYvw<S
zFs>0*2m_cncqy<X(@_snAR{tIItnlg>;hpbA)81bg^;xQ=ajdiL4o4h5ot3#@I+SE
zchcyRJWu0?@T42BPIZWr79mHL3S<Rv7N2q7`e_Tk!2|G9kVlv+(0%Sa$Yeaub4uQN
z7*|2+;#jC8GHFp<o<ht4TZ9#uFJ8PP<{9SU7i@!ko;4n>^T;UL4ue1Oit@*^)(u1@
zRte49Bpm|RPqVva`T29VR=O~m?V=p2ba?+B3fBR6UNCj@vXmG`f|iZvo4BR9r<GJX
z>GFU}8A}+i#w$g$c%wzkx)R0LID;o7TypZLufS?XbvZ4?zWQZ?G14{fFH3P<nDryt
zzX|EoN3BU&>2G~;<lp6NG~r37&NA>>g}`U69menakTCe-qHjA*J5{IkjGGQRb%DB1
zz1u~Jb-F?&!u0aft6_D!36IG;_>q@1*j^mTTW4MP4VG-%MM0p3ZL!{?__e5~4byl!
z>*EH~fERlxAJC%j$*GoyQi5$0mA@)=IqRT)GA;D*97P1$d{+taJv<$8eAmjL@)661
zbzd5|3dh<OEPhd%ldt(%FALYg+2_70JgE3nIqE5E@+QxEU6dZS1Lc<TNo70Jn#e6>
zrb>qcfLIYJm%^*cT)Mhkn&zz43JsODyh_c1cS##eKJ8D}ojDc@elC3FsP#%0*_x;N
zNm<X{S068fw;JcB?}folWB;=8-bd#zncEW2Qg`nT{W`p_3%e{w<J*$jP3iH!u@WM2
zlM<3jRU;sQdY*<2&D}1pG{4Uo5;nT8zxpaG&|QV@E(<SL$CqioIw*Uf?K7th_A{b(
zLE70F0+10!H&91~&at|}{!P9yQRi_u%5b#N&Jm%P)H;=Mbi|@O8u=bVdXwoFw*o4R
zWJU^$cVF@hO`B*ThGk)?abz4<rauT@t;n(=bKN44K;!!W6w;Z(X$wb?PD!b3IYRPV
zG@hWSr9roUT<av-v_9{25<!sRc`}JtGW3;}?bhjxK>fZ3e?GG!X6*pQz_S<Z5LDt&
zC)R#E_%S207cZWpm@JeY!xZQq6RlGz@^neQL=xejN`#=zngN9D&(EJxp6s3hzGZVr
zXLor2FbKVQ{WjYmokBBRFIL?GFQ;l0*7~3IB0ol?3UN=MwY;>{xFy)T`B^?)X5oNk
zami&IopC0?M?OBA-`pj=9)#tV->ZL8M^=Ixmlf~imHf=x%LwG75Z0x$=Pu0dyp1RA
zUg^ZB^ZNCu*?2lTKih+KM@1%ehH^@J;_HlSJ3OJo^P424gvf*9ZD<$Xh(9UWybIw1
zcpEIL-gwI{9Ts?=oK(kYXQ^|F#yp(JQxU4Ej4qQzsc_oY9*T%St&=KNI(O@sURwPM
zu0@cYF0rIhCygY2INM;XAmPnWLP93~3`*pOQ?}M|c@=9$L9phA#4`H|Qe_P2p_}_C
z6Pv6!c(p6mgd0G?<wsKR6DP}o{9T8op9-LLwc7VM4+MD1Ol6Yt&{I;ihIBaxEa`+L
zZ?ymkK5beH(LgA7V8LmS6i*}F(SiG2Qzu<9=DnHA%Bc-P$8FNS7x?cmLf3^)MY{6k
zGIv$EPO(?g1rGJW`q7S7+mEyRfZ#fiuCcUn8n8M`HGW~Y1?2G~7MWtAuhULh_(g|n
z<GW9%WXG#eRe@*)#D$s)Q5&NpEmy<4HpD#Kl4qk-330S%N96VW4?kpqr=xeR2`=7J
z2~nYG*Gh;T*jq;FhS}(~%(zpe7oY9#)j1{Nm=0f|3tgo^j8Pbj2^%0(Je5#kmG^!Q
zH^$H4)l^tAhy)w#6wvLUpuw=U$69f~&yLtmzt7+Jk;BatRBjzqKr2v%<>`s%>>9Ac
zQLQm70uY?c!b}`-IuhK)9G%5(z>Onoh3?}gkJk3?ng2z$)4FKUgRGvv_&Gf4PG3Rz
z%9A$X+aV@vVcB75d3wvI)usF8^B37h`Rv*Eyw_BWct8L7S?a*UhyOpgBY4^6V3`VA
zKDz_T5uM7$Cin@9$MCoo9oKf|D3+sq%~x7=mJ^Bc9R6SOoWPYI<Lahz8R3DU|9D@$
zb^K1h8h+jTl%{d6f|`;DyD#fjcgWp2N%K5^;bm<M=~DLcgl8#}r>@^H{xr=)xoz=P
z<Fhk9A3$3?WUID|I4uj~nU6cAtW(0U^K5*=;}O<%(6(#eeR9Gs85p*;g*AoxCSJx#
z=Rc?uj&)nWc6S-o^WwS<@$@Kv*D5t^7blBD7P9f6vXXZ*#F<~#gYcjd6)WW^wept-
z)h0LAkxwb_Y8y43hxOI=oAd!$aQz+i561DbxOyOA-gT)Y*Gt`yQPffE=Y^~4@YM2=
z^=-(cE>p`@`IGw3U7F`*ZF^JBDIXn7Wk3Vj^3^Lmh8x1y=~wsM?)Eca+%56#QZ*e>
z0!bRBWYBUuNWb6Mi!!gWzlsxGNBVwMykElqvOHG(e7Dw4WxtYf>IQzsnT~B=#Hk7|
zh^jBP3e2-;=8kCfe0xmmC46hshQ(3qbZV|?`10$oXmlR2&S|G~`Zhk(loC=F5Cvh@
z8ek2uLwajxi!@LuI-1F8p$J7+#T(b%GIoSN{J>XRX{_$>Pzia>!mkKZp4+t&^5!jH
zs|3D+K_TEmG*_ehtV#Cb+^G?#KeQS!Re-Q?S40>G`Hay5^3+~ZJ~%>AF!0ubyYkeh
z_ysX}DxS%^!AA!#q}^wFY8PI4_JGqDDp3lcpXt<l*O7uvh$>C;&XMD>rzrMzLjp}N
zTJp|3#7Z73;G|#t#K}XZ9ICpgP-WphygazK%QVUBoL%rU(<#rjX7GI(r)s|a?%TD~
zSEp+)v5+_wV&|X&>9=JRmbFV^ZSBSS&*A%<BfkBC1>^LDRsUG8zWwg+q+xCz)A*Ce
zpHnA3kHEE)bS~Z#R$WITlTP@|y3V7hrNeSo`FQ#IWnk;CvZ-M1Q!kCLA~NH+UdwKT
zFMHNoH|ER2ZsOPFIfs)H9_4%;jH$taU)LksQNwKI!@EjGlmcm)(UZ8ONu1HC3L{%e
zhY}(_c5+sM9h8uZGrqONK)@+E6~#ByJK<QK!g0-q^~HMDYc0ZG;8}}ollsSE2QE9b
zKg;jbH5i@;f)!w6yB1ZaKBf(55s)_PD)?!gqt4pybk<*$1C<i-CAfGpXf-Y#vT5cD
zUNTRr3JQNrM;iZZxjL15m4*}BPS-DR%NZs37k=E4YvO6=FD?tW$WMOdk~Sh0ou!00
zJ-oGE>zjH}sk}A5_1*!N)(_XYI7qW>#MyAmS-ChVAv{N6lUHN<%y-E$zOL`ZhXuC3
zh9~J-a=ce*K!S8j^N8T{#P9jD;g;tN&^vHvVK;qL`Ejk&l;=$o{sQq9Wp+1UzXbhv
z%ehj?vsNTl;ao*q_&n-tl=oS}%jnsfd!}l<Q&M=?>8U8J*YtrRO5ykQ*Iy#s4%YtT
zKmMae!5kZ&5i1SJ%x#cw-qakac!-xB`lM_yQhxg*TXFGsor0qv6%0G$AFvEKvN)oX
zR0#g#n|~k_hlPa3Zv!@=!c@FvNCYq)uDed08u}LDYsY*a9CuJc_7TJ$FaDTS^7=nu
zp{#<~(V3mNpOTMRS0M~83X-%4T19=_jtBDds1ci(yyG7i4<9Did=1Uu7lBTw%7r}o
zim56sA}T}D=R#0NRw`Ey9<ZZ`%T}YQ_~ZmZPMTfl&e|4p0kDEke2u3I{U{{n>1mO#
zI1;-@i}V*P6h6$ljkB}Q5q>H`9+mwzipYQe_kW?3yhu5^-s7idKat*1S_yImyV*qf
zfCnm3;^bmL*Lo=6Pfy;g{rJ-}jtc*dkA<P=(ecP<w`V&gXL**TN5#l*dervcQ_8YB
zjaqbhM#o|K${)+O6~B}RIJBied~6`X+A}}?002M$Nkl<Z@=D#qFOK&(nQ@IvTEa){
zW@S?7H*aXhmAR`B<KKCRhtI}^_9ZuP5vEty3E|mRtoKn6cz4^a<*5L5I^qI;w4#u5
zMnT-*3jV~&4iV~-{E`}UI?82h?GcN^zk2#**7`U_V|=Y|^448iFtd~%@ChtZdmvr#
zGhg0~Z{BvQwH5`%m%NRWT{~JciRgP=tb&vqI9TRU_QB2W#Wudlw219Z{%XZ5H#&)i
zk<bwY7NCSHzIpfGy1h_-3?sw%mrO8`rJ<5u+rcrp0{@&fEv7x8z3IvVQSI6ALID`Y
zg>Ef7(j5SBK##wrISNj#<Iq2gPl@AJe9O+VvCiAhrBUV2dMj_Ox5numYVM$MD$}ws
zz3Jq!N~6ykJ440{UQGNA)_43VM&2<o$%;AGrkB>@J<-Oq2~vbUtiV{ME9s3lobRh~
zI_!V*U4FIfYdURVUK0;zyo~=md=_SsTf3F?mzBX~e&*wn2661N?6Q}hRj2)O_%#W$
zD0&IY5rdC3wp1$MXaaTXr@KEMxayq6m7Y3i0>7%s5O&Rvxz+j)%(-OlWoJW-2!XTH
zI(4<Ta9UqooTxB<_Kc|l1lU_FA!Yff`|**ZGZ*gJafwq2pJGOJMvk;J@!R0-2+!$A
zRYL6SU97qV?k>nwaHp|@*A6F*+M(OYOC#ccv>yb4qca7o&OG6nJdtyR#?#;MvL5m_
z|FI)cBnr#57Fs)`*J%p4n$c*s^a0&Dd>8Dd0)pFxRt<35<N&rDEHwb<oQIUDxDEVx
z60Yw~ZOIp>WLz8?dUp&&q1j`R><P*m>o(lR`})nx2+bdV_@3~ywIBI{t!r0e0++|a
z2No=iP(DBRV_~&B#`8ZcTk|?P@|+ju{f}4>3?1^<t489rK1FlBIH+huz&qj&Z0E(h
z-k|*OtiP1W3zZDtCpUlAvj`8KKF>#7Qr#)To&hA!rW2M|KacR~Heg={V~W!cBkqz?
zM3`wCf1lG9$WuS9AF+Z!vv`=+4m*x_+l29hBc9cnSGY7JFVcixMzg6>JFVw?S+8?{
z?SQGa-YI3-0yI3f{^`OmVbp8N!mxD2qXV@LtNhv?&e*qWy^fM(8t{|!@HEziqRk4P
zsa*oZQ^8c%z}q}bXnJWB6j$K*L*4gQN`@g^xe3_+anFsngykpcBKUZ9L9Ob^I->~J
zRn%#$g7+P;#5Pp{=NDYM{(F9jyMR0qRlk%S@A4x2VSr@$NtYH3l{A$PFQ-)k6?j_w
zjAy-9zNy66?!9!v+UG2tO!J8A_#@|p{}MkT^$vh;gh4T~pwFH=&y?fHd;YwNj$xf<
zHSBlu4*$Y~=HqA^&9`^TV}w&dKbS>x2=nI~175|UX#v2TCfhs15wcjxWg#dDnAZfJ
zZDe+!S{lB5`c>9{xC=%Qn$R8Hy~29oy8=fA!_M6u6;7)tJQWI#1~kL}G(NT!3kZ$#
zh|z%!+g%(g6E;#|9AyniHC$n9Qq@^L5f%sp8@AW$BRee5$#{j}*<5=8zODz!P64J4
z-f%Ru!r7?^cNaY3{E|JVkvuq!4%3l{BNBCwOiQrHQ<r>XCRvxbQYax{249Rb30Q=1
zo>6!dEYjns%??{!?(>b6{Zkg>GJWA^%=6JPc;w<;I~=EI+--87>)cNTEA%-4Dhde&
z=bwVm{n$D38azDHBa3UvN5Skd>h0u{ibw<~__S3a3WDoDD5E!@zm6jE6V{AxzWFB8
zH4>M+C>!Zev5R1a2afi>{PGL<%6cHC<X*k<Yy``dQRXqinsZpRa=gMMt)<89$?04j
z!GoB7!pqlnd3RI|aJJ}Dj`(}&RNBHzJ7vPD5b>Y?7_RHOn`eK9e;)m5*YSC;logQl
z<RQFXSZ9cF8uk>wVTjYrMS0{sBVg`<#a;dh!@6$SsT5xfzE;O%FIozS3Zo-wryx~O
zoR%`Zuxvwyd8xFSy9-oZ_$&X7W83m>89nj{cX^9bY5(?ncf3$8uWG9{5sOdWai;TX
z*p`tY!593tPdo$=;jzIWfs<Grb5w^bUey8jphM)0>olv*b!2H*;%-r1zythi%2Phu
z@g`z$N=*@lalFe1l>q3_BY&CxvHk0Xj}dXwAh1kp@SrttgiwW?vI&U9HKKp5U^qZ>
zZ3x*>N9BLlf9kb3_zoUU`3TEhWxhx0pd6f1@!4r$aa7^Z(kU<Bvd&AncEq|hr)HHy
zY4`gm?hKmpNe7tZt{=A!M1cd#rc3_a)fN4BO?`d3@S7+TdX~eeOYuv28KGh{J(odE
z+cWKDfp_q0-0qO{gOa2jjrw>OKvAIOSHl?hc{%#r?-NgrH~4&)e6LE=`CJt~$1`!2
z**N7v%5s!xm#gJGtqCdM23?hn4OHPwyP#odhIe%96o+dq>@Z%@$TV?gssu<8l8zks
z^Fvk<s5DF)=Q+~_S}U9rby`5>L#v5vK5X=MPzv8;E(JqO_B4J^(-g;)1;|KZ%(5yW
z#8rW@;5VE)LQottDlxrEg~H?XSTm_-DAcNTl@SHT*(r-xY4pZXAyLsdM^GvhFY?ir
zd4nJ@Y(hxEkBMtM$ZFrK0`NS-Jtrfb2p_TY+ZqaM@pOBx{HnDxeq@XF81P5%p7U_Q
z0b^z+%rm~(A>9#5K|x;d4(*nWg0UD=!hVYh_wZ?fp)3(xS#aP@I!y3zO^?F*F~{kf
zC?`55YV+GX;2-5I0Ie%3R8D0%y1swN>60ij(xa8+n9d~&0*P%JJH=XG;t}A3p5bq~
z={DiL&T{YRpW<WjI1SneGF8I7U*?9tp*aDyn6Mmk7H2SN5k4Bu!sFd1a*w+V1O3Ex
zx{TR4aJQ_iXDT3C5v&&(S&umBz^MZ+)-)f(I?fa};CH$%$G}Z?v;orZK#NE1I}LU~
zT{=WDeevpfI=k@4JcX4sMO6#3#?w6d8$_NhVtGlM3jaFojoRSR%fMj@!TZcv*20+<
zUD~meTZtxUQxm`?Pai>qD2wn}-4%4<H29aK@(zEhjy#&O2S(*-UIQ?7RyuVm7}i(u
z>-Rk5rJoEFCYNo@%eqM>Q<;qN^Gmnlrp&V)We`LAh!qvwOe^0USjFmvjPs;al@}Ef
zm6WZMS~p{RIo(7FM|Mhsc*Yf~FsG?#Ja2I?CZ^19k?QiO)PXqf&G_~+VE$Y&XQf$X
zJW6JS)=j9FQP|`!MZUp{urABzuJCS<<u4Yc`4C*2pUvyerOUM6xi@W{eipR+*f9G(
zLfb0(j%AX(8ul`v#hc+BXEp!!U&L0=b|fEH9YX|Cr<|uC^ApVbw6V1(?1u69?n62r
z7hb*if9&0PcV0(wCU_78xJaZ(YOTG>$M*E}%<1|5pQ7iF=|0Dvo_5(KsiZ3H+zAlm
zJijM1UtEBoL`t$<Dt7|+z4zu485tQF85z0U`}vH08qUKUwN6Q39Z;rcX+ZJLH9D{7
z>6mn~8kwCL7CU~E52KkAwII^^1B>Sf6C=6`-ojJxQFv1j&~uK32X7r2b)=<pbeSHA
zz{xg46Vs2LC|gGQQFfR{@5b`LT>52>B6r4uT|K6joM)muPlBKM^Bz$iL4Gy6X~sxz
zC=TEgI2EA50R`7VN&V_*ayxl8wx=Fv&4ntt{>GS(zclB-@7=!}TIKMvSa^<~J<fL8
z>%lKaJwLg-_vxQM%7WEz_WtvK{y+AvzVUxXAsEWS<xAFpC^z~VDaBfo(cQ@<>tpO-
zG2foD$KQz|SNE?+;ISgvjgvZvuv2~t4<%zLiK7!_>(Rm7yq;~gv1HuNj*7cE_vP!X
z)gd$wqHR6Pmx1wsvilg=mFz~ic{NjVIo<N@tOvRhWNyEa^iNZ!$IRUkq?FIpGbPjK
z+U0xwT53ZF*{#!K_msM(sDPakrZdYZKAmTuL*aHGt@|iQDI-@u<Pa;kC&^%biMYT5
zkLw(z2k$O)6wWWM>QDS?FPLE?B){G_ZozCr-T1R!G-l$vi&wgb4po3tJ{=KDh<wp?
zIrWLJICbm-pMr{x6#XivNpmUhOHtKc9%VuAlgAJCF2&+^tvkP>cvzU4zvC<%wm{SZ
z(D&ZC9cAQd6~}9-hyQ;6zhxBusIXbdSR8e@Ph!!8zwf!2o(vatd;3YIKq=49o<3>2
z+1rcana=Z4)_7QSeC1gdQ|FTj_<AT{_%!uNVa=lKmJ7p%fAP_9rFdyIJ?0IClwiaI
zFA3u{I%<#3sX5AYVy9E$p$yHpWHe}!M&0Kp7g(*kJZrmT0$+Sbq0&~chBzasgsBhh
z-Ct<asos>&KxxDuamImg`9LR{wJxO$Pd^<WrVP7R&N)7x#aNs4DTFV-{3`1~#u}ZY
z4Cu58Z16bi)3_@c5M56@RIZ?0*HXc8yh3<9Lw>qJue_9QrCgs|w!~YV%S+>L)6Gbr
z<LBiTj6vn)n>##`zO?2YdNp6q8ys8Ec%4^V7tW${3(m@K_`cq6EebP<aun7Emz9Sw
z*A~x~Z_-21>$+{;^4-EO9%u#r79W|!k6b)Y4?%bg9JR6NQn-09tpjVv1M#bq$HuMj
zm3|Y{NBsd<yYRewcC9ij7`wd1N$Qo;x)&$@;RVij$yV5G(Go4M#u+~wuV#WC`Lnpw
z;!h1v!`B!!%HwGE6oR7A9zh{SgC9sFo>6klkzE|uGc(5S7du)gB^F8|{7Eb&(h!W;
z;<T_Zfkj$sk@6!BQ?<$w*E<1|dpb`7Mkn3!M*up7kq|eUF&$y;g2v@Ic{@I?U(foJ
zjBHE;=%|Mf4dIg&WrSd@f;6Rq+zzr;4?;^&4C{>fa;EHzkdg;s=48j3ai3-DYw$`5
z1K-z$qc)^A9LV(G;k~`jKmT(SwO>RTOXqSa3ueun`Fd#tx-z1{__&{uV9FkPSMf=_
z!zdZbA57z)$CRz3?vy*-L4wBm3TJ_!i4Ia#V+(%c!*?k=gb3zb=-)UBo9CWQlDfCB
zm*8~B+>x!hu_hf{iZw^HT<WL``jQcvb)=4vaTF#*wGh(Zlaxc9TllOl-ML-eC#*3A
z_P;qaPUGdLY`1*Gn|ZC1K$~;?<x{uf6n4M{r|mRM)33M-?z~S315ADwjoceH%RBga
zSO}HhTB+tAxTO;uX8JzlM|xLB&7mEkgHAo`Ch4Vc=|ElL=_IvDJ5Uazb>MGdXeRx&
zv|FR}2iY;x?elnfQ?}MkU_XljEsZZ&XInwe1u=D!ig)F>{o}xpcRyqZ$*K5LZcBsY
zHR>#Q<FyP*SpO&mE@5!vM!Z?xEiNNC@mJd)_nrfi^juJ_GShQP;w6s(Z}3fC8gGis
zqTgZ!W@<p@g%20)epuv2I|paL8utMycqT4du+*^*sRgoV#Yf*X@i_6wiJ_3C@)Doy
zb52l{j_%;Ii|=6vX0|~e#uB1FSVxpe-KkGj6f$U#B){VUGUGA#<~Z-`x%kJgceD`S
z)z}yRta3JV(r*`Y1OLMC74clUi}G9f^nNb>0(1U;5uEdJE_y$cS0Z?SDJ*5YsGR5h
zF3RX7ek*^$@?G!6VU->H^Uv{X9BH=UGakQ|Wt&bW-{Mx@XQHh18_(jo;KzTZ+u^*$
zXQi9**Xu?1Rh}J<llPh3F&8CTevqj4GCK|eOXJnqRuI<AsBuQo$g}vDvXGLgaZJy!
zqU7+aj0C>PC*?m&C(8<OKl7o|`-Hb6*{u?92Op&d5F;ppN4DN|eww}G%e{LZxKLau
z8o|L`C!f1s=}JcHz7>IPB+(PZN*Me*H58ohW)x|)x)I%1S)4hx)j443T6eP8V&`Jk
zcIHCA+u<w*7MQ?QhND;zy0<_0>IgJ`U+yG6OrbcKNwF|n7szH*`OdpntH9mKyglm-
zdOpl3$-0bB{`g5oOi%WH^P69H?ac>2|M}kaD-LxVLYCl4wt+k5qHE#EmPrbc`TH*(
zeHG?KD}>5o)Zp%pYEO>Fw_I#vRzLS+%_96<6w1d>@)5!LtrtR7XF^%>m0;GRcY^Qj
zJGb`U$*vb8wZp?GrM?iHdSP`i?@f?elfb%U0rS`*J?6@XV1!Qa3t#M~v-1_zc222}
z9kT}8*K#diQna#9Q#QYPUkrKRb9UshV)l+_fD3#bdLPk*SG|WH|90lNTOUQa#T$6>
zXj(*8xD=*~z$_i4NgVv|8!Kq89{D|vf{y2u%e&45OuH){rXA@->DnP6Zc9Dlx6qLi
z59g$1LAspnwYP4+6@~x3z2E)K-{kprw!eQ7`0`6h6Zgv1n=O+aR&a3!jJ$8;%n^0U
z;)5o4vv~Z<wHtefv4maDXBUoH9@`<~Px<2;Wn(#{Bt!uZyj2${cbJ0?-Ia&K|MRi$
z<7b3TvhVx{Ov-Z9)dcoPp4^8L8|uy~Da@pL9v<U&DHc6NjXk5Z+5r?NK=I(v^YH_^
zhK|8|+ksOqT-HAo;YmAfL*nCDiwq8)q<wWUYX&1<e(_alZK~OY=By%+I*g@@>_J-t
z5xk>a;(;qUc78t!)+x889N)(&*Gh+Yq1CF3^Y<0)ukF70?Ayw<Xz?m>zYWg6dsymt
zw4CpO3I3SwB6^(nD_?1Y(kwp9`+2w0;XO|LNVQjV7&cKC*fz*C%rPP$?55Bo`1+?U
z4yA<P8Lc{kGaa19Wp%!b;?e#4Y3N;j9>LnV*XE@|9a|NB2jVpM(e4$Nz&KB81>4Xt
zZ=<lgd+_2;q9h4{kJ3H7k`rB%zqJ$xPb}UW2M`%G5<-A>zBhsbAqdzc&y{j2cWWR)
z5l9YBDxLYuY&y>bW=C%Xdd%XUtx0i~Naya;X}_82A(@PmW*<d>XLnEN)w3Ov|E@9<
zOTI3hQmr;IkyE}hdy~8<L_<i|gg+yzlV!o8bm-V9kLsYK!uVDA7VlN-!6-@~#m%lB
zyP;UbC~Q5FJ^8>HZiJ`K!ANc`vSbB9XSAx3O2-*I%1euQ&F_c(ENr)QzYt<8XohT`
z&*QS*R~+x_ImuUkAT>|nxyqZR^RxoZ@OEg@yJp-q(m@OLXnON`=(mFre{q)Sx;zwk
z_*MJ?=d|rju%#{0FgoebpHd#4Q+IfbBKu7ihhL^BB@z5tTsU=_3Thqs+GPU;B^-S1
zc0Zl#0Un?6%2(nJN(8s$c?pg>;|w5@cJS`FLw%3|Ou4&8ar&_NY4GM<+=*&^@s_`x
z8=wktKJJ2eUc`6e$tPqgy<4u?wufArG#=t0vGb6)+8G)*dq5~Jx!?$(X9E`2Pjs_L
zzj80MGO+5p5A;sEp#W8ZOged{24=Y?{6PI)?hLU4>p!SdEB|pz|HCSMOE+|U9k<f1
zyR@EPT?r963W3XlLuU+F-=TpKDpfAhxH{@!jmVz{#Pleh^&lC;#R_t7?~BjB$_c2M
z3pIj^@Y5+-luA%N$s9Ik!K3v9?_`~Z(e&H5-wy1op~wi)cDAo`j)*lR53;smb*M|g
zxpzb|n{ImM!lWz^8c(zGop6W>)EFIQrn9^sSUSzmGAc1@ldsO!yzNbAcVr}ZC!Yg%
zLM#Dp_X-OON!Tf$lvm#Uh~3pJC=G>=o<3~eUH_0yh0;;jnd_C;z|*wQmA#)Q{k5#A
zxbx0?d%yYZ-|YSKKmUI34}bWVT139Op7O@JU;*hnZ@<&2qn<$!p|x)nee!*jH95MG
zFHbT;jiPcV>oizRe7A*VhVo!!-K~L9Xoeshi$)*DJgg2LWQ~FAzPs(VmBy&$3XCXd
z*JIWA`LBMR4-($ld;e#@XnF7f3yy`lzOsu(EFZ47a#qEyyzeLN(+J|D?8+cydrn3L
zba*9q9jbin@=%_C4X1u~E_p834seC#-jGfBir0I>ThCskfp45$KIm|qtL2GuOF>xc
zn@;l1pS?i(uHOMG@1}rU){9sp#dlc7bb4{Kw>}6xDU;W-egKS4Yo;9?Wxd9gtjD;T
z<A>#O9MivhJFwJ!m3etl1n;nRBwqN<@%`HFSK;xy5ALOHWQSF(C18j9+qU}cn7j|=
zC~c?;M|PowN1jt~BNNyaGK>hn<jcFk<E*nbEb&4omQDO0aLt2K%gJx+)~hlsw|(X}
zl=;A#5mq=%b}q?Lp<$^W^kAI=*1*YwKe0Fp0A)`YrA)gvr8?GvSX7#{p&J;Kb!8fL
zWj9@(@-DvJZFoF@iXT~Ks!|(t41qOYaB^9{V-flx`H=MZKnc)WeV{jjS*W}#Y_Wa)
zpYwDs+}8*5lS{F91LEbE(ETE}=p+7mz9`;L!EamU*TxrpRwJB=(W7!|(QZSdl~-wt
zSI?#0<-h#JnVzqugfvYumZ8gN?P(gunEqf!rtznlr{N~TY^WM*#6lXQT@#(UiLz2t
zb?zFo(O&1YlOB!=pi`y<=yYq+k5VFzxL_ZoK|jsD|Bd`uqzZrt24>*$*&>|a;64Aq
z7=JpC2KB4Dm%)r<<zOLzA3+T-LdF`BE*{G>3xf}ZnS$x)HZVK&k<njA`+4tHV$>E^
zxTcl};ZafXK2v5>7nB|FKL7ml<aMpW@F)s`MTF#<?}5O@8kTl{fXE_ONLayTWCiz2
zzL~=_aYWO}QAB*Hwu+xROgSwS9AQI%dLBEG2D1#R$Yoe@<b6F`bM1!Fv6*&jK@v{6
zx^iWFysHZqDPEx`K9K$-0vdg)bWGeKw_7|an;*fUH-|14y3nvhabNDD=N0dwILj~h
ztyM?;wWG?t?Iri}uK26W9_NL2>XgR_9uKWw{<+$)-NQe%R^(YHrA-a~1ANq1VXT5T
zjiAAI7MGe1v|C0!*x_ax^kJ0FPTv5do|7&=?KpFrE<BwyT@Mky65kXkUfB;HRmrKE
zRS+vtpy{wF9OT4Mumq0oR@opc6VLh0&yEfvPPZZ5u+&VvMp4l$^}sK8Alt_Af1XzT
zz+auz%CzDv_#;J?mC7acm-MU{L$+y8<PMkvPq+?i2A|}Z_$!ZENJ6|p^C8nHpW=vH
z*J8KO01m&4erNQO>7wxe<#!<k7TMPQ!bpFup4$fVI%UL%(k^~n{C)w9^K@MtyzVa~
z3CxB~*sr`M1fUAR+}B*C^DVBV^J!<2hS@E+xp%>3Fw^ODC3r@54<Fr&<>OusA^W<W
zI_ro|a5d-(gdk-~!l=!OqDK0E`?r7Fj=yJ+M9BG4?BggWzM^~i@j*w81j-_q!V$<7
z_zDS+ESQbliC<q6b#$o&6xXA{0|Ojo)ZtS?2#fnU()`OD8Sg;0o@AO){aZP4)k&0h
zayG`>SsUSCu`zEu&PeD;FBTw_KFKy@Fb`rqpiDWdWYJN6b~ku;Sxjh)tn)oS_}K^T
z9B$|R<E#yFmVmQ6I2I%y3H(*k^ee>HljONh8B#7Iwg>l<ceeUHeAtB}87)Wo2;J={
zAS6qpBq9*+-pvk+;QYnsU!;C=Zb^=Jf0Q*LhnKC%8FWLgbTye?u}<cfJ%8f%+wZ0R
z(-CIfK(w($+0Pn~Sc%>o>p@J_sLv{M$wQfYC|l|XeRZf#2bGC>LkH6kvoxbjukpL1
zt#ZwDTUh0l^YCW9i`Voma8l>KO}mmUxb4(<Qm)ixXj^_xIh3W_%X1&rzSPMCAdiLT
zf@u_tKdRWLP4#S)_+1TsICNUdhy1M}pp-jb#?k+eV+FJDG<&6scVa^+bu2H+2Oyqf
zsY+T(zUNzy-rBpD^tZy#9C(ki4k@bQ-j#d+F=O+++m~+bcm>{%eV7n+(?-<21L{mI
z9H$;FX10E$W!zDZ(n0x7JILUH(%wbS$#2OD#XO}8MHkN6U=g2TMe+2<v}6t$Sai+y
zu90I3-m<3*>bqm2;A@Qww2M0;jwc)FLa3=MhF%uWcKRe`qvVly)^BK26u#Db(!kGW
zDkvI0Tf;q#h0zq3cIM2FHzPOD`|Yfs@*N*C<T#dG1B@|vles*H{BfuJY0tHkg$I&R
zFIGg<c>R$h%!Vr;t(X5GcQiv+cTUyUuWuLssN3ZRtMqqy{Uq+icvFtoLK?lK&7Iqp
zo`3mu6TawG|HN;crku1+ToheY-Q5HKwfPw`8leVcp<Sd!1X`4d09zwlVUXzvDT>3c
zAx9aZs1USm$cZD~Dkz+-A7E|JxsF58a-@11V!suWxXL{m(k#?09K2evyfYcmI&`kU
zYvvU^XGkfx<=~KKktJ=~nNIo~y!H6-QE(ed25SeCI9%Z<ts}A{5BK)1bBO3BT#H5k
z?kaOr){)&Tf3g;b;3e4G*~B8_pfXb)PLI{85%kQ$rjHJyh*&Rzjz+R*bdXL=raB$v
zaQu0u<KxCC&a?mpq$i=4BEiCcVD#&ssUHi6eXqw(C+6nvI3Sosj4O$Prw-GlGaJoY
zoPrxTC`)lc(9jUSL^vBcTMV1p$s*U7umgMN5X|)>xdjPco<|8T0KUz0J?AdXYFsN-
zVT4{>8(ayZ(;96;J6QaWuJvDu-531n*F5s9?(vk4hsl0*L=-Y^#cTSRCq8ed7k+S1
zo%Yx11bIlXI`#tK5tjlelg{1L$l=xJv4~txeZSeM$HPpi*`cOvI^FVma8kRSYE60y
zY-_vE+WF`l3q0whmB*HL_B@$-f%-mEMedxV`-e*&!3kde&_a6CJp<OHfwFPA4SXh`
zdahRPp1<JlP63v0WzLid2vtb@lUBVAS(G==tM2A~?C?^@xl(?_4VJ05{K7Ln<VNe?
zTY&S|Mpz3Cs7#$bj0Lfa?Gph8+dK;-aL&_iN7q$$nN2d}>3_NJpa61l>wf;Zfq$((
z82|e?ye949xdmebZ?T{pJg=OV|GI+_QS>4#evvJ&rG#k2+@q|>H@|I{i?s(Q=~Ot=
zFk-9*Yy@G{?)v(huhN-*Udx4b9RwzL1Q<clPBfz=J1*=(`Pt8Z(T3|JPVxTn|NS2k
z04x+^U4y9|`N>NmHQWl4z#F?p#<A`w!&JH@Cc(i?(sX3zt&P9_gRG8+H;v?E!{_(R
z?dM#Rl-b9)tO;R__}%aRrX!s9vv$Q!7`PkxOKZC3@VfOFITyhcL~yOyI{6b$I-V<;
ze^(|`7{Xlylj~>M4f8=PLUzvhp38?He$-X<EEV^2Ibxj>VD}6m_p4w1rlYPKH)3Xw
zpg+nEiM{N|c@o6~z0KiUZ{UoAB`~WfW_0)IpFZhMix2<sVJ#tFrJSa1lmT6;(1bRQ
zY9~bR<a;pIdidTB3)4xab563JM|B)$eZp~eCmn}QM^798H-!38Ivi6jSCZaAXV<F4
zMCr`Ev`;=imIIGdliY&5{P~Sq_LS$@xEh|c;-AN9bQtN>5~iNYyR*)8blSyfXNId!
z9e8*=K-)P9S$T=a@_{FwhJUmn3y?b!4v%mhfd|LMBi5go()=tVY9Fcl;<L{(dfwlA
z@BR0~OX(cb>9-5sAOT+rQp53P>c??Azy9@avrFr}u3h;oQ+@aT@gKTw;rOdqM8d0Y
z<b#wwUnbKyAyo1m);&Aa;P50CvteyX+2q}h*?nB_?{o08rpWVw-D>&knumm=MW-&3
zUY^!5dB2?9GhJAnytET(SG{DyF>!|Mkw=w-lo!8^dKiPD9?u@UYjtTD3Uev^_#dzN
zRgdI~byEgRWSF+=2mw}~XY+iMYgeO8z%3u~#A;8oUJGq0^V-(c>|(QK?3ceF$KKp~
z|NXI!i<Rj2zyGK5n|w@#?#C3fpFzi)DR)1CKsk>*fLq&X*06PFT6Q2cv&j=Xq+jD9
zAP0<BNfVlF*yqgit0ezRinH4H%W!I2d`aMUrF}`VpS<@@K7|jua6Tnk{IKY?-WQG9
zHkEwJ&_#su9q+&@53lgtUrPxoz&7S|rfICl>Cn<@_l~A%w;gCAK}drnpa{>=keMm7
z79{>{SV`aAFpx^o2r&!s4gyQZN|+HwT|1Jv1o-p#`L2qS1z82UW5fb5hy@PVgc>~D
zM}ZSITHd&=S81ld#}{3oqE|-h6a>}@9xM?QlO>ezr{f_QA4Vxrl%64yZ)#YGCW(<9
zp>M!H1S2IQ;xo8{=eZ<Er@;|uDv}^KB1RV`T;J-sci%De2UY6O)CcESV|@J2Nwv2#
zJ!Y}#qlY>RW@=W7Jf@B;KBNqeZGq$PvQbCYfE@KuGTWr@X8nsPDDfc!XJi^U4g_^R
z2RSkKa>}T?Hu|Q_jQSi3NsnTWzd<8@Q0L;nr=3*FX0*HQRCbKL%Pf(<k3&T6A~>6{
z8LK#r1W9r(UEgQmQm&o3qD)H(Y3H8z@_%@5eFvq<lHbZ_R&MLJehZh}rL_)c!E74}
zOz~Fw?VW;hk_%n&`h#pQx2}Wo+rt9G0yuW0S-%Sc%tpzchrdjjU4MSH$}}a_x+ISE
zkw!$Rznt?yv_0vN)n)hF({$=xA2LFA^{HN~z=n^@&#+1M+O9C|ha2r6L7_p@OrAeE
zu<bm!`94iLXlu1>1K!UL8x{ub;wd;!2OJ26@&hkQZ~nTbDgI;1CqC*%NWxb+Ahq}o
zJ`@8uvDRt>ln(UmBG$>vtx<IckHGWq#8hcLQ<ko0n#NAIv8CVF#w|=8magWfDuP>O
zVX<Oe8fnw&TX(ReSizsG65FTm|09pBlgD4Z!>q35)ZKTu6+<e|o7CTt@-Hm-70UWj
zjJIjG&nv;C>7v#09_=D;7o-{e=q_d-UTaOr=uFJ%kCFAg>~+@(Mk<^(WDP5zjVK5-
z=NLTbzWAJH5TT|MH%H#ioWM-S#2sGG{C67ejdbSycv}iU7<4Kk;Z83#I&{VXf!z^(
z=3l?aNU&Q=BPa-Aqx}2##(5i*1&uualLsq4cDPb<DujY@8Eq2vc}6CSXKhn_Gv|!x
zI0=s`Rr!dX9Rtj(7KuL2QS&KC&kr%TPKaO2?wEsEU<mFiA+(SHJ&8bn76Do-ON5gA
zhq7V%W(eZqlsFc%v8w#?SHFrNJjsaaM%UAz`~4^+AAkID(g$ql!4d+#&dGY2gRJFv
z^UXVfIfZ>_y*nT-=5`}J(<10-TE|z4tvmVjlTS-Wv?mG)a(E5F=?*Vn2USm5ESzE;
zhro}c{8;tx8ynW|I31OhtQ}Y4FkVw$C>~whn^C`!KUg{+PQ7x?t52{0qaL}J35^w=
zOwPnx{uDEAivzyrsu%=EJb>oz!_+@IIEuMA-OegLy5+8X;Hew;b%Du=(@}S(RjPDP
z9MCF}+v%>h=Uqd-Vj8fWnnSIg?)~Z0Pt!i~9KLKm8BHU?pshSd2Rq83JmcbTfvS^K
zfAz~>bv@7LpMBo7Az$D9=i)=*S67$!-zx5awfwPc9cRkU)s7q90V@wYZB4`~yK>ek
zY#arJTXYuBg&#^QWsAaXJ;iDVI`YLY;%XD<h3{J$KY8j~p5Y|mS>_~Fp|u|kO<RP&
zH6rq&G%QEqpn&u!^}y~<vhelueoLkqw8GF>8LHMqsp*=Lz{JTu`|#fT??tJ*vG-oS
ztz<w_i)YHD?rQN#Ov;e_lrzj<YZ!T2S_Cf^W^s-VwC7l@>{?3tALNpL(Hxxa1lj;p
zcYlrUe5ocIzO<;F(x0I}RhqPn-?u&qhBlfZuNI-oFB2x3;-y8~bzih~-&#Hj16=;|
z?%6ViS8_5~bDug3>ALVPOl<@QXki6GT?qkTw7xEajDWXW#6U!Bo!n4ha!VttxmBkb
zSnULJ*GY=0<GPx;RY%3k=f;h|kI-g<?zYnijQbCw6kQ&p_{SEAMM#>KSW^i}R=3Iz
z4iXOH4Ml`ciNE3|v{sqM9R$aeln7>CijOV1cfYyXwH6hU!Kr73#NY0gG3uL&PT=>i
z#L5Ub^a!=up)El?vkvZryAdG4oCTkm?(wlYN}3Z&(MetK(-D<c@iQtmZSo|?)w77y
z3K4u+Nh}EN`7Eh30?@m3kD}DDl&pF}19&d#Si6C4oi?j-#EOyd;AozFj2wp*3Qn!-
z#6v4^B(C)+N*#f^I+HSS;;4j`7weNGKiEMyt_@)V<AdT`x3YpO%g-qLh#_w4JN`PZ
z(mQ;-`|iv17f1Wx#+szFxvV_g*Upv_RDN;jx27oJpba$*_<_5iwejZNs5;1uK|ta@
zjIvLe)D{o3MHbizy4gUjZ8Q(#t~1i{Q!=xgEB>~ln`XS71I*Ob&ChPMjhS9kZq_z&
zhf=K*XC)XGF1+Mw-6t-dy_{{iL&2VWv<jiXDG@xXj087rf4uA5b5TxH#wtK*`}n0w
zNS=iY^EGC>^Hw<vwvZKG`A3HMpWDW(0+0cbJWYoz-ZP+4j!wr=)brf+AT*e`E<!bU
zQ8^<Ux6}nbbIkhHPRqU##m?7@FXu>mWh+YuZ{;YRGpLfMwLEZGxB^AdctP1fqh#u8
zMk4-*ZWiZ9g!mT>YB56Se%_t;QMW&7+FVe^^JTv1U77Laa^-O^8^0B{6xm(u#97ZT
z@bQANfuna@u)g))STnAV$vId4iB8L?n^4HLgOy4)LSY55e!<B7NscGhQEF^DS2IDr
zX#q(M`)WGtD>3n3vdlbF6wbT=GorPE?A>?Y@3h3PfA!mr%A90sB>zpmLGvi1ZeQ8`
zEIT{Kw$@M6dG96vk!QXwTC?V^fSWQe;;#D?mm4Blz?ZTG?=oc17||I;faKX)7v;Qb
ztw=^HM>#3+ovfGm+rR%^x3F3_LU@iW2lNW`&`qaCfy&q3Ldz&rvvaP%&D?nfd&)@p
zU<Crp6v&-BZ>PhF**yMesxx4j`R1FfNr+katDK?YER=r9HhCDWQ|ef-8b5h@uT`pd
zN5xY>#t4vES?2-iPCexAn>U2NZ(~q`IxVJ?NWQ)e!}38X?DpH>L<vzoju7^(nd=!%
zD+l3h9Y{N<gi~hGw@jbB*0J<W{F*-t26%OY@S~&PvHVBq>94XTj55$^v*50t2EWd`
zaAi`qq+AM%a(Y)L4h;Ehw;UH6xX35h;QNuMG?WKQAeyu-p-G%mio<<7+Oz{s`u)_w
zVbWN<OkwyamXL#tz@46|gT9uvI+$1)bo8wQ+qJ=;m$K=g(G1)>Z{IG@K0W&1+h$Ei
zIN9^Vk3UM8hJ_1m){;oq9d4#8!hb`#9j6rIBk^2SCITP74JA&V+EOSIe;?PAe`%6C
zMb2dx9GqCw>;z(o!8-;cXf)_FtR<~;G*aM_I6rZvqkH6wFkhfvaf(9*08TP)SZR?q
zleW=zmom6hR>YMrI&YO{^?zACC62OazqfKWkLlB&eekm=BNUP`C2EHr{3&_w<&@#y
z{_Zz>pZ)pIdwcosX}9v3{8qZ9j8k`NS^SK8#0{tGV0MA8_<&Z6UXqkA`!JqXm@@x0
zxmBKB`BJ#SC1*?70mKa3#yuPBr}4J%|3283HsO_>yMP~-@*&R3>kOTx-@;h&@Zb(`
z2DY2$?|uYugN%89zzCQ&$TS?`9n~d_V0A4Mvr+wU5j0lB#jS?5ucMBzl%91!T=HH9
znnS-gbL5Egm@gxoy8a`bCE@5B7Di0&@-9A2OaKKxQz_;RxK)M?ac0N}Z<Mn>6`(V;
z9nXA53R6qw#vjD$a*!><Iuh0&f^U>9WhFOKkXBUkC1fvUnx<<cQbyz6DrQLoW-TBQ
z?k#zQWaDKn935C6obj+uE8WlM)Q6cmmZEe*&vRJS&6}}|L^-&3lB4iJpEUAeGD7w^
zC&7*bm_obIX;`SzNhI@*=5<oRQGPB+!JXf*W<)UyuHK{Gr0!kHkjr!bX4;Sv&YAa%
zxjCa<BV@G6yFYF*^BD_4L;h`ll)toaSC-Ayu<4(GL7TiUSNeEKx$J0ZV+=2Tr%QaU
zBm1NqmZn$%(qSz{Wz~gvtrN7Hmvnlo$FYJq;=k({boeuJ@zBu@O(yB^rMesnPKq+1
z=3U+_Y;R>-xM_iR-hH<Uz~`TTwwEIU+s0TyTol!Dg;t>hK5Co9Un#ea6ujo29IBKi
zo*i}Qil@*@+iF=72PGV1-?Z9L(m$nqB#w4ZA#n$HC?N(iQHZ7@(dpFysoIvi_E#Q@
zl#@qN^)4ey5Cwx1*7{Y3z&E&A7LkTa;6L$L@w9<@fDsU^990~G`;{miEVh(D(@`S_
zk*sTrBA%&Q2Ok<pkQLT-S=fta=%^imm(2V(&ni)=H+AS5ALbj7;rH4OU=H$e^KJM1
z10Wa1Px)s{i=T2i7ViJ5<vGVkE6;U*)x2L_2}y=J2@HB9^DwYeQ;pYjY#L!l&1uN|
zH7bpamBLAyzNPR~M?D3Iuhd#>sRK1iGs4zLj|sQra_K0Ikg$y4c$)k^&+7l9tYvsR
z0@A4H+QVysd7P6Tzn=L#+WYG3uX;JZ<VjYuSJ*9zY?@kyRQw4QZr%Hmc+<PDgw^BZ
zTOpL6wMa}FovYJD39u)LL*U)Gbu)s7RcBZXB-GZy6izUPkW*pEnC4nan=&g)l_xZ<
zNm$1;1ioE4@>3r7w_*{Y=)9eiQ&~W4v2-Yy5mx8!BdJ(t{^ei(km;hwUARg)Ye$vU
z`^vMFhSf>Hjc8@zP{(#6wG)mqh)Vm;d598mErQkP^-AhXXLOR&PpbsN*|$b=uLUH9
zE{%c<rLj{msox%npAHru0E{+x#|s`nC`uB#|7$onajV?hJov57;IuAuw!@O--m*;?
z$^~D_g0_M7;InS&P}&~0c_qF0DR8qDcVH8`ho<NNhUY4~fx%C_!UGcyNex0M7^N=l
zBc18jPCU(aZi=SS_fYr~Gil@@TW586@PWevgMj>><fDjPd!Un4FS&mE+u!Zlv^7~@
z=G!t~#=2%F4uwcB37>A`PTrJe1C5?R6TURv2XF0z<xZLo5`w$1(l-1K7svi9->8q!
z16sK#AleFAINi1P(1x`Mj+Brwoir5-G-;kXYjEta8CI4mEch_iMgH1<?^PxSFO=7V
zUzPNf1&%H-cj<$VFAIA=%D0u!036rVSaGe%d5a|^=Lo$OIRHQKLZQkuQ#5bJa%K9H
zVxb%?NWK$wl*Q!~ND5HvR=o$_<*X~y_J;CDF%9h#@B6lnkhc7=(!-rw?`@Fv`Rn{y
z?S7T}qEdca*hTUGV*O^_{@D1T&!Xdb*rlC(*EJ)&jHdiwT?tvaCD9tOw$bO^=mGN*
z;2jaoMy#=EgoIRNT!oiRH7K1>KRF+vG8)*hgiv#sdNuMyjbs^L0X@mb&Gz>0?|t;q
z$2E=G4oe8=jK@|w3oVW2BNQ?^vU}s?B%N*4qzsfol-10VgAZA89da0b%-3<-bqwPb
zyrBTJ1BxfSjFd7LkuvBw^B<U$sT(&ONR-91;zXzsWN;$%WoJ=p{_7X?11p|^Pbdk`
zOjII4DDV<SkhL2TH-XlAI3&%~mGTo>6~>__1&v@sSM$!MsP5;`v2LM`qC*KBN`ghF
z)rBg2)Q=8BK}@f$75NyAck3)M^;=h?q*!F9o>VI3#&pc1?5cT^(ReDmmW=@|QSzsg
z_7!0rEc|3%0bVSEN9ZOGmN4nW4T176509XXcXuwa$6dP3b1i5yjtF?PF>?pASNy9`
z<h}J1oZ;hYn*NM*@DkU(cqGnJ1maoP1Cu1mdorw_YIJT(<0d`Gg^w1ws+gsN?Z`Vc
z1fRLSijDlIgCCsTrAKf1NC&SD3{nsGvL4Fi44Lcc=vgU>d(yOQsbkxyM_Xmdm{y8}
zCxhnZ91kVh&q}dNmky)UzMJj2Z|wbF|K-0_x!3lz>8~^ONa?l?rAwNE&vCY>QZ%Y;
zP(;JbE;w=&)ULy@FdIFRuO_+93E4EymhW;-9X91FT^EN&`Mvyv1%v{Eet5=}=h<1i
zYpb^Z4h3F#%A-zlU&@3y{49DcUxhaCf!DgkU&@Yj-o<5Me41^+7TF`+^-R5*s(JtY
zn`x8zyu{mgItW2)D`CY!Q}tTQR`PZL;mOG_J5~Knl)R%@LLB0#jxDw&Ll1(JX<%s#
zzMfvr!GVFV?AGzL#sYldjnC9`%a}CoZEL0e;P;j-an{0VL5r8c>HEC*B|plYpOhf(
zT-b^;f=2rD&nu45W$|2o?%;MRZsXZJH}C85W&SJtW%u$Lo;w#|mAQ4hDgXI=yr1_2
zYNapOq(2Y;Tz238b-_MIfBBukA@2*%b>C)sk>56a3#+h0Gwz~v%g?>Mu?3Xp4p#8a
zMQ)Wxc%EN<cgzML2Ip4eT!Pf-TSKB4D3E7LNOHo*voIPoAu&c2<69CGk~A(IZhCJ7
zc$RO?NLQtzxQ-6}qmRej>a`phPLOE$T^JdmYpWok<Wxrf%$&Yp3C_53^E_XlO(7A<
zGra_yb>JN``L+|!C?#blP*^LJ?fSLgUBNIK{q0*fa+>N}$s<C6P>v-bW6e}l&H)M8
z@`2kN))avkpQ52cXdlVP5U356L>3r7`7fu!310`84P__7EO;4Bp?4V~@d;xVjiU%k
z=JkhJm-95%l%CufA!qssT=dgf=?y3tGJTTsS9ElQQv%YFrmm7NRp@f!`S{rt9YLPa
z_E{7gupec6?~|OIsoX_gg|~qEWjKxbd=?SSm7?Pl@1k|;K_0nA86cCy=rsAr>#mW2
z%ZZk2rE%{_JGiSS!EFr5-(9k9pH29hhBAWR<stvXfz-kUj}zYOzWCa`Xym?J@mueK
z(eHquvAFW2K;#)Lmrg?4(M~+CPGeWj02U^HX?J+WpE3ZJa_aExKyy(Jtou>EmN{`K
znXbjN6p({_l{FkQXgCWHR7d&qtIlrce400N$|ruOc-(mNMn>&l?mhg6?0!l<4z>eB
z{jvTm1;o0fp@67Q?JvqkI&ZvY_muLbM%wu}QF!Irh<<`rm36Sw&RE)xj#ei=_;%C}
zSu%E8p(iC5-ljoV5H5wO`<J=^lb<$LD@6DZU7J_hXzMe0EneEzQR^k?#aV+YvaYaW
zeKh^@vk!h=>*kv|qX%vjr;%qfOnndh!Pj<KfWP?i%X~xYi{4vDt?Sgcd$uU6TW1_S
z$nHDo4m0quQ|Rhd<qki$Ziix7{!*T}wzo+SaX5J6r6elfE`K!e+;{iD<;Wp<1lV|A
z{`0=lH*o$G&P*%U^KnjF%1BW7Gp$?S17F^G*@djK?mS-iL0Oao+zZ?bcy8jsqOxRz
z&nhR%Tv*OcdfA;XC%+%+77h!JJQkf+e&mHRulG`P##`y!FD!r3?WomuGT^UtE58ME
z>)&VjOiyr@0(vH|<N?Rmw}h<RHuqD4$o3Rjh#Fz%*2ZJ{Wh}mo0#czn1hI~FY5hG|
zEKtnz){S-rY>7gUvXnf37-61tPqRWEx%}ynjdCmywPiKruaJNI*eH-fGUB1Br{&hu
zmS_!2{E=(3AC-x40idBX4Hste)wxo9tk+->agf-9oVW4dQRc8y)@%73H{ociW^_Oq
z?Z=@_fV46^4&m33!d6MU3Xfn>nT|QvG3f6JpDu`OJc2HrUj*LMn6Q1_^sO8Lk6xy1
zzR7fz>6vu|zK*R{q>N7O<_Xbjs*lPL$_<OfzyzK#O~;=QWOgQ0l|5tKP*O*lP9`))
z=QP`vd3CJH4QKCbp`tRaIzUe>n^U*zs+6gp`MxWEiIpgu3X`t2T3o_!p%J1sY3|s^
z$ZzB~=mqaO@n&3i^{_y<UcIllNubU-b+mPjlk`<G60lenyy7zNtL)OM*IMz~c@Kq_
zKRo2)Do*iR@Re!d(756U7az=H!7}}6+QfQ}pZ)x25Z&Rb1A@j%udV3#__Z@m$<-;s
z9poc)?n9x^V#@5~P>!DmrUST=tkZq(zVr6p*EvM*ZWW*?B%!l>MqYz<+6A7Z5LmzA
zoR8v;-Y;_rqA0X(LI=FQXtM6Yl3i)jE&9qA#0&^10BagUJcr}KVRuj(oZ>$QNCS!H
z6~A>y^HmOa`3f5rm?#oa&R9Th*s30@BgKEwb`?trlO%bkJi$wu#32JL?)5Ruci(*{
z3smp)*nI7n?9gErj{+JBWJ+2k&;9T)JQBbZhExS2^bTirNwMA1Nt98?c}SWCTiWI|
z(-qk~*KJ^vIh6MqGQ$5Ie%lVdN4|?j-y`W?XA(RxOI998{k~AQ;%s<f5d!aU>%RP7
zxSxZ02L8%_#yNPcik%3&km2-z*chRlN0joRk){)>U>qIN>ZAx6rsy`}gv+MeGy|71
zQh|`s9hzuh@*x=9!6RTaa1G6!*%DPLJQYySGHQ$mi%EV@8zg{dsWVG5`~~l7D|yMM
zNs<qojdW{*j!pV{OuOc{t#zRUe3tLc5WqL{-4>&s8xiCdjy^lF!zD^d;#W`*{L|?`
zw*r@tbX60FN(mF<5LN_k@`uk7Fy31Sf%!bj+Krsn`Y6^A&R2=Q`R2>+dboe@K@<@S
z8FTb|2tctL9}uJ11xDgmDa^x@99fS@7?K#`@1)4XbSgchD*l8&<zrp@IiuwCY=>At
zQvU5+z6>;jQ6DV;6jN8dbuC2@YWc(?z@>=@Z+&Sp{`}X?J3d)FfiU%+_eG=L<M(Y&
zsCcf`J{H{8*`)IhM&K=_q7`y3Tv``c+6ULsZs6nRRnP9Dos6|1<pu3xrYTM4GvxzJ
z%ALBl!-#d{tMDD(Vs#j8ZCG-&7X*X<u#61Tzq->Vl=<XD8x4M)j!EZV#o_WeqbD7%
z{BN<Yh0mQi^L0-1{pSz=u=j@#KP(ThhQL8yi+;+=;zbdyVxc1a&vGpKinp9PPh{EU
z2_9#akVQ}B=@#M1IM!51S{g<U)QXq#_Af9@zfdA556aDg2v(Kj&~vq)g~zrGl{1O^
ztcRw>8ks5A>p3Hcg4Suq*>+djfupw|1cTrqZ*7^9crCDQ--;6Q&byt$v8$#7j>K>I
z7oglrqf}6*fK%q@Ilztaw7hmM9@>k)Xyh*MMd#(OzKdDnb~+^Swr<6D7kA|){j0ha
zcfnfemj8PHsoc+(<7IfS^1W*Om%;h5{fpNH_4jXq@5^W*?;m6dIRiM%TB#9+x86g8
z;zq-E*Z8LrqCpxtXprfsJG##FfkxU>IK$_IV7Yf*IiHDtZu{vNX^l>AC?O~N8Kn>a
zf!9VA;YFwr-1U!;Hzjc?QwPV1>)en_w&n#lD7uekB~ut{<ilr#u3y3Nk5xh0z$VOw
znK_n*SYB@CV^vOk)j5Cp$sZFpi%4%jsZ#WI?u3?g4;I;hg*FtH(k(gf=S0Yp3aTNr
zbV}BIfYXs&;-ZVrNq;PTF)pna-U<~Vl(2n*KNgT*{^IBTFy8ens&kN;DGR5gn$x!^
z?yH!Enbzg!wA9YQCmx}m6*mDll#Hj8uyk%l{9W9d`f#X|dHZy$Ek+efir8Tm%f?J?
zQDN&N_<&QzgOw}Kwd91PcG^hX@+9cEQ*kM?dT7w74>aKTj~X0xvUsI1z=f|)SAJ8G
zIK!)8B){ENsN+Yx?Irz+x1v?ip0wvz9+HnM?{)%7L)mP2V8R^FB%sffkTn$}zs66v
z`_6%(9w+U@t12OW(u?cVRQxDRoqY1?C+&PG1Hw(qp~X^gbf%p~qkO1_W;SrDA{vic
z6rGE~A`(Y2(ROcSeTYxa%L_cm#V4=|<`<b(_&i(Wzi3CFE%wPD4ipbg8h<Uke9;x`
zwuMGH{YgN-DTEfY{yp~E#i9^vgZXG*TuU(HwzE&ZwS45hrfJY|3?xFw@G(o&!mlur
z0h+mrZ*X4kOQF-|@v~1UyqW1KTW$@Y9BR05ZQIO`y!K;ZOt!g(RjTbH@${m#>i_^i
z07*naRKc@0<oz6K_NPC+*WEHcBuLSaPQC-*p}e$?Lo*hM!P{tas$8f{IByjv@A7H8
zDwWbo;}>*xPCnd<BmLHaQP?wHN5}Jd8_)BCCH~g!{I{N)XW;{jjpw<?RUhZPU+J}F
zc%_&ZNBX|$zpLCA=l0|Ie&76e>0VOL--hA$q17u}LNLjiFT4<mAgqsQ1JWoX_#iaa
zG_Y8VW1lg*hH;p68#+qL1I6Po=Y$NQVnw?+a?u!#+{SlXG8RY!9P`~4aG{Y7fu(`#
zcNfyw34~|u%&kMn85l-}>G5I!j}Xr;j-=6;N-*ukdv2?I`@+2E>aRv6POh3I1Iy@)
zfWM#4*ZjNFOf5*YXcNAleEO$!GADa)KRW3MOox8+#*^+Uv1oMc3JRW~)pBUFU!`i$
zXlxZU4YCB9aKcGv#o{9+%;SQig|PAfg+O}0wtC)NIVFSuzLgOd|6lz4g9!f!suaf#
zB8R&<A@$>r|G4+(&;HcIuD;4H3!T|FF|(Vtx#Z;Tz<nC{`xXmE0b%`loO55m>OxkF
zV+V;kRbxu(N)`^w+tnlNLp$^ih83bEDFO%mXBl}{X-pjShTo<Bj0ou78tQ$d$!}dp
z*P&baE<V;AtWH^e>ZbYT&b5Kb`~t^U+!cR4OS64m=JEg=)Nf%ZMk9R%zc6^sIPN+i
zaN3FF-QBeW|LO>~s8yK^oRvStWj$~4lD9>vpM3I3l|LV$TRdhuj?yrAXKoEI4T|xg
zNpVMg3JwJ!b&;t~d8CDAr^aD;oHZdkzf!;Oe)8nb7}aOnsr4eZrCKP9Klbm1w_CQE
zha2e4chCv!!(8rX9ok7S!i1lVk+A$z2o5vNx!m9ZvSxI=qaKDu1PlvS0fGaXqe;`W
zO`XzM+(lz`;vFpY$C`qtZLz(XgA<pcb3optO~TV}ElO32q#ueTrIZz9wToLh6wZL)
zop=6}#pZX$sjH8*_ra%RX4A$Zm1Yy1PH_+7dza&}Qwk=|>SdMB^TNC3NLi!_&_%M{
zvpf^hvI=ZHB`!Z#D!g%Rq28nxa>fa(W#x?Tx4(QFyl+qWilHg5I{ZGasWxd#BQ%T~
zzZ0rF>N#jOeg7hEyE1IbaS>vr(X8_ozDv72Uh!VqOge3ooT-ONvsYF^04if{A#O9t
zbA@igjrg8xKxyR1`ILMcr_zY8F{Q`SQP4BmS*8JBxk_QkQRA^FXfWoFAxs$A@nGi-
zq4DH#tSo5|4&EBsCtntVYCt7fVe5Q!;?Bd+_?={kZNRjxRefL=#9AV5-P@zazr7)p
zBgDu2x!41S_jU+LO<4GV-MhKq`8X@x|M<s07T^2#9eI5<^Ysrav<SaznG-e&BSgm*
z#eu_O-<CUYnTtDGQ@;F-1+WpOQQ*+n6pOsAf9W(z6uLTP3vr(1z%py`#378h3g*!F
z=36(~k-V1=m;L59zbQQ_N{*jKw?F^+Px+eegT1d~DN#<A5_r9R>rM}2bGRJbSVC+e
zW?|9sJ&0oS%)Eb3icAL(+6VP2L@ZM3mh!?GjkXdi*wkJ8Q;%=fVl<Vl__l6^CfyRS
zdF^w0y^NZT?uOvcZv``@W;>>XJ176T)B`_<ccpU62VnPrG4S9t)0A0~2n2#1Z!O2K
z@sdVaJ?p4eTWGF<rG8kzz{kh=DG2$n98yfBL+im;1Frh2H3Y6n+m1bV3u;kOTr|WB
zEE)g&FaOfKebCMpVe1YOTw1a4!EY(Dloap9Kd>mO;wmGzEIZ>f2HK&{ySA97p@_ft
z{s-wm-wFS}mo+2#;9@?Y_x=a(_t3!4KKndVfZ6>K{;>8)J6@e+a!Ng=Oo6xZZ>GDS
zP|kmGo-}5i3BH_PR?VlusfvVhQ37&T4?8|f#4(LA`DN0gFlL&{ffQWZln$Kzew@xC
zPzR1(>(*P!1ItI1kSGws(4@FTMhy0iGSUIcXm4Z*dJj1_1`+a>m-OoA?RV}}32_X;
z6BRHWINGjwc+;*w#jlcQ0af^k!|TbUd=kJNj94c`7F7wE_sW3yS5`<9SK8-!<vM&D
z(C{+Eh*;Xg<07|9x8hzD=_P(=(q0hfWqDo@^<T&9C1rkDX4~>yOs$ut{Bixacog!c
z&R+-<VU|;v%R!0Ug>k>dUmCKb@MfO;>J|>W*R<F9{Jm~l0x!e?!q~Q&>L_XQN}+s&
zsBLRT&26)7oFl%`Jz=ba)krSK$^c(xSyL`{(E!vPC%OBel@XrCoF?9|wj_BJc!Wys
zI<9Uh&S?ATF(KuGCFddWxvd}k1E(Rtt%r1LBc~BgsP;YhQY<I~RgiK<$vCvkwz~Tj
z8u#w`DA?t35@m$^wTvDQuM7(Up{Mh0o+GoOL<uqXK04u<bK-TqMTC=zCUCnO0PSWO
zC`cV4&7xHx=Xrg*#S|C?h|_7KZ9CTFO=)Anfggq9_FK2xY2J<H1Zr<&ilryF2FCt=
zI>1ShaW|SwId;CIgM&XoxfCELSJqVr39hY>fxKU*WwxUfX;xNg6F7uyJB?XyspUMO
zT_f+XX}S$9g>oaWabF|0L92<=Oe3BP?Uuu9Xx<$*iC2up8GuK8?>fQUi_{DUhce7U
zSWBGnxRr<c(AiP!l(lVThVO8ZAoYv}{9xS1&YLwb>3mSpcZIA4I*R3OW;3WI%aLtw
zL#a)B$RAjOs_5AQ8jn8n^O6|#%i{tA2WOJ(?_aNlOnWQOrVOUtjx);l@jG>icIh6w
zNEn>>jQoJU&?P`vGOO~GL}T63D1e|7J8!aGw)~Y^c%B_UeU8PB3nvdc74gK+lx+DZ
z?L&P~CMr8NIF#1f*lZ)LJKFV)SUOEDcKRiCB`pQPmiOTX5cF1F7T)31RnUa?_$dzB
zXqA<~bjFXX?KIIAI&f(_h-`$nXReF$@l$Csfh!~U_;vDW>#b#1cL~o`MI`aMP1il~
z@&As`!HDBm%W&>|SU{G4AI|^GbmzUl*ZYqN>w6(^nn+!={HI>K`q_A$KAy)t^!s;p
z=kQUMQNshPZQ1&~;?H{jR@_(Imqc2$I}?l^=hFQ!C1gWj87>W*u&u!6mys&L;QdLC
zC$D)qo$fWC;BKi0lV3T=h$E)e>sf8Cp}0z{AZ+w>>rDbJYZfxJHxeM2z%<Gs@H>T_
z1Wr{qI+Et8L$7iljTy{LCs-3AzUvHiyeP%SB|@ULK~{)H$aa(zIP!&uqnTM*OvPAy
z>MR9&{t07SCBOXQo8+5AiSp6A>k($Y@8Q(McVl5N!g06}<?Bk|zz9yrs-rc6>?pDp
z9cQOJu5gDG1@BJMTJZU9=9Sxt1s+;hlX5jvG4H<pW-4G5!bzsSr?a0!;0PS)9X@87
zg>quSDR=OUP<+1ktKa0a@<y(6YvqHS#@Ss-@gt1yWM_mqe-;pDwmA0Pw#7G7e+Q=4
z>;?)S5XSDNWRyuqRhKfFyGLJ2hx)MVh6jOJT6oSM$FCndO#ES)(orT}MueAiM2XwA
z06CBfjCJA33e)(lI!|aVAb5n*C<lI=IIDy5eHOSk@8nfW93_Dj0zL_suJk0#(KL6B
ziobI%E^-TAaZVa=N~6$$GRT7XEc_#$F3yMqk7+yCQs4X?7+5$mNH!Km_iM=EJ%@{Z
zl4&;a|L*U{+KD%!B+FyaRs9dk41TV%9+)Q`Ib`=mcmUnOM;vM%g+#Zmd|7H8I2LD*
zx}*r%@$v5dJLTn@X$N;=DYzN_y!PQ6J^KBNFTY5e2)~4n2F-`%0IxP5<p|W-&-QKn
z$Tx9PM(7y8EO&Gqf1`Ysw}h`PxfBR8$Jc?;*wk;^ZFt=MQg)Y!L+NXM<yrf2X_NTd
zhbgtw>#;E5>6_UNL-7z!Ho+V3KkIs^tQFsPP+f3H9%G6*IOM%Z=f<MDUY#sk-V2O%
zrjO~YF?~jP5i?hdJ`TaFQrh~k4hzj9-@+$t$Eg$T9{!xTXu7OsWS9I-f;Xhb;4pW&
z<Q<=AC&IjM-H-|cm977`J+Cz17H1o_snV_YndkF(oe6xEmviNMl_V?9DSs}Xv-j`9
z?Yq*Sh4$mWE&SXkXPwPt(PrL@Z=TQ8Bc3{&>PLI~K}yIASSTch4znw4l>>fxRxlre
zqAchbHH->2(j_sZ(FiFpp0CXLTLMxHbE$}Mn1-iei^J)Jq;GqTKi9kpKcg>x#X&qV
zy=rXZxt>xR!8b~9Oq#|^Akr`Lw{p}>66+9|lKaKnjA95p!iVzV`yJ!NQR_2ehK?oY
z!2@TDoYW+L6lFk%X_O@=ox>XurUS3hxhSJfuhSiYDJ|=YH5r6HL6XE%XeR&A=V_E2
zem*OHJtNc`Z)7)4-ml~<oYGtC(NS6iw2nWMRIP&+n{Y;#$$4+&E5M~f@KEkIg0p<>
zfT+?GVI0sE&@4g@P$Qt>e2GAwf}e0B=)qoeT#88j;UZ2lR>z$0Y@Bg2a&fuqh=w(7
zK<swz6aqgTrW24?<0X2Z*FFBLVJS2I407Q|$`Jg96IQES@Ilwbs~x?@`eQMu`Vp^}
z^zpM>%00k#u*9np1hi-OxTcf8=iJevSDyKzc{>kj=C$H@$N%bs5~<R>gJar5XMg|U
z{qSK9s?+AsHaym~M1kIpUOw{8g-2SN#4G0QFoT7Bg+y_G7KOv~2jxErljTS+t<>^2
zwP>+sUW<q6xo^JtT7)iwmQfE(gFtvjIZmEO%~FQC+ygs5^WEd-x!A0@Nv4elpQou4
zRI_GD(v~YVZF&*SSWH;ox@~vTS^FC_8MIT++Asx#azgp=rD%)E9oVLwi;u_g>VqiY
ztcGNoJgcAwmqF8^6z~s&<p$@4xB3)+yraLhY-`azd>1?s3nD+KP@36*nD7M$owTE+
za7<Od;VTZ@)>NW=Daqxj(9yHJmW8Bk1NvOT>fpSdhtI{ouM%gQa?)b!wR_%%y)N7d
zWLb6GU)M>Y^skz7#hny;<#ex$*AI@rXu0b2BHFAtGV^><oF9xxWG%*g6}OGIp3hfI
z89Ks<K&o&HbFb%qSHh@8<5!q`=~JP)ULzw6+X_3S5dk30<#b}gSsXO1Yb+!&?dnYB
zZ+j^#N9V*-01Zw4!d9be-f7^+IXRV|unMdA5w2-@NhtB6z<3ci|1UUc8y0*ze;%QF
zD=XYtK?o+M^sm1DGCNMb7@bl&#OF>Z%LtQl0#|ne&qwBrtfcLStulHX<(|sbEQhjN
z{7HB`n#zb86PKU?!;X~-px|e8Wc`LhWend%ioO97ag#nE3uq%uUGlwk>((w$qeb6z
z2@%`soZ_gA7T@Y{u3n428lqgjlF^rViGP?wlitlCO}Dbm@OIV()KZnHfTL{dJ&N^)
zN#1CT;(|7&mee1`O1YMTIZjFqbwhI<J!MN+>l4jWnvi2QgfK0uO6y21j*9X=+=x%v
za`2t<jPw9_4eng<5H|A2d-xAd2l2;~I-Sump)Xv+U$HJZO8X#X5qBwAJ>MY+;&D!X
z#U^q3o-pZs&Oo}+d7hfU+n#0|@yoZt0e{zOAZ=(UZIztFDW4||OMrN*^wLv8{_x>P
ziL;+?lYHH>uvDNeI)Ra-g%@llPJ@>GJB1oJqYKn-JfYtx8Gh?1JNQ_4fUg(t+Ck$}
z_AD(rZPTaUymhbnJ<RmpSFzMkhAnz74byQ4rGzI3d9%h82PNU;J6y|{$i*V}S~&1+
z>I6^r(*h}5>E}H-fNMc&EhA}n<WIvBwyYWy4<Cb6=cZd+reIv!mo;6|!4L0Jc71S?
zHH6}chq~@6cxXeNLdm`L+NnonQ{SVEVTST|3!b9kvno1Ke6Hqm6zB#=vae-M+wC1c
zrCp50p<u)}m%F^+A?<RwZ`IS{Iq_K5ma;4#e*F-iyLait$@{FFQ44v8$7zj>oEmRD
zE|2Y75)AHHo~LbfMqdH*z75|g@Co}qN(7LDhPf-x_i&~2Cf&C|X<m~~O`?weo!qWo
zUbWma-mZM-((U*m<=NJe;=NEx{DUvgmHqrXIHw$YF7kK2A}%Bsk@GAAB3@%q;qH`?
zg*+jsF{;!uM8XJijK3@(J(@TTp0$MA*YNU4&^*hQL2E{y#<I}7lU^Yz*uhL(c{VI*
z<-5>BCUN5>q+u{dJUEr8lagD4iaszoeM*j?HK23G87)a~yRMTYj~+V(GPozRlL({2
zONTtBiN>mYow3df?u3vI3>_*&l5WR9fkVi*GuQde0*PzXtAO1KnQ}A7>+F5G*qI$q
zj~{nb#~)q5tdPz8wgr*|k-VAu4RBXIC@Fm8b*Px|i9)3yRHVG-Cv_XYYr*62Bquqh
zUMMX$Z{7+%N4-o}>7>vEj;6yXN%FBM6%7X6OQF&UU6+E_MWV?on8!Cbv{RA)RMg_T
z()S*;W@U+895G}YLy?vi^5gEg%5vxcr%_+w$p~`w&@29gKHg=~sud(*c&%r3B+OYI
zXsL4=f5|{S&Aggn-d9J^{PQEf6{lE%$sazsz-WEDhgRMl1z(qEuH~L!q*W%<X!o+!
z+81zryUBuBAFUHNL3QE<-GOiN6;^b^2y{2F#*YVdPMpyP77xA@MRmnGM+vx;MM-GK
zN~crDYo;i{{yNXfay8R-gO@|Eyq{*$9E(3hw8_&jx#cbKD>tM|oTccTKZHzOQaXA<
zYQXkA@hKA0bh;#Q_?wpLVR}(SC?&Omtf?4^XcS(2(7H{&1_RnG-fi9B$y8NkUFuA}
za0{}zJeCvZbHelfjJ!chMa(6wfym&w;FWiN!sVNVaY`?0%O9|tsu3yI%m;ksa_wth
z!Bsd~qn$gv*B0;ALGr@a(l%f46Q}9G^}KcQ`!2VQv3)+D?i}Rv!J}-;_Z9M8QMdQS
zH!qZ98|fv_@Pq&0)fwpmJQ`=low0iR&hl*||0_Pj@7wBO;lEILS=?_e&xTK4mL8v;
zcafeidY+He_Y2}}D+ZL#Pioi^dKmq38UR84e6-TpP&NE<#)eMVu7EKD8GG)no6r!L
z1;H@w!5n;9e+v=j_NTLvYN+C+u@Srr#WtqAv||Vi-%e@A244OMCviKv(YfTP@&q-k
zW%ZMAa6D9@O1}rMj{E|TvUELXaQr-$fZJK&ez&^-j<U7&?%u<U@~&Isk@#0r)}&0j
z(TV8H#y4fM{WTr)qX!QQm&Kuq7$Kf~+c`o?YV_cjd*of$mRS7AdW43DF@tkQ>2#d^
z@mW^c!(Zp~?D^5&m%dMt&j0hzKMS2Qosdq$mSS_=OYmBuPXND_xo?WqFMjdMjLve}
zYTz>C8xg|6G>&?=-h~@93N3?+k{;BQ9U*<JW7Bbk;)$=rVhK=R6eU6&t`wwiGCD?A
zBV_{KapCTiFKK;wH69g6L^>T6PUz8sswTCUDC1c0@I=GXX-g(_(sJEP&(uvVNyRTC
zRppDHFg&{GzS0OouTBjnO-I4%B{Y#nM}lW-rGt|C%B3EHK+pK7KulR$mx*843G+_r
zqGUMmPF)mM))1i$df@Ztv4+WOl}UQ>IQ+xHMdA4UKmCuaA$qv?kN@zWvis@RRWwbT
zfrHm{u9QRXcxAW&*A`vvWUbfeX}F1Z6drC{W0jF_7GRY>Y_YZ|G@Yxao`Q$J@T3E{
zeo@O-x3h;oKg<?(J7!pRDD6j=@^Q%GKib+3#tx@}nP1TwYZNy$)5cS0lNTt|rE6%3
zPw+Yfl@ZM(PWN0z!~tqmNV0~=)Z@(@rS6;`(@9Q+)xOmmezlEO`yoSo>qZ;CoW<Pe
z-4l>AKw`<DWK$$4WTwITwPNxrG`W&E1BW1ydW>MJ+%a{EpWe)N-&<J(?(jA66;|G+
z+pb>D&ZqpHgwE>wDkUb&zUtk80on2FQa~18bq`k3h+AG+gmjN13xZg1ZN;@+!qZhb
z#NTsingr|K)Gz+5wA&DM2F68>uj*`-2FI7V;xlo4S)7;omlmPr_jS^{;J_7L?(4mN
zepa~0z79IB_@exKx>W<k`-1v7PyKEAc6cY<juw+HyRgjz-dXhEeIX5`PuSW}u}S{&
zM};_#=nJuvIpl4dbnlf&c^dwBI?PlC7tgbc(9c!jAO9}uBRsYlfCrX)(ljj3$*F}*
zmNFXt9yRua@^8B?_9Db1>Bcn^y`*IFaP?Oju;7(PqbR1_t66_>Hy?0wK+`%_et~0Q
zXle=K#t1J!{OdH+H7#q8X)3VW;1gfP!ESYkElB(|C;p`~3Age9bHWhgBtr4&GYU|I
zS=L$DnoI$(*p-l^Na%!htlYKh*>aeU#@zlJ5yZz4;>X8?Yeu%$ho!~=WSsm@ow^-w
ztQ4#!`xTVIKjmpTQrLr>^hg*Qd77$`#IvvFvYIG+X$$u}!?z#wLa1b5JpqcG<~{Q*
zt?)!R0Sy8f>^?&*d3rxhH*w7kA5t=fd)BcmzLn>y1G(mHz{PuDx=XvR&`DnEPdQsZ
z@dLAPQl=``2`?^5mjX8oMrhTxk@$yT<o!75-Pe@MDg)SD8+))qO`V?(O~_+kuZ{xa
zdos=ssbz~o2$pi_b7mc-w#q{T;wUKu!;Nm`(-g;3FIl4)82dGwNRK)a2?5_)r?F0i
zx2^YRK)He&E81Ak1#030&oztuw0vG(;u`NXtoTeQo#tcF!8Zj}gkfA7Ev2Ptk`G)d
zwW9+c@~GJ-v3j(trsoqyo}i<0SiMvwBQn5o0=`hJBL}mYCVwbu@1<|@P^RRRzwmic
zO36#u@SVB~y(qHUAb(0}VI*yNNus!CY1b73aLIdNq+XY6T)3=Dz7tj=uV@mM_jyn*
z;WzjxL3ma`;<Q9F-M~P=|8iRjlt$CzFNsDk;DWJT%XwZrk>{PY=zP2j>VLyY3)-3c
z%0)cSFN^=mYeI_nLiS8nr!QsxAa6^oZH#Fu&3u$cLmv&?2*wCr2WT|k`Pg)BtOvHL
z5ehoXHUJGN!jn4DHDD<TxK7z*LL42oMjL^dhFF0u!<{2YEM_&D9*rKJmuwY|fEZB&
z)W7m2sc<f_`q$^+M933taI>1-ig{~7-2dnQ{HN?n__CursaR%mi+Lg=J%T5l30&=h
z*}9rr<5BZ&@ath^E!P;0uaT8``Xy%ISzG~19!}`w4x$hp9FG%9b?Qv==<D<0OuD;K
zzmab8vOdK&TmtsbfBtjw&e<ce?pSm2|NOWAo*g!advD#jRVC%Ozxi$F$bX%?j^tWS
zUtPi*+0cKWVK+jAX<$(xhUJCHKFU`5Xta4VW^QR6(4-v4p%JCziM1l`MxOC=f~2(r
zKJta;wx7b=!c_|m_p?o#&>Z#U9{BmGAWb~<T1p_k;1=DQCULtnry=Q7g7;V~tUktC
zhpxp4PUWjGNs3yF&n#>^sb^8<@cgxmhLb_c8p}oQ=m2J?yYgPkPVmstfTPydmzLFq
zH8bK<3MnpN!LwzLLV-spq&gSzIPJ*P+ff#wKDX=R%FRx}{OQwAvHRZt_xs-muQz)<
zxRHN(Fl%{|Ecqn@O)Wza`HYS#?~Boz!)PODdYI|Z=lS^L^Y91U5;IEb*oK?eSVlr8
z{@P+L+w7f#p@XArK1dmyL&IX>FuaGCAMV{x9Y3ceSC5I@U{OhDYmsTPEWThT!~TK4
z-lEEmeLBrA;dbBUPaCZQl5!~5VR+TLkT<SnjfVx>nIeLGKQb3!V;W>khmcLmrv0c#
zb!L52KmHiFtV2|d!#BnmKQ}UUqiomHCghbENuS>?xTTb)V3i#dmGE2YE~`ochs+u@
z8#o3=)`VOSeJJIuA=-huSZiKZxnt&Q`to#uECp4y;lm|o@d%z-NBmp&vD(XO2gwN7
z%4>I4H-tAJf>#)=?*>ZJ;=fmQ_#3SAu5jbI!{K~{3*Mz!=`INVH}P8deit6!75;C6
ztV}{Pt);Ykg-pd0ziFhypwpc`EuZhWgOzLRyp@J{b_RHV_30Q01?Wzr_PlbC+4&6f
zzN{O)C%s0aqhmohWJQQ1G;(Vrvv|y0{|@K?4oL!~X8oi;sen<4__ZjCGm5l(N*d{Q
z;nL?%!v+T)eV*}}r6SDo?1~bA14hC+vPr|tU1h-;{!#@@*E;Ap!Eet;xHnG=X*G1B
zD}F1VnYYAZiYs5M{l~G|5oi%uLJ9N`Ap&74BEi4uWb8`9fxF)VY5A11YcE(3A{&hW
z38ppL7Z+G2;OCjPA-p8Nyzg+SoQj(AbsKMlu(Ky*WlhJ^bdXO2r;V+9qa#$)QBy3y
zkG|B7S~--tay@=6U~t{fdY>u^%`a(^r_RMlkP?Dl(3b$!ao$J=ts^ckB)n#JMV)0W
zVtFe$O0hhzzj&4x&}`%Q!6(1H>l_I1F3L<;!3#@-T^-E0&3A^?3)xlzFyh(bmKN#w
z<0p4{)#@_iRG9!EuJjm>{jK0gJpQiIIw8e-;>Qn6IHUa_Nobd{*qwzR@20cA7b~69
zajQILfhwo?%dNbY=gLF5&p7$u>$U|rg--HSkJ{A+D45}d@K4GfI#BL_vrgnu6ieH>
zTi3B5^!$#DF#1*NL{gHTcjqZE2mSM#XL$s4tHAr%p%NPF`)wIUtEL&4qm6oP=e(>H
z7%sT<m=W>wibqT5<S8&9njaqYS%Nj4YalY#m1G@<`eo6X{0t6@l-k#FJt<57<TJQp
zAeOh{8rp#gPuq3L#*=>GIPD;DLh9my29op+!+~I7mF~(MToY@+&21?oZP$61rs>34
z4${jBpYIgM)@$)jT0E6Y`ZM*Mv^$<o+~yI#LIo@Fms>o%-ha@21GjJ{-aDAv@a^-8
za6Uf&O}taA<Gs?X`-*oF3}9BgbI%*^Z+~2H*Zq8K`6p~sbB%%a=V>g?nRwF9^fOUm
zt>;}_dAr2jIB#ctHh~_`#XlE%;p@Rx+qp2t#<R>9#4N<)%<*#sE;l-NibCc?(?D;<
z1dExCA|7Qm`NQl|xt_HMc4atO@r`KyZ)a;K;n>k%8er!`Q_OyjJc4w5@iq5H`KC!a
zLmyrvTpwfyg$~hZMn|io_niiG8v-M^rNh!_2|0dF_Uy>!DCccNXfp3}I=sk$Xk^2b
z9TjCND;bd>IRRH8E03Hu8q3E!PKkW~-IU92kjGhf@pZl^dbb^^(_cqjPy%9w2~w9M
zu&s^pu`zUH)&&}V1i3kSNGQk9F%nX|!5Y=>E4^LZnR)fg`Pi7X5@VD@Tn(O4PG<PR
zkMBw5?eBhbxAox+mTMN#CWFq&hkiyjN9Z0{@M0>*-nJgfmj#1D@>wh;50dtS4?ZZ}
zA4Z`;`@0#*8ZjCXThH+J?W`@iYUjaF(o9!f4!*Wr+bJP$^`=D5kf{I;KN(+L&WG>d
zkf<@=U(FXrSxSOycH8I-EN*38vDorq*3u~Fy-YXUe{ionaTI06G)@)4z>~*u);+W%
zncc%U^p4_{QDz*a($>stSxdn^jzvL3Hd<#QCwyyR3$U8s<3|x*PF$8ee9?6S!GG1)
z5(sDwFV+sOokQqu?FZbnk+u`}#Ish$VR1tlvynP!Sb-95hg9KaG!IUX;id?LhIs!l
zYaE`QWL-jd`a#O_U;pp_8oaD&c(V6*fA{x$fB*M?*Y<h;QE&>}DiIRqs)Dawbd6NX
zG0p!meu&xik;-Ojue=dn4poxp8|62=jdxbLrP3`2(%#!qLORV5Y4J`@=`}@oH|^zq
z+B0k1Cx84T`Oe*Q!TpU`<aEPNQ$OLx*5$COjSoU<FGt6wn(nqO!>1j2XhNQ-KXQhZ
z!#8g1Ub%KPWr(8k^eF4G^4G1gDchCs>fx2p(j1Ge$gyIWu2dI23nzSMQM9#WZ{NAu
z9cQ<)0NWtsF#gEr9G>!06|*Q5Irm5!TW^IwXT3;S`q_u<m{X?fkrUUi-pIjnIn?mr
z>%D9F4$if!DR;|~^-$T7rX4?cbgz_TK?|rdHr&>3EA0=>wJT+R9!nbDxtjRouyo`e
zD~C2BEel}zC-jn*->BPh29A?k<s}YfWLRWlmBW8CjeEnBeksaet@;OVqFY#C2$P0e
z@z#6eCynxZ-ld<<3&#bsK;}>Bvhf%>PmiQO?^d6+^%rXwQ<|AC1vW^xt-`4b<`dVF
z15UY?<Fh_lqRlTF372-2x$!1^VavaHdY`a`z2M+8-BKFY`}SV!HgNi$w7sv08(2Z2
zWiBquf5l(8<}P<-X+7ba`3~fQeg1O{%U>M%tP;p;w2+2;$ImJZBo=}z$@V?ri+M2D
z5~P<8vK!>kqPY1zku*+%g64084tpKWP(p?xv6zm356()1v`DUkJB^;nQwO2qUU2R7
zXOQsXH5wXta2!l=nzC&pQIqtHOUwwwkp}gcY?)xM<UlUCP>#l$2Ey4w+?9L)8m~I8
zAx^>b7o_SpVvo4>-QhavL9qGL*+8<DcnU$;%0K|+8Qt2MC!R9GQJuk~b*DjX-dBA#
zd?3^GI>yo|@Cd^$`fMj9kFmZ)he^Ou@GRtWH2bqCNZpzonx)S!PDaF(1Mu2~x1bQh
zPczfRjUHK^p`08<IY6Q-nRcOQ;w2qfKM4>2q^s#YVU#Hsh3S-Ul2^WeqCPG=#ZGpi
zVaYzTTxbxY{O4G5&<f4vGyhtc&@SjC>V%hY{6Ilb4a!$o{O7eRpa|0?w-GNm%R^_e
z1o{%f1(-03rm_p~TB!vj_wqx)!lA68e0fk#TWPy4B%Qn+Q%@-0vyH9qBL)1fEaT}E
z$6Xm`eznrI@|Fh#&?xO(y{4lry#S2^sW9?D34s6AELgpMG`=;%dgf3t{Hc5tXzh&h
z!AgQRq$v*u)>7807)JRKziTxnP61JpA%XfZxX?}=f>-)Zg#iZ^Xa4CxSTs5f16y(>
zQ$cvBiim+so?ACjnhP^!@vWGy0crUaG=*%j^MfAAt9(#T9JF<VW@rVjvUi(0{2(8H
zJiF+J7HD7qk{HcO*m;L<w!E|IqIuv6aFb}87G7K23S-I-8Oc|Ns4PtfUw&|(;SJ0C
z$`Ag9DO`RtzB=H%bB&X@=iLiF7sNTA=FGc%rJ4E2?TmFS@AW&A=*N2j8%zG6JYKQP
zFVJdJj78^-*N!G<{MO5)*Lv;!oS!mr>ptFI{Qa)*ZHiMBZtMG!XH3Qs8r&)*LD)=H
zey)zH6L)of6p!oIbCPCOl-mho|2?y8cW<P+j!yF3kytj2B(FHiTpqOor17Ughusq%
z3rNBrJ#-RjMt0H2!LKP-?#T~&oi+&$Jzyz93O+ipj#NNOq>N8Q%&vxv<cj!cG#x<)
zhNj~Hte(I)xv7fMRG$eaBMx7|{4yVWdz2Bm&JWWoJBvq!Ny>4U(M}s|giE}4uG)2Q
zk^Q2p^pSWIr=f@mH{VR#c5qUKYEC5W*M?KB5MgxCU?p<;`O<~Jj?@b9SOY>gbO6I0
zOrvrO7E8n6Oc+<FP7IAYpFhdy-P)YMzLRikM0E6O#C)|5?#q0Ex1V*7B?upUnzaZw
zZ`frsj=Q!XQQeO-7mU!x=uo|m(OBRGKF2cU$xaH0dxmQyb>m#Sk}}3D;5?a|w{BJ;
zp#W6@Ou6jV_$KFKSX-h#l+8{I=Vnlrs+>f?>Qo*F5A`kUK~uYa&~NxP+XS+PTuMjQ
z4q(=+4vbUBXu(1i1oy7O0g#MD)v3`s_`E1_?I7geJQ9#{u__v=a_|*aC2~I7fg@hq
zOWM)0ri_j=>5~`5!U$hR=z2Bzn(km(co<9F=Q*wP!;e17cV6DBGNKJ$4^B&=)tPfR
z1I>%HTYX}ce3Edq>9k@a3QYak3F0);p<uM5R^YfrYi)(nFE7-(@#f7c1i$^8--Ult
zCt0sGtZCLT-RlBx@vR@>+I9j{`c5t4%lNwnJ+sy+^fSeG7%PH74W1)&Y7xP=dDhV%
z#?q#90$=#8x9+5>auj>kOi^r|usn7QsRwnOd=np^I&oHgvU;)Fc42MV04Af%c}P8k
zS6$P0;a}RI!KBajz76AVZH?V!gSTUqNj-Aon@p+mm~6@?;r-^7vPfq!vwX=QZGX{;
z5}OKXoXUXkR$chTnk4Z~($3M-wCCc#WKv5Y*qzP_9CX{^o^+mjZOS%)V@HsO8}*;~
za9nYdR@<*~GM@u%2fryNj4iAMv-oYoTi%uDYG*4viPX}v9Fz~_wF4$X-V-03RYt)U
zleZ)@e3u67`2B}=lZK`3qCJAV25nwl3E2?)Ol6;|0b$Hl7BY1Nk#z02Dkz2~Oqft4
zB5wEg@|9Lh$oyZkR_is)X==qY?LQ-|;o_li<Za*+o@beJFqPP4wsZ_V(XfdL!=w${
zSs-S_eG;>^Om?hTN{*anlHE(0GT~K<Ql@pfB4x;N<;pY|lIR9jXBNmhuHKVaKz?MC
z0{R`#3Fij;&uBsnhtg3JpdGPA0Q4LcBQc$5;uK_4pZ(!am~c;Lqi+XUA7F=tgeh0&
zM*)F<1y<r{c<^5IwRWeLtK4<S1UdN9*3_H&BwUV<eWlfzA)_;8k~HEqE%YpwFiK*P
zc1sux$2u6(dpT`2e`Crd)Gz|%GjTHt<@g<#H_ss^Mv0|8e^rpqsEZ}15`s=TZfg?Y
zVoJ#UAe|iC`yBdKX$(zFXS7xpVS^`}OA8Dc7q)5?G#iIgg&xY2^e$!K(5DI(7`e*l
zNweC8I2FS2s}(7o+rp`xi}2tkv5r)F;jI&O1Hg3f<LVa#DRNJbbEsd+?%bLOnZDA&
z`Z(dU9Lq2MQiRcO)icz4U|1l9vMB>zNj)eVrA>suZ+a?Na4(N1;=*B~Gzwk3Mtn&R
zLf@-77Jf{T?C;&q7oN2#N~-BS2kkje$C{R8*%Ea(OI~UzNj~^#-zliUP@LyoOQtST
zrr=ZBp=o3enPTxS+8yMhiktxnCGJWVtd1~)lA#2amxG73Yu2b)q)kb*V4kvhC5o4|
zOS`g8x`P`Kv&^|C|3rG>D&5dbTrP2nZ{upy>TvMUXq&08(7i4<EGRx6YP(24k3s+k
zuf*YLaOJfw(ng-eTY8uGLFCGV!MlMU|88+l*x=tyITAK2W-Mk+oWc!Xa=UUh+Qb_k
z!Rbi`L*Ol()QK{*zleXa7aqSvD!em46W3jcI4d4{<dS1^-IQy^X?Zu{RXAq)9n6IH
zdBFw8^MvPl(X6zb2sY2nGZ<T#Gw84DwxyBsPCAtqE9@0Y{OU@GOaM$}V44c8#q4RM
zBPiJm4?~J;^rvAs3&i5bVZ9ipdzOu{EzKxqzZR11JZ|^jM-Ys)BDouFrZFB}&gebd
zs-bdr&g>W<1%FQQ$S5YPl64c9c{Iuj{xJ905i>@;cB0@d9V#fbgcR!d0Yf>2A+W|j
z%)8mwpV7Th(bqYa`t#2|&j`Yf{4uv$Monkvohew_B}j^E-i2~ta;p=7PlD^D@o04C
z)JDfhqmd*8x3t@aAl*u<yylEG=qe-Xs=HE>PNZaH#5C5`j1w=xf}b=LM?#{VWb%45
zYY<$@EO|$&Gc)NZDe4-2zNh2lOxt;NhQ3P!=jG^Y%DXxmRu^TZEKmk+-Mo!-dDj_a
zjz6Q-mNz`GC?`*sbjk;FiohE;uPa(j_h|4i>qktNJ<8ex)*&!;G)}dZ-`(sWQvXTX
zye}t@;*#kl*0ADPg$oz7_U0%!{KcFf9_3ql2QK9fEJWIWnrqU=yYIWel|`?(2OnD@
zh4q7m31_uhRz7g6z=sC48YZpuIwt~MK|R9@&iJ=DroL|E+zU#_L*Mj?!u3UV7~P8^
z(sf*+vGqIqI`3e6oDVK`k!bLM%W~+b{MO-=4uRkDH0`ix#V3(c4@oBuoO;i@GM0xZ
zEAyRof@1J)6p1IXhT1~R!gu@5+wD9Z^G?}#k~KRMT&C2rh9NV@V>%{l8xp`eFfbkK
z6>6Odo$xL`97;R>8@pjl2lb@glmU&kCl)#ihBj(_RkyHH+F}hjq;&X}Rg?}^sP*lb
z>v@KMEo75W@L1);uifqom-u6QH0#x{p#Yf5;LxhdESys>;yF9y=xOi>o@6P01-~wL
zG^IrBPafD$85ga^?V77eJaoWA76*IAKNSBYXx{jCypERD!|;-}r&>2!@=aXjkO$hf
zekDpg&tNXM3tJw!!(3_6l`EXUs?5lJDI@5*%Hh5)X_l<xmw;LK1Gexn=p4$Kyn4Y;
zU>v$GS9}sy=bkWch3!5s;rv(WR`?QT>s^^op^x;1zR44;zc$x0&oZ{WDd3_5+AR8f
zSGiwH2?1<nww`yH&YXl}Y(lA{w!CA03*#`M1(gIh;b|V0iIWnd^HG`B78rTyz_7+)
z1giKf51P-xk&k?3<Q~C6;P3G9P-KchU%a}oh|zYRavQ=o@r<T(8*{D+KIgDHQ6iVs
z*x*YjIi*qO313%7C@MrY<7YzS0mfwI#jq-A#<=JvtPLmSKeeU|zwr%>;5LPtc!Y1$
zNEbm7;#bLFk*XjgJQB7OczIt9TmGxA2%paR1|H&6Xr(?(!t?~n(jqhq+$F#?;tKx2
zq=ZmbZfE<gXF|TC$#m{R$W)xkvh|7n@&?Zc*J{aoS!l|PzfM`bo~@M>v1`{7sLXgG
zc?|_XsdAw~2*9tf^CIMGI-Lr+RFFIl4kFmosp{x-_U)KbzE!{Ya;$%eNjvb=(kN@v
zQ$`@1BD9eu3~%Th7cF+9tHf)$Rvo#A9;KD|!L514U)`w}odr106R-FL#;T)r*Qv5>
z^#C`xD71U7@rOiFT4MR3m{_1HFN%b6DADd}QlY#Cy@n#b=rQ=d%8fi`ypevS!Paf>
z@OH%;?s6&4t&c%F${yOY#?<1F`&iR4>9?`~Iz(2U;ResF5~_y?n8Z=%gKsXyG6JTQ
zI&b`_eO}HYP<XK3X}chjH%|C1&EOPX7<G(3NWsz;<plqR-s@06ohBYG9|w>&KX44T
zDX^;0R!6&qn^H_CFmC)No%co46=jB_H+)z$+9p<b;m;TKEq&mW7s1P=e0a3!Q|{)U
z`-(>vlL^9=cHserI8~13b9{FA3M&lR=UF~y@P`jC{uem@HYm6|Xt7Fb-tj#=YB>)K
zZ2@kjUGjm0{M4a3t(@cU5??qc|1A!Au3Iqjvf>KyJYzK4`dM-MzS2%uO%uPJ+dMCr
z3y)o#pX$S+-#Pp@r9Y+LqQizx+j!?dyzu(kN(cxQHWwioWK;%WAkJ}^O+(gr@85gS
z`S~@1ceoEFM1$5z7@e?scqeGP?>?}~d||-b`6_2S`>@ZW2nZwG=g+g-;8N~sv;<sM
z?cvulWPX!)F`fnzfh;grk02ME_!zgMEBG3LoVvu-g27E30)et(Uj3K9{AIuRd*9L6
zDVO81&9lROxe5r$OD7$b7yrXL(GDkGZTu3Qp=a=dYv@z6@C<Lmj2{I|gS7KRCu(b-
zFdeY9B51Ayd>njPt<=ZL2af2B-YfxKVV~!&X}OfU#vWK#YVjCH!B=oa>9Wfu5+Xt_
zqnenk9cG61V4E&bZ?|Gz|Mjnbv-h9=x8Jo$%D<bA$zs+D$Ka*Dj*kXS;exiz@o@cP
zc1k$-&F%?hagt<V=gf&(g|dr4e#1&O6p|`xp-Fcughx*F3d#FYl-XxlNMcSOiHTk~
zz5C>yc0P6tcz0Qs^vNfmR9RN0b_T(-oowzpH*~_oq!@n^v}<ef+$o*VZ+zU)dYXJN
zFmMv3<#5D(&EUyPz7JDIFb`4>6kBJ{ECoxIQ&N&mFP$Nr_(_XSwG1R&o5Lpxs8hFS
z@~UqVg?83i5cH!>rVRv|%M{bFL}AJy%m*E;MHuur$yzNmz(bs}jK6`CjD@LR@Zq&w
zN4eUzdCxc>ZKT{e>%|T%77t41*E!AiesFYth|@?{XTu`Zg^y`CRU$%jSVlKZrm00t
z7~%w_z%~VEs%j{laG_vY^va<jONiZgV2$Ev`xF=aK+(+9T+%5czOt@_HR(on!a0SO
zWwx>@VWxeS3><h&ng&NxHS*9|f>oEhx?5#KXB}u2f5^=6G~}eq;A$#rt6hw=V6duE
zDo342FOQvan`d~jgeatZ3<BEGhxfESIB?>SX~-$f@C_<(hfn-_&oc)Wn3RVAD*ns*
zHTY)a*LY2F`U4LCo#LGOYnjw&@<UheW3Vvj)V$#ue;ABGhdyH{>=c>Od1gai%3~Z*
z8C0i#%i%fc_^qjwEuPCy`6_q8nsFO1@7#j5VApf=)3{)4{C>)h7nWnu!aZf}rI6Eq
zcYlq7pZ1!sk!Y9s%6OZ0BBlVfRODUdVKk=L#ay;$@+T-Un9ijQM4wOa5eSJvponJ_
z?rI}R+Rp!`(SzsUDw1CHkL=xzZj1<qKOm!l1(;4G&bkwMWlkp-ACYq7)%WF>mjKpr
znigW370+6b3U_5pf?fkqRE=mIl7U|b=z7dEL{BrBK_cXuu!h2>6H^X~%vi%<yKm3h
zh`>U#(k&^+2yd9~;nh3{nvDE35JIRO=fszCL2*In`CmMWG5m6^#yZT_S=gHV2<-N(
z5&B~kDQ-a&7Yc$S;j3^YPor?(UCF?=<+yao6oQJ$d~ml!<}Fq!<))a{;*wxWl5*^4
z(IA)=xZ#r?nIB%N;umvyc_`(m5VIvX&vgNbhNj>uT+k?aD%$?FbZE~~{^K-(c!XP;
z<W@dEZ%b>H#Q4|xkunh2rZ<k#SymwhM4f~ulvCXtL^)D}6e_O#pEC9QhA9L`-J&>t
z8T>1M^-CeFGTqeqk#+!z@Kl!3UhuK?4i4)ZZ7lsVMKtN#NhBhk(D{@WRX7q>{z+ZH
z*&pkOa^!;d5~AuC9%E5yF=#mYQ?$h8hH-K7Nb7EvCt)LR?MyvK!Az%Tk@JHn?iMtE
zorTO63_DOyJF?K3g+z%*TN-Icn@x@@(^6!<KwS=jKh}5+g^~iIOtxFgn?fs|=}hp*
z8m_y&a#z-tJ!QZ{+7U&Byux>Qfzn@^&NyngZAqoXfh)Y05OhzxRexgiieKdwnsd2>
zvD#$)W;zIy9q^I{gx(TRe|4R-$+YR^zu=W069%8ve!(7e1uM^R4D9Z<f)`u_<lbw-
zD}U5`gtx<nv-lHt&yBbI!HF9=dB{`Z3>*g^3k=0}VZxUJfQO~E#zR`q%_GlMsNf#I
z$^&>#`0*Mv$$ijl+;`9MC%;;T<ad$d>)I(MBEj7~P8iPIrtb4g;7Rk>>b5j#8P&sb
zv)<P0x81*&5`yTh3>oFxGP=wv7{frcZOVs+VnksMb}>ZTZgAG%hrk?-+&sLF&4ETo
z`Ctq|H#(m%2E`?ch)(J7P-iGD6E=K;5n+;sTQhhX>2cs_kQUk{o2Dfs4!T|=LMW{K
zJPOF=q>)h%6)VGGGmJ<iY8;6>M@={KeF-D|k3ars@1u`C%BXpq+t7L7n0VX&Ax_o<
zU|y@w^As1)(5ixbywx`dj=RQD_VzpIN=b02n=&~t$=65qZ~dxg=s-Z~fC)*kJ=-l|
zI>h@o8DY{Fdg5#n>eM@$3togRGc#rHX*%7Gij*B5Nn2VXY@CkVsM2MW-36loY+hdZ
z(c38ZayoT&N!Y*h&b#f*SvQ;j>8rd~GfGz<>KD&Ey_$|N&&N-#<A@+DE}7!8ZX;F`
z>luVwH{h$vU+$FH%N?cOOgVKFE@ej#C)ip)L7=U=t|C%_r%F@~H>NzEXPXFNszY{?
zt5aN+$8H)?@v<odbb9bGTU~>TgWGhfc6YF3v~y3I>!B&yp(P!($_LuSPrYNaU{yi-
zJR@dPjOZoI4xjruRE%Yd6``32OLXP3ucN%~dp>BW9Ho2mH|=ngMW~MRK6x^%Ug%|8
zyi-g6fA;Qc-I6265_D)u1~r-aRbTYX6ZA~K|C3ag$f^`%l1a_bV%ERl*s;%HkV#h6
z++>!W<J;X$zyKHw29q8^F22gjcWs`A&gYpPeUiA0AD<nd+~qZ+8arY&BO6oaf_Lz0
zU5pGcM$p}b-+@u1Bl0xisc+$hH4cxHCnMC_DaOXX{rkUlMD+8||6F<e%U`~TTtB+<
z^;ct*C>=6*%hXBed>Xmu(7Po^wq5lLtL+Rz$5SV!#-E(aKdHJP`BvnEO!`L5wcPK#
zW2!20n6r!UxjW~QPJOeh$SAB4J(m%<+u@<GHr(OCjABOr#PYGwnVa)^z&L16{_w5-
zIq-#Z>z6#`pF6{Ijam&KT~WifhSAiI8C?nNz{FqE+SPG^0e|UfEVNf%1Jm=l5z_Y~
zeWDgO4RY&!-bq>c8(8i&Fmi7hANa}}4%P36A39uHKJEi3i4l`DZP0U9Ki3Ir9bFIa
z$HkbQq<+VhU&%-Qbl8o^eaqvZ34L|$xw@+?cX7OzKZ#HpbDx7Cd8|)U53Yybc3c+r
zWqi@q)WM6p2uZfdKiA)iTUuJih(g~%+}=;L{tCMPLq<rM3x^c2XGj%^BjZI@FSbIQ
zLNdI0J6q^D!hEy|lRm}Q!?D~C0ifhnwC0ugX$RX=dKTq5Cm04$tOLqVPlgysJC$M@
z#f}Cf@7OFOu!lfaFcbf33=1&i!6?#bF-i=K<gGZ8EF!FN*U)W)_awveUu7s+xJFMy
z*HkeMB=sGWp#=bV&M1zpfE+w~&yW1kF^y#sAC7_M5Gn@2%Zxf`Y#e8On!`>!f}iBQ
zG+ugej->|kK^ja)PBZB6$aF&+z4B}fks2jQ)5b118TJ>4)B97I&8J&q!1#f~T!ykW
zTGPW89}V}jAk_{CWmG&_O(_>69E?Y<MI6K#F&Ymr_<6>cYuU=JG5FJGf6k)GDBR?O
zz)qbcjwumxrQt@eoeSu&W%R3T$<;`}0Y3NczLn=0Nf~(^e&6h3SIR1lMdH6;+$Im{
z;L%2~xKi>%qmPssqQ^%@i75m!_Wg$pt}G}$jJca|4t}@@&noRa8%0qr96$N>3Z5&w
z+jyEjSQ)AU4jd^f54n>?<y3y9Y&kISis_tjhZr7C@HG+zk5@68X5@h(V+Tq*0JUF1
zP5~r+@qlx9;GOFZwa5VZoN}zUkl?6x80nnGVM1mE^M^F@-`O%9L*&yxecGv^KmPF#
zoo2JOT^(m&&Qy&pqUyVRBJV-T*JguwFbhTDRj`xg<e75oh>d!bu-55xHbihf3gh3_
z-$yqeU%CC+XE8#uRXl5p-btROF~!I0qzoU(rp`?cNz0Utxc6eHG15ErnYinWaI}8p
zh<v+Jp29JFB5Q@2yYTKZ!@tDIZ~5BtTR{90MvhvC#6?H0%Z}grgC+jdnQ-JRzOr*&
zh@W>d!k=3><W56eoTaOtnK*ts6$Vy#v}u#Lv@8tw)fF3O@6Bv~@I>D7%Uv9KY$LvD
z7OrIH&)z$wTe%#gCl#RjNm_9iM$0tsDct%m91C;H*~CXj>6?uF))j`dah{b=-ws%(
z^)eG-MjhoZKEHDpI0Nb2^KbC^zktqZ6@DdeTChjvEdT`@SAmRwt8DzXvZw@x&fkCk
z!^tU$8YlM_cH&at6uRw%dIA)be59?A#mq3@truWFf73+)bRK?maQ(!uz_#HKZ&uc;
zbl~g-krfht(R>XvMQ~Ul#Z^!OrjbiU)#wrQXX$Nymo*F0-pCY9@krD$X3`rzK@a|6
z&rE$p@nxG})lwQzY^&fmUHBGQo%2lqe@b6o=<aAn`13p&818Q)aT5I?nYb3-y-0&>
zeF3z;%l&%r)L4VJMhGbEe~178KmbWZK~$I;Z4H?V-{q|o8Ck*^C1AjN;=0egviMb;
zlwtUd=Nu(tr-O#1AD;`}8bu>HoL;`LWYw!%d4rd85hjLMSF`@-y%;<<^F5w_D$W|u
z_RNxxylvx5uEpVTX&A$a{NW4u7B(Ape=lo^?9S?9;xsZA6xZO(^HmLJ43y@bgYVES
zZUnG=fei<LGI{TQMs=RO9r+t0$d{2#ZJ1Uto$&jNjxk1V-5LX$Y9hya=A3)WIC#qQ
zTo#oNKISH*Jom`pqpWLm<sI$_gsbxLfxD}bbqS^KH;aN31CO!RhA#eky2{`BrFS3a
z>$WL7WnbCHuwVp-(wgU%$RHV<r@ppvZ9Pf!<SxGS3=^=Nxe>3HfqH7x#R!M;`YeWk
z<KsVzVd<Md-+cE?%kO#O87*^03`40#4Sb3dLzQez-ISkaJj92|KWUCXi*R)+b_DeM
z@3S5z{BhAg53=hDeGeYI-8y?ayI;^$1Ge#;#gmbcT;#_{)v1h<CoVC}%llMnBMBCA
z!-_tGm-Sg-wmu~c-_ISgT-@j?^&{NQb;@gkPrsM_x9)8Od|=d)lPM2%Y%*zBc_`oE
zqjW;O4=pLn)BtTjlE15SS6+h`&*GMQYkB<EEj)Men`Q%Z=fBQ#Wj`LBG6d(iHr?ux
zO>=%9ULA@)0Y(>1*Hk9uA3yYy7kT+Y7kwn1yM?dZj%F<V^*nwq(@{zT6WjnDdU9X4
zPmLGPrQ;NS7I*#r7VgWZpQA;bU+R9j5t4#ftGgK_lNu7qsRf^c)=N-evu<D-2^B`e
zu@}TBtqx!2nbI-sp~1*gzDc6NJp?}CqjV#57RyC}%>3;%T4aNewNXw!(>PP)#9}HK
z^`N4izE_fvAGtDYDv7X@0fCmhqxXKMlitY!HA?8quW~?H=9wdQ)3Ani4Hw03L1dLs
z?wpfDRX*ER^=@5^q{nDsD5b$@?@vYPR1>2E4NK!F7C7js;d*(>AWOpsK_AsJ!=#bF
zoiiYeBB0-5-ubexFX+x$5gJ9$c(R6wvVl9kwZaCM(~u~<lCN!yH9~^hGz`r;KgtA`
zMhUMGqg<xE@Nj9s%f`&lq;Wf*rjjQFea`yOXqtk;r<SKiGx*%T^Ss0CMnIGY-slzA
zSPR|UlP56%g7>ShEzV1seW^haVpJxd6nN)BVLZt6TBrE}mkf^58W>`e$<h}j8HHde
zPsWZ9f+;`4_0RK_VGDKfa(lPm|M1}8d(M$5O)-j<yT7C}k|aJ_l}F{Y@<SF?&5X!|
z<`==YvK2$A$7v5CWhPvDR*8{{7|=$;;EQMKmgxj}yBI;=<Ou)w-g~dck#jH5Y=j9N
zDW!p#{H&vCV}S$lRz`}MK^^4E7yOuAXW|S`=9u#*J;d(w&%cP#@uyB(efsCmder_8
znaVOUvNFRc#E&-a_-!p0JBxBL9MXr}0)NhHnZVYE_!Z|wg<-J@A7Wa!977h0x^Osb
zzw&hsZL=6xN5W{whdDkTe{W{NHG{-jBl0VpQKdXMGFRQ!x}ZL!e!wpb!B@1b;o(ka
zl^+;)xRwb99!@xS?$1Yg0{uEA%I2zU9_gglL1=t_hrj&Q>*AAVPP)3pX+7tcOWxq1
zZ>E?gn&0A;`>xBd{>XC=V?3fY?UC=MIf8my8vByGHqV8fOiuKdGjHI}!b1;u<*zz_
zyf6Q%ZxZ=h%0t*WUuvwNGvS*qg*$xae~AV#8FfawlHu?OBA(qj=?=bjl<uIV;b-tB
z@_+yS90z{65wcJ){}2|6qe7tI4?l%N09ueKq*Z_#KNlf#j>igBrzujjqXZ3WM&T;N
zXblBw6?78hbx!hQmHU(3<}nRWl$S=mf?*UPi7AoXcP9=K2zw1BFNbSBaDZE2_Y>pb
z3y42e_jX1Vd^g4P#8){b^kEuE7*gWkTXd%<M{rd(^l{d>P)fHmok2M`Fv-Y7SEr|K
zHPp|wawVPy5|(A>L1T>%JfZv(hq2}C3yl}-3}dTO)3eP2yo^SeLV58bbJ`m6T$~1I
zd*+lae3XxL43oSRSISq9o@-@JzFn&j+@2;*jgT$x#tO!HSDAadYtk!+mT3zLBj*wa
zDdUueYh0B<UNz?Ea}sOMUx^<cS{&Ptp@se)9+k!dqd>5T5??!omAqa&PUDcjr5)WA
zi*17B_WgYO$Ku(KKFK1~z`dPCrA9^IhJJo{TeJ`SIx=M>CFizR4t-TNVJq*ZgDCm#
zI12sNm9*i{h*=k9W)uQn@LU<rbj}!sTOO?!foWPu{*0Cv*B!qeLncOx^!T(5q&q&7
zkra7&H1*(F;xTHSckwD^%iv)kI6>B08~muUj{LsL^ERdk#_N`6)$~C;T<|&sYbMul
zE`wT8gd1&{JdAw&{`Y^l^4X^`LUL;BS6_csUN|&OIbj@r23V%UW*3KghM&G!lZ{$}
zS{`$cpZbwZ@Vjypxx6NCFcNqD?G`Dn;ej$IM-0QiWD3aiq>)P<3L|cewwqZ$qkgUd
za&ClFCc&1sEBtO<O{Vg^er-l&<5~P_tiaEE!s~*Axa*2LL5nM%?zx7h@FZzVxCk7^
zri^!#Pg3y3nKB$>NmRI1cF*7|+tMNaflKvdo?#J4tyd=<4p1DN+VM<0an$=K;bUb3
ztm6{*{Mq|uCyi5COK+YzI5dyk&s}jn^Vs)O8Q`r8jL-&#k=ygpBx%Gy@#w%6r&?b(
z>=GUDgBLmLfo+LDC6$N6Ek67Hw|d75?oIT*$?rePhhJ}mEK+9zcrI8iu!Jf6El`z!
z(xU_k1_ErG?1IX`ql_Zb^`}%RmRGj$CZ3)gmlWCZJB^#SN7wS9vG}(!2z)(Cz00Yn
zl(e3Vyj09E)9@*#;;Zi5@#u9TC4W7y<b7gr;x~@~F^~S?2OngC-8Vh>$#jgJGJ0}o
zvZh5Z>v{4-O~?SaVQA`+=NYN6xbNz-^p>KmgKW14S9#(~E~7**;So1{)xgBiNh7nP
z5$NpHR7OIOq(P&|;nC@nG=`awn6)ACGrX(}Uc;QYj1PR1ugW~V@|n89cc)4|$Vb4O
zYWcwj?{~qg9Sjy5)+k6EjheEx=n*^(xyxL7iHm;0<MKXFQ$|H4`B5D43x82uzURS$
z#>z<e!>}wL6QgmITZm@7X;@}4rIDWb@~?(Oxodp<N!5lZ4KE%%Jl17+{vm5T7;p^9
zey1im;iVBD+cKT9szGw$hqVNZ6=ebT$lP%pe$w7$1Sj9y$|HB%I`PJ_;P|HTM?Zc@
zYih(U5!+S8r~Kg6atvSaxW+~ZUzrr{8BXrAkd?tioL;0M&B#ZlHXi3YG5Dj=++AD@
z%WsdP(d|m<$I^243S6d9(jcYoB~8!?{88{NJNN~6^&=6Qcln?^k~h2<vgBb6kbnEP
z-(C5l3r`u7^WikR)|9WEL{fH}aGuc2kdkkFhX<**{L~NOU7HgWmKWqA!rL-4a`Qa(
zL{&F3RURe{+}GeTZ3fm?84;m3Iz1U$1F&V@a!whWhN%oEOkNAC^)v6|mvoQ~Ef~-`
ze1r*C9<Fupn2XQAJke32N7rn+%W3JgjtB;}^hUYJM)RuQ<OJ8tu8nH^rR;WV`Gx-c
zw!`GUU*J`a=wcQW-@KD)F!0Q%+2ZXkzjGWmEXgmrz^LIM?>N7B*1Qj#3rAr5&tdc8
z;B(o3<Q8w}3Rp`%Ym!<f@QXjLEN>VWw+=kc;epSISDln}<5Ta#Jueq&f6o22@SjWj
z_Z0RYHbO9%P!U+ysj%XI60Z0Yj8#y(CV{2iq`6aM-ES4HGe-1e4e^cvLD|ly$0?cX
zQREsf%CILECa;B0krHj_fbWz}4F?=wyfkg0!DGry9C3?uZmL47c7;l%;z_Va$ov|<
z{U8evPew=-dUqA*NfuNpLGaRuO)501OnS~a`y|TEapa7n_;&Q9--6H^j2k5;Z|?)4
zN<d>2WwTpkJ@co$otmk#Fd*<@l@en>17?m~IT@X(A|9|!ucG0TH1f8PPaGF_KaHFQ
z{efwyOkJSef>+_Ts^UMR$Vr^e$<VkCj&=`(|G8+gm)W_%(7@~M$+aO)*i1W(Vo5<2
zS9#-^qCivAB(W(zp44OMPnh)*8a@{T*Qf*KjQ)-=B~ST!{yrlKWW?Dc+_R*)`+bbF
z$dmHH3vl$<t>XXq<4<acFh*?2wiBmvh5zMaevlQqJA$Ijk~f8Y>QM$~%2DIrI>7J@
z4P?A|WxWJk(KAy|b^}d0m>N(?wjPxB#LIu{j`Cw**gYkU<LTYygD*E8W?Ju{4c*|&
z6aCg?Fo2v(QeL*nNi_`LY3oSr@C}K{uS4*y4{A?n!a(H8GsZU<^1y#a)u*3+mfd52
zxbnN-{Vw0~VWb*uVyurG>wNUwlYq&xmB3?grGs&7*ACT*8`!ZZ`1d5yq=`+GsnV&d
zMIpRA$qD~#{f7GW8*^sQtpwr;M+Qwzw#QtLbY${g9g+7=Wk%+ehpQu;$+PtXujS>Q
ztGds7($#QE9w)ye-cg0+^hPH>hj*?%^Q-@HpD3p|o%T+6>vrzsQGB>1QaBLWgL8Qo
zoEG--fIoR$|K@qTmoI@qerjwS_r51$M;wZ4Q<YxumLGdxy2M&H)CD(q!r`3Wjg<TH
zaAED8KN_TKoW!N~PH+-d+0Q){usjh@`fHI({9$->VQJDDT|egm6vW^AhE;M(^HR_#
zr?~U>_x)S09lSf|L^*bp^SS?<o^N3O7N!Uip8#9}(&kcp0^zQ~N*t8|9157*Au>I1
z6_<Y-xd?<20m9fX&+1i25vZ_wm>o*jaOaOAEv$#hD}T?PMS<u|`UXaNRX%A>sdALY
z3M^m&uRVZTacu@vD;*op?;p;vpzlD4>>C@e%(tiahz^Zu1u@Fm0?7~Fe?P|1-5!wk
zG>c-%x?x?7@!M}rujyt!DyQdYetCMZ)*u9zC{hilZJpPzKMt?-VCT#NdGySYQh{%b
zwqukcI9TLnWJ3kRud7#WJ(LI7V_<;cu&0~hV~sP4v&KhoQ~?8nA%REWIAF~IR)4t~
z;~{B$?);;VJ}S)`k#EyDX|OaD_hN+numAP`bOh+#tRv8HTu&M}DOWp&46lRt`}8v9
z@wP>O30F3&I1~gtjoxs1tYf4pn`@b(YN>0)W5hYY>bvhA79Jj6W4Pr`A?k_Q?K8dN
zX>{cyZ{;TMDxcsdA4Y<awYOtP@vt6Z@*E@Rix^Q15)JdfCIeTl{Nq3VV--3B#fXNF
z=Di;|_%OWshu{7~r*ar)8nGKGb9rhW$*@JR9R8{+rtsj#h^vAR4GxvVI}Wcj;^^C{
z17~SyP{1-}@F3@xB!k3bP$#Frq+Gi$An#*z#V9k?5!tw(T|B1z81476E9BMN#V+Ol
zLyUw!XDvj|S&2WR7mZ-Om~COL6UxiBYP<zUK6XCg^WE$qDIE#NGkN?ehrBst?w|kZ
zpSqyf2ngOf4@fzyN9a+G<kZ>_dAv*qfiY?-%hcuL$Ft~Hn1YZ7;oAr@GVy`K)9Y9B
zQNL?}5j=YUTMSOdg>-H70;BaX<zfvCTETdfqw4?9|M|z3DFf4X+%8^D-H}%IW4K2j
zq>PobG7iGczD<5mVO+h+A{I17i$sTHnk(zt@Qag<gEDEyOZpIf%G|TY&U-(!7-QK;
z<I|M$-RPWacVn>S8#e&-(*wyr7<Mg-juE)<1k2PT1KQ&1*45A<j2^%rJgo6YRu4MJ
zoHBN1$oet(8E{~)!$UZh&xfaXm-KrvX5!hhYn@KIcE*xM_XZC-#qL?!T+-A(`I76C
z{taAo3D=R*o`??z`Eueb=bL%AM(b&07_Rh}a&*i-e4HHvN7KovaIKp=(ztP}H<G5b
z0-pF}u=0~T{&3npFc*)~E`Wb7aGNG~{52x!eWDCJdGQ(FfXj0!3fw*)d2PC-w?K}#
z8-M-E_tQ)G?D@#|%=?!=o~J$VI!`SBawK6ta&PiWq@`!+mTqdwIRuFpK<*1(V$Tw{
z5ZsS=1ff8Ww3}uWNFye4t6G`T4Je7LI_s@*MnTeqT)B}2dCfy1B#z-je@-@}?7Ch+
zrOHJp;H5W0nR*@~$8eeN%AC`np+FynmHUyV(?NOOo_b**A{W@s7(fd{U<5BgdSx*h
z2FCOVO#7^0FD__tNejO9Fyb&If^6vKN8Uqgi*y8Cqww0$n8H>8uLgH~mnQ#x*6<j#
zxs{3*z};S&h|QP+ur2@}DJ^sMXs|}%>Z5Bh8caLPs0Vzm?z#mT1W&t5G~5>OX-w_L
zps;+;<=OM3i()7&v?NS}iFOrp<s)boS-c0fF8C%z=+8Yr^Cw#5qGcO{g<R<g2cBe=
zpzw*IvPy6B03*;6KUEJMuB4fTlY=vVG;q&@$g3EU&(dH$iUIX)dh8UQQ426n<(%i4
zlDXAs4U2*Agr9bv7y;4v-^>U5Y-43uB&WhJ52AqJ2`5TN15iZ<zW4_&=d>u}t|>|R
zkU9LtZ;k)RMDPyGsTb-^c)j%Ij!)?B5#@Pl-bcAWOr!KD(?Ui<?C2qv?ae1`6N?m=
zEUQ49SMsyFpbKDwvk?<zp>gQ<R)VV$mcM-Q#g)&$_`H|*M`=(sh|@{YODFG>Pe%<S
z%iR^DKDI7S9Ucl$5_yk*N6tJ$EWcy~eoENaY;9ZjPNQ%$8kx}K*Eqqs@(@`<D;($@
zJYYyaiX1q&jC_cvtSt8S*L3kCgUV~CGvG2;<SF(3nh~M+!PUvluQIB$u;4Ud!7zw!
z-KgvqaiMg;VN{2&rgyL1nzc!cCvkuHVdP%E)<(he1UG4wN3_bR%o##tI)Iz6JQx>f
zd6D#(wjaOiIy`(z+`>=V(g(l9Z>`C*I$mbSuXH5-)<JjjGI~m9DskZmNqHu$IMgF(
zoyBiJ@RtubTPM&&W*Ku!Yx9vH@!$w{-QMe!vP|A+J-LO;1CK+?=3|$W`lj}7naii`
z1e*Dyb>SuK<itC@h<M0R9#LF;#W~XUzR&0B&Yw5@xSt0NLz4~IX94HW^?p;@v-rR2
z5v+5XCH8*V@0S3*gx8OakaHj#`6n`3rk*G9{3d@D>8&*8gz07~z8;;PF-y@<sgaK=
zrWhNz`yz)`aRh^)SeU0m>9xK{qiaZ9!$ZNHN<yienDx3syn3Edg21S`gBFS{e=j54
z6&S*reD0^WTT=6`5qF5E2KvKHMNnMdWLHejHAqiI_-sj>25aD?vg%a^R`Cituk`){
z$9#R2X)>_eL4`ETygidUcZ}vxEHlbL!F<#80^ha=z*^tMi#Ic^qk_E(A5>C3zZ>C=
zUQo{`2z^M<;M)IhDhR!J@7cvT(b(%D{9xf~;xaULt-<UXn7Gc=IF)Je4l%6&*v$#k
z!!&d*&o(p@^s9dvg!kWm>&m@+KL(DrTYhgXMgHVxw1|`EG#0b-qsk<_y_H6wjkm@x
zJo6E<D#PGR>G_6{9wEBBjwH`u;@genr%@u4MuUvF^$d$lCp^yd#?YHG3!m=a|1uxJ
z`>N0IF#=+=MdSbWyAO)PO=}d=NNR}n(g;`i7~GeiWTK-O=!<{LNF8afodG=G&ACaT
z2Z1X2&_aIT^E?Yt1FihW84aa4MpV=neBmlDWG{Z``tFDCuYCUJ&udJ5@Zkr=@uVyh
z4=icnWIYslxD#CB(P<Z>H_plV!|#88<qv=SW7mOL3-ZTI!+iSbr|Fev9f%PLMna4M
zBQHjiOkqxW%vzK`fBt#o&^nwnfN26=rPu#FqYLt}2IAQ@GOoePg@yD4Lo|GYFZox&
zLUtT=W|}dLNj&J8@>GLbxisP<B6t=pXXG!7wg2^Z|JpS8=(7wTfBbRYjjSCzi^7MF
zx(25jRgslfI!_00<$T$}joW*lz2}0H3yx%z95)=_b9pDj;&Ak2{o&4wfrR$6{M9+S
zr1~{wmBf{I25lG3ANbr&8TQaQX>&ct8~GOO$ROD2Kr%Tsy6Vn4D7b2ns5iN)UjuLD
zP1^2&Jm{2~Ox8d2l?ULXKj}&-dULg`<l)^t@xkp!GwYL|10T40-o@bEYMgX7#b1xC
z!*SshcEXqxHM$P?K!USZ-~GhFyW4vc*XgpQyCJXfYn@B_)-*7Kt8@%Kclei*BYl%6
z4!4V6m!7?gFYVuh`*|ZI94Pbt3pmk=rKgyyLaVosN~KsMx-~=y0j0xHq%t4DnCS?W
zaT@aoM$&fYNeWtVCFy}|6XhjP6?&9cs>29Fd@G3Vac$3%(Dz0nC+)>Eh-$GXEu75_
z`*N&`Z0!T9ycIO*&DVoh;DLiz@Y?k%@*sf0Ze^>L39eHy5;yrp|0XZ<<r&(K0-=CR
zsdUIX;dmr|{S&s^jfX|ku*kbZ)=?hKPu{_0722j<#Rne6LWygvC@}|N={?;DVDa=K
z7*ooNJ5`GKpLzuHaV370PVVK21pg5i*l=acp`7wD<d3FwIO*TP12V3N$;ybb>!Eis
z5YcCM2>g}Pi)@vSP+hs2DL3))$}|9VL5_AL&0<_-_~hE0?IC}87mtj7P^6DCRZ^PU
z;AOfj!(vsGLFcB?k>n#^2lVa85SqC1HDT=q5`@y>2ir#G#L=MCn2lU`XK*)a;x%^C
zPSX4anU-^=1r`sxfZ%+B8`zwD$+t46)HO(^l^9lSXuyS69Jy=bmN4NKVQajf<)l?3
zAWvdweEn4xqGk<<lTWP&F}l$8Cq`M4hxlPH1cRGtgiMKq{xH9Eb*MhLp&jmw*#zdx
zxuzk>DBXbOQhVIdm9m3t;J3G17&9WOyzxUhwhjk(@SV{E%*lWWAJpq_vzXReFN<m)
zn6`^kemWq+?dmw`gunqa(o>9%(rB8W`WbE}6kl_}g=?L`m%iisK*MvmSIC;|D3cMo
zrWLv6<xRq-8QkF4Ize`G!JRB@a`4q(20;B1Mn2oq&oc+7)^YEF-}t$&!O}QKJY@@J
zagevb#Dzy`-1p=GpHNmO5BW<k?c!IziBdesKYGrmHICq+GhrdHdBR@$mHnnkdbr4=
z7dj3Aa4oOmw{qVgFb`LMX^%SGyw32gA(8x-H-(wB9YqNo^lp9&)4e&I<@2MkbC18_
z=Zr$VX7<v#mE;t*g?B{Q@QvSsJ-xhsxWM@2v!elGsAT=-%Mr<eMa{3Wl%9vV@3<^l
z>0NKJ3WqR7m)v=kt%rg6JkuHkuDzqgndeblP3Xt`PvTaYg<Ok6#iNKT`pKia3}PV3
z=UIkl`Cm`3%<hm6vq<qz>Fs?Lh3nuf(=gj(jV35w5%bATEy!b<vZ+9K#?D*AWu^z{
zU?U?@=xHqUhNrQdxbjZw7(chWI8uW|Nx<K<RJviURQ($8!5^?5Q<mrq3d#tNRRkxU
z=GKuCN0UFx9JrpHg{{_s7~OHKzC3(xUPEjy)#c<!^RZjz6yHhtVJ*OM3>9;^_`*oy
z->?LW(&q~TF@91fc43I8Tr~2A@}H;2jz31p+K?p=dB4aM!qd<<ln$*u%07nMesQ-7
zFmbvr;@R|+C_qM+Jj$Om>=c}#ZHsKtZ(V@0(dcV<HJU~_jPf{B<gJ7or7*AkEFXC5
zm9-J*`##?)k#DE20_)wZApzrV_^;tLwIU1*{MYDxpAiX$4P&LF9*L(Rh9mk+wO!2x
zPLOVx6Gb^R==Do$&5~&dC)+ysk1_Bfqa;+IFUB%(!DJwrf-xmy<l@_`5n^P&^DHu$
zx`bAz$fDn<m~wslL8c?)M<$I#TI=CMbbtE8r#+y}dJsEstVxgud^CJ*{35UNdzP~F
zxrE2L7zyMUpYe<AGs2C07zLpx7~zb>F(k6BI&$!=LJ;{2-TA33*llw)v>8D(3RQzG
zID>0skRityWMrl#4eaO)x+Zr=u0MMCqKjxrp*1|}zwO`75V@DnOu$VYYEPT&MQ0oW
z-7?sptsnbd`Bz3s!})C;@DlvRXS8=TA#u9kK0IqgVc|b=@nZDmA?*pXex{2n4;?b2
zLG&6UY+$#ngDZUZY!6`!HazONAIWzQTMJd?8T8aKaYtST|GdP&&j@15VeuDNywaL9
z;`ci6+Kx(jaO5k^bThR9gKLc<0o(zn{L1Skl~Ekwzx7*w@^42aclAUX{DE8X1~XS-
z;&Gfb-8Cal;CWx%n~wkCYDncBujE_aB}|NRN8DUsaPu3l<W=J=xNKM>h;ziv!+Kq~
z7l#Wzf4eW!>it6cUKh*71aI^|D@3uey7Q%&e#UpAQ2~g;ao6DVPEq9jblh&b!4`Q|
z6`G5=Iq4J@W%a_+`Q|ZJZnIIR&-2O)nA5~3EQ*LrRXCI#H5)~n(HMf)MP5-zra=gq
zIrKJANuwfYxSmIOPotc)(HbDWoVlVZ5<<%fBmZ4Tl6QDi5#~NUGxX1u-}t3&M5)Yo
zJ!Uw$&<`3K*W%Oz)>a@~$e?ExlHq7hMSw?p)~Tcnh0+m)(uVm^tZ1sSlQawpmj*=x
zda4x3|3@zHXBWdX<Oj)(2Iu1aqeSCTXvfd?wls9sTu{u)7LJpv;m!OlpW+WMfAVJ>
zpbZUpD$$IdO@oK$8pg`N!4v6Trcoc-Q>LL8uNVs&MR|jVcei4c^50!Yc+P3mil^_K
z1ny(}2+xfAF=C$G@eL)rRni#4FJWlx$Xe)B7V_2LjWL|?7!r(%u0cp1ox4|tjQ2F;
zPDw?(#v2Xtd71M|oRrF0x53B<vx+)?47!&w$c!H0?VLm#xsK8O`1X@4cLTrW9QZvr
zFF2#sIU4ck=#kFC4?jGL;rcL(m}i&GqZnVs_b3mB%X4_;-koMfErPGIm?<KL#*WmW
zDQsgf8Oe#M<FiZGUqt4NH)NR5H==wO$6;*O@>~}(7$D?<fnoHqBOu~2m-60enc(2A
zOkZVdaq$Q)n4pm`BKGL{qZnhE_WbH@r)L;&Z^z(lV;tl(`p+Xzv&KyQ9vw4?{LJrB
zSA)!jNZwp1aVOi=2S@zMbkd9rllN)hnB5z;5T&ael7(C?KXA|{?6{IeGE9ze5ikDS
zly}PMcy9xrw2b`5O`M52IFi+--@Jq;?cyQ){5>zSnS3IP@a{+Rau=@+eEhklOWvJk
z0yBQqTW~qT-K9|<;a=#8+wx7?H5ispYG`RY;K>6{gH@|S-^0YoKVh3zUYs>qXtrP%
zU&}{+as54>$K!wuXIIh`UgB)r;?T%HPG3r;`#W+!YlL9l$>sS+3_1@_#EpNJ>Xe+d
z{y2@e^77oZ9Z^c(W$u3z(+I&z2jvA`g)jnl?|ydL$S>&?rv`$O?n?4DTC*ENrE71G
zF_SozkJLRCl)*59KkpRRxdXofo?exOmmg$)Snt>F3iH<15m<rm{0N50?erk8S(&e5
zQkayb3^fE-GkVa<GP3y6;>FOkU+!(Z(i3Ho(_k$aeUhCw*RRc4D(JR7`JH$2K{ZMf
zT^i)v4gbIDbXF|vREqX?65j{{p+~ERK+juc)xc3`?Twq)7bjO6r@%8B!twL5FVivR
zj`hr~NJpzOm3Oa%k8&G@tQEO^H|J?&zQ2vQ^CjRYz5+X3UgNiK;mgUN=aj95ieLQY
zt6t94Fk)a=lm;ziL=l9xJopBW#)iMkDY5Xx3m@d`V%=mmau(&(j5R2uyfoZbZ(Z#+
z)fYJ{!!g~ALcGQ+dDQR-Of<QSlGx$m^jo6^&!45t6W5|fjhk=YAdqp9!cG}cilZp4
zHF^KN_sdT-SZiW?^TRaYRg|H@AzI2!L+~V{ZH{0^3qIbCA<%83_floUE;N_BiN}b-
zBkzn7cq_}8G*@oiW!TJaCGx$EL>FeJ?%9#?dDdm%v-J>nujhF4I}e%$PvF8JewOtU
zUu8@2H`zt>FaP{69clUWvrpS-D@Mj&jn?F)9PI9KC3*3oyPu{0UA_9W^4ag&1n!M|
zh*H_VcpiFmBvMDNhi(_nq4UL!+`|L%0-u$uF=~9-I`R<XV~pi7I;1nA;ck=+Kgf)-
zh3kw+4nG)S`0(dH|G9M=j~H3%{JZbxoE%60-=5y=lkDCR&Rl}ep%q)+c^#+h4}8~9
zEhB`+pPxF)0CLfP_|K4I<b~{BMs2|9;coFVb(Q}ygp+>tCVdJQjbwFe-i_QC1!Q!S
zMZ8fy>aBF(a@9k}vv)d+JFSE0+Yv8ttckhclD4`a`BZ-%bd~<2tk%GxB{_%Zl|*uy
zcl?l+pSrMm39riI;N20?GrYyc!y49b2n>04YA*M>-0?#^qr2_o2#9~_k3T0|{$!*^
zO#GVf;opvM!q6$s&4Z^#vg+jL;=pn9oVW|yt31VWp|feg;*xG|bMMRF$&Z?Kk<^13
z7d?-+OJQ$h*QGcMr@j6tmR8`@q$gP*XjuJ8l$+t>3PmD!7&0(?s@yZBOo?jvx1ivx
zA*O^h0=4?{tda`z4xUQvox+Gj<Z&urg^gc&e3ghiS2+4Z(M^M((V<Wc^}(B<!=bw%
z(kN7^rT2MK?h!5VoG$_u0yywhrcqZ35>TbOoOd7iX#|fgX42~2qTLCuAtSKh&j1J-
zdF9EwcpcCDRW^KU#gd+1a0e4S!|YECwOc4Atibb-a7x~_4eySGG_NCGQIQ|s=Z-SU
zo}zOfZXPP9k(*62YHsu5xkTir&*K5!lczC0(kNKhPy<T?AD+R#wB3NG#&(P;<=LsH
z05p<D@l-)a-m0XOTa3pfFO5|+*}cH&<yM(Q7BmP?Bah;}O2fA3+Ft5n=-jlG+$cjL
zY5d!eq(DHcO5&eq{O<AOc~-WrXEA7RMCKU`7R6dSVXES?Yrco)sgVdyY-Yv}=i3<s
zAyS#VBKxUB%E9^;JZzaJit><m%SD>}ixbTI4JM0necQ$K$2-s7%`_+DIlR`WCoeK&
z{f9-IUw{2=zlQteS2;2G*qV)wfE>f8JWJXGM>sIfIB|khWi>^SG76RXziB~KLa%4k
z=0z%pc#SK5oM+`ebvF2@SaEpNbv)rfN2V&b!be|JZ`3%6@fx{gm^rN&#1}WW!Hgl9
zx?Q<Sh))EVB4dmg32~w-L+ssmGosWJY-31ui~cl@Nm2x16w>PEdFmw@8EhxQK9?`?
zQg2<#hs^H#mK`XJCY>#A41~Z8t%S6Vzp^IZaq{!LaNHq=!(HDM9zc_S^N{9vZrZ%p
zfq(80my0ZP-Ip-=9+wfN>Y7cLc)}frAiw5y_T2h$2ETBRH04p^O3$6b^4#gYgC^l(
z;6Z6XhxkXj`U}I?z~1Snf%+P!Bx$&OHa#3RU(bv?w4jsgcWV;=5|<{82xvLN7tZD@
zxU|9Zz;(k)dn5dP{oEjY=WYw}+^^4x*O-^$3&UtkzX<8@KLC+WTl7=_O3m34=E5yz
zqpUR|l%8`Y+Uv|IlX@)gruV4fynEMNUj#*^3=ECS_8L@D3&@h=<fc>!jT_%oJ}<9&
z?!g=p4d1jYBL|T!Ff}R^62+s(`^iVYEd+Dk@YQ?jX^O#TrYbaq*-?|mKrhuug}HGm
z-mVP0U!J|372yymE}^q{B_DaEJt%Hl30k4z#o0PashSqjNZh`2e-x%hFZeo>fT5)^
z1BT)Sli`3L)_Qn$PDslrar7GR1YQyrkL_W=Nu@#;SdM;othwn2y$HvEtJvU+*8{i=
z??;KF;jMew(bAEjgoaN?{Yp%Z!oZb8+<;_QJ_z=3IC;miOti$;Sf;!?4=pX)GR1NK
z{_J`{(>CyrpXB_QTt8&=gtF?ohJ=$PIFC|_0?AsVS4N)F%hjmcb)wO{7bCYLAc0?p
zR>~)3*quKq-)Wewi^vX}ggZ62J<k{taJlpLd=O2#%2V(_Zw<fj+<6oXOXR}a2iEXk
z1EJUYFz<TpMon5@!s+dK-XH^LNkDlM*kG6nDQ(G07<$zQhCQ5I+gPB>NL~t}_!Un0
z`<I*`N`~Hf`<*MdVqCv^Z}|W4@xvH_FZvyrZ?Y5Spa1n=dgAII{{7#>gRG}W{e`Fb
zclnYP1}@s|98-V08!34(;^5~{HTC%AjTm1o?2!V-K-XghZ&O7^2EcimXLw3m10>O<
zK-crgrEg8;g-`1UW6LOBbx7_V{R>X@_eJI4YQbqo2s3hp|F#nw;ZeuwBd5wb56E<t
zvxDwrl<$KNoCy;;Lste=vRHMdGVyfCfP>=h+uV<5_`$b$7T1YKcC6zv5^0pRhI9D9
za6>y|RT<NHoRKYt*On(Eee(kwjbuHEPnUd^&vp<xIwB9aEq;NGzuZvK$NqpXE_|w6
z0uO~l<;0WrLV9HdAHndECh5sb4Sjxba)dKE;iQolSWB1k5=VL+lI8=~#1-Feysi<F
zK56$W(?)Z_+L53mgA;wy{>UYMc_M7DM6%;X3G`!;$?rzWMEHs3JcH*QkFJLvhGKUo
zN)PUTyPxF1e`+M;C!r=l5uX?HK39NKD2A#Dc89-nx2K=mUYJE@6iFAyr2)3<faFW#
zP%!DPXZk~-YfN%GGR=eDQvWAx5E((4O!Dk6g3{U)SQQfrnh-_sHQtntQ4mul6a=@z
zMj5C^T_l*tm&dApf@n>M(F6-LdlWqA!O5Qn+AaTe9lh~{j)ZORsf~6F3`6n`;$exW
zvp5yh=nMEc{yiBSbVw0AT=*M4Q8*lBw-qE0Rlw+LgOzwH3A)My;Yl<2Pfr+bt(=Ls
zu6fAk@auUEDJO7(c457aEfvo*ACJofaL`}il-@*#GdjV<Z)M%0jo&)?lL^$%Xj{sX
zY~@qyGdfb%2DUU}4UZ<1(9Am-Ym}~L%|M!%#>-E6Mjkur5<^A2&1XiW0u4<N&*jDq
zzHpuk9h7Me*YKz~gh)zNIdP=}AN+xtG>j7;5)@xP;+9s=$K~B+tr0vaZ0}^L`3Eb1
zCzr?Ok^GagXKXMio?OqAga$E20K-FCaLBzytfp2N*pD0*SmSF(<x+7Qw<BnH7`&xJ
ztHS`hM#Tun!3%c;wJ|Mac?m5_u)Y4U1HC)S2Jd1o9T?vGnCIlDUX0Tqp?~;X<C)=>
zICRL~E%LTFyS$2yVc??Qmg0v8?llf#lriERC}&#mK@Ci~lgA^=Q_h34S3!%B_sh>s
zdf=>qTqD}Obt)-Jzr;ssJ}58ml6hsOJZq%>2vY{tX$OLvw>%eRd1$P`*zng-LY=h2
z<5?MV7a{S<<YiY`jpyE5&UtTs@CiIur5tR22av_J??=4qxx^_vxE`>?-D$?^fz3m@
zJgZ~e#y5I_&eD@O8{SAi_jBM;I0x+IRo8|DrgB5my3M1bNoW2IPkj8maBaf>zLx{P
zd`-w1!_EtU_fNf_kFvaoQlu!aImAbrL%H{))HE>WZyo;hPELlTK%BW?1mHn>j}(aY
z6a=&Lz?q)78b#?G=Qds%C<-W7d%!v1L+|{R4~^yCM$toZhATO7Z?7-jj~|PdG)ZJ=
z_d$AuZ@-m8yfX6eJiSxr9_X1(T0Lhyk^o7XJ<T(l(yM%LYc%5DEtU)qmP2|nav%3v
zc>4~Y4Qd1EeS8T2lSkiI0WgeS-AiK@K<!nA=C|(8Ax1sEJ#qEwC>eK3&DsR#hqQr6
zdh0vduqRK`Od3Wz0!mwWtnt8e28tf|_E5@quoE5trA6c9Q})jBfcCvx_c~>UFARm9
z3WHCTE90x~H762u@~H{LBZXzh362?ExqUMyK4#~KUWJ_>Vf;+fgg%W0JX~;89vv=b
zuqHdI)FL+xG9K`|VJe0HZFpv+Bn=VXGRWFX4v)Z@`S_rhXRtMso$5(maO;9miY_pG
zLx?;Zx#CI|T)+=MeS62$-h0_CqfEa`<ECu%f(t)swlPhz=7;x;D7|5gvo#BkAFDs3
zY?TGPx}`Pspwk{$oU3P(T+e8czu@&EQ<o0(lGY_HLqht>@pryD{3IhHUlfiLV)5d6
zMgug`fB56?b7svqS3dotdf<GO*|`SCuEj_mV_;230o)hiBOYmd+Jg-(>xyRTD%ulR
z`i$N%eyrg^%ZvcbG(yWXyqtAnuc4F{hDcyB=I`YgdN?`#)M$rYNOXs_QRsQ+?RUEq
z=t0hLku~{6H{9XsG)<$kcKKLz`-eaMv2{V*$4~xeM-(4KFQ^}ti3F5?i;{c%61_ZL
zZ}fIZ9}V~~POfKJ7xb7>m;83*rv-><uA|OecYtGnC`*nkh}VWW@JwB!(|Rs44X4YO
z{91og_T0)3e{OWFOp-S5e3wQt+p<jDHDHWr)No4J!rDA*&<9Q%XTbAZz9fzNy5f^}
z>sO>Tl1rbrT;pGybC(zV)FF4_g&9B>)_`~y7Pq-{Er|Cc-{L3&oDR2PUWJ2SgXf6D
zI4_Ox&qbD`t}!x+2LJG4(<$#D`#K-=h<mt`;gWYe{d;ylYlQqq7_&vY_l2R6ncmF^
zP<yr+n%(binC5U9uI`ja3a|oo2;9ny`#vx1(_N+s96{2a#dt}h)_hgkQ`)I$t<))h
z$<%l0>?$zoVh*?!Is(G*0i$pa7~-Fcj<Cr$lqPWV>WGOnMbd9lHQ{=Cnb1b@-2ePU
zUv#mg;n)bc9{#IqF>-R|g{gtmmZt2`V=|-^#2Yn}w835FX=ULL#v$-3T1P^XhLS1%
zL7<iBsOXejD+oBbnl|~=;x9Z(<=|V(YZ?~uP*pOAsN?OT&nYM|quxr3r&)k&SBeHR
z3N~d3FZ2eH(z6E0D#FPu@sz^>xHy!y7*6y-{K1jOO}+ELqfmAPp|r~*2`#MkP#B($
z2lo>%8XA}iJFfoXE<fXM@pF6SktsZ96fv4kL+J;;<#pjC&t%e(42@1Ms3=4J%5dZ?
zc*a#B!zspYpOd8FhIj_8+&ttrxa@k3F|tNdH2%-?HCPR@brD7d@I^V8^0W>@9NbX$
z3?C3pL2CSKUljLdlb-{J(5nu=BCB-=CU6aY;B~Xh3amKzku+$*>&m6~gEw0y0_82<
zG)zW37|yesjS<q*W>em4aCeb)46M7^dMkaWa8p+q+@@eHI8NbBqu2b?K&Gr1>W?2k
zO*!T$^}A^>$s*)_%5B)*;hyK#o3k{^Q@Lv3TjmKXnMb*aievN?K5PS<*~B<CM))+m
zmEFV<??;YbE_rZbc<wv7@*aBQ(T3Bz^$!g{;*sr+1SG8S#65EtU?;8oPSYfk?6v8@
zjh{HyZ7A~;W+7I7g^6=qXdn4be6i&526quFXdL_qoIV}zC;!0kET3~fp!dD$5?*)2
z^;SNPZ-13#619%%R5(AR`coP)7RFhc1G_VyzrV+x6SB-+@-M$*EM{Hy|H=2nzw9Vp
zm03^4W#`4qm#O&o(z{Iw`Mw2%gtbf9^!7CldR`hM?<v+)T8&hmvK_Mf-!-Bd>A0>d
zX(OZe8zUzEba5A)M*ys=m@nl{rEn~|C+HG#!?xBz^jk4eOi%K?8pG3uoORr&UJ9WC
zYlUf)DLgn8hLGLLw_ymJMp9lkGc7Rri1qrpLRlaFhbxFP-eo$cRi}#aX?9Ws!sBb<
z-F#^4PDUTftMC#|lrcWng{Us>&0Rwcsu2=p=Gt;#i0DD-0czkVa(v*gm+>rcPp1(E
z2CLPP=hpQY##ROzEc52KvO7jeENywb;NW=fF3XQFMwjo~m;$&KdAON!InAvmFZ6ZY
zhlVr8AsOjfjnLl4F{4>%bq+=Ii~q}vj77R?fZPo4YN$vT`tVHS%;-!Zrgv&bfi($P
zIGy`xQDC5gXH??+Ah6OC9F=YT@hdo@jf^v>A7poh-nJgLwHKddSBVDD`kcY#FoqHb
zi=?m8jSN%J6t1*J6!gO3Y~iG`S7r<fYZ$ncZecp}gTEaY<WHL2B~oJ}1_yc*Jud4o
zVsL!^d3NhO%tBN3EuDe8Sr=5f3SQ2P@ln6$k$1aE_*1%1@*SV2SvSGKg%bv>vU85l
z)r`W-qP{dL@xz<DsTZ0d)oaqFC!8`+U+`dIbQGjebwaI?U=YpX<7?Muy2bh}Yf#>Q
z|AUNx+|5GLjHX70@gLoc4>J08MnaqxY;6uXv*q57z1ID#4|#C^?Uc)m^f0ijKe9`%
zYuB<fh`fu-xKW-L4kZjtTqQiuXz2UluTc^MT^t^eFJ<vOa9c+!%SJsC(|<&=x-VtH
z;PL@X>7NH@5#WNZTfv7c7)bzU4Rd#hjC3~JQ$ru7*JJBzPa2KSu_CI*vT`}W3(vdu
zC;99`&s4I)fUo?Im$~PcF^<|c&cUPQ?4J1iYv?4-lU&Lvgx479$W(AgtNgq3E4n~F
z^vx^=AK0hNE<M1fcm;0xE<EXXmz8Hu9$O~vON+bfLY(8ho=4g~qcdSHG}8SokKl?f
z<y1(UKKF&acq~li#Vt;!c&7+|dk=qgBxH)=mlH1ju<=qF%sm*4u_{;BOr&y2Ksjjm
zoJmm2KgFgX^uVetqs$7J;?yW7j2OBu=SLCy6V)}25}=^d!!lQsY=^e|rv400vreL0
zFr$R39HJ-~K+kh{K`;TKc#tPmNj*o`V3HhSIH-=c1Ml!L*vJfFF@zF_5c&hd4;SIh
zf85aBr^N3_3JQbslA?#2svV^h#h{@&__~VdRTQwVM`{$;RoTcR2v<2JfYK`N0sJ%{
z$AihPhuM)9ibTb7=elczB0BsB<rzw4w7{k4K=$;|q(Aeslu2S8%rV81vY$#FWfy;C
zj&}It#u}X}I(#cnUE~qky7e<kHD!lRQ)}f<_|IXXTp1O<-{UNb(we*(9=j-z0v$s^
zVI@H?&#%i#gQgg5v4x|Bl_JVd4^@K>*J<>g#F!j|f+1KXl{mvLc!ajV>$g1ug&X;a
zJSP)yN_Pxu@UvkmqLDJf5yBqa84-ze;w2u+VB{m~oUYC`P-TZ4_c16eG`*UgFTwA&
zgU=XjvsIjdhYyUvN4K+cDKr`}u?^l(O?h<i1TB?u<<gUV)Q(`7fEjuYuBJ5er(@vk
z52(9OJ!zFJ4B>6QlX$kU4)5WNch;TU%Lnz0cGynOK=4sWhBW-eIX6a>g$s@jEi@h1
zwNn9WrxZh7+0Uq$!|&#>KKcr?G%$?B)*Z}ylSAk_yeAy4Gv`ftk+oC(O&mt`#>G1Y
zSNX0iDO2S*aaMNlp3V`dek&t&ge=OZBTM4sDi*ohEs{u5HC^xVZ#{SCeRkou%SfI$
z<OTfV4k_0e{6wuz>-`8Pe|2~gN1mlCPI)gK=-hbmTU{kgV_y9tZPLMy{{<(nGq<ql
zLNRT(g=eQf!4Ur}4MU=`o4l5PXW)jP=N@2s2Mb*X?9nB!KNz@Jx}QIbUw4kf5$7T$
z6YX#D_vej}Wzc_;@;0;zHwC*DXVgrF#Tgw5GaSmQ_eD~rq%yG;q7OQ(bteT(VA>On
z(p7=?if6+rpcMx18n7sH6+H@5n7kt`siad0maHnnRBXlvC3gS##JRb9TPWYn`V<X^
z1*-C&hKo|l>3Prfn1fILh?AjrMwFG{Mrdx-7}1NFwFh^~2R%LtK3s@$BWz^Q2+vGK
zz(0*2ennZ2Vpch(61OK1{un05yN+5U?(N{-9$;`Vx2_irw^rn2;5o!MKXi7-MSFZ2
zl<cxdxQ5YCwzV}XxL#*3Y2$9jPX#uF++k#jOs`Nw8yZ^?6E=V73SkKR87B(2<x7cb
z42+yn_(nwvXQGAXV;g1~jp?x(K}e3|Ab-l!cS?+`nZxg>Ps&y{PXY`4Z!+h<y()64
z+yXc}3eO@h=H1J?&^>vCACVuIbvbwiNBKU>=l})z;N1^ekc=q)?WG3pz2t2)>C3Oa
z?H7-~&NpX_Xnc~91qQA|)_%yzslK#q$CI5cgh?FRf*E&RW0bs&W{_iRsFeMS9JECq
z9HHI@(#RoM?kI+xXnM`#Do-hIc_<X}>he7nXWV?3g{RgRy>;)KjvhEI*p8f!y1Ob$
zIcc0(;d27MWc<Y!p9jZlSDvQMqL<Q!%Qk?v&)Z2tYI{Ozc&1KI!!ZUnV}ZM#!^%>C
z$s_e1z0nKm#;ttBO*v?+XIdnqlQ~OAGl@U=@LsmkS`YHcZ+=sQqa#oW>&dn$m-{(m
z2+tUzj45)__l%mr^=`fn+;e)45gui0-Pf!$8$<GT%87w4zL625SV6G3|J)D0aDF|b
zZ7;G)R?sI5PGqJXw8I)x=a&w0DBiOhbJhypjgFv;Vju>NH9SUF#6t&|;*eX<@-99J
zGhLv}`hMJ7%k!)rU%nV?kB%^~mnQJRRjzx74-<#cIa+Is#>!j#;YY7DcX0w^^by%U
zzS$FAH(u__nr@3gpYVcv^Gw=x@REnJ0#i9DvquLk-;=3PmN2l_d0yUt5$3hSg^p9a
zJbMq|P1d_nBypN9KJst8-ll^lr@~MA-VYcH=}l90!2CV9oQhtSl`-=)WS>7TBeuv+
zVP}O{rsutJ0`oksXCWHzDeO_=p65LR8zF&<;ZXMjt(k{Rfl;vH8a8&eK^q+T5AYQH
z@s~<V$gW1IUhS6KM9F*Frz*%K3^t)ax_N3tPjQ1O@0;0POWF2B%T&BkqY;;cT{{+u
z88Xxx>To^W4uHOg!!(|EivFzP2VaH;I<95Wtl`aBOvX6##mi{`H5{&~V2kB}tXKH^
zMSmPc_A(;?*MoDoTqP!kN=uPc$rbnb=B~0(01TD#Itp)l<$AVg_pTCau!~RlATFBu
zQAkFLD7`9((BcHy5O?qv|ARyf1_o7)6XVOg`WrSuKbP|2#s~|K5>Vwh01`4OR#sZR
zNdv!*fHdDkD=e||M~Md)2a3g%4Lq`!1%@#u$&w-MlRPHCacyz|&%<erkUWBizvX3n
zugMHQICvq&R6!r|%qX$A*;Z%#$7|bZpM<w<Y{-suT|Y!2PMs(Xc?Pd=6EH>><2G_k
zIp)tOg#5{DBTtGU0$Knz?Z7EN<Ap6Vw4Z~kB5!ywbuY?0qkZsxnh!o+f1bR^6C*hX
z@5Ot@G*VBL^UEmr9&4Tk{dvBo+y*Q86+iUGcyEK4Q3~>8he^)`ih%`}ohmHTgZuSb
zCowu=$PN*?#XWs8df{q}g6r3fmL6sL>S+w+Oc`YwceazymU0FN1FVLlbzMQ+6e*rF
z_{Ol91{od7)^_^T>=-^8`RJG6X9Nh}rN@)q9xV-6iBx@(s3oCgF)07l-zf{c8ePEf
zZJi3A@JgAm))*k*Dld2CgwN6^+scQ`wgEdCfQh$Gy2VYpk=Ue31pa6%K1YI;xpq7P
zhru=SJM~{Uo^gH)zb|>?FTcPPR=oJZH`g}O^5$q_uDB$;MoH)cU;NUWd*wbnu|BCV
z$jaf$32Bn2ym!QE!|^T{S7xWK$Tg{w<{+R^(XlR=#-K|*G0GVKZkY}q(FBcx&)k$k
z9waTn^-Su)4;@p#W@|ASD~|#z!qaDvn*5air0qiUgporIKhB-E`tU!cJ&*C@G$#s=
z_?O?$Q=ed;d;Xf|#px2>mqEZGhWF18hMXr_AzGq~Xd<)#lUd)B`Gv0{0w9dVU#iV%
z&uV-^w}SX0y-9*Z*%LZF4|f&5ir|MSjIIeuI;V>oHJBsP#bNoT0q$;;WEACTE3d-4
z9)A_>`=>ud(2v#4!IM~NtSnShVZeMDB(1LhK4%4d^~Ki_#!SaV;nE8h2zIn2jl0nY
zjqlBy)2OuqB@eyhS;z|C8OoN&)29?j=ngzPW`g>a>-TR)*mT+e06+jqL_t(${mCaC
zI(|FnqG;@;|Kf|k^fb+_7-$aYj6fW{#N@k`1AR`P#Qf*kRZ?T-_Izu^Nsm~LHWgX}
z48F^DONN9|pL>}jf9JjTlvMqV`sf84;&yg|9;kHyt?2K)TgBAviy2M8Co=R_@Hh8P
zVWIo#RYpvDb}5IaX)rWs6rM6oU}a)<LZr&xj$iQiRpTcy8kECR=XmH1r;)yX>zI~F
zxd&*Bw=@JsK{A4%SKE=GG$3tgD$iHDz0$gshYvrm3{XI1z#(XO*0U1=i!qs?j8r_V
zfpsruN#OOaH9_+lD^5A>;aq6Q<-8T)owE)Qr);8wT;2@t{;&V}e+SR&-3?=%%V(c`
z-g0~Q-S?|`z%$aNVck(M@H$;!R7`o_Q6KCKcq_ckXi83#C2L=P`Biaz-;SI7If++I
z4!5efu4NreP`Pp|X_fc$$d(0C4$`}kDXIH!-OIGg4=ESxO)?4=`S6V(^E(e8ejgs>
z)X<0DCC_JR<TL7$dPwFN_SSCPxM2!p=x4aC@u7Utpq_PFExb@4d~c@fVLtuSu{O-~
z>Z>acBUhtWW@N)oB5A*iY*{z+b#~?0<zkAZ3w)Dr`YD0=G<fB^PWi^mZ$AEqp6m0Q
zkAK^d5Z}s~hC9U+C4MXG&dA2SjC`nz8V5LDj}dbte7pJ1oUJl6sB2`nMO1keR~cvz
zrXg~!%NXL$rN9?oV9rDJR~rl!4~+N#Q$Vc0)nC|gDL+Ps=_UFFo!tU`?3#K`S40nW
zsw!~F2=`seLpeRoB3bkpv6NR0llVI^*(gXm2Ej|cguA=?-)W~p3LQsz-by%EgCVe_
zSBHIBn?5o+*N#a1!FiB#oODnYHav<$(jwWlu=j8;_<@Pe@<xW29F80HY}w!wp5t8{
zy+Kzz&bc-6qf<Jh7752cx%%c+>ul3ZXO*ndwLVb#;700C<nVDu@aWU0DL>~pS=WO{
zm0|4(TvI2WCeGv@-V_Vv7r(?e@`(-~ScK!<yUzYe?4ZS8`GR>fqjc)g<0p@zXS5g5
zQR-k|yojDtLn~K-zdQ@j>M6!(5+@+KzWM+3&+_SVjLI_@hbvF>yo+V4i}Maw?i^0(
zh|K&)-9QWpV4vY|4i8;$_~jOe1m_SgKOX|a8`-Y&_!U0L>~+M!bgqghVd7y%zxQy&
z>F|FTBFvEE1?DEM_^t|5h^l2@xPw8%1FIHx!Z&`Aly{U4A<&3+{YM%vi_i$Iyqu6n
zuqXgS_%Bi!4fh35nFzm*wj{0!nx7t+l;AGbg!i?praMA^%0xxFo!*y*<>QZXRCb2k
z-+Awy4hgrXo=QL&^rEB>j%r{C@bW0A$KmwBbNJ=;Mb?U_6whx?rQXGTDsw+a9mLRO
zir`sjq@22FH~x4CzDB#sE%8oelV{4rNCJMery_Id!Vf%r?T*riUan$^A;;xi$=Cmo
zl;)|?l4o#014G7eG<WmX^pd)*mJy;Hl4p&ZW;Y5TNvDwrj#U_FDNTl}W5^`e3jUE+
zUK&0ftd}``G_fcCS&yqI%0ryUOY~bBiisC&o90^H$rj_yB~Oif8?5lT^c+eLJ>XC_
z8PrJJ$WG|&bWjv4obca>LLOT_%7ILF2tQ>2zA~Ud4b6++5yQY<#e&~0$M6bn<8U5t
zbH%Im;s|k*U(ji}SN@X+`sI_L@H|4CcLtV5US7pF2{i=5OGnb#28*}%?mm*F1|y>g
zYvseCgTWZ~wGIp&XbEQn!$Dq;QdURwp*eY~Yj35_I-IOmj12YaMowX!sY^;eQ@vTx
zNj9fZ9~$ri?fA1hk>WEYxfwN>0m2XvHF0y5KP~aVSK+`ZFvJYZdElS)ymGDE_xd$C
z`0`=sSzwyX0%*T|lPsGjUR%$myu>#u*>Vm-cv#&ilLJrjIAKHY06LK;+$H@HuL&km
zADdU=R8~Tx{Mwiu_f1~}6Hi_doYGwY@^<A#Si>9ehM$CwWWElQj+2fo%0G|gzTdeT
zA^xd{hA{g&kXk>&H#kphI0|#&^J_QkNcud~;5dRV<1CAT<u}f45v|ATA9db;Pclx#
zIpLd;G*dV_4_v3YgQLhmt${GGe*(CKzt9!tFO2(N77at!c~)>|Zr^^D5mJFW19JKC
zEX&J5<z<yUMM_8ygopy6Wd#?0l#U7$X{*558KFRa5HJ1|l(cL#@Z{MFM#&_MpjKED
zSDY1mcZNgbf_bC^%Xtlkaewpi$K9?s$5Ln1B(U@X7y~<k0AJ?`+)p|cf-)dFj1v|1
z?TjEWW`6t0Cso?>aj$1rKL6sg7&dzNzS4Q?%10l(f92y(KECq*{_p?YUfyqh^GVKP
z__{s6fBDyciScs#${+vmho0aV@WD4qDe3x&`)pe!)T?M?b|gj=c^W*{zlPMvkm((b
zv<9)tM57&=`sK{v-zfq-4Ly3n_yac$iQfE$o=f6!dPn$S&C3%sA3UHC@#3;0Xe(^3
z%*nTq+`-J%RPae$lspY=Q$1}sVr(cI!|T=!X%J>^Tu##nx0e@Ol+nYq=nSMs!IK<_
z6VRoRyg0$L9tjcN9$@azlBdtdcWp?*X7ORF2V2u$aZP~6rUV7w^v1Kgu8l0(ZrqN+
zlwSD9A6e^htJ8UK`|kU1%Qx#(5MNj@OPZb>9DX_}*!dMkM;>Ri=1~@h*4T(aGTX*Y
zQv|QjsDWqP_&}ZVP<BQH(5E-7;okbBOm5zKe#QAUaB+cwc1E1CAmc5^qfzrLVEJl_
z^yRC}$!G5b+`B$SW0?`DrKQN_$LOFQ9~_!rjHt=i6yME`@U>@s;Gitv;-pf}=#2bb
zt#hHzNZeNm-@M=rhX*-R=;Kd*d*#CqKJ1rrA2?emGQx0GTmhYXMy&1_83`Y<4r~48
z<yl#dL`Em1?r0SBaMukhywA0b%7(x3o{S4ysy-)G04@HWtBU}SQ+G_A8f9%mn!0^6
z4JBh+<KL0cmRI-^G_EGD(LeQ?qmLLM9qr0{ckU?5Tx1Nt@r?dZ7W|e%&!;OeD%ycT
z?<UceSHbB;J|@V314Ewf^vb%`#fexNTYiCiV!}ZuBkFYyn|kiWCrmr>DBg?Xc}#|T
zR8zt*B8G7OJkZtX7)=<jiFD;rj6_E9jg)=syOEQ`EupLl1r?@Nm*#Gfy7z0xC<nc3
zJyUJXjh_7QX_BWycW`Ztpygn~lMj8p9Xvp`6aCCr;`6SI$b%Ni)Mq~<k!iqN3Oerj
z?8ol&n|{#zSKSKZUyb<l5w>j3^L&%vIh3Jd*nRGQdJg`lH6bU@pLsaX{AZ$F2wEh0
zO7Bt;xDCTbL-CX-L4mmDgD}rGWLUq$>`jy47QJ^BqOg-zde`Q$@e-kpr1-m*0?1ww
z1`XQ>AAVRkUuEH-p<|V|>jx<Q;A|vAugs-Uhu`95{fOreDaQ;Q|HJ?IA62>pQ}6Y|
zFFxq#3dKqRnSL>p{!SDpqooy>Fh=;`qA^t2DO7lX{XC;0nb1kYGsoO(kn4gMr>9Dw
zosZHU<S|tuEm*c(TG&W|4LyvOjF^PBZo>@>!}lt6zs;k<4(Q@15A*ugSr`UwCEI!P
zltC*L{^sKL-k0ETfT|Mcv$(-mJn!(7^6dxn0X1Yb-r;ZfntPUypozO>y>~Qq-Z#pw
zQ)tq`CECKY0{Nl{Wx`0nbGv7xZx0^-!$WyyqHPp!8dM5OI<Trp6Ti5Eo!E?lfJQ)J
zew5KI@H+w$gRrA&)}~DK8V+EsA}{_?nw`QRbK$YYsov3~oN*7nZ2)XPPdUM*WlqV)
z_@SVMt6dBj4(!P*hQWciJGcoL1sJ#-xxhCt3%8tT2?TY$Z$0nak983ih2q2VLO%G3
zcJM2=x2K#Kg%~jFGO|;Mp_#hyeT*jY7`EyK{NZ7%zJ^7k=c~4+D_S2DR~%Cj>dwdc
z=-&J9zn6GfP@ZW;@ZrI!ubx}h;nj^Saz)21{#IO}O{0?`a#Oe5p7GO{{@{@ZFh_RE
zmb@#IrqA7)ET9aLxskcz(LBI7<R|y$n<z%P=o$6dh?vo!Zhb#=p{Z4HcG`0Aj)4;T
z=|zTE8(#)@_)Wf-f2&WF!<IXI_-~mNO{lsK+2B`71HT3=n7J59-ghx9IN&bqrOA66
z)g!Mhi<U@!dY8!?oUmR;`CW)2E&m|PD5Lt6C3%!c{uas)^XEOD;Y;dv)(5tY$WU9*
zOOzz&H93IOKk!qCDXETZ^Ar0;rYNtGr30a%i7rsLrtTloaAXgU;f-n0!8>r0xq7K{
zsE)V3M!%?6&%<w@Q5x-=^vNv0<+=N1*D{Rf;<Ws~5HIv>IR8GsiMJOL&5O8Oc6{Ch
zUwCoD_g~`@zbOv<e%(mOrHsyzJP-OQe}tR_7Xk+o5Tam&J<S~h__e|uLalcjj~0q_
zglZY(eJh=N^XHeV+rS`~XBCuSvPR)D2bm2+^x%B3ic<L9?|xUM*b1CVJUvIk#wsR6
zv$${@5&qK27aR)oH=ksszUiHO^M)a!w+4=LM-nXpcm2v+F;*BLl<q3n+1|-0N<-s{
z!Z)Q*rG5zRwa0jeCOy6={$u!=vaQ%<n=plCt0kI;?g;c-rZ#4t7%uXJy9<BaO?}{q
z%6nRXDXZ``5eAP`q=tu&w#9Z>_{<0yc1J)HpW8f+mzJ#%Bp{^al}F{z>EXhod=DOa
z;taVbIm3gp+-b6NxcgjO0`pcDG2#)!NKVi5mDVb0^Xu`aOv~SdGhA+5k71KP3V~x-
zbiGM?qiK}e0BfA1Ff<Z}3?b%K#Y*at%MUWjr7S-G^B2VzpN(E98|l!ak$>yfTa{b+
zDPQ!St~5v%;;me_9J44|?->5*#W(mGDNznaWx#!v1`%E<K@k&g>V)Euze|2eQTVM(
zfrU1g#zx(l)>o`NdS#p>tcD7?iY&ncJpSOI8*jmPwPTUHX%oi3U1IPty1|$?0z&qT
z=-dtr4XirxUcMOxZr7<Kf8kq4<i$8)?0*<uztgYW=8T{0a=|A!E&X_|esu&bWv`SP
zTwaXffi~{P1_-(ZlF@?)M_h3I$%kv6eVGV72B7ICAvo|}Sr@-kanT#jD(OPrjDRsn
z_Jf}I3|X{qxk+ESOc<jrKpN*jqx9oHC+?;j{OSqoUec_;`{{~587R&NEMa7$ItDY&
z((ZH9=f$XS*8=3Z^d`Q0pD(48_LsPFJ22#dCq@klvv|@=k?Vs7@PvulZ$)J!$43hX
zze>O`V91#471Ng2C4_v@*>}YPboiD#H+AGS4_o|YYWxz;$dDVyAf^M9^K0EZc(Ob?
zijw-<P87qcIZgc>2nACbO4#BOZ*;whx9c}4U;qA;|LYjP{%K*he10bRhM&ie6#wVl
z0z8M*Vmb<o|7NaIQ~?S*7%4pg+EEAbDKwR~SJJQ)t-a{*rZ5r~w|TptUprDEj=VSj
zrU^XG{2XUdr6GSHDY`DiJd_@M!S(SyZdWa%fpU<huoFiEct3C$7nG?6L2s6T^%(S2
z=q!P)ps|S2X2IepCyRB{`wiUNncM#0gAXde6pGKivz&dmL&HQ#&5-*Htxhk<7DhXZ
zMsPJCYk<HH-bNBAWd3_^Si*M*m*SfqsRhI{>`M8h3WfUNK^v*iW;nj-a&I{$bN6!x
ze))6mc{~>ygb#;y71R`ko-boVTK+2M(s>SV;k&ozt2_epY7dE9*a;I}T$FkIiI2yz
zQQ|Xxj4O`v@SLVJ3U6q@!^vkF^0x6}1}?hq=bQ>VH5iV-jX^Qf5fsw7BU8#j`7RCU
zMk^WAXm&V01}4}(Z=ms6gMnOf!i-v3$jRxA<3kl-@B&LYjx5Zy1Q`4nbc+|>-V83}
zLYcaYH#p7#gAN4aCk+07&7mvEs2~sxP16fUqw>Wci%!8p2g8Eg%hS6uS4^Ad^4TI)
z9R|j#bycP@G>+n(@Y16-8lfPH>oj03*TCg?Lv8Mf*?7VxeBzA1um;Cd8)1mQutkdN
zNJ8%Q12>U!i-9=zNk4R=-^h<0Ll%fzENL;RDYWhYiQEmIc<0D?AAXCWmpY9vS8s)u
zz^ySHSg@t<jEvyjIUhXBkK21+{>l%0!5p4U<aR{z*m6D6uk6mf=(e6&l)5w*23VVC
zXu$Y$IGDTP<-UT6+x*?pZ?X1;bO+2DTS+tV8Ruh!ygWK;m84`~<OFDjx;hrHr+oXs
z<B|i@H5k1PY+i{z_{P6Ddjms0$u+!xaWx%R_gu=C?jITkZs8Os)n&>YzM#;>t5FL%
z;mW>wfiI^6=gWMNz683TJM^5t|AObgmIpkB{$BuWqa67EDkEeOSs0h^KNsEtNv4==
zSgLD1FhiwMK=Erqr6O`Fr3$3Mv!6ZlM#KDsX*X(VAIy}mO9j1rfwg(}IX?;j)8Hb$
z6BiwL#FsKQwZg#I!@gSa1E2HERiJ|77#V5Aqo6I$bmHn;QH&G<MgR0!jya8^DA7ZK
zOH27c#GL{|@8W(pyAiHm{~!(2oqkK_`<x+Rt$^=$=tVofgu(JCyBNTF6a|Hr=Q-iE
zjcwj7a(w5#DD>nBzR@2_pOV5Wc*qZLq%{S^*f4J_jYFV}nk;YO&WX>s(gXL=J|jQZ
zjTRjI7jI!U!5e;B9`Oe|KfUu?o__o7H`$qCr^!@Aan6Oj=@Yhh_<J7OlPqAD&dt6J
zS{t=OnzVtse{k^(hjsgW-GOj1MC1b}_l^YBiU|qH5I>w^+Vv|&TMVOT-k0GA4*Zsv
z?syMJ4$l}MrX9GA=Nm95W55?&!)McJ!q*@w-NDB$B{(p4*8m~Q%DnU4fjgsM(@@>G
zVRUUq9~XwanqTrEyL}Ii@?<O-m1uoU{&3)=eUsyVx)P!C(KO*-^6i0b!S9nyU8yT`
z>gnzBP&)O&)P`K}3NFjDF+$$W4z34H*RxcDr*MtTQB1?fK#;8q+%-f8{h<N=Gg<`a
z7BR27EsLS|b$odxW@VuqNTw%F1bF{)f0eS8O8UWXrZMOj#z&2itYb3e(vhBYHae}4
zG>pkt;enlLWKNlo6>z{*wl#hr9r%n#GPQDrUh%h1l0*8m%t-p5b6dWfS27(Mk_erx
z1HsAtRL<}SX6+2%!CvLnksH6`5zI2&d;Cw&8-9LIaE*qH7OdgiV9D|@IHZ22b%+Rb
z)Q&-Z)s{>oixkI@Y5aWtqDIz~PQ$j$`f~omSn+qjkR<<c^DzFjhrjWL6?SOXQn2|?
z=OS`{nqpnv9CyjM@-qN+T*;)dyRE-E;Mzif^V_@zkD<Q+e&%P(bfa8;$3yP2|Hj|D
z@Qu5Ed*8IZ|4h71FW!I2nvl(S5!w5>|7MEGhF`nLG6)d9_FOR_R|_owuBXCa`U}b@
zcd#zQ!!mI$JMp%{3zN5t0+L5z#ot{8wKl^J6+^>5uU=deuZ>XRoD3$<Rm#)r0$atn
z7Hj+c^vsN?y#L_`RgM(Oxx$-@snP}ob>$>N#AqtG-W5f|?HL8PZ{Ms@1DAKRjn)wE
z!;FYf(iR~S?jK@|^d5tz>j`40Xe>IM8(4Z+(tne?QJ!5~wu-%cxO*5U>BTc(?!+kb
zAv5^22bA<UjbFoq^oHYO(V7tvu9PHSWenyjsPa4h!d<8GPMkz+nXDm(E{!KfO@kMn
z1U?vr0Dzps#P;|zI)S!%R=!a#xrtwGl-MnYz^fr5EEn903?)S-5>F%E=>p;To3edv
z*fvti`)1O<%sDSy(?up2j31@!HQ3*N_if4}2cmuRZF{lsgAd+Of;5PG4c&xQR?wR`
zcq`s4N(}ACSr<2A*pO*-Pp-|rvbx5G<LPf5w3t2;@A~x@EpLsyH9@8b8AcWzGxYSP
z@7}R(aq^c3gF}N>{uh_vW35O0tGJc1`Vkt%=gRLsVdpf?c-)JT9K1(GCq5n=crrlb
zXC#C1ix+Tt^e_v<k8+iVaIo-)(Sa6sCoQ*e4!w(mqF*<UevhZzF2>_eCqT*byr0r1
zKzwumT%X;QVeYT8-T*CA*WJYtGd#+Jqw9tD;vI)SXfuW2jxXSLDl`C|M}`4mG>SVT
zk%1Ukm7xQs(G)xZLwvjupIi;jMWhDYsVoj$8j>`Zopjt#JGhR=(7;3o>%5FTzsc1%
zVb#%rdxCu_{-$4S_Wc~@&=a2e04l(nWnj&p*!2GBehb9MC0~ExnRuzq#htNIQj=+S
zt^|%L7vbv`kHL$!<MMmteWIrxZ}5j(ctsA#e>*PGS2_^N6d-x{BR`f0SR74n<rO)n
zay;R3l+Q$f&qcR5UG%;9R$gB8y71P#;Inxhp^Jmp4LkR{0OS|GNdK!wLKf~Ku<z%0
z1=iuBbI-Hqtu$?jRH_i_8P6fKM746}{Z<OG4nuN{K)ePh{#FXla8p_L+L(o{!Y9r;
zaaG*+(zxp(Sb?vH)&fqvRm|cqT=CD{Du(xReug39E;h}=N?Q}79kN>@`6ZvNJQ|T{
zNa9_%$u4p~@TY+X^TrK}@}`Ga!z}3-7#3w24XH7bb15k8Z?pTt;>Rzupw>yBPP7!S
z+d~sxWra^^bSQJ{S1A5AHZk(-h^fpcUyYy9jUHERY9WS^hNoM5@fUAXffy<&xUS*R
zcueCue8-<Hi_82*XZc&&r%We)jHM_~>jXN#8~klmH0rVqGTyiR@l)l_r8iGj#t6xK
z8r8f^`KP5gjS4ir>fYR1w%&@*feTvkhXZ>Z+S&_C?#>m+)`f--&DoM01H?K8J5-e2
zr=NXRWAB@9tas2rjPX{3F1Tlmq5RR4h5O3-l)I6K7=ccF)$?Uc;C~bszU9AZ;f5B5
zm1#u{&rL_Ua)}QXCr`LNh)g<Jl$@H*)4R3+n`}4}f*j$SwD{~&CtT!#LCbi>f4P+Z
z!JXT@Jhz?|&zTAmuZlQ$!ae%6@mDU;y0Ta~OMJB0B~uxKL*z@G=9zHx%u(%gG<)SQ
z@wO}n!%9kdKh02(JCJ`JujE2Ic_mg_w^?X-q?P3K2)_66fUCyqMS9dMk2OM+rF?>Q
z>1^YI@H2;wDR;7)q4~T6%jkf6zi)E9lk3TA8`v>s@kTucS-4Bb4s3FzOk8WU@Y}MG
zrUqtU3^ntf2q!FCj)A%10_q3%Tw4}}d4aw{KV);{sRb|n3Hloz0%O~aHCn`yCx<^Z
zFmhikh2ZbG8=6CR^q8_zZ^dW_CT0Ax?@bAK!t&fZ<0f_H3|#Z}&N#&gdAU6I-47Rx
zj&$|OPm{&F_?2fy`{z>LK`cnA6Ve99z#UnM?y@$7&H);r&BlA2*YYa#Ig{%$4;rZv
zFP7i>Ujnz@zv#a5FY=a+=fYG->&~AKzp|*j9*<3Pu|-_~zwmyJ(0P&;{<%k<>vteV
za3QdM?kzM39+x;A<!DIrcKo^e=DBl@-c!-GB5c}a&N+N(6@ITg3!~U5Ru`pUINV4<
z(<i*?rAvOr`M{r`Z=M~pj)K0Ix$S#d7-#3m_c0EPs7U|ryKfze;20LYGrb5~udFjs
zv3%$bjo?Z%jzXkJD4lIkOm!HR&D`h<-@uri9dEszodoGs*(G4qqo*9E@q3&@kMwlA
zvmo^GKX>v%n_ltNPWj9_0etx|JB8Zlz$bCQ757H+VGOly1h*P3sql;iig4X)I>yJl
z^GZ<CGVmUFoL|O*kH`6Ze5ZSk(GP*!NQ(>3@W~MUER-t`k4mdi<}^&Okycqfxj(y?
zTKN;wL+z4ABVQ%n#l2`9o@}};_ZlIkPkGpBV+w1^%wcavLb~uQBUNAJV`q#(X(6E@
zP?pm}Ej!yQdXf*nEuYYY?==9@7+C{i1gO;H$LPh#9FQ6u23c^4u|Sc6<F2vc$O)K6
z3``R-j;*(_Xtvvc$$9?ByLC(&7G*3Cm$c~X@O|LnCF8TaPr0q)D~-)>>c|=}{LxKj
zD9n{(qba6!)E7Kvbc*Y_@{_n=qf2_zTlnT;Fu4>1xNzy3q{JIO!cthJtj`cH4Y^Cl
zy~ya7^cr}M|7+tTTs=lT8d%DOABm6m07ypuFaAxM*S~V0v(V=m&gVRXhhcgh3uV`l
zAh2r?!E@w?!E5{R$P%M8_3)^ZVDh*99t^x%KSrJl{`^dNtAEufhaYT9--$*(7oG6Z
zlC<#OtK}Ocg%?Ji*O<5nVan{>PudcAc%I(jA?(b*O)`I1=-|rvL_3vC!tkNAC(evy
zj=^u;6p;OTb03mEbQJ#r3$&thmT%-^!pCn6hCCm-M#c>xM;@XN;Rjdgl)G|+&+u19
ziJG|dhY{O$p0Wkl*-k+8V7idk&t@bY8k?XgJzji2ya#?^gEFpVvM>Jy_tFqJ#o>Ur
zehdCQZ2jE%ySA)0p7g(XItFQ!#A^~K3qbi5c*40KJ@Y%lWZnvinOyuE2FVCTzenO(
z!<a(Q$a93g`A5h~Kq{bgJ(W-V#95&dzhMXL#ZkI*;`OLp8W6pTZ@)F75M_VB3-sbQ
z=|*7UQDrQSUhbVtV`w}m@==&GO7ZEZpF$&e{HBUh#W8L1;@LgES?-udEPuCOrRUvj
zV43S+OTbShN*Wc1yOz-r{?~J!unO_=m31sC-lUs7=nRih>K3Xp7!12pP7I?}=)HGU
z{_MhFh_p8xh(=j#0lb&SA_TND%`xYCesg{UJXJOd?li5FF!3+D_PXNPMS6)dTj?T@
z;^Sj3xNErW4$zBX?2OYWQF<dE8nblG!^>$TJ57*L8Pg~hZmQT^jfOm<o5JqypZIs0
z0*dkDrh#;L+%Y=QhB!Rz?v$C!jUK5yRiT8wXBk}}ad^atdDm{9@L?N(ufP7Pbexn=
z$xH*`L|)tsUS#KoOgW4ZH|sBag_sh>PoodgQ)OhIfzw7Oe1L~(8vN@P_Ruq<EOwFX
zXAc;N);rygA)J#%V}N}4(T6oIHRMM8G=9FzGjg6YI{1ZO&dpIS3<`!vOEh$cjFzDz
z%+<A)HGzy%r%elW1SM%oN7CaxSTm~Pl+TPBrLOprZTS^`+aAssQfH-O{P?z!^(}VE
z2va{fY2+(cyl05x)4AI^&abqke)KQtB=nI<GG*#+u-9lL#>x07#pfW4+sPw|S}sxt
zj9b|-z|NyJ0I#fc+i7Lxxj4G3i^{>hd@TG!SLNCEYYXA%8T`Q~FvXW;b!7a&#UK1C
zzY~|vBujW!J{^Y33CTQc`HJKDoQ&0@=keveG{FCIoQ53<gsWrV7Pj}z<9NQ1VZ#D5
zVT;cRKN?4)UXO&I!Qu}OG|o;o$@89i4WFk^C)etQU}Dx|^pFe>oFWr5B^STAp;3o9
z8Rox+xB6ElGc+|GT%jvcT=gB8k+11UDhIpK5`J`YmI-)iJkbsut%6-Y`AIoXr>?n!
zALkce`gZ2mKw+Wx+_hF*@;eXz+2^D?hqm&1o~HRNKIZ|y@OkO@RU;vniIi6XEurU#
z2e}e)gjFCYRRT<5K^`NIOM$l$O(hcM6Y|oys+bOCqjK?k6$Q62muC8)a05#|d#z!!
z;YnDSyS2~f$Q>#M2TD`rhL=XIF7c&VM_0Xy&=^W(#8}}^_$iRz{qA4eTl@0MFH7g|
z{_UR=mc}{_B_-l`^bY$+L0CWFX!TZ{yfb1b=&49EcZpX<2GD9wk@4Xyfn69$R5WBN
zS3h<~c+Hz$xt%SH@07=5^ifb&X=eyHiipwSqO?rqtlLIW@7~Bo7@Geym}vyS(MV|w
zABJb}G%v5=nFcw%yu?|V*vhy~#P+1O{PDDLDaKrn(l}ZaOsb|A6Y7$OMz>BGq+B`$
zlSWzZ@#>?gtavP*o-xI%9937eDOcfm=sO}@cNoPVE~Y*1<RfS3xtX;W8mOl+#K?gW
z2s{RR8f8ad55F`juNa0=N>=TE{^{o(WpId@p?dL?G(3!97v6v0<Io8YGDiND?dju<
ziO;`$4nN@2`VzmVMv20EG!5S!Ko{lJJe=c@Gb55l1LS}tqYWDI-~RSD$5|ThcVw%I
zG<;BxEb#0WSbUAU|MY%)%8#?=NWAve@IQGd2lYou6tASwJ@AaLMp4K`*LbL_@sp-T
zNa8UpGmF*H!ndEib>;mSMGXG(A>rrG7M-{5lDELzMYvnOK1#-D>~10S9}XVe!hAi_
zHvSFFqxDusANOn!{*&}9UL*h1Ps%8=(?i6_dw8{ZuaRIobNv$5GEICsjr$?|{Vu!W
z+?AK9r|;4MGD<Og7cV1}kCk?C3qQ*9$YsFPAX9I<CI{kp00+S1<O`OM5rRjaq#qcA
z!^HgQU*(4>9lcCihM{`&?C?*)^j#b>A<sfioKp_c=jy>{x$kJI@bU~Xar4{9{!B+g
z{)yKsaPaQvn1r>V35SNaBKnj-J1EJ=D3Z3s<E^8E26ypoXI}^8j-2u<-br&}<PmX7
zA>mF)W`rCdlnb{M^F}WjAip{eDKXYg-^Eo1HPqFq=oFuzAlLE+r-o5=uD9?ugcdJ}
z<M#RTeL^Sy3(uHv$>II4bqhtxmi59|_{Y1jy1j3_pU4m{uWRs6MEWUjSrwtP2+dyR
zGF~K(aLBm`Tv@Ql<qi>*#l2SwkkEklT!H3puHTeJO1KKnG)xIT%SXY>Tl~Vzy#gLT
z{;k2egR7D;3ap1vVO`5WmD{ske&HQyN?+1)i>qD;IIZBx-!2C$-tAWC@ch>niDh?8
zdVnrQfL^N}oOBhm)T(|IEa~B{QQXrg397I*0tXEhZqF?D(l1q5o;u;x+@{{Z?HE2w
zlkh4e6(g0I0Wyv37$K~7AQ=iOQNLmuPT{6>l%57b8vLP9+9(HZ6`abHR*_=n@an{`
zA~ofIDzoA}aiqXc%F+dqNq_t%?!r=@RVaxAuXT-?c(>Eop&cBJ7re>EHKQbChE|Jv
z90P^{LasWB5ZpEB-Tqiy#z3hoB=?>-lcaWO$zMUei~%ykyBf`WIoSv5GP*?QU`)Ai
z&vV#OukdW;1|RUbS9s?RKd|T0KbkUwe{l^x(#w0KeDWb9Mw#%Y{L3>~3`)DTI@)&(
z<!>FSd`AWuBNdv!RrXCE`GCXLiQQ(3+uN&^rsbaW#XISxVPKm^(nwoe-Z+iR@JVA4
z9y3%NHYN=JTleGdj!wp1%NdXJUYbL34O4td{ZMA|I+ySBXT0O!ybesI&v=`-BMTe;
zOYh}Fel$o^V>cINp66tMHqaY_o`XMVVc436a#yE%c0}$oJ=7Yf3QSo9#u&5YE@A3?
zhVP{VyYR!`mSNt}<!bqXeUwidKJfwmDF;qm@eZHecl45cg1a)Y`Rs^A^9MWm75O|j
z4tQ_^IIg)LI4F<Cbzhgb;<z@AI4_JKgnoK^jE%(D0&QG$M;5_J&}4yM%TEx0>@NF?
z&ArC>NTaOr-m9Mza>qOWINs#+1fltloD`?xpE{B_({bE(EJ?&owSUc*(0raxqFitV
zc){oYzn9=$_l7(d#ml_pf7$O%2>#;fn9B$h_<`7kG6|B4)Om&kkszy#+t6S-hWDaq
zQ)y;OAr?1fo-~vmW%l$?TAeFC1f7+xH`e^%f<d_mUKLwlRtS^6?*~GQi=LpKGs8iT
zM9-1(q#!6{L&D;&Q39WJ!kzd3PL63ensGZL0u;}eS={=~H{aB9m*<mwbcsUv;kzG7
z|5q8RC*Z;;>-!ll*I0pUv7I^mHcWxfK(a35?F{!jo$}or>%GT2>kU%oD&&oXNpI^T
zMfKeJfw1CP7R|}4N+1!$QB4WIMt~tSywd0l?}&}~K60jX;BM-`Sp^J`??Xo$`0x~N
z*R$Th)JT;sUIhowu9J&DBkFYJ9WUj>;e{RmgMo46NO-;enPSj$I4WvuWUvZORR~G5
z^zFF}Px5guio}$Xp3Bp0ll6Y*g`GZ`6HntsMtU@S<jrmzUr%+|7k--75r0OyjCREs
z4!`g%$?EVdQBJOkI*pIDH%6)2X#9WF-T9U!xsfLJ$cV^@Tx;LRswR6G9nMTk^N;2^
zG><FkHMAXdpYGFKn$2P{Ys;+MB4g3_`^<6UW>&IT#ih>(5q{kr4j2q(27|!>hv#Vx
zx6s39?~8NrR_9K)q)eW#nR?k4f`;cc@<Q9vI6SHRM)@&3@gX&D-A$=5XjlDDBR4mW
zvdA<zpgq`R^OH|Lsf^h`Z}>O+iJ<?zELNq|O?4<wTT<RyYosnJO-}294E{-&H-j=T
zNIq%RhrG&<<KUq)!}%(o7m&{s-jWHlA;a)FP8}Y^U}tEUGC)UO7o;T<cxcK>Ip_yQ
z2M)<^3e~3x77oeMYiEIE;bVWR*auhjw#a&a1T6i8Ynfaz&dLC1Ki~zuKmr9*IhB(e
zypsCj8QmuHR+m{oO}{;S_~_twzx&<6LvK$-)@vvSyf>H7u3Qf9`|-H&xC_!#UiqDL
zKCm`8rChW{^PIbYxVVOXDTj<`OAhUwG1_Uqq{p|_{#pU~l4okunAV}5#qd`?Fx3a2
zrK>k6yjyVKgi7nfbxoIY=*sd#89fHRU+Gs}8Vg49$iTp1(@ox>mz(j*WiZMsk_S_w
zu_+l#yu^vMwyd(5woA8>PtVZmIt)I=A$hcML7SqH^-J!cy!{m4RR_`6L9Y25B<7mY
zMu(6$1|(h2h&CYAuJEsAgtq9=wJS*<HCXx|E!}rrP?CfF-Ql~Q@AK=spnTtS-v#5<
zXG=!*Y4^Q7Au1)~>niuUxMA8g3ev1fi7mlHc#JNOHzw@jaXts(k&nQ){G0`&C9BAr
zH}?y_-pjkClXj&64}1y&^L6Z$;`#YkRfHOxqb|raoSQi?ykTR!{NOA@LF0B%KpN>;
z-uZu?j=;uGN~Wm9>qJ=UJ{V^-Y9s>&B~;{7EpJ5V*t;+@4XO+HVwe~yF_4a87~GCx
zR@PNEEaKh{C8E4H5Op<<M?>o^u9Eb6PjHid8pAX?_3<@g>2IW=6k6(Ax~1{JzoWi1
zj=IL(drB#wWpHrc^nn4+DjW5bvO(R+k9dbX&Tyzw3v8}=bn~!EHeZsQo2K00%MyGS
ze<nrhp)n7Ab#6R1O1SEdiQy6D6~LZ3M!x8@)J6RaMH*V09e(?DYnees-6{V{ap*T4
zlTE*ZZ!+Myj-v~CtAwVbvX+6|S<gZSN<;Fs9g0^EN_a<02qd`24~HoQ@cR}`+|frD
zft9*?gNJOB8Afi6MY5bQ^cdW`@HBWZPD;zvMObC7I+yC`oI@M~VK~vK;2#OdK`tc|
z+99W7Xj=-MDSy0FU+#GZ=ApGs>JoUK>#RH%XXj*iFh*OhGLvTLrI3vv+f|F*%1+We
zO?s?OKuckJJV-a@`e)(YbX&rep6y$hB}h{AhMXxSk?;!5%fGGD08joJJksT0Xp=V{
z5Yvu~k0p;tuT!c4TU{{*EATB}Nw)q^&^|)Vow2**#<?;X8afM=^U%{rH<ew3V0+dC
z`BAMEHj>S4;Ne-}=L}+HRor&3UW-m`St$cQ#naJf1O6&wofRO>YlHYL@5&E~^i!TP
z)0SnA4iXvz-y(z7V^Ce?n@3=cNudUlerVRLJGiD*cLDz*&yv?*-Fvimxl3Lmv~UQ>
zJO|OM5mG?OnKC-)43FA2sykE0Cagn~krUU4-oA=#2_P5lqK*5#piaU%?>+kO>k|HZ
z;Uyzv2(b6`)yoLpewZrn$K*PMD4dNT>nQ;xoRo+WPc6jUYb2BoWnOo9u36GL62Xph
zT;b5(M_m~dag}PW*L^$8?H(?0luy|jp<oo|@Jc0oY#O9t`(}y!ALZk7Ef1XBGZ+MM
zMi<L#VlBwWAAj6$R($^8L8oDCB=$Iq%-&?oq<o9`G{(cNJ_In5HH|sXFEd)tsU4Y}
zQQ^`?oEEkgyZ-jG-&F!nd<`^teNNk`=7ZpA6SVi=`=F!OJNK<8L0<~vnpdvS(`&;O
z&=hF*OC421@Uy-GefnlWo*NBQG!!P9c*PSf-pyMUt(VbpXs3g-9^tFJ5u$T7vbWHX
zL1C?fj*wT&$F(S3oeelCXq0H~b++jL)t6uNBYr#z$~?(eb9Iz;zXs9rGHJ<<`~uGf
z8ZQI#;&9f0=y)!oL~bQNe0e#XjjqAAeZF%?km(R>a0aH)se0Vb8WDBDuQ6;QKam5S
zK3W*by!-BZH6)B`?cIW2T}PDa$<q;Ra9y2X*V|-T$Xir88?;d;Qi?Wf@g-TV%%zHL
zOG&fn%p*k*@#`pC(Zi7u6e@H@{jGRJ?!9sE;FG`lxSo)a@2oorO~FJv9lLJ;q4}&S
znRc=|4S9shO3o;~uEh)PjgEP#jMi=)@=k>Jkue_;Jk8rXckkN1J>Dif%yE?Rc=w*9
z?$$wQN6P0>d}nKmm5snp%d2tpFkVHpA)D&z;AuqLkMi9Zbs&EX_xXGQ&kiGYrsCjA
z)<^NH=TMv(eVMmpQgqwYh?AO*F42yrL)Qt%gBIG+Rg4E;_w~iw<D0x=(;@b4r2y-i
zgTPEnQuM(|;yQJl>vKFvU4{?QDM@!6L-ba>R+CzX75YV%@Jac%vT@<<(|EZ8=XpN9
zs9xS~B2%~|p4_R(v|pWE+IQ#_1~g7voVIM^y0kz1$hihZ($NpJZE;B2r7NvR(9ZCv
z9;)q`=MFHmJ%RUv{Q3&G1@X%4LUC;(aiu%bOFO4>qL;}lpyk4$*IPodvuJoC;N1o3
z$;R`k9k-tdM6wVtsZd^D&T|F;TjC_`)iwQ%w>}VhBiko*ARAe#q0sFVhNk3+%wp16
zbTR`H9Wo9KJUT@ip8;Paq-C}*V@$A{^QRqk$}PI4A$ekIwvLDK&|dncO<<C*MW61N
z8{Pl>(uX#aJYQeOGi&bPqs>)k<zB$o%7E*n>me>j2N$8u9~A;iKP`YMEM3$yUM+s*
z8F?7OH)&;~-qN|<;YjuhP5LdoWQ459-*+BVe&5qAegR+SUGyMb7$FeB=}0eU*Fr11
zRX8_|Yen;XzruZxA92l_^SWNJ%D-1|KqO`8kgWGujiheBz#}v|J&I0)_DXAas9QKC
zKcfeObqb6YuNv}%5I_oXm0v42>GI}&WQk=C>-=nWO<@(hcCdjTaojrRq%TdzXj0yS
z`%NnqWt5JHmxNGqOA|`<G4;?<aYvb^F*5I-Gz1uZ)QK`yrZ2{hG{CEvRgq}0BgihG
zJG5t%2;I0X<!KDkX{;KMEvPjp@9QW>hwa|xSw<z45=Aiv5sHl9E}pxmpjU@eem3v$
zHjkTU@!45muzFYCy%>s->=d`)nd<f$Z}X%Ek~(#MJz|$=$%#geF*xmxN`p~0w2mEp
z<+<>Yvm{*8WaHjG4$gVo<e8ms2PYno;sd7~?EulL9X1(@GRvXEg0(c@*GLP_lx!El
zC4TXu-m$Ej_Z^|~<VEP3QBrxTp7@%|v{Q>rm})d71C~w??B-LC)TeSKE@doUqNgB_
zfL?-ph_qj)+_e2}OxftTt;NwvCtv5i3n4#A-+F?VYsMTp&DxF_voVO#N&VEhhEUQK
z*RA6Q3S7nu{xQh9jkgDRqFs{T6gFjrpVzfe1v22-#7c@|_%Dce?@&-_-}ZvOlWu4@
zUQINw(FEUIH+|XzKy->s?IZ$j(NsHdl%`vJX%J)jqWWXJXZWMd*`$tVEs#nt$U#?G
zV7AQME3=*8R&Lu~qS22*<5`cCEzgdpV9CFN{9xgEf^!!~%AfesW=peloOF%Px#ey3
zJ4_wA$T;bh>9y0B3QJsx4!~XjYj-GXyJLuq3=TcDbs$3Q#M1$A1ZePg$#a{Ak#r1-
z$W!YzHAYi-!6`ZVym1?n!!yt4@D`%<<WyLLiwa%KZFI_1-_e!cwo!$xnBwob%_X~c
z6nOEDBu6t({R|XRA2C@^M$QBu5?pd*z#vEE%tLLqg$$hatj2-oKQIiG-=7I?>u3cx
zd67dr$H5M_Mz%IvEfWFPEu>x}AnwbM_2kce4w5>d&i8GwmDh?uxC$hQ=@k%L+N5hk
z%d;*_4Gcbs)%bM21aYc2_!nmKt7I9P^hU`TD9C#+%9-Ktr8igt^WV13r8?Tly$;l2
zE)_gG88!qx>sB<hyYWQaO2=PC6hV6t#mfWmX1x^iajkaH)k%4l(TfIPIwr_<v~4r4
z)4Y~xRApU4JCfjsJt?-GcLlndRi33})%iwN)PH^O@Zjr%Pk-~9))oA#G&;(ihv5F*
zjM#&x{UOYyzxn1{F;;Bw6eBF7)suK_rXfc`gziQNchY%r3wxdSBJN+;@SPANjJ%|r
z%Z$WPwC-fIg<B4^UOY2ue4H|7<TtNxdZx2fF69U^i{0y?Mk}|nkWvRkxl`;*xtDt>
z2T#!--yAyE=g*Tad35ZzQn$)dj5=lP@WR8~-^i%M8U-Uoa9*@7rUkO?d2lx+#uMTV
z826_~r|qzLP)={=jg!#v{r5kp5lTiFgwLPGwVu)U*V!}<PTj3h(J?xBuNbuym(d%U
z@TQde^L*L2b&2etfx6U_LZ5`^WKTNh#)np?=kYGY7<zn`ZT}-mzCC0u!8>{DYPDsr
zzuv!hzaBb<i2RFQ%HxGW8zXSIk*g^yNGHDSCUnN1MehJ8AEm=jG#;Ash=rc%DC=<u
z?M{5#B<ALmXBSx`^>s(~{q9lf_f<AXvuK;a3AWACz&X#?hINw5I|6?)%+OPwc=UOc
zJUVB1lda1bV|fuQaumGIHpR`P4-LoD=LspJRw_k3OSwKWH?L)m9_+&9B(FDdqs!!#
zA8iH>`e}pcRhlKylp^({abHE8Or>dPn8w+Kr8W}OfGXdfr~c0(o88_!?YEuc2Kyqi
zr`)yoi~*4H8IP6mJk<8vo)RyP=You*-C^j~>7QH&NO=dJwzJ^a_6j$6q|VNs;jtiX
zzp|Y(LrL2t2bCE#YAW=O7b1f)?VFs<lpDP;DQ09obe2W_SI><}er1x`oTRKCvRHq4
z-r(1|Z}`55t~w7*%U86Dyred5yYk&hyM{N9$CNro#~?ud8Awy##_j!Rd-Nkc-D$=x
z9r*C;TD)Aqxv>4HTaFr#0ZUJgeong3hTf6HYy0&6@n+yOGEinsUTA;GfGX6KrcX%z
z<kdflSBn-`c$03IK5>7Jpk9ve48f%-{!qHt0Qt5jWUKJE(gN<;A5jQc5Zp&{|GF?Y
z_$6q{DIr&x$;+bCpcg@$#l2o^6xI@Qv4z0Y*a$CAhz>>r)sawcu9wF_19FImhrg6u
zKxIZ?iwDE~Ztg!Y#IH_R2cjctnz;PNDjh)?tXkV|U5-?x%*YUz-(dtXfLkYGKI<|x
zuo>N2TuCs`VqJF{YcoPFvu42D`t5X_-0uvKc18iLQGi21!1pqp<x#dG*BOxP%fPBJ
zlP|Zz((cqq;8d)GPf0vKEx+}YjDpCyG)(!WbLc6Q2RYYL-V~U0@Ym^_<S<R4L+bh*
zX<GN>NgC-^hwAKEIg@j`p**9QQ@}P?Jjg2f+w!0$g_mbGk8?lPJ3}W8J+C^)Uz_-x
zQ%@ZR1EWT5c+a)VxJ1XOm>O+$lSf&F^Ci@hhsr{F#>xhq;RkF+MXeV)Zbhv(cwjx!
z7arx04AwYlr6h-El&?-J`PTtnJq<B<T?=Dq1YX4`^-bL3<uEc49&X|KD(T=WzPxsV
zd-#xeaNMKGZtgtyI#alL4b=|5Pi%&#GmP=u&M0ZhrYiUAjtoD*h6~y)14;4PPwN|*
zfRk%~#RvHUup{AI=W+POfOaFdkn45hvbVd7d9u<1;CkIR`ARG6a7K)<muw<}>vifa
zDyJ3aY4awdEAlpNQpf9w`scT!{Lnvfm9tcEI>eDBbeQ%L7)gVtlY814L!t*@o`H>b
zIoftYL+Kq2Te^1eTU4K~e2H@p+CX$MSMuN&9V;x{TE6={Q-1Nyy>oM}vrIzn$$|XE
zD|x_lfK|g>BhT}C6GNlaXqR?XrcEq)4dhIFI}81Ftxj$lG3U}bQ!3stT<HhLJm1PJ
z{vx-0HB6#w;qYpZY-(93&7_Rdq0drQnAYw8sf?5{>2Y#srk<shbMy33e|o<?Z+#LA
zwmkpdVZX%v>pz6Yw~UZhYE@#Z^a3*7$h!MsXM~|Nb|J8rpf92!^lC)c0xKq+X)JJY
z8jsGu&$osIro4r-D`1#GP8b@V89f=vTGwDS_$c$z3=aaY-liIKpr$tDq4eY%7^76J
z><7b_a_vO<JR|tW`Bun-cvHIB8G|6O?7qqZe={H1A`snoA|2K@DN~?!833$o3N6Lt
z?1vx^3)QuaF?fjqSKBd4_soJ-8O4LerOk5zN24}x`N01{MxQ!~Z}O&rQ8+rNV~w$R
z4?H`~+%uZ!Fl--<#tzqY2WFH8M;j;g?T3^xI3yl!TLZ@jZus&l{C<{WjJ+{<)J@ra
z3m`O>p0P(Z-hV&e-N|}}*?BhmbZiF9({y0m(@!)13zl!@lrKSuLZ+PT8<PrjgzarA
zI0@XA44S7)0qK|%*Ul{0_^$&n4Zu5g5(DT)?yoW@TV8txkKkwodV1$pJsQgVD(goc
z$7A#)qb>$u*P8^kXC~}VvrYmX>Jdvlmsf+K0)L&(GSKSU9Z?^|x8LRR%^0w>A}>D5
z#TZft?E($S9>ZgK8hF5b;LS65yGJWBY3|<inKw1gjf|5<nBP%+uDRQtJHpMmIHO(Y
zQQoI)0j=Rea;?=sFW!!?zWDOsAO7(lL#MC%mEU?lq*wPGXft{{@~_88X-AO_ayV;r
z;>F~-OPr&FguqVW?QpH@v9{#yoqL7lgMnb{EXXW_R_l2FY&;3G$<&nJbSZHcTH#<2
zS~%9)DwTuTYKy%q?$UylpDmAk;QDJT&b55M?~uN-n{?#hEM7HdxI1QDiow#f_rCre
zT<+Y<TQq2tLwQEnWc(_;v=EkYXq`=`=g42ud85g*bNJ=B5#F8WnQJ)UMe*4P(&^e<
zFHcTirAGIZBOTdDrltl%fYWK>)SZls-ev5hjV6z&>ykOwiEJDg;0e2P?}qN%Wo{Uo
zj!AdIbFBP3Xq>f3d6UZk?C@~B*0=8BO%BK;tcA|yO(MxCyyqC$X+DK9Ia0k08eqmV
z%2GIDoTNxo>r;h2yc2J`^_N26(=Ey_e==AouQpvItx4{1C?2Spa#PkCyfQg%7zCy;
zlvZE7a$lZ_56zV-4i{FPa?v(6gQ`6Ep#eMu{GnmPL$+6_@p~bC%Lr*o%oPG~7hFdO
zUy9p3{2mwkup}n{&acM1e>E=2G-Sb<E6SPcaK5v!hi6j-{B~|0#LS^qEcGt#(I_Jv
z6eGwB7tHJdcCN*@WY>Wqe_hyNJ|oJ)BO^Q;1351|r?G-Nm>9)$ldVX)n8N1X4Gi33
zsWUiMr;{Y@*fp#!5K6k#SajX4%HiD*)6IZVyQR2vG)u8)M(yn4!e1lMHiW=+rtmdl
zo`VYpNgBb$c?^&kE5|qE4YGhMbzTavi_Puzn*4BE^{wIzE<?XidZX9AAgd#apoKbH
z%OQPv8QP+&3YkLCnH@(_>R7ltxs(}Q$~nJD;fw;IJk?b_oJ;XMTcPb8iK}5k=i(hD
zoFs|fLU4+*u;`Dg6z)7wGxDUM$bJ5ml|u(QoyKgghS!PgNF2arY_tG{tYbrq!3O{0
zk_x-NmOC;5zmc~*P4=Nt4Vu*7Dt-#mGmGHh$B=U@1I=c6@cJs#G(KLZPNu3%0fJWp
zAkU$+j6T4?002M$Nkl<Zc(kDx0gE<HdUbVnko6i4;Tf478JZleOX54??0JkU8^S$}
z=L1fR1^BJb4xMyBaBmG$FVCzSu<0IJcQdgRvG5fKmo9h<&9vWo0r6*rEq#(DDZ<NN
zO7*G{ulX4kDJg%+M_0(+8csFm#`OlbVD_&U+~Bqg``25fEW>ZO<w$ZgTt`RNt-q%c
zt&OIOPuKX(^VW0GW|g<W3((@*^&8p;SrxY(tIXajY>*31TOD%WtCc>uq@h(Fc4?Ar
zOWk_mBST1B;U^DWJv2q(35%b_KT)NLJnQ*(b|NFw!0>WXfWuHr&d7^;k?XAr+gmZ0
z)>{;JP&^}7c$^z;Y+^y^^`f!p>*WLT&{csIL&NL*602iF+jF88%_bJc<(+nDn%ryu
zrH6ERsLb78amroJ;CKB}uI)=SiRaniKG!PS^i7jwg`4y_PrWuo{?e!Y@&{Yh``+Ot
zBV<)DH7k&MD)tI!Q-6if-?yY9wP3v_)2{68+3J|CuA*fV`aDF(#0!lnuy)oU0pFfU
zyBNLRS7-Ht*E#TvcJ0LhpRY4|G}n9<B}u_zoH`qgd9G6|Gu1iyqnf)r>ihZp_`SRL
zYMgu?Wp6cmKgpeuh|Wl6<z3{JNlI9F@+9TN6Jj;D=iZcn^AQb{QZ6$MLRfW#uLdI+
z_0_ovHH28qw`Sqt?RYVGK<?hjz9$(e8#P)Vahm7qA)pB(#3AUk*P))BUk*NBe(epM
z&^PtqU9rO}+CK}8;9$X{w_a>jUYO}vjL5tx)R9o0E&3~G@fEL>jV|6UI4b<n7;C-E
zo!fU0u5Mk$%XF^*b3ly0F#2{$(i;Km^sO5~BRq5O0B?>lZXRa@j^D}^W(1NllTKJN
zJ+hOjxfo#5PG{z$VD{MvZ7JF)w!<6hmp2$PBENl_t;F+9m#@F{#>UyfZ+`RZdc2mA
zc!yjC7P>l*Q3QukKBl);(8Aivf@b9)<rxJnym>W7hA7wkhhE^7x9Oy<l>iTnP#|xN
z_yY7d2mkW(pSOsA`lo;D+f^Tb^2x!6d5h|!53_UXM<4W>Hw+ve66K*MI+dQ0lV|7`
zC_Wwuw?pkAXWoh3Lhal$M0O$hya5GgofCZ7oQyem!mVqpl0lj9SD&LJp0}IXR{p)p
zFnHvLK?9%1Ph-3=LQ|gfXzRJM9%jACXP^Ey>tWvM`VTx?pN-LBq?m@0$Dy+*Xs>6J
zVsL6*H)BXsYjX=8($fxnJzA^Gm61^C%9Pz}f-(ia>$(y7S$uS6P@Qi^q-(~)$t~W-
z)HR1d=1W)4bGWY@=b3waaZfL<wAz5_8TFM0-EM26+K>u`o&vZ*FwEX`Uec<cJn&x6
zNv@j8>&w7o-hK6^{K}FqaRyWJw%rC-Wuxs{0KIS*C(pWf-LTFZ+(k*ox$>s0$d18Q
zV9<+pF7}KpxnFCj4Aj)&Jf2QCStnCNBzc2q<y89R6_SSRf_2?mrw<x|V4TBpXfz1V
z@2W=%@KloOlgC-N)w)HW!dE-*x-dMpPAqZmd561>Citu}vdiKC4?Y`pdS6<jUG9{t
zeGcxSL#`L!Ci5Thiw>`3EBwT-&^ozaaf@#&!S_608zBIE7m)RAoxf)RyJX3?6S5$^
zPtaYOb&=<)s?w{1?zz_c-hHl*BivwUbizLDl}97#_14f<!^SkQHIf%G9Haxk0iL-i
zb2r9nuoPUCM~oP5Mb}1lI>x?Iy5X@J$p{2thTjNw<62ruP`3zsQp|G*Y)CBHG=0(?
zrs0_4xY<{9Be>n7*xHt9aO)MtzF|;74)&=BFVCxC2`|_L)!|y1^K2BuS+*YcGKdaJ
zc$6t8=%M#X%00_#csdGiC}{X^Wr3lyt-l#1t0RU7xM+^1t!5g(9T5AygcLk93(bu%
zf=OzdbY9_|4U)QuG<d3~4&8=kT+})#MjM=TbQBJn;)`)j@y&uY9*rCEsFdbu;7Xyu
zteNRzzMRE@t-iKW)<Iv?FpPo-&#hBYZyt!JkDevneCq=aHxf^=@G_|HdQEfjZ+Tyq
ztfr{aWEmZ$d-w~sAiX~|D4%kFlyyN>+QC$Mw6-tE`OWd>F`mexsTKwW?}GzAjK97<
z8D24jc}Cud5&1@pJ%%F%Rz(J{(0MHiUoxWO8NT^VATQ)H`PMx-+IMk#X5qF^Bfx<H
zq7&+Z!Q=%A-^nOgJc-AFk)*9NURrZg8BRRqeV&K-wCMUQb$Rsgn|kT2W$IU#lV-K&
zkZtgWWHQRz0FL_{Ta6O!smS~f8ygdGYIzcNyf6@^KZ->2=1mw{fN~9Ud$gTk-?k7w
zYa=`QZO`cxaTYVXzUKdQiUt<)3(8nVAl~5#8ab~(4ptfRF=|Y+;PFbe;H`Y(m9P%?
zO$+XVtvrXiCvF)X$^wr(?cmCvG}SY~M?MR~`#kr~j4$D(JdK$j`G@$*%8^@b^G2=@
zGhLB~(^}vXrWapXS0zH0et#=ECmEoSQ|KFuwbOu!;C-S)iYEgW<K+*`i|7ZkI(sNZ
zX0t#JjLL&Ft|hI8%T>HD68M|9-11QSZaH{-eb52uwrH1>4XX}l+u+<n;h!V-ooW`B
z>z3d8Z?u*^_x~;d4zCN}v;3VALJ<^{3Oe?HF5y_`y+_~sQr^`dDpbEGX9X59BLw5#
zy*4rso-laR#dhulW?6s*x8dGvFt$V4EO#Gd<LU=Phd{|WHa3rtB(u6V4@DsF)@a~w
z7fN54)zN*Ej+xuLtMr45M!vq+V5CWCbZo@4;t;@*MYq9$P(^GzbW2<U3q2sVYhLkb
z6t_Dyax<S|&(5jo5bvfF@rHnoG`UjGG`!Gjd3?IUKAo<w`KpKQ+u?S5+n<7gK*>3D
zX5Rj>Cy7qNw_w!Mz8^;8ore#tczn3OzdHV+bwNLCPJ9kt2mRw8|F~-^yn*vhzSR2e
zJMV6=179Z{*r|PK%Pom2M!4{+63czx6i_3|Hs@@^fGDF}bJ>7*q4hF8W(`6NlEBho
z>692K0-7_H#D9}@C3wLg2?%6s9$R3K!kM)x)-v3TvGZ<t3<rv4?TYKQXmot|FuUT$
zi$-<;>-Td~KX8F2mjW7*L551@dW-iAjgL-@<c1Hc18UyME?M&}ImY7}49Sb8;vuPS
z_&^Sy$MeRpLXU^Z|KRh_^F5<qbWO;IAH)-q<D(z_=zxdhr$6~g*EYSGSIx-*xlq2&
zUK??wkQ#mXv<#KD<J`+r98!+<z|io_;Z2zjvX6(&($MK&{^jTWKFV)?^KaeNwQ?2M
z)<{WTFF?-90~77=)`#7W;@umgE1g#8FAdxtrv3fqvriBH_M5*+yUuILX%}{Codw!6
zZK>tKy^I--rXGn+M{&v8*b9<D!6CRO0HhT@Xw$l<Y#I_=NgA4V-N-x7_Rotp+FRG?
z#K=|`hji8?m3DcP!x`Pk&OFcBoV-0&cqt1m3_8UcAc?Q1A^G}BdGKBGM`jp3XeeFl
zk}}a59J;@F1lFz^hyX8Da+WT6D+4(X9+8v8FMdh7<hc4JWd?GCbKAMRTO990%L|ME
z+!6*a*Tu=HIm?GeOB<*!GB#q&T%}E-6<MJN*N)8tkN)MqPWE^d4CL(X;+fJ~gsSUn
zn)jM!U~OV>@Hjf;iibKGTA#et#?&cljR~~E-0HBw7bgd=%(U~g`Ow5Vj`4&?w?$rg
zVir~CHao-Gh!E(sgeeD|)dTDtbQ&5jnM4D$T8DVgrTt^W(Ej`7ezla+>HB5vYPU%W
zSryFPs{p?ztUo>39>FTG7?ML+)^)AF#A!&l<9UzVOQR0Mweqlx+cnY{2D}OOt?)*H
z+RD?tFH#0ZI=DB>@SGb`4rWd#ZA6?sIWoHJh&+mdLTwpITwIdA3Ly8CGklhnt@j<b
zVlHA)>yQyV%{#adu&t3KPC4cVZ)ew3O2h^`RWuuX9SuPl<0EBcieL;Ziec(gA!PYD
z@Hh1n_p)$mgh3|<;qJ2ln71@EqjaQM_@#5_;x)puAlKJcze;C(J7*oW&d-*_rfYP}
zJO?M|Sx^{7dUPD`g|Emio(9g%7hI8PD6KMMBQ|l+?BRr$+tFy(TRb|I=6Gggd6Cx<
zqtxNdSZYXpo}SFq4Zbo!ylU9S6+GaDAK>&A#=xia?aja-f=EF0th|1(EXvLs9_o)S
zIz{Uk$k5rgowhp4Sju6<K&Z0mO@69)Ql2`(O<1&nU!If>o~^fK$W3t{kj=A{KEBR$
zSPZ4qaV=)VD>6tfED)7WeHk1UoZgSOL?^`a6GTe-!V2`&IW1$PXFMo=l8@pe>$BK3
zo)`IYkkc-FP8yx-n=zo3$xyMIu<1Pt*Tz=hxi$w6R=J#V=wSvSgPS*MWH$Ugj=`C9
z67u$xAAxf>Ms^rd9MZ^!{&*x|{_#UI$jN`(n{<L|IldL74LgOP*|6{cvw)q3g{~5=
z0kyrMR|`<a7}dkKNhc01wQ=$FcvKe*w8}@y)?qa*oC1@)KndNXS+bxa;>$lEHpJzr
z5s`AxPk#3=1^2JUZ3(1YhN!5Py`9~!AdLn7<WZ>1Cazw$oL7l8DEIES#DzP++kysv
znQA-+b_$mF8W+Y<+9#TLSc8JmGy{R8&U5`ma3ECqmk$0f1z8(E&nh|aj(({$MGl%T
zc*s0n3W!Z;E<nLhOn$UA`Vf9?^MOTQ(e)5457jGw^HD|d=saz)n@6q!wsRg!GUX_1
zJ4%>bJCD49)s$P#?|&LPeE;-+O3J)I3eP=cUxRdo=uatSD?>R8);hOr&$nY0Q<ro)
zjrBn^JB}K$UfQxQQylSORH^CohVC<GUYBHmkcDu#;3zN((AzB5E?7xTF<ZbHM|(T_
zHsYiq*c$V~i*hvL;DK<wlQ(PbWG@Tff3S1oH&KEVD<uwwj;)<>uEDy<v<7T+;snk-
zdFvkI<1mYzSll<#>7AUij%OM}9l+ws_8InNnCMWzu(K+|XGZ--iy7sEo$6wrwH($b
zeEjiG+6X9i@K&Q;e1HeqF+LboI^ZYu6bvmjAPrIDHcern;SFcTN4}Ax-d|*-&ArbE
zIboRO0LOPn7*URvo${53uj{QI>sNGE?`J&#{v3w}aLms%vT`r<O5GX)L!j7icIt+q
zc0C#g|3=makfjFo%0^<X*JE?x@!sf=1CJNSXYap`=fihoxbhiQc#jWuqD}S<Obp$~
z+>@t~o22>ti_Z_f3IA_p^bLP?g#Y7b^U*?kKu{_TL1s<6TqC9q)A20Y;E|&=3V!07
zCU|W~q@S}WHna%nsNfw8Z-MaAz}-Pk)UzI-t&@hYy*hsT>8IIO=F@{;{_<Bjewitm
z`v?E&KmF%^|H#`kWSrcAbDBdiB8S^DP|<hr%Lp{|AP1lQ?!m!l_S;EX>3P<yANsq$
z`@5e1IyBPe>^5utl&{8WE839wFG5rKJX1E0W!r4{c{>K}?RW=FdpedJ8Rb5B@I}7E
z@*sJ#|IhuqHCTv{W38|9McWl{h0z%5hnBgkXp!=SbQ;=*yuI}}ys)US4JDn0d>$(=
z3`mwqeBEgz4LM92vRrx2wN6tf%#f&ON4wbQTUv+iYn_advhX~H6&+FIBM|NK48M!C
ztuD$9?b|5^M;^Miq2$Gzv!vq>EdIe~$yIl7hL?5_p86`TXAQZA_ta>j)~_^D^=*v6
zZ=RIfx~}|NV#X(LH8{#AH}dxF7-ekbNZ#^tzaLgv#cC^;VG-?-WQol`Ohf+ACIgTB
zU4^a|aacf~Q({ab>*U9ikfmLftGwP;81`jp=?k9#tBgk`M+Sn^hL*M^jAz;abJmyi
zwSM?9BGi|x<d2@oZBV`nPwzyQZ#TYiZNJh?9(yr~Utx5A=p25Moulkcq<(_m9*JM=
zuMhH5et1rAY?14{()`f#g85zf(CT&J`{Z5uOfm-ZRm~N>ciyAW>x<h<`yH2SRE6G-
z!Wyphe(^#4bvdrqler|+;ZCP9I$e{7bea3hOlvHcVe6FS0wQ4;=4zyvtJ7Hv5hvC5
z8+;JYx|8OEU&>P^ML|hWHk|^90<fM$dTC+lT3Ng`0)doPm`-OD`_vt7N$Z;*QD}@9
zE;=q#W%TvkG%SLDm0j}~D>`qD%0gK9+q}ptU`iu=Mz&ztA@%fRT(`+symaJCAr&8V
zQBDsua?jw{>XG^`gD%8v!_r9AwWHR+!_z9iEv@>S5)gDKH8N7CMSqGo*=y8+fgd_>
zjn&zvThg9x?r2i!zHf9Sqbar04v>w;Eo|>zJtj?`bS;HKZ{3>uTW5k6^(-kn$<DIq
zR35K<cAZViq&x)iI%c$>`0dS7*$8jUD~E#hs#T{IT11AT*@EXAM$Nb48@JIK-SDOd
zWgB`_Ucwvc$Wi~LV`=90GJCqf!=I5e1%ER#AqPGkS^p7wdp*#7y+bM!7%yUcJ<03Z
zU&oNtL41~N=Xv!kw!OV|{^XFnE2FelR%p<Baa$<vfzLSOkzgP*6x=J*J(`k}ljQFL
z=3L_|<D=WUC#`)%l+!gifsZdbSooc1`huLdlZgCRmw3L>%vu;9?yDi4c9d+PLIT?(
z*WB;|eG!t@Sz<6-`sH37n&G-yO@IA^14)yBFDuNI-+8u+N2_&{dO~Qu3Dz?A=qDYo
z8~suPCA^fmIBiIgRk~v3C5zn(K5*(8US&C}OOB}<=^HwEq%5?9Z82VWuCwRc^?r@k
z^?H^>I~edRuG?ALS{)%COM4@70H+Qnav$cWq`e?6VSXomT`RlSNwc;~FFxU=2W{_p
zc65bqGo?DULod&*{fPdm?vgV{+86_@0FtP*&Ga1OK{`6Ie9k%MR-VVeah){jX%aJw
zqw|vd&_Uh&T`<n_Ji<lRgjJ8-jpB1NbW7TeZfHCBIIZjHw~Ai@c$8;3(6;>8Xe@4@
zKTRIk|9=SAU(s#F2jvP?F27ZUy)-|xa#hO_&_#5wY$3H*Y=sgKA;m8Om#}@@AB?7W
zEP+4Is`RL+bjG8k%bYZ1VZl<CGPrYf(A>gCuI8n6lHCO`@F+{KTM<lobdX?B49b)C
zovd;2tr<dUF8X0c(7qER9feI{I!xt^!DECTuSJx-G;h3-P67_zSP4>eS<6UgrepNB
zjgHunCrx9N_#mS^%9XNy7KP5sVnLp191AEJ9ry2LPCts?&Zxd=a_?SVan19G5A(&_
zC~?XjO;NgvA!l_nH*Etm^Y`ciAJbI);OM1uTRSE42I;8eN0Yk#lm5*dMrb-Obo$p{
z|GH<_((WJk?1GtctzlSmxM=bpfA$}{Q>l@&`hW1jhh2oKF4F4^87yI6;dMWk)IosD
zPNv$)gm>E?=67_MbYwtKN1=QUExoOT|GsovuU9)yuQSj78PdbSs`cm38Gd<NCbYhN
zmA5;LsPmlnM_B<kaJv{eQGBw|x2AWbL-SR3&*8|zLp{h=>yvsc*ZWPY;{)c|nKxLX
zJfHKRjRHbjIC$&plRKZ(K(d1>1AvT@8?s35pQUr=5t@!aG920Bwa~$NbA=J};YU9@
zcr!fs+rRzW(%hO3a)E}64)AvE(5ciQk35piN7*yw>wG1emn;b??>ryY<$U;Ic9{(y
z-wVuoYJwX>nt{iYqNVbJ;KAZ4FAKS=cQM9a7vIL%=qQmKCXT1Tw|f{VAAR)E81j+n
z;Z691{^1?naUL4_+VV-hYjYIdT&@m^atgltAx$~F8Gfz~1tTQuu^A)iupid_(qrh~
z#j812jzY`wFTfc>fT6KL$gBz3)|D}~(&<^OeSAE|VE61vmW<W=>L!_{b9gH2amm>|
zjNqZ{w)tGcM_Hb$*lIHiduBm#m(Fz#c~`nHrDgCc<IkGTY55z^D#OWhoHWW)Uofh_
zl7{zDP)?7efp>A+;^Ek`H$0mmKib3WmZ~j=MrnumLU*<ucsqs>L6;}*%?$1eDLmG$
z&R@iX#)t@fpvjc8!j_Keb_?+`H6Npc?UBFRM+QgmPP+>~PPTSD6dDZCN&7;Bvoiyf
z;KTireDGOxG(a)P=-@Kva4!JxOaTVUbPC<fqm@BxbTs_Ro8a#Hbz#vo2?GnRGZ<0s
zfA8Vi2zec*h_@GCLhRlDkjuq{o|CaF@He6&8sa%Blv{kG!_)~)l}k8<k=d{*JnV>&
zFgm=JS8Ellrt~v{M{#Kso>00*Tm(i~Mh9VW)a{&j9j$wcn^5Q^=T*qKwiyHwIxq=_
zPM5H2Al~g}fCzW)j+^}Lw`c40&RYlOS&WLhL!)TnZY1X*Lg20uq6R!h2v5jNCFm3m
zv%rdrTIXoZh;pqFdGhQ@aG6&wpJY4n5Vg;EM|21}^cm$(!w3aa9>;A;hIeR!M@Kd#
zOCBF2WAK=&fQQAs3<q#1xpn=cEOxZ!;Gh5bpF0vpV~wcvNW)@WaP#YQ*7nnn^V%{a
z$D1Z8J`FJCrHo&1bOjrI`<<LTt5M=?ZA{LLKOIcz5o4?!&(>fybi&*EEGbU?<^>#b
zHVym6ExgKFp%`N7Z8T)+4F2k98}}u<9nnX**|MKUWXXUxPVj!YWxdIweD^|e*>p#x
zcayuF;%KT-hSNSKU-NFbn_3_j0(pZ=C|0E<4>vS<^2|cG8iwWd?edN6ShwLIt1}(N
z`|+ZSj(s*>dw3KX=aC~*+C|rmL~crN3NrBVRGGyG-+~Jx=B-ZSC_iscg^%E^4yK6b
zhs<b$;0f~F4GuMw(k9=|v>wCb^L)$@o{wVePg|PTFV{4U{7+-xpFMwCUg13t3_OnV
z%_QIXSZlo^%Hcse@9;KVOH~admOcwg#i<DG{iKSiL-M-V!tCELH|nu+ZYpOUh)u-O
zWgTz@2HvQ@fVQAEU1XOmYjg5kUCjsJBydm0rBlZ|46Ydj1P;TcIt!9pSbTAIpg%lY
zPOdAbiI;Cb)JwUglfL=4Ga^V|-iPj&;g{#aDhFKrmhLnMeOun=nlV{Ja;rOB7Cxns
z_{8-(Fy&kDTS$Hm<7zy&Nnh4}F+934UD{YTMhyL`+j%BKBh&V+ciOl&>n`IZ;YB64
zrbdcu@LE%wbnf6Z?F+$@e50Fbn3#R3pE$A0`*PU9#dW&{zn<x<7`tJ~i|n_=NI!qU
zJFJY!4=_F8+<^+?1HF=9Ue%o!2FOckSY#wFfXV>s_t@io(<jZMZ-}VCzid$6Yf4xi
zx7R%SR@7*MZ$<xQA_!JHerH5i3_>m9=C;D=K8%r|jdBm2(ltLmeeAk|Pxy0S!Yae{
z3Sg8$5mte+VuWb;;^kGYP{SaN%)R+u&nRlbO|cv08L?52gj46r=&(S}eLKX!tl$RZ
z?d*MF#PZfVZ&hjBkH<qJ;6mOR80pZ)g>A#L8T}Q%z}69)1Fi)fvBPp@!rfa9?c{SU
z=nR!RoqRgfY(;LQYj;-$h1W9I9A!JAjP@BoXQqs>VWS^dEOLg|)29^6%<D4F4nqU4
zUcQme!do$K1}0_53qwhwKODH`ckbn`yNN-TPSIKc^rZAU@=aQ!DkB&pV~016&|brI
zD2GyGgsAs^2TQ3~|585WZIf}OTBJ6McJXhg&;nB@M6Pzz9Vertz5d(y{M6`6KCd0W
zola7H9LjbaygPFD;KAqh0DPIzKQED;7ARq9DVcr`g~6OS%EoyeXx#YVw8E0XmBx7~
z%6c3^n{cBtd;2`2V}IF|)fW3Y?xqbew6bmqWQ*X)8=Q3L>u6kL7=7%vN{-|=oyJ2}
zBNNTTGk7ul@V;#+cm;VPQ(VwQd2eS=8J;yV=G#LI9`e_=7uf=bjA@0gmFbJn>iNz)
zZ^r<A(Qn}JsMyzqAqY1CjFJ2C7P_`3<arF(<!zk3c&tZ(XUa(3k=MdEj`L~f%EpJ_
z;-SD}RciIFKU)<&EPn2Vpt0rK^6pOCg+%I^G72@<>dVurJoONkTn-%=lZ=!}!)T?4
z2CcM%w8I^)$%|gfs4TkAHF&(9ZA-~3NT)mpT03jQHGo!ynsVx-jN&YBt^;7J`*uz|
zIAlktd@48hOU9gAr^K(i&Qza)&IIw&I9Hwmvt_S%$&|kZ&$5|H;-b?|^QwO0U+{#)
z;|PYS82W13e$pmgr%Lmz#(D?|Z~BN4yYQ`Nk+raeqZb<<DX;h>9z_f|2>T3j#Kag3
zuhuK!HzMO`b(!fHdkfk8v1J7G^BBop+oP^<jZErGYAlf@MpOE_(UsA;@zOCm$u}I3
zdhv0eCgL2*aTfOJDlgiWc76K8e-_-8^G|`bV7%`9okmEl$=BujBjXTBKr$V(dqmWL
zYN=9YajCku<ul=Z@PMmzHoR$O&18?bLnsWP>>X1^;f6Wj>V&=4YS%`MG>KYvaY<R-
z`7URp<|UGv0HPbA+2&n`{8c)ej@t8#vC+kV1Tg7z%JOqp>I8Tm&~Jn;IHe&xd*Xv`
zHr$HxOeaklZ~+?Ssw<C#5d#8w9WQ{RG&Fb(mIoly6KP25)2Wn<%E_NCxY11KDWV_5
zONV?fpHk;}u$UIUX!7{+qaJOP!PBUMH-$py&eU0{w{k4{%w=Fk(Vz{wC1UY$(NTT|
zf>G_yGPUySOvz9P;Yu<e)C1$$d)ee|t^44DH=08;M#ga(oqbLe$RJq_K)o1DaN;G=
zco}S_kk(XE4Up8&D59?#_j=?D3OZsRMC+<_@{_ROD-WKnX3}=eMT7uMe0h}Vjmkyv
z0INJ_Y$XGlqSt#L<dtna<Pk}_$B(Vmi6Lf+C-Js~$ETM^2V}OW^hbIwS_%nIoISU|
z7p{_a3M^%!ziVmGMmUYJAQ3}?H>aMFbY@q{-_9*KpJajOz2pfYf=pc6nI`EXy4D{B
z7;i>?F`)7l&#z+yJWHo_lJ!L=`OxB%OkMdX9OJCUL}2-}&SBQ4+{p%87k4vl7VszW
zLflPyw74Hn(g|ZAI9+wGmK396p-K9RgHsvYChkV$`Mz!OlmGIQk851q$%pY6C%k-P
zv}a08hmymTQavSM&C`59@XIg0svO?Sha2s+F>k=3d+HkwGo+l)o~J;C(-mRl&MP*e
zQS?GaVxKUb1{|S{=8}cVqboWs%(Kyl#mn=qU#e_HujmxH;8L89bG9F66is{6xgFiY
zPvuQ#XA18yZCD6~Ip*QDtfsukK|31lC`UVq@JrBcb~4_W`5`xJ&6)}^`rEd(vB01&
zs<(m+c?As_v;hQxzlE9Sb~4WM)S?}y_NXo3(+<<5hl54;KFCJ(DiiGluTzhX%sKZ-
z-Sc=i=2>MwWw+Btpj7)T(`sJtH-I{fPB=Q^Rp!O8CS<nm_c1$iZ&5MWxOIFUFG}bt
znC6H-u<f&SOB{p6;6z*0ZoNs_`*>llX^C%W!9yOJ<?o7-BBX+7sm)4b1Ip5^562+j
zNh7;-hr07RDxVPoUqYa)8a|+J{TW)TN9p0b;m~-z+Vas2g_kla1Jb<`z){W$tN!bL
zox3h3HzD~;Vea$mpWNqS;;xewJiM->dNt{iUf$khN^K#(pLQX&kX`rJjMz{8!>%y^
z97<j%^f;p>jI)-)sNWtOf@h3|PqX6Qg3P)iRVqleg00?Ev5X1KVf^OR$UV5>cJ~ip
zB>e8dgTe%0;eih3Y%#S4iZURuoi$2(V4#7$7<ecIqY$G+N<i998RV=pC17vF=pkea
zr+QD4_e+Zb(^wfD$}!TXfQ@7s37#=FluPMSz<tj=>F&mhQllaxPPFpe$mZGeuX}di
zQE>88!a9DIzpuWkpGAVpC2wlU!PrpV<G8jj)A1-@s5PF3n#x7lL_q{MhdNPy74YCq
z=<BVB67m}6&wlo^HYQUQt3F1^cI~t)rn1md=h==vFd2Gi#uz~p4N%@?jHuV5qvvpt
zel=!iO3F0A+o==d1ix$Orqf_Rx~G7<6?0%LS>831Hj2%Qk$TCC7Rp|-uw3AHPht2j
zkh(Z@LSJ`}k{EiCX$Ga8Rc+>Gqc^h1h+6nh`@5Fe^Ys4D!)vqvU#FsR^NgW|JN3Mi
z@1(%PhY_{>j<U!`_TvEaB5){TQ(f|*b6=T`Je=N&A(CBRyAMm0F<EB#@SHK=?VC{}
zB)FO`eE2AfO7q>7<4kYh)#>TedOYrBFBnrDjG3sqbhw8HcX^WHm11z*(z%2tgRd|3
zrZY{)<%_rCX`iLIPfvJSbS%7x;Z?qiawo<DcnsCMx&G|4&oUT!knh}N3;5`S#k)sY
zp#3OsJ@Mq(nfQwsA-A*U>cfxHDTNm2&(qljHY0<6x3jQM2RzHyeewM;-Uns3i&q^t
z0$K2vvrfQ<Sk{uTTUO{|aor}sNqg>VzR7=F&l<y)5w5DEkJ>pThT);}R0pp6Y44$1
z_}-CuXx~wMJiR=QVAjwKFw-@GaBv2(9F@KB0i733wDZ+AOY4R3cGhl$!8&x0{?I1Q
z$XH;JKl!8^Q(?=PlQO#QBC<io;3lonyFo||Zn}-}9eBZ{nRWOFH%&n^z*6TLfvHdA
zf-Kv7k&d~_fD1W~GS#fD(m`|pqxdMYVX#9V&U%o+?s>eq!`~PJVG!dZkaEo04($-H
zh2!IyF4y5(xPKmlidR->J<{&Txpr}NkwFsAYGmBf4#vF>AKlScc%J$+1iTVWTkDix
zFu3tKxUaM!Q_EnC($Vy{0qU%2QBIzN5u;c+-FPuE>eeS{r?7>?c8w?awhr}J*JMli
z;$8~5XJEb>zkgqRgP$@h_enp0iRd+mVp8XlrPzsVAd@Zj)v~v|OJ33SJ3S!_;Gg5X
zD!3X5r3{gF3a&MdAgn`kULBNAFrP&-n{Pt|ES*3{0lD^#3301puGb+=KM|i%Ai@ak
zycdIVzg~&cfTiziYy>4jxsKu*d^OJTNX*vR%E1Ljg*tij=2`B$TV-HtjPg2n%0aqZ
z>(B)Pr`*QnydRXq&0SYr1Z_IiEZ|Eg*H>kAB7wsY0fS%{zZgf1mExIt=+x0qz2Slm
z8nh#~?TF+T<Rip7bd6=aOC!kY+Ppavxx!4xz>v{!jR2JKK{mwVd6)%-=bb+B8n5*Q
zc)RGKJTTQ=U?2&_1!4)B4?GwiYbVdrJn`VHhQ<wQ1jBiM^QO)<oB1RUMGt4n3T<XY
zWj=j8K;#U(Huh?P!3oWeZ_Fsg^VN)dz1=HcSaGIwj-zZDqG-b~rR1MJd8_>HhpeKc
zOrOAy;fHPxa(|fgaBb5@lfb8ff^?N~N6b-T%A@R*uN=p0OgB71lbs<KI68;(?5%;X
z;PYl&-6iXB&*wSs5k8Y&ieE=mx}<0wqv;ME;nQccpNeN_$=JIW!{*MWtWDu)FU~=6
zk9X6489WrPAl(WSFWj6_)Fzmh_>wrENFb98YEz*O(=?3wp#vjm<e5$o$I$TVqtUWZ
zT=MKBZu`QQF7c774acO>2_zCr%T?zz2{fs<C@>el!L-Qny8`%*PHP0PQRR}8KFAMV
zphN4Cf~zCK4_JAoliqbgfhFz8Jfm^iam#^t0yvAuuEASnItvz2s}mlswBp;r=G?(i
z(&Dc&=tJ#~eBvE%0ol?%CQRFb=T+Qd!MF?N$?$MGbjRxye-*=+tTIRhbsQYVJ42@7
zX}_Qv+GK*E3vS}bX219bmcQV4GjU)D?r+3ms4TOlC$jN<B{uh_tdGxOIoI?R-0=q-
z-cr!rSFR#kQ3C@z0s-oxRd4bWv|FYr#I6H#PP^e9rEhNC9<Qu&T1L+30ADgzE5|8+
ztrx&m$HOyE&<bs;3v*^fB%<&ZkHN!p=h8Xn#rJ^!^9JQR{!sWHY5v`ekbPXCK69R%
zaj_9=2|U7SBpS<d-&^(H4m(`KtaeHf7_MLhUmclpMwrrYB2<{C^IM%O!s!@Tz~)df
zQsY)On8{qmX$0ceL4hh`EZ0q#q))(f%28+XC_+tP6K;4dOS=jrY1e2+M?u7tx$v)0
z29~ohM#!C=0IvA8{zQ55I^e%X(K<i$Tb)Hm6TyAqLcy#<Ip;>(;H2Eio6&gOgjFng
zE*{Mnjo=IydDPRW#Tpmzml0ye)MbRMhM?oHAQC)z1vGG<N_3$ie&88C$w!Hg`#7@F
zJz|=zKSq>E*!r(n9XwY!8o`r?NwDZwo$-M(d-LtAX$g(2U!W*>OO!_m*+;_E4iD6F
zmwHw{p;KrHKY#umc^rDrbW#oJ)a>%DH(~&UMiz~3qb_n})L!qygE3MWUvN`zbnR$)
zQzq(Q%@MpkXE+LEamkI*#4{t)Es(L!{pT!Q4|Fl|3=GE7@MP*lPTcF<Riq1lQ=VYz
zz}KeRHlvtNPO?c`SZAt^m+j!q7?*F}4gT88c)7sR5x}Dhn_Z(@&Nph<Nvj^x4FPp<
zu6v|eot)+kt>EdEUVG4xD{sLl^SAMiYGYGQ<Rpff{AdOL)<5;t#sV-0WgP~^t052l
zkW<pn5q8hLB&pjbR+Y<+fKw0n53VIkGLX;h`qk6k14*i{wvMjl=eZ<s#&`ITAyXL)
zMnO8VhhP1WU*hNkcszTS#g^ggs_#N$)ot=9W1cL$o#~8D`3Bx9TRODzymZa8^Y9;D
z{Vq+;9h5i+AMI!45N%VB#J3Yq9iv(58SC^=;^~ljIF%8PRSh~kN#Nbk_Bt3MA6`38
z9{7tdUCj5zv@y`B;gc%iWmuZ}WndOul#NdYFLcT9iZK%Oa=2$~OY{6FQ;^A{?Rr(c
z-br{W*T5rxWB}|Abm6#}Q>|<OMjR40Sq=QdPEnyXy)bP`J+*a%9el$Ia<Bqi)eO%E
z(bQ>!DGgr0y%kM9?FC)*WB4`O9bEHR9`$tedD1L7@a!)UiryapEX^MZ*;D4*$yVGF
z;y(oBk9kfp3ahLN0Ri6{Wf_;}_pHoM!&z@2lxfqfcT5EX@fW-WgVl%;Mx9={TxDWp
z8I&{WaP3Sms%*D>6hAkqg^)T>4Hz5^RUVx$i<IDt(+RgjEhex3S-8RleWQLFbOdPN
zfS>w1s|(zSFZl9V6G3UXhVNa|D#3@hL3$$(xCuJ@D+W>0D!ZL<lp6(4u)(2}+Oa16
zaXQ}`FnNZ4l%kGZqYrX5?5h0KJ1~u$D2&H(jpL1}voErqM~AlR+Il9hS3-4;oq9@{
z4Z**y4ZK7sX7RF(Ng`?T1g+Xo(@V6I+s?`nq&kT^-g=p6hQbAlLVypRStyA{9ntaJ
z#6XA^OC6@8QvWXI4K0o6Dcq%?x|MQV!yT~)*{I(9gg1_2kjUV|Z9Z)$24)m%9=A>u
zDV|x-rqh1(=rp{H48c8hgo6$gjKWM6x|Mz6xp9$!;xuA>nt5?vp0gM<Rmk`ShR!@4
zq&gn8?8tX$k8<mk&zNCNt6lIOwF6Jy);Bn=?*PGd7B0_QR_4>mnYr#4nYw%YICaYc
z;ES^?Ku!6Tt(?22CAIR0r#KV!Smkbw3PxV!Er-@QhNV9=4u*LK*PQbvgrMFhS$x~6
zw%|jqt$#7S_Wsv-B{)-H>a4>?H(ne2p4`eJTxy+cY8P$eg_q8z^)GlOwG4OY+XyVk
z-*jyFy{99S2ELQ97|Xf~E9LM%umjX}YB>3lRfkAl=%j?oC|+7qlxO617Xq;NIoD7K
zUg~OLKfIJz&tc9ZL+FQ$*Q0IlcWp88Xd$0?vX16-jrRN`I!-*JPe;d4CqL4YNzIng
z4F7hd>9OI}X&v|^uK@|U(s}F51?Ta?*hp{c=L4Unp3aWqq0OK2QTB|H!{`WK#lZ*v
zwQI7EFSAHA*Ir{z1L$AMAcyL@cD*JW@|)UQo(S!OXQW=_ogPYCi-+VqW%rGi;E$fw
z@fxr?3<ra0t`fEGLzuvh?u@L_YGjVc93744uS*FEVPvU%tL_T^+C`a&0=aOV(CeII
zeoR>@kSz5k{OCZ({VM_NC3~+t`h&OfUWe=J!3Ay;cm3o#^}1TmcAz?!^D>?*&B`-9
zRqkFwAeDPL+e-eYo!b}@E<(}HF_<o*VKz5*SJOxMDMZ3=M@OCN<Geyh*s~Sjbo7A%
z$02x_nR^X@c#Fp*uym+(iKk%@Xkm>gI<Kv+Iphal-nt(v^y#c~Hd^imVVQbh0BB^R
z<T-RlZdM1|>7~>QJjb^~Sdd2Kna*4jt>M9Yv6TsD3ca1*rZCiv)y@#np~AI>OVZts
zk;GWC#{=9KlhM|^q=qpHLkBt2NH;TTM{@><X##k6WKjcUls?>Q(QA<K44q6>Rf#1%
zxD=9d?rdv7D4(ys@(Of1E}0VW(NUTtEyJa}$n~N<0q?X}&J?kDc}k<41#!xXnhvZ0
zkAUm6r57(yRz~ujf=WC&Ga7%KH!fPgEOaCnw=>PfV0v~Vo)R#_6AFtrW4)VC5rRee
zGDa>k;>iljmanu$=N9ZfFtnU8Xef#;58U#99EA(to9WC`mB1$NI;Ye<brHA1d|f%P
z7#}I$Bk=ph;nd6e9`$}67q_=SZe>wu3QpSb>K(_(?83dn%!wE9(0cKLLA2I_EIQx5
zoz5t-XCX6LG`*$_dyY^<9RZh*=x;BTbeb`So;@BqSxjyHNsZ^o;ZoqdSw4Y|N8}CN
zjyvj&ip}fMHJ)y&qq>IINubV+Pu-4UsI)Bvr#mqy-pJI9bxsxlGk#5Ljm)H@i`-d5
zQe!+ca7H|a5pNWnZk+^|z)x3}zXWWREa|jZ_!&&BDX+?w$OPP~%wmXRes-s3PCSV%
z-P1vVTbalOovf2Hpy8oZ$99O3QS{WVbz16+<~k{BwOqg0c*#RonbjdjzN<^2k|1C4
zqhA8MdM5F!ji5(illi2v=S5{AXKA`;OXBG`>9o_OlZ#IbP`Ix)?0MJhMOKgq&7S4U
z$K^-z4-Z3&)Tw*%NbgM-IGv;AW)#6E&){6INAS^^S7*R8&uXAIXMQ@R6a2ckIQ7Ph
z5dGNu*a<iC94tF((@9G|RMyd@<&7ay-Iy}y&O@IB;IYAnz#JLA)<5voW-<KAU}H)v
z0`v!Y8ojQ)OdZJblA(|%bqe1ii;asuDBb*$xg0BmI-Lqq4@Z8cPOscI{g$uOe;a_o
zEjQ9Hpx^U6I3;n>!TBvL{MLEJuk%XS=kh6gMXp3IpLa&c%h7YU%li8Yd|jsB6SpH`
zU?B+MLiE+i+==qLLy5)s7~!7|(_1Ya1x4_6z})HtzXnN&%UWKc3M5EhY4~Yu^(drs
zCODX{V+a3!@)0}%)4_GtC<)^2<YB0A1B3A>TV7j*|1h1uQM^&WjQU34!7;T1uPWT!
z!%rPq-)#(=;OJ-xC!;}wP>zn>-V^FwBO>|Wvy3Nva8E(-B*=H{ZIry$XbR@T?9owt
z(%^Kyw!~JKl~x^Eueh>Jx9BJsV7~CEVb^d<8Tg_cM;l*orWhT9dh|2i$!n3B`Yy$>
zUX!G3rOSVx8Lb<&U8d9HQDVT{-Rg&?U{h*>I<$<ODK8wtUAP12S&)<TI^>!7dZ6-1
z{cptn(a>9u;A$Zz&qvF)xHt9xIE#%L8PXe(S~CK6@lB&?2NXORVTXD3?)(PDM1k@|
zS#!b@g~us0XyM&09|%lYL^8L55qUtPOqCeDS%6FsXVpCKQ_9W?`!fqe^I9Jx0}j(@
z^NWTlM;xW4j_r_=j?%5t%NtM>?XA!joz^}BJUR>ohDPfc#k!+k$q0`zhHY)%XXqPw
zH9dEdj_$`FTabOwt@(8zU*w~P-aZ>XjW-7#X!b07(!6+n-ob*(8?`cM?&nLr(xdg1
zKOLh5v{&hjo<4b0UNRE2i(7XX$604$Q@=dHZHp0~=FJ>5L}#C0c;kNd6WL@_Cx|xk
zqm_*WEsXx`v)?uUXP^GI#>oA&XX{_O2^*P69ZgZ)NSlZaR2d6N%G2p6rX%l2HvFVc
zl@GCggDB9+yWo}Fo?W|WI_c99O_Ip-=osEq?S<zM@9;KM4C<TAZqsJK;1Sw8ADxXc
zZ+m)_8QGXL+6o$=F`a5k5<KBNXKw(yQ-5jJ+7PHodm8ycJI}~Ao(XiXc7vDj{pR_6
zABd595kbGqo;c+zBU*-#N1nFX@XIqsR#s<)uQ51dd>-F~Et+k*5v_vUaW)(Sq%=!h
zjqk{tjvYVf-G-zY{hw#~CKVkEM`hK>%01qqg8{`=%I%a-3PH#6F=@JZG_<F~z^kVu
zn9@D;cvGfai>ULPqVe8Zn+<P<G<>DcQ@Wzwv!)I`!g$`s%O{MhL47V@kiSHFBWwZ)
zWx?F$R~qf8hFa-?-jSjlyf_63Mh-aW7utt7YYPbjwU@ge>G#9y(ufPS_LIGIJ<ms8
z6|#xiRv>DT_TaWX==C0iq31+=rxDU53;HXF@4@@tm#SU@1lBSF5Ko+r*rrFb7$`y&
zAv%r_YYZBf&Q@BTmTlEBLjbdlw`Vcpra|aPHHdmd7$s>)3y0>{K?XJ<Ec4)zgCHxH
zfuK|IhQf4KI-Uqx%3i|I0s~{=umobIS>LTFqrngCihyg4n*vN(ye%Ghw%mkZXdKO(
zGI%v~j@+d3-pOKHn`w<3of~o*=bE;0md=7zv_l8?{Wx0+YJ(~6%FQ2b+5h2*x>;b=
zx&ofC<|2Cor{LH|ywPy+H%u+wI>KCb%n1h_#f{WgJ)R!LZJUK;&jJJeI!Z|c0y7Nh
zsBsX2U9L8mrE#9i(@w#&T)S7^x$Tn!UyBeYa}^ck6ge_tzvnad!Np2`yUW5&J<ftp
zaV#xLm1$Uoj~3v^7;#wZk-Wtt3i(#v4k?}spPWPGJik6llnqz;w1d7Lozc}*7Ea1v
zf;lI&gGybLS$?5S!gPLQC<>?YI=odsbZoo1Ob3Uy$$#MkUJ31}rj*{7O449b;;+)t
zc-!PO^|~AH3i|qzEjjE6G3iVPOh=LQ8&7JyB@aUjo+mfQi*hS(q`*}tFrCg+&y-dW
z>wX}WEQW66;_@;bMye>iFiJgVIEn}l87w-%w(ZC^zIljS(pjXd<Fkt^W84%Lo}wJX
z`HdPz(FN&KkRnN-vhn%l>lT_c+6)<)w_!#uiVpxvj)$zkk|zHRPQ*x(MkPb_O$ucA
z5>XxH<lXW#KNxeAcRI=(SJ}rRB2xHy+6-{0jEC;In0(@Lv@JS2Jj<DGYnjmvwo_Lp
zZMp_ro|7B-cv->InK_yVnGq|{y+-SG(`DSWz&!D+w9a$h%Bi<@IkGx@NB9I~O958$
z^;PQDJ2Z(KLYrNIKiB#yq28nzHP*^pWp`4H^ybTSz=4%1nya|B0&2<#n44LPL^jSd
zeYo_8H7TY>)zj{|rA1)1uHnlDz~dZXYC|!nGA<}>`$9x93ijHvV+j(u!iCYF()S45
z8K@0>xRwUM2+H!a#g2R=tpHE5{z_199}I+VJN>OJIG6sfr2p2f3jQ%cwr^!EB@zvn
zdEDt@*9)0f^0enFf%lO3R@!nCS^oI2V?tn+XkivybdUt!TOc~(n{k&b=SHXMBe+R#
zVctnPYeYWDsJY{E0vZ5o5p*SG!w`J?bWl`Yqy5kS<KK6c{ll;4^Xrcu=A&6@fIcJ6
zTVSV8QypIaqtJ~O7$9%w<5@cLFSC0i<EJBrDEWE4Dh)T~bX)V(rExir(w`9kh4VPB
zY+=0}aywHdt#9D#1Z!O7%s5%QSa!-M9-n3Kv~y0~*TzwdDN5p<_ulQ9cOnIM8+mJ2
zKF>$x;DKHYlzMcMo+pC5@vvNE`)?g;8%^TYh(`h=e8j~Eqqf#PWoQuaEdO{fQil9J
z0z)Um)52r$Ji7+cvYzRf$k>a}jvOe<nNf5Vyt=A9&(TA?w9{Dzhx87v^B|p5>f|~>
zI_a0iUztffugc}+eB`VGHi`2am?>A8X!4^UeN<xv|BYbqnL**q!+`(13g}#=kOJUw
zC4J~)7f91$UR5+gGg@w^eVR1qdHbY1rCv<&%v%j38^I&EpIf01Y0y(gm)##B&s=xf
zA_~XXW*Gou%;9GiDu)jzsRJ4uhQ{qgym}iPZuU(T%9R^BMwJp$Kb|^p9d&mntu{iv
zmEM8VDT1W?Z9Z%8EM9}Z|NFnMtbTF$Wjq{D4*swI^Z#ht``daMZ$Wt#-QrPuaww0d
z5F9eXSWyp#vOMm~<2*aPolY*{=4C$amHOL!N_ZNt$BQ)T8;60xz}jS@^{Fh8Thlst
zmX0aw3;GrbnB*&ecxdshwM?#`X7qlReJwuyw|_f$kipG^&mXi+{Oo5x3!dZh@BTs&
zEyhT|_gq7E5e>6K?_KCMT^5;5<YgC7Mka#?z6ef3U!&x7fWh~L&jjQQHhqEz3cN@?
zdFy!Hx+psQLo3Fkx|5^6rW*M03x<Fiv9$e&(H;}i&b0S|50`c5WOwac^2qb55xqe^
zoymiA(rYjHYTC8a8YzzsFi5OX83^)K-?Y$_9~=R$XE?A7-~u#*!1-jwNMW8A+bvU0
ztNk-RD%)^ppyYn(Ayp?Y-he_wMrv+CgFHw&ZJ3@q4voPu(9t$m8)RTJ0(knYak`au
za~aPJBf9dIXY?1HfDg*<rh}<tU%@3C!8_GKKwLODOIkRglcO7{CH+}Eo#a8i;NU<S
z<?z&?lepBd^aHJf&cO1jF+QB<RbhoFXYneRh|p6IEc_0fqpRq+<ymOknN;8TtDILu
zW3~eOJQ2H8(&rGr%i4#2CY3fSd(ivNAb;MJ-?A-easj^O-ofzv4?+CS;;tVOj=~B|
z3}QJGb_%}@CP{0R=G^-*%RQ!bBpyZ!nASQMcQx74C+)Ik3Bg-YsP-T6YNA(Y84Q#Y
zV_+?)VgO*8^xnuZUE;aMRqu;3;gEy!fVXQMsCal*sB-41@w|;Q0^JZC(b9q`iig6~
zm^<Qtcgh8aVle7e<`NEgfn8T)&b}4G5MjWGTls3lq#il})MzIkSn5QP89{On8imbS
zgJhXg_LSVLCt2l!u@nsaTECoK_gS7hJIIXs;pP4arqA&wbXwClT+np7QGAu+R#$im
z%Jeb9B<Q`oDSRiBA}d2V<ADG-alTVRiBJj<YbOzt9H06Tv7}WFLxfRB*)WvOvrk36
z9O*RG5AQ1v!9jxpPrPFVcn+_4oTh#?$WpMso$Rg*U>Py5!3?_Om`0xTXqf7#QH{62
zzX-n=5Yi~SMtR_v%U5T#(I%!-pw70xjV|)RpK_$|u7c;<(62ntMqU(mh!fe%=sf*}
zy-jYN&7Lh3wz}IO$`_N>nVfk`g-40}F5|YwLCSd^d||;zu(m*8fLxr|ABeXzGF#q+
zo|y+}XA+)~{mNSiL>V8N`biz>qS79qP4PEf@Iq&FuqaR&1WViUobluPN@V>-M!?UW
z=e6E=uNVX5X80PIA$T1#xeQvVZ;C18(0uRD-3rnP5#i>)k%hGWYIVSO4f{L~d4u*C
zA~9eYB%6c~<Y724xvs}fX9aNn!D)QzHJx)J)jLS8KL{^;V7Pys+m=!KB^{Zr&d7C*
zYUzW|7#Y)vf=5Sc%f3>p-dC5~n|MiIJm9FF>7+}RjWnuLuU8^X*925$kFmh`RhRt0
zsyyn2pTyCbePjwo^>fqq3r=wCpeX=(Q|$49re$0(>~+qJ3EQ@#f>O@UN3YQ}XjVfi
zUK!KF3Y#5EN2l-fn{qSFH2OABd?F+;Z?x@1{=2Xi&FL93i-xIBWM5mV*Al*w={So@
zuP&4~Wwr5f3=;K(hic>*Ugr<ZrUrqzt*6>DQUCxz07*naRJB<&OO^UeBK0d@gKyh+
z%I{Vec?Pcn%CkRDfYDa=-no(eL61pu9k*SHhQ{xCyn8!l867(SD*0!tzz%AOR^g43
zL0BDSJJfUt-8nN{L%;B}2gC`W5i7%i(Lh0U4Mqf`olUtBK?*|&avNy!m@}2u?xYUZ
z=gG~C-15eaAe)BRZ&0Ko^g%i!H;df})L5n0Yev9sNS;35%AoMj3Opm!n;AtDV1nVJ
zZ7O6P0?&+ZTeQPWTJUQyBtJzgy-u-<qEZ&P>QdKp;8FGr7)Ho65Tl~2@#rM4{X;B}
zg@=w-J{^Tcm{tCPN!dKk2jmzz=&3XI@j0D|V-$f#7a)O+-e@A;p>r0Lu?tD3agX9m
zW7lydfezs0#<+E*Z<&F05}1^WIxNo$dVZ5pzxqc1R5o&1<$?}ySu!dOcva3(f;ARW
zR_pxM+lA4&*yKY~upgUJh@tV#HxE<pC`aknnxb3Tyr^NTSY@B=RR9%l#r?=CLlRwh
zeR$9wN46kzJ3Q`OG`+$l9EAk;bDajhCXcBfJhBm=bpiO#n?gyO_FHcMGsZ{q+WFMF
zl9R$FKk<a1Yv#G(^epKR>GCl1<(q<A6=96=s#))X4(Eq4u(B9Xel)vxceYHo@YGt9
zH}WAwJY=lvXuNh#R)6u!UmX167r!|8<u88OdK<B;yNv|@%m4IW`eDbDEYiD`*J`=w
zX<l^EX6-opH0wnkK8lx-v5>qLR=$1j&5r!vdGDRd)oDCD>eCK7^|tON8L|!{UX9e1
zfsui0yrSt7QZH@f<Yo-{TioN}acE%e2;EYTLh8W?(LNZaWbaqMj)y0_divy>=9y_C
zdIR1=wS@z|;Yp+`tWA3e-|*O3nszRcVV!^nH@m3Bm62^w4}HN*(&EOLv_8jz;%C{)
z+sIyLg)iQU!Dsl>1Kt+#{UJPX4^QodjDop3IW%YhWI&t0+Zfb~m;1_Y62e0UG$}2E
zlYvVO$9!f1j_AUZ#{<&q+~d<*TRI+SYj5oSdl-6vtz0rmR=_|fcsf$s=$w?9c<pP^
z6OV2j=8c==Z^xUI>W)6z*wTrxMq3{4%P5iK(16xu*x0$Zsq^eOZ=NWZmqi&XpqZe}
zi$A}Z4{0(8(1w9ckKMkQHAhX0*WqiYHbdu5Eo!e@J864};F;nhXD9X|$vNsI=tM?r
z)eg}1Mi+ATNYvmRIZc`9rN5*1cwrbMYp_%q4i0nyUCLuq{Lez$!kIx9IQZJZcxh>X
zIl8n)W8%=bX{5{XYFO0Tr`t<uHhNa~<-v};z1YRRlx5wlr!y6kl5_;Zuw|}a`CXoy
z>yl7UVa|VnP^e&U<sfR^dl-}Jw3zT!r<1~SV1dgvSi&FNFr0EBgsagN+#brFYb-k2
zud-0HYcICd@;n{N<?WEg_eZ4j4HB#CsU5=bC~xE_do^?oMSuewi{e(7GCtR%l!oST
zP=-(iya|iC7r!)=WrVCUq=l0PEGUQI>im}iVYF%FEkDn7h#GW9%(?f>;Y|-NMaF=8
zK{<>H2GBCH+8L+3DuUn!F5>%>9uA(hA?5t4@VZyN>5|fBgMcRTHT{-`hlAt&C=FL<
z1TXbFPuiA>w%hd<G5N*inFH+#I|<16T*F^u#643^Q6?$N_R{C+c)LK9(#j~tR`|`=
zI#0QM6*f874wxyV<xYK2LkFlnNfS6H=hhcxL>Ah6t4lgl06KW<n&316oRQP=sC?p*
z56>#kqu3cKFQRZRQV0A1S3Q%tV9=iOA_vX|-^T+#m2BIw>uE#sO?in=T^2sWi%Bwz
z-jeMbb&8SIH;3Y>NF5a5J2HXZSFY{!>+8qu=!4_)TeJBa8LC%LXF-7nPN&yWW~W2q
zR=$XPMF%I~cWGqWA@s72Mx7WL=t*9ZHFC#Y6xt-6*D$pTb#T<1)U;@lb`U=6Ao`c*
zmD8NHdEqi`P#dve)te0$dBej<KS>&#cHAmWH?mK$l@+k1Svn_G-S9zgcuTzEdb#_N
z7cbC?o~aR)duzPN!0Ldi1ESoPE=Y?mUy5A{AHGBvhL$?nw$0FchkHsaf8nu}KHjy^
zZEODw5WAattsmXJ;{B|0HE&bF2agPvr0q0w?w>_xo@HUIE!)+lIBG=buIi;5UEx`9
z3R$!josp0h&^AcFGwp<KU!Sz$@^+ufhbi?U1vc-1!{d4oy8Zst(Z*=f)c{KSnh-vu
z?ZKz|Ip<|G7H6G&484mOFr99StfjP#mcct7m_r`H)=#}sM)K4E#_$A~4z|v&k_VrP
zW8le?I&`2-zgXMCSaThsv>hg`Z=Zp%+PHOL^i|tz+KhYE>SA`K#CsFPk}Pu+{#Fjy
z?RU0kNxn<h^o@Asr%Zb#F1H)I_8v^kE`D9C^Ipo`WuB4whM+49t>*gG-)ke})hF(Q
zC{PZtO2C}|)PPaD1{A{8AlMY_u8c4wiKeJ6J%$lq;?NOnkXj9PA194n=bjIW*Gurf
z|8M_mjRCtZTJPbL<rL@pnWE9q>^We%LW3hrx6ktNucs*~MhJW;{r#{S`hFZG+d4&n
zg`>=YtC!bi3Br1<w&~!P(T8hqvVe!iaB^rMMj+||rq>e1`(_MWrh#P?oKddl(uwm%
z5cl<)DHjas87$(^+(?XvzzAKv;O78S`oaz-Qx68kGz$01*BH^L&yyeC8jYYn;-7gQ
ze7MKEpf^faSaL+B7(sO-!Z%0fmZNN{T$QEs2>kr_D18#Ij`FwR0-6MI<@G*KQ|5dV
zLY=r-r}NJgln&^oFIgsgIv^d~ag<?4z>z4lxS`{v5JMlZ0ybfA(BMsjpsx?c-HtZ`
zPH(@Nw8`@#FlA7Fl2rkDKJ|~{%Qn@#A=WS86~24Z;AWJQEy~}_LP>FsjJIG}`DSLG
zY#V^_JhZb;-8!<(vlLyuWSi*6Pb{M1SsdeN?dXZVjDs~NUYZ1li;TR@i&J1T#bZRp
zkf1=v2>Qo=_=n*1{NVrnfBx@oo8FOa=w$SH7&^S2ZTdg?_~V2B;XnKgu@Yv}J{}Yw
zp8Gm$uh0kIo@9G<e1Fn6J+2OZclURxV?M6<ufJ*={P4q%vi1C(gLg7QHwADni>|?I
zn-7k3unH$8GB<S(qU~JCB>x;4RUadJovByP1#iS@Pe1+ZpVlzqm0&nBTJc#YNVJ5J
z{qSJx*iu62Ll%S=F{q@e(HeTMDTWs><~bucNcx{6aE+|J?))s?oyYMA@kZh`gJgLc
zO%XLD`|9MyjnSh+4^On)B$B6XA>}jJ*Q@R1nB2i{1qMZVocuha@4Wk7r_$c@vBA&_
zjo@;e>t!HGPghoU!XwkSkAM7e-qLw;@Zhsg<01RHuWe6786369qI2W9O9y(`@5-3|
z^WjMyFqw^>3R-N6;|xI1Gu0XT6XZ~9<a%pFTLp&x;D`4}i?&TZ7oK|n*LoTv)52N4
zU|T~rk3*K|EqY*dRTGdUi+yur+ECpEX=l2cr)P`_Z?n;l6$!<vgb+kr9eTs%GRQ@j
zXlr0FC~jmRM5nf1$-@w{-pPR4*7Jh)s6O+?+&m-m^(^H%p5g1_VLivWA05ZY86Dn%
z<i=}htJ9Lsc?GiZy72qrUzc{pfmCpC-Tq39Ey1fWSJ8XuEKs{OEW3z*mkXFUf*BH(
zU>o69`K{!-S?A<w=asZ0uwnLsRX~$)HAI5Ns(nalL>QE3O~`RP1YB&1hypyVz%WL<
zZL(X0l&6?!@u`knfRhHk-^k%E{%BKp%W|!2jDSRlb|K{xurAaLUKB|ME^!Uc>qvz1
z)srCVfR@L?D|a2)Z@9n*1SHTDBB2$|c3PVdd#=0+M-sw&86g!~`4U=upv~XJm9)Xd
zn?Ct%n)#UwrFZIcku(~Zz{{bd6<T%xgCi6Nb=dG<iX0xhfUhgK1|(s@nwwF$izgbc
z1(Hoyni!P?<0u9lZ9#yg^UB%LuvgSX3@ul5QX*I*#g*soDH!#N0#c{ruAUzV?%QdT
zfs}9LX3;26yeR5!@oN=-XvRQ92MNG0PXj}Zl{udcR}T3L?ea(&>yW;qA(}Gzqqab*
zm15iLtFHZs@nJz<{`$5BIFc|vcpQR9IhMV6;GiQPSG|2-W<MCUBa`;esZq~BPCbT~
zQQV<7K6F}<ECd}gUL!2LVT9?F(+o?Opj`vWcafkwA7>1$LJK&w4*6Y5S^4T$neFqT
z!-}7}7nFoRwk>$B&0pp<(2I*%^yzu&7<@0u1e&&EL@zm7RPmU*feYpeu6yi9u7`VP
zJX<tbqx04Fm8-5x{#U)1*QJ}|Enb6JqzAzko4oE^Z)i)^Ft`nD9TvXPXKT>GxFZku
zLVxWQZtyToOMXkw$j7ZKYlN~NP^NaR^ErBg&-B+;&c+J_)QLmn>Lw4tb(@|Jw-_ap
zwkwg+rz0+~cp}tfzaG5?AFMKFJ(YL<h6&J(+@}cBg6Md}dyt16U?nd@rG`T4CfyX|
zV!|SA+m&jh$PTn5Ud6VsKqWKDCXaKI49T|p^*lJXTH5kf)&@)61$5u9+ZZlMTRg9g
zj>?9DasW|0!-wTPUituU<ef^kh#&Y<=sz&eAD8^)A^sl+<9p>o903u9QCib^x+nNL
z#Wk`PCy3X02vI}e=;%P>#=Od?C(!y|_g}z|qO+k>G+lz-K^X-`AYlDV=C=i9Df69-
z3f{;%k)QnZuRA64&bx27Tn*C#Jg!S~+(rzPv2C|!wBj|f2t#n9u&)>*xmJd~3<%QN
zV*RUp{0?62q!ZuzMj>=jUmNq}BTU}7N$=EW_DOj&xG_W+G|GdA5sVIkA{&Z^R;iQx
z4(k_e^rpd?U+tO%I3_)Vt8@)*Z^iIphz&i{LBK!pIwWxog*VeCMx8f|#F9?g((`<v
zt&uK1z46AIrKh)azWCzv(h9wrH?YwGot)t=-W=&$;e!zbS_`_-cBLs_LiU9+*QiAr
za>O933pMbbnL<b!N><(P+Pfp8Q&SIu4_BRLEdd(TP_iI3>AMg%&+#tmtK~I=pNd!N
znsUygIHyCL`cIlB*ro||>Z1Vr_1P@OMB@}uKooJO!$QMvzWKPuhbh2ym<GgOYdO@@
zT)h#obsgw6pSBOaRkq|P{0>fd0QU%c`FCm!Mea-U(2v~Ukq(46jHi$%0>8-j?ZEiu
z&wt*HzocCUUi>zje_6vq#u++yvzFwq{_3X(|LuSIZwEj9o4-jLxs|#+?jAm$<;|0S
z{nf7yzBXEpH^i&ZMydD&W(}RR+h<R8Jn`mz{zdEXP4?pPy`iV^nBdum`I<H!YP)1Y
zz9qAG=crdJ9p|%nb#(9yF=@K_9vNUXM+SJCl&w=$_a~319oVBK!V@}sNiav~#oN%u
z=HZt-J`_k!hI^5Z;E{qthfSAhbP<=ewkTO9ZTT%}jB}lPJSaR!&r>F?*Ui(&QsDW3
z+ey}xtaiBuLX3j8GyLr_qh}*`T~G&B>Xk!r-mtk5c#C%sK@V$~(30oFUMb#GBBSUP
z-(K2A;(zj!jYhM^!_*bt+m{Kvd(ngTPdd#u<Q6cKOH&)?7g@iOp4!w%jn>GOd^%|D
zir)3LSv0uFx+Ap8D{)XOFQm~pt|E)iBPa45>5#XeZVU<Ej-dm=LC3SRX^$OfhBm@c
zUP(Vmd&8X57_p4Ei}P8Z=VO!hztL96ur@iYqATozjgbk#MLV;bYZ_3ap64Tuc^5de
zgns_u8N3WoOY78e48<f5{2H>U0^Ys|jhDWgs)m>47MuqIFio`uWMGC@;70CS7HyOV
zO({VFnecWPT*dTx=%y_@@Rn@cb#9Iy_5;2!(C2zVaOe5UvFqtPbGLzLo#uY=bm+WQ
zr5V4M5z>k_`49Mc6$_>2m_`*gggG#+(hsrP*@UDeK;lc-q}P$Vt^#r$6!r;Ui!l^g
zF0v9Q`AaA_$)j;7ixSfC2!%5NUhvh-lmaGd@X9-l3(rWABG)l!T*G88$_Q#2gCYtu
z&%k=I1T*n{7D3cO6tBcB+~MMyz>w~CgzA?At_%r+v|P+JF2LOCmO^Vo@9x2tHv#U1
znlJ~cX&6BOp6fYWdnAqY&8bZBxmRznQQSSRO4pY}u^ahQ?mF&W-T;=q4KB59L(mBX
zo>EJToVTV9jmiL=T8^?q3w5AO#pS2@rdCtGRmTGGlk`2di1If1ieGTof%Q4fkhJ1E
z5|T}J`I5fM5Dsp|?S*0ipMZeZC=m(Zc%9H?;cZcat9q^2k#7dXl11JR9nyHM=B+J;
z2szVXdd`)9m|aQTGdzc0Q%LI*j7mH0NF5m{c-rfcDc9?Ob7huXratn@r=H>|`qCt|
z(MDdROru-*;bYq5{u}oX-p$Ucce9R4JVJnF>WhJ60pRnzQFD4W3zL(m9RM<F`!8})
zgG`5&XQpZBf^V|3Ej;dLA@S*HI_MYy^0a+yd6Xlrf>rq$87i-XbJr9Ur=gLyBNWeE
zyQ|k@gUh%qe(uu&RyRpC)fyfTh{-Fzzr8ZR7m8QlY{`Mmb?6$n>&R_+)V32oX>9)1
z+JP~HL+f<R;I##{tcB2^R5?)S;<d!+Q9SbNJ_kJMk{WL@sK}c73q1vFr!xYF+^PT5
z813|6^;GhcCtg2BMW<yM1TnC(hhN)+Jc=tFLCrq{%q&G^D>9k8#^bMAwf>?qRm1q$
zg<LKkkMR{fqm<x54wQo&4sa&WRp4$hc_^ldlK~CXLH(Q=jfIgc7cp?ovrc3xu))Yf
z3uE2c#|1uxT)0ZvbVja@OzWf`+72G;Z|cQTE-X@axd->jw>GyM{i)5uZ)nni#pcNg
zmdUx*G4Pfw53Y%?8*S*-@`FR`nzQ!M>p;kb->r^wxy2N>iP@gFoR^aB9ygDlR7fB3
za^_vKA9PV8WEhT26BcqMP>C~LhfY{fvHi2_<}Rahz0QeWusWp>0kAiR^${JV<=!a&
zP4kc0!n&?6oqLG4M$H<}ag>DgIz7)Qjd?pJM2;{CEr%6KiR4D34rlXZ_CFv9tY343
z>M_dJ(yxQQ6~i^oO=+}1Dc{b@dix;2+qH7#sZx%jofYlRo7d{`ud~;P4#H>K!8dZ1
zzS|pz9y}^&v}i|&&`BEC;7>!|;uB||l6NylZeNq<p_|4dEH8x)lPAJXs^EafXnEJ`
z$m&6nB<}oK7TD%_6?u54ey*3-qe_XA$+Pa~5IPtotE;sHStWDue&*~cv3qgHs;93T
z`~HVIwH|nqGVqow_wL>4T+GK`86AbMIx}Id>|T4U^cB;T-)S%S#Zcfe;&nnBxKf68
zKU{Pwp!}$(H-O-)@qYRAK@CQD<N4!^=yagoe4>DEWz>E<ov4M3j-#YwxIEd`|8Q3e
zLr+(!VQ|8?>D=K*x}v1w%VzxI2RwcHo5Yo;a?(r<TYic!@F?<;N2A}YF}XdXWq62R
z?M~6wm5}#``9_K<8|y9b<zYMr;;q@hM|)_ZlhK!aB-@DfBD<KvNgWs$$5}vWp=Z=&
z=Jl<biYxw3M#iDdTW`OSH>FPF$@ulia^55|>VBTLImnOcq<7za=itBo&;K>+EB^Z6
z$3OXT@VBU2D;h(UY<%(HLDqvjEY0nSp)GqIntZ9lS}W8JCis??!Nd0Jj~+j&{Cx4{
z=iOW9;~#(0wliDtk9Q7EWW$=8o7y-bOi1lh&v>4?2gxiN#uw{R%D<*fX_1wn(!iVg
zDC;yl7uSv;JkUn)Nrytd;V4K)7VFlJaan^W@$H1rJ#S-3(?!Z@u1TFZU%np0%wPtn
zlju9TOF8&R-d7uLM;RD&l`0wa!ht8D9sdn`dW=U9F4B{k+mT)CMd%!SaA;d6d3HbE
zt`FXOKVIK=+jgJ)f4tp?l3hoVCg=oGKoJB;fTBotiB{9rGi%dk`~FX{W3%1W)zzZN
zq5ud`K_pUO|L^w`7ndlK#j-Km$UOI`2oDc;4-XFyal!|Anuk^UJmpj-UmUk=`u=@$
zD9#VQjaQ22;aOG;E-#+?{r>y!b;9za@E%Vcr2CB+rG`YY#O7j#|L8t){atw5IuQlU
z2)KRcc4<|)#k!5YhELBjuY?{&FW#w_FC&BbQ{EmampXrVICFF=OEOJ99Rdn4UJomg
zj^fogeUah!yq`{;dPhH`9;(YRtSxab9!Ud7*Jmjcqas#Ukt<(<rA~tU;3lq565z||
zd=t!h*rG3jU-fWSjxor_n2FI67}Zzh@hmdvUNE_z&UEVNwza*CgG%c>84&P#@tpju
z=iV*dBN6Xq4X(gV=Jkd(ZPHZUa~9?Up)-OYo?O7}Mv1wW+)wL3&kVT4%Kv4C&pfZ}
z=D~*Iq}kx)w1?x;iJngQS)Py;87@PHd9r`K;{4dTjYkBX0#)hS=%?YQggY}ngcT48
z+`7EOXv9`P3e2H&eV@`I)II}xpERLK{?i+dkhxx=jePPh-dkEs(hK0OA9rZBG*)?&
zkJSO^Y2+-sl&}27saJqv4gM6Y29~hdKY=TmaGu7Eshr<!1_EzLhc5C=qlFIesCzf(
zsKw@4+>?*|uBD-O)>vw=*FG20qY0Xpt~j4~%3%Vdp(Xg-iou{VX*9)mD6g_<m?_vL
z%oV!ep>mWTh=xW^iLZfosO0&<sFD_WrL9nJ_~g;#m2?#OQXJ?3o-|<Bm<pW3(Caei
zRS`S)gl62rmg6K$;>B}$3V!iBAZx6)Ou>`lvl4<^yA6EeiQ`x_tYHDEqz4C&d!>?r
z88kdc39B;fU9sgB+U6x#WX^kvCRgiTyg|`>Z}RIw4pPpmz>p#H^=O3ly^ew=yDIm|
z#Y_bD{`Axm%^6Z4hj_@<ZTNs^fh>?cd?sI&k*xt6BQ!DqPe*WPl&EJm@=7hM!fl$+
z%!(GvM$xSepGG4~c;AZm<AY36eD|GqVxR|KG!8tXfzMNZiaCeIV=H;<Z(vZGhiMeT
zk)cQuh(p<*z06R_K{mw-&W00tfXFbpLEE7Xy2v;6B7b4YybqMhuzUk>zh{ki{wt$l
zU*#Ax61;<}fZiHnlX}JNN~6dDiY-}d9ZKAA2*_6fTXQFGp)yw4As3-_8uCR;vRC6K
zWrFicgC`x<XoJPriT(<{8CWa+mRKLgpy81u)AWw)7!w*29W{BiJh_K6dec|eYwYHW
z(J5bE6j%6?r!foyiJl3b=&as4%d2>J7kRZh<}5l!S)N>bTHSb-y*OI>z-{sUkgw3L
zdP9C>1Fuz?{3tV-R@bj|yg1pTbN+Z^FM^l%%EVw{h>#K0(mo)znyD^ohG4pwILWH$
zR0g5qNSc;Af~w38xb@uYJPQm)fch^FE7gF7J9+b68Z<*=f+u~wB{ER9oLfHPRO06B
zGS_Ii8?s3bm1X(o5J9f5f@gC-^cor3>AeLs>6TnBo-Zi%8U;_Jpj)BSP$XBZy=U9y
z{z=poiVh$<K<mxkdA;9DxA*o(FaI(l1U2`=K9+mW7hy29A?dhiDaNH7_3$;8Z@h7<
z4JKuR@H`Qej_-;PbQMU@u&B5#XTY{W4}3kX6{>{tO#Xx>oDLJFB4YGduu*7<12gdn
zyp7E6-M`l!m}g&P{{!iL2#eM4J0p4}Dy02B7$2;A4S`B7Ke$pZ3>Zp*6|bxhAAZ?u
zpF_8s<m0a%*I4;3<<n@GWZKG}ax)y(GF-+Pn)wDs_Y?@MR^-ID8I(L2I?4eDO5<)C
zh&!2&@ta@&x^%SUm}2Yx14&mSAom){&TR>NMhE=8YlzfRR1^T3p*a^j+zGnLeP>Zb
zOXypE3Bw<_pb1`~hxCjzuxl)ZCeq=dfL?id@H7V+h(qaUc;vreD|c}LCkNPz$L=Xk
zN*N4Zt1zf(<-veMtIAIB(Mv&3o~&!xTKr1(nII2_uH=0peM({`eeq6O3apAJGMcBr
zHdKm|meL|OFXMGsAz6{U;1uPRAqDvA;Vry7p*6)?dx;#|0K<Pzn{o;;jZP2@jB$Uz
znPCu0yvvN`X--3w^AmEH#@iS^L&SUx^2amf-gV27xF&|$k#TlqG(UsUone%1{|b#i
z?*^k9t;KsI-lM<yo4@I5nE&g4{*P_|h~aS2qeMG(wFenUeL1>j2q5!s7{pofNiN`^
z$7l`^<K|g<n<+->+sK0*hM%QTu!;mPedFZbz5AI2`&JEnM*h{-y=%eC+^85`DYu}`
zn%B|I@Cf4;rUBHrg%8Qdu9Bria2N1SSu}zSj;k3%&=DmUmPV4RaHMgjkqVzG-VBj;
z9}))~Dkm3-8*$8U>YPo|GN-_Y247@F1iqQ?^L2J_u1vr=>6(VD<>pEz1`A8S_}C<4
z?`C!!Reucc@jNUYf=0qR!<!)SEZUC1uC_cO=I2C^EBlQS438s|FN5=%SM^N(NqfQ|
zjvlxb9=JEBg8uwAUM9NG{Iq&qA|tR>mph+^+~7&>J5MGsgzDq)9{vu^qxh(&Jb8ry
z32<M-Lxw8nI5lsI?1pFcU?v{CIxjFdtE<X9JEHQI1ZMIH_vRRq2EL2kO)trWdWt4P
zm*AYXh@30~#CzVt%0XVYqmjt1i^=*bgFc6a5}pb5xD{@24UKE0=RLgpK0n}GiP7H_
zc?{n~4o&hh_{@AhI-B8!NC`=QsJyxFw)w?ldzaTk#<XpX`l_teQQjxMr`>^iyU%GC
zR*DVH9lj}Taq5pelOm^v(j`|F|2&(H*u495zkl8ck&9!Ga0}>`e{UD!>)kwj^^{Oj
zh6tf?p_ELV)JWAjS0oHMj?jdPlwYm#<Xa_>bCpQ|tK9v~X)(qVWbrq-Zh1?WXF?mo
z<j+d%lUx%Bt72~02q<ynyY@#ojIx+jG@}qG1x5+w=&X^iyNx?<gGo>+WJUocz;z6!
zc-*AifDDvxQ+|zs#sTKaNJ;mI0;qA2H|PdO@4<C_6oJH(vs4miLkQC8?Z}7XH+1pJ
zfZLRAV5uM#!W6c&D-8+Fz@#kjgYn|oYQSB$B5b|460m89w!4az^xhTkob{rV1<#d1
z1zB;Wgo9x=XZ*>_rY#@AT1VbDe$q_vj)KcMLHgnsn&u|2@EzS^AEi7diSlS^%;5uA
zqBX#$&yuFzfxN&QJ+H#deesl+?4;}E<ULR*TKITPu6nWE_cZC`o1XTR1`tZeDcCmA
z#2wt^oA_IPBLn6MWc5fIapxKVp__R$3@yOOsqDGaO{yegJs1~}J#PAE*V2H;u$(#}
zj|{hD)d-`AR}gHnk1jlHj2{!Y?`B!<8!-gramMgVBMohy3P!~;66I}kIL0A&xKKKW
zAto7KIh@AJGqq9KoX;uk#Ki~ozs7QgX<DY_qcK>)mxoL7jc#xz+hi{Y1s{;ops^Up
zUTB@y6Bd~5emH`jkv|wz7DbgtJ{xZVNPaEvv;!MJlusxw%CxaNIFO;PXbX*7|N2as
z$>;DruO<Zie61M`l0ky{ZV2r%9tY_=+??{z2hgV%g>HSI{E4p~d(OC~2MiB6)~b=A
zc4)mJy0Sk+vpjp5y|fj@a1OaXVGJajvlxIkB2RRh4&<|EhEx(SKa3JO9yYB<d0v@?
zYs*SzBbUizzN(BK!67TRN`GZb=E{+zJNmqWF?vavAqm!8Z@=zi^w|;kGLuxxw;)4g
z7@d<Q^hpDQMl#XEuVEBmW0>Mgo-5QAMh65=XwmvwNbut+34M7*Vu-&?y}6R*$2!Q8
zGoauukO(kWS>Oi$5UTR#USR?<FoK_t1SMEdP`r9KKTYeMTjh8yc;0v0^H(LA=glkc
zYOKDN_ItPc3AYym7q^5z{q0jk^|l4<<xl=(|E1S*%6*^YJfK394cHR)9q`vP*<k$%
z%%GqF);sHJlQh5|e)Ly$=~A4mCj0EGd^fYhCL!rpCi71G)i{gWhL|u2$$uFZ!_j$M
z_fsnE3dJY5!9k!@%g9tNdlKBw+>g6=?q())hEL*o_+!2?!RqfSmBdF6N>F8HJi*CM
zsJ_YaRur>_|E<hxan@54kC8&)>_H;PugWN8pj<QrJPc1W0gs`=@G+T>Cx!BD&ovxi
z1hAm5Zuu(<0|zbS<G#Z}DHBhKZ`5=L)KrL@@d#AOCB8DWqEi6D<;i!xjlwXRZQeDy
z6dc%kL3-C6TTVj-S9oDJnxJ`=ed1rHmkf4gBe+n0+{DpFb*Y<EBj+l`q}M2|K#`Ru
zfhip7(FVWMt$|wc2KQ0Q;w!UW7Fkd}!SljOLrbY#%P<DRhhY}&8(M{b;kDkN&FLuL
zq-a`3m$@r?RZs1fTV9c$Q54B8VN)W_dy~I;in)zDJ*B}#idI$mBs2B0A1ceeExsqX
z$rb)8OAcj#|ME744dvIurF%jY_68y+d!?LY5+}UP>A}bI^LKLZJVO~v=DG?eJ!c-a
zr{OIbeEjHfd3<&jFBd~4JkpqO&VWg--_HD`|MvHPpKmbT?=>Fyy0m>v7<fMBCym{Q
z?MnMiJbqTl@B&?=zSs)|?uY4Q7+yHb##cDt+d#s#Y?LM(Wu*#SU!3urzKq8p^Hvgv
z2kU88t37#=Z%jq59eU?)W?n%jU}FJ%L-o#u$oVX?guEBYS6$31(+TC5=-%d>vXoar
zrS#98hWBdt<lVUM6SvU^=9SZUdjjVupMD4X*j^EhGL4RFjRzSgSIVSeX1vtEPnpI*
zCrgPOIkpE$Vzo|HUx*U}PL4HFvqA%&_LLxV_yXUf<UwZIVF(ZKlq`}f^TMvjvuL&6
zThVbmD6Q*(GdQBy=%z&r1`4koj}l!8mfdGHh{stS!&_@^7`<Z!p_MWY2BRHezy8WM
zx}ra$$CQItOg**I={)5@1EGc-`nEBTE{`5;$dYms>($@x-;{cLc^n*rGX2p70h`<?
zgCHM&3}3uEG6f{@%?TGpne}d|8pA&?F7v%7buKhA=SDq|zKLQK1w=0CM4e}KD05o^
zNt(n8kmzB<^a;vVK7>xrC)dYQ%W$LH)eHJcThnu3R3;MF{3vOZi@eb3;xHCrZWzB_
zg}hTnp(RZk_b5h2SLiibc<*@*sl@QiJ?}|Vlo+`dZ!h#dk2>^>B>e+ma9DWkCD~nW
zkds2a9(;E%f8=7Q=;b~l``7yk*Yk}S`>%G_;yCu=6qSuI3<|sPS@OkIXjc47!Wa9X
zL_Lm>MMw#aiaU+nRD5s|sDI67-EYO+%Iy7qDAD8(uY^wC%e4kqB{RfgDPFgCk5Qmu
z){tuGrIm^>oDs6%M5r{-j0P@i`y}|>ww~kdl#RFJIAsH$Q9#g}bMoYdXH-ao7RsZX
z1n=5uhGPPc;!=S%;3_-=fS_u~1<NcMDH?w17pyfVFTI%0((<V^8ec*3fe;F%;@C;m
zj+Cz3eMzsjUW?#$b{?aH@(jLsP=1O=IXi+cIjgV-cIhN;t}oIXxEdVblS~4C8dzy&
zPDSD;-op1XI0_vfPMY#-D;s>38{A&^e!~I$yaP)bxQG*i@1DcmyP4p5p3l)|I4r#T
z^wUpkP~J*!t~An%kycP#mAH^|ofnlkxaF#4UBqmn{7j^#*ravJ0hXujayf0%3Fd}-
z8Ix>ls3a<7D@o9vw7toDxUJx2LUGBt3N?>PtG*l8KH2lRb@HPz<%OWw$sWU+@~t9{
zJW-4qFnr)KdFqn`30A`}raOe1#%zao^0Bj{H0?}~7mY;fFQsa7p%pz2`IC-trZ`qM
zGVHMwIl~v7c%Rh$^B6<$QA%~AvR|6zK^g~Kzjl%`r16Kp_s8kLw5Dx@ssWzpHIT_f
z8!in-pl<Zf>z0RH>ioa~kBSR=mg>3N_&4$KDCJ}`w?XAOQy0N1=$@O01c8RBY0px=
zkq-Pq1LrM8JVsH+EpQX{9iG9Zy<xt2*8zm@(s@sZmA=Vb2f;iuJQ5f@^b5}vozY32
zWFha+je)@}UmAhwg|v8^=XhxZawe6_GwGr?j$=^xm?B*Rh9R2g&oj9{dJRs7I~gSH
zM5T;VXMhjf;yxWLc_nq>_6P0Jif#$Z7XB&+Ip9Ijh_2A9<kzp!N(X+9Dh!eGP9EjC
zw*lNbTstntW^h{uA{<NCuq2czEn9E${8f&?ZFjqeE?ynV=&rxHX#Av+tVi+MlDOUv
z%AI)d1al}p5x{JKXLz&bdi41ME`sa5@V&cLMsad2tafoP$*wf7@<&<pkzhaO-t{0-
zJ%#}zqkHdv<Z@?(EN@2(Ag8^8EMTiJ3y0nFE$)ukTV};~D|C&CJhvH<ITj|indEkn
zne;C&pG8P0t5mKCn+BL-qsTP!@=-~uY@8#EhCiZtqD%=mdLqyq@R$Q4D$`(0JUzEX
zIC$L34!r~l;fv2d>*^h|id8O$JdJ6ahjAHPtg?{L!o@oZori=naHia?2GCRa>tFw-
z>Bp0j_vt~1Z@cP_mB$!EYcQuWjMYu}DL6Rz`mctQ!KSBb{|!dZX~yz}YyDh0*DH4N
za(ovd0y7FJw6c1H>(F;+SlU(mA&L8o>@0Y$QPLa9AD1nojG@Cj)yaihc^OrVB5@7(
zw4n)I;LbQ}#ZV5)D?FUWBgR(X`-BYdQe~uAM@$1xIXk?RXUnsL4&<@-d5%}&mA|>h
zd&(OgYKwdq?F@m{xJewL!w|voi_gDkL;1-kpVqkhG~W?nAl}d1B)ob%vWR!SuFJ?m
zixrA<;8ES|wY3|mc<4Ou^SlZUK2nDcn_Q}p;#n;qF5DS@=tE(z2A0|(s^rx<B*>2p
zbKc6c&@$n+^hkqU9*1Y5$pbyEUX8p4K<AWzm3Q`ekfvkzp)vU?4B&I%6Fz?W*{5x^
zq|v~<llf8_8N0D+*p(B0Hc->Zyg17qB5CBz%{Y$%f5D5G3A0@_5jiqGZ%+=g)_hWy
z_OQcC8VgNqg2j;OGUSx)+pIP#&y#>L^dz2?+x2{8I4E(?(gQzD9WXCKS~C4+;!Hgo
z888e7Hn}bR&>^&;U(n243v!137coG>rlAMI`_GwYf>BcA8S)8v)q05znX1sbbu9#`
zx56jnQ><nwUdw2{les(bGN~7l+nd*Ucw>l!zC(R*BIjG@i~Q6;N#4?+1+S?*eWwO0
z=5k1*-or^fQkD%(1qd&!Z22x<3V&Q0sZ*7g)Dz!jsnHPr)W%ADovW5Ryr?ml^P-_0
zMC~_{T-$*(G5JnBQ8R~#kz*gF(|ji5tMKOZ;NgSOQ<h!R<K$bXhOEHX92plU@p3q;
z=XsrYn(y!>*Jtr;#FPyW$J<1Qj}a392|lIJ@G|X0$}hC`xy~ycAZIiLeEyP0<d(dn
zzjrmH)w|H&)(kq5@S|6{b82Wrhw=`g2Rvo%j9vj;9;;J2tbG|nMm<q3X^Pt}{-h~#
zeKS8PU19M$fl*!3%=RM5ciraQKK#K=DJ2hI7m~S0H(0E_Zph~De#KpIouw6$4hjpG
zr88u*qR@p6O~UAN3_s?@kN@(xHhj+-f>55WcN1YR$4RpLoqY4VFxuWAL=lFJQY7#e
zaN$*$Lj+u+bSs>xV3Fu5BbkFsuqt$0`sRUnspX1&SORTCs&YhVjm4Bfo(+ko@)c$t
zU=AaJ8{SNO1oHtn%Uz30;w%ypL^!l8@JQU^6&O_<x!3S<`+{q1lu*A&g?$)<q05t1
z`gkO6-N;hWR9-_SDgYd$T@COE4&@&Ol|OW%oGDio;Ov@leO{82gQrJNTeKPSfbZ}~
z-Yz?4pE5%ko)>{tS_VhuMGFlCy3`0st#2<`#TAmSkrJOm)N6n<I!JRhZp*`rUrwXc
zOoldjUk^(7Dz4>R_yGq;=?8wEO?ih`eJ3xtZSi+;M3*-3U{^yUxG$b44?fwMRJ*Ng
zLK}vp!!HKi^A}Hx!_gnZgY@JqC1rHI_2^ND*}jVVzXnct#8|`^w4TN-n`h;E<}s!0
z$!p;+$1cH3o^V>kgNHctK5>G}E2XipoHTf&95oI~yCe$}lxLT|twrv%a`KUt@@-ir
zcK*m`&I^sEPSnynv_qXJ4SG)O&J)P9lPtki7a3=0$t$Lzd^3$d4xL8scqrPS1UQ3d
za1KxKD=_PJ56!NJPS?piISHRS1Q8>pJzMoWhlU0%+nY~3U0<~0UFom^o(4(0o_RgT
z2dAkoye2PFm*7NZ)GOtfZQ6&{RyG_3-j4RcAvcMo=y)ad3fq02t)Mf+_-K$+`n`Zp
zYFv%Jpst~8IohdbH0MZW<!5M#o(?iNX<;&9aM^H3K1bVEhOca62`@9hA(V?OrlVjE
z4*XM+g^>nTXxUN*^1JE<kFLqeSK|SczhPtz>`R{7$oI|2IirYXl%C-MzhD{|h6z%3
z@4}aUzeHUw{m>R)Vnn5bfe(yoc}stKGQ+%`M4SC$bXdUYUwQos|6;yq93)=VA2J+y
z2p%253@&S)R-_6$A=jTsLgc#?n1>s2<QE`?A*MnWW|(81O_!&RI0m@<$aId-fzIG{
zic)FaA!7g(X0E19$dknDa(L$)mMFvLO7<QBuVKqqy~N8&8(2XP|Ix29w+a=Un_Uts
z%E<A6svb<dKA%V5yY^@>%{g$yYsfVn;9XwK^X*yV1osW0UHVD*Gk*(M-t+e}Q<<Ec
z!YFu<*#-9rBMiCN?lmMo7(CCh-b)9X%0cik2GPBL>#e)lne<ofor`y_cMuDD+=bbp
zj}ekIgjsI&Ky2y31cSoU@R|v)A}!RLPXu7)AKlP!u_+LhPg#vuvTA)hh8Kh<!775!
z;-cqq00&UQRmG7;eF=fH10K#5)SUGoR<E0o6#qQ`BA%Yc&ogNl6&i9wQ5v>88K>o4
z@XZXy*;*~6VXTpqILfNW`4(CRue)g!jC-Fx53a2+LuVfuV}Nw5KgvfB2rlCZ8AHeR
z*z#Bnj7H#AXhHcXi+p{tl_;S@m3-)IkChuaj02zL9&Wy5d7iRpD6S@7bYJ|QUhR03
zls`mh`N8WXJv4SCELts|GnT~LH_YHd9?IK|;o#QAoU&5xj58}Be1R0K(wmZOXEVL=
z(C^dC_rc@e{>yIzYm|n#zBH`IZ2uD;Dsz2!M;!O>Sv_^X``NseRT^);nO#65OLktZ
zH-l0Qc8V#GEx7ev$^j=Yr}1daE#j6SX(?%CSgU2M0w^q{PTCrM&Vk+Md9Oaa?B~VN
zJ!wNDb&ibS!|-sgUN>bQM@1&UD?#JNfH;cbpi6u&^FL&3^%~lcQ(u<;{L{~J{_f!O
zdX;8XQhSHN$(*XY8M4tRp&8nn2wXYgtz=llOVXj4;I21}$My&U!?%2R_v~tESoE71
zgPrIYx$=#q=b39_UPy;`BKy28*CLO0AVr9obZNd<WG6Doa8YmMaU)*GB~Q+9tV|+7
z{<!^lIy_tQwa(u6KkvL>>CQSF6MARo*E|fJc==3#*4a?*7+xteF9F@4Z01eSGb102
zl*qKYLp(x*c09~S2t1(;d@e7{hX`+VUIwptM#A4`ybF<;=Yg+rSvm>L@!w${-oxl1
zvc*IB%fI@ou9SKA-S^^^y?OA9@SG86&c_rq`GFbO>I+YbfNtn;^|>9Ft%904S9F=q
z$#HNTw5_DM)4X{lmA6ZY)A)i>!zF22j=*&ml-W?>+-o4mAhmmJhY3SVGO8R7v|u`=
zb`>wU6;2Ju@Cm-4Fa&a?ZuC8w&jYlnSC6r%-jK;WQxD1fMLHPTCOj0s$$~35=QVcn
zo<}Foq~*moWCaGE7j~ZOID#i$bz)#m>sWLYy!1oE2_htJ=q2Wqqov&5=ae*x+akN(
zV5|r&YhWz>YQ>JYHMnxE-4xd72g<l!a_4$*EV9EB@C1*x%Xu)rt4t-0NPj_~o*Nbd
z>;5k}OS=xl9>$C!#FTY-u7X$<lO)q<6RI#8v&GZ6s$htZa4%B{cE(JegCTJlE8hy6
zM>DrV52X1=SQe9rBf50|iy9=OEN*8xYkPeapfMs;Je3&?_~fE_xL;-5KbKBhTw|4R
zh2K=T&Et@Jq3KdS3i6bjI!e-3+LWbjk~Q#F-e`OsXyw@o+|I2a$pZ`qL-7oaRP-u}
zl)6@Ya552UDJD1#-UDr}DdZff{AV|&5u;pGh<b8NVrvBr@d8g!*&O>k1#67Kp^-}M
zy<$60LxopS8^8o#&&4ZTyiR1eN;GLw@$feHsHdEZ2WTKj<Jr2VOz)%@&(*IIw_->)
z`$GkY-peQ)6?LU`?xeV!HGC~|-ap9(W|nEA`%I`dA^DBYk63byPk5y9x%cinl`)^5
zXEe-QoiyTTs-b;-^5S5wB;o5UiEK|V<2YB-IPprXo>b6IZ%|{zxJWq^M$wfo&t7Ja
zjg~i$+EWcr(T&nyVZj3<<S*<bFC7_3t$Q*`7^6gEolT}9n;J92fHo8}uZG**q)|q#
z?bhEg9GS*RCnH$qd?)FAG;nx&J@3fpv!uh{gKN_$*@O%qXQE!{p9b|s<W9ZlysKrH
zm>?Y^<jJ!q2VXIyliyo!-^;G8`3Pl<K}H+7ILr4+MwS@gG5C`wea;}X1FK#?W0T%E
zi~%=9YLRmpT+uY4*9f>>Ywmizb!gI^ZQOo?Czp))WL%jBzjYuow04g*M?&N2IhvP)
zxj+jvhR>?EaKOtkT9b3$6(e;aINAKn#i}pTR)*1a3#&t9Ij{pB8b`2Ao;Ru3Fr?vS
zlbP>l!f}l!LsWr{FE1~%k}q`%Z1qOslrU%c<<9e#Cv_@BmJWWUf9UKJ2^ZOjPlro`
zZFSY?j>t&h@mz^N#(*|+;2GAVgBHz|e+_4~ek9&iAHj<<sn;_=29Kc`c-<o^*uJ#V
zBq$^K;1xROuh*IUsou3LVr*q%{5i-9&jr{TX8bE4`6g~RWlK3OvYva|FKt+^(JCIR
zl5Ddrly^Ek(xTPK6PYa^D_0=~dS+<QvUovR@O}<Y65J*z6Iic@9bW=#Hwjf-{CPd{
z-hE*OUg0+W275n?9|7<x%9O7__F(_i<<1ECsYt7!>%8J(q~QD~xTP>E*s@%gkWkjv
ztP?m5XP2<1a%eOh_M^~~Qz0-%gTOFo`dl|Zc}6hh%PX?*SPk@4>{rjh5r6fp30@nK
z^z<|?;+weG1}N9QwIDywg{pwPeB`HMR5|3jOoLnS?q|%C2CWue&flY*^cxIc(H(>0
z%boF|9FBS)Qi1g-G_dcy_f8wU8`oyiYa6xX)2-q+dZ?6bsPeq!4QGmPo)12YXCvUv
zD{0Z0_-RNeQGD@D4n|cQ?5#}GfXyT~6?E}|#lOmoZm#oR8cx0`Q}Xe!!(GX*Grx0>
z*FmoG06Xt(%(je>JnDPzz-s*MdR~oD>K)G#8LK-#71^Ez6j)N`pio-RgT^!l%8b|K
z)b-h!NtyQ7$UKLX!%>O_Hn?W^^O7(;?q#k7@6em^rkUXS5C8Z-de1nd#BSfoO!=fS
z+!2!21E!pm!NMC}<fkt2M1m?sb|v@}uNdcb;TGRKr`)!`0!d<JF$YKa@v}5a{uJ1v
zmAI=s=jS7j@<EePh`vc=<wWE^jXK$G+C<S%@v=RN5%OI;B9CK)@Q9dS!<$G!U!)P!
zLq;2oTpM48MCz3WS}#y};AWWcFf@^e#ts3VUD%a1y~!9F&vSInQ;ZQF7_gokKgm3$
zC))};%jC0DYhd#7sV8VaUXEf!3i3HWPdP%5pjW!cpKO~{ss8zBANg%v)+kK7(Day|
znlsl02<G-I7vR%VFGE9rJT%^3W@7zW8n5Mz@|`K39p2v3p8zUw{uez8HTk>8Yn`&C
zFXUtO5gh3(C|#z94l~aJrn0hjlLkXR3`zza<I$c=UBMD#gmI#y^EktgjAVS|@uK_C
z)LcD!gW-T~%EimZd)$zZG)}{qQitY>BhOV9_>!BT7Wj5-&c>91#rQBRt9;#;Bx&sh
zV?IXZE{1`I7n~P)7Hu-aQYol%!kZz7)`)PJ&SrAX)zqif@gQvYhdeurcUcb?noj^r
zfM<uH<E6ul{v|yF!gKYg6ZiA%BKp94e2K7kGQWF<^B7LNHSo`U(pO&w2XV@1zu=@E
zELzpG8T=xUDt~1c#_Y>Pg?UrM)7>&!a?^nJKNXrEK%pmD$B=>b!>LNu#CKj{-FdeE
z97cZ<<dx94+x4Hs`H=@d%Lw_Aw7bU&S5lQ=LvX}a87OxG<XWJd%?qhjnR_Y~<1}up
zNs2|<HVP?>iqy)R^fS+3m|S4;-8(#6m3Il7XOz>fvCK6kIlWh#03CJ%F_pDFqzxWX
zT?%|DcCf&er+nZke2=-(@5NjAiCYG3X;6c3fK#hKLcc=e<~ha15T=m9;U46v=;Y7{
zuPv70BMu`1ou!5Mq61tsdR5SQ-pZwd1Q~QyX7M$gf^mF9G)sxD@=|~f2Hc_5FBEP5
z;4Y73v8CB2^M=QIpDK%Y<`!JJa=rNJ(W<Dm(u2t24W!Vh^h_f7xmPC7-~xXE{>td=
z9mNT!DvEsE3jUO(Ey*=f@5X?8`ZSXn+tbcaR%LSB<;^$Bap2Awy&Ny&_Dv4`2M^xg
zMRA)uWbz%iu%Q6P#^(<o)$73^;dQ$kuK>k9%eb?oG@dsNHP1ok1w{^dZG17duJ0@b
zv~NQ3psaJDOwFEh%Y#c_>P;N3)xne;g}&jRy0n62oI$lRF|si7N@)(=@G!9htDC9i
zhEna4K}G}D|3-HT9qsYkst^s_^sKE2JE<XJJ9j<Z7a1P0vPREWV{(zrtJLpKt_*wz
z4P}j%<#SNza9*&tg*YS6vg%tILgA%j;0ytxmLm_mB$o9uYC73C%Xq=|p~w3#$4Dm&
zWE%hAP5#LPyw7s&Om5*`xeEQ&1BRgc;x%jtR&%korw+7D_+XWHrQxFbG)yKg+i$GK
z7rmA=>aB^-hOCso4OZwob$w*K4H&+LN3O>h$-5Xj8ZD#<dP%!Y*3XSH8a_muE1@6#
zxpXZ)D?eJu|0G_bZu7tCSu(1#K_Aky_C2CQ=og=iP-l6>RE^y{6!MsI!^L^{I66_j
zJgB@_^wBr@+ypuC5l@B&V_e7S?%Q{BofR~9Gx<I7J7klFo}onkGLLn5(DH`Lk(CZb
zg^p<MPrdP+2doWp;;K6g9W+{vHX5L@_5>}ew<6zYoM_EQ+`e<pQEypROFCOIbtQFH
zFp+=e#K{Xx5Wt_#ilx524o&5=^cvXYZ|WVwm3TnD3frSqzEw9~McBLAV0i-W(vK?!
zK2!!$ZnE!y>7NS=zhamK&P}(6ZtoE|p^uaQ?s~iMXkyH7BmBhAUu1+pviWT1R@gk4
zN=pEe>57x*70NJwHICA1MD(^{NsWRy*5nX#-(aBx2`M7CvLb5WBm7VxU?PZ^yw9{U
zUCl}GyjJH%IQiesEPDG4Xn4K<>hnik72}<Id4Qqvx?c@n0p@2h%RL;+WH{uWVIfTb
z^z50i(zG0b?_NBWbrj%O=qs)>90ZkEJe9PY4@G(MjC5{Ca5U2%LC^|ZiYdUGcKvu(
z#T;cuafIsO&(NPjzEN5jri$?p<pWRrTQt$gI*jw{87<sQY;AdElr(}g<L1OSe4s2D
zJL#3?@7Yr@qOfu_K3e6Ex-MVK-sw4r-wKt8&2yujbh&IXDUri$sSG!YTLWc_WqiMx
z<)x8;gR}SEtKq=QK@l5I*I1${_$FJTF&d&y%MhHwSb-~8*$vZ>!N-@6N^kk<HU|Iz
zKmbWZK~(SIf8}3(*K}`YdvmnnxiS&(edF&@^zXg@i+bMP%f2*vhvqJzojDnHDCK!l
z#>x<y<lUmPO_AClvq6$yKl+7EK~|h=QJO<5G^d27-e_dT@KASPI>uQiqe>eCW68O)
z)qu$EsmJ*)5IA5DKhuzhK9u>37x8e!7%(JrJ-yv`;{|yqM##N1E(~XG|0VCuGYXsW
zht9rv1E={gUA}LTokZK)&a)RUG8`HCJB=)k5fb_bV|msan)IhR@K>XGn$-(m2i~EE
zEptn*$77&A9>xeXakrD#or45?o`<(#T<N4t7x4%v-?49dWVtUoxOWhO@^mhON1mZK
z?~!-m4Y`VzL)3)P8-{1dQOERC?(S)kJUjO)Mu@U>ua?k`u2YVwYg2dN$td$o&mZqD
zj)o86R^^=$qJcAvyXrOCN{ddfXg@keU8FBgPC_?2B5<GddpZX-6c`=&V~z-2U<(C0
ziKj$ot#X(;rrdcP_^;~qYV1wY#z*COlBKzL{LR<jMy8*4IEW|ZM&{mGdGb!ZEztpy
zc^#6+kF%O4@#u$oj+A>V2O4*1kY{f*5|mvBM|c@|W2BWgDZjci^(V-BpJz4b;;FbA
z<M1$ZMA~rYdH%OLmNcnf!+ZKZ2U(yibHwxry)&F2*<@UptZMZ~Onf;sfB51x-lIc1
zY7Cj|i*iH<hGuXPe>;_)dC+J54akWT*-|{o;OJtC8(F5gsoLq7p?^0&OvFh$@5lpO
z2n_!9U4fC?fXg%Q1it~>IJ<K$Oq8WxCWV*@;N>J<>y>AM`1$K(?xyu{;Jf;%P$LAA
z`(YJk|C#F_d*1Yr182cXfv0kX7$L|aTA7;0qLy_Uu+{re1aTar=w_*yikgf=j8TGZ
z^ixRjl<{h)1TT57V`%L@5%ZkzLW`iNMS3JEm$sPTJdA77FpvsOz}u@x0&(Q0cd;K@
z&OB&-(&V~nW#V3*669QiNx784IXDEX9nRu(c!JR6pd6LWwMxM#&`_cSx`0igQ9v~c
z0J|Zp2)qL64JbnupcZRr2OR;eC^=^e&U^*<uL)G0@D&{N0Czx$zb3&)D|k@ydiBPS
zpQX3zgMFn_-VLlVDkvdks6h~XcMVy1F0Wv4@HBK)Zn&XK-OPETcPPj^5t?=><P}%N
zLJKr%SvD7^@&0`OIx8YZ30*x2Jp#vWxbS)Y{LLu23<X+Jc$RJV(@^TEnowDpW4wee
zb@>)Ayh<LSk^Jg5&Kd{`uWnUJB{*Pa6+w#QGwR~|NaJC=SLEjS<Y7MpfR=_K^}g@M
zW3VQ_Y5)vL@YXO)#1qHrDKzw4Bf?u!#Swbr7y1ygLs}wAk!J^uicpY()5s4xMh5b3
z4s<Jk{-hX{HSpkT_~a}Tp7FpwAH_Gjq%u&r@%3^wFktccaOKwkz|-Wbf&I2~K?alS
zk(H4Ln=mPlF9S0|DDxSLBO~H4)aVBvocqNueldC=#-Zin@>LHUtA2Fn%@n`)ksWoI
zflc<0k7quUJa70|T<AgO)Svcn6Q>i@^Zeo@!)C!9KJaj~Zi|o3`4>DxbBeZ?@@`F6
zj_9G#8BR?QSgz;=J!o@SfWhDIzI)o?K4lV@++WGAuIe6L)p-xWj~8Kd=j1(|hbb>l
zm?iFArdv5o1Dtzx3Ge7Ga%;Hi<m74TO&^Y76Q+kw%EQ<Yc!K43^zXl~@$od?K6U?o
zK6*z6%!N|sHDtmNNmd)Z$eK@c%%+q+vS<tWhw%*Z7=HO>J}Mc*!tjiZ0xe9Sziwud
zx3b=j49VNXe07IT(D7oJRHAc_Zpzh8XL%W<^ZejxXdcAQLO(byz7^T>G|^ILd1zLf
zBb_Dg>N|Z>9W5jA`{z1gGH;{L0ZC2^ch#r;*ukL_(L<w9wRGvXmB?e`Pf~Bus<MQj
z^2qy(WI)#1@aK_h|I!M%Y-k;Za}F|XMc(@URNpDv)|Mz&J*GK(x57S+z+La0BN-kv
z>PT=d;px<07r|~9XL~3~j=_J+YnNap{NXdd{k+*S`oU<4BUTCbLlO26Qg8%%{g77W
z9&0nbk5tU|%(vc036QjUIU3ugY*u#SR@qe1Q>4fhrfm^QOiz+e>U{}2Ov3Pq)VR4(
zp;S`tXNu@?#=7NcroJj^UPrwHX%uSN#)Z7XJ`I{z;Psixoa<KUoRw36v$CzD_j&Gq
za8108UQ2k^b7d}`d0$s@Dw1>>E{&L8b{nR^lcvfw6p2cxV81}Cyc5TlCB>1xZeoI(
zN@e!BXXB{|Pg2n-CKaBs!|2c};6^ro#{Ko@*V@p)S6XGNkre#Q5O3NV98-z(9>DMF
zBuY7O%gc>6?$OTP76P|0xOQjl1lhpNgR#7GmaiTko|T@tCJ8k*HE!}t<8zYUpN{~-
z=Z!RuWX@`+Dx~BGrhJwRfrZZCa|@5xBr6-DP;I;<bR5{866rY&b%R&m3>0!{w_iZ*
z&7fhDM)}A;+Sb58+j!)7<F2Qn@Wo~x5{90MuvVDh{hb&98Wv;pcvD3jSY!h~E9}Ah
zB0G%2U^SqglRzGimM||V(Kc|2U%KU)G~_`XhQ>wqW%xSZZ;{XGGd()1UB(D8jP(3j
zc2W&pV;faI;K>MKEQ4d>sY%vo!V|%G(hxHU?FVq<(if>?bo2{k!HFj!JWys{6gv;&
zJFiez3dN&i8TCwzeb^=M&ojqD<MwUli#!cJEprq*p5^;@?Tve8IWgHikIbsi<P7~8
zsQB_cG^?TZGV({+lZCAEKobqcs>6NY;V^omdme=Tty|uahP*%TT!p|0Eb*^|en-hK
zq-*{f_LGy$1B-kz#PE<&N0!mNdE#r5ceo^-Buk^g^R95=nS-9v7%ZKG&c~6N8Zvks
zSyX2AZiYt3$Rgi5DjK>Z4-Z~uUzU8!CeI9wFc1uz`NBEby&2=6o`DF|+9*x_Xbe8Q
z$b<sZ17F3nVdCp&`39GX%sMQ*1J^Q_jUmn$l84E#aORycw0S%EkR5z9@95y0P?e4f
zz6{_RLW#f1H=O`O_Vi9VXp!F(qBPb2%)b%_=kZRA*Id6l&k|QBY8eQL*!bu%N9uTF
zLNgN#3h5-qOm{urMJlM2Kk^dqNhjWu2{=Xvt-3nPeBm-}1v#nS&beQzwd^{yOl~%H
zZGig<z1|*(Q~cq8bI?qBbsJBJvAXTZPmNU1dn84&{4-(VP9CBvSL&#gDO+{U4wpPG
zI@<%$<(oi+e{ML^$O4Qg&0wAPAgnQ^1$URF!X+Ld!a3O!42A@_gj9n``4@he1BT7C
zY%zW#0&Gd7o%W2g3tWIoK=G@XwbXO3hb^z>kuzK!t#}b=4QBI=U=dms?L-_Bvc+8<
zkhOW&gNF}mfh!-^c*_N5U||{@oL4A)*3EV<$rI-a8YHaU{G<|JV<A}T=9zrZ3S79U
z+{!2(MG04xo&wNoP`Op&uTm+iIP%_(uhQ3&Cxv;|p0D(wyI#9MLFCQk!>yS2OUbDq
z*Le=oNZE3`@B+(3U;M&H3hv&$`@KhBY3{zArK^YcGsoh+3cB7i#lvXSfUhmUU7w$)
z!3nN-Pnqcfk{Au6<&_#$jRAT%uAGd7%~*KkDQOAJt;9u3y<{;;S8^8G<gq8gZIrEf
zMc&`MJ>D)e*k7bEeVh%HC^s~$p_RM^3QF1A<4C$2>D_Cn>(SAGhGu$c8a~f>3ebhI
z{fGxG<z&dp*H_KZ?P)#)2j)q%4Ox70{-nLJfK9r1HPnyHKzsJ=;WnS-T2|rw$HD)m
zK-2ii^X7FMaG6)~X6w+MH5oQz_2!`5k4u|zV#%=8A#3vD(bfodY`m~Eb}<0b%dUZM
z{|CLUXle6*(4+GNg46MFzJ+un4cXJ#1n-ZZJ?OCIVTL)5PHr9i&wu*AyXVBO|K@MY
zd-QuN_}zZnz9Q!b559hwt;h3yq+5FA<H|Sxi*oj{xf(&fNfLJ3p&0APMo1OIfZICt
z_*VACDgDvX2461@-ne)B;GGxm9{lR;SJesOTI<rYr<s3{qpd)m=sj!D87=C-<89BE
zx3c_QqxaU`x6`YSK8npjF6|41Pw-&yIe08?g;o`;@PlkKj?^QK-;H={I$W6*5rYLP
zFyL~hAnGHCi93c&WIw#Vu>E?LX;m+Ue|PTQKDc%J=E1$ty&q>|RK@typxTsd=`yex
za+UvpRb6a@-g=W8G&T1|uJnRBe;m1>OOCF_o0)43l6>K3sX3kL-eF?Ap%_Np)xhKR
zKr4Kq*KJ&<Zaw_+L5<~mZ@t?k{PMZTke%Z=TlXL3xOp=jhs?)1c=^ZFiOYj8ADI-J
zG6gm~{`)`vW0&u~c{>xkV~{8lI!7yIUzy_?ySGzs@5WnczaDc+7@zdto44=gxU<+&
zna7XQ%r(KIPyU#VCeyBb9m8LFZQou!z`GdvpN|~IW0j#hY|7i}!`KeP>tJCAQr(&J
z)aR$+i~2sD9R@y+ZDckwS{<5utE-YE`OI+5MZWDr4(OT6Zsv?AqCmgbP?LFrVJLWl
z1<w<^{idFTqkT@F3fJ<{LbB(p)_FN0FlQJ%qVHR4xgTjn`@xB4E@%<M;2Bw@|5NAW
zchofMpZKW@m0NmQT?iyNue|auA{Ia%IAGJ0-&gVaJns=Cr?ycO<In!)^9M37I*r^6
zUQ71+vvm(8fkK`Ns8VcJ*ao)^ZQv`sBad6rf@+oRG{WiC<vHc9@;Ed=Dk2J3SS3Vl
zVSDDzJMk%QIJEI>q>0xwTYMFaQ6#Rk^R1Tn5Je4c3tk@t)^gQtx%Tm&{J<-rdCx`f
z6|{Iq_yjZ(`b;?pVD4+2ZTR&%&nO)YjNt<Uq>?d+C<nMNBMY-T1_!|aQz<LNTi)i6
z$WPhOsAY7Y=oAs71S%$XnO)zV^zu%;0PEQh0i$H$Els^J@ro!j+$bx!X^<7uoC)zY
zCT(~k#gqVHmqz*Y-}<TS<hzWHD;c-l1%ML3(TjFoZcH%pApa>JJad*mBk3&VQ9hO4
zJw*#giq_f7*&crAaM_Kaf=zI3Ug@FQ@JyHqr;O;p&S6yt+Akx-HNG4jrP0W{%a?Zk
z)o_jr7QTj!eUOTPlg(Dp%(-CnXmBW(i!{(0Px#ftm+Y<J&{MwqC|-uh0uKSB^xL?{
z>m9o=AH#5=6(YPRcf-$=IWpdIpch^lmYS||N`x{Fk{5$BWg<__H_v;2j1CzR2(E@z
z+9-Y(*$WOg)15y`Bc2%LV+~e{Iq>a>`aE+*1oS;jLw=I)j2PyFPezQ|P@)4Asvw~f
zjn@-db$u9_Pu_>w71T$_uD(c*GQ%Y|@(i3Ywnvvdf0hRPY4*d3+%bl^{lk#NQN1Ep
zp(mZgfMi&9qPO~FzRRXxmhM6;2Og{UBLR#RUJ6E=JX5<U`FO!<yf#*`FJE_w5CMP%
z#Qhe?uDVATnF>DjA?1v$`0mbmWJ8|qNNjb$oFQeB6CF`K8sNZ_Bjr>M^-CGhulhFc
z<%P!%^})Ok*DZhIkzcf1dZ=;ftmK1+>NNS#u-e@3#`QPLysN1<7iT)0<Qi`Di6Pv~
zA(1}$4viT<k*zeacF6VtL`KcX#Slr;C%*c|@HT976r=PcUtvZUoh=?%2l+t<ofu`k
zk)c%`JgeCd+-e{CWYJ!Ef<WsMa`r=Wd3iN;lD;H;r7gPVaLsdJzKV2O#<pp>R)1SR
zivj(fEE7g9lQz|gcu(euFT;i@r}qmlaOetg@CvNSTRQKi4CIaMg$OHY^9vK3K;F;+
zXD5?rw2(`x8W%-u!z1yjro0&Rl2yZ~g<MX&b?UoSlst6#3Vz>~PW$m1eOLa8^JBk%
z`T@Jd+)!&~`w16OcL;2S@F6FestBvW!!V5+21>UqZw(fgt>y;t38ym!VS^lohqNbA
zuv}EVPx@9qxFv4kxlaBABNYH6BmCkuNR#y9i0|CbS*NFC=~V?QxV%h$8ZfCUusN^z
z;I&-AXTh9{S2SApCHlH?PJZ%igWlU*dmdbbMPui6-52iBC3(ZAF2ul5PUWeR3BER#
zs&yS@pHk<~xF|jZK6C@rj*G>?br1J*?ZSAn`BfU#ocfLecHGMz40vNa9!O)gdP_VS
z8anP#jU;0Uo*GHv{dh>)v%^b@J{3)4U5`l`1J{f{e6^D!<3MA_>t#<E!H`6-sZ1yj
zWrljn4w+Ig_(x5)S1NzK3=O4i*eP3dqny8vd)J4x^!VT-J(}os2pVy6+x&YU=($(<
zH3a53OhZH|qyh0-`oy)V*UPuejkq0J#e4kigEZQw2cLiOSrxaT3%KiDC#fP$W6|Y&
zk@sx7|044*3>%cs;g^rLSy@tJC;4<pA?5olpA>+%iGBF;MuwHl8JXROGvV*pUMAUg
z{=*O3(C}0-Qr-?<(9<D(3SNFOCK(aVZ)Lb8r8;<+YcB9_vOK}w5<ESK{o8;2ZR$vR
z$bm-=En(;4zZ);iMU0`di}a$w3lYWxqIuPb<y2QT(jkB__18q)3G=+TdAsR@_6qaL
zCRhzQK_3mpbzF_-DHQ9cAjG*JZxOoLoNM8auhlt`<3FZXUl1vS2K9r?JK@z}ye$J)
zV@4jyl!ND_-hpjMgA8dr5C+dqvKPr1D9Sjik?0nN3ol$OgV1O^Yc*g(BO8OEC&LIl
z*BWq*r*B6wy6bgIeS^O`B2UjWqK3c_!X!=$XBi*Ms5lS(4l}Gthnh1|xz796iO`G(
zgJIGwyOXyG`a<dAed-W>xIEarDy2hkRK8FC`0?;Kp00Y0GVz-?&O9dz4$SZP&98nP
zUc6m7;mPnp!{bc+eJ2LAA;{}Db+9tL9wWwZgt<rLLnF#CpxbU|ayQ!F$fWx_fp5~W
zVapms@p80$Xq;<0iI<x`G6#taJdM{?*}qBNHo!}Qk(rcTIc$9o2W4A%doGZ@4wof=
zI2}Bp!^X4X{W4se4xPiB8jh*Qcr1|5layH;@=yb(Wu#li@IZBP=McoLTnESMr{J>;
zrOPdj`j|+KPX-NHZp>>w^5-REoOX^><c$Fd4ml?8^g9_)XF*!?Pzt-?H~7Kl#{ygh
z`2J8RZGIyCej44By+ZB1TEKq-(8^w;sUZ{mR=p_XdY1E4UX@dik5KFrJZ`KI#=9CV
zxh5cGV4f8+e9D|rVr68W)i}u+4jKk=5qEvWsYDVl?=_w(*&-YmM1~+L%Zj@`a94vy
z8Irca^?um9UodyC;Va)fcr8_<?B};j16~eBYlRkwyac=V!J{+&b58-*fMX?BK*d~n
zO^$vw)dR&_VY>&z!S@z{W$}wUZeHI{h@lY$r9q@b&tf=O-l#A~sa$7g`J8){l|0aO
z86(bUd_4vLo;xT7M#zh6hS*}nJoVwUaUrX0l&c2pGI>D{ykZJ9Y0)Kl<Dm?z%yX||
zE`@=I8ddWx;853m6o-naC%oLB<#pn<!Y7l;K;#Xco*8Fv-uZD~xy`wXB+r6UDFq+%
zIl#M@Mxo>Thx6%q<Kd@Q<9d&=p&__-mCPwZ<UHlK<QkJHgbCqg;(<|_r&yv)!j|A2
z10{_uMFNNNN!}TEm&U55!*~ylV}#f^OoOW-K(g-j5Lo8&+`N^>Ep<x0!Z(f4i_nSe
z*ddpEF)q-!{0=X^iW^+{8J`+z6KD<Vm@G>+$lrOsM*8CHSsN`wsbH@L9S&9Eu+H@=
zt8<6HwkJ+Y5x3XZ`(1+4ZYYn2lZfmFl`JJT>^LrRjXN6AC1emS$ktih`Wk$bCB3I-
z@Zs|P_v4RKk0MvY*vdQObBwCcBRpg5;W~_J5aVUim=r(t6#o(@g+#jvc*qdjI%4Wr
z0O%~BrCz_@wsN4U21|p>$Y7x31$gRtCrbyMI->{A{f`iBzz}2|fnWL25Jn02I^}i7
zSG;!~+%@`Sa6E?SR9uoTxf0L0I}j&7@zBlCWR0ibj<=z9>MQSsl_Oc29v*ZOZqRxw
zL$}Iy9QQxjW(YG<J5NY`hi`tsgG=t>g(Q;WwSj}j<V+`{-!4-2%NSp&50x!2&@}jE
zP%UMMft7l1GHQ*tucL!f55ZBsk+bBJVaw7Y&&ULM9C)J#y)%RXM|99qNZ$Zh@s+nr
z@WOlNzMH)0G_t6t2WsOEgyP9Vk!&hM-W9rw1ae0;uOM%2?Dl#+&V8OVqBpMLpuBWi
zZsncFMrn^cNS@@!eVn98kxV3=)N}Fw^Y78>1(zC&$UlaJ(g<5T6>$GPoJ@feheFSF
zH||1|TvKMOSo1;5Z>TjE!Kqs^3eh{RL9R6o13{MN+vFjAnGlY#=;eW+o>3q*E&`*%
zkawP`Xs}g2cy@d}y+R+*(=f@0l2WlKWN}n9X_jK4h~#~kydp|F__=7jmK6b#wgy+u
z@)b0QLh*pZhL5=3>^?q~JQ^qY^s9)8D*_xGBbfzNg9dcv<N2{P(a{PMp<Q8zmj__6
z?DL*NlNSLW<)UItWA{?!)~H0$oo15*JU|O$`RHV~Otiu$c$H?sy*;Mz#5@+fc_Td?
zJ#uc}E&C};^7fscsEPK5z<y679VN{(LJ^c7f&cK~<YioyoOq7rV=!b0C}XzuPJ}P=
zG*r^YHHsZRaI>+NaaCvD97YHj<e<s~ofFUsT>0QNg@~SHiu{?lE8v>>KI;2<daGdB
zp4xC#y)RM3FV6I!=bJQ?;SAMeYw~QfPl-~%@LrRyFFzb(8}2m_l2-lrHlB!jooYl)
zS&esh_$D|(ong0V!`=Pz!er8uPUr(qUX-^6!cuPgXxORPVS*_`#g}n;OFsNCONUbq
zrBvp3;stx-#!M8|IGLkhp38%WUsMJ(L};Y!@4b63OD?y`shJnRO@Dl(G2rs&)maTd
z<S)<q{S!3b8f&=X2ObbOazOGZPY+)8EQep>te}40yczz7SLAx}6_3fO&l^;BBTt4E
z9z1wZ4+;98pFcYI=l}PAx3^snY3RbML?0RUYU3D1ZqC8e=UH76-YB!i7S8CX9-hZ@
zrNuWfQb0f9B%X~}IH_GT3H$I$JYNUAYKfB}jML}QbDM#peqN7f=J+H$2pudJSJ%GH
z@C;q`<(H3IjwhL4;rZuZd_H*u5Akn>53r~lCxaUNfuYQ!8?Mw-@J4uE`X)_{tk9al
zj&9^hP$rs$Y|-C`p@FUC<%<{PU&?Ff?ezS4JdKYt2j{EIY5FWHrS7y&n6F@&>qj4b
zSbB6f+m!L`7$t^wkB%5D^jmav&b&a_KaH}q(N{T+GJi;YV)Q0+#N5d=4<Y(JJsyt-
zBZt?uG>CzzULTrU6FF0sW5Yx#ry)mLPa~b%!l>7(x@|g4XyfG!L#HDUnJ~^<T@hH7
zObBnGdNF#l9+y0mkHc`fosUynRjtw<e)J3jM>cJ%5|?+C+~)*BjW<_67Vu(>+98d)
zQsX0Z@~%dH^nz;SBY#>$!^9U|iafptbCu+`IsiYG*H1nFUR$$8EM?GVk48TQW9JD`
z_Ll!AVlT4q;tj=`!n;Laf{yVjB!WZK^)xgy3CcA%EM*g|nkIrUAE8Pmj<D)Q5&-Y=
zQ$@plEMT5M`FQ4#Pw_}xm8u>M@7qvq22X_W*=6?KfX{|g6(LwD51beX71*2!hhWYG
z*oE%_0`HX{9zu-=7s00~69-=KEq>DE@ZmjiS30y)M(@<O!X5x_cKON$ZWLGXb-fuF
zdDWm4DZ#5haOBlL*Xw}Wl!*c$)%G{Inu@E@o;fp_jG1f0F*Ow9rk9-JUyU{T-HZ!c
zBdFnh^L9KQr<nzhjyWg@N`-OY%y2-Dlek3TWpXmw@Oa3>a1KQ(;Jv*oygyIn(t~}T
zv>oqHZ(T1Hf9pOC9%v%2#>eoUG8iJM3;~#T0{K{R%a`OMPAhjxIK80oJPm326ja;c
zka&Zk9Li+=MwKf$J4<5~Jh}%6I1B+3NAvDxCN;LF96)vDCoQ9?-Xm~wSRL?9JP;a6
zW#|{3b7oNJ*)a;>E5KJAli{ptIm^Bo=9Zu-9Pvr8lvV?4^E4}2EUUG%?3>AF=9kQn
zk;WHI+Ww~zVcbm*G0P^ykA7J>`7@M+!{B-|dq2G0&8H~%;RU*YL(pm<!G(O~ek2nu
zrwoiWacw}Br;Q<AHEf<lRrh&*O7>NDamoGF<V6m_aLz@&H0Vvn$dfXVDZ_v2*-`MP
zuo-sGk7F2wj&g2;opN8xgl#Rc4>V4L9=d0)204gA%dV*L<e8HY_}!xda*Nn0R&_w$
zDWT3tmr)+Y;7q*|-+M-mB0JEe69<zwFzOMUz<-`=#;0<4hyNzt)?lV%5=S{p@4&5m
zkmF4@yi@1UMLn%Ck#}Ih!+ZIBPk-iJhj;S4v>|Vu+cdDquwj(Tctff~Q<oS^8gu#H
zyft6WHu1R~p6K_(Oo}vtT7yUS;Zp*^BMq`^^@}{varCY-RW}EZmtZGf-x#U^le4_0
z9+Q!|M?rWP;yTY#a4_*<jGm#dxIv=N+dc&jSZ%yhR>kcP9LPcP=hcAol0Q*a2Y%^~
z*6o1qCVPo19R^o!axaK0uaoek<t``Sg#gK0p3>$h9^2WqJcX_5OxZLR(&hPFf<O7p
zQv{eeJy-7XCi9yLtH1LD!+E;*@U_c7_kKke@8`ztoDhYs5GUa}FJR7#@cYm9pD)37
z*0@uan5*$vgb_d4RJlHQ@I|eC&u-tkT}4lcQxFsd9Nx%sMHNH?gK4aK6;4kDKJpp1
zPJ>8+Bl5!489stB)Y+UYJ%+%r{0ux5Odu>^$zL86*b3nEAPjk^T)14}&ASz(1A`Uq
zdgmGOEXM^!koZd>D5HES>|Oa%CLbH>21m_zXjX-QwqS#m{H6I`2r5LmSVWm2wh9z2
z*gWOsP??s|=3FJ6%B|}Aw6{;O+dBJqzx{2$o1#%eYqQ*|Z~|8uTHaivF`P6^di$2L
ze*DQSIaHAxH3Cyk9umso&CGu1E}wDVj6bJwhTFBkb^Q9Tf75hJCK#s5d$}oNN?1M`
zMU8Xu*@$7IBtuBwWXYlBfp|xe-N=FFi|3-dmliMWHGpQW*K>J#-y`Yq&+(FS4Ze3X
z?yV8;9GFZ>^*tamzsjuvBUAO-Bt9jIRz9d^4Er~~`AvI{mWGp2!#bAtcK%J4pW@9|
zDJ%GTzZy-;$r)FlefC)mzJLGy?^A!GoUg_6l)UhcTNoadvEZ+XrL3*&v4uCI$3ga9
z$;IKpdznXK=th~ORY%aE27AtL<lRNy@tTxhp~Ek-0o!};eNg=0&W2_Ej!e8^Rp`O1
zUdN<efsxy~$+?y5=kc%z<gX2JCt;>u1n=VS8#suTfH#HfoV27pOe=n?o}vMMlWp@9
zJm<B*i`)6|pkca)j~>Uk%Ln-~M}%Ad!NUg!Uw!qs2IalXDSA7eAafT?gwz?rPo60-
zz7Aa>MD`e<WLkc=Ze>3gvY0s{#Wndhr0B{fuH)eL(3N4F&;1lX@_~olCOoh;m^=wS
zs(39!G=?$o!5$sFBy=6R;`PYawx>`UdGjV1Sn@&xe8)crey*_SMYh|}CtuHaxzJsH
z(m&6<2=}}jhTHDlyWcvm6L1zD@Dx7F1mS1dQvb=5Zx0?m_$(`kGAt9>GOuiT3f237
zsSfS7^YROCs~_U+z(*m{5x98Gc%)~CMjCvfZ9?i2_(8Zd*Z8-sAs0ztIPJ#Q<8}1K
z<x9hvp;hJSP@9nQ$Uo)lks88C_Utm=I&>(_qDL;H6B$PICW#YDJ5a*+c1{x?AK<Hu
zK}xjNk=(Syx#;cOcqv0Y)WHEhI(j<I%L{`iA#ciMvbq?PcLC=|JBFC{7g+)A!0(Iv
zq7nWkqbY;B6h8I{qG*pM(@;n?u;ABBU&~WIiw~|FlAj!3539VdKmVb7-2XodEp$NS
z+(xQ`FoYf}+N)UqCB!HU%C*Wq*mLh053QRlarC%wvc$|8Azg#j$`IxV;5AOJ^}rC@
zec{S0$IA0sV5pRcMYuGaj+MTQ$UFB{Yf0OXJf#^%g<0TZny}7uxBK<B$AYoF!SGtS
zaP!Q0#R=RI;EA*Fnxq6v<(fF+wVZhdH%EULJ~=mk=fH1CT(2W<=lxc$q*ihBx^b3)
zLt#<8^266Pv{u}crZD7;CXU1tC?`u+RjSET&mf8}JkSW&SftPr2TjiMeGi#yfG!iq
zGqiD(e}Qf2Nu_g8)VEaD?S0Qt5*joU0pU%)y8lUVYR}3}iR1=seBbbfpZq!umbhru
zJhtcG%R~ObB|r6gfTt|v(-=O*SK5h_fIhD9jZp*-vIiywLmn9-x8J<o;hHydwhzJ>
zJToznj2<Q4G#)VmVyv}c2wyZ7@}}fp#0!G_vL+SXm!P?8&?q3-u$`lzLmQZJWuKi7
zz2PMe1D`U#NCQlExV)7aXYse?&ovzJi^8Nx^>&pwViX!s;E9{~iyn9*^mkC{4YWEZ
z$;;yC<qsVq2xQlvGJ&@42lnLBh=o^K3S6=WKI00JZoe^Jjc;RA9iGnQO{vfutxam>
zy=&u~JR0B{7-*<<5Z9}dv-ybRd1z@>)s-y86)*o=Iyg!@H1Y!N8fP^K>8=Ai1S<HZ
zU2w>nPR%N_;TPQz=9?i1p|Yy+3mkGV_u)-$@C_`XG!kLs=l5uq_sS<OnGmi<uJfRE
z#Fw(kPjzVaZ?WMQ-GmUpk4z;^3^a9up(gL=k!vvN`FcUP+H!xCZR8CDPw&4_HUt$S
zVBy&?@Q0Ef<oOV_MhQ`P86KB6bXj7w<C9L2I%0^~JduiFe$bJohg%Q$OBI|}16`dP
zG*S&b&OyEhAHR#9uvyn&y&j5Y9`4?3c}la=HBFw=72l7qo(>tV+CT&%8?3q7Zu<1K
zYbA=z$zzxI?#h!_PgnAu_aE<HZ;@Vq^!okxVg2OWUDQAEgzVn@z{L*{S+tO+HWVSy
zU|Df8N(%uo+F0UDUUaXg*WPU?c`M`66s@IO)`oxlU;kBMI!!Mo-L>wQfsnit&Q&NF
zCa>fxA4+lG>eqwrI^~?bW4%-OJTyLY|83mR+}NyYjNO$hu;kJ7$)lTn<?Pv#hPvR~
zzZR$%u*CJYF}C|;kZqx7vSr<~dB1BMf&-dqR94`A9lD_%g)qHQ_)~rwinGv$JM?AT
zym~vMq%^8}kwdF4d8{BpZ4>$^zFN`A7Y;$K=~?erqIl)O`(d*v3P(DkRPPX4pmpeK
z4BPYBmRj#)di@Lm-?%aPSQ*MEndEqUID4JkOk;qLdU$AUrNUC0_+_}jWJ8;lExa|_
zHty2s`WDNgNof{h8wT>VSGzUp?N^BgcjL)9wNVkb4J3SoJXcZSc;(9U61{=TVdOO$
zY4F<4;%juR1ma1sIz$iY_rL$QdWy1sA@IjDBvZ0U&K_kFsZ}kMG<?XTxk+{f1xt?|
zUmrc5ZQ#+(1Y2b{?*zJkl~1rMlO>3H!T<WZf9oC!yb@o2`S{>J|KsE2WmS#dVPqib
z4OQJqKISuM$VR3Z;Hf*otx7L>kSF0$_E>n5`43jdq^RYi=N<MylfY;%J=a$+HMUtb
z70=V-c%4k3HI!tBT6`6E@*PDuX=8wee#(Xa&eKCy2N^EY16AHNESsyw6So-#vt9--
z&m3UuPp;6oxnWQ!GukSzj}3OtM|1%L>7$Q6YT0?19>qKIINp%&Quc=re^<Q79NzGZ
zypdHP$C(eJ&DDf4s86380{gl+*aL?SvVq_jm@_Ab98Eo-6Qf4LpI{u`7@FX1Q0DL?
zJ@j_s5{c1DcB~-c@wv?2D`?Uqd4KTHU$uVuiZP?#Dw&DL2+z!u%ukUgI>5!50Y+vS
zSfxka%Y(5%t_H_tq{=JJbjqV%m!7%jrF-%utIUd%$?G!--TW^eG->eB9GZ))DEThl
zQ}yGs&p&Bh{?)I38N)q%XF^4W@Scp+^G3etpfZtO&VRf?p2)u*klY(?>fET(2#!3v
z@uFrPQgv$L!<im4y!hVZ53(}u>v(7$q^t~Yo~-bJ*DCm0&YTQX?8wo`bRL&B=kdnC
zK!8IHzQiYQ@T}(~>AZV&G<s<PSD<TT&Eq~Vy*HQ)<z!cRBvXf$^9N=k28YNlFIlcj
ziNsa^0EIUknt^iyI13MV?OZBe=%bYmo8!4jXDE4SSB~s+5?J|f(6+tW@iciB52%3r
z{ji?(jdZWYalh-Ze?RG;-2Hq{$bKfS7v?~^Ux)kcow6Z{-8$RIq)?E<gepM{%`-j9
z&aBT_p}3P(E|t2y`NXO55m*{2aK)*W9~i3vb3Y9VW)WlpnhGDd44vgtUA_G>EU;FP
z_gLl0A+Wv|R+$2q3-0yzV6AdD-4=gcmlnC%#oK$oXwr90FbZVBD?{Ovd|{Q`OUJOe
ze@)_Np}fm1&!FWm`SoZol1NYGF~A7E;Ju1a#gqmUorv!O3!eqr_3JcnUt9-R@m<gG
z1sq?;RXN?O0OzUDscXHa^zfHVoRNh*Z;rw7+Qs!6lK9F9@~sLRe=XW*n874lOQ|k2
zUeVQ)9}YVj;BWAvpciZvb9yOL{?617e%&Q8aVbR&-eC+vJSv^?eCW;aLCcX_WwXkJ
zA*Qjm?3HX+CgX=rvhA2U!!0kM&6ID37&WH#@TFnV(Ab2@1i*gIBJshYu+XY=a-v`v
zK@^d4!Re|7CiqZvh;0>yAV=xRf@|KV!|mG|o#T<GKJ2COM0briLq`32dhKb41T-*u
zq2|j_?!_^A8HQ7qWAvd=C%4MuW$3JzI?8aA^9WzAHz9=Aq;cM^=HWb8&vIV;LpMjB
zv|yDnIHae15q#XEfrH_#q1NEEK~I@1@xIn!mJKI4!6!(aXKn;6t|Xq}(xVs`a6QgG
z7dEpxNl#x9;X;PCVl+<WBQ&^@xhIGAASsRtxYg_C(QKN2ChPJZYXoQK<g@Ts1~uS<
zGajp7La7_ts57e$z(GFbyZAshN8=_9JcXWfy=drJbwf7(LmS3?;RFvI2FAw<g$<89
zyOJ*zUyYFjzO5NL=d8RBPT{)jjv|K)kgHkcHCU4aSNV$$caY^an1wWiDK(WZM~&OS
zkj7gdLS)pLl<jP|m`CJFCf%<xwVp$|WlpNC?78S)0RgrjAhts28X~N_(s=L=?xme~
z3NmT>E00}1yQF=#<U_pjX~m6#G>4-lvwSEK21BKN751+iDX|zL-S9N@>jy)t%fgo!
zA?Ton#}`!~{eAT#lmD~uL*@Jdi1Bp4643R&0=RDW&Wqe_ghSXi^vS4|38BGCd`hOq
zLn?@S?<pacI0ih6CxlQq_p|7E_lqo-dl4c1U;oels#3Q+@LCl6+Zm&!xLMkSky0h_
zta)MO?W`eCF(?)8ITkhpgcrp17a3PKw}GM&Pa^=o<?$%M;5W>H>Tp3hU9Uqt#>zgA
zibtL=dX_f1_gq*OH{5n1FfW4(x3~Q$RJ4g7*LY~ffYMlKzEz~WY@WP4RRDW94bD*v
zHBlx9u;9@~J7;w4%=J81>4%nSw}OL;%(6#AOBs0;?5(jv%a`}<wdCOo%S;nAp6F28
z<l3quaBf%$5<V4AadN<4XoH%w-in=OcV3{pmya!*;cC)y%N2!Zb;q+`WzuWBBxIrQ
zL!apxy~rw<M+blW^wZ)D9z%*kCMRI5;JPxo_yFh7(>x0_AgAW#ETabgf^=kw%xKib
zGXd@*J9^sY+&q=HvU4a;k;YdL7j0#apLRX<b<+R#5C4$!-Gldk@%}W_iC<4Z=p~5n
zevF2}eU-&r7mcLVN<2OqezN#ZK9cA87uhOXd1eeez#>yN7{j;D4+zd?@%sqc(<h-*
z6kL0F@=Q7gk*)4|eBexeuBW{1)#MsBXQ6AxP2&YPJ@}7rzB~B0fBpC7DPM*~(*%a{
zzip3)<Zahdo{0PR?stN&l`G`$^n~n%*M?-1C!>W7#E*a!m1hm?PL2>qNSqF9s58lT
zdpfUzXM^X7j|Y0ePb0JSHpCNTiLBKcaCrZNUmSe+{KLw=6*LdNc#s`JXN&mn()+hs
zi4MWn&TESCayQ<Qc5L#U8}$1&t7wFi$N}TVBxGK`HD{!9V-j_q!G*zBV~BxgPDIL|
z8P|Q6)}rOUlJ6lY_vMSJKZg9mrR>BzGWPac!6EQ`b4NV&)?O%|rXExV)Wy(Z=^XP(
z)R)$Y#7CPNzk%8vqeD|G0R?bZ9dpmSV5Qu%XWzz?{ay6;)7A&|;NJav<tzSLCM}Sk
zF?41_KXV)Y@P|JHZdL-tP~%Oo*Ns(Nm#Nceb}I%Oy#(JGvKj+~-jAjspi_6-i;qVo
zB&$~<cq>ciMunF2=KXu`99-Ou5t3ofCz=0a3wg#Qj~CiM2~Q0dEnR_LXj?hT4;^89
z|LUrKEjJIguyo<NclHeKy$%k%&Agy+K}Yf$qO=~V<zuLg?hNwnbo&Y~7|K#_mAVNp
z_$VW+Ahs98U18UOZZsaw<wZ%1JkuFaH+}rFUP^Q>ZfU0jKiM^}$>bGW?}U}t{&`(1
z!>bfyM0xge0yFw&g@W6Fwy=Atyn8)6i1*05bu9@Ck20hfFiAR0;@&r*3Dd+^IWVBg
zCvn#E5-~!^ZB^1BV#0nNt_d=Qsxnb>!KMlbOfaNnDD|9Y@K`Vqc1K*#bL<C=Rkvr~
zBXU*dewp_ZxX!yC>-m<)SiDnKUrHr}qQbxUg$_#iIu7oY$zKC_xhP64%sloxao2tB
zWU!@mC1oMyIeJ$y<-80ZDH#?jkpMir1@RkJdDi_ref694J6yM?VCXRj{dEn87#(=D
zT#=MEdQhTz+~o~?9MUEonW}<?cMg@@b9wZ7;kBhvp1v1<$yxGLb^*MlDEc5j$AMKD
zAvZ}c9b*R^_m!)lMCRaN2ojxn3_P=&qz0MG{^re0dW-_6VA&pE%rUufucOM8Ft{cT
zSH3cip=ke;F}lU6yhq89nPeWBQAfsDvay^-w0s34F8k-FM_DqOhBNpheyUWxiz(2-
zm8`Pyy$EG$qY$_vUDlfuE^7FTrm}_ZagDb=;|+S|=w31L_)w-MOP+_P8e)0fd}@`D
z580s!&qy{83tiOL8<#V|Gb>w@$K)a_xK#cEJ2%aJ{ghInpU^k1@gw*1)Eo3Ax5zWb
zi3SGUItexgmunMiEdk%rIYu6OPs1Jeed3_wttj{_sjsi<idqzMW)%$@1lg8Po*jY+
z&8}aGX_fSN%dkrFeFSq&X3-C>c%MJItBr8t#x$vn7|u}A^p4Sv*D0y`?pm%5oiZQ|
ztJZ^{E`<+x6yV8cj5=pwQ4@_zXU{fzSJ!Rv<ZT`ZgJN~~@ZUZ>8gqL5IICdDDuG&Z
zlA?RC(J|%8d&bW3HA5pl8!&bH+Br}7ruU$>AvVc}Ca40ZD5PAiIyZ7S^=HbY1EbNs
zbvYm4gx@nAhtS~p36D(bZfK!y!mT<(hAE@J3G2MwzJ5X%@wc=3AHpTpW)yAjKv}eP
z7QgxBJRXzWZ*f-^-j5F0o)oN~3Cb?K!f;*98*g{?=uPg)Rpi7!*_)2ff;s75|6OmJ
z+0Xe&IaYXewVwR&`6rE#MciNu<2g1sGFXH(2ZTbdke8TwnKR*A6`io^sR;!5;e&_m
zCD^OLo*=*c=);bi_oGG;9xh&!RP~m@F!o&rVlvo=W^JdU=iLx<dOYIju{ht&y7lhO
z5P>sJ$>@Qn@pggoQ4SRpu{+y2>D&tys=1i=eY*F9CM&?}hr$*IKEk4zJP2MVa;Ab9
zZ>2a64`=LFkDtra##UYOnKIE?ngidq(F^MyAyyOQS^1I1dR(Pxv?FDU4sDoHR^_26
zdN6iUk)?xT6-MIt8mrzld^HLk6W;QazQ$nA@ZSk}0I>Y^NYB!{*ago4PiXlA)7c2a
zKDEfk4p!pUaD->@sgesmXl367<MN()=cu8UxYEO^^JP+o_WBdA$}-Ol-w9Sb1c}P3
z#zOh}Chy6y-lXT1feqi%EqUqHnx|65oH(7^6!|nJel1Gx&Yk(v=#%V&pf32Z)O-Z*
zSq+a5fALX2Quhyk_xGLKax;5zSmmQ<Nf8^?L7)6LxYmH-mu0wmN}qlDMZcx-@Bi^%
zRsQFhH=rKaDUw3<!MVe14t1Dsz`U7dxPSL|e|zx32Ol1M_`wIID_8#4@uuhze)7r3
z=_Nj?hrm39en%vA^UE*4sDh5?w)C|Lnx({d?wYG}t#B!MbyB0C(Xlfo&kPTyZ+&#d
zP+)~I<u!u~zx3L9lzc8;@AqC-)Oh#qJMVT!;}1Xls7s1{%+HY1txS^L=XJk1gGQkc
zKI#1qt%F2@=iTKqagKtsA=+z^gY)FAj8;LZ-DJ#x*m}f^$GlS?e)M7Gz&xWbqNhGj
zFZ$Cze$thDpMUX1b;x($J_$`S|HZ19z_!sCSZCS3Uj6y(^TdnZQQyr2@`1v4-+d=v
zrM(|MX3kW2#qcA0jBN7u-P3L7;7wN1*D!+>9Qx@`XW0nF6<*I*c3*@>JTB@1JZ?wl
zy_@v!fShyXDn`}EAAj7{ZsL$D4KVt-hBv+9+Iyk&$UV<qGRuE(bi^BScAk}B*%RkQ
z_Rf$7MbGjvKs;sa-n@Ccx`$r3SH=w2L`EVPnn5dm^u*~J29%Xb<ow&n8Tp#8O&ey6
z9#6T*x79KZ`CZNP^>yZUAVLdX^bV<I!$7#hgMntUH{D<-XUp{;KFI2_#5Mnr_w;e*
zkeLVMQ2)$j!wdCl=_K$_$fq6FGvqx_1wj{`1H+UvS7e5U4B@uJ6dfvPXuDN?aB0Vj
z!5R6m?LcAX+Jy4eKEQ(!awYDbdKZ&Fy>sZBFr*w!0LI5*b>ik?O<+xw6hCnoJt8LL
znXoIv+`jgk$bnhm)ZqL`*ue;J;N1)Tbq&VP3_Bxa!*786%)|>33Lf!Rud^Qu3c+fH
z5Yil&<v<a{>>d<pU@=@eb3(Y;O*}kdtT00Iyox5US7Xz{ft|p}(c8O>05C-4(~z_H
z^@K!O8W*Fiz>0((99O6Ss5w}C5O1X~o`FI67vctYgFbl#w<MTrzx&s@1+Orb!TT+^
z#-3F+=S4rX0a3nHEKxmZRZmXOxsFn>`HVO!hxaRQIW3-4DBxI#`*^;3q`cl-sfC;i
z(cue^TrVp<PY13G`6cY~LDM`Vuiy=cS9mO8EG871IPl83kes(`bnuKKtP;#wx?MSQ
z)3NW!%tgw6mZOTXvNaRzIs`-(qzl~T!KuRa3=RprEtA_8FNz#5q~qypz8h}EC-3bn
z+To}O9+~2qVdTsXkGk&C%^yI#2$wz@brHjWn>dZGJ#qS7m?$TCv^>EjiQ97var}>4
zJ~7DgMVP@DmzU+`Yl%6dnxSbv!D$pU7|p%%&M2E5Lp$6O+|(`0PBn@q#$L}iIyArx
z2S&r=yk``VH}&#`Nqj=eV#0C9^n+`86MAU;Z8W7^RmN4M%AyR>3E>MxD|C`vaB#L`
zrscwn0ADj@l*sSaH#SJi^2*?$k<bYFvMRh8=<CqauYtYfKk&&AUZWTM@S*`8F1(K}
z7|%h{Cr|QGen%#x7m9b%bXYBUnp1H0;;r)f#*HUqm<LDt&JZ5KpN3n$0}l@7kX&Y^
z48sv<!+T$Um7QDZGWq6!!^ugQo%)An8e6n4Ql8{_v}QYR8pubF9ya}rC-HoQj(YcK
zp(6ta^t9!?4xBj==tW+2#5C-Tgl-0@&b}VD@>lrsZlhlE8Qos}9UhTEjlCFaZyvm$
zuh6sYK)faM(Z2D-oHJGwvGfV#;BKE0^+#KwF7g^Ct9a1Jg*u8JXp3jb4Nvk^nDZuM
zDT;KByurtX)`QURto0(V^KLwYv*h^Y%h|XMA9Sqn&II4*84{Zz8b$%w_%*q%{M}7G
z^FH)RSxDLFz~rTu)wKN)y<y<=RgA!t#~~^hxyJ7mI)Og-j1G}1Tgpd1no~|i5WM-T
z;UHra%Kxlf3oW*I>P5vXa4P$OyX1c5>nu7j0r8iGCj85ih~3pOO}3-ZD?;pp>(ecc
zYeGSk37xTFjRDKA9+3#E!l@YRm}Q%z)J{^NSfJd}TzPtuRspEI_pE$Kdb_F$wMui3
zpkPL1`y4Cm`pJTxvGbV=B94P#8K2||?l#)iVFzc=RZejzJe7+DTjh}aS9vt@N2$Q#
zRk-Au>+gmA`_~vCuL_qq+dFit<sO<*92!RXuQIj~Oup?Yh9(*YljGJr4o1~dHWZFs
z8P`e}c&W^nRzBoxKHTCY6!uY7N&8phO(oK3i!}U1gJenra9a(ZgJ)h3opNqZDre6q
z|E`|d%8Gwo(i=F+B%nn&SY_05Ca$tHy?g|D<hhSic}KCK5Bh14R78!MhEt;~&Y~T~
zq1R{YcQo!Qp1hY==i>xC*c2vs&l0zDAX0X4c*kH`zT`c5Ud9hvJA5m~N4Us7C1G`f
zeJ6}<A05XlkngCVi?P@a!Gu4E_aZ|Zyjj=tL9QE-J*$B_-z7@MFa=|RH{eEKD1*A9
zfrG<IhDI=t$HIK3Pe1vj^G~b-g760)eo%ftOs~!ZAHE{jRq(;(^Uoi2hym{JWb9rq
zU+<82Lq~w8=952s+<6b5q_=O*$$w^rns<-0>PD~K_&F~^Kd&Bo2~#h!S4!qp*e}JB
z?8!4jPvO&Xl;=e}G8*BTcf#A4m2poq56EA3*#zUuG(PHm_v1(%uNxij^Bm1>v}Rt|
z+tCZ}$D3d;6Z6QdTH%>e4swErXjlU&@C2R=vOtE&A3X!6##kVu<nJW$@r^8LWNTb)
zFyG17ihYL%oxS@!9uB<t{lEQtJwM;(+a$_yQ|~@Tj{RP&8o3jXgnD(Fjur!gzVrPb
z{Czw05>Id4(N!WzYjdxc$#(^EA<sXA=jPD;^KbvTxY;1CMo7x1%=R{+!_WZ_<njJ5
zKIk4U^lEqP4Smq4-mlFgz4F<eL!)w+e7nwZHL%Dbd9)XZ?+MXcJQ)ld$CM)<muppz
zOh*>Fxn1abWY{W;i5Y*_ufNecjQ_;QH)Y5MuLF4CFhEb=In8djHr2DzC?1i_Awd?p
zyLvSUj|Y%`QJvG6N9IyXvw7XL6SGq<1J1`ZV-%(>(8${u@k+c`_rh0Rcf5R@)oWy1
zotoouWP#T_##X$USL|;d7`_Hwqa^t$Dm_QGsxQ$h_|@3a`AEG+O?(=C>Rn(&Rs=Zr
z`F6eNjkY>L%Luv3^ATlEr0Wr3ktuUApkA~y@?fNi>zLu#lhlvS%}J{{Y399cF>D@Z
zcVo2Cz6d&SXawivR_|C}{EVNnelOnQ)83;Wy)^TWMk!Hxou~!qd5#5K;aasl%XJ%<
z5KNq2CtV4=V1Xm<c6g^@M=;0pr-mR5k&OEbY>LRIrYVF^KK{5&rbq+~SJ5bu9Rou<
z7$VuilZB@d?{?y>=-gW}I^P;#d2Z!M8t;@{<!tXMxDph33|p`w2MG|w-@>b_d9d=|
zf8RJsvuIEOBV;)UAOU8;>77Wge8$o#U^rPVU=kquDx(7?@TpN|Ty$C8fuI+45dkEa
zd^p385vuvl1%Z2YPz(wP7ir+&ObFpFy<ryk=Vy*Q7sT!JoZ+PsiPJLXzImhDR^G-f
z&^-&?yQDdnHlb9OU-6lR9<GZ=aM7T&p@?F{qvO2OU^tXvEd8<a!2@-A1F5)*8W?)*
zN-8bm$s|-Xp}@s&UOD%t!s5FC06+jqL_t(L3pwXOC)w&zIw(Sli=u~1WnZODZ%BhX
z25}m`7=s#c3S1c&%szDIQ}E?o5YmIvLu{`^=Ih8sJznxppwR4`s+Zf3mgS>&RcLsO
z=E<MYBOV#3TuYaHPqP_VhkRm4z#ANOgO}!ZJ|z#gv*-8HK+nWgUJ~)!fajeFu0gc*
z=Smt>%C3gtP^<i7_%IIPk)O~knWtd>$kNn#g&Tg%JCVZs(5SK!DRMsDT>R*a$LE>&
z%T3J~hl>s9?mU|9!TVL$%&pOI;hzSvbwBmL^(XklM^0#hrjA8sWfwY>xXD4~UA)}G
zM|`l*jC|R|Opp5Bz56wmHTueD*)v1!n;43|aYA1)n!~8_Tkk$mD>L1YF|g|Xj~vxo
z5!u2MdFX7&n?xEeG!KJU>S-pyqQUbR1IooC!3Z&IwRF^~4?g;a_Ud>&NQtAPFg<C7
z#+w{lBg$w%kEP2!52I_S!GXL_Jxix;GZszm1-st3<U=;vqc6`>_Q0=9ga(MtuXAZK
zLfN8h(|~Jm@raIAZZe8qxlGy4k_RJQ=Mi6qM&3yiq?5<w+x#*#;)SgX$e;MiEUq-w
z^SN-(%fd^<_)%Bj=uP1TC-klS25+z$A$b8LJsEWdvy$X_j^<HV$v5Z1OT5x{d%yC?
zZ6Qm?$Pj063!Uh!q*KO=^T4uvH#^)l_r!&f2(uLfJm|g|mLs#4E?dEut~b&YR^=*k
zKnGcQH@YGjB^(CX3ovSaOc}**K3loPtImPAu;J^DU9f^qpm^R(-!$Tgusw4r94Yjw
zjPoeJuSG&I;0e6zsdxp~BvndzuJS1AMLsap!bUMv074jP^=9iW3B$kq@<9l35XJfr
zRUUe06yQJo(?4Z)`WLegLMpVKAq~~&kzBtTH}V_lVI(~Z7Y21<<z5-&sUg3eo~a&y
zbjm`gxhNB%f}`OPC>X@L5k!=W;s9HPcAuLnC=6Yp&$k=b`}Zr3^HK_ie#99er#SQs
zUp1Z@28|hE&{MN(W_M}~y)Lr8{oC|RDS#@A#3AfpQ&^0Z<5VQo&A(-u#v&=Iml4>O
zFSMr|Rl}p3+Xh>KZEirl0)fY?@G@~I9rSc4GsS+IJWJ=qb4FuFmqTymDh<#g<yyUZ
z{I-gUGhUCgN{tc>?p(zxfwJ56(cUeBWvlXqKMtYQk|Q`hi{Yx0eG|8K6>AhA4~Rn<
z<!S$g_Po=$0Chl$zlHGaol2vu#XI@;dFVF0i)`n;d?|scdMT&;qa-OJ4bJ^|5zss!
zyK=(np=pd~<p7?Z=6f290iF!F`y#72^@@bvfBiRqn|T#?TXt|LY;@4m3ld2~@kKG`
zY`+c9l=(dK3_eXg_#!Je%;n&9xRLn?WaczDq8Foq(L-4uT(RU=5A-sgjmOy;lt&MR
z6+Jx3yIXJg?n`!jy<snmH<QQZ!MpFj7ca}#>BYZM-uzcQB8;Agw;$Fk_gy*-%41~|
zc|HrCdjGRm%y_IuN=DvW_B1SOI0z2o)r;)KD}?^1nOuw5N85zP=P65x967EmoMt7^
z!w036dkTITgtYY4%z?48&g!X;vX_dw^G@cMxWAve#3->6Nm}wDnD-+O_)Q&QoKAh9
z!%oO8KIAMOkD2ctbvbq7sju%@&6i&O!-o$````WUcQxMr>tFt}GK$YUF7_e;+jmvK
zBdb<6ef;sq9e%c#1>N^%bk^IEF%$IPjd#cA>=`L|%%HWRgXc}1dGejTNlfO?P>adG
z!4JM<vW9Ww>h|qB2fzH~FI#3C;$8p$NV^jzJ&q$y&;SZ3+#m>mmq<2AsU>x{+GcHg
zZT|J_ub5pM+u5sUBu#78I!O>DaS-=aK%w@1UeBm1QfeE?CV;HWFES3l!o$Pk$ZzfH
zn)L-}K-V!YconPn2xNjbFF2<~nf1uU$oKuASJ_bo2le_4z?B%)3_s<f??F1qF*J<U
zM<1{`x{e`tQ!n0EKLtkeE*-Zmy>?(w8a&m_m|J{S-_;m@Z(VkRYz712MKpbQA8&Tb
zV8ql24{WC+-=m|Zy`PJ5P3|1QAB`A~<mk88(RmI*be;~q<NHsgZRmS1-X47Q4J#hO
zCr=*+cXB*>3|y@v^&51cA-Y(Tp*^3;H<ZA`$U;@{*$$uNm34ZXo+SS~K-D?{lE(x7
z;B~69#CyKP9a`<>dqc7_DzyQFsd`4_OTp1h?SJXHF*@TdGcd?MMq|ye>bU5-<R@1>
zp*Lj~sQjIDF<!C&9pK6CHpWEmp8Nf*<Y!cX$3c&HSc4<KSE#I|j@5?NAT`7&uN$Al
zrNPSdm)zG}xdLjjr4k(v;DiO({6JJ!;<r0wxuvBebamMWC}phjE~XJakAT0GhGTSP
z@t=mJ@v>YI6*+3*hWP&V?1Gfp@(~EPQ>Tt;U<J{F2kHFsXPz~n8e@o=ID~OARxoQ)
z$uRi_0@?G%)w<{XqQxqEm9^5WyuBwM_@q$>A*gFMcoH-nuSIKzX*PI_;QL;Fc|$|x
zqp48ZL`ju5spudUFN@x#A^E~^g)rqTW!vwv-~)cNJh%?cz*;&a{i20?#P5*z-^a7G
z$@>*ndiO~iyu;t(mU;#CR2c;qX~5e0@~m;|d|CDkO3!Cr%<<J^!wA^|Pb2FEi#xiY
z%*c&OR-BljyPBIeuGc>2w}u~dJ9=c?Y5d(<8Rb=_jYbJe2EZP7P!UMh@~#F29Ks7c
zX(yUwrhJ0Xz^M`qIiWD1d2nS+gC-p5B1Xo@^K=BOo#^~!CF7=ls&K+zbaDuhQJOk+
zU>4%>qTEj>cIloefD7q7EH<^SBZgn5fz){}ZvOr3Ca_pk=h!LK<n<YL&!(Qp%gfXm
z1Y4)|FpM)iaOoKAeDM*)G;~7Lt7zuWU`oa)ue@{0XcJAk=>&&cXv+ws#M_Q@(2};@
z^#n1@rX$LZnsf@LE#!Tij^CC{-jRFhNEjZz=1Ms;P|yW!wI5zbJgGiPx(=c$FT+=9
zs<81Y;Nj0wt`-bB2Sl0bb#R@j1IG>&-lQdqPeS)9_rg8fywAii+Y5cri9Mvv@${f=
zKV+A*=)dF8!Th_N&X+gn{TM#H$4{P|uTIoq(GlXn<}wuoSNY)%KcQP=LlaX$I??BG
z)$?q$&E($0;gTP_;7Uf(|77GnMDH`+(!rJE&;%_EEW6ez@+h;WY*&gj=t(xFji;=g
z@I8gA7mZ8>wz8CEaN@*dyJn<G46rBU@u|)?@-%HKxW+I6i}oNCrljeII>8-S{hQZ*
zkG{(2EO0JEW%OwIoHEc~X$%BLihk$$ix?ak(CtOu)CG2hqBUHrF9DFC?OA*}UK~br
z+F#0)#uvY>W5cgOx>?$WVLBK5W}*D4SNKNOTV3E9JP&hkncTc;d|`^-WR$IbLry4%
zCha%;RxTQ)4Ign_Z7r?$mN!4uFd^lCC-|MKX^+3y;zfNgN7ElB-O4+ffINKj=O0>8
z1!Ou0Ob|xE@(h>|+KTVs+>2n$H9}lMX!Ie#D9JD;oWyKGuQ9(Bg~3Am$xnV#i}dIJ
z<-gSPV$M!FqXt4_go9``II5(KnnxHhT6s_Ro@A<nP)B*BvCxU=tp4MHan;lF0-KIo
zBW`C%%93{p`pWa&+!MR#v6#}><kh&nb1Abzn{r!5;wq%-Nh{g4fGoRW&P9O~hQxD?
zM^*y!xM@4u5AN%IJH(W${2GB_fHM~u4H`N6?8LGq%xBV}pyA$Llt*w>hK@tSdhO$9
zXVTF`nLnr8h^I!aVd>m9#a-H=Uw9z+d)y6ICWs3>E67tNBkmZ^9a*e&c#tyjwjEC5
zySh9yK>yQ$i&DpToeV|-1N=QyD(`hf{lt7=twKpY^{Rgivvg`QCG4jo0AcB-?kEot
z@RpZn!8sD#+FbEu>fs|Nvfgi<U?Q*J1xtQ*PPGSyCC2-Xxbv>(nkYZ@FuWO2M&Oh<
zc>t00WDnoKvPN4*W6zTQ_TAh03T$><B<-n;f-Z9*hcDJBblpL4R~E$&CT#<Llp2_y
zM~<v3aE^#Px3gB{lg~cs2>pkjeOAxG#eB!fItw4SLxUgw=tq@}2bnYHjkD`RP`VC@
zA>=EwrhM+)x!dbzcHDe<ld)^`{4}rz?TUBWeg37jmsWf6fRIhJ!5<wmyx-1ug(_pI
z#SQ_m53f^e7T|IrzubG4H7xXnT}kKTCHkiH{Njr*igR~7k(=<0{=rA|{88|+>+Jn}
z@5cR;cnYeAlBQmT@LBsZl9W&Tz$4G?NW!BU#^sTz0*ig$E3K}^^Gg0cfB&8C6#DAR
zTk#mZIPu%x{<`_O{H=kz7O%{?EL!CaBnJ=Usd6_82c!D`_HTdO4-MiAefQJoLLJMe
zpMFxk_Bd(p<jkZ0{Gb1iE>iybtFN2S?i;7@+KKb=$A8mxC(6S+qi;KNK8|OGd_1-z
zEA8B3S)=_&@qp0{;BcQEPR^0BW{S~}>F^loj#|feW$?oAgp616T=PELc^X*8>oPKY
zD(hk5Nr;ZIZibvc%-Rdvhb^$y{&=%=+}3N@-DORfPXAowN*U*32t0k9Q%|Fh7?)^i
zjgWSEfe{!!i3gLbZ8&;gUHkE1h_myNa9G4G7(%S{+9KV9R_LvKA9}R@$Q04-d$+5H
zzxwJ{y-g2!L90i{J7PMWo<U>sV!`+8SLAoT4A0Zd6JQE>wR3I0gOlXb=3j@l;3h*1
zErzA{7tYU-_oKhSKREWDd`UF=GP2H_&G_J%+}nvqaSMziZLPAN{Q5%qci!+84IQ*G
zMxMJ+D(TF(;P=GkZ*XxP{$7bznq&9Fw9;>R<ZJ)ncK`@y{_-#e#}OgrEK&86DKP4@
zg{@BKkjW4ZqcKRFXN(7kJ^2==BEB*zP*F@=tkVBHKw_rte+Z;g)`8j{H}ryB;0P1i
zQEzpK(}}f%4O0k>P_hOF_fdA9O3uXPU)~{1-{-mbt-LsPn6HnAf9iF}FC{Ss;HJai
zRcIMfircQ1l=7Al2Nox3QjQvwhOU$0x^$Nw-HhT!InLHEwyeddsc{-x%BhPlLbqJa
zl)E^V`QSXbcXVC_5ThgwhpTgp6GqI?$f(BoB^rvWlnM?dI-C+WRPkfc)a#+a#P-EJ
zCSD_xXBlM5P&V3=4}3=_4a&hfEOzk%I#a4s{?W7+I>NW~48O>OG<c!xS2@37@w>d?
zs?58l2{h6JsyO!Dtk^fTr^DrGurSZUwL@<UcyiU?0axnv%m8*d`d*#gCLS-zD2N!`
zc#+D^<SW8>ih_)VS1DsBvTdQ{;GYlJMS=U2yUsxyp*)mX9MUm$T>;MowHjJ2I@!LB
z7Ucd#ju(HCX${*eePqlO*jg)M@t>pMDPJyNvgLW;o@qOy@OEU+rcBdRg1Yd~&JSLa
zXDP?UE5%@;z**1%cay!et!**1@2ns=LqlB@r3G`~A+P9yUKazCdc=)a#ORdQh0$df
zPukZurDJ{6`Uk#g8y2}g=;&PM%1Hg<i<|A-e)TW^(t3KHL^>Xwk4{w^G&QUp@+hob
zduV^y&Vip$xecrrv%cnRrnWvj_(8n54&U0TA#3WTybYd@A3sRv;tRxqGBiPJUa66y
zOgCnl%Yj<I{{Mc}4*vF?Z~F1Q4}b7cJsg+bx>#CSY_HzWKKrZ&Oc&e+hAZhf9Wn1f
zJiB@h7js9>O~1k&UX{7X^y~AP!VC-sD{v;e&X8evkT<&1Gz=Nyc{JU4|G_uKtwv3z
z$lr-!0}lqdZQ76G`Eh8OHf%bI&OukxdW??!)TK<ad_SKlAfxm)m?qyC6zG8dWb*vp
ztod=If_$ck$Q>i(a$saUtUAg4$N=u(4?ULlOP7QWUBBd>w$tBvHUKp(pNPtu^5_j~
zD$oR9?dIAI4dHk`(>2<IMY=UKLMwXyg^vDpWH+)*XTzQBlJ8Vpn3Z=p_(~nGvvbb-
z>cUOWnZloZLl<~}HE8rs23tnzgYATMpET&|l~`?#J4AbcqXCs?pAC@0m%@W^yfsRa
z#(kBy?&ALK&@5owLgCG`d!N0_)xJ>NQNq8@2vJxIK~P9;m8@s$1XP5m$1emWl$cwA
zjnImdkLwFVYlI|U&v8g&`5>#iyNEF5d5`(ZrtA(P<Qfy+s!|7{EOn}9I@`F_wsHvy
z=6f#`PIm%=*kNYjMVX`#ze3kF9LswgFKHIB$GwHYW%YOm0-So$Ouj|C0NZ-xw|+qx
z=E_<9hek;7fRmAa>AS&25t_20Jb(*-qZH4@C+TX0<X$=u)HH@JQeqJ&z9F|-$5EE$
z3*N*+9wQ|D(SX#M#tFVE+m&}p_|D&!8`#j+^P-DHejnDG_h=$8Ml`0@gBM%6D#_3V
z47_V6_s2+)2MzJHbd6!)zRE)z<>gQ4FLh{=y!n4r|3r7R=MkKe^7a(Q;A_+(-Uv^p
zXexp+Xms)*iqqNvFeXt;W1zTR#hYaPLl^SJFtB!tf~D;7NGI5JCZ~gcWiCnv%`!Tx
z5p<C4s_}Ag>+>S)r92~2%37yj<Z5jKK7$Xgvm1h;8c;$%VC4p*bRMBq;9(3f$hqFh
zhK|rFpnhLYXZ0vMRczU{?n;O8(Z?TECU`~A#7g?EPY6HYVAq!V$tApuq@U*;9>$0D
zSFV@&Z7mB}^S|=#8GE`=)yOYpEF*-WrlTbPbOf11lUJ|aYP?Rb3ueOuw0L?tM$fb8
zj?lO#9w#kYlW7Yk)nyc*Gx#>Rs_S;vnlW~iZRBmlP4=`sYdEYU<1KJ`wvNLt4c?2l
zvtENdEbmukC3$HIMyg#rsU#g|raIz9bB=|96WJs;cc0x2+~eYB<nfBw#%nE1GDdIg
zcRdc=x+p({`uy`h7LI@a?+-dM|KP*-VvOYT6PdC$MfUW`XYJU><1qRI>~`L$gPtI#
zaL0GX)bjG^P{|J->j3e!27aEUHyz<U-RU<xd^hHco4>6-!J|bhI)WVA(ReXae&;fJ
zC&yPZ<>%eGvllmS&+j^5dj29tYG`!-TLv0kKyG<LoH83DByjR>zId5?WQdF~>a2ew
z2fN8@v~TK(XOS_of6-R&ksE^<{YYguaH6j*=I~&A&>`^Y044buAZzNK+*BVXzd>0z
zbK=^m?AVMUZeXZA>SGwfHNFFj^4hjjHoD5M5t23KfkX2|Pxwm9u#>O4F7KuHujw}!
z{0?w|J2GbRH%#&)O{Y4pn+@p5D7&qQyx7vfEB~qoV_?v4nU`-F?N7XGoM|E)$VE}g
zzj;JXbsncttykTD_L?O9>%DNiA1vH~XThIBnx^UU{8xEGR?a^M$bi36eD^MW2_FU^
z0H$y;8!cLZ#Nss#fq8^>H}Btl@7<1GxEERQPs40e1#3Z|hKGTerc4*%pj<G2F%=|-
zjLvi6Bp9AI&Xwk3JOtLTt9Nh{loh0RsdELtrSMG@_oLL4V$rIw2LJxvGB<r1hp?0H
zCQuNgoMt8dqMQ2)>QmO>{$1fKb{mWXYdV1Zqny!07<i13(^<wlu#CXvHDdA4!m&^x
zc_``D?jAVW;kyUl!yw%hUiuuyAA11LqHXa>85Pvr(Q{qWQGAp|Js}!O!zfdEcp{cO
zE!vbG$ve|u$)ok<_i&RZW6D`|m#?W?=yXzeBfWCrOkq>Bf{u30+1Fc>_tJO<?}A}y
zzJ9}NU^c>ioJFce<ls>D8Z9BN{PNa-j{K5C9v6nUh1nL#A;Xy{U33Dr4vtH^c!4J!
zp?E$fSc&gDks^}D@7Gy8Xm`d=It<3?!*rUAEQ((ztu20e>sE1h*c7<d0eB2z{C84o
z7r5s6UXC|+>43oOBYa?Ew3Hrb1h&8*h?=hkSAa^<p(8KdPK<yYT9}U0ap$}q7U`W$
zd%lr{g*CdP8=M`Y4lus;7M^4=a6TjdIHPZJ%s7H$*LsBZ77F4s$%g|2kYX0Z^9UiD
zXYy>cZZR?M1w)Ww#1q0>(XF>>pLCHaFWX*CsTd5=IkpO;r+f(h*4tSOnl&~?=8TZb
z`2ZeWr+oPHNFe|k;hnVBelSYxm^0dc@PP68<itP!^Z!$sxpViM4pgpPy<QyXN?s6)
zWT!1z`{Bp}#v6~34pLiSD9u7v?W8maZ6isIN3+Uf?kCf>{3b;X=K`OO*f~o(r!Hr!
z@`u&a^e?)oziSX83+H3d!0BvG2F62&#>sy%TVhSO+0OhZQ<^nWgI_(oVMje*ZA;-(
zI(t)xympM`r%%qEc*H||=}H&xs!JOsGw$e3Pmr-Q`I52Id(qRkr;0ON0z*USpicOR
zacLTO@1Vk_f^X!(wdim%Tg~C6JutR-WXKy&k11E}nNd~Vru-Rj1dn)p7*O;XSiqr7
zGN-%+14z1FBXOrT!;!HWq%v&`j=+<F3Le*ZV!&sL%wdT6K1t<H`D@W?jUPIHQ~SI@
z$|9#=X#Y`o$b#CH<Ny3KeF{FbsaHSG|GiKh2(K-8>aWO!d-bh2X>#@5mi~X-2nk_I
zSOlE%TVa9k?^Kw|j9#TfETiow`MncCbS?o`+V$+&K9`S_mC0!!?_^%n{J)bPrT^F8
z{H9(3E<Ff!J07DWOzX%h%qM&rA)#*PiFSdZ0)Xf+Pu!tU=4Zo}4(%n0ROUW#BbfB-
z8es}k6Pm4mR>2}ih~*2yHE6cCEu+Hi-FlxiIx`)RQB)TUrqQVny;#h(Si#*|$Ebqm
zQ3!4zmq0qa0CVW!GSc%~s6dBSuTIYNRHqJ7hf(*-GjrC14n#TuzJ)ok2pT@vD^%)8
zoXe?>aG!e6U{I6?Jja8E#C@spE63q$>KWS3mokmccod4u21DVHZ@X$R<XuO=8d*jm
zl!Nz!e_-EWoW>vg7hg4OG)sVmt!em0p&3!}2rMH+J>36m8Ujq;$7#w*S@7OUc3y@R
z^0ro5k70KM*MD&hyxgL2aw)GyNa#puJoxrO*MK-%f+r*tYI~3l9P<sNF+$3N&|z?#
z@;sw043nkhlOtc*)hVKpPMBfCK%(??u-crqMNy%_;t7VrrJkPTTOHk<+ji~T?#Pfr
z=FwbUO(b0l1tT%RmrM_Q=!Np=f!=Rjx!OgZPLUm>C0>Q-1@rPftSoTp{N-Qt-!~nT
zb%G2v9t0iSbWGy`U=*M&*x}2_KoI%S=Vkaxv6C~M$j(l-*QO30naLK}1YhEuP0~R^
zXk+C4(MKN?-YZ>@8m~a6LHa$K?6Q!Kk#cY)9*lTVbgCEP(!YKCZpyn?W9s3<$Hm9U
zKkRN^yG`i2N0F1goIv`mb#Q5m_hXEJ;SYcKT|2&?{lmYjhtAZCPS{Q$9UuIp!)I-H
z@YN>A0AV<E@DSu`w1xu=88p&frcG-GT1oEP4!cD|YgsN_;!$xb>hKAV9b#unohJkC
zycOg|J{QAMXG~_h!zc1)Czoly%6<662uhZ~fKR-RAeiZSIK6q7$0OUw!z%^|9glD7
zBa>tX?M|PbH6Z)Jbqp>#Oxaq=v7k`|HUZv;Z707QyTLBk=}enS&In-uTQ5RCIje?Y
zUY=#Z@KxF>J@F`pG9%LWq^i49M%z3Dj|I55;KX%s6FO*z^Z>l*rR<Fx?@c{9fh~6?
zVpvq1=)P5Ur`(elEy?~-*-DyzgDl{A`-TqG2B0BDFwcMVFa@51`*IB2q1EE)y0=%5
zd6Stb4QY?ki~AG$R{u`w|8XNEm8}XF0P8M(l*U-2h)bX-6G!$E0@k{b*}hR-7>jwe
z)IuoC{lIb4xjR5f`n8zJw9M-;>_UZoI>=DHLLHP_uMV7mGE@*B&yNDQb=?R!QzoH)
zf4@1IQSnli$;2?Kz$Ei@cq5duHvdtfqUPF|vO51VG-VZp;L(nhBqb0c!rg0YJcl$*
z9%Z<Oo@2DIdI=l2DDsxG$#jaZARQRhfITk<ehR!POOK&RBLV|FHP%_TLlBq3LU+H_
z&v@v4D--O>W^}gX{>Cclqs+GWrkTpC_?Mz9Z1UtPt&#?9+7OhRi#(`DP>c7Jk!$MT
zU`SzPNZDMhEqg1YnQ<!xxNzS4dO?x~mEhYQIH9MJE!cHzIylNkU2DBW%gcLdT2KCx
zQ+Vu<?O5dJqNoqjxjoLIUq<j>ef3Q{;!i*MEM+C{wyVYH&Zyc5o)Y8jyA^MQExATl
z3`A#QP#&fq7>*P*{I#zm+`=dPi99a@Z<JfwU2yC?xHUG(Ag^hZ8hTaW*;&GPPaWpE
z;aWwQeA+EM%a_o|^bVMfwjDUNwuV~!0Gid_7{3HAS!3Kq7V}WPqZxUN44=<AAL_Ye
zw@wTYyk?xy52sGedNDh2DyxBs=b`6Qr1O39I14IYJ*(TCoU{W@Uh4?Db}IN7`Jz31
zYkX;Qg=hO6YWN~^B_eQ9xacH^eU$}{a8(96*p6vx;Qsx)oAH$Cq8JiauG*@a^)!(i
zhTNrey!6+_7%gZ?hnU*4R^!sWoD7{cSavQkC_7zd#GKJTFQ4hJ-RxXK7kCG#Ok-rx
zXhzn~Bl-RBe^&$X`t|#j8Q;6%MK+Z+bz~Z(ozrx_bf7>M_9H*=Spkk1RO7rD=Xcl!
z<_b#Ri)+fPpU7g+@X29pbhpb9_KsvHJ39Oqww0z#>pI9Iewj9Lbh;C8O`{l;T9ga6
zIre<?4cU#kTUy&4!~mriO+WD{2u@1g39V<Xl8p-Cvv!~QNB+}}(_YC0`DRG+&d_o6
zCLNBK^o=wPXs}%o(mvB)ZC;Uci=tm?TS>F@5qk52SeR-TC(j5O<56@+Z}+t&TwZiV
zr`^&@{iD0YgGC-~#B`6}SxFgPw!9cvWxYv7qIuglz)QK?rETHe!VQKkd#kI@P2k0#
zfBbLH9mxyu9uI#1xhs(qe=PBms}<F<q)Wu{XYc<yBLq<E3Zae{xDZxo@hV_Cp*j|7
zy|Q3?rEwxOMagA_Q52XsqbN)^3e&Kx@}_|Q-+%t+3ND4GBYff<kO-ItFJJ2o(=Y;(
zJq|eW1dB4Dl)fJ*96`f3of4C!@Hz%QT$_$nnSp2VYM8nG%I#DnLMio(<#}0+i&=0H
zDl92Wx!x^>j)+E!gxR+uIvPC&tC|*VS0jOA8xf&P-eGO(sQ+!4+_X9Sp@v0Z=$v4%
zDE!ec%BZjc4wC?qcpbygFuX$_FngP4m(iopwI3-%P^LV7Em-?v`=ens9(}~6KKGWK
zyexS6S6$_2(hQ8?-10YT=p8uFZ_%)Ir{Mv0<xQrPH~G|E!B=-e>)q<>`k=foeDM-c
zTIW(XV?)Pf*8l^ALZ!?oo;5Ae3WG>;mk&l~esvlqZ1+78Cp+pqPTzl+`SdT7CGK`p
z6%5FvR)@=o)_EcK;{o|P#>&m4y>s_&I{BQP5M$-y#k8H|d6o`@%em)=B%g(>J)a=P
zMHO)rr*@@-!_z9-@ZTX^jBD>upL_h-bAtm5Kx52>@5`V;N5O~|UEt$BYhM^2V}yXK
zBQu?xJV`^|b;=#BrZd&aD)*(+WYfl12dE=Q18Y)rGIr(|sq4sXJ2oiU3$32T*g6@y
zRt6$rj1VJ#o&El6Q$N;U%onG{!<Q#Vo42EjJZMuoutyd~23K+2v9QU|t9VqVy+UDR
zddVK%@G9uk(5+KFspI*}$I&g>qI=;=bW0Xh#)~P>P7W%wZsbCyUyPhT`aw>pjq&mE
zM;~_tZJ{&1=s=(HkVQ6L#_O?{T}J9cV{My<ilL%Ep`Uct&oez?Cln8d^$O0jpf_~x
zM(4KhIz82ak2-ma{+E2=C(L>lO=w%AB6w`JMQ1Oy$r^jPze%TiGD^*sc<@#>lAfIC
zZ0#C*{q37hg*}a-eDChJfywToF{IV|C<}4%{GA)`6sM0q`Y8H6CuE1`^!&ZM=hBw4
zSh~|MF|4x<op;Mx2S!JYQ&T?StMwr0z(5tq+5Udk$AlIED{U$U8zFS$PW~-}<z$RT
z9?}z+th<YcApJ$MhDS+<Uw$N1KSVB4Xm`%k0Et-Rxv3gOYhOao*GS%ayP_+|ADv(g
z5s#-cUJM?<ao57+&3ICX`tWy68QSet-J^okRoM?5_(v8mUWgH!J40D{(AVc27${fg
zjo~-)y2^r2YTKBrd^DPSbYZc`Z~fc~A|>)Rx0R-Mh5dzLrElM%G<kFNe3ATiRDFM^
z?X2*pJ@b0={d)?sct=)^xSd>nt8^jP7PEEW5CVureinDC7(j#%!OGD5w%E)9Iaa1J
z+mOR3jCK9&XaCTV+5i22{h$5Dh52y|;t?U}(yaRBtx>{)jVqDiVc)rieKL#m4nv7u
zz#GNLFr8pkGB^oIL!Kw#;NjD}%3t^pOZaWC2bM|NMh_;ToxxTwJWU&P6q<SoBqeeo
zina5oX>igAZ75NwMtSY{!$gAID;-Ejl*e`@X*8ImK50smG(H`J(YD{-ErcwTq6k^?
z>N2viCIpSacu0%nC`4OB>?xB<Nm(wmSky{%;z~-7r41_BD}Ow6<dg`Fx^&vgXor&m
zieq51CSXE@IcZ!*_Fz-DybD*Ii}y~H0fV_#<r#?yl~{R_rc_`!NRuDRVq`EtDAfI@
zrYxci^3V2ZZhG~+j$+Z7<6j$D%5}Gxk6ce@vYpkT?xf@GyZOZzh38K4GqSIwvji);
zp*3Ut%WUoZ<-h#PiCg&``=c1#I`iFh(i9n<P;#APCfo9*J%u8nf%Y>7!ZIfEx;O>6
z!ly2VTX~h|9wHS*_4tWhEPhXCkFMoMMx>>0rtXSg3?OxY#cv(VZVqNTdv?CVQ3J4b
zPQ9IeNgf~Ho9{afFVWEY0Ug%Ww{k$2wHN1SnxktBLfcdM2%z#FJ+jj#J0Q;F<ArSp
z8FXX;-QLSt<Xe0O(bAr03!UI|Asr4}EOPg|>xEJ#eefb_?xz!nN1$y7hpu3ghe6~r
zO~VuBus9=H+iv$##;)}q5$e}jF!(ygm=Cy-6Wgatv->*iofCij!yj6n!{HdEcJ<k{
zawbz7?fKQ7a>(ELlxb?pdWhYuLtyxfu0?NCX&I4+e=k!XS$UR@j{bWRx4(t8<lHm|
zfii%j3-M9g*8!7RTajaBS2rUYS*Z+GE`m=0DNS;Y$?J4<=<GL|)x&neg4G-Xmo|Am
z-Ua76!P|j@_ZjRl3T6u~-7)%@@$q)xyLLU(7=gh8Yp__%OqLiiI&mI2`p12yRehVt
z^g{S?7VXA(;Vt*UIqetE=ud-LgVoD9Xs-h=d_db=r(#4hBpHi6M<x8-k8$H<TQWt5
z>hPE67XLb>A47~XWso7NU4m$}re#gt3UJa+uV!6S*U6-8@PGE{XZ2`)mA1^YL6+zf
z@~LlhEPU5>1WFqR$UqoYn3dVziE`)psg4T{g(m!xoFPALm8TP5$KwbWbZGyOdmd6o
z+EwmoDqcQz%9^rn=bkcysJ{u@9shH+Uhlx}t!HwR=h3cPoV43@XnZi_5=yb{T_f`6
zFE&C{RNw-Df>e5FTzC$pO2h40dY8CEfKoZ5M90^6$wNV7Os&n7U5Y2Z1PYOK&Vq%j
z7%O0q5skdcDs!4A7IcKc(ZIYKl!;EEOv?LK+7jl^Ks>O+t?L@{s-xpA-67t`@_7V<
z06(E+RhuL}Wi32Pqr6v6nI>`G_n71;=n9F(k%qK7x%Sn{5CZr5U|kRia2m{9wQ>T0
z-$IRxWGkZ+x4asZQ?2PuV9}%7v^K@Y9c)n_Wi{Wcv6w0p41G1O?}V+c@5F6CiesJ&
zk94kMcofP-8|hqu<q)4O{o&M3euLHMt2A3ANj~mvji+*Q?B4pebV;Uy!qK`%8XMhs
zv@EeDIWpSLGWg_6(D;l7cXA3T#mgJA<gH$Xyf5BlG8eg^VDP$lsWZ<yZ0A8(MENuq
z<H$(4BZj=EjGe+^QKvI;xc#+@GciDPX3FOF)dt8t=rnNk%?Gq}7U~}bkV=8ykm83d
zeK)X8Tw8IKs(h3Ix)%2&=#H;|FI9@3cB_tD@~eWfHfo38oQSD@)ehcMPkz=_7=9(a
zX{&a|k@?+p%4cfeS@WS&Nar4$=yxOEy=*N$u<K}3_`D3B_p+!O&DsuA3H}x&bV{4|
z)m(Cf&OKo>_w)HJz0Q0$v_(h2!WmuBo6Z0qL)4|S$2WKFOC3AOe=%!3$VI0fTIO{A
zj0CV$X{YRmp(`0Xoo|4^jX~X=H(4{%4i^1V-)=_!E3{3N4_lk|A{FB^x;pzv+d>C4
zV8r9mw=r(til@bthv_Z2RIXzTBza|x?Dj`sF1<p7j!sh_TJU}dVD=ntCz`%x<npA2
z&*|*IK`uK5nf9p_&>eP<Sj&QzJ`2#ZUFf$MKX#4?<IR}q4Lr83TL*1%u4#5{jDf;y
zGKN}s7a3!$f(74R#!$mc9!h%ZX{O+ee@*=`rq4pyl|TnOjPZ25EIM8J>K*CJ&^UNF
zKLl22(CFlqJFP!|+X5!uukchD3idI0TF<e008J0YXuA||7+GVq@Zdj&SBxd4(ywCx
zrtH*dBT5RH!aTP2^^=|pG)dH3s;`tm(La%1D<>EHToL|ym9!3|$T+12Rj243{1@+O
z-=&(tXzE+f$<nZXmp1AUo4d>N_uq?0G1uz<JU7Vqy>KM?-+S@reiE#@z_-Sa@C9Z|
z+)glYwZIbB_ui#q%0oxnMvy2Y9vuh)B6t?@j5{_Ug@9ebpZ7jl&-M6|pZqk3EPYce
zn_*!Ds9Z{0xohOY6<POw8itW3&x?`$@eX2bi#>>Nl!zN)DzpA<)L+W@sIv*t1KW2P
zoxB?Qt6bhW(8upL!B@KOn#sNOt0!r|&zkq1_r++&N4mPNKELHZ9=O~cbR!ncG<=r^
zNr;w<G8R~~aotg@b*Jaf7(xHiz93{RP8z*Y$5&Z<@Y~<~YdaAO%kJL3(|GUTQx8na
zSe<=G%c(_t0ByWV+nIU*XBtzLn7YIzT@`D76We(XZMOVet)*G$q4Dr$tg@9WE?G-A
zzf0E*ANkrjB)%PWXuC%Hs#Z>~{y6$m7}9u`JbGh9<q#x>my<3LoH92}=346@o?o&q
zAO@QSnlU7~P$?TD<{H@QOsLVEN5^K%YlNr1pa0jN_p|Q4B?B>Co6m~#eEcCJ=es$G
z3LTARZON6#sNG0>O$Du~A9cUU=;!3l7$FoiF+-z*kw1)vxuhMLFMw6Bk(G!|=gaH9
zrR3+JBWw_+EW92A%y(uyGd_$`?QnUWZ(CUGT;q1*5!`Uj)sbZ?vew4b#XC3NZM#Kp
zqf_!vMo-7Pr!zSl<HWR&5n`t;azLGFEIV_?2qMFdtM`G&X*2EudC@teKVzUr3WW~v
zAR}vgrH%n@85=rm!R|EEIiGy;$%#)t`J2kOGH!nUMRn5ibc|^H`gP!m_v6-=8RdpA
zU)=n>y2d#fXdFU>ui8Njgw(m0T|93x;@9!qfB0>@B6iV`lcWlcU~h+>wo`deejWoI
z7&^?LK`V$cb1G+x=y(#0@NcKha{J%9dGo}t-v3p1%l!Cne|+M{KmKv2W^UZL-a!xg
z#Rh8oldpO-xTsHOq4b8}%fkTAfgzoKy)xQXJV;sNWaki>v5x2JTlu))J2}QaYeDF8
zr^fOUzWbi7&e{5$X$tFBY&#bi8}!MQO^3b8X9LuCJ#ebCYe_;w?HIn=_|wN(tokGi
zV566wMHU`DnjKO44~@vWse^-@Ar<e&@G|;^jE1&!(CfX>F?CwwgVuJ0)g#5b8RK9K
z!jW$>QE!Tdk-BT_#sE{UcS{$Nr^?i(bF?gdhI4*|^9QCEcTc>V&j?sk<YJ^PLrov{
zG&1-qx}$bw06iOctY<rQN#7b{Bl1GtcLQT8#=O><O>Xl(*qLh2|Nd5QcNT_z){=C6
z2g5R1*-j^8e4s1iaHbyfZba!YzcS#R+<mNE1lLqt8S#6?t!H%b{_s*+f42u{J8WS~
z&hoCMr=Gvu2uUW5u~!Omug4&FL4oZ&Qz!&hoVyg1z77ln{l-w220&m5g^QBVagP8d
zUCJjkR;Lp{XEf9p+VWu}0n$i)R(!3bHz%xYMu;h&jv7LI3c|DWds+Ba;+0N=2Ta<5
z1_c7;z0nPPJyR_@|Au6l4mbDIRT>0VBDgwbG+7Lq{CJa*Z_*2l5Q5FhUth!x1byev
zqN%&i*4^mN1y;mt$G+w8e!+phel;$*xHoO;M_+D13X~ydizP+#AdAHqNCIB1(2ic5
z@^gVX5B&`v>Ow%pH6vF*HeoDA{Zpv>YLzwg4s2+mEVKk`o=c_tF3ln*?!w`8O45M?
zujO-}#cc!2{2iK16KeL<KaEg5D=<E+H6R;|;4T(T2WicM&Z%zxlUWU)(N@b!R!XzT
z?4|A$1iGUV{cu=W7UD&=A3vFCuRHhd^i*4Nyl7zCqLHi}1f_rSjWRU)34Y-h1BcRu
zBL%B$^Z>6z-6*SBbCKUAQ}|P1ELS$@auYB~qLh*U6FlU<@?kP!o_qdj3;Cb4=`6Rl
zdm$ZZ7s?qiN2yWVsGzv=uI2bOGNTN}o3$@(r=jC+T>6YHUl*2tc%OC=UdTy_pA606
z%_bsFye{Oo-n!hSHJUx_9Fk6`vRyPnr(9^HQ*pH}?G@hcDZ><o0S#}39eSPC2)^JZ
z6ZDO8U%Y(S(Yi8iqkWjOR(MMo@2|2JYoV^A9+x}g&a^;ak2k^2q_QY*l0hJt+L~kV
z@df<&tE`d#(L3XH(7E7?cRFxuMjp5a9{SYr>>qyk;fbqxCP!d~uk{4FW%O)Z@(jOA
zveGb>7PiGh`N-br&M`<j733hXc!Zp{!k{*lR98df1TQ+xmNcf_sxMka>d!kyskIZr
znH<=Krv^4)(WW}UssN(RskDv9q0Qr#MSl97m7F2c!_;En*HaMO`pu+>FRz?W0=zjn
z-#gQOl(~3q@=RT5K%VzARlx{8$hT!6f0Fko`K?bWJwlU)Fv{92{6U{ZBRGxhGPpH5
zy3t1~o!>GGjN4ZvPS7t&MqcVI*vgWIQQD3<Ra(QNjv$jOl@b)1rUQn=dmoP4!n*?Q
z$s<XVAH5j9=6A49`wjq#Y%F|~J>`VzeV54Q_P0-dZ`Ys1HD#gON}1;=pt%p@S6(&#
z#YTus+`$B-qrD9fg){_P#sU{;1tPy!Ck`K-c=M$UK||N**O9i)L@*M6C+CAOSO^C~
zVlJeA_`}a4fM-tpx6GGQ*lX132xH~!hSBC!-;FhG>c%%6xKTAoqeml(vFbAinLO~2
zS3PU)weUrdi<3Va2Q6@`qK^<0v<fdnCX+s=()h`Q#ih;2ttVeD@VApnQSz-ru-d3~
zK*4P%9w9DTonJjgi4)(>G5J$j>FQ%k@HATD8F5*>rw)o0xIJ1uaMZI9<p(!FajOz;
z#Z}=cKj4b61<>flDBC-8<sHH2HN{cGAo0o;YzzipM{RkQ2R&9gav~J|c|82P;wXYO
zl>(=FB@!2a3sV;r<~OC#JgE;ZE^-8Ju&we`Q03)4{^UJ=<)LYd_?FL2D0|&gc0*oJ
zMwBVT?CWej{ouVU^vgEL>(_FCRkn~D!JN!!#>lcG#VYXZ0tn447__cNyn~X!aX&H*
z2lL_9Fx|;{8VrqJ{Nfk&{Qdpk|3gP%6!IUxyqV9#|91Egm~7Q+dkS#yX0)~=fMXPo
z`Fe^q%t(sx)4!rT?Z)EmV!6>N(vxISWSJHqGp;Jdq^i6&wNe!6Z{kPSB2szYQD(et
zJZ$xB7@aw%AX0Jq%yfv&m_ItKb|z0|A?{ul!j^A67sOYt=i4Ueg^#CCpDR9hbAXlF
zcS93{mMfR9#$br|A|4Or?Y;Ej%7M?|!!}+E9~p|)@hIy-_(aJ+*Z#vB^b=2RJ)7zh
zoNTJi`Di4=58v(RVJx)6N_s|!#jsa0-%rPMgq$>&ujDyR!p;!8jPTA*s$c%%zs6hp
z=)^bK*;J26JH|{K#ON@BXQ&*}U%M>lijlP-!oeQ7&BLI%JqP}DPC5$SnEa~jz)5E>
z(f)a#st+Ba8)lSm3NZ#$WZ-4+MicZ_?(&2%#=xUJ*GM!HPnmeUzdKXi=s}Ko`}X!S
zr4U@cj#t{@b{59E_xn8gc%oBpAB86L!Ka^o+(Yod&@)+ZF8J!i(eb!pViBc9sbtx7
zOmzx<8@TE29_fB9auzt~zEsdOXl-!d3=*CmYYT(W=-a@+P*olts(S?<ofk0%_j0^?
z%M6Y1+EpJ_exE)VS}>Sv<ZlMSv$XY1jkGxV=`mA4@aEY-Q~XpXTJnbCiN&n%=R=V^
ze+;ObUwqzvN?+B!A|T2O2e-9T>vg_Ughtk*GB{rb*Cc9v+DV%A$Y$WsHp|oCUELaa
z^$*SQzIp{Ms`@rvnbMrmGIi4d$*X;<rLY{wxY6JJ1K49>jL4L|;<ooKf5qiqI1hiX
zM28XIf6&>@71H$OVFg=KzZp}D*ZsXxCNBmNF7Mm;^1uSl`nM_zYZIzGlctF0uESQ2
z`Ezd5E?kQ;Vn`?t<C+HiDNmy`YT+(a1|g<s+zG4ngujh1>HFI~0@nJ~rt?ae4zH|^
zrKzK(G{UUBV|kv+Z}kcN&PpMj#Y@=N{C%50;B#%EkNGJ-cOU`PSPna6S+;8<Sgm*h
zBoF#wv43bPFc833H=Ku6o!+TD+d7^N=4BWyz8(t5f;0C8!)m~-8w@GGFeFWCD%{d~
z7L-NsDYlijJQbh3Ehmko_sP;c%?H0+Gxf7|np>K6Q^vX)R}}q8fO~-&IM974G_NGt
zQc5uJwDB5I&q$NPR*sI%fmOC#I@0=C6yQM?816<vGuFtf5gbLYBdkIQ>}ZQdE()Ha
z3Zp86)_ibJKaP>&I}VhTsR0Z63YR<^Jmj+qIe8cTv<OEn_qgReCwVa{mQl_OFmBu`
ztu)9$(KN@~rW_<_Led`pT?-ZZXe?cL5v63IVLOoU%{J1bk=?X`8k7o5Che^`lh2~r
zkR<S@Bzc2djp@(?tfnYH()D6hu1T8_>Dgp5RfBdlqLa?_lky82<r~^MXsw-Z1~%di
zD$x3Acw61zcjA@9aN`9acVoE36BGTQ4d|f79}mhivfYhNIXL4Wv{`d-{W=5YbPh42
zS8SW!j}h`HQzPFb{h1mHI!~S{#)yv~k~KV3zYZN|z?B!F1AQ?(p7x9<O_8<`y^vIG
z#>rpfCK7#W&e_gKlZU?`2Y+d~={s`Fz*<HY!;R;0$!-Tm^5GXAsV7&B{QTbS_eHWW
zJGdat>(>qm<Ar*5;!4UTpY*LkWlxyBYSHQ}D!vkLl7q^W>1(&eA^48m_%)#;@UC}A
z1&LOV#m78IOSfqcCqr)rfyK4;1SUKAuy#bdo%2ebHP8A&uXQYM^6jkAt7QQ;bg-<w
z;18atM_Z*|_oIjR-$YiDS7J1^4cMa3mA0L1F^XLewheGxqZ6hR4pYg0+LJa;maKbX
z5HXIxuFcC^PaoM2{%z-x8La}W4Pa4rFY=NX46%4XXGWs0TL(TyA}Y6ocOtxIjG%w;
zQD+_-)HD@MWmA7(KdgMx9sQN=dsosP#vISLJx|F;B^-as06U8~HOw%f1qU#rZUC<l
z5hdYB*R<irojkP`2*ssXVqkPegCQcz&0M)HXv@V`t2)KMPNQK+@KjK20Gm0l*C)Qs
zfjl}-o{)<X?CVi3Jx({zLg(vaJ^s@_{V!d|@E`xT|5(Af$nsCaIv>IR-M{|k#P5Fd
z+n#RdNMZAl8iz8u#Eg7&5)MGaj1ne9Vm(&~qO3H|2RT5^BC9zxD0{$TrBV=GNEG3K
zV}(&1BIo~Xt(y5;iq1I)?r7^eoy%|Zwhi%Ob_ZA-#WLplz3k$_iN&m8yVL<L%263t
z^Pa*rx^v=ej};Ez9>yRmje;BNl|qJJplqW$AA_v;raYq(H1Po}hde3U%6cPyYf{jp
z9dPcIQRRRm>4efOu-G9$iffD{9$kxD8l-twE@{ewqDL>ClDT-#XYzq7X^lF-Wdvif
zTXb7+CK>6NL!%4X{y2P!v6)7@pP{+><kO+({Otxnt5X>YPz4uLKfsOLz*l7(Sfy9z
zrFh%W^}Y|Ip-0NHBjzVR`rFFW*ICJqPd|A72PZ!L^wSeR{pn9amuwRZZ761SQ|RjC
z-7vK$KVJU*Km2{&^}qk)=iSAEcShxRvz`Obu4lxE-aL7K%n0Z|{ICC^M_>E?23Nm8
zp{>qR9rfBJkB$`&O8@+M8N9(sHyIhd+IKWLjwhtIita0(-7O(1Gzq#}z*)w!HlHw=
zGJqIGC$AJZA%twX(*Cn}Qd>S3Ba9p|&YW7h_cDfl=z=zICaZg?Ov~*Kjo`GC^t^?i
z<RffXEXq`9fI?Gz`6eF+vo`7n*~!9t=}@a0aM?wo)9u%3QznSv?NT2;8$3w!>#S`e
zyB~e{VcSLuFAdh23y!8&C-}|<k89WGyC()LIz}B^%S^e9io17mQf=y_d*J0dNc%#N
zgPfOf@FrJ0Je)j_ANyi%jF_E#1n}ai;1SPGbfeCVO$SdhU?9}?oKdnfVR%uVJk1gM
zF+wd`#^W@R@~TIVtINNX^<p#7&h18q^UsI;z@4#oX|<E}G<S1w-0gVrZr!?h;y?YT
z|9#uCn;l4Li@X*NRnsPI<)sJ2fz1cw4zh#r`HLs*q#7Sw@sD@oOg`3UJ8jowWO|$D
z)Ov>B|2_w(MVHwXLjG2{Z6Db_{rJ&*>x=RB!3Q5!ZaU~O@O)%lA6)Y%0vD6N{2_bG
zAd49q13z+o+Eh~{Dt$zdEsSzp$SJeuV(45-8&oc>j^<C|b>hK!kgf5iJnm%*)Ku5`
zz-h2&&Cv?r=Dm3RB3`oaV<+CbneJ58$s9_S;~*jf=yux4)|Y00%d<tFSwA*77@W<Z
z5?{`oj_ym@w##d3G!#Rgr{aU?7xM9bysdQ9{cn93GalF+cxHRA4_A_tHwV@c86aob
zL3E-m-XoJ~_t8Bsp7T=CiP{s47<&wic*w#Rb=lEHM~q<<9g}jb<*HyN2N`b5YF{22
zPUBCW(fQ@Y_D<d4rZPMiZ_i>xM)%A+x+qVDDRfM;Y4FmU-X|)Mh)YakkG+}qUT#k^
zrr@Mo`;YtkSVfbTn%MH7e4ZCl9*mX<y;p<40K_uXKy_9wUXE$(1S^kGHl{5sHhq+R
z=9C3NS0;Bdgt-_oP6wugqR=T0orvvF%G9U?iccEhEO=6WN3mcCbCqGVawg@$hLR(2
z6d3sWJLSzt2c9X@q94r8v#r5sszcF{?xzlQ&V>$yaYlp{ioPd4hkUE>sV@&LYq`xe
zE`n<msInB_&=FIFPf%ye(kLQ`5)$UtpSsl2PArV*2&-kFH32?$DJ)^*=;1S3%F>nA
zNzWP=XX`G$s%v0NDhX2nMe1D}Y8nB!N}F_k{ZL(Dr?gYvMytM0o<y`0^A6oy8k$Q5
z19#{<sMUZ~t+%tw;|Qlj6{KXYlF=dRXviPsD&E-=Iz`H-YmUH>0>MRY+|NcH_cO}p
zrYDc5l4mdJyNES%C$2nFpE~dhT@uI;BZgw5@O8Y>DA#>>6FR0|0|fLeeUn%7&I{s|
zYm3hhw4FB9D)J`@dP%q9+&6}7QE1$dn<qaT(}#_2coHQ}c`-o9b+-)$ZU?U|e$C4J
z$XUu^bTD3Q&yDmq4UZp#MNr4IT_fxm+vpm;S_ek?7m$V5c25Sm002M$Nkl<ZmcFtn
zbHc)|bV>ReKpP#kW$UKUuc#wHWiofqiA861R95}YGkJmwI?@B${G2(oDRJ_KN8C!e
z7|k8A$B0hTZyY++8&KX1ZR$meyld+YZrk4{SLqBA2YyqljQ73J*BLJ0JM<z_?19Gz
z4^fR*MMqFskuvYJT;KdaN0&JAtOKFPc#s{GX1bir;SaCEUSQEFALuBzvUI@O*O_<&
z)ZvBpg_nM#7sGEqZAD$r&ikIv9A8i8+OEr6G9N}$D3WRK+O8=v@aCVUkauLD|3bGs
zOkJTf*+yfn_*8U6lJ$N5%GQ8qrIVx86Uw{5H|L=+HnMvr`18Qk6A&Cm-jxs-a%-pR
zK}a3-4$<S}C2;7Qlso;!(4eWw!jbYk0&0q`x4gX3RyW<Z)l&M}KyYvnqbOe{=9$;7
z40s{;?Ad%X=TtsENmnqg2ks5b%4&&K7*bM!*AtEmM=m3~F}Z_}{%Z6!B0_c09cW5}
zc`WUx1`%7rCI%~gbc7~$G?hPORL1aD?Dk8{zY&_c?`nih;{$#IyvkTrOu!4rJ)1K{
z5bpt`!xi$gz*mXp$Jqj#_p$nx(&2`3Q4Z?pm0Oy)BI-R+{<9}pfRy~Nvk+A>E?J6+
z)lR^Tq?GxG&+-MySmHelM(5GFR>eA&de(YycWX?)`qy8#K2thQ&bxATK9coTzDHm&
zm7|q)I6M)%@zn545WpG$fdQ`8u4vR?)KQtEW(ixA#vm|~B}n0RcL@*xjrr{;T=|~`
zetGK+NE)4+PKtL1T$&Ig$`RTGUW1pud*&m6c+<Qo>ts&klmABM!i@y&IM69LEfLXl
zZj_sms$ev%0oK?_I%&^jWHLK1GS?eE-j7?*d^Oy3dSJuP8n9(BZa^=}Dlk-WyXPVR
zI*qb5d6J}xrgE}1zRowY(vwcdtU<3jz((Pm-kTji=!|dbY2#IA=y_OA%8<Ta#AIAX
zCZ!c>AVksiSK=sK`Nj~_*^Nt;XM=Hp=T6v^-rAvCUw_q+oo{4Z%+44v@<KQfvg>66
z8--35y+a4jct)vH%4;-;cVvLe6;2j4?gek_J322L_|$2w2?IsvyxNdfQ97qQZH{8B
zatxqQeG@#?y_XU7!CERGEP`s^qs4zONQ)mkr)E9Ce7GucT?~R|-UqM9F1pSEg7bYN
z+bz-U#cA&NZ7`4u8KFV?#OcWAspKIa1_n>Jj!}7ZfOQLig`2j<8+j_}q!oJ7Xx_if
z4jA~Vzw))^A)|i3jJ&=IKhW(;+U}){{;sC|qUX0=D>1v2@K+mNa)Nd*9g>xNFN3Jx
zF8Su;XK6QCGw>w3gnT+nN1d0iX8I@e&}HvM_ILneK%BofUS*VO!Ku!w9cKV5yn%Jv
z^5{Xhkw4R~F8adn>b;b|mrvJ|A9X2Lo1_y?rW4T~-oA0I{Cb+RQ{XaF(%PW5(f@5H
zuj4UzozANtipy44-}V_eBC}cOYXC&=Dbt|Nx|(Y@<_P#3*WZp@Tf6Xf;`7ax(30LJ
z^XScU>oWr1+`gL=fWOLhtH#=N^l1laQ|BYcJcI9L?E}yE$A9xTIag)Y5!G<bB2YX%
zh>=s{Bk+V&Z4;@Yh}1BM?pT98i_JR~NA{x&4tPb9hp|J(7&nX&bQdlMxAs-JUk(j@
zVVGfW4II7RE%2c~xamWe1+>dR)~R;gOn9z+(#PlHaX~+N6+Ib|WSoZ(Fz7>u3@8}H
zFTzLLxTPiIgVWqQWuEi~Ky<;?9R4QnrSwU5L-C5Ze{=h*>WDADyjgf0=dYao?DkoM
z6ZOf<_+{+vo@E4|iFf60Wir*!4BLPzdNtiPn6w$eDZE0bZ?(_x@H?Tr3T~k@K7|x*
z6Zup9jG3?EMI-y<dH7+Fh$36S@b87j<d5|%C0<K!qN(V3`qdc_c9(5FwRN_FTUja`
zHY9(bv{*N`dIf`U<VOirg&M(WeV}k97<E9v_bDhO9ZRUOUN;TGEf=C~beM1|^Xtve
z>VtIttAo<_ERU~jsz`_S@~nAZi-kt<o=L~2(NzIOacY=}C{55tegfJNntOf|nz@2;
zP-BGXgejqR_I{`HK`V@@K@(Uh80FXFl6yOz)JY*Mfv7O0eDy3Qdfv!XI4LbSa#1_F
zUT2sLNk2k0ik8@g1mTf?C5C54L0}P8!$G112W=51<(A(QeORw@ygv-v@79&)dTG$#
z^OnrJaRVk!N%;?_8Byh#oD|6ZDNO@j?iy3gpH$bJt#&}2zv@uF^pheF2~*}6$zZ}a
zInWSb`5u&1<pa9&fH<@WP4>xm?$F?!usg;G7HKH1dMCW!%42-+vdEy=(k*qMtBZ`N
z-@R$WC(Rw-D?j8(-qty==5@;}5fgvB+^UMidB-3DtM_Oy5cE}x`0M__$TeX-l~;kU
zh7lznp7c3Po3<91gM)8tfN9}hZy4C%mLgq|&gL=N>%z%&<jU}|Hh3`1$Yu@Zq}S18
z1~_ddZCRSof3*quT*LpAzSXTmAT^X|<%e+Ld%{YK*5FA2LpzU&o2aB$cJfFAF6}|P
zT?$`ml_m7Tho&Jn(&!9^4lO_-NhY+;c`WS8pbNa--yOYrCbFhJ_3!QNc3?B`rc()>
z836D)6(jLXro5#a4=qD(j1l#8ikrNq<5pgEVdBwjZ^w=-yv*<XVeHVUIz<C9{GkuS
z*^It3?U1MHm*A;x(`q~`I!J3Vd>5w+LN$_<#~b@9<(YnYdoQP|hIVje@aH8c`acgy
zy<MUCif|W1#1QbN|B@HHbll?qVX~Fn9shD49@#NTpQoIYrqnWh0fy-3)1jZ;N3MqO
zG8_-S8`Gwl^_P5-osIz`;`B>ioybVPDKnm_f}20;&!1k6OR3@xz^#1sXC21`=>lsm
z^m!9}&!6Y5AESuT`aBD<>8Xnk3^bkGJhKIF;zdzDxelYJt?@=|{SCg)W$jU!T|Glz
zF}&<*!;Rws*l6yb$L|JoQU1J1HD#yjOFD#U&Mm^vKPlwTM#l450)=rYsd*4-3#&m{
z-|;4+E%T(5!O0z+da4}lH^Lca>#Pw0G8%9aw4qT{o&U`Ipu5q{SmM*@K>H*f5Ng33
zy1ZSNkb*QmEaL?rP&YD{NwM2ma3>xCTT4~=Jfny;4MUN=ZX>bdBaJ;}=$u|8o%k99
z5omQO>zk~Na30Bwa9NT@q+wkel#%p@ALjctN#Bt}EPH1g)CC>+ww3|B<32N&ZDoGe
zZ19kzVcyCS-0o*H?+X`Rin9?e9g(?IBkqnyQjgBpA!bIAMx737v+|sxPzMVW{lB^!
z3peS$**qd2rVQ!Lce3Q`ZpDw(fv2xBN^3*t2qYGM>OWlf5t>NjNNabr(txEmg;Yf;
z%@&s=6}Kevm|bw2-1wbzT;MuxJY{T<oi>!n-*n21I^`M#CCDG#$PFdH8b5a-^iMf{
zYY_E#kfk*GWxz-uQsf<FD{n?at$Rj#T;58cKurtq_DJ`d@=m(T*}{lUc1B1y+b{Pr
zlFn-X$k;#p-A}XP|8lm5oB#E3u%StkmJf`8?ifm$Yuo1I?E`Agyy4-PRq|0P;inEV
zr#n<JbW|A_z{VKHC&mzXz-ScpJmtRvZ=PK}%r6L&Z8D&vX=f9~m>g-7rDkc3-laKU
z4}<q}m9$%Yk_M)fkw`M{T~;t9W!fGOiH?UWee;$WT+{<<Aewf(o0BXr<+Rf(yuizA
zM7CV7b8*{)->4dnz6E6-9IjxrJskbM&Kf0Yom-O^r7=0>%~Y0W`N=o=tfA+g3(X8#
z?1<y-v*52dCO$7K7xbHUgaSMrF8Zp20<sAIyzKN_;u%3Na*X)t;BGN4p2#av)22KI
z!pn3DyrS@M*}B>_DwSu+BeS{4JkJL_`dRvP=H3N?;%kk{nH*g&?own_$Ii?1G<+o=
zM&JMQ|MX8a7%E$NPnRfD{q(`pcx~*^d6I43yqb1j7}zZt(pD~C8gI1s;^|#_#}|Id
z;_Yv4R|n96tv72)#PhcD*nafU$6c?`T|SY)eleGPgl6g@2Mm<9$MC{AA|Jf_Ue7Dp
z(ZMm)>av#tJ=2cx$zylN^f4m9u+#R*LJXjE`WJFO4gTU=d+zXnUc!emC+!@qkB7u$
zr&YWc)`I1Q4jA6go?fgcsYX<O8$iAxXQ6B5tGbK!i3mj&Dw77fsqe{-@4|#HksE1w
z#=v!sha}aXu>&kqX4P@&wDGi>tT^Q<@K$D1cxjM2f?f>htUro;7nz*yj|SJSjc4Tg
zwRh^h@o_v}9A{?nT3eqs;Dt{8D5<HN@s{*(Jl?>-gesOK6#v|{>vi7Gtcfzv8rUl9
zK}6da2<h-PrNUHQP5!qliOQ$rzr{7kcr5tYm}AjLeizIGX~TMu@>pf_b4!!t-^p_r
zefV&g9I#4q-x#nGVWNt4utpUcRo#{m<}T(e)X5z|QJ?fa@vT8T%f)(MR$tLmhb#Xm
z%*H_{dBtNsC7vHO4wOehold=Z8DWS^@8#we&og))Z>JAxjFgn=rZbs??1(*#V#p-A
zLMMHdTk<kKuEbm6uJdFNUAb&4nS*k;^?YB#hr4ERSYDS7U>Crq&cYp-J6cS9jgi2L
zPH5t;4tK&8eV05@koW`V<YjE`<S1tEHE7#OXMyF=0_}xmRP<N!RzW1M#@CRzN{6yn
z$g{>Trt}OVi5Yn+<8gt)82)u6wxu6V4ZcmItkeZB|KKiomap-7C&1BDe~FvuzFU7<
zig(J%)zb2tLAyF3GJ$>?AH`dvApFsB;G4N@&-*#g0PPc5xX{&aUJ{*#sYJ?*lIX|b
z66YP62byxpiah9LktgM?v5OZ6zFx_*+Dc`1i!Tmf*sdx!v|R}QRrkrYic|Z@>CisJ
z4o%U0?%*O9m8WzL2!8r>@Z$jwO{8O(oeCZ0GgRqR8QIGFdVk1Yu40*DI|9tE6Z}Ab
zoiv_<Ok2|7!k<jYFWndpo4Yq@t2dJ744C{*IfXTG@`8ojv|P_AqegY=P>*M@=YM~r
z`*f!1R8y8RN)ZWCzH*9v{xHZW_Kue8#l`dBFM%M7e2HKv=m62Vb){Z!!suQM@ZuQy
zyvbrriBgrUt)BsQE(VtRrPD!^FViIq0J0>#b|IVfYU(Nc%F4gWUh>r=3XiP<@P~VW
z(K?dwH2=g2)r~FJ2OS@F8rndMJ}T{^Pa?%F3o3Z6t#?gO+N={@c@ETVz){gjGO&d>
z{;VH){__F<QgHGNpN4RW65V&l@`>95;~Xt2!!V4oz%bcA%fKIR!J?HcrAwrITI>{p
ze&u<}E&Sm*IQD&InCt{sX?9|GXvbkrK{7o8KH$)v^@6ygWnYvdejJ8+vqLk*I40f#
zXmyy@BKHg;zaPY?N6)j)Ov!J?a$!u|82*`hOvM4M^#iD+Uc8StCgL`*Z8R+zlRi=s
zc*q_g4ndo8@>3KX)Zbt3eU^5sFt?*1-HK55cSEJt>tS*dH?2&BcP@(6UVS4`Zo8dW
zdeZTZ2;=@70r3JuGyqIKnFGN9dMYC+ovuCYI&z<^ZwCvdRy+BOCMbc&*`Z?ot{$2Y
znNrYk*Oi_7XFvR`yAtd?xR*Wo_wL*cgTjEoq?4uGq_6VIsH4K76G&MjknEZ$iv+8u
zCoDn@8HOn*PyGBBKQF92Elzv1#R@JLb1W|}h6S^JQv%CSF_)^Yzx%tNlvxiRJnT0T
zd|{Eou#<=~w*P%-pE^^aQJ*^+g2HO(=NaKpHZC2&jf~#C=SEcri<rXx^5)IL;)rDQ
zdY%ttbqbB546e^pc1Nm4tGO5<lzxqu)Tyl9;GwguK&Ot^5w0p|ts;T9q44G|tY>*W
zOG_SH=xgMzsZrwiq2sqLwVh93R1aL%T{#!e=fJT1>(;x?@Ze`RqEQFq4}S7g+0|GW
zz80*+OOwPcM@PZPP!B`KJ3E2Mp*ZyWCacn=H{vsDLYJ4>^${|K=RUILyA=-A%ZJeF
z?aIyuBhaLC5~BI}KYaejywB0q`7Pck-YDNfJsk+cnZnozSEb@taSDAYu<|){laH|q
zZWsPmX(qj{;b^g>wi;>$=QgDL{nLO~TfKQgBX!Q!NSzg9N;^)PD4yMVoMHeoC;~6I
zv_t0}=wRWv>N<Be-!D1$v|dDQz*LKT@UP5;wrHtsya^4=bwA31cn?AwWq<tPNAZg0
zJ33jL<GVOKiEvSdyEG0ebk>TtZ(%t(a<|plh5j()T_Ima^%SIc+KX@RFphQtH#q9?
z%pI-K!NT+!=h4{Y^+k4VoJpHAy7k_I&NDe!OD7-?{Lf~GS2FdP96XMv^Kpz&&uB#6
z$*})HBppEL7doJeDGx69nHI4-(FoO97_(l45rK9F3Qp05n>rsq%>X9nTYUTN<4sm`
zNLaiBK3u0GL@(Z5G`yGH6?d~1<iUL>b>8o^$^LWSiSl8!7#8T5!_h8Amx1L%zGlq8
zv(p5g=X`aRuGfnt2V;<ofARU}#i2NcHVkwJ&aL)NM!|v)XS1^@s79837iku&>SP^Z
zkFNRzBG(VXwd38U-_eU9<-U2Et4l&7{Yqtd>qIjz8TxJ=Tgud)AsDGm9Xz2H0tfHF
zPrj`g((dh)GNn%*t?TR8s?*M_wdpig@~B^6Jl5;4s>Cr-`c9)s9dM`j$rah`?ys${
zAOm3P$)%tQPzUi8$HZ$H{d#=fLqTm8tDcH+b5b9hh`p}Y$Zo6mKT7(XmlHX#c=g=b
zOMx}dA3RAvbg{UQ5q#oJyp?ZI?PcAF<Lb|yo9SPibUZw?Jg1^wY$XvEC(qIRn}HIU
zCC8&<BMW(E@Y^XHPE-*wiqHZ-1+p{Gsbr&{_@O1f`gIh<A9+HK;`3B+lVKBtB&>B5
z3<mhAa7?RX8P$NGO|L@4%{^BcmzBDNcP_#Ttp#A19YhdPXK(MlBb(nG<ja^FL@*!_
z0@p0}mqlnV<92)!H>8HV8b?o%OW_D8lnCJK)u_XFPQ^nTuY?lbeM6HFSrlb)q;k4$
zCCWr6`7%m}!aSD_M#G~hDeZNN<6fpngynX|Ag4e&@>(NjzzCG(3ScgCQXa)_q!oe`
zAoJ~SVtjbd4JhcOtfR3=x1C1tp@dn)P%w{MxC66;$+%KWE3`~HQ%%b4sKz}oN-uaG
zq3X}3FRiwpH4>AC8c>%}<_gMa<J}_AwTib2Vuag4Y8}<7FBkaXKaCb&M!><)yir!t
zC>viq3+_f<fJo9@o|Quml`_GCIYyQ`l}Xv_>=_~grGSsO(vY7$^TpifH57Gl0kASz
zn&cPvrs=(Q7U@(UNBJ}Cj20=)CtTj)OHXi3TOb#Oy*PmxMe;9j7^T!;OnR~~vYh7_
zP>oL>ZAkk<r-cz*7jEF6kjMsaJg?sqJM9+T@;<N7o7a#6DIJ$lm(i8qE4Pt%02taT
zDsU9e$d!9}k~n-hgdeQA3iC>zwALWur`;J-2gMwIC?`eFh?CK%qnoM4BrH8olH0_~
zr$fQJ)`>RBUr$WR@ZMz`KN{~TCs%oy-{?%6Fg)xcfTw9YFxaVM!L6wUcyspD08e@9
zu1uu^T<-0|U2-WzoebS_DQT43W3qFX*K|z1Y{4JxhwtHT;HTha=qPie8~9$n+=+ql
z=)~Df|5$iWmOAZ;rsv2QIcnBJyvN9ycvTB%3ofHz+P4cN;gmS?)dLU%^WMpPXmo$P
zPRb=SmBr8oY(Puy;xn~+OiYTZ97I?4(+5d+>&u%p+}8&*9WB5hxDdnMw7fwRdNCMQ
zKnwiQ$$18M^}?!hCCt6=w%_YnT0L(05S6raht%a=5P-G?!lp~lrw)37cXAnv9Y94M
zf^cCdo~fXGRS_cOzq2|<kIbOl^cWAJ&^C@P)p=wm8C(sCFU{z=?G<7-{pbYwS_xed
ze4MK7$ORwEW6;_fZCc+BP##$81s$Mw`SO!l+7NwK+#@d?00n09T>XefscZWA;6c~u
z<LH?HbP!o&v>bxFU{2gAFH#oBk>Mz$%)@Z>0?qy;h{~1Sa7RpmzpxrWm1h7mPe!SZ
zwH?6(d;)h-OPL?d7x2rlrAVhiM%mbY79Lrjz*;7RuXkhlBTg#j&ek!o+BJN8=5=6~
zGXkn-BOQZu&RCGgHb@4@-~Q-Fg@^mwVL{IyW$2oqUI=3UQfvrL^4`R9rQD3hD+Dom
zN}>`qdF0LeRR$)F+tu2ScqLNLG>!;bid2SClk_ZM@KKKLJSX0xA!EfVZA$Q+>({3I
z<iP`ie>{lu-4dtk!JmaY`@QpN^5j<+PYB$;{PN2R-`BtXbqx%A&b_z$McVh@olj}c
zDDiRW<hgW=k;BjtN{i%kFYbwx4!8R*cl-sLwJW3#-I_jm^3XCVEF+OO;rYJxD3KS+
z0M8wAg%0ZG0Pfb@=pMY@&srl5hViM6c7G{zDeSZPvMONI+J=!)LjWxd;FX>t^<Ej|
z)ynTGvCKu21#V5F7T>Q>hm#hKEFNUVe|H>&zm%l27H-_Q-Ze;{e){Q&PqQn_?icf?
zdTb4q%#p`iUw)C#s^9BZav9MNA3vOqEF%%PpUp^9#{+jX=rP`@djLnkl$PIb0_)4j
zzOxEm>}7R)T;AG3<(DBE9IUYku|n!_BA3<oJQ|hX6t%)EAj7$%5K=ZYZT{d}dDf9B
zUeAf3P98>2r<iIdIx2Z=t(+n#*u>{u=h6?dHiRrjk%N`4s0Sx9%Ls$pitmUvvX0jH
zVbo3*@B%>$GVp7iiQ;rjGU(C<tPP_lz-JxQJNYh$`SH$!CoeB5V8O=o#Pji9)+DHd
zfvEGQJMhWDV$xIIQ$HLdvngkJT<l(Zk=-wN-z~K%7fnn>P}&Z#bNQ045jUN}Yo~1<
z24oIU>VR9a7rsS*A8RX(4BA3tZ9v)?q-D$N^9LRE{^r;JTKy7+M5b~^PS)3<ySCHz
zntV>~WmKc{ctO>RUBfSW8G)p4drY3R&fv`(lR;pN96UCSQNtj5Ogl0#1P^{3<ac3D
zxortLF7$w3igy~RbA*2TyWh2}U;ejWmhSkWqksEtXOx`nc`m#W@4fe429<Mav~pcK
zSh7_kFSyWAI$N+<kJs)mc(;Q5L9dZbRHujLmj$U#wQdVZ^3}=Cj<T%Z2wuCFcXI}l
z9g%1?-z}pPtnDHv`@CC$!BO#KcJNO%F$nUrd92^s&m!GZFCChe!_$(Mx9L&_%H2()
z&sjRK^wW2uap+2GMF-#${7QqQ4U#cBW&r7|5f~XnT{?5Qo*ah67^4Qg+1{HzlQ+%|
z9y@*h<yXH<d&;0bdB}5E-mrGV=*LX~T0aMOJ8#_cCwe(NAS%?1e;Z%Oer3nHzm(g7
zNdTYr5F8?3xYPCop%sojikkOO<Oz{*UEZb=FZ?&X!NQ-BT$Bj_y){?lDZtJ>!k32)
z$Hr}-7x=D6h!Wln(R2V5LPv#BZmCYv6c$FLK1z3QFA68jC$Ku_8VqR+YNy<cW?o0R
zK!jVpi>Bg|;=+E~My0?)+BA%c*3-D)1#bUyM_`?%)Ja_YPj2xnPjPKrUT^txLxmJE
z^pS=F!W;_BoeNs2buZ^`L{2d=PIw-a>HVTvAxIpYC~nFS-Ws2+hurxul2_+%Q6z+X
zZ$nz0Cgs)<e&|BDsyIR??(vJ}6BExnqj8&xSzeFwBJc<V4K82+k5FNmhvL8BRNo5d
zBEDXSl(F&*y_7XVd^8$aIpx!kqI4;i*kR>|*Bcxj=e&u?3^yj+b5$8RF>(VG&3k1L
zsWdg7^1ANBWT8+s+TFT{$IY%2ah!BJ`s6_y2h^Q@8rce685kCu%7;!Arf<k_OY_v3
z##`=D3c%6a4i(~lZf6#w;Z#m*d=uGcXn>C*<8Bp>k+6-wov(#^xyTW(o;V&jcQOOF
zUM*qK%V>J2ExM13Fy)%q#Zk`*W!S{Q59@XAvkqoFAfw2S_?b$Y@z6*QO>DW;rpYR}
zmXV=7ECtp2Hn@v<()Jkb08xIY*cs&p2*3x|=_;GExUF>`=n7bnqo<LqXPw}F;z@EB
zC3}v{`%MB}mDv=pZSr_%6ANCLm9FTmtkmP3%X3GIXjvYn)Z$~xIcXWXUA)h8gO)l$
zJfAZCCa+{_c2lLYyek%4TIuuHyjA$oJ>9)!8YuE8t&Y(k=W>p+Cr6`yY&Ugi87gG3
z6CJ4I>?od|OLXZJ`DY-(6YP^5oCXC3PqUc!RdDH`D|mOH6`1XKd7ZN?ibvqA=Bq3y
zY+%O!6s+jVa%CY#L=B-AxNbm@^>Ts1AQT?tn*Qb`t`R~{rTl&y2IBy~1$?GEJMc~a
zGaq0JZs@IEuDmUH0=qzN0L9w@CNiw8XkV-R>Jt3g<^|63aK1j*sWo|DWy)$N>1+SX
ztMgD+0utwM%kE$x?fhlxWN@hZdE_NP1}_7z8H|i>8h*noz>i#{Bu0U-+9chsE`xaW
zEPk3w^V@-94%U3~`e_HCJl#NJXkW~<GCG+4)|M;F!C!xlT{}DDks=@RtfT4MZu7qF
zCh_{^Wn|V-m(1F0_1VTFoN^^){<!j^zvup)zus-{#h<QX6PJSZ#c0sNQqZdaJC>;V
z-H;v^1cTuG8^t#fhz`(Ym=vf0N?}oM=VQh0o;{kfn7!8Wq^#8S@?hrdc|cge_cFD>
z;II>=!kW(5DCuEbk_6TCgAw&R8Rd6PL_7^f5zY=-fmO#<M&KDe8icxv3m001tqr~n
zh<pxZG$_J-CJW|@=dlJ?q0h(>0~}7~Yy<f^Z<uv62hZi1z<|1yoi~)2T^CMpMUSg#
zjMuWR1N;P}Fhy9F@$t0Zn0Q)2djH+`WA$FC0NKtc4W(m5WA$^Vwt{=l5DEQtki|Fk
z7%7@-|IrVBl)BPUMB&`YiHZb!DOhPW3NVZi4IMmJQne(H8*N6BQ$q>j2qgNsr%rM0
zFe?iiJqc_U@=`z+P>sUW2+{x(Ov$LfR!DyH2vm8c7AmG+Y7Ok>j+Gy_sV?!OVA~i{
z_rNhqX(yuri8``W#(AEYot(>ZwMMM8<iJ@k=J8Wu?|=N`7d1vqwVY2Qc0%oEpZ)N}
zXFvQbyE@+MB@V9Ed~ko>iYMw`_T)dwyf@|c?74GlLZOt2w%pJf7vs4G*Cz8xQ2hOQ
zogFGhu%`Fyim>pH_pCcD!fWM?r(q0|wlg#f994v=s|uhcmV0?Zf7kZA#x_0;&%%{?
z(~fhd2920_drH$}!xwP0Eu|##e>QDhdm?}4!MW9!4Cw%MmfdX>`4`keb96^hU?7X{
z=HqbXL2wfogo}@8SVK8HLNju&jl1)ZFs9pf7(lt8RoZN_syzHT7hK2-gHc_)m~_;M
zr(;x_rlDi*nKsjo^NwLcPTF}T&Tc+yt6pWT5FFLrE&bsKude*OIyzSQ@mGHBO3(?Y
zJ0#0;oALOJj>P)_p8se2{hf--ArH}l%iU?E^=<?f>&X}k*WY=so|*C{G$DiJ)wi+c
z1nR+=-l5A(F%T7S?1UEdRoiMHFtYY(g12%`1>U}l5sm)7>I|=bdOk4nEEd@Oh2Of!
z!SYPivyyb`@~5Ye28L%(W?GV`T6lACQ-b~n$KoX~CFAsZJdpJMTfyISgVUwKWFXN|
zeekK_o4j_R1g**^`6o_fl{dzHFZoP)U%i%oBn_o9lteEAUkBKUbJA?R6wxnmKr;pd
z9Psx-b}cb9(b<57j;($Q&!MUxk+%Iu@}hHTy+bcQKXLle$itjAY)#?0@zU{77?cMO
z>4h4+k)v0EiwwMd1>7f2+|C*w^`z+*A7@$77xekfOm~?+mJSW9X}gw_aqYac&I%6t
ziH8p#*Fcp@z1QQ7XV{^;u%_u9nn$)Aov)wJ-*hcuJbnz6{4dY&qU)vtXJT_r{mTYu
zp(2lk(esmTMf+x8RiT$9!_Z>hCk|q(Q)#64B2`)K=jU`H$~b;WFL|?su#~=0?LoGR
zLMFMIzLwWk-itJ(w2&%d!fBdD=VkPQL10k_>(42_7-)wKZ-~aEqiPzI+CuXs55ZB5
z-<VE;9Vz6z+5{m5(3~Y}F)dyS3%n^8kxkD~Fx}SFa;H9xs?9Zb4fVZju@zYSz87*e
zjjK}v!HtknUdn<`7mLPAVf1Zr9V5ckt)A-L>aG-o=BHvY@_HCK1jHCQZ<Q7lpt|KX
za=&r?M)73a=~#JJ#@#v7VH9uK9{eeNw9%l@S%W+t4u7f8wY`@I?xWm7>)_Pwxd@(j
zEhm}7+Ga`WJa`u#MhU3PISb`ay#d4L@<biV4<J)*WG8u*HHQItGv$ixXhPYI3qR5U
ztS#1*3>bB0XrPfX=*05^=xFfcRSp#4jWc4td+$yfE3e$_K)Mp6P+E)pbW$2=)5f^0
z5{WU4M<8%5T24%04nXsK)NZ7gx6^)%uuaa}p0v<S#>n1W!yj#EC^IyG^MqCV`aO4P
zC;b*5x~@)yLeVBDN&G_lw*9~pX|KE=FRdd;e947By~3b17y)O?(U!=0<j<B!9wu$E
z`L+TX0hA?tg|;JQhgYRx=%K@7K=)fQkzw9JxTr&0T6w30DMKA=u(1hG+mS{}zrb(h
zSFh*7weZM$u%dEGl%zsfqbFXsWgIR~Nck0BDp$Z$Qo#9j2v<q9m6f+0^=fxX3;oJa
z;nFGEzRm3)otID9M=(Rf7*l}>{xamTGS=NHfe}B)NJzV`yrZK^{r7JS2~&i4O;0!u
zfS&b|p7yOAYa1S9eG~dI0?4toarH6;PB@;-yOxOuI`Gt89gz1;TRx*JyoLz#p|k*z
zwxz;X>8wxZ{1PMdmpN(yrL;A&QhgO;Q3q*&*=e%qdk5v=pY7P>0_=<^Q#Aq~cj=Cw
zDX&uwjZZ9CwZY(HYIN@D6SUh*O@)2siMrsD;t%<$F0yJcXv;<E&+t*Ud>$fOtv(>=
z{24BE@0lDuPv6ry0U?G!j#J1d5O_lvQ%er5&pVUy+Xgad1K-GZWIQxSN7IiC#Ba0k
z*ns`+-MckVd6XD!`a%O;gMIyp!NH^5N985l4U8DM`2ThcP}Eu<bfp{hYcH}epN>>_
z4W%%#{Z-%>CsY>&asTGGzeD?pY4pGGXHuN*$e@B80g*Rl=Vz}f#1b(IM8#w@onQFY
zN^42x$s!ebc6K750m$5OBUaU+xK8%zvsXbn0<>?D-za^|keA>Sc8bUx>+QRDva8^E
zKc{^m);o(GGgyB*V<RJ0@;kAb?L?tC?&te6=Wd;w4t;+PK@$He<+K3>2W4oW;;U>i
zJi<`L6R412KaD9MWnmWOqwx%08XW;@ql}Q$`=Mbala8DaIeoArCJiNx=imM8XS0?d
zcuO;iGi9bMf@z<=x$Zl6?&i8v%XSHs{1ilY;G{uZjwi*!SAh_q2|8Z9oDuuMd=G*m
zSl$hDutrW--pXg_W88feuZ#{_hl|dg?+hOBS_^7f#%RIXhbwwvP_R6D&hO<Jaff#e
z8MF+J@|c#OOc^8jpiS#!Fi?h(kd9uBHgsh8(R$*YVh}v-$S$-`Uf=52jUw*+W#vAw
z)Gr!<Lia_7)d*_P<kc8Kfh}IE)2ZL;e|Q@KkWt_3FtV{8K-zQJWpMHJr5J42>Qypo
zV_ZDBZ_P-)k@36Vb(hBV>scp~lWH|`@M#0SF=Bm4JFDQX-M|MOHPR{`GL%bW=4K~Y
zKGKH9<fyVAMS&+qqm*zJf7(UmF6E4YG9z|MFokb&OG-&@qx{`2F@krw0?!K5aZJ=0
zAK5jMHAqIHZ>K#~+2#~PJbZNG;f$t2E1|r~-N-ZyZQIIIaZ@wdM&p2&a%MzH#*M6R
z<%G>A*^Qv>;k&!gHBWgDXHd4C1YZ0_M|5Jmn-b$u)M@f;P_W=%o(J`(@*=kKl65j#
zgElbh5E+rqusE5#=g*I<qqCqsc;n0P#-K$<Kz3ch(Yp+g>`=fn&n|GjiN$gHRoX#l
zFyGN?>uK5k=m1=$V+v#VlM9Bd^I$B7rVrJ3F4HlWFK6+6<jXD~UY$Su{*R@b!^^Bw
z@!0@reR-H{by_xGyR8=~xW70RzNH=>Wypu3?feYh-piRSzJUAFAOE!a>gu(tO-l#y
zI{*F;zc0VO`ReN$+!kFAZ5Z6945x3XJZ>2BJmBq-w_=Q2XF0W9Bwa|X!w6XUm-g~9
zW$_w_{w8OY;N4u~o#35o-%*`-A>NXFIXSX=<J}uIRIWuHd>`kn(28C(&;r9u;bhlP
zc1kh!=tHuwn{`^RUvE<s!GVl;kB)QkBws2#JK2?o?_`afqPfnQUO<aZ$Az}MFJx!D
zz0uD}XR3)V5#M*wkLY5+2$pN1y*A*aRGtNL;ar+i`4)`yp;Rv3k8}A#P&{+_aG;ZQ
z@8_F1c7Ofym%qrGz?=CF*jL>p^}z@4r%$|5Ptp(aEZqF^i@?qRakk9k(!`FhfBeUP
z?Abc+eUNpB!JP+_{PMm&&ly0szcL$<(~iRzJBWCVyRjj9k*s%GGwGMEQ`Z{QM_nbf
zHF_9X@6&hx&=H3VQx9JeR+;=uBhq<O(iH6JlQWmDLCS1?NaCh%I-`cM8jXTS8Bs{C
z(GBXUB97&U8Rj+x4divsu`tK30qO*W$F|$0{wo>Lv*2`=%Bzc5-KkgS!UEI*S79S$
zI$aBNY6nNKQc;ButQ0H(UmY(a#QPd#tu-PCpWrOLKptG`vVCebvgt^((mMu`(SxJY
z2|gEIU7dMe4N3<~kr*M^QK1nNX1F+Hh_Dh`xD$3l2-X^axj&8&P%Nv{H`O9d-OPbk
z2SNdyx&l9Rr7(jB<<s|+&U^i6T9L{hBMgTc=^zhB>}Z<C0=6zR%I}_HTe{D*8DT}6
zj(XjbX6<NTC^m1>i--T}{K?dGfbm2{5x|FAKP8l=l)eU|KJD)`i?;zLks(*gRfhk)
zVN6~I(_E9@oH)E0gstDaXyBkUj81GB<i%Lt7b7W3>_$9S6uoscMqb@!i$`ggjFJ=g
z<`V-tKDd;=ff?;O+6)}z5Ih@jmES7vOxLaTAcP5bmb_M(YiC<qJLH8laEoEzQBZ!g
zgAWZ_$21?6{h2Uy*+}x8_d}(?5anYuY1F5azZhff?YMHucR(qv$PRhbo&~Z+X~7R|
z(U>A-G)qT*@x6*Y@6`p{=E19kcGfA6TnJ5@J5M1#wVf*?v^Vl(FoH>aWF9@$Wwfk=
zmPT9gt|Rf>wUhq#oOiaPx-*pFgrDlIJSC0thfm`rBg5LLGQ69CP&%9FCte3>dxcOl
zh#KRwQjzlM5AV@pBsX+OS?WVe{Pawx==eO-5vJnQU0oAeTNCyYf6^(6hqp3$S7)TI
z!yA3t!BBP7&>3w7_106N-u#vy45offciy{y@5IHu@rc-Y_}OQl*0{PBqg-ce`~TBS
zOL70#^A{KlEE!H<hUX!zHy)uw|MwO=gNv=od7Xea(@vaJ%vkTB7OetTUP@Fr(S<ib
z8u;@7*m=fl=Ywx#r~|Hej4ofCk4ZY(fNs8UF{igizAr@|n?^HmC9f7=tI7qIk*lMQ
zA1-xk<h?^WrISWk7O@%-FpTKtM_Js?5PkCm9%H!oo`v>cV>pwu!X7!M<8}CSj{ePn
z=En6gs(iqdLCrucOw-KTM$z(d>Z7x7f1QImvwqEo?vU&49KQHoKGf=9&A0Oz61c)s
zV65R0!-pK}D|wPCi-8v`@T;Mc@++J9{Z`5me<O5vqy)Wg0AV{b92jLTx>h@Za|717
zwx1I@zfGNfKO7n%Eg*!KZ$`PBvWxPoAre3+hmK@h*|8e(+pjjFD#>96#CwHG6sL=D
zX(*I<uULx!)(Hv+vAVc|mLk(R>r7s}{i4jdcrNpUDc7h9%r1(hZn?l$felfNKrpwK
zS%kk0#66J0W0Z;oJ2PLW?xiqF(EL`Gu<B7p6>!UgiiM-uW)?x~R7e!DulJg3z7hta
z_xi%3w1n_s+-J^>;6AI-lXx8v3sYlXMhK-#2*LL>`S73$Cg};6i}F=2TB*yIB)f_{
z4b68M8X0_xSKezh8U#a$pu2k3LUz;mW{rkU!Ra7$7z_f7$>6UMEk(0B5b-@}Gcfcy
zbudI`E;$``;=u)0+lqNGENZS9l6RDd&d&l{z;a7!%;=$B`1>H*VW^OiBXivpMlS!h
z`YVJg%mYJ@2Lk;l>KdF;^elWk8LZo&L}$b~9jP;0@K3%gk;4ap+33U>FWm_f9`V3E
z%<mc;sR;~2PpjOmcVT27Sq9)6B^^Vh(@gVDI@6fi1Vz?Md&zIcO<tC!)3W>>MLlh7
zo{KMfPV)Xdq}AaAf|MxRef`cou-j=-r5BksUC0o*e(gr+7M_J)=*SST%^P2}MZBJl
zehe=BWq9-2kwxvH9Y)$9Pl&YE6tq(fEb_0mMy>=hMHZ|D0FSc7X%A#nx#VHxB`5N_
z$gH*n2K*K(HDo7*uF)QzI8T9*&S2tgxRr08FkVt<%U*4Cm0J%-@K*-IW$=n@MOT5X
zuIW5?sww)B(J?%63&--Cd<_q0`biy>Jp9Yc+|xo*#Y`iU{e_#Ib){`QR8CpO8@O`8
zYXfAaAKol|Gcc+LVqE8NJ|5INomyx@#v)JPGo|R4H6TW}2cP5staRp|e)hxA?Zk<X
zKK?km;Yq);t;6~J%eT6N(eu;j3?t+vlge+KEgejmTZ_QxHz1h$QO5U8z+<tbCY^<e
zmHAgE=^9hqsW?0ls#o&fKn;)kJMT{3o%ev$2I&&sEi~c@vUAd%zH@#0$tUfMUuFt|
zx8cNrFRIi3^avKak1kpF0+)Atg@0t^?5-)nSz|*+b-)UEcA7pJ-AOjU>Au>Aj{dk2
zQeB$1U?&}Uq?5=eBh8wQ)klE^P3t)$H<=n`IDT>Siw;_OtW3EuLhKa#!4E#{V8Fnd
z(Tu_tsS0*xqKUSsZ!E7;rgb9>K;Hz?zUj8z^cB~xzE`=g*S9*xVRibbM<;Ic+V9>e
z{s+(TFR<}YSJ#ASC7xQpA2z-ni_TQ}G;BnZ(cLsbM`X2bW^vO&8i|8)EFw%d*t8l`
zGE~9K4H3f6%JlAK4#z=k4aasr%0>AWQJU-1p})zriMVz?VVZ_6V4yhO0cFUe4x=q~
z*jsMo%0-6&4mLwT3PcBGOPwIEt&tTh5y>m4(>wP~Wu;MWzu6v17SfC`U}%&OA%_D7
zyAuooR6{0AGa9jVu0ojx3a*DSEObJuA%qUPa))0>yKP|kSXBgomju4>7M1{ZgqH%q
zKseDzozj7yW$eyqFyBr&Nqgt+z(%>_vxe#u=`~a=D2xFmp7m|yuZ(rEcy&MwGVqU)
z;sCC=<^zAd2zX?K5<I2R*?^70z7V?L1!EA8EX=(cBWX=BF$&(!oTX9M+AXA`(UAMG
zxODcO@ftosxHvPA(V0Pp20BYLaqfilj(mrcJUac|(>YZ)JlPHJ;PyC;o;*6!1OM%y
zAQC3v_*aZbDaZLhTa?58EJ!qhYUdN0pfOqp@#L^0fm}0q#_&^yKskevEEd+lMFB8u
z7#rZ|0bo&Nrmq;4MjH$;b(zQRv<&&qGkBM$Bo%6{p*x=G)EEX)=&6tJdWJ6^;twhB
zmt2D{`j9E@m5lxWl>KM>CdZZMca}xL5QYQ^8nD?-a<==LS!whNr1$?|C0%kWX<F^3
z=>ZyepzMA>zqqpivS%J_R^>nQ&J1(n_&8w_KL|Y=8|H}2dGGn>pQ`hGZ1FgJ2e0VL
zk@Vl8onXV$bjQ&j2~lOBpE~dbp41Nc{RWrq>AlzU^NdW4Z*IIT4d?ltcSNzwDLEf`
zgS#2O(qBAfQ!`~y`1tX)jD*8PUuAv~e&I)5CM=?(13uu&flgNlU-=Hk`ScFc89meN
zXFanLg}g2-pR*AIZE5Q|YS9<cIo3yxq2m}7&V^H3wBy2^E1Gg*ouaw;K?DOHY>sHS
z#e`UJ{JxHi&X=}|#J*<;qI^};arB#e#z^|-&)c}m)j#xVR>PM?gS6AUClS^E=O6xI
zhTO9=_on#;)$y~R|9tTJh@Ots_3_SQT+bwGwW<$q%g@2#<dJLo``5qvP0t(poID)E
zy2w-=QP01$R5?C=63@wppOoiBKaL1KB=2yEWQG_>?$iA~y3jkr4%2N#K<uaVhm18n
zsH1^L@H^d&*5(;qLp==kd4BAbGV%*X^T6ol&~=yopv||!J$=DG%+bh65GiqwXPtBc
zIRhKJA(~{4)WKts1q&{ckMIBBy@uc4y7QBtv?V}rkxB5*&X_k6JSOgof^g7Hs^8?@
zc%4lZsgWKZ;~jJ$cj|us{r4w-b8srhrxd`WGX~d&!@x{_MaMq=$Y)RD+2+Y`yx2T8
z+tm~SJO<BXhwkN&9sa9xb#p|~>(7pxoZT*a^=D29M!@Wx4IcaqWzQ&*W)H>$XT0{6
zN2fQlF-A->LNS6FqKIKwfDy3#Q5g)x8*lXOh=7L_y~$^99fQwE5x&QH&6FtwA~q!-
zxo7k)<-@tqV1$3IF*1T|%v28^vJf7#B7pbGb>82;(qE%O8BXuOGZyo^{O*-SkwCt-
zVN^`INJ7S%w@v6>ja?HY<*aS%UB_HG26_P>*K5{eGR}{{W{R)=7<j{72^`_`#nsCR
z@%}o-1i?_IDNuE7yu4?e5N%MGx?M5cq9|x~J24&H^Ojwm#*^PFtf=R2g^`s6gn(0k
z-h)3H42NAmG+Vy<KFkC5HQB9Js6CVrT8k8b&5pP)qJP?CLSnp^0O>W@_0nnz9yvqs
zgTjaQGBM<%o<E8>#2c11>wW#9XWH$Hv6fB0h_^5YLvr;@N2}mvwDoQu_6134C?{o7
z4De;>2QPk_(V!f8^YYp`@%RjkVZ4R;`g1pBzaB*OgZa?5p@wj5wSwIU@ulO*W8v|P
zyIy}~Uf*${EUbRAGHlDHbB0Zvw&DR#*dm+Y?o7eLFCIA#N7vv#1~6F)_TzAHe!@v7
zC15y-+<-3?)mZW|dgjKE=O?XuzMS2=?SzUyjuY3u8DCQBbsB1)-r_s)hl$LVbu(7a
zp65VUzTH7hex}ShHH->;&nSd!b;I=qyDM13#FQrfThz0185ixGGOg}s<hA!?iR^0|
z8Mad>eu8`QM~QNv@tgb?-L>c4ci){O@}nRAXa?oi%{x&B{s!BxDr?L4<1P8{g9FUq
z;KX?DP#*npwF!>tmlL6?ozVe<p_#~#5cV8)^ob7F1WEblP*fIqvZeQwzcwWc42LwF
zkkLQ9fI!c6!zQH@t>Y|!2Y80cUVW4<iEbPy^X3><8I|?^kA5^Gf!=%|{KLJ5d+&A9
z&}O^0cJKFT%hSHYm5z91CsyBR5lQvklJDI4+0TA9c;EZUUkzXHfAYhkwhhHZ1GI3`
zH_i)3I&)U(e4Jr@oNW9)-SJU9Xu89MZ+e>}Zl0E51NFfxgNE-6e~@K{?T{rt3D#%N
zbY}6Cjt5M!6QXmqvxIZTKRXWEC!gARx9D)r2AXgN;dS@nlTO0xq7PQ#ywf}*PV83Y
zq#MmU!ha6|&|I}c9QRckjLP<kb76kP^XJXq%SkkBYnNL1vF8<8hI=%4y$%uw<k_<~
zie~+BwEtc6^vp}MCkdPT!w-JA;S@YiCOH-=;z&Mxb{0~wQ=+l(HC-PM<0)-YzPj*?
z3CwiVi8xyU<pYS?&>63uPEZ}m$LLhY(#pKxCg0@u>9ckrkB(@gqslq5>c*-X!(Cw9
z4&cebA4|7YH9o9Y@%Rsc>;Ov-n2*)t9bQ4;erww$eCPFl?Qhx2<eww>&!M*S|Mteq
zQaL^EJppv~Xo^=ESDnDRm|_J*EWOD&2v2RCazgu<fsrPxn{$!i&700EBM`hvMvKB$
zCSp1#%n?X_L^&ToOV}co->=`sh$)^6CXUYWB#+^C&rmHtCcY^t$_v51$my8lM)3x(
z4aF9MjBu13k9QFoN~+&yW_ZYm$le=MH75h%r;sW~+3Irzm-44rA~wHLIh-SVhD^C!
zIPW@RP&zx8GB$Aeh2Q1lIeRb9W}PSe-~H~p`$!%`#IRHPR~S?1+W4Wvm@A`<$+c%~
zjKL{G#vSbH5G6XhvbLRf>20|7-HgQdnn?Cu^HG%Hw2=?v44+>;Pl3f37i`-Nw=!Rk
z$sIoah7P>%Df05@5qdT!H(EOwIL4>F``2!KHwT&Ek&5`%EObs7!{Z>JB|Anp+)x^L
z5&j9;MF&>`|5_0)%26Z-?Z8a&JWd%d%v@da9e%3Lgk?@bblB44U_H@-Gqyhk9{n$8
z9BoBa;CQqRFLO$Rf7e0%Z~v^m+M#1Wye&<?=HE<V2N(K)S8wJ)d~QNFdihAA@bx~#
zg=cQXn~;|pX}3^A78$}Ead4rT6<g$PJ=YuF;RycGgrX$tlVxr9dtcu^g-Z)&@TiZX
z10P<;Tk`oWol~9YjdlmON56P}o$$rTf`M!vS{}zjt#jNemp<5i^~h@|)m6Fjqt7LS
z6gqgoGNYva?mziBJ{bMsIXFdIGmgIHAurvt_D4763i~qvC+vTQjXhVd^ggH;bJuQw
z*5P}F%nIq}Oi$sS@IL&H_n&6af(8G|4=A0$tqvL6!Cwyn+o@b_-cP^7)ublgJv`A0
zioPhWOfs#PEW(2hWW{dGaL?E=OxkhrIVS>+R(RD}W)6>Z@aiLfbMVpfM0DZR<WO=%
z$2_g=kBTUAHtydA*})rq3~!t*x`ce7GagnUT)@fjRy8JduATNB>f!x+brJ#50q&=5
z1@Wv8<*maSjM~V#{oSj-n`21-X|E`;_E|yn?QefK80+u`gN~Z#vunYQhL&vyQ2ZG!
ztII-wr%yyu+CYq>7Cs#u0`!C%(cf2Vw~4^7ePzl2*)81gmWb&8{6GJXgMrSr+T~Au
z>VS^@mw*1}Enon%_Ptj}&W4swpF#l-c7QVx&F48T_~SfxJ!HZO+d#3j=F8Z1S4WK_
zL4K57KQ0;`8;tf&<$_~+Yn?HjrJ{ocDb8@Kmf~}a?6@KS{P@P-ub&%N`L7?C8S|FH
z_56{%AGgN%wR*hoILZ231`UUi8O(lDPSR2;E28(tn%&!Bj$p`w$qY#BnxH#)m6Z9H
z88i9yqzUlnG5lvm7la6fFO^TIoR97uV*oiDN=e~jqV>AEzHxId^!&hxSjwi18Dxf;
zL9~=fd6$yYj-%+Tqi?2gzY%87^_n8w)AdkmbJ3$q887lt+w_>+OV8ssx$c`z$RYM&
z95^IYl=L<!3m5W%*MV6GCVbD^tb9!F!*kjrtua5(c-+Pl)kUEb8ofx-5MDP0WL^o6
z`fq>!?>0&B&AUyog|ir(;C&~j1x%a)IKhkyxn@Wmp8qk0^@ny4eNyjMy=H53K-5FI
zZLBh70nf@JbTx8{Oiu_LqhI)soKsXnrrTR3xG6o#ivqHY@M{=qTRkAo>QO+^wDK1o
zzwsvr>y?CNy}fWo5aAaMwc|2CXrSE=O2==}A@{=8@OLi-A=)P-FFzw99C<mvXX$aI
zsZERa<Xl+tyLu9IJi|#<E<7Ct8=u8Lc<t~B1xukGS`VPV_yb1}h0oy6A^(orPmv#f
z5i#QEA0xu4qio-6-w8dxwJ9ZY?Y6pQgU0O>E6Rt`WDntB(Z?z!duTvz@SAqvxvPG#
z1AaZK8PUx-X#fB~07*naREwh*q@RAc!72|tJ*f=WIb(H57=E(ryqpm2QttRL8ZEL@
zy0yJ>I0qa>!-ilY%dSpv;jR9x4EjLZ(HBfxLL4t2opBtT>808)Ki*U?dd~@op6J9G
z029Tp9H-yvlOFzD&;6e}AwAv!7Pt4x_xMT9e&rrLW9TX45dM^>J)9k%et;J;SIOaB
z$14Jba(P!+1$yTh{dpaDz6VElG%&*`gIP(O0y5(}G$zT?wQuIA***5(r56@e=)K7B
z4D#B0@T+}3p}|3Ce9iU4f9i+R+=9Wc4eR8D&rPN;%~!R@?xy!UFFcVYGW7=hm&T-2
zG$NbQJbKb_e6&PsZQ{&T)$lTY>fLA_Zo0{xCQtEN{E^BT4i0UOjnD~J_8f}p<t+W(
z-~HVjRXPwZ-hR6wlG40h<YU9Z=_bP$N2k`l=G1UF-e|6dvUO5k>X&qrp;s#ft@^Xd
z>)4hPJ^$c$MViy8zxd_9P7iB?PVo6M_ZN5m>7V{Fr>XU7(Z*bm_ul)-78{t9iWYEk
zIl6{s(D>`)#uGh|4!_;2bMfX|r{n*vhMs4uLQK`Br`V`7odW-nVa3r~BEsy5jz9e3
z8+<r<S3aJ@%>RXBkH7J*;MO^|d@aCsj5y#|>8W@G+n@7XN=&PwE8WSQd<WFL<p^+J
zNCb1=%AO%k8GqHb)boxBkasuvIG=&Pfl=LrP$Px%7~;<hxncx8bM>!B-g^oGBT{CF
zZnBu|n%^w6_Vb_rd`6ZcR@cLX48nN4Q%>sd#Ke0aalEIPly3&Ia@7x>DTa)Xc@Z2R
z0zAWqNJTi#zhel#5~>fr;kRb#scH|8n4<InJ_J?An6dgEzHug0aR$IJr5SuajXpzF
z8f}^K={aU{)QTO$zq`^i3RWDs>fu=Nu~7F77h|ZNFFKK7)zcwN|D4CIZbHJ~znKCe
zcpHKW|Jy60_r_Qm0U=3noQO#iU=z@VP{Ykad7PVslRiSHTC?5&g(f_E!ibGs*EY*!
z(bKZJC{olUg;iZ|7haz;**$tW9y}<7oN{F#UoG9UU;_^&V*UkTH-zxVFVQf*<1|G_
zb9YAT%0iC^(XvXa!z!$JB!_howUo&<2j~GsSe?&)_``UhA%X7o!hYK(QbO{-`~3%_
zJx6N>ulMS}17Gy*kI(9LJxKu>Dk8j}HmjZk;zU<`1U6&-wp-q_>Ss`TlZYH4!Cwmz
zE59Befxp+vweelA7I~>nV3?xG!N)7x$VywIlhs@AzyDVovUvNQ@1*eh221>?=W}{7
z;XS%nT6ol(@|zP9FZU1pl#sE0p<ML9cjxzkearbvt4IH`S6n)dg_RxTmr<ub@H6?l
zhG%pDDAjKjvL{;<^I0C)gl%Ow<*82cvuH)krA+E;&SvmBR&My^tm3cDvnVge7H@lI
zSOw1+s?#n`D*d3XItt3yK0Q;8+#k&0D(r4=1-w8fNXr0NK&QXK29q-Vc8^cK?j;(&
zV<=8%LYZVhZ~7S4<K*e-`>xJ*EA6Tc^cvcVP_Zv>=H$brAx-Vw5M)slG@`49*YvM{
z<B4EWp2*nqWMYwlm0joVs{<3A#IOqWGh)B^#V-bzVOD(ozy5Fk%X;R9+?`|>&dD4a
zlkGn>@5(#_%X&HNzxPRmaBJQfn!Vmt9y~Q(3YJBD46U#+J%^K5?p|~8GL~8q(vD0v
zNtZr~zGPfGO<rCnbQEvyVD$4`WE%fmTl3Ek>p6USe@=20@1b$DoqnocG-F=9|I`2U
z|Ct<#NZ^;(8^!_uYfS~%wWZ!{k`LF3*jULq9n>>C{L?1Iiga%HF1(7482V7(B4!Qc
z>IBleI%ISlo$#WL?EhFq1Ydmj-S?&+KCUwkM~1P^XG?x|=db&u#E%+^!)xdaZn%A#
zZK6ZQY_$;nYR-O+E7A)dot=^Q>R_*Dei>qhALXuWe3d?;N3G1{pt28mAHUFD=*-sm
z(4kc{Wjtk8&JGoJ+%A=$tX{uN_&cwA>i;hN@nZ*e@z3FJsr^5h_SADgyLzmlaeZle
z7E*?U-uq`v`92}L5K%#pcfLonqF1KIBNhc|T$j)?=z75vmr&(cS-IVhxoS5kd>`b^
zl-+k4>wc?fhI+wt^;Ez193lKLCb-voj>#x+h{X`D1REnAN5n+JulL$G%v`N&f0tnN
z%zT*PMBB6ojI{xwltsCIS^3&Gj6qpxF(l?AwAwi*Zf(@g+EE#IO9S4<)MQ%S>QFv<
z`U36+H`wQ}g-g#yi_%_Bh&<|pKLSK46EKboSl)51SMaTjv?X5zmEhfZRu4J{fD*i%
zBKePT38em+q$-rD9iwb@?O#H6FBy^dMarZlG}Rv*4Vj$KECD*3N`b*X7Q&bmHaMLe
zBXB#xCta+z-rN*j1yITih297}@ahTajCA}5tmnDocyV6@?Hxt%<Ze!9Wu5X*Z~T>D
z)58%x11Ee_9Fc6wmm{GksciJ+Xu~bRebNx4Z@zG}85q$7Q4#b2;~7go#_sw(2PDNf
z9;z-pevWa|xedmn%yjU>S;yHx@W#tSdx>r#Qh4z_n=>TWxIBj~cquF^8T5|u$Nii}
z1i0Vkf_QSqYaDkR_8I2#XrFq=Yq#Iy(E(DuKn?DFnp}yMjfUi_w`;p4uUigYezF7S
z-ZMtV{lCe(U^U!`{`g$o<jGkj!@XYf>3OT*$OPS?9DIOpCg(u_euip>A{|4K(hcyW
zJmsRt?#qj2jJ0`CCa7*wy!Rc9Bu`cQ_b-ESWbM#piRMh=i*w!wGxCt(;GVn-Z?_e|
zJwp?aRB6(eH2E~7gf5KWZHPGM0xl*O!Sa${mUE!*nR)<dW{X34CN)2A`)?J=cmJNr
zp7>rw$s9W!vLc8@Up~ut3j3SbX}*O{%c3#iiZf~sh!5@QpxH&3Ga^cjMx2nzk_Z=}
z)86W<_TkF?9t8(nNOPU^ws&}tgXSY{oJe*6+*ZBCIN1j+JC5_ge{?D1gJ2s?Y-UzH
z@YMf3L;ppNhF`b?k5Udzw>ezDKkIAJIybL<{OQ5T<eWo&<_o^_dEdl|Z=QUoZ~CO~
zII{QezrBdXgC+&5|6w!}UAfoS<u_cK-cdgpG;x@d^R)BJaPE@;zQ_%J>?OT<LiFpL
z_{p&OisqVdLXVy~{B~=bM%E7aJB+Eg=zPParvbC~@T)GJhh1&S;eCI`tKo&Ss8dHa
z=vu>&<lfM`D6)yp<mWfP`t{0G)Id;92WSo~*RS|&0NqlJ9jb`1GVVx$Sj)KrP7zn3
zn~2xTD+z#|lo^Cm1ag=NW$p(?CeReYXU4(%#>Q1P<nwAb5vM$p%1T2ZFq+N&G$L$*
z&U(gX0XHG%^}NR>b{e0ii1l(Xx{Zkb`fq-^&CYB@g~<%vP#~Nca1ov<*RCEMxKl_s
zrBA3C63d{pN1`#Zw85gxtF460J4R?}D#HP%es3QRJ>AkFn%)-g{KF4ETn~*WX?WtA
zT!eug(!zl<u48rfgR*j}363UMrIlVjOM$`lxOzs%(lGdX3EE8wtA{Z6VEHzur82?c
z+2z^7ndA8{@T^DygMY?|(Um8!Y)Z*WlviR>_lvCb{*Az&!&%<Dg^bz5z-+IX_{2>|
zAY}kg+2GT&5Ei!oijXYF#-UBxqn$peWpdf~o;PmW{YM}FVe>QWG)c6!Girik-VNTP
z6w!|N|ExMTo?JPP6QH|!L=+Idr6|-x7M=@ncTj#0QpB&ne(GkZFudi8WaTG+{gcs&
zK%&>5%wwuf(E*(B;fH^ia(YAbtg`j&@z?2H=wTd3(eJOy2R8TGIb;orfBOgR$>NyU
z-@aFK@0IfO{5XE$EZ)<m2ag(lD5CW4cfJ?RqED0M<_Oh(^P4D{%v)`(#~R&SjTLL?
zh7cPw#=FL@zewh=C8fyEVW^Mu)I)SmL0!k@H3S3ag9As2&S4Cv+Gi6ixFF}u%5e}x
z&CWL2*QznRe=k13v+7;lwG+N|Y`R^90557Mr}H<@elw@}c{p|8ZFL%Mc&$3O`)27y
z`yS<Bn{;QHq;U5pG3StTx~CU2W^~LVisjcqL1P^g$G-kt+sXZYhquuuV*^GfTF?h`
zHcI=v<@9KMzpwT>7<26u+29;J{fgs*KMwC!;?lidCrdl8Ui4i&P<!##rDqQR#Md|4
zoYi1aIJzR<)9M$ke(pP2piVZ>ievrcan5o)ftLNR9rs$A`MX6w@Hjk*kifY=!$UfK
zk?87Eo&yhU?q9SXT~}5wmhTuqe`DevCY!<H6ZGSyaOJ~#2WOR8Svo>L`qBHFBW6eu
z9z}8)>|ggWL+w{Jcr74+b54_oXyE`E96L8u65h`BOlUuND;+(*jW0`Q!NzvUjgKnk
z#Yd+s(MUQ&6hst>Q*ey53HdghJfAkeQ|#Aecjz`YPhnr?M9^sulg|fFE9>(d^w6PQ
z^wlSyoCwJ95Y1nE{wVqT^v>V^!#{55%hFkVaCku-CL?pu<Ez>{zapJCh{G4$^uOQU
z_0)@b4B;i6tlf8NH^LigqVwrsd_!(ueUW`o{?q1@L~FJTkCF!w$A5mcP3|5v4WJDV
zVOz&f6?CZU#l!S1?Q)I~pisOco1^#rIfa}Dd%D|G$W7F)3>s7V%ac5{R$=coSVJ(u
z1hLfxFY@TsrmS9dTlgmcmgWYgv=qjwD2;93hYUS|2&h+S9MD$;HBWx;dCG<iC?Z1e
zB4il9r|5hOf&+n3)WvCWU|w~duxIV4hoA>d0S;}|FTL~nJ>ix&ByYpD@+<p=9+6ZW
zoDRxe*%+O$8Z#AzxX$&aGV02n&;}EugusXb4;Pv*e1Mk$e;!jDyoamO&6rh(-kLNN
zyMqF7!jC%ETf8yW@|Q+^hboH1#0Mu}_It{<YXuB%Z+v<}!Nz~}F4wy*MEgc(<e1S_
zMvaIEqIHD6ciVuBbGm=p2&K0MmLfXKTf3q+MXLuym?j(5KOA(`9=(%SGZ+)Plmt4X
zEtxWgd|K3x%0Kk0uJSoH*17U3tLw9(AI~S9hnLWoARM07Zg?cb6i9qiG4N@~!tz$-
z;wNnu5y2M}n%x=o)}|E7cku!vDXku~Xo;R21BI8Qz1n$0&0zkwe<nd%bDDL7alny@
z+T|pFa603Wv3*cjpJQmAO0PGBX+DZJjxO+3sY3XU3s-n}VexQ7nO$<ULMlj}V{C5G
zKbbW^CHOA<;1$a3a`u!_wJG^S2k_!$__>TcJbhC>=7dzvb?kjVMC!ItAqDT_VDMTw
zwQTf3FYqiX5??ESutb-;?NkXi4yI?)aX=U=%tyA#h|QaPVq!9ie)y!s=zZD>cVQwB
zu-e@mL%f8Zhac)q(haqHpMp;|M1thKhHT+Lolngrim#ui!zSPNZWCvR$2tR?%V0&P
zQm>xrrRri#hRfbm+6@>0x%V?D<BPl7d9u1ElYy;GMeWue`5a&iCOT21LvMdF>Q(Qi
zci=f3X}h@|boBk46~jgf0K?Hi^iamK=jg!}nAlCu23r@YwfEWt5&xkHN6~P=JFWJx
zs)b?uL&FsC`Y=Zy&SrFL-^H5`AFlPa_waIHjHY<ze)OfE?>#w@7rbK0?{r}}B)b(E
zK!B?KH{JdsS<&uizP#C?Y`Vb;IBk5;q}bPrlGw81?LInawHAl>wK~NQTD^H3!Qctb
z8GU8k&QLLXcE67H-SohhpB?+5o)_>HTy(J&l^aTr2f%_RuH+J}+*7y8-QoV_Z%=P$
z>H2?;v2wVt4)Rk`rRrZ9W0mYVyD2Ffe4HnFpFWKDR!WX!QR3PAo_odzczG;u&u?|U
z+^lQ@KZH{-6k!6<`O=}N1c-GgG!T=o^IIr$%-nkoH?DkMRZC^AjHswhWgI}IS}7W&
zMU)YyoiO#Al7napg2A+*l^&E25eWyL<Me2GmX<=AsAVGQyG@>aKTr9`owg+YRlDl(
z(EgM&vEhftb*D7IQeB%It0|bbQ;&)u4F@pQ;q?gNt~_hVns*cn#XQ53fGS%LTuuh*
z2=3|qiYRE)N5P;cY#$M0nZ}u=*H(wJI4XLb;JBEMp*=>UYjXz%&iz}I;+Af^If5aW
zri?~y6pN%&$T=2mEv*?(n>|esDG7BGa`)=K*Xbb0!i;(xq74qPUS6^Hn;cdf^x{5h
zQfd|5`SFkcs!&_=Me6l3ZucK?E=qg#kik_RpTTzUq;b~%dskb{_Q${+!Y34QX`j^F
z8a?{+Fll|Y=)`L+BYmeWo8SN8_l9rfDT{L?`e7B29;_Y%g<~Fpc@6kt?Qbfy9-!Ib
zao^zqI1D$Nlb~l9-GuDF^r5<*-~E`Qa<+t)_s2i|$>@EC>-3hJXHXq;F2eis7k`++
z;3U<go*x6NTs(u$&N(D_@K(RH*D=<+eq7gZWhw1i3~zG`(8*B3DF;8l{dTLN-aQkM
zP4eZCYa3&Lx9NZoKABbY<_+}dVM-q^ubm^cSsn&paQ8c;5AS#ZU+57&EP^y%)!G1O
zBpDsi5OPX3MB0?$UQ{c@hWmCFF3Mmy=JTGx4;iv@fdj=sa1MWPR(|h|wS$8^(IX!>
zH$j>|GEiO3oxtCe_Ug=e1uHz@d5YUrZ`FIB5BTIZM?|z}4kemo3`H(2LH)8Vv_&z{
zg?wm-l>%gGx}`R|@Bd%{j}>xJd$MskQodyC(2gSo(JtD6?L;NcJvy2*#PK#^(Qt&%
zHo&*-!s#8uEA){)eBch>&#{dT@IZF7&m1A7aiE*ynh$gt>uHVBj5oaRG!a>3aX<PE
zXB-K*<&@|EY8QMDu3&`=a5zL!(C=Ma2IQ#Q?+G7>3U97aA$>!W&zl$-ahGT1XCq9c
z_aVuD|L^~96X)M;vOCyCL(LoH7>d$;)VAheMh`lZ(=_L&PbG{_;R{{x3AoAgtKsH!
z1ddKe3vhh$sqe+m-|^Xv{@(d65A;>aV05<w?7ev0^{bCQi~rgTuK7de(BKoi0C_rF
zA|>qC!v~_OR$QK;|FhcaBv$WPCe9w0tjC{)BirErsmSlIfAz163Yq-x<e)iiZEug4
zwA8>NCl0Wo^Zix{DJ;@;RA}}0cldF0zFJO|$5WM|ltoZ_`7)qTVg4<uQ#-czRP+!f
z=TzQl13DRF+dn&`Ux7Uz_7sj_faGn?Y3TR$uafrf!toh1`npU+hL~#yJY}sh>Zx#T
z2L!_TcfCY}^^BtmW}O0vX?9egFwsj#0J2KQ7z%qSS5YgmdZs`)BQcbvpNQ~U1U6fJ
z;l&W?3yPco1{Twekt+B8lM@jce6@uWvT_6bFmB3m4XVyrL~S{BD|<>1gLY52EuG~E
zI9?=pK%JrKxv0bx16X=z&d8hvsj6ejOTO}ZPWjCV>Dd&=ly1-kz^S1ZdSAPoHc@oy
z98qfDb$B*92j@1OixK=*7P-(?^;678S;B&L9wp_S9U2{mstEIxch^%+FbChWjI2qt
zo6r|5l=NxdD1P9oj4fC0s^@obf4f$;<0Z;(_lzIBti5X&cts?lUuWIec%w_tKra0Z
z-K$?Oqt<KWgy1*IU3n*re;nVho^udR{TDxQ3h?|4Q4SDCgrTg_lNXUQxDdXl&@>h=
z=%Eqo(W*A#KW+W;hQm3O;Rl`P^acxFVjR}i@(hw2KHy|D-Obm3q*9;5J^5DmzP<CZ
zjGT#JI}<Jq!M&aw7)mfiQZJeUqK(R{PURe4LFXQr@VMva4_>f30x#p0OTmF}d=U_?
zHFyQyG|1sG^pdV~dBY_>8t?Z^y^NC2r&sXGmX_mk4dcwXl+H)r=ukNR>ITRCdseCi
zS8@!uGs3-pBY5Q7T$6q;vIgI!Ip=Y117v(m7fwDI_8S}<B8wk97ZI2qiC+6IQpjTv
z4GY8LWtiQg$@rjh;97#|tm?wU&Jxr)I3%WKR#(m?=kMScp2>Z@RUL3*N$MsJ_Z$8k
z@P~&OugPOLk`Dp#nVXSiVu`A#<Erjc2AFs3aepdPI{$~4;rj3hoHC@Ml~HN<rH6;>
z@ZqXB(FKtL8((oI>1fUfr^H5CR`F>Q<GShxt1`g_M<%hqwv{S2@v^6fEy|VuNzMpa
zHd&YK7$OY`<H0y;&kQp>NRHulyryLTonY?<&zhqHUx!|O4&zaBQm4({Kz4q8867&=
zgN~H<qZ4(&!2-%pXH54;1CGR&-`{vi`fq)4CcD1zT9b9dAwH9D=<Yq(kaaRu4U1F-
zKd16(pMluCvUR5Ftb)6fjpI1zB-}NFxpD>@61v-!gr5hSKh|?BC|y|VAGtqmcu_sd
z1(!}C+&%XN`LW8;J8Xv!?>%TQqF0*Rg7@L`)IE+<_>Sx84V*g<FzE~qVTG^>SV9@|
z&-0Tqs1JN|nF1g%drt8K%-AKc?`8xfWQ33LMzE08@4NR>!1aP15K1932$%{Jz1eJi
zy=;U~zTOOY%swaBKl<^HH?fHlm2czHl||X;k(h;TPQ_on|Nfo7`J2DF^V6UHY++f*
zGt}{<@;D5q{$Pz6pIJWJvtF%EguM<h&bYy}Hl~ok<-CNOFir{eUOm!XhrFBz(HCjp
zM5y^F#a2et12=YNpA#^wcUZfG(@@v9Q`F^P@RR``gQ0sht+v&9nC0|#ybvKOdJ(Rq
z7tuI$O4)1Aj)MdHMZmi;cFI|AQ#g9<DIH~@EXo2?sZ+c3w9r^PZ``=LOtgI%^U7y@
zQjgop8EfXx7&n}M6L#5e3N<(!tEWU@o?_TjzxQ@<?$Pb}nT*yC@D)1kl3F0SXNO%p
zW)`<YnG_5MNtA%{JI3ZLvo`<7bNuBU8eKzXgjUb~*{2P0)w9y$LKoiumUHI;;Tt_g
zdDat+rx;l2zKGH8Ctu15%01^0Gq{%ncpT06tbAxEtz%9C_!$$|gA6wZ@9_p)oe8Q+
zKfJANXkm!r{U5%6=chljQmoDP8lV2Be{MVU6#tx_a=h8Nxt{Fi)6|a1b%Jh#DaT6#
zeB;0Is^{Pg4kudT7wsKgQ`*{aweN1cg<rG-o^5s{pS?<yr}s>8Ul9`c0KYwB@Dn<U
zUZKy01ItJ0kl!~4#}K6P>G6H_ILHBg^W*n_9KW7nPI#hA@awOB^~<dovkcO_BXE#4
zodl5x$7fBhwDF(_nmIo9&9DspX+tC&2$K^RimzPyiM(k4nK+3?(Y>}Pm(0AHanLU3
zR&oWu_K-Uye%88Z`1z_KJbPI%HrvoKSVWCc#A>CT{hq<7e0qvbw1J+vV1|;^X{bW9
z(*7=UsL-%NW^7T&j}Dy^$2}W##S@OhgU1FiShhP~)H$4fb9nEpB0D;Kyjc4$KBw1z
z*s3y)`rrQTzb^vtMv>AV#4Ged^K%+jz>6<l*f^`sTF(BHjQ*onIWs169!E!H+g9@5
zudH{zZIiKRuLGMN`*u+hJBcdGN;We1d9uahGI6s+i?*Z7nU`_e_w3n?U$am5`rZqD
z3?I^I{VR{Y6P|q2>U@p%i8&&{yy3G_oDL97ogt7J$B4Kxii$nDC(7@{d+GDaW@q8v
zq+LT+_$uj*Hf&J1H;h=iGZ*h}df7@S_SCz01pNYEBnhA5`A?gp^~+!W^3Hd{!?s@d
z-Veg}SwV%Amt#Oa$>WRg;kbL-zsLL|v^KiLp~SmE(H<^!!gNAnk(YB?&^DZJ#G)5?
z62B<>NqgMDx2Y1--_>>56a27$Jv+6FgjB%A{A0u#DFj7eCJlpf<<D><iriyaP<2k=
z1L)<U%zn(n3en!5^B9#Wq|zS%yVf8|;pz=t<$z}}^<YdGA?5MRL_#p0(D!keFyraz
zHSU&hU!ENAe@?^=uJz(?l?=|x!B7-u6^uZ}A;YNV9idIQ5&kj~(h}@3gL2E~P_{C_
z3Gr7i%RPAEVbPPA*bv3pZ{M&&J&Eem&M^s`+|Gl`!+295>bTY89avp2LW9HazdS&Q
zO0I);5BdZ48_Z|}_P@L?efch$?0QSPVuE$`d)ED#uOlpX+T)qy0)245gz|#+)N=UZ
zC`c`l=N#wedheNg$4PVXOFTk}mi*vQn=XDk{3F}LhxSG|g8lTB6XKq-&uR5uNc5tG
zI<DU5r!5P9+6n?Zz!^g`3I-3ri_ood@W$7`AK37~Wb4*bxIsTmESj-~NE`g;sjB}A
z2VA^F%q{&*1}o$IJNU7xfum$~0z=LrVU%?w7=RPC<7hKb_-<<9@G3s79onV6`1j%g
zKwZDxd;awS=lyc-@i*V8w=#S6&HUp{nU?VzCH4EzJid&NhGW{iW4fegi^$#bU2>{y
z)!)yl@Sbyx51(h;#^2owse30X;(!-U2wuKh9q6xaI^=W)lG$(I;hQvA0e|u5C4csO
zosi+LA1~YV7a76|xSjM1{yI7&+>a*?4h(B@evaBbxSjF5<z4>WquHS!E#dj}kSwDY
zK6BjYb6t>O&#S|A`02j)8*a;~;DBtekaX=P3f~-sW$Vx32fTb_&uSV&vo`bMgnX-y
z&5<cXman|xLwf28*Wj+LYcD#0W%$aeD{5ysIb+CyK~H)f-9%VyA)wQ1?ghZVtn5m<
z8gs77)1(`q*VWJ6udRL@c<>5?@7<rJt!@XNl!d)}dAJQ%d*V+pE|E449Ec`M=-h3?
z7TQQ2dsv&i%VEjRjfYkcsn(kQ-grR;s)cLOkFUCl@N&RJ0DTtX`Q~sWH|E_;4|X5u
z76CXw^3CHw=DEAuTev9s<V_hYi3hhdf7W(|c_)witAZ=}@?Af&?F8d6{OM`iNac8G
zn9~{4#qhG6nY^cG?)jJ{hB1zIzc9`5&qHKo&KMh;48uHe9_E7(XRPrc8|H6Fu5^MX
zTz9u+LCQT$7xBx_^Zs!Wj_>8+nR9^27TxG--99ETVO5XlZ|Z><GxP;OhY<Le^(OW)
zB_C9ZsA|@-)p5`y=?J;2`(S}P<Fa2gwvWgZ(-hoI39bFzza*7*D`4$7!h{(J{d$1P
zEe$1xQ7mWE>!So8eXCjdtp?%25+uU<DWzFi8@8zoJ!$E8^jyS2UEg#*q$iwJbn97U
z6r+!Wu!43DldL_vXuoo+<8i`_ehe#yTTien8Y+`gJjx|6ytJG>Y4zwhNb=;mZIY)5
zgtl-D^$btGqg005#{>;N4ximMei&4jP*<EP`xE?trFf&WsLJ7wThBw15=NaGblPg3
zohNr2_mo!8$8o`q148>0mtM`2%H{dL-Vh!d>+R0y)Lt9BS*kp`#Y6w;|M|bS?Xq2b
zA6v3mq{r@~k6Wds=V{*%k(Tq>bKkEKmuQHq`iGw;ZK2T|<=`-EqHOie`KaOAqUkRW
z9fN*oi{nxd@G~5l#-<H=Bt0Wf@Mxm2D57ss{QPG>yYsW3{%o`q;Ss$--%sj|k*Tj9
zrm#Zcz1LCD3W@HU1o`A?ll`g$4_qgB@#Xb<FqK9*c<SJpQ+9(d*aC3!{Jea|v`51-
zG9=@J@|n{Bh%53YYQX3zQ`zbS`z7~y@2y)UhDgApP2h9D!J9c1ma&tQC&}G!Tj8PH
zcalHvZAHyFJDaz)KS+4-tr1Z%^(jjgdYc=HONSU5l5P`uBZ>JKxihO@hv!MrVGfLY
z@^6{3p1yf2CheVN+VN&<z&RUFTM<DHw9|ev<}m!B4@;WYVq+@h=*eoY)fQyyjJuzq
zQ}A(o@vSHledj8g0w?gR=WXtpI>sa~|H9NcX0W>r>k)>-K_`<T@^fVD>=|-o+p>91
z8;68$#s8vI8^R3dXe_$)@WGpR{`T+AJdS_&@BU``;;py7ovc0GMwFj@)+(*w6REL>
z)X|OMzH-4HnJU97m{*(R`R0b@4Mo}B0Gwp_mF88LhxShU-k1cAN51^(@9%sTEjHJy
zbet%|I*k|v+pW#R$#Jk5_sdYjw>E$Nxm`@Zi2hcO<(P&O?|l6m|M(2UWG+4-!(T;H
za%@5=-dP7Y2mg7Klj$9Lm`&4pRE7B&$I(x2qqF0cq9=|AO_YYi4de9tnww&1NJsCr
z%3_yIGPmL98HUXfi4T6&2O33MUrE<+_6&O+y>mtr9$wM996QL^9FFSH`HE?yfjNPY
zbaQ+wL$ut)@`P!8z4?>3c4{+x;DK(l$)!8lD<i}^uHTTK6NI*#m`XVP*gu8eeoH?}
z$={E8KfQZ-+-|S*p1VI4%H4g?OtJRsQaYX!2m~R36f@#!0KuJ6z3S<m-iq+C&2!Eq
z8OAY&sIiRvLdb<C^}6rfZ5~LBOUWoxS)1fU&?CB;=4Q)#J_XSg%v1Ya`!RH_w_4qp
z=n~?!voZ+r*Eo}<JU!E*Zyrb3&y(S@TUQR6^n<PI)$S{dJE4w29>hQtuw(Gn-Z=|!
z;>@u2_Y^7-03=s_=TV;L-fL<1Pu~7r$6ybRXSeoVMq%w*zGKv+*QPzM{HJK89k=s}
zCV}lJFUI!PW2?R(T3$f@C9W=5P9v^gQeE(axBI`ySykUS`hl^(naX>QgV%9g$Jb>(
z*bN)^<HPdWdfS(Q%{xjs4qb0o?PLV>{uluD;Gs<{OA$SeSEh*LO?${pKzl!0p=0G9
zTO6NHi9N{aJ98A)vnjL>Oy!>Ax@v@bSI70-J%gg?qbL}QlmA2t;GeT&m6)ZcR`A$y
zQpjEi)zFJjxjf2nT=ZSYIXU&s4sfkUUK{<8W(v07<Fn<YL=S&}adLkHb>*MJ58tQE
z)o1R2>v)i?bt>ENq~{kes^cY}^}aW~PeF&*-S_v3M1k+m4((D$&u56M+dYR`5BgPS
z(IN9qu4UzhgVB&erK2cc<xXa6n-zJ6R37J~X)Ag%E?{&FrugY;@8N+iQlGPzz?q?b
zYFT)BEJKyo(AYkkt}~dEq&KT0<lX(g3nVihToge2<6!R$Z!!Ffl<v1@U!PzP%x@eZ
z94LGIXRkbmduRA%l&6PkpB(h?<Rd)`f13{yy{^uLdPQG`gUX<P)o*^Dl_JZZ0q2N4
z)g?bdQQ4rtH%s2N``YDH7GB<5lj>)98F2Wj*Hj&~wGe}em`~e+{Pk9#;Bh>1TG^%J
z)Nt<9bMDHkuGM|6Y|?d%Z@j;qfwDuACA<Cpy6jtxp5M~AJ?%MHjtBf6wUC{u>^-ZH
zQ_m&W<Un<Mu?t5^T4hV?JsW~fuK0oTY_BK3j}y`mV9!L7%rSz;*K;0bZy%kZCo)$d
z{tr)NMb-FP9pGK1gXZ9A2`|vP$}`*kGHEwGe|dcK{`Sz+;dx;^yfHBY-TfH228@O)
zrF$~q*f1#|4iYE`b-p|>e@`rRG*jEAE4(zy8-kpAiV(ODpH&Z}*)>zoS!0zw1ZE8M
zr2pzCKUweS<9coCGnPx?+xD7*ef?G6HfiobV^IsmfgKVnOFF%y%>(Fa?f(&6a2~_z
zH-T3+#p9ST1tZ*y$d$QCZ4nETUeBcEsPJZ8l_8H~Mwsx_4&^;5-(5Wh^^^u|1pKwy
z?z=d)Hb1kQ-wyW13hPxUi%0uW5gAUGHVGe_fD49haY!(smvP0s?H-Lj-Gj^f6Yk^!
ze0{9ll~1{f$T&D~RuE{bx>i>3JCt=fPv{3eFzDfFyXe4s?|!?pneQzxJG1_4cWdk^
zc8cm6k39lp{>m{#M`5H>rn9~l`q8So@3wK%i`odadI4KGa~l5=56Z`xhaZgrH?CP4
zxNv~c;kg40@Ozu~;K`oD5t>o>@Z#z~FO%OE8L4-xhx7{NTJQ9edam|n`Ka$aY+guF
z61XwXA+W2H5-kFfgGmwF{@GB8o}yJdCb)7c!1VvmF@6X}hDo^$fH15TEPD7q{_&6R
z{J;Ohe`qZI$2+Y;pu9J6Ex?S^uQv{NRSjl`@pQP?+k{aD=c>zcW_)l(COpS~oMd^t
zTPqaU>Al}pLg9U5;1r?ApLe1RpY#lD_|MqCE4=H`pU%c@Uz+e|l^hu$Uyp)+MzXTW
zo>ewDdBd)^wt#7LuB<QOS&<BfAzb^>eA`FNns4$>5mX;C^Qi`3G#xDhd=c235p&e&
z33FI%_ik<%+&wPB!Xd@u>H)`SA8$<Vqb(WHJ3qr14~M%456)gI(6rEbaQfJr9W_5_
zxv8CaMdE5|z5g?eW?0Se$1C-c$$$<SxrC?3>D3Jf>izfZNj-3%kU08)&LT&Q%AxU)
z;12+HIkIEV5jeGCQ73T29r3A;P@;)vuA7f>vs!S);X37WK8_x;@6wa(06u^8N=x^1
z&KRoUG=8xnM#oUZ9nU_h&L^$-BMlsRI}Y12{mr+~=iBj2^UPWahu3{W1rLe>sr&V!
zl2%H-_uhMZZwOArL<bcTHY8X3!(62wS>xnP$CN*y8>%=?4VxtshpVtgIM!jdC&tP5
z#k2FN3d3UBt_;V815Q3%Jl~{lPOpi~U-JfC{aQ8)9^peia~n4oD4NRq`m5}r4H033
z(|&YFZu8P_rBu%8p#_KV;biIe%~$e0wXHa9P7PkP3Q6bbt#F5b-u+H{E7i7ZKGylb
zDMv9Hg&X`xcRmzlsf|Z31ZR9T`Hp2~GkW(591OxCAUd))-QSMC)M<0_G_0cqOC6Xe
z1dfUPc6fi%?598D(=sfN-{IV|F|-1&7q`qy-~G$?3Mm=nj_?&=cDq8JZ~0|CvhG!@
z!R~9z5Mcri1^)PpBjgyD&?`#_iowKulnbNQi@jg(1EY5`AP7uBU>H}2diLA1>-k{1
zJ(sTXz%?B8PE-B9j_dghbni>JJTsQ%b;zSmjHvw>o-Pn7d+FTNZf%kpQ(o{aU+raJ
zN^$2|={SouK8x&0c@mJj^(;1~ueQN)YJLg9&4WJBNWv)$kLs=Mlo_lzu@y&dZBLPT
zqJ1k@v5tPjo$}AvHT;4*2QwO(EVcYKR9jG&XN{@PahTUVI)V$6!>~40+xq%ayI)29
z0LDP`ur?hbDaR&ImEcyq*Gf7RlQw+a-t7im;n2CZoxFR74&Il?W?*ootSR@DX0<LK
zxEFyCWeb;_?&Ay{m{q4O`&)VVm{O+@zpUO>3(kOb4z5Z+hkSo|cy_%<dtmn0dAp~S
z$b!9cL_aw9hRhfZir%>hPEn9&rIr7AGt$PFD+OCcr`k?&DS2U7<p<;(^!V>gpf$&$
zc0g_KMzwHNA`e#D!0WxCHF+2pd;q5<ni1JNn($}$TymrSRlK(1GfsZWxW`~uCq1!F
zQx%ZG8v<A7YVH~S#JBj<v-QMHb_^G2VNVgtnnGUBKF7xW87^CE$`v2yzb?YcIV!zz
z@CTO1SN8tBtP7uf{p`{WgHLNUt3R%5jQsef9x8ohN%;(GJ$W5N3Gv_z4Lz5k0mpD2
z4_-!eok+6R>BEA<#gNc#HGyb15O{m;78o0%?K-jS0V6R(QnbRVDGsiBH(7D~j-wCG
zLvQe*w<{tm@rJi>bRxV~uAK<V(QhT!?#^>mTC`j=X?XNrDgCQV5ssHUECQ^GDx6GC
z?$G(@0lb4}(GZ^ezUurl*jB>%qpqEmxi5Cd%m=m$z60Bd3eZ#0Si{}z^5<<i^>};3
z1>LcBN!Qcy?!hZZT6xOZJj&kVYxHH)z>cONXS_SH=zfv@GWk7RB0z;-!@HlgAJ&)m
zi+;p2Yz)~!L-O$GOZ$`@K8GK;Qm3Idws=lLR=}LO2c;)JD64!?o7LwEU)osSULJAB
zuV`7h+Aqak1m4l0U+0(SfeLwOmg^lNCTJ<hG4{<A%_FQFUj8bFt&ZD7>V)-;gQzC8
zl`Mer903?pJh-1z@LCP*@#D6fodT%A4gQAM6kQpd1dPXN%#U{k@p=?n;~xRPRWzhF
zG*;U<d~h&s#*KgYgCB0EL&}zSx2?oW$7uiVcfa0#1HWu+?4$MySd=2z^<<`)y4RlT
zxDc!E>nUVn5#d*ay7KAKF{+3qpR&E9L@46d3hnAiJj_e=-gvOy+#j2{?Njhyecn8i
z%KdgTytjJe+iykiX5oKk?}5*obMX7%|LvXM_Niyz7%_(=GrlyOFUq`0<ipd8#*+i1
zdI@ak|B}F<qxS~}*I23NO{9E?;m>*Y6v&9&Zz5L{m>71~7ey|fH>RhJjyd98r~Gcr
zi@DcJ=sCvJJGK%*Pxr%mC|~>^|NEW4|L1?&w)rAF@4x?}t>-s)hti*LU{;DQ`0xOx
zzUHtH9B@8eyR?mBENu$u2;gxh^|(g+2MN&VL1;omZP7jlCuO*4lOWduwWA9@@r{H#
zue>pwa)9KsGb%WYy%wvg3-##EC#}Nrjh}z`mtWlZO?%HgZw&g4x4+vy8u5Sk`#50V
zOQ1+NB_e+_v<yYKN0bMpd*_{ZCkJn|W9jev=EZM*(|3_F%H|gk_N6^w55_X?zjhpc
z^|ELF?p1#R+wjR(4V`c{-v7Z*)*Jlq|A+tn&QDrR;Do==K59~J{7<<UwiJC7woX{*
zL~OT2z0#QRd+)t_WbC~k1XG`8kB4pC_fP-u4@F1X|KJZFgxDfkCrC6ycCQVKUJ0D%
zcKK`-!F|JA!6`y%!>A`u+8^gLtF@YlSRGFupZy&E*ggtq@%kHjhYy?KZ&zPB;6xM{
zzsFBLzVqwfe6(emmS`dshws<Px9BCq4Kr6X22VQt=H!chi6Z>B|J{E)N<LoFJAbFL
zzFqJ7JEAM46QOi{Bj*i_ct<aiK6<05mU{l>m%j)XIyUBsJZ@hKn`Is6d3hT?mY0)*
zX3B6~L?OLJm&_3jCgsr$AAR(jMHYR;%)FP+KDElHh}E0;<zbPTAB-N~dc%qk`}uI}
zi-6>`JZTx_C!h3TvGk&dFS&V<9Fuwx22LLy<5<9}IVz0k<-Ge!tmmPmt25)1T)y^N
zGM*x_!Iq&)@^B3$99|rX_eYx^wO`Md59q>V>?^dpd*`P=`I|-Ke)hAUpGY^lRDR%;
zj>Ab8a2kpj-m(7*T`6i{b>16`mi>n!dH?Z0{x@4~N3P&Y<Oj|XeM9c)#JlZ23w9B}
zD!KFSd*92z^xdW+=7U$=ouTH~xB!Ji4lZ27+uoP^UMqCu`SQzfXQQ|#l920Hr_;-~
zq8Q%F>)z&WlU@!(Wo_5jo~z8q;B<gb2c$A?9fC*nQAmHh{Xt687r^1<yX|X)KWuQu
zIUu8czxk~u^Y)vKMCae8heX#zP&kS@OLXL0ZxyYr%}>6}u`Ml}eD>K#R8Ze5^E<l`
zkHZmtgIC`!(qvwgQ3rd_{P0IV4$jw-w|DROlqfjaH*#><kJFBHa`l!XdPMv9$9$7l
z%6tFer<I-#{Q{2CZ}B?#Eee1$5weE8kK#js^-8=^97gQcuj@FB{mm!mjmjlNfE_QB
z0K5i!UZXiB$8%L@X9S0wa*t^1MfH1v5wV|}?3xfmkl_diZ%%koCV3ey@8@|9=p*dr
ze^3tB+>!Otd$!>OSBx{Iv10$tpTXoQd`P=n@OR4JxqI)waX>J*Fv1Kbsz-+9@9)V2
zzcEt3Px|9^69&vCjq*>#gz<{eS7yDDW;mNU4Zee0c<2W(S5H663+D{5@+)=;yq~e_
zfjX9H1>NAShrv4@>?t^5%!S*OEq!ShG7aC9;uwC6T$L$$Ema%^K|m~}{J2SV9I%fY
zO8PWGG0Y>>t(VS;<eA&B2MvXrk5W8BV}cYf`oLP<wMBdENH~l#oLYge*84jODfMw{
zo3}x9!Jl)$$oj3fv)*y}#*^{X9Pjgb`Mlqf-1tOKgrU<e3i8V){>jJ5I0UHHcy>{d
zV8o9UWTYB6=*MAoJ@nP9C(t(qh~_1A9Ek{!VRMge&pd+RQ;s*M+``Q{`*3#*$Lmwp
zhC1GD(xo0GXHmVUV~~TCLYw1HaVc}Kg}2h|XSeIY=6u<2rxp$OXz)?Xt|=ezK6%mv
z-sfiEAKsRKk+q>c+}_Z`=|MZN&p({}=bX3c6}X<YJdjKX1)EH31rO1ILlfnnJ?(oM
z8Rjpl&$30|&@j>OdCMyef$3Nno>9k!zgFhKsPd=irKt?%Xgi|`%Xs<HGq8K`gfFzk
zZ;eTof}JxI!s-;|w8j%(zFAqp<MZ_lJq11a(OVA|v{a^N*U+GPojQ9A%jM{6>zr%6
z98VvbB%8?w19LeoXZ{H>FI{QrHjb9zJY^_v^r&?}9{@KR)4P{Ej5f6eoQrzhc=_-Y
zSx?qFWp!NbKqn7h>e5?%%GvX9HM}O@V1C>1Ry-q`q4NQ5$8=No)2H$bw)1YXULEq&
z6FRSFE*3}N)H4Yw4R~*jT)FP`?nTSsc|$S1QP1>=JmD8E4y3nigIcMMm40q-&#zj}
zTfa_Pe1bo&-*oTF^=SN~*6<k}b>iqex&qvm_tF_Bg3l?BhQW#6JJF^zGxqT;XINb(
z@Q%jOpG*Vg@awS+gQsV!qsm>uG``ANGmLjFz4v^4Bl4!)&M-W^F`V08A)CQ5b{aax
z=X1*IY?CEXL(X><%~4U__9>#9o%>gzU0?F!4QCI8@UD>Ro=Nu?$B;ZgYjBAu6eVHM
zKRZH*LugN<Png;@h7AyiyK5hZ%DRSA2+`AsK!7Q)vvNc4Jwj!K&dQ2c?)?744;7r(
z*RF_FZ#M81LPew@IoOUNZ;Uo0@o}@vKkAzllok)}5?a?0?9YHdKSI&JIS_jaB8HfN
zm!AO6iAhKxeg-Buz%oUc;CN<mV>p{}N2G+ha(Ki<;1crF8~Q4N@fAg(&$D)Y<FPU1
zi+YhyW55SbZ{5zIycp|Qy%R)!YMpm@UT+p&F;n$DeL`?CY4C(&OEFKq2bLL|l)m-&
zR-R~sFGW%S%LpSk=Fk(MAcQ+PD7@)0@a2S5kyQmBv;yHcn%WqC*7H6qY0Q109KKWU
z)0Zi~Uo=Yu>)m(1J0bh)R&rQ|XUQE$ORo<;9@ImI8$EdHy-vgU>O}K$B5JFpvg5t*
z;Rqca$C&J@2e@aa=L2>R;6+>gb0~vyx4Y@SYMaAgp8#`S{@8vI9~DuM7d+}vq+SDO
z<>E(?4T|2x&x%_cl*@@>G@PPh<NJ7OfL$;gFoI^lf(TeI%ch8xr7m(HGC;Xe26}-%
z``OPn*8B6H{;cSNokr~v8tsGm^%v{cY^VlrejW^ln2wVW81?8VaZv(vc&FajR;xsJ
z4yHL1_|=&6Z+`o$Rs;QJ%G0w?@4<^X2tgiVK<)lA3UGmMyi@0&ico^#_17&Ue(lZ=
z>g|*H?|tvR<vTri&U2B?c*+jA@bA1Y%6iYdfL4!$%kMN#gR=g1k+wNq<sq-;Vo;c(
zkoE!*E#TyQw_Y+w+LwCI#P<wjd5)eqxSEl#47h=BlUEUwZWDnJS<xf6{8an(NWrQ;
z$Ea6&<LaVy%He$A8%Bmvl2#u1=_>lhfwpBBz9@LB8yKs7II$#XVbbcSpEaMTymnyT
z25RxDn&bK|M<;`w9{#H6g!Ub$kke{%a&nwZ9e%nPFPc@iw&G!Y?fomY;o`;1VA?KW
zum=M;x5BBk-?m3k`1;%b*Wa#pjW6*LdONfkzB%Kf+oH*Mf_^lK6^_i0dpmlX6S6rC
z@g#Wf)tOMr+7++TXL}N?Rb5`X>{(HyH#ev0IBV7CQz0d|zfg!A{ky?4>F=El!L}i%
z=1UIP{;1D+>9xy_$iH`FxI21%zd1SH^#jcx!E;w9hX@=d*gi}qUVr_p<|K2|+eMwU
zX%U5<eJ|ap3^e3Kf7CZ}I0M)B!RSf#yueG{fN?rA9*coa1FCav3%|o#^}EV9JyKNE
z$4DPGZ|AM%JN@n7{$2H*h>OU))mp#(?eFgVuDXvy(tOPHjv<^M{_y*gTag>bI-=2Z
z^2@G6NQe4%o}asaKuDeQLc{bD!+ggr-?=;5e9yh8te1yd`G)X*jF>%xFo-00FEWsv
zoo6{Amg0`^Wg1ctt$Xtbj`vjW{YJnsT=_pK%&lA#S*L)Tp!2k_pJ<1JGc^K7Fzq3X
zCJ%n`sg(f75twozln2<C?fXVkR@Q&*&r6u$YdAY_&lN-1G7iDrzZBrYQ}S&rF??(?
z9lT_e&#tBLqh%fNF&wyF1~$Z2-jvtPm>qcgD}Aq@o7odQi>GT<|53puNc){&`&|-(
zKYD{DZ&>uH+s>n;)Z;|o)${U7B1^Y9C!DR*UTt$$^v>ODgXe3D;gmo~FFt$LM+U2m
zF{J$T`d)kOOcLgl`1CU+<#YFXO_cVOZ$hEB`eyqJ=n=i0V%nkJ+X>*6O^JD_0Xv3_
zwQ)ahkB1TMDV-r4Gz2q)fQA&H1Kz&Iz20AZI-?Okhfq94fo@Zw+QxDAmDfZ6>aMIC
zb=T`r_J;24ycnz$@QFeWy{q%&Ik(qStpNJ?t1>7q${K8N^Gab|R|edkAfg}cjVUuC
zjMZpgoo6fc!zSLrfnM)(bXRQ;7iCKUx!z-_44=*M>;rhoF}Q{BpZcg*POT3}GQf<{
z-MeS9<;JlO3c~HdLlDDgX*zp$Iy97EkpcC<dycM2i^Be!6H)p_=tP>p8DGJ@-W;0s
z2V!Q>&G0|`8^8AqzBoPmNMh;Wz|iI#!<xzoFtPY3czOya-|!qh{KjL$y%*~0pi?&V
z3V$57_!iB{+aeR;fid7{%A*XO7W}g4Vr6hXMQm++$FS_ov0#|xq5EKMPD=&|Pi_T6
zZ5mwR-D*MG6<kj3I!EEA+A5;!jYltiRl(Jq%vUyebwJ>PqkrZ*Ty3viy(i~z-qZ6B
zKKDELD?_^;)&_J~FTeEWtD%RlXIpK=34Sdcufv(_Xq#h@xbOcJ4dCS2;1C}$$nWJ?
znq;c8a?7x~4wVk?s*CL7Te?j>$9`~};P4EI!%8gE?~)D%bzkQ|`Zs@iyEdK(Lcf%M
z>9^>#_gAjUTDdCdQ1m6Q!#AAYPAd^5({Aq!_mk(xwfXLo!?SY?qB}fr%KRpGc-TZ}
zI%knol~vw&w@Rm%21_;ce1HV$vGEe0<FZlSi-fViWY*|`3DJg}$*Fqgz~{vOsbNQs
z{G1SzvS)Wix6Y=BZ@&4_^vPMMkZ!|&kaWXIjXW@2Ky-b;o27WIA78ruP!z-NE#V5O
zpfUDd)L`XmV8=yJ45_vgC^?J)afHVhq+X>4OE!cp_xW=aq;Xx!9Yd!8N+*=2o?kWw
zi%5jcgt9m4O+P6N{P4r;BVp}|$8e=6As#}1{p;UsuZAxo*rkA#Jr5S#G1QJ*JKSr*
z+N1Js9P%9oKi>NcdHKzQh?6FO;MCq3E#7YbzWCy6#RCskNIY*2f?izicvlAHC;-)Q
z!px?A=8Pt`=7nYDt`||erUWSH4B{B;tJ9_lNAG<XX2^37zNPCj0X>WsTJKnjp4^5n
z#;iJ=;O|_|p(u_$4bT)bI+B@lmVs&>Uh$7!OeY1*sBtPTJW4YjV{l_iA$T10`wu@@
zUH|wGXT3cm3U}7ui?CSbq@1@~y}+RUpoj_M`d5F|m}Rrzo%U*AbS;-8$XiWReQ-@N
zS&zT4a(i|7ml_yd?iu3^|1b0uLQ$eJ1IiFw3Ijb&#(dQ2UT+ai@v#U71;StdeZ7RQ
za-8f3B1&_8xK0mww3BYaH&n(GXMCCyW~COVO!<FyBFHz7=s0*Nt=_5Q;7YWjJsXk-
zD-YfYS8p}k_1*8aS57@hD|`$inWQ!+oSat|I&7FD{xXsI!2^aQ9x=hN1OD%a(+9WH
z5QiOAwclzIyl3d}-S2*T(Tfj@RDKveiPRiWJKX>PKmbWZK~zV5>Kx7Ca?S(Bf_+i6
zc+1Bog`qjYHmriAHyVP`4s*zSsO|S3{64ldf2-9<wN3PFauq$59iZ{qO{oWKg)a((
zPqdH0*KRy$=!x?$jY+IJC*!Xi2Rj6t3qrP(PhRK+`}!mv-PcYY^zFdM@c5@_zV8l&
zGkDwgk-A|-_5L<bBi%Y>Tpn|1z$5=#@$mh6=;X!xli&XS!**@`Rgs>)ebTTFLDUAr
zlOm>X#pfa*=EZ^A1Yphv<DgR}5<>pIdVJ>f9V2;jK<a=@-%S@BX8|hk3cX_v-kiVW
zeEfxBJ2?~%k=XP6;Cc>r^cyYEGB}<G8=A~X3fFK*2YfH*LG<I@oC;qkHXM8R3;OQJ
zlX5K<;QWZV{owmQ*h<HDn%~E9@#R}9H@^41?*>V{0AHow$}dh+`9-qKRe8`HgQp4O
z<Ihj;n|w0VOc~GP={gVOJzC0V{vwv>AAFIASo<4CFa7THUi+NzPA}8D+vE@NDFMfn
zscXMI?smFl^2H8{gdArck4OXe&R66AYYyC;+g`%w+sU#hF&R+5Jxk2>_<d0c9c>*p
zHE}5D_m?m!yu;sP=W1hm#sg`m8>1f_0-Ju~%!;O&>tz`Im%rqM^u9cBuOlP+BF%L@
z(+7^-kQQyV8NJ!?$6sd8$<qxUzT#hKwzF+Jg9Fgha{zU<qL3E&bG!9uKdw@w>o~^7
z9g@2AcZv6_68cwKJp(2{4+kO`F%ba_OV}aVxF&|Uga89r7KLac2hvdf4(TZRt>=#*
z43>#@yn3)4B8kAnJk@=m_3FTN{@$9YK*!alZ`{9pek<eZVWm;$7(xDwch`Bg%J0-B
z+x>#`eiolXF2SjPNwmK*I)!G>eeY4K|K%?4iQ>TDLWDivd#+-WOsk@#mEo^F33^}u
zWzBe&lD-;!S}!vrbA-qQs7nBC6m&O1y_ZttSUqmzu+gD!L1^DS^H8*@Fs9m``MiB~
z;~O+_PT5sD14(H1zTd+0X8v1hM-a9ouY3+T+KPwDJUvT2+9ot6u(nMX-EtLXx8V?@
zL2+4%clLBhS>=RW_gkJB{`NhT@cl|dIurgEXWMz`U%jI<UhEf~1r9?+>2vbXlasKt
zC&Ty%AL@N@4)6D!1gpEg%&Ef9cN<cmpzy%x=~-o>uX<F4j+Y_RUU>6-hqRsR)%9Tb
z_5$PH5_xxetjZQ4IFXkb0L#<Y>&#ds3!aJGY-irdI615+EZDz|Ip37v6zyno05$Js
zb1VQpLdxh}`52KDJ2>E8gmX@2c=_a$&(3n=<kHuw&oEaUS38G0?Tt@OUUiJGg9Y#4
zw-}<@j+~D-F?c}3MTlAbIwPX$(;JmRZe|oY8{Na{A`CeJ;C(Gzgp3StOFv7GuESmF
zMHv`l`5g!UFPWq#Rdf-cve3*a4DRIMs$(!Ipkti`eT-nyhBJ@t=paryT?B?NgKuN(
z4HuoE6#J6Yu_ANjH782M3oXgPctu^E?hS9l>6w&Crr`<>qceQ!&~)$gOf$^$Lbnq=
ztK4Ljtoaw(r;qW*BDLyDn@E2Q*Xf{4PxY<<T0o`03S=2P-cJX1H5|ij<Lr5FYkqQ}
zF0k9TVmu5#wPO+MoG(r*T)`#!NPF}Jn?S!b-zLMRt~oN&^;_FE#F>#a)P@%s>J3+k
z5>>}^9=)eM{ThyDeWB(2^LWS5ML*qicsRwIc=Q_TaDU1;cva@DxaCvHuou3MPU%5O
z<*C%Qp<lJpU&nbqJ_;|3D#f#gT6MbVOi{*9KK^VSEqtOAp-9Jy@uw4R1rp%TajSMO
zu`4<`@&3W_p(Ce%9hOg865jWhKKU#s&JoP`jgzxsx7pmYf@}7==PzmB2T|R>77uK~
z|IOY2LS2LC#*3%$cv7L1I%Zg+F!`)T%(=TYVopjq=7m?-n7vXx*F;RN_%W6|mtZdc
zNjJq%JrD|0X3Il3sQNo5vq{fqg?cnX*lF<SB_@{EzDs{(tVYXKkI8I(W5PT)#!rvN
z#5Un+<HtURq-_v+9hdSFes&yB@z$==3D2JwYwUsJ_RJgk19H#xzA3OxECSz^H^S(R
z=^?|>>-F*ysj}6><UI7ALE$)IG-LUcBxQ4}eQ-ak-IwAzKq2PKyVAyNHA72$4fw#P
zM`xJLQ^Hf5I$!iX3PO#3`;JccW;4UB9rm8rOL+Q(g2;n?U<E}+JbY>&#frY`HT4`F
z)J41r{qol)H1qYwv(~(q-cyz_rSBwQS`$@y)*Q&?Z8!s5PxV|n)H@i<hivXC9^rqH
zG5d6&v!NOD2t-4iR!IGK|LuR=9uwC@E@R1X%z2?a-OtDi)7AjA9L{?8usYtv`?Xa%
zeDk<A!6g{LyI}_kuG#Fbyl8S}*6_ntY}xnXyF~@i8jfChrIkX_)bHyHsoExdsrSN>
z(d)$rm%@d+;ko+Z99%Odu72DOZ_r44HUwEOk~3gl!gt^O-p0>GjNr#tMdt+8mP12y
ztnK|1HP9}*dvZD;WQHyra5O|LIL?cTCF@p%jrZZM<UTcE-2Xf8wtqwWM|`(&>u==*
zZ3o$Uz>Mb~!_Da=aJb@sbS%>KydJpre*MMmqGPoYKe*9zMf|J7Vn<SwqP0Pf(PoUc
z0|)<>(w0Yi%=P%7sEi53hHCNBJ4Mdm>z#h%Q>%PLwKil`<bvK9Ea8ab#R2%*;6$@P
zX-kgr|3@E3`;6u9KlpGoSB5%7K)|+Pp32BAj`GRbch2gSN6jU&r^UzNP`en&KlRKU
zqu*852c7g4S)Y!mOzpx$+NF*!;=jAejXld2709XA(ekm#(>td}llA_YybqV-A9$?&
zLhk$QdphQ=cUnyteb55`ZL&Z93a4;kj~_hJkNw_|&G`EE4XMiXjs+MmDnA(UtzNR7
zs{dHqyt8`4+!-AO@A0fFIdNPD-l{ESUeD1MKN^Y;^zz&Z7oWxJpG?=SjOr}YB2Nq`
z9J3uFqHG)6@I{UZa2@oPW6@&m!)6{NVCoa$KsOoWfzNG_xk&yo#C}VA>|c55j4mR5
zU~%kkzfDF)!*yKH_Esh*10Qc#yEc<iLzd)A^cKG034fu-?~@}@wSWEBe;sezRTq4C
z#Q`3;n=_c~va@ivHr;go<*c85ihR8|n^YdWMAzXjeCb<k+HHRq#c}`<`n2-k3!fgn
zesuJr*lW*mXg6P|GT9mUyc}2jpw5atXB9Ljgtsi=Imiftvuf*kOPojIk-Y_^JIvj+
zyoy?p6?`wvpn@~28c`DlF0tvEHK<oRHD?SPAs~MZZX#*1i<=M@&Iz~~jf~Aos6pm0
z=-Cl=p`j;w>y>LPn&+!6hmjJ}hgLE)QLFJ*%saR$h?B6hvWASVjv=Kw&X0b1UfL@k
zhIfM7;kkn#PhsB#3*ns<pitG5mMX3cy^WTSQ3?o%NXnXXP&tJB>H)(q?;Yd9)j7c!
zae!8vd5KSQ!r*w!rpGwXF(QMrT=2*9FUgF&)dW*Yx2T4)^pZIUk6TavtmpT-2S<T}
zwM(04Bnet=2Cp7ACWoS{y}V=(w22^Gku2at6|?s&hv8wX91i{Zw0a5@Up(7nQ3~hu
zHjLS8yI$*zfwDVJudg&ImPMSZi;y!Y46X@Mh6M=y8@Hhx+o@CL@ATPlLT$K$lDnJ}
z%cH?8Osaj_4u1D0$>~u~G4xCt4(^G%oK1wl^+o(=cgu%GM&#FC(HCLVOF6)eAwn&|
zvoq}4rOxVCpW}kTPxmI<n#sR%yodA%Dz)<{nIc#5jIzI#BX&L*Xg*DjRP;r^wXYoe
zhra!LJq4$3{PxuYn<FKU-FpWo=7MloD5I~U$D=w4l%_p0j`J2QjxNIP9M4zV5!3QQ
zA#~UO{4f7`%G1aG^xQzLKJ#d<9-D?FDtDedNOnr2hlYNjq?tr2@Bw);1aV|BxJt`e
zL9exA^h!@8@1h8=Hfa^F+NE>YQcpU<a2TNf{_p?q4CU{mQ}c}me-3T*<=Cpvd5lud
zN#)362|Z_;t^tSg;Yv>xjlYbi9S<qw)oqyR)icil?s~>RGAPE@w`4zlhffB_VQ3K!
z@Gb|$@C})sv8dk9(`kkTr{Bmf+Sm5!@~(p$Zw2poJ9;nTlddxCWzwb(qLGL3Sa82U
zfYSE>2sOnuZYggK4Zk9_r^BUua=#&<jMkw?NMFAAJeWj_Y#|>o8)kZxBXZ4qfP?W*
z<*9P>0K37<+DW&ZPR5xBLtYJA<2|@EbaUuJ&W_D7jE{Fd%{aq@VLUqY)3&|;yhx>b
z=pDS~M?ZJn4H*2p$=PHo89UG7yUOC&;M=u*d~uZ#uZ6S8ZRJCk`|Hro9edup65!S0
zxq7P-0$@5kT4iz^v^R?MqMb+?Tw98ZXH4V<2Q{EEM=u;$P8sRn-~ht~hjM_S69>2S
zmkt8kwNiRHdR8uT{UrP;UU@LH6_E|sDF;tGrPpyJk8DZ}s}B2&G<PN4hiAa3gagNc
z34R>P$DxkCqMNXt7WH#2dwIPYp@{%e`7#H<vkV^QQPYMeD&6-P5W4!SI0aHIh&G|f
zo4OF=JueV3UN$D9EFEulja6REXM-EHd&kByPRilss+%)%zQc0XAIs}_TJO%XIqxXO
z!+<xmYHz2|n(MXu<JI09o>oqkl^28Dkl6JSI@imqzpjWvA#g${4T^I;w;1f6Et+FE
zIM;&_raMf8W<wfqrd;*9IcOKo^f};D0&q*`tB0HdIP?G=PBGe9d4wFrmL}!j5DI7E
zp7d*EG%DYd4AnjO2VZ5V)9{fbPdG>Y&O<@el|@)Nq>j<A^3ccoYhqx~lz){>@N6+J
zx_~}tF-7qt+E&)}u{oep9Dm4>KwoWyuY35cvZBD!f7!k^;Kuxv7WnkoIVS(t|MmY}
z8;vKMpW(Y8da2*<dqSL)?>CFvva=sF)=a_c$rDJC5{}cu+XOIenbHZKdns$YZ>Z09
z(BEq1ka;Z(f%Y>h==iYivi4nuNfgO*bVPG)LByIo0feLRR`|FM?ScQ{yz88f>t)xn
z{`tA`@zd#@ZGOyl4$hDT3wq--MkoN5d1VC48`5QvY$|7+e<03ygxJX@2ZVB2FOCv!
zI4oi{Pb9d4Q)vIW{bcHOAE&Q~NVDpn{OJ8V-wp5I=^GxRe1B~G`d7dH<st{Zj*3_G
z-jxAIIy>a#Su4NhG&Q$oPDt$sGMthrW7eFPgX6Q~=9iVnU>U*?)qD1goW+N=89$Iq
zU-UdpFPam)IxQR#(UK3p{XxS75AXb_ud>?9M!rX{zC9ShIht{Wh(HdHp#$FpLf7Fn
zyfG|z#cAT+c%gblWzd6@48}RQrMH_dgW?n=5Q!FLewcyPHpUvi*|UMcFt^7c>T9^-
zi=r{~2OV)HD#Gb83g~^{KAggE!vW*{D0s%_=~ZyyQ#uL{lS8`X{<g2oD0Lrs>ba9T
zGIonZRi5&l@`3@(PKUO`gLh;Y-QWcOc{kls+Dor_rrur-$MoRU{^`V(np2k%d^x8g
zd`|X)K_oy~bHFOc?!I)x;OR;aYa8Ps-C!g^2&&H8zdb_do?UbwPLc(~Ugn~_j3U=2
z$4mWo(@o%=lX0tOiF&qnSMJ*0J5IA>6sir5GyJAg)$mKWGO5>LhhEW%4;s><XYfGP
zt@EWd%HF@O(}6eKE^~DOCCp9M@FQN(`B(%$`Y(TZO(yo;F#2By>B808*fY9&XMD_g
zxp-jnqQZyYTBNgc(G6TLaTMQNWvv0#S>-#|o%NCzqY#I8`(xbKKWSgSaK8i+-A?GZ
zj-}ii%yj{)QJ4r)Fuh)FswLv4$TNa7x;5mLCdip}{rJhX6s+=Y#bW|a1qO#a2Zp~4
zHie;{)!E0<W*{M_w4Psxy&&C}a(lNA=ShRPx_)gEN?E=c3Jli$Wnj*@e@ZTaz*J+b
z-s=Gb;9&YDobrO0v7kUq)Lpwg>;$v)j1|Iuk&)A0Oye+rBc!d}VYLS`LQK@C-SELy
z3p0Ak3ABvOv3I?9jGjxepI`q8zA{9U>T$1-+aaf+CC69Q<wJzp@422cJZsm52eiG{
zDySt`uJdP(Cp`9#(!i|RJ!Vz?KI)8KPzZ<ZA<@qb9Sx*ITlk~6(f+gwpK50Umvdx$
zcW`|fj{fk+KNQYAy;rpR<*$FSp0%Z948vPF6pV>pJm<uwRfNsZh63Ss!#(vNKPpV=
zGvijk7*4!Gj`pyDgGZm#g6fcdf{5m$OLU8_<NLMuhL74W-5g|(N~;~?)1L3B45E3z
z`B2(QS&csD?Hp20!snlVS<f?`ss!bcXQj>Q1VfqETP`gnJtwYb91YIdXp$V>5ATn2
z2*$6yC!6?DUOggFkTiV%D}Sr9qCvF8x6Ui-0>3}{q31p_`RSHpGyWg_$!2`b7iofB
zU>xHVq~?#%7xt48Av<%$4i3QLWYpjb-5AveEx~-H?}Qw@3CW+2MS@j($PP!$#)h0*
z-_rQ3IV+nhP#Zt^@PnOJ16UgT-5k@u`pMbYO@z#;zA4t)i?8Q+MB_Uz^Z<1T;^Sz8
zU!CiTSC>h%U%b|wCQIH+4-O6o$HNfJZ(HW*lk=W2q<W~Y=H#AU|G_VL!VhTWz;AHy
zRYQR~I}EW-m<|^?I`VMRs*_PN*_ZJ#$y=1?IO|7`kzwC$$D5m9P(9<tbVx0j?7e8{
z*bstAJYcis_xQMa$q5|85&9fn4)4(?vE70HIr}=CV0T>foNV0m9-5&M{`s=B95&@$
zhy8EZ4-rqDhIRJhJMy-UNNJV#RZg12J>2c+Mx6xc*sklvf8i>Vq|-a4&heZb>w!+s
z!LCHq<>AY<r~2tP2Y7Y_1HHLn%$|W2z8G<x?Kxi(_HO)E2M8a?e{lSC&If+A+1lf5
z1}7PS%BuL$e|7m2SYX)@NcwTZu00!F!YSv*DGg@>J>ZGW!ap2gJcm~g_PX|7t1kyf
zrC>v^>25gb*>Di;=8#4^ZS!0`0PE51xZTLp>DUMCX0S@-*$UhZf)VAHxlf^GI_b_8
zRaVc6JAYR8$^zi=ngMdgua07TR<DZ3wf#au!X_>Dn_vs)ycU8E;0GN_*63@~GKNS!
z6TDS;-b)8bBce&6<r$1#1Sn)(tc}>(`v4KTUM=<P<xzTs86gqq#z?M&a)wo!n8$%(
z<VRpi>6E2j_~K}s$!FKM06cdh8ioBLr9fE=`4Tu|Z9SYZiKN5NoQ>Nm28~-2ZwgVb
z5^;It4{lD#03%h`aW&L$ZIwPCzCMqma3xpAdh%WGRyi6yxPOF5JK<Q53<DZpAMO{O
zK8y&@LWG;M(v_mqmMgD%)NN?RA>4|tCZxh25U+!C0Y7J+f%y*j>j;E;@s0`16k3H;
z#_1W|dHx$LSUpF<SxJS4oDa%)jz>L&pZxXToZ*{hrHg!^fz>&Ts-6h?ne-}u{n)j2
z2BGI4XOu7gU<B|CT5Rc?YxE_&!*>mWhn=Nc6HD`w2Db)}R2U@OD(g~w%2*imRxU@u
z<h%-)5<8uWn-dc0??*EcuT2UjAJJ?It@L`xcv+-lGI6V0nq2gH?m7xV7@i(IIv;%-
zo`ZkPSGtQ1Q$optS^bt+bI8bqRSxL(%U}I^PRR!!{B|@l*8}}HvGl<wIU$c9qYdMB
zoH*dX3x0u9v=SkBl!16$o2(F_?CAzW6IV}qn}d~pbSv*i$<%f(E*)5Y(@v}8_IK@l
zU^h}rK|jhdFobKPH<1y;c7{nN??qW$wUNHyl&rTLdBEmW7XyPfueCKc`IQGBaX<`9
z38~v}0SwYJ(yx50G5VZA@^iI!+vEp79IH>c^d6peg)=amZ_bE7pdY?U_J^ld)HFv&
z<WA%RE!Ihjl*ia*Ec&+%`EDJY=Zs+dvaQPl<e8<8J<v|WF_y<0z9*jyDqNt=b%4`q
zB>YKly4pP&O!9cY=Vf@>J>5ZmHj%g}7kO76+8-VsOIHsadm?9W=_|YV<#G_=Se}<(
z%N4Dgz#rvt!FcrX1?^v6zi!9gfN%DT&V#8A@ZGYJl>_U|J_NVl{V!erBgTo)w4-y5
z85lWGqJQYX9ykJfPKbG3bMC8S@*G}1uwp7bXf7Jql>smKmfTDSRL1J9PS+RYJSXuX
z9F&#=D7`4PiOy)D9h_SDr#BsFdCKoSMk@NHe7ba4ERRF_MclfgC%7*Ac;r8QhM&p{
zo(k$J=l-}^c`L0KMo>BqG64iJtLxO~Wk-qnJ>lvG@mDbhRl)upL%JweJ<(8c%IR*(
zm_h7BRE@}H4M9`PdK0D7aBX)`_LZ<nPVz8vV}u-w!56F!#`{9RQCdepH~@%rn!BV!
z2&9Y%UFV5a*t3MjJ_5>r*=+Amw&lC(Ab6A@C5f<>A-x>ut1JSBIT=f<21b;A!`Bt@
z0h{s=oTA?3Pd%;jY_3af)!y|o!m-{jXi6T8L71_-2!Hh;#*i)5g&g>r#H6GTeUE2~
zzN4`+ZihSYj{!5TW71%w0M+}@uuFx!Xit`k?zkDk3=$P~;RrkuXcxFS66>Ya)&NT3
zM2pI?x5Fho-lHFdX_mXOYR92nZS0+YoCI_O<MZJ1jPjv$!2<}w02doW?YaA&1Vh*O
zbrbK_SbP==m07#-qI!(vKdYzwkN@!ZqwU5`>vf<rBSDFSi=q4VuM90@K$<Y;8J;KT
zqBD3%n)}fbF5rd3M6U4qj5SnM-vktaI>(Ey&fa<EP{wtPcj2nj^QA2vJ~S7^HskOz
z1Ku+SvBM6Q;N!njKMZ>2Z}%?MdVXolsoT7Onv=q(taJ`2rM@8B^3MBD&rnYt{dgoi
z)?SgPwMAJ~iYDeB=!pw6E9cdQfEZ)Tef_qfARKal7}iZ{43CyrUPkEZh1uHs!MB>v
zlTlMH@U7~4^d(+65m6H&CwIYj44DnO&ipK`+NO9d$+b%_M4ySmPq!H&2Z&=}HOKeA
z|K5i1Y?{V7u`HQ^fO7@~k26OHQ|NRvlIaF%$QgW5;CRer){l#N2&aSf&72!fpoorM
zF#a;^!U@q^rANSu&YPQ59>!YKhLc1;azG}}H_8o$j10YUB00C^>y>ZF>A@zOo}nqx
zyd8(=;~~622G>a#<Md?AL1z}VjEDZzd@Kk0axS&y7@i}e7)HB&Z$vxjT0;xu16<03
zA6Iyh-*fW%)uZ^V=hJc3QL?3zcKn1HgIAeJKsSDzgxlsI7>^`_@W#*ODSP=8=IULS
z^KidM#fy-=Q@Xl!&V3U{ePG^p3|%*5UAjvKwSNxD<mOhdimF?&SGsQZ+q32AhEqaj
zH@7XCa4gA#xoyY!I8H4=Ub}C)@ZjWzOLbKisiuG2E8pCrt1~38Mrl605BFV<Jvh#S
zHjZ~Md|R3hPo2ZTt(}9TdiIG7SLxTb_+I+Tm8TlNK0NO@ct3dX8*W&bDyGzAL2R#1
z26ro|pQ@R{y%pw_kmGeMcZ|8Jr`)tz*{&2Pgk=^|HVmxv2sbZlh>OS!t^%ur!=neu
z1H}AuR!RqX-oGd>FlE&cNOv%B1eAb4@PYia>uW_{-YSxBhcdgr5aihf-$|ei!B-l}
z$4Sw<&grOqghBop%ra>g#cy8_X%L4YG^TB$la?+7n}X0=qG;`VVzQT$fqu4SrO{@}
zueP-}!i!>~;HbE+jHn?k!#MRc3Zr5|Uh3G5^#K~;443dR&))Nw2f7^m8F#fu#~C)E
zRD!){V7=zgRL~U4#ndIbl@vfXoWcv`dgn`ySB4i%!O7^u5rqSmt;*_pkK#^A_|A?a
zl9!J12u`roUTvn-<e#9ts;%K7+`!$+tt^HF{umea;#u|SAxl%m*k%XF!NpxM6vqJl
z`q#f&dpVkn&J`u0EXL2hwN@VwtK-0Ucw>%dQ5d|3AID$SIcH1N)iD^racc(-90qTA
z{PX)GKXAOVN_&*o`3)Y2sDy7l=<#mySxW8V5z1>yB3Lh;JUBdUUjac72IjZI&6(w2
zz605vXPe|#irR#yAGE741B!-sAJxNp!~u)97I`FVrC|)Cm-#r1jETW`^NqfJ!3p_R
z&IpI`U;g!9x3a-kR6qahLGKzm$zlDn4^>|6BOhicG5*fu6t=Rb=HuiqdXF9=XI5X`
zvxi9ha^)9opf{k)rcZCR>79s>p{37l%2agb!{&LIC-OnxYx<{u{Kq>#`q2;D1@cFC
ze$wu}KmF-XyZ^y>8ojm+KyBlgt<d4@YhJ;x<YjUn-|6LV{{<U8WyHZEx<W?WYZHe<
z^hk%qMC#GGa?ws64jbpza5P6`D>lOObR2%FZB@IFx<~;hWc%oJbw8a|Cxe+D57PH|
zP)6{>SO4(6`z`CP{0+lGbaIH_MR6|Pp7U7yE?mR6HqY6p%*_vrXGCCx_|eUwekTXS
zNw$UuI=koY;r66EZKTu9%K$FDcQ;#M80<{KOb5ho!#9(Ln~7L=E2k^ERpOp?TiIXh
zKyY8Y;vGG;e-)q(ZAa(Hl=sft?-}IUw%=Zq2U|Gk+P#O{(!V^qzi<fO=B^nExfgw&
z_U$JVdgp{BABJJzW1XhzIrb$Sp-um8kgn*p7fq+n?K*q*fca)y>C80a(u;(^v35Bw
z7-uUgpT0Ql)OOL7qtm3HEpaX1H5@y7#p^;4vp?TxGe~z-pF_DTyWev{APNBiyUz~4
zOepd4OyOJZ3P8OYj4JH81f>T6M*$Hu5g{4b>ua4<n&y`<W{&xK&$CR=M`;ss3gM}&
z|H`uD(ZQhb6b~{EvM{;ckUAhIa%XT_mi4K1)E{pNp0QR+U)z5C;~(F7@OteFX7H75
z<-G{04ofqkM!lMQ3Dt}@cynwnr12~*q2hTH>^X1MaT$6ZKiq`Ko%`K0;VDB<9EPww
z1*EtAos=_$4KU0`InIcuj0o5a9j;ChEL{K^m!>kF)@zz!>pfv6a2ThqQ{^2)&$vHt
zLaMgG{h6C^6Tlm;OFSi^O%yDF7J6me^*FWtQROQFiS)dUHLIHfQ7485_uDPiB(vaQ
z;PDG%?wu6~u7qbaW*ky1+mj+3y<KnKJK=xz5Mp)9gXRn@CzB&L$FzHRpEK6|(~OUh
zwy(lM|5Z;nxBD7eI*M*iUT`bx!aH2O_Npz-4-b@WX&FnA53neU0D@mn)M3kSFxV2?
zbLlUBfwbXXUT25jd*=T3d+ELWiv$12hbNunRNBi(d1o%m51UwN67CGZU^vU28L#UA
zx3;dFXtE)Y+p{@r@gCVUmP~Fod%i2@L7t1}@uA<~z@J|<L3&O{cz@V4@2+-<Yz*Gp
zGxHu^dGJoo$!l{$4bw1Ye@bru)JN?8(|`I;{XWh(SQk|)@BRD19n2J~s2SXw9Q;`d
zn?dKOFy`<&$8+sEZj{aqne`nVhX0*!zq6>zhrj)Bj>p>#wOOSiN+H_78Tr>={Nm1U
z3x)s7fBu&{|L6btf82SmiOp8ce7niEn>$olHc&fJmZuG$o`}vSLIwk)3lB4@;gn<Q
zWMD4iyIyO=@!WitG7LVE1NsoXEL;7s&miEb51K~;H(xz1ZFR&L<JF@V;*UBFTe2QM
zTH;AZE@E4`aDo@i3&A7uBGRH3$(gSMf%na15WTLU9`o%uljgqRUT{c<FUh=9dlt<t
zpEPH9GCfy5=|GBKMLZpxvdKX7^$wq@lYY>yBcIubU|xsJwf!d^J++m=YjY&&9wUVK
zz!33QkHJsRrl%@X#8m@*;BZnG4|sylhr2zzFxZQJgJCW3uG5}KmV!S>w*P<3-C46<
z$#o`X;7phb1)vIK6-82tP{<vU`l@~54m%vt7yj|K?{w34SfU21nqy6<0mGR9zt6ie
zKb$HSDT!(_I?vhPo-$XiDOau>w~3q5Hzb}w#?@`<a!igX+of;Hn#snTN;Yy&;>8pE
zlHNpN^&456182iJPUlA-eN>roPBjigP<eKqp*VI;eV~l$;bQmMLyysC88*jL+2Z7=
zOIxmkb7@Sz!gdPDra{@Mu4Xk6yP92$QOd>&V3&WnCO<giC8wu2=6N|2yp-L_SV9;F
zUxC{)Jf8bHt$pKup7nPB1im~PdMcq~K=mphC_je?6PFOgj*;S{M1U$~2$gp^Q!);Z
zM>Tljgb>KVRYj>wRlHZPX0oT&6NE>A^G%DRD9KR-BscJlZE9UB(63@hq*swT!}EZy
z3$T*zKttkm5>4I_uIDP-GBVm;;q_h8dEdkY#OZJ!8q;-`SH-7H$+dKOUEDZn+bd-V
z99oo=aVBa`zA7{YJ*gWX`i?EYsd&@k2yBizBSC)Pi1^wVZqwp((p6V-?JxI~7M}S}
z{N(8m4KeQJafstQQqJ}U$Y-@MNU4S^xa*G7+5zXyrYGk*vxZ-i8U6)YD=CGILm_X^
zIb2tITY`3tR4NK-<>`Kw1ez<doY_2k6Fe1i%ZYEt=@_{LUU1NAX$1-2JV39!(6__j
zmP+E>GH;HOIN@w`2aM#jqAEz!3S%^_-T+5_JD@=r9KsnQ%Ogg4;!WJ0L_WU}zyPM=
z7T>V|^-i97LO!|Je|aV2II7`Kc+R0!!Ojp0<2z2{oJS7T<%P3+R7_(mrc&U@+Y>;!
z0(<8fiBu+Z@-1EP9Jq{8N?PA3pUNC#1(ud{^Id7%Vx{vO=<pmp96}Y5eMf$k-T*sY
znwUM##7sWapxFl<WMjP7Gr$R<xAwuHJ6D+$-8`eUi5fJgJlPk4fi_R#JUn~xPWGQj
z7x<&_s`?V%WcP*6Ppb`y9bvpx9`-O&Vd!zeYrRhF1OCE8FO2JCT>DV{4G-PJ1)uS0
z8VzH-iBRy)(H^|ug~#|<9_Be*&Qtl@^MxJ;8qO!5k%tmSHeF$p^3|RrJPxcj(gJVe
zAit?p@@&DV4YRyFWl_G~AFqbXVQ-c^6Th-XQ)p>pGWZ?mKvyyY{y<MBQM+G!u?wwG
z9s~whXf3aj*6zN7x`mA`v=vX@LK}gxO&Tt(gXEhyVV0EXA!xL|%ZsM!FGJaPesVY^
z55ezT%^jWeq`KbA{3Px?pR(APi~lQnU=Hk+LvQhv1I&hO-g&WQ#de!?YiQXhL3blf
z<IGSD2cOd&28A9rP48}~S^0R*Ua1$l!VsE4j9%&u<CJqk=qKwH&gXz#$w~b@@-PGs
zWE-%dbL5gZS*2E);#2c<n*8^9>;ZVguYF$`%V@G&=}986`mb!dh|Tu6m2`+>kauB`
zUTl+VI0h=I2M-<=rfZ=UYBFg7&Kn9}Wb7mv;1ot&2%AWBelZgb=X()DjJ(Z2E`xG{
z!#FX`qTQ>vGyD6Kc!(6UNJnsAWou|mewx(`jPL1b#&{DLdaVKQ+6qjXwhH3mQ+6Qa
ziGmi=JRq>*m{tY>5SobX5%6BcnL@ISFc<+LxA0^<SxyA69RIWq6hMV&cjMc*M{C-W
zW8<YNd1KsKU6h6c!3e=w{tO!a!Su?K)D*H;QA%`yNwH^A<x#uh8FU3x#cooQ<%&ML
zO>woJ@Rg!72#p&ehNdV%#){!Ue&-BE;@wx!6UP!U!FJb}X1FHnx^p>|w3aG^gm1ow
zz+hWBqQ!!zoq(0TQS_L6tfG^ac1j~Kd~kUcJf)#tHp)7=pCT5Tc=ggWe%@)#;ZeKZ
z$t5yx{^7xOmi_8MIqy<Xb4M8U;IDTm7j{r3I}?A!d++_xPw8vOLm#~cyBm|$Ak6mI
zbXo=2GLksisHO%L&2{4Ff8<l1Boi&D&a3HM7i9~5B{`_jue4(FKzC-S82zdPmR#~2
zm(VY+%G^qf7j!dW9N2u`JuUQpr8KR3d||w{6NjeG<B~^?I?rIa&!1L1V}ZA$XEA0e
z%qBV}T6IWyQmNg#og<+#Z8?D#JyeO{ih3Z#EeBZK4wEH)1AzW>Uq03@*fcBlCF#s*
zd6J2sPqTzt@7Q^I!z_)o|Bv1(dd*o;33CK`M5oB;y(fkQrqa#EM>VEo^vbwig06e|
zeJ^@xSoo7qKCYfxImP+;5C7Ajrzhm|ZV06}Mz6^JHiw=LwnycF=Rf)TzpG4s^us@k
zGc$X#;E5G4-v7tH{*TpH#=pHahLgjVR-SpkS0I?%bSUegNBQ7PYBjffr%c#%P4A4&
zqx3Ft^%#{Dp~}yP8`-p0HgWJxqBY5wJ%bA;#t^um<xL&|l6(zCDc?u6Uw6W*FSBY1
zzIu>7hks!Q79HdO5}^hGyd5en$KAYOBjrS&Jv02NOb5vHlH{5`iZl907ovGN>d?^1
z(3qFdEsW94X<XU&B~~HwvO0zweL5id1P|%aBfrxy8+`WI`7E@so#q<}%473YsU1SD
zbKmEA)*+mvQLfp`_UHyJt$jA4z4qe4HHSGn+Xezz2P(XVQ+I668E@G+PQ^Q!-21~n
z`IFXLHmF0xKY#Sk9S;5?A7*sV$=!THC*CWwU<j>y+11#2WvY%xWT25Z-<dh8&XzyH
zQ5g;KEvLI4h4a7yFSO(ybn}x&3ew=ay!{NlF3UZ#^DKdEZ-&mBf%M^dQyk(MacPLX
z=<1jPDzGR?&aI?#hNSNl$P&D^xO0~ngAe9IFaHJUCkH`Q5YqL{R5&Wx8p6C^DUue9
zWf<ih<ls=aiB;l4-|nK9#YrYfDk$(XPI!7c43HQ2TEp$)Qc)x*_%no*)cNm{vpgr5
zDpd-od?-o@%e#RUgBfEfuZc@U!7TKuu;s;ID6k!7!LJ-(!P9#IJ#_@4(2ONRcoO3s
zooHoRrz>&lB!s7R#AJf8ep9eG4wFan4?g**Zok<u6=!%6L7iTtC6RQ#vZ*pSzRL4B
zAFb$$cko*<3)fF@+$~<M=qk57!@E{Y8S4jgI}h-ehL=9AUHnKM!CTKr3rzNGVDRuw
z*g9q0e>jt8`BT^j9r@<KyTHo-!VBN^q9jjwL|62j>0)MB3|EnZICZQv#)1Q|VOI@N
z9mqUuTOenXDrL$^0n<Db*|{Erp1B6WL)R$#jUM>m2R@bd=-kLTMp#;gF9$q_4r#)(
z)EIgyq?aeBXRd>1{%8k2_;V#BdT5Z}JXuf6H-@zi@;nSaTEAm>e5yciz|70wu>EFF
z2zf{Y4~A8Fbgh*oz!_+~6lP*xh6J-m3`5s$U$iDozi$*6Z&KFAP;D645zRr!KENtK
z=a#I)tCN##B$oI`40eW<IGprK&x3gnFSGQRBVrhUbM3f&WEGH#U8TU8<E+k*oo|_3
z#W@U$`N}fgRf&zg+0N3782Rqw69<ESm?Wy=q!XM0cHm`U>n)4syg1L;{FIZ6I7fJt
zc6%y{ueyQ;Z6=*^1{|FO0jKDRT_}%I4y&u-M;-2*Z4W?Kr)<!R=wH&Xv8|YMuM)JY
zu5!<rUn+uKy(}*y7kY~qzJRS&t3gxS(}=JQujop6%$cLBBkSau^5fndgO=w^ZdayS
z(HGy*8T_s-1UtBZTFA)*zUTro$`bn&y1dVQpE>8gTO#Cqv%)G2?-FP1R1?<eZrZ?V
zV+oT}AUg*vc`0jj8RR<g>}q(yxHn%zqte6ymH7b{LV%JMzQt|u0}YL$1e@x6dmaEt
zsIGCilU7h0l$O}FARj!UZ^73g7d>;mUDx!G4Ad9&T;thrB^c%{na`$2P1$7Q1kO|O
zTt=a%zL5j;HE5R(a6*>zNG9ZIcnY5cI^nrv9?bLhS}YksSpCUPN0e{Fe*gB#H?E4~
zjVFgZR$f?rGaTaQMOJZ!@!yKDaU2d}>hN~VX7ub^@d@%~pcE=3aQdZK%8@9H;D$X&
zP?%u|H2{X`o_7&w0ycSfb#0!&R64g)8ELH<Mm2smhGTWxM#v{Aj1P`88S6q%f+-Sd
zF=Z^xjC?JE8N-j@!fb)j;qkU0<$k>W5fG$Qz-MQ(&Yl8sXc#Zox^w06RX*)rV;RH2
zz_n{Ne0ZJ~cPsf6Y-7pqy_a?NQ4aZ`$G!<Tpn|EfNh^~88G-Hf=5=|~c??OXl}51)
zJt>U77a|S$gsS{TkyOsvSW8|MUuDx32!YQK)!?Xz0`FBmGIaND<_?4qX%#ZdJ>i|?
za-Vm>_xLo%FvB3{DzIqjMTDRbG2(hi&Ilz6b)8n4f@K&Dt_-D~pK-9p=ql(n#Mw!f
zV`joGIA3I!LrxAn&}TmcJm#c<Ri`gJ=H(wIrmo7-_bKWi2KQ)o96dS{zQ`jy(hEn=
z_8IRT+j9^24*8tEo)5z@jK*azuQHadkjFT1OwimOry8Kp6I$r5VMN#B7|&;zE-z=o
z9sKEk@ris+(&a606%?Gt;YsSqr9DWI+3gHR!cduTfRB!}2d8Y%oACUz?EFc$&1q2y
zsSNQ-To4Y|ZdN6P2FKz+aTsxvPxgda6R2HtK*pila+vaJv$HoGIfRAs{v|=}N**O?
ze@3rzy^2+&(&u`2WX5K++|%0}_anDAZ)S(u$FFT37CM7ZH;uyo@Zs}kpJgt|7u|zI
zL*r*Z`+2uM|LjvM3+AJh<ykmD|Fl@yATorlk8Ei*?maTt0~36!UnwVg4D@j5z1)P@
zaa7P_@H~@f(JlVttZcDC-o1CSbU4etRSZ~S9>g`>PZ?C{&~LpeddrkOWr(a!AD(rG
zRG%c!b$^-4>8sc_8>8*Lk(5c5&Pn>O(bEsyfvw^R&n~k!QmkEYjs1wO!ePrpKDO@J
zPR@Dul%bFFy_U?$NpA}|Xa%2WrRAFLs|4vznjGO<>}uZADLRIS=rgy#5EWfw*YxPv
zHC3LDSBVebuDDAY@hWR_&%JWK_0$>Ty!Wp9)AQi8A1L<_IKHNPu@m5<H_xP1r!^00
zLyw*U@ePl7ZjpfUHoY7oCt}hUKq9gsDr1f~8zh3b%E6RlaYmVyXE;m<-_R#T^11aU
zdKf!xzKfMsHlTBD?xKBSOhBa%mX6m6E>7&rcmQq4g&%u&ayy0#va7S0@@GYrvPtHZ
zdw7B0<>kgB=Qa-DB;fP9KFDL<{Lz1-&e;2#p{5I>A$FXUTqkTAdfg}AOh3Y(7m-&L
z)dbY$F(6i|ecBkTmoiwFPVI>iB?>UTGI1Odhma_dLZaKNijo8jNvq;iHmB>$pk1P=
zz$YESsNe~Svq4c596bi3loTNGY}ga{aM&$aY4ot%?Z#GSHP3z)jLAj1#cY(L(1Gcw
z!p2x01#vAkP6dMl?yHVbS_L$dEPcO)BSO&`EXJa(nH$oX);GMSBD%OVhayJowY@H2
znq!}%rFK2-hvx!A<sVZpxG5y@B~*4`s_<(tREjBlY4eu9^KFiRZJ5fe=!8GT1@Dr+
z#wTYj9rtw>Qt)({<4t>!0z-uZO7luu>ebS;Xw(`b*G^0fP70u5i1YA`3Rr<LbWWEt
zr|l@M^kX}{1}FN4ZxnW9mIo3#z8(6UUb=wDH-kdXQ5i!Nw#MjjJ{kNOmz~PH%W^p&
zaL;hdRgB_Pa%o8%S=LMshR^6mbD)l~lw}^sy*h({6@SkknpW%JCSGz04^%Eg7JhSB
zzRq&tJK?wZyP^?Ki)!<%{2oyKEiQN!QsmzNh4p7iuQ>FT?$b3oAr#--D^nZyIwFC^
zp<Kchv+*~JajiMbMRVE_&^Hqb@nk9udmR)+^0~^!WLD?UJ#@D9pW!u2Tj`jtZO=IA
zDxJ=sNILSugB~HMa~}fMPRfinSn^Qb+|xVu<jkiZ;tV`bIoB%_LR*od==t+Ze9wej
zJ9K((7{xH~S@_+r1SigQ;Am*Hl8r1XkMNNVD&G*EYk6^W>x|H^@Vho8`pwbixO7fI
za3GtbJJX^zl%yg$w(2GHfk~F=ZSyLylL7Qn_WW>Qrotav6Pp$LDGyEqx|BZ?_`~7S
z#WwoJRvmk=3NrEu>|F_Sb~pisW9h<nDk`VIq@(EMn98FF<+{&v>$&&fHBPP@Ja_z-
zKojA?vPp{v;|zok(ZQ8l=*cqfsWzHVZC85BSfc?iA9lV!r9qakkOzGr$MPm|<mnJ@
zUgxH80L3TvpFPjywU$Y9m-K*WEcnKao-R5|$2t+19mfBZD_>C;jxtmZ6ZWNWu_Ld7
zQT3JXa1QD1)Pv)Ul@|aANm^opPm>qVT%&jX`+R;R5K9;w=K2r!+ilyM8Kxy~Otn?7
z{LA2FbfCbnH!S221h~H$A_3%f2hzavK|GvjSJ#RcHCY04%Ad>=$V}Kmu3-y&-_Iaq
zB7vx}S}h}j?500yp%Aj2E(zJ8<qbBEGpAcG;%&COH-(ze_QqQF1<<nj<u89(1EAvL
zFo-*jM`qS%^CE`f(IY)BdDmImcW<4WQJ+H0v)!$J>8uh_N&41=+1tiy8A^sqmwNVw
zY`QO>*|s^G9L1P1oE)R0qf9cqXOfmKu|;=yD7~H|_K=bA4r`|%@Z7`Iu@a4Y?;ZoI
zrzaB@^QTa>i#M&rPQ1I7S<nH{uGmCmn9h<%jln8-(eDL+4Bf%-X-%fk;7?^08Fe$9
zw#rgO7@ok>Vm*p8l%G&DhdD0tQb8J$>JSw@NpF{4sjIZ;7~bQVQJWT?RssCOx9l8y
z871hvnee^6SHWQno^DMFWFAf8h9-p@JuSq2KJ!lOq6ud4m!kXyC!7k)!MBn@i#n(!
zWkSf%rD}LuWqg(8si^}(V{|*O&+go>5s}uspn$1^Tc<1hOzZ;ELp%;9pcBM_v&_Jn
zo-{2{_{Q;wUf?r>FYQsja>@DlRdn{VPoLfT#V<aM5lU;?ynyKbyYD_u3q9Xpd6<@Y
z?(G>c6N6{@?if~h97PwNm+#}ekWnyK=IS&C8Ius=?BNX^8kr_mga_w~81!-I!=Lnu
z*pyDod|Kj3A6QKbY~&sYS-s7{NiRZLaXE!gxRp0~-s~|U!B7$1WiZiw5}l~i8=l`u
zPv66|a_{9#m;U;%|FUU(D}<iw-d2ek+5t1b0zc&@dfSFToE7KG*b>8|J~Zg8XXoM~
zLwM;NjA|uKoV*t)r=eAEs&WcXl|3g!CGg`v`?ESjY!HW4<#UunCDCEtWKY-HV$E3(
zLRU_(73q=j0l+6&al<ZICT_LOXZdKEVV>&@6YUC}4X`ynJMvy^5{05GY={rNsl@EA
zEPtER-Ol&5^p4Ofj)d9~9@SBHC_L#NJ%(_F-0Q`BVa`;_un#pVKiAjk0lIzv)}7GP
z#zvmiQHp+c1yo>?ku-E!(DP%6i9G~EnW*e?U7c9*I3vVeTOS$In;pO;I!^fv$A*%;
zriU%7g+V{S<Qz=O6Fr`wh8{Na`Z{_BR_}2wAu?p^tJA@uPHOCcd-e&B@R4ocymgKp
z`rw<4lq&&0J=6G7R`tB^@NVQpzE-MO!B*!paewv8?3kP0C%t*fxIEz_KlJ7%$0EiZ
zv*uA~khij~Ja@UeMsaw}7UJE`qrIug>$inEAva#4L1;G3d+EfDIDd#sT>&L*<ve+y
zebc|sqHO_6lLYHDSnqfTK$KvCTNhhFc9cR{?*eVbzHti7WH}*<P^5ZF?AodbjL`Gt
z*W-Nt{UUQBF3wWm2t#36Fvq)0_i~hqm(D<8_!x`yN~EGO=KS@SK7K^WQbAIz5J;$$
zO&o{oJ9jc^?%woz7}_xWqDMv{lch8i)2wj{vl+w?^ghq<Zj_(bOZledpKg;OQ!*(_
zzn>V97=10p<Mi+d&7&<FjzB7j=Cu+^;XZupCOgerg+I7C?_+Rmwi3;ejQf`Yj3&-P
z!{K=zy52`#7S$EaDX$j(aM}i9Dgw0Eh}7OpN>0@{UAdQ}KU<0bdbVILJa#67>+rP6
zR_3kLuxlw21389`?&NXsK=SdTQ+)3kA#urNSi(5v9b-`(;J@h`Oc?=rS9hX+D$Onn
z58};7!aT$8eR$N6bbSVZ|G?57#h#*nGfl1`uVT`Z7Q2;+fYXXsL7b;z7gT7566x$k
zh5`*6n6yZ5;0PDA>Uk=ka!)4$y@A63nb*-by;v<raYme@NqLiZ$sAv-$kGzF=^xpW
z+DI<^%K!e}5x|RsE*Br1Nl<P>e}&Ts`5~yhb%JkoCug{t10jD7!pVtB&oWnzmPyia
zGD0U?Ac(I_b4)o>CK?)wc^SO#WS7?VO6EPjHblU9=+N*W39#B&$TJ&R?TS-HyBjSW
zyX(9oygFmKC)<`cas`8a1Rt`nYRBPPJWhp*NhQ)=S}e)YN+wRo9ttK)p{6nC+qdH|
zFS(E*UWS|o6*m~eM~s~Nt%v9nw>So`a5+cHN_ionG?8)c$3`x>r+ifY;sLODM{jTT
z0D<e?bLj=wi_@U@SUGv!;Q0jdUwI~d)UmQGEHcXyZk-1+zln{RR3&+X*k=#F^Tm)r
z#{i%!xFGVB%?;pNXYp?y>$Nm=XBJLq#6>S6Th5K)xEW?MEXGO1jnMHe(BHlVv%HKz
z>y$@s^rCXhyS|4vVdq?0z71sE|E_nxCCMM;@yYHYS8xu3DKRNPi>P{5Vb671q%{H<
zo<KwmOQ1C%^W-;vy`zHI?5~EzRTOK>@0*DR==dH<8(|9ZlZ%5{!ATmNrsBFV#+mMD
zU9sj~=pOBmM!bKM5ee+ejIn;DfJKRQt3P@Cq!rDl8JGJ!8wvgMfBUE~c->kzW+<Pe
zHK})n;&4<x&y0HB=lkzJZA-+sErt4J*6@388K1ns$Ip|OT1}xd8<Ab6lw6;Wk>RkI
zGhibqhP+=YOrGYmbi6aPGWIq6VYdS8`w#f<dU{o6n_ra-9=>?-)&rM0g(M{>R}5m9
zF>=MryigfwMV_Tg*DwM{$GDgICuGE^i+4Li9_`{wMs(%q_H+}Q7lEH>WDxF$7q_zt
z!53WB>jFLURnclp;^`qGt0Em8nn|zOFCqNXYBb4Gt4+QPv#--?&7_(sA^4mB2{})1
z8JvvYPUU!=)f{x|zy~d*#8qMA6@I%%p8~AK>a2p%QXWI`GUK(gI!I)4+6I&y%*EiO
z$e%pY9vP=~wMw>eJ`{F1C!PHBU)}T0(O!r=D{a9;@O&n&8IKQ7^2~QAgtqGP{>wPd
z9$A(Of2|{DIzhH~tY(T%b*6q~ZwOKE1N{V(!|ZFV9284r%Y*h#D6h!}Ox?qgnLG`5
zTxHm)j!bx#euzZk&`&GuC`+Hgl+J-3!s@eJ!ejV6dqdRuiEM`|d`YC_gFg+CAs$c5
zeuno{%w%+yK6YjCqmMr7gg*OyF#3Wt=8MrGJD#G?W?<!M=5j!*tI0x_l~Q^!&0(U0
zy9Z3U)}v|;nQ~7jeEbXTpMUaYdXSXqY)F_mAAT>Zl)6XG;D6#<PvP%<^R;5c?!K7S
zd)v|2l@rIMt=Z7g#!xm-`z-r{*qPL7I*w6wGkm{yX8ZW#^m0AvCU8$4WdpeI%C_|y
zR{gF~a1-A_zSXoav%O%kW%!yTadIQyl!?31k4%?}V>6E0S?)tY_TY$Pmoj=gjjQP?
zvD)Uk+lcD{akP>~fB0hc9C=5$<3O+Sg#YY=G6E;~9=p$Ro@W|Jddw?tLRmpP<;U5f
zmqDJl+v7pL8F~zk7g_bjPTk4-u@#&}G8m20TZad4;*1AObuWnLQSi*W+Zlq^<Jv|@
z^JZ(7{xKGz8Cn{7ZP*4U0bTUixnD8Pb7y(6cW|+5^oP;CAOGo}N>KT#XB?kK#(M0`
z(}OB`_&d(fjWQs;z!4SLUO4CxRjw^H-#XM@|E2fhjy((>i3i6W&?W5ie!kE8Cxh((
z06+jqL_t(&X-GUQvY_yf5m=E5r}zzyD_;l^=36v&3q+4d)7^+E>V%apH;1cz;d}$$
zN-z(<@jMZ=BtTe<sWFqSc+UGT52m%M`!p?xFdj1r>@4pHqTT*6ux(W&xt<GxFdo`g
zR0>gB9W{1QY6_ot3;;zxPhs}u%d@lDzh7Zs)C^6igc&O>mln(@V+$j^w@`ztO>+d5
zn7qU_<j|e(f(t`6m1`6`Mmdj)<4uJbL!Nl%jD#*YP%_8JX41$9tG7(%dhlt*khBUE
zqX1@|9^PIfN~qv(=g=8$3NrjtNQ}1zzo{H@@21H}MI&MxkTD_@y5+t^rBVrh8If!I
zyyQiVR5B+YEmV997Y3i=<OgKO7>Nz9S@&o15=DWY=wKv%Q{^XN=n`q}UHBV5(w!9A
zt$TT1(33L1VL%>$hvw3m^MJ~E$s4R~0QNH=su$8m2OKq7;N|*e{K_aWnwNMnhK5$a
zvK_p<_d$;mQPk>y`^hss9Es1_JvkqF)0bDeaRaN9+2SZ}onpx7l>4*GrB^X{qg#>L
ziJrP?*<CSs3uonr9u$G3?-{K*(p<<W4oT8!UGBrrBniFBgJH8Rh@s)2OL>bfp|b8(
zDINZMo)>yDzRK3`K9c7vyXI4w*94BTiLb8lWXmF#xrT#x^Q|H>*mB$a=UB~9hq*SH
ze-=BT#{wT-WFohr&(|sYcxFO0zCsR9>a+*AJUDK33h>SQ$Wxr%dqOVq<m9kT)3X}e
zqjC<6(aq@r(KtFdd|43^Ej;Kh2U=b#r<U#Hy}4U?#{rcuy+s$|W9`~Y#~~q~X>^3S
zk;|J_(Bo9L63pW|fWd3S=^0Ek<1sk}F1#JzzV`ZNX#Ab-#*3t!uwWEUo}}aq?8u%i
z-94M^iC|eg;VS0>2P(n~eu3EM3oq{`sw+_w=8TTkH(*7^o)(`x-#XO9!3{6biGJ7}
z9EHe1PnBVk_C!Wc$d!{zFQ=|}Ydhu9d?1x@pnqM}VTz8l7cQ_3dA3Kb<ty>=X@$-E
zEe~%T+>y7jR^)YRuIO_gn7|#J!*53Y?|#=Mx`Y%iqNEw%gYHqS6b6LT;kuuW-*FO?
z`Gs^nFEJyu3J|qIh{4*4Rr31SPaux)(gnZ&4llvV@r_W%#OTD(v>QM8ma%nw8N$rJ
zc$s|^vcyRtRIu-+2Sm&3$$QqZXXm&mz&4m#o?XkHOgc=Hv+-UIf>s4G&odkS;>dQ%
z^I<A05Vk@1UOZeatd1!rvN<IR=)D;8Aka^h*MOx*!E+OkRIqwKR4SYrxabPUvoTWn
zmQ2#x;nC!z84g&5Tby(=i)TLoEt@+z@W94wiM$-COax6_ve;@j1ti8S<L3<6S>=+q
ziC2Tm!(NKF3p2oZUDXJn$Ua8V1G16sw%O(QzT_Zo_<)u=gK3@Zv3a;?!M%JLp!wIk
zFg-1)U{BMcWxNx<vZq4mfQvxR;Y3J_?nVDpGIZ%p)+ciib*tM+QxDHf7PSdbD$vG~
zGI)pgGBvzGs8@RP{4HEn66Jplc=(^n<T~FSxysj_uTmLbMxNd~?%bJ4W#m>T0!+}o
zSMe!B;lr-@7$14681(FDnaaC8T%k#&wvCw81rG58E_ypX1wI|G6$meUQBlDi4h}j=
z24PD1G~>&e%v%RL@j^d7D8u+<E<yloK$E{KsOXuG$8mncC1m9JOxV7kazI{>(&`uY
z{+?%&FR1ETe*aB&9yPoZ`dVQo540(3Xujh*y;<({5LIT)A$6NMPO17h;OHPj=T;g)
zC7L+=)>03yj4poQe|0yyYIulFfBEdww9J2X>;L{g|F3?P_QmR%+QSvTjBAbtob)#6
zsl+q$Ea<Qa<|+y+Tdb5Jx98D8Jjdsj-^4ek#CGL|T$Nk%(Qe)TX;-!>_gbZ7XNOYq
zB@d3DvV;aav_j(IfjI_oW;&lO-_glDF?;>!mRB+TA|KSVFNis6@X&A}v-V;{Uhn0b
zH+2*v8xt_C_K9SZhe@>A?KC|k8b*%ttNcra6}lc~cUP-2y4`!^)m>P_uf5Z2bto|L
zip=P%VFJ2=SN8sim{NX5SE_@7L%-O|7s0RPByurNh8-10V~51?fu;e{1~y>JhD+C1
z>#WDEnTejGIYv*pR8?0z4sX~|PIem>d3LOUl-R{1dZaL(=~-x2AO&8tip@$x1Aa4n
z6I+qzV-J;O4KvGBr=evoh$_;1W!-aWclqrwe4a^X$^Vn|h|w)`cJz*+U9X(p^4$}H
zZ_0klM_{raLWj=te18o%#kO#*;!<bXO_vT>fyLor*OUe2z7!^HUX<7BK{%BGh5j%6
zB%R>5yww#lZ0qws)X!mcjJ&>)U!^sMe7GKENCY$<!r(ku@9VxCCVb=3A<^XX7R&-F
z@_vZ(@-F})&xMSGz~L|V6k<%=09Oh4!%I$-VFQA$fRelhoKcB^F-C22rpPLLK}&{F
zmd3MISjqk>6SbzG41&;Nc)re|6(lgK(zPY(KDg{1Tn#Z$R0inMCR$)~PT(7wTClX<
z6&%PTKd>e@H*%O)UK=v$aDs&5r7+eRPzV>hLVCCcw>&663u0IrAPQ6sL(XVxet}hZ
z&1_mxuo4yujVFszE&h!luA>uOL$irw;D*echgL~8-SX;Dg5@!P&8!)NFSiL}Y=U>v
zp%d?ddXC)Z0dSKiy<LBKNG>NeI+M4EV{~LMt$cPKZ&5-_;fK#-pb`haa^EC=9B`HY
zg5h@aR(uBCJ%gNxkrke&f(c`o9_cvEj*UwzlBA(e$0t{<^ucfpPp1b$CCVV%o>&i4
zbqn5!r}tqY+B8tb`%Uv3bBlr>g9v~4>FyO&oYc!1Cr#eOG)^;zNLkYBa2!Jyq*_s)
zW%3_dcBN7W2fr|+Z*qdtx<23$#Ml;U`Z_<Nr|w8Z6>K5qMnIT&p)F^XfK@896#RKV
zhNoO<aHvc<AaGH^%WL;mb`?!<zISe4k}NNcgTla1<Hgs6<M73}m-eK)lkXI9D9+FG
zPA^^#6?^04%u&LtX@qfVQ@$^!m*vD<1##k-+0vfV9FA0u<Sma^`NopX$PB%4K8k1Z
z3&_BV%#+R`F6U4PS`A=firzm<WA{pl${?>y7?ut{I6SXkb5_$>OI~cmWt{DRj$X(w
ze+qW)<>v(%_UPL<*vVgcGWTcl7~3%o9e72*^UghcCzzMB5iZ(^v(Jr?pXAp3_WeGK
z2c{suIo$M7;Y^&J#l)SrH{vM^4I`J%?J6EO<1e62ITj06x*I3><Y;5UIS0Jz<cHTB
zwl;zd%jUVV@+$V96TExP%e#Ucnq1}z<IB7Jls3*uVTXsD4|+O|s5y^KQU1W8lq_5z
z?|xCVa_VUaTbJKBCCwpzuK!q}`-Idmw6Nb0NO73^F3n3tw~Yp-l)!~(ib;_#W510&
zIIK&;Aw5O~!j0Ve{^rkCy!}kUtYXK=BzXj{V8*bw6)k@=xKrR$5o<ZybHcY;3^g!P
z*sCxy9-I))!`EqbFxFa542qr&Fm##m?mx}$i41fnoyPdSd?}9^M$wC6>{ts3+-U`8
z-bu!mDc%=>&0s&v_@9d8mEA(OvycMuIz2Zk(XIkX-uF`YUmKiF@SLIBJP1W)p=ao9
zk4oZSXA>pEBcJPvP9C;)#v8_Twy@?<1)v7a3S{I!?;e{SzQ$>Y@oZ&~xOea3N$^>f
zw+~toZPSKl<9soytd`b)j9rYsO6mMXX3S@ksY9=d2$k8;gg>p6Q;D5szQS#*XA<rN
zr_|4JVkiy7{SFOy(PEj$BWEqG7&*2j(Z|V2uTWa8Rt1oU6voM=wSgPnT>Ab|F661w
z&{OhS1rfZ-g51H^YjzQbQ3c<5CNaKRoSZ*Sr3tij0N>G)pX3KyJc-d=G@{|av-S#z
z%8;*a^c7D<INO5TFl$zC;jy8Q>!j1l<|N2ZkOw@SErq_mq%+x@B^4W^NB$~<EBtMW
z`Nh@BG4H!Qd+s^p?MXsQ^8PaE#iqLrxk00zH@zQLa&VH&!MK;!^_}2m{QF*dM;@ft
z#9kZrlYp~%OdTQ!kCoT9%mQyVn#-JocV^B+CRdavZQVmMI8lSy+atIpEs6GzAfu@E
zSoq0L{=U2Va(wk%sLb_XwrsB6kH7hwzo`tYlGAhZGWI~3eUx2hpJr~?fBH}V^Q{lF
zQ>ztlYzLlJhM{o}f5@F3`9=03d6r!*=};ZT(CR~RdL|r?)1!j#*}<r?s$PnlUd%gj
z04VF5ba>HhHqG%xR!RIky&(VXqmSBRZeuBF^g6-Q7WXP2cB+nI)J-qe>n>U6+(v&R
z&j<I@YZ7~?M{9<B(#t4ao;{2#9%seW<Hzr|EIfVsw7oi?W=|D8Kfn0($HCXU9}St<
zlI+7&U5kA&6kuMK-mN=#(p#J|#sTEqIp|8~w}oyx=^0rN6Z{2o-^+meVm(tf@Y-4B
zTII~UHYo8ihn^8-=5}ZXkFB{r%e&NB*Y*j?r5--z<v7kbSP{jXbFWTNMg*&c;MGa+
z(f6A+N*?Ul&byl{ve_vU*H-3{T@q_hs`RH9CQsoWyy$_(sZ~yf+TKeKn{Oih{O3RK
zSDN)csY~#gtz-w~rE}0~p7vhnnQ)Oj>Bz$eGq=%dD03b4X4QU%S8%Aq8alRJkXBiB
z+zy^_3-}P8!$s~VhvAP`JM8D;um^vvaOery8Sbn*@oms=rP#_zeG|(te=dXe{P$UJ
z7JyxkU;&{|OVvgH7vz=ef%@FLUNCx*$(%JjdG0aB3lZ5NPf5f|IfP#0$E4L5h))T{
zNC#HlMd>3PGx;f3E0A2D<xm(1_;rjAGotly6=ACMr4!7L;6)R@Wwl!Yy=x5k1%}CP
zlug$yMHhc6#?6nRhf?<d;1Jfc10UuE&?+xRilbubphDkO06DvcPj@v9GsH(o&j;8W
zmWpRuEI}-3(XQlrFYUr1adMUS!vSO60Jhb-;<yil@M&=1)Q~q&oIt}%coHxRJRDa+
zarVl)R?@j>m;K<`M7;zp!Ks()0^A(*L8}>F2VHA|JV?g!lgHczuMMUTJ(VtoMTMS1
zjmP9i_OFAWG4mRy&_$ofbRQW!Zx=J8HN7M<8kh@k?~C@z$NR>BdvFuzY8{8drnB>`
zvRHKXVSZGOjBdMJL;ZCKzYLzosf2xP8Iu-=$}iV)Y<mEDjBB5ieKmqJcZxhlcMM5Y
z*MkqeY9${d3Lll5R^6Qpt(~Pe!Gu>Ir}NcZlQ9i{a6&lgDn^^AX(_6VRis+&oD&nj
zM>Z-yNG`sV0zaVx?YNb94UIEh{rA(lfB(soE<x5q!^qq6UCa&%#F<plO8+by18NwU
z0NM$>$=_0YLxb(%i2l;#W|u5+kZj}+p0cOR8-n4y;!`Wo#L)makG)7=d>uf`)$vJ%
zd78Nu<OF?`sOa$GSzvT1BRn)|z7w-o#i5bF$>Ds+LnX`J!lSKY_kqUoImrQ@*0-~s
z($3o(3i7L5qa}7OakSJ;Mm8Lz;(L+G^CCQ8qkJHZ9dVD|@;y0K2b*{FBeL`92|5TK
z+8olU4C_qfy&zrfQtsJ>KI?B#$~*k9qN{c{G+|QtjafO+3r1x@COLv~Y3xeD@Wx6I
z{F-CSYnTm|H0}A2*K{P67p?S_>y3|y2`z`A<x)A%QF|79YhKWl(dY@>f;aehpsw%<
z55ts(Cefp@%W)-Jy&!l4kIp+wK8JkQwfow|oXf+U_oP!B=5;c8_&hX)uHYz02cI2!
zmHfvU-cLh&_&fDFK4QvBv#v)v6Tc?-Ey%Y({s31qC*;76<chfwx%_8ujtFHjpO-oM
zJi^-elpt!d9#?e$;8m|kQvti*Pv*ri!}UBbGjH%mVJfex1S3ES8IBGkqX<4~*36^B
ztw;mSSl<>xXCMcBz4F9OSxW@o@a#M~-Hp>CBq$1{;0r3FFMAyTK}1XHe64W#B0VLX
z3~-@<f>u<)Nq1*kLE)WqTN!!Z{T`Be9fdqMGn?X-CP;yid?@}_wzL8IhGXTQV#`a5
zWV%*!pENcnGqH?uCkx*DO^2YcCxf=;?L<2BEyCloq+>ACW1s*TJDx%_1@59B@rz>?
z$A_VNop1HXyR88*)7_#qdN)S;GOZ~1FbXHZsa}6tQSDI+;&t{yvx?}fxRPaLo0d+;
zXtQ64^!Z<hYg&mvG5pED#@4w#FzJa(-c1wEDCnFY#+R`S&{UeUUHhI>beavHHoTPh
z8YRw$Z}~W@5X@IF_rwHe(!B^wu<TqZ&N!Hsb2C)gTO!}BVc6*1;6DV^R779f+YmT<
zS4JnIh%qW!V0Hq1liBjG(o>^iA~$?;jh=h=(@G8bU9sRIU1xNC^i(S{)Z{sTPn)gV
zw?{s2(&L8CR$z-qWZntJ-Xy-&A19eh^G@~@QJH}0{g<Do(pEV=pO&9pCSRK`!zsb%
zk8{M}CUJ3TOESDv9_o~2Gq8v0Eiq)|8&4{Hy$!ZCf0Q{2jQ3@5F~oqD;=lC4m#vjW
zAE*q=C*W&&%O<_K_5RcBt8x6-{SF5?>HSb%yYEk0p#S*Oe~#1eb-w-bt6<B(r-6}l
z;@rKH3MOgriN4cKPF60;lgUs8svMYi^(y7=MfMW<dEkE%2P^Y4Ozf>~xOMASna4p^
z-~aG~Rv^b|>7ItB?U>%L9T-CNa9F+fMRr5wAek%hvtRtOy>EJc?3${=;V6(tSDpk%
z<xM4m-<_x!DZm{cId*t1?6Om*ApAUz+&R~F<m|9b;(w6cI}!SWKlz}J+oMPCwxay_
z<6pKn>9h12WqRGMm*H9G1d@GL{1k`e!8y9G498ZYrwz8?w_D_fy6HV!*a&%{g^cNT
z16p#ptA1j1-j6({%#ZyNFZM7to!%u%ow6gn9QkC29_ID{r5qj6$Ak_=Cf>H8Y-et3
zBh?-3_3QM+?s5thd9`sDJoNOOTfGwb?_GWIlVjGMkQ1McA^%6&w?!}J``>#%_i@nj
zj4u89*T3plaX<d}<JwtxvCR(oOx~d$d6}<Q2RKfVa>SWeFX(MFw~AAbe<m1*D#&J<
zKoSZ$gM~3@+((z9CmuMS)_jSRdpyJ&p=op8aw%SkpBkm_|Ik0vS0)$qESlU>422(>
zRg%2iE#-~8c#Sc%%x&UrFq=mJzkMp&j<FIt-FQ@fdbFRdYx(MC7F0<+CsIZXthnkE
z4n;J2z(oK+6im{B#{ggCdDAWdX#ohwsoWI+LX$=%N!V3n(8CZsaI`fX!*ZlTI!^cZ
zGM<#rJ1wiha=6#b1AE4J6^6hFp150>xo0HHTO{PqTU2>ARc2qbl53AhoCuXg=L94^
zhbA0K67-M3-UVx5AYml2=aIC+#E->n1;8QS04c0u5!wSbhst3bl|*c-XbO|}cpUhv
zfCcmN5>KVY`>pU`D~Kv)hRt())7sdTxHw>-wT^JoD?rU-;RDS>NPLC@g~#cgfDgxr
z9+#NXNGI}V^DQK%@)3{m#4APx{C4f454|}yKMDf4q^+^cI~BmB<BTjG?1%`@(LAL~
z(A_OBIx_lTNC&Tiubvo3=#aEdE&SJk7kE-WCJ*OCYwIgFW8|`%8;o}{fY2C>b41oH
zBQ$EubetEvcAjNe#a8U_Q(2#8SKgUtLr+Jyl#hAO86&%pSlvx|NJWL0o=blf`I@6)
z0=J$DPP|I%i!W5Pr*+mjEqkonY63kndN>&0dvDTzS)C;J(H|8GK1lZ<-PNzo13#om
zhi@0@!9gN^3hniXT;Go)By^=)V4JM@Ko|ObG^^N`l*%~Q;Z}%K$?0{ZD=H`OmAly!
zg&fz1-@+G@r;oDBD|YMg;2`+c%IokJ&VbrDPNnTyuZ#-YFp@+ac1aqQyEu9qE-E|7
zpghT(lfOA<w=D(U)*-ssveA4-FKRcWFKtPaNtTzHV{~+vBR@-YH9R;W<YW1Do#4>@
zJT}dzC<;@V&LNC0&6zGZwlfAdG|&(3fE3CzN$fZs1`d3@MYDH~hCGvIAD;DEe6|AZ
zD^3e=-YcHP$yownT?{{Bx8a$HaAQLZXC308EikLM>__;jSCOTVXI@rz<VOb=8(Qw^
zoVhkVGI1SDaP+v@a{pBv8Fmky+qTFnTp2XCY!|^!^!8pFP2w9$bEfOg!>yo%zNXCu
z9y&l=XV+E#zMH5fvG#fB-!fcD%42p?{_KL^o<#TqhvkG2{0P7Kl}XY0)b7(Vg;bO%
zGk<R)--r1rQ%k_WsIun1g$~@lo613&1-{9@4G)AeGm&~0KSwLZFq9$myt<K}_Aq#!
zr?ZBdcT4fpB78k9%9CR&3#J9v@%$(WoaL<|p=i#}KKs0_citvL`82!S)oXl{vGKg$
z@wLJ`wwwZL6$0hhUOvOy0+^P6@%N#+;9Cq+2<(@^I88xKzAB7mtSH(x@L=R7|EXAF
zFg>rO4Xn$nSEc*K^ZEEu_j(C_j4eFn#txYTyC;cGdWDHR(mF{c!BaLNP2@Y-?ehc-
zt-BN~`73bOmg*(`SqvArCo!NreHBZMNTee!<Ikv#JUvSxOT6Zrl;Y?fH@u-$Mt*1H
zkayF%)x&@ucm?^s2?+a>r0(Oa*b|`J(1%3%fw>QQ1K)EWfC>yPqB;+G2B$mO#12eO
z;uw<Tj2EH*Dzc&<97$ukV0%}GHfic)kWJ_*jp%$cUD0^cZtZkITwbQ>sG8V)zNKMw
zjD9sgE$+6iXN44lMTf6n&zGE406x|@UF9&wAu^pHK6>=T$Q!@(P@0EfD1uIe<j82s
zfk~2NFmeHdlf~J><3JAR>UiQj%$IE6NyYr`JMT8%`wucy5gqv|lRu$>r(|^*9O=Kk
zEP}#n@gi?z5UDgB_zbZU@bl<6dC>dn`{JinDJoiD(KYV<{iol{`N6FZGDm|fo@RXA
zve;JI#dkTS$-|0)yJwHOJeSk+?&Iw16IdLJXOBL=_5c0t-_`l?m0^4S{O|wU|5``o
zS+>m9Tj%3=Px9{j-~XWS;fFVK3@zpfU1nYpJ}BE9hL@>W?w+NGD$a&yj-USYpSok}
zU;WizwkIRhhR*a(yfe#U?<~E-Kl25wo?suXme5^qH)rMdB)jlhQNjKg|2GWtb$aDm
zel(U8oZw>hLL^fAOa_LDi+}WCoPFFM2SIry1J1tX%zBNU9LdA**{llU^w6#EfB$Lp
z=%V|9_-+o})bnTe+1rNs!lUX^GVKI$=it~Q%A|bg6);@aeDf^m^#~Ewk&EC#SFV#!
z&w0K><F5PB?@7a%1qF@v_83Qm-R6X`6H_*~y_qtc6o!w3_TAvaPTxxp10B@+=N`Uz
z%TXd*d#>EhXLL;1Rg?`w=WXyzp0QozC&*{E+P~Y~Pz^gidGaU@+(lPvk?G(6!#@RI
z&hgs`!&Ba8m<N5zDSpxgG_k?>_Fm@fNN2UA^2r$)K5HPV_c%*aM{+heF5o!WDUN$e
zdKzOnl71QqlUJRY6&Ee&7dFq7-!=}rlCZD-pg$Rokzw)(A)_g|41yDKj(gMwP?Wqa
z3-5w7<2blej)W4`=-~*(%X<P0qYz0~C`@>gNXg1Yv&lKI-R@_iOhgwGcz&XT`!?j-
z!yKMgX+ux1MUuuroAR)8SUgWf&{N+ioT-=+(NitC`Syinw_mj24(<cXw$5o2-a1W7
z`YK)1gc}M&<5Ui9G07rJSc?E8WbV|~NbmYgS}P1xP#lxo<|SMSM1dRSYVSav?E=>d
zI6~umkY7XMmgg!tWBGZmGM@Lt5fv82B@ZyA5b)kRa~!mUl)lb=>D$3&9FDwMyrS4i
zkb4fr!lCg8CYeeDSB0sLR|;J1SPD=yp?b!ssFYfuIGAWAuUwc*G8F?8sxU7Dm}|Uf
zVIsroSK<UGu!I)wF!U-DHj#<O3PA?R!4QBvtztTaM5?|z=a8m)2N$q&bQmc2<w}04
zvuH_E+;Ec25;MQ(OQFvbUFSRPd=nz^(y~sJ9CSZ8>p>~mY4ybLR3F=`!ssNLymE{a
z9$gLY@Uta8gOgX~8M-)4-9x~m$OkVEK_Vfv(19s4(V5^dAN5KrxsJ20_UVN<NhO33
zj;CPrzmphi-R$QXTDq+K2Osf5dE@m8-oegtQ%?<VkmgnJ<oKx=>4zSKm(No%ry|s<
zG*n_Do*_@ph)TO%+0h~K9j87_7T6u$3a>6BYrQsjdUBkNqCWUR9fZf3r_m*}Sz=6Q
z^j4`{@7{S;$IG`@IQT6R#V))FP2$m&k$oz$@S3`yh0ZxtUUbC=$SjTLR7j;8om`Z0
z<wH-$?%~se@jMNf8L|zxA`3Ft6H(e!jycGt`blQVXC&u3M^vV=gF|G@Mzws`UI%AL
z1Rt;ky*Zul#BRhcYlIwSVyo@)#id8e*Zq4Ll1oG6{2^!k_N@;-_+fi%?%jW<e7&6>
zMsPS@9D0?i9N2?ae#ukWgX~tjV&I0(g%&u~nN2!pJtOqD6@1>0P6j}(Q>scg9*c#S
z;*pPd{*c-qwxo1RI_+!Ue}c%-YCMe-GQ%4x`#K?D=eW?L5jk;+JFg?<*?ha_FTX0i
zFQ3QOh4-~((cgC?W6mPEsWZP$13*K{oT2A&R5cW)hdemQqvZ{y8b<UA;^<%l&^vq_
zCrOW>jV?=Fc+LLjeTJWE+sQ7nEI(5UIzmuVBl9xed3yJn5D${!<jYpU32)Iex(1uV
z%X4pD8zO=hey?}#g*mO8dGw9ngAdUt$=h&6f;mi0eSj0ktCJg}h|#A+gq+E4Dg(qz
zsDdQNoMW7X%NyOv?g^T4JWh{;Wa3T%)hjS91)khVx|pgsQJ|Q=UbmuTQ2Q5PY)DR_
z`x_x6>U};M_x4q3T4@#TPN;eNI`rhG4nfXKpvRexyV-{#6Oe|o+{Ex)TxOTdJow_-
zufq$IxE`0L;X%gbQV`Q)nsMEHZI*F~A@x;Xm0Cm6zKH_3J6d^8g~}kM(5iIsGfM9J
z5tNZ4$aHZgX`^&`c9Qk~9bZiz8%YI0o^Ny!=MT@x?)4ixu)ax4+WdjYq^(bi&G4E`
z3)kuDH`Y(S(y}5K7bZ@5@8e%8Y2AU0{ZxWl24vTvle?L<{pjJX*YT9K9N?wpW7#Q3
z3s93xYoJq6sYKq%q^<MQnWU@mXt~k}W8e0bU<Aw$*!aQS90GabE&9wQXaSU#j;8Q;
z-bO00y$axVrhbUio`j^)if^kCPQlZs<B@C<89bh6he~-VpbD5zX`@jE%fM<?X7kmy
zP8cRb2)C`U3GXgbp_0S#m5F)5FDnNW-x^E`9Nibd=NOn90p>;ceUu@*(aXR~$F-GF
z*Kusm&N9g}^HoAZDrTw2F}un;2IS;qcHX>rh1Z#Eon17YFVjmDHNJKBAbUYX-e;*4
z73w4NPx4}h6QV64k}lG@m!T2<ukHDg-Dtmho^LE=Nbkw}kCRVYi)o3&_v@^}VzbQt
z_xx^pDzy0SruXdY%(>8IKb7(-qw>J-WTM4}pEHCJp1qnLoJ$BKywFpUxX+>o()=Te
z!>j$ZR))Jfil(JG--ddS{W9La_5ORAW0FdPoa+QfcF&%LX62zU?%w(@|K<N^h4?T-
z4b|zCx6WI-<4dsYoJo}$0rPcdc`1kVzI!62tgvb351Es5*-ATQ<y?z68HUb1?rz16
z{YU=mfBk=N{rJZ}y7iy`^Z!yhR7A?%$63jv42HxEMLp{Z8*>JPuB6CNkxGXio`e^8
zVqOw?od-shGuHqZ7OHw>9@158_p4XwiOpA^TX}M37^fA$QOUvmc~-4GxNlCEPgQ`C
z{auc;uaa_TXWgT~c!!P7o=Fd%w)f{xfAkkgo8GW|m&RTffB(0Cd+XP~`t_|Z(o3S(
zj{VbP^Y9_qhPL7`rd(c!uNQG%=><Jde)VRNjq)KA4i@?h)v*=wac^D>zHM80a6H7X
zjSyc&F>r=dkl9q_TwNtk^JVO3ldl}>RSIt`$!62=oSf;!^dd};2z<M9ZFv86dW~L&
zhM)Z8|4Ci&*{zSF=T==lj6;Ubzxa#4s1u^>KaT_QkN@zKt_-qji+y7Y)s1k{`+1(P
zX~R<B0A--CJ#wADmEsIvtIW*JdYbu8bn)4<+1Sj^$mr7hC)jU?zM6BS=lV`$NGI{O
zPDAME{KKRPh1DBH4h*`_PO(cj1G~8cKB$b_08YcS_J#AEIAl~fU@U&S7rYZ<wfbkS
zON!)(s?Vj*ed4RTWZ4boTAnIH0EUR<60QWv1+@1YNl5g$_`+iCdf(-0&0^jT44FFQ
zApTCfuI5D)@4Lex_Tl2U#QQDp=kd+oN#7@-<&bP&=K)+W*91-lX)}7~TtVbLR+S_|
z?n?rgfYx3&UWB&e_O)CcZ)5kd*35dZvZEYx^x-I(XOu9?Vyuti-A=)6Ax=Rdy0dic
z^1?bVVr3z|qsl^tl_SFtYV$wdHLOxGC0=G8Hi`|K8a)o2N{j*0MLqLrMrXjaJkpSt
zil?+C3&SuBOj}R6wza;>@6{_~_<0uxR{q`44SvZeRYCHDD<>ocD)byBO$(Yc<xk;a
zjI@5!o+@2!?FGh3DtU!$hJa$Ul27G_&MgZtaxjqM;>l6EYBu+$)u2++N_31<))q?e
zGLOKHpeizsJ2{eOEB4?3H+s|MgOo(!uK^H~ahkDS^9F>eq(;xGkxMpWGd8};x<lpe
zhVz$rqQd(G%o^9=8$-oNSb|%bCpx&G)JZ6|-pzq`9ioejyl14rwLSWVk3KilNpFjm
zAVbdCAb0X1H{^2q?Tdl041X)1r1W<h-GtL;pJoYrbw!00Sw~OcZ(fW{IDs!RA@nTE
zSi!)1Ln|si4Freh___mm!pYDQNB4VAvu%4SK$BxRf-0cm2gTqvjxSxKCv+`NX!Uir
zr>8f@_6<wWcgGiTIL<OzRfP%0$ly(u+Wx$F|KJB7w#)zB3^#d?Pu&PDaP01Cc*9aw
z`BXNE)5;@xIa~3uD-^v$U^JiLe4cCbOv)?HTgpEiRaz=~c~#djki;Xjd7qE4B|p;Z
z@N{~cq8_(SFD&C_Ggj|nV7YeajjB^dmeH*@fBxtBNaDNMB=1Qp>96zviJQDNY`*$3
zJzA@LkG$h#1*bQ8|1z6NV!U=r&r4*=pcOv1`>BPL{W#pO=}Tn6E;z0@T-O}NNu-R?
zGt5TcQEbuc$c^Jp7kmKJ@S(<rp_liTlc&+6OeiN)9+Nir_|wG#<&82Oy~^P;6kmV%
zgRc0x9Xp_B1b>y;HiTk%zWVCL@FFs$Bh}{+Y5M_mPJ^!{r}>Os8NBO^L{|+rz=QKd
zX3wJ|deU01V*kJ(O`jS-F#D=p8!9|YqZZ8S(JFs*B@ZSIp69sM%O4=g=|Fof`d>Y1
z?7=fx1q*gnUEwa#lGxJvDc=M0X1Nn5+U94{K==U8jo^IqMP(laqVzkbq^f9>*z-Mv
zA=!=tZegQnhon#JfX`o)rwRh5L%e;x&pX}!;Ab=&|K5d?ARGkVfVK#R(Vil)c`|3e
z;tNen;d?hc9OHsLW7zSk?{hAn$tLm3@5FJ2llaaKOSQl@TiSDlz_$>JBX2x9m&XlL
zNr?9rX0N18PAY=lN9kXOE)l`pa}X|6Fex1N&@lPxBKx0&vc_RFR1)IPA<Ic?^#|jv
zWvvzM%(z`9z5rL>?~txvgiJx{xTm~B4?~2F#Ra`7c(%4FJqlh9yns_3oD^opOgBH+
zo>gvz15mTAQLAjxnqaGHm1pI^D1>>_lM!bm20-gVOT9}JQ^6f4U*Bg@0bm_G2H)V5
zJVCA8$#osgDA)4EF9uJVjiY$z%ul2G*);S)o-ggIV>Y-Zrv+1`plkhg=oi4+as{zn
zqKSJ5d9UKbZ*o?#-VUAhtKx8)Q{_FQ7eIAB$xmo+LioT)TDjBKS<An-x>{3l<*)OV
zlH4~+@bF$%&Tx93*^4LnbtolH0O$7j$nXMshHmq&Za>$WkIp5ZRsg94loau_0#&LG
zWI6EmL3V!CRsO>t{kYeE^_Tyl!z3ISeCW!q&~%xN#WsJ<p~3&{Y=Gmw!zaGw61}(g
zj4pQ-8Ar@}oW|!*KKZ!$|MZ`JnmswPbT}0jIqD7MOj&)U%f1s*L+dUR^0%oOXU27P
zz_WE<#{^3gj#7?fB0ux0I)OF(P@%cj+J9ToO<5#=HW?4c(Ms=0(yG+yAK9U`6KQj%
zm(45JoDXt4JALQYqqNeW<QqJneg58I_fI`GdJucW$+CKn4YEwRy}3E-7e*_-;~V-c
z&o+d0JI(?}#C*AslpY=O2p@v4Lz(Kvk9882lj)(vrwNf){x~;!SL$R0e|Argh1Ev(
zwwYcJ`WTVm?co@mX^(#8J-loqCj9(<<_5m=PUa=Wfz_*pPjy^EhhA3mo;a4X%WZTw
zIpDuxEvpRaEI!~P+o^n((C8^!LXW3hur-EY@=d7d*ED=h64wxmUgBNs8y=CNc%C1n
zClfxK9`BPuX>?gQIjf`S4PN&TKu|sZj(|?g0<Surbi$i8>EE+MNTmMB&`}GeRQ$^!
z^I*a<jE_JLAPZ1uOi?V97LK7VNF=*&ARyoT1xE8vie9gHMOJ+2dpJpv^E#Tz@A{GM
z_YRcxK;6Fyii$*+BY~)B2+}nLKtt&vQVJIO1v#PFd1KweChbbU1wTBC0*Sv4?}}~U
z?{Q>s_j54tA3o)6U#;}HuJT0w3_&6366@6sMJ21iFUDwrJGd$6Q~1%!fr(*ATKd;K
z+ar*clgcFTO&XMk0%`)SJtFF*CvSPm*@_F@jPk=LPR4PJo>pmF$P7;MRUm6D8Ji4?
zF*Y?o=zu6ZbJEq>R$1uc_>8U%J$e1jAWaT%LtYB)^oW?4sL*zzYR+wGh=3Yr|2gAj
zkL8Rgz|A|jUWXU(^MN=+HI@Ox&$iX=UXF2Rd@RSrN3AfGfC>mQuk)VKS8>2=9J6Wd
zk6sK8@W7j!^4<=M)|!A8*o#vqDbE?ItsmYug2>^=jhiZIY<=>#^26RXnFEr@XAdIN
z<ZUiU4ZlhuzQ=in{lJqKuzI5CK>5X)i+9*8@~nED?3g2~PxMUaKBhNkcV6H6L0XyU
zHbI(m`dKE2Z&)MkB;2sWd$QOxd4u8Tlph%jP-x3A%XD7^$Gk3&?d5SV`e~bX7514H
zt*-iv?UMmna2gyc;hh)Q;-gnQ?>W1s8@1A=Cv-PMBcEr<t%~RGGJIq;&b_qA^`fZ=
zuG6h;P7JzkI<{KB%7Jcsm5JIkB|is=D$gM^M6tnwIY!EhN?qVM(uaB5M~NdInYd3m
zjErV32*-o77rtE0oF(|3W#TrR?s7()%?FZa(+|FIyOT2h=GNm5;pk=XQNR~*&Nx`M
zoPU-I86NHJ433-!JwEt%6gtUv%N3tqhq(43!Z&|AD6{_hERWrzk)Yet#oi_|mdLMm
zmBXB&HL`mgybRTSFB3eKKfRS?(4G{qR@!g!X>`kC0Hp~`<@Q0BOmoWS3+(-HWfWhd
zNqPR{lg~0)Ig1q1OHw<Nc=9R#=uPq)oh1k51-^*`Vw?_o)(@A4rqkRTrteA?$eEWU
zeW@KCS5`;oQSKPqSDwjWg*vc}!ZDgZV9<qkxu|^mQNTQSTdq7`aw_ig4xWN6wr*9H
zxtDiw>GfdMJSxR`)wF(A===QC2{Fr=vDje)gz8-9P2Q_qid-|yqm@ARFk_hZ^=62I
zvx!@19^y1PSPR|)_noJIMU?MM=V5W$$o|$?-~9qjm#H*r+!Z8BwS=)#0F8Y$Ag;Yw
z{M^^|Mw8B&RcV?Kg=3{H`PAs;W&}LEHwL=Hh9!L!7@Xv@WhA|8g@|IRtT*q$aE+cG
ziPv;m8oA!vWve`!f3o$Vy4NYt%~zgD&zM*`sWNfTSZRTpIN0^<X@#lSTd7OA;uxEi
zuhz)Lh5Z#~j2J&ncD0o`<IyhhbU&t;6@Qhy?r5MCKGLkgjRO)%wDOWhyv+x(cJQq9
zp(_ttdFRY{D5O2--RWux!lCphHbY_h4yVIW!;xon0nC$567#*9aV8#Sm_Wc6`G8GV
z%w?%;m%NPK%Y&n_oshoIvvvaqetR9j0x-`x7UH<CK6{+Q^B~i*DIWY!t_@M7b(<AU
z&ByyNFLhgC?ew{82#5Ly-x+4hFhuf_J6g`Ndv5C7b~U$h1RyX+-zZA%E9%g+&Pmo_
zPE_;DJy@ry#5*>hilO-aU}XL>U!=8jr&cusdozI5)*Jd459d9|^m&(oaeO19;`EZ<
z1W&tG(zj2tmxs#8W?gquLD@Cbe3K79_%JIQ9(ST4e%*9J<TH9ZPM|S)j)-|MdUE7X
z|4;BYcb}$5z&kfzeEGD!2Ifkb9NU$7ajYT7R={b7m^hw68l4dmI1Lq8_?EaKBVoUp
zu6NEGo^nR+p!FMJDiSg*|KfP)39-v+T>F4Z-pA=th;yPx(nkx){OX?jso?ee(IZ*Y
ze?1iU@5gbAgKHx?4v6_`Dr3i|nI|G`x1bNso^g;~rYw+~$-&NaQFvpPddk3FCa=;)
zBN==n<W_lW+08pNbCf7_9nX>z-N-ZF3*+p=jcnh4|9fpPP0f^aflu49a(K$#$}R7J
z%#DfSl|%6H%vaQ}<23s+IC{JDZXETmKmH^=TQ<WpKPD?{qQmIxMAh&MJ^j2!oFhDJ
zhPhy?ZikoNE3fpR-@Hm14muu=6B9d9Y!{Bw25Gd=Nd~bpb(!b;Knt3-otO2^fPcd9
z8Qs%}K&(37YwspnF<<wSZo~4PD3y5<-Td6|ojn6#ap2~wb<J@UadO*G-bUQdJ#kH8
zzWvNQ3~<7jgxyO}rX>DWs?8+l#*}WQm2Sv&HQCM8{3hDJBm8?3NBPm6+)>$OX%$w<
z4jtaD!jL$dR($RCLcC5$c)_6bc?%q%gWKkVyLqoL<Szfs6Cb0T0zQkQ!fLu3d4R{5
zn2&ItZsQ6+%uee5jba98G%-qvuTaQC3pSU9eVo4dZSdvQ!UTul4=M1?HQ@lbXAErZ
zt|HT-=7`kcN*WGHm!-xSwRIk??O)z05S;xyXnycdpMX<w!(Yay6y#%f=taXg2#J*c
z?+ODlxxhKb3OrFucQONG^MKI-w=`wK9_Nj5+5ATb11Al`Z+-w8XFh;d@;CzLXVWt6
zMB){f!y(AKE!pJL>)^!zk)cYhGEW@$4m8Vu%b<oMX&5NPnDQ!(`6iX2`yAqve|tl=
zLCd`$Uui=~&WSxt%0m~^b7l^WO}|uy@1}xM!Jb+@l5Df1Mbp=z2}};u`FZ96WuhkD
zjV{285g*>ysg*H<8CmM~XSiOS%+mHfKjSpKgQkk3!$NUHz|-RZ@fz-cK_6cDgE!Fa
z4Ra3QylTZc?}yhaLc7TNz}?YBB0T3feqE;u4WE7XEY7r^kjw!{_r2kaEr;0y;$G$6
zVcGCsZ=c~9n;_B&NL;sE<Q)25hR2^~a-uTucMpHk3j6yX%qJnpwNA4{L8W{O?gCiK
zUtHuVXT1)13`^*6Q_kB4YZThi8B7P=<V1qWv8Z#Thb6i@`GPY*+vAz*%PY2mw1azS
zYfoR!=UU{ILGs1#(036B=FQim<G}Enjxwxf7~^ivoV)wZ*~-E2)#vHeGDpG?6xzsF
zm*&VTc}Ko<oJzu3aMm!9XJws4p0y_~FzU3Yd~jsh9K$2{$?+hw$5{cTckA7UmLLyq
z@(@o@esDn9hg0oQ;OD;qF7rJ6jQr@e&ofvBs3*;eCQgXn%+J^4TuYwi!TEqQhXO-s
zAKPKsZDpGr*aNtmWX*1{D|ic!=|vfSbA*+nmaWx5!zgmKiRkN0AWzKLlyPJ|j|5AD
zbkiVCJOr%(M`6mw@VaCtfAnt~e_+N+hyUJ7J|N!0Tv6U85!XX;`|>yA?lk+XN!IJ#
z)%WcrKyELlMd1cflX7L2=S33hBU}N|g}d@_zYmeydYzag->bvVN^`ge{PzsMe@cce
zf=mG);!+4xQFr(XF2Qko^+hdl@uZb+AH2P8;Yxf;ng}DP700Oxf(PgZXPv7=3S$0h
z4Dy^2YW$UMUrSS3qv$HhbE`N4vN#58B}swvaE>K|<QLPJxg`Nu^kT%sPQLPl(@dC}
z0)oaGy&Avdh)2461N$obo@gO5h)o|DD)koVoTrsL^o0#2w}nXyun!hIc;*;mF^H*1
z@?@UbX_oUa6bGXf2HnzH;PB7g=Fu2*Mgh)~N0*PS>pD>}K3e2EFyf3Cnw?+R+B&o_
zrU8BHJcb_JGkZrQ5eH`XD3yN;5FF&WaRH0rWLWzy^6z~rMBmvEmqA9sPE-BK^A4Yq
zOWrFx-dD%M?+!HtE*%iRx&#M#JFUYbu;*uF+TM`hPgl^#*<&;ZK3z)s<j1hq3CWxw
zPG(_;2P4DLof^Y{m)20gn=W08K~MTuuVSnN*B3|8cqcj&ZA-7wgX&6GCq>@$NO}iE
zhn}H{XhU`7g8wQEPfVT*&|JHXPNo7u=WUf??hO@y<NHn=HZ6PdBH#8TrTnM_9KKv@
zm)Cdm-4%H19pI?Y3EC)Ko#@C5h8!q3>z&Xe0+ol^cg4IGJpp<S**$pa=veC6^P4#F
z@GDjO1-zG#+QS)u{2h<#JOpRMuF98#6N10f!k!*W^Kq0rJW?j5g~L_uxwB*}&m$|a
z^4Bj@$4LZ3WW0(?k~FiE@Lq*OFYe$~91rw7%)T;@vuchL@H|5_cQQ=*JgaKbQLvnV
z(LX~{=%j0KC7&{{w8pVcg{oJTZp+)*<n-ztT``oG^wQ=xdYYct@YUpFJzAXlcj3|A
z+Vr%BPD6cyo~PlSVw1Evl#klZiiNI@F2IjYwNVwBa?TAcaYB^w&$1kwo{A?u9N@Z#
znAt(A1Y1TUBddV$j9!_eqWo02p`~9lUN)1A#37ff>)4^l08elaEUOT$q}zc`Iwoki
zgm1}LI&#5VTGe{Mg^+jM1FZ5Dk=)5&?{Dst4!+9Q4)kz0==QU>XYrE0<K~`@t)}C`
z&UWs;`P*h%vq=srw+^3#34k#1&6xL<Cj=%Oj9Tv|1DQbD-Nafib1M8YBe4*7;}|CW
zE)*9W>3`pVh~G0vIfidwjIl!pfM^-OI}4LJ%eSGryppnoKYZPllKVqgiGivS0@Tsx
z4DS?@78Y@;9KkE^rO_h485m1x4{|CvDxAMozEk)V>O)qM8ilhd5d2S5a5r&*q(8P<
z@Rn6B9VSX)g>QAH^rk{sg_aSn(M~!A)k>J%YGE{6F&TDAD~i#c))5>~l>9dSJ#Xqp
z2Ia$C?iV{r4DmQXC#RYC8r+OYZ@m?HjPUeAkl*WkA(gS!`X$?Zqp*BaF%241*)kZF
zLmnEl&D^XS7`QMxXw_=RU-K&DIWTs`8Cl`g0SAV~Qt!;N&w&S}dFjYy;)0=)lt(M$
zq{STpPr<qfo`>?3dvWQjvE{S$U<~YhTHqpcSn#5i(NZsZ)k$W_BiHo7mef`$k&y|8
z=Iu<2Qj2<8TV#%2b9>CKvHUiQ(DZ0<FrfJlPnN@%yvY_Fsi%X0y)$m#`iJBJXRYW?
zj!Yu+Mc&I2T=eVa-ZIq_L9bK*mR7?}1zXKjo@5=+Dll^Hrgv#HoCR-ud^d(42UXOZ
z2g3&TGMe6mOvnvxTD5|C4tkT!bNJOSf87cWX&-;`aa+ONFDLOjx{_SGajArePVmVD
zU-<o}f3)%3-K+xoB-_kCEnRT%XN!K0k8%v>mfM18iF37%l#^U`oAU6UgKTA*6;nyn
z-Z1pl7K8@0o#fmKD-b%zAbF4<8OvWUW)rvzkB)r^E&yilp_~KX13Efspc4HJc^Vv-
zng3!Q&qd_h$}F!{a;I@XzW?FV@a0i9Z+qBD{WeH5XX0l+``J|5!5d5ybkT2xhqU8F
zCY#XO&vn!$!8I~Uyo+oJ&>qR)sg@ZVGW_dj0y4yQC&NW_f=$97Ju(8?3fIL7yWEty
zE6%)@!Nk*8vG>cXG(0~3<kJo@{py#$tPasdFy4vd+_Ib@X*_KYmC87_2z`z&Gml==
zfv0h5O?Krp={4oV{32WTb8yI(-Ya{`kmc3z<75;iWj2|xMe1vEC+FQ$t21QVfa5H=
zE<TZ4(<-AL3~41Bbd159_O#`K1}o#*46QE?3Bbv*;DEV7ZD4ydQ5kx_gLK~&<Xb1?
zJ74A@Vv69t{;r5OFc|{Bl^Ga9&feU7``fM?|64yd<9_Sazv^~j@vn;ejYz>8A9niT
z<)S1vbB|9_t$1D^66Csyv!kd~#LfMVGq+(*65c$-$sZ$iz(E=b<(<n$?^y+xCwa%d
zY(`J4+<TPBaT|KBG8G0Wq7O1Lr@0vz-~bi$$I!@kY;5CCN-w3&AzdEIvr4y+Yrauj
zg{ux!l~}xWz(d>{Pzx_l=A2A-Hih38-}$5POb<%V3g3YjZ)FHpox$L(;3>quvg_Lz
zCz-5gg-jh!a*H8oOG80V{F3UqapPhzFvJg8ExZTBz?UHJB;21r$3pU~@yNUF<bpRa
z2SRz0cq-JA5&rE0Tjwv)81q&V<(~(Ok8mvP;N5d1TZ4vyo8R&q9N|-g6L>0P6NlTA
z61vmI3#`NpFHA;$qst6Er9sb2jAfqRoL8o@8l6Qg>;>|W*3eI$$I98C#)EsW^$<)P
z@a)z)&Q5yu;uvV<%U{?|L@Qf~LLYn%fAB<e&hWwqyygrD;DFbbYUxOC9i$Sfqx3Y#
zED4_>#^aY7Vp+A2rNWl0-^sFNd2D`*j}FV%okM9VpEx#MO%ul^c+fF(QdAi5X-hj5
z6gqFFErk4Z6T7_Gq5Mo#>Beczo5+RpvULbP%2>FZOO68goB#6+ZhI2_vKqygoQg`K
zcNb=JZ8S83ui@HqDEz9d1G7C_iHq;T=(HYz>=Ik&c!pxn(j)RRy#G8s2%P4#RNm!R
za7V|{5oa(?VXlMW0T!^mL(@8JOm~&#_h_5)(Tu~-=%L0y8VJgC%SUiDWEPYc+`-db
zc}ZuY3x{jZaiV#Z8@ZldlouI-eQ9-5&Mz#@jtmR~mF76B=FW5#SsXMv{x&DM%`l9i
z$9FRfW0zh_hBXkrPEV1>*}d$+!f9#`VEE7>#NZa;Hobg=kpkN@e{{TdX}ukejXniA
z1l1wy>%H=CnA%66&T2VP*Pk-utsg~B{HDoyU4})&|9RA3(Fe_NapiF%*^QX{vbijq
zwg|jAbRt>_&%1+=6gpvYnCrv+Z)R7dya&I6rTz)IuJOE)WWDkBJHx;9{yP)xhc&+Y
z=HGTk>Wr@i?Iz8y|2gH!JSDzEK<>I3<NfyC=4D%oZy%ROy<NOvV0R_bl{MjgCo85<
zR6+B8?@^%js?gyn_zYc*Xp9p~>bd5<XRaBh;t$Evv6SvWz^s&DyckZARSd;FDxho#
zlb7yC-PblELP5Or;Gj@4wA=XKNlUj|q$f%%vBb2u2IFDpS>ec;zR4p6VHbkof1bku
zo%uT$50TKo$;mtL(zJ`?kiwbkE>(@8G=4pyT%OBj0qdzy&E9X9YIyxBJlz$U7QczN
zr7rp0&1b|_FedAfk+EQO+M$pPE>~q|R=%|F#P~35j4UTY3rNs4i+A{CXkxa3PU}AJ
zwVv=-9OvVt)$-954i<8PgmOIS002M$Nkl<Z{)9Xi@J8k15YX?dwO3bJm2+WrQuZ+D
zw8ZE^-ftEU2UITOE@+*h<EyWpXSIZl-=t+oBdes;GZ6$w$AZaOmMC|RkQh<A1^1~1
z4V9$A2+wBz#ONJtuQM4A%*%9>7njh0PUYe(8_GR;^rZ5?e6Z#qg!^?IGW26Js87-=
z9Y-P#O5h8+(ZRMH_AL7MAiIB(;k|rc27ljs?`bRO&!1%lN9cPBfh53xE8upkR0T5-
zce#iIr}eHpg#SS_yqy>U6hVYO%Z3s<L6zR67fSrAH(7CXex1pSnYfr%`6rpocqhXb
zc&d!+i6PHuZ#o54pwV%TOu(D4*$OfItk#FE9Bj^$=T@Dm*vv1`(ne!mOgUd?v>Y8-
zMn7R;T|SK*;%q6W(k7fH9h#i5Rd0vZt@dim^~$4oHD1bsBuPLgZ^v=z#+4}>9LZbx
z_K%GU(P~)Bxf|C-R$Bdf6ZE{n)2_45dIUVzD`nV|<3aC_!c+H_!?vOi0k-=$2g9l-
zytvk@lK9u@q2r9a^I(QB_JmJ%2#<YKlfL0WkSzXt80K2<MABPwOD8`5_}9tvi|*cQ
zNXF0_9S6U#f?GhvIi6uyd+l&$*e(04DAyl;_~ES&ayUCW+>-b3GQZ^gr{|Gj^fkI;
z-ieLj@YDu~Ld#dq$_G0RSJ&ohRaOy#p+H+9w4oAw*rrk#`Ji#P1?XiT<th4s>FeIC
zqMHYLyuhAV&No6{C`0s8s6C#G-ot&r>x19%vl~`}Y$AR$;YM{Jz)`sQ)b-8#M({%_
z-xZ4dja<Gn@pmQscRm|#{+>Y4@*Rlw(UP3&MKS^0B%iYXM)2%gk9Q6#Q`xmqC1K+x
za-D}LV5|crqy$w+%8)c#744o&WdW9<6UNOqfDWOs!58DZ1LLDYYo*ry=D<nU-Y&9$
zZzV%QkKl+-&s#8ay`2!}Iuhu33*~L*8>fM^;v8}-++nEn17WUuTUx78DO`gKhiO^}
z3}=@&Ht&gl^A{I}6I0<ZES3?Yp-14rf4Zf^&G0g>lSV}}@!C_8hn_LTCW{K&$<VD&
zR$A6zj&aC)dB7`&=i+rjDY$vImu<W5y|gR7G~|<nm1TJ^)5S|Pd1fqGH+-EWadC*Z
zkqD)`43ltzTa{b@=WMgC7`;?}Z?i??*t0sdxyG+3WdNpfjPiKC^=pQIqJyolWJ*$_
zCtHWW<J92`{Z$G3LTzkxbT~g`ndfcc2N{*ui?ks1q0D@Ou_s!l0`f|GTHu@pt5Zx|
zeD~dV>I^aZ)zk19J7`*UG<j`aVPhaA-jp#0J_1QauQJ4Q?pXfx7Ko|PBxooY2HZiM
z&X>yMUA4NQhr1VdQ^`)R3}iJ_T0UZpJ4rgcM>d#_%h%KG|0Y8qvKa?;^vqryoDX`_
zumpuZ0smFp@DR4{4e#+cc_nTul%(~%4I1f_8Br!mpMwC;!VcYrVG1=rjjjcqi`_ub
zGl93KZ!#>9U@nccDoW1kE0uh7%<!8Y01kpL{Te>9=g85y21?xK#DATJg`Ub+Z}PFf
zI4Ybmq$-zoe6@_6<2m0uOG6;O>yfWpx6HvJbmWKZ$asH~rR65D$T|H>2ijl9*)?3p
z39<b6>#wu7OC6)+%_gGLQre}P?&(Nn9mm`Au9`~&#YAR%)0hY1LxJyR6FrU^S)QD{
z4DJ~YUJaMCud^3P=xYyDWOtPncyMgwoXBxx*Q@e%t^y1l%3<cq8Dc}TUMslW3^(sw
zSN`JiQd)T21XgYfB&Tl&>Gt7554uK@E5HAikbPrfJZahhP(ZK0pirf!c}QST2ksEW
zKnVML8&zD$mdL~rImg6GMEBn3@|(xoXWmSDkbg7$+e!LO82{cUgYAZbMx=#a21E%J
zLMw7xsff_h8(zdfwBqUeJQuam6Til1(&HuL-J=RXA;=}5bTj4|=R@IRZ08u~qe_|g
z<v&B7LMFV4G54;8BWbpLC|VvzX9vExgPE%dyBqg3%RR+E$_7#<<lf5Mf;U;(r*&nV
zR)xUOPs<`O^Gu~oue6GUwk(sO&!ICY(_SBgulw2#i|F>fkg3Fi&q0Z>HgIyJDhf13
zg?{ELFpX_bi;}T5heC_Z>W8U#`&|&t@*2-1N<Pv`Bj}y6RM0laU>j@K(s`ASn=zs)
zr>n@&mq#~GY*y8Zb|Hz4kw}(uuek;>oTCQ|_vR$8M9iV$P#NeEX-`M4_t}=)_*us#
z>C7vkKa8>TdP&eNj{D)r&LI4V0qaVF$~LX#q&HFUQg29&<M7WuEcB0T=|S!mf2S7=
zFD$X7JMTTQPtJ0(lDAeWeDW~l82287ts>#Tj1yCl<p*xc_`;Dam$soxOSZT0tvnMq
zQEZz|zrOl1A9hp8gfI4e(CVj0TBIf@-oJRzxdrm{bzF`UeL2m1kw?MJT!F_=-fN5W
zae7zatxKIOp(O6?&oMG+rJZ-?)ZnoSS;c}6l^<&I?A6OJQ(5@1TP7(7&lhVu`c~3`
ze0{xFV{o97Px!}~c=i0^)>kjSq>hzqdks=C>am+LBr?2`&#)wXSWZY7<L~t3LFdC9
z^e2(wWX?>>%A(7)Ny9YWd;CRH9ok%1kAk~ZF}BWr9hzH7Lpb?do$x+WQD)a<WqNqA
zbvu}phUv{jUmzCg`RH4&Qyx94jZ9=UilCRMlTRO|w<>r%e*9i}XjlR-O(Lx>1^3-M
z@#1;D;gUlyi5?0zMR@;dK5&={{7E*3Q|7<QTp{}ZGI8a9p2_0ymY*I{!!>$oyH`=@
zp0Y*0=C@>sE{=yuqMv;BX}8+9vgfmGsArA_-C|>=mkTe`Fbm&GZ-iwdI|PR#afq@^
zI3Kk8Ap78a|AP-Y55!y$8|eA25$C23bXcSZ>`_+r=(*!)T4wzG>sbO``vmH+F3fa+
z)A`GFd(;Uj0iE>P1Y;liRDNy-@)AfwsJwFCv6=^Y^{w}0H+tGsZesIp2R3UV9(ylh
zVFZB98#`PVt@~YJ;~>s_4JN>4xGC**o)j53%Sc5N#o}n??ZoPXZ(Qfep$54bC(lvY
zV^V>#^W0DV$2{Mpqg)A^1d{*oZpmhzv7iYzQ}n^jpQafxG6;=Zp5aF#5@h(4Q+Vx}
zhmJWhhTh9}!dxg&9L2Wg^QcBcMU`hYLQPI`^S426C?oK>JH$^ub3HhKNgq_gyl$0w
zJ?TDYe4~z~FBMdJ4U`#i?JVeByRD1+Ok&mY^PW?sFb}O1m?0F<wg}yqUJp6}?i#NY
zz=>L!gyy#T*mTbu(7mqUE98v6xt$6Jf8mP1DwX25;<R;&e-dn91F5{IbK_kSE2a+k
zXeEg5nmpiEM=kfrs?*|G^k9xj*LZy%JaQ)AEgRW_xzTXO2RN?GR;8k%Q29*dB3%jI
z3D-T`;LV5=V<&{h;T)MxT6x#$%lXR0PJgNFwGO3YSj{OxkIKi#AWLEMLTR#Czn+_r
z>A{<}A`=(iy$^g20f*3o@+v4;@zCX%aW24b`Dq&YNjAe3VC!}0keSLmEziBTO<U-3
zJnMV~CP(vhm${QvGWm{=xcyReo|j$>39a*y!lcRP3r^0{l^!10wajoOJt>RB{BiK*
zJA9)@^n091-Q3EhN>HU3Cm<6;PeSih*rTwCf)P`h;753fj+1^MX(lc2vMp~8sQW}Z
z=oo%@w{?bY)1&F33T>E{Lxn{|XUT-qLY5j@*KaZ$D80fJSt`@h;$6BJKFd>&nLP4e
zenEDZm6k_#G5<o#WzJ|PCqY`y1il$=rbDx|T3O7;MDO~_u!+*M6qmywJkEE$q<fg1
zT=CNoj>*t3GtUI?G~8M-HlNWKIy1Z<2MT|@b|xQleta{=JP?ODNQVB(gTORFSkQoK
zc}`+<D|@Kvyp%Zqckk)}c{fh{ds#AU-xAK_3<GhltoRW)aHm$*#OdQ`qg|t<HZ5~n
z%2)B|F&!W%XNjv`t{h{R@;iSN2x1d+oti4o@(LiYS2zsQEIQ~#ORs!aKFvGvcaAI9
z1&5tWIU(!c6~6U4WoK0a5>d#6;uS>Rgc+T*5=tqauWJhXSG=LYlc1Y8eV3oN&%Mgs
zK63j<5Ay!KgF-C*(1t~&6F<-!{teBmk_qKvj}dyC=@ihncoaqqlUA4n=INO@NN>gn
zj4@^;W-dq~Egn|!lkX1w;91^zuBBrsAY)_p`Rgi03J^ttZ~ltmmPej7IQMWjIQ0sB
zT58$4o3S?6qTQMKcv;6~b)BaxR5zob5NoA$-he!eod@o52=fs<1?*6Hq!4yF7$XoK
z%f7{u!JDhX{LUXJrbIBAhn}IdShb+=-tMhhKR!665;+abVA7DJBez`i17$TB=B!i{
zji+l^H@8-k`5^X-Xe&9!c^iHkSLqJEdndcbhEK+(r>ANhzem}b({e$r9~HU$@dtHe
z(Q)o2$5sxBbDb_^JxJz{phX<t-N3h2G@(UYJZ7Yh69?l;Tkhq9UKId?Vayo}-?<^b
zGCA?+V{WPq5K8P@Z%#))nm+e-hNRuTxH{<)FK`)Td7Fy_eXT(sb?bL$&>4p^nf0Cd
zBCQycn_RY`qXSUT(sY(HC#g<GTFNT><d6y=y<~|1Tix1rnB=(V;XxPvc9-7ZMfbHZ
zTj?f+VF=y(FVC~mAht_BoSD<eEeyPMk)D$PZzYG<v)W<s1Pfi9i&Jxbl9<fq6Y!az
z_5M4>4Zml2haQIS$%7;IGCgSmeDU4K7*FHSopw@X;zrr@g4BM5hipl0MwM_WV$u)g
zQPjLYZViwRRO&IvyF2NHlwrT%8+;8PsO0blU+_oX?FCq{<O7z{P}spA-g`dD-W~cZ
z&#v>WmPjR&Osznf-kLjsx68p*-~fkLCvnQ~tKeG+m8i;fzSA<g(>!-2r6BB(XUSpG
z!#i_^UdJJfc*=Vpp1gnWVS4jSm`=SEK9PsrZqaN`h<o3);fQ?t>8C9VbdSuZR}+3g
zIH8U<IQipbID91T%Y6Uk^JkymdX_yv9PKrWBcL2~-b|8*y$#j<z=OFj0n-ukHb2L1
zvOoOck8b_hkN>nEG=u|LaP}^;T1c-6dACO>AHY)$upQ=uc&=x}$3ee*_GRpUY=Q{#
z;<)Av7MOK-^3F4kZcginjM==v%%eQ?-0)Z^jx*T@OU>$J=81gqBcwDQnAOjuJx*^J
z_yuNZAPtPnrj|-7^n<Z7D=v+khyOxYQcp^yXNO$3(u9O(C=17Xo!fTH--u7Jhj_iF
zIsM-ddzQmK_6zrajPTz*A9Ui2V}%0`a*tbw_u}5**N)%wg3)~+@}OxJ|CQ00>Je~&
zPsNe*%Cno8iK8?~BOiq>=TU~=Mm4<iu0{<OZ)rH(Eqtp7xu#SOg~m`pvrIbUo{W$}
zF=(agNa5BkNa+Oyp&OgQQK3LfP*yGb8{HdRi-`y>JSrYI0e;`!z;<}-`TlDqr$FT3
zfK9O)*a6bKL7DhDE4YsQZCuJ4Fk^UI*xO>AmLA4#C&W9nZ{AoMh+kT84216Tw%oLS
zQt6rB(0Lw}(KvI`N`r<CSH<QY{K8uW92L4h5AogOPkEG^oW1MeHb)tW7Bq%%x)D<Q
z4#h%5QtEq?AN&E!#AowZY_B~-7Bh!pUeEik_y~UhA9EKva;<S=*oo6xLk7(x<Y+4k
z^2G1%|B&m{^+a4e7XpP@<BeZAl5+GX@k@s1NqBVIMBh{<8LA-z8{>^p9H*R)8eY+I
z4J#8ruyO>AB6XeCy~;tw6v#R9xG&N2HgB@3;~AO1$@~PHbgA(50=4BzHu+Po_K@hz
zL5Sm{>t4QrQAZJPOV{KlS-2zPz=l>HEFAPe3|<5m{Yu$V#@ZVYMm*0D1!vg3Ns!k$
zw&Qk6cTNjD`6qqiHIL>KM<dVR(Rm|zb%F+Y<e5AT=WzB7%PP-$F7-@k$+z4Eq`e_}
zbhv{)C+Yvs-o16{bsTA8Mxm-uIFl45wIz++)6;YCg>URD-+1}g+ke5md?}5mmqt8-
z00Gnq^FD86e)Rz)C9+#uz0~8Z+TY$g50Q}(k&%(djn64B7`>ynubJ~-rD?Q0kMoAd
zq$ufv*J;XK{TDw|L*+HP2bSqJvSop<P7ossWa~~IT;-cobWqCD)0RVXGO%@&Jf@y7
z9B1V0GE+d^`kg6A+rKTA)rotXb8zrNM{LWW>4<p6DSU2uCgU~ybjV^DnKrb|{CPSe
zPjZs-?kJUZI$s2_3DIYGfIXGfg{x7#;u##t-sG2ip2s*CzdJsF71{-#F+lPNHwrCu
zADtgC8)=Vm%9j{NZs6YedKY8#REtf4HyUH4A-VP82+ogtJY)T1ICvvHDFDF@_NP&v
z2v4vHd3#8eu!%dUCRrawfF@fWK`BBY4S4V6(Go-0!(CA)?S@c{l5Fk-=-`rWqrq>$
zZ8&{9ME&PJVCsVmmsS|y3T=Oepz}-^xGEI5CVqu#^>WPddvC*%c<x}$l*z=Erajfb
zvI|0E1zz)vf{T$pFyjZ-9+Kbu*bNgrJbx7!8e>}<3nS0%WhK1nqdXmU^L>;W1)Ywf
z!l|I?b(t%Em-95D1=6DoEds}yiM#Z(Vx1JBjk#UDviq5{cUA?hsuIG5Q%^Jqddy>e
z*E?}0e1<Hu4x(}>t$Ad<P&*1JVLgb@D)^w4bsrRfUbKALQ&%@i=TTZlM!3`a!G}m0
zRvxJ@PkJ^+dd_;46!M!WCB2RZfe*(X{+3?0UhmAc21wBL0lv5Q9djN<_wWe<w^xO~
zK_L1f4Pl<Iqkyer=n!-G<69$k)o4sy#Xn_m5~m*7CpjdGOgK=>sD}o<Um{I|gNM!=
z;4ag1g3nuWm3Pm0k6wA~I*?I-WHfn3dEaDhLK|GlENMoTQ{N4}FBt>dr^iG6;$tY;
zQP>yB+X6c{Su^5DYv)01#jUR6QG0md5gwMW!3n{LfUTXc9i*3+_LAWdMVVV21v+Px
zJh=DwfA{w_<b3RmT)oIHh1oq3!zl3SrHdRZPZrKX6SAQUMzDUJFWHibM~}?IXSc@F
z$2}UJOsj_%-IhJ0N@T|Ao^5E7{sQrWv-&Te`Nl``%CNbzef|YJUUsJoSvG{P?Bsxv
z{+r+YrscC+qx=aiFGAmp=d`zL1PE+bWgz)k7XtVE$EEH?4$<fV-jlW9SSIKDC`NAn
z@P|JX-_O$PZ-nH3k@IdW^xU-R`4s7y)1Mk@azmeMbm41+FBy0A;4~i_+(wf!kkwbw
zAw9z+c%mU$5r+(}iWK}AB^lxDUwl_N;LYeV8YjMw6t)0)_SEz#7(Bc)Y|xwz5he|T
zB)R2u(%<)-k>BU&{m*jd%HRF%zwKhrr|WCV+F51%_|w#z=-hw*_kUm77)`K_qibX6
zhVUbRd(j2w8U1spoC64d|A*h-``vGU*OOkq{PK&+Ele1~DMeAo2ab_<X{5RM`-axT
z?0WfC>&Acl5C6W4p?~qqU$&0xs9>G)?Jsh_4NQ9GGDp4R@x$z5D_&u~(c!PN1Bi@#
zn>>W~Je(AAD5u^G4{j}xjxI;3U|Jq))18mv9Gy-_(nTrr==O<;#5Wx!7qt9>lRBsl
zOnf?e49>79c)*{G7}W{9$blVPjzZv!wl^tkA}>X$A$LE@BJbhx_#0=OCye<4S0q0S
z-|xEsN!yHbZ$|Dv0PooU`1uE-|I2&tvKHC*F+P+vbYzUtA1GO6f;Q{o|6a;6<2|1#
zr@U4gyb+C{6TW>Y(D>h_Rf##c$lF~S3OT7K<FOP)iz9T2o5%+!5`K>p?vi4JRYa*7
zs!%0}$t8ISuVYNalWg*W1!2x>uK5*Szl3)yV#49f)!*VLsIN+3p_O1|QyF|Oq-R!4
z5-2sVji+}`rGi71Nb=TTaGNhG`~mO6P97_qu}7IRbj0Dm`3(lZh0~Yy6BsQAW$HQJ
z@D0M{$27{pUtSGUo+*U<@{&d?iaxMa_*IO_EAheOw|C&Yg_rc}8kiyu%mun>z$BL)
zhPs2C$nom$`bj*wln=q%dLD+@7z%?Ie#2M8)M5W9+V=i2Ems?Ec_rSsVwT5!=euYR
zugKBJ5<1P)PB?t;-Q&<F=qN{(b`&i7pr7y3q(H-i!05JcbPrE8xEk9z&f93!&=w%{
zD$S;t>wWR7aECvPr@S@~w9^|MSxMbm#^Z#Kzp&kiTTdd7j;&nBNXbdU2`*pJtRraY
zsEopwh3TYetf57rPik~#Kx+&bCGb$^Vr14R8C=k#?~9Ibo!)bj#%&|yaeDZ0XVC24
zct>&YGkAgRYrYJvl&XBst%edDe3@EzP`t`R^4Tub>Z9W*_-4?e>j%+M*@vEaUTOFp
z+$0?cOUUb$Na$)YIK%X1rb}LA`|39_fB{8+v>d}bMs_+X<aX*$8s*T`w^88ktJ3c}
zJ&X_P_LSGOn0J{^r)YEyJpHxaM462!CsO{4y71tNpZLvC?>BXxKku_CnHmAX^CtRB
zeJRW&Hk#AZw}TnpRh|#TkZVy+xkA;-VeZK+)H}oq8p$6r@lPIt)bEkeLo~6Ps${=Q
zepe1r<DZoChZbP<E|>S~Fc5cmtTgp&vdRBApM?D}ch({N#Av0=y%i}1>3axZ=(NHw
z>;p)<usiJs!F`ox$NRqUzjnVRE}MOa;<U^aj8GtVyJWboCeFRR`~;oqafQjIxY{E?
z^E|6qdPo#T=NkiKpGC0gEi8=c{4l>R5S2+_Qejac>dPH3S<;cj9(O)+Hfad5h{0zo
zk^~>Ql#(%>t0OpxbsIQx7<r-O=2@caG~9{Ua(f>`f|4|}Pmu~@NXXRw`stO~Wg+lQ
zEO(^S2ZE1<kC)k%!xJn!a;iX#%0x5h>FbI0ER2+S&Is9@nO?xgPopHH>o{ccfMF6P
zl3oMe*abk|^(LI6+A}AT#txHdD5>C-JZYb0s~sg5DDlat#GCYnla@kJ4!!Hk(BW0$
z8bRqOOYnvB19{j@lY4h+B>t`^_}E4wFinYw)8Xd8**%~cCCZ4x8VRZF1TLANNKJno
z>TuFI1I4_!(G{Po50`Qc=E`rx-y6Jlc}l@Fc1F~$F0(@@Jy|qB!!}mRk>POPG}2Rh
z_;98PDwm;1^+s^lK+LpD(i4lw4%*6*sH~hryPR12+JeH&g=aJfk3avyC)8)ST~E@b
zUKnLP%_xTvq&Dc}jZ8*9y1wJ~&Al&jIF_*aria7IvMa${eDcMwt4p%O=!ujme+rH_
z5z&NYm(+PixRRc6Wrq&h_tWetk8>l8{&c%(a4CSmtnm<?U4G;}gm3Y-G~R@sk2`ww
zQRPk})dj#Q6B+X!u?H#NDluXe`n3VP%_xpuu?B3;ZJF~Z5;ce$ZFv*i@!r}i`Mcgm
z#$KjZtsdd)KmM10tbXY<OJF%X%V^Q#+|laeG@xYd+4|D3hSQEL1RI_7sCT@vU3t^~
z*x<2~C`zcb$wOsEhs4?j&gdu^0HZ{QS;D(LIrKS68ESb#IJ8JmMg+gU_uJq7F5l$&
zyeF9Q#CyiBdSl(3DYsw$n}5@(kbv)qkTpr-7#&eB_cyV|IGIj{P7wUnC44tp*s^BF
zm5!V`O@{40@_dpJBiqO?p3gDu&qFh7;{M&={ku-T{EvV8ZwtdH6jz>PTB{?pku`E2
zu=h^t7asE?Gr^Dik`<@0G75|~R>LPkG?m%l1I#`IpM!xMk4}MRRE7aQqoSziS_zxz
z?x!IHKmYH=%F7sJ!AZN(Br6_fa+7!rAX8^~Zzrb}X~mhCfAsf^qAIKcoy=soL4+pr
zyy6vFo`28@`y;_WnC@S3KpU4uv*8aqqZ6USIFEVm`-~`H=2+mv{>EK2YU7e;WG<<}
z&eg_m$4Q91?wprVg-ho8-f%+oVchzOp9V}MN<Iniq7A%BIPv@zp9HnccafWou!gU<
zLpeeO&y=PJox?0FvKfz+u=GKr!daeu190A_A)kz={O>;tpUbd!@J%vv-4qoVRg8Jk
z@OW+moiL5=zQogWGL_JDfeAPKs!_l4*jVib!M(jtnYJ7<4{49$H;u3rchiVJSKM=N
zX%nvsl9G&Hs8<?Rnf0x)nn&r9^0p@yP39dMT%YH$zr62Q7hG234}WCph)rnI2B>i4
zyl~5V09Kj8qsCB_2qms8b4+|1piG6xcSbXam+7N1CK(zQAWDB~xVlPOdga6a&|3a!
z%w`Tfu)-gU=j36eL>ll0FYq;1^IfCjk-VMcDc)s9D$dhzjsdBG0DovC9UhxzLA#C$
z1-^zBo{1W4QVzyL%II!11W!vv{Ec=dOWWv1Hw}N|N{8C{p9Wo7;uW~pS(q6>efGh!
z)6;yLC|(R^c#D(&amnvAhyC>v@XFe{H1c5qX&aQ%hQSnrISs4_WxKj@ytk1U>t<%!
z25-DG!p_t9b{$5H6EeWyoC6Uv+7o=lIm^OdX-1Y7Rv5-D-fvvta?)WE@*sS8THZ>N
zE1(8m2txzm9T_PvQa3fQC#IX<48Is5jO=Hb&P>vtKVl*2LkpvuK2VU-x)peaMy=0<
zn|hqIru-O!43<vq!X;#iG>3l48|1|ACmRBT?8(#g7o*x?Y0moq!xWYg)FewPou$Ca
zC`_a*NEFN}U&>bcjC?mN>AJ8UKEXu?%SR+D!{JW>&<}x=-<DUI>++D&bi`*mH+&DF
z6DR59!yShlKOZ9OxI1vauk--49tYNV;n%nwWxf93?M#p2V7(xksJTxi3gDEqCvt{q
zXCX*k!b^d9JN$~vA^dtQEJ!uv;Mr(@H9xAzD!~!XDV2gPw=3ZBy);zjTzYeqs@`N3
zq@mPQUg!9fZV@U<Q{^(eO#w`<c#^-Qr6j>>1C@OFZJI-)&-}bVkB0DwU-{Q?3hP#{
zD7yRU0W_|_q~qXLvF;G%3~R&XJoBe#>D5uhq5{b&W6N2OD7%bG2-0h$G@5lh+|wCP
z4M_zP7LUS`TktY6qLM*WPi(>mU*)32DZe170EwHbma8;KsyVPI>oan|5oOjXl=$zX
z#BVZh4Ka;qQI%63)M<EHV}jE6tr>~Zvf3XX93<36H@@5T928}!p_2HE<`<8rq0ta1
z%j4`)l6EL7pUK>vf}Aqx70z7lhR=}cmpj1?&y`Uc*G#dcSE*MPnR%VHH{=4nyjzfm
zgL+$BNSkt{JZ-SkaOUtMjX;$+<(Iz{vrdmJ2?J~RfKYJ%HhK8;_+=L9;eo@y-ensr
z{3%Fh8pxZ2t7oGjUF(p>uIFq7p9Ezzf`UEI;=PZP9u22~(W}*{zs&bm>{R$VXET6h
zYUbT#bWj@o(mnC*G@<N`L^)@K9H5PcgY4NEG3PL(_nF@2x39>IMSm~L;{fdl*ROv4
z>()sN5DoR)j%lrgZNciR2Ea};iwxmTu3moqMMt$9-K`-%Pa|i?X8DX*aSYgpSs>YZ
z7iFkkYpjd{TA#2_Ohs<uTw^BdbguGk6b86aJNVXUNqXGi;d}|}Eq<2sCcImBMHkfA
zOCGPk%}38dA5$qtNgM=*9^9>U8{zGqhCySpGzZe1-uo={Ykm0bZ%a#!YS>@?Gcwx9
zhVDQA#m`F<4KDgSgzH%rUOvrbAvU@(&cPyQ0JM&#Tt@h~ED37C-(^SDL1q)yZ{Z|}
zUN9Qb5w~R={PLH7bMHU>$NxDeiGFqOt6bJj;nnZI{IB5$zRh|b@L%Q3kgJR$<wG*}
zo_zGQ@@#|y6X)AkLlyKToABfRDCcuLPkG3WQLZ-`*;I|r<F9GeS*e_-Y~<}pMjYs>
zPk;9F$n&#|hJ4(eF<`MZe6_i|xDw}N)%ULJ2~p_0GC6SG#i~(P_K2<#g-^1m^;z18
z$7|cN(_+zuo@%`aV(PfvV_k!nxce9cG*EL-0+U|dLLeQd&i^U5SR`^CS#W*mQuEsJ
zx*@ZH-*|Fwx<8!{cxMFIH!VWd<{}^0U2=@dd*V0=kZ|q7Y@F`m!b1^vh%El~yCqI7
z=qQXSGrDYA?P;%R6$}W!?+Ew4q^m**tttP#xX;&KOKZFuF2BbQ6{GnHTj3vOB;Phz
z0xDJFZd^PJW`e+jN3T?XBs@fI7ELyJ4c|3DvGtWD3P|JF-fiAHtT|De1w-}dVdXm1
z+XjeA*Pc>vU??<F@k!&waA|k}MTtyf)`rc{Al!rP>Nh0q6mkrnr14rW>8H{R$yP-I
z+f}%e2aLBvlWk;e9)l;Qyn7`M8f%czL0voMk9Ue$8vfqJSAP8C*L?CW&ao30O?SEj
zjQB|#>L<GL-|-N>y%sz%N`sPp@*rX5w;U4>c0Sfdd3N|e^q>#KVT2g^rwHTI|IK^2
z4ez7mDa;x?@*A3V>LCi%wAVDe({NGhl&eM?tm*NG7A@<@2V+EfILv~`DahGrhQ~vD
zMnDhGiU9%!<y@EN*>28&ieVjtio9A^Q{>4*iya)%D){xvb7^ggAtN(yGG%0F9!<`Z
zr+R_kA$#D-i9xt{7k_j1$v79nkxipWzTAsXC@gJgUBZS|!AIE=8L;AD9nbw95@-Y{
zBNho)zbYe%Q~5$SQzRO^^DNvog$b!#rUI_6W5nipXqmi|9=#dCo-0#<A#eD|U*CPE
zjBIzdLOzXrKJMr92Br$We)4q}CtC|)ai&p$hYzwaHayzKI-Za&qby(_U+7i?E@k)4
zCeG+s-YPTGFpq)xJX0Q~LBQ#>F0v**`Jy93#eN3Bfn%3sHkTvZPdZ8?Vt7G>hODJs
zJ6+fkJdA|Qs74OKixFm_GB$LCC~}Y1M)b%IyiN0!-Z9efv_?&EF&#)I2DXwU>;8sM
zk+(5Acb%BfV_k^~=zAL(F#;!T4`C!{tH*$cIuf=JTfpj|KKyx0?<UNpn5(FH;>f~0
zz5je1Yw(%qEAk;2E_Y>V9Q@3~M49J<$04uA{K212$0&5Sh!U0nZH0=Tf^e-+9x{K<
zsQ{ijg5tj7VNzZ1KN5pU@z?viXi;hP92&L}%`;kd{yOyND)eyYBn#9k01+t6!aQ`N
zM5gCBJ0VWeqv()aD1ARYq?;(Oj2e%CB`(-^ZfFfgfTi*Vj|gMm6KRRY@Kz&|q1~<v
zY1|~CB=aoh{D#MRm#OS~(cxCcBB)#&b|O}hMhgf?8Jwo6$E8OO-Y83rg`Oi>vdEN-
zo%H(~ly$Qn1tASwa)pu&t1%#-=|WA~RsJy)q#p%AF_bn664#;mz}ocE>+r11q&q6G
z`PE2Lc~{S<MowrIAy2HKHDv%dVKZ7F2kSOw_&@q(=onnlg;R&B9HJmJ1j-ey*1Ds|
zg}xes>-g_-Q+gX;JSeQq@4yp}Q42O@l!rNhD*kO`!iQ+zpsk!yp1Es$!_egGntHw$
z1RMP+O>#<U%P0Tfo>*-+tz9vCW-F*XpX3mxJ=)V9D=)0QFf^@UO=A{C%<$?7li^v7
ztI(ChFUE`caz*+)-+4z8>opjB=fUq?^0)1B8d(Z;Xfrq^PPV%`2uknwn-}?He0Im6
zq0tergLp1eC&|0AI&`<o>pcDs(zET!vymCR^NpOBIlJN8jE0o9_<U<bOwF0Uj_hP)
z;YH$olOyAu<O;vDGdod2)BH(meMfurp_e*?zvzK(8Xs$89Cme)vt^8mTxZeb+tV0v
zp#cg7rP4^=PSQlAu49P*18Z`^yDcBK)8d=_(b&j=5hYi3NqE4)i?LUOk6Z?pQ%CWN
zu_sTb7OT^5yX%b{4Zi|-_?fUb-=1Wx&ez4MydYQMnGq;?<Ec?2r<m%&fBN%Zv{7fE
z2)9l_y?3}6!^4TU-m{+vO0BDKm^CG#M~ujK;g8)+==h85>@ni-ZALoK^7QoUoNW48
zM-828qF}aXJINL=O_x|(RGP#<QuJ}jt#UD)B<UDT@LX!Sc;TN>mMaH_1-?;H{2;5<
z0g*ZQ)Awj;^aRec_&xHFG|Fv64bSiiU+6~Xgczmk0ca`9WuBG0@)LeN2_3;cKmD}u
zyEZBBXBi=aqq<4Asgqx3)X?aVQKH+g!o%#ua>kDW?>1VwQ{=e*csRJAPu%j4pNWzD
z^Pl*k>GE%m2`qgU2)D352_l18C)0DAB;@ry`842re%A40lW`X|7Nh1Z{{`xOkPC-I
z5@R`;9|8^!g|SL@7rfK_E4fpwLkmotXOwXbL(#)qbz=;b2yGa*{Th%TaVZn`N`W+m
zH@6+HaQ92^1N%LAAAqlb1s*}UOL81(!$@cztH(X2;xI@C4^YU0hH(0D@N}wc8}C%Q
zR-{xk736+*1n8}%vZvx%y)X4mPOBWcP#k)3W1MIdWzLZ6&@DO?ExiyVS&A{SDv^rP
z@PQ}NG_Xk@ocvLeV$Y>sCM-1Fe`8QY5ih<0=#c`_<D~rX*9gL1145Y`Unj8+(WJ;E
zaj2CBByi9zG$~KxS0@}zb1x0zx%uw2P-Q5N!SfE+Nh;nR{eUke*YZyAj#O<!f0T(w
zVMvu^i)LuD-QYen=9jQUoTwv{*HN<94x0<^{ScqXp^e_WYkx0g(-nSbrtGzGtwY(_
z1+fH`E0iu-@|pJ27^h%$<nJI8;bLISUjQT>I72@Tdxp%j7Q!f@h93RKF=8i=`Xo^}
zFXuEkcoQ-)3X;Et&K4!wLfxsI@LrnAL>_9$WOs%g8V<r@Xrn#6?sAv{4X;Cw&`X%}
zHuw<_w@EQ&3dRE?x?t!va!B^d-&I#@C@19@+*6+#9)M`Ul_tqc<8HUc_QS`6%d>`=
zgNFw3oDm{*{?$_kl*UJ+2VsU57<7O#*We~x0kG3)_)klhF+3IaRSb={sqYMcIVm)Y
zUt@HT{Z3QnZmLV;+|d|lqK_+&WT$0pE_{Y&@aQyI3?2*CKkfqB$YC07AfB3{%ku+U
zsZHO8hNW}l36#>K<qGfP*D_abrekDujFI#xQ;v)(`r-A-7-%hV@UqzT&KL}@LUWwx
z=wD!6=cA91>F8gKKL%6RF2!h}>zZf$bgClmX$w-i{A`(`O_UQaXYJJRRfhy`YurXR
zorPEEWnn)9)8`D%&VBrko}Ho=5ie8v^;_SgIKp@Qd7k)%Qo2@*Ql^61QT83-6Jf+v
zP5o(3mn)Dbp~}+Fji>M&A>a^=n@ZQm{!HkON8xCg3eN9Dgxr2ZvA=RRW!@zp#(6tj
zDIx_MDH>Wt*o2~CM@aZ5&Z*(mJi9hsQAe2GEM+_uF`oZKH<Ta#^m(D(H6ekCy}goW
zZ6bdAiXz;@j)XpM2^)n<9J<>faVh^%h|>r-U&C;uUhvgfPW;TS3T1bx^lBZYZrVg<
z$}4^qb>gVdr;qhGQmqpx&$G&5Xp-InggZr&yrn-ivXKf&k3@amdyw>u1tUl3dLAW0
z4Z&C0j4r{Kze=qa2WCe$R(|+*o<)lfA3yBb5;7|Nf-6)gH_Ftg%M4Y==t`Mp?8n-J
z^m-`QTpDl6ML}zP1V4%yF7VTED!W6K^jkfP*DG=DX5m)_7y5Kgf5nmBob#qH<XUHQ
z->N{I*=>N9DF7G0$$NO=(7SjhAC02Nfinh9&Ua9TG$_`7=)FHmFW$Ni^Z$m$2W(2z
z{LyvzlD>Pm7Xvdga+zST2fs<e_~;3dDc94S)1WReRw>ap>-+@tmZs^7T;9VVy-h>)
zdcD*0Bp;Ep9BS2Rjl`X_!6k5$Dm0C3WkD?ZlM%*;sfEweoAzZ^=Z|Q3T*^fL(UB~F
zlX#|`z6}lRQmH&81NEYHgHe;X3>#^$Px2A7<WW;7xO6lpG6FYW6xI_~$MIjW47Z=^
z<tEL8p0lBj4WFPdy)uaFR8#UJExL5fX3?l}7rxk0cN5v!hOCDEA~L3IWb0|lN^VTu
zaMr{asgoYu^WZ}#Xf(HLL-$S2#vl*zbU`RZseAb<ln8JB@%R52yv7)UA3X-{&+-X-
z_3f!sPLuB^pUjc#&+~2+%|T<PH>~-=BgV%4_&b~E=8fr~S(C&-_~kEu(GL<nHX;_h
z!JVnt*$wqK33sMTjjYhFC)lPuHD2?6nKCG^5ia<t|Jxw)-jRzC-Z>+mfA&Qidip|o
z-<@&K&lEg7XW)jMneqxfqR%RPA<(xOm0;`{J%JCZse9I_v|cktg3qiMq9fJQ=n(Qi
zUV|4J_Br_U8$+R`ku<XTFkgs<mxaPQY3hS=g}uc|+P2DNSUJ}Gi+Jk=saoY&w<%Ov
z+YnP=6D9XH6yS_s^=96Tt_$-WE@k{FZj2BqNO!IUzW1L3O3J(V1FJAG>#y1=%OcMY
zmE*XGp5H5T605YKZP-*QHM%&tJSRroPC9;xuTe;%Hd0Xxk=7ytPH8*I3VXc513WCi
zfhvcCkTsID>pDEhak)GeHJGWK@$Qv|MV+Gy%j+T`&_j$96)E^pP-4_jIMAUf;{RP&
zREo1WlUIqd=p%EDjxrg#EXd%}GD-*Eq|?B*>}ez|k}^E%ecUK}@nYUIzH}WvY_8x>
zdbrC(<=WQ*G6|D<Mn)KBtH&yDgyE!3<*xFN5;<NRTq7jm?r_1^7ASr-Mxw<)na01E
z1jpnj-omd_6o~^qzbg5}k!~--&EvqGWDNrir+``>hZkZM5Jv%?$BEv&Q@#!*N1;~f
z$NRR8+$<#4$S4<OH0Rk&=E`_Kqadf5N~khyK8(VV6NfaB9os7*5i=s$g`UlR?jJ~S
zX=OgV2fn2dqTuy3_ZozecY+fj&96VD6~K%DlHEpy#e~L*GL7Zz4mpbvO0k2rr<NGl
z3~8f&Z5Wca20A`B3-YP9z<aRfUA;XhG@bIHC6V!r_j=1tr&Vq}<F1RSQ5ab1uq4{S
z$$NoYxX=5Jawp4&Bz+hZ!-b^B7x8w-sJzR2jo#qHC{U=|>^PEn<EET37hBI6km`D;
zn);mF=Ra{nhj6<eGK6cc`{_Wuc$waN_^j+2c~f}|d1rnwlB~mE6g_>CbtP%&t&1?7
z!O!N1EA&tO?#`Uxgw~7?M(Cq_{B1f-k)_1PUwC&xI;mQttZJUJSnv#w8X+l{>ZDE?
z!Bp>cAe{YyFBaS`L7K88UiDk~tk_fkbT)J*W^s7<7(-*~OhzV>=fFUZG44jMhqg&4
zym8zU;qbm$h_<=ZmAS|;8t4?w8YzYwx}hl=G{H{q4u$2w+A_p@Kkx93Yu}T}l=r)_
z?&9yKKm#1mAL;V(#5;i;LM8iVI>G}^6<+Sg;rzvW&&fQwABZWA3%rE=1ian@`!D*k
z=vmxbzW1?7$>wj-46R(7H0|9c31QZFZNX4Uq(l|;YDG;Vicp-H4<DNE2Gt3p8oLT@
zU`jVU8KI@jr1clj?E%We|FK&!L`{4ls?kI$uY$w$4svJx=h{0EI!~MUQJa(2VlO~c
zq?A!lRwH_G!M+_@z|oXVKQC>&A{u3(UiY6vVqZ{+LxQ<bDRfG&fFbA0szI!$n8Kjt
zH3pcg6*JoYBs)0ZPFeBuj2C*9o;5C%j9zH}lt1}X|KNce%C$kAQ(0%Ess;n(;`C0G
z0Y3{njYp3-k3TwY#VJldaga2qOG!}ndZY4kc8L1tqPVObp**TUWE8y65{h=ij6ySg
zG4fKGNjS#<4A8~Jc%UidQXY-M$PJz;>HHP{icI_x-No&2&maiUYc|}Armrs2Yu4Ki
zt{N*lBr5Y7^u+aNzT4EzeLX&O<*yEeM~0rQhtYul&Ch>XJTEfp)HM>JgPzx<!HZDT
z+M$-S*pv=NF`k5uzsODzr%=k@qC&iFJ!@1yQ#TZYQ;u)4uETDRpJnZWx~1ps44a^R
z?^(|9aMZSwU`-d;DPz%}bt65gH9T@myJ35L(p%R{okm#?I&z(|&exoi|DasFG?LC}
z(YavIBwFL@IloQ4iNsbGZf|C|-*kiI_p=5@-0gZ$U3kl@^0Q>1jbLc4jOa%ujBKM#
z=^r4kS1%p?#UZO2-cCEZ=UrU8U!>b<!PMCQ=FzWeh~Y0IVowFwsX{(Y&2$@c<OI_F
z3ym2AUwmQh%e#AjAeXtSW17bpu#Xs%q3e@pPrE+DG>&y7>dlkz>v0SX^k#Uxi(#uy
zo~OZmkaJUf$WE9tsdu)9?{yZ+!+2C*)IYp3Wirz-IfsdyMs^uDVPK5_=Q?%mN2=G&
z&GQ;m9kC7k)+@zZ3;~zxs8RB|&6*+f>gZDF<XKtNZ+VTp?9?Lz?Z721`T%V7(MhDI
z-O8L=tPTsx$N97+%78aUWf@C$C^-Y?irz~+`fb*zq}--@f*<(;W4m^4^0M{V_TTri
zDLRhsZ<%>)96<zYmA%h-9kvTN%4h@pfqUe638-WX!+qa9GD13pUx0VU>y8D7EDuQj
zK#u=fy*G_k`82?2(yxdnjDT7UI^rsT5@C_&chMcad&F_>oSp^&-f0>w>8#VJFeHo{
z5o1Us$41r^j)L+aeCM&Bdv9705((Tq&G6swv-tB*S?bDUvEuqHS;FBRMI6mgI4W9u
zkqIMT*H+Z#+1E+kZBZO$po}pN0&IF_OQ}-UN$09^2y7MdN?yrYp0j#C&+A=y8{16J
zDC_ht5)Muy!7cC$Gq~mzHwsH}f(J$5SZAjyUO&)NGCUe(z_`l&ZKeoDkrd<1$&R6A
z__JH@LX!@sFGbEz149`xD#6#QYXgyY!`sC*_<9z<iY1DWLf=jtc<_UVcQCf{!wP=+
zDO(n6aJ}?x5IIYk(OW#`9pzIgjY4lqbV4L>a*12gB_6;OAvr$Y4yi|LVJm#3DV*HF
zL3^&a<$Zd11H*zkGGHV`V`K_u8n!fMX^<Hc7)P!cI?acS1r?27*D55N%Etq17$PGt
zN69*dTjPj6Rrr~(@cd9C72NPpxxUEp-`0Wjz_(5B1dL1!jqwG%*TI+Jqb#E{)oI3U
zjE5}BOF(5#efDg}j}Z{ELoOZPj{ly00^W!~>6N&hin*WZn{Yz`<AvCPQ)C)9KKTaB
zOye>B;Lo9P>r!Bcy@g3e_9cuf4?+V|XsrYA&Gl$a&&Zh}jLzy#>rLwpd>Emir_5cv
zJhG=8=+8*FeZugW%i0<7O=I!9jDd$=*Xs!KmMze#7L>YESM)@Vh!2n7UcT<hsEj;D
z9h`jLzPoNx9(fui!Sf*6@?ZXq!v*u1h?I}9*WEd}8yUa@@|NP)$f89&hbp<#52L@(
zg}j`v(Vy$;Dcj}K;OE1GA;gp`b)a${9{H>62cG)t@AkE_XXIynase;8=IHg~8m`Kv
zV+RF!lDFw}23nr^>&!Wh{aPK>xjMmf^^qLqsQJKztFLJ!uyCKmkiWiq7K6{OAk$ab
zhTRdz#Hj;$>Ihyc|J)mR^c?>L!;6j#gZkWhk^CyRO#&Zq?i@X;jwR`jb~Qq#LKAvO
z4N)<=RRXXG!FI=JAM*JiM{)jpb7)a|phd1t8$A_{pq9Q%=z8342)u<)oGp;<9A*;W
z!T@u@0>^Vd!!2U@mir2qqsR6L8}3K6aApb7q`Bi%1x`W@P7+6l7B2thTcDt9Cmx(B
zkzB#~HZmhF#i2)&$G#t6Bq_g(D3mJBWO8hTfIMj^nt|!Q73t($pv`-~qxQzP%HcqL
z(dMsfS0d3Jg$b_K-4O7W4OucmKw<V96KN=}uYK1=4{nB6H8y(1Z*plA?Lay|GoKtq
z5ryMa(H<qA-3}BGS{v!02)gcpVvpY_^ntsL5VEli4E!b&HOf*zl7JP*;mR#dard&|
zRyiC$3rp%T$ff~o95e}`I~o=G8mi`xQYhv*o*Bi7uq57z=>UJB3x$3ex>yt^U%at?
zrV6;_<_*@+e$oY=Ld1<BlezL5A#-S#!0xI_`dRqN2nmePR-@9!i1C*EDLHg|bMvO*
zrhzWPw;CJo(z_3SV58$jWDtL>6VlN5_?(8;&WYD}8yd<FpYVB%*g<s48eXP%X&Yoe
zW|nvg*0mm#ZN|^+obaJFWyUjGLI*<llk%$Pc-6&%2^TMIc4f%3>og4<ecPQ&rP5Fj
z_wi&#P~sO@_+ivyyYeyR3#a<kbLBq1#1lt8?mfptBPdPq+KWZycpCcSr)fQPnxAyy
zRW6k!XJ@FBU|FAV|Nb&+0xwaMAfr5~BN{(+L+{gkC+I3Y_{%H|Qs2R>^PC53hK!J1
zMh-6nw+3WYAPmNx^_(K>SER_-a&t<l5fLA<Qy=lmXaIj5Amt-Bp4H=Tv!*AY`%NGQ
zfHf(~KyH*nxn20M$!iS<+P}%kx}nI(NQ{7!8X;2$l>mJdJ}W``Y>TjZ|K*oowqB^a
zWXv_wC8iI9qdeV%M@#DN3apNPlQ4_kgCpF0UE78R<h#weGQ~IX)e$rRyT)IP5S^AU
zzWBWK1QXq)bNO*Y$2%wQd21Y@a0wliy*TE*qbZ_+)i_@6T&NN7L$CM1-$nV(Q`Y<?
z=aqYfBLKR*KbXMj&<mDy#rqDF?V)h1fQRC3_<t(xhVfT)ANt(UFDyIIy$n>AP1`(b
z___!XjpDE2JfX}7P#K+CkR%Cnt@K7d1`I#GL+5!4+z7b{6cJaR5|L?{TfzAgz5Ull
zE_jS6L1C2C<PgXOE___^p?ub#O3qIuCL<~@WvpTtrJ%H$CUCX_OOlD(O4XM(bhng7
zaGCfyjH=t|Mk#>TkFJGI(>seI)$>}))=(sar;T(PBZrphMR(Ul%G42=;H(h<->@hp
zYP&Fl#7cKG=?DEvk6q%SmxgaEVWR-%eA^g`yXd!RIy4A9(8jeuwN51NHrDj)I5gas
z_Z_bzGAUm&uA&O!#i0hPh;a-H{^SUL-pl`@6;+V@x;7&`X-O8Y(!vW3!mR}5Ep!u%
zJ`I%wRhlMoopI6k#PcLS@Ws!+XN^vcDh<^NN2l^7VKWL4m@yL6gGA{^{l5IdNZ%Mj
z8mPVa_>o15p_8qO`0bQXJY^)BN}$laJGAQdHfsqI7rGQ6B#kaSeP*2ixtvA`Obwi)
z%;EY)K2in-SRDQ+%jP%S8CTXLl=q=<58?{n7@_2>yA|>-UB7Cb27%MKpyq~G{v?if
z4&0IVz)5;|Zde`ZOq?1<x#K&hK?n0n1DZShzyPx{v);kmhKO|Dx0?}xO#h`mz{K>6
zbs*OHFsAV!|D{ED#s$7S<drI(^$}B6G@6lGIBe+i00z%lo50Y_!qU_o@!(~|Yj{0%
zsqz9F$?~T@-FAT^xZ`heDvP}Q2rfoC-sT*XcegnqHRUr!rV*Yg6bsI0wCSTyK59J=
zls423+M^RpwQU3U_SO`Sa;lGF(rYoYI$Aa5hso4mVKG)x{tSec_AT`t5$cC0E+ZgD
zLdlz}r>bVehD_rN9LoFfpmKm$xyVbZ%Dr#0uvJ}soR2WlwXL^7k+I<&p3*b87);<k
z%=F^p#};hUu`N!?SW+t9k8o3_M9jmMQ@r`(jt1Hp3?2V?{0XfQvO!C>iAOixW&UG8
z9m)yFBv_2u@P8bXzmix-Wo!Ao=kk1b&ogV6BUDuwxuZ@ML+Ip4Wq-4!a~h>4$dCAu
zzXfxItuRDT2@;N9=nW<ddh?7ju0oFgVO#EEOCw7YWwE^uxxfcp>2h(0@)RF9QJnEl
zKznehyfu`PLWM9CeP@KM1~mEI&r}I$%HH8f`J|V9p5v(plTpUj7<5QgC5z^>8ofJf
zLJWD1#Xmj1IVm+p2qhZj=KKdZnZHGI28sflpJfO>>8%A&k;W+g{r~_#07*naR8LQ5
zxc*`0lqtl^Y$sN+!QT#TtCZ77kI@sFM~Q-)pViQp{3VvNIM87hw*}Wlq{f#<!xTR>
z+VW5yix)SFec-m?OB&NRZ5;DlUW*8?T%0p{z`8Rk>z$to7fl(aUazneQocsBZ$tlU
zl|2z><P1;nBMoWuNZb+OM4osfSHo)!G1&M+$?h^FRbh+~l2HaE3rq$`I=(SRa@uE}
zU*>FumMdj|lj%r3+=po-E}~37`Q&HqtsBL#Q-J~TBK*@J*C5GA1{{pWK-We7?235u
zWO~JV|K#XV7CdSc_AvPupMM#Ar-8Sy^?n*BxbVa0<`KCvrKUG+l*JAf{L?^av=|QM
z^CGPqe&O4|(pV4g+_%%1w$Vv?juF*$8B1mq#E1kK>XLW(aOX*z%!^~iRh(SgC_zVQ
z<R_2Pw<&WTh*6^WchPA4(DgR-nfhZiXoy%jO<L)}>l8xd<@{VdnNg!LC`?C@FL{xx
zk?+zVbt_5ZriTtTIvOQF<FFz!#jr&m@-adZNPXwyK{_KL9J(1<4AE;OyveRA<#`!n
z6#W?G99^gpc8aM()IR?BXI)?NJYRe^!trrvX-$!lAW5b{m+$B(r=!}bLw6|u0G)bM
zn#NB|xS>5f>vHFZ)9wu1ufO`bM&C9>lz}mV=Wr)O&bXl;4tg>C6ZY-v8NrbonYI|a
z@EKxB2bZ%m3uh&V(M4fhUCzhzT7ZW7f#bsS;P<B=l;5BKUf+`hPH3=hKN#<Ol4hEo
zfX68eSL34(GXxiD3BVzKaVSEKdRTO~m8^TvI4H>3`bT%AGJ#X&CI*m0jCEV2IVR9N
zM`cZBD@I>RkA(OC!!*b5VoT3&!{PT6oV=Rf&}OIPmpWGg#iSaY{9>7UU*s(Lq#<zp
zr#ukXa3p=WRwgfHQ~?XE+LH|s!;K%hatZ>Np173HZiI?Ug;5sr8G55YUrUGJ7^NH)
zvp`QTz4#G;#PJz!h5$tdzU^@h&v3JCrxcJfVk^ayn%Lg3g=u=uDzR{Bd_rEy+ki)V
zpJ#2w?YYlaJ3uH6O6G*yNco8~r|irK3Pr^NPlY7l3=Bh$aO}IhQjX@AI_My)HjGIh
zHC(zRG-ngc;z>P^ad_t7q(_Z&n(f5ipFjJ!%FE)>8UV=;>=dxmNP0&-qB?MD;3Nc0
zG%%`y&mZM5B}&+_+2}{fB;e5Np@i{mdM{v)!}}`sG<wchXrrZ&fu#qF53THp-UcUk
zMvppdx&V*m8?NLU+Ut?e?ux{V0DvE45|81<P$RP#<sTgr9K(c5Y(mttA*zJ<CEdGs
z8A(Vx_~loB$l+1XGva`@(g&ES9SsXW;(AV5XLvan-r!dn$@p`!6E6}~zWKEJ#Y3YQ
zIrkwjUcUS)WqVnkx0jcB`q!^<)R1p*kG%H7js^qyB#rNd7&&?TIBO`vm(MdStu!B9
zei9nov{6!aN}S>PO}@<fJjcJ2M?57H8j2UMU-VQ^YeA$(r{XK$Jhc9=`tEprb<nhu
zNXl1Q<^>-uGG*jytj+ItTk)n*XeOPJ4~C#d(|QLZRbvF>d*sZBVF-PZbu{8RXQDEb
zyiP-hZW#}b>Qs;U4jv5ne$1?+NZ0o+GJR5w5Pr225`!W|?YbATVI7CEUgvu<gX`cw
z^*Z_}{5uW))}_RJP3hEKctp-qSJBiO5ImnLD5EC~=@^BNo|gvaIh;*ir#XKE{hdsT
zFN~W|HT5tug1CX+QIq(0%X9ELNxgpb$VkvM($0LL$%1ikM(5R@Ip#bx`TE5dsXONx
zjfjyF`aSw=KGdht$8QFSy8bu@s%J8Wj_CF@9Tp=l7a1iv4c$`Yk&7JI)~VX)x-uqw
zgl{LsSlGGVh2L*G4dr<JOzFLO0B}H$zm<A)8QFc6L&eU!kd^`X>Di|>o){Nshi>XI
zhu04uK8d_A0@t=+0im_2U-8~XSKiOGSCR>+JdhP~Y`u>WuGcx3?)qJ58`}A14?PzG
zHZJ_>$%<m>nD(3OCJtT4^(>x+)U;32W`QA34vxb+I%M!DZ5n#~vj9!v2Aoy7OP)*t
zqOdqoH5zsC=ZjDk4JD4>J}Fh@+ZO^6NTuv=;tpKIbxkwe48TMTFDQ59>gRoVSHbl4
z;4!hAZl~RM5s#iVk{i&Z`U$^5P@&`2E%*HK4)U{T*R!&OPOdypCCXjPAHQ0yO%%EW
zH`@fESS;f|XJJ4)LJ^u#NE$c2!cks}vG7O+yD@?nLG$jCf9Wki2EEBhxSLYowIS;X
zZ=|Ciz8d;%se}7AjhsY`rr|Z^ld|UWeS(T|<mDz6?VWjI#zErCk8ItKvaotv+1{LG
z#3FQ|?7Q;1eTu+{vZ1gUA<7|NXC3GT!6`Y#U*)hmo^eZ-@D>g#SLnEQN>CCTCp%Q0
zWWE>=k|h4<G)9)KeWncxH^$P&S5E~laM*ch*iKJ#D&(^{=K#qljR#Q>Cnq_BA&SfJ
z_D9)W!M*vze{DEp1}%hf7It#mF!d~-mNzt%uX*L2zi{gu7#`b*(yQXU8-*~Upz(Q<
zDTjl@;~JzR|FbO-Um^i%JWWYCTcEs3TK;G`7o#=dd3zVbv$D>Zq+l$+i~N4|)m3<S
zdhg@2=Yc2Ll7pnfE0D^AHW;I@=>SfnEyMC8?;2^wYPTCk=AS-&+y>~g&ps;+#r@?k
zf0-WbvwJTM;Uq6O;Z7IdMV<`VgNLq;b+_eH-Ek^wd+(7|_32Tz^_q@23k_ZekFURe
z(Z&a@&{N~22kUfDL+P(#giPH|As3xW!=N#}e@0cxj&BUR@bWZAtmAo&!1&7_E)=|8
zH$xD0T>NT;#Hditi$~U3OwX5We*0?V<#84s>KK@EXxUS@Ohdd$V}}m?u-%d)qXIp1
zBXUfZ?QDAY23}=ycuUTbZ|H6@HX{}tuQNgwB=Q^snj%~97(+h3tlk-|>Y9egK}r}u
zIwAESv^Ao1bDah^#>HjM_K=4<374lau1L?T=m$oKdR;@s0?f$v&GmG2V(PS^q&Jeh
zJ4J+B2f=W^h*9Hx8$LP-hq5pNgVksZ^=3xg=m+rVHGEO0i*FihoeD<po+>dCYXs%j
zzxp5Qcb@5_i-$4(vOeST^8URS-;i@g&-?Y5UT&+GI->4eURet>Q+fD%`#Pl$twPdv
z93DS;eDC9&odT~JJ@qj`<udX!hS=jrG4ScO*5ee7oZ)-;r{iah4n2I8-99xw0}Fhk
zBX$$b>8!yagse^tO9vJMFE|l7M2=VlOc_jf-p}HEyacOL+j)Q&xfdDEXc|HsIsU;L
zb(90}?{k4sFe_|Ko$rhIp<i-Wbq0q80)}xrt2R18%$A@!80XsI<M}V}dtBn5n*E<j
zEByQ7tU_x@xSJ@)F8C^oF!TLHjfc^R+mDu2*nWU=N!#>`zIxiycc`1+;!86={^9>&
z*na;2A49+3ZKGy{Zg3tWB#l5Mf?|zcj>~98;=`k(TpFvq?`zO#>5BU<e@h4lX>*OT
zA0e04`+H8j<gqWx46OZb8iq1}zY&7nv#^Cw12ql6j3mfYk5>cIA!704x1Hw=*BuRd
zB|mw}TOPvXQ=>ik8rtSc!-R9<>w&nRFy?ayYlq2e`|;O!femNQG0a!|b;lN7Fva^Z
z`iF+aDI%v_;7n=AV9MHQg$^0CUUp$zB4~Ujo}7-|(7rey2{~LxnWEoyr#QNO5njRq
zqIBHPK&HI>k2@}fo;VHmlV|vAVC;Z#pYOgz4g`nBXaZL06GMn{wL67AJWu21Ba3?E
zE}%EAMhz%Nr`w_nBfo}f8c(AR9R`me&pH+h`{ZqekvxZQfv@T67URH$8@fCwoukl$
zi-r*Xt4NhY3jSDz)EAKSGDjZA7#==r+(H#3Ilk5lLm*8>r={8e<W2b<d=^xCI_owF
z;Bj_i#49qB^1jX1bq85pWIOBTr}7xxN$J|SYe2vUY(^<HXc|4VgSQ4L^w*IIDtO-p
zDlg*PCO<n$w;XHibW$FqF`LoL07$$o_lzcnr$%*;0x4m6op_VCQ8aXc$J9a0*fI|;
z!wi3tByy87<l!u%VfDzf^mwrGa)<x$>TO25EOyRmqYKX`PoHGlxepb-&SKZ+_x{)a
z?SIJ!<vt#RG@}Zif05(S1J|K(17qs$)I~DPR)9kco)~@NhSK3f%7!N4hwbF18yQvY
zjKr^oeh8=rHC_ruj_EUhr9n|2dgDjwzRFh#O}M=I%ikF%?r3npHF93P8d*Ri9R-d&
z=uMb%D~*zWeDaJd%IF&YPMuis63gM6AT>d0J3IsT`y4pi{dglJDQf8*rH0xtT40I*
zFg%XpG<?T%|BJYz5<!52LkOIQd0qFu{5gLD9`nk;1$T1IZ}RVaVf-D|qo%w|qfv~4
z-!$G5Y0xlj(o|_xI6!h8yD1MJ+i7*}ea*GK&?rsi{xF5O-v{Z#cp&quB^w13&C^D^
zl{a`-d7x=73b%Q~K|u&U%RF~GQD(^DE(gh|F?XU3*1lAj<LlTzks2l;Xke(+XrxlX
z&3#`rVuCBBc8HVkZWm0C%g9R085q{oXzZl5Rsaqhn0=CG^68al{u&E+_(<0&CUMrg
zFvps68fJ#bhAXYIZ5SHT?c{e1U!3N<%5Ruh{i}cSQZ`QGc^@xuEkH_#gz1Tr$xer)
zG;lDq?P|jl<=AN=4M2yL3vOsLkKxnEgEprD{F4V*b~N;a#?a3wjk3bG(~F5zV+38`
znfOFR8K)r~7(DE59XB*r5Acpr*v2P4zS29`M6~MX-TZis2#vhPDUBU`+Z&|lX?!x7
zYb(zh%xm{jjnD9}jYaB@o~H&IzFi*?<Kimv4}CH<bxj!{9pg(~i@VAw0=~RU&lgV)
zMpH%^I|^!KS(2m?J`Jg!>hKaj_bHe;1OlmAbvz_R@rhSqfUQ07`FaFZkD4cW6Deih
z(_hxFpbrDYwpRyz;n6mJLX>2rVVrtsx5#ZiAXo4F^gA!P>rdk4Y4||4-eyEaW@ySt
zBj0t2bA6f~bk=u)5wMl7?(E2DgVB?lYoiiHJy}QoGFo+>uR0qg!4Ilrm#1=&{KDFn
zjc&Q==ok2mW-?`TagKer_?RmXlL>=@oVxA@N7HW*Q~%%x5Aub_J>e9s33>cY)flB9
z*B@tU_5b)k|Mz>p&KH+`oteSIXyEL!G9oE`Ql=~!!3sZ(AgE_XdEr=|C!KdjhwG8k
zm@_hDx?RU&^Wiol5bBV+Il5vxEM&gMTKHNQJS&TN!548;`UzTwqxEu($)2j5a$X^D
z%2~rSGI8ICVxZjGfrB8lYw|I2Ri{w7$ehayJX#+Eiy;D5$tHdNjkJMvV%xR!Kj`iP
z@3i0#M!pLe299ONx4n`gKLCdi!Knlr0oa0Y2jwsEgi85eggWF<D!{D)2k(E*1I4}>
z;klrv_yxNK<4|r|EHt{zrExxsfYgeNkl_-XrO~J#6gNnTKfM}{L(9RtRi%%tWM_}4
zxGu##j$2_VsN?hl1$@H3opfB&<k8SQfQI@g*#{It;;3Z&@3$ONsWmW^|7`@&lt|@3
zWzVG&A`j?#S`;UxV`@T06Q}vCyv5b9_sh=7PorQ20KO`la=y{%X(U{$vQC3FWq_lG
zOxis<)Uz$(e;&W$WTYX_rn0~>?5^y4%7koFusdACWETNtQMkeLO_UFRaq!L@ECsb2
z0=;G5%Yh3A|M%$}z9(JFl#vdGko2w{ch{*Zx4_u%8ZQnFFCzz!-(;v;gNs%fD!%tF
zMnV4CaV_0uIzU5a%POV)?kxPwJn<DJByIjm!O&sc$XYmB+VBHYToJD$&xZN$-~Xm~
z!3_;9K<+3;=*!SDw0j!^P($1X1YdI%UgGpF%cSMM!<5aMlPdGTHVwfkqsHJg28A^O
zMqUk1TSw!&GxxqRGI4$HlQbae+AlJVgT_93XIjJ>kZ&`OK88-#h-57Qg^i{TOnYTZ
zZoZVO;Z9-04?R+iL8cARGJbQOPUPi6W08@_l&^BpgB$G1GNl4f{-!~OCin`DQ3R;s
zKbZXS27P_~b=KY_JZU>J8giMkQiJZFo5l?6@Vy3D^2fXOy5W~=49%HdvRJ##Hr^!N
z5D_6=`N_38-jSbpUj_z0(>>}|hv<X1o^@f%GYuOBhPK1Eyho2IOAnq?#?b02!~Utg
zskbx5Sfe-4J5msQ9)yM+H3=SIRrXN=ZNgLa@J-fKxJIXqoK#<SA!*VxLV~dMN<@o3
zb*~Ft^pdtZDLRUz2esyr*m;XjXpJW=lj#%&_Uo7T{^9@mhaBDhc@~&{)!kVC??3!Q
z^~G<0_q!T))_zrYFt);HXPPL#uiN_kfZXUshO_l1rlFra&h$^FP9H|@y8cc*S$tHv
z=phZfbu2-;y4t8$Z4l+mpMUg4uVfM2)oXg8p_S{Ty!R@4DZo0~BWmfJvYtJlSEmCE
z&b!dr;(BEucRpMxk4V(eG0#K8kt4Fy=}4I`S6Pr~(R%y;vHMRmLIBP+N+KB@-0|}u
zDyZ=U?ch6H4=Mi~5172&&T7BU{Vx1{|9{#WXa|Ug2e4M|ZetO}hq%W@x=qIzuG9z;
zj>3s>h)`Os%KROWhuQ?tooj`Puuuy4GL<*s&9Fark@pktb$uA)z4zn^58~Ga#zl4m
zN%2%<<wymYxyV?-iOiqC3(L>l72twWOR5NK2yDe5FIVEb(4u0${0VkEtK{KWrH!8G
zm3PYDGo@stgX1@H+SdbZR4~qX2o_P;=tkMWsYXQbl9uun=Dzb%rskO+2MS5I!>D&K
zipxsF2$J431b=Qf1jTpdzl|#4r9t9OKKTs{q}%lA^DalqOM$?<@Jqj7k~;vUWx_X3
zrAfl3H^iXqf8jT8Kih#(7eL}cC~j2Tz_zBqD3eB)VIY!tb1sW~CQY<kWJAi7D>7>i
z9JtM+9W6yoc~ln-x5nSjmTAn<O*y0&4^nE9fA|RlRgZrR20I9BTa1C01Qa?0(bSxt
zap@H}Xh(S~E?R1E87n=pbkVAdtI2iv1r~f;hw~tvMn?UHtH7=~%Daw&1I!rJv)ibM
z*N;JFokbUzwoGJr%3k9md6Ztsuc=z-RZo&1dUSdy#w6GGKFzcDeT^Zfj4&TJ)bpzl
z{F0BV<|z;EostUfP9Q$DP9Tjv+DahFz6OJMfRUG$1K#UW#-XXY#t;jR49P$sHbJy}
zP;c>l<9U-&0(CIjEqBHY{g8YO(FDd|A2~+%hLQmJ27gFRw1g*SG8Ky}8Hj%x;jV)e
zzlB+eV$>SVFxB;04z*)A*;Vy2ofYSQSSY@CS()kr`$3FDhV#|MTKg75wz85iG#lf2
z)+~`ZS9yky!r>QrfkxYi^0=BRvK1N_RV$8D?nq1Fm&W8L6_5OG*HN2{G%>Ila+{_*
zPSYjw=*k$2v2aE^4*iT0A)>W+=*ShvbWD4O5I)5S+&hwl7xp3{RDX>7Nk&K+je8Lk
zWLS7D1oI|Lh&u^zg>IC_k42~ezUPNvea8jpRD}uYR@6$~am<Oa8E_o(Q=WG|;Oyb)
zHl)R1mrl6{ss@JBKzpoG3Elby*+f#IVPBUBTOhpGxQH;zOCwpa+U1I{Hf6$^I`k!M
zhQ~Fq8aumJI53Z0^SDE?xHd-O$qT%l1`hbCl7QP$j68RwAo0z^>z%XgRUnPpO%~(n
z%~@D>lFyJgT`JXf(>s}6F{T&vs^kd{Jd$I}exCQr8o%K1-m`c9@QQ+(x4iSqr4gv&
z1tV@6rohd+GU<H@H$;gsUDS&&;^sO}Wp%Idk!KAWqo-eXjXxM*;)<S<a0Ud|5s`!`
zi#V_HO$=evd(I(MfphrUEIgBcM-0S~Z_?B#N*T92>}{`xe1-RrrNHl(h7%8-@-yrz
zjof`zmNM{;K3Ca(Y&VM*FJjWh_(6`<FWPI3MC(UsJN0@|If=MR;K)eg928ALG$@xj
z?bCTJm4(L7C{X~wWtX>i3Nlj(Il()lFyR(_Q*KVV_4wc9G75M0rJ?KWX2`Jkmc4$s
zE%_TJcb#Y6|6z8^82)a(M2WzBk&%v*Y}qt&MFBsG5rS_T72hfO&;Q?lP7gIZGh%3v
zTO(L@Dv(!rye;32V1-{Ai{PQbkcV9wvZ;ZpjJ}v^2s&jb&*(b-Eb>r)jEu+&n7Qz4
z8fb=L;FjkzYR5n5cUqH@hqJnMmR_+8)RTB7ol(lvKlAW-ocz>Yb)~uqF2%<^;o`tX
z#D?US%rrh-ut=7}$C3G2@O=4J;}_Wp&iL&F&UfiST?VH!x+@LP$g7-^dv2O|pIA7}
zA%60`-_fBM8OCptw}sFQ5q#h(A1Tw#t>OP-+%nqXW=e$dSlyC5U%q4jyl8#0aNRBw
zQ#Q8T&m!5>1tSH)RXHkTfvWCtMoN@N-g1&(bw$d7fQ&|!ql@eV$J+64UVN1;%`fl$
z-~ZQ($e__l_+RCGnHTMRyv&XpIG`n7<#TuGm}LD;Di+#UTyEXVgN%%Hl%w>AMRE^M
zT-pg2-c;A+?zu*BU~Yqq(Zz^IOX0UME;{9A)HBXAewGp7=Q__fWh_v2HVW)soPQM_
z_}XrE(<ILaSEGMT-7?y)qa^6Yk;CAhxTb)T9p?6D<JcW^f#FjejR0>^>XgO9ozM8g
zYp8#C&+AV!LUu~`Akf?T{cd)Hg2E!U>|2<||1<s^XPj(~NwITj<vWhI1OAD3@CtLM
z2fOe$O;XlA4m0vrHC`&91qHCuY7}^tMx|3g1wP{iOad|S^ANxJBC0SYjb9$OU`#pV
z@kiVi{>{JRsIubl$umd`jlpS`8-5AfQ!f-ndw{A)yEUmOiJ@YSe<BB>5gP`n-oo@g
z6DJ{6wkdm`!A^K-EKHv3zSq;#5A1|(x4+RMt%}qTX;+-CK94_Ijq;f_7dw2znaqYC
zRkEqr(^DAgNbji%q)I%>VBs#lQ*6yaQ-n2Ma`)VH3D;X@C^A5#(>Mv2mIHg&8F}nc
z7~#z`ad-I}KmHmS!`blc6kBLu=vRDPlvaZ<JRjvtxjCS%I+$Exv=(gsfgTzjKOt%h
zKY<4tP=N6&-K9t47(%6Op1Dr1MNZ1=RTlm$c;e=rq3Xagr_*ZuW||~)3-2ge;kI-#
z&Ww^UeroIn5DBVrAYQIl-^LKRO1K_p*JfmtBt2WxE#09MSv<*>V6>W%kkBN<@a0MJ
z!3UDu1_q5FyAYlloP0Jtf;)11Jvi^VaJIZidGIJ)Ds44*w}D%JOH@*UjGkmwEekDc
z(8wB1(MU-Hc#VKaDsAyA5Lmdz^({Ye(5MKSk&fiA0M$X(F6BLR<GTC8y<sc9XS|Dn
z1dgQn9|tk@4v+Ip-qpf%n&ZDCfh~_oH8QWhdmmuL5I=Qxx>ea|Nv?vAu$n?+Ksxk}
z5!r>EbdIvgrsbKs1*S3=pWx=XMuWH%Z}Kt{QYt57?J0awzsH_ugu_(ji+pQ{-jV>X
z9F|AUtOq*4$Jfx0J7KuxeLl9g>jgqA7Z`FW&c%106;k=#TTk5KxcuhDd*Ufi{+y!9
zdB&SvW;vW?7m=eF80lB%jy)Ke>Ux{th*w~~i-D<JjJO&jp=)KOWo|A33ztAmc1!w7
z>Zx@w7(9xr*All6cRxcy0G6AFA8mv*ixMy3bB81%y80ZjQwj2L6jm!d?{~;;1pfSY
zRr`R$5%`Xf9KHW@9-9Y-_>08I?>Ns^tPpDyNGeo%+{jV!SaiYGNKDX((#vn(180=@
zObKYD+bbpz3*l44V0muJz`BW0aVk6l?*_7-`PqDtrWZaOaJUN>f5-JKa?6&-C3oK`
zXUknGO#aHNVt0<P!p_i01&Bfcn>;(`fICm+u%OI#OhUhkk#|GyS>NFE)T_dvvws)m
zG%u4U9HuxB7L?dLnq+8`x2AC}2AI?1%{@I7jlTxZ!abi)SBA5U&S;?LBti`U{w6HE
z)-%`G<AVBb4&GMRtR|?^FiKGaN|wQ6@(!-S+el35kc3ceSn}7HY7i+M3bp0V^E6gQ
z8qy<AyiqWVPbnu#Dl4PMaDTrw5kHDh+@2GW_dA*;j`HS0XAStYH<VL)qk4)(N_|d3
z4x(<!V;*>|=Xl6#_iQQU%1~(4IDq7WBXA>M=%(S*5REbAGw#Z~(_1S+CNwnYVx(!)
zgranLS3exn{prtskvbOsgl7-4Rhs+9oTFjQ$G2Ip^6J}Mxwi~S+auVs!vPK&HwxNX
z5zpv029U$ai~t1l6jy#dh>m>aJw@!A5D(p1gI|f;UL&0l9Q^ia(-LAsi1omOQ}0t(
zB3oIP8oVY?Apski%&1Eb>Y_tO-hCqACV8W5{q{Z8V+`N$zj^8TN5&%Kmsyu_o!v3*
zBqZ%Dy0xHjhUsrIqN5S2fs=f!c~S1ek@VU5(sChyr`!`IbYCUSyLx2Xvu`|+at$}5
zk&J5q8H_g!imuxT?#z_X&DUn3OP+L4e3Y&<32(Z7tUUVJy-)M8LULiP#ful;bX22j
zR6;)niF_Pr*QrhGQ1B#vgZIebZ5RK>KfKncI+(RaVauNS(S~@4*R@iNtyC1S$XLrB
zyww>aFm+*4Wikso<rS*5?8-pj*!HdACsT05v$sYaa$us-26eQno#KD!MCF;@feDxF
z>%{GPA^I#MQ*_p@*Rwk@EsDswPOK{l-5GQEWJgX%wlb|Ge>8*3+36fACvMxVOhsu+
z&N~v75>*a`c^<s958~E%Pk5yO&Vd2xt8^mN>FuWe@3*`1@cGe3$R>tFKp(&J`@VO6
z`x0ju<moTuKFDCdeu&#YJ9mH!{P6G{khPo>jL9zU_;-|6#6L{iP{gYs5QYfbR@U(h
zj8RbHD3364xt33abxl6Ye@ky;-tps4(c}=mV8XWzz=oImlws}*d~Z{ZD6-(EL6;6a
zc3EqYw@icSWT7%>^eHk^J)uD=@pc5hN<fkzNBF&0(FMLd!J+iHP_p+$^rk_tF))nS
z3>~MOca`54N?WlCZz`3_m+Xtzto{z@<cV{wn8Aa>*i$`}0%Ty&@H`ud85}hvp~qlX
z1wt_}Kqwe-j-86EVc{!Gx*fwaWojCOk^>d&L!`v7QiEU0W%Nb(@Fk-InOe}1DqW&f
zhM&jK@1BGGF<O&{`~$CY8ayhK{E}D8m$1D%tnZP5=8-sP1h<Ze<!-u0di*gB;`|P2
z9kV{Na0k;=9(p+o#KB$ahcZ0OjstSy3%IECWkyWob9*~GI>?Gc_mY#2-J-*c-l6TL
zwUH3Gj4PG%*b*#1{0{$<rZP<-C)iQ$rPC3q5@i{8csCayc(q}QKJ=Z0mFeL9P9rCe
zl{bI%!|UeXJ80xA^~}hMIJsoXm~)jsfmb;SotOyxZn{om8k|ly1|EM6Op9N17IqQl
z;xVszPM-Pg9~^U)9=Y$5p=TS&@G}e!0DMZ^@H>b#ZR6(I`|u@U8y>zUjXW6mHPVw-
zgNg<WR{RIEje6oUutBB=rYvfqT<~+~Av^gE4t?os4fxQDLucvSeSzDn`U4y5HaPi;
zg);qvci!rkFoxXLb#N&!0v{hr7k+r<f6<lB19t%)*(t!G7kZciIn7!Od`kbP<t!S(
zk3Qg9&k`&SV~Jcah90EzhrXz7qz2zQy%x;3-jja(Q`&?lAot-#`4^b{T{Ddpx|5F;
z7qPDMt)L%r@@|n<gdc5$h>vhGY5{TO14kJ`02zc1;lCU2gJeJIbt={>=uwJc`*`&E
zxa@t7FGjyt>>~g_;;|$PtioT!9wv-h5whJTFKe~&7oCW0s{uyEv5JM)kibk8ZeMxV
zLlDROqS1s3d<0Z#y-4o<x>JOBGB_nh-do;;x8w_lCOZ6lC<+q*Cu!u*X&?=yYb+=k
zJmm<xM#Q3|``M*%o2_}|Loeh!L*Kii7MD~!@D}7G4(#~pMT4UlyiQ}0>#0?T%0^-1
zPoem-?cUN#xpgWi!_!?L1$HY*uxGdw_05T)x4hA*4RG#?VOaPo+XnT17(PZhoSeAn
zNa-=4QpEaM^I-d0*Ahe#pov^G2r97E?v#il*X`Q5NYCR8pV27q;Bjc^+Vs*p!u!HY
zet0PipB<`24on&B(G=n3Lhc<cNN>uQUiouMtTC5owDJ&FffGEqFvc{#MrZUsXSB|i
zTQKyVxdS`^(V&nUboim%@Z1pfw<%xOPz0`FbGrs=9LYNldL7xx`<5$d<zq`I9z00T
z(ax4G`b+s8)2>wFw0<OxkCc6qL!O)_DeTqPUzeVE>lpq|68<E#cle#du=JcYxZMpC
zn2)nF0X!e>V>tS@i3Z4Fdg#gUK`inf@8YACAY*Oh;bP=0G(OAPCk=z`qgI{Yzd!g+
zUB-6`L)?rwb?97OjV`q~b5XR7b!5;;MdiY|Ct#-WYx+gI+kU||FK|m)YXEs|9ZVjP
zLt%6e1J2Gi4UN3;RHHrR82ux!HWuM6<4iq6BQnz6D#?QzIgqEX*N%+mCHbd(iMz!@
zyyBm{@VRTIG71B3N1d`hN8GcNO{41oLkyHgfNj>cO8f9&&&LRCvTw~oN2NMBW&|P!
zuw}!?MbhoBFkggc`1bP2Hyv#Nn0~RA3SPYuf7+4cleH2KA?w278ec|*@Hlt`2X#Xj
zyk2Kl5BgOXAjooLVeq0$)SbXjd@_T7<QcCwUAnk8y7HY%!M6(;<kU;O4Gv-)T*<WF
z;g~<8PGr_s^LD!ocW#$A(IC>%7^AojV8e%RzJ6JK><iM$*OA2VlYWu}&hX!(f87ZN
zFFY4RSmJ?gG`QJR&KX`QMLilrtnavn;E9mjf07YW>{ddJ$`P5Gk@sFoA^dydG~!?I
z=TP+Z03W=8(x2bS<dEY5{2{Ld5!B|7VXhT8w>ugZ%8H|5-{^{0K~Jr%X|d9auc>II
zQMp#`%_D&mO{^8dW2IeD|L6_x8b5*uLMs^|q%aZ&CwOXPxZU%hP+YB$Dv8t!RAJ_2
zdK?rMML|wfV7Z{?s@QqA4Vb@$w#q)rYAT%yF}>%3m)E2$g$hSH5KGtbDc<#mBisWQ
z{@}{DRZJNrf_m<rho)fXMVY|P6=pVU;z{HZJmF0jc$CUslTw@#J}{R|BwWfidMmw*
z#o?M}z1Ii)<-gaI487w!kq<cV@K=7NP{liO3VMFDC%tZL-Add$6*&11P?(H+&$qbX
z<rr1M$Yt|OKBJr=6)3s&Ng7wuXk@sa-x8(Wo?r~lj(#Luytb{27UJSPgAIMiodv;d
ztYY*~&ZS-XxagZSm8-~`Mrakg>tw>6#_xyq($E~_p@mGU<gaXGmRIpsipqz^9*+ET
zM}q}_WK|=bu*le?Tj}8;jq;)Sh*UwMTRRsqav4doQvReU9?~&-P5n%x7kX&X<9r8b
z_pFC#iQJUK|75qy0V$cTawVGbF@}n32vp;-$>lZa&tf!)r&A)!(k0JBPhnFog@fn3
zA3dksB1>D0)0lC1<F7_d_{X@~BOvnIvQ=3rP9p=-il;-;G}6UzxNp&X{BLuif3p?2
z4uE6l9e?&kKK7+v7SqI8PaAh26kI`zy~jNJCmV*w>XbjQ5x;ShEJ$^_EBJYCJr68*
zhLAd7YqUHk4}jniyz1vYJ|Se&fYWfEho?!K_ZkKvHACbghTu$v$9R4-y1l?aSlM-W
zRGEtJa7RaerhyVK{&auuL{vHpsq+`<7;wkGn-uX|Bszo!(oZu&4B3@nfaN+o6Usnu
z#d8;Wl<^_QztjUF<j+4gar<}f2aksR)SqF-`>Euo{DkN|mvUoUU1*`Fhr63K2ZfR_
z>GjrFoyR*p(2>w~=S6xZ8H*gnyV5i~&){+t^0AW_oZB1P0rGM!UTp>eLoy0V_`-dZ
zz>1Y{3WWip*Di_1r9+KD;U<;l-OYSzeH4WrBiztvEOt1zR}x{iun>L}2RLu@5h@IO
zcl)GswAb0%#^e5)3)iDjF|ACO4o!PpdFNWwa0GUHE*dc-Hc8ugTxxv^rjW(ud?`8j
zD2usec}t&`FOLwhYsR_&L%JE|NE*t)qQ%$dI`w)~%3R|{0Xw{iLxYZT;9c^6kd^*u
z)sc|AEAw`{4(;7S%7;hd+@>t2*##oH9TIRv)jBw&adV`&ZKsqP-nCIogn%i2dS80L
z%8Ca3m3=$#j9yfc*ddWH2TWCdVjvznjem(%BtlQYN$3Xt%j`T+Mpvga^oT<X<>?T7
z;uS_{a1tJpsn5UoqJBOm=GbgX65U-L4NBR@xKyUmF^A#o8{f=$bMN=R|9yM#;Q0>7
zqnyKl7Ox{`U1O5`-i1Eo$0*AT-A@nEnggwrZM&~JVq#%xMo7|_=`|YllHVB_Eblam
zGpZsF^iZZT!sn~0eCH%j>mX(*+&4bRF}}9Pt??nA42S%+bCP$g!9xw<*}1qH+Ns~y
zHP}N9<w;rZ2T%E1H3o-w7OobjcrUkUCU09@jW&`waslTsCNU$Ep-UHI(pPb3kt?#x
zsG3G}Ml&>y=wg&jJv8E>j?2G#EOaYXf;XAGwM#BE@XT1`x|Tv?t6V8(=}3;^XJiWh
zEKEJknz8G*BfECQynHw3=oqD`Wf54aVC4|t;9J=7ya^0DVXV2(*<c#C-llCwquZRL
zbA3(D=g9c#X_$oe*ZDjyooXZ}WDcG5?(gbt&({dV@-fhEf;*ZroW%!2T}x14(IIxW
z<(@xv%^IX9nKE=TYY!7kSshwdnMrtQx#|`E14Wf_g`Ic$nVxIh&`krwDCuZh6Q$V*
ztWL#{l^DCek3?@;XzTDeqmh-1tCTxNgoVzsx}saA;CTa^p@+vt=t>Hxj0+y|O1QA(
zKl3uBwFrgdUdSGjPco^0?)dk9>m}6>xC1cA#mRX4)x%H1#?O6x#pCvU-u`mGhkWwj
zxp?&9FPG<ryo$VEd#?f}z`5GsEL@5ghJ?9g>`^K-bWy4VcitDh!CK*)$U}3&0eD;i
zuIKR?LL}XfaZR~k9iYe)XaH9J>cugx+rVkBQKd^yZuewVLXBq26n~1rP9Mes<q+z$
zo>f6;{A3g@t<v>aae4=g<n8@7r!(rUbtpUOA=f0S6c+8-##p+fax^gMe5PCcZjB8r
z4Wvd!xVk=-5bXzUjXkY`N)Ep~!=d=VJ(sv}PczaAEuQjJ;U&+~3+y9z;8XlOk6f&K
z(%_#)yJt(}P9ahzosJS7*Ro~*=k)l3e~QU37&T8g!A05NSl%?@#4DKQB^E^zCF1a}
zf}eDvZ?~Zxfi3vy5hYaKr4JgR%fd4NJZogz>k9y|C{OY7(`)1+xVF)Q^WD`38Y2AM
z7yO~!7>F7Kiqh3WXow;MuH8r_`N&_=&rTEN#=FENli`~wI!1=LT=7r~iI`#r-GMV@
z+#?3)OCB6jrRv=38M?>s{rU>9TuX&{YW(=*KBHmt+|jcb3A=GaGx*F@<TMI!2TLPh
zy;6ZBhjt=T4;Ze!G!kf|Y_e8}UC6vHbUf9;<=B-D%9!*BY7u%HR&}?%esoLyT)Zjx
za%&Ot2-qFpf<pXF8_8ce$fEKxX!yuio!$7N!R&Nu@JbIzuDCYNN#9OIJi#iyDKs$L
z6uUVW{++yvu>J9O6Jda(Nz)R7;I^Uyd2lUver9TH@ZWk%d~#}<hViRD4{uf*kg$}l
zRA^mkwk3?*TDLd94;gw8CCK^1|DkVSg@zp=3ckXemTT(O`kHR_r`NJFcl223oW;kL
z@ze$N&?q%{H0Y2H?Dulc`yDOhE4}-Bv>Ec%YkoG23HZ@Qh-B7$6bHf_*G&5~KV@MU
zP@*gLLTChPl;ctKzt|&!*R8_;IWq#gL?+wE9Ko+eVcDuYB=#~5CTv)s2+}yE(I~0a
zM22~Tbt~O!#M+CD0PSM*?wv9?z!-esLZGEDUPriXT#o6GXoKVO^bX%9TM8YVHWcD4
zp6?3leiyzfThbSQ7Iv_vFepdzphsZarSSP`=44Fqo6803U3#i|W_o;jg4UWSt1o!!
zUtM2YyO0r;^lnsAmU(-sfo*XW*rOz!Nsv*KR3wVL3}_K>Vk~yO!nZli&?msF<kLHW
z!%cfm>FFAQNZvhnpyxSczO?iB=IzJ08s!+$B!g*SJ*~49@UZlZPwpK2HhuPcdy}D8
zN3)WjxbWy%DS38=h0|)Sb09-|E3$@J54~k(NJSaI=dBM~k&&ewnqQ1HN6AxQ;7`x6
zDD`I=FLjzan8tXBHh7B^tPV1&$0c0oR&?YSFZGa5l_T#ADaUFgQe{E8_NX0QB7Kje
zk3T%gOZnN6oAk0%F1^{k4Y!SXU@NzgmybTmH$I-`V}e-+Vuaz>zxq{smfvJV>Q#2Y
zym^(axRHa>E$}({TxIP=R()puLC#M|-d9<BVI;&!yp0pgG#1Lzo?_(dI=Ge3ZPb!H
z`EjX<y>BDL`$$>6r%{Q=xiwRf?VB{X>S-IIwfh6D(6gg`tQ~Lhvvy@hmPSihvjTto
z61TL>yQ0B$&g=;+@9+(-%I>TRXKol>c=gi9<w92tO|J9SGknQtYw*iP3b|)J)6@s&
zxS%^-gjS_XGB7HN-{cCe!r%ih>uQ8q%fVnj&G%fK3*);!8b$mneoHsdk=q4KXv3%q
z7RlFo7Uyk{O=l0866#m2$v3h^S7b0FR8OJvZnw;C+{pxZGYYiUs9YLZG&-0%oH7{Y
zL@@U;Oz{W(Qja5tbV&UJoPkP;)N5(n(I$Ib^ERVCePytUk}rj$n|e9MM%J@rr0?<5
zNBL-8MnW<g;+zwsKQ$1NX?I)2c<8!LFaj5>l)Ca~@w0M{u^>FJd2yjB{pQ}f7DK0_
zlF46L9Tk6;buiB|B0^7||9(cVUtdM;lgDjzBG^j=2X=6|4MB!qGerb;%U5b7Nz#x%
z?fhBSGYu|{x8k7Olp|o17dhU3Kg|d!TE%m6KD-Bum02-@RSCEvG_*O_L+0DvJ#C*K
zxXa>4xD6ZA`Y+o)9s%GNSk&mwU;OcFz#%6wg!M1w@>f?F1oRH>hIL4su!zv%HJq;^
zr?7z$f*C4X8~40}rUHBCF!qo|85bdOU(e!r<~EFmEx!hQ5I+g(4e-YPfaUVoFehU2
zSokz;(kr*9;=!l*ZhtuYTZk8*q;HxiJ?UEJD8WfLjh0HTX|e!sD3Nlevdq~9h7K)q
zn+m0JU!BZQYmI@V)A&+Ga_@8i_+@@_gm3UVJ)5bXrtg%Cg}33-13MxNF^|%<t&tKd
z>?jksPcPf{MsX?O8Ev6p9913@DSuH`6i?C`s#ZBSpS^QnZXT7_i{G2C0p9!?InQWL
z8mjyfmxB)&l*6nANSc(jyLT#e%e-ZXF)AtB1KY>A>?<CHF7jw5MK&3%xE=(xu7tjx
zwktoE58oz9WOxO^G)z&6X_y#o8YzAS20)&rso#E1x=zKQduWS}KyCRd{3Kz7IJoR#
z42Q>$(vU{kIzt9}HFmQl_w726Nft*aH-98$Ds*va%1&;$=`|Ott}Q5E6VIGFF#@Lt
zgn3MUkYk$o>*?WX4Z9^>JPk@3Sum7=WQ2qLZG#lL=nZ!|NKh_kh9qT3;-bi>4h^O7
zFzPzMka2qa)OR4KQ5AojG<nK(OuMFwx1A;Z&W>^iKKU%Jxr1>QgIjuhgHD&MXovw%
z0Nlt+jqYXy`G&Smp#K;j(|{ik2{PoMauj%_2T=KwKM3u$S5AW)9w}>(Xt~=TPlC9q
zJ7r?N<J<X_!e&&#&xq$XE*M36`qnuVeheXe=v-Ky)Z@x)ITP<7(lbz!r1~Vq3pR4i
z1>a+bF7c0iRJW*$(P4AON=hi-0Hik@3BW*dT>H}}&uT1pp}P?gv{&xL!?S2Dg`X_Z
zD4D81WEMT@2A#woT9!^VK;$<(6@`AOj;V}?*7;)u+OGQ~U#hk!_38zEmRuu8(xBJK
z-{_FiPu~4Y1gOck@$w$rVC8C=Q&|p%Ta3k`x(v91C2c8JU}J<VAOl#0CMzO|_1?WC
zK$Tp>oFE+{Que1yxOzPELbQm3*MvORW3!*<cpdVPA2-FC=jGAc#trGY-U~JY5gafJ
zdR!}}=lFA7SR?;5gnggeJo{bgh*pc3bd$V^CU^lqrj>k}S0T;MemB=QZehIgtWr+E
zZd5zO--vhd2(Oa^H)49_<g-`>v8j>-EuM9E4OEyIpSg%ZpCV|3HyUhO1yG*1ks+p}
z8focA*<|(nWu6mn7XHo3Z|~(tDl!9stAb5MKh3b86Z<rZT_2Kcd^%fY^_52Y^yY2w
z(U7NN@xP8Gu0hqyE!@Cg3`k_3tG2Fzn_jn-{PQ;D5&dC$78K8$o~Rh56dLl^%bN;3
zG?IsPAD3^xZ7<{|Tm8%t`z$)+!;XsfVnT<Do<@4pUZvy7?SXgt<J){?Q{|N>f1l3g
ztbJi{INDiH<av5M_}n!y2?rmMDOq?hYV4ws-<F57gaz%C*@(bRkBdfI>x4MHS~QZ6
z5d%Gb`l&s|(BJ3*o;n}m!S%(xM{THvPF?GeyjoX6AFF))G2ti)7iCf&@F#b1C*grV
zjWyoGr*VP@9cp_se-7-t-g|K3w9goq%9eXaWOCoKgO;ArPuX9+dKo$x9kW}*{Cpbv
zOgmlYR95lG$?J@CNawUxJ=*8bpM^JP_dfgb^QQgc^FK7b#-TEpUbvoL>s(+k0_{_h
zCs}gthF-jphsi^j1$>cLWlBxM=S}bFVPAwY=)HQ0*=?bm8dR=RCV4(f{ZE!@+(&Ls
zl9$m5dZEMO$%Da#_t!b#4NL}rU0&#AB&dydfb3BuGGj_Hv5FHqQE6mSUdpuja(0=u
z4KcipLR#a1M`x{v>ScB%<<wb*Ns8Bew&C+A4JKM>z*=tW1(K(30t1L{I=xES{pDFY
zan*@$zs+f;;bEr~LmxCu(yph1U}Yn+Q|jj(t-(NVBLd{Dx-K$XzGUr$DVy6Rf47<P
zDqX~bcwq03X3|eaMy_+nSSeA03~nMCc|h+sZzI=9JA55k4+^DS$XQn!hL-(~R)D71
zd7D45v^P8-I20^*Yg}$}?#jCu$8aU{4<0<qb&(I@ecbEQPk(;z)1UwR-p3z(RQV$d
zbW=x(7QK~+Atxt0c<|vg@~yo+iIIrk`?CW+n~-RwQ+033uYQxuS2<z!b#Rz+ppjy|
zyZ7u#MpeUiM!s)9o!tK~)k80H(lFYKhG%$zZ<VhYb}f0zSv{Jx`0bpX%0uFk33ZG8
z)Lk=Z-)4j!8l6YWK-iyu&L*k4r2FHp%=|FQPQCp1>k+WS<Kf|b7(d|sex#r9w&875
zBrWPZ64=}E^Mbv+8lh91(v3bjq&uQpQw=GT^7}Gfa5te3aqb>=jNVmm9N(Z;1{Llt
zL`9CaQJ~zQ*Z3r5KlkbAZGswFIEVK`#P=Tt;DXy^6~w8a5y*sBn0KDNtIz|us3}K;
z(KQua1Ed#3q4Ou$u`BW*9olp7D~<tDp4E8F^WwwMI}sruQE#A*lP2nSUCVRd#6NPA
zT!N%d;l^0us~m$<aS)~WOq?p|RP_5%GSjG~vIZYR$eU=&paD71AkW+$0a2+D)qP*X
zo5T~>MfoS+RNcbbhKq{N-H*{BO)EWk$y_e_=IB%ct^!m+21obyn1X+|_~AontO2lw
zq0>?z)VQGZH9!GM<`OTQ3yfS<zS7EfoZfZM2T6jvvY_XY18chi_9D*CYtbb6!bwA%
z_PC8TqYEEO^TL=ZHw|8gveS_HR6Y94NR@LDG!EfSj9zkf$<Rr@k8{*}jkMxJY33pE
zhR?)nJ%^SZamago8_&BlFHSR^gWwiV!d34R2hPo_$A21H4Z`rl-1$xDdjDqP)$lMX
zkyC4RB7E+^9X}ZuC#f&!LK63H9>xvH4zC`{{iKuk@GE{Pb4L;nat7LVB1$!B7RbCU
zr_B3`SNR^VEicO4OI_H{<ST!20Y|;;Q{<$6%9SVlF=eMNmA*-zfZ|Se2B-Wkycxtg
zGVMJ>H1gMS%0wCSa^nlYdEfFeMl{~4YpY>RCRE3)&(pYX`Ni+Rg2a)7UrS&(C}oo!
z8L7*QY5)&eC~Ty+xS#<%7z2-UWV-2(=O4K~y7%<i)A(fpc8oj?xK9u4(J1sfO!1To
z@mSe`>5mb}kGydmrAvc_22gXi9?rt!F>GU)LM~w^oyr=cN!gsIV;b(Gm(OaX8cAZ*
z`C_rcwru$s8DhkTj*u*811}j)Cw1;&=p;(wk%`pnj4tEbY4i(t9hnXUyxMNY-h{>}
zFTe_N?CO8)@!#3=LGX^tjVN(p(SGQC$K5m}HveQt0CH`Z8y5BuI6rM4B2fG&FS#{u
z9>*@PA9^)-c>>q8JU>X<qON%6!rO>Op~^p1IX6mTlqOtnQz>rJdu@*<N`k_55@m;f
zBdB`As!&H7+R&vIry$*~o6L8Q4=RBk;fGMLpcHyjsf6&JyXPvkt@J5Iu%Za2@fCF%
z1<7m)wZ*qR-aPwWg?u%9Gp(@(36qN*4tlN5FVL`EJxEV98B#iEaS^2gP7RO1)R5HR
zSV~kmIinVM<C3loX8ep|_^!;Rzk8=%#7$Z~7K)o<GZG@4fgrygVNDC7OXXnT%WpgW
z8bduLqeRMZk=~o(S3J_IEd4?s-#pQyf#2)wJSm>5H`78VW~axJWmHyZN|~qe2`(Hs
zWWbw#avO|1PiP*N6USc_dHjs{Z2G~|86c)vs<6|O93y04>%BSS!0D%7fA#X-i>I#|
zcZPxkEJ!HBv*%A*X1z@1)Z^4Ueg5oOd2m(!gpTQ@Y4BzpQp%z{Xc;(ncq>_GOui`9
z$Q8QU=|JJyb%n35lSlDLTn&X@G8p}`@@k}<{o+JdL-g1A*clqDKR%Y$Exw6w&B%=2
z<t6w8sTjKq7}a1IN68o2BMNn`_oBzIGM2#dFO4X;p1Fp#-jP*um+-7>KqI_u9WFiS
zm60cR6gW?ww!tdD?<|oWGF2l!^gw+0F`(h<M_u3_1PbqUPE3`DGhEP<Jf6mQ8=8*6
zNN1p#I^wd6#U&lr5fHj1IID*z88MRuQZP92c{jo};0LwF5pR!%R}L3UxBZT{JFjp#
z&x}QQh|{8ADq?mdP5j%?!E}rWMziep`ps{CbMJrp+rO<GKh7wOU0@6o(;@1llZ$s9
zf|n6Mqez{)4XE!pc!gj1aAsZ8I=84gHu&Mk(9A9{pFW5&z6|B_lkC!K-HJSgY8Nr=
z)w_!r<wn<DWT%=Dme+YU%KYVLUzB(I>&|4L27k&|V=u7!O`hO#m;%xcB)ki6$wl<Z
z)UA;ve2GLQ`xJJbe*Es3?-1kv_xSTsfMkE*_|$j*kbF{_YpS>BzD_*9;w)H)^b_mw
zTX5@bLFn7D|BhV0z;mzq%d>BIG+in>-8elCSr1-SK1txqi@QSC*f2bsv`+i$js;3F
zm1agt(m+sz)`CQdkANnZ`ESpqnP&8Xr=z5cLutZG-l-VT3Ekn5yeZW71QR#dESRGN
z!I-m2C`aLNUDtl5_rXgjC#zch%CFKzfwY&K0LBXi+AWDGmkR3YaCG8&mdns6igNRN
z-T7106ha$_`1dS^q%jpksdQ8#@hBCpMoykP+QP5ANJ*0aOgo~%y7(oEiz~G>WOR`-
z!Znw2Ok?3K#c}`uKmbWZK~yun@@|EUPPysO_NK(7S0MQe(2)^oo5jj(vP0wLfQV-y
zHS?go(zFK{8W;@_{@=>G6D2)vBumuFmDE<6?rSss{!cg8=C(~~qeL<n&iRf5k&#KN
zic-~B*EJzC{)oWh8ypS?0)dEl+sRjsEtmKhHKWD|G-$`T>8OnS(r1MgxOCR)NJl=q
z(3ZyF)p2$2jig&g!64GV>0IbWU<3rxbY!zn4Y)eMBqTlmIdt1PM90W37fDH2Czq{<
zmOyVJfXKz@=B7W;)K3sMUV83rentl>?VQ>P_C7G=kL-JHet?6x^g;B5r}U%_Ia+;e
z+d9jA*Dgc`fmxm+4>Y*PlWRicxNMrS);Cj%VgC%n)u|u@s~s{JlE@`TFUdXhj>DSH
zJMEdPFHUC83*3O950B_^1jv)43!ZF1Lnu{*|2c(*ak@Pg>kJAJ^-u0x!Rcz9T^GhW
z&oz3akd3?}E9*;gQ+!efca#SGnI-c%3Xhgv=kcs9)4O$h1~42;^*HS^avlK<-;^$3
zb@KRkM(-fORIc5=45e=(>2e?6#t~t2r#*l$vh`@(p<PsdUj|PGPv71BC4-yTp4b_K
zoa;P_X3KtxJ<zb|O{*wB(azAHdR%{{3k{Q@r_y=Qnu}J_5Xy$CxUn6sL&Hm+&iA`g
z&T)Un>)#`Xt<=9&W~cA-mgiAtB2=i;I;IJyE~iD*N=}tdE(UeNDo0tCQfyc&!^)O?
zQ~$Z&(4XaaqG3SRD9Yif1Hu?djp$Ja<v&WhLmN->+-D?fG~13e(}NV75_%M1CgLs%
z-jP}wjVaQ*JI}`%zJava#W%EVIl0y`x!!R-uk<hD$QWJuHU`7dwoUQtpu;h+AK@&1
zG+-$WEj-2S<-l|zH69Z_rp*JBarOM1F(KdAw;E@r_p_d4)-+_qow_h9B6v^V{xA0U
zkZ(G>z+60tv-U$B*=6;I96K<gf)U9c!OmlnallZhyaUx2I^)zw=1W4d*$sOWWa``g
z6srR=Vy)cp5*{Lw%Fu4CvzTulU)<gM)<rt4z}%xr2H2h!slkkdbvD7vIFS=#BE;QM
z>bymN6tLb%9YZd1Q@zPeJDV6zvqF(v9X#?jf`1#iT;#Am>NQ8HbdAGSqn>&+zUqiK
zrLQyEG!nH7D>?p>bqZ#Zy!`^t{vxV3Xr@CSVrdjr!BD?`>oPL4TW^4u&E4_mwUX$C
zPHGNW%MiYD?1gugk>bHio^>X3O@3s@@hI<+W%V(#$kL9s0}%V1JtN*NCo=PV{VH{F
z^WgE5JxgXLOWNqfKF_yUIFns_l(h=>(b1l)se!w*58=UjD>S~(z8u<U_fkndG~&6N
zXVFP(Uvh8u6E307Ih`U4QSxyZC$(pJ>`|hOw`U478aQr9)}JK38;zyxIBAj3l9^7>
zSI3RI$(StCop&*CNDQ;uWpGdBjSMd4+fd%DyUb&{_xTuI9j`joDH0yMPEGcOQMmMR
zbW{C_cQ(r~b_&n4UZ||elRtFlH~7t7U3~H7ZT70k?C5{|$A9n6zP@5>zye?&6|Qd?
z&EUm)n9P`^J<|8eDYHTl)mbs{3eK*PA=|`bxgqytE`#>UR#?mCY~i+rr*LeC%xnG}
zq`pBEoZnb$6BvK_?%O)#fB8O-Wik`<(@&XQ%=cw}`sq#U>ha-C^VNxk4IL|Z3>MBq
zUmTO@cWqqKI-`fs$amAM*wKd-Mt+iqMh>_I*`4ErXVU-Vzm)CohO^K<Bm1sR#m+^~
zcqWbF4Zs8JuD8Z-i=R{-z)4Fs+`on6jL-EhbB^C}IyM1%X0#4MqB`@j3N%M(umk61
zjq4a&`K=x3RBQ4w$_&lk1kXp<%(QS?&U|jgEw18DY6xh&^TLZJaT=yk!O$0+jN&S{
z4X4I39qcT4EY3Jo?R-OX&!LU+L1RZjWSeq{4eTZJMMv?*ki?3#gZHSXuimBMn*voQ
z9505Yqkpi6sM1`6lemWR5N49MJcs_Clenk3JDN<oavi0!{s2`#s=ow~`$UH3q``sS
zeZ&xUS{pyo*3tS3>~XY0YwI-~pwZZ-A*Mz;VUw?T-zv#nuBQ&5&wqm7x$5oy5d~S!
zJ5?R|dlm{vXXG)B#md8rdYsp7<=()CWmU6v6YuB)wMP=oe0ffe@ATf$z37_GCHZ86
z(obw@^5RJcH67B-h^TX-o%~VKB7O+0P?u4pymWp#ETXYSCb21nwx|wgFwPMAbXsIr
zy~*Wl?k6stqCd1adItjXMsOuH_XBw((G15A-tnB@9A*f&K4_1rK&efy9>8CIqZ8I{
zh3`6ayB$f34te3g2k*jS9oRifM{XDK+WJ{{Je5{-p*O#gLml`y5oo%+9vR@kMrJz7
z*&HrqITF$3p*>WpOGo19TV;UpGLH*Ki;M$QTEb#F*`Vs9)G*UFD4*SEId<_NgDdb<
zxbkFM%GJr#%iO09N<&ZvAecM;$*v5~LlmyZ1{tgfWCzAW@LijMO2>T{p2gkXTK|-E
zY=^G+wi&GDCl0js`92>Ol&^LuoTe(4Y*Hq~(D^-3aHe1IsXz3gq0SCRsZJ7yX*g)z
z<mM*r90z7>#M#i#%m_!y0L<d{J&+lkAwq~O_N(>}(SHNUI*_5uV<EcS$5XjC16kRH
z-kw`)gc&y!K@9=WJqW`1RLD+1SHzm*Papn<2qN0&;yG@Qi>9l@b8y#~7x61IjXNMy
zQen>oY*@Q)s;DdfGb~{5qY$2cHtqw<%6HtatR3Oxx<>e^_-|ZIX}{CiG5VvpTbGOs
zG$1^fLKI)V0@0{#^Ol4*8i+KiRsmhy@#`#}0lT^RwfUxe8rIsAfNX?4OVI|KhH6d4
z%}u_f+eKno-L8`eO`8VSoQOISDNj6REf`=hOEbEX=XFMjVd;yvF(CI0P%<+B-+cY`
z<X!cL4rOl^pRO4WGzoY??Uq4DYpX2hMxBje(@`d8ER&fG$Sp7}k=X@Nnl~bh0*#K8
zB<H)xUZ-H{)rh4|Ot4xghu#mFy#c=~=ac^?Z&z$N)WF6t(SwUSjzY=@W$Q*=tojIv
z&w}!spPk?y@6MGYIUDtc*Uo}O9zSFkR--zj=l}R0|D*NCz^ad&68w-zV5Mw+OS@4n
z*n;c(81uWk`2@SyQ0d&Sb_Gq*alYUj=W5fXV^ter=_%kNcjdpGG<*3});Dz8X(bmP
z=v#`E7Ik4yoULp9J_|ibxX#-r-N7>sK;r|`S{5^J)(n6(?LS64dycs7^LWBP(T6(p
z?~Y5G&@pIhXx16jLE>m9&YM-XeZPB8pQD4?V*cLWX5B{qs%I&0U4XjPKDq;HIxgS!
zF>=PwW1R_W1a$r*cl8MWtPco3c_bm=rFGu4n|=+M-o4IVCIUJ<J~C@iUI?%bBS&aD
z_=%TWZ3xbA%xoT+eW>iRW+xr684<Ii9O#GZ%%Vk4?xMf<ef+ER)K_0&Mt67}IwPAn
z!tjwt@i%?yyr^L(^YD#SYvPi%v!L4aPuYuT2aj3kuD4$})ZQXgH}-LXbH2?JJD+bb
zYQNwA?;mSxyk)}X+-IiOAg=Z)<-qAISL#h&&@njazBiogGxR!ZkJ#I5Z`g!?wE7gr
zb=J|qX|~o3*>_)j`^8<>D48*_2Fto7Z}_pgZ}LW(_EH@Q{7MY3fkjTU`B~(-(%vLZ
zdwLnP^Qvpk6JjIxr3lw4n~SGdS5J2m|6HMpcaq2Fa<8HrJ|(=BCdbM(d3Tw)quF!3
z!2hhw&b!XfVD9q!+&ruF1`_%-uOZ}F=}$mT)KPo`I3Yfl2+cKwX^?3vRs0stHADB9
z$1Y?<iFaqAqpYEu-sONM&av~krlS?@I`=xp^h35|-a$6TrBm7qW;^X<(2+5)8Za8(
z=hfVYY}d>n=v8%60va_Yjc4f1g3UBWWd$9O&RpZZvKTkzHAIIz%Ay+%<<Z&B1aU3T
z*p1O+oHnn@DhIE#vO9lny0}-ok9g^1^epmv%)(H57OsDWM$tKjhSHc$rH)h#nlr*)
z)qs%os*_!?nAZ=o9f65f<0{4;?c`hsBJrneMUH;p8jg!3;cZ3lGCh+?`e!4#?!6Ji
zsDs(BCM(0K(23r3ZwoPlqYL};O0N!ooZZMLIA@Iu=o3bdrc-1{9b|V9pKO8nM0bnL
zeP>i6a}G{OEqpvaId!3D&8Q56t6Z5KT0w^k&Wmg`V&=(X4$dH2XOlK^Fr^-cI&V@a
zaF@q0^v&ZY`rP#)!Ce`<j}D5bAK>hD2FOpb2F_l0<TXO3yO-LHy4U6GP+}MPxla26
z_ku$b<?>p-b^<S44|(&49vDemr*mP8`0^E6QeyVcAY<(*`ql-<StOkiINr^MfYTYP
zIx^Eqs;AnLI1%W7=zVA;@8~Z4-Q+89Z$ne-H5kRSHS~eb*GX`WI=J6Q<HKd5I#XL2
zsFTpx$7+EifG23nfS8RG*VvrJqwkUv6+YzQ6p(8}PUSPaDs}TdPFa-P0#W)zw_j(w
zdG*C%2)M{Z-N#lAmN-q3MS$TOUvDjB%{6<2@4o(>b|TnXpj)?F_^bWI5y8_q>vUwC
zY%{Zf&TLE^=Qu{MZeRC}9{S_8d3?Nm%PEDMY{Op)wJtN`t(~gV))iPw-Xh`bjMGXd
zk69SZrk)?ipju2q@=j3t&mOkm2)5P#IRZ&eHJV^+@{>dx=ZB;#?IrV_Ff}lDJI_b8
z@S@Bvdz@GLF2Bz^{lJlns&o)E?q$YZAm^jZqI~#Uu{oV(LlH}J=x8UC26iZAVNQeD
z;58yagKoJrx`Su1ScgQkvzD(~y>KjC&7bRS6luOccy-L<p))?}V{pN%qZ0S_9Ybiw
zCPpn~+A%k0P@^aC@S*{FGecSNI(tsYtPY=P<~kusFYn<AeAhKnfoV2FD8GS=ukDCb
zc8P1}k}_b)2cI(|?;OHUaK|Vx0`u*L?KBy&kc@*)4$dy-%=BrWPY3Qprdv^Dqqd;V
zp)*jU{yq-ms)js*V`PGDoR82}SuJ@1R$755@|=*e0@%`@;XnX+)SBlqFl#`QkC8Uw
zDy)?KC0{q4#~a~4Wi}im#k*`dd1oC!%5Pp~l%C?@4H;xDwA0b~!Z|68?@xffps?uM
z87Ix-muOfrrC##a$AEZUg?kCcIzTJLoJK=KL+DWd>h3B6vc3r4I+Q-9$Qu;WU*}C5
zrQX}0BV}|lZ98BW_tG<ZO5buEP6&F`SK!k_y6$@+bx;D+Y!J#Wvc>I^O6~$x5eb7K
zFgeB3UhUDMy%kU3J;oW-$y(QAn*QRIdQ3^LY38i+Lq8oZKRU_P@#2T>6koCxXYjT?
z1c(^`vl;fPL15*adYtc^gwM=sD31nZzz2km*V)cBQ;dQ<5h*R`Zre;k^%UGvnBTyd
zF!<aNL-PeM8n%ttI<8~Zp<6FkI}$|Rcp@kH)pwos$Ozu+%rNFuo#s%`?{B`D8IsqJ
zua{0_mLf36@j(_!f*W5s45H}3)IpqqHJb^KnI&%?&5TQAZe}QD@75zjjv}VCkwV2*
z+JL=v9<wg%FMr8bcQdo~{a^q3#hbpxBiHCSdhmN;UzFtGd{@7c1C1VkTBF962=pK~
z{&V)Peq3iJB&Q>+^rng2_s)Vcf6A}~FsqqI0vi3Z@Z$Da+Ue%VmP$N_XLrAq7LAp!
zkIi<TDg%W`3XR_JfV)23Kj9F+urH-vcLP5+yUx$?`}%zTO*sSr6h#AvLkmwZa%T_Q
zBoe<#>2v6$dnUT$j5hjSe{$<g5Z10D@tl#4M&w`D(>U5K6h`kg(j|vF5Aq@e4Dchj
z-N}zpiXlk@7c{6b9MhQ8fGKzR+i|1;GrFF)K0M5p(55b>GZMEs_UQe6%7JG@FWsXc
zFlsQ1%}8lRlbwoIpF(J?Q<6(F4d+n$B9V)9U`EtNK6tA<;4%Wr;kyufhfbY3Jjx0k
zy@xNaQJep{%5*<D^bLi`JQ5oDv85?PW{#0T@Wr?#p|H@}BX_Msc>@C{Al0GVX*r`s
zK8~D@g#+=BO}o${VQ<;q4mmWF3%UK0X=2Zv+sQ_ro2P&pJR!$|P0!m21V6pVZ`UM6
z2Q8WwbY7iB%DFNLFQXy!uw0D3EEe%+aVvOx8hX|s<Qae9wk;-Ay{da*>y<Q)#W)un
ztaQ*`p^b#R_0_Ve;`(5lo<jK4589?opT)0wDhzVZLGadUxHfPwGem<PyPwHCE>v}Q
z@ADWVRFEUSybe4}XI3NS`UY(BwS3Y#YqaD6HvVSe;&hH2j?1e&x(NyQVdg4wv3QYu
zjr{R9x=#NA8t0}uK=;s_KQ*AkQa|?M+4a-c`E~ZGUNxH%wFTBDR7V4&JSRgPB6Zq&
z9*0>Qgj1(m?IbIC`q!lWGw_mZJ&eFX6n33<_N>KcRw-Z-=23}9Kvb`@U6tQBYU+yv
zsEui#oH>u~%sANG?CU(jmHuURR~>$WfGh&Mv#y~5f4hCVrq5vPvqh*o|D5OkIy*pf
zc7rk?BjwPgyzAbRc$^1tJbJ?~Cv`jPwO7g|J8f$Hqz=WygZ;z)L|2NNee$Q`!-vB!
zJ$dGYB(4?FGYkTW26)4co4Y^FFX8o-qL&;m5$)&^>(gC?w{PaW<oBZ3v$S16p2_Ez
zmZZX>xaz}q8+K?z3tVThQ)W_KpYMBjilBVLb1HX+=dmv#wcyU$o1KHvIF-e7j&06c
z#B76H6jXYt)X@&0zFo0kNG&w{Rlz|@3D=3+=U&4it{p*2Md29=jbJ-CeZ@HU8cg#B
znzIgdMz_JU?%SZ`U7hyi%g(2ng)j=*4g)7dC-m;^7~<`OY`8b+FwKP6F;)HDBtN=5
z*4Z%R6&Mw?p+{(g&w0~ky@s$QTpD8lj9wTgj)u7A$yujzI3c~ZD<wxK@=H3jGfR_4
zP#mumeazb!WJurYyvai*sW)Y|@g3PRF3&^jCsD1F@X2=^L?c_dOLt%m4Kgt5a2UCA
z?9zdSPFqYXD~Eg2>u)1BU+iT9`x-9en-L&N<lM7I4Ej7`!A~HBP$$#<08WRy3~6L|
zrb*_IcZ9iqim<?wJ-xQsRO*X~1iQ5MV(QMww0NORykqLzsMQpCMxE(Q1<nZiJF5mb
zq-P?UYkr`@gQFexf?a21U6+<vg;{9(gquw07kQ%rfj2i9QR63%Bn6wWmGJytCZ9Ki
z$OrAu<urWJg=i@z>cG?2t1R@!ua1WUG8_Db$2w-}ZXE790?DUC+Ee`;*F?pE8Zy$n
zR!1(v!5##59IOkAJ<D@s(V3Fy=#rauIy6lEw%z9;;^v!U0#=XEnOynHeXCkp0z5h0
zuQgC)fHr#8<kpiS^XDpJ=nTk1zM$(3H9V5LnIeuSdxQoHK-IT(PwvfZ*vJfQGh#P+
zECS!IW2ovd$0>;Ps`umI%=$NFt%KvpSPP~d?*KpkFD_;sibvO-^=;inhLO2JR&gXR
z2m3aUX4tkVJLPMWmb|?o)(<Ue100q70no>?x%O!Uy!TsJmDA|No+Z|=K3fyQpno##
zc#{7!2jclVB49+$;rZP;Wv)L;|CEdy7s%?NmH$+#S5YG8eS8|rr&AAfE*74akK#0}
z3|lj%Ip%JrGIC13Q=U3D(x9tbq)8_DiNR>z<^5>vHEc`2?pK4E>E5by6?$-v=>hi<
zZ7EY@SUk|7DgU7bs(I%4)BsHnPKI?#BWvR=p^>dA=D-B5okg#5BUss;bn95yI9ny}
z4SZmZ)lHf;APmz?Q8V)B(_r@_B4@ObfppFc1H<^hDKO^Zc1J1=95V&z8>5*~zL^6s
z0@EWQuRq2bRJ@RMdOH_*Gz~j=1{WjPssGUSn6!@8gP#L)$w&r|y#S1=z<Hlnr}1dY
zxUyNax~}68I(9rRcJ~$RHEQRyA+W6Wr*tm9I^w#V5i0)1_@=Y>mpq&{Mw$ciCOf6R
zew9}tefE7#LC>s2=*<ks%&O?rq7NUeTL`-1Su*N<;&QIxt*RuiJUJNMeNAut^*?FC
zq(w%NB_m7D_<V@L-u$-XS?#DeKmn#W#$Xq2hoaO%XX^1-<QWye%et1Xv$&>9@=x%#
zLLXO!c8;N)V(E=eyt7VeM|H&MQXi4T!HnEEa5IyUdQY8kSU7xOw>&#iK`lhjP9^-H
zeRpU&s4cI7Jvt9eX~t`KQP0({<hNBkz3zg@=z56GzekT8qvian74ld-S@ZO?mrvm7
z@B;TH^!w&c`p=|C_W}<+xodg%W{%B#!13yp^(xki<WW}2kXMQ?8b=nb=fEBhNWGYq
zBdgF<$dOlQ;|!y>vy^Bp{nnXzOmde6sXw*@M~k&SW@*SpAMuda`;WC4Pz8<|8?e}h
zKITf^pMLxyv~8W5G<92@Pg||?opg4_K*a0?**fTsS*$vy+Ci(3#)XdHRHpq^a0=T2
zq=WdX9&^A_&?4yC1G3~qCEWo=<wxXUZcKa5!RKq{7CjHgq>!U^BW8!_MJe0&G6Ceg
zj#E<O&i8+&YZbS{NNC@^I6p~u5@_Au3a9kHm9uFp?@L2*+@D~bQ7Sv-&#|X#<#IG%
zGbSzU;&|WAo%o3vd8`~!tc`l&HP&gcQQ?$%$_=YD@U{YfU`-Nx8N~%%qmd4JeI1IM
ze@>Y6aWq`Zq2!z;GLFsMlu0c{Rw}`ibDBm-VsIMO#@6SfDJ!i=%22V@K%l7`U4_Q(
zN*IH4kr{(@7l~&$C%)e^$tZ>)(V1&>9_>iiu`<S>8lmY>8H+adq)&%#nl`)2CSClk
zhMqEJ29z^;6yC12EcY4hKgKBd#)K5B?fH5PHmNGBqhO3@@gD<adNuLEV;k|;w`0)T
zDKH#09y7g5p0{@yvEkDULe7kjptkz#JBFr@63o4a=G>I;-MLL(1#s{@?7Hisxg<SC
z9wE64g-r?HXU`Iyn~@c#jf{TDUI_L>uz2oO_h>N%oz0=L#z5ZdYaQ-=zqiv)%m6*G
zX_UXMnIiE(pO$(w9Rb(z0%${oQJorxm0Y)ek)OlNm3j`W05LMup|lQGJdWY-0cY`^
z;*thG9qep?*Im(-H+t~>u)a*XnqI#mt&UnI&f)vKr81knt<EC}Xwg|+rOv!f;|&iT
zTOH8o!{}smy7cFqe8s)mUWr}p1O2qN`c$6CBz2-*a-1C!z4_(UIDKv3<d;9~?N!z%
zc*I~KD!qS|IvX9KN8^OFGtPC->LseC9nh~!yKhF&-UL%Sem!z{G=oH_iKE-XUQjAt
zo!Plos6HgC_9=fVxb3YATSKGAY+)^I1pf4>$!o?Svh7TdQabm?e1ZGtp*wQbft$_X
z?7e&cQ)U(4$AS1UGZMaDY(en^^f)-0>hK|TlTlUqHG9KR)MnL}$A*8kPRk!TC%<;0
z-TL+mzGg6C@Q`@2?%_JH^T=&pL#JPnUB3U7LM;PFby=k*5MT8#0uL*>PJU$)4<@;q
znenI={ha*DlAbMPk+%hMdWf>BA&wV$?f@sLFVQ0B6OUtd?wm`cTz~RVIYht}cEnMo
zh!iXSO8#R3si;qe--!QIev*fKryb`F_$92D_dDM|a(+?&+YoM$Ps;SfFEXAEMquS$
z7v+m8y4OMGth6nD6*N6LVda!H5e!g^_;I~pOfCR+oya`&Xh?u}{{!g{84;ZtMN%qX
zMz~Xpw}qNk#LNMi`?R8t>Kc>C6`Ce}9Y%Tq()aG;;<PhnY>n_4Rb0udV{w1L&b?Pq
z9o9ZP#yB%lC0wVRbcP@Wx)>2%QfnxJGaPjEb(QrG$&&o(WMb?&2(zwZcvP-VNiJ~b
z&%LeDJ(@Tivs5x&gO{;pY`i_e0oV>mKs9t!$w&uhzeXHheNlL164|eSU!k(>#ppFd
zTbYDjAK$ZH0M1|Qa0G7Fe~_VL2aedc-+Y^Ia(tJ?bKj+mMa5Trj_&+6)6hDDpYs?j
zU29&cPNx{!Rf!oU8+DaG0K@0zNqZhyr-KhYSADZ19c|7=bmb|MYrUwCzTp-5q!S!a
zI+uCW2BuB}{2=crpA*_gL|ObforJRFYXr<GRkl_pGZ=JkeI(E0n!p;34kjaV4Y8dB
zL$AIP7o0uHQ}{}py)gQ>__Pq4j&NKyF1q!WGaDMn3ZL)ecsfs~9S0jf^d)(!iy=ds
zAn)zusuz6uDCL9oAkkxKT`U<r{q~#hIves;KIq7KF^b2O*G!3N_L3MqagRQ_;4mAJ
zRTcv)j&_~a&|Myq8Lh&8>!Ow27*&vx;Y-|;4Gry~wa)p{ukECZ#|Mz5%jRK`qIQgD
zk2=JE^X;49=e*6#g>_0kCGb8AQ8`ufcr$WMJ;(=lafE*K-f6qxksPGcJ%bak$Qq!)
zrEU?z=`hRj)z_SwJSNFw7v=2b!LegAr@fn{Mb8XyE&|IMDl#;K7GhEdv+&u@!h!xp
z_h+Jiu{m|N@I()zXY*1(<P<m_^DHR`-rzy!$RPaXZw4hQZ|cM>D_IKk<}$JzxsE*O
z>r=~aHb2lH?Q_D;H>4@?boDvqKQB)yT7Z`c_+&xM$+ogS&$CGF>3sb>*YCSqh_+(?
zy~0~s&9`ry_w^?U{|%_8a=ZAmoTo_k>i2*?!+BZm3GBYw=Tj+B=R=apSz$VKH{zx<
zbXTLDf=5RsQWl<0{u#mM4=h(;?(@D@7mdqrNjS)xvTm8V?dD8wQ<gi9hVkCh0CNTI
zj2arfex7ULjJlLUW6m+K*mYfFmPTmmwV!{NfbIpidBi2XjX8v%rJYL3PnWcLTJVxT
zgYeK50}|s<qaUN-+5eRhRz`~3fH8*KPzOb2GG@NO^Ao;1lM+)<=V5S?qrEH8X)4=H
z&UHE_BR<DdK;d+F>S>Ht%7rn!?Lh8ANG=V|6DDQLNsU&X>$^4Gsk0eJW<=GFBOL<A
z#LQ1;DH0E-!=g$~gn;LE^bsnaIs;Hy4R5eZ-)(rN$Tzh^FPdhRzxWwJ+t7<M+s<P#
zGmb#;z@;8J0Fi(3I<Lb#hj@I<+dGqw!${8Z&oT5rQG=tYpPuD|r!}C=GVF;@0^%mU
zysI;vdYrzbS1)qbfrilOz>=TSq0TFhf*F|T(r|$7G*9laT|L&hILr*u_2@rmy!)7>
zLT0P*jN;QFaY%HgyS_U@#{u{an|HjC%XUKO_nH-mX6e^1Is>B6s(_YH!n2Sxq?3V*
zI+N&jXs4mCv__xm2oG)B7L}~D3*74T)!Cle7zfSFK^~99xfSqX#>bmx?;=~Bzm_m<
z`>B4+N(uMbvm}K|4=3k+-s-~hA}~&21Ri1^UUO`2;|R=b5<Rzbu^GV5Y=#G`hg&Q3
zC@?Ow_Cwp);?Zu=5hD?vRis7JbS^xEhCYU=4$&J<VdBUv0P38FzG(|ejlJZs37sXb
zoRxLJo4KlGm(0po^JJsI{elM=j_VQ?CaX*$F{gngu8J2zJgX8$pa)MSOnlH4*fgAg
zRl|a^j{J$)(M|h3Kln*SwtTPuPT}T(al%&cp(L<E6N2483yZ)u5ZC#aXD>i{mUrE3
z*e}z_6M&eRp0^z6z=WYeMZq&2tdzmq%BB!C3V7GZr^5GC+NK{nWE~Y7c?SpM)Yk_S
zj}j%TkU2jEGz>vN$TvUoAA~mh=2JeQDDyiCwPPm@kbyUi9G%s(K&7)|oYJtYt;i^d
zvyZ3O`M?uCO-UAFWq6R(i_*A=oa>F#IWZq*bmfu^E3v@odVv;5V-4^O-lN}{BEI>0
z$(GTKp}b7PY`#)Zn3o=L<dgLCK`}-v?IE)cHEgjVzy50KR0l_#xMf6UW3+CQoUgcE
z<Z+2d2joDX80>Fv@(KEkN{s&Byt&CS9jngur=PuknvrQ<$Fvru-*yP(pMM@5umO@f
zF>|6_`kE~IA3tQ5L&zV0=}7G7H`yquoly9R0mfB3ipb=bUv2yC`Em^0{bfg4oz9ju
zu<x_snY>?r{aw~yJj4Kg-O;*fW;|G<p^6yK?HF_v$ylYFI^8yeepaFite~#(>f8<|
zd_;rW5f)Ai=!X-mmRx!7(U+0k-QAln{`X&ht{&}9;;U@XgzlRd>C{&8(B;TIYenoU
z(r*-n7#$~?;6xVqrldQw5Z-m_>c^<lf<H9fG33Eb7mV@+^gHmVb7@=8HAkf(NR%nR
z(WN~yZn^N+`rrsehb%hpg3ZW<?npx<nD&Ga=$#S2nX`5*+E8R?rlox3hn|sXx53Yz
zY~!efo{ZAfg!@28eDBSG=v37}YHn}_hZ&GMH-Yc5Dd+A%-h6}(?(&VO&{tX_GmkVb
z@>zRdFD}oKAx9QJX1ngYP}7>1;7&QZ-PxeXPKQ4GjA&mv%`!jj?`pPp_Za5*p3xF<
z7{baIK0h@<{pCt;Ex2|{q<ZuEDmlpG-~e7`yHa~(sKa;n)*pq2@OXbeGiRM~OTFCY
z1D2a_I61;s+1JBIFkij;vNUTO|M{Q)8CcQL)Kk$1QFoC}o%+TV-G6xawNAvteO}3p
zY|W0)vtQq(-G}#|ew=+q{_WrYpS-^NO+U)$Z8SQLp0B_DDm-E@?Iw1qYxE-5nt=Rp
zrp!q77522*$2fBDB7YfPzn&S(Sz|{!9$B$n;W>YJu<qsFnxi@s(S<Oq%YsiZ`)kVq
zuSU~xvWzt;Ewc2r|KzithCMKg!&X@Enu@3lp3Q`W^kJxq57Qv(ZxNjCWsQ=wh?-cz
z)#TbS|D1Ch>=T;L2O46~gOy#phVcaid3JiQb5ljj4WseT2;*-|MIBPo$RLKV7cmg)
zJo0DYq={#IqEver-v$<V$V0Rlg9NL_J(FvKQ)36sx;#$L)7%Y=DYdVU>_@5?sk{7B
zj>ILQJg{*ZL^5go;@ncgXkeWX@eam92T0~MZf8=?z!&VCo`v2A#>9^Vd$Zzd%$40T
zD&aB2Y52;FewD)BH+f;0#ka|&Y@g+{L>nD3TL*%lJrc!l?LdCo!E|OM9j+<U7;~f1
z9q(B+N5V5?-_GgiK621tbr*!^cc!Wt`f<=qgVGPL&~-7RPC{<a_T|q#WBz_kzlX!=
zfMbXlB=bD;`k-lJ1Q%RNs<fPR&E-{nG4LE5nNm+Wq8ROV#^LdhU1+ZJ5%@=3>z3vw
zxK1eeUM;!jngr>A&M26xH)I<?g|)KFC9>!4MG&XqZPthE^)>TofCIv`X)l9w^entv
zcfoE21Q(?=^*#~dMIDZlFrA=!y6R&W&cis)qpx&j=_**E6MNNxfS+2+S^GwF$snH`
zaZW})oN{`T+$nE?vA=#pC2fShNGFeq&Wuf<p}bIwa2h{+*hu!ICP0~EI`VOBIE$&a
zB*7)0_L$tsSe_XYg)C@Up|phuXLe=qq?4j=@(N(O^b-*9;}2}d!D}&+ld>y&U{{xN
z-U}jU_KpLB9taYme?63o{Q720;|U)=2?MkFj~~<s2cS|xPw+f#RyjNtP-JA2vDvyD
z4SlSVxXUb>*19gUCyMqk>$6gak!$(1=X&`4`s;idF8SEG_=n+R4XRMl*&9w&+HbVp
zQzCDp)hUc_X1*~``O<zeDBy(roWX@Q`E?wS`+JY)rfrzL^yZ1PfET(_Sx1jTP>-6s
z#H&*E-72C4fkV!tf69ZBGQp)?!_67PI~kDA9=zi*#T)QVcX=jeLF?ED=|%U3F>8wA
zWRD;?JMamdSr3CM$Hu6DUi|C+918z?NyHcdDR(gWF~5`E-)2mms*e;o3W4yuLp^y=
zstt+{yu!^TrD<pbc)CBLlSf0@fgJ9$kAg<G8=i)d=&a7B#G^pPdv3%5AkW_pW89jj
zd4H<_c@Ev%$e-cPj6oVlm>B-jS!y(ls>Yk^Ho7aLhT)AedsZnwI$_?m;oNINIbe$1
zIR<<04kuxhed;g2MU(Sc2tm?MfhX=m90@cU)xbUtk0B)QOkbu!c7z%p_AZ5BX>@zR
z<aLa~Ilw(eE{H;_=cUIpQ&G6WXnu^gb>N(kcJi48;h226J{AAVTaR5Hzj&Qz!0^7!
zvuB299E6O*GQwyF92tUV)(mKI^0$+TK`^^wGbL$8znqn~*VzF!2Jj{$S2I&OTXc5|
z@?^C5Eg%Y~Fwq#f+~hfRv~C?aB}jKPD73?>D-s$*Ii2skYP;f+r$!CEfsY2A-41)U
z#d-kOMvjb_Po~?H$xMj#F#9?5+q^}CznL*$@MpoW83LMax*3Htb*(mrCiuXmql_k8
zM>}nsqR-<rkn^F20oK_Mis*X%^(yVvY&1vKngVUW=iJdQZ&I^W`h4K+%Y1p&V+)^s
zhg;n>N8hf{d1S2&_@S}&x1RaGdS$D1@tmVLp>cMN<T-X-KpLFlIY!zcx;e368JRxV
z@i#Yg8s0=lBFSH%+Wg3;j+OfHhc@W~=Zfxhsy~q#a8F%xy0Z=?^1p(+%_QkTP`!*y
zI|~#2oDOh0y}~SPGissDqee0feSHNtX=Lmd%w9J?y4M9J7amb~@=c&RI$K5-TW03w
zoz3miMr}rD9R>c-Z{1NBKBtYgeP(8Xo{l{;x*uDI-uIcUdQ7`jf7XY5|3f<SOuOp9
z$=Q$jrp{XBm0trcb;G7q*P_3bS%{VY%{y&*Y|ibw+dA7EZuGHH=uvL!bImkl#>pb>
z)^&iW*S)K*GS-H$31(E_s*WW7*SE9Jk@{r^-e%p9K}>BlemVAd2yI}KRkAj&b|}fV
zidTpAa&xjq3a-8upRAqX+<GKm!f2qYAZaMjNt}p(f!~nRhVTc-Qbe1eP&+(b?U+et
zm<X6)UFVVRKNtcC;RQR6!U0g+3S0EBI_7tZ9lwDSJ!v%^&c}kGA>)tov=O@&_k!<@
z@AsckR3=u@p>gQ_7@9_5L6lZh65S`8DcMdB9)^2KJo(xIr?CkiQnrnxcfqpCcPcQ#
z0;}OkaEHt~TJK$~Qjqt{xQNI!q*HwVRdi_U4vfKXaFBjt{Um#Y*Xd<K4?IZ#wn1rt
zF&_AntU+GoIjiTM&D9a>G*kv%dS?wo?=<)rg=A!~JMxclrF3Ld!xlqiG<=muOGflM
zI{P>HCWVex2hI4@2<fCkAIC*L#$XYf20x1k>yQQ^oKJiN<<=L@Q98JJ?1<Oxe3{!O
zr&*A;YBSjKmpbuPWSaL;h+G;@_2a&Jt^UNXG>_istl1!8Hy3ioA4B&9j^R&72o^X!
zN3SA_Iv?p!Ysht=Db%U`$R1x~G_^ro3@6hu%pDo-^&r)+z@Lt0MsLNpaF)B!5rIbr
zi-<-t_=&J4C-*KlpSj3i>NtlP4^D`K<BKLUO#96P9bS((RvcgQsXX%bk#p9y7FGx>
zzZ^9k^KG0C`1+T7>5m>8<_r%RwPXyU<PHS#=!D(4h*Ncu86VrZ*8wZ>!vuK#_X(?&
zU=7UBgav^<%(@_Tmy&cO_4tt4Hjb{67V>WJ;Oxbab!bA%z%RJtr}dUJM6{i<A-Qeq
z1T1fPiNKdUc7rX!`2|rsZ7Xi`NKEX(=-1eJGU(f1SpZ8X%>vB8DGO1vAk+6(`u0`i
z=#dS57zfm&Ubus&gr{wKgqQD8-Fge?7kN2v<M^UQK7$dj?Rw;??Olcb>e#Y%gka!$
z?-lFNjvh0cbeR)k;6o;KpTlTj>ekU{Y#og056EG{@S7SkJCU+rwQiHFo!QSs9<&x;
zoR!YJ(hq<1XP68wmw^*~k-GN-v4EA1bDRjpHuLOE=3S>O{!&TRzg1V)HSCK>GEDB`
zf#3*`o$rr?&%X$S#lYv6**Q@XQx3u(3PwCadH5|L_}(;jIx-lS9_t+bPzbJ=_bFPu
z!SJLLiB}6l&?r#sMo}exYPmAovTtS9x-B5;#b4UCJTbMCzlxNG7tjO!M1A1G`spC|
zq=LJz<=v&w*(?jrQqPjxe2Ks5cER&0?RLOnY7DQ@`WSV26@sl!i~C`|9dGG|E5YxU
zkGrF>?R4iJd3E)T(Y;16Q6DnZNf5TW04Ef*yl*p5bcTmP@bM>WReYa=EZ=butjLdn
zZ>7bcb}BXUA9T}6g31ZB$HA{zV^QZN@-MD*ij0JhgP~c&)5sYSICSJXU$Yop{}H3m
zjfP^xJd5?%rt2KyjEwW<dmHmXwAWwdnRfWP%}j}W;BI(iR)-=&tWgMkGjlQYpbh?c
zEjFD68Dyl%$kESvRnI;Mo@H|oBF_*Helw&z80yVE9Z+9rXuw<ebr6=kj$WsX>(+11
z_;+3TBn?b(=sG7Lgtq?Bk!otE1B=eydIzI5qpO?D)>yZ}P<GFj$kH=y#&nMwOGC<b
zlpnaQgWSUbRrL_Pi$?j)P~GM^|Lh0ikyi93I-!H|_0fwsg#roqo4#+j!;^Io4Z#tx
zUu2d&h&ZM?=5#97Lxf-L#2Y<mQrGl}TsftE9v^wwwj8bE5nLxQ@>dSL`{+F%)ChM*
zoFgaG`(1RIj^TaE(bu%d!nn>O1Uh^5CXQxw;2}<nPIJrXGJBQGEH1+3#hZ?b{M;wD
zQLFAl+BoPTHIBIshXX=BK6uv=b7;Uq?P|()Rw6V!_+o6%AM*I3s#M^lM_wHNC_u{j
zu}!<{j^-2dc(HJ0Mg#cE&WI+xvogUExWZN9wHsThckm{@dCd~(<U?mh|7HTd$}E+2
z7d}AwD*MsAO1sp#(zPOqoZ#w4T;Y!$;Y?V6b|1aAp2#4=TPUGC_#!iWRKJIwy=J1V
zB(S38t>ciGpmYB!HpH5-H`UMYzyE7yO|r-*p|dlY@u(w0AA{HRK36{Zscvy%qHn#9
zbAg8KM2UB=tw<*a*KdBwTQ!jbIO<J%w7asr(bAkPG1B=|rSoGZmyJ*TgqG@m<c04p
zOf64DH~Q&JKb~en%2wf3!Pf1+%!xk!@u09H((L&RX~#LQT*+E|iJI7R{@;1MQGa%^
z!MO2izWe$Bfj&2g%rWX51&@M=>a)A#C=YTz1EA!yhy(0gNZ(;!%l|~lxpZ`>fkFX`
z+C74#d}aXlJ;Qc59LcXiE62QP%{6c-)5VB|I<zVeO*_GT?R+|>eA<Dj)kuU9PI6%8
zJ_s&r{k4;ELO2}3lyuK)OLgH%M@M0oE{Nv<uSaiJq&zFksC^o#RE>4drjSzxJdJBU
zIFsI)kc>cU0Am396<Ocm&<RNo{oO&9Tmmzv9&-cns-v?k5?7}ooe@Ld9C574#DO0?
zl7FEhF8|F~_~fz<IdaV??rlI8#-RfVryx!K9e`~hxEEp9pJ=*K;pSRSY1Xdj5X~$g
zZr8toQ|YV0%xuppyW!R7RrU<3nUvvS!w<0M;UqgawvK|wiDQ7#m)jK*AG#{z#E3S7
zLw-cLxo_J5dzAv9G_HGM8rnEr(UrDeo$%b#4V||&NAy@{p-guPO{@73C#dD3EA(VW
zoHCr?rhxdqsl#a~?t`M-xgHwNv5UmxG`Wu3-@d?8QX}!`iw-rN6&gy%SSpLM!RsJT
zPRi9hoA*^;1(ck;oTQUMfk%$ol9|N#OqD$~;cr}0QuqV2GsZ`yiHCKCz+3(O)LoK9
zjqXlfGgUVktXyR+$Sg$Go<h4e1Xk-4yn#FA2L?DtUAIiim8QWjZ*rz2Nx7Pz+N<0|
zVu?r{M@YWryd4QMAmDPWM;~TCAI?XegUWJXb@nTHVJH@GBSTIf`LIK1M=Sm5quRv5
zRlBtGaHtFYVe>L|@oeG(nJEi*<<L4JA7ScQJFp%^eTX3~Z3jSqoj<MsU?8}WPKX}V
z;(CDj4~7cj55lOT{-LtLdNe*<VSCOSoJtioeWQpG6w8t{-(L7bp`uIW{EgyMCD7z*
z(eZSxGl5wPJ5vB~{(;UpDloK1c_iE~en*j0^o+kn^Mkb$(2aXz*)llC1H*GwlA*31
z>ETCAjmLtK0dHJjP-nG&zRZ|W<?iIAX*lkkg*`*ioBV9&V>+YYL@`_`vt@)Dk-g8Q
z73%7{qD-Pum<DpQ_l7Wyn?wsz595~xX(m92cb$czc)xXT`|_k2jT&KF^{-qvP0HA2
zw-qbv(`*N4XwX4bwkj_|Y9mLddP$v?w%nFK3=pg;X=QR;d?j~ABsUr5G8XTwV3Q~G
zgZJj=?17Q<D<cubm6cvlu&bZU5Ol_;Ntrdt=3ZHE&=DvK!_$m{baqDVc#A`nH*zAm
z0;_yWpb<1<6(o_T5#7BhIDAXbz1az0X5IBw2O<E-8?OR7%Xd5Vi&+zq_^!`L8n}!c
zBW1LHozdt~*NNZtOo(Pe9`cruHwBEq1nCU4k(`;38QF~^vh_?J0WCekDo2K+QC(fv
z;<^oXMzz!N&@a#Tz1@^LPC8kF)>jk@b?Tsz#nT2nvP?!Uq5P8A8*SHT)qRL<I+d9z
ztMjRi<>-#6>5Qg*(3^39bTT>=Ga)(+>13`<os*QII~-iR(cQkCkn}EoL`l|{p#=Tb
zU5tXN;?yIR>OoG3MmOrUA)St664coaTdtX48wVPvQG9QTpzV@=fmK_wa0TC<f$J;~
zuqiiSqkBo`(8E$4fIm0=3`62Bvk)=>6BV6Ip=FkWKk^SO`rWd?#vle7=<b^0HC=Cn
z(C^xT=+j^R+rQ<Fl&@>cN@~(ZU)b5vMdj~&#4Y@pN#rEFi>*-?W;q64@~-tVBfAg;
zmcpM1s45=vWd~w6e$JO{=`#6wJY-K2`fSaZdozk27dh0iS(tCX{kF85>GTMujz;7N
zEW5VXTixZYF|dEhnh-kJN2I~4zB*t^{)Whv?7LYU)0Sf|il>L_^DZ-m?B%P-1FhAY
z6t))>-hB>$d}|vLqb9NwzzTA&5=Y!l`!7FliuU>U611hL92l+U1!P}4KM%#0pJ$!;
z^DKY=F6BQT<4bVUa9+mF{Yi;icY~)i!?@x;D*G{;$+9E&c~ksU8YNN-KNpiTzsr!&
zvBtTG@{OSwrQP?5oOod6U-w4m8r=?!Z=Pmi@|r}+cbRr%@SO6wPPb6jOOGGULTrZp
z7~s$_L~AJ4i(KaW7VSvTQfSI7{s+jy8G09fD1}Y3h7`!pg%dCv;?5_-na{!i#|8YM
zZOaN3g8avZW4YJaKx~Wqj%wm$mF(V+qW~|Mg)4FXOMg3ee39zvuvW%I4GqdmM-JM@
zXe1Vml}VCi`9vsx44uM}i?U!afGwL?bZrN&{DLFXDQsNMM_t1&4;m6UxQyl)=X5`r
z?Sik*^wSAQ@9_e53!i?9Blo3`ew^z8&fKG+6vDmm;{?Crdp5c(F5TOr^}y3y0|TxC
zTjiN2a9W3>O@&w3E6+JSW!r($YYuQa=gzq4q%&ah*R@BDANkY)4tz6(LdR9%>%m3n
zZW&N>EwS~O`~K(Di&g1Uv*M;6rmoyS<!X^5X19;v&i?~FOPcjq@Qzb3e2g6F_R!Qi
z0XO%cmv3bRgAH|@`k1$6&^<JuQE=btT16mko&|4mDWjnB;IR`KB-L@7#|o1V)$-)*
zvh5!E;R9c2S#>z{PO1LNzmFrjK#n`GCljJ`iKP5#JKDWRM0G?H_cnt(k7~C)Bd2jR
z)5)(t?LxdWC+^URIML|yTE1YjxITi(K&MVf?l~W`er@Y?vzDH@3YN6B5$@QXcuR3M
zN^$kEB2H%VkLK;H!&?~~NY{rknEN0J<GXY#h0w+RSB6w_X?R&_pF<P+<$L0R4hho~
z=34SgK8f>X$Fp*O-U-U!s=-O)KByPBTE@_}`FjqwdD0qoxy{E%bS6RCz%0rDN?a6A
z&7VgpFl=~pjhG$xB%___$U7_W+F{3r=%NZqV=XH-%Sj>E%AnC%Bk5y+jvA#8J^5{;
zk=_<x9`G@aLW&OH0j_^+SqRE!<0p@MxTKkZ*`21SOV2wqMQg*P%7{FZ`7+Or>*cHX
z#NFmw5?$1p28)J$bP6%~=TSx)I(`9?$LpRk5MHfiggX6OGKL;Bsrxi{KMg6>ywyP&
zz)C%t`tACV(9f8>?OK8<)Wx-4--{0QH9cj_faLx$oBx1qZGjmYi+?L01mR^FB5~dI
zGb6)ZhY$RE4R_1dx`FOin|(v5&O`QP0NmRql`%Qz@CK3e>x)^#K)*_9=t^{X>O$OH
z)=1{;u|}j(V;UfpZz3YQ%<v?T_dXMgSqPnf*D_dJLw|DL_@tx5b8`Em;8`bZB|j){
zGD-y3Hrmyv<P{iWfk7TV)}m06$s>-lQ9SEj5*T}?#ySpg>~-;It%EPClD}7{lg}ug
zGFk)0(anzdjYn6@rF`|19%e@1GLCSkz5{b+AEN(Jj?%$N%-@xD4q0qs13h`{_==-$
zRG0hvI9wj(*y`UNPdb9mfJ6t16YLNfW1l@MLL<2050II$O#It)Jjv(HDg58!BW28X
zb;dAheI7?IG2=++Bay(_PFkI?;2DirnVQ+i8t?RH>cO>YhnKA~8>ACAGgO?rI6Ufu
zq6&#G{lT(7YMGS>A_Hsbe^>9~aqLpt_LQDq4q9Z^5xi`nnXEjLsD7mGy3QzXjHPca
zPHk6EP6V1JOd0!%c(s?a0G1#}X-9t!EPU_-2Ig}q&pFg`W)CAN7us?Df2Gr@XLW7O
z$@|Rmu@`^2`|gX!uj!!K%;+`w*)28ujJRf3%y?)Yzhuv#ry0U|D~G(TW#fGO@Wb~>
zmsguaTkt!hlfgo0c}{%VqxDjDTfNVE9L^EBX@M6R*!4{&eA4rj(SFq*b~+G>jR6|$
zr<su4|LX?S3i~{Fpa1)*a_6c)U%!A_LA{9kJ+~*+E@9eH`#SnRk-Zq;Rm5Dzh8<y-
zo6NbR7*CN$zGoLZ{co^1R86U!?Obf=oRrZ)*BVuSTX0fypNo<e<pN`J4(vbts}cNG
zSa@`VoP*M|iAfpF$gK`x0=jh&jZNBWBx$USZ>>Zc%FZ)W-}0!5b(Sv;?m2g%ag*74
z95XDJ*AjP<Mz=d$1+d0&cVsG8)3=RM%IL~-^;|YA&A=#f^ViXreCt{Tz|j#@&Ny{E
z-A}r=5|owJursbR6iPDq9D*^T!I*1@ID0ni@=qi#ekI^1<%j3F-uXdixa0|Nw+W!^
z>qZ<uQzx5_sYgY-*e~VDdK>{A!Ho747N=}!K+nc;5PYDJ`RP!0m6t=u*!%qI<k8tt
zu=^N!@cY(7<LO)SFFlcl*P&&g40Mp6hf)aFqP2K8U5VFm3O6}}S-9fl3GPpug^_7t
z(a$(EGlI`G+VU4|P$7j)n@V3#(rK44@w<lHwYo&~X5H`W)0wWR=Owya(%ryLJowhG
zxNlF@>jle6`@@*vl=Pk(p^WrI=Z{9_;nUo6s!%RVY1AHqXXyi)$ZtC!&S=1^c8Rut
zOkC*Z2u9!03e?yH1f=5*jP8FD=DP1m%1*nKF0{ltH)8#isB^?|EoF*fEVP6+j+}6t
zP9L8fQX8Dzd9)oFiM!?~vu7D(aN;|o5k4%w_h|r!w*S!mZ{lcXGcxT3>~XS_9^{A3
ztWG4tf8>U5wABV=k?yvkNg(Eia6eU~K6Bko+H(2e(Uo`A$46`4YXS%G06=)|6)pg$
z#)$pHVU+8S!dWyOEdD++Hu;fJgNhQ4dm6Xmp!jF-3*rvAt{1;xvjN3{j2ye!cT|xG
zX`Z<<@~syDN*F~?{wa`hQ<J14%VJOFPliF4-~P{8#>V?;!o<&F$V8#RX)9iv=XB!i
z!ZgHnzhE%DMgwT)G2?^6J{jS`u^ar3-)DJRnMor603?-3L_t)A&Nw8iMNPEyVvR07
zT5}+6(Qad)F<2Gy7@p8^Yt6u-2W&LzTwkRDTf;IVb<>HnSkr98!+ln{-`saJF$)xp
zT<6V&cFf?ULSAPlR5K;NzMYu?d4!8JbcXf)uCT_Yu3A}B&#lk=JobJ3hm3A!g}YHc
z!w}(R<T#7Y@Mz2HL}qPAc1ldJg5R^~U$Q5`WoAq6GTmMy8|`3J_DqAkH6CP;gPHA`
ziH!W<G*UEri2!0SM#qwkRP!8nXz*&{eLhZRq-0N&*<CavMKeO-@i@vLCVxASo5Pz3
zl=miUan!phW<gp2esT900~vf|Y6NKH-hDuFMl(8KGSa(?F{M-Q)>@G2%;q=nJik`&
zUBDS!)_UPl5D%vgrXM6Egw-K2*v*r4`#_kwW=JosicBb}W<pZeMxAuS7Uli+3|(GG
zjG(>2aw9!XF&&$|W#$cn3VXq*QypAX$qtuf+m987F5h`kGY=_`_kP{Aon`1LI_)u?
z0|(KrvANcvI6tBjkNqfKGIu>g?yaw>45rC*>;SKW7ue(rCp5e6q|+pu`OS=RB6QM-
zj{RyKR=zN@3FL*IIae&Lmq3BT{wLv|9`2c^5nbe!PSxHl4M8)TG3GAU0hu=h<8%b?
z{?vd8Hf61w2}#X_R!&N3h@h&s;nUkk_QX-I^hf)kTR;3@XW+~xM9$j@Z`rWttj!|;
zpUU+EcG1DNSz9q{LS}YgI_{-wf$b03V7Q>C6GGo1-zBT7*Ui88Nnuyty|+%tzCQUb
zOmyMP*a~Yw$Y{?vXfJ(jcMZ0{f1iC&(CH9npC7Z4L+9JKzqTIO3H9*)K0EX-*|{K3
zIITsQnaRjDu<r8~6h6SM?NdhPgx#)u1h63=(W@PbOxsQ(!&wnH$$yexDj?){F6O#%
zr_0fj$aemA^6xnVPr(MwQbKWoy-`C`l`OtdPW3kpeVh*Yg^8OpApMgTg=u<Xl9~hb
zbv!@Gb_%0d)0F0VE6-RnSpmr11{oK*jcPT-o!(6Ar-mer-PCy(@I}#-8Jbg~*F)*>
zgp-8L=!UbRs?SNQ`B-o!`xuOlY7#$F!<o9ypUz~YvLvD000K1)COOjL$M8;0bai8{
z1Am;7<n8C}bsFU(23_ZuPROD-qa~vaaaYTrn`HrOwyzG_3a@Ajxith=%D02aHG12L
zhc16)*an;qkki%supkeP7|!7%n**&*vcL;TQm`x-6Q5t5i~~J9Vd=>?e{Q3r7ATTU
zI)y#j2TJEN{^ycoXs%(l3a2wa2bREtzQ2_@S?q)1Q4j8gtyepJ#}8ddMpNhsj(hSa
zvp8qu*0yDaYr)LLW_!{Wp2EZ@K9Xk~$W4<{p#xm-b)euNPMOH7H1u}CNaBG~z6N~d
zLivq1m-M2UlVp2qjt?VlD&3}Kj!TbpQbs5Em`4?ylN$5ZQ|JU14oDLx%{|x=Pn=KG
zraZh!ZzqD;)^U<t4@b?HpX4S9dFHNX*Rq~MOYY+6?=kw(Z0;AdDds-u1tfp!FxRA7
zoh?4IM4=mhN$<|L6s-C=rH8I&>(9{Pcaa!pZg30>_Rt}z4_={;96sa`ffU=jTSz(Z
zQi3x@P9D!D2b~}Jq!VpjPdaFwA@JFRe6wYLI|vSP$42$B^spo8jMJIhg5FGn0l+wi
zAPxF`?j0IJf!}8#x23Ck7@d?TOYCkfAaYaR^T@>P#>i4z<rwV!LX-)L9PH9`^K~|~
z>pou!zxfmc;L6f*vp(f7?I0ZGm;eJ<e)^IJ0z;2>wA;<hEW3}x)_n3rzF~lfBlz6r
zuYAO=(%Hb|AeK6Lc15_{VH<*d|6h6jsIq<sF<V(jk|{`_2{nksUI`OHC2^#?uk2U!
zl=^+YvnqJRTQCdvhP|N;YLup9O+`W6?TJ%a2VJeiR9w#LYc`4r1lLXrnEWk+ulX+@
z<-hlMUTpRvH<N>n>$Q$u7#T^OJkihE*~vd}n(_2gg^pssib8hlX+|>!Q9N6N12Q+;
zLc-fKuR%!a6Tbs=uP+$lA|diw%np_Fi>9kD=c|p6c|NNnosKeaR^vbbJoYS@kq-|&
ze1tKx9X2IOa#y_j%>6^(h|!RBQuAEX=n4S=WgeR0x8VU9uQD5<<M8kpY>o_heMs3&
znS7I5$!Wm*OQ2+71a0O69Y&IUQy_6h8X9<KMFOKU2RXmZJNyV`-aAr9^haUn+0ovB
zU@Vz7ID5}HgiqfEQ4bw)uYCj<Q%3AQ`!`%<hUGC2Z`+|n4p-LMcy+M~6M9n?Y|qp$
za*%_1+#S1oK{@r2i?+F>n=ue>Bhxwn!8a0$mL$D9Ek<|9ld!AX=`41Kah=&2#+P*8
zFPvaI%peud6pX$FK)}rB+NDotl1y<7W(G=rI?o`Jt|CE~a*&s_;+mNZI~<5lzUYqY
z+pL0Chut3{aW}W#?n&|dJX~k=8b{-+ufERLPqX(#o~iq+z0S+X+4qCg^K^2V^$Pv)
zKG3`TSzKM`<Q4vmO7kXIh`Y<$9J2DJ3!WTN#?<G?kgQrDFgUcr)`4rf=05T0?*h-j
z9%g&trm~0zL{H6DjGjz<e%qlCPDPQQ$gaoO&vu~cZLIYWIy)K$e05oURMvsf(SFjM
zp9Rb=*|+}DiC@PG-54!8W)_RAGgfcPRn|$F5#p%HPA242F(hxWRqs0}Nru*EV$uKe
zFF%Fm%!KB9Qq|4GS0X8lq;~F&h-AXsTl<ToE`P~q1UMnus9DNpRv?G$E>lanh<v6x
z5<8(fn#(>0s&I-RB4j<5UIy3QyV#4wnH_r*df1=Zx&;uX$||tzEu?;{2bCMUnAzc1
zzOgiep%4=okyCWez9ej8wvldMbh*9(Bu@ua7)V}k>VeB1PMLgb>2+o(qpxPO)T70(
z>Wc4&p4kvC*K{bj^>I^EO9213PpRU60L4F2_KY?HoO(^<D8)ofgL#R?{mZyBw7+!?
zjk1HTYH`5YY0o0u>;gPS3L`{Wt2^tyqvB>HQ60mAwu(1%%eC~2A9?4XON+QP3PHq#
z=tpna-iba(tV}OXo7|x)tfa_{%vQ!>A~!mXq}O<*!8xCUd?juorIw8R6vq>p6W`xt
zO1x*9AJRc~R2jGz_6<lwL6fkHR<;3Fc!}-44>|9K+srK^&*bZwF^f^xSr0@o=Z{y9
zl{`A0U`qOg)7{|r3s=q+Qtooukwq@Zo*`|6pPN+<{~&7A41K2d@tY`MCezY00ArB0
z@l0y3Yp5cRcD80Q8b4*^5q1ZU$8jDx7z`lC4zfAIL22gTp$Z8Ww{#KZ%2p>9FQvu#
z>Bvp*6(bq2eJUROb~3<)|D#Ndp6eRq<ng$!;sGE5F9$!#XK^ST$<}hiPdfL!wp{*8
z@Z`}k)hS#wrRr)pAxKp7ANrgNvwWPKy&d*V_wq=+Brn74@gJx5rmrLp4`gAsqB9+V
zi?<$yWfU8Oor9wxueuWI)S|~L<54c7$Z1ze&wpQgq?huBk2=K4=Rmx?b)t>_Q;(HL
z$~7H6O542fnq7poKQ5j^3i_koyMAPL-$$o8qoH0KB+W0fvFXs~Y=JEBkAy60p<3BM
zcnat*ZRj4oaPExWQ?9wnAb?XPbZt(`S>y|DRJnAZ`(n)uK)YchWY7#T;-_46-@vej
z5ah^GLekTXwoI_a-i%FXyVT>>K8(Duc!Yp29!%$A+K0=*I|5YK+FHQmD7pY>TgKi7
zm?E`9$<Vfwx&&vpwOvQ|arU%Oj=`2;GDY|D%5>!QxZ0I13Fg)7$RcQ}e!19-!9zjw
z7e}s<!KR_F?1Z-K$d(h`0bm~aU1i{6BI-kSe1<y<RwlpF<vDeh;!goHPj~;^D}ZY$
z{|H!xiRF(`x9fLfo(f;eEfVO4#N}LJt(brGG!+KQ(B2ALK<RZQ9;8yGp#c6(v(d*v
z4@0vuDo59+<TU1rSGo|G2U`=fqDR5NH~NK7KF08QiHSdcX_6eud<t+={S1v`@S}4^
zxG>~@Pft=haw+*~tn<8C+{B+)DAdjWKDXgEWq!6=k*9Amu*_DGqhUt4S;?IdXB+V<
zN3L}Wf|QY-Z?B{r;}Ibn+4tGBDe^Sp2MhU9mg8X04kLb{x)i+Xg#*+@U`8d$d~LB@
z&eHN<ivmWcf+NR?19uFbeL6gIHHv<h(Fi`giPFb9nWje<$IU3ov*D3(>QhG&!0eL_
z5RQ%rb3L;*iAb=WwCBLv@ZG-3{t>3D86g1h9ilp`5&o?w!H30fj0)I{|I8re`FSLG
zld`@!V?})HIb1PMGaDA(&=VjXeTL@%S^3B$PE=&&j2D6Yg|?fNyN$yD@61BsbJ04P
zWSj%nV27TWl8#dmStd5Jz*=zSjzRTp7V+vog(J7A|5<MXaOhCKIc5e7XpVDW1SVeW
zplMH<-9GIOiVvMoosjGrEU$0FY+m{WUoi1ZDH3KKn0gLevnh#ta#E)45S#oy4M0x$
za!4XZrlBk{2G19N=MiX~qopTiw;W`Ghxd2+=1A^G2Lc1mlr^gFj;R$VIZr23BG(Vv
zI)aViX`7WdagN;vW%Y&Q4lvv#%-bX>v)T+DB9lYjs(i^qerf8pu$^<jy&Z4#Qj@la
zEytDbA*1*5rtTBB^+{bfYn)u=Kcs%>rkU!(p*w|v0_f>_WM?KL^%dci#$?G1g!Uc!
zOI5jF_9FEY8-UKeOLu`myy;Y5DQm;F01fDowK@k|sGReo4gLJHZ^VQI?K2s!^QM#x
zojnR4-RmQ;x88Ly6>{r{CcW@DNXbJMV0Qcb+|vVmngMY(Gk2RZ_8lR7c8H9wZ?qTM
zoe&z3SsDY&Y<%G)rn_Rdeas-!>`)h>2A3JBoRT-QB)PD7R|F^iy2_Rt2}h-FX)J_W
zfF%MUFTy73MfAEIxQpyhXZfty-@^V`g`Z`@rx40|Mt{Q@#Sys^|IYbbHd#09<4o~^
zfV=-M3bbk=r4X&_W>w`KF}HNeBuz<RrfDuXo8EQaw@^L>u>-%x%eBPT_l}?38cOpX
z&d20E8=Ik|0P?fwP%32^Vo_tPR#}Z8glh}~cBG@RtCN;V**@FOc_;vX1vKRV<B9I1
zEz~^`%<5j6H`kmsyH0vGJg@4m0x7ubr!s&OjuqNK8Rr;)diM-_8jn1UTZXF!Dv;PI
zr5QO_Iy{&dGG)f#GXjeixRN1I2V?F>K4f8}t#l|nrl%vDE&_#UXc%_oOU{VAop~Y|
z0M0@Ue&`>*0xPmNT1Q_E^SX9lzR=xHE%au~?%XD*%j9qUO&!9U5;gcUa_1<db1}uP
z5E&#>Da%itr;L21?niwfGoAw6@(!JiYI#mw6Ow4EdKQyLuH~P@k&#?r(4QKA*QL3|
z2mb;$b+!^yzMIPBOz+f<spxGREP@oj4;{IB9iGMKDu<CbS=hIwJZGk%atx!+`x$-t
zLx1X2xrUrvf5@x8eKnLp5B<}DkPSyS?IpTf`>dT*2S;Y37IboK3)(25(X4#!Y!R0*
zG;P`t$Mr&TU<aqN^7P}A<!NN92FVkhNp2bC&^2Xt96L2Qo4oQ&A@S%KS{6xT8&>9~
zw%y1Zm|*P^5bq#fhS1#Lx<MW~8am|P4o~V(AU}XxHnTkzG2^#s<Og2vx9+GS`^_kJ
z6`f`adXoYd84$E=XAVX#ZExC~`bnX-6|&ttZ5nU|XywgW2yA-h41dy4WpgnX9yOXV
zYHW+*Tls|KA0D1$H93dR(i|8Z9&PkCc+mFtEk`Ljm)Dr5d>&y3$NPJqX~@9h-7Mm#
zSGAd5r$r3HiaQs9L&s{fMK^>l{QHYMrMC<^YZr1`8<F!emUJis#XN_MusiyxLh_Qw
z)3bmIk){YC1^66>fy5+FL-L#h(GY5hIVb#qYq0)s*z^?uK3tT#oK+;)i-(*xU;8Zg
z4y9v}a;B-ueUO-xrOjDY(4tGVDvu4z84iDx9+{lN#%-DdGpbW~8d#NJcS>zwa1COq
z!lwI(FDzIVJ_g0#!X+UT6n_EahjMK=JI;CI*m3hH<zT$gmzj$8CIfx3v)>LIo*9E3
zz-xzcM0k@W-u{+@c=YEWBNS}9=N2bo{iKtQ@ga^)KfN|u_~n<M8~4fxD4Cl-zinu#
zt1ei}{Y{=}q8p7C51L8PDLCwgXg5w%?WQoVa@Jux7Sd{!9SRv0qh+w{U!g61lN}5_
z^Z()d?_+&3()jYaBW^TOZ$?A9JwL8818{wDll>dA`(oN<KmM0y_|s2sTHm)BaVYbi
zEL%41KNpu-;jT{JS@9mY)*AGUlGF#D-hP<$cBlj)gci7mm&oLce0VE#OAhFah-4U}
zj30~d@T-F*7c&m>(YM->6jY}S-Ua&H2Z@?8-S1u<Hh<s=<YQEN5t%nW<>2a0v4uW)
z<m-Bi@S}drbiK)Lk`~>~8xGUnz~sydTVLQ6$Qa#rTHSiGVZ!Oy_vI_{|Ni?QDof5v
zXGhSQ<IBvj3BUXr-HW^*AANd#)^3nbXJZy#9vFQ1b4XV=zoyRKL_dA2rz2YVqd#3^
z5PiPyS5E^rWgPa?K;z6TpfGS#f4)=W6YwKjeCQ<584Hq52LI#*_k<65L(2C-)Crx{
z_A+Cb#KDpJY+B02LDT*+F_+p1taMOjV926t0wPaxEv$CDSV)1`RyqM^ougr)tXTK-
z6v$*z<sygMM!65|==#mjjHlG7=n+yDoh_g1`@BI{CnRy~ZEdb}I8MipKh7&QuU~x~
zIpz(uMc`Ry_Tx0|sjs-^?IZhnm__VI=At`DtAk9Z@s+x}%{mis`faGRFM0s?=uqnT
z!#fr^Ylp^$%o?_|=gff0M|T8|5IH*ZAx13MM5VX!)}FO$sh`vhT%7@oF8r_m?|)UN
z-ep}0sm3HmKQ6xbHT<$$aLL~`nR;(~2z~C=N~<GvTgN7|s{t7sJNjm~*xED11?H=i
zbGFuq4n!4tmQ3!|9l~J-hOw!XS<tn;vt$lq0DOpWTh}Rk4vkw8=j(GG*MCQiUS>bt
zV;#=F1^tx${|}ebBFfcsX+9^uBPE`Ac;}V&GwWLxOB|%LL#4EC)#*rmBHT~h&OK-K
zva58?L(!A5dY=F9-mQjq<XgZCFV}nV6{MT;eTKQ6m%`vUGY~n^Qz1G(A?bw?IOSA+
z4uxnr6kg6?^*1fFBa_541dWrYfdR;3w4Z)%nk~U~AP$Y#q4Q;EXLN(CH1HE`g&DY=
ziVdzk5*<9bhjTSnM!#)O`^&K?P9AKhu1gdCr7a@20K}ZG>4?V~oCuDQSA(yz2Y_-T
z`k`f<*vJu$Ip!7U7>Ham2t>uGKCF>9gNUxj8}vlR;T2RJ63W!0*`6X%2V^0SLtW&I
zmj?H_n9Ze9B=_nFUV&_On)|}ac|R^dqw<B8GNm^#j6mDTNrwa61#>YE+{YpL)HhuV
zm7^~yx9jA<m6Cn6Hu})dlDm)9y~twFn~YZRDm?h&XlU^PG*i#6J=gC?=Rl8ey&W<$
zDoUiIi9Ws0Ht?KOuXnzWBW1v5&j))r*bvQZ)?FRzP-O5C8a+CJx2-49Z7=1SEJp_A
zjlM@u%ur2mKXL*`9h9cfx<{?;z>;RsXa{F?KnJI)zI_#bu|sb1scW>5Luo+5(6GOj
zJMCb~EJ8kcs;*$Oe**`s1~51LPnQ1eNOG!O2;r_ti8l2U1A)2quJ*BV4J*M{bg8#r
zGw=wf<r(vxy-4DH>S>&i@SglrSF>>!hkq8OE;uRzFAjj$QJjJ4tW)=RB80VFnf3D}
z?aBqs(bp?){}4&wh}UikGuK=1ap~IR0zLZbE3dOJ2O@PXfAN-AdC6|*0@|ZUI-}m`
zt8~1vf%tcDR@D0kwt*K?7#z=;z|by5(l=?_`2f@XJci;p$GNH3auAi7kNovHelVBN
zoXc~7S&dr;0{&S@`DelXiztZyq3{ykKLqT*zA!mY&Ha7GwsL(W=Bb%yvL0wrGJl86
z=UQhaOq#eOL*xHhKN<(axWn%3dY||{ZpqCNf69yK-RKG-_uZN~`D^S#a^q8`GbqX`
zBVY2gVd_mA-=u{$G^OOsc%)H8pW8s1nx7QW;NJZFbK|AmWcS6Jhs=hg@j0x+s9}*k
z_{IpQAh{MEBhNVK+(J|0_GhyjmwRh9%3tV%V;{S%;jZavt9pEFIK)qI-p9l!Yb4c~
z44p>G`pW3Ykvx&7%#29!dF~ABNUcr;3?o!!J+}qhHIXtlbm%&34;q*Fj<f^0vLKVd
z>7h)n>mZ;w_W~O$><*<)Ow#)79=yoM$la(5ZUG&N^$wiGx6{jYdC7U?E?;PF=Fx`&
z_{5Ja%9<e133YG(oU0dOF%jL8VBe)vu{VR=9r2gBs=N^sTB8?=8wKkm9<#&fMRvRF
zmx;3<M&AYq6!qVk0<c%`O_;0qS*H;vghOHkO->e?GUnAcb=8@W$cWPuWVheW>`R>p
zG)9j$w^J`u-%lOEup({hJu*_??$hDSqXc!0FhA;1n~F3$E1nU1+d$5@)}JIE&J=K<
zX<vhYBzerfSo6})4zvcP{zg!ye@PzzAD0mShr;ME40z*^bXF#G?{CkrG58Q_OLFP0
z8Z_PCnLW&X(>Wo-uQx)f_wfsW=J<JB`aZJ(<mzo8rlGQ&EP(+7D(2W1m0vQa^;CZ!
zvUWn6H_))%*{s07_J}nR90m2($AE#gB{TIPe>O7ftckYyb^G+FV9Fd)4(#B{-xh00
z%T3#V^4^Dr?$XQoNqb}iXU3Ea#wIFXrz}nhg2m(E{w{dH$}xC{?kz`)t+oBG+3Ug<
z#Y6PSCyFVju&4uU-%p;o{peS|o%ftr$!{^86aIYHTgsdBQ5v!Hx8u*}%KuYxJ!g7(
zor)T#V`|KAF|1Yqudwm==Ja#a`6=#mFLkW5%KB8|(67N+Z9qf#8cHTFdZzqS?51(+
zWu1{^Zw@d{us>u)HrjndL$h!V*NATxbi%ZaaZ;p;D!h8=40M8you@8<q5(5B8ZS%1
zsp+=LMy%nvZ;#|WDPxp$K4{<r%z6%v!0sfr6EYpZ^b6B?mFpspKr8k+4L`XaeimiP
ztAjIBb{$!CM4ryeD7SDt3L&F(?qqi^3qm<R$q+u7I>OGiQS!V6f|nTD7{K8v3pI68
zUP%sIM#bo|Zw~+)Bl*~!EQhD^5n6=4b{(8UPtrqw>nd@}o|Rr3dkNUzqir#cI&ewo
zs5AU3pP%VOlD_pqS=Tj_Ig<sv`%wv4+sMi)l$Y?+kh*UEIC%1r<IrLzETsbvtPd97
zxlX=UU*?fk@`^|N=*^M?8Una52E(8UES<K+hw#~qO#k-it(eHYU!_f18;D_b7A4ZF
z=t=-G|4ANLp?UZme(ANTZ?fP(AiJGyaZVY(qc1w9(3*U8WWuk4iOIsnm9}IJVc+OU
z+47Wg9n&~s?VQk?bue^^Be&}aE&)9ga;1?Z`0W)O|Aa7LIBJ1QK>ouA+GoaNWW;Ii
zJ)F|yO_)dPuIDI-T#}WZGJa?I((YzfF89e)D-z_LWhxOTeio43F4^kJE$vC0xw)k)
zk>x;18Xbs(YGG$=(B(xo@e6WoAlQ(bL;D72WZyL{NqhC>O&`VlnpcMV2rO{IX}%-m
z7oQ-Ej>9o^j()n9N21}=TWfW=ienrQ4hP+%s~%C8hv-@LP9qMi4j{q@I;&H;9$JgJ
zW#U-P#y>}cNDc-XfC@B))R)jbFF^Q>kgQ?ldHC(z&-49h?kW%!uqE~R5=xb{B0rY$
z$I=@#Q_(L|cjKA{F!-lhKcAMDFwgf@wl<D)EahA9b3QP+#@I>||H#mb6QnavNhrOE
z>MYBKvCorvi6)uTkU`M6gtVhx^~<k`8XX_BlVjj(SnlU}y?5rbmLd&MC&HL$pfwh1
zF!vt4n5yiXG}6(H-kgo@8LD>Txz-?OB$Bu^@oAWo2XD4#pN2Yc`bY!q2<lJHrB1}~
zCti{%dkb0h7aftM4LctrA9AXJs%hNfF1RtJB<~nhU`~gWbNK1^YXSq!Pb>0c6dAI_
z*EYl$eb8|$oP6EaV#$Q5-0=}$jW<eackm#;azfzcNps<H2DVIg{g=v3OYE1r9A3$C
z8RUUfSo`rGT!oRc3};vW=UUh!yIqW<S!Xj5B+dEo815R382RZi1`fG?Fl`<u@O2g~
z+PlQsBhE<I58=-@Sd=S$i#L^Z3TQ%&wHf4k)i1-w8IH6(+OThfK#wA@@*K#mlR(TL
z-iJqZiM8!`T6Eh@_vm0ikwb_IG#DKTs!g0|^^v-xR~8fU9l?5BPJZhoFtpQh(Q+x6
zJYln8-s~FKV42z3bvZfwlFi&nJ3_8ccSrOex-ej{8ry&Cn`{Cf{pTY%n=Z~@@+kBn
zPGG};Xq--wtYqnS?YY)pcDLPJV~W1jQAu6Nd*+8!9Dd-mQ?vRPSRnWnm@;upQ-AKR
zBgfh<YYl^MW@cxQq5SzUD~FU<+570LtPkl9$+2I6;?3+9JokCbhYvcf4yNBvU8gQl
z?Tw!u+D_`^j3dG}Y9}9V*~RF4)@hZ7l#Q}4T1S@(n_Ob&##zcayr-R*5}BzMh!R)E
z7>&Ikb?Fo;hZdZj!oG8MRJc2xkbC_scBg*+`326;FZi#Sb1Oe}dl}A+Q3pUn$o1$@
zR)H$2>lYw4Z0ScHSEqbC&jR&NJ)vKxvcpD;2SmVq){~A-R{ct&^v(yt%CxX3*Gp=y
zH<ATg6XwHfsAG^P1KA0CYQNi3r@V#}8xh$u6w_Eh^{QtSe&{gzOZhq!a0h8e8!3w(
zyD}pGNKg4vm@=hX0daxvSa)+@NhpgtI2>~ws(Sr&|BPFi8qoXP_y90iR+gLwE?6}N
z;kR%@;5^4QQpT^2CT0DvW5a0K$||t(*AY`jyseYi26m#GVJ*jjLB32U22lBuJ_Izb
z8nW+`hGLXYUUE(y<lu<Rd{h#p=cwaR2V|TOhI1GKddVaCz}$9g(+MURrC*Lk1D)ky
zfn6HJC#d^bbeRELqWY!U7;Fq%kmq|MQ@72&kSZ7)Nhb##u*HgIk#=1rrg+|EJ8e_=
z(4lwM9Su%=p$YusQ6QtA%!d<OZa9cKvPJWp0xjA>9uDmSf+H>0LhjXPFSQdQ$+;bB
zWMZ!kk5=dfdiJ4S!R|SQ+tF5L;_@FE)$uFjt;+|d@Bop-rHi>5Dq1`@`>p_vw7L5o
zzX1c4wpho(b$CJFroFmx;w3538^0b!aY$3yk-M1crSwqLcEH(B{TPh6H#?xdCQD+%
z2ToqF)8tSwf9hx94~F0z9enRA*~%6|FC~`-<ay0`bQO7HuUaR`%K5<KU%vb9i@$vH
z%@;rZ^kZJ`_EwO*!51f4-Ybtz5@+VBgNzz2u&hFk-bGKXiz*(xrA$~ZO`H%M&`s+^
z=qg!Ru<af$VbjDi3LP|fba__?%aq}KlR;vYJNX#ti@{Wb42ZPm>F?=>kejyB=d?0}
zibV*r|6F0Kj%vdD2;%%(#Y?=(BI8|I<TP~=7!)9XF##g;4}t;%bDG9L2d5r>(z{>?
zpcT&W&v<tF`8-wq$LdQd%De-bhTfb2#^=QJx^c23po?NZHOPrCyQ@@bUockc0kaY*
zK2bAO*kaLTGSnx+XSFsQv+F6RWsEvQ;_War<|5~0&tHv9c#@A1nvP@)hkB6?E(giB
z==#syF#)BnT91r;3>(1ipLkFDaVt-P({B7Ls_|v0q~&&WXLes@R5bk1Qk;|f`7b<+
z-qulguop=T2FGf2B=i~m&bz#c54a7FJn&c;ITwFm=ophJgZLUgc^<Ne?mk<RSNGfD
zC>KV8*IEie$6wYeFgi1gPCVK&E0#f0oyCP$4$s8<+&!Ip9g^}gbXEr8WgF#zs5(dQ
zoV6K-uxa_xImv&1w1ZY`4?YVG$Fc1;^`6=5)T68R+i6>D<`EV-D%V|UgNFmsbsuq@
z%#axTMXOr=eZD#LK3}n=7U;Ummy!7rUhUSQb*3$P;BkgvRtE6V=+H=p@U5A&5Yv1_
zjf&nSeJU}L5+=twi&}9#uROt7M@NU1`j+(YXl8Ia+R+uzBUeSxA(OV1xUh5JhF;Rf
z)9bELN+vg<v!`4MW5edU_x+LY4~78WrAKFT#}<UJ(TZ|01QV6Nre$r|O|DaC{)VBH
zYw<W8k%j$V+IdGO)#-lghaOg=F-AY1F?iyG2nmavJ>JZ{ae#r7x62eW6DeCRM%9ux
z8lLMoAxYSEf$r@L+M~t{%D?~Hzr_*x_KW}d|NhUyFu+pUuD_7=gS`Q5V?AYc`{0==
zM8gcu&=EqA;7l*q;r{(qHa<)H;Xs?z;2hnA#tJ>z@g(#ySnlP6gOmP$03qXL$kNB~
Q;Q#;t07*qoM6N<$f};@l=Kufz

literal 0
HcmV?d00001

diff --git a/docs/2_39/index.html b/docs/2_39/index.html
new file mode 100644
index 000000000..5ba34cc12
--- /dev/null
+++ b/docs/2_39/index.html
@@ -0,0 +1,1129 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+<meta name="author" content="Stan Development Team">
+
+<title>Stan Documentation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="site_libs/quarto-nav/headroom.min.js"></script>
+<script src="site_libs/clipboard/clipboard.min.js"></script>
+<script src="site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="site_libs/quarto-search/fuse.min.js"></script>
+<script src="site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="./">
+<link href="./img/logo_tm.png" rel="icon" type="image/png">
+<script src="site_libs/quarto-html/quarto.js"></script>
+<script src="site_libs/quarto-html/popper.min.js"></script>
+<script src="site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="site_libs/quarto-html/anchor.min.js"></script>
+<link href="site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Documentation – Stan Docs">
+<meta property="og:description" content="Version 2.39">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="./img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link active" href="./index.html" aria-current="page"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="./stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="./reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="./functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="./cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar zindex-bottom">
+    </div>
+<!-- main -->
+<main class="content page-columns page-full" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">Stan Documentation</h1>
+<p class="subtitle lead">Version 2.39</p>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    <div>
+    <div class="quarto-title-meta-heading">Author</div>
+    <div class="quarto-title-meta-contents">
+             <p>Stan Development Team </p>
+          </div>
+  </div>
+    
+  
+    
+  </div>
+  
+
+
+</header>
+
+
+<div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
+<p>This is the official documentation for <a href="https://mc-stan.org/">Stan</a>.</p>
+<ul>
+<li><p>The <a href="stan-users-guide/index.html"><strong>Stan User’s Guide</strong></a> (<a href="https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf">pdf</a>) provides example models and programming techniques for coding statistical models in Stan.</p></li>
+<li><p>The <a href="reference-manual/index.html"><strong>Stan Reference Manual</strong></a> (<a href="https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf">pdf</a>) specifies the Stan programming language and inference algorithms.</p></li>
+<li><p>The <a href="functions-reference/index.html"><strong>Stan Functions Reference</strong></a> (<a href="https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf">pdf</a>) specifies the functions built into the Stan programming language.</p></li>
+</ul>
+<p>There are also separate installation and getting started guides for <a href="cmdstan-guide/index.html"><em>CmdStan</em></a> (<a href="https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf">pdf</a>), the command-line interface to the Stan inference engine, and the R, Python, and Julia interfaces.</p>
+<section id="older-versions" class="level3">
+<h3 class="anchored" data-anchor-id="older-versions">Older Versions</h3>
+<p>This documentation is for Stan 2.39. Older versions of each of the documents linked above can be found in the table below:</p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th>Version</th>
+<th>Stan Reference Manual</th>
+<th>Stan Users Guide</th>
+<th>CmdStan Guide</th>
+<th>Stan Functions Reference</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td>2.39</td>
+<td><a href="https://mc-stan.org/docs/2_39/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_39/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_39/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_39/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.38</td>
+<td><a href="https://mc-stan.org/docs/2_38/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_38/reference-manual-2_38.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_38/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_38/stan-users-guide-2_38.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_38/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_38/cmdstan-guide-2_38.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_38/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_38/functions-reference-2_38.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.37</td>
+<td><a href="https://mc-stan.org/docs/2_37/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_37/reference-manual-2_37.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_37/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_37/stan-users-guide-2_37.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_37/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_37/cmdstan-guide-2_37.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_37/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_37/functions-reference-2_37.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.36</td>
+<td><a href="https://mc-stan.org/docs/2_36/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_36/reference-manual-2_36.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_36/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_36/stan-users-guide-2_36.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_36/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_36/cmdstan-guide-2_36.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_36/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_36/functions-reference-2_36.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.35</td>
+<td><a href="https://mc-stan.org/docs/2_35/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_35/reference-manual-2_35.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_35/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_35/stan-users-guide-2_35.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_35/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_35/cmdstan-guide-2_35.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_35/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_35/functions-reference-2_35.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.34</td>
+<td><a href="https://mc-stan.org/docs/2_34/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_34/reference-manual-2_34.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_34/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_34/stan-users-guide-2_34.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_34/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_34/cmdstan-guide-2_34.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_34/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_34/functions-reference-2_34.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.33</td>
+<td><a href="https://mc-stan.org/docs/2_33/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_33/reference-manual-2_33.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_33/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_33/stan-users-guide-2_33.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_33/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_33/cmdstan-guide-2_33.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_33/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_33/functions-reference-2_33.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.32</td>
+<td><a href="https://mc-stan.org/docs/2_32/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_32/reference-manual-2_32.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_32/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_32/stan-users-guide-2_32.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_32/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_32/cmdstan-guide-2_32.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_32/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_32/functions-reference-2_32.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.31</td>
+<td><a href="https://mc-stan.org/docs/2_31/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_31/reference-manual-2_31.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_31/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_31/stan-users-guide-2_31.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_31/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_31/cmdstan-guide-2_31.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_31/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_31/functions-reference-2_31.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.30</td>
+<td><a href="https://mc-stan.org/docs/2_30/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_30/reference-manual-2_30.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_30/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_30/stan-users-guide-2_30.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_30/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_30/cmdstan-guide-2_30.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_30/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_30/functions-reference-2_30.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.29</td>
+<td><a href="https://mc-stan.org/docs/2_29/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_29/reference-manual-2_29.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_29/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_29/stan-users-guide-2_29.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_29/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_29/cmdstan-guide-2_29.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_29/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_29/functions-reference-2_29.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.28</td>
+<td><a href="https://mc-stan.org/docs/2_28/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_28/reference-manual-2_28.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_28/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_28/stan-users-guide-2_28.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_28/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_28/cmdstan-guide-2_28.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_28/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_28/functions-reference-2_28.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.27</td>
+<td><a href="https://mc-stan.org/docs/2_27/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_27/reference-manual-2_27.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_27/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_27/stan-users-guide-2_27.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_27/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_27/cmdstan-guide-2_27.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_27/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_27/functions-reference-2_27.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.26</td>
+<td><a href="https://mc-stan.org/docs/2_26/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_26/reference-manual-2_26.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_26/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_26/stan-users-guide-2_26.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_26/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_26/cmdstan-guide-2_26.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_26/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_26/functions-reference-2_26.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.25</td>
+<td><a href="https://mc-stan.org/docs/2_25/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_25/reference-manual-2_25.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_25/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_25/stan-users-guide-2_25.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_25/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_25/cmdstan-guide-2_25.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_25/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_25/functions-reference-2_25.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.24</td>
+<td><a href="https://mc-stan.org/docs/2_24/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_24/reference-manual-2_24.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_24/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_24/stan-users-guide-2_24.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_24/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_24/cmdstan-guide-2_24.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_24/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_24/functions-reference-2_24.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.23</td>
+<td><a href="https://mc-stan.org/docs/2_23/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_23/reference-manual-2_23.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_23/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_23/stan-users-guide-2_23.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_23/cmdstan-guide/">html</a></td>
+<td><a href="https://mc-stan.org/docs/2_23/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_23/functions-reference-2_23.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.22</td>
+<td><a href="https://mc-stan.org/docs/2_22/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_22/reference-manual-2_22.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_22/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_22/stan-users-guide-2_22.pdf">pdf</a></td>
+<td></td>
+<td><a href="https://mc-stan.org/docs/2_22/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_22/functions-reference-2_22.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.21</td>
+<td><a href="https://mc-stan.org/docs/2_21/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_21/reference-manual-2_21.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_21/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_21/stan-users-guide-2_21.pdf">pdf</a></td>
+<td></td>
+<td><a href="https://mc-stan.org/docs/2_21/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_21/functions-reference-2_21.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.20</td>
+<td><a href="https://mc-stan.org/docs/2_20/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_20/reference-manual-2_20.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_20/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_20/stan-users-guide-2_20.pdf">pdf</a></td>
+<td></td>
+<td><a href="https://mc-stan.org/docs/2_20/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_20/functions-reference-2_20.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.19</td>
+<td><a href="https://mc-stan.org/docs/2_19/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_19/reference-manual-2_19.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_19/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_19/stan-users-guide-2_19.pdf">pdf</a></td>
+<td></td>
+<td><a href="https://mc-stan.org/docs/2_19/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_19/functions-reference-2_19.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.18</td>
+<td><a href="https://mc-stan.org/docs/2_18/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_18/reference-manual-2_18.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_18/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_18/stan-users-guide-2_18.pdf">pdf</a></td>
+<td></td>
+<td><a href="https://mc-stan.org/docs/2_18/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_18/functions-reference-2_18.pdf">pdf</a></td>
+</tr>
+</tbody>
+</table>
+<p>Prior to version 2.18, all documentation was part of a single document called the Stan User’s Guide and Reference Manual. These versions are still available for download as PDFs:</p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th>Version</th>
+<th>Stan User’s Guide and Reference Manual</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td>2.17</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.17.1/stan-reference-2.17.1.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.16</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.16.0/stan-reference-2.16.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.15</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.15.0/stan-reference-2.15.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.14</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.14.0/stan-reference-2.14.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.13</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.13.1/stan-reference-2.13.1.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.12</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.12.0/stan-reference-2.12.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.11</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.11.0/stan-reference-2.11.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.10</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.10.0/stan-reference-2.10.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.9</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.9.0/stan-reference-2.9.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.8</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.8.0/stan-reference-2.8.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.7</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.8.0/stan-reference-2.8.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.6</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.6.2/stan-reference-2.6.2.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.5</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.5.0/stan-reference-2.5.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.4</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.4.0/stan-reference-2.4.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.3</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.3.0/stan-reference-2.3.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.2</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.2.0/stan-reference-2.2.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>2.1</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.1.0/stan-reference-2.1.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>2.0</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v2.0.1/stan-reference-2.0.1.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>1.3</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v1.3.0/stan-reference-1.3.0.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>1.2</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v1.2.0/stan-reference-1.2.0.pdf">pdf</a></td>
+</tr>
+<tr class="odd">
+<td>1.1</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v1.1.1/stan-reference-1.1.1.pdf">pdf</a></td>
+</tr>
+<tr class="even">
+<td>1.0</td>
+<td><a href="https://github.com/stan-dev/stan/releases/download/v1.0.3/stan-reference-1.0.3.pdf">pdf</a></td>
+</tr>
+</tbody>
+</table>
+<section id="copyright-and-trademark" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
+<ul>
+<li><p>Copyright 2011–2025, Stan Development Team and their assignees.</p></li>
+<li><p>The Stan name and logo are registered trademarks of NumFOCUS.</p></li>
+</ul>
+</section>
+<section id="licensing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
+<ul>
+<li><p><em>Text content:</em> <a href="https://creativecommons.org/licenses/by-nd/4.0/legalcode">CC-BY ND 4.0 license</a></p></li>
+<li><p><em>Computer code:</em> <a href="https://opensource.org/licenses/BSD-3-Clause">BSD 3-clause license</a></p></li>
+<li><p><em>Logo:</em> <a href="https://mc-stan.org/about/#copyright-and-trademark">Stan logo usage guidelines</a></p></li>
+</ul>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/quarto-config/quarto_styles.css b/docs/2_39/quarto-config/quarto_styles.css
new file mode 100644
index 000000000..c796b6023
--- /dev/null
+++ b/docs/2_39/quarto-config/quarto_styles.css
@@ -0,0 +1,299 @@
+/* quarto website styling */
+
+@media (min-width: 1020px) {
+  .navbar-brand-container {
+    margin-right: 1em;
+  }
+}
+
+@media (max-width: 1060px) and (min-width: 991.98px) {
+  #navbarCollapse ul:last-of-type a.nav-link {
+    padding-left: 0.25em;
+    padding-right: 0.25em;
+  }
+
+  .navbar #quarto-search {
+    margin-left: 0.1em;
+  }
+}
+
+/* navbar sizing -- important to avoid odd-colored gaps */
+
+.navbar-logo {
+  max-height: 44px;
+}
+
+/* Remove top and bottom padding from navbar image */
+.navbar-brand {
+  padding-top: 0;
+  padding-bottom: 0;
+}
+
+html {
+  scroll-padding-top: 61px; /* height of header */
+}
+
+/* controls what padding is on page-load before JS overwrites */
+body.nav-fixed {
+  padding-top: 61px;
+}
+
+.btn-action-primary {
+  color: var(--stan-bg);
+  background-color: var(--stan-secondary) !important;
+}
+
+.btn-action-primary:hover {
+  color: white;
+}
+
+.btn-action {
+  min-width: 165px;
+  border-radius: 30px;
+  border: none;
+}
+
+/* stan-dev custom styling */
+
+a {
+  color: var(--stan-secondary);
+}
+a.nav-link.active {
+  color: var(--stan-secondary);
+}
+a.sidebar-item-text.sidebar-link.active {
+  color: var(--stan-secondary);
+}
+.sidebar.a.active {
+  color: var(--stan-secondary);
+}
+
+div [data-bs-target^="#quarto-sidebar-section"] {
+  font-weight: bold;
+  font-style: italic;
+  color: var(--stan-secondary);
+}
+
+code {
+  white-space: inherit;
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+code:not(.sourceCode) {
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+p code {
+  white-space: inherit;
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+p code:not(.sourceCode) {
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+li code:not(.sourceCode) {
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+td code:not(.sourceCode) {
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+pre {
+  word-break: normal;
+  word-wrap: normal;
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+pre code {
+  white-space: inherit;
+  margin: 0;
+  padding: 0;
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+pre.sourceCode {
+  white-space: inherit;
+  margin: 0;
+  padding: 0;
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+#quarto-appendix {
+  color: var(--bs-dark-warm);
+  background-color: var(--stan-bg);
+}
+
+.align-equations {
+  text-align: left;
+  margin-left: 2em; /* Adjust as needed */
+}
+
+.book .book-body .page-wrapper .page-inner section.normal pre {
+  font-family: "Lucida Console", Monaco, Menlo, monospaced;
+  margin-left: 2em;
+  margin-below: 0.6em;
+  margin-above: -0.3em;
+  padding: 0;
+}
+
+.sidebar.sidebar-navigation > * {
+  padding-top: 0em;
+}
+
+#quarto-sidebar
+  > div.sidebar-menu-container
+  > ul
+  > li:nth-child(1)
+  > div
+  > a
+  > span {
+  font-weight: bold;
+  font-style: italic;
+  font-size: 1.3em;
+  color: var(--stan-secondary);
+  margin-above: -20em;
+  padding: 0;
+}
+
+
+/* custom CSS for the install page selection */
+
+.grid-container {
+  display: grid;
+  grid-template-columns: auto repeat(4, 1fr);
+  gap: 0px;
+  width: 80%;
+}
+.option {
+  border: 1px solid var(--stan-dark);
+  display: flex;
+  justify-content: center;
+  align-items: center;
+  cursor: pointer;
+}
+.option.selected {
+  background-color: var(--stan-highlight);
+  color: var(--stan-dark);
+}
+.option.disabled {
+  cursor: not-allowed;
+  color: var(--stan-highlight);
+}
+.grid-item {
+  padding: 10px;
+  border: 1px solid var(--stan-dark);
+  color: var(--stan-dark);
+}
+
+.flex-grid {
+  display: flex;
+  flex-wrap: wrap;
+  overflow: hidden;
+  margin-left: 1px;
+}
+
+.flex-grid .col {
+  flex: 1 1 100px;
+  margin-top: 0.25rem;
+}
+
+/* when we think we're all on one line, remove the double bars */
+@container (min-width: 500px) {
+  .grid-item + .grid-item {
+    border-left: none;
+  }
+}
+/* otherwise, put a little space */
+@container (max-width: 500px) {
+  .flex-grid {
+    gap: 0.25rem;
+  }
+
+  .flex-grid .col {
+    margin-top: 0;
+  }
+}
+
+.row-label {
+  padding: 10px;
+  color: var(--stan-dark);
+  background-color: var(--stan-bg);
+  font-weight: bold;
+  text-align: left;
+  width: 20%;
+}
+
+/*  on small screens */
+@container (max-width: 700px) {
+  /* Move the label above, rather than to the side */
+  .row-label {
+    flex-basis: 100%;
+  }
+}
+
+.highlight-box {
+  background-color: var(--stan-highlight);
+  padding-left: 10px;
+}
+
+#install-grid,
+#prerequsite-box,
+#install-box {
+  container-type: inline-size;
+}
+
+/* Table Styles */
+table {
+  width: 100%;
+  border-collapse: collapse;
+  border-top: 1px solid var(--stan-dark); /* Adds top border to the table */
+  border-bottom: 1px solid var(--stan-dark); /* Adds bottom border to the table */
+}
+
+th,
+td {
+  border: none; /* Removes borders from table cells */
+  padding: 0.5em; /* Adjusts padding as needed */
+}
+
+thead th {
+  background-color: var(--stan-bg);
+  color: var(--stan-dark);
+}
+
+tbody td:first-child {
+  background-color: var(--stan-bg);
+  color: var(--stan-dark);
+}
+
+/* functions reference */
+.index-container {
+  display: flex;
+  flex-flow: row nowrap;
+  align-items: flex-end;
+  justify-content: space-between;
+}
+
+a.unlink {
+  color: inherit;
+  text-decoration: none;
+}
+
+span.detail {
+  font-size: small;
+  color: var(--quarto-text-muted);
+}
+
+strong a.unlink code {
+  font-weight: bolder;
+}
diff --git a/docs/2_39/reference-manual-2_39.pdf b/docs/2_39/reference-manual-2_39.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..d36bfff9d2373ef76c9a0b29f03ff82c44e35fab
GIT binary patch
literal 900313
zcmcG#Q;;Y@lr7q}ZQHhO+qP}nwt3pNZJoBQ)3&>B&z<{jOvJp1H$Sg_G9zPGR#onu
zD_3ohDu{^DGSaa^k<KrTtU)m{5HJwf8(Bf|@IcYanA(}USP(EVauEFI1w}7rY2#w*
zL_jZQW9VWkVrpz}VhY8_2j%SIWNK&&<*^Z}z$*{T2($e_%|kZS;VJ2}TKxw!a2P;@
zCSqSjNu|Oay!7@%(u;r)BLRGK&TkGsEpOs)XyeaoQwiGX05X?bf`Fy8gEZWk1i_Y+
z_<1DW<8}kS?BVdg&F&fS;L_PcvblLz{@xewhm2Y?S_c$IRp}7WL@e|5a0cmZYyd9&
z9k<2w5@USoY^<C1nlC<-o&1HW&%~F-sf4jZ7!u^%*)N3&t6Y>x*)#!*DJaGyaf7uq
zwqP&S%l+aUF29g+rR^HB-QIxcDZu8eCE12u`413MU^!PPQ#+IY^O3*3|5zaQ|FTdj
zo(`r2^cwO;R;I=-Q1r^KMlSz*MAFvK+!Tsl-O}WD8JSsFpy(w`EzK=l{<AM+>Ef(l
z>LhG$>tJtZYUe`00YxutZ)5MI>|kj8JD-TDo29X-lDN?CjIxH#)&z{Kzt2?qzj|S0
zVB+{s&HR7s<;vI6MMbsKt;Q8K=VRk#@^n?*);3iTYh)DA1S(S)LJ~rfKmY^S5Dc;y
zrc{OyvZ6#OwQ^Dh42g<aGD^kjh6+#+MI;4G5G-+`0(bM{YsOuUiXfq|Oxd;URCCTr
z(!K85ZPjo00dDTYE))phzg!7SQPbcLw5^wFSvxX|*tFc(78`z@EqUGQ@@=cyY|HbT
zK4atG(J<`Gh^DZT_C%OuRh1)Y#%$%wivv=`aL524(MVFuWZGWDiYrOWVvX%N)0AY&
z_Yx&(ASQ{7nK_ISvIqBPIYf=sLAA9+*s&^_((Y_p?#sBtbpQZ-!aXyi8J}@tjc)jw
zDq=`b5`V&^rC<6*=PTY^3@LS+;=8rQnYA!hT_F0<0UUsM31eh!D+#5kX_Ox(+Gzwe
zdYEytC5;|U{4h0p;;$_V-q4`^WC0wIM%XrI#X1|pAa5*9YDH<(ZeWOAZD5)j(k?9s
zGDp=uR{$KyVc4|Akhc|NPg3XiY=}rVZz8{`n5rUd{<L1(5n>5BfCH=(ZJU_&e#BgX
zn`s{zeDE?7L{HPTR<r-TM5`+5!9@p7;K{Jp7em(6m(4KIpKaYmCbyVKa%%0UV9RvS
z1~|{`Ys@1_N6Pen*OkC&lyJh_R+D$jv}xxL_@lVs!_ksRS4BX`$zaQVUQQGlYO>9D
z!QFzw4uyaygDzt!8}p`|!QX~BSfrG+A`@d?U;_dYyJdD!hQ%cb-t<I7ZZ)x~s-kH=
z>I)1Ks?@}3%b(j94r^ps&Ug3-)YuZpG=XkM&`;mI)ZLpUH!b=U&mLwy-f30Ir0Iwa
zxJ%qL(pi$TJk*Z_sw5{vN34BOs0Zc)Xxydf8$d^6{Pa-I2k2zQ$ISx!2DnGMv}vP_
zeU5VGKirSOtGJ@+tjGnxK?8&lJE?^gN*Az(QE2MUWYTI>26+oXmP}c*z_l)5s7$m%
zWjzi?thy|m4$%%X1Uc08<kAS}Mv_`_s?*pJN0puB&<1Eezq6CKA#w0v!2P!x+s2wC
z!wSSJyfSpV58OMk8^F>W`7cNTH~hShgC{6Rpws@EOyi#zyQ)xj0J~J6EIFH6QbH^g
z;A8f#5~3+;OE4qgb>f!9(IO0rmm)yfZFtQ_nL_65bk%;SemS!A1>Kee(kbX)8*fKT
z$M22~v}#<jP~Au%lfbu9-u&cke=x8|hT424W=NokF^{SEh?GGK4#r!7S64|&6{F!x
z!eYvzVUBhj5<JbboFb!sR(cqFOs~?2qNuPE$N*c#cIj{DX)I36id(p;o!_-{x!+}9
zkV+6>?GGsfSo<i??;sQn`F^;>Z}q3Q|Eu3S3elv|uqi@*Q3wa*{rQ-D_^7Y$&Gy6n
zYR|?X5J%9Pw5LTZGPtJ;qn5n<A+b#8btheJ(?{*!Be}7u$b$lYKfdpI1{0*q_r<U0
zy!d%Di&*{hjYho~8aItbod*vPykxVYZ&#d7=V18pA_B@@-KQrUN1$U<H3l5y^z&HG
zz9>x2`Toca0ka1M67H02Im4F%V)}kAUqLh~JdB<12i`5WS%(|?1X;#uAxRS=I&SUV
ze`!=?NjfICkYD$<^?5_|_T@cId>bDFtY(9hHZ`h9yRIRO6_-$z{@U{8{r2J74y2bW
zfD9R^pz!xmIgyi8#5dsU@yAJ5<HTS6?iM=*`gLFh3_1@rxqJ8}n|ko~rWUbP$m+l8
z3axE*>_zkhnvpjXi$x7zbdAD)<4ni-1PzT3|2k>JmJIsIQghx@zMXe!%>3+-4Mrl@
zE8n1j*vLxGLq+8yYIT;zvEY)f(jil>`Jr0^zDMo^&)Vc^0}%_2$m1vu!1Mb6WFO;X
zp=D`+lqS4l)DmJ>YUa+RN{bw{?v9{-dcn$9J5`|&L3%jogyeq<lY6ed-EPQq3;uk~
z5U$aakMjzYoe(1-ESCRNMc=o4)qeV8NZn(s&!}A@A@&Zlx25`hcEB|Hxxe+PBcQg&
z<22F2-M~S!-pkWHtr*miynVPUV0NHFvS^l}_aA_eKS@>?p?(u3y+MX+IGr^b)sL_h
zBb)~)a4@FZR1oLXJSNglykN+=V1d?%o1+gw-p-we=3ag48xC$Yd{MUpjm=}lLQ5G2
z^N@dwhUD>rXFvUg0kZ8g2r7LE(kOOhGy{J%FksW{if6$qWbX~CcsBt9Je1%yp5e<V
zk;hC})_nOubLG#Mjxo5|KG^2MD7y*zF-2wY-zf%iLAn`v2IHcGM*FABMBd$)w1i)H
z_6})6bN7Pn0O~KWY%=P_Xs`Yq4e^ELq!0Knx>j=!a1Bf_8a7(2XBdOGYwl+WGb6iR
zH}Fyvw9;#nB?7F&OarcK>qhwTOO%9)gZRQ6IsGbh-D!f8^s;>yUVo@@c!e+F?m{B_
zf&l9%ERjp?)%}|XMrx^favrjTP2+QcV6W)QoJ-7qu)mH4EYWVVN+DsvBs(K|<PZ-O
zz^mz#&38h;=`-6#Cvd|gE0mFFjtbzX6RIk)dbYkt7mamjS(U-u9i%_o>ill89z`Oq
zv=F;FE>yScwY?mUGGGfS#J)s_S!H>%$YmTtXYIR?-55k5k1$Yma4MQX`I3Yjdu`dD
z0%+WaO*2W(K$;e2i_sjV(KvwL_zQhJD!L>M_mU+ucxfEM%UB4I*CEE>iQVH=;9zc7
zqC8f^ris)jqF_~uAQ_YZ44?}YQ>rYseFwt$#t^nWYZ<s03F0{oS0`P_t-hy%mH?Y<
zcSb4!Hx^}EU>-AZpPtAcm~_P=M-!x~FMXo)=uvaaVLo-hgz<)TTO)Ym7cFKCr2#3x
zg6=V8)LNPiH`j7yDu_H_5nT8MUsq_v_!;_OT6ZV@=vZ-uWy5%BP$1|JI6%9_Y#oGI
zbwfcprQrGymt6{I>f@(E!}s^X>DUcjfp(~%MV`qvFN=Wyld#46gp~n|e{BVFctvo_
zrWM$9d<2G?9{nTL(NgI3o<&ca+=lE{j~66VSRovE_M&=VB|e|=U<G&_5K{(xflHYT
z@zLZC<bie>EZ9N~o7Aex$qgt&O-Ft#%n?W+3~a2R)Eea;b7Lby0`SF0g0Lg;^lgbb
z4VxT~r+^9Kfx(5SGWbKo&!$MNOwJ%FUaLQv3-Sg939=--MX)MMY|K3ILGFMpn15e>
z@%X5@;S_VGbza6*r~;WXx&rA03)ae7Km*{_A?lHv$1gvFzvu(Kts#sAy)JHhy;NIA
zVT~GuuhAu6b}Yz``i3MA6Rh)475z9bsbyEpjb-UX+0|Y9;{EDI`lA|j6>HM1lMZ9^
zM~nf$YqBc+Y+v3oG<I9LtsrGv<?tUbPQGx8w=D$N>%raJKDfZ@>lLckJ&P^J;B~od
zg%P^4gA?j1)*B3@QLH(cmFc?==V7N~)xgf%km~3^$!&0!s?h2+Y2(M8&R9e(uHE_J
zbo@1yZL&>+f+ylv!20Z^Je+dT5W5QEMd)>?q&d2yO5bi)xUxytECM!oc=*7_E^R?S
zFJ@ge!Pl!S(SH&7N#RrjWmwTR!GTcNL4ZEjzV99GVN);n3Zxm@M2k>{7I1n5Nz$@6
zYm`<z!+zJf=eu{a9Km&M=ys`hd4V>_L4c6=r`C^C50|HU_Ga9J)2Dhd>^{Cd^1?x-
z-lkYEuT{5Tfi74t;c|a@)!yxJGgW=kCFHwxbWXVW3s_pgPRCg=)^wr`1ctMG78#$N
z^OZyo?n^xRftw^7uOs5ALYVyeUXLta0T-4YZOGTMcFMH4iWA<ej+{i*ZyppS5)=rO
z$?My3o=LY1b2is`A3EZfCSH_THRVxv^fg;O`+P!&)8Y*sg)X<7Mm}x2vz^6(8zwJe
zgH2JHo`97LaFIq)@)*2df)2WQcb<TyWy=&g)F7zgi<tHn>2}VUOYR`5rcZS%$$M+n
z*kWnz?(*<G%EijSv1<wWePACRJ1xv_K3u$kGnEHddJJdt975S!LYtyYy(%}<K!772
zMuyKp(dan?k4MWK8$TO9RgdmPHEPEkzj3cVB0b<XKeCz8(Y#E79D8(3fhV~z8o3?6
z!^<s3J}#le$!62TwB&Vkj&qj`%o1g|LCa?eC>_`a#B=Cz<YrF!ST)R>Cmzyu)d0!o
zdHXsxZ^s57L0$*tD95lNSJb9lHUm)SkRKJ+gHs{%si`^8kVw!b?l6#{Oui!CTBfS8
zm$KJq)|}M-MWx@p>AVNFRheX6UvX1(^<>HX(YGSbwj$w9Bxj$cJgl}Rm}8gGyqZ@%
zk3aSCOIyJ)%b(6r6&Wq8r01lo*{im$FHnig1(K~Yf&P%mvcEZ;dWOY%a}XnBuL#Wl
zi<U*)MMU^un(}w2kbcfR?qr5iP+Z=v3CvFmt`iLV-_aW7ng;Q;+rIm>_z&`Pai0x+
z;*|x4#VHb<Qm|^O6Hn)1`H1<4Ivt`@P9=Q8k<!^qKYx2f^tGW+?`0IQvy=$h?Ttl=
z#aS?~J;gl>o`#lbQlVV+o3Nnzt9*{x>UEE<<RHRr@T{X>!+O&?QPq|ljh47yCmqKo
zziU{^#u^lVlJ9(OW-(Q~P!Q>Hqcjv0B%<83J<o7&1!CMa-kD|PMIwF#>-a@`P~${w
z-RsA-esK#eDrpTk+lqk@{!!9^RgnVKwlh$4l&upr+@;MCAqK~b=i-Xa)tM$`JY00A
zzrbEFz5MGY%kj7hvK{?6p|JBd3?$pQjNCLw&!V|+Oa3WU#g;x|pWDY`AoB+djLGy5
z9^d2S@F{Xztd#W#C>ZDdffGA8tlGE?XT49|YH(oPTuXO)F&<jmYhn7B`!Ct>@vvBm
zZc-L69l{R4yCI@*cBY+x$!sa<X|vkfGMBwzZqkdAx#p%=!+hvtXT#V)#P0tUbdbv9
zwkjHsDz0`D_dd)_ijP~F_A=jG1dQAk`>wrDG-L!$c8i~7IFt&$j-7^KX6g6`zOuAt
z;wt6cq0Q=W?J|{wi|G%W<@wW3FgMdz#g37X!sz9WO~-adrLNMEgH^pwUVK+nk8|HM
zHkYE)<zUc5)#X1N#0MPA$^PU0dWP*tPpcHa9@_=msRZckyJu06v%@&e=6M7-Y4rk=
z-+K@^V7SBfD_Op;97Z#P+q4;k4aYUqQCk&xq)hTlAf@#?OJBIX2`sI}`W7Jhenoo*
zA}WH@EgU8U!(*mh8&uYI@G8@$)XjWbrtk_X8!lrlsqryzChNf4ll}3yA5v3@bX@D*
z<2F3kJlX5(6~qoj#pO7yzxj1b&qqS(<s{JJ-rS$Sgw@dxTZ4HWi&jFH40L04f&$M&
zhcjo6*y*w|QTFck?}DBH+#B^3?(1QXe}5a)=_p(YStTk`q!Z^2yKYZNA983b;=8dZ
zq2O=9!NH!OwQm1qA#M5G((%`ig{~Mj@}QxFA0^n=JzfIRwUOS8dtE+ZBzgwWWez4B
zdu{pOx&d5v<7Z^iM|Tsr`oG`>OrWfoTH<>PZ{h99@$qrp-AK0-GtV2Qa1-lPz>k~k
zFQ0D9qO)bh&V`anjskC$iu%x=ybashU)h9lj;Hy-q{`JW`YY2$PrQhlzR02Haa&`t
zn)kpj1$^EQq?7Yt_c^|LyD~s!Djlmt8ocpoqJV3~Z6AZHn2H+FgSHRz1q*%Kzcv<9
zkW9uqddj^8wlJ;9VpS0H>ve4-+hB8g=dzmxi45?^W^DNDy^rhCG9Y9+xJ4@a8MvlC
z9s+LHv_bFkZ{3#c%6NJOaKH8M4&_Hld0yrAFd$Dj>qROf1(|x6@<;f-5nX^hrjEe(
zApw#*kpEa;RtrvOu7#5s&QiHmnkJ|a-8pNRo8<PR)zsF_k%NdL<H2D5q(`9A_vr>0
zik&-J@j1%z5Zra|%}X|FR$VbuKs;Y3#P?vn_#CwEZ(<wr{l#Q-u=~fLMQ_TVpdg@1
z`sNE3+CQpmMfV7;$zjp~=bb<z@IzB%Vefth9z@FP*j>?lQkYKO-@JV|BwU;qyZbtN
z;#*H<DUTm6sU?C^LN*tx58(Wq^TX-%9#9gK*mb)WC7~J$M-uv-sU!#W-9QCTHf_s!
zcb?G#0WUHJkQ%JCKgdM+bI6a|1U7mRi+M`^nH)y5Xc#9+6SKgXfK*eN;zW2haVKrq
zs+Vi*Ww-(-fgC!QDGH|rW=%p%=HvBOOQUM_G$Sme$yrAG!q6_rG>MpUfl88Jo53Rz
z%<W&5n(?hZky>|a1CkiI03C%h0(`eF<uPHgq9r&U9u_F^)45l3La0U>1<NcwY20VP
zurx35N$C*Jn?`Qu-m+^cgFv8zT|+R}zElM~X6(AHClLc=*aUR`t-O~jEgK$$!P1}c
zMEpMq--O*QO>YhR$_A-CwW?zQ%zZL64hiTq$zx4nLv8Her{m`gl6V(j@?o@|#?q`Z
zN%;JG7S?x3P8~QSzZN17LLj*3>2qH3XBoo`r2t<;hpyN>PTV}>*D?Zm?|Y%n-!M1D
zN{t^PZ#ae$<17?2FK120C%eN%Uvv^5$uPr3;iYB=H<0n)0)?W|V>+FNpTE9@&;~I=
zZ{LxI!T9!E#9aqPO;uXMnLQd#exZ#)tT!4=P=2f$H!~{IjPDST^CK`WN^tWRMl)kj
zpT|E2k=bMeX1uIt#|3%nn{y6Q!t85)q=%(w7>@MaQ4$`*ka|Sv&mwSUFljfT`R6Ct
zcWZp9s6_AQav*bj!NYm;+ij=qbR_6F#}<&mq^TXGZovkt@Z&N;7ZA%IK;aGokG&nG
zwi6=d^5Zlg0Cu?dl+74=f4W^*&^p%n6%P90tN%0~q<G=TNZ9)MjkbpWAi`E1H5dVW
z6~ce^Pg3P18KWLZLpR~$hu(qYX+hSB^(HArc%MO@a_;a4c=T(+Z7O`Ls6Zjt4YK+2
zmp$5iidBocW;fw3L#nH+IY`lwLDPPnmV8+bt7MeNzwLfIClY&9*mIO4lpm4l>uTTg
zD^}rQ1l{(BV<&}=7=RXG$4oSFd5np`eg&9QBP?oz(=R+iONGHqnEQqhPPw*w&^hX)
z0rMkdMCC*i?+<vl%*lGcEUX$ER`Ov;dN{gK)#f}r$~!+EIeK^LqrpW<e;ezrGdEl;
z*gmr5XLjTL10tXBxyJ2v$4Lub>WAKv4XdT}Bs1F<7A7Aw>lrRDQYsK|%R%{*<MqNb
z8qjoK`PqMezX|FesSy}e#QysGx_eMw<{4oWIsMZ_AnJS-6lJ&=-5QJZE{eP9LI!<0
zok8Pm`%qr{Bt5@CWQaLC*#;GoI|Gbv!iSi!WR8FX#O}l9hkUInpQE;yg0B&`VElc8
z4n}Iui`iOov!}4u7><+BjLlvdUTA#!DM0!Oa6>74@r%!`VE_w_;L;0>Ve>tI))pF#
zxKGjyh6M!R_HU$F<iqHPDp}ZH=c$py`_#j=UF?1P4ID?|V<yFN6#|$<nUM*3P*a0i
zi>sI+{$TEGHcUEE(MPai5t1U^FFj0AKjcSgfrjt76cXUBj;bOi;wF0Gq(its{GLcZ
ztZqZZb%v{6H|v7K5xBDL3~phSk%N}gbOVu*k#+W!K75Xw`D3%){sdv!owmew!=w$)
zjd@%*;Av@x!y*&xDa-WJTRZy*2H)<yg|92KW;*a;%+b0L;Av__Din5cYYqMiPHKGx
zCK0Gtm;CPNcA#B>CG(*C3?l(6g)S?kJ<Sqd{)1`bKI`}S;^YTR3^C)5g!H${8>0vF
zBy9wuv$<n@!9f`55!MI^$3i^)1X}Q_h0WaYl<@GWtJXIO=$%V6?X0d4$%!LUA?And
zeD7g2X=JccY7?L6`APD+Pr%XAic~4;r+4;?De<uRMCVF|SzFN|BXncH(vuDSI&nU>
z#>dW(+IZf~a$bH0dp;fV1adR(>+szbY;P3?=n)79q=(>PwRCB3_W#li)!O4G6wrfZ
zLT+bwPRnfk9$_8DA%3^&M7KUdlC7f|t)V)}mEaV>B@$r-v#AG}SL#-B5K#0WEe(zR
z&@p(I(^}Yl->4nJ0@?E~g%NhcIIg9jBf5SIgn&H&oEO||Fh=$c>Hq4tW?tEQfl_Yk
zI0ZL~9VKAx3@;EiRN721?Ia(gzoM2nzb6(ZxBHXuvDM&@cBj07a!u5LKnQq7BYtg#
znUeD)Vvtm@<0ivuF~sWhV|f6B77DJ>i~j3R^xbgO9{L5SpGP&I8@Yb_mB~@!*R({<
zYcy~!aCqoKGNA3#D(e>>)Ry<(WBm32B=0%f>u?Hq!XutIS;VS>+)rR(hfgrt7=#zs
z)yFb1YXaS;1NZoDm1r)fn*RYb2oUK^)3HB}b4?M<DV?4tzx`_8P|u!&@AZX2T6^7|
z{p=bIhdVP%9u?2nkJ2qE0OtiAjDF?wrM*LMqwMuI;STl#8k78XNHq$IaLCeAZSE^T
zAM00c2a&>#jgQ%rw{1CiEKH67HxXbcYe)i$YZQvf7K)BxJ>UVPXVsPkR$_9`#BV=x
zwV01b$I=}L#vJv~V?u;h)gdJG_!hVzJvnq+rZM5+_0RcrJSsT(=Yz<$ExE%b3LWhf
zCTPxE;Bb&y4*3~2oUZN`y+vS^9D&=y<PBZ6JvU(3jmi*weatv1)HOoQP(=_Q3SKq!
z6?)30jvxV#cD`ddxWVhP=)d?Y%uEsq*})|;xGw=GOfXmh{a=z;^n=A+g-*`EX1M=0
zhjr};1k+N((Gw>pNSHnVfDM0S2!?qSE}V9O@B6hQTluvURl|bS=>-tRuUCxh>}W?*
z?8NyDTwRY~4t-xP@wxf<@(u@QiX5m;tn7CEzAvTSCKq7H3kJo7QPc{fhJ~tQEQc6}
zew4atZEi2w<yhE44sURpgQTP4T^>Nu9We36p`AcZ8WZ&!mZ8%x>G1{#!}o_HpP#UZ
zRsRGC#^9|CEvU;>5|zs*gyRj!&9>SH*9bdNm@|*|@>>ERhk1{aBt1H#{8s`l&^2F`
z8`Q;?ib><gVKR?S>@@5W^X)YRO5V?K3J?^iJsF6TJ(!)L7a$DQGmgPq!6o9~dy)m-
z@3OC(3dpPIftj%l&@<f5qhx2?^;7Ewz+DA*0Anwz4EE2KpMmJXA^}Sr)@h=HbodkT
z>$o!;FMu#vkOYN*(f~6<6iAsC>IN7)N}x7FM{3;hW#)(N91_Qjyc}IAR34$&a>NX2
zUYNI^i}>q^1stz`)>Na~H`=aG{Q%4i<)B}%&CU`|(!LCR*7*bwBact`*09mph|cwf
zQTc@d{xG~CI!9UPg0JAGbFvh?-dP(s?nSMpC4kB&^7+{UHxAcjI)J9JFbjukqUUYX
z^NI;_dX&>02F)~i&!uUwEhD;txZa&KAP@y<aIGSpCm=VBZEI-rfV%RQQ|yWdh9Aal
zv2Btp+iHA3p64Gh*93mug2ZrXu48&AWDkmufg55a$X+fkwUWr$)07_j&t~-pz<>Zb
zGSus@?F=rl{m^`ZIGKl}`LdlgHT5$4bJyt=DKkRv4m8MMlr>Vk4t4W7j58g#)3_0}
zKM^0aBCR3keFZHO9;&yKPMj=4)(ZvsMB|h3nm(UvYKT$6d-VXUL1*)q5OTiKUPDZ{
zs0^?(c#<G9@w+gRrqe0(PVeSH+f)qMJE9xIiVy;boDh&CcznqF1~52s#GlOFre8Gg
zLkyTWs*6=>qYn17q*#|&d{?45-LuS{*q=@;y3mzvcGkdP{1IY+Sj{B7O#?f8SgY$3
znBLWc-%^oAba+;&f>eYahZm40_zq&`{&v1ZL5+B0{=DbeUXU0oWZbS=RD*zN0Fwqd
z3N51kLS5j2=6nbZnBK(${*Bs10OCyBH4O+<33+7bN9l{>H2p+RNmQhM?Lkd``u7O#
z{)BYvLc;H%lta*C)DOv<LeB(NOBBFNnwj#C3jg{V4Kw+N{B=#e0X3}l|FQrY+d&l?
znI)q3B^F@R5Ld(=eIciyFXs#Aam%aHUOFLtD%FAlzo~@=?3Xvo&L5Z`^u_RWW&C;u
zCl-%Zv4vQEZu-o`L(tL;;}f_|w#sf^VA`W9L|;oi;y!ad!4t$u`{Ts^1o5W0DTHDq
zB*aq%!h_2_NSZ-&x+Fm`-?u#sdTJRAco2Z+lo0oB37kP5w4hBBdw(qU6(y;3%ZI2v
z+`R7@WTuh{LXGQ=qNf?FMu0y&@2MYt!%fqWXik^^AlKP^93`;#7Jb0|9WgxvaH4-v
zihvq|k)^<J8jS3y{!_LO(cu&cvHpBD6tEn=CIA`_=N(LFHD%EvNysKMVLfDjYUiFo
zLu>Bt<NGfEUR7_qO9!xHTsrak?yW0oqZ)w>>4CF8^2c7Bm=(8pR}ZoCs;MqKr!=Nq
zica)qj{I1#$6_*gdmQqw4WR<Y^P~f;XL*(ye%a^si-X?l%Qs|JYrYKIUfTmMa8qQ$
za<CjN9d#J~8QG=2ir78P(XgdH&C7F6I=#{%p;`_?h$537)33z}eP)DR78rM$t07hu
zw&WJ@`_KYy{=8v*oiSJ1n*P`yo*pJDav@o>NcF>uSj{vp$FD1Yry2X$J-N7^xc_?F
zX85q6Ch`H!@=53Sjo3#<owOKn3rl}PHNH2Y4=O0FbFUPOjB3}|!PWW(F{rlj1!HVQ
zB_pE)R&EZxOQU?JcU2fR`4h4kFr-bHWN1Zu^h=yTxe?wVyo3V6gRDs#x@>gO+edZw
z#;cxRBw<dw^sDE0bL;O%2ToW3`)&1@S;FE~TBu4QLVnpGe7ulvBsGy5d}X`3h~jJh
z&p6xO3ANke3HQ;~jv+#w8I-dJ&h24q%HPrYQA5vtJacaEtxJg*yf#uHrhz_{2`s<t
zfFltmpDFR-fJw=~u+0;qZ)%Q&4BuZano_o6&<lTIH8UJD3q?shWB{ostUWziYge_E
zMNvd!Q+%2p^}K+T;j};qFT<!l4mAUOM??4@pTX%dsNDy9U(pfA4BNt+H^U0Kpxmsj
z@;Ib2w1<1Qi6&^v+#@8<WuHr+?>>7?<TigmCLd+hY>6J)8Q9g8^8FjkN2tzK8~Eoj
z+q12{%;R{OpaG^CM_~f~FJ3El^m6Ri<G7W@Lw<yIPaZ})H{U^Qx8%>#k$Fr@+rcc>
zyIO|oAl5O1d)M@{_9XZ=KFmu7)?WX7U(CjThgr1>0BR@Dj#Z3<q>2*Slb|_dK7^el
zz=|n-1R}I~d^zR*;&>(eqAl#@Chz#;fZSKT;I&*qlUc@1#g2y>;smEaypNP$FB#3y
zFI&OBGG|FXVnQkz%?0m5tES-T$Il%GN(VL`tzNk{=C2#Hd}a*$)ru{b6jSzj5mV4A
zmjJk~P!9-sB+oMD2`8pV1E{}~4n<cNCjIbd{zCS;EAKwmmhJo-C^3sn$VYe&EPst{
z($3#Sru-u%C5iD<^EdzOk1fyE?0j1zbtvHh=uS%AsZu~JpsE|R0WRl~>)P?%Zv*^z
zQU2C;_q1i~y=Zu+_8KD@H|)zEbP@AxWGIVYjQFJF7yb>=8l#3Shg<&h%W4icVkRG_
z&BP|HZJY2=#>!*-b{qOPCCPgdz3UKC!rrv`3T*gVfF7<vo5*>F#1);*p51N!x~uyp
z+`?TRV!#Z4#@wTzf|nCu{d%5s+=d-erR&oM>(2$kLx7J{Rm8|u0(`yKmdz$7<m+=M
zn3pR+7Wub`dynt?Ws;B_e_J399%==0_?!^y5v9*x+^=iB$0VKcvzg3|7`*bmfMJ59
z(9uf-XlpOM4xYIdV3JqChZPVw2OvFSt_{RY)1gIR!3Dfg0>vz%n~-tR8hT(Hj{@x7
zJW-+qSWh@+7O8fG6*!a@ZIUbzYdow62i0O+1eXq*VrxCH9)|-^-=KR1t!9XbUcmuo
zRiO`tnP$oSR(YL_kP0BQMZW(a-7whpeFK%k7V@p#uYiC<QeMmiFI4qDl+*i<KAICq
zg1p#8q)<f&<9&1{X{!Ni2`B-+(*O+<W|rZAK~N{^m&p7WjAF6^4Kf18jug9iu@l65
zJr#!&`g^dDNc92-E&K(jKwa_#C>V8YJqIB%-yBX-@Fd|G=|DZHRCrJT2SV~xFG8R+
zwA0ib7E+OAsBff<u0jAA(737dygX=T0tI3XF@_=tuSr6di1j==f-gQH9sDuSysX?b
zX|YNMK^lY^ibhZYE0FFI#K_r(QS2GCgtT>ueKg~qSFjwTkaQ4h22M(KKLJ}lm_Y^Q
znzXsXNE!$s{BJsm3q6>wX#HkcKwQkM0IPsl#Ny%|=O2uKwk)7^7A0{;HVX0Bz)7~A
z2%&HoxxdOkGJil!WA8Lwh|Ew;wc(c%B5*ta3V@|df0d91+?BAp<j@@VPDzLeaxLW0
z$yo>>Uos(t@f$P~uaa$l5QrBAZn4z@ybJD;Xd`A215ISuLkV))s9zRwk_MItz~U9`
zHw7`%He$>sh{Px;2Q~5q2YtdsaM|rI_`$%q-Glb~okSi&J^7VKBblQ-q}jyB?FDbI
zH!$!8-eBEa;B>bhXZ({9z#Z=l60u`NBZfSJ8f^T!fzV^~7WLPo7y%0w)ASMY@jiqK
z&||2R=IVr}?GRQ7nQY7~SsnOp{z3-X-sg{CeTkd@&4+u}@83QCxuAb1q!AtsapnLj
z*nTB&gHdCZn-&gBK9~*v7cqPe`Wbg=A0$H}z!d`UWj7NA=Ij=%tq62v7szoLa12+h
z##V|ia=s{GoLSh}A|(*SBIh6I)@t9-DC*><-(H>*;72&&KCUZp2uQ%90@avli|k)R
z{mwaM0Myu-;y|k^>8$<_JM52bzW_<7OV*)H&=5XgM0kMOaB$1kH6El)Xy_-dAr=-T
zdq;a~i!pKjGdceL6)^6gkhngu?eF+)b^t#B2mHYGR!h)ncPOx_uT8+@2MK{z@%SjF
zG`}veS8-p1f-n0ZJ86iuN5E2{eq)gG?4jnB39skp!4O^=fYDR!(Nhhv>-YYL-Owjs
zVBpRZ(C7F4T+e~8acF?+|DfO@vs|q4AVxtGBai%Ct4K-0pG@sIkj?C{A8;7%zupMD
zkFRxpH>zl`x5H<k9OH~a%b<->kyb$oOfR$gESU-NeXOYn8TuZalJ2LP_@4MwD3Gmv
zBCdnLZ$F#D0{$St{;pqtg}AB4;k|?n9%i}Fih(~_|6oR6eiRII&h7Kti|@d$9<oP4
zbi*$Dd|#6{#~M63>MiCkjppGS%Tnm?%_V2wH2+RChE1ENK}tBjU=$#uKf6SPfiS`m
z!uK9<Fi<E*K7%I*xCQHD(Kcu$?Io7Gol7&Dob!1(ytk;IBnoH)GQL4W2HV=@{{!y@
z5>Vh1^p{O*87*1f_<;sdD_TNS%2?GTDD6u;VzwnlQIVfEEdVg!1421_^s*JCf-+fT
z3`?S2Avpq}g;Y%b@Ns%~UJ+Bx<_8$S-aAZ@s<j?1Rk|D{q2kR(2(ggVnAum+11H@S
za<Sf-cfo5a1c%Fw-TKMi=nx+O0+^U7{=LGs*iAc=!$*cJTUmy!o{lK)&6+Qgecl2W
z`A@PW7*;89vI~ih^bjlBiG2n?liSRv_Ow3%U;JoOAo$1X1Q=E=aWV76*|zPpKkbA5
z?5F%*4&Z;u()-UYN!Sm(-Xv7(f6sRO->HZHPhRN%pXrX`PKKWUH^Y&ciGksNnt6J)
z?dg#6ggbV3V@3!`C=>uuC}zSDd#qqs5QHI8j#3J0SdS5f^HijYQqKYBp@4x~I50?n
zFoh_|A<`d&6O7bBXdq%lBt#%Ugb*hHB-uG-_VR`bvb}xzgwtpK!pV2j_v(DgYWVW@
zmA5Mxf*>>=v%T#;jm>O6H#4c#^d^Sg)QC2h&FSgV*Sb)UsJMy9%zl(#)7+)e58{0+
z|CW6cCmWa+fh0yVN!<h~BNS9aV?Qy@ZHQhFUgocAW!~tVvwin{Q*fqwS+8226C*s^
zNPCz)_+S>P?5@sGt9tF%lsTdJQPVlDa{h0HC0bH8TXxo&0o568Mve{tA2=xuOzs>5
zvTgGD2oSV-deaRfsMYl2;=XW?>N(3${A{OfttsMoD*Jr@A6Ro;1H`KPL$XnotJDi4
z#F^C1O^-@T@=P$UpQu8G8fALMSQTFSQ3huf&DG2>Cll<0Xg0r&WJ{V=vF*8;;b~_E
z*prQf0`ui}XV!}Mhs%K|j-E=bzje&8rxL5DF{{j}GKZ>tCH%`NLS?~n7nd`#TU>$X
z#akk{4RY=b+eaB?=*w~=GU~ajW9b7nf#)=aw!AY{WogW~YOE5Ql?&Dxa#NNChAZK9
zOxw@K+CEK{O-Fm6wv}Z0CS3L{Xygd_zq6>{N1XW8KQj>@eU_^#SL04*3U2J^NYjH_
zXBtvn6=JTyn3Y`4Sqir(0y1k%H8Y5L6l7u6en}#l;~toRT~2=?n&r|k2)o$|h-juW
z6svt09jO$%F#xlZE7e)waLmALH@+hp%61RK?379H*t7Nx!tBOz56*hcv6=<Jg?R^V
z`TiJ}ohf*{FuOM}CO;nP8k$_47?f4~F<sNRX<<foZ5GYY&|YRxcIDuy>q)9*Om5-l
zgQ=FK&N!cpRLy1m24XhUUf#(up*1YC$(&4P_{fy>R##n-X=-FLOo^(tXLvaA&|zH}
z2#VBQU!J9E*tsz&yH4S<w?wwenAj|zbR-+gIM1#Fx`3)@VoT$`XcFj2tl_??5F}Ty
zR~vNqHkw9F*s;n+9a5ixw;&IASk<Z6($HQFTGXX%TLmZWf24^<zED!|N*@hv$p?_U
zq0)9Xh^+xK*>-dGSbIl`>6KNOe#b!gQdtx`_a1~Ic&8TM_Z~hbc-)mRt%JCwK`3w%
z{jit=Vo+Fz(l8ICcer==AT`h<kRUgyqn!lQt0$Y-R*b=~DbP)R+xb8_3rZ)9c4gTU
zI{HUu{y=@AEz~ML`p24YO^98D$vOwvs*sz^bC3bs!U&5nu?2C(FydePaI!H!aArGS
zVrwRZe=QL*iS5G{07HW>VPH!rv|bT7IFj0+FC5*SFIg5V)<ZB-C8}@W9FBpcWuC^W
zm4@U5**)g7se<H%tz)feLedJVVs#&Q$`V+#<4kz+QZG7l$cQ%Wd9Ws8Gz(KdNzX{3
zZk>>%mnxzDI1E5iPFSeZOmJcjlu!?Ozf4-Uel3X%?4Q4bd$s5UBz4gP@pvE!TU=ny
zx*+%+reMxJ5cnOsaDy9gJogtUYfs2$<My5gt4?R*bwmW#;CmDm)&#4b^@@N03e=AL
zHu8WPzo9jObsRySI6UQ2^m<+CRJT|69xzN()elo!XpGYBi#b+YI;oE~{?#D|#SKLN
zT2==|E!&iQYyJMBZo1yu<btAZOAjyAQr-+=71CYQ>;;YMQRAq7^SS|pg4=QP;1{5b
zTMI?qRbYmFML*qD>EA{bR}K24`Ay~Y_9_lxf3cWDQa`oKyQ?-BBEQwE3ojHRU(UNr
zO(;a3b*GlYJw#nkotQlcyiSQn7e5fZF7MeT7Xn^yW)PL3vUy<96`jZ9DYyJW(&P4D
zPC2#R@?P4i5fL_z$MOBvMy#fgpXW5ns6ep()mE*$V6Z-Xhwg`fa4O)cgvITJYug2i
z#Vz&r6&E_zr`Ca+E)1++&8V$+W%I;UCm7b}#<klQ&~J0&whaaA=X>b34GHUq>(+H0
z4Auwb=#fW->)Pc9{X68`bs`k359(q2M=E@)cC(;x^5KHa;?|n{dR?)Ue!G^saG);M
zZdK-YD0cIS`1hHcSMO*nudLYWj#Ym}+AfOc_f{2f?n<P1w`*g6g<J8uj2-^~_hETg
zlMf%?Z6VQcWGTej<XuEb?+sK*=hc4f4^#=sucF@WrLOXxDen(Z*ZDfi4+g7SwvS1q
zcE|ZO8yOUg)&r>LI$w0Ybik1l43;z5ug3naH=VsYkA)-E_GSyY$SX_#CMz0|k5%9l
z+tRq+3(pKO!~ViuXYHERaOvDd=Qk)BPu7w1KBtVwk@v%_YJmgBh?b8|16rOiXzz<q
zH#q9?%$=(*0@bASlWWVtAEl8Fqpz(#R~$Mi>L>TQb)j|&+ZO)S>tEZ!|LV}X3{XsC
zUbzu1-xUzk4E?@2&V|7=m3_F0$|go#q3aL5Z(o#9L<Q7f1h1-w4b<`O7n2MI?odK?
zI694hJnO{cy8qxJpe_{>kF5i2+`D#O*Num}l*-;GZle#bKM=5ZgM~SXY72L2T891?
zmCS6?%1A-YhBaQ}?VwHtmY6deYEY_fNK#H&sDr%(B<VFfTBvgeL0z7cQ2nS)AxY2L
zOnn`ftk)qLa)H4iX<G@Q)@V=_xI~#AW~5`uR#NhWuUNeVLcK?FY(+SWS_fOwqDewc
zYP{N=qBPuj@C{4pOP-agR6yG6cc6O?&_4Ve45IcAhQ8q`@gOQ&cGyh=iP|+Nquoow
zxfGWGc~LO3dI@T--sm6IGOD{ngk`#knpD4LSD`tWmbCjwv)UPxNxc%yiJ`E{wn7|r
zY>>|FXkxb?h;F@{wn=B%o+h<MkEC5y0D60!Ns|&z96fBtm)u)z`x%9UKQXBxy;A@b
zZIn(j!@v9NCRWM|R*ZLRFv=@%f_F6|)k`mSwBh&JHW$)k*%0IHGQ{|MfOFm6&}eJs
z%j?I<m89245E@==xo#~W8F%ZBhNKphoIV%ExP0oC!TO8zKsECNO)L@nC!9Oja-nY>
zG;_5^91tpF(I?L?r9-HCwHAVJSEesFxI+4h1&XFFIRe-B`K@aFWUWjuV5K-wR`sgL
z^47S~1h0{NX6lPlvl6oz*-nhvoxG)*WJ-KvYwzR@QF})Gsw(~p{Vlr1Snx0<4lA2R
zQHDVcDEY|HGAG~ELDeZD<c40s{i>0dGDF|wK~3c3%U85fWe@BlCeRBBDEA67kJPhw
z=FmQH8!>ZNKEcF4!2_pImpznBq^X#hPm?_;6-Y2X(7uX9qP1Q>tW+1XU?>Yaj3;@+
z)f7<<#4xY#+8((?`m0jK4IA&tXjXRh`gMnWgtjKEOGWLj3niIh*k%eBUYX4*+03H`
zJ)Z=YRbEpu{qL?ombHkT`QeT+q_itKr_%A!8wq0B1&5453}aGRH9dCp7aXg}%{foL
zy5AF>e~Tp#Ti)dStI4m4aUWx5<rtl0B8hZU>Lul?y|0P3AICL&{^a2ccu`s9(DlMx
zD*sEIwaTb{3pNayyC%U+>(<?@-YrXzyQtkib`<|ei1XGc=1z{X@3U%X#%HrMXv8!9
zuyDw6V7qz_ztEqYyr)_#r1MIVU0%DtE-JV4XPELa9GGbe<+5{>SvqbLoV(i4G_7x2
zt;4^#dH&V~gY{}uC&r@ZDVw`lPlLY7%FtiGY&fg*o#D5jq^>ne_g4I^5X$`y8tHQ-
z{adN}bH5SuCMUQAs~%lS|B;fV0>60GB)F9SEOvV#FpjNcZqxEj+<7j=*h(wny|emY
ztp0ZsaiiVGR?s>h-ub>XyOx<bF7HOJ*?{-FE4(etw<gee75{Sv1kdB6?4mMv+1s9{
zzjlVCXZ3d7*U+b8kBj95?kC@Pzf*e`_)3@A?(?o!2<*?meBIE!IyH(cTsMzLNzLJ0
ztx9(gSOfS3#z^d1?R^mNN3Wx7@ORwxY>V7%$6b7kX5=aJIEQAJuY*X9tpPd{KD=j`
z#%5)1TN=zG6&IrXXr<(a*)_SSip@t=JMb{qb)DYp!QQG`ry6+@pfuLw0-{`>$zbS{
zv9|OpKJ)b=(Z~LD`FQ=vb6&$~QITP=gX3Ja-(84Lw5NfK*|+L>32W&3oO^VrKEyU#
zd|#`}7$aiu?EPoU4cx^Fgjo`vQ*zg{x7Xlhs{YjsEr(P0<8<jgb18Wo%F*?9>%Sq%
zA^eN1EdVhH!VrYN4nN?LLXv&|{f~)-<-Zh&{hvxY4kniWaV1@Bnza2U1I*aX7YZL&
zxP=#g+=a1lI1qNL0XGcypSWx+=bV1WMz?&wX4>_{au!ZbMvL!pen~|RO;V0GkI?N0
zFUP0rU_wa62&94J5#>RX!4|>6KPC+_er3^^gd{ETrhM7Tfr~TBKi%spvU;pt>vbdK
zw*$z*F(r+zfytc75ezHduRh}!!ihiUw4Bl?Oi;-yN(>~aVenkS2R@0!`p)J|<~Wq6
zw|;Eo7>ADoYE?bDErFJ#5>lHt+FYa(P**ub^(R85>~pQHn#rrapYE!ynrO>ZNT`7~
zn8$oHd>J^QrYaxYfU5!bBtW&0ta{maDW2DKQAq`#*(ByjNoN9Db@U`HKu}+7lj|Wg
zCmOq1oy%3ish|cnn4pGHB&^5^8Y)7dOt+PFDd)5mbxXdfsUN0z&b0)tVm<bV(m~WE
zstG{2P!Ftg&rh`+?Rs=_!CK3p_>BqWRErxSH2PTSqVgLXC!BO>!cfrhijEB^gG!-B
z4TUuya1orV1S@CSV@e1FXx*H%0kBrQD5`w^&0Rj{?-wjl4ls=W*fn>^LOeTa3tjzB
zXGm$t5Yn9nvn<G1{;GtF1HNQ^d4xT0+uEj$sT@imob+UCkRgfbh7eD1FV{1T&C+K`
z$szCyKd=$T<@Kw+^#)GRWUVTj61~IJwm~{xg4AxAv{JIEHp_un8j;mo=7JD;IUzvk
zmuqjX;WcGo=JeVbs&w_WI@vm9o0E=jiXo(`D-Dl*1s#Ux37RC;VVD_c=ndR!!v(VH
zDlnFk!mx1%yNMBQ$PcG?YM{?JXsCX;>Y-=S(!&X2mMj(U8!jN^C}IF%uTHg2>(fCH
zOomHhvBO2y;h3Y`cj$8pG`aHZ*CWEU(v3<_-e0vA%x6l)jYM7}+M|e(ZE!NseC_B)
z-&e0MB}=TJJs(XSiw0RL4-v{9`X0m#Dwv={$?dMT`1WLgtMkr=EotVQ#t7y8^XSYU
z`Er|ox4}n<y7kC4B8t0=6)M47Vq<}&@>v27$Gzt+-WXU{Z?-0`o~V7rRt*M<jUyiH
z{Nmn<`{90OtoF8m0j?UsRTV|_acpaU2aiV@!uap|kdfiPR%-sAjmXHr!TLXSETpDS
zxoxr`guVDd(N%;Z=s+ZBjK*zGS?1=5mFt%1&ZCeL7E0hIIpwz9B(y6S2(0TWruiLa
zW5Rp`IyT2UJTA^$ot$g~4=^Q3Kv5OV&>0aOL<(^1FPLHbMzUA2MDFf>yY*=)#~sGQ
zzk9yiIXeHezjeQxKU{Q$hv?S!LmFiO8t4d6vPRg|8;V(rDU%>0)$C_V7MkKnYG6vE
z{d6R;%xb6dtCoh@P8CgVcB{)?dDM$UP4pAT6p#?7jWr6$SW{~ygFxCPg4C^uC}8fg
zjww6Wwq?C^OW^m7*Vt))$1>I|3mLE^MkGvwrOhiCsIfrDWc2gJIH=yR-g(q|xi%)Q
zb8sW8*X7IVrW86bh~{+pN@lD`9E1uR5<`T;gA7hIs}Mwx?OXc%I9W~A2D@|t5FkgC
z4A#sEpF>44C3&n5`U`G>B>D2-PpP4DMN2)IW~&n-5aJk0)+JYui!TTygE|{h9K#U^
z0npy1o0BDXm0dflrq|4thIH0QpwX|!k$@;>aumUonTYT)GfvBQm@R!3UfRfCVj&Ou
zeKv6Bhv`Axa?(uT<b!)JNBQvKOGh2P)ck`VYiz?li<@A)(S9=WdQkaSytJ<8YaL24
zhhSdo%L$!FMto0<vRG>~Hk75Fz90}OR11_!T07Pb(G`XE*1ojoSJ8GX-M%%?ogvp@
z0{z8m@!7)M^n8<tNFjX>y?gTHR||oX2D3t;%P_FgM8uC}S>(yFIugpCCznc<cEtq<
z?NCc0(O$2`@(?*t+c9lGA=u~3-?8OZYWY=ghnOR@$z@RQ7-KftMEBXWfzM*J9TQ)}
zyUv~L`6}bA!R)%n;#1aQAy($bfZH!NZCBg)YUBQC1lIBo)9MSSnUM&>WdE6Mgm8{r
zv&j7L?Cq~a&8h$OLoZKI6^L!p;2-y5cGylZ9Lb5#S;^evu4Pv?c~xbfMSSS``Fm0c
zQlAW>)e5mg<%$+6hfH5y!!B#iq<t=uH);;zd(UZ3!9r{oA+?zuJA>_tZ|~a@iIlR%
z6IzV~w70z@|9~<6Z_pdIT+>PajOpU_nHy{b2n;49Cyt_dV1~cqPQ71V*YAkW#(X+N
z78L1nYma-^L#L)kt(y159MLTu7ulm62Syvsw>+#w`DW!K;L8JBUR?5{&r;a}{*L!h
z3lTm=xESzm8tSsnP<ZU^e-+YI$k3@cEPRCA1Ti)U>ETu-L4pC5(_@DX7ng|gNz==&
zsE>oA%}?y&lV~!J39>vVIM9?Ze_Rz-@avi)L{wy|ZzI`?FsI@|5*%Hb?Za+tuA}Qv
z_N(A72{RA0+y7ZC(C~3{*v*LfXOFO~RaVZ~Qz>T^+IS@<?_0Mb&I}rm#mL`Y-;*VW
zZgLqb_mxVkzYDS$lZHKszqQc3vtRcaT;m|d^suXZbnk9`g73a{c>lkk5j)#|rv;3R
ztc?FtG^)vvb^Z@Dy7@-!RONygcoDa2Fm^9X9?jFbE<F!SBM4QFQjwHu_=OJ?fIHy=
zUc48Qsix5fqlS-bGLOd0y@@*X^yp}K_8eIlrliRUBVI!%qIQvt=m{g%H|s1sGk<!&
z4+r7;a}h;Cgk_jn4lr7?D0zNezmB|_ivGn->=#eh4_C)8SI6BPhd^GZ{%mEjghCKM
z1qtkPIF*Qq^&APSGIq&alWQLBH0Gjm(eqw4SEbFGnTtuu5NFh15^<t42y+ku5(=V>
zZX$sW2U3Cp%ps^`fnJ4J(4nx{)wq^2v9dtdWb`JV5`vjj4ceeA175;dqy(G*Ol|;?
zBeNIP2Cj{#^f43|)$-9@m;9MN`PJ==ZokYESILNFS*w?bT=eesXBGH}tLC2zCX&Cz
zB#Y;C^@kd{yr`Z{zOhSa;!1OA6~(sr)A{t`O_BT8*;GP(WLPr6{@9Xp2qb(vYKoKy
z3JS)J6_cW@3<2En>gw=rO$A+Yw-);GO0Ai5<XwgYFz?v3B*_3{B8qVcD0T$O4O)lv
zp1qCjl!mBeSPDrU1+kJKabzIV+fVt8Xhnz+CgQeL=>iQ?6b5!$zE|6dZ9}7G^*71i
z#u|R48wohx!my|EU0N$+XYC;Q_s6d;`FpM!Wo%h|MF|MsvgVu26lW8Rmpn_IYRJ4d
zh4tc<H}elz<1F?)vD+Cw{T0i^c+VEZ?&Bznun3XC$NZTb9+n(_eNMX#pYtQCuu#I$
zC2JM*rSQLiOO8$y(;ixyb4g>hW3x2tIWl)t6OAmp9P{s`I!Ld?gxVP~Lpmz>EA6;!
z9`EO`LD{~{+C9hoH`Vx}lGpkuQk2r!I`NN1L3&5I;>&l6dqwX;A}oFJBUaURoJAVV
zf7*vW{_y=XQLL$Jx5jhcBi})Cf?aHnYv63gvtLgc?%EXnUyQwTkZsSlC0w>`+qP}<
z6i!*EY}>YN+qQYiwry8+-LK#K;&tC&MBg15D>C-}W3P;rIo6nS%pCc++Gqb&``A4>
zP&)O*)^6;}Kl`L(zOW{wl2Z_KJznfH4M8|h#QR6mW&f<vrL$Iz5~@P;4<d`T312z=
zz{|OJ^8`#iuUcQRq-uw}BA(6$g{$j*stz&{hW4m0qz>z~Dw#RO_R#H~e&Qf4gm%L`
zn`5CkV|?-NsvGJ(^+ozqwyFd<DfRbSCe*8Wt%Pr@dhM$Tqt}hiTT9S1{%k}=$-GD|
zK1}q}05GhwW!FdEG4{NafpW~J4p$Q_;0q(M=qh<jhHl?+A(L?YW*D8$8FhUsm0AO#
zlA$%Itrp|q<x#^)-%;LM{T2`IxJGD<MMDV)EQX<1T6wxQp9VN!+DR^4G<IfuRs|qy
zF_Hb9^Wx+UfN4*!N42sPQi*5+KypY+qeBNpV90Uu{U?oS>K0g#gFd=*qHxQ>dA85U
z$R~W>iJqs?j(q<*pPNn9PFVnS2^SdOq0dEeSP%^Q9q*p3<)M!fL$rWSb>jU1#Yd_o
zQCKf<;1;_D@Eerx;#238)si?sBCc~dWkX<*S_h=t!uNI!Z~1f|Y_6?qNu6SOVvo?o
zOYMW!=N^FFpHSi!pA;BK{onEIUyc<2r!4eif}!ka1w}7Mz{tSxFJlUOhaYDP7PkLi
z)L~*{`H#`AOH0-Urwy@dre=q6%uiMgl^;rjrncE|M8F>-8NP%1C3g`Hi8Oi`BL?`o
zXNJ1Is1nt%lw%5@*}XkP$5W?rW*Nt>&v;}`6`ivUL#OhLh?B_qj6$6Jf@H_txeW|G
zqC$~jylj-tz>;DJv74X+dSAm<j!r5HPROA7ytP2WDUi|vb##a}RuBUVdHObv<pM$(
z3e2H*76%5y7*rs%^kv{ef_a*e1T2%>b7+x~&aQ9BUmEdwm2}5&s7UD+AR3i}WVgV`
z^c&~~l1>yD-_}p57jt;wfIu25yLcx!y`1_*m-X_dk`jp4h-~~KynXN&E=vNEf}qJb
zX{!{SDn!*qya5`I16(U)KC_??>e>N{b5kqLi*m&TCOq8Xu|K#vjkKyA6a?U!f<5V8
zAF47F9?u?OL~Rui16n02LrN|;MTH7aF`6DQ)&N?aS$S__Ee_z0#${aHi7S|?dnG0`
zE^x=Mh3XwL<Bs?n!W5lAJeJZbHij_RVPSLs2Z;CC{^p&Ra+8jDORUI9gMOk$e0`Ku
z2#Y|SRFno5MmI?Uz9v%NdV`<MFsrKmqq$(KU;Ftw0h1CYqGORsuoC_<q5a!|oCa(Z
z8jL)R`jV~G7QSUB4Rd-L7j)slq=qzo@i&kTu^8i#oD}+f2E#Z;YI^A-<rxc+h#=Cv
z<PK?O4aSpwwey?o+pUNFKLLvkoVD8#9dqY-lUDY2=2p8+&(Lzm;S~M#v$dPkW6Gzs
zn|mworHwrPO(vUdOXrpkhbNEjHGl8pS`h!GPj{QU{mTKJRiIwI%H2Bd75|=?SuMPW
z9gn9cJbH5in|y{zSEn|OO|*Io{?<D0a!>2w>Cm0!_e{q5=Q+Ck#@*knX#8jDb|1s9
zuMqsH4x|T{_uI{mmJa%uQQLsCI<lOoew+}Oko}?eiu100zK_Ml(?<R+uCtP!Z%21K
zyLPTi9ZPiQ2?w)*G_ah3e1jh7z2xifkNNGV<<E+gZ(sfmpZhb{*6x*@4bnn>kE>+%
zcX@Jb;6pyA=g^)~ySNZ6^tT$7ug%L#Kzlk`I;`v!{h>8@t4|u5W?>(ZvlzM0If*yI
zN8uD99VjQtg(BOOejO+m%8mcC6YfX}zwj5&utnSR<8)#A+tuOydGqeT_QTD`5(>g5
zg`b8#r6i>;D|NUQEhV#)vz6qI2JOqU?k2rJo&@7D%VP(+r|s)}(F3FRn&)!~%|;o8
z(X9}5kXVDrpVv6%<xUjE);}R8gILd)r}dZZeJ0H|iTq(EUNcA>LmBB17H>VJ4C8y`
zMgYY%vk&4R_=bx^>!Tf5WTsdBtY#g~n7#eoFK`Ak4~+coBA_QgxnPxDw?de$Ed~zP
zgORf_f;xASwFVPLLBt!l1-vO5l}bGzmT*&5LF5bnh3MqkA+1}##`CsFTT6V}(q4wb
zB+M-f@$yCKGE?=SmSr)toGKLm1SvL^y@6gWx8d?$*@bZq*FQsN3Q8a|h|!Q0{<*mR
zsO}Pk6xd-A^!98?uB*D(sl2&`(O6!v7CH!)s?!$2WR+4Xg$!}lxD=ca%~a-rhVLtg
zaZ2H*1xoU?qYh4`F_!}r;Vv_<A#Z_e3}_-0a;8<pRyDAac$)-`3Z8$0m2UG-E7|_S
zoC_$GQM^KkH!dzkxIK6R;~-ZOn=I(`3hpibV*Udf{I<fEOqB6r^yZuN6b!$(qJY0H
z9+U26b>=&d_NU7W*TT?)I_OWVB4^u6czh!f9gI^PyEx!710E<Y8&GOHOb_l^zGXB}
zbe;)DvL$r8W*wB7Z~hk+=$g*1zWB##EG2E77`d@AG6Hx}bq>ExQqKxPj!mE;1w6OD
zp;vb~YI1bPLw4i;ZjKK{!1P_@RX^>sd+kx-{eFS2$eFT9x=csvQpqAEUnks|a;?ZL
z1!@1yIHg!8-PwGt$oyX&M59lPW#TIFt$M$12joBn(GcsfW;EukNH8TZvhsEA@O)Ok
zcnR7*C|knY5X5MVW^jKlmfX8MkY!X*hJ&>{p}i;CE+a8NI@+{YH+&P!2kUilMaiWm
zEab5M0w84pMNdB8LrBmN+oz~2k*YrnZMka8h>@4)up~(I$!iK(rW`%CAlV$fbzWVB
zf}V9E_>~D@a2$4+XNK?Y0}yx~e+EwA5+j|2KABl!E~()RU4Q9Vf-00opSRLYBX<Tl
z)I<<FP;<3Lp6~JGo#|<w@8M_(j+$$uAJ3deTiM#yL{znqCqmNfw{yr#(BXLK<JjSi
zZXHEY@ADA1*=qbTY+MGkOWJScn9d+e@t}jr*23P5omFe8<W?I6St3=>xdc^LIKVW_
zgZj&tHD$m44oxa{i;O5AXAxj$$g$+9hHDIlZekv-_5^xJf|)VH@-vWZ;&9!|$3jiX
zI=-APxJO360+?8fFDS24EN&AurhdZp2L5p3H_8L99Hju%&@Uf`tFc;N;wIa2F(YcX
z!c@R$NEVb#G|HYih1D&MjCr`nI<NL93OV7c38G6{&Uc0d@ic*u)ptIRMMLxI#uQto
zg^{OIw)k8ZYFba*BDLVeJU&Mt++v8m=7c_7_ElX((JPGI)L82&t2}tIw2dc^T+8G<
zJ~AV%mnx9;L=fH;Z;VH$%IM1huJl%eBYknI>d>0DXg!$|T`CJW4N=HazkJ~-89NAJ
zyjX)~{WBcuC;q=W4EoIb-vn*mvVmT-yNi%zGY~aIzuTUP9e+`dk0Vxz=p}ZXoAXv@
zh^U>|9bQ|4t8f=5IY?0-C#t!bC#C#f7j-wcq?CWSsH?ddC#joJp@V_L#_ad&rJH-4
ziAQ<FKV_2&t3vuwFW=i_G|Q7K&C+7J(T0-+`9Z;PoBgOuci`q(>wlt7#-5b&*G+ME
zQ+vM%Y?GQYs?fmj1cMhbFW@8Ki(>*C4PXw5S8BQPaJmc!5W40_S(s#k&m`_gYo#Rq
zJSrM#WC?y%8l}bs1adrLFjY<y_^2_#V9b*8M8i~Q1;=z`iBq;jk%nHH;aTysYy{t}
z60rHb#hJ0%7Dz>86_|zwsHvG+N)c0^TGmuI>V>l{&F3~PV`kJTWGpw|W3Y17^CTz+
z5_`#*T`FcIT=tW{_+GZqD~v!42Y#|8b)lj*({PIP66O5d60)-)A<Hl53NM?~TRk^3
zJU1+4lPcD`%YsyftiI8@r&q{xKca=yTAs!jE<Z@*9I5v*U(>|k82x_72?pnp1fsOk
z&sl=?FNQ+xI~?$Lj{y)MsysYeEAk@yD-Z?BWA>(COzha$b$>=$0HZlxwRIzI5GH+a
zq-YENO@WQtZAexN%gQ)Ntyr3wG%<I8KkLyG8pOyFQY7tqY(N4=HPO3c`ai$Z7&m6|
zKfTk{ZG*5l%M4?s2`GaVShc-Ze>#qy)YXYK!<Qe?G`lD&x|3M+s9~J%^A&$V8&r17
z{yTqVV)>s8_P?`(Ol-{mX*`~!A#1bGhR}7a4skNXmpWk+E<m5bLix+`r;>bKx`hEF
z!+BvPo^Z^(zNcF_BHl7xmouanl4s9@+4J56C9@JK^TXYVx9iPTyjG6oZ~VAXyG*JK
zND72QoDBNdcEPE}?cB(@e9Jy0s;i^Vvgg)@Pxh8~Hecj-+Sk`^bi)8Lh`0<QzzY7&
zo=|nIxC~h&Vj!Ts5UNy@r9XsNU^IE40`FV8sT1&>!LgH~GXj)(P*}h$jN+IUNWWy@
z@F2J>gt*cJ!{opQiZLhI>1efV7OGo7E2C4|loe_pNhINz8N8HXu9ZsYIQIg&vI$wv
zdISI_@2FBHu%jYjt{GrXcvDsOHdbqq9cooNf4K2IXHO<FDHJ84haA;#SrioK*t4zs
z_N78R5sU^d<0go?y+|bTV3|FPpT?OXB$j-mLV-Nl@5{+?_1#_CnyEh*y0v2Ag}8~K
zP=raLL8Jts0-z)+J8uow1*+z3*vewp_<MQrz#LX800-%kCn#&`^NP2Y_ZJ??HCU7O
znbFSYtw*67=%oWP`>C=jzGI&DEj3Q&WU?ksZ0zSxQuk2Zqjy~4l_W@!nFRR13Ffe5
z0$!mP_>$kCKN@IoEP!3!Esi_7oDIzVe@W0AVzMKo=*((1cJ$LP(X_{>d(M3TK^`sE
zQ?PHULn~)s1SJy4jsgU+QSAuuhHUpzF*i(iXE28ewXr0Y0f$MAf`ejQTJfb$IbYUu
z^75ieNG|0}=SR(gfR;G;ynQxLvmL)8%k;^?BZYe8#y~L)c7v~+&@Tf|@8U>Q4AcBx
z8`jUf9@<DDmJ9aH$OTY;+xVhDTPmPsY8a1zY9}oB8jtI>fS42qH7l<=sNT9d8$Ff6
z`6Dol1<le+0r?jG+Wj_UEzQl|^hngxP`s7&q)iq7tgKZ2667abG?K<}!0n5WT!;Xj
zZ?_IIsl1wY`S!Yv8Cfx!wlUSoW*O%<ZxnO9_FD)YAEu5O2XjI8Hr8m}{WtZxY$b%%
z7+`+JX2T;J5Fput%Zz=>4DM!mlIexYN~O{g?_am54)m~Fvp>C`mR9@wfyJS?k+hgi
z&Qyed$oea_m~Ew2*r)QaxQ-ANTG1lq1%cI5d~<<r?dZGdos~Pa^LdCA!rx&9&Ptd4
z9&dW1UPQ;&r%mF31+Q9vJ7taWR2w??gCZ%x#}Z>a45^M*ryaUJ)Rot(cGtE2j`^NA
zE5Wd9K??Ooe{(}xPhn6>guT^gu=vYzFc4gkba?-ltGL0bF5PNkkJ0{X%YkdT!XkEo
zuT8nxpO9u0VSS;ic{G+5-=w3y*Wn!&8iv1)^TImjtX9=<nMR-YRNXmfUkpx!6*4Sb
zKfgL%m`Ho2bNsADkA5|EoO|L<{BoQjA3TU}fIo=56z+9G|LcC}P6i6Y{1PPsx!&j2
zk1?tDi*yIny4{alhlsJPB?jW;4^=c?+V9X}LTvAk1M3EdYh3!~tS#_;qt?(bJu;9<
z0B6U$v~VR@>Az(}D*S(*s<|~+Y%u0Kod)DW`I~`v2$etaU{yPc1|0P2*a?o%;s}!z
zj<tWx3P6|YGFP39L&QE5ke{}S_cEFs1niaUFQvWjK^h$43RuK6%tf3`wfRM()hRB!
z*;U`a8cw^`W9A%Cmu|{k8Q3yV+I{jHb6Vyrz5#nYrTG3GKbikoU18z;kMUD|!u|(8
zJKw2Y6U&Kc@IVA`tg0?J4#la|;)UaZ)$)xNlnTQOyPA8vL%;--7#i0yLF&a$vc+cf
zLKCd~6Dk@a4p&Pi5`Q2^P_%x7M23V16{3VkvpT_tY{ukm`>sVk1*Ad|<q&5svJdZC
zaH33x{1}}*;xy8pr>&>4=l;FpaL@X6_R=;O*T|Q>Ad)~7l7?hFu@*ZX1(7EUQE2ht
zx!nAFD7AY<wr9cpk<|wKz3}{3&zkZ5ujaR*dQgb4#KFP<0wm)8R&r^Q(*Q9Z&jpxT
zP<%54@qId*HBYNNXL->-Poma2GHuhpW!fI)qc!JKt=cA}bKl~Si+qK^rC7wqK!YMn
z4Ijj-p(7qEBT70s2Y;&`t~C}@IUy1YfC$D<z{8+$B9QLq`bwG5h{jW4dw9WuFbpT$
z8)z*Sdu=$)#+%|aOLkc4En|`jtbGR9vSd-xk+<Bhw-r6-y1eHj(+Wi~1o{cc%MK2X
zp&$|QqJWPAu>!7rl)Dt0Rs1z0AF)zxzUn+`rz^*xOOYsi0I80F@+|{@g0!HI_S!iA
zb$!QHvmZi;0AA;AgevN+2*~6x+LeFpx((3oZs7x_-H%7y#*9Y~kB#bE0{1~hf4{bl
zSh<XP@l>{y1$a4Ov0W8|s6dGg5o!q1Ug`_r0>YPZ+?gG3SsxVM5NcI5F^nT>4D^I8
z&`|mfDh4y(JJsLTb}r*)Rlc^DC0y3UMh3OS83?;rY)d05sz4S1N8&EMefIX}@dCwr
zfh-$UEtJ9=XjB|jVr$^O9A{AbP!Yp_MS>UAri_lt_cpkzVl9rJO?SBX-fruIX?lV8
z<S*-vQ&rE<I+Y;KJaVU}O7Db+TivPB_CW<259`sv(h)5cWPpX9FeW7;E6Pl)il!dL
zDU1_5Ydy1#T2c%`I3d{^ouC~HXX`LSa!A|QmI#04yo3^^7-bweftQN!>7}Ec+ap5e
z_LQaTD>BUv+ITRWQ)EyCl!l}AzLM<Ml?jU3s8`5_>UpOb(7h@@pC6p?XdE7m^C-$V
zH1i=G=RNMQ3+yJJGvvU6f;d5;5DzJ^5Y|MJuvToJZ0uvm$LfidCaYI5IfCSPEWrmz
z;m_OSlknRO<VrzTxgYSPzw@#08d<O~Hq2i)6*kjV#cW4pJ~uUS{<VgZ5Clx5;1b2?
zwt7NV7ysOX!N1(jZ-|}<1}G4<*!>oN4a!|@K0OI^W69m>ja!nv!WtK!k3yPXFXSFW
z3@7*?Ad=D<bNj?APyL51Tt7QK8aRtrO>!kfqeTDqr90~#uWj~?*i7;4dQ!gq4p7G!
z+l2;_BICQ0NOZ;rC$0kA{Nq)yuDeS834+{$|Ez5{G70BkC<eUobw2n_mu$KXcp}SE
z@DE7C`VUNvo&CR3i3I;|3*4gQDf<itl%QufsAN~ae4kmS=&VTYYG&jF)p<}xrX>*I
z0K2truOQOUsH8k*9lu*#uj7;K2w&YJ!RuQStnCdx0ty~Lkdcju7Q9`&0vv(yPL4<K
zms^s;uT4kJ9FSDO@6$f*>m0RPPt+Q6{(YMk_{7W_0pebve83V=zC?=%>s(`Zz&;qg
zz)pY=0;G*ZeUKXx0U3es>hd$l@Q=HeHS#Db=m46QqqU2|w$((@^0n`T4W80*Vd6ik
zOGl<kg-0einM)HYT4hbHzW~tV&<S85h{Ob{w~VWujnkg8g$RT~f5}Gz$!VBE5}(}Q
ziXb2eJ(RbnK30#8sTQ&;X;mbsfUSk9HSl`sQ@7RoRDp|v5RCJK5~k!<TjYjXh~hx|
zkY$J*ev<}3Biu1#uAo3V5(=APYS^@xIFMP=f+Yyl&gPVnP&@7~4BPgklsV=hfTU*S
zr8a<SlZWeK4Ffm@gO_?mFkQvSI$3IKfC+*WAX=)k?ifsT(MPC5*g03u>}xkYgQF+p
zE62K%xiDER`byF`=t!FHe^Zx1s~QF5NhFZf(bg<_Z=-P(LV)!vXiK(PGKdKs%fPeb
z5j5^MYS1(yLF9egOVm4OfR-rKAY_vvtg(nV>l@79&{VX`#9%)b+b@3DN&2HG*e@aO
z!T<p=T*Sw?OQZy<u9DceJ>#vCijrO8kYFdg1$=%Io)4&dg+Cm{)~fsGIXL73D1`d`
zw)mu56hjNa+hR2(qt+~}U!gc9X+U6jYXe{5{c6Xhz>4I*eAe3}VGD0V+yyv&BF2lh
zwVVT(5AykS7KXilk_*6OJ*oyPp>uFdl`zC1Ne*1E7+;1J+s=OcX3VO!LAzYQ;MI@#
z^7^-b1+qaE7~>o+dXq9v2_KgYcfMeJ-aEg<t<9|k&^-udc)qNg;9l>whz(a=;3c*B
z=Q%~KN4e7c_^lO)eF^70j!h7Gg5u&}yg>^oxV!1`OjK1qig1AYHyzJvDvu43i@Riv
zu=@ZeR?_VL!<yn?{vY}K|Bikb+5Y3&h$1!Ve?>pt$Lc3)n4xW+p@}Po^;1AY`Oqv|
z!gcGwR&z{<UAf}zS2fo+gsb+MH2&EV(=Z|TQ<LmS(cuu{Io7rs*KF+gct%KZD2n|F
zW(_f^q46?Abn|8j;@f-lGwmL1Z#j$G8Qc6kBsBjjps1u2O!ddci3_zOAqhjGiyhrD
z0cB?gpRZYpj9aN;!Gl|aPX7!qymg5@4Ikt#kgB;>BZ+v_by2H0*Qu>m;dMtwy3Uvu
zvp38c3cW9-&UgDhF3C8_xPRa%<yvtoM3|xP8WF(gg7YWCL=vG<Y@{nKK2cFkYrpYJ
zE~gVyJ;@;f{V<NF|I`G48>=3x;=j?m-g!tILlGuR0I5r)HCQRHqMWkTuA&^VTRdpL
zu=(A`@=nYmNYrI(W|4A0L$cbgY&_>^!lCl%q|hK7PHKR**d`njPWE@v5kos)PS8E6
z{GLEXehVP5s>aFMk{G7AjF+giWU@F?KomnXRr)*m(E;>f6*ilIa`jGa{UP8gG9p3Q
zHh<jErrM9!L4-mDPFK?vD}7iHh#rs4w9Z@Yu#{{gZ~k)Fv<-hEPh7#JADFh%`s~lS
zK$DfZqXLX9L_N$}Sr%cTvRtT$IZ?4cP;1wkxhtvT0x2t_U|3OWNpX={dYiq|=MH?;
zSwGwh<R2^uXrSLkqVpQJlQJIb?57Q5v@q0)-m5Xj45wO`FV?>|DYY)|kL%|h$UC8N
zk}+U{ktKEa$$oak5N_i5KPk){d>x1(l?r?PGUI8W4fyMT!q(QDx71*Rs?9o781hG8
zw9Ig*jE4r5q5%1H#Qk_zDI4Jad}F1+IdJE;aC9*@Hs()5JPBX32{^B<He&ytmJN5{
zlcYh9J$54RR$0MS6jZg`W%k|anKMrRR39{z(*uy|(^P~fFjyRc9MDh;8)s@Ojvh`6
z7KYJR6GCMpi+npvNHOFwBxA3n5>h$9vb)usN5-s8>I)b^l?{L=+nqkV&c4?0BPvAY
z;sj<Nhp5Q-IdJstJs`L#;xHn`G&wNNZ*A1*+u7FazLfyGy8u9G%;T?<TaFZHy4K|p
ze1;BHm(D^)LnLF&oB`{0DcLF@Ww-U0{9W_H9oLJ2VUwxqWLk<y5t^Sujz%)f{9f;8
z${w3FL(_0!TaKCiJmt1(mFLD616^+*jKDNM;b4uM=PV`f+c+JE(!C*W6l9=ce}i!w
zdBpck@z2VM;wnen%Dh|iN>wM!eZdU6(YGy>Cb)vDhnK+QZaf_#qtY#+&|j)=d%U@>
z$RgdCAAQ%*hTOwe^vWyr()S}8yq;~}Z>~)|=bR;`qHmVY&ZRH>0pH4z==`9)Rz!k_
zlk5!25|uuImB5tX6n!-|Us^gKeQnatwyr)s-S5=CUnk2TyRkAMlJxx<74>L90<k$L
z*H$dgvWwBpB!Dv2+d(3>iE)@?n}(q|RW3#8t50kY+?SY|$#+=<e8Ll5)Rr#|31O+#
zIcu*A9fsF@uFhJK43SdG=U-evLdqnqp-U*_H_KF=%JmFiRu1?0j-Obt|AcNF|A21H
z{{?RU0b@{|`mxPsL=aWnuEA)d=HkwBF0LZghXnB}B@QJM!1eF_79J4tcR$L~_6GyO
z7M`dMPdmEsg;dqd)$TLy&xPw0iR2UH9I)e9csFdMVX+)hUiFgcP0c?awiiRJ%`Y`4
ze0Tpdvqty%eJ<;86U#{b$EBZwv&#+Yq?`5&`P@Jkp_csa-^azo@IS}(zc=Ikqw31Z
z_;aGbf1x@P2h)Fy>O)%V(Ks!LUO6=cvFOykD>P^|ef_?cqS1aJ2%S)|BC`Xotg-02
zQ#2;xevi+bA+p%*aHMVAV2uXPGA?x*s+AM^QAVbjaWrBq1`oscT&Bp{!x%CU7M>7f
z$>AQXNzZ#xt=e{ZVx30XAW9I}^aRLMClXBY3z*)toTAz&#EZAoou??@uTM_>K^Qgu
z+}E7g5biR_(!HQx^(j%$1d0ZI<gR&Y8?TlsC^jz!3*<<ZVgUo#sTdb5<zmXiTCQ@?
zt@q_(e+IR~7u(|5@ucCc@MOu%_OuWLN<Ax5V-syuk(rG41DNx;Y5~sX?80MZEGvA?
zh>QzfxGZR}@jJuH^yq96ph0#Bn>2{=hD*j&Wqu77y4h`!$hx+cO6CbcX!uBto90Ve
zJ;lqE!zD_yt%7!|%AlPAL5+`ts+rn{B1oeql(EBm9c&C1lcCcgARmu|3PlICKeMI*
z3EJ?G(Jge3ERza7A~aJ%T9!K$D+HenqKqr#$^AHy=wU6?wi}d-jjo7l>p7qFsf-#I
zD9B6&rH>V>tpo|}?BV1D4-P@16(F>6Jbu8Cg|M2Op{OIuP=mnD7pMV8A$aaWsh!lO
zNBL_Z<^>aCA;RCQm5P)2Vh8DjiFB3#I!KH+qD>Nl4wq|0OOVndI=HHjQ5@Y&P3JQC
zdFi3XvDL)Jqj_VSJFGD5A-nnI4=K+>odJdUnbK$`a{cYM%13M-q}LB_l+*@2R6W9@
zs&8vuk&)NkzM6c{>*3F2n?J0;m%seFUVhJEpJxx*zu&y2Y^nBO`?PxVetRlw%$Rx@
zxH>b+9yVYl#RztSyjU_|2Kkj41iPOHezeywlVb9Ab#tIs9CB2_-RbPLqGi{-#r)f+
ztFLEhCDa1CO8~^qasY*v#%eIQb?6nsfO#kmchG;6T5Wg_!7U*s<f!ZbVvj%d^!N7{
z{6)V=sl)f#(UGRlu*^1|HQdO=`sL#E$;Zk5<!K}LRaB16_uIs2$He?oBlBOr6`2^g
zcjf&(#HhBOM+i09i7S_t>DrwySYCR|D3?@ccTXN|ovT`v$;X?xwCoFS&NNdS>Fe94
zA#32Hr;^PtM{t+LB|*$hTwUB6XP2pfrkJlf5AUweqnikGC0K=DlaTT5oxFQGt}m~3
z#WCD6<D&v<C!a^R%kQ%zD<cad8zX;5+~Oi3?kT)39z08qxp}uwU*%?8+`H%Rv$CDt
z*G|n#O8&1-kR1KwLpE)PJc)jQz@Z?mjaNM#+~qud&+n`j7;+_lr&u!-OC0pz%#%Cr
zW}^&>xO_I~O_O>zn)q5C;v39yv#fAbksyUm*c^iRRvvQ3H5?N;piT`5Od?E&_~d<&
z0%L7xEX#;isUiU~z-ZnvaL%?ja}Q|vmuPmK(F*m67fMM>|7}R99=wB2ft%60$W}cE
zLSjfjxJW(jV=VDVnjW7fduArVVE)_5LtYXlBN6l#grF#sa<Jq+5Ew{^yP|urdVh{o
zABJ~T8`Q8WCcrsIT^w&vU_~;5(z;X#^GF$%szgF!YKcMch?Bq0)gnh5W5abK$<Nnk
z<+K!8AXN5~eqAdXAE9xFpWNkmE1`|C>SpD1TKvhQcUQPhc~OZHaLxIpLYH7+Fqo1P
z%>Ft`=ESx{ENVHJ;UVQ?5!{IQW|cJfJKJ#{@2QpYubt?cTPkr<YYJ)M0(gU1wQ2LE
zk#ne>5ccL2V4w{B5kxgsx_t1Mnp+XB;OuMxT+M!Tj);a~H25Y8JzD)}NIW<mRxm{x
zGg_l}d?HBAmUmNWr1{90uN@2bcM#MXp@9~VOc3!4Dvics46BI56Iu}Q2-vu{jYhz6
z8NP~pKf_?JaW#e3$1Npc$sE#RPacJ0+ar3CGS(045^AIpCK47)j$uaiV?z=Z1{3fj
z5S=vi`}y<SSSjeft*}8NMct^OICN8DH)#n?CMKrBP)$WG*j0*&6ogV7Gu&>1g|NO7
zW16TpI04aCZ&BYfD}^)JRttUK@U}N@093(4dZ5zKLH3>E1`(X@s3KslB4CjseswMZ
zmH(r4{XJcjt8sR3ny<*xy?n4JQRnVj``1^~4-joL1rKx2it2#5N8wuoI%0)rb%);i
z7=3EK7Jr(;T&2ZAQy$C_R5U<!PY_TVGJ-#Hm27hekb8KxkTkJwDv9&Thj}xT$3r;V
zHu2=0LD6<nsb=2%0}#p~{nFHUa?GoI{RkgrBeAM;Eot;3(v}EnLjR~dG))g7#;rrV
zB8W)}i>Z~KD7Ki0kCLxV?)eIwY7kMYJn(3TF+U9g^UzK<21f>>D&U*^f;e62;W1rE
z+{H5*l+eZlnfGOEItwHZ^Lx3>^^~O+;G<Bdjz|a>ftF#M!|a7JkHX!LRSRiLpW(qh
zqEQr&tW@YqwqS-|jk`-45BL*l7*+{DLof(Qsstf@7l|K?6%rYE=!dpYl@{r`bial`
zaBz4`%UT1_NN~Q0sYcPyh7g&TnIf(WY<3H=U(|Tk#9E9%TO};E1n{fY)X8M9PCSI-
zU-u3`s~~wG$;T9%x~oOKyAC*8@0mnGG2_}#jhd}mc!^UJz*SZW*r*qI(Iq&0&)GQb
zuMr$clamtGFBJMBvPCQkGdd3jPU{4kSR~4E9m}Z93*qUlSL7sF^+5##9@59wJa|cT
z0bDOREj`dNv^u(2kK1oZUjGiNRpL!D{W&~%hc;dy(d;Hj0#{@s29R*yx76dn@`eCy
ze;RY;`l<v$*uA{p|5YXLhrp8T!T|DA)3nAt&0}tZrug~T{ZRK~nyo3~(GcnkBXmrM
zU4UCM*M?Z)R6b=`g3i@7dpaaX@9aKT6AD=CU&`u*1h!)gnu8WcNrt~F5)xm`GQ9Df
z5W^1Oi{`ml?-DyYNU-FpNhk}mT5x7F8%e^Jo4bQ8MO+CWq-VEL+tvgTRQOPLytxL&
zwd)6z5gga(G1%ejivHkfcf8E;7aTb`s!fTJ=m=CplI~}k#6tL2zbr*ruI7zaV_-@f
z1!cppnEQuN@f4vhe242*;)9o|?f7~X(>N*a2Ty+4TxANL=m0|gc;`COVK=hkYN><w
zsjH8MB!aW^fCWy34mGIaWgL6z9IFEuW!iurTY`m4_MsvA`ax^$yC#CNWk9Y5C%ifY
zhgTK+Ym%NbVb}kVNT6LAHq>a+dLR%#Mg=0w_+5GyB>Sa05jC>`$cr&JStd-y=-9wW
z(t06mxF?B&SIg+;=A4+^zu3}7uVm6_%7}>7wJEi1xow3FzVA8PYzRl!xA6Q3AAdi(
zPJYfv@l<*yef79<bK(bH@CGaWgt-}TdkA4UBmjMtRj#}N61IM5MsemQc&KM&&wbi6
z)OnVDoDYbIVC#Fx)tqwV{Jamjetlk)FPXpG{9Sx{xcJ#I?`#7<f<E6BlRBSDy~_#w
zY#-=CeK9JYk5<=&48O*oQ&F*R?a}^qeRXzocc|~P+}hFBy>@x+>@puybA6p7KRQ#A
zu^DkWwYd9w<zl-v&2ZE!IbY-%)PpGuu&4&;Np>EhWxG0gY*pRU=IQ$T+Q*~8%jubW
z#dt*TKz;{di_#{Mc06Y=&(aRo=>_*}`^(kUDgDdQsnyG?)BAS*^vn7C<Lk85tJ|ZE
z_sjG8QfDisn!kBs+b4Jk_nr15ek(U$i~|1#los@?<izGZlKu(y&}%!EZ-U+T;~T(2
z8Q1^cnKv`%KjlmQOY0!`4<ZU?1qpQq1>0k$UO`|0LkiG;%E@s4(|#3B#{XDRiPDg^
zIbcKTo~bb-LFcra_Y8H%X%HhzB@lqy=QT$2qnWe5ET<HOcX`_`8Ko*GUA3ks5J15n
zqFPc{#UFB+;qHDvMBeWFe!Mxoa-ef2GFCcqeZFZ_9;2JKFTQttR=E5M1?hO58K}7r
z-~91XF~r=&5AUWBpWnVHI+~klwyap=t2xRy>$9&aLY{K*DAFu%Ab6!+%@lvG!i~aJ
zw8vz_+SurrsardwkvG~St>>B0r`-<24VBaD(I}eKv=79^gByknE4U+Kj9w%6M1OhX
zLPOKi@gc4+^}326cE&03TjH7ar&88$C|2e7>iWJ5QtU7{t7Uo;6EXaZayWI|9t92x
z#^UQGu@R{Sg!^qlmgO`JP*bj;_$Q;`&z$v&Cd+2t%KE9<Tl?VLkp3=;)^6{KJdit?
zR-f6byK<S0u2aGp1dXD0JZWwH$@#4<3$f*HA}X|OkWpJ~p}Fasf#sZ%9JX*lh9#%m
zkVXp{cR!{rm$EmfvYhGoK$?3<fIK?f_NC}B$%7ljX3Gei5&x1bZ?@ej<igOB4THh6
zLBVVfKHqf%doBq^${&)lNM_O{1*FC;c3FTf^vGI&$Yy7>U2rSvQNUS%DgX>YMntdu
zbf4F^cfJY7ZuA>cU1;<<06Hi<V)HF*=ptlSrr|b$Alw^wmj$UZiBTAAW9SU-!wvD}
zC1;%4sF9;zfeb|?7T0gX10wh1{cFjEd{~)kfboL!Wq5GGXgvB9xadThvl@jN4of%F
ze{H9O0>m>##FaHdiZ#TKo%_50hETD92{*{7axZM|i7+Q$pn(sgZO`l3(s01Nt>JB>
z2llo)h+qiVw584dl5IYx@O32HMWrrQ3PS4?7(}tH?n{Cgb0%8IS?lJX$;F)0-cBFT
zkNZSW!Er#%aX{TZqH2LW%9hqvHNmi3T2;AlH!)%QnK+Op6qxl5p7h-KG2cGUx~;!@
zBdC*y6Pdj0>5TSS+<9V5a_^g0RCi33wS;ly4Q)mIc&#mePmjAQe6R|x%t9T@t4KjK
zO$bzoyb6#f6*|wwS30$KbXbST<XIHicR?sPSmnsnvq&yWcYo_;&=?YJorw9`hX7hn
z5!W|(H`98nuP{0VdU7wpItO_IY8~9Jk;Y1E(lFaYs-oXivXXOojj|uMulYDO5(VSH
zLYkAlXDHNptzT%81M-bIbZJ63ca)P9&h!&;6)27D;Ru;xCC{SIL58D&%(a)IJbe_i
zn)K-;$5k;yeod!Au<kgPN|)0<GJu~))UgWK%Gr1ZxTV=9OE&d;Hiz23r1r`7&g732
z(|)6yOe4%_rxihH`tu57@8aS>OvPjtwo3Q7S8Z>?+oTLfC=%({ZB;^zPC|H48gJ8(
zDAnehhu99ooxCYE>!q4!7nJI?E<q18ePP)11fXPA1%2M=vV?2~0X&yGc2SX7%jzyo
zAao4jm7lv0XT#YB)|r^qp=#^{MW^qS0?H1V=^_NW`;Ea)_;VNlZ5*f$)IalmRV!XQ
z*{j=-Ki3%YFO%)4^q!`>FnM(EZW76-$}L%jiF?`Qj|SqzRDs%-u}uR?A?u#kvaP`w
z&w-9>B#M~O%S;B?XewR%d@zR6U3gCiLZ6Y;r%G}iK$WS9avn!4%ZcgC0Yhe|_s<__
zU4R^dTY|Rlao3ejKx-zZ8pG5HvcwhBr)EV4mz}S9ljM$_%x3dQ=r-^I7E!1>Dgr3;
zz9pt4=d+-?l^<a#=#ALgx!uu#bNzW&VsQ$>?-RVe$nC@R4a~If3h#FUB{D0L1z(C<
z=4`GA7dm}xUq5$3p(`uXs5zPPk$vq`gYwtzSsN4Id>pSLcwnMjrp^hkGy7uk!e8p{
zply{nrioMO2J|R^goM3gSLiS9XN%)(6?*-H_W)3ImF~}n#vbnIH8dalnp2C<VIM<_
z!M%hQKnClJbu@`h{}7>E?8+t#0dMSfKO4=jem=-5Uunf2#V5rSj>L;>lCncV2hMdN
zpzFl4FjAdo8P=T9_dss~F_tt;mnOhX_VHE^lh}U`N`osxoU{+PKlTzxeQ(Cow4&8C
zZ)jNixx6slz+Xw5_Ec!owC=KunK{boD_6O!VC+XL4_<@Y*lvI^D^Wsa4h1=}J<xFC
zk`6Xh`)8xepDaEmWW9wcAkj7tsjrGpnA2j8<<h(rL5OArcuvA+-uKzT;R+7;7D+nM
zz8lQy27FtdZ@V=k`yFZ}uTTFjaLh`M==l7(ei#yv{gMP0DujAY8L^th)&uKo-12Wb
zIrkOR&Nm>7!&u$SQUCBldemY3EsNWbv#@!ZF)>w~|6~vO61w+z0S?vo_M=_CM=Er)
zzqf?1PO3Q$(FD-U+9ey^L7KE#`9vJRwG47w*rev}R1V&rcm$81PdP>ZN&5BLM@C{;
ztoaTHBY+-U9peWbb~tgB4%+$6v5az|Rc2}-O*CZ+%KCmLG@;LZ4dm~m)}a|Jm#^Is
z#TQ8-r7tic)`cqECHoIruK_EOba&^LWrPH}7;Y$(nxl`8n^Sy=VuP@M=L{?ijQ@~@
zWMuu1E6gr`-se9Xe7n!pX=a7X$kr|rCPZy$&vLt1=cXuOeif1cd6tqGi&Cs@Z>@q;
z(q67aU*>bZ<ZrHybUGhp<Uy<v_;1aO8hw~^dbs3K=0h=4f&QvSjER$AN&b~*985<X
zeC$KxgKd*L0ZQsG+Ot$)pL3YC>ghtr@mFJnh7E^;3}zZhgq9g|p99bYmEnD*<k7QF
z5~;{|6^D!6=4?Zj53yg0{~DWJ*Q4DQh*=yr+*=E#@~n9K=+X-9y}FeBD-yfiaI+-m
zv$Hgu$x7T@Z&byJkU@Q;Y|CaIZ(0elYMm3bh+6#f;|MmJo@8!LbiTdl!{OKE_y?^*
zrr4%;%b)fQE5Ce`esJEZXNp4#z~`Zj_8ID4kw`ZrDn(p$ogCV;;;O$$F2fnm@A$3|
zL_%bm)`81zE=pr--X<KTsS=mb*p!+UR$#}>l0@WNQxhipWZcI|=MU`^5nm^04V_nF
z23<AkUULi4Zz3eC*Bcw1O_<6(DJGh8=<IC;YO+?2C%a&eeW6VTO8KWN*0U+ie4Rdf
zRlk!P_)QCJILH=SL-q4meyRS{e;Gm^BcnJ>1`1e3FEwoElpn<AVfg#&74M4MyHIMp
z?lh#3-O;#C?$2Q$A2zt$xS^BWvK)|%NRihXUw<5WebfS(5%v-y6he<P490hCA0gl%
zPliJJ?R5F?blJ~2s9x1Kh53y@cSKs>slrP-gI^awMz>}0>3){oQ}jXQdmG5~*qnQG
zji5&Fray1>N01|6UXka?dDia10Lfm}AAy^W%?$`6<w)M+J^$_QK^E+dcI9Bad7C5S
z^koMwjS(jP#(SFNZ5fzx&c+-^b|o_F!-#H6`Ly#DPdEeTFE&0V)X#uph=pt21=u1M
z&OvJueQhAF9?$gqTP#aW4{#<H%GzDc0c2Y|Q_*?V1Dgq$v{Sc$Vv1zE`7FdiOsK-Z
zo-qh+QET6iiMj`43W?S~GUif%ty8?MTHFn_Ufd?xAG%p|Q|Ok{0uPWetilVs(#!EC
zp~ro=MCl`v%0I)_YEj?Wi62Z*mjq`SBvYWFgnRTb$NQK6e!bKZV3pu>0IZ9EAoH0T
zX=FB<Nv@79qxf6H8u~;fRVuhsZTzSS>;6|7-5e08qgJ%sr$*(b**l%Xya3oi%eHm@
z<pHCWCXl3rW9h#7vvIG;u;l2VfogZ_VyILl0X*g+NAV=?X42`nQ1UfJU!S7pg10#U
zC+CCn@A{zhzeP@?9x{g>w7HNYnYTA*5IHtcu7|oq1r4?~KvJvhjE%&y3A*l6;0MWo
z_TZb89CZ4jh%W|y$wT412h&b~N82bjbvIVM91lT|y{8~UQTLR}-!2^ztU(vizt@MB
zoo_KO_ac(GzkY4!Xm1ZWoW1+3+!#WmnLg?OghKNrJ|nwBqo(m=JqRou5etySX`R0{
z-<%k^T=GLQ;0eK%1mlI)5xsJ&x;rNP@#$6jJIQ?KL#4@^d2st@U^GKuYz0vvn$!{G
zX0s|p3gd_pCCCSG6w2@HpAhYT>JR^|8Z-UGM*rXE|6Pf(_=9NwQjOn#YG(Cu3G!bN
zNU2)bWdEW><jZc%H-nl<M}v;WQ;&^-y&rQgDAboarH?f0ETf0VxXpOtq`SM|Jy!o5
zSxDY6A>hgwCXDBPpF}Ps4@Q{h_iz;UtdEub!S`+FJZ8qjF&N-Gm<b&5H2Dg@{l(MM
z%6Z%NqjSo6V*WjSls~@q`TW{jnw2k2Vn;f<7eGSn#Xv+28NJl6!Zup0RQ}LN7EM5t
zjbhfxVqk63I<#;^V{d8GErzWA5Tl$?_75JkB*=;PrJOMhx7DyI+P3yXI7K@xue@+k
zqq*0KUm6``^hdPd5#%+=M5mA}JlOICo~-UeKm>4#+HbBrw~Q;{_#=hm(xh_|)Ywqp
ztGFL)8pNxX7?gvmnQ+SE!=tYv8#R%XuA7XSEx~^_#U?MWj|O&<uEz61mR|g<+4ct(
zFzVS|+}}1&!A5$U(<g=e`btM!V|j_#1{TP$(BO<=m-0r9;+M@({4B$T(s2ebT#$BX
z`)4Ia(DL|5*zFi58Acc}T@gE8#gA)TRYnE7lFlT#kDU85q!8vkP_;xa@DHO;bwZTl
zsNn&uU{EP*N&mQ0Bel9zAnLOcNtGRs9o1~Bdz|d`@(7s-l|b_Vj1;7)T=t+3PLyMh
z@!llyR#d~zN@$-8Qf^2T0H9O}jU!j6Zm@S)gh~R=<1?*N$+08S@KtJQ|C`Y$S>YZC
zX&pS8%5ak=>nqRJIkV2kw$T!agKNUvn1;Z_bSEvuETsqa<bVVj%~bP@z)BNadBxOD
zduR$Rg)Rr0xt2CG9zrw2&4rUtsY|3H)P^U*AK|Gsb#LE*>3abM^eA<KjCg=0QaHkg
zH)uu{$yFnLUj#;Bwv_QV`y3jZOksuGGW`u&r$qr+dZNocI74-YK|?DaMj{q+Cr=OQ
zcJ|`Il>>-Zxj2(qr8Oj#)*V;BUdQgX7J4urd=wAf$qUDpghShkg-eKoNf+Kgf!LD}
zMewy>Oua?2-Qs*mJKDs$+J`y4AgmfGlkgM00E3t41L>R4TToS*u@q?>(vVPX?7bPb
zuhHcuKZL539o(IUuWJ5Q6>e}fmg-v5W%k95{yCQQoG=3oBY29CWUzF!Ej%pWQ{})K
zpV<r%Tn@0&D00w>JEW8@^h5#Ib1(au?<G>m%=+(Pb$P@4vU9X8#-3H5Suco%+X`m=
zib0p=jB?hF`npVi_An)gX~rQK(G*S9FL@vwu}&*AH=_*EE0zc$+4D`bn^)WOrE@9?
zkkV6%%J*UYjeTgdt0}vZWSLn<wFPm%<jf(4V8ccbFVdvn*tDGNizD!PhkFVQPw>x-
zkq|`+90V7;(Rh(3*iS>sZUjI`k0sGmA2N*ji5$~6%%Jj@6nrS&pY$Qy;Ey}2F`Ni_
zVE0ZTQ}4mZitSs6vLUY41MTAY>oS131EjheQNLeZv1we|n#B_wGteY|(csT$Gv~Cr
zSM?86KRqEbT@rJ_pfF0M&P*0lDDV2}1_fRvJZ$D!ricizTtG5m7k5lh`yL?OI&{AA
z?V>#PbI*D*cGT0{Zw@<iKQzvFcXUo#swQO!W4*}qPN7SpyX7nfPCjg77j>5_BGKiz
zE}(Yx77u+77pDxLo>D-{xuW#&L~R@V&COq114)Gl3m2F7{%+#8+@kP1_8SPzS`Fz{
zMmWKJ8`*_ln0y<MOk+2H!UX7Zx<ki5{&@TcbY=X991|<&e{LL<q#_-Y!SJt|oEHcs
zJIerAzNb2KWmVOp*?DVtT>~@{Aj;UNyK8O+GHAu73SKv!v{$@r!bl!OvF>WVqKe00
zW?7a+DT??b5)~1#@Op`ONBJvWLC{CwvecH(*VmD2dKYg3l6&vhU#|F{9Cv?hxWDD!
zyt=~rIU^Gj?O<t$rFar6Qwu=l@Zv$jRXI?=m7NF-(#?W+GwSQTb208=F5Q(4H`Xo|
zoDUnS=u#=pE4eN&w>ICV)i}HI<}+>Oni36i4SSLQGN3Sr+`HT3+k$CmiHHq|&!cbp
zL^72>>Lnn;6_){(Q^=JbKuW^<W}^`UJUbTz5mKFRHMfmfhCBzf>DgL(WxGx}gcbHP
zSNEC(l}%jvE_a$Uo<y@z-%D%q?`|f!JYA11rR`hUr*t`X@xs7Cl*n{SL}N;*g+rm`
z88u=-u=3DlALma(!MvOF&i5J!SYODxl}O10_gw%R=GWs^DqxUaF;<_38Dj0YI!u0c
zHeR|9T3c{`TE6}4CNpO$lpqmaA$o{sFL94S#_1Y`#PmPSqKHN%W5=e3bQJ<U7{*;y
zEvB~Br58Btcm{5SgexTsUq4b8-9OkIlp0ZH=LG|c*(J45CLFtiPTz3IqmYj$oqB`S
z<cXw6a9;(P_lzQsS8-^wbOXq3Db-`Kv%iO757^z;@=mSn$rrK}Fx-9LfTBpYYqaD_
zQ4I==kqh9&^SWBaLPti5Y3A4R=nxK2SqAWM?g)ip8L>vLp>6O=X`t7~Xysfv{5|Km
zRaD`F*=YFuPgC?CQk0xORSEwo0Vq<FjrlRL=z3Sbp&+MD$0riam-Qt(0i*B-*nDXo
z^J_NO3Z-d_XNyKW=y9P?(p6rqY*~+xfF0ZQU@mYYG7ErkXD2MUo$_SrSVEb@VWwhY
z`AtlxSj>XPC4P_EF3XYao&D{^<!7xd#gyDx<-)DyRHc=vV~S&5Nt!SXJQQfZB7<<*
z1sPTZWW#;)RL8u?Vw@(m*74bG-3!=xIJTZbKi|xqyhNWYYCIogemPE`VR49#7NzN3
zmjzoh;R__MEVsa0>TPEl{q@k#mFRnfr<|;x?GST1{hJs<+kwZ@in-{=u+@GAEz9cJ
z1am8b?Sf|FjRA+cT<<}62cp62x_pbp1YLX%Jj;L<<VP;0FvfA14o;m1&zF<9`{JRq
z?Yo<!iM7<DsVzh?CK+%W<~iptH@vB1MOc_-^ZSL{U{_|#oL*VNQigk-zve3KQu<l!
zw231l&j#>EdfwXV<|Pbb4^fE%1{8txnh?xY26EEZBhZGeLm=GTuL-ZtzrrlD%s_dO
zuRu6{Vr@U<0&UY-8%vuxlksrVNelbd&Phy?Da2gkMieYTl%dt4olA5`{^IA{fE2dW
z4unS4u1<&5dz;^~%BiB}>(n`Jl@4SeSTR#zpBAS_o6Pc`oH-xNoI$<=$LC;vZ7ODM
z864ax2p8Ifg0|C43mT2tAD`N>!pQyAvOkd#VJkn1rnOK&zEq}a)Py1&t!QA3u`w6(
zTk9T^Ardt#-RILaVQ0#<7-f>)0t3^e`Oyq|Z0*aujJHqy1S7gwT`sv<brH+A-*|;B
zy&ueLNdDK>sQr8ZiQz&`Nm*d>-EPZ{sKm;&PO-!dk|xmG2aGIOQU9VZ7~jg4zEg`9
zK$N#`%|NS})%~KQs7@1NG&vY8lu$V(m}1zxK{{CXAmY$Ed`?dG<vO0Zp)99$*R;2%
zcPa}BY`8N@@$cXR^7h07f)V0w{xA6yj89OWLNU(dHjIT(-byi>>A{zLjJc2Qk@lsm
zftBZ#^KNMqrX^=PdzV9X*&20*(~FkrYmJpFJuB3uPupS<^X<lv`MwHs!PD@k2=;5)
zIoEg6sJ(D~&|-+Uiy_kLbi*+X&+_HcxHE<!K_~xEx7yEd35;vO=cPsYa*w^UFD{<4
zkwEqke1Ff<QFM4JS{8S^GnSa~vr-PF)(}BZNyx4#eDyiF*bE=Q%~|)8e^}g1|CFy_
zV)~ESeUh5=zZQ4*b<M1z)*#I9{KEaGx}m3dVHglIC6Gkfbs^*N*n>?cAFq(iINCOB
za^A3*v>4{?&cOt${)Fx8m&NDf!_D^*B@!Z2B_l3PYb0YNokSYyd!sfvzLNCK@1~Fa
z$~aG5h-%|Eb9MM?{*|N*9pZF7<K!{l;e?=ZjcionR9E2Wd&&(V6CRyQOH=iL^EZO8
zFU5*NqoO!6U#MYGz%w31$Xsk!rHV|@_vf2Ydvt+Dx!=XPTp9ku@AN$TOEtIlD{buq
z_aQMoJiKvI9b~n_TJL$_sD2*nn-M%k6Z+!vnZJb{zo^_4+{A5Gs9J$den-yqxu!8<
zYC|s-V9|>=2?SrZuUUs-xwc4b+FtNnN$xI(0davw6h9R<*;(@@_(0rUINzkk|1ZY=
zsY#TE4cbK8wrzK>wr$(CZQHhO+qP}nHdcGq-Z2w%@I4XpM$`|egNnM9d8J8Xd?aPb
zD8M8>biUhIL6Zzm_AUg0nMr{toOEF%%;O)SUp_ciPB80&GFOO$NomV0FgjC*nc73B
zz?aSU0^NYG3tfnm#?KI-;yCcO%07EqI<`A`z`-y(LR}g5-Vq-?J042U>9_NMT=#ft
zGUxz27zLTK7I@e70$U$%u^c-B_>Zb0i|7@SQ6AkkT-|8E6^48W#}3?qYQBSkrYj;u
z=WTytuX>BzVEtY|Kk3-hHtgvAeciBe965Z1I6VeFN~LWg2X2=-`-8SwRanU0<ssW?
znL$h9s}5Pr%%S#d^<vUwu$192ijH^_z=X&Rz$zIjv1C#!t(AX<Ksq7qlz8P-f}^-k
z%V6_6k<NjWksLgD%a6@cyNlu>PP^@<mjHMh<BtJk+I4%GlE6v^B-p*SwB}9$yarYz
zKT~sdT=8XDhglZ})o13tBLNnmYw;8PE?>dd8dsh=*B7Ie-R7$FGds2`32l=JjoSDe
zzJ_yu^Ae9~gOW)|GXu60n(MM`5eqNDmq-ve*zsBiosNSMj<~F!=DN+$WlPTHavZN&
z_0_c9Y^Q|rZhcL83!M3V8?(0pM)w5XQ5t0+#ob%@7--N6>5sXM7JIfy=|w=-CTIU|
z8r;BDu)cJ-diPeH8?a5`v7y4qOC)X0uQAa5EetYt#9qtTt8^yCAWaGa1yQTw(gZ{_
zY#Or)ATEXG*H*5BPM&r+VqDUXBTF$$=JJ9nGEk7cat!ZAVbjEm{?la~f3|ojzJs!l
z(QFLF$Ha;b-{Z`<T!K3EDzEcyTIQ~hr>jl7xnt=_R85kpD5L|=wmLN%p_xtjnHI6^
z60&9a<0&DPmsn8{SKTC4gwDR5`J~(F(uj3&TMW$`UpvSWHvqWiz!Zl*j*peY&L;6c
z>9ZD+iV(9k23|0=ugt$Y^`2G#-c=gqIqexGOEldw-N8hb0|5HG2;np|mSIKzpHKmv
z;cQ>Zl}oWo{5El;LC^6oF&C1m41|jx=tldW=n6J#v~1LOlX@yhn=at5dYXoz5T?s}
zHb;4ieYq=4LvzzF*CixTdZp=R+q2GbmgCM2Pw=F#t~yXs+7oFGS5bS!9Xh&Obo93?
z31&OYvSQQjZ0xfwTf3~f8H1j8*<V^1rb9n@zz1ZLfyQZRJwdA8!!ztUEcHqp{zIHj
zEWyB`IbYhO5DEfTghdMF*<fF--EU~?Ir;y%%QFAdPyWBGoBwC)!T!JBWkb{^61Ui4
zdd{f5u{#!3g(Eg&lUgikC1gBhCvesoZJ26Ch!t+x_4oh)B^$SNINGHHupzj)0TwL#
zB31KrMbv)n?(}-z)gx2Jht!K5UHY_0Hzd(a(o^3reY-x;iVr^aKKn{>tL(55C)nNV
zsx|EnnW1jiOi;4pJoDvNY%T6})mko7Sai7d5$$*<ja5QT(9GQ9Ac`<#qg4%4iVeUd
z$C;s<fNE{_H$Q^{fRoB0DuXyRcaFp`m3JiGxDDVlgSE<{kQ4AIwq1B8p687E(C+zQ
zWKr3yG&8>}Rj_N>8j}aW+25s`$XBpm<gGNB;5?(?+vPE<Nlem@AuFQJMl|s9Wg4e(
zrGPrSw%#fU1(@Jc(Qua~^sYBKqiRI(&i_b_Pu7Fzfkl#OAbQ(gG0Eg9^Kf<z1=@~5
z!$;1I{UPyUbn&?YmB10Tkf3w#Uo$ny@0b2e)q~$_J^BTPMjH*NyHmir$V=hX{2*&O
z^4!7c(&b1RJkw8csSRML%sf`m_gaCgFBmvG4ECbE!X$+j5y3oVBo=|xZ|vTV4J7Wb
zS%9O4;BSPZpb&JVHh@yQ?d-hG{X5hH!=|j`wOGSqh3NbzB3tF*d?O~t;;k(j^Mnv1
z*|}P!3x8#ohdo8W25&*z0;}8W`vXoi1Y&bGF&qZE>gxqW5rLg+AhO_cX5p~Kc~dIP
zk7Em%>h|@u_^4W6v9tgIlrsF5K8Piy@}eM0=)no23V4i3>c=#w-eK?O>7vRgbcGCR
zpuH-FVi2QnJz=zm&0<Nz$q|QtkE_qG-gv>)?+moQx{!fjtp{x^^sWs&(h#<HOV=SW
ze26<rBP3lJVa>|a&O||7lw)2Nu~b45?49!hcnx)<S}=jD#Gux3bb#ZH69Y%59*GLU
z#RKPSz)IR`Ag#N%=oeh(?46yCV%FevVtr_31H%;COn<(WrdByk17j=@7$|T7QvoE#
z(|9O?)EGR`oCO04PJna3Z-qrV4qRsX4|hlufN+Yh$DJx)^ry!ZlAZ1xNr2y|c2PgY
zeV5}jST{N0+rA$Wi3i<J&zm8jgyinMZtFc>bPD{76sv71XeYS^-#i<GJ*+*6uPM0c
zE(dg}Dg5AsIM(jT_}~tUh<i&BV{}x8qfz*lWXxw!$-}(MpA8PD?SJA{Uq%+Y`qG7Q
zt7lM-SAMN{Y3;)R#B;x^Y+qQ&njUVj3x)(<a}F<t$69~%G)!ZAQ^QqAyMRalJ3!8g
zh!|ujbpkJUY{+cWdIX`OJ<2c|1g4e#;3BN}ZmZUEAVba4fxRe)F*hHMqIekqu2PV&
z2BJD;TI2K*>FGVfy7Hbn<yqt%dL12p16os%^KMW_r1T2FU;6#}bVqP|7VB*R7@6sp
z>U;hJzoH5}tKPqmK+{cGy|H;Phh(#aWJsY?CX@~NqOU`*qWTfcC;gHX^@)36sxZF_
z)6IY3P{O#aG|yJFKcJ{*l^8pbRVO*|Px17^X)*_sV`r)7??$myvh-;2>*Hx)uyLsf
zQn34PVDC*3m>Z^twCS%puHm$j4i~NNYyRRJSx`Du{75I(R*vR|Dt(cJ#Nn%7ng{(O
z$8{bki2Yv26m1s$13Wk(Im-rruu^u^^0<m4>4U`u5GNBAnV<mvYnRH5!RFMr2($Zh
z<{)`OD|Li7V8Mn9O%nv$Oy#3I!>GC~QVC!l1z5N<rMTPji7nplh16WzX`)+01LQV2
ze~PZkLi=^qh#W|Ux{t%VUHqA#0RrXdZxofXyJJE0RQ{&W;aF}esX2%-_8o>K(E*2S
zk|3s1Do(%|?>mOi47*7jisqq+449o$$%^vn?6&1}k~1zpy+Zip(0&m6_l)nSMj!l^
zDLM18dNM94Yw<*vwqn~wRS*X8j@oQmw+v;N+HS?zge0HN5S{*UVM7haL*A`2-XK*B
z*z}fsX88iDA&MCibVNZYhMM#TNCPjYr9x`3AYpTIW43KJ947MDLJHhPe+DT8d<I>4
zB-xK4_qT3Ck%tOEPPv(xdoSezdvEI07>Ae~9%jS;dGF!W`#^b8cQ0Z<aKIT&+<65J
zirY<SfHlLHtYzNcdk@AXVj678{*c7rarq{e`^k?*5?zLBFVcx-!-;$9NGhn{$&M&d
zZ`J~`OkPB%^d?P)mde?83IGtFUQRLubXrak@J8%rgr8al*WjcC6vBtNn8v`=TDGD8
zeq>}_!JUxSS~+s9Oo(#!703`87T?R{DokpcL)r&thiGCfCic>W7P$8pfiEfvxSF&7
zb43?4@I?fh)E`CmMJQs2lmM3J)?9cw>O1d!9&)1p2Itt%G7Tk;q^*kU{KD>1{PQ(s
z3!|y?89f*s=jS6&9#oz-;L<s_M2|FIV<M7{0@^0IFj5HrR`&z`j=C1}A2SHkf9${*
z>Hl|M+6?BD?NO_1Ufuj6m?2zVVQ_6}ThigKOc|z8q@gBCWXUj)HD$^%_vw|_xX@%F
z4cBb3$lYjR-1yU}M}(*wVvKM1Utf=h+sFlGNeQeccFkQ;qS)kU$P_Bg(}lDQqUdRK
zs?w%BUcdHSYL?HhJ@~$9bZx+6s64*zJSz^Z%)B=Ujh~N)p|bb8so$QiM8j|$1;G%o
zLo^N}(>YL=>+UYe4gIIVJ6(I3MCMT@{47T;*At<D<j%9t!kqkJh!g*0(dkHGP>KC-
zM`#RQ7Y)#`BxBdTX${6I{7D>pb@4a#tdLMOFGhg5CfXtN)O}J#V`*W?_~;PQi*tum
zgc4)%X3EAH3)YTf5*e}X-Bnn|`pT->YS))q5mvm$nOUYBm3K;CVH!x_8L;rFN-qJc
z2m<X0A=mw702dM^Q4~3^>JgoVe2aLId%S^u;#H9==~RuHE<QGwwoz2^^+3XVlh}IK
z6q`m<eTzV8a9_b9Q1fiyY0h!K%bV{~ch49vaZpWU?{mDsiv%wW@zA&wa)Te`uA7<c
znds$HI?qTA%uVud;o)ElT*bF?xvogCtmyY%kV?TBytR<EJLa*5lXm-1lH^ffk^Yr9
zjYEx!&g$yuB#9eXQskgTfOsf(7N9DefmhpsrfpL<lN6S@zq|DEY9cE-u?s705FXlT
zRV_mcUgP`UQkN+CC^L)lp+{@Bbgq>U=o0R+IPD$r!txPz<l5KsO~15$Kip<8+g2+p
zw*-|~Xj|1*lWQEDCc;6-LSk#y*XLOCHGsutA_0C6a;+NcTsg$^No*L|YgV4J|D66r
zl1PkC{y@2C{+ZnULqV*9SZJy*rfnfn=f;A;WgLJp;UuI)E^6f!_`mFv;gOJXig}&3
zJuLDrakR%Cg*^pi<wE_&5_!NmXylG1B%Hj(W^Z->_}J6Xj>w`K@u$k%Q>--;mhoM8
zHG~86`Q+0c_nVuX4XvzxI_2jwHu_$2O1!9<L3Ej#@*=_FLZdhVe6x%lQ-XkjywD+o
zLiEe|YGT0I1EwA>spRyckimzDaK}%@YA{8N`||g6{D(;GSbrztOF-pTqTaESArj0I
z+lqFXN`@HpiV(;N03>f1efCYufR{f@CvxZF1fdFPxG{?9Z*K2DC$r_0q933IF0R=s
z%{6LRySl4yx2IaX%!sVk_$?c6FJfETr>9(7thBI;^sO{vW>rU&!t&9q<yTP=3?bAv
z>><2tBY{`=hZPv!DkpTp!yL<A)ppGM`O#}f9pbyy1)jw`27A)kZ(X&-<9iJ|AK7}{
zC!-ta8dS+DLbWhr@@90LuUB5;_S`Rr9uMcDbfE*~>#^Ak<QvUfdg$#Dsn_%T;}Vz?
z6qJP9(E0qRlWJk}L23a(e*^LDwY(_&wLfP`+iCU5+BNqNY8Cij;69-Nd>tf<`?-He
zU?@T2t-3N69v=2~9?~$Qdua^qN%j0-;?@5yxt$p?uEih^9Pg2;ar64TM=quMe4CD3
z1Fx|~5$@82vZDZJe6rOH{mnX&+>H*XVAFD+pNQ`Hy>hdQF!^pK5ND_--T=3zWtKo4
zwSwtt9k!4=(KyFXjvrX4nC2u%rZ-H-%B_bK*u*i+D^*6ifYHno(G_&JBRtWcYnB73
zA0`9G3VyT-ASd{%9)2}QZzZl}SK52?$kTdLF}x8NnT#ZED%9Fm2FikU65lSu9j|Df
z{O?(mmE)Hz{}UDnv<(f#Rg-_;9GFSNl|QH`gRw+dfx&S!e=6uhiZm`Tq&p~04j^EE
z%qID6;n7~fdx%KNa~kGUu4RaN3w2Eh7KEYhc2(^J<&d)vQADppaz-fXpeRTvb3(00
zT_E4jh2j*JJg6f-7PGuh-8bccXq;@C-q!U&_Dou)gtduru$5kitSzLp<|fpe+xgH$
z<W0|8K};E*{JlXeX+S2(&_|geTwKtPRyPO?=%_ucv8MXjZ}z)iZPKV^M7jJj_`-~z
zwYcgv-=T@ypM;EeG}+=<zq#tsz_1IJ^E#{A0hW0{OlK$K<rcL>vDBc3YipA?V3B8^
zme<tpnA6g3Ua(AS@1cYei-zT*#fS1fz#dGRZGP$*As2!_S__SSD<kNTuwBQ%ly0*Y
zjxwi>s?mo1caC07-KMtbb2bj;O~;-Gr{x$_dt!#8XVP)jopUR|N0yD|G#&G=Ik#&|
zUJtub+els)Pzo!3&sF%%=G&e#x9M(|OPkrS+e-M+Zzm`KEE1%G(J~N*59POA8^E-%
zI#X@y!tkY=@_72}BWlaEGUVQxk{P;?9Ub(ZZhS|=uQ)o4)Na{hUd4nW0|XX~Vurho
z3K^u*!9tlM=i5s&l8kWWkWjit`V<XL9l@5K0I${>Hk&p5y;z%{p1}Ybmngk_5^-j$
z5XJ6~k1?qH2GMxm;}C%FD25hNMt%G{vC(yifA5wxoEAb}CT<&<y1+_eSV5&(Yn-U1
za)GQ_o6(n3xf0UO*%<L&nqhoJo<(9mZk!M3G1MTO`_nn`oIQv$0-s)P@EF6vcFdDs
zeQkP%MLedTltl>3Tv#<_Q&vn)q)>p4o7HCq_%45i+JU7~U34d0t5kV$u7iR1Z1&w)
zT}O`GK-SozHyFcRzhG1Z5n!;KPJ}<uk#~6MH#$qj8*oGIjvxO5(_3pLm-e<<jMmGZ
za&_3adu!EW;RbgX?U3KdC12M%Dw$rzH#h;^%}FsPq0<-WDySI$e{~of|K~K%$jtsf
z#|^G%XgD2>BKoe?^%JX;#-8=>hbJjhc3GL}OpGSHCpPL(L4Z)M6ZDS=NPJ!`@$d*4
z5=3ma<b-v{JI%5AxCMc4`GfE2e%ZJ@tgt0OF{X|%!l`EQj|9XTjuPWmuMWf&#7E_+
z?fJaK>~9#S#8V<i_%lrxvibFW_`ckFHY8_;W&TO!JU!T|5<v=~{^}^qSn=!idjn>v
z<LxdP-Ey=0ad455+WWm4f_L8nZM!#JXP;v1P68j?M*$$33|xUD*p3CpqZoAINZzTn
zl)OyUX+yABcXifrP($leok=CD3jGP=>he5)EZu9~sny<EdL-=gHn7cedxx{rBH?U~
zdNhrpg8(@al*xCTQT)p$-)5rZ`mP@vk$8WU&L+g7Y$Er))%nM_r<fJb3@aD`8xGX<
z$R4;(YS2|I$w;YL23~6a-PTA7x(pd?A5uG+vb2Q6{+pLhvJh1vjO+4#Dah**yLh@s
ztrM9gDg>fXn5p7?zN%A~CPmettH95xc#*Mt=Cd-NfutqJXPcN|8ccm-Vex?0u9^nO
zLMc8{9aE$EO_ER@WG-D-ySd9phj*S;E|tu`B$X!C$nJ>Nj|89%?YxyU85VZshQ~m0
z7~eYjlP7ux5?d-YVi?Mti`7c1y`hb0eCVeE^AUUtFN(|Zgy#Zq3aP?{{`anRKuT$C
z?`Kr!P8A)WE_NSvobFzpMB2<j{+iZadjxB>ECA@ROkqA{;Ph~>54BjuVOCx)eR^Mv
zdh<9&TZ{UH3AX~9H2=rgztZ#`1{~&lL-GFQ=-^seZZ<X0B7tK|bo;d?F-~NZW+wLP
z*YNIypW#CSbmuBY8#~@G_M<M!WZL8J!PT57K21@w2oY?mAR}lKn26fx3%)qwU5J>*
zubzBxAAdsU&hB+zFqLKliIZ}Jp(909!OahQXL<0*hZHF(&!WkDm+;BsZ$OS5=+_wC
zM(ydAADPAZ?Pi@{T!=SsZ#oA<IE!*B!w8GUJ!&UWIO#m`J^OUbl}k>iT|>!0K&nW}
z8nd+zR&%yo&?#6DqP4$}Zt;OS3~^3ZsX?M*S2{Ngv+Mi3ipYkLGipf#Xp-~*L1PD9
z?11_hnKh%_-v}oUe!JYih=4HWWw@q^{zg*(ecWg6pu_N(!%@33J}(@(Y9cg;O*z<M
z2G*4U@f$<B0|ou|QzVaaLqfK)o_?W66^2O9Cy;0)l;g6Tgce8*>=Lr*G*^uqaOYhq
zj@MjrjFbbW<UbPM?Om9gS#l<8z0vN0!i7bOi+sW5Zh-+g5JqJIp$AA@%ZvTwNSCkN
z7Q`cp;W6D+o3=WxD6+jUNE4aEhm_*f72c9+;c3d%+ZPa4I(RBVq8T2`g2B(U=SJL=
zMa<Vx-Lna&rDj3*=n;AR-OykQUAB9#2toLhS!sGKBGd+T^9g=n3wQw8=J2?K>9}_F
zMNMZsMeSJC1uQ(TJ}@<T&oBxoGBAkF)$bL&m{vf*qqf}oT3k*bMaHqSB1@d0u~P6N
z3$8mcJa^r+0mh>H45e9m29c2@N#4GD2T!%Yh5^*F+JPe8;>f`3^QE(4l{V({n2ML*
z!z_3BZt<)A1e0>9vK*k)tE8GT^N)+NT3yLNpOxNQtRApo5Sr4ESsfqoZYz%*SA;DS
zN;E3ct)Pbt0iZqP_4Iq$5b~@6A#Fj%{K-Ly0!`z4P1i9@_VyX@rR`Q%jlVAsrb$I#
zyZzX@3byi=sH8uXDL#Y4u9^D-JbDyl{BhHpY_Fn|p68lb#F#Au(YXBlx;rYV%Jj^H
zsL%1rtfWjERJGf&`Lyav?adii5zg~Y)?Lk!5@7Z{+?&M}`VMdo#>O^24D+d<lj>g0
z{BN+7%@WfGcvApkA=y9TIHuda_Sr`p==QWB_IYBw^<~*ty7q4MPk&jRl&MOsrDR2|
z2eENx^U|5i^e%+^^b^~Gxv6h1DTAazkA?|QHwDA&aO7YdaWV>J4apA8dg$&`-97g3
zQv}r)3(dO)H4&YZ<R5L*Jx`oQ&mj~pVahANm8htzvDLyKP1Zlm^!eV+a4awO-p@Q7
z1V;B0TVA+51fo86%K~f2Fp7+wEAZT8=3-SX3{#DZGh%Dt`WzY#^kd~k#rW6ZS2~LU
z>uSk*Ix`*Xf(bxEep*Y-B+Z)2{Y$L_Br+}M|Ai2-D<QOdLBvpd?Xu(Imw-qS;yu6S
zz&j~hXXmyVX}$?{{!w_cPf63v9Wa7li=d?DpB~{K<pYKd#o&Hma~Bbx4klyiXljc}
zZi1N2)9Fi2>^sdQRilXw?sHKb7RE>-I`xPHT-7sWqDtp9w1X*l7QU-^0%@mFTD|Nj
zHiLv1Dh#@UmhL8rs|FD?q6(PTNdv{gi#H(-rZ)_Z7psFL944btnFPP@7#`Dy8lKql
zqInA3>g5A<{)^)Oc)p(-xj`O%dL9v%xQ%lM9o{s~$slqV9XZUIapFV2PibL2{bDZ`
z3uEfVUcSezxf83y!p@PnD0=3Zr~pSskcml?FcL4dg)pPh<}rd5%FVtV2Am4(J=Wm9
ziI=@L*Dz>31U3X!@f?h<D9{_i0-(5RHDw{5%TI}Huv)BNLrAI5pla=)@&uzqoGR#C
zpIuEPW&{JA88eO%iW00n|MfYbx^&}pyEXex6~FrNoa&ICvgKxT_j>zU56wKS682`o
zFg;X^iD?67x1;-8wv?2oCOxA}1`Ie$?I2?Bm`%Xh<*Ku4Ozm0(2pZ(}J|O#cg!ixj
zbgZ_-Ds|Mxe`)Tl!@9qZk4{BegdBD<*1&nZ-x>C?a8vA_3RkoBhvixo+o-6c5kPiN
z?=%sOG+t*3dq2|Q)TCiiR*;AIuXZ81GVZtq?~?#k?lv_OSul&`9ouY&%IfEU!PTTi
z3CzxYfu+?a3xp?YWAN&QNAvYx1P|}<bY5ZJ0lKokc@y*@0B&PN6X49G0(Dv+cA8A_
znxaSZ5kl)sIE-90PN2gLY3X{U;!B}q7fSFrOK0lC3@(2xkV&nZKMOVmIxor2r=q%E
zG3U7hw-wu;A<>-#@KgKYPnK;)cWT~6hdOcw=Untf0h~l!*uOs1mWhWetcrrzAEOX$
z|0Q7<)DIE-88Vk{LBcxRuf&rx;Wwbcq^<O)Pj=UvJz)SAyOqR00S58ukr5I8D<^(0
zjniUhg#;lmp!E)2&ORPF^G%VDbYK%QA@BFay>as9y5{M;N3Zqfz|>b`c^t*2&3k?I
zdTt|o2v5Clv3mB=5Yxm!E>gi?NfnO)Fy4(+^Xb>t`Ik=WtxEGh8bId%(6BJF{O^#A
z6b&t>t#(x3x4L~|9&82xcR&uGN!y%Qt{t09_pJ-#w*WN42pvUbkA}makK7<9{=mk=
zYgtpGL*LrHcYr8;{P2A`wmrV>ZXY@kL==UD3L5uyPzSVP3A6|jSiH)8r*u)O&Yj{6
z2tM!b--n-rpFFoV?3Bp+vOK@;Pxs%Uu0DmQxjS1IB;I|$yYjPxJ^WkVou7~UwYIyY
zbv{DWgm4xJefW5W*$zaS#802k-uQrIg@jTlHJk0mm_>yI^jznk$NeW!D8Rv27p=IL
zBtnf?C94llE!O4g$%f*(hj-&pYQe~I3Z!)fki|^laq%)vck_e!ytuAf{4yhs%+q&O
zNz~qfnL~SSj1Tipa-2Xr)K*gGA-~&ab{hP!2D&z*9b5~^wxWO!_RO9O@Lbnw^Y*?J
z5A$tH`ICgR^7_`spy;Mz30+~xazoOaSdtJ3ITOBBlcUTbXU+h)(!x_>w_l=Q&efug
z?DJ_Gp89yAZPyzMEE&0)#=0>fEB>nRYVh61YIJL3Jj(MX-A*@8offBC$UIYp*o3e_
zKP)M3&e=f=`f|JM(1uO!RpOT)BE<x2ygV^ry~Gc0=s|0HWS5}BXM;bm#4qgCX!5|x
zw1PvO@Tcsl=S6^re5nX?t6lUpfBKehmNln}zABUmVww{L5=j)Ws1*+@tjVxZAce&^
z7S~p(z<Rxp2{{+I(s6(p=;uK+<J}Jpjob+XBT=(;s!imZc3hGzEpBMfK(PivY4VN}
zq#ZR8q%;A$Kox}@XiId6i7=(chbe|+=KoQ~4yko>L?m5`M?%COOO7)FZI)h|`8cK;
zayCEA|Jd0aWwheeLUW6$TE1!D^n}+w4(p|eHPGO28pQhu4XB^>?d>|$77Cu%t$!*a
zKw>nrx>5C!$^fRj8;pkY6Cn}l39_K-BczB>6-fMYV!$+#yd^<s0+{fuQ{B$Cg*1z&
zp`0glmN}e(X^%@$0kZ@4HY&h0{psfCdq?CgH8S})?clww(AgZ{!vpYaJpzHn&D1bI
zIS7g_0&)=z_=P(h1$ePxi?*W|K_*9g)ocL!N<HYHljA;ez_Re>UIz9NV8!FQ9SI)Z
z`&ND>O&?H>|B4tr&U5#iQQCRD7{+#W#-N%={H%uoB}saK+DN3~*3jwl(QB=du{s}$
zxo`X9*L1q*zZwj~?Yn;q3X{Kf5w-V{KsK#g805%lzArQO=)r@_yh~@A=9$17T38qv
zSlrjPq(nuzRRMT*kq$%*4J$mw6jS8xV^xWDN%E3T2Y1+B&9BD=OYDO?NuGOFeMFf~
z>yXMxGQtGrwrYYK?PA(A7}>o#{qL7FQ&cUUZOGLA0UEn`R{Lv{KbJ{OG?d8SLtv}J
zz@ybw`j69jtD?f9n#i`yCmmP&;{>1P8Gcf+p^c1;p)sw`uv!Mz9;}s9hO#6n<hvtD
zL5j5qJ!(U5KBG7YUa~?zP#lWB?g$~I@_kIj(~R&0W|Xcr&_^K?<cmMOhyQ1;k-T$;
zqxj&8C);+4jD;^Vrywa#>uFHJ0VyK-X~P2>$k-@HN9Rr0jnRk0mB-77ez=%cod9kI
z&b*5wp!!!ozXeU0{~H(Nl(@dm>iT3*D*i|!^^+qvoJcrG3n(ath7c$ze0!7Q4FV=s
zs~B_v0n=f&0=Q{@UJ$8yPSzaH(=j7h8__F+8YuLFh%}-7_M=)#z;!A)$=}{^6W^bz
z7Y#tK!Vc>-l&S~{)#6W&;T_^9y71OSp-Zv=r1nJ9Rc0v{d@{My=+55GjzXF&@+{T-
zzi)LgVCXh2SAorXE@WPlSTmG7<6r<N3SI3{-r~5d8jTb_AN8Ak?vQN*w+}gx;+(iR
z5B5`~k!T=zzve|FU*OgAOX44feZg|WN?Q0d-U}=GPvP*6b+P;<_85vtvB3Tc16r7}
zJMZ!qFy&P;UZ$rX#!U^dCTvF-y{Z-tU$5*o0RE!(&p)jar08BqSFw9cY6<_;A+*)7
zpda-avb{<-FSt=f_|T_=O@w1@RRg+Qy+GyPcAqsB?R9D;Mx)Es5l;BI(Or}LI_|Ds
zTvD5>^}rXP=PWwn1A(b%1477m?H63k2|!ZnC3-H3u#83?CwQTAsKJ9#;`qZVaSq3S
ztm^R$OO}3X>I1YffLjJhR9)&RD$(4r>lIS8L>hd&1ZxM*BZPy)s446jV%vcwQ|~4r
z#*)2BYryg=w1%bDtDYXxLn3LHYwETwNGdk%4?s(F)f~@6>V1zutNyoE`l6+DE%YXy
zh^NbVNCy1{;>VK=-5`q4UX6l@Rcmv%t_8%GpeIG(=txm?NWeLI6oGlKTwdotcO8EU
z2zoTc-$gW3fiCcX|HT2y2jZ%jYK^GTnDZv+B2Bxqjv^L2ouynGji7p&a>eGA8rGOd
zUSs*0jfbCNA0c^R@*<+1|D(nL+;_}Zks22ScrL!=W&z8$ZuVC24|y3V=%)5VfY^*&
zb$Q%=B=~nWq)YjR`;%=`u%4LK0^6!#6%T^omzLRIpUFQa0wD5CWjNb)5j%_r2!YHg
zdE@ooIR|;&1`d3T({ao#mjW}m*Xo>(<9C~NUwbfrKEmH&>Hr11D1v>dlr<%3SMzJG
z#uhCQr^7;lqeR7>L(cVhLFS=`Q>W}L03VzOxM3t4n|fb`$(8fY>n`->d6W0*LaNJ^
zt7($==6;CKoEyf1VjN@dS8NlTjBS<rA2t+u3|HW)UYDf~$~Z9}LdZA+b|aXbZ!;j5
zW!zRE1Wrr2RG`?aBDs^2r$-DGneH?256RpGE&BO;!b_beM?0KOt!9tDAJl`<Xziry
zPksRAL-_r@<%;EkDPJKxTE>4zK(P-6jO&;5&T!OpPX!|bN3hMzIdOCNn`<Z&5~s74
z{&=p#R9q5dzk(rd-J#92d#`jA`&rT0p|+KgEE$UZxESTOAt=`273Z+;I<x=N<H!Ke
za)w?9ia!OW4&e72*I(yk*HKYUjQmi_)jp8^-pSPT&=r>TyX$Uv%0Zt&v);;<=Z7i)
zx?_HWA^&Dg-Vz{fSb8mUe{qWc%my+Z#Y`3NB%s@G!Xg(%kfuP0?0fvyt`?ANki$!f
z16Xzm)Omav17c3fS!zo|aJJ}|9mpUKENK$KC52m8!5m(h&-(Y1NY7mbE#p*yC~<{I
zaNOG?(#$}YPZu&oy<xNBvUF_p-@F@HP_TwHzeiaO(|SL65YGgGj<ALwE@^EAoz3Bx
z(NK+@Yj?z)b_)dLGHi<boNcwAi0+UnaIuC{mcoIlR@F5n2Jb56r?3PMGkmt#=h)}I
zpCz{4d;3_=Ti6Nl_Jf8GUsUDC@F@84>w&Ox_8;9M%YUT#GBPp$-;(G5Z_87QhE&p4
zJK{em!+e4gSKz~blQxI+=uwQ4R@baLN3p9CrDPyUDybr2MFLMB@6I_GsHL$-Q-wB2
z{=>(|PX4hmPeKyD-4Az{#}hAWk|0WjEP2z)aB&ieY)Ev{gf+)Ii{ir#pP8S}lkNnO
zdWkcwa_K%y+xY?A8@^ebT$@tPI0a(c5=qKJ*XI^Ztm$7TIi@B*Z|^VRoqxCxg|nTb
zA-OlX=I0?gKEJ^&TauwH>2DFGoQQ7mi%$3#JH^Y+%Re~}+P3hm#M#8A$?;fliO45^
z6C1{57&315qvK@MURId0^_6i{eunF~Z|hkWI(t+!jHh;6nHTIPGfO9LQpH~k*xGEw
zf=c(Y=|Yx1W^L8FYjhu5OY>5dHowps@(N<0{~#FyFRZ%iFly#7Y-*u0=##iCyROu3
zrrxY3ER}BoN&_TcK(rv8Eqps_tXO5b{0)5)&lPujEUxG+Xf4jL*BQN{<)15=mqd1s
z*T>uVnO}ua$Seon$CmnDm{DAc@MoS)JlA>AW>8J1tiYrrzgSRss_U*VeLY-QCM`iK
zhCWwfZpohSQxEI&fKE-VFz3);q|^zqaZiz50f0V3NY#>(X_t>fi#-kw_BEoz<-6|m
z=avvo91HFi2#OMz_~dD*x2sx&1P<3}sTG*6kQQP%va|$R7%)d*dd^8NcnK_6P@mxc
z7wrTs8sH^>1E?f?t%&l^?KdWR9$lYDe*pq>x~_jFjX=-fjwj2QHZ?Tp%B-^}d~VJ>
zP`bZ*e`|8&<`U+yml}mvXNEXM<8J}rHpmW%>7F76R(jJ97%>3J+`;{^e2v2sr5c6>
zXP2F7pdnFM)G@oNYt6pKCXDw0>m2E3XHjpdscF)e-1WV>>v3=@;jei@6f!Sfo?}4D
zaJ|mfQuA54jWl3Xw6IPTu<xuZ3>#C4+G5o|#U*Pn`3EA9Dpu!wXm)~qHdTk-A_J<3
z!&Hqn)Y6aG6_`uuqa{a%@Kmg?+Ou!F;WpNM$^^99jhR~&Wk!YhEc`cvEOjQK4Uf$^
z{PAK`K}ypXu!GW&P-So&9Co*Wr+A_EOZruZ1fo(%1I2;a4(_GVx^K~ZVR>qD3o5#;
z^~!t-kd;DyZ8dLSP$58K?D14*0v7_H)!M>f19=g^v{sl5bwD`YW(``hBm_#aB3L^c
z8kRXNkgs*7_F_YX^nPuNrK^ynE2OpM<W=$CBn+cHIUFLNIJ8>UYivXX@6!bAq{$Cu
z1Z{{PmSuAmGo~XAT!(S}HlsfbTJH4e8Pc}5DL8yCe-4G>DK`9GP!<ahLu&wEi|Wr-
zF}}{LM;IDxfJhTH^o}f$0EjZ8zd|4?KsxAkw#o0;`X{gL53Z0E*RK88nKcz}0t}W{
z6jICxFg#H0LPIzcExmTNw78~uQ1u2$<DZFc*pW<OYOiO-F>S?y6Pn4+vJI1#+K<=|
z)T*Q+*+5a#%tn0Mbv6vvZ&jql+-0zbdO9C|Lngj9y^;{g7B?o={uta16AHh^f~AGS
zz`M)AaBicrzg|5xyhk1y(&+p)kM~3H6pz|rlAs^?%IaQ`*YpZoX5hf*#z<0yTpp$m
z;!h#5rs^Vw`UB^&W!hagK)&{u!{n4uAe86zyN-6Y6@R#tbB7WgI|EqTNk)u;U50Q3
z;RUB)Q1*0-EnE-L(cs>+`F2!mg?W{VWT6Xu0->5qW#5vJdZdD9i$@6tq7aR>xY(2t
z8o=K$x@K6K_Cs>q3TxqwvVg)28mi)>NKixORLQyW0wcISW1sA>2#rI4mUSI8=C5t8
z9Q1t#jH3l=oj@pA!SR^ql%s8|JB}+0Pi4WYTBp<&V~r9%wwLYTXtp@KWZt)xrOMba
zaM)bwjWGlVZ>Zv%F?g1>8|9&>iiDk7=S4A!D78oiqu1nobEqgyp6B%^9z$Y7BPUuh
zxGmagJ`+uTT<rRuI;Di6Aeg)Fq4WTP;2)M$G9=)-6pvgV(8^O_BZ;;f_1V?#sP<`d
z;@CW^K@D7Z%9z~2pM=4q1;hd11+AoG<QPPp^8uVdt5`GWCpL1@b?S&i+QP+`Gt}pq
zYjr>ifRP^E;Wh7E-lZs$S&wumB4aaZ($y#|J{k)H8|`17uFgyLPqIwOW7@G$;TbEa
zFJmZy-q;$B#aQC2bUJD+)%XwW*JrZM?zzh_-#7zb+bhtdA4|5)e~dlTMnDl2iu>)u
z-#;T6^o2At7itOND;i{|-KDYdr_8b%&)5QzXn2uWzy4Y0iK0Ux5W+z&$O9HJFkP-J
zUJUKU^I`4CviW+H9w}nVo+|r_GT`Il;?4j?OFH6tv&|&9UCvTbKnIG(SO7#*3@qK*
ze7e9+${ztj9QN?g%8dSo0Oo+MIs^MSpqXx0Jrsr-%WqQ7^W!)rf6=2IwGPf<AQZLj
zP`?{5G}L&$ozCArNC}t2u%%83Wd@k@4*!L*^c<9`vrpyQ<!@GL+F=#z!oe%C@J9)(
zVbrc`r-Lgna1KK?974L8V$ljM6bs-M4O7`eNb2hz7b%>3y-#}kt_rh&pf(FlKQl;S
z6H>mh!3%Khy2Co}c%Y<o9KkD!0EDf87{A|C7#BRPPU_PQ)ITO=EisBf7f?3d7?3y!
zlTTpJa)_dpfVQTP_zFx4=-do{vR#SWlM#BqoXU*ST*55gI=D=WQp+;PeX65ijIp2H
zbme(pU0#awvtK*9KHXyr(u}T3R*VmwCX3pEO@xJsq6xF7Fh-Np<kHyrx`{J9o&-t(
zbEX(97KgFHQpo*V^`@Sw&x(*_GCjAIj{sqniXxesrnUAJaY7k@_%VUH!&=A<`}tfl
zRQh{EsUNzsA+Vsud;A&Q1yKpw{U<^uuVU7qWtHh7v*qSDM9MEI!aD%}#7>$3hTl=J
z7{*K3V{lM@K#8we)BUPy0Drf;16QYg;Lc(FK+aj#UNDya2Tycvh3n<*hxWB<hx(Nv
z>gRX;4PDpJZ!&>=|JUy!4m!_p^hGoqaPV1KP>AqOc>f1z!1piUQoFv2(wwgYzAQMS
zOCUV`sF^vjm`Oy!Q6tExe5!U;r{Nz%1ncmZ6KwoUu-ksgWIP+petBj*sSK@f!(Tuv
z+^hKiYQFx<Z_dv2zu#)rwd}TAk^Zgf=CxY<(++eNDL?jGWfw}ty|Q^gBbh0Vbv*wV
zlXfJoHYh#!<rRb`3w!tkC=uL5Q7B|*#0?#=2`!pqFPwK3)v1Zm=`v=DU`{ZvEnE^I
zaVjbzQ>ip>3fmcpkZ91Pls;tH$@+40Xwh^OKD3cX36D~LvJd30)bZu@4y+}u?Hx|h
zL653?JHCETtnuY?4lLR6LrVJnE?u!jxmDrob|)J~@+%0600DRbNN+HREeph^{`@AA
zlgu#3Ued}$pmk!oFp?FbBwGe75pkpHni;{Zn*`dAe8qQf;w%m^5-t1JP~0ssscY6V
zv)-g=xQ(=I?7ZR;s!|n_NEu+nU`S`2M~g5{iB~R~s{|fjF7^i-Bk5gJBh*#}*=Gkz
zBUBof=U4gZt(~mQZsP!OFw7c@wyPH0ch+v!Y7{^C?zDsxkBwek^HutI*kXk><yq+*
zsvV5X1HaEw0=6CgGVY<N>~6pI>IW;{<P;ERkO@m%6eO#;m{it=@S`yITw&II720}U
zS!iQ6IlL0$eY|c%-e@%q2;kGonAGxo&068!9Mt2O`LmN_Ad9r~hpX%u+}}jbGV|d4
z1vp!ed&S^;Lfa~I4A{gDPK)wvtKvghLF1c|vE`Abzm`a~FDdlcvM&1c&GC7Yu2@=G
zz!w{WAuPmTrX<cbTj=C&{1&9f$P=yS)x0WaSQuIG%Y$CLF_$DLrUWETrU(!en9Crq
z+=z`Wo(vDJzjNDUv2N&u>^QXfdR^%nQB;D``jI-DAFBztPjnz!Us6UdQL7;c4Zz<Q
zz^sJx<Y$>Ck|ArlPX*ERd;n5G5(ILR6R=UK0<PR4p*3Yn|4d>kR6q5&Vyis=B<HfW
z&9neA<&jWrs*uO(5WDO~dCuM$F@1>u#LlW4_R|H_4$b}iC;aYCMmQpZ-C-`6j)>5G
z5o^9jWhL&>)MGil`E>hdWZ&N*`gpF7hYuUsmXi{dC=}Z0;|&o5<BHEAn*esCBiCey
z{-@c~a0B*Z8AaBY>zU_G4{bC)Rwh)EpNdzOv#Zs9wGKG)dNg+MG0?(uYg0Qd2p(@I
ziJp`<in}&IfEi`Ni%fa{gLS6_fiVZT(0aK9@q9?*;hn=K;@isvsQ5~u2xw&*;cEF1
z0ha_>?TQw;i8CSm)o454JwrA{??#y@FLjR<NFqR4@A_r0N~Qfe+0ZfTHUg%n^)$At
zY(*Zr6l&QprITyb)y+hDdgmuBStOrtVwk(wPR1$2=O&4a^CNXSr5wG+fw<#NB<RLF
zTP7=u$XPafMm|E0d@}Q7K_3z%&ZxYV3))utxzjoNcsrKx-skwLpE`ELuVuE&6hHg8
zSjKRi^Qaj*!XpQWVX}0nJ7M8cy{s-flt(!sF2)GemY0okz*G?v0MtxF^(1lyvXGb=
zIW9n=zyOo2(#j3!D2K}=4Y=6A>T3N^a87I1&4Bh=5KJCLI}M8ui-$dK510M{+e+O}
zI*9i~lbN(j`R)N2@@gY4t&^l?V<4vir={g(>~ThY#f51?Qa$G0`38mZ1(*wT>aq{Q
zo&h?S@4w23;)Dsh?tG?@i|r?gk+y?x3VeB*ujDAG6x$it{Pfp(f4T6&p1X%{0iyhV
zMc(l{D~>+fw4dk_VUIj7>y{%QLHSd#gKM~Sg=r(#53IJX_e-mChD$p$_IjAH+(F5W
z?S64DhQQ>t%d3lS=U~UR1Ge<g)z3wR4cAq#qbt%AWi*EPhAO?LOsTps*<`@F*rKZq
zhYmj<#lvh3=$wRfk>S#Z3t94VqGe6V?evQuAS09^k#gh{7%)4YrG7RMng}FM`(h?F
zQ9-?j?aDiitv!UzBP~w)Fsje;Xo`;nmu<CLfcDN49;a9c&yf?EQsRI=EW5YKRz|(w
z`2K>{5-HXhn1_S4wmF#X>XW3lO-9#Htiz7WyKA~GCaF>$nr5X|z205Nxl8L>aR2+m
z#=btwHPGeLU14m{cvgRpsy5KL_UjwD+kn2qxuy(m@TXAo?iJg)3d_kEAB!S9H}|XR
zaZx(OO1T!S?Ipzz$o!f_3uiI&;F4Y)EL@xn(rn1%Xp{L*E{-NTNZ<kGl&&76fY;}$
zoon<SaYXozo*r5I99}Q(#o)Rwp;An4<4#J)MK(v+e9Wx_j<`oOW1{J;<KoU(Frmh&
zj?%WY;Hx|9ey=eL!MPY)%-l>5Qs4s)o*$$fBf&Td#ga|MuZp`tCc4n&E3c<J-?_Ni
z@D9;zGQ>o8B3j><jPWnxXzk5L+jsVLRnC=AOEgd^I6~2l7c?esGG9DdzCDJ>Zq^pO
z3%tHj8k>>(dGo2xPuoC#3OJLVVu^v?MSVDjyGtHYCT#cL4}j_5;-eIAu9IQ|D><s!
zg2l?g5GX2~=mb|+=rhf7#vk~bMVRgXD7&nT|EENXftBNb22uY1BvSv3p#T318H*7i
zd$^V7bbG`O*^+~jn8!JZPd@4TsW1_0!efHo9w^{FW6pX+PB|S>cvRHRH}7?=y))jy
z`CHGPsn=XA5}HLZDC)vBjUmxt1V8)1f;D1KR7Cn0m)Fa6xc*LtNV%Xe(t1(u^H*HZ
zTGLR{M6x4}6LW@Rz!5q)a>g*zr_;Wu>2Ei$_faPJ7@hXf$iWjIyzKUJnjdaogWlGV
zE_ZZuakkjd*(rCY)%mfN8t&{X@9hTiLAY=<p|tpLgQ%kD(@O4*Y2c7EY61uYZBb0b
z?CZ~uCT12g64<x$5WJbKLfzRgb_9PN&^I(%HHaYuX%1jx_a!B*CmSlp*ce~Ot{u|D
zMgke`(vi6&W}&*r61s<m%_0pOw3$4@NOo4tQ)aGJf@YcS0Fh)d2!7x{Ur<(>?@BTm
zqi(T^r4~`G?P{rWPSx`$i);*b)>|(Ou*yO1g`<Xc$Eiu@b|<W`_OVo-Q0ixdZN7Gv
z$19!@7Hgx}dcUF<SQTOkmtqtMWo^!KnX6Fl@YPK4d3*abnuhOfr{Wuc#HEtJQ}g<1
zl+Tu_LuOa+ji-+ndqh-BmaC>Rx<8qyu{X*GBfCrHTuh!OK2|O?lU^x7Xk-ky^_&4#
zk5ml%U)4K|G2+onTrG)Y{{gS4{UKEzogl}^_LCS*meCHOOT4SX>B{THXCDy$zB|v>
z68uV9@1yE^QxKm*yN3g@qQGE8cCH;6TJdy##q+Lqf}f45`>D^N(?~bgG7yJC;5jb4
zTU%a~P^e;OreE}59T+MnzR32wS5H|>!U_3HJ@PmSL!P@_GAIeEPi5Awaml;9sgqSl
zDU-?l6bqL7IjlNgc7QpqHyl;AUf)yyEq0#>EKP-h5q6~h25BeZkgzrDLJjV~8$2Il
zSgXM)lg^37r4zFQx#04I?N1bhi$gdt2qdH!1)!ZCmjKVo=b(q~6N=SFjnuNNEP`~>
zgMKyLBm(aXxTlF_G3w4`vo<K0G&qk?%1CB(+3-;P0jQ3?KrGc4RmU0B^kp~CrRtrq
zfYzD8CwVMxg1)HO?Wzu0wX2d1SX;xuK$Ue{M`YyJ%-=jGWIV~9UvWh)weQ(zOTb@{
zgahTEa{ULqir$&CAqWQ}ZF3hc?&0bn$I1i*-b{?!#PanA!fOz_PQt!L*j8&uBixT?
ztAJc90-+^998({9HZ>6*tA`B$D5+ozz0OY@2qdyuXTKcSoBB|%F?J2jSWHH|1OSsm
zRgX)D&g<+N9Aqsj5JKB~Ja+4pSaB#phLykRM=z;lN^>DwTcyFVR8wsut4pRbFI@tW
zDBTM7Ek(uQ0ieZl4MCqUO>LW+a(M$ixJ*?q^;Zsgr8Rnr;t=%&M;IxU8)d!|ZB<GT
ze%n&e^6rqbG)MLH0Lozgf=9goCh_Q2D`_{r;EBzzF}sre=7Ab93xE`q<l<zDGT89&
zw_yr~HH6slzE}y0L@TOVc)@e}{#e<u#BLyY@$v?_*JQkuO^XO+cd<_}gdZ2+DC)QE
zHrDOCoa75oq=cfp4{ODjdPy=eBJY`zjt7r5T8g1siX$eSP&HZIl65|WBYIsnkxV8e
z5qhUHxLVfUHzz!&>t_J20v-5`gfm;#;nY}&qzDq;9_Lh5wAUAu2xAstm+RZoI?uMJ
z8X2e_7{MOYN+wQ!VK&@yc!SGwqGLvFL8Gn)W0c^#Sx$)39aOB-p!??S^lN=g*NW|V
zQlvsrRsvuM@2XlkisR#v5CsJbhE_8ks~Ouy#&92cO0J$kf^9uajy^20PC%X;wL23I
zs~LT>N+f!!*h>ixw1I($2SW>a(I_7qzR*Y>Gq7W)3Nsv-5?@zU+FBvbbL}db-6k^)
zK^kP4NfF}Bk&tYc%t{$tkC7a)@!IRAXVvs#^|STdC>#-hSSu~ejzxH9W;{00Dg>#{
z%X_wAS4ol!NeGFnGx*crf;5ukAq~g5iZbmcUOYAC-)3ICSCzYVm6i+L10h~pDE@%Y
zV}PLbs29So&#n3ah+wdNwELQe1G0%~K|Za+x%_!@$fCuTU%^%a`-1gX+i!Ti+Rs&N
ztmlv?z(eFGEKH&gy9V1v3DR;F`vp*hV)%TUPzt&kAdS2*TBfMY!?Iu}uqm{--q|@m
zdA<FWPXxIr4j4Zz#E>FKL5eBb8Y~Z!Y1iF?7z(WFWVE-46CC|{DnAb*U@8wxr2B1I
zJV9xHkHmTT^?i}!ANBJoh0*%2n#cNo);vb~|J~%ff@u|Z)cWxG1ND*>U5nW(EGe$l
zFL4`54QJ70MM9BTxWCybGn|McwJR3&+U7Q1V64#Ame#cewgul+96EF`ZJ*05(+Vl`
z<NmPnco?-7jV2)gHKOt_*iw)*LpgYo=3lU7UkPS{G3y+POkJODw~qGADR<EYTU18s
z#p1UXf${rVUw7x|#o?Bn{o{8uLn>|UE={lJCzZAL3k$-)j@{4K&CB)O6raz7W+VeR
zCaJQy{ybBE^H4{AfQcm3OcPB;?3m4CZ%LKh&WokT!xLzbk;NbqA%qFdgt<Votm;Y#
zFmwu_XD+>jaA}fJhBa;Pr!pQmWCj2`-M$`~y%I>rB^_KoN4ZSI&!InTiq-N~1M{c2
z6Tl}fknaE>KhK=#atD!z!n0eW-R-22u=k`)^zWw!dwSsFz25BoK7`{*hp|#CU0%e^
z(!srhwIAA=U-}Gc$%Vt3H|Du_tcvuv^K3bpMnNJ6dhikshp*DFIAxuUT47gBX6H@p
z=&H_Vco~DBhuscU0w=jN>O>=upxrm@TuSnVj8o-WO;%L%Rt8+bDZ7z}=AoF5t<1jw
z58*~0Wcc<?5W#ZaILA^-@PCZiWs7`AbIq$6(@C;+ZMe<{>%#&hc1--sUT{rc2Q9Yo
zJFGyBSrgBl46HRc--3>ot45J<wM<N9G%ay_xb{=w@O134#-nrm)?M|n$vEbs`!U{p
zFZpg-vcwF}@8t(s#0-KA<M7?Relffm5_S6F&G6No*;?0G-^XS~c=irT2kw@W%^BxT
zhqX)oo!M4)uI$+Sz(=HB6sftZlOeD2mNKFCL*43V^GPI1%mZ$C%^fZ7q|(w62OCbI
z!)E$+i2>0<Q;_m$4LF9tqA(iP+s>Yv<WF!Ca6H8}0=bpErLozv?~v1BoyenyNt<-&
zfkiE+=2O!<>j@vvgphq*V{rAfDxdotmRe?(S`3`cAPrFMXPB@irhr`;4Y22M3-JGo
zv3HCSt&O%d%eHOXwr$(CZQC|?*|u$0?XqpF_euIDo!l?!^LzbW$sF&PV?KkUrFDmh
zYrS=eDkQO+gP8_O6hwKTSIo;nD4JQ2p`;9Ml>ANOXU7V}LI{m<!29ojA$YOKQ3ACg
zbO_4mhezc;LFNw`0or>;GXWJgkUgq`6K7eeq#}&zfIWUG&;-Ghcgl-m9LvOU0MJA9
z7)j{7msqF&Fv@79X()+Iv<9ww>mwh?K0J$VN%jWLP3sk&X;(a@5&{@CV&MgF`C|;?
z2zj>7(_MHM1B0=!$phj5y-K;Sd9F2X{(%HVyCsAq;{6LS4kW0iip=yOQoJa!FMDOQ
zju78*ptt!5mo5Iq?BA{qDQ#`dQ+B^XYlQVraR@QEkjN3$YA$*I^$lOl)~m@@m@Y4F
zdq>OdJq>nTqH4FYoxE4rT9yN~;Mqrne3?%L+|IHlY6QjcW>(a*l@22Gg!9SVj+UA8
zh-d>CysI-WfUuyK=5Xzc!CF4vj!bEJ(M_B9GNFJjMQvhKWH}n^m`);o<k~FP4*!_f
zY679xcjN}ue9%v06)VArMk+#RuI&5aXucF~Tta^+Va}KY3^sz0d}FHsNmD;4qKZdF
z<8K)7y2(~pZ=tdqP5Pq<_X<c=LfKSd5N*x9r496u4RD-r>Cs@E;e7aH+YObya2j4P
z9e;A;n0{oDd#Y*{4lEonv~P9W&ZUJ=|GVKxZBF8Hqv5^;5oz8RMNI<NOz(?3&LTK{
z^NWB5k3@F5m^V?CX5t25&%wAvzkTbH-6@M?>P(%?CaA6M6M#-e>aMojb1)<`1XCZ>
z9ArwNkE(31-k^V}h~bmQvw;BCRDk9hdCI-ipvWC+-viOQrwZSs_;nV1!-Ln2CEyRG
z0!$cEY)5dw{Dz{_THz>zC0Xo0*f?$((Q9w6bm{ll@wlYVQ~g>LmKdGx3?V(=6*#Ly
zPJf(@4{UWF|Ae3?q*_Fzu#IbFaxIg)pAbr5A9PY*$$#q2{CUph1S!cbSk%_uk)<YY
zf_mi&+n;9}gEK!vfNMN2^9`ZzO5!|Hwh%9xwCcE=gWdG1_M2hZr}Cz13L*<5!G{`c
zl5l)JlKnH#&<BRa4e8UQu_@aW{?bC4zm&|oWeuc4O>%l=%17=%8iDri%tH>~)U*-u
zU@mi@sde>vYFU36L}piA*YFb%H^6JDTSU3B!!7fDZ^ku=@{`?X!fYC9kE~yo=_FB^
z=|jdGi~L5OP4$Z^%UWB)X*%B>qVur1YIYlwhPGp&nzzIGsmz2?gM_eEg?*V^>jcl-
zYa941SsBV|KQttzUra$G49GX^3l9UEkaQTpL;F=2SB79nl(*5P@y3T_82Xy%h-T_3
zsM8c~yR_V+7e>yQ8eD@)zm!}MNy?;5&#|ZP2Lima`vbqtdEv=al(S+&kPY*FQZ(|4
zCBwx-d#gKe`^g`$6S|g@kSx;aDqkYsO~P~ru1k-dd!I)hA4p%u0_Yw4g27YghWd%8
zsjdA7>w2bwelPsb-hELX;72+uUCvh6J2xI33uVag;}7Qrw4TZ8Gzx{dtgf#4-o@9N
z+w++^V!5Qnt_E6Ew&3A)GJjk^Ap>uB)c0k2*MZK72-7RVRHVU!Kem45k~3K3svR{l
zXV>R(s0AwY<E{KpXw+y(n0of>PS9S+B~d3Rg(@L)BN3Tvyv{E;xQq0w#}UerpdQ<6
z=VOEpQQBowj0;mx)fc*P5sS=!Xa@HGdMC3o{=dY&|Nm^jjK;S8zjw0#tA3$4W`w3+
z1X5@%@FLr_7K&~<7I2swe6!3yi(?}7m}R5>`<nZKLR%@P8XximJ(&mZHqL#lD_=A?
zIslAbT)w<moSq1=ge2AzM+QAG0vK~7M2sn8HW;2z`Ea+q4!_5M<=TO<7D5_h+(Bl}
zkIR>xmm6Qcy`+%jFfzvS?aL)cWPkZ5=V11!&E5Rg(d*-dQ!X!`*W1}EuDZ_8?{mir
zThV@}zg4ozZ4kL=5hFD}TLv9e$~vvn(j1IDlM-7wJ`-vNhtgX2G8NG1r`+zLQpBov
zK+$~auTE=s{-YTuipC7AmgHiHG$zd02T{B2<<~KL9#?i8Uu<=sWk1mW$czXPnW7BP
z4J(s#qO|_HAQ2=&Plap)#*bs|*eAe(5Xgc@6u**wVcyA(Il?lN0s@5wi{uC=BBVgV
zO-~<Y2zp?{Ph+Ogvr4v#p9i@gd*~p^5{cmySp*`DkW=g8=)BIoA(AqdZZDgPerZXh
zPvV`~zCMlPyV2%4Z%|I9zn<y?kn)Q=x`Z{Di8R;IomkvezXOHU7pZi7N|NeZ$RxH8
z(Wzii+zxer{OfKdLkF}&pq=H6_E}X}`QRQcv-K~4c61zlQQ}3`#f#n?`I5%@l32KE
z#gTsQj+qGWqDMwqgc0StA)xri@JAFx;)96J?w%&pG4}HMt+ZL*JI4!W{o@I3JU+(m
zYfy=~2BQe$YK%&*UZNb`T54y^fWO;pyZv%hx14qsrFO%sLa+DOR`8yKc`2l06n4-8
zI1p42N(J_zQ21<*$s8@V0~-w+(qk{TNA;3j(H7Ba7Y9hdH5?7Uib#oO=Rs@fTyrlq
z4Z9apdq*dxJw%^r?5%XWq&OLJ%Mms}XxaCu$T7A-->8zPa<~MN$P87{)EqV|!;xT;
zNm!^UhT8Z0VPo*{2M7b0e-ylV0;x@n(a^R!(EqmdxqDR5<AO9Qs^exSr`)ZcWip~^
zMzZ8&SI-+LMnS+(tbWflZt7{+kfIz}1Y?&@>?IYfW`b}#KPD>aWp%^@+)c;C2KG=i
z5-@C?J0`&CkYK>-(>j=s9XL;QWlzRkJIW|YkeYMf)v=YeyRgTr{Bi#<XsBBvmk>B-
z$jmg=n<cN$vjK2xE!SrEy3tC<ZXcM8o@J`9xUYUOn{AIXWFlf*PRupLt4#ye_~&zQ
zc-AyorCHwvMk91vgStlSS9v#BFROdC&`n3CufhmAn61XLZ(3D2xCGu_>Z*t;`p!z%
z^&p49azcUhR|6kA0s+h#^1n1>UW)B*0D8dudM(;peur-k+rG7Dyte}2DZwva<KbNj
zT#0(+YnXc!`!Vc6!t{@A4|pq0^R#QwJKx@7Y-UK{;dE>brnkzABw<Ja11Hdw3}P|s
zc2Xf+L$_V?NGxzP+u8vs_+LhGA-FR><HLBh+3%8+BLNYAVhw<XYroIR&4NY(7C*&x
z6oEuxe}@lbGLNM%LSF@TI`VF&TIzlPPtozq0fPLsv4{~55nHHz0ixFVI=L=v=@`3g
zbwB=rW3OKyr97vmFaz(NXN(|?ArY>!DKA!hgPEjtNyFiWTn{z~rUbzjSPzC3$1{N1
z0&j;g-x|f=waWtj6T)Z>WX;{}n%i9KZ0y|A`?wf|Jd)fC!|-N@GDbn1h+`p!07msn
z0E4F64kR7osBefLk7NXUR$&!AdX44;A#J1lQ4oqN;;|$W9Sq<Cc+mL$Sk|?K^3&kS
zlln}QBQUyYBS)|+v#J6imLflU!QAN5mV~yp`PbqCCm?M^1x-c?^M^}Lb3L@mvV9w4
z?zSE*Mi<!*s|soG54b|6w^c(wP9*19ub2=x%)q3uR0b{X7r0zJc3|_6U%L(GJuc8`
z=j0Thtc%RB>rLtA)OtA)GI$^u26d!F(IOi$dn_IEKjhBt8;9Z5=o<7HDUs3dCwVW5
z3<`XlF5DOQv+7|U7N-I6PpYp^xa7_8=it`Z>-=#zvB^>&dv0F9e~pg+I6`1#=lEY+
ztlJuz@&9A7?&=q|P4;K)6G{;!p{(nOMiJN|7YihUfYLhK5;h)bJlOp$`t1>hGnI~7
z^EOFD@kJ~p<YYJJivoKL0OQZ)$;{#L82AH)qR1E@JmE=jARxwAm<U7Gw5LlUDI#W!
zxpa5E)TC~1FJ??<`F3TKBC<csH=Drpez<mOfB$^-%<S{-JB$H6x6{qz{czE-`}Rf2
zFtAPU@A>WaWcctHjGlOAML=0d1PGc7B(uRvzFZ{}Db=1b;Gi=XTvfL7*LCf2V&dN+
zsL}tK0Y+jLl|eS1M*n6&Js+%52C8K}+rayTy}PQDIT5nJDHNX4UK!r8EH-CB*<PE5
zMxP{Hl?s8T$N`Js>G`vkz4s6BLDw37f(tO`K$oM8Wjk9ZM!n<HwXihoD2-d=SPb$2
zQOfsRY-j*!#g&H8XN;OxO?nyQWoP*y`U~|S0TQ3>Pj5%$;W${`ir(H2Fj{#0jSXVF
zWTi@#wxVv<v!eFLB|iP*18=GeG`yRaIcZUfh|OuijFqc2>C<U05`Ub|n|gG&Y=8t3
zUGWg+9c#Fu0?yvu{fmGU5;aExB6*}#@B|Wx9GWL=C|Kpxv`MI5yap1M`~pXHC71mJ
zbg;hXMh52_;Op5aedqk8jy!bmam8&~D}h6y#6P19s16>bIc^r`ugKCso^qZ3`Pj>}
z9;Jq2(gR=oHLDY9J?3Q1tRjJzB>Ew=uFZ5rD6@1ru6w@}wkgCv-VVM8q*M`$kX|U|
z>C8dKsse?wHh)(aoM#m=iD5{^0NN!@Vn9=M5)o{6C?zi>WS3Roatqz5C+Apvg(V$q
ztvh!jR5`W2T^jIc!i_hT*GT}p29&7z=`8Vyu)OTjPuR8NWntc9dmk`YgTcIktynZi
zHTsHaWL*b-kD{Xm-{o=T9lj<j^Qz-NdK<@Xie|EKb5$c>maQ5*mABWMmkITNBeut+
zgvip4+2ua*14t+lT0<`n4e#0%vQ8q!KU7GY#r?8&t$Gy^yIf?CiUUh1Oe&+D71c1c
zrYaj3z3r{z?lYP8UiJ%$yN>lDS-4T|1IRnwV#v4#k$QNlNPG3(cxj=kxRU{=txkaf
z*;0%D)Kqk3yF&}~ssZbgtX=T&$@YF|!xGN+q*d$_7*Zc!-KsS6r;%ay*!_-2;`+yo
z!VIOwzPsu%A)}GH-au3uzn|{WN*P^iz!ylFHA=6A>jS6f1*?vS@>~@V7ZKhfZm|^{
zP@+R7x+uR!k_1Hn(TNBo{_&L6cLP3#_U2S(W@6pB&}$fWhd~0)0s=b|oJB0zE*8g&
zLm;ew3IbY_eP|m(#8dsCDugnBH>a@|?bI#!esj3Y(kk8`mJ|qV@kxbXyb<dJmwW6T
zMxHTn_=G~r{S@<71rGQ9yK!jl%KZ$*KeALez`NeT*F8{2^9M5zQI3L0mtkAvXu$7O
ziR@F=SN=(wnRC^d;Ri|_5o^jEL!|NIxy-CAR;${~wt>8)6R4O@@%6o1gNUonU}g#n
zgI-%yL_1{&1!z|y0+Umn3#KP%3~OLt!^<XK%HEw}Xp6^tZ9~JgsN$_Nz2zzCwWR|&
z?=Y8A>z%fod0VqJgSaSHMuKvfh!qcSfUIi!&3~A7GyblCNSriHu*KTS%;RZV=U}zk
zaLub;&jW5}L;j`1DM&@zRK>oV{-S_EZy<7TKDW-s6;t)TObSt8T~Ap%hKGGI6|x*d
zq)`na!+Utrfx7D)SokClyTo|1EIGc`ID{C;&ONA#CS4EPm!YAl8qlp@sdMERdANR`
zf4F^T*ckbgrkP8^P}&wQs2na}hW>VLha}{`fi<WBOdu)EcZ{&%A7Xss8=t$Q16_Wx
z(!51R+l~4v#yqm93fy(<T;jSWCPWhl@YW`2X0RxJNpMI|ukKkNK-qEIGqnRa^rGH>
zToOx{-Lbi^s$%kKRfuk*^2TK2)0+mUA)TMk_~bvfchrR@ban@VAS258wIgye^`9+a
zaH<xV!})lMNDNz<1Txlz$-wPH^7fg{2x`Q+z4uJIvH!h+zJ<R#xPgzvGSbiJn9c_6
zwZysjqVH?hY>xrloUSHNxB7B;@6_k7sT{PBcb<U#315!<<8bHcpxO|d?HniNuGrF5
zzx-b9T0OSj%p2C>n&Q5>e_5FlFyl6(j$ReYw<q3i7O4w9ITcusS5Pi20RL#stB+;9
z=MR`EeKyaC59C}x#`Ce$t~2n=8RhHWTK;m<zNglH-hZpDopuX<z(BOIO&FIhZMlbE
z5kAcu^{I=ySh!mLTJl`upRprm&9=N5+}S-HH8cHn4tKC!2WJdg8QG4%{{kim(Q%Vk
zKNj%%X6?=~xc`M-n9ASsAJdfMzo=2n%>V0YTB^C_bT|&{`>K9NtCYI3q^{?NuQVG+
zo~gpL;h|C?H6vGQQ9<e$PSitmEUonZ3ILKtMCm=+<=K#BP>d~r4^VE2X-kaxeSZKy
z^$*@#9m$k9!WEa6JwBY+)R3?$I_3IgS$6#O_%LU`LzxL{MwtthN}*Al{WO$c>!+i)
zJB_|Mk%{~m8hw0qy~5MbMEcWF#6IQ!<MEvsrjEa}xOe>w@88i`SaJLNvQK>DM!3O4
z*(kz|F6$4!bylj*-lux$<#|5^>P13~ib-6elqefhzbI1m4LT`%W#EU!Nr_}z930Pv
z?*CgO_CiMm?36ozZ!Td;(y_0sR1vMj5wIR&tfV$<DMap&j{neO|6Pa)im+_ZkeoHO
zx4!6L^IO<aS5Kg|uA-4D23;DenD%u0U0DNP3@i9i@kj00@%lG4-M!1FH=;D>JI{}A
ze$KtSE~H>FUuB_}81m+`LOEfHmv2dE8*cZ}iVlo5STL=7Q^ILU6!^e3Dh#LV@dUPp
zD&)-y0Uc_V-%MAy^VgL42cJTQNfELF80T=DiyTG3sUv)Q$DV^y+zs^3=);P46(e7=
z{A<xnicq5LW5&l&g~uBH(Z%g4{m$g7dfO9)yveucg>=S)LJIFbue$y4UqYIG!`_4{
zsbZ>*;V%Zc5yws9&TD<uQj-xg;~_bvY_F+sSII!B30Fk>3ekj;mRWK-tXGY#l=WUf
z&nTg1Y;mQXCBom|sB$XM0&d=&3+lX=6U%^tR^*ZVlH?lw>q?{4^y4KRN>1bCXg8}9
z>S|^#WVp)Ry4Kltm68pShMoCQ*?W7q{O*$aSFOR)y`;p!va}RDw-GDxSfks=%0k0a
z*e|~2m0OSY$|^Q`DzEV>a;(3n#|!L&?R-m}&)o!|#<+Q8Q_Ev7H|s(ksSpAS<i&?D
zN^n+zlPs7$e+^sDrf1Gr{5TKTGIK#01wp|0@v-r6vHLjFSZsmM7a%t%DK6%fYM3)#
zw_$_`rq6In48K>h3<xS*jmE<K-o*IMMTXh1FG29Jb{lC~AyHOAG7D^1oX<@fWlAK)
zsvMw5>;8aMpv{^ZKd%KKPz8)%Y0&rcWA1_CJp@c!X*cTjK6&f6eDwy$iSEYDtl6v>
zRJY8ne@OcNsLcx0H%X@0ebe>b-a&h(kD%jEW1hlVltQv_k43tzp}D4}hirpc3Hk|m
zxk>^-@xpu=FEsaR-5aOpXW*5mP-iJS;hs?Cu9#vb9le})2v{UT5=2eY_pZ1YVPxgS
zD%c6?%ZzoH8imwi9;b;EuR7>)sP+4474V_<1Z=AFvpf-zEBZGQe5{}F+{gKB0yz^2
zf}3VEzRcsg97}eF?EpTq^=8Wz0640Dd1h~QpC14nbY5&b1Oq!U=G_2%Cd^!R=O#zq
z^^g8V*9m)~u|Jbl6ud`Vuds2zFM++ldswy4_rU7o2+FAr1pkE~|BLe4>r*6J(VA%h
z0`cMw@{FVC3`PTuvosnEH48!H7a|gSE{lO|Ib_L>ab7i)mvZy4lJ0X^E@C&?%p{|e
z$Y~KJGT)3i$}kZGMo0z?fF4}%3+$DfrAUd1{^`4e7bT7*-0l3bcRtHL2jB$3=b*<$
z5cf9{gN+P3PenM%O?>|BAz5uqxS*WQilKu3gr@f!^UAh&s3>#xVTjP*s?*H2ooq1~
zZX{e+^SvxM=)#BX$k}-{^+3XW-rJNwzCBE_ob?Ry0?2`VuFlFSC~wk8zd5g^7iFch
zQ1=oxk|K@6Dgvo?llc5`*e(3NNOzw7%0M$rZHh*Nm&qvWYY(WNI2av=sNR56AoI_R
zkl|R>myzxj;-1Q_${KT_SPjAyEY!iQk&obNs3inSR@|F^33w4ShC4@6I4&8_de)<m
z@0E}<_EB4$GNEwnq%c9GOw83P+TMc#GlBdlJe=yR-07{HxErSkTeXm(>Ar(FSph?_
zWt45|ZrV?WA51!YV*7l&wN;?&$X}XpRR(}aNR-9h&!|;eHePr-IF>73%YRuVh|{{l
zZUX?2!0(G`oLsHK!${dMBo%C2N)RBX55cOL6b1n~F)EGLz7_LQAg_v1uNNKx2k;3d
z_PkYe<fFL(<%j!jNf|0i5;g^3yM+ve!&>QMgK7SbBNtTmnD&4k1nlBakhzY+UKH5w
zo!5Lj8v%rZimD>skyNw4;Vog-sZE9O;X^fAJlk0fh+mv+%Q=EBQCg#+=_r_8AiRI#
zIUSo|&e`V70$eW*ANO!S+o~aRH53>%Z@v&b0mGJ>0EDpZForf`T`DZ%Io0D&=MD;Y
zgWlkA1_)Y5Axgc0RMd}`n8^$FAINMoT<|ORJrA`)r>6M)v351~dh2^}UI#;g9zqec
z|ELIfF+l9Cm+>x5I2upGi*{{!k1K*5=JLY?2cS#Sw;nunm*2#jYX415*yg~mlQ^6q
z%h%9rEDIUU2;b4Q`@Rkg?B!IeN~gUaSyi9Ydsuy=wb57cTw$|VDe(H?`UKNu4SOL;
zmLRlU`K!tOdrKJ5KV08=zGLKt<&%DySAjePAOP1h*5t8=?w7!!-%<(heWoi+0Pc+Q
zcag+>f>q%|*wOna<Vt(!B?6@mQ`hh~!CZS!7XW`f2MvpjXqME9=NutZcT>%GsN=bu
z=5&Ivi_$cJSf#wNfs`)Aare4YIE?a(Wp-x(=xiIxbAqH8^>1Fm=JU)|2X_{h-v^IJ
z)CmhStsm79CrnyXD1i(eWHL}7hyIjMZi_D^1Rq!ybAc69pL^)7>TSfJ)=@NXOOY1T
zpkR#y;FX&SHBCCrR%*3$*2bGUgBy2A%H!}t=)&yJW+^xSU7nrW3Tup<+qcJ9c~eH7
zG#Yd=BU>aM&)6^bCcnY{A=FE+){tr{5di4E!egwE@8KnAAr~9bL?hlZ@I!_1$jz|V
zaHEm9JH5y6URPNNbFkRl@}Wn7f`@WJF7C~Kq2(hmJ~odnSKJmRoV-Y*L4ammXVwIC
z0xf=*wc+!k?>b?iWASDEV7%D=X4T-eoGU@+?N$?0Gw3*W_YJGUoo3(x0R4v$%%L*j
zC}uajxCsHIE>_q+QwPQ`dbodGI{Gj0H(X)fe`&7&$YwJ!urU5_t3XyXwf}X2q51tg
zj4KWr#nB^C5^{$&u@j7L-ID!lguWzLINnSgL(|boa`^MXdqn8!m;+h11hj)5$;Cs$
zpL9xxWdnqzliQ<{*W)o|fmK41sslFVN??K@o-|JZmsYi>i;<j>LvJKGq`%XxmDfb_
zwf6iTG{(~Xb<fJ?*UqD^*xvo~!vFS-)i-m9L^<{Q^K^3Z_5IL=ZcogpOH&pC1*&nS
zm4lhbd@&E2Qep`ti^UKfO&TI<S^pQvKHc?SVFZJJN|<Qq?Ty@FP|dkr7tx&5mF1=D
zzp5s%KCwx<DSi{|YqR#=jO*jisdV)9`-jsi?6v7iRJMOrL!E@<{-xC0{X_xW1|biS
zT3E>-J7~6g_U^vr*cc?q4(5)_*<@RG(N@93OjT?>fF^3nMAg(Khj}GX))pO*Hvmz_
z)}LQ+k($xB>iWK+zzFmTew*&@uC9Ru;vtGu64RVSG?l485#EzoWLs@Rf^4`(A$6=k
zy9aCdSp1MXHj`xR(!rpA(d0RIuINu=8Wc`-x0}U}0OaD}`i+Q)(L|8!^2H(S*l<?O
zJPrf)%ciJGV$z}UJ#}BS$DHC1dq?xsxYM+CSK-%n)?gdaP+8j48tcdU9E;(bAO(Ry
zkbzr6(*dc-#BoSPn8+L8cyfhG_oYrtiK`PH6kOcn77Xcr9H#pgGtfkmqb2!e6m48r
zfX=FEI(hFHqGQADzek_OY`UnQeKIs;!1$2h2=_m7-iUarq=$aCBU|Qfjpirey?zz6
z%TS>Jjfo@#pi1hM3Y3a0j%ot6Pa;<KLaLUV-g!9*E;~cKZ6URglBn>;2KKu)K;&I@
z^}i<+?bFQr@?pU#AWR8{VL~w7QtanobjLB?k^@9VhN&!E>iIrj^FhyJNJt&RP->y@
z8u|s9CtKJbo0{0zrwjA(^PVkycSCRg5-I`2$A|1g{rhN&7GAuYPft>6`1_o$6PkS!
z(Iav4B2*F;FdqLB6!$`;w9mF>OiaCnrQNoQNTTL4x8O&SqIc%92wj8IEwg4Kx}rH{
zQ|91Yw04h%1wr9$LA#IaI;f3Iux!mfP}mqr(pWnigdSZHw83~+dts$`4m<63+kDPP
z?d?kxL`M?Al3GrACR7&kC@=NF|Cf@l4`e!jc<CKA@!lY2@8j9$L6}D;X&8((?KeNa
z*zgJAnVr0<fXpd-SZGirWt#Ib+ekY=FU4CyDegoY1D6R3%elNlpz#ATM0kw|Y;ABM
z8M2sUzQ@k>bbRFklqvzD&36N}t1-Y^u|XfM2h`m#<c>~CM&(UGRB#d1uAg&hNr8u~
znwCd)#2CuBy-(;qphItqnlIy^$%Wz|e2AV8IB+FBQot7y$#HSh2V>e<kRu|qXF|ON
z@i)H7;LSsUpAnF!ki;qR?sWuQfgLb&8ABt=k#1fQurlk@6sA~z)DPyq>HV(o$Cxhp
zC&IGb_K&l*5ru^;&0!)7+yl9gFOCh6BeK@gaOi4cPT1;di}nFevzucGKkFGj4^MNn
z8x|dm+bkF%Li!uWuqdWeM9~Ve;uDBB4sz-pBWUQezc*908Y76p#V=v?3Buz739oGN
z7;$kWlKlOB;qUSN^Y`w4Z@%iA<tn?15lrdJzSSud#J8O~`wel`%X$XLN`56CgU;`1
z<y{jhTBP6Xj({A1(;z{$+oFGDY66~lQbj)Ggc@jI#{Ox&=y^?WA)d%y*qqp2+-<EL
zuBN)KvWM4}y)JIKSkTN&YCYJnIt&@t6bi9qk&j~8btBapdb@6(2w<@YNqt*n88zs8
zeOA2Bt4u0rH6!>!RxBWr1#dEv)?h12b5Y<ZUu2MI4#j-jM#}9r!T>PnGRmVn!yiZ<
zQfIA)8D;~_bw)1^)~p#2TDuE-G>C=4?V6V5igdg)%u?`6Ph(C7$Tru@_6yf?U9+_|
zFV|u6dJfVCq=+F62|?9%VA1spIrf{akhD?7BeJ%ur!u>GEK3Wqc9?ir(n>FF18|Ke
zpyi_W>l!gk{_)U9W0ek!jb^cUF<&w3GFH05LeqN?I}jkGu1-tqo>W%`MhFLG0JhU`
zDiSXAmrj6w_fpu3KrT@;cs91cFOM82uN)*AU>of3O!>+~!iqF7%LM|*?ezF4c^Lsd
z8>)G*`|eKDD%;j>(Xofsz&sod5x8@Grj%5Jb2@G!hknu|!bI=R+7(V{_bT&|X1{vH
zd!uIBFnfO_zT`mnkeE{zUovA#L#C19*Ya1=Sth?h&CRJ_Zzv_@Sm}KS-;K3wd{%aT
zt4ewUS{d|l&r{>R>RhoXKZzDD3iEBNxgE4@C~)bN1&c3JfEmdgyO-oYh_*mw98PMe
z>H*Vlz2naZc-5L_sk9H9H(s&(?X>sipd~Y<v1hXJ`bb&9$!k@I#bmAyOPXj^)sNlv
zpzy9DP%roo4(Xxlw)`=ti|PpNHS4aKEoM5hUpzhqLu~#)Aug)^d>!b((*29>YM+H;
zeV3Jg52>m;7bPy;-&)OQ7$3m@)oDV)_;&EW$WwrF-iLXVKOU${ah?!5V0rGaYud^9
zpa4Sr)$vwOr<*->KtWD?oJB-RNr~S0`?0i-7#tB;;leg-GgSEJvr-u0B?3m3{S6$%
zxeCIr*1|S-$tBJ|R0iEfayj$eym@^EVUPEtfgrMoE<Y%EO!q|8`g>Bv`Ww4?;}?KU
z@HFf{lmzF0t<gD|{?}JtH5eOChiq58|F-C!W;GAXjJ{H@XN<f}e&@}(aK=lF=A;X(
z$0kd4^b!e;M;x}D*U_Ls01yXdl5XjVS5sB=YU=qt$UN3jWhSPrn`Yh9B^1y^Cas$$
zeMd30e%ii|8{AS^kf^4vg49q<Kns_3wf;IiJ(@I`Bpozp(wa$MGtWgx?~}hehqD)_
z_`AL0cOReZ5Z#`iA6u8$B;Oz5x4WN9wGt|~GVEFiU>->e1!SPOBFQye^Lv+m4M#ag
z+|F&l5v=c{E}ib9FO?%T--d#he|K6to$CNdEpKGn?4QN26N;4MAwZ<Q#})_zB-Scg
zSHjydp77=zPv(a0i;#xoKCMkkEH2E77!pNccMIyz+ptPst*>PVoL6=>f-&84AFUCG
zJ-cCDu>G6$;n3Ouw~jiMCCCIwLES%jkgjD?8r2l1Ys5mdZex1avPB}yIK8?&L)r*%
zQlHYvzeR_jConGf|Ma_;gU^u(UTEEdqtX*U24gh~J6OX*a2(J0Mp&$XqfU0WeCNiI
zz+cSczCzqS53I+&rs~5MtR;+tK|6gJ+Nrn{nae_pSuG>OqpRAz2A@yoTLmQCWia^z
zXdmjCCl`ND_A>_3;Lu+Dsn|WguL@-T1EjP%c`mRP1GF!-$w{(}0siX1CAD4^e32%2
z?#U9}NOvJtu!u!ZMUrCyR=gRxL)7MI@9PqLomUL}7wtT_wt4U}@V6WqLfHE(;9QS*
z3C{+Pha~9`z@c?0qD<0IQzgVqC@nN5437OVstbr(Oqjev8fukg14$_jkPjvHHCXpH
z1$!EcdcfU^BjaqLX*mBw9gQ?J85105tk~DnkPnay$Z*>;5TAGYF#$F}`Ba<8AL7I^
zNYtQ$kRL&7>t+2J0$jR2!{r4K-CQVs515B`6%`+?x7uPQ`{zcfL`f^fl*VvK&eKgW
z87e4O1>mYq1ZjE&9DtJO_Ss8-egWquPN(NOy{8wu2g;<9H_QcuBX`Kj{fcOkC`jo1
zB;}+kS8RB-wZ>Osaw++w+Y*Y-6lN`+uwfxq!(D;_%7>#f*)dH~=?=>W9FBIt+cbG|
ztS+P;S`XZ7atx1H0Z+^1d^QLi7mGMv4VkhrFkvbUhA1I&eQ6#QAx?FHA@7KPrGSQN
z!-E%zTp*g1Gz5S|JR%MOpR_4xDKSzW0!bw<p?%sVX^yV`5n8<_s|YN@n~`zit+Z+q
zT0OKKw6JtM4aZ4E<mD1?hWoL8ySINxAy}GY519wY_zM$#F4$&WG<Q^>2L&hyHDvr3
z2e{i)6QMqn|6Q5PixA>$DmY*!^Jk18;9^+@i_fJ|NbT8j&?t#4c?ks>PW@taC!CKW
zvrZqHLcASAI%60qYzScHv~(KYc#Ljx5HkGa3DuALc_6*u{2pe=FgGq7$ee8f%isa4
zR^<V~DEZPsv7Tv_M&<HVIpG`c@KPwnDZ*5#jNy5yzckoKuo`J$fUjGIEj5%vd}HWP
z&-j<;5(VZgi|VCo4c&^Mt{g2e6)PKy=E`lJH%L536o3hFX(QH-7@cj1>wZLTxGEVN
zFdBtJ-n#sAKo`p$b6=Fw*jbswy9`rjJ8KmtJ|YAW5~YuKbb>VIkHqN#40Z=DTsDy!
z0sj`<=G-*bVu!K9n)9@og~i*L8$(0Th^jqzMh0DsQF9?Ex}#h^6?z^l1Gg<Io35M(
zi{AR5M?f<r@ZgNlyW7K`Gn76i#LrwVpZZsM5*1IRuGZ}xE@ru?WQMYY{g84-9zKV!
z`3}Y4K}u!7n+43%Elx^bZhx5}s~iK>OPZp=CV<{CM=@4MzOZZ@N4U7h2NJe`s|7s7
zUF#znc{RRlGz&=8lu;ojPZ!y)36079RD{9_<XLd&(1{~UyM7)BWKvXDLXD>ls9pWJ
zR3OR%+?<B5XeS2FoYh7G*6e7nT0RlHe6+^O6CQe6^rh$_8d^){bmR+gjQDw-*xsNK
zvAV5@b)EWxC-*(meO5T918Fd*am&3>dcF_<j4D7082CK|OCo>tN^XRVozSs@y1i-`
zNSO8Wp1g)QlKI}s4mJcVgbjad*pn?i$^0%*aaH3v$EK}i!#fOm&`b|Kq@uvR2}Rb1
zJXxo}c;v3|BvYQNa+F`yxBaQ5E&#F`M{dsYhgm?F3GGM0IgO>OL&I)Wgoym58X)mL
zK>?l#ArIRKuhrlBHJw00Y7aH@)7>H6m%pns>G|W&Zm!Q`D}PYK7r|~tfYMAiI=db$
z^gi^|g-ycayyiDvF3ROs7=MzFz-R6tLh&aOEyFq@{ljL4kg8YT%I$K#erly+!~wVI
z7b-z&gZWuUn9Vb?1*YdZ^uM8ROd*2{IMVO7i1vhce17zKzUF?1_MN+pLPL_W^bU3E
z(T}`f-2>kZCO>t`<mE~jqM-TAr6~s(N!MveJ)|ISkaOefF6N!4tnwy(WR3|+pi1ZH
zWs*Xr8K;gX?ek5wApzCYji;Cs4z7)_te31}<SX8bJwKKsMQM1N%WjK@7sL^pUL7Yd
zQNHh?J6_RgGD5%pYoPx}xPgh8neBgT4a{iR{v#|R{@myf#4pE;?4{DXGv`V=OD4-L
zb!}d><dP%@Qfe5*6S;zzj=ewY8USXMkaEj1<@zN+1oht-y>6h>M6w}9{<yr1t!y-@
zr~HE~qEJc~uec5`Nkp*_3W-&@KGCH`%AiSSCcUkt)2l&ejq&T#q{$?`z3clA4s4x9
z%M4dVv;Pji=HG4q@Q!7!^!a&ucD&VpJT|EZ#b-(;A>;+p#GxO$&0yXyCYFq6q%3@G
z*cK`0)AhQoO|5L)nxm{IMTq5)hb&4Y{d05v{LS64%#+f{6IUV8F;grrzHI9At%(_e
zqzv~|>(4)Gw`6igv;k@2{6l*@AVwgTW&WqFw&&a8qxCP<lKSyj?nT$dl1OU(8Lk`&
z2wbo~r)XSqCUjfIoO}K+dRICnVMB)UTfM{Mx7zE4xcs%Y3wtn<EClpPvLw_AGN`PG
znn)F{gTEb`v|!^~>qQ#}(~6z`rTk_NF+({OoI(MOJP5k7k*PMm+l*ssK3a4siVRtK
ziAs}S4DS?Rl+EZ*Ne8~F<fs6tX5zdGQ6~0^>P6*h0cKjXhU;oai|>8|r_Q%_I<Mck
zc22jVYm1ui3fn1zVuS|!$-L?e82_vGF}sDTog*A;Xwz*ZNu0Phv~W~1*jeUl2m{KH
z(!vXcng`dLF;*mttdpdolReiufs(;OMz6B!5;lYi4o%ml8d%@9bj)of%@#0Cv%RLz
zwi^fRmnZa8){N6G5Md4|*`K6Frl09S;BiO}BsnOD)xW{URa3h#qdh%00TXg{_s{iX
z<;oO1sC+W;-O{RS?mxcpG`VtLd<dWO6^s5iZ4_x6tu5g2)DMZ<v=&Mk_lm0cfngGj
zE)ZRj;(H|U0gM3e0jZ=jggA^kUK>?am9<{<lhE_NEvvc84W2+j1>~biKzVZV_SLcx
z2=}d<HtCmE;1)nsl`l(Tud77yUF5*8G!ks3U#>S3O*|7(nF;dk<-EwGl9TJ|-ejOx
zCPq?aeOub@36^NL%eZj;Cz{p%{3-*o*CrJXA0G;bBK{UR{F{#6*Qc$1t}8m$I=d-e
zafW(FkmQ6^E7`5U1-CiUiV}O#K_hrAvzNlv0&hfQei>mtlgT`D<!_#I89CKnFO>yj
zsL4p?nt8%*g_x7B9l-Q~%ru~~fYDy|;!M5OeAbUc%3fh6aEUHLb4K$?Op{<KvjiQT
zAF_vE)-6>Ry|z@x=)JbtdWV(hl~vj~Es599Tj*?+x$;j~vk7LBV(svH`!IjFuZ@Em
z%=-;3J2nQ?)_yK$N7>(ISb7UmMo)KW=xqLqC)TfbqQ`;ZGfyllNm~`$0?Ma30DPop
zMa|lhnhx@Y=mBfHOGRWVumOdsv@P(;@z6{6+@^*WdiN=OO9I2mWTeJTK&DH>AW4{V
z8*LbKVRKYsj?Fp^7plH87DGz(WX1y%qBpa1ff(}lpH~H+W7~EE)||?sarzYIuu<>j
zA+7RTLEM{$0XRk*k#UnF?D?|1OK;gI(29P+Yo}D%oiZVI{}y#s>f)<@XVA<Ld;;}A
zZnMiq8VX?A2My+9fVB~md5?jN1<bvyPNR9ha$A8$;0e7gV?CUpYVdj_{A-Wga^%2D
z=~d6&JU1-ds8S05UK?xs)5zZr^3-pp?xoDSJEBt`TGAw(gfi(D7_LKG5(7Bl8}Kmq
z*}JKlFK!C+p$tvc6u@`P0gNtYD&)jy`dImVw1xP7oXX@Sc*yGa07>i)Ikd*#gMJ-R
zqXCG>Us5l~wT!>IO_w68qW>)X%n<AAo?5#$=4QaS9KS7}ExhfYUH}F@Vct9JlqGS1
zJPLjH=w`0T-0#04-2Gh1y)OjJnkB>`d7kn&+nS^W9Q@n?f-5|V5kq@XL$@Uyug^25
z(c1#MZ-Rd%w$e&_$r!7Vjt6>yCKOv>Us;heCk$v83WlExL9SSXr8a!8wU%JP1<YEx
z&A!!=p=i5w^)M-Y)~2HFI-1sXN=kU!7e)n1NWT4D2H<>ZHS!yv$POJ17bW(`9Rcxu
zyFa(=dKyw$wk{s#qCdyWzQGq}-~d=DD6zEp?gDWK;7;XLf2dr$eC_(k8<=b~Q5yAS
z!$RF^VJ`y>a+H)tA7IFqGKeVglwj*9?5uHLWYaw!e_rjS8)1o;Byn$_g}7WSUTQ!z
zvBa5??~OJ`6f-3VwABX*D5xJaw9F!xpT`iko%J+Eo^o#adYY-q98=<)zSHRZ8n9z-
ztqBi_2%jKu!xEr;MvAr_BSq$C3y=_ATgMcX*;+q3-M77MUH65_yvqO4v^pq1%6ow_
z<$&cibxbS0MZZgo9KLXQBzW?jJboWc;qd5;EEJV!NLV_a<{dzjyjD%2`Gh84>adfY
zJSx-yNX-P&xXCyK;`xpP1Czr<1Apkwak>npKF^%bW8!$vl#PGaS6ceKs|CiAYJd7S
zpH}$J_>TA1RjmiV2dJ-9k>YpEmJgbRPbfrUn{<hgXxNkMfd!f(KpWB9rDGP6LjwOH
zooTF<<)^C4Z)+ex&bU$*nJv6}^~K{p!A2%L>}4QP4rkdga$1PImWs8YgBglPb(|%q
zB;hr)MB*0P?X-0+_0zlZ8|iH*tPuCags1V6ZxmHHScV@3*##Ogz0^g;D?DviKp><n
zAN!mer{X@SQqvkmG~@}gSc^T|@3S~oy5n|8w%41MEAbz5WiY(-B;Quq+st}A$M5xL
zoe;u6+~hTUZ{7W{sTVu{b^ER4cC@eLmQ6$s0?ac@-t&??YbZ!icT{dX+E&`fakCM4
zd!KhPXr|_Zsym`XwRQn`WSb^L$VRRszCe64uazjjl{fw$oK8#s`(hK}wVfOgRjoXH
zyZQ88)BQTR$4c8@@QOqS?il^=*R`JeL20p-d62h5*m}J)GJHMqiNV%_TZq%fNNvh!
zd|pb;@Kw_VEZqVI1<#6r?n>LmERTQ|pkhWt<owY-Q6|G6d32Hjz|r~HTY=^Tf!y2F
zvQ<@VG{5m~P`YC%P@O_yyBL9p!mTpHIpfwU`wty-l-5uK&b;`HP8Gj1L`l6Tikm$T
z;vuyUpy`*^S`)UqhmhVEd`&xH;{PqUSQ-AqL1AM2-y^M8G;QPmM{xD_3E{W&XN|q0
z<0$+)Lbr|<rrXA#*z6K6NbRsDY;9>dx_bBX4aZSRYo2ABAr$x+DH<>2V9v=rC6tZ+
z3_|1A`SSF<-?pZN63;p5)Yw54LYJ}<Dy2%RIyg<5n$l1A|GZ%>wj&v7C!J<^6GQR3
zPrq!NQfE(xuC=HbKNhmo>h11jY19eJx5m-@dAzQ2U+?yNM8c-m?|Hlb02BTFSju)c
zY=O>N&OZOQ;)DXT7=S8_DC$;#K}oFDuoH~+ug;DQpOeylxm@PBQuFKMWb698yP1sd
zy`k6KxVd-Lb){IlVH-24*gIrh^2&i4%*mm912cEX7@$B*Yv2e)qHD7DD@-X`CYRmU
zK`7mLTB4E3ih}e|W=>ZB@$;XsIrnWngv)oIw=M8qv8$^;b;l>w_PK=9wgCxm{-mQH
zYw{<A`$L45=U51p4w-#ou;Qh>4@7<CY4hP~Av(wIebvov!)7Ve^xRcz>Aq5S9-odN
zLn;!P(CM>Hwr{rfyi)t3Zi${jLDMcBJweBzun<{c)z(`#+vUL!M!Al}nY-$uY<y)P
zvwQH!mZ@FnESHxE=N9AqCHHa(q>II55GXg&rvImi3Do-&YX_tEFCOD+B(u%{*<9Y|
z=Yg^?JHO9<wfa=Fzdno`gb<-Y>He^j`tnxvJbg2er#6FR2-C<8y8FMJkP@jfmNHN_
zq~!jdIAjjXd52`GMd$gz!YBeUYZ=nptM6_vxc@)kc<|T=5Q(20Ua0}(k4D&(te0I;
zeb(I1vr{2u?Ecd?{zlg02sBAGPG~}sLB|Nf#~LMTsze23Y);m_X-YtmO)ES$Jmre<
z_@KqBNfb?$F4nf1xXs@WnjO~HE*{K7-89;x8^4d-a#cYoIgVxaYA8cc!NMq(mheBj
z?x;tsC1c|y3^=k;PLzyBU7PGH*Q4$V1t96iW$$~AB$P;5uF$TA`mBl@&P__VUPTo|
zVZtuX>wsaCejx`<m0n*7F>272QWP@bS6l0SwO%-Yv~la|xgaVV0Db_Kd+sW&wEhZ4
zf()jgyxmZmJB2+?Dgoq6Nmap|FjjM&nBB9><JZ#&$D^gtQRX1D78On_H#{%6cz2lW
zsA5z;p$D7nt}PSDo#6WH0CGHFfDkIKzU%7Gj5B)0Bio(Z)ku?^-f-x-qz1Dp{QkW-
z`B#Bl)yvZqDjZaccy7Fvp5X#D(tYBm#~C=DV0@#ra||mk5zh*Sw+@qRYlc)>4udy-
z_w#IyfM=mHLeiIHd&q(_geGE{+B5)8o)E<fVSbyy#r?~LhdiEOekcnHmY^zHwy0EI
zJ<xAsz=(k>>V5)C^PDD-&7X(CV`|A|-M~4+XjXYJX9YMe>-wSL7|ywOAMJKsPm~DT
z;S4XhaJdKAU|nvz=CTGe<)i`@1l(k%sw*pgZ9@6ljE!S=0r&MIkcWm2VfQ#t2N~?8
zs2PeP9PGMe1!tPAq>$|S1Fel8(^y6=nSMElfahb)6r79=q{7j>Rb=RK6cJq=Syv@w
zG^V+7T0p-D2;h*@KTq9R3c-;_%&f+n!PrAD^1iv57%ovVd@%yQ+7au5bV)h&#ei{U
zElLv5AWJ1ioPbCf=Lp^=9kyM3`y;xy?h)vC02RDOYU3edjn(kHaAP6Txw2V{v&~Vt
z*)P541qy`}aNCC$y5n=zgOm1Vp#V<PAWFck0m|VS=iTxg`2GBW?snot2#@W+&O@i>
zDFim;l0QK`nahEuJVqf{2?*ynj;dh}HBKI@=KuDq!$*!KB~Or~r4z*{ZFqU+Iqavo
zyAXXxa>&KKihh(7!S#}q8)bt(tP-Z*s2In}A^W39704CcVOKG(YcS?a7Dg7TgzG~c
z^Bc^CAV%a$NFoWDp+7^l=M5{Oa39%oOyi7(du)+$9A*lIyuc(n)`KbsNtsXQh|#E7
z<Z+Na`A8Ae;XCzCIgHgfUc|J}b1l0<g3D6r^h^~o0g%S=vK$p<_S0{o6eDcs8g;(k
zhT{`Q1{Oj11*@>b#dD?mZcD59NDC5~pY3)R&oVZ9&630U?LW;JK483_TW+$uK2Pc$
z3O<n9j{9msfM%pxEjsxsi)fc9JXNyj64`Evrb1&w7LLKBJOBR1{5XKZ*dwy$M(u(P
zJ`V_m{M7&s>;9i{?V%Vp$Qiy|hM5M+GDoQQ*=vxc+C@!@VCB#@6d=t+e)pZM93eUJ
zGM|E^;Qf8I;0m9r`;HiKhjC!erI%rklhg6!M55L92^PHfXj)Mtrq-1EbE(ltZ#`IO
zac=UwKLbD6SZGJ57wqy8T)-srH21qu#1DpeHRcAga4}+O5TnrZxt%DlU(x#>&tv;<
zk2P;(1DS!s{mi>0Y|Jx}NnmY6-f!P)Am=OUV3zfevoJy>U0e8ws<8aZ<gpm))Q4d8
zfaftM1kA$D2~%-gs0N>r@P}gvt+JISWC5o}90V~ar(iIW9p0VXKrqnY(<m?R9~8i_
zFtE+8tf`Xh_vayk$yiLg9gB?K3GZTQW-#b7a*RAhX8m`b6M+hBF~{CUC|~>l_t`J^
z{(?5T0>@z53vxL&%O1rm5OaH`Zh`Q;=RmuEOb}T3-VTkS3e0bh0IhOckWpW*56_}r
zGL36phu14U4E=L)8ihp(s<&ngyQI@k$(L%w6mWG@Eya<*_CfXd1~z?ya@U+WzkCTB
zilia#cg;2E2>B`nJnP7UVz?SGzIzzC+7IwfpnFesC{X@Pk7nV|3g3*&Q2bquHREXo
z@2<ce<4<w7d}jHYE81u$I3WiZ^#@=WI>PwB)d@2v!~X}ZpyKIZN<c4fWTot43q>zS
zz{tSxKZlY|&MpKTtQ=7E|EFJ_oss>2YZv`fFZRn~2><M&2Bp6K@ak7nGyO^kN&(<h
zEl5PFmRKSe9T*YW*{9ixv%XtqKCIbOlGC?1vQ!9eZ@k-Cp2tq>{6?mz*ayLP6qVtG
z+{R3XBqxunCe<Cr(Sq@SSxH2T)~Fuk82<&-E7f`eNzktY=BfbKag0<PLl>&8>I_vg
zO{yvwG$ut&Y^I3nYUmOUipyBaaR@;%M6{Nx`chKU;Ky}X;zL|(K@-CMY>251k|be#
z%tEO4oPi|+1}BPfCWwWhee^LQj$<AbNHpUpDg!yV;xP;cAds*OkaJvQh(Cuy!>x1_
z=)|ai(PYqy_gT_3B0z~Dp!B9+=>1uu^5gK&0sx%$_duQo@wkEjPWb_;89+d2M}TaR
zWqbpC2muAz@oFAK5F1fi!Geen0*LGLN&wRlCPeW8dIV<?<%FX_1@xlPnE_~la05Om
zJ>C6xV2K=tA!Z~mEn*Q6+U9qHL@$&B{S0jw=p0OJ6c7YXca7N@6T#@#&L|@dDCUWZ
z46L;9Y7B`YECuvZq7i3-;G!JJoQX355Ul|T{qvFQ9@K*+DEo$pgNjZERu9r@4Wa~?
zsJW@))kWl2a2mvb$kphY0WxSHr$Gh}7=-KtI*gIx5-241x^KQr5I_LMNvO_d_hyOE
zbSY_$2`WjaAog+r;061%2HID*=*P~kG<>GZ)~Y`rTep+IAO-AQJpBBWc|09?;y{en
z_?<qLK5F@EkHzxImZe`w-zz1SmG3eg$4}vNGttFniz<_?{3yqUyw=M<y;Sf`0&OPT
zA*feQpftL@S!+whm5O5;x9^2iyDItEuB6-9ef+A83*nUk08M3Ctk4`fpI(Z+wP?@B
zK3cS!Bnv-J&aPf;o?8BDV_^Id*+W0%>t;-2T>=CxKGc)u?$iRsDEi7)$y^O}aPey7
zHm&aA-~)$At#@m(J`mu!gD213@mYxY*>5@ST;{LW-_<<L1vWBU9=XeTd6^Fq%#wzi
zCS`PU&t@O+dxJER72RHGC2&hCG*UaeaO=%*)hWA#{Yk9628RF!UGKA{1)R4myRQ8y
zCxb`yO{(K-hCfipwgvH_8!wGz%?!kbt#-`QxOkT9-7jJ8kk06v*}QMmXT<~2zIPrh
z9)<uOLrzaz_+Bm6&bM6o;y2HXyZ%fYe7+o8?6P>)SnLG!+|1J~);{d1b9nhPT3T<i
zJ{DW7#h%~!De_~y%$)q2GA#WvEb}tDeE3Oh&6zXaxkj7gr$VcX8TJQFoQ0T0;9D-q
zx9RZyZ4Zig(`Kw`Tz2xgL{tCJ-;YH;T66MMo?U62l=^A>X>$w%mY*VS`9fC&l5|nh
z7u`Gs)@h8`xw-YYy1mx*ChtodSMd)9{(aj8JQ-}}&(%Cs|IW~gzj(9M(Vz8`A47j0
z51ttRAIjb#JkuuZ+Kp}7wrzHjj&0kvZQD-Aw%xI9JhAQF-~LDY*ze%`C--p=YEq+G
zb=6wu64PXkCsf$;Y3n~On6~?9%D&i~wgv07+UwGle(JEl;F?-FS}|TO-zKaNbJ1#B
z#)lM+xa^xf;vlny#D2mBJKz;=MImTTQNC8%NVKSOa13;TQml1F;&f`kNU%;A!a^sT
zzx8-qW6Q_bO+OxSv7i5K)|ux|C4X^N<m1n6?TP9AaPAA<Ql{z&VIiRMzz?=&>iYh?
zZQ{t`>pfB-$bosX8)$CNB4{n3r!u|nLJoFQQD|-(qho+gziPyA8u1IwSJ&tE+&59K
z<v_Qt#+Fk3jIqsfmVCvj2Q9djnmV2JK3?OY06vyfwYOq-2}bz!^!TIGu3oeB%6M7U
zg{=U)G;sQqh7cDDQ$?ER`kBE}LrF(AIOFdK#W8*>>+c>9b50KN#$B6Wgm|4ivRnt6
zjuTO9Ou({OXEq@3!yOb)Vu)|gE}+XDl!FJqs4HId!uADtlh7jg0vLBn!ywofioEx*
z)cV-(BabRXz<>>hY3DK$KkxXjUN6&jG^O5+_XTiYn)5NJ`1q>HUIRshZ+<MOFULYB
z^nvf2%?pUh*o#WTZkPlCyy9ybQtU&oo!S;b{gU7cg*Rot+WT|+l0+{MgXGS2Mzou*
zz$i}aUPH?Kus7D=+3+TNi*q5xLv*fdNnB0q{oPi=Ojw;9){0THkF9)Xg_ff0oL>v&
zwp;=uWZ;w*UMTAfXKU)ju?Hsg@YiAhZNSlrUFck@V&~6qC*_Q-m4;Y^LcjR5!L^*G
zeeWlJor>xfl-%UuX>@zKj1K+7`%|3sn?Ix52lft=zj^t>p9J%i33PAogQ+~WNB;yL
z$Ift~2ypPb8?#PruRTue-aEfKP5J{Jjk5elB9xu!Km3%;TwMR_qJe2mP5VuW{{!6v
z*U0Wk7ux`tQ7s9`T|th*kU~;GF-0x>m0nU^(%0LRb8M*@cSW`w{k%ue)Wo{$(^Rw*
zD%qnSpEn(s*KNf1D1u^D+p|7_x}p3JbDPSk?RDcLj+Xg5wd^~eF@KhV?shjvzD!5)
z=jak#O-`Od;=So%)!*;~y!C1p`g!X6`|Eh~Z}_^syq?tj?(csE=)juiJ?)LgaTUpN
zNplXR-Hw_%9&Nrn{-@RHO~w9kST-{J0BB(<<Gdnb=27kHc$>C4m@r6!WHG8k^oBN`
z_rpS#R!S3)cTXrm>Ae!+Sx;xlIoyDaW{Yd6;kiKJ9`v{8L)Y$#d!quu*LwtBs8G~B
z`F@A&e+&7s;Dlo^Uzswkq`#(>+vbercsqU6XxA-uyje`dvgWdD3^z|a)zrjOOjQ4s
zNyZYQuRI#r&>f1O^Q2uj;kJUH3z6T}@K8+6M&D!cW+6+3nYv<gDlf3@Qmlh>9L~<a
z)41;5r)m?xN|fhj^{BZyBnEgB@fvCser}d@iY@lVEy(H^`4Xsaa?R8;n`%g`qtM3i
zNsR4IotMJw5xg2^^(+NN5Oq&p`a5NHd*iN=D)jK9>5$~C0Ha_@@ag!_p{|la4Vit;
zE_%OhYFr*dM)RW?JG;6HHV~3OF}{Fh62da8JQGO-Q3xk(i0@X@1yoP8<Gd0-tY$Y$
z1M$`_kTKL(i4yWM(K}bC{^e;93Ng+}CIu4)M=rZQLxO;avh307we2q<SR$6#f#wLn
z2J$K9`wI`<Jiz*!oG|bgy$>z-(@0<d4Evt?y^wIIqloD9l@K_@NfiL(LuLz+aL<Vl
z4r!-r<Bu1B3FBBGDPrvH6VVL4`@sVlj0_6QYtpl~#gXsv&QHdsx|nessj+)7&ojnU
zCGD?x7#itg72c$Hp54m2L7}mY=`Vo|a`!mfwhCWdv{LT{9N82%w2D6OJvYvOv68r>
z{-pj{aQ5#D)ghP5Vb{xcWu_xJjUV%5uJ5B$<bYJ=VALHh_Ws8a68*XdQ<Xp_P%)6m
z7P4mntH>&u6gy`}ZV{mZCr!`Pbg9&%xbGvRB9aT!1ahR`@crOj-X%LJ;0E(E^!X=g
zslKBf;x#}rwK;P3W*DTbNHG@ZY+DQT<Wx70mbh2w&`}xo)hjl|hB;+sIjxX)8#9&i
zmbCc?jGNth4v{GEiDinooCe8-crYzOg#rls=Cg$c$f$90X2k)lRp0xXzd};Djz~IM
z!BiZqq-y9Vj`V2-&Dq&o`5!8NV`z*xOFnN@XHpA)sdEpHf|zqP!f}v^t<jCYx?~2t
zr7}@VH`vdg<KeUr2%nIm*m9vEQC2vw_Torff~zc;#;)?Rfio}EoAq<-mQ02P8gow&
z*?`d_k|^8@p%Npooq^ed*{4@5Z)1Gi9wqOhkk$5yCIC6?2RYzef}~>@dIJSZ99j@U
zL}fvjP4|nZ*I$dW#^L5ks3y`KXa`i?bmQTPN}udrvBAmvD)Gog4uvUfj4$`!u7aSU
z&S4_F=xz20?h-SFz#8>?L9!OJjUDk^CT@*~PEC6ebX90lZf=9Kl~37$;Ul<@g@87@
zft$=+v)HWF9FjBz91FBrxI)Fvh}`lJ#Ngwjuzvk~a2M1o%y4%```6n?Z+cv=PSpzU
zaX@(9KVD!rJPFHzC40}hdd1M@ci`vwBEAwE&SkQ)#(*iT0Ja_uo$}*x9DvQjoaNm-
z%<ZcJQ2q|?qsh=}M(E6@DPCFkNX3MSS##u&JCHJP`B0JoIbTPPW>74}aCjjSHR$t=
zqR$6zG%OmV?1{y#xN$~8jz5dwn*4-7lJ@X-i8757FRh-!bAZP_CNzvJ2K=H4r_H%u
zAr7`)hIc$m;n2!hfpIIId5%UTl7?gNQh@;aEBuN>bVYqGWJqfYcV`p%L;!9xPJRJS
zRVGrzb+a^B>h>r2xy>=l$Tk(zI^E{<`s#^?CdmT9hh^LTYN}Be8G(~`8$U*CA4}m+
zqK~@<WO)`7dybRa7z18~X6(tA)P<}jJUesT><0fYYlCf7A7o71Nq$k$lHvY{bDU5i
z3ivt9bZQYZj1C};$9sB}j$+8G1kWr+fmniAz#Yv^a%374Ti!Br%m%`E937|;HgqO7
zD!MSbSxS)fi<4OwfcZJ*fo2_`S}Fg>CTL(#%k)_$x1xXToW+s<=1&o4NKsBmL2;nT
zM<MJf$i4)8`4))oS1uR1fmE>OeHd4O{V2|agY5_od12*M2VH!FuV$XCy@?bWvKpr)
zO4;(N<IhdlkjEoz;DV52!XLQeex~+yNpOO1${?nM<lPr`0gpLrzOPY6=NB`xdknMZ
zTo2`ZE^Gx>XRKF@{DVbC)X@z8OTGIpA84&;sv#HwkXzh%Ah<)!lyu~G)T#$Gs;xe-
z?jNrB-KOK@?$a;FfcP2-4i+)H7o0V8$tj#dZV!m|kAx|>A5<*C3tE=X@~zap&>!Gi
zP5GPu|FxE#`9Fs4%-mf6d-~7+hmt$gE-be(!=)bIZ27P9;n0jH1#C0tSBa>S+?M_>
zIR+A<veVY%c(hbNzma>x$5k6kY%K|nulJ*ipUb;a6cza&08f^_goxz-%-f>?Ri`8X
zuV=UKp)(nY#T}Ya)Im5<Y`giUn;E~%KCb*+2XSL@GfL^j*SGblagUvE?jgp9T|dt|
z2>n?-!Ljrc#=I{Nw>X);pSP#%=VsFL<zg~+oj(?z)wM;w`7=M)wU`gZFOoIlY(|Ml
zAuJfwSt!Ou$AF4l46T8CGNU9Y)1nE9bjiMNU3_;87T^cI(0|8v*}zms9~ndw^s9cf
zgG@n~9GqT1g>~iUVZS@m2fCckAaYH_Rp(#(m5!QGquysDDlXt3GPn^qGldwOn{U7Y
z^X7=h#~rC#zJ;_gciOER0FW<mmn3k~IqE4?jLHC24y`w&se7U>uBT&Y!$~esg)MV%
zFB)5XkCKlq&{>6RN&<T55+R8%1xj7MXwe%OJfoQPTOsRrE*qAYrJYqKI#)-%U##+M
z?dK)_3}9r!Pb}B4wc|N@^|_I|V+N+Lb4L9eG%sVAivQR4T9IXJgoeE7uM9A}Q<d86
z>V_+t0LTH0&&KLG;Ai934$V#Do2HAc4{APpX)c3?)+6WZ8^T~PDbzu3T;m|41i8OI
zvCvAw5*d$36gr<SF>kjUC*sdXi;5tFxsD_m!>E<_LxgQ^>nSfpha|uB!Ombrunn4I
zHS_y|Go915fy{IHi-+Xo6-kIq95&YAi*477_rkA$<Le|>d<eqe7Vy-_h=<YUsrgV|
zX_%}583;jRllK&D*ST+Gp?^1Rdf=1Op;&S8Jf{YP-KC7}Lr(!JNX658mm^?QPtC_M
z$Wr0jKz9iEgWQJDgqPzCYnzZWi3K?Lou%Ngg@=f%Z}&L-iSfD*yrtch+1@*QPzGP6
zzEF89cuq1{qHK$L7nZ5Y$Ez|DVk}z+2J*wZypJc2ti}zzggen!W`?vrj9)M0K8&lI
zvWkaXj0k6tIx$!S1m$F`)blPTn%A?m6DzJX9?)jzHix^7UZ2!Ii`PqnM-njL#8Tmu
z3<L_wLOK7iHZGU!!s3;hDE86G6o@z*F!o!#=rK5{Wf*PX$5ho5_Zzi&As{AqrCq7D
zW{h35g&#Jl&<6ETOCIMaa8f+iF`coc{nPciF{`L7D!2yUU5vTJxLz%++N;NcCTbuN
z^zK@W1PYM_L%@2fh+74IFqS&_u5p4F6`)!QYByFU=yADR%XZg%+)}*l*0PU}hqB4&
z3~g8erT}nT8#NDM&3*FaO>x^@`@4F4d+rWQY!DAuD}@2M7E2S<tlN^V3ro7g#u^8|
z=}I_=SGZuAP8;JV6eh28sG7*R{L7a^qRm|g_nczuE9avJ{sef1+>E2S23xcnIzbB%
zNYu`NSpjkvIeii0%ijE36%c4-h0Q>}(yg{!QyDzLXOG8)74MAZ%WbCPkg_}}`AH#p
zZn!(krpC#Ze2q7Q>Ff(dyPl?i4R@pwR<ncU0i_DIM0%8GSIr^@e>@v#w=kg`7@kIm
zGp|*hR};ceu5f89l-Ws82)OK&Tp~nxZ>f*;{=bDh0kE&>>*da%XY%}%ef~qBE1$b=
z6Gn*ZzR&jcTZY9xL>FquSOlxqV_P~*7_OkpofhZRDFp-s8eB&lRrl2z8<5CF@`L*m
zTGUI*)A=HjH71uw!n9h44#^!0MfC$=k-2^m$O}IpM9dwRv5;sOs|0@NHfOZd?$Vth
zi?5TJmm6LlV;i+dlDPm&dDq!`zPv}%raUx^lRR@k9^g1{aoIgP)>gH@a140GZQnC_
z*Gr0abK!!-zPh|c5ODKZZW_5YS=&Ux7`^k?VEInYeEDojNjPO*nP*biA)=jb#hpuE
z<FP@gBDr5Ux~gU54?5C7(Q~;>3CxwP&Sv*LEnW^Ud@eYRP<Ftw>tf!IsR1r)4wi-Y
zwQ+XJ`_Cu`M9NkPuT9r>xOE+0WB1=Wj`h9(1_ew{#ym8K(EclL*De8-1X&qj3Gdv_
zMN=-kF{0fHg&|*t3wTitCN?Gl3J?8X3I815QmiSrFz!OmFq8pC^oDlXAX%qE$s1rf
zDX<sph*f(01p@VxOXK_d!Dws@p1_TJdfHBwaF5t4$avo4bOj^BLO`;aE%LYsLPUjP
zaL-%@EQRa;c+nI<CwT76fdK=<G}ev8n~DYCYt)<i1Slz33O-45q7P|VqC3G{%xA*2
zG~k9r#wyX3mEjTbR1z^|Vx@;#!(lBE;abDK^hW>w|9yM3QbgolMTYt9%6+#0LsljA
zICqBle?%#-aRpXHpO=t!fIr$9er<nrXc>!J<<|--Q(=#OMS+R_L^ofnlVR0tG~z@p
zFwf|k>w#fu?#G)mX{-T)TSiR6t_C5B(kD#}fS}43?=~}YSOQE1)<)gp&UCFv4Es*G
zK0G3W%rp_wx6_NMnWx||xQ9${6JZDvUT=W0?O`8lLr^kxQ^i~D?9(P*|LB@+ZX9rS
z)BL(+e|2%!uGEj#PV9Pyf_VC{4fF1X=~e3D<h2tjD6>2>_EM<&Q+4rlyGt1ziE-&5
zhys)CFQWSei@vay`Co$IzYxN#|2Kj_+kSKW{}fAuYcx6c6>Ni8a%h)y*j%vx&lP?S
z#kMJtO0ww6!{2vys{%W!$c+=Ylaa)nE8SV}gm-X6HZZy`_v`8YvWpO!BUD&INmEA;
z^EXD2Oh^Y6L;6UU5>cR;GqK}a9p)^fxmdcO?i4^}?|yxg@N+WToOR?M3n%^O`H=8)
z?VIm`HT~ypEIQwukm2WhJ3gm09Xf_dStz<Inx+c?2YJrK5(T3Om%o{PuvQ+u+jTK=
zynez8vU(7N_CMHs;(B4FWTOUwrib4!8g;OP$R~|!D<e<qg8qz&{Lvu;eC)TR8M`Ud
zI%4Dl)es$07^z?;ziPVoFtV?n4?E9Wr_+B2-Ja#<3t?&83ic;<QD%2IM6Wx%gs<xu
z>3lcW+E)sOCijNz27Ws-m&181=f#@#XsBvi8$ZoxPX^e=a;42{;VF_zh{GFg46AOB
zX78z9<)`{H^(C@sA;Db;Z|{RpAFRA!Ehauhsv<{5X9&((^&|IIcA71|OVuHp%u+#R
zpv6Cv!`AkX((PJv9li~iF?<5Rp-5u#KBEtVCmT1eva(zC-ZjU_`8c<2d!}FVKMWTc
z)LtgbdhUik+!`$1w_B|)vE%wfxFQo5woudaK>65Lzo9YNwSPKoR$RM42i^HMqc`ij
zt<t}bfoC!Vmo{}nyMM#~g%7k~0y_wO|8#4rbkEQGT)D8Cw{6|Oopwcg6>YJtbW@~k
z)D!Faiw>byi*j8dAtp1sjeEmJgsR&KTBoLW#1pjzmUzLEnvs30v|6gZqg%h#dG*ie
znD~=d4_eI6%j9s`zd@sC7?~j%T?oA_MpTngzR@(TF?S!6qHmL+R^8og>-G%G_=WoW
zptDL_uw@4*va<RMYR$7^b2RY+!!B916Ap7})y+WH5INmxuiEPCg#Tiq$&TK}LN0@t
zIEM18Vke0D4HR?x+)@IHy5~WNZ2pcfkc^9V5k&FNXC}v_LSj_rJrMv&qJJ9I8%h!x
ziJFr{@Zu=~n0ysALjCugPds`HIZz0TD?TKA)PSqK#32%KMC?f6z(9TAdHeYl^kAw;
z7=oR?ErB)Vt*5!Gle7I$UGfgkJ9zyQ0W6%I-~bLg+~0xqi~dYzY;z<{6ZwaHNks6W
zRo@i*@^vFMLY6-4Pdg=KI5c6&7CGWm@?n^WApV-0(KwTcIf3PB`2-qh7Bh#tH?l${
z5^0y-Vt!c~6I&~Xw=Za~W(gb=Yd0Q&TTX?V1JLat@dSaY#2#}P_)F@j7^uI84QF<u
zy^V;f*VS#>CvM5^*}9ADh{UDz+2>}9YP5X|TH7VOmQf+(f;~6Zk85?WY<Z$y5|M%L
z8h<_kQsPQBk$*XTaL7|h#n1HOjTyichaa!+(ovh0U?eCXvhM=VsZtHHf=%WA>bYMz
zCGi4xg<ax9+n_m$KL^G}ks=4MGNZ4t91O59<Le-vB^YY<=D=ygErE1N%ATIO!igqw
zG4ZAYJxk;Pys=S{$xa>;yJrekl0JzX99cx_n8xojFJC>EBKB0P2vmIqX+Wa9aJ_7Y
z_^x~*SEx&BV%J?0It!1&Zsm?jM#R&g(i8|zd^(gu+BdQdlT(4*cmOyb*G}-2$SLt~
zO7e?uXC8wB25Z4-EwZt;i_0r&!pB<{6#?(YH`ni0Wa-?i3EZOK9;)u~Ej0-hCYC}$
z^WQ##a#rYAZwnZDM$~Y1kpRQyx<`Y#kQ3DheJcCCxR;bl9GZi2!=7QRVN4Oq^HRp|
zuhtFI0zj{=z;3v!F}ce_^bjULk)s9;hN^-$YZ1{WlK-7)wO&_wm)U+X%W3V#a&hX#
z^@A_^g#<6Z+QU0zVeLv`EeG`_Y^dvR+>b@!#43k|VB81LLPIKm!?-*~OpZUi&X*ZK
z*zNBUpjZ57qzr|h8v>AJ2&sR}D-x>9;?-qIrtZ3P@>|lLkk#F?_yqSF5mMj_{a8X&
z`SAH#J<f-zZF!LivOERgCs63MOY#tualK>N?UhX{@W2>rEw^}F!Q3Fz8Ibxm#uePH
zpkdrCGmNbPKbM4sUKRE1a2`9Cik;q-f05>TC)O2VDZDTzsu5W+)9?xZvDu!%*Kky1
z{eOh*Mo?lbnE?@|v?Yi{uP7pzb5<ZYIi)8Gg{S8s>K29c@J7(w`a*=Z!8q4iDKQb^
z7E%7am7N<MHk9UWw?tSyz5-0LbGJsm(!Db>!F#H;<d|=mdde6s@MjWMF(F&T#_?Lj
zhE0~nyV)-6J1D=VZv7DZKgM&+V?@|y<paeQ9W@*Bixv>jBQVkhLJI$252N9Mru=0%
z))>>akO)k#_V(DeZQ)-Xj*yUs6#el15KF>Qh+YyzQO)K@WDiIBwLum4dC0##=9B01
z62E+K!Se4#mHd9~g}+wIFRUC2tAPm@Yn;ZskRxi-4yQdR-DuNF@@VYy_l9h|)FYp3
zSvIiyqKRtgicIJ0`W0asE5ede>@HrCa4WYtLainEH|5qz^eA`L{<S<QX%A<wlS`=S
z9x_iSJR1%WT9pc4AP+L+hdssEyhd5tB=PV%7}_Xl*9tb$;+kEj^|@!ScBB8TYDeJi
zx3sOMt1-9g!sHMJIjg$8(+&ckh9cNfw@G;$T{F-2D;K*Tvmv$LBM2DX1KZIXC@8>&
zYYCMi(TcgvG>icdpl&;;3xH4Lx7&|EyhW@q_U{x^@EHR)`CV$D#eolu&Q?@1WZZ8E
zbV#-r5!V1hhHEF)DjO_ZPQ^cY*`;@4d;*lTYBh(6OC%TzhL#P~NRG|oPbMScNdvJW
zia?A?HKk?KNh88ADqm;=`nax<axD<qXxL*b?ZD!45$s_?agc~`_bYa!_#wbvt{L`E
zJ#_TRqYvs-x4j`16st#+X>^utqrhmi|7<tYB<S<52+sWcqL=RG6sH<AV)P!+2us5H
zh*?iO?G8$ufi!9EJ%33K_P8GK$kpW5Kp#1Csa$Gw2=(TQG@_u9B0qoammPY}<(>cD
z<<I^fiW?OeB(s_MD&_Qz+CLMFJziPeBmW!@PqX$Np+?YpIkw!_M36sRvLE5<Q~fzW
zOGY?2Xsg-iWQ;#!MYN4+{61b2KVB{9?C+ulPJfX=T|UFXj!|HU8R8A^ubzQY`rgy}
zy<Wl2z7R_R<P-lP&{+OsV#C75_P;h7O#d?)Z2iBvltTMdSjHhtMaFhLU`|GNdM`^W
zT>1?igK@15S~1#xeS&{?>TVKPmy|YQgLM>}_0`bC5zNOk&GYu}Pl*ra5zYPExxMcw
z6cq=eRLbV3eM=U8QBog9#@VgRe`o+s2<^Tf4`TPknWV>aB7X<Kah@K>r{CD=>F3s!
z02<6rXmjshmybQuhB=>4Q2Mq$54SI@ZS;D5Lu1wJ^uOQkVZy)f813t>tju$kyzk18
z1}bU$E~U-r7?pHB%bhMSCg7;-w^B%3!?x2pd3?T&-VtImHR-TUVA2i~BT~%%ASD8#
zj+su)VDD@0r3^=E{&ks%%!m`XuXVIo`?p^4)k0oYb-H1FG}=vX%vN)(ksRwx?$Irw
zcy)UF6urumRokrUXB+eds3Mz-=XvrjP*H;|HVWZZ)~?%Xw~51`hnlbZ(7X)I>ku0x
zn+jm|$A?QDcbjm)#pywF{f)(jqd&MTOfTHbYS_V1GT_Yk&<(!&s2zidBkTnHu_OGo
z-(orTZa!n`sm}S-{^AD58~i?mJ<<PIZJ=bK^2e@5F6&63pA_^?bX)PKRXf|PUVv8R
z^F?+W>(U+94!6tA_TB8Jx#32>45y~CwTBwg#>KWtayQ8n1$ssW!Ox^;vWgsx7LfHd
z#a_Or#Nsd>QmXVm=Ov?LC`Oga3J)x0Y+(LsIhMyEZc3LJor^q6oAWD;#<8T4>yN@j
z5*_qoV!3JX28qIl%(J=Do3LPRD<MZR<*tE$UgHwa1?L)~U+x9thCdjw<(|JK%(+xf
zqniL^t$7%e-NS2M=j!$|h&NuVNd&!e!3h*gYQ7&7)oU|{swjR4YW}%n@NR$L1pzPd
z1Wp#v{HoRoqb?rw0I`|8Sb|)}jC1o|?1LN7DNaC*4Xv|qIfY8f5J3seRvU>qPbO?T
ztMS|Cn+DV0CnMcxWw-T?#MUvA&IvTA(H9g#nFzf#a@K(o8bI{v;y1hOU(Gy*N1pbV
z^r=M8fh3)&a&pOcv|IMn2SwU^rD|ag5i&{bRf4Xm0Q@~kIRghPq(+p%IcS)zNzWRY
zKzHw{y6W+gmHGR1K{5Szm#3k+oTqodrM`J`QUa?%I4U$Z=yUl--nw6^6;{3^iJ`Q)
zqJ7avQ{T51c_fbm#Li!AGKzk@&}1`y^<h7No{hUANCrB$2yFY()!Me!xq}QPDhe9b
zU7lM?F!?KJSjX|dg?9CNGrg~2>s2Yvk3|ZD;m)h}#+O|`M-G$|EiP8bNiakb0;~Fj
zs6{SesyxG$Y_)CivA~bm$d!Avu`nkkd3S5q{Lfq1f_@c*T7ups=)C<Menym#s;Y)H
zGK#AP(3x{6a~?sx%gJZq3n-X*qd$CDq^KeW7QfkMXQ)3qyRD}Tgta5v{|MrI24%Ga
zazJOya~04PC}gSir=S*5`yV@p@fV?)*U2EDc0-5w&SdFMZyMN>l|jSrL?SQ=OFB?P
zx*v0Wmmw)Ia%{M!Z;nIt8w4E2iJ)nI1<^nkY(=@S{gu^rr$1};CL^^dY<%L0FAL{e
zM%u<DkUjoYz*$mVsjIK%-3K|0$En*yp%5;MptF84%H1?CpgflY^q48kXB;|ylB{5C
zihkxspWEN;TQ+ue5l${g3*=MVEO_HZR5l6S%uA!wud$p;>Z8l=<|^X~Ka8xUwak&o
zy5yrq<w%w41f>#5<aP#Pu4QtaGsnz*a#l?-s0@B2rxI4#m!r<ZgLI^qcqa6V^dlf+
zTQJKIV>_#emmhoEE9y__ZAyr>KGvBBRoiH}#-24~sweOh9J+Itn_`}+9JLMdV~LE;
zxJ|64tplhbo|+f77Zv{K=?FXrkHkr{8kuDmb$;1duYq>Ggk}ld30Un6*4Xy-L}6{k
z0;z0pWBs=6&2K?QsL9^3Y5IWc+~^Ff-MuJQKdzR$48@>+JHC_#1rOw#PA*}TKrYcD
zGS1%3ShuS7MNEvETloH};WWVyBQ2yysRg{V(kp>EnK|VVwpl{nxH}yt(Lq@LIM`0G
zS3FwPge7j5X_sr+_0QTffy{63IEWyz>(QDJlLD~O+3P#xu!voMU{Z@nytUE<A#8Hg
zjz`y*T1RZ4j>;64M`x)5%G`!$V}ojy<OcD}@bn>P3F$pW5eAid34yxyslX$V;i)pk
zIF(6_xjWaM8sVDJe`s8RGptUamES~~O3QqjX1jnf3-7$P-xWELCJb1~_uj67V847{
zVJJ_8SUNN&)(-D$laPpEI(P%(Qg~a5T{!}Z{iCtn?2}?Fb3I->I-ksPln4>X5<3{3
zyS__osFanj##9o%rfwbmz4bS=qqH^ln=QO2i(Sr7bPXu_mD{L^xt>rV=?W?2pq2FP
zem~8Nx55>G>Cpfc;S8w^=Z-*fhU7(T-j1}h7mn>?^2wfEPZoqO;5?|bvR%}kRmPJg
zhJjkd#HlACFqGuzNG3I|<unQjMdtl(GbSdPnXY%Tk%r_>)zluBR`F^2V`;LjHf<N%
zPa(7m$1*jc=V&;l@N>X9bhW^i?^P%HFVSSOY)g0%mT@Vfo>#xL80ux=&x&#-RFF}T
zynQ!sn_DdUIL$MdLrznq@l0oMLhuKbwUMI43I8J0Ck7#D9}<~;1x9|jfv>5AuLy`Q
zHK<k$kdD+Ovw=bg*^o0VqR(%G=!wM>$G|G~iHD1FdVce6w;7v^6duoteyPX?aIP4a
zvI0ZW0>c2QHOVv637kVuc)?eq6HgBt&rIz*9FZ-R|2A0}=vDhJdpxxXqANNU6O(f8
zV`;c2;oh<{cYpvPqADzgdO|7GBUzI1=o!xWg7w#iUGk~UkC}u<u@59Ccf&FfNLORb
z63L?upyo{QIU@pFhEh<-3~x02!F)jz0n*;H@{|YGA^FPJSB4@7LXRk%%YD4JI-u7@
z;M+g+_F)?RlpR>bbe~6cAa-iltgCK+p!GX^3e)-JlG{g4`wjboYmlq*+o|)qL*(S7
z@BvcOk@utOOnKG+(VqwVeHPlX>r2BAKcRxp`|GV?%%AMj>#^k0FmdLNoC^FZ<HWv`
zU52<_d8C;r?XK-1XiWIdr5}}h|L*|_w$~}iI>t`HXuuF5(Kz(a4OHTkd1XM@pYa4u
zpPq+Ei#y-%y17UvS<qO$lH=kcBzxonaM96)?;&IT`jrGJ6n7Pt!ArM*%bq8c&xC?3
zGBp`${9Zf;bqIY!ksY}01wmkzZ+JI1b^ZU+cmFYfW9DZ2zj=}Sha^M#KS(l)7>~1l
zH`tP^vNe$PR?T9_1%twI7ZR0l^`zXAuLpd+#)@&d12kKGx~Ky@JcN^vUywVom>#?M
zKK#5~+(W_#<W$g7r_&e&@cc6(XiV4BfSuA=WXbR`KDf{PU0+t-;f9}~k$vX<K>zoB
z?QJeW{+JBGE{~rozaO=}$$aH)){m<(cS6KEv)%k0W*)XtP_#KHAH~*RBx87Bqr8a4
z0YbF5^-i#3P))<xsZ4D2Fbeq2ebYZ^HFjmOF7l`89j9W-fYgjhcfhKuS+7f*+pmq%
z&eS$9B&rCX#2H!pQt!nH7Z+Krmgr9E5ua+qTn<k!-2hr9&|CLI*M3Df`Qs;@PXaKp
z4=}?Hzd89^Dtnivs|C4~qI0bXo{>O*wQR-%0s8b%u5)o0V?|7H-*k1(chzaIywG+j
z*pi{hqzH%W*!=0(wyVVvcW>#QY<wSqQ*(L6fQB)sU>E7`YQBUowj(BiXOcLgF`RvR
z1iy_{$z&?vL9}-%5WyTP;_{QhGOwu~tVYHXie*g@=yB9sKP&cdXef{ZGg2WDG|I^7
zmA>CD`K?#@he$lA9w=h_RY|i|GqNu-Ka=KFc(b7f!;UvNCLtDjOsQ?7c0jNt{zlbT
zJp7mwe+@35)WgRNkzNS}mRD;mR*d{icqh7TiP?9tp7_A{RTvBnmU#KiwC4M#jWI~W
zj8VDDdic5L-;qV)Xg`Lpupp^mbiZae(!<g$lhYdWnt6?*pv4n&rU>4EpOYiH{Hn!Q
zmZ!Nk(zCg&Z+fxBKcxEvFA1OaI_~_Y%lNrvZh>GfMUX(4$u@2+)1cYObr*m0sc5g7
zSfp649=()_VRX945y5jMgk)F#Mul63yr}&2D+HIQvXo#&xERrmd_u+GQV|GXkd#n2
zyDZ;Wt!7X&E$!k!9qP?37;Jo>!A5fY`Q<M3R_Deo3)K#)uT#hPcNsMfbySEU^SOI0
zN6n-qPOz1k^HiCeG(4`DnQ4J4uwSAp4G1Ena9i#;<N`U$``#dt5g#xDK~Fh^t{776
zH<Xq>tir&9PKiPF&djn3X_|{c%M2&VgwAAc2x+;s4LBe2$-t*DOOWwFYD2POW(_%z
ztg+3VePS?d$P$d^_yFm=zQ101qk2&xc2U=x9_hJW`TV2Ju3FW11}#*q8%3ACC9a6c
zi)NN+ppTMD#33e54Tb`0bMY&F%Ke%F1_%$uT}heN_cwPkXzt+SztH5r{P1W{TkYeL
z6~!=Th^^aZ7kOr;rGH2Q)`I%dI$>Fxu4$W@%}3v!6&W0IBt?Z5NZ?y~M^=*!%iTYB
zxKO$tSMH9n4PV>RP=LW3;Ejk+q^GR+qZvLh_W6mnIl39^OqJ$FKYqi7(<+{n*U6@)
zbSLXow>Jbk$+4t%$!3(|8XxrnGqYfk;u%I`9O;DLZ_PvV<{c`bO2+G{Vo^BCKL#Z>
zkh^s=xtJ;O7B4V`8)R06zsP~kS0hWUr)g!B7^X0UM|z?rr$1{{XbKArSdz}(9ea>V
zk#`x{@nsRgTfsJ<h_`W-L`J<P(_|;JoUVwC%@v#Hy%X*VenJ2n1Bg-VjH><Ej?yr_
z5Gvlp3rM3yInT*o6EwRAVy~rMpg>fN!EwB$tTt^e^VMB;CsMap63mDlr4{?b14J~r
zK;BVBQbnY#_Z#T2t2oRvrYJ*zr6*dL`x$i(8r;cg=MTxZ=vwFuvbfUl1JVb=i^|X_
z6MQbhZ(gw!Qift6NlvS`YsMrVDs;}b4GwzSzM2P{1C%mq%3yKPVU0(cppK<#8)BU8
zj0<SILn<+0lK4?(AovxL-KG1u)1AEZERH(b;*(mtZ!A=7uH~%wvmxS9tT@acNQ@xv
zB}5K0VzdLCrk||PJuYD~j$dhK4*p&YxQ+pc0iOQf2K0i#rPX?Yr<?n3{^@P)-gZ*U
zYr0Q-UV*F90X9#cs-CP~arU#!Ii>pvAj=_O4IRT4S&sv06&>5zwl4}1^jN4J1@TgF
z4YA<`v+K`E#)-sg;M2%{Mc;6?#uYSt2DRM`Qd5kEZw?2z^1~u4<epqpPR^+P5E1Z<
zFp%6&CoN1kl^ZdOQ37@M-`4~@vVlc@W!{Sj6B*wo2vbw(6ZA2oxL6}tYoLZ3`a(?b
zln^-g(%+Os0M8BV--$p@eM_s~wpBIb&UFDQSCI5{rt5C2C?qo()?To6I9{+}3P6>3
z+(X1WxbdZ^v346)ekOc<`WT*NB(V0_IqX-NJv&lvL}I|=80W3d;FvMe*@LSPOy{QQ
z>2LYZ`EcM{!LxsWb0fFv&9@~9!jSi{*I{WP{YzrvHyh87z`>Y}aNfsW;*tjcRJ{Sm
z_=RLz-gS&eHJ_m(A$%5vYUR7aXq01Sz&$e-XqMC+sC3KIYD4ollD_-9!HsUh;eEz~
zS_;72vCP87tybB=odRZ=-W-{`*=*w>)XDm5rr3@AzfvYhmIXSCLF2|yhwMHCi^5)b
zlX$Q;Gh^EkH2ZSB&)K?rTXz1v=O+^zy4lWdYhVA_%o1ZL!0uIg8JWqP;tRC=gqL0$
zT=<WJp7p;5IZW&<|J%gKKd-FwrUZ)LYF$662JN8pex+<{RAi}18;ZS}*+0y|DKJ<#
z2?|Pqqe7CuuUDD35K!4Mm)`2RIM%a{udi=RhWiQgO;@yi8YW|H>f-`Ytk@AfXI@OV
zf;g~pVVjP(z0Y5>wn_h<b^slI^#KXA$J9Pt*XiaBrXYQtotZN-iy)lY1C!DB<OrSG
z_rj;<-qd}1ey!f2!`(SKf}Wo5$It;p`uD?bonH1P16IRW5F3W!schd<?n~jH?7q&6
zfAyTtgTIFPuBEV<W_l7zh4H-Kz6TRb(FX5_4`~t%FPAqqbv{q(?$>0-$>+pj{6!+O
zB~ox%P#~>9-s4M}Q0kG%nQ`_i9$G6UJ3-I03C$lOn1AeFJzb-B$NCNjXN`OgIPf0P
zd5=Sd!6uZc5mj=$Ci&|dsR##lXzrsMm)$$_?uPU0Ts2$7J;Pbw_YMAbT1HihZ3q<G
z(#G!A&RgTz3hD7$w6U$C-B>iQyfIWYVa#-hMRD^yR~a)OesAbN<1qbh8fV7$?6>JJ
zM#7$P<85NSi`TT}Fs;h9OHfJ~vvL;`FcNH{wEirF-76b4y{MjMNZeY{Tn8HR-eSv;
z@j)Yfc2~`VuWfz-P9lr`I_9uGI|nQ{AwWna1D`0Okr7MIjCt80nU!#2748@aod?Ye
z={Dm%A%HR=Aw;q4x>|2I1T1@RpJ%0Yk%l{If#>s)nC&4Kf~3nOOe5wxmdn6*bGhxh
zX|7|~8m2J2++eO!zi=TgFrkeC!y75tGnf#SiYWx(LZJSGt%3s4op1a#hZF7=B<Szw
zSc&DsRpDP^D_{UGZy@QmV?8Wqxy!zGs5oyO(9h(~A6*`uR|Bso=XjaG@^Pq>F(x!N
zni&|zS;cSCEw{86F7MK2Ilz5BiECZ3n<hnw!Rtwe<2EOOQVz`FOueH7UlX2%k>>+V
zRVe@LrrDfw0;ctvf%r&q{Q=w750tq=<c*yS%zNkAWPe57*5Ur-41t5wnJx|X*2e6h
zpyXs0&jm}MURxVMh5Vk_rze4?Ga+GfJf!dnK`0Vey{2m2DsBf_i~@;_ER~~q{j=-V
zImM1|P7hNhnbc2D4qjHaanbMsdgBO(AYN#?nED1}t=4zTzT+Y_cX>}A$D-BDkcH=I
z<th(sTEpXway5i*-&JJ3n-`CaIorhQj_;u#9nsieg3@W40ij4o?UJ3lWq(uUdDFt7
z6k7E1D}xb->}!RcUa2}dt4^Ct!96ZS2i70_pOxrUyB5(Fa^Ph?w_#Kq93KMMc$e!1
z2^7BSpH7nTmH{I=vr-XTCmHB0N8TRgIiz8J@ZUK{DQA7hIMx!Js~@NXeyyz+rY4cZ
zwh@BnmhmA~g4<dMJ@vFzPj`uP6bzSTkBA#w!Xr-6B?|)2PQ*eoTc*Ybh=hKw_ZQ;5
z2yN%*V5p<^&_o^$vrw|Vf;SMgY5rwMT(H~D(T!Wj=1_->#XiWn4uvw3Djc!lj5*eh
z!1J<=5}x_A2GtN<LCj<Z*`grlnhvX+RCGNI7$PMQDrDyo;6*Kh+`7d&r*zFLh7-Ku
zZp<ZxEfa8D!RZkg$VSU>9G&V}*Efn)-4U2B2b*t?nq`eb5=ko$3~3T_lr98HFJ^#s
zu}9@-O$crU!=8FPAqzko9qe_3%(LIkF;yQGMx(IXrq+OK^eSnmR6=$@CYAZgb!7Nd
zR)C!426u(S9?#=5#lw|^sFh)dR45GUpp_*T6VO#eb0o-3#rip%E1!Hs-Y1H&Mn-Z`
zB>k)9E1*h!ltQ$Rb_qeQ%mGuRP}K^qVo?HNKZ9B?D7mfEbpTEWqUf3Ssw)>-!#@t~
zC9q~Ks|v$PE{L&b=VM%j(KKa|R%|BQB$n;ZHsM=9*mDdCKC@hclhX_Zynzpft-?o^
zBebu#aqd6|kMP0s<g=a<<*0@{$5tkysK#yYD?Cy`u5jAN$Rr3Qy~FgXE4X9at}O_z
zG@w$D<lw^!bIUdcg40hHH;T)-Y^T1>TdNk2n)jZukU%Rk0yofNRW`zeH%U4N6`03Y
z6e#U}>R81y>AzYdbL9g6z;0HrJkP`K*Vh>Go@I40Z8D*W(BnA+0)6#0UYfpGdh&=j
z8W>O+Mk3}Z@rSu$ej0PSew$14Rpr>^*joUg@srBzh^Njr?ZKKg)&l)=k`9Uhp07C2
z<ioQE-3$`7w>@O|=p-a<LIJlf2t9@?frgv=>CpL+=H{Ox7-37H9VA>Hn<ZPMe_BX_
zGe(ZET$^&PawXDr@7Hp{17~H<@oR!sd7AGxZ^;}u{ib6<oieAPe?Y#{rt0%7wr*@m
z_lE07QWpHf<J|8dNHhcmDTU8hrmQ?nA~CdIiG0dJzX*h!-}^v9a;|$nL$+5Qmd!;c
zxB=K2rusl?Lv0Wu=rSMxp2kpXhFeu3d0r{@sR=?02oWmQ#k+=^Jioeq<R<)O3q%UE
zW5UWZi)d2SBq$=Ent=S;BZc7Z;#(<DB=H&)VGb39MzI@O9Vimz)3UNv=N5q4O<Y!g
zR@YYB0^wxREX*o1Tg|;^5zhJh(&M5&)ugXLh2R_5-MV2R1YthG{c^qlSPI<oBTzOS
z!XTRyc;275i@&s2YcvU9C>B|kU#s~={v4(sglm?4A0YEKGo>&|V(Tv^11Ki$%%_L%
zkWNt8%VPnXYQFBjX8)Acj#8`(9M&sCuJ^^qd?QNe1MLx}E4FL>m#FxU01gY&{~kj+
zttI1p$br)Pp?*uHu8GAbRaEBWK+x3Uz47##<fNnAEVe$SP@-N!_qR^~o=ph3nQSaA
z>D4GnmZNk2^!uH7@l>E_zHi2_Tj1j{R1x2@$e2KKwdvPzViI$qlKAT4saeUd-!o%0
zWyO<KRg=!6Yp1IZ%<3+W567M)lB<AOa-~}TuNy}FS^K)5hqGw3FLrct6*8g)2I`DA
zKnxE#)jYI14UQh>%Y3i8`pxnC@%!bEIo$9<gt!ROghojPg%pY`z&wo32t2<WIa#cN
zOzUX=^3wlixnYy|U-25wH&x(o;NuBGa0O*1ZMsjz0364xg6+l;l1;EO&;xiBDDdMu
z`3VIPp^iqvXyTWeLqFZVQ!_ke9<K>X+a50I$fHsex3jnVCT#$gHlhw<@7#C$;V&x*
zlcM4a?ucAKGK+RjlS>NcX>;LQPc|7!k#Jb|!!ZKq&0lHM`dlc_jg5cRS=+||8VSwN
zS&!;8mDz+Pv{~*QXdmW3RUEXHMV*o?1fs!=>f#>(M_|xz+_<T4e03WTM(RvZ3Z%sY
z@qkmY?e7>}6j_aGh1qHws~B@ibV7h;7Zn9}EmXsI0x7?~`NJ5CAmCIEV?wFa!2X-r
zSaUQeLM5LjEfc3vzdLYS0`E}raG-=Eyzkqr$Jkd($K%28bvKd@j!*7U<X2FvP<Hwp
z3xEaR6p)l5+O*mQpR^Bl*)#^!4h~m8zpkAZs9jEimPiETI6x{fP}l$AoiP$~GMRA5
zDgB191OrB%@Rl4%_N|n!l8^!ae8~sQIS&b&6~|C@1-R_hjwTQSim==O5BM$MYR?f5
zH3lqj7gFMQP>(sE^c-+LV>sW~Fxj6gNpHOd802T$pIkA<%2Qv#aD8HSh@l7}zYjjZ
zIt;OLqxrXq@E$HNa8Eu2fhE%&pIx79Qwmo9>?L_g&0`OT_@5OPTAY78ZrOR-`nW=j
zr<9>%OpOtADWm=DAm39Hx|z-y!xItcC3Aai$>)Ax`<_v&a<GvI%@cYOBb^({6U^6;
znfVZs0o6wFMAc*;4GT?dBui~<V+gDSCvZ+`k3!hNNob#)_5-Fhwd)C?mm;UrHM$!l
zte2x;`g<63)V$7#+ubHn*G84$4Ej9{LW}}1LzNJ^5hCNba&VhMV2c^)7m8nLX)1u_
z9{QlbGKNk(W&DNo=nRV)2$<da%^{u#tU<&08C@+_hDHmml1|ZnPL7vH!7HRCsD1uP
z8DPVcq$=(XP;FpcFd5Obr`U}qR7SGUSSk#s%v>m(Vy_^IVis$y@dFlFN&SMmCdQ`1
z&N88WI_PwiIyEZkaJ&!N6gdTuOoh7y*<J<8)R4b{Jh{0q@1Vve2C>tIwp@$#kl6QF
z>CG5A%N7GO{@Kx@2jI(Os-UHuoR=Vm**#7fW}_(1UlW#IeOvo>AWU+l<84oCDZK!1
zEf7M)H%`vMKlJ*F+0f!xC20Zp#-L)jCqumDiFab>eDmjF>CeE?S3W=()V^P9vHN6%
z7Pp^S4d_y50b~7gmAUtk2)5g-Y+SJkP8Jr*Y1eq7zo@{CuNNmCR)`U}A%LullxJO-
zL;Te_R?*jwq0#d8#cG1|yM6;3l=608E!Go+<7$Uyr*q&N3fFb1ogl|S0e$?*Fe(CB
zpwmBw{jBt@KSVmaGT&jCM3Gu{@;ec{Y>~WEGtVm{L$;e!Qd#u5@<KfJ5wApkq_8nD
zN9Pb8P5>t)yr2*~!Yd;-GFUhG;;)!HGu{gq>4i$QQiz6Q3_utfrf7?leann{TlnA*
zC_-0{J?vj=E-v56i4NWvEw4G!>~U$$UOINtd!pcid>{=5iG6UOhq~P4PiADylY;LM
z3ik<p$+yBP62eO}6xskd$G%fKE&aK|%s$avewAaeMugyABMU3(UvneuV(qcCG{v+l
zw72K|U4S;%4p?u#No6ci7FM!2+%nQu`9w{+YM#RygHZ0%2R~3&V|efrJGW)rq~-sN
zgiHiAPkpWiDYY$17PyustpE~SET$IFf~;15RUo7T!43th@oc~Tcl+~T-&2VQNI}0{
zTQswqf-*9HGV)VkN+H4tqDS*tJMhJSL~wIq>ui}UZjlz4#KcNo%(HvG=B%_x?lP><
z%xaID*fsVy6R0`oAZ)fsvx5~gL|BT>=QD}7?G0#IugxpCITS$>H(vTi(F*eBW!ZsN
zZ|3Y~358qqqC9dxOzYZ0tvqnlG^@#PG0<b~wIcVrzmd5KHND%uGfPEusXrU!WpxpB
zH|5m))|!H4<;Ojel`~E%bM#N!$==&KMqF1mEbLv?dQ+Pq@2|53jER^=Vp(ZSdfoTj
z#m*lUW~2LtNF{{*lLpbkNbECY-)B-tiwew($+FE)cc8XjpD357nsX%KQydXCqWZqy
z544J>x7=*52PT4;J{F7ByM{QPEQ_D69~<5N`X6aN&oVR>oRLh#vfDc#6RFq>4Qywj
zmqLT(@)@|$#DN8dC5US)<4>S^YGbkN_JYQWD4ib^NDY3;P#xOCU2tVY^c1?d$5+QK
z7jy-7_D5H+MsvV7?AL35p?-hjUHUs8*?*Sr{5;HjWzQJoLvN{W<sA?Ty2%&t@;;x;
z3GUk~8EZ4jfAm4^e^u)F5Tc>Ta=5$G2B1IOFc><#(vn>W0u5FIIsS)eVEd0r1T)M3
zo?UUJwQ2Y7$>D#R=M*7~!Y~t)>Q95SCtJlEx}I>TqHGk#F|MEzYE+z1PHuP?6!c^q
zo9@W1c{eem2@;*TapC|-Mkc?pJa^wac6oYS$e}F?LMgtG6SJt)l>syh@#FuDgdd$h
zn`h+a0}FHvJ`?;1JQGspeJ+Aej9yo93R8;B+lpD-C)=$jJYNUp(eCg5zMZczL)YQX
zcscqxUprU(OZtL-uR0pm%ojUnrv2&FZZu8!`-*~d7TwvV`^Iq_3&?tu`*hLaqtZi`
zA;I3PIi;F$$jgI(6h~kV?0qqsN^Z(1sq>+dZ<F<is`-cu`sP%TX5a<##ZWG5=UrwG
zTD%}WPUM}aa;n^IEVQHj;W&ap$u6V9GL{?7hmFXEE@2}lEiqA#<1Pxj{y!+d@4jhr
z4_qwRqw!@1Ow!3@z7gezraCwV$3sleHB2m`Sm8}rONu^XY_}mwC7#oxM}2lrJNE78
zMZ53iEfETX#RtQoRfJS5In6K~OE7>BF-IDZe3P&kEq|Nyl4${n4lr(^zZFUC6%=Rd
z@WaCAqBNG4?g<yv=K~=uf|KUzGS*0Xnn>7O@qz-9u<TNGIds`o@+%tK^D*3Z%0n&7
zw^vX#5ZNS~A%%H7ij@evJzLmnXuvcbU9>icIGF>0@$Mq^A^7e_Z(>%%g*mOtoW9Mn
z3c4MPmZTlSrBJ20e?|pB@t5GHH22-5nIeMoLvw#FpFZ6XU;Nl&Wadoe>SWRAQ2ubt
z&=He#Nl2B8c$1ijEeud$Zp!%bu{;FuQ)qYsB9o<GqqlDj+>ty>{RC>MVIG~2k(;ON
z0&Qd9$4-*I02_Orx+`6I`Ki3weV0ktAy=C~ns_N5#?3~ZH=jKl<x@ZJiU|FFcvp<y
zRU_j|&~DVi1E1P8O*O=34~aVt^~(UU&1{x%xatDjf_IbO)Wgt3d<%O;k4c8DY@3+(
zZEiNx6NQsYyD4JBONRS~Nk7~V?NgGDVp)Rc(?WPqd8BF9IBgbH4%$XEoT<?4fj27s
z>?m&2nD$ZP;M*YD<l$diBY>><oG(~tTqAU(0m&FHsPkf&XF!rOB51JEBnSAQT5((y
zjXu&}u&0u6VIt(v)-tjxab=T6sbK4|#qp^^UJ())S<LXV=#M!F7;m0i5(KS#z}w$>
z6r};(Q_dc<wo>wCoy+V5m@DO);5M;e%=)b`qJ1J#W0k1*oTdC0Hq(GL0vDa)>BEVI
z@IL*-+jxJQLBdSNm`F>V8Gf6II^D{kf;WT%5ykzx)soM-NJaHE%$PsyG4nfAg60JM
zg`7v9(ZQeQxPS=>`U}Y}FoU{ke|RW$v{n0p<64o)%h;E?2v^P$E7PUJp^3y6_J}VN
zO|`ko^gNaxAdiP!T>U!<6t`v!J4vh-9oJW6{2$=$2Ajw9oV%K-#W%f|w)a5VVV`K2
zJ3%{iyTjtTYF<;<oW{{`?%J}(J@;dGZgAp<lH}e0hp~6;5v>i@2HUo6+qQPwwr$(C
zZQC|?+qP}Hd-j>UlT0#CGUp$x538!KyXrzQ{9@qfJ8B#-VPf-w(N}223k}u9FliZ2
zc@?fs#O1Ocq)M-SN%`gEM&*M@v;E1<3!8_?4}_mEHK?GXC)wQ;%1Bc)+obFKF-PkF
zEf7^&Mmbz-4-oDb`XsyT_kVT7|2o1lbNsK(pO4l5sUX-;d|&_NLct8fP(~&P8^B|(
zINLDhOMNN24-jlN6HBLRO2}UQeBeIFHx`%6WV%NM^rWlhV9)Nq&M#?jFL`i(m~e6R
z8M8PNAxX$_Ev5{W5PM6M(`TPAWOM|IGvz`^Y)4g9MKKypeY51c2X^-9Mg#BWT;xG#
zXZU`+op^dP`}p~uKekE>LLLK_vk-BvdWkJz?Mq)gN1(?uB(_&!G|f%ljwYPu7QAVh
zoS?*BxG>+<kE^ZO9#o3EnnN0GK@Omdw@@+=X@9g(RsmXp?i=*T+VdLLJVP20LVDwK
z5#H;MUwvVqt00w?^V3DT+jW^_R5!&;R066bix7hAEE#a!`BH=<<T7vj_g`Lbg;TcR
z*y=oPzife}*t^x%i<($Q{3DpNj;C6t8Cy|P3@u-*_$F7Xuy({Wk-|+HS9nr2wLV3>
z0AKN{<Junxjzus3RXB9l7)e<W7RR3{_sT$T&v(5)ajLv3&}1&}npkA%Yco}3r)-c}
z=8}aZgNXo=gDBfYvpUt1QnSl@U(Px|n=E%oM4rfxcmg}8VI$_9E%7%v1DWf0K%3Br
z`EjpJXhow4ur7&Lxg+NAUCQXFjA~6<{y-5C@IE(SGW>Cgv_6^H5ga2ex$)ZcZErB}
z(Jo_V)<c0}>ZV?qy7l$!B!O}{M4b7$&G7U_ejR1IcQ5s+kwHe{+vj4W5SGZnj0XY7
z2dfFX)$sI<H6RXlMtOcLd0n?Ph0R}T;4fg!XExGy(BMa5FaM`Dq@Ig`@<l;4<G*WZ
zI6x%Mitj5{_ey&Qk%h{4J{WKj7k2j2+_g1%P+7;eX`-zH`==Uw@%)+&+*~KH^gz17
z&@Ec9{c8(0H~i?5y2ktbjR`*i4FPM?^Ze~k18@LVcls%aiPOkk`ReU10=5!w3U9_r
z$M*W=Z_f`xU@h<di;~Qy?<qh^=RHUqTfPKj<_ci80%nF&W)j=;+5Ofvd{YTl3;e_s
z6$6v;S19E#^=yw)6cznN!iaKdhoo@_RMjhtLKIr1BOPQ>D)d+G6_&opBYZW)NKA-1
zQIsY&S};S`Z}mg~15<epnPvYo%alhnNM?w|H2PmN+Vy`v#?WB2Y6i_?i`QhA)n|hG
zH392fiX4&HRNz50b@fTcsHp=$Z)2JxkIKu?Yr1L<p=Ge4P30yW!-$n}Zw0XHPp($V
zkmUg8-mi3YiJJNSkB_Wv>zfhPZ9v1t%}W2`b)sET(@_?z$TCf*moTL@@Pnr-f=a1w
ztu$HT!iPM(6a><nX`mz!Md-tFV;lgepsomRgJcbzC6D|Vc2Yw>&+FOu+iD;$J*}K@
zO>eZEu|joHRJ&0FKYOi_uhdcf6b0K#ET5ju3(!AEQ24l(RP|clb=Qpowj-Bl_?_D-
zs01;aP#1mb=ICJU)FuewH&qO(-2J12NZfJM1$XnSa5JJfhK#~~x)1#uC#Lq@YT?|A
z)d$?<VUxHycHUUdXoJk~!_4&Jc9@`8Nc8Kb?Req7*y4AMjQIx{zQVrt#I5Ok!RDJu
z#&wVXyiBUI`tg0!K7985h)$x_P5U0gF=zAd;*7~tVu>khB^94QMg_e*myTjtJ6j!E
zeSFL?c;S<H>?TAkuQ{0QIFS3Iy3veSdNIcf$~l&KaYd;)l+9(Is*f65`8Z^bF1fW_
zuS)4j>Fn7S7Vb*rdPD5y@V{Zb*@l}!N8=AYxT(DImw+`d=S*n&F+VeGirDn0!hx5&
z!LL69+*}?f4S4U81L}LjKai5gYt_NLCE$UQLq>o70!BY7^?6=qdO`Nm$O3$Wf4#-g
z9s5GN4f8`kqLuzf>t_9b8@2sEMr^E1%>Uc7+O@{^zt)Z5_o`RWHZCxACxsMY0I!kc
zin~!O_042FlD`&BJe8^_G4tp51J|RNn9_EE74ibmt3me9gS%VzC$3BBJW=6%)%@(@
z{ErkRk&N@iqPZtp1dF7xWVv+VIS*Gv2##O&Pv>u<!gj_;I}wc$!H|sS{p7&s`TV{6
zATk!YF^_~I7Cv4yGp5}2EMTnkue;w9;CPT*m^cnzkGqq7w36QM%T3U>jYyB@Jc%{S
zKBfZK$n_4|s%S_5b5(auREEVJE&XY3`Q=xs;Ih{XqmTyBpv5pzLcDSlMM2^}=GHls
zvD_@Ic)Z2d5w%>7+N4L&k%kH}V(WHOorbrI?FFM!gWdr7qoK(}Y59bA1WjGMo_COl
zMvHhhR&o-%OasL<PKS*NY`D5Sdra{ONkgc*%4aF}^qvP~^>|4|R(3vp#Ed%-8GP<D
z(Z|BveU+WI3k$9fmdz6Co+X+D997x8HEW}F0vUsjXazw<xFvE<%s<8us!zQ9Ajn-u
zXr9PkA>QfQC?{}jpn!i2<D5xqqO>Ipm7Bo#8Fz$SH#Fs0EPv!a9Pdm+7ENf?>(9uA
zgro&c7!sc+B)rxc=@J-FE&cHQBYvMQQj5hOkfbvVbko(YQ_N9*SXaSt1tM#IdHRY`
zKnVl@s!I?72BrLr&rga3goRBj40hn#qxOrkKUKtpI>AukSk08FQcd+32XONRzyMNF
zx67D(-;kVG48W&5kf#NN$bCk$*^9ZH?{19XDz_R<q+s~qSVMuW4q)iPk9Bp-X{IZx
z&#gktfUgv=JE(g)$)}S9L+jdi;p-9riLDGK!Br$2S8|da^xX_KO1%^k6);UZHq`bE
z#G7M~1qQD<$E0XD`8op<=52dx?B4B)syZ&T-C9P2XEP{R0|tR7+!_`lsR*TSNq|&d
zYyv(a1?5#$ecsJHe_PFhJ>T=fJ~m2ANQE@{{1ab;de?C0c2$Q$|MHQ3y*7AafA~?T
zjxg1wu^atW??(6c!=@s~9Dn+I)3$k^lc2m4@=EBcHQ<YMshoQ$uuE_c5U&OA3gDwk
zyIR#4c6!&-f?W@5ON2SRL^sH$Mmhg=elzeeUq79T(CYK@3%@=aWli{=3)6CJTl9kF
zm50>M>V?tH?<Jp|g-PJPgO&o`x54s)WrwL}4yQ100jJ?!$QHzO)sCais5lDx7BFIJ
zVCl`iioW(^fwd)B*=}xcy{*svhmZpJv@rcUCfjnGsA{A(DH3~bo4%Q^mF5cP-)c!7
zcSFf?F?XS1x5F&WC8yH{d%y#E1QaVl>z(MnhlYTS-8=t&NUXR`1w37llH1o|AF9wE
zRN<^0gg2F)V=ed8CV%Q$u^5UTaCZ2d=g*AFd~VqG_g16JOo&91%}KuvS3$1<JG(se
zJ+n;d4~g?B$wEGw{Nn^m&x{t(yk@J%zS&8M-IR{NQnfTRK6~%L>Was@|EOvK`)UrO
z`2u6Pj`Sy)&YQ7j`hoySDA2b>PMXa;S}7q$SWkT`MJYhjJ3~HyR;OqF0H=M5{}*`Z
z;_)+PE@mbrN@%*#UKjF*aL|+tbko4V;NwsNRaMNVqBGyY{@lL)K(h=tOZdo$8UVF;
zPG&Eh+sH<ISk~$i-tOGU3a@0LhUg)^*$5_Io#%O6<kg_cA2WM2YxhM1q<+rN9W`7r
z&ZFdw99;^ZAKf!#U%{s)y2arj+9Qx7v_-sw3{}9oRVfi#?$8E1a6;_6&q6mo!q{|U
z5viSoyxlkeY0JUI8=$NaSE)2(dE>k3fnHL+8X2F%XGI8<RGXwFzNprL1Ue25Kw7@b
zhR9XY8H=bC!P?1>E{@#6YfX>r?q{eYx-jk5_~8t2R06uk<BdIabOLiE{128@bp`!e
zYKIpOO`Xi1PadcyoOoxIak;h8&+Um45-x=v6})x*Ri!%9*t!>cazPWE3dzOKFmr#~
z2oj6GJxp|{*XZRz!dQJUs3xj4`4FfwDmO0sGkChUy~lX~;=cf3i5_+THA7hbUuK{G
zXNLTXh5cW{|7lE}e;mvQ-aa8aa$v?GI|=ozA>AxmPxZZxB_9e`bnuOpzk}n6)Z-Fc
zdVa#N74+QWz9i*(G@jycVUN7RSX*E$Z|4VkQBhTH(PV-`NYgI$Z#WRR6cvyuRO;11
z@C8JO8Ph4H`v<RjJ9@Jx+Vgj#ZB62#(4UWJgi-vy9l1W5U!NB_D8kl}NK`}|)uLp{
zI!=Py_sJxPYE<dIRhxSa6<K!ME<NpUkAVXXEcyZeV}L@D8BYPFx(N;aKL#iq`iT&z
zpek72y>HK27FV`oApIUd|6_oHc=Q(<FrTqr2AUp(KzJ1KVOlwbGg=+IA0C1LC44k5
zeIJr|Or|Sc)=kfM<ffyP1w{!a4=420?u<dXVw_xD^X(yK=!rKwmjbU^yu03o09nj9
z86>L4#8fv;)k3tr%9I9f^4T|K1&v5pX%B{}8+XTS+LT9QrzB<E31nqmuL9Wb@2wtX
zVb3t=q)*UYwSYj=xMgs)bdIiP{&q?*<<B#9B+~D0(HpQB1->@5B*Y?A-SS$A1I%dk
zA~?>;IqjUsoxS?J=9&6xw+urr0<0LLP$!SUA~05ZeE;b<o>vhGWZL<})#cMTaZ&oP
zzW*UH|D2p$-b4jKVAOqUs*(LPul6r-Nw@gwURyE3zG2cqGCCE^ZIh}mgJvnV(<wU3
zTa782+@yWgAK_Rl8N(5r(Iy*eI=oD*z6_;+>WqdveA&<w2*e_oDdCs~g>DvfoRf%f
zK@?%td5l()L)w>Ia&=a&$Pp1J*MstVrdsjpV|UDlC`LRWa!};`uEs&HQh3v@Nd-4i
zbNVg^vAIfL&08mb)*Kz40$}Yw_Un{plfSkyN+$656@^l72D92q{zMZM-r4a1JOA##
z4bsGdNL5mLbWn)EKzKea6d)Xgm5EA)`oOSCNZqE<O=H1QFb$&Xd;NQ-EGY=X#w6bT
zKEX7mJ}kF|Ob0_lV`IX4JyO+h2_lrX>@&ooEGh%pL6A_4;)rCvISTeyQJ5Sl_VAg3
z<kd2V=5}yhPb^&rg!BYpWXi<!zE*R;mGwKft|rSIY1`^wtTXx0s`(w0)fLh%_SpS@
z+Ni2?nEh?s$aX(%GK8(~7TqIB-ObrUu%5;s8+&F;UgwC>WjGb`fzCD<P#7++6X2^H
zyQrg@@9YJXxGk=%z=w-8+GW*NDq+&qQA(}KpglCU4!Yr$mx@MdOPOR;(Kv#RntVmQ
z)D9@k7L?!&WveHIa9$mp7^4)@Q-v@&clz^55qh<hjJtenGgCPLFxnROPJNXMXX6oY
zse=ItGq6iNIzeTZ@YW5Yx2p*xS8*=}VK169GULOE<|y#dBVAn6YF=#!jH5$9p5biS
z=eqP~s{M}3=lEN#C*vEMSHYn)l}t=6H0(7!-h=%V5etZC`-0uRlsZw3qXG4}C0a21
zu?TguH=w1$ex=}FO}fm#i={&DkeK=SaOv+^mbC@6fs%`2@+NucZ+TTFgNX5_^-t*(
z0Af}qwko;4o$}DCnw!E4oUI>VJq)6{m;YqsA8_-#^^FTmtI4LVS|q7%R*P>B%5bd(
z48cXTQK1LBo>z|Fe0h?KU;wqKq^{Kb$)_vVu*YB8%i~<2l9eDyBxFz63RvGCur<t|
zt#RYs(?8ECNOIrWbo}P^xQkK|4HWvIpKYc=LJ<md4JcbYH_$Lt3au-bw$}Ypw|N>#
znZ`B8DxZHvF&{l%R#P$vl7Q>@WfWek*3nO`Qln@&iT9o>mVxAmQDEA;G{3$4Fc%X}
zv(a*Ki|1kWZKUbG_Kzs+a+L*?V%yl{=q#D}gD=`Sl>@_>UlTZ{k&d>y2gAxQw|GFy
zI?i0WY^9WNtece#wl%mQoD&*Hz^-bhjsrUr#G_bWOYiIwLgDB8`}0xOcB@w?@no(=
zfEQ*?#~DH7N+t5?JYH$;u<7aXFK8$Bh0A7A8=Xa7f;#Cq$1+gEcDh8WO|p$));8+!
zZ6A1PXeOk-bMf>;p5<g;dTH%tI$txpoN-m;_gfQ2nIyvV7+_v&;APVW;6gkHXT2Z(
zx1*!u$AGaR;FL1w!SH5o786`4sctepK+WC4x4Tmg<7H`%yuMY3qO+0w#yjwZln`b^
zL{^v*Le*2NSpr8OM$6QGVTC7ty0gT_PkgfWB4D3DDi{>`$T84cduon@guF9ws&yn;
zN?FI1`t#}~Vvuki#4=#s_fZ-+euJ(B=hh*lfI9Q|3<<!e6c%%IFOD1|&q^3*Uvv17
zH-otFQg*;So#lz=ewc5w4uy$~(fsTv`^4fRKanW1Y1D&-iUlR2pDX<uQ9^C7lkK_H
z+PlO)GlYr=h9|EnPmqnnCpjWqHA2q~k1)gmOj>a<o@mcCU)Rq`+F+v8W;_EGmpoS~
zwBrZ+$|d%jpYUih?%qqUal^A%%wvC&{{C^6kXn~7A=@}HS_qKDI8CCl{+8G^hW8vB
zu(2vwjIqD>*}5^s&*4H#<e}VP&roI+$c0|wxpz~?%}$ErY<su3EnzK|4e7@Fz{5QR
zLS%Nf85s~2Ch1|ydoT*YaiUeX5p#a?e27}Zv~GQ`19@K;!Pm9N`xl_<6I|US?`q@L
zr(;$J-D$N|bFxI^r1tV$5Ab42x7P7YGJBY6j-U8UeyWB=`@w9p{U>H`{t0M;>JyE%
zxC6BumEhk`UG3zPcIdq3GZ6d(?^yF!=s(~D+kZSsnV2~L*OHEHO|5^)CaAu7{}T?&
zXzGqA0eOv<ZcDdaIA)6cTChd~6*JusGPW?Bm?Q1y1Md=REO9tF<%|aTVW@yak3a78
z&NsoM4#wl|bicYj61AEDg_Me65x6dsFP%?3<X#-KPV`2rZ2ItdbxW@co%>`7iOFmL
z`ZTZnac-w}Yg)G>SE~3Z-!!g_jn~_qxki!wo=5G?{JGkthF|ag{6dPR+xu~juXfvT
zG27Suz4xbU4+K6Gl`7moz38ic(RHh$nMM{SmXCh5t>g0}!74x<961u@kjgfJNskyI
z+6g`voJc-1Lj*)2v0i<2z#6v2G$}Yt>#SKQ{;~NE6j7Jk`(G*?&$hG(PA4eed*U*6
zr8b&Mk3>^pZM}l#mX8K!b&Se|il$M8H6DE<(e~DUl1}INf)$o#v+yHA=_vEbnh87b
z=1f0`f^|kLkx1T0sRGW*@imJZciQ2#i|!1+K?C+s9j0~K0b4d9nR}P3g`f_AGfjW7
z<!cI+F{o}|b>n*^UQ1oM!A+{g&S3u~)i&n^YS~<n9cvQQ{NX~Wiz=EqG);rzrM~wz
zO^{LDPM?Qim0DBOEIKY70}fy)q6C=<%G4_QPX`m$#b^M_)SCWbU{;h~USvXbHZz0c
zj4=?ura{p_((6&Vu-V;ZDKgqQgpAjr{>bMJ@a&JYl)Ep$8ShHZj)AX3VzX^hnqI%Z
zT*pcMPaF$f80k=+`d)hw`7L(5ck~B+QpG+`b>lJD_hOJ4f^>@^TZB)tzU0?#&BXDd
z0(7Zpn4y=~=V$@ipO;46j?T@=-w_S5c^@i{_A-WeM3f{Ru`x{1)#w8^>=^Ge+Iz`)
z2uaGGyZaw$8t|;GnR8T;^;;qyhml~sKjerF6imB3JPu<knT;&9w_{x1ieBwO{B7*Q
zPbh3ZVvIKrh>y0~XBnrl4G#`ZaeZOpWPJ(*R&U3AjNhM_!dQ2+y5dPf%n9N!77?O%
zYL6XNmDTDyQL1!qR;R#AJ~z-oq@H6=DIqN(tT+$aN0~Zu)7xA<mr1&N4X_|2-9Q}F
zOdB3KyE4J?1*Zx29<a%1UZ0tsy1o%^HqxjD9VJjcrQq<!>O1i(7}kd9cQg#n%$b6o
zy!;Gy`lF7sbSbd`I!G;j7V>FU81qnhsrB(?DlsKcrXb-w)>cFCE<=6$@2!r@l<tF?
zYqYF5uzQRSSouexr1`C(6j`EIz+Yn>Tg@02MRam>=;+cSFMN0&rUNKpWAWKTz>~^o
zQL@obY21N0B&@LNG_nCpnd=VF{^hd5(So!BGuSM`!TKWM@?PcU1<-csNJqyLDo>Sb
zwRPL2gJ737P1QLCQgiGrT3n3_JXPCd(-jZ8Ap<oJLEHP#a~7hU0SfRn5J#549i4y|
z;0g<HtQgK^SRF0>Y2o8(8y7o%-8YI%e<?r0R9^6@0xnrrZ%}0sFWcx}e|XIoAaq>e
z*+-xuI%E(m4@BEq-P&pdp0~H)@y>N`piqi)hU7p9G_(&IRZO}xEPN*9jfXStSGK|)
z5pDzZ2a*)JtU<PCPl~A_M-Di!Tojj1aU&)bsut;>t>EZyrth;yGsDl_*RBSXfG<H(
zQ-_Ot-4MPfrtL0XfrTx_$-5^@M|rpnBsXgp5Y1BMt1clM9*Wclyhy6~@Uvh3xe#iY
zIx}&cW0tAzJAs3=Rrv3L)iJaP>FS`xN!KXNpuFOW+Lz$)ikErsA$t48FXj@KOxIt(
zE2z>m>a?zvdL+q;x=XY;?wR_>*y3t6PTmT+i-VGx-%!FI$54K(ObBB^px5)xmhp^j
z=ZP7~F}(3sN!MXBLDGYPW1K%5nNwpwu9RMmGVGQH#9P8RL&)+%W_y#_({P60c*LQJ
zb<Ab)L|fD5H0pZvwRV!W%2Qpc2xC<f?%H<ma@*1jZIa!lTrMO{)>`BmS)>G}Qr2g0
z@3q+H+8%Z=Y<ULp3da-Ym0ls7yt9~)M3rE*R!;ttOc^ThXvVFy!;8n`p(bWGja_x@
z1H1Pz(qJ`v^@~F;p#_PRnUS>f6Geg3#*%0Lhzq?7gK(QmY>`b%O;I)~@#FG>4yxP6
z6RWfd6gRb&@|(*w4YpXXoDP>dOAUg;LcrhLnzifdqBj^1A_V-JIb_Z7yKvgt<vn2@
zl%{=FHbqKeKr2R*U_kO+7TjtlbQtakuq{=Ao{Io`tz_9HZ~}SNOzx1QMCGC8V+P(u
znY&Z>d%Mhi9)m4bK=Mmgv|bnQkqt`5q*;gVqWj}I^uuTh|7phQeI8VKHbn~C^yt}^
zC&!5SxAs3a*X%jCqf&LzWT(U*Z)Q*W@de-q42Rg0X>}_0RN6RalfrG9A7-*SuFv`i
zce%ONY@$FK660X?NX3R4J5G=gBlb-9O)=ALezTv#C|aoIcz~^imk_dJn;+8ck{5_l
z=$PGHiB&#xS%ww$&V!{WY%`gH=_RraS5sMw`Sg_HW3xGyae~{q36B==M1HKvZ0DJf
zV<MssrMApf!cO)VkqGU%i9Jh#3q?R+cjvKoi69QN2L<JCA1*@;mPd*nj%7eOWe_-!
zKYS*D`KC{^zg&2~D3%QpfGW1<@O3ryh9u$IFS=KI!(7~39Zceu%UT2~jY|yW6T9fu
zz**bHo|sa91q_rB*kJCzkw-EMyXGzZNVq4*2_l=Gh{QV#^D_Bx6fEF+RhvSeaBT#M
zbZ08Nv+vjQQXE{R`6B~gf)(;F=~3gY!Q)9(Re(!Ls12v^+gfI!<TjVU{N``nCdZo&
z1S+&G<k!g$I#Qido@g1)3IXF0t|-zl_*9NLXb{_6)Ps(LVP_Dz^Nmd$o^jtBpy3TX
zy)2#&yJ0(bT^T?R^^4=-pAUvta0WB{o9Ug@X#X>zGCjM=jTRAzU*Lk5WQra-H`>ht
z;9?+MRZQI8XSfvO@Z(u6u29{%h^^MKCb?}+U9YibTV_ZFXZQ575yvb#on<5Q!#+}y
zdfZkS_{%ALT~;vPL&ZIqGsRCovwSu?+uiNus>U<%4c++#)T^hyZXyxF|AJBf#cgF|
zW&Pi7_D9tJ1;qVD{PFDz+?51e$^4W=5|r20<gU-^9M6d_ba@E)b1w@mRV<`<n&jtW
z=4MKnxUxM07wL_t)?~-r#5~K>CE<)tI{%8kTSKSYdm@2%Ov64XQne|bDNa3vR8q9e
z<%w^V+AyM+Me^e@z!^thx1;U*I#e~Hct)c8eLAYPy7u*+X>e%w>*?0h+3EdgE6SU8
z`x6^;$^IKM-&ecS>s@o?E|pU+QktKzt59k?B`U-vmQ4zzDKTt7!ps@RGK-<+(Y!Hm
zw_JKrIid)KhkxOt;9LKk^GSVBczLDM^Dp3Z=%=x~;WU(NNo7^dDtTQi`8f*;^xL4X
z!~u4?h!3yr#dfjCT4~7&#nn}Zb6HS95#2k(A9(*#g*K~`-g>)$d+?<8`t<^125;Sp
zD5XT2It04gRudyHvfY!ymEkbtGmUe(M~yR00nC<lp@Y?Y02!G2>|C;n`IKQJ-DyNR
zRrYiZj-CEa_{U-TWW{H|ZPe$vn5x5AB2^~s`SL6AY<Pti!#I(|^>?FkZ9#7p1hK5o
z<j~sBKC!GO-eNF%Z}ED}%s%<~t6sS>BfdC=+tQ8Y({-S!y~8RB_b=(BQh^K0Z>!t9
zk&fD~-*L5CRn#sW80df-PIN#xQbzd1c1V&`k%2cF<~R(xf|h6-I=@d%VzMJ-(4tiO
z*!~}>Y?cIV(6jXH#>f1bj0GrU%o(T5@!RkFb-S^+Sv#P`RBeWF3sH24(^T*LF%I96
z1*Ejne6F?+^1jL}0s?B0P-O4n*aMe21C)xqe|YPp?M=hy@lRm*Q>G|S^jH(3kWl>C
zlz+Stg`Ou4$r9$lMwrVoZY%GPT`m|RASFrD`5?6`9pLDNJq77j`FV4%ha-6oQ~&Zg
z%I6lJSUpR|zf({c4Vn-G?b4Rvef6}EQ|co<qlUl(nd!+^*l8;xrV`aRQqns!nN$K}
zg0F3ZyBI3uAJgV{q$E=FJi%GxiTe&8#gY8rY}R#^ea!5xeKHv{_n*%~0qRs>c?41o
zMvvMB<0%+nJjOKUTvLOoGUh0AvM}H3UwsGD(%?Uqz(Z34?%5|)JF0EG|MVE`siy?J
zE!@saUDO|rK{8rl26^V{_^QyyIGHlMmygciZ#hR~z1np)c3GWkwsHLsq<H>(us4x9
z&zK4SqE&@xQVb=i=3+k!TSj8a)Dwde|EwZ7Y(G&@zI;}iE3{kPt<{(9x)Xsvl+sC<
zN(7?7f}G*55`}{0oZ*r?MUP0>7cjbHgC`u(EMr!B8zx&TQt0;1;F<--&kusrCPR(u
ze<OzFqN3%J1SrD$-c>-OMCO0=5=+8%H(YAz$NIQIx$fGZuJI<ap0k9Cz|D~^NWkp7
z&M9(o+_hQ|D%0fp9z@$nYQf+<7hSV21u@@{N3ydH3(LESI5E=czZK}JUB8^2zt=xc
zZ4yoM(&V=VrC34J-Fi>0|8Nvvqg({C!E6Jf#cct$;N?#XhRU6a!nt_LA@ehLds4K8
zxp0JxYQxOKHXC0UrZ5dWuZ>8CX)BptD4Y?-b>g9Ht$$YjJsInmyRBds_i^MH=r#93
z@%YnD$Qz@9&P#jBbJfVk=<EcrZts-tr3-XG8eDN%xi$E+UmRHxcE4LtWc|K~+2B7J
z=`VE^jTnRiwU{>@4mwMCea58IY~zqkDEO6|o$mkemgKNyK-BRNbnN|7*OzzSi%+L?
zs&0GlXIOep`PUted}z@j+JI5aQK(`{F5rcPa5|mMv(xk*2F1}LJ@MTcN!hmV682hm
z>FGJ_7+^z<4^bd#v`us!JO$cP<HMI`EHl9e?{7W}ZT^kElhW3rW&x5Dho*LURw<dX
zthiL}z&9PCe(@20K|X8>pxd*)*znnjfDt11R?N%&`6)vtfXj#U5fOD>;YA<2N;yt}
zdfpu}^iU}kzw%`tUtsAvg<`XRKY5|O3^hkKj%q0XWo6w)P7R{$qPtIZR1Pg+cfkj~
z!ilXZsDWwMSigYvtGwc!<4M|c%HIff@Qn(Z3qWf-bK%n%QcQ=FiqOgIjOcQq)gi*<
zToPJQuw~?)q>#vz{YfE7dT8A}1->D0xUQNN1UIVj{hv@n@F7fmM0K}84Y&>bAs!l&
zTog{h<%2`fDCR6K^XbQg4f_`0N@+{@k&B{j+i|;)>q?0nOLR(i$8#X8uW%jTx~1!s
z33pAGskfeq_l6P(C)P?#OJeyddz}4d9k376a!2y-CkFoblLi^>p_oW;y!}BrxJf}|
zV7*xPdO<)9*qTNT_o{7?Fa0kXFBVJ?^4C%<K_TJ#8ye%ibZl<A-s@}MG3s;W!D66D
z4#_}$f5%%GX6l`N7(ixspf*4XZ(AsmLaba_zkqDcE_)yX!@+XD+&|+<zlv@-$cxu@
zu><IGl(zmS+~0d_wGg_yxy#rT_vfC^!5csXxnkdk(HopfL{GpWR@cINmG05w2;JmZ
z%-nx1QaKp9d8?qc2byy~?R~KVE0V()L;o6E^8CeB#|!oLa;1i2C>)G$JJp@i$dsdt
zVeaih*&^lKZ+e4T;qG<(=YVfn$imt4_gA0)joqNz(3KucA4^9+|1tRR4%`7vzUU7n
zqA`2*5L~4?!9bd6h~1o{K9pa9%Q+V??@uXo;^j^mzc@sIpy?+dvl#s|<zz7&{4@70
zL<K^gaXYkdP(SBzTvB_azf4{3=M6Q@U~{AyxZ*yon)o$sW63r;oG^9r`o<6$fUggT
z$pEIu*@J5Qj8RF7sU}D|^`zx4@@Kv}|1n4$Fe4?I%E!%KZ4<pIiK%$r4TM|AgYx*6
zci*gWR^tFlWvj5j{rFD2meB-eZ~w_Jc)_kc<9}cm_Wz*RGBI%euU*?x)Mb)3*^qiZ
z)y~9K#8`UYTqI?#?rmFJvz4nSR3~Nti3nk-P%20%W61k_vCqIn9a&^wh=2&@-Oamw
zw-K~(Rjr@n*!it@yT9{~D#N3ci>t1ENf%WL7t$+?TWVDEh%@T^c7HCf!Dk}J)wGWo
zXF1mH@OwT_zh^hY^j5GF>M0^tvq((lKR;bc3u`_*M%g!aeSP18r$6Iv8}5JhxBG7W
z2ITy{&JW4kHn7F+(ALzok0`tKsc#${0SY3QUmy@BUJALZvKBbHn7A$qn)DjBU2iH1
z-QsL)KKOGToa-h5cZ?-K{7J)}@n?B^7v9{rO9R<5*5{?RUti|Zp}d4I*%}X!n`t7P
z+&!DI*!@_?h_;p2+SeAW=}IqMu59{ip<P;v9%y8fg}C#s=2tqa@>a8>&i>g*m}md#
zdf8GA5Bsy7d73cK`U1g2e&-FQ%W=oCvG=%@Y_kZT(>iVA6`Vug(3d}LjBSA!RAi6+
z>%ju+0SIek1`iG2n$yFT9aoBJZ!E(4#Bygo!7u&HTg!V07nhM<olD|SVEzmri=Q~F
zcIGhFA=GFXybu&uB7Q2RcZzlObUhFAVm`Rq_2_S!>6$-#N&YRZ1j&kI3d9_dq+Oky
z=s*>IEVrYOSTcWSUhm+Nc)5<3S5j-G1lXXGD+aGO`LV_Q^rQg%9eL9CdBbh}1dKy$
zTQr!kal^IEJG***h%KNrqbB*F{s0oNv)YoCDnibXE_y9{Yi$!|hJ#~cu(~)}#D_oI
zf_F8t7c~)>MC%^+(w3f4czHK{R}Stto%tv3uj?d)a(6i9AxgkwlkWhp57)b`7Z51M
zAOT{X)+qwYn(L{w-FPg2X4+RvK(R9m`$13*TfRl%9z|7Mu&BvICRm^`N5;nT$uE3V
zNU<aWI%@<}nRALO-0~VE^|PBGk(pHmr6MO2_U;w<VggrT*b+=T2|Oh^9cVc8^yVQa
z4*9~BV1l@xfwX@BDY*f?22xWt2s(R`7%h#=bT6;hWQoHE{bIclfqYq(@2=~g;N0k{
zlBjQV19A!d!cVZa1TmFO4C@#zxjJNmt$*<hP<e9pWkfSf0qwxNDgNan<eK&|n9`)c
z((Wx)T5({RaUVjvS#_m6@VclZ#?X>(GY<lC7po_GDY%(lx5)Ld<knk~d>EzyDZ?4|
z@Mu*ef0od7@?aC{wr4r90*gn=)I?U#2yZ^WAFWXG5Uw-P$Yqm=PAu4p=;CBM7xRbX
z<0x3^&L+y7j}t7mTzN?;@~$c57Aj=CWf$ejHbUblp`ke3aq@cvhqRA#WE?v{L??ui
zp70bvPmJ9a@7fm)Ij<o@(aUsc*eZa?Rrnl{{@$5K^-Gw8$_~1*6bQ%?9iwefVk+Y%
zHpbha3AI6BxQ+yG3;Z!h!YyzqF$xceFcX104wyh!EWX4F8pS4QDUVL_|H~E@B4Y3g
zEwwi7-#hdg<z^|7Ny3%>*Ifdu$Yp{jp~Qhuf@Q!*B=k4eg?Vit6A$37E1tbnWsybF
zl&2Y1da31`xgEJmVhl~Y)^YAeS+<Ijkd$PHGd}L)(53~yU`R+Fr~>pyeyqcuKlrc<
z==f2Z1+<eOS{yZ*7THwpeF0ww?uzmuXv;k)$z*$ceQo%Nr!qqV9ggt)=A)G_4CMNs
zaKK-sn#B9odn;oGcSMDktV5_j<xeP4)EU{`-cZ8Cc*2e<@_CL@@KKvGvV#n;;PC#x
z?|DbZ(0Zuk81BtpV|Qs#W`qV+^B*T;__e+3`3688tx_3t)u}&wN5QFxL$T29YN@lc
zP2IG{B8eW2RO0wOL)JhmaHCQ#h<>6kBUqmeYhvsefbbWrnWS(rDYon2PK>gk(M2R{
z5=NUMVtP2L6G)=>&iQcE;0P;9+k-(mQ8%qW>STvg9Ki&g^N$50DJs(D=9qZhc$8n1
zf%0sHB*J-_%X;{5z!=sB@)yVjYi!CJcf<3+Vwo=7cm|L>QRuFY<p}l>3$Al<;<yZQ
z;G+fEQ5B&h+T&A|;?$*blK3HYw{5qS8|%()&zh`H*)kbq+FEst;0sY$l%HmA(%RUe
zp_~yERTg8_Hu-35Dzg&Bm4sFVJN<kU6C5FI_mFjX2%>!MaEoF_e9qTiYot?;+mFXs
z7P=GZ6At=<qwS*+KC`aw)b9th=L<ft<P&ert)A6mFat+z88}fg3mahBkJ?8(bGY1@
z6+C{7wxtKzf_{d`zUXZDXo%@rNC7y<9uG3Zx!E7t>b~Ln{(66XCNqg|)*kH4ky=U*
zcu)uGG8MDtugY-`hc(|H7-(Y5g1V85>B^4vBWoYAV^qE4kSHI1MDu4SEP7dUdB4QY
zqIJ*aPFIq8!k=RQW!wXDBw4a(ifDzmCZ}5b(Wl6NY?WH<^5T%|@t;E(?%DF8BgdQr
z9X*Q#Ey}2hJ(6}b4h%2Ol!pRy`JFyI-An9F*BU<j9SkOb0w`Czi6xF&P_MAI{TmV(
zywFr(4GHpW2yr&2=MUq=)cV=^j5!C`Yp#_f(0}R<?9gRJ^@v(MER{g9XD@V<`$PD*
zC~Ha$X@&{XPv=1Q*1QP)@E6&;V^upa*w`RZXXh!5k)MwByh4X5d%j`D?vJlQ_lES(
z56u=NXZv8lntj5Eb_D7haVR!?pu~Y8pUtz!5T^Bs+HKIf9<Y0zC#{8@I*lL{8&AlU
zUjDP*YDoBrTATcjQs?_+AGvGudQeYvOOP&??bokTTu5kNx)U18v<mHq`{3kH`{teZ
zTJJl4s5Mgde<6_n+MY16G5&A2C)5AKixflp$BP{J0~ODvP3I<|>jmDvdKBNa6!r+|
zBN61Es)~ddsVqJbtM%k}J2gWmx{zp1PIsd_d~DBedV0u?57pg*?*7~UX5f2&0Sm84
zZ&Cr3!mj5b<iVs!6JE@&=XvWhuE>@lf%?WT$8q{|{Jc4dztfXDC4~9L-~02~HF-HH
zf%E>@_S>83o&Ni{oF)F$mIMeb0#lZ;EfVQO2ju13zrc~afX#o|;&!q4yjttY+xO-@
zu)o%Sn2SKOw-tpn&T-eZrc0nh(QcFNPGn$#<k9k?ivIhqN!CT}2>8Shf=_8a@9%iM
z1fa~JM!CP#Ph6Uq4w9C3E#SEGnE?g#d%<1Ew~wICc1VcKQ;2K#Tv&54U;6V|1&`eH
z_`Q0Yz3k%pt8dh$0}xo67rf{SJkgRtGg`&ZlY@yg#b$N+HObCM__wNTR^28&98Y&>
zjy}h_Z=_x<_HT@!)}^X$F^Uw)>xmYJYX!Qfs+{}5B^VBS<Tng{gkRROhpUpg<5mG^
z=d^oxn`9U_p3^fQIFEEXR*eO_hE&-pEN))POON%l=4z6<e>D8ph};2va$tnL`_eb@
zRY)Xr?!%w=ChhpLbcdEgxe&V5>k}M3BCrn72M<SN#xYvFirX{+u<O4NfaGDewZ;wr
zZt;lezCucF#mAf78|)*z^_gNm*YpWQTn&SNaX{#RKr8EanKYkB#{W=^n2!w{-(!bv
zV~rT|Xfl@+-LkzK3XfvL;<N)=$hE@*<w+1B46mu^bMm~@cze<zgwVhca|xN5T7_$L
z&S~kiJ{isgn2Es)#$|bViaim~7|8zkfKsp&%GaaN%w;o!^9BSI`zHNb{+Th}uw<ux
zs-X+NBH5a=(p^}j`Y2{Eg)&US!Rws7u&&b5VIK|gi{}$L<bVvTDm(3gu7(j0Lbb_B
znXasCGna=jmopU@Mm&Td69$`_o(GwIM>bD(D%Ee%Xjv(#ZCYu#^T<oF**W<@fu4{+
zXu?hw^lQ5AwUCfHqS%anN_dN^wBD$0)@urR*JXtxj0^<xh?+WL@Ypy{Xa(ZTAfK+9
zIPkSDCz8nkC?=?$(X@*80S;IV5tM3{m6`Q1`ogSP^F7zX=+;CIMid0NAMGW5=~h1>
z1>iPhh*CHq<%CxGQp`vL$zcw_2gWW3Po{eeB(aM3h!EM6PMYfW6n#lS579LP?ZIds
z9?;QMN3sOXt0K~>%TXYT-$TyXfI(Y!O+)YenWTXA<*5^7Gl`p4m~mOb_T(HIYtxTn
zQKjSly*}4ii3azt)rRfhNGU;5^7FC%84hv)dR~w?Uv5I4L{`&QS<Z`J8i;O&NAb@D
z&;mi>zsR9W^XL#mhX+)@J0B$6hmDUpChUPe3zc?|i*^G3GJI`-7XIzCF_lQy7U0kV
zUXhDPY<pTl$Z-mCKUprMbX4I7;P&4ph&hMzw=gjgz9;KiHrkqyie>_9D`k^ULwkgw
zTLoCGf!+_Z5{|!Zhy1b(yJUurOet|a1qxPk^L+mWj50t77>`v*00#{6{PK&AG@vq2
z<Om+wtRBN%LtQ>dD%zN1ZOMC~g^gZv0B~w5EMTl0ew*vn&aaF;MwuO%0G#5jc=-Mq
z7iS1HL}Vsa0xvX|dDd8hC2W&JX8+|ZKjQ}&?sEbC+}WQ!{J^DlD<!~8YXHZDQZjH<
z9Li8pe-YT14H2sm#^$g9TGy!QOJ5j~VkI4BXD5{%j_rQ;Tm~IYaa~B6>iTJ<c#+CV
zB-mgpYW}<L$lq`dphmB^MOfwSA<fXG*>?RssskJN3BkJd=TJ)>j$im;R7=qGmtGxU
z@!Gct+$b8X@0W6b2U9hcCqQvHh2RnLANE30fRh>?Qw0!UFG#!zLbJqlG-_neUX4pA
zx;*qnkuu1w^3A`Yt1%&1*93!8Tlx)WpdGU+Y<cD0ZugZrT*-2e^N?~6f9FAqzmi%t
zZHN$cRZ<u})sJ6y1%CzEhsOK3g%Gt@<WJo5O7Ls6+K^dZ%Srt}r(l$lG$25x%<{J2
zM^$OD3YSjccg6@24$_Y?!W8)KZ#<AL_RcPa(DI4s<z6!$Bfx>nTu}xk6TCQza854k
zFg)rTCnOF~L}0S_<C>ZitI)7iVh#}Ul|E;va{3lL5bYttE1?NbGSUeOaEM;4pwqYr
z0!Ya=2MJq>zHp<s@@jEI0t}TzzrlIId9NY>X}%mRL`}WyD9^P7VPizmIjEss9Iw=`
zuVlz8YQ~tf#|{K@g6gqsOR0K8S83SYuyNdsY51Nea!cwYi`@h+=7|rnFr%}0#(Q|s
zFjFZU<&slJ*=!wE4a!k1S-g>1IfBmeDw6LTYH(=(yAjs{=IOB9N3eTIy;EJ(aotPj
z<+TBN{7`RjQ1@DkZ))bEczO>}2Q<<y4W2_M5a->&;v!n?*_f=Fx)3SLo1A`?9=MNd
ze8pmyp*4<yxJN?`^yXSGnXiRN&~dIF3%4K(Dn%5<HFIv@Scl|DldlA}pd<0DeXmVT
zCxQ3~KKkWc*oW<M-IY#i*nX1Gt2Smt@e$hA;dzwKkW3SOIFN}G?U>N=5_;_GRdiVp
zDDI((zmks1I$YUh`jP)z-5o5P-a=sUtG6cTm?I1VY!0o6|H}m<gjBsZu=ryiZ*Fq0
z|LB)<OXVc4uR;YbHTRXj)~p&xj67^V=e^?=8hlG8te%cRJh(QSVQQ9Ds6J#k@%+y)
z38EtcOVYXmP_4z}u~g`DRH1Hp)~ivK5?Coe1^l`~)=O4yg*f}8S`oA5og2FD@ws%U
zzjv1k*wZ!5b@AQh`bn#!G>9;*$tV{bOBz_P`-NBcc}wO!?d2B#d*sJvo&`j2yM~~F
z;?@ghmhDCN-rr~|xqNd<wr`$}(c<BG|EW^=_~#d7%0N?g8pLha*<O#pXwRN!EO)7x
zr=PvSXwdsWuT`Xy9D`9)pp_C4j^NLv*FM_I_Ss08L#9n$-6^^sfN*9x@PD8r4(9(a
z^OuwP|Ey^Upy<Uctes693FyVF4V+CxOpNS|O`!Prpq!i?O$=<H+&5BGw4Aay{%_$Q
zaV}(KFkSw+tJxOWxowsVy#c=KLUb6AY=+?xknF|J2d_8>81~`Q$k{_qRPSSsQQ!@t
zpdESL(0G*8O#GmvCsGLXFbR^j;UFj!vh$@qQxEiRPkXV5i~JHa!!zX@Lq5CTG+!us
zURf=9IPu*QY9hyJQ^jmf{d>lG)?Hs5I=d@p7hhY{V-&ku(=6J@YuIHL*+y{l`Sr0F
z<NaPXf#pJRJg^#twh$P!P9VQTjyk%E>aqo*-Ibu%A*Vv_W>Q%BRAHTmW{>qtC1XWV
zwfN)&(DUFyVWEqHN#DW*UE2KLz}`~Y^2)|fBAKh&6`MI1#I9+j7bh|x@H<w{)eJf;
zByV!6G8X%r)yd{^>ijsy<KW#)2%${$Ws@X*!}-Z_x~4o=9Xr+d%o+p}3V>P?X6+V1
z-WkRq6wd^sZ-zEq+udD_%OVk_*nfx87c(P7OqF!d0pH<9Z&&e=2D#ze;N)=X(2&!X
z`Cbc(+t+Dt#E}-d)4QGha*(S0XF*T|Qjy(x%=)VU=SX83o@XB#QG2SeMxG21;yuCg
zUh=9~F<3ko=LEZr;^+q~!cAWDS%}&p34w+`K$j!&_}I9@o0u^}?BfDXs1Oe7BEd#8
z-!@rxm*CanLfHeO;`pY2>RT8dhD*!n2gl>al^pqA3S2sxlgM;sZEqnY_ik|K_NnoC
zv5t@!2^$y|#^$Gw7Uza7eINr;%*$3tpYkl~LJ`ya@lYKr!UU($7%rW&0THE*@gG7U
z8tF23<n$%Hf-KncUm%gwp(8qdYn^e9fu%c?4e5dk@^e!Ce9ig$kz_k}bWgc{Y&q{J
z(3Ek(u%TGhspRx2mwhWZ#NC>eze(C=^ve<288nRL0on6&>#0eBO;yFFA~OqL7D6Za
z9lz64>uP_RTe!RR>QwE2?S6A{*V-iUMBKX1M+zk`)$n71QZKt_R6W3Se>l%eU_H$b
z;rQC51^H^}{Rz61K`@vmg(KHyOp}R$F}z$?hnUrw?i4a%4mjkj`)BQtP%n?E<o4qf
z-ujmpXI}kluWbn$!kRN?5h4RpwzqV5Sqlbi+xV$Q$;T-6xtG#e;JtZ+K60ihe!15q
z8t^=Mumf{wO6iiCrITpAyiTZfOk7^#XGRTFOsU+CTy>0}yY-5FKA#<RFW>8$m%Ier
z+FhOwy{REfGfTa=7#j7by)W``Q^erecH(%A+qS@i@Jk<mflbiH{;Tr;<9)==_<uNh
z|C`EJ)v?QBK=GNayCW7KX}Kp~;G*2#sIXbE)M-tli9{f!r3!`-|0wu=%K)&#fw9R2
z_}p^3<$e9Y0=WSM+66018o1{a!@J8L)qx{}7U0lN78b!w-wLQ0$p^Oyb1?pHWc<5w
z_hRVI#nptHMtFI-o;H8yY7T8FLuXriy?*rYmYt}>6-q!P<gwYPplt|N{^G}m`Gu+f
z90>p>A|S}yul+oX0y3`8CA=qltpfxSLf*fKmO-Hihv~TN1>+0Tt7dU#Yeh#x;AV0B
z>D>bpNuiU92mmff^pEq4j-(A7nh{gt(kDA;(g0d9*8%q)gP}?Pc*feTROiVwtwb3K
z8Bc{KJE6{C7|#Uf>LF&aHwxB)Gevm3aKs{>=8^C||A4S6Bvh;Fx-NUj;$F>S&uF}6
z#m(HoloBZk2$}F%QkZ*1pMqC>=_yYDRjwZ|)doc@m3NZGK(S!$$@_^?CTHNv#8p{Y
z2;Qw*v$QiiTgsb5szdWo&(ZM?z@jlK+8jwd!o%mU0UxhC$Cehv3&Z=@d19Ep@oe3d
z!$ISl5f{J0F1Y$K<62$~X~<JzZKw^{#+Zd{N$eGJb6ki1YU8HmEtkuWI(!G|*7K6q
z1p8lEb@5_&)n?VrP>CP9G-jWQV)49rAc8tM?ZfD}l*{zWo``b!!8SDy^$(1xEJzwn
zsIk6OUNBS&%{GRfS*{SL4md6vi0?3Ziyt!hd1ugnJ%~(CBEUUw7p}3u#45*-hRh%r
zk5j?rT{2s9r&kD<(aVlT;-rPu_SlA31NO*(X+fE)s{$8wONfrD!n-gNHtX+N>S|hc
ztTeI<7v2E%$i0gHv`I57Xjj_GD-?XwJnEG;NqWk3SeBvvEpE%rLxb8T#kacR_Ow2X
z@0r_qUokPtc03aHh3k#&eDf^JAfvsIZ{gk7`~|xJ`UPjQ>iVza1?PVZMMk#&XDa^x
z55;LUoBxqi{@myd*d+y|A6Mzty9jKMY)7(PBi#<gg9aGRw=!rsl5*O7T)G`4v>thG
zZx{;nrK8-s^YC?h-As)IGCUYQS#x<h7LbgaqeZ1lpNjkysY+)m8E+<FPe3}{Jf9>-
z&o<$urBj?hhGWZKc+`{Km7T5oi(H^leC}RogqD*Ni;OCBJxeKB>etQN_xJq#7*lp;
zCa;gHM`&4ZS6g>;QEaNqjSsg6dtC`*UxbWB0dpQmtai+rx3j~f7+_s$&^(d!sAo|d
zxx%ROUxaz2cAPJ_tCU?7ISU#Ty+cEdsg8!)3NVv_aw%qpy&&l;RM~WT*}P#Fo62c2
zhRMDBv0UkTtL0=|sXryYHd(YkX&MQrp5b}DaV8aRQg4phGmReM9dI8xPbrmb94=Q?
zRjbuQ<40mqQu5#Wg|-)KmDwL57?83|vphW4t8yc`l7(Q&3_QM+iG^AI<v9sLrc4G=
z<wA>WC}(23%&+^C=WBN)W;EV(4K;T5v^M_bM-QvcGCSG3fY9E$wyN5#YRN?&pYKP;
zkb>n-N@*h~hR@s1tWMglhk-4h@zPdpJJpk+f?A%^u9}G(TesO?eXp+G%ZskK(u%On
z*xuwRmT9IR8;KE0V7+b~cWOA{>I!87oy{-G$OSSCkr$D25Zu;9jte~Y(vfA49?j)j
zhaFnjiA`@C`*S|K$&55CfBM^Q$sM>C5`+rrx2A(J?=jjH3d*V;suPFQ8-R!bH;D|C
zGv~D3E*mV_awzWkd73(VP^5y%Nvd#s1?^b}K;D3HuyEq2DPaveE&9e+(A?yDK2cU&
zj{|`i@WtlT=q+%~rSNq2``)Y6887Ir8;`oZZTHBXI;&9)vMu&k_B!f5<>Y?T9|Hf1
zCxTXNW@=jh8C>X3;@HjBYBY|V<x7{H+`X^@=nLqPc!NMAn$W-mn4Y2V5qI8!{TGOH
zSbB<b$>BNKJFZbd=^2D{JPP?b6#KECHO5n(LO_9erQ_%gZjNQ>emf$h00A;M@i#~N
zfp+Hrs3eba+&j{Jf1d(#1wbQk<2StR)_XFcBdwet#E^`dxmZe?zY?ZJHK1U^+<dNo
zR_w%D$fg6YPSohdzBi*_8hIw~B>URHZa|J-!qBMiqHHuUvS(7?OPdTo0|rK2-K@Y#
zL22d(29u!o;AbpKTk8KW#@?|>uyEPJEZeqi+f`k*ZQHhO+qP|+UAAp&`VL|y;+}{(
z`_uaecC5@?x$>z4yCNZ(?<2RC0bB)!MxNc9$XmUX=^5DqqMFNMD#m%FIKtCZ9}m(s
z(DqF$FvO3@C=T-v^(B4TciawP8K+%o4&dK#RCq5EM8mm1sC|jpJ16Ug=TYB>RpwLc
zwld(YkBO7b5~vce!ktfxLE?f@Cy^^)k!gY7SnBoS&Hf_(kjDg?Mn#QxLe5nLj0<|!
zg7A5R!O}H2=87#iPvSiZ!UYdGRZ2>g7nU4>(X38}sI8Fa0|TF}60AOia;$_xf!g0W
zQsJ)4a+#-ipyL-*-I^HnHq!@IXktT{s!?-{fC&t#h%P#u0C2oxGJ-^uo|m?qL(aJ$
zc(Kh=!wuFH_}qCBr+wB*bkqW;Ahz=_t>ea%@WXBOb03Z|g?tl-6U$osB?yH1?bI2%
zfDeHKiWx&%{Hx;EWgF$1TFmBMEGup_j&t#?#)Zix50yL#Kp-WL#15-ktBZ5_P2fUk
zV}V5;ZiR`k30_x&m8Q;<>&cQ)BKQhyvW?5C-m!7bJso^DX6F%yiwKuKelnOF&T-^i
z?Su7n(EDtFci<xylvh!K^8?A40%xO(M>v!Vt-WD>Jj5HjxhFqn6HmOhBOcYNq0bqw
z0R0vJ*+Hu**wSCpjlCrhk~%||Vt|&uu-kg&ow^$KoGEO4^D*k>6weF)Zh2X@hWTuK
zGBirOaYHo(=#Av49o$#@0iiqJR4RDWg(ZIV)5nmZc+R}D%6?>)ws<ywIO-}XMxAta
z-%Xnd<B543m2?SalIj_KJ-wq~in+5uhqHy4QVCpAs!h{1vB(?$ZL5>AyD1f9c_M}X
zVGDX`b7}T^A|943L{BPAzd|j=bWE4L)eZyg4x8R#^jYXr(Wdu0>XPcZQ3pOxLat3v
zpt+takGe`1cvX1~aNW)X{k6LVUkairw-D!yM*9f<s3Tt&`ujul&X<91S*YAQ$`0nh
zYpna0^`5dd^#gyyiv#%32dG+>&-)(=I?n%4&@r)a{MVNIil)^6Snj=Zb@K^Lxq^3K
z3R+(E$eYKscCJQpjt-9b6q1HDB(frthZA+XU|8M&e~qq`E4$MOLGtm<SvvVyLT6Q(
zxx9Ry4{k2H)S@>sNL8J9yR>G5b1DDjI4xb<(tXBJ45GDnsbre)b$x$d-n-cOdwYA2
zBTaul-mhB{*0-~MyL}>f`?Q%ynbIG=MmE!K8*V%Dg0B63-@d0VcW-{)x2Y*hiT<6)
z#F!tKN*<i)C=?vSBu<7hU&MUeXvk);FmuyN*Yc@$?bq6<)?yD^!RlZpGv%yb9c(ZQ
z_zPudA}t$d&Wt)nLChItae)TPHbwAu-^)dwDgntOGHAT2gH!{WkT_;MV(Xx)JS$Oa
znD|oPLl7}URMOyQn}VZC`6#mOqhd`5xw>b?QqU!s@mNU%XA{ETZ@5yqq1VLI<uXY#
zU$TKDQeVBs%^AzgjI34=d35(j?cMpdqN=qfD@QMPV$v`y#v^1GNK}n`PUoW9svaFO
zYUX;((5YYW@K2ET?vMJ_rD1P{CI?mE$$D0Q>zZ}E(o#*84}DY32DY_uomw_hR-C7l
zy~+sZ#WgLw)k-_MqzBP1B0VJdebt!6r)SHJ?dq({*KMn=vcUyYMr72k>Rz_*;LE%#
z2NOKAzw7GC?1}h#{yKT(X|#PN_(}pJ?q7i*X)d_u)FHjc%AK0}j&eye-L`JfdqZ1F
z#K~FP0hM+W3>x2#R-Vfk{3r_7xQZtm#LmAXV}g8F=K0yyy~I{vDs8q}m@~SpURPCo
zsGFZME<p*A>cA(NCA;62u9j!pmvzOS9&XQ{EBbbTKeMI?WDuv0s<w67ee+N!rwH!-
z+NP~_rc+gUy{Z!k!IS|?asyX)Vvr<}o;%MIB$05lgv9|tF<C9Noj%``u{ZdQ@B`r}
z$z%Kojy=KUr#X2viWgb*`MoR2raZ@Yn=QBWcaQL2)4xAvt!xZ(2fB{qoJYinaPPzG
zIrLB*Q5M?clnV&@#7qGTDMNN(VZ`8O>BC1`2o*??dW36ukQ2z9XyxkC!%8|BDIyxE
zb5LEEE%>s}9o8ls3N(x&4!WFRs^Y4_0WwbuRt$N+$ur-?t=`HAKJ~~@@IMT$qH+VQ
zj9iQpU){Mv0$j;VrT2g)%KH|oj=4Y((#&2P0(95+40=p=LFvP%BstqPRU8^Q+c<0c
zJm`!?@1X0^4u;A?D2d}FV??D4X$eYjh%o&Y)3)J+W+?m9NR~*GCQbNBbf$N;XO;0c
z+J4BF@A3vOY_ZOljR<QCZzKr)U@3=mkXC|PtTWyKT9cz@S{dLsXo_Mgc%Ad>c#}wt
zuMhR>JV5TH7_+i@7*fr%MhfI?Ch{?Y81h1+BHyKJ^=sbC?X;~uj+X~~&S=*f_$64{
zyTe40(4LU?Xa&fHc<;EKx>b1LK#H5J1#Y^q=5sGwD2zCKoaP7h0Y+ft-Mam7UfrdQ
zzyxWqY_F@FU&F3ycGP`MOkRvYvd6>_-a;S&s<0eQuttCsTYGaa40ct!@B<0{VDo<~
zp_~wjb69c~>~$a2@0HBF)oE~GYK^6v-(aF3i6hv+!WP`&j%ai;HB?R9r0FTc<A+qP
z((N!4;ST|o7uSI9yqW=kDvZ$fq8Y{s<xpB6%&#IY<-xM0;5_jRYPGXr4oL<<j6T3N
zV?huao&4xy0nWN0be8~imEvYUr$+CM3B^uV`sDy(ofM{S(M7EQ_8EQEkiYQetD(-c
zU&7UJdoXyaFCAlMTow$AAkZTL;>nXpb^ugvFL_1O#2EVa(2$B{8GexizS*Oz2w(es
z!|jmOpuxt8+J5EwKcL?SK9FO3KkdeE)GE(#vgYoTgwQ{s^|?iK`M9{-lGD54mw}Iz
zyI;NefDFv}A+0KKX5+-w@#g@6mlpvfErf#DRpCy(nub-Iq@A6P^L>xBImG}M#oOoR
ztcF?lV3KGKyn7;nSrCOrSdKv%(HO|IDTuMqMcXfo$fh#4PY}}}9Y&4gGb~{^@UB9p
zqI0Y(CnA;$ECB%@qi2y|fYVtDIUF$0Kn3U$5k!d+p;G~s^9cvUNRB-y{UK4Nj1oQV
zX9}Zm0!k<kXfCEDW*?m=78bbcu*%{Y2=hmADu@5jt6-!MbjYS>cf(4T081JYtuy7h
zi{!bxC%elUvVtN?{=+DG?a#V4zM43#(a2wg8=sRDbQV$sU5mMS8-l88iDR*q^15jh
zE>+>J@RW8dTZDSm97QxgBoYh={JD3TGTKPNL86^QNgtCUnKq*-<yO_FLxT>csEx@Z
zjOYH&bf8$yX?U|@<`JJckXN<grwkC^@hFn%IFCw>DkkXUweI}zedO3H@hI0QO)SU}
zfBE{5Y@&DLl2GR<VYK$qv$N7aw9BFSWV=HHH%_7cwmG0xQ)CD&dY&8wjglMNo50tL
zaD8Jmo$F1GDpx{yL0@4chj*N|ENuRQL7dnr3qyzI#F+hE!)<-E<X=V`A=#B5H%8o*
zZ$7dQk5%T<DH+ccNiie0)d|9>f+&i}mRB$dlslY=AL~Dv97TBsWz41$pJpNxDrRet
zH%_-7o+A~uia>!};-@nw(w3oL873EUx#*<4@5FRm1_5K%3)COW{p&w_ao1=%bi8c%
z=kM!hVWx_CkgtFVLip=iL%DVe)~<oRzQ<Y$?p=?=AFRvV<xFsxScdjW*OeW9F;3Ym
zm8QC`=T_d*27_TJ=7wMSZ=llG*DNgPKdoX?bbuU2`phVKqDc5=_!oRW`LNM#fIFTI
z)}gE|8`n3^hEH%7-!`R0)Q(AHa3}s8=mRTU>Q-v6a%$ftnCmHFKHM-dsrG3^fw71x
z%!}UgY~RG|!1F6zWq%pSNMTOw0|1M{40oLLV&e@pq2e7G>LB8#@3!~F|K(HYp7?j&
zhX;0G>$y#cnhuNrF=@rmJtBK+DVyF)DSxr!3|bB!U)8`U#~4ykShf&t;N0od%MDQk
zi!8|CZ?s;8r#;kEt~QTg9r5BhIDGT3YI6OZxhoFq?Rg`-gB^*iKTbONHPSz>U|l<z
zK0v99j6iv7^F8#8?01q62G#)>pa@HUl@gD*MG5`B_e|jD$@L403zvm+<RPqGO<aCv
zA=nnZX%u>Hy{(yMS9};2-)=!a?E|^5G9lP^L^lswYi@EZzCs6*BaDWUbsC;K2l^W^
zi;@M;Ot;XkRI?E8#eOUH@-9l@e}3{1>vww6)j>PSBM0hqp6$Q<5N?F7w}Oci$Ha!C
z;feP9P#*dZGEK59sXE$uHSs}VUiq7?%EkV79mvM;AC7r&GBf;Ndk_D`U~fiE#_8Ym
zqUTa=Pkcrlf=?X5dSkY{HK{d8LNO#wb_4j25K|e-d3-YN?<Y-tzEp$aV&}OsI^e&p
z7x=X{cm!gt!SShe`17mvDEY)-GD(U=+i+zPC{z;zk;TAn_(A(0uaCzj^lZ{14Q=yD
zBRIaT;wk(zyQRw&Z$ct85$E1;6zJO2M-CXp&!%De^E$h_T?4oG4=8T6)wH@)>g<oD
ztkqlVX3WPsvNg1Sprh~#Sro9_snVNy4Gp6#6;#*1DoMW!3R^;hV!asdOUnK;oD!#y
z(0H~nZp{{&S6448dONCMVj&Y%)iVr*j~?N13Xo&XdxXt+I-ZBCCo%@T1KSHEXbK{w
zHeIA;d^Ng!4l6EPpA1bNhQY{`l!w{DRK;A9;=NfKUCM~ltHXt{hKpE@8Q%}e@q7-^
zj*zYi{c<D@3jrp_9b*HP`v56VN6;!lhigf7tQsx?5D&fu7|?vW(>kAbU%NPG4vBz_
z!Nq|V!SbG>>F}QqbXC&ZhwXN=#A(`#;`bJGA%z#eawWzrcu-fyRADH-oRL);f~4^L
zw-0^#A+$C*IEqH+6$u7QvmJ2K;DHW|3I)LH_Nk-LD*WZbY1iI$R94L9a;FomFR7+Z
zjvDX>F3P6}uj-d{S=*h}Q=U=Ow^Ki(UEqDXuWV$OuQi;dj@e8v<AdQ;_S7i=y^x7F
zc&PN!fz~>zCcpl`<>=-u{LpU7ZwAHn44qROOj>HPVFu}Dtgc&-{SR9P?4X*w&e&C#
zlQum;4^Py$cG5ldXEu&4cz0C(Fk(m+!<IMpLxg0ESVLMUC+wpjro>PGcxr={G<5c^
zCpLeQZf&f?ps(!V3F$LP=@o6y>Tu+)sq#iB&ZKQS_79hPqIUm<MouZ3H+yYfT=v<9
z#j+BxYA+DjAJsBVTo@VVZA!emJ4RWJ$g{_0NIYOV_D?;SB?!yqKcF2$(U;tQI5K29
zKNoO}aMe}gE{}#a7`E)?p1@r=pWj%O?@@~4#6`ZyrT!$!f8uR$-dA+>&V*vc1`%B{
zUp?P^vyzf~-%;fYDs+LJ7UB&bdOrVrm|A7Wjx6F8vcg7Hj;<g-Y3~OL(+FZ4`mY?W
z@d1#xkTn-NoV6J=ZbhBstN(y*;3}LwQqUFbX~2W<PaPmNxVSa82HO3;Py=RPnjhih
z3UNt<`cbJk*=<btMGG^8J^o7UgIY&^e7>*x)yis#Ka&p-5V8EHU!L=SyLgP8Z2xuZ
z7o$Gqc=<2cv0FE<n4#O`9T8{3owR9zuG75UqC?S|Xg->%Eg@_B(@Oxjkz75jlYs(>
z01yx){CLo@H$t!XcJO){M$hm2Scs4+p>Te%Um;&QKif39d0?RSrGn-9IyYoz7I9D)
zv!|K*x*I3V9BZ>1WAHs@r-_>DK`bLrgppRCgI*|)$}EJ^$u!L|?*Wj*@8X7CQPl`Y
z>Nc2C4$;>_{3A;b3pe^|TiK@6HF92IMZO_{<np#~oTns%AE>DHGAHaK-f5Y*&Z@(|
z4avcnR5Q8i>1VmNE$X1N)5kXCbyu-B-E{;i^2e&QS_r<D1|9D@f#)LlT&ro<F85=9
z{_2h!6GDn23nsWE&2u;f>vZX}VF2Ek%R4F{U{?qDMQ+C4ysOwF+cI3PXSeV0%TCt;
zL5lBe8j=*Vj)$w^8h*?bAFh%>^tu3$^B%P<?ir(HsQ$Kahoi8EUIOA8MoMmfHQ6To
zCj4;K)BJMG$LU?stqT>_sFp~}i_69SHrZeSInRTLr)eXfoSaGZkW?A!3SyN~`u!4_
zZn>GEuLF^vjNiGVrYoM%;cIR0qIO~scDKnMx?r;IEoGXIoCreEKg_(Z9Yy5WxS+$t
z9^3p^t7{h=QMb@$%s1U$R@0;7w(U8%1B^LZrB#{|bb+tF>L>&$m`F|juh3xtJUwF$
z$J(1B{y+}{;H{Uqucp8XLpbOguREo4xbi0NSVZ9^NXL>o|I&^5`XB8vnlrg#mgS*z
zyJcs>-s$-xBx)rJtMnug9NG?iA)YbPV8#J&l?i`=?ltj|WJSo4VW?OK-5*~r7UAJk
zghMnUh1*pG8`A7aru{0ExnR}n{dPOiz^px649uq&up>l_1XPMP|5{L6+_0B<*=wty
zEN*8oleu!V9KaxE_t`;eoX70OgH>D0TqC+In9#HgxY`Ci8t3q+SdOACvfnJ*!8uyV
zVVOiGEq#g6qJmK<hDD5&O}Ot!i;B~*pi1l$Nm6waG($wHqp{Md#!-}!iOqRgr6tnK
zdv*Y_SdcNC+~<Dzheau}99dHSo~JSdAmG*Yo<n`>*k^gUZw9ce=t*lP72BkR;{vh*
z8+cHxYld`8K}2PXU{?&c87CPDB(+>_MZ6D@@vT1-63`7zRWY{5LM-v1c$(5Q%`>9a
z_kd&i%r+6{aFA#>J_OguhSV(oM!eSMUt#TOktVcC*#=+U7o9rU>yu@)G*5jP(Fhb|
zfH~~?oZ8s6{Kdj~AK$%xK@jBFqC!@kkPim40ON;MADM@XpmQg)_NDFQP9T{>fFZB~
z>G``dSOR6deu=4GX}!RZO|#rqmGG4$Mo@~X(OynP<sKd!G0J1~&(tac0J$QKLX^az
z17{VoK6J%?RJfhe<pn>>*P|$h24L{$GqBeX>)dRs>(nGou~?U!a-*hMCWzlB?vP75
zyerAN0V%=O`f4TCQ2VS)zL*$BrpFeF9zaxLe1JjKvatQ{t-QlY298$)Qi;&EK2Y8%
z(G?k41AM1@jg10>r$lGC{r(}{d<*LsbQm-mi{vj9z}*-azO~w04B{GN)p=n|@ey)y
z0fDK~>RGXSZT_XA=xF2jf_i`$Nwt!N!buD!rMi}*@0j%AGmf1X@$$TY1H?I*;glp_
zQb}3D5onL^znBy}_1GSK-aRWdK(x7^w-qA;wlPTkOJYpd%Q*Kx=887ngJ8AzUPMnI
zQm-G#BpFWyfANG1f*q{$0SUM=fid>VNArl2P5Oo$@I$m)Ew)+(@HYB5)3g+K<d#VW
zY2I*CVhUD6PQfcOKh`OgEPWmyz;way@^Z!P<=upSELDVHvbPE;fT*Scg|jWb2%4C%
zgH@z~74jlN6kX_aAaHU6V5jLS;F6&E10<yShUN4@zChYW{6#Ik3d;48(2WMD#(nf6
znBBc26()zMn=+@i{6XAxtW<PxiEr3YwS95xk`HI@p=Ao|w&B3P8M|W$5$|ZDt=^!+
z|GFhOg=$Ea5(IuqFeVn+AAP@aDSB+YG{1x+&s#-14xgo7<J}%r28Kgfzn1AV=`))d
z-up=epxNp+Wan56cogrgr$Gduq<S1%7lKT)U39zlRm=`5vhtV1-$*MzmLc#g47!}s
z6AH6JzBb5WE0GicyPuj;Wnz~>`BSOQ#9fEUQ7f3A<S!f{y|&3FAmEs1cEIe&b?dq7
z@%9##oiqN8Ti2plMCT>my?YiiJ)!TEp1`Kda+;jkMPy)%)Q}-Xidv6=uvrA<YR%g)
zdNT{`0>IL|d(dhg0i|Lc9P-%o{$^1P_z<$2m5Z=&1a9HE`(tBf$gb(>J_zZ~MI-ki
zxLwGMv)Gh5VL+1xm?X<McKraNj4y$5X9<o=-8Ko#zVCEhu(EaLjq?SZuDvCb*f0-2
z@^wwigDi<U%A|r=*0JA~Sim@KNA`t$$n2`HyJu-nk*9?@<$Jsrqex{NDh7hU6Y~!n
z<VqVj3E-SH%tv5q1vA6AKUepsqjF4$s6skM+phzXbg>lWJb#?&)O%xyT8(0l#C4=}
zDhGqCM5gnvVZmXSPm+b^R~87p7oZlqWD^tuiS5FQ_1yPR+AjcU`QQJ5R%B!RPlhEY
z(|>)xuSH$ze@gz|Q?)yB8EFuBAduMDF^O&AjfNcQGKn!^srqTkKc3=Lu^*RR{Xj&N
zQ?8z8kb6MtSD#w(WS7BY_`BVGzK=J;bV$O{2UH!sQo)kOwGcBVjy;mSr5&1IF5j13
z@X?tQbhEGvg8=Y-=T~V<yLEH)=$L6ND#i&Cf0iEkefxO&Iy>L4mr{)GLqoZDeHg3<
zsiL}>bP;S&9J|+&FDog#a389x23$?Hi)fewC>gQd7K2HJkVe#rDhRM%wC(&CC>iu=
zY+AFlHut}3Yc5|FYVR&x!w&8392&dnyBr~3K}RygS++`pyJAAB()WWMnER?cHF!*6
z5-#?pC)uv`y}CQC?W@T)@FDcr1aGKv&p$1<{n_l@1|g0YFsLSAiIIYz&|$BMkcrOD
z$@*7ZSD578;9vgu3`l_T>UUVP!?%1ph+IbCNTZ5SaPprE2bZiOYjlU&{>;;+%U&<m
zR+@e%N^3(8K5!mhm>6(etz^#4Sx@5bcw-kgpkh#pD|H5x5-25XG>RH-{;^szG`tx^
zm^DTy#=P3#re$aJ-VTGq1G$(4a#3R4`K<Rxgn-`rF2!oCx?A7Dt(F@C=F7ScEZOl5
zu}?)^E<WX#Jk7#8&K@sEVbq7CFEWE!e>T75>ayyJnS&%yo-@X$W1GS<34%|CBuiJ`
zV`P$gVLU(OE0!2#*$*}9Y`QQ`pp;a?BhT1NMA00vGsWDtBkHWj^iDe5F;yyqkOYCw
z{Q;>su`U(3aSU$~_~_`KqXdc1ZmQ8m3_8w4QJjN+?*v{|6}G$VVt@yYh5W&xE)u7M
zQx^pF_yaV{fkS;VCfV(L<3m2e(yjc|3r`b4c{#2Hu|U26%2&c-+RqdI@(1Q5%Z7jN
ziY<7~NDVf~=C0@}u2TSW&>W9347Y^~RQT3_GI$Kf7Q?s#EEdC$20*>sxGrF(yBCU9
z0uHsm%8T{wf;P%|<s&+El6y3Xj8o}qs&toWhL%G!<bZv-3sk@s3ULll%B~IAmfivX
z<^dJ7cY|&R$!d0-LmShitEItFW{V6;uzW&TxmpziVnY~Al?KNwpx{)6_C^P=)VuA`
z=|!yk96-eeISP|3A1vW!>b3#w@5=5y0~~U6hG?ezo<$6j98mH?!^5}V2vpVlCOK-9
zy_Km|uXzuW=zNm^3REt#sSt%=>@!>>g%p0(h5_6$gsgMmYi^{utRGN6D>0|iW0Wsc
zKPPiA+WXg01T))m?5E{E7a7WE3yCfoN@N(rb=$K)G@LySuHm!+_XYFz1z_q8;9pib
z!vp7&c`Y2E>SRDAlh}8{GD<R|W6EN1v3RVjt|3yEx>NaA+S!}d8OMGO&`?5x0KOSz
z$kjs(Ln=A1%_p|S$$fF|QQ_aP>vaAych8p>TG{Mrm1|Z-b1C-JHkY+CX6Xq1iWOJP
z^H*2q?SURd{j-75`#W=l--Rdn8nC0&NBizyhNX5oh~*>#mzIVj3wC&4M#YDCYHb8U
zkdsIv?_U0QsD9Xrdy@g}#|;(PfuCobfZp6wPz4_RPtTmPU10F9hxE#k@Ct#)&}9rv
zhUo=32)Rh)d2b^S?|BJ<vniSMcOK!;0Eab_m&SM>r*Sf<qkMqqz=AJ5w9vQ<BQ5kj
zdICuK>>?`olm^(~<E5&mC}@&rVQ{{i*Q~HcOK-yt9~-4GFo#`^ozxzyH{S69=<&-O
z+9^JoE}F<fxx-W^L&F0*e7uuW|5cMN%Al(jP>kD;r1^bSEBg@%foD?}`Or4l{6Brx
z3qRtYqNF|z=it>$@)2f2+c<G~4Z(mXQ3rbnTf=v=4Ug0hEdApJeuk@#>A3R_<?4w9
z6`k{|JcGzY<-DxZQ(Ttlk4&!gCaK>}BJ|~he6{xQo>vhK0y-MQGG!Vsr;MU^Acmj1
z7R{qOaz8ppT5W9RCWnjVJ9le%PNu(+>@Kp#1qa9M>~l|!mQ;8rDrAnL`6$zj9<d*c
zDwbWNaN)AhAy&|P>%7Xl1VREQ`7bO`Q@$G%Pro%X5WFUsAAyR)6EfL4Dwi_aJ6E=;
z@bK*4^;Oq0$&P)lU?F9ELDHW{$NWE0g0X}X#Uz<v>F-<70BD4l4*<-&en0Lmfz1^u
zM@rN&Jce!5M#<wCq#iUP`~OZkn=4;a`8Ub6T`{Dn(}~*-9vb9D%W=O2*<K-e{$mPw
zcd+T#nf)MTcsrtlq_KhLblC|>M}B!03m9zfTUWO%ehXd1Uv$RT4iFffqfryZwLaI#
z7lR5nm{Z{4@o(WmZ%*0UP$W!WHr=kxl>0A;j5aQXt{S<|#V~->BumQiruq}-|5)Q4
z6t#^s2x@zIav<E+n|4?r<AoIs_M^lEQoltFvwHb#MP_9)(JMR8iwb8UU=y!q(CFQO
zCzFIQHYo|KvH9r0?a*tMzMAmn8*7eOmXKNkqHMlf$8`~!emL4V)L5#nG)9+|Tsoxf
zyUR0<V#|yEM$6N)%Hc?kl%6xHCY_y1#MFfH9W-<`*&H+-15mMFPligi2A|&tI{SW~
z=q+T$tp10#{vVNufxynt5{ichivIsO?as;YU$^x+>QYIEt%$w1b$hhgm8fX&KqQV$
zRoxZENcdL~PT@LKFrbo8^2LK=)3JU&SoD$c9E+ZE*4>@hF*kKx5lKFQNqs(VKlejr
zb0AY9M3m*?cvZm>!Dc}F(d2PmM|~9nD7@C~{pa-X!SPRBc|SkfufJdWr(wHzzxU@(
zqp4z?qUp!|>$%g8O{c&wjdY+kgD<c%3WOotPj$9sSz)lp95yQob>f5&O8Vi*KtKuc
zB>Nn~ph**lh!{M0PeY*!=doIqRnu+aWA)C?LNFSYt=FPr{O`vW^W1Jiu(qNdb5R#o
z&&zK%b<ghQo2-9F^|OPhTKM7h+QkDCXaX;YCP@BlZB3N(!X{Z}vO7J}v|uUBMr$+k
zJ;Q_40w%`U9!R7?H9G@SZ(i-3OUlfgnF=ynaSN#{`rm^zjXoYwf(AX%{N1mx<%cb@
zoI<UlS1tn;{X=%xbW*CCF>B?^b?)jW1v&~cX+Ub&<Ra&!#5V;wIUM3mzKS`N?s-nC
zZz-RFbkWK+me1K5E9i7U3xV8cD2*+bro&y?S+n^)LUzj=iX+=uJyW0<<FY!~)dW<Q
z*R?}?Nq^yP`6R|Xxzy>{v>GjLk1(GZAF?qhBDzltQ!CoUbLKwO7)WRU)5W9Xg3Dyr
zg(u%69|($+J!;0!HSm1aTerP@1w+|gG?)#E`)Ko|%NpBE599@RF+xxE$|;tBnX%e_
zO^!mGh^n^gv8qBu99X9>doWZ7zFo@+c^Zzc5ia(57k$dt8jyZC1p^dnj~G8{8zMfl
zWV?ZHD~eAkpPHb4P;;h5AsHe7ns%>Y&of~Bkq3_ab~r!@3}Dc*KCNqM0E?1Kn?{pE
zYZU&^-NoeN!hoA&Z8rdr7RFyu$HO@nga8L{Zi@rJC&zAQ2kmba|4jn>){T}tS)=~P
zk6Fw~Q(RYYb1uq@In=CFxb{`_>WM`#3%8FTBnE+OkcLS{zRM{+*VWO+PJL$R1jWA(
z!q$LA<3cJcdg8+3>miQz^t(?YFMI1%8#~kpUftLOE;x_V@3VR}QY0GfP(pJA#6hKV
z%b;IdJ>ZQ`Q9U%)``ZgSO;J{by$t3U+lgDVVOr>7e^>*-Z0FgWKD<<Bk^Syf==JwH
z&<^iScWsCQ2&~<nNcBjb3jx14X=TX*%oHF@L_wmS0=AH&;|?Vr_yQ_=pDNd#cNt9$
zcKZ@peHgyfwK}W5>j6ZY4rh4=;MAH4j1Yw}k(1DfA2c+jF>6a<<<>=?ufLQDX0$B{
zOa#KaFT2xM`?$KaA<HeM6TVX5-+nKlPqHsSZ^h6chz85NjtZ=&&LE<E1s`k6&2E5j
zYdEw!t|-RA@Yn}Ed}47F2KYE0oEic>&oDInTB<Ut_E^Fw78;}fUjRXdyryNj^1t{P
z@JZ+n%67MRYBhQdwRZ65)a*kb3e!S#clTd1vnp+eS7KD|vg;WHA4qtTX@^&V!adrV
z8=hOC&nan`rH8Ipt$t^{o1@lf&;Ug~xE!wEvag<}Vr91FV;_I)PR~&ZeO|Gc<1kYH
z6|OJuHd>TTCR(oFowX*b(qpdN(K;UTz#PW%(|3KjG3pF+2a<oMtQUIY=%K9cgD<@7
zBVF|S%HJsW!y)kixuXJjrShJ&l2z0%0I(-dD{Mz&M*cWiTS_VQyvvv|gyi^x^yc?S
zgHL<0T62|&oOENHSi?K)x9l~vVmLtX51&78J;L~Xqw(`SOZ^kjIS!YLt_zsd6fBjd
z6(4?XsbKrQsz$F$wJDKD7$RRVpcZjg?!LsQit|Y`vA^8nQng6uKE2xNUbM%a*hwKV
z5p9h3KK~5W4%!%rpJ+z(I*=AJA`d*p^w$^!jndBhCpc+L>uQ2sD}aAfK)<BGtk{-;
z;T7Ye4Zy_>2e_vmTbH`g4wRpdNel32DLr8#a8`6#=RT)d>5e=?s<Uhed(RvDra|V$
zBP$y!B;k&D29#x~yVPpBvHG#pTn+OuBe+)Ruv<mFJ{IW|5066(J?j{(^leLM5hBRM
zjJor-84I=T-lqI{pz*Wiu`HWpkj<ks#aOpu@c;svUR|hx`4J{x!qRRxOdweA(MD)a
zY%rhF&TThjI)uJXH4i;Q1%b`E-{BdOuz&ulgO?f%DbR7W&(H`33&TjrdE4tPXbIya
zs2(zhbjfu|Jz%<4uQ=1N15HQB8CWqZeTPlwcaA+v>EM0If@EC06iNE+yDv+G#DP;*
z+Y*_-W)w7&B+X*0O(+4t>zuD-N8B$iP_nHXT#LbG&aRjTn0~6*hjrGuEd&z{;y=!q
z^Y`^lp3p2!Ey*^tu(>}Xts(4ll<l_6bsdJ%i*BiT0<ox1C>v>%lz7>!hUp)<asT<~
zPyXE70EA&G>Jyz??R521UMH@dQ{_)AL)&pPbakg(>A`vOOFwu%eknaWr(Mejr>co4
zio8jHlhEcq{4kH->w1##3Gt8BuM_|B;7U7>NwOKynA*sspR_z__{s0b<N{3!#nraV
zSZmhL-@CW{UwgUzw*R2vOpO0Q!`a#YYdlK{hK}74+YN8uU>sbg0`2=JBW&X~%C&1(
z#t?3;7D!}G2I60%j#l!f^E!UXWR}+b8Pu~R2}vT##DlOReSDFDI5mFjx?aDuVL^$>
zXp;d~^kI#7GR%lXmbJeFw+4#ZzYX8#pKdHv<Xsrl5{PO2L)M*MexD9rzvrGqixku2
z&tx-&cR&32{M-C|d|w~?3Qg^i)zcyUF^4S%oRpI=oUz~mTbjPj2zFK=SSt4lm1vAD
zLv{T3gI*JP*ikB9EaR?yYf2R_PADUdSQ8u1h5EwKgOZVx-bg`bdLljx<^p#aYI?zH
zQEi{dyF;1#99<-&h`_Z90xbbaM8p<~Ae+PtemET+5_qmXJb^beRiaV$^qLEgh$hGR
zj<g7gw6^V^H(l8N{mZ98)o|_Y$lg59t1R%_f}ENHIoE>fbb}K)k$2TFHF*7Q4njj=
z3VMbWuDJ+k<mIg(r4HJmF>8-t0}A`-I;DiMc~N384L*Mayk@hx*HUsUv0T`89KFHX
z4YEh}1#?EC&KwF<edqK|a!r;nQiv7NkW-i?B}1{;11~oRubT^L4D>OF^FT!*;`?Dt
zD4LHpz`$aHfWFd4M&nXhWiW;<84rJ1{K@n9wtgJ@kN}Wd%rz1zWK7NY#Vkn1bGu2C
z77~pmI*6LMdD$C1wWz53JrsJAC38UG7Ka(r0q`La!k~!k1qxN?hjZBQtr->=8=QbT
z0o{iFg4X(=Bb44OtO!MsG&$Q#KtEb_v5()uK8W6E)eaFx5e+ZT8rb**<PPV-VaB3)
zrXHHW6I8m@E6%`<!iF1>ltM;Uy%AjDiL<Q?41n**31srdAhVdFH}3~2wJMI)-rI9W
zJ4H04l&cM|*5=)K1M%WaX)4dWZ9vA|yj672uo*r{Sw73Q1ajPa=@O2cu%Z%dBr#*0
z0V0xaPu41I4fUYxq<XQ|P9@q9YmrweT<-+_!~M}qrlarNPaYmz5s>z(7Gm>q1w@Cb
z8Ml^|o{O6CubdM+<>mtXn>j{*4)&~f>c$iXw!5@@s8$tvHl*#*jx~jW0I9D=WsHRe
zUGTyHUUI@e&=(ZBg)sA+A#MRJ5Yw3@upN7+rsd;+4zC_#9SXwXT@jx1o(Ha1gW+1s
z&mI~5UJ!&cX~-d5lSk!hmUrXjZcn0ULidgP^4veSJr@dSod@kz=f(PQ==Bu!%JQ~&
z-NWBHczy1lr)}o>ezOqsc+Ur;I+(s6UccOYXAiu5L!k@*QH8GD_vb|W2(){F;y2=D
zd560hA<X?~iPc9=!0?QPmD$TWT(7Z`%Tyr)+Z8O)wQw?~-B1vc8~T!sHNi{h@kVSl
zNW3n@%uLo1Ue%r`G|AHQP&Lv}jBg%ic;%OV0&sP}+L56LczeNm@HS`3i^^yG6djm0
zR6Lr{M%aLQyD#OoAJ;AnLo%mW5h7xX1<2mO3Qu_m1<Y&!^UfpOCwBgHf4!@xv8S1Z
zn6lZ<*7%#H(6ZLibIxbxz4}bc6K2}CtNZQYV!`T|7+`@BB>~e~5v$hsiwMd}b74zP
z|F~z}iU$lOd6I99!ZL8UVn@B*V<t)|kpa725t}SHUj*tsBU&C*YK5~iPcJEv&`{zF
z*?uqcjW#0-)f;*ep4zF?Fa$8x*1o>?T9<N~JT|Y?;ZQuSa6A`n0--SaSF;KhFRLVK
zv!njE7E-Ffr%YBem98XZp_74!EeUO=B;{8PYuX6f!lIXo)`%qV#3J=3@9!G(QViEp
z$?gk7$X{BCM?bWHrLBa-Xdtgq{g&Y}m(6H!myJsUvRRjgRRL3cO!-*LdXjCbeH>;1
zTj9(&<1HQt9Tj!Q2y=qT{VW=8idD7}pzYEaX?VDxyf7lq1wQcL#c*Sb$c-Y=N>;hi
zS<JhT=p{Y${R$_gs+$4_YF$)_hs@BiSs2?&;I%Emyc}&v`Gvu#U}au{N*YX0VUo*t
z*u_iYCuluq(KbyvCJhaQ^gDSyoM5OWCW7T51Zo79AXe|3>qI7=2l_$VW=I%<?HwUr
zu`WQknhC&yynrp+a+B4=SG`TQETDmN6Zt)&%fW0MMs5}VKClMqN_O$P$J@dS-o9AF
zrXlaXg>jZ+RXVYvEbI8MoL@k<3wM<3vy1B5Tx7Vc48g^5J*V$JfNmLldCKTb3n3)a
zsI9)bC6Ll(U|HP?wiZO|Las+Xe^3x6-f=&{L?_`;;5M?F^0aAuSZl|PowxA3*`;WC
z%c?zlCbqEpJz!s-Vw8`vAahrCadx^khETzX20@v2(GRyELdmJK<@I)DH2h#*HXM<q
zHJvvBK!iJ8OWu`rP`Bz-V3yIkszWm4_Fq7;e^HD7t)8q5tp6YNRQ9kpA)uEtv{Z7o
zfuffsU}RwUpDd1}lQRJa2iyPKA#?oqB;RdKh5t$F{Nq*<)Tg}={yRp9H$g}d-)hCK
z4;)wu+=}WWj!O%hx)D+Xu-&)o)`+aGfW%#y)gD9}o*27)-xcR3rdWV-DD7-Lez}eb
z73jo7Xq+7t5#TV|6WsVpCNY>HgEqLHp-pgFbWb-$@nPIOhvyb}A6zBA^VV-3wDu;B
zKo%AVf+9~1nMt-h8K~)g<+Pw#u8R@3Cd_;*nh;4S6cgSWe*X|mUQKVNlGzduGx8$K
zlALd3tUQSrSwuMr+7iTmWxPQ9f+WhwF%%p-h5*}SPKXJo5CRMZ>KS6FP6WllK`A5<
zrUV9pc%2-qS&on`k~9u-`$Zv23sR#jtuKI51fW_2`z$;?PXu%pT>IyV@2GfiJs9Y#
z=->)Pas`beqnN-GYXpsHMf4c>1LIl2SpTF2bVU?i;gOUjshJrv{IohWS~MGrh?`i~
z&`Q$)JpLZdAOx3J*l(c+Cpx?`NkJ+ZT3&+0iV$<c($HB%k~;e0kB5>zUfCH6IClz5
zBxyPd!J$KOs#ZU1NOcH^qTo9OT`4FAf~@@XP$(t0x%&J!j1xq9Cay5II*r$Uf`3*9
zZt`G43V0w?$8{3%{IZo>(}culP4G<rXi#29e&yJqwm$MTNqaO93r(&CNjzoo)>B|0
z1D=P{ngyVnkQaGYA=}a_%5{nV4k@56VRO=e=YT?T$N@A|N%YIB*vh6K|7s~$x^(GG
zxhG#v&T^9*oY6v0oo}sgh3*~@@#%NDr&1jDBOO>3nHd@M<Ee2D6GJ^Mq2Srsfp8F5
zmEIiLuQK@Or_JNaKs#raY-Q2KD|_GB{gKg<ckZvcO5Ui~<B7XrS<}*mF;}jdXNwzS
zo!Mj0QdCn?XXZ*-lP8~;%V)#8^N(<qzE7hYA1`kfyO|Mp-R5W>pBLMpp4StS<{;Q?
zH{%Mv-;W7L>vEoUc3z*}WctO(lQ$Rlhc{0xEWHSPTvlnW3=A!u-^XG%pJi;;22CcG
zN?mEv=+$l31?t&pT8pknyjI*XD#bwl*+HZ|>f3{)#*tF5HyAkLQ6rvs_(HEYP;FSf
zJCIA&(7Lm?GO*X#+e1pl&R;z&r}d<gZ@4(QOGGYKiwg<nRlus174ygQ5o_%^t=;vV
zj~YHsZePA{LrrGz^6-7IC0x8Yd348u6fIC@A1Q_|0N4C$fRiRJwqD43x@l@M45+*^
zJntRYkU4%d^tu6X^xFg2yY1QafjbHOUIvWbcIc~1$1AwENYM%JA^Ff^FoRzg;H;17
z;(_U6F~bv^9ozC>92nEQGM4l*XLh+yCWwy)^<rUo;nUd9t!Z|ck(5u8V_1nvsjtmo
z<JdOK!J1BOuVW?yjT7ur84lZ7J#R0EjJ~&n8q-{qB$+!U@n)M;YjJ%jMHx6vk~ryS
z&kt8i{Z{fRl%dGuveGQGdOf?R*l5$8bu!;%jCltwYHtkdQ{U79rsdfwhh9?drm0jx
z3#bV+Y2ft85u5;h+taTZ)84oI-;xGm-`S7cfu&fnqLc!j<>~@{@1gI|TPS@4aKq{A
z9a{#G?{UOI{cu;Jp>?*X!b{;~aA9Rv`mkP~jfNNuEMfRwapC=_Za!%jV-{&tt$45B
z+wQC<4@~x>lDp@8R4x8_y5g48^j554??yUAdqZsJl@f?R%!g3~oA}M5H5G`*RiZVu
z%AoMGjc4UcL~z#R4^{Tv+Han6y5W(;Vrml(uBq`uHf3^+i?Nq>V%d{cm)M&6zP!eW
zt*CEme+kW7RW7FSHcCx~w}!(_+@)z7n!XZWu#_$N<Rvht=;a(E6kL=ahcb+yhD&Ca
z3Hjgb!9n`65Eqv!gRCTp>D`<q8RA9|42UGS$W9X!8brH2<hIM)Ce&@R{$<rRf!adQ
z($<D8-WFOmYkpY*NrlB4_$h*-%0C%5MWRzzxEkKNGymMZ5^IA2PN%r1b$AzyjKAE`
zu`rFih}iSr;I9}#fR|noJRGsdxO;2z1d@52Cc-Dl?%O)4U0a7AGXux{WrOxS_5heS
z2#0AodwhOT*MhXS-#&~MTIuwy5d1iEbEDMkb~}VL*f5EUxFpKr=yAPs^_m~RrQ<#B
zjZo8I5qce|Z=}0)&9zCw^NX-=Ihn3Iy<&{>SO=a>Uf<&SYvn6)2+HUhS9h)W1lEib
z1qjwcUnc-lBAt@dEZhCvue0v8O}#azsMCf`zQ%=J<S1Gr6h~od!IIj=Ug=S2Il0`%
z97s1gF9s;#hud0Y-xshz8r`K*k1X@&;Zs&QvTCCIa+9ikPa5Q75eZy&F66eTYct-n
z^6@_OzzDZtfmf?xBel`K^JvvkUuWLifm$yW@FqAgCEjfsCGoew3T!|^XCD_$G<@>y
zSxUH=_L1$+*Ks`(os4Baz(SfP|7jM@^dD!@jQ{sq>HmKw_F6;Ze=O#IO1r03p(<pL
zK(H>2zZIQhVV2q4TAC!AIWdo9)07|{QVjR|NmrlmF3#udelYv=2f*Qq4$aFOTH$oG
zV0JSpFO0ly785onJ~)!v%>5eTSX>VALc)y0m9H}rVi8;X&o=w*F05DYY3|R%%~c`$
zUSg*izHc8#SNGTPch}TT*?wK!EbjsdDP_SRxNK-5`UNZb@`Eeb%9sBq2?S~_mE9pF
zQr<ByTKhW9EV&!+Xu6c<r(RCjv_W*RqT@^q5UybG9&Cr!7atr(iG4H;Z4pU~(-Bgg
zsfRX-L@JtAY(K<teHNEA_2?rok;WcOS^u7O2#{(PcwpF#OR5ogY(UNwlVk&XCFbSa
z)kU7WGTL|U-=HkfC5%`L(?XuOo5~1Kn#qR?XqCocmrKfcY>?^^<w*OBg7l^G=_G|}
z$g@Mya}4fdZ#Ip%*yd<cPo{xc6>GCSbCVtDt@rZi>%Sfe#q2FFSds+gLhb3AA@=k%
z1URG9Jzo~Ji{d}Y(D@`VSmelo6c>jRbbYhL#<h@?uT*iOdIydDbr#NL+`YUJj%gk&
zS^J2|QP@o$nm$HK!+NI<T#*iDt}#VL)?7OD>3=+oS@(k?$d-zw(Hc5{;#r@t2OYu2
z&?v08TF~e+v`lHG@aGbkg&ov=Hb0CO7BpmFzm)uC=N!<IyCO^NJA9^&*lI(wvgB8z
zA;|$us12%nkak|f(d%Cz`Ch=Fw!#WJV1ps1GXW1p7~tvC7J7C5P`dK*UMUC2mBU9<
zK<tN&o8y#eU|OM4E(bGZ-YFng7C@e5aVvf#ySHNSA4>4VEG;5wkR96u$>PY+6qqyB
zVFbu)8b}6SUDp1Cb2eE~CyNBW5h;sMMj#DYK(4e2_`Eu^Wo;Gs*_)))aeE}AHC5N=
zaaI0thv3aHz&23Q<MytGhyI?0&1?w0Mxmv0EsB7xp&$c_gP3StFKHmTfwM!FnS$Or
zfHV9H^_jXvggNHQHfU<04a;@t&QhVHNYC(NV)|2>{F<J%kzp8F0M^MC-6DPg!5q;Q
zAoGTU6dKcq3{j#YPhH<LOO|A<vDOZcHd5l;Y2$mi#qM(NI;UeclvGD<B?wv|%}xk0
zmA&vJ5(iCc;In66kG(r#boy}==VQ{08H04E3xx;8ueewGcl=MvzLst+DkY&X1$M%t
z^?L_fs%S4rn%Eyj4uBv3A#Apq1n8!8H#fgQh<JJqSKkZ4YJXSsN(|}F<j*d}v&qnc
z^A^&_b~G6ztKkmfa;dGN+7NXssp}#dhx*3uAj<KYi&ji>`I%^7S){kd!0=t0Dak|#
zaah-pOJQ%X7@XS?_e2T+sz9X!=&aaKu!#k(YY0tJef*(PI4jo(8H!FjfN)(elULyA
zbi#wReEmIQ7$OfDa=F5)i0$$QgOG=f1f0dvGWddH8fPR(V0KEE%Pz9up;wY##&@_*
zvbp|cCb`KsLV$h{;>AN7Ff;&lyB|Dk1yXNo-~LfAl4~b2!XI~=k2Cq#h+&rm01c}w
zacb(}?IZ^9*8%MzwgNt(a;^~ruvM<Em`U)4L~_WW_8p=&&*J3xz;S`*SM_QKelp}e
z;~eS^uVA@;NHgECIGOpHEvg7PL1q9;j5~%)OKU4fE0e@cMm~_K`2%k)zwgq^*p*sV
zv^V@>*;)bqX*7nia0zK#;M~^1IXaX@qfS)`DA<oD9ZTbKu_)1u{Ju*j!8%PfP|r#Y
z)O0&-rGGDkGgW=Pm0AdZEcKC^DbO7{ba&*rN3dM8EE5#_2P%xxb7cmY4_pkg4yz18
z7+^~qj8b%Uj7@(<oINjuH`If_XEZJBxCYEN+WLvDr8Hd3-)V3|92lg~J#;WeSipWx
zIHMf^VZvbzAAD4$92gWT<;Ig-<;MK>58C3jzlQybI}g_WrQ0S7n+9swSA6*PK*FYk
z(~8*Oa!QZ6r0hnl(HTU$BpL&}$mRz1_no9x?V#U?y#*fOCHnfh>FKaNBsK}<g@4KP
zz9Y{a9wiqeLTK&?al=#uwfS@BFYb#YEiT>a6xdpkB@~XwB$$5g{{-`#2yUnevL6W!
z!vo)Hh4Gv)2N}hxh;5@cpy9(+6~V!)MDz8)91tD_h)w0^Yk~3>IN~zwBhn{d1HEKe
z{fy{=Q{^oZm**`SrF<Xs#TUM}qQKhL&+Gu8K)ySBU8=2a_!&g(@=@YacC}by<<o@I
z7r$#oECI0IlP#?e3?7^$eSO5Goa=_bgNR~XqzNVkThbXEf^sBh)q#21XB_L<Q<hSn
zp)9bpCY4h869&x+e|AR8?ZG351)ocF&>DpH0?u(CO^I?Yz9m(7X8P0p$h02Qm&Cdf
zWAsSOZ2nA@(--?Ds?syBg2C9sZd^mqUOY1QYTqDUu^$Bbh-|R3k${JxH^{^f%o@bL
zg+&RmKX;J^JHff#Hkk#$WCx*6&H8zzfbudsuT4e1l`gvV_S33511Og_yMT=M$JuV_
z==|O`in*=-AOc9Vg*-jd3(^zE{_lpGjrl)qigEk{w*8;qe<9<VGIm@4!Xf|ea<|*|
z!1VngJR^k7X5%fcu*nDA%8zepR`XZgsBLzLCiv|UcBGK9TCXuGV~y07ruzdIR+RB-
zO0jfAR4@VWwVQXv>qYt}F+tiXCp(*5h+NWAs6<j0ZQ@*d;oFUzpVw=|oGpFCLo|9^
zAT;mkVGP|)4leHHRC^(V?9-{_{OYc)O)P9|tj^EZ=}P2EVJj(-=2()n>-vPK<9cD?
zw3dr!{nIIKpq0fS;velbl_E2NMJ9437yBRW)j5elB3LphMrQWa_gxz|Ow~l11PrD$
zgc&2q216+#(5G1<Xk9i)2LklZApX!GS`lJAOcLmXsHwE<Y3Abk7ar3WkZ61Sacuee
zK+j0=2V$^v&e^7yR7b&hm)o4exUlYF?a~~kdyiLy6CFeuS@$rzjJ;Qvrkh&9=MI`q
z6Xf?=?Srk7Z+`{;-z*<f(^i|QQd?pFm_mf+iNaN?Lgv)*WrWhEBV$mnT$-As;ac`%
z3Kro4jAkQzkqBw=!DID<^IrDjt}YjYNaUi%`^84s&ZDZP2g6P4&3rjs`a$=$=KbbV
z8S^1~ZB=$$y7J1C#e+g$7p>Y!s$FF@*`^-|e?=ss$$scZt}D&utd4E8srji*N-qrI
zU^c{|d7vIb+m^86;0e5sJ!z#<(_wR$sIwry&LmmDFfY(YtIGv*R;(J)K}Y-KfIi1e
zBm@q#P<!`-5}cb(N-SvD*jUXqi8>GkfFzXt%V*9D!|OphxWPE2XDWr8&Hyz>)q)LN
zFe${@L^rk@Uo!-q)a?gCz;E*{LQHZf-K#z+yxC36Zpolqd!to9kjF1s)Rxy!ozAE<
zf>2&4{n#xN*$89P=+rbsn0frvwf7vwUW+bD@9bE5fj<l2={fd)lL2N9oK6d_*MK*P
zz|{q#Ma6Dv|GGM|OGYH}NN%qEL;A%t7pliW^AYS{A-Urlx}yZ?vDt@8KN<sxQT%D2
z`1$Tf%nW=zzwC!R@hI+p$MTe8eX_ORr$F#Xe>ZHJsH=tdHRCg_Re~2JO(48??{KM@
z$Zm3WZJofi`S_uV3u0?d%ku7dnSlKo_bL2qefh@CQuFO(bqS>AJ|*DA+!7#YCuuG`
zxk}UOkTM3`7s8zpM3B}CyGX{tT_7q~A|u}hIcio1bSdQcPN-Ky?KNNK;P7Ur1JZUF
zOH1g1bN$YE#0}F#Ao>Xhn<LkDl;5))jg&CaupAN+J^p_%_6|*=G)%N^+jjSE@3w8*
zwr$(CZQHhO+qUi9efyp`gEM#sUqt<Y8dOwfu3YQM5visTl*XH@;Nit6>H;t{ac%Cg
zz^om-)cvUYsbuZ-4omPdtc0YljS+23pmyGQ{Lp97#}?zbw$Y~6?uyQ@LAayS%o?f|
zg6-$Vh!P*$OQe<^X<v4tP=)k7=cuY+=YB0iy;e#wxOkGSefv}-^1q|%#@>kPB@~vw
z?>W`OZL1Zqo<Q19sAZ90@sC&?B*7E#5PL3&3=u6_B*w-N%S~5q<1wwOuobrc+voW^
zx&Te}r`u<Y+4wGF`-1~s|3%x3ho{NpC#oWsCjf%Oxk+{wFIOV{wM(9Qrk#Kxc~C>E
zWK|9S@3-WTuUVu?vq)bm76DUqnh(NBleF6fBmbLrLEVJTk@DFGUuulZ#LNz8cYvau
z2)znwero7R=3H1<@2A`6!JqK+%6OeA>^LntOfOHUynYGn)|4m+;I&#vK|7bd(uH+P
zSDNX;_9nlD<y;JpKrQqfS2irT^UiH!<Q!^=K7CUrTSQVK!aq;w#sJp^z+Xm8ne2lX
zOnR7HuVbf0)b^zSHUNICbIDK!Mp)_#k9&)Ld?q0J5<JonMk8@<oOgC0NAMxxv?2MX
zoERUv{O5<6WP4}*ra59fYIx6jkRA_0pSn@_c`Have^xc<X9Kkm1N`_;Tz@5Q1*93{
zD$N<0^(AWjRD?Hs8Fu%?bi`>7dQM1|QI)M6ZA&q&8d?GE4US=|bkQ^&)$gzli6wdc
zO50Us<Dg;_uSo`cJUuf|UJRlL3S+=tUu~e+rpv8mF#&KLXKNj8kBd7)ISRK2vJ`i3
zeMealfj_gDHUsvlt9jXDK(d_#vVV2v)E{v>g<~FP?eZojh2xg2IJ0(MjK}iIo?0KE
z$iW+L?6<otLY_V4*VRbb(lnBUeaV+nUwnKV+Z_irnU-CGh50tEU#6`zoi_8YE-B2&
z##%Nj9E~t@0EutwNCG*y0$Wij8W0snv4%zm7WEYNDx|jF;KrEef4E~6Qn|c|7k+@_
zJ22?677CoQgnf@!%x!RVH-#r*$DQvWp&NfVFA=FWl1K!idUAnCMlBuBPTbJT7PB9y
zxLnXjTz!<5$pJz!tsn1Qu>PE#GI$J*Q2wi~+hs+O1mk^kUm1Y+ag2XhuOCiya}eY9
z0~gtF*#{J8{VZGE2QBzp_%ZUTlxGu@tN3`|eHKkyc@I^-ozuJ$DcoxZ7kJC@68PA*
zE^0`PJ-bwB*@SaTbm;r1A1&{Tc5vb3>c5Kef6Uel^sN80eS2C%({Zx}>Ho~u$LSFL
z$@rv;p-qWS3!z1$_k-#8<4zdiElkB?%0(A#>EEx7eKb-5E5i~8<~lGApjXF(2Cnr?
zvNlLLUk)#e$Lo))0jJ_K4@xED$Ib-{LuX0$)(MmD4$R`|2EWXjhw<6l9i1FaZuj?-
z>&nQ-awe~bi{4)CE{~t@>CV>MSzfL0kFRDGL6s238Y;1etSCFj8MbkmlKt1;v#67b
zS<fd}>Z``=w5PLo#_RKI2W>VRF7Nde_tlnmS=KYcEaZ_fXTs5j?HaDGLK{|49$nfl
z&`M#)Dp-k8qtnt2x22J~J$B2Lla0KyAZxS8@npko2;no-iz6hB%jt!cOMPEdh#BO0
zHWNo5SG>9vMoqH<#F$nx6$3nu<4MY4pCwv9H}mg!tX*CrV|H^8KSn^t!YBywTGEO$
zq+C0jm5_Cay1=ydSnWxAT3}BwjOkKM)9GUU>Z<rpF7gHNexnyw3)EkMpvG-mO*jgZ
zvX)sOJWESwt@JGVfqt}-$zjGxoLt^{DRtzGD?^qi#;P_dlO0x~5zelb3(f5$69Eus
z6?SZ-cHcoL1`i$UAkY!ZOXc(fgAQ}d(&{E46WXe6iJve`<cTb?t3ojVny{jCwsvcC
zz$JhBqw4;;;9l5tmiN2dvG`IXC+{nDh;p=|O7}5d_}JyO%?~>D!4W%{9IG+db$!^W
z85<L^IbzX*L^7{-GooJeuh(sunJBdY_s$Ab_{#9fL6<nO7+?^LF$y7k5#mD>-L&+F
zzB?$d%+zvc@Tsk6y90;pNd%-6_Og?R=7T;q87WvSfxT(xa~rMT2DOub*}<`;A<^rT
zeQImB>s2{EdHn3b2YEU`iSTaY8N4dEDcQ8>QOkST_}On7Z1|0~VAi*R1FHjIh_;~3
z)Nv{bZvRWUsdKNu&*@yZQ%gs}8nlp~b(I+Nwkw@_OTRTBV}Rh%hYG1O>B*LNp#0gQ
z`Jtd-!rR$**KC1#KO$6^YTycFf`%dfI(YBCLe7s2MS-Tz9zbjAggKHjX0P)f8ltW*
zzx@!t5#Q@U+~iMtZc>hWk0<GzzF53ZDDmDxX>7nDaiwp!VX0)RMz4qLraIL|pxJ#}
zyGf9!@ibLQ2Lr;Pc7pj!dhGyOzrcy!H679Nlnf41SRsIwLHtQ<tjuw0d4KH^j*Ob#
zV#u#qbw=)2+~qmcI2L2QJRDVQ$OZSV7n4jiO*ZXj#9$vQCDI9a$hQgDtcild?Jwb%
ziH5Wor3-96Vj5RI#cuBAr_7FYV@DJma1p3Poc*yUpAvARdt7D7VommD7Hf+>xgJo#
zxrX8p4$3{G_I$h#2M?(uZm{Mm|4vTUSLGr1&X1M6m%m%|a0qh%2kvt>N~VP(PxGQ2
zDj^6ragOF!7j4;9iCi%hj#c~Z_kMT?ec+%auZ9V_O1fl$*2F+Ar3$AhqeVCoOeqg4
zw2IpqpD>&IS9C;;=#m~Rw$*qD>))s4R4V+m?~{q%{!fr6-RQn3Hcw)w+&cZ>elEJK
z!#F|-yfXQb6r{SKB~)XonoQB4SQ#uaD?U0Z8ZP+>&|7)m>hc3}lD!kZJxG`Y_yqj8
zBOV0dzTChSP1}ufie2{1vDm4%vzS^THRHmNr52#KR7!AyZj6#syO9ipaQM2YYu?8h
zT%Dd%EQ25v74aLEaj-4&)92@KnZm~^BY+Z%ROJ`>?YC|YAdwb)GX1V>6(k(HUI4v5
z3YpB6c38ausVgtYr0e-}bbzALv>;Smgm}(=i&hRaq-&1zS@84l*T4ESJetekUVzt8
zcpJ#{lEc>?=u#*s=MLTK`wLKeTltoJ2~mwk*a4_4PD_<agN>BBk~4wg-GSxeszGiN
ztw)h~gQa!Li}OIXenkjwG*s)(X?K?OnGT&U%^SP>RzMl$YSFWv3TIFg%!<hmoh2I+
zhmB`C^rAgyMdj&hS0pJR6Sb13S4=6PnRI5SPaz^a7npZQL;|A$ZSdFKY8|iK(Z(Mq
zbm0>J;8=s|V?ki<!bFs>k2?~)pw_&xCnDjrjdUO_z-+j#Vbs4LJ?syIH66or$-$6t
zH|Vn47B0ehV)<eL?B0lFLF~oZQ`CnfvG04hcxI3is<y`WgsP8h>pkJ%hxH^mVv1~D
z7DGn5#O|~ll9;oim0mpe46@*S8q*Hk4aX93wEn4}wLr4$8*R!`r6-qq_WI)(Bm^{i
z{hM3gJEB4V1{}{jo8~uCK9qWj9n2_+^zcOk;0+OS;h=kHM@kKIOXNuD^p&Y#DNp9g
zkwx_#ge`8qT*i***~IPeGo$?q0*cUgX?{v<@;(>Jit?b*5Nyz;n=3DgnE;3foPie`
z5-yS&$xtR8{x~|^t4|^r%B~^z6DJaKlN;gq%ka-I{QcK4+#)Q94$6&UGlx6Ku29#0
zmIJp-lWb`b*cGbAfq~EV#uP7wx}<Ki6EicJ{9(71fE|GhHv<e-tR=)xG%SlrXLG3s
z*cAtQchoD~@CWRTvRE5FEoI-XyV^9{md5j7FA<<bPt=<TLhw2!ATI>}il0wCWgtBt
z)-)eA8&&}-$%>uIPPZp}Ep_iv0@O7{ua=j9&*B0JuWaA1*I>M&?C=T#fbXW)cQ(F`
z+qJZ^r=PWpf6&W92)O+EhJx@T0&S$C%(i7Xn)XUxr1>k5<jZB1`Zrq`GaGZmXa#mM
z;HRs<o$B1H4+^(kqNbt)|9V0TY9kt*y?2894pQmIbadMyLZ6?@eSpy~L}U28)yt^%
zk#X@<o`J<1M?Qw(mp0f-{O=Cv$YeUx&J#)EW?Wo@-x>2#gH<%4$}+&xvOxISrQ3#O
z`)%Zq&{smfDHMg48o-?Mf_4FdHTVa3USP}qKN=dQ|1iBVFtM`zuPVc~`Y(#C4e=L6
zwnwc*)emrwpPqF^?b(>Qw1FwBtib%@|5qMq5~cij@bEt6#}n)K{+C1sCNE2JwH63M
z?2Sm_=cP>;c%~mXug-6%%<V&A9N)M+T*|H)t_ZS;A&L@5t2*FPPPEZgm6z9>+>Jv6
zRhV4T2#RT|0Pp+d>UlYGa-DF2i^-~GE+mxp``^dIa3(8n_s4NGv3-B|>0)X*M!bfb
zMkJYuxJBH7@s5jh{Z&MVMZA@^{wD$DXP0n3NdojpI~W`ng8$MK0Qxksq~828oS2UF
zqX|g@9XJOCY~#JC=B*Yjq&^cI_z5E-$9LV(eW;7(t_P5`OX&93A#QmyH6%=ui_c3)
z+LLw9fKKZ)hgK4^u;ikpXu-~U;rgJ^Hx-FoxgLkaJ@z&WZN|EQL!M(RxH2CrAc=5u
z|GcpZ%_qYp6)UT<D^8rV<3HTd1`A=L7WV3i4TEsQ%Bo2-G!y`?JHxfaetOnrnp%z|
zXCkw#1~D^~rgUO`-IMR{m9?JphKTG&a$G8*0m3$x20!Aetv-INt+Qv?ZAEWRVDj6U
zOkt)5C<35j12a^p_o}NF=iYUoXeP+Wm<_Z|n%!&8Qngpmjk5C1E%{I|m3PgVLW<E^
zm`svQtF7!!0wlZ+#rxAdC4zw}D*F|RFtZKMI;;~a>CqDaaDqY-=&&Yi%qA6r7gDUG
zea#wo>3^9Ye{GcV@hUhk32e+Xt+HB+_6?N#gf-b4kxp6MlUdc(y(Y3;f#mzI5Do6?
zdY)G(7vj6*hZcrlPAsprgD*R%E;W^E7W6@7LcjCjYKp~F>Cd4?ZJ<jj&!`)Fh#h3i
z`ABW&VzD$jr_BepsiXWXI}<l;q7@~T(A!|<IqEI@)5FD2m9?p7&Q>+c{3eaU+^C}e
zRtFxotb@EtYk$8*cj{<qe|gZ0QSAfeiTA-K=(Ts&4KA3e<$ws#mmlVh2R;w;_9~Cq
ze<28fLB3_iqwpD3z=?ftTIsWef-w*9qkM^jFr-HTT<>jIBP8Yl2A$|i5`5S|J%k1P
z`QrK+K`y^d5zS3D6@f)s1alGh)XWN#7NM~QKojTcpwDLl9)0QmI#(oWt8)D;n7S{d
zbXfxasmScr>!s$IKdUiwRd3yr0r}$D<QTTsFWJ`~2}mcp-{1CcVc@wN(Qny3onh5S
z(B77#W7)6lowc8r=J5+7h@uPop<@RC?oYBGvl-uB5gmi5jI5YsWv3US`(3i)yIEx>
zji(J12^&X*B8y^Wu^VGEF|Hqm>CW!vKS09_y@K-qy2k+?jCz5FVF^lIa$Z!`2QSX^
zCF9{BgK={RzS<e^?il>c3yz0j63=8`NWv2nkW1iCWno8TN2VhM@UQ}@a=U`BAnvdR
z@wo2ZX)iOyudf{q#78FwhWEjY&|(M^_m7n)PyyD$L`<2BZGxcz=d)}gT3^b7pFfLv
zl&e)4Bz=>nibKls1-Ykh8dVL~%H!I#QFdBnFimsE2fR=-PSn|&-BygTC`XG|5--nl
z#6sI}W^D@pm?1zf*<<q%69Gslyj78F2xxS8<7ZQrO?z$ZqO41&&)zo<*4=oJVZhs7
zUdV30L2R9XXHrK$E2}20Ij5TkH~bI)F{9gc{{-?)!!q_A{BX+M_Hgn&q@6EegO7np
zo1vgVwyz-f%<PxWD}2}JJ7|bdcQr+dHH49pJ<>9_V=Xbd#;z!qwoO_SvcEZ^IFb-8
z7tX>1F6l4i*Px+@=7-hPcMs1|kzsEFG>rJ2?@zgC4V!z?;6|5<P4rxyB=vt{`qDY~
zaWM~Ym2HIp5q<>3L7)3DahE(s>Jc<IC_*FBn_G%iZq2@uO{MZl(4~XK$EYJo2i_uh
zB*j86gyM!)Y?Lz7^;CElvr<_%8Y`}KIEowN<*+)Vzx$88@KvV|b7bWTF$SMOH9Zn5
z(bwetDdHHgD@#m6L*d)>v0lDVLw~^m$WM!^DvPM8Lp9*>zmqk?dJZ^P#%nZ-ASBZn
z@@=+gPys!Zx2P9AC>b9+_V%;hoiGrW{WYGkhs<2o`pv}Q-TH`j;-$vYr8G1do=ty$
zOzUN0rHmn0O+#pCTCIAk@94WG*G5j9klA2P2bM1fP4290J%x|>m|hX@&kEe?eQWp9
zKE5DM<spd{@re(W(xxRAq4pwbS(|9Yc6+QL!qRd}0=&-<^Gm{Wta%NH;w8kdArS2i
z;`k-N)S8q-+@si7thQ!@qZf#Z+oYRXpE7R9gCIhBIp!j&5bZ)>2&t_skK8l9?43HS
zh2m=w?ajymNMWdNxsHA6`61gyDoD9{+~Ficd;$ebJaC(@sG?x9C<^yP7qs0GgU|P~
zQk_*@nwai?Kdq}`kR=EOA$0{%1X7t27-Olld+rA7gh;O%35Un_-GQoq&@{?Mg3;u=
z<?<F)iV;bX&L`zAC#UTeR>x<>1K9zn=2*4(0unjds$|givSQkoJYmY?rV2>_kBXIg
zW<&L{M%Y;DK3+4k8(YgppyX5c&07Yj?zSP_)2`9*vO9v*p_3!UYP~MiLIw^ufdJ8T
zw{<X-WC%&Ru3kOmA`_@pI0xAQPbnYliEKPfATdkT8;YS}_uV?L_}N-NuJn%8dK0&p
zU-;NkpGxY?275y>2Z_y6ycSi*(5^*qv%CO~aQh-h^zG#UoZ~~npe>mgYaTvSafqT%
zN%(jD)V*-!e+(v)z7(`W+I|4qjJN@3BxHEC*GdrODFCtJ>dgG><vQM$(?38caM7){
z`#J-748@V;$DTvJ3Eg_c@C9@KOP1Q6Zqw-7!!<O1l4kEQCC)V>j)``8-qJohIpUiV
z?Vxb8k3o^4`#Ii_Nut>^+xfx|&dP6q@oN73>|ThxbCF~eE}uvpTQ(q8(g!=rxx+m`
zghmDQl}X2*MfqYZy5rU7O9t~E<8Amf90I7Raf}BKA3UR{2a1^hnd+^$Fj=Im^j(+H
zW<-9ZX_)Tgd4z#fM<8>&N+u~$_XkzB#mgruy2<w;d>nTn6p*U^X9fGt$m_P=sGinu
zgXZ#XIU?JgX!$OmSux?e_;VG;B6*-2o!lEtFr?R&DLJdI3~a2;UL3~jz4zt{PYCbT
zxSoom3;LR<1+Hgwwxu`y`*i7pH=FCzBNy0Vb-FttaQfr!5Y+YEUphPM>K=jz9~*IP
z?KiWRgoqJFH=dT>NA)@$rK7C%csidr2rQ31v9VTUXrY-YrJ2rPC4=&qr(GU2EQc3+
z=AF#N5c01a^OjVSr|4!$0Pz*i!sR^gzuLe5*R}b7bBhejzo$n2$0bv@hV`#k1nHNb
zo==-SW*)ScOvH0?lcmFv*mP=El5Nh(+9b4|RVbuPWPQ22OQR#^mqOmz-VP55iO|Q>
zsdmK!ojzk@>hv-=yqki4%gaz<nsF9H+Ialie5EHq<||C6T&~+b29Y_m%Ry%QwtY8!
zot{=6m-gz?v<1AR`mUwU{vM2G*VXa;SZ?wCc)!?i@%}iUR9$@?=Bf4h_x^ElsrtAd
zdmMYqD}e~5nj{ww#PU}Y65Ej<XX6(}JNPG?9yrlBNSs^KM&HanS5<olZTeW?&S+=l
zC7sYEeag38$w8;ZwZ%0Tk4Bw^*5z7z9%Y*IESUaU``lzVW7fHq*0r_8G!yT7QhJ|r
z<)+$gOZB+I%K9=N9mw*p2go$goKieY4%*Y<4v=|9!;ZQq9f=7h1`(ne0xzSYkSSHp
zGz~gr@#F3Fp=$+MYpw0IARjuX{t*B|Vgz0C@MXX_$n;6BWnan+za&kgd;9ikD9dj1
z@_hn5hp%a;y!hjTgEC05P<XtQ5Z>h=tVb>Uq+GtavCu@gjHpP*WTrS>MnX{Rb;M>1
zg2X{7=&wh^9B09h6O`z0U6Z30f<+g1Yivm;=K845cJ;O7Ue}t?>2N7Y{Ace^nf@ro
zyo7yHTOA?#Zw6e*PNR_VaWg>P?`t0*rk#4jfjOajdQ<S-Z1>XJeoV9P)XyysVh6}p
z#+KiWsu+pLvGn2ZTbz3SR`am6KWsVp`U;_YGvB|+XMC$V6kM6Be^5Q9rmWkz+Vh(e
zZ1`PUN73`JB3ntFUv^F&a2|&>3I)yp$_xbyO9s)&rjH(hvOKe^mkrLIBT<Di>?p{X
z;^`5o-`_#YKtxg!0?%e}9Y36r09&QpN0q68drZUuCHDc5<_OqdT;mI-{8pV6)btC-
zwSDo%LM9A4kJp&kSX<yJ?g4`lPCbLCmk&KwI#EJIacT|oF}=AWgh;|h30$E?Tk81U
zS@2&GmIH_YGiMnzwN;EfN24Kbyi-ba{_@Bj1A>181%aa#M+H6%0d4o|<G$6#-dl8B
zl}|lpZ&V4zDU4xGI~P?h_8*gwMufS>GsVF|<wHW}n?C=FdX>N&gv}xB;IpPbSLF!w
zVJQc;YnSmEUa?{sx|$yA>-P3o+FXTc+y8m8ncG^^)@f*OWowD9+=s{n52ZV<6VYo7
zf~eaHZ`#r0G4=`z9yaXRRvqR%os(X)jyIJ8Ma@otk@_RPo((yS?VwR{z+Sen^zB##
ze4lj9PA|~cavqOe`NQ@S*zvjt{4Ra0xQ)qIhaZ2x5PQlHcMw$iq3jWv6mP_J>Lw#s
z5w1`u&Qmyy2*?s}2m<j)L-u6i#wuZ$*->!;KFBFcB$&Ghbg8m8l|1=w_uZz@c>id}
z>3OsG4DV~dvixu`8>z!@2N<%SGpIPLvI%`P`c}RO({2eA<3?tZ8R`2^IWA^~4<)qt
z+P5!sx~S}Q1D$zZ<PcJMZYjN3+|P9?f$F)*go}1i+HZ{XFxX5A1HD$wenzQjCDJ7N
z=w8vH+q+uhz}EoVWrHE6+<?ZT$;m?>Z`A+3*SDIAXJ;;6TW*7<A|@cmc6q9#)0Uq%
z{NH@71uo|#-T{H;=XIyv7%Og5slP2m9nt(4Du@vCXhaH$m$7=T27T3e)#*%0xgg;s
zI0gqea_*euStJH~(=Yc)zkeOA^Eoa$&<MsoMC^pHfxHCC{4y&Uwk<qo7rnBD=Pe~T
zUhGfrCHSVDA)PPj)H%d<-ayUzYP5L;V6i%7CNID^pqY!+>Y7{*el{DZve^NdBAjH4
z%J+^PW`)22*wM530Z>hpy{p>M^m)-SF;H;=2exfCns9n2{Kg~Y?^88dXYP)a?0*hY
z8aQkx^)-$}#<7M=Mb1dyOS5rPcX?S)L<cS_meJ19#eKsqz2<hqq+BgxUN2iV%~Fd$
z+fw7H*4RC%eM;f#d=s6!IM6|vb{a9ZNfXL#3w>%m<^wzh!zM8SYzv&xSmA1os{&E7
zTI^uwJP4WmNdDGvUNO<bENp4I{(%fZhy-*5)pNDxTJ3$Eyx&75&<{>(tb4l>BkDHZ
zy&k9$tY}l?H$p<Y?WpZXm`pn}xjojUd@gNyKcM5x={S(7Sd?R|wq2~BUnnR-h3}P|
zDNX~^OpVN-bIx}C62(tky5Kg0DKkblGasL2?J5$tRB{&s`s@Q)5%Q(VS!le0;ggUK
zj2MECLTe?v$wU}K#f)65fFr;xE~?m*(z@xTk>ppQ8hx59Grx6lzI@xvyHChBl8h86
zD8mn`Q0jy=Yfo>$@jK?ph&D)lvAnh<zNSVQBNU1-(97ZdWof()xm;y)lhn&4d{hm#
zm6<3YlKz4aRl9<VBA<E`$bE&tm*nFm{4&WzvT1?Dg%<`Lwrlol>tq(i`>-NU0AuYs
zj8@Fxv#@V9KU?GA?(`=IQOp>2gs<_-SmTi)vWqoCBS{VFh`S)pY{iZIX{w;#YN}yY
zn0iug4L-F|Z1d}A;f8<S=Vhrnn9%-=%N+MMnd@(cON5-)Quq3kN#1(+#nkiX<rQ?&
zRCKS0)!&sHnVxN4E=m|<WanWDbj_|gF{6V`<F7;`&spxKiGO(Zq+7A+p{kXL$qhnb
zI_ogSZN@<oGL$&dNn`!)9l?c+-Vy-fntanh`;Y*tk{@A?BTc;>XnxkH;|vK}k(EF+
z2y8g9_r1R%#s^5Gs)szuJ|D~Z7?0ao(n}~!c$+=^#7`3oHp}gJ+C3($FGV3>qd^vO
z4Gy#W$DauzaXq5pAiSEp3yf+}`j2ttxX}p|R#ohHw!-?Q6)=TfJwuf7hWM4xkNsW6
z%tb)}NTPl&ycv>}uH4aP)V%w%3WLQCUE_=ron*SB9=6}FQ4_w2wN1A1lxNM>Z0Go9
zG_mvHrN%C6+j^ry4ZTxYc9#1z2AbCuXN3)GTXg)Zq=!~UJN#J=SfY~NVmMRoYd!fj
zZUMiSUnelt;B<*43McnFb?Ocb?jt@O?>f~)w`yXypMnH^k5w!-kzS1$B*qg0B75$~
zvz(Eltm&Ca-!E<fKFo*hI(;xvC$jHrDV#`Nj=rZOqXY$Eo^cxw@u7&JVHHGN!mMCG
z6NtdeB81hEEeKnt)o1)%B?)3rbbsTA70BgdG|sQYFbXCf6CFfZR0(KWZu4~7>B4AW
z%OuEBCM39wOBym%4QFQfF9`#}g^v@4s5UWyQv;oUz$JV#WsA06>c8gcd7RWG1To&7
ztwO$BK?|NO>_YpL%FJFiv20(y+d*w>2@v)fbX%-fQ}7l4A_5CpaR50ya)sJ}_h(9Z
z8O*M}`o{Yk6GKmjO&A;BB|oek?>LGr=J<J_TPG{zY+Gg3ZI@G@SFDoK0SvC<Jf-%I
z#@#B3=MGKOXkD_7cjwv4<6&RG_naim|LQaU3wz1>zhf__|1b9HQ}s@z!jQlX5hU|G
z|C-I*Q%pfKbcTq14u#4-gb+2Mq5Ax(*VP}?I3UIFba`Aj53Ms&`^w-I*sLkS?CIfX
zb$GlHT|o$Q97T2hfC7m^qKHt5WZnR?I~tkYv-j<B9a^Lj5&aZ%o)8!(=Dc;+*S(RA
zGk4luz);xaQE2|{>X!A?6dm2S%g0$?IZLjW<jjC@Ax{#r(vpWP^D*g6Dt}_-2(<<~
zZIn3QO6`TC)+I!(CM-e9F>f(U9o-0w0vU&zrO##lw3%6YXwzlHX-g|ra*<$I$;9yD
zyW}d0c+4&F)!IEWF;javb+dx%w==QG@)bq47>mgIu!hvq-vP>TC`g!#@M2L-8aWch
zrj2DoXX7UxcSF1I8^+8?S(R)uRo6+)Zk6Gh=7^Uscw6G=jC70_<776?f*`R^9d8V!
zHVIl-s+3kpprcb8Bs12llR7l9)M$;yDGJTw(5Z^vbWHmb3<}wf<Fr+s*`ixm$#?w+
z8>e+$#a%zH)*{22)$pGDyL;<d2iAeG*@mmRieSK5btZe#5Z>yb2y!1Nn)ab$Lo=Of
zhsm{~VNGkD_m=Ef7;-#N$db(Gm`y9Qg|fe5B(XZR4xSQ&3ObVYuAW{ptDF3@8U6K3
z-vS8mirfX&hH!(YLDN2=ehC~oF>u6vHMw%b;yg}n#Rmwy)1C!8YJ|2=P?WLt-9$58
z@vD`hHab;3qhP`tM)qDPk?L3+NqW=;m7B@=`+>1|28|R?1`ZCo42UWLMgbN|q@KOr
zDIaNXHQq02mBB*V$@yXR6ze<^mDnzb%s)^}oKTJ|FYA*K3zzv1VtT?pgmz{@=sz4)
z{=6RVXIxp8?=9QYLrZ@uT$y{K>D1i(cXJl>*<K=0xN7_=-#H+VeWgOse;ZPZZ=YkF
z@fHObxhFV@LPY1Q<DG}>#7+2+LsQnW{{C<4r*lQ!UV(R0);&i0QC{adH&BO?Qv?>;
zJ@>k5oSbJRZ@a4<`X%etD>ZYtsc@Cn^>_{jgnE|iukkok1On%OmBs6yerhB*(v9MR
zh){XlR1+V$r`cw-L|@G&4I)3rkBvH56kD)7l#Ow!B=f;VzeO3B(B)=sH%*?m0p(b#
z6?wY_&h%#oc*CrTg21`pBed>YeDH7!f%VzM_}pTkFo?^cj(`4v2_s~zYU`e=OORDz
z0jQ(8c3z{^#d88fHeFdO3OZB4vnjWmdBcD(6hj&r=k4aygX^8+Crb=0DM9g<M`Thq
zfg_}-r=x}AMv-~004^OZZ}q##f-J`GI0Lg7!TrHC%5!+?Y{Birk`&1-Bo-PfZ1(AO
zW*g5Wh_!;AO2(NvUY72;jY~Q7vxI$NPs(spBU;8lQ@^j-w4U&ImQRCSErSujQB%hQ
zpLf*(1Xz^{6>x#(Z9xDLcEFvm!qpUW9C#k;0Hl-cEAE~VODC8mYeTWovrgAl5&Ug9
z=cxP}Cqf%NJck(C-iimm1b0m4w4eK$e1*mlroz)p6HDxum6RO~fzRu2S6(Snqj$=I
zpIWs?svaV|(7i*7$;bCHc+MF$1|gSqD`&ZM(qjLjvULd+V+RtDEU%-k<gam=Bhn7?
zwp&ZxGgxg0j<nUjlde4xzl@DuzYk6O<5CG~CpO!dBJ0+d$+TUNuU;p%u>hvJ=?(uL
z>`s+i5`pKX8V{53V&}|c0U)jeB*Sbyw%Gm)8R4#|mSd^9lN{3tmR=_VAgPAI^Zf)o
ziZExd2a$Udn1*z8@8EqGB{*12xD-+2XOHJX{6ysyUZ=vk2sE=tL-Cb3rG34r4196Z
zQbJZuZ^FJb?Bcj2300hhXSc=OOLft9n<)tK`VUxF&7<rL<d;y`V^0Tam%IT^4!gII
znz(vHZthmQ)>1+z*@ltUiOaNw<*hd9^E38BBGf+sV`Eo-)FvCo&-8lBQYh>f61f(u
zhbZcBHyE=cwd%wW!-|%-n^u7pWZ$)wtjz0b%cwE5-E;G(nRV_m{L#mWpNGh0gu=my
zw{y3_vg?_qNIOm4!pzP`_tv(tJdA@j(R$cA4a_D1AIl?y7~}2JUg;1UtmP84XvruA
z3+tQ^fSE+fQj~W+Y~=6^0fdKZF}#0~ly^n;Rvxn(6$t->&M4BjR*|b?1>lRVJgdO>
z%%C<faD`_Q9>uBDQTgb?i#*XyYk{hpA|@8EmrJRB)q;`cx!9rpN-72z_k^xH*}`JX
zDdcoyAwaN;xv?)9<2rXPwupZyA*+0Uhn4Zc+LM&8{DzrAOB<lRkCDB#;3TYRW6hy<
zmud11(WjK`%I)*a3JaxdY~4}!?;Wvj!gs;G{ro@bCi8y~;^~>#>Hk-Kra4XXH#Ljm
zxBHKBDb&T`4`Y$oc7%)IYKbNo(>M|@irHu0L=!dNFEe@ul<Z~Gbxe>rp-^4rlKWo9
z=XBujg`B0n{AKL??c)5sr-&dWQU0mhij$4~tV6J9nO*hy)Q6Gh5b<I7rTpjg@%hs%
zJn7<M>tXnzz~>c?3Qrg2=Rl55j?d7``|voiOLk7Kc6tvZl`M~rR*shTxj*e<Z_F<A
z4%b$3kOSV@(|S$iNmN-rsL~CS38MfGMjqPY5*5Bw(25f%<zi9Q2(k=tr-p<w<h3)s
zi9fX=d!jm(JzehMnpfhma)RmFpXya}B+=*?D-#M~#O>&IQ#}4n#21Q+h@%yzsj9i2
z6#HcKHo|d}2bOWPJVC?&5kD}Y4UtkB$h@VqcMUYeDq8)(xc+#oKYt{s*atll^wZH9
z)zdDpC5bD1X({#~M_W%j7<JRLjf$8-mMPS!Dxcu~{1R>Lrp028DRy%{)xZ!P4mjW@
z^jzcxpiw*yg5b*7rR!BDpeUi%=eZ)sqVk6#rzK33%{!7Tl!<4cmqgoVV9PE<+$W_q
z_HCB8HB#U^-&=-!l%#yJr;64Qn#zpwiD0@y=m7u1!OCG=^T$m7(J9bq<XQ=n-zL6l
zZ|K14)Loz}vG6IOkq#pgJ^PHZrvz|)X>&1k>=^{T|6&pBo^|XNRPmSQmhkThfO|ob
zXf(VGXeursf^hfLG$yfgno|zt?*>2%D}mdSXuSzJbY#MZ*_u`JM@?8EvX70GOwv3j
zW-S4Fb0COBDslia*`-Q!3s;<XTP_)sxC(1nK_ccZ-^V-CJ_w#dq5+}(dmh{_ifNa!
z!HSPjy~l7lu}Z&41GFq%t_^H~g**|glWx1Rmefj;kVHII>VlMi0TPbv9lb;p=RywM
zIIAEyNTrp!lfOHFzyY@%Bdw7WLL7t-=N5zpQVUcSKEV)+Ecf87z%TV+T^%|<?(}56
zq;(Y(<R3U(IbTY<l)>NY;L7_{{fFDKSUXSw{i58N`QuM7RCODb7m^<-<Oy;OQ$@hz
zdqx>0%G*nDekjW!3=d3%6f56JBvV9j2Pp!d*IPtaqOlf{<@KNJt|<Te0c#yZuuk_)
zS8=+7#yKPulK>*DH`Xr3AHDmA!^cB<AkxfO_(FXD+fJ8!gPd~JhOvSH4EAxJz}anz
z(r5A@k~S#`sxU*GR?h<mB#x;Jx==flxE|lzJUe~FV`dSEO~xAmzbw^`i}JING=I(*
z)6#dK8*PiM%CnQc>v+LEg4vvB2N2$=d@dK=xYb%u{x1UjhfNZR<$r7Fd*l9q30ABD
zqI!WW1C=yDS*8$O{4IWb%Me1{5p*^06}pJLf%2YYAxgys-OrDg`p<#;+5@U}7zHI*
zCw&$dCGU{d-Dk@woI$i1`f;0|U*r0-w5+<7gb}ZTbTwaH==?DJYaV-7m*hw8wWTlv
zapm7tJ4D6q?N@`#`gw<g7NXp8&*+NW({=+eIb_H9$~Zr>6&un^`9-DRF8N+9;;hFz
zxL=q#3f&-HYHTZsA5MU@gse_2PWE3-pVvjd|G`N2$)7y)oF|GA|7`N-t1C8!=4s#v
zcaPoPBcL6hZlgOxk3B!uBvZP9$iTmPsy!2zx!M)99x1X9&ejT4wH0KGs=l_nA%-6A
z>Z6)oqu#yRr<huAMtVWtv`H#Qm|+b#9vtg?yA`8Fy-yWkH%#U_D$U%%e!6#r6B1`q
zVB})yU;QndPETpY=Zd0ZPKrVkt^&9%43tgDhHL28i>cjtEd~oV6jLwsBUgnSx_JxU
z&FS>@tMABv+=4={PQTlgA4=j0ox*9QLhiPX(D+hZ7z`L@BT-4v^<WZdC`9w`K!aNo
zWh(pRlnmDAIL!*PH{}<dYWy!%Q{OE`{WaGiWk76?{zX+haM+gP)jLPj|0{@QSS(Hy
z^BAKFizw2lF-jgX8R(`eN=B-Yr0eTlY98VIuNv_mvlRo=ud(BQ)bOV@#D6Uus6D@V
z&uOdCfck_!BE}lGi5z>l@u6zL3KI4@Vy8&daH~iqTII*X$GUhBVfkxfW=Hb~fSrI)
zcw586BbNSgS#;Y}^x9p=6tW_Q4C2YmAmi{o%mFQ;N#_SkiiXJ_s-Kq+?tY892!&uG
z#lAtBkNeZ!?*^~edk20{8Ahje1WcLTf0wDTw>339Ux&A<A;X|A45C1g1408JGGYTD
z6DZ;!@nLq!RIYuV?el%9k54wf;txu-Pa3)jO&aU^@o38%)9BKQwG#!@x0Any{%YsS
z*Up9qY?kN1&wM=Y#uD<&iu)%zBE{aOD!&C<i)XBsAv)RJi{SN$<#dqjemylmW`@$+
z3O1$`@f@^E=k)-9R}~F*^^RrM>fBqVmD(+_&if)6!E@#(Q9d=oejCh2m&XTkRWem3
zm4Y@cBaBjChuF(A>*F>Y9D!DXsC^WXPKSxpZtbFG$+gj}0IJh4Nu<Y5R<t(7g*2!B
zw5Cb4AVQHg;1%YozoyNmDDz#wh3u38i<J(lHj)2c&Jz$4dNLLes7NuOopiES<EM3U
zQYh#Y+7c~bLC$zt5MX5fkeNc964GI)5tYr^`pCcri_v^edwMP4=5V2hF_^x#a_lKS
zIJvI19B;R1^?Orr$?3Z(GJT7>cg}l0X2RO$c6!jQ*g(l5#o6dprG?U0h0$(sN%R#O
z5h=7lWV!<EPu17~=;Pwg`m><vh>;*9Le7#Y5a9u7`asq};<KProv^Zvq>7HdNimae
z@xi^)8(jU%g5eRsIvnWdUhGkVDtZe$PD*uq=fc69v;uKUp(mw-Y$YAs&oZ5!iq;(y
zMTtKdIr)2CJx%J0kc7`Bs!n3iv&YHS=d#CHm+^vw1HphF`iWkjAI}gQ-$0e2**3b^
zRf7<ZFl$aVgpsLl{|kt88zENQ`>Q@Rt7c#HhUhSjg|;ox_Af>GsAalgI5D5c9uB3i
zaLK&%_yIu1Pe42uTuUPTCbFWvGLxp_f#++`x{Ol0*?3-7Csp4?6~53DI;ML2gQ(w=
zesU6ix%$zQg_3M@TBdc6XnV;AY8p&3fm68k*}9#g#SV^2_L8%^6)L?*?BI!eZZq7o
z+v3t27%tKH1!;20i1d*Bip4C%Rk)K@wIe<F?;uKgzF6UKf8v;xSb%&^@LxB2cs>z=
zAqCXNW&k5(cEF$qM}R4$l0Hn&-g-K}bQK1{qFF{=srx3LZukO4yFf7<1t~jKFn7W8
zMzCpzyZieOK_~H>(xsUDe+D|lK&B(=REJVW1SBX6YL%(~P`!I(bPXj4A&{JmBrbno
zO-`3WF=p|5hDEaF+{<Q*1`tCIf!T|hheyieX6smrK?dk!$z*_C4peY3+htW^6}lWZ
zOc5YOL|P$f9czYBFiv=OEPf9D8J=iG(X^O;iH^w07+z9U_*Eg`tpz1Oi#SSR2?x~5
zL{?j}sqBGp5C{)Z$%KGAmol(|lLNjmLLnb&@jV4F3}HAkVNMt{9Ib-(<3EgcII!2)
zUyFF2ZfJAWb%BF}O*x81hSA2#CTz5xlv_lOzB|4m&abKMd8uJSCj^*R8vGpTsMOOT
zLH_;b6Z0sWkOjwq8+}|ysn$KtwZ7Ng4ib0u&?zZTQTFg$;~#sb#9o5A6=BuRrUIwQ
zg%DE89l%t(nGh8-Xa+B546beA-J1I3gKVEfTs+@saa7*}apxg)P_H~EVj~u5ZpEpz
z-O6}5mzn*50?SE{0ad!j>@M>|z#qv>e=T4t1=vNLerD&FGa-LZ4@eTg#G9|0fiQ?8
z536?xM4Mcj>R~q^Fw1B3gB=o0ubk5<$2P0wb_2Y9_1Jw+YI#wudUzB@ZD8LIC=_Wu
zQ_kDocF*Q|2R8ClY{`?mnacv1efamPxQW-{bC}nKf}qQu8lWT-EH*Ox&)UJ_;!svs
z?AgZs{=9&Zu<;*GVx?glL*;fbrl}kpC7E`gE;QYv4k;`8&zAymDD3c>Q(wjeHsVwP
z&tzuV#g~novrjRUqcCL|Q07cH#*SAqm=w;g*RgF6lyR%Xc7liMYZ{{r0aZ??7=m3L
zN<G}_X$BxdwvvcJfg>y00sbgM3HzqAp8#j-U&37}&5&u&N9>mBUXyI+0P0dJwHb6^
zs0z6cs!A9O8j300a9**(5CreL%tY`5{O)i}l*fl8Av;YGhCif@+qx4^d8eHqQESw2
z=c=Y3$WjDkT!cuM<G2$dnTGmQMgTxv+0`cdsh6#oP8^qma`oFQ@Nkt|qaN`(&A~sY
zFNg~x_U_Qn+(m{-gK03B`|u$_?>mW-OlqtaT^B06-|Jm~aDCey_O4=FZ6+s!{*we<
zF439wbpfaAc9i(1Q|?vU<93_PiVkByFbkndJf4u#lJl;BJc4*&<CW#$gfpIo`R;6#
zF%Db|4WLQWfT3ljcH)kBGMjc^g;}!Lx&r!|0ypNk>K}FGtsEkJ*S7QOm5z%>arF$O
zF6>LYD6}9g==19Ee;46(CTGs-Dhueh;Yt%WEm$TvxmEu50?4F%{@Uygc0Q})(6lbe
zYi$zl3{FeomNjl$WsRLLO=>h2acf6vc8ZXQ7rZItnXh}uv{_&tFFQQuJ8~;O<eF;*
zbPhgbWS+<t^D$shl5AOqvt6F^Ck<D$Ip)KoCL<0NN3{V@X3$*Odp4mmKH5wx(8J3L
zW-D~M#_sg>a8GL@2;$nIX26wU1{titqppS0>vfXc9dTrTlBZy<E=rsQH#=;a>Z>qW
zta&}XOiB#n85eW(l1inoOD0<EK?FsYqTq*}+{hns-EJ3!Pi^XBLv!JbQ`fN)--J=g
zHq3khSo`%Zk?LHucW?!C=Rnq8^;LM$Mm&2y1c{<fphJre$n;SlOfP=Y>5$;kD-h%d
z-U3-6!IZb~#IxeejtZ&QVF7Vh?@jJncMG~6X?KIpGvm79JdZEQ8kgmJe@-?}+t*&`
zL5S@ShxHzN);=IMS4FeY0ZZ$tRgM23cwa7T4r-6t<aa>frTVz?2@p$^ec{|e#H2&I
zKNY{khL7{B-b?SI;N~~#K5F;=Dx-78pDkkAeBZEuqy5X^>BfGCEoPf?)vg#c84OXN
zUmq?dqh&p-1UuVNciiVuk@=l97QU=lkN(A-5U4{<CG_6RkMO_h)`Inr$mHqs$APax
z`dI;33Jg61!?Z<?bKT00*A|xKBF{gw3w|s*A!)aP>Q6lPx*+BM=<HbjD+0_;|39aV
zw$;WR{{IN@C8Zm*@=Bi?1kF9cF`LT-dFnM;$2t!m_EZ=#T)7>lqy+J94%pmYN&S^u
zxr76zUY*P9rO!49YzrW)56{Q_)!F{9V1oZ!W)JtQki*#G<fY7r0vnz0NTS50VpJJQ
z_7hcUeLZ|$zLqCPX=7Bi7VefWOG{&86Un$ZP0Zgep5C_CY2kB)4LmZ6f+9d@8bim2
zs@MzMxQW#VieXZ$8)12voIlz-b8>M0`3&u8VAb(K2w*}J848(Djk$`NXbmB)l>Q<n
z0x+>~tYUorX%U#BnGlNglhog#nIMKQ4q*!8)EY<(;%PbFWZY^e&Yr9^vR_-h3m;JL
zX~+*WQmNUC2xV~G*C;JstPs@HQM1%6uc{iyqiee#e%t;zgmm^TUL6dfMR=$L#39Zy
zV(0;O5+alCA)b>@Ko}n$_4kn^?nbv#>+6BsBV_h-OppNix>rG64j-)ToXZV+(qXlh
zS3Sc!Bh_dQwz(_YXm%-7u{<~f8n8tHFV-h-BXC{dDz9{&?;o#l`n;svE~^<))a+JN
zRz11~e9#YaxCI9QJrgqwYsbm@&@8RE<r9k!qs!gF_nR5{GxhY@(LR)%cNl$n{qZ$G
zO%o}3JrV(z%I$vFH%-pgu=9ci!6dVIcO}qy-<tGLave1WR%k^*JS|2ToT#9#Y5kd9
z!%;{R=&8ARxZm#bx59xM;D~C|14FHCMR^-_dFA?!GO}V(lkYbm3qg_w@I9mE=BRvB
z;q=A;u%kbQC~4BQ%H+9>a?!P9gA`?yl%}a+1Q{+A!XYJ3tIYZ~rgt0^O6nHePBFnZ
z5c~?8su6c8e<GNzC<CIY<5R0@2j(@}_@2X=dTH6*NpT;`QTw`)F@3;Z;|z6BaGrbP
zeeDmniyNIJF%TtmA|q5oOKR4y7P8j5!ZQ##1O_f3i@~xF#KUg)(R-OWF-$24m1~7L
z(D!fIQ~9N0@eFi3Pbftu2|!00(^bN0+8*nrta-mNz$}1EVR`vtWxcFL>f}nPy1UgJ
zZ$>1_Jpq%%6ltM7$Z9V&nqqTMKnk=Dq2bP*xr}+3q=X2FJ6pr`AAr`<NB=kuEA^+R
zchWF@cv+UNoll>5DNsETAy@kCJxFKY^3aU8O5l6?vWSX`tGz?~bv2>*Kz_Kwz`_{h
z{t`|cR~{=$f>qV`18@Wh5zYy3-V(2KusnG!WJ!j5erE!Zk_7!G9q9nHvnL~`Ls8|l
z5*xBQo+Nw1kX@A?eLOi*$20(<^nj$dA&2>y*ruI+=-(5<#-c|{)A#WjjE6+cQV!N0
z#ncfXle)hAmo&h*md#d=#Z)%!_WDQz&w8vib0Fs6kbuo$u?V}v5*z&G6EFb8K)KFJ
z+;ymLjk6#j{VRwTXJbXErxN^r-B4mDs`+TUL9xx;*~N57rg(o>m&2XnC3^refn?J7
zk4YI;RDiZt8f5)|sXL)RKlS0hoa&ljt=eO+ChHsVHb*G*)|VQjK~L$sPbdqGlKD1P
zi>mFfIb)l{Vf1Q%$fTJF@Ujd@qA-SCjpuYaDS&hbPU3RtS|zAP&X!hhhK(3a&D)(V
z6aBh>Fy(=Sy(2sM?tw_Hxmp*4af#2|yrcHbuz762528pPMj;WgEvtI2mE9Tsx}mjL
z#I8rcB@frwaO%3EEE@oa>0D7q@wCk+(J6PLA?oj|7>>G^n!-b%39V*TR<PImFDU@Q
z%m>_jRxGe%s4<U0C<@&^0Ou7?PV<-KKSsPXL%(Zt+q?vU?rK<+_S?gk29?)3@)PE|
z=i1IiQaX*%7nStzQp1V*s??}g!g&H$Luc|Wdz&hKc0Bos@T}y6XDthaZ@+VDd8D1H
zr^>T=1uEvQcxA78TuJ9MrcgQV?gdU!RdIJkS7;0>`OLN*byYn|CVu>0qPT$J>CoWr
z2zQmXtDX<61a798R7fatxZo7W(zvMiNt8O@`0~_ED!$Z)_1SZMzb0dP;zBc7c@s+r
zVq@%>Xp6&833b?6zM;#g%<loj6wOy86GSRX*LtSzkS>14in4W1eHjFwS2Eto+}H;0
zhM0iJkJs0{ZQ5PDfc$WI!9M-M*;D#s2ps63vnML!L?<GRIAp}|S>orUf?L7C7UYNp
zuRMGrms4_=ocvB)DOd{emLc{JD)##T!v+^2#PH#`lp=_)6xMRnN^sEhyAoA$JkfXb
z5Q3D>{S;P8V1B%3W)mWqf{spB@(y^n#bwdzXsI{XFLh=|HccvAte&k=pWd6d3)o9)
z!OHItzUgN47R-kwxWe6?ZZ{5*XO$_sBYhh@(DqIaYQ%vTKbzZ$#lZrdv19A|X3^Ng
zaMv5FdSaNs<?916hsf~J!eB~9K5l)iycua}tuXmaYdREu8ycD5=$G{%H8u3u?d#LP
zA!&s!0AsbXLm4;)A-OOMh|!}-m~v2jzzcA9{R6$H-Mw<$y$TH%#0anRi9o$hNMoDD
z;oDxqkV{<b&C^T^o`}A5xek4C9d5a4(5YAg=k|n}qnJv4ce%SkzykVT#r(gb!wl>U
z|Eo+>t^RxYsRMQAQ`L^#l?nqsAoiT%n#h_X@nvT%a)vDjj=CR>6s0^aiMn<1bMq4r
zyhuzPX)&`31c(j%^gbILAD0W1>uqv4*M_!+H;6(^9c!Gd%^OAu<&4gNYpMU5p)K?L
z=;8atb`=jq7QC}aO1(hg&Mx!&Yt^-PtLMjyJvC#N(1m>!F}M5Yls0#FcgyQ*bk~+D
z68dKLh$^~xrGMY{A4-<S$CqdM&aRGECjwgvIH6pGJwX7SG)AmZoto~NJF5AWED&Vz
z-#f<s!{ORq)iE7m9DSX2VVW>ArpNV69U}X<zKZw;mEdi|G~FO7G)VQAj}Hf2dmh86
z=qLz<e6S+Ne(Gy2v68wyvmZ$m-lB0pj+3ll1Zx`N$RyD*$!6-5qYrSw-ph&}l<_Iw
zbkWcIV8`|`>2<VPrhmbfIlK#PRU3v$*4fF^B!%RXkXZbZkrw7(<?4@rBfhz``aub~
zgOrQXU<PwRDce#Q(>2)039+WH^ozmQ_{2gH{>pHb=sh{gAt9!{<Xd+)l}!+<?+D!U
zrf{TLmK-br(-d{rHtn;1kXL$Sqo(+{3NuQ7H_cEL@*+hNm<-aJl%jGkC8W8ROV6h@
zN)T=&HmNRD9+8UjJ4uxK{`_$K8)8F=a@_8hZ6q6(rO0|}wZwMsY(PXQVZ7|<nyzG^
zyb@xt+^Ax<1X)D2CVQw2QQ1gjS{_+a|J^POrju4mppRgrpH{5}m>#~S<VZfZnOaf<
z#|zUs{N{(G2K{BUpzcgG*QgLgJPH0RAq_uY8JeRrr5(dss~bhP7v!UB&E_<6%;G{F
zQm$Yuf190WRgIk9h8@feb$2Nc5MKi4>pm3ZIg)$wBc=Y_b=pHIUsgD?PeAF+ONiu-
zng4%`y;GAYToa_*cJH>$-L`Gpwr$(CZQHhO?6z(D^ju8DiJ14|`xC3OR%Jf9{;#RT
zrNqBC(cicrZCq`RK>Z9oNY3GcNp;JdK&s9fX4?3Oayog03VQLinpXL1IZ^fTV;2%t
zUHfF@p{AvJr<Z|5S{-(GQ2IHk+5K&4T<>$dl8Sb`61d%YOaaFgSK4G_p!Dvjg%sDt
z>iiRfEBEUkAtmYtVWXYpA4zOEY7jRIbDRN$ya;j<@(MZ1(T_k*5In+r!*(W4&u#hK
z3I9GI_9+@Qf7^dB6@ecxDdRd=deBhQE9#mj(2L$YnV|c59}YA1#D~{t@3)?p;Mj}F
zjHBy7Be1A?>*%uG)>+P%bLwjg%zmP1EQ0av)yxZqgHRkVfY`yP=rb`aCUd^te0?2n
zT|!*R&}cOB`-?y2ukht2ggk8KuG=Yyo=5pkG;;PHBD~&K`;zhweO4bG?0D7YuPT_O
ztj0T?p6}JL4$DEkfN4zFj=u=U$<QulUZEx_b_GiT&s26qg+*iG4VO4{7qvH*kgRN!
z=^j~gYcZTfx+JQQx*njb>!B|MP$RQW;oed8B(;S;6CK8sJl)1Cm95vB31phgoHdCE
z2m+H!h(jYP^)YX?9V@&H;cQ!t*fpn?391D_9?c_J%0ot0@wZv3sRaPnrtKQCJ_QLC
z7d?`Ek^|NAE^PwZV`vS&y>{frj>$ic*+YL;LVY6Ul!!_>5vFN0ZK3?W!shiIi#<F|
zhE0+Ye7^a3|91HWO${E-!5(j*rdOR_MId_~jzORmlks>LUBuc|2~KSk8WxewCpOk-
z6$xGmBKuq;d_giZ?a1v5qn>Rp&%I;-RC*QGbO=0UeVAa90N{>EVw&baXKmSU8o5go
z1t6&OS~IFrea3`xXnHX#wS!W;tXmZ#`oSs^;lvnc#?L^a2OCHwvD(V{b6TXmo?h|$
z67J=NriNc{84>MYpL;p@n<_4M9#4!W@E#Xu<cguQ%+!!N1WKeI-tn-C?O0j(Am?JP
z?AhM37c&It&jIDiDL|D7a}4BGP6kruFVJTYMdz^c`(9T*aH{UTSHR&bsB_Po1s~|B
z6t#$BH8<ShPW9l0zj&8_2>h!y6Iw6bS}|Tsd_M@EX)R$o(U@9Gm8^P(9P-YtO@YFn
zlQSf{wUHQQ6A{3nOur$yAmrL4QYR)@thZGvks$4gA21}s9C$#ZIdT_+n?+$T3Fl$k
z1XRrevk{F{q1yu&02p~uG#!DI6k;eL;)0GfW#3{{o6`8`Y3^Hosid{fhEjG=^4@sK
z1_phf5bIsu2CvU(2w1>eiS(U<LS-r2WKa+U_HykElF2Y<dLY7OQLuA&KM@iqUo8%t
z+(2P=?WylOoLNB=QZiiBb6hc(pbJ*9blMJ@4#6>_QcJ=P<J=>?(5D{GB)YjNkr9Lg
zOv9C{Vdz*W%Kkpf3&&#<4m3U;^s#P0T)$M)I5K@G?@YP$T2$|58bwI{#Rd<koBLLB
z=_UzAXK<0W)z@Mpg$@xw+bd$tBUqdv=82{PRC6bl(gG49QlT1-Yy+H-AeIlPa+Gx4
zhsqSuk+Vk{dkw*idJgTiBf7%w`btHrv=yi#ZfzCOqc@#-h`3&-ZgU_7i`w0J>ol-R
zw0b^L3CltQ#98i%vIZ#P^0QJb^w0v{eha5=8FGgYI8(B3ut7l1n6I)y{Cgwcc_PEx
ztn<)6>$}hgID8(kl^O2i0YTI<QsU^kmc-3xU9+nA=REeRal*L@Sx7Ku|8$yzxEVA^
z`TSTNgu~h1<{9LWO&_vNTJNHUG~ZXdyRW)6bKSZ+qOc*UJs86UGn^HE-9F$->9tf{
zR^G{-N@EL)Q9YX}4e_yEAxgYd^$Ny#xOi6?Pl|wDXqY|XnzZGe(9263fm}8=ujWzf
zat!Le=p5<^4-10Jb*^?^uB7*aC`h}+jaeo8_qU%{QavcMKQ1idu(19{0Fh+Ado=v|
z3K8i;2!H>}GvUilK{LU|(ls`3RIS{Uw$iJ<pn3CQmdo0lb`D(ROg(DNRpe|3;1$4w
z3{w9E{@(ltE!uP2*`LYl&G}Bu(P?jJx8HV$#l-zcx1cgd(31yg<A-J-9_H2`Jeo_s
z((_C{&t7(s=)TLd>Uz7ZG=m%$*wGMD1|={4=n|G0g-GD1AoCGa|8b}6$J^KQj+fnq
zboM_)6f493k23y$^D1;~O#iEh>Q?{1d6oXyrL5rHWDqY$Jcm`{1Ivp}$K~{yQ!WgR
z0-PE7A_@ijO<mtkw|t27ltj`iBSb#@I@>jG7qt9wZ-GQtQ(xVzZjZ?_aTt1x*@R(j
z!Um!7D7o^4VGi{YjCms#RSaq0&$m!@n#JLxUSBuHOh@6x$;zAUPY3s>nJ*9Tr|e9v
z-X0Ivt^<?1SykI&bVh~Wa7rL{N-RU$NtDMpq6k?ihIATAxcNYlI4aiKKVLFa$A0Gy
z@^|g0@#)Hu=)ut+g$Y;f8=p%PD%lP2mqN1@n}y^`mMK;1?j1MGt1X1;-+DJ()2MS~
z>wLHO*p{<O6J6$)w*I28sN(8O)C*>d1UZ98F~Ewjka-F@<_imDmy;QasBob`bq0SE
z{cow6up$bcR`rI5b*@6DBOhc|q!b)`6RyAMDWaF0UD)Q!L-oF!DWyXC=>!_M;|-R0
z+rCt)-kqG(fF~IGT0@~%u=u}6v?lMfqxxf)L?{}mQlG@^gCk*q(CKZ+oD3~Ye&myw
zLc<`CU>RFH+44$81LWA)SgN0bVUlt5$k@PDh>?X>WyK}tHmFxbq_a2kLg)m>hbr`b
zz)vz=_(*7BsTxCv6@;J%S8QDekeoo)egz>TRn#Z*?EnU#RXjXL8UQIc8ho#zy+cF!
zn`$Zu5TzI!WyXQ>KmLb$Bb>l86m+%GF2uCcYwl*ZoEh8AAKE|t(}ITwHYeId0e}XZ
z2J@ZPGyzoUeDgk-x!AX+3Q(2nVwF-z)<ugHeG0_bQ@Ic)dgkW?Y~YcZe$|=bAM!${
zx7k8-ALrG8FNW|S@St^$G57_|;=2Z6@Lox1ghM(r7@)GB_PQ_2Ezg8NOUb!ZQX<bg
zDAQkeG3Iy^q9Xetc1rNkPUg{s@HUk_7GJeJ&W7M1mT>UOi>WV1Gv0NZ@E%)xGtn0f
znpYK?k#H`Ft1#d2#R*&93{$n$B`K+$VDT$?Q$1~XYdB1(ZmB9AYYH@v)jsQDykKKU
z37fLsd(P+28`hl{-DR2a?r(ilKv7Xvyr<_rRY2_<dl!1B*IeT_PBttI$*}|S94W|r
zGGu?vw|GD<whg%04l0m`a<uTSUcT{jwl=N@ylq;Z3Uhq{EWOZUw%cKXMwp1Q;6sr5
z)4{R$wAZG(Y`A_)|0Fa%?jG-7AMnu}9o#EjZOS(Q_Vfgh$EiOSKl93Yy?q|N!+=7F
zb2HS<$l`%kk~6JiQR6W9A`3oBYQ9woh0{w(+W-1k5Fm_Lt__84^wz=X)0Y8T^5dvE
zmjpHA(^kReGTyMasdNP2FRl<;tvm1jOAIiLBLm8EUt8j!Mp%x-9Y5V1u2=~qItF?u
zLgw<m;;GVTdt|Vxt(-}$Pc?%Z*mBp1Yv<@aAN@EtTyLvgA>}YaOWSN_K0W6*jT#hC
z*K%?|YjtJ^j7H!|uU|p^tKL^i5A$icVQ3&=xxNEU27rI=*N@;0_R2BsN6t?MAU9-8
z&q(;x0hd2{Fh(1&k&+9=RHo}wrUi6}ugtEin%U5QL1ht(Cbx&G%;;-CP!cxbaFf_c
zBO|T-q*;a!@SBbk7gzx6x_vvIunl(ttqw8K8$%LCN*DrAvHus@X2wS{4HRFIVK4~d
z?*{D|lH^h5MHmrhpI3r6)DUIe-P~n=kpIrU5P~!w3i6Sq{tI`B(_ej249Yn^V~uRo
zogIiq7e}{(PeM!|H)>|r+@3Rn-GN7NO#BPyazeU?2qh6Sb$(dd_|vvcE73X<r}Y2=
z6l5gfN%W2m%@UA~!38M=aP*b0l8GE`X-}0Bv~c?{Y=B%jgIysj5|qSGa{`Yv4Cg|5
zyelC&Qet6f9A-2~Qz~G01X&;e+=mk;126Q0^GTr;Nc2E&B4|Ft)Y1SnNdd&psdD0^
zUdlcXUP_Uo>>TJu8k9}DKAltm#wB)dce+^*kY_9#GVT*3NH9j2GkNp(ZHoOBKRAX&
z!Ai`3sWi#F*})m2p|}m88M)OWoR`12(sX$|k1KKs<J@2I90KC&lJ8ppXGz~TfWtxo
z2W%Ij=42Vcrje4b#6S#4C)Vsf5<AHc>`!u<l|L*<*sF`gVQzlVrvS-)r1&_7GV!GF
zT5H#xJrtWkUi=T2PF$8T&%Hh~8rnH+pDoS(N%Tk=Sd@ux({i45X?g^6e}dbDNnC3b
zCOhqW4aZ(5w!<(UPK0ie9}pY{x!?W8$<eCId-PH6R3MCn0hClPg>dYo)vbork}w#G
z02%mIU~y7@BfLWNyZonKWsl;lOau~WvZv6q!Gtqh!mc99w6j-9A5^EK)CQO*>))vf
zp5P>5*081I-s?#d<IK9Kd-)?S2rJWJp+f&!YbK?_3kPM9B?9I@OIx(7lV&@hmN_;N
z%|A{br>$;?UwnjF?8N5Ng4CxgTc;b)cUqo+!Hh|&QS@T2-zW$9Ze4ab0QBrkUE_CQ
zvT(Cr9X>CdvclWsyVn3mFl)>2q%{#!+^7uVK@*AE>6`wn-!cf^mVyKjdN`5hK>s!b
z=O>(BZ^FiCjYaNZ7>mXq>=Aq(a~d8+ie4oQe*k%#xC&4{+!9SjOKm^TQc0vzTF${`
z*?c}4CX+3pzEM2_W9RUb*Q!^Q7p)_e+kN0fcQjG#-YScDegYP+r(Y|{4@Z%2;{kuZ
zpDeT7%zR?y@U$1YIsSp~Gy}JpAcm6298A2Do9MH$9-iK!?{Tj6WJ_J7Nxwovq9*tm
z9A_TEe>23sDicP*ul|6OAeAHkm+1J9pBsACUu)+7YOqn0wEaE9^!$=o%Rmh@yd=0D
z=TK!bfXo-&<dN;3@jZ<-W14<Rto2*EI|QLgWYsI;NJV4f$ig#Tj9H_iB=gouLbWnJ
z8aY|r$Xp9TlqCg6(nHF`!Gz&)^A0%LBNXaYDZM$}Md5!`=u5;$n8)fOeZG%ITPGh}
zBNa&#)#8WCmBhWiS2tWw(9mjqf360?*u=AD8dEdC3S_#g6jKtUD3e}Ibe)e{DCCUV
z)zM<vtRgvfoQ|9$>qOSb#StMn62e5Z5J5eR2E@T4Q<x#Nw=vuaqivrvopsPMX=<m0
zRwRdKJB3kis%>7K7BB5wv=5cr+lNy_8#)JXp8SxHsTT@{WsM^fwQF0;D~<t!-~+e!
zjd5!uf=YxKqSbVF*m@OljrH>PR|Hx$!~_=aLL9!_rvB=LE`)Yc3E`ZCj+ikWgZcns
zvDmZ>p18E;sc%-jtZr|6Aze<^c7eA_Yq}UO`d)(W2cal7<zWDY{^ii2VT#1M$)Dpd
zCykO2((#(A^xj6<9(G5|0{A01T&{?n1x6+#e-}BP$``^r__8~qec@#F&_ca36JyfQ
z{*up(&sU2)ulMJDs}og#W2egAk5G@U4FuRG3&?X$nyxrZCOs%Fm&>(vzFzi5$9u2l
z;b?s4Z#R1uG4cRB1!&-aopx@Tdkt~bEFlF%s(BM-L1rRv53-S3{)o7q7dd(URC3wp
z!VnLeB8j*Zgt~T0KXi;wA2>}nmFk{F=Q-u(`f9WJesDhYu(aU)`rr=^VZ36*i8UB(
zlnYx|pc#N&7XO=D3y7TJULPpYxZai*?5$iGvQQ%=F4BRBYfLO^K#jwgSC$4B&WyfO
zM;G3DJxzXDOPswNdFNl#jp;hG3<P8;F^~YW=Pj3=_9Qc`T!grl9l;_;Sf9X?AKWu7
zL02Pl__NuOXVxf_iqY!i(g?B&(Q4(P=@JhHB1rk=>n&nbc?{k*3TphRR!s;Lue)ib
z5$EkIiB;Jk$|$<8`nKx~ZOf0qupcb|jBxu2NHZ5aJEz`JOzMM*Pd`BL5p!-39927$
zaDlZqn^a!)cBb$hm<fVMY%=s^BANj08-lV4A5q=WQLLIwi{3o@?(}uUN4cFFd~)zA
zx#K1m>cNi@5B^}Y;B)P?7>zclxeOgJIxwFEj~|N5?+RJ3hv~eQIQ<^mP>LRknoN(r
zihwvV=>{`5ldu<WLZl=hqKJ*bKrWMjz(g>+5*HXZ52O(jxKnTm7IGeJ3j84;17{N)
zTsr`CLZdS*r<bNHS)9rIO3jas8P_A;;eU9~MU5{obJJKULOF$_A-&;%%#25T>`+Eu
zFJVlF&GifOih}uN<tM6R0zm6YjYI?GV)?mq^&yeg_Szl^UBEJzrv*gA>vPZ^%lc5J
zcgoklEk#(OpSaghR<0%s^(9m7cP_r?jx3Wvm5oVa;@Y?XwyfKRYAU<t`@;7GI(ifP
zfu?B@FoOL##bL8}4<3<V7zl&p45ry=Q**?tYUU|3Ook?D<Vt4<bDLM_?OaV!#z5~o
zMyt?EXOnQ?bK3TrY(n|b{J(&AW3N<+fwr4RQG9ba@jhqr;sa^Y(VY2wGW|25*{{;a
zw8}6sxTm(NGwKqx0(&*VFQ}PBpBQkzoCGHe`E?&CJj1ycj1SouJwk#2>8OsMCW=Ah
z#&MQ_qH<a`YaPza{Kv<mLDM4pMHfq~+G_f3XJzJU4=MUx-<tE~&{V5M9jRe-dwkeE
z$W<D$5<X!V&<-t<c}$*kA!%n!?zLd&AvP`hB=Z(!C%~Z7+ZUkV54O$u>46}QZyh~S
z8X0qnBPritI4bdw4BQz;46T%qe_oqlIRJ)=ibNn;BaI?bY;3{6L73xaB48FR8Ts1_
zT5&-?vipsyb|w4M>3bVmlkG8#qo1<IA`H}obeVRq`%f*65$t3sI(K#po3CpXVDL^_
z|Nb<MOL15itRA19d}S%*sZep(rI;$fl|)YAsq0LiGp`zv#GGe0uUw^5n5>_sf?<D}
ze!N3OJgYxQlVm-#_s<6|Wbss{ZOn00xNZAvI_0GoGvNS3cCK56$dQR=-PKAx%pajG
zpqb@&l4t!gNI$6)Ej)RdX8l<EGA(^gc<4mHFW0-;PCDp(O<7XR`YOMO5-omw!O5gf
zcsAL|fuS>#L^F)|pm5&QQ>pnn{<$8BdS(hU^r~<8N?g6%frI_wK``*3K7yA`e@$6c
zn9gG}qS?HZML7pridv_rLfRu)uEzQ9TRTd9sYJQ&vA3@r<mG&J;iUXy639%%PY91}
zmW75)bsN^R;J)}5d#~5sXl(pHM&<7&{{Qq~Gtklh&!e*VH|KZoixd0J`H`uLp!?y!
zmSvGD9#l&gvt+6tF?E7#1X!sMFG$KiyZL(aH~}N$k297}2LoW%itzSamrUU@q<`5x
zv@TD7aFS$JAxNjEu3KcDF)#{~8|gPJlYTUi2@~0*mLx@X`u;rHKOOM;c4!}q(C~bI
zKJfZ{zV5Y9pbC9jKfS!T2FJ)UD2QqA5%P$&GqWz}W|I>*J{pyr$|yrCGp$?17U$+{
zY}IFX<d14gH>0;~wj8YW;41v<$sVz>=5D@XwWalLxiCiuwm)>~QbzC6QlEcwG4xGl
zxVaj+gR_z4k|eTd8e*I&M|1JUhTCe~>fWX(M!iu}dU?*j_^9#S-IHYHcKB^H-XyST
z?>u{M;oUt0j~~1E(k^5u7BZB&E~0ESt_fb@P*5b?p_L>dX>z$-)k(N6C@s})HJR73
zfT4&#e(Wir5~hLn*`Zaum?6s;M&fpMyJjJFmea5qY4NSwB#G}V#p~P}Dw?qk9l)jL
z(K9@0O<b5`m%_&@NUt{izKt2Bjoo^mw>9w3y(ps7SlW(a8GMJ(mvI@Ka{5YN3-2_>
zw?z^Uo+hU3UnNG>G$WEiqI#y39L5!MW-F+A<7v`yI=vE~JgusSpgjVFgr<H7mf5+s
z=0LeS78$f*?V4+(>v%D3*~GrQ5#6X&XHa0$yBlJZBCUmez;=2JH4g5GH|Pa$=5+%^
zN-qB5L)>%E>_JD**S~Y^(%!q0BW-gVH7QA&u~<_tAD`h8WrvZ<WNdL69IRa!(-LDk
zk4Pr94G>*&PR`+r@my%+I7t3`_l}>(fEWjqui!QF_-@oY+*8gc(HezX;+cJ9Cs}Rg
z-e7I`p??pMFXQD|5_>2h!>U;e>vmVEo0FUJa^-|Q?o8*0zJVat_+t9U)RZ%<`K;dN
zdMevVF>+yve&`O=q5guDGi0>{h=Rf`;!hu(w9a6L0@+p&^D4H~s>12Q;00~1-7qtM
zAh7B>o87W8u6xLebAYXon-yHoA~)B-T=`A4zBTJ1Sst^S7uL+FZu`OEzD5ED0;%gz
zd)S1ASm^knqHJ6^Gy(2AoFLt>O9yFr1*$W<B-Bfa-(|1e0nXk9mW>|3%oA0g-UH6A
z&Erj`KO94wODk#v!|II-+UQGfzo0sVVTDjQa;XeuihFWo1x*79kXH=3Mtg_#;eb0D
zbh#(~V)HjFHryPB=|<{h)OWZhCs<HtlwLh3U;*@j{GQ^4Kqw0b3x=W_G+_Ht2(1I7
zKW%JaOb~!mC90PkU~6#Qi?;0-q!n4384geD3uh*7`Fc@ZUgjS_pYt%?yfg|f(xJEu
zki=9Sx>F~lwp*dvi-Wa?HSOa$H5J_Gt#^#*cqR~d0B$Hs^z80b)F7l|>c3j-FTemA
z*{V}sn3Aj3cntZGyYc@{qk)I9DL8FMLoch#XL&!zwR2Tidv~U2{aDF@ibNEyJGXb(
zI%r^5R?{hj<dfFT8n32%zo`;=7!v__<{V#Z=rUMIl4c3$Wj9d>XRv_F?s$*|q_ggX
z>-LJ@zA9_*j^O#eo_^%Vvsz2y;}%x1Rb*q+w7qA>XQJmwObGto0iF~7q9iI!khiV?
zF>weFndd-49l)lSBswd}+Leld{Eo=|;)m2W_gMZqiTaE>t!5j$rT<xK+`y9O=oGB%
zdhq%sBL#6@0qW4kc!z%4h%O5R@nPi@Yt2>vLY{LaINUB+O&uLe#6;~;^Jo>gLuv!r
zUS7Cng{`?JUHOIS7ltg3lALS&OHv7>JAfZYJ$9TlS^qisaG2&ipxR68E>KeNsxED}
zbT7u&He;Vn+NwG+=jLs5_*V|ulCv&USN@YAj%c2RS0P5orqz;XSAg^bJfcSaq07^n
zMrg<I=*+4VnCNNBpcG;4Jz|lVr7QTH!5gbsB~TVR(UbSL@?sLJJy_jRIZjF%KfdG_
z7PNlM1A1a;C@;^%Rs~gN%cL<wZM+;3&Njn#3IZ~++G=|EsJrgjcpV>oPL~ElvSPR{
zdiBegJd2wU?bkh%IGxxiZ53Smyo5u`k&k;~PA!>Dd=ZuHv`F!Y84R03j;J5z{6j$m
zOz-@S-!X){X&bq_O=HMcfzA+6hA+>KzoXvbIo$IF=V+(}#&@;oml1*p+hKRxNF%&o
z<J+E&RJMB-f*eCSf`Jrll4>w@_*6HlTqHNq7w&emk#AZTy~j^}=*{!gpsT63H2wxm
zmgUeP7wKUofX=q9tCW_yew5zmiw)sjGU&=En1o7kcv-3~b}w1+FP`naVoTmc&H)(N
zBK8rn^;n6xvs_L<Z{EJua#U645Wf&Fg}z3CLz<F^P80Jr3wQy#mXZ-)tlAp`Rgh}a
zzgT?<jaP-@eG-0rP<TrWKOZ7hg3LI9GH3gh9hq$c!5+cxfcn>~8{_PAg&J)z!!(Zi
z(XHu3eolCFiHv`1a1Blg>FaTOyK*pE6^iWE<)=)nXBag8RZ9~JlWB|cZA=tZmB1~V
zt@wattl*_u7~GG)?TNUxyupX!-q8OY>ssE0P-dOjt%TXvK|Njv!;WIyDF(@24ec!f
zS3fn?`!n$tN*KOI#iI!%0KjgFh%el`T=#7h#1uY;^vxsk6L7F&LHH7HJoEAIc65)f
z*e?FNFVj}yCnyB<?AsS``{sz^zvKejf5-(UR`&mSv(}Jw-29ab7phb770G&C;wmzC
zDc!%pl@S$Xh2~6P(-6i%3Z6p(^5GvI*gSq9Wab2HRb$y9KcT`-K(8-n!_(uo{7l`N
z?H%pz53*s>zwO5!j;?DE-SP@Z6wAjpy>Ha>pYJ=azTPkK^IWeH4_QWp;26(^xj*+O
zx;H#tJv~Peq#0T}6w--sKkrWup|rX@UthMmQDKx3jy)e~jeJYhsm_y%2}l3LyE<pg
z@2`5C2ybS!7h#oGJ933PU84uru8*%B*Il2?d^MAaI??m7OG>d!q>BQKZog~%+qY9Y
z4KDR52%HvhrW8q(Ig>{3FF8VIxMx?XBqs_+bSiV-UU7AbAkW=^y(t218;z<ax4k!_
zzIHV<u@#>M>#Hljm9DoU1w$LTb1JsR)9U0jj-$wOIExJzDbqqGxz<sfwF?|$mIc>4
zt_PkTk5=ogT#GN5QZ#HeDq{>4TLLfWL!Rm)fTB7$_b?wj71wv?8ro|+ZG8%^SSzVW
z+;J^3kAoPGn!Xs2RTQt8f=_l2?~7<Ltx%qm{Zxl{J@k~HYd{^>PG)((oDMW*BwNl>
zKv944h0cOKgJkws*xB6C$O(?&s#k%Dnn>BJ{Fw)&ADpXZls$!H-EWwbHJJ?$e3}d!
znKeh7x>G~-7_r#f)cdB}=5{OGHGXRpHmmTaTXq%x9{EDqdAc9?T<KXg)81+C;d(Pi
zfnvh&QhIr)@-LE8P1{snT+d?|yyJ0R=4zXO;%*qA&-crA5i#u@4`Fi44g+JhQA_#~
zUbPu6!xdY;J3_OtE9n~~_G2s$<j7|r47Z6U3$9k<7)h^1aSic9k1N(JnY?k~vCekC
ze@?o_yxX}xJy|#e_yrqcLD~U<JFaS_+vKBo<yg0oa%^QyPVb%L?NkyPkb-9B;~nP-
zL9&vazeTCxaxPUNguZODUs&Z)2JKJVwq|4;o%)nU6j18sm}@IJJb(eZY%>$&L3}J;
zfX+`-{{p<us6dv!`N4RSA6-6FjDn88Xt50J$#x^BY{i9{K-A*J`wphErNQ^5{<2J^
zgY_-^0~Er*Akv^Vs>&VF+sF1eGl&z$Kh_V$fbSHgJ1b@Dt^t-LL9p{lkw^;$X#+uL
zq6tZ{KQo>N7!M1gMhw>gfi)@o1od{KflTyJ#%qk1Cs_1SCjuP7_(p(a9GfXNW`)i}
zA^*!UsK${s;)CF=^K6lVm<-zYu0cC6lpqUSnNNEm$j1)jeJUMyJElWbxW$MSO+Fxe
ze&m6#CN<-KQ-^*|JvrxZihP;iC4Sgukkgl+B7HaQY<t&iuVzzNr<hGZt_L{I2#gou
zLG^+4c^FifJ@}w%WeE66DTGRSj@H-t^9IYfk-;!99RqoE*-^jk*$M&>m*Ggpo=@H|
z>`B*%LxImX0?P&N3zLEfvBrgx=G^0kOhRH_V-Ua7`NWP!i~{}D5&7Uq7vBG@or#x%
zQ5w2J`3F}yC_s@&KID7>MjkN>__3&d3hE!@88o}x&?G1^UAYLRQFv0^lqq-uHAL=3
zBYD)<WsvUjFx{G$V1wBL6S=+AMsvs)V7J)&q+GIn!hNto?WSYCxd0|Arg-_goM2?m
zcHA|jMSYUkFxuNh#5N&#+#a-($MnF+DN}Lqa!~96g_v}k`KsA;JdnXVhzCtX?VKEj
zAX1ynb9Uqab^zg`KD-Z#hAeoj1Ye)ngabyd@U<Kkm^sX3gV<d8>B%RpcgBzQrk(^~
z`BHqB+cb?H{~_rbVb+A$TvL?Kn_9z-gi^~@2;@yb^#~*E9g{OqU$1j>C0L}hgp3-v
z^bp6!pF}|nQx2Aw<#A7Bx=|>pPvlb<Soyx@QU(9)y{C9UL?c*FvV|}9*9<Uz{8Z9b
z2=Kjm&36#C5JYtHoUOChZi(;}b*|&|GHe-O(mJZ>(5*aGFt6upbt}uqROlwm)uV;y
zRX)#!;8)hsZ1Y74em~c0ibzixlD}24Yb4LG#Eejh?%cf|H8y?xjx+b>yghc?#xlss
zk{-|d2&+>x8KdRPL>JF%8}2LAfe?`Mh?p2fEK}VdOC7-LuCNUG+ViPy6448T<yf($
zZ&m-$S{jTog)h|C)RqUWgBQG<!=%42qld*hpt2astb>AT*i;#hGVRo+*s2)*zyzrC
zu05qui1HJwDIs1t;7mBmHQ$278&FRq5vbmA#mylrN6wIESUY2P&Mz;cr#Mq3Q)PGp
zASte^lY{ot5F%}4O7fIbz8Tq%{~b<{<w*5f7$VHcM_`*NKlbT6<P1&3hqB~5>w>yA
z)nq#QCQJb2Jv`2u>9C1>;CP&EA%`C-;E(m2Nt}_5L0+~?D<xx7t(CH7piumxw-BI0
z332&GuoR#Q)e6wAf$RtR*0m6zU6R_?L45ZI`wq*PlVq;NQ$vO-=jy}%)a%=D3ONe#
z)dAM7|0^SBE|b*_ln+f|q@W83n_4RO2KGE*E`ct6pgw=af3<fc6lCca=9JI#n}eM7
z$G_)aJwL1^m~Fg3AAcK&!2jZZkw&(XVcyFlIO^+w-FE!^^ZBv~$3wgiE2<0BV_0O*
z6*&78Vz9xTbeA1cE1pO3Lh{Iy<aY#95DCk$P>OvC8%c`HsuQLRhc=x6tT*~h>ADca
zX^v)T-#+RsZ4IvnA79>kaGNTiteq;V>Q0}WQJyhaSO7Cx5M?PV1aJm6iU=qBe5ACs
zr|3MI8eA0`^aAf28LFr4k5ruFP0PAGDGafD&7mn)PXnJ1y&Fm-Be13i*ZZ>Ef9lq^
zft*<jd4s(RFTB&-k{16r_s@0qh5)|7w)0$P+XIqUhA)kZ+AQQ_)~-&dIf>T<f}Yfb
z^h`zg<KsV?rD&pQ8Unt5H!8Gx3;p?OTAX{gqLPOA^%QnyD2etsS@=(f3W^7~i?UF@
z98$T*BrMB1vb)f(;^^1>2x+AAOb8jI$CJbAwDlP#+i!64%wkEHNu6y7D7YokjvMMN
zOyn%fWhxR%7(^Ju@Aj*hfax(83L^<xnF9&lCrPtPVo~`Ua{ZHWwc!(sW<93myC65%
zampfwReN;0V;Ll?Hmar+`%UQ?Jg)F@?H_V(QPxk+S)csC8n6h$$6i+y-tLo>j%>EE
z-yE*$lZ>_c&YyTl7FC#mxG)amjlkv^uNslYZK-v^K<3EiMcb5jBn)!m3FH{zcfR0#
zRi^zl1V5(cs7Z9wt{)DnXeR@i7MFDPrzY{7Z;#OeAq$(T%`eM_M_YiQn#iM*si^U}
z@Sx&DNuNk-&%S^V=q?xkq5po}um8_?(eJ5=`G3`a$!e0Zzg4w8H?_C&U@H+~-SL_M
zTlm}{{O6MK$bi&~EC}5c6P<UxH*t(fWVMavvL$1Cc5Jb>JRBN0s`KZATb);1nWCu4
zl?+sjIn^zp%30M5#L$Jr+SV_Xwfv|JWLmZhR8TFx-acx!ZkV8Wyxcrq?yryE`d=^4
zv8=J;my}|KbP<kou^)y+j%f4Ywz!hVkhWQmAt(CsCT%B?W-r=}=Cji>f86%vTO8mE
zEn;82*;<#&I#hP*{DS3&_L|IwFP#h6sQC0Y{LV*XkXf>Er^7(E08K{{hx&ES?BzNc
z&s!DWH($2sCb(_Zwe8+#<(pm>ul1)R_zHpwqJexXZ2FY9T(5F$`U3!IEw1&dXG|&+
z6N1TU=g-HHehEQB3Nkr<_`~uDR<a6`NkUg-&VquhW}m2udm?g7&na}awvdV=^dkdg
zRM4}1T`=sD^L#tTb>S!djhM0~N%6Y+61i~4dA9J`x#%n-`^td!lLlC<3Ns6AQuun2
zNIh_Nd#Ux$RtD|{C8--#st!<`YxuqqCAGzn8dKIzdiR?Zf9c8SC7Rl<LPVC@SlAIp
z4uv98QVqmYf=Q&-h$!{7;)?rc$|ey{Y%f$8z1^SgyC9(Dn&ra5Fwndr7s!Av8p)v0
zz3MjJZmbTdAS1ifEIgrI4#nr1pFpVxAo}yrfP;LKSQz^pG%*|b8p0m9IR+jCGJ?$I
zcL{vhrHtiVlk<@MX+6<A!IR8il&q4=j;IE>5Ohngjf<ri#Q(J>CbL<(D^EKb?F}X$
zAt%Hc)c5t7HR}tHNq%#(4Itjrl<2~VW`$tv>JG3MS%<{@E65k-D>(rXtMTb}eE20;
zaiKR&x^b_Cs{!X%Kr}|^UCNm_okZv9;^8;)4J5ms2_xC@PC~N07&H+?X(#Oqs|=p-
zn$E{qC@CDg{L#7>%v6$0<clh_dXG<;OIXupV<BZ`OK47+^*J8`Aa;%>>R4{az0+()
zvh!$0PZdTl1<3+JfVtO@trjFi))ew%2BMHqNL!<{x60=%gX+a9Z)TGSB&K56BUG~j
zaR!vX_*3cvQeek8A-+6=9wGkkYF?BM(!Ylyzb$XYG76k}Wno$;gUL2~MtH7p_id@Y
zh;>)Gu_2qf4V2Xwv0;JvOxKSx6qN}&n05oX#4+ngKZleiJK2)bd5e1yy*Ag)4l1$t
z+G27j#G1ueO0r-@<PS>#j`>^#`|2NUy~xQwv+d?k-k{S%59c40r|fJVw=!BqDgkj7
zp-ypu4w2j@`?wX7-0>HciB&#Q>?NBb^nAGf+Jl(zFTH-#CqMQ}(pCDn;bH7@xoPn0
zt64>~%!66WR@ylr2gRxYLoK3xh^UtP9)J*38K6{ZYZ;&ld1{{Oi4A{BvCY`}?jVJy
zU=H#(OU*b2j&8v@1MrVoP?CXUo{FALH=6YIBTQE_gYq=fg6-#V)a-<)*j|DT?!^=S
zNPwbdA*WdlS0vtz!O{##3bLbT*wj`_FEXqTvm=*Pzzo#|wE;HDSlCaLG^w9w<h*a-
znksqLdGBDt?U)YGp+7sD&LfAosiIQ_Q@aNSVv^TUYRjYk5mQWJ&#)3yz12%fK^_!~
zm6HGiZf+OC&<BYQd%}hhVlL>_m)i4>m;MaEa58qUYdx}dab}}#0%Pi4D4MlE=2Wh{
z3qUgalSxt~xg+y>u*?LHv5UP_+*IV{=8O1s>fy}df0e|CXSCJ!FQrh>eiCKUPn0T$
zd{p0u?)5OxfH5&W#_!v?tWAJ8=Pu54sj}{MAa$0cERsQ*-5)f`3ET0<{@7N{d#etj
zZO;XkzVorM*&)uF#8*(Zk1&rkT@dD>i@qy1OgM#KW3A2Q3ZdOFCcMGKKM*Pn=rGPR
zk%hhWEL7bRRi@>eK~#V*pTVjt{_0dq{KG!zlM!w%3TExY<ZOUxifGWGknT9>^3cAz
zbWVw8iPHgG-03ONviuL46as}PkLa>kz~yiuGf)!^WHScX{=pOTluO$PHBTF$AyFf*
z{|@$xEl$Eypv4l*(|M{#>bQ9PJy}C8vLg$%xF6$CXPAYkhc?VYizp+LPWh4SyGH^F
zyzbv+Zr@x|^iT}42>vgv>ntHEr-5*;DD{B99Pz=Vzla~NuPg9$OFS765I9+fx>s;n
z=X~mTLqCNEhUm29pt?<C%rJL2CovPZLd%3xJMG<AJluajE51hzi>xp=1VvylNSzr)
zPvf`wln{DUzzM>({^j}11ve)@9Cf11v;(n1|6CZEH>dC@v3s83qWTqMpU_C?AkLw3
zxwZMqS}qER#p-oo*6O-96*z(II;wL>VCqP;v+A9P=HABX+L7)?ypE%vQCNQon%oU1
zSqbnaelW>G35uaY))pQ;jb1&@Ie;;8*PHK^BMT(67-y3MOhe#Nb=<_+e`HMwJ!PrN
zOGAmg6YXPSoqr9g+37kFrhe$4&XJ(}G69KpDFgu6vm?KBz8E#GytsXFLfaaaFmGbW
zZmnGv2eVUnCu@hRNb%@B5>w1@efZQ1BUyL!qeWd2z5JHTDJ|BSz*|x?ZmXMb&>Xd>
zkpG(WEdMd-ndtxLpPv7782Uex{zCOuY<dl18^5Hq>7Yow)ZF4?4r91bpLonbIvh$|
z-NnZn4Njs+N`6rYt;+=s4>i&@bzojFsPLsw_irKV4O=ABLc)YuLsVK6NGgOQf;8$p
zQ;L@t@7wD!PJai9aJdNNqTV3n*Yn55Y+~ld9Wx4BikTNRag6Qvo0q4f>&srPtn11U
z#4jb6w9u9aNE<s|jB-svIDn!EBKG)V{t;IcTI~Z}zo;qVS;hfJz9eo0@!j2t^}Q3^
z!!Qy8UW%*X-Pm$uH3%_#XoEr1)cM(nOQzMx{|Sf?UEV$-^uh=oj(|!6AkK)IFE2Rg
z7+0>P91w0Xs-5N%4iy03fLJVvbh;Yi)2mnXczo)jzzH0<obEinvf+t^brxT2C~wy$
zak+7%g2ylkw6OE3ipFF|S{+-d4VOtb3Hp~iYmB{7;GoLHX*RZplrsDsrm3Ea^Vef&
z)m+<THu_+o>?G=8-+X0^qmjFy3xRsFkP0J38>PJ*SJz~6i7>05Wta1r*I9y%vJ-@2
zr<%2~g66ogsiN@>kU&2Wy!HBG-eB;=B2=hwU%&t&A9{N5h&=Mb$bqlg`^Kg;Cm8Sv
z%*b?=XiY~ZkB}eoOFV%PVo@&zk{A+@MjBq8Bclp%N7Ig3X8(w0)_azFw|RZZJR1dz
z_e6a9`FOZwqN3xz(BN!iiIuvi(~5kHDkJ7W6RR|e%+7)^cod?`7#HR>TLQP#IK5}B
zCJYTjt+vl4!}KXq*N1GBI9yu8(;Z=za;=XVMSsL=Lw%r8(RoiZtdL&b9xy#f98*@s
zL~4J+?aey{{H}IdDt)>6Y&aSNF~DteremaD-Ik!E@kWJn-fzsPSl2$ZeWr1=u1WkN
zThJiRyM)3=6-4|YOT$$s_;4-%i`1%2DeP|*G~uhH0k-Uo_eR_&Uy)r@+M%|jU5KB@
z!Mw*J%_#zLMIcfbDFo1>Wp@Yh^+AXOPGbK7`~<?D_^Q^J{@*#x4<s3a#>JWxK^QF7
zeqhhs(d41m2~#VqLcwpDT>^TS2Ia{eUAp4_6Eu59^TlrTfiI5DFd(I*?7JqVZ=pE`
zwDo_&EtzcISuQo?ax*z-=S+m=uBx;8ioO$-GLF6kV-RzhO4^P*V*pZ^bSTZfRnzK5
zpP+{CM(<q&B=DQzj9OC`vaBtNrLm+_S1m2l1Dwmz`m3`nkybVpl$qgVT<Sj*gZm5d
zQuTjORrNc<Yi3+N25cK1VHOwc!Oil~u;r;YbcDS_jpV2duZL}>xttG)i%2>;&e|gs
zaRB<YM)BxEkbnCWD%DYQn7S3_QbHGopK)OMH;npLNePPHA>!piibzN8Ow@qVt-oTF
zQ5XV)qIh)QbTr;p0PC}j<4=ZpHn?8*2BsdGxQJ0O7UCYw#c*jaI-Upvf?UPqIJyA@
z;1w&J0pS70d?oO!%i!nDHF<<iQfzj9@>EI@^<ebH)-j3F+XVV#7$Nbnp~B*cNR^Z_
z%pl8u!0AI`iD9I`9*}wJ<MedO867`*ZX$s<OY$pjn`ius|ETCYyc5**kIsGD<7s)g
z51O4bX&5VP+~)T0mh#n~(egT~_TK{VL!*sBMR4MbO*E39PzRl~F(QNss90E*4np!P
z@~z5`QS@qp4Ue1X=UlBu;0y+0NE8A^MbO+jWq_W1Vf`DOs}byM#}g4v7r~B<S@u?`
z<szIV$ihLQu|;2=)vTbd{CBLr&1KhU_Bx2tw5KefF(h68&U|>S$HT6GY|Ge!!%h{b
z6ECM38S{o)n<cIf?suyk3c=2Aj;*YYd}f)JY611?vieF`bKF9+X9&d>9ggeu_iz+N
zLApEp{e1Lqdp91)zABeUcIeuB;;-$cIYRsYecM^-{;NZbo{9B;-Qv2{w_*!f;C*|0
z32jNvM^0=Hp3AFUDpi)xmP{y>R~3gzFTt4ok;ksDe0S(u0FfX%m2YqfA=VfV-ae=O
z(BPoEUhk$~E?8}CAy%NHn~EIjdvxjA;k5{-h7G#4y~2nI(WzAh_RYRVKQ4|Ups>Fl
zuaDnXFI;)GXnAXVeLM%%bN%`t7(pMeeDqhk`eP0@u%taARehc!D3C49U#e0uMQ2Bs
z7$)yOnxr10KQwcm&6x*M@uS#p`=A%Fff$ho1D*gOzE%FdsU~mfh6be)$8gyrCQq_i
zoU%f<oaLe~Zc^8}`t0nr$-<fI{B>42Bv3{~d3hG!-!96HNfB3WuDnU0*veo;FE{_o
zv<d=QQ9Wlue_)@K78$^u*^B3w5D}r+(_CB8)ItlZ_^?Y)-gsz}`_Fg7sZn-9WCF^H
z3^II$^QDHf{Ov_4$#<i(v!Et2u$vc$`pI+hS&!s-AF7B>-L4pcWFW$es!>qIG*(9b
zqe^Dd=;o=unKe7=<ou3>;?h|WUD+2?JG=@G<T?mp*`uD2UMu{m_{G?s43jgva|((m
zWZL78>WwGo36jkDckX++*;T6bWp(oSLmlq9k|&DWcYs_GvGaobWON?C6Z`-wJp`Hh
z!6LGKzc-n&JNNMmeZa6yxV>QMpy2&wkaTE5$%e`_Y&W1{URXZ{&Nu@obWUUOZn@Bh
zuxfU6AY*OV0DjIqy)eugQ{|bZ4EJ!ZVAsX0o6U_U5rzx_xyH)0I=-)d41S-$L(fib
zMWs0JcvowxxDK=a!HP=oEStFZv7lDn+#m*K)tN%XkeDGTkmaGGzR5yj35$NTdB62C
z48lY}b?brpDkD)EevMh3@uCsG-YPn9jcBSw7JU@{@FYk8F)%Kg+|O47Uy=YMfZ$W>
z8r37Bivr>FzUrMK(7c*A4QQPjk%%y#kGDEu2|Mwb-K28daCfL!FhslyC~}9F6ubl=
zbkiOEF|6DmiaGF@YTqiD0nFWU0|sY4pI#dN=Pq(br6i&N@HtvUd8MSKW{0`axx4!E
zbL`||ixu9&kP4o~U`;y~xy%xZ8^aAhOdtqTk`Y(v%TOt{)Ua{e<Pp<}upPIuN-zla
z`&+qgLm4ox_nvBXj9Mr&dLq2+5X0@<&>))JK!Ok>IEb&3Mo<uGgL5t`yr6yCq=^8c
z@~BfH23GL~C8bCX_QF@AcP2XmQr(E44L}-Fdz0mj>|4nU-}QibAL>o0V(X*X^D&%R
zn32@6iJU^1ib}zaSIo@CBLVeb5cm05%_CN@=a}OO#XM=b#qN37QqDbq3Q&dQ<d<XT
zZD?)pGnrXt0b|qYCN?yL-umxVI|=C^&d6ObnjA6~nT?aI)ZeSlB$caUS;Vc;NOSE+
zR=vAXmPHtNCF(gNI9ir1p(4~#1^eREJk{FeS=#t>`NphZ{8PLOYox(Z5Gq(C&12nY
z5{S<^Mbn${B+`pz0U}8l`yy`Tstv293)25YAUIPij~>mei@y0fJ%dw`5qNT%1U{WJ
zKE0I!SQ-0si-x@<P}z}O*I$wWccLwa9X*mCYxi$IDcMKrs~MlUlaB9Pjs23ITHx_(
z>huF_apu7AzF|lCm-mJSc0$J#k26j>xbD+Ts7;EkEEXKfMK)~GBcYA%FghgJ1>os^
zt$&@&3IZ|Yk$Q{RJui{!DP}Sv@b3F6v8o?=RN0@7(9+@(tbh=jx_UPh1KOODbZAI)
z@^2~8KkP3<u-$0@=LkZ<qDi&v1&&R$R`cRD`NH0!0wiOIp*|ZUvI;3WZFCCQDYpj+
zL(Rl-PcQmeTBlN4AB8%F3nF=N5c#o_$wO`wzOHlu6k|IE{Ppw6*q{=JQhQDnZ<d`f
z(P41}%lT#(81rT@GBNIAhYnyReoKaQDs{#(pfHL>YkG?#2fVQ;v66`03eY9XVf`6^
zd+myhuvc(cRsuwdG>D8thTzuKf)<S1;!&xAcmwrbBOxcKX}T5PVDyHF3C*OP4q%yS
zS&tg5KwG#mf|x7mO_4g%T<0NwaSnTmgt3WAL=GDSnir(?j(ZmR@(<&ZWNVS8_-7;V
zU!@syaP8CdEzrz5Ae;QMYD-n1f3g@n8^2ZropjS9NekNma~f}p+2mj{r05AhLe;IC
z5gB6;;)#hJQX*xqU|LEU0JT%)d{l0|=4Y@Avc~9qr!7#a#5&$ti$}=YWOt1OS^zOJ
zF{*J?E%)KQ870=*Gr7uTU4Wo(7G<3?I%Q-#d0I4KL^>(C{BfI1(d(}7+Gax@B8+ZM
zRZ2UP@%qwFp3+swkwN*t)I1Ys;F|aQ<t70(ZHx#Yf+$!doJYQ{6J*I@Ho>th3R`*W
z`&thtGGG=uJ3GC>X~#e&p7{xwi8*l)n`uO{Bl&me%6OKTBW_9CxjYde1iA52-Mne;
z-a!!=$MMzS!dMph;3>VOCQl>@+u7s6#%8qLHe;ff@>@&yeRBY&@7r_5f`J68aiUKg
zA1bKua-m)hlD!^gsG)c=sv(M~KgjdJeSPY4c`GnF*Ukc+b~Zcq0#D)agnkSo(BI04
z@nNJttgjO$pAhrFLQ5KE;}<6d6kt%p2f0~XAy=m&-rUib&d~X+`3x<vQC~*t(t@J$
zQH+E{dOO%7nJVZ5qeWJy15+yk_&Zb$%AkTwsZcX%+4pLSD3JysL-ha_w7R;R?o#DK
z__n^P7ZaOPFr4nNBlLV#3n_8GoWyV&xX5#_KJ{<77ypoEwX8+8kQx11lzTi2JTYZd
z03@}_NdIoxHf4S`D3+V!OeM%a6TnjQ-#@k-2{YNu3+aqo;3+yTCGj?(C{y|N==OP_
z#kWba_z%rO|6j#dY;^y#W@*;Yv^{7;@tLi?Eo{JYbT55fqcLewL_TvtL<VKx&tHq6
z9%|`I$&&i^hO=!LQ%e9}8-PnN%NZQJ(Q@x^T!xwT>3qNVTBglq&onP0NpAc&!ANB6
zCmGP1x=w8jm72I1{cT~oPuJVJ_}xj_7yAZX-k)rJ-@d*VS1;I6*b+}YXozQczD{0h
zbFZ&^J|CWE>e>c;q;`eZsK9j+FmxXJtYlConw_XlRvp9bMJQRlh>7^qfGD|$7{d`(
zyhz1jAXXbcsN+y0Tj&ewtQ5PGj-yS3E!CDbzDho?|FI}*;zmqIzag$K);mvIx2xS$
z2-mlht{-nTD_e{HC|ymE8hEskx~mBM@TkUWFe&Z_n+On8800U%a-}1O30QK?Z!L(+
zUBWzt^F9s&(IwaDHdCuC*RXYQf<}v#fh@4YS#ZPg2r32PUj38l;i*}Padq^vfXnl%
z*v@whVVq?yh?+g7gR3Y%xWBQKZrqnck;U?Rrurk~vPMuo9z|*Mw)VDF8=cZV@$aSE
zw=#FD(JoeXcFUW3a=64!L$f1A8uq<qf~5)krOE{%mRqi@%B%X^@IWKS?{fIX8pyKG
z1)6jpgV#Mrd83G8KLJ0`FWo<}%&u-4qu)oyuPhg-Yn{En@z$d1qTBOg`cF^Aan;ok
zw)skV%fIFmzm%}7E==}tK(7S9S&RfF*IiDsX`y0p0h8h5YByUCfQiet%5S!Jn{LR?
z5Yx~O5K)b)ElpkX1<S%8-^Q}Qj@$`hdn>QmPL?Nde`K(F4E`vHzt7tk*Bu@eT1@~&
zz=zvRga{w6A2$XE+`ex-vQibLNnmfM;Jm~`mvtG=H@d!#d#VKaVbE0xggK2E08&Yc
zz~Y6bmvSLMNzk-<GHk9)x-OIVp3R!s^YmbFych(FnB@Ckr;`tAOZ(VI4Ee7-Jbrm}
z^=SkDz$bCSZmD`XrIVx&{6biunYcDOn&nY+QkPpc({VwCYF$@b?r#G}-@Y&UnxCnU
z@J9RctG^D>uQa{<0wjjq2`{cdyrPK4#iR@yAWKJ5T=>ftdeKE$Ow!HpRmT=NfDLhx
zaOg1R!%@Z1E!pqL@F*M(98873pasa7ws>M@!_DR#aK)A1jLx?LO{D9oy|>M^paE`>
zjJPcXklPFh6y5RgDBd9ic%&jmkpm%6q+ds1H<x6RDqztqSAa1Pv1YtEtYT(I?0WR2
z(2Udguhys}7rQ72y9OfR2Bt2;LN<5!;sJbG+;de&jwEv9a$`G%(Wcy8TsLnnO0cUT
z@d+eQpfz&9518ob{y)mzu{qN&?ADEK+qjdCZQHhO+qP{x>DcMmcE`4pj=i&Xt*X7N
z*7M<A|KXbR9M?I=7>5?%uTjy&?3gH38ka^=2$l$9>1kyi2W63%LW4`@L5zYz8`XlV
z#XiGp|DboFi+@F{s&@C2^A;yGhYCE)cf%q2!YN%;{!hau-a_C%awx~GUPbDCKgYz%
zzyLw8L(V&fc$`kK27&g7BD}g=0csf+rief#7|7BT!orYmnv`3-AvzB?gps6ZkAQs#
zz&0gmm~TfuKmk~`O!)~EVi{baI56ZG9KsCiA8%$=1aXH13Tn1NLA@ehU=@+yW=vF)
zu<tbLgFWR6!Cx9IG6M%)M2m`0moeLk>d8~3_X(Ag#kZsh3>)Nk;7R^*lL?+06zX6A
zF&$;BsJW1$oP>-v-yfbhM9IvcmjucnHoLUHPLG)^b@$PU26*SiIb{Oclcj3a3LgQk
z(&T-WR7)P4qXCmO8Vn^am^yYz4#dR7DCo>zShQZ8)EVFqm6<&ZS`GosMSxM%LM4TV
zagziEuW%{OVZ8OtP6<6ETA4uVePEtrQiUh75*ZS56y%*MH(skCAyPO|8=ktzKr*-z
z)E5d8L~wU?mdx$RA&xBK#0d=J%@3?GIHR7FNVF5(Kl-4VDDd|g=M++x+hD}4>$0F5
z#$)eP9O!BNZ;+e0LMpEbyTT}hSrp+8MO`62fA5iSAGm{ll0x#Fdj6g`Wx2{F(A{(Q
zWHq@YQGq;Lu^_ePJfuaMn0}SonV;kuRn1!qPeBrOf!!pGqbpAH1Jq_ZXab-|JcV#7
zplOKn>tL@DhYG!)&yB!>D$p0R5pN^*z9rHu8C*}_vZZ+ioZyO(aJ)5eVKT_>YDWR=
zT?ACO#z!TLRP&nKg>c!1cLHG0MUd0VApA6CQ}>2+GH;#)71ijxp4$StpOjP02Mx6A
z!mf7H2$`u>+d6}OW3Lgw(d|6^qA^xXMlOwr<KrwLjkWI;oN?yX5Y<3b58#TVv<Co8
ziE8O!2pX|BFi^$=l}+I(AdWY(!MDR7sBEBZ=8F00bE`Q1t4gz=Yvu{#51Y5JPY0co
z`%vYB1O2Lx%#1OyBG?HFw54j#t3W?(5Gje04B{fEV*y|;X?neG!xel&_C#Ubq*69M
zfY;cn#8k!<SGMM2D2Gi#jqTWX^A_4snT`JbO>8*B$RK}=7N;?SX%7J(kq3=)Vub}Y
z2)FtWTS^!fsmuE9;3~(6mi4fI&STCv?BY7(L6Gd>e`3D>Odjpt1h&99y6YRP$V;6>
zY$%+G$#_etpWM<ILPAQMTQwf*C!2~qA7aeRf2Mb3<2Db1QCoAEoZ%sh0)OPcy4-7G
zekwm}$KSZe%0X4(R$3HzBpQfLk;jal``3pDA4pW9e}B4m*3W$*x_igIDwKI75+74Q
ziO6ZsF;JurK;l7gLSL|J&O*Vs%0#>+M>CXdW6qv0n;f{*jCOtBk)86IRc{IMt5~PX
zqb6xH13&nm8&^%%1bxxflW&X&Ax3D~T?Ygye@vcW#uY>TVnhRG7>)r0O?qGDCNa$!
z%Qt+N2YVT3T)WIt_~6}kWqRwzxFUU%d&>PdnyHO&pV#cKH(ZSNU{t=bJ~=l9NV5b9
zmHgo$w${%ZTU~ZSwo_0;7Zlj5C$iSw@|)F`(O^R0zcgeQ?U!=hgH#pQ)2YT1Fn*@K
zjXVIG#xkudnBDc<NEq){PO}&p5}OACMtqG(*91p1P%n|@D!tKV5i~&^O5s!!s~Wr_
zPWh2@!cX}nI=p{EJo{He5h;!$qC%l~S^Cn>YB$@T{iuJ?3C^i>+hv&+`8lD9Ss?S@
zJP7iKZXCoe@%&s_B0td9yJYHlyh}Fm*t=&=i$Q}ZEh;8IAs#TX@X|z&OTLHkjt3xT
z;=EU|%+^<F<+6BRaS4ulz4U`_*>+j-LOY<D%P7Y2c0cc~23D^hKMx5&e;s|HP2m${
z-JS&%Cz9pV({7p1%Y>lAr^>HBJl=DCnCwR*A0Lyg2&52;tj?|A<C8~e{*5+n%(!kg
zkAIl4gjQoz8Bb1<=>Sn(HL`6nkr&f7p2gdnl<#JuZG~Z6A^T`4ba=V|KiCTzXyuzH
zti8W%KeHorHM5P9+A-~=`bZ@QMNSbus^Vc$Wc@h%I*+d==z}4JDLpdse!Pt^m$qOV
zEr;`Yns!_7)G)U0n!da~*Wdch*c7FVVJh*&S<~4cF=aP+wjR{GRWSV+O|~|Ogrrmu
zaj;OH<&W02+@P`cN|nbd?maL3YHVil`goG>8@!Cv?DM#jE&F+_h36MVPFKS-Fi{D?
z=iMfqqW=#-@TH@=%UxGm?!p@G*~F0V5irV0+npL7&t9L<y=W+(^C^pN?9t*~TU~0;
z7-bcm&Xg0O^_y<Ov_m^2n!oNT4D?TjHdEP%4=tNGH>dUUr?=ppZ9!faXL8&domKsZ
z)yXslx1gd4-5KiCiNF0oA;*8{Xr}*gaQxo{ep>#&zz>G!<iIYxQM}{`ck-Z}=}4ti
zs{4efbn{RMNeh|j?8m!Hv0!8gm1nt|GEV>y@WcCs-9MA$+r#qe-fi9PF7!>xJkMXY
zPR}WTY%E*s6^agR%4gWg5Q7zI<JbMc(R{u?-_0LKi#t31oqAG(m~G#0ulDYQ{;#k1
zkLg|hf4Y{fJQDX}enGmN5`=r#dX(fi94mVqL_(2?PJ}#-Zy~f^H`+UNqeYMAjQBd!
zE{JwlclpTv(v>xSNia9{4!%b_;hS=#j^9vww9`M<KPaowdY)8GM_1i$X8Fbsi7p@O
zW247Ov7m-xrwL_*TJYUC@o0g_8p4)+G89_*nt(r7hwg6wjr*48Kb@d&qW<pT;=FpK
zifLb+Zy^Fw;zg{I6m<PtXe)Xwp_yuNJ+_0o%5Fpll+olI>%__@$_11q*seMI96~Lq
z)gMybq4X&Vo9gTw0?sXr6hB_$mfUb;^!ykFnKC0ACCcs=f)*Gd|Ltzvz|9y#wP#|3
z%~}CdjK0s~&$RO!1f5{@)n3E2x~`)7L!g+f7xXX6E%eXPqNqDyL}m(0w(+}9)S=1G
zjFH7c5}4DVAV<zQTk4EMD#pC4AIR|p3BD~RuX(Y-AN|d_mCGclP$jr){4M_xH_*5c
zM;Vz6nEhmN)UEU)nzO4MZ%vmv>Dp`7MYEqChG0b1GM1)+Sb2zG(D~$kU<yg;>&CP>
z5G1$C6_-V3z#sQzMggninwvXDD<A<@3!srm2m@0+&`R_5^RGO>zd_B$nT8ZX!%Vzv
z#3MA4G3aQpl-q^BEPv6nojC-&U_FqBKB4@|QZ;Ry`j%@#vR$fxcst8pr1Z~ueFO+?
zEakAa@v}UbUO0ME9;LV49`sb?9nKqK@)=TqSd}pqQCH$BUc($!CQNXYWtDp$vb3~j
zF!|>k)I0-Fzamhxm`A4%2Z|Y>mc;_<LiRX+A@<iM_2H&%gW1z&q%XeD8XYClTcQW_
z5af3eXVGsMwUDz#jqP&jr=lesnt>EXU~P)UA={*xyaW=I8g6H1oRP%c4R`8=w$I2(
zi?dT#v4WBQZfPpA9)(jOnCA?x?<YEBcD)BBxgNJ*7j`zJ$c=tbVM0{d&$S8hxv}E~
z;kzG^*!-gs1&O68HJp+(h+`3xA>AMNFWjhUI{06jYtsxaN-gBc6Z|gY6o_Z4R+)=P
zbe9Ldg`OjJB9q+u2exn&#jE-NWW;)bg-)<n*;Uqsi(|y0Vd_+*KEJ!B&qb@I_)eoL
zCiL+11|UH=@b-%c?K;C1<cQ^hls~})vq@tPJYzMo02VsDTOO^ij$C#!(p8V#l7Bvz
zG2!@A(fzP)qfj+YD1lNxTGT$L9wk>f?@u=^n<@k{9;-i$CU!L+7yFDMqf{6s;V-_7
z6wzn^-*R4<L~0rRB*~aP=@<VpIgx>)tr6E))Ty59&?UlMeXirFr#+|_2+y>re73W0
zM>{^OBj$o$-xeb}a06rhwK4!DB)^!ogjo>-dgmR`(~Y+E)VN?wf+7se>5+mi@}NPY
z_!!W_fw8_RCLTEa)4DE;Yn=dpv@w`h=2`<3#-gpI1y0ctIUEH~ggBUC<loO{-Ke{*
zm>zfmAA3<V2?5tVQ@tPz54-H00CC=mqpK=tVJ8#-A?Fws;(+G}Ggm=GMM4&E15U3*
z!h2O(c$2?Rf-KP>I>t_XJJ<sM_v;0x04#mN!zi`IQ{Kv5Kz9Fb)aU$cK`^n1L?NcI
z`pS9GS(lHnySsM6p0O|>i^8*nh6`-q^zG)R)~^d^p)pOeZ*lqh{0mn8Z+)C@);*mk
zbeCH_the$Mv+2Lr^ZX^q_-ZnaU7&&jv0x)AgT?XA0Z(TpYm;`*VY*DvYP_qhO-Pj_
znZo-}Hcbf#UV|y`@+hb_Kp$^1O*r@4`d-WJgz!zCT<FT&rY@RaPWT<kS;jv`V6!6%
zf1Ac|3>h@AC(F(Z$$kJed-v$>+b1RtW9m*q-vHzv!g1$gunuI%T})E&+;ir!Rg^|p
zEO?_R&CD6b4}~Lod@MN$;%~gx<>!XqW~&%;I^5`Mada5_2r#)Hf;Hw=kk!QT1k%s1
zLMu$}>q3`^WYBfSwTZZ6!aAkK$#~$B_$h_PpZb5%Opb3F=tQdJ&8UKH3r7x7H}X<c
z>h8n+^}o`I1&A?7^s!ElUmLJ`mX4%;sx}UXG_Pb$Fy~&zZxRXEKrp{4x4Z_YN+%`1
z=R)%uc!oM)6}X(<bX)x`{wZ)!)6a|~`wA-))R;WvgMrpLC?x3k10Z9*F9mV?O<f@0
zlp#0azm}k_GUz-po@HX_ao(No^=-FbSVxJm3dyjx87sH~iWRb<ZYtRVLxpkQxMt42
z@lGg<SadPvH7bP+i3)VCzAM#q=YM^bfL>&xHt0?JbDL~2_%BAFZ^joJIa=ZT%8=F1
z6oY3<0pj^5c-&xFzvm)IzTUX>G9=)|<~7GP1Tn}>N0G%WAaoFKmi7d`FFq%%EJz#7
z&#PjQSXcr~YIM}URS<NdfKr&-&`4t2ImfuSnqGdZ)vMB+NsZ@kGvo048yoJO;9eEN
ztWc81bI(@y#P%^guGPNjq`lDZ1M%gxG32_=XDdY5ntIePVfb5uSI}d)_Emd(zW2D$
zwv2T3+<(9gUHw1g0yD#Z*g#kr{>L_uX$?*L4KXCYmAX}N7={&ZhxEx#s<}bCg>bfp
zNRj|lsF>!Yh*Jp>>Nv$84(Jn+C{qg2np_rUr1x|S^Uf5FAhXEaO|bq#|EubdVi4*$
znN)d%G#N>xB$)!2U|XQdVA~ntiFyiItZods#vPc&s%$Uc_p7OwCofzkQQ**^f&_|A
zkDsTnr}O=O9lt$XB=8*%w44zI*!h>enZ?e+qp-YgS#Ufkj16>AgtH7llm<fsC2V9e
zF!MyI&o5M?q~1V4q4;cfS-7;uwFK`NTLTLA8llCkw)?6QMIum>ye>wb6p!lsl54AW
z#I#2bofdR}s-T6NoFfiZb!(X4?gkm8fB_`hAyjCBp*)d0-<EC7H=3{$o}9S<YU6;a
z3l%q)_6;Dra=|W=fI4<Xu9}B0PoB8j^e@}m+dnAcFk4!E#x;{yK<I0l?1N|BX9eSf
zXQDs*Y!0F76cr`TIT${KtLk^AHSRFAorq%vkn2G^?bmAuX*44r6fiM42ZrSPZSQe}
zb`tL|#H1uV${j|ttMj?L8Xr}+8vGBBhg8?VR7Dz#+**tuI0~nQJcL5x(tEyVPYBU>
z@t+PYW`s%$_4E9k$c1R-0L{G!0^q6sE2^S~>$CDf4yU_aq^)R~_<}Y2>W4FhCJ(DM
zf+HjOY<Rjn7TlaNoL8$VL7Z3qo6EZ>U+4(cw~m**y$=VGyj%m}U+d~a_47Qxa=&H?
zax4Jh|9FvAj{~?Dn`G6_=e>%=v!4?=GTeMZ&k?h6u%(Wg3d1$N6;Z5NM93u;C0|Vy
zZJkWFY6Gm&s8f4+$_CGxW6^}gk2Yv5>QF42MBtnXU7(Fhoq6FnBhSM|VmMxg5=Q!M
zx|Vj=-6-gN+SZt)m-`eK&jKc77YvBpP4YQZgsQFHx<-f@Aj>DP#v@xM($=Bt^ExxV
zEVTtHfxHJJ{e+>k&m^^zVX>SrOhU8ijxV}Wy9gZg^fE<+ys`JlBJsa{j<ie%hhKy5
zobBJwH#}?B*k;Y0>4(?9WDQKzYtV#oRte@4r?zQb*u`cLv{->5LHzKc-|?QMf2~kA
zh8R_<WJ_gsujtiIeQMsx0)^2?@c<O0reYs1o#GI-%d^rB;W!*;4X>Z9oEc!!X88P;
z3?N7+YdG$Mmxp|yUK~)>49(Sgm+LT>Q;3g!S}f)(UQ#xuC+g~Vo?w918w<3jgxfD`
ztd>gQT_gKcMCG=6_u3<~d9n|UDR=M7oIOSp`c*%j4A@JrbyisW$T>C!v*xjatzBM6
zi{gEgADgGwUpwLKCZ_zij*kAoAh#+ny%bjAnb$fC%Z(#QWMtw;WGH%m3u1a;iOF#0
zFeX!|u!rvOH~rPL=g%;-5&lE6`A1P>!s;f_by6Mkh=MSk;(S_X+hBdyGS$+eNCrr{
zT&j@Y?$N3y*IfIULWW(lXH1>>=1KR2q@Uv+>#o+BNqeg+K2&eIFX3q;tGWB~%BZ&M
z5e9>jBMz&ts1NzD{?5taJ*@z`CQ~As*yZMOhTMoTxOTjx7`zkv&$%Z}h!44>k)x)V
z<NLG~77Sgrf}MtAxt(F#oY>1H*$%xsT5vLjZbz{Qep*+u1-RPoS&H;j2fbeO<9;$h
zH$hm#A<s+6D0zTng$Rl|sPi-o^O|HQ1hM{epK({@<)}dgMiJK@HyfY2J<ZpeJfis5
znvwcYNWup7-ljJM%$}BU`(L?m*0s3DidDLPQ$c<4dXiKxI;!7%SUNEv6RxF8-pB#4
z^=BrKR=7Y#<QN&y$7HhJQve2a$XG1TB_}is{h~&BB35`pKr){Z4M=>&YiUh94rHKp
zm_C>lejxfjG9o5lI(e)P1)0L5==kzF2mtZn#OBqoaCUQS1<JdLz4VE-OY%%$!tP(W
z0<xlpErG4UuiK|i$2wg0<+-8g7PoKL!ricBAFdKfAPrH_<kTb}B3gHBB$oH~>GS7H
z3L~L;1K<)(JnZ|=IXmGG{`LcZeb;K#|C$>AWgcW@{T~y~S~O(sH#m^G|5pJM5aS@C
zZ_x(0TbtwtZ}aMsw^VRI>zPIcEhkDYi^juGR6uw=$q3gmOOM1#1cA_en<q{<cq0Jf
z2GTzc!W=W^SWxa?2xY$oMg&<zamXA1L2PlCS_aDa&;*N;<>U75GyHyX?_|>bB%M;2
zVDWZ;N8Ue{&+qHu=+Nuoc}<jTJ<dp~Lc);nAvv29^xUnCOx_a@^3&WG_wwE4bwTx{
zxPL34zm>%x5GjPoLqUpTHhA30;F2WL&=13bP9n`*@ORMKZhm^vW=<k>{UBY+5D2C#
zO*$W_CY}jWyT}V%eS`@1;MnxvBN0s1H5U-5>b?~>RT>GHKHcW;!H!uVBskKDh_N8z
z1Ce^Dm7VIy+Vh#P{$++jmK@b(b)3bM_g)xla?CEBvqjQBtLX2ByaZR)^i>@T|Iia-
z<DJd#Gu)bpFSPis|M|LpYL$QaDdYJb7cF(^pdYFDLdMmjzMofOQP%5a2o=I@xzVJj
zEoiP6J;<5m<rZoSjrDtBF-a|OxJQH2eHo_hmX};SjYcajoa}ka>AahtcKG4k+_ssY
z)YLKMBxW4x(PJzJS0RohQxe5wzuIs0q(EHD9}g@y7<@MIo$oM;6tj<3sECm~;ibaV
z#r2Q8SJxqLkn)1)(93(L9+C46`$0q1iPfoncfNw3uu8o-wbdnTxyZYX@`KyB)5TRm
zy`imMM9`nUOd`5T%<X(M3q<MDCG;vG+8rxpdo|hH-zmy2%YM>E$qIw{&R3=*u+L=+
z9zbSqlNdmglEimH_URdV6T9Ci=LhAVNqWn2O^qq<V2_fGYxXgVIkmfrO9QcF{aQS0
zl0}X3?X8j5z}T)$hru%L#f1L?J3Ss{-=bk^ZN9)IX+1C(#zhXU&6(eBRNhn|Wx`2w
zI%7~{Zni-5+5*73aH#529hGLeA%gZ=gB9-)!e?cBJ}Pt`q-2m4xLgh<_u*$GhlocE
zUIytff;mCo8rLWjl<sxcz;e=3P(In}ZO-7AP4Jc7n)TXr`Mg&hc@)0s@bID;Bd~L6
zaAENZ{!~l&vgDP38EtXrf0W^2MIyptHGef5yNaeq-8S=cUrn(Av#!|=Gs^|Ox97UY
z-HAt<#u#^c0wybQ+jT0tTBs>uVlzDLJTtlV*etv<)MK+t3DqY`y{H~du50{2E~~o7
ze=8ih2>%8x#I%%a3~`-l%H9$N=KY<y-o5FFdmdD<(oz~o)hyee>9Ja9&N<Re+zyLo
zR~^>p3(+|~|0x=+RqsHaE!F`)I%E0Utg7{Ha~y3u848pDr6Uz8ahNs|Npm3?>Rwm<
z<$*Q9&8pz4P!Llz*sQ}NA>Qt?U%M;2oEJp-(6yL%v9Uyy^`%?B5V$pAW3>iC#35=R
zjPe|yyJU!_m3A=_2UfwVrS@0XRbPg&W`KlQA1|>eRsuS`2g;MiO)>>NkJhPOMly>_
zO|@lmqf_J5aBp>fu}L01>*jB&nyhw1w@Fo`6^>>$jPV{ZM2ElNpJXir>*dqV42xNO
z&3B=V-}_h0Lqvy_sA=@ejmoVy+V<S%>gpWk@lFFVtxj5Jt2rqhK~@LbUO5AWHbX-c
z??iA?`}|rofm)Oi97|VXud-_a9GrB0V@|+%=kF7F!21_OfsRV4S{SB^+xy<CbUC2G
zz~WCoMgaETASVuy%RnY-%hIskJ1Xvywy<akFztawCBx|eM9fgN79D})olfHZp5`hL
zP{?_6H=Vw(miNCMS^pGj)uIQI`yfzb5b$HzPH&f)|B|j6VbD&6*#wau-)SA7VCp(3
zsm(NpLjO&zDXi)H9jmz+pPAbE6vLKfWmK0%pr<UYF*QkPDq)<&yb?UNaCVG^=j+^A
zb5ixR_t^OfhMY7LT9C>L_!m-s`Z-X5hP<=Zi{-6NN$Ref*}tVl2g8i&KY4S7rhzaY
z$K}2LfxqNOAR2OC(=@R#ddcGkr2w-GLnnv$cft?1v?FXfB2&H_P|mm=wGjto-L|;O
zd&Q5BsOuA6+cQOemb~%)doS<jX8pF;@!9x~`SQcc{6Au7OpN~*1Wiq8`@{dYwdk}0
zf(Ajq2mr=8CBxGMov;?W1q+t32HHiwF;G>i$NM3X@bQGEnX;tRwvil**r9QbqFF=j
zr}~&!)~_oNJc~*A7xRWs61jv-RR{cHL!g2vElL3!@`C0Chr;D?@{Rt6k1>lNW}k;#
ztE6BM{_SP(J^!!zN_E62i;4;Qa8#JO|M%J3`SI~Azh6fKHC=_ivq#T+N@D<H(g<T4
zSFxbw5Ngsnn(D6sj?8Jq9<dlD9KS#8qM_1kzL^Kvl}tY4|H?cVbng<{PWz^$#Q-gT
z7XvussYoSrz%KA>Udn?rk<%qLcGk;by58}?O?|`Z-_3?yH0-C!0MXajp>4h`S5ch<
zlA2$!N_t1wew#0X5QJS^lhSw=0_rHoFU9uZ&@-YQt*^q{84GVlzfXei7$aE`OXMQK
z5OYD}kFjYmanw<@LfLaG4b)L;2U}bR{NDR7qZ+HL2UGwr*1E2oao-kMjbfJ8ZtqO?
z>yxTvoRH3~rupd0_y`~W?%$44AcMwdIAFZlMb#*vZ49Eqwpb~C{i||zL^o95D2Kaw
zBbiiOx?x^CqFtrN64$<pK+Nac*$wd8LGRipG(h@dGree8eat<5R9)X}kKNcd=0KcM
zO|mGP+uZ0Bq`A6i1yz9Kzbb11Qv^rKQxFY~upCg1o9nj9T3@r=TlEg$a^CIGEqG=T
zLnNi**pn%?<X1Pns}?u@GML!b7;14IyYxOOC2VC5(0@_3xzd}VB-rQ4ES-_rlWj99
zyyJ)YW~{FM0K~v;7NFRQj+Ga#-AtN?<GoK(CKG@HH=30SPno<ZdKcc!en2<T+H2Qy
ztn-LFLtAZRv<iGL%=a3O|M)~f#STE7ErpEY701*RyCmm(=&EMy6ye4-U7ow}f=&uD
zLvFOc`}s^xuTCegQI6;uJ~LV&3pxx(q~|<6I(|<!`ia0^Jv$EWX1gK`)KfXGq%mSx
z%6K$}SSfBDSCp(D2cH(w5q{t_4V+$3OQLGfYw%@v?Uu@~<olo0R6h>eV-}+ZDL^1n
z6oj1_bamA`6&hjH1(Q*L;0Px}wqg3ecZ(s=hBE*=RS5sc7s_d}l^%4afY2~Fm4RoW
z#F8QRfo@AUfxfi7K{v2ufn^pdIcVvJIqdH<;g7dBK{D3Fgk5A7o_?|!d#9S$@GW`4
zd+)d4K~NVV5Uf5bkn)d<u5_I@=@V;XXRecKT1iQ+d-P6rQby^MsywxsbN_?ls_a~<
z-uaByUe;XSw;N=>Jb@v)Zd^2hKz|ltcC(}lMlu`Z@;NI`WC~8i#qXeM(cw4%^7=na
zxs=<<KlC^MWR0VKQ%#<r8`C_*t&tJ=mG0r^I@<1Uo7<~3>nhuL>6ZyExXqEc&+VqQ
z_nGKVpCG~ww&rxw^K!4h)M-#R`emCNd%yWtK=UMu2!d)Zfl(Avkl8DvGVGW;Ky1ST
z_SlX%xe+(NL2`(7LGIxTKU;xXQ-oz{!$x5yqih++YHSZ~>X-SvI5?fMZWJuA2j<qq
zonS$)!~4C6rZHSz8tN2G_Ya~68%KD&q71um353ixSdGM#NFE0^%Q3N*PxE3T)uJU^
zJCyPLPP=(yk7TBiPr@{n5?CHn3cK9#$kF%y3aro;E^Eojl+Fsh)E7{j=NQgCYh+ES
zbI>KrAiDl*&JX^LBxw|H$Ly*NYSJM(NfRj9v5^8c4gr&(*cU*qpD{py3jUMw7ei}C
zF8Ab3!EXyWA-rlNJL-F-uCNxc$}tcV*=r*$3TH+bXL1^nCKT=m-L-=j?hPDuoG1VS
zjUN4&)=8R@f&*<Q>#O06jf~0s&8Y8R_;aDqGNU>8c)AF?*Yi#t%PK&yE7;c?O{s(F
zW<>>*QXwb7TMsB&0uhDZO+A5ESTO*HS$dzQ4!(N%5A<nl2s`!}64<N_@JUj>w*%GG
z8-pSi0J(s*({oG;vMn_$rwN{_Ff}_F^t(je26AcSj}BG{CZ>f)X<Wt$_)QUt`^YJx
zDjc9-1CeyZgK)=zgad}zs`Ym$5PmuFA`aU03t~#|d6a!-`M?MZHOtMjc)%8{jU5hO
z?aggxA{k)0kmllHkBD^@=u%D3?2VqI+zr&1D~CyfbohCBQ@BouM)fpKE4cU7lM%Sy
zC69yEY&PCoCd(b7#Aj|qwvCZjOP85DTOzPvK^!V04KMVkzK~gwmt^~}(%1+4akNB8
zBBq@XNujIkVL??i;=VbV^>t_hv48hW8E7SI<id9nMXK#2W4#45E?|;~l(xu3nLy{5
z0tX0-)gS~3$UguiGw|dQE~lczy_xJ?pHv3ZywbD>D!8^$!BtQj7q6k^PrL^X_gdn*
z7w3%y-$zV+7IZvRwFp)x^;^>On%PmP3I%}WG}+Y6oUKHzFz-Wi3aDesMI3ErSH{dq
zG{&1Sd?w|Kd@aNstoB`MPfN+9$`*Gs$)rRZC5qBdVINl3I!XJ5CNH+P+axf!m8{4`
zmx+@b9vVe6`yuhgmy&m`OV8ao6}s2ON0_#TO<|3pk(cyaO*~a^qF^ZhOsN;0aaoTz
zj?J@Dp-=(5Q^th}L_IaskC=OSalZHMkf<2i!aK+QoNg!5c)Yh%{IO}zrVC8uR)SOx
zQ>+L&U_{E(_P=x4NWTZx<fYjaSK|z6VU2AYhAZDd?OSxn0%1D-z@E(<WG$7tWs;%t
zu3IP^8dikIyt3*ffW#)?PvIqX0rdaQlS8?+m23Lsx-NUzyb}~qJoYiW$@Gpy4FIo|
z&y89GAM~Jn_R!x2-_;bmsLP^E=dp$GCq)GoVeeQmPh`>Crv0h}iQVV-0i&AxKh+8w
z)Bm6Mt>WomN<=SjWTot43qvnQ#K^z^Loa6O<m^Jk!NCSY|GyatjO-l$V{7BImZCix
zJCa|ozF{T(Em9qCtNA6h5X>Bq(l276(pF@lQPaSRJ%lY=>bINs>97G?wsC@N(otcj
zChvp0mu&mX245MQ9K(q+vV(3cLLLHQSU%9qg4zB|EbRRG5;#}iAFpgI!ya>Xj>!OV
zrcTh_X^cCzRN@lgK%Xt|PZb6fG@B6%pBMVAtJ6AnkCdV|IFzk(bJ8hg3<O^>d)zGp
z>}83;SfxA__P=#%MlfV36DX7r8-N7k)Qkd}iv&`VTNap*6KdEy*;`m~Asl-KFBt2X
zP0;-tMfPrCFtWg$9H-zIQh*Xv&2%)hVgXG**b~bg*sTj#OdO~=dJeP@b)S<XR%o9S
zl>@q-7$y+Dj0hxdPH?0HP7pgIU;<mel%_tWE<+`ps2;}12%!p&o4kTaIN(6%kL0#E
zR6m&$41bXXaxDfVadm=B6r(IwZAr+f<2ceT1t!3{W>`=Y2{Leps<?v#YwA`G&jqFu
zh&xnOK>9=_Kpp`Z4N_gFI>rN#Kzm9sp>rE%ptm#$<S}f0xg3{7(~mjtK>ayn+#Cad
zl>sVdLN-+eZWelD%|Stugwh$}Ck@cUHYXd9wG6EQ>5S9_Z<Z;6n3oDf^Bs$$0LhXD
zHUtllLS}_3`b931DYuSklqLucWfTB|wr3g##m>c|`AVKeP{c=y?{sI_*vnt3_x0Yu
zwtRoq<)y@{HI?nPT@w&Ygs(4Ch52aZ*pstd&4#_9-n&d$z8D@Yf4sUU8X^xhR}C#g
zGmqibu<U(X!1}$4=bXJkFY`E>?~l!qqr`ZB-_)eOa?Vg8#74F=&wP7h)4=n_k-JQx
zQ^B9lCfl8Z+v)Lfq<nu^k;&F__menvjUdka^i{_9I>P(X8Wx_unq@-}oCf5Z#<vli
z-M&hmU91CRb13ra)0v688pqAw9|$XV*YL9*Pjb|TF@+GM^!uZxEcg5L1XLclL8$1{
zxvu%~SH{ikuw32OmEZSLE)Nd>E75B$+p0dde5~<ux!izsH49TTP-?<1iJVEA;qQka
zlOo|k2jDmN;b)$Y?;i~i$vmrBxOHY|x9;_TG-C7#utU72^=l)lvnOl?_`)BZIy`0)
ziZS!wK7$>3iRnW|G2tjTORL}F%6HBjyZm}7x*1d-8Nl4S9h#r-M<FPdbszkbT^3tM
z6c54l7!YH~lHdqRWDGf$F?_Dn>fE#K2N73_SZV`d`i~cIAI~uBNjtEa3cHGt#4$`@
z5IF*pHo)-A#RbKIo~_WZ3IF^T=$+ug|GbGl)=Yl_Hz1SS00LRsqYX8!mY%*%Kf6O;
zs(}!ed@;U_t&Z}M4FmZalQ~BWv{ev$c8iH`$mlb7I$>+UtoYT5re28GVAvP6-ddD5
z-4^3<sNXDsTd5J5Sr#xJb-f?FI>w~1GI`Xg_u;U_%_ZUKu?vHE4t}B~@1E<n1#SRq
zN%`5F48KU%HEBtSTaxRt7-!UaI5{)9`x`X-Jyf#H9>k6Is<D^&s&;}IQs-l3n%TVm
z&#9Ycd9cE2T2Xi+Ap5?WK62yO*0sKOXT1(#U8Fj$YDb@EvDf=1`}pAer0R`fLz{K6
zLziYpl`?v*%FCP2-c{Xu^366@aO&JjQ-1d8D>E;5g|co(M!mdgMZfv-{djOq-hM`Y
zHmiH_bsMhEpjLD1rp$S}0B3T`iiTJgzjc_KFS;;^u_=W%t%LJ<ZOC|$Wzy~md<B>d
zBV|E<EeaP_YR%vduq7I()DY)+Lg}x)tG>sV;>eO>2?|hVccjw3IlKP#ofr^#K}e9w
zgu>7bzyhhM9<+^22`2n2Y?=B8?mRTd7TQrEKE#G*@Dqi*T3tHW3;CeW%u-?2TiUZ-
zUp^#f(^e|(DCH)%ICAF_Pu8bLh3PbOj_ubYN$c+@2^$csesGo3`Jq$fV;9rXuZJt6
z`GT(Xg;FWqy5A7cy|u|Xw)P!(R>Sql2Z3_F5*|@|YRt6?_IC^3Ee-uWSXDZB<*h58
z+HigmCa7#8jF{EW(cdfa<DO^<KBl+%tdEdWrx$^}rKsavkOR=IVQ~*e*xDno`9<Yj
zcfc3j+i4I}ftvJ7zx7hXV`Ocx#{wqZ6}&LHGk{$z>!OIe;2$E^&S#<Qmfmv|FKN5}
zk=k019xFS1`b-$#pY6+eYQGL<@qtU@A4|b)7+1o-;l`lRaPm7$%pckRBzb^crQB7)
z)bHJHD(ty7&vW4Qytf|kIp(U39?QW=IUO9aduS2J>`^W4NRTpe$Bpqt`z11FoVp-y
zoC`90IIH`CdSn6;++6T@Na~G+*LrhO;pxrmU`)5o0E#f=4}!yOsSMO8Aq#EfiZ~MH
zXOQ=0(Ffl^y0fB%|L3e_XZerKGbh{sUOW^1uSsj;V(LUhFJ@!tVk%;4Y;R%;!^a2X
z?BZl<Xba=99;+&CfAAAqKT$sg03!>Z3n~Iy1v=eYsoa_b_P3-rKy4S<6eV5ACuW{j
z+=mKn)+7V-SaPsueC~HCacn?wd^%rE9i5;1EU;5b4xte-vWboa^Bjemcr<E)cZB6M
zgo-#$SUz4ZJ`a@9C^_NtdwuhYUA|d_4cYwtJ_vqlUEV9k28IuwU!@;T_S?F`X>vzq
zsR}KD&vajc;Q82>Ks&twYeSp_-3KK^MTZ{dWk;=yxpu7?o;z8lrRNCp;<oO9->TJ!
z3+KV-+rR@*rPySpz@Q2t!uvU8A_0p<VKmXXK@6JMbgJf%k|E-+psIfhjAAAJ95|L=
z%C%UW%fM3(L9yXC)C)UW%dH#ZG9u&fL?OQ=03}CX;NV#?799(!F(pxwll-L<+?z)?
zjPHwL#6hG@L9y5F_*>Or!R%cv>+CuIj%i4F;Twt`Xztykc;T-wExdK<z4001P`GdR
z5EM0A_=xSt6{w#=X2>~^4q5pNI{5+Ln&EIEM%J^TQBac2nuRal)qgH$J@>4Qo5eEv
z%?odRp{==kiji8=<Kmz?^>P$BE`3u+rBpRs?l>icBDNJKIumVxlgctEwC2&Qy+;wN
zRFkSVtG4&{_mMsfd1QWS<#Xe5t~9gbkYGUlrK_p&)pTQ9Sy-vYD{t?Fc?%v6BD|r=
zr596w7W4qNkS|tP);|)JI=iL0<YpQ?0^JT)Sx#}R;B=j=U`*>KwN!Rod)pBWt7UC=
z$>&QfX;9#sxbo|*pQHqliq62<m!0hg&PwN5*URIlxObHZnt<%lu&>43qk)oKQXmJ!
z{iql3<>ZyBGcDh(A;N;Y$C<1x*?`}rm%E^<`9x<;JOm?ZLNc-DC*E+*Lc*r*@^Q2I
z7nU4o9Af#(YwJr!QcUFJ?P#vvl^BkTN#nKpxj#v6h%^HpaXX_hGPI;-*!l2>X&@Fq
z@6Ig2iTuxF#)ZxGhjWrp2^lF5)$N3NlW2Zm?h#7s_}2>%dWor4$y&zH7Lo_3UO>%V
zoE{=PPDyZ$*U^#V5nv4WDzETn-|RqFWn(}LvA+q;4{yH_4rf>sL=Q+f0wh7ik$86#
z&X_*D-lfQs8%Rouk$6W0$lDae-P%vVPJZE5M99<a7}oq1(MZDZrb<;`k~{Cy%FM93
zjXSy%+IDc<W-f4h|1}6@+bpVXM^p*OuoT<qIV8}!nGqey#QjlBjhnME-WzpU!XPfP
zuFZbJHnyQ5Cc8(klOMuD#(pndaC>Y1b^ll&<$)XVKW+rp|1vx>GXIa&b*YAA`~e$A
z&r02#t%`u=UKxBp9+v=@M8c#Dbw|kw(MB_EGsb9=b*6s4c#n9LN>?WnHkv|FJ5SG}
zcm7n8@jw#c_3HWS*1;A#p>^IZ6Dp7T_yKXEXi*i{<OP%GCS`BAsoQt{X#Pj!;WNCs
zR7vmDbd5LfRI=vRadDDl?sDkQb}4PSCFqJ~FLJ60#7@C85gmQ(Qt<dda>rhTZqb(g
znor)0X5lWt8J=%92w41NlDBT*&co7cb8QtGWuDy|M|ZZNp!|4&Z6i+Y*=vP0-OO$@
zkj^HImriy&%|4b9dKT+f&f6|RDU+s@itmD0UeyA&jlgK1e%i_SQO3(%@9*L%Hhbb2
z_wr+&i^=tTvmY$9!>rz{?1t>xr&uHPr|aqZtc&3;MA8bv%g6N!p`WKlGYMy62ZQmH
z>))e5S0xjgy75l(swbHpAG`2k?3GhJQb<BHLXC(6W4D)=zAC@@u8W%IhKUmM+&H%z
zJ^g`4t{-S`SMtsq&Lpc^jS@|rNRdZwYL$R=9{m;1MgL1;^h4V_4hrmIV9}0#`C5!e
z=Q2mrN<J29uqt}l339pobn-~1<)!6?@}jQgKh#N0VWxc7mA#tmeuhow+d%?z1sd@N
zHUdE!jZ8rv0EP>$^vMcyMVhnE#`$178j$l>pdzZAKINap()<vB;}XGN@h4Kga7Vve
z>uUmqh89oDI5{%6Te}G#A`_hpl-uBH*GhKft}2&gq3;4t8PqW~9VcybcOBex_DP5o
zVmvLE1LPel?6&FFtMylOb;K3JlQ<aMhO-*?-wnnGu}IQ*CYmOxMD^>ey3by-#L)zR
zRzO(oVO#AX(u?6g`fEQ-5xK|fkh_OfK1(J0g_Lv)5;H8Z)-yJ?`={svW;D+_VAV2m
zJXHh9!p0m_x6c@yCT$$%y2Fqlcc|W&iQrOhjV$_SRj;k-TCvtTx-}#U8L^i7>a918
zMl6+>abTq7z^aZh4_Seb8Py|rRaIa>gK#)jWxIXHkdQex8^F8FUeB?-1u{=O#fO)a
z4ZE_HEtcVPR27k%my>sJGr?6&m9t8cR;0a&p=f{ebOa<rHs#Ky1;z7?6ue{YM*Dy`
zTWNxwTjR9z&eI8!_!)0kWcP?py8za(?12GPpc4d>YJ!HIle&jR?>HtvkUiG!u$f`<
zW$AORG|K%c2^+~ZALL+eek2j7ywJJR-rblTfw5cwX5fc@8(NAuE;c5ELnHz*VDEr5
zl7^&Qgp>V8tjW(aIeW-~yo#yXxE&;(jae`{{!Yp2(C0>2l7kv94#H7qF&gO$?wiOX
zh@i{;Zj%*gli;1gmlRYK+AG=!eXJAVmQb$@IRjDNmTWh59XQ73dPy~dnIljGmV3-C
z17;Bc1F6MeF6eO2C--c-KawvIa8pIUG~1*UFbNw$UoM8JTeL||otyQwDX1iP1_+2n
zY;7C(fLfVL8}>Tl)jKI1^^_X5J?h8tV&(WWxGj=sBYMJUsy{zAYm8eaZ`Yoe>Pb<?
zd#>PhuBHV%+pJfV0xa?@`wl!8&nRVONUD6oz{XY4%5&T;sDsH8(Qj=bm5CYam#v`E
zGuOk7E*pDi&%tH2=y?uw2QI4S9uwH1>dx5WG#-pk)7%0=do;7eoA8OJ)f6a1fMaqB
zY=~*1k$V(%g(vZGIGat8%fGBM$nYs5H!>a!v(^vl0~Q`;+?CJ`ki&7SK{#k|p85e)
zu)9ft5}D$sKd$W_)E-S-L*?K)!N!4XLzENbAzNCSzSghx>EBx{i+Lw*e0w+rNq&q4
z*9*V;i1eytIN@M|#D+rn*lAMjBmP}rKD36ti%U;w=K94qBla><_p&o#n}kr{DoQ{+
zqWMPsHb}zkqr@8-N*`=MamdKa5MXF;9$oyyS$O<&48g?xV9>K~OLHar+EGkCY`RY}
zwWtOZYUuP$+0}3HK_df2BnbaVQRE7cYiGbM4n7=V1}DF{4cexcnx$sO&9gu7#I&9%
zYni)J71XZCE8pAG0E2yWEAQ}ce>$FArDHR@5(StZjn}q0kVfORszawMkdbtp9y&bN
z^;R3dEJ8wB=?HRSB>BDVDw&!p8YBh>!vfCylf*Onx!&vcJtSFzkt`7E(V0qizWDhb
z10J5_X$sf{RP&T)LPSkHMWj9O?phf#r%$eb8sD`?RV-rbCdWv$E`>403qV_4#iAOX
z{eBsFj;x)}f%<AM(u0&q%KT~x@Vy{SgeRp3P$&okK$gCZ7I6s^+&EFLdKrB~{f6o2
z(7?Fk4-xD$j%Fnw^x5vPwTGxbBWxAQu19OzrSKQihaO75BL%Y%)-OH>Cm`;%1z$pc
z@-B4!8=C>r7CbE_X;N64=#}0@NLV6R&^IB8W-d5&_|tZlcI;2)r?3%TsrGb*0!r;g
zf6?iF8>S3s|6avzE(%K5Pht*TA57LMFk(@K&BCBV>en+MkY^!_D2-<oU+7IDPg2FE
zS1DP?9q-i`jE))S;R?>I1#mB=+fHG5wj;})`X<er_dedLr%YGVsFQGkfV&EB?7|eA
zpyxMiJMI8^-mO$Y3snHE3Udyw*dI%o!IXrEZ`ymNczB-i+u;q<R!nEk`h$o`wg2&0
z5iEjG8u@02ltf57qUH*rI=-ge=~l<!`K9S|C(6FX&92Ub>S!SoH%a|@r&b2ER#DAW
zvJeV03AJs5fe`|@{mR>>sA8QvR(N>#y8jcRvpY|ftQ<N?HP;8+tvcB6SyAGY(nEMB
zPF+lQWpR54-IF{r(mXotX+t{`Hc<5UiWvw8<$%>;qQ)(y5p7=2fYBf1wjxaKzm&><
zV9S^ong7QumT67>pMMt=|Jm}n;-Hp}56VcP`S2~HrgqTUZtKl(k~R(ilWjGnbS>HF
zxbGXTlZ2)cp7|ona8Jg50y7?hxgT?hWj%t$?d$%*{{AO=v1LLM%dum<1~>@<2{tNq
zu|Wfz-9U(g6KA{;j$eDv*TxCq_ubj(@KKI9%O5=*3Z@SOKa701*SCL@`!_drC0J3>
zfI^bMWrZvTXSxb){sJ=Dy=aIjx+ye6CU^|(j4V~@CIr}~AB0zGD=&c?t5R3&NPJff
zRpW+Ia9U90?O@g7v;~YiN^9r2L%~&}9AVd+-pRXM8*|%$3X@$lBQjuBoo1?MRlpht
z8T|KhCzm1UYDv_l4^@Pc&y+J@Vxa~rH2YHSAYGXRgnXj={e#dZrqnRD`^TGm4sy)8
zI;*v!wu+%zzT#K^l*~hq?)n#f*{i1-wmDeaDr`r(dvO_sBI6Uz{g<*J$&BKCr;s|J
z6NA4qJQrdZOOQ=SlgU7qWB}Yrk6%SM&0RDv^=Od1-M|7PO|haMgvN%To}G)6N}LjM
zaKrZk!j4r*%~kX7&%^MKH0QAErv2Ah^XuT3N*KRuc87Wj;zJClq@ZJUg&l&Zbqxs$
zG65U-%Q)SnAuBO@%{qQ_tqf&9Df3e8S6!_Z8J+bONlO(snQhZdX9aPVu#3-lfPCn&
zMkdU=>xX1UKqN$!kjL=STJ!U>QeEdO1c=2phO{{IFZ5*;A74uZsL1WX@;rV7EP}t>
zTPFA%$U|=)bcoqT-PpBkkC=r}>U3p~p9pQ>|GID5Im-1&cnKEglPwq^ZFZ@Ie+mcg
zZFbIy^JO5jI5_R@ck^d>S=%{0RtR#&rUNB*WjCh(5F#E259z{3;ehppTCr;#wZq;-
zP>rDD6NfElqV|)W>$HFBxo5t=!FR8$7Ca_3eC};^MLJp$p0bFb{=J~su#45!Fv?WD
z@NNAom*DDSfharAxzYi&B1zO7Y`IYrTps54^#R$zM@#3|d)h4kgdk9ik`w_LYr_nq
z+lM-?4uD2sCSc*x(EYfGr{<h6CoO=$M5Cf~9J<5m8(=KmdtPnd<~ApSiQ0=dYCnts
z`326N78tv@D*Li!Eps^}voL~{B9enbRVqB|+rt=FPNX)G`YHv%xUsKrj2z<q(6AWy
zQZcs^@N$<vyTVMi^;8&|@U&}A_uNrzCNV+yPw8DTB35!Y8bE9lBg3MDQK7O8Nz{I4
zXS@L9*7m5BJ|vu*q^BpczvPh=)O^-w5-4<pd|iJrNYJ<qLxttw1Y(Vn@B+;L52sOa
z&#Yf9TBpmVb6^Vkk1S|Ye;X1Iyz>g$bM3+X1^y*mPkGrc<KYz4vYuJ4aCECRoof4R
zS2w7r_Q2e2l^qYBKu@*P)x4!^BMNYKfb$E?MW=>+6RIebeB>#mCyxlI;W{6=wo}nU
z9;~}h)!0FZFj<#n6G2+FJ*a!ZUnw&G5N?!xDnI2-Lo^II$P=ggA4qEqW0QLE5+<S_
zJ-W!LK9ENOLq`Bv-JH?y>uM=PLSIH7nbgmW$-~vae}=ukDdhZ|C@2<%`5#5l9O!<Q
zZhrs+KED;z-y+OF89;jv15T%Of%)M2vr8L}%q>w`-#7kjUkk|an>-nZO-$)Q+Qizw
zZTPOCJ8K8I;FOd6)!+<ZJwvi~5XvW&Dc`zrb2U!zta8A+1yD$F**4USD2*6lxMzT~
z_~zTXmlg3iTN(u=rpSeu$#3R(<)sV-#d_Au6+_)7aVE5qd7uBYw7BndNK1l3aIc9@
zbY^zmTd)|DGfxd3K>N2KBG$s=?Z~n#P#kGEOaJo?rR~eJ2Xg9BNs_%dJfE-=4t@YQ
z!CwzF2H!RG%iO}(Px#yuor(Qu)MOrEmTRl^T{UNjA?+WNd{D<jn9Y=5$ub3k;*;B^
z$a8)ri7;lp|4e<JOmG|L_pQo}i7ShlNsN?u44?b+jM4;-kh-{NTi{&|4|R!Q<>6%1
zL?x7l$PDGw8G~gs1Tx7H8aAxdX7oms(Tf6dI#&dD`RH{?i%I<qqkAy8&U8XyPlS4+
zH_{4fF<AD;J^s=4`<TkzR`6ODGe3mNzQ;iOsrw7&3kj%X5uaRFg5Ev`aD?4Cs3|or
z7YDYcHq+t!7BtZCw6X<U3Gc=*d_W%%94b&7asDI8HL~36Gvb_CgoA+n<BhlOk=>jS
zL-VNnFO1f~<nTh_R_>^rO+1Yn2^svQ`13RZNnPN@ylMj!NtJ~ait-t{K<LGdKw$m&
z{T&1lW@XYUkElKeSf-U>_7DRM$K=D`k=Nq4FBcc*Z}gPNFD*j)g~a}_@UNcmH{UUM
zjJ7-ar8c{VC^{i~<iVZNBmCK3+V01U|H#PZ`tMl}BQ&Vejc$==|9oVgKGV|`ubeNT
z8&I>*)4?afnRSec!U7#vwrSXl4StcE-3+z2WwZSGc>(TSl+9j)v_Yq1%fzK#UyHOB
zEAdDGkf4P3L4ryHVRkI%dB#q=<9MwXi)#hb|FX#s)k#e<X)bwn-}AYWLFE`jj&F&X
zpRrVX^WK4F0dA&112P2sL|(QJ#qaa@4w**#ji1qAfAO1&q>ljo_D+v97J9z`;>#;^
zO)ho~RO_PrWB&UWhsv1~;3<+8{?VY!0Lf@9v^JfibZNe?n@Qi$U@R%&INWGjwq@yg
z_7t?zZtPI!)oLe$^4d1Ghh5loPAwQI7Y!znC{hOENE9xzWxT5OG>ZEdSy-3-6$FEp
zQJgpw8qa_R-T@E<@726ye+H}E2f8Ru0C?=7MsIQ=!zQAtNMH`J@`=lA4#=nE@O~mw
z{4V0c$v5!JZ`!K=kST2cwNPQ;<oKVq$Nm240RNFGJ<oNwv`lkmKa?kt%Sv~KrtDz~
zD5f6r=+O0|jiFOR2`Bz9-zp7JUY_erntDiWlSCRd^ozU1yJJgK`!C1m=g+P9e5Dq~
z0TapAH9~gkjZ+p>$13)Bt9$R1@{RAy@5cj5?Cjwh5y`?95CXr2aDTp!&hM8ExAaqv
ziD9?$CXTm1JI9s}0simt{Qe)N-`Jd6!+%)oQ>>JMCS`~xb$2_YofI=-I0!qBbvD$K
z3~4Rsm~4Z<K+J873}K*4JV+H2m#`F1jhgHYM(RG*Fceo@iZ8TMRK%ZLn{8PY%r9@V
z`{S<KQx(s&HV~ghaJllFsl|&@eD!<oCa%XQw9ZfuC7Mwjb!tK#>ql3pX@}*VsF<*H
zX}MEqUW>^YX9`m8q}HMUq|t9H;3C549dnja*$=HX_E40fIk!<X7)*aJRF_NtAI9Dx
zNR+OP)-2n$ZM*7}ZQFIqwr$(CZJ)Aj+qS#zjUIGFe}jLJ!yIJBj@&EXz1Fi_uaIwe
zVXn)qtJsU!4e*k!ooUtIc_+p|J{i9Bilvmvj4hTYYG_g$qFOZZ_~hMHzqFUpxgjMS
zx2N-Bo?=>7;igh8<0;lhE5V@&0D7u*aw61AE1Pe)KMem+7LWQLI1mbFNk*PFT59V&
zKW(5DEMfB*CG%r_cswKO3*6OYlRWClPx(BMFi{%!Y(K2w4cGikD`1S0{OzuRJia3b
z+!>O{qrg?tVtV{Hu0a$v89w2~Ulr`52d$kd(;gB1x#M~Su7PSbuy7O3-f4>in;7l=
z0zu;y*f0U@9QI<@+(bFDjeAL8rbLXMd0j*$tk{RnN0#6H(=G4N#v-%UyA|h(4h=fy
zPOkLou)dk=w)VcLP_QU%4r5#>K(z7<X*jW9>51b?o`53?rnY8aZN<e30fB!^vacS4
z);R!sthat$d_*F%Po{sIG)U_{gfCHWY77o|g|ewRJSRsR?SnUw5^Kb>7XD^I{zsJe
z+Q&X?-*hcCV&3MZa=s~E5=DdqMuI7T$AkCcB|FBzmhZm7^~lItga5wD-KI<Rm7b|^
zI}%%Nvbh=A)1lm^oYO4g`^gs(@L$p1vHpySo<lHTV!J_C134xmhlO0doq1*5;oX(0
zHkthC6b#XKn6no(jtC<tG1p-=i%C!b66Ztw4a!QKbkit|4^uq045UdfYux#q8uYld
zIi3z!Er5JiU<oIw(rM!G%i+54<lN^oYn3_<Aq*?JEpuM(xM@kUbh))eAg8Soe~H;n
zm?9?x1(Z$=_xvQKLUFk`Ad~#jT_kf9tkr2*H%n{_dBGp5y!uFlctxQXm^mE-$Z+S!
z*?D;iqw?coOcRa^;NN;B|2?zZe{^a7;03t>9D5hehPZ~!vjU`of3ud+&lAkwE>;h4
zUZ_3GI8Ql642Zyr&XySbxOnGGH`O#fpt^=rW74&4(dKm|@M{np`v7T+-YgOFL~{DX
zh}5;v#cfsb+g8(_Kd~QCi6oUF#m2D|$XaA|bleRkk3h}8<?8`&oiIFwOHpI<ruvap
z0t7C$YItvrW6&SH0Tvs_jQB;zL1qQ+EH0P`j5w32E=ltPNGXB?lXjGDfFUeG>1`Le
zQ<SC>AE3tNP)#q+b3$gec12Hs_qI5p*SvxB*N}nbfrzn8eg`Um$Taf$wg^)bezF8g
z_a+z+JyBY?;Z4i2+Q6fSSRC8#`qt&G4k)%E2^vaO`Qa_xP0*ia)XfRLgg*WRWP4#E
zbQNhJ${`>i^HBtO2ibD&C(J>NNg6W*uyz%-2n=1b{IuHHf9rvIhM2{{Ml1P(^3K13
zT%oY;?e;7eGY{)$Lu|>cAXv$>gPTu^fCq)@wuo!X?lwd0fb=q5j?2=F9Gr<P*3C6p
zQV*~LEQ2(?7|pEK-XkL=9l%5v8x~Ku7Q)SgLX{JKl^=`b<*LJh{snO}e(9WtQ({#b
z7w=7s*u<S(Mr&t4XPj{GUXytI9RK12F14=uE?QR9%IcA6(t-921K|1Mt@UXP5&(=M
zr)eW7v=Rtfnv1cZ#=1TPeCRgsdy*A_d8_c>qw6@vwy5n#xy<sfOmIAoEo^9QgOl{c
z0)D){nK}Cvc~U4XxmR$u0+fiz=@#3~usBlXDUo+6Ljp5dUmZ^sR<&7noaS0Lf(^#S
zCcA#zHk)sji6|wl{UK@Q8K)^ZF?yY!V9P@9;6n|^ZdlL-83AzDD+xCZ6e2myCgp4s
zeZdz5KR7kNpPUd(L87L#8T$a5Nn)C&FJ{Ys$Y!LZ)-@@6g7DFRWGk&TLrgQm6~=CN
zX)g1pGn~fn6aFuEDHb40UET~Cy)%d)8XHPGO{RV7Sad`+HNfY$j4hg8Q|z6nK1@=<
zFs(AOKu$lERc%isE~pOh&D3*DJ7NV$*N=YwO@{@~ehpHMk4)#pihVbL4$T7u1G27|
z16DcCMYU?^*t%}dW|xUWsxY0*4+Agp2TrBx_36d!cgv!Fnq>9jhIBTbu~xSd5nC0N
zxwJ|U+xs7=DCsUvPT0U63>i1=25r*R9_;WqZ<Mhf<UY_^=V$m=qs&ftxPGsLruAG#
zeFAPXM&Q{sDPFA(Dk^%sj7eAb&z=Y89|H*2ThDg=W{8!8tbcrJ=AF;2+wXJXbsUGO
zmD_d&0oYxfLMz#WT|wKkw+=uP{;^{Y%!R96-~++vUW3~22r$uNTDVDnwRDlDpQQ{!
z;)}vHUb4SVF?gnYT^#zyvT=suOYk?tjTeXWA=1b(+7khFm7;SE4%amB41<Rgbc8JG
zxYA$pFEku|<Y-Dp)BCF&32te;N5us9I<tA<c5?-9jC|SqzL2LKkb?Y|1o@9i2_q{9
z!~ZHlrv5*0a#pq|aDlkQv3~PJ>2(RNXlF-<JQB&{BqG#=hU(9Ex;6ljS>+A4!$EYU
z8Sk5Y;Ih`|vKO0s54Lu<3A4)~5;*tMDb*YZUd*5%u=z~dZc9Bx*^(2cqVJ=nr4Lv9
zKV5GenK_OUUzKahpx?#cliR0zm1n7IH|Vy&bL8Zd`GSD#MPyGIkgT7S;i#$e*x!}S
zu2$y@My}oIGCi^wb;WoOK?F&R^kJxwMfYf#;dtozK~M1`LP%7#lHE&(wIk!J=A$4{
z4;XxkyJV(^iBCXLqWUfJ>hyJfLGgl}D&`5?DGxO>LXuw}<{zHHjzIxvui+T?4x)U{
zABTQvKZ@}DCt3DH>VDd=Pv|=`y3q;XP)tF^f?5$+P}30d08)PQ;8Gxnkm|+<C;LDm
z9WwVi5mOGVSD8eWcX~;#tNNcPgr+X=1(}HB*1{4<!w3pgWHdpWYkp0Q*l`N8t9nV9
z*6q|Ot5OwJ?Q2CcEq;83KJzk*&MkCo=5zYJHVI_BPKoaBmJk-SPs@)Nbii6pQW^7~
z{*5kX=r{T3L4@HXLSw2XIjOE1^~1{3QX87dOCcDhUx%APR47V}lQLxDqp#)(jy!ij
zV+0vS<e<8ojL-YVZ+8FGgEo%>?6C58Z=jt{xoGF9#%AhD4+^Yth~_=M+&Qv`Ng@Wo
zWkU;4c*|$<SE*XhleG>Wzs|g!Wj6{+aEJYVS-f-;j*r8euVuNL)eqyQd&K!1!mU_$
zh{n@OToM!_NeDj)X-CU#myq`t2G!_H1PR>Dx}42?1+tK#5>5>{X_-gv9kLG0?!fIH
znW}Ah^&s^e7pj>hSH@aDS2lg75ma#iq8mzCg9=_0pG3PsLSt4#RV6eQz|h#uF*I;p
z32ehbv<N`bkS2v=<3Ix07ckg}Ff@Hm;*T0I-i!fIA4m-pL<)xqtCtqUa?6Y6J%W@c
zqD`GB%py%-5~&=S`i#g%<9-c{6rB@OAo^gIWjWbxHN1$Dep-QLrr@W;6u>YC3fYs6
z3nQC67UbT~Kot(^oA``TTfpYS`~7$Yv0r_@TB;$m5s>3sw$TYszOPXe7_yMhIt{n-
z{pDofWK;?v?TCB&3CuX7SfALoE>is-paqG#lKx-mL`+a|ZmKj3AIr@G5N5UGRpmn2
zWEfqmtYmS4DqKR3zUe+CN$4rcsxU6JE~xTKe5x84X(vbfw$`Rtn%trK+s2Jc5z#@g
z$w8GqixB6#)cx<Jw9;e~CVz>bjuUL!Lcg7Q)wmwH9Qpc$RcU4cLYpCL%KeSL0p9M?
z5;5#ts0}kVtdZ&)zsX-aWzc0R>-mD7k7wJdbHVWz?Xf3J-Bk|k7#IUPqhP_^Zr%^j
z9p$6-nreE172XZSzb1@s%s;_*75lh=BD#TP=`G2xRXZ!@B%;ekN;3M(MY3rZ&NYGv
z0ufN)=gI@{Pv08@<D@vmII-LkSyWtp2%!FC85DWh72NqU;)YztbF?PC#FlzpRv%<$
za?!zk9xd?~Ds{7bzJ=DWnNpRT{B`R`aMZ&^`ixHVJ*{^f_)S0!Q9WFN|6b+EYq^@~
zE2+J>OMiWpGvnKMpIH4xu6g}IqZrnE$#41Jo@_d4D~Ev+@r3=xzLDc|WIDq6#QnnI
za+Y0;z@w9LO`Q*b@SWW-0}cpy^g)J+3k)B)7E_CVw4L`HV!uQWs#8n66R)pAWWi1k
zox`gGS?c4-_;y#&ZVVhFH`K;}RN?brj0ho<1hgK$D5g>6)eNy|%<!FF`;-7toaaQ6
zIc*gi$V`ol^E+AqcN2z)UHR3a7)->{cA@I7UUirr_aSBj{Y`uEY+|+FQo$GWo_D|u
zEZ9u067_Siz@%^O;j}4`?62$EEdT;qoseLaEv@Krj;EQ6uhAzEU$Ohk3n!bWMC$JS
z^quV@0QShDu+wNZKD45z=(zH>M8T_WmRZY8KMpw?0|~B@-f(MDjy*P!>V)qcsx!Ba
zJP;J1{)up<_MkZ{ANzFCMjpkRDq3wNcY2-mY^c{HtV>Z2bs~J)CV1Q;sgYxUY&U8O
z*b~`UGx_=v{{U2qJ+=ME=EKhXAK@cb&i~y|ps|s#(T3=qTf3`O7A<lhh#=g4ziZr`
z>8!#Q&a2{nYtv2;!90*g)Z8^Wq?ZfUArPM+da<^s3~MjuTKMq@I9RX|Ch|x3>*@Xc
z3BEi^+K?$;WQ9F`k|_Bvq?C$NcQoG+m)4iZ*Y^{AR{9zRp?g765bI}N@7LE=Z_SRk
zdUdd2y{xKba{cSUlVs_g*2~TN*6%r&*VkON(~#MP>xx>hjy**~n3yOT%|bj9*BnoD
z#MV-&o<>t}&kLed)7)GjetV>FMFNkKd`)4j+7Dl$fO=%*U>y#q)r*N~Xoa=v#U=(W
z`jyR2%#=&>aX`JBmCd?`mQA;xDd+(W;<<Xw;vs!1G#Tl55k=~j4K+R5dfEg<=v}(g
zK|2a$YGO3L+69d8*Mr{eN0X*hN@;$&=0G09Fnwp7A#Muj6!MrvjL#-aHmxq%_`8;{
zMz3<xmOA`WPYu)ulX=evnt6|BIm^a*&$%AS$c1hE!IRVx#f5_!+kntgDu#*gqV(9L
zERnZt&*26vX)faJQHqqVZ=#vwMmFp+N}q(?7ZQqrBM%GDJ`bZLB#`i#U<x+SpfDLU
zK_p)_C$PVdMa1+5z*Fsp7L)MW<C5KU0u7^R!lfc9rWK82JVbxt1+LYvW!jqZAo{9u
z8&za;TYq7spUt<R<-VRqRG57QH3ZJU!`=2`4i9V>i>O7NR6?ZdMwKUaBYI1@IdS!6
zJ4yzzB;pmlM7@_j@*I75aQ*DC8A8&pnrSxHTsg1K`KxJPd$887{wSt=XAb8E1LjyP
z=*qxI7N~j4u>SYCNrv7X(kv;u+jPt4mn}_$9X!>|Y`HtBLmMu3+(g{|4N4;x1D29f
zF>P`dN~TOt?pd@wGq}JyduDb9sLi30S!)lCHDk2cdLraGP_>`8m42i3rabra*vlnG
zw4Cj|hLVKAnqL#Jx3$cI9er=ZErTOhH93Rg0#CE^uG*>D)SG%@1fK75xc0>wOqMvQ
z!e?`~qecBobgGmN1w*~@pOuPHvq6RrIh>UZbE$jcECgo22$Q1caS!J^NzLfOdA8_d
zYN_e*&`5r6aonjTm38zirTK36!pTAhVy=g+X1u!HM!PC$fL*?P?lt<;CVyR?PtlPa
zBW4US!UUaS93kj(K&v0%FYRU9Ic{SCB;xCkU2SJ_nv=fSrX(7Otr&p2OemDuo*(^k
zV?L@4$)-dke6}f}o+F-KILR$l-$Hq};amRf$x;fnm-4S}#De|0PEk#7yraW%gLZks
zr_2R;u;|X)#NMMbH0Hb#mhn<H(Dd#Fbv1uIiqM_5o*^;_5ROU_1tH6uqVxdaa6-N<
zNE?W!@b;F;zkx#|Tp4}(5hUXZh{x1(Qh%(PSHl_1hbXmA57;=4)%B{SYOK#$%Zm`Y
zi&SVj|5iNsQ91axqFc2U)kB_f@#Oe%l@0Qs<$;a|S@S_S1Ex5Ka%?};qAjzN5QQhi
z5RoxusDE!<OgaD3(*gMO;tu+;!v_C7j$;qL242AxTjmJ<50M_&?9%VL2_UCa7$*rd
zzILS7Bhh8(0Yi_B>9tui?cn>I(}2z)^f$G`gS`Da;_GMH=91nYEOf$egOq(<MqP+K
zLi3##ml#pUr_CG1oG>U6nF2qk;a2nab3D6Z97$l)s&1=#BFf1(ZR0jAof2UiI5@X$
z__vsY=sndD#c=MvKRyrRkn>msdc3==J3sjc{{W}HoRf+;9qBqRC@qn)`hcEPKH;_h
zh#A**W`3cM-2b`UsvTzSK525Q1>7Rj*v?;9x{I2;Y$d7N2`t?NgID6dT83Mmd0FV*
z&YfBg@+A+~0D5>1BzpF0g__`75(|9KHKP{fka`N@<D=dWAyrX{%6vpiZ^ln5d9|LB
z(5e2^9R6`D7;Hw#cr|7H+Kk8`_~NB&<jJ7tVLQwS@ySf>3dI#)juBWWE&O^uCBKTn
zG-M2ux7}aeKzP^3!Os{2kux5chi`prJ5!ZO765$8XaRsd<249U7!Rrx0``CxO0aZj
z2Eq!2+YZwzTO5EZ6UDI?66u7W_@1~vmySDQ@<DjQiL;c;^;G*p_+$_&+*F(_%q)Um
zoRg7xM-=6mrPJZB;izG`Y`2c-11OHQl8Zqw5YW`~_Hf)?3FC6;_Enh1<dvqnrYUj=
zi`+=)44|Hp%%)wsm2Q%nVj@M#^g;wjz@SteP}_`B_H9<zZM7>+n-QxJD`;E)M>=&l
z<&Th&f%#*IsLC{GHs<V_@U7gbM;*8ju+ew_0t5Sy;|~xnyRi!~<&f%>gwF|+j6+A9
zjCRnFD<Qkq=OaUNPQzzblb{{(+fK()JbY7R`{9#(Mwf2(F!TwJY>qFmQvp<4$_e?6
zufy~7uo9#+;~#fZ$9A=Yx4#lKFxq>7vs<V>L~w`TqRc){bx;oJ_DX(ERLm3TNCeFo
z+&t<xRhpc$q$L*}^|pFi{QOp!c$zqEFFv~4YFDd96TI4N24o2N&VL(Tb9J1xSp9ww
z$gi6mUFBr=jLct=*xQhEEyvK?_H?<Gdr*3|eZRdlOd#{h`v%Y#@My~G^pj0zgG|B?
zl}K9B?3CL~1Px3xE&eQDU%w3NbSU3iCCjn54s*wMuKQ`-wO!U(WaA$sGbZoBf_#Qq
z1vi62hny=huZq!hLQjx-7ISK^+y}#nfsFZ%Fh2uhAb;mHu=U-YTE77{UXJ<btNCFX
zDyq2Qyu*~eyYDn(r`n4BVAyxS_qAt<uTn>mwNZo8>B>}fY#=JZ_eYJiSl-+Tr*)RS
z((fTAfoy)3@WLB3vQh1kX<Gh-xbzL3=s@@_`XS!ajA6uQ{ZutYaks^J=8cl`Ou#ch
zh;*nOG8O6Ad%3sS`D}{o_gd6nt=rE)TRk-30o`izqh;5+#d!cZlOIJxo4qgreQPS8
z`shm8Zkj58SXcQ$CBsf~x}vYZ)gHDtxs1VZJ@qC{lK;tO2JB9$naMtqhMSI_v3_`B
zGcgeR$)bKZc=@|?9DVq%*=1)qg-Q7$3Wb*7UKcl6>S_QEZ65IrIhDDL{oB^Z5VFnb
zjb(<(!gf@h-HSSTnd)z!mVujjhyfOd;ElrHZQ(VXun@sx<LBd~;*UBiH>#;UByubQ
z4+5ev;l1MY)~>IQQL|gal6y3JKhM_=5Xv$U?Q??z+v7cR5@d=#N{x_9HFy%79qc!8
z;1K`aI_FZ%Jhi}_$C@Z;tl>~pSx8%4d;l=-1(}b@zc}}}1Ap4C1Q0#T;MmhA#J$|~
z5yF!<=xRhEqu?W)og`aSkQ%Sls_KPC6oH=;)v~MpvfRD$yc*agH@AFSOyT-RC?<|b
zZ;!`nGY0go%3LB2fuY;;9}}s@%|GnPj6GLCpVwk!TR^M$3OnFF;&pRL;=$wLeCJp<
zkyp@>N-rg_ccr)BRFvyi=%j1ke(V^~|4^O3!;SxlymS1|&8yuSJB|k}h(0fR{bKFS
z2%lhUvbyuzCJE-X3*0kK>ry2`QcY8EB9uhVF_~jO+wj_PL}ro>4(UDW0B!1*bS}MH
z!S45Qw}%&%wKlaqZq#lm#x#xdWUVjZ4dHVs_SOj-&iD3vX@>t7-kIs@>+y43YrDOr
zOJBSf+535AiS_w5)d|0|`Tg+zv0>ZmR=Jdx`U~$2EeuP3U{T}2nbtTiVjO#P{fWzZ
zfBas#_Hguk8c6P|=RpitjDn6g%Vs)}ZdliTb2foJT^PhHiIRR6aqxY5(W5@hE09-D
z#r(z)$t*~I`NAzClQ3YfeCI~USX+$Wb9v+2L9wp4B?30YPQz_YkAkE?sQN;l>k$kv
zhLkdkCzB#fN6L7<iR_Aqk>aDHJ&8Q|^R6=MaBS6ebfSjOxs(Pj8i8T`TldFRS(zZ+
zahZC-#p^~1%1jjbWcKhSG^gu&{{21iz@|275*SJ!h=T(s;;Sg&AWXe|jE|0BKHbw>
z=$KnusKL;&G>O4(`!r*}D6vb&sl*>#oHSWsH+f~WCf~Pk<uu@YhkMdM_b)Zer{W42
z*5$)8p0369O-Cd7ikWsB|6E1O=0cX35O)F0r-((0H2bjCRi7vc2a=Q~v7q%ahQhc4
znWI1waps#XO&NI1b@P^!4=bn8v*0ArOVkPkM6Pit)gNtBw28<>eqVgb6do0*-W+Kn
zYk6+0uG453QtAZC26?CoBA&jpGkR7^a;RtuDS1!7kk`xmGl$&L>k~~a^60Rfcx%_K
z4?<Ms9#DQ6P{wIW5i1B3PFe}$4>s<bqLdr;%YEs%G;6dQ6u#pCZdQk6AVVDQFJx|&
zfzGonB+d@@2!R{UWleu#!Dy~sm2raCswqVyaTuFE>xf24c2J9}Rc@^PPRaHqvy19S
zYZ<<&t^+O7daKEOYhLgpEp*O9i{-1vr+gLlDPtG~xO7l$(<Pk(>G&<6de;;S0P#xT
zdznWB09{~Lcil0x6GnT8o+-zb7`!b_vo`V}2W<PV;$I$*_nEjOB?V88(Ml?S!G#{H
z-)WBoci*%UUe69{<>xF4%Nm@MK|a+5(au7f#*vCWb7kq|?mZ9!_)Wq+QrV|y8j);(
zBu;9Vlnow@U|o1B2SGy==V0&s9iYMVuAh-$B?h?R9jTZ@0t6~#)X}w*`o6dX1%iCl
zD<cMM7;8g~(48UDBE#mWt`TP}4ga5_K@(|i_yW9{$yJ&Y#>Jdr`Ra7h0MFnnIzx&G
zlSn}drtGZxa959$S~i(0eI!;(qq!wR;y{M+Z^ieBr4y4O#q9ak(h-Lz@#}3d$&W5-
zHS+zlax47S&=j(u{1-r}3l)fb7j9E;NS^!Ee>3gi``lhLnVe)&yAi^-N}J_BN}D1E
zVpdHU_4cQ(G4+(y=}S{M3!f&fYPEKrm^K<ZE}mbRZQYWXGtxv#6G5oCC(n7{&}Tdt
zs}DB02^*b}HaCWWaT}2}TG;cUEZc9e#j)_CMz<Q)@!(vpBM{S<XhZO!30)oeOEqS&
zMz8EXxJZywA5<<f3j=gy6XqRUWbQxl30*tgS2KAevA9ABHy{_yam1u7>u^PZvf*^j
zgpRTAJUb4Bm!%mSz6E-C_a+){jGymu&|k)$JHhRx*lwanlDoGNm+*Oje4>UuOdn#N
zx_^Dc2dPCU;Gp~z5FMPgY$A%$uj}&XBmBxfbgbDsg*Hq->+-8Myn8GjIxE7;Bjwzk
zM>hh7W?)C)=sa!_XePV}O3aW!-?r338!SGA1Phheu6nw-UnPs%)oa%$PfYXtA$-=C
zbg5V)xec#EbDAZc2QzveaTS@#So+t#DBy?aWh$s(_>N3Oj&Hp_th>S8Uam)X)9HDU
ztQkG{B=+&n%@RcxLO8SLvGAPg4H?B2W=k7%RYIKwE=UN#Hay=Yr^7+p-tEQI&)Y+n
z`&}!~#HGm@<s>GZ71(>Kg{6Ah`gzN%l;D7|&jnx%bv`BI7T{MF^REBp7{1P$s=##D
zH;U|v5?;H4xoI@xUh5Y>NPVU_ztBSOkbTqBh;`>l_<PMGCi?EkQLQWV@`{GxVNZp%
zr-(D;H~DNpg0N6^`styvvpjiGJ{P7&<<|&9$7|v~5Fvkeu2<ijWc$Vmas=PX9b6HE
z5&~e>@LpaVQz(QcKibU#QfyhVpTW!@)Im3Z{8HLX-~Nq4J<)g^UfV@f{fXPuEh^ky
zCRoeMQT#rmFDmu$q+Jk^Us*Yg)IBjb@K5=6@Gf|g_x|M2f*`mAi}(S7a{D|*O@`#`
za<cwA3BR{AE4?#We`_)?pT<f)UKCZD`*F-Daa!(cmB9zn_KMr5d4@06#)~HHB@3!d
zlTHsWS~Lae=^--P&Ba(jSNhaDV#YzKR%W=_gdo_;gCZzc;KQQ_iC7C0EJb#}o+J)j
zF}wd#If@k{@$e7BP*q|TUy#;l{p|&#AEn0P@s?#lN-|V%e!{S`;!#?s%)e(pixsGJ
zk=n5Y+rfv07WzmV0x-WxWXJC^DJPN`$IMilsRwvg%BWC3Lmg=4bf=b13y3#uLBlFS
zV)tJs$i9tL<X<`{8m4+q9b)`QuN78(DafFuc<VxL3{&X}k*eNrBT@&jFW5!!)VUD6
zcld|ZKU~uMz(4TGbH|9c(7b>?Gzhmw{C%<gh_AtfyCa5p&vJ_GVV0$Y#16b3oWc^o
z*-hxn(2uc}fdLgy<vOO``(q_p_4kmRY{peI#(9X|wJWePA8aJp8U9taU=8`5d233i
z*rJnmP3)5n6t0_$iG-0*h!qX8JzexDk8OnhigG3@bMHE-XZ45NU!h`}JS?H#BK#lQ
zUBs)0&gy4UxpO01E#~9$+80J+XXM^%P1MVctZahkzF&JSfB#F_aQug|VP<CgpAU01
zG;9yqkbGWhXA80H7kHdWxS=^R(=XxT^jB;yC#>|#mK<7#xwyEzNPPK*q>xP7ud+eJ
z8T#TB(Z>vz)bv!wRyRCXKfW(Frq+hQr@JtQ0AahPsz-q2ZUGGFx~8T-4nyb8OQ%L6
z;Xy`_DaRp@0{ZbJPw{`cK3|T%mMmmIpM{z<BLO~EGF3*uq~3Opzu<3tH?=3Sr|UR9
zT<ApYL2l~BlAuTggGRPXg0?!b*oN!%$XwHNoWYqT5{a%NizY6hbP+Ch$gIInn592&
zGEl?6-rXiEyHjK*3py-1i_Ry9ARdV1iS-p>4W06MglP<Gh^%|)k-N`Hx6CmJMBgrf
zG?a33xnLJXlBQ%U&r75uG7J7(+sNUM2Ox+%9Kk#5T^@Gaq0~{M5n!5yTbJObd~Ke*
zNW}Y#EIO+BR?j4YVZcWLqYU}C{skH`D8kiSqQ-htl~c!iQ`Nw-kJV=&B6MhmKsbsA
zo&2fm8zzX7o!$*|v<b$W5+z{Vf~W^{&5Z_8Hw@4hKDCo?*XNw|-Q4#whmJ9$s1ok<
z!bdqca}MkBy6n|j)OZgLDP$RjwipT<()AO>b{7tw_?TLoF(sw~h5EO8Kqvx(lbvFk
z>+^3L&gT=tPv0&aoK*84>b^J~b=O7X56$3Pok|}9SYHKk`JDs@BFxcbA%=*8*;A?0
zn7f}P+^b(-{~RTJNEjIOu$>OTy)sA)wrL9lMFFE{BrOGS<!}w-=7O0xfNH#Dv`ar0
zNLhSDJLl@|G(vbBMyEZZhh+qaWA%q2N}y05f7ZZMd(7-afqJzB&F4msB<Wc5J7c9w
z#Nr`Ott=ySd?(Z1rH4}2I0cEHc+ztk4L8Y4Iv>-VkS05SeSB{7h$gjDguF-)ko4b(
zu4!wQ5(*H!jGE4w3{}RS#r?gwQg9<Q{d&H$1{Yb+ETD+;{O+Q4^PGioS8w8}im)4f
zjQU!Zx8q)8QI}K|>uug@2*RF8`Zx<V)ead+=7}rODXM$Vhx-<r6c{1kfLC#--6V%G
z6cjVUi0Y(8@3Bz=k^=imA9s2%L}WNaGs^l}%u)dD^J@RbzGuxxDr|?Uad;2ui~HZ7
znT#=;)L5p#_(ZgZzbb#(wz!}WAxa?MjW7(e|0I`21ykztEKO6%Boz24+&m3>i1PC5
z-r(fF*Rb~TsOSJOf}jc+1Vw&MU;gQa#I+?-2!f5pBB>7}rZKk)DAM<ka6_Z&Zq_LA
zO)8rYhW`w#7_)rdX6SwYp8P0UmUi~T&Ftb%!aVh4-07N9tlW~9E*CCAJI=<`qTMnd
z@Yis87n#pM-XuVz2s$<H3lzXPzxcfaL}*hhP0l(p-$OjHX9{gkyg%G{^gw6+vXS(K
zIp!ZBfYqgPGbh2#dQ}@M#ryzo8tEEZdU)T(`=&C(OZw1p{`CvEH8zFP?!RrXq+{Pb
z^4sCmY6m7{BPb$ifG)<=aETsM|C&sQP<7aDvIiSNbnpCx;VTUPEq`<)NU{=iPTLyk
za5*BzyV${c0PC`YTxY-fpQTQX_-hYUezItAFv^&nPdq7lb|6c~VXOwW=HMXE(0+mV
zt3<<)@dwP%KQpo(=GLZt6UD!o;>X4B(^`Kb!8~mp0K`zu3wXeIu&{%_kO*Z?hHI3Q
zi1)JaaAuDT{{+x%{2X54IYi9&y4n@{0^ZFfF?XHW7B|t+Jtl1Fd%1mmavoE0l6wzH
z<^qes!T<H5XElDBp{m?Q$w0p9sh|VUugH8$8Lqv#D<C7YZ_GhN2_7(RWK-mU#F3`c
zri;{IAqv*wz*R#m4lHh2q;Mv(OvPkJK!dqYcPRrl--y+B%_EHvNyyMu+se6gfapg9
zzg(`U{B3aZe;(B!gQl)ldn#HQ#fv37UJ<&-b7Y4l;?J$??xIrRc<L-QA{z^I)G$Tq
z-h-mET`}=2W1XrcOGJBq)8*9u1F7e0?KVrjl{;Q!dC$^#;k4FQ%$s>wzVx8^mrHh0
zC+eIr^OVjhK(*uMUJ;Gg85FgmLZW>7d$QfnDT^22_#6Nu+G4qpo41~-Vv_0;e!WTB
zhx#m`YaFX{FU;@ROZTtRbe=lv5-@`OHb%@f>>S^rh(Tp&zC{yh3oSJ`*v<KLLs$9p
zv2$By1psXV4XAQ05+9W?AjdH`+P;}8Q#Isc&UbHKOI8x=j4V@xD>ylDap-S?h4(_M
zm8-tp;-%fTvsI;ssg4mLl7V-;_%nn5F-DTvO@5?rHTxJ6D4QA1(6#r3^4r84f`tsm
z$B8X%9{99wV!dgsb=>(j&B{`R^q=z4{7Q3@c{JoT-aV(pg*DyUF|9{~d;zWItRu{~
z9SC`N{tnQxlHolW)H#lsTB+}KX}AG1tn9Ind*{L*jU49HAc;if>M8;zWd4yNf3~^_
ztj4vLvITBr_h?wzuZ5y?wIbHXX-N~le}ck`#0FJ8rf0SuNFk*TAWiM|yEW~YRUMnA
z8xIgM@*qxUG|P;Tj{w>H70vk4`b5DW_~5OS5g7Wv$*6K;5w9&oY)2bdJ=>42)8%El
zg-70ER3s1zq7rf(m>nWo@Fc!Hw|~SW6blV2dCu`H6+6J&Dye{6A$a*yFng&Zg8iE2
zVDwJ13WB1>1%;bN9N%7XjhQl+1RA<)249y<G=2?icm`!-1=2a%hE_NC2Xl&rV@`^f
zjP$gb<KNAr@Xppn4J{KB61hIP*bNo2w}F(>V3m93RtNxajC?DbGeV;!;Vd)lK3v9J
zKWF5q%nW26apvrg7rcXhJ_bGGqvg6OaNcDa<}fs|V*>WLfb?u*r9D~5l3UoH)!+-P
zH+Vli(gX)npem>KXz+M0jYEsU7#P^tJeo!u`+ZJG-tC}NukZA8f_4-y0j9JdPNhI}
zcjPbm3Q<6kP7!$ScUQEhwZs2vb8o8ix%z1mNO=O7lxuB^CjOfEOY<EqL<7xDsrn>h
zax8vt?OT)gbbTGX{nSTl-p$K?ss7vvOz+r4wUg)lg%GilHsVvGH#tD>MzVug?5B?1
z@v8rAa!M?eEbN)jQT7X_Kp1eH_o-~wy#Ra%I%^j*ulTc{IOG*n_5n*hk0Smbf{T;o
z|1Y>0S(yH3I`5R4mhA=`%KsHyFoQ7k1tbL*esGqNaF+F~K9(U-DX3b3bUDJTupw8i
zH*uj6q$6<iPPEU4id|U~u6#L;(t$Gkfh&#pDUA<mQ5F1%s1F?Ky5K}`#wf^GWBToI
zd_j=$#yxS2u=KiJUq>hGH@Y+E+W%G@a)kGwKN`K>o{u_@hCW-5<W6dd7}3Zm^N9d;
zD5Tbyna`V*3Xteb>^NF7cNdpGO1C}Qv%WZu@>0NP5Ptb~z^T8bSPd3`S43#`ys#LO
z6U0HS@w8`ryes07{n8;6`$*8xkO9~Mqcnu~Sp>AdTk1fdMX#m<e$6o~>wwt&K&0CJ
zu_$F84M11|GE`y~r^>UM*t~#0vov^#%H@QLzPy_BW#I-2?#@6rt?hK==K=7MY4de@
zydUp9*G`p{n}^mfc_WI5V0%6RBZvq=#Y)AcJi0o~@5k62)|xu5D>7O~7wt-);wTvC
zB#8GC{@`C!@v)JgP}LqWLe8E`H!kj;%@*$--cO;^87lb~QdH%@msZzjK-8Dmd1|Z|
zU0w*}{g86k-SC)!V3BwvD9St<cM*y4S)iuEOsp!tKG$DY_iq@qTYKu-A9K5I%vMzo
zRg1L0E*ahY;jz@AICCV%v`_AwbWLK&6Ia=_S?`m7_g#A4H?Zz5(oKdW31fi?gqn=q
zo;o651eEx|Bm}9#LBnJArw$E-x%7#$W84rAw?BbGYgQI+>K#pwML8LlG3n%#4+uuR
zMP*sNidHBvxBiY4#<~NMG!vaQW$fyuQ*0xUZvKhRx~{?PyoxmfYFtU;6w!IX)D+2a
zo2jVcy3;BsYy?4f^4zo>VyM!qLp|qG)>d#0+-1mQfwC`iH=o_dyFP5D20JZ|DIxCG
zcFaFXrrneoSbtn)6tE!xPpsNT1}=K0HxUo_SbAaMh5+RV@s^b?qUNs1*x3r$;>7yG
zNa8<Q70$GTZIAYXN;}BF_rUuwh$SVj|J{(%5DY*Oie4=Wco?I+I^|dDxT7~u1B9Xa
z&EnQu0pgU5qw2zuZj9-`8HBW|>|G`}&f%s)X(7^CTayX&-eZ}D{nRX`LlakRqWryx
zE_p^JR0w_ZJP9&Vol9>~huy=n<hkYa!r;2e`z1T{4UPDJ+%?T#gNzJ?A)_d!n?xK9
z2G7X~qqxL@WiE4A8M`f@4F_TbbBD?`&Q}%)OKv=PCSQlpb%MFR?@^9>-bU4C?*XCM
zsa=XznNU7b#`^}mkz>z(N+pc?F7q{xy1<>+d80+llDu17HHwe`>XLj5Y{@D<@yV%m
z%Ps2Emp}GIbI^Q#g@51~T1Z^`DC!B3QOA4`@*2OanQjhSILQ4T^kx6~fiIL>{~Q9n
zHtPv=>QiBwc8!PX+^|?mdY$DYfi7z&0(DtGCJO-jgT?epiux$dSLZG1uriK<D>>z)
z(QygxsFZYp^9>9QHdgXtp>q{r4{h6F4$Tp}lg8t`viMpQ3IwACKC`pVC!tKoDpD9a
z1JYEF$mgYH!w`v)t6qmN)?Lo<NBZpmhWXIzEQ67I-N`jXsSau;vzkpemYxJjz`wiW
zPV3ZoS((5sdg_G4NPz%Ktkc<Bi`5Np593_M;PbS%`9`s>vJQGyx=H_Yw#i9D>ypOh
z5JR2PwR!o>>!9(1to8v7Xh1Mm$JPOdS^B{+Z26tS8!UErrutw@eF%~%IyxW=b}~nI
z0Z#&(+)4Y=++QGwu7Vre5EanJGK&Sjnw}JoaYI4+QC@CPrwFV?-2eku@C~mD<Qt3e
zOeVn59?3<E!c}V#EVl)&p9FN1p7MzG9+GXC#05sq?Nkys>uJ+pY1BRK6m>OvA>XMT
zJ_<k@CvvZbww|oE6t@dM|Me3Zme+~+zrODOU)tRNOOs<_`QKBdOBzy%zxI@#ncCg>
ziYR>^ah`<VzKJOIP8IiRm1N247r%G{SnG(zh{Y)K(61}=Jb{QLj^$Nd&&0?u)55va
z?{~rsE2IoB&!?&V^Px&<k`RnBlLmEiK{9tT$f<kddPyH?*6gpQ*Y6=}(J6gY^r*vj
z04#&K-0%B){K!oFzm7Qz%JFMFsVv^zY2`HDXTE6czTIyd`^}|JA_sHiC)3;sjx}Ly
zNRS&1<?$8PMEC*lE{r8%22APL1)w%-dt5b*K?+1HMm`3zZZlWd84Hw#n0r(gXuP6K
zb^DM^wg*s!q5H&9nKnacYEjqe$iO5A0T+}=Ik0QhW;t!q`5tz-U`+F5Uk8Kk(#VpY
zIh7v0&A$NwgkDbFA7+z5Iw%P}q=Yt#qEDk>7W>?GKuJw<#}s6NH&Rr?lY7n+YVP)R
zNL+>*ZSUd{PcX-W5p2d6mjF^goF|skC#&fqNzClS2+V(PN9X{^(@buF5+omP*x!Aq
zy>)j0vY>f_cLS(QM8ZK+JG^&~T&81Qb=QU0_7<jro;**=`<LY<dQj8I$iqwpsff#&
zICl?_k>mWxDPcG{v83rmbui1e21`;1)+GxjIw>}5XwJc39#_3oI;b+8^fvG#=_Sq8
z8yywab-fO@6i3MePiWvf%OXp|Z_H3F{moa|1O8h+!@HwP25Zly;M?3iCJ83DNtj|e
zsZTCyYIIouKBjK}km<q_0SH146}=(xoq(dF#!rL})Pz^^QM1Fuk=>old%FX_QN3oE
zahf`UyfEj)+&<U&P8QzYh!A+f+0rowU`a(A0XH%Hu*buJ5J4AoHc_|VA9%dWUTy@H
zN^FMbJ=F|K)em?IfLU6e#*<fR_CqRc?DZkhcCJmiKu_(caW>k<udKB;JMZm>(*UNA
z1Iiy&x<^2-ySJ_-U89>a25i*WZD2S!6>#`M5~3&%KkNAjHc^%uc&cz*h*DW}fc=)?
zMVkAs8?IGRRH9<;9y*%s2{tcYFzkfP0J2C+cfQEnEzU!2|JbkmF5r{O?}A;wwBFi)
zNV18mWPCIvJiW7^z*e=fZPJTMUmvQ-0TQrnqPaXCMUbO-5NYMIs8O0>6apd%^LJ>U
z*;H357#MkdIA8vuf9-{c9Mjx$3SbL+1I(_Wj%$>lwO_vcS77-6!u8jFz<YDw%u3Jj
z1Jwnwa>0<`RuP0}*O+BG|IdGg(z;p9Y%Td0Y_Txqop$ad-ADE%gr`da97P5aKBDsd
z3E0ycIU~DE1o3eW{-UpHk=x+4-ciWf5%&GgWyp#S!(=Q%u4RrLFt$?+!jGLA%}t+%
z@FE)EPPcAkR<RVcEYL|td?SEF9M8l^(Kk5>R~wxYz%luX2qtvW-5pnmz-Ps<lparf
zxC1TVJSi@k%P?Mul0+IvUTFr!KzHoh9eGBT{=678Ccv1GdPnew{-r!wZU0`@N;j7x
zuj#k=H^^Ux487S{f2Ut)>!y{LDFikN^Vv@2>*a=_<4OTt+Y5KA6*p|XH+4-@>yx)t
z3;a|1aX)NRKx-tSf`;<+MH&L(1rkx@ukp;i4E_izGmfNu(V*rqBkt?9xg|GvZz*4M
z%?7M3Dl#+96IQP7zzv+Fau-p|%bRx9vj!@Vqu->AF_9RTZzNpza97xvZ`(tIHYlJA
z(WB71(;Lvg2njh32)@z*=y%2&6ty^#ChA8a1e`PI;Ri@JP-#S6w65tySol2C8w(F^
zeqGVu-!g0qf&b;&OyxCfIl*!BtVN<ch>DT<9s+yFRugR%B)-b5VEazg4}gIRsubrp
zLV(fbxCQV=WCcy>yM$5=U!wt0Q!eqt%gh6d=<pYS%BTeXRHGXk4aJEZ&`bpK5fl*D
z*)4DHm-sXkvwJ$<;~?Kw>{;=*ulwI1Hp<d^W#f)iiMGG5N#7kzm!$enEq87#L`k7a
z;Mq_{2>t~%TwGh8dB46LigajJk$SsXT-(YpXXMsXend0c9}Fv`9`4RotrPPOZ|xhI
ziFnv_$^G}tfN67hU7_z*bQUe<0td!z$FF{+J}?&c1Kg=#r6_`gZ1EQ6PNj)Gl=*L%
z6{`qOlwN6X`!jFy;biM8Rv`7ceI7}j5Ud2r3!Uzp$AMUnLP$L?Va~-vemi-t`=9W>
zSyxn911yc<10MfTr?NU&+TZ|VjLbR|chyft3^G$4s121#W8sbcKZI5xd+NxY&8;G5
z*05tD1JKl5q#gt9g5-y(C{xZnW!^-G24Gp);TTWyNq~QxJT7jb*(tp4am3(rYE%P*
zM>$y#Or6NfZR(t{#d{BYglV&xW=u#*xQ)-LeZMOAj+*$Q>-h!J(1Om)rl$;ep^pXL
z8cw18EhlfmHQpFET8>J#(XX~z2$9!Ej76KmkA~VH+v<)2y{+T{>KF68$_e3~RycrY
zo<rppB+jruPBVZgEB%)DilaCa?(A^>{&uVPTfps=L<}!d2kkn9QO!b^{WHQHf){qP
z_q^N|0#O2Q8CVd;!qewHeRtVK%oHKP3V#y8{NtjvJmKp7G!h0_i3!?@ZT)_HB4lqX
z<yIx1vS(h_3*iKzR4;%ZC&)JZl(vdZbqewfGh?bP6jw0(@xITR<9%&rz)(ixMeEZN
z+Cb}KQr|R3rEg^&cqxc|!CkPx8IM;1b;R{Y$)F}IeC}v&Re&?JY!7;!_=hVWXb~YD
zo8<AA*#bc&it4e0KWJJAB~8BzBj`3TV(i-<CwGXMe{Kv<w)8FDsokLydDTIMU+?Tm
zG47%(%mxgml!J$bUgx>}m00)h&0gaVO)$0#fmc$9g0Qhd#T`9p@-?ISi_A@wP*TCX
z2)j;)^JiuM%)a_dYH0TNiBNPBxQmT7$o_^nfk&y|c6W6!KjN93M&mT(l||vV{@6e-
zcCqITH_h-Z)$naj!Xz=pfMORN3y%6S;j*^A+DBU?a<Zd9?U#PE5CUj0g0zP_R~ED@
zK2lRp2IQeE3kHzk5~}bQVf`_33%vb{u$zyBb3>6Qp_vD$b=Jjrfmy<8r*mKE3N_Q+
zeftc<{*U^O%IOb}o1X72;n6FhiPIR>)^J+27th^fHScg%KjP*GMm5k#W0c&Vs_(xI
zv1mMaCV_wIIRJMrioxTM&eyyC0D0;oP+JkA)H@|TzJ&wy!nQ=V(MqNW>2JLQy#lfa
zpSOlpz&A2-J{f<EbIHZFu+xya=#8HKKwJ440sj9b9S6gIIF1=P+5YFa%de#Se-pQ}
z3ei`8UlAyw`tgvA$<p*)(bDZ@6AAK6*3k;YMiM$kA_;nVMbJp4S~jw$oEW;%rKtES
z)MINa%&@G9u)IB=p7zfpSEBwZh%ybSG^vXTiZCezkuhmj6~Nm|qO|uUM-1<Dx%axg
z9ADt0*?vDfzo$rH_3CMtN_-}NXmq{5RJKOGto)(v`Pg?RD#4cgZPxM6-H~HvKZyW?
zBod_vWXEDmlPEflH_oUdi9|z{_NC~cWsq{}qFT36qxq;Yvk)8GZ(+`^w%TguAhvyX
z2o{oxB>A|rah=*-Yhi;)XV}<1spx}r4?QfH_KzF(F(M)SgX^jlY1vW2T+A)X5&@@K
zm|*7SQ9toFkV^1*YMv&t;q+Ntd#wZa4?kWlLPD|j_~~o=hHxxw_s27RjYo6a`;!HV
zjvuIc(rA>~cwtwEgHt}q<VyT#*!aAeY2nk-i?7P4zNKUu0N2c}7?d_kVHY-!;)Hl0
zFeM@n(;PvFAo%&DjZr^S#QZD%PZlR~RnQk4buoc?3HVH9rjE&Y?nLL(yBO6OUh69?
zzKU-ky>)##OjIL%phiy0oBHlX-%E-;k^X&Sgw9joCIud-<zDPxitQ+%eG&XmAg<U>
z_`9i}4q)!Jea@#Op4yx)+&{S$bmr$uhNlqPxE_DOVI+}6NkAG$@?Ns<`bZBdG}ao1
zUc5-q@b8W=)Q=GB##a}qy#&;y7BSTW6@p~F{if>{szW6?q=(51?xPs^(da8qK+Q!N
z@O)KVHcECl7E>o}0q#*b0cyyjO#NL0FK-KNJ(W^cXtobt#=pfWYk4mROvOoT^c}YA
z6ci4S5%F|)<qH=*h**?_E&T;Ivd=t{c89Z<=|gWOrAQuyd0x$~cxmiH5zIG_Ti7}a
z%<U%|Lr(@;Mq4=!2x%%O`JymumE10{vJwKxfK0G;XZfy+Ux!kLTFh0`N+7Z;a>#_i
zrr5bz7PHs#n^iWe0y()q`HV*$d|`(Ky#gPf>Xv<2mUO};0bI{%V>~&tM_8A1ZYEb3
zzdvn;){E6m@$1xdSGjf$u+ZC6ti;tJ-5dI?cyf%+uZa$VCgL2_N|=#dT>(RLDM}pP
zsm^8WjFoIX8teBe%b-vgI6k#u!jk%D+ZI@IA`qDA)^X%chEUG)tr?}wz~{_eD4Z0_
zJcObAv+PiQYRoqbHA0{c<e!7$CZ>b6U9cP$Dn`Lkl+D>tgi5S3<@;cgVFm@afe-Qr
zHP##4Iclte0*K-)lNM+)dj`gjxD(kY{w?;6hC%_fgLRpPjo4abHP<&gerC#F1wom`
zaPngg4a?mG-NGPBWVfo&7sw-NBPWxL1S3g>#my*F7$CAtI8Vb6_`!(CQ|S7Pgf5gg
zYRIdas#6$45bjMS?tIH&o<c15@A1O}5=OrFP~I?}lEiNKaF&mFH@&$S_{<Az%>wzY
zOtEj~p3#>8F{CGuBtD4HdFt#ZQYG9Ii<XU%x7iQ-Wpe3;+n+P7g4Q5-@hQu;_jpHC
z%0j4!nK~_l(@=aYm0cll0x|hm#y{>pBnH+)C-@_}m*ks3<F0bLlf99|2bFmB3<~7R
zOML1<-(KH!3UHuF0yGJvJP*`JJfLvPYsYm4U6l=CnfJWf1(k$+9!PMI+=j)(v3KsO
zy2?aT*&X0<Z@n_Q<Qxmv{6?#BK3!re@&++^AbkaE6%fAAie1bm7G9g>*cpOkbPcq$
z0h=Wr7_h(-vDo5cu>yd1Z;9ugQ>zJsBGno79Bs$9Tk!mjcF{JeO~D@1lH91onUdn<
zTx)-4O7H|R7Wwj#YfDp+{FWS)B)aYa%e&1~ADh;+;BEW+N*|h`-?2{3xx|D~I=F4M
zAlgq%A!T+|nQX?72=(&dhq$U|D3+op;?Ab*0IDbdgcu4`n6kdL`HO^OUu2L^a+Q%&
zZ%nBcT7Jl{uQW|7^vXONQHC2&>~GG4+gdFl5`m4WM)zm}qb3D~P@#PJHKm=j4GaFy
z8wwFV`9JSx)dUz2SqGy-SxBCNZ}Euv+ZPNr%|}>IjszW*af(1HtlqVYK&K!aY$mJv
z6-7%oB{~vh)M0vhg>W_&J2m5sMT6K|H+r9B#jUO0#`i@zeNb662#$5eX7ZT)8c7vk
zRq9mNw9>1U;s9Us;udjL)|Qrox4jYoNq4u-Zsh8CGW9hi2|uqAwsC^&2UIM@)9aab
znK*PzzGi7;E4Q-T94q5;U75`<uR2sS(Xi7)RD+9j61|2mp<63w4pB|mbd4c$7bdJy
zLe^kbSAoDF6f<GG7mmOOWVgmIlO^$`2KCn|$8({(kfH;)&6E>I0$Y81SVARnU=m2%
z86jn<Dn3As!1}mhz8Ua+TV1x2k5T(zSFJxT(+Di1Y9Q3f`U1MGTeO!W@a7qkS@)qn
zeGr9~D`PEZ1z{c_F#TwF(r45r=SZc|$$R_{z}vhN-135OFS-Oz=@e9FJL<;@#Ih?)
zJDo9%+o!}<4(ZAD`6?Z#=UsazMs3-x@@I7iekQ%cO9m_(YaHnw)|o*a&3!I8@f33-
zS7|`6CL3(keGojPfQYAG))#;e_cH50Hh#wcGJP_$G5_z@hzr|;HpHKtUO{|RmcOql
zL6*>)Xq~ihZG3Eej|10x1X@l7!8C6qoaG<y8{CI+EoB_CFV_)QsbOTH1L-?3C;hXB
zeEDNHwc9kcHaimWC`Gf{<P$PPP6b6|$`!^Mr1<k$dcM7%$IrNhLqQSHqDe@71BhSO
z@OnS)Z_i&(0n{0P4?Xf(WNuPc(xrP_d{b<Dy+1~dvX7iZzUGopFx9BieLp@9i;G|W
zIInCREyb$A#1unH{U%=Ph35j~lgTgsAI9FXOR%;H)-2n$ZFjlLwr$(&vTfV8ZQHhO
z>(rU~IL})1?jNx;?#Rfvu9C9QH2}S`8H7ve;<rgMDtbPZ2xF+2d_!-6z5&#16j>g&
zHGPk$#sIZw{UK9HxMvWz>Mn1l+`48Hi+kRsL5QqCgG;I;Vk<`j)Yq$blBeLOs@MM&
zhU3K!{k?10+S%1By=ft>Y^aHSno)2+mWIhi19TO3>O)eOi#J!>@(-K<2bLjr)>s-?
zU8(p|bETbzG2}?EG8g?&EK>HFs&CwYeQmJ9BEbd@nS0y-3~|P;Dr-Euwl@LSx^L7>
zs%S480hY0OzIu1MewB+xgP{Hm`&|f`aKwcdtQHvpPX<xVJFoaxUP0J}b4%vnMItXH
z0#|cUUiN-kyH3p)>#P*)oHg39Wc3)wu?CDd&;i!dB<mY!ZE}J0Ld-n;k_G-1@+-UF
zMKd{Nh;4HdSeyXJ!$W2m<g>my`r+`bx@)AxB;Gf}#S)ozIFQ^uk9Hx6IS;eF&9VIv
zQ*5~~(Os(Ax$Yv!$+2S^SFhVkKGc%dul%`oHz{QeWZ-wa$TmBZ4q@xMW8p_+C~GOn
z0y~T4ZlxWCzqGg4Zf0*{Vz0xOIx4!h1N&3c^$|G?37<lfGfGEzEAW`kH;o!9UoDD5
z)X3llM=r&iJhctO-ptw77dO>hA$6d%a+agl>{AY2Qm^NCx+WcDU6SE(yngFnOC;!y
z2*a=buEbRn5GO3ihq9gsOc3Mv1lD8h+3V*hO(3w+h;n7~cOOB@w-pvxjAc`+Q*5`^
zg)091cKm6K-DqGfd!ob$bIbwnX7yUPqliO3ekw*CqF8}pG3cOIg9IEu%%wfo`%Q?@
zs(y4jTdgHV`{7i2p=oN*EoN}A83t||)CLCK=2o?kY6J=R+;DB>sGL#1$cox0;=LZX
zQa1H`h-Vx7CNj+CcaI2W$YvsF@j!W8$`D+b_|LA;_Su;kk6Ex)(IOurNyr_b_?kzk
zZP`IF)?FnFbQsRhr3<_zVAj5mm!?2S6FfchUa;XWeoINbeR=31)4ExoC#fPu=;%X3
zb4_*;6p>md<KQ-6lXDm%p;_=Wu`mDdaT8%Bs~^q*I@cZzluYY<-GJWQX+bqIlEaeF
zs8OE-MJjtT%ML6?9D3aYKuEzv=l+vg5T61uR(X#&Dv)_KV}|{mFB}x4T`6_X?mhHm
z;;SPXFzs2P2B5#Sd+%G4>8^jz*K-Ay(9M?vPG{_3vc=XhrVA6Xz_{<0i)u8h*~J{!
zWn|SZuLIp>ls=2sNj_KQ65Mf9#KEs^*ok+$k?NV^tIv+Jj(9I-JrN&Zy1DmKR>O$v
zbyuGyc#jr8GvQKd&w`Sq+iHd0OgrCgwUgx4#=5&Ya1@W5Z|g#{_Or+Dt?h3_n3rwH
zu?Y{Ss=EKZ?aO*ln)LnNbIhR{wlS#yvX*{5_K5BrcIH(fP#QIQEjxX?X*LXco;Y3f
zP3uBV&754#D2|3DkdS`%LGJ@5genopO8#Uj#H`a|MMNEixwXCT9vG{h@%~l|x(uY=
zQGZTk1GI0TBE%a@(^~Zyt|n1>ZRJ**yZRAL^n?9ZZEXDrU4u){ShMB<;JaRWrx556
z!JZ!oMr)@Dp?<^*x(`VJrXYxxy%jB%mVp4({IxkNrl|Pp`30VR*v;t@)uq@qPGSLh
zlwQ@?mt3n+3=jH{+F3nMjBH=K2>mYa-(Yhem`e_SyXVdVcg+^ZN$<H-)}DY%-1k#o
z<MG@hZ`^te7{+?Bjr7{VjN$q~4n(rd1nwhYDr6J*+2|fdNHz1(Zg87^kM}JXPe^Qa
z)u>`xpE1b`D;@>7xXVAgkO}i0m>!j3L&B2zQB@616|vSFD&7;&3Mt_OITSg<1Ft;I
zjyr=!eh>gO(e4v}d4ykX0_pxHJ>yGAehWKbe|To#H*|*#h(@`oEr!K+@SZ~b*Za}J
z2T0~sib3)&EFJzqjq(}hqP5V`hWY6mRS%CvMp6~8zx9()5{_2K>>mH_f;WaBC~>YX
z=CGdHr_IE;U=SP_K+Vb)uT`&$L%12J7f!0-LDhUTs|ps8Ht_UIcGUvtF__$fP%v_>
z>FP&GvLq}*X*?v6<9GSTD{ibKbl-vYiwTYi&v5&Nm*h3}r)i2xFnT4QR^jn$G78o^
z{tGMIzYN8Y-&SW`m8DBP+gX~gVlCGS4}<Qo@pQ!EhcNQ&T2=eZSkZ{AWl8c%Pu0er
zJE7cCfCmrme@C`{jPk5LbxT0+W%}}Dtdx^H6XpkO=2b>^gcse?j>n9C{6h)Uy1=qI
zX|>WQ-KR<XvKW;UeKA@)F37Z+CDys+cea0-x<PKSb0}%{#+Hm@9X%q0Y*lKR^RxwH
z=vn)44vE624VlFhC*3O-_4G@r!kT-mo6&w$Wltmd0XQB4>B16qJJC!x(YWqSdGqM)
zVrn|Zmv-^ywQHUtx>7~IuKRO(d}X3HR57WC?$a5QpAi-bb6&k))&z&RH~F*3!*a8S
zwnAwsP=Y?T;KCVpexkmdvpI)(y=FKJf56puULvX??FLXcLoIH-NEvvzc5ul@LR#yi
zc_hfw0!73$Oy6&Lq^VMy31tD$WTVZXqe;c~tQJUIcQy8URF|*n?F=21Ni-5@a&$0#
z>cU|>059+mo|%0y+EH~2dKfPp{-)*@_{{Fbk~s9rp+$2f+io-Sz9oW?lsusB&1OI$
zm0P~3AzNyg<>y`ab;JiuOg9B75Svxxk<Zpu-|6>K2GD`u11M8vf<sB^u*g!KkO21p
zL<xIak;-6Yp632x|I)p&Lj&t~;1Kj1m`o%xR9^(;MHi7>A8#-d_5a5?5Qr=u12PN6
z7`L6h0P$`~_5+SJ#7)4QrQ|;;G$WSTb}~Ps%C!s*aeAV8YJ6()b$tol{ejP1r|$n>
zTI@f_5{#^W{^xeqD-B884H1Oy3)Nc+z$Ft`lN>%rF}V^u-(s=K>`WYz2pN#o_zKaj
z&kwkqeFe+-YRVxx4@P5)Hm(@ko;MRpJg+FX$BW%7qucYWT{?$J^VzvWP61rXGRzdJ
z^vQyWJpJ^jVTp%5`V;g7*%RBJ)Xbl6eHO5x5cGt&xPd=oIb&<Kruby&1@s9oHQVBq
zGNtSXW@cSZyb)umfHSnE^ndAsWBM2_oStWm4qxqf>Ib&&E!wqddQ2{M=$m)yW&c!s
zxMg`|(w(e@Y;9@V+cvM$o@DsiXL4AMA>ctWC7Kfw4UsmS=8w_Rev-N41-o9v)ebCK
zacXYTH19NIJh-({{<M)zG)F8kX$lf3p&`Bj_1hJ<J*l=%WCKtW_hD+(`{K#N$e|E^
zOrpY1QKylZ-t+tXWq>y@#+pL{+Vk}T+3CMn;(V}A*REEPl<2c*+BmdVvueH=vPx;s
z*>1}^mQNd6bJjej%O)pFBqVGU*H9!aCb6_u6-H{fFldWf@sYOuwYmEMl=ilbDTGFQ
z6K6BPI#Z;?HDO*Uaf}(-F&Lq-1p4NUo4Y|&<fYo@XZuIoKhqr5hTBHgj4^l6cT7&9
z_QH@Aw9iyP`5TSrc@ls^q1S~PYq>*X^v6S;)b6|3BWJt;n97_8cs~${Mr3(-JvC!1
zm78uCi6AEJQI#us=#E(z>M`oc<6s$wExKS0qnP!zsh2Mo>WbIvz^v7%vogG9v?E$5
zt-+RYpbId07)6ZGjQ&XNMf9UI8%ppZDH_(=ymCSS=fV*3Poj`M>WBYD8i1cB3-}8T
zI^r*vs2fUhSCp%!wFO7lb89A}+fDu~YFhB&bZSk4`df$fXKrZMuo^BBU6FsD580er
zP|t3yK1YC(MJiXScOQV-oupM>-F0r2>7;ic^rl_<GoOa~>yf~pa-qXh536BhBb)it
zog!RKfNlw6$ZGo=xwloCBuc)`hsK$^tf4;z3yGqyn2J?N%z}cdG^25K+=f$-qqy&N
zAqOWAFijrItjr-J{ikhGPMuUKtg=<Y5YhHv@}+%NE@)P+ZC0*(cJ#XP{-WiT&t0xO
za5`yZY03h5rlS;2pxIYZ$zMl#goc8Y<ctR^!Tv9sqmQgbe<M@RS-{&<-J%fj1F*>?
zeGcS<B@~U28OPY0$+0ZKnZ+W#@KC-{y^wRhi<BICLREx9(Av&9NN&GMTaKt9lUOex
z>P-XnSh<2p7o2oenvgbd8a~&QFN4ro$<<sun0@EI{6;tmMY*{`hJ-Y`nKvMdV0o*d
ze>|m;q~!w_&#B*n_aD7s_PX@&8a@rOj4$bDq1?+`i;#&7YIp^mBnT&xzr`h0=N}dM
zDVaxEDqSMCg(87y^(;{w!N)JpNhNQ81?zIon-R)VM`FF2hW^RDA%9Vq$x8fiB1D1*
z_TWc$F^aob$9x%eJcF+^fgt#deK+=8N`4TNe+@|+f)a$9UlM)yshgm_@9#(&3V#an
zr0|-QUjeY$wG#}P={gDWy@J*dVgNfhj2BtK3o|bcjtFDV3Jfa=@i`G3EI8g7JcvOp
zFUI<FfP!9lQ>gJK`$OW!FDj8k@otB0$B^&g8M8+@*Z;-b28K*PDh>rw1{WMNC(sub
z*a|hlti~XRgk9vaJXtqtB`zk8UefXi1vu$UE+^XdhBL(m)G%`Y-I=3I6jG3AV`AMM
zU~uP-vp$KG%O@erawTQPLXow1$5APne$g)elz(EB<BV>OV;BKZjh%20z9;8j!I@7p
zuQ7FrJ}JUk<i7Ohm5vv9OYHeN0izGZN169CKDcs#8LHqGpnSW+yF0ZTl$;aExQri&
zwULL$DoLh~Lo?ZEuyhkPS(r0fzy?`y?2Zfoe{)uv0id5|z5GSe@*c%Wqtbnk0IYlh
z3^xmQVL}8Q@!s$?761Nf1O%LR)n?yu+5N{oX1Ok!vjPy0Q1|oQc^rFZjXoTrgYo13
zBbHm26H5Y-vyc~UtkpiM@Ceh~h#W2G$fci_D;`N$jC>XH-hV;nl4sM;)qhivks1X5
zMk=pbVk3~*nH7ubP49EAqD7hV12BFGBmN&_hv`2o0LDKY|FieMrM~qW)<f`_(ame2
z<Qu%#JRE>>60X<p4K{f4HL%7<(G+i7SJ0cVzTdfdB@GKl;wo!E^AabC<KD(~fBD$L
zBu{~q@#c9qc6r>7Y!D)*Et4`qhfE0DMbf8>Un6qQq4NH;fBuMI_J7htGX{a|;|tDu
zd~Z>;H7W7IHC5VW5gkd3W%uIZns$F5|4hsJwtqjltcr6A{8|=A3p8R8m5`*j=vb$V
zXnxn?G>@=YsIGXh)?{t^aHy*~IkrAPiD)E;K#q+@Hu&YPP!=bwL4o{05k-Mi&-Z^$
zEL=P>%9}gtWU5Y-cR3sKx3{Fv<vG!(i}}@>)O}uf3r7pdTtfL_?1_uS_K&Z<OiLGp
zlY<Z!Lohy&7yunUFYxO5l_>3oK1^TzNvPMGu>lJugzpG53l@_x)TB#yJ}h&Vr%zA%
zcx+Qs9eS((H_ET1ND>wr)V*P^gF#Fr5}w)0Q{i%ek_K-J4%503hdDO3yUTG&KM5x?
zytrj=J3RI<Bz#l2iZz9}N78z4b^eckNvri?=_9M=`|q?W(bzzu#Q|{Hr!QmGM3AR7
z_RG)ah`Q$y{<TODF75glFggcO>jpIjydxOY0C7pVYTCWHJcvA@XtAgYq);6MvP`j&
zkq}j@Jm(mO2+{;DIr&TEq!)0)-}_yra;NtPB{g5v)ZjMT<%IoR#1r?_M{fzrV6*CH
zcAT!2z0j|z4|aVN;0#W9LhO1BbJTnlf->=b%+%_(h|RRFgv_Fii;H=k2`e3|&lazH
zgg|<vTw@YvDtK5FLVtp4QA;a?zGCay(wR_T>C252>Jkg}$Q7Cldf9q^^GF9;gD9Kp
zI)C_j0#u0To<DrIk2_bz&ay$${xsH=5)tKe%PlsCooma`FBpr!Vx~fcrfBH9AD^j(
z(pxG|&1)WGS)yPQY~<Hl<mh<ua;OMw66=A4bYAl|bRy=E_l=&f^Ob+!5|^zu=+dO$
z;N3v|65|=e-Kt+C!FK4PO#NNfd=JBb^kau;*ZgrRdT|E*XxBi0q}22LtN<|N3n`;q
zn->y~)M7~hmuV=+_|dUYdi&t`I(0}eq&m1C{Tvv>5(q*vV0-_#?9)*>QQ$9hPLOYS
z{>!9xMyEF905V@5t9#)Ui*RHDi^m#>DWb3M>64O6l*G5EQ)y+uF}`LJ&}^rdckAt#
zLFm_YaI<?85z4hDWAkp!rf-RGXk0fI|G*Hyz@rDg394!N3YWiSgkuN5+%DHqbEa>p
zsJ9`vPiAn;8uZhDUW2$rFrE4e7AoN2D+!||$XZP#@GT-lTfI$xDNg0kvn)nfxGRe2
zj~KY+apR1nrW+j%UzG{tGmn^#zeZOMxxY~?pbAGDWhCl<w+Bs=2cq|F$21gEs9Cx&
zaqgm->UuB>G#vjq^`GU-kER4miY1I9<r!iKW!p}RaXEH%IhN<>)8;@thx{li2hML+
zB__#yx#fToW`KmeY7mpB=WGa3A7lp;3k2`)y%Vg&*74!OY+cQ~;q{}A^%rE6+rtK|
zsmjJ=M+t9~B7$8x?1MI?fCiPDo#D#_m&3&UQ<M#a>&t#(xL^77HTZU<ujWXGk<Avl
zn!hSblAy#ZnhO8<i@ky6%dg3IL2cSmT-cY92pb1aL!d>3{JgvIQ1_~D#b75{0Mn5r
zl7ZttBtQqlJ=|~X1hm>}%?+7J4s{XO?Gva#6<nyO&rT`0LlUU<ik%j8E;$t28Q8A-
zu)rRc1|{6DS`qd<ato+>xVd%6#}Hz4PB7%96M9dO<#|b`b=+8XuO@nw3E>-dC0W4f
z2Vz2V%d(J_^TG`XQGbPNZQVCk^XJY3lh-3bs|Jzhnz#N=I#apMuvvR3+!!$~(hmY&
zkiHpFVgDx@G@u<c@LXj#dJPqFrMrg{8jc$qc2=_T$51O#a51A?$d5hZ-+LxeiG@&N
z<TT~;rLR)PX8ee#Gqbh>fyK*f-}J6cR9l0niCAVl&3T;pj&{AaJ+yrEaDQ*-?O)H#
zkF;(Nt=Dajt8NbfX8iSgB)+~o4UD>dHvZwxRU7}@<(k#jm;NrB@^=UIV4$^l*cz{B
zaRZG9?oX5XUK1Pn)yWApl_u2@Mm>zPUX>!%px;Tgxq2weQDM-)7)4zvz{p$cpp`^D
zQ6ekSkf|u@3&D}aq)X-6#gtk7&$8;KOp~e(0{5jHZ*R<`Lj%<dQ*m31zQTQp&Xx;o
zM;bWBqxDUznjDfEG1QQ?yPPpgPV5o)IYl;P6<6AAe9yZ-x_T60M>6hCJ6C7Z%U8%^
z`3qtz#CCNd0;{CpIhs!`Oq7GYg~J6$_{Ht@Xr)!PK&u|u{l#~BspEp7nmF>@djbAO
z)>QNd@ZKq}`M<^~>;E_V&cyJWi2r}S|9?K<73PM`!SK`8Z;Jwyy7aL(e>%{X%b!&M
z=~*?vaL!@>c#1R#VJZ^inz_QAY-V%?BH;-gO?9ie?$Cj~H0Bd;<mhF@=<YT*o{rah
z+CDO-Q8B85H7WxpBrT#zVHQmh)j9zr>0*KXSKAvp)4!dbUk8_4FFVr3O1rNPZ1UXS
z5npLtZyy;qCMgLQYc0CnN0$hBy%5ZxFC70g)KZ~wC_*9nQB>H8t}XHr_sG(%0Um2Q
zrs!nTFWG`F3d2*>x1_|hj32ZUl7e=$uEK-7D!wk7(-Ck}TGr-q<Lw=hWzf%vsxAuK
zUux9n3NNU}covDq(66iE<P|kILk{)*Sh#WBCmzWo&GqLT5&tX+jqpj%7fSXRj#-;m
z@dv{^yjrWn_VJL{I;XJy@lHx{HTRW-3XvY}psnqG+F;5~=`c|5fok;?7QM)pv%x@0
zYz{;_BBPEU)`s(#u=ug!P<vl_>r%y+j{|!Ww^<Sm6;KmVBtk@9zqr)}jGhxC%nMy;
z;1`N8lZeZk44A@IRt#<=^5r8e4*Q9}M7)Fz(i>y|j?@6DtpYDd?}$l9fF?)KLa00=
zr@6(Ti!R-#oxKKRk+=ZnHiC)a3l2jY4Lbw03Iya_6Y%C4$LF_-xx7k+;RuB?5{m`P
zank$2>6$wC7XsIljUV`jg1$+4R)<5ed$uJ8xhAOp=?m5;$jch?&lYYB7)N3#n&VQK
z^15Hfmw(%iazxA$3_3K#FLF)#f?FKU25TAkq@~*V^b@cY#wI!1LsuatJdtUz2?(oK
zr;hg?`WDUFzf9r6Yts(iiIvRzBMYW*XblixldN6s%K`?JU!(#tVO_C5mL&^SE2bPI
z9|R-`*h026#kKw8)|a@U+oeP)!yerTDp0JiRVib;*^mZEq<_Y$5PvFYD&E!vy+^B2
zKg)pH0MVWZVT4=5imtlazA-RnN5psZ^mN+KhDclsWmq0hTj!Iw@C{EG#Zo9Jj2&!4
z<rpz0Nfu@JM{pif5J?S0tt`+#06&-%=6mENEt?WJdBCgq<tS*E?UP1}G?C4`r2$^7
z*$zQY>s8h{yUI5GbGye{!e_RJc6^F$roESJUs>JylN=wy{j$$z&Dgu{v4eO0?iYSF
zV~XY_*~O8oL#3(5vEPs{Q!$)vkY^Df45utx8hrVjYvEsI4Vkn_-P7GMhSnU2#DXxg
zkHnzts*XEbgpc-Sb7pRI2SveV--gMWnJhLOTNMjW=`|_)xG*V}GbBGs?wnZS@X7+l
zaTN0b=`|?pL1w%gcxrjEMEQg?dbj9Ci+u#oov6deq3E9!L#E~x4MNT9<&TQw_xQv0
zKs`76#D^3_geyMV8}?}dW5$;=ULz!^7Tg@de>F<-p*vG^Vtk9KZzxe0)`M{bKm>Zm
z{%Z96o(Sc~TOoiU<78DXh?PH|h-ifA;59k8>y|2Q&lTXS%oR|ZC61_(xaG2O{RSw6
zK^{1eBCQZ9w}zi5cgrL!Sm;b%tW@Xq2lRqrAdUQiW+vUU$DRGePgPZ?cY#$g{E3r+
z%W{qiMR1(#3nV;s5e*WnBj41n_vSLZ(%dpJ04aznj9ou_7pNTn53~O{p^}nZ?qvrz
zS~=l|F`@+-mWlWXa{%2f0|Mg3juY#7LTphf@AH^GuhkfNEL>ICBSXN5KSk&wqwpb#
zK1#amZkf;L`(sj~zjxbpjyg%6#zPl<P<=ghzeYmAN4~AHhUI8@)6R>3Hk_%Jl7rdA
zKxHj9<Zsm)EQNCd8ti?6h!KiX_1vv`RY3^v*X<Ou>=P2yA_@kG1jpgOQy0RVRWCBo
zUdvK$0*%sAi2~HZSW~36j06A1t<C5l4!R?(Iu{IZ8xqFop9B2u=#5*D7DIxmAG3Y>
ztE*n^Wrz2I4Y?lfyT*<b7)GZu3}S+vB?G;jN#e6B0Hj^S_Z^;?U>r2X2E6M$ZUPmT
zXh+sBit0Ew#+|+eID0LhC!~wZR&J^wV@of&A&iR{AjDDIX-R_f1UH6?{gDPLJn|2`
z*{Sj{*bEXnJ^7Ol69)6Xh}&*06SN5UxLJ_mZgmuUwz?9Y_KaCgvWTFZ+m_RN2OcJ;
zl0X3w(k+?}I1WEoj=mo22kk3Wb5N9u)e)alecSn=8Q%2L+0=f=gwM=(<u1qF-xvmX
z#75zGJXvQePRZc+Mc!;Q5%RE&*$KsUzP!(~&45k@dw|fpdXp9`?{m5?3lT+r-=x&5
z>pt#oupDyD0#Q{w4iWF82wvJ2p3}TbMOUWywPN?yCYf@*IDbr&&ko58Am>x8AON+~
z$K(X48ow+95Ypd667$?R$gZ@6J8s#FFDpSj1`>!yWd4Vm1>t(jk5>Ye^<$TwjLC_O
z+8optWABXM!puhX60>>()$4N|Fjr0(;=szg+=oZ9vuM|rrQY*ca==amliYJSs2<7d
zjxZbDfUeG9&jd7_o#Lg=O0H3h2X`ZSxTqSWfx{xyc!lq{u?e_tP_i7a`}gy;HE2Om
zer?k|w@DT92lP1KS)A`1)WCc4A2)poM|SV8r30wcpMSwLCmHk~cRMp^izp5)OkU=!
z;!QHxF>etyvZmg_iD%z+6!T3Vl5burl}x3tH&ql|!Xk<kRG^8*T@&7(Y-e&x71ey=
z51c&WltC)fPMqF;y+HX7VpdS*Z(mHBRof-_%_<t(Kiek}N9TrFx2po3dtQ}AU;NXT
zwL;_1ADm+Ji8H|2GyB##kPG0$whyU4%tnLnx=05hvZE|Lz~|dwSN|nH{^K}<k&Wqp
zmmj}LzTek;_s1`un3!>(goMc2uW(gbK3BNojusNFUdnUnbWhD;UO3_V8BVvJ)U)M|
zQ{-hffHs`S^cZ{E&O{7{1{kLkt1B})TiXFmjD#2qbV0Z`pYIkzpU(e`+YOm)?se<&
z#54tDB1=9Ikpu)7%Vaj@Z`Z4}hdpbyoZkRO6gg4<-o+XF@gDTopRU)V{gpFIY4bqb
z3qajSBcLyK0uhjSf_`HLTEBP6q@oNFKxeA{MP8j2b7Relr_|9}o$7^om&ql~Nb{of
z?ZAAU_Zj7cmX>GZy0Ut<nu?O!aj{gipsSgAq79U?@53>G3=$`S#b{U0g62g)Ej|RY
zIXD;^IUrf!0>FGIrvW<@nJFPWXCQV`WHgf8_)zmC1kq5ra08-`#c~hr;vwg_DRdVb
z*^CZ{*g#^o)T}?rU7)>&C3vT0=w0;ksn!w&VKFv^##TL355?Li4N;=GZ}YVO=rEBI
zHIi}XA(3gMD%z|uc%^1*76L<KBM*4>5xzXK>CslsXO=W+?4Tkr^>L<mgK(CkQsTlf
zl!!kb8*L&(pXHamkEBmHZEYo%j%!ngs**a*7u54;zHx6M`BBK>xGAPQM7!8QVj+dh
z_qQMs%n9&f`t%8;F9*CB0B(T73`q)aQR3ocK31>qg;2ziG$H8$$sjN=eGpB|nV~zC
zwPQ3d86m4CLNU<lgGrqnHI4N}mV0;A%cNMP+q<2&lCa$l!ph6(9;mE&KAK$&jMTfz
z)_JO+B8etBt2Bx@9W<EL>(VOPyfOfT+K#0mAe&AQFS%uB)MFA5tmv%2ZV8P#O?QZA
z^`wFzf@hwXS}zusD!`}PQ=@q!IZt7@(?K%(krD`~McXrkfr7~a0m0GjYgPIT9zO()
zH=`hc4?vojqmi&Q4cnVkUR%5>&pXY}C9n$>5y93jbp>cvbl@3)iC~7h>;Sa46a7GL
zM?xiD0|*So9zCj^UJ1fh5fzXIg%7X1I3ZZQ_<-uBSGIA8S#ZNYe>vqMnjrAvou1y|
z#<+0B$e3wec>d{I0W9-r?c#saOCgf2aqUMO(;J<tWY4Os3#arZ^XbCh6?JH@ELCPu
zJy2}=2$~sJQNg?k`HAbeOLQ3^`I8nNK87B^xxnDa0*Sx|B4gk7byf%v5fZmK8gJp2
zTt_tCXH}(6PG*wRiQfm#t{Y&ZeR$M_k!$g4kt@h;@><S3$}}Y3#$H}*$!krRo;fQP
z=Zz#++?z(JM_HPXI!Iy25JM(8u9-jX_2wW&>{~9%^5tJ7B=j|7x6ZS38AMMv%&hX6
zTTK+-L8jfOo$IL3N7`rfE4Y>?uk8dHHoy@tsSy5ptcGxyfdkGz7!qEty|yvHMBbvL
z&6K<RIKl}rie}DzH$NY3fX{ggOC(?jL{TGFE15_N=R>D9cGZ}OV-Pto_~_mKsMKAG
z3G71*)jIGvkgbt%e;O_{>QESWcsxNJUbQrDB+(RKSs<r6x3ZlX(3ZqX&GL|V7KN*m
zmO4zGW{2^-oxcZ>hb9!NcEijNkl9QIlGLio=e&yJxTJMa{*jge)~KZw(_8c?;l?-=
zJl5Llg#thaW2Z@&sJ>oxxEG&GR@6r2jpLkrKO^W`uT=T=j7xF9AaO)Ov)J=65!UGP
z!3jezLb>}Iy!Zr#ERt)(pX|7N8?3{|;N@xX!8w{dq*xTYJYX}MmQU)1Si_BmMByi_
zcyZW&i9d^KdD6F*c~5GkRnvIYC?Dq-imFl>^%fSiI!f>?!A)d6yh~<%-;Gh^eXPbn
z)0%bkKo>_Ed{o~fcCA^cAo8|QKAb+1acuvINd4&y)h|I~-16&`%#m#j3o)KOd2IMl
z{lXrBT#i8&8i5@CmnrvhROVlxM)F135#h+*cg>C@Atzw&^Zk794|S}-bWIGgu^q<8
zRWYHg50t?5bbJg2>Tq+72HO5?E7M#AHuxPl%6o7PpgWH{#TOa+G!?`(;G=(-uMH>B
zLP=ithpxRwlKIT$`yB4E3|#ojzV&frk7aUxxapGdEkxP|o9?mr`QUl@msR&ZYE6L;
z5CxIH9^H5u5CL#W2SlJX^JVUx-qmw(>tKwr+~Aaq{%ix!InikT*>-fBt(O|^ngX0r
zZ2I#JIroO&x7{1_AA^edzar-h98CZ7J6X;8|Ju&q*SdW%OlWog5Gj$?peofkje0p7
z=W5}#5h<E4%@Rmu<2Nubo;O_Q#U+e7hCh=@HG9-1<2ZS_w;f-EBBdaSci?nD0!-;S
z8U__Ag2?&;#iSEUC?8NDhYza`6s1F|=O^Ee@2T7z-}Xm#??Y4tRVj|OKm~UkyQ{tZ
zBW~z*Kvf4^`&?&qJnp*%hbR(wFI7L7y_-yWu|Og<bSZ>Ya^4dES)5OG+iHUZ?114k
z8wVm&QXXVdn3UZl%vs5$6{*y2B3#Rt{fiehVsh)EAJ&O%BLNf0H^w$x6~M+ftjB3j
zXFQ?*H`&liQU7RKj1e5p)EoRa_@J5x+6qjD<VqAxYS;e{ot}?KP!K-pobU_~9YCS{
z3!jDzPtk@Yv`4opd|@fe)p$#XFIq+H3<KEBG&X&RkT)r!c14-rY=?kCA54B%n9ab8
zyx*JFwg@{#zdG|f0-ZeqS_lEBMz_R;qAia;?5xs-B3b1L^*OWqdmzsamm#KoTW-83
z1XH%vlzMevo^?Yr&`#$01TLI>4*}bKb@sdA%|L4}<<7ILQt%rK)!YNRuC*zN(29%j
zpQvJe%$}I)11F>s`?kvH4bY6%^#Z7KZQF*wFTvB8PaXY;q{5A;WdJJc*LZLcY6lST
zgz%B68Ab~i28<oi-Z`p$77m}BX)|OUWbYQ<i)Osn>Bc#^6Y5?8SmQW}yW5~Hv7!Zc
z2mSx(A6g$vO_z=rxhh{@oMBPG2}_f8HfCrX@AK@@&LB$Tilp0OCNyzqj~^&T_N1G|
zMx-M`GlEo&Q*98qGMVPktc076oP*f-LiCoe)BV7IPkr#;TRj7VO>D0?)YEC^(2C8>
z=X}dV995#M7ixP`jqvTZ2izNnLDS>S6xb5PQ@^1+WT-AcUL7tv+oP-3%zWv~=7H=}
zv<veYvwd*Tjq!G9t_4&R8Mq?m@9iJODNE;`_kyw#d7q-T7s(fBa)QRHrHn3+%tg`e
zMf0I88Mq42onU{=7xUmnRlS_qXF@jF*W4J_i4ZSX<?F=8WhkFhWa>d)F-sIb9TTI7
z-zmh1d`a*YOHt#MBxLUnVh?kSFp$bpjQB|OVV`o-ua`zTDg~$cH-^zx5gblLT!@ge
z@d@DXQhAyhYWBUnL1es_JAc}bc~}f=B$K@!_ccVgdGoD%9*Ccj)b!u7V3g#BKq%nh
z{wPnB<pfJ`Ma`pBtp~_gJ@2lW%X*hQ_inPF$cNP+x@2i^_1DX*H9BH8Qvnv$434n~
zEP-L~7Pw#?OJ>52^w5OYE7&yB1IhP74VN)i`C5Bo34lOcFDwLa+O<Nz-LA)Cm-gQ3
zj``db?0D}-H-yZ&fP24r5_!PgYuL5%Pm7bvq_5zandW_i?||jtXYa~tQjE<AKnBil
zl{9T7&1;C@JT0t;1x=qLm(T7HfpKd7*H7usf0%bntpBrmPEy~AIcSCT-svHVfnzvG
zGLme*$*vx<@dWi?cM9-4Ljxz3C!^zbvEX9P`nZq`PbB6VyL>jTJOvky!~d&*Rp+4*
zwxrAZY5uTxaKf#rY;K&WZVgv4S?Oe>5O=X;((`VrkbL1|{r!B`fTGMIY0R|535@AI
zpY!v+P?I71qxP4IHl9fPuLW^DeTH)!`QN>qsgfR_cjE_8QZIzgr*lRKcO_vBm0;nl
zoS7)x3v|)>=O>2er)TbEZ7#W$2X~CJCfI~cRYIkU=faCiJ|<KsHWcc!8M6Dv$?~OR
z%;)zu?Y|}h<jv-#bqxX7c)3+VOnA61=|-0;#tt^FXv(?nPkcMo*4w2|=l+p`D}Mus
zSZ?@s#FKfl-kWF=t`}EaoLJVKWEWRETuMcgSD)NcgWlbX6rxT9o^q~yJxZxKd_5|0
zKkI32a>5Fox&FX3`xJPugJX4!4!?3W-)Zgg`B$oPPQne#*ix#6D~St3eb49RK*hD9
zhe(tP&^`C}UA5VWU{!9^6-?6?SeM<HfuG->8_wAH<l_0vz9B^3J&I3(e`=G^Vl2+M
zSM@2Ewux1gM=Je+7>HMb$FfYz1V4%SapT8)tXLBVxKEpmpPG|4<Wf^2wK>Fl{#drT
z82-Qz*8Mz2cizQViB6`O8Af$4%wH~}b5yXu(Tn-L73_2u#~29qs&h7Mf&JNBu}pKb
zhFT`0Uv*$;zgVxcnb}G!8l25tVs_=90g1vF9}w%{mQ?Ory%dPd-gqBkTE78#oC`K1
z7ZnKjR~ZmvD!8y(DB%%NFnBJXTvkqkIANk4c6c?2-Pgts3+-!&I|}JLe7$)cg3ax3
z#yF>6Bo(Dk^mbVXN|!kq#?})I=>YvqO97f^ckWMTLBu`=QiAlN+YcY;$dDn`4lFsK
z7zb$7OvP`Q;C`IZ2+CX<G+w*W0;itd$YRkQsdn})Q&CwyFJ+m&#kH}yMrs#Kr$Afr
z#uQyVR=zU#s4b%WC#rm#x%}O`08N>yzHR%(nv>C}NYT7H+7q~|PFyk<a6tZNqg?@V
zNRjW_9p{|qp#5OBm1+26G}?>-7>mXU4_Na!Y)BR@?1=^f7~0>L0~jGy5Kpj&;wXV&
zn#A89lD?UP>m2*?p^Q-eSB<W&y*C2l);zgYZmgYia}LMj_T`#Shr^S%RK!)|E{r@!
z!8$s*(vTn`AhZl7;vvIFu)?G`4s5D4K!4kL%Zv8d$axJKCNR{>O}Vog6%cX4#FoLN
zc{5R-nZNYKP$^PVD)5Cy@)HLIcRSF5R1#vCw?%I17Gr`Cs5<B>SKsKdQ;?%3%2DM&
zzqn&79W3w{U=u!jC8T|Df=iUegQq`uRJw|$D>wWZcATk-SW9~=rqut_KgUGKe-@vd
zhq2b)?iIfEYTJU{7WyV~{tC+i>U23=UGtM#NMo=^)zpP=#>u)13d)^3>pIgZjn<r|
zXU<*>jRX(SC4}uHl7uxq6q^T_sxft|&7d_^DbRZj?A*W#Te;rR?LLq|M^nx~u9SIj
zaKO8{FR|;QehN?p_VC;!pZEyrOU|Q}e-zBer<0t7?&nGZ)FK09`s~*c;CL{V`#qBu
z5a-3QC|d;zf%G^}Ddc>}*Bj!$X#s-#M%Mewa+t91HE60ki32`jIfF#yrN6wcy@FL)
z??@0FF38-W+b5OtMuF+x1iFD!-Ch9Wo4R9IdZZ}0S|8Cj#olPh>V~4V+SzFF;}HlC
z*0tFzVi5jj=$RK!I(Ex@=qnfnWXhTe^Y7);M6pXd%>OX;iSK=yg$O#}a#!2e5z!6o
z?80~JRqNGwGv);qZlu&cuXU~)Z)X6J?wsI<xL<VV9j!42q_>~Amn&U=>CEq{4DCXd
zDgZDcb*$WS$&<m`yjFPMMWq6IqBVpBC5-#gjeEUICIDugaUIg-7z6yc`b;8PBz$s;
zvK@Y3T$B&)E^1dOlV8XN4umpzfSq$Syr~q6%6``2gTJn)LM_T{=x7I=wb%ob_x*B;
zR6xzlB*ks?u+Y#g=6$o=<kL>XEI5k@7q4JPX<$deFHWMJfO?q-WkYkdo5h$S&6^!e
z@QB)@<C--oCTM3krc)|r_me<nHzYfSv&+9Sw&?|oFvj265~CKIe{(3Xy-@;ME4AvL
zHiU+QbbaR8CXxy&<Ywn|$b{BIQqP)zd6ixc8n*K;1~sDotMJmYkDNpSb22dZG($)f
zLWgjHbJ^hkPd%Qf^@MSrv*iSf)W5D|?O*i%Z}I7rusSRVg>?3DP}&g{tnkfrmIw{C
z82wb}5;|H&sHiN=%XsQy;%F#!fv0yWM-;L_TZQrShKo{_RKu@qc=|(dWev+0M`lD%
zpC1Ty;&k7qdIuWfL5~Rf_0Rzj=`kNfk6Wu@q}?l>jrqfVIuiX&!hov_erGQy_Z*Bj
zeoV?+<xHNg1OKu(Q{FhqamL-Y{G`R6jF3S84P#u|mpj2q7b`Z#M*}RBM=T-j?F%<K
z7(gx;`k9`x(2|<ODcpZ(pA}o=FX=w)^k?x3?hoqPA4<VcC6uLq94J3MHx~nYuAN$4
zIcczRA&U=3-@C0()cy(-kyC!{7T@#E@*u%Mc+>5(#vu{MEQf-cr4hRf4)w2c%RS2j
z9U(Ae@tn#M-Yb}U0<`y(0nX%)pN|$zoiEB{RZ6T~Il&u?bJ2=>E3N8~;*0_Ci3lh$
zgm|a?h2iuR=TuLlhimtWMtp}-Ec@7?o^HB-pOZP*C#7uht8J|h@CO^_HF+nX7%cxP
zz-!?27}{<CLIZV*y}|u(X1`HO@<V}h_V2y&5HiODBZ>CAVp#sXM{gk3!`mfWw<1hC
z(d8l`QMEUwBSp?;i#vis24sW9A-EJ0%hWyc3%<S>!P)~#D43=4kshD-EhgR((@Ry3
z^s$0tt*K}CgZJKDEzzm~$B&<>ud|&@1T`E>8604JIi&eI`YGPwGdvt3bmy?)e}lu-
z&8+{2JYxB;-6jkB|16J=)FmB%uZ4ci#al5`5*R!XG0vm`8>;N7<;1Mj?F3gU;fTaR
zjqt_zM4s}h!<G_1Fbe-=Lrce0K{_IkeEidq71&s?3SY3C>TaL*_g^|5Rw0S97O3n5
zUn!qR_^cS1totp_FbUa;W0{$o=Z_uFNqlh5&CSo9+t<<ANK8u{M#AqwaHwBHo|Hn@
zakz&1_1hhrF`ywyQoLn(8*#=%YN0qM5*HbK_DQ6|Y`(dCM7eqM79uT1p40^&unC(W
zW|%P!x3%xZflY^XJ5U)0GCz=!1O}5N5;f9DKY2$70a;imQkblwORfpY01SSOx~2)e
za^6o_t@^%I&rY^-@DL&*RkmIA2lRVwB9b_{QA2~jf;dTznxzjvOo@N!<7IP3`!A%d
zxU{3btO3e3r^&jx<;JIkFi<KqG?jkh*8MyGj3Au-5A`ujHcdZ;1$3;t6Y?fgGGiPK
zY(5DoVr5;jzZ{Cb%E^h61iM-OzwxQ-&PF9fx<FY+ei*}W21>8=YAx&fhkP{>!q_(<
z70;=7Iq*D*;MhexMaFiZV){0J=-3hRNP6saYOskkS+G$e1GMeyb#Lv)46vn!=Nqf~
ze3gsJzt{u__ZIm|UOL18Ar#S&2b6TWrt3(IuouNE0t>&y0*lP?h#blz=ePst7N&uo
z>39<^@mHs*{-%P!s6D+kCFKW|W$HpoFEm?W<*5|u4o0F$2gAs=U=HLxtHS1RvHrxU
zAl?h;(c0~OQgK7LJ0`E@irvME5Luttb6Z)}9vxT@NMEp9dP#AhZgs2b$4)t<WpY2U
zP@rA22tD_wnZM<Q6(k@6zDT#&y_tMOw#R+S%|g)v!q{{YVnC2)z5JSPTKdI#Lur0s
znp57z@ypBoDMPZx7bMX*74^^1#Cdv@pW+zz4<>lq7sYR3{D0WW68ENXODC#NprDRz
zLt_wI!7UQOV?Ivaq3Pb?c`YD%E9W)0E=|~tb446KnD6xy2ggkxZhlB_!47hy6va`T
zU=Rs<A~A<Q6U57v(#QZr3xF~tvb0YalE_IE%4QybG#El}NYZD0qd=ssqCL+$Svojx
z?Edj-zcN>PO_gW$=~UV9xUZQ!{YJbZ33$m=^#8b`Bgmd>+}l8ljBx%fW$?oW%pEuO
zztBzp7XzEh&v1qS2L_C^x5Ky_m7-4E2N5MCXMTIfZw`f+%qL(USvvzfHaeAl;>~qC
zf@{W#7FD5_O-OLA@VB^;ClM<o@lfqYkEE999P?$|e42-Ku*fU|6c7g!nrpj13dKmz
zq0fs*TRRoPr@Z1Gy&^UYTW+@opLaKq6)2=}D^Yn_#j_&_44rp!GkMmtAWJF>kN*Ob
za%IJ7xl9ea>829;-avj0pyipq+or-n{+S?y5e@6*yo(h1z>(b;03xzd_KncCbPLzG
zSg3!e=<M@s7Yswai$&Ydvaf!q+>Ef4AblaA4$K9|8bH#M{Gqp|F}~rPb^~7HQMe-r
zw^)A;y=_C9(}K8{mL%Frz&OhhH2q(T*944n^bdONz_tkOcylx^(XGL~WATRF-o)3x
zhk>wEEDeTiBvk1|FX{@YA-Z~PQ{eVLxG>T1{4xB4eSJd)D>)kP+Z#-KegKnUZkI;S
zmbnOXeptgQ2$3Rdq<jL;(CN_IEUeSUBIc>J-@|Kw3VTCPJYb=V8uFG0h*N`f^xj*>
z$*@*emuJ8*VAcUGVKXCDJeSz-+Ho>w3_MRd5!C~c5AX7D<~7O;5mo)~t*iSBlrRGi
zPy=yz(^1IYf3+y=X#7i4XAu%zxXf-@H>{`^@<gk(ge~ko&wPW58J^TLoR)FHTvSZd
z*2#O~iv2iz^2H311}Ik!ECw*09QL@E7trXr{W^qHh(9hCe9}2TGHOlG+(eQh)|B*s
zq1F>e?Ux5%*1{2xNr`l|yofxP<d6swat}R)=!n3QdR>@5=`We<XYzO@xcr6!g4||D
z31tn!go8O~40Q4guPfiDt?cl^kdCkdx89Je!XZQG28lUQVQBpPI`_!)`x;^bfo=C2
zoOg${whRn!=XnP3s0a<}T#f<!iaBAdGqySU!wipULWfK=xBy8Y0~%W9^n<Ku5Yj{-
zB!OYIwP1A$1JI8O%NTv{KE^mywm3-ri@oWeSQ;&hWd&p=5_r;g?Ao$x{Gan(L3bc=
zOW}F;<%dPFyW{d<P`hj$P3vWJ37(ED*KWvN%p<DUtgop(_jO`Li17jG9tQJ=^GfpB
zXw6=~WOMLL+5cT;jWKV{yC~`tQ{xcywRZ&oD`_8+)x7RKJAf1#K16oZB`k>nqMe=r
z%RG(H1>~;dM~B*|1{8|BcC((dRj_TmW<R&=Wt5JhS#kl-Wjz@y>N<S<>B-Ua=U>`g
z_<)2o+Sk;T1u)?Q&ilTpA-@EL1#4D+LRUr7`B7%As9)KbD!*Fq{`QTj8Som;xxE8|
zfpbNF?7te3oLio%E|-|vg7_(Z!9#$bEftMyvqdLcOU#LXZmrVlFTQ?8QfM~=EJVak
zlGX@n6|p<p5mjWyS=S+N%2A!%JY8PRdLdZRjaZ9=O9rf*Uu^DpUS9N%gla+Ma}F^a
zL_j?zb<kBqP?ck7xN6tZ6Ia|T9)soztip^<%L})ywTQAsOEV5Rt<dZ?Y0-Fj1>X}p
zD_HZ40+cXR1(!<_6rM|&rPtokA>Xr^ySa5f$>|E&0F<?j_OM(%kyz0xWNm3@)UlUj
z?|Mv5IZ#&~g2IHL_Qk&0g+SDIq9F`p(5)m=h4%i0Tfqyz=k#=*iU$66(KwhodWX20
z16ca@0zKtcIqp1)nE2#8?*6T=spZ%khG%;<^g<8Xf!|x~-z_DXRIU@RJ3S>A!e2?c
zrYIQH;`r;6#}BvAE{tl2=E)qv1_=C5d}^%{3J-7(_S@eM_PY)dk1~FgMNZ=n8<LH)
zl%ZJ{c~11BsbpCX0i`R)Gwvm?{((Vz-)_hl6Fs)JCUQnNgE0>@Gc*L+Z^_Hlbdk9K
zbKqKeU#moCG|qO<rTsEz^*EZw?kvyXYzWk`Gdzxrbo*}9;xkuhK?bVA4rVc4+6I`3
zugY_gJFQlDo!MtrP$Al|)j@%*{@r2jqs|&gG-qbiSX4PyYm5cgsF56Q+9QQ2MZKDZ
zBrTvVK%Z3Tz+wN>DSG6H0y}X8<4Ol~_$gb02a3BnD_<H|H-AuGMrM|x?}*zK7b`@E
z5duH)=jZb@0f0P>1X;}58-o7<Y|2ejYe#92QmhUhDL?IX$};j)^;K<sr{stH`w#l&
z<C@2Rh(Ol=%8Roz{?Bdt|5pV5zZ>RTu@*FcaDJqz$6c4q#w_#VCTB+`Tn&E`DN2RC
zf#%|c?`OJN0^xaw`Bw1^bNYVIx>fCeJO257oVk4y3AV@NGde;<azaDgaK{yVgMN_)
z0>qe;DSghRr)L>&@H09L<zNl1dO{L_Oq+$aJzPKL-uCXC2l2s>99d(j!Kx1nBv3O~
zGoSk-5vmW=)MtV3RpcK`0vwMEz8t7(5^9D1Z)k-+y1u)qL8{U!nQP@R4ee|f1{D#-
z!*!LM#=jO=-py0K^&^tzFum<>^47^QrTzTl1)NZvzECaEdd>``Gf)t6tpCC0FAbWp
z{6!ki$|*oRNhWB#M0~&gCgFB7fn315|4?aoF1J>ayUuo+L6o$oK;_kD^HKYF^J?9g
zMWs<Q!rm2SsoGVYt+|TIL(+w#DFqe|<1+|_)E{+23HqKjBQNQKZ{Q<JXd=?kL!6`+
zpx^t=<<rCZS&%JprT!wAV&f(_JnZ&=Z-dI{4+$;gUD>}ywI1QAt9wJwgU=Ca{hugS
zvWXB>xRo;A-sRD}a+LZAr-9V3ezC3*B0MSIcy&r(lpsztpTgnsAj7EJaj1b??BAGT
z9Fr$@1*zvYi<VU8d$6D1S0$be7&fxx<0u7QlzDm-pW>L~*iTx#i&NvsT{c!E@SNQq
zpy%}q>Qd&f8PLV*Alr!M42GG%mH*3EH4%T<&){inoRkx-v#~ujLI=ni9J09}7i*&I
zs<4NYbq)2yH+1hEtg$-SCtR`rev+cSe3|GvWxx*2V33M)|MtT4AkU7-!<zUSdZ<OH
z0SV3lEEoC~q+wY;r&D%5n&LT+*xbfrNQ+l4s^MsiR(U9ak|_uzjP={N4l0GDH#b?}
z@JaN>f_1cqS#L=*eIc^QJk>fUY^83pc;Zw6x-|}2MC8G&_Ts%l7-=j5jz%6qIGww+
zP+MBRr~o<3b9YWyRMI?IN^q3J9{ll8epbqDU4a3+tSo=7c&OjhpX#OLp;<4EDneRT
zaxGrY8k7d>RegJV!k{_e1CR@cJKD%!85v&3^fy+asRN3Kt~|4<>q}n&wV=&LI(bXt
znbG$Qdft*L<{3Qs)&j9b2Jyo>%@i|ot6+|gc1~0t`f$r|KiN7hTgaFf9W|J<*}(Z1
zakW8X>wNW(rf;Mwg(V8l&cxcrfd@^cWka6Ac^^-W>zD))hLv75%gak$xs(l5L@0qK
z{ojuI%3Ng?-FVR-tt}Uu(c<EpvbTp_4t-mZ{^(n+{(=}mCou!!!8)`MD>?BTST^H?
zBV3pQ_3at5z9vloFU5VgZ=>Lh*@2%_tpeg~`=MOZ-mo9`uQK5Nk>60Dbg0-r6CZDa
zwD0X2iJb*N;+wpI<EPmD^~G;10qz!cW2OkIn_Ou)`CfG<6~HQ|Og641KsYfNB~}eL
z(|eqk?Zh{S{OAzoYeEY<?kvlH&2DNc0gt|#YS*XmZAq}Ly?Gm4(<xL2;35jwG5A$G
z%EBN_Uw6M?Mi}$;vfP`7D>%dq_v5C*1)@Qs-g>)|lr=1*dg0{1BX|eu6)6czU|&;k
zzxNimO<DEzu#$PuhU&IJKb!qMhglbL{o23iS#@8Q81_5`Y5sI!q0vJ{f0h=4tkIEN
zfc`myZLgdk?@dcMcN}T*l}u=W)InXxTg61Y>r%_Kd^Sm4S98^%r+Vv40RQZ((b`)(
z0M<!qqKw;CCn$?QItan@1PM(G8=Z_DHXFmgaMxqfl`Jaw{YXZnF+md^amati#VhLG
zqZGU`+<wB5g<+||dNdF*`92@~lZhmVB#b}q)8D?E6h@EyI*%#;hp~4G5-r-=EYr4a
zJM*M%+qP|;v~AnAZJo4j+h*k-UEN&~bsz5gj{USFzP;x9<`@{hH|<+HcVDN?m2S=d
zCiYLY)?;}%q7#u_b~<Zc<%9WDJRKoeK-q%Vx>7!@6UgIq{{qN}OO}{{(9hk9Xhh@N
zJ|TK$D-aR5qzOdXAMt&&gDC&X$BPem%7k9+>W>{kE-@utz0*Z-^zeIBt{SvPi{&qz
zBI4OsEB;C6Q+p-O-F__9Ukcn}Sa11`Z{c$xHJ<jre=MA-KH5zHIt!u|xJQhY-*k$3
zMOz91H7$y)xx+-`v<b~!M9BPIv)j7@1C!Xlh2gq8`x_eluM9-doXQD)E#Iz?7SvQG
zM_PX?+k#aco752a3mF0gEgC;<%Eyhl5oGe{`XHil;xp`sWJr+BhxH5M7NZ^`*T96g
zcg};iajzhC%X!90hXSsr4rb7G65au!lmu)AfJM7EzaOci`Xxv*kc-U%`jU`<qY3Mz
z9~D95<%b{0Exh6S2N2Yv>9l9UW!#n2-L2@0OzZR9tcq;ZRnF^&#{(e~(%?EA-T~Y#
zseIe-8aef=W@XKz3@yA3Lxe(H<WINF*QlTWoBSX*UkmkN8q;}yH+uv2p;Ob2we^Dq
zB=0M7>@gtqe+t_FFl}LEXZo*$cIy8RT2)WOfuTr{b)|DMT(@=OjMHOx&GiBrSCV-|
zO6m;RpTBN^@<ir5)@u@8hyoC+I+t|3UeF``G6>+kcwfKMMsF!2T=N3t6p%oLzC`|r
zpdmr9Km&Wk65+%HaWV(@=kG20lfZ|U8{Hot97fDgeDKk72)%=u-6?##ZCGABJvbjO
zZ)$h7oG6J&C3&+UiwL6_Lo<G9=kX$#_n9@Bx~p3F&}q*<?C)N}0ya3aeojbASP-P)
zMg32Y5S<=G25nGnM36Fp8hBAL-`tBMOHh=dJ|%l%j$HpP-eyAt{_*>7AZ>^ELlL1g
z`(Ug5nQ>siE2<j(19dT5oys@Nq`Q6xFTb-QgDfaQT|wb?K3wlr?4}WD>WM?@fb)Iv
zI01s<@F*i)?Nd8?Cdb7@4A{R%6-GgfL&I1gCs(a57-j=Q{ZUnEu^kn8Khi@NVQQ^4
z@f4&(9cC^)Kcx1Y+xSK4e{%&aX}ElxLlPdcDWy0-U>ZnijuZ885axs`m{krH`Ua!N
z(y`-AKQjEZ%K9L|J2g?_dLZVy@HDWmtS}tDJFwDR#df(6<E>l@GUV>WWR%i)S5p>A
zhR+F|O*r_egwr_hN}X6&;-Hf(qngzI*CwU4Ju@DYOSMIC9GWsLrdWlef-KQe;bRV(
zf?U$TmO|X1m4la&%*~9l%C?*9`u<|Xwv<<uaX|m0(||k^2l|<EIVMddvbO~)F{{Xv
z^ioWnDVnQZ2telzQIBE%Rht}O_goa4VhN!t8Fd17T&X}cPK(_pj`BDhakY(EG)b4k
zLjBjT;W1O06BAaPR8|xX%0+>;P?e6V_dK${cimY-_F#c2)9A|ph<_@-N-9zD!jXh`
zsPIG)TS5|?_(HjsJ!QIfzds+aqd%w^L=@o46Dj3hG6XUw?PbDetC>jR^~6_SHs8Mi
zBI^tl)7<8$n(XrJ0auw#6YKxkr%CkBL;X@zk^PL_`m3g>n}3`#3wkj#&<d>_=Ff*Y
z8Y*joy}IAeMFCY~IJTd|rU@c&%RqWBEA}Tc`j-aWfUD*&+R8Be);Q}@aeU9(RQb(z
zwGcQ?zHTwIfr897UCrYn7FI!uc~bf^sbZ3NMk34dgk)ieWFH322Cl-jPlq)<^8svq
zETW}&b-dh<i&C+##?nse%8e@U0U76Fg5NMmID>TWZEXEPD3ec18+MJDKa?D70la$b
za&VwD93w9@#WX<7B*Q`XnpLeNgwxSoal9+emX&5Nh^!luHP0LgtjQ?fA=M&XezgQY
zyxJ$q9~^3r+H&xt)rsPgo*x930LUY$Xs@WkwjOly6pjP@*@SE@uW}0Pa>%Q{XR>EN
zH0s?s2-;xUzHoJdN0>VG?`qach_9iOG6!^gje*#fu?)Hv`jXdv+)~UZH@&8#$<TTz
z2W^T%`mfM<q&S;9IGtB(ZHNK%I$4HtH{N)s!}?wkC2)j>OA|kY*S`ybJ`4VU-JWk;
zg-mw_{7>4GSuVF>i-hZTs{PyU_@sn6VJY#DD!a?qLcfac<-rgjsHu0It==lI<RnXJ
z8Cz0XEE|C%$!R&AyJGazQzT`>TPsCF@wU8)@2^`k!ln;1aDBO4BgzSUxjD4EhNAvS
z`1tfjrAKrRE&dAeApVeP#|^Fv0wmPf&##O(NGc2z;q?Y)4~9_t)+hM2f#H*s;MT2u
z|1aL_9;L*)pFP0mB{-hd*GdRFM*i|pJduf<q7=>yA{GEt4OLSH;9c+Zc~?jUGSuJ0
zlV>|A?euaDQ<38usFsw%JORivt99it7nnlIMbZYd?DGEd1{QXey=@>!E3dRmNPz$#
z!a|r@oROHr+FF!b*rS6+iyz$^tFCK((KPg8>toRxTjE<#f3)0hp4d5tKdmJ;-tmdp
zjt!-Keoz6HX!0}+`i1J`y7CaPa^~XnZk$9*U`*ZrK2WHN#wu##c*F_W!C0%kk_0LK
zgd00<K&{yZ^Cz(9p!N*7MqVpb1Asn;6mQo)P%dgjbY`0NERbPPBAw0fb32;uh1<$H
z(Jwt46Q%g*x|oftX>F~U(j&@DQyGPc<*sjLsYD7PB@V7PX#-Y=BQb?eS@hRYT<fXY
zb)#rW8O=~d^F=a0mnC0I@;lj7-AEm$UJvx~(J<>?BGTqtv%J)11nBUKVv=V)Ory{0
zK8YAH>{U^PiY&Avo+iC#`nM%MVBIip*{qI^BuL{eKWuDRn$|+OvuDUf8T;cabX2(C
zsQTC15wvkB=qqN+IBlpz5p{t_+)G?}o|~*0>sfC!=pp8Gdy0HgBYfmTu{E4~l?)Vb
zADZx5t-wx#&Tb;^y^Gvx<wL5xoYE$egaF@i`L4=<02DgLF?xc$2tKNt0I@Jn@*`T8
zG>JK=z6s>|Up(gS-u?h~6X~1$hsgaUp8j7BTt;?Aj{mllHmv^tw5M-sXRDbMo!%A?
z(!er#{lILN`QattwY9cUiKJpkPv-P?2+b&z8r&kuXdUav2syIejwK!rdH4Rss_|9T
z)aY3zCsj=vr&P6sk0TZS8EKYK%{-LR93RfD)T_~h(iQD2LYl+?<^C)|=i~XYae45p
zPYNmJ(4vtUS=d-2S~+uiY3z7=I$k<ocIeo5W-&9Ni<UUv8T7tioxh;xacS)E^d;TZ
z%n1{w90?Y08c*M2(zvO?(J=vgMwf_ZEIr7VY*Fj=EsraV03tAK2*L-Lz|WA^O#~G-
zNC4`aQfBa<l|~1etIp`j5O|sK`o%^@Jv430$_}X1-8m<N(I_KMzNTRJT(8UUEb34j
zkZtZ6my;v{Oeu<&1dVim>4#dXH3tqsr~TWtJ+j{g68U}i$3!G|hZiOLL?h+zR&a1?
zCgXFhO3uMpX;m+5R4oMFH1azoL=NPzUNYVAA=aVGKL|_HpxX8$sl0CEjz3-}6guoq
zC4DPF8#IaiF+@RK1{#or^pa$RWv=!Q=lPs8!tttf7U|Azm*6;Ch{OA(wHqHJiO%cF
zuE0u?VM<%4Wbwf36<ONB(+_RV1@h8p!-ta{^`kxScqXr}^0oNlvL7B)$1PSF*GA2R
zRMqP#U{?-q;1TZ*t-tDtG0qILDvCczeVKn8ztcZ~vZ!#T{*XB{^29(r?ik6`$p8cE
zaAI;3mVuBch9^pLUp>>=?~a*mB2<nkh~%w5OV|j*O_$J@JvLEELZ_UZC(y|+C+(g^
zgZAqP4=^fZ^(hsbmdHC31@$JXfZxm;v*xL5mT%;DynqD^!?$pFW)2RFdb;ylZZuEy
z&T4RJKYVy<yTHKxbw27hI&53mw^v6wbLesmbQEJkk5g_yGQq=x_&H@tdz%=G;PGi~
zaj-{#aKtjEQbNdh!!UtkiL1u7+|4LuzNvf}t<YJR^aw^Gbzes+GEY7IYa6@?k=nXt
zutL0rgdHK<%4uketqJq*NLq1r6C%am*PYUzo+8EHURLA6o!!U`bo@qD^q6P*Dveuq
zl>1!+j7-4s<i(oG&<6wpOVfgS^L!8xVlen`)~o{@aK6gfIpsUk7UX@HVyVzc53g1G
z*`gVa$s;Y?0K9l1f%Bj@cuWw@xxaf&C$Qz(O}28}EB6~nW$rQ$UL~)}Md2X9m<>_C
z37kfJ!N0eBxza;>PfHH5up0Y?)~tz^Yy7U6L7ZA5>TsGbkH|+fhhT+P@5_k3fL*Rg
z2=gvSN=hOM5E>uT-v!y|l=}q)bhPWeaCdhVc^O&C<>&(EnK=Mk2LHVwDuOA~CA0uJ
z2vscN@{jlkO49oIW0>)qs!l&DxU;X=(2jCb^ykzCV-$WSwy0F)YFqH!8Oh}5(XSpZ
zchF6kQ|yZapTMM8XH1yPnvI1Xqnqvt*g3=LsmmDF=iH5cpMm;fWcnN)-~%gMK@_-6
zH5(TIb2GH$^q{Dm*!Z5tKkTDb#}m>sJtfF*GrHu8DKx-mX0^zWD&d)SU!3u`{F73a
zd$R>{Pu@LFwDhi{rO&DZfr?2B&(Ad*Uf;g&f;`n4Nc*A|xJxuE^ur>Xr;TXw5p9eL
zh%f)bD;#zf6r57-VJbq>DeXBNinnc$zDIz;NllKo3up+N6>)MZDsBo}%oL(yb`<A1
z@G@pSaW6a0vsqSy4Jn~cH&L4**^Hut4lB!Dt*XXu^8wUm^V<Q$h{SiD0<b&3r@V!#
zUHos3@vzu13_<~Bg6`LN))7$Or3y?z7w)@6C9{n<Lu-d)Zw&2E^I+4rI*Uqk))_lC
z5rwWa!kR=PLN!iGY0Fy4RYG;Pa!C03W_u>#d1+mzNJ(Ls7Xu<@Yd&^A@F}~2B%uU>
zAU<ly!Cj0yakG%$v)hE+H8y`3C5dD1ELxv+Vu!|7wiJW1_~r@M<{#_EbhIQA9$wRv
zg%ATy?BTp4(RT!}(We3I5c{>jE+Y`oJJQ_wq`g~QZH0z!j8k0=^O3mgtirVFhTiUG
zi<1z~B^h+&k?LMrMxF%)1lixmAcy~r^<jvsixjn*Z{eLS0YM+g_^cs?mOKR06(Gp5
z$gm+JzuMs9a!CQ)TB0a#%YONo%dTo*hhxH1@loG1O}h^i=@%d&t@!=9S-r&_V8S%h
zp65*p!mQ+kfy6}2sR%&Aw!Yq3`sfA+3Cll(4y59Fj!P_Vofp;}=<s~ks7?wgqbd>0
zmT2d+z!`FZuPMRoj|ixzNc%^Tzp2vS!lL2oRR;JZ!2>dbiSk6*DP*$4Qy?72UlwJ8
zU~t#Y$lMi}@z_!!djv{d|1lT}WsGsqt06nvqY2KZWI<s^TvjGkrtWr%OmNcl=ccsV
z_%P06qzLhD0xHX4Wmx8}1j~Bsv9b5{6#8m*$8s{u>Hp4JHs=4c0Iuv|Z$dyXXK1P9
zYy(9vOTfs$@IR~4j!w=5?Cfk%^#5Bk!^r;McQL0lH*8VF5WIJ4Pe*U(7gW5<E&mDb
zBLOQC5lR=4puvw=6M?LeJ`=_F&XnI_xtd;XnQ+%}_dQQ4uPEp8x?C_iIP^<Y(e06&
zjJEHpX-@|292P_%M>$2csj97?EJ+O8#ngy84DDA8Vb*g(p*R#8`jsXCgDru9x^Wl*
zq>uF<B``ABtbvu(#1&>?xG{7aceU}SE;1)3;8+BOgF>lT%j1AIXq=}(EDI7MOQRc%
z#$U<0WYt0RgG@(gB;Y{z!}v2Svk+65lp~-~ESdxN^Gk=|s4)dE4eK*W(k1K{kJB*?
zDl5Q1fmx=3LQwNlGfoOOMw&)al&A&l&Ei=#oPh!_0GlI`?*d;@0%4c06)VFcxl$T|
z0lp;}L@IAm3jZlTPXIE5?Jvf%PSN*Ew$x9hTqkIsTm^?+<pyKLssQd_9iuZNr~okl
zbjFI9$e56>_tQ2=+P5WSO%DT<SL-UVuM=BK7M_JMmn2J+z|gNmb&ISQ^!HJi6@hG2
zUXae00^(O>kcD`qG+K)~fm#<Ih;j}5^Q^oko=S)eYgnaNctxad;WqZYV16r+Z4ngV
z7rkW83TP%{8nKo*X5DbMwr%F7$zcvQv-m7kkFROJRG?ay3k63hokhti8l4O^V8sHx
z2f3sUer%|ZW*E<sHjt%l(g`fP0)dNSlQ(PNlduZ5w4`nFTRXMF>s0d3@9(ysy7B#l
zoY&7?L}oR)yZic3Yy5bTnP!Pixk*JQ$Df`)q0<fL<I}1>Mu#3%OXc&`<l)fBu0D=g
z$8J1!Z&z<GBwoQWf^5ohfstdj)7Ha0IDCg}%NQ`@a54a>?-{@o=I-q<EHwG_0&NJa
zUP<udT3AAjvdht<nUOoDc7uw<$L~Bo`dk>-sjc*Q+X-!_J-T^=k<zX@j$9&j*S|X*
zuJT~))%NLP&(&*DDC>T@_?+5k44sg>UX+)b8`IL=V=+TdvnsMYT0Ylue*o(eedi=?
z(qU)wzvk}*JXV)yg6IEdz$br?)de0oj8q}JL15c5=DFcnYkOomHFz|f`6g2Nr^_;j
zbtXfsji15OL^W&hlSp@WoDwuQGM>kC0-3ULh6sV{?bQ=Bd5Yc+{U!rkOza9e1N@}2
zbwb2x{_}Z1{p;iG>Er0FQ|jl2&+fWE&f#|{8@^R=QBD`JN~1tdoH{E)t=f%wBrVB;
zIx&Qajq5?iaZFj4<T}X+Y@Vl`i^)B)G@G9h)XZdT?NcRdE2!WS(D$rQ1c!%s*atoI
zMs$+CE(Wv~2Dc~5^|fniZY5?<bEa)=rL6mP-)apUg$l>=;+3(i*7~ui`GC)Sj?-Z(
z@o1uWaQoAl!wz?95c{UR#weKwr_veMe-fyChg?dwr%|YtqHwDqY4EUB*Wa&#F4#-y
zdV>OY1c=N5AZ=D%jY6%m-bY_Db!@m$Y0pptf=9?eFVF?~@a|MJ*8P3Ov3+3UnAgU0
zXzOzJu)3gFut&d`#obC&ydjqHhJpA6SINEA$kn*TxGP7S=t*<mwH?Px8f~PpwZ<5K
zO_Q63$6OsFX&%|#FC-uQbTy99CDx|8d!X=jaU;MEUGw=yn6q1~l|H`(w<n+0BBq*9
z2-t&XO7qOIHATdwj!*M>BNyu(QL?rw<%x+rLf)@05m(y$R(!XNLxG~OlAruC-8Kd`
z6<<&9#O4kmlTLORfji%rK+>iw%zjgIY6bfGVHqhvkj)TFPsP=#gBVy=Y{-&F9CnUa
zzcp}s*~(77dFHleVKX6i^uDtMu$Maa$gNo75B)1n*v6GHmucepMqrmR&+Yp7Fa1rf
z2+Iw*oH&q8Lhy~c^T;qn<cQ&E?dmejl~rp3*26l{t1w+DbTYoyaGEUdoi=idcDvJU
z`OM>VE7Ke{aDOGnX%x_0Y%uf;t)=LZON_jjAxK<Ilw9W{7&of5H^0i)WV{S@j)0*e
zB<ORL!)MwTve{O?n+4mB?C>;Rh+O)gyAwcTeEO9Z^f=JP6*A+11!s`J)f{d{D$Eop
zoO(iQPpssVf@=x{G8(n{{#XoKb+{uH@7h1!C@C(i3LnS0y3YISv8qh0@SZ%sFW!7x
zdbz8UINdXEt2Uggy+Apj=Y9)ggd$*Xq}3FW7H8@xlnjZ|iGSmR?h+rvQYLtL8orNN
zUF0o=bCbZ#95Et^3;u0u{pP;rcGXN`g&Lve<L1opD)h2&QaGyq@goCBZ=>vgasbdU
z2NJ{v8$O_w)WB$UVt6TUQjDId8kAyYyOyg$^(|)VZ@zK$78n>xYN4aBO+5A`CT>u?
zX%B<E1*)+X%<I2kCGO%cVk%_cRdggfJr?5IztXyhTWh|dpr+;FKGdKW+^z7IaV=&#
zh~c|ft`ww7uPIqYo+7NDfl=oR^Ksr)V~p}A|01xLZkImo2m1!W11;n@TVgsg`{IhR
z1*mSb|9Qoa;JLQ`zvG&n^*>Fc+1UT<qw&@MH;t}ivm6Nk3;-E>HYS%PE5DFNm7$k_
zLjaY?y3G$u${ua%?S`h~kAEbV$%y&0pe@Xe-Wj=N=SFy5&2!Gf!<mVn&27MZhm;V;
z@BWZH8jKzsFw1a%EOnDi4{2n~6b<sEH%-p;C^t3p`Pup~@n(!AyvWz-es{W(y2{(_
z6$OjI)!lcDk35p*%lX&U)72qHD0T*4Cu=%xz7ai@p4{$uMlL6)Yeq~M@(5|wdQH4d
z!%3L-;QH@9i2)u%ny`YttIV%{o<IVL+#B=#qgzc$lrn6wOccGtI<8hkx7g0gy3#Qm
z7DI`HwRVwCDN@_Wu6#lJN;y%*#5)osi9`aXR3X-%ndEpfl7g5iVY>^UCg<kjd9_dN
zDt>aOVk>Bvx2|!nxyGSE9>1=Bw#`!Sf;5I1HKuCAX2~aV@cj$zY_Zwk3PWt+i_R^{
zmCCyk`|tUw9{@an2e+vCE(n4y1~~}`q}K7&To!x`t4?sPX79!Y)6#0C>g!jTLp|=+
zUaKEiVvmkiCv>???z_gBtYRVPs*PP$^GqLKvr!(;C~u_ar&L`afxtJqXLhoQYK_35
zd=m{Ap?&^5YzrE)sqOa_qiZ%I{{p5!w3dRmY3K+bbL2Bn3j2b!a-h@rKTA#*O`ENR
zbEu?+7}w1gc67q<bpGk9CCx}1%+T7kCDwN+=C3Lo+Bz(&E7Yt~v)b*XVG?m5=ZwZ2
zdUhgeCyP_$rDxVm@ML)D5O5NGMy73Y^QQx2^Cq6YHP+@9na~!QXw)V}SFmMDLTTHv
zZDlK1$D0robWbl;J+W!~cSQz-xR7Q03X{J(4rYI#I?J%lF}D0f9I-ykI92SU8iAsr
zzcC@RBvdTe)}|_&j+Y_-dQBXAnhPR^@ApBMlf*awTvV;8yC_dcYXuhrplfX9X+2g=
zSm2V~*JrO~SKwH_c&4&!HCL>!$>pf6G!Gjlj1fe1-MO_sUuR_Se2wdzJQTW#WplD`
z*0!(tK|r8dI2ez#hv|D?-laj1b!R`TiuL-^CeleGju_zZhbDv_l=P2H4)3zx8{QK2
z+rS({5(kW&MjTuIPV<+6lLz1Z;=@SklzU1LX{$r{d_P~U$snL{%d2oib#85EzK3AW
zU`0*YJpMeok78qs82p?_>8Hw+00rEYE}NGUYQx6@Mga}@C%f*1svTi#Kcf4LTdI|9
zhG|~LI+RT>2Qw}D2yjLrQHB_Qf3k(FQ2Iu>SNBxqLi#D=g#?A7)tn>-rN^1&d3UxN
z<gK2_v`p#XoDQNr{;?k@<`oWX+2bc}$Rk8x4tn@I_x^RKUlHsnwJ^_`cY@|Jx`pY^
zzRPw5C>b!lVG!t33x%-vaeKrcV!vj1IS~g#_Svt4sqH7q$Bp@NRH06@w3@wAVeXw;
z;nk_IoR8jd_V(ushRC|Is-|)t;cVI+f*B%2A3-=EXr2uIcXI)UiC_Fvasi2smMxH#
zb~Gj)7`bYWR{fqPNw@*=ORlxlIEEChKRC)`K<kefRulsbOU%Sd+~k$zgnWm;Cb^9?
zs|i}O1W#h<Fq-wGwiYKW*+N%^i{38cEC@h4UH{dhzWgyA3x5)vKJeAnq4;&!ZQF_X
zWQpe!1@3gRR53h|poSzTbPSRKvXLqB0bThYi)k_|=WZ1`RF}s_7Y$rXKiwEeR5K6m
zeS@GX^=_4mY^qB_4O{5h2FFR&Vb%}b+0<bGVF1AcX(~d2&)HM#*`T5N*WSJxnav};
zY=LNmmjiSq01o0?g@LJTMd}m%p>vt%Bap^$lhXD?<RUNL%<i-==J|c!h)rpRG!vdf
zIOCHn`5$!lS<kMbnlbe7Co2h^X)+tarQCrk8AY{o4nfms&y8i7`IM@8G?fxQgPy`!
z`o6r*Nd(*Zk^GC@yO3Z=#Qd)Ke|K?j>6C>17O-jQC$X+oJ#4S`keIgIMeibLuD*Ft
zARrX%j{G}a_auOu!q}yLquQ0!HX9fgdG$7W>J`DiOwQ#^2xi)o&5#}iSgSb$nkI&H
zV5jF*&wiQXz}d7$okoQttfs%*G(I}!;{J9YB=v$uRM)ZK?_0c`iXNk8^t@j9uk^@B
z^S7e+u=*K6@X<do4B29rV((uzP!`)#ZFtQ5!te+LaKSyd1ndl^)CW@8Yi6Pk37zLT
zADa_;GX_>T^_bZnyH4SoNlIsYTCm$tCO4#YA2)k=+Q&(O<Riex;~V+U1kgrtC3pQf
zRwi%SmP)`$*U6swYOIx4+Y%o$i;VH%-j`P(bu-BKx7>*E<o@fA1SU6p$@wLQzNJj1
z@+Ra@a4gW2Tra)xP^6>=ec%I|#RG$QN_a3Y7^vINv8cNzJNiJObN?w#{}ZNzo%z3p
zZZ4^BC2lYy`1bY=baB|vNi-b7W8i5XYwE89xp`P|MI};%_ty(uZi2&nUR5<!4n2EM
zC!@L77d1C&Ue&nzYU0rV<8^U+@^Wx`<QEVaY0w<#H$W7y&r+FcpXoEfZ-=B~b?18d
z-V5GV{gDz4u)qa`XFJQZ%l%^K;K7bAJ~kK-HNt(k^<t7#ex39oE3K8=k)Eb4C6?8K
ze@2ax%6s5vM-(0p8V~q~)u!eGos~$DmFbo>F+S<Q&JyX#g=ArFzgbge^sh!~`RN#=
zgR+&6n$L#hGeJiv+mmtEpoVp<(;}jS2seyETwtHw5$Ugk$fPsb2h-1Yi)U=)HFxno
zjf~1K;7K8+?T3}V%esBzyAHoB)G9^rVmYlE8D$zK!?ewUj5G6IALg$(!W`y>3eSWT
z^O8sp?Y)JCJu_P))6Ow%R%^S6>vY3nWS$0%@GslrAB6~G4j?k6#0!#3Y>~<~0p+N}
z!f~}mW(BEb;2}=vf~OA;kj$XM2;$0~8*_D@#Z!SxO3-yx4V|>gB?qc-?<Sg!0&Tqc
zR9f(pV`i;HRLh8k(!nFt$v9kv3S(69z#oq-ac$sgHvt@t_+}m|cN9vNb4paSntQar
z##&dbuC(quS*GbvCEEi7_ZjSJ-|+9%qO~E3G#6+n`pLkYo`B3J*in(wPYP6^z<Zbx
zP>(uOQQhEFWZ-HUFrX+*D36mWJ=;l-J6jM0`?RYVb!#}39NOU~B49!3Qx)6EkMQ&;
z2u?FNPe5DUx?s~)1sGY0S(wE@X4)re!=9}p{!bGH-6MeIA)%%QB8|^!{_vAkD`B^3
zWGt(ksk2!zu-PIyDNm%Hp-28=rWhIODdN?E2Aw)!ZeN^#fBR*niG@2c&zfO<KB}0E
zQo=Zs%VH^b!&t|q0x+%8<@#0u!*g<Ss0jshW(b4qr`Ws*y;}u%M^Om@;x6RN8h<3F
ziSijjXl8FX`@8V5NsGYa-yUyctaRf;@V9;1*n-qnQ`Ava?DVGD&7t_p`G0w8Ycgvz
zz2@|pr+2^usdoI@j98;tp}yRjr}|ZWAqdc;f_RHfEW4lH-!b&=Mw!0BbEUllU;R_P
zBPC^YK6Do*bDFfrN2VgZy?w{yWLPU7IGL0Tb>ydwv2kBdKF@kS&n*EPT4*Vq8=rNq
zH0)CQJzl;(V<H~x4??>CtSzO}6R?N}K<sRr;7PX-{>o{h5gAC7d`!xBzZ{$7Bas~x
z^-h!G<0?epgp&-7;RaJTh7zos=9Q#%_S_SW4kAWV&Vkvc<H;^fp2O1GBdUpoRxt_P
zV473Jv+DaHLr2_(1_7Ho7KjRl_FiV7QOY80qytUp-PJ}%rZ~4#nNn-6K4V_ar3^bE
zXpmRxIOPrinlPH6t|VU}LdEt?IdAHPFqJX%jz!MrMmweyZ!jM?xsfLeGXZ6ie+-GW
z#BHNB*hLyij&hC9LNMnm-K-C4+NBh@NT`L(VA)w`%xaR`Z<Uz4LF9$Ka2)Ci!ZllX
zZusTfP_5?*!AjGBC|NKw*Dn%jqdJM}lT3L;4+R*DsRcV|WRU7&2;7QXex>55=GFci
z+(L@&&|*RbNJ!|3a!c=UV^!BTlCU$VCm>WaB?xH6y~piBG)`eeG|Pt0D9Tgh*5U0H
z6}*{>E11e7&C0#L!FRrcoqZF#30u_D9xfKr9y;@@oNwzq1z@;jMw-O(2Jq6ylwtwI
zH>>&qgF#F;#?mDL#z-KW{oJ4sCW~86JT0d%{^}JVtmPwx@<<Q`8#_`G8VJ1qCbTMx
z1%_<o1^S!BNI(Iw+M*V^0O*M_FFwz~UzN68&e#7B66&v~>+|56*-M6rG&LVea7}Fe
z3k7opHx_P<Cq^=_1ecLx&-S&hL{z>1kC(DNHPUdAZ09bk)SyQ9HcD|IB13_qz_D(t
zf~}H7ib4dSp^<6xUl22kR`WoRF~b)VTbn=+y)ZF-=5)&xQw`rOlerrE8Jx^!iwrLb
zLB&1bL)S49r*iu^KoJ;r!8b!GdY4YmMLa)V^+iu5rDA|Qwa%%?ieeys8Bo7_)1}Jx
zuSsnekR;4Vl;Tsh)_M%7<lu`eVKVd4K@}f^+nsDA|Mf`P=`)PEe9pZca93A}u8nwz
zTDFqA$_H$}0bUPx5{30{X{G(cN*V4;0)Bbvxu);YXiJHWwOe_LO%*30lMFXSQQ78M
zg_kvnyzHC}atMc}?*q3iqA%CSw8x2<aziS`&<bJ#QO~u<Tf2=k$GP%_aG|~UJ`bTX
z_Szcy&YLzK8rw-bE?^A!a2AJ+#yc*Bi`}kt<cAKXR=;IoCc?AdeIfe`EWIT5fex*G
zXcq@xF0L(1G*f8{L0O?nU^QkLh+-fI`7ft)Fea4~G@lG#*fmPp>Y}Sp>VXL59SUJH
znxn^UqSW5*)n(|qMWE+TY|PP%|0KR*`#-18|L=5}<G&B!VT^x{8?2AlzYcRtS`g@d
zQVV#A3+Hv&O^X6qI0?8U5UnaoMO2B6bm{ji?>=CPfP^D<i4ExRD_VChJXktltWW2A
zYuEczjzW#Nc$#AehE51!oG}VA=7fH0%q~de0rqrh6X`)S`k9<>&U8nK!O|1H{oT-=
zX*0guABT`ewMGe~V7VtV*XR99uV>49+vkbGlXwipxFq>_0N@aQvZr)pON{~&skWs4
z*WPZJ>vhk!c6qbL#9mp9xMD!)|M6cD7)A|o!=havchea{o-FwJqZH6Uw@|gQzkNBi
zU*3#?R6W4(DiXxC-p`klB?~|WL<TFtMv5mfC2fjyKM8}M08U(Un+Z2e{(GZMc8)&b
z^Z!g^jf9NEsVKlIj|wQe1@6N1+8&?q-e%^Vdl=#Ns=Iv!t(zcQNSnV6+=eMZJQXYI
zM?oR<Lj``>OT0ON@Z1$yih;lmK%M{EChGlhM2o=}e1;{`;$v*h;{}#!9l5gW9rX<!
zEx~IAg7^KjK8h~m-L$QsZ^+Ef=}p)ZcH5YGGwA$x9shBa`Bc8t$5jd>iNq1`(Y6r*
zf9YkQjxj6&b|>DpOCpigPme6jkC9gGe`sAya4zcLo^2QDrMteSNIUM8*(#!AaEtqI
z#?TnykmJ&wqFD&$@)kmfrkITGeHXT2mUDVralBO~5?Q69x#F^@r_WdS%onNRNt-s+
zGW@sktE=4(>{?wu7bp$}W?GB{P6S%RF7{rj{f9tgmF7I!RN(%$#9U1zc;br6&cPUh
zUlaICgJYoJ85;QacL*#sMN%(c5zTIn)xDsno&r!#cR#~OGgr%Q+)$zE=x^4_%6MBA
z8X-OdekqCnrB#|({y!4y^)8o$1WJ3AlqVSloqgN6b>HBhOqF`|y?=Ot2Yi?rU^oo#
z{wi*j#VPzCYhCq4_q2FpSN&?$bMOIJ-{6xIX<^bt50M?3WwiR-!-4?n7D>3<i|nUZ
zD8x(Se8%bAq%<<wc52`>41^YXYyh=1u(a_t0~(M8fm&~KusGleo@Ase>?P9t9>j!+
z_BI`c>c>(0Y=yf}Gc8oTf>h>_41O0aB1;<k-2)k1S!7rqj6-$8J8KX79f}JrJ&MUJ
zFNpJ8UFwbrB;m0QC)ez3G*aOJPLKLesIjvLFv@Y!`p2Szqv8Is$h@Y`*}bjbwk7~%
zhTYw68)In?LBW||W;Qm!LTh-gR8XYz(==K}7%Z=4dD7q{3On2!uEWxQ(ysH{weq@R
z%d(36CITnt&RJ{@ay;eBH_*gDwE>A)q@(p-$>-vy6r+ExNI;ERXD^yH(e%lF8G~W)
za@owQ)7!@5z&Q@$1FWJuRVW>jK8tlo66QE>d83b)#(6h*QuuIm?Q~ST3+4<?XM`>e
zP$||)GoJouHZGNDxRRrOOKmWhLOp&cw)KbT>j=+`iNGWMiBUy#I^2~TsOt%4<lG6n
z-Ww7~u8}Q51}tq|*V;@IY}h6Rq)~IS#f0mpalfi0-eYkG;r*Zw7+l1<PT(0LdFlPZ
zt%D|cfdGkEx*rQxT4T{5_*2Kq>4ztF%3F5Agc4oR!2MQM!@xPka#%8vGDhVEDs1`Y
zQ|t3Ix93YTJFHY0d5*24@qp!}MbdVIPtp%kpdjs%K{>D`*#Ivd+KuEv)GI|5W+q^D
z3=o4qqkhF~Nty>`l_?dJEp-1x5hkKw4~CLB#EVw<ZyRhcEHt#sdDaAU$IBwiEtB=o
z<>UI4JX4hr!C`+cQ{;6Qbr#H2GgBFK7H;R?-7kV*Q6aZ4*YTRcQ{K8LBA0VAdhu=P
zI)iDsPGwVzJ5-PITVZrjm(~N9QVdkk$Uq*kA3AzEfQ(MjBPu9UEX`ocq!PQ_QpzPt
zZa3@TcLgka=)o5R;8OIrX#yko59S7Zx)oJz-{#UF2;$*AhK&A3L@^PG+UrJ(*%*l0
z3}LG(z&j9F0dJu#sXdM7QqvxPnUD3=9t?R_7QzxRipA{KQqhbXnHn|eVh{b|fFg#I
zdA4kB%ci&60@VBf$#{6MwsA}7yhg3v+!1=fHwCQ`;}T(7XZ}ufOy#etmU<hMOqp{f
z<f;h?84IYjZ605Y>fPIZSjv<NBaj_9%up^J(f#yl$w~*jb)+D7u}Zo$c@VHKdgwmv
z0@@^ffq$KH{AlommD`xRZzb*TpR#=RfIjoXVDIoWQ49YhQ%f*i2S4SXUiZktH2?J_
z-#A8q3^(yCo2p-7sffm4{s7O^@ZPXq70%>U*8Yojd&qa)ZvRtHMoe*DxTSR^?i07$
zqHrvL;rpgF5slf5@5?Vg3#x4!*4G1|Zc5a!%PEq+qct6Dt9~k8+X=l8bkPjqX2U}&
zyuLy#7OAicq&pxq@NzqDB#v7TzK3DGS8s~P;Ay>0?dq7Fn%$ty!m?d)B`3&qQQ4|9
z8_JgaOYS{BlKIv(!vX~LY*nXdW<Fvq+`9<3h~7;^Y+<~IUPDhITEM%x!Su0r7ZWS+
z?UEcQ7&AJS|7+OC85-&_t_WXL*OgjffkWMHlBP+<Gjf&x{pNh`CU=F8G|m^>K_T!T
zPT6e#0q64z!({(&>s(j=yUz7LFw8=$V0s0^1rYH_nFt`-^&$S>l-Ewzt&AsCgTJwj
z_<VumNTRi1lkJ3@$8oGLBINjkGd=k-N74o(erx@F;P|{-zUv>6WXz;q156jFJ{A_n
zu3i(MmY?vzHlc^dT`UZ1$|zgNew;Tb^L}<h-b+7|JA+O`W`Hz>UQe<YdBlylhy2mV
z{qywMdIVi*A=-5ou7iaL=7=PJ0whCBrChfe@LaK#_D*F7b1Ij}V0ULxBeUFGa`&x(
zSCl8?SUOc%kR_%ay;NO!&Jr#G$%nGyGlodDKp}f@=crRN%j>SPzcFt7lvR?ymeIg-
z2A{}e$iOtOp7R_)ieSZ}0nCV&6*LhIFS&5OoIWb&t#I`<RxyBp>s7Z$>X0<BP|mc%
zX%&gh-L*^^lqzZ`Lq9$t<+TKhXHbNMHroHFp$CSK<@tS?Q#qa5*lP;D;vq}{8|3E{
zPc_!^hx&vm<c;!@Du7{$Onv6Ntu~gILjojJn5bkIe14M_OCslEJwFa_)F>n_LRo}T
z+TO*X`5*M3fA0>4_GtCOPcDb|jy-WfL9zI6^2n=1-$f|6m=b;QyUAJtq9O9wA|}3_
zAnFX_BozUmCWxg#GH{i9`I(|4$X$G~w|l*4KT8~(E^s!q6yZo`u<v=@KS?+r?oQ#v
z<pja(S)_+im3~KLcE`{;V+;!|6SpLAe<c#_+*;VPswg^2s$DW~U}IFRa2!mGK+#PL
zUp!R!g5FUUpDOZO8t#F3n%zKoY-RH+pAN`7AX2G*EkWUt>!+=5&IzEQ74?l`rROcl
zy<|ZGZWwAl3M!otr~-L05!Sjz@&J=BpB2^*4Hp|CJH8p6nV1R(uMf;IZAsprQ{`79
zj8$@@QVrYhzZB+GOPLX?&~8Qk6*@Im5TZLm0EOkH)!7e@rRi0pTFET)*sTb7XzNLP
zh&@2zvb?MFhP%Yd1EQ5!v9M9(QyWIEOWWi^4%vD6jUvKvGtBx5R|<zWvH1fCylp`W
z#Gcaz*y)CWkxEXIbRCz*i9b4Rk0CQPjMR2zg5mAc5w<4stIAr=IOj63<CF%j!1AM;
zX8XG6xTdJkD|9eR%RUa`FGLqm=wo1AK?(4=sw~qW7y(JlX6z5+dCZj^hQS3jcTBnE
zd0kWy=i`OQH}L(_Ww&e2+qZ<rz^(Zt>;U2Vj2+#T893A%WRDKWB$E>o;inzTY;Gkq
zP3ou%J-K>2@i}o}7JNR>Dt(m5G!2ioK&4P#VCL@w?wq1xWzhWMl!{U4O>KriByc*r
zS);2*`9s-!1uLZiXF|1m2Ont(Kun;U?<B;x1U!ZyFDx>o3?sTKx-pmEnfjZOcC(}K
zJzE2hH<_Q=^>}((tc>~iFY5qtn_>LjQABCAa$xqCMNTIwZ(uxcdyS11A2t&#{&OWE
ze$V~YZ|1Mn9#E(c6bk>;H$D=VZnMdqSn1qu8SaAJ)nV|X9Ku?`V^Y6GkK$vB&lqWB
zxjT+1Mk_8!e37-YHS&mTVFzUWAHFYJI>GelouARrIilDz6vbwT@%}h04}ccRX9~Nw
z%{B*{5O}}^1&4xD-xA<ZWA$W>0;%tUf#34(l5yqeo3!BLsIs_1E3R^>1<p$YUq9Up
zMr-&FTF8gw*|vwg6x|<)RWP=FDV$y^%cO<ST^F%R^kT{;^qXG#IdzfqEjtv+x0iBx
zHm{hNmMor+B|UdcKY<eQda*7I+23>(W1sbXD2rO`muvgZfJ$#07}HtVFzS`vLY;EL
zaV*>D(NW`!OBq8vV|a$OYPU^l1^)4*6QE&yaa!j5Q>oh;2}H#{ya;GWn$$ODu@Ym8
z95}~&0J0o@*D=5hUK3+_m7xr4m<Z8^@fYcBT3?uZPSdu-bRZXQ&M5H|HXJ>W3zj#2
zPIt*;!%%#c5ROJ06Bq|`M{1!dC%OMTbML&n9Gi>zO(;OJ4gBTD2FHAvfTI_#gk<lV
zIDt|@a<6LF_Kpvkl**U{={cIhn^-c}pMvN6^M`lqm>Zv%q`o8amnBnU7|2UP4J3Nd
zc0vU=jw?T9;m$7wJ5|pMP!vdnY5Non&SB)K5DbfW^jf2a2CuMxoyAZKz6kQuz(~>j
zgPk26r^4yfeJ1lGXOlbl4JmVlUl^Y0^&i5Do#{Vyj~Q70+pwBa|JDDoBlMi8L3AsF
z8HT{~lszxf%@;sh&&j}E=5GaBFSN8Y9BMi$7OwvE2*FV*L?oInpmnAO;qe}4?%5w>
z&*|gN?mn$etgQ`Gi#8GvYYZgnw?GLa4JinkqkuP>+!9JdBLCj7B4hP*dOo<arr)Q9
z+?XXYL+PFJHBRi%?M!dWeP?}n-cccaD=?BO7v=rSN-Fl8!ffsU7NrP9giTWzCoxb!
zn{26T+M<|EG~Q#Sr@~1VDK?>my+D;*L~mwq60t#7uCQ1_8g(&wNj=w0ZmU+72^=U^
z<YJI0l6Wl%ka@e;rVszy{CiYEdq3T@oz{xx$s<Xe9F#U9+UpJF2vIuE#v|Dilk{k|
zmhTwDUdrpLRw(MB86Q;4XQ)k~&{?i1&uO99lEc7wf0OUy50UL3xwg2kM&;lVFy90w
zx{58(EVNg-p^d^q;Z$naPNG02Io?c9XqAIiM^tr*1P*i^==&>m#f=v1YpopY<W~kh
zHrVw;H~^M+6^;KCJ7^eaHR%AmjVJAlXE)tR?7$#~f0%nqj#04pw-rG`lsy>&cN(w~
z9%>MU*$Nf*+Y<FXZ5=yrZ?k4YQyt}zId~mjEuV!W&Bl=1NH=OhAIg71z|`>b2+vQm
zR>`e&G5`Em*zh}MTj@L=y6`^{2sSezkr<dAW21(ib^4nW9Nhfp27UxmZ%by*b~3X8
z+CY<@+YJdgh!IhoKLH|es9-k=IerM3!pY{AZT2B7o_QMV;}+{?5d}J7NT{$g^DCr*
zhS<8S(&9nYMrElW+C>F#?zX*qO>U|>4GB`a!h7wh|LXE9Ca|DFyUv_E!wv9X{fMww
z14@oI!*{P9xSNblx6fhC?>BVP2-i!H<WEjQ%GhRD&$Hmz#-UCREZ#s8p<bufe}4Hg
zTF@ksCFlac5h^Wc=${FeK&@tMe7V*5%uouF0KuD?DWgy!;DR9*vKv&IPECq$A?-4v
z&!C@ZA!zZs^o~;i2DhwFDZ~x@FPTHLWP46&Ix4q>vkSK%5}G>&fvN8aA#ikges2z*
zSv4m5h;Li{ZXudh^+j_WiQ?TzXxK*RhbP8DIta(`%yB8(-d)gmS!B&A5!H{JpAIW{
zbSJh4;JuC=n>drn((x2tlo1LbR`D#C?uhjjxgvT%p~uo?dtq=x)*B+{#!<*|9eO8m
zR1{Psv9Jf@7G#?^RrkuIyc$rR*UJaUQ7K4Z^E&cLb8mCLhI?D>0i*6HEV?QE#J*tP
z13&}n4M+W1e@B~eWtRU1851vw?v@)b&NQuyavVtv&jG0HIe<q<X_QDb8R?wj-=m6o
zvpNWFw{xUn*^qn;2r(pj+imc3f;bHJTwz=7F#FunB^X8N2%ZA|zI`x<zD}W64n1#v
zQ8Q8#KRgU!0H)H|SGvts6MJ!H;Jy2VWD25yjT@7xT@18(Kr<yy`r>GAhXaIRIz+1J
zH6g14Oi|Z_*qd$Xy|+_oAmuHADe$hyn(g`EdV<YPdtW{+{Sk1>q(3NI%0(er(sPgk
zf!Y=1``q<JY-{Tt8gE#Jz~E>R31ze>%2bnigBsL<ti}2hJi(M)RnwQ&`HP*aMq|0%
zWrdfn9yLxzO&Q-I$QU!Bq4Sj>6g<6s56bqdbR`<oy-u|+mf8hwcV+YHW$0K7nbn+L
z=oMV*wWBwD+sEonNZ>*p-Dv*5gs@p=2Os8I?>W48G#K|v3lwYdWACgK%ZRPfI<2`W
zBI}WM2i_3bP{xE|##WagXd;joJ~Wt)Px!b7!us9~^cdg2Zr*l=3oUmLZI)3j-FC2#
z6tHe!dfqxTGEz1q?p`FJ`6$ID>BDnCFB3j4SScb_ol38O+ks7MyF|`nhe!nCH4pc@
zC>bz7KTGH!g&F10_$i)X{P~6hPlO$)OTTwVT)r69yt^cm=+MoLPGkU0b)42G$8bDb
ztERQ^bZ}4Yy?#*k6FvYsFFz(4JEs{85qPI5Bmk_ho86V0q;(1>#*HpJhS##2=M30r
z)zLK`-Q2PmMucyA%wm*9Mq|mXp2Yf8`@Oc&MFfhgGj`P0NjIid;wL;{Iy@On@MG=@
z(c-wp{$Ib2J#6tB*~Gtd21bQl2)utkh~0Y(PIPL60N?B*YH>|f2C0il7|9%zdE&G&
zC*|t4t;Wb4PO%<feEBdp++yw}wcff)TTO3~*(;~H(?=l)6vzQ`H%55|l@VV4(PMRK
zq}3ZEkj0U%!@hrPAs*S(Hw$KJ5P7YL23b%#DmlCWQZel?%LGIWNJ-PZLwKhZGR^<w
zghq$ID5yAdOR)!M7&2c^6bW=T4HeaqjAQ^8(#hO?^G`mpe~-1FqCTQS-;4XsBa+E#
zKR@)yw0v|?*r;CRMts3roZ!PW)erxtbp209L3Z~4I$gIkw0?2ZC_Xc_2=7dobsD<D
zlG<~jE#ho2tFz5MQh@^X>B5yX9tk+)wC_7^?un#pvYG5t0sVd=3ZY^r(>G!4Z4mai
z6LS}5a}TYE&?E@~5&E8Z42ZYl1UPs4oba-O=~1$!y*$2Bb(iuAq6jB2?idAZy*-|6
zJU)CG@-e?!1TyX2+d*d(u|4W9HlFXV*Vr@gsWXU9HTZ3=I~sCk(QuR$hi+-JbK+xT
zYmeMi58axj&F$>1RV%0R%U!A|HCy!@_?+?!6BZ(gvU5&q=u#8|0U#m>)1(2&0p@2(
zj*OfnocbZYUX`LnU8v@kKyGT31{l(_av6*WdmL%;019rVsvQ-Lxz*st26L26>j^=A
z^*GSqehv=g?>hGH6M<(y?$XWsQS)^QE75rI_T_7q5>HJj%6bdaf4G#`EDF)acXW$Y
zN-U*ZQkH#qR%f*vOIm=*%b$6TcHRD@n?+a5Z*Kb$JA|NMS=oTykD?sLVixG@6aVm4
zA{j6BfD1u@Df$PYyJn$2XRJpiZYef5C?N)0^vK$s!zE)^UN|owvOSlyOJ`oWnH#KV
zIBC>i$(2NHVo3Z04}eZ9QV_e<goZ28CxMFn8(&JgyI9f3*6<Lsg&VHbG@f~lbS{)(
zJCNOBZ<c8mXwT+kCTV46y%Ucg_^A*8Qor9xz}Rqb`!qhhXVTu|+mPN~^$e**J702#
z**JZ_8FOq|4#l=XXSrE%Y-)iFaVb~48Bf3Ll(|{iILxHhiNqs|fH-jCP)xwvx)3P}
zfh?z_+WCqQfYi?;)50ay-(ZuyC^^~vR6)Ztk=?z{9_)Z!>TRTTpL-s27f)Sw8`uS3
z5t)bE>jugy%1x$<Pn0)%>*rsp<8suBY8-Yd{K|kRAY4WnLngW%z$4vOtdhJP)koEW
z)o?30h2@Un#22p_+koJuYqG*?U(Nc|hTfcQtxqRS^HrhRoXt}?zec{aNY|_>K#<?u
zI{I~r2AGI6JY-S=4a4f72)A$7UfyA0BNJ#W-Jda7+47hprvLg;&6$Ju<+La{P4K#l
zB*nl9X$pN{LdHHz4j7Jb;2>YV`*}L^0ktLq@Kv{LDut0i(_-MbZ|(15o!%Rye%+0!
z4>eSTu4J0|`Mref?;s9;+!ul)3x9W<^Eev@3TRx%D(SuF$S7(<nHL2H0&EUoW;wx>
zG_<G|7>lu>g9=PMph?8KMl!t+J^?Waw|DpIfGHvlB;Mrk#s?uy;rHI3%^VJw5{N%3
zU~O@K;B+ZdF4Z*=WZ2BEq|#mKxKYkrX$fw){jRxljr298vheIL4IaT0>{SSP2^<9X
z5DhhS2z-=D#|frs@!5A-us!+F%@HnuMOYncIxY5f%0T$A7V<9j*J8XEry6YCP13lD
z7!h{npUx^nQ!*%4W3N4Tj}0t?$N*P~pE+pM5c6u<UmP9)^*|pb{z7WDPQcMzna+n>
zkHuieSt2s#JVw^+$bJz#AeURU!<G!%v5QslC?a>W*yv?|tUoz^7TrXx;(~c8gIpin
zI0TE)si=h%S5a?ByCht8i(x^@-B$I)K?_99PI#!6tUYs%&WOgHHZp!9fQbOK3L7h=
z*oP2p$RfM8#LF>YfkBb;j7*8HS<`HJ-%#gD8(*vAf;5J%EaR!9je|2ZL!^5&E>hDt
zxgHu6A7h!Ip&lA^HXg)#m4;!Sl$SHYg<iCs)Sh2t)=D3&#f=+WR`Z7Zmc7nRs^%3<
zWh{~}v{T$RIPV-26Kmv;B$DKFX;$j&`IK?>UY7%Fm1yfyvxZE^P-Zh39ch?W+p4@I
z{wUU+tMofMtpO4N@o@!uH8|rq%0paxNbk207mMqLPXnf9$ZLDOOT-6!4PqUsZl*!E
zaZ8bNK_r?BKUfRfYNh6>vO1+AFAiAVoU)1VN6hceS)xODwLBmIoUQm%&x7~@>}UPC
z``fi(A`pq0^bXzVzOe!=_Oh{ist8|Raf4T4HvfyUcL>a^?V7fOE4FRhwv&!++qUhb
zV_O~Dwr$&X($U}F;2*vB;2H1vUUjZob<}kL!|RF%a{u>8;viF5Qb2!CWC&8O7L*r@
zT9SEXlDT)CJ5vuRN#nt~Wi8R#sE4akc1s^`dL1SdzrshQuPBg2)HnM5^M3ym0CfPF
zTR95g9B#5WS@n&ct+Z;@VwF6qZw=&gIFtF0)eF(Oa(qHre$!fcT%lQ$Sq9vO9L2e_
zToFa^*HqH~)bvRxKuCa|z*8j7&9BpEPxlE3?s%W4;SDLz{HvKDkvIa(DEZQXmut8m
zx74+4G=EVWrx5I^V9M3c^A5OhF)#A6FIIL!LV>j%HYCKg%Gu>WPNSG#*m=6zo{-A|
z-c=29h-b8(&^dkPhklb>+aDzGW@g7}HODoM&hVy@ImguU_Fqmx7)qt3unj&b>#JoG
zX|VDkx*;glF=2Y^;>t(V1X49PB(oZJe_69*e;$7dmgg6}!j*igdh_iYwtC`})Mn|D
zP<Rf#3h4E%hkRw5*u0-W_zctZhq>N!uAhtIjB4Cp?JqahPndA)0N($Y*8gfC{!b2<
znfZV3B>tDfb>8Ge{s$U$aE4Mfod@bSm%S!hmo4R0P0DO;E|d9x=f+h9l3{@BvxZ(@
zoHKB-i3YxL+!F&B3wG>SgfGCO8~&qTk5_L;$NQM(&9P8!NM!4V7xDc8C6orLcV+DB
zx;!X-&@i)H`p@R}b=k5J*2q+^KHb8Yzxj7H|L5QR#S6cC)91<QQAruGVOj&?(44Ab
zUJhRMn3taS^hc=21xJQK*0Be`>c4h~V-aOLht&o<2wy`I^dn73;fztkMMB_yCKMrX
znMzPe&Fd5-#3Ue0@gDvb(IvR=U!QYsrqu}zco)KHvayZAE3P7yt4%9s;AfDzZ>>Ws
z6LJh~`sgBLonqJ@t4$nqU`+N>JxPU%y<@lNypO5^Dlf_S2x}fyL(xw7Lxyn@RxNeJ
z<JAjNCkWj^&E>sYe|E88teb_n4IyG~XvMV^E_79iBtWy=$BYQirD&~$lPaw6b40lg
z*We&jt%)@Wrx|cZEBreWlN}j;JrW)<T?iCTk!P-F<!FMz4Di0D@&@b2k>78cn3nax
zSJ20$*%D)*-1U{B<nEuSiE7@)px)SL!9yVXi!ZnBT9>DA5y`za*s5+HiuRKU{JDu>
z(ZxrN`(;Frx67dlt5uK@vVvaSk`2QE>WX5xgjbI&^*eN5M}K>~1A+~~0SbJr#9+#a
z&Jz<7pNJ8s{1}XoTs*ms5R8)q2`osA(p#5pm>^$^vTEdH)zZeSV0S50*vlGr7&fY#
z)|$~8DfFfxk4K(Ta4=-{H9OZlgVD@K96uTSo#BF^5LQs?bikzB6KReVfs;phZ_Y;L
zc|y;Cg$RNqv(MyCDpZ6N4ho4-^;e7bCDJrJRks?QAZq=ZmBqULV3bH5+~Il;k7}iy
z6~xl2P|)!O^5>K3^&oA8=F1agnjGV!DkuA7FsZDAV)^vTtmo+#%dM+~-0<fFebrxe
zA#_XZx?~j>WKqmK(mvWig?1Y2&j1Gy*?qzn0_kpvMA3m#td$huh1G(2=y8MA99q^)
z$Ah04qnE;S!4sJNNi*sm&tp*_$$^=bKJl%iIm!_g6LtdEdc<{9b|`h?$RO4Njdr!a
zishLPsgr0oF=jI6P3Om?t1}1>oU)ooeW#ETt)-)k6*Z1g4m0e^yU@!UDjsIE#PGqC
zCrQPCH2KLjQ3<z&1lb$li}1b@Zig97XpFsj-VC!lN-&rN@CeFa9e9wIhB_*ko~C48
z)z|#rBAsF`N`&o63z?(7npYAhB*=<#1rrdaM(mUFor4Qz#<WYinyf+~I&{&ARLJOk
zxrv|3ZRcS&;<1Mm9Ol>AA<5fKf`2$BJU1-gYq(oJcDD0nZ<B;g5f-R_5tze6-rz_7
zmbtcQNCpNsnhADAO}CxssV4R1rc`L6{|L@7OJ@z}QafvB3^6hZCQ(LOW0Tdr2rJH@
z@sX~~6xxiMiizc^Fu+J3B95UESZ142rX<o)Q`s)4GBtgxM@CjW`eAO8cydS=>@HoA
zeT1mr+P*YYW@lCIfE;7#Hy&@8s*p9V30pZ7TrT|uL}t4M0J(N+Fi}b^L$Ph79Mhe^
zk{!iHUcprau^iYk@sBm>a>87KLt<Sbs21Nzta>5jFp1Ax!#7bC{;Th>>zkg*PuCa*
zVnp4XgqFCo{d4*9@^`CjkQMeI+vjg@W&Gq7_SXvGi!e4M6WphEHN*snb;Xniy8MB4
zvT(#v4PBqpWiu;M)E2snTG?yG@8IW3UMD1?c%-~7PWWkrYFRymA<c0wj<Fml=b#ef
z90>gp?%L55uqeXUKz-_`J(R0XJnG_1?Jk4jxjs_*Cwo}T-yLr3-xTD@xKvo0&W+`8
z{^Pd7SCX*DNA8UC^v->z2rGhti$hq&t|cJ~((4Ny*gk5zTGq`Kw?uBQY01Y-OXkTI
z#epQU)lFt+jtlH$2Z@>K(X6Ztm<Ht$bjUKcYr*=FS9GeP3R_^Ms75Y6lR;7Q_;W0V
zgDUf<NGGg#hWs#XL`@!3CAmGeP58#dwy*)Xd*3jmK`ac%4fycZV)IL4=M62g%I6@g
z#4_s++if;yYq~3Kh!*9(tjpiZRi0*v+!x-J_hG*}#<)Pjm;*RaAStpUa+b9gaxbte
zZ<h#1?0@qjp5+vZrO+wt4}x=NC@q>bP=<kC)nwgtxk@YDI7UDwOxlvmUGfq0UcHnh
z`xhifDO-GnIJ?e)JuzO&ZguZHD%$Yq7`ew57?whYjK>ua$<f5x8fPxmf2!;r6;!JP
zFu)7t3d29m_;=>KZk2WB`8zC!FE6iFDziX90nZr+;G-T6?@!G2ZwClGA{%ZV$a!v7
z{(SyPd`cex3D{<xykkIbvk-50j^}H<WD>^`Das(;L#QAG#RLO=W(gb<^!!Y7+%gnz
zZ=BkHzhGEv!}S-&$Zf3}y-l3XOZb|-MKr?TMS4P!^1vH<)NymmXUdSxZ9n-PKDT>X
z6xoRmctJN{`;8GhKy<@bL<q32`p9I7xv?g<24hk}!nU80caJ4F*)jjVlS?gL_%w5v
z{<$j))6-Q$G5uNctuKrEOxeU|1BY@nx?`&8`xgKe#z>KKnBD(va1-(|4F}9|^<5tj
zj05w9>*rXu41Z{=X*~a_o9kvrD!qY=gzjHiK2}{F=Rc(MZ4}dSXOT6_!#@^}3;hrA
zSnW`NOW`d81hRV(BA-&4A3{h-B>ANzKbr(A2xqRnp4IzVafN#1X(9c7-jJwLAYr&K
zzpEE(H}L}d7GtGa)K4|V{{uZ11gSJ6GpVkEN(}mOilI8R;1rkCp-@mfH^5BYQ1BMA
zV{%>5Y4d<>6~#64`6^M|T6U_EzR$#RQh_Yen6h||GfF+gXz069b=*-^6^nh4WMxun
zk`1AFveQ(GDN$q24$A7xqJr%GdG;xpz=?N`&GqZ6WmIufa5^R)gK<zCrsoyzkJymJ
zqm$o8mRHRf{>{OMd!O$6DfH4}y=-f7)JeWhsIqF~s1G4lI4vlej=n)2dh6QKOc45)
zu<L_=zgk53e@Q6z|Bz5j%>Q#<t3*RH{-_1XFHfK7jSSl$vlRX!;4r+o-kL`;#-wHU
ziqJs%%5H<(%XLa-^v+*CL3QakH`1^-B3#3yKCZT6tUGD6+y16!W7qeaD6T8}SSsD)
zqX>g6Hv(nEc>h}Z7D8a}^LzKB_uuEecOy+qFqsHQ_WH`6@0V?X*IR;H_kumYi5qv~
z{qNB)U+j<Dseg511h>7r{;WLPj@;hPjOL#3k6SRb2tp%4oUbozSKq!ol#Q-O2<BV{
zw+rVZZX&$|NBLIEuzcSY$7npV($S{Rw6%OS{Bvq0Hw-CV70BToHLB%?MHgNKfP$ZZ
z#qb39^CJf}JX?<RXb<2+p)P|Am{cKa&9LD*b~)eW<TcCq(rKUdxwk?C`R#=}3Hml8
zikE=FU*nVYoe0dmd_=kee_&{#1>2D1zufuguSO3(1!*=i;g@{{DuzL>q%!BH<FmtK
z!3#F5r_>`JBO)1xXjE*y&lQUnoC+NjE{d=+_@e5u9h@xA%eZ4GuBbb&{zM50V3lu-
zoAYHy9#mPr3Zb2f`(vYRu1Vl~2=zG;30o8H(8p+?idi^Ot&L{(eHcQlOngZ^qu@q)
zGuRAj-A^3FfRt^*r1p8%M`(zfwMyWbeUAbshshy|>VZE=nVd(m%VHs@7vnYDy(9qZ
z&`&p_?GG{J`Us(TERa>8OhIH}n~>vdubQP7y%}OZyIArRs^i(sA-Qf&e*{6{Q>z9u
zLG;jPi#apFkirQOGI4rk61&Sc2zN4H*hab8fQL$GF+hJh0a{7FP;SfQRz&X$qY`NG
zSAJb$nd$MY9(-$ZM-!vNEWa+M(p3+z<V2{W<DSs6ZZKU1g%(9WuPDF{nT#9R0-;z2
zSpU@c=Q4_SmtRGY?Y+<#0g865ygd#2L<BkO$+MTf81rHsSa#>WH8#LLQ^gAC;crcx
zhu6)j5a5eww*)^dS@)na9X~X{Y>Uy4v7|-C{p^R;o~;wK2?*H=&0>uNn?ew2iaBFf
zj?-Bx92`|fG?Nj3$>nC2S)+#E$8dR`Ago5yo0uvBzNy;GHMNQ+;c}6{*u^%JH!O73
zvT{}@wXF?G&@f^9L;Vgx`s@;!=pm7$*$6xL9OWApkw3$4Y&k+LbyY7+5Zv|_ar)pv
zen9`~!9OHDmnb2?gQLk2Y*tw1Q7AKPB>Dy9r7dzTJ_x_yyrIjO5V7vHA{C53YSJZW
z8y#x&njV)MI3~8$eLD@7f<-gG{xEcJA~^&Cgp}bT`hf(5SkrwdLt1S+lU_G!NA14}
zmp{I?%Q~f~6Tr40v+2RqX)kzK2*^NI%nf4pCd*f;ZarRW1tv(v2xbI-XNUeCVw8-b
zowqtkSg0mMfeFOHiFQhj0*FEPyW#;jD3kk2+mm{wu=Q@-p#a^ziP>CR2m+b`GF4{y
zw&|<@9~7QxGvJ$NnCs^shNkXm_$rd=Ze7F%NWleWuF(N?d$dqmna1Plh+xR4F=X#R
zL@tf#wavG@1jvhef;w{)fkmeA+#EvdPmpkl2puvv{hm)7nh1!$#(;;==XzaOGk{tA
zYa!rhRls<C64xAwfp8c~6It=*D}{TK;u;YXDTl}NyXMIZxQoyL-DDlyqtOPgi!qD^
z?S)Vk#^>U-p>)Ua1X?`t&r)=xBq|~;UD(fQ*q@rExDIT$Ofy}x=Fn|KTvlmglNU~Q
zFfm6#B|j7?S@<WnEqGD*Kngwk?IvBi${&@q2M_H46lFumqE${2aPD$%nMJ3Gheu6X
zRQxy`%++aAT)Pmdh-Ddt<SU|w{s+nl+K$W>?6z~A4LzK}kp$UtnNhO0QshbDDU<2X
z?4-lBwPdm(vkcetdt#?d!*Y6s8AO<qU!`RA42KF6LVCYBgzNVoie#xh{%jnaiXE}0
z^lM_Ambj>S;+w|qfr&A>Ihz5W@lC?|KtzG@KJd`o1AYmh#26cbn#DV7M3G%y6F9`N
zCIRXH2FEik_`PMKLyw|MU(^NZbtyd{>Tfq%A?l=FgM3c@15LzB#&jG~iE5>*yNx&n
z4zX5kgl7OYagLZjGB8op>D0T@cVu3>L<ge?VL9W`HH5rDA4i3T{Jm!UP*0P>CXrRs
z|MQNCdX-$z*PBNhq8&Ki<BqDa4N38|#!*gKPOQgOkPl7awz0N-S4}9D-$ZmO?3xj^
zu|diylYNZBpZQM4pXq3bY6pedQe?XH#QLSFGY<>*QyR)DmF7gq?45OdF<#mTR!eN=
zh`3V+Oa1pan{DKrhFN}<c~&yt3|raiQBE;F#dI1Lov@p*)ZeO=ZI=)pYnJM?c<=WJ
zfB&=lAU6x|A47-ZTo#*2+LYjHNx(0+6gQgFSByjO8gwRFcX$C67SD0!OV`s0RZXj3
zyBs-WqUP^}5NR_E^;>3V0f-E}BoE+w+m5+A52-2dFh7#vLaUxg>%1=9>rWC9>QRJX
zv{HrDJd?2=unM)7=BxNBME&#@4`(gmb-*WY`dr#OQ;v2@VAO^2O6p%(LAH{Svkx6B
z)b#Wl6lEkn0{IhVyFr;%a<*mR&eyq76%%-joQquNW{>HNX)iDdpuDD@ZR4&XYrJS6
z6DmM=WG{~96pr;NZRw=&-Bf;1A)q`GCXf(r!RVXox7XZ~6<U(jxSK4iQWK(?)%b{y
z=L&KN<m8K*EF%=kW41R}iLmQB+rfs6IO1`?g*qx_(Xx4B(dmwq6)X^<<Unw4_gTAR
zL~*nOy;x)@QE`*KakyL5-a`-P__mxV82IUK7%Jw#fwnf@W9I<s>c(<aZKldt(g^Wl
zWNEd0&W7T2OQAoYwrr0N|5u@KG5!Z^jhTt%e{Pdm*4&QYY(w(j?G=u9sxy;JSb|tU
zw+NmtfGF;>{cOQNZIbnNO(9b!ST7d4wZHw%l1w(*>Aena@nIw-;>4N#J`FL-yCK2*
zcmG1ZvU2qnz)%>3W|=5nsY|*jS()ruQnbw85zRu5s!i)Uak!cFk7ZheJ}LV6HfWVl
z{uX~@?)vzB!cd729bZYH->d92A#FM03a#y_76ij_9<_LrV4y>KH$k%hipFK5CRzz9
z<7e)r+^<I+ZM#V?iqBzt+Vj{}m9X>l+A_TAx@_|;J_zv$7Fl|MjotEgHQ{SZ-8md-
zoyp<Cx^?9M^mbqw@>>ypRH}q04zdusA3AsXc$>@f2fC2`8ccRdE>o(s(Wj^DQr14r
z?${D%4sw+?ye!YD&UIF@CS%;A37{sGke_LPUHWdK-B?4rF=0LXZN%Zbz9hN9>S{0W
zx^1dR!LFEL6sbblZ)*UOnd-R@x1TGge+Lp}ExUc4`Ld^SB_3DG2C=!Bs+OJ!4nMu3
z)<(HD6JJ#IrpWQ-ljq2WIbm%@Zv|~8rJ(ftY;OG`$u*r4)d2`>)%TcI&w>>_U+bJy
zk86J$cdG-Z0-s_VM{Da47SG)Qope1Xs|(b2Pcq$#8=}5pXOZH9BSDIkGpQYs1(z!@
zkQMFIe1>aXBQjFeiChc`_xx1!+?+NICN&sFUQ>N~$2zvLP{A<Q3=0BErC&*~DbHQh
zUOUTGGvzBS6j93O+8}8N;q~}rr#s5SxX+f`=LgyVL>Ede|7PnGxEoV22C#9m=CU&z
zc<_b2(&hj0O{XNdKROXS>dW<*#}$!?1*BiNXZA5U_YAQSm;}pvLpuO1cFLmAahsT3
z$n|nu@EPt{W&mtJoO4AEJtCy**DCN~kgx!?KYYI0n@ux^;x2->1gcL_<tj;O*(Wn9
zj}4XxmPf8fNPhiq^Fzvxqb%gnnJZPA)5;Kh{y6UFF|oi)5zj=nuLf)vKU%SsUe^w;
zVI^l4Xvho$&qjw_w(-wslj!^~`_sZ{texq*?Lw;3i+ImBQ7w_8`hti$8=X|#pde$0
z8tQ}Uu#ZjaL1`V4_4cmt@{0OKCDg|RX=2MI_qa|<QqK}mp=<tv9u#<_*`<XlB_lPM
z+=Q?}j_{C_+E6~X{mjK|n#-6QHxG#<p0z(Z6Ai?$0x-#)k{ff?+y|V6H&oo&%JS|R
zAda=Z8A8sQO~f3m58+*zcVc<*hKl%0QirzvRu&zjYJ5^$kmDDgodHp}B!mf2nLR~^
zQj{ne*R%r`k#Qga=D4Yg0|ySL^y87hmInRog0N9>V0Hk?2m^pTRfo1^683DjTCS2w
zo_~(lXF*Q)G!(g@A##mtr)${J<+!q5)|6MA7%rVrF4++}`<}_-!JdS_AO>Qhi&Tvv
zR~J2+cgFt%;D7ChR@$kl`}bnck`uN48~=&N1}D?|6Z57OhJN@^8NvsMiq=Kj%km5<
ziO_sA`EdsU<ZtuH7}J!k1*p@{#XC+L@&HbLH1?!6K01=wNmgCPgkvK0@1&;BF2KCf
zu}mcH?>xa4)wZ7I-??V0m)p$8dnWdNJjUO&lW*?58Jq2?iKtf`so()3oBZXu@1MI_
z0<@o=E4vQ4C8@MDL4`*@2jw;TJ&gu`RjlAf4_mLh$bE$!&Lm>mco38Rc{N^lN>h}i
zV(j}JsJw<ys01uWU4F4sSjO@bfJ#lbmKS}7l`$-zdKZ5yZ>U1^;dY1~K`4{xXl{?^
zivrt=uS)edP3_L>yJbJO<!GANE+VZzU3JNxzs&e08aW7<-%{8EtYgb{h(u5Kc?<Kh
zXIP51yb$(I=^ek|D!+OTvV&=A^x$c~zmT5uio+<N0n<uj(!{|OX1gg27$u2|v{3-u
z02t^MxAaq#Ke(T4p(jsG9rhTWyg?uo>>}MIhJcBpG=F$_mK5lO2bXFRF6DQd<rVzu
z-syudy@(=By|$Gol?Ku>QzGCF5DAM8L}y@OSUIl;&_cGir(+b}X*1BWt3k#Ug*v0Y
zs1^D0d(Q{L*)th^L1_7lWEmLpg6@p-bTU-6W3N6Mw6Ur2s6}e0s-g`ie}Z4SyF-q_
zaQv;)RXSg{tFi`*q(2olLEcd_J@U&TX!Eck)UI-reQcr$-}-<G3<^Bpgwin=N#e>f
zX@2t2pj&t+0X0^FU7sJI<4lUbw8HIf25sWT98wfFa`pA2etjq6AO^SKIC>)XLr;zL
zQU^EA9!bbR9@39zsbDbu-b5$QtXR6BH<fHiI-9pK1?%_*rwh*@gWX!Q-U4F%3l7cr
zG&Ws#ZW4|4_J7Vd!Y3KW1-Dmn?ItM!h?0;VENE;F_vBI6Tf8;lgBxLbHy#=OG4`vt
zS0#1n<}z_e={0)PHG4tVJ59=Vv8+n*;^TEhX_eA&Tm^m6%52BJ677QIi#_bg93(rr
zkzWI=6UwQpX&Q0FXW38%Wsj>E{~8D6lL!Y#%}pHUp?q8imNZo{mi|7_&BQe=rPQ<s
z9~ICGWX$md*2kXPoO%aZeljiu<>q)1ec1IGVZeg7$5+y7OcCjm7g&-cJ++s5C$K=7
zPA5-Zz~kk7esPgW4(rIfP5|>xQn39q=^&ssv*hUe5~@Ft3I7$aZ$6-Geexu`$__zr
zhCh3_>#u>!D?J=EcBPZkCM}g`2l^%+;fq<QF{Z@&7xO|-Tu~`dmo7(aL=&p?8EqYf
z)&*aUnWp!R)EG0xXZ&I+8EUE;JafMQks~e%@fxzJ^Aw1+U5Z=PKtL}GP1G`!`Ygfv
z<1A|{iMgo_U8eHZw(5CE$HlgDQXO^|612CQVv-O0UowUD|1MIPIR5v(lM?mKxPN^o
z|2TAr4@z8h?Zu=QLFizKMoBFe(X98{$f!CQsVQd%l1p~crC%@bTqN2HdQqha?BzsU
zyxEgjiQb{3-Dw!Jt^e3Gg>#b{=n(a-+unv(<CJE+i92qm-OpeMYPYQddOe7|Su!xf
z{l&0Cky+2%ud{l#wR*Mefk%D=YB*TVKhth0qTehZt=)g$FVe1YD$j!!ccNp#4uQwo
zDR99Pn2}6(TEMSMb!`U-pIsa6J5waTbczV=FO5}CCeHAnC;n|7R8%vSUMVecY%U9!
zY4+1NDrgM(p<hLkvKnl?qBX$35m?u8e`)1b>eUwDnEF{-#P8sP8X#+on_zn+N~eN2
zK)kIK*kpmg8Qpibd>GTOS*jY|wy=gfE>J%Mr#19Rx(ND;2st|i*l??|ga78qYM_NG
zeoD1f_dtN__InSltZBi>qn9f~z`_`a?t@m$^z`&KJT{~?^t%{Nz`1T}?R|Te7E)?+
zASc|?_}xzc$kk>|`p2k4FMm>ktmJGm1gp;;!Z$E>jH%bU!7!v0oc0Wi5VB<lf8n8T
z=dvhL%Sw0t!Ld2Jd^ESAK<ODi!Cry7`j&$K>;5EthBH2}v>6c)%DCKr{sfVeC2=1P
z43miE2Z9(hs{wmcA9(^aMG6FTB9M_?VhQIq0j^ysQ%ynP!z>cC8Ck3AWkVbhgYpSE
z|0$+mV&mRwcLGu{MQRdl@{dcU?&29q>7}fO=9b6beNA|n%TE>%yV8l@x!G#v`}M6@
zN%tgQb@yr!Q=_&#l)be+{axY`mI0U^PXpKb317C;zt4~SG3Vbd(ND;J=Rk%+(F}4i
zvt)`eDjX*Kg-ZvLRdWwh6^at43kA6D=#%X@3$U~dE#m$CYFjN7UnKrk<+2+;p&G2b
z8Mf5S(SZXY1rSYt>kzbRQhqswic6tT6>R_nQ5_&jFCxl2spJHGRc|N&{QOmFk|XAN
z`79(FuqC#_)c6^g7#k>=2o+C53wA;6UvD`Po@UU|=)X>b3+N<YYixU%-RDO1(!pdB
z$6>b5<Wy5=Ff)GYE$!>wy%k1YlzzljgT;w5A-Ib`^Ld;Dau&7hhaJvRS?4o#G&!5d
zM;n(qm|&JwwP`PN+3eFn!lf#_C90HHFFF9AiuE|rS6xq<ofqI$CAwgZNSSgnNH7so
zkj;AA7Fs%5_R>t?#F9%lgL7F(HJs(@Z<fHOe%>yQF5kHzsD9fwtSS3i3012d<cjM}
zO&z3|h+?Hv;n2A9dPnO3Z$Ps+^w5oqa)@yRS&!aOQ^jWgervtJ$m2P>!Qo6C?k4`h
zOR_!>BYrN>$ghopc`;S=V^=nE1&MR3TbK%_OK+zkJ-q=&!@Hu%Xx(hFvbF%bT0$tv
zE(wCwH&aGjTl+XnQsp7dv&(M2E?GE(xGq4NkPA%i-`^BS8qa9dVd>T0Dh5>9$*pVf
z-8~FiAW#y8s1Pq}>2ATKAnqMnc=3Y&PQ;M(&d@*|fR7^ZA#^CRcd8vk+tUjylb%M-
zE7tcVHw_xV;fpT~=ws17vbh2<qhtrpy<lpX=nwyTMgc)C4f$>tIC5tOA&q|p*ThZ2
z9Leb;hyhEOMMtGU$qQ=ri=|QPAqK62R(S}%Cem3=i#z-gWJt!cF9$N-@&wu7LogdQ
zxxgPcTH}=5;|AgocHgURWW#)h1CVvopSUw>WlNGHbLn@rI{1l@#{ZMgm^a1uKF?6>
zAcXvFRPJ%)1@@7kw7QqL7mDkvl&<u^OkPiGZs5J<e)XB$$?>8Z=U$CZzKG6kiO8)&
zoW~Xl5~oLc+efL$e~7QgYDLuJ7@`ij)GdZ@xd~=yr4`4Ch@?GOmeCyVg;dQ-iNgQ#
zghjFGv4dfQ(f*>W_M;Y%V?Jie3#FA>d^e($j&0x7E`tL@=wI*m**`n{XhkMA(rjA0
z^bzy3Vc9zqrDf*{fASiE0v^F#2}l}^L?FL6X(yh2p6_9ixUq;p$t>tu+0Idb+(#pf
z1jl_2Hs^<{#`1-iae3fErS&G$U<{BZf(Cvnf2MT61RGfEC@^vu?bLRpk_yCM;LISm
z{GAPtpzH1t<@7dW1(jpX!PYX>DiSG;Y4OouF|rK*Cuf;(W^3mb6}_@^LbtxTqmun+
z&6t4Z580taA#n^5Nbs;Fehjq70)^e*s~lC!X6-sAFwpb{>J9j{RAzZB$bcd$0-vi&
zkf`y@Pp+fzc4Pc8@;0S<%A`XU_*zQ$c?c|?Ue%nE6I}k5JF_2%psH|*{~Ch-#j9cB
zV)|dh3ES$@_J{0<{|++F(6Rnt^-$q;3c3gCZIrfOzX)zN(N&RgB_h8De*eD|9WB?o
zNkZG%p;4UF%w^WTMHu;J2swe-lZ~%CPp4mAV+htjM&kH2ITCk*rksjh;->2Z)AZP9
z+U6}p+|gl>;`ky#me=N*Q_o&h>Fvw&h4baDZf_d#xKvI1%K+uLGrSM0VWfg6OR8i^
z<cy#P)Ijg^C-GN%BbrjU*-V2ABHcpaa5FrFwl-RCrz>wo$A&U<bK87Jr?+~w&e7VQ
z)_TLK1~^ZzP^sO_LXL=+W-ZLB+<RQ7QS&3eCqu&yl%T{f;ut{)HP>ZyTL@fXqO!U3
zU|Z<UEuJmi&gC8U7Mi^Zb-D)^<=fY1CJiu*DCqSL-6M_ll~}ThrfTW%v^lUXKJ;z+
zT=y2|lFAM@Y&uwal6Ix7(csvvy%>e@x!9OGiUAqRzk|Wn_|Wb{oFuHQLzh-Hf(+jR
zF-eB+B_=?=DshRxGD#}PBwW~zIG<@ivtNGkDwo`(L1r2s1|VFrXlt1Hz#S2l!Gu2-
z6N5c(qExq<pa*qs!6S&j3pd1@^R{<f*|Cut7EghBu<R*mxhN9NEk4IGBS^>L)Tz1j
zvuqzqin0)DkN)1cxpClBp@a<Z80#d&WSOYWeD=&i-<U(KCOg1^=r&}0J!#SWYgJ1R
z@XZ~JM;>EcZ;vDkGo>CUL~4p46UUS-UeYjPe_f{Ljc?pCCA=Ie0Wj_3f#(VkkNpZJ
z->(ji0DzM7a0tOCXC-Ba0s=%+X)!5+q)^>``SdNu+=U^LyJg?(Kf@eH)h@p+lvRe$
zhN3-Y5Ii)7&dUD^6mc*Ndd89EKx=Kz@K@$oCmz3pDVSi}OzI2XGui#Xa)W@m##V30
zZiiqaUwK2-WFwsyc(CY*2V$M5I)ELh{7VOpQXDwfv+@L_lw9gO4kS%Ov0!*N_)9dz
z3L#^FpDVWbl3B%wlssB=eZ?w)#%6!_n&@iQZhvcj1)<o{t8O2$GJJQo+3Ftm&G5tB
z-q<epV@!7c4kIMyNc-c<<>NfmTk_Ifvg^4u_h2c2gO4ldcwsI^9<KMV6tf3r*bX>e
zcsA+?T5GxE!e;;L73Z}&xOAbIFDrgm5?7}-e_J1OMy_H20`3pzrNJgQvob|HarS5n
z0n=-&HKt!v1RfKj5X+P@7=EoI#;`Z^#RisyM~*?ff~L*!*LfSb9GU1UpQZ0BUjw1m
zwD{pK@Nvn;ohzmo9WzJXu>nr-ZLigX-MQjW@m`LioSk@3LG-DImc>ujFkGO0<zfC<
z%mm{YUHdvNESo*UBV@QZD^?>Hw2+MZQ$k$7gDpa~t#C>2cx*}NY+V^>P)(e#mHGI<
z{rV~5n$eu=nhlBtA=6?4LoFC*B?0R2y8#m&f{wx`mR5=I*R!{KPh?Gjj(F8dqUNu8
zME$j!@Iy}9SCUVs2yVRDjR}DZsU>gRC`>N}4RJvvxW&^oD&QGMCZP1|yI+?6gVm-X
zh`HO?@h=K)*ju;&JeX|R#(h6)0$%Q<^LBY1IhH~oeM(8LXk+>!BPXn|)?T1SU!jrz
z`0p|O2PTA>>3_$B#A?VU9kRoB&(_}A9#>exNk#tslkL@zVRKE+RxYLfQn&tVyv4>^
zA{AHq!tUq8msBt^vD4cL88VR&2m~R2=%Nm}l3>@@$K!Qh8Mpd@QC_9NYpu4>q~wqk
zwO)_!O6>=#IAZn3_w6)XUg>CunM~ym2^f*v^6O9A&+&dZ!zE9N*zeFwo}7riG=@C?
zPp2OiCV}s#_YeOsE75KgpSJ@OENBw>sw3(J!^FvyR87i5pJ#^wB!p4q#6(`s=FSFk
zC*{`7+OVmh3X+w@`mQQd^t*$d>xxZKBy*l^yC_*z^{<rlkAn3V)y=sG6Vv{hPhH66
zWFAb^5-+8F1@_~Tv&o!#bFu;6EL3?kR1EH^J*mpKJdyb)Rs9t43YDS-C2%p_R{5!&
zl(Y*Z<m;j7+(Kt%HL1Fp;S$aiU(fTM-~!P)n!iEgh3oT$FE&*JOzM^8PoG&6s@W0+
zYEes{x)zQ%6Nalni7u8HkXG`d9D|mit6G`Y;b*^tK-1WCiy=u;YHu7osnt;l@-z{g
z+T5^%EEq=aP6VBHpdq0`QQowMi|>46DI;PMtg)^ln^y7qrB}xBm|!pv6Hbv(iin|a
zEZ<vfJ04~WY+CHrbIJc24=bHnbBf)<5B7KU@aDu?nhC9HdvB3ny3Zz>pp9Zx5s^T9
z5XlEiw3$g9{mY{uni8|YZGhgqcOya6E`QbI!AoS;di~{B9YF~j7!MEbCWX~_Qg*bg
z;x}`k?yTWVW?fSzPyd)^nL#ZB8g-@TYc2v{ObB(`%AZz@C`!U4;W3|F%_TGfp`K>e
z_c1Pc(3$;%ivMxz#G`N8V|9@>Lsg00R9vW7nb=Oj_(^Y|$=Wf4Qd60hwHg3HGsH~*
zBb6!7|Ik`7)n8)Gtb`S>w<M4@E|d)NuXP1!C8!oor-{B!ptMe(=FeB!WN}xwy9`1u
zF}7sCQ4OI4_RG$%&)Oq&jf>}NE`d!7F`OH&<&q{@3!DC#u$32orDKv1sMkkChu?0(
z{i@mGZeK!CZqv`*O#W~aRq<am>6a$l1Cs-!M=m5k{n*mM-RV{sDW?H|w_Etqc4zz}
z;um;P%b#F`Q%O7}wIoECN36O?;=1X5-a8t~9ZWPGXz?}S&kdz?-b4QkoTO&pu>+9!
z;!jU-Y)f?hWRV(HD3{$BpvJIdr8{z0$WDc>q_=8~(lZ$+ZHq1LGs`s*H%7=S&)_r4
z52r-H6(kXb&)5<bl%r%ivUP)%OO6`ivMg!clS2^W`QX|-Sy+QF^s|6Yg<_HGVrTBR
z&FkGaJvJ(ZTbe^lGJ{!Xrp|6957mKp=etx1kH5>F9)M)gAelOr(*(u%OsozW47cFB
zg(Pe=!+Ke<ZaEzWa5{2A1kP?gTs=^qUQM=}U|uwYU0Um1r?Y?Qq+WXN26SMVB@(S5
zTrx>e%tE&)$YF6&(^4=*m49OSDGH#er$&&WKi8QB;?_nLR?uBfx8=WXF)S*;+)$j&
z9Aw{2g?zb1Ux);zGZgXUg@U~`)<wL7_mFH@S&Mwp*waVsm#NG^s2N8&%4&qbYO!UQ
z`pJ$J&42QmW%WOz4LOt&uCzx43D!&{RGOKkf^}-kfKrs5y-CA-k?J?uQ2E5hQBNt!
zP1b|yl>M{cFd-15>PccnuA_GBFLeOQI#{}=j;5Sovof;vSh8Q_u}~av!)=yhE#<YM
ze2y8ctgSLVV;Pp?(3Y6g>IZGtHB$q<uBe__Ztl$~hhK&MCNYTcBmiST>fQ}PtTcL*
zgW;x*&f_yVM~E%Re8*Gik8geLaju$@{D-9%OPAu%2_EDIsLM_^kz4>1lhUH~AZ+`2
zq)s>3ezzjybptZW5eVXn(-hN|L{z9KY>)w}xt+rdWKb?m2#R@UHt$>e^<Rt*iqNY#
zR`ZFSMLdr>I_%x|g>s;Y0@TC0^gB^G50K`7X<i3)RmK&S*{{9br1R<N5P}32%NF{c
zyB)<b*L|x@s1U3NdEmMBGALgR-73ht7+~b3ZA?n?YafM#464s;XZ<UZi@06+-!7-Q
z@-I3-Nw+ef$wl>;Z*<v%;^F?a>;SG05ZY|l*MIjR`NFu41Kn|*SngN17)rWs&dx$+
z@4!CRs_R$c!BCL`K8choSW0A(W^^Pw65MLect34EkFn`d3VXV2n=g3grvWp!U<^QR
z#S5)OaJ;6;r9vE6(_?Gl*`6)rr2r>;O=7kMzUb3apH1Y9afL{Wo@r9^nxMRmK|BH2
zZhcGzB(lm4{qf;)JuPDq^5q*~H*IPc@bJ;nuf3OzpM*q;8dj_C1dxLl3F!1c?Sqzo
zqIbv<Ue;S6P}sE|>NPMIgzsrX>T{De2$QC*>tvSbE<$4#7-!lWTZT7|3jNnpGYghL
zuCG(fUVi22CU`va-`~tBb@uA|FP)~*GYZ{ibh3rOxHCgMoq6bSee1DrzfqVJ6y=E;
z$5ZGSzkp!Ig3T@SL3A&k=$hUcvv@3DOt)?p&TS+3dV}cLTHYRB-Jf7w-Z1bT1jI&;
zeQ)5<cY6+ENpz(+LfJu6e1x|--H8tzpNtCfZVoHHZV3Gv8Jl)#8K#l6$1e*|!5Vt^
z9>uV{A&Re0GfD;CXe(_{XllhW!eHn7*<oRTt~CIKm{;tWSy8hv@Yp2Sw&&eFH<r%r
zBmks#*e1(hIRG2_(`v22e#}?vf7OA3mNYZudEXKy*^NdSdhS?~B8zaf9Eqnc<U&W`
z2Yf1@1ny{S2<-x#Jf&|9*K~r!dnKb!1&mm~lA7Hp{ko`MJ-~u#Xf@FHR2te`2#Nzr
zr4ncK=Du(3U%&SEw|{^QN`!*{ORRAI-%TYmE6e}<NDI}RblYM_-oB)1gZU5@2I^^V
z^0A3$IpidH_CS|HM~qDUeK&t1f_;2h7C`;mppuS&4tuB)I#gS=l@Namg7?kAvAg%}
zynD}7p@BKN!N4E#fL1ao=AIPp#q~~GgR$RD-Ew4bR)1&vYJv6h;eOlSU0o5H@Am%H
zt#>l%Zkj=0>)^GPBTrzf>pg^E_2=*BOI~ZIz>C1o6Mv+yl+=VrW+1WJ@4fO2mwu^M
zK}LL`Xb9p6<<t~nNid6~<o=$}_YVj#kOy9$<lt2ISnO$Y^b7b-3d~mF(1>7=Vzc@L
zY71$kV_Gm@cJCFZM03_>R-|$Rv0R(9Z8qO$j9)2(a;QX>VYZrwtNng2bh4yu%Rl67
zE)9+o1gB6CBYt6Eo*GoR*rykPsc}!%D%<H$H*sZE0s8;E<7|moO+9f#f%IKW^z~V-
z*uOnVNfq@UVI8r+HrCl;y7)88tcO%U0I9@;Wy7sj?y6R`@S9r=KI(oOv3Au*N{Dr8
z#=*Ca>qAyUQ?G`#b$k<PYX%yDy(-#<Wm+k$M%E~YNOwBRD}5Lr*%zNVOjB<gR7^Vg
zK?3os$`jyKHxxJ8%)`?n8#6Iy|5GY;?3SW>$3$IQR$37nPN<*zb2|8E*5kCFD|<t%
zH=LY&nU*(s=ze0ow>NpPbUpbE5Xx9yG|{yQbDJ$HT;2mk0$1BwJn>S9tIS)1y1@bD
zQm?uuV#L%rQ4MsUN$KqOr+gZ%Dyce~k#AV`q$&fj?C}Z#J_1*=tdT$Irs%1f9mMDz
zRhlaWa7G0m;Qo;HjQg-g#_4zUgN9_)nRCkH^VE4T9d>7rj27k}8pu0H3o#`wz#S~^
z(4`V7giEq~UkA$$x9lDseby(H{aul^`jy~W&CKFwTg;IHvi|XK9fLfcWjK6%i#yKh
zvK_aW+=zmvu7vnHLRC@uvtjFBa*#;y4LCvtE0#T8kD3UmWr8tAceNg12Q>6~!pRWo
zTeKNI?^vg`fE}?WNuyM<cQ=x|a`j*z^hQd#9+FS*dvir1YUGhekIozzK}6i!rC=YG
zHGJ1f94-x?!BdwNq7Va3ajB3ysQuDbMUr^o;5}xj+5%8#pf~BCkR)!g*rX?OCdloT
zQ|Ci=RS|=l`q8y1c0rl&Xxx)jU;%2(c4^UH>eq)2CXShfFGYX3D{BgWFA*@Z2L#Jh
zggS6#76IMVsFIy3wrez7*%3=G6KwyLfXzY-&lgj)S)U4}=i1K}v7chj{cELjBxnrK
z{Tpn>Fn%uAFHdB02!585_<1R3>zjnWQEcv}A|fq)ca&?Mx9$^rTVF;87|o9Htd^-G
z#0Qq>R?opcD(>{=t-H#tS44QQ>>84in}g}BNd}C3{^PiCgE<C)w2bv=vxr5<+d2Mj
z;AIxTlGgdPGZydjZWL&Y)M1Mh*JHvJ=i3~jN+3qhQEru+in@~I<-##(Rz0i(doL9q
zy_5SoEHw~v2&fKPw0rWPQg^37naLV!(NS5;C9u#~<*glaqmx9rbo6#05m&+d#%x=2
zAfUa@Lcm}b1@@g5JCW#YL!}Mp#H8SwG=`=z5~<K9QD<)$P1sjEV*Sunv{4en9nF%l
z?8*DP&#?Gwj=mJ84W&7^FF3yk64(p~>5e4YkmaB~@tw7$**l*mgZ}W>Nm`4n;6n;b
z;nty>!ze8pmm|-@^d2{roE;_AHQYQV2m-c?HRDD5nd%hGfreQWf4Vd;<$*u@TtCR9
zM+`0p(mE;lpm{dvsl(Z0^GVpGB0Ldb4HiCKc2M7!8)WG&iyxQakIv^9qx?2-hq&n)
z>HUWCpTbR`vg^PRSS5EmAa{w(f*SPYN!8~+K=|ta694W=%Ra@6Zf`K?^Pn35$xU7j
zYN#)g-~7!}u5P^4nbEwXPhERlb5iPRVXwheY^5M?#E;j6`A85ApZYchaFMV&<j%!l
z<5n$RekZt_$%ZRFOP4Ydjbn4?PNHWdUf{v*#=C+J1GRak%e$6GOXQQ-E|+pjqT{L7
zWGc0+FfrL=v=fDJ{D?#EMy*@xbO<<P@ebd?(@z#6ltdJ+O_DP)G-=u^=4v05(w_%x
z<GQ~BR-~rbQ4-wV*U^QUXp)S^TfEkaWhQh#N&h^^GM?U;gq>1v^kv6bHJQRVSKTsi
zBc!u!ujreLtWfx-9@;5^_KRV9HI_FMr$y$}&gIUUKioujoic@6aG4*?U@7)XNh2g(
zf={?7y0U|NN>P)*H;tAU8CM`l;}w>+8?uaj062RSphlS}soJ>8#C&g}Mzf<NG{E-C
z@9%{@TDn}~(Kw+}SSz=ZAg@^C7qR%nGJsqiNH=I{wxY-DXa?E`@g<X%79c*wu;$*1
z$wNNs?W;IM@~VCSjzh@%on>MgHW(_Z!r@dc7#V~o@~X0jkpb_J<@?3pfS$pEI!fYH
z2k1c5inuSztj0EgOlrt9Wc~ZFnqdgW&?yj74(7#TK82WV6niB}hK95sV{0|Q%?E#`
z1j5{KqAU?D2>AntxLmFyzx!!7I*%UkvSuyOp5S0ly=Lz$+jB&9tCDm<zbzu-o2%Tr
zcTYT{sifrByBYzJ=gT9DF$1Pnsq@Qel-He!{OE5`-yw-yuJiRo2d__~4LH=q(rn#N
zM^QwPy$#ht`f$MAbdM<2L#(U8?(vN5nQP!*C%9I(#7WSl?u+B;Ky2s>@9otNt-|f{
z%ljBKvSV}yJTGOc^XsJyRBmi8xwQ(y>Sk7*=O=h}d~aIv^J@~C)sohv;`1Yd-;LV1
z4H#?(`Do|0EEbqmDm3GgZ1Ye-uzai<INq+5LkX;9LyH$3Ex+X!RE;ez;irJZaPA+^
zh2|%94jTBM=u>>6LfK_^(Cc|!)zi5z1z`^77Bvl96!`jCLl7kIC}#8ES6;Z}f9TZZ
z-HVly);Z)w`0<_&hCzrYswA|9JMfP0*Ur=%?4E&=f7@DxyMIL2V(ef!3A*b5RK4#v
z33&e$Y|hoR(FRfQlsC(>lleU6L)kHW$Ntzc+TT=7IbYI40Wr}9!&XQf5;oIl{28E<
zts<*+nw26X9uJyw@cV^NfPjOX8*cCaiR4|;L#|gdE*o%z9k4x%pGS3OJ41g)%_1rV
z3}tqMFWNUIZCI}YuEOhq@1f-K=8S%XqnY@&yd7;m!fNIzzICV#uxz(bvBlp`J}#)w
z$&w<_^{eOLwyS-5!l-%s$?w{C{|`OK{9n#9C*%KoAXlw1>Gscg-u|FzE0(jxL;xX8
zmC7uZ-Ar1VKPMjtNi~RO6Qd?IW&8RI&$uU@NMO}K%m!k11AF2qguN}!>gE07<mdKI
zj+8*I<%l+IPh3EpK`D%aHJvtJmafS5!R+t;9;V2rH9|$9<e+e%-u?G@IQ;#xavdy<
zb!O~Jm>S!^&FkmoVwix^=l$bDz1P>JV#b2;+sp6eIi{Z`RwKDMgc?gKH!+!P-pKdm
z940pPZnrZcQKf8trdCHnxR_LWzd3?G_serAR1PgkbP*O<4`cbj&JB3S0-K}S@_Hkc
z4wAE^UbD$lm3T(l$Hpc{T^iw)Qbrsq?Sm2}hc7CT3CR>lITA@`O|)1J@noS1e30)x
zqCe7sGXwFu0pxf2%z@WP!&>Gs;IEPC19SHJ(4JTsco%sPSdy*n#7SwzD*6&-#7HnE
zGI6(r-zLXq02i64&8=*GXaGMB;36YnEq$ZP*81`UaQpb+H-Z^Wv<)Z<COL)N&(Ty_
zZO_QHHk>{g39s&OEhR31GG0fWMu~EF?K!nIUdbB3cjrhF9XpCl7uf%yN)WhX417@3
z?E<@a^TSwKUO*Q>r#-N^K6&?awwBfoUO*hF)mCot%-T#Ed=sQC?U&9)<7K>8x#L%s
zdPEM$B>q`#I{`&@Lv19KVlMtJaRgsicqbzJ2;tTXk?>9*OK)_xs$8~-iC3Tw()y!7
zgw<y{?*nJ>^T1|uv897Sq$98?2Asj~vdL8w2u3Q3U^e0SECV&5!raWHe=Hl(XF$e~
z_70Lb@RfAt(b|AX_xZ|Cc=1v#6K!wSRZ*q2<!hVi6WCD0#I50T)MQP>D<-H4XWa}9
z#%Ek#Ysp}xw%np|@X^dtWZLxveDAic2%X6l%wF}N7)0SQ$4&8>+9g)tzE$}qL&ea^
z=NV~K*9jk1iGvC!d$QFj6uIJb1ffpq3^(x3l%5PL+ff#e72g^$C-uu`K+A`!E>5$n
zzoLhI)ia@WSHYu^?n7~sd`##vklu#rdpZ8r*@we)tuCJFE!mVLPXk0*vq&EAQ#L1U
zj@iduo-mX)YBL9iR%u8~Z==q|_JU`K;y`z&M!_myYT45n1HIj9<m&i$cOPtiL%RGf
zP>6`hGe7p7F7D)eJw!f&1^DJJWc2{f4GZ*~;Xu$_89HQ#h?9fjLGZ8(F1pJ;W$<Rn
zHG30U={sed;A?&o5fDMI0?T6Nfd0$Q%C0a~i<{1kNIhL;4G^dU;(9C{XbKklld87O
zV5Pjdc_*xRSQmNtG<@G9Em?MHJbQ;qqUeQ$WXa+ox+1#~t@2*3^a-95?WOi+!qmvt
z<{fv%-x^~)h#FQY;9+CdN2xvN6NLD-DadJ5qo48nN13)VNzU-ogg_X&*fGC{>=6z6
zffJ}xOzjlrXolRDzlh^|nULFbPR9_}{ixfpV2#Moa&qQ1EP0O5X7isc+5DlpGP?Xm
z%I8O1^m#)T`Kxn*^SDreIkSORsYNlt(9v1^D8QA^-BUQvK$EAXkjX+`%;nOkN*O8A
zgFjdXD3PJ&L!Cf8>B{5hx`0Znb>hfOX6i*bFSaHRQ!IG&c!3z0^XKS$TK}@LDZ<ID
zkciPQp@I9`%i_4DXKpe@Mj9;M66ow3Fqsg)s#_V02is+&m<#%+AGZOFmvr4xwX;5&
zt*7YKef?}=KX>>8uU`th9D9QDZ1I(G)wB<$bxqpG(W<i?3;h|%!7Mt6xd497+!$ic
zee*}I9=eK#b#z;Sd`BQq$J$NCW+cC#vNh%QY$dBeQI84RfCV5^tT8xj&5RSsELT3U
zXu$lTsy4_3_NZ!Od4Nt@++m5-RpKy~TaQVgdEbSbXxo070EEUi6$~#ZsqJr<5D-yt
zG7<uzH6-Y=DhgC@STS?J4Kx1pIIVG>AZJOkB-V1P;*Z}mIto^%>jYy08VR%!1!4H0
z$#D{bPI%<3geL+<Icyv8Y)56x^Xt4r^W$j==SjKK=<VSjA150veF8#w)(?Col-kDM
z17d+NX)SD?NJBg7$5PIb$Vc;pTB@Bh;sJXD0)yU~EVMh(Z6M<PAY@m)e`KxT|Kw{{
z-O>33@U&U1`(GJduD=E6f#=Xgu-2t@_mGH-j%_D5&%pnUBjuL8ZzR;Q&u1DBDQ3pP
zQ|DC8QCexHz@wF}18%i%b2QU+|GT^k!!mV28q1-&c$8jF_>2WNOJ8IdC^Sx4w6C5v
zFPAWD5y7rjryTSSV^8G2Pq!tpNaQcHW{dgo%5r+XyIHCqjZrl-_XH@=j%586u*R~p
zOZ%s{Z=gLy`aHkC@%^rA{bT#eH?(eYe`HHnU3|%d?e7fWkrxfq1M1bn3j_&!2KEF~
zmGAnloPE0Ni{fb2-zaT}>pzC;zqQE!K@2dlbN$cxxND979|$zJX_9Z|kwjUz8PH_W
ziK^RKB3@5K2R@vNSI|+Mn9Yv!^8bK9V#71dVqu01=weByKBoiD{Lm1UKoATsxl=DM
zc__4?ktQ;S1}(4}xT3t`=@R`mNVwrpg%-@QMz}pWvH%2xrP-^y>#Mu-{>t%WaZVo&
zqL2^>iEs#cJw6s?vvYfUz25it*YWW6$*<FFxRg7{eM)dxEr@<$x{X2fRPP&sG*Uvq
zIAx*L=F)YRQlJSj1Z4PT@F7({CXl;jTB|O<JWTKfukFKA@!0>z*gFM@(nd?0ZQHhO
z8@p}W+-=+DZriqP+qP}Hd;0t_7ZdS+5p(LgE~{!~tv55DNv#y0vWoUjhF5_FF#JD?
z4*!CI^Kk<pk>VgsNlURMNY}lv;6#ua^yD<&1}N17qJ;WS;i6HPgTfz|x|Q&nBsGl$
zx>)i%lq#N$bZI9czyR!keoVe5&F$l8xrxnm(J$&1O9k9*S>#Ce5m>9bv|C*6!k@N@
zGSDrU2_?)~^eX`MRn`#jU8oFf6=wLGSFUbS*x$|<1I_zCt(spfY$25G2HhCwUt;)o
z^#K>?`?7>0b{lk+`-z@t0$MzK(i30e;}`-`{{hBQMx@Gk@~cD{K@P#O7YgXWMy<oq
zL?{sLGcwU7WDp%d?vp@q6!I2<t#($g(^a)En<D;pB4HaAvx&ekl5N##7V5Mg-6{kS
z&C`1{%D{y#h;BtxOH{aMuopooS8CVlTqh!o8crGrAQEq0>Z+M^jY6RxPU-3y^&^g4
z=KYfjXAl_z_90#cQUpV~%F}m5Ss3j%MRVD1c5nKP9Ay#+wJo@Uy=j~3EH`Q;3P?iQ
zO*!w5$=<3-=EP%I;^e8d`gHEB$S{b;*ZBrh1_)*jC7t<3GX@d&#fNhDKq8Hue$v&G
z@w?vgSLNLw>SS97(^KX{c2uo{jD{e%Mn+g_C`tGsN5S2wO$Dl`n#kW};Tn-gWr}ky
zA|2z8=U2~YCj<^xVzt_kWw@7~>8P5jv@|6!R8vuRiNy-5tErL0Q?WWv?x#2vt>dH+
zMji;BgoM9v^wZbpOH)5&tFj^bj7RoM`(pwnBWCQh7p$DT2_>L34e_)CVy=&}SK)z`
z_ced}p75?ID1dz=!1_@6|8UyW+c<kL*<gV1<P@T7Wc;0_VOlB#MVhFEG1DZjJhO5Q
z&0B_$0|$C;Eqeew2v9Tc)=5`;u6luqiZ;_Pxu0|eYjAk?6P+gsaG-@f_2*Aah|o;2
z!tj~yP-2zLJh2kYIX_t7{(Hh_M>WfLvf{pl5wn!N^v~UfQ66ndf1UBHKz;y&0WNy(
zt_AR@j`93KKdx&eRrPI`U+ln-9J{~xQvjTbX!my3t1+U4Hv>>IC&=+imJoI#C_KSg
z*U6L$yrI0BjlS$nD|C5U9Z?jngpYPz>^41Y)&utYRM_ID44c`OZCuLPBq&|!R0*sx
z2@<D0rK~=({#M>ID&$IoZ6?VK=VMti1REtRtPJ=cyXnkql4%|B)?mC1RCgBoMX=nN
zg`WDzKg>VU!(a{Z)^-lEu(D<J!04uhrL7Rm5BmctXx*Cpm1may?LzO#jFVvbgrzE5
zX?NjE`ZJY<gd$FvfwLhDjr7r4!Wz0!HelA@CBBgTgpcYfd>a$4;nvY?0Yu)k)`6_v
z@Ut{Ma%MQ3t#*Y=xTp|?|JZn?!~KCfD<kv2t>AU`Ef1i$64Lx6^N6H{mb5jQOE%SO
z!0D^sdn>BtVkcYW)#S<B9I<pv5Sk<U`{Yhax&w(xC<tUh_WNA`e(IG>^e(HUa@dZh
zs=JgSFO#q(0KvI1_%l>$FKP<VT{n-|MV&>-0L>_l-$ROuHE{awL2R7#GeO)nYJ*(d
z#Qt3yfdnl2yFf-kSBIJ=usa@h7m&;m3=zbI1K2=Q9+Ix@9)#p1pwLCumpJ=syy(_w
z*_)>(@&S(1CaiiPL-zF4jn>if9Ry`gr)#C&2QEbtiX4NNCM{n~;aL20i+UFmZr9~+
zR$Npg9#k%@P1>JQe>4=Ag51xrmHGJiv+I`%KXz8t&)4h;W#EVX6DC+9lb{<l#pUz0
zcZ|rljF3S?JG|4j_&Kztb8s8u^)SV2WN}9qIsSxy^j@I_dT5<iGrN3y@zKv7@q<Qv
z<A~PP@I+u%H>GCcQ3G+tt>@JPpyT2bq%iqs`sZUhX1F^yYh^9V@3Y-8BOu$4P+;?W
zLn9H*tmKScJg@oVCv1nCgIwdKwysSBc#hLpWOLZ63vi!-?Yr)6ad$+uZJ)ZLm7W-v
z*B|2t*n6Q_=f7s&e`LZLIetOh|L6K2;I@^Mi30)sA1ebV6A=?5TVoR_K0YW%CkGP)
zYbdw%U+S3SK{KN7i(a1?mt+54AVE0W*(9lB3Ym4nqCz%Z8VpGv+6jdsA*J4;?0bvr
z@2^BavPss>`z$(e5ZcGrhE6J(E2K<!_q&77)4dEClBn|ZgRxIW7|*Yh%bYCXjmkZP
zNmKWS>F4<f`_EEXl#*=WZttLi_tT0R|D@iI_DpM`g+#R(Q^^@T`;M>o!#Td&>(Nfn
z&$%tXrmET^bcqP6_PBAoi|d#X{}#{7^2OVdts2bWd=RM+!nk^WWM)`A`NUKbFM1ou
zJJX(qAQVFn^VroAe@?~GYF|`%A;`)Rkz1+>2|e`V7XB#;kVv8(sBfXb^&BJvsB96A
zeg0UKJ&}4yRCD_RHgjrq>?ThVJ*t6zaqMOfeRB3>eY|mHRItmRmAKZ!E1D9PtxKP+
zpS6c%P6otf{zSr?(?KoOC?{i~xcbq6czQiu;sqBI%Z7e<vulz(x_g7eXbps5)o`&Y
zNq%3KOgCA{q5wwck8LM2Vw45ny(7i+SN@_GLnnuW#6qR2pMYtsN3o%Tpa}Uxe2kdT
z>g2>jqDbV^grc9>v*I4>4w=Ha1vh=|fBoy<dw%)^dZ_mjBA>n|vU*$Ak0mqNNoVCB
z2o*an?U>~Tv3mSU5;~9i1KFcx5Z!zry1xn7;F*x;E|UwtTcB7|{6k!#lJB~8&Kf@_
z$=-gs%++oT&eW`@fZZT;s_|XpVazyb83w~Dv4fw$iL7m_M(3#NmcLiH4Ge*NSnPdk
zrsMg5lV|JqemQI%=m(W+mvCNupu>RMCk^=4&oQ-BtTSj+4g{RH<>?X??pzjlC|LS#
z<`!1z(!TH0q#X}gvHD<+5vd|}Kh1AP_BS`gCpYAfR4SS3zMt@pE6nZ8(Pxhg{_ZB(
z6l2xZ$1_x845<#udTfQ87uEAq^InUkNG-9@6&raPI`Vs+exvyX#|m+gP&;Pu6RtQ`
zh~OVa$D3o+KorWD#aBFg^(4x5#@&GSJ1nn67|SZL+z%bwR(~J*`C0arZ?$)Eg_u|k
znnti=BTNA}0lo621k0a2POBF@H=A2p@xZX@G}`9_L(rr3S-e%?FI-$({*+tZR+z2Y
zPSwAsDaNpuRbg%`I!?1F7?T9XBU)G#W!J_aj*IzBkx)he+0vZ5yusbla{zA$fC@Sy
zbY0lDtUUo=FTZc^${yTTTkNp6cC9>3|FkkRum3J<{F8R9_8O%(xVdi#|6cHS2K;r^
zDef3P>(9!#G&o}LZNyp(4OKVwu7lK}bCdN<QEjQKxOxPxhxLh$i2MQ(|8>4~#cEVa
z_gt2X68yqI899g3b1mh&MCIk;aSZ#?f>w~kP*3(^gm?1!uvJJQ0>@DpU$31+;Ds=o
z^h~_Y{?+K-eKl(eabgpbPjBlWzuo`q;rTL+uj}ALfVCE1L$$+~CaJIaHo+T&JGY{g
zQnpvWfIKOF0%sV8t@Efm5%%&_+A%}{=rzFsQ8kqhd9X(Sk<-#&++!Oo7@0}ZSY81D
znoL5)<6`LN3i^OEusNCS2aU`@V2l<{A?H6N(7r>wjqe-&C%!hPh9KErH1lBZ#Kh`R
z*x3uL^Fn*gL$hTNd*RW;x^));OHI3wgJ*=9h?xWfHscE_9K3L(AC55RIOe^PXm?T#
z(FomDYPm9Sxi`+j`)pt645{ymD3t>G+OZ}ObOZb)xI>WJ2xMCFI0Y5&cqWBoJ`2zG
z1$Y!yzR2C%lt?4zi{o^#sXk~!M|40_G^4(HcC~y8S5{8;>Cvq({4!~!8{^!Os~?F|
zykKP*KT|Q8qEIp1pXmVNC&;jao3BdD;aAb8a!}7oXU+`I=xOZ3KCvYa+8=yPtGb2T
zL!^w`qGd&_e$V;DQOgZeqr3KVXMe<F8_P@TXEV3WPk|OSz*7>TkjKAvA<FX9HZlkq
z>mV9q)E0bQCJq=)1Ek;<k*N;`t9(z~V(+Ta9Q|~=Rqa^xt!#lyd?zw!{s-!OVi}@<
z$c%B-0SzFL*su1a{|<}wJ|(%=AXu4A=`OnjCaziQ^0C(Oi=(8Pz@s}GmCtOmmStYO
zPr1~s^vsNrfYX2~(pLHh@BxBkSV+Tkhg5WMM6lnrlD#Up_;3Vs&if6of8Xy`;>;7J
zCymI2k|^)`?IWl@Uq0x>#Oo#wgompsC`EV=yU1HT2MtLRFYzQZX|Bjfl_`xb_iAtY
zD5Sp4pvTK*wd&%?MdyRtr?f%Kjj1i2a{jGpTu~14snI)j!>2_%RBNP|YrmaAP#+{V
z^ehMcc7C*TYh4a&T!pIReVIxMp>1nf<rSA45ml&i-+lpaw=(ZD@vN*OdLnxBXc!M3
z5yGYAwi5{9QyqX@#Sz7oFGew5Iy-0D-4s8(J2l9C6D^|M!;EZUX0EGba305NjE*w+
zNVKW8vk-0FmmLQ};`0Ztr#oBk3*CH11?N9zFw1{c+cE#|59zZsHXJrv5xrN`ZfOgs
zVA0S6Ql=7Ik!`J!R=9Lkp8K{X4aB2DOyDRVh%|h2z}ovJCfyDt&kO{#v7;AC;0f?}
zMR?wij!Ql=@VB9|#W*LB(?niA#7W~~ev4Ufw^HUtWQI{Xw$6iWw|!r)Qs8a3w`_`z
zGjhJ4HfDZ4@1AEqe0xDOuuK>P^EW|+2BoP0A3pAUJsp3*I)*IkqLedD6hn3O123Eg
z#%jeG=<)`M5h<tQwM8_Ir3c-ee1pDULd(I0nY!ICZA6<b@VOiZBj(YX{KuV)$|i0%
z4fj@)o0~~tSr`W4nlEn5v4v<`1S_Pn`dsdx?nmx==+W7l7&J}eNqKbPnPuFqDTfTA
zR4*6<-JNMk9gCZ^V*`aF-Qi;DyM|z~Zz$g};Fe|D)4^o3|0eg<<mA~vGb>vI11s)X
zzkoWm(w6&JVtYzOqb|X;7&_Lf+pJ%kThe?W_ZoJTROjs^l!^Ho+u&jBXg?)wIL6n|
z`b^1DnYw#1fqfP-?R^GO<5!-AjP*OB549O;N^;?_ONMh?PN3#gq)T#JUQXTBRi}f0
z<}+HZ*Hk!7=MHE8nR*?(;RX8QaIWM!Uwxjfu;gxZF|Z^{J;Y4cRQ;M&>6pu+fc`+b
z;~L%<sn;$f9L|*5w?bsI=gM%tauTUb1Cy&c?g%&>gNy4&JbU86KhqzcO)K63P7R=J
z5Yukx%TaP`_Xj9;St_HKwodt2)8(?5A`C*sgvSPxKScnSKm9n*6Wk@w$Q#lWY6w}l
zqH3tPQeijAESi>?NYY;}a8J@L<ksxM3-2otw(?Dw+s`bL@T%`AazySf98n9PI<iwH
z-(ClF^~#nXPC!ni%w4oHyr0Xlm8}l!cWsk32h-P>ab9vxKo$C1ycmv&imA{jqnCsx
zsBD<kmk8dpq*VIBlx|uYg8unVViHkgc{)Hw&~C4vy3r4j+3Ld?sZyqqZ5g3*@?sRc
z8vyA-{~C!G&@YlC|IBmllh*`2b=sSq)t4^oS3)zk&pBP=w(c%fd_f4`T$hCjq&kY3
z<PAzqWZ!7Vlrj?0z#ZuAyRXgG%6#wnC%Y5OTpt=Xva<Mz&m8K*@*V|J=_1dy)`p0E
zCbXO*7xEhf9V{9BROo1zBA@1fQ>z`fHHWhG-)r2Z$tVcZY#enn?sxKYS^nikh!^_B
zFT%j0S~b}VJ%a+T0Rv>vAs5+O+tvqnHoII**c4vhnb~%z4WyMotCd74>f#WFS+SDg
zpa&R6dk)R?v0T&-38S_?-$zCYm4r)MDNu)z2jq+=WwmX!lR3xx{0Y~hwC}q?ln}R^
zQ(7^6GkG}iA_XzWMp~z3hOZhq6Ybm!&Tj{|DA{O3S-2w?NnGRFYYSsHjr>A|N?k#{
z@^g@NM}uXh%Aa3m!%C3PkFHM_IsKY058=>~2Dr?f5Sb*L8g~-;sHA?Kd*E5b8P=Kg
zU>40f^t~<nQ4i-HPg?L<=*|mSULxRg&c&&<Oo`ewD<gTaWVapA39r<)zpG?AujL8v
zlW3LHYYc@BG11Y-S=g-;uz(j{2r<f$su#15@lyNhCNQs>u>b84;}mS0wyW50-IV9L
z-#93_`Qi>c772{tGS`%{?maRTSVAgNgG<*)XGj+8b0-seg>klGqKslW5k;^{GTc&y
zi}A*>hMrrbqq&xCnTSJRKA@UvmT0AnC9n_f!;!YU)sTps3ZeyoV|QO48a>yka}MpA
zfhom}+DnTS!Hm8IDB$Uu2GmE3#5ir`#i_-LjswCjA>=s%i4qYi5oSq#zh>h(Zg(~D
z`7gyZm=2d%YYsqF2&>%)dUf9{Gzr90y2*kQ;`g5Q^&niB;jS45z+WJ-6HXKWLP<Kh
zvx<+0k*Zz4_!pg4%)j}oXVcZhfS1LNATY*#BIEjD{Noc;=OPl~^6yM471+Yvg0g4e
zpZESJpxcG@aL0%RD12dAxyl*nizincndU<OM2gElo`RNNZ-W&_@nrp37ss&o3ULr-
zahIQfwXehUw)zM5P)Zd0ri4#cj8IU3XY}rpqWUy05_myn0d#F6&s2|RJhMTsZ>7Vd
z`h|9lf9?PgZ*NLYD=C40Kcjbv<UBduWCv#)oRukIH>e>R9!!TDy^zp-N#H1bmh!m;
z5~kuLE5Uz|9xIt8h&aniB`!k%Lr5kbxA*w|HAsgifUY4qr88%M%;2|hZA<s6&$R&P
z?%6lDzE6PE^0-8zK%x3?<(m5w_N*J?Wcs_N0zx*<zE3;lY>=lP>GLqe!_2{|n3#3;
zix(sK593x@el=#-BK3RiqdY)nib6^L8QO!~<vF@I1Ffp{_H_j!tY0#4f+-Sptt2?W
zf%@t!=E<{^hVkjHDaNpDpK2t)1p3>Dzrb~lmtXeT7l)rqVZw#_%5tWQmO!IOV~<LQ
zvIwbBE#CqQEO1q~$s6<vF41UeKSQFLwllH|Y*K8L`&2P7Nne~dvD458a|CL}mp~A*
ztxSfOq3X)zq509-W;tCm6Yi^yD7cM#Rd*G}DD`2EEeDv3Q4$&pOgUE$*u0g||DCkW
zPs0kc7Y!bdp9&!ThMs<45cHR!tCIjP`C&HSn(^0UW9k{gojXTk-mWp7hn;)-JygaJ
zun{Wi9<V(#5+AhgfK%izrR<)MtADGXWuFpeJHZ=h3y2?qoY+0&!ig;LjB89!XXm{Z
zqQ$j8W<9HeLs3kIyfXH2^TMuYgEl>hTDnO>6~m7IbEp8hBnv?;-#^p6fT*9Rew}2G
z`{nas>$L(7y_F$!P74y9C8#;EOCF0}Crp@JqW_a#YP5>?Pv`g^1Mz7v`jv)bS@=nP
z!><Y9>uU>XLH@}YGIom=pvr;*&`jLJ+wU-Cj`yEi-^t%{(=Su-fob#(tr6JCwHW8{
zp)A-5B-$(`4=9j4iLS<IEe(5PG8$fwxcESaTHvOuJ9KN;n#<ntFlDu`%O<015Ggl2
zrzKr+#~7jwZ+z2gr=6}fC1@}J>010y#0E5<nH0<(xN{f2yF26{>M1*{C_CU;9=M-2
zKj93ukpes8Pd7rz5|E;kFFJpZcj(?Pd_k$k)Bn<1|7D0|VEtd0{#U8*By6z3`Rw%c
z+s;iJV}2(N%6c+ekZtJRtediRds=+tHnlYu*B{0m%6R#H`Gc4%Juhcie~`ol2q3)P
z0+QrZnR-2MEwwyUu(xPV(KNJ8nmvT3jhp>BiTcZ&L2uS$EOdLZbIjk_hPp{+7%yd-
z2tq1&zV%|uKe@A`vm-vx5U;b7nrViA?el$oeHF{`cGctS^HLM5yM$UYw6m!hSK1lQ
zWoz6gnNe{3qYF(~UtUx3F`QQ}x&mjUw{#_b(4_UL+d|#wSdZPu@$%<B4TAe1B8h-u
z$M5)M$q*K*aSBi3tD-Mf3WJ?HH0PW!g5{|mYYQoK2`U-?4@!*~l4iT8($kW<D{VNe
zf?;`851PqwIVp#vea4mx?{vq^lIce4iVr#Sw3R!Hj#gh(qq)b|?vcA{pD^xKi2x)k
zbsumW6dnPp$=I(*UxInERN~!hGBzQUBrD|6@aa9W?^ywY(Tf)xAjK8|9FNMPy3kqj
z+H%U#{(=1&F)`{*irO%I|M}ZfKR(>_cMjXA=S3glISO}L1^BXQs0{&zKXhQZioh_o
z&mE5*ycI)LoN*!8L=i(f(l3u=hHptG(AICzpR58cry3qHq!x`Skl><oavP3B*cTK6
zro%J{mLiW?0NoqwMawncj6>%!n*?pirU_4iS)#4n_BpgniK4HW!-{$)Ex0Sh+M3%A
zes^R*;c#ulLy2856riKiH+s6<-YLUcfGa9MZ?6q@?gsCoXaI6En1$*BPuxi+hcC|x
z0v0Cp-4~7&AL#Q7>s9^t?v}qPudL=UbrSJ?=BDHE@;uKgCe2z=+>ey;K<(o_DpRw2
z*rHUwCQWIgB0@#8M-q^R;ZXn)0<xh*MR94;1*Nw&diH?88UDwu@*1|%pYAiX9$!%j
zh4i%DN(Qtxi9?zca466}f5GjBhEEjXJoblYC~5u8;n*zm`bL4e81?iD0gZm??g@I|
zgX6zytwgh5)?h)@|5N@DMLks@zqc~zjcSl>@(1d3b3T*^VU4wxSKq!j&0Z&(WMgl}
znmxM@5K0WsqRZJo3Wl_&PrHvxp{e5a4*5P<ELcoV>d))WGD(l9`KvyY(@3E%Af6vi
zQIfBEso#3C@Sqk86&H76_?|_7!`QKMUme!RltB>#6NWqCWShx)=~<5+*0GbffuDVF
z78O*SPqfS6Jo9cXS1|z$1MpFvY@#~3ez6T>+t`ZOFQsrvhJ;=`wSZr`4~o=Wt}pa9
zHVh99QF)<AL|YO-PXKqP)ddG=M<RzwtN}D1J=;1I%CL^iFYFS~zOFCsgQvZ_3QKXD
z5Smuoa2ZlNiA@?>GNaiO3q4F=#i)0+51DJs8y5aFDJXV>OklR$3hSTJ3iSO*0|b)3
zaooij5IO?#6%s&KKF1rq129cb*>meRZ<{CkptwPI#G>i@KcW>W9QhYz-Qw)y*(ZvC
zs}bQ%dRNTB@_C`?xT81~Tav^YVmEynP7kG5)BSjQQnPXng|`4Fo=5&#T&4B0>{Bh<
z!9EKHf*#3zK?C@3{{p?Xy+J$W?%*LXvtHL)SmQx=I71^6Dah!!m}rFnE}sytd*rwa
z=2GZv$)uj3*8Dg{O4B7}xm)E?cfhSwTL>cpUcr(n;T5)zRi4-7N19;L+|7G&p~QaQ
zHAqE<w2KrrklD42&M@COvYCuEKMP{@=_|m6JnVq&M0waE^e;dqG2h-^Z<k11JfC}@
z&IY^e?>r+Y?t3Lck29eG?C!G48euc6MBXS}_gG?ML$^Fr#*!F7Q=*mqQaCW@ksjHC
z++6k+vb|yp9HB~1tH$DgR@$g#sEb!QVP5qMeu2T<40n0qrCQG-YFBIOlD@l(;B3p3
z*nhpDAZ6ed-`_d`I2>7ZcHR3CI5P_cz3#PsgHxzu7`i~10}0~9j8;N!dN#F+oYZ-3
zS(TYY>)f;W3IU7J|D@kobiVJJ3qxcNagIvRk=ZFitI`C{xWUbWJOoef5!5_{)E7LU
zGYVw8QV)wvfpmWy3g{vN?&irzL7^502O`k>27G*?fD%BvM0lu@8IyhH56VYzBsk=3
zDfGYspynT+ARPFy1!TSYmKZxks`KKNTnt=`tedOx&MXW2JEDsMzs;0do*0%RA3X+g
ztj_`tW({}=6hPhq`iiyopi%Qb&bw{S?Q^!16Y~y6P|+i|Ojk%&7*@4)+%q_tO_ys-
zLtnT#;z^AOql~%8$90$S5MMfoNkSxC|KXhj*(-~nnM*m8-eNkQIT3uv{H4)bL!yB6
zrE993JR5yz-Wo}8RF*h-jCsG5u;4l%EFgz{3vNt?ng$e-P4SiRj_Cl{!SKvKnI%0>
zOhDx#Th>qBK7rJI6(<(T8-7%mDeXN}x*v~hEX8f;H9|R`;Kvsy<6x#IJAs%@%9P(d
zFxqm*iP!A}1j!=-sFABkVwEocoQ-TQc|o}-$2{0<EnXL?649wzk3q6>_$z2)jZvQD
zx%U+Ord31VY^S^7MupKG1_jGwsL^|j6=4#g$T238h!&<OMbR{q&yp!9&8cxHb(O$m
z_D2p@nht0-CQSK4shYff!-e*j!eVd97{WFm@*6R(5#NOt<wXz?&))^flFI}b@C465
zTZ0l4ZpGJ~FX83mP1S`h6TtsN_V~K<^9sk07<{rO_mk%Pn8iD=cXh{cQFIH~2(}wQ
zcfFJdfAH}2dH$2*&Hi7yg7v>rp)5@Q>!3xnhSaYy{D1HSx6KYJK|={3j#9;B2{JD6
zi&61X_2A-#R?G5*LW&NVdOff#0)UBDCqz79HZ1RNfca~}e;VLjcWkb|zi<moRO6XW
z+?&_Th-MNeA(N;!l-F*kWl-AmDl)5m-X0%9<#xO}^&CajcD`RuT={*xe|o0idj-(b
zz`)vSVK5Scpus+Izvy=Wraq=V<Z|rmQ07dv<@4L1{$d)r%;7%mn<OTt3#XXUAV;7i
znnYOi8>d;r3D{MJ-l^XdA8Y%X9K~0h)eTS2!5quW)dGyjv}iMIGIyQh>to7VZTP$l
z<%r*kkE=s*aEogf8e)u;ciZ;{v8V-|Gm&~LY+G+R!LQ;@?X4~+ndH>v!ZIMMhzMW2
zTII~q<&ZBRwjedrEs46e8K*6RDku^S*}jm1k6|}hUm%88c&xlPD7%DIO@>)h;p$vT
zuRJ09Y%H&;!p2`0KXb~*P6>y0o}OE+EAa*}Lk_FucxDDYcph;^4BxCgJzzc_4c5!=
zYJT@lYuecNjJRY@D&wEq+zaK8r7bX~Cxx#n@EU=UO(`AE+1su=p+ziYCPtJDmKiak
zEM;YoYL3+SW=pfPDk*&ws67E(UslObB+b1!%1Wy4vUzR}UlTf85@6s@GP`6t=jyDg
zCYR#9-iLT_3^u6JY<XcxvtNJg>8Bkfc30OOVl1OK$*w&O{6&I*w1GBA{KvDX!|T_j
zprG1fGnH;nlsf2zRQ6mfPsk)CTL=_aNj$sS^Cv^)esdh0_C+%|X2H@X`U$)UM+c`^
zr&R#Dv3Tg36N-_*PO^oXd&a^=b7s}L1IC$>WpBMC%6q5zD`5tXE+CySMvs}bYX(sD
zj%yr{G&L@1+84J%2F1I+_cU;Ml}s<XgjO6J%gL<nA@utYB)A)z%iQ&A&n<m3TYSms
zq?@-O0hQp=FKbQEPNfp9!x8=PXs*lxuzG(XDXEM?+h;%*%ap-@%<6B0XJePw((<wS
z07#+tuY4$vZ9Okb!^56|N?>s6{(T3zWi8cz7HTqPI07Zy?C|nXS8MJ&rX$eM$3)A*
zZe!Z5$)Z|>ednCp7PZ=#tZ$^Dbf|xMyHi12^>-fKp~mP=T3}nr1RHITjc#bCBvAC;
z&pKSmvjOr)e@CnF<@@heU6+RXyvo^DbBQ4Spl~QBVi3Of${9?H?sRlx6Uwn{O+v~7
zU{&&N{1eg@$A}|_fRM({m6f~z{JEz}%YIj<XpB*Ts4QdHc-w#I;$*p6bdA0{|6X-X
zSXty|Zp@y+S5~kCGEZ%hRGTt5HluQE^g9R{fHI)&5aRSbU+Jl4f6=Uw6%HNSkqQUy
zNOGp{SW)9ZwF|2F7wVH~*Un~SBxRZFz_cdVlVdr<rP<X22hUFp^5=><d)Vji<@bfW
zW>bX737Rwm%2qLP<+{UCP8PV}jJ_jBQ}1%fO(IcQ)f$^MGw*@1it7PSDxQv=(9N8i
z7jt&Zh|g_97r0rm_17R?R{%6>$57@th(UUhkZ-#HI|VwZBZ_{kwrRkji+NVPCt#tD
za$F#!^Z5F&^j+UfF<kmMcv>NVZX5+saE$h{%H;O|aR3&(EA*jwjzx@I76IB6Z}voP
z0O4#)EYzndirHP<W09rUTIND@NeH;<RppO-_exTsAm%|)bnJ6=az1w<q+VQ#QgTKS
z*4`kRiF@YPxn%(E?iBu4Lzl}L#4GDtrsL4!b9feiWv7sdW-x>qP>8=z9RTzas(i(Z
z9xMj`5(msULzM9lfP@Nl4iR^tWKTie{3qMX5O!z*^{X*vI#M{EBItozbCoJO2hLmN
zLlK7?4hHvAt~uy_zHzVNm*=o{<{d!|O?yatN=3I3#o^=;ylk3>ya^BzXpzx#F@7TT
zI^QlyUCWaqq3Acpi!rFigYmXV6tFrg0gFaLhfX9VI!;~VDoE8e0io4r@B?*4<AsRK
zW@C>Npaf!zr0Zh$j14I>SU*4k60V+hQ0Fbnao08*3@qY`K}UdHMjG{OQLneDQ1j{P
zE{DPl3oyM^mgRUqxz7qF*vW$w%S==VzeU3R3*znu>PgN`rH+DZXt`gY0iA#b-%~oz
zXmK`s*aU}=_JAH=f;U~U&G-j)8-PJc3PgkyZO7ec0XdRt(88Rcj7N5otP{#yT+XoQ
z&=?BC2}L$(<I2S`q5KLO97Mlj1Ev?4(Y6|z{@`ZB9{E#~i$m=wenI7~nhQuTOivOE
zF=u^YXB&2jb3IjCi3j&h6^&Jh-VNQ7FF~M7Osu*ZL0V8*4H)$jisi}8yiA8Pv1tpK
zuyt1Jya(;w!rvJF^4AtB)s>*9T9%tUX8I0BQ>T?yc<kDjrQ^N05e?*Z{U@smE<meF
zUHQxsVV7s{Ya%6o?T6C1oX6zx;KEfN41ER1(n-1=4_E&K8uXMK#J`AM(l;B6=lM0A
zA``XnaEMlc1A##@sQJM&_h{(T3LH*Q4;^Ekbu)P)h*67bp~x~XK@?hQ6<BBTyz#=I
z<?FsY9h3o;$p}grUH=8zve^Gme=vBwicmwgEuXa$A$O{U`FB6x`S!En)`iAPbHJbS
zDF1YCh1w3vo3Cb<?SS%lyN_|9L>UehhxiF1DD-_MEp(BS@8ls{C;$j}D}pdT+D=|n
zPhsXv$^1wjl_@5x{m3Hfw#H0Hl+wx&WvuHwfH&KqWDr8&>gPTQTVtWA{eX%E98h6!
zH<qs2Fo4+Xg={mO$zWizTUc<e*yBr!mnS9OPQO@;6?v3u#^-dUjEYURS;rKwUQ4;J
zvMM_$KS67-OR-pI2Fs6c=zNh5E2vJ(OOYB5VkYbMiFmc0F^*oF@BGn~WAKL~kv_`1
zHR;?M8cSN(CKn4CFL^3xB4AzDVD(WV^DjerZg<8~T!YA&SR2DM^nI}c#1%H;X|ZX7
z+ow@)Jh|s1vC&hN5YhSd7r`nBSpBcD$<j4x=+7uyRtS9CyvN5bR1v=l#TIqu9a<c`
z*d0|KM8ja;=8L9#=;``^1fdW4=<G#;OJrQg7_NZls)`~S2$IUw8!fpGo^gCH3G9w|
zg#I?9((}j<xiMn)L>DHR3(=Mnxcc^(uY{7iku_QcIoWWURUny}IwQTR{fiRxlTBqc
zb?_wzGt2>RS(5=_wikZy)(M}N9|P^|>jLu1+nUi|lzFNI-H*ZANuV21;|;s(5BS$H
zW#4~nhgkmu3CYCB#{R!`!(^$C#s9`Dx<06#m`f~0kU`iFk2&fZR283<-*yiH;{FA}
zpa)Nkj<1yX+sG4$u_ld6AV9F-VQtmmBUr;_m_O_-wZPYGcgnr~2_lMYm>TdIXpl)a
zHb^YY2(^45)u-wHc-SjpztJG+G_XerrHt(h!+p9AZRz3n`nj{gE>VyVa@}Bw4A|lS
ze%?FS(5rjc(DU(pc_24BR}qQqai>o;bGWdwIpCoY2}$OEte-E(os*=za40`2?LbqY
ztrg3oU}C_vPTdk48~_62VbMPUUpw`<)TGriqJs0eT31LTgWn$OzOHTQF%r%F&Sqf|
z$pyn<z(e)nb2E3Fy{dEl{uE36qqO1TP5k#OwtSa=AkW#gmyo<es3n10$94*jjV21C
zCtnr>dO@AQl2=z>QxW8&jP|Pfs;2=jc2%fzKq86QCCZdoLN+xQn9Svh;aRbzyuUi(
z(xl+e68v{2(drz^6Ul`{3<nQ<dD+TUpCKrG<l;bjB!elXp^992&o1MVWcG9G9Pl$t
z0hge<-Ecr?fzaY>QJf3Xz*1hp-z6#4T4x~aRniQlD^tZCdrNBVe5~ua)~D2`=1{%~
zA;h7u?lUiWZW9-)3?C7U$kh(lzpOe%=YShdG-R^ef#=i)R{*XxTTzk&7N=VxBapn{
zOMUi_ny`B^^yH$cj6rAf2EEo})aGpD;D#)4yywa5DETxPl1o(Sys(W@XkJ(*AM2P^
z@H+5X7Ea-~N!uAF&rUX1r6(|r>d^4iGySWF)qq+TI`o8O7@6(7pQAW@*Y1naf1N?_
zz1X6w7oxdo?t9Kr-rBFUQ@F~S9jZlxj@0$E8E7x-a<kHLF9%3<_XT4)Z^7q+Wv)mb
z6p!!{^8Q|MqiJvA<A{zWb6zw@lilm5x}0BxBRK=~z`p9xlb=94JkYou!$#5saKJu<
zStAYi(-4o}LJrT4^uMKdmTc5#Q=R<sn`i8WN-E1a&*xkk(!mbr;r<~`-C}Y3#GAAL
z7Z<sKiH)cYV%GAZ^BlUzgi(@32(WDhk!K;d04+Ix5jRs2h48HE)P@2aiU4(Chkj6X
z&gK@hPg7Ql9nqS$IdmLD>h^-wgY<bwTPk`Y;)QS&&vBYPvwq1Ako-kYS<-m+G&6B^
zXFTm{6$q_{GHYC`2dbwVj(B^clLH#vN$V@u%?`yrLUKWMznk$l^+U4H;8!e4n;4aa
zONPs}XzgzdlF=9i(=jSTpnC=Ib>zoR<%IgSq~b*K{SM1tt!JNfRyp-rJhE?UrxcGU
zHdIaR)}7F)3Nd2@^8~~)q*VAwA`!*O>|-uqdIcp$3k`{Pkub+J+X^2ol#U-n$>^_5
z&=bi6F3*xeV)!cvInr?a5H<bX6}|^%vzr@$uKWo|dP2E=$iw&)Ey&5HAoYd*NpJ;@
zr9ye261fm{UzPh0xJf|(!FUr4kQPsj=CDki=r*W->ehag(6!%**Imu^s30zYvt`d%
z;Mbmdol88LHBc%;bjRXqM=SBsHX6l6G{pD-7m_ow>{@%w`@9AuW4$)t3yjjYO_@pa
z>KQM;_$Qh$Stx#>C1kv^juDnvqcjvwi@WJKTpfc|Bv{j1g;cqL%8Sn30Gf!k&q4`Z
zdzf=%F_^0dr{}HymvW)l-oh#BWPtMli)$KZqmFW1*2`?5kvh~sqsKuJnV$l`IC?@V
zGg1bzpsj839t<Fv@GsCxNBV8&qJaleRF|g8{SYX|&#o?}w^%}MRt5OGVa*)r*jQX?
zdG$gO=|`o!5^%~zl`$JE)DmYn21-l~%4j#T{d!t_Bh#GwIe%z*jl``3zpma*El^av
z&seT)$gy^lp6rhLEDEF-S%*&>Zrj0%U5@Mom|rPCU`OZ;847V0gg`?_0=uC#`gxyk
zG6optw6y{Qj5<h`4z0OjTb|k;Cv=TDL}OX<w0J+XXssf8izQn-t|+Xu_&$xUMbtKV
zA@~+31G72MTt?h?mhWJQZ>mTxX0>Ka3$G#l`k}7C1Fs;JIRT}VfUAOcD!!l)4;=-1
zTN4H|W=px_Rgu{%)+9A-AS^*zCr4BWr3EYlp?C0&Q6o^?6k%*~Gt;Ec6$)rnlZZFY
zFSlt3NfUZoo7N$)6H^=Ak<2FEwYxp(*wF>MOj@UjYcxHGgNoLi1!BYp&pfvt9SyjH
z1lPob<ve-w4a_wt<;ZmEm}X*tBwvy^G-=-X&NTBmp3LZxuU^_1VW~?KwDR0@g(kM#
ztC0HTA%7wW9EqtXj3UN&E>@MUO*mvYULd!AZwUzg<GTk=7gwd8t2}6$91;uk&Q5rl
zLHteRU%>E3c3hK@yY^w7!Lw;{)h@!u26qZ<De_Etbr+uJhE2}mQCnG>il)3)0qkji
zVjVPAx=KQa+_qb^fxI@ReNyIdH6W&Ld=xujlPbZrq8O(k3h?_qjYR~B45O|y$TYx2
zAf(jc4h1wHF<@UyL$g-MvO>13YFB_nmn$`b-Odpad}LcdRl#yrP8n<q{qk=1u?L?A
zPUfyF;r?CI-1T5mOv%)2nd=N_MaO(Jw*g**Ok7AAwV_%U{qa=C3#FhOlW!lZZ?%e3
z6YIrv3Nk0`xrdRv%+AS5^w7a?9Uc)jiul4qun2$_kAFe?`<aJ=J=}{fwj<tfb-8~J
zwWk!A4d=xB98_Q_gb&V_mi)Dsj((784CEGewr6a@^t{R2aLGC^)cHIt3GhfvwPi)(
zG$r2|+)4RFJ)ha^5lVB4426x`tpp_uKbM94gr!fq^a(j43Yza+$DyM{bSF6evy$yi
z*t1+ws6%}$ys>qU`h@X>0$*B5<CB)I!NvR2m8nk~F~sxJZTW&%+?7v{uiF#7QU#mS
zf9M#t|4OoOGBEvbbxhU&z%u_&H{&hEZ<Vb8lGh_zi$t;YyfKoihq#BI_~>vDkxE>B
zzV8mGiy%;Ax&#w1;>=K}w+}(&H8EOOdwW-h$8&rGRrx-Z-2JHyc9A<tJ&$Qx*me5p
zL3vc<s|T$^v&YN#RmIE9j&==lpK43Z&iChs+=}nc7Z}hu5tX1oHWpx}nSH#&Kiiic
z_m7Fll~y{bF30%9ArqoarfH>y(+EVS7w_l#eFCQHAo0m&=k)N@a2`!^QSGE8@cA7)
zZ*9k6fQP&u{!v`R?MCk}_kJ3p>3tz(&VcCy9toUuot@HWAagwCh7eI6NzqJTmOc&$
z%KQ#os}4_jlB|;MKjx8sv>K>3Zy^bih-OI;Em!a>o!V_3Do*TH%fCxN1Dx6w+p`XQ
z=bbS$&)ECiqhgiS-v*HJKo)58m1f(G??NmjA2tYLFsPZWrw0n^3J%~|v{{C6g)3!t
zxLsPxnMyi=i#4ajIGXvSYdg2Ww?r2Gkxf#4<#p(JB#h)MQH%P&cAJ$os!ql$u^ir9
zr}oNOqBFdQqKFTLLWd3lH>(o^VYC%YufP09j+l=F3^~bLl5i%v1tiigvQ$WTeIZx!
znc_GECX6+TPquSpBMCEB*avd&sXgsW4B;5?vJQK@3`=($bMo8_ClDQ(zwOy1Wz9M>
z^T4oS>WLDGf~k=t!k6Tcv)b(=H@{^U30T?ybECyA2zj9UN>8E%<bpYueX5dh(EhRY
zORyY}<%pjd^g@(UlHxdLIFc76)I1~o-A_4{BnQ2I5BTO>KmNyV+kb2Scn()9fn0%-
z^2=}7u=7noaVU2vopNXGhc9m-ixC7nAgX0<o<HXPEg;cd)vr?h`eWl%F~SVzV;I1S
zPz>J1nch2~a0`ze!Z2Y^(#rp6CdQ5%r3k$fvYAU)neJQ{1)Mz0i41{(6KUn@A&)yW
zm}%<L-Z3J$McaSyJfnthUOx9l$I?;a?h7yR>;x2O49y=$|G*zOm@{mV-a4rFT@(JJ
z>kNgIHzxw8_{}+Pvse(q;gg;(49<X;d2FGLy4HdkNOLUOdH@c8@&`;gQb1LmAomOh
zW3!)&-D{5~LHN?Cpv80`*+G(hKmJb8rMk$@<c3X!!Z8(yD~*>Lm1o*9=pOu1;Dl1r
zDZ4+|R^c^;?O?medZ#a>m?Y?l1B7XOiNN&(jTD6}b|MRl_D^bHqoXq1LBF;IYKg`*
zIS!+m;=!~s)*`wJz8E^5_ey-vI4z!0`l}!bT_7Btb-P<S9XEO3<iOxwLT_&FKd<hK
zs_kPD5Y5#atuRxy3WM=v>euTJz}?gD-T6dTQ^4WMBxE8ntEKq)(~x3Ih_2z8+U>Z8
za5PC8q*n$4JxsA)|0f}x_5eH(YI)IY^tp85WAMOLgwI40L(KUAOi2fhynn+BPl-v|
z-S+wx(c8FMoM~|9@R*hx_??-;M2idoof)@IqZ~Zp{_|eAcv;+geE`bQ)mdeqO5D!r
zh~78;x|b)jrlN_5K%-6pfBcm7GNg!6@VLNc)XwAjJJgL0N+a~!um|u#+yuAqG9<uC
z=NOQqx2K{9E(zuz0bC1k1NZ1&2-h55MX;d%u#A<iT{Fh9F5m6H|F!AQe`l)-b!D}D
zxwa&Bg?|GoIO1Q-^!WlI(BA@Z$+*}_)KC5+8hP_E@C$$szkYC<e_oCMyFK23)p)5b
zQ}yJR3HaC2mqX8ST8-c36q&TS>+k7I`4YVM4o4vCIimjJID0PPX9}^UOT3gQkYrt%
z*(vHxXCv(7R?t2i5c<qm@}99+FsWE?p3_ybpc61C7{8S`%o`Vru-?iw=pJ1xC6T{S
ztIK2cJsx)_KzU*uIB~MaAXl$HSKRrl<qqD>%IH5NS4h8tTaCm@zNmPzBl}AO2g5Q`
zW+AUd*!+V8@Y`PCR;g=JGS|?S``cq2b)-1|Xp_FAgdLo}gV4fLg4tkJHuf%|yyM?0
z>Z<y}qZ$Sv7jmGRIWdsRP~K)B#4t{Km$ka{3fU^d{0K~#XUB$YMG7bO4cI)+%(=(2
zdzA8gFZ&-CSUjTGiiU2?ar23cIgy2gfDPVSQ#QBmk>_Pp`{=E`=KWqpjOn-n)yC=D
zk+V*BW7ZRnj+899ME#CHcSL4DyyL<Z)_#drg11m&!*=sLbU1R<WN5z|>v+0yOoT30
z$z$Gj16@AJt$F@{`C#^_17H*^G}Z8PEzS5NMzc@ID%RolQADy1!MfDKL{aVdC_S-4
zqHCSz3-VHiavElqn<&kMiX?G}Y@RO?pUxlrl<h>mLn}ea1|}~UAt!N7@MqUXx~v+7
z)F1XirV`}h`u%B$@(546^-vv)R!Uu0sg-U*YlMdcK<SsmD_rf*Gkpw0>X|;cM?yJR
zXzu>VfRu3fts~b60hu1@kkIc{wmm(dpOxVvvjf*RQKb1pn#hR=uLI8d+s1vAF%^t4
z20Rg_$1ky_Gr1UBu9&L<)P}-T6R;kmNudOQt+JGFo$l6-cl0y=M)Cg|->m<45S5ML
z|IOqO{Qrk?s+7iTHW+@h0@Q95Aj&MRaaUuI`zSj9lu<n(J#uN+7^$RUgj@yZeCaI|
zn~j}hJP=t2(ni=Pg!MrYTBD{T%$bZ(C`v&PbJc;U0L*}hh6NP@GBAviy(n~LO)<&@
zF$6&zBxy$Z&Oe79+gnEy675i=QVX?nU%7hPCrfNQA1<Gk-YQ(-sD$?J5>giMm`O1c
zCEX6|H5c2ZYNIOViL1(s&l}ICI<jmg*S(&W$+xPWI;*d{Fqu9&JapPcPR#*e;tpZ~
zz!HW18TBOv$4DvIp7$(t$SFyUA<iZl2_?-d2d4Z|nStw^bKG|KqpVwO;Np0DRb7XF
znxuTwHxPENCtb{V>_L)n5X_QWCkf0Zk$rP0(BRhvhAfb)%(hyYBQwsoAG_VR==egS
zj1DRdzS^JfLQQ#^*o5sDu<)5PBUodHLIt)GyDxYh0o8{I^RgS;wu0llZ8FG*f2Id^
zg&Hn36{7YvHr!pGL~Wa$pC;*1l`;2lGPAd!KF_Ioy!4kv5w}D`g(yT^TuYrkM0yR8
zIr&jHho$tviJ5`gp=b(P;7Ktb0|<(j<r2T_&J+G*^FUxp!4)ng<qFP4QH!TOMqo)x
z60m45)N#}e@#X7F(-6-=e`}e)=V1DyXz3u+T*CXKFvIJUb9umTUap;<0H$t~5IB1a
ztOCdKyk+N2hB%wb+pQsIN_)At(|kzEV9SzIy&Q->ZTudrTe@OE#eX^8Ve~~+^iX+!
zkIpYyoOxP*$Ko+SVkRFDb>A#yw!aU`+2D1Ylb5J$!Yef<M*YXL%Kje*pNy=`9RFLX
zIi)TY`#)0ire;>*kE%8wDPkMz6&S^=aLY9%CDeNSijdV<+@Tek;E!i0&U|uXGdq1_
zTk0?CX29K>xp>Z7q+mtOiRqiAr@d{7#t?H)dd$2#LIi6}8@Z^#Tvx=_P{!1E%9nRD
zY?~xX67Od%=_=~T{@(E|Sy4A-h-NQhxaGrRO-*KNO^dxnQ6&8|u_T#=I1v~Ti*ZD>
zh-Aut0<HCPE&~g~n2Rk#=3%yvr{oSV)v0xpodYd9OHU_7CDJSdTuvXAgrGrWo+Z0R
zil>f2r)iLR%V;YjmZOFTJg}tQvSULz8IlKb8$luefuM;9(=0AAK1@3I5MMxK7c3<$
zA_Fz%YyFWUkQ{}hViOayr(aI#<4KKP*RL=D>fhoR(BFxswI`EE{H<#3d;uTVEuP)O
zTV**Un*0MK{s@OQ#%U1~`zySDuy*uJ`Jy{e!6LiKGz3y5mQgrna_*>QX(L~O7XuOM
zp{i($L_5Kt-9FD%$e%nhehHqJaxqe%grQ=MZqH@!pktAHoE*E`VZ+mCk3xc1mJS_@
zN>!EXoFLQCKFT#KuoY7pKGK}$`@KeO)>0!ms{=|uVrgYM2T6<s3fmhmr0w<I%AvDq
zM9XWf2HexUhjU`C14+?^x~|x1>#eTuajl-onr%H?67(P@3=C4mQuDU1`U@fzJHA*&
z2G-S5+JdE}8$D{{QHdZ(^%E?g#9pa_B<`FV_3C__P9ju3y>R7zi5vS010bwcN?^?q
zEy;hA!A%_xV;Aa0y%OqY6vBXis9j^=3M+}TM2ty-1tc-Z`!!eMJKEAnDPK7VLC6Kk
zKKn;mw&|}TA~`lSRZcCp;s%RcNoayg<X(RcavnPQO2!L~go4keM4x#`+m?3(qf4?G
zl@H{@Y#kd)B!bI~5*<dVq~9sJcvGV%Xz<7a;8GZr+Pm^3uEv*n)+Kx<IcwFI_O^*-
z8Qf9+0Wl$lswq}5c*&;jv<*_4<L$6r1<G9XIu3HB9b@~k#Ct}0V8IJa6}=nX=ZAF?
z9?(DY%AOC)-JX7U9|;~k*a?8(?55cFGBa=7OCJe5AqFY!4w1GtJR^i)z^~hwxp(d)
zZbPtfb{s8@#`hML-*mYLKn}!#I6L(0Bs<5N#hf(w59<Rja8$Yww3M*x^H$;YQt%LS
ziew^l9s2w^fG@6^1UmGA+`%E}p|Kxua2uw+tgD8Z35;dH3*)QfvC6m9ujd8AXb^zM
z)^dzkEvEWR17ZmTm597$x#unp8&8eR9S+obn-kDb(fvWtA8sVk$d>Dik{~j9*9%;u
z$O-uFEdfkN36aqfl3=RK22Lz?mqPuz<ni@ibn*QHHQr2eENnmb`71#8e6Ka>#U&Se
z5N1Ld#|8a#kvHGeH<DKIJBLMhz1B7r^=&YzdELO3P{uu3$>7rY;$p$(Yn6p{#z$K4
z%6+1>KSGBmSY)ZaFmkT_E^sfSM|4L0OO8+i1dq5)F{B*?|4Jg3lb1KF-5`~~jFSMw
z(VG;Z`umuAS;tc_gfnZkfhMMh=DU@a+dQL`PHQ2(uSVgGw_kmuBtrD7aM)jeVMQd-
z;5gWL2h6v=wq@@Y+X7lYEXw?IGSlRDWK(-wg@=I(mzZN7Tq}TqoB(ia3zC3~8<$UJ
zEM=R`x4}|Uhw-egFx=ImlA9ywxnlPQ)xaJ#zfQN;a*gt!QOAE)zYPzLMY~(~x@Es}
zX1HU3%xKEFYt`I?!toePIv4_C^C$is9Nie62C>D!h&00?z0T#Myf9hAUHrgtlii%k
zAP(9k7#M1n0DpXtdGD>hvHZO0i@%qua0Kh6$lccNHgS)LH!4wQZDtxk8m#U}WZHG0
z`hyldAuwI^$9ML1k)`nFH^>SDl0sqv4$m%n_u}&+Owg*V@5l*i2WgP;wey-4Xs)aq
zV!L?wB4qcl1>j}S_pzYn8VhsFjQe6|y>_$lZU}$Ibt*i~w))W3Yp_QZP7%i$N5j8$
z<dcv#!xD^19s$fxDW2}(6IC|H-An%i3Z#w7@qdpL4z~Yq9!lBW&V+zo&d^fH$r_4Y
zmVl9g;eP~>gQF7x2M0S8y|jsqnUgsIBm4ip1Zzuk#}=C%#^*&ZZ{$goTryt-NYKAg
zXB`Mzo&~p-_Eb2zUAS)f8VCLVVeFlPH0!!x-LkrD+qPX@c9*+s+cw{_ZQHhO+qPY&
z&yF}(`-?dLideTR*4@mRBgd1c-yhzcoK-U^>nRlvpy0@<6iUcXcO)_Kv|8GBe?}Fl
zUu&W*oci~P_Zo~+3U?b?X^Tms%l5grOg>Q6P`Ju4Ns_}+Eky((g@8e&rlKh|<b(Be
z)p!T>0;X1)!wNQ^4WOvytjhbzNmmEVg}DR~MSp>XlcQpYNFXv>ktv1D#j$_|=Q2vJ
z4khLm91?F7rfbmxJMLYm$IJZABWYHxg3kgH45>G{2#!;cQAmNJ!yYYyLIhJS4jZDP
zmB5wJgHkAjIf{47PldQJ^a3{?J5LOdP6rN8ouv=ff@k8);0y%;E<QA)OVpMNx(3S{
z1Ow*QxcEJn{tGo}@yLRtK&0AW{J4zLOcGu_UGNrba&a`8#BWiZ5XdKL$(|TxcV*Bx
zpGmlvnibqlof@nWH9VyK045X7&%qc+gQ&Dg&nQ>~^^*$Naz0U;Xte^UVXzzp92f@u
z26nA~41yIHmiaeKyATwjRT0(!)aBA-wR(d=92=(b?zNQB78b}z_%R4DzG}3dp%4Rn
zwMlP<o(P?wFq0Wwq%%`ulvxD0a@8@&EGro13@mV<ITJf%u707+Ef+A9BswmT@}Hr0
z#$+Q+LgC-5CJg8l5Fi1dz(6<Zq*#1Q=vb4m8L8?NkdO(-OTGjp)m)iV#}kr{pJz43
zAD<H)YXS+!b6$5Q?VQkIQ@8Ez%e=PuISL<)<~njS$=9TR?lkBgy#n?m0cuTS-gGck
zBCdA3>4q!3Sv~0;$KfAmxL?~-rrjP92%7_;D4%I~@|)1;eto*_$1eikPRBC5GyGk?
zKkpgb;J`>;v(?#__%k~*7$XGoGsX1Ur9GV5^d*2a07i^}uMR(-PBwi>^m>S-BWupT
zDO^l1?B6Hud_M&kk2f&WliMSn`bNTk4S}<JODKe6O-cHH9t@ZSc(Ot0SMLr-*=mY&
zQm!Y;>3MJWA8*dULS~W|*6wlntwNbIXHIg^O&1>6x0+;o<aRoFyjg!P|IQfwiuw!m
z3|Ihv4}iQPUJ(N2aRJ$wtD8l&1=}I1)n;7L|Atwyixmg0@YcI9(=(>2<ro=+r`6ou
z0M%oA0;*s|uf=v{sVfLXG`(-yMn-9=yZH!gU*=Ebg&i;Fqnz;G)xoH}36(HFAci(1
zLSCP0j)vd9q~$koZpnFR?ro1@Oy++%^w{huMV@A=NJ2507)~^-fJ!XF4_1pFVo{L5
zR*+Rs!s0z&rI`pbwLN(tGs*B+t0XRjwR=y8H)N;Tj)>8?=Jpo%i(&qNL63aucm|ac
zhfJK8J<j|<qWt^rcQogBXT$8`WObnI9rfmt$_cv~r(gl)O=--U=)tjh(O-|0yVl&~
zQN;~wNa$_w%li#aOs0Pa(rFuP{rOA8Z&`ueZ&|vc#jd#`5Svc#K>TsQdVtl8V3iwN
z9Sf-Pax+A`C?(;2+|#+2J0moC)N`~)s_QF+$&&OC^`=<jTU7@SF5*#$Y6axT^2r3!
z^`mlIp<1P!K2J;fc?}<A&EM#r+Ftyv80RPhkKr0N-Z^XHp6*TklztluPwj>%WyFU*
z#=f$G<Z59FYz3XCa8ie|12ew`4hQ_s-8AoCW_!uL7eHRD|7It`kRL`Z@muprb&(2e
ziD#Wx_pA9FxlGe3vPdPLzskw#udIH{2{*SUZ%bZ^<ulvX5JV^F8{7>u>yrzXizd@3
zo}|b1?uRRCBSNf$?TtkX;|j1j9UA%8g(gEqj7y*T*v62G!E@#ieGsqZcG53%7L4O9
zaTRU6!`5<%a^sF|>JjsM7ZF!Op`|G!P~BwDRsCZh?ZgT7ki)NK(M!;0<HR8hla3}a
z4Y2;iUFki%6vLvC7+_;Xamcj9F-_r(B8qdz_z$P;+J8plR}|QSx{cjxe^kXNqTjD7
z6}dIjT7Ll74jD4R?K|DB7#n@%>Qa0K7cZyTh%LM(H|xlhEkn-6kNm{Y+2!Yz*WKRc
z&&S?WAiXH(_BQZXvfIu*_Eol1|4vGp#eur<m!DDNi{;J-?=$#B$4p`e_vmzKl1MDU
z9%jv>wD@wJLNaiXA(lF~%_&)xj&RqnWrU7C{42lh7+|N}f3I^(7yRGu0)xAdnW*=d
zMef_|cjUq*r=IrJ6~KiC@5MRSWz%oMC#Y)5eH7xO`x6!_^*M4)X|Sv8%*!I>a5O5K
zeKNAE?X^i5Dg?FzUnjG$mjC-;<l_9VyB5d)Ub*z|uJ!-f`Vz8q*o3c{ne&6!2Z$}r
z4pQ7$5kaJlsAaeMnrDwauKwD8Na>EwCJzU}Yt?C=@2&zS|7uRYA6=ddtqmnnRIY}D
za@3;JNP{KcfWtp{S;lGDT1^M8=>m2bsyP9WR=j$0r$dB;c3rPeZv=pejJ3_DG(4V$
zHUh?vr>mo>&!eLy+ZuY^{qdy>H3sw^?Vk1};u%b;zD!Oh@11m?iY8}^v*Um40{ZFg
z9DB}WjZNf!{Y*`rLmnc8Dksn46VY8zU!(BRyH*uROXug4mGjk|*k7rxeLB$$b_-_5
zAzFR<>>wnN`48YPnS?NN<4H32aYvOL4}MHMWA(0A*K`k9)jq@K1PLzR;D08gmEx1-
z$*F=T?tBF_u`!wNAwv!kEB)C{s!3W`*6PhG44jrp9zx}I#kIpISPI7EUp1n|&RFP1
z(>=M3^6!Lx;+A)i1rl(N;2k><pR))<5aESK9Epru?Dh>voZG=&*befdURV!IdJ@>;
zkEDD>Iyw_nbjZ{GF-4cq%?rl!hmR@;&+@IpVn?yo;pphjY9Zv|uDI8v1c8W%KiYGL
z8m%5CS4(HQdhfk<_F{d-&SbicCq;e2`U1K&iW|8x(Uc&E))65g(`W^|nF`%^zwpc3
zapr;9eAJAyQ(m50UeC{Nq59DTPeTVu<Oib>m2s5kc=2$vke`Pz9fW7^Ljz+CbBVff
z%z(rXqZCC3s|SSV^27`rv;@T}45Ey(N-uH8keH!7GcK8y1gGyr+(xq~@d@(n5A6e@
zT|}8`KMt&|>F$(4u=E8X%JNGV%T52F)s;EN`K6ff){_Z^*~x2w^8#D<o*veZzU0iR
zgT)QxhqJLLKul|mm1p;7LmA>pg;=+Vzvg)|;jmp1W?+rFM1nyeAqxeyV-l$u2jkuo
ztzA2~^CJcMiA4t$5@wPRnm)^^YA&%de-IJM-vY@H5cqehQ=Bg7WfSyv&H^G2isxiv
z+-Gf~80Fn&3EC6WFfdZPsArBm<}1C(t0NHuqELQ}Xp+3#hy}f8IbSay!0VV!vxHY}
z3U*vA(m^JoL?|MNC;Kb9$q>6ufr`YC?I=W~(nJX!q~$l+q3>1%jl(6i45pXrrONr>
zSryW&q}REu+8*87ZV*_6kD)LPo%h1`L$_O1=A%+M%+{P1hJ-)viIO_t%lIr7dAD-0
z5P_(Pcn6Q7c$YPuq0fb@^kO4O;HcA@%DmKCHb_=EXE%f4Wo0;O;;Y7AO2f@Xy6iWg
zx&4JoO3y<2?V;Ifbajx+(W11}B$ZUz_IAF+if~953Qy;3Bum|DyI9IL6oxq=8}GEo
zrPKrqMD%{~#jr%FOcSc>lDcmjUhW~sL@`5V`WtXUW;z~V7V*zK+=*F8M!7CusiMpB
zW#;o(Ik)yDz1V7VfZf2Yh((Hq@5ofgvr|ZKlDX4Zo|s*U!8yn6DI0~b{mjx%gldyr
zybr2S#7*eYzH-hrX3jR;s5A(bQv%+!%bb3Pez!vPTfEo+Gxu@8!T+i49s|k!vAKXK
z)A}M&l>2jOU(Ra6d|0pI&Z0)Dq4}-eeMiaM0~Waq1{bN|_eTCU_U@^523M1kutl#%
z-lZ>ElC!sZ1uNfHp*%n871Q}{v!)<+<JKURVO}xK&7Gf04ut?ZVwh;SE+R&BP&Cq3
zcczsFo1cFWFM48<)(lrP&Z<tnHF_rqY-UBOb7GcxT0VPm15IJjcOVGNn=5!noDl$4
zsu*jATR`Yf2HNH^UC1Z%2+y^v|42FWs4T<_^=T5Z9rRZ*)IwhTAVcIK1Aq@l+J`5v
zDY;YLG6*DYC1@?*`aWlqz6ljAgF{Lg??0}*rEIZ=EY&;N+h}67-bWcEG3V^uFKoKm
zJp5A&_EO-jEgW~gNpL!CXf~cNM6!wHh&gIjo5V$+N+GemThV`E1OB_A@9N4Ukhf`k
zR6aiaxRJ>Xo`tQv(h}rN0*$I&XGg~Hh8015_c!p7&<^|wnLq1Uw@fm%8eG$kWP0-k
z`${P{s50k`dJ>m^i1vUsS@|`S?(@*Ip@y)X2gwC<&&7+&(CGn^%i)oItJSHHB57(n
z{`WYNfmU=#(gd0xo4ZD3)Sdv^7NKOx@}$+mGT_QVGhL9En1jvNMh~G6VHpn-XAnH(
zFB^+ZizEgKqq9J%en0OH{7sW2H~1zm)TB2;6a68(TaDr^>^upmsk(HlITAI<K(PQC
zy37R<F-X6iLZFBVHNd8XSS~6%dr-!Ghq@idrVt7tjjwX3wzkmr{<KH!5{$Sw#X4`>
zb(F`aQjW;80;WlYz=+TN6z$4|ii-8FMu*_=wAkad{<0~DRd5Rf-}DF6scvWPzs}SD
zs4!$^X8fOlk5gD1HXCBkTe?Dex@NuElUbzb-lGr~gaBe|r5hzz#mK_A--#H(g_o^>
zub1pJmPE<)_)Avn2V!^x)9K6)S=RO6|B%XeF&h9ijz|+lG3F^}r;g~L=)}b!DYUKX
z{G9zrC<pf{)~KD%@8=EIn;jh;N6~`}-yUDbkL!*X0iExcXWbf`Ld^Ww2!*0Oh#3F6
zLF@M1z4_Dj9lI%%coXcFuCfYt-8G0iS4{xGSsV4qTgMvYvwK+7+cYDeou&ik<9Vrm
z0vK{zv_6OiO*UtWocsA~U}J{&X=Vcug8{2vfCLC4OstbAj$lL_ukW<8%R{@AC#epX
z%O4iI01)t}3=@h<r2=uz?21G4vIcmaF8!KCbt(|8(HRey0F&yfE5@`uDY2BzTD|yD
z5DjFU(FHHPFV7e-s@Cxv%;6MU0HHMS_2p;B)@IR58e^C~XHsT3c?z2qtNg}iNoyr1
zNqk8z51a1&(?nioFlP<zwPj=d%ZDNe_Z6kH+)bsRU?5z>Dv~Dt51W|U(w2-HW_EnM
z1Q%Bmo2&mn!u?If=K;Upc<1<pcR0g~rsiI95fu!^zaPD6ZoTj8rxP4^0nGys3y04x
zCn+N}BN?)kbsU>_YhbWmYLa+fbv6=fTp<3dREzC0lvQrBnurMfd)c<9pVGK|iKzSS
z<Fb92A|R#MLHOZf2sotk?SYPe<I6!hHp^ekpCk`D1)J-A+pv0K{fz1N6dQk_ATE2_
zkak$vMEfS%GfKDehlBlUT+3H9Vj<obN;)I!!1406jA6UrcXi3WN3;(8zm=-_i>ZH=
zsfl41<20Eg7j|B9Bwf^4;Og;Fb>K7H`d++b%(4`$ixdp;iKJEziUhE7n_zpo3uCg|
zsJNlq<5&`KK^-3a3%HSL<P2oN6tS~QfYf6w66!3*t&H)p39y-7Yzo$Q3t`y#F2mKL
zk1|JQi0|aRJV`W0D7ot;&re)PKVgUEk6olQJ4s4D&T^#*TxwL!Ri{i(MnrOi(RDAS
zG@F>~duL^GVo9OO&x%T!-w=|&?-$NI^wSxVss+4FgowI1gFyq+4L+V`sK7JL&7{B(
zG+vtkN8$4PB>{mDCvgV}{@%`{4Di=ef8s&xoUj{OI@Xn6FB3RA&2owoVvCR*azKN9
z3glC7VXsv&T=_+T51sr_Pz%9zk6Z1y`DeluA`4?LM=&qvFE<30wx^5)h?JoCplxa@
z_JIS43jDugK00G?@XzPLf)8PYYHOj^;ZQoIGqpO>p<>ePrd|KYyO1!ayn1e0#r|w@
zO#UpC(t52?3b(rppO%W^Xs^bnfy{F$!&LXq=IL-i4!s}nRmmG8z(|icSLPz_FguqT
z0z+zTYZc#h-zNI6wf>z5%u+FA_D|7g_LqHee)#2c>OosYW&LV~k9Y{c|8Zja2g1Vf
zcwJ{bh?2w9;N2a)iwzeayN9&^u&z!QHxn`{z(DHMz8ct@%c3Q3EjzWgx4de3xcUzj
zEDogD>jOqE@MTvQ_%8d$eb(2EL!0mb7znSI#;k0R@N2)%+hEdjCASvy@>+5tS*j}*
zuDtC`T0q!hYlsaU<oeoj9`ASWaU*XT<kGYpZNGh0Lobw$i*&zl)?WRtmEk}cj)GXe
z)Ki4Qm7{Dt`^|+Lxf4}_5J~rb_!)!|k+39D*yLtxR!*j2qNkW|*KI*`V4nR1c3bdo
zpzc?XDxm9MC*+yX`OHP4GW4%@$sH61xj{giNF-*Jm}9w7zwhm%U(T0~6reD(wR7!k
z>GNMp5J(%8G6gC<fEj=S|1h&+-e?g{{<c!B;C1T88i|FR75Wp9>sY>`l*vdx(mKPD
z1RWGiC2E>Jj=CQ?JuIF*H>P;DQ9s9|2Bv`Zvzn#@C8}8f<kIbM&uuB+;)$hSP2z^d
z1>(VbN~qJ@LPLg{1@-FgZA=Wc;FbI|+an7bB;;_!dq|bDnCozfJzAni?CSnvH<o+Y
z#?5Dh=ob+6Z|iFmC~5ETo8=hDJXB<dhx?wkt`@8x-ps~n{z$+4gSua3)(olIAp{l2
zPaUhzKtnBSjAo-%C=&EBqYm2@+|4lH^ZZHji5xNE$?-_^Nl-%)<f(NKg={-H_`JKO
zwYDCifX0p3{fxwY4eK6rRUv)+&Uj?%Q;=X{(-+;sYdU5Lmu*APs1$RH(aO>#Sx@q7
zUH)hH!TU8#G=oHdk_<1&wxqCPK$!|P(p8~xY3eeEiF7V>(tYqT_2u=>?q35CrT;@>
z@qj<{H0c8~gNzdIW0H31XKbFQmr&uA5%%Bo>ai4;W^unu3W<DWGqirE-H&3OR8bwy
zl`I_SBmLdBaZSOlRe@;1jcF$r#zVe1)l8Z{*y&jrec<7K{1wk#c(gRi?3Gqc_c?7H
zKZSR|G_i-`7@t~~suzw-1d-A8cWpZ2;{9C8HCm+d)zmZEWWX{Mr0wbXr5vM|<L#$!
zehj5j>WRBKlmUTa@84^Are}hkg6VP%aXA&cC(I$s?!kyV!hI~Q!-6+JxaZpAf{?xP
zyAREKw`eBj_Tlifh8(T*)}|SsY8~n9JT{ZG@v{c-a5Sr0w)&yR#bX>B)vnP=e382%
ze8)_NM8_%W4AIt!UbJeyINQ#AugmKqO6;`sIdkT|weSZHA#Mz)J2TU+TY;Qel<}lv
zN;eA3pEHE}Mb=u}Lb(-i1Fi=h`Tz!fk6q~gVI?pJqfPpYEWCrw7Lx-KYNZ8B&l2U+
zMC%>X63~hT5bkBf7CHz*_hId(4}r`Wxd2E!Bn4j;>O6cYvCv+&XXD%%TB2?6`Z~On
zuDvi!Z&*To3r1My`G3f*kA5dieb19$25>)Q;fxCRyF8$A?uxu=bs`88$~kEi!xt4K
zQ2WL|0yBLa*}rc*j73(uDbJ4>=&)A4_Mp3-Y;TvxKL11UaQ#Pr1T*VD0NMX}{{NE&
zNg5iC8=T1hcoVZ5^D5{NkV42!;wmNvGk3dbQmpcl)ZlTAgLn~(VTpJ*H9HX4xlrd0
z;|Lb}pR*$?aH-R(EZiS&$0I{i2ncnlRQbr+lj!m+ag#=wh}aWr`KjI!m=iWVv_2nu
z{)%ka!^l$qU~<GRi}F34ABOfndoI&tQK)TM6G39W{~{iYzP>-V7YARSm+r3*cO^qL
zJpw+g(%TA<1_6FZ<9S)rso(?5{dxsM5Y6tZ9PYYJvaUQB%uzba)@u2yr%?&x#e#Vj
zxOU_wSe4mO9qpRwMYYut15Z8QYA09q^ESs^>RJl@6iUu3SA1YuC&nj?&s87V9n<aF
zd9_<x^|Tf#9vVw3KNiinvuS}!lFY9|OAlT>Yb))xuBBs9KB9Tn78G>SpIffR{`1mH
zHA25sn?E%5OERv@wyZx(tG4~`-yK%koX^>R0P38HrI&t`(|?E$UPn<gO}0@psV!kv
zEUfi|XqqKZ;}U~$ytddXP)sm|45SdL5A4!g0@s!{`f{0F7_Cbo5e4ByJM3d$GK|{>
zo>kToc`8m(uqDX$`mKXZ6Q)rKIhFj3WPy98QCYcCLH^=_owZPJ3H#nSFQ|D3uP2y=
z@x-QOs>5;r+PctN9rWH0LK&K8S`HTn;=J&G$p4g~vh5V$jiUXPqc;FUtAE5|qD-al
zwHv?^au6xO4vYOMJ%)75Ne-Pd0#ClSt4HVoDPp(kLbnPoN&1W9^{*|UWC5SuZy!}l
zrf~0f!PKt|`|0r9u{o2BiVR6HWHzC+3kdmEr;?YwDWH*yiS<<lQ8I;UJ>K^sZ>pR;
zs-(FsvPiKt5IyGOy{j3T0Z2iZ#GH_a>QZ4vVSmGz=5?)&h`UsyT2)5J{nS7cUhsMn
zm_{F_VgYpP_54cl-zUkdSyADH)ufVe@=V5cf0dQ65e%WecxB9)jB6tepJR%E(MQmU
zOhjKv{!F`6tP%RYzL`8jhm4idT=jE8`X1HdHPdhO+l_J#wU5a^XEGXrf&^-X>#7*0
zihiIO9jbxIu%AwLZ!p(5U9Ic3u!Bs^mQX!}$spyqnXxdBz3v`IB9W1`6z5E}5enOp
zLW&wN2hUIz&2?<SjCP%6$>PBvLI!}sSJ4=9Crrzk#<+p8sglZOdU{_vKOl5u!)nWl
zwKvlz{I>kkGEh=HZW>;tskQYxc!KS(zgn?@z95I%sfoxnHtHDn>2buREil!-VJcR6
ztj1xE<9^MXe3c1H9IQKV6S*>Bj!FI~zdkvA+qX;Px#W}m;N@j@b(7W_5XU5l&JrEq
z<)yKz;NjTJdO{x^|2q*3gkza4-5(@fB<47rfm{;L85IFz;sPYTk!uhsCWaR)iE&Xr
z&<AyJ;~m_6`Iqe*bt&P@r+*-Srnh~|a^b?3vrtzP0(fjl^gKHB4Sv1P<LH_)z#okB
zlTnWO<Y<OboZ$kw-y44dmNEUIlyrQ>I*A917~CfhXmQG-My>iu)Rof8Npat=dI-ID
zr635B^<RBMW(8FZQ!1NFZ9!~dFLAcLn0`XbM>}S!?H@{)=0bL4i=l_Wv--cJ6s(5#
z7&DxElS_ubDm}Td--N3r(8l`9W~^1T=m)WaS@V=Rr<M&(MvAYZ2(A7ip1y-iwGZGQ
z02$o6$kXG~|Bf8>3bzj3n_0yjn5#6xgBF*Q>CGEfaf%;<7foBRRN+iYnf83ZR|+$p
zM2eX@wtL$hM>>vK*PQX(_a};xbyF<b4iTgah=_wTCe{kWIjfD4X?3Mq$kR4xQ;!1n
z7Q({2RAlpR>(}eByE6!2glgA2-fA+49KfSXt)n4|)83{;kHxINS(5<4gTk5~-I!{o
z@6lfSRbGf$UU;)A-RDh}hIxkNRn(O)K!iEgZAyV8vYvPL+xu!_h2yH&pw#g&{E8eL
zDb!+yw5266k(n$Fbq|$N+hqxjE-)Tz5u=Y+&l0WSw2rOP+UA&{=xO_{ehcR>TNHUN
zT+vO7W(tRR7FVHkYnlpsQHG+(aw9>L#vI(23>n>6gci*NS78!q5~TQ7O7N5Idk#$t
zq!<hG!x%_X2cL?wcC`~6VQL4J*PuuBdrnPaXv1p=k$mmH9&UmrS%qWb<>&6PtVOTR
zx5C)OAcpStbFD%@$oc^hr6XoFyWBHTJSy0*00lK%IvcdvpXK{}PUUFA^-5Z=brf2V
zeeeQXdSpyaCm7iSJdcKUuJ-_;b%|VcvLH_LJ2N{SpUGWf|JlU8bjT)9!4el54eJaK
zIf)#;+ONa3Pj`ll0p9+?`6VcTl9|6(+hfS-M#2SC|NLZ(2Ct5$u?D>DFMT;!l>5SU
z4vT6D$?Mw&b8j}5h~Zt;U2&(~>#bdugiLP83UVNP78()1n*r`e>$b#U5_zLj1{v;o
zcaAQoNIQIb5OGs{{LjySqZ*g@rjCrtU0bU_80jAeCFM3x3mB<rmyL~uEVrxrl-pIN
zF^7_e9;v|8G=qi95oHtRO|g$*_zj)fa#mT0O7s5liN;l!F;?LGdmQx~y|;J2D+O+^
zg%B9}hJTO}Q=(p&iR}j@a<EP5y=K%&YAM3HdEk|>+=}!!DrcZNsl8VV<FcR;ku4Ac
zw=pHinqYL#IzN5LiEVv}{Rz0nGl2h?{}P&VYv1wmV@@|=TPBss@FK{QDz@n!_BIjs
zU<ux4ab`@lPDvwB<y5GvDCl6PFw7zD_?D1=B|WwRnZihx`o+RlhWt=`q{67r9x+t0
zuX&)Mx)Dk3iq0*nvacN%pZ^WZ(dQJRD99awX$evF-Cb^`ye*6d3|UG=a=*xH=}*b8
zeP{6Jjq}YBH;j<RlaT41+vj%v6k7jJlwt8B{*wY}x6mCoSqR_?ft83&&;YuD`*{w8
zM^FNFD!T{yw9ah|dUyNI^Lx~q_7sq8A!ul}d(7jtat2O3)K2PDRc9aR&?#M+t4^aw
zgjCddKhML4Z8|2ILRMIRFd$iY2p9SKWPmT0d@7#qtQA>uK>G_GyQ=)PK9s<SYul>E
z%j=$`*t?Vk#&ef}S7zWj;l<tJ^Guq%MHJK38Ou;{O|hZaZ+3v;N9OUl#|ISH->z3=
z1fgFkPkP5qU4M2XIN^#bn>>pG1Yd4CQ)O>2c3a`mRVzQs{|z)S6HvTzR#RvYxxly+
zh=#`n3W-R)$b9+`Ts`~0C$vaOt-wsu2py7+RE&g$_&|zA;`&2dRDL{rTHgc$`=j0e
zOWOSx*^G(#e<ryXscSlJH2=TSZddaZ3jp3rQ!IZaxg}|uE{Bp+!is<jKE|6T5u9sP
zk@fv@ZGeF)IWd;`C>mt9rhWOO8K#F3wtG@Ra8upgkt7VsJ)brn&|!^eDwJ*%Co#Wk
zgW?6{oG7+Z!HTkX<-7BJay?~hTcbvewinaW<@2!caS`L%<K>z$s01R39nK2+e%+af
z&7lo#f)yr#!2~Nxh{vMwn|ZEp3!nXEr5;Lljqn#5f8x{}(1UvMUd_?$N>W`ba_H0d
z?4k-EE5*C7E<YwW3$IIO(N=0n`_^2KY!^pWRmU=OSBIzXt`lSDh_dWj3g~6Nr6d+4
z_$CeDPRD_`L81VpTke=o>iKccBMBO5d-jCHdl_SqgoH?moJ<s{Z$C1VFSYzz1nTHI
z-DuBF7hX4B+yp#x*^yfJDAQDt{^(QTit|{1ERvvOsg%PMC6)H0n|H3+Mq0^}eF{$2
zwj;Y`{wEJ-mobcGNiY9^QYGS*TX^@+#`WrgnFTKs(>9ze@AANxXi_0eX>@hVH{lxo
zOg90;DTSJ1K8)|Hf?AFkbo+r|U_8H)0NS=Zo+up!s+ts8oLqt)xLkE3@p#P)cU34#
zT<Nv!(-?D%iwNk0avErRK^Gxi$eeN>mP~07s)wf)&B?GJIdFouG~RV|uk$X}er%A<
z5bQ{Hivi>68I_3ddbB+Ji;1o5!!GFDm~|?)20lc_8(fi0e;6+zGVz^oi6R7kDk->^
z<Tb&pA*(p5rH{0u@kDz4Q_iq0&zzpM*_p!M3nTxw_9Mq<t5)!{d5yTo{rUi?opehq
z{#5+xHOz~xs2Th!dkWQS>Rg*F`9}g%J$?AXs6HSYxD9(Arcl_Ir|aylGU|*F@(F34
zAH=fSk@M`L+J>qRLzYDZW8IX%yx``Gmu)hR(w{vrR8qQkTIOHEXdPNIeFA>kT!<W5
zb%0sV<B9^Ir65A11epdVaSq5o_IX)kcb?sbTBxV~Z7+ga&UHrSS9M}fyR?5XdkQ-C
za?+s#>f~xDiOoSabGMs(HfHZGS^O6$F>d{pQBSy_y$QJyoLw8$wG8sw=v&|5rZ}5w
za4KcP<!X~}RZrqMDRBT+qC)wAveN+3&>?+Zjc5$S=>_&i;HYSpT6^=ZR=*A-+&d**
zavHcTT0{3Lag0VkeN0p8aYwmP0liv%-qh{6D-Y?Pk9zVsy-t&{xp4QjvcK7BcTR9N
zwOqPYJ*{O2=JTLDVDnB-SmGb+)%2X&nGuwdfI7^;{5T_83hrP5bf$suVSdHkR?D7H
zq3E)zW5?e5Va+=m2gP4^!xyQVxZdx)$=(;#;NNQ^>y)mOC=ZwO7EJwFPB5<Dro*Wi
zhIB>^aRM%()6sF>mm~E?(t|>Var^sgLQ|_NzR3k`7%Gyq<BD8~71h?9?E(Zui+JCs
zSHB1i9B`msS@%}mcdp=6d}3g?5sW_~gofVLIR?r@KG&J$Zl8Pdrewdy+y(~%;Rc~k
z@qL6vRLw%yNsvI}>$i7vu5u}#ve6HO2AOd&4d>|}HN3#{Y7FH3FuApswCbG-zUJ=D
zt`vz&0ZfOU(^ITm=Zx8+1>WJY?)yfDT$0EZ-hO~wNerI?m{@IQ8Gg-QSdK~9@AXXz
z0WaN`GRtBTbu3lDVHYiaFhT|Vbsvo`Of57Lh-v$bo`HW?s8R-bwPb1VC*`dt+@sOS
zx1Y#lDvWTAHJD5zUfM3>bYz12icU(P$bc+MT(g2hiz_=bEdx`7<H4$vylUpRA{Nf(
zkz;KQ#@tt25e%03CXpHSG)BomZ6B)1)G2=9?pt;o#f7q8U!!iBc~TjV#vUz)cDeUk
z>DiQSL^xH?v#R@iXy4AycjWZkEy}y9<<y9Q55*GpBBx^Z`YnCKoE*>we}OMnQpO3{
z23#%O>`;@u#1VtcAjJ;vQbe82S{-NE+M42rKt!UbGf;j!g`_7q`o8x$Lv2)xybg_j
za?|W86{7G%Z#!9$cV2w!&NGOrX(*drNp_RIqUiJEem_mk*Dxz6ri<h$99|60sT@(w
zn66J1(Bj#C1^G`iL)n48`&SQ*F=n`j-iKP)=iPc@c+sdH%cS@GMCHWn&uCIxt>)7{
zzc`2QJ=7hueihYr#ZEN3ROPG;6@{=+JM>&P3}Jv~H`G7SOpC34T?E_>TBZa-A(BEH
zH;`(RW2)<)Gw%Nh=twX2Y-F7<hZA$|D5@T5Ov;OrQ&u;y`h#`ehJ4-jN4F=zZp&J3
z(u8MuC;ZYJ-I%N8SD<ff+Or^aBc8y5;CAO8T;km#*Ma>U`}bPdIA)6`{m;N-dJ44G
zK^Ql+sJiqNpw+`<#~dzJ5_yQ~ARBz%BzHllcMD|%^fx&fr$l+@i4X>I#aPA$adVMh
zST_4J$dYCT#x_J_=B}p^c|4`szqOedT`R&_{5`p4(N0ETYd$9|{HQ(26K%KqUDI;I
z8>HEG2Abo&g;PKi{|pJ;d|QyO_mOHj(4i{ngGl#A6yG**T>5mnFNwC6UrdC-^yy8_
znO2}VV86}jL@n0M2EU4Mpe)UK5R?DA4Ak?c8c*s*Q&tTLbD1S4M<40ulL-_L+GQxz
zeSoRjDuv(0!*~QEM)DDa5G`G38T<EJqhGq<tOM+NIZvwSd(d{Uc+9uMcW?o4=ZspQ
zF9~&#EWelN=@KJ9oU>r3X*4KdwRBv@k)jSMN?Czvc=%3{Q{t)HJc1zy6F;!*-F0C~
zpKq{l>NwTWY#kik(&lGva|%V*kRv@mk4Nyvt|97G?y`lj-|a=n?IV`%Ek^B{fEy$R
z_W!L-{$)1*%Xeb=*Jk=Z`c6d}8n%b*|HpSS#7RBAofkSZXE!u$fHL98o>r=dTHI=+
zi~M^oVZk$O@UYP`N{y$in@0ONt9(~#fTyc6$LkT{_P8~*qsnn}(-Z;A5-yX!WLOwa
zxeR^49j&@?dJ-J_`Mgu%+u4J$DU%!?)S!&RIau@ld_(Zv)b81;J)X({+o3%St^4`-
zcCG(*cm4e#?@C~Q{9PiyQ{DY@QZjQriX$gHNs1ULo%<l+rvG~wjPdW4pR8gZmqxpr
zH$D3d=Vf*kjuBu)r0FaoUDfJ}A#<YKE)iYw&ByiuV#G~h`9==I#@SrRuIhEafsP<$
zs%jM7#nEhiKUn>630>_5ev61<g2NRK(GfLygf>u@;qnux)tA8#ki$ja7vsIvJk(xT
zvy$&wjeu8Gv29kn^^S8y=3{$RJejdcFs6Vx+;$Egp<(d5upx2iZ-GLEseytT8D4~N
zqqdTa@k2z=PV)p$M}vy^V?InxfbNBdrh1QXP)mJ}YcsI>&J)Mlb3CZGn)0&rRN9zW
z-hqr5cT}1(<-<h-IFJ8PkmRgE{~(~=M|d3nHBGc@dPP*j$yQ>LBV*uN;aGcX_dflZ
zgHX6@z}|Vaif&3^ZJgc4T^jOLiO#$8P&6&jzB^4*iy|H?E<d-bNYk||*>fvij^N`g
zpmA+I9GM_1vJiJS6y4hwfgCq5P+S@c%A9m~3v_Gv>fH8V*eN65Bb_H(fH{mqp*u}W
z!oC=6Xya>97i~1lm2=(ir6Ugi@vO7}%!1q>Mq4)y$GoI0z$KEukg)gf{$pm(>UIM~
zZLZL-%n?cxpAKbaZRNM=c17U;?)0i=x@bR={Zs`-h!pumNF44zt(F7seQiDLgkN?2
z<ndwZwKC^+8yvliz@2A=cF%Zq|5Dd{jie;hHftA?Q?lN%zn`ToZ@wL84luW~k6_iM
zGL$m8n)z>UO_gq*Ma@&R>n=bW@AzS=Q_jJ8B8Ex-01<%U3!v(5(&H!m_Tl6NiiCvR
zwLb;YiXrBvM66Q^I_V@WZPwi6=z{U$CM+E2>%5`@2Bvls%LOE3i1ADq8tJlo6{pI{
zxQ$|U$gjd=*V1O(Z6Dc=ddMPzt8Eg4&n-t~%5LHArCvWR)_h*Ys^R%{TH9~6S90ju
zQP-HsLUP*z)49~1ak8Qc{z;cgG|zCQ=WY}@Ri$Ou75t(;!beD2*6G=vVI4aR^QPp^
z2c;7i@htN1!u2*_Xu(~)mc1^fx6~OzHV_NC*FoS)qDG&v!#KPOnivYTz$c|XmKnk7
z1A{Qa#N@~4L;c75VH%c7-LqO-UoqvGvIe^jb;!o)Azq#C?GSYP$s_;V=1BNF*05r4
z=1Yj*Sb#xANmL^?)*y~n)VaVv>iS|o(zh&A4-`NjQG7s!I|#k8$<O(kx88#Zd-6wx
zNR-@Tq<c2=-ug^a@{Uf~Rq<Th_zg*kpWbVq#;E;Fs9&rIY=>njwOBF~aTMzcc8n)K
z5$Cryg!IlzN!8K1Hc?IArZ?aAt{kR2@s(z^383qfk_zqUT}lE{BZvp2ln(1Um@sib
z$^;+Ze#b_y$B7BGi9RJWh(LT5D}BWaHF8LBb4V!A7<_tOEa4F%ki+7Ebn+lOCOLLu
zV;sI>1<lY_8$)=US0tiPO={cLW7iVLX~YgX>3&*{PN~=PufbECCS#!*s7SX+|ATJh
zFhu64Z}rc+v&yKxKSBl(ePxZBAx<g;D$ZvWk09C<sd{?^LCeHsbBfhgH3w4Bzm<fg
zP5&N!t@M444;b)kNViLo58&^w>jSc1MyE=J$15U?yd8`B1nOH#*$fjFv1Fu^FCu1%
zkD7$V?w#9?#diXV-nFzbB}2%nUh)V<k#i)wBF-5Nzy)~a&aJmq7vV*D_d6k%*XI$I
zwk#)&)%TSB(#dY+JqkAzj)&>ewTRHiR1kK!xHXNB-L;gFC}klGhL-#T6b-2rNtuUP
zxa^4J+y?h1vE|}SuRV3XP`8k_hnlmu2yG`LmIj_Hm_AJ5fd+i7+!D0nZ3KjN0Bok#
z4qVEV3k^3NYPZU-6DR55r-&4aGp0B_oSjH}d>poRP0e!32g~}=H>~S6DZc2i<>q^!
zy&aDXVc2WnpODeXttc5m8n~ZY4n{B(@}ih26x1ZaB{xmp*qVn8{yYh5Qw;G^!?3n5
z0+5DYm)VgU(DSAox6#N<WRpuhO)#3Z9F*LNq4}9y(2H1kRJd~@1d7pSinOl6CIA@7
z$`g6;7B$8+0`qt*$>S{v-H}pdBmWJ|p%R|<h?&Z%$O`5-9O@PoG|%lv`EJ$XGKH2t
zvijBTXBZ~ycbehn4!1{{;B>;@WeKiXN~io6V%*yPxsHC~{R#c@6dx)|Yjb7l){Jc_
z0=c?hob;V<=L)IJJF0ufP@Tt5fO^3{3~?1N9VrrO9<C_z8_xvD^Jka&(EY%K!I>_W
zI)O9x?y7Npghq0uZd7kZ>CT`+!=P;xeUTUzHBKbQGOv#3m<CR1pu+0<acereFeJp(
zE2>_tFBLngS#3o<zE#Ia(0L|CUw=x5IxjWnr(O7G5_R#**Lr2^{J14UND0P5U{nPQ
zAe4}ib;G2LJa7-|l~cwUg^k0v{rRe9T$t?7!kHt2?4%9omjyoZ-(C>`a7zoP168==
zu(30oWJ&aAF<a)?d*Hq*UD9X1JYUnu-SOyL7UQfp?$X|Ph2dU52)qtp4)po{5I}N%
z2>_RJ5QdS?D46G*+qjueb!f;UJzw<+fNW3k;=|-?<G-^Paja~gJdOrim9K4MY&Z=S
z`-bbD=)W-eH-OEvE~&gwlAv1EZuH55Fbhs-+|a9%%tAO+l|W|Eh!x8xaO>M!6n9iB
z6BA3OCbO`2SX#mUR>j?9_JfS6UlrRgQF9WGA%&2cRak4UxjQ>WhG|~}_m(2u-z%q9
zPOn=e)y;!>>9$ZOj*lB-6D?$vtJm95M%u-F1BT7=Eb(^{xt-*Ah=o*Jy~u0=%+W50
zgJ^d&1!?KIQO-5}RqmkMuehOSfipt61Vykd`<5<KkU9&WLkbuvM^ZY&K>YHJuGgCX
zFO|jgAJ`9O7N-9hZa<}=S@W;V&qts~*fwd7ZGfi;-JO$)T?poIKFXh;R5Z_UqPe#D
zQ2jnTGxyZ?CYGgMbjOKZ7T{ec88I+1$)R>{lyrMo5z1hKkmF6|mVHRuGEd%MoNr~&
zXsnfC)GXm6<7l#GWI-$6lXJ-I?e+3HwXwG0Dj-ulZ+F@~G_2eG@f^A~c=mX=v35w%
z(Bt!5q&lF|q`#@Y%1TRJc$<LFnMHyWDQ$N5I|wu^5;}{bfo?Jj!+Fc87O-@t6#fIa
zr<gX@rZ3gJq9B%@aBp$%i_2-<(-zSk=#kPq7f{Q9Yj*Lk)WAJ%v{5|pZ3|_fw|=gT
z_Pd~I{kGb&)uH2Cs~?oT$ZjRxhbM=ChsukGRo6>0^~c<d!Ab7(ZQT*z=oq?(nc$pY
zA+8Gvrv;2fc-Y&f$goUr4asf9W$7`IJ9xG&DN}c1vC44UKD)iStSkb`b+&jSP~(pX
zry3e?#cJr948>Yw+FKfv%cRi>#B8fMom!)zQ$jtqRb;pJzbepccB%l;V#%C;78!ir
z%!D#dwU|iAs+WLXo*6X(k3eDDwgA~DOTpS?pwsrMc8pKnO|)*b-g92MI!Cj(-$0w|
z7D5G-m^}`F(oA?ZquN899kv?3(rK^+UYuaF>>S4Owy?8drt`?|RSTZxB*b)RLxX-Y
zyG54Cu*N%KCO)-Exa!`ZY#L6$B}n7(MZQkbi0I$H5W&hxbl&CxqlODb*2DE$2WbIq
z4!nGl8ym<ekeUvmOu~PYdGHuqlnfyt5mkY5v}8Dm!GX0VpOU7UNirmXOBzixo(%jn
zA(?swB7m{ghJX%4Iz>1}iR@f@RSnaKBtdUG>eX|1sMs@(u`;>>gdSx2^bQO4$hQ&O
z`xIZfPT^e|?+t+azZDE@;Tt;gum7S)vmeR$Z)64(#J35SguwULC^c<$dvHI_?o-9O
z95q++oL6V?ts*8)fC_F%`aQh;_#|z<KVDbtNIbbb-#!e))kN3>Q`7WL22ADH5`4CT
z$WCh~`^8fo#CxaF5_~^vB47<+A$@i#6oGwI$9ne26(>JD?!2~GVc+z-UW|M03jA<*
z<&$jIeM01tp(LFUDg|PaXcD@$(}REo5#kOgXj+x2{2^t$v6GtHcT<^}6Yxm$e8Chm
z2UdrceJ>}iKPlG~|5wmtuKG%6Hg_u>bj8A^h*erRQqTh`^evwQHp*KrV?r+op`SP9
zOAu=hp}wCRb8Co-EJ||YLD<hlOh`SVK~4*T6odkHp#M5DQ^5FyJ0Qk{p@#|u++bF~
z=KwwZ)8F%h7j1Y`Kxh!(MfFX;I>3{>V_^bJi!^812}>DKM_!EM$PAJ=aX!&CoEgcO
zE}1kG6!*pMEn`F*vOD~kz4I6@yp&;Z@pPthIN-SYacLt|TcD+A_s8-+wK0ZD#*TiP
z2O|Qj6+2AU+%`Hkj-sWYAz`EM`k}q@rXGu#0}g4<u8)MahTz8yKCVOWC!`VWiU*Pd
zkY^;e<|hEV4fi)SN62tj=xY*3Ej$t(OV<v3`Ol*FMe;lo@ZGS&R<-W0c-ey_d<xaW
z<2&R_smyB5S8nm*p9?)Zk@`S(+zxQ(>!F5-BR!M+C7py5s4(J0UY_ApVJe*Xa=?@r
z3onHp3la)E8qBp64A~^r7^3$xWZ=Xwr1s&m(=N@F+4VVk;rw*A_CqvkDDybm1R)nd
zVn++Q$C$n^FBUWi^uuokPs9_7o{1ELU+(EX(-4-L5dt^lobyta3#Oar(lUL0q~T~L
z^jxH_uXZ}FVi<Z!d!V8h+oDy=X?-OXGm9lpTZRJ!)z}{bx#6L!df%wLSNmg4X>=`V
zwyRi3=H_dq1W8H)^gREl1WQWd7gbL}k2vL>Z!~FVJDjj&8P|bGLYU2y8FHQ{wTxg%
zcg2VMFs|!dzAN3q*d$f4P^(D4$v+5F`YzdY%X+lUk)GC&<`xv|q+723FbEtp?wE4t
zhPS30Dfq8!Nu;=LF8mVy8EDd0*%gdNV(5EAH({4Kw~v*RTD6EweVrGr<iurkHCrhC
zB3SZ^gUv^{x_<4dE~vViN)xZI>r$a@s=8m0`og_5M57DooEY^kLr0=I&K;a2X(6#>
ze~(m#unq)m++{eAkF`UVKfD_5_F*Zbrw#gnGryuvx;L+IWUnrcyPp<WNC>xDsM8mS
za(=g4qaTTS4@1TX$GN(%vp&2iC*%dlh=#T4gdcd5eV4hFv$u@1<`wtY{1r9`4?gjE
ztGLUD3niA=49E*wB>kYo+;+&BucsBL4zh(jnIjjTonTu41q-Q1Tt<tO*hp!o@gX9s
zz&KGMsr;*tw2nca@#VQcci(NF2r~<GZQ1%o)6Zbur+!3EOv6r>7IE;5sS?aYtE+`{
z$z@Xsjw3j1uGx5{GN$A`_^^vO89y@3VC?x+&M&QSfAHJ65tW!=RMO2hUg%x&{NWS@
zi>DIXas>c9%%o9&RxYF^%<?;lkxE_d6yw2t?GN~lK@(Ir{6%o%@bm0^Ig#~cfqc(l
z`m!*(B|zVlQ_{Upr&Wypz(b)jHHDWV>ioT(>YQqs53(*0vd+(IWE3dbh#D>M)c)2C
zlkfJ66AiaN<p#j#*GcozXkI5!Q4@W%Iw?re!jJeX<^5yOjPsH^wM+)VCa>PH=CVhC
z4Nz{?r)>eHJzLz@xwWhAv)t^~htP*_;w^TlnQ_jC^v03HxcC^HoXc|Uq)Nd=<pg$%
zFw{+*&0gdS?&ov{5W(P6E_LESG9M&zp5z%ABuNoQh$=rHGhPn}n((RQr27_L6#WW?
z7Z5-Km+1M6w<99QfZY%O*c%Qe3~zZxG_kQ4k#q^yU#+_Y>d`_2;DhI2($Q|GN@1<?
zZE*n82HaH7zKj5)S6Y-O4aDwTeFx@*`0;S>o~#HtbtVKQLRGG<Wr5*UGc3Kx_iPsF
zE@(%~yG`TpF{Fro6k~Ytd+ma!<Tn_Os$9ig#M0tR@GIlfm!}`{qE=xLtWOPCY-!~h
zm6nZ@yc-h-$$*y2ys!4Xxhgw9OS6F$5?WY@kFWiR)ISkSyv#BFr?~nLC<`+q3*-N4
zVNI!P#%#2r`Q+&4iDO|oN=CHnZ-8tV)ypatcA%K*#(T$Fd(i$)##KtkDmeAMA)3jv
zah%Y>hXGCgjUT_$<{e6kU;~2x>2~+BG~{Zlf_g|RZ-iXwB9sVLCxxkIU%L*GBTRI(
zzp^CI<4NPjkBkCsS~jkaO!4$iAn@7P-mXJyJUNg_mu%lZ5JOO1onl+P-R<f71mU^u
z<Mre5biH#Uuu8?i>7PoZ9`T!zzCn*lDftMx+D=nBYgpso$HDnsw7bkFiic|q?Wy70
zmnc4Q;W}t4V0Pq0%^!`TWgR(_lXRMo{J9^`8WA;CGttXnu=aup00+B9_6r|ud-58Y
zFrm7~5#f|D#K~cak*}HEnMWGu<fzN<)H53BrhVr?pg&7iFp%6JvU0|C)s1yI)V!#s
zR=Sp6Hc!RWxF%gg$@xR39V?v0*3ql;mnM82Hbnh>k<?mhQ^&t#l@`2A7H}oI8CMt6
zs!La8I<&xqYk8xK0EuxH=KTg0jVg4eXHI&z=2V;~drdYSN_;LkfS&LvE5FF&V7)}<
zA@?1Qr~76qV8=mG0^DJr@!Q!nbe`v-s^Jz80=;(D!lJ7dbAT#=lkELfx0yYBV;veR
zt56i)&|MglBba<sJGbVpV=1+9TNt8+Cx3p;@6sFlN_L1@VGz+jAvZ3ok~a5$s3i3@
zC1*egK)FBh%k%|W;2;p)I`YUHgCdV=qy15ZZ%GanC-->?VwFI!q{jK!WPoquPI-li
zoHG`z8edk9I7&)BXs0L}S|)LH2Jw9-k;y>F{-E2BSLvb)n&$yf%l7SIFs=Gf15>6I
zU}0{xAcRtXAB>_)F*8mY-J|^7og2MU3}*#Ljw&I^XwNsk8F$%nqVriLOn_mY0q>PH
z)%FP2wq#&o;`4gx=0M0CLlabOC4K{=lhE%I`tZ<foI2}76ea|1U&KekP;eeLqK`Wq
zcK2KWhFht6X3FbvIfCx|p2bi8g3wZ=)<{GV@v5x2|21{{zEmL{IfY;?1UOwEMiAUF
zV}ZtStlj|OZF5J31<#rSnn67ny*Owy`(iFG*mN=h8+@ZIAR#2kmJI4NKSu{`=w}Pb
zS27=O;8!8g-6`RfTiCjkWXm~5kbpSiOS&Dw(mPW!kP$<QRZLfqLJ=?RlrL}EM$Mv0
zoL$BF@+pW3LP;+P7q@=mEXxzxxsc^ZG*k1=eG{73Tn|kA2i(MBNAS9g%85DvN7fWV
z6Z^BCqKw-Iw^3h$#P(E`Ppx;6a>5_XugG#h#fpy@XsV$5zNo-c(ceg-&wv$&mucjF
zPGxT8i^T-ocsN^=Q6<ZU&rMUf2h*fjCM#L*D%LFhm4AC&)cnrE*&;?^R5)79lXj0@
zFgU*a8gw<+Cqsd24<4cSIk2Hdd3SWpTM{c_m;9E0Q2|7$_-=GL*45Iz1dFQW-c$3d
z2X5oh_K5R`d;=7z4%RCslO)!WUm9zM1BvM-Tu%o;4!>WT)|AekBG+|f1h>z&{=-XY
zZ?qu<EH7-x{iU?XD45f}eX95#nZ$(9NTd|6cShz_*liXD9D3xIYn0awZFX~C-d9C_
zGj?c$0PeN@#%z>=hQNeOX5&k;I;R7x&U0c;NYY0v31-kF(csUwxy&O=u^jMCWnpGl
zdetEM<}HNBx~F;*QJp{6?vQ7*Qs$=~re9SHVENdW5|e5>JI+G4i_t!emWOzE<ED2h
zV7``MCAf+Vf+7ulz?FRxiP{hne*&6gX(>DeW2bFyG`<8-Ve(ikw-_&ARZ0?S^N3Mv
z!ezk>Ro=k4bnJgmHcOc_Yl)(mp^8EWZJOdc0{iZV%Zk?IPNu~x!7(5KPd2ZS&}+5=
z>aV9*ZZk)ji{17~4zYTv10q!yfaR}2#R0waj5GUZt@fk9kcL*Re{V14SerN6&txtV
z=(~9l1dU*xbM~RZBsL<<a2Q|lEU>rH9`~aUZx>(4h}g4Os)ppyel_?g*R7lAD1qT3
zy35I7k7nF&vGO705aj)}%cE-XaOk~tHczdnv^_QtkgPk@3-lVF$gMx)68P0I{;=50
zT2Su>+$=+8_5U#TPtBsWP0%JfZQHi3HEr9r*0gQgwr$(CZQDk#uVe3yj_Bv${R`uu
z?#fY_c?q4!c`DoquRh22Ixa0=99)XnHbP+VgJ?`Na6qwvyNV%_GKQH67}^pT+b#6p
z9Gvb(#+M~BLvV1IoW86O4+tct{Oyp=04sT>%!+9%+9U}Zw$$abV!F*3Ibp_pJTA^j
zzu2)4@cf=-p9OCt1Qvq?)sC00gpB)o$}ma^fRx_Y4b?EYK^)qJ0BJ=Ry`e{jFkEoh
zrY11c4E7BXoJ<p0SI+m(5Y_d;JO)NDr3v>O_HQ`7>m7Yk?4vaV+wb)q6Bw+hDUn5e
zPr-0t^X{+Zzu=z0yT8=?`S{&e2MTkE&T+baw9)zj)5&?d$1le4kcFg!)W?qFX98;X
z{zwMbGH)T}3b=4S4T9rmL`>qjETVb?P(HP7riCr3St|#DV1id1RxB2EZE$kVES3K@
zO=V`R%1<CbWk`rIw{7AaumP)F-N1vc&uCj*J+gt;wBodcrr8^p@7w>XiA0W;`WTP^
zSD+4u9UVwJ{^)u_Uz<CuVZ9k8Zy(wY(f&Zj9#0N6I)?uW#6%tki4AG=Vf+&p3(ujV
zu<JA&{i1eJ!nY+&zo}r<74?Aq@+coI@!j)tQ^xTshUu&He`egd&K)u1Js);6A;5p_
z4@?aI-%>Y5rhkU)|M3Jit*&8v$cFHLtk+t{1*RND;^E?21lEa$=;mlLU7rn<P{S$K
zfDK8_#U!HhzdQLME+p&rH#a?Lp<|pqm@@`9GCVimJD(<Ne>ebbxCf)qHn3&6%Xv`E
zM}UuEM|aG2>%yMFB^`ekLtT5H!3FJqj}Fgk2VL=HXD!^1Mjj3LV7`bynYn)s-<;fb
ze*eDx?2*LDeE1t`TNvvlf)ifn<GCW)n+oP9t>U69-S)_t&2~F;UwNMcHXP%*DKf$B
zfsz?&dlb6Q+DKL3xg9}*3<d2F-II0m<x(3#JI3h-rI}kW3Umnt;R2Nu#@W5E797C(
zZPEiHNDI!rJqTiQvF-2hLzEOB+NDPwSf<A*OlmZo2_1EsbQCeto8yotwoOPfIcZS?
z4Dv@d7bn6<*n9>;V2FV`g^H#he*w?OJk}c}(Df?Yr^u?(&wnp_34<X*N8fNk=22zz
zCZyVhs5_7$V3v^RSYBscKaqHK`=_)e1)UOwH?Dj>@WQj4{uEKrwH?6Cglb$9I!6ZO
z>B;@Nb9dv$c;g2y8+@a=eLh8~{|FCW?-yzi8^>GnujG3&OEngej-ZVMK(e$xszV>r
zLf#i9`WLFr$$^jv{XMlR#Jr2?`EjhfqGCblL%}FbaSU>5P$Qp%0bEl(KP|0zV#A^&
zE39cuS|-}HEse8iu19tcWAAIl=5Hb{;@#6{!Qn=QpC3NFqpPECl!{WFO0`9M^tZ2v
z`-?@4Z8)nC<NoJI3j5bWP$N&YTbF*aP5MMacOd~u$S*wqN_*<Sr2SKJhf~5fmJW3o
zbp=;ZSY3YUPwH@!W<q{iI97qsP#-`zZ_SA#OBTR(3i@hc!Ndqlcat>1Q*DD=($hcK
zDpO8U+T{f@36ksz(f(3|F!-XEcg4yEQfz8CYO0hkhvO3Z_gHP5M+%LFks?h;LA>DH
zTnIH<FrfvqnTGwNOc8gOAX$JN^>Kb86mhT11TvVM!&<IrgW<<3$&6TKZN^(izO@*h
zmvC9iiOPap0NiKv@ZZNXw{NSqTem8y1@iVh;9Edy&^H_7CrQ<SJ(;Q_W3T*d@u=HO
zXZ2Cbhv@1RRD{mvF{_x~qBtT!dSHk{S!<ua4d8#{ek=ElzzSc;&izF9>&b6TecLX7
zdbAcqw`@Kyx;F@(qc+%G6FS(xaAHkw9OGjbnP?p}b<qw{<1>e2ygJADiEQ|2w~qoG
z<CVB=N@V0S&iUiOGMNIMr-D>~04ytpIxdr?CFkD?3z1L!3*~$-fc^A}YA|`qAL5t$
zSqJK|XIG_DmyU>PRRduOnfDZxLl2CF>^pgER=6T8G9x%sY|OhgRkW}ry7T!f6}070
zmO0xPW#<?`uG3s>793QSNpT}BYJ+)f^hPy~A+&4Et5Pg@x7RQKU<Ly^SudhJY*=>a
zvC6CKHrW=-VpA<#STMHB?QP&PxUpq%4UqA$2v=(BEFWXq{Wejs|5^>9`Sn#n%j@QD
zmg4D;!3?rpdp7-*1m-d}FxFf>sek+?npGtkDNY2+oon0IejD!YNC~dmqPPi0$Zxhj
zle4auz*Em{!Qq-ES)!SX0I>_^+O4~G3lA<d&n>&(1yAgwQt0P;*ba>PE4NM9a@ejv
z&?vr-Vc_KwdveH>Lw3#Xx+-Voh-p_PV08tTc61}xOx)qp{rqqP@b`R^F$nih)Fz@4
zk0Izce(LvD@_)|P|IlZQtW0eG^L$nN7j5~!=j)zU`G}z%0ymi)=M@gT#b0VU338#t
zxL`;T3hpX%uJ2nNeZX*I967RU=?_65Xnp!CpH6~)YyN;#vsqQMSb8J8BZZtg67_1}
z91tQ|BBV%t70_FuiG$(XO7>WK%5kwxbHM_Z8UOI@=db0(T{U#INK#~ysls7J5)v<0
zZ~JC1d;9l~>tSf^cCXi$_lxI$R!P`N=7Xd(3;qIbU@piql@L=zP=QL?Fn6*^!!r^F
zBlk61<IX${^OCHf3L%F=d$_;u6rcIlGq3_d+wwMcQvQEJYTBH;t_%pyOLz%Ps<lX*
zD;zS9Ho~+f8*Y0{&oRnMgo|4m&Q%=$!hj0No**Ea+_j$gfu5ym?jGoy!Cz(`h%Q&~
z5!myvnAA2k58f!GG-qn^?C+vgFauo|ncvZC8L!bTa75h^IdZwU)PaCEX6^In-Wr!j
zY}hyL4BD@vcQ=;{ktJdmA{vk*A0~-KnVBE_b;L8qPA%Qimc;UCVc=^R*701nU3wG7
z*|AXB*S!tV<I*la0vA*-?E1@Oi5t-`=^p3r1KIlFeoPUL6fei>q{?fhLQ{}51Lu<v
z+TC(fEl+6T%^bOlfkbM?h^KMcvEpg*^2nmRSzU#~(KmuKJCF%Xhr)4vq?+8Z0;|Y}
zAb{5pgNSF(l=yapHz*3llprEOR$|7F02<Gzt!9uoWw~ezRXR4^uS9}$gdU}}?mY3g
zbJskxtnRj^44hLghy!dK1$5&InazYb3Qj|WIn1xDm}MeZpUw(XeNtyFz<Tk!|C5w7
zD2U!z16?1q>^Zux_*|U)PBe$Bs{N+AyMFv(Teh`Zfhh)-;R+gi6cwo=KFX21{$C9~
z3`14(Ptd33vjow!7}HFQeOF%oGO`9g)meUSXY{E=5($t>*L-pUZXPX&i`%(BzYk?g
zr)Ex}P#*!AFvD+aYkzA>jqcB0dF1l?X=7mxju&*VavH2aI}fu{PMQVU9LE6>tR`mF
z?9ZD5>9$TH#DYISNWYjJH%q>DI~ThQ_Tf~NuttgzOf!<5Et(Ky&>R5!WBvSfOBvrv
zr{(qHUFQ*R->d$@)p&1v(dQ)j5KFiiAhF_@U8*@oMOsCTe3T^22tHB<P~ACUQ$7(`
z&wJAEmfCCSmI9AoS{UFaBz#{wo-+j@W=Qoi%S?Ak$;h%p6uL-v6ND1(JUeDO{MdBj
z(!O*fHjOPDXv2ViW!tiRffNVSL5(j5W2^9(xV`~1oy*>V<<0#TMZ3YtqFa}dc3aoX
zVM4#m4gXhEUHr!ki0ID6Ml<*vLUkzC>I?Szp84H0fb%o}sQ#JBA;QLuNL)a^07u)m
zrUd!4GnrFGWiiF8@6XwMH=NlVhC^un=m(%`=LaD1vRlNEf1oBta)$%UvTK9xAGzKI
z&my3fWF9P=8ERLPVm{wPS#*<fKoFJId-hf}=j$!ogRN$VYqd8hQUcR(xzuc{3L~Xs
z$HP2u@Mc~6#n<Ns9rWU;IAUSWS8Y%fE{K<{zCh}FxzcRhnUfweHV-m)s4(9<_c~ll
z1p4Q>ckwpzZpNwDE}ri-cf3chE0B~Dg%RyZCTrfvna$5$3i_$9MK@ht4sD|E_2pnv
zm>h+Bxhgv7KVO%f@BWB$&@pv*ik$x7K#>HH-MBqkc+0uz=Kv=}kMx7ZiL!e>cfkC%
z#c~wpuoMotjJ*Yf+o$>F#rJdNE?|>uTh1p<M-)%DzdcnkogS%kb3;X#e55Ryetbr7
zDQSVE1W`y$1E|vS>P)8BcLJT1F1R%=4rW|ZZUk*PTMvEGFEAb?w$7`>BJ5znDOnJK
znGbMjB}Ck>ZNl#q^N^7TB)O&?)vf8Hco-_=O)MFkqgZkEkUh99Vv@dV<2OxVa*k<f
zf^fBGiTf!Q;IV59*&wvoIQ}$8&GFzk))Aexy6{VW<p_;tPT=}30nOww6vvQ{3~ka=
z?HqST*uFNu5DDMDP)k@|nwJF*x1S1aA&X{V{|X_#fsFzboB!(_Vg8SjB_<}8|DEKi
zE|qxL2HP`RduM&T!4`@l!UNZCOx+zzlA1oF>*AP4+EGoV;SjF)`nC&B!XMt?X{p2;
zCWipx505|PTKlKx^BuYO^XL0`@$^*kO1NJ!zhQgQg8WfORaIJL_@;RMJ04=@<=6Y&
zdcEQEkiad4j1RUuTXg$n@h6Tdr|T*hVZ6-K6Kf>W*^`&g%t5W5O3slyy|IEu0Bkor
zTbU-*kvX}sD=3l9Zeg+~VSKO1B-{sk11!_*d*!8eE$6FA8msURj=OCsbIXMYp_I~T
z3-emjwaLv>(*)<@5dFu3%i>&X2l;rbt{$AMXKgLSM}SI0)VN?ljrBFWf&_Cci3+Kz
zUJdS3jyAj8R3@e^$0NBK4olhwJ5P%T(;Pa7<8on7+IZBOwda@RmshrYEq6Oi^tm}K
zE)S?t=Ex;A2D(L2tu&?@>cu5&>Tx(a_-?!2lnEBjLj`X4a03x%hU4ZUj3zWz3tM;n
z8t={xSzt(AJ06`{-FF*(L(nyR0{Xa=`bo1cthO*zWCW^43{<D*e&n9BiNn%LCDgP(
zUSy8Ki0!X%r|jGEH!O}4nZcWnnene6Dc{<jp07jB%k>gmcc<T2LT&TSllbFcTLj_a
zXv!%l@a3UM)h_s?6d>7+sV$$mr_5Gbd8xsZr5ZQpYk_pg>okfBSc3ih;@XP9gvf54
z{Z{0VzT;;D13x2BwT)Y7Ca)z8^>7e{g3&g6?zRP!%D7iImL*<jyDg2nkO|YfH@#aj
z$oE#$>JY0F_{tUG+K#lCkXYbuw5d&B`*8<>V=4-$`fg@Fi(I5$ZJh$YSPCVzd&q?@
z_-4momS?iAAf^;V;SV&2-Srf^QMX#kc(0mKz-9rX!Fqhq_OQ5w0SX$27=HK2h=1g(
zB>f4((}{`kVL&5I(4m9nTs4peFVJTO96bEBN;QF>=7xOkGb$i3mzwX--3psP&WPOA
zfLotkfmGIckVW9iD_y=M8C1Z@I+z|8C?87)S!sri+O0c|+?`L1JWz#M97;|q&n}K2
z*e-14Z2xYz^ZsVqMsSoTKhB&tvguc`IR>4f41a>82yeo_jBV8vWKEzTnj>lffLPKo
z^n`=0j9a`~Ul2J?{RnWdADFQEb#?eTf{%3h_FIS6hMt#)Q>CwKyA`N1LQv{LykDFg
zk8nUb!EbTeufB8SYi;R`XvfMPbdj$gE_mL2q{O)LT^%w(*PqGL@g^v@j0EgH^lI?w
zF~ly96qh_&bG1z^yLuCHwlO<`{KfCfD!u5&4CNh5UlY=`=K2KJep;Tma(x1ud~oh{
ziii=nlFSXy45VHbjjP8-0Bv2$@G%l<LNSLUu1bpD^GHs_n6^AG5*AW(H4_77PXKB+
zq?`sS;yj|i@A9|)*wQF<y!M(|0)FKQxjtX<0o{u$SQ((Ahk<Yz3MBloUoT~`)~m8X
zhNWk0Hq54g=IiesZeec1s$&8)Ak)qm;hUf!^EzU{<Se0{Sz{R_l@tClDcBl%s+Z`}
zs6f}?0EsswlRB&Ece(f&oC^v7=*~2(XkS6|?>;4|A9u-vKH!jchF!n4G0G7V8dfb$
z3Md|pPmjl8ATb08BR7yKz^VMKX96BZ84TdJKM?&E3QUw?2qN$~=uR#^xdr;Bj*_}G
zpIF+rz=o*uK_CYNv-|RdVHN!-T!bvD<RFk8jgbvR56>*Kg!_n>7M~rN*vMPmjP^fO
zs(tx3W8$x55e#R*nRNFLaplyxOl)C5P*ir!OmSJ4tOV5d0yiUfU6thL9B-L5p~OWd
zowR{<R!CKfRL{fHaDMRGq&gL3_X68P?hmyu3$k2->85Ov@Ne|T){T2V+!r5cNGN_T
ze6$uvpE;LwG)!dx6v4k6e<F@4{{F2ufh}RidO@M_o&vQ|C7$d@?PIy9=OGGBj492E
z)`AsAm%ebzCk}6r_5o~pKdAta>m@~r9=yN1qtXVrES+mG^@1h@M1@fH0UlPtReWMR
zQqq6BQ-u~pRu;iS=A&hZ%~0j<NuSv>w(cf!Q4S7MY>+OdHvY^YP@3nZG1kO-O?m)=
z3KFzzQAZrmW}b@MfK!t^sU~Gg=@d%|tQCh}1&lOTj}<q}#cSv^tdP!^Mwb?U*o%Y7
zV-q-liq@RvE$t**(tkvGD9YQw?AfVlI{OFygUp!efNvOlBtlLWKGlyn+nQJ~F&52>
zHIYn@J?PB)0H$oBCI^a)`H}SVA?RRq>aXL4v(4u!Mkx(<$h2L}2Y=R*K{dZbsw!Tl
z-ry$Vv0b4tleDVNFnNu+#vk)~Y%{e7-KN&M6){;rhQl}WP}7SZ!M?Bo;4fYW*K<2^
z5j2%nMwyW6+2_2s$GfmIuTndGKj8!7+mn*lG{1H6Sbjc(I+-F@-6#L-G{Fz+%M09o
zQVUYj71;oerIy0{Mnz2Bo3BS|uY+2SNjTc986n>dIoTQv1ojC21>SjH3w*wba6GhR
z-s?R4_66FP^E2Sew13CehuW0GU(Z#OK32J!VS){piEuV=kGcg_>WA{h#|Su`*V&X;
z39Tm?#*&(&PH^1_N==31;9)U574ls-Hlp#b2q=~ehJs-Ax7&LYn6<YL&Tabu(F9nQ
zeOgV5ENCow<Nk~NL&AjOr!X!$e{oKhz}Odzq?mLlM~#`tZRVb1o%Cez#QtM%sAYTl
z6DGAdE&Gvrt|qvuwwS5l3e9vshB*mC!71|TdWV^K$HNOeP2#1isK_U;C;*sUX>6lR
zFFU9>Fc2WMgbb)_pnXJ`-QAL?6R{2<5t(#MY0zVqN;N#7B0R)QTI#ai9#Z<3zT+{U
zB(XL>AX-HnH)Ce;S81Xqr7t>?bAf}DcS+Z9{_Icl>Pd#iFvy#j*Q$JS>9}vzFMNf#
z&srsSJ$y1zm+6c@rBchgyG!y#A$;*3+uv%}(HUV?CQvc_JfbN#ef#o_s*lmh(qF(~
zYDU%nT#5cGTFA=rKl{p-HUFQ}XAdR+6e*=55ww**i+GDfX3=z;NWc~ESc_@ZN{T5h
zXX3ivZY=tEl4uM0FiE7LU%W6b9$wDnDDmPMSHZj-x^Ir&*PS9sJQ>%q1xs&40!soB
zsq&b4PXzC9RH>X#dhgFZX(1PZ5ffp#bpMbd&+lYo@2p;zFVkVdNK^tjbNa&DduMCv
z<Lj($cXlUwo3H2l!}ce8YREC@vtSg7evm`dVwgxcNx5mfaRftt;X9ht^s0!GU+&C7
zgT;%8(Tz&hZ4Lt}`c3{@0)6B*k!@shq4rig`ZY3IEzi74kJa|3OLnWx)w_%H`bjdj
z%?6fq;~K5O*jLAMPxtyKruFqmj^59tl_dd>vIMC@o94K0Ns=v1BTcyNq|5b9QfJk~
zWy3SQ?UreZiPB0kcEg`B;V!D0hwqUqU`g4eh+Q|rTStwPZSCI$`)^b($B6TFG9TNv
zb@t(!O$aHtKZsm}368DEm2~kR8bpz)&WE8F)UMNNl?}m9?;||56DNEzF1tso!%A|?
zPqwL_CH0E_MOAQL*`<52;+WSpg+D*va_IBqQZ98F?q#i;U7FKBs;a@8;&qV25Tz5=
z2P(^Ge>ep8vmUok4PV=7Y`|+EnV23_-iVEarN*Nh7tBYH6G*niph>f$IHoX)!-Ac4
zb+itRXe&vy`A$2T^e}Sk2-zTjC7WbJ><K`KnI=Ct7^K~{u54{CbT095%<xW6fFgyE
zpI2a=Bj4>p#ssF`gVO?fO$(EGv*KecBv`t97{))@S9DNSlv-R%Q;rq;HWvs(GY6w!
z3a4h`NDPQF7z^TJswTRfw8J4b<WP)4=t#q(fGTp!i<FQch=dnR>gf}!>Wt+8*{-74
z+75y)tlCl=>^tt#M>MVHQGg3c>*hv|nWJ=;JM&*9iBO%USobXsoV(~!|DH;iT-ID}
zb*ihVHSR)T-#25QU-e-`=5p<1C7?{1Le?yidmYLqdhFpu==+n!P$s=zWVFbHh`jR`
zGS@hL6_SH>`mGiD<Tx*>DM=A==%2*HhH499|9D1#S7@xLHQt`MANbyL*F4K-DnF^d
zF9taVsNz4##GZxr|KuEb1ca}9%jq*`*?XX(gpnZYpGH1aRx0hoc@Z+PX4$jis(wTD
zVdFfTGmMHPJB`a0ptQDZ?7dj((Q2vvxo@J9KCiZN?k~hf>+vuk7|UA#IV_hZjVn*W
znYiDNy3%G!C~$!z<9QaWjl4JtNGB?r9BjuY-xfh8^^|+FPoNZ3e^g+IZqrX!_DZDS
zt1bd@W*J^mLaUL_xsI;=y93+IMKT_RZ7U2lQ)D!s%mbVn_kMD@U^Z5ArUL?h`LPlM
zjK9)qGbQ*3hUSuXFDDdo$0BXi{Fz8EDHF-As@w5UC!={OAHRXQCRhf!%!1;3wA_!3
zWhgCtjzlUQo;V4%^TE1)eF|-u>-TC4FG0uxcE>g|5rQCDxQ{}p#;i!6GVI-4vDU7Z
zGn6;e1;T&h=Gh^87a$H3p)`Mpjh=lvjAYH<dk4FZBfw1}fw2Cgq4Um^;ViC=2S60`
zC)6wEZXoS>!B@5TIpu|5Vb!38@J+6aUzkgRPA0-a*02^PV9|=<a`2p0*|buXW01?W
zBK{D{wUMfIgpdb8&R)JkmHfI^iAbSSute#EJy^kc|644LE1o<LIKu5ioHJNPdB}h<
z5Ko1E1D^{C(}vBz(*@;Ljym0}?5wO#Npr^2WOl<NWKzjhd~-!*dqt%k$CG?DYwU&_
zgA9BF*z_U+_fFH@gy>|#zpcO$*TdDWcEI8|>`yV}yj4_M+c1-)MePO1*uc?d(;{hc
zHx^tPIce5wMwaAW00#6lkSyCpdH9DVN_XH8M>drG2+Qy?JxB5<3!+5qbBDB#j?ay%
zmUIrYDU6D8;X}WSL`qm9LoAfFevT+jh~d_tX@=n*zspwjK4SKetC-vU5d|Keh8p3~
zv5<cS7r3$pN*wMcpe$ZF@1-2jFJ_|IHzoG_BF`IPoVO{kZfyA)$ryH60)OKiNXs}3
z-ECn2WGG_+=2dq2`Tmu_(yLeYoG_+=`zm*x`}@K0*THHe@YawiMGd7;x@5ue%fTW&
zi{^f@054?RY&GdNmh1r8sEm{QJ^boAc8z%&Bpofza`^?C$QH*lrdt_lXtFXo(eLg`
zkFD_<Xo0T~b}XDsR0a#RG8$gi4Lb)X`@iE$G;WYx7JY_?kw5v2u6XXByhOC}nQm6M
zA??j(ati7^c{Q{v$Ji8{?Uu33?os`iKA>&C;IK@Sp8H0jmqVfv#RPYQPZ6%9b=UM8
z;xeJ&N5d_YNw%O3f%-MIwcC@Ls+v$Me7}R8EB0&R0}eC7lD)!!P_zP#E?k=AIhG!}
z3d2;}iCqV^b=YJT(wcT(xT^?E;Le39G*+o&y4pvRZi+{!F`|8Lw!57g?N1+9SC9#~
z7XwU1%-fme&*>AJqOCxH6J^6rSuuW2JGc7H6ct^WL6?=u(~ws8PU>D9gOyK7e6ZIh
z8rFE}5Lj6pumI8^E*gJ!p<FiGO`C+qc5Eje3taSe>(u?4+r2~Uy?VF9HtQ|{n#qk>
zQLF9ej}-qN5mAok(wqi|sA@y_{63nS7WVFokbm^2jPymB#p$tcbyt$>ut>y7$f-8W
zP<kl<<77U<XUm4X1u?Lq4VRh)I0g&9CYy@}#|Q_QXPXC|9@;I78Jrs12a|d1?BE81
zWbtg?uc)965Miz~m!r5P`a%Y=n3RSwg*1Xc&L}_!r&b*IwPt8Weuyv-&I}f8+(8E0
zJ@kAJy;V{PZP+XrG&W%OfWm<VeEeA#R7C($Th}7@Vi#47x2u})jlvU<VBlK!G9hMk
z!rs1NV7okLgFT+c0Px!I?)cXYklrnmE*yS8KRPY)!C9HMW`)r<j(UV>-W%{(B~7-4
zPmo(l$BDe5PC4QGMEEn#4f+95F;jmF^Cz)C;@F$S>RybB>bos?lW3>BPIE-B=?U>s
zXkgj}GuZ(@iMwlcvA_(KxB77ola+^RH)L5dlcQx*Yvd(%%a{{7j%-5Nuvddo7SwPN
z)tYG$OpC~AW}gt-ztdh=GyQsj7GT*G*u3kZH@?!zEJBsxz2Sf7>#6N;kXb}!pDDS%
z(`S0(fJew<a7&isyr1h2-&!YO?(^34oqu7ceT7YCQo#uOh15@aF54T%0u{Z?UGE&T
z^{Q-$cu)=w+vNu8(j-G2Cl<4}g%Kfr1UJz}%O&}PKX5e(t}>ta@s5qMK~BM)mhfE2
zR8MDzE*NKrt|~PNz;qA^8^!p*m57{HohUYlLwLN5y?9L28jbL#IU1ZCX0cRTfBw>i
zYH%1x%BT<C3^qb@6^~XK_`C_n<MwI=3i@C}-^~HZf%7Sz*U+D!|AHD|5JCUfs>$*n
zK~YA||Gl8>SmXb}Nmc8zn7;Yoc$jAvnPah-H;N#Wa8k6mtcKCr(z+*pzkZ8HtPE;y
zZlekKXVEAO?%!pm+lH`A*Dz&#JG~4JuOFI4bxa|aw$ISfG3PUrvgRwApg&$+2PqJZ
zy(?RAx!*s`miFe#wHGAf>;BwdCYH(_OjhFe`rd!vp5D^u@cEF5sHG|gQAHOQP}y|N
zVOp!sQrM%(u^6Sx3v1+1nnEVhb*c;eEIQY7nz<;xX~cymXd;b@=sD;v36vDhK@CKm
zN77Q+)@3ZM2K&bS>kGF<%RMA;Ldje%n408*tFn!d(WPH8Nt&K?-y%ZpY8*92Q;T63
z{eb4RHNz-?fo3Mz4>41c%|bv-o1_>aEP*^_3lM8t#3Y~Epm>f^>;Tq4G?OlzVQdU+
zn>RK&!8OB-B-$BGN;_I|Ri3Y(c3U)FaAvQNLZnki^2cL4c#XksGqOwNJqSI0xB51S
zPqlBmOnyEloC_vI5)kr7Ra0;UmlCRSp6&suLlgFdzcN_wAj{+HYiMPbiBec^DlBiC
zPTlOHJ7GkB@R-sKc8`MfyQWmNo2OHVvkjOZA8tQP{5mY?!dj%HL9CvcEu3ix>d>+B
zVvits9gsNEy*mQ#--H&Hu!isP#S9k%LiDdq7%M!>BRestKQSjiF-OipiSo;SU}}FE
z+)m3^G6{Eiiyt~ORnimLkZX7DOk>)P6D)BAei#jlM0PViwVLn)51@p(U^!zqk;7po
zAG07zAUAHryi*wE7pOW&Hx9Z0VOsTIvG<=M81`VqUv!zsn{&~&e|b8%s%_M<xce$%
zvf$`HTEDou_RN^RA4jJFT;v7qR~bPNTKQv`-p%z%t9sgVuA8h(mxy&7U12oB%q7TG
zxRU8ZARssTE{zPC^Xxz4Di98#uX>eZ?1kl@DOHgCutd3>%C7}kguw@D6CGg<KM-A2
z(s!cV2Cn}K3DEg?1b%5@^i5BkbR0fJNK?9JL#@PPPK5oZ@nQ`_?JjX>V8-kHTnL1R
z#{p_bh=bGOr~dagc+Qysrvy8elZawN-G>Ayh>e^S2R!yBV&#{6zt2{amth|dh=XZn
zp!FAakMdX+<ja6^m0rE9g$jp8Ij{wzQq8pJ1PAxuH*&s6#`*`4^weH6C%q-ZuTV1w
z5X7%y-k+-{oEJp?dL*4_l!k3t+BOw)!Lss#oQ^@?-g{hjNw=MWg`ZY*_4&06qQBQ?
z+ntG^l3cp%Ps$`{2oc8xU%@Tx_?JEHQW4&8O4B-0yJXTEsi~we&`{usPy!)VWS)b=
zDghfgxoTZq+mO9|eP%W8&UTRH&}5uq>&OL5c(;KVvI5e8F5u=NToFFN?O*<vS{H$z
z^-gYRF$jsR&c5CU#JcbH1>*FvJYx)ii)Lk-m;b&!pZb$klJIOFG(|$QK!qHE=vbto
z^kVl4pn~Lsf)q;g=x1u3JGIvH?g=h|zl_#@24>a2S&4Ri2P}-Nr=6ws#{yCx3Y$kF
ziy3AFW|TjDbyKECduMpntFbKNt?jO>V7W+hY17VVVJZxr!&Gr-HXe(?C$jxL)I>3h
zy_I|41XuTDdg#57=dvG7c|;xJ)LR+=mjXj!*`1U6EAuI|nSo2MDxx7kLZ4;-fL8WR
z{;2ViBBScPBtq{(Mpsq7lj=l5znvD#d00ka2D{&x(-6nit_69%&2+SaaPYvTen-{5
zsFMo628$o)!)HH2m^!DasSFz`*IhTuL5SSlq3^<U9pKz~c-}3PscEr5Ju}4N&k?>0
z8nm#VG^rT$?b#%+o2P#BX9V)f@q-Oc`(373xut?G?P~<ks<`{M_NCP%FMk<~1?fVk
zvey?({WOOC1p1?tp58osxbXcwSEDQRJ6bnoaU%Mq+;9o!d2Q=4H?7jNL+s=jEqWS~
z_I^ggU4^VG<YtAr4Z_1)qqbiE%7;gD8fz+o1^qqikR8g_Isy=@d1w3ix(cf|Tg!SY
zXtc`(TBs*xjmO;t+#GZ1-0PVvGojmBlzw#}{MMuf_&}mJYb|{31&#qh%}&x>rv9-z
z<0GBE4N_J>{G4@P+T~YE7%?Ol_XsV1M<r_AENn|Ue*0Kbk1X8FQKK9I?M0hiQ!1>3
zyBQq;O<`Y)G~r65y9kWfou+U0I>RGj6KG6WR%yn)=~Kso1bZ96<#q+U$^i5XCkqh8
z?RnRyY16OW(8PDgPJhP7D2YD|#&Q6WB;{=r<7<;pL-&OzkB6ayFz4z;exU~e{do=L
z)b06v6X(8WeXuhPZYd5=l0Z}5K5X@>f0cGp>u?EGOzD{qq!}J>x@B7}Ye1xG{vH#{
zwojd`o>5L-i!3(wz-mhD*CkbN`GFIDolg<$+MNd!ID~aPMTzZECh;dz3dStZHYcz9
zSHZh^gM5?44~3mz1H>Sh085nm#!Kw8v+iqcpr^{d5Fi2kHuNX0Iu%&|iWb6m>RRwT
zPwS+1@Sy}fp<Zk$Qb1;BA!U1D-kTS5jTkVPGdHP)ry!;o#3HJNEX9&c53Yz*XfFty
zU$BkVhVINYmD`61{x3V2&`8{WzVrX_ONo)?e=meM{{Nv&F4gWR<b*J8NAtnikz3ED
z6IWEt4(uKCXf{@&@M<X5r+aS$04M_jD5&a7Z9Iv@fp+Q8KCkJ}$Xy|2eQ>@Qu(EO-
zQ!tE4lE=^3444qLiKaxDGDud11fR!G=Vwb#K{7{Jh%gcafy6m4ujR~oV|H}1ij?;V
z`_D*Fk>GcIzU(+VI6Qw1j(qyw29MuIBlYNLj{rPLWoKb6_)Ae;BvBVcxggCB4LLWf
z979>g)fPvsc)Xu(C$(O+cR%49kCl2H2Jamj)7ti(?$c_a_)NMd(1x!aSY1D%tKtlY
z^m{4y@1C%tsIIO!R9+yQf1zx(mz!u6!Vsyq2`ZWsf56GMW<Vp0fiQ~?azCfR{G-@N
zu!7Ph_qHgAZBB}~^IZ%3bAf>fI0)^Iz@qYzYbX-Nr@idHG0BnNKot>~x!kG9XK3M4
zU6yIualTHxXp+qGIlT502rGn1k))Zv4xS1#v~WrwDbX3~!X!x)t2V;>;vEme?0_#2
zt^^)?@ecg<4wRK|Iz^zLkMeM9$%a{x^&7!p{AKi+zvx21m`JS#<Yo%&a3n;kKbQYd
z2)Jg0$R_to#ss+`f?J%EA_CjP*k7CZ;Mem`g&H*X%Pu~eG4ViJ^{x_b;V9p56yrP@
z!+?QNzG>BaItxBAfIFnjvd84R@?VKW&L?^)NU!P@m`U%)Y0%yq{1GF}598w!<*389
zxRN_m5_XKfdKw{tpked}fz;gO1_KGuxOTREu_}hrL4-sCFt<Lf3<Hrm@tAZLhvrut
zBNjK0FM=V~cz?lr)VB%5XHo}ca*f&5PhlzHsiZDe{#Ti<g|^zKF#j4PW5>{0g@m9U
zKk%i8w{o;!H}Vq13T5suLje<z&M{*(I8GhQY7h>r%qlpS0LRbXx55B`pT~Q?F@#wo
zRihNZi1+hzhmgeFPH!HOX_Zn<)WNSF+O2fBee00@PS8U6Fp$c4Q$Kdw;2g}ZwPR9w
z{^7&h9lk=%#=(`AA_(dr36h+pA{dCYL^;1Rw9gdPcm93>e4#kTk1#=iH3N9XlcXVy
z*a=;7cZaw%Gt$U?k%u<`QKSrgqSUG!fI1Mz9LS2WFFBw1HG;KE78$;1*zM$d!>k3n
zLBDeK64{#iee|-EPp$$5MS%r{>q;%jDpHM#aKTwJc)k%{-BfeEt-5gnl8G-dcz`1B
z&qumrirDrKrwS{QmbOD_gd<aoV<|K+5QI1j0(PF!6MBhsckfoU2Vvx)T!rz^sy>lM
z3m0>sUSl*ru}K?^lF%SFhbS;*v0FlsFD_v-i6>lqUSN28iFiM%)e+8*M55T%WN@Q1
ztdrbkPB0+l&U~<ij(?XAMZ{3`uR)9XsPA%m!LTawWg~=r;4l6U^IkGgER=FLR})V)
zLDi)lqI}6Pik8wwr^I_JRZT|{5*ukoO$RIvpps6fW}@f1n}!bi;nqBwPRmx43hDyw
z+eZ(K(d6{@cg_JU#ae;k)<3(H(jw8J>~KW$7Zp_VwUBx*RGyefP-|7shS11J(hh1L
z9X2KjA!5V**v`nw=GdoUcXUp~i?#pSL2)DKgG_e3I<#;n8pO4gzq&I-E1UA{90=jw
zOCMF?Iin1Mr3Y!bHSD*hy31GhU{0QvtF}^5SJ>>AYbVjP;--mH0PBF_DrWaQo<a?e
zVc2O$Nwg?XS5iDc&hzdx(1IJ_R4&~F4p{Dou~h{H(_Vo;m0BpZ0W7OR-3VH*tdyOj
z<|TuU8_(l%*P2Rq^5ONjz#CNFHajG+-sw(VaVJOG`(?^B^46%AovzdHGS4@mRpwcV
zD<wiLxgrocCs}T%_h4?l_hUr|yB=<gbFSSNJb2=a+n4XziLzW7><?IY(%=7pI5IM@
z{=X@X|9?s&8w>mY>c?Bwkof1vL+SayC%iPsr^8sk#epny`oDTy3oZ$`Cel9{@uX@A
z*C)HLz>r}epz~6x67*!CK(@ZSp715ZA_9@QVzY9ha<Q1s!NjtTI$2rhg6NW#g2hy_
zWW%S7g~IWHiWXcx-!HAHxA2lRXK!cIktfH8ov!xR@0+o<x?M1r_&!`7-cLV!m*bVC
zql!-FQX*vw`nz$bGqzas1PWffJT6uGQ5?l>e5xakr0hc)tEe@eJ#X4fl7yEJE*+M_
z<!Ff~#~&9|HQ98v=!q8Udbui(mdWwS0mx+w0n4WOQ5Yr>iBfjFx?qxzV`0yjjO=uO
zyw=`IG^Z;T-<`bIzJBV;(f0d1I^YJ_a=XM8x@gy@D>SQ4m&q0~>FFnqjT>GPgodi~
z7PyG=#c9pG&%D|@NTLH6{)+V%7^m1(EuVOGcbRi6D|=4$$`lTuH$I!XpjokZwrS6h
zVIC$YK?tjkljM{b*ewt$wn)#^qMK!|GhrB^sgf_!Cq0O;06UqtG4%hjN;3_zL_a+5
zSRbqh?<b@&@<}AEfaSz%QD5H%iXggq6GWXhCxKM5hYp5MXERD61<P))!*!q$l0~LH
zOj*2bMwSwSWW|}64KE=C?(S#TV9P=x+pbwE4<?9=_x>yjLSbY>Z9Q6v&NV3#?UL_k
zMlqhOKY{`*%gYaEP<+Jg=EnSHjEJ8It{6_;LLyeTcj5f_N=8HW@(Nx;9_!xpxHReK
z6!$E7JQzG%1#3YnZ)w61%)Rh~TUTf3Yl$hfSTS9j-3)7=@tC#z6&o{^%hrW7q$r42
zAHX!U;EyKzX~&sGV_#h6`cltfpOyLIs(qaEwNM1H%|^K&UF{@~O9HpjJ|#sH)b%Jr
zq>&(%vy!6{84nH=GN$^N#xvqF!%tM_;gX1Q$s+1BHwo9bZgMQY@v|}tXJu14WDs+!
zY(otqY)eF@EzB_t({xvhNfo3tL<NR%5T{SHG8@yQ_HH5Ws@|D)S%U~rFEwfq^3E%l
z_x|HwN;px6ke+3GJaL#PoUm3PZN3f6ZS&#>E_mg$M<@U{Ki3bM(hph!>6K@Us|8#J
zu*ftRBA=fhG$Vk`XdDZIoO2`(KVqAl_2Zr#>is06&BL{ZJV@T(vwP<{luJ}FvX3VX
zh{9q&;;IA}Pv^1h&ZIMKQ!#x-_EiITaf3&odYi=-z!4I5m_dQr5j|dWiTPM}0p&z8
zl<$sV$9qNKJp0bW0n8X`ewc!6Vqy$MIlAjeXDYnk)tdZquXO*pD>K{?^C*k^Ypd-1
zs;a`RysTXzK((?iGDv5dfRnH3{;EWKAS&BPnk!h*(D^7pI|5Mf#V2IF`W0?~O~o^K
zB0rJ0eTK=#UakKWXuz<S6o6QAIOXt~*LpK0=YulQ9=J4)Kt@65ujX@SCB{R$@m~Gz
z-lAP4;IPu9qP|9xd0H+eah=&!r#ncdo1(T@$D(CbQkiL)Lr+|fU^!))QMz#Dnr?-w
z2>bW!dd%QH=x$$LYK(SHVf|_7Jcpt&V^<-iqk&`U4(~xpL!T5I+<1)|X3VWJW7LKJ
zMJiIlsz+xu6ap!luqr#?U+i7{L~cGi9#PZsYtp?D50}>*QyM3rA!)nR%M7ddu&e_J
zm6MYhdO*6w?1wF2-5#YShMK!qMMePax;W4r$g3`s9j7r$CS7F^2<kBTpqsgWDmybo
z)7^>t`PFCTIA|g5Z}7To_d4Nb6E<9{w^$eUT9GS=*6grWz=-NTUT_Th;^v9TTyn~a
zkdgj}C)$V3A9L_>n_d}g_G5FPsW>B1z)BFtEz%-aG7{?6<UNfZRgUaRRLCud5Y?IY
zPVkFIiapv{_SBfn0+#FPsZuLKM`2FSSPc!uWPZ8Y{Lwd?xo^hEoWYEK0sOQ6XL+^I
zchu~9N{o^}h3-h=e};ECULnpsU4oMtirzJ7XUrKY$J)m8l{Y^I{$BcdVFs3FnK)xc
zcE18osC&BFiOFDh`Ad~G!T!h^!a6@s>EQusCR|p?y=Q<iJFY@M@Uo&5_mF!SvsgUt
zK?)l=*$M=$NIu1#;6uk{XOzPFS}|_~@{#*))x=>f3yp6Jbo|+<=!(den)C{QP?M`7
zb;!Q-%N8rWJpS@<{6&1$lcm^}D!VXvxHia8ZiU8y53X%z+IVM&w0ogfLWU0bFNwm+
z_8);A4%Yv<Q1@Cx!u9`<?LYrXlwMH?RUnS1xDxx+*NmomxN|9oJQJ=H&kf53m#^DT
z|9=vth{nV|m4sYjPv4Z`aT;vlNJlU+H^wIomxni=sD7w~TgDVUTpXEHNw8Sb^mP0Z
z68Y)v`f7EQJZb|DQz@W`3N$g};(S)lCzr?5o%=XK7K3zDDS2Uh5q(z9j6V14!RthC
z@BRIJJTkR6n=3_(g#vJ=W6%RCUO0^<-9aeoSa$$>qG&T1S4pI5<#6&O&qN%hi&zmE
zFk~(>lHjIm<!=4ztd<yzjKx5LDy*1HK^f6hhDi^v+ea$Vu`3XRsLn}Ci4K=Iy~8^4
z=;;jJ_DEfy3UWB0DFTBQE1qHDi0*8)83I{Y-`^swkO#8Vh+HB(t`5g2sAM70G$#i8
z90fUqQ7&yDzu$XTfTd8sLNuL8BFWr*7U++>@KuxbdUw847W|c<iV$l}DoHt-mT1H^
z!YJlt7_>5C0%?6iu}0M{>T(JyjfIuWFr~p%fI7tTD&7@~LI&uhLk+fzRp<>8TefJ|
z2rU)lYJ-6b`-C)Ll+YjHwma>P9xxh9LG#HfSwHMYhI?lIL<FRRB1o@Z@PaFzFh${_
zUxI=+UsMHCzJjS!VU$mGA>lrye`g2a*cf$~V32v7;zIj!EYwi=Zd^v-_CJrSnDabZ
z<^>|NbwzM%ltA^e|76KmNME+jiQ4Ft^h0FxulEtPgLC?4=%bek+H&@9W6xQ4bxngx
zc=hDh7X6m7nutrw*eFduC$Yoj6rWgAD}nhA`Y{Ib60uO)KUPZ^_@-@3k`_HsBog6(
ze3mnby}q|peI4m%ReX@{FwuE*F$FMt6Q$)SB1qy51ekv1xNfvQFee)0=sd&SH;*6O
z8p5uJUu_9E!gtg>GX{|47zVmlBP>ZLj2zz{$S%XZt>FQ2q}ztrObyYD(bg5+!9g6+
z&oQe4_~=ny#4F8ZB?swB1%qpWLP`>{C4nzw1pec{hI40=6PfUnM61SeFAF^tACn$k
z8?^UN`DORA3m`3dhs)%z?-5(dNyBcBBlOhEh6@{W6MZwEW*DtNn%_}A+lPPFhjvVr
z6vcWKJ&Yy(;;xVNiD!fbzw2frvTNgR&8_!l6<t>h^~x2WW183B$4o2xDYDh(sJg#7
ziHGHRE)SWjm`B47jXf0&Dxfh@_2a~~a)r<)K?Y2*C?AT7{5YmM@Zv&c2lwShnW^>Z
zdra3IVSa5J=w}}@f`87u<~atMBkhFG$S0jQIc)$1$?@!(5H<ON)gG<tAv_0k^BkU*
z>KGBX4HwpGQ-g_sZ>TuIXj8w9sig)s8bmx?HB;JQ^ig5-UK<yX_7^Bd<#)E21T?X3
zrF7*mm2<e;?M?z;ypq}3u$}fvqeA&{o_TNI6!W0K)pCf~x;ASkxr6i?69h_XOo--G
z32!v>3C&l1)<a2O+6W1t)*KRNBP0LxhGiwuIvzzY^mfr!OcE^7cTw9ssDzz<jG>$8
z!uE|JveS{Zz5|!)>C#siHd6Sp(00H7t+buykJp6VSFfzzZnl^$@U(^?b0$}+v<o1#
zoM<*5C2-|BLy!$c@3Z551q^t8GX}_D)e#VY3WOLjp|OlLozgV?xab#*#fH^MvRc8o
z)<0z^<Ij&dzflm^4=9=%nwsP(8F;dwh<t53(pL}UF-NTnC9itQnr1B#T11K5rgJRE
zf1Egf{#xiAe~B%TCQ_+nGuDfKC3ni4${b>a=qA$u&9D4=K{z)zgzdCtke$zLv7kI!
z+GI<OnI69(+Oa!A?GHcx>Vd?NUuWw%Y@{Mb+J;>aVRq=nlBrLj_!~mm0!|F2)tc%x
zxgX4BR#OSs6Ngov@VZ^*_>}jJ{j834nx+2_GEyh}$RHXUfT}}95JzNF5Q*%U)=Zm9
zRPGFQkjKPtuj(qGC^z#M33Gq=+a`&zLVuAqD6}k(hw-YJpD177gloTRITj@xEO#g)
z!pQ;I2*gtob$;d-RmgZSgR0XoS00^x)dwP2-gy?Dmf{IVR_6cvwi5_x?6S^`g{{cB
zH<tr7Jb`bRlf2JS>1=npj)mEk8&_82>0_H{qotI}m74YN7)!-Bd;ykx@2ANk!qWB<
zpAR>1KApS+k$SznP2k*Iy=HFdAb~EFz%M4~S&wgq$X?~RDMiflI|xb^^96_<4Jl@G
zoAYuK>o7WGQG}K5=!x4bu^8BBvlT1yt^Mke0F~h2IvIgC7vl7IRHaQ{Q~4tKT+ueJ
zy}xZ=nc)pYFqd^RAoRkoIdm7^D@Qbu_IYJ~N%KIcm+~~|U1#zPLyR094q)<{$h*d+
z%hY@`SgmqC2EOT&)F#(bURfkVdoe|gI5pTNsE+K?c_?xni~fvxFsui2of}<(TYTFK
z881xvLQuE%i|b(PgH<UWVNP5S<tOfn8zTN|hOY{i4nz3Bp744){9n&8+kfdvb{3BR
zRZniKYuWzOlRmSxbDKC#hVF=zNb2}4th6mWC}UCiLapG-@yLWd(~28MySIfD69v+v
zvRqDmap3b4^ZV^SyMz`F*>gwR>#>WCYPOiH;e?|0$<n9c^5Ih``RURnYU87jg(T#s
z<;qSEKM&Kx$HSSANm5+z-v_1VoQrPmKdVovYPQUb$_WyJ#RGV<q57(4Q7)T?a{b^3
zl&48{Rf{lkX^TG1_7T<Qvc@x-3TM;O2~39OH$y*mhT3M3R;U?_<YpAQS3^HV2{$M=
zFEbFc%(iEPfI8Z4kYxrCer$=WjMmb~$x`%QeZc)pm;Ns?K4CW%VR0~cM|Va?fb1T-
zzys!~4JOKZvQg)_-iS&d20c=m+u?n0{E;(9bQnp)1%4!y?%X9b65y(>@zK`h5TitK
zPn#hx;F~M)4M`-c-VkA^{O6GV{EjpTV#_8+*B+M@UFz*-?s_!rkqh7Iv)hlR-@~*S
z&7LzWYT4qBO3OAX&R9NwBbKFFt0oudO=X&OoxA<{h)f!qc^0bace6?HwdV0pTGsbC
z^^zmOU*Dg$=Ehspmmi;9G<4fjQ%&k2GNwDZ&Ugbm1pkV+MVHBAgAfH_uz8gcnj{tc
zwyGVk+&*Hq5KDdvm<Bzp0Grm6OxDHoAEfYB4RH17rr0zKkD)DJVD&OFTSR=ch$acw
zkOd>3y%@Zce0$B14Ij*C>i+`I)~-y-R;;NGlao-aB<^6nWu3=M;K#W2abl^3?yfDh
zfUPAZ`R=ON%%?vWTiR?aV1Wi;7jQUYU35;Hc3a;8CL*KwdA8^h4Ry11Rps|;Pfu(Z
zGE9=)suh2V>BYZ@{YA`~mO1BpoH%_cCu{7y88+3a&XDcGx~kfqY2R*d_R;5>4A0w0
z?@`xNb5@$xM&;0|-7c(An5wCkSMn3IX{#*HRqGK@b#A297g!}bcG2jC(X!T$x`|q!
zECZOD=t=aAK9WZqj7Ap^9mc8<e4Ui`?Ej<yun?4?ALtIlF=qsL#B--JDyxN@bSDZk
zd^adWWMuK?imQk$sjyy6b)NcAHWh~age099JoL_T<^bJs5OgMYxg=Z)(ys)J2d=*@
zgwz)*;b<~&`ll)W>AeEEA~s@KL17n7zi;l_(J=8bMN@}?raESj99)%|!*SC+-Hpml
z9)7#@^<%DMzcqIl;=SD%2u#bgV=#Vb&$m1ORC=QlyI_(`1<>2&Y)0UUVn$~E$beWN
z7DZyMlPlqx!X^j=2=n^lFo8JOw)lve(sX2)wY+;46a_hnO&;7dh`;GFHK52+5xW`%
zN9+8cK9WUt_3FO&zm09C=zxCw7ZZlH>3Q7(5mAuXjsnX^rz7L^kuujLT5kQBmaL9r
zic7TDYp$XD2B86%1`)w4_@a?t6q;qo^Mbgd&~>m20dXZ1qKoyd_#)gHJJla?!)cyJ
zv?F&>_xDF^HVBSDSU=D|8Z1*dST2gWcwbvfENt$K!peC%I0(Wa;P77@n7Md?LVVAs
zgjkB%<S<&2H=8PVZb9KQ{)2w}$>3+0eL2<|%Q=WeyiwjzQd!bqip$(B<zybAh^^po
z;D44paYr!=*AQ+PQ1ZsMs{HQb0)4B7yLh&n6w^=&Jo7b^kq*~?`}S!9F@gKZ9X8(x
z1We+laLZyEg4p&RG@e$VM9yjX@8b5KF!~1l2%*w?(t_vbB`Bv52`1QX=`>Yb<qZ@n
z%%D!5IeCU2l`9K_IGqyALQycCANtTp3F#0ua6;<)BxFR@%QU^|TAkgoNS_-yClImi
zg9e?LZ<>m+DMO)5_~jK_m_<A7<SRwxKOxfX;`qo?_G92f%YyKNHA{MY?r=r|sZ*cw
z-D6a`DjjS^ZS_71Clv6e2h8yc6ht3w_dml#JI_b(cWWhGPj0BDnB%a;KY-`1Y_wd#
z-|J6D@E8PnJAS!%6Ugu59lZW8(%!LGv~Ua7T()i7_Fnd0wr$(CZQHhO+qP}qeUeVP
zlY2g#q<_Oq-cfUms;5X|9^_*bEhfdd^d?R3<Ygd3wSPJBuctv$a!B|Y^GZt<#~1zo
z_~NXgBBo=RJPzk0_PREA?G8zx9*uybJC^FY@Rq{RE*=pk@=8kl!gFEiY6Qm8IIe9n
zUt?cWcK_L6Xw0fZ&Ui=cQCr%U`z?QJjmP#qTu{_iw23VF*wTeVbKbIFi%Mri4Qo-F
zy$C`kKN4elB?$|ElJaAuH&`i5V_IGjsR6pAeP<AmN;!r53U_GJBOG6bc^XwBLP4S{
z0*fAbp{m0@<YA*oT$>ELVM-sd;RO)BE*_S-damD(6N|mUL@1&)y_W_IV0Ko3dD4La
zW3xBd$n>7T&$aR=+W!{zWxdWtUu^URIQ8Si{;yN=zx=c^u`vA4Dd=VOzlob8i2t^(
zdtucYQ76#?$}a?+Rb!Iw)e8@CN!h^S2JLqIdP07cqqVZHQ+wltstzdRoQP@WPOnc-
zhnHm0E{M@TUk9Dbk*T%CVN40;HHFLGBu+&|WJ(og&Cx#L>3gcay4~LI8S_?xBS{1r
z=6VL@J-&|@+ZUm)s*tKICP`YPl8bMjxwf||x4PeMoetP{UY@TXr>}h5gp=q?)b|h0
zBV<*HIj#eW229r^T^g;@&R0;Q=hi*uE5<sVt=99Xq2B7J&ZlS@p0AsK+&)g~XyJxe
zx|5z2?ajVrT*oe7dN{APDNS4cxxMG&L}`#B%;ps2Ws<-auLO#dk8Mmgiqes7Jof7t
zCYZ?|`nnYTb+Ae?<yo>LJ6(0{-hi$QU>)%#h=QW}bZDGiK9Uu<a7an1j8kNgXQX{K
zuO8Gn*RqWb=(?@4f0Q5dIoceZ!8smp)Q&?dm4;T^a)<FJZu+flC%sme+>d*xD~JT@
z2QmBSjYJ_;ry$qRomV@!0UzCDv0`ClNMQ`iDh}>A+IDJ_Ci7_bg^8E>0<si{q+dv*
z8vN&Vfm$ZRj=p^y@sIKaOt44=f#`22c4LAL|3D7J#{}M3l)PWDfySp_k?MrLF(hC%
z@^r;74qDm~VM>#1yNRBcqhg*_yV;iVl1|}5WRbGU1ND?I?#v<q#$criWKulUOi~Se
z!>l?YCrIqa!`i^oQK(~`KCgIaBm_M4Q2;K3B*v7zy<wn$vJ1Ckc?-<P%_Q78F?^NA
z?W6AJUuudc2KlFx8xb&IB#U#rskrmON786!@u&A=_<qoj29aUnuOYbNYZZJB))I(9
zwmxgukH^O001E~IDtn;W1~Ro_MlZfVdQV9%m=v26j`fyA0Zi)s$ggB00U{Q4RsN#M
z$s)n7CnAw`-TM3GF~aIw*v}M;-_fIzBJ1>meowSbaU_5mY5=sd7}zZu*zuX%Vm67R
zj1$BJPM-G)bw?#dL1gEnb*&D3A|}E(PmwqP@DMP_>LFmymlE_Vhmm543$>AeT^>q&
zMJpq7WbhlbYtfyimOw6+$nvK?w!gz4Kdu*3B^+b}+3fiqCFi{5?ju<dFN#Q~<CBq6
zHnvkQ|1PrQS2#w9mz$#d))$T0DkxBVhKPWk9TBk_=Hu9>%f+n&EU0O{+^jhpcLzpM
zz#hqGn(R~7BCL(G++=72o;&yX)uGqCb_0|UFgmcMK49>+O*ckiT--j@>K}2{t$=ZX
zvzl&OUReO}SVeZwcFM|ppA^*mGjju><S=@PLedd73S4J;JN`M7qi@2*0@}>+6P4vu
zs&iJ>TKJm=1F$|2;tPG23}IFh=KvOf_s~M`K7&H&23BqfB5ZmS8bI9tZL~xl5(fo_
z!;>+0HBmaCy5%g#w3bDj5cC1%J#-u23js5xr6i;LmZ8S8KRirQlz!=9GqdEJePK%D
zHZmw%$F(#Dx!ev6ATuXqrfY`5s619{gCYA6U1TTCth2jNW025?1`5)hvkp|yKwnL8
z6-hmYufqTKYbZ-ZZkC({=a9!bh7y3lKsm(W$0H>e1O-pYUy$yL_MO0BH}Iwn+zyc-
zq-XS~WYuwkAO~g#sRqeSIQgZWY#R|;vVn&74L@wnII=&dp$CaI$`MfUW*<ji2J)^7
zV+s0I#CYK!Dbz62*-*Z1s-|matI1DszY#ysuzQ2YGsw2<DPa9t{t)TjnP*YOL%ou>
zb`0n=#AyUPWmNVRE*K>gyU2qoXO{;td}(-!wyg!0n?(^aFY=!M{3)Wm2Qt!K1+961
z{V_UP5)#dHdo>;>aM*hBa#qERe+7UNY%s<Mlke2=a3d929K!e#wJGPByG9y;_oCkv
zlznp3Cfh;wu7#5^%wOwmQUK1FCdjtb<FVsh1J80__($nc$tH58x+>OqnY4I!m&F-c
zL2{KWqg?p&W>l5T27H@FsP#;1J@KH-KypKuJsFvnb%=yzKxJ}vhVe`NRh4F@775Mu
z&W?gz<nNtD8ou;|RY|}n44OB~^Pd{+NmGB5)GGpHpUQ$A|3|F=2}}~!Zn<YRtKWMw
z6c38>6a4pkLUCt6z0uGzRUO<&on<Z*G(U5R2<uu{=*z>bgZF#?^})S>UHw4``}((2
z##)WntzER4I?(6!e)|kfq;&o@+=<wjz7s$-VD^rnEs#Bi9=TVxwWq<tgG70lV}TK;
z4XmH1tVPQ{5)-;@d;`S62|p$x(uUbYd^|6OhqdZ@c2?XPj!Cb(wS`Z%V7Ky^&y8BT
zDRDt|NNt+HK-ocsi?=l{@IJs?O>?+yp~cy>eedknne682$`)>W|2#rmA)pILd~Ect
zvBMq2j=UbGd;7q2l*6zRZ=7O9`4(KT%u?j&{$$w4#0MQ{(*630%jij>g4RSW5(LCk
zY}2R}UzMXw@K(nva1B+c;z$telZi@GQ1^7A8{^X(=dP_ryz=mZJ$fV`FFG&B^++Qq
z%#7(Cvse}5=$$y%)FERSf^?X0%?>_byDZu`l$i*V9$7J#7v3Om7=OjRUt*sHJ8Yog
zje)yF{(mjc1Q`NKSUIm0YzW!Oto=@={jOL5yNVbVBkH&9`}b~!Qi@jtC8DqDPgbcm
zWx?WvtpIOatPF_`7W>L>T4c;10mSQ>3`g18AJggUjq^eSpZ8O=07-o<lIgZ~sJdUH
z>3#33TT`p4qtG@czn#%MpNA+Dl+4W7jm$z^ydN{u%wkMod(%=v9aHODda{YrVGKe7
z)h?KlivVZ~ifXpZ7in^^<Wp|qa>;D|4O4;^U(Vr4gdOUwv3_scnbdOXS~O)~Qii>2
zo+e=i_pgngWrT2Hy%hxWz!_XOcphxUzXh}r`-fj^tjdkl?C#FEbGoKOMT#^7>Rj}y
zcRS?PE(+du@`qv5sES{*l@Iu^TQC1iZ=DW%rr=^dTXjpyE$8S2MYSZ|BMt%UglozX
z*A2@aLtzW)l9Xo9U!Bz46D=&OHn35S*J#iU#H(rew-K_IX2V=!eUzzbWme}4_uB2g
zRaboDcM>|z!}&Ay?-rguHF<j`n}6*B#}xYr^81rLEUam*v0O{$Ev{_rM)AZZ<D2MQ
z+HHCKR|G}1TEKA5m)02eD@@Iwm({zTr7lYxhv2B2M<*6FPeqgwEN3**e_lhEBYgp(
zJ)j2vn{H$OFWU_!c23U!b-OXF@o&2!iuRxOzQ2GT>={$hj^^@un+2&fb10W?^?hU_
z{Qnq?+Tji)8!?TkyFI&hCT4Kb5;j*R_(%$e>CC4-J&q(9*`_R?_h*}ng{OQB7R-OE
z4yd@jQz4U46%eT+ubkq$XD~@qaAu)I9#wt6-CaDUq<$u^Zg+FP9C!{vr)6b(+}|%k
zb!B{foWCy5b8~+QhU3y`AZSr;9-4yeGGolzcnyOby5^&d3kWhAtMoz`%o`U`UvyJ5
zJFHHY(O@9*AJD5i6_1}+{zQJ6beq$EDKy$#o^18Dq8nebE3GOYuG(C*IV#&^!JIlP
z(tS(%M*q60fJ$=w{!rOeOuPph;#HNvECygu&@yy&RdnfGjPPwn*41@uOm94287^!0
zzC=DcH(IkBZ!1OCT{dKiL23S4RXjFS*H#`|^E94|D@`_3qZJuKi8geTDU(H$1t;Z{
zqvMj}1hBXMSnJqy*YW&&OMTnRThZ<7|Cr<m;AqfS7^`(dXp)oczdN(qu44b{fN@QG
zG*YEb+Fcfels&Jbrpcf}`Zx=;-Cd<j8x*y7sB29Z2Ao`)(fztB2b_f}8Q#{U$gBTa
zt*}q0?y$DA0`D3nA(OhZQBzQ60cMBVc#?+zv0kzb%!ryKFrWzrd`e5fmk}Byjkl4A
zTJ}bI3>q<mDG=d&v(cdqjYN-UU{+HSDeYa$A_JO)`9o|W{1pkSC#c+ui-Q<U+bl(#
zBMRBamOKy2iW<x}CJQ?r_FI{MEfu?sEvQdh5mr=_(iqThc)vJ%u4U2Oi=BZ6Rq|T2
z))pWQ?w@4xbbd;WGw)m4<F(FX5tBr)8BC@pUq6jf1Dq*h_!P;jOXej2Dn$#*6$TNL
zoRmfL$DK_>{LYNBO_<k4f|8Cx3?)$#FNoj^4<b+|nEt~*fLWM8Q8i5yFNx2V#C2xb
zhKxuETHv=n`knz`8I#04io}@U7myC4oSpq-3P~{o!YIWX9G68XVa`bTuYe{su?!>P
zopb9RQaI|yZ7hblG8~{96K~#HLaIV<6-NuWF$!zT`neS6lmZ*=L5&u<qR6j}Ip+F>
z@RECsCoZxa<2Z=3yN{EW77wfF!Y;03&-9Bj0L3(n$-$OQIonzB_m3Rn(tgC98>WGX
zTNsSD2nR|4vU!B-TMVy$#6}@r^yXEf8F&)S+;@@-636kLrAnSVzpvfs3Wl^dyMIYf
zCcomYbY8hFSrgQ8dsSJOyhKyJbp8R{3pEKIpl&e_!PR0UUQ|gp=$#id&!sA`pSIE}
z>)jXE6S(vlNXq3&*G3e=&x^x04N#f&Y9a;O#u$-yInl8tH|e1qnwpYiHXacF?fg>?
zbg<sokk`Y&P*NT60)>!vblTpxH##zs$uOh7o?}s6eJmlRD8A4cGfQcBYIMd7KefXP
zW<{Ia8IX340I_0T1|N~=UVcy~OaYC{#WpEjebL`+3nSQ7R^}CV(<cK?Go5lzH6`Gi
zR^1Pmn}&_1=39_?(4M~@jA$Aid*k)_m)V%pJ0ByqTH3Um9}e!=9_N7J<Dzdh-bU~1
z*KK?#NTPZ90#~1c_{5J~47{;k{^#LQ0Z`cAa@S8nBY94d(9jZ139%1@IVwg!kY9HQ
zO9lW&E5(572qyTeIgmmiHAq0x!b<B&_y@EiQ`U9n?Qf&YEx4?Y=C63Ia30?+7HB?#
zcJ$8#OdzfDvWn}|uNV4*sXl8#Tz!#5*nI4#>ot7YUvJ^13t4`0Ms(vEN-)P#`zzZ2
z7Eq8kfnIIp4X$#(<PR7lW=RYj@b(!3)=!}c$%!22gBAWYNf3#=fOH=dv29+PHM>M~
ztF}TMWP=^6@0Y2)c(W+Q@1eCH4pV^VfwsPPzF)PzZ&LRmXe?>|f?mJ5`Z?o9$3>k1
zF}GeP<<`5l8e*5lzi-2j^k0*D)-TvM=WuoR+5{s9y|s9UJpM<*11+wQZ2&F%DTwEK
zZGWwNJIgY>NmT)T&*#&%NqKh;0aD}6(`}^4lyy2nXXq!T+bcHUyu{i6IESmBCjXwX
zUv^my<PR6PK=%G^vFV0>P+o5Oxn)P+9z*qQ-&h%?z&QO$jJ2~$Gqh~eM>I_pEf!L)
zq0O)9z6_~YHo>Nff&rBn3L-462=BtKY|)0Lc|hM^ljdly+K?lTdnB7fanR|Pni#zf
zstD~8k{pn8Ob{~NALy$ctIgP-a)pN%WP0LYB5f_qEX6fQtH58Zf*KA(BPyQ0(F8^S
zK{d9DNyAFG?-mqQ6z0ajD{b=(qK<P7l8)Vy>2&A<bXO$oN+6!$6p=rAzHm^Vy<Cy@
z-adWo!2XIXaCB%*IIwF)ijUfKYA+{DK%?7)SaerCbai&Qgn;>y;^uRV1lH#0hIIn`
zB2WgP8SH=6<K+cW?D`D|aMeF2yH!Sa_yck1&SeMh=z#Vb%MFXi4a+_SuL~$_)C9aR
zc&J64%h0>MFZqor3?&=8r2mCqWe0xl;tsr+Bo3>3%CF63R1WX2y!7Z_iOo?Elk@Xj
z#1eqPH~^stD*+fH<k+5$1qtw-XW<CB0<sGYBF%2p4PG1~B+}F=^H+P8g<s)y6d`z+
zt%iH38NAE>wHSHVms}DrjzDbC>i9>0PiO($F)bjP4DLJ`ZMw*QyAeWqRTg<8m+-Ho
z*(@PA9{Z0c(748SGULovS)j-ph_vLBY7~P}xnEA#6Bi=hF>2S);5O;6Lrt1*)?DZ3
z#z>lMmn6{p0c-mOAM%^GiXFtzmXp?NT;-2iv`31QPC!k}>kV&R4tD3p-7xaSW}nL!
zomGN#)~+?x*^;qYT-uwot8_D^nNB+<d<l{XLas^QWD#{fXMIa%9H{iROn|qjXDjI|
zRRaDJ129obSYQSn`#MpqtZpi;-gv{7&1Ji;^~yBO0ZO8)Id<i?Yd!!9((_jnUn?iu
z`*PRK0|y&zMN8G-CwFeGs*kCAizrnO-02O*Sx*&Q6$mC+HN6n53fYp*YS!p_JWCJ5
z9_Ll_>yay2wpgtA4M5yCrjiBdGpXqTg@qWKs`vtfy~EK<A1l_1po;j6+nKr6^M-4-
z^LMgL?2Uxehg?DDVT9w6+#jOp5mAS84f*sNjU~S!xE7mScfyQkogqpJV##_5N|$<a
ztw+-Jnn>KI;*wjc5I!vmm?s4@A|(I?_RJ1kSqn2>C_whTC_eB!xE(Q9?C_y~1g?U^
z7+PG<LCQdq<u6#&B@v+PEC^zB4x4lmmXta6q(?~D{2mmnXXCm=-pn6RfO;uDJgT<Z
zU2U~|B#3-GNYpuZ1{QZI06F`1Y-)%l!;_c&a(qZs{z8T@`?k*b;ad>Y1>L_8t6vK?
zCRo;#cqWsk-e2iU`SIXwaPo7{?hRvq@M5a1m|yX=TQ9n`N1*&tkGZO7v;dl0e+JjB
z&pb2v?`wXumE?Z2!>=5!<H~U_8ir83m7S%e_H{}Dx#R~FV^i9sVsJc!$2y<7A3W(t
zpC*D|J9lg;!fgNKMO|<3srt->vez0uy)@p#tmYk43wA&r2I$w^Z`}6ahi{C6@3MdG
zn!5di<gj7wLk9Ag^1Z^q`)5quH363ay0#pjbXk0JT*c~=A6DN#)glOMGTU*VDsFyw
zDSsULre$>oFL$*6Y=hh0m<9@S7|-+5s&k$T3x2dAo1+;85CgG=-u~q9y-LNDq{lX`
zoI&Qk;|Hf%p#E0@{r_|ES*@<=y4i;K9|2vs#frEEn3Ka(#M$1o`Fyj%_1UoQI3ZUz
zltP3X@38XzjH|~Vkwn2Q{TLVL7ZPZD!^|MP0x7+t{blWbe^8AgA<8uE+|czmNG541
zNHUp5b^I2Z?C2PO(znZP(u|GBP!ge<g#l8jcDMJ#`2FOB={SN8gEX~#!Rh|v^m6lf
zy0TZ7{vO<|x5xYQaay)D=NL4afMOg*m5S5nW@WrR*QV#Wl6^8#79A$081iorK~lZ2
zhDe#RM1A+&ryzA+Yxo#HG|zhb(apiCEhuN$nX^dHZ#>hNS%V|8b$^uvjhTAn4u%xc
zAWg;laO9+tV#nKAgx4+?OVrrNiz22z0Pn`J?z{_fO@v*}Gs<<T_NzaYM|9S2Kx~5#
zhIPrCULESsboOK!A=RQR9!yOBxt+s0(PlyUCS7eOajqT^%b`6IshoZyu3VaFt>~4a
z3FmwiIc2J_W8kRMyUjHh&qZizqL3fp9S)@^LLzKOWk&O=fM!i=d+JB++x71;xLAJ*
zl?2BkfSRA){8%o#aecoZWybbR!g?@u#|2{R!*cJd_3FVv&HDb|*6^7=5PWPAvjNIm
z(hgB9I2Ju9{r=Z^DlNE>Pku@fCQerRvfO838oby&oEp3b?R4KNt-oKdIv-K5{w2ti
zwDJJP?ZLOc7I%Mt33aQqQd5;9BTr`dh6_54BBKXiUu!lxv_nBQyEc=34?KBI6j$D!
zL_bY+(A_uNahdeeRd#|@#SxcB0yhRGN4sdDi?liA0Ba{}Kzd;k7z9TH{AALbDZ<34
zgG5CIG3+i(E&=rXKlg;*8Ycbm6WqpK-2Yi<X>Xcl2$Ms9Ot!r!U+f3DYf8tca}}c1
z?IcAY1h#5a#a}%VcfrlS1@~h8X!CCKsS999xb0XiNd{g|yJ;)wH%o$JqqVg1DR;rU
z*_^dP3<ZIDr``svZ>hB$CTIdU?MLn$ba`2{Z^z9Zl+{nQZy9V>0}|{3&uQ=)?9G&!
zq#=YrjIFY6Alyhnc!S`DcDuH17i%PX+q4()6`kzfqGV&QpeEw)#xV9p3mtqqgA!UH
z@B?k3NCoP;23z~@zlcccv~^Z=fcCvx`|&pbCFM@1N(r8~dj?!<`G~t937F8|Tgq48
zgchv>%>SH9tJiK;680VS(+A&&OA)+yH3NbBKt!1EGU5N7ou=Q~CY?mvYn_^05k+qG
zoZ_^z<9?LaWPPRknagai;zp^md)5AeHAG0;0~XP~{R*$Sj-E!_G&!!|yhYInBy0$D
z<!T^P75h<M0R_SAsnwDfoVeEsGbayrs3;V(dkRzS4pX$8I{M3zZN6yQj&JHEns_D0
zy;yH4%_(4+e<UaQlO^4iGH1mR^-5wfL`s|jm;QcY2bZKV$nu~ULOvJC=!nO11~ehZ
zRbWAg`>*G>{e{8=-1O!o=1|TP&~d`6ewK`!DGZ-XoGco$I2yT-R?fjdgk=bva2aPc
z2~om0LBqh>=p(lbj@lp2MZR2kUErM_TmU{QZ^Pb^LV?*ue#k8gfjdLmidh7NHX*go
zHF(rc6l}>)ij24+^)VS)$5SqeV4Nhpm<ISYRD}c;p#P#28FRz3j9~VJl_Mh1P!rE`
zg=vYZLX=0ZT!`<a)esP7943WS*!{*4(ushVP>?Vn#Qh%5#iD5Ja(#zbaBsm+@r*L$
z-ERzn?+K$3a6g%{!ITIniP7Qtl2dj;j?HNHB$PSFD3<3uE5W@aJSc%Ho?!(_4*n<`
zjTPqt=qd^dXt+QZNvsCW-l^hWC!Aff-EM&-^5bE;U4y?867@4PmE3H(9$ngqJ#@b`
zAuh&lJa~u44TXg8i`CW!`5{^rlu#Uj!XM<8q+lnY<a><Sv3VF_P%E4<of*xvmy1^3
zCD<}uB#be+?<QL~aH=!BDn!MKp6jG_!qFFh@aR1@7s)e(V8=?d@W;0<%ZiG39CU4F
zLfva<^m0qDn@fA<2Vo_s%f*B&1MjS8Nr+w336v%>9+D*#C_jd;^Ru82cdQ)bX<!y?
zdqH(mIkEPqS&lylQ3b~%rZh9>1-=7jb@q3<>PO0Hsd}^C8>#T_#}{y`0>N3JsDg?K
zHm}O6JF4}|``)A=5w)wuD*6){eqLT0baO~=5Ak~V#3oW@{mn!9{ecLaMX4sM))~%<
zr#+~tF8kzXs4XefTld7$XE#N+#j7B@az%eT*U)hzU@}Li@IFdb>FvfWTiOJ#aquxV
z<t;_c(!oX*<QZZ?$*aAwGbQcQdUmcKtUWZXT8^pU+=Y0Ws$1C-)tbL#e?F;tRHYqb
zyF?p7q#WS{$z;NYOPt`>j>Cd?bDXmM5+~wVfMm|7Z(p<Vm*vn8*G(^5rf=$P6G9Mq
zVrRMjfgWTdd#bTu%kEI=^nSu=J`HZX?|P`VEzI;mr%V1^J#e@V4av87%ffv7=p^fU
z?BEkGtvL73IsR0V&I5J8&DdW+YL9y9Ot(F`FE3mb6-hyHkoY4multylJ){Ze0~QQ5
z&<KW?_gDJ7C;$tFnC3$+lj&se%nYVz_;OzzdLo61aXZR7D3o2<@A^haf+|h;yna}T
z%8@uNs{?5;5viOqR7l<Ow*&mP5|oYQHP-zgJ-zaoH3Z;$OF^5p@qsT`1HFej=EhYV
zy&SBjcOLqqqk~V@F{7qQ+EZ<>O)x}IA}qkL&ol4k-<Tc4KLu*K0O2iI!|wp)XB*^V
z&9tiY*c!yORiT_#DuM~Vt>_ACF`9Ae)yYXH!lhi(TZPu)hNyyLn!7Ogk1N3G%g@T#
zOTqrRj7v6Ev#$$?-@ROj#n=9*&X~QQD7h6?!<nOAogSr@_vrvulM>1H&Vhyz0)dev
zWSClYNz~o!%x%LD3}WXxKfXe+4X-yu_u}EEjwyKe^v~(w3M32^8U*vk?&3W@7wyAv
zMkz_`+Ea%VR~Q+9LB`(uOy4W;sglo$YP>2R$^&IESVWFRxk*14HjPX8m;}J^JW1lU
z-*LR|+`kw<vOj8bltR&0!UZDOGKgp_yL#g~<)QkfsYrZ(K}pWZ<2o*cfbsDYpGzyQ
zIYB7>kpY}ab-<`h_h{&_O!hurDl#c4DYVN?8gB&Os13(_?_KFlgvKIdoVpI0MKl--
z4&^iEP0hG^m61l?`~adF39B`+?I)D`g{JibgZ;mC8z(#a|Hovl<Zfq7Krd@xq3C1{
zMK43}uQwcuUew&d(TRYQg%yha|49)sF*5(pG}dhmYg=q~q@NnyetK{c{3+7VT15=d
zg_A9`HsAmb&>Jiaf8uJzeXD9&4d8a)T~mtT`H6<+1h&&oAp0Rw)A9YOM;Oz8Eh@$+
z<b_p#-<~286xkXbLXwROvr=U%aXvUijwN`jAKVg|%hhrkB^hHmz=z2WF>llK2>rs1
z(Pr?dId~ZJ5s5vj8g|5qsj=4z5tF6WQtkt$MaaV8SPWbuU{p-a1PkCC<&Ge1$KoP`
z@Dz@sSrQ8N^1&Jl&g$`}WEjV%5V4yIvimJL?;&Wd7S<@d*GgbXs>^VWiO}_c9u|n2
zh?>~YY=skxf_p#^p`EoJj+NKwjL8FaIA3H^#-KEl<*CtF48YCXn8XCBlT#FTP_OTD
zUxy<=6u~ivYw9{E3>Y<G1i-o<*1@Xyi+KJa<fQ}BCLC1yHROYEGz7qK%l9Tol9cY|
z#i1SijlI#8#EU^o!5Gm#x+2&NkQq!+rjO|p(78DxRp_dV3UpueS%l}e3X@a3-IDdG
z0{n@g^P(cn!gFE26352yyyV8ib@k4VyfBkg)ceVk1?M4dODXVz%_M7JM4OOk1yVUs
z-Z+z`0Z^jE0{@^vh#FrG+12T|$e08QzsQ&frOGmCj!}VYnhaJFhC%MXW<@|Uy3dq9
zapr9Ug$5WPLs0kSS+&6gKuw|&g*9slV5#|o98Q7~>Gf&>e0MedIXRgsAT@(LiqwJv
z<+gG1;Hud9K5bE1u3WD&7h8Vc9&O?IwdVR<E&Iq%m%VuR1Xjt6yg0o`y%>FM;fbnf
zeQr~}t$yT!P^B;nzzvHs$%7$TAY(q#DPL4n&aS*DJ<%?#SiKkqvX#k9kIB%fxb4N?
z-@T8hTLg_@GlOr~JiCVkmRsN*CV$quUVV0p9^dmW+^M+0|NWANjM*<)-+q1`E@aAs
zG`8Lk(RR>b0UB@}mw(W}X!$!MT8T)*<xufa-)?hq+M0T((O&$PIoVQu%G~VnS)03=
zJF_|2qSE3NRncg&f`*+<NEARIawP?{V-dm}8Fn@OohS@EbU|+zM#10BBM^Qu-G!OE
zY6qj@<6TnW`h2RQ?d1J^@}T%9prWxQrx|*$Ns1Ovo5l6fzMI0n+v&SIM~dFpjJGlX
zkUqZD?Rokb=awqQoKkhKv9F?QK}sh6A>HlSYT|TQWzBjJlPI*|mBTUI382(|g8)xV
z`#^)hTG%Lf18C!7`P@9c%R^aDC0n((=Ue1Qhi2%8y}|VRHgkMZjABkGH4pY6UXSOX
z4frTKh<23A0Y@Xd8?$|Z4pV!EGqlf-KOWMnY4hFf^`4qkVsk$rP+EZ7w-(yHme_1t
zHnpidGLGRD0>M3S4sj@!g+!kFC<$fqU$>SP`3+Y$ZQ79v3TD{QSKQFK91`&IXM!^L
zm*yjnrG8~UV=}+m0*!e>wA+z23a2MyI^D+m(_uEpFTK&qnmu>B2jmWf_>M~LZsvv6
zhK|)95np&D@%Q$#{pcsIuPe_M>;|wK#`;1@Xgb_8!zMXT7yCT1CFLI+@@OL{ci)xy
z$shH`?RbzN>6pgN>H8R%g)f}uCGOe2`&wSUwc7W5-9sH8ZFZ>({jQ~HqpoEIwJ&(Y
z6MjVv&NntIb;%yyk)S=Yzmn>dA4wGs%V9!D;8pXG#lQ|`w9hc^rS>ZMWGfWBIYw{^
zvHLO>Q1;_{$#}X7RX!1I`zY$h^kv7*N!jvKgrMDl=~uiK0B6TOd)Ro;v7upbHLD1|
z3^mo;0hT-PtCu15CZ=g?2oa>oO-DJzGCV2bxHhwQb2%%xh~y7(^8NWG(*ia=UgNpA
zYW@+uOC5HfDp+4$%Ph1hmHUO@Jas*>o!+f_CP4vhLUb4u7|JbNy!4G=Tzq3z{UW>p
zrdy+TTgU2l8{9vNcTp{OQ89a%^L3@RqNG?yIU19woW_0XH$<G4=x(<b?QV%5cF}BY
z|D5w=-;@+bTSBSao<Wm~RF4H+7HxOMp`pUS`5KS{J@^FLH(BY;LVB(OK>8j`?+_XI
z)BYb?6!k9@<~WQEc(1IVf+jInb-&3Ax7MDWZ#5M*BjjbdrXA_N6=c(H^4i<e5myIB
zI=agYFKsm&@5fg5%Tu<zTd^0ro0*q|^dni)8<Vee9JnAq4YVLlznRu?&BL&Wr3?~G
zbn*f{VRP&9zuIQn;V<N_ksdu9?@sL5OUuj7j_Wd}xiU7H=T+NnuhjAV^09aOAKU(X
z2qd{rb;b{ct}EHlbOSO+nZ<z22C1z<>1)hWZl45GZS=`Gd^ZIeGYE{_5P6BwuUc>l
z^Kx6zyoZ?*0&_^*hbJc|4s7psf3lbHGd4*VLfgh(!kf1Q)f%`)m6Tpnda5h_x|Q(i
z-Rh{mkx1CNg)}xmpTi8g_keTo+}w<&#dn@w@5v^p<A-|p9;Xa0`mE`;*17)K+S?v!
z?pW1dvwS9RSc*1VV00Npv^+uOJ9<D4T}E)hqk&J|`k*Bo8ztVF3x)LwgdH)9nYFL3
zw>j%s;#gbq$IIYK8<fVif>(46YRPTbg}Ijn$tj)N^?yUiCmW&-c@@LpePZ-F+qoe>
zir=|LYvP#G`l3&k2S?1iTzh_VstY|pvk-Q4BSHzGtxynNq2Ux=pwuW55&%E4A^o%Q
zP0BEFKuhX!HMnEOz7PTc`p}rAs?Fg9S4`eFFM-`bDT3~4S;U7UqUnbB`QX)_S@ZmP
z$K!JT*Q)7@%(^vCfaK&<7Vz_w5ckwMz-e`QClx;vR~@^UmXI)gFgyL~(W|p3zSmi5
zqyBqjmXa$~1KS7l!_u9aZ_&%_`n&uQyg8zR`Ty21j0_zA<#@};!S+9klz*nXf5+SZ
zB*m`eYbK|A@F~5_{7KCGD9WmJE<nJj7ulf0B}pb?QQmK}kLf7?Ax|}A=3Wdmj_h)F
zr)FRvI|M+wIX{dYo(@<^(bTO_70n!?+XNLTq=JMKsht=LKs3Gn@fSo-dwV0vMwg;?
z#C*HBJbiH(B}5Y&b4O2WcXPYB`F=RuKM$JsoJ|A`h)~BKy4swC#Yn=$<RuUoqob8`
zXl817=qJ>pF^lt3OehMl8Wp-nVT(r;76TLvL@DO<^@GVLc`%uQ`~|k;F3wm~_~nSX
zK{IJEy$KSWEC?8Dqd0F<s*0|YV2oJDp|F?ZzWn2C;z6Lv{=pr&t63MU)S+!?zdWKr
zPF7N)WGeKm-zcWiY6O?h8T>*0ipd~(P$|l5%xUZ8R@(nUoIvuS`unnSes~EBfC->5
z+AzvvNrY9Rph%yJq2oh#1cchgCz{WD3vh|y5b03x37v}dg08WB9i_=n#*tuIwlH5H
zSsF@me+tqvEEbSl2*sLMZ01L=$;-HUDj7&BNn>v8lbiJI$E5@bK^&*#R+WE|5M(%&
zW(Ey|dBL^?YFVhQ%4KsNmIYG6_YB2AJt*-%vGzp^&>`I9O4}?Z*Mhw>&8VzK9&i29
z*pkENrqtMxR4O~5;u-&V<c#k3o*z)%{|;BC_Q>c}rh`1S)=XkAy`Ei7``fZ=&2W+-
z;F|#U2XV%BerieNv4>f}olnj&39Rm&05`+1&gp~PF!QsgZ1pa+IjW?Xw^<VKN1(qc
zsZo=V1SJM5xTQ)x4mYUl(G!V`nc<mUy1BG#%+fy&AEz3%w|!CP(EUgqRt^0U{BePD
zr<l|H`h2yEiG%7WOM`L5`uAnsQ~l~qeJ<!LpZ*>s)rLn0r~&mq_`*CJV7tyx#Wazc
zJT2M{ODLYo^hVK*;CEZZf}Xzoja`1!Q+l#v<aG}~-9!WN;R7M});Vd0z#nSRb9yjZ
zJ)mQw@6s?tZyfaQmO4;JcMZM$T*#GOO2vT#6_+NqELXR{jyQCNM!S|WL9uA5&J|nl
zFW4t3UM^2Ifo&JcnqQf|KWwoNSoVHeFz5DxYAc|{gM$KBjuGdNKQWOz7spm-qni}5
zABO`s$ZeYZCjkf<$N1jSp_OzOw%Ad;yC{-uRpMDXJ87sTrd#fsoL}GBQo_z?(wlMk
zAh~v5msXn8HM4V?TZNwR1NyNcS|m?f`p-kymKv8V5!wf(SQ~#Y8z;q4d%e1+TvU+G
z61JZZ8(48T;~CWER5HB0WjvK7TANlFQ%0CWl-HSaYER{R!n6kWu{-S~L%{Cn)N2x4
zcGu=bh3HKtM_|Bc5GGLuL-==@={sacg~PDK2(hy%csg~aOd>y%f8$kctI!@vbW0R}
zXRPcSqaK|Q%lrZj0T$`W4W2A-Y-bN;{iWyI{-tWII>sVSfR$m31ohCOT$e7}fK1F)
zUV1W3^0#xTAyTSC>PBbQ<qa(>--DA$8b&}f2BzigS6G<?F#PaFBSO+WxAxtj14)}q
zSSxsj7++GAQFT|h=wC3!!uuo_R~R-MX7rf)xY}pAcJ#Fz@l@jA9?Bo>2KnX!EZ6U5
zE&g=F`t?LV(k&A~4`#^^BL&u-R+wsB$YWenjH;Mj2kZa!y1>Q@^4=44p#pQvj%?|d
ztk1U$TJ*QE^v7w+7dh2Mi|FYd>BEh&y0POTx38qyi)WZrfHq-77(gjXqA*>TOSm~$
z=Yo8+yed`wJ%l=e8)%p&@NCXGIoD?L<S9vt@quZNAD!fxw)v|4Wa+(@f}}7jS@SU`
z#_lnDTe5|`pvK!cgsCw0=5yRQR)fj`HP?Tt5UPvD2NMg0eaj!bAMB;~!H_y$h~8H-
z>-2<egUws90hJ<_vnzM;WP4fKmh?V7R%6r?$PZzb_*~Vo6Le@5actHZGp*_NOW(s0
zw5T^Y`R#tOxDUr@7Re*Fh*0d+fD48C_e{)FjYmpd){VA7QE(wwAye<_1zWgi-Y6(t
z%gRR<ACJ8%c`_xz(%dmjny2P!fQl7qvjK4g+Z>xFB4$YeI6hQ=FYTNf01y?mu~?Z#
zDDDFUOxpfCNDc^SM#IX<mw7@httB7!iQh^(j6!qi?+xd_J&|)ZLm^Yt#K@FyOUOhr
zbrFx_GE0ig^KlAo^jGFXiyjC_^qA)@<b07x`IM6ET9|^)&ov0dcE2h@Y<m5j@JKt4
z-svmD)&$ZV*ZO38qE5Kt<0nx&XL1H;d%-2>i%@Jt*OY3nZo&zw{sxC&NbG{FzV-FF
z3Psz5r)@v|0<eWr2UmHA6l%APZs9;tQrD^V6h(VY7P=kD>URMCyTU29Ji_P!v+VKI
z@K^^#qC9hnHRji+qylcKH*utZZc0KuJ5~`2TWtZwY`yYD)aa{_kJMp+fkZ4WH`(({
z@T{AzG_@mPJeTKic6yQ{Uo7bBNUN=rkMjk5=8T^AU&r!)ISDaxF#gX0?&|-SF(uAQ
zDCh?QS+XdU5x!`oP~7Z%&VdO_D!|5!bbg1pdH&Sv>JLg0kfOYLFOrAWcN6NZU!U)~
zAH1_UUi&y6naJ_Llt4;ze}YXf$%98u64lPG>v_r>op65o{)tY`G6gnC5i34`tXH7g
z{d2vH|Hxjt=yyL?N|+AD1Nps*Z)*CusCsMSnB{l7zCJ#>xI2q7DPhJRg%Cu7%n$#k
zqF+wHEX(~7H%xs8sF34cRe8PXndPxlSySoQQ3Ulzq2{3`zD@X~^6&HTw?$ymivm%8
z@^g)VMvu;9b9&7bCovmOslWh|q~JwC`?{?-q-^<4ALJvD<N=N7bW+<{4{`$w5hE@m
zT?U?gcwDB<`pSDnvmX0%l8H|<HyEcRCnAw2he^L8I5RRM4G@}#yi@^A<BQKquEi`V
zVH9MiA0CGbg1yYr0RRsP*y%j=Jk1`iajWJ8R&VpPo>gSF)m@{Ry%k-hrO-vqlu~%%
z)TOn)x*0WcQ2c!Ui$D(njUmPFazlBGwk3bptm<;Z80=uPl2!y0{U#zE^hakEbztSD
zI?JxncA9&g#=2F7hwB<FlTGxn1Q3Ox9XJ|f(=tx*9HAK39)vIE2UMIRPKyYnM&Ip)
z<Vt%qyjn-{dj|zQOy<OTDe(kOA||brev5E`yl|dOy=&c2Br*p`B;u`(^lEBw3~zCC
z$tuQM)g3ASY9XV?vPN}v3ob}uOB#9*Nq|>iNP$^d`!c5)ms#Z=H&a!N#dbe|ZH=7E
z;u*h!c;w4M$E>>Lbq+f{qfk7JahB&{eb3!ZCYSf;n6~XnVr`;MGolf8GOj_w$bjZG
zSp*&_&I)v_AkzVVO>t}b*SlI0*=I<QD8DJfW{}uFv_e*fWdM+}-y>PP3^=4j{|}9S
zlRnXVrdG4wmcGU7Odtu&x3-SUy9;25X4A6Z@YKMX<^3?oR%u+Q=;?3tX<417o=EA(
z<87#8P*Re^v3-;<Lpi*XA0IvfgV0Iu9)P(eVXFiAWQ%<0+<~u8+uqz*R4m0jA3IYn
zW)EQ-@DD_U7w&|4Y;p{zx2P!l3}%u@255wesT-3yUrD6K$)*sPoR5<={2d=*h+!Cf
zC^#X(8v%%@89Q-fG;xT)+`<o-vIQMJZ3dZLv8Ia1`IvVY)5sc1FNGuBZGMc-3?8bY
z0E{_|4YQc^D~R$wN0EJBU(lLj`ysIwBiBvDD%{oP<G|3WEC+vzCX4Q+ih8G=30P7o
zWFzm=h!1C2iYeIXbAK@VACFEM+iu%yWYeNP14q=b1?Ay1GP)pN9}YXu%@%EKyy$I;
z;8huXe0DG;nmiZuQ2Fn%WS*ML>3vLM38SGYD$V+qRCjKw`mKu6CV1>qVN7C{=&SIJ
zuAVq++po@MD)2P@mRSz!q#~O?Ne#YUu#<+ZSuJ5!$1Y`D@MV%%0qr6m>bei<4z{VS
z&i0RWSlo`5RwG1Z>J9qQxTc;Vcij`%Lf4nxWx==feER!!2SL5)4M7{8nqI55=&ZKK
zX`kLCrL0y)>rNvf9SdZHo9LTS6%Db(c7`u3&!i{{t#9FBHOu(^YyM#c`-OZ&UW0^K
z%-Z`L9PnS0lh3#})NbV06Q1c%spr~XER%coLWsXR-~nk4zu5=WX}$hpH6AraFa%q8
zo;wO{jRPdk?K5y~eE=ZFjQ*=Jo#bkR#En*a*&viUfn%3ky~DH0Hh#3lgnE|(Wh;R`
z!T_)}O%qo``e#JtwkIZU9uAw>N+cwOltlJ^03&!1!p5s-rI`f<H_jeltvKAobG?9=
z+cd|#@*607XH*|cF5EZ7Y%GYCW!=6Zg?$PhORS+HaJGpgQQI*hG1n14s8B`5_aYZ`
zurm3EfVjB^%N?2$o$EB^7cUJRavEzL-#Oh2&VP-n44a+HgVfvh5o`vQk{Qg1A^Cbc
zHg*moIa=sok1CSH?QwxkvM55o(?Bar`kz})NjwJoO&e)dkfIQR2l)EO@W+_;4FTZ`
zNx{%Vnraa%rS7|CNxV~56F&GDT(x0?GL{T?yW%RO3#(*B$#hvlP&3Gw{daiqPssLl
z$31fQK7hBL<NFbpn$sD01VO`%GDLe+L@bbPJt(px1X84;ZS$L<ntUR*LP8~`@x<FN
z{o$rAU=xh;n@=<P$fo5j1DRwB)|w2c7L_g(^E$v`e3$7dR$pw?*d8pwD^_YMZDK5(
zNL9cBj`=?xEaooX2sY<iYK92L2%A=1Z~KZ_{&0^a%h3r3C|E6yfF$Ho;i_5WocsC%
zDx5HP`olYrLKBvJR_r~=HOe7m4CzQwbT95JZHF(hK?~J*MDQTW2*sJ}w3K2X!Zq*f
zEt5m|3kf14F`DS(qGFVWpgQ7<n0PVgenBV9XNvr{G0XX1ijA3AIR57`JFKA|dpLsV
zv#U!;LXTy%bkn&GaiEv&b~vq%G0wrTG7c<uFvE~C=0CPR*vqSnpR}fQ`9O!H2OvLv
zQBzpi!^PtC!cf@A*Zp>JGmKt~(frT2eB$Bq>O%G;B_Um=)Y!RvHOg5C;>!+QeE#)m
ztGl~R(^l~4+x4^Q>ihi>n(4dy{T`{R5cT>Wz843t75WHbcV7IV6(a&o8)5LWf0dQW
z{@zbL6g;2mCbr7d=3TYtPEe@pa)WJayo)B=zs3)2-DGWq(S~@AsEM&^f{cg0>r?(Q
z=<ruzlqgOGp(@tEt=>qIfLI9(;-<Cpmr(vee#0=PDP|>MY;H{kBI_o-#ACAXV-q%_
zPjqFLLO+tNG5^ESEBT1D0t|6Gm_cjGKV^MWXDoDKD{|v|0qSdamECB>jgSWRCj$vY
zJ9F%u#;}Z8#>vvxEqV-*4oO*W#RQbolv)V=E*o{(q0zKcWu0<b!IEWdsnvHfpkhPF
zo0`E7SwI;OGvEcvBYFuGR3=);H3e(EGh;B^daDLjE3u<E;&zM-N5Iio*`SGw*^%FW
z$ejezRN|r--S_l8zey2S;7-5D48_idrt|ZWR|mr*5eqfNKo_aoz}sj*MyJvDw?9V_
z>Xm15K}fmDe@JARgah@2Um)3%?hvP+06U4<y2e5?qHtmN(OhsCD?1@Knm2#yz0EQm
zaC>|etJ#-Z;C0!Id#m|`<FNf2s6RU#iIh3l4*mX^JpZyrA%O#&D6(NQdZrteKq?;o
zeP<etLWG0R#GwC8I|M-kBmwXNRFxq3(~cE)Xdj2;KJ(d%0CWKKqvl^33HbeRZvI_w
zgXgdzk=M?djn2}fzzNQ0hsx-3%S+Z*p~X8Ask=IP#!9%;aq5i!T|NUIjW#5=hg={B
z&*qufjOo}<H$x)O7PK+{W!LDJE|wF|f!QcGMo8R}HZ+Np6#OfW-y0x{Ydx?fLSj%(
zuv9kLP?x2lxnDDdu%pC9RUCOZz?sOgCra6fEOT@%6(nxNZGh?*`|K5+1Q~&gA;L{z
z_ySZIHfx@=S60G7hZ7ahFuyGH&_@+6M&HH{YJyNFULbgB)sW_aG3ulF<{$vmehP~t
zjqcuc+J-w!=9gPC2=I182Y*q$BpG0*jx*ps!mn5>feBRrkVRZ!$mynWB^8^#<f0yh
z7k<@k)L-n$PdYYy7guwK!yHt}XfkwzOv~b&-xchll)ETst{8BU7%EA&<HPTf*Sjfc
zI`#Yq$QahO!HWM*xtS)H8@!39z;@(TZOe)@-Yp)-PJO+8B8|RC>sIf?$Rvkj&6%A~
zqGEhHR^jbUCHqQ-1`qIjGpbyS%oyoi2en+?>wHkV!Xp%@v`1GAxmEhjiB!HMI#SGm
zg#k2IOZi_L@?Cwludjv@-Q0I^smtu^wqkqLi@`sJa;LjRXGg3q`>~=Mg*@I;GW(W{
z#;M|fF)M0}FcaSj%A7Xh!-;r0kt^tv+@F+vLv40tOwEJO4?eRvOIxYe=8Iib<wR%g
z7>eYt0MWxbao!Z+S$FE(=&~A`Qf8RavxGFM!)AKhbc&d1jFjt&?Ih<FUQn;pM`(qK
zF`iuF;<QMGh<Dv06mAL1P&Ws2$@$bGl+Q@I?>WJ*gd~F*jWF#vMYV<Hc%y}5M3{VQ
zRE1dTARH7qb|t`&{6X;%%ki+QQP9GqnO@tAH2|0A6LW3vF+kuC6y7}83+`}u$?j^N
zgRjSt2lw{szk5|eu;%+u(yO-n&&kl>>Op9O-XC=U!;X4U#s2nh4{JTKAhP<!HUp<%
zxZ>Y4_2-9(l^vQPK=f69CECiDbcxRH`(U;LTMLsEC$HAG9k`UIvd1l3dvVrYhd5@C
zZ1C4DIv+Nek%SHRd!XhU(%LnHI8(Uwp4?e@q|v*06(t)7msE1QzA%$@qTSN-TKSQJ
z6$~{oD*}6^*4sA3INb9;YC=_`&dahL&*8y17On8ktE+#>O|7^o^J%U*Z^4QlR1-2Q
z2|DIDMP$GKqJh)yimVv;=GSOu$q;r3+`L%7d0AYR+%^CU-jKO>(%lKM)#nqg%PRT4
z=Z%o0^OTZEl_()aZa>b^zs4FbU0gG8n7rS>zE&>tUbIh9<wW`UaxoQQ8J^xdz5r)5
zvb$z-8}=7Pl4iTS;6HTZ%bNTN+zaa}@~8C+ByOOeos~L}ifd;;^D|pQ)!Ou)m`2US
z`qtP7Q4ogRwq^kY4r)Jd6lT5H6w{JC1Bu+W;}=inUNHv02n7z7Zsr+7bYVK+$iy#M
zq&Xad32ECoJD)o4J39qeMr6eKs6_ne<P5%Skza9l^1!@r*8=9xM1nx+cT;Y6f|Y_y
zCIZ+r?B~Et;!~E;pwK#SLw5j780+ZPhhWm{N*{g2V;RBr^6#zU3Yf`8dsOB9!l5V0
zYfx8d4MdW<UIY}Pw|$$VaXWv?qE~q?vET~Z-0mP{wwpi1F?OLDtaPBy*|9C-ebV<v
zM&C)kJk^ij$_|AMkcW*$EqD4fFv&=Kg=%la`Dbk+G(s!G87&rzAq5(zaO_LSHXMgG
zI5*n0W)W}XnpSMz4Ly3cpgL?6XowYIJvUDYL8iNUw#XPrt{-nH)4@WwqqGP&swcWg
zOuG1v&y|R*>EBqEiN37O2*s5^5C3?z&V%H9Y_&;-tvS}cQdDpsSy%FB&6)82LKCfU
zC;<8Uy6@%-k0pD;qIp@;bmey)pSkA=tjQi}I*yBTT!0`>@jId!Q^|HW$03$sgnT#W
zF+8-jEKTlSkoCgLrtcxjC<;3fp`_IG);16jYA-@qkf_(s3F#-p#OjN?8xCoq&K6fV
zK+An=x~8rR@W=1xlIFxew7Bz3uXgfo)ELQ0U=>puV#$DL-N`c}+KG>R&2l+0_&1oW
zbf>rtc#mPY`2x^6s(a=T^d`1a9gu^5su-%E&awA1mc4%bIeSI6lZ#)fJx3m%Z9nYv
zo@{p@uQaO0GW2DRHXPAzLr#b!>Gj(2Upo-~po3Wmj&`6!k;}55=8G|~$~RpvuC~Hs
zU5sg0?dPb}YFHIv^qN42n0|ReLSXp2YrgM&tspI7C>KBJ!sqMc^J4{<xe^O~P{4KC
zXMCqrNF*AnaYrQ5an1bP{i68}Sa3mBgNdmc`c0&?OLmNy4h2q8t@6AApQfzx5jX1e
z%{k6ln=hP#ss$wp@kB|d!D_ACSIwZ%W+#3&*`vA<TFJo^U<|k`hR<JIkG&FWH0fss
zrP&+bxR4^eA4c;=dhmnTW*Xcxm|Wf6?l=0n@O;|;Dgyt@uH~QglIefVuU-GEh}&vG
z^~u%E@0mtJ8{{4a0=xhV*W|Xh@QmOk5y;D4zM;yQ^bj-je7yX;CkkJ%FmbUg6^O*r
zi4u<ILC4E^%obX>qAFZ(T-QZwr0<C*Y8<ktX&g2tb1L}{d%(&*{+q$HfaBBr{wvZJ
zqgjHz1Y};)5NY?avX}S!^WbvTlvCAdhJtf4x#Y37$M@aj{^NXChi(9zr`PLo`n9&#
z^ROgqerzZVSd>DHds>YQ<zJ930yRh?lXivR*~v8pI{#$axw&$yFiRDG@_D)<0{>#!
zqSw-9vcw|*tXO;Y&RDh627fTm;W@u6Qpo1w|1fq=O`-(KvL4&EZQHhOYmIH&w%6FU
zZQGtTwsq&+h!b%h_QU=O9UUE=Rh60F@#bsWHv63yP9y*BcJiczr59!jq5FixWR0PS
z6W`H`C`LacP$VMgQ9pt-2z&LjZ&7ln9yqEYX2KMVxtVSB#jx`xY?L@E<?SzzVd=uE
zv9c8#Vzf9{e<zobdACM8btxDjT9T*=L+-HyeG|bn9Uo8)X4U}w`l;|23o0^-cL+Mv
zGFWa%9eDnB&;jH$+|{uasy|xqP`QWEyS|cjl77yKJTy1IgfD+P$k`+{ZLiP`Ns_aS
zLrg(Ca=*W(47DePOt@yRu#-_ttlcWztY3W{2uD{@(xJPWbx!2X)e8;1OW)BHNP(e;
zkiQ;s1=e`b+9}x#DMw>o{Uw`aAy{QvDwB*v)*(kTSFMIR1*ao}5W13FcjfOJ;902@
zteA2fjZ$Gz+AXhvq&G%$_Gprbm2hiJ9&RZ<C4uRAM?BOaXHPSwVcs!ye>?aoFIb8h
zP=Hw}o(<!cM3jvN+@560SVUocN%r_y@u+M$;B|U0aP^k*p}@YXe0i#aVC9hteZ_t&
zT$E2tiPOK-rt>J#;lpX8tJau6KupaFp?c867phahkDl8z-y2|?5p0J%BxT&yvbVL5
z={Qn9?vX7k!8)vY%@aR$Kfs9C(Q}2m(ir-s>%4g~3G9t(+yfR>hL~ktxp+uD_B+?3
zS@q2!ko9ri()?Q7$=#!utn*Pn`)v_OEsEabWVb<_E^zVn&Zj@lS45p(0&jntK9Q>A
zgGKj%6_S?*^8a${>U_)*o>p2pE$f_)X1g1P8@O_O4?85PArp&{ya*D_$>_98kqo5i
z5&d9ob!F-=-&D$4{JsQPS?WGpul1_QxP8#EK|k8V)-Nu*mNyOrh&ppw7?9r3@D5li
zAAMTvV6~r1@<BU=`Q%Up1(3@`Jh)4OA?{zwTwZ@ul}tM0+A%2fl(#l+%x)wyEC&;Y
zBfB6gX~A|49vtM{KH29bLU49AO$7@4$25`w#Gn1YXNbNU1>q=v*WpLTegEKvCsA%@
zOBJHv%S~TX#xtM1tmSWhI5yY13Zem|P|j`uqL9{ehYk7_MaoV1`x_huqo<s?(tVm)
za^I~K;G!+{rL864@?>PTAFc!7mCc9eE_7oUnFJP~au3HHi&V^Ka=nhbKVs=s!da`L
zv)=!1ty;7?xkMAvTg4cb$Yg@L({0)3ZCi=S%=RDl)xZ`%xhTs3quXyj;D;q|Kb#uB
z+D%VS6X<cXpz|rRH4DYKVv3M}+q1GI+qc~vS-D^x1D=$LSv^}Ju5evty2}j^-z;9E
zlv(O4Q<Me;cYsX(02V%-xJw-*{nH4dU*B@-w5{!L`1G!p4)f#f%|ZdJA4Ca{lsF0I
z!Hddxw3rZEQ^KF?N?n`Ltp#6`fT2}xBrJ0z#xq4&G^37S^EN=_q`xF_3q%?xHHx&n
zfLk8Tvh-RGA3opRw=L5;16-C|1{%OjDQjW9Twje^@6<Gp-rMC$gUdQmF`0b249l}{
z4TzhufzUR=_g-}YlSbUAZ<*Nn<?aAY&0#tMwgF~SaV!-aIebC{mq9})*P+;&bCRM|
z<^ba|yU6$8sX1a8I6$f4$&#WN)DH2eP5!J6_zZMxgS)E0_mddFJ7yE6H_`4$dGla#
zON5*P_B0~yY93v89^3ffDY;=sL8GmI*&IUQ)ypfdP`~I*O_Ir>SHLrmVuj33*w!{&
zzdbc-J`a*;1NRn%?-QOYq7fQG0lB%?v}O?!@hCjijb%s9DCnuD)oV0riuHZ#wy&pE
z8k_PCE~2D2Wl`J}W(P&?^4u^@-P1fMPk(+Zt_jp|eulY>zPvt8^giyzaK#<KfF&8m
zFk}5JoE~#wBY7^*?lF0o&~D?@Y9W<U5|@Jw4L&R932jX$zAW-2_5{u>l`h_tLOudl
zXIu8y@88W1mmGmQ;eEDX`C|J_#CdnZHM|2+Fx9%agT9u^C>?taxuv5QQi&1~_BW3J
z^oz{HUSg%L%aj3q<BzPc*60u;xe1q_X@?LC7oU8629@*$X4U<%8LwU#NiEL}K5Rc6
zsD(g53F}a&SHlqLbh+5wtdXwG^D}FdwIkf($P+^e`9?^KhtN3XGt=MF(f#(&<YI2y
z_|h*gO`d=}oY+tiw@Cu<|BJ-P1yvRe9k^Icd7(@EoYo?TYGEkkuu7C*ou2$PxA!Q3
zL9Mad<e2@N3uQyg(wf<b7@dL_J?D@$ky_=mRNhGO0AjovET(<~_uY7z-}q>uutliB
zUZW(K;j#F)>Z4S=_j4cDJ$M1jmHzOv=cxPawV@Oj6&_O0l(%T$M4p9Z#xpnIcLmi-
z6ql`vb-0x}T||Zj!%s2<$`MF*-2gqrQZ{#nKvlE=TnU`#-`J7<?;oM*;YNN^DfW!q
z6F~R7oJ}6Y7ArScR6Ix{oI&pv+j71WP<*IUSo)T@b!}3Y`b1^ec><$}W!(HBZw$mW
zG(?WrkOHAj#~>3V4OL*gyn3aAhQso_J6lAnzrgyZtg`KZGR@(!gpv%+?iEk`HMOTe
zy==F4R$f@m@usH}DtD9nR_P0nM33!-+1-m#=T>V7?!J>Cf)Qhs1Gq*2noIg83W|7b
zsvXxTizk!_y2Q(}CBhf^D8q?#)L|6!$Qlpr5e^TpdL#iCoWrxB(>c1`K(^=RzgvBO
z%<WvPO%Mg|ox#`mEx2n~j&%71<`$2u#fNRzX1hY5+_cFnkL_Rf^a~C9^cTr6f18r*
zce@-S1`_F&-0S<$?0s&raSD%c{$2m;>8K<VNk%q&nERPPy`|(!gT{QM=g3KQlZ8x@
z9J>L!FMaFBw5a=I7}V<%_|I}pdv(I2VH{59Fu)^y-DO;YI0{&R0ZYkx3p$E>-ckFO
z(2$g7GRyD0U!8mCfDG4pM+uR?*mEd@qc5+n%{hRs!j}JY;ZV?aQWS5^0I~yOQO^pW
zqE!_W_R_+?+;7}|@y4%KgT5hZa>N(XOV2>Hx=Wpia`1q^BcFzbx%^VuL2Pw<Pg6Vs
zpPUJ){tk$mFZ5kA5EI+(JUZPm^!?_(#|LVz+k2Yya9e16D(tmxdSAEw?!|gtwJjLr
zJYq;HgWwiqt^oCWk@|7ms|#L?z+wcCmPsDrP3!1AyCq&ts-Id8m)9{*r_zO=ERg|M
z%s?1IdDd88!X?pWq{)hPA2orJVWiuSk#Cr#;8rwia_;+~2ah$|vIPxqJWtna4(d_%
z#HAcU`$k(7MgbbDc~T5%115W|o-K(V@1Uk($r&c4=h3r;>XeeZlDkGCyb1toJpl|c
ze)B?)6$kaB3$MiDF(|-OxIMjg1z(AH85Qg`?B3;h#u~rTyP@YR-{Wsk7@OTzH0?@%
z2!q6Qq6-OQ=JaijkXEuSv4hb#y4+?|=QX|5(|r8L%et-oN}t3CuU+}iXnBd0RXUO{
zy?%68!KvK%ZvD`<4I`2GX?{UHwSM>S2|8_LZ`ow!RY|m_VSD?(IX<By(_CLy$$p!!
z9&3O)BP?~-Rnfa<(H=VXwWoFi@p5dOo!zTeS>2GVV%yCHHt6W`^D9lp-CCp5*m_47
z)tuebC#Lpc74D@*ny&z<Ob618)m{J~4#vv=pzIkLnE#JM`+ptS49x#^GsV@f1N*lI
z`}<nIAdU^K-rEmz4iI;x=4!)GST|pI5#F|n<l3mSF`H()c2_Z(Cb7hN*r5{+8=)`O
zn<(xDC$IAm(HtFf#uvw%ACJdfsBRpFDRIKuJC`v7k`D2N2wT3fRX7mwGscnd?)GVR
zcI0n16p#1Q<uN@yUN2@hHoxC(F!PTO9ReCss9P<waR^vjE@FxC67npisf0t)F$3k_
zq3)+GSZh&Kq{y}eL{p+k$gYbbv%fkO`{3l#tg`}&C6M(4Sc)=@YR`teh4&|9AcnNe
zg$B(}L!}yfuGeZR#4{{Yup<N;5j!Q$DB<>~=jjWvMAgBiy4*Gf(y`%d7!7z3B@bW?
z6^&v?D2GM&1hZ#m`5))B!rhWub^=YgHzpuvgYF#M@@Ha@g;zw2jZj6}VV>zagn-^`
zf4wEjKr*p+Dv}#aYcjsz)b=DO+c!)05VY`37VL4r032rvafKZyzX}$plf^uU1GdWx
zbzRM|82gPo!@wXf>5}FQM6`?I0AU+|`k-V8L<3MM#G^a&a%pj1rpParFIRl{^=nS9
ze{=h49d_nPTDLy?d{LWmm5B3$g)~v+UM!SJ9ZW$eG@Rxbk~Bbck!W-YOEg;UbyV@J
zi7ZHq>qkVKaEuaxaPI71_CuRCzMJ)u>Ea+nvgk-`*zMH1M}f8{Ra;7b79psNFKx-2
zt9zs|fG+c|EI5cU^QmCRIRIPLX|QMabM8z*k+J2X5HV|_mc3ijX;!Y+@3(vxU^}LM
zp`{XySe8)N_TA|n;;X;j%=s>z+1mxLmoTEtb5X`T`JU>nxTSL>D+E+1Hs}!Cvh1Ek
zqnt=m9L+-ujkGlzJpd?<5oGV=p+$lcHtvfoxNM6~{@N?5P<C7nOoB?A>_AkBwJZPl
z(W`V4*2XTAK-9<EpAn1t5AMb*7r$aQp*XJOgKXOmzdE4+7<);iLJDt5hP2*3I$`(z
zR&WtXdyqm(lR!fhdjg812wo{zP&R@od>RdXY&n0>vdH^L+Ma&0^jj`Vc~DoVD@m@)
zav(1@8hSuqLE7!02gaAp#esw;$%Xv<3`2prk_i!44v>N|isqyOO;jUKMwYX|IUb{b
zegER&_Ncs;g}_teWK^ejpw=%XF&Q-EV4nE*_9exIilaA@G8C$SRDGqq$dV!q>W`(k
z<P?QJH1!w06FZW`5J@ex*OW`Nd!z$J^TWdr9Z^6coLa6Z$^t)Q5f{T|IPL$8o$CIZ
z{F^Ud0UzU*PybR_8yEJCmQY5E7x06*2}RX!4Wr89PhfmCLqc+?z9tl(YSc0-FC){D
zkSB0XB%0k;5szDy_IU%Sp9MbF2YfCuz(GRCCN^*%QUeV(Mx#p+Kve)sX`pKez7GUB
zruR}x&fBKndM4SS18}9~S_#2Wx>1BH|C{*@DX(8!i(Wv9kIxC5#wZNE>c0cPBF|ov
zHQ*3Y)W#jOVpy`|1W}u|ik8Dfr5tvQCsl<qb<@<@Xsu9`uts9^j5AWS6%iCvkYjEx
z#ab=5(g{f^<4sM-_x|IdM!G@^VhBfwPy-jr)8t7SwO7`*SYW_nIn`r*(A!TIVLk9T
zM&?h1H0;(kuVdtfAG)h$G<(VId`6R|;4ml$Mz|7F5~B&g6-1+tgFJIcV-y8+=}^;K
zG>dC|DVTI;Bx4%wI(9`8zksmFjhp9Nu#I5}`#QdY_QIKU*=-IVfy2nrArAx~+9PKM
z5TOhOYx-Awj<RqJGlT%7p{3p+Guc2Ld#?=eg`kvAY3SICsAdP<L3KBry0<2Lvy*+n
z%2M6}!|kc>bhZG-D1F5Lo=!k_YCZ1sj)}rUg8wggE-ooXqYpp+7<~)xPrvu8erU<1
z0CTQt1i}Qb1Gu;YIgwo<0i4BV1YwOb!c$roEKEC7+A<g#83r>L43&)Lpr~hGagW5C
zk+i0R`x_GLLSPA#BT?w<l`G^);WGfLl%?k8#gkEst4+Nl+mE~IQg`z&^OvF$Ogk_H
zvBXe5i~&Q_s<w16J$f`8sj_4`u-}d#04nsg`odctqNQ^mA}B+a(|1QvlwsTp=d49M
z!j3DFixOEkj7}M`e-9LouF1+sQ@gv?hK*LO5qq<&3eVRuuM$yFc$1)$Qtfx0la5tG
zzc0DplXfvKs>o%z$yPbFyT_ot_7a)j3Hm8OQN7WZLWEtlP->M<VB~zvsj{5=Q2r+g
zIcv;N*DiOLTdNEU&)q2J>kzH7lu{_nI$oU>dN+q%O|vXkSOOq5Kv6~3gtb(kmcC;q
zl%XW-5wvA$c~GNEX$raXPaONBq(AN_JcqwiTsBzTXb?YGjGuy8r#jIhR=Bc>`9uGv
z8%NfF-7IrDocI2$l%p*(t)<(9>a*4DAD?oHD(;NsS(6EkGN44bahyo)sjC<q!&f;b
z@R}qpvxjRLsI1wcllcN*w*m{{lNUF<mY2i`zJEQ69cq6g&3hhN@Ax#zgmgqUhj53C
zOrqg{gX(ywg7o+;J6^d11Hn!~GVgYMGkbLFGUm$8F^2tyx3LJOQ31GgYF5%FqmGRn
zkA2<wDX}qs%=$YNo4RsOFHFDFxFvt9MA(PdJ=+}<SNr-*6@}JkAn<!KX=fr`C&Fzx
zLv&?w%)V(;PfnzYw}3F*9bFc-tSGw-J{>gC_OY{eC&#<ibZj8~D%$ZuUv4wy*Y5m)
zNdiEU|D%xjzfrIM1wyeh{nvNmD~<nkN#9Zcu9)pk(#8vCj@UqHP+H>xaR~NFIm`4F
z!X(5d`Q_Q+7qv0batbCW%j-D0o$~HZ#hc|ZXMTCye;gbQXP{(|syXG**8DHlH%ir&
z)i6<xpYGe&VWbQf=l|#P!(Wwae>|UN52lBmM^JMZq)M2J&yJot-;NIsmN(ystPOa3
zdA!^le9t6u=O{#8ydBImM=<K897c^-7&9D%t!@7bB=5-x5p-rUN}U}wY}439mQ<$(
z?0U@GR@MrBib#r4jLK%#%TN^dt0{^_Gp!rJqD5sfoll9R9;Tvwt92F97`98*(y+I_
zT)ZnUO5a*uW#~`Kmomey&+y}=1__H-+$UQp*1|yo;YK8IbvJJFM29=z4X5KH0$-=K
zlOlcOYJFVQ1T!{gsJOW6wTSB2l+xrSyY0^>yKDyEticOG^O+!VG8Bgc_6AS$Z0)rH
z{zPUo`hiO4x-mWnDDg>Qh>d+n@bFjR1FMV4PVlZ#K>XM%gM|$I(WJ2(pmOJ^;#Uy;
zDxU^UrNJsw=I6P-sDQ03$!r++JOObkWaxzE71kt+A&lN6nAo;C14*!IXm^$8)l|J`
z_@xXP{$;w5O_|G#%PKsbMR0@)NYN2GhoU04yXf;-<nnU=MGos7t2$w|6k66TW`SK;
zU^Oz!)<$LeYqV`@<KTC1-_i`Mq&cn!Cb6#&P&AWpF6uTc%XZRMbm)YL*s<?>D0@>5
zR_rV_c6{nIslI^TwO3D&>9tXA)LC7xQpU*YCK_<`4b0f1*e4^~dO$StOQFZw*E`9y
zx#8xDx_{E35#AT)172s>)b7F4dLnnaFcVzutO75{2ZyN*IVLFi^5;5bTdbQS-^`Q|
zs=_6hx*ZvfV(gow^Cu!l_yc)%uBvA4g&ws(m9ikp#FW0oCX>NE<)&?CLlZU2JvrYU
zPc=c&IWZnqyWb3~Rtqt-sY0Cjc)5W@iVH@B&xwJu$u_%=U)+bykFT3yf8s&LaEuP{
zaNe+J$6124S~N1p7YaK&rnoMJAfb(e*Nq=rShp#A#(fzBun7a^78M0^9T7p>Z3N(5
zwlZ=t=N%FO10c#8|J?b~kih_j5fer<G2v8t)pRN+LTp62aEXC;i`l|u0kl~;v+aRU
zmpSNbPIGBFV{s*3(7}xGlKRd`KT)J)m+6wb`p2XAEX1aSZn?1vI~Cj^A!QDd>LAq^
zgST8rDH#{mL0@#RilWHBGC;y04HYi9i#zoJZ;eXQ9rVKsRWKIbI>e4Rg^DclunN(E
zD*)+l@(t+4f&`IyuS(T1LKk3pHDWo?Jc>Z@x^O%m&CgA~2*Xodh2&0vW$W>wy75K^
zk8n8sOIty)-ACC@L!nFmvrALq?2fsD^0I?&@ynqeX)XL-u`B!=@$p;sTtYdBq}pzK
zeQ>d7=Kkmd)RJqEPCz5Vj=+l!5+fMD9bh1;P+*K1CxDSvPewcZ<ysq4IV%hhldsFx
zqmC(!t%8#N5=J2T1tcMLaDYH^_d|Hr8%nZDSRCJk_v^2le`6G3niU~=;}J>hg-JMq
zCt5g+$g+gFFT`%+Rl-<gwba$Oi7$6?!WW;iCwaau!Jn}&6Gvla%ii}71MzEzi8_>J
z-rz_i?h6g<;a});0xO+3ny^`h+_7AG?nNJY!*bus3?5s2sK=WbeGGR;t1+fDeIDce
z=SotTKKuju^{Kg4*!-LPZm~t3M}8ns8tztY6;SUzE}{J4`J@hUlEYx&+s7E!!~Ib2
zk0Bk)KdSlAM?8Le;fvJtwVCuh{_yQWokYNqZQPw<7AYy*<`4MTYzf2|r0c;9POZh5
z#X#9z(VwtE9;YdZF0d~Y0}46ZT0gv#myM3kbTqQW5gJ%4(2RDP&N(VuzMBPo5jcYv
zVI5d4!nn|gPUEw6Yg#Khcj?b@%@5t&<lp)&B5ZIsv6PMooDbdMgO&0sePi-*3!7Tc
zx-lNO8KLUkk#=W#ce<NCIO5K81)Bd)KM_X8|1gcRvi{dp!Rg<gMC?&i-_=@!U9~cf
z+z(O#sG@4!A%9&KwMYP0genQGjZhsb5>w5$r>~QSt(@cdWM%$G(XGj;-;%`poSUgg
zIa_k)`?#oHJ^e2C;j%1_1D4gKi#X&0^8bMgX6nq6-%yJ|9v(SXuvGhce|$Y1CsS*E
zeH`!B&}USCzYQMGuGZ>zB^pr(Q3<7tKq&^A_|AdKM)kNKuNgcxVe=h_YLp%p-PpEV
z91q$erg`bkuCU8)k979ZyGBh{SYEiI<9qAa9SmR7$-ee{HLY1M!m^lFZ7f}(0jq<C
zP&%v#v2z(B7leoUMnV}e{__Mit@*2wd)8m+Q3GDPg8x<fZ)s0;v9iB%TQA5yEBR!(
zFn4>n60@bt>TjUTZ9r(%qMU+I$mBA2C#htFJz3sTk?f_5?sBK>!+I1{k}SKXe37w;
zL)wYW;qB(hVqwf?w+B{4$L{Nz#w@;BhT$`@2&>)sri(Ux5~vnU_gd}-2YRf+siUsm
z3p0#j=<vT^vp^mbQBHLI()HD1T%w4=y68VuNjPaV7~<HBwnilNP>ZhVCGNE+$FDYj
zOC8o>-;3vGLruGbEFS}phsJ=j7~<^mqtiXYqe26S^Ui;eTyVl#F7#$kj-jNkQ662j
zt-C`u_SPsKw7;N2;teBmoQSdrRT|q@{^BmOWGL&b2osGeoyUyOEAC0&Me)i8GZAHI
zd59d#;cA>%C4X4frHjqNU(YHt|1626-3My`51{8~BG9wkgkOF#TA(=c%u7N4f)P>!
zsKY|Kzm+fU=kbt`HwsxB6(tz6LTG)e#dKr%)5j&N@b6V$vzNCp6|>S5EH;3X)o2;P
zSfy1^Qkt%bfeeMY449I@m%f}1@$pl$y=3Mj6Z3duY+RS(QvF>3l@_F&r29jIvFx*C
zs31(7Q3moxdhO@%L>f+Q2$B`EquR~hj5JC`tv#rBO+(X4(kirkIaP@odhds@nWla!
z)=4_mDGHf}Q}k-@QlPO;tu(^{SuWkh0fm;7AWStFy&eDgl4i<C8u2gy!tzE0=UG|O
z4VdJ>mviQ^(YZC_IztHoD5o??aTacvG4NGDL{W(>_8r+OYG`*5L)adZ9g1n;4w~Gl
zhqN3xvN6_<*SeO8inM|?U8P$mcEU6wEFxA5)-1@7rQn`!p#;FL%u_2W+QdRdpwiKl
zi1Nd<Q+xLM{OWh1vA|q>7Ylgga|L!BXegVk)cT-IO%1w{e@~h9beL8%5zS6Jc?`lu
zAj=fmMC=&~9#Hm3JT+Ii$w88dQj4TL{5QAyIU00^<+VQ2J}10n^sH^~uXr^RsG$(Z
zH(>_PG`~sN!#T}QoiPxT7J_|Xd>&kI&LP@4N^4(FJutjKmEd3&ldFSKx)&RL-Go_}
zfodQr%b#Z)+PYHd!uy~v;4hr#(jL4IWk3I~n2GKcO__KO3Oh#5w6;4!GXGS{Au$`n
z9m$?ea=k7;C#sJlf-_3!98ribb=`$aWa?v?w&!?%Xk9{Xm<D3LZ8-Fi$@u5SVOJ81
zHJO7J%--FVN@vwl^-VLE-Y%2ngNNm|Nq=td3>dCmo+7%!o?d-6-HSTy(cK#L;x9RS
zUMUC1wd=oYoW<@IS$lBqg5j6mh8)KCc$XOq=c6r>9o||~u63sC@4~nJuxYUnN3v_v
zZseVoW#>Z{VqRvbzq@S1tm%xItmVROw(`HAY;dS_wbKOt>+=VBQ62zHaxa*}orG2d
zD94dk7tip;*+DQd??BKQjOw&ENt&_aJgc>m%Ldx)5ALWyz`g9{ElZ4Rc!d|9(!m!e
z#_zH>*XJa)_}}-2E!HPdgL3K6$neEoc$1=Dot>SL{q?$TVf;@uQ-+3kO`*zw?@5k7
zt4#EZ-8uaDfKv75gS}|bq2P+_BZom&;EPr_&DY6Sd?bC77g4)(&u|Jml?JXbYsE(<
zEgkMZ%sEqc3(ORd7Tz$z>bY@}hphT?ztvmRIp;9~n3iBC{glCqE_K5_nFe77Q7c7E
z&uNZsW!3biABuT#8A9vzluTThFaaY0$Qv$2y|`L|%F=x|hLFPPoOQth)i*SN>dcn%
zngh^_1E{nKDY6O}=geYQd}BQFn4g?WD#)Xzddj1`S;>YyQV&1gMYG8|F@e{Uk99QS
zp$%9<Yceqo5@ug?nar-7rfQe9CS6n=nxTK`-%(vTAt?2VL|C4yA(TAql@f14HR<K7
z*vP~*A3aVlGZY09nEU{0(f*x(BC^VcT;y{NH@5EgKU_~`G@kFrP)2nmTHSW)THBbg
z+VKIxw%(O^z0S?7T%iqQv8|ZDWqTgtF!HU2Li6I5v$Hs(kYC8gC7Yke@afrwvhC%Y
zsl+};kxwFl5@+OxFxE~0ZjIOXF^xEKitp(~ev#$I_VJ=$vg~L#ostv6iJ}68;BRIW
z0yqr#eaw-u*!Mu>_;dIZv|geNc^J;z?STA+h3n7F1O0HP;1T&Kb=s#<71*BwRi6(?
zkeIOYX{^jbnSE+oVTQDpOFk@+020p)o{#1UPAjPOxI?jpP)xKx(+0T#bg8Hz<j=<Z
z;6XPVK#U_Rp}v(5X$X~*^ao@3)T)N}z6Jx;BZuO+eWOt6W)by44kNjY)$TFdGJhV@
z^7U#7+W*!GJ3i%7Ln>Yn9C|87Ugp9Z+oIbTE{WZJF)@g7UwhMX?qREl%7R1s$k=rE
z>a{%}1g*JX!LH-huSIfhA_ds|BdJL2&n3(ncaXs|n%DS642JXXMB(1Ju`>#b?aJGs
zw=d5XJt$9MXS8>p5GCx+PPVUSPG2Lc1K`tcu9EfNKLwbsl<aD}o6Na9LlIo82GQ)u
zmb#d@*1AKUo-Fp>u={`)K=YtA`S{hTtY<d9aQZ-hGh}*wKA9JMNN`T*mh;33+#vYp
zI2B&+riJCKuwzF=s$wWvVyB6azn2WoUah{-pt7pm{}Oua%eRgl_Z*L>51}uaP$Mw_
zsiOR!&gd&7Xcdqxjmn{pTTuWaF`Na%K(8)v9Pq#D%KQoC%+5izC(hq%<n-0EuIrj4
zck}QU-PYbpe0jEmgwXq@6yB9R9&qNEc<L?VylLkYTp2EbQF9<2&qcpX&4XJnq-`q^
zwn@7{e<OZgm1w;mCi=z!#~c5w^-<W2HgE|qc*(S1Gt8Mk>}W^Pj|_^~;tJzR-!Rht
z#VQboDsK=w%TTfw8k6&hXoDE{0H!Xh*MMrFC&av#6j74<$lA=(LXX{pcu+L)_w#4?
z;|wjaHx8DMUewSoZTE9l!Px?Q)qPrQqgteW-EnIG|1F)g9oi3S*gWk_GX=O&3RZK|
zaSqD^{93j242<fh7R#w<U-vW3@2XJj8(u`w2Qz*qx0s4u>CneG>TRtwPXCW`@;}*@
zIM`YLTieqA$jQSI)Zf-Xg<Xrj%&C-Y$~k~d(FHP5_N$rz9_xDn$;U{Ih^5Hn?wQ&f
zcJ>Hi<;2B83w;{BnQTrcy506U=VwbXbHDbthp%b&U1>(<p(CEWZT=|{K4m6ym5TK*
zt7k^z#NBVtpF{RtW-Blg8OX#+92uPD&EB53(J!|fyHlO{XhlbEb<XLX@5k%GuMc+q
zuFe;F`>*T47yj;_5AezRP%uI<BnBbOL1->xIjj=MBs16EVZO?ai<hT^xm6qB+=j7@
zQ7e296z?%odjlDSYUhcQK}%6VmHsU#QK-3njjG!)D;`ug6)kLW(g+4JybpA#a`FL4
zxopRQ_E?rn2?ie#G7`lu3D965(CNkxm+d#>)GXN#lp6T`R=YzV)bLE{%sr)~E`&3&
zY1Ff}fq-F&j<<>(HCb@%WbEFD!pp)U_$yH_h~n^9ezq_Q7Db_Apy!K*)(f*_3@0u~
zN{TGXitx}pu7U0E)U&GdiZ5f+rvmpD%QQLZHtUM_MbCMzOqb0XAkpnt{nn_?jb@ic
zx5H9Ww9Kk|_DR{z5b0C!*XoJm4gKaXEJef&@aA&LKS$B_!48h$FUlwWY>sDB3FIdd
zC>dp?W0L)1@9U|ZW#YWy4wfKB{@|vahhqUqBrNLJU#dJsJh)1D@E%}5Dsc6p9h0be
zlRuk$>`MUDPy?*;piKyUG$1P9(F)r&wtOML;@wR?3Tn-MWvO2$>$_UnG+Cw^T(?s8
zZBA(Zvkw7)ET8^XzfZ4~3a1Q`lA+w7du2DdFWVIFS6}ww(_UK*5g2*duQk=}L6!Hl
z9HhNv`<2sL)D4x|l%~Fwf<A=hNO9GBQ!R`tA();@X^!uqMs?A0t#B(_+7)o(DBG0a
zE81O=*#mbc2aYE*+eXo{SQ74g-#a1*NuS*KW<bxB@^KpNSa^?vo@o!#Mf~-)q7iPj
zXXw{dBi$(6SU~_~SCO(B0M3;8b)sAGHHZnp31|&>twg{TeTxOR@1KJ<_WA9IfVuUS
zUeD*+ACun1CXJcrLi~g^7ZvTaCKxO7%;$ub4JDc^rJnr+{NQCWFJWRI8|;LeDAfm>
zONwTsDZ@H11Q#sw{Lry8WYQSRlkFDf{X5-XwPXUJ(M3=SVd<u!(In3ZM0R%ovr*!t
zmkroOc+Tur!+Du2NpGiYZjS)2>R5zOnWMIG=m0wO@_~#*^;LZ@HRZlhRaJP&E4n$z
z9^+S4gx%M17P~Pb<T@k{0(PBecGOPFeraPy&Z>aQQPmQt=TQRw)DBLH3pimT-PdfT
zZC2Di3&u<~aB3MV4j17*-@787Uf!QjJPBAxMJP-`Z!at88oV!E^LxG8!prVq+nhPo
zU|4%clEtlEN9RJIzBKyH*(j!sq;M=4S2-{E$T{wM9F)4?N~20J2&n*$784oRue*;~
zxG7$+XycsA;wvso%QX@}P~d;s-6RB>;lGa!jlU|;RnqNY&6y%_1J1U=*Rn^|Y~#H5
z9gRA!)18l5m|nG3XEmJ7T4UMBgSWH-3ApJJ2)0;rsZ4HCmyiF=ug)sQ2{f<`Rob`J
zE@XeVs&zTZ@dMU(=?ibwW9h2M+AesR4YWqj0*j<tCE4WHM_YTwltLRvGA~X-7$Fji
z7wQ#)GN77JNvZqvr2UIMfG{;kI<r~t0bY#5cI*pOt3%lL(1g%Iph*}ZmKIO*!IhFf
z{{++ju}N7Hf@;c_ADJDWGOq$1<A`MfcIPwo<kwLJCdPUXCJkzG7{nY}T`p5@O$?q~
zKlgk|pV^WDZ;*NSPx|>DwFg95p?X2LhErM2a3tZ=wMzhXU~f*-`IM|&iBMXfb~UDY
zkWJ7r>>XgLBu2c{)4(}aEd+c>@W_|5EmSK{h{cF6m-qF-j-KoZXCW3d&tdH^K~=s*
zvP#E4sk?m63DqPzkDit|x^RQPK3jc)<}a<Grq%ra8eFM#@mk_i201N_p3~UkT*J<1
z^zACXjUIv2NO5i2@|>@5%4zW1*{tRP<+F#@dNV@+m(kg!-><(<CZ{3k9Ft`gqWm9-
zlW1Dm@%@S$QS~;gDHT3=ITvV?V>4H4HmCTuJ{M;Pe}Nfhps3_$>g2d_!r991H1jPx
zoY+U4f)G>SDyx^I_c(y`N>48R5)MobYA>7vU=`w|K_`?B@#tkl(41;M6G_M!Q4&D*
z0`EP610DfDk;R*v|4>KB8T*f;2#|H61FcEU=09MGLG^x(i7nH;3ZkXKX~FA<h+?YA
zeieX5ckXVBGDSS+|1Lc6)0v54u-J_)dEx&$;(4$pfs3BO6f-utCZrgt?<M|`&iVJj
z6ubtKH&~&J=g2pREu@-I<A4C|#HX4y?LDWa;Y0_QmV@Fj@fyDSt>r7eJ9N|p<A8DJ
z9W}$6h2;*RGM*fv%&_iok(BiZMbiy)cCJY|ODT+|WPd6aVYrf+6-G8H`D#)|cAs6V
zR3)6E+~ryLFm7LPLyvZ{#t}0y%olt%abuaUuDMblE!MAs>gsUp&#AEE>h%Z1t4q{|
zdOrMWnXiR~WX0JXD66ZUv2;orF-!<r0|aW~5fH+kCX8rCrN^{BoN@E5e0rg#*nWHk
zy#C#LM#^XMPEk27@o%WH*-|4-BsdoFMhog`MPGYm^gPLMLD)%2!l$TDhRcV7pf~?>
z8HNTPqbed%I%U$`H%;%$#v=B=Ur(qUJ3ckfvdwq!x*taT1h3lM+&U>UJJdfWQE=_P
zG%qe^SVTOQGzMX3Kd+TAMCI43=(xZmh!BG!|H=cvw4Bc<PT6KO73{XFDs-2tD%SjP
z`6XUwY;P^c*Und{a>MZApl_jh@q|($oQE$Dv-@8fTCaR>-b>e9I<BFr50m42O46Cv
zHbAO;nLWG_&_>@+?8wO_83t$Pj4^JRquu$;lE;FbXJTaskir~}@1lTxc-d2?q7-wb
z+lU$DTsEHPK$&TT{%rEvwWg@9>bBV#d-3R2+Y<n0>k**NH$7pFg0&X(8~Z(a7)?;d
zdmf9R@hkS>K03g?dvGbK+8$<{8OU-$-u#<_W&q+sPn#h7i3s`rcIuWZJ9;z-Y@7yH
zsZ|p^U&!`E&vJ^jMxT2ogk!pFGM$M1_ZfA7q%WB~uWm)-3zaatI8Maw&4E~&6aaET
zkt5T{8O+Y&l@6qmE>Ll3kK;QDWsFSC?ER{oZdu;`HDp33?z)eHDZwWTQJ|ymsuGKj
z88XKOy58XpW2hEuE?e30Y2x%iJxFJ6o49u<r~52_o|tH}`!hI(-8mWsr|>6>B#v7)
zRaWAPQ<~2-`kWgqeb$mO-DItQ7PIx&k6bZlK>CM2kFgk0oIFGulCWg^oFthonDC2y
zYF(G<lLZ1Fhnf(xgWzG;cwrotxSe8w)h+FZZ~%>iN7rG;@R&3V7s`kmetaZw$^((?
z+4d3RB(^`pp&8<#Pj_><jA3AV!WZ8jS7ci~K++9|FxAk8fqFqJUE)xasjx#S1L|za
z?UREkMa-$mV6-kbFV6w;<M>%ZLykDFE86>6U-gIrXObBpya8p}Dmy?YoeZw5XQO^S
zffy>qP(DWMHCTUFFNRY}&v*1C$SS3C!uYjW&TQJSGaUcotBS`Lw&sBS{HQT}>T{Al
zY~yEfaDqSP2Yc?S!t@J)&ovGGLz_FY@#lvb;lg^q`1lVLy(t~4o^Xo2F}Nzxx_2R3
zQt5p6Pm7=Bd4N}8Sv`+UOFD)pi4SBZ<e7pE$!!G0j#-p?oBB*5-svz)VLsax&;?yA
zF;X6b>oqTrua?cIK8yVi&^`#Z{eKeUO#dNB%FMvP^52T_Wi4%|&0jITS}TwTBs)F5
zL!VK^Ts?ZWCYQA|ubl|45lFIzQl6+{^0xZ22~7tbn2$qk^G<|RIeg8zn>*<?lZd7c
zO8f15-`bc&-)p*r8p|>5;^3+Yp-WK?nNq>gN#RR{Nh%Y+7_!jh{r#22PT%u!QJ!$T
zK6&%w(Xkhq`V&5Dr^n}acX&o`Ufc7j7P;@Kgjm7S5;ZBgPo6VJBG?ac{3-^m1|iyH
z9AcEpa2DB!J;mQy!(KX-%F4D_W!>_fl*)dw|L?j(YxwSo3U*r^-od0*&Q7KEjz4XG
z`bjo9MO#xq)DZJwp@9VIG#i36kOmNnesWf2{uK?mi$MevO(i*)I<_}*v+U>sF3cB+
zbfd$_FnpP2<{$}4!iXZDKq#C8wzC0J2fWR}Fai_Ut(8l^wxv^cx#~O$qTi;u-1c0z
zd!1|ziAA+`$>8f*0R~N-O`nHy>~2KB8NAHeYmQ%eqI5+UVmO?wOM^jRY6zoV64o#t
z>_=Gz5gHhlDZ7?^9xQdaM)UA9<b>mRyW|bPX;~_T|GI)@Yio<{BFnRRbL4c}1-I8*
zeOJ9Q>gns*r_tR9o%XPY(@&pPuX+*1?|ir0cNtF@KANrwnZp_H7;vfb>e$l>dE2P&
zzL6ubI&Xx5)E9E|joz1AuKeL6Asi~UMLYzEL%Qt{vF%i#K)A0LA2ED4Yq^%=vdNx)
zmv1SVgIhsb)2KSes(WkSrTp)yr0U9YOL?m$g9ZXvYwMGT(_1_kZ<{ZR&5deGOP3aV
zW#wz!4Co<C_CyuE6bHu8oe`$Y+u_>*0N*Qe!pISZYb-jTCo}|dCZYf=I!v5#7TmKK
zMM}_M06j&L!oMk9g;U{T{o$vJ&HElrx<dn(p3chD$uO4U94z?*pYF(6hPB6PP0@9o
z_D*^Imd<xqY_`VZ^>SlVcHfB!t46O>yc=7))w+)M`~!-5xswSV906C03?T$R7GF;V
z#c=>!KHutDw?FW8_Y5_z;a)pCCooZvp^XQIs#19zWBfBUuZo>k&E^XTf1Myk7<UF0
zXm8H#C8pe?!gp%zc}k~J&iY4BWN!_-=7&RUBgh2LaXoPEI7BNX5npW{;PTmfO8}f8
z9o{*D*SuovVpH{-$Aw^N1FVR0tQ&{3#)AWE(|2dy79S0P#LVBT7ISwjC~@)<g0@7+
z_rwC7Du}7-%t~a(OWcM%jp8rJ;^4<q?5X1KLF#!z;{7Ls|Ja*|j%=Xflew`DJyrNJ
zr!a9EPN48<PjFOz+jCa~P~fS5)R|6>$KP3+O|&mX?h%}Zs0l<~5wmjP6AZ5<JV2^6
zWudHadN{b=-z3o~D~My<*Vn~L&_+IFCH;DChC#pIpk_!Aif;y&u;(xtAngz)4@Q&x
zM2WI(R%(ykn*DTG(BqpvA5e5MA;y{80GOF~_^Kox`4pmQe#DXJlKYopgKE94(F?oE
zd@dnKvNk;q{>1|{nwCfsJ9e0iN)VqYuMj?X@_{n9<rKW_M4(QGL-z_Y#jk7Q$5|I~
z#^#vUl4KKs1Br}<5MDV232drVR(6`E<QkX7F0Q0GTz+?O_R8(KqQZ^vGEAC=&%Lg0
zhH+%RnusII_=Qdg(IJpmD%JJaxvo;?wek?gePV<onJzhV{1legtW;7VLe-tgA|Jh6
z0u`n+O>eGrKAI6hluZbU86CG0ihi{6tK=r}W0PY9WM5XZmAta0>6D@nlgmmFP(!!c
z34iqeeE`E;<*p^-jq!5!_qtjfG{F$NG^o%p5Un87@s+^pnuM1KioSNuCW1IvGo}b6
z!&0n5FJi992ZnpCkV6$~&Y>GskU@dIWspn~z=uD=qZBU(XXEcf2ZKo2^Xl`<*QC(!
z50`+HYz>O!2rl%7L6pt{R61lfA=G%?*lN0hG_f>^3bIu7H=yE-EMLbQ&hPBYpyGx|
zIjl~&Ev5r&<fBZGzp>aLgU;Km&wBFUu8<j}bXQFHo7?P56p4{C74RM|)-2BDjh8rJ
zz-@+C?#2YhSljB&)ZN*<#MqrOzFg7+?LX!s;}crZb>qt;hw-Hy5|P){ytIV~8fH$p
zq;{*p+UIc@VK#s~J%{{cIF?U+4_$vk5F<blzXe{3Fh07CfZ0h%=OU~`ik?MSXcIct
z@)CH@q=*}BQ=4!%(DW!e!V?rwLjY+?2WGTLpTcA~G&&TvzJTH_FLYV(DDLiG2NBDr
zjzQF{=@GweA1-p6oKTq#`SPye%W#WY#w)Mc22^VLR)Jr_ZP`TcX{c64)Q~oU3zD~K
zu4t56kdI~a<<McP)<a@8MCxQkrW5qH4XyfyCrc|d?4Ent-AzUu90-JbvIuxSV=;Vi
z8)l2ni+p?{XOgo`6?gr|lM4fZa>;lY)KH`i_p0k^gW>6f4Qq_<p5;Q>V`gfma@lh^
zyLIWZW#bZyN9MKY#+=zVKL|<>Rp<H*em1E$V4|nm?Zu@#^K_<+yPFVB-7YrXe703E
zs=>`0Bz<$#-HiUhZ3lOKlw+-Pau<@O+kA75X>S|R%{yYRjgTX5K*$}e=#kuzu<fDy
z`Z9$+Q6M6$!EsY(OLdaP6NI7;&l1E+Lw{<0kOP)pUuREmom#($&M3J6y^kHO)_4+O
zc+As{*UKV260IkzdTR=Lun6>F`Hg!tT3Gw2MP4}*^VXVkp@Y7w@VxBLFBxic1$b5R
zjjBbR)}6hgvvtyw+lywYm2-*&g(sN%k8V+Q9bs5+rVEV0_hVmXvf10jW~ae+ysew=
zl^Liq+H!(dk%6mFT{nVdx}Epa1KhCWhT;e@+dgvRgP;$EF$J~go8$5b*Xk(Oi2`5q
z3n(t&Evf!zXr>n!Qq%@)gJ~2CxE7PzWP8n)0L>DWLg4XaL;da4t`lbNtS4IW%()69
zgBnj25`0DFbOWRfIFEB`2HIMU`C`6M$q8{czg%hyR4nt@g*B5MF~)Q8Tc?M88ym#U
z-=HWPU&WrzDA+&q_{Z>bPS_vQM(^h)1~4mGTIX4_!4AQKi*ENNGD?K%Zpd)MB;r-+
zi<w~;@LVLb!pf@!ttK)ZwmuoQtzRJ}{G!`|{*3Z*hoUcg--vNnvK}DAal3#S&s1%a
z)PXu%ApW}6f+3rGw|AeOuvs2?w)6Cr_MJ>wjuvGT%>6fUZA}IHwzOWQDaHrS0hF`T
zzq|1XW9A8V$LBH`H`L5`Zktmk7h$?XT}0Z_&3Tz1tx7@t#7W`1${T66B=D>Ey<2N5
zSV@*NVr@*xAPcJ}V-tr)UM2B&*V|1=&t3y0Pht~#XP6el>#t;E?7+zVm;EQNJJj;J
zUGtJR5}z9e)uNVa6I!!koX0d*n9{&P6Tc}w$DPyYbwNHE()K?$Y5eB>rD*S!<+jAa
z0e_(VMz{3{<`zSkR#K-;*jOooDeyVRL^zk%uk?MjIa7ug*HrX4LKhZ`b~DCmbl0S^
zJbubn&mc~%x2|}u-#&)&)i1MxBFE*GVg5*@epL0AH|)C+R@fYT)3jyYv%3SYqz$$B
zGL7H+r?VurIt)(Dv`^cnjd=Axp)dvBxU`CKt<iSvUW1JUAT{KOdT}sk+;(Bee4O~J
zn9-z+W1XmQ@0v7~d?<pKCgr$b`4$HUtq^J3q2$08C|w3My9Et<8FPq~5jC!OsRs}0
zV<><)2u#raz{Z<#zspoHi;xR95k$u1&<CNgO|*wC*{=#tkinA+Onl%K+3iN%OMf0G
z$=dJxQIiFcHvfs!Wrrt|89ZCBKbiWOaS=aDg*57eN0TA+Xq@7B-mr+9X!*2g$vG~T
z0@ugAWX7g&YT_qUm@)CM%>%g#5!v?j&jGfvhnf%?Dw>%Y{PuF8f1s#=9L$Wadk*s%
z1#bJU1+#Y}*;qrc*-8MQGiq~?NA5fRaRWe)gW9`R1l;141wfa_>HUHYach!~+8rHi
z<-}9qYZdO*FLdMAvg6m!&iMqdUbr0jkJA9le{gy*asJn(s=tY1r_HwiS<UYe^Tr_P
zK<3LbB~4qm$ZlNPR<%y^%`xFR5l2LdtlxdV`w|G;8A-G#u_v89ih}eD?`_S73C|xH
z3nk^r`KIUc_|S?PMNGS=%FrVxkV={TGJRy|6L*!!j?Zou+2!?_t-StcFJBH5#iCOC
z{e7Qy7I!00t}~yk=*XprDXahWd2Rjveaycj`vn}mo74UM_SHMHl}OIQz<A(~IKe27
zoC+bzG>J3}W4J)J2-=dVmy)#@fxy*W7HO7TJX!Iqe4%W%aPZLb2+W@WUNP+IdTegC
zY}8!+&{388Jzt^mK_1nmm!|(|QeU;&`bFxjmD|u{OSx%YshUvxb0Q)|QD1V58(9>I
zp@ZFH{^s41V?28-ZUKC=tR8~zub!pGj_s#TH+X2&4WYbT+-sEOarR3ytVytcS+UWa
zqG$UXg09-=?3G6U5GU%(2mdmhma@8FcK92!qjBFw5=5bnVA3Q?AxfsCpOJkDB$H}9
z0hI;mG8w1~e2H36?(g|6j@F4-amXD?bwKGVejQY{Cr0K-Wr*fP5>1xv@W~A`b<XqR
zKx#Wlg}5k>ZdGJrBv|mh7Udo_<U-D9{5Z{Rc9KD+9`fti(|0>R9~y%xfPgRzS0E%i
zJ0nTbDLyuxveH7{7B#B_Xmu^)TTJK*-^fLzTpEJ_9^qhAgS$+Zp+O<UL>Y)M5G1TR
zTqzAJ6NIIyaQooh9Yiv91Ht|fm9IPX0!<o)WYYt_R~$%(43v_8H=jx?c3J`#0qJEY
z8Y0YP7Hm7}4(F({NhK^?SGV}^$L94yv5_YcmeSQ5YGN?N>;g%#3^ZkJc3Y|mU*1S#
zB*Oj`;IU6Th=zzZM4r3C^Y(uChmS`Ol3AFT1Bp%<5(U5P7bflO&u4B~y4yKa-~cQ8
z#dtcTBM9_9mJD)p^4|9eZj@PCvON_ggH0Kq5N8|}Obn&~k44UDa4Aa>(8RTwF`5cs
zNVCVWKw)C;Hh-(tqRr6rvV%TsDr0MVa6*^lGuvXZA&f;IB&w<OGhdf@5v-vofHj0G
zv+NO~C<R)qxWDUkf3GlZK!ZZjh|ckbfkjDf7Pj&b_v0Cg0z#P4mWGV)6|F~-AfJzR
zgV+#<a3=hfTuNopwxg}Ct#XkZ)$!6L@($8k3*jDu0(TkI8&O;Dw%BB(WNka?DHTgb
z%WW)V(>7E}IqC2VB#`t;+EW>#h6n~10T6wuYNLwLClA?STByUM0q4m8ir?;u_YODs
zJzP|$qozx>{oFEUtpvI;g475+-2i=&_Yn-&&kn}Jf!kp8Pn`UiD_8IsW`G@ClV}ZG
zjR77s>XYl(gv^O3#w{fN=pPXHD^%W8OPSHk&YiL*L=#72D1+a^s?_eO@-}#-2)iQm
zhTVa}3Aq>d5B<>CDzy8L^QXiRnf^{<eLkEY$zh!8TMP(iSrxa;_+>)Z1!(Fg_%<~a
z*gsW06My*8qlp%y26A90OnYGDGam$nwO=E0i^*UYt48Zg;yH1o4qUU34exo{`bFnC
zgd7krx45i03Jo-LV|l>;s7yMpyw6bSuO8!xLrEXSypU`loGE3Eh3^q^5I8Jh#>+O~
zMsm&+dz!AAKA+%L3k7;$KTsbPT$ZU|u_eKg0igZ4VuVp*R|*u46|1Yu-lKwY7r0O;
zIlC2rf4-J{n8lx1Q-JP}9M1-<H3O!N21N)z0dI)=muTJhjqmW>t_p4g^39Ab`0b@~
zo5o8uM~oWnO-hWE0){O1Z)Jum6p|nfi~gzEfS?l}rj3O31`wDA3=HpQ>k67Two0PD
zI#g$d$<@5Apwr$!E8IWw2I^-_7wofr3;}NIWh-PDNYP;Ep!f|h4p3q3ka70l4+XPW
zkFob~=Y3BeYW?EJCU2>lA^y4h!uRe2*DN<qmLdJW7I5*l7KvAT$pZUmDAbeVw~|b|
z3x%$oqsH?t>&m^dosmdfs10A0rY^dbECJ2hhvSn5EqbEtR`6?uGR1j7)*||)TAxn4
z09BLC{iKf878n~Lov7K(RMaiTmBjbThblycersTrUvjPlu0IqQUYWsMU@|!_HnDR*
z;Iq!3>AP(Fzl*r8m-y5${!A{Tps$))NHv>18GkTBJiD(>s#X%(4UV~71Hto!mmJ{0
zJDq&lTey1{VRBP&`T7Qf2MWWX7Vzwzg&>~+XbR%`kv>2Akt)UoVN>Vehmk%3H4Y3F
zgnXeX9K?ZrdR=P9f4}*i6(GN@6`|+&l|XCiJNDVLyAr{X$2Lob7R*JfIaUs!f?Y<x
z%OgMUNfCUy!IJXhubL2UA<@5}uD}BJr*W5FQ<(1wj-lWBvzRCU4#DTddsp}PI+*M?
z-}}LHP3tApgBAZDWAD_MN!tMHCKKDXF)^Olwr$&XCbn(cwr!geO>Aptt$n!OgRhVJ
z7xZ0Schyw@=m)#|Wdn7tcC;WvI9UU3(yqa6yKj$DHSW;Aox_-<W~xZEvGaX4QEu{W
zMx%ZQgOixDh6sNuiL!#_{ahds@B{p7+KWO6deaSWcvpNsdN%hh+>pAHGdCR%Y6GvL
z#b@n4mwT^o3cpFWo703l$|ILY*pj)|E_u%PH5gG#5J0y_VS}p7y!5`yhG+MM9$YkR
zgqY2hrn0nqj#j?KfK<V;o&9*sRe39QjK**2qr~kdewJ1p{n`u3*|4G<_5xe)1&%u=
zKC63%h`c7T>KNmd;dd>1%&_RoHaE%*y30SKl^O_Gzs>uaYdD!8dNSLewl5n#x@<;e
z%-EYqs{c`icH9FHw!3Sry!q7+CeCh#Guy1L01e`91F*a6EZ;T!+^;D|M3>O1bfUG_
z*E{Ql*aLfnJLupu_IRH=p~)D};!slpQMlzHeLzvAe5j4yomBAQ;nx09eT9Zs+llv4
z>XxDOul&cS6&pz|9y%|bh{~ljwC8^%D&JB75B47a_5->!L*_Yj+wn##!7yDeSC02u
zG^g!^6#t2w-V_ds?0<<Y+AGr*>W%CC#7$vwzS(ZC=ll4Vc0>?CoFfgS7(ssD2E&d^
z*OC+#$ZH@|9qsqjpRw9JxnXyh!eV_Cmv<Lvf8o$K)qnk$h+_T^OO*-00Qg@m)oBea
zyRH8#qUN+rmJHp|#)+xou*ZpUnzm)4o`}{VJVaZXxtJYA57rX(b_X+sE!h*T=k+6j
z>W7GCIX>L!2x9FWwD$gnHumad^LWYB(XlMBR!3H51}q2fu?}dXs!jeCl3*al%~mTI
zIXdF|c9Dgo((CcMzavPko%u~||N5MIl8~x)y`$aZ>Gw6c96PJ)gg$xj%RnKJlOhI&
z`4rX*20DNtm_j=)07}_P5wd{4n674p;rpxxjJ27X?Q(I32d?vJhx_!-)Js-*y>|O$
zMrQfwNT=5_bazIg{rNs)xMr900J1RNWvH?wC_iID`Ps;nE51Ex0Gp%nxkKaQc>2`L
zly%dlk<SEl5S7uN8tLVz2?^FpxzSm|?|_9qL#Fo>eR`sMhiy{_f}XIpPrOnA)O_d{
z)r(as{af{Lvkq%#%`WV5n#U^=3~ZS0morBni9=xax5K}c=gKRRktUsv`19`fb5{fe
zf4=#nhHckOEj~x)%!Qf_m(F#M^lj6p`vz8MKsMt<6a&dVY8ET8GO9K+DJ73@5i%oj
zpr0;#sc+Xgy4!yJc3Y|Kef##+7Q93I^~SQ}7#1ODf+iZ;i8uqr(<F-=F;RFn1mAf9
zaovo}ndm_Y2daVq#Q~Gz6fC;w$C;`8Fbt%231mGuQNA-HmFND^7$&PdwF<`GVwOCc
z42K#OHCW@vLT?BoG=9<c+C76P=|U7<?n^4{8><*{Wq{y>!i3D=iAr=cW#`4MWQ;qH
z;9tf<2D0_K48<8SX^#~-$iGpuaZ=WW;ubl;ei;fD(U@$dkoi(=nXhSF<Oe3ZCv_-^
z!2yk`^BQh!GddD0%;Ip1fn6A4@&iilpl!`LgR}!lCt~&Jy&&fiC>nrF6LjZavO;}Q
zzsEe)&-p-I1|%+q#ihW@P%gK>NeYQ6par3loQec}qH)!Ui<{KO+Oj?{4x-js3!J$*
z6BfZ7eiu<vlc?dDH=YLqD1A`lser`vB&-#ug6`Qbhg&ttP>Sr~W<Vgt0V5PMWZ*%?
z9ELZ&8Dw{+{d9k->V)b{;luV}N0V;Ij5v&qYM?1khu)#54I%`v7~>R0zG1{sqD!JC
zLFWs1#7=f77U%M(_^HeBiO4v510o>9aao{8{9>I{f|6(V{IK~YgfL(US2hE_JQgGt
zMG~BIzIRFSoX4VYiH_Dlk)xr?5sQDe2(|hr>#S@q-7hVoA-llvTc%ChPFt;>&)U!!
z!t?P=nXjLoFP}Q}TwuP18r#7H4-aDu>9%4?M36#4uofx)Dlx!OKoV*t@;E>b9dXJD
z0@GSLQXB1w?GQY#7XYj1&PIw($kltmGV5cwahQ?T`3tTwlJ}fNG9~^y<|vL3_TeyO
zKcTn`kw--yl9(ZJF{`ba;mwv&16%7w_HSTd@vi56zk#7kxyf|E2Os0!NVi(cP&Hky
zaQAy^?6fN0!61b?Fe9Ge1Sd0vRUnd8`?QH^>L1t$;YC*hOb?d{vXUZ=X~9;qq|XpV
zB%J2<GA7OdA={gPp7+8dlF36HQ3VB*Vdj6kpot>}S1E$v4DS^?f+m>&E%)&gdT9N#
z$l{kiBewWu083Jnh(lYu?P7RCFY#~i9raB+g$*GyxcyO^)Ij#nXWF7%kzcHS<}T0Z
zr9#agzuTn%nBLC*^_{Aao#;LhXA`HmbzPu+7q+ctExv!uSrT!}0;0@p+K}j#MmBaO
zl&7TJUU1?<$cqD6qtV@-3~TgKZ<rF-`$8f>!FLVcTh^VXjpjN5pr@O*^YCJ>wFQG<
zlx0C@`USI?Pj^efNtrO^(imOSoXmB?c&|)6$CHoh+dIC)>TRr4kR!sa+b4<rx-SP6
zlOwq*uuvb|GSxZ`Y*Mbb5<+lWwX4+vGg2Eg+b=$Rec7Lld^fk~*JPPJ19L4#m)Wve
z86WupdZBcE-L<(o-+PlMb991I{;4BA7v@{Eb$Bq&x{QjccCNMsDEFuNjXvS7?CW@c
zhejL>=Qpjtf)bTn68Ll$sHkuFxJPpOUPJs1Jo$x%K7BI7<)bVa`s3a*NQGTY!`szB
zVw6DFTBh?;h4Yl4y3p^Pb8E&|z+W<LKuTf5CzX<R-2AULMcf+Q0;g9unB<rBOT{H`
zn<)u*QNbU}8Ro(g88gt^)%^r*(_?C#(Nt1a1b6Z`%wDWZ+#P38rY8-n3bjS^dl0F%
zcXov)z7j<er>aP(T!wnKMJLWp?TurSX0F4dwO$Uia=o5gtT8?}g7vPw)NKi?d@o$x
z?uDi59cXijv0EqVV%Nzpt*$+KVr98UJe&Q?0>`(<I8$|;HobbHMpgC1$8Her6zjWb
zM^4=MvZm?J5`Gm-?aUib&6XhB`+FwIg)N{&`#CkZjDazTWY@TJI@SmT>bHPoRzX3m
z9LBJ@5*RuOhH5&;S7AC;5X3l^^XTOs%t)MuW{IPm`3}#t(qAURm^`1+!;ya&#Ysa$
zVuX$~YY+goOG8(9YTv{tiZ{};w|J{Es^5q0jzKGgfqMZWoGT7V<;<U#YEfWqK0Uo)
zs1mnSp<y(%hxQu>MWrOP%INizh2hRqMIpFpIaXaNI5y~&FPIw0*i}JwYp$2}`#ang
z8N+23{uWMR;f5^mKDfLd;`p2!gF@!!*H;-j$N(rWwuL`7Rt#^TP83-lt8(RBG)W5v
z(5V;(*gX75KO?P)_P317RJ45jPmFU)hVIBPK@;rsD*Pn*$JW>g*vmCHb2lEt+`{Eh
zx|b&)%-1vEL{aWJ`ZxO4UA%)yA_vLZUJ8-E7K`ZIbX*wBk=vHJ&;XCkNoedW8Bgkj
z?Nq=PR6#8*s3f7%;wfWF3HUv}VxWgD`1k53Rv$%_hi&?+D#>7LhI{p=sTi%l>+Mqq
zNLb5zntPBCW;ndDI`A-4Jp=U&bWQB>??gCxha!NSA==*r{FGyHy9ZX8)v}O`v@+g+
zi8X(oE#SbhJ@=5?_r;3zZPv_kTW<T+OMwRk23Buct6Xjpyb$>_PCcuB;1wZF%1j#R
zC9{}9b|})4Nt?&^Y+f_PupA3~O!C{b@sKFeI&Wg3m;x=;IPW{D3MFuBu+;-9gD<W^
zQ6h<vP6_72AZ5bg!@3AaQlSucKC<1lRcRv>0zfv2%_Doi@nRNgA$a7VVjS_9VFUZ&
zVH%*4n+D9no!_&@iLDJ)RH{V@Rf+M21#FJgw8ewT<{V+e+YnKkgPf|y1Xx2saIEa1
z&c=ukl~fESQ@%(VZELg&Bej<owi8h#Msa|4?#&8K(QqV{zF|Ee-B*3(QQBNq1=c<9
zQ&?&Mq9}Y@tat{R<HMh2nowN&s%{)ui$pR)d^U3HwJ*3RCs(c_Y%`V&D^RJo;&L_B
zeC1HfDcadF({8D4+=2zGYQ~Bi7Mc+zLYd@{owj!|e+qY<WtN!7;@$iU!+gmPNL@#}
z^*{9|>;G%wF>)|6{IB|RTjR&X1EBx^CLRh=Ti`#I>~YhKE5M>L+q~8ygl2(F4B3K}
zf<u<xE&`B5JSt6j^``T4cxzyw)2saqL>##ql#DO?`{>8Ts}&XU4;Y#=U5{Mw|5mjm
zk0%dLk|cb-CO>{Qkg%FT82=9uP$)e0r}Ufpo#XfY<voHXgGF}2Tx{`nw!MjP|4je=
z=!(y~_jMV%dcE`g&|l77SdBDC`@-K!j)gWZtvJ9~y86wr*2kRA-#Lcq6o8cvIG1*x
z&)`rT#?05SJ4(IBYQ#;=gG~rAPKQfKQZGc7z)Tc)>m&q3#E^rtHU2?$P=o^K5;7k^
z&JSy$iN_+5%rx!^nfvJA2W^u)pL8-^lv<0QNip$NQG1>6iMLQ-bu+0y9sX#BIxv1a
z>(E8Z%nZGGW71=J)LGhHfuz$fKQ#K?BaDlUIa5r?gmS?6y3H}Uuo7X@TvjRL^s{wc
z?;YY6P>iu@YWQQM!$_MU>XS8-X@}>vI<fr?+zYa*pVXi?{9#o`iWlq;N0GGCO51tv
zx+>a_PRPzSrs0EMdjV|=-(M0lrI~eavj9(Fh_vmjWHwOoHOr-rNvVT!#laz3F%LBC
zw8kb)9+Y*5TQb>Sq4Kn>-jko@9(88`nMCcQQ2x#nZbY9rNhx`Jl{QbLJA1sWFps!7
z1SWJ&0rW{{sNoD5!X-gwUiNs{?=_Bgdnwos{8RB^1!)*4GhDB*rsxP3O<bA0EXe+i
z@B>(~e5?y)jt9oqU~mInk4sna(Q{We>S3y_X{qd*UgsRmoShfPj!CZV5a$sVgi_X6
zgrbRzO}g`)hzw0zNFQei36-R{6o14ljo^<UI>3V_F<0wr;Ej0N8KBlwfJ3fH=$F<}
zVST?jOcJLHP)D!OXPP7|9_+O1belr_t=Vr!9Djf{eiN`C;L(LCvZXJisVn*$6uIi%
zxT$|{1)itBdj50c>*<B5k~jtU+f!R$k4O{?s+Jw8>JfS`T6S+FS9x4^teUesp_-(I
zA&IFxv^;tgT>{Y>dxKGYB;V(jG(TH{=B7(?WEDzqTSWV7xjzbP77R)Bb4DM3;M7eH
zQ5qRkz72xtP~A9WKSHrUD^qVd*HH}^W4P?&^}XRudCO)4!AB=%-?CN*a!!nFJ??ca
zUyw_g{UBY&$A$jmS;l?Qc`)OhbIUs8uPcfz50~TaoLEgMwgAsO&;r3+eB}+5U2PBA
z)YHEV*`T<f`;;44Pq5ZoT?c<RMvFPeSM{Sza}Y6J*N=6V$TD5YGWMlDh#ZDO{>>xz
zD-hc1LLcS$D7k_g`ygu078-I}PeDp9eDiKOw;bC2{$l#A_K;by$L?*OT@F!g8p|TS
zu77ArDoX5h^2d3YiRAR`=;gn`L$gt%%u6WH+!tEP|6$Dy3jui(cEiV+xGdNFIWH(g
zP)k&~r2#Yw#%44<iCL|DJ2_$eBG2ePwuSR4GBv~?uKJA3=uBgAlTzpy>F$2<#<>ny
zBXw%m>WdKVCfXW<ve`OtP5P98)i-P)&oC7N>a#?zJhk<k<kI;T;_gSxe@UbKAmaTL
zvj`A~`<kHv_69mv;^c&pOe_bO-q)<q{)HvlC@bT`&M&;rkSJd6oG!cm<x>o|yu00u
zQzG@PT&u;j(6kU}gh=u5!Wb69)=BK>xNUcbqjqDoW;kWDQYQ~KOszN?Q#`duAXT0i
zQ@jL>JVt{a<~%N{v0$U-=}}*<wplWL6Z=}wL&;+@u-IUa*H(2`!SislWVHr|AvZ)+
zN}7NLr?ea=e5HZGFyuP21-?Yn8rjxfrB>%LgopaxL0wVuFzRSIrOfCXCB<g#X&gYC
zlG2Cnb@(CRnzbN3OK?bKJe>+)%{ka-pgs9_lq7p`cX8fCRi6^InS&7O$H7#bPs*G~
zbv=2^Db2~vr1a^rPws6U`5t${I9%spVR3x0j(-#(>iC#x2{}s4dS?!I8p%YAt#`U`
zBV;oF=-2E9Iyz*`)5duTez6kBLK15Nc)%nWo#Se<sK?z7jy-V)Rqoco1=GvRjso{k
zHaHWA%GMZ-tz=e<BS-h<F#Tg9Oz2!Fu=+}#78lwM5qjaiz8Xu_SC?rYWmcEz4)`O(
zOPv^tKJtF<=)glGCp+wlQiy;RX1f+2*a*w!F5Tgc^1DA%XB|9nC1=UF0PTaaEuuUt
zpRFGb(d8lb_LNa`siu<O$dm%lQrMTJ#f7J#v(Vc58$2EI*#8S>;|hw`@$wrj+lP<u
z{aPcI?xG*HWO~l4u28*d*7qxzT*}+&-5$93S=aH>`A^KAYqB@{yW9&KYdUHNtQk~T
z&T!~_)jMu%MvB_Cl0Y1n(T2QwaaLNq=^KGQMKD-_br9?o=@uV~*FQDEkZhoP9M3|z
zHH|Wgb&U^+Z*g*Pt{sCAKZ%bX!;-pCKO`k6-6+#DiRuQi{8?X+r*#yAUFkZ$^0hD0
z<xkXozgp>v#aPH5QSx3vq)c*0NAUw=vGk9QUK}L)$<Td@LV}}hw8uv^?!-Xx|2)Sz
z{@-)_2k-m8vpckYYS-;pKF_sjJ>zAp``JQy@O#(}nKD^<^$O1M+YlNli2}|Jv>f5j
ztJg#v7>-l(V=-=-v`@neQSPsFp}WcC>!DG`-)ryhuRDz>lG3J*g=rHrwM;ckRLZhJ
zQ#169&m@wL)c2qD%Erzgv#qsyS>10Aep94a*%wxfoSeM7rNjR&JDI=lF2ldMec!Kr
zBv1eS+ny|B8>Ro4TqNN-`q-X4z%YB~%l*E&5WcauB61=|qrB&Wv4OaZk64R34%){V
zHy<&nft@C$b)=Sd?df~Dk8S*9#PEI?H16o2q>y?a5{(=`F8Mg~)6jXHjV727JzsZj
z?yAz=kxn_UT%I1_)7@4dQH`95I*6@&wf;P<!DdOfRb|V;n9%5ht)^08Nqr$BeoO^@
zCUX0k#Bw>dR;49N%Z}elGxa79pMvr3$X{GO8HWR9$ku3&)CAQkD^flJGrDEcLs16r
zR3z`iiatrQ>OC^<(S5bUks1%(d_6SPo1FjS_gcK={jf_nIUc&~$n-4NNER|%!CF#@
z^RF+;^xTTFT-&z{y$Zc2CyC8Uj`+7Qc-~#Wg8CDO2@^`l3Gx)jFF)dhB2ps95)Zg;
zkpQP`jE??efnDV3Fp~tP^z(wUP=C~nHSr_KC84Y=WG%)Nj0xv46HH{uKZly9uFT|#
zkU&=vbjwIQ?3T0w5MQKvGrq(@v}UPaw|{F2`cV6a>i6`E$3|eU3AN=QlqT@bF;2X`
zw<}qT`w%9BSU8xOGU=XnzNH(aK8zh^>d!FP><zHg2F@&UvM)a;aq8UFXiS2}g6c9y
zyj(p}ImnuhFH&8DW{OLaB`<A_GGItL=FG*$w@{gYqg_FV!wN`@D@dGA1I(8EHC8BT
zpp^Boy`!slP-`U!lm;!~(>B{FqblU}S!VcUEr$E=ho0C!k^Q1r0n5iM!w>*9SEp>8
ziCj&>Z6(;*gEDGrHRrTnScB^!9V#`!YatBO`NFn2;6|>DP^aHS0wcyVRvF=h<H?xP
zSkoiaW+GstYRl}v3XVV`R&GOB*08ed<fvWXbRX`P3X_CL>uq&w4?=v+5q3y0!RXNz
zT+-sdl<Py}GE&-?;9n!)CbRp2CMXi$OIdiZ@1Ox2@i`^|`0-A}__DDvxM;)7QE<((
zu1k6g6mbWYd{>~J2x!RFp$^C&3V&{@>vgk=18_s)`(T>j*4pA}Yz~_g5x8082&<Cb
zLDV$IY1YcRB;&Q)Q08vOo@q4q;t>c=Vvvw0LW-dRR)?sKOu#N;$V$E0s3vw*BZ3cp
zul)m76dqguJ$xdExO7ODhwl_H2cu|2jnUB{Pgh<?G;?U!ov~|MQ3@EonV&OtZKO<1
zJR0e(^9cFeT53G!9_+ThhkgG?V#@WjBGQ-+_qA^`5~iym11<%Hw~jpJvA3Qb9pW?r
zc2NjNa-_N$tyF2g>e|S=snb_9;99v+M5ow-6K-^ixro<^KR12lHTZt`&Tr0k2NgZ8
zS26}MH;vMUf5tEoYR$8Gx(D};->rm;X6=%u4HXjH9<G@ou;oqRh%k1N$2BtDnmCJi
z1$|y^UB53-b{#UERc~v&;^1BbU6}kbBY>jik^{4DJdMDl=pH!d$VSp!*dks-`QGza
z$|LsP);C07dmbP`VT{N(DE;}UKw%!$ND|&+zp5zaCGy+BZ2%wJU$o5t2>WJ6=K-m0
zuM9lJA-oMsxU*v#YU|wFw^c5@UIem6MkO)P+1Z6FchipMX5op764X%nu=sMe20Lul
zG~4mj!?iJo%+JI?rMjXQ4Y>N3VgACRW-2b0L3}F(OBwYB1p$Hn(k1e3KemySo>G71
z>)GT~@=w(Z%ti_-7Hw?Tp@>-WYJsdP!`?KgvQ5-U+XoI*nmhWubJ!3ZxuoSm==Gmx
zGacK}i4Ujcc^ow&Xer&ki<F8<!9u8vb*dmKFr~g?fyo<md_}pgqFY3P1aFO@3+#yK
zs_sK->teLuej&?F3}hiE`d<?~S@V>`!u8ejWyhASq(qr^O;1RO0)MGxhNhS-3A`_t
zG04CDj{?+YVpqlcfpOJT)vn^3>vR_H1e+DE4Tq)7ih&wTOZ*_Wqce!EC^JkGDc`x;
z1v|7?pNkZ{@U}PM0Z#Av(2VNIR0%5S@BLU_%Jw06hq>Y(dPMP`PvT3v_}PrWAM+FI
zJvlZl9xmZo!uReIDv~>xq0Ii-<!2M#H-}FyO02;crnko!jWHB<f_%?_s<}D_)4{Rv
z6~d^AEe346Cu)x<IB--X@{3y76C|@><n;$g3G8lZ#gF8wuss}!@eJgH$ZDhp&ZIDH
zA$9iqX(CxRa*%W;gYj!~=MN9S5*&mV0|(yg7Tkr2wL}5)DBB;6ZMf^p9uiRIiK0iO
zR;lRUDotWM8mTSVOfm!SmU+>!GR;?nwTUA7>`)soQG+yCc*RIR%swz}phz<x8I<LV
z0uT6rzZ@J)xCLTjapiN6-l>qc*B3&6wk$cS12=c_bfEHbMy+&9@!(4bR8S<2z`hcL
zNTn-id07hII-dktR<5k%CbRGV#CTESTc%CY5bF-OvVk_~V8f211_s8ciW|Z2Sx+u)
zGL)_X9b>;?O=Z`Yl5O#HApQc!Z78|Q0aLzUBI|N7wcDCD#Kesy+r+WK9PT*1j@6ex
ztZ_)(>qp(6&cQcFG@xODE#KOV<#}TfHtw&pVdKQM-g>MWegUyHs@fy*v^5HA8O;Pd
z%QxpgPj@GAWqKV?b+3<aEtb|j=RBsc+wH7UK>=A;iQMVjy+yH1`$v+BBZ?FMA<JGF
zeTGS4{4FhqQOn;z2eH@FNmkUECK%Lx8J)MYCmBR^ow8SfW4Id_2gGw=UR{bxbIrQ^
zlPSTJwiK50lp#o2n#)ei*v)6TOX8TOqk{W>N)r8!p_(>=o|)cWFryerLE7iI1Ou1K
zXwp%}G=pcqCGLR1(UJ7wD~U6(r(N<C^rLmgV+%acN-uI5*bEehiqh~`m56^F5Ysw9
z7~hVMG|eipgA6b*bq(W_#<|b2rD$ciI<s%`7-MHTVy_wAmjX?KR4k_L37*Wd*-a;I
zj%#2xJB#>UyC1le89eWF*V5T7_B0N}e(&rwnbf}B^nD@jddF7V?=-Q4rg0i@gWqz9
z<M&`Wf8<Zr8cdf!nbp{p-R9BUluWa3*}V|~C02HuWCXFQO~im8k}Vj}{_v2)uf%Lr
zZNzd*Q>Im32In^uLh}iK27p19H)}wf6sD?e<_F%0V%$#kUqbI&TFjH~^5?k>8mYR0
zRiuR2_~Wuj{CkS(+skvYz*c0Vr~~LAc~hM@$t@2*b)^paSm4WDvHohQ<hBNN-c;em
zE5L)w2OJ}v9c-{W;urDi0&NQKDCtoo<_{ffExWueJL+Pc%lAH&%oO0Z`cwvzk4jjn
zhjl%2*M`<uEuWF_oHc5qG?5%7_56c>Q>^aUDT4alvD)EPn`vb4C2%0~ril*k?)CX;
z$Ftf%im}OWytg3ZO?%!$&r&pP9cyi;+Vp)$RLs@-v_0epicY|%ioW~lsN$FYvP;Z|
zMoWER!^$?&e0u>Cn|o90sKlf8?zOA%4!!<Y9s@C>>`cAiPb^HqMowSrEXVle8;R!Q
z8}fxxEw!pM6MZ-hf>5pUVGFLv01Cl4c<F%<{?tam?M3VtQ4290D#&k2#cp>uGs3n?
z!RBtmXCI#Rr1Q$oGd=#Q{^k|8Y|EkF$%!Rr^<6YJmlG2(Jw7ekyPokX^x!_^&p%m<
zX)%wfo+mDqo)eTRJ=G#GQ7AdQ!b1y6qQJ^(>=p%7;(q;~xcz)^04rXH0@X{LD-J6n
zWUvU%`_nVe$gdpE5zUQv?MFT|vUL`ZOqT2tE)yiZ(Ab0D4T%mndVW3ZyYFb9&yCRk
zCGA=Mi|@<C{y$$4FKbFUZ2^#b&ed*Ol!t|He2Ws4spz~`%Y`MXIE;VK{7GO(w_+_~
z!X01udb2_F2Ew&B?MkP75)q1n%bTyQ?Ig<TlC!^?O}s{@(Cay-VH^{uP8?=3U_#O+
zo)T$jll7H?S}pCR@A<xC?H?f_2_T{(rCCty{#ad~ja0LfW2Uy3WZ;|*8akcKouNm!
zo88Hx--2-G?REcL+$5mCNT6slG8_OFNE`6ViVK1wg^~cB53^6CQLS5Ew-|wx(c4*S
z6}iu);4(aA>fKnfdM?)OaBIiWT}F8An5wKC=+-^Tbb0;PQ!N)R8_9Pi<t}eA?@VE`
zWLA0*EU7O#7nknD7HFB%SDp@TdWjk(OrK6qBk~S@m}(%*;6|`x@MrWB6*;2m{Wb;p
zOYsu2lZRiF1Zga+oK#6%>Tj8O63QRU6)2&odIPluM^dCz?%S8wU(Q2AL9}}NVN4sj
zXQQ@k?gzP<gf{MS_wWZVl9(3xUKMSitT#3z!ML-F!R!f1vMcG$q*O>~+)Fz?E-Gs2
zN^|_FkTh{2yD3HMRrd|lHQB(sN9W!u))6j&TdpDQoBO1J%jNmma<J%@rBCRVxz)iN
zdb=!_ZEP~{u<xqV4Z5|&Cj(>u;`M4U%S5?+=>Z;(vEV6r8)iVFARtTbk*&FBXDuXm
zcmwL8_J?f2@37Rx=cWhX(<N=^$Mh7={r&n|cvH}x1D!#PM={TTDlt}2O*H>hX^mh?
z*quLL84yxee#vzH2Dx04X{RE9etzqi+Fe!gJr^Cesi&@h`$r2hDM^1NC*cW!5}`vk
zTO=A0D$RQ}ANq?6#w;n?h^W@}8f<17d4{!tu6yIiIKhKg)XZBNvt^V`{2F(GkWRX;
z)r?E~sijbTV?wSZQuSE7=+)#?-Z-b~5B6afoz~_bfh*|=xG$}y!na@?&gp(3Z{PFO
z9vBMTB4L@6jKPVha)IYMCXmDNN~`Cd?EzkG^wvj+ScP|6{UvzEP+;3R(U6^I+6;?}
znXQ`B&JqF+b`yQ5srr*nWNv4qB3AJ89kyqOrhZU?DoLgoq-KYSvumml6vSA@KVgAw
zr+J=ez|g?3aOw}lE%3k*MA&Nc=2dY)byt@UGOclU91toN>GDzhurbnHv2sjs$_w4^
z9e!e{Ogz*+XpHifd;Zja!YT}MQ((2$|4<}Pru|C;*23a1L;+_t4VaRw$T>bQ1K;8h
z`~#XLfKTW_sKidhPmmQ(EY7$Efm8pNy+RDvCu{yfz%$wFv4VU7k_4y-Nc+t)dQ#;b
zw)PkqU#@BwSELFH!mY^{SA#|gl?iScD(gDsXueDM@-M!2F4e7{q*H3VuNSDrNCR5r
z8YIY}zZ{XcZESo~(lW@Q=<pXT_}b`{)ZHKEk-lJMj7_|rsai;OgN2HJg_fy(4-c8V
zBd+&-Xn&5TW!SDkxH}Ml#xLz|-PP3m!A*Lo&L<3UiB2<*WV--uGOi#kxld&IP$b|`
z65Uh$XE6%ykou!NiiPebnc2XC+Z_BB{<|qUFeUy|b>gfVHFBO?Wh!~!>$_F~#<0I3
z_)?}By*8oSLCfbI49nw#WBO&iZaV3#cbJTqQ|0QL!KC2~ZsfowXHh`0K+szS<|w&Q
zblHVI>;m}hA^RmD1NPJ%`r`J`27cezcl8xot{t7GHxY=jhm)c%?8q7^V0&_j1;O5a
zGXGKzPRk?8<K<CuL=IL06MU<hTk4=*Cd$as0X$xQmIUTs-&ejhTri$&JR#*pOjpW4
z*4-m|Gj#oJfrZ*j7W^y04!4&K5e7Y&*%duE@HBI=RynDEct6Xf?7shx?K7a=&5)Oe
z<a%ao5rOS!IXY7lpa62(MzJR!;eTQgNNOUm8qMECB=6u57N-gZtMFHP4LkT=QB=hG
zbt<KIV#JtgU3Cd!bFn0ne=02;2fPISkrGc3;uKE)eFO|iLjbH6@fzFnx-7{C<YD>m
zVQk`lk^Y_JnmPhn-ro4~*<tIhgCiV)&1z780fFX1axjQMm9Jf5WTKunid#eZBY}-u
zv5f-iMd#&$dN4=B$?M*vBnOS4D5R8VT9#*o+;CLP%a^!-{((*pIJ*T;<g@I<7N(-e
z60EB~Uk`s8!&H3!CT0SZ<p(4m9TcZ=%dZw-8uLT%iV!}ibQ>3_sm2pkQa^B!ilNO@
zM3J*u@%TjPEr5WJ2rqB#14!dvaiil&{6`y;-dbsPu_y4!f?<SVQ#B$A{xwAiW_+_N
zQplae1c5taak&QE&DKAhgu7<q$Q<2d>#@I;ejn;9Yt*gm*EG5)2xG1H*r%JK1x7)I
zToP434ZE%v`0mDqLeT3G8B_=?6rY2-6?&gc<3PYD=PR1g?@+F4so5!;yFHS}cs;gU
zAImhe**TvdM{C~Tuh+BLfzIxBIj|6)ePjcfL1=eL(8N05w%j`s-H6U%8{7SP!jIrB
z2kR{3LSaAj1m*2`Fna!kJb}=TTMMv##$v&u7YBu6dv0_-H95Vx!0#Z{l*Kl66yq{s
zafc*L9OS(+K$85$AdMvpjZKUEG5_fbPrXo+{2;&b9R0rJZ~BIwCNW!3pNv^Vb?H`?
z;kdX;{j9>}wYnZg+rogL)*cBGQ7sIt$a00ere`MUWSY#JFdZtu%Q=3Kwb~FH9`H2N
z|70c@54v5=^Vh#)1fN;mV01W_bEpo%MMjdU4;GFsHaT}1n0Bx7Yh<Po)!L=7%d}Ll
z_34)*aG%xc697#_eb4!$H4&?`e@nfSvCS&|(@S)9WMg0a31hHS*a#*cTtCVN3|e98
z0)C!$Y#efIT<-i^iZV|6tJfG|qFhKn1{c)ngF}Djswl*J<&WlrYU&@TXr8fB^IYw;
z{?-Gy+bF}3rfP%>+BT-#E*N9w%yeOm_kj1s`R}+r`nWh#JIfiA&E6w1$?3r67;1F@
zqvotr{uQ9cf^e#Tn0#H!+EO;q@Sa=MI)Eu`P0hOE05vnwYP;Pr(eI&qR<&K?GY1a(
z%ccI?r<whgfcCn2@jvy`e^{f;j4bT`>lNp+rj;FTE81s`9wUL8f6Bg-c#Et5P+caA
zZS=;kfPwMQdhV1G-Ahic-;;N>*ZW4MV@PglGE2XLfu@aK+n(=soMy>FRa4QbzZ~yA
zU&q<#F`JXgT8>>@zVeY>ij98;aF<`-&n^QL1}q&rlv#DS-}jd%Pwaf(PEHekcep!L
z9R2MyJ1L!_(e3s{(e3d>s+s+H`}{a^s~LZdt^LB+{i5aW{kmQM8Kv*d`F_mAP%KjV
zS<0;-(n!<LT>Td$Y7lzPMo&|TNuo-1)j=ETD^(ND3P|L_%k%6sP7-XXoUXX!+X2hx
zWx?p2Zb3lEMA-MNU$@;<X*>s<-~8=Zx9N;*p%0T`q#7wjs|+G$HA*gsGB2*Yw}QOW
z`8NP6_g1$CoM4rI2hSG_kBgTV#DIy&16!X2SMnOHsEdm(c|PyYhlj2zHE`_2)xGU0
z+aZPK{Ed8)Sd|B^gc%v(U6qR<H)|<jy;FET7?ca}X+7s{-9Pm?U(#{weA%kn?acIa
zSRQ(kY1_(^xmaLL&2(uD*o7uX-)+U1mLcdYggcy2t*$Am7ZzL)j`FO<G`kAcfKAk4
zIiLH|u}*)%*a82(6ZIkR-L)B@pC+39Sw)*LOSp7{dqrGjKzM+)MU59cajFHhR;w97
z*_p4TvQ?UD-NY6!4G3<ur|aQIlbN`aHIqke&K<RE>a=s1v;YjXuWz(EEo(O)YY^l<
zr$sIrJ1t#9yV96tk`i8%<|EtV%g2HFwkrCw<RUwmoA5Kcf*`XHI6J|XhA?q7p$p7)
zp!7rGt9yAGcv;CBYMcbF-#h$AY~;HUSBW)RE%F&C|1Q*V!eR@;zN`jG!7^^ak@Nl5
zRAw4r7*Uj06k@GinoUgBH=91&>qNGs<;W)|&x$R)Qu_TzwJ<a{c|?Gdq0q~ZrbYuU
zaFROxRA5py4JNHRg6ZUkEBq@mQFVCOEpQ$BW?Z@D={2cI)36*uD&?_p$B(^UJht37
z9V4H|b`;f1bwF?=JaL|jfAi%Zk~Q#medM%_`^%Q6JQv~}VhScFj!iAcB|ZkF8NBIU
z{_7a8gHYYxZJPwYI;;fm$KQRt)qJT3izdzH^U3XG7*b30lR@!)n>K`Y2sR|Yj%}SV
z0>SbeX-XPUh^>*Vjw`p#60LV{1B_|Qqv^0s%v=B%!;mVp{>r!qrJVgShZw*FN>d4m
zqgB`{-3{7ee1yd`epM9%=rvnTD~8eobX|K^IMm(!(}1G_k!ONDjaXWSG>)}Rd1<Ct
zVHo3+T$_jUohz1=xX{8|Q^1XH3wI{(CN^W-`>q=cwd>X${HU2TK~SukgD@^L{p{HQ
zQ&Jf7>3M&3*S5ocF|;-F@vJ5*1RMn?=ohOm6OobNOE9%&1e*i{`PXmT;|fxLAxm}8
zJR{=cn*yK&^9kVDfnfuN97C{4nqb8tN>y5&xrMQLyA`7!L;0zE9k6;?v;9O0YqZ$)
z3sd;Q35Vi0(rf6=3;lkNmfr}sjk`bUIev_b4m{oniXHk2g|*!gfm=M>)BXm)XZ7jU
z1d^)ppW8H`2b(eqi+GIGv8}Zb_lps(W9}%9!2ab=-h>n-72g%a1>X(G6N4Bc1RfD9
zS!)PorIXi;&o51{Ty=Q%Ogd0#Uei$NqpV{@e18^3(|cJr6QocM%YW5(_`VV<4X8)s
z)^Hs>nN?9d=*>?;&|LlShH}KL2W6p`)^|J)I_oqMQ-~s=_A0-LtHwgV9o0!~DN{Vj
z6Qr8*@`>gJ<4E!E+2wh={=PK^RX~E*RFLd+TfG5PLxhg7g<y;SZkM{g+cuuj{Y9wb
z$3bSIW_igH3TuplcA}H0Hg(F|UFH$suL~D0PiW?bG1}F2$^Vd##AjIQ1hak5aw@!t
z3rMEuAULb5&DbJ?yBxn;P}8WNPyNfTE))+t7||1s?2H|HZIG>6_kK48{%hr5lW?(D
zo`JTO&Y+&@2!xhSgC^bDzP33dBBaU6BM^49k|vs?`9^G%wJ4dc7>9!XubyE9qKdB7
zhB^xWD4)%0)&1!`?fuV+cT6q7t<2wtyo}6A3(#B&?a8skj!74HBtl5srp^~ksDWv^
zFI2<uKzZ=`k#>?@mm&+EgfCCUh(ckvSHggsch}C2^Oe;6Ns-2#s?NwhxnPjPZK@=^
z912aRv*BRetCQ)s+5P7p8;F?I5j>8M{FR!P)MfYYjqADKr||KRbEGTMUxrZ@0UN?>
zOXI3TGUTCDpWb?LQg|9<chiri+KOFQD>b9)fJ{g_Paur$G_@6QR8P0fmU<maZf?#4
zu*v*=>2hD)s6hLKj8V8md2jX_M>L5^5H}_>(@a-piTn=`OClE$F?XkXsN|V`=}_xE
zM4oe3w>EzesrcsUPWP`Nkk=DKN`}wMe4q`K8)-~C61J?$HiF~_ZtzzbDYfA8>8UZN
z36Nwywo05<?lR%?;Yn9F27*;^;#2RbS~1tSt%`XK0=*X*DMYNmY)0@LJ6y6L8tXL&
zB7L@On>8ZEUhu3f*c+DRPH%f_cqrU}m$MgLlA?1PHf3SD=o;0_xl1}iT+_Ll*5Vv3
z`*wFE=wB`7y)3J_s=nu{5<|`*Vj4sFIjYp6S?oDz6eECLGXo`ER7UiE`58$mC}J8!
zL@6lPlzR3z%?A!Kv<vTDg_C!+B@WEtfwCblpG#28!}#3EfH{3`O-8+c6LS#*QLG17
z6AJ1y-qHqye}+!2IE*Lf9EoY#EgigwCB4<c07q;sDX_|1KN3ESAD3fIeSvyR8G(sP
zpb9;fE-8g+(BC^5m|_8nwZ(G^YdFxfh>W7|3<N2N8B7X1lk3k$V|YBSS$iYEKWLtj
zlvoP`VeOHv&&c-vN)KTc61kvhrZX$}6`T@@3ONjOVo;UMwk2kRpeBJTji+MHI)zVI
z2L*16=$E=Nz?HD(7}DV=fG9D-%dnmV#V}c36wPMkvCv@!O2RRST)#BTn5lkujLNic
z3{){SqRZVcc`TYQVazI4+EuLa+=6+R<e*dBz;#+oj1$3ONMnuP3M@|wUd~<g>iDi*
z$dHV*ZiQGTTzcxYhb3_XtSFi$@3$~51a){I1<ICP>oEHFhEimI%_4WGem8l-NEkp!
z7^g`RA!du~YF6l~H*)+)A_9CcMW)=~bURnF4fG*6BLLZEIBa$#5A<r{eHB$+H&zTV
z--2tKjWr^@3&Oqbq)if>k<OMn6&#AM*=4h1!`u3bv{DP^MW5X%1p5LN&#C#rcC9fx
zE9c84F<vSFH!hC?U0o4*p4i&EaK_{7JfzuY+S+sKQVbM2Kr~+A>2w!+b2&T}VnBmh
zw=qG3hHpUF<PIt3Goo!M0sj0VBCf(4Y!pV+<|$)xRc5Y=@>4|#{B7kHFcLNH7^~HM
zDhw0i??s<zDktegY#W_JyXB2h7$U{d*}#!eq%g4^LV+$}?7^7}E1GvT1y4E<HaI^O
zE83VQb{`m()QBZz$xa44V{|DWgB;j=aF3ZDD27ZTM?fYT!sLKY$PpKVG%zL1+>;VX
zcXU?%4W3{s07Hw3OvEj^@=k&-96z`jP1=;=N27lrkA20~w){<}e?AJUe*2A6*Guxb
z!<@j0UKfT)cX1871|o5}DF&s!G@WmrGkh@?&;SXPUYQTsl*`r2T6oVK9(DB$<9MxB
z*6`5R1?wK^D00w^4b>#51mFUbUlJ=Z+-Dg_8kf(*Ig%acYdx>g(R(4@4yo;kdiOo?
z-0~6XR0rR)?S@*_>FmBT%6+Lv*a3Y0eD)X)Qh1;(E=BrniZ(Q=WIO{LH3+P<H@$2R
z0lj-Ox^kB*>E<=p!<uK-O;72r)=$Pw8ityN>koLwXDflkSCL|Bu$%9&0KAYtvTYL`
zs`r=~ngZxH(2Pl~b-|bu$!uUp&mO;|!}*cYe-Dv8uts9_w%)r$XqmJ!qS}7)qSYj2
z(UG}9hzsRjy)9N)y(jdOpNJ7wN)qp^cQ;lpIbT^Nk=;`;-~L<<cnggb+7^=g0}!b#
zUYhba^#$b4v3klfT%9s?Bn*KOGh1Mv=<4Vdmg4+k>FpI@F8R!)Mm#hVARn6WKliL8
z2$Kvlc*F_0Fen?<E}tC}40wx{O&coIMJJ)%xynO|j1OULPpp_NhQ)6)4?xAf;aEp<
z#{TQ?g!MlHm`rT{dvK%sr$v46v&;NkJ6FivjJQS5EMwNfwGPO%STOxeIsEf`9)Ke8
zAbGdZ+1o<^>Hmu=%(Yb_M>wFbzkNUJ8aHgl*39|o2bQm==lv)ekxnA(#M9Z64#lIS
zh(f)5;mP`LF@xIuarAbnW{2sr&hmc}n1SMK=Z}1PpYLy(1oRl%^BjeTj`=Lvy<hw9
zZmWDfZ%^;%H^8&Iy<Yc&vEQC261sCgqs)OK1>Ji;<orSSU0aaoK^;7NVf{!L-ocOo
zrj>g7a3?_hBXcpgG|q+pt8^WU1gPhvrJi0fPlk{C+tSkcGA}g{7oLT^Oj-tQ9yJ|F
zq+C47Z>JcV{F$Qeo|q!(890(9OF6`?5DUCT^+x>=N}|C$*&n)zk*HAyB;Kg8grQ7Q
z3-b(=*?3(l%y?SX;$$$=z>I~&L@+71bU!I6-TRgjgjI-3BdNj<Ilg{E6u3rRKyhu+
zX`H<t!1SJfA7ocU&#V1&zl@d12y+yk-V~=)x$fw6yGZZcGQcJd-Q2Nm#OB0|g1d3s
zp)>nKyDr_E3QzijnLyDl<!iKeOYD?c0ASE_Og7~~W66gs^&}5t0g<BHv)>VonT+=Z
zzp#`@tM_e>n4A<f*g`HO#iv1v>K$WH*ezh2n#-g*bZfN!OUglFLOd#k4`;dJ$%R8g
zabX<>&QpF1WAVi&B&I2e;{D<TTg@pH8J}P71fmpqr_d+NYLp;8Ib#_nbyK}1#Zp9j
zm@6+YcP53#5@G3nD2Z2EUVu-OlhoUYJsM>%$T*stUN+YIA%K1fz-9j|m!<rh!9P~t
ztt&a=mVHG!s7rm6C)mdp>~9ARtcNBT9^AG!l#!NM6Tq+~Sp3oum&Jhn?9o-}mO3?a
zZWKbhuJ|r52{GS4ix~uQQYa+3E+a50&L3vOVq_^jJp1qYwfSFxxabN*Ws|}>uKs7*
zVJtpp0T~qA-O{$$gmF#oAP{mI;GX#k(vx_QpaSz`P~);h*=RX5)Y7Y!{i5RXt7O~e
zOl7`!|1INb;zM~&ynl>Gl%(H993s}+9Vx<mbT2fD5;}HUf5Bz1Z>sX$%l`P-rnw7-
z1Q_Mi15NjN`Vu`V_(Z7j!{)X}OJ1#)b#s#eUxMcRlNYvW^9BcO`5pmb*;38G1M}JR
z?bx8*R@J2$j(O$r>B&Bo1*&pU_E;9ENb`jAz<KIAb-4rl33X|D7q4QQh=F+fqh!&s
z^VSZK0IR3|QvO5iE>$-1_GTR1_NC?YCmuON+QKQRkOn$N6m@`waF#nVh7WNjF{u4@
z;8KTB<f&J?IXYZXZo6sI9HZA7x?M^&4vLKQoJK<Nsmz(GPel5?`|X~)9P1w`SPgQ_
z$rfZ@`6ho2m<Hg&_tLn*3~}J-ynRy8FpQr_+c)fWXREmwv2|K{nA4>{&rylR`J1!v
za(wb}pvNRHbA>nH(M@Vt*@+f}>Nl2W*sx<`o*@LT3U=bq%=rS#UK(K!GoFTI3Snxr
zY<>=w?Bqt${u<M=$R0&c!cLF<JN<61bWO1w^S6wPf}@>pL8@N&!91Uek{2ZzSkof!
z2H2_fkw{$4hA%kO#mC8<?+1p2-b^QrpVh@#vE3oDm|)wHYKr4)kY$RI5|SL7Dm&w|
zWmKi)=!V}gm={Rk5~Xt8FN{wXXA{%8XA&333t>E$koQZSMM*%-&mQAEcGc~&W9R#c
zil(dI;VcVkHXGv&x^w&Tj5WXq$Lduk*eBO;L~8s>T72wiAKqRa%wvaVYq#~?bHWRj
zkiKp~e|?d8Q$5QmYp*{8)uTc}tr)hQA70H@EeIc-*H~5yjiN94d`F|#E5wFekce%K
zr!h({INfBxV|1@)due|hr`*Y3NoaV=Gn+rc%%hc07&`Qgm%D$<2m-VMEdk6n2D7x)
z{TbW!5%{RCv*|wc$bj#c$n=WX2X>EF7zsh;qY#IqeazsPn7vCLV2NX(80r&=CO)AI
z2wFuNzVQXgR41EiJ$|4njaGhQL7C@!ozgQ(IfU}<9Qk0tMGI?4h)urFVvUu4Uz2;1
zmB60I5{um&q8L@I-7fBM1dP$pQ4jHbBnw-0XpkJt+#*SEi&%xxrw6h$0rRTEL)QsK
zgx%7Ln!NHxc(Tu=r%Z-uof7cEOGQS$9KZ)tE}0<$6?lgMzEFNPDfzn$I@o{(sDfxI
zWCb2u9+#{RxXRuk<)$V}=H1Fd4i18ugaG#$g5?6vF{HbwlpOqM$DUcQ^tZ9EP#c<`
z@aJn+LWqwusN0oS=f=^|YJe9dh!riUK6do(TU>|0I$9+~#z6o#YzTk5pSs(T%+L1c
zAEg3a+z)u0_AXMxj*g(2nCwea80*DQ>!jKv_J8K97f%Lr8a#PE8#3`LvZ{*t3K>Ox
zIbv3L*~tdLp{4mLl&G)mtphzi@p^{%r8pzn#znljHr&#8w4J)e;M!{It_Sj=lE}{N
zj-^hH)Jqa&2{No=C{&i=E3Or0h9AVU>6ooD;HbbcKunYdEYNkfcC*bw|K`JIl$1z@
zbL)~t2Aw{6e7j<(H^D5~KAKpWBf4%tMsh45;ppSYgE3}d=(pA1=&dyT(tjC3SjTJJ
zOK(n`^ltqtE&x)=AC8Ymbz>T7Rl9z^d?U(lX<csc*;Pi;D&g$ezz;bR#87y8oSXpR
zsN(V&6a|Vmb}4Oo9dSQ@iU)wJa&K{0tvXN!mblCGecO*8yCG5RhVFI!lI0PiyKEuw
zwS#xiU$?#5q_wPN)Hcm<VJGf88j`H24^>&s1c&^f7P}W3^41L|c}4<<?3a0uB^tH;
zT0Jt_XW@byBjJwDa_2UfZK16dWvV{YxkqXqA0<8L@pizv(!mgKr6$f%A)2rp3nC!p
zqM*1fk`wbpH4&p;%6FTkq~Of&tP#*eMP832q67ge^8GFv5Pjo}-n3;>EBNShNR#Er
zaw5j_3kS%fn_Gr%VAda&duU%S=PwLD!7%kTZ09Zqq^}4<R7h1ncXbTNW~~{Z<9g(>
zz@wCpA8eKNn_IZ)plTxaSnBl4DY%3RQDP#%pA)<ZtwrOLMEb!s-?i)hD&=SI!+tf*
zE$#Fl(}|OZPs!?c>s}V2u<!YMnl8icAB1;@Hhf)-U$A8R57YNk%lb?Dv84H)eAdl_
zYdVb~!+dTc^<~dT`Ee|2BR7n}ZrS?mvPa@`*MUyXhe&Tpy;nA}t^ekZ*^O1t@&yot
z$7UCkNQrvty=+Kv-SDx?pzhW<mK#j(J42Yh*tnWJ@Xz5d-*4!5Jdu49g_)KXq#>X6
zER9Jsu*&0D8A~0uO_k3iPXV*2QQQi6j=HzkOsi{t6u7+)cKNcnaU!q8QBBwFNNeAA
zAreTFtMtyH{&ff)psuKzSWzYH{a>m|$Hmz3B+)M|eSOa#WExp_tp(8RanTu9QlGVo
zR?xl1?U09#L{!mj2tD!pyj_nvKVxVtpJd|i>etY>0usdSvf{9$wgfB7R|{b?z^u8~
zTDf~V9oPy`yZZ*9);eAMGLQ<<j6YPG`PmB#W1qJ#0HJL*Io}TJWG7?D!}I%Bu0Z`k
z#`D9X-=;s0SX+Ik6pd;V>WLihw_U6f0#*Pb_xLz36?M`9uih2@E7-fwa-Tla#p?G7
z!{;xcPmBbtK!N{~#DM?k#4`eZm^1(Pfu<L=uy!_aB%~L$HgGl(HZig@Hi72lg?4gw
zG%>J&cHg+vkoxH)A@;nh-B8L&Va5(yk)N_WuGu6u<!o@Wpq%WjITpy$6sIhbem;Hz
z#i0raC7e^{;Csf4TZa4ibiM=sUCxY9<GcD`*jeX7-BQc(AdzA1sY)tm!5WcJJBU0x
z61OLZ-K=ijFZ|4n$sx$%_Nv#CnidP~Oc9lP8tweq)Q*u`7%Uuo==O8=_P+U$K*aj?
zez5X<k?P?4^7!5}eqMe#-61OPv0R9<0+~!Or45*aw~N~5N|IEq6XPHX@TLtm3fQa&
z={FYkyo?=~7>de=&M6zIj2xFiMp1X9f{+`3UGXoKFwRq9wvkbJMz>;-6!YN-{k^#Q
z+BPp<fE?k`0H{z_N1ziQsg?{VmHhu0dj}>#!zNj@ZM&y!+cu|d+qP}nwr$(CZQGhV
zdw1i+iP(ua-}@8ZsEVv7D=YJ&*cq)i(hSFRL#WXZZAN(9yl~w!jN`O0%_zd^a?rIn
z7-q31(X0yH1%XakVzkJ*z+(awRG>@Q@+9!Oc=Se>2bbD`3M)$;GmykMu>~S5@O~zj
z3n=nVQc1yHR7EA@H>v;8V`X+~t(&tyUVbRWZAGm7Qw35=I7Yhb^7xD`>>E?Gem#QW
z1JZtx3Pm5+M<<H-Y*v*;zu=zdJHSf{BmM76*0tRf6-Xq9w3sPc10YDD1Mk3M*<xqr
zEk{=dF&k#FY9Vlgz#xd&y#=8HA}VwCNs>6Jl_xI&914G3C1omuEtXzyFRpetYYT-S
zOk$Q*H&kZ<1CoWH{)jBffG)+G$v8AL<ONmYdD)<iUkf+!8=i#Iq>e)^*O*Q6LEt#;
zrRVm5@q$BI(BoE{b!fpVfMX%=f@w^WbJZY%vYGr1Q+hxqkW+iB)y0CH)(F%Qz=H;9
zE;2a|Gn0<|^oL7&aB}u=_%{PWC}?BcPLxVPNj0!F%Wo((46(IoXcE?BFK07EhGE#!
z49GghbK~^+^F_|Z29q&@4QB*mcJmX|i89um%{wtj!{qPOlI|_(!LMU-Tw!XPR`dmJ
z{EZA#&9Ik7Icqc1n-U%Dxw_psGo<L}HwKKFQntLMnUn9%9dGnpxwB|&tETdGOU2w4
zdgs@t0m-a0HKKYQ<kbp+N(_R#<&ZJ0EI&P3hgn`#_B-q^1|t3*%)>cr6suh@aR}#I
zKbn@~I>ml}mKWY@4j$duO{`9@q2L9R2Z2Yi%>L*sSzwTB$XX-=n~9VVg{J<dr-=+t
zQ$V2<IMZ)8?Tfy<bhDW!HvsXxu&TBmQ7(=&pn8}xi`R(Pl9kZMa}N_twcK2(S~5Yv
z+#SA|Ul=O}Y4E#%O1AYL-xr5m;04?n!d-y7G}R7E6~uQC4i}eOF?XnjdcKs~)hO3-
zgTmn#c|`bS=|yK72ssE0WL4Y*7T9u09v~RkK1nu(Z>i71i1b<Dht2^rp%+S27J2*b
zb>9+%5)Nm=MkHb+*B-r71WjEhokTqdffE9A#NPWkBFZ;30;au*xo-LD?-0=FzK-$X
zO)wC{kzg;^qUfr5sI#NLHOYK4%tN|9;eLx}wv*{2_O;;d3oqDCD9J(f>&mVR!V7V3
zxmycf|Dj0FFd%#`UWT=HMKI>&u3Y1mNiV$a+s^)Y?zd+q=1i9*XCxFiMm4|y8j*=t
zv%TtKms#I3VCL1ZEEODe+ii$~`w7@x7?%|Pa8w-@S>84`pa6eEa|;`w?P)%`H33YM
z0xk=Eg+>Cdx0*0zXywq39Yf}B*0$7XW25VS&YYjAue8bH_?6dyi)HXoED>hVS?7DE
z)BV$x*N=w>>u*=|#LV&MWhWQbNN`2%k0gQCggmS9koRM2oA)1_9cfp^vThw_HvGg!
zP^m7;5+rUp%<YgPU*RMKKm@s1+0n9|bi&5-hv(BKXoL-Vefj>hbA2|1W1*bihXNob
z_>1oZ)qM3!%Ccr$WGPKPmDgS+kM`?<P3tGYRxhQ+cQ_p?55Tlv+oE|ffdunX>5EeY
zYbnf+NxwB_nYT>}B;ci(!#S97gCIxf;+3EcRTnso564e6n!qJ@lONcMDT?ye2w78}
zEPQ^i-2&MlDS`?a!*-J}E7sq96=_3Ct#fpvk%iSkVh?0U(*wHn?m)W#=z>C@O8XTu
z271iGmpE}sfUgwMBFeav4V2^N*!`r=rXH0%7SlXT*E;t!Mc68Y+fESdPW=$iYZ=q)
zukV^%vq8P%os$bj{S&6x|M$ra1M6Qym;Vr^iq$q1HpLOVXI0PSi&hfb#T$|g${9fv
zIsv8pU&D*E4Bf4i#u8MBkjGYSXP?H@1~3#fx!r2EO+OquK%Y}^pL=}rx_1x9LuE}N
zAJ^jS(7<+#_QS`Z4}yZm=Q5|dHFFsp?hGGZPa<5o^qBX4+No14+m*n9aTWfYl$#wD
za=e{WBj^Ywgo%k_lSsh4mW^^FCO|m-btNspD5~ZEIHFy|CkXp~l2M7SuxC`=r8o|5
z@D%E8S8U@=<os=4CSQn?3jMX6)oE_zuc~^ib*x@L^!$a%JhkG{NN|6uR`P1eLw^g<
z2giy*3=xI_734*+%EId*ze0uWT}r^y!ySkh0zIiwqq2})`OrKP?_p)Z(Kk){+t_A_
z)JFsD7i&X-{prMjxf;58Vsqn3Iinve`voipSpeu6vHH~Zu;g+Bh34=xV5E0)MLVU!
zY5BB#xpTS+z#)NU4jhJ|&+0q|MijCBST$KUAi6L_Pk5wuyg)g8$Qg!!p_*S)bsfN$
zPy&6(fYn>yZ^bT%7dN?J_qH8Gg($h7D=da6@v%7;Uv(i+S~$Hr^2e^Pa2>coH1#@x
zm(X-kAH|(Y!9z6mXjpGoyld6{1Tb-n7_9U~W==jnOzmxiEN^F>*8N2hOwb?_!~yCF
z1AzRV38-|eYJ02Z5fv(_r;e6I2koh9LA6Ytd=d;~9hj1^AeP|8K72!pw(`OgyWVRx
zMq<J%0jBwSF?jG&(E76-j*Z$-N11~%S8y0Kh@04{!;qdI*KaGAv{{QWp=I{Xw4)7h
zNdWQ@PxyA6Z#N>RAbf79&Q>j;z=f(bAs^Nhe*k%PSdH2RyUA(-BKxWPhH%yNOrr(~
z*rFzIC+zqjaBaZC?Jzbd<T1C6P9eE<2;T|?wh#_H>$SdtR;_ZueKod}IK8J75sv6E
z@dVtCSld3P{{Dd-8dG_>4r}?e9|erV88x`|iY?**{YGYOJo?^|H2|o^TZU>sqcEZu
z&lXe(@i*wrmkU1xIgCOK*IKO^v_4i&$DwS=(PgIBpk|dA+sp=Ik*RB_(hqYMD#z^{
zbxMTOGcKwY^RBE~YnnMD^j<sBw>i{hCLwALJ5Fh@Njq$edXcqqDZd%o#uP@H*OCg0
zEMT;@&um9|iMDdT)&||{M87dsZZNTcmbyKX&nFzYa{F|ul9O;iX!&kTKeVa|K<F=S
zVA8T&tz>^bq!69%C)PttP3Ng@Xyq+q=UUwKtFUzXb^VzoaJfgvQu5iG<sEDQT_(y7
zJNU;bVl{C*X)Qb)7{W5wT)scyfnr|_pJa!Tk@1==VthRUqV)hsP#qH|7IlMO-2N{V
z;&<X1o%l;9@&io_fQLUw4n3^HOHwdKTh6zkaS(ZG-(8anT-*kQD>=z6piH5rP^m`B
z$LaQM>1i0OnZ=L_(Ux-QcJSKdTEe%}a&Dq9G|3R2COk^9%GFIF^n8NQxAa&zg08>(
z@i#fUrogP2+))sJfmmlkuHm<~VgUi>M+_Zo5G(GIxM-e~NnrRx0#2thXj@G7u4kIo
zG+N+K%h8{lLH&(ru=OHTeb>DIj55l{E-k8<<RQRRa8o1~-weqRmjCv1mAL+eYq6>E
zo630SK{3Zr<p$VDCdPD*?eY@`lLSK~FvGC<>%7chTgg%ki)M8UJpI@qrSg@$%hA@i
z++OTcoIR-&FmivjIZoBtyiD4)hVQijBaE3bmj!y2chD7Xho|b!fiK?{eiO`#b8_<O
z=S-}VH#PdIRsdYr>`^-JWO1hF-iVoako^e^5)vr%?Zr;0+#|ONr*w<V4{>Fp0guz0
zAq4`z^i=VzE-&APaaqxslecjJjE29ioW^g3_WJR;iTa_m$33T-Nj0)CR@;9m+<-@3
z##?r2oiCLvv3DeO^z<;%w#~`Mi-(hAyYT!M{FOqB@lQ?xUE%l{3yHM4G@cPG=Yp3;
zlH!}-M9MJNQuo1o&c*oi*~5vIm$&_gF&k9oJs=5xK{mg+zq*z&Pq1i>Je6K0&KwvP
zAWjgS!QVx?g%eK)dv=!2?pKRFwZ+>n6vW+_F}M<!Pvy$)VB1NT&pZukM8B|~SWrzs
zoTkn9Wnfo^9|w=0jP0FL=g9JoVVg0DF~hf~kP5u!KM9cTr!(hAgTAm}xk~@SE4F{=
z$k`eHM+Sxe|9)0gt7toB@S}Lo*6ztO88=R*T8)^1q0&U-Q5a;%9MB8T(X;3QaxKdE
zdZDW!7)4J@?P9;|c;0Zl-l!9_?&Gs{oT;O$3V|?$0;+%nH?Cxr*)<dakdJ2_JH*li
z0RUO+sOVL);k395gjC0uOOs}B*wR(cstQR|mHcM(YY?nBU2jMO)8>Z~pNktS2cT?A
z40#mcrgQNVuWPKj8G-H^li*(csdo(+lJ%l%$!*=g_>|66#kbPd^_}vKdMdZENa1~J
zm@7%q$NcDG%XzkaT(qBE`taBQ3^jt`q#QNGmk1jm-k<q77p`gD*w8l`h4lA&4mdht
zlx3TIsw5CaP4SY4E$8oR+Oor|LgIsQh+Zgw+vn#UdPgU-i!3CA2=1Gmt$4^vfJ7n&
z6sIhSuBAx2_P8_@>4;CgD(sW-x#sp*na22_%|EI2rYp!Db<HT3ecX-JCfisu$L8@0
zV7!_gOJjPb2g3O6{5%1I6hWJ+)F3&1BQRG5p{qntQEnk$vTh@B>|+AcFeL*n4pr0t
zCm-Vz|HU^^Wc-mSjj#jEQMh)<4cjawrFtqs{XWQ#0)aa~k2r-s!`xyw20-f%3oSrS
z_L{D;co`4^;soX@L@t#QYyGZux1_+B(k%vBPxZaLbkA7=jSxu(ZAfz)*5x#g5oB5|
zzV`e{RlFZ#WyP73nwF^iNQH^;1Scl4KVx}?qN$!#wHE_u9t6%vnv^zHu1eCA7>ZjN
z*wW1Ha}NuX>}h6W?R1>RkZ_~a;l364sh`=JJNGX&|6^$Dv`KmOb%mv=YFjGD+KoG!
z!4Lu#J<i%sR>87k;hx8?=b4T>O0VtU&p&_9YCU<VC;7=_7w6`tTU|#O(HD8$CovvS
zvBp^+0$+0*%kZ1B*<aZD?N1COSzG>uv**u9!MK()&ZLx-mgML(osr1YDh(<%zi(m8
zTS^<GOl9;~CYrC^QcbrQ6h#13KBFI4&Aw05F<Mp^<C|M~yk492KJzANsYxg6+xE0>
zd!0A>h(K@x)zJ3${w*~!QD4>^`P~72d6E*--nUG2x{0Yu(Ifuu^v{~`r5}7|n@Ah&
zFVM}Cx$=K@BisL;9@!ZGCraf%=*DK%|L(?D)w{B6L6>*ZL{6b8QUU>cgvFv2eY2tv
z%s{v_nZ>_G$Fx;o!bu<>gi{W;9HzHCuC5?cS%6c@ms{r9S`k2n_8C)?QICm;Mxsc@
zaA8P?Q)v=dk_M(1TUlCydBy{IVC-cT7{uWh+LuD$>o_T?{8)lDgOZG9nIC%=Rl503
z?oae~oDpj)w~%_!a*!C9P*D?r>WUDIROVUa<}i~~Sry|HKkD?3-F7Fms+W409?+|i
zhkCfYpZgzNar{pxhu~KUxw1GaU}><=(K*H%z=9)nasXKahDv1)6;KoxiuAq@M6TZ*
z50EhK9rKd|vV96!i;HyUlj-OpJC^Fy1nUlauwfK~x*455>NH<M?fBCH?M>EO`8?Q3
z%3A?k&^d=Uyk29!F;v#cjM(nqeAZ|*WUl(Vy3|a$ndLt$;GD5w$(@69o~wLxOKIZS
zCH&V@q$=>^p4pPT9N|vM(6UO%T`DqJV=2d%4^u-gFZgJovI!$WzNq1T-k4rP(ZORO
z4PTj}?|f?s@nriMuTK6F)UOW5xTkBv%Y&|iBsi0<jNQk{Qwlp#$)uffEjSWjypc$_
z037P_ulH->Y>n5m-a{vSFI{xG+gwYI)R`ZzTPvQe!yO7aPnHx3*51{?DT5c6G1gnE
zfRfvP1jd^<qsL3!Sg7h%4ZkB0z@ihC8K}4iX9T?npcrLwCT^o-(zy~-!^=%sM0D79
zK{n4*%?I)yvH<pTCP+w2Oj$WfaZ~lXV!H6PTw_y81a-=fBh{y62YmO(KK6w`Y*)b;
zg5#<e1z4;|zs{Ihnx3JAUUW*bqX!SzQ6?04<EB{vRxbREb@8#ZndOf1Ss1Yo;g$ur
z+To+8npTk5a4}>Y!T)-hxmMYZwa^<8fxk2x7tCJ|7R#W#P<O^>8iPO8K%EpFQKsQK
z|9R1!JqgcR;d?=>(EnGLu>T`Xo{i~0?~-a2P1{U*6raDC#gHCk$)%TfLCJwRzSd?G
zQnmTlUqhpTfQA@K>Z0iTEk4a?D@$;ZO#{z2^>KaFe*qp+a&d?JOj`<pnHU%Y1eroj
zQi!k|3Q$hIcu&6Lng|2|V~o=OJDv-3OfC4EqA*4%ov#g=S4%2M-@dZ|s4M{{`&MOH
zTOR%-2ovj<B&XldoXasIDTWzlY7|QzR?zL=`iBhqt{+Vk0ygp9_uV8C9mRWpReoz)
z50OBsKD7M78&PFyiIQOTD^Tj=%?<o}&GfSc*=0-mxFHnNU1dzSbOXNg#`6qp$^CNG
ze&}Mnz5mPaY%n1;k?}~dn;-hoW~OzQx|$dSa8QmI^PGPbeqoU>)9$`5Ceg_gZ`fSM
z!TmkOcWa}m%)}G#&tEs$_X!+RQvO*uFQEZrzh{T^pmT%gPgRYOagq~yl5mlQLi){4
z2FyQ2J+{)@S$>-8Sr(gKtd&;V_LZFc2&s<N09>LK%)%OCHZ&DCrwC;ZIVEt>NIMH~
zJuez4IOO8BLu#=<7$t1!ZzgvQBOH!gbtWy#b3EF>;y5|0yntSo?yxI56l=87$fTRg
zJ38?tfWa8aZnv$Q!-h?g73&HkXREA%*IY{ZdTqwP&!n9jbfjRE-01do;92Qc4RKw`
zid_|-d~R9{^Ugxditu2UQ+bVL-@@IkJ!|xIEHMzOSd277+anH)pN_0+K;@#(b&Of8
zU3f;j)yd3bh)z7zYE>@V_M47Kf0v3lIsUk+E5+k9k$<;CEuQJ441-HRAo}g6vsJ9r
z(6UKTx3zm-XN2MdQqE+C#`ahYC!dJK>Su{gl5d?{kLz0urC_x=jv6hYaT$^}V?{`^
z+;*MXQ-{jr(rnB^JffNeQ7ma-3$<=@+-!U|&=~i*n<k~(K<`Wd%jb%d;9xg<)J+-y
zO_YST7LXU#wco4dJju1v7gUYwEPk;B?(j_pC8_<hU;d#cVP*Os=KMcEm1<>)*bRD=
zo<}N&Aj$hR0{bh(Y9su?ITL-svx0L7c~T*wV32rh_NP%cs_0DVG9HqL&y?4kYg2M;
z5OQiTMN_+1{8&&Z@|3#0E=WUaaz-#Y6_8V9H(Wx{Se>j`S9pj8C>VJN1cGq!kPM!N
z<id3#d4)X$0(vA=m{3k8bgY)ShvR3qMa=StIB#iC@H%k<c~L`ufUxom%E${0IYQ`y
zA|OHgnm;&{b|#o$Sm780`$OGXGd3vh`ga77`L1;D9o+^kCM6F6;(#*B<Z;JZhpI92
zj#AQOy5(uytA)M`;xv?S<gIV1U8J9PZZLt?(@r)VCnIxKC8j`ej#osNikZ{AsQ7el
z&U7|PL08;L^IiF}%Ml1e{P78NfI;FSWJq6`uJtmYKR-X-KTOgy!lHvnc@Bz1QN``*
zt*IRgk!5O>Mm@)HSp(!Y%q~_YHL5EbRK*+_=*$VEosH;!xI5}wb3Fup$&Nz-LP$n`
z?uCNe_C8YS)eAfNuxRhRY7h_m5vx9&&~p41x`dcb%}sTzcD_x0W}juNdKYcon3ciP
zUChpbK^B%xh@Cnws;#0Mj;8V$vj*su_j;w_B0_?ULbR|mIq4dFHy?b88JObqyTz&8
z{;k8mnxir9g6A$DH{8(HYa`&yul2c~LpTc`nA6DD=rn0H9fkEb4yj5X-?jW?{*Rtr
zB*<>@MWa)6lkV5zI`qLdeDBV2Mkgg8b-8qR)PeAE-E;RE&+S)@mZi(GQ)T~EbZObc
zP-L-hLa#9NwsTkHb|Kaf;%itd&&hdOvY(WhytmTAUmsVv*c~5DK<0DHnLV~<Rd$1a
z74qL033_~60}Ci_ZYVk_<G&&$Gki8i#{cHl{)0kpR#~^n{QFq>J06*d888N_6a9ir
z>?i5hgFSp9Xhvu@4?2=&t^SycrNgjO8zGo^FoJo_alFayggMG#AB*-R+URUSSYn4Y
zYruDWLT5|jkVpbVA&H@FALSJU@CT$Duj6y52BZ_O<0Bn0GC=U9FMVB2VXx|lMDk0>
zDmdLqNWgs76G&^wCnyCXx0E!QfsOzI0X2r_Uq0Hp-1SQ(xVp?dixZyQiwlxyieVkM
z4jI00076kda4P~*7HN;fCEaxHD%-$iu(9bu(TlwH=-vjI-`(cr#6Gw#shb;o+Fc}l
zgly@2FkYr65;4aOY3RHEg)p48Fp9kSggra0Y|gJT{KZufib$t;^`?q_4eexVfpB#)
z8|(3V9-dra@i&tfNX;Hvz6!R2q7%#Pwm9hX@`lFIlFQBA@c@=2bjqo?hSg~j-^KXP
zs|g}$`g|8vh6eo-zM<6PX*ts)>~8DmO*{W(UR-jv=irlQ%YF}ZLkK(JnwuL-?|J)@
zi|N}koT5Qzlef}`4W4bn-6<vKmdi5KZ<hYowJ(gV4xP!o5M(_;?gz|@^Qf-Vw@QF*
zr)WYEguh0LkrX6JiV+N!GDH>RHB;Kk00|*5DUyge%RCa6A^J+UsGb2_{x?^)FE}?1
z18_WnE6kSr_r_Vx@;>4Vfxb{G_DI)*%1pm`r6awr)=-cKDb_81k^vX)ij7V(z-WF=
zU?L8B=}G<N`h7~rhd-noUMm6~(l#S2YX(~|j&~^ME!r=A2fi6KssgTi+Tk%PA-Q;O
zp168Zd6jT0(RzHFal8P*Jv+N7SR*+sG%{y+sm|L+b+@L_kI#^g8<^f=KxD#A>t$ot
zTbE(5vEXzIjw=={+Jw~ne*|kV&7K*)9rRgflcrfw=I&RfI*g0m1i;jeUW_#GG&i;H
zNDUBy5Mk4YKu?(7&q;ND3^qRD(L>o^|Jft|$VUIWC;g8P!=u$S9W&YgYdF&TP}l-f
z1`3~?xpLZC8TVjIl@T=U$TZQ2GS1Y7(dm;|T*vDb!xr$1D>?)ZDst@g0#`J&mOFgg
z({{T%3F8vw2U>NBwlik0A!z$ejxO59Tq8yLHSYZOnMouXIKzS_H{gYrw$!@whxHp{
zhL$virfeJ|U2UNFWc=ysZSl!1WHZ$_Cm1<4b8$9eA8?tHd6rEKkjal3VL%0`N;JsW
zllTs}8jgtdd>%^`tuQHLKZwEP<j)xOd%TfllKPdRdchA56$K;^+|iO=^KcU4U&(SC
zC}OF+OKF+^I#jz7RybC&f$o8a?gU9qf|cF%_BtZ%ib&J8s7Ux}CAajW=Cp@>;?XDL
za)Qcm(7Q>GH6{rtuF<?oO79s2{pejk$Z9!}T`lh`@IZf$#y9+;-N((**Eb%K36o%Y
z<+)cX$kOE^sifkK#c#cM(q!rarEZAEDkvvvY%fhw<)?=`ieW2B$k4ZE;x0Zuo^0O2
zbZhXT*P%8GL|(}`;$=x6F0XK?k<79%gE_Z@^%H`damo-Yr!xA%eEGa!Ns!$ew}nTD
zMqm{6QQ@0C#I%MabyQ^uol%XmrM;4<1K64e9Qx`9pj`{+eVryP(g;dPhNZ17{W9AG
zk`wsV_d&~ALzBT+2z?aGLT%lpbe+f>NI@6Il;7#YEZ%fXS~7oy_SUkfGzuJLIE}QV
zg9StWNR4T-SA8Q2nHN>bt(;VwN2Mk~3il+FR9Z<Kyv`1I7oF=4L1q-aau%ivxz<Ln
zb>-w+nzxH(eX#G^o0=(mO4L#DG7GX}HP$fJnZ9Y+!H<RQ7V5OjvAXy6sOo$rE4r)Q
zY!;xpwYvH({7eT!&E7{2m^xFlF|sFf?<HhBqI6@G=sZxelRVI~yxF{MKA?;>x`!Xo
zs85!zSgdPhZ|kSAh7GkxRSbL0&fjRMIQ>Lfv?hy*4Ba6YKKgKEgG4m@I+?A_5M$hc
zG|L98_W0P%S$}f$n2r|C)E6IdznO$ph&n6}zMMR`O$r=dfgkb4<Qe+9ji!9@X2b?L
zZA{sWL-R`CIEU=7(5|!Wn=-v?BgO>1$`jgb=H>7hP(ZPX)r;lf*Ieb`yCBsjoDhQ&
z{W4s+-XYEr=)X&<Ar2U=Zlz0=st4evG}B8~DCt_9l{b6b)kQFI&udz9=#ia9y03ex
z?34pwk1ycD5xn$T1-JE<m#cdctz0<{=(1>;_bqaVD^^q1Kp~59!c2M9Zq8-aT8fMd
z_^s<}$5CYHEpU&m8^c>Vl`dFS{4cPsD@78HH;e2&Qo{w)_+P7iU&C!wq=~FLG)Yz)
z`YM%m%*la*xx|OK?HVl$d+d8=lPZDEUIFYi)t8U%B2qvK22osW3kbt0*E<KMH>0QS
z9F(QJkh?%#Ga9{eFG$X=oRdbUL%zcbXo;y`S%#XT=F$BPZyZor?`TeYICS?(huV_b
zu13*V36%hJ0mK!Gj|mLePeKkV9w8GUS#1)ThnEF)1z3G2)i62wL`eG}#PSv<z31Vg
z?0dFJ%%Pu9-c&Mu-i-VkoKyO|R75Y)+ET(9d-PRVFcVMgL(lc<ilRCEW;Gwg_g&)N
zyO;pUH_*Pp-856&Vcg@kh?lpCm{jNPA;b1!a05;;V`M#$P^2jaCE=Uj8Y;WDWSF$@
zT~?FW8<q{%Yj7vBOvo+U33EJu&E{e|Qjo*(l$93WjQ`@W_Ual4eXJUuTg(%B!muH%
zc682cea!dVedC6@&gh^gx~NB)s83qs*qbb&bpkiZWDE)MYm6=|#r!-y;VNa@<rsX2
z!;FS?2zApPl`xM_ns<q>dX?kTs1L5PDW1}pya~i(A1dw0yac1hZ!VN7e%SNK<4tjk
z4a#q{#``KCp#f!ht%vR#qz80{#5bBcW%=NZW`()a<)?HCx2A#Ba}fSz2v`qJTCV<x
z$#FL?uwVWAnA<!V$W`>6KSpnMZ2wM~_~S)yctcuyt>%Rrf9691)ru4XN8=+)%=#Gb
znc>gkJ@9haEmt4_FO4uNG<u=Fr!WNj2=W!8E1Ut1c%S;?>EL63b1*c^2p(9gE2Q6)
z^9LmzA&+sNscK06I05A=DiqcSu)L}nN6@caMBK3ct$FL|@C!V{U53WHJ>L#@{fmEK
z6+PSkeT1vzZfA^7Cu?A#=wuB=Cxg#GPyb)^vxB1(J_iTG|C#p~85mjq(`D+m#<nd{
zE28hJu0S4Ld+6bV;TtLj;7tN2WL!c2I}|YsTrhR5w;*wKy+-NG?>{d;Ayk)5D;g;z
z!z!P4r|J&kaK21~=>)~f5Q31^f+(zax=8rV5gKZeQfdN2Jo1ncLqW*>M=AIP_EK?S
zJM{4nO2E|#Xd2*jQ6pEPnm0pTBlj?nf#eRF1LU@9enC5NWD;lMl#ifMY#9_l$9)CK
zh#^Useu9~b5u429Xu4EY7*;2d1HzL)c}!rF(lTj93=HD>&Bys05SB1*^<tJ?hAqTZ
z%D7Z!WL1vou-!Rnu<Y++R%v5khv@@;!d?1fk`tshk$FI`zW~foV*ylxY)W9r71RcZ
z1$G`G_$B=r7SI}rr>yb-By+&ki0YKla2hy=Gv-r*m<ZtSfHqHut59Q{Z~=K`qmU`~
zfqsD1@PP(Fs<f84lcMe-o-4zHR)-u6sC2{zl~8h+Z((qD5M%Amo*82A2E>HafS5jy
z)S=I$C>k*lHOT2b^_T>)q9&WVbwD!J3`|AvGd&X_c<KDp1@pGB1Vkc9xJ@MDH}#0+
zP1^vao7AHP!q2R10%=0Fz>G)I)a3Ru)pWrGNp=9LKtZr_Uv!8K-GlvPGHC#0;>Zti
zo5ZW50srO<N%?Vu8fb%nQ$AGK#RR3N8z8HwgZg(17=Mn(6e44x-U_4-9X3L*QM(x6
z<yB#${cevVJH+fV;aE3Sf|i_u%RrQ_{(XmZ#foVV#|<mxZRfy9-h+IKakh7iCwxCy
z`zJ^O8obK1HF3POpKX_W;0P3x2K7$O7-hJ1xu{~+CZtedE#>=Ee=+G<D(&U5)^+;T
zbh?frcm_7`pecjp;7Ayy6a*8!P-cDhZTA2gO~itruYA}9XzH+MwqA4Fm!V~@{{T8i
z|L~hVkgnlXmpzc2+H>vw>6=_mX1CTl52v7*7y)EsAO5QSt-VIn)R<_de*RWQ<i5Vk
zatNt4)^{r@s?V7NfU&V8i^WkzX2d|wi~MxinuqfA`aON~M@1(l^tO`Aeg<|Z2j`27
znz@Q0<2jq;+>G=h`eIX&N?Y2=X-T9tP!atSb!kOuNq*NwavjHgH<0_<zw47Jes>QZ
zD^54q4tYr%wmP7jsY~Ly59`&B&Dx&UIgLAjNg_s%?f?+Saogw*j<<7HzU46~iJCii
z(R3+prmW@$H;~fgn+<Cv{ib_nQav{ZnI-}aHay6<5ZLhgwMb&G@2^=OER@^74jWaE
z4i}2|LidLE*X=WS)cX*gjYY*GZw(U@S%xHZ$gKL!@iPI}+A9<3<UO^#DnO!l%U>@B
zG4+QQq|FHu7i=)A`bG8$OjO8)%;I2C3JZk!O>dS>yNVe-$B0$II<&b_P4wVLc?(3x
z{7WfzdPQvumLmrBxn0=@+v5^By>6+hbTbjYUT^V#yr0|6ZBBI2hn=2Wc3?~!DrzGq
zEtu)1daLLz9x8%*#Ev)5?OeFBP^5XXrfVxaI8$>o&yVOH@1IN*cDl7&CWpaxz1G!U
zhMlla_Y283HJEROKa_1tTGrG$n4+@3022YWa|Ur*k0g)PZQ6l&RC&cxLAcq^ILQu(
z#%^Ow&-xdy{Y!m0U!2nAO!}qUO}r2@cM(ooUqVKiwzHq}oMUkpZan5Jpn06qxKm#e
zpo`IG0a*!q_m9@&@LW+vlP=jiOJi0PfoT0)SK~KD#kN1@-|*W5cpBdB+(cJ-rc@r9
zvX}LgVt*arT8P&o1L5v9X-<D*bcBuvH0L{4MXw`S9<J80zPwTP!eZE<dv&_c%&2YR
zcCy~PtJNxDd%e1keF%c4d+}4sDzt}n1QY@8iGpKF`TE}vDl4<7)d-unXH2)2)QKc-
z7(ckrqf8@5!TtRw=Nx2)OJdoxJU0!#A^z>oz>O}hrY%lhGhO>Y)sfC*f6h)E!TUL%
z^>AMLIBoW6a7fwxUK!)LS~RD{A_tsBo~W!$i{)hR#XVe$<2YQv&(z*s+6Vlx{XDS!
z25#n~HfWAZ8&dFn$KFiBip0j4=YJQ_y_y0DB{+Vos_5QSb^zYcr<(fL8b#lUJ_wOR
zgYk41;rFSD`ZP33g86_jX*?*><q>A*K)pAiJ(y<-MI9uj0$)Z&#@#kQQR1iqb{;pb
z`9~0~A|GzIFzf_36U}rhMD0Ey)Bz4k%YOr$edM5Q5A)1<S1&kvI<W~K32mKzw8fOY
z&I&oPNWBd+LH~)$Ckg1V&G#IzhQOBPqs4DcyfRhc7$-E#S~O}Sgi`pt>pBgtjXdyR
zPmi)Kb>OZX6>?Wb|I^-4o?)W3GTfBqje98TqpQYEnM72%bM?h&fRkz~gCUF>_*aO1
z0+i}_|87gPdU*199gMBK7_ck8qm$BWYG4!@Dd7ybep<I`pc5NW$$9zR(1yA4*pRX~
zj`wL*^w-edh*8EZ0j@bp07N|H$F;>j3nGB7HUAN{if@e9hm!2@;`;N$Mdj-o=;^DK
zXHEqYvrj=PG6>&6rlB(J`EIuX^%)HzX5%q0(#OY4)8G5L{kj8JrgTt))pa@$oaIM*
zaWPyp)LgOP-Y-5Xb?uEn_F@Kk!$L}{eR1bK43<Y(cCHh=`c6aEBpdn%I4Lmy^qQ)T
z$M3#Y2@ZEa(qI-?NxU-<arsYp@=#6`r%k5(D6ZbwbM+B^UlDF(xVfv|0;1Z~jowoi
ztj)J$ep}Ay$Z&0tlLk`gcYE9eBflm}RZ~ccJH7~+b<A1pE~DEyoX@(zPJvVCN+I!a
zquNAQ7D~pV!5|;ti7p%{F{?P3;7>E_9^@=yQS&jZZ08?Pw$_~Le~d4T^o;+Tj$+{W
zyHWrDzyJ5{JXvi!c9Rv!cUE`5b)qj~Kb5>39yB2jtH^9o=pYohW;lCgWkb+tJo(59
zE%bBOb_{c(;Hr6%O#aBu<PB>E{oSB*FY>i^r$(o{2c28@h?;GIS|t|R7%iMN>e8rw
zgZc**vL8irBz%8<_K%jg2XgP9r<2y*9co3KYTho-$4kZ($k?6lpL-^1pSQM;t}n;u
z$xF}b5^m=KHv@j+1}TT%PZ5b|wYQE8`7u21)QTy`U~JVX9I-j@%8er7*F-+6*FST(
znXo_B6VBRB!*b2AGAO0VeC(;x$0;+)N7S=Sb0Ks=fqqmefc+Nm1$=Hgsl+}s9Tje6
zlf6WL;7UVqu7Bmg8@pB5mshrQ66{1<eOM@JJfjELyDzAo29Pj{c~9Dg+E6r<U3r#<
zB5`lveFBPd$Kpz6QRq2u=3Th>nAz`yw;-1$uAhf42-u?E#mNkS)r8C}nl)$IS6^7T
z^%R+9Fi{{WIc>9k&!H+a&FQBSf>wvtXuyAStQkPr;_XoZd6`e=jL|M%ca#ZPKDQaT
zYU1>Nq(4ZKQYRbaN9%K#-Jrm>i~|5<m1nglw`NA|sY(A?bMcg|rwBQS2^(NR#YxGN
zRK1$azw#g~2rM_#WF^6F7nUBDniOs%o|K#HjS*o203gZ!0|@t1Z0fcNjD9Fp@=l0`
znb_WSqbstCI?LYl3U#aGyB^q(=m2(PPSIB>*NO^MxTSq-A4Z6>y~_R2wS83odIq2%
zteX1SfEm^&#7jwTik^G+vj-{WEMkI&d1jfk>?*84EM<>s_6cl_V9@|L#)voY;~9RK
zv_}Zuuby#~a|AOJht>AWG(Aa=bUbNJ^-Tu!&n*`Vv=!U^yf9<_L+}X)Q@H>;fFPh=
zNyI>yl(Ethi_E&4LGMgPogXa#GLM!Sw!BxMh^b^6MVR)pHG9FhASsWVl|U0RaEMvL
z2;-?#TKWWQi55@k0oYVdaOX|UxO^OxG;88wGJ-}8G?4As;yNo7H(i9Jo@~J&4=c*b
zaVyq>!*Em_`jyz<V+xsmwPOm!kcO?Zv;~5g8<N6O?dF|*JMAalj(2)7AI~cAh`=Gh
z@4D^Qz*t|5C=wm?w}MAmwQKyvWS;wbOvJMc<ZYIrC<j0W_JrTmA&=xg<2ZaOa`I{w
zeZ^UjJn5f3QBEakV+yyDolu00T<0FIRsuaJmWCV1!keC9Ki(K-nTJ1s1Aky21LpgL
zoC&?m8Iv{>4v31h=1NQ2R9mOp82JJB_lwxO%3GHG26;wgQCVoFpM>V;@5+C<upcxI
z1^Gm#JJWh;GpDG>)$=>U=2R>wKGc2?faF_WoaU`@CTTA#TnD=2<6DeOf&4u!jckvJ
zLl67vY*}w3>`w&|Y2`7{Xb<O>RWBeH>9WS59a*&)(aEH+Br3RI2pJ}&z045@SnDTY
z_!6Pk2hdSTwcl<Dg?0+a`?*w$u16vrCn^D$<knZnm^8*$$=6F^7>!jvDSL0SJ2<K&
z({x(KE$fk~P^mDgTx8)+AK#6&6y<0Kz99gSD+Z(-+{E>74q;BfOqY)QYXNZ%032(D
zM3^Zz&Ox|&Nc{#~&Bq}_dSM-{Qd2o<`sKQhxDq)Hk+r@0@qMgCyL2s=wf+m$E&#@r
z4J{`T3k-a;?dS?TCp68X$uB0)ZYzbI_?1A@;sGBAR+a{iR!Nz&u9x(6zVLm4xMMKA
zhZ{xBsbe?P^y}wz6fX?G@A*S^N!19TvIcFgidl0Nv;Xai)q-Ie?8j`;?9JGM(4-Se
z0&IEU5VBBW7t~ianMzFWiJu7${!M{&9!b``dhOkJH`k7bBQOX$IUYJ1?x4cCqtH^b
z`t`V~m+Nxti{q+ehKxsNd5Zkf0Ar%~*<lM}YP(&;$@l6CXRQ@tyUW(eOzRA)NKUGl
z@=Dtd8r6<<w@K*iSf|!!+e2BKpij~Ti5m>RGygR!%yr(E;}UOV<-GXIEg$_|jIAYT
zwZIW(X5d-{R`-w8UVVSp6&twzHOfp^lDV86n%T_U`-V#9<V7J|BFf`uS}ty0%Nt`S
z`HPOyv^`h{%kitTvY@F{(K*F}c6me|7fL!@Yo3fMiNG*nGF0n!nHp&5jl@(FR;-6|
z#vH}roI9wdSftvfgbYgrc49B_r1(CUNLqb!1hA3^S-sKqh^_aKt$(iAQ^ng><dh)5
zS~?w(w*x1MIa-%)_gzf!ei&gFrmeLY?kl)BF)MNvh5SO|s4MeI9wzu9Mucpx=BfdO
zh%Ch<H|N1(cmbIDI~ycP;jEB=BFj;{VSVwtfs=(Ap;z+d-3A%>^%#tYUqhojEZMz3
zAUd`<a(Rv&81<8dyI?IP!rxpELJX^L(fre_gshYMHGEz=0z6)W4H=yogNf4EThlHZ
zG%tx66dvX6*h~ZKM5?9VtPAF;Jhhx53sR*H6PLR<R^61E7&9>ND0Xmq!}<PKW1N29
zg0)(jc}_saU`O#_p=_L_>HeJpCz)dqxFuFZMtIdWc}XVJ)4<ZU&n+*AarV1lSvjp)
zL*rO0&Oy}hhz$;_hziM)4lu2f#fDOH6LX8CQi4);wMH%rIy_vi=NIemdmUNZUAd~Y
zKk%u~fKLA$1O6rX=3x4dt;DMTjsb_v2)@0&1+nFAzhbZ1UnIDx&)E4P2x`du_ktE!
zw_%IMDJ5QS-1cluzmiAHH!S>1*J5~>a^9y?&kuYA&~0_R+I{ZdR|Dvn=M%<<J$m;X
z;-MtU#L+Le4^}~lwR?1ZU*A{xpUtq$B>5?b4P3pSr?+D-SNUjEsf!^2)N2$Z5^ryP
z+#YXkrY&2)UQ+Q^CZ&j7`LF`$%Fy)(e2vL}Q=(I)AQl%F{>i_>!%pVpL&?TMd#<Uv
z-EOPuPU=I43YV}6GZ%ToKfI-}X~`rEff0w|v?7eRBEvk~tOZ~sy!)V%3o^Ge-CcHG
z#C@pvs$AnJcQIaVJw91WmENAd0~72Y6GPWSp-Unzmy3Mf&&r5Hws1LT%TTDr%5on4
zMq<<|vVt5REy!Gibn}UlC=f<t`tIiB#eHETHbx3F;^tj%QcX9`N-JhE>C~KJmkvjx
z<_%v3b3GRN8Hdg2Xegv!N3k)^R7I_T8>b_m&xn1>f*feGYeL{Kqe^*#A8y7HmuW+|
zpBHMRs=$07y<#T9oIP;NtXThfU#RDC5$Oa;u5t;k*SyXfOc3kbuzHYIPyQxrZR1<f
zwDk7HNVe$(13L6v>n~YPo>8_t^ck8q0}<WeceFSmjtt5p^RfBZ&)?#O>2aOD_p|J3
z!O*Jp$YN?F#?QHQqac&4U%T(AZdK77&M*5UAU+Zp^6I4Nuu;2p?0oMzJ0;f5XKc!k
zeR(;$_gn8uNFvSkwg;^3-+@r<M#wrazjoSqcDE(D1fspfXmIeO&KUjlP<K6^sns~#
ze+3g%53va^5SyYCaH3``WuJ8Onlo5>v;{vek<0!sUd#M~{LAxBPNrdVZr^^6Rn>q@
zzD*P)z_ra)^Z_-I5Qtr7l<|OhEKG=u2#$6PBPpgqA+4?x|51WBL_$noDNN?r|B!~O
zG@|HUK2$Cd#>%K`B2zI+7{&ld0A2<lR1VuM=-_X9@VS~kniB}N14k-rPR4Dtre4Y{
zKA}Bmg+CcRll-9~Z;H-72n9m^^PAl6iF51dlSL4#&39U>n(C9t<~BX{2?J>(CZGzp
z@=TB-$(jJfh{@Jfsn1|zk{pRQ>9A1sW2_;%=pRiz71~R_%})B#a{2NBMG3}ho^NX(
zw<<5MWBabbEJHo_rv2B7cxL^GQKvsSv(<sv_`%hb1wuWn`~uif(l3tmBP3X6_m{vl
zKaS>!yKpsUr{+`Vk{^8K5sZW-A5itE>me|_XoC_L08F3}2x_hY;1fPU35`exy?#SF
zDtuVPr|nASK%7bAQ`X-^yR4ZDYbOSREWfa!%3!Rcb|Q}{6EY>r%^}pxIn^}Wen<8%
zoKqAZR48s=%ViQcUIh`J%tGU)ar7^!0S;bGWW#}h$i)I&AljKU{}9}gi8g}%$y%c5
z5c0Ti1ry2rO@e}$M>;a-U_sqFZj~No{c4D%T;!C+uH~pQ!>T@#2RKX-?^mJBuXb|z
z&ZVf=Cp8@zNd(6}_*^he#iT#JzxHhNPT)(qRaOrDx~MF@$1{OoQBm=}H-I8z-B@JR
z7?izGX-VIg7kzSCKXr!MI6e7RvBoS%ty1Wy>WG?(IluNL!+~RCzYsu9nKKOe$==Al
zM7_CwnMNxQZ}a2OcZGc}VX;1=cr+E^2J4wVltfh`@>7GMvC*jE;zY%amdcyjbIm|<
znhXx=0)dA!BE`3tX~v?oo!0v;dG5Y0!Jj@rti|S?r}V*W9kaZzrn}WSJXO+N3D(yD
zf~-bOr_koyA7wXjsF;tV(xcRhl$~lB1qpNMwRdE^%s9Te@-nZngMvj)5+NP7{@B~&
zx_@_o6GADYaD^l+HL&GRtx7nZu(~3i&a$xJ*2EArBb9YMY5kt|i7VU5IqAhfyo;_+
zu{N_?g8zEdL6wG1#@arwMSe=!$A~u`99T^8eqv@hu-K2z%Cy8-zzBgwd!&!iTYoIm
zUAtxKxTq>bAPgK}jd_k}g&ZT|H_L>69diwowe=ieU#QWS&e+u{b--&T10otua6@!k
zv&Kr>E?^*fOi`GWt;t&~aOpA<yba$KtGW9YEadnhEoRmiK1(O726hG3;Otg0=s^eq
z`h^_fBz!Rz3~)#lWQwg@Dc~SXYW+N}+D9MGNcP=<0zd98pGk}c1JVQLcSK=3>80r2
za3Zb<F7O>(VX)g-EH~{}YXUH+s=Bk<LJQ@c*WE|X2{ht$jOWQZeD=OnW*!PR<`3G)
zjk?LJM7Dp){MH}IwGnsgqBDu><joY%4U%5pP?-9Q9--B3`m^PW9l0N%@+|42?3Dc>
ziVWsBh4S|b8Zq~v!$wyeWX1;=%<EGwZsa9wnER6oG8$_QN2n*Jw^tRHgh#RIkPnXL
zSy}@+o$J91JnbH8r2tdO2t#kz$`mR$akOGFQMTT1fc@{8iAI`%&%Hotvo+~7%F}X)
zYOswBK1SB;XBKMTxx0aJ|F&nY*L*I_9h7WlfVd#^v5i-%Q%^BDNcgy~Cs(V(d1H1x
zHKjvvMBttcAv3%=RzHhPrBNZCKW=Xx-v@YsXk<hG!gz*%NT?V%SpQ>;PsSLxIb?mx
zshz8a%Ba&4ra1$WD$Oq7aguP%eBsxOqKQIVO-PjJ?YX^=ZzywPB=<si7w_?Ak#&KT
z_31dMynTID)$OhnCWImlo=Nx&6+V_wQIS#=anL2ZRU}w{-hJ(zl1+GR4KQW={${_5
zDe3G<&M4*cY1&*|g5lBmzR)g7h)`~yLd@6G8u*<PCTyMp+$?oUO$X=;{&xF14IwU>
zZ1Z|TzfNQgH)<(gGH9r%>1%by<Wslk5^I?b2M@9dFFT)oJ%^LAveNYF$jDROk>5hY
z11ns8$gRZ_E!``G8E2UT-XcJvKBAN#c0BJCibmUTsX09P)H=B2G~GKg?n~9ki1Rv9
zf2$Vy-PHQi=j5ZDIX)L*lktm!%kSJ$gV~#jr6gKj_Jw|D&ciVyY!x(DlCg&B0D0`i
zDA&WTs?Q=U9dvBIubZg4wEKRpF-qLOu|jKIj$Gx)*l3WNPZN9vWU<_O+5~`7W*p-7
z{IWTr6kcc(PFClK`_13?+Um*%{1>=mRu@>#3#|O2-lG^9sFoY^=!liZryz$`9ZWa2
zd(CwVi&!amschW354T%mH1rue;AQr8DFFUwpTQ#BCcI8<^r|Ks0#vOlS9(b_a2SwG
zoup;>eSe~6d76Sl44XZ{0lVz237I-01CBHwFU+V7uS1P91Drc5mi%!*vH~a(rz47G
z=v1H(nx%#v%2pfx=+i5pQ2fT&#PHRp+3g{WgN%#N<`|o5iz4O`@ePhAT93_C4c<uk
zdQ6lsSosWLpoPddabK2&ynk)QS#c&+e4}KTxDfYuq@z1F*+$-8okFs!@{MBzy!y={
z05!`TdXr-DkmQIdYyosIRyHV`M-Q4{ahhR}xNn-R^GSM{^;J|2ccCTsy!q`IOzAf7
z5mWmd=`8tRp&B_X$Yk{bdHP8h&{B5KwS`nR0FqOf$sllOC;k@P-g`E~mi&ZrzwG_Y
z?JVH>>S`-y>bV(@^j+MDA?YY4ybBjgEZu>sO2YWUsIBQoqPh1|yx24#TL?e^-XbT*
zO@h03C?bRkT^emrqX#7EjB<fmj3;_vrJr|oj=$rWV0}fx&}t?cl~i&fn-T#)g`>tV
z^WTtTn6;!~joH5YS;*M<6#n#fp+P*#@__1k6ojAWdGJ$`<LTrFs<eTK^QENpJ@;Q=
zqd@3WPVqli#|-t%O`91#cqPy|qy)o9!L`Ui>eFeqb$%uSLXPvV-k<<Lij3QtE>Lym
zvh+fJa1YJv>H#d?+g1dDZ6Lv|-obvwvCM5vN80*cH8`*d(G$(CO(xidg5)YIs`&P1
zushsx;DwHHvxqD46-{)JoVt|#{&-)m1-CnIo*}$cu7G1`7AQtW;#jz#=M0?T61;*;
z@`eN^snnq54G`6XT<Dibj+iHAktqtrh6(~23{Gv-O}S4hW-o1J22Z^`Hx-1leS@mh
z`3U07Dj<SWZ@+*Q1dqv=q15@jfq6oCd?tH8rPh=%k=4I8EC<;Qdj&PdVGz#$kZ>^r
zQLkP!Tut@~I7)%I=sWZmaGPS5-D7*Ui<5CFkfr>t5C^!N0{<O%&)iB^=hy>rI^(O?
z`}j0Iiir6fX~*mq_6^?P5cRG5`xxkVx=NB5fYsRwAv?g^8VpSy9`N_Q$Y;9H4{#)T
zyVc=Iotzgcn&fAl+(2=vg$y0Q-8m@k`_q9QZ8yNw-03*BrzDx3#|%rvXggSLht}aW
z031VUxL|;6>YOEHE(E%uDXmt%#dE&r!!^U5iK|TN+Ja8tz((S8c7kmMv*?q#P1BTc
zCE1}ojp9^$tfteBd;tV|fl(QO3o!_>i*C;Yk_V31!SG9@F}~eWz_(ug=Gk?HYxDD|
z#;Wo-DpUmPY++Gl(H#e_VSu10^#;+Fd~EQk3^;^?P(E}SC4g!c4I;EHSDn};Viq{a
zhG()Cgg8O{^+WLiw>zbDGhQz<?~)9z)H$&CrMfL8u}5pz+EPBAYGeU$_ndW2#M5+}
zN2i{r@(;)qenS}{tF2B(7QQJu6mm&!P|7tNYrteXgeT~)_7Cs`H_vi%d3YUxLB>Ct
zcG?3&*zVCV!P=8)C6)*QzzBLdWrK=%*B&<#8-5T66NZ@A!m523wWOceP$IbXpOx=g
zEwhX&CdyA^vTS45My^WwDS+YYMME`|(Bs5+6Tr1TYjfOBAa)Cw;^c=YBg%r;tZNOP
zl{a73jhLZmQ@BzR>R>eB9OKU?O<c}co<k+Pxaz)Ly&tC*`|`m~zHe5Ym9M^!MPT%#
z&Y<ew<JuQU@%ow(>OImwT_1rz=|I)yDyYGW8YzuP7y58eeyl%RrJpuElx=F$b{R}M
z{m|nRTC`&9)r+mVJ`#~ML3}k0PfJ2;&9go+EBn4gYVv=I>=hJV3=cIqV0V5qA=F6h
zV;Y^Wd}D#;?WOL^DI$j48h`|C12=c$To2F(kC|1?`v^NW>c#hfk_BHq5?VDI2)gB}
z6&#&!Pg_U@7yu7g3n|$S_VQwJfwLrRWf#c7^ERikKXdmHeuSA}P!;E5o9mReZkBuZ
zJe}M?-;O?lP%o@C^0FlPhnhnUgIxq1?bjx0Rdqk`t=(85|7!wZ{D=PtBg21A_BUg!
z#~iiXd-n+1uHY|f;TFIh0hmN@PyZ6Pk{uBTu*GHHzQQH2C?_g6<%WB>tuctFk>B>z
z3+xg%s#mL`T#?ARLdxmt^zin5zOeppFi-rnE_e_&<v2{2xnUchEgw66=DY3nd{()$
z2qBFKNEQM6{}}tHCfgb=>Jm-cwsWU#+qP}nwr$(CZQGeUZQGf(zp8WMoQQfZ-XAcp
zR?Ic}9MMMW!Z@pKe!buOKQlYel9stp6SZobr|0l)yM0|<-_QcPUxzbod^>%=KXu|a
zkzQbEWzeIf{T@U}B^tECg(4Hes-&-8>?f{Vn?If<BywY7QNRgsMVNIWmanb2;6NEa
zn#mQU!k<w(_+nih0Tq>sOU%K|4VR<Dwh&G`Eko;Bn@jr^Ye<v;TLxRtpZ*;TMFI?H
zUuZ~g$C)Yl!eh5d=T?vN>iJ|+Joh0<WJ}BZW48Pn8OrCMGy^&MOu<+k{qL*GGXRR?
zl3oR0imAMrrZ8sBSxlV6!)(9}IBzpa)r!(m65Xw)c6Mi#dq+hjgb7Q~$L%Re^z<9c
zLj??U<kH?<DdNHUDXU0zX=GIYM)c%Xsxjl=_^)oH?8he`tvHtPU#f{)7<xvOd=ke$
zRgvU0cervbX2~dkL4YL1h}$LTKsA8iJE=c@*fO6o@%0-nK0LfP;G#JG^bFvn_5|TU
zd0>|)jaac^LP}b7t(h2E=PvC8GuzR7n#BN}k+q;Rf({Xi0N;ZeEYf7ir<6Roc`<uS
z8|?#m*Q2uD!|dsJ&Op2&t|h{;WiYws&Ok9Up)dGhojVIal;|~)O_kXa{Zgu^e=T}8
zNhuHQaU94Odvy3RTk&3+1x-W$dJF#acUXU6_{_cKm=dHv3V;KiP#VM4Lb@-W+k6T_
zH_RO5cFcMG9(x{Wsc=j=$#`>dO+wl6X>cr|(MaO~M3K3bvNY2Bf9{M-BSB!>4&7@d
zo)Y)eyFRFa!~G2(>a^O6W)f318BInTcEjanj<s(<-lg_F+`H?<6&LqMP&{Gi2Qg#o
zYQ-M)lzHsI1#4<*Si|*X9~lES!zGHnqzofT5;M@bp!TVsZMMg-z6mX5l3OIR8U+=4
zm4j9&A6;oE`fcEW#7{jD9u+gw<KOQnP51B4KG&d9HO{rV<E%sHJUqI3u6yBrz`;Uj
z7*}+q$f7lzxV%}nmHtH%iFwSwyoHel%?74DG}O*;q9ZL<wA-n>-oyd7>!$rMSuwtt
zI|%{9fVZ$QS|HXMgQSj6DF^_Mfe&sYt$j&&SF@=0(LP7RxCvD?vp{`2K`mCc+d)Pw
zG3LaX#%e;-o-X!*lC9@-Q_t@5t}U*^)xYD9thM1^&qcV%4Rufe@c)?pG*+RfiPT3G
z6Q`s@w}bfBxqGr1=Vgg!z2+?5F&k$|D3UQ#N(@eiLd&H7{ic0iU)<@c+kQUzx!6@B
z!$e(A5H}vxD2+1i=@V?9WZ6rTzvtgiBJGiBRpaxeN#gx=`h(JM0*YT_t_0HY$Wle_
z0>trWE`?6hgrc9N2Y7~-msSkqpP2MKGNKTM<;t%+k#A0*aZ)lh0;yY-k(6<BS1VlB
zfg$N<KgxnuB$0!vp!2?)Wrry^PHqC}@OL3^YQOLd_<k)1Ti^yK$$R2?X3dGFA6ZtL
zpXK5AP?NgjXX%;pl=dw4<TW>O)*wny=`dFj<Q@ryDK`HnxKjD(ey7({WEj&ZFB#mm
z$*|$I?>Fkvw^s&eX{;Tu5)Ebz>S15`YeGpN&K~_>Z_Wbr5=Bxx0cYS*1S}+E@_;xy
zluu_QlE?_NUA4Jx04qjzY_CpxN9<9=RaAbuiA#nu!#_|3a=?7FcyLt;_1+-9x;>3L
zEA5(8dwSxSZ(Q216lX2*SV};f3R2=>;RDO0J-&s+v_?J&(~=}eIuQ#F1o8uPNBqWx
zSe%Vs@4{48OuCbzNvbn|GgLyJe|YAWw>14?3Y8doK$ya4aQ&4k;pe--*4li@YByO)
zC!`*H&U1%XAh;FXnT)=uaxqJ5*~iS|Lx1&mFd-CVLrJL_kKvfeT#aud^^Sbz5bV=D
z-}%Pkgc&|vpRh8h`Kl7sAD>gY+16Zp%Z$+EPzC*9`DFN;7mXC*H$C|2RQ-Flx+`o`
zM;nh)0*52w5r@Q;MVUu4uxk<ve3%j@(^jm!z?mN2MJbN1xf4_?AIwdb&rJ>(I#Qo@
z=sekEga~Sa($DNAt%ZkRfzI6V*5AAB_hVL-AG9Din^TLPX?Np8G!joDlHP*rL+{KU
zk^8bjP|n_oNHQp9C(|gX#V7Qm53DIl4+?;L4_FaCX60t&5H?rR%~H=)r!XG0iqbB}
zVFn*=_1$Br;1CQ+o|UDA*0=`(l60?o2(bGvcN++AD@&(f|1KwxJ?SY=JHotqiqd!}
zta^{2(kh=q!NWqn0vf`jv-Cdj+Ls|p=o@1Xu&e@`Z%qa}GEOCgdUUu{JBz}dvTq^{
zAtO2eGN#@E6ImgsP;q2y$7+}4ZIG@u8U6gai%>0ed02i+$b!C+{@30>Qa^Z1PM0M}
zXBgZz)=zG-MSfdnG98E-j9|>Iqwb47dCz5i)cgx!6!Yo0uY%`G)cHv59i2uva9?+!
zlL05rRi<_c|87a&ttTQQ5+?^fC_dMG5~YFX144+7|D%|WyKsemZB)mOK^OZH0Xw!R
zwi{sEPlCxg&3;U<Y6Txnjx>EqeH;uPXJ?BBUq?k-d+W1639iN5#Or~&cwDOpWKR&3
zU8_=8PXqXt?N<~?54f1m_eRm#d(zZKE(Jt=K2JGoQCkAeCpA8)gusWkz~-(BgkRPy
z9HDtEAvdqfj8b^jjS6)ZPr{o#hcn8P^k+l9<k0Mxl`Nkghv@AD>6;lxpU2GHA%WiT
z!1>HLvk*E<e^c1m9sd+MUHhq<gRi7^Qy5=!dmkWBrJ>Y+#+m;(6JTWeCt3R+*@YJM
ziNt?sWzU@2-HD+va!|>t{2KuUccS%#qx*;P26()1(%rIPQ{>LlkEaO{a!b8cF@)G#
zFb$yb@xw#G(MfYc*6m-P$G4kraMCIn(n{^qtg*6~2{w&P=nT`WvBQ%zVGDy{*R*5C
z_qS&`zt8)yKHs+QwZpt!Kk*rFwM@v2V|#L&Nm!_!B=0#<$ch3oHcIV$$$~}8$94a!
zu$#`Eb4gNkua+3>?Uu58`&QuRE0?Ot;RXHOoy{VyhyIlvXy}G|@AaGL<@y}9w^jJR
ztIY1L&oQT6WPkgI(6l(NyIR6(clO=3tLT^%1hJQI!>XQb=F7?xfAp=yc7)tWTRx@M
z@Vqi{I633sl{`ozwQ1y>^XQBXbV{*FLnu-Nv8nDNXA2{KDVRDA!ZS)MgDsM1;L|R}
z#%NT)Vmp?=Z58zwrAht1uAy5bMJgwcZC&j6YX}RYH6%d~i+pB_o;Oja1PMwRrFu&%
z&ZN%3^aQ(>pg#xiElh+<S#%y($uflVf?uRu>0-zy+f3I)=Sma{qMQ>Gt!4wt>qIeo
zJTcexPMW6rA3Cd7Y4o5dWRg)Q=grn%#&InRQ}VH^L261*{2D_k`_nRha_FUne1H9V
zn-MD1CnlT6gISrR87w;bQIgP*-S-N2+~#7tt`k6k@&RB7B2g$nNob?V0}rf~t_o!w
zbROzUT({YT5Gg7xOYLe!N+Q|F6VDMv2#!^~i#NnFxw#FnW?`ruP9jF}>2mDA6N>@T
zY2qwKKfhE-Y0{x%gSD9^L6jOBQX)3Z3ga!4Ocjq3WqctmUA?+qTkv)F+e6;*%?)>L
z5Y-|39H!r5cIwG66vg89(tDUAk|R9xNk3qkUshVbmE3I1|Gsj=URu97nz}1Ng)=K|
zwaw+@;uWbVx{>kiZFGu%#GbT63Cy)F%RUPqwpp*>s-I5~XUf!8OrD+1S)Rys2i-O*
zm~zrq=FLdg4Lorot0XM2MgOo}=_iRN@W|VtC-UfYV4zXN%WF4a6onB+r%NT9goiV~
zgE@u6pY;!`Z9;}xYNTe~laCxWL=Nrkw)wgPuS|yHR1%tiob=zrEe7UUakU>}0b!Ku
zh9X}S+0nLqUIpf=JRF2EJ^#5i(H2k-mR3u6K~*W{UIa{sh~S=EqTJZ}`*>rLyNI&^
zC%I<T)b*<htO5$DB#)+ebOF!QZb}WuX<P*NKs$d^j2PKte-Hq;jW=sY>2IcCAF(*$
zHK+-iVhyB~gASO6Ve{J<IF=$l2DG=OZlgL2v`1SS+2*QDOUZ*4sUxR5WJ%AdUcyv$
zOmRmF_h~n$bNwL)i2QIgffq_EFNOmm3z;Le7Sxrub90ve$Lg-9C}gy70(Z*D91W7r
zA^7bVN&NUo%p8?vfQsGc1Xj#WP6wKy1t~ED&0>^LMnn{hx_@+~$7|qgp#2@{gc*sz
z8L@WM^%m|W6e0wel4>6FR(l-Czd;*hSpL$O0NDSE8$%BA?vs^OsXr)Pd`aZ>o^{Hs
zV>))hjWI``sxE&)GMPf`n|Q-6A+ZTtKVU~)hMqXB4n&<3WI13qmxbfNaEKx>Oy0oK
zoVj$d&{!Gdu|eF7No!ovIP|>e^uvlSZS&9!2+#BjS6$^WLBX040p=o;Q)~^t))(KZ
z$86raQwAvL`Po}#O!5Wp8OVB$b#2M?>K7MJy-%KlZ)=J_l`$F*i`w(qRj<p7?XnU9
z>&)}5IpZcO3>!geQuFR_a*JbA&~m64((yIlc8bPfCY9ggOn*QDVS^J-6tvP2(<EpR
z2Ku_)6_rb%LPJv+<#d-Pn>_dunwEr&zA0!Kjq42n&(*P*sP1uo=;@_kc!<`vWMoi8
zNJ2J^nVMdI(i=tsPVf8xjwN6O5VT;6P>6rX9pQR|%O3VTw1GB)SJ=t23-uQK3+LO5
zaFO`M$=VazJTeJA?LN(ulb8C+nm;k6sf~RUL%x`sv$@2OJdo0w4J4Z7KD+VYIEmf!
z9^JT<0u0ei8@1#Js#z9oDEBP>H}c>GvSM%UtBhU%(ms_ZFM~SfpM_964|wug=mhi?
z5Gj2{Q{*&$8%~Ia&<F|Yw}1m)<k4Il^&=|H+Z-)<gCNR4+$idIV0?$pcVg&-2m@Mg
zeN>5PI|iexJ)=qiuRLQYc8O|rw@Ud`X53&8H(1}SGF@?w65*5qt0+fVY<G^!cfSmi
z+8dN^uB12|u)ZXB4(?-r>C9wErR^3b4_O5O8v5*bY>rd-X89sJ{sDF7L+`+lTbsj6
zz~0eDV|Oe8tBQ{NQ+O6Br+MiqBjfEbY93Z)+jy6GEn;Pt${~$|A1CN=V4ce44dqOD
zwo(&94Sf@8eCF&*88kf{4n&{y^nMsHnYDhxwouXW7G6Nxe()VHZ9unyG@9=zi;dm7
zuR+kTJbNS#YFI$`I31-v+Z06!atg*&6i_`*g2om^<&`Vp+CktJx}E1GboE2yHx5vo
zE3&oPy&y*M&Gkh<As2pL^Y@Vu<Qgsw5>;lIZs5`dT7Rr`)ByhbKH}{SSinhvGo<y^
ztBUX>Ie@VvrDW4iG(-40znt2*aeB6nSrM!RLVb;Tqueh9X~aW7Jc_n|wL>TQ;mAz-
zxzh{4f&syWZICtw$Det<KI+-Y4$XG=eFbgaGxsea8$xo#5WnS@+uf9?lt6>!Yuh3e
zUE{;bMdAX-8U@aq>RetPL{EpL{K{^DZ$QOB^j`TBX-#Ey%>LQ^e1fttdj&r?-M!~`
z5pW$H5LUASkWVy?8&SBkQ@fISQ0=TgN(X$3*EJhroanDFu*f9rqio$x_}d{xs8Ek@
zIMKV>4{S|)lnQ`}iZm0GaR#r3r<{tCOdd3KAnUe2Ls;y@ef9ZnWr)*z9q>+*kU11)
zgLp_ZY3Kf}ekKQ;rXaBxIsP>8)7}FUVsrhWVUN3kU4AQf{fsNg)0}0-p=!DZ2;c18
z2w~us5_;-+1Ju%VdhI9?9t?PsT|`G{Fdf04#78H{2}3gw=&muYW#Zt6m+e{Z&4#Wx
zLIuA^U{Ka08bw5y?k`VikZPKiV*oZgW<SKMUHV3z4nU^l!+&#c8y}<We7chdF->X3
zxZ2OpJQ!lH^YC`Rp+4e^@BS-^{X4|}e;qY5(*N&GbRqwe*gvst&nK1hk+KG}0FrjB
z(U=yct75EnN+k(+creMjHcDb<!^5*hJ~oJWOimI%b&RfVpDp<&4%@=XbyHf`ze>d7
ziBgy;>RHyVp+6I>+%tpZ$5}c(?@am{p0BSrr|5R-%AQ;5xnUx?fe1~PD_x(vu62An
zHb!`$)Nr@0+);tMyx$)$OW)wxKHp=PYI$zAe&6LIJW=8SE7w|7ZrDS183j+Rb|!2(
znkdcaMb!@LZf{&UByh1o<)~L9bPFq0&rW=m5*tQ*iN?1uEuGkCq6#ojV%ViT$*Csl
zcZN+?0$pt10Ah=b7%E=2B(Yn)sWWdld>^$P2jY8k0=x(L2?yR7H|mH_Vn~b=kKsdN
zEIINSC$XSFt;Yijoi;D4ex$evqDa?g#nmRT3F3u|XX37JrYIicODHq^F$P|f=O-jG
zL6vK)K0_9Fe@Vbgu2-zYAbB&vNLz9w;CPXDP}4?LnPKrXGt13JA*UHsFX~ja9qQA{
z3(NFohu8(4=W(DE<?d@LrwIO7bXysM{i!CVtw#PFnpq_12jGa_SDM||Q2#^v_XltG
zk3@jk^B&_QpxB{?>II#DSnQyo9Js0OX3x%g63k^h4sM}{{&W%v{7<BNu^^^poT-C+
z&WCSubLmWJ*EI(}`$3`aWFRwaQ+g5TEpIo_hBR@a6>i7`qCccXx!HmcTlrB_sOLgJ
zH1@b2R=EEF=J5CTz+@UbKi_!`8N-3{+S1CbQ_~Ce<FecVy#dUMSb%?`#xApv4^Q}U
zvIx$zPi)so5&q}(>Ge%5Kuu`u3nZ)^Qkl>1{o{?>HJP@rGS-M^%__TvidbRuG>bas
zyt+j8UiXE4+l%E<H<Wzsh=(FPQ*sOK?H<V$FTZBZ+f|MA$ppJ$#VWgSCT)gT86tuu
z;;2b9_eWIE8~2k}gIY{xE!Bh+P>rzC`J3e^^CG%&kFXYyv7qw9l8JmyFC}<>_(gD0
z$~Q-$BzQ%=z+%_c0cM-0=it%)qRQ~@U5MrozYH-P_w#;LPc(DmHW0*N?qolmLPEjf
zs_MZs^?FY7{t9<}UmV#v1}k_c!+{yW37VyTtffWLg{oAaTwtlN(-pY>CK~z`O+4C;
zvGaHkuKaj)sOyXj!+`;wh0y{1oKf@9s#E7t{AoD}Pi}GqGz_Xr95k8`#`QVOxwc`Z
zzPL+9gD0OJRC$T)$*kl#{?Xo1btCD%Li#2YRORyyJFK8f>cQGJPKb#DXBLyc2hdl4
zBaX;by!`=x6@7YpKe&nnh5^!OYWe|a^9+dSOJ3CAK>KE0C%8^aXhN^RmM!;_&?oM@
zRF?I6qRM*u4RQJ4%pa>V)PW##TJ?3n-og(_`U{6pK_w8Wxq;wE9A9F9eULkJK*)Vf
zb_Dl_`p`||iJowgdFpk~aKcmCBe1#(xS%Nn*)7F<@2^x~L8tz?3zDijkSb<+kQ*TR
z!z*3YK@@JOE>pBZfr9xaUGi?J5iBPM$ZKKgbKd8qeBA+J)H>v3`=J%rm`TRM!j&;4
zK1FQ6Hhc?qpj^={Xll<G4(TLtbG<fgH|*g&I<qVfOA!*eVisvZXnKjAir9ASXQ``-
z<!9lFift4AN8iRlU+%P7@&JlpXRK<6Y8Jgg_F3nm2Ehi3KH1;obX<^M>Pp`_ke%=~
zcdV}UizVMo($s~HQBX{kwN)@{=pJkrzjfVKHuk-Fbzxzg(EAGhu7~y~W+Q3{J8N(O
z2}`I*#3u#ctmp}}Qk(?HY6F;ltv8|>4n5HLPk})1_diw8DkQ0(uVL>P+72U#u4hRk
zZkK4;%rre0IAhbnP*OR-#EB|2Rif?9K^Z|43VHJ<;_EbKkK_*8z)HGy7)Gv1H;^u&
zyntSk)xanz28!|ycgCUvK^HRpxx3w~C$o8;tQa}MOdA}Z5D0HdR1R19-c^x_JH&rz
zWMhliC+{G;FDHj2_eL4Dl<{mS<;cj_tdIATHNO79zbQfH(2j@KRiPwgjk@>ok6%E+
zD^smCc0)V{I=T!1#%|?4T=bq-)!ooc^Mvt7%`%pskQQ9}Y9}Wr!pKJ9v;AogxoV&@
zW7%gmZG_{29K(ZRb>32K?J=-S7p=~&2o3@scAW*QOKoU|(D*mOH35o=PUtE7G!zcI
zEe0{QPR{W`L=hcgLbUgCHtNTA4ba6Cms_*zwOmx3Ds0e+5Eey;w6F{jLWrbP;Y&<<
zOQjA19@g`Y_5DRkaI!3T`q|*~^3<GGp2M|ntNH-2v6B^Iw_-E6$TUv9Hb*7KJJv)A
zUX^wlf10gx`Ef%<!?!cqEzP-n66^bDBabw223ljG;M933rl&#p`O3iyb~RtfnbqRP
z8K0HvK*HrP$da!M&l~h6w8i{_kkWylXUUdO91J0{o9mF@9vr=fG5J0G8m7RvNM~M~
z529|m>K?O<QnpMGQys2`_}TE*DjRmfOSh%0mVEql<cWHuwBLbkw(xl<F?lKZ#x2g~
ze2PZj<uE;RbzvT&s!M-8SctrgHPVNdKemIWrl-<>Hs%=uFimR*<_}aN+)^xXW?Y=r
zo$By5VtnoLvar(Tcg35bl9J}X%KEH$Y_Dp2%!SoBNW3D(T;9pForQssw67@9pf3ct
z|Eh|C4KVA8fsz8_f7h!Ch$w$f8FPpL1Lg3mw=xgm)Nw0_7>ZT?_|?3;KHqHk0R_MO
z{%4~24_Jkfftllf-R;Eu|8_g9BjL6%{(?3INmqC4*Wvf%)RN#d;PIe|@s<Ua22qdC
zvIPHjI}{UTKuB6ALbUKBY-O;!UynPxuD872DBbHQ)#4_d8@Ox^of68E?UJlrj*mtp
zBudtw*~Q!+Uo$tyOH=TAc1)<(y1c(;pN}`Y_`H?w9h#)kWE9r1QA@)zo@3|TG5Xqq
zQFg(6;z!cuhD}ulwidnz+N`D9ykinr*LH8XvtZaRTA5JH5{J~AIW}>Na;9gkL~)x^
zTA&%Uewd#HtTX=K1iP1bZgYjL0t*E-be&BzkX<TXjvuwl((0olfA%4#1Wf8KK;yPA
zdup!qWqzA|QY8>`V({r}(`aE6jpJd!I!61PZK?p75ffHmZSFl=!<b^ZTGXIF8eC&Y
zEaTjW(_8_pYHGV*vQa%)tMW_@iz3rW6$?N3?t|G*iov~D+6$Vt*s@JVj5lCSp>XD9
zB4YGxjKIA5W49`PZI*pg3NT$R^YJe|8Y~dfk@-Q{FoYgU1_<d3u5Gml56&!DFMi+_
zV<th9xW#~h909Vll8}`(jI*e=OI^`#hp$m*pJ@as<7hqE54AxYSe63}P@l*5sDWmh
z#=i{Vf^4u+o=;t?C9~Hs+#o2XlR`!<W*-C#Mz8njU1@Ez-@ExAQL{HJ)?f9ruPX_3
zy1oAyym6kfh6p$}su5v40QAe5gU&6ZWWgP8ohkF!R%H==+P54Ay$gI38-wp^qkC}_
zh;PZb7tt5*+_WwMVB$hkd&4Q1J8ke{l*G77c6>g0|MnZQZu{5ffbNmPYZJ_7OjBre
zx3_S_eGF)qoAhCX391TX-NJDB)KpFGXQ3l%ufUq&&rrd|r76Pu5{m2Ds2Et;-EkS&
zNNyVQ!L>K9&e*ITETF+h$3s#zg2eRC_+isrXc<D?5eRnHXyNBt06<{z^amwLZozkQ
zV4TX?2jImz2NLJ(i>KQ2mks`%=+ed-qEMO*RAIt8H<skgI6#^jL_us~MwL=B%}yrC
zP9EDFGtidi7f5nd3ugKE6nl)Y@&QO3gfxM$GXJnmc(}iLL^y&xktnD55=E!uP#nyD
zAYi!p_+Oa1QFp&sVkhTJoGsgaUeXZu!QeJ=C<36p`j8EB-HpS=qBZJ~u0R9FPj)OX
zX(b2e)#hf40Z-26zVslMlHvLN%p>=;T2QxdW=z%T)De#XcWI8ZEx@G`n0Ye$;dPAU
z1XjmQ)ejp7O%B7DrzjV-ID^nyx)tWY1v-MU?3#{6S1i7g^K(t9>}FX*w+XUY#Z-kB
zsCGv^8?f$Gl5H~F*<BJ51cByP16QV1FJX^auD_T%DTaVuL|SL@N754}oeu7x$>s98
zo4RV`16ppY_|qYU`Ync%w#;@XP=N=r`H<{`id*{!qh(a_IxZ}Vh$#!F0T^SCSp6W3
zZ6LI%jD+MX-J5^Kz8H5twu)M>Z@~(r^8?kT4O%f*?_|IbH;XqCZ++g>!($Mt#gWd>
z2+bsqu)`(M#_kjKYQ+!Y<^rAav(g4uU!vgP9K~4H7;qkwE4W3C)TjhFP~RDspT;@L
zq5RwN609rA8~c#ZH9wkxL<4sBtHE*FfF%jGoQ2|Sj$9#2Ky?-$Nct;?+I4xDC@P_e
za$(~`42guk4N&iax75Whuh6#@j;pXBW&4Y6u@-T`f^OuRJ=Gf`9Jw(LAU$Q!30yoS
z!3$k+Xye;=8}&~Wk<`_55upE6Voe?N#y^vXkeu|5H9W>%aCI)M)U+{|<bVV0hiW#%
z-Av%^FcF*t^UX3fquFs(w+Y!*9u5zq^%5=xl14606xh@bizbkmVA$QQyal9#DO+zD
zto3klTTK;|Vk9H}<Njcfw_4T$Kr}F|wV$1Z;hD5Sb#pFlx~Ba3nKF`<lkUP18tnW0
z5KT}F93$(G9D;q|ySGS*E?q|d5n}lqnX~d0-O!yaQpJ+i#pCLV@rS;BG;ce;NUx8+
z`!B<V{{H;X;tmkf9sFjMf9?lO+fB=Jf#GgvCPG~>d<*tu_tEDj*1~qRSpV#+kO{4(
zM}F$ezn_&U)>J8NZR~3^Id|${<g>7UcZL%9N~f&|qcD?F%;(pXX%?IEh!E~W4GMKE
z)wwGTPh6T@sKy!(h^p&KR{Gkgg<ZOES2Xd@mU3DEdx6g)F8;3b=l2WC*g;bE1=~M+
zn<uaVt;)B5Dbd}<i6|m-m{|ytmiL((nn^r8L$dlpT^-BW=A5V>2x7Q9U&BEN+D8Oy
zYp+4Q%7>S|`CcLb;n_Hdqtbj>VSSOOJboA9ylY;Z%?zaCk@Ib>MJ6wY8f2o^HNg-0
zA;wmCGUm>l!q{jnB&sa}suRWH;#<Mpvfv?w$~7MKdf4xB7XX}(S@!qR9n%je488J9
zqCL~8y?uZ#55RgR<8Vg>N1zN7Y%aYN{q`M@q&M(4!-RC9kSR|y3LGUT3g09R|EB#l
z1xnzRwaeSaFWP(hSZ=PJeD+IN;is#`fR`Y=QXr|~>DD*3pD!TNt_GkUj6*ixP63Zs
z|48gZp%B@F<tph&w_uQ2QF2u$*M>+9IokV8>?tdUC?*cY&U7>sPpm?ZaXdyQwaV?(
z<Rl$Er)kpaGhjsRA(M2*A|w!W=6zbP)~7rQJrA+MW`8e--@m>{ob}7`T6B8O+FDjv
z*u8wv0?)8L(?hoA2lNYB8e;N%?*|<%vJuJhQck<Q<YMbkVxLKufBL(kwk~>0^_bH2
z?>Fb=1QAJwh@&x=7|rPfbxeO{p8GBsUabfU1f}M(=AR-r9m$f0Dg4Q&!W#Ju8hyHa
zxalh~?3n)9Ydhw&9(={`a{C3|ptwEwuh_!;AA}79GsFLEuCh`4e;R%+)$UX+M@1px
z3C4?(8g`sFIa5}~q%(n~2arm=$w^ShetV&>5`l1I-~dv)qU`YB07w^jB?~`SKCdr!
zbW%l0shkqTwob`sh%=6<CybkSMDmPsPcP}V^#1PL7sD*KD^|;6vfpds`}V#!_v}>H
zg-9FHu77BrNPJm5+V^$5!<MxDZu`DDectc8o!X`#_*;0psXc@{sY+60|6W=ISrw>j
zB!h_3kD$FQoLc3QNPO92#g!o@&>l$iV=*U%f6gP#CMz$r0@}Ihz|SipV(Z}Azt|;)
ztFE=LKYy4YO%-2@w(U8r<ZLdqel{)gFDdaw={fIa$Fk?OLMo31;W?Ckvd|eZMc8a*
z;#!Z@X;TAuhUTJ@P7yG}`*35_Q+4AWt-`TippC&b?Vk>nzZS3Z9KJ1TK>?ucL6Pxp
zD-OdnK|Dh=LV!}F17y?<^;pQ%i8TpL)sewsV%A{Nq5?=WrT&4E+08n&4b4rt$I~Ii
zQNMgaCBqm}fj}xom@&s%iUqE<i8`#VhtD5i^gmgXb5Z*i<3fbVGfNF8SkV|Yrlemv
z>8g<3gZJd5QtswyH!hRFQ={FoeK5A?hP72=b0YkRhzlj5YbUILY2{c!aAV_GFs?a}
za9-}Qoi_K}M3@v)_rUU(0Y&ak{~QD@J1E`C$Smj!!!vO$w);ROlPiwy-?6?jzjEOM
zH~~u|Oq8E_&BBfo!`E`VqB+6~F>^|bZQGeH5K4qmFsj&qPl=~K<(}M|KFP-VJ;9Xc
zR~LzGC=G#;x;ZkK*g35?*><BOcUJxLwjs6B0vv2(){|3Vd4c!<6OsXiU@r*;u9QrA
zpw*m5L~XSm$|*mx%Y1ZJk(>JAjp6vFx|K6(DwRa^^A_z0z#*@SPVeWe-?h1+r;r0@
zn&Cd73=}jOfJ#bKI(c0%I1JZXd%m2}+dzZ1?1uqA%ruOldb?$dVH0vW3IiNJBLZ|!
zp&38EYn3w6FMI-I(T<WPfbA~>6~;&t-r2T(s=>dWQPs>*xQ<K|y<VPS#vLSiXk1x!
zADo&Y^)&&*d%3Z}A?a|_fjsTPlLg{3ePXM?u!{EukB+rjhA(SfSY9wEb~$;in;Vz=
zs&v9<`Sw|?9HmQ@=5xDxItrkZ=D*M!J6s&;cTE|-T>xDt#!sg2pH$%pf<8)avWJM~
zjz5W`BNUbO0tH$F4ndiEbru;wTp|`zC}Dz0Q_;@JPh?jQf7}6Q#~`n4HvuXRUn?0o
z(E5mK=#@QdX5QJ4<8RPw`PAO6a*CaZab1YyP@dFmH_eT{6v5DIP;>YlpwsRN5)YX+
z9EHCl5)XY)&P{sZh&lidF+QX1#R(fTHMXE9OJJWR<Y2kWZQn)L1ql)b#l9tf2*ke^
zOLPb!-!>~62?{#2*hX~4*-$_17VGFDMVfbc($~&(!D4R@St&7>D7nHC(~B1!#A<iY
zW!RW75jNJh-DPoc!FUFqFyuoU&Y5p_zKIG4>5$1Udw4%^mUFHUyb{Q+=}8k&QBDwu
z14}++hRWxsrJV*c*Q(;PnShVJF{^s!uh)q`0AuTN0O3v%y_z=?5jn@zo?b@A@Rgpd
z&=+^|eu9Vp7AY~3pt+AoBaB#LUvV-%d~lHT<=XYp2}L0B%1KvPsp0dJD9nY>C0y2z
zPp4ZE;hrjZfSb3xaCi~-z8hj(qUh`NW~MOCI7#xP+|FS@jmtxee~r}YUz80wy7c5t
z9h`11a&!;2wo8DkuCD%8*w5yMRC~8C`vI8O@8%5?Ft7ew{xV3WnEknIx3(`9BX!1>
zEe(6w>!As}F-s()VkC2u*X-v8=F7?lOKvW`^jv2(yOgd|Ei9k#3+UsTGxwkW?EiHk
z$;$FS!v-_zQg++_cX!(vIk0XforH2Cnt4_+zZ1m8TPg@x&(w}e;g~2qSMO7s+c;rL
ziffYuE(#4y`f|L`EKHvVO$;kyj4xkN)BW1zzlH*iD0NYGDP!lJSYa%Qlq8b5!b_DW
zR@1=jzP=wL<A}es7%AMowyrfA7k{f~(YXumrHVcaSaNo}-xVXt$vh^NyhunrW&<)r
zsgWQh_60MFNmT4bK-IMEqGs_1je52+JXO3$>l)8I`|5EvU$QIKmwn#-i%}u=#ETHP
zb>W{I=`uCbL~h&88!On@d+0uBW?N@?WIw>f*eT*>YL}&gf(zX&wKh}`aH)w=5IrId
z`_5gq&~NNn@n+LY*2E8kPtoa*nQJ3EEfPLr|1^7*dwqK-WO6ADvJIgsQ@BwEpy4W}
z?YC<M5#Ll~a+R?nfWWzxUgKn;W+O_zlAcl&L>3)+?qE|zPa85DSU8*I5{<;ecJxz7
zMmz#L0oRl}JYfyNFqgWf2J%cIS;$`9R}#Fet%EjRdZJ#z+haB)DfGkfGfJUW9^_8<
z@n(mKhwvqmD0D@PzNZM$%Bg#^L;_HUYm%w61$ju%q6b3qmjw;K)uP47v`|T!4T9{R
za^0*5?JD2ZO6h<>z;~fyINj*FmMpgHE>+)i{G-emiE5fOzdvivVLbJf4%h8?_E)W4
zI$okKHxA|tzL<5{d<(sBXs|x)OLuP?j@dlF_B`No!%c_ddfNT~##-}%(PJbNP@|TQ
z(aGvn96YC?3k-g@s@pnrZeMsBal*?2ivl?vD4ICO^g`gz4u$rYaN#Jp?BTd(sn#Kc
zQ2Ks<fy#E4g3z3<qG6Iso{*a;1RYrMXTNI9FCXHVx-5Zr;bRPJ;T%0@vB0JLAXM`;
z&|5}2Rms>(zQNEJHxMjOGQ`Yb%wbL>1$XUX)3g%^hBIX2)6=I2M<-HG=(LBPDp){S
z#TI<M?VE0F8>r{u(V!Y+Kw+yP)NCEN)+a!_F%7&n_6tf72}FZR5mHkcd-BFmhII%@
z7J|ULad*Jo@i<~5CjjwAmAfRD2#fnFN1VSW)4S{R!9OD8fO0uX;{rVK<}zVmka4_D
z>_a6<fI?o5*LY1+5=YLa=kXsY5M?_^_W;PpM?hpsZ}haX9ox3F?hyvFcjqPtIf;-9
zr%XujR#cFzK2tEbXCgpy{DfoJczx&s`X_pia(d{v@x?B3IEdRh&Ph>(q#rD0buE{$
zT*Jq6xMO9%ZHNVXh71>mriDfh^(JajWWgStOQyB=(uKtl?Q`SH5Lol`v*pO3F^3uN
z2bS}6cBO|Q)uIsWh9M=}d+#xcJ7b0gGvM1WzE2_@rH(oDwDW#CvqFIxfD|&MWs{c_
znj9l#_@mDZbco2B|0!A(zHH*dr!p?*g!B&!gtWFq8D}ogevfp9VzlarP6FXS$X*uQ
z>N}w}>hpR2@lH?-PZ%j(l<Uz!4mS~2kLaJ<E<v6|4ehtS6%Ehu7Zv|;-#;eJh4N*N
z%SILe&Jt!N19AKb338LZOj@xIXx0tOIB#)TD&qqRQ)rCnq9IR^?hvLI)ji$(1>a*Z
z=T&qC;5OyKYJRIr5JOeS20d!IZp(k|WXwYJ-0gLN%j9ah%5wGnXu9QvZ+g##(%&|o
zwli_E+f2d;WfeF!lKq|7m9k;E{GqrNM2@6#G~!JAI9(v$iF-dJN|Bc`Q<CCYq}ut!
zxu0xxaaqlUpB>_NJDVfan&$D`jl=&JM+?r}2KSR0eiX=g5O;HF?7x`LR$!lyS7s{b
zv8_<eZlZWlzN83kRbvGTGEar??$N=rs+rJs&s*o)jwdoH!Rs5m2jePptBo+y=4Z+w
zwMy{sG}jP~TWh_=DC%_h0F2i%jMriM%{HiY_5qU>WZU+zj;WdL_{%d>D83i^pApk^
z$QU?Q)g-^9V1bVRnDm}rXz94Uv{Z8gO(x{nx|mqor9p1}{L$t&3bi_!ebZP8ewkia
zdl{;1q1MJUi8!2i^Vo$c=`%dz-nH8Tn=Flke<Y$vy7TN64U~lrEU25%g1v~-F*^%b
z6*BLGtKwbLd#{{h<hioWajj+n%$soJZ>yQ}uJ&u%&RmfXU7h*11gLWg51<YTCCBDM
zsoJTQc&uqo!yRq{OTE8xWAVooz5zwiUVJgA)sSkMz!T5&1KxX?*ZZ#-#_}IUIc)#m
zo$CMKo(@rwvCHB|=s8ups>zR>@dZxt6i4n-kxLMgRY0w|-;6&P0m>%-?H!_R?RUX>
z?{CcYcs#ijm<L0Adb|*Gb>TtBDPdQUF0hVi7gQvd4i?^k;sI4A-Osyvea=pFHWoDs
zg^U0k*4X*!?3go+qBl$gNj{`cj0ACO`W(4>`4cS?KXdby_n6`5r#+45xktq)&%s@H
zN+FVbpe(s67715Cgm%a*0j`O|#C<O&he#R>5?6DdaPiE~oe++MGqubR#}T`c@aEL=
zsJfINZ9uy+VBXW_p|;xk++V^2{g1>a;d^tAX}#{WZ8b#|g>t`kE@~@flHqc9Vyz?#
zSSEi%5=Kl@SBNAj@W(B(hPHPzT?=N0l&<CCcZNvnjR8DBnIg#r@`fgsMO7|q&ML6I
z8&WJZ4T?PtsUl7?%w&S^6m%>299mD|!<pjekn|5bu%!osix#IQ7=er^*eufZO9wU@
zOmRAZC)7EsrAIfCilj(T+Lgn8AVkF*(hpBkPF2bI9XZvMU@$0>9Yu)<HY!MV(@xhZ
zacW{|<wABl${uDXB%7TVfmE8GdkguZP)Vm?aPje7@aUm#>A^vL!SybyEz;Oa1ntc8
z|GBfk#t4<k;>VR}NAv#n&44LG(v@9r#;x|J$|F|TY-RF1*4*D@t>mF@@jqGct?*i_
zrq{(a&abNUeH6oh@*9V?Vf$0Lwoqj&*;5VIT9xLr^-QSccdwf4u`kB3wz^9Le%Rew
z)mVon?Eb1Y`l4=Y6*mHygUUDnDHrPD`Fij$Z}Dk@fTd$|QF;|=6D7B5TB!=soM1Hj
zu7LgSVRNNoERR4pIyRbX1AI)}$UG7aGxV#kD^o-TfEfSQTirQcQz{?Z*N0C9(T(=8
z;+|B(5Hw*~XnD7?Ck)!Q*I37C#}K_P(!v>20tDa5dV~3`=2mE*=$IGT(7{ooqL*nv
zE?!^5vnlZD1-alu?fvq|GFAHn2$nhfUmXODEdT$i&;O6Qoq>`0e|BZ9sLTAH`re*@
zANS;PLLhl03#(zSr7z8GoSh}O@FYV>X5@;f5?w#9ItTy~^2tf;4wrb`C<4^=>#r}@
zgY284?47!M6OX4W2j3|qh!TjTfnrsk;iQOjBuI=gs%{Hor2A(Fb1QZ|5>E_&(wB~7
z{;I)$+z^!l(bY9uZWh(fOuwpSiy%z||1sqWqz4tkNeG8b?l;vYA)nBH=b`S~JZ{l3
zYElXl#lMzHSc}0tb@bRx&H7U(sd6M!yK`h(hV^-gC@tQ){8B^7cC$7Bx7O*5{J5B9
z*=1iaX;al|`_t+2=z1ZL<>Ws!YZDG}8eTHpseStW$Es$h(zT72<n}6@veNO=dpave
ztJ~9ASJ|jbsNgGc?=Nq3yGS?z1qch`h=MexVe|_?LIvQ2urR%3$Q8YRznKKkE@#kz
zy_Z!0)LIGVB*c9a-ZF#QrQlWuh$}Yzh&Yfha5;s4B!?tN3*y5zyUEZxBPn?aWAoP!
zk!dh6Ri}A<1$wXuW_3z^tcES_AO;Sdwi)~3#&v`^VmF9sYLvBx>)BmNz`SB2%}UT@
z)NI<#bFbr}TgQm~QoBc1Os$pG?dg?D2L7fXCSDN}Vu}I3Mk)cVH#b+{6>ORr{IShI
z=pPOJOCrx2DIevL3NF{;&bB#I2=4_;xA28FGyQyjGbC#tJdj(u6nk|dpenov+a*9m
z{J(|ecruAHipeGM;L{b65+}68@zNCt3r`-~qNlt&+I{mEX)6bHBDV-a=@`_QK-b}{
zpCOoL5&rKQAQV%=Shz0EeQ)0-vazJx!o$Ks#t?Wyy<?E0Q}{-oIHJ8x@ZUpg!BuHF
z2q&_o<g?Z2*eaeg+zP#23DUig>*4(D31syQ@{x8R*kFx=>}U9*ff_~nN25Z}ig<!j
z1fv;4tqNl`wm7M2OOy!pK;`L^(^H>T1bEQm0@<PMTohwLh5AM0OoBk3(#7mGiUWAb
zz=FQ9cDO(mtfb9{sw=i`hl%ifla#7EZ8lxE*ucB`)9=C!{5WO4LfAKa_5F8ZrrU@7
z1gqfh@0@r{l!%AJDNd_tPMk{)fW3SxF1<A1eLkJp$zJ3%<U@;IfEs*C4u(pskZ)MX
z<XDRZF1?OTAnFrenK~*1!0ju5l~TaRwtqq)!B4^~*bI^&f)RPDYny>;zaw<yN*2(1
zypWLuGAh)60~(cN5uAEDpiF$eKPYdQYMrky=LupO?OTT;7=U_#w;1)Y@2$OR>kvVf
zNw~(^6}x0ReJQb@+rD>Jhd}q4(j|fcWR85JrZCAm@$UVGr4l8D{lkAb=_J4=VIFw?
zoE)<qA9Lr8UcuZH_oX_0z_`m&JQv*8K!RfW>NXDOhAWMj>Nx`jeI7Otf7>Y7*ujrk
z^0?s=%b!LR|1GywW|~Zv(;bH(9S^>nqB9OSJ`;SAzNIS&o}b7qD`R$P;!$EB^OX-5
zOl+|4*#Fvpp0ailIBNV%-NJr7!vpr}vdW~r?Doji#pXANe_G>f?`-jMqI-zw_08mB
z8ICCCp})plVLw;Y&CiIRhH-5yl+dHjWJ~g+Z|_;_hb${N-0Lk6<qlD0m)JGM&lNOF
zzpt|%+EPmpT?$)zN@+n{Tw->uhb4!T(k2ln3exbu;B)5gi}|mt!OHyqM#30aSpR45
z@xSloe_}y|-rG8a`6yUnnB8JesjVGm(8yNovb16d<1AN@@x-Dr+t#zVTlr*T1*$IB
zMQ*r2;m9EmzidLdRzSF1z3=*VPtPF@gaUCzOlnrZq<}pn1KPk5PIn$+sJUOwuZO45
zJ@fi#^N`^AdWoisvAn$R-uB*{Fy%eWeRDjh2(z}ezMq%VXF1!KlaCGU`;CK@>u0I1
zLsVOnv%hk7L`VpMM1<ugh_2j%mGZ?!s(7*bpM*T&r6T<FibRf-5XB`|!dsaaHl2P!
zprK*xI^Lpz#1rmqzDj0QloL&=R-)jVVThucq6C@3%$B5`EL~r-)jJ;vGzt3ETW_Bp
z;}E$te=*JCO>&A=(v5ACj}5e^BXVsnb0#R+I1)qZ`G-ytS?~}Li88YlmykoDN%)^!
zJrrH7cerMnYACkdywS-lx}{+=n*onO6xOl6fJF&%5MM018Dq}kJfH(1R*8kWDYb-@
zb3|y{0LQ|`dSENt0hqGwf~u>fyLk&qmBQtBEG2x@*aCaTOWDbb$0B>H4^}5o7+52B
z6sDc9ABG#lN}Az492Ca;WFD`T^Oa(l&udy(|D8X=*UUAkR_SP`_U=znd?Ex>SO)_e
zdPs=?$quyZM~&PV%fdjQiXa~4h(Y{84xF=nFf@yNGJq>*@?wPN_(Ln5KG?)F9G-bR
z<(^R-1LWEWkIXg+{kni$lhjC=XyU`1%xQxWQk>GOoW;#o1BHxo&#>a7TV^%pV?o_h
z(qBy+uo6rNb-!L6`})ajVHhRWG+9cz*7V0-KYMM3au<`$106j=W=_cx0aTZ8+w7jc
zMwkw10mh+3%+aJqy2L{-%X$SW!EXC%w{&7%RoBXut7|Gt<52XoJ>0F7QN&~ML6Mj`
zVh)$^315(jowQt)S0jbX3tv##U>(y41#v(1DrfuBlO>e!ap0TD#DiL$UQF`&9c<q~
zX%=V>7^G@=ebNUch*^|T5trfuE(9)q%G1cBZy?>*o{6lWlXNpn1*y!|9=+@t@eFL4
zAIOi>xo=&%h(WQNPpm8uwqN1-t?5GE!GhZ)`<lmQyR5}7v1ZXT$9U_dPD90G0gIfD
z6Cy}jFJqB&Z^C{~R1UnDTv0h(|BJSDph2%~jxX`T)8YB&0)P$z8T%5%LN;*aW^$H^
zLcgaK?MNFe1d>(F&$~K)XmQpJfgdgj3FFe~;05_TP8)wvLdMytexdx>FHiYaw8KoU
z_M}K=4tRV^;&z;qpm!>k@}~a#vMWtdm;iP31{J`5!OVp&c~Uwe!0Yu==1}kSuOByE
z(eLg|`+a@B%{e11ihMg>WQq~{!uRfq3Cm!_>OUt)1|B3?KOt@8n?X2`&A_XTAW+B<
zdKFz9gz`{B)iQ<fo9M!xL&b*;d!o>F?RuJ=FL{6buy^`b*Idq55fCu3>DD13h<8h_
zgzhp^&1#mB4RU%)xh}sJJeTxiTN5X~<lsugcjhP9LS40+3zXQJ4)&vN(E~$FI?V(7
z`r(1-G;y<UesfpNo(|<hLiblfclRZ!fkp3DnT!D@I(%9AP^o4ZK-6#n7f~ePU^4P)
zJs<8maOm{2rz@}Pp%WL6S5L9NU)-E?8C!C+A>=vkyYXx3ty{If&g<s5lOBnRiRQ-P
z5)DFtd@jOy4Z4h&2yg=esQU9b{Jm%KB~*BGo2GUDH9OY?;TqV^lyVxNRdZ_IqVc*p
z1n5y}<S07*cfdCWfP!;795%*vcWbnM0LD#M+6jH;9#BZm!JyTbDj8)pq$pwlNm4E-
zU}v!KSNP1V@(ZI4V0$sIoYsw%S4HQ%9X+3ZEHOvQ{FG3~VSe_re64rqo{KoWtt~=U
zGxXZ%%an%O<Xg@FKL>t8M5;KV!)guYwW`PWyM{XqE_zULH%4rj^NVKm{st3(NrBmO
z@LyM7dvGw~qz=V(Uox_U`M+nBdHV@Z_~H^aWRRgaiYzoe$7WJ4pRI0aFFQLLFuwt4
zmj%x9QS1a(mhi8TEu#n(dq=~q{SDocXRlb?6bqj_A6<kOVL^T+n989(U(rAfUP2tN
zbdB{t@c%h({%bwUPX9k=^BIjDyMN>6pU0LkpAHJ4>HoYjZMOVTrDyccS{J-6BK_me
z)WZ0=BwY7+FKxFZijC#tvN~N`x`G-9_r15iWBRA71G8IAYD-P0X&_;;;es~~W}I&1
zZs~J~FBwcZ9C_LY!~wH?{~XgXMD`%oN6mlF7>4WN$C$VN>ge{pKM!FyEpRT3h1a|8
z9*@1&mEHP!)v<jvx6MHd9vRi`NB}9?aSn;9vZs3XTw{b3_|vTu!kL7F4;u&*9E8m>
zgMLu8NE;U`rl=54e=G+AyXY(RvFFJPZVJtk;St-m$cyp`dW904U>q<VJqDnxN%($(
zcnu2N`foTNTl&fde_|Tl3OMz}RgLhz;6Wnp55#{vvib$~ok9<XI?g=`Hsw@QM_0xC
zFzA^mpq^?}_zmXr04bah3yu&6ES7;31`>hw`jQ~T+sK8nkHA2xXWdp%V^`GEkhmU7
z*R{N6hN)CUZnh?6RL*2oWXP<EcA?rJN0*~>A|p&d^WYY!QO>N<Ja<_OQ}OW&HUsVi
zqqy=ZFNz-#N@C~mdknOBz9S&u6=~MJdX!*E?)Llxi~Hi5wlcF_!QW>D&_7ISv@7t5
zzcV^bP{}-vsv3Oho)<4_{3@(h9m1P>eo8>ztqE>!+;=U<xlshWwVXx2+7|;;9Xh31
zg~EW@P!y37G-_+ok2Y#3pB@FV%A|UdrBFv07x)xOnToJnorDmv107qmPefKcoA8;5
z|J>!5u25KUf4+u~?!EgrV^7McPJlC)n-L!I3J@CUA{)5iDdr6UgZg;}J#5tl1#*BC
z`G<5G3ZdEmZHzo!qojygW(8-@1JPl3wZ)He-6zJ9hj4@2$J^ZhQCFKJ)H^4bEwkx`
z#n1>nk|%1%NZFAk*;Wp8&2!jPhZt{(M)pW5Utq31j9-@t3MCv2_A^;JD+g7_h0f|+
z;I`*WjXCM2;h2tb9IxC2<3!8IqEjKH!VTW;(HmY6^ow_iO#9%N$>KDD3z{yuAnc8p
zi^xikoR-ONOhvuhrdHWdNwK++KXU|hUK5@7?xkz^2g&`hj?BV6p(l~Q5IWc*Va&Qd
zM3M&ySuPBJVL}7}1@xOPuE1!h@I05yT5OvJcd5B7Q_%KOL(V8vqcQ#*t@cQ@hS}Hr
zL#0un=x6R-vdX!AhMZcQO#pX+V<-t_k^sBi(l-~V15pN^kr>HoEtK_#HViZkOddR#
zqJ1-SrP4pvLK11&0fN`89j$G)R!+^XFX3Y5>1wXY<`VGy@?Xci5MuOV(+AHn%nLMt
zVR&OdMKTlL=b=_eGeV=u>n5!<fvj`{nD;1ytKP-Pa{XMofk%PfPC83q+EJJh^;!$S
zRGq(!H&JHq{y)ez*mB{(@0R#Z6rl9u)d$2q-OF%VOp%W7U&1~Paagp#VUET7#N$Nu
z+=eCh&t-UkkPBwbwda+A02|SxGbSvOY-t$<!z2_)cBc8+!07g;7`-J>fF3U7mXYZg
zxoMR5%*VW)Hqdu7H8&$i*kNyj_IaWJhTorWqnIIQyWaN;0(@Py1p=<Ug)er00r3M;
zkQ*36Tj&POlE|)@i21+Rd#5N{mo8g0ZQJ%r+qSKhwr$(CwbHh2J6GDaS32|Nu2WU5
z?m7FM`rB={-F6@Le3(!3i*H86h=@LVj4?WO)3StR)_h0-))E5r5`dQYemP^Xt0_4^
zEHW7!ayny3?TA%@vQ1QU{RsWW{Rh~`nN4$(1(#TK%MVci5*PYMjVzHd3)YB73<I6e
z$8jhRTf?H;Hfk6I(Ae|7sK&A;_#uA+5cvBuLu?ci2lesRlY&B(a~)Qm*%J-2dacqs
zgGFML!PA7@mwQBpvxBoq`#KhFbV^mA%h{~~fT;7eFP@|Mz#q`1%$k}$GZW<<V}1e9
z&PTFM`gD;6u&g{}(L)tH;hl>2c+CwWPtPLPS_At!k1aFGNoI1*qU{Vo8|Z%X>xgsW
zzu+^~tf^d{AS^Woxxl}#@BDDs_#dHv&l+i>uXbFE?{Pi$oV;@bm{EpA^Z9`$@k;E6
zu7Ll_Z&&yGV2*!0vOs4iKq+7;G|5^tF2HYQK^o=7WHN};{qjGp(KqO7E2q?AW$*u(
z2MRASw%Wt4ha)Q$%@gd)Vke5{7AN9NI@tp_3(56`%07|J-oWp9WsVl}K&QtEuRTYk
zXS7_I2iN-(9NqMFNN6%BROUO+e_ONPzwc&>;}zf<#YIkU37TIE>|9kqMHe!HXOy-R
zFTd7#de*APNA#hnfqFY&-o)wO&^<<b@Z3wh1!;{<b0?hant9$<DRj&^zAT3M>h}0-
z4s(f1Gx}7yT!Lc)Y4ItsjJ@cE@zH$+)+F`=AkZMQ8H~4|tyhd5*lQ;rkT-zm1)m*m
zw)=BN@LD=L`H!SvzozBSB(A2mWei8v{V_XF31|bei<J`%8&FHU16Y3Uk5W9NcIgM>
z3$*$)RDcD((T$$JtsM7v0_zOS%>S}4x1_%6u+EC&y{uMiyBIYDFeu1q&Y9TkkTC6}
zl+?|h4nr+~WJwE!zAEe6oeCfxo`fTo!mX+Ufpo4t!$VMigjTQf<b(H=ldWU`D%-1V
z6K#sNi}q_nAW37t%vs-X=F%7E>(2$wFFix7KRlrf_^b*IKJNEtHMe$)_^mLjX`CN`
z`=fZUzF+sbyEW}kUeD#%3ZCyi%wB4jD?*|nF#v%U=e_&CdmnN5c3%C?Yt>Iw=#Lmp
zoY$>vHd&ZGK6+IK?5`c+1LH>f(Fo!CR|<kSy03pGwyHfMPwREty3hK(r#G5(_25$#
zGC(-c`}?r@ffIWuRG1p!gyt`U57;Yu!nMKOhOwrPGTN`#Gdt;E5ECKj@Of3rnLS1=
z8vWM(n5*p8<vlm(MXVnvJh{kkPekADMhpl8sG{ddQ7!CvwTfMm^zx|HW~4084nc&c
zpWJVIvm3bR3Vg#TZt{IUk|8MbYK~XN0GtMdz|4xAw@i1Qt;i0h1)qgFd!c)(<2#G^
zP|`JjFpkZ`&=bVPsuh#V==!KnNF<tqL+W2g6ST%nl?x|-zUk}>4fUpupzh?Tc8`tk
zBHtl+((97VH<Cs(C%La!mt#^OoiCAS_b_NYe)4!jKm+NO+>p@;!vRvoY#p?ZuQ4<v
z*>@r>$<XVTVOFrNlYqjh*DsrS0}hD)s746a0n!)cG{e(_HYC@ZR0*ArI5Kdi0<>o)
zErydmO|gyQf8;aLS-cxGRt}mFKdYfnAJ3#uU)F6$b;p}dR3{VZ7ILeNpdyX%afo!|
zkK(ww!w4!(?nNC*SY<pn+)k;DK9j=YxS39@<K##`Pm~F@;Uf?huIoZscQ0Qz$R82#
zV*2JzRDY^)s=Emp=#Ah^mXO5s|8Pe_$&rVG+bX!(QSI;Cmw(=T$eQiF<#DV-LZ9Kg
zQZv00{}w=P<CdLo0<O)r(XZarp_Z0ECQ5N8$PKYnIFxz?sUF&)?2TD#GTlx!-3F-9
z25bn1e$}|#Ut5>Sl%6Mn2UJH&{5+R(KLi~aRqQzzw8uv>@!qD*uXUEmyJ=D!tl@!?
z2Lf|(--+wxA8*B%|Cx`}Fhm85x@V7lkd^G7M&@Az8&XLMWgFfP!(0z5|9d;t!C)_I
zy?-)39Y~9yv8o6vnTz0raYi7ggtVF}=GTP0$`33iWj!ykGWw&?F$JT9Z|!2jZ*9#7
z?){|j@8pM0F~u|u5et)%)ZI>)$aUUfb?ldF!ANgC!sjyRs*aA?nDU=FYaXNpibt3A
zP=NHX&=3HKhq@~9fe`{fhpd+%z{x~8+q-74;lnu-DxfxyDwHl-f-F7R8F(Y+jcA!Z
zx|1|;&=9((C3b!s;n<}9EJCL38XS`Qc!GXI;K~tb?D(wRSe$LYMx9!1|CX=0(G8xi
z$+Y{kqzi>ml*ZhH!b=chv*X`H^o85Qs%$P$2H$qV$;))gP;L1ss;A4g(jXzw(W^`m
zR^zDuTeQOYf?T&SWHCEW85PL&7m^!7BrS}gP^g##X1IiqU`x1D!<wmN*>8*Fg-PCJ
zJ6X#BA!~QwRGRzOHW&d{_8n>0q_zxUCmp5H?^4F~^vlNJXGjOt6d5#1o5OqK)Ru5%
zl`T`*41aK9r1LWY5-QHTu;_Py{fi%j6#(={TvuI#fFANG@{*tRZE2dF+?3e?GuzH<
zLxB4GaHz_$e@if*!(1>caO|$W5YZ3|9ak4>{q~ZH9p;aBlJiDCJlo2M$WXBNOb^NE
zb>|N+kXRjYAt|rVJeSSim6V&FnD9Lulnzy9*<d2D?8Dq3?ERFqgH<M7wD-o+xEE;Z
zfabO@2Zy1O?Z=`45EE8Lq<=)scxjNH91y3n7`<!}@qFN%@-wfX9qKD#w3O}E5$8{^
zil)r$uKFpy_XYTqRe0?0d@44!|CPdHV*2-8|6jN+w2qgt?Wczcx&A=qC=LbV(q&$t
zwF(S}^at6bU1vsx_%;8U-<<dH9$YM)cZtLmf8QNzO^^#RYG{`V5(pVU7=k&)BH<!z
zR#0d{fGW{^kX&Fis)0cX>%@UK)gSgFkB*RNe(u7(^A3KYE%qt$Wq>)I_UP9tm6u!&
z7&0a8QjtMkBh%BKkUbSt-Z9~*Uk40?3jqdGQV2;~#nc4nVA>izlAc*=>bcOutLN*|
zTD$oAg;^I(0YTmU(%wO27X_!U`CyYnyUiAVUkev~6f;2ABcUAITV*Z2iZz=?V)9T~
zWyS0IDkukAh&@e>OnotU71fN+lE&Jf%?mAC?fEONcipW81pc~}A=$%Ob-rVnHTB=F
zRL`JdAkub1M8{LGPlh^co!o2IJ!O0N9vnvYUR&eQh5~sNtTgZF(?OkU3mh?tJK+1i
zk$+py?e9e7zoSk6M~=+DFhMF)4AA}b2qC*XJ$u70qb#m%S_uk$lG!-*;eiCI0*Dr0
ze9h{hCAJ>j%L4Fl&BHVuw8iF3fLzLl0Ea+HC?ed?2n??vhcKZQzF5O+4t6TbOC$Af
z1bt~lQ=_kbg&?dH7MinXTYRDGZr+_Hcb2S|4>=4!&ZSFWA_ERd#JBN8UYIfsw&(yU
zhyTw0*#D-!3&X!>01^BP-D8p$SEHA=J7(+^00uCm0Nwq&mvH=D#4g)^Sowc1fYYV^
z9kI&})BU7&LNPN)t&3E&x;oIjz`B|Kp!l?S<sXkJR;81CQ26--4ONI2lhDYO%|i($
zfdInm1t1tG#2@HNce48QU9@PTg%UcAC`*KKWfO!5&L|;)LZ#0J-wRGdnDy)86E7Ad
zw<Ssv?^`CvqigjpECk<q6~*1Tc3r1CnMhm$>wbmm7m3Uc3amN-5z>GRV}wMO-xZS6
zwaxwX*^a(TMdnuHA8_bQmw|*?r{c3s2VYCOvYr|RhoW`c$4xi(8oQ73YLCGA<#w8l
zBRb}S-#X8X-w0T~#ohO(i`&&odRG&lUXX(=gPWivQ4+u9Eu5QeO$D?m+?vRMl=hK`
zu3DQ*rl`<~5d<zwTh>KhBJP!G(+Hn~7l-(towWDqvi1d<joBIvD)xMZh!i*z$dnM3
zu8+pcSnFUuE^Nwk<F|peyWNGWUo1yJBJHJwQ3JV8dMKYwY+cilaK4LdH;0Q`aa=rY
zOB#ZePPxlhL~k*D!|DFiq>{sCda7!=6o;XPTsWFUFo<K~SR!5k3bls;5QL)q^{C?W
zxsd&wf`>*$;uQMQ1<9tbgN1Tgzbw&|5s6AgN;k_8%MAjkbXCak_*H_6PHionj@#<$
zh{9uAuP*<L61lhX{@IBl2(b1kp^4ath%V4hyDV+*<l8-)svD<IayeaFS;Rv_;PrHE
z9=`5V1}iYLi)>qknJw^YFxF4WgsfDaqK+jZ;(>msYHk^b{>@$1Up=Q1+%i6h3AtB%
z<wtbaF4Fd>Y&r=_i{(Io-g*NC@YDFQ?$RDI5ejYV5V&p;bPzSQ5>7VVjzf>lo4);K
zS|=_wAQmOaozu>AHEEKqiMY026ild2il8HQ<w0-v4NuLh(rRyCVw2!yiSJJFDK>uM
zQ`w)wabxN5u&(#uo*>;_Q$E2>r}Z~(2s{COsiM9!bzcc)r#r&2-<VK^_p^$dZ>DLB
zO_dNLNTHyFYRmI%rC1kL=T_}akS}X7XlxR(7js07bgZ1!8IW?$T#z1QxKYrbGis)2
z*kVVJ!5%_GyTST;2uaA;gZlcgS|;K&P(fI)V4GB4z`@Rg6LYSka0t4(#qgWgpl#4W
z{@8tMdpL&FSv-T|*u!w5vWKN2R}lkhDit0~p=<_bFVl#5;J%0MD45ZL$aRQ$+*T7&
z*FH3yNdZ>7Q=0*Rn06jFksJs-q{GgQyaB?u@VE`GjT9XI{rqNiq-nh$r(VAF;?`P~
zi`~CJw_0BDNZC%>k$lbHsbLHPW@sFaZYW5ZkEA@YxcmI+FiDB@C-K4|T1qjuY_tzq
z1^T5eNOvV)WwzCp#1p7k7Y1_vpvz(GGHPS7#|3-q|N04wW4mTEm4QfuxajJ9^l&{h
zL`g;|m0}7&!GK~Xjo{fc(<j2!=YLA^G<3MA9zp@|9v~`YWQNl{Prd+~WE70d!G}m$
zOrkHf*~1}gcWG$<+HQh*!y)$(voN!Ar&USG%^;By%gul4?6Y>R!v|{;mK>IHoI56l
zr$lsk7j}8VE@V*ubc8AjJNx?>b1~B8*I+C3Qxu~1tg<jlFjrm-#gUu4PB>w3gNLow
zcDXlCrl#@xL^4Y$%^?*XAbC+tCj0Aa&3<g>sI#XasIj%@q#=i8Pz`j$?>T?0lQr`w
z3)+t~AdpDbi`Z3l7vH0prOpGpP{ncJU)l`9utG`e^e%r=t~V}`cc--eG=T}Jjk-am
z(kRDZ;D9NFn+K`A7b|OeLN#ve$hGY5F`(Yo)E}jgn?E00aFWd0iTK>_(;0ppnwwHH
zJk2h%;HHoy9B&Wf*ALz)B#>nBeHDL|@dL_5!;$^SCI;pNl<bT0y)O1gV%&CighhFj
zerv}4{jLm%>AK4)Y6ezO$$mA0XP1?2Ixs>6x&K0q)Rx24e-I{Q=A<80=f0YGZxSIu
zTe#uej>?>~vM_JY8W2?Oy(eUO=vc|X%i>+a!m1~iYBJ4^2+!|(v@cvves7XV5(L#W
zdElHy#TM6^2L%q-Mh(3a<OcJR{J0$IF_36?GeUL1exv+=;B>4PZ<t+(Zg~=Nn-gPE
zty)w6^Pr~x4)D<)&X6t+ILa~z7;>7<t5SFz1j`z&G;MYiwGL2od<K9DY8$74Xgf3C
zz#j()J}~~WLMa9r8-huz$&dj`56Btm^1=L_BMQsqQ{XfX=LVNBxI=GoDyl{3={PM0
zR6*~Lo^4r=`-3bli)z(NqdEDFQCe^4=^MOjf9v7l0V5pDSsx15bd~9sr9RXRkW*_i
z!)*FJuhOl-a^_~FFQEP&A;RArj0_zAHQMF>By7vV^q<1E3>=LAayj`;4NF_>5!8=q
zT?T%H-p}4u_YHScK>+gV1OOTYb=q0w8}JY<O5q~y<PPd2-)+ax`bko?Jqw9-Ye49n
z%`PU#>$Vyuq5bb&Jqx)o_;<K4vmkcS?eu*w6f`1?_)4KWfI_BmQeGUm!1GilNq`En
zbt6(Ac;}p^@DT8!Na3ihc;O(l#pLYTT^O4KzH1aBuyh^SEgK%_5OG)nnGP;KPF<zH
zU=9RwZDVX@&M*bW?_u`*#6zBFRCxlq-n~i`Q2~tR$<`o&=E)t%VtsIE@y<QS+6j=M
zS3z843=%^^bYM%HE_zBcG&EI8>MIs75CvQMFslI2ptX8PR>(;BP6349Cf;=g&KczN
z^O%7B%BdJEX13m2iS&dAJdh3&hpr^z3O=bIn5@J5$r2t0+@6$Np#1SuOxcJBVaudm
zkMUIwb_PZm93WZaz50YBvsnB`7@EMkoQd<pGSh1y$TwSU7kn7kohIK9dGHQ0aMW*C
zNsk#Y0eUs|rYdq3w04ETBSYby>Q?RKGXHLnpuRKQ4o6wc2{AAhusag~b9@VM)Q;~b
z2yEAki=Hh$#_SFQKpxr7Pk<b>+tXjs@^~pP7%K!9$jUJ)X6Q}_#&t>rK~;zBWW84T
zws|Md*rL-$lyrKCKX8de!--+bgUxVV!qPH`1_Fa3pt_sG+%R&6kVIozMl#}22PwkL
zfNcRsg!dlk{SY(;m=N;j1UQ5f!nRh%R#yw%e$d9(!$H2wqWFG2;zOW4Q%XjUExjHc
z4IFe_=b!}rASLW-0TsCw@>Z&$y)wC1T#0_BYiOoXOQz2p;oENE(`jsIM&ay9LpgR{
zifCX`$a19So#b2w$oyiJlgGF5c<vVf9%YrYqL~xpm}aZT7Rgt<ohZ-IO{!`Q!n2(b
zx$W+>9r~ql{nh$Iqwm}SXsDvOaxKJCM%7ud^GBvfS4QeovHGE_FAe&{>eFLa>jApf
zsMZaB?VMHcYN<FwMH?&*`hZ+7q0~+*k5h*r1iWRDAQ3K#v?N&J3uM=<h|fAXE*(Ld
zJdcMLBi_?WU<V)26Ftb&EUfQkVbOrYL9%jAxAcQ6@n<5sC~?&$?|f!gD6w{DSvgjP
zwv7kFc|(IWS{e%a`S#lTd#dr{jKr;Xmu6B$XX~r{yYjo%yIt8m@Hvw$rWFmcBvnL;
za@)Od`g#hZVq)+{X4FbBg-_-7G7y}{<AREI%Sv_9T9=7lvv-L(KY>$FPFY|_04vAo
z99a&!6O?{%fGvq@JF@7Mi>wRL2@0mUvRu}nzP&-z>V$Z92s@E86Ac(9jCU|^;MO>7
zH61tyq<C9Sb)YSQmTXDRC-WA(&o+FB&Qa*47YlX;W3D%+{Tm~ylgP=4G-zLZkbyyR
zU@SX&r*yamn3su0Waz<6n{Fou*8~@e?Kcfk9TB-_d+l@7qjqVb%A)po)*S!UuHE^W
z)I+B-DiK_4ln1_WD$ZEXDIXif;m^szAB~MLWD-hKMuKaxhJLyr=SVc!pBR1R)HQgP
zvm(838Kg2(@CK^r%Z<Wgy|hN0hcVxBA<ISNgoO(SZSrXYKVeRNBD>xYLO(eto0pSg
zy#K6eJ|NKs3pz$=NFm8fWMNt~0q#4Xb&Sw*>$x!uXI{n%d%(IL0G|Y^kM!#^cTC`n
zLpToq+UIqJDm=y_nNU1>pF1C(B({NL8eT6ZAniN^y|Gxu{pXjbi=$C4^Ohub|1Zwp
zggDw>ZJa$Vm)iFw1+Cd~nJ%hh9?sg-XYNWj3r?F!6($~CI<(^ko75HO)Xb?pEiG)@
z2fGh8D$SSp&$b#(2rHLO%m*-H`g(5H5SsAqJKZq2*9l{?$UYgGG1EY%PYu1N{u(>)
zK4`6>)OKh4z1fY_7J=2EzY_way^<}Zi=}8sO0H-kWzqu`0@SV5hEf?tL7;Wx{f7$(
z&c&dl$o$sMkd6A6+Q;jsoK$p@d);)<B9ZL*#2{#jV{;u<ab&y^Ip@Wm7bpA%O&XM;
zbmyY7+@@i~_hV)q%{z(3Px8;s@ey(-Pj5A{JnGpysUx{shX{4Tgt}%?NwABcuWIuo
zB#YhDWjhF7#6rFb&bZeSbN=w0PCa{EJVuU%IX$=hr0Kn7_Uf}*yo;p}#X^0H7<#D~
zsyA1g*#4b$u#L4uKiFcG)A>`YOB8gs|GoOVRDHy~@tb}LfEyRV1E=6j@Ky|gy9v!}
zZ609PZmu;Plm)0>)+b*jWp^mPCtr8*Ga<;-wN}o)bYF`RIH*kh`~#keD=4bOXWvZ&
zu3`te53Vlnz+ukah1qp5o(_7<j<W!+bLrfi6nwUH<7mZHv^UA;_U^13)bHC-mg1HM
z3!V<jaZmY}Vs|5~?TyLua~8Xdt!mq~LnjTIaZenQ17MyFV<LSBWJoR$A!InqKec%R
z8qRY*(YnDoKXAy|RYPZy-E6#8AwC_q?s&ZW8iCi)q<`gy^PuPHpMaV*dd>>75D0P!
zvX2c1f#vY>sbR$`&d6cJ$MBD3^ZL2Ka2Y9{wV6lVLL`Q09S#AjxqkuT!87Ln&B@Qm
z@i(<M7}&mp$N%l~-@C+98WP`et|;AqC9#?t5UvQ|c!+bZa@H|fGj|Pg%p%b+p-N~3
zUA%5T`4dnigrJn|lXa&>;%&cn!4A(c_Rn?9ov)r8{tR#~j89P(Z8#4b6SoVeN1wAo
z^u|Q?|4#S_9^8MbndbA#qTtPu5$FDNy4_tGGLi(v%k^zf>#}d@>3Y52FBw~2y{6lC
z_-#A^+AkQOJfcD=#oQlFG`EN>T_#wU5_9nwcqlm>`M|ka9p(9KmCz4ap{T^65i}Q5
zt4=(gmg;;ZP7w!ip7F$jAT>l3LZT3JhGe*LRuv~-SezI(-@3%9mnKYQlJ^;-<k2Cc
zDyyQ@Zn^iMZ8TO-{T^#>T+hu==1EnPDQA;`et$>1Y9zrRw>fgpsB0_JUdt)0BnUB2
zU@u8w5X3PSc}eyoCFY%X^<bUF0n2iJ4r4%!p&+9@)`TtE?@1!3N?OBj(l1*n3Io7c
zl3nX?XTjE3HEnpfv6S&10V+~4Om8>4&88peJ`CVG4MB>Sz#(T3DUystc?6diM`4I(
zjx6s7KU#}C*ZN37Q7f!Pf4pkpFG=?2dKII1pdhn&)oJ)KypliKjM0c8=x7gEY|rla
z%HTv~kMMU<>`P2axDB^89o**#Oc5aAu8Y~!>yS$nXjQW+Z7^*S{8UJHR<n*lu_p*T
zbIO7k458FWaUwFmLq1HDn8{`~Y=N^oth$5U=!IRD9zaHT5*8cGV;%jn=<k?1@3*0$
z#0sOyp&b0j@rlWt5QKo&4F+ACrvffWoH%cP3D(F}|3EpvjpVU6RCM3#33$(2m%2j+
zay@d$Rt%sIHY$OcZ-iX@kZG@5P+DM*zRn}tJP_c@_ZikZNzx~&d)A>iWE$Vor!eg5
zS#-w0qi}wAPv*RP(1DuNM%f|GAD5%75%bGnw=9lwuh;AoUm2Mv^}2IaEhNkAKsVpz
zZ}IZWbz{m!i8Le}Wsp89gzx*nd=iEflM{^fs6OE+8+%+wM}a5kL#*S%oCMM%i#2ga
z_AYk9_V&~bo!eSDsgX)VBWuV6Q?f_)1-;Gft!ER31BHTS>szn!&^>`O*l0*quUc?7
zIo6{4bEZ$wy)|w1L9FBMk3xnRJO_T1uKDFW)uCmRuwgjcwMuLIt$nHNlU-VJHfdzs
z@X#hwhFfb*?->bhCAz<}2K*!Ts)qNvGW}aknyf*B8XiH#n~z|!KbSMH_e5qu^zPRA
z@`BM-j<#pVq<gb<UbjZ_W{+0G`g}Wxsa&d4wwHsl&<Hgk?TIEiF2ip$hp{Ep8T^Vu
z^+RsDQeL}bk!r3z-<>u)VQNOi<JyL5zgewys@!L?kFWXMf1c&5$g#0TSp!+pQI|Ec
zQzFI3zFo3r!X~m)a{XC(y~6gi)7Wqv4{2IyU%ru6t(%wZ-o2ewNBr#1t_v_39GGo`
zjBzD&>z__9vz_Rr;%;+jW1S;}0qZLrPlZSs2VLfx2$6Q#T{(SK_^V{Gs<w0lTMc{!
z{zb4&SOh=yqD<IaE)~nJ@*PE>ebhb0hY6=v_rCfvu8)+L2o9Xw-DV!))oN)M1z9uh
ztpj<|fFVRxo+<p)k=?w(bq<8wV5qu~0K(7`HgAQLplnX5bsa2XQsti1X`AMyU%hL7
zDR>DYtNdoLFKH^nh7)~N<z7?ek&r0x<W;(K5JS$n<@dWpA0x~h%O;(CLUrMf0P0hL
zA#X6maot#xmvuZv#Fszd2k|~7woJWTN+A&Smli(gSUuOtzb_HLqxjK*weO7P+^jlW
zmIZE6Vt-uH*0}^f>FFEeAPk%9-nnv*1Of5E3+@e97F)nclJwEj#TS>ftU#NQS^<pZ
zGeLu<FBeDEh@Wb=wciij(PC$bf9#dnw7@m>X@M|@t&qWJxhv733*dI-SMjiTFkx@T
zXeQakU^JNerJb1N#CCfP%>nrYI7DI8w%0&9LZfTGIiY&veazt~z=J6QmQ@jz#Vu}}
zgA7N>AUGqyn~y0>2I;1rBUhjk3Y7{%c~KI6!d^>J8IQ67*)OHkKLIAP2yvs0tW5H2
zZnU@qNW-r4hI_IuduTKajDf+X0mOdwU2MW|ds?x8jw1L?BB~;zEC>74EF}RWo1uqR
z;^CAcNuO)@N?PGY#)@2!MXAG>gLBfS{e~AxStf@CCydM|tHrob5ED=x0v^qAkX&g|
zCI+m!fkgBUl8O;z22ZG#T3gw=5^97!wb#iXaSP@n+=;atSMVc82ndwL#5k8<vmfUT
z6yQcbV2@-d?iy(FAA=I*aZVch^BjT4tl<wR?+%x~VcSNK`6%z_@g?=3Hv6n=!`4;#
zw@2lFb&~%*C4`Zg<6m}ymo%gt_gRs;-_&l<+70vp5CF#3#tG9U858T3RtK9rM-TX=
z;-+9kEX2oYXKH4^^C;sPX}Dt7%DYwwAXr~Fbz9t=uy0S^8f>d)w!BIbC`{Yp$=Y?q
z!owRGqkiCO*A(a)#+w}XXn)-tarNELF^rKWx&sk?F0Q-kemQ+y&KyIRxl=Q9>6#7f
zSG1_v4*aIm)%|dK`5d{%En|{R#owIN$bs(M`A(mjVor@m!$hWvt!8W8?>Sx@H1<8F
z7;o)qup!)&X{dAVSq~~nuIi}wg)z0U(%269sMlGo8p>^l$DZPRC+8$kZ#*W7wYA32
z0hbgUeGUktpP&`T4#%nU(c?0f9}l{;e&#o!LdqgdN4SEXG9m)I^xnx>DC5SCSCKBZ
zU|*bv)bbKGWxsvHi4KBHhb8Box+_r68Aqf8nm2KOU&<<s6s8;|+!qv22_x}(kehf;
znhxUeHQVC~u$N$+wZqL8&9F@AOYefs`bZ-BX?W}Jxaz`lx#ohbYV7S~bMVuqW8>Ky
zcH3;?t$;67YGXZ;qYf-K&U`#dQOSi@-0&!gOJU_PDlOzx86JFVa?>(xa#Ba*3eW9I
z@99`M#qPO+!=U~mU@@0wk0>%TlHk?Qyg~)y$2^L7ERbK3mN1oET1o;?9>sbbsJOp+
zJWzhuRbCj%e0XmrnCpgcYoM_}%96m^KpRL-1^Kl1pUcP9k_La=cubR^s3eK}RAx<^
zX`hd1u^*l9fG=c#b>O~Ezy)P4I;tW)Ssz>nL&Q0Zv+{9M5Fv%^c881WGwn19etVQ|
z%tY`dJ)F7+8c4nz<Pjji!|dUOcpA2%9LV`KxllBTAo1xvq9;*Az9bKbYT+F+GQnFN
zDqJI5f<@AO>1cLdX*bEkmdDX0K`}{3)Dwf|`-iWj>yV&){CAQ{w^ZxqHpR`EQa$YC
z4GJ;)Ub6F9Wzs#<jhNd?Z!?e!-IWPA0cQ)vD$*PaG0qN*h6mKh$l0}a5GtO^1SARJ
zlinv8SqEK>LoS=FErib#?JX_Vb8MHBr2K%Zgy|Tws@$R7Wf?kei<1kjR9>G)WLFpU
zbeS2=*1HZ7y6C&!U$fNh-6y95Ds3(=R=D$6^PU>YyLlgtKj4yH$=<Is&bEMNNpR?k
z=`2#ggTpKCh!HoK{S%pO;^5Zf;N!a&3Ba6EN}>jEbb?&*P|CPz*?ua=jkja%(_PW_
zM`|C7^hQuA932FJc@uRP)xc{?tCZSlyxXbKUZ3#9Q}kzgv<OUVduR+Sx2_|qyM^FB
zV=-m*QAR<)B+3H#MVvL81NN~3yHLH-wbx!XyJ)nxNBp4b*BvTf`QsuInt9`kUD!&e
z2C}1<nHllBk8HF0Ec~L~d|*$zj<TY=Y@#4X6^A}kvRD}_?GmqJ%cgzD9Pr?_%b1H5
zA~TwIIRa8f<(e^#g&|O*4yyt#7JSIllQb~-=4@1Q1=EsRpqY}_V_|Qq{31nQBc~f|
z|EI?(rt41t0!Qi}eT*WHP3!#o)8%hK6+x!;UEmYswc!XH^=iCDCUPT%{OXS0gZ`0l
zgcPjFD;0#~rrDfu&z*HG#+!4m3^*E1L%_RGcv)z_pM<Sk(ea+_E>-}I;9BG3Tfo?l
z?<eu@!DG4ELGpdz?B>z1)%_5ln47T)#^UE4gfN|>3dt=bH4+L3{6i?fTw9)u8;DDA
z*r=U|Gdt|#fzk}aCj%wR<(vUaecfG31tC=kSwOuuXhN0L!s_luj<ihdJ67Mu5N}eX
zkIyIra^qCGm-3msnt23i_B`c1SUi)t?vO@Zzojws-<T`lef$(BXVbv=+_pmp!IhTv
zRg`NTqXl<(=%TM9EV}Fj&~>hxD|WTfa7evy;bMOR#+9q-*KMgRnPZ(g5@=+=h-7uT
ztmmx?hh#j0_|t*N7SI@$ai2n-bcygP7G_wtlZ5IgE@YFYC7w!+isp5pATL^&%0sV&
zM2eH9&z+%|NvNN7q#~O6gGbRrD@Y<lCe#*X?L%855WoT<!F{nIqU*ufz5Gb?1>F<$
zBn>vz{Y9Q7q9w-ByYsr}ygmx5Dbw*xJh3la@R@aZd<@vzZIUa~o#iav=TPSg$Ax}>
zYTxPEWU3*H*$3dkNs*4nwKBL<yWa;n<TWq>J9`wS%&!+I{pMorPjoD*_Eo|#Fkp9d
zJrXtQI1FC$hCehpG1<duA`G`Fi}B{D9Rc2Yny(y&ViT`)_i+@zbLsMvhZDxZU@-L;
zS;Xma>V2HKZ}twwj;I_TXFvoVIPmA(caea!Z<p9P#QhMmT5F-GG3(IsY}jlX{&20k
zw=L;LC0lL`P6-E56K2G{ZB?bzvjP$!tJfWC@XqaIFe~mue(-RVewaXtqMZssa;~_l
zp9rA!;U?s&7>yJ+KgnGNGTIE?!ei#}^d*fx{F<M98)UV)`N_G%Go)*;xUFRXkJq09
zcz=3>*?MSS=v07PmCXsDwaI>S;lj(HhMp`=1Vj@<26nbg;9TpKlf;-rdhn5%lDvUk
z8u&Oy>1xOkJ!4;zsSV*2kY?~ASWa{itgj~x#~Y3qZ{y9p1)U6v$iB^nQ9(SXab*Rw
zX(e)TuzP8!*?OQXMRu<Fsn^Ia$_yHNyQ|X6b{B7%9DE4MV(KA@jY|nh+Yqlf0>m$2
z0XH*ffNO(~u(N59TR^S$nBWrf6Jc4rcT9IjuABZ)AjUq}L-4uT!sZ=EAdF;^RruXb
zj5ssV>c|W*!E`mNxJ}`lpjX7^!R*2N*;+gbZ27Hw?Ycw#M|lhxHvajMK1LFb5&=&K
zh`LYnwEGvTj@FvPy}hfNMuub1G*D!&HOApAOM)YdQRD}LF~j~&Sw#wc${cfdzC<K_
zcR9q<xrlC3a_&UyRC_5N7ail)3kX>;6TB6CJi<KnR%jc{R@5=$CoHU9)|^oQeCm>o
z2#fPZ18_{M&8nliv@Z%onOCs}J~$qpZHBfy8vd~GuZEGwWPI4kBa>7dvpHBga|{pT
zx8_w=6^UriI;}Npz9fc@Y&tA25|T3!$-}*mR~49?NVidTn{$1;7ka7M%Qf2Yz^-7g
zq_7+8q7uH(gZx#M!e8x>zk1fzmH_C)Zy$O}@k1l1MHk3W&`HgCo6Y|$&aF1z{5DbY
z!HWv1iqp_qj!exj!EM*OCLAAZX>OiQ?XM-DG=Cb*aalL_g3r%hN>C8RAEMP{A`GeS
z2{DdHRtpgwr|4{k31;@{DYI68!GF8)T}PYf)|$7G4|iz}aedtN@05}&021!vYv<-e
z*lr4PEJkGdob!B!AAAP~mCX(L8%)UbopkWO7DH$LkHCfho<3>%#)Q~x|Aq<0Rl-Mi
zSYp=_*$!8>>RErr*o@(cl90%9!)GuhRldJ|_<<c25hfRQmh@pqi#{#u8Wt=KP-%Ag
z-p{PF@x2fJvP!O_M!P-x<3qZitjZ9Xq_cUY+Ed+T5}<AEgj2=0@$S}y$f?=$@$h*(
zgKx{XA^*symag9O@&4y%^Ri>6N9*J1k1pSP;ene>kw&4Rm4p;oI_?A6onj?>B#7i(
zM8~E3c<Z=o3+gc1<&vmH-4)jBQK-<%MdU*p!{%*7$@$T2lD<>-W`<De@=u}1W~NX}
zltinKMmC$)>(f8ocKnw0Hdl8Y1=O59B93y~bi+6iuli{~w}h4G8Vw|SKXA=c1ii@r
z5QZ%tkx~j5S_~1BDNR>NqL?az1;3-6j>eA9$8Vt^heHL%_ynO;1QnSdiS9V4CJb@I
zL5nyDd_|b2SF`@|Up{VL(C?^`TBl4YW5S-E8QD{pZXP6j39w>_d?V&fTf@1Zz8I~(
z$~j%fy>F*!FDTo@T`BgQw{2==Q$spRdB?q!6<&xE>*AfZpw6+hst%*xC05W`0F++A
zS`L|r3YQiLMvNq`87MDJmCNFvS+zhGFks<(oJ6U#j2^as4h}fXv<)390!>jpO)ui8
zH0*S#9CEvakNi<AJ-F)Hmi7GQ#PRz??3V%Avzc<En_E<@u`SK9T#gTV3Ha6yS)0$&
zW#W<v^UcipY4CdPs$JjZq0vO1N=tOsI$|i-grkkWGch$^MjwP5`L5t81QYx2l^8&?
zIB8$I_z+5%c4{=@pPy{Fu=#rWE_NXC;q)r6Wk+>pvOeR4d`HP`d*^TrgH7B=hhtI8
zsFvtr(yslrhL?FT*-umU2v$hkb^z&|=#Jo}Q!OzBHm;MorAf<p2y)hbqS$bkEv5dE
z?i%m<UT<6IA<hxg22{O?@exeCiuNFR{wgK`4{yz0W-%fyHD(r`<>l|sopvnp?wmnA
z7?$M_g#rtNHrXih1EHN%<8ZCn%guy3+sw|#Z$6W;#zIvuW<MVqwS6u$sgASp_d4KA
ze)hfwDS@nY0eRJ&mg?;{Ulu_J0<Dn_;9vRK=c_R$QbPj1pC|CZxs(uv;v`aj0r2?s
zq5LUL*ZKnb)aMXG*;mc7Hwx@BinsvPwUlG;XrNHGADO;lUHDZnl8*2jLrj91nEfgP
zst|%34z6B-ssSD0PdEI+9q6|IIwL8_*q1yv){hR=^EFLd=VK$mr7~n)M$#X)N-e*&
znYE*`^|R{6otKo0uVhhK_%f`IL1woEW`7?o<&y}d^fjy0YPD_U#2g4ZACu%{XcATK
zU=s9kNzaywOGwAR6wS`+eif}Bt0$X|IW5(sTIc-OmG@vA$%bFgh2!$&LfBnIG8LGM
zOs<kj#OL<E)dua{11xtg?++{Kw{VwB)}FJ0jA-A3KOU1}D>F<n+WKlQU`&mvWGY@*
z#c^ES*#{v}lPPKKNJP?eB>EX2_glfFQDc0Q7Hm9k0K+eyY=M+dJ0*ae^&&DIW8HfH
zp@BGp77}lBxy}=XjNfQzgJ;=nt(moHuFMewEB#5dL~KnZqHJ*_&Ang0$~DVQow&K=
zi5FRvz@Cy~bVZZi^O~3I*~8M-g|TGZHAHe$lUrC_&HO|vL=?Ng8cnqjC2AzRC5({v
zIS(lAa?$`MmN;ebR=0rCmqkPU`<sj_{7Z-NfyV(^ux_*{ph4ZT8`XNbgiN;1Dq*KR
zZ-Qfj`I+JOrV#>a9Pm{jwR{xvrG(D=c}So&oVzP8?9BZMYdo&76j9x_UF|aUFB+S`
z!G+eS5QJ0(82gt)pSMm}a&m4)x&yn%67RZAr8>^0oRE!|zOcFFh6`Pe_iTjt?Vwpl
zZL?AXKxl~qrZ8}kipmzO1~YGoEa|&2eYpxBd6!yM+O&hSdMo+9x)oY7(VM~Dmwgif
zC;*29SjC4%TdJ5NJ?u3pVmc{g`qgUTTP{6DO6o6zGZD9tY_b8S-jbokiG+7E(J^8c
zz1X-ciQU_mUtPq(W?qvm+I1|;%FfJwonS1Ewzf<2+1C1JG%GO2vWR+2t}-np3Nb&+
z2Jf!Ri^yDA*HO)?bwBj9AVlM9UAPD3cn}zB`|q7dxMyZ}BoN1(;jw^Sj7o-bde~Q)
z1z512h6c8HrB|DOw;o5(Z}l~OT*+9P1v%yg;f$)veHsf5DVYjx_e2}EoYaX&4aWYg
zZO{4D1@IBWG$YK*OyZ2X(J@1?m0555a8B8=Wfyr{OD3mOT+ljGNlOn^ls=U{f;^27
z;R8&Q7IG|e6vt}cBO|1yd^Ej|A25e-WH1;dRE8P)ZL$T_We0g9(fnz=xRfRZU2HHt
z+5vt)`M?cv?}nR8K*somog?na^61lBP<nFF@Xii0{LWlgVpbV4aYoe|!wS=l6=jxL
zOiRt;f{DX!oLIv#K+FKm3c0clNmO)Fmi=j&Nh<j4Pl|`}$PXfwAY*Z=6Y)+4dpF5j
zkdj!NOEF#)k#bx{(7`wZ$GGrCz+;qH^NVQ8l~#qi9pbLt=LzYFI}Q*r{TG@=JGq(?
zF9fUbv5sYl6Hr~+C!|KUO=5<2aj1SXVX+)~&aGi5M>=jMGC0d2uH}#*STe_b`&xlg
zT82a=)6J7v$*`whoZqBmB@2Pf_xc;T3rnO~5u`_`;53+SeoFhgN@zXq?rs{jn3yX2
z6)<G{#?l7+UTy4W_>+{WzKDxy+j$%3gOPWSqXJU|nU*NUBM3tzI^J0m=wl@*78nz6
zwCMq}J~uv9x6@fmwH_hdvE=qBce>ReP6YrxXfKhYi=4HufTqwGZ+<xZI>@^0)5@XC
zGN7oN;6<D_UDc=J+f!Lqa(v$t>S+;fPfxuTVEQrH>)MFbRYbZ=0~=i6BdmUxd~a1E
z#uV?LF)^0GCe_Jf$Y3=;e{1wP45dPzC%Um<e=z<bSE$A)5|pYe!{(|MFK&wVjDr=*
zO@TT46v2(bGD3mGl7%e*!1<&9#QHuBANu_557>ek#z}p5s-z6klaJ&d%iJ^CqN4{i
zmYh3H6SyJLQ=<^mP4+-wYp))Fs!|j^-nj#^a2Y25%(i{i(!*)6zM2;0r<uH@T+QV%
zcE4mN$RF0ShcPJ)&zCI6=+nnXSb>~Sxe0&ry~jd=0)TI&NSHy3t~4y%;5J#801wVZ
zfY7?HZMNUxfduGt{}#mkO+G0jJ=4Ej$Tj${u+V=i_je;cqYi=R<#+iI*KU6yyJ2qW
z>|y4aCoaARBT`O+I`s0Wtp^C9O!7p6pnl=mqa7xi877Kv>!Y>R=_xL(97nXT(WoUx
z6gf{q%q;QC821B}Ny7F^<MaK>e~yS?KM7d}LeHS|3V-{n{r%?cGN6RngtAS$5HmXW
zI``WC{P5$xCO60DrBq{6{>S^KP#;hV^oV1hBT*y_2cL7|8qDG118#jrMz&zIiHDNn
zDA3A$m<SOmX02=)0a92B*G|I*mACGoF$?avYE5FR{iLl%)I0KlEk)*>CaoJmm<A!W
zl!U;G);M`*A2jC1s6>MJaoN{G_F)2j%FODP#*)IoV}3;zpYTS_Z41|hss%q(3!AJ3
zpCCPm)PCyLI;blAdA%aBk6hR9F|&AfcoPLh`h+_mRJU|VV%lgimfeEJPpYT4-ITG`
zulb8{iCulBSWP71%X&5rDgO8L#nXtItdy()biNK{7nU=6>ncWPCe=Ix#EP<)GiO24
z(3v!7^LI9xvOlRTN+pA8pqUqn;^|Mw`rOe8odj7$rM*QE<Rmi=sX$}|D2(f55312n
zsVW@hx6`)!dnoD_z)}pgt=z2Ekrl~Ze!J{WP388#u3}YH-8qVs1RT=EpaFZd3Q}m3
z`_`k72r(B56AI0bU|#wdCxmpO<im)A5{q%hK8v|KNLa~1pRMAN1sn^R0efC*J71oL
z$-zwJc}0>)_;V&o*EZiD;iJEKPRbty)M7&Iz~mDF1Le^=Ha<dTnPM{cl8%U^f(g{9
z4FOr0fY0QldN}cn_pAr#dl`w!z^Z063`322poPL20BNWshM+mC8!!Q$g2Wy=%!@(_
z7zg%H``(mk^#Gv`=)Wq$lskjPO<yb^ru(qG#z-)z02?tCyZNMf{P>mf1Ku1cTrC{N
zxNzHx*R|rnD23!uQ6VWIW{!IYn6Waq9hRuH0B!Q&vlr{Fb#F_z1Fth`&Z^c9vGk2O
zkxM?sxZqBw2+DVHl>>+J1e=;THM1wm6+LV76eIf#h+FfSLE9A)%bc!g3gE(<28nLO
z;OJ91VF>|Yrn>=)-9<t2<}bGHQ75u+DD#Jh-lw5<#L2A^3|*z`pVi?y2JoJS9a`8j
z27G`OSko<cpy3#n9t{r7qJHU>_*{->$EJ{#O?*702o(59@;wcfOrYKN+EXsJo#zh+
zOg05m2PPEnp+21r<S{K+VOLzByMQu-#>g6#IR~A>AHC6gz^Mvixqj@$4mn}8Gfk5X
z_f;kg$Q2lx{7gf8oxq<!pU^IG>o^THS-ZNqM|=e+ob^!S)N<{;Q)u#0YVc>2Va}9d
zsr+faLrn{bip27K&pKm(ZjmyA|Hr#&33RmnIe!%gOoVpm$IfWNAdEg!^R;sxZf{+0
zK0&`J+QpNKGe%`=@xZ-#B#+kaF?%vLhMxL6t`|B-3eE+aHWLe&!tPZ&4msO8io4#A
zs8SBOi(UjOWBrwA(33&(OM&$iDd+(Uvqg5>@w16;kl=PK>v?5s-^S%?2uxr~CI;HP
zBI3MGM^5(!C%{lb5Mikm(24}^Myhz#Eo74r4&M`SEpFcJZNj<o+~(F$OxT~Icb(l;
z(RLVXg_JOptC8TbB$ni(W+&qU;1oO>K)Av3A0YGq9Zi1M!%XUc&D*<ot=8OL9)jnt
ztWQkPP66;0$d&TsJG12uCsxlZ;khnqOP03Yw4tK=99OPTAt!Z!d|a{>YPPcUQjQpB
zj$<>E{=eColheW6-wyy#Jb~c6b&yB@=sU1Vi}zbqG_Y~6t5(qrtt`60D=q#Ul;IV?
z1mCK~9O*N5(fmD7dbpWeJ<|&<J;9h@gw`5j(vUrR5pNp85#R5Wn=S<kF8&)`BLzWJ
ziD$OUbM5w|1TTcu<59buhe|dLauf_95oyc{^{Y+gH{!jbV<2#>N(<m<sT`tpWv0vF
z4_Z~5T7Xg2-`Hs8-cDUl-77nG)N@KM4vE>OvP0^I{v9K~$HHO2u9+ueF!9FS;LGxI
zOIdudBALMixzcei(lPpWMY0x~`jvP-i36K?q7pGB)GyLJvbys4ca&1MH1c>Sm!wUq
z-Ln}d;Mhlo0m6MO0y+(Si=5q3qR>nLa&MjUTnXvkdEn$6TwH8J01z_2RXN4(A?^Sc
z+k6FW<6ck5m(MMr<vT`Sv^xC@T9(iP5~z{Y7Mo|D#Ec`C_I&8Sg>b?rMRv$e!ZwcZ
zsHw5~q0|rRG&B7>YmH%hd}Q%2KbxJnCyl_QcENF0tl!3&6XZvWFV$=sP66ay$+WL%
zAIpdW)d~Y)43}EfmKMEKFVCL{lgLHhA$}QfJ!%_baa(aeZ5qv7P2^wNhc}$~EtuK^
zIXw>HtXyiQ^IFt+Vir5}JXT0U$XRKuC4|VDScrd3a~UvrUY<Bw#oaA()49aka<qqF
zIZT3nS`u7*sb@@D8YwQV-B+P<5C`i4wnFfO^O2sy|0>gYCKU3`(-*(^aUwi@ifU7u
zs$<^uP*Wl_A()avd{p$|cU(1*-W#RN_v`XN7trnTrgK3KnhP}Pvhfl=v+DYStOkv<
z2Fp5Sz&p;}&M|Yy`^esRd6{XiZs`MNI|&6)@kw!@eOQg=-nMxsm+2!m#9FExJCL>l
z{KaJ9QkkvIm50diCe|-h`<r{=KRb69g2g)b6%B~v|63ICUn6P$M+~jHoPmY0p%WCH
zlCy!6yPYusorJZ%sWB9tnz@ma836+`8wV7fxUspZnG*pMJ1Z2Opt+Nyys?9jt+k!4
zjj@dr0Xr0(kgb)igOZ)T;rHXh#xCZD#)@Ks-%ZNsJ6aMjFfsfk$EWz8QMnxK|B(~)
z-;+<5o9<EBD!X5|%|et;zfdJ#?gG*LHLa2n#3}PstcVbb2=aEosQeQ!5TK5vnEfHb
zlL+MJ3n(HfB2{{<0nP<ipn`?~pnwSF(TJrmlSx<R6(w(<g2*>uOw%SVAF@1;rro<f
z3D{lJkGI}#Zns>E7rs8x0sde7`1M`nhvL2B7hthiEY`YO)eWIpKf554%i{d0u4>ux
zkmoIm250wqcOzFP$Kv&J`+$nU<@((Kc**ARaPr#Ihe#~y#tJ|o&w)=--FAeP{A&u&
zZO4Dpb=&)BGCh|l0oP?kf3Z}7SS%)s-QEDZd(-FblI=`;Ns5CacmDOI@{_~+)#<zK
z^3RzHUJ=(-WAquY#_d9h#52Nddi?-XU_YSb8~=h=J#}xq&a?6g<rvnE>+ZJ@7)*yg
zP;@sc1CGz<{fZK&i<LTXTqpVO7gW@AYkGHRG?=J-X?1(O<atSk!r`{h@V+U-+nvrA
z$=1p;8;>WIDV3<!?6e){oUi)*;d(hptoQ;TmrlDmmGfO|0Oa?VeCD@5uWC_+znPb!
zdvs_?FGb2OH2{VTql!x{1>)ZoMx94gDL_`SA^=nXy+cv~pvWo}<SdYMDu{q3At>k9
zDS!e9Mph9QQx2C^uJZ>GMje$>2F$Z7fI1DMnhU35jq@u7au25hK#^4@#8n{WR2Tw7
zLQo-~Qvmhr52q?5E*B;*U*QKTfI1*82aso12zC-qJ`qO67Ux&??-WJ_fFduCi}~kI
z{WDVk*wjC(@1HH<pUv~1W9FX&`5%<<{{f!DMO0z!%x&5B@zle!{-IDbM!dEHp2=L!
z%$L^OMH^@iOx*o|E39lEK~(OC%2G&gRv`JcMOUN)pdg}rI<YqtITDM{chDsog;DqQ
z`Q~b-Y<?H^daJh}l6(ZdFuiO%E1cX)*Hs#zgrWe-d7@oG7~F@b^7`ac1`AD49!51)
zrCwK1QBlGBbTpOubd+few<2VH8~erVE{y%Cr}TB3``t{=vT@u;qfTY2j&dJmL@5yN
zQ}t4uV2N0=jC)U#6QD4vLQ(RJ@#ATW)oQ)JS60=RY%*DF=_B}h$adNMyWZMOJ4`np
z5awYKx{6)IJdpw@*kXsO5Y%Pr9<az?o%so%K$;2{LB(V;St{4lan)`~fNy&K*H~D5
z^feRyH5RO$w1ITx!tSSB=UqjmdnlyD|JzpHt~zYJUq4=EeDOv{Rz3b25KZ4&>+in8
z7sJ*}>_k#2=)A%am9vYCNh$jSKK|90x1wkMxt=GfQ#5|%F_FJe_GuXGLwHlRFlq_K
zH?XG9+iJp6G~@gCC6OGn!k??YF7Z7r3Eu!7`0TH7{qr9H{hvqwyJ0E`EEtc%WI9{8
zuI{+P`hPY{MS%s9k8PE_A1O3HS>s|G9!uvB5Su>r))7=P@TIYJQv1+V3OXOMAKgU0
z=MeKh=dcF_pW@s8DK~+_sM?$)mEGTOM?zt-RdanxvK+?SozKB8@|!VKGPnz(J!;E-
zN!|yV&zv)k7D)cz&A)#qjOxSlKg^%5=XI%O_}iI1=n8?y|16X9E*JVGnJBgx3!X{)
zm;JhQ(*}~24|<=9pk@_>QOycmYr;~(#FWl|8=bZKUyB)&S@Ub3PoZ%)_6w#MmIV6W
z9YHNf{BJ=Ot$*CLe;nI?D1d*+l7ASSf4HuHSiyg&%6~}R|KHNz<NOUiAFnKyzhPob
zb^d3<ew=?{b5pyEiQ~MoLC52?NY?jA<bP)3VVE~JH-)ws91cd2<+yLtO><5+{<fts
zs;27d>R&OOXa3;dAy$9)8)V_+kktm$ue;Qq_isIoC4m_)T$S#dV7f_?gfnf+?6u}K
ziwah+j;=pz6|XLa!%=0QFLSXzPTw~*OG4N@9!`%lXAN!0mB`T;3<i-j&80ehxUL&>
ze_h$w?+u>u^-i&_SHAK=&wv4dQtbi{*>qg0^7O*{J`dp9H+o*R9mf@mMq@P%z;St9
z7~-#cUU!{m{$K39Q+Q=j(=Hkt9e0Nv+qRu_Y}>YN+vwP~ZQHhO?A3ek=Ujak-+ytg
z&RI_`vgXPh8C9d|t(sM}s<6EtW<eRtHhjK60ht_Z9?ut9f>Sq=hF4B^iCY+uFDx)$
z{qJg9K*D6J6y>62KUz?IesPf*kRpOcA;Gro0S4S$p<26lFU0Ki_F%P556Hj(NB~9T
zvR!wXOlP+E-)W|nZSQyXUXIy+*&h|#Y3}C-OTKrnkA1xAZ>$ZY6fG_+Ak>mfL{!!-
zR|8SNbM96X;D-HZ&a1ZbDveI}M&1z!KxPcJ%1!YXi{;8h8Y3X-bJ6|vSZ&k&*;EV*
z_$-$t)mx4l@C)BMTv50DuYZ@UFy)L6M&d}bUH2HLx%SzC0T*BR0OyH9BE|N9&YT#)
zE5JPwOVv8v#*>)Spn!M(gO?rmO%JH(^KKx@AF^XN)h0`|&l1{}9S{?6z#=z(*<Mep
zk~FOc6O7Y<9NAv1%Vak9d&m7$``cOhxEt=#bdC=1Qvf2*irT-eHC(-Ke|r;_aRG8x
zURsyeKHu+vCm@bnU>06`xql0u<V4W4l^`i;dt5f|%&~5J-UH5K-R(H5TvEsLb`Zf`
zlN$H+?r{A38AnCituW!=qNbbEK414Dif-?h%>j}s)!Lga4fS~v$GGf(4`-0PZzFUT
z4)soE{&v!$DFH44ZK>$U9e%xE>(m*JGB0b+i=m|Gc)!NGe11K5`)qr7oRt<=0R%Kb
zhUb-1=XET``yu~uJf-__NrSuk=3!oj)~y<l#%u2~y<BTtY-R)G@DSM#VmYrn>YuZC
zpH>WD(W%%D6BW813TeF_%XPT#Cm3y7c7wNgwdTj-SlJ_=7dJOQLwG-SGaaT%ahEq*
z9ca5=_RgUGZT&rc^7<8!)a2NYWbFDlq{Z{NsEt6Uc7I${$@2S0&)>Ij3ers5zEpYx
zkr>=fF_e6Oz~5JqD-?k}-w$Egc)qW1S5-Op07Gg5^gTt>ezZcZ{_J~|5og17KWb5>
zNIGMwW&MqHV9Rb8XWgn<Yd?bb^Hc_lRjt{26QKL8=T+^>I@f1sYXHIf!+F^th{)}@
zQlq!1Vmy(0yAyzDeG&KU^t00naP8~kHKTjKrXQjGZj2Hj@*8Iv%B-t;Avo4uuSYv?
zUYqx0NtB3OH$C{*%x-rF@7LYi;!pr0vdG5pzTKWyO`g{*s?RvKKaN<jcsc3-FxHRA
z+wrz1%5xV7IMSS6X$0<v84-t;*Q1PovT(NDylR}WC0E$q-u|_R>v19X>9K##x?wB$
z`3QiC+awk3)9tutuE?#FD$UkAl$;kK9EW5Yt+f~T$Dc9WPb+WY?^7;a^Y6t*{vO^h
zyCKb+&g*vjLs^b<;wqYUgTfeEirmR>bAYSVFM7Cuo#qk>g;6*hO_WyEEGlbMs?`@A
zFXsSo=FGZ0;vaK%NQ$%F&Apldp`=tMv$+npdu`X<ptEyPJTFHbIU|6m_GT^c!YRD=
zB6`ns6WU-8y^ACf#&FejIrjR4_X^u|J}#N?zOTN`0W!aBqukS(><-6SE>+on**>lE
zk3!N}9F1LlfUsLu$iKt4W8p5_a6hknL_3<wT7RDdu)My_6aWCdh~=38O^ovB+6$+|
z)ZzX9xDP0~`=?*uo%wGmJ5$~7o7YRvwg+*dPx~VtxFXjj_QN<X3m8^_4cK(tid?U6
zIj>tiqLmq(f4(bWK5KG6omY2vJx^jm`n;VMb0PvXEqCA2U2zqFsTm(VL&v<|uUm%&
z3_f2UZ|C*KlV`7w|Fj4oy8pyZo&5!trX{y4;jI1P|Ng%^0H7^kOlGiFS7)$#uN`bS
zOj~_sTP#YSQDWN-AWaDZ2s~lS_^Gs1sj<AcsHC&ov~GRw6$HTYimrzR1)0m&M?_16
zO4SxyJulLn_l;wgr;!Z2p|{8L5JWs}9mjF%<_D*Hz!b{G<x*VZi06$O)7iW?7(iw{
z_WlQ8agHexktD#uG}r~f(stZ@yQi@`7<--84&$!h2Y*IWSS(j*)#?udz{>Ty^Eumo
zj65uF3}6vwTihNm-`ZaN!C}wNr7~D52R*D?UiKn{(f;jN6R_iCDz&+mJ|t1>+oUdP
z^@fsTcvhCO4uDPokRX-CYHh~F5+G^+!j0eAhp2LgF5T~1`$xSt+m~j5r2eDRzi6sn
ze{ht$FCg&_1mxHt9Z3<Dp`!jFOx73`PO%I?!(<I>H7&Qf0&!Rj8qG#?rO!u4pO53a
zaazD34Wer>vDXY;msNuKRMwY(0G0??2ha+LOjg?k9cOz~JzlT7E57dIeXD-ECx|DI
zIJvkycaUi*s&#p~=`<D;TU50cbxpO{?Xx@qhf@IVpP6N$FRM^*ENj#A3R<`6?)={?
ztODvE&Mjy(ny6a--P~$#w14CSz#9N_*pD+6bW8wc6Iky+|A#TAllfx+5&H5o(=X0)
zE~};kNXk;TO@v(GEv!L#zXt%(0wgnpl9Z!6UiARcxuP!9(RgC7%@2+3H)XFZgJ^bR
z1H9b-O$|=vF|F{a*Qo|TUHr`UKJKThtLXq%cqgTiWmaGD^|9JVOQ*8D+yVf|y@gRF
z_5kz)h#f#USi5ec-tqvb^5JFk{&rp+gT-3aHRZId)m)2MgQSGOb;Eaz|BvOFIUfh4
zli%qXxBJ7(DLkJK^Vc4T=gZYOJYMczS>rVA6@%9vL%&QGSkt&YoD<4F4{3eOu(zaH
zH&Hd$LMi?QfqDNCQlZ&u|8O$9D4b!{wEAwGMp@laH!Aqj1L?ZucD$&P%50JRe}(a%
z(<0IcKno0ao!yfB=rY<)K#9m(Tf#qf1486&Q#ZT)DFE<#hdE!W^Z;nCyZ7M$icH0|
zGXTm=?Ohk)vAPwE&K63!eK4-V(I}0lvpWIErG#NsKgsgRd3O{t`;tVi^Rb5!fY;~t
z^?FPyldbs;HTe%?Yo9gQ0W_XdO6Bluy8Xjs9K*-6MS#US02s#YPznrfS3Q7`WY~1A
z-kZyCKQ4^K5uV%H0Tja;0Q>;2XAvX7qa7ey&YMolLwG~4tJB_1p09U+6ZhWj3vxJ~
z`kV>jaa@oSU0nY7XP;LLVlKPhuO5@iIlBSe(s4UX>iRmF$@zLH9zwEiTvQ&$c9?kX
zdhTR*I8q$0cfH-i`+gaI6kcR?za7Nb2G}n2n2x0WlCMv|@#E3!3poXy$#s0aoj)#r
z-^Td-)62KLgV6+!Z}%<4j$@^AWp03B2DlL_UO{JH*8rD<WEwyw5S^{2&*!7y+swNS
zq&L9Q1GuK8!{L0fGReCAQBgm2=9N>fQq6N(kU%1xzHq<AvTj)qsJHR{JlE+4ko)z%
z#cQ>&{{HovKL$AbhWjZ_)+&I*Rs+~J7XWn^g#j?~c|gj$+H$hxacy-tl7T(6-1XGB
z)6RQ0LYVV9Ve-#GJxtcX^}Oi?K*CQB$CH6Dl;f$acYrYjSNZDk=(?yIe{1W*55eA>
z00RQ@?KL?ZOZsR37FBe-o;O`szth71VJ!d?UN3EY-bV;yDm0pD#$0r|y+89~c-}Gq
z7IljA`X5f%o&pd|4oCHKpDe@vKw6Pi6994X#FEx+r$tv+*2n)tZ}1-n|EDiL(U)(E
zI@|x$Q{(?sui5CE!*FrI(Eh(#G$(4W=l1_p$uYD3-{SuNqO<g6FEBiS<i6amQX51_
z)c+~<|Nr0rf0F|}0l*!BYv})}Qk4Eb^g3stWBadrt`(||#_taybv~=?7&{KsC}Odk
zE;N~l*qd6q1+$7n#!|&Y8T_qbDt>$%E1Lj;)wQ4s$|FqYt8DF<@Px+ABEaqRd3k@@
zEqC$9kn0UsESi1#S;{2gIM(tf?-Kuw9vwR2!Y-kJ>+7kX7nhY`_jdmI`uqO2{jwSF
z<}qYgU`Pfpho@t+>+|LG_V}{9%e`31wdl`J8Tl6En28{*haU0p(E2>tn-M~!(!j!?
zP~<qqzY(VEJ;9f&D;b6xjs}mK(Mro}E=boCt>%bqtl8zjqO?Z5%Fc3&*|t@jdlM&!
zZOLAYTBfb1<<SVeggZ3|$75|*RcwwI?${PP{0FtQ9KJn7EU8hXR3SQ~bsBS*MS4P5
zNg;O(9HYId61LSSF(I{FX@1JHqb~vh!&5NyHHDLbh8%LU5En$S;=x8P&XXv}bU_%|
z!W`48dCV}cS&aF2#&ULx{+nZ%I?=>1)q1W>pY?b-D%aCS!lb#IsB7ccrh201ZBMIg
zlYfS$HRTrjYd9P{TNOQ8$Qvp#xIxJ%5<ThmvF6aeRO5v{$<uRqyL@b+z?2wwe~>Ya
zCq|=63X3$Ur<#MWBpRt9TlpU@n{(xJwl#V8lJ=;r{-lFvH|I(@xb#}x6SNx3#*Kjr
z3nppfUP}mY0#=`lH17*%ugGkbgE_dz#UM-e9NcFu=l>zhA+Gk9EH*p(H??M7D<Z}u
zXt9L=ArH;Md|x>vW3D(-hE9=rFeO~}nf0*@w4%F#D5@}`RNAzWLn*oIPYl0=1h|34
z^g;%HJD)x1ZI>Gzt`sk&CJ)Az{kq?>JA-fr;3@1Yr%cYN4Hh>hzBkUd;0iQ|D>yl{
zMjpNoe8bET{%?gWpq1?c%6_r$M@l~_Mo~jAi7t9_x4H{s9+_#Xl%%V+(t<*64=uGL
z-pWhQnL#us&WS-5Fpze@AZ{GO?z|BNKs<wFZ?RxnM3LTi?kPr+U>KyN7~t&LAzKy5
zvE!R9-%*b19#0M0uzOVWv=NjMST?I7v+aGsF{t(vHEjCWH?>}4$M62y4>pJ$-O>NV
zvq&8+k!Ly_pQDd06N#K}6M?6`!r5v+1x%nu%1J97gKA$jcTevr@RBXGzUxw3hD(Ab
zVe>0}pyRIxf|y#&uKUBNf=qi>i1ph#2Xyd{ZU3^WcQ078C4}BkKPSqlecPQGjU@Vu
z+ebgaQG_e3tRF>7IedRWP<*#T-3Y3}PRNZbmA&nik(0HlNdq1gg-TyyRYSiyo$KQM
zN-y?!Dem<}3Hv<wF6am6ud)=IX2fo;Kv<)xma%-V`<wk~E{>iSa9L%;*~Ln8-K5XT
zd>O=VVgH5$252CpXk>C!!m{>y=bi*G8U7KyZ550Kp@WW@zZ=gC&>8O!>Ll=fAMFg(
zpFkla1Sa4I9;C9e!@zX&w@`VT^K}Gz!pR=W&XoubVVv%=;KLht>wPl(vfS1^3}%_c
z)%-?+&8tL3fkT{nI`=9;aPi}Ks0`q5NjJPyMCbwElxuye)g$ab5<cQh&H9|E5>(^}
z<}*(<zsL`0=6tyYjriok=055{64Ptn7Q^bG#kV>m`GK~v`j#@eVihTp$s~YthUz}C
zYPrmRTY|GYanqFrlN8mKmFtwk;E<q#Bm-Mf$t3D1eb|sas8Y6@VAq22+iR<a*HDS#
z2<b55xs}uAw^lfWp!$oC-(%MXa$m^MkNcN0IZQAR|9K)dHY1m*tx(E~t%xn1P8Az}
zs1)`ynKPb>qUiF)scTu!=y%K);CRZIG?A(<qK~fZ`!Nv4kV7eU`0{5}$-}$|pIpP(
z7~0Rw52o{WkXgoN$lnqoeJsPKRuD*F5k9!XOrk=a1DA9=;4cDBBkY3Y>FV;^k;NYI
zJl!?X^=Zqcb?gGFsxO5rkWvghlwgHC1W0i*)uEzZ<rK_Q2Wz7i!gM+xQ3E&<lSwI+
zrh#FK)|)-AvIi(bX(miJ4(pXjhTGHO7kEtV9$5^tQ^0R=%M{h?m9*Z5F6VW<o(oEQ
z(pW8f^Mx4@x9Gx29^(=oxTpc^J5PQlmRu&oMI@7|hP|0x;~FA_-x59dQFgAr_@?LT
zFdSUjP5R(i9Tl3}x|w|1$e8-Rc_Yq}S?yS(O@%u>TeX1-l!c$GH3NORC?f5ZUp%5K
z!D+|0oPlP+pDstE-))N{y=^X)t+M{5M6glN?3X~$wwS@}!>>}8{G%z*p#~<v-S~U9
zT9TSu`!KKyp_G)f!Hk0uM$<e#kZgalzxW+IXlIOx$J<7K2Y)>hyjEC*?##d#4z>lC
z_?oG*qE`i}GW^y!XGrOD9J?zB4UdGzVVO_-2`d>xHvbyK0d3OZOP$WYLsO#1;_`98
z2?9^TV7_NRj`4gX;Lu(*OyMgH6^Tuc#o`j1OF0Fx)(&1e0re`HSMW<Y`bT1A>7Qd9
z?+bR9E7-WHES2`JB;QG9Nj4|!^GcShAjg;5ZukMYMMuc;MvrTRBHWtLRhJZr9hX$+
znV4oV$_pT()miV#vs1Litt3veoPAgvWdz483r%}-%?qGM6gC$OqBP^Qm^ehEZ(gdp
zkEb-!3&UaD8&%r=K5wLwaxEQc_zYmXe)t;Ahz#i=*L3hg`=E?mKjM{wwc5-Y9*VB2
zu<$-%JF2^vKsdhDA7}KAzuReBq-zklZJJwUP`Nw#NXijkBFz-!P>M9WT{9sB=Mz^G
zKLz5471jmF8=I?mQAr}t0b_-(nCi50{J7fQel&aq{rrY6%4}lzPfiEJe{wn)*qHv?
zMz4jc<Mta($eZVs#^Oribhu!6lIupp;)CSdVw}THumtpk1D=r~1%`*`HlUk3vH5XY
zPW6~T1N*t}52r&nlh*yL-I^V(ovx2ON*VD<X%j5X>rrDY9Fsp~9*>%~JpTyc({^us
zJ)L##9O|bZ#E;7NlISh1BUbZncD>)<shej*P10054JE|b@OBOU%~;0c{roz|>wc-E
zVZ-@4^CCyruvbQ>n_?yK{Q-_**4k>N_nt9(rAL@X%zA})uEF7^mGNmRvh4D5n82R8
z6#Yb}`~4&z%0!>COwihnDsFI!DUr>q05erUu?)GrrLkevy{1K~v>g>9DM#cr-(nq-
z0O4Pn;HPUGmZqoe24p*P8?IRo;@vMEW8ERD(Yw^|N>_yiiA|>!6$Ka)u@Jf9MmK2)
ziB9!)^|ZGxyumK1rd^QHa6lC@u6#LoOzWx8-@Luk<(6iH+dtu)-xEeIwVPq=;BcG6
zXc_8p2u^7f-IXm3T$Hqtk69M5o=*x!tE`fG{QC~0_b7a{f(<;w-^CkB8LAQ3RIP|>
zjmrp|E7jQFNPu8%jU}Idx!O9}KgDl$V0#wK<cjLob3XC))%%bzt6~zkigVK7S8rr;
z)pME-(qp%#c5&GR@)XS8I?OIJ4(xcUKke>Kmf3oaAd_U*`#v{Mdj_V1?Kjm>$QTnY
zKA80##xj?H!=|&RKk#mQ8&II;H@1Fwh!6Ao=QnIE{WgZd(|gx-1!?d>*$Z^v2Y6uo
z;T_oSi;gz}-jpYB4k$8h+vn2$CPFsY&Ol0Uk_54Tx0JDLEJ*HG`3Oqnc)nsup6-rZ
zVmPaeFFtr{?SAN5!kHnQ1b99rg`kh{y5nC%QyOUeCN}2qzf`Ho`d-#vLp`vm@@B~n
zo!X2~fgo}avwd-8r;ye7zz#7v;Le?!TC%(+6OxW|fBHiS##&U>NtY`Q$g^EBNHBg6
zK|;3aL`|X~FggFMVcx>JzGE5&g_T!5nJ*m_FxPVomL=kc(}mhwPuLA#q74Jr!PA=G
z!4OIyUB~T+Crv60xET^=fPZ45E|H>_el;<bp3hAhOV##-or4woZbZhGN7biD7!%TN
z_E4;bl)RU2FJ=Pg#&Z3_IIgpAD$llEz{2Uav|?_9ly9wBz>OJP!aa2ulc!Fda0ujq
z!~mB@Ai8;)aM+`OQ~cx;{NbTD;G{&C$Sg>ggr_b<?kKvvw~5d9&1BtULG2QC)o10?
zV_%3rXxcdRiIPm2RWDzi65IX>pYhk{NanCma#zvT<XbHHQlrjpT`)P13Us3Y`(Z@}
zp7;y*%GuQsbTVgtf$q<uiu&^W0wk8F&SZPPx=YWI*1U?&Wb|0@{2MG78Ue<Tzv^51
zc{&$@<hS#F#V^Wrs&N*0Hw<srA;qHGzUlz8hMNLQOV}|yPxgs;EFSOBpuIpBRB_hE
zr=8y^!ml`&2&OY3AaGp<)~_tBpvXO>rXK=&dy|4E;PF@_yqcnGcfytAl4RHop|~KC
zxxYa4&tU=@odtuAqlOjF16CSQw{W+BU5{I@T1R1PbWqcF9OmHTPdyQiJfq9VU|y*(
z`IYnZA0X4M<v@F{jM`cR@pCr_7lVk_TgG;o4UrQCgoSjY`4mh?30X`5KSHqKtN!sX
z#7G&<C5ho|yYLEvyo0|WW#_x#%V0a*!x4~0e}%u)gMU>RmP%o$_3wM0Xh~0AL1(pH
z2M}}Rcg#vN&x=n8H-u{55Ik=#(X0QdCf2vYd2R6cGwIQs&s`3!$@PaVVNjB46tR|&
zA1_d~3-`BO+~3?Rd5Bqh!YhL;x6lrWe2AAwlk6HhQO>kB5rG0BC|3(9mweWM%CIWI
zZVG7mmBV(6;$&yj_fk%I(_awe3$(<YqBn0JP@9nWX=<|453bh~?hUbEZ)@99a>Yv!
zcuN9K7^JI1$_5W{=f<-H-Ko|lDY(1PIyOu^ZQ&&EN~C9$Xo~7V1uNO0=6(oH`cp+u
zh0RNI5J3Kxd|&0o=1YI8bkb;KP@GBBB;pw4kQJ;wvqO}EqHAEdvEb8F*CG-N@I++k
zKPzH$5t;1H^#71K9w0U7)A->ew}N#|VC}UjWTu<ZWgqyX4O<=uNRj^%=y&LEMshZl
zv&*YTojip+|Ek5(l;*Y(Dg*<5AT%x471-{q$CrLP`q`YCb<BAZC#K&SOc@@3nrcVL
zNW6$HTtXe~37zwMeFJ}-cvecF)AUJ0RhH!QswAQyY4{=FPShN8(Xb=ObdT>9SDKGc
zUzky|Q`a``SZatDANSg7;8I~+imKX|6Pw;e^*o7q6B<*rPB7%>#W6w3n0>UODM5r)
z#2v&T2EzWyc9VeBUHLhA_IwV0qc5rKukX2^e2$nBDKO?*9#9@0n|E3m!O%5$FS_c>
zkzU7e)Fn9_Kg%F8AshLG-+|j1;z(GNmA~}7vT{qp!JA;HI6hf*+;U9)XX1_Gne;eu
zhTcnTdFEyMYdn#s)Nd<(f}DSsMCbAT-jY@V9f)aU=jzO@VAl~%I_HoObIl)46mv=p
zFO26<ktLs-HmFh=WoirdEk9T{G?(dwY1h#*IgY-JHEt)puqm!mzlhYw)6tQ%xXV1O
zbrl$zanQ4&JZ7XbrgFWwIQ?AQ2a}MMt=M`l*_}Zk35f~Lwv(YwzX;dH{oMIDKL9+A
zo=-(JAhdYcZr6S`_N(5n&jd_mzQ5bpMaVNM#rHKD|4u5F&hdc#d7vfeL-oLaSPf`>
zg$3-+$6I~b^#jFjNR8^o{AjqFLc0K(Ya%Ee(c9p2ybBU$e@f^Z`+Vm37P^jrP3frH
zT1K8at5w2bH-~FJ>VCnxF!Po-8(98O^9YHaKc{WKn_7R1fnyj;pY1%U5Zjz&sV3&u
z@hHo=qx74W#8?IDm&mtZ3_U_j5G^<MOI(n5$)y1Y-t}$t!y!@jamu(jn|9^b*%s7b
zc%Eu^O|Ie6xmr#&bIz_*&t;PF1a4K<>Z6t?=H!6RiPH8e@OO^t6`s%bruGNo?y$K2
zfAXFf8U90`H+uH}+Ixypl}y-YLFjs}o|)BJG%kT8uC-6!OqX(}Ofd9lV+l#aHVm_>
z7eU66yvm-c*6{^0TDzPq^85iG4-`5K8OyLr!tiQ)A2XbB?JZXLr(woOnIuL=T$`kf
zEQm_+ZiP5Ur4%`>MXOxC%KPJGdfln$;A~R{&)coV=i`>}^Yc;Jq}&DZny1Uh^ZVxX
z>3P`4>sc^vm<yJ;fP$Z>w!%Tvkx(t-MtaFDJG`MH!AUtxJ*DA8GP$(=arC$`-hMIm
zVTwhAm4*8gYr?aJI;vZYnAR98^;2XYgY(+7!6hN<vc7d_I<lr;s)wSJP=e9<cx}Wm
z%a}mszPq^U7pa5RvMV@jhIWYSVo=dhMB=Kc2}%xvlJIM;63rQAWZDwJ{4X<BPWAR(
z?n<n<(I?jy>GPyh*Qcbe`{)!LQiqnx@T8ke1$;)fbSZhbl)ME31=}~^mu=qO;Jm#o
zI@N`c+N=}KOKA(aij<IY952=m@)w{=*@Y3y0_Zt=kBf|ePsR)HVBX0U%%iQIJoCZK
z2_}u2h+7`Xf*GfR7K)%LX7VmGl7o~p_d}5na}I!E>I*S$vd%Q^ixy&)ZYLaDXis@$
zV%6BHJ=J<o3O-rwT20MfIbZzbcIZUCpTiE`&kmJdzA&L>RMk|DI}*kplZrdiz7R?d
z_X_g^O3N5`f83%Bf3lT{5cm4oN0%C^21Z)TAA2zL#T=P@c`-M_<o#Tin`U+(jgNkx
zvWqH_$RQn$tw36sz;)GOTc|9J2=TZNQsH^C=Z9Sm!-zT&9?;rDb{UFn>|C@xzwu<s
zDL9~OLS(=)j}T7(+sbt{;of<p&5h+EXatnJ2Ma7=xg9bIA+v_YFq)V5@P~=QutWYn
zc|>ap9{j+SaM$Yq!q0`aAG?v_mO+Ch>Z0MJii&lSb@WGA0nn@J+J=NxT|J>cWBt<r
z!)ZQXm5O~^y}EC_Gh*E;g`>rS+L|YUV=-RG91o)zDPoPulxg)ZTMFGce#80mk#N4T
zJcQ-qd?9Hvtfyo>l7l$fV0=s^oaUH%x~)W*j+9G`gvS{=(?;n`v*@Yvg2)cDdEIGe
zB7=LI%PpeOzauU!L2c`{-~UY(X*z<AH#dC(^Mu<mq)dH<SL++yNQ<}fz}|7K=ag!o
z65_wVuoNZY5x*`vjRMa#A)}xHiSD<SfoXKEi_Rb74w3qme(MDNv;iC%>#urvF;Zmz
z4n*t!7JhFmY;Px7+VUwfw097!+t{z4QOg-IBD`W`-f^kD_mg;LY+hhN37v0T@NYOp
zmj4cp+fFcd0iv)Wf9@V+Cl2t6X(VwU%!tXi&3e1^{5!)G6@uaL7g$5MpE5X+@(LCZ
z%xX^#$eTp;s1%t}2(}>2FT<g~{P_*O?I(#;8moHO3X`wapio3mOa78OsGy|m6Rs}(
zV2>emH*LKG2Y@l}IXzbHj8a`Br9e+=VpY7+$q+XBl=Je2!u%9IIGf`d$8nP&5LxSz
znzaJTg#xtP5iye+?`5ppB`|PRjGvCfLBR(Hahd*yCo8n7vZ3fOb8*x#9-Qxi=J<0>
z!`{YAPhrNbWT>XDBebwZZ5m}8NuXrZV>9iJJ1*BGY67{epOhYV9`>Mv*b`|s@Yyqh
z+R(_u#8AC>Up);7@Vg?WUd1;)p3YveJ*pa2BtP_Bp$GVU>%~U%kljobR)T+601*Pw
zHv5u(1^&XAgyT%BHMoG=sRC)~_gtQUzqd_XQLDkCJJ(59A<<tQrUTUtc_=W|?O>$3
zpT98R6&O2BPs86gGL;dSCc3v^hgM`Di;l4E1S(4j(6KySkpXALPrX|~zC(s-lZ2R;
zS`{3miNPMO)KeFJQ7nXJ!?6u_;DSg2scHZ8S@GYs{$=DI%MefKJ5@47jVxub?omF!
zi1W!j9gI<P$H!TosU14~O>n>~u-brK5Gj=-5x1QA7)!e)x)74=YpwTV2X_c8%*fWt
zPmUc6H+sks1^%>2?v7jnhWF)3cb5}WzBLq_x#P)ILO)a&Rvd&w3kaR*eg?tT<^)8w
zQz$8n3OYc=!o5mrzIHQc7D7R<33zS_AH$>B59x6gF16UFnBxn(Z)mDoWlYnvG@l_Q
zS$19$C#kiLN9$y)EwXG~C;`{pHn<8-4CZHmVx%zGn+Xtt4O=&XZ`roXZ_G`{&6Y0v
zT`47e`N10f;VEB)nfxi#B{w%}?zr&VxD?INSX+uTB?y&ur&0+H(mz*J4Y*$vt5;Ml
zZNg}uRS@BYy-}#svP4@TqUr^CG(Xu+mI?$bA(|&Sw;_P7ApfFkqkEcir#Tmq)8w>#
zoksd;M@=%_Cs764rMg+@7rZr?Xm{N7^~If~p;MF#n6WcnHNIh%^#4fb8sFCK!<1g3
ziJ`+i0PBwHL41olsF6;-?GM4i%Se({Ix|3d2(Fc|mjhXPHRbUUc#BXC1}!WdvF}3u
z+yhpZtTV&0gb9q=?BjMPDWhFVns#{L-ixVbb9ZB#D{jVCl~L=Z<f<905S%}S2Sd?r
zN-hR3KAP8gyid3TBIejvc4&?BeeVk@>^X*_`LP*?$IsN86wGKx6$E8Y1~r4HGuaZn
zqmgl$ykE&om>clUQ*&6U7%SL=?0)uA6y!Nf3{7x&E}s~GE?aLhvesW{bOdqghzDOu
z4@Yt#Po-BY-}Ro0*#*yfsE>vVL*qUi{cPK=V@p<6FDP^k>emdh<&qL6iw2af=?kK*
z!#?<^@mQ$8N2R^ZyY*p*5=N0U(kpvud;jW^=Ex_aapX%z>pctP8~+Xonf5SRoZ&M+
zPYu?I+*al}(+^*v%;C3gpq(TaZ%cD3QS?QxjB4AeX_B|}q3<?r;!y)3F=5)Tmh!w+
zJ`HKI5(D;>!zEboy%^K&_7O#OuOlply8R!0oc}Nant>58clf^?1!M*Aan{)ox{g&|
zN~$m|gwC`)B<kEImW(Mf9u{z&pZFv#Gv@0~kjP!!y+47#$i_E$!eoAc`61#ZT0Lc0
z-k(mkzq9iCP~AyqG$mEAJR@dwtR)I384I_zUbB8n3Mp)Set+rk`k<<{tCKLMjj=%D
z>n*-d+jMXFyf`-}86}4@bgm~WC40JkA9BNc`Me*Vcl*3PqJ+CA-0o&rCDtW5tu>&H
zz7^wU6<L|2$mg?~+lVsDUYC3ta)`QXnbFQkDgNeK54BCG5X<P;+`!|%a;fgH+j8WW
zk|PmaKkt=$@BOpj%8`2IJ4s=8=88GdQViWsDKnpCirkL%fOO!*RH6%L(BUL<Ub>qj
z<GOYuwYh~mS<-POXQ!y7%|xx9mh1|-;246X9@n$0{#V;7+0i+V5rS<oI$A{BN)>Z4
zxG1kQdlHJ!kk;JAyufYoo1PQ#i5e!(Q{gMO7(g`drq#U^$&A;L)A(X;ee>}p^<@TC
zl0f{EJLYcqI@ffcu>GokVg8Q`C-jRlpPBgC=-c#_LJO91i@g$FG8vCW^_f4M)3A!q
zYy%28k-P)oTm6mxeRAszm5jCI6b~)IOgyFHeo+sGU+R&t*65ZISH|MRXQf{iLkQC_
zoq9H;Y}Y5=dBCh8E<NPv_mjTGk@?WUAz#O|*Yg_q;3SsolgdmtXs9sXofgEJuWUJ{
z<*cHU@{1#wj`xnr-o|J;AwtQ0%(S#gD!;QoBlj7(MqQWq2ekVSO@GMbE8lp%#Gg;U
z3)-6fobj1Ds(|2j-_n2e*vT?~dK%2o4lF*gkiPrlV(!fUoU%5C{l;$g&Q6vwe(+o|
zaq`wG=KXp3#R&(Meugj>cpT8Q+#nLKT;p6^%DYl*3v8<WU=iTWVU7Pij22KYx05AS
zWi&=P-f7PS9Ze#nt^~)gOFeDPRGM(g4~s7p-vv|3P}u`{$0BpzR@C_n=^4ZY+Kmk<
zdyvL)2KE#1l{zb6@CUJe{LE3aS_{`+E==FZ{M(Loo|%`yFt&x=Lk@SVWArd4yE+7b
zEwaV8zZrUQzl8C=hqzKiE9kbj=E+_G?g2uol~_X9#jP^3^aRh`Y5c)EeQ}AzNk)g%
zGaJzp`rW#+#N!%UF#IFlpP9DjIPup@NdU*WsZ(fQ#F$dT60sRq$xBsDyEl@l9@G_`
zsRX*GSk}eSZk&2=%Xw+Up}=!VW<FY?#@PWfiHWot=+sPq&8C@R%Qqx^nw=cCPN}vs
zc_8blbpU9f1}~E5Bhue&f?QPWl^drY$P0T1K`Y{N;yozRB^<m8Umf6>gCLl<O4Wwf
z#y&U%xsGxL0`^C#s21?;HrROCct|;0Y<oNYWu8>2s#+UVZ%fRf=Xj)IF}g<REbkT}
z8VGPS30uBJn#p|BVB}wm!K{<B<~On=PyQ`jeR|%vJ(<$o*&a-R)YO~QHos|K18^?B
z*sBJpLa8P8cG=#h3`}0BnP69Mt4^!@PLbRu&IjOYGm`z0AZ)?){!+vV>>Ts+>gB6z
zLO^1?>+dKj#dYjU2Q<d`QG-KgI|r`^`Blx6R{Q7Z^3IVpG6viN_u6vry}>gx#*IbZ
z0}cFex6vS-`Vbyjri)!fD#R2+PRsPK?=fwGiq=jn4pZRc@?hyaLdp~UVbH0(oQIek
zbfNgYSA#}+;Go$tm>xY=B%TrM0Bl4D<qdTCv%(iFyK8vW&SH4y?_2iPGCPh7L>@{q
zAu@wfCGw{)Ebs~UK5PQADY*!y%@ZUZA~!zCtvqKZoSd49nHpnNEKD^N3JT_k8s%QE
zrU-IMuh6G|+s|u#)tDFU_cX$1l3JfQY$M=}Tx)&%g#iuG_D?I=-#Bx+e^7C&!yHDi
zXdHoy>E9vIWr%Pv*tt)pIM2w?LB*4`e^M286!qHiT;-YG#4T&XCW6ZI7I*WP&x1nm
z{@~|z##V4e2#i;L`J*D|-0AD;h!-716Ce9SQ?N(X(!K|6wh{lhq*sgf8HFfe^mIIW
z`2$qE4%I5%fkE?-#+S#DTdXz&D0AhSI*5k}*wPVFyR4|^3s#CRMimJ?ep_7s5}chz
z`^OTPHRwLSaHa}xEF3m|pWB{o#P6|+KlQ2iKNi}1aKHj{-yM{N&Rus9bMY3={=yLv
zAp?_-O_)Wk(g!S_@*Ezq!d{=_JoB6an?gB*wbuYay%+|Iy2KH#HHkbEVD!s@348<>
zck;6dX#mM|1hYgs3qYD?R_-445t5`f4qqmOOBp0N^|i&4+TI&xI>VZ^biTqWh8%1-
z>u$8M)R`Kp8^UQMs|}LYuY-?D!!i7M9^kk09OS3f%oo>A0i_dLYsAQ_)lXAGCew}s
zWIHI=g!oH2_gxjLkwipL_=$dccqMR1)+ckPpRAsViI64YmA5hL4mzwuJ<IWZJYmka
zA2o6^=f(`y$=>7VQ2X7>-9};T9_Z3qgqERW$t{^}(k=6`CZv(g49;Pd5Z<v+zto<H
zv`)nS#m^07qU`4d!PYVWi;^9r=|NdMfM4SD*xEJY_Z2$9YOvn`2a$?2@^hY%HJH)y
zO`1TPF+60RgHkIdI7%gMw{<=teH<8SJ2Ii8D-naj7=W9KC+>bRBDYar^@YP@*Qxe{
ze^MME&;z$U344E*WUB~m`XRigIH4v=R(#T}IQlopYGOq62OA!s<z-J`m1aC;bA?u`
zsS3%PWeB6vbFCrScto1r0=J-4%|I4>Yd2hXRkrGS#%#uD164Znsc-X2>D6u6O|1(P
zs%6A9V4l#>Vn+{3^nyH7Zx@bH8ojR9id3v*p;ubRBa>7Zb!J`OalBbaBy)O!`_~Zl
z8dVZKT4(5=CbTSg+nxy=+$1WZWK2Nkw|VnJ{h(i}3{<uGN9MK7_tB*b@|=Brsh9ow
zRoqn$98<D?V*Y6BJXD`IB<+R2F2)u6*-$-d3m=QlX#d7L-&&>hRjd5Y*}!Xo(87XX
z4Izs}z+>bKa?yfWz|gD)=^1UozL8-;kDRV`ULf&(Th!4c!%r`2{jz&(Pdt32_+Xa3
z;YV}aNzD7>&`Uq<oD3_SD9)M|3oLR*aFqs-EI2LU9I>YwtCx4ZD~N*Y4JXpY)0Z)o
zjLnV+p8oDw+3<bw4~@vvz^`Q**062FXPYkwOh=yU|Ky%A{ihHVGYj*7>z+laty^!f
zA$vWmoX9~~iV!$~KM|`DrBSh@t+oVsi9rBU&u(Ul>BrDqvFUURE6b%gPh0-@EsYjo
zSgY!E78~vF^#k0dTKlp}yCW>0flMv!mTH9?i~x3y$UrV0qMhRfdx#w6;*Di8@8;_8
zef#V4s(<}lBU@zGXZzCo^>H47*Snp@?d5KN{JGktns-a<yX$>9f;CK4CZm8#f>9OC
zgj&Ab5iJhCw{iz5B7vdhkezZ2ZV^{1=3esSbO9Qx-;wue*$#;!2&9?~S4=kTX{AJN
zk%;s%L(Qw*6T6-)c+p@=h|2bnHZw=<^_3+`*<#_04cD|UBu4|U8^IlW>OtdGt6AaQ
zQh$}{sKDG99-2To89Egiyq3G9U_>pQsJ9B-f>$DS+zmfi%_yYOY-STT<Dg$JQHwf@
zo!PlV#WagOT?CN~OT?+;Hi6vIGmKzs6oNOpE`|CcM%goQQiwCHST?Gx0<_|ANBat1
zMr~^U_o%`U_}W)VrJXQRlLF(}RKpsz!olwE-l!Jv6CSf;bkn%6&ScOg5eIb^qwcIl
zX9%UV=s?I-F5Pi1%LSO`U0~4@>r;oOYOZWM<so);?x#D%mA|Jsw)@qnElmsxynD|n
zWKpiDCeJgWzl4!L49qYe-gT{gJUtHUz!4Z@w3-Gw3<W~@YN(17#aQB8leJiA>XMSg
z`xvtGlD?U;xLZCK6o_nn^0+Q0E{oUg&Nx3Dr;7TQHuKGJ7k#;h&9IMmk$ouK27*tg
zin97s6to8<b7TU0AsuV$G*&JmM%y9HK2vXG+jWpAung@TS^Qd_IP0|PI9!bu$K^5*
z_#_9oHgK3u_p5y;E-H|WOw)%7Nk@i2C5k?Q!I6il)SwwQ(1q_wG6|uMe@26{>|5%@
z6Uqpfh`|NRdOOZLVK+UOZ4-#*AKU`p+g>xG3C{HeeiXGixw9Yq9_>tP!*r1kenySV
zY?$~pg$*qIW}=64NJR-Hmd(Ga!>>@@9u^s9;z^<WLIeY~AHUyF4Hu2#y+<@)Ap5{P
z4Y%b2ZNV&yMzqL={{Us1`>ZW66mGJG67h^2GKj)YFBOCoinqd({)H^ti*T1I>=wou
z8)bK^JcJu6fwv<2dBliU)BkrQ?}wmmlrDJFp)WH-u(|>trb%BRH}yo2E)vkE@nfyZ
zSnJ98GVG(15eqF&jkWakS+|Dl3WZOX?lT&|v;DRajtgLmYVQB*9y-GLvh>cP9U_`s
z!$p{i84XnUCdVwZY=_-i5C-GBAn_-_NjE-$^lvRH!OMy~@Q;SMB0i9~8D<%>wqhE#
zd?bko3{Ye^x+W-6jbiH~r6X_|;|y|gNmvBlOk(oh1##A>>IKqQp;tN}FnB~6YscPS
zGO)UANaF<hN}kRqAaBq_h#}twH-N!F>*;pKgIF|dk!0BXAoAe0(1~GeXs*O<bXG%b
zw8mSFQYfNQsOXlJFf;W#&W#VoRf}n$qA0hohDNQ~6D;)u=N|!m2<V#9gsP(^G%vgU
zv+Mz?AT%R+c|w)a{yJ-52U6Kaz2+k+p`<A+_&X1QI-k{l2)@D+RI9UVKpK3OLOh@F
zGhiiJh18VSn3^Dv7FV<|j9S9%W0J*1f~XO3j_bE82N_qR3Q7;V82d{oFL0y&F7Ug<
zM<M1oyTEyDyQaNetpFJX71X9!jZh=JgT&JQnu7Jsixk^J3>psUu_!Y?3)9LhNelzO
zwhDjlZvlU{2fbS9vxn-oD%UN}w;zR1$Sd^u4X&kvRvU}7c`El_2QuKQh@lES?RQTc
zXwFt%C<XR&$C~U|tCCV2j$FLQ<)B0v={WUG6guDYP`raV*ORHpCfm9^;8!LVpU^sJ
zCI7c9S-qt^G73lt+qUn|V#|WTDOm7G6JHL0x}`Wjg8Ks2dgEPJ49eeZ>HHdZfsv0a
zyJ{~ME-WK6-1%aDGA>iFM!|vDphUw!e%^S$Xa0T-yjeL-P2Pa*`@f3!f{{Y-*RQta
z&h93|tjQdNHyIJ8DSGAr3c&}PG*-FwEXFZ}6y3XYIX7olhC0(j1oR-g2)U(0u~lt;
z{>HGiisk|>xE~gTj1SNg&j`_ZHFb=Ve(aR*_idbp4mD|^QIdY*Gp(^ss*L0U5hBA6
z?lP->Epf+;%4n!e12L*1dqeIZ@XqW<%O~0kQ9_QHLs#x3<+CFiPnHST0itN+ijFqk
zwhXssNm+B3$F)9DX`5N0%ppmGL}S2#z<J1~lU7E{WYH`(<=A-u62<@lqq*|;w2nft
z3v6m31Se|XX%xVtUm=uJJ%{v=Lbs%7anr{;0K#s1s+UV_cyZ!5IrDXQHc$?1Cdr^S
z!tFYjJ<(q`=Qobdwyoypf!(V!L{g+O@gRl_R7bpL1biin2ps5Pa^omBoiIE!TtJdK
zW45duodBG-8v9-frl2ra$`B@5<Au&qgA*eCB{w<Sz6(Cb?xSTyH5+-Lxkwq_-kRIN
z#y{13;08VV%*>*QAendGD~uODy`eaw*P$z?VzpDJRDmVR(B_%CFJls%fPFaCRjF`~
zOD~L!^|yh6En8}--S4ca*1zpA&7j_8&A?S{df+#!o_;~-ilK-5e4kKC$4Y2?zs_AT
zueAD7QM_A25Kgo?he^h4z0)EXr}TZ^bvTI;R#yo@X5dD07^GW0gn{~qH`@Ep4zDjh
zDIw%hwy0wmdYr6)(!;j6l$&$AyBU15uuU6=aWh!C_IGtPM<*-~1F&+UK6-JDjKjLx
z_XFY`z)P?(d;(BcCZOqyMEI#iPmj%uhl=Hf=%E=rM}4o<-+wrnz{Gpe#>Ae=)@BR-
zvHYc33-3CDsF8!VUZ3Igbih-cJrQWnp(RWA$!fWHwD5OH+v)1jUjq1n`oy5S%c(#;
zIHWlLEC#)ZOiT3s!;ApdR|p-~U<B51_NWVHKP(1y%BtyWd%~?70!sQUtM~fPYf4BQ
zQDYfqs|Z4c=ybAX%?X0<m$CPbrW)eI>>x=OB$8PMe{jfiRB0DdNP)-CVMXv%xl21D
znv)h3|46T*bck>N#mW%dn|g?bs?V!SjTAiM&xd<v++?6X^MJV&3$~}(sjw{}7{|*4
z9^XF#NHt`Vd<d5|SNmau0<Lu`gBe#hZ5SUOug<2+(a?L~u`ChFp3d&C<zxd^-c*4)
zx!DU@LQ!_ND)Vxoud2QRe2{%Bq}9@nU9rYg<C6JdTo}6CBMLYb%u1(OniW*>Ykfp?
zp^M)mxb-wbC1fIT8`B^vCXLwGsTwQ)&Nl6}AHgwzK+LR1T=vYFV~xLCz$Gp5ut9m0
z*3urmr8S*CYPYL^z6*2BViLua4GW(j*L(kzR6W9D@}{rq;%SJV3ezWq*)B6ipUbzY
z^S&RfXa5eWf5RKS`uiW-7Bl^)siyP*XY;>|`faLd#H_QRd*|r<ZBFu|my~~YZs%ha
ztJ}PQTPyYR5)1H8Hq;2EVsCRjLw(<Vjy;x5X8L2o056+MTja`le51)bx3S8x;pX)?
zy|}1S?Kv&3iaa)>W`#}b2*PW)ftaau)ZrwIO|w<`^|?sdn=s1&7;72^DKt@r@%?rc
zvN+y8C=aoL2CP995Zlg5YpuOABeT8vag%dxqhlejl9@Wq=8#N;BxNi5lU!&bFo&(2
zd&WnJH9gkxUT5)3_Go$K+<X(pB}2t6zYSXyjm7Hg?{s;Ptpa8>t0S^Z^a_kVx!c5E
zPtINEAWlxALU<s!HrHk%^I4a0a)vR!e)Cp-D$z<?$(h$*iHl8)O@!ih2#EFVgjrY=
zbnxE9%?we)hrGZrS7$mUBlq>tI+PpV!I|_|$6|^FtYvpbtMy{Dt$1ICFwqZ!8Jt)W
zn-^Pt0zC9U=U!&y3+Dkng%l$%9Yu``Tv>~CkxT01DBRaH2WVuv2ZCfFWJh|NjLQ4c
z()Q))jYoT@$6a&Tq~Vqs%mk`XWYSHYQ#)EXU9Sr$TEj5ZKM9-}<#;)i#KKS}AOdh8
z@ZKPZtkNP^WP!riZ1g7_4`jigfsD9BC09~T=ICx9Bblrtv4f+!UE5;aG55~N1cAL^
z>Z4q#%<mc4o*okjLtkHp>G%CoMxd;t8N<`a>yO*WItVO%IjX<6M@O=d;-?F{yt@|s
z!RKIR@epK?INXwF^l2V=(Z^#mcFq=9a||T1=7hzX9^?}9&bo5R`N{Vjq|-cqDa0#p
zc}9zTq`ne-Sx_D<kqxLp$P(qPs(n9S1l*G2YveE<y#7^DQgIaC`noa$j}^<iQ8PP@
zr!6m%D{jTy+UoM)*lZU0pgjYTokgL2bI{y*gJr)mIc{b>@GQ2;rtX!00Z#B;CMh&g
z;3XIqy2^3uDL0YiD~wt1pws(6P2jq5`F*0{jyU(W-2A(Wuk$mQ{u~tfS2!*E>#5Zv
z0)|&alyF8j)1oro%j9DTk7;}&5E!Isnx+^K8=CeXcptD$&7KMqrn6U`1Un|bZ=g6A
z;Ru~~j&Vs3cs^|MjU`P4)5|U!&JXo@@ZUk>d|GCxt(Wiv3(%vnm3I8wN6Cm@CxhKb
zt9Q9Tkw;ovKi*lcIige=y&HCV#q4Ez)VLEpvx>t@FxmaA_SQUL#(cBFs~Ai~Y{Njh
zoxj_mKrpVLoVF*GXP8BE(sJv6qk}HyWyx3DbJaD(F?0dB+NME6f0s+)a4e!Z1wm<U
zV-%%!o;_P6?s^Kgs7!Y8?zqgJCL)5V_ubg~$A?ygL72vsEUkD8omv>9VZ{vW`@88y
z-fb=nwJ->Xn%=grlVft9qjNk<&pNpYTbNreGllUa%#QKT&ZMnuNF@_{D=|}#-Zdx&
zNHcGrJ@7sF3S(D=+ku2`h~7>YG{W{j<(ao@nvsL4wul}x;N(JS$8hs7@L=1hp<g>-
z(^6<g9r+;!yiP8(Rm?xHrL{QP+#785q$6M1Et~ob9jXm8z^EO-qSnoXL=rj@m?Ao@
zqo2z&h6*m^XIIYEP1~z8DF1q3C9&0e9Apd6uYDdua!bR9x#3G19QF*YnVt8A^E#3o
zgSW+OBSsIX&MvYJlek3*y~eHHTvL)cNaY}B>1o|cDGmZ!)LOSAyfcC%h8U;~7|`F(
ztCqict&$TqVr?#+lR+=AzKZ#SELIfZ%PuZGD?|$-qRrswF)}XzYbcJZw0d0mXPZPC
zDudwlH3)1QsR(U|N5l2)ktDT_6@0gE(u=AN4jdMRebe-jrzB12w?)M{p;u8p91!OC
zF|?RFbOiEyxWHeddL~(d0CIQYpU1s&!5r^X2JZyhMvN`7P`+7I&^vd*bjt1%_e6^a
zzOR<bDSz1n=)J6ftj%J~EGEeOmLMmBQnt;9@JE2%SD2m^ip#p?5Fl65Sh13h7K&Bk
z4dSG?JEy|b<iSQH2}b#ePFz`><kj*p5BZmhtb@H8t3{QjIU}g|M!0g+RQlWvbb#=q
zlGZWjZmebI&Pza9F$BXTTx9L;MA{!7TcPsnn2iZk^}3Ow&yC)GzVGeZ*09s+iSq__
zyKYf<5Ol|thm+|(HK_e<LR-iuAQpWG;|(5y#K8K6r>g{4OE%MgC57v*s(T6&#QgsE
zJ5oZrgnK0<B7s(}4NZjDg?^YCvQC?UpgmGxnJwL>jIn^xiAqYs&(EKP#Y-*whH)Oa
zffgN8-moF6h-18_dQCrKzNdU~#z5PSZwu<+kqwr3u+tf?gy~1Ws6D~ui^c<7IGBI{
zI&4#xIR~tz<I@7Wocq;t#%y~zAGAn~m@Lln@3Tyu>e*T9J<pnElDy2ehN^;;Sv7t%
zT`d`~L^5jBbUQ4<6_gzydmvj2iA-hf637I1Ca+gq2Egdt65oVJQOKPD)E~=n8O4p_
zglR#2{rxu&Ke+-abnSqZJr&BhH7fPp{wUMI{65pQMfEg5-lpiMV+?0Fk0N0PC*$JW
zkDuY60X9|?GE;xkuY+qi77bGSCC-JumeDV}a+~G$fX(Y)=TKM=i|On79DkFUq~`>7
zT4ULWpBqc|>AyKg#u5twqYla5c6NgbU&RU$_zP5!(B=w-zQ<Q9?#kAVKd|@fN5Bib
zI^mw|$nPZz4;<juK{EOfTk%M8jEna=${L3EIFJ1F=SMl85HgF1eW<k$MmEMB`i3+j
zYy=2NW5|1G>OO)yrJNtE4HTSnIWTn6qs~Q#eA`U>wWqO-((1S|L*kM$<k;Yv9({hu
zI93_c&A~(a4}0{d`Q|V;dFUn$?2ly0g4}bP%kvrBjkGE%-;mQyRki;q493j#A1bjJ
z{`+A~g@B2h>ukt?Eagd)VpJqBpMQ)231F&bV}d2|pG!RxgjlH2l3hVb`6sq+@Ag?B
zou!r1^ayZ&zWmwG-E%~y?MxQW``bxc7aQ-d->$Xfs&QkE4lWx5ry|7U`(bMvNg3ZF
z0=0owHi~2UG1t7E-*-<s>sY0(4_Eiah|PH2UT*D^@h4YL_oQcEA2yagn_u3~(>c6E
zw;Q&3KUK&pB%SZ!AUk)rlA?4q$xQE}Z2yF2h83)AEYiK$R$I;Y!-%7EukGS$m?^I9
zD;pwxgh_=^&%-?oCC@Z;+tElL7bj(T7o>=?HdUUMnq_1>^*Sc;<XAc6^?n*{XMAf(
zInZ&#Tf{sVx3Hc(1=r`rGTOB2vwv{>FUG!UNw8&svRqwemu=g&ZQHhO+qP}nwr$&-
zo{5;Z^Kc{b7k2KIi?*EGN!H$rS{K}Jp!(s3dN8>OWlVwR7Mpp8#Lm~bW_TN~j8YKx
zvHi=xTbUsooz^(o86jUmrymM1FIw)6JQ8O)X<CvIY2@VN%v+u^o}otOALsoz>wEM0
z)s}X?qO!W++h2vHE!KVKg(YEqkA>9~7X2XJ50BlySy}gtr|Wtk(ZpwOCtwTI_yw?-
zx>M~9;M{91?Lb5T6KL+k3j$UNN_7LbcQ%@LW!&g?MVlt}L7d{Ey$q<T7X2*s@iH|g
zP^#smM-#HR*H_csK6TcjBv@Q0wY1U@X&wVo?Ot><i|R570mWRIgP{A$LWCxH-VD*_
zPjemCc*t%D-e%vDV@OBdbWcKOyowH!ll?c=#-XLv@B^R#A!vPXNX*W@ETR4mf^9Ti
zhio<Ob*%lT=mUvtfgr;>#;tOSi~f^tOMFTQsN`xE=dQMAtrl>;ZE=Hj4c=B|u3$q<
zRj|N3zQwQ;Fid@yIQn$yGXf0xhugBe^1y@olwn&KUkv)fC+b6_S+h|D2Ah8!P;JTU
zoe;nBAyl0De9dP{$jMRe)YiRAIW2*f0P|+!Y5ttIMq^jiT)m?9(WsckLX972xwR?o
z0JIxH;69vHacUFVoo@%$&<Gj1J6U<3y0zphs+P6@>r`{A|BU?oecrCUb$MFL0oU}z
zb!8${Bd_M!1`(NhB!94<L6_wo%51^C0<3~GsL=52*z~+$Sjkp3;Nc_1?!Vkh6H3h)
zAvpr;dN3uP$|F?X2h1xZTsV6X6WliH+nM&twi9V2UrHbHHlk95ur|OEzLrA6wH|l?
zjuUN%0@OAgkgcNq=Bw_O-5Td%^9ZdGhCQ8>Z+i3-FmrzyE`G6QrLwP0n;`z;`?neU
zS#Z85*;_(b%Q?GWa_c`-+sG&>)MIpJiyv0p|A(anTKK@3Yc`xNg??!|E6WKz02>By
zk!N;}`M9DN$Q5$gChJ?sute>3PsZ0PL!#}HyGfeYsqb5G?$3$b`&0@(VWK3ZMR>4Z
zN5$B42X(Y5Fc)VEdJCrc@wlkZ5cGpYV@+y*I{d^>9%Z2Di73U%YN-$1^Hc(?)l)X&
z@{gO(P}H+ASZ$H%_h$d-PN<}+w`8^B=BtbDdlO84Y7j~e6mPyd5v?0P)*&d$Yu2Ni
zE!^u8ml?o7o5^3c>PC!|CsLxScHHCjs7NBL!=z6NI+`0G>IgfQBLbB<UL3yjMPh;c
zfEdX}DYPy}2uEj*`zGi^J)uS5Qx210exyG{1VB4EBp>PEIIy);N_Oa#SagTO!9>zo
z4uLVEqlI@|TbKSFJ&DLR=jZB0`d3jk9vH$8{IAx9<l%Fu2a=8{E+3Y#q=*JMMkEUo
zRQrEg+(7D?5%7uFTre<M18bdRd$W8-#w*cu>k~u`m`z7o`AcSL{xY0UNpo*k$z(lA
z@O=~z!7tZcf;Sf0%pfzD1qm3&^sj&Q(oylDr_~MInTah)Zn6BhY2y>_uP&|@K!@_c
zf|mPM=vQWQ&mr&~v@@afpHNx{e4Nt-UD!iM?-aOwC6X!Pr|pv%fM>{CGPNf_;hk!9
zxZqnWoYO74d5PCm|50%05(!c3RC~vS$3<PQW>Ve-^YDs)9&%j4RDO!B+e*L@N~t4y
zdAcKB3$M+>Ide218sXfMAH23?o(raiRR9eE<r47Yyg)Yx4ROB+KUB9psHvDvK8jRg
zK%j`>y!9(L-!<@Q3?{v1?KyI_l?jXT?=GIU4a=#XDy%7KXOLwHDBM84frMinXM$sO
z8_L8*nS8Gc8mCGZvHy7>mMNk~60Ua=24dI~CaUFNxFG;Gm{pCp-}7}J<^oh)chE53
zpWFBbRg9*hF@9<RA~0ui0I(2m^Su@mw1}va#9-dc(*<>~7o<i{AZt)Z*}pk$*>?&g
zyc2R~E>k%`<@f?r;%^~?_wv@%2K#NY-LF`eH7<SqPV`c8Y_2Fvn!ctse}TN(J4r>>
z8Sf-$eN;7fRa2f*mxy-QVQ~v*SYl7%D-c90hVz|k>(@+sN>@mn(!F4GYg{mmC}>RR
zK@SCy92UAUo>W3yYtF#)D|xtk-a|bc8gV=)5_@ig!?d>b3K-nBw>8040@02=eO}T=
z1lG-~mhV3BX3Pr<ognbq9;$I4_RPs7(5HBOo*Jf}!TaE2d8<DGRPBR}KnDuWDUIG=
zRDf%{e?t-#wdU;WzI52#$ir&e+mnn-+uq|5V#lPWhGdh1lF22vOLA6t(sNnu2{=md
zZ^1iV^_VFC2p-VXYEL16<ru-#jk~6u50Og#{9<t3yKPu+qfbAMGPTgpTBGs~o-XQK
zJGE>bvV=k*SG;#%?X1N2de17=({a}k&>I!M@WW2*b5#iQV1^L!P%8o&^{C6f!FJgS
zOKqiwEa4-YTFq^^SH+^)KS`@_qmN{af9n8DdVVONFcq&@iR@8eBSMZ+qS6+|&!BF?
zyKyhq`Y!&Xi**mys2~AZp>_q~LIT4sKA6shsA`iiS-8hD71H^m1}uj=-6-%_112^C
zN+$K(HAZWY&}RGS109qN0E?J&{O)`%*B8~Ue4nA*^7hL2=I#4>@{6Aop!1(-=)c$-
zR)+uCv*oR>{(sSs&ui`8=7KT{kRL$osFaI?m2`;-lI^vi3635_LR*O#3>;^MZZC{e
zAc(Q?#4*9z(ER8s>QJKhKq4F~FZWiLdu>CJfilhGk$$R?dx7#y+rs!D%X>;hL)Z7=
zzOL;Kl#cKsQL2;*y48Vc_vhP7YiMro){f+~V*dJd_l*0;`}V1;*vH4qqiwI($8+H3
zMabg*{`-kHF(STNakG=>Zs_!_aeqYvfe2He_u=z+t7kvE?5{fSrjP9=!Yqd8BJ8u8
z>$P82?a^T!Vbw{v7{jy!o4R8^q&5+pDBJPA-Zl-#w^@ucql>k*hUnPhtTyLDt80lZ
zM^T2!?(?%78|>S;Yt6f8l16FC_Cb~I@8jLyo<Brz&-ND3HYufsYzS{Zx!Y0NzAjdz
z0ru!<(@4rHQvP^3<2K-wkZf4Ey^b`jg^6w$aSjaMYPJ}Ww%^rc3Z4O?=^KDgO0A#n
zIqVzrFK-`K8RH8mwBVZuW@$Lsx^D~|?U4jn-WMXRx#nDxd7BjxZZwc2M!{U0D;{7k
zrcDWk8!VfgZi>osK%yviXT?FMllePXb{Y6cm|j^O8<^NVOTg&7R!bY*=zwT%?dLE0
z;{XRM+Of5omw65&;xjIb<Yc;GVhLLH3P)t&s{R7kl?2-WGa`XrY?c@p)||BxY4a}r
zMzSk{KqK0bxS2MFSSr|Debx*)rMdvD+X$uDorGj=90x|U#B!H~jdj~#I_(i|%L(Rw
zF{jNo_fl(x_~e$pba?$B)i#zvBO~|xVD4OpRo1pT@HBi|@9tw7!4IkBNIsTdsnkAZ
zoj-rV^14IsRf@d7+1j_&`^JOeD2CC6GS<MCi{Uuh!N2O7VR@>FwF`uuKEy8JK_>WL
z!;oW%|7OAbfwJSqo>#}`$hMxB;_$A}gsKUmMF<HzFl-ZSqJa9_PG%MgTeYwc-N)$x
zmC8Rr;oVSln$3I|WWonz1J4Eg2X%r>9Q_!Yo$k8r=OEAg(^S%hUOF^IsHN)<!Y551
zSQu@-*|oB~o`0+1s};Blv4*j7zmZJNFT~rlQi=~~9sn|c-oN}NElT;~9ysExUD8Mf
z625h868qS~6zsSpmUcC7`n?}fsbDXTF6xQBVrevGSL%c@fdYtyZZ)$*IG^xMLF&7+
zCb(01Q!chfEtUo~TPA9aA4D8*wENOu`U?WrS;%n*;y7){3LN9Vc~k&^xqH#tl4c%U
z;8J{SRxrgNZ*<u$c?uc7ki+mi-X_I(fK%!CeUx&hLo)?B@(wfVB;&jC5W*sqQ0SBp
zW3f(P(BJ66(2R}g_@oZ%afX20JNzO7{iPk=HyNg#Sm;d}RP^P0cXGr=fDfn1>Jkwk
zpu>?2YY1RzR_Dh~78kf~xwu!Jjn(?QxENEgz&M*hfn*s`FHZ_^FUc18StxgQ1R!E&
zu>5X;^~P1d4VqO#48&YIANpOlbY7hteewRbPrbL#^Lg!z!{gaPcT6~(ph}_Tl^+y2
zKL(eFi%w=I+gUbSPqZt<56Wx!F2*Ab47zA30QYjV{B!&!IhWNi>Yhx_g6O#sR5=?s
zr_cU28ISN60or)8`BvB_UlH&oCvhYMVn=fDMp}$I@>Xt=;@ir(Jd|rLg6^aLvML0{
zPSVXvS-t{S1<)=CThGE>_r<_Mv>UWDgC3y!WZ2^p4F$;fdAk9YNEa)fz$Jg2L@mK4
z_&3=klAJt}a&>}3DBsDG9B4X)p@G{*{Tv_@P~*so#Eiz%Cm;rWNKXj^N`Hy9JJ^t-
zJ~u$*`PZD;*p@yAv7h1YM)8JTKaFsF-~F=X`knB-O7Zf%j{79QN@(lA=PQ2>3mV7N
z<;RYAGI$K9)iL+x;uzgspy}I6)2@^u8LkYnfSPp$zZ-mj1W$>PGBg6C&Ar>nDUK)c
zUxf|@$LzxSRoF$u<*DZ#`@)e6zA}KRWBejKsn_l$U3xs#sKev!g0-vF)8(z<)n~@Z
z`cyAU`sFQiy$ylxmrz~!Fsy(>J)H(J^NGtS(pENZ2@|x|BM4&^-$W80=8_`HTpgKf
zqZDi%bKYkNMwCaIipYd*7_8~+f2;@7XkiLRG?Wcr0pR3r%k#nPXpsDV3^GVF!tP-z
zHaDQ1Fp#ST&kP!|Ylu|SAxJ=-0cL_3b&~=u7y#xApEKDq+@FTyUJ;9;LT5<pY&LK)
zG|YQ*Gicg;Q1a6Cnc*|ar#koPXo7#Z+xS0_gf3hbrN7?-rHF;cCn}-@d&gjgeN;;c
zIcZ*In|=TMtX_Da1vxo|=CW+xGew)+SU)K^WqSZ~&yp|!{)3?|hLbO-$_+>Ci^U$n
zJC><%IdX};x_yDPcYA-}T%-kW42r0}jmH)+j0FtnuSMs8HhoN4WO#3~JHBgh@4d0H
z%oFlYBhNMc1oJ+mU;@Q1^>@-w+=|-TG1sg_aA#3VoXPQBEL8p;9XSHS!BE|GkgYAt
z1Hh8@B{)Y&#*~|Wc$952MqHC<0~LzpUg|SQf3G(9`ghM;!XdB=oo79?NkYSXaYCKs
zBhDsUT6tDPI&$waAAxYB@p+;2f3<he`O4hD&RSdldkf92$iYuJ8`KQMd&GbF^vIGE
z)HLg*mqg)IkvHc#48kGUI5;-~AMJwx=$o~-iUIYq889F|5`)xKhq}=sETmZqvB<Iw
zbfWVC8!>-TEN(Bgw4|r%u*Hl~j5NyK6W*}87<*jN##=KSYKwU(9eA_iYxVTtI?S+i
zRiL$6vSukVJZ0x)3KakTad!(?pZpezLK*+pIdkXxPzsjWzXw7QMw^UmpAWBQie*c(
z8oIj=g|?oUAIy!Bd(-ql%pIL4R!SY!o1B*@r|2ugjMyr+9c(DbU3Ai7AR>>x92}4;
zWsJ`$XIXq|lOdNU)}2ot*e?vCm^8sKHFFmfN9XTKIO9~dgraTS7?qw%zoLQy$$a(;
z!~Q$vT|a}aJQGxwlH~94ua9JRDqZ77g;kHJuFq@q0C5VN_(*R|5>#u6L}pnM9g|Rx
zyk^zdZ&lD=TicARK%i_&ZJZoz-g#h4)A9l!SyIA@GnD6sry1214EPN?0Z(Sp8&n&T
zzm$GOO|^wa#}iRtM4MMmq8R-s^8R@+S8*=vp(NsRHy79Drz>2m%PzG@xq*Gh>+_#>
zh{yS89Jj6eRb)tz7gXUzzec2M`@{YFwU;2gA8D4{sGusYQ`Zm`X;U#xxti9w;RSJ;
zAt(t(iF}1+DxU6m>da4Mnhp8X#i+=5uzJs@dJtN(P5{v#=9+m2Y3?laT5+PG$FA-$
zoTsIZP5m}?O`tS~fNV@PIYgDM8TZ}oq`=n<T^2rT*Eh{5L2yIh%4btPJ#|W^s?E|a
zBk-pk9@nJ)qOR>S*P;Vn^KZ{w>~Ws)bP++Ff)n>TPC+KTddEvqJ8?|`5i{ybMveLQ
zEMeO$vNQ9((yFI&g9bLH5+%OdvDAb^v8KF%W_~{U#ED0OE+crE!qI+e<w3-cY8GEp
zyp8+NxK@H~dD<wZ52|V2+7r<JDf!pPVl@+pjJu1wRUqr+-OfmBZ-xZ)hyB?@jtZkP
zf}75+zR*eV)=16I4$AxOFW_X4Tj_s6UY7qL0O=Y2_tjHtmH$fsLjOMk@Ih)?0q%mH
z-MZC0Gkg6?>fpw>6I`Q&Smx<2LSgG@6_toj)Wu;AK1l3t5F8#4?!pgTx0c7chQ}w|
z4Lzx9>VRtPDv$_DD1$yHLa!0(2Zpd8M6+6WWN^0icWcYHyM422rG9V!<ma_BbaVQ8
z*A=evm&MoTD}h9kvs_Qh$PPM#T%z+VVsV^*h%J`yi2rCN3KFH*xb9r9{s#7tf<i#q
zys+llT)qvv1dKTR*RDa4mo3(_xlnnQO$O$hcS03LJUJ^`0JTK^=BX%Dr{!%{>cVa*
zgk`b?M}(U>yM8T@Sc^sturQDl#>~1jH)&8M@x40}274wZ+Cz9o*3Zo>L*K@g%=YhK
zBU<dYeSonsa*^<iP{u(#=B4hy;T_$DOw24B{%jT7mA#hU1*ww%Vz<2xrqUVCXlA>`
zv&=tF(SWV*xBmyOmYp8%)!vWDLddmNCaYN^-ZW?=5+pHIbYlPzkO1>qT=*Z;jlH$A
z4)a`jA0S^R-NPd~J~0vQs3uRG+#@E!SX{o;(*2!ZE|v}pIi_qYO+@@6qH|Q?({-!~
ze8gM7SpyZ3_E@SQ7I8*9!RzaEmUW6~r4pd@z+ICwWQnjU5(j?#Pi#ojb$=H7r=<XH
zp<|c9>`6D2otyjE!&&U(EqlZ^!4w(!u%Q8@e)W;WVAA?+<MjHeVs=na5W-HuN)>sw
z1FuSoyRkjkCvMlCK}3gL8pob>;%F<<!`2Nqxs1WC#8@<`NzjU(TKoX+3M(6C4-eMp
zgn(htWRRc?O;28DR#&$4JED(fMQ~L@J|}bV!|m7U!08rCg<JF7R*`Xi;?WXKBeVEI
zs@VaQ#$p!;)v(p=U8B-X^D}0xut=czk$~iOzB@4Z)`gum*5CoUs%*P{T}x896**YL
zx%Q2B^~nxg=!HL<@h<R6m7Lpw`oE*-A(=Ylk22ysrL5#gDMv|Xttjf}NbEF@WyB)s
za4$KjD8K#NGIOni2seeC@GFFWi;1x)CMHs(vXJVFWfdz0w~D9BAQ`nYT2n~Gm!#eV
zRW4l<2+8$w<e97R?hvzD$dyA1v&1+oR6+dGM349UVjOy&{!|D3u_i8Q0F+pUbKHL8
zb|@%*DabmjF^aXYQWY#-g1xZ@s@XOWu~R`g2Xa=S%SCp&HjK|{YiSq)MEb-NT_@KB
z@ZU&X#(2x-D-e>PSC*~(&1Ht<T*?3x4Y(x=4xp@j=`l2JMoUS4-tsu5eG-l#B`_?v
z4amMSu0CoR4uR3CSQQbyw!85Z29)QmKWYoVp;ZyU1Xh#V^kpn7Gc}Vk0!7%gv@?lt
zAD9$|kwW~e!W-eGmP`!WvHLoxIRpPHC&l9Y^;=u05or$nmbwMC`j&Sy3q3zLuD{ZH
z$!C}-#NVvh>}Rc(O-lqlS?i%D*yYu(8Q!sRSfAhSH5PS^09;joK6q>u_44o}n)|cg
zeCWUuEE-($%|^2K^%k~{9Gc*SSb9p=g$YIbPybiM<pv~j_=IY3g44PtUaGca>V55!
zcM5M06?vV7Lki(}Nv4Cm?S>somNi<!+5<e-bzu>xk<%fek5#+sB*L4c-cw3)`7w5d
zT<fH5XGPdgMcB_dJw{rKWzDSAeX_t{{LicMzt{mrHrD^u+cEr4D>-CA?3u0Ivu1Bb
z2|$H7s0$reYKV<!Rp{&?X759^q8E(Eh;KUl`mpX2^D8kj#aVYqh6p$l;pJiK&}8eJ
zwR?GGb-Vw1Ir&$wDw%-bPe$(M;+R^`aF{F<jJ!Tg^GQxh?D07{-Aj^iRSMIeb;9fN
zohR$;>d~1GI<wnc;eK(sQ}un~Il~ZcBrZ9KKuAtJcMzp^BOE>f*@o-)=?lJ$0Y-%D
zxt)nHUAufP__}A(>Z$w*x~k1-)_Ty*!JM7>jgI>E@%`-ShRDizcHL9W8xw=w<r;IQ
z)A~__1(k3sb1_fb#@S7*yn`KehR&ZR(%=x2rYEzaZ)@hBj7zxOKIZDon{}~iU)H%o
z-3FC=Gg4x&xLZNGy{e*~R=ddw(;1=Fsu{gwgNEO@QRnfQa-`$v29+^5XKn++ta8l1
z0F)eHlqDYobzY>60d@+`ebt$*`XR4b-Tc|K2-k5Q<8i8!{Z&^{8!C_9?eW_G4s;s(
z>Cx#bdC~9tAh}_4*EA@;{Q3!)#!ay%b;4=7Kh`Bg&@XAbGZR_Pb5^%0gUkG7`Hhpf
zXy;Q6e%fNV8dTk^TY_P!U}eo`)E@;SSEr#Y1f&<x*6-bSpi4!W!ds?j{a1Cw7ZeVs
zvTAe-V|mN9)ME{I>vDelQSKJ&;f;zWlip+(D?xAA0CUg%$;zjamD0_~jv*ZNHIHjg
zBG&$~DM}w+5LqX%SI<pO)`ag-vsWoS@w3yJyX%|dB{27P@RgMH%&+s*)bE&dbe|1Y
zRdSX=z98a55EhJmCD65E6`QZ5MKDbPQ0P(*DAfiqO#1O?mkX?f&tE$)%^OQxL8(k{
zcww06mmeFmc^Qx)o>u(s*k<d3X9JP18g^7<)?UyfH=!=?S_b@7>L8munvv8rJq^t5
zMQx+{JuNj%;Q*v1B)+&60~&UFCbom~l?ii_9g_d(PE&k;#!`x>MU!pn#HeNEkZu__
zIhwx<{+;)B#>P||u_X}h0$KoF5K$GUppZT$E=5epztLrpMWp+4K9nqN?psf)xOT4@
zQ3o0?YYNs9<#fu<R?$M#@EW)=Cp_p497>g_1jL5Zw@k^hN+JCWRM5Efff1hYa!=T;
z`HJ#lUG+Skak5aeUk#O@tJy}_1kqL-P5lyx^_bVsIr<(b>o75xxL~J8`kSBix#!Y_
zCQAs+PtBnEbI^69L-nSU*OMxM-WR1ct?`u6MP)xo+=4*R&cThhT{TZF2$T=D2-Mgs
zcVInZSxgZoH3Oa~GGIf9U{fvxtx(ZKK*a&63lVLmn3$l04`Q6pOi3S?a)%uQXtNsW
z9!Ov-;B0+{)=G*F3w1(~#5p&pT}O~%F+|LGyU*AM=&*AwaGTj-)2x;x_>I*Tb;`Dd
z4E4b03FPlANCkyC1Y1Iu2{cOT%p90lA`Z~sfHi06_y9o5(<o*cXc3~Sy3(%(=CZyp
zhWH6LXL<5|c|gkE1eK>qGh%vzsA@CjN__SJEh23o`9QG?9HQKt7WAezfe{Lb`9mJ@
zt0o8cji`8dKznp-Ng0GxA^}_gXqpJ^YX>CH+X1DFxPBpkLI?AjEC&#@nY^aj*XKZ+
z?K#o~6LH3iq{XM{VB)a$?X^<<BEA7EP74Km+&|QWxlh|yqjA+c)%>MF@B+ejx<P*f
z5GwH?`!M`LDhCpJtAXI9)RHN+R^b0-)*V4U`I5GTHmh;4V#>&Ry4Ic?he)96SwU<r
z(Y|j06DukDzqM;Po|HM#-*70E<HXB-YwIZx2!~6_juUrXt~J*0Ad~ItSfL$w@v*1v
z|7L&+jhT6R&5VCB4D=HaJi`#-Fz9h7$N1SXn$1KC^C;IW$=fv&7*u4T_KoUtG0tu`
z#bw~*OVs%^;Hm?I(bI1&L<o@F;o;HwYi2>|YmdVd+rdK!|5Y2JH*Qih?C{3}`m@d#
zJixu^vQ07NB6nbyls2dYy?j7b!V{zwt|ao6knmfJn+hKDApVe=9Loh^g_46DPU@?j
zh-Jr!c|j>=qdHafM#VZ}`bEM+{PYWmMH&yGayOplD1mScRMd1lzkajDyOeJu6~EqG
zlWol5lFux9u(Gw~7I^G<8!c1Z@MCT9S9({Vb+(qhtYx_e9rSq|g@WR{DjLKb0C9~K
zQy_AJ2(gC6k5>tK{hTLr<hQ;#K;2|H-MPP!#%*Qz<Uv%isdQQ7r)wV|IEY*WO(g3r
z>1w2eNoi^rFv?rqV034S&f%C8S^&qmOsh6HrN@HPy`n7FoXmTrRi{zv06{EgNPs7m
zZpXFLqcdk*W(li6j9BqvBa#P%+H?EGqZk4B9vtB+QHu0>w7vzj&|0{TqSxC&b`%OM
zaaa#aqu=BY#M!uu`d)E4T#NP58OP|m9*vfa)wFKw>rEr2)99x+<8`6Q1i5vHY9{cY
zVs#EYv`AHj8p5(*hS%m91XYjLqcxD*hx2B4;s~-swJu_Vp6Zr1g^L+rf4Lawn_J1p
zA0S0Ae+eaW`K_2Bjz(uYB(B$PGw}6*D&>V6w;8*}VUWyeLi7WICdG^%&1o`%gaV<p
zGR*jNH$BMM#l;8*`fB8`r1KP29y_TRS&S&fS-`u+m*=gdN9{OGXH8Z*Yk?G1=T&Nt
z_>u0{9UlK%SPPiET#una2|aj#pNS%bA80I2Qa)UcDRJLmPS4L7>D}O18&Mn2C(9I=
zXi$87d1+QFT3lO&^rARYB30DaGNJ-Y&Z1R4eZ`WphmlHFHlZ1Twj%e6Il&Dl86!wz
zHNd?X<37nnMFj#_#3GO47MbneBWbP{&B!<?7QH5+grccluY;comJw586<=@i$}d9k
zm&c_lDJ)MG=EPdD>7K3|)Wr&7qS<E8l0eu{AX)(VSH-+2A1?~Y%?=4)%^=R5SS4Qw
z*OF2YW|OHy%fXng8ov_90Q6ws$YMQ@GCOj_5_hD~KqEm&$T}+|>Q6^_3Z724yjo_(
zXXU9U47$ijy%>FoJr@>U-XS(dj%{b@MeYu$6jesAR_#xK*4-z`Hp|QOTvVo`1q)Vr
zscXx?rx$ezUA}xojD*(hwVRY^o7}dycf;;)I#QLxxTjfF1I`@7PVY`{$X;Hs85&X?
zxI8J|Z!fOWA0as<05le<DHu!w`-Y3oOx9mu8#A)e{{m^O%>Q9%VP#?eUqRZjx_azZ
z8^S*W?!4_ho=rxmkodTwl^ZH%pN^4h0_rVKUFY()gj19ARNLdF;xK`8Ggppq79O8o
zpR~fiXxFGJF1wfG#@+O*t1ZrrjxVWOhq!u0@eUUjtAuTV93zfMX6WFFAs$oWl@F#l
zPS^MS;O@@%(|0GF?e6yP$KC1bZm>lMT@+d@(^#={d7rrJmUS~iR3u1-7vU8Nfj)s0
zD;%-ZPdlTwj%rE!j;%2o*sb(q><7Z8d>wCQmxcM46%Hc?^WSp3q^vW#ZD3?QhKN&4
z<J^y~F<0rGRHFO1R_(-Fk1d~LIYoblWGabeExf9Rk0!f-Df9Hs`cg+9T0Y+kV=#?6
z;EALCA$Tq$h*hE+tEwobvzeDX@wp6!!u$oMR{{tbcVkG5_BDSGQy06k8wscw@8;3P
zt^#&z>A~%Fr%IB?(!~h#?3v~EImq-N%>HzVpgv-RL<u+E-vPoKlj3n9n=vD-5Aj?P
zc6!N#rD-w2I)Ft2od|)yhA6`U1d;uKU1b*9NfL|UA_mH~!v}0TR;Qd1&b4qj3FRgP
zF()qLy;sG>vRR|YXLs;Gj7t$j?zY>u$#H;gzv!s{>aWzMR(;n6cSraic=z&l;Ye_W
z0Vc-ygVd1n^PUM%FHt+zI)%t)K+;T1RD-y1z<Ak40r>O<$isO=Zx8pNrR?@lq*k!Y
z-5}-Ae_XlW50oT(<tx5bc+2DlZX9g>+FE9sx9rQ&U9Kol%7~ir7%5*e+)2iS@$*lW
zoj2IthK8fr2~pWpzY-?`yxhH=sP)>>R9*qRgc;X$F4X4ZLMvIY<8d&T1dYMqr{&}5
z(+|Y1RNZZA)4uN@2mdALg4e<Te<rpp`yJ70WGPMH1hXG9fNfdKAf8x_?6__qsnySp
zdUM1H%MW?zN8DXIsaqU4S&)7I%wUKJNARkloj0Dg%{b>z3{RYcPa*E3U4}A4eh$Z=
zR@;q!MrAE8(G}$V>;Q6*cb?M7ZFCYS7@*UaM@zW>-8lIx@mK!t=lZjCZGk`C`SfFM
z!A{2+89CG^x4r(Ryd*vJ%Z+vc;JIfP$of2z2A2fiy^eaKCyGOPdw!#EeY0=?ZG;f^
zcp1Ob=uVfJ=z>ZyStL?pINwcY><@vPu4;DUsOBOvb99FyU4UO`V`fwbi?!&Y_daz)
zLeB<)bs@Y#!xUu$td|>y>GIGJfti$Z+2p|mb(YN2zs|NWeUzw*u1V?m-^8y9ksL_t
zNyTz;DjLXn*Vs5Ay%i*Ch-_EoI9F=slO$xY66K*Bwg*xb{!IU|L70bjCFB6}o5}Gd
zEXao45-!!5)TN0sd$Xy;PmiJ{+H%vgQCy_Wo3u5m>13hXtSs+sIX7g!5ICj~H9yCO
z(JAUcwV77RerV_5@uB=Jt)$}yKl{HJ^7!f3^Bxn7bY@rha~P6Q`HDxzJMZP{(g_p0
z?aZAWuK`vD19B@a?&e`(>pq6vCjQVJ%GXCL%Mf4ef&6k@^8oY+<Tf;^Ko27vF@=X4
zS0p^zdNJ)m*@ONa)|c{xx&<Q1-Y)$z>XC2p>K|(yPHR^C4g`w)WI@AyI2Z<AVAEz2
zidvgn-d7hLcGKS4Iv+E{Y5C*Ix;G6WAs2+-E#B%2-Y_J4*Wd*#yREfnm5rug-<uTG
z1k*{9<PPQ$cZM>`*2a5$SXk8+cgA82XiwrHgRtaz=kY)!df9wWjcC@qh#h{s17zsb
zK)7O~*BJr&X0;6}gEcmG{xMlKb7)S(m)XB&NZ@}L^yO9zX(6)?JtLJk2zm#o9w?#l
zyc8xiR0$hYG~sZ>Tq?XnQli_Zan#oI*L4)NAy9zBG5r7P&C<Yh{~(WOf(T#0f~}0}
z>7nJ6;Dw+?19>SiHCLVUl`nGKg9iYbG28UOR2z?kB?T`OcepdT>$|Lo(wJ&VAeVOd
z4}N&z+Zak3XEsL~wDX>>8=fGM=t|ZLuG|2s*)C4bh79WVQN8L>vUqm6ePb6P=UHM1
zBT8}~p)Y|JFFi#Yy=>guQ?i9(nEdmtz5acuWJOfd$?E8Gb0juoBZ%+B>`utwbAQ5^
zdB}s7hI~!o5QXec@PNHlJeaFLo@?Rl{FladR%Qd8s&8b1&v0*j@xgbV8W57+r44I_
z?)S7yH9oBn?muKPbxfguHL!p5raB@l4{-OX)IOp3L<1$T<7p}WY0pH9W5ElROY_(J
z?d^4cQ(GdJG-6|fj1L_Om23EuKcwi593z)WAz>I*W2lY%ocN3|)3wAWSl-T>TpYKy
z!`{Dc+Tq>C5hhq|SiL+h-vp5`Cg;~cW8P44BYVnAwS#@Xk7pk&&w(k1K5dGB`DnoG
z<kl4*cM$Jj{2e+m&sv1E>Ium7w5*{gef-CZLBWw7N{%$Yy)Yojna$w>CT@Y|v1-0)
z^<iN-yR&M$!?uH#N67+Dp)s19*x87nNf;Y3_`>7t-)Y-hn%Kee^HqRMa3P50YDe}0
zN~-Z3m8R;}=-XG>leFL{C2iG71H-?j*rO`P@XkB<la>Uy5ct>r8^X<P4Nwb3Yu0GW
z3=EReT74uXx=)Q0%r%c6gqpJ&47x=Ng<~h&{tnSd)owXhf`0knj98whzXbNUk$EyQ
zNbrM$gTaCR#%qMzmlWj6qhoe+M@=J_(iEW7g^<F+fy~09yUe?cQ#M&Ywy^iQbK=pW
ztl0gSbBGH}T6K^Wot6<}8OnYN^mfo25+MR>OF@!NekZ5AlhL}T)a3YSK>9Cy?+;KB
zYoTq34@s?P3vZXz+D;Oo;43Jc*qcdkO}hD~tqS~e?|l}%vp6&!Cbf+F1a6}R#Asc;
zw|paGm}k1dr&+{KV~uEA9eiO9$Q|fh&5)%>rwKBGJxO7(vPs7Xd`o>U9mr<gFT)?l
zJ^vTY)M+Qqw3#Xo&H#t0#b9w$3?UTrEs$K6yIA4*5_ap57uhuL##Gz!fh02joGu?(
zhv#-7gtXB!m4>%myg1>Eh0j?J=ZDI~uzRPTjqX7%Oh^F$@;<q=szFo=y*TRYXrAGJ
ze*@~0jf=O^9E!Zpx2@e0KP4)ej~{pD;*hSPu5=pvAoN$vc?h@a?@26}Me(@Gx7YA-
zVBrk98f@rK2XidivTM(Ne~Hf8jYD+pVDFPCpqa5{t5`Tzi9ax{ob<WRSVgdxh-94{
zIn|iAU`QHyBxrw*&sjkq-cpB<1)%690buM^<=>n0aJWz{+hnGqzxUWD-xSKeFyXr!
z+TT7q(*5s5{p#Sq?TJkDoBY(hB%w&%rjIMs%BH;XzHnF40H(sg?=qAw%D+i%xsX0-
zZS!w&Zg{q(`%tL<28O>JZ;$G}p8yP^Q9~x|jmuB~*8RN`V6v`%f4`x9di9Z`4ZuKD
z?4S3$Dlh(`$ohVs{$Lq)71mT30C@+LRB%0guAP<?vvu@x<nw%|{9xzJ-mYZbnuc_R
z_<{zX7DdtT2*%a}O4_y3A$_q+rc1Rq?LsHwhsz3mh=HkyCEeoKHVh+m1x=tN&7xx=
zW5udbi5KBY2hl^EW-S88L2LfP_L1IvDVF{TN-LCVDo$P_g?Y^`m+{um;62zL=?3ku
znz^;R`qqcfUKrz*?aP9!0lrppWBLP*lfgVV$u!}JgyA><^;RGiN<V0qo9Dv`*L+MH
zw9^Z)jUL|rUl93U2NfOO|6d7`GPW^wGQ(qFVEmukH@DSyZMQ}cd{%YsZMgw^o_~%U
z<d6Ev50-=tJ5%Ng!XEjy<v9vmMb(eQ?S4Ld8XJ$|p@}P_p^DrQQU}}D*VAo#&=nl)
z8{;BxXM4K2Jl{7mgyHcQ*AE^%^c#X3)8kgCX;>q1Qz_0Kb7_7b9-0&M^GX~D9SZe{
zEWXcYYj<>gUiZC<v$ZOuvTC(RQg3&)`QA>n6M8*9FKSQuX@A}CPXerR#*c<4SNUPA
zYB>%muZ3|Kb`HF^Q$kxPT&$|pQ9Bb-U5`)NY$`wcVq#(lg~#ecwaVNWAO_w3B}4mN
z!M}7sJH5zw9t*19J8I#SHFt|kHygH7&P(<tJv5cTb!;L-P)you;9lhv-ycf0qOSM<
zEx_?&S86*ut75-Wttm-+s%m2wzRD{wJ4Kwea<kh6BFccfWKCIs(_VaABe+tj%QeH$
z)=XI2>&`=*)hyP&V{Nn_Yum3aeQuvVabltRSNo8R78-Nt22x=hr=)IDq#suvKE~<k
zwk~_m2m9m)Edl3x!)vg=yk}(5x1a1T9DF*o(!zFyuy(!KbyXx>6*?&d=h-V$wR7Y9
z*&qR#jq{($^~q|!EjKNHfOf9DA4l0C&Kygq@lkQS%2THE!OjDOX^c(H2~*CJebEKh
z@7HU?YBTb19-J9sP1aM6mBXG`b`oskfvS@PNR@ns*)3m97O*?)kyW!enG5sFOQ@^p
zHiqT$%K;n>ODlBI-c%ga$+*g+Qe<`><eRI#$Te$Dkd03obQ_yj<hGe_AA0Y#HVz!S
z2Y7C691U(uJwPu!u>s&b(*U6joB>5fGQ&bVlWa4#{EBq~&yS1MZ^Gn~=<8}aEjA3~
z1?k2`hn<yN-O)<Nf@Ja0Rtb-E9Vc=rbq?vMo?58G5=us?Ty>%qVugswgZ<$TsBQ+5
zx?7Gj@c3*WYw}%MKT`1~BP1R>&LpS!sN?^0A-*8EUE&qnPXB|9Q0A`gB=PAjN(B5b
zFp#c|(Qq7@2>t|hE0hyF{4WhTBI--eQ?^@mX_GjSE(ErUDIzh?(3_wsWL-~2X2y6_
zmNeENbcX%crG1pZl;TT}Lb^p3fdUSbWUEZ+xft^vT@^w@O*K^7JrdWCS^ZAX{<Zca
z$9JO?`k~TcL>MW6j>dkiOc<@)Q2)^)kRZTyb+p6?aKL$0B^0x#lh#Gl08PUYHBK9~
z4+sXjPIM~)%oWJI48z9NDapwf<ncBEc7WG;6{6ODoSP(73GaL42#W;=h;lXv3gttr
zaWYq4?t}Ln2=L07War69;6^LC^TR(?x?6U@C3DS9^Ol8)-@N<llYOMdxLsI?a)m;(
z2c@&Dz#w-?p>m0I{OJCq#416@s<i$N7Vl<TE2mrd?#sM|PKAa&odtzdyTFl$ot3$u
z2~pnyVs@DpSsGp^xSoJw-6%~xhT2w{k*!0zGmME!b2PobLd|o05vCUH3Yx3I@vJI~
zsZ$5(Yws%^MTK}p6BQ;gOI@c)YcK*pb=2j63BBD?ZOSoOgQfMjqvL04Rx$rB5jLQy
z=fvI$`rV9Mg$qSFQ_AU|8)Z5n-~%+a)1$`_gH%`m<C*6|*+nazXs<qY)mvDU!(2@b
zYtsV`P^6L#OJE89Rogd4zu1%D#=4=Ee6<tg3P$o{P}+S9jn*qC4C@;}3iihwlU{KG
z-6KmLu$!iS*2&312Luic&{e1WG}ZQ{>S21W!<*+tZw<&{lzf650JnH4)XgJ|ebn7c
zHt#c}BKd~CYN~AoTU3&mt}({-Yu}ud-Z)|_ecYDAXLN_l5HOo|Ee=e}qy-GgLt=Jr
zsx+|?CtvO!RmBKJU#Ut&f6KRj(66=C%7MaS^NC=n5N|8WP)nv(U755%wSu%jxGAGD
zeA3p`OC7zm-~_TBlE!_ya9>IcnqutPwuv0f4u*d0P?KyJ&u=!nZlMGp_=ppgig3<=
zQ!&(CvBeW7-E(9NlB_j{24nl^KJIWC9=u=P*+SM_LQ<JAx)AJDW7pb8xdqPHUCo*B
z-L4`xJ%S-$(3g%W9nva3B+r87U#7ILFr)82jT-p(Zc|Yx3FN6f#QaFXo+z4Pcd8?X
zC$SwAZ__&`%$@i1=HY5lotN5^8t9+%Z4Dhl`*zmnOPq$qO{~g%?^jT1TTnQS?c^E-
z)ztT}79VNO9XRehxP)YWm^Ha1<fP9#<E)B<Sb3t|2+*iRHAA|yK|r5Di6@3^5GJ`8
zMI!I+D_aPptx18yhkD;S+PP6}0mxrde`d6^A6~*x$wc5J5xVa?5&cN+6zP^!J;*oW
zrrUlI16>%1DOMH6qFbG{j+0k4Rfzq_xAx7<{6a0mq@y|}{tQ{P_+6fY*H$|HbjdWO
zvi&2AU=YBWMaArgi30#X^*a~T*y<b^`Mi<c^Dv@*hL9$mRk4zAP2UvrtX2Xv-!#A!
zG+h^_>1`3NyQnZ}t`mA*lp-nAI-TT!O>J))BN_Ow({hb~kE~x72u}S1A)gOps7Mln
z3u#MpU<7VsbqYV5s4AhIdw^9xH;r$EC4Ha1haFV#D`_$u#1AU+c@Td}2=cp24ncTx
z?s!~S2((oEhkNuVeaMqy2v&b+$t1m?kNyf>Z??K&aWOC%JV{21RD}0$_k_YoqmuB;
zrAqv_Gy~e-gh=R;Tr#4JwkJEQbZR0dUFkDtfx=&2dGU%&C0ttKaa^_}k&Y{)2((lM
z8n2UFPV1xDT$Zc0l%Gp`Lk+3IF%vhqC>@H|=DCub$d)_aldE$hJ+m@2*!QiixIVo`
zdq+>Bhf%1+i4VWW()Ux?A4Hs2(wc?Rou^s9C-K%EUhf|&aNLhh*138{KX!k&C!M?4
zcLLx88PDGcpusVSZ|Gncyx)rtg*^V#;M=$NS9zU7M_3Q#hn{rI*#ol^k;3&gL#BUq
z=Cynh*u^)dmFp)VCA(5EU~NCUL)(AVzgbP<<^NODW&5uk7e?m)+424VG~G6Yo=?>~
zvDFN<1L+bAxST7t^y;C+DDCyZ8630lhBUb%k)kU%UvC}&X^ABB*wTh_Eh>03rjK`L
zV41>CvgjYDmz$TXPs*GV`56a7DW}eeqVOgLDrJriZIRxgExPd;B<qE@kHhPKHumCA
z$It!4XLkZ83|;RpWH8oL-|tP`on5V<THc@chr#DNd-rccw6C}E+n@K7uhq@mo?eBA
zH=|<p3K$S5@Y4l^>|!FzBsL&p<)lB84O<`$qt5M|r|Mc@F<KFXDOr{zN-@<T5SgLa
z+ioq!8xFE%1jBq-e}?>*{RxcCYYK<-0%UeT7~;#rN~KJ)I6Jb$D+&9P_WiFy(UpVs
zIFRD?U6>kIpTDoyQyF2M7sEuiFiM<I%BAQbY$B)EX?siN5hHo|vC>WF3rV+y1R9|i
zO60M(v%a&pIGRDyjuS*D2F4PY+DKsd+=~E11fp%;vP#aQ!0-+dVtX<g(Nd|9WME1h
znc3qgZ&VzOs|ls_s9{O%_nZHW4y-#$A<ZxhStigMhBdLj0ov~a1H%S;Kk#?qnsCMV
z!M_OoAgKxnK-1|LV?^bTqXDh(stMwEEcIPYc5$&MiF}njBrPlPLmZj{SJqa7u{4gz
zQo%Cah>Gz<uOit>sxbs83%_&>vNLXgvy`5_Q#EBXf@vLrNX3ZtE)xx0GQi77J+L3U
z<Y1;=z^Krp=Qd9)J<q9_b23l%IY44Tm)&D=l9UKK`FT1B4P*UnD-CeawlHxpDjldn
zCbi*n>dYB!8l-_?Vt9926|Ap}Fp<w8og=2{?Abu}?#z5V^G8Tc8DElOSJvK>Z_8!c
zQTn}k_0zyi)i2Nve)f?@idF8d$uO#?sR(IdnXGHN!`!UltSYfrwQ6xjm$I)51T`TO
z&$F+RcjZ5+ORN*}P5^KV#XedXj%o-KiH{@$v@&-9dYb`7RGn&b&{LH7xHTaZVgy$g
zHQEh=S;P_|SwoQgWj}QqZ$oqZIS2I*${O+|Xz=puNrazBuTOk<lpvn7EeA0OC|cwo
zQD_?jb+wQeeEhA6flzWP7AA6>tm3Ajl%cH7R#^vF<()lkF#@;rWi5>t*to1!GSz4E
z`X#BBt8{v)P!OSmH9koKN+ZCEP4ZwRq>n(y)rSAa9z%Lq>{1n@`>!c4BRhXP&9KLo
zGhd#{i*#17ifpx9PnOh1a9o|EfIh6i3Ennu(SB4|n>EvZK(;ZahV+>o?We)Q8f>K)
zhwmXuw?pT;%JPGn^=YSwr)GPUad2fLp+V#n{JUyl{qBl#bpl+-hr})EhLHhhF^8x*
zh&`7Wb3g#xO3Fcv^2^Fj0ppGh0r+CK4EyXs*2I-0OY~66<+i}@b%W?}^rX5GMvheV
zs0gtVd!b^&YY)_3qS;ss-Dg0Ck{6UKRTX9JnKmAlw~)-Dq7735oFK6lbFkfq8V{w6
z{2_nXT;M;px1P}!mQ@?hr(0@1;5Ty<ocRa;PA~-f>Js>8eBP}$NgpJZS<8KVl)i6d
zp9oG1m7S1JYB}y-9Ct3zzo-)U@uBQK1y4#D8x~y~=%P9yrg)vZP#Ol1^0X!{Hi2Ce
zRl|W|-lYPQ3DLw;6Nueivwt)=h=~-+Fm?SwOdd9njtDLpkoFMkS2~sU*)Z{spH2g_
z2}LPA*Nn^Ly1_NCqqzue7>ric2t=7`9DI7o8qE)NFH0eo_UP=`J#Ki|sTh1Mp~4_Q
ze*;Giy=VOWSg%M=TtOo$gINHxH9uRUq718tbISsK?w*nu!d204u;zO1XY)A}yj+6P
zeeE2$FUt)X7Pj-8mnSm;NcG^2`!uoHyAqOC&v_(NXQuH|fl1-cA}S>nM>~8X5+~IT
zf-{pSx;*Hw%-3t?k;+Ga2^JWMHP&8C>GbyPaNYlAq)QjF25a*lsu9xo>;$2Xibrq-
z{o^?0&uwC+KOQA8B&~imV7AM4>R4lN+Oaf*n^8Aci(vhCo;IS!i6eq*h{>2%h3p3#
zrd{n0zX&^36vHY~TXyQAL2Uuv2q`fycfYUqH(t7ETrGmeY=lCQ9AK>7*Y}QS`v*4t
zAbuP?b^<{X-|52vmE^;@ua)a$|FlL6`LhN~#e9z=N_#aF&gh+DPq-735jgD#oX*}z
z8!c#WQ11^9^tLnUk-u+8T-F3oF*!!Q!#vB<LL2buMet_kdj?LG8q+UEM4ye~(z5Mv
ze!WKOa|%YusPlT%_h!rtWk`Pq6;6&>d)dVe=qi>MlKtxl5xMKAMKSEM1k8CY*te2a
zO%|z2jT4Nxgepgr^Okb&e3d2n&hFM+_0T=WB@&MoV(5Gkj8b1Z02)P&3aO|qP^jnX
zEYB#xHK(HuLUBNx&!)?tw2EzKMXh)qnz%6FLk9qu1=W~W+0>$~We^WLtO_c?xr*w`
z7yF?`U%dMGf=isvaV-!zVd<<abr2#am-aMw1UuOvty8&>h5LDkbo}kjX_s5#8KQNN
z#!h(7Gz^1LOom4QIje~{x;LzOGR&qatdqB}fu#y6ILAyn0;T*z8HN<f$P!5~bp2!)
zkEolX;d2wtJOeQ}A#hB`^Xm7y=NDVlxYQU<tV^%?Z~YIyD;y$(CkUDdg}Z$Y{HLS|
z;}GIwXdXSUoB;@Q;RR(-Wc=JvfVGw3fPip0qcT%R@29se%PmS59^;<_L3Eq0sVG`o
zTTjj)tkGFbQP;pEa?Zz~H)wn?&O0pCf+wLBPKrilS@vUKyPJ0XvrYeRg$eT1ZKzoH
z8<r>EB{j6@v=v45LEzOjlc4?<>&tvx^_vJ$h7k&oQ<E{v9BD&HRc~XmS|~Wb2`pxZ
zGccOjM<yRK*I{pT74m+1Nb_KR-FKmpGl23pzGqEDe#xxO5=vLNCK=)Rp=lpSRm563
zh-gd6o~u%E*mRg1Yjm4PjhrSXjxLZ7PpM^j!&OD%j|SJ_P4RJ=pY+3=<z=Sz+|z<i
zh_v$UIe62M8+6v5J{`G#A2UZNPks%jq5M58%NT^fK$|AK2huc7Hd#i^XJQGUwN*Q&
zM)=^571#8=n2h!D(NcuDFl~7z;T#5jEneBi-^lwLc&qaW3)L6JCk!8Zq#w9`;bx)#
zBBR*<tA~??`G3}Ho7FXK5C0V=&DPFM0>Q#N4n-1O<ijl*l%QVObS45s8fYLlQI8Is
z9n>Eco!9QhaurdrJ7fr#qUlj#$A^ZVK4eQCobf~y+Sc~;bb37MjFXwVQ!Jd>V2(3M
zC`}k73%O>V9R{3Ue5`ij?sT2n^Jt7T8)-n&*;n?w-(1VO-tuVC3M10GP1K}{W6=71
z+@lQL`aV6i{`!2*UTRVK^nCxUJak2LBkRGcke*C3>6ecc6f01LcjNiSp$Sxy8ETx`
z6dRNy&MVL*Q);|OW)11+-A$<_*ms~JNr2m9QqC3R9F7^5G;mRYu=;2O+bTHPq!v@j
zQ($C&Ci5!Y*Qc;WI%A}Dzh-4=?Q>1r@ei}f4usdI;+n9B=Q02WyNT!!<wVB>j4PPp
zrGttHWFgyYl&NR%H_zGKsN$S3W1d2otNcu7fVRe)Af?zZRAtSNiM7%nCV;y<c#IDp
z!2F~_U1wcxZ!e!xesagS5xM<q1uY~R>8`WX{s%yrV_4qa1cMqKvAH{nor~;(1{D8p
z>wJd-8CCG11=v&9$<<TIoPTYucN$S4ou7}?wt?er3r?r06G=gok%nk2_Rl$l+$xBT
zrQ&lW88JNHIGY036}xTWz<t-S%}J`?!mVo2ZiEK>(VllQ*5yh1MndrDweDs#UYviN
z;Yyn3=C+-vHm;+_$}KqO&TTG_xEYc7BYoO_M*=ukCN!XnZ*AQ@$CP^oTW_`@XkGCT
zYhbv0d@$%>>hYoEBBas!{Ij+QR8Ob8y0F;7JqM(9g=oBuBQ{{O%YsSyBb6jYqzp8Z
zBRjm^vJ7_0yZZSQ6UGA4Z*+R1hr@YGbJrP93f+l@3IO`IaR`jx8{!)=L@b`_`1}_k
zzd3U5L?P(n6p&*Yj!N3C2`k$}(1oS4rdki_7J69F{G$OTrh)jhqMP#itD$2psiZx9
zQG_N9Y={Q5kA7Td3{VCNg<m%w@IlpLKN+2mpiKtcml>&+2Y64E7?{pccwSGQZT5-l
z-Rimh|4{Z$L7MP@wrF$O?rGb$Z5z|JZQJIwZQHhOeQn$3nY-)0oT~p}pH$xRm{hWA
zWx<m~T8RX#uW0AT18>-gq)3hGq?w>bFJw=7s|)&b4BNDsmK2qPO^T-!by%cipkTH!
zIxj(nr?+p&eCSwu&Dv3e!;)r%%m^5@89v``k;#|ERS*pQ-3^t-Q9!~YG~RIjs<J_E
zrakv$5?BBPIzF5#78K`T8RwRkP%d#}h`z0NCd?xVJNTH=4C<0Mp@UAwjU+2O=(2?~
z{~ktxq9MARc?Mn3pPOc?2M28?zfY8Q41+als<eSG?Ogpei}jFBS(VU<N1BV=6;sLE
z^B6kGIV+zS2nITXFjdN&>IrkCeNek8rlpUt!MWp#6heySZ@iEx=i<urPsLBPsAK>N
zS7vzNFQE}m?cPhF0*uXZB^7u-OT4nBHu?+~!J!6BatuUI<v+fI{aF>+zGfNo@SO}r
zMa%W?B1NuV&lbwAU-^&c_~m$4kt%|e;+aqC3`i$t3Dr!fpAm7F7g?Q#GA2>3;ELHa
zo(ADd`E2FTM0+hd4(<X{J`}o<53tRk69^Ghe}GPUg0n-Z?W=KIWv>;SeKazA+6dF}
zySsYQHe0jFAHQ>q2YtLwd9!JfX)PbTH1Mqogq}{MT9&K!7d09Aga=-kkdHa3>dl=2
zy#>E1+$b(&58`@gk`Mrs6xnM=jlIHLdfy{a166L1P3#cL=)6LN$}ntEq^G;dtv9QX
zVMm<+tsPS5{4QZ3blJANyF-I5`^#K)%T2|#w393@RFNwyc({hRFECs@gyLr)P?vT0
z6RaXImv~3@)X%{OS21uxgrr&V#Ya*uOa$3h(LFPa*HJ{cun%+h@7pWY$Ap=C-uVsi
z8Y2&@bDtxmBeZt{OS9CbX=6I3XTJoek`PFbT|^aRAb9#}h+dr!chlIse#9zlG26d!
zVZX}^+Dr%jK-d-?+_LIlgpd0y*T}d4Q{N+P0&|Ffn4xYbZ*OFl%oY6}go>)WDQ^Ys
z-g*up&k$ZmnkjXz5>3&Wzo~f}5aigN^nKr|S{?3Trxu!W2FmKNnb<yq8A>lofYS#R
zUmy;tOgh>Gl>?~A6I-uDUALV_`mXVZGWkT-%@n4iW168s2(u!N9V+>p`AB_8%@Y0C
z_K7b#$2pr^)mEVQLK0>^|KdN#A+uS`JuKBm{p7~5yA7VG+8~&0)DUp$nf5yIM=K=L
z1)?#&1ij|vEYU>m<)(M*&R6zQvyHSMuPNiY7YKV-9Q>=qDQT7wR8nH5^0DKa`O-s>
z96nEL&bf6g>aqFH-!O3@1nzNIV0jY}*aUu&yvKyi4((BZuBDFm@=ZoNKgG-}0K)yt
zo*V#pVQ8i$T93+eV0%;sX<a;k*UOofrMCbgvX7yQ27BG^5Ip`OZv>yNCyva|#R`de
zQ+chWMPeWBU=jw;omSOOF%!m@eGm!$>hDuiI7HrxO&D0Xl5R;+BJQuO-}JM!NS8tr
zo-LH+KDb$F`ikznmnlG>`ZWp5zM&2wD5m|}%E_6i^zBlHdbtz{Bd0o&sg$T|jtTWu
zzG=q-;!%UI9sJ*arUOCZFT{S~R`!bl5K!ayH!kt9=M!C1>{^b$2ccGJBfRV>@{>+v
zh-DO^J#T?4A%=Hz<q+C+=oG8suadt@g2NoY&dxH3|L({{fvyqfO;?n~7sj8gZ|<O_
z;LJr837$#JMc8$gdhWDxRAZN#yAd%o@aWw450nqK&5kxusfFFE>z!^bNu&~q51#CO
z>mA>g`_yS8H#`9u$(6oImh%1RXXoSR;%=38YXPpKKm)|j6Ih|*|HAI=nX??pP&NlN
zSisC~zBha{Q9g?{raxHaFy@@b*OW;(jIK>JY&Rcu5EMMIe01`%sW`U%4mfW;{7Yhr
z9`F7vy*_U@R<%;7s7RxlvxbUBVAR!Q5iuTHV8SdDB(6#>eq>7Si<&;#Qyfnitn>TJ
zNbxlvm=%54abz@EI+i@uo^!$G{#$65uT{09nz+=U6bxoHwRD0kVBe`98Q<gQiBu-^
z;>rL|FFtsn2%j%AVE@6}sY)$BC!Y$Wy;nRTXb7i`#mp;^U(?rX$27l%zuHZ-K~^oi
zVtCrZEsZrMOLz2?WI%m`wRNc{v0&pWudb%31KMs3l_YCAZ<-{iLb<N&WFThsY2>2k
z1@n1*FQ|ROLYC*P5xyqgp`^(341r5~UAwB(Zvhy!EdcjuJ?V_3)d|vSUAf@tjAPai
zkL+WRheT|<A8kfSf0dyWyYF{@nk3(hovQ>Q`CnX=qHKO~t@J}<bXD0oIQ;M1eqpUE
zauNv!-1+Vi&gk&-4tl5z1yT0xkKEJjDmKlZ?DEM9@!(Fuh-^l_e6otKc4M6D{y4yT
z<W8$)T+Gj4cx9P5V#xWZTvHi~4_9S6QhjC%ysE3)W+5+Z$bD%8CD7+1Qvy!RUps3X
zk4?Zi7M`>4P5#m#F_*n@maWo&Un6xYuH|k1kmCC$!j(7VJO>t1wu!!qdDaJ$9i%BB
z(q`84Dpt1Z^%Gt9i~sR^e&s(w>wk9p{O?)AYK?#0KL4B`AOBLbvZxS*5qB?=9J8is
zT34P^cvbnTT9x9#V(N~C_l!`#KAa1&Nhl`e;Jq+`WqJN_hU{ePJT{5bJ9?jwulr#T
z2`ZUXHTNt#Ja}wOJH}LTD9JW>FPT3gL~$nGZn;O2kJn!ZI{5`^uP1lYA8u|h7AYbK
zH~e$E2bHZq->dZadJC6N{JPy9C*-x;Uot_t{B2vgeA{0yce(sL6qwsp4IJZ+`q6YX
zOxlk`2D<E$+Q|%3Yy*sx6Y6D_4{LoJcV*&j1hNGiY%6f!h;Z{$H{^}j(>bQI9XYVo
zr|=>fYCTsQ3BBDbK3jL+22C+|oxa`Dt>{?lYmZA8Hq%-N%AMFYt3)5zNaO{OYjO>3
z!Fr?5HCDN@e&iFU<FIvVzvsi<DzY?$g#6URP~;_$W5R-Djg>^=#JT$B*d|Atw%nVp
zP>`%XLSAxRc4Wx0K4R!CI#{`R;l44BefG3enPwK2Dgp7F*)8iM%=0+6o1hF_i2al_
z!o2JA=NV+6;r-Z;G{7>hI!+bS=I1cevz{DWi+h5L7=I6{eAw1EgV>8yrE(rzE5tvl
zytsD|sXDgOS+%yc2jOzTE=VV=2%{dPtW|P`+NmT^^1hQx8i-P$Y9tSs)S8Z+sF05}
zOZwMeEV2$Mzl(MWV;>Ow283!1Ra30c6cW&QoWn|}6_@(2W%=ilzRH|EuYO&8E+A#S
z>buJ>ws!n1@IGu5wsc9gdg0B*sjW;_&OYxx*2O9oPpC^_a>XR}%R078iiz_w&@!y#
zNYtgOVRm!-zF-_n&?%0E(d|2_mqh6_GHp@(L)sV$Gc98Lw95Jym~V8$Q)&U;g0GZ>
zwK{Ms^xKCe`3xeQe`)OVp4~W<`{=_hyxv#LJH`{WYJ?b*9TFIV8NAf^X6_ogwL^|C
zblIDZzZZ4D(N{k_M8y@?hIIKQ9gbdsbfn6L?=W+nxp6vr=5^OSob9{Z;HCu#^WfFZ
zv>MEi;K;j~33q?bFK6&*A1Ge=EiyDL67>a;s8uBRD+OY};{M(%Dctcf!${r@w(9DD
zLc4<-O$mvd4E8xPn}mo5(c%#^r52Cp#CQ|}ZNaEo+u2AH_r&VdSp}_+l(|~sYO@Co
zlIt45JB4hZK|D71uaEoXwr(Hg<4>+?GSu^C)6l&2oWWF7aco|>CM-xUU8nkQ&X?ze
zvWDL$l5YGefE~51c>BECuLmVShxX+rF-aLJK;<JDJ%Dv^BH#kulFl!Hc<jeo_;nb1
zO8Ri4<P4;xE1*V4;F~Qc_A*sy0~!mt9EwP-NY_^TgC!_X+yFF0AE6p1$P{#NfpqB0
zwGX@LPI8{E5+JMs?wg@W9yc$%szex|5)G%#ni0^7h4?9ij}B3~Npx5eT$CW4%EDu2
zfoHa!(fR&(QQP`3Rp1K@S=d9kbTua3Z;Tlv4C*LzxD-v_E8BtFkMDgueeHjidr%Xq
z!}GZzjHV*BS_USnkacVkE|msooo#vO@xV3f0m`E~(W*c>^`_8p?{9Tse;@_I7HQcw
zg+OT3RiEr70BrkRj*Sy!Yj>?&UQcR6Cj1Oij2)Eqq&%7TPM}&8?@z3bl}U%Y%oX!H
zq=ja4#-u>2%>plLG>0JX*P?16s4w4ryIQuWcz_)xQ$L3CJ8_szvZU6-N>_TNQ_q2*
z?NZRqC?vmnxmFC#E9PE{I5W}uaq8#C2F>cD8p4|~J8=je-=?O0++Fj9eILW2%yt=-
zzRmn~%DPTJd;e8bg2)s>??RB!4yWQzEaMV`=|we9FhSYw?A6(L5Mu5<JpPz-&pDp>
zFn^8l_<TH61F{1-7FpbZ&GKiuTR!$f^$y~K<)qQUl=oCpNn-@u>l|;Ox;5TRV6QL0
z{!IBTDPFQ|1998}f)a7ktT%EKP+j^ds{Fu>!#$_ISToFF+ie3o|776oXl!nn$oID=
zP%+E93o#!U=q#x=5<WZCE$?e(2k()<@l0?DM<?-BS+P2`0<O$Dw|ZNs0SpD$RD|yC
zaJOt_XCT$O^Dg6tv^<rs`~f=YGn60dH@Pa70KJ(h)>I6R7;RP`=PkJ!exKVIO4bEU
z*j{laN7$LG+_;(Uf!nmHR0=n(*=V1>Et<O-E2y>=;kr(t0?2yNE{$P50|uhwT&K_q
z;<d|1tAhn-X}jQ~LMt};4*xsbw$I8FI2OUgw~$Dy6q};dX*PxypX*Y2?ulCv^UJl@
z_dXTQEANwi`PSWsKMG2jIHbzYSYB-|Hi#&2y1v<$W8KYDy8JzP=yj+(JZ&=EHD~^X
zooF?M2u-A_(-D!4FQI-ZiGo^|vZ@H(Sjb37w^M6o9gwqak|@5+BJbh@aOlXT6(S#E
z*501JTwIJ(zD9_Qsfo7fa-?Mc>ec4a=9jT<fIwS{p#4^6aWwU(dfx`9BVhj1Vqq>e
zdgOIESUfO9w+LAqYoK6VTg{dlHAyZq>To?$(D31pbNf2_M2{mOs4sf}v{YVhEV^E7
z)r5SWcN+w-@26O|B6t<_)PuG+@0u@$4gpER%|d&0NXqD|t1k@M!7QDD5<CA%4(ZCI
z=xaMNo#bn(Q-;QNo>VO?Pc7dC%ue@HdWF_4yk@*;C`pXdvniKNomZ@@!*b!)Z932@
z_};tpI|DhAx;H1IL>qmKEiEcM#9&%#VvfDv>TSL7AZ@&T_z5cbA*p6>h17#=@=!b)
z#x=ZJpAR>}?!VY!;9a~@a)iV+OC3@I@kbx1$E3?ODBdL%YJvV_m4&^7N-xf;g0S25
zi;7H2De?uwXxqtZw-FPnBsr7-Kep5;7GYjEF7rWDwV!6pE{^p}l1MFv0=BJv3CxUy
ze+psIGm3{R=I9nvoLX$8x$cU4+4P}uzlpc&hs;mvS-VEV3K}_-_60d_IBd?O22#B*
zL>^%MiH!4=SGWDVO;g!lA?(wq<Nh7R%&qI)zW3navrblv<wPmT@=m`HO?sP#keQ;q
z4w5CsNbQB>A@gB{pgph4_q9KFau`moCI0B=8xEV^21dZLzQeRca2(pQkN+WMoq1xA
znI5M^&)t{8?6x?Ylzg#Y3jWqB(zWES`A*&S^Yie&gJ_N}E&C4u#=`Jl5;k@Qj{mL8
z;~#721ZeyHFQ{vdm=}c@tPh66M{0{?ddAkZ)0KNs($!TTszqJ7fGE**OK%5(0|J$j
zT8^vHr`}&!8-C`98-Lf?p;ZF?X7=0L=lNZy1U<P7T9u~84!uMDP=j!RMo+sIEf-ni
z_ernL&+U7*OyxL(MH7RjK?1}1eXX5uyYI)vTgb$s2-^N95&gnpYHlsT=PmQ(j<@&2
zfk$zopRP~ay;V+e)o=o|Qc=TDm@{Dz*8*g^YSaZBTZ0a}wRRYIBhlGv%S$y5t#|>z
zp!4#=b&}XtySuxPP#W?g8IU{4<UpOM@hCH(yaUL;u<R_>?JOm9nf<bP5UI{D{37r4
zl~ZoM@${pp?arXcJ+4z0oOp?=ZWT$l-iQG_<GNhB4#(VCpE_d*Rex~7o7<hCgS0xm
zHtKMtX<ksZ>pq<7XxVw;jY!w0e-`*~_`+n?RV?5*PgBu47L$c;%gS&zE|1h~ZbFDQ
z{7dizH1x5>Jx7?;0!)uo|9MkEwcO2Ke7bS4Xtd7{RkdSRb(R%cdN8|wU23)+wg)g1
zhe*L;b4=06E;Maln1DMucLP@YGvf!1b+-5wPn<GJaR5s6u=}-hgx~Hi>C^^@>5a4*
zTQ|LnMF(jyQsYc09SRQ?YtjofE$ag03$_<{n|zj=+ipwI0jIKtU9ApaodtG;ICowb
z*8B=IZlzU97zP#5s<y)Ls9a|@Th5y#y}vaYU<nZvg*#v4Xr-;!_GjkN4uGlisoKY+
zD2tN31xO1Sfngv9wJ8z?FkTfU&VkT+eFuF@5Jk)XnRN5jCq1Pt|JD{Jh9>iD^PNXy
z64l0{eZ4jiRZNMd;7?ne%Wi#0%*Z<W*vj0WUiz?S*jSQhTM;f^{+N(wrw-_^r+f05
zLjQFhbd=><|GSrfPINMc!xcC-L8AAo^Xh_30I{Qg;kRR1y|j3NzrgDjK-`jN&D0*(
zby<nj>Z9*qj5B$*UIh>q@2gHJx3l))u;^fJ#@;y+18SBMJfT6RqL?`2&mnG)$^_MH
zvKV_CybU%nZ{#ogw;4jl@?B>TUX*mALEXI(-T3+DPT|f4fsj2?9j$Zwxv$>SLrSt!
z8{P{5+AqJf3ZsEmK*h`gn<zW)d2?QEWByEB-F#jKxYtlT!oxWkbYGM0tp0G-L%Hl#
zu}L7Iw5Cm_h~)WIwUGLtN;tto<`i+KAivdltW9<$JBPHTleTcZ$tPfoW`(;NbnSof
zSZ^V*zvb21%1(%NyH{2Qs3)P{s#KoOE)hD(vMVFtL#h}e?ARcO=5`(h3NI_fXjUe@
zANbtMD=B6u&$wWNLnssJzcVQ#t_f3VsZ_l}z>khpd>FU{?=RVt0F7dMj=<_cn<a$u
zs72erNU-E1Av(+159ZkuZrr2D)Wp-F2$Pq`38=9cf_kvZRl-!xiL=31;)(-@sD4lJ
z&D{vV`oZG*Zn9?oCs6z+lMovG@;XV`)-8f1VXyIX&Zl~fu+kbuNE=)b&ygYh$Gy?W
z*$dh9S(K!Iy*e)>7J$M1{WN}gxdMe6)Fo-zLh+s-2*LG%A$yDg1$%Q)W*a090(`a>
zA5CA{p4ec&@PY_rAC(cAMrU0`!Va&kfI@RSh*T;T6{D)=P~*o^g8YH0#*`6QY<vqI
zE`>=ivABO;q4$RIT19{*B<A>V{lb5#ULb+Kgli#Z8cdUgZ60ujh}Tr?benu?(G6V_
zb5jN7x9yJODHOwgWDtPp2^i<haTzSgrSt13S{Zq|U;~2Y0#1@)71PElgv8FXzbhNX
z68z?YMU|)n`yk+ZYkHUn2^SCPfT{#43+>Tq@q#B4c#S;FQBW!uHqJ|2z!)@21K9@{
zZPk{yW%y^kLNS7CCj3TDsThUsDgw7gk7Byn`z<eOLIzwQLj_xkJFN*Py?WQ;VJy=w
z7r*3hY>i<ZMekgXA#U7Lvw}Uj>BXckRXR^PFt_S#F@I2J;o~bxWJ+Q#D#kRA14|Yc
zUhm)v=q`}(s`DzETNw5oDorT+j38ccFxRcQ%6Wz1;)gT^y&k_4vExmaI*Va4{3ocZ
zI4t*1ESgZ2Zxuwj5=7^}(MH23m0~ks)DVq8F48{eA`ilx1DVkXre&dK?)V~LSYh>n
zFS>|uJK#<rtZjjNK@_|LT0aEpKvg9xeIcwu+nby<dWGJ11w4dHW7HHMBfBl~F53%x
zryi=uKuv{$ZD(W!`G(+sN!D;~GwGF^z87M_Wn#Db9s7VcDMTZq3Z5NCPJ+>!(q+uE
z7_0@%D3b7ijtpANMU>`K{!1vZmYK1yXew_4^JQXu|7a<{!QHYrlPXpAtJK*Ba2E_k
z{w98N2f#4{QTL$3fRG6HvGcDSWTq;7zOC2*SgweDz!zNTA`v9w@t2ZR;hWo&Tsljx
z=}pDxl`ymo)G3p`@yljls@y6B69y%mNHM?+y&=Ng%`+iFGc#Mv>PNZferh_y4WRR1
zqQgQ!lNgI<F{Da^PC{cAd5!gO#B*4rlI}wJz^$g9yl@LAy;iOFiwB+(sNN>?aPmdP
zNy%Sg{jv_;Q`=k+IJ_KMxPX)?woyK2=t2A(dIW4Q!spJ?IIa;YwvYudZmLu|c+t0-
zgeArbiiosv7%J0NhUcSs$k4io3V_AplB(TxPYJMEdcNh)RTx9MB$cEHbTu5F(tw)T
z65D@c^*CD{L<C-i-oY_=dlc*h?bX@UWv7Y(LR|fwcJ8-5y<9|Rpc7wQ-1BayDV?+D
z{&>itIx*MiBByzBC*)k}exqIF?I4yXNly1E$<1xJVeEMSy{m?OJqBmL%`TPb;WYyC
z<!_-`Xv<I$rW6c~JWsdqsqU>G8g(2_mG>dQ2n$#K8_xa3L`*%0_OHC*z#<q^^{?vh
zifwD9u_7MNfl5Yd!pLUNCt97LCSNh7zUYwg2g+V(1mdiN!wwDF_iGg%CYcC;Tly$_
za1gCjLMjui;BP8fe^cdoitUR-FOiBPLnf6hqPP)e2DukEkw)mvc6c;uv4;2)Vh#Vt
zwq3)O9EL1(!bJNgwtd6?n*~vKyuLP_PW+qJ_<Q0F8UF13jf~l(B&VO(C>vJ(5i6$9
zKnGH~t?zQ!V~XH`n4!$Xs(Kjqyw+7Sy&3pa<0WzT4k|DZ%lgeMpLbfFs@v(x<a~Xe
z(1c9z8ikS9Lt8X8Grz*0lBw*7&Yt**=*aHF`Fn0&E@s*Ef!>Fp8(N`T%zv%V_hSwA
z%=-<%sS;1Wl`U{93V}5k^jQB`0GYuAY38cYr$bjB9+OHN<U<c}SG2!F!(tH1qB$8I
znmg)t*(cZT{|k{J_VN{5nS|0{C)POeYcwX~vvj$h?6b4Tad9KjQ3NmfA-U6{9NK-r
zF6nNP@yNR&pD<!CsLAM^Gxo?ce-uEeBu6|4(t@ZaDDk@my8HJ8*dQm=*1PHWkJ5Y>
zN@UZlpn@lJKKa{9DUMl|tbGINg-OhSuN9!C`z1W{o%aSF9Z-9PKq@_iTnXN!N`;Bh
z%EmdL_a;7ZEizj2l~VnLDsd_A<r{Z6zHh2&kWLItiQQpmz|RK}BQj_AGz{q#&HDjE
zK|}!mC(8XVFEc08{~F~kYXTe(TabKK)uy!Ce#6OV{}cDz2X&Kjx|+1^t4mf~K7pG3
zN;H2cz;*mn?Ck?j*QFnfNm-9Vj(}`8Emk=1^UlI?c5a(o)6?;Kf4tvll~IvQB*oO;
z(XQjrJ`qe(KK6&ba~T0DStOqQLZ)PV`RQpHkXifV`eZuY!18t-<>vRj=>GlmqGspY
z^>*B7)t@>cBDNr7NK-0(Bz8*lG+4VLCU9ujY|2TQ<ZKwXY=5)#y`*}zD~JDVXP)4k
z>8QNCe0ylSjkj^7;8Km9;n;$a=_si_F;b781xq~?SrUw%gq0u7z$fJ`(YORtZkR8D
z_zR|cZnO+bP(=>i(|ia$aZGZOXHYEF##ur1z6s+fGrT7dl8urEkiGa)zNu!?xy+cI
zPNXdhe@33oIqrB)BB6Wxn>_JYE?ks6G4r>UgladocPW1nrB!pQ$?2v8O0#K<IRWgo
zo|Bl$-l{J@{m}wVic*(AFtVI=I34H!TXJQ{+6mG_IYGOzDVlOPAg`;V{ngz7jZsGt
zl2Qm$CxVEa38m7;>25IY%rY!&r;35V^xj+)o0+d!w_0sS$NNHwDyv*^gduGVMn<Ij
z&{r%{8~i#f7sg!pbInDKMvIV^`M9{}MJbT)1v99={AHw-Jd4aE06pq*M;f%tf_c?6
zK>t<#j9GQh)3MDc@`|k{lKP1!A7cq@JyUr|p1}}{6_36iGb}@S{_jzLn$?JWM<KyF
zPuNcdw=e!8A(WZ2B?2QZ>`Xy#yEL&t42W_UNXeNWe5tS(Q^9MI89MkvF(*DT2C)R;
zzkC2<AeDyZF@KoZaFxCk8RU=yS<w7UFVF&>;t1{ru-}1X8`*Z9-L!rkbT;jcf!Mq}
zmRqq48(X+|+-?os^pz_Ym$o}|JpH{zt3h4{52&;m+aiA`h#q*F9K!R@cxX#V@<m@K
zgjdx@4DG@+IwH*-BcwA(i<yh*Za$Z6Zb9`_8Q-L8f(6P0xU=C@3s<|h)izdea$r4o
zRBY7T+~iFKBNji4w^PD&v5^vZEXm)U0K8(v9$7#Vkbik_Ze=X#sdj5)odf=+!V41C
zs|6&KeCB4RO+kiS+OgN`p*PBLi|4x=F6kG%9FgM#u}R8^JenNXu@=i#aR=kE-bwuR
zjO-VMbd=%{=syRje$~~oL8M0HrIFb>9IF3v^L}~W(s}_qV^g`p-OpBpbu9}!&-Wd`
zB|n8_I<sn6*<4p0QwmWDA@#JAYc_C_!^+?0_85J!pY0#Gu4Cn*otYx|iwz>2Jg#8%
zs$6`bgFotcVsYyLV%8HGVNz>kxL%o-#xXraLdsbXfKEVlj*iTD8VEM)G&wqR&v1u@
zN9K!!z`c)x;|U8_NN*2q$r<u`^hQZ!y<0W|1@_~~xCtX7{)rGY(nu7Snxg#CYl0^Z
z3%~4UizA=NUPK0Bli(@Kf`IC`4Ilb8?#KY1-c85fo_BJ;cuEzoQ!6#6UWnQ!Eym<*
z8(7q?jm${UnMjCokViu~DNbf84J!Ru-M^yMjZ38_P*#qO(CoLV_H*mjUo)dX8QGEv
zjR}9xr#nVASQbo9fYa3kNR8{p{*-I=)Pt{CED4=j2NZWcdXZ{vdEGhm0wae3htzNr
zC4F%|_X%FiIXEt<EMluG%l!^h|LfaC!8Dyh`ySC9#KEgd*_CJzk27|d`t3u)=NO0x
za+=t5tK`dA*dD<}Qr7<=`TnrBy9-1)o67mXr&Qx85Gk}!2QqwhIT9eT7gFfW&}>u>
zqA(&@eh0yJeQ^&#*P#B7-jiB%SKtpruug>eL#b1(*ht|;8|cedI7Ne$cMHpuFPcAR
zJmauMy{jTN1cNP5yY>x?nBUPU<Y|gY7=MSJk^qbe%h1&oiCM<VZHOy46jvcIRFYP#
z_~DcY?W^)BGyoRLJjkEvgJ(S>nRJO7_Ic!<UkjfGWg-7-O0f&<jRML@SvFD$t`DaI
zqA~A<ezafl1`eGzaM$C5R;Q8*Cd_1YYumtR`T8+vgunGf26tDtBc=ZZeu*gJqWEN@
z4=p)BE;+!%-e_DtY`YR?<a_}yBjy>j(0?s#Szlml*J0^2FbFrY&-|$Mmt)i+JkPRq
zD~=XyIv>56B^y?pcwsWTCTZBW!phTMFKxHDjW~`?03QVVm+=xudiXRaY5Hv@ux$oL
zUs9e!$%;%$UiLjg%ypGSi^)xL=PL9gkMNUs%dlaKb<63xb<{9&xkZa<*2<P+=aP5~
zJ|2Oy&sViitvN2?!aS0pwrZA4W#s6JEmF;ShH{2CZ?}m~4D*yJh8{Btdn1zJB{AV{
z)zIRo?N4K8^ZWt#lE!|weCjbwU4Ol4*dA@LQ}~6bXv*eB_<-cj`Y{Gb2Y$uK4M#fp
z0ehMzyDMF}#J4(|DzCH65p3}Ezf3do@0QKgEt9EAb7tmSZ@ZNizRN63>uAcXrmgqd
zi@iE_&ohilub0a?z_3}l5#Q8Y7yx;zxwDQj_`pqg9ZRiYlRLisMk1WjdCE77vnrCY
zDv%HBm`N;c9S=bQBeeZ#hf{m5*t8W!TB&_WhTO8Tx>GkSSEPpu!RN+1wDQeKK?hD4
zchyy7Jv@l78O)D>H4Q|$C#Xx4PVX3$DT+X_V+<}1%*^e57sZkd5+*K>EUz8))&?qD
z3#CWf<J93Lu4gMfbFr0M+o99=_<A%5s$)jaV0pvV9vQpVnjt7DbEC2x(bnz5V_k^a
zX%K9<OY-<d3c;t&%F@E2bqEb8X~*oS5y^Dbc{85&;;R2w)Ye2&_{vwp1TlT4R$=ND
zXBUpJ2NiSDz<h1EzoQ%HevPX^(x5b%B}6Y2)!&iT)J-h&>(?b0j0nf0KY0#neELT?
zQ_s4K7w>;t(x;>95fUSur1+ypN*oPrLKo-{L3TtL_plQG*oi*^$zAw%*0OEzQ*Ujp
zKSbFML9Ox9g+B_E8{is~3i~BtJ}2eZr{L}KF=UKj<eq^N1c72>ulEjtdS_rw3Kl_Q
z&Dbe(tp!LTArqiIEn!?WjtT96$du)OZQhK%sR~^lYV1dha6rrH?8|+G`aUe!T%w@6
z={<ha+S~%H(KT5$3MQ|*y#N!rTX=MRj)JnFi+m%msRYP$T1<7vL$Q}E2TA!Cv%*_(
zsycC|okr^~VWwFvppuM#4vQc9Mk!ULxqzpZu5fRe8*V|U=}P5QZVK_|EeHn?Sj)dW
z84=Pu<Tl|gpd;NZ+wpq^^|Nt2G^`oK3pRRJSrNhvpq;`q2plouR<rcl-bnWGUy?Px
z8kV>6Ta97m5x&|0*ej|b$yk+NsxGB~9ZXW308LGAfj@d~5Nq9Tt-!@Lacvfx#p~N0
zJw2HGK4{}X>nJ<l+S?XxR~M)vlMr!d2p<(qPk8M36WQelehg532_N6ptuMR49l5U&
z8SHw{!pm^wtz>?}ICig6{1&sK(ULU9OMs>#O4*aHiyBu?ipq`bIOpN^@f<@<@!C-}
zNCYu+d?;***noj~fT62ftZVV@n!RB^IXvr(-`xjDs443SiBG8RXzyPFkZE*HE-tpJ
z6R*O{Huek?j0ES2u*o#ENXneA_3;G=6{*`c9_OLm+fnZ&?pErOw&IFOwXoDsWpG6e
zj?IR;2q22aJDAh<*{sO*YoGG^h<{E}mf2;+i!bXR!HBWi2)P-ESGcOZ{?kXMAs#ZD
z$zD`J1*PH5fSEkj%A@dV7TDQICtU_^uMlVm4>!Y5j|sgyUR$}uBnBi)y3!`r<|?Zd
zl-jZ5r2#LEYXbvi+0H;>#ggYK-$pHZ?whk!`Ph^9C2nls=TUJQjN2RXF~OEsFzWlc
zu`>vDHn7#2V>vcVoF~)yzN)iQ`-8jl<Ne&yns}K1MDYI=8Oi>?uh*K^)UpGN{SQyA
zWiphLP0jfb53)|sDLZGpI3l2E{Tz^Do(UuEEHY94aq%%7atJG`lO-F~quw=@#d*U0
zBD}bM6+Pq2_i=ZA8J1g=hH1!@GhJhAN|Z^NjzqP5Y03HFG84`Bb9KIjz|Uw|U?Kxk
z01koFdj6!R>(k}=@bVyCbDCvx_^D*#{C$@y@BaFndi%K?rsl`X?)^P0Z^y5ZVa8$)
zyze3Mk!T^7WZWp7^B1T0Iy}UT*x)%4#kMXMxdfy)Zg_<LP;<typ!x1oDseoWgfrQ9
zD+<anT7jw$DZ4(MNyN;sB#lfaV|^LgtHPy|)67$szBS76ETm%PyS1D8YdA<tiSN%-
z%bk3N<NTjy9rEd}>aRi3ZqK!rBXi~)UA2o?*Ls@^(e6928yImg?I7;i!sXwpbHc}O
z)SA6FJ#BDYL7~6y+{GM8xs#hwfG7-uPtj0gxM3qrK`>x&j2(Z@nax=8tep!uiJFze
zU>WRwi8&THqQ&kdwGWGtPN`=zN(Gxv)dXQKSfkN;aTc>gO0O_ZgAit+BTX30c5CNS
z7AxN}h_y9XPtZJ2|1e5MJ18)BF4+$d5TVq`@SzOZ|8{VFK0d#k)F|bLuQc}fXN?Rn
zeYLLV`gmgJ(%L`j#DoWH(c4=1O>rW#ckcUqc4@b5qbx&P9St=XXms!Y{XszPO#uvx
zo@JcP6r!T6xgaN<XkJ2W`KNQ>os8jVpjTbi11G7JJ6>0gAPH&aNy4UFNPv)8f9Y_u
z){<;YLO-iMkAaUAf)w@H3ChwV=QcAf;a6w7Xh_8YTTbTR=*`SFoDZ!cmQl~XUBgId
z;$PPD;WtacY55{2m`kw=Wy^#p3!I3l#MqGsbmIPb_73Tc!8$G^myY<ZNB9Vx$V3z(
z%*e4MGc&s@p5abP7=aXr1DgQsV9uPdvhl{Z{hG3s7(aJ4o|JD(oz`zd#aeu6Fh%BG
z+EjFRvY|TkXtiNaq~|1At?I~@YZk1*MK9Ve)&56XY2*dYoNow*VSfdIP;=D6`{s_`
z7ojD=4FVa&X=u-NLMs_C+9g|=R3V^Y(J>;E`l^*yLgA1m)nMn7*||H+=d<bWq2Dd^
z%3Oda$4CVaqui~vAt;wv$DAg6i)&2V$mKzwwPlqby3%OzG}s(O`}cZnRKUn}`~Wiz
zw^QG6M@@$tU86K-*8_681QE~Wt1U8JJwlqJTLbb$?&r=##z$2!wWK?h_q>d0kF9Wt
z1mA|kWi(Fl2FAW+_bnG~MnnAy$#NH@06(uv)IQ&)^$aVGKJaS_9^YY-gpKBV+SOy1
zj)C$>)r4V@Q)Y9z)+Fc0w!q|<fOwaslw-cKZ{<iMjIve%4qD_ae^P5mquGn$fcQX)
z>K&?x4oyp@eS6OhUd`Z*;~BleRY=Tsr@h11l2<_VE!ZDU%MJKh)mz)MZcH`HZv~ai
z5q|3p!^KciOWy<Pon${VY8f{{`p*l6{=h!Y*Rn1Z=^yiWmS^0qvH{6ffT3~~paI~o
zzc^e@V~9`MekpZl<%W~Vza(qI*nQhtiKW=98<tAAVV?{9y-cfNdt>Ft8|>u<;N+*%
z+xeaO(QwCp6Hfl%^rjM8oLgQ1ix;csIv<29`cLY+G)-8MBLv4bO2cYJBQwR{%qQY#
zC1_x_mQeYYi}h2ZDeQS7N4&?`5latPAZP2+vaV>^1bP$rKDQ;Wf)3D(3#-fcrH`*8
zp1>w@jv<7lB6Tu}9Kwjtm5zw_n;g0QR;ka(6{ebniqOZj#a<ZAB;WpnK~eZfr0=9I
z@?4?x*EYEpH0s;0c5cltM<yaicNXZm9#~KQEog>A8?<JSzb;ioW9nN42+B4<RLqdX
zWzLL`0&t`A!HhX8_L3?>vN1i$vu%nsOdOiyQrHOXJgBEks|FjrOcg4?oXd#0oowA<
z4;MOPC%Sv6wT=MZ4OfAjzPOBivm%ighMcQsv{BE2B#8x-o;0$7P$-C8YlEuidOEdZ
z0norunZ}|}{w3<}wMFWq1Kn}Fm~W+xXrq2%qCadl4(w^1Ya2S6-gzFU^atD$2;TiL
zYffHmjG>r+7yw0SKT+%bH4uyaDT$_h*Kp|xJA)Z(V;N%Icv6@wUiG}yB`z@(k#c)1
zs!l+Wujbb6iMt^ta(+7Gkz+y4+4zeVUWl3;iG`=_J5(iiA~}s9P?oXBKsD*J!1<#;
z95%lLNz)B$<K@H}bJ^3nY}x`+<brUOQd{PcH4<IVC4;j4HEF(bB9x0J<P)hr#Fj<T
zDkM@a)a%Ck#v>+n;p;G68agY6bWoHPg8N~Nke<cH=C$uLjYWWok&k3Oi7NEDAvk46
zY?quwBwRA$FUyk1-|}R<Z<nGI;B~@DvF|RSg}M9yYDr2j%gz7x{&z8p@ZRdHm)e_w
zyw4VzL2yjzs{FT`gWBq}$?9Q!*`@KXg1R23CZCW?Q*I{V(mkF_v{7739ZVBjvUmk8
z*;bpcv#A)zMc!#wht+z8Cdrs7mDsVWIy{$*=DKA@{F9xX(Z#vp7|f40hiClQ&j&`Z
zxeg|_K!05vk&Qe!BjRpR8H5z7Ph)>$Tp%ydzd|GtE8RQ;tY*JnW0G(;>U`m#4F=Ua
zezC=R{R|FdF%r^Tb-9oL1X=V8<{CY6w$}S$Czs&u0!?ceosQX?TyF!*ng=Vlk=)G4
zrhKIsxH1HzwV4p;B7RwPTHI(uL}t#_*`cq1q7;UOU`PzGc!!LmM1O0Xae4W2Y_f+0
zNZoPB6WTjTkFC1a)%agbOefMXu#anP`E+R|8Qfc}R!ssKDc?ZBqP@_F^TFOe=00@n
z?KXTMax@(ZS8krAxZlZqR)=oH<=aMHBDx=(Jsm9BZohv&eXQ<<x2V%c$#T?fTAxqn
z0Oy>isnNAC0KP65YH!8Lie!Oot?SOjg;F|($Fd3LPWb=u3NtZ~Y&Ff=(z6X$fQ6J_
zVs*v~_DONPGMTWsO3@|uZ@!1yYp=Vi7dN3ld1`3nbn6~=oqTstvd3*Ln#`kxY3r@Y
z)9aJuf3Zv}@YtDUb)K@el^H(lw900B!<cgGFQQtnSoQ8Cm`~E_5&Q7t!s@wt!(2$Y
zd#<6BcsxR>%A(0oP~G<^dgefj`=mdTG17eaNWluO%~|kVnQ&*!QMk$oqKXiT1cq0}
zU~CpBtDSU41id(|a?q{SR!hb04<Ics)LV}JU@-i7EP2Jkp!z)QwjJpHVL?+W^zVP)
z%N&cH4)vFuNz{;G#vLy={risdGMes^Y_PsvFU#*>k+O)wBULokGmN~*g@qPcquDO!
zDMD3V6!PKO3iup#nVIW`MPsuZbRr6_<L|<rtbQTz-yIIWrfP#(qH13+*N#UsS4%HS
zbV3~vKKs$0AEkx4-yZ*4+1-Hs?-IH+T817~BKg7#OTPAVH2-YppbPRAQ=V{<5*9-7
z{@mewmi*b_ygsh^w#=4cU>+^@Gq-~L0QeQ~s3TY=3eMl{eUz+t{w#Uw#SAiEeIHPg
z`56uRAJsw<<s>?Xh`0xx8w&XmoRFOYkaQyZT;@4c`nRro86*YGU(H)Kwp&$y67qkl
z@WF@5$HWu;$S!<$w6nh;QYguw{sRFs{a^kr6%Tt80(yBvOJ!#pXnHvUMg|6GdNB(}
zCuagqb`EIz|38d~k@KIv<A28gw`%Qd;z&R*W^LeXB4T1>XKVt^#|Q1?>}X<O1MR-C
ztSM`U&5qc8p>~_5-S&wAFWCpVOX7l@CmO{8zJSHY5sB3nL%Nz0(A3*wN5Q#bxw_!l
zT)z->HE-$4@o_v2(f*v0q_|RCG`4R?cGRUkR@Eg<<bK2%9<QRWeK$g3nnJB8dT<n}
zyj}-cRK>`0kLHF^e+uWDekc;zMvj%h<Xe1Cdrd_B=Lnu`zVY>UR8Gj9b)bY|dk8ND
zH_9TU(w`Yk#v@pg6z(k5neoJJPKn{1hj^tVucI^XH#Q1!#Trg^c*T{z1Fn`M_)~{l
zb<hdA_JXiwrg33U0z>2o5$NqmUj5_hzC$128VSig@!LB=5669cy#6o(t5;a<!eJ~h
zFR1uFpx_3uetRl96xfq}Cdu%L10@Bp8&N~iS1cul$iOPdBJ(?*If9nx-+^jLBy8$h
zifj#hS$UVk+5-C3v^x#ZK#<6NuzUj=eYj&1{%b-7j>5qze@qbMUZY8GF9|*GB5;c+
zj{8#TcEL{vMk18Msin(EJgxf)pzyf${)ERDRQUta)vNj;;5qd}PEkkqwS$|7fMDku
zX$bm{MePxlXTaOjg0lm)ALF4(#>-CCt^E@e@t}50B{U#I%0$_a$uPnhh>^=Nx;Hbc
zBRP}o{z~QLNobIoO}6Ti;t$~|QA9wK6e4>7W3MNrk*+ah7{;4T0V|<>8;vI+IZ7l-
z5LDZ<WrB##P-*SK6Z)@9Q$WULl^4M;Ah>xZBu?NgobKwYbgPg{oJCLhR#v?9e!-Un
zaO;{W7<xXuJ)E7cGb^erN3EZ%FW()lyBB;4eWf<Oc{{SI+TPuT(Lbn-J~Y0GtUDU0
z;7*q~x!C7b5EFm|iC#%~t7~!yHBbOM-Hiha?M>gK01<q@LuDM=u}%i96=S-qe>t*Z
z<I5kPgV}cTC@pu6r?z*v^mx9^9*KN@v0(0TmOP9*`_ihGr(vtwk*_vpR^A<YDh^FF
z=aS*i&gRVal%WK{AZH@Jp~7VA?AewF!3-ZkgOAEka|9kiZu?_smWPnpiPwiNQQc?W
z;zG`E2AXdI&p*;4RU4=ax6;#F@!*3f+IfR2S|s@(CI9Mznnxq0%&6k!M^5hy(m`0j
zef+AIqg8ieuj*!n8CFwk%#=>f=o|e~@JLRyYN|2P8q_fjGT}tw*!lU~n;mzrU&)MO
zDhQcY7lym-S&M8xB0Dx&@P<4NTRekZkwYWSe^n(DLRwj&&nP^9Js#}x{=V~mUoG?G
z>wM>Er>aa_Ja>62@pLepgm8eJ^+A?wgjJrm9B@vIp247`0JmXEf`?%mt?*>^VpoQI
z{k!TnW$IMLI?u-UJMEVPaeNBrs74_i<nS+D11gt4nHa5a34gvQvYPUR8OtaQE}3<)
zM*+Vnx%5rADLSe^rDpz!%1mZ_wiE}=GhG=f14nDKn*_5Hbj2y{UQ3T{n@lzr#e3;P
zSMMKm)D2V*GibpXeb8Af*xJa@g4eKjY_rZkmG}iXdW>}hqtyn<{Lr0J{5yUw&?cGr
z_3Z0^kR1z2+yR8Il~R+i`*~Y+cfIsM@8V1?B2167D6=2er``E^XiAIlMbANr8%!Hn
zsf!U+jDp04^Ct|b65<iVsj@|`C4!dcLgyLC@8W%P7yD?#TF)pOpTtjb%~)#K5hkE$
z*6)8nnU2sloHw_}6GpjgC}_f1_u)P*h@*jxpKG8u@<4@B=I$v8hcr@yhNd=xo!^rK
zLjsof`UN%exNv4?%@)tw-k+|p-#nN)vvT#ae!R$1pLWpkbGMmWZC=KV7?O-wrlRH#
zYMnG=9cU(8NY1a37`|Atb!KNrSFPY_%hEa!xtGJQx?sgz>9Nl?Vh8uR(p@zX1XcuO
z8-BeoG}H`WeEQ0{UGplNUnZ+b3HmK*D)C&#?*~BOy3yQ24G)N}Of?`$twp;ZH6$2z
zS2M4PnL+9xQ!JrS=wlmJ0}9+*8^aIr{OIt4VK)<nx`d(cww25HQE5H;xevi_R_@j;
zoKc9fV~l}E5U5b3TUxxF+A;K^0Ugk8lF-)~s8*45h2VjZaX+pGiw@|f9r_ToMfGov
zg1YoiWHL>Y8k8M`)KnW8DXDTR3350SmqYbkFAMzuo8QZDZY@U`%{a_TWLHH4lZxo3
z94IcXT4ugGWtRzD@ognT6I!*aW$P7Yd)fnO-5Pf-pn0mro3g2xov&YAtmk*Jg>JSa
zfJE$eBip8~C5hu7=a!ylG^66jM~J2)NQQZM|InL#SyS;fEbOSxRP>-Z*4C^P;|B(3
zHE9p2Nu``nUYNr&Vx>uJRFt^hwQtk68=R<V^5#Et7n12vO(wE~s`ufEF$-6_;%t%z
zYS$9qpraS4n=MjuF;Y35S=)!l`LNI14KC{+Fb5xDQLGg!8rsx!)R_n0EVlm|z4HAM
z#y)vbBM9>d7Z>6ke?m!qN*j|KTu6&3HKK1mCA_P%TQ=Pk)_8O`ze1qXkM(6nZv0t&
z!n)7DeVfX$F1IVk4$+*sua(Drw;FyR$kf4P|KoXQVf?QzY8IydmrfMBga7G|e{F$B
zY#dI-Q@=DIm-xk-W;<mV$SdQ9kDA%SN-b?GEl2in@p7=3^pE|I0Y~EtFOnMe^i?hT
zyKb5?<=geXw|Bn9D;udH7@1Vu81@uh%1o)02+8TCKFv36AqCd1JI0c}v$eCc#n<)o
zvh@@=Zv1`0&;NPfh~N8jb^Sg$D1UPe9+wU!iJWdoKUqn^GOxPdl1s}1EPrI(&y<K>
zPU^KrHR)11F%rAc?DPD++;ipWQnnG4T?2pQgf0<7P&Gy@SvCcEvMGJPyr3{+ikX1P
z$`V)<?<*KV5lm`D=B#Ni2|Z5WC<IoVa?4yezSNR|Ng<e~yOt(vGv&xw2({O~Z5Bhn
zAT&eN;3qQ%@lI5%t7tSpsVjq+0Pf{2cUbfpSSJ%5U4q%KHKLPI;D4DIE)uOFSoJqH
zO98W`P&tc3j0HJdF5S{#D;(nB(S_WC*xW~QhDqARsXI@DQJi8ia2SDRTc&)5Lh|m;
z?VK+u1BL8<H<9wsSleg8bx4A_&`W|HWtard%J+~X#-zT;!ctN0SbA}_Hqld<RmQC9
zdS$W>#THC%QpH83&iV%q&!i&5*vaTw@Up|%^o+?RC#0SN37q(fdcd`=bm(6%K3NY$
z!qNMiN%vn3iu(~9cfR}%+*jmq!R=qVWFaZ`*`==_Dn}8s1{u?i_EI5H-9#av1|TFc
z`QOtCP*~%7h#6M5B;ZDP;lVuR_K~&VRTYvJB_?-*2_(we$^-IBVz6QU(2t;2mjtu0
z-USl`c_&T1<_UH}`T1e-ks3g8k~gR36yF_gy5ktH9M+8)iL9j24N-!lbSapEj(-ON
z#*AUSX*m)?Ts`F|T-k2$CTGYk#(JLXmBYw;yQn@KC#WdCgP|E4CM8TVg6qh{B8R|~
z1M#!-aZ14PI0-~RY+XyC>5g^u*eVx*#wTjZg3>i`5FeUxCFonj6+~}Dw&<GTNrc%R
zx(GPWQ`V1TDkh`$GtX<hw@~UEXaf8e4a=&*Wf*r<<SXhoczV`~^Ok|WOD-Ff<W<+z
zY!2YGjko*GqRs3B?KNvFN-dAREZHMBw|#gudu(m1x1MYDn=~(8teP(gyABktpPTXZ
z7CoLyg|ZKm2C945qL1W5ePxA^%HQRdFY-`goTV_&l?_}R<ytivbu0`|(9PlUc;z`I
z>E8JmCHx^!=$twzcM2Ir$@FFhKt#!iJlP}oAR~ELr-;isJG|Q7J%58>jA_|}*{8&#
zrgv}Z_`D}h{c$gsC06*(g*&qiP2rwp4P9!~tNlL8cr;j9*nG_C5A|G(L(xH><C}CL
zB55ylgwKtA*u|d+;I}>{MgiMAVas*vY^XuPKi5B@D_PU9ucP2#%I9Y6MjK0Y0c79Z
z0yBn#o*H`6W?}5=cp!oO>pJymKfw>Sjpd>}onuo)w`YvBPe+u@PCJj|O`HO26Zmd+
zzc>6Y3w5gO4fi6i;!A#VND3M##>nZ#@AQ3q3Iiatx}H6sT*%DeUlwiNTv!qW+Hc|L
z{@VQdm^+>eIteCSI-jS6fNKTZD^RB%^CNwHDt1$Z0d=GuP+LNmYiNWW*Tx`O7QZ)C
zbgBVOln2f$v3})OT@}^1s^UxF_Ttv}1l1Dl^moSh=6&01u*xo5m%zi37$mGfL1)@t
zHm}?mkjxBwuUlGw4$6U)xItRbDf>Y$-+=tel<c-@j=Owq_dWv&k~*;c+HVK&C^m-*
z`JB&~uVGob(&0)79Z4$ua({Gy^c7bYZIX~N<|AqF@VEq|_13UFi`wP7w5MAm3IWqW
z>Qir5<+c#={+2e@8~L>(gV${SD!~3UC=0*1TYbPdvk}#xV6p2)5vAx|oN`^Xm~Owq
zb|$tG<Y2qq0Pl|!{%CW&*Y}2+Jsd2diV{(T!apvHqn?hl20NI1)Nxx}$T0Wm2v<pV
z3cP|%V_dEbJBAqPbDV+3^HaE_YZji~Bzu*z{Fke%%LMF~`|+DZQk)$*)_KYE`E90X
zZ?o2Ua6{oqMa>mt==iavi#AtzGgVJ__*KF&$2!xbJ=d$E9nEHN$DmDTQHRl%7L2|{
zJ8k7iUM(yxV(!OD56|QBy*enY-sMES@`ce=8P1Q{Xpdsc@E)443%NR9IpZ!<1aaC}
zHa+n8(FprfSXreqj5TZE<S7wNpUrYr@>iY)71M8Spw}U{#+P^I9A@tlm^UTDP%xPu
zu(G@xXL|Mu#jph|$ytDyQ8rZb^IHBl;Hd0sSs5WduQH=QY^rU5iK7)l_~8by<D#{C
z)%XC<;%9p$Fi3@|ORwZpoP==q!<+_J&PQupK%J^MmY#54c(;~!)bTP5H{#yFwq;)U
z?v;>yljlRPv+|^&1BfY->5=ct{7DFBtFE>$s{^3SS~&?He`Ji|qw-qkaE*1qgyCJi
zPp{zmZ*0dr$LFWD(izk%Y>b|z5!zV0@EsMwLC-cg32AXKg!~`M-l<Eowh7iP+qP{!
zWpvrLZQHhO+qThVb=m5&ZLL07XX_nffBP@o2QzX;W?XUQ(gZ>nT95y$Q$Od)5DI?@
zhQ^ts^kU?>4VOM$OJo#?p%?e5k7ocb)831et{X82L4)ebSlhO3bcbjq)-`a{D==;?
zzf2HiX~`I`l<w|9jonZDWx*TSUfP)DDT@YV41Ry3u_N@4?v^I3q4#LhkM;@Atf{3T
zDO*ezYYxzOJB+Iu5_i&JtB`M_j5rQU&b1;Kgb#hp$Pf<Y8=y^DKc1owzlNMQN_p1N
zZ|Xpq`|sC9qmXc{!m4vs;48eWr4geGL=6HeCB3&ah_3S)gX!s^Die1O+g4f60tpyC
z)$p51i5RP(54Qk-(tA9a&+c-w4u2OB&;5Buvz!kZbQ>$OvJbtyj+)<1d@p+#LA~H9
zHRJ#?<yiyGpnR90+rB5Tfh#ys8jwSTPX+JzVElm>fOry8M@UQHKr#<mhz0UEX?)@C
zZh_>_)8wcLYMQMC>+>KsV))}OC?I~fU;0mZ>)vU3l8HF?c;t<i`Y)?XotewDyTQe9
zzAHEf33I~)=pfat+_Zzsm)f;c#4CA=Vrhn`z)FdNI>qKknR06kmn{A{7J#4q)5YzR
znX;|TK?JQ114}LOr_%^t{V3D`)?VY+f~$=#GM;O<qe3C?uktKj3Wv1JmOINEQ>K~L
z&Xp%L*h#2z1o|)Rf+sG*r1fG!&pf96n{0p`y6@ZaPtV@#7fL@UQP%^j&8n;*$^mpC
z7vUCgxW)5acX;&=2$58P=YI}!|Fy0BM~nVny%^nu3A<q;G|`uDm^31GFkYl)m+BH%
zRd&UhvZ^jYfLtP;jv-~@`xCfv^>EZj*#kp-04dQPe;4|^S2{(2@vISq2rLRjQ;K1d
znIPL9{Afsl`(4M5;*fvqJnx_<3eWhG>x%;3?`QDZzcGT7Oj@Qf1H*kz6HWpr(I<bQ
z@38vROph{=Gj%_k1)1**Ku!<Cjx);eB9b?i+bmK!os5rMp_H^lI22|Av0b;d(<ckh
z4=b{8)yKK;1XAwkm#A%H0AoxJ%et5;09UN!VK~@=8L)+3&3{bX7W0dO760GZUyaf%
z?pPR?5C+b5svQBiP(}Szv|na~6)WwnRRm9_b%^(m>-1Xw)SoAMQ{CwE9Ac`@=>PT9
zEUf=8gPWO?3-G@_b*hGRGATP;-%b5^Y+hC7XuKZ=WtvfYLEBz~@(3^|XJj&)OPBCv
z-sSB@B3-9GcE#VMHdF1iBl-Nph#5yx0O4-r^2^slcfG)p6X}%71n(v<A)sr`L^b!=
zwbOIQ-cQ!xbpPdqaEv@#Kfh#L<^ExW<K)=>AiHlETHk;Cu}tnl2)#nAOo&E<2*6S?
z9o<qPc!aIu_gv=fp`@9pQ-|0j^`35Kk1zS&xk1Ey)z}aCb~(mwS8nD}eBdBfX)k~7
z%y+>%y^v4C@!sql?N=Jpy{sEdb&Gei7tvDpGP6u8?OHN9!&p7kG`cM=-X}5`j#qb<
zBg|`=PM#BPfEg_!Ut_Cpak7?V;_Nn0jT2*laM7@@Cu>8Pwf9l^qo3Y<CPB1@v)W<M
zpR-=4N^_rOgQzzyYQ$i3iKzl-=Sud%7r%JY>)BdY_z-!xz{6A-cr`^QX_>Dw&8TwT
zLRa+`dS0WYTQMuufF)vTMDO}C1KfR^tbudN8+$vmS?vw>t8$)6<t@4QEpK@dPXl(5
zKq~cjCKL$XOM6~Q;yDIhXFn^yyfisSW`bo{!Wif$djL=VU%As$XxI3|+ZRH9zvDt6
zE(;$op?!`eN5EE$l;63#oBhV}mOjD2thF<*=lKtXQ(*|z9<8JkkhwVT?6r#Uzv%s9
zzSsSY1MfVE!^8E44yRTbJTP;vr9Lw0FBt<M%gtGEV1?PyMh}y)M`tixe3zpP3mZP+
zP=@FhCA{O7_Lz)RgUdm<+VWYC?GUClg*#RMkqaUkxv#tpT3s5;TF06Rf<Kg4^W78$
z)D;3X07T9jpz!#mU4%Tg+-Y6BGZJa%%Hl<O$jR>!mmR!L+=D62o9t_e+)-zjCyqF>
zh~nsAX9mp_4fQhDJlYcJMo%T~Wv%!U+T8+O;-vVtW_6)-2{+&HF@A*ApMrhtYglN)
z)KYXHx|}QRkYEfNJTchgbRw+c#COUEqumS?5NDnGq_WM{Vg3$dW_OzwKE%x0i_O92
zSMlbH_Hq6f8PEZVhjBJ<lS$`A?5>kDjI3+$H!?>b9#}2#rtYlSUm>}=IkQeu$n5Lr
z7)l|wGi{dC=3wIajKPR<e$jBPNR?4K_n=xTB8TWk+}cRlUxdM+!e5yUT1*{=Myuk#
zZjHv#h{UQ4OpW1)ZHW@1>9E7qDpA!?VBhyQIxSkq&Dp39#Z*nTnvb0Gfy4VU+zL`)
zetWUxm2`hg1s*c+9qmLZLC0X98-b7koBpNg!fpzLj*eeh6zm93Ag6-{T0mf^J{vkS
z{H;4Z+c&r`3O$;BD_T(ebXo@SQeFL9k)`g!0(Sf|ebh5N4zz6o2L2JinRO_?4|8ty
zo2{)F&gt3E!t$7y%mgXx+Et7Bfm_0ka@J-GzZTPrMWB933?1oSfgJq79?UX;#Ire1
zmE$R7@pLmwmjf3>i(km3Wl5W>NpHta)=zW?o&A?vofUFiM#4GtB@&|GyN(GMn>dxQ
zTf`d$$*ZGc+i7dNppXGc%nqiAmyn@?u}3~S%#E<p84S#94mBR176x#>DYh?E(B3eM
zTGSt`uF~B}CnptM5R)NbM{<8?JdbG_bC-Z6Kjsr@TqW65a^O^=-a87P<pNwd?_n2J
zLGVd9Uw}V}=p=WF@a;NiR6V>2<1NvSE0$Ke2N^;M-J4b;v#M(g94(`!?7<<HjUiKK
z0&QzulKk;PoFXTyikX6)Q=7Yo5z)7}T$&x3M%e~7e3%J0B!WYJ>xZggS8HOmI3R@J
zjnjce*kFtX3uz$Ej?9uceO-P<&|8#w%FiAH_Tdt<hJ0HmIskESigD{GO4`VFc~>oJ
z7NX9CQ@I>>R=dg@v%86(VfqLvYGS@=L7|#OQKWr(hk1888Jo>4G%X{WC<vOFGOp6O
zv40N>13IOg#5dY=K3+6$(nwL%qy)Jij0$bKIwIy1D}vdjjlSJSUSH7dwwX)0|BPvK
z)7RKXh4Xc4uTs{T@hf$*o(XRUdhq8}s$=H?)08i>x@8ip^FTZ9R(P3F<&v^OU_w1=
z*Dy+%jSZ><#^7fN{cPN^<6-EQNQYBXbg5z_voU^UiPc)d>t`i7Q^I+M{k#SVto(ad
zNx4_oYBWn<#CYVA_?B^m(uN&9zX8<(QgQ9Hn?Smy_{uG2!FiXRAPGZf$9KQWD^*?d
zG<-OTgoac@lV2U<U1bkFaGamda3wzN1nw43UU2qiGwqc81XG|C=2Y%($e%V~`h|Y$
zsDqWfV(gAfMOH1HhCV8Tnx~HZ!|o{{!ruaH5?3Duq38V79_GG1Q0L${635=mCzie<
zIof2d*^bHhZVj>YMPs_Ce#+DblVsVmM6yX|W;cdxZF?6iMvHEkuly1pxzA{0A)_Cu
zIOhAx+J`Sbo(i2%(JJ-GQ)iCB02vi+Lo+o>W*f7G_2yF+@Iw4Y)aSTpRmZ1qrfGT`
z8S~6w(j}bWZMW7)5(MH_6pp7fNmvfLGvo1}F8Tvafr!@sr*i)n`-Pc>{eM;ND=kgu
zO)(U|)%pQaRa3cR{wg;@r_D-}1KHP>^%PiS%Sh_T6j7<Tm(?4Nj6vqTJ%u8toCDF2
zfV{js{(idvjC64NKHd)#*XMamCA2us8P`THT_|0uO6b%ojoN5JjGX%iLUn!LNATi0
z;n7Tr=^$~@$Fqlc{W|`iYo7^pT_)k#bg9LWc1Cr^KF_=S__<h}8a<y6#<`Q2(cEZL
zlg7g^jdIZ9^dnu*W-hyHkm~4vt(>dV3=ACI(XvQn$mz(FgYg(ONKqDJFrws*L+I7G
zx7xb4yp)X{wb+CpsMj{>(k6oy<db(PW7!^lgRsIWhRZOLDXt?)iCXtp9S$*ptXGol
z+u-|c3DvxIFWqkCF*ac+<nMSCZtM*uX=hnYN)}77Y(8ssy>T-}&Cf5de>)K8>L$a}
zsv3v6x4gRf^g^o#t{u8NIm2)@5G!#8_Z1f$i+WuAeh$8I#%!l2GN|Z6R*M#BDaATB
z>IL7cnskmcqEN(fC$BCn)KB9SN$8AIIh2J8pUC~#W7FQTm83prUIQs~&a#dYww=vx
zY+-F#w=62YrKxA$@pGFC@i&kIl|>PT5~0>bjJt8sR-A_RJ1Q;L0=7_SY+ueRCtWS^
zw`7(szmX--%+Jmo)*DS{HHH6(WvO!kk~lXN;5<Ra85C#q$kLnAbf7+q9f+Y<84~>3
zegWa_&dhLdCZSsnB^pqYNu$JtA*1!~<~EPm`MWFY+5fa_9$>&C*1x2G3(>DQzS#9J
zf%Ywx_<Tj_I0YzbZ;S9~uD2$W=rpXbtE?awB9Z0|+;KIEkt3uaYaxx|n)`A{TL27U
zX2F^3%9~d|x(kksZBsl|1r;ooA2ayGsD}v-qzUA)!$}H`(SZxuk!7AZWv;BSnA_T~
zj-Cyhc_tQPNULNEG}6adC}GDuGf&qU*YRMSWYsnaveu$fM@OZEjY|t<OAE_W!vLva
z&$P1Lek~;qN4?r^)|21m`K(6LQW-f_RO2Kq_xk$EF%FUW<q|NZkliR0CPX;V``EvD
zDZ`E+F^0>Dw>Igxyp=kVxc_pLFKYP|$x!82kYr<{ugj^M@Qh(yL&qS`_FZ<^^gYr7
zxh+3kzNmda1Zq1<M*EjBBTWL4M#$x+k)#><E2jTFC(Lj8v=^{^^X5N$-A!mASr!iX
zlkQ8B-{TYZvKlzWp5)W(z-I4y-;^Mj(iAvXe+>}kl3^`s%^f$C5h5--@8N9w<O%-D
z9sJ(HzPx-`Ue9jkE~p^}iIijtOxr@k<S-Zbl!#`}^947@R5;#@x`&`}j6l=T!`#jP
z;$fw4jMFGV3;fOkHdN)=N$EF190<g;K#HqEMMcO7zK0GW{{DM6VfGd>uG)(BNo-HS
zezi`-%06)!ER%D?ak%|4*&keR^t*?J{U*t>8AxW_4t@$NM2}HJ69Yjr!m<EU-5qq;
zAz~&#*oPkv1i?c}VL2Q+3~QMg2%T$Jn1wxd&Zen~V<Sf13=ABTP75Y1AOhq4zD~!U
zqx*nVY;c*k0rZV?!}Kg^J_by}8hZ-p>2o{zTLl8`B!A>vB~P%I0U7?MjA~Ax0(w$#
z?_-s|WdanG6&L(nEDd{eNYFnaX9!)<rz+kRilO7~)`Trq?R1`v{>zkMm`W&-$Lz@O
z>sLG$WJ}cjl}Gb(vrKYf01#)Wp6z2tyZU_Y{XV+77oljJq}onv;)>N8O0}!{u>ENH
zPa5j6$YjU8oA7~zOt&4=-t+egVqsM$cHYQD^qIUKTrBH%AvDhFEe+X{$py4!rx*eI
zS&~ybL3Ws<b)*tNm-RFUaFS!WvO50Ubdd8ijT?u&oHp@GIg?tsFg9(w9jqTzg0r%8
zmW_vxUV+fZ+PomZknjqgDigFdg8G()8p-q%Kew10W?v;4r&%nqig=ATW%cE#F;B!O
zIz|t@K*&0uqgGB!2fTM^A;_Za10eN<cJAtXr=-Rgk*3NBP{OP1-M-PxN!A--t`<=5
z*}B;<$BxC8)_~P^!a3-$g6pd>Z*Bhog=^g2eX3z$%Qb}fE44_!EJn62DKeH&Ub+#X
zU?Tglz@k82e29PykJhWSO-uzEDDQD%Gp}~>UDW4oB^H9zf6iB)e+HLAU|aRLOncf}
z$(smwl%u_t7n^3OZ^Bd8ScTB|Xiu`wx)xb!sNH?b-FYycGoQmLt=F47f@n^Wz(41T
zN5I4L8}g~+I8j&JtKnaQ8WcykWp4u7>nsI1k~?@ue?ojX(<NabAc1^(h7%x~5F>+<
z6Ks5%C@C}?y&mQCO!aG)fILD#didR14Eh>z-fA*Z4K*FYk!jr;Mx~K5Y^1+(^{{-e
z%U*(DS3?ev?d8_Y1-5{{kne~cvt#508A0ire7NhP65`B~E61XH{-+FQdhW+)oO4My
zbsR^p6mZjwj`Zf}4eEpPz1e@w=hw%D*JkR|k9Rvi{W7g1oTyaZEx+#$ikDL30oZtP
z0G;V{8F?%@<)2bUHr7S#K!C9_L?atkoJkYFe_nlQHN)*Fsfu}K=W(Rd)kyKf_Z{(n
zK<<BT7~M1#Sq9Pwt2TN2649r0v`eqST4htHudm+*FRt}F$u7}6ziz<!{*Fa9^B@r}
zbvdP^!pB6kV5A+0xVV_PZW>DJsB@Veelk{>0v_`@_7Ac9`aND?DVN$Uf?<_CADFl5
zSY+A7viqt{b^@4>tb~Xx94%yWdApyt7YQJu$|epUQ&DED<932dF-fCIK1&u%_&z|)
zZhesTbZp&z5@gkhRC`OSTKnD(^YLlqTe=c@n78<XMSJ!)eLhP0uknQSzj(@Q|GTkJ
zs$rYF*@pb>*Dtg?nVo3RgG@%rnrpd{<DJVXcX{Da=ElEACSgRE)FiQe^nTGj!0_tE
z-7VW6ffN9av9^7(4z7QR;dA$X{?T2n&Jlze9fU5HHL~qiI8Cw~=9tTH4Tz499M3RX
ztzNo49PZxi-ptq2uQ!=Y=CJ>H9*)0h^nG~72kd<NxPKf!>-%@kx;VLiJeObS-+bSl
zi&fvfbDQ$AH~4Dk{p+jNbzC@ImL(QzR(E`tGNWUJgV|m&Ro?);UU$2esJ-949*WCB
z7lnv^2-T=a!CoUeiaKICs|#AbenvQ&^3&9vPKU28=3Ae1y$<OfCn2;YwSY}@G>gf*
zHta7<A9qTdG@@y-mFu{9;i%uO-?%>8Q)oY4OGUSEwQ}rfl3(j6hrXJxos!RyZU^3^
zCQjO<k8}BO5cPZDjZHVg=t465X{g8KZOzqZk(J`CyP7N;?a%$X)Uwg~#LMy$H#R+L
zc#16FTEYDhr!2Qlq`aQ${5`6Z^=#Spx}A6N)R61l5bnIz7s?LKAmMgi>${FH$wa&u
zItxP~`rd>Mc90~}Vir_gDY05W@#4~g&^xl%nDt=i??Iy5pEFOXjKXr)rtg*-6v_LM
z?-KGRld;}8Tf3RiS(N+SCXCOZz1SVPF#ob>a?tkdLxk*qwxr)SuN#%vjx`MPhw4&y
zCAa!3={bOdIJqt8d7)i&2>Zk+@+<zssdvn5{V%5PP#cjS;TEpQ_zn51t1K;nA;Bx3
zOr^x__tt!gQ~Qkqk}cW(dv?Lo=|;~=rI}@3@d_la+fHD}Uy?f>n%Q8JQz-F)zF3|v
zFJ&)AL}lx&F84KyYk%jxCd8Jdy#YTak-(elT!3ExEKx(Qewqq8scNeU%i=s}FUJ{g
zC!=q{8J)!ij|Exu#(UrH?BL2DO)l+rHsoGl?s?xrYX{g<d`PYszO)W4j3N9FpNwwQ
z=l5HNOk-)hEO3u_#^MKwe@n@nszOTxn$$}<H|3z8z5eKZzndQ6fH(Z_awwIF!kWru
zZZ=A<bcFC~4tfNwjY><X1d^6vn7SR;51>Sf&Z}<`6!xH`Q4KhA=Hg)>ava%m&HRBH
z>lklQFh_cnIb_W&y2flDRrowvC_NIFDh~=MeC_$nDzgKG!@o$GM|Xmt1ZI+o_E2l^
za!>@(&iPravfeJwtnVul?s+Vw5&B(xm-1*RmB)>j&ubeg+Ww{5;m1O#zuK>A+Gki;
z5ORTaazQdsKL=lQ^nVLbqfNy`a1D_E25p;L;54+gE;9NM)NYX#r58e)g!nvS(y`By
zkf6>8#MOw*nGpIrBv8lOT6EZF(|!E-SgS&^I&51d(<FlofoOeZV!wGR2rRlP3$D?o
zDH<OdxHsSJEQ*<puvcN|wN1Q-zX5%5%8xxSlHrAggFgVu=LW*sg4v*eJ==mt8Wqgh
zHb$DXabs%AgIO4i<L_>(y&;GvgyIN<AyOBBebvWB2nbO{1_;uS#Mhcj+YM;A;#L%2
z<~1P3J^k}xX9F{vPumt<;6MfvU)sdG%YJoSo620=c002dY#DO7Wd}bE;CoD!%Yqd%
z;l+tI?ibU2k(DCnC_o=?5RCNfas7@8&utZNzKIB@TE_t|2!dmyr2-N!nSxw65{cuL
z)bz)+Y607m$60?#>oK_PW8>;V1(FqI@UY23MQM%`EJ;qM5a1@c-V3qq9L#?59Km5;
zYeT#3Jt^%y(Y<9Iz(wX-{b1;jr{FZXy%_(LsOiM|R6k-qd)}51>#N5Lv4G>k3Zqd^
ztN==H=Yb6W12S~^=gK!OtQqK5D@WLJW}GPi8i-4z^W<H@PplnlaNYMtYhK2-MI*n;
zU4BdglGk&0q!M^q`?+TcISAd72DAf6f?Cj=i=k%}Bs!jUkhuJHZ;Ez~2ieo{DU(8k
zS>j?27}+ftABNbWy+5e6I~m+(BY`KVtq}NZ+NY*Ic%%NL>RSsPPZ&jHCxlJ9!lbDP
zghw_7UYQro^q=YN+ax9aU`#Ht>2)R4Pc(ZEilP2cj0>lr<$pjh`AtDVBg83mOirGX
z8FYKwPxIh3)8s?(8hb&7mV9eY!qTgK5IM8!7)ZUjY89@-1Ri=v4o*gP5ya}?vM=hS
zcFOB?mfv4je5e3b<<mu%zngU!hOH=L?zzPTlL?6+RHAFJCMfPI=10o}KFyR5DzW}9
zK)?e$v6=#h1H*IhaPwp+sEyR1h@UrGLq(SIvMx#sQ}`*0RZO84;-ozbw&|qas}S;p
zJ7i@b&4*sH1uaoSA$C=}`DUTS3%Nr_O$k|LNwA4c`kd-3b9;ATcrOUn>5QK=`~Ih8
z?p4&u-krTpmuszl?kbCQerAXN@pb;hx^Cxp-WZm^sbYW>X*`PMXvxBS`G8xFon|~U
zg}KyY+>7M-<3cSAg-E}Wu*bUdXW_bT6v)-CV-%Gu_4U1i;0QF(L>!+q-NHaINqP+C
z?{dPr!f$dO4E+&)q3X0elT%+)u-F5ijcn^aC2xL$Y-P;ml)97elec&#@F|v}`KEg#
zKuBa!I|S7cwYxXa>?m;bmo*hsHcWS!C{>)YZH1By&)$VA!-t|)Yzw@d4pGv_zhCIu
z0v65LuRq=q0PbTh{-AK7KrD!=<Kqyxu_!o{$dOO)4SGEsua|d3sV@ijr=zM>74T0K
zm|ME^Xjj*xk5oxRzdocaOR7gXyT;!4qj(z0!x?NKr0@k$Kc!Hi<UIE~E;kUBZyI(L
z2LAB-Jl?(*etHT1zPBeYQ%>>A_v@d|MSd>NFkfT;5E3c$kdN0tkmD|8O9^?{%p*6-
zIzI|P%E7ckADOi8?vONr27xT2k1Da_0!WB33>HOW0vAr3I?`)pB;x^EwqRNY7fWcM
zW~HAbZD-65aO2PSt`R9`l@usOS{ZDKm`|s^@<Ixr%W@*Zfp@X+gPMl5l{&2TwID9F
z@9m|EUqk^to){B$0c5sFl;-u~X765XDSX%<5Hc24WaTym$9}@ltE4i3ZB+yg#85wv
zRYGj_A>Bbsa7Lg*)gKCcY6_^*Oei6wb}>=w-cBFqV1~)X6`Ay>o$@3X{&B%$dDG*d
zYO^oY%)7H0u#W&0AW4{?p~A2pB8)=5m{Ij*WHYlpiA=oG$-c15d_i}Gik0Nm;%`ko
zvD=q*mfqI1?Z-m)*@l<9)Mw<!HQ{mmztv}FkwjVuH;QC!plaE|J&H0Vza?K}eE?o9
z@>fpZX(q;c%#y#nk_Z<{j@XDT@n=p6#S7{x%~kJTe<g<7IH^R+(^kBKBfKKYH|#V1
z=S28l6=y75|MP`)?Y~viRvU8Ph5B@&a%mwj${uOEM3l)$(v<3kQ<c$C0KSo#21z6{
zVYlAT`#)GQo>O|8^EkBN!JyyEDq8o0+$Dbx%G@Z2dwrj`{$%d-frwRWH_z3AO?ec{
zifLbWwOhYPQ^d<p=f{bU>C4nFm{bvg%)5Klh>~iXc*s8Q`UPF=bj_rFRu&BC988`w
z{UyyD^jjlXMn7!Xb^$W($gndn*DuaJDY1hQ6wjxYU6z^krza1~vjO$Ok$%Q9p2dZ@
zeU%=8lE^5OL^21lchmR3y<}C8IPkhYOun`VR3mb~O&j;LD>-MB|9aQxn&Wg3H&wJA
zjHq(1&QCmvu&rA6r<)%7Rco_u97m1^Th_~7Wri+C#<jFmoiFNP%tK=Z54suKUgJvN
z0OIOZ+HlVs+dwBt!Dt!??*`$+7RfNW$in2m6@REJdux~;%c>(%L|q~0vdX@<3vB8G
zUzM~`u(^UbLATFzcD?BLlJ)k2z{4mZisIvDlSC5@K08@8U_2pCklunUHAg!qBP>xc
zRLh+jDmJIas}a=MKDjds`9CcP#l!tdn#Og<x=g80r_n<^WV7S<I7CFaa7YMp+dbV|
zT1>;|H)~)%>2uW&nNL&KvLTvG@|s;dkzJqR^8Fe_>bfB}5qq9`@xwN{z5PDM1-)y!
zt++R$5Vi#75&Cn)^LKiFZbL2TNY}{Zi#>v8T9z79z#HGDBQj&g2S+Pk#6X0Wv&Him
z@W^2up^_2JV77+OLmMNf&vkdKw@4$RocB%3#{Ur%qRALSNYo^+kitpc4Xv`goj=`P
zO{}4U^D3~e1bl`(%3Q&#%-l&nfM>k%;?f0AUqspN{+@@;KePX~tZQ>kuhf`w$;?;E
z^oBi3^sb<7!B!d-4^Ji$!c86~UC;Hv6^P%o{yJF68TxJ$pi&h(6_4dBC|(FWfkxNV
zIhUbqq~mHL4x6`+uE@pt!^jup*W>k@2gW3Z<h|(=YbV$lNkbY_Q7^yE{3oQe(D3$5
zH8CNt&k2l2{4r_0B_>+PS){r6EVX!W(lw1@M-3lUzS@eY$UEYhbpE@9si1%*mY?yd
z(bE`5un!(l;$Gt<iL*|dPC5l&sg%pj&J)k_dQcm~%)}+5p@5Sw@ua0ih&Hh6!kSee
zBY41Pg$FSPw{0pU>YCw+VpwS`@;J%Merz6Ag91x_EV`YTj(ZjVI0pVgl;ruMv*l0H
zCk*Ax-YeG{f3xx)=nfr|Az=WMZR8$2<U#o`MmtHQ^a!%NBs^Q3PfUrT2Hb(mr<n4s
z99DEfW8&dM{=TQY@%EjpvZ`LI!_oOU0&=Lp*vBdaSbzxx&hnn3y{<R=j$rm1KecF1
zXx7d6%g^`qNxTXd=i^>6<}GrOs(h|xZ!*Z&$DF;TkRD&pPK1nqH0P^p)p{@Iq>sXY
z+iQzFm)}@^@Wn7*JRH!Z%b>5`@5*T*J3isad>Vw(WM4Q#!95U)9nb+T--(fk`}@Je
zt?ViY5h3-fU`vXsyL!ufRDsJ}Ng&p+F-u+9+e_vFkQl#J`%0cEB2NcTTS<vIdjfpM
zGD{oxvLx6Lu0(IDl(hMN2G`+mF2h4<waC3Gniy7-X>1lVIBazEOd#B|@d;VoR^UdI
zT#&RQZXUfqs;EqdH2I`~Xw>;f7H>nSO!6`Z%g+UQX)BkuP;mH!mfHh__l_7=vHrWN
zZJEj;OR#FGV6T&w9cB$8FHE~&rhVSRJQcxnd&~+?piM<v^_ublqQAtWT)_^rtIjx=
zq5`^JNe>UjA^l!?k~4q<k^sMVgN{mKL`OX{O2o%<y*fr<KX9q4?QF$kgKrJviVOD}
z>xLA8y3L<`t7wB-TdyrEAd?&`U4v$#*YK0r<Q84*0uRMd!6ZI);m@7mp!e_kEuxrI
znzL(A-sw_PvLj^61Drn((!9~1Z`fGI4wkYHP-`XXj-Dapg1@}^{X-5pVy7Lwth+$2
z{jdvI=72Sv#Ln&{UBIiB5>`9EfelA>Iywv){Kv(E1<>T|Z#%iiC2-3H9NP%CS*a?M
zG)&1IG)6Ti@cd*;h_<DHt~ibe!O&ho=x{HBF5j;h-_^iv8`7-b)~QHMK2EY093g;&
zKvL+0xQ8R&JsCx@*I`q3vS9ciRB?h}q(A8^c=H^3_yVr5NqP0?z=1&Hh=HJ&@bbC5
z)?kvocfcx?YxBVj{)PS;I4<v5MkEKaXi7Th{euvm3ju6*Uzc!kK3RW&wCJu-UT=HX
z^^{5|NJ?5+$Qqr_7&;!C_73x2erkRtOZ{dsclh;GJjjp3%fGyyhRlXbtP()CyaSc+
z@1!V{&CSFp%Ir=B98eLD*%kAHc+0}Yip1pf5tvVgaXhs1P7;NyXTvLZy`|7h_}D)l
zg!0ECkse&@924g}p8xzwh`vDujcQ~{FtIa^T|=XcL`~}ZCXnQZQz?y3+TyfCLfyyt
zUX6HUD+fK2JzhAPyssQoL6(Okn$w?ex9?YwgFL84O?H@sGxp$P6^0a8NR4VAkkf?}
z$qC%lh|zOSSD^rs^Lj#c8qisEK6eA|^uR8Z=@e2liod3d@xc1yc^Gaqo7@7M%0=Q4
z!KgNFkDIYR+}Lep2DiMM@h45EBg}mA?}r~I&ih2^{YlT?gxrXK4WUh?VDK>uCBOSk
z*Gr~(bF`%AysU(}7s8)+9BL5}{=!OeWcwG&?RBM?23YDG(UBE~)+QNLkL-?#jyi26
zhwX+cw~sTpg{C(;Fb92KcHO>fFLGR$MjIrzsagVenFsA}jJy&`WxYGVXEyLJZagpP
z%)<96<=DG+PGgv?-mmy$=BNb%!V&)j=qf{YS-({)@X9F*FLepqNWpkcu<yN;9Of8Z
z15ecr>CN8=uav^P2aWe<(B>BqY3pdJf%2K_<q!ymb(|Ed9Qs(Da#98@<iFw%`=>AC
zIdSv>kEL^Ac5)m~+}AURB-H&tVDq`gxBvMt$Oc>rz;XxvqNEy7qf$IZf`Aqg&H}BN
zp&~#jR4e~>d^zuL4imrouOW)<KY+u`%q(pG>(;MUZN_O6fYkRv?Ga)|hXn+jz}sb+
zQl;FTC)Ue$4UF;&+c-)^l7`^qta~5`hI}fMzzZ28`fJxd-Dm<}I8k1mucpocI~!F}
z5Q@RAsVZ58O3rSoL|)DECgdRpo;#Cjb~raT^&#KfsSv|E4fD&+Jd><iA7knxU*J3N
zbL1vbV~0_FYI8TuA7AyFf6F)HtM~oMjCiq9ppr@%X9)TlOj++VY6NQR!A$Ws#951q
zyl2B_DLXm_!;=}_tt#Iq(|MHWd53)ys~bCy02@`-lIKvxmQ3dW!rbcRit1Bt3C<_T
zH_btxNfS+lIkUzMX}AeQHV=6d(asz~`|s1B3#%RyPm8lXBAQ|@tXL{qq)||lPGPM0
zAhy?0WkyGCuuYoqtjZCdvfsC(i;hLruOl*%Y0My;JUWzIY@JPSE!0)g`rXW~HgU;C
zq7e%>_>=_Hs{~DZw+}BF_uc2<tF6<D%W3I&Pvz045*JS~t`@-WBk2@Wdlj?aOFRwx
zma;POJ+_-^<AZ1ct}85iA!)f4y)EI~=Uk`nJ45oUd$?j@fI(le4g4J+H)aFo(Q1oP
z4-!_%2fE>m@<=Pu6^t-)aFPL_S!%czF|0#Og9Pa;%Twhv#mc51gy0uNCTKhw8S;P&
zvU{r-><_3xd_^D`%(;z_Dw``sP9WeCQY01tjzW1cyPLL5S1qwgZqbkNQS;&`hpi1f
zi?#VmE8oR?%Q-Mus_2;+YHbPvc6Om(uH)=D`~7OT1%8lU??2l`b@lYB`l?9eDvv&c
zD;$)p1Pi@sAY{c2ODHpnO$_EYHa{%aU(_8V`oMD;Wr#gszLgQ}w>XK8*r9(1iiNDi
zoE_J}3AyQj6oLdd;#ssd6P$MGv-d)01*la3K}`%V%Lv^z#Ho{$g;@^gxIy>}%~JVx
z;iV3dyTu*DE$1V_-*3aOsGPar$YXIiOIqg&$lKAu)aU$27<~dHQ~Obp8Dhj#3MG(3
zL;Fv{O-HRv-f~SE`Kg0<X>hbCD`JHrvELz?5}L>#86e{h6iUDktb5t{r2b-1vG;ea
zwl-u{choGSnmHS5_yBo{e+TYCFDus03r`;4GP;u?<!q*0W<l3WjqKk;2YaEu1PT}6
zdRw<4Xl_>e*|FQ9>^OOjBVH(XYiZ%KExnd6v(FOYykFQT>I62|`Ji@MuQupTC{)6u
zj+)azmjp#RiQs;XyD{k~CGAY@WuqrNi+t_+cZLrpI*a<^y~KHZ`~>Eh9t7)okO^X=
zo_wFh_2RJIheCS<s@xb@s+1#H+-n4IaNh>S*&0!cJ#ccBCnNpTK^g~+dz!@M3aGMn
z;unUv!E1G2*DP=c+Ds*xWtDNx;Q1fV<bv=kpM};9FEa0Czr<;KL_(_CsepINo_>HK
z;$2>~C5PyUki@qsHhvOQEp6Y(H0Xy}Jn9<)S{g_KHMDuB0$Ak(dv{U~N{1HJL-juU
z-yB_^7+IZm8vgUE^&jaRW-iYExm>YSLo)uL4W;j2xgtHwA2W%ljksi)w22_R(QzI!
zL}2YATXHQt<@3CWdig}v6!O;cy8w3-_myW~RkV0H<oG`CVUFE;1pB_vHnha?Oti`<
zgfXfK64ews`iD~m@&1t_lkOh^zl?T~{wr{GmTy<ekcg+&d1KbcuI($+$%B`#c^z4O
z?Dmr>dsURwnrV|`IgA)-&LHt2{UkFw+Px$MhBu~j0t5xoEtm8amomcIi*!e<)kRJh
z2i4H+^5v0+I!DYK5#*04M`>@Ix*3ma0td;`&6TA!KORCzC$XH=5h0dRjU-B{_3q<S
zxm-lTJ1ut@NZb@F+kX*=+h}(Z3PX`k9{IOT_x`@{`T<t~x#h=nOTuj0=a~ro?WS#h
z<PK4%#BO@F$G)+}QD<XESJ#zv<m;*C*TD;tAQEb6SnNs>mb|oz>E>UI%cE+m$BAAC
znK;izRmbkLrlyVtoPS?<d5woH18U{`ZMu#LjXBJDZApzhJ{MJtA52yZLkuZ3GRqdg
zA<d=G=JhD(BoFaC|H@rDBJ8sS>|hiFUvuiZZU@*ml-nL{)I%CfvH<5Rv5>>IAW)}A
zdUV7zJ7tm`8GL5v_q~H_%l2mq#b3DT!-KXwbl7?EEtK@0Mr9@2zN22ZO4A!ytqgT1
z-jQo11+fyN!9>BlCnVhYyOqYsHau1In*6yAAN3n5gDsKg?@>Z3@V9AUX9xc~<b_+F
zEImTYFv3vvh$1eT#>#Sq(kltm_J&tsp!C8H@FVcD7qsxJt-lJGfq?jB)V`w`kbbu>
zQ*AcTZ5L)%ML!z|SOJ@gDowGRRM{&mrB2Dp1(!V}eq9ORP|+pd6LM}uqT)24*-b7c
zWK4I;+h>-|q}-OnLi{7VIOGRI<!)ykepG5Qt3*$x3XK|=9^^RiJ;+(9ffL@a7Yogr
zJMTuKb<eu|N1*UaUL><&QfRIHT?&}V1_Zdh{r&je(PBmwb$ho0(nOY%7UXK}94;k}
zN63SF^oK+QZ^)MTdHyu9#1zWJAR0Hpvjr18ndvz4Zyvqi)6@|+kR*NFb_hWHW0xEz
zuRc>K&B2gDO$M0D3?!~SkFo&uP~nX(e9D40e_%@jRZg&+1V$2<noZB7!zYwGLqN{A
z6%n}C07kQ<Mg?uouK(XUs*8d-49*40_-lzTi#Bl=x&KL3U{g?w-(F~H3V}y|@|xo2
zmQ2yY_T4Z<7z7CO{up`+#i(se{uG`#sRGE|B+?zSR*igW;R#|qf0$6MyJ&F7-DC~!
za)OL;Ws^_IIa~!fS(o{|sxX+OghIR!)XI5DnFr#f1+^0nu1!(;IAM(_w)Q$``pHN)
zG57`m@F<X`&-<2fg7^J^7EjYr@+h_+fDAKdi0oP%M@PDV*z;TM5e(>b+2qUEI`ijj
z-eER}iTNGf&=lA;EMDO|fSWfb>q>D;ocpFVtUXi51s>ywxR$U}NRvXoul?Qs+IROR
zZK~5m10YHV&k(8;%#*_W*8BV|-88HaxI-7Sg;Gx@dIL-juL_#eAk~|6n16B<qpIP>
zId>LK39BC+LoH|N7RId7@l6*I4`hr*f+c)`4nt#))-oAJZiL+jJmoiY5#ee{1WH?9
z;xg73x1JqjI$0vM`mwHKO}7PE>O|IQ)I){^T$a37L0@J#Zpf_)D>JMyz{Uo>I3+Do
zNF*`&ME`e|#e=u<sS2`d@o%Y0@36#c(y;0ld|C)|pUC~P2Wk}`dOx9~?f@n=NE464
z2Q}TNUItaP2(^Zu0t~Bydsu)B)$P*MfX~3)SY(OTat%{sO;56dC!>n;?RnSfn49yQ
zqs8Vvb~LT~Iap64F>_+NjOk|AWvf-d+TW{MA!qw6?a*T{l|o+%)f8)PBpPE;wL0fM
z7?-^$z>8#WvX{ZIuX)#CH}LTQahg&e3Df)e9ghMy&yC>4F>w5=-cPY_*>foc{VNcJ
z5NGu*7TPNo-12^tgyT)6NF$r9;vT3dHt#OeE!ZfG2JozHWm)%W_4Txc(2V=aqa2jI
z=))b0#skv0kK93ia8F2;0QG87hT<;eo+hv+8J+A(_vuWx&)(xcvDb>cS?OYPpWC7-
zV4CH!6bzxT;pOOq#8_tlz<PDYEFuY;ih&(JD*!Jn%ea->ukxFojRNHoD!y+uCMiWm
z$fMTetw%NJ5k<*43FE?chc3s8qa1|A!}MI$S0B-=5ct#`u-MwTpFW0SiD!oxZ(!Xb
zKu&f)$cJ)`4xe-+G%Gr+s|u8=iIV2w)OtbX+>giRNmBapdBq-Jp1Z)~3lbswwEZ$b
za9%4i0gtFLK+U&z{4#sy|H=`%H%A<kO#YaR2iAXn&Umq*a@7UFwHLC|9^u8YZ}P8<
zT#9bgVf=>-t8CG;*=(ZW8!4h`<RfC6`mGtQgSJ?}@<9M^TWAof?`fTK4m5Bu(V6U(
z+^@n7?GJnJi$Yw*<k<+MPOMr8@ZRNy@gK?OL(Sn>rG3a5h~M~)&~I*j1-Ia>#|l4P
zxBDs<x@F8g1;-1Zd%nx%f1pEOdABpWp#32RwR|<Oqw+irsR|0-_IYLKmjy<nJrP<L
zYVCsD(XmXw7A+x|88u)y1GP%f$hr^Lt?oF&IetKRyEFbXe8Bb}IS*FA|IHo{{eLqz
zY!emjgP71lcfZhmwI$oUU|UkeUC(vWGN=3KqbnDHVgHCK9Qb{~EWsu7ZEr4aI`?gX
zNFf&-S<r|dfQMp@kR9`j-=l)#H71}!VJAU*jj>Urcp%1r?nuPL^yKl$0-V?%2z($s
z-T<CDGy10fSvF?2Kg9WC(M(ufVKG1lfELy#VykemJx(O@y-ft9E1(J@wwOe^t4le|
zW1A~*gXMGz9uKymu34GHt@f;e$U2$g<Y&RV>jhE;93n$(k3TF5n1ZBoLsUJomNJce
zr$*Lp`agS!x1RyXv!9!*_#!cLU6B9Gs999sj=$WR=#^ZDvTEGhkL=2n+h!z0jaYQR
zSjcT|z$<qqUzhL7xobC2Kn<5ZzxI5~s+OKSmsg<X^vkTvm+Yj_Ka30C$=4%~-+R<b
zFJ{jdb}PM3DQ?RYmva}V-ILSb;;o#NUDqDtxne+veEQ89WS!{PI^RyMbEoDRJt*@2
ztA3D;{lCH^T>n3)f|}Ctn{607C+ZfJzl_u56MF<PZRZ~YDZstVS9d~3)5e5sCjRwh
zzrB3DCXvTpS_T{z(MqFvdJycsc_7F(kHwe?^c6Jk-dyYZd`&jRSPo?-t$zVfMJZy*
zC(kealD`t2rLMo-f2JD9K8!~y4&UmJ9QEs}yV2Fzex5yeTm6^Ao9Z`5Wq;mN(K5-~
z_Cir7p^h|US&xsZgU}%E5fgrP`)sb&gd#pYx{SH)cUQBkU`b71vD@yA#X05<)32E5
z9-MyG^3MI-p_=(G&z~50>ykgisrI*2vu~H$4IMxJq_?hm*Hxy7SRk`)=S(8XG21@a
zDm)?Wd_r8mV9DNAy~#3p*>v}DFJGO>MdxhQvMs-T)xi)lmSJ~U;2ir{$lG@_8EZ-o
zLaPF()_Ep$FyA^g+htet4Kfyg!@G*xJfX~q9qZzbzU!`smwR3KwI=ViMmSAQMbEBY
z;ZIR2x%$z`9~A~g=3>$(!EoDeEEaBg$5J<r#;KU4S!jZ#Voo1R9lB!rt*S1H#{vw0
zTi2LuL4k(5q2`}=aQY?5Zw1)zz$+>>V;^&{7CsrWSLCs&izuf_Rzc?wv<O^R2QU7C
z7Sb9(?FR50ua<2NLF@Xw6p>5E-33dp7<bSjCg82M8fxmyJv2%ZAY#tJR6`k4oB}kY
zfc>bFNFraJw+QEa^*Q4A@tY72BYP9d4O!z8-y8jIw#i0B)2lA{HijL#bA6u+??MJ1
zGm!ow$iNQ4#@WIdvUEKX)9w8^XfhIoZfDVUB+*2!INJtUjA=2~LLFXO0t;8HO!6c7
zsXXuMipd#4S}Nyu3fUeiMdcC@#gCoFA|GNpPQ$`FFt}d+NT%sB=pw6PS|5**XLD~W
zB~xpjPrGbW?igw?Qa1Lf8pOVM22voS%I^+9z;?PtL!xbfK9b#5ho)DH*;9g|!aL{y
zE(LPj%;=rzx~HB2ky4R@g5ylN9A3k~-sEXgiK7EQc*{Po_XGI<ezxO1$1fez<o)qm
zy{^8#8e{6A@7?n3qe`fZeM5OHk_4H_K1b$y<L_IZ@ov!Zi5z&FLvcdFk6Q~o3^^iZ
z=^xuOBv7!d$yY~G9+ycV3H?{+A^gdvm^I>=lZ|HF1_dgcp|);Cc9C7`oY6H#W(og8
z2V)Fz((DsuT>b^-r4w&PzFgiHK*e+V%{ZHdj#>Pesc99F6pWlmrvT6esBt1)(g~r?
z1*cjjQh*idYQp-_1fX&f?H`6OEYl+x)NMs{Tnyb9FRPR*Xn4Tbn_2iet<3jmr+>JB
z_{0cbn>c3n?g>a?sRMWT%r;_jN*Ou0#TTqRWk;to-iCo<3jX$iA=51Qi8e=+R)P_l
zQTEFQuMiA41pD?^`c-c_fb=FEdJje%1Wa;X`HpId625}u7S$Yz&i&)bRnF<6Jcsl&
zDa02eLxdcSMpOi|ls#(`pHP~TJkG`Ee7*%+gtd2i0jp0+cbp>H1jSm~>@J-+#5=*u
z@Q4ZXzEoOVZ~=WHD-9n8o*&NzW5ihH+p56<^lTOJ$ojipAoulT98+R+Sp`FzMG;vG
z&PsR6ns4RYOApq%U&U+=We7>Lql4+jUEoD^JLThRjAt~X{sd@ya{f4JhPy?o{nu2v
z`7LHk`K{`8Mmb6+;*3zy^}kACfS;AQNWaX3Nue&agOnp#-G1Ztb1vlxxLmwd=p<iX
zC=AUL;q?RH^V#ydB4bFKuI$r!yJoJOfj52MTwUNg$)G|DP#17e*XiC`FypS;iT$i8
z^i(SJK#3wgR+<4v@#$6?@uy5<^f03ymSLfU1C}~~;kOJ8i({!z!qRM9_6f}mhy(<!
zyiBD%fB;99#Y|4Er?VG&7IF1UHq8d<s7%;|M_oY~Pk|o?LbatITu?_S@cUZl<Fp_l
zk2rGJjnv%fZKf<Tu)E@w>?jaY=}%@ff&5Wf9ZEpJ!;|NY;|4hCyT#Bt=J9M51DwdA
z6`7K{JLBRAv}yXK0z`rEFy2S1hvi`wc)mhoKGZ5a+zEv9wjJK3-QP6ej?==?h5BJN
zf4{Z+>IU{7ur^}>W}sE%t??9%#0CkyX<bFvM)plFEg;IbJO<z0*0T?)JP+Iv=Y6wi
zYy>BKj>^&F?Kdo(N%X3<D&9~<|GVH_;LCw7G|@vIAFTG6?`2~``EC{%h6C?C2HsL%
zBl5V!0sYa3)hY5&c2Sw4bOuRwmCdP6BP<vQe%;UGXZk8pcF7}`$b)s+jymW4#!siY
zkP~C|+<&)MnX%n%kNR`r;ghHwdO8yJD9oxAAj>XeKZqqy&IxVJ_y20l!v&$GE3!Q2
zz&E`~#Bk;1-2Y4oWuQ+wqcg%IocQV!zFZr$(70#<beS}{_&J#IW@XEZhRadjsvXE3
zQ^hLDLyZ>k^>+dB&T&nSWu^T+e%P-9=MdzZ`~$-KInUE!l=<ebd_mVkuFs8ponW%c
z!P0p=O_^W@CGV?v>&4ui9lAZ%+%C2C>GY;03B{YiDt7YZ)GjFZtclM8RqPZqr<d}$
z7iFD&NucjSCH-#GCM$?9jNcF^q3TutW!$iH{=e~17G}=>9jl`8PvQSBR>g|wI0211
zDpMIm#!SB%%IyQd3b|fnYiIIL;;-HOzA!XJrJYg;3|<;63S{pzYgS*-->mxoR_E{D
zcJCs_+xxv*j7VWogsN4KZPPHSRkg@yH+$8n_nRp-gzpmmcxgAiyV7?gt;>(aR$HqJ
z)RZ2t?^Y`-+v5B62;|OgM+>M0-)uCzhR4(BYns7ohLD0CpgjMjZJcDM-Sa7hRZDUB
zi!A5EJHX$WklBr&812JL0;lTmJNO?2I)jt)xs$WLoKH_KcK024?Bg6N|B7I&3;#Q>
zYp^xAKI<5#1pYpfjL<@uIxQQ|l;`!!Q1_$Q+MBb*YHz4&+OLkm_of=7Ka0i&a|7UA
zHQTqr_A86&@iH8=jV5c;))CdaeS!ni2Uql9I_*uQO^2n9e$uVxZ&WX2=jW=2bmfB$
zXNR#xWlcBSzhD%biekLIJA&HgbgPAE>inDb$vHu}yF3$R96V%Jh;>}5{TXj6ejTMd
zzv{_(hulUAYGkOX7&=uW?iSajt*9Ei?#0@$7BvFGp0fRdP%EDSVw^(*J{5NNpHm_M
zfYl{F_eKy#!MI_#o5Kg|TE%GKylXjRA05mbJZNh1Ek}2aX=d=}aAEgla7tOhaKXM_
zm90YKyyjKdBnRI!<Z8Ce()#@1I}T6#Jahr@j<YY)mRUi@76umf?)HVg@bY)1J`zuR
z7PKNoqkJsr4?dO6Q0<+dg^dJ!_w~Yxr0%|YoK{@FEnEZBH}2|Z`kp7nru#%-;V%3{
z!+aG>vd%rB@Id9IonMLvi5<1(1bP%~wG$^uvVw#ExU}&(l3}2cmV!s$7}!!Ug;fm1
zq`Z3^7~i~rOG5rsAjSS1y<Z_8wA3rNTdQ@$+m>z?j{922SsX9u%dYDQsh@&D;bfXk
zF*#})$?GmYCnXs%6Q%sTnTS(Tq=^{?wv4A2kZgP<gjbb?;PBY8vB78uEWhggC8MoC
zM-_sdjZ99CQbC?XrnXvIR*~EEr<DAVIJ&eY*YvNSpwX~n-S-w=8hF18gg_xdkC}wO
z6k%*hk`mLbn#Yh`rVZHkB<v;9T)Rej0Q(Gd-`b1{L7cMK0e0p?mnJwiv3z9Kd}LCP
zPpm%EXBpZyOOVT7^nYLgRnJkeoiD%ZQbH)h4`Nb6`b8UbEGXCuFM=>{Nj`2HFxJsy
zn?pz^k-REWnhTgga<S{*+yE0qhhmAU$nzb-j`opJ*^wnWEY(zx_DF9r+!=qbwedDP
z3=0Nhc~s<2RUzsPZeiA(&j)7a2oN|tWy-e#E>^!MR@1=_2+Tx2P_>E)(uyz~g3(i)
z<6H^@K)JV=#$QG9iFvrExo*)zo%9*_D~X=P!~^tq+Z5&hT*?ItB(a8FV%Av``xHSN
z;I+cWDs~sXc-?POM`udeNTtCJZDTO=sncu@k{1>>=fYaW`6}}rq7@esUhnI6XkNz9
z!F&Azv+C6CApD%#a=W!8iIQ;sS_QfXPhZdPK50q3xgi$B0Oy1NzH+FuY7a{Q&Ano(
zPWtXTrlvvmvV!!gKPxWwkwpdxaB$KlWcf-{Vxxe+RhL>sKl%VIk!>sCH_bps`y%Wv
zaIkK&#jFExe*R<K4tPOng{V6C_S;wQ2dTw7Xewy6QQRSeiapIVJhznH<?9|G>K=kH
z{k*|Z$7v)*UO!eo_zLU?0}&Kw07Z3{O8kUxs#}CPG$Zz1#pRKVnGtSvf<syRIMo-B
zpNRDmpM(pPJ)QUHZ*Tm>unVpoq@oa=lZYmLrSWd1qA`<z$7kjKAI8q9NfV}Nw{6?D
zZQI7QZQHh|ZQHi(X<Ku*ZJXaj?8E&WyuYC$s<JY3t*bNGyeoy4+GR}kcl*ipCwDIN
zBaLR@w8AA1!48tb+KNy^A8_4Qr$N^_$YOxZVf)R}H9*uYcpyVVRFfatoP7YM0Nw&7
zauw%r22`;G#JVzcU)ZD`CDtE!2IPnFxHXx|rK}}qSs^G)QAqnl1yCWL5YC;bHOyqh
z`gqAx^g-HHnOVRMjZ<5m=(lCQ6V}315RoHw>S2}IALbId%CX_)BEKqvNRl1)8q!pV
z|L)-e9*uX4e@P?*froZ=o1OAL>%N$#QU(N4rKCH0)@^nL7$JuNg%+s=yKVoaJkks6
zjr4+)?MXod3}|i_LlM|R7ZAnUKME4-a2Lc|<v+Z|MJ9^#s(G$OoH+W$`?ru+5o#<}
zm*PxF6-d31$kYWUM_DH-Yu1wS@R^l_pv0?2CfSZja+<9NBgjPaH8qg-y#Lo-`VE(7
zSBQBjwT4JAa#*hw7aBTe#Npd=dPf%e<}G$h95DhR>?m)|k8@?gD`6cM<-PLj)_m3<
zGFuYS-FR7T2!Au~1?d9mpOCKW(;GRxHE=g!qw#3AJ$#A<`=#y&6`0%vAcshbOFoYj
zAk6lW1=QN?H)y5u^DzfQq`{iPdu|kdISS2>@U!*)U4@{3WkCc>w=F5nP}#@`0jiGg
z*BuZ|V8S<r1EeykhX4nTWSV)K`OCYBsu4D1xS$^MHh$GLf1p+L3`~GRqA~+ILrW>p
zcEO77J*u>x$n!28BWNvDS_-EuWW8QjmhoamH<(+yJaB4X<|V0qe^VeEhUK1W-|Kop
zd_26#3O9Vc1{))>^Dyg#>k?!A6tVnw>@{e9z9p{l-6BvvGzuKsyhtv5o!bp{7B&HX
z0aIwSB^4NaM+|QMZ;HeCou$Yt(Tet8{-eH{@D@M!Sresn1xm}!2S06=v_eQ<0S(+K
z9CB#x6oz8AEIx_}^X^ygv@}r!>~ReLAiulq#ML${!)8_LK`k%ug2$^i-u$k|XLT^B
zAxTtOKDaJ(unw$D(*z`I780(sCcISY`V7*O-58R5Xy>mGNvi@xkYk1P1Xc{NbPnVx
zl7)2cAE~Ni@+B0e8bkFzk40t{$pPNJ4}=2tRR4*naQp|N!pg+K`oGT7S9G*o4myzi
z^9=^u=hZMUUX&~SZl^sxI5Ml<^-@*M?uBGiM)jgv#3%njwB=4pBvj|OF{;UY*dRd?
zpniT^flkipyN4$l^$z-9x2P6T!V<qSH|0<%PzT9}bjdiA4-KfvU-x%713vaBgXJwU
z5+iYm_Qra=Jf7Z;-`Ap%(Xl8Bq7xIidcSYq->(1{?R_6#D<}U#&g#D6tLO3qz8>cc
z1a$SN^0q(k&x-Z-(GlbZwPV~@aj=8f7A7KTN@5>vwtea<j-dx5twb)Bf~ZL8bqpFT
zz2Z-u#Lyie`5}3jibN%eO;mXgUzkQt5s>#Zp-BloBTtE#6AYPvD9IA>u!~TH*+?91
zIbkzgxW!UiP~k{qMo`N_c+(=0RyLTq2Y$0ex}c~)1Tl*&Woy}wiY|SaDbw2}m#B5d
z%MVc-7qvHvMgW5*vhn4qF8iN3>2ti@2WPP5zvk841^l#Z@zj2wV^7KK=*ipNw9F1S
zH2w%J0va-l@HD1AtTWl^M~iG}1j~Z(L7OkWG>^uKv|cYdvP|WV+1-`p?>ikZRDL*p
zsjHor?=i-X^C0ou!A<UclN9H7l~G`d)F&dT18|b#Dojer4AUlUO=lolaTU~>-RJFT
zH#m|q>i<aTmv}ShB1>gkCB{Sw!8QNUfByibHS+ZdNTkpzMZQoItT~ove%GYO-0;ni
z{4F#O$d0vi_P95K>%-$w3#rhpe}5S)MH>nxbg8*49p@>kap~hfk)rA_mL39L^3#Wx
zKpLJc6WSS`axQtx8ab*%#WX1W<A<-MBD`%y#C5gH3FgU`<T!)`?w)gVyRDk_$iV1w
zD==#Ey48hZY2K~|;GO6<l$+Z0hC9lIef-uBYK8&*-KB-mUR_opvjggq9yH5`WNSW)
zeAHQf-!M5@X8F5^K=nKpAo^<|)WmTq#_QPt%>c1GZao~Cf})0g;7b~{zL%2?CYsNC
zS^!T|3}s~fmSmWcM7W@-{m%n)nPKe8?&COqIQi4d<1{m`V_a5VMs5VYdmR_I<;lCx
z_1S2jEMb4X=wE0l#1{F!0SKr9u3{gU-jDS|;2jKR6DBb^<m5IW<mgB1zg+Rd;@VrQ
zZ}*K<H=m0C+(_%F>|n_Bknq-&v}j3ie>tCI`*$xcx$A$70gFW)lij^xHyxk`Iu37^
znN08)QXu;+>kcDic7AlpbiMV~>xJO`KF=al;zJo3f30RZ1IVrjpq8Mrw<Q=5iCq<>
z`ApVjj95<iYvUAw3NXw3nz?on__3LOpk(cgpD}4hcdbBoG(UH|WSv1gxXB_jBGJL#
z>ax8X`$H*4FJJZeFWbSY=o>T`8^Ef_Q5pc2V`~S442-+wY@B}j64UdKR>vyCzKq{Z
zugb|L{F6TgTp7!O4f-SqWfdWs<(QD5!c*Hu=5B&>7BV7TWJ6um&UlXOrWY%&OeH2I
za629Zlla0|38~40SJ`zF@SQ&JpsT@Fuu!p#GrMjgKr>!<k^+q8YV3FY0QrK#Ka@m{
zNY@@8I&B^KAM<u(hv)?bG&AEfpBpc7e$2xVD#b&K^IHt?ev+foGQvPPBR2WmwVK&v
z5x%i!o>yPjLBoTx1anx=b2!vJf=a=H2?pp?9@mg9sVX=w8uGn{S(z%R39{MRA{GEO
zU~^g>>2B@J8CJ>KIKwSv6b2<J{qbrq)T&?y^8~%)BfSA{j{I5-?vf&}d3oA;ysg5i
zR-ClA9)wP$7)ZCpGnk~c8fY2al%-wtDW_{lvxjbLEw*NHzl}KTIgR9n1O`@!ax5Cx
zJX-QakBy=XPBpELzqqcx6*BDvY&)h?UPsxhbcMuA<F;C+tGZ2{KPb#*FZQ1Ab>5S9
zNwpcnxmn;gfqT)@neceg^?Xm0z%aU$#q;HVec!Q?7YYb_t;)d#0K!5Sy;*PqS1I0d
zRBeJq^xE=fXzf{(k<$71Z@#YCb4`*u&8>vFo3<wU{IL|HB#btcytw9>5LTaiY=d!B
zIEto?ICJ64kRe~ckudLT7;I&Rp%m?u*S~9PeOhpcgg$<ph`E$<-W0XYevYF7zAR2X
zgY_>PUwqBKFT*!?ct<O~Rp@g3xe=gkCjtoD*{T2(w7LN6QWLZYZ_VB1eVAF%$a~kP
z@VfSF>%Z_w^LvlKAbVZ&<6KtTI^Y&qv>$3F%M(|P>#sM+wDDF2oJX!Aw0*olrJ_dx
zu5ocCoXz{eC2r-;h~wOqle;Yh(R3YEZ5<8{G#GCs4gPjIbAc*Q!lvbqSSt$NF<A-G
zBI6x2l;Rhuw&PAiCV0{``{Z-l3(SGJ&l;|1`$KrRwd7!O^2XmUt0xsCfd*=P2V_C&
ztfrRnsQ?ScnCTztFSigINwtjvuuzW{AgmQKF*QQp4r@}57H#b|e#?EDilRq_I}fOm
zu-IKMB5-Jl4U`$bVo<mja{Nt9aZ-di9Tw1zLBy1q>A{p@mo5lsN2CP^ghM<c-OcPf
zdG7IW>}=OReLWGsl~jQM84lzG&KmoT5Re$}PsfB(Q!#;sJbC-e(tm}l6O+kSPhwaD
z;P(b|=<~tjvNF5<!&X(UB`sVKlbR-)f0$rhqWYT$aQ`_0zKGXGLxVRn$LR9fdqBpe
zUaE<q0%#>a#&VW5Tbb@Su38FB;bZU<104jtLTg$v137Fq4L>faNAyV%J8&Vq@l<Z)
zy@}3Qhl<d7Kfo6{UNGFdHAUSKp<8jBL1Cc9KOd0AIGDS!NW_s3MBB%{XhA>f4AE*k
z4!V*=$OP9~WdzTqixJ8EVX~7c{E_0BR1!mNX^iERpz!ti0ta+zE!C&yVKn03;=sa0
zC`thzlkkgyL?A(Snfxn#+;;Mwl2Ms38klhv{CG5`A<L@qA!_)}a$938TT$`exY`ko
zE8p3W=LCqOKKu`MYpb>HgzGp9iq-Dk2vVt$=Xd>rlgbxQT6_m=`qSsJ=gs><PTv_b
z8hcT(=xR&#?aF^XSMwb7X3uHxKObc{Rij%RP{L`pMHyuBCd5Z?4inIpd8szM(C58I
z%!Ov4rEtPKJ#es)$b+a;fUk|OE_lQaytW?NG(~=gVu8~jHJqYj33Emp5=R>jT4Jg6
z(gnLUL%l3fE-~2YyTf4hs<C4!=n!3rPSWFz?587vA@vS4eV`0!pR2HgNL+)~J%MAk
z)`3(mV@6%)ar%$5uf_?xu{eAf<$3`K#cv_Cga(BsR%i<^PHffIMk%IN@k15B{){Tk
zoxf+MOYu*8X`+9#YhU@aGkKrBFiwhj>#!$?&5D$3xFExhxZ|a+GJ`0Avyt!LZq~`(
ziq=J6KIks&IqMi}F#mk$r2|n8`tW|2dBo#Ohim|2#7TXG(n(q4Mj}fkvc2lp<2V_K
zPmn&)iLe&|QIZ;4%r=KELjp0R@0S3j58N4V@v>-$qYr+ko@P{|ge+{P8QJ`BqG=yN
z01H`U+aacgL4#Y;hxA)Bx=3XYu3KbkvF3U4Qq)?ZSlnGjFGe}UXiSYyXWdP9m*1jr
zRJA}aa#}l=C<7oBk-kXX=WLg=ra&vu9kwoR`t*j8?xm)G{EJtg^~!%+<WN<kTw|~n
z^;|d+D7h;N^AbzTQhrZ4@}ubQ%73QUJO}jlv`WzrK{r2Bo{xwqTv&I}5_HXt?Y@OZ
zoq%n%+L{}ZkLG(#CtscwTti@pFHJqEDIvya(HV>H+w8g>u#7v#T1M;Tp3D$Fl$Z5l
z)mC?Q!|(BeDbW8}1abqag>EfP#tj6o7M*ptLvk>S-H4NH&B-!d#(qDT7d8gy8uWTg
z!dtZMIfMQ{LGCq}ngSgqM2@dh0cAlGC(SmvMCh9%qzxYl(6(A-aJ^!v=N-~t9``;T
z1u}+XO<4&A^rIteGZQe(VU>&a?YMVm6Jed`_4RTkR^^Ta>*$A_pw(69?&^Jvs5q@?
zpVmB--KC3cm5#6Qc)AK#7m@8P1G|p5b`$6x{KAE+(ob!jzPT*ce?{ThJ|ZybB4p5Z
z+@PN9a>l*Mrn}KE;~l<dYR^`n{t-k9(&p5%{L$a(d1pkp;Me-EE-A-<sdza5_c`c_
z_Ld8P10~?aU{KPVLL|~**Zw`BJ38Avt849b&pb(B9)-TKR4TRP>h=3w$58lp$*HEo
zWeKM@QKv3Y^r-)*dnK8+r1az0gdl^0e%``}M4INPs;d*a6S{Pv6p5*8<L~Y97!(Bs
z%wObpqxJRQPu@=c?$6h*gQh6?`Fl_MGd4Y5Pg9Kjc|Ru+?ez++k&OOaKW~ri(S!zh
z&;mII|8fn$ry|ZOO?ae(#TJT8G@0aM)c&;9s&ByN#KQ1o#4Uz0Rc$dDW3lv$N{nfN
zsl0O6k%-5jBod{B$NEX35Maed8sgYfx#eU94nzZ=cznD;*rp_nJbnajBn_C)C<!3R
z%4q6&z<dp)oIe#sEibde7`Mi8DYMyTdiD17alJi+4$(-9ky>*x2a69$wQn3fAAUJV
zw*-3#N)Tbu`Cf3h+iP8?>WbYce8-;fA;TU2<4Hwmhn|!;xF%ocu%V4==8;Jy070S2
z2P>gG^psmM;;jhd5PxxN>UNhc%X_vixuu3)^?RMwdxM$idCCMW=A=<)?8fguHpvR)
zXnEtRGv%d??HDJcJHNQN->I*!w-E3zBf`BN2i<pXLzVDJ(c_RI7)m%3wVJqp$KdYd
zQ&FO9`W5!n#;8IS-RMQ2*80Hvb8;e253El!4^Xo%%@D(5A+npF`=hGiu6I$$K!B2T
z5E{oq=?{=Ar%8x758<t0bTDvaHtQg-l`oHIx#+ccbxJHMp4CaE@A-8Bywc81`s4vU
zG>JK!d+hpc_Mh_FCsw#)scOCQussImX4!ip)l4;`Hh`$T{Z!1H>#(%lCXoRJqCl@g
zhX?nNAQ!E-QJEB$sI@nF_5G_OeEs2N>6~;<t*y|ABR9X~wl%7l_Gu~BqS~;)45UKK
zlQp(XRscBIi<m?-SdV&)pA#D2pgVqjD)G=ao6X*DCFoFo0H$|srstKW`&skw`VmE$
z7%KbSUs3L|xPT7I+xfyZhK$Bs%$U{U@IZNw9*10LdVh)L%-qbNx5YVl$B$;@?E-UD
zfD|9_ti6Pn9Y<lDtFr?|s-)R3;S}Ry@)xI@-5ykCC5ZB7tq_#aQO+rxh-b?F&F^KC
zD~D^-%2TsZPw4f~*kN)`pmU;~2*)$fJQiYtb7el)!=lRDYlgYssTOTc+6a)DX9aB{
z9B^tv>t%U%x#<F~L)9cdq_p?)YPI@6z)Hq$ou^<5u9S_xJKUi|id*cm(ghY_V)u~h
zHsrN+%ZdNMYJQ}Q`D=CTtaz#MLcN3h+XxxjZGmf!<xC=W(qKp0-a_+liN6vW(;Jk>
z1Rx@*F$xsQ4TXUuv~+|ide(*bz6;N}L3zxrOL~`X!g?;1zmt;m`p!xPL`ILGVaoH*
zXnzq2_<(R?hh_B7D1cU~$OX=lfdGG+xkH_JNCx#X{=Nz+GDNa2*G8Kzimk9vKGH<Y
za-0;TL=tTG+N_q~s2}^lljj#*&qH&;NpmTO6^a$P{19WepfgH~tL|ieY*2ERrXs{1
zkS)u@db~Rj0S>^iAw<hS)cOc~va}JTx?B$l`v(Oz@IIzTjmU{0AH4&GN0$S0KVI;z
zsfeOkzhLKUF@c#e>Ge$%pxsYsza_yj$m@uP>+kF(N8nr^8G&f%m8l=d)$N~K-}eO)
zn?=9ul&{Gr`-FljGK};7At6f(?=aPfLMY)-S_}H}sT|bR+IoGXS}w3-M7c^Ufr_$4
zj0$YRbk=0#iBD6K?*LDMr`pe^d{Oj#T&W4d{<nbK6~j9@IA#_td)-SJU??-&g+cY_
zFJ&EJA5nBy8hQ5b2h2c|<AtlHIVfmRNf|hzjYd{IPGEpa;DyItnWl!@wtZVJr8cvp
zI2E=g)~4=_yV{nqkNJ+M*IF4lx1OWSfd*s{=r-a3IA7iQW7{C0It~MG<Jtrg39=D1
z&zJV3=}(puC={7018)QhoNRGTK090;!-us6(W^kyfM~M4N0`d2E<8!<%^ChpN<w&2
zr#S@aD*AxWzd$@kl0r!km~TrVryqr*;_BA}oOV(->LPFDhOSEmHw8zT66W|*LEjnH
zcbtTqpK{(GpvW^ponAgrUxi#5T_}h!7l;gg<UnQay&S~;(pUkHUW0iu>HTOG7~x6<
zjSUbb7_jhaBu_qb?6v^>{f!qr^O*gDhhal4^YK_`Yzx4YZR?N_X#~GB_YP_L9mC@)
z@g=*IRa17&SY1a90QAgkXAeRhs+Ts#t6ISOLD)7F%~0Ml=Hy82v_2w^E0$JsVyjUB
zmF~NDk{L*mX<2ynrauq3DcswrWqFAMxg5G~hq(1O7aOPU`)GNV6rD6w2oW>R-3s$+
z?}yQ}UvXCCi3;t7TnzB)j4WTit6OrKT^9o5e8t%}m5)izGehcH&iQ$8Z)TVMp)Tg)
zn4U%R*6VC;af_6dGm{ciiXQ8^3W(_0-w3xpF*2YOH+8$_w!UFXMn<`YiNQ_3Zr4uW
zW8*TSf-yfYc%~5z<X9Ei#mcZID`=pUgJ})XPb1f!Aykuf2XoCbP9NDA*`Hx~2L7g^
znh<G6k25#v19-92AxLJ)p5cpZ*KABXwd1XhcC`(2m$p*u6!?xmf=gXcUl#<1Z<LtB
zRTqXb8<IkP^8N>DYe=c}vP@1m^>+&)V!K1@t+E$rF;EE4{o8zg<2n$Y^Vx)OfMY_?
zoNS!r7BOD@HwMo>w@*4@SPQm3ncR&W>m%H)=@9@Oy|Vpo*GvLzD`e>V^2i!FN_K3>
zID4O_{ebdtPG19rHVj7IRw(talG-MUoq%UBkv~vVJwaDqtm$`N+|^cvQ3Z5DJ1aCE
zFoV-ZyG&El#JYsjA)r&+K}V0I;^AUSS7qAl_7^ZDP+F)Z2NnQ-Z{Asopf8M150?U`
zPYKz*taT(O`|5W@)uFno^L>Py>H-Nqk$JG=B1Nxy!%)y?Liz8WiE=<%HT337(86NA
zoWL?V%`R3kvH=Qd9*{3Qrzx9JvI~s50wkn+Hu@<4IQtvZXq*_*b#fd?$2(rcYdZAc
zrVJIvt7oz)1CO$}9LnT|OZ`4VE;E6KqXy{=xD{bxSH}1{l0OLRp<84TaqBQEPY-^n
zI}vH4WPy;Dxmz?9=&|lmEF?}ew-G83UE5DV$1D^5!z*3H&|g_fCevTo>DXl{^A=!}
z4Au2j-v6#tMjs}T6K2+1QzXem;!uabt%N~`-?&HAq%M4Ld1=U1qOguX^r*^VuxZyg
zeG%pg3wR=C`7ocdjA0g#q=@^Sk~i!MunhU$v+=ys4h8tY6X$wuHAKVPxr-xXM%I)8
z<-3EFI`Ek~VY=)(!neC6fiFaw`yVUpAif0YYMoTgZi)2|QzY|EIGse}()kB?s<!S4
zj*CMP#=B=Sa5=!LT|iOkXEVAOzp`H6ZEK_~WU{;&>1p%zTrp|bU<s4?=@E0RoU*sW
zJ|+ai41vB?{=8*2Gl4pmbX}d;f-IG2@}Q~ll(Q$RBSS{NL^0XWoG^0E%Hcbu4mNhi
zc0dMnI6mp@p2~*}w>csNK>iMX%n#FW2su(K3^=mX!VO2i4#?N_nPGa}+cDUr-9d2S
zgdaGx%7$ghy~1(Ykc=qJX@;-fzJ7n-KMc4J<Bk8X3YhahurwAn_WxN%*{ijcyu^hP
za8m!LSVc0lj}Fl4h6oA=>oM?adAUhvir*8DocUm#*d5FK=lvp0J1+N5uA{cw=Qmlq
zcJ248TlKKT1$O?|=i%E*vPFGMmKh!E;&h!ltC@KtN6dw%duz@YvNZ9@PuKf+vqrE5
zXwx)PE-<X*!~WA~MD<ksLs$&At>J^Xt*7hT_1n=@dOG9RJs=v<fl*)gePT2Mkuc!x
zYNk8dtEjk|v%hDYG|v&`{PAU@S`IEw-!yoXXj+7VZYBnEZ)t2b&!fH3KjFsx<JhNj
zrn{-o;omku3n1Og;_FC)W|vNO&rkrB6`d6{VqCWo&9bh|aXFP&h-_g^R6lhKUqNoU
z=CeqU0Au^On}#tG4KXL04rdZ{O18WCsp$!|-?B*F<pT3}l~t4$=IrJ30~^o0^}_NL
zO6L)}KIq|9sX|AFm1@d#&vJL!)n%XUv#C5eB<Cs&TwO4?N1j-jHwBkUC6W{>>xPQ&
zF*7BJ(n=s=aukA?K@^ArPJ^^&k@xr;3%MMedMxPZR;ZsbMF%vPB4JW-xu(4SI1_Qw
zpx?-daCqi7$Dmu+*w1$k;or7to8y;QK2;`$5b9|;N*L92Z&u$dN*e2D&H;I9?5)35
z#;BBtT&B>h(j3S4ZCZ>2uj1b`B!tLnN3pe&@=J7p@&=S@OYk)J{L2^L?&?T;aQzI4
znbrKfk5NSVjkZ%it_==Zowq%^4Z9u#W`?H0FgUTuW&M;Ytq_qG2;_5IyuGJS8N0RJ
z3o4S?x~Z;>wB1VkV^7|5WY9o_QTA|6S%AC#{!w8;n5bTGOkR8!AqeP-3fgH3wi60i
zg@3YmI|93-Dw__{JcHGibVTvy517rz#_T9E7Rm^F)LmqOoQzQ(+%VqR)wY50P5G`2
z)j5*6P~{|3lx4WH`@h_fKh-de;@GODYIRdKEhMwjUXy#lDk(Awx<b|j8}1e-Bs7x)
z)Yz9|yf?r<7<$bW&X5W=`>JdaT{e`<Rwzp4szjW^#QOt+!mJI9;2u@!{<q8y?G5Kh
z*kwr%uas<b@&36MfqhRDH}_S-%&d<6i|%GmT2}B76;_QWPZzqkY)#n=JLy5WyCc`)
zHV&6}w<=M>)gnwLP#^|?=P@2Fa2DiVxrDzGl<TvoM@ZSQ^k~0uN1!-Y7P3=+u9dm*
z!}Bea7OCBucaG8Zpeu55*Hv(M9#6@xUGVFMpY8-`rZRq;+$R~8x%iB4T)zjt%wyjd
zEn_Lw{L%4AVrcI?Xk;KxnG}Syg>m&Ub&*eL4>Z7{zP2pryZECuQ4Oh7xCq7qE1us@
zZHSs%Nm3vGtmJ0a`z%Wyyt2Bg#~4I%nY!?-WcxWuGcJU%>0HFy!k_&O)!o}RRa%-X
z`isC!7KwF%E)hD)t7_%}sot0>BBc}YE}`V1G)$zhxq$HYi`9^^pbXdom(|z)#m%Ln
zYLOKViXH@M7fPw+((*JAOENph^whSo60``(D_;l%BXiKN#au*2<aN7qA@LyN69%H>
z(DU+#>4hPrMC=w`Pq6C&HF8KtHU;n*Zp&flA*PcPd)auJYilV$t~p7~Nw8JcM9;0D
zj|K_WtrFoRjRJ{q&v%8=0KJqDS-BX}YDJ0jTJ~FepkY#lP`jH))qle*BMPQ8oxqw6
zx%v*hPvDx&(<D*&OP*_2e14%!*8yV+CAj<vdZG4N`K6Tl3*T+oSX#(??d!Wqf1_uV
z?E#gcMA@eH#~CEC!8j^8M5?qy795x&9x{InTQNfl8etoGm#w#!yYXm59jq-0vW;&&
zMsli&eqPa^Ovo&&iLI3Hw(qQGI>ch2vjfa;ICsof{|zM?QKjAh2yvK{Y#FhRpuxY>
z-m(>E%^_$0TbL4ADa!m{c<v2~Z`e3B=A~mXVG$$GyU#P2C2KT_+fQWR8cYOp5KJb@
zMiw^mKs9@B(<|>61zv{O&6V8=><pQ8HjMHhvhT6(s=n7Nf-_YZ+^2@v6HN~AE@$FN
zX;9J*l4027S|QT5OKQ66qDUs>*=P{6H(Rf&U@r8pM5(k>Q{PE!&`2~P7!aS9a@$H>
z2mA3L9Fj<y`_$eio2g?(*)T6hj#u*+EQHytMQTn%W0rX+RoYAnA*;T(MH3SjNQfap
z4Xbt?s*Asrs@UwjOa}#ENZZS=6JAVpmqRdNq)eyUI>;JBc)%4<qJ$?ci^T~s`u#pm
zFi5}5rm4KO;gru`-5u2RJkKMRCnc^xA;FYh%$(@9T>K%_=tEu<EG?P%x0yanh`G>$
z8SNnB76~6Ld5spIeQ$vv*RzE8V53!OBM`#4&V4h~51DEs&yo}Z1v@}EFeGJ#kdocj
zTH1+u5%LsT&XfQ*(LV8rgy$eK?5GLG7SXd4<a&u|ZzMDPbgJ`FDSmLI*nfhIdd9>q
zj>5!mxR@&L$WgeCrx+vJhDqjnUv7bqYNGK@U<-ydrV=L>+eZ6(+O8_wBgxEZxNE})
z`wI*w3E516i1A)99ZN6oLs38rD@Er{ak(U~cMONwmIp&xUKZly-J2AglOc|xH$+Nq
z|8cvt^~WIhLxC$M`wHzf5VICw>nfm@<f=>huPJcWuC3lw-fi6wdAcc+y+!YfT7tSy
zrxX6_2GVevcwri#dy08|B2H!S;QVNcvOn`4xfrjlO4@dQcj48zIH7X4M3$gHV?aHm
z*CBbT!CuT6qEne5wVCq>`{#Fk1x&xKu;Vj1t1I1CnWIKxaM`!-0~}goGh?*vs>kc;
zxO<>k_|tyQOgPi{%(P7p@I4T>HFM6!;LVU7+b=_#tFh40uXc%Mo-l<P`+iz@mVs3o
zaLxQm3Q3Q4n71~3R;mJl(2H?=kkb0>z(WKIh1<qLuW^P{Iu5!eOe(Mib39My1|svu
zpho--oQ8r~*kh_6N024Fm*`gf%9GD^3nlU05{Jk#fiYNzEAM_jN(zN;8ee7mxQ{`Z
z(_`TK609)3{Ae1>7@|<<&;2G(;F9-!L3KXUhiLe5)qk8myod{lz5vT#@G9Ehf4)Wy
z&UW)ZQJH#vwm@IznjXiEj%sAT8F^oUw8r}@^Az~-fX2-{cY3f?f4S{xcYXWC_si<2
z_9Z;Y8?SK2Q*Z4RtRFt{*oKlEKYjouIeOc5>)>XzM1Y*Pw0*kQ^P;=-Anqwy`1y%@
zwoFzlo$3N6zg$9JS<Xmwh3vL$YHVl>)iB**kX0v9mT0e!4g*Llv4j2{>B;^1L_e@X
z4zJ&T7JC2H>f~VkpGn>o96c96$K6(cutVtyUFa*3kn4NWEV&(z$3*L@oVQCaq;@DB
z1zIKQ1lMf+4G64KiQI<5xp@+%4KM>n^ysy*V2;?DGRoK0W&KpM1*26FseFqmi*_9;
zL$Rt<QY4!;#)1QLu~@=`I8E~FIOdcK`Tj}0e;1!Y16$Ni&&bZvjsel_XS%`n)A3Hd
zzrX!mTK#s9z>nWsyagjTp##ycq!N_pm&Z<V9$7xwIenH{k}RY=W$Qc4>81g%79?n+
zfoh#+1ZoxG@e*h=Euv3a<9}IrmDEMQaFo82OQOaXHzMPR<e~x}$AjWXw95FyG2p&`
z(ZoqB8yY04E(|37>*hGep^@rV1l|o3HGo1(ukWe4#(_avx>E4?I4h(i5sg640lU#S
zuP4TZrjWS|LHN3NjQw=CoSBr;6Ea?vC#-~ico;(P0Gm>ni&I3+Y6^Wjjd77p{K7|y
z6puvFG~tvY!MRJ3f=~KGC0+VEO=m#cK89TYxn2Hg@Dh2py6vu6ui577unTtQ<%BI-
zzVC0!Q|nKpAn$9DNwcMgx!TYQs|p^N->%CvnkDc=lI!FO$B(A{2R%yd$CcW8rxs)%
z+^w4iNb-RU%!98Cb`Em6)n}Ci&lbf^a83&WhJ=XqykeY<x|a<DkiqF|)09nUS8TF~
z%hEpAcz(98L5M2De6ijWzCyd6m%>5SkX6dBwVMkLBvvR@uM3S7qN^U->~}>1i}ag^
z+MtcqT;eE$+Ph)9rEXHCq=p4(p}H&^K4F_pdX@DG+eFAYDg_-B1WQhN1N6wozZokZ
z4b%0Elb>DoyKJ<IWj42>2ABibQso18Gso^tY|t>rT=<RDV2O0%9BW=$nZ|Nh9hTPJ
zaAN=LTQq(p3vwv4VQZ%@@ZKQN$!yyh))Dja8+J%d?HlhC@V_{?#MFU3?d-eCB-X8Q
zM8r{7SL7pv%<hNkpLmZ6)@p8(j97E1FN)=}sy)nCQVFLwkCctZsZ#p%wfKQj_hSyc
z^GrdLF$hvd5!7sN))h{^o>rH0(JS$b;OIoIDYLAvD%H&3&25}1<5wUkuL|k9Dggun
zKuDgY)+4qeQ8HxC7jQ7kC}ev9wfRDpBc=G>rG&CW@WVkCeulR=bfn&05B03qhSSss
zS!5FnBt7NRTglC|9T`)*J?dU+TX~{^s0Kt9cKrbMTJDB`mt)T|=(X?jF~}u6Rl~4f
zk2p{&gO%2<g-oocc@m5sL2{{X%5_4Xv&OBrL!iF(`RXTw3IqPUirAEQu4`sQ&<Kxz
zKstYsDQ_#M0Pwq75yz}&M=$Xm8#s`Gb;Pu}{YN4BzVe31!-$2vA<+r4q7qP;q0Sha
zK_Jnm`rK<C$%?*fI_~2ND9Yfl_Nm&ibqEB&a+sm+heda{c^I%dK5>8@#33Y4tC8+i
z>f!m8w(E!HvDadVAUVZ^uUEszK8_#aBH`;~lh8~#^$xJPJN{%gSgorK>Q<GZO+vcQ
z3|Pi{z~7%`K?KpfFw!Gz*$AXA&(ti*vP}qmV%hXCdhos+O1<DYQ&Z~yQ8*a3*+JzL
zyPKMjHB5(R4Epf=m!>tsg+R>(VHAtFx?7W`A*i+kjb}R2vSN--5?|Fdsn;<~EpVK}
zO$amGezL20SJ@)^p}Ozk%h>9}c!jm5oc+jz359Rz$!Uj~n$K65Esw3<=6F^Mo1rr|
zG<I{h8pb+&jD&z*nkwxx389&WLh;7clTO3;gz@+Z%I1G_#dX8eyyZq8xA`w0k1N(S
zXn1p|XyihdzoH8VEA1?0Vk;*^dT_hh(n$0*Dw(@<7cb8Zy3SOD8`0P2PUp|>AA;oL
zq#&m|Ql1AqZ3ygF^PA3X!GXeH);n97&MV<?;)*2FE7gJUd5>TkLf)h5)Q!%8*->&&
zp1`^+1~8qcKia-ID>oLP4D-X2Aq<PdL?(EJt!mA%ga0(Qw-M4%Fdc{|l5$mpAVAL4
zk<R8rwUb(Ps>X_12au*{KyVV%5lv@bIFbE1?*L8uVvcpi{o}u`xCiM#ha_x3evb|s
zPgmxzzIZ^)iO7!8c_-&x8=^Tq@LGT1t-bdia`6J%%<NEAS;{Zq7UezbDDM`$pYKpZ
zLkP%R7_VOULhz`G{P@ghtjAQZFVmDU@EiOy*h8fCj+-C9GwiIcJ_B1n2sJIpRL&xw
zlAltfnb>Jc<Z_WiDI!EB+a1}PPeSOFBV>7a0?9-Wj4rJyV!|w;MZDKJL%57krU<}8
zL0F2AS1V!LN3OSDbB!^=Y;8i@^34Q-=I;RM-*DW!Frj=fYUwVv(tpZc$%?-CW(3o$
zB;G>X$+(GE=&;4yjU}_a@R|wfcLj*tGy~5<5s!2+lJ*~D9!tKtjz~hlL<N0Y_n9@O
z@+VL3%I~}$C~j-H_&J=s`{H~>)Yvq8JqLU`=4GJ``pq9AR)|`WS1oMteR5VSZ{#no
zJ3FKHgCb6@ePMx_LzEO(MqXn^K2_nx_%%TbSv$X+NU6|2D-cP{VNq3uWTuURU}A@T
z-pz?vX!>P{Pp3keR>D%KZiS+em*Q`QUn&Mj6@S3}YML>pngwUW<=~yYp(P7wK+7>|
z+bsIO^2ES>o1@quj~(!iIb`<EjNG-*_42+@yN8nTpq7J$B*aT8a>_~LLC%PzTblf6
z85A_!e_<&BPbR)qUFqOH0S*ObjOF+|#*TiN#t<MC-Fhaf5gAe&mpJSilh?VN4GlcZ
z+#C|Ql}Gvr4r=>Fo@TZi-e+NQRq{@@g2FRgOI+!485uwgE@@YJ^7a23kX0wU@zO)L
z&s}Ol+xgL|I5iW}HmN+|S%~~R+6OCe_c%{dc1O%sJ?5?~t3^;8;lCcoj`@#07xJ|P
z${VRTF&w1N5Gbpg?;O=_fqVAlI;v&bU1Svvm?*!F6_2=j_f-BC<u@H+H9?{?EyULO
zK;sLz;f)rJZrpG?YWqa<mcyvT#_??f-#8k8%KRMvvnMFYmJ7kTyNQWr46C!=lJ3Us
z`X~9d32>6z$sq4jbdjPE?VL$bS#pq-15^*b$OuEXU-4S*t8qm8L%riQILMDc&x-!i
zmlhJn^{Vw`w`Pk>z#|FtVP3tz-Zg2wmN-{^kLrFH|FI>5OyNt<=BgjjwWB!#D1#c(
z-_bZNjzLwRboHRzmK^K<lNN7cIQq0lNNpZjLdL6k=lvZ336Usd3p2Qz@)PYdmmTaU
zlEJF)_>lp+6v364pma`c^n(aAcSo|F4Jg>I8-kecy7RVPRgczXZDw=xn3S7z|Lgh!
zzcc7~UzPRm(-9N;a7&20)%R&BTf@R&H{RFEZq2K3z_<K<C3(wjZt!>YoIBRRiDA98
zdesCF{?^qCS_g+hll52SOvjZNsK^=wNEtXbyR75<r9Ik&5aflrlCfOhe%e2KEjmB@
ztIswBPRJb4cmLGOM<pP$7vCo~Fa`Z4M0)!iv=fpFW(B01q1Fs<HuXH$JPS!u{{y0v
zZT+9Kc&`6g>SbYJ|DOZxt~8}xHaRf<Sw?R;RAX@lC6KrdiAJ4kqD*+56lYYqxs~Ll
zW|TQM?3>6v{QU@S2h>th<j5<ts<@Fy?m75;9SRrh#EKmh8PX^XdecZ)MyTqFj^xQy
zQ>4Tz<ciD){C;yMmkTof+^yJ1%ot`Xi{TLMz4$&He-7W@k1v5MQsEJmM~f}I-hV_-
zY1A_q>^%Y=5FO_FJ4f;#5DEQ#e{QO8<`h-18SA(EUFZ}nk*dBF$aE~FE2N6S6{9#{
zv_PBp79x=+jl@VXsAb^fi$kr3gjSDMrMv1l+WL5du~=eVx~tl{s~ED1Xtq+~s&RF#
zS1}+RZ1pg;qW)1VA1jO7=JHO2-yAFSbq>FIyQx>zQXv7*0!pFaz~VGftIl+L{kLEd
zUJ_CBF>j<fShG4l0r<~AD}#AqyT41UQ0NWY4MASM#i-!{u>??sMzS73f=;E5sG5A_
z(vzPA(sETK!f3Fo$NZ)KnB>bqOK5cY`)zahwRr|2t{^2P$tzb*ZCTB+Bss0dSLSw!
zNDMrKRw5^Vq=r;B?0IIhXvra#p>)1kEs93EvHi3uI|MfDxUxz24o}@Tr>lygOwNP|
zIi@0?>h1RLU`3<MQS9|l?L9jxCdw3}Ll+BQhDme01?9Yc!WrK=;oLup{vJ50ZCUtu
zLRUAm%w+l9na#fP1wqOrt8d?!uRwT6A0F8F+6uf;X2+L@Lzx7B1$+F~`sn>SG2XUx
ziIam*>_PoLht}Yz@mUGTrxe|_doI&;TooCS68GfPZraA|HAS@P7Bu5&b4d{wL7Aky
zkF}&B@lF!wcZwxxH&bC7FXZ&B1}4H<3TARtHdM_VC^3d|k|i6&=eYCR+EvuEyRLm<
z-}?%Bn;I!~b6Y>;>O-(VY=vhXh4&9JHHMmGiY>T=;HNt1#Oe3Ue!;k2g;vFg73YU&
z1S={+%c81O4tp7}{ZU$ui?G+S*(>>bpi0R)y-#i%_j>Xr;ZjfLAtTb0A~&jG<exwI
zFR&fDw(0MxXH&o8j}BYMFfGVVleJ?ZpY2tjrdjq(V=s~&M3ztvl1LGSoBIJg`9Yb;
z13LMpDC)6xcAP=K@9|UnLM%1}GcCT$hyB=0_#$&Lv4Wf!pJ>`_8#{uA01@T(HmY#8
z+n!y@!RLH+lOxaY97T0D_}^xT1QVx_b^#`|<XP|@q?OrT&wH^V5C(xz+~N%^MS<%k
zKYRj(eC26kdq<6Tst7DW50OkfKITBw0W_c-qN~m+QB_}yc&-ZE?%~9KGtTDbZ{WdO
zq%^!H{Hz-Ja5SFPUAg2fEj+U~KXr}`yVubksaCJVVuW`RuHAbVtAt?Qh}1+72mN0>
zY6&?d2cj0eXvx`VlSmT72VY;9ftK2AHhAFW8yjk<2pTFRTA-(qCO?jK#J71f+CieP
zL57`<2CiwG57a~zjLU4`cww@RYpPZQSE`dc8SAAXBiDz{UiWV$KK<IFzw1efxt(m)
z0m+JjOH>i!5mX>$1fQ$52ec@Ze<|KQ|B27hQ6~a3f?MRF5N>qfLP1gCovZc5xL^%o
zyH!{x`6USwQ7GUnMOB`h)!%+v?vHvbQ654;g>Yz+*91^i`PvMcuIx>=Y*eVj4akFU
zhMnEuh5T-eO7dO~XNSRhECr75edGk4Gr$NsK$gLO77ICY#*wy1am3M5WJ#Ybn$!Qv
zOjGBS+tM(H@qwqDNuWs1IiKt%hyT{Pe&CuYY&zDBSH$~36;+W6D=XN$>+u<qYyso{
zuv2`UP_l&{PmWsGN>8Oqr%vgdi}QQm<1eCe>25ZQOMvoeV#$BtXs8SRwYb!>F<D$2
zb!QtZnBm%8PkAkg0V%-D2zr^ta4HvA!Ac?ifavA#FYJ*A)Rc9X*<APn3P2K=CiBgx
z+iVg}E*hyYA|lKbqpgeWE>0-GZ#$4`9VT;R_Ou{=U;n}gfzxdwqLVHTD>w{*$4yRe
zVE#I2L|!ys8e6{8agJv|zjn15zfHqhmXH%l38ouyGvGQT?t9vI#x9{VWs!6i{)*PI
zB*N~p6=$Sep6%-fuPOYNp#%;9^O`&W_QSaXf+9)^ywolU5CepLaVjv4Qb5Q6bF||3
z7_X`km4nN_>-BHwjBS!hPs24QCUON4%{#nJfn0CEfera3mDqvvr2K2zL+^rw2)Kk1
z=I55v%tl|X7Sg9A4oOwiIT1&3a!d`+i^BUl6dd0jA;lC5=;Ct_D|}$bGb!Gxh|Zgf
zCg$D@a0a3WE}nn}HbhudHlj3eA&8T5XakQWv?ph0soQHhqcg^JWm<RJ@wSM+`e);J
z&DO+u^qJSG6Wx-sfzz?sMK#nqcUOuAfWcpS>nPotiw7>a8NFaFsM~0WML<Ldu9HjC
zonbS<Z_^}vk5kb89G5B#V2K}~Cl(5Ud=O5m**LyjsL*Vr)ias2uW`&ZvdHeYZ`dc-
zg2dXkKm1ZN_-&_AG8o0!Ew9EFwEtA5y+z+u6+mQ46X|%?m)zur%|)u~)FHTXXS9OX
z#f8wkjaPIPIxOZ0VPLwT%qUXfuW=eUAgn=%URS(Ny6qT`$zn_651ot&W|2lJQ%fQ$
zpCUUH+}o`Xc;fD2n+;TmyE`GcR<a8OFib_BiKfg#tA&$e^1Sf&g(_~g(1SeIm%rJT
zdzP;^BwDkqyaS{q?_z*2M{z~f0f$6b2WX_UzpCJ8F?G*ookUyzyAzT_yJWM=pKu1^
z!7R-WP{Y#mvR}{5jn{?Y>3PM82CMzF5<KCcs}Z;KgjSFkD%$2`=1?K_v$gIh8*HFT
zl#+?Ki6^D{54K>b^B^XF$74g{(&=nL){nc5HMK#e>S?^RfkzTov3lTINnvTM^!4BN
z5O5+f+&-}TXT0F!<lodsLoiW4YD3PTrz4od3wdKh=>beuig1hvm9*bqYZV8Qoc>$@
zB<%9nhNUkVcpCV?z(M}-Rz+;p{pIAyzlqmlQ1l14W;}KNuNEFR)Bn$OGjlO9{jb<l
zspf?JCKpog2d!%~Tn9D=CjFWgi&P*qQ{+j?VS#-m0H>lfE{$t;O)*hiCS}tJA|pDR
zBzn}CzMf#FSY@PG(Ej!KCH*6N&A~a6$PA?A_jP=t97*bt+=-Q|fdrum@%q)thsQ(O
z=>b9q%AJFolXbUI4PrZQq1UH{hPuS<6E1mi&QX+943DAit^h&MN)2k7X~kPimSBwd
zhjqu6^ymT0JcSuSek@1#;*R;<GEZZ$`iw$W$76fK=vF1;EZd!CJfeabGlmD3oT+a6
zlP0G8SFgRIy+IIfRpU9N$-}BlX@(rRmR8|V)i|%YrR+l5R;L$>0g{J(S{xhel9HB>
z`*11Gay}QPnl#HS2+D^xn+p48W;~xnDp~M&ON%~ska118SI}|x_2abm%rAt*q+qAq
z<k4Bm4`Tz)C(D?bFJ>={wF4p6)8?kSzfOkjykQDm?c<@aEqn91j7ye`!O<U-S>(lz
zLXAwa)tc+9S7PEEkk?^3r8LAQvF?-_zer;7zmn1|eW|jBm(hyT8`XE~otVAbit3vL
zaKt5HL{<o@29FsCdO~8Nbd_WrRML}X;$X@>ZzvG-Y-YLSX;Jk?VjJUU|B@GCsR666
z`~uH8+Smuml!l0!s&8e6UO|p?{k-@mAYIjyoqA~pqG5QJKbi%mt|KH1WQ|j=Q79%V
z&U1FFX%n!tJiGnW#oxTK$5qU9POID;Z^j?%t6=uK{==H$MZveyn32UC!yr8MFTKdX
zUL;H#0LGo;@Go>&IqCsRG>wFeU%Qf3X77j$A^UWA#vs&$c2k*pN=3DlYzjThuMb4K
zK&XI!xUinlRT7ApsD;2OSX2PlZ$%+WIyI^`7e@3pDHn)ka$X5F*`y@sNOSNP5ev>z
zZ0U+w1#ZhZcMe6Cl~Tpr<QLtSm48M50o6<s8mHx~p6>ZGm%tvF;akWrX4U*qDU@_I
zT3ioL*aVnrDJJOIT(uoi5PjxE(>FRZ<wB3Ad!aFSM7@K>{nNGCQBThXs9K<<Tfx8Y
zd^CI}S=B$O{F)L}0@Y|%lPrk!4)J3Ydq_x-t2bUTOHf<A*xoy^BZ+s7O&tW+ju}zK
z(W&hjZ`K9Wi2aXd7op*G2|!P1_HMENHVBJGyU{nHlG$XEpNu94V9P{xc`yAc6qkX{
z4p-xS-D|aT^`J$3+mj#wC=J7M79PSvbaps6Uen@cO<cKE&hmVb6^5;r^mNTo_7!0F
zn2=6Vn!7M&d!OK&H)Y5^o2jN8d2o8vY*X^Ue3JIu)yY_DB1na){ar0v+3HS;D{Cpw
zD9B~_74pR`((h*?u**=T#QY`87>lJB@W#r_8E<X~6`ldGcgZKv31DL(rgaxd$}%5v
zsuN{C2PiA#b<;x^N(UoD*0;l?Jo{yFJFhcpPM!r%;QlJuJKW&Kf}7lMC(};Z2&7y@
z^E`mz_kmCx4QvXKY)6hzUwGisz*xY5GE9b6W&P|Z<#CHUDSvu$IX1A6ejzCXpidp)
zK5h8*xd904MB9aWBwv{iD6;+@LDDyn?!7w~H>aEh^nDMqC)WaYLQd|A*eksfAu4pz
z_iVlK-}KR~GWmUZV*bI;PAdAx*=c~eQ&mes>xB&to64HGAaM-cPgA6*N&Km$D<wj(
z)fABfZVZ*E`^xX8!kK{3Iy*;`?WD>QGRqA+x_H3yvnrwhPwlgnC#Em7oAm2fd8_!w
z69PD+p3dJgjI@^Mir)a}>ZhJ^UnPl!hZ86%A)Z45`6Y_bQ)Gv-2aMbgvntF<Ym$Wp
z2d#}1x7iv=oq8{&`kmhoVSJ_`F0rPwsnA(w>sIOzedYKd9z?*Y*nwL^*chsRbn0l1
z>40y4-2~j{i2W|Vo$2E7d-f6CeEQ89B$bUWIk~X5b4jZ_l490;0O5fwS9(jLfU2HT
z3MPSHlIc*F#rPtVsSxys3MFat6Mf^^IvOd9y4f;9j@0RhrcS#c^U#VcGiA{SRcA1|
z4kQZjc6^;J$8vLii`=}FSw_Q2OSOuixoUB-8blEvDq`_2V%OdI>ALF}$It3S05fPo
z9bxw?+rFJ|zVduiXVL#MdT?stLyW=BU3|}O%bq-~tig(v`&shpwzdXAe&GXPzs-ES
z)$jr(^A4IqA)CK&>$0G*oPMkIrlc#e1IAN7@ZsqRJ-LPZZ0BQHw&;=KF<J;%6aN5(
zXdb%*V-+n2juFlKWy2hj87GvtyY)$JZuL0y;PhNg^^+N%h?Tn)3iJ#2CS&w#$^3>!
zz_sA6<tXn6{ke*z#$NKOrhEisdV~XIEymG^dD=HD9J)|f2Sqq;9Sf#&ANsXZEV%D%
z2c*Ktpf;*hk-%shPjfSYh~M>zT&O<uF87`>Mu>3iKVR1W(A;vdvHZV>^?&f$rK-xh
z0Jz}#PBc!X=9cLQy$QM%{?5#nT(R}F`V^4=J&qCc5~m?}S;Iu50ysHupb7(x92`J4
zn5H)zPHh=nbjRnbWU97Iis#j|M@=Ww%}~oWn{JT_M(w`s?>q?-3c9H4b>7J=_WU9G
zbJ+U+vGnpZ96$QYGzLl-bBfO5@cV8ly?!TeEuNN*&#yZGYVcZ9E+9F3p$djrYp9&Z
zQ`$?L`~rdH)v2Ki+KKC|9acYyUcc=NsTI<J=RwD734oa4$m-ql)=Myi8>?|}t9p&+
z$fL;f4j`C20Lj6SZ7>ufd@9Vsn(S9ejv5=Q$i2$tbhE_m>Utbw(H`P{T78Nw;~Y@s
z+sjecqM$HMDos$8!4?!(x_cMCcbUL8F<k)|#6)dz$#EI30B3IVe2#NQ&WW3`MU+_O
zsSN(2v+d0XlATb{;jh48d;;K|agV!31tjSMBhkb36_PMl<qz|&h)Z{FEc5!IURpi1
z6hV=h?FXA>Noz}F@o68(Q>mJRdnsR4;7{iA6gDjEO-J#b{tXpy1+`5ylMu<<1vbmw
z_ESwTSCWh_y2r9mYCGdK^vz?ruj%D<_lSGYZ>28RQa0>-#o81XzFfC^3}o5AD*r7e
z9am<zFb{YRUDWTjOi2-Ej73{+mF(S+;3Zh=gd0LD2hYd#kX=+&RXs$J?H#)KaaQY_
zTR90Ci!wi~fZ*N3U{IojiM^+p-7@TA;OZ&8wQ)7mTF`}|d-I1wqalm4V@5?lJ9o5{
z=U>7^TY><JpC>Xz-NFe>z+_s<3H?$=va`7wa;DxO*WB`mSjZ0%##s_yvj5BTwNO~d
zu3io7&$JkPH+5@PDrNeAl>GybXiXahh_-E=wr$(CZQHhO+qSLKwr%UQZQP#R$wlVP
z<eOixQ#*O8R@IV%{J2#0xA`Y-Zpkt?*`;(s{!q^Fm;fw2k1jkMJGlm^P}D_C0w+Du
ztL=f;R=`ipNjFSCI)U%}AS9{{-;c2Q5isS`<4>i{A!#t`OX?=$Y<|UK<FaO$`mip)
zWy&h+<y4|x3NKM5w<MG8nVUz9c(R$L|Iw74dhTR0(>=PL(JsDq;frot*wvaLGyWj#
ztO=?9i@~ykkw_mQs}fa0g*vo&EiiNGuDi&xd>ofiaAm{JEZ;%17Z2>iTYc{b@K4xF
z<A2O1%P+pA$G0=IgyQCgqLVhUHFGw{XJGn21&aT}Y^qH2Ko9(8P44sjJ&3)E-#j9H
zA*#|Rk%e<W3JYxQdjCl<M9rRY#N7da24QcnzpCvpphhImhhS*NC`2@!h*wug66oqL
z;gX3{yNM7|v%0FgdR4$!X0Dw8XO`X5khc~(J#F4veV}8D<`C#7B&H(lAAriGHu@|j
zKhIlddjZ4CBy0cAg)#rf!v0S_9{+z>7?ZrX8oj*zDPx}iFn|#`=>C6x<Uewt7&zGe
zXWGByx4Y@^x4Y@)H^4i)5d?XUOU%wRXPiXq`RHgGP)HETQe03ZkVwwA2kQ(B4|BcN
zQFf>c>+7ADekaHE?s`=2Zu%wcRBqR=;x+b<^ey!^51yMuuhe9uiAsZ)?Qd}tp#l7l
z&-ZHgZ*fz=XhpcMVp>15QjMN-)m4|)!&e&fx=Qtg$hGVN^$E#kgxgkL<}o$h{t}$*
z*8--6i=QWZICC7W=nGobNg=K~b6@d4G0gM3%?wf4rCJ%cjA~5t2y9$;42sLgkxYRs
z3+vr&el(W%*S@GTEo^A^u3XJT6Ifq|hmvJ*H|Vl9^@^NQY6>LfuGAGsio`RKWk<{k
zquC{PSsx;<Wv%rCf9z%+w~2Gi4||@@O<b@EcFf0r)(*7{tGP2wY^S_Dyu5~0V~Tag
zQ<aQ*78wv|W~i1G?;FhKThuX{Po*0?tvNZAc4TM%IrR?Nns$`tu{)nFuC0o&UhKr&
zak)L18b~@|+v5SLy4~wjQdOMU%S5tkP0a-*sV(?zD@D&c!30j-X4bk5gwuDJQ8im|
zh&=%es4XTj3!{i8UVqZ&6$<ElnEM-<z(k!~c7f<<3#F*8=3#W$tL*9-GK+d>3qXiE
zyZWB8`6!rGD@Q<qtBNXhrO;?m{!BK4I7XDUpPNznyrP*?zOq^_mFcovxKk6Zx1L1G
zoMKul#2>V%d5n5sbMOs|JjywN>cPV!ka}$y&k88M2EKU&M=0>`#1w&n8`^cwkve&;
z7LRFAkh_dmI28MafL!_%3)O)^uL%ee#<2smG8O&Bbk^OugLX)n%7Fihf1A@*44FpM
zt#KZcOeI+#N!%k~o`m4I8oocP>>1Xd_;o)6C8GX4SJT*}Z%1z?3~KGa5UV!D-s(zj
zYT+eBrIwqlQQT|gQn7EMkt!f--NZ-iR&J~cj~#i-lP31AVCp#^1Qqh%QE<3wWHuAD
zM=8KstF`iw`%7$sVSV57eX}X)hrlq|v4-P54w4_N_aGB~iH6X~o~wf-h7ia*%@$?I
zE>b^KGVrC8og@B*y%8@2Im_?eJB@QI3vb|-+%j<8#vbM8*yKh1thcL{GplpxSs!|6
zIpty=pV)l8AbV(v&9%c~!59_1q>!}*kvRe>WmpMXb19d^U4d|$zk|D}w|j*lB51~4
z)%p}^V>5$et!i9z^kS2X*3_76k1ihX?z1f`GuEy87shE{*vFUN6h>lou~E!N@k`9*
zd&%a^n{t_;YV>8Ekn(fcyw3=D;!q;XP?m&q%N6bG?ZR)YFDD=xQy1o{v%{{8ld1cM
zq=wh;MTgvDTcoXauXO&qmbHppt1eHffIH>`YglvpukYjbPnHznakY5(a5TR{w*M5W
zP1s`@WPq!vX?C)@1QQ79>$f!l@X=s^7@Wv6dwYi?*iwz1bl8<bMQFO1514u1e3{=A
zSIcta@=uq*YJO8WIzk|*fgSa-IMt0k1R%zcNW^9?cN<VXi{<5yXpXEV45hi%Z;n9p
z&U7%a@Hv#5+zUpHus%5uSexWQrlSgt{>mn7uX5!U1SA2jn)Rw4;LR;M2e59t=tv~G
zxra(4V|`c@c7CW_&Es8?7@uWSOcVC(+Wj?$$BHWGO6u(OOi@xlV?6K(%o*^#gRJf>
zFwC)3&M?CC=`E~}BM1b&v+!T}e9I1rYg}}b8HgMCx@K*4KZ(m?m?Z|1zkT7CyA1}h
z1`(M5{Aar7kqdv_H<7>vD|sk6XF+hiMgPoKRR}FMjNAsrO>1l+5a<*~1~-pG9r<Yz
zhHT{-yO5K)C)G6)KzXxp|5Y_YtwF?5)k&is*6geJ&#TXn47zxqbTlA34;Ns#7$m0e
zTg2nY$`I{9p3IfmQiTFFLtr@I=0cz6E?5jSCbADfgYcMVK;$bWxdbzneXx%4_8_oX
z(yF?}FQqryLP_DwZ~_os1Kp|u!H*}gxFG_SqnbA$`o6k3{sd5%3XW)GY?ptACV#q-
zn1D_?`cg@JMi7sN5RkET;b(n|BXk2UoCoE<#(|PB6w5Yl9|2@$(`Rk!!fJY;0v__%
zk!D9^pA&{!W8enN^?@hmd_s4@*s+;YAI-Z39DU3U(@)aVIrxo1k@HvV#h>LWcMwt6
zE-!^KiXY5Eb8=;XtpfT1p*KkSlt<aa;uoo;)+#wn@82mLNnrtr^8QuQ#Jn-4lY?kj
zPgU+iPO;Rg(o+GTpiBmd>_dr2;4kjdLnS_vvH3Hwc)grC877Hzdg}-wn~tiXD;^-i
z$AdgUp|-(^nkx}{23!!f$)Euof8xi$@74)G57v45i9uXF@jyvEkw`)jaJlFZ2gh!w
zmq;)8%3j3Z1UL?Q-)b^ukZ0jt-M=EAfX(<E9A4CD^*FWKT<$zI*iDQ6&VY%Jad;cQ
z(J10ETsc)lR4MG8d>bQ%nSF|1`-0gXwk%EU9HiY3yP&n@z=bKuiUNc44Ad63d1Wz<
zQ3NPiRn?C7G=r=f-qE;H7r>Kv-QK3qV!Z@=(I2npVbHO74{`W7O7d#Jn?+{@OcM@?
zIKNyEq;r)=Zfc2z&0PVXpYD)4(#8jxs>B9rASCBOpbiFe1z<P36j1mxJ61JaIeIGh
z7jVHuZ~ec(0K<P!tPIQy?EmWnVu;#!;%^60&#c-lTIFRx1A=z!^KZm7v8*=oTYRI_
z9*{(Jp*SjJY_+!!rnx^MN37GpFF$?eds840<ij77XSZWxx5F(=I3|wvz`k}3gb2<U
z2^nKtUlT}2aO(W<@|?y7A8pFSBugP743(u`)dg?IXLFOs+NQk_58c?2!T_6h&Bya~
z_x-Sa*fqL~+vb<yV_VMU;X+N4u^tUZjgUBm0Zl9g8P%xo75vo(4(bJstJ*Om6BkmW
zxa^<Q`Rk(*#jMd_KN9;}%he`zI9^`PGQ)hX!LrwxZ}B?!#9q;RJgQWro27uX!lmu>
zPxLeU1uHeww#+vv)M#_%YDe3}fh?Rdg*aqor;K$*7@>gAWhb+zh}@e}jx3w!$Yz<`
zsUw_&&!F{k!)zJeW%VFT@L!XnZ&6<Mo0_CZn$=;@m;L#<>S38HH(EX5%pfAW*UNdz
zQ!1k5`tmk8van<9wd%IIzYr;FbXXiIM50F+!6AuzcN?74O}#b;44y2js!l~HAgp!t
z{vdlhD(cW5X;-KsOB^;XU0F-_t9LB413=Q95leI|o;_64>#}J3%)*aV;(iR9vW6zS
z2VOgtK8R5?^f|N{`_3?-ggbZzW~SU|1d%xhaz}u%M3Ty)^rUDKCuJ$P1kT)Z(I@tS
zrM|D!#h?Y!njo#jwwX>0PG-7fy*L{ln4(6Gs40b2{B*f?;!y*wv}++1_+$39Y!7+k
zrgBM1L&~|2nLd0;Zhj?hyFLl<I$~~TvrkJRFl2wI7=CkmATWt?Z1&2A>Zs0&Cr~~@
zg~VG#q-FdOCHP_`J*u3SKYwpfsL@f={)8h?>{l!%b<Nl+P^8KJam4(NRH%I7LwAJ=
zPE1xp6PO&69KbJ#PLP(gWuXQ@#TT?M%MPqY1JSd^C}IBWS>(9=2kO=r*LcIth3y8G
zyc?n^sQC(`29rrm(%|gwXXPqx2<!rFdv+r7Vl6@)BRxP$-B;$>$ZCUh9bk})$zBBC
zA&ND{;*3p}&qh{|!@epL)N^3jI6<$ygKSKiKo}DdPS`!9vIBE53wQWafS^=;IYx-w
zcR7q&>Yr>|<=GZr@h^eC2V*dO6oeb_$|<I|P0Z;4>bLJS{B%grf`5gQah^Q4CLRe_
z4tyAX8zcLIO#-FYRXXE%RNjL3Y=W`vaVl#B@NdqXu{wPLB-p8t4z6|^7S`z4mPq{)
zJMT#i*{1D=gCkdF2M#2X!#`WgBy5wDRG!F($8zHnDAmP+E;PMk=}(dxeuc7-kzRCV
ztxiBo=}=A<1)x*c^rRxb2G?$jbAiLHd{~qQPe4`LSEnBJ#6)6^?b`FDC^8WUQFQrf
z%yxtt4@eK0bh^lACbR%Od5Prfn7%AOx%<lxCcwPVZ<P#?%BC}Qh#^U1TMP*$K-e=J
zh-8;zs#8Y`7Ajf8vqfV6vYg`UezqwFhGYWp`5|1vs(}DxA4sIm$$*a}iK`W!-ceUG
zJdZQ)?hp6VQttzLCTS1~VR|&yK;E5XL<}?NvNJbA-ns^Ju=9R2slEBPd4g;t+~FwH
zOFy{%{5J27Awu=wJa9E_34GnK93fpA{UagR<)-3VG>~Nn4%g!m<ph)>-DT>xN7E%Z
zeAbyc&p@njfS~u&3J9iSD(>7gykzxy)1@+nzrjfKAtrN<Kq^D<^~p4JV@l*ZO(f~X
z0Q8U%Jyhw%bi)SEP^9SDU@WTJH1g$*#9<WG`VRBDEtRt@EAa!7G|!RHt)^;qhK|R&
zKrQZxwA(>vtRJ;?Py691v|6e%%#Touu1=V2@;1QG@J|-gdPD-(bi>uQYB!xY1rX(0
zSeZBcA*1qPqt`Oo%KEBmixibkDduSzk5^yMCwh+dseu2rw=@2Sk?(hJ|DOYYYt*EY
zHd$f*kBeT-z4X*A!MVfq5yQX7o@<MwAnQd|EeU60%eLz}I&t~NVs}@wg=ly}0R)f$
zcsz4V93o7g_M)hptm*BG7tknmQT8N>W)5M(n8a@U#<9X_Ue9n3LDf7qzF4ioFqR_7
zIKKt1W#>w*nvP<J^-G6a)f|;3wQ1Qwq)V?aUPJ|f$mFsaPNL)>3`_>Ucxj2uK>-$Q
zXS+$$mQ?)>tfYwWmMi&2cvTu%svet3G+Fl+%`)n!N!Q_fbx@t{byYjBs=yRn*VcAd
z<(T`<QoLte`?`%R&ns-O+1hUsNW={OANe}2s+{F%L_4kNAbZ}TJ2BM4hMv9ETxAHt
z&+udgujr9GJl=BIE46kK+jh1dwiYG!pEdI;sVcl>LJS6hvQvj+iUJO?zkHlY7KnM1
zE9W|T0`-izgI6A;Jx28D+aS*Hr6LTZ!x8j@`1cYHi4uxi&eBkoITK<b&@lqzWHZAv
z*>owK>gZ%U4w8}DFu78R3TPzK@zm4v9|VP+i=}rCvVx?&kDag);@W5XU<mmN<Jenj
zaBKh(EU(*@`G|!yhcjYb5c8=GyumnRE|AaQ!8)^fdzTm3;)#M95rT=t(62k$%<ZJ)
zF>Yn|);PqVCJn&JzuF!(gXOZJ8f~S6Azl($?3I7ad$P8eD114O*TIB-ZfHLU%;yWa
zrCM}UqE7q+s{lwH_|&j{brUCRX=2CtNvFUF0Eo-;%^SJraHrwmjhbZ4A)A!RswQR*
z#xWK4%E0FMN|rY=rgz{*u_|-B)lg6-rh^uo=xj1B8mVPt<l{*|J(pbtX9(T3=Y3P^
zl|(}S<ZEIY3;>4tE=8i{#U0?f5iJo5*-!b8q(z}9I*eE_YdY`+!xI?SVHb*xICL@&
z!9)TXQjv6z%r84l?7c<pW&J_exZowpXPxSkgGgdi^G}YXFECgW_b#(O+RXK;-Bcj&
za=fm|^2&>gcIeI_Ppq%UGX<FQGxDimbOfCH&_=;)cyk3QWlQh1uSV4S{6iM{F;cL)
z+gYRH*ZOyKc`Bixyy_8q%LE)yU@%Qq{}JK)b5G9lieMyQJWwDhBEdD2h@XR8i&`VD
zZ}MwEXT%dq0nkxUU^@frzS?wmGn6*b0fnN7IMzVn(+gIlv#QW$RKeo`^MXX_)I200
zbaSI1m{-1Qr?8f<PNW9J#@T3eft?Sv(ln6tg=$R>TFm)zhI^)N8Dr<#=Sw?dO0~q_
z5zS#aTV^x+)M)Oxj}nM5TCvqf@(%VrP*4Um+qw3Vbvt&adHgi4Sx?r~GSe6JY%K`I
zYQ@9;LD>PaaEJB}tN`|Td~)Sa^3zLRX~vn$*KY#pmG6ym8+fTctb%_z(a$68iH4hH
z&*rXd#j=9Ff8e9^Al|onll^2*>-t+zGzILWxO}jVV$VWxA%Y>){a6^^^`|@&LKCF(
z#ON(z^Y}#;Z#l}FXOyA4u)yiw;tBH1Ufstw_zDZGza1-f`IL0+%mCo3bNQ#;0qO>R
z)f|<3&we@$!q!F^gOy)MRUaZh4U)5^u8Sm7V<6uplx?MrDA5Ot(2w=!ypr-Rc=z4r
zxRyoWunNomXf_~$4}PQ|Mf#qDHMq>Q{Yg)ht+mtJ4Wf5*%cmo2sMr5G7j?e2*Xvcf
zh$noTNu7Vt?_QF{SPY=-l)kp`qigrh4%1&&@MetrEHU}0F?=sBQ@l|E94e;L{^pE>
zl>|GXSRprY=xg4|<ELq!cL;4ap5h@5sUk|xhc>I~Ri7t$>$XAwb%;Tnw87Q!j}^-5
zDXv}zeHJTNGR)o9A?qJIvJrdQxy+K=QU1ckT9S-OHG$rhk8XA*Sz_?^i9qAwTi!$K
zjQk&f(V}b1|Jd4@|NmgSvWLA1KAoJQrIND^6rC(S13mr!(YNF1<c!b3%JTnxA!T9x
zpLGG#8k@FQYzRK9YPStOXS88QFcN(b@lxAh<^lM!0j;45?9ZhT=i(LA!2fn%U}#1)
zEof|1u{hux#kTgfXSCU+6!5L83-pWb5PvAp$qIwf=>#C{0!c^`5$n2VGS3ogg^@#C
zhBK0(5T#{(5AO1)U%Lk@QGoyp)D>|diuO_lGLchi4!F%q>if-F`VIr)C7<+T>j$~g
z(br|`L4*ufYI0+mdac7p65#cv@CzxYM>rHq)Lmm$Eb2$z46W@)LO>c2PSz@{aIw@O
zpnzC~tKvZzXf?pF0{TdA8W>6lr?AqN*AWu$QkP*uH-}7M^A#BqCuvn8!Fk#FLCU-l
zR3_?^p@*Vj6HzondzF`fQS*@1_ke%^S#H_afY2TZ{3(ISfo22=baNMl_k;N=kuU0%
zZG_DG^Pz|{5PHBc;Ej~c5GLZqVdGZnT`CDU2aD^2Fi;o=gi!ko^-+M!<&R9|E9xWm
z<0n>()9qGom7ydP1X7X%#Ih4PVevV{wi5yor6}R=5GN5i6a6PF8H6Dfg(#pYE1Kj%
z+?f0PH68>0m<3BLo0g#a10wBfyAp<xoXx=K*rQ>_TPG$EXI>`)?}Y{=I_=++YRU}x
zUnSs&Gb<`ItpHD*MOwTf_hgQsVKGPNuL+y>fe;Jyqswd3^W&ln1q35dlV}|YY71^4
zA~U0WMHkDlRzQyvFMt@LSQFAHOs@rZW@6*Woc}pN5y_UHtdsfr!1E(4^7mz3%&Mwt
z8c!mdFg`zRs#^a%I(!o``lPw2dKfOukc;X1dOs{1mLrRL5HTvJmOejRcQ%NPmG-G7
z>DqFUuGiz6wEIQ9ZF=&AW397s@>H5OlV$*tP9|@#v(4UV*?<@Vd3k_#r6qi$5VV#O
z7Ve6FGiPaKGebB+-soawzDnPk>D^>k?7lzN%Jux%kYBlUm=}jzo+}H@6#v@Rnz21Z
zoxHcUq_pT|O}?DYn&wSaunqz=u}@&2<jgi`VL)TLVM48-^pdXFc&jDY3AKaSL3@_@
z5X!TyRUujt@qw*$0AI9&uf!#;25r0fF8YWKZoAo5ec(tJZnqOMx8y{#Q&xr_tc`~!
zwM6EQ3V-u1s}|8?zpMpt-2`-T%W`e29td+nm&_E}zDT>8)KB2_7|;*w*brV`*$4AH
zja{N8%@{T`rZd7V*r+i%U)@>;^w3aWJy#MD&HHfq-8N(6690oxJi{Q)V+|Ve2^`|t
zoW(41>UBh~>h)B^fZXE8**n?a1KR}NLQ`UE2e0sVK)6I73fu+mp9AloW8zswcK}o7
zyuc&mQPUi=x91|QLwEsT<C#h3f<oV72xEW+5Biugxa@VU;Z5hSX{)Dmmucs#4i(Db
z%ul5Y>h$HznZdt$e0~?Y(ooVuCw*#K|7^_oWAR-5-QlajH~L4K%gnM5RN8Gt%`<7A
zTLq@uJDHNR=lU*1{g|u6qsxElJ?&-D@4QS*OhoAG(eTcfwMOoISNHGItec9;->!_*
z<qN!%=`Ju*J!Y4OeH$l_b?t8#m+w0$h3fAw`!^nG8xq0GB$@^@JxwEL6sok>xzUpm
z1m+QW3LK%5M;uBL>*R_@3`!H5a^mhSb>2A_iCN37@`Hxm#gfAi33vFT68QTC(cBKg
zG&tw#=9}~l^|r(MD;(>V`B;R|H3lU~T-?w#v(o5Rvr`QtXWW|QNn6p#TEm3F<Ru}c
z^XGcr&s<jSR2&l27xId=vSP!o;&8{~!Gdd|Hl2fK+KTx)FQw6f#AkM8#XKPrh1pC^
zv*y}+Y*(hxW)JYyX>nzaLc0%cE7U9W;WQhuxkX%oS)ah%4=|oOoo>mHKh@jOu}-Lf
zOZ{d&HEuCR(!L#sl=v?_P-6~6x*fAB{BT$;WWOqAm>D%r9Gq8fEGXK|Si><3c>50&
zA>GNl;3V<;wvQ~p&H8}EZ(S}FlCLkfN9ZJ$QP<97PVGb<?ejarO;gUxmeC*{D|^Mc
zliNPyFd}iD&@fiy4cHTFvG0b_Q-xdSr2w;UMs79RF!#4jE{I;Dqu6ja;m*GT?|1`@
zovHtE60tD-2N%c2@xSxr(<w4e8}taFH=j^=)6(+9>-VFN!2!k2=%mP7N((^^pyr0w
z2XV!pAHT=sTCnWK<HCNvPEGZ;&y%>mk(K_TD?)J41=c2`_Yv0;!$6Dy2UrtMu#IVO
z#GSsPWEkIo-6|W%M~=ymW5O|eeS|j48)q-H*~=iJ$yzZhW^8)U^sqOhR;?)&%Woe@
z3PaBJbn4=|BV!oCB~M>xrfYZaU4#I{Jw}mCqe7OAu+EWJZ|#(5eECPu9X%nsk<*h8
zM^4-yPCH7&+!Sq>`3qDcJS2lx#C0~(F_T)vYl9aRRFmnKrD`c%%Bl?q6K7<E2ZUj{
zdDSphkAi4MIX&T46<I=7p@R%8U1s1}O2iV%fxL@yBoEa<Q?!3}GBOIRCRjOyHnU=^
z&ID1?!WyWCI%?bsJu4am&?c<I&qD-g_uTM9j|d^HC82T2jZH-Uc!pEtts~>TX$y?<
ztm^-mnnfKj0~mcGN*=;7cx$xo)o#Z%NFb0mT8H9+m)D{RXBBip6iqi-4nOUwhg{Vt
z*mjwVR#PBMMJkHwGL!+Qo)l=bh3hnL)Nk4m=(^WVEmdhWCEt!6`T|koE(DcF-5NLZ
zgKO5+g)eyBPx}n<mjzQza>Py2Bw*b@;}V}A12u~yOu@RD<MJlhWCro$6(=Hw?Fzsg
zEXr(Q;*nR3HrzBz`xT`>Y?UIFMbdx^WI1A1Q-Q5Mj``_PF9<1PClu5GqJ@-&EhdzX
zhu^;C)o3nxd{BW?PYJ-0VFeJ%Wq+gaZq(Zl@3#Gnd|+NYN4*^NDB#g43!uL?o|xij
zOW-zrI)#I%B0}$UnrNo<CDRSE%~F(+rX+N+?*%E}0K>)t)jqOCP8p-xJbGnZa!~22
z_}qDV*yi@G1XEZ2xFfa$Mb|c;>9OE^HkB#x3*9Zy%nB7%lkmnqfN-CtTvg9%8nxxf
z4fCu3L+vDo@wJaBHvq^sk}1KN-sIZ%Q*yUXA?0Qbf_3*6^ws`0=xF5ITILqNVQ2hI
zQ+-k>@~IvIb)FYhr5>W|e6Hq&1nCj+5)b^>q%r@;q%ku6KPg_`bo}k_@tIXSvq-DO
zR2T|j`)fVydN~wfrJcakq;=@yPogQ9LC8#u8h8KX<qu{dnZUMTU1G>m3-aNd$E(k*
z8C^KPoHh27yWNwaePTVd5Z|O8dyOkbh%rtLf5?rqJuT)3k{2~vNL#5fpZ)o9-Kn?!
zB}-d^er%9?J5(DxfW0a5I=Ut6fJBPns%cQxRH(8%Wqtmfmc$UJVP9k>rINfqu1=45
z8%{E@&fR{ht!>^?E?ra=k~1Cj@1(z4t4Hydq+(mMRZym3p<^y>9WGR@VO0^z>1+(c
zZJ>`d^s7V{OD3W)Ay*Ut_o`Fvw3aDR2{B=Ww}o*fe9&GlfpoyNaLd$Y><6#o+Db2x
z7*p@<YN`CG_`5KH!w384V0aK4+7+#lI59ox$l>fO*3SSt?tHksjZd!!;|nlMq`%Q-
zmCOi@xhyH7QS|Fi?j$`@rYA!K&v`)xz17^z4B^eCr|2zxq!XtI?u~zJ?yWw`+RHxh
zng(tuV*$@zSM$fsev3RGdlpAI-xOex?-tSuBlJdzlON{i22tz|Jl(6}=(YqkaL>I{
z%guHW(}O*(`!YG~U8nQZ4_tG+LF$S6h2sH>vIlmlpmP&IK}g2KA%XL)#~Hf7ss4rA
zI|VyPV0gnNJ~mP;Fm{f`ygPK=G0<B`v0fY2=~uSa%_<o+?36dt144cq3$xeizq2#P
zvFpbo?DN^odjx1#_)kly9(@WDyol&94Pu|?;uQmJpw^$&r!QH)(|44zN%-5mc!g2F
z-jYfVn3*6f6+Vq9P&G{4Ss<l>4ghjA+F$A^WYBW-$~D5RizR1s+?)9+;PsCKTH{jF
zgO|?I1dna&5)+{Y_WV>-qX%jiHtEC&gT?%;p}s%;MA4DCxvoIz=IN<_f<rn<uJvq;
zmqR^5eLRRjR}1cR;vT1S&J=KPF7%TM7ZWURXGn<cU(_cF%G*na2^~=hJI&;H$rxKr
z<#`$!rnu%s6AON`WkYW(`R-25VK)UN!sssacWvgZMw`oX?pML#`OH}Q8Uv0-;)kv3
zmCm6K?a9~O-83ymTIqRZKf4NtL$2N0-`_=!2#|bGTj#C1d6F{dfaZv}xL^CImUUP0
z_zCMLy*%O8Q<HgG_#GBiyf1M@{px2EzUGIGqTVTdXDG1w6BF3dgk_e;>2+{9p1;ff
zF~8pYyuLoM?oUUq&E;%!b!mAUJ+OYPKVw~gV|lBbslnmpO@3EMJ;gy(XM-v1R%K=~
zFKtQ}z$lt%bR+iO6Gyl&G~BYvsykK;;cOs)lU14rhCNolP)_W$5;E*y5d{1jqcG?M
z(K)6x;-SQisyZ3aGXtY;evQei)GjWc?}U+_VpH7zQRqR5srF2iMKG0Bds4kB?BZ9j
z)zulW%r+zjty9rhK&y7xuu3NFI065hvcVif(szS1%@h%M5H$ebE**&wm~T{Jb?dy`
zkT0V)8Va?gj=J=!T;6$q&fFNG;r!@Pt}XyE%5cFEk0+wYZL;!l{rCrq`{R!y=c=g&
z2NUE@iluDM%`%(H8zcxo=#NX{;cJVgf>Y#!X5j&=KVk?E3_T8v6KMZ&<J!y)H^1#P
ziw7nSn@ATn9w`v+sX?OqxY>!W-)84wv|iN$SL_vgZ4@f{;z~X?>ln!2AxXZk81;RK
zyFP_z!v?p+t`fg90J)PpW1d{EfF$BInbLyr9@fvi{1Fy>JVKvy;9Q5$HgWsTqWnJ&
z=#X(=T_JC*n`_EVk#C+tm{fyOEH8wny+3WpXvct+?65nPrwiO{a+`N5724hE|9n(-
zqJmd9<T@2-UVGP~>7ERON@DUv%k=|M5r?UcW;x)MuE&3-x{ArCU)jngPb-}(8AX#t
zeydH7e2r8^w7BvK2>39No|Zbf#vAihRf5(up4>QK<f)IRmRYu-uMhxu;vZo$qdZ)D
z<ZB4mb_}1<F5fMwHSRK=F!6FV(4GN$&^jrlEzmZB<4*~V3-u@rGyU*}6Z-I0MEEl|
z0T2|Nl|kycgSzjOy(%W{s21bL>LqeX`)%eGX+NID1pOl|QM$<lYd*;dGBuH`HcM1j
zsw8YNFrO@2T-$D!5Cze`=@+Y|&AP~f-5`{Y^ZcY{_5P6CiU|3LcYkRcc@Hr{3{|fh
z3OVNx;e_sb@dQ!oOPspsNta%1IAwFoLNuau=N)ZJB~ytN3POQQWn;fv`y6rAT{*SG
z%xuwZ+c4L}=kFzYmiexrK)7tWR3Z+V`Mw$|v6T94C6y}VjZUgAZ`fJ)H^pM17)83I
z)+4kGk9N9bh;S5&-G+nbvB7sXy5@d#4FXz&Z#JeArM|H8Ds2`PWZ|?vQlLigmcaQl
zttARsj<$2(SpSQ2ZfjP{=DscSWm_c%+TJ50hf>Pf5JkzRTlReNNr>zJ9W?_|9<`qs
zLooT*vy8dfw^JuZ+a7((SZMfwf$656+xT|yzcqc?PHrUEM8b{5U8J{ro#m8Y;_q00
zjCD4!r80^JAHz6EtA)3`cs+Kmx4m{+-lP2a9W<S9PC>bMBXV^rjj`XuUEU#(kOpq4
zf)5HN@n3+>QR_8+FsMgTz1^(RvKbOhp5!my1|me0!|76Nq1C=cZV~I~*{T+MG9Hbk
zKQ-GQuoD;C(L#9x1h6*`^;^wm^z2&?2&7*0TTd|?0Tk@EQpBR4DgJyU589<UJ{OTr
z_NS~orqp>Buh!!!TRay<`IRNizXJ@&>jo5XELX`$s5RW^aY+g7_|SeXJl+rWx6}4W
z=Ail6Zmn*vlAzR{GLCcN=Cr!&GLd^NX1JE|<U9L2FHoI4mYk0ZEGIR5EhQI^QtT)T
za$hAe%r4xFueoSW9F=Q&C0mTrDwkBCn6AWF5o&S)kve3Kz-lOpKYUjjY{?ABaWldD
zCSnfdehu;9?+Z674BQ>M%wbw)Ow7_9A&~X6nZ&_wK;R#W@m0jYLs<HOSfD)Ad@Svk
zX%mg4>IRmFd753S+DzQNHMWLR{2gBJ5jXi#zRE5<^xsAp&?D$pY~1B`UiN#gxCt?q
z`?84<3)<Ys$JM{Tc@B~OI=U{OZN2lXCYUbP@4|}?{byO_b|GgCF1HN|nn^qXw_S;!
zNMc5Zq<<>OWlF(FoKV!jA7REJGjQit+SBFrfJYw3mhfLN?YDF2{~}IgWc}YQIWZdl
zZ<v->H%Fv{Dl`;>-!0deJ#@E`k$fdBb7hRD4~dAORGeCTdADl^469fy+Ne_z34mbD
z{NebXk0Ym=<bMCw_BwX+_GPyy7bwR(4=F|Z!bg@o1cN(_%(*`<ypSwkK=yJmt~YP?
zVR(T&Tm(;6Cf#VBo6%VxlQnl1*}ltOsL6!xh$hB|ez<GdWd0CBVZS;J86*9r&a;m~
zd_>agrj$|xTZP|Bc-zYUcJsCTh<tn*9!O0}IB!IAqq%q@<7bM!P2;FVOFhB)UZaVC
zI@wSTe~sfFu4OFU?Ega|(a-1`B#*r$tth)S<EOrggRq#x5^^HU$l6$XyMp4WFO%+W
zH9*J3QKBGS(PUoTBl0)d#z%0mT+_C{Q6&Ov6Uxty^P+HwoBO)@Le*`B=a@ORH(9K@
zdr@0*K#5iLXCZkqtkP2$x!;zRF-e@lz#-DBD(_>WL@Xr;0PqWrtBNm(QdCi7v<!r6
z!(?ZJ1o!?n(8{gG2Z}7a=&o{<^fE6+>aTI$UCwN)dKM%1r*)+4q*Wsx?1V*pF>%Pv
z6Q;IG$&(?$c^^0XD-+xDKNsA>t<<Xeez90_NKe42h5)0VkXc$81?khaheHFFF$Dr-
z!XQ}M{-s+QzB_1lAPln&R;S$&4@{l-`tueis=+!pr%z9L$x?1Z22|fM@1t!qG&Y|a
z*}r)^b{9xxZl*mH);=pCK+9ZM5|9o=AO;bFux3(C77t`v?yJP}IX0W!+8q&*y&0^7
z$5!;pSuE|*wBmi#7v1JnjBF&lXGCNkSj?U^>}tVQys8(KfwOnvfQ|IJ0^J->yC8@)
z%k148B$|%iZVAUF!)&!R6_d9d63$jw>0A%AjlhUJC{iTC_v+LsX6aALmUz^V7Y<<G
z0TyIWmJ4xq1oYqc4aqGu#`7oEoL?=pmWm#*s*5q}Yn>Uq3LMj}27)03;QF%iGDx`g
z^j53G<VXfr?&jApunibNH5tMTKRdLoajouSAB2R)%v7uf&JCbTv<}{WiKPlS3-Y3E
zCuuf~w%WDQkO2pI5eWU1M|IdIbubzbU9?k5y+bWoJ7q<5Cz^rRC@MU3*4IpnWC8a=
z(t+!(H7oK4XzEq8j?Klx)RdcBAo8JMe%a=5$8c$#j*lgmY;>3N{y62MIaI;eRP9k}
z?4zSqRkdJ34zoEDK!S>*NI3F_2qpP*SPGROOX>(Nf4#Ra98X<p4aw09A3PU9?6%Si
zTa0GK^V-?p*ErfW2Dw@vMdSj3^N$YZxBzfjWX2f6^QOTNzizO%3>_SdjkMn(E#^`x
z9cZ00650FuJajYPO9p-C2>D}Td-o}z*}0M{2l@{2<de~M{?-}D`4JvQ%d_%20q}Ur
zQ+g@S6DiNzy;-?q%dsd#$Hq#{JRb@{u<i>YkpR~eL@dDBmq!jG#&x}7(__ta{0Z7O
znFz7a9JM8O6hYMS^c*YN9i&^1aHCA=S~ystBk;1>J^-;)%Ly9Rn@*I49&;ot#;sbI
z%n4A=4>qA=Z{m$&2x+e)##kx_e#QoeF52t!kBT$^`wBOa-skfS9oBX~Y<LN55jsCT
zJw=`-ysxsLnuP*vRa-1&fe*=H1i;HVcN45<nxMxp(kEAT&B&|6Ytv{j$_2nWZ3S6q
zHV3YOp9&GQ@C9Hq1nYZTIf<$UA4b3k<6gjl8_3U$L2aL0-5dyxgclqo!IAyzPGMj%
zk6n@W6$XgjXe{?yAUMKYU&Nk&vQ>?D?%mPYukeNa_1XhssL1B=5$H7KG4|?qN1qBM
zsI-jnj&Z5Z!xqPpuF_|7Hjvyw(+P=oT*;CzP6<ft16!5*FkA%b6<^dpD5b<D?KoI9
z5eSZgJOj)LqXXmvhPPtA>hEa8JA(o`f}K8GqhtfSn-1`)%HnG>k6J7I91+&!wCFCF
z6JEY~I|_Nkyr;AlKmZ&R=hxP%y+Qxc1A_#Lu*+JNI?CW!UA+`f{EnAG3zq`t{@4nD
z^iRs_b{poooB6Gvy}%zL^MxgP#KE2~q}Z_ns<n7Di2|oxog8Qg4z-J+3W-wqxI+}i
z`VQRRd_8zQ#iJT`@3|B-9wQ%cW}g%2q;YHC@Dlx|l|9GomVQF0Rsojur$40}Z*qR;
zf@b1F&h!=y(CXAZoURB<>)D$H2P+x6p1)5Rek$V6go-_m+5F}FwnDoR4GQCyDNQ=t
z7MI3UFkY#9MICnMDf-|fxSZ0Undazzw<EVhU*p({{l7uyQ=Ha5f<N{IRFyJ#bw(~2
z@u-+%6CEB61u!>FWqWUiD$5>OHF2wult}zG8zTM{KT~<46ulJ>jj(^%UT&i~PB5W!
z%7Xw&BGw|jK*NUdUCHAM)twuHio5S&4uMzl1^7Zl9^}dUAHC(t8D^okn6$nt`EM45
zh?h?ByZvlfLj;r>7&JI}^gSw5h;q@%>Jl))K7z3^9;5Hr9j1{}unQBihwgE>9RbRj
zQ-~GL!-qle_3RknRzo_vHcK&q*J=3ig)R;8!~P%y-!UW?!kX53A<eP6xWQsiskV5q
zh5&0^e<37&T_RzYgONkwjV1xq&PaKw(5$<(4TzwI?E=Sy7_{Blq>&1xk%noX1KDN&
zFc;9KyZJr=i0{tOjAr`e^oAaG!C)X5?Y3)e5HbUPl#X`zR7@OKb=QQcjLE>ISQH;x
zzkv`VIT5(N`~VkFDi*qzf9hSe;%e7ddT#3t9uT<P(DAs@0T^5aEfX5xtGOgh=o$1?
zV1iJ|ag*xTZ_E{KDvRuZKz*hJ_ebyD*np;GhZ#nxX{fh*TmTAMD@2#O(07LO;%*)H
z69s6Jd7%FWp1=OJ0SEX{2D~N<c9mf<6VW?DUD17YfmMwf?4Ur?B8D?ax@qsfA=8gF
zkZako2Huxac5ATMS>7!f=rSyxkA;VE&IHbRG4}mApZpiBD5<<!_>l@I!9GSCKrFuj
zjB?S=on3`3Hg}1djyDtlCJvDaB!Pis17@JWcN`#7)Rg`Z1y{VWYwPf2wG1|P?Fm_m
zyzh=e8**ek@BjJs#s*G6KoYRn>3n;|in403*8Y`1DVWq}E$N@vLX*{N)nC*lgI2%4
zys1vNWTwE42Axp9lUk5IeZm$-;Y$ZJ0Xemh01>*;4xM&Jr;QXH&U}r-M*GiByS8Sh
zuf^&gn98-MF}=D>foK<CiJ&aJek>AkN`R<C3G?dESNPG_c7uQqF!Ffq@bfaJ8yf(A
zFi^8|ZpcL+QnCa<DaJP_J|s<f*4s38E7US70`0<`$OMh^pNhG2H^<o4cdrr<GRe6F
zQ=g(;sgPhU2+Ui)i+69IEWN95umV>b-;_|y*K<7MzT@~fPvW%_g-TZ(Uxf2J(JRhR
z%fEy0A7VIyESI-JDL0&-l^3`0-OHeaTMQ&p0wjeNx4z;5z7L}xyWj9!`xB4<g*@3=
z|F6iCfrI{k)^f(EOF3<|Bldo%)l>eFO^N7_0@=?d=5Q@+v=%A4H+4$^Hidx4xKr^S
ztNMDf;1*BVNT>-C1_0^8+Sc>iM#shm#`g9O`g-_$o_+>43lL^3kg)%t31>hG1YwXs
z+-Gc`4*YFLEM589lo2+<uH<wI`a!9T;6=4{EV(G!<rjApvq+S0jzI6<7pdKlT=sLb
z`>B|4^m1?tVEyUo$(~Nn1r%irL4*_UL#ihcrQo)sqCaT52H&<3$7F@a5_C6zYPeP7
z5-=4Mifm~1JVc1rU)52lVD#jxT<F=2`7kPL6kH9`YF`lAG49Fow&G;M-xpEhJLB6r
zXj$x_b6F-rn20&vgb$t%Sxt3?js%fd9nq^Gh?@$l@3#TwlftzhmzS>CET&>X0XL(}
zeGKeSQCSizb3vsXR~X<BZxSFx>4LwpkB0mty3qz8Cn3de@J^ZakVF`+c=2$W%b?ul
ztkY)2Lei{?-Js)Ow#T>z+dk=ZO_DmIelsCTtfz{7+cN^%mzb0L6sr&h$TJI#PPy*3
z<L^&%C9gb?vsY8fv!A*bW)`4`U0tYV;okZuq-3_!?6jRC@^!5oorqF$+njg30T;M$
zkM0v<w{!>Eh7|m7pvEE4u<pBb{lc~$GG*_!$VC;&SHO3_dZya_Yr7j^Ura~v`!$XC
z3*=Qb@IxvI$<;UL^SZG6W;!DsCL(zNUWT=gjD6{R=U)>#xCl_823|AI>wv@eWwsD~
zaMR{L)3uJ>5g<~o@W}qM9Om7v2e4<>6s#akxYV644B~8H*Ue;+o0XXfqeWDq?+K71
z><Hy*8?GuVm1irWnT=Lz=V$I>NFqQpkn!|i9|DZpM<I3SD6;<n<Yg|CO!f?uVEs5#
zq}~iuhcHNje&l2QLfl8`RA^hQ(@bNnd`lR`u)RK$F>9om#!2`M+iYl8!~2%=varKV
zl<!&Om${>n>nT<dS`j;3rxAUeU0>yH&7L=Mf<5}i_qMASvr+g+yI9yrbX5xYuTfM$
z&>#RkCg*MQxxuPX21FMy&+eju5qi@Vq>+q})Q{@jH`uz05?{A1PF#FaxHP!B+ob71
z<Yk=fg#m*|I(+fRg@~e$NiB#nE22hOP42KRiybbiamm(G=d6JI2K`i|<^naPfA6w4
zy_YaJm=jM~Z4Ds(U9cW+L%q2ssvj>NJ}B+G?ma7NDqEEw)P0(Qy5u)EraEsDE?Blo
z_am`#W91mN`wi~SzDTF4bI0nve54mVzZz;p;HAfxBO$d_{?;HcsfDEr(7Yu~aEXgV
zgeT>-n%)1eZ1JcfsT49kZ$qlYv0h@le0ef@!xD1Ubg6rWtb^r41hIc}8bJfUa5f<|
zDL(>40(4Fc9h*T*(JMy(J!t>HQBNhbPG?M4q2-6Er_;{m_MV^%%^rFn^}*ca<Jtax
z3+Ch*;Y~EO3Qxnibs0QgTtc{yoxSfU$j!ACQCt|)9JvMn?CT@1qsH}QuGkg0e$2p&
zE;NjvCESDTg`B_3)xL(u6fe6ah<~^VHJK1aU~e!3zxbtur5}_WA~u99Qon1w(c9gL
zavhK|&t2Il7ecdu<&8k>X=L(yr12k`46IijZJllyOTcfGq)(@(^1Ouu7^p!b_>z-4
z(Fl-Ru~RV0yD=#w1AlOtttv^bAaO)Wvps_j09{Gay%Oeb7EpBYxm)Pq_sR;aS-y&6
za$X0NZfCDa@TcZO*QC<@HRVb-E!W{eb;_TG5L|#alf>I7RCV|cKs2Xj%M0_y5yH^t
zP<KLbUn%c(hI%beB>s}+pxEA}TQ;~}1y}j=4+!#fffPE%)npD*op~Z^FUbr@J5u`D
zL}wF>lsP5?xw|}nJ(<NBal<oQI2aN~yBSN>+Dz#*KH?q;W(nehqm0{0(@RSw{ut^Y
z_C-Y93L)Fq8QOvMt_{NFDO+awdWQaHxv)=Wk{JX86o<ZXNg_nq#1ft<?i&Z(uvQH-
zrbk`!%?n3CrC9Ov4!ST-IFnC5mGy#=lEqe)n(Ox~U_`9}9l%%)(Kq>&I=a9A4l?;c
z1uFFUOg2dh?#+TbPYha*_R(uy<QK$tH3m-fXnHY!{z?DlFRFBuIkep1){u<7t+2?>
zzH6H6v4el?bE3!%nbt*4tTXaY3Rz%Y<M*m8@;@8f0=hT-13L~Mx0YA2n(~sFl*ABn
ztZ@p>PB4*x%Vj#6I)ZdehJxlWJCNw6`7kQ)OO9ya#LTPgONJKqxK|DU)lt>=vhl`&
zR2)e#{wDm;s@0YVgq|#$hxdL%nY<q#Sk4jqJeoP=Vrnqd@aqrI&(ckyz2;i|jj0(-
zU&*-M>EUNU=|nl`Lh<(p`etE}5PrQ;coKQk@WZ3dArYmL|1CNfGMTV@*v_rr-m!v!
z;KWQ81nyD%3I;>4m<O08W9~B!2qgw(&VFYJ5OkWTw=4IB_yaZFl?_d5BFBaqnskb=
zhxGLc7vML>m25-d&>gg>+#Bc)G(b<72!faZ6J4^S41VOX$!EL!iT6fXg7IIcIOG4X
z2lW3=+Qr83|B-ev{_=qTt7BvNSK75Hj`S<-Vy>qf2eNc^dvY;A&l)<E_Q$YBF!NuR
z`Q)c+uF=py5vP>>_LXa)?Hz6Y2cpqe8d+obR8h-SlG7>(dDT2POurAB(I<?s@=zH&
zfLw=8R7fbm8)2T3pD!GZnWstYE!H1?WC0?Iq)9P3ODfNLKbYVAA}v5J07t&H5uYCD
zZI?kv;|u}~YA-x%NI-tq3kXkY^;?4~oxqk)40FWz+x<I6SdBVjY66kGn&CJAg0?pX
zofJ+WK%q#%mjHs;zzyw_9Y+Mk4qo>P2o1AykCvJe=Z(S&Y~24EC^%+?@JB{BDxbTK
z16yeV02_oXZ8f@H5~F+{B6->hO?kpwRy`+f(?g!2t{-K@_!cqQY>XYlD&@qkVPr}j
z81*$jM{0nyq&`e-pu+o_-CR91=BlQzzkwSaUGolCKLD(o&YvsosZVsqgh$`9jZ+*Q
zOYWG3sD8$?Pnp)3N1y64=?$O5l*c}i)|3Z-ha{;%SUzB0jHG)4>=85B%zGZP&Zx9r
zbi@cA;%VIgTlg(rjwtb#p@FQQT)<<8NK}5k{5B+?28)G4ew`%|t>0MautC4^hJ;FA
z(v=EZBHz1nJWAi?B*^%pk^7e8*%FZ*_t*MANbnleLt${(XL<{@3F?YWPAXW;4g}ge
zwtqwbNS2&FKss(hiQmZ05VEFdjai4NlZm?T6c>6pVDwi!`L9d<8Z)lhpPGlFoE(L_
z?~6jD;i$LnYtiA!&NpQiJd^`19JIe^c9ClB0qKJZceR<$2f&3ANK0Wv60J=VvX<ha
zx8mr~-r$o(aEwvAr4K)7bgr)VaKlj&Y&ECD0(YZbY@a`u%@=4t<dL2V)0>{_(v%u(
zWV3E=Vy|8xv}wdJA#2Hq1ZF4_3+-&0mY($G4-ZZrsOk(^vi5o{!;?Wh3^97s6DC^Q
zpl&@`W1HKtx~EyNo{|3W;A|4(5JF1#3;2B!8Wu^wyU-0yM3KM>-L<=`g=@&OzC5)T
zaGNT7-n3x2+4F1Y3FgcaSR6Ez5mT@ANmh1iPEOXt3S5guEWB4Uro-|?zrvDaeH1vz
z8CgxRI&jaOkbqgVriO%bqzRt(9;WDv9P;na7r*#x(6Y%lM^>)xEJ%(iVUVznyoBx~
zsDIzOqR2%@C?0f0u>z#TA$ClhB#_JDJu*5W6p?CA)wXbG$HjsJ6LQ@&UA2}c;zn=J
zX>{8cLA95Z1rGz|YjvkO&x=0kkmEc}S46l^l0d?L?6N!^<@?j)+uG<NUzXEkv2|U-
z+m0nzfc!nwrn^$D)JLXl>l@MQ3i3}|cB|wT?@hi$jA!>Jg6{$C+_hNUGx1odMlMZh
zSZBUK4#%&CGSv%su#YlU#46+!*#3dXL<(Eo4nD};IRERJI;Q9Y4BspvpWo)u_)YQ1
zU9_|wu_RXnHpWe{acT@jHvJ?Z>w?N$B70ZbJ^MRhP9g4k>V<(=wZLtWtvh7eP>1Ta
zbAdxAa$8iivV+vZjBgpS(r<Z{nNQOIX}$r%90KcviLy>UsHoLs<KE6p_UIXB>CUb?
zaWeVn^xS=4PSf6zB>8rPi${po;7`;Lj0tDclZsei2s!y})>p=p35Cmn(2a|c)||x6
ztm*CP2^Efsz;FfQUk*cQFoi*1Kx0;=#k`q5gwF{VijH2`V2lSpkDka4fs}vuIW@j%
zxN#&G34h~POF$e`N=Cs4kiVFnWUzyO)fsUtPLM>LeYbD6wlccSaUYOPSTc1~%u7}I
z=DrT($(Az}-$?e<ozptTmQ;86?(WT9HEGMqhe2n_&1hTc%5kB|-EW0Mser~Dnm_EU
z$`V8rivrKbJxyjo|K4!2(VnNdsI68m$5GUKe2G~EJ0|qa8-WQJ5U}wc46w+P0m*y*
zE~#oS#Q2p?fr&8>`DEnE9)aM=NM($&s3d?)Bbu2@9}Rm(>qR*RoV(0paCT<ghq4y8
zZb;m~V_@$6+>8f?j(qk3Ph#%9M}o3lZ#ud-Z8~}}%Em=o`|;;5f%ZmqISV3KE9-3p
z%Bu*$=en+?&D=TvYY^3ho>>#=pJb7a^32xPY8?knJiNRVCmX5p!Jk`FF(ZVzPeR%E
zK08+6zqfB4`IgC`DT1`%#t2^h6A%{;1jL?O7v+C9YIO;0yY9_ZQNcjDhMDuViB{cb
zZ$L=m(5yDe!cGqMK`)hfruX&~G_};^At1lV{M7l@&H#Og7-5K|VElnPIN9HQ@ts8P
zOT{9(r<!H8;B?xbmXa`XksRK8(%-HGyIa_|MyioLFLKj6-rrEX7`0HqSCO530j)d-
z=p*#R&T^+{$9F^-z@W}^q0WcuW<2i;UqKytGVU<mG3igEZ`dyQM8>1OO~<32??~v*
zU38>OWrk=3f+%s=3_O6AvF)4{{^2U|vZrKAcP$<(*e?Ci%ku12pNf3DkN#%}m}GF^
z^lrD<s-wEc5j0_xQ|K7ZSQGX~0A}HnivoT@prbmK>EDkGEHbKG?ZDi*kG}-^K$35u
zAug27D4fkypP()57U_{5ru78htK=|Ws_;<-J9b&Yz*$2igARCR(*P&$U!6c&CY#F0
z6W?00CYjs+7h~@jD@qh4=pNg)ZQHhO+qP}nyvMd}+dk(W+j_S%nMqHk)30|bm0hX&
zxhrec+FyN()Yyw@Pcz8QnC$5z?%z82*AtcB4C;!5*RXz}`nrOEU6d;<ZyKXN=9!;J
z{0Pq1tFZ^El>&`?>^CPxAhT3XYT)?h<}~{HAyxFeZHmC%y{S62f4a#+E^auyhHXUe
z*^{ecx5v+uVs%UJ$J5D4i?09G-SQzmd5%x%?hPJ4Rq6L|%Dh*_$Fj9;>3Z}15gQ5J
zbV<@zc@w+WM@8-a%-)Cco#|`}Qa^30#YdaB5Bkb>QUCtLg}QJ<3AxurJ$1d{%@mrz
zdBx|wS=Ax*iY%&iGE(@0+i`hojGnExedmhL6XN(qn`64oA;*?}4;OrA*s(r!L5F*N
z3c+maw$@hNuFv+?Ej;z(7kEHGyZpb%m)ICN|5MfDzs;98|4+We$jb0PmI%*i{uBAI
zA^fX~>}TLy?DzD`kTJ{dtjXvvd2B}0H<C=n*h&_MCqJ#ZqvHh!K?J49l=5QeAsyee
zb>Y`bOK2Qmm=8+L#uWS}af<#rItT<ADtRE$B=08fXCw*U$pFey4m20ErJ9I6Fr%<A
zi!I^~RX30TG|@rC1(MBVF+<c@59rKSqdrE>Tg-s08N%rw#JMbS4<<WhB_!oBb5Jml
z6f(|8)(9SpfkKkUkkW<#s3gq80h}{?oRbnZdYoGhYDvy|H5;-fD=j#q0T*<WL1rw=
z+jEe?hK>V?N5W{TPytIoLzV7h7NE@y(TLI)02xGK2eREEj^J9l<xrAZu1coRyoXsT
zj3&XFD?A7`%7E^#8X{Gb`@}WwkAvw^kxN92)54)_3hZA%2C~aIORn-1pd<w;t_86G
zpUFs1vr1*i$Yui7X0QbTjj}GK=%mm{Vc?`FKtni(C5cD!fCp;vkb;JYOe|mvF%I#b
zBZF~<l?YA{$%AZ&K8sy2Tw<2R?xqG|&RC*tL=1@X3m_Hhw1;Z}2?79{QPfiuqM<(#
z`o=6BgOkm@1*i}L;Uf^x)<ED<Mk@(O4KKraih?viAlwqNCPD|1L?S6dCFGz5iT~N<
z63jwuAK^$o&VY3UGa(;MHFFd+WSO9AL{7-F7_vwi#FoGk$Vvc#aYRNVA;gfjCjbQy
zOb*c^cndV?2Nky8C*_#r+aKCD%>X@ZS23@vMbY>Qrk;JDW0|(ZSaOgbN(+O8RmNJa
zo7?R*wdbp6=cl7XznTH{Zpn*dUorBdn;WryxzVkBn0q_ru+jz<Wl>?_85OT>KfQl}
zb{Hp%!>i@h)?Z{OG87)NjqbR=w&=*LomE|0cW0ro@Q4^<0a|z2SMT8XNg5@!_zVlX
zjBvvxC+oV2(T5;Io_=MFF19$JyV!w=JmQ?}j6A%}$faWKl54_%k>Ao??(o1b>pnQM
z^)aBAX(>__KX#>i7#pgzY>KsCUB6LI)t9PYPN=^jrXD?StU4U-&P6SQJRwqL-eh&d
z^odTe3YmkCF)Bg2&@0Ja_a*E2%-~M4cFOtW-GNoVZ;X27yB>r}vva$Zfts~<noUPA
z`}-<BtUAJ~Q+mUHn%O41=2K&w<8()o=VHGw1wml(`K5ZFUSR&x1CQCniX9JU5bt`I
zkyUwk;)>M^GSq)b)^z+UHDgoMstTWvr$o_WoG&jXzU6YTWzhvUlb#G1Ko&41@nA}r
z4vetba(oM*<+}LFxXU@`PVo5FuZLUxf|Y~^Qi;9c{~@~9#{(ozQ*A_h_{}yhN;2W!
z%I-o^yy}%)*2Kj@R#e(%F`|Q=&B~Tx>(e_r-#GsP);KXETg;h}{;_vPM%9Gti_hL~
z3ZjIqi{EO#rBnB)?XD=RHB7a{vvyKvp^MkI8|4+PJHoR=`P#dkG{}O>w%%E4i(NW-
zb`E~S@*N~I?wU;hr^&-S5iLG9Q~}6J%CuhUfVV_HT{fxvPbdAN*5X3$vHC`2qb0W_
zo8`{#L$2k<MmbA5g5bG>X;N4LIWiOPKppr^mMT01?^~nPnVI+oWk(j}4}XcO>vq)>
z?UIH{xS!yy&1`om_V#7{zf~?N$#PxI!7av2dQkqyH_cAE%41#CUMTrX&ufV}?Zk;x
z8*Rtdj9)&ZX(qX;NmH{lf3CPy{$daiv)PBjPAPPWcou+Ay&qob=jz_|bo{J#zQ}%&
zbrn{nbvdPH#vP3|Wxf_XI{$(vLGL+AAGcF8PvUWTu6BrL+!b0MapR^qpUIS8Q-Bt{
znikvj(c$WCl*{9uk?t-F{S%Q=bhEj-Q1KL@veW`rTw?3NQQ4F)rV69jR^1W{j-YvY
z!VFuBx$ed$s#3T7!qUsY_mjxk?+(tco#M>ucURM|DGS!(=h(G$4p8ofJmT|rt%2Oc
zJLwg3aQE4PL*1n6spL}yQa!hiGN%y+TK6OH!}t{A9RwrqqQ7s1blv>Fh$-ys{~5LY
z?_vrE`+tfl|8p_@jE1(dHamjfKRH~@1hC+%G<kAnD`oC1i*5Nl-~A!RmIqL)H8$46
zWmnZQl6M>lws0;F!OQsKO&j`kElCb+4D&&bfe;r73Njf|mLzRLz06^?bDSl%InWZI
zIgj}&1_?<Mp1bFW4urS>#wjY@OB&=Xtr8x^Ng|XCM7$I5Bq|amMza8(AaE(LQva65
zHugR^vcv6}Ant&l!9MB!9ZoyQbwbb_z*_*&7Xb*Rf&hljG;XqzWeJeN_y+WBNh25~
zBns4n5Tj)^2<9L_Xh1gr-D)w50ab1UcR+N!Qz-z|WNJSth7LKP;9a{MR#`|~3h?9z
zZXRU?@3;capLTSZG8C%6sYX)8-#q=QXt77BrjwG#MFG)kj&uMg4syWh2x|lIS!1qR
zjI=BlfP(|=j<ZZPERz9()bULH>zfqAHXsTMbYSi47OcgAQCY!tE%8~x5f=Cz82d;#
zll3{7e=Xq^jW|Bmv`1nPEcmN36M;Gayorlfs*S}W+lc{COF8$8Xm1>l{#d|5@Edmk
zV-^H8V5H#<E_wv3Ipa~ph`R)!CgT4w6-@~&goG250ycF7=NJj{Omq_Hc&jp3<A!4b
z5@Q8R3jf=@S|l6T)1n9l5*#~l5gb|&?-o|jKuGX`ka2{O3|O=T2_aNKqOi|Yu<sI3
zLR<jjtYowvG>{|W7wCP8@&JvHD`&k;4WfWS|L<8$5v=~4*+Ha&0O=>ifZ!vV4<-SZ
zLBJDndO896H)r&mM`au^j%sAr*dfq0oZol{jWWZ$KQlq7JKb{Gc1Wz!{MCARJ$^HC
zv+EP`kB^+(o>O`-bzsrY7VndNCic4u4t>fE$xoaTU+fgU^8H7PgmCy&Nb1E0UR?P-
z4|O8?r}iA%iZe%NEHmGa3JxgjejLtXut=%DWSUk@Rt;Ttx37z`FRt9w?N(C09I{9s
zg%0Uu>6YgKKsdZQ`J}!w`{t2RRJ&~~3i44Al`Flz?H7&6zqC9b03NF;Lo}Y|GV=O-
zI_aO4#&89*L1EJYLi$U_t1#xD&)C23DX!oOFJ!dsGUK({1z&&yvKBKKieZ<HJHMj`
zX`Qs>O{}uJu}~-<P(PVM1j3x72ew(*<UdVJ-T!EeG?uvnCp8-p5en5x-FG}?eD^y!
z>#eLfYr)h{;hpgu)LHo@o=~NGU|OR!mFo#%$D^}5g+eZ$bR)AQl9W!OB>x*_D$M7s
zFr2*_-~?6NrFDHf)j^&e_ghQ7%$w7GEw4S!+?VO*t7qb&ICf~?&Cbw<gw)>x54+@j
z%?`&-xhgSQ$xxL~bs+v7{-LLsW-FH3_?u(qA0e_CNaqX~Mv9WxYjL~eTF^S2kEH0P
z6wCLkT>i2b_7wE)lq!G8BzucR+^gfNDIB&QAI4kLMkOcwXB~c-AzD@6j7EjuTBf3B
zht<_$OH$4IUDW4+k{rm-wOU<!a@)O%sN2)?L-vcE0O!%J()W%%=xV67s=18pKQ5)r
zn)6zY1}bLo+NxCR*Xe7cY8S8VtH-*pvlGa-@ZLb~|6oDVmW4)|r?}IT9HmVm>-0cR
zrIIC$-RH*nFMn`MDtUqqMDDVy!!z<Iv);DwHWf%c&gI-`*&^cf1kf2P{H;cjOGb*<
zXZ2>rzugeF3Ejnz{N-D;1jk1xyG!RnEt^54v5l8zuw?2dNrOYURZ;V_|G;V1Vx~O!
zILnn?&vL!8><{0%C5)8rCgB=9Kf#@A_2N%JW5(y{mfhwhhm)z}lCz*DlX$t$HLjnc
zWz;e8PyHp>FdiV+8y|KSIW<|uP!Z?6>i0mmx}}r{>UiKOllUDk3B?OSf0?@dTqvh*
z-GiFjhaqu&FD*o({HN)3RNRHn5#k%LMilDzGTs7vXy15MuW9j>8>q{jy*0t;AX5L@
zo-S8BJHBh@@}$yn?qWZ!l-<EB9@o+y9R(M5`4KxRd%V&^1*wE@=R{T<bgXG`(6Y!X
z1^V68yjgEUrczE9v8|)o3O!cFWvwrtMzr@P$5z!#-7@DXd^>M9{!;SBM}SsV`O06p
z_dag-ERg>q3UU5l{)YcH3jL>_d^s4H|EHDU8O$vwWcDXs{X+2!N<nu^a||=3#j<IH
zP0IP+yTj@=!|Z6@2H=poTTZ6ckLu<Ju8>9W;Yc;E*D4yS2~6{%q$J1#gn5~adgBZ<
zB*3v|HV6+h8PI^kX%!DKNV6U>0Y>y%>KUGNZ~%={)Lo@GRjm6zQL1q6EIOzM*f-FZ
zh)OaN=7>S?QN)<A{uct`Lx{wvLU#kwz#uUI<NLbAna6M&3|W!`iSWi5DLhx40V?Z{
zGFEZyG~g5rrl84e+x1Wcv<N^nyD9=hS^-og*0{<#;a#W#f^cJ*NMM#6kO}bKQIrC~
z34niurU}v(kw~5J2%=qZ4H!`u?}8|+9Oi)0(3c^j9|ZFQR1ORS`eAr5hP~@!gyaM{
zwLpAQ4YahD@#~?7@5D+o!9oOz8x4dI?FtxijlzgGArwH)0!9G&0!RcX`2NkjqY>T&
zLAwD!G!i%j_91n`uu++X9)TwSdk_hFMaW{6Jmf3_aPpABe8h?Q1WaRueSqrM9pG~2
z|D6)B)CtuM?2bqefzz9HGz$o61<-DTl(i7wfJd|_sRH_GYlM;=70N+d2um_piLNjM
zT-JeU<O)&*wmYBk4DRKCVI2Uq9k+qO9T$nrYBH5^c3`@f&ii?FoUOkDwK*;kg8Cif
zigEUH?P$XT=QXAPYY(Y{JDJc3@LzBqi44NVA23jpg5AAH@2E3Ysbe+zxqB)T(5*Ap
zsICMfYD|L>9>l`y0!Ac#@Ba2Uq5X8c8th8z&^f$=r@sn?USp+m(%}!W{R>9<hl_`e
z!=|BAvfJE4Cv1Ke#BrBn;B$Em<@W1K$?sd7-d@vqwAo%Ovvu&HTq*Gzk7p@W?ujBz
zvU~*DwzoEKutqekebYqDYl;PY+KH2=O4x1e+_@3RUZ!XZA$97B8|N?2qTL$@f8%P2
zP<rX2t&}RV;*P#WMHWRY$x}cMF4r11B{eJgWzeclPZj>A6!H_dP+tfD<FKU6X;X<x
z;(sWV9Jh7vSLqM`_9=Z!TrZ6B&mDf3Nw6MOi!YF;A)a|I8<<~TQk;cl%isH~|I~k9
zW75e-b(`=Mm6G)ozWi-!z~y-ye=+NNtngf(59^})^441|S4JT?5<Fw!aJsjBTl*d9
zz1K4tEt!U)wBW{OW1)3f?cU<sB!x!;QNQ0}nh<UJ4F8*^!j@VYXEQAZ!P)QG!(YY)
z<$_u$eIZ0Ei?OHWhNqWnXI=y4!&`X6SX`J%Hq<a3D~ZEhT_00omeMjOI`Az}OBiNP
zwA1Y7@9d*S2~J;h*Mp>OUTM7yzeeSFXL|VO@$a8fnBA%_;}%K}s%iX0tFvS6x7Lb0
z(~_QoNaxH54`KQpQaML)(VQ3rV&!0c_mi%9|0gDaGl=L~xHIe~x=Ow!__Gr+yGHla
z-37mKrBCKaqmtY))07^}g@9ESU>VEf*9YNn{-G};r>4ngditZrLQj;1oF%%pbf;Rk
zmWa$OFD&4;U^bV%dAyFtP+DsA7$Ya%1A7#6Y2R99UFqnZuu^3&t5lCuXe@)vYHbrG
zIEKs%<HmSgJfCBR4lw7KNH^6=PTKXoz<U{EPa)#&LKiw{LiXbK!?azR)N@Z1>pi)l
z<IRO5nP?Z`?9*l;Sh)IYUo$@LT07Qd<uge{zR8km?ny04UN-de@@0Rz2YC$EYs^i>
z)B27AU%?f%R>>ag=H?CyZ(PZ0G0tt8y6Y2#c{k9%<&IoZad0u<$``KrZqyvejeYZU
zEfY_%=hX^nsX0&>y)YPLojzqJzr4t6>L^Dhb3dymJRg0Nnxef-dB4s+{hX1khT+29
zQO1hr53k8Y$U6S!ljj1cmwHWQnVjb0Pc!(_+C}mofh;_Qp>Ax;Tze=kC(d~uGTJ4z
zj(JNKQmU^3KB)Ln_#s~q)~(IPI*KDYN?mXaE4P8rX|pmt{;)lUY2-n$_vtL%SX={f
z)M~1;;=A*;o$4ZKDMMb_rOsIEJ2A}jOD+(XxpT4DvXj%}-&91}^J9OwC^5wERc<w(
zm}Ad*i8Z4F8{JXQzU(~mPR-Bp$wx-<OUB=as-_<utfrDZF4&k0%Yio>sOTQppX~hg
zO;h~SE}Uc^0h)L)6JD;H-Ftc5cdWDvc1T;HD__#$T&<^2WD_Yz?3p18B5{XuUIM4Q
z&;1^DboA5xGPD=yJ|6su1*Qca^j7=`lJj4RnhH2P=lF~RoapEg{`LQbi@up7^6xV5
z#lK?#@PXx>qAA+9OnehL`tF?1O%dyad_I3lEG(Tn9SQHcBdCyqPdZk8z{S0tb%(K2
zO2xDAW@L#ytvbJzqLp>I)PI3mRSY-U#_&(Wm+{tJV)d1h(0{h368~T}y45u5*b9!=
zA#d{sJOd03`G12i6XXB24g0?w!;A#%EFAy)dyna#W`l*1`F~&jgZ}^b^xqB6e}0bt
z82_Q+uyHYUBA^$uF?2B%F*UX~F@@segK~CpGBvb?^4LsuQ&k}CZzN&hfVp97@A@~U
z2*5eoyI^i^Zs3BQ9nt~8-2aVr@7rYGPc<U8kMc?;YU-+D{;VJo%~-+V>0J;4OZ&6I
z{=xBS>90A7VcdNid+7ZH7a#zP<ty6&vtb(>ni!nwtKsF};mF@yfh#ZvC$P<}dn9Lu
zMhAw*rT#~RY(QDv2w4C$0d9{ne<#$~-vMJK^cA(l`O?q6h-|>=-&h)Bb36I3TP(9$
zxZlFZTjPj|-=WZ--=#1<Sl9;S{`KkI+0n<nX4GrlfHD9hD@Jc{dGdnL(kXrvnP<z0
z->OOf+eDX^hWGUMUsp8nd;VDD->J9XRU;ed=KHT-n?<0nDTL==L=Y}%AYK5oNda_a
zX(MN2Z1euU6;quU+ki46H#9Z?qf>us%k@ma9Gd_)08?-Cf+_yG=)myk#N^=Q=e~oB
zGy!F6?tms{Ms9Xuc53#iK8JwZ4gYLfLvt&8u;wNLHD|_FW@b00&$+=q15@LI-005o
zA%Cp5v(ev4%^&kxgR4W3CxTUDGZQnjckp}sGk&7rUpBd!k%?Wb0#Fv`hOfG)qM{eP
z;awOUoM9N6YMo&n9Uh(GZJnInU+cDts$}H;*%N*EGyLo?aXYxLv2OqXJOJglR8eo`
zUW2NqYM6ActL}ro!_8}51!;h*RKHSaq)TKAng+1VN@eBSI>$A70nn`S7(5MY>~HM3
z{0JL695`^cPCB=DI4MBI6C-ZkcH-s}0d@1Dr-kJMlfS+LIXCg7G~EV8xaQ;UD0kR^
zyB!ssx1ewmtR`l_?e*-6o;gP4<Wz$sGcvag<$GpI{jeFYT=O5fP%FwYgVAf{q{DyL
z0&ps>;kkF}U^0D?BaVR0ivq<&!a)Td->0Xp4vj$Sv}|47FAvM7zNy>o&dGenX_S3x
z(q$FY2Fg(Wq!%&9`wxp(06RB_r=QL7Xr$AdfrD+@>*T=NQ;1t9s%CXpWO+R3Sf%tS
zFDDJUpw_y7#}IUH`i&$VC}|g~Z7tmfg8tz%h_JGPb$97yKP<hU_NPk>jJvjBd$55N
z;c>9Yw<k3}NJX=6^0zXEx_tOWqu}=l7idZw9zEkh5~9=O9t?Sr78?c#YzYh@N%8@^
z*;7>KAw;eq^`fM88f4DbiO?=BK#b7mbps53z;qf;%~WENcVkdzR_1Gt9kEG=d8ebi
zQYq_-L!)v4sv<MFF^_=LUcRLVQ?08xh_d*Mu)x_NXf)?JTpHT`du{hR-n<antA!Rm
zgaDg@@MBS0QE~h9In)yeK{;T%T)gs;^g(C4o|w^X<RoOcW`R-4dC0<xf5&KQv+jvP
z>uUw6)uAk5h;j+<V_buQy?r*$vQU!6Q8)cTwrv3#JuxQO!qF(9+<!%~KDs4A$HZC^
zw-yC?5!BzUOxLYkO!18QMb15~H%t_!odrR4XGEz>ZBBZp?0Azn#Bl{JbbfLxTA9b;
zDMP!*cf`Y8*xzq3X^=YK{P*jjX%^3qm9lfknkql1o>@I{HG+)I6T1bg?S&Y%^0h;c
zH&&)Gar7sqYg;q3Xu^)&>7nkI5ikx|)Q71Es6tiHFmmHMCqyUAI7-c5*|(ULBHz_Z
zV7YMe8V-$%6U~<2)obxH#*)-!G0kCt`-TN&@EC%|KH9oxW^qODQlyP2YGCY9W*YW3
z2fFQKeq^KtLy|-HQ3^P_x6>7SpNqjP`aaZyg?dKhapg+uiS>HRi1znp+NESQ0nFP`
z`)*Ahi91rhES;@^dZ$xwUTONz481{Yl{j_OYbFl&M)G@Oiwr4{#igZ<!3L$eAle2Q
z)=yK|l|?3rF3!RUctoV<q~;%h7Rd-(^Hn`1N%Es6WVUWk{|>vY6(VxIV1sLzi&6;9
zD?NGCQiy5!gRL;aWqRP<qbGQED@51rzpT+oleOgwTXuFEPqEW=pGSSgj>?PwI#j2<
zwPk!gTLe@_WKp)*$$?zE$|Z&Bw-J}%1=5COc;&&$R-YQNn1IOj7jbi(CRAS@hHgx_
z-N6~=Ce<G#CfK%T+e|Ya5H=@=+cOyLU}w!Afn4kXJ6rurrN~VNjN<j|%Jh`8ehW+n
z!oz%Y9?L}k%++IMt`z|SIa`7)Yng7e3+rU%fl%HYX%mB2W-OZyqLT%=TCs}r2(6}Q
z=lCzC59qg#Q5o${>KCOm?p>WC$Oa#01PNQ`I%N_Zv0r7d+)QTC;cQ%BytF7zZw2vR
zx0*04!~f!yr=(*NjScSIb6w)(_*U_|Z$PfWx6c{6$WLAsvAZ)-Aui|_F4xe32q0Z4
zwXKcR%25usZ8j2LZmq#sSH`+4><_XJPMdAbP0=+R-J9e;A8O@XL|e_znm|q4?61ki
zN_JF<@#R?HgHcnLh7dDak)ZeZhLh=6_6*39bN%wr&$HFJT0fMc5k*fbPV;!E$&bo?
zgT@j2<7~ERV5V%^Ls0ur-NZ{EYKbE$_!L@p+1jOUlnBqF0;5?j`j5Z%7jtNpH)gX)
zQBgKTiY?zCrbqWx+fLxTf{*cC8tG;XS*MquXF<X+Q`ykc30GcbT_Q)hj`p`RW;Mw4
zX?w=O0*){QqJs(3Y<n9Gihj_(Zh71zU37GRQ7+Vpp1a+3z$M=NQMBX|(UN3MH^?<g
zm$n`A#nJMLxz-%#xnzWB&Y>uZBA6>adbw=zQz_=p=9n-kRjzQSJ;wj@{WVhM&LSiP
zY)j;cjlkQr-=!8UWi~Vr7O~Y~0rR>SSZ_-^H(OeR<N4UA!z!}U_3C|z^K7z=^#zST
zBk2IV&dWr7($4B(Y}vFpwFR>y-T2BLL8QG!?ND8J>Gybpb(aS}jbn4|dt9r|xOe}t
z0FO2)%`>2QB-8WjX1Z%+WH}=93<6MD%p5H>AZ>vxcb{8pRMHX%<?+Fn5RIs`SMra|
z<x0HHO^Ec{5)*S<(ip;m&wh0@tMP7=mU??IOW7>L{J=(7r+e2qvwkjk4?Y;!fWHGF
zBHhZ~MB$r0p;8NHHxQ{hT*cn{6M3<VFu^qp2bV%Rs#dXFSu(QX$J!m=mh;M^tni23
zPe;hMBj()S<JaQ!clw*J!4D1ASsnvHDV_w2ZGpEkYEVbr4a66@(tLq>KWXeeoANKX
z86A83n%xA{XuY;<7zH)Ix>Rr>Yx|p`ilj3K<u|(50p(t3^1Cuw)YQPorVjd-D_tN5
zCz$)?)EV$DeINz88U3ytg~en2f&MtCHkqAYq#$4jbA+Cw8gF?!CGdJSQH9EvO{6t-
zG=CqVeq1`|JijU4j=)6+(LtG5MrJ0sbf~P=-xO`9?uQs!$*?S(-Zm&rULZj_16QgL
zB#N`gq;2(-^@M5@359Q9>(aFZ`1oZ|bAtur77bs5gjw4G7Smb`15$``Iq-!N<!35g
z4r1g<bus!h(pQl^?^z7a6f1Dirg7Xl8KTo-^&EP=Z{-X#PVaI#xh*6O#+cE5VJFl$
z453U&rWM|?w(JTg9`lGEu0hGA_$A#;LqWG>$M5}r-m9><=#+~{{381;WqH*X)Km9+
zhmKRpiQ>m`1)jiQxtxlGuA=FT`|p8Kl8j{oFd;WOV{w{%);^cHzs$&0aapvj35Lvp
zBtnM<#W`z`ln7o{3gqRN!jzXXbr}ELTKL@>VeM?P(FF1U*(CFKh+P~WQ94;WP2!>N
z5QMvg&d#o&O#<d4-+3u!OW_L}fi||xM{DW#2))!|&t9KL-q!FZ2X!1LGsq~SZouOZ
zZrD2t0*U=09B=H<(rfEry3XyE>+QL44!JoR_GNl-lYiWRd9s}21UYG5=NQj;%d3Xh
zPtd?$ojqZx-Wd^ds@jTSWla3<k#{f;Y~~97qS7^eO_%?16jO*(?ZC7MiXA2n=i)?3
zAXF{gT_(EQ3YPx%{aYjRk_zwHOWsONnOwT5A``iM=g5-hLQh=+0e69PE*kmzPjpIo
zbhmn4gs%X6ibmdF6S6bMSyzZpyhXuZ&|$>Pp|I4}aP;`IgYPXt-v0c(b51<5!hnI5
z2q-CYGwR)liMNKVgmQBz-}?e0|A=AyfQ^<Pq;2N`!H<HGUNC<0m6{!vPxOx#?_q_)
z9brFdAuLYdhv6lQ<jZgwd$7yqtpdsulghgr9sL<FX8E(Lqa;BTCzFyrXOa9G!dVDu
z6+-|<GPrGBnt}JssQs9q8L=yajIraGLv%<cEk-gUCmhq$F%;B$c{ac|<g6Yl-33QI
zO~BX&7vi*uu9yNkC?Ju5^cM*tAveW=A69bJ|D>~1kNqj{j~YJtecCCG`K^<V`BoPD
zKAM1|JnAQtfSLVWCvZ1;O+?8gh_>BG0~46!A|M3FTH`fcuYb;ti_np)lzO_ty$1Lg
zT490Be%Y65|18q?vm5UGx`*G7=#N^ImCRP$MdnGQe-S>QD|6snM}^9t5o{zhlJrY}
ze^xP*j_)qndTs_3byc-dg^O^D8s7r6-NZU@Hh5a@Nci%|n)~D*(~>~=mVWL*vuA|;
zvn%a2;2*`LN|#G_&+<l6UQ{B6vqd~AAx98@ktG*F#S0N~{JIq#D#j|HK%Cs6dzZ-{
zfT(YDcKqJH^Xa*yP*WqYu$={~<--SP**`~pp)w7SuNb-Mp*0C}QA^<Qqw-;hNPZQU
z^!upJTA=0%wqD?{!=;fo4;Bvki#q5b;lG}Ja3tGpAtqE`BFnXruhNtCvc2yEjEpzT
z#78h6z)OMV%23!U^(6ELD58)txjtT&Ar%&^US|k%^kK;3;i!W3K5c2MPS>2_6(G>~
z<`Nb9XPc;xFG2$TH)dFgk3``E<Th;eZ<!RkOWnvfp-e$uJ#9$!%V30nZBY>i0R}Va
zHZF)c2##Dkg+hOg7G=pe*WzShwupS4O&kEYWXCk;DnkT=c+@<)Y-|-A<lc-X=Ze#!
z!Zm&Di*Xj#kobpxxgyfz)8SWj$**=JFmaDygEhgvE=C3&+eCADTfvk$u>q}nG_oIG
zEvyDQV~%=gQEiJve_7gB(-`}2BY^d-0LUk%)+cd+3qQ$85E@+Vi7fzzY?p2%K*pWQ
zOzEnDVC=WV#Pei=J|IW9k%+q^YNAW%-T2j6B{zjqJDisoUp*SqS1IJ+1*@m|zi%`T
zVtZTuqMG@w;t$sJ`<7oCm*k2Rj%M1p!Hcyr1x%H9&*9;|6b{x2HgKX!WBJ4=-}SgH
z9AVq5lyr`1Z?nMt4z2{fxk`}~=lJhO-&xBP-cps(r29r1IwwU6E+Mfr>yRHzrn}4y
z_WZ<#)dwA{Y2pv6{QBL`5Xt^=%ETH^HT0o}(Y4{Q1qg!*3Y4YF{LU;;z)59=l*2mg
z^rM@YnmE=L>!<(cJ3f(T;-T9}fZ@cozbgG#dnFy;VHr?Km(=CW4K6jjJ9%L8z1W9+
z1^I1%A5N|$xv0|MxO=7op!&9`rq+g_qI-HLwAS6d<7;U+T}GMfd3C-*o*^ae%?8%Y
z#n5jY`7k8DX`mRnpI~!1z-ZWHQqfr&`D=c>ID=IatmX(!^rokrg5+c$ncq`4<Vvg)
zXKL`2YM?vqBGJ&ZqFIkl6MWD*mcc%`C&<RyJ|s`cuaLA7)+NjxE_tt#+G$IWpRO&T
z`{C785*O!CB*`<B-l-u?g-9yQOzUkaCubf6XsCq;ynU}9jq<svYksnPAX6`uZ3C0^
zGN$v@9zhum3?(8^R!4JVA9>1<DTX1Jj9yV3p4WUR5P0~fYDlR8Njp}8($g?DnE+AF
zX{2a6p0&4nQ0xvt@0tc~r+VH4boI}c(iiddhk;Vuz~R(RzNH(x84=p|D}#jY5`kNM
zFCNvlgG5x{=TRM6gzvwqvMzzJy=o~--5dL)%F&)8pB|0}FpZgZf5G-=U-9Cz<y*Et
zsQby3$KD1|p?~^FT4o6M2;BtwkPyKa)j=1kg-s-~!0ImUIao+lyXS<u^lpTao8+N&
z?dJV6hE7CyFXAS{H;eeFgM`fcT`O9+hi^9wpEwg%v|{%1p63qE)DtsUxYr2YGVh8R
zC%Kwnuc@#BF@jveS@#UL`sWaPLxon1#QJY|ELo+|x{?tGmY|hm)0r|wDG=eFsk~s@
zgjKMkSg27#b{awHyJ(2rV)j8w)Kc}U<}R<#TdkK?$u1dlaf(ny4eL<wVs*6__Z5vl
zU#rcrp(bLZI+N#l5RT#{=P_kRdGrZKmCW^*=J*A~1zspKmLO`6H1b~RQY<Yz;VsDh
z88b0lqq^fr2e>`%+SSoDXtsTXPCpAWu|Yc!E+}CR_@BU@2%SYGfrYY7qG+mr+n_Q%
zJX;^7A_@ic60*`MzYXtxv}E)`VYK7A*btEP{?~fL_u*z6!>c><6`Glm&~IBmZ!IWQ
zsyX9<gNGbF4l9*e>7oPs^d|wxR9EcO598RMoZR)Bg;q-LbO)1~UiaSbJ5kQMF7R-m
zCE))y_SE3cAY!rFAbcn&bK91Ry~XK`b7;-@68Qh@dAOrsHyOt(ir+qFk_1~V!&NW<
zQ;t$<6O>y8msx53iUWUGw?Y(&%bNbi>I$lb<^MdMymRnEa^ui-q=Z<Urb(Q|G{TO1
zve%w2ctY+XW4FBY%(9nJa^yzOi1oWI4$BC?8q?rO?lb-s(D1^B#$5@c;_!U^yux%N
z&HVu|>2~C>d#MqK``BJh8y#UA0}DS_4W80J8N+ujS`chobXa25NZM8)Tmpw~OO&*V
z$AkJBigT@d;c|b!>b>BTD>T99Ax$a@7YJCJWg`z}c~uSYc2%g3@bOL!Fq3KgzSm$G
zp$fi?2akbDim`esZ(k~WhaV;yk-UCb(;zSNMn=s75SA(F9mkR!{8_Oe!iEJijTqOZ
z{CwND8duc+3`Jd1T7e!RkGD*El$tHLy&$D0Y+$h2L-G6T`!vLUHO34}m$1^>kAjl?
zmh1_zzed=Q5itZ9jl{MzU+L#2e}h9WQku{~q<GJ-uzLV%KR*qjq{Uj7?z=#|95~Mm
zrupoPikc*zKF55XWpaAkh{*iiYcpyCz1TqF%T`0%YVdVATh0wmonD*7Pk*CtqJSLT
zvBsw|`lK2V8cp&yigmkJ$wls6LI^q4!lQFljFAzb37sjguASh|Ig3c^o->$9KwK);
ziJ;u0*g&7M<wcGVP?#Wkuu$}yq;-u4qMguV<=t!E6I3#;U20}96n?P%a}mOMDS$*z
zWD?&wZsTU%a)oG7)0KyZ4w_|Fh5Irqu0MY4Lt|gaf(4TDR7dY=I1s;BZknxsPf_ow
z%0z5g7^h3ra+h&B2A0UEAiwj7Z<U57TJ!OxUWL<DT-*vd|Ne?EdpoiiagS_{fF1ab
zrYGa1_0n9h)yGP_Zc_7=E_Ia?RIn++U4N>J^JLajbUAKX=+grsBR*4Wx*qGOTET&X
zD(?I|F)FsLSVQ1x4K))yDiS!JKbu8w1aA#knR^ofsB!fmCl5^?x2iNxbqR?FS{8yo
z&87H)uGR!w*&ZfCrM0daIpeS9&+ae;6Fv$}9~W7MKnA{yfA`V4ss7Nf<XTMT6Ro|A
zNw>b4+WB?H*M=!Eb2&(`LX+C5(Jk}C2g|r=VB#|^F8#WkzcBlJ7P?ZdRhKSKFkdrL
z!y7ePiqr=Q*)4$h<Sf}VX5hk5{wNX6{btfSdJC*3(Pz(`st32W;Dq4E)r{puSxd9F
zcyt)U@f*dZOAx6tIB%)`z_thq;(>A9v+Pc%Vhp(xa{d5A=e5G?N?l+B8RGR0$SL;A
zL4v4?!|kdCW$sm5U))^Qvo8YxRt(<X(7br~_K&>pTC?~Wu^UvmOT>rat@e5D>F>ST
zW)#ihN#t5M*tR4d&b?7T#+XYyN$z3#v29{bJXYPv+t@YnARdzr>9>@`C2yaW-lF?F
z_INQ62{59wL=#Q2$28+A>SqFDQQ%w_<Q`NHkqM_9ksPZ18&tYU>sdBM6R*8_vlH}R
zf>traKm6NQ87^xBY)g#$|AZHrZ{erDWURzTo_$Wv^-IqVi^&@d=m_-I#T|kg6l5Dn
z@^vWDQ5gQ}pSGcoa<t#nh?jr85N@5K>p6%;5@_4FdYjvuAg&BNUHTF^^aChzHX@Y<
zOO9>bZy$KwYT{$0;S62x^Q_uwk_&&Nxj!)uZL!vrRNRy88%O)>6)d+;X2JWO!eG|F
ziwE$j2;VMZ`J-4NcYJQM8kzcozg5~I-377F+j{0%!UmEm+hB~{2{MZL{S0Nr?b>h@
zyGR8mw0l;*BIfAKc7g^*RT_o01wAPk_$Zi2QiT}(c1L>$y?baBUs)pQ7uy4(G9;Fs
zQ>t_=E#TE%1#i2+_&Ao;N;A{W4X7ZT73ho<mE?T)C$$SqGr@Yb9((0VgH9nX-4Eiv
zt7=p9GG<B`+{vNVR3D2J{+Jj#!dWF!=~ntQ*#h!uRbYVVjtpOCzx_ZgA)xt;DltOP
zL^DwKMp;8QOtt$Q6rT0?*fd0EU7t>%BLB#Mk>kQm<CjOCE?P2sA|;==L9bMv!voTF
zHQ+!8u%)rkq>{~bE!jUj-wHfr8s#nCr@cSzdWH4RqkV%ZRB{1Tj2#aIIC7t?4^JI*
zko%7}GE0KPjodn6$0dNmhWTuk=lfhv?xTK3pFw{`ni`LHsNH24;u>baQ4sf7EAT}o
zE%a1`V@NZi^HCdk8`o|knrOq{q<^5_-UFz|3KMP7HRj8A?EA9iF86c{XCT3eRuLjY
zyL)WNtmiwr73uu&h-26J?|s{`45q0JYBw^12-AQNG<xMnEKzRI39eFfMcMZ^UUp>{
zks~QQ61ate8@mHTHSC8~-$V)i_`M5?`*8<XcMd9Gn+Dz({(w99=lnVm+^n(`))6nv
zQA~)7nE~{jo2&=hNnQ5|`k6G-*<lFCK@xxx1VCbEowNoz(&71Mr*1(=4iX1IciHQs
z$(Bgz*?3lrZIf84TENgGj30p0K5FcudyP{*sZKOwUNW8)C(TCZAdMH)<ElFSbXa!B
z7hfM%BR!qdTtXR()a(-07bQcxrd<0Fq+d=i)@MH*c<j-2QHj*JYw0ix<tOJXe^Urh
z$c?pqTc20v8JU>*p`tyXseEQEiqQ2CWS3q8((>$7p>Selu6!le@n<<8?;k$jQ=u6Z
zBD3E#ma8TQ5J*X)7KpblY6_jyBa==)&jc*<rCuX#rad8yFF|4&HQi8z)gYU`IDXN9
zVJAn?hQi(8SuxXa1>Z(pl}gJC8&ial$&TGj;hYa5W{K`(RK<jpiN7%{py~A4lBv?w
zWKk!yLt-ccwYwk$vg?)5I0D^gK7F?wf&FOdbk7H(U50u4IfK+CXCM-PgnKHYZS~2#
zym#TM&i^1K2PXRZebcoTk>_3Fq*(<A0oYQt+6#Ppdul@P+I7@`${Z8E8<0`DEMhZO
zhzGNdw`35-IH_e&a}b@bq>b^|GJh_RGpWRcuxp{DpYK?#)Dg}HVvtSj#06=S#WU7y
z4Z<wbcz6~d0RGXH2mi?GRp#u1+(Ruxxan`3OU}^ysD3%_D`Z!^0tG7$Hl@AZSQWU`
z0%FIpD7y3))`3uvM^G9dLx@2qg_Xh1n=OPvy$Gh)(9G?XjiZCIOnVTqg|^wRi#rNz
z+2Kxgbhva{eKW(musAMr?_MFo`y+wRRfDtABR{{!n&C17#&eLi%XWX1SCY_2H&+iO
z6Ksr!P5xoWCj(0`RI@ps6RXBlgEaf%1by#7kT$(K_J9^S?ACkWXx_+dP_V95j;dEX
zUE~HPcum)`ARpDo9ihepS6(7<i4R@fU;Q&k-H++-g3Uwox34Tq>l~6?Yxin6`E3QC
z3@L+&g9t>Htg>^fST4s$DUt#GP4a0;<8S$f<y<@3({cG?JV{aDe&2K!$uX!`zT)sG
z04wJVZz(ASLCjI(TkDlO1akf7R+nYOULPZ_?$-Fhb^YyAsc1$-TgPsSeP07jcKe9h
zx#p58jd44fL=LasjQzo;1XH1|p^T2zJ)+L@S-N4LL0+=kZV=s3qj>D;P!!X**LPN(
zsJFQa%%^Ea^vThbp$X?f&_4-~i6oEw?EI8BMFR+;TWlPYFP*=RU|Zwb<i*rK;0#53
z|NTZ`$U1j9Yef&MN1<&J*I5UmIhZVAn2ZYrD@3~+L{X}g|M|X#IoZjup%rGfDyq-~
z<Zi&Ejo}ZRH@+Y9^+Iz58y_m*TDL|_rgUiEdXfb(&X=SAW4^mq`NhMcnML3mqf<d$
z-KTrRMw&jdbqhAV6acDKJ*%g1w#b?vR0qA*qbbBNI+Ppd)J%#x!Wr1*^q_Y`Y0*%O
zxc<V1hLV;LH6RK(%QWhdu{DXATTLdgJ|Zbs=1A=U`TR;Q95$he5nW_a+(SQPkJLw1
zSz-5>C)tYu)^d&LJ$oaG@T)kM-}gBs1<z;~Hmh>*QHp7F+Lf#)u~@AU`gy`Ar`7Uz
zXuBH6HE37Vc{?KA#IZ$`{TB$>&P!&^)tPL09F=4J#^(LYv-t}uRzjzwgM;ri?@pHg
zc48i(8=AcZp_8FXxxL43P}ql2|I9p*qgacC`%q_mvL<MP$Cz<m+p08BPO(&VRc#NU
zt}>`I`Fk$AS53mdKNDH0?8o5!)E{onbja<}Dwxk+N_TYMs!2A+pIq9V!MF&Ouylm2
z!c`FHMw1QE1SsRd+zU`uGKfVAmvOBt6d!Sv*%dJE!X>#_EouD`|Gi^0ZM-cyqR~D-
zDrc%{ZDQHJblt?2BjU}x%>%MtZp<(Q0~wD3gF7nHX|ZP{J(DDCif_1lw~b4WM2uE1
zAyBu^oJvk7sJ@b<-Y7C{Fh}^eG8wSNtTDENE4+=9^-~<En%XZB-4dCt+s&#Sy+SuJ
zi)4tV*U!PQj=P*o8+u<Y3#m5znFv2#beLGeDmKw%&#Lo<LIKQD+vpxp_Bp8Dhzo+<
z3Mrzawj_6+_6_W%ddFVcfXA)m-_X<{gc&pbL$}@xYBOWwYFoIatxp}63!$%iuxY3D
zM4t4*Ka1wVMns$_EdJ>Ud1_7US`3WDf7b<nai^Wq-gO~d<*0LJnjIMhsk2&cgS+yt
zo&9Pt@=M6akPQdBX=8BCusjMEB0lSvui9-Qg5oiOfh21v`yrT{iK7r>?iY8!I^HVQ
zg)9xU&a{vt=SR5akDxCn)`_wv7bqGm1tOy_zDP7=9uynfv9+bHB1gp_Y<Xp8BN7N<
zc=BB8_-(J%j&U8vyjb03@W;F;N-z^;&&E|aP}W^l>0Fin^_Rq@q311Fp-m?$c_9z5
zfe2Wshq&X<c8QAiI3GlIv$Kz?wFJ8$ao0A??vDVV?1-&*u^|<47fGfu^z$VMRWqiT
zjn|{qT+b&`tnh~Epy^yi1xcvtjn!0DKwQaY*~2xXcimr4{;eqB_^oOEb_}sgR{KyZ
zC&ms?D7zAU;YoDgpz54-`g!b$zj}hf#mC;vRz}<q)!K{zlC-WjU%M7>$NULlPQQ|6
ze<&Wx)wxe!k)ayJY}_s5#KRSh5vI7tJG_tzl;r%l-r_{bo8IIM_}1W>yx8bH%2f*2
zY(Uuc0Os>4LO|aYCmhj4pR?Cc!`F?4NQS_r5xkvn0Kz>UTFA88&L9(eKF8qg^KuE7
z-_yfUQ=ZdSDlWnQ0>TYSHFGbriFkU0Yex85*x_CgFh<%-jf!i$pOofL<@dMmyE)It
zXph}cx~qrh^DFZg7nV7XOFxsn{IoqUSXDmxV$Q7T3Z(?2HjJn)l?>uYd)i*sC|ziy
zb6KjFwmHZQArN%1ZZ`HL;5JxG&1J7OFByJG?d}6f1g+<%YqE0v;1+FxKn47Ik^tiL
z<FBz0Z_L*;fk{m-IJk3l*t`6Do(U{>QtceoS&P&d*x|=2n+@URx0s;fsCP0h>Yls*
z3Lmtga)oK`?}aVPNx?NNsZ@{mryKA$7E3@&Co7Durs7A_hO2A}kiA~~e)WZvIK7P-
z6DANa7?QyG?gn)T2_p$-#o4K7o$fKOK5Sox;f?jTsUKxeds77f=bxo>U?|+yj_*&E
zKPeAW0|}iZC*kHsvQ;faWalcBtf=hVg}hOUFq*yXMSvY;*)6#oP_*x4Twy)m{VJO;
z#H$3@>ugYO?`f@(lZ-yV*?1x`A6?zHK<53e`-ugcqPQ^8N}?&-smew89K+DO$g;7b
zDRFOT2cgcp!~|v&f_H=?VH8GZOx&cgI{5+JNq(XrMouD$w_Y{k+gEURM^9l~+fR3R
zv1t&YiB2EqlCWMQrB>otaYdDhpB?MtAWyUHn{QQU#Yk<44d^X!+jaLLC{FUNT*;i^
z9%8S~x}M!CVq*LTvhHs|i%tFZwz%<gUNw?4b8`0tE$it;e48V<V?-yvx)VIT7|zKe
z))+y4j!k41O^V9hTHDO|mJ_cQ0p~A<?rL+mL~}=lbP&rPk3(oHv`Ej{(){id%ye{O
zWt&~DcSWikuEHw0g2vobPN}%Vy&Tie*coD)&GD^vAFH_esfrE=J%fxI1d%1<73NBt
zf_WW^&eb!FRzNlHVNp&p)Q@tD1jP<?P^_o1vQ0{D^RDSP%+Oxlfa+RSPR}F26=l-l
zf)1hq&@R=7#yXD50U-ZP*}JLr@uTXjh;Dq3*0iz~kIB_`w=JE+wgci*8AmZ1l5BDR
zI7FHjCP}?MP`-LT*d<1R9<2T%$Z6+2g_QjbP4O53jSd-GrIxw}&zuF`R2ejv3z@dx
z4?~w=EK{J2G(B>7Kt~a2r7f{tO<B#S-zS)Sc-4TpPV*XMb1#dY(o2ai!ouU3#nMgS
zebe~-?2g4vYb{j+@IC!uf<J1+v9V}J!1)dtf+qg4>s+fNX5}Gzq|?O(V_zUNNzYgy
z>k=`~Ee2nYpn&~z8J{Pjy6$EriHsuc9Jt+X3vmPw>p(h(nZ}`X*yw~t!Ld&A8|vN~
zb!w^`Np&xAXy_J7;QNooxf~R$qB%Ma!=4W<o!Wfmg%pOiBSQXCR_mXAPW@UI`IN8$
zQEnsgXwFU)lj6_Fs1w}&9SH_;MI=Y(GPttXQ2H%^9`y0w6S&W26xj{-?{1*9o>W-6
zv#}6+$Tb@claE=3vp}J4E=HSB0_@#MOx_t?Y!N7Q=)IffH}b~O*6Lkd31#FnVLyZv
zND3=SE4<-%&&gy@#9~RJ*n8?ulv9{%ITi%!uMZg>#58sED1i{wB3~<eT{sxD*a*+%
zzqH9UZj^P)k?h)M@t%`zvd#q``P2<-4)~90yH1E|mGIQIp*#v)HDN(SQwjwR?<u@_
z_a+<GM5$?qvu=5zVJblmu!IzG-oxp<lf>q#?4}Zrk&+pYCO$IkqNaW=DA@+5?Z>Ue
zS596c5eqVWDWS_+$<gKBWb`CGViUGA-N<aaVG;M}?eeUG^o_FCX=@Dp?GHEBhgbB7
z>&b1(<%SW`8mz)+h8x!pd0^=Sno54JWAj6fjJPHFufCrY8tm598-*ndM4>1yi%s@}
z2<1Rc_0;P_c(iU=dXev2*@!#f;s08$cMz3BL-qiHb0HTP?OcuRBViwr{9R~ShZg;#
z+dVtTo1p4&^LXij9bbV|vzy7|lQl@ZBakt{uNEa7WVSR9m<ET}%ou~EOH<`i`ehJv
zdWCzrB15ZU0l&r)P^NJ?P2!ObmwX=0QG)Tlq}NKk&FQEy>inL9^`#OXKwH|~9P0dH
zq`>83NT1Yp*cBZ-nuJYQzT4FL$g)X{-fqg2rXPZGfS0h4p$>U?VBy+!Q{+?1^vkgU
zE+mWErI4qWP*oCk?j$IW9W)9r8yob-79tEqZaeFw%|Mblw0ib!<hhG4u%O$+oHT?m
zcNsrYmLBq@dJ2khPH|Y45}0#MZ>~+~{n|s7=m+xR4LUyv+gZRsQWZlTmhlKv#+C<c
z<kyZ0KyA81Z&Xc9fe`wGU{9yl>yyf;mm8AI8)5)C;x|&R*wFX3e?u$a&vP8BCCC`&
zoo8}~RwUGzBhh&zErN1&$^ytNsw72~(+4d6cajdVPUnLzMc$4TDh0TgkO@@D@fQ^g
zV}-8+s;!icb<%we<hud)x;!~#b@$bHJ{iM>j6?e70r0Z1zv-`yEFbrYtisOvTK}GR
zHQqybq(ZtH1LHCpx)6?Zx9*P<t_3b<G=`-7XRX43MV0Pfr+Dz2)w}P;blcp`Za9bX
z?&W<VfjA1q{VMBW%$%xc>N&c6%+_q&u5H8I;Lnp_NAxn<ueS%UUk51xRML$?A?Tn1
zw?^<&EPBpn+MbG(L1fsX{cR}-5t9)78sYi__HK(ey2k?JV8If7sz`H^Tel+jM=14Z
zb8^@y-P}F-;)@?pn3I{PKK=6E9gv#lOCntc;*&Geq?~?uu!H)R;5gfSZMp_D6Os`o
zsmyXa2HK*q@To$YxCHIGKJ~#2MCjl>jcTu2eIJ|XBeOW+m6GN%x*G7ySEJH(5}j@D
zW+}pz&<bmk$!Z(!7FH@m9CDr6hz$InAN;H}!NdO&jQxKtga5!-=Kn*&G7+#aGco=j
zH_?9vV_BG4+5T7e(f@xi)~5})+G1ysw9|I0^*<lJFB?94J?#Fkm)UaWq|B<tRw^@C
z&ZefaN<yHfT7t;l#LA4`+~`_jTwruSrrzr0#>$4?$jYGP-dcp-*vLWt4is`g$kN8>
zK*Y+<(%cju{-slxn^{Qn(i4@F6IP19`Wj?qX1|gt&CE^Y`)}}j!z%~=R_&QwoZ6dP
zhzeiIc}e{f%k#tkLA>&_aC)Ol8!H1v{icxCyO1%EI5;w~Ft)I|G?LY`6TiKdL8kEL
z>dM&WLXup^T<79k<-+L7$U@LuL+H@RK&o8R{GL#DeF6Unw=oyCxf9j@k{^l~*xXv3
z9hz8P$h_|PhF;_H<kHO0>OfHZv`q~6j|)83rJnyTq{Ru$kA?}H>w}mBkt&-j|09|E
ze{1Q@EUYb$-tA9*%v=2}#qQ!Qe*ZrJS3s!0ooeYUX6a#SqG0J_Z1K<RuBQKiC0z_{
zER6;2%xz2w82$sP+L@R-*;v|{D%d+)y8KI0T1JNdj;UB!8e7|$Iy)1v{wHK=XYyYi
zMgJMc#L~|EUz*vO7&@8!FW}!{1w%_a7Zp#3|JC<jXZvrT@qhcWhW`lbL7>e*$H2h&
zPvSq%|5&>J)lS*P$==#j-O|Lx;{RV!NXXuUz?+tdfrEgSg_VVX;~#^OfsMiU|D+qc
zIysrzx%~U3`e&5?@y#s%(Z$r%!_*jRbIsnEJH#p@JJY2BJ%4OQM5BWcRFv6$v}eL9
z!tv6QN+Z}vZ;hNt3hLQ{HUU~F?*+4z%+2dV+GnRLQ{>o4Iw6!0in9L57>d2pla-)`
z5HXKmp$F=(h2v>OB7VEO#gD)P_~_dexEg~2ZBr~z6AkJg#U8YIT=4-O3j;Bz{>R7g
z$Nd1-asq*bbm?S`QV5`)+mSuqn|=n3yzx@M5}QC^K$c-~g6cjF<w|gb3!t1~ufmf^
zvhRr@IM`=$?i61fcj{@lh+-S*5NKBw*e$dP?9yQ-*aqi@oRUpcp$^^6zil}j(K;O@
zYICT(q8Ds}85Q)N#zR&{5PPX5=LH1G*w$FxR#*{hJwGS~D+_qG@3-`;#e-B_rv%f;
zL~$X&2h7uQqIn5X6bZewoGEQFD$}&8)_whf(Ve>DD~8NwFxMt#`tUqjqbCM?Ls@e#
z1wb*S>wu`53wup8YR5u(fiOznL59xy(R`JGSp9<j%T}Y>^N-mcmL%gc<5VGx)2Xhj
z-;#_foP@oe+mZU%5qG4Ax?GI@ug$D$<8Sax;Vuu~VtTkg-{rQYqm?021BF4Gp$Y^0
za2@VjDw=OBkb#c?LTu!ISp$<zuRLrYqm#)=od_OV^aE*HW{ZsjvwmPefGgFQ!%t)y
zNw-!rIahoYsQYijlyzAq&0iMSJcPP31u|eUGka^ld@Gi<Z@g}jaUu2TW@Ky>->6Cb
zqocoZ0&{l;2w>^noEy-hG`FWF6CzoVwk43%!XwdLD`iy$e^Dp~T`v4kSQ6mNOte!Q
z4nga5O9-K%oNJ;^C&bARe{bOqom#Rl0)aZ54n$5G7}oiyiGnJv*;#WVj)kxada>n`
zto!Tz`T@VarTNG@&quN*_-k%0GBBtAntTu&E~Jy3hrC(BzbC_YK4#s@gQ}9e#Yw3m
zgW4UF+ZTxeQ5M&MYiaK$bayi;U>C2F;lIdcokq~b3B*Q)lZ6wov-c>@5QRk4aTZ1V
zNNhZbq~~_7mI!VtCh}hc)EL7c7jr(gJg&`aQ}o51f;^E3-O3dBqC0lPW)?V`6w7@)
zcq&TpF?J#@yCNVnT;f~H<f#KV`$>)Ta7_2s=&R*8E8ym+>FKvvWXqXTHvXn|68>rO
zyhmZbyux)Mx;(WRUaR9bYFwiSUh<ncqMyl0A)4Pn>pEYi2Ik1C!@atc31#asO$+Pd
z{e#7<oe5K<ckjQ7wHib_Y+q6Sd8XsILY`~dMfw`G38BGI7vb#vv>1u5E@aBg>E>}C
z^OQuxt${jaXOB1_+{`WNrkJI|@w@_NoX&yj|AIII;Hu;s-<)4+;!d@RZ@raX9<a&%
ztcg<vL?KOuM-bFPb+8tI>;d<bKobl&MKj(<U8On>i1HW_;znw1-AO2!{-Y%~#RLUV
zpM4s;B>6o{EHmGq{mLup2+)Pq>``G8eS&wIudSzppN+m^TY4MQzd$C}!l_ptOWioP
zus5xA6+^sQ)Ssd<i_I|O;jf)pvBd3jBZX%f-m=&If)vFCdNc|~oEqaPy%S0TcXakj
z2&Tk#s%@QPHnmVcfqqJYQGMsuSN2r*#``vdy+mm5ubh7z=rkIw-Bl_Wd!6Hb&x&U*
z@T$}0-Kj=DejsL--l#RRv;j^STC@j57rqQa-z%!0-|3XXS*6|1D`yq{<eM_^Xv}(!
z8#1Gd5YpKXBY_82fOA?qf`z{2JAe;yIyU9BI&sCbyd1kPwlPw+SDA2Lza6R^%^24t
zX<L0%DKlk$V(XQCJX|B%h})-#Q8n2j;4veO9LPh<$dO9r;Yb7-f+PTO`z>AuhkEPo
z@_FP)(`Z*S)#GS<M=aCUeBT2^O)|qdpM}wf9dXRl9_>|0>oBmU*YGBDTS$O1(%-#0
z)$iSJ@5FxF>=5r*&x(&csf|30@^oR0JaU#K;(i?hZb_4gX;UK~Uf$9WozH2x5z&_|
z=_+{p%5=}~psWW?{En(*c|A2(XU{dR3WilVOouo$anzJK3Sg6&<Z=7mCGUvJJ81(s
z|0G|tht%}YO|(~}$&3$MvN~?10aQktSNdi_5t|ynTkx6SN?v=wM<hSSltu@^=s>2%
zCEj`?xw(9|gzh`g!+oFNd0y(iy5?1tjD%fWvlQ9Kbj=4*0P)bBC^0zqO8gmACoF@C
zc0rFQUAyTVLiR`fB8v#*h86x?^4IckErv<ne$Ys_yuURb427%v$>wQTSDyJ|8r{wX
z^L@;+7C`rOiKZJeWGi2_*qo8aXdVO-gHRxRFJS9eKG;xEX49lNgr+YPDtkaf)InGj
z%*ctfp^n1$6mI#|dS<>THD?yfuIKvh$|#+qfdP@r{HHqlQgnC72UZ9w>O=+ND3g){
zARsF$fw%K|)Qm_vc`Ymt8OzTLcMa~8yuW%-)<WZ9NvM)W+o&?9xRGvI@1d9W&?oJe
z(5NcY$1ge7H2(Jsgsd<#iXHX6XY~StNk0+pof**EW21g!&%Yn;(@|oTO&S>b%zaIy
zP3YnF!oc)wx?9M5Jw~=8R4cnQzegq5vlZ_-Lb$3vuZ9Q`qXnZ6d~O~j$-r)PkO4x^
zDeGF==BH~%eGM<%oTmfJ6yqtqA1wWTT=Tv!Ln+7#Oeo&*P7a@S*Ojflw<?trUd0uo
zCZ0qPM@y|W__Y%Flt??J)JXgBK`0j8*AYIozlWPB36>?peWk7Ujv=#n;^ua=_@vPs
zB!p2%SQZP<{82lK=hBd_GRj90p+{<Y093efb6$DlVjl|Vz;E7G6(<YXGQG_n|5%Y?
z3JOqPA&A0>mN>E+!YJj(lLS@N^o!cC#D-3L^c;Z_uOJ<r!uV!;nwbdHmaNIU8Td25
z@vn0P68#q|wBP51!ReYh;{%D^n;S!;=Cdb=>LZoO1+X-lENv_+oV~f0jjYThlRlN9
zH6KabQ|n;j7I^yNTKKWVje~y%Ly||lkeExLE%6q_Jm=lzaBPJbRmHw~)H_*z!7tXV
zunIK-3Ya%$#D*@YETMsr>`(qNZA#JKn7u?P@=a8`?FT~eAYffwJdguQc2)>{dbaqv
zldyGUtrM&)(W)g<;-mI^-GgGGOzGNO)252-NSM|~hTwUc%K*ujHY-G+Knw4@$P6j8
znrdYZg^Uoh1-@P}a;WQ7$9p=|8x8TLcUVp=)}gL9{57MwKzX8Z1X&ISpgF-$bV(H@
z1c)^-%LNBNny<PnX%ibSCq5~Q$)^ZUw^!_&5T80LH?5^l`(MSea>@uaiX?7u{B-Dq
z3c`ZJ7{(TWIc8Yx$zZ|Apj{+=yJ>dWndJBuEQL0gLR9OablbZM|5$jVpxeQ3FRIgR
zH6L(8^FS(4RO&qpI-zyt*}r?5z<__)+cD3UYc|`PLkXAnuVC1^S4{+^x+pi}vwh@n
z`j7UYuA6eLd{YO#F>Cwjz5C!R?t{}_VE>K*j~(B-O{}R1P@ErVtBu&%;7vb}!{B$g
zGB#NcH4tm1{&`>^*IQDSaM0t<h-9k3&m5RD21b|go>BMP{Qa$?wV%`gJfS~5O`GI9
zqHzqK7eLyijLRTqbhN<q&D-t2Nx^lVL3fbt8s8HZVu)BGftSAtLB(TFY-wlD^W+Jb
zN{}da?^GGM8R;u2F?Qx0Yo`R;K#;9_8TgUh@|shl2slMvPEr2R4lHGU#{ui-a>3=j
z$Wb{^KT<EF#Qt5>RyjpQLe`i)BK7}}$2Hy2)7N(8=9vf&k4>!Zk1qOn>0KG;>0%n*
zLcatjjCk$fSCMec=J{in(r9rU?Y_0b^n!FU%MrcYv=bZhhqNze#OpHOm`M7ia=tou
zQNxey!9^Vr@h=1L>5M%f^Ne3~|0vTlnGV=lH@$?r3LwR|=P)BmDxoVtgRKqhnueR1
zf0NjYBT0KRW2ubj`>irD%=fpW2EZcvh!H1My(Q6>JQH47faYLcE!W5TcVG&lnPAvu
zQ9nf?NTLu;iHfV)&d9x2UGtp5wj!QbUfHpt3|@B#v-7+#@Fl~gOfXnJ^c0hY|B~!5
zD_e<}jBZG)#ZhdGmvWqE!Qf&ka&(R|bp-rpu`TWya-7DgPydLRt3PxgH8OZ>tBfDL
zd4x`vXT>>ga27i>yzr?is`D$MAIV`%q8u#cB~On0v)B2%fAsO7bFBm=aaSoDDd?9t
z?KoEo^Lky+Ty7(qsK?^(NP0|vSvhOhYM4VnlAvGur;OGk>Mj~}_5{te>1FaCNvYJR
z5;$=wsJHkL(k0C`5u$+mR0XLvk$t{CsC_n=l^rH5|E44hUaUTnfnYD;%x_0B+X(nm
zfWJlh1$b$KKNUYM>nU9sDan;JWk4dmyz-8;Md;J@TbwAQYhDIt#l4+{oJaS&=m0I;
z!tZKu*h8Q~&KN=iH-;@YmK!Zwud;vr7!P(-aj;`kS|ht90G#X4zrSGR9dw$HcE>^T
zorL5fZKQH8TUP(dfgm)w&piaAT_7bEXODA~D18Rz5W^AMMvedT^OHH|=qu*FG~w^&
z@5TY+?59<b>2u||7iWH;rg5V7t`w%K-coEYhggDoxk-x!7lPeVm{8>G`~+`v3r2qB
zpbwHO(xY^{!@hdmPt~!-H?E@6GQPK2-zl;U(1^o^g%H`13jB?4#GK0Yj%8g{E(Viv
zmPV}H*H$pq<^H#(+Q&>$!*DZxA!_#|o!)QjPdZiW?y*_91e3e8DT}nj>*zTado>uF
zpfryUA_OH8@l?lckn8MD-!1lEV5QWhk}w=uG)98aV`2uJp|vKqjk}1ndvRK3Yg1gj
zE@>kRlZ-9a%jmkD3kND0g$nPjlT86FVI?wvVg<)5WJMEOg;O|J95j6)ThQVJBzXHS
zPaEhDv)crzyKnx}%V@mPmd&a}dLIv=%J#w=Oj))_?NZx=Bbw3UZyjh0kqohG0{+9o
zZK5=G0!Ik)F#}kak&r)K(@o3%Fe(g`3~S@MyjLR!YaYZ8jK$~F;z^)6XBw+&bQtTf
z=J^P}_a=o-zUmruXxxjt*WUHerR@=;^}bVBhwg6IOp6T_;|O;3U2l*$&vU#KGESkK
z8pcW6dq2kgbrT2O%k}UMOmgRv9;%q5j#GYE`Sedf?tLS@aAR25z5DcoXTfa9jOe~N
z;=r}4dMQ*uK!4+BNnFz9xhQhC#74)?f&`I^)fnve+=njzc386kz6<Tzc&`&qfMG9q
z&V0mHTm$LL#-v`54(%g@qY@Z(`5lvROZOaqNFTRC8U@zzkDg#{1fge0us)%j-hPuD
zhM8p#5wP^X`{zu0k)PV>`1ZybDE_9_kj;-*Au;2T(PELY<(tL2=#d1uCEdJI>_IGV
z4LYs!?6(6kN|!d2e?v^sWBY|xW{shjkf|sxGp9QW$lZOxK#71zcs2ru^l~JB-56Xc
zz6+CzGy$>azWcD=WR@&yHgoZDbh~{r=T~HJh2)ev15;PFWgS`Wzk9x}n^4TnJOyU5
zww+rMKyw9NElxWG8p$fax<tD~i7>!<QTk%R--p@VYk^Wv6?RkZvH*CjG2eox;(vSj
z3QmEw^>!AYnQRw!gwp5>#<75-FNNVzIlUjNvt1$Qp3HqwZx^Wbj#3n_daRom`D(Di
z9T@+jf)FERV6#rQF|I!9<xM{C=Pyn(?y^9qC+?^A0c}{rtZozq)2Mx23&_4Fi08A}
zdE<kD?7+k+h@gh306n|No&|30?{HDl)JW2*NIIkd5SYKBr1X1iZQSKYlrVID?Vapl
zbmbC&ya#EyEY3sOTv1BhI))bR^vl((iq>R7Zyn|7RKD!2Hy|y3kgo}OE)4}z-V|5w
zL9lcK|N0@(hzr(^?p?dqrPo;G*H|ZN=2@B9Z)WM6C{oy!O$<|(XmM}R(X*;r^pC!A
z*#I{#X+~I51@M45=TGt@)(thX#FJpG#$$_APJrlLT4!oHmq56}D{c|ZJqJonWvCAi
zi-TwM=SQ<ttZR7C^-O6=edD@sGJJmH&a!tdH{I(I26wz-+ZH&Z4roR#+lP1nnr23n
zPS!>^<Ovpe0%<9%g(Q|7Z>Edx*d)b<Br=w}&?540*RX{5Rxve>)%Epi$6`^PpWy14
zT{mbhEJrhM65bo_y`s%8CGG&iUQk|=p#nr)YH2S4oKdf~UWoO+43Yyzsgy_0;b@8W
zshmVJ(<B<4pUmCzcnSk45I`eqBAEX9<O4Wmfuid3zLthG_@i1egv$Zl=!XsrAese~
zMAr&%X8Z?pRlef!V0;F1v~iOm!CDN%%vDetYa?6{cW~OOGgMV%sTSYZsY(4$vAo+t
z&=7K6{gzo&<ZNC79$>&&{D2l^@w}5sB1q*oD$3!;u@{k6r~9e@(<aiD)?WqrJJXpj
z+0u*^E=|A|Kom5ilQii1_ZqoIg+;r}a_-v;MGKqZ%*@%A8W`b+%L%QtPx5#YCA1G=
z7?HK}eSN?tmaMLdt&**Qce+SQ!zA&Wiz3qW@z(p}G=$gonUJ`-ZHCazu<NQ6Fh*h*
z&7&n|bORyFNy*<C#{Rf6;qdLI34}2pbc_=3X%)5sS;-_(53vGwP-fkY7M=OeaMa;f
zXael$li%H)<<}72IP1u8j@+1_5tBkCMZY-1qAVU6dwOoIo!Fo`5-GZAIO^w^3TTC3
zE-4Z%Vs9(mlA+>Wq=ypA$c5|K@Y?=$YDxeyj`H>Cu)#B@k7WOF$Me^=!N=%7;JeSs
z5}#{2&4X81m&Iso&SNC}giZ=cMDxzvNop~i{q8+j4x-fSmKSQE7}4NHKj3hABy73+
zSL8XN-s}j?f(vUB_$Z|FYz3U(Dy#Bxri)tHXg#d*_^dadPQ00r%Bs%4LvJ8}!DG?f
z58l6$k)W!@Z9uYJ?9))HinGElze>o@Q^)Y$Wwac!w%8d+y%!JQK`TL+l~XxiD2isQ
zHgMvTuy)!*WN#;mLkhp<eV5>wT^B_bJXf6sxv?+KNFPLMg=`kUT-E99#=;W{@yO3g
z8MWd|{Z;Rv5RDHhuH&Mw5<W!9#z*EO2%#!6$Hik%4nz~h;k7B@$+p3A(V(ONEuDWw
zVTlyafV?;4IvIi|Ee{2L+D~OMS4tSIsx+l=TPLpogALiUgtf?@9URvQo4Y|9Rg?Ls
z!!U}SsW5k%oX78c0yKGf08efgFW~$m)8B$II%BD=<!+p?(Lzx?w)T9?0x983&#ms#
zPyr~WJXh~}{E(a%c2HN9dVb?(`4E8qj$4`xeLnim0TiCbYP6{BP-;C9$?iH2jNuU=
zFGPQNCU7iH-QKjp$a-sfyB?`9ncv{oLLzWoO3gT<yB{72-eg#ito7{s`PY!9QF^P<
z>$08yb>YD)Ch-#vE9AwCU~nQ6H7fl)I==e~H>aC41jkxGlKhk7bK+a(J1;So2QHA^
z+oYuNT`a;^r*5>O30CyOB^aV12Tgu2L{-1X2;h2%I=<pmxx1B(++A2ou8{a#>f0=|
z#+w43javF3cKz#Sj6`iQzh4IjxVDvKi`OhQVENnaCnHpaDH<~?wh@t*7flcFl3iuL
z&bcAi1zrB)201$W4A|^YD*kqE;{uIMLv>}Hf5n>{w1XNCO{d0P1s`LbKAaGe6uAG+
z=k-Leo)@Vg!$VdLfLJ^h5+Dx2k+iXZ+f3?&HDH?Tzor!Sz`^EFnM<e0EDRGuUYumD
z#Sv<7gDY(BrVPZSZW|AUaQU1=%6`JTuduBGcLLtix<&#&F(kM6C~ynVIsYbFRR8wF
ziV(ZDY5}~dc1_skcFbFIPG*(_Oe$C$yD&$ZJ;tGEQtN%d8>RDh69zhxP#TBWXX$!g
zUEb7E>I#9%n?7|Q!@#dWvPj*9fik<B$S#YU&0?HG5=CWmrjm+Pud*Bx*L^%p_OsO6
z=nK9E8OWLaLbu)#P;s=mse#5X0ppiu!hLu&9sR<jK4;eG{2O3Mof1J47|)<^@vzJf
z{{@tcAiOWxJ+#j$yok?*&BkIt48Z}y9y>k807DtOLMAs4_g%(PBZI=bW_3~gE^{%`
z@5wv7M2w#K8EY(EQc0#|jvPeX#Uu`F43xOW9-y(q0Z{?Iq}>Q;=V>^XOwYY1KuP1A
z^V4ryfcQY82NG!VW{0Tz+TE)jm5spONG4Kq9N0t>b|GaOUd}|9^t`1UoH`)629!e4
zuqYrP@4HJC!R-0+jf`9iUz`2S%8Bu9V-Sfb)fhx5WA%O13a#yblrf#$<{=EH(@^oT
z{k$vbW^m)_XRKK1r?FB}Iz%5TZd>xOz^h7{=KO;kW&l_=F+@ZcH8az&0E*f*oCtK{
zgXpNQ*W=Z(`O~kU&#RYuM69S6zT=B~X3jh5*r%N4PAsMjRTB;-rAu}?v>rw%a)E}%
z3m`|mNE9&GGmS2mBce}XCc-4iKiOa~WZeUK-rttlg|3b7Py0u6ILNTuz5bA)#akNc
zIURDPQRGqF)yP=&TsDXG;~a!wfR#ctCvU0p`o~kUL$txx^cm>j(pV#lp{&+IaL62a
z1-U?5^;%Y@Q=}$S_fABE6fuzI$zp$J<F?!(e5utULYl(CmN8nH;Q;Wu1An2-mP{B5
zpW2&SD1tr>yzY<Rfv1TE;e^psp3wr~QRyejWC#%Pk~}K|C8MqEG&|2kxe_$IVk~()
zpJKRdI@K^l1V3u-K{!wSv9WRMbbLe6<DC=mQKxCTit#f20J35dGNtGws!Y7Fo8QEQ
z`85oSeh14a9k-bHpfr={Sj7FBWCi?}Me79JVlNTzAJVDJtZSYOY@5tYxuwolrqamZ
zR(RoK5WZJ9fmw{EZpyu>+D$L8ZtNpJ-s0#rn2c}Tv?U}Y$O$Ah3C&uq)myD$DDXad
zpy4xbg3bt`f;}RcMx=-ou>tk#A2a2bzgxuu*j8C@WvAfFdB~{?+Mb(RV7VS?7#TKA
zGb+fZcSa7^Jhj1YvN+Sg(hj2-C?@c`DCuEi`_$&v_fea8dp$EvnLwZfCL?;k)q)IX
zcz0X8zjT>!gSSkz5CX6EE5^EdP{Gzzqo79GEyvv%uc|+S70IFj(kB7<l9yiUcU1$Q
zJ2x0e#p!~F?_|Rq9l)eVBf!v)2#_MNG9%MZdH}>|DbU|t9{^z!H2{4a6Z|h~ts|@>
zaR&L{L&HzAVSydzy&rYB9aO9J@#Y^oHv`;WLRjH7bzt{?!e(Q%<V%Sk^6`h}7rxXx
zY%QKBHAmddgj-5sjYTml(~s3?HAik^H@==0$7nh1RdE4(0@pjXFn1{vUIIAY)tGbq
zy6{~rylB(1dT9u_U%vtY-aR#w42<wiY|=!!PmxUrtVWaQ;_nN|8fr|`R&@{`(+I$W
zlpNLM_q#PJ`g_oydcE-^migq~)Cnq1n;rV6%sHP6O+0wxamiV?I+~(+!~o`LURdmQ
zW$ZPzNALIyOC*5bz?mf$ovuZ;zaLrg!p(Th(5D<P#|D>D3g05Wx9*ofn7VDfK365t
z^T$fsTpK)v3|ybYUTFQQK^`$lsVl15XXepV5yYN^Z-&!Nb7?zUkbc3F0Bkdph@K`Q
zF4Z?MD52e0kN78#T!!@u0e&lG0Ek_3onD5Lly~Ie#LzG3f}^`WROiLTe$iqM9LztZ
zrck^hu{o$&?&QnHHg4b6M+k@uq&KGi;(I{Tm7F&lg7NZQk@L5X_z3XfEOIoXloDi2
z@5ho65g~o^mWYWPam<zoNrK3i_fLMmdvY6pLis|75UE4;cfI$LO#(w%xc4i5S!_+!
zdu^KU4S%^6qn_Y#&FeHP9QsS;AEn)N5nhh@pzNHk82!3CE|6P6+<8+iXEMf=FltfD
z#KC2I{<)Z}7)TBPmySgzOgG*6as&)RDsCl!19Nvmubm&Dmr$A5!@l3(S(dK;35!yz
zogu5g+dP_1nS>?u`r}c`XgG~m9DVc1k>gyzO=SJ@pmHUt06HJ3k)tMk40)^iDYzlr
zJ<!QYy!j}*fTH=0eW%%j*BIlF>4c_<Jd4%&IAgd$bdA?+TUg;m9bQ)IZi@W~{36L{
z(a83}hd+6T+-;Okn-h~{1gkJ2J2Vl>-F(zo-IR(Eo}vGN?M0p|{J1khh=Gz^V#h;<
zb2)(}=lXtM$aV~>_Le_zj`A_$fezIw)S)RA2wkq=Mf^USf7c<j#hwezac_GDoKIGQ
ztbad7l^W<4&Wj3K8y>x8ijgZiLK0k0`hV`}Z<d~(zlO(SIkf@r%VF))vW9Sl<f`$B
zfALH4l?=xXTGMJ;keOv?HyYbEeVO30-_FX~XP|}<v>8P-M~{jQ_7|IBXwgqC9vDgR
z>XJmk)}3h(QLWW)@4<?}CSJG_h`13vtL~Q{Pw<@=S30WY7+jm2g<rB9)$Y=deI()x
z2IOTAJrr1s!&VKCXr2hCZ(_TytL4qD684JOpJs4Wt3J$D>y@{}!g^wZYls(tj1=XD
zU2yM1!e)gOKYLQ%aD%7SQ`cy>*QMJ#^9>{d5i}V^E0V6&bcA~PJc2CBl<|~)m39Af
z0qVWUv#Z_Pu%9VMghxHhUx@^f`IR*0<rO?8RJF`X1}Mv=zN6%a@d(BUE;x1y8z#kj
z@Q)!I#J@A&kt%(xj9{o~b@Iwkwx9A$`WqVrS!{2}V=;K_IHLBiE_24m805Ae#b5>Y
z=-}6t1RI9ewKjI3Ziy4YtjeWKfm8{U05Q*#n3oo9fj^gPt6H@O9b{nJCui%Q6aAvF
zm}Gv8TT`P~ROGWUr`SUsY0)ig>kJLpu6}zLH5lYgbNLrvqxKqYV`9lhFs}Aqtin1p
zR0j=e=f=7k?oUob@srU3?MHctz8d5B+;1Sf3{e|-KH5KhI9}I`YAeco<-TToLliIb
z!yDP=JrKf`zb2b9Y1t3eTRuI~<uMafHxr6}xv}F`eOcH+cXaJZ@T|>P(W;2y&Wiy2
z%z>;trBicis|=>7oW`Gh8<xlUrHnRm<7npNVt5dh&FclOA?obIJ~&8f#2_PuQ3Bpm
zlI9SJvA2+DUZ$$TXN`~gj}>d#MTc(i&aa_FH}=b%UFLV9P6v=!HRC7hFwOUBozB6^
z>V1>}4Kdr^6szdZZg5+FA<1LtD9M1fe_bkWM<Fw*n}K5?@yIy!f1l=ovHbUld??$d
z1cZRHg5g{S(*RTb?>wEa!ryINmU;mzx8o}DiLBnnbu6V7RNV+HX{({KeL1<K#f6`|
zF+=j6hsx`{6?ds0nriIEjumCp;&gE*R2reK{i=ymtbeZSlbm<MGs<FFQGenZI6Wzh
zq^3?yY3o|~f2<6+k%TomP77T<;?CzwK-n(%2R>ltx$>|IWU39Z8bqD!f%wWc%fX8o
zm#9a>wB1P4*=gPDP`s>0V=E6F0TC8)O??<mEbB~WW|JY5nRGIN_<-j=?`E@qF2v$g
z$#83PM*xs{)pt5Pb>(KLlEj2F3U4p*cI2h{IH%zV)PhVtSGbs@Zo!m;tH7Q;uF11P
z%<OcFFm4#{k)m<F{!D}1#f^CB?ZPhN)AY9^6*ISTH|rdGFj2zB@vX^eQb;r&-LCB?
zS0jclw7SVuqsU<kU&2G%4u~=6U$3l^IxMxM{H5>h%uD<(dljAM8HG@C;fmV#1Ns^m
z&|S87kBzQ5F(yY&c+bF<JF&svS9YafS?Cn4F3@~SHPjNgCE*q={$cTl<cj~&Ub(AZ
z1`T0qE$3FPvK8wFx)q8##f@Y`{axZGpotA-D0;Wjky@;n89w{uZT+o?$=tREum0p(
zQGAqVQmoRVM{VL*@^tMDjg~u&VLQMWx^q0~@VG4`=sk6=IG$o_qXJ|tTSPDjZ#H(+
z+Cn64y`9;dgw~G;>>VY9tk8MQ=-ZPB7J2_$BLP7+krJ8u_H4ye>d$Gf$%3`)0g_WH
zVonu`E1#s(gv@YnUK$H+fmgbN4%-ph$$4*M3RRGCTWOIP)AMBS^%kAx&T2W9Cn{7H
z5sl4XHnBC}U647H-S`uOHuWHGeWT$%TLfYBnqXLy*^K*b5Of3!7UF@M;d%{cKd(*^
zjKb=fwiHfy`I~5G-KUR9_FUdtKRFffE7);AduJhxJm}am1wgWHVXvC{tY_z6ORzrQ
zsV1DA(J9;_I&^I0MI=gF1h#rW<txn#?&DnY_S-%Z9ryY6CpXiu@7B~&6EpTpB2(+j
zMac`UJzNVLNy&UOw2knK^9{6zo-B#y!jxzPkQ^PkDL!lTd@)Pn^<h=boazq5AB&S1
z*9{wU?%Qz1uT`3QGbd3%KV~7Ir*%|VeT)yQaxPrlQvj5%JRFH<90<gNsie&hjXtz{
zaZ|S7YH_3USln@lYE8@kCp4oED8W6AYp3i}=lK1CYqzRnyA1>me2k*@1G<-dCT~cF
zBpqv308&D&kv2iHrMO{yAetF%!m}HP&no9nvPbe!mE#uIDwOejdN~jsPW38Ab~yMA
z^UM^i6W7-@rJUqTXZGa+*;Hz{1F3#9o0q-jPSm@jh%B6mi^YBDLi5h|Sz=F|_tSSj
z<%d4dHGlcU53%bnFrT>23m8hi*K$s@xYWIm`&Ioz1&84(8pa#ITIiX^3Q{N?@jDJV
zt|7-~A1?*{-S6{};YghgHc-1|J)ggjsSVPu!U%}u8%eT*-xog2L;t|S?vXn8dXC-S
zSB_S#0IGmh3mN6b%f+;4Y}J698_ogZ-8i;^yrZliqd1^%-fp*h00-f9UVs1{ylX5X
zt||Y!+-3w#%(_iZJ69=WrWIS%=ot=h)(5KZu!LyGufempalZNmO|dw>fSQAWY2Cl^
zWb)XXvVNK;c!94n<!C3QlL$F@OiE7V;VSzhGSjYBNNsAsNS8(62-r)GN)~@z4mL&H
zLJTvblc9qg`kq+kdo-sl(zKUt5}QSpA5q#Gl>Dz12uX=1R#0scmCEF>1zHj{ok!_r
zqO!y{472+7Ah!4t*AdJGmkyaZvK#hx55*SrQlYb86<Pu9&54x?btzwX&=B=D%5#Fp
zE1`}#@eAJph*%Qvi?$w?MRj@?9WkK_IHeHCe+ryBF+$A7uzBEGARJ!KRu<1+eSL-B
zk4KIY!o3$9G&yA;PrbK`bpSInEBO2g+eeHcGnthdwpXPE{;DKSP30ny;8T5YC5UXT
zFiY2FVw_D-lqy4wr`N(g^y7Zao)CWAEU2)G1auP%;U8*M@*lT-BQln+yUw15$NI43
zTT$9+tKKNvLpVPm79Tdo_PeQWVR%xcAhzLHJH}iUQ*jx<=48>_Z#`SyHEd%xXtDW_
z0(?aARucj6=6!BG@N(!qrTi$3v%R0rkwKRsOlXF@4o;(&B1W-wxX5G{t%N}dJ{#Kv
z9a|g*Bb>I|H-Wm#qr)GXIOSMSIy%N+ZgAt6)vA2k9E!@r21XYpd$6TiRA=5jwz6EI
zfKSBvI{KfONg<52klmwr?cf(d*agBmk?`8eGGFOTS*Du1<Lv(s;#_VVA_Vi(u?;24
zHWt6U1bBW$6;ZcqgSty^F68|fk<TfIeShqCMU_^Mib<h4ai$CkgRh|+fyi^HQ1^d~
z%aZhq*|GaxPysOI7<jv_5p9!yes|2Bf#V+LC9BsO7&j47oz&-16_=7jcCizC=OQHR
zey{KL_|`)Lqxo885gueM`*@WW`zY5Je0WNVHs_KA2ok^^wicuASiksSOYx#p$7F-G
ztYGt3hF2+Mgi2)gw%?K{84wZN&-fsPW0ChX6=}_3J`m>u?tnUI{&QEjg_w3x@z}La
zr`~7U^trm0I9#hRBP4@HtM#W;i35-{MYcZ$0HcM0yZtsFhGZZIwox50k08s^6>zRW
z2rTHfC6^oXFS{uMvO48Nh&f&#TW2q8BK&Ci8r@%w3$NViS}@RUv?S;hX7Bm%ajx$b
z>P)3?BLLf*k`tT;V7!z-t?Aqx@ehCmzc*dWMcnw(?D|R&FR;~<<4^^E$<)!X9_)XO
zbt1CL>Vp9GR(G=oEba97Nsr={w{qttQL<^Rj>B;;PE;b<Ywg_28u+3>u*U|pLS~Ek
zd<YS-o}%l@FK$^}x0~grHD;X;#R$wq>y;)nzS8Mb^!lh&Eb)Lo*NAkjRs*6s5LQ^R
zmO6`6lPELF&FeR8mo)ux?Hd2`vOwCtXWsB4w-K=U9h)0LJR9fZ{di+K=AB(kc#EYY
zaP=wn`syh|xPaPLN$xLBBgYfNQDLe?d!}h$oF`q(#E6tC9(y^dy{bpt(aMe#pF?U<
zrh1aqr9DQF;aMvRfWj2Ri0>VyLaOO0n@gORAVbUm2B#~IN9Uyvx>TX6BRY6nop#Os
z@|Xl4MFYa4!+8)&xB^%K@Cb?7Y<1W`4ARR)5{`V+gY7q)ZL$y*MFjH&U`739Qc9->
zigSKqG?E2rg_g<EsQC7(4;)+#IBaPkt4<=O+?xUMCE>*&2FPJM8V(YEG<u-ZE%b7{
zkX#iUU8NEeJmuNjSeI|hc-jLa%z!6`J5|^pny@q*w!%H}ZqB`}{NjmPmk@Ohj-5}<
z$@OEq6aVRCF6$}ev?mb0Y;`H`6A3vzq$2+~dW5;MNwbWET%HhGljTimoD!R>eMv9+
zjceWH#D^YZtPVE%6P4U?A&P>#)AeYiQbVUMo1T%P>Pd~rkElHk&}@49H*b8c)02go
zwNpEo6A-uCu=SZBO-F+{oMz1;F~JEps`4HHmb%ibnr!6AMQT1@C019$(<7n+16kcE
zB00h`bd+0!6C3n~i7XoNCUoIe4#X)5_BvPpl<Z-QLQgto4+1vdG1I#@KjK^i!DKh{
zkx5)h5Ez15j(6!uA7_JA#uI0NJF@C!n58o`JRDH4GJfA)F^07CLE{^L+>jYuZbr>6
zLY>MFSPmnXl3Z=y!DP@+SqAD&w;L^y<s-Is2~C{D2WQWhkD{+^aW7?Nx?#3t0J6cW
zg>$R>9Uoqt8Pov5_tjZPxdMxq<SDpNE%yKkjH*y#bs*ZP&*^K>vc+ucC~0p{@?J-K
zWveB}mcH{+NU)d@T!M!yYS=m;jL9C50r8JRYG@R^=oQw2x7GC?Q*`Tukw4-EK8593
zYUiSg^S!(EEba;FBvIKY$=!1np`YKx;$q&%wnZq;x-3tJRXK?6N;lI2Pq^NR_a8`W
z!C+oxT;aZ;Q8`BkNu*j&tOPl^z1j85_@1Zu_aY+YThDRSZIi#LE_e_VUymrMEGsWU
zrl!h6X--lk5R@#%g9%VmmxP!cq^o60J}jJP?H!uxH6#uUKu>x-?cqRRh&QRL-bhP#
zEr&yBdXn~!=vm6n9lApY^42}uBgBrtrbOZ++p({a_g_@lonP(k(R)s#1WG^X-{H&8
z&-c1yXzGqrr|-|LAv9$@C<&KQ=jyEc1z;oJ^D3MfFdJ|<vcO%1Q|{f^aD`AC3SEj0
zvT0iDCDjD0)=q?0(NDQJOIaJ7m48wp1_?-Aru!XDF*wEgAymuU321Tzo9yy>XK-(w
zy-k%1b-&ryGn!}weo1X=!TEZ+g+r81)+-e36iuCzO7tLjIA}42JV7oD&IF!18q9sG
zy4tW7E}{@qT0YAogxP;U(6?wo)*p0BotHYdzmt`Wf#?rud4(wL>&qlZi+5Zez_E34
zW0XT8uy6H?;t@*qGr#Oj%dyo?m47TM$(PpNTdH@v?Qaz%44{IWa^vHH&3!)rI>*Ch
z5M}R&O?u=NvuSZA`CC=>`7C11K;xpF&;tvR=FD6txE@Iotj%UczE#q&D~g7X3`5pp
zb+JI#Qa}vT*M=&ASzTD`9NWeb>AiAllrdOjMt(1zwo}UhosPdTwaw*EthtG2saA}f
zG@Tu~+9)J!EayFC2}CYCRw1UM3gF4|w*uP%ge1nu&O}%CC*kKnr!GK6{bc`G)E2S{
zVsmsp!|&}>wlW#ajKb_{MFJa%hl|J8@HlJyq**>|Rfbq5s*|P)b5`BAwSD@LVK-qm
z$;)Vs3A}1Dn5VS&LFp6P&xrbBU_#yXId&U0c7fmu8xP!rVJmFfr>M}qg(;PET^`7c
zGGHfO+-h@NNDWvA8l%i#X-mJ%?@&9ho}MDCUKPk_t{dgc^ib{A+=DLU5Xvtt2sw*y
z+>8L3DfC82N?I{L)vnaK4e+2RBdz?njAgJEW1@M={S@H!L_J_GX?t%A%$g%@mgIe9
zmv_!Dq}NQPhFrB^N#HW5E&_lTJN0Mm4Tp{<b4F%SlkR`3O&%}sn>XE`u}1f?<DAr%
zWn7qwxV9rTRq^`A_~}YH)#d;?>3)5IzH-Y5Eg{?0M~s+QOAAt+4bW~%zb+I2fLu@f
zijn^TtK(_+1)-;2TD#>n3(*+u8QLQ)7>&uTbh#M3X%&P{EW=>@%}?2ng%@vf;>KTy
z!8#bOgfN*hAh*K*T~iLkm`fwy4e#MSmogfgfkt+Q;&?+3zU!klc6wSLz3;t{etl&X
zgb!lXp;9OHoMCa+_u!ZeFgV%lwf=|-SYM4{aM3o!A}y|L)3cs4<?lW4xkVMtHS)?c
zEo9=ZL>+|9_N&>F40yL6+FNZ2F4^X0)HMJ<%Q}`>hlE!BiIU-VG}s4U8xeoEiQmNb
z#VsaHbK`Vr$$Mqek{nsLGJr%GkvTO;&j343S7-#*Vcj>8DZNw!HJemC<!vzSjk()S
zh&9}CA%i7pF<pT;FSCM6xSjLjdRAHBnz$sz#M-ii2*C!yq;bqWe@%Fm0NAE6vE5Ss
z#wYXFI?r4Rs7bR@y^{&e>F4F3l~ySm<pDhRiif>nBOw5CGE!4+9~|+C_8lD>99=$R
zz>m!>w_|T^fsx65U)hGbe%7KoL|&73K;Us>ZfP5-m!bq0j9g`F?nc1``EAYq;70KL
z^8xrNazjNds;p~=i>yh6f-vYBdS49ni^Q$>wc4XR8WVd|{<4gOLV{~lXr;&crl`$!
zpB?rQ#EWrS5`9>ZEPf)<4qOv&lm-Kqhdtlp*vaKC{O&<%l7abHfYa5siTqaYuOsWJ
zxy&<1J%fOFr{}n_QCp2OZyZ#3q4uFTRXfko6#}V#Y()xDR@LAk6bB$!2iVJXk7es+
z7dTKY8*ee$)x#WJm}}sp5^3G$R?gAAX`GWc)V}~9B7Eyi6ezhX7$E&FzUlbAogz1{
z-<$(m@Hds(X%gg&ur<Q#cx_O{w;-R-GaH#?{_OqA?X89gNzSqtuftz{&=&VrP(y?v
zRKpi}pMRsQ5;OU+r57cfkPw7LeTLq|Mi9?)Mw*gr?L4tpF|z}WNj$@XxYi`+^^P>{
zP}hgo15<OW@d=g*#t@8>qlt+S*@z-wU&_Rdf4{Rj&!@1vuzv60G@Is+{WUX)!HSGA
zO<GK{VV|DRPL5DHf737N$?Xb)g`M_R^A`w*f<!u%6gIkKR1Q)!3H~LGU8BltcjPo9
zBe9MH&bE=){Qw79vHlMT(>9j{^ok4R%mE+!VH{88{YwUhU4cyUFGcz)s?ls9AWO<H
zR1&=}K)@Fjr#>s1%ZyQ?q>i)QM+rPKpSC33_YOIE*=T-KDf9ImR!HFtx+ait6LfU!
z2?&Zw2k`MV?g4AW5q1ptkA!{l(UECzSlV~Cp-HH5YM4$?u%-5~nngoVd2rdgfo_1x
zWWUe&vM2}1B1pEhv?i0hU(j^;@9ag(a`Rj^+wDs>>K?<=8&i-b+x<kE=+Sfa`sg60
z4Q`5t?IIQk`&urQ*E^<c{Br-}rf|#4e7+vM^+i~Q-kx~BFAvI;Th4GaIlJfzb6D7+
zx8cl|9mU8X;S#V!z+Y*|l`&|8YE2P`*yDZGwwR$u(E^*pz<7I{5|ebtKRP9JE?v?M
zR;2Xv`dCA!A8w1F*wU!{IUO_oyg7fXUQ}^9F;5H)0oiI^eVTCerucpr3a0k(abh3|
zjT*}08PvYbzZqguOE5fVlye|=l+LLZs6rg!fk?NIOB_6~nDm)?+4_l~K{dBs9<S71
z7q4_nkI8&d#Lkke@}=#g^tfDDbU`70u#-_!tgq~ZNPZN@H}%RuVu!8j(Wsk+nJ2(^
z#~%t8w*9NlW_yUzTN=q)`(qOtpp=9`(4q_=rDNAHa925}!8<P2EAoyN*+I6bc}y13
z@UX{or=SvS^U3|?P1zHeRD2I2*{`BK02qX}LIcxvWl*L!$D+y$cU%@2ak<F(t44Nv
zGPTGu1qU08E+2-J0^>vNzXNnUyr6fNf%+ziGF1lD3U^ms6|E`3Mf0We0m|ybJ1$Qj
zk1%=Md`k^$dqogAcTV;QiCO{<i3S{zsmMhR;XeE;9OCLzx!DzXqnVPC&5j?)ssjCP
z%Q{&Fk$1ZmmavO88ZG!cfPbb23!VHrCrLm4Y|nHSgTljH+9z!8aMZk2F2L)SO2Dtz
z)10HS`|0RXm2$uik`ERj!FU+y%eG|7j#u&%Q9d*?tv6Aky_W@OK{}|Dk-u>fx{r;k
z?urpgvQ_cXF(>C!<U0Qq#MqpZeCGP#DF*wQtZ&VpCzb>6TGS$HK>mJa@NdX)0aWBg
zRK&~<Gl9E`mVAdEz?^Q#>FItuvgVs)Db;S~gK(tx&b%M9-OvDHTc}wVw&<<d3vBhd
zFb>Egy8@ZOtt|(!sNY8wq6-bQd&=flRZ}COJtXIr@Prj0J?4BbuPNl7?M0Zx)cW&#
z*@#OCQWLJ6oOH$h_DD^ES)4*o>c??G!<zkcAIL%K{yMa#`zIGPT1F=WLd<<AWVjdq
z9c&GCtfBqU0GkHjFPxo&eaJN<yPh9-m8Fq3CAti+#)x-OKGf@acdR5n%xMT9H{>==
zZ~x)s!a@dY#VAY(jzwm9HyCl88|$NBwKIRX4SS=nl>u4dbEY5|YQ}`9AU)#=bPBT0
z?)XkaJw5ArqYkAeGuSrWt6C#GW|Q)EGrj8oD9$6;gke_UJ%Z_sC|jr_0;%8iK)3nY
z^kDQGIFTbOwX@peJR4>O?S?8hwNxHhB}RuWtL5!ptwFkrZTqycEH2=ff+pWsn<@kU
z3GcDxNKCd{x#b9K6cTxO@+5b|@?ZSv@9|nEnL7tB<DWZklAB#KONBGJkDRDjwT2zL
z<l9^lBK|>`^%(i$)uqZwU<(=WCm=Eq+@*d&o<<5L7ooLWZ2z{4>9^02#f>-gT;IsQ
z8WjSezu-KRL#3>D@MT~RU=1t__gZf%A$omU2#TkAjr+S_W4lO0W0_jBuv(H1LzKVp
zTE)L8(OQn&STtbd%Hrprb<qg<GVygl8f7wk_owq!f<|yHi>51@+@2z96-qR+-#ywb
zJceq`4({bY=0kMbLdiK&uG_HP6~^StQ!5&`n@eM&8>u<Ew#DlG#Z|S(hRiKET8CF>
zr{9{J+bo2CVxMMq3%4{G-&0pCC5tsyTNC+EDK4Q|CfJC7Gv{oSR>#l$C(e%5dD}j9
z{#F)&VRiY0MW5plbGsa?8muWKyB;w#AbAC5*^!UqaKf;WoOmYY2KzQ`5F(K~zYIb%
z7u(G(`eq>sF<qp%Qu|iQaydkj;`Y4PoaQlWfh{8Ka4n6wdeY?WO#^XVKX%fM`Hj%O
z0f+FUQlx(+Dr8#&Xs<~+g2n$Z9)eVqQ5zB#UM(vyRk(&IHP}D47s1P=D8=r0xBE&<
z_QM5=bNaX**;K0UuBuPYjtDM<9ql{4gT<zxZd3_5b}=M0Ru?I0{Gs)@Od<*c^KXKm
zG)is+g0BqSZ94HxNj1Z~hFo@PE=EKE6^q<iav}>ed8!nxo@94UoO*9LRl|CLS3a*N
zdLSJ?zD>CUsfw!ADeLKVc9TMEm4A@m2^HV&1pTzOi%nPS6Gg@^)e26#zoDQV3ExG7
zX^he~;Y4GklLwC(GZHKETXvr|qzdLlhzDbv28NL4I`d6rD{G)EbX4>$G!{(`5?b)+
zpK8xAdYHR5#zPuH!Xd#2#3vYYu8-`2%DRLF6wcKkC#kX$bD1?PD@d^bT})oH_wi2*
zZ_7vBJ#Vf&jLZd%G4j&uQm&_6LF|hYT4@A+$tCNT-4Z3+PI-TccA1f^%O7Yv-dhYC
z_uMW-X&k0WXS^n|HvL3y4@s4cj<<j8PxYCgCuy!2SQW+!tA{UU>zFO?2*}bz%YHDZ
zS2UrDS8>u+;VoGQ9N-*8>>wPWeSdPJ?p2Qq^^5o!VZpysv4N@|c%qoRwX~zW2tb!A
z@0(+BMaUyn(#SvOZ9Hr9k{Bl0?q80mS?lBWcLV*{qt4Xsl&|^DHK?H1DEZv-3=g56
z0{weqw>zX#*pnQq!|GJ_S7R9LyNL~8;h%4H9eL%H%9uMc{mSTcVBy_f%43|5p2<~y
zIC;2bx78ZeJ06~ph)B1EG{C1yR#IpOyOYs_W)~qyaC)=ZIA{PD8@E9hw*3yALaPiD
zy1BwYtip>e0c=1^F=cIF-sS5Kkg$GlHl4@>=7r4B7|W_w-nB$ntDnO4x)#ab#tqi;
z8Az&Kg7yV;$JlD8i>=7*DUN7I2zpSLvgo%^)**v-9Si$5=P>ONvlkBW;a}Kq%I9=J
z-oG|Y6rwDTR}h!0^%kL@LBvM#RjN@6Lm}A4o%KE*nq}*ZM)j@lQ$cJ@_$i#7-hbz~
zA+z!+xf{JU3n=pRVQj~h&=FTz$jY{|pz{fkF63Nc(0FI2q%UWyEaMcqq@loiC-uQl
zhbZZ|4|^55;<N?BDO5y31X4s!WqH+)J3a5m=SfxSnQZq1AE5b-0U}$%QVdU2R2Obe
z-y&5)0LEJIq|g@ULCCr%n_6c_9&^^NYo~pkL&v2>9&dtA<xAk2LU46>f;i*hW)o<E
z!yJ**JJVLJwhVPOMq_G6{fxn)uhV`><Wr43(l;Nx2HX6URGD>9V34Cpxj`ii0IalE
zssrS%D9U8qDxvL8KzsRnzMEJ@*C`V{GNoh7%V0!od$lT!8mNU>1`o*UH(NIVEdvLP
zPjhuOSqUBx?vFC>T=Z<f)B(!kQA3lHSL@rTrH8FXIdVu+^9>goy<AmzXbe};oLEF|
zBrv@64CwY-M{R+z;h~G1hg=N<4p4+VCjn>60a606c?o!srK26}P_HYHMq%OaV8Zk9
zx*(qykqjQO4XE=a8{g@u<d<43@$0;#L{XxB41};YD;I}mLp%4<<`&Wzf6mJdC-yoV
zDq6m5$cW9L4|u^YoNC!;X=Q<(_Y}Y@6Mvp`1^iiS5C^j4<_31aB+Xq_-W6%P?H!$B
zeVQU?*^K``F5YQL6h(=$E!(zj+qP}HR@t^~+qP}nwr$_;j<^r|{6fCujLb2n+*f$X
zO75ak!RyuA>rUoSd~Z_!xU&F^x{mZ^4HzUdd<q?ySvkWsSYiFm0RKE&j>#hm=tnr@
zD%m8lWEP2C6L}eB0tS2z>zfIHn)K=GvXspg^D=^M+PU%=fP<OmO9o=*(xd26C^Ms6
zZDV!to<G-8C<C~98q?*n$KHy{v4u^UpSYT5X2A@k(rjBbl~Go7v2_^*Zfw4;xv?>c
zi$nN<!bN^K3v?wIk419;gsORf-_@*RT*OBYmVmRh-lV)YKmh2dTj#?b_()aN&A`%f
zvz^5$g3E66;bSyZ5M-kOKtR90Rbc5Q0XyY>wK=bKdBn3(h`%figEBj)cHW$*5HcSW
zZ_p^#Ungw~`x&5LDm$I>U2Z!PQ>9r4%B9JZWn8twGrII(&aoIBvC3Tlt@J33W~~;=
zW5b?nil^)5rrau(i6an9?7t#qG^$G;csnN!xtlVn@LAJH_NCp<&)1mBOU%Lz)3V^p
z|33WpygL#Q<b&e(>)PHQm9FiTd^`l$p_`i6(Sg>SclnhcW{VLw(YjA-iEZ06?aTK4
zP3vnhO5u_Q0%Rk*;{Qc4p*OTX{e$|9M7lW{I6i;ZNm0inI}39x&AMq3pqBzhht&1x
z6onH1$16w}aulL$s@ph}U8WCURIDCv2qo<tKs~}XPssxaIDR`Z4{L0A7bwvN>treB
zBsti~;<7PRv|i3uvKFtFy7C609xOnSxmLp*<y1F7qwZEbG=v7xAccsStZ)UCzpN{a
zfTi@!Y>trgXnx`-sd!YDo}U>`?%(!H_+t%9S`=J9^N=}Bbyx@D!(<zqeJZ{uBV%*S
z><p~G;f=QG%<oIq<y9eR>e>kQ%b2?Dk^EG16$43d`U@&hUbq)l9BM7vY@Cd03nHk3
zJzdVk?GbIu9l@|RwC1s40sqjN6y_jDcsj`A`S%B-#a<x;#wGZZRxQS`fkKiuPAP1f
z%UlOH{Js5}i_WM-$a_zUxd`QJW$t4~eoSInW@bu`wv8%m$ZbtB?k{J&^^?lls&Z>f
z+ro^<K8W8bHSZ3j;^k)BcQGZZb?*rpm=W9A4eRLR+l9dK#Lt``n`=m>0+H9eai+`I
z$7c`pLjqZ4lQ$?rsGWmwC!jk=te8xVKw#ag#g^lR##rno3J&_ZNq!v(Vh56A1E#zG
z>IYR7b(yc?urg)mE7(T4i0_{PY_EMW2VbzC^-L2^^j?GV(wDE!PKoS?=0*e!OBhAf
z=L|Y=TDHrI;qoxtqmq^;zHe0$#bYVpzfu=EP`M<=0KOXuDxl7QFg<_{M-ShbE<1k<
z9^NjELW@jqblaTKVdr>qD_-|9)=sJ#`5d$L%X8w6sc@TC5?W7!5Q(QACzg(RLD|98
z0n;&{p*b@)G7K>MEt|~-p~EPVZ!ONmKzsdO&W^iHmij>blZVZ?s_d%sIQS*4D9wpc
zg&i={b?kqi@;FDCY*i$S(7%X=BU<HZQF6uLtB$G;TvStg6^)V$f$nTIVFd+^M<GBG
zbLfzQ_(}p3*ih*AQWT(r6TVB`k@0xxLk8shG8jyH-{-8t3s5>vIICdhLiKheVOZKP
zFIjRz?rx_6i_lT*7Upl4XSs3!vxc9bu4q6E!c$FJ%CK#HN<G$C`Q(M=iF+Z+Rwz}9
zCEidHb4T1q-=Ag^k--<M46dxN*f9YOg6FDJ`<>BW$~1`y)TWShU3ATV*hrf6c2iSS
zw&PgZAJ6}wiH>W^#9#+rfDt(jPo)j1azfeFOXN;TS!_01ggkW&5Py{$=l<rVclqEc
zUr}+z;Y5FqDE4v84NmY2*aV(9F)<CHVIDtoG&|Hi$uuPuFjg?0DLc<uYw<Cd>9n@}
z=ClcnrOXR6g=P}JI^IYv>8BcVFl+^Tg-WyAV~mz)%uqTmOapz4JbOcOl5nJ<;rs^Y
zqL>!Vn9%q;DEgJEjz*cdt|SGqat@b}8fu=H7q|Aj&$ap&Cv3r2xah(HQae}ljbOt0
zImKBrWvK85Vvwc9YfFi7(KR2=zU<JP-N@O{CnGDHLhy1a#@HfpB3I%kO2?7fUD}L(
zaME;xNlGgCT(}3a^@3P1u7cqZ^1>Z+nmuKqkgyZAWbWuei?37?W9oC6;fu^Wk;xx2
z0~PnVRMn|rrg;hyKsn2&p`%UE<HEOXwA{i%k_9^*r%tNEM3oEY2*KVKLV}b2dqmU~
zRIuG1(qb(?=}I~vaTp6Vi%nNl9!%h9=h!X_ywF#PrWVO`s`|iMVD_RRZ#@lVj}F~5
z10u=OH=$=;KSSyvz%<R}zp&|)4TU5Bud>zkjIb3IoE!(Y05*gb>#q=sXuBLFyYSyx
zKBiOTHpyC*Uq;PJJc$YhG!Ap(G6}la1#(9Y3G^9;nn&&Jjzvh8XG~<$Y@lZ64Cb5g
zX%UE4v&Rnzy9-$$9)8a<4xgfb9qmnxuwopwnz1)!;S-JoS|FrLTO;>#3$YW#>StU|
zNL@W7o%aaGj}pOVs0^0Eru34Ct||PYgg2RjvV+ADd}yaOMI8Dv^=pZDi@;-^|2_=7
zEQ^e)5ab-nAjItpa&6L0pAIfz8AZ?Hm$49TAtaQ(n#PN}fPTV^*U}g(&LuK}YuwP0
z(a7ssoF5necdz>BF`KW>k>YQ-!O*nB87MfE@DP(P%v1T$t+aUfnkf!2=#Il2fI!#>
z@-57B3x&+R=JnI$KWh8fOniMlQDyW1H~zTi@r1@Ge}Wyd4JMQpc%JQbVma*5%q3VZ
zV4r${?vv6c+^_zPpzci}0f<d*zkbV=0#Giu=K#O%d$G2o)zCRSET5)vB6GMOLjuu`
z!`c^<*Lod6g^RnOPtkset(!&vn0oO~v2<TP`qc(k#w$0~Gy$qZt<}H-9zoZYlk$Ia
zWJ$^J#P)e2msmvN7f<VtUyU-iz!eME&|1}fdMIlIM1c>O;284K7-iu9?0Cp?9z8A-
z%kp7lqE+|nf!hTq&s>(H(SV&`G;lGuKr5PM>mGvmr?bl7c#<xhIFLR+CmbpbJhSha
z%3I|}-S*%4TES@F+M>^A<-+9*^-78{jGO&_4LR1z4!_ZK!FHo#YwIs<o5{ommc-%J
zFrUXf5n)LdL}?p8?3Q7p?+Td++eWqI8qqIT5E4T&3{@TE!M%y-%7tdopa`q7pzA&d
z@8xQN%^Uom3slAnW9uqB^(Y+Q&@DmL{BYtdi`}S0q)RIyC9F%SZ(neI4Af68o-`%7
zLooI398FG<sL)o>&&QO&a&zsiEU~>tui^&DJLd&Pjp>q?<jQx#vk@Zdx{de~ieP8=
z_+>|#=5RI*dW)71V41NkTy+APSd$PGnk_<g&lqCkMx2k{x4A%#hCQuK#_~(4Y$tOP
zhs>ujFUQzPZe0?iZCjsN{o*l}#VcMC7pjWXUZSfUv;Moq*o94H^qGI&)12U!ZHT|6
zzSHz!z=Em5|6&Rl&eOB<xz5e0UjE^xq=Md+6fmijI#$y?hVKhvSLVDEmNWYFR{pXr
zWwOHMk@M--(>cs{=f4GOD#-uT$r?DJM5%Ry{iF_w;H2PhezurHX%AI?c8WN0r9g*H
z%yf{_m|HrP64(m-fS_duhTFPhfIO=Rs-!W})RI*qd-1kzc?vIs@Bbb(T44irhB|a)
z%1y*t=_>|Ty$q&<!HM_3tN?)4(7Aj;cRU9RuBQQf;++DpdProNf35P=MILip0XN;*
z@;>TrT$^wwM;Iwagiq{cPGCG`xN(<6)e5@UFE(<G1xQpaDKGTmmk?0(Gn$E3+N5Bq
z0N2d3WCr@8d&odFbnUGCyp2m2h3vK{eUwZ<824M(fiexta&$9j*fd@?AB{J(S0m5v
zQ$Ww#UFqLP0&-uy1m_0FQ0GTYJWGyj1D~f<g5qti+?Y=|Z=;VOM3XeHbL+DUd99H}
zD8Lq#$s{hl^!ia0Ig_k3{khFE-0!+GSPvw<?xJ5F0`jS>qgjX$s_H`DNYhA`9#|L>
zFrHDa+57|U2yBC(qPEeVMcmG3&Ae?(KRoZM|G*@d>jx>m-HGL~*keQzupjm0nKtR?
zxx)^3D|LTKsP8qxYo}Ht;&NzDLs%1b&lT1oVScSXcH0=om7G9=PrxO6Z(bfyXS-$P
z6yu@neajsoQSL7qa}r!Wv_HElnM0&8<I+nPER9h%PpXN)I4J*{<rTCOFd#lF>ny{1
z>G{3zY_;BhkfK^SKp!;I@34u{+CJ2KQZ1s3Wk!*mCYM9nVVjevBCxr5UuukQq5uZ}
zswc$ASKZP8RmBp@F{Xu=^Z>QO*q18Pwcnk-TjBfD3Q*u9OnRS`tUB@?<&i_51md(h
zhkP$J_nW4fx#aDl(;2q=CiD;aO#8Lm?rHBr?63@n<%4qon)6W$1%mW=oN0%`b;j+<
zbij*^!1qp?9aH>?0JO92^E^AAw@d@rSxfsCxOo>^)@#A1ZdEW`69yCIRm=ZTB;+5=
zgg<W$!#&V%lar9FvKdYVxaN25)*x7Jwr}|^tt)PVsD}s7MHOGiX{U`D?rIx-o`g_}
z)-$@lhuN7uR-8Wp87btd8~}f%amg{u{EOdq=RL*5meRioRkEOu*_1J;q4~fg5x#Gz
zl-_rKM>Kunw98?@Jj<#JL<cwoM4SETGBqF#M)-5V;KY5~*Pxe`Gzfmf)*OsY@L>G_
zXcXuB*vz-4y(L||#1Ij)@x0c)_X%QgA5}Oyk;2(EC1fVJ2n6I%xc)enqTV(oj4BV|
zXCcE7(?ZzehhlJZjfHjNxY2|R=0$guW`;)TA5|^*g(TAGp>jIk!8xow^IMOwEP7g%
zkEd2Os;s#bFBb>1v~b@K)GDiZ%AY0k;5}jEG-q}=hYK1h;Aqr&Qt$nuLj=b&rQQ|`
zewdAu(v*<qBsm`*u_-+M*DmX-*jZ};#*`rJs&>na21S|dwhvMf;O}&FxdEBCvuhyR
z&|)0LZ3l`)+WUeGX$pn-qptiRCmi4atYzE1ztd9s$KS=uG=9X)I)kC$9NjU$9ZjGO
zzDB$a5RSoYtsoxa8iih{5p98p$4CGG!SaNsyn<dUq34yAd$s?(Fx!$b>~k@4VbWwY
zXOw-sTO>hScp@)EAI@T7)~|}@XGaEjAx8V`h$8F?WFGef55;l>2jq(17rQboWBLE|
z#A+W;E;W<r_#Uo;92PU!vkEAfTZUI6<!*$gRK`>x%wkOq9wyc-ie&ii7s&IGa<Q>t
z`CVPnJ`GFvQbKEEr!v+X<~(1VkR<?i$5893|K2|i&hAhtdO9;BlZ1f0l<LzdAQlT$
zsSL16nheCAkPJyl03JSG!64I`W;YYpJ}Zftu@{br>)~9}mhqO##nIr}{f-R93N<O%
zhP>@Gu@N0r#n1f-JAoM$h&b!b0eQ)=`<PiHs{-w`;+c^x{We2Lp?5BF-OPa$TjCt}
zDh<C(b3)Ypnp2<rTJ0&LFECU}25_JHcC(V>MbKg=%NV9fUVZCvDj$eRF*aK9z@G(w
zPTd4JrvmK+N|-LG=pJEX8f#|-b*m19I6y=NOxfbdk1`*^Q>51C`KC7`!rdqNv}hkO
z;@crypetlfmt>~)<B=h7Yvq5t5w;6^db;fAu<X=?)VLm1e(c`JL0v{CGr%7m%ke_0
z@;`Z(<JHj%_D5!i@3HBDrOrw^SaC=37z@r$E1_yaOXxpvbWUiUgla}~5ftYMA-KFG
zGlS)h%P#<Sb_rCS+Ng$s$z?3%bYb&DWf`)UAwb)t@5@ke$XMl7xl!MJ72Sf2EUUIE
z+IN$+waiz2V86h*1iW)!SC9{w#A&x3no!ux|F1$g_EDy!O=^_TKXaGvmV@3F_Sr|!
z5rH6r77s=E{K$scGxP)y)(y8bsYoNt4OHRs^379vUEy#fyIai^Q2R|FM4=_jaKgoF
zbw}!4Ce1r($C=(R$TXL>Wj2e0C1K6|*U*!`zBIY{fO(bjg#x|Yaj4ZXDE^c0528vU
zbvP>5=>gaA+pqJXw?jbZEHl)qXjSJ3N*^)Di1qiwSsKQ`cRkr3R+t*{wFHF!Q)^fN
z$vC(0_(B;o$`C=4OFmxWOIa(Hfb%g3MJhuilW!4alT=_phHbzNcV&Tbzno1y$OLgC
z18>z{&6YW1ndf;Gx}ARMx3oeNl?AA16EU4|wK}A{ny}{3$Dm(aA*aTpI|lXw($Ia^
z)v&Jj>TP|<EyT2Xm!?yR|4Ss^EDoT&S9P;uz#4^(bg0K1>~ePkM5`fzr!P%5slY``
z4ZYGThVG2dOD!}ZfDpVJfelGm8MtyE#WH@c6`0d%G<(%d9b}uH-<d=nbFFzR{HfZ~
zpWtvutB~gat~^nV{E9S&^^w+qrH()Dp9QubXq*CnUXwvJrtv8ES}}BA<pLoLtS1bl
z_N<<_1PZ?`qf0Yk6@jMR*^vcS1Kcp>vCVrri1alNG%&2}p5o3242-jZkCW=#(fdN?
zd^;U~mvruCG0s1^$F_oORbs`<IUXa?Fc1UAhS8gp9EP)M(HTI(VC2RfV}wo5zz*N>
z^UaHBc&f%u^*pO(2ZecbUy^G6q3l#5X89*FUf(WB+`eTe&kPZtzi$k<pNq+sa+Za8
z2Q;7y)#!A&e>`B3N42*v)Tw&dw2fB>k308e6edZ0dj*pOB|OLeteUVl)GDW9)T-wT
z(%KgMO0mi~qS-nfvlyoVyB0ZbJ8&Yi6X^vMI4}NQp_4P<LoN^;%?;Eqew%Q!<x0g6
zXY#eJ&I_O-pUkgNspF?F!m5*QsR&-UI3XFe@;Nea^X}ars?&nu#MtD_KrTrn-9^o+
zP@O8PITeUaUBst&eQNeYh)Ilt2F9JSHv8FJUKOcpiu3tkr?Gv6NOg9@_TB_O7VM&|
z3#iuk@uJOfSJ^iLid$jVnWgG&t(N1kE?5M$M(YLLYnf7&|4t2kn=S!}{XHCn1rj+a
z$HxMtT5>1?28TzicT)-(Bgm7Jx&>ZXguK`t5r%t|0XnzezXYy{aT<jYB&f>244g*{
zEV?`&jLJ4?k3gW6|F_c7Yj&?GWZ>}v&v#{|GVE`3s;<NU_tTMMX^=pQCxUSHAR6nD
z4xP|273OwmZ8nkPQ=@z)62q^-W((I*N(N)=;PBga!QYdWE)CmRuz3>nCvY=QBz>&x
zlmF>5gymwbg>}_7Z}<wG+Vn(3YCb(t9GY2)5Bo0`Ei6vq?c+!VSz}<RN`k1pFNRpU
zaW{jj1@m95&?G|Ivh1w;4`nK#Bq#)Y#GJ5p;6h?a?OA=*$ic!PtRw&b*|tVLKK5}!
zgSAQFy^A*Xu23Q>ye-GmG~zSo=+@RZKW677DsN<K9>EoTeXcbe^GyS3yy)1|G7NiU
z%V~?7RjLP1=;{3Ax6aT~2;n-!c7(Ccc?=D{9l=o2YA96nyp}ql9!Lgf3RdcERBlJ?
zGbT!k!HL+NXo(nUcJD1oI?>%6dD(-}D*x)ogv|su(&16ueZUQNO3g}j@>g%MnY%te
zLUer$<!q2UQ=}TaPdz9(6xnE_)CS&jnrsqa%k8}N5Y!$z@$QMmTV_o+GC76?V`fHK
z0e=M*Tn84u%T09a^5nYSvEDnxskS`GqRcE9*<wu@*P?~3U*&j3YXChxjAMxn=JNDJ
za-G6bZ4jWrjD?SNDe_aj=IV-3pLxeUN2zclkeQ{KOX4cOqEzg>$?AwDZNmeVgcW_@
zCg`Ld&jCowsQ>ljIhZ(ap>LO^Uj$&VBfVF7g+dd3BY^+efhnz9zYQIaT8nb=eYU;`
zew)-QV~uS@3XyOuPtAD#0)&@p8!-d?KOUFlT%u||tIJp}ZVAdd@FYaLV+6XXwgF}K
zGj3ma8mXD8o;wv9zH~!M2u=QEbu3LaDY$lNlc`%E#HM=%j>x<th{*{#cJWzxM_zkg
zHVn(VwqTlaDLFexu<~>l<A{ttDIxy`&Kc+mYa5&z3ChF~@cxjEX6C2b1rWSwu}-U5
ze57REHJJh$JpP-$8*DS=S{rCuwPptBB9bHN;Cea%1B&C17_g@Mv+(<fc6waRoSzI^
z!FBk|FiHGYQ1^C)-@S1I)rl}Dw_QELNR*YCEO%xi%fLl9Z1ClGRn`oMVV$i5_`9%S
zT|!nGQQTyyM4#I#&^%SA4MUn?6E10f5a@cRbm;*8%Xp_YNNmOoy?VLZ)BB;KmQo<h
z=sJ&RQ50!GFXdTx(kDfsW*QIfYP9_&ee+6*2xUCsbyPII=(H3Y*LI<Rl8E}mFYZ!_
z@jF+d8P$di5QK9cH5`SR>Ol%-x<=u-2h=ei_=;}`pKDSbVxvRW2A+Y*sx`p{Aai_=
zr1^cmn%Hqn&0I#LU+eYccp<UwiQGD=y*Cl}sm0O7v|8()LT=xOi_PeE_LXcIqNkEA
z1Dd)kum<hi#+i0vPx)`TvsD_Yw{L@H6uDaZMt{O%0SB=aPvx>5$1!sX@&@HJ!#s!w
zfbf1do=pjA)cSGeOl@hG2AxwYu^pL(ohA2sc*?xn?GMtYkp7(|Aa1SWn3^c&23u-^
z#++kY2?^}bnYk$dFNyCL(whq}oJXx-z0NgMks~BCAp7`@{}sat%}r6Wn_JBk7Ki%+
zzfa6sGrC<|J#Pl00UQx10i|6R_YT0@AT%J=ldf!A7Rxbd&m9rjS%~lH`A=gzQbe%r
zgt`L7eeLvlgv2=>g1EUtI|RZ95U?e+k>!+_dMv%NdzBzu5L%aTcy?0PqfoMvH8|X)
zQ6;j9K(vXj<C3h_Ct%&-zc^u-rlcqe%im~6b(}ZxW!H}%;>N{CcL3g8GnHgZxaaZ$
z?aK8IP6NXi9p5MCy>;@KwHO(yl#(dwvXG>mXh%@h@!08}hJXul3XyjQw+7#mZ3EfU
z-~CE@d`w~&YDoYG6qL_?pmP)sxx7%t`!sqq-1B3q%!`-sF2~QV;fH0A&>IyH(?aK?
z3CRD*xI{P7tpj@rru(pzpi-KMv#r6GwWjOS%sssZ{RD=49pKhlM0jfCh|3Iw0i%I7
zY0|O~MQ^7?voRB(8!2g+roOF_-Qc!#389jZ$Ja=Ja?EYi>n8~X#06krptjefBDwT^
z8gl^LC=ngx+o!aoemKE@n+<+oF_}E<+WmiPO^%x|nN&>Y&*hw&<=AyS0#)^6Vd4`)
z7x=(PD10x6%>V_f^izd>ZgHm}nv?R(I!a{Eno#8$8j<)O(>aB*?6}*p!?u5&Aed4*
z5QfpVq;-88Qv^?&6V^-qT*84}f*}-F_R0#awY5{<r-M*T3D4{@8zU9Hl(9@nALN|X
z<^HauX4M(WW6+4_RU*Wru1z}ctQMx`-1{^&slAIG?0QPrMM7*^+6-SnH)vvy*<&pn
zPGeJ8=@FanMbF1?M%FwAZ}SsVcR!1KKH|bhF!@2{jMi(6a@{bH@J#Y0hf58WU#Vdx
zf2KaRq;?j3U&qK|pR^HDVOqggr_?dwDYq^K7R4VC2fI@h82x%4KZxp2SlG_#EZd6c
zh?-87S9Z^t$lv=xORL}eCT8FhUN$eRdPU^-Qy99SLZYxGEh9NmbIjN4ul$9<S}Hwh
zECyG57d0jB?r233PIEFl5y#9U7i9HL78pGbRn92$7*lQ{I~>2s07HuFHe23GZ4@48
zJGJ!93@V3r2z*1%gBPEQAeiOgiVr(g_Ruz?S((_z)oLvJdTOABYxj}<$gzavWTVcj
z6fZc0*v5<R!VI^`{|ra=Odb<&L$5BF<nVpn#>%R({VzjyC@*6Tn`J#=#aPOz@n5w9
z1@K8#h>$j|&`f;XZO!L`&S-sVAeZXIv~B=(?~?C(Zv7%(!3s{A4<fhPbE2k1qjM_t
z*oPI8b3R*q+L=!U!IEeFxPT{RBMEv^mPo-Xt;jSyrxvqU-~kmw*~XA>sMk351@6J4
zo!yf9m~Wz!x&6qq(g5tT2sxW)IJQ%AiFT?Ry|52mm*R3Y$)CXlpqgi=0r-%_C}Ktf
zfyiZLKZ6;Lxe8>1z`L8}q!Z`~01INHGn*XbrD=?tjBuJN(Sr>fQeJ^9T&cOzFD={*
z%;uAg9{7!YNwnGZs~{QmSM9mG>3KAwg{I^B%Ne~^%noXUcs{BHy;cPW;y31qMcnkv
zd71i7CnI6da`~q>YMqmSbx?4$;1*;eieO#DgpeP0y6h<}vRUui;Ji4oK8#hS(nM%v
z3c=9*$f2aGyGi%%YDMB*E*>t()htv--%I2CFI8GWH>t?p?jXKi&J&O(C0;uyfcg{8
z?n^wt<vEP5(;#q}{!kpv$apPrQNJBg4)l+yV+nl@F@1GX@YXv1cdEn{t3<O`i>#Zq
zXZqgF=~N+yfV16?Zb>z74SmcUPhLRE04L)+&xB|;A|&J~R@1JW{NnDDX9=VzL6g+q
zy_*W4-n6Pu!On5vT9y$F#=JG;?l)^|U#BgG`1<dT#agp1szCuM$}ElB8VIp{82#*W
zi_1UrDPk~WkW<biTOQVdrRs`~)#jS*E`^mR0M+Hsjlu84DI|7gM?ApOB=Z_i1maEB
zV}G0Qa4IVAsxo6FN>^-@^wmW_zHA*j6}`Jzf|BnYybIP&s3V3uN0yc$UwfOjY~Hg7
zAfnPKgdl&UMG!xW+z*S^-UrtOVr_dSMoy_gN?jJ$lg=o({Hw6$<=En6+5=omeu1K{
zP53_wi)I(nu6@~ybtYPMt?tA+R@p33S@Kt?$V#ZkY;*lHp)xIn-W3%ri2gkA55U~}
zD6P171VVeeFawV>+&!ORKPQ^SAa=KvjS<CbVU$1FpcJkVXT*#U`f0-6VnbegSr=N<
zOI^!{j=Is#;ZV-V=f;{4^OtLO4pXk}_9@r<H6u%g_ZTHAM3tl$K}V0j?NNjI6m$aY
z(-&-ALU#mz$$&>X+XiS0XggsiEa|G!N*n35#{M-rbhn>u1K(Roc9u{CbGJ(EjQ=Ws
zBo5Vr+?IxdN}wk1(XB-YR__nbgd&OOBOu|66owmFkctQJ)+}yxzq%)uTAxQ~j7uyn
z0q;)hkbD`y<;spjRgWw>(DPj*ZA84AGJw6ksmCgJ!5X;d(IXRO_1CRz+iXqfzvT3<
zm0o_PmG-NUMI9!ow@>ro`Mwr&5Fz==DApzqVpk=qft!t8E{Dduao+g9kGO8zP79BU
z<{=RxMQ>BGkBPadIwX*z+klO<xN*gbFjykV_@WE^&!yqnFO0x=8GivR--<+%maq2p
zzZq&tY1ZJxOKPh}gxG&Wv?FC;Xb*-6t=ug}tRTnaZ@mO5vEP*5`l%$8Y#$+mo246&
z9I9(qqZ6BdNNvTgnXkq`ER@w@mS))r4t-9PCuY+W!fk@1<HEe5x7%y$puThyROpg*
zvk>qMWN0{Mz7kW|c@qL%4t(OFoX!vX>^;_G;+BhQIy)8X9clDPDDwo>o0V^6EhWP?
z%K!H5Qcy;q2-~X35e~|G_xx&&sQ~_f-Pxkt4%~SfRWU}Y+(`@3GOxHvJ)z-E^W?@$
z2SL@=0)^JE@n5FwPrui1wmX=>OnO#LASTufJydWL$Gi{eFsezGQ9KCdHTOEF)r=_t
zEU@TVtZvAq8vgFL-C!W_(j10%hSKeRX-YP-6ugCQR{3PfbLPacfm){{S6@}-1Prt;
zxhRtMpq8-KA<fw=&NB$5;zPVl)$BBZOFPe_g4c$XKxX915IU31h=t5VAW?#{pzO~|
zjslyS(Cm`<3oosNgr_s>C7(#%!<=Xdhx@XXZ8cm3M$B?^Zhgx#=K1hIjl?gP+Fgxz
zT$4X8Y#$eDfn~s>?dv>Ym64|Fody1X*6?TFC>Qc@=h68p_t4DHwEec*KsB>)JZpCA
zR+Gm(nwRd+f7!T52*g|Y0mW!w<_lk`aJjBd&n-pkLg)GhRPUONe|;!eB-7y?WtZ4u
z3l}!C`FRgIS9&x5N^3Xd4wHZK`=r!$ivkWnaDhI}$y(0viTi3m4ZD1W?0glKb0$21
z2!MJl2KMH`@Rz*x;RY0<fKnX;a_zW?TkZ~|B9+q>EAvv#wibdAERnKL18_>IrMXAz
zJo~iEL<dxoid^YEIB?E~L9Ht1%i!U0^Yd}h_-ZW_Imx9xR3=s&BKR-TTeGusEY#hH
z3e0d?l9RfMQ)s83qn<a>T-t3Uu%*KsV$T`G6}{=J_mT=?*Lc1*aI!@dzfM<J>7UmO
z-a&?ui|<5crXUeSR}~jI(88;~MmF_Y64ln7jjO9UDfSumv2Jl^z=?vtcyiGc3<aMe
zQt0~7LeRPlE-jYpQSHxzP5X{YfWb-J(-;(|!$_rlnRjv*Lj1+8rH=Qc1Js?*iy;U-
zgL;T|@5>{JLfvqJv8??}^H3QapXCDjwMk#~QMaY_t>K!22i1$9wjY~Cda-lV@4S<A
z5PKp%iZ9DhY$YfqXWckCYr&i>&Oxr&C>GS47;frWj_@5_{m*A&jB(Ps5@05Glo>$<
zp5`L{$ny%IRUIxu4fZw^mD;Zk$aU0T4p62#WxJVt<DN?c>5H<fLInQ?;SyI#7dwd*
zIO4qfR2Ns^WiaH$C*F!_#kh;<T;Exf4D%HupZ&(D>-BZ_s^|w1_`9OFMM>RknBm4l
zNrV&?^+24w+RPEDP65^CdSl9Ma>?!Hgi1eiLDyyfVkiQCgJpsY73SF?7Xuz<Kj}ds
z&E$bZaryA>or=Z}k9={cFfTIA_6=v@Od*>0Qi-(|eWBZQlCciI+h^K2`<wCnzPCjM
zG@sr&dk8|2fN(~NoCj>Fdswx~0G5JikwA6WNGvPJf^E0C%zMG!<Bg=>&LTcXu}3^1
zlg8*v>tSxs?Lu-bHcpA8a3u}b(Qy>qoU0{RQA*z>>x`C*YtgnnKY`ZIlPz0hqf!Ns
z+B)uraW5PE0+nRq8rG;#M7fI=J?b7{t8PK-RP^8WJ~RC{6|&z%zuyQU*7O!x9azk(
zG41XQZ@iBo1v<NM<ymEWhZEz4bNh}9ps~kAj&9&oi5_76MNE)Ml4UfwD({dlYdtzF
zmzpaF{`<jz!QbHT6$t~WisENOYz(hLr))@?NLTXHKu!oD>wv5o=z34{4$W9#bRPEB
z3d0+2vh3ab*yB%eYA5YR^>L#s)^^qGv|6wFKMgnhDyp{F2)b1mZPFY8XlXS~@&59t
znUM*6FXbwhrL8FC&@@yE;|mR_C?1*!iKX#|Q;nE+kEoVu`86MJDJ;}LvO50mb_}39
z@SFkXjR`KARi46J<7mXel08|C-jmfUuQmN8wv#5<viEAKcW)#cvGQ@XxFhLg5#I{J
z`&j4@>?npBJarwbUHxL~wMTIkGnZ^wOBl^E6CaOvg>2n{1Uu8O&jw{Y=gPL$&&*9J
zn~pK~kyZJfc<A+^%yGho7Uj{`K6l~U*(_K?Slbu5c|ZMh-rjlSF``ChWV?(A29|&p
z%A=D-Lj!Z;P=hB1O2&L}!v*A?X$|s8;0#smR|dL&%Laa!@>}9?6U=g*|CnkvwKXl`
zT!0&9XEm)@Rz7L~$du3spRtP8&i+*XQK(g&GfHs}cF0Je!@(vD)m>qG4e30i(oi}c
zTBMNngyq>b{JBZ9+>rI;|3Nnw%slL1k8*na5AQ>)tC-U%&f|oSux%GL_V3O6pELNp
zhhy|0V~|}9dz_=tL#eUODB;eRu4jXb&Rusi@u?Sv33u;&tPtEt2OE%lW@My7Hnq=u
z216UoXP3#@G#ur;FfV@Q^?Kdiousi>AaSfSd5XUD2;qE08!G$usr$IImbGK~k^G&{
zzBX-9Kr+g=>9dAN<)FQTVZDgxMZHuMrcA;H)tPX!f!BcCCO@z`+7=aS(QnyA6nh>T
z?j#8*n^hU)Yuo*-xc%@$Ygi!^^!RF#1G~f)=1AYs8)4Www}E!+PMRdL9$-@weR>uw
z%QtUL>(ABtoMRf%9}e}I(0p^iwIp|&-x=J0y<b1&EB1nLOy;oklMX#WuaVHsUd^ax
z+#eC9vcpJi93i2WmZKzM+g1I=7DuuoUyjhTpW79#%{^J{Phu%x;#bJSL&+tIP8ba0
z6{7Yy5}xJQB!Q4q)n9Z6iw9jpAm;e$>OEfrW)N^M<Sc#cs@5fnq0j?<xNCL+9}%%A
zT`!W|V?q2)Em8$U?G;-vu4PH`uuupkqY^7o`L}BCU+K5W=2rC)eR%#o@(0qO-MxNp
zQrvKQAL{-nVTTKiQhdjTd13Q{$La`C;=8@OZY!(2&#RW<ZR7?;5Ww_EU8`EnxL(yZ
ziInjums@7PHdZ0G_eze@)rhlPD#KWQwCmNTU*r~m>kzlH1Cxfki#fNRd67?O2Jyi5
z%=~P&3MfB+znv5YsJvsz@2U3hau9O0uMHEB4@2iT(&FV{p?Ce9ufaHaFTdMf7{Rjl
zEhi5Itl?CR7Dys=04<u^RPr6!N$$2Np8+WWSaI;{7~s}^O*xk6d0@wt*E9Kmo%1i5
zk)#5eVtk<6GlXNHDX$y!Zs@|E0kZ)KL*x*#@W!ag&Y~JvKXY;b_-rHQIN$b&T9)Rt
zte3-1K+85r4(;%bQ)~7Dr5~l@O*R%O2~?P{o*ls3Op$$Pp6K&aD$U_aW`P`Ez!|oR
zhpZ8~gS65OQ_!s@kouT?*G$uVh6z29n1&Kqggs6xx1sB%U{s=jEI!B|YB!JXGCC4D
z?to5Im#{N0SaR)tGKSISB^{(Rg~Tg~Ej>i}jSqx#=buSq$3OT_A}oY7uRf`+TaSaR
zzlTca^yiky`(aI)_K}==ooz@nV<Pd6ZmY_J9pXLf9ER(Lju|<XMR&b%1l)~W;1>=E
z?(BLk$L+^+O*`@qom$v60nBW>EPhHzd~j@KLa}MxATKLg$gW0E!xH1b5AeMBg{zhd
z@4xt`ijLJQ@0uI5jbX)YLBZ~FKOq(*;KED=T*(b{;L_7jE0YuDWgV16`y3N!R9y-m
zz-k7@gR`1iUgO7!`K=q@QjA$5ewpqa0862~WA;s>x3ew@P+*yGOLI3z3N+N4cYS>9
zVZWrrop$li5?w)0Lsk1dgib`&<1$+`O)M7OF>*M`=PvENF))6TC)1nWv(RGlT<2uh
z7Epjr{-PT;pPg!~D~uNDZ^ETco}G0Cj6}GV6zhJ^2|Mw<-hQ2krw9ywn(nm|DIW#d
z7($xxaoCuowoWh=&u%>unZViDVIqb=uyaNnVRkAyvq|wr2;cD~cFa@1r1_WPCBjD=
zuRIl!=2bpW-bZZPxziH#;!6H77T*<A90C}r6x(-yHId{G*4OIua*@>_4Pbz2GL;fr
z)AIjGGUfxUnG1G{|181soH{qGBQhJe+oQ2IW=wNCX&$OdO2#~Bqa_ZlfpQQ{>Z8K&
zNrZ4w122hrcjhEHD%VDLrUiPf9<pKIfbi>(rs}nZhxxE`6Jp63Ml^&liw3~e^K~cT
ztuftyMif!EN6UFZ&2Gmpz}i0!?+>irJhn9BL7V!@x5h5npHqUk_S?Z!)r&vcir%XE
z=IU)(GB;Q#bvGHDwXc}#rZ1IBl6JbvGYv#595jcd_jYIbUDjGnd5dKvT;8I>hOUcK
zzI;<F2C!Wx95l)%fFo`mH<$3QA-D#z{b+vZAFn@`ib;xm<`a}x`pdWzF$m2uL`A%Y
zywl>MnqRg*<EaN;u?0pVnP#3Ni{6`2I*@f$KuOq&tb2c{?{+@0#m&AFlD)@-XUtfr
z2Zp8LS>D(~-(*o-t*yT1c)rQM{@c3PAa|P#Pyp?-L{gjop@gUCz|3scX6JEw>0+pZ
zJ^0nBXdGp-%Xx3wW7NfL5cE2WgRLq(M))Yz<CB3`Mwlv>+xuPmB<nH6JSHalGAjVB
z$6J?un5Uyg;J|#}l7#u3b#F5JkNo)3TIwOf`(U#TG~`1|l|~rj4*CnQzRV$<Xn`SZ
z*u+OENCT+`6#pE1ClfBB7f{;BB`M?bE=sy#s?kan45@yMKwdM$Jl9$}05oA-d^xC2
zok4yDKuYnEp=w}zzHdN#U839}qfRP1+clHL?UYi8E&+yg8!tEzgrZ^ze2(?sUZ4SC
zO)$C1oWZiRA-cloGOsX;EQIg=9?6rN&y9fD<qo;uAXjk+c&s}f2f8rceE#*Iz1Nk*
zmlHce0#iamy@DjuBKesT@=-mB_hQZB0+m8$L<jxA*7o}4_|q7Tc_|SHZ@A5b6UGmh
zdgRE_NsrPmgt%D*t2DEx%`YKt44-wj9NFv2DTR*=^!I{iXs-Q~B|Vf@`mv08XwO)+
z74o7CD2a@waEW83LX#F4J|&zI4eCGz_S!q5OTfaSM`#3CKV`tqig?vVgCUm4*ogI0
zLlVRoi`;mfM;`^R+Y8XtZ3jg6!}?D|R^zyB@T21-U&xG1ga{V|x7_ejI`9ntGv~iR
zK7rx|<&ij8Bf9KJ?f^y6C0eCs{F8YGA&XFZfM?|ieRP|9L{#X*IP?|oJArJmm0^kP
zenF0+4Z#Et`rFEaU1STJ$MC)pP+oVIx`cgfc*dXP|9Z)QD?0Trwr>jW3xDArE9{H<
zZF%!%kw5wP{uLUG4hIFf@R!^f&j|9iS@b`3;;w<mAegz^c%=zMcz??{e@67`_njWU
z@9dr+of&?>u@w8fC;+QzE)@O$Ql*IhIWl~-xH$cd$PO)6c?$w-vS>q)-Opp+qQs&4
zcf<aw?cO^O{E{VZb-Ir8n#X8*Q4<TAuAMb}P(U_@{ix=g86kA=8%)G<M*S?3txk_A
z!LOs#nkQMr<1De7A$v&Sx)<KWm4fgoX0GzWZ2-Pb?w!(b;wyDsjoZJYd)dMcODP}+
z=s#YkB>n^Z1ByN@EzcNKf#{v5Yg?_yY4)0#2{03aE7$XXLSAxz1w6L$sHsZ!#r>aJ
z$MgnFSCLJ&K0n$P%|W8Cj|p0F3*!1JcP-k+G&XITECL+KHg)cdAicgmBbOFbzm&iz
z^jr6L{Y?&f>h8?#W!{88$|wndM;y=VMnz8qu__*CVR-}IEsG`7G=&uD#=!@1{H8**
zTivgJy66Qj!sc9Xg7GBM!6aCIN|I5a&xsIGSHjWVz*ds6T)*w)b~L6m@&*nx)gRPk
zcGfr?x%Z|)e<IMMF&M}pG`jjF=T_*&xT5}E?5BP<XLw`_EkF%a6SKzi3-wpiIwt8p
z-tiP}BL6oH0-i`(?CW6I;!rFY1gqMBxORVDM{&7?YveRsj8|{v@y0IanGvhcjn08i
zM7M0fWWj}U-JoeFpAFkd%Lou*5^<P{5Uq`eB133I@=a{#zCIk(1iBuVln@DtfWmkv
zC}!;!@f(yQhG}L~AmDHV1bYv(s-ReXjcDjFZt_kPY88$GhuseiBJk=QmCi+;5mIht
za8>j?X?wwke)qgIJy;?6_#ipK#A<t2FR3L<_bCn>E<m$nE_5eMn)A&}d1%RT6H8|@
zLHAS^w5{|2TV=h4`1+~3(s&ejgOq;;2-uM+E5eD18d@WH{~F&%6Ve|2b8nx>K|Yhm
z%yu-8kM8Q9r0T1i33sgE*^F2f7+jQnbg+B#W2n)h4x00&(arLE^url!adS1dza<`_
zDhhgG2@_y*w@41hto2efD(Zhu(uW^=m#D>zA5zlGhz<ruu1+@7q3v|CGQZ!~C~l_<
zCKU4vBGf-b)e;G`Cu*O=x4fge!zA}SjRL6gtM~-b5nM_CS6>XRvwfB>_~C7lac(x^
zYeL~geY^`)L%muP3l<a^wbVIp&6Y&7>t|C_7>1htHh!{hz@CDHLB4l{F!&A5>8;Bq
zBrL{q7ah1Bn*v&Iv!$o$Eh8NMtyGuWSQ53pdt&=lPDSuQ2BP$1AX2+Rv~Bf(GUA#s
za8Md*K0Hjjo#!YIh^QqYWk}%-^C&PTj0o(sw<Y>AR2DBydVYX`#F!f)t<v2%vhV1G
zz|+rB$^)`3Zagn*corb`7(f-mP#uPSpg9;T5E*Y2+IR%|DQE$aLpY1r7C;wKfrS|s
zuu6IlG%mlr-@l=d8IMKdz_=AL96`H;kgChi1yE$+=2ST+U9c+MvoxJFB(qi8Mc<ad
zsD3+&wE>SZf&|$?CdUz9O-LDJIX-^csTMO8&}97u4WkgT>`)DVLHxj9P`fQ-18mZz
z6l1sxM}q9L>c1StsN5~T1DtO`;7G||D<P5=(rE@OiB1O@>B+`K-2sJ?^Rg_h5H>f&
zp37NJgfnj{299MLay6+hLQ1U@`6#o&LLl#J5T96Y5xgJVv+G0UNzlt&UAsekCBbX4
zt9$){(4t`*>N!m-tH5Xf@x-d{EM$F*N)tB4?2$Cx{~4vRO6;pNT%`YKCCEx`3Hn1M
zgahn~n^W8WdYNb%rNipqde@M;by%fB>-=FPp_=Sy+baq32_0YKuobzb{rO+`ua1~{
z&m3sODoo-PaY8u(_t4Uj$%!gvRX83x=cM%qftU8+sYzr`Yc<HIzFq+@&l3*1fO9iE
zQ0eQl68A%SN3um-Hi?;RW73+O=lhmJ2LtgV__!tEvE5Cd``_LG`LjmEmKeLN{mYKo
zfQa-+G%)mh;PvM)j=m&`j8T}o63)QPn9@M`1&NAK#y}*L;Y<0FmUu0^D2LG`!vRaq
zanD!S&?{6N3QRp@p_D&40Q*GC37X~+phcLv=RX=%-eyss<eFpB!f;pe)`J{=LY<fZ
zLHGbh%%RVB?ylex`{t|^_03k@)%Y*M`L`)ylBEXeJBeg>6FPqG=v!tIXy`j(UPg^}
z9$4#G@HRwWjBGX)T<!-~hacgA`&3&gKjmX2TWmZ(%1E2)_QR`QMl|lK3ock0690ir
z`OD;7TgY*42)Ew$pLl`9^44Ij0<8c|U64fZAh~xf23b9vu}m7X*#1WcWM}MzY>G!P
zH1nj<8c|*Vo2PRt0i@k07yQ{oCm1mAif(d<d(4wxpG@~WA)8;jexM}g$IOQZHJck}
z%!tx7;e53;<wZ^{n`eS<vqUUj8O@Bv=+T>>bWop-l(+Kmo@3hYTWJhAL5X2IprJ>`
z9kd>EUM>gOOZlX8LrJswDoC<36jWRpPPuAUWntl?KriiRa29Feg7x}U(ieh6aG<6v
z;SPMdZXtk>Z+c~@pxwGQB53$;X~~6here08MQbhiRsWq*piEO;$*%7gF3OzT8ORhB
zyq9&9-JUkkoEdatN;_4Nh}^6}@qbMt0mAc{I@%6f9*@-B@j}3M+qh82sC+VehWRLs
z#kV&0@*5ui*!G}<w#wQW#TpZ2%CxPZi6Il?c`*)@E=D#RXFXs2Hl8~V-*@n|4sQqn
zIo)z9pK6?_U*zcT%yQtzQ$jjd8#z$k<xHDrE{Hn(4mHyP@~QD<PxbdnZ>Ae}@ifEa
z=<VqNCQFkZ@`xY@E@_G?_Uq56X}AS0_skv<h<a9C`cHjShrc?bUD;&rxm!%&KqIi#
zD63&b@gQUHTYbE*9lLwPhUXiP6P*pt-AJfFUQunkXrRQ73DHS8s3;hd+zWQiBN<An
zvwXcVR=R%6CxcwdS9`p9$?H2`yobJl5=Pr~PbM7E-tPO&u3)=K2TFz&yDZ1l$!u-T
zx+eg2+OMNm7wU9HzY<+9R~Wqn=N4F%0DpHG(%L}do-Uj$+4>61{qhhJ7cab?YsaSN
z$>LCgY0<#QA@Ore&{~dLf}_@cb@};JJnJ0?&#f@6>{tzy7PNFK!dKl-2pLOlZ3;~-
zRG5Q~C)F6JAK({UpJhCtM_?W2yehhU@uDc&heC4_?K?xh-JMO)(u^q%?w2kQSVZms
z#Su<d>H&>595gSOVw6U@vOu{CKzT`WxyrSen2d-O{TGJQ@b4uPmgQeN5B9+%jO&RH
zfZdQIW>Hsp6OV~w9bmhYjvy$N<7=0ws`J*BUH#4rHDI|p_n=%Jo$fG(MrcWA(@<{J
zLi}!K>b=rZH#J=B^4I9;@cPQ|IUrv3NGQdimKW8xcgu8nejMas)ucJ49emj;sXT>+
z4FEYUOur+yCb(<%t`R~vY-)ub9UwNEK9}cNODahNj0QDCvO4s&3i{e{?0Cgm&-5%X
zDTS9)D@5*v29-(1fXa7P^#Unwgln40>cdW`!CcYi<kfKC(H`28IzHzg_?oI@oFus?
zL5TXU(E2YEz>1bAntBITcVo)>--d(nyKI@6rl_H8SNPFpt><sq5CId9))(l~m%9m3
zihBi`U&sYffc0R+5w6T6;}osH3vR71klYZJwViCD%L^j4*c5MHmncA~;Vv-(WC7h|
zPoYGVHGS#({e|kg0A+u}`e5e6YO#dLGkU0Nq!%E-nG=t2Nfg>TrXHtxC}Hw&NEt}B
zJe~noXu9<)fX>=K?y^>CspMU7(Pt+0CT6&8#K?3A3RDH)V?=DeA{Wv?T4%Nz3&j1&
zJ#GqRJXDzPE$L|MC25+5(3IwRo7BTbd2f@~GGhRci`ci7T?rG}Z(vB>gxvg$Fn@Lb
z>j_uqCBheAQc|Oyy}e!n%5zcco=r{BV8U2uUApkHCOn8#Y-!@rQ7S*0k0yJ%^n(3u
zNdJjr%joyUdaAmMz{kC}<++Xape^9K@mh~H64E&5tJ_}2MQ5@6XjU{M8Lad#@WoAd
z#}sUZHe50Qn#4w;CC7%<kYbd`2Vjq`txAydu1(h$s_E`Z6w&?GyYK9wE7_w{J)&1w
zdhu_{w)yO}Mv!4bx*m9@dOEqByF0zRXxrD<D#j|^T7jw>Q}pRudY&_QLQfncVs1mg
zn@9O-vq!il>tR;pB@@d*K73li8DIH@ceo@|R(e$gyQ#cech0tmnB4=y3Q_&FV}R<4
z_`&XRy-kutZ1`2vI21eovoBe1(`$D_Q8pR(htx!z$F_QajZFnYc3~Qi`Ai;q0)r75
zaC*dxW=$wM;zZBiQ2Z`!PS+L;sCf~-!vg>J&=z9O0id2Cn~vmMh!ii4OwP{z;l1gp
z2C#6kK4W#TOK8r@Za6eeLgcaa!8Tumt$~yJz(lj69ERYwZGii4s(1Aka$!}UhWga3
z(#)4XA04PSZGFxoT+vSd|D>#geko};bNTp+r&j`$k_64g*emHb9QDJ~#!`K{8^iC*
zafju9Wt>q0oW58$R|upWCQB_5En!{SxjXVHbs{-YwsRt><!1AfxeO5c;}NQ>B!v`4
zBfI`3A?EdzD;MRyVx%WRUG*(Xk3QmeRJV-*ZNPxtTo>8)s%Uilht|_FGdwFWI6hL`
zVg_xetki&o%ZXlPq!3l?1GJgKxTnr}Pa|Nul+iy3U8iHDn~oZL!Y%Wl>u~!2Q{XSg
zu}2UJslbjJgGA0go=`re=?H~#VS-GKi|pj_JCn}X3+p6tJJYA8ZnLg_8q27o2CNnL
zWjWLdaCw&Q83MRc?QqN4ks9XU?b{agK(iuGr$iRAx4@N_X%UWSe>~bdwI{|>C}9I5
zhZ}G`Er{6JXd>=#O+3DKRV;=rk4SR1fppNrl&NfIh?XSsU~AU-{61idx}!9n_L<XF
z5ZzxvYx0H_6~|!Fhy<v)WXpAr7}wrb>6FiTc3g5|g5)mjPQ94>KKaT2-+~_Dg<!yw
z#Jap;U+2Y$Z3?|2c#O9a6NrRO#gU8ysL2_-Oz<Y_#X#94Pm9EeGiz?H5)wz}n1$i<
zPaI0UcT;b)wN=8RZPMMtT8Z%Y==8!zfu=7%0w@az`Uh%@I%(R@b<4}0onM&Hf|lDT
zUjF_`qr@7&v}Dtli*f`w9<IT-c0q#DfuhX^m5#eyp|}()x+l8G3J7~D?94}dV}2r(
z`4_ak^}Iau^Z&1_Z)(y6K$dOWwr%TcOxw0GZQHhO+tapf+qP}o-G}>d_7_xCR76!~
zL}nI-D(Vr8D2Jb~C+>M4ga<ML6nVr$zj;MiM*-$e79FK2w)Mq<0S3W{jSUN2AO7~w
zp{EXUm`0kOxs7a;?Kjj6s<;j{YIhe7Kk<&~T%A@P>PBCSyKNxmC}TrHNOVytcE|Hm
zZF1=s$*A~3pyV@iz1ZShKsV`rgq<f;J6a#0Z>Z^M>qJp8K9)RnZ|jWVNHaC($sv!v
zN&o)!KO6-h6;~t5VG$WF?*fswF$#MU^UKo>qj&5b8QidT0A@g$zq?=OCYX-)zV_ph
z(7xgI4S{fijYN}N&7<1!gO=k<(vR4l$I`vjZ+d-5t6E3X%qNnT<ei9u#+)&5LF6hZ
zG#on_xRbt2kS<+v<*E4y4xfpPXE#k~;ujI8)oCi_g0%I@_cJaBbQ{tqfRXnU0B*TM
zd5f}_8ir^fKw^f;6Z~Nji1<hueSU>la7h|Tu2OtCedsKS)d?^x*TW@aDF5m$y+;A{
ztzde<zXNB9Cu?<)xIgI=w&mKumC#JnzbMZC9lU9bV<|D#q@o~W6U&n5_15TT6G(7}
zLq3vUAx2k4`3sy(8A-)HRRr;RYQ4TG*u_TpYFdYZMJ<e#uB-F<6Ow9U7tMbSGs*^}
z@1v?9^ubIyXa1qXzXj=>CE6#1YGO!Qu50OIeb0r9zkI?xO-F2G&d8VP>-AOR?ZG=F
zjLb;U>GCS_5M)nU%&3BHqXAtSBlgBxf>~Jd^A1u1!We?{tUflr@5V|kc~#|X{VE+A
z;UW)zf4G!jzC=$JC6bB0%OFqU@<X||rU9*)qUz;Hh&vibJ3uyn7}dM>f5Wei<UA70
zs8;${^to_oVhDt3=rklBI-u!dhS}mnWw$Z`WJgDnzf&ubomCiO)loA^Bjif#WUD#D
zAgN$mV`oTH3e(P*h(@eg##6dSUSQ8Cobel)3BgYR!Wu9(;RJohB?)SK1bB;<x+osU
z4>btRl@ba=JyGVWkDWvA3#Zs-V~x#5NOBuh?QT%H5&>q5Mm>2imx$1%E|D=<l`o(d
z{1e>LHJ@gU5icO<>9qZ5^Fq?z%1MJe%<8%>5=t&kWQRD78b8P=y=6fCAhgC!9;tEQ
zgNZWnMIqVy>VM2HssgRK0^3f`DQ8EW$o{V>fW+Ff86H==u%_=R>^##%4BcI$v_tj(
z%6jLioArG&hK$Edg_ngI2{F~l=70@GO^hRH+P1v^4LCYoapMb=JIzKqf<cA}?TbxB
zgzW$1aQy6dKbs0;REn4w+hH(XloyztK4@#r57ZcIts+8(vlv3_UclLA@TX+ka(&X0
z<y$Polq%Aa8Pd&7!eJiGjhU%gluMbiAXZk5726#Cz?-2UqlEnfS~w%=Jf)b#eveWs
zLknJw(&^!&*5c&Z++`@t=TLb73XPxK;-jE~;zTJj>B!I~`(Tl=!fB7v=mn|XU?JGT
zL3P{DI>_7iIY-Zd9p{A)xFm5EB$k9(*|NR+XtqpwreOy4WbOk&WfAfPJjjI%FWfKa
zCJ0X7c&fw>^QyUGeM}TCmXih*4Mph-Y)zYl>cWtWWy<iJvm@NAcYBf>aB8q8ThbhA
zuYPIeDlCtzjX9$Sf#-PRcsbj5URa|Sp+ojjA-D5`wS|#IR?VK3>w?*y&8R(sca#Ln
z<Z>Dg6UnW??LLJidr1Am0HVX#;zB-H?Ra_HGV*eLxLc0b|I0@U3Ab$9Spkt<Y$%38
z9!2yiB|#2$^6O`2=^Sq^<c(=NIUmKYE6V;zvS<dnjysg%?b&!fmx*(UhU&6@il?ev
zAfu^dATlF2UmTl)Zk{OBgb0p(CwGkB!>j2oMWAwy#C~P6+jVLV3;LO=Vlitl{8__*
za1e|Ta9))<_d_H&w20!tim(9_)M%Z0B$^GbNuC8sy_)a{5VbeK5*fT_tPczr(>2p#
z$f#$>af%1|MUm>m6Q8X;cnjumX6BWHZObe%nwr?yh8v>XF|G&^fIaNAXH(Uhi?z&4
zE!~i8fw`>1kZesb!F<YIzHMS>CB#=EOTH&!K}{+7%gXbsuJgV^*LY)0MC)KY=&JHA
zk4W4Ne~q@`w9VRS^9Oz|_XV63GW}=^?x&d<hG8M*vzMJomfeI0B^#Ki)sb%CKakeh
z9-3iS5Xeme+i%M%o#{%+Q=jb(;a?=}(m=My(Y?af&b5`zAY@h^`VJAjT~z5miFFRq
zzv*}H?+$lD>mf)j)#flUAv$(}o!{Tk^#Y~b<_Kx@qYWN4y(DjGUyq~Vuq+9bJZMHo
zFnw84yla;fPE;FH2V6?_;XG2v643W=$;-}gZ!#6J<=9;XMs0biGZ8S(BW>gQu4%s5
zaS)9%I^3kM67C)z8uNhe;C*82Sl)yc{o$yJbMj?rxkxvN;+9>s?oqHj9yb`#WiijC
zqHKPCr9G|C#J7xd+~;n|KUSVNXC^`Au#)0*LS>Z!mmV2k9#Tnj)GrJ_4BV@N)?k@Z
zhP%OEYc-+9c0Jn3L@hAqoCKx8H4%2)0WIe#$>~*~1~RzvcCk1@KJqs(J!)r^VtG<P
zWN%ZQnWo!bOPCQv#)f=HHFpq)ksClE2NjS@pv&iikmy>UdoDaG)$g`C$-2><u3!D6
z7nip_1YX&_G9U5JyM%(q{FESXc|O6WEXYJvOqa*eqV<>K4ZT8DFRLCLMnOs`(cgR-
z0sZ>gr>;zb-J~M_Uw|B!Zg7ZPyc<cGE_p5SC%rEj!UCSqu9Hk<Y@;rL4KAGrn%%ds
z*^X`*$B*~>81XqQILX=)CkM<2<a_f&L=^^fJ^BkM>5&`B(0M$dP_btZ-sR9uqh=ol
zs>lH;Vtz+5{-0oA$F`W(FNsaC1vIk3?ox#te0sYLpszrcPR7{Fu|$qY7W;+7*m+;l
zU5+%R5`C$Hpqz^7D^4<rEc~R`R^aX}*pARN<*n(aUEL5JEl}H;Ps=N_KQF}m!zPz1
z?Y2X#i!;k74o~Y4TGG}U^2Ht?1;NDuU1~`>+mBBriSn{q@{8=R6>(wo<Tr_;oF0jJ
z-wkU81T*c8s&;D;5=z_#e!KQ5Hy_&&Vtl*<`S%>4=}~J}k+53R>8|DJuo~eShnsTn
zCI0rdIiBPwI7EV{6C1&1P-^^#Kw+Kvg>rlNW*yH6Ksg}RKO5>Q*Zahrjap~uJQyx_
zki>}0Hz|RnXxf!O?LJ;KGthe?b0r3LQyBr6H-W&cF|hX=h6RYrO-AAPEMD$VC3VY|
zdM8t0NGV1a4~1K@e+|UoD>Q3WIFNF3J9<16)w)3T#LeIH9{r<Ihdthr<pulWp0_qs
z=k<M%t1IXts@^@1U)ka%CpgStN8G%PXp%CD1^4pu^mU?<!39VPQQyRhZqFk<-1%jt
z@_@>GT-W0!D%nnJUT%mNnND1jNS~P^N?{fA?@H&7Ad2hnUFi6UExMJ%4Cn>n#s2eO
z-sF~t+iZrm=pg4mO^p6*AMe;+2hAWB5TS)ao4{oJJ}VK-$Yd;G96v{zL2Qd=1fFT?
zP>C%V4kfkt1B+7F&(2)0y?I#)_7do>1;1pd%$*E`=ZIxBSk&xH(bWM7rT5i024bKS
zWhL_c<tR)4CZe+{%-3p%gm=`ZlO=$hAQ{ET8<TzMUgZy;XNu_so<V#`@XuBPlPIfR
zD9p7Fp)9&vnqr=x5FyFZYe!l8-nrt}F2+iFJK4<=u*~yK8mW=8p<F?y=arO~=Lhv^
zg;!!59{>50WZDUogig}6dFDAper!Fm*c+czhU!6GEC-?LzHP3yO)D&;iBSw8e7csB
zvju3csz9DkWC5OUu>xx{LQ3B)X&@&xyIp^E9acGqjUn<pX5R0`lFWG<LwV$FsA-=G
z!sg1)*D1vA^DAyz$$dTLt@n(gd-fN4>9!6~_>`HFu(4QWdB897^db{c>rSW06ip0j
zq&AEO_%~exV6V>nmeGG7_x^R~cr7NOW9PCt%+DaObnbR(l{Kv4dj6*@yd8BTe+~%a
zpzSX9BaeZ?>~^e#y1IpkNBab5+ML2&p<rB@q~Lbc(f4-y5e*mjHnbU@2i}F*bHzx3
zHdN3buJ157?Rq%G_Z1Pdd!(Bu7QWf7XE*uY*y0MLZ)<*}_2RhXgMz5B^mmPN@j}jM
z2B~ArnhK)sQB6Wl`>(vC`LDj#Qa*8qFtEt(XrV^xi~hgRh%6n?EWMAfMpab=CyUNV
zYIS=?9{XdxWbyec3^l*#RvJ{KLV~n@<t_&~*Bw!#G{u!n!dqG8xO~2J=yNKI*n)GH
zAs!eGBL%2sTjw=Bj6N*<vRam{FCV~C1nGE%zk`Efa3>ekv}HJJaprWi{?wL;4{bCU
zAXYlWar<|r&~yntD}FcBE0S72dzXS!3byfof%55I)zlyp(6X2n39@F~JAlk54UB?S
z!dp{ApC26*H_LpJLGAw#;>eMHnH7OTa==pgfO}L_7R=Nq7FWr}y3@7}#q7M4<bTs)
z{nTPcIO_b&>=csm989c(o99x-C4vA1mCns=J9eaHMXIOZdWY?#%pZj~)19DZ#g(p!
zm{bel8>E)&cbzP9c*>#L*>dW=mCpZU&gYXvq6Z#H1XD=|g%f4uw<$~js5+cjo>UOj
zb8Ch|TK87*G1rHE((8N-fe)4^avuaFDspl!m7`^zDCV5wma?n7rJ3+fWbh-R;Y^uN
z@Z0{g>%U1oc*M?kLp_;l8gq+W$cxQAAxBu0%%ahqYh%o;T4rqtH|7cnCzPqabQ5m^
zU^q>LT{y9)jKEqn70*~Rrd6Xe*qo(d?Vep?BX0Q2dMZP9cJxJVa&k#`zb~nYcwmu<
zUglU4V0Xe@@V2YlmVnt2u1d*lWrx$1uOIXy>CBwV2^Dzt$Aj4XeDCz{=!T1B<D1uy
z=f=-8OgvJ2O_c0+0y&djn{d(nAC;fv|J!@S3Lp&*ykq3c2t88Wqtu^)(W2(3oSh;Y
z9Hjiz-5#f4`TbY68J4A=5!8uv)Z~?C<LbRL@!HNSh7h)N29)(}`V?>vC|}?m*+r*Y
z43!w#eg{*fdBEnzHMVMxNVkY2NrwfayJfzv6vfbF;F!yoa@V^Syq>DQ?<gjSsjd=G
z%wnO`S)d7%pXR^@nz&wwnLmR`s86Y%BBv_yqz{B6^-&ZPSt&=(GPDEfYLr@HqF5yg
z5;BOk1wP@$r6oKpPd_dsw8qyS5|t-hRf+~57-dV$WUwUZ)h+Uq=0AU!`91I5>3WX&
zhr2caUD>R~MgX~tmT}93ptCy>4rwVX=GhAldTK*IfWmFo_;@F?1W|H57XD>=eNfq-
zA4X)KZ|_UH2)Xq>TkV;<)SQT;BTWk%779+M7bQE}nJ#cG%vvhzVsOh?&h?D2ez+E%
zl;&SO%bm-N60+Q25Z4g%^-9b#lECn7PrwgCb8PX-%j6p4qh7o_-e&!cw;1!{JEbvA
zU}{|h6xpKoNLFv%f*NaKajJbsT>m~37v;e6lbxdWf`fO5{chV$8X2n9-@heQ*)pVb
z6Ti?HNkr}p&E(!0U>J~4H_*K0NG5xRYVZ<lWVGSVO`|Dy*F6<wUp{)D0DScN(|ecC
zGu%u+y`^ip5#g!isI{Xa$~b%)%woX;R}-kz8XHgqL(M8LEUNQ1M4$xs!vC+ppdoXY
zICw4qXp`rn2-9drJ6@4x7jQa?bnsoi*fRL~Hhy#Jj*N*8ay4qU=%!)n{O*1oX&=z3
zA%i5B{_vl%YPObz{{8Z<ms@}C?#j_of2x;Lr0*`T8O{_CJQ!=eXz2{EXfE6k6|r#0
z=}?fpNxb*Ud3r=E8H41D^)Q!6M~Pjf+|Q+1<OGBdDDCzprya_jW^?%DJ9l?0`7j?B
zFE}&TZ(>bVIGgcO83vh%jiv4~b^|HRe1dD)l4~alj*{;;beU~^SuSi`3I3^;!@|NX
z%!A@QDUMjLwj>m@C33}BAAdyJ<z!%gl!~*=m<uAtmoCLgXccUnl5ch-oZ#K!8~wiA
z&7KoC5=HAO1*<is)6kCi9Z#Rn9Mtvfjb&knoTJ0z+nd)W6=nbFqexTWBUI=PwaA88
zzLyV?xwM!)vuExsNB<wBevR8Q_g-Ibso+*LC=j{0-+89Z4NoB$1^HV9FAccJTLCUF
z9Haf#_>g#cP(Md#vzpAEV)^`%)=N{|H*CIWj_)h{GP>BPPwJ#3h_KJJMf?@E-}CX}
zd1@H&(#3zCZV#^CGZrtGP|D>??n(CpVx&m!SKP@lKmLf~jgzC<r-bJrw}^2z7GHA5
zH#s}?{5z8B2fXtKP`|qr?X|)v%NJRRk+5rWKvvlLwJnmQVR41v-PaATC0}20VM6D5
z&ualQK|egqhDx>G6y4RlSX8}xyUUISpH$tMd%KK|!K9M2^`$7AZ;>iY;p0AFI;H>}
zO6Zs_Za{85SvXebk0C(F2G8F3i0ngzIRI-iG@YX!Mm|!mJ3#X<h#Mq&?RB;I&@-7+
zR(-=)*~UehUh^l#B@g%+bUk4q@UXM)fC1BbbABu8Fd&0Oe?lh6!ACy;D1*?syoW(^
z>V@{*T}@bhkT4r3(s(?sN+UV;m53b$y!MBn3R^GP2dl*(z(ctPAEd5Hm6KU9Tzgb7
z`rPkr6HMh?@8d%M#op#LJ{XLg$)AnAh3`k+l*(S2E^os1Ld(7iX(S`VS5hw(kGNRM
zdtrl6SjIIGvt?2V@6|D2Vb9N2dtt#H;<AtzFzx1tsYfJu0eW!P&^RE4ZrtEY9z8Ab
z^_3z4X!%_8i32iW*UCB)ZV#yNh#C?=8d2vBj_L{JgdH+L3=8&W9inpXEiYf7Tq05^
zOgxVO^@TbYRq-_K<c~4O#AIEdRAPwV7k)Tim{AwBZ?XC?I2^UFXXiym=EuP+=v`G1
zU}!7q=$)OeJ&OEGKKQ!Y5#CjI(`6ycOM#Nzg53-@qmL`s_ZleXQmbtqJ2lWIIaa>^
z5NqFUVrnRXTp=>GbxURyGHUfdR!-PYIi)8lMy=z>?>~wPtOs1hF}yQ0Q!mt~WphR$
z`Qv>X(5NEN`A2<xTsJ;`@jLmPWoxVTC6IrfltlA-J$d(pI(d$pK8_x()myw^g462t
zr<WKrp==V-In^1_@aS<)A%~G;?p8*t;;exvuKw*ISLdLeDwnDx4P4m8NAjRH?M~g)
zeaOOgDnA>e#6y&y?OmKlt~~d$GL$+;*MkziRY@9<%j0x2W&EII^WI|=-?vM%ssyfd
zEOzp-%ig}H6U2S@+4k#ve2~eim}X4xtse=-*<R;>e_wPNC3=TcuMHS=Z?M3>9JrG&
zy7&o!@M_n-cak!cQm6x<El+L(CT;{+k`xp+8b6M<U?U6Eg*}Rcton;OdoVu{mN_f>
zZ>SM4syN#fEdtLkC5K6&hGzy9G++M9fimmA4Hg2ZOE#iWy9yAxKv5&nXeIF*)bxUL
z0p1V-gQtlNk<<&Kx(S5sSzT3LYShSL%vIWpdF)1c`ri*4p~?<UIMHCu>|4!b5QSgJ
zeq=K;A$A)ZZM9K~1Tgk_cR_7dTV4MXR;}=gWmIZjGIFYQ+20e#Kmo?K4uSx*wYt%~
zf4p`W<8TgtYyzz&;IrUJaQqpuXVXC8(Mr@0vOgL~?|kt*?{q-0n6HCS%yKXp<K~ch
z%c_3U?>c9x*?1+n^ik$_R@3v=c$-lewhgpBl8kn=lOfbHIvJt4ho>s4zfIz6hze-)
zKHC+TB2b1Poi`6DiFg9mH+&>crcH-8TBb*@?J^?PzNioM3$8-Zjj+Z#kjCGoFGDX4
z&Q{|iK^=dG=||N$Ek%Nw%8yvD;O-7OAc61bu28KNld=VEa>I!OQv(z;)~CU&84I&v
zVw-m1!r_(pUmmF(ZT3wBT;tE0l>uv9bQ-cQ(YnJ!B_w9zi+)yMq`%8vj7g8nX?<SI
zWsA_U;ICRCPKb%pW3&*3!7w@P&D%3kN4Y)b0C@9}rgDH2XM9wPQB1GE_-s&&I6*GP
zIg?Iezs<zwrBVsp%r_Vad;~XaaM3?8j(|*Y8kSOo9bBmq7G9Rt=C*6ug=E>_vCn<b
z@BKb#zi5lbq5grWv3>N@sc6)!sjIS4;ZJSEU5Y7Jj*osK**uzPD$(+82X6!uK?N_r
z@Cqr9WQ~OzmcX|lEyD`W$dtu}EqWjT$;3uy?sPaOsQYefCDVf)jT}aEm|L`|e;m7+
zQBaJUOD?j!a`Mq7EFte<emu=mIa~r1Jq|Pvi^m})PZtHk2jf#0eE4i8aBN18IBfj3
zFEH2nYwr>WAN5fT43g)Rx!Ng2hw`*Npi){ff@@jlnYox%vGtGrvj4!9og6!T^$Wm|
z0I3`p4P6w_Gqzz>R3a$s^=+feFEr}dbs&4Uy7rhYIr?#=Rp<7djt3!i7G)dX^5JMy
zQM&?A13Ia$7kzsia&Qsu&+B#>{1e~bHd?Sf2z}|FX3$&h??u>9lic<k$2y_AX&dx)
zo?@<RL@$@1q|qiD^R=aj?*hAWn{x;rn6>GQnZ#QGW$d%lNyrrLp?O)2YK<nh8X<Ks
z-Yzx^w$r@TS4jqWOU)s#s6pA>gE`k^D<h*kl8dC9y=?Qp<9ENT5~mm3x~Vj7$5?8l
zjT~1njz)rpHTpgbc`W`Q)7!Rso!E^#0ZV9lxB%Iah{7QQpw&FzeD3ru^B%xdpZ_xT
z!gibV7`av@ZJiMqi6|$H%C0)zufKoXZp^b~2qYUnjc(>b_9lVxDQB&CdGe(Y6MtAO
z-`yv1<%u+4OeUTVF>9=_5As5BNoV55Y7egzja#)eGVA@5ZgN>1X$XTN^9Gunh!zy;
z=fodBkU-SdJ{=$a!yI3rw<Qiv9r#UM*HcQU4E6o-7$T)t^w-)iMJ|VTSlA%umIwY$
zSu6bHm-U0(WtYnfZ@YdsrW8_8RppzhaiWv&;ih@G7E8?#A2$gCy(~J=N&;;Js%d)_
z3O9mC_}8|WH@1kS9^TPqq2tqnMJAN#Mw<yCC*BPreQm1<U!Lq^8<bTBoT0axWNG}P
z9HOw?J)f4(C7Qv9u|3xcv=d@KvEl`o;u*ekFDm{g&3h;{<1<=eRHYX*fCz5f6CN5t
zpR>8MiZ|x1wDFHwUNR|iJ#d^6(rLF4;@S*}VK_hb#gZRaxi-YGNl;9%>%;}IS1~Hk
z%$iUPSrkg2Fw;BS2V12e6YGqxKeGF$D6i>Mq7CQ~Gh^#*{UH|KStS7y+d9lN2x^w2
zDwYH|{%Vf{E~c7et_4e&Z1iAoTI1YZnJ=0yFNt|z`NEHr(l!Nl<t7RJNz{0Po;et{
zA$zJ!eiXet)M;^61bR^Ux`gL<Yo365&?ZgsFCb^h_aL|66Nhz#{3zUMHX|L+fN$~z
zy0q;v!r5^Oc!2)Ff&yseMhr<hn)n7{a22Cl)pk4ea+|~>S}ULPz`!5hfakXLc4-Nf
zcrh0GgAc}})WyG~D-)e@N1Kt|Uo<(!Tt;)d&ZE@AATo=K&T{8}AA$4K<TAnY`qD;m
z-(l%#30b@QR*YTN619WEXMqySS6}IP<(5YZ?fv~T9Nw5anhA_DCW6gwQBNN_*k-IR
zJM9s1Q9#v2h!K?b_(RDH#yo7%cuy^IJm-1YU_bnXvj7Wlz)Qcypqdo11bdlVVNzE8
zW1g2baAO<idRnC(?eK{dMbs<&5cOeidQ8n@prs{lCHR&w)WV)(r&~mZavEJuMNbo_
zNY*_;39b@s%yY)n1Bb~DYz;MF@M|`pt?+HbI6$HHp;Qk6p=vR6dIcz(YD_Yj70cX2
zJdRf7$3VU_V>?%AgVpj5$*-(0jgU&ap!m_xrjoI54EHs}+Al~G<g{#Oc8<==DuBO+
z0BZ^}6MYDLouKDIMxxC&K{aIX(Ldb`8>@ccMzd;Jm@U4#g>XrKQ&iKUhMYR|XT>e9
z8pgmAA0TDXKt}9(s&eS=JmK-yx)W??ZS@zqZ&03;kkf4LRf6RwiAsCniLt7bedY)z
zx;2r)0gz{EP*Flfj6{zSA;q66dr$$s6OEjnx1RYs;X1C9)K1CY=k4t(W8Xiyp>aBC
zu~sRYKDqdrMfAoZ+wYg&l0`;Qa7I=>yhylf_gag|b;@#od#%c?2Q7uYBd+_*Kw$V|
zBcW#WX1Ds_>gnB6yXE-(i8<Kn5q-K3pV^8A2jEjBZ9!ia5(;VQV!cnYdSp4zn(dxg
zXG20~{$X2xF-+|Sm2ji3u|70^SqZsSN9|6mWpz^z6h~eOFhpD5DwA;pTTmmp80J(I
z4B<2Zi6P+;Od+{~H6NpRHtrfdLCmQW3^yu*w$*X<Y`av7%LtV@@u%GonUoD8g#Gg6
zPsSQS+51<|R0<#su7c^&hO!xV<mO8J_?J32nt15Didzw~Z|2T4)rVpE_v+D72mG;v
zJd<ScJZe9a6up_=&;$-D%azuK09aAXVJ9gj>X7O07h-yejFj*Xn|7EIGB~U0TkAkY
z2O{EqQtd7nq9Wm92VxK20O|X>bvSvj1`d(I!?}lJ$bsm13=P8cfLrS|^GlWB6I>Ph
z`!^00*!>pfjgeC8Q?rf?5A*9a52pDRuWll`JKIF)!bC?&G+#SNoFEcVJb<6?F(N`K
z`6C0gM%#OX&e<3eRl;Qu^iC1kfoQAIi`Lv!;jqe-#g1g2haqZIScm!e6Qi&0(7c7+
z1DAk5*Io+ixBk^ov18z|q6CD>CuUJC0k=Kl<phZcS9MtIZv8gtEcB);?`}~p=#Kt&
zBG=TIkyi;#xZf^G(U9%z3wM7DhFh32{GwZ}RyPDvP&1vmy2s;Fi)3GjIMYgSo()Uk
z_B?)tjNx*P*Yc{sW`@J&l@LcQ**dD<1M5ejDM@Bs3}RNvgjjBjFK!7nDl90tj3%Xn
z_s2*SR9@1+E-!k3^QX$Dl`Yo4&EG4l!Y3MFb5vZDo;vnjflNyuWu*q1TRZ_Q89z9s
zc=F3v)cp|~udRx9N?cSgNnx_Lb!=Y=m)>ms-l&2LLd7C&;B(P#-`S87=t^MOz}i9c
zvlN3APa=bzl>V`W0#$&A#I{m?+!AKqFY*cw99M{*E<?%)h;{d6xmFJ5TMu)I8R|u-
zFRx8mx5>(^9ktB(O|i~iOE#|bK8{I@i~|1kvxse@H}UAtb^Uu>EPFe<=Zam(6`2={
zz^HE`iZ;C)l0(<%0u|F_@HG0ejK-QP=(%ckNU9K+Y1+|~%I%<f3Yrv@6KN&^wGEk=
z85^A5JqCF?no;z3<D{>J&Ubv&x+hzi!DG4WLQgGV$=BY1Ejwzq-2kRBS3*7X7MA|H
zgemg?Gs^ohL1|0ZoSyuA91dd6j_l7{>nZoZvsjv^Lpvehk%KIRRnm8tTBI-Sa~rcI
zV%}3OFoO__hhcTHWuQ3~S$_BCmtZUH-=7opTX<zPXeiAD5|yl|jCc*cHkHR>+IOB3
zf>#+57cm>vwKo<<94XG^P1uTALuq7WQY{NU{<EZJv{yAl(LPQc;nD9eHNnmL^j7X8
z)GCfZD<;vY_~2<DjEcMq5pvCFLG0q0<DKV4TheLwY2T-@4VSqzw05!U1YwCjqkXB@
zU7`_nSk6qhJvDGg0@oMT@qnNa8}dE{!4j$yGjvIIE#yQ7XBH+>huDjZ9Yd_Kq$TQX
z@qhz8RIqhniS8yhTchxwBU=fk2})@Pp#zH8^atE;Bxl>6w7@y_LaNt~=u;%F%=O$7
zN|=BrJtkYN`f;i;-YR%W>I^8iL+nsv_xZ*n4WTYdOAxj`@feEJ@xLnUo=OzS9lT4z
zQHwmF{I#P&o*e^4&BdrEM@o2#9jhCIQCa77Kim>?`w{C`>*_jvf78iI9@n3q3}E4k
zt;w0HB0^~d60pU%5%3pV`N`emx`$rya)G;PgBr@uNDZZ)es^X11o{>20K09g(QoYR
zsEDLpf@y`yg43`jwul^_ph_y_Lon{j;~x<%u&F8dg*87I9lIc$HtU`KMj2L)v{L5M
z@PR$!@v5I3nql~NE14eWZ7^GPASRP^UA2aE#6&JH>@__-AfHWThn>vBx8He!57z-1
zqa>>tcigLVrS~Kj!P0?m(`xZO&L`?I<G|@&8!)3Iu%Iq#Jd7eciV)jngKbfKi?Uux
z6d2u(o@n-*O@u&@K)=rrXkCX&m!Y*>3X|sA(Uj@7%(N>#>aiX@)GNMAZS%P20p<Ce
zVe?3A-kwgvswazF5C{MS)!2|MajyGb<}IzC!pdd_$9o?bm@S`m6#L)s<onOU@Nl-7
z(SvrFY<Jd`f~i^)N3Lz;!6oeS=b8;2Q+Ee1OgvUB_k-r-fIJCA#@Rr=IXR;a?e}aJ
z29Tv1#Tvzr=57hVkkIn;5WdxX)QI-yY<WWTwNi}^l0|Vvr*|ILfkw@zLIf(XAm+k8
z4&DB3nx8mSMP9)9XO;5Gl++K^Xu^c&h(8+JWyCGJ*wK=SR2RbEYERuy6gXACgJJ*F
znW+nj0(0wah6E?GQ)lygLyZO(Gz4czDuvC-ssq1YVY=6!3P^Rm7Ax8_QDP@dg&%X_
zNsCi_fF((;A^5e<uZHRlv|FcXvXe&S=ThxrGvnwgOWp2!ct$POxAqZj(zL?ZsXPTR
z<lKva+Dl}U&yps>UZ2{)x78oXCNy5n!Y2}r!MY54_c8=0>k`$z06FMpzdwKIxWQ+Z
z&!{s=?Q_8}>L?~K_FHC{E+s2%f>SdZvn{K?c;k1(+qkj3Dp~Z-x-XirIQl~}3TKmN
zdDY+EGutGyUQT$3k|8jj(bpl|*~VFihs9$Ul=u}V1wyt`f&PTvm#?0~5iil4pDOQ<
zZPx2dD-rjqdJ1=8+Y{j?{1hX}{57C=R!MRQ{ToWZ`5?Q#E=?$tjF?pr8K?bUUmJw<
z3w@AJRxkyveyc0w1*m-UbrQtS-^r#sNdh2=ielG2p*x4A30}|48ue88^{x34Z=PQx
zevD>L8k89T?E&p1kVzf}TFRo4Qk(e+`u=DBq*~{WhOW(u*pr(gb=P01xEglNb`aU^
z4r5FVXqNf{Gw-W*7jI&Sa3r*=KN2gfiWSWQ6}<Cy;JtXpuH<N99*nOax2_&)M-emT
zcd@pjP)hT$IUFP|#rTpTpXLFK&wMgD>mpaRVmr!9ng2jXsa(r&yRCB<72Pb6Y3nJH
z7eM2sR_4;Q>8(r)5bGnACilsKQ@5s>J@@LBqvU}-M$u4Q;9)iWenqSX;wOre<e|`;
zB^~jL2l+N1avfo=wn-4)yVMt6K>Je3HA@`(eq&E=O9Vve!%`E^*b55H>X!t~s}D%a
zE6V2@xS}%hY~aZ=D#_f06u}lU1{Bl`|C%E*X4~nt<okI&X4gq522F`?x0#Cm4HTZA
zk{`In_%0TjhQ8-%V5=w8dBVUtl#{H6s@X8)e#*dDY9%awo8f4DB5&?Q^0T}1rEs@a
z7XP@u^aoL=n4KO&X2w=Zmwmvi#1ufJ&FgP_pd!0ltZhJkYa^I2Thg<p0Fb}fB&=s;
zC6=d&gQ#wy>Pv<`cxc|0qsD#gXKUypd&>NoXm%?nJdbQWTrfXo&ZNBrQg##kmW$-o
zO9yCzzX)Cen-8v%$ckWD`T~qzdVX2r1C{3Yfqg^4YVrp#FHxB!V{Z}io<+X3<M(Z>
zoG(75M=psR0N$S>XQH;W@uz@-qST<4kY*_SF??PC;%*~?2d`UR7sLElTKM7`iY+5I
zghSBA)o>8u!OJUhK0zA{-e@7y-rrRRNqJ*6NgHk3px%ze&WpOeZO9UoDOm3u5ZvCb
znn<%OxkrmFg7AlkTN2;uKx~1c)W3$Q!ovA=)%%qzCy1D1R{uykQ!YapNWK|GUO8Am
z{;jBmp)x90HWlF^`B*N3G$UkV8)YrCvI<HiPJh3y&fl9vH2MtQIwGxPqV~WTDa0sn
zTn>+wJyrcbIn?kl3T19&b98cLVQmU!Ze(v_Y6>zjGB_YGAa7!73OqatFHB`_XLM*W
zAU88H3NK7$ZfA68G9WWEGcYg;FHB`_XLM*YATSCqOl59obZ8(mGdMFdARr(hAPO%=
zX>4?5av(28Y+-a|L}g=dWMv9IJ_>Vma%Ev{3V7OlxCKxp&9W_wyA1BIad+3j-CYJ}
z0~>d@!QE|eclTj%cXwxScenSQd*c50zvo4~=!lNWmANuIS65X;6(y;%8l#Ary$Mj#
z-p+-Qm5GH9pl0vtWC|3sHv=j=*)z(x0RBM{cmZZ0Qx|{<&;n$K!2GXZMPpkafLT&h
zLq%PN{{P~b0nPs%CSzxA58(MX+sxJBzcaW2ot#1Tb^w}xlG6g5fW~I_b~c{>rYYLH
zfJ}h^ntvw+fSduw02e1?GoY=plQqEJ9PnQs3{dlQb^+Qt17z$>?VTL#os3<8W&n^K
zz{L^>(69sj%Tjas2mFU_XU1&r1hD_d;{-6ZceQhI0s)<w{<E%_y@RI{$imVEK=VIC
zHWpSEhJU@;{vTzT03tRv0M&mHoB^spXP}eYKj44nmo&Bo*?9h!ZvWlezZ3)fBiVn3
zf}ABm9zZi?kc+A1KfSvG{|(Ey7~6nMMeHnWfB;5T)_;Q<c4j~)8;~7P+1?rC@-I;t
zSy@>A!>4WuGPSk?Iy(b6|Lp?Wnf*siiGLb01KC;pOEo()V<)r!5BM*)vN6ccMcvZ@
z2w?f2zx!V{>;Kpa#{Wp`0nlS%VqszZC-9%=zeoE2QBKXp$=({M1u}E7{QqYZ6}9&O
zcr&u`umKo3I5_}3tQ-JV7A_Xw|4lb_b#emQx%}6(`lpruw#`BRC<6p~08J5Am+VdX
zgRRmtf4S7*<PJ@UYqya?OR&2Sc8pksJDz~(w1Z3xmZ-^O5$-G*;}J!3?(vGL+`L}p
zd^XyDiSL=n#fOk0(ANGnMc^*?<OEcaqUC&7?m&34bUY|aAZc~C{1m>19eg~6)nqYZ
zY>0tqV8D7(>A;>QlIjt#G?GN<zPSmz*$&_=1qjEd$t7tQ!$EXh3~ULWce7~c3>W(q
z+5ChA=jfLLRCe*GmBW5JhREvo%HNNmdWU&2V)el7o)E>>z>n>o=Eg%XywEU=<`d{d
zHoJMPID|>&u19STcq%M*t-lpkl&h)5IHzFGR!L_Iq#8hJX6Uic1~(vvZ{AW(O35)N
z5f43cP-EX0tZImdi-;PE9Zv)AYiOt~n|QK?J8I*|nb66#jL~?ZajC%}-=TF|1Q6ri
zv*=^U<^EjgDzrHsGo3Vu5mUlh%jm6PUScDXi4^&s!Ca+L#{`mT73&2MK4BIiJnvuD
z1)$K$*s*@JfJlH<aGg2VNp5cut0PlcETqtA^pu}0t<$`E3#H#x?&5I-EEz!>qO4z<
zH_>$b6b;u+8-+)pXBj%-0DdO+3;*Jh`R=?53El|9SDjKX>p$Q{iW$9Ew_h5_p`7`T
zsYJn=UxFb6zeW8XD#}h}B$vj5)UB}?u%%nLxR?#mrrk(5x=>}n<c7;~z>T<BCtRCP
zYf%SgOP^FreR_zCL=AEvmnQ7Qubk@0H>P^799HjXsKT|9%JqF~b+Hq#A(3PC!_tZ}
zC~g>@HMH7L;r&9<n)H!MV&#O7(t|oW9OaY`lkfa_3`)!I-YhgyG3Ptew3g5420Nfz
ze#n#@zVNlyMQ1RSHL}~|kxxlQF&J77Wb3Ca6(f}dm2)%mKL7YcJ*>?4BqGgSz8Qgw
z&7!;iDAlEf9NCW1drRVa;0iX}a}Naq98Np=;vc-JN8^F)_n1Qyh=igW8FQ}5*A3dJ
zIwyhYfktAs^49yc+POat+yKh`h|7#@?h<YIZ+MVP#lucy+-{sjVhqZA__{x6{)`(7
z0-{;oZ=1T^QkR+<^|<km%DE@jWY{>~l6Eyvp652;bQ!0V<s&fD=s1AXI^Ibx90?Nm
zcxKV6Py!Z|b(rTQDQc`LmV*8kKEJPrm9ya5gTfqYymnB!_3ZtMSXgkgGMiYn4-7CB
zss{+7FrtubE>I4`T>=iCG^%&n^YVT5J)afh9*Y5UK|~#j^<&^au`2Su=2W~VttLw|
zK3!&+yVb2VWz->U{%$or1X!l%fxZ!KSBYhL)+-o0blI?=ugArEY*z33HKvUN_*`IQ
z_Z)?;Lm?^_T*DcD(-s#d(E3z(8B^rA{UD_S2wTvw9w}w`WZT$<R%8pnt)?E1vCEM0
zuv#%zk5(TAo6LSecIe$Zy=)r1%rrT(Kdfo&7gdQ-xHiHW(Vg6wF2U$uCEr?kEFLw3
z+%4Uu8W)%-I3s_&OXJmg_BTf%UhZp;_-dx8g_0EVDw{W}Zw<ueev+b<e`xBt@EF_}
z1c-!Q{6)u!OdY7aGFZ5ElGZM*m*@&f`&&NwY@%I=ibg<4W+a$lysJ)SK?rHWR0OA0
zNT^!&E1k(v@QOm^-Ihj!^mxYaz6mQ;Mf(>EH5S&pNfDRwdB&|&L5hS$bahO97BTnn
z!a#-J!psun@=7o$k~4=KF+^7<dg7`F>|Iig{?#?;?RAv;nm7fywXbsPvvl!l4uck~
zq~i%D>iFToFZWhK5)}tR>u3(mV$xR96Ou{%0=A}ySB(XF@0a4q&ERD~#k;HYYLBit
z)Rrhy1-JW!5F=2Yx-}@<Oez_Kj=(7F=W`&?tZElZVg@Jjjq|cI4C|}^*xF4gh8mLD
zjh9jvDgKwkZFldyVxyGzbc&*8GDhMmzkQo*Q#kj^sEm)tU&w|(!F;yTj}Gzmy8Me^
z@p>3a9^Q{?zrcLibaB#-QS*QGFH1_4Gn)|##X*1Rhs&dUo3~^d3|6&ZwOqaQI|bWG
z_^Ir&s0BNHDQ@#Iv$fP^p^iD+05PjPh_Fv~U+uMCq3Epp>0tm~_!e2vRKCZ-Evf{P
zB&QFez9VTxov|`}7x0y4=NDNd3w5`lH=Ac?&7<Z9-`IJYxrjsp?7qGkxmi(M59g^g
z|M_%W`b{Id-s(>FB7JmfwOi%H8$g?MZ~(IWb}f?_n|=j8UZQ9x_PdvDi~Vf6KbH&!
z2n+ZMVtU~h$NG4;^j9Q~;JKxv-1X1&kM7tC@X0f+tuKOlW5Uh~69-SmVY&TO{lqm+
z>nm->g^u8u5z`{+-E(3A^W^BxZ%2eE7F(l9GH^BMg;s|o54H!$T3!T5ZQ>wH68-U3
z=MLB&heV|5-QU6mCp#4Q69ODLZ{r3gfsd$bLP=%5N;VZMS}O-Oh28`ws4Tn|m_aZs
zxT&G%Hr<Il)@J9=H*rqu;2OO3TmULw)i!s2K8KQO&H%S8Viq|oAw`d5uXMs*O5Y|~
zZ0|t@@CY@lNj1m)JGWmMpfYNzAFHsGD=f3W6MFun=YLn^;S7V5`6wz_UZ(s`Ff)hA
zekH|-Y$^PzQT|Rco1-cHcgR%BmS*FJ({da28h5>;lP9gaQ=_z|tC=)LRZ`<%V&-%+
z;E33X)Uzl{$e4GO@NEYc)L8!X{4>dTzVYx6{HliP+T@0-SkT}x5CMy=P00e9l8j_9
zZgQ{8TCNQEH$df3I_l%6E}B{{V&ky_S0(<g+Ce`d?ktqgZh}cib?3G(OA&YUX=_xb
z%J2Ce^CH>9oWI0`s84+)R3k=N#JFv9yjKfbr0(KE0X>_BBz3W@V~j1gW%%tkLCtCO
zd-WotPSkrieY065_OUgCee`+uo@$)>?&G~Pgh3-qM~=l%yl^)dfV}KwlKGZTd5Kse
zS^VwHx?sW8umFVb#-3{N2-He9d&v(^q_IDK`&gQuj1iVI>Y_2vuk1UF86`aC&k=&J
zIU*C@>>CH$Z`wDQHl%#MQQ*D{^?t6=hP)q9)(&+kcl&Vwt-X|qnk?li={I1s4V3$%
zdPPXHwPUBjdV^eq{PZx3#IGE6g*f_J^;Vi%g}aOB9a$?;2?ac60;L_gTcTxS%+1_S
zm#N(VI;8|)MJ2np=k;(#!_HRV%;)$DCKI6pyI>n4WmF*KDm9jwr6tZX@myOG#G?|K
z^Ut3X*$i==k!3z$nRw4R-+h)p+Jd(0OI<ZsUr^FEh!!`v0kW}pr)fv;k4fC464>r+
zj8;DeWC$9c{HxaLzRDa_+<aeiCu<^N<6ECB4f>fBQY@t{7NB%FKd%^WO7}I?Zz@5d
zHRY0fs20GkCa4laswkgBr)SU&hG#)u7jlKH`o81?^dark{Ueh_?z|{T=Yy1!Q8mMy
zAGoX2k*4%=O6mqhr;vaiy@$|OpObHI+1E*k>Gb!qKjb9VJkxJ0t3bdQj@6cd7mr9|
z6dso7iVaB)-hJ9a(vqpnn>~W_Th*!YPWnqq*v-l)54a|RavZ%VGy$>EdCP7zNXabg
z7@43Y;5i3*aw2-EugbsGRhLn^L8fI;=5r{An5qi)@Rw;Y?p`*+9p!2OyUea}1I)}-
zRY8ZedTy<h<KZtNA^FBzNV@O>Tn9&~6E1z&mBo%8OeX7O=u&fv7k>U7xwi*5uB18H
z#B51sEg!5}Yz3Jr*3MAL{n-&Y@A)#2K^h7|DsSQeAy;`xB6-8vq)?p&UdVLa_C0t;
zUP77m%y#scl;@#6HG1PsmcbO?%sgANfoq&K&1hQr=>;7w!e^6rSlLZZd+_^lD~ufv
zji0?wz_ja$wL3u`pT5l&!4+h$&ky?Oj&<`XTzs&GMVP(G3URtGrM-8O*xJgZ*|mU&
zZx821P8A0m{__b=+0}jq*8JW<&2|{$KDyREj;xU#L2Y{}YXSOg9KVC1v&k8p#|#=U
z)9)7j+?BYLY23)xYc#G!nr6NztoHHi?Z)sq<N-N>u`5<0IX}M5Jb=G8Q#Oeus-hO=
zgVgs2m`5N3$IS%}Zg9Vww{pu@Yi5u_ZO4kOVLAJOwl76$Q>l({X2!>%T9crl!}Y@E
z%@8Xb3rdwZ2akgiNagHJlCU=UjjG<6CyK2%y<?N+BXx-(<1@SchM2>M6v$x1x|s@Y
zhqjzL>Y!B!0e=T8WKwjPjYBhq$7dOiXn;+T2y&KM0U%3SOj}~~lv<CQVT8bY$rei8
z^8`LG<u<frk$VaMWpODdB{6P6)a2blYK{skWA-weFFa*T2G;62IKzGIXAFbv^l+=)
z*?@7x_NtQ92kCgy4=Y!Yzfq9Z<cbI#v*fIlD=g|rKN@}F-&<LuoU4euX86D1oxi)0
zCLR0H#>q~)IRpauOBB1CMUxx7g0o<k&p?X@B4B?wV)`H`_!gR!AJvA1=eG#gbbAxy
zbTZ8DXu2TZ5KGZcs$dkLrAeg$*Hxyp8cMwIE)BSI!;2eII<+c9sBW5;!RzmTz$B%3
zKsV302G<>^3%04{LuWFmDnE>|2v)%(m@Gf3hJNo-jlEWxBg_S>*5i1xK1y|B9gQf#
z{&M_bJT)*V;(3^Pjb7OEL7UefYcz<AjPkZw>t$KHb5_-Q*2G;@g`*g5&zXl8u_9s;
z<JK3R%R6qf&^B*)0cuHjn=dAW4y3QegO92w*#Mh#opo_v<LPBEw-ikRlLbv^+3)D2
zQm3wmj6gp-w`d<U+K>`~S}@zA{8V@8UQ4w8AmhdRekM*PR-{IIoQQUd?YvUk;PCS(
z%Mzv85%nFTQY}J8$b6}_qf0R=AaZL`uJ=q2bgxLk{)&+QD2dsI|AF0m%AjEA{I_IS
zR`~}eE6ZO2B?S3TipQ>oY%xe4SDR^Z%iW@5Xp|KV$^v2#1+L^Syn1n-%3k9m{Xmrg
z<cjZlR<~nkpNB+adHdY<2C_v1llQR2a-1X{6@LG3qbM=~<LtL2ThvY_UPnb-9g=hd
zhJwfwIt$V-<{NYW@h3byhK-d8X<fuszgT@56C|qy7jF6Q&%}OXFfvP#q>%HdTIPW@
z%6*2=;&SGx>S7?h00!H9dbl{D=kMJ|NM%_nFMYoxhkWRII@LmZR+IFuDbeO(b;>L$
zIle`%a$~sVw^M5!czz}`Hjg&{_7xI{wuOx~2d>+unaBM&VD=ry4#-GBa1$+jXv|?I
znx_4*V1&T9FmIKV=lRoMccD0AU1y5e0#ol+C_q#Fx3~-Q#Uh7klv1*DjJOyMF7T>J
zk9MqZRf|3?JsM?)Ip%Lh+m~6%g}$HRA`uCxvDur(5fRFvKXHssV$Sq&>>(*P7Q$8O
zh~|4&42(^@R_qr;O(cnOR*v^sE=9&mOxciUg}tCXVVJfv2DwKE`XbW8NzG~iQSz1;
z)!hy_x16#*Oj+Bdxd>>Z$*d}VhLP)dP#Uibyobl(xOg#}>of0$`w~H1x8)VDorKm$
zZ(wH1_^^tYOzNesbL7;^=5f!Ps%rZ%L_bciZte0ilRL*Q8iXq&C5WnN?ZP)a>NRW=
zRYNa<lSyJZ`GxR0!WsVCj%tht%4fsi^wpMKcLDt7HWux}WF3XlqECk33xWbAY_V4H
z4>4;I&JdQPKa*x~JK2wU>&i#!!E6nZ!vTSe5rl~Lz5eS`;8)PXd?`XEELd2-6ye=a
z^X4^;2Y`>AjfsbcHnGui6Hdf#0S0;vhY6!+LPgZe_g3-^gMHQ}men;Zi1O5bPo1Lm
zS5g~E270V=mpYT8S;G9!KB6?nFLxko`~h7Y+zDeBQgk2@sG3HSAn~qvSbuI*wLd;w
zDZ{G_k{lk<eVk3+87ybm)J`&gjsS^4!NlkgUcrcy7MttO<p=w|^p(FI{RCzpA-<2a
z3Wp|D8#pFDo;d<P(JBQ;XRfKkl+sUuqYetuW%MoQO<ugI-0Qp3Q=gTQMwwUa>bkkG
z3rj$1%ciBFo?ly2T?eBmH>T$stwtzA?r+&D@>Ny@aeSqXj4Vd`CJ4ppz#lMSNGZq*
z*mw^@zl@SEnGsXq#3-rr7dzinGS*sx-!Ca{-Zd)RQI@OI#g3}o5f*qw=#{%K<2fVN
z?m}k;I)(Vfgc+Pt)e3mG4-Q1SUw)(99$5Y}>3axSbE0~n_nK)FN*YY6<QIhwKijH%
z7m&Pyr9(}O&qd?igcK~5$UJw%PiTg`kT@rQ5kJ4^N+)BIaMddlKTxnw1(V8IEeEoC
z)Lq2nFK=+|4mJqa6s7^!Lk+T+%jERKvJB;WSq52xCFaL+eDYz9&T*VU3uom8NnEBw
zw1IzD+!6(|L{0Doh}))}6Qg*`!XCxhVf7G?B`R8<b~jxAB7;+0P0@$f!xe$q^Vf|`
z<|+5X8r6zBTZt99n@qjQj6e`7`L=_T6sfG!M)D=wYhd@ZQ1lB^<AslLosepsox6~q
z&u22)3RhS$Yo!{%U1QOQTId#U{5k93BO3|3;lD4cn&B_24;Z#lOO8<~uj&z5NyvLP
zs`189ju>{Z({fa-r}~C>a;SoGGHp>U#wS`yD3Git59OxxNUH6IP0JsxnUx`T4W;wV
z^e5b#N&;nSjR3Ux!s0zP>yLf-;m2AzZRXi8KoCsZngF5iPfqz6T-qJAD11>2jT$G)
zaLB1^E35Dp<svVs62CDG?=PA4;|S(feXuJ~P8ON=CzV{Brv6yLY6K%ZY{=mweWub!
zDf{@!u}AA=;z^P`rDdb$-{m(3PD(w8=xxCkRw>GYOiQ?y5!B={yeO<9!9Y@*89Ls6
z|1&f1MxwneZmUuobCO-g5*2R7bOzH?A7h4Q4CmhJ2FIU+q>2&T><+$KYjE@fh<3~r
zI}HcMm@#2y%;0-Y>9%JyzW9TY_QuRvrk11A21bo*@tx!XfI%y~lw*}Q_zlz<IZ1+7
zc3M)7P6_NLu#A<{8NrXwP|Kb#y<6}7tOE3mlZGCGmft^vK$Q{Fy(*GLA)RdS-PyKz
zRXn>KG=BNl^f7})qyvsBWNeDF`?SWL#1Lc(Vi~1BLxae5b(I);PX*0%tI6)e=fuI^
zsuA+i!!3_6P#a$^P+^SGFH(}}ISnqS=J}6f>=Z<@+H*Rsz^i}>h<n<rizQ{3h=&z7
zz5NN1%BdsyeJC+bZ<z+=?DFoaytR+bhuP+BH^j&umeVqP<zO=+M~JM)P4mwv+Sm%n
zZxABwCj~-2)hdqBwOY0HSt<8Ab^_9=(l2l#nGV<sfH-IGEL9Za*bSH<(xKwXXo6c(
z2`nI&$IGKyKcw`==?WJ8lY*op1-|hiQF7u-8xEh1G;vp$E&=-DpegDAML@d0@qP51
z>4}$Apl%uQ;`v6NRh7w<z!K>zWqnZD^cfTNi-!DnNZ{|E-to&Engg4E9uJ8J9YQ-Y
z-hxnM4<aUAc0t>FXv9RgR0mSu_}>f)uWY?2B@KCC(tM<=L<1K^`dp}LXOy-KI~|$k
z`we}k6KWp$m~hAr+Ot0*+L5m>(+C|~ta1#sO3>6u#L?l3?yKD?Mub8|$=XVqo0p4}
z-#5OcG+NjltCcPWd+7Wc&(S^!x_{3X0h(3MN(azYMallDQ&fMtP~HW8P#oN}P>tO3
z=1E)LR;5r6k?!3fQFKKiGR-`^Q&d~QD9I;kA|VoHMb?<>LC~T(*0Ad_BBf&yjv&-L
zgpQ4&D4rVnFXB=E{s4{cY*oqZd)c*0HE|2ldkzU;UYRXEir>i(4MuYJas<P5;Gb$;
z_c}OYephkX=t@Apyot*ssDX;3*5S)ceZdnCARl_YTEdJZ41u$H{dBITs48`cXld5O
zuZShP_omduzCxIC!O`YbA}@5e77iI@I!>rSFV2`J&f5R$4(krzT*hd{WDi}+obsC{
z1pTI?xyRek*Ki<XPIb%tsP>lnhB9MX#KX*hiamv-gGu{&ve2xmxr+97>~c<du$x7A
zx{zTkEqa%)E(!imSyM$#_1loE`FI->e9dfBetrSl*eO-*MHp_7@{2pp@p?k|AN%GE
ziq}QYM}kP_2;7X@#vWtPzU1J^iIg(wPac0{k&^8D;!Vw8ifjw-_-PcFzAW(NrUdyU
z!=eQYaU$Sv#Zi@i3GvaJP`0a~o6B(QePHc6cYt$vwZlJ+p#yZ+as#+yFjjrY`q?x)
zUC&Pfp2;6`@~VjUy|k&<sd$K0aOT(+iIVf(3<DOsoT@WzJ*BRmSse_PjPzlV0;${X
z!$$112pH7c6gk*%#i?m2k%b{xD4RJcUcT+{X=>#i#;Rn!*p#Z+2A$gY7GI`09%K<@
zS2$6j3t7<^*`|fHm>E^p5ysA$z#7`%WwhFQD-4Zj@3hzt!$xGzy>gI=*Sw@>{=A!|
z&V7M3(N2m3{~P++eIM|`WAzZ0#&@Q!`uLvm^2gnZZNnyNcfSH>_zq=l-)%rUZ#!x8
zG0g-iPDbxi_Zk+!YmLSQDl3r&=WLzFdJtZ_`)!T_1@|q#vbG$UnTBo8kSm%se+#LW
z>-_G+)#RV9>gaPJ4r2(N+FlKonw7X@haKOF?QN>2p&a+#80bK=hqiHHhE}=!Bb{)f
zGXLye!NW@(k<!~*h0HuVvP~pZW&i!ik|<L$m6=H@*ZK7K4;1E|dy}<_RX!=XaV0U1
zm4?FM!r)rvmY2PSY&p9K+Qd2kjaMK1#J=y<*BPJG`smM2#gRHhWX_gQJJHQFUhc}~
zP85Uv)D;?rh;=*Le1**+&PyhxcL50yG3_>6W!_-K43v${rYjTBr}lv=3tA8|QVgP{
z6+h2Bf4zIkCM^;V-zvK<qWfM=fmNT0FX$3>_NMy;F~{1qKHyN5+FNv-OCeXQ3c`P)
z0>WFoyHYjhJPby@5v}GxKJdWtXEi0zg5&4(n^()a=vt5eWh~}-^ug3ndnRu~1haJ8
zu(&H!c9p-ng($Z8A&vuDjZ2MqIf8Pa<fkH<sF<vl95wxEJ!3!DX2(kKqx{S`!*%sr
z%VMw@$DLEMlR%&zva!g#Qhd`xymB7+*f%6D8uV0EFdQ+MOmpztgk`g|k8~K>GDrDi
z6Y&qFn9i;=?UlFm)1iu(lm6^h<N$++mAd!aE6l#0s!K@k*vTrB4d;wKDBA!!Ge_-o
zJX=yqzr9GYHpm5iMY>GW#Tnar59>bn$gx@-s%kG9>RFzS_O+rnVt@L$SX3@`g=D3B
z&y9v0N-Svm)7KQ_in`<Oz59y4`Z?hsh3bPMY7{No;GtTBP#|gCl_1@8CM)(&c~nQ`
zznY&|0Vz`Y72&wwSO2zb)Y6)*FlPMq?GMNHdTVO6k2}+!NtmtmuI@1XCReW_ri7`+
z-P;q5hgoN-G|ren>HGZg&35!mxaIBlZ^9-(u+EjvcH@IxqD&bc2|!*4aaQ3@qrZIX
zK#u@teYf>lu$PXB_+wR)=dd~6+KoUD$KWyy2hKdWgHg%Ve{Pk2)^Cr5V^Mp47fTZ#
zM2UQxFBQD6m`NEW<!_l;zl!&2FD+#7crYSt8*6lkG~{}qhc8C_H0u<F_I)NcH<Cl^
zhNW-J^GUN$l($F(vhzVv7Mks=nq**Lr0qjRJQw|mbY?6nXrY=l;sJ$JqQilPP5EI%
zR?2O`tdx;cdhj%}tt3J=?Ud@%FpEasp)uS&MzruU%(5lz@L83FbMw0lcEm9k#={6f
z{FiEVFM_7z=0Q8FZ-nIzqDd@E`55~1Qw%1`OD)faIqqP?6m`u*u)YVypI3cyC`hb*
zLkzWZ-f*3F#(O_W+&d(_FV5$32n$~f2fy}*o#~zek|7JiCdSpO{RTJujmoJB%`PqT
zUfz=F6ltuq;IiUU(LhNSfG$ls^SW={e4UW`*Dpz6e)qtveE0Eu8C}{q^ZRk+F~-ZV
z)XPP9+IJVWlI1j9ih`A1%8uQ*m>jyReICu>wbs(Mqj*|ER+JpHvxLbXby;JpUj(@-
z;;=>{W`uz^JV7suLbf7o9Y3T}kStEj@FD7dI*+UrC{SDL@^cA9HT`lGJpKCz;g!A@
z?k-wY<qVlEhv$!&@-F1yhI(S*7aTnD;KQ&yD*4q+N8}k#k{+uBlu)O@bix+=eB#g2
zK3)0MP=QkeMRioss6K`UWeXu*8?(A3FEp$T{0?cf3R?Hu2v?1o`wT%PQfe4*Vvpcs
zb$rsiwj|8s8u*r7%GRYcU?4yAWhm5+xP{NuMRTANdRyjDDX;ap2a~wTIloco!l(t|
zKvYEi$XGxsU3T-vn5%VC-KP^f+VVg!=LH7h%k-zL-0tnzdZo|j#ziSe$1nTmuz_&z
z)9Pq~f<b(#$wHRx75#FMi%+%5!HR^O7vT;ZBWkCG78@8r>*xHRSR~nLrEx0%`klig
zY(G4IwPFIK&6-?nwkF#mxUNcX@DoS6<N4_wYHFX(Wa7(Xm$pbIc^;?`dHf?Mqcr<B
z7nhaci@{d!i76TS;sUF0$maf*ed^(ta)?7QxELV}`i8~FRyajO;&a!jOQKbwb^*?J
zjkk??^VFSioGofqWbYSVr47zQFSygDrmpmMjt&-6r^J6I9nIE!u;+&ri4q07q#WTL
zx>5~i3(gLGUR<JdQL2MaJ~ATr0*HI4%+<Rj?d|<MCpOITgARyO06wib=}2_e6whZr
zMp3w5S{(t1P%BC~op{!I;c9U6Nd!8)yc^+}_a+YCO;jGD>V!7eijwG%-G)cAVP|dl
z?E;}{3(@L8>b<`{$~SJXH%ne08-+H&&-$>9HAvVwFXVpeuhuxQ&?S74ldq9a5HAT3
z_T1@rIGCd05t|FMhC#6fGOtr)e2=B``TdQF;;HE5qhw)X0lONwI@6p?_x*VPJBnNf
zcx<ugS#rtgyglwdrw<9rIURQ=`D)L?@62RV&*svjW-(s=H5RiDtc%n8&W}l^@$db*
z-ZFE>KHT_oYD`zVw${eDGO%~E#=ZphYU}x|Uu(-!W*PW)TTE9L#F?dtxc!w;j1#}q
z0&@1^=!JZ&ZGX<jX0b&SLOw2R?;{CuVZNj}KWWF&nyqL3T&b{=&d_5Zy6WPl{BlcK
zori--he|`!dzyup*Te~#w`zc+z@A=+IRG`JXXimF^H}$G6)(Pa-;q=Nr2EbbE1l5)
z{g9#KSI-V3VV?!WVVw6f3)#Q{CTf?xntX(ys!(gte!~}kmeTN;n#f|hU0o64k2T~t
z+&`%(UMgA-RT&n60ChmX3X4Ta6Mvph^1aKQ^RK5WDDTnz9m74;*RgxEzdR60P0)(9
z(c!ikY5y$UIQ^}Gq0-gXyhoH8_Om^Yo>Ll-EUvUjj-DK8`czrkpO_@q%jo3ra0Jg>
zeM854=nshuKAy~AWIIG>L;CbN7nQIc+17Q{gBO?AXR1|P8c`~Q`dtDp0vtn<CqoqM
zCMq59C)+780#M<#xBRX7FXGC0IG-}aVRlM35#6&<lVs>?foxgRRk!2}=`Nq^ryy;G
z@uA<}<~SHIN&<VYe!5}Et6g<DEZ08?iPX@477_BmgXrx|wzotdE<)gTCjn%pM5(?r
zZOyT<6f<PPa08Hq(|iHwDFUhLS~6<H)lrpcLpzM~B{P&^q&*ftB>Fs!zCS>qsQK^+
zK#TGAG-rSFC8DoaYh(`RGWZ3ltL^>)*7C`u%zOzENMa26k(@g74CVeuV5t~~Six`)
z9EEB+t?i7KM^rpxU;%-JWxJ%zPD7R<$n_NN#lcCmEceBXn{wt7ckI(%@(;v?XygG=
z1Q*uV+;;%EbQ8|!F1h5yH=$+?h|qT^o$NHasNBBK=B%w{Wu}A=cWD9D(TgDGpk(|C
zdyJ`6s0atHp#cV9<|g^3&vSVhrgrvZ8+8rCmqRMQfy}R@Z9?R~{XFs@n}?b;irbC$
zWE4XFQoN*6Sl5hB>ZWl|G5`K%E{MYG`w95wpL3==67p(YXPan&r{SPX!LBw{Nyvb^
zN59)H3<86BZo13+8)gwz=;h`dZ~_VHa@!zTMi__60LyYRh2Kt6-&v9UCJ%2Uo#=GI
zY}k3e_Uhd|cWAp@qOaVr0$IlxJgMQ1@G*q{plT8(?i@3d`nyR=O^b%{p|N6f<`&0@
zOJ}>7n-|-(9gc`xrlCDttHpi&csChJ-B&cM^G4YT_mQ;lV;km?17OOtdoqu~a-BZT
zmWq}R^2-eX&VA!)lj5h@lbqJLmgY?Y=>=D5z>iIl-VGUIc}6hDWC~$)Jx#KCQJz9U
zy)}cpLz{RoNC%1HFm&=;beEm>f6g4zrN1C#=+`w=)$}+kCMa~~;7V<!!Z1&q&mZ6I
ziWLSLYro8QMJ~m-ys7Y}%Athk{d+=`R0<OL(Dy@I4vojHt4O?<g|FFz8~J;51zbew
zOboXk`y(g$@d6(;>Ji*G{u+Jq@cW8w_kmdv^=Axd+`m)k(?<>p7)`<wN<Hq$|E@D5
zO|UEnn=T*skDFUzcU5cCei5={qGI_TJ7p-D*E>6i#tV~#@$xw{CkGCPaiu|gqaL5e
z@GW$-P!RU_KwHG!9`s%82~1Uo!)6G%m`kFlqrBDo9EX~$9YltRA|uaDp{O28L`mUi
z;Uw+yHv3h=h!2wLHLn&2Qk%{Y9DMmxtV9<Fe5e&P{c80nC;uydw}t@Y->GeB@Cg@=
z;xSwjO!Bz4)I}`*MgGDXY(3+A&}Q!p%#Il1`4|mn_sLtTFqj8Jb9ao~K7uA1^H_Wt
zu!$sWL;PbL>z(DqJ+Qem=L!+OU3k2E8;1s%nlI$V4PPCB3g%3w!$1ut?dTJ~;ZjA0
z9(EKa*B<WB@TidETGy)MvM~yC_PQi_jx(>?@k)}zSZ!);Hok`wfpZmrE$0_v#d~ux
z@Hs&NTaOI27O04oW?eZod%@9fN-bwhK<NhHix}0cqcva-48*AlX{vonPoq(PkX4=*
zw$)v!`1n<+|M$<Jx}gvPbm`RV2F%@bE>GJ5eL1Gb^5Oj;g-bjfdH0VVqekGZ;kAdd
zj36W)P4ls}Vb?Q=iE@E=lp79m3qSm%5o&IcYnWk<(!=8h{iy<A7e@}DI~bzD2ZuEe
z4f2*v$}d%KhSAp_m}k{Rg(UVKtK~yjN!2BcH0$~1TZgg$nAfchml`AFsMf7wP|p{p
znY%d}GR>@LJwI?;eNNanDng;JXWrDaZ~R#iL6gdCb}wl*TcxU4cm9H~^>VLwh-MHc
zK}`p)TWtg=Y)WzsD7Qr<#so__I~(k4GfRxJ{^^IU&s`F$lfiD=ui1m#WYb2|&MVC%
zUhdQ|LU$L+$?;bLI_R;o8PMnJYezTJ-)LVKxSon<uKLzVngbj(-p-Gke^+)8g=P!f
z{S+u#2JhkRMQbA+V0^xHqi>(2RQYo;5ii&|^l>I3O=WXDfJYlf7vXVOS@F_ozZ>w=
z97|sdWyYxi&(?7i`n4x{mSau*don&GtG;pFK%qsmG5ar$04L;9DG{D$sU}+pOw5jw
zBJ)jEBv-K|nRb6~%aWw(MSKG8Zze2BNW&$XYb^6Ue-&LQbCvN=1RqYELPFlX0?rgy
zuQ3iKn$j4FyViqz9@FxJ<b%*D`UB>?h%jyF1#(+`)f#RK>k;mExPFbt-gsi)>L~mM
zVfD68<a5hE?+|u?5XCm#)AeZa3JA(^X_<u!vBum6c2fy_ACs3WPczDC@VK^Jo(PTz
zxaPO{rNSPjkkTpyIqD1CUPWzgP2>?HK0~}B7pKOTuLgLBNgeBMLI^(`G)^juD}KoF
z8!s9W?UJf6I+r%+r)V_uOmY@|R_Y?&d;a;+1+y@iy3dQ2kPMF8y5>=J>Hx+Sz%5;&
z_PZXb_auQ&@qE6Oz=v89wtZHYtPbPjj>W$&-Ki#>h@=4yx|@RaZB!@G$S3!*;+ex1
zay`d%d)tm?ZCx@EOb`(Fz+K2FQownQaW?)%_bmzgoB4zGDet{0Iq6=iij=zheD-(!
z<COU4W$7$fa0GcEf^NM*c;GLjWqLtm?31OZpgMB6g5Mgf92pmtR)nBqWR!W%9R+zw
zVz59{3^o>jjCi8!cT5uS-)9hTU+Wr=3VGiHn`WMU73?1dsw^g~6(oomtY)z!4mI&s
z<?nt&rX5^?65$D(<=#o>+XXgPlyMVL7E?+5t=x$<@;=L$A@Uz4@J7{*F=TX<_N^y{
zH}<2HItY%9aU}C-q>4C5!l#EV=-2hi^Wh-^@p6T))un41lDxJ6&l^@p_6dG+7YHAT
zQ`$E#AUzn5#UrwJY~PdMon3Nx_e)<twt^hyR?pw4%i`4_b|i|To$f@ps?JnH*`%d`
zK0;!9>cht!uy_MER~AVnoZ-DFb^)JxA-rXoc~&7>g_;HV%StFAtg9%`Rwj(*=eoBY
zYLgPpeke^Z-|CEpSilyrxkAkDBeXM`QsJ}HYi@$HJ+_@l&P0T@I~(Y}O%P!)!9=D%
z26?C?gffw=&6(VcRiJ2BS#k1R3l<Ksvt#jm;Wf2y7O$4BO*R&KX+Y}PCT^9$m#C03
z)s5GuKjJa_JMWM<UH0gr<DWZc0~par>YRxSeW5Sr`%a%URA+H2`AA7*1UlREM#^@!
z>EZTzL?Bcd|FB6fCng&=Y7q?L2BfiK`1Z61tdg2%o}<-;{HV^z+J<n^+_*y%YtJ9L
zEN8RmcG~+)G@|R}oI+%B8HFX&5eILNa<XT?uk;@Aakl5I?Z0bUoGI}p;9Q?<6~t<=
zYP!+x!iB5>b*$b(5ubdq|Ex`UGiFj0h{Br0;O1XL-<fJ_ijzGt;t&*vhHBgdipZ9Q
z5|}vOyRTng<A`VNDyWw%*Ng$Mmsp$QjnRx1`Z;Xj;#O%%wh%$(T3GrWXJ>yZPJ!iR
z@j)*d!^SV$9SytqCenlcW*!3Lx?k;ZMbj$rk!+;i<K}(Sl^Gw`S=p37Y4y%aPbaWg
zRPo&P_fSF$GgVQF^<EWBc-9D{b<zw^i}JquFB@aDISp57&EqGA`Ib&SI6O^Mvtl1x
z;;BW#H;zboDC7Q~ZaxmqxPh)y=Ag{=1nsDw{p{q9C4o)n66UY^%h<$3*0wDUUBOsn
zKQo2kKj_^(-%FXO=%6QtcNCPk!HBI*VsSUowH)-dt$Vv4Vw}jVGP>X(y)}hldI=@Z
z^jJ`ozm51OS60J+D_E&Q+MjucRYO~{kGn_4SGyt<RdhkfgV@8ArLM7xS!Fpa8VP;i
z_s3q;MZS{Polm}{3lYox?A(!gp%pD$r#hj?Nrh=LM#3KT8y)2^|G2V~JI#C-u|x!_
zkn4=-qn`OM@ROhho{SQn=4B90wk|<^Kj)i@KMUbr?Q20jq`;n;FCct0fpA$zJjuxU
z=sz05#a{F&2;6t-b8WhVtBei`mxhj_T$@_^<BuAH0keX+D&S-h)d-oLm~CUbF**Tb
zgLUSM87JXMwlT{lI0KG#Y@F!3^)THJO}tEODVOV^yiL+WDnUjY>c!%moDN&hcRyle
zq^})Seag`dW~rGhXqi$LJ%M#^lgtYEZuQoA_gQD)vNR$1Te!hd-V!vUl?zP|${O+2
zg{ro}@Ho<6Gi+f*6Fz}f*syK<NI>Vsp+gri4?bW_4&rNW5SWQ8Road}7Q&dZHsC^k
zi538y64uF%-j+3V-KE_tnu~5ijha!msnDK4G(0wz8LYhC<UA7KH`$D#Xq}P$Dnrg&
zP62RK?+w^ur0gY?t*>ZCWYon|Vn`@*W4E1ykthBMSd6COjbvw7JP0-1AW=~2)%ShV
z#`vW_!Qi@e=K2XY)lOQTC13Q4FDks<AeEv8A)eBA!A4oT_T~c)b$TS!ZI%c14y3N6
z$420{fE$09K=))vSwjeFv`Ge2n@)`pez)z?HX9f!b6lZOD7VHo#bu7N>ODfMF<w4(
zm0xy_`r>%OXYKb)K&Lw(!uG}!cPKh5&^hU$U~+#6j*I(NJdA`LM*}u@8pf&Ld8omh
zCoy5yvSDN_li2VetX{_yv-oW&TLTd$Lkby+-^ZJOChoYi@Ydd{UmHr`uyu0Z^^XNS
zxV~6K?2hZyb6YCZ5@(NLzg!}O?k9zgV{Xy?>}V=vZDthYTKnjF$i)WK3X*=wjru00
z6pSFLMZix8ELSwSqTo$RkZ}wxKJaXyhebrWW{SY%IPwIF8hmgL8@89b!mP4e&EyMz
z4w2uHF#QUfW?b9~>fUEmB7TJ8yC%q~R+e98&(SbUe&GmF^2{@CWO*mi0b4|E`;=zy
z7=HN2pv~NFY3P($BWF4ek_}r8r~PDI++=AeSvgo=a#7m=Mop4$L~UFk!COgwP6O*Q
z<9VQFel^<V56@C7&FlYtvhY6!a%sbvZ|fE0@4XZ}5Qx|*X-h)(=Y(P3N;-2%`>Ic)
z@&ONcHe+n*AWwg*d@aVxa42(9%R<wP|KO@@7JR>|pxp>(<Bt1W3PkN&lM0lB6)|#B
z-y5MRRznz0t46u*CVHJbSbe34=37N-n%DIJLAHT+Nx(~wyIKm^1*It3^K>!KhAQ9h
zy8Z0+CYH>a)K$ajbngiQg#|I!S%&9hh7)A2qSM;hmslTlSX`Cb%duvAvpLaoj=V{6
zw^#Ptd;A`b?~QzKPK`1oR(?!eN@f3h`?4OEY}3PkA<OESwg5V^9FQbL>8_oVvSrU8
zF4;gW%tLAbK&hd`xn0{8xNo}2>+<II1)AgzblB?v@+pz*y7x|XxMOFJaHV4ti#_M#
zkM141G_(BOsRn+H?dczcNPj_kSa*mjKFhlNqDCNnQq>)eK`VRcQ5lRuu|E~cQZ_Ee
z@NPN>qfgLzJU#NoeA?z$6I!+{P&zlOz}FLSjEw%@m!#rML(Z|huMiW6SBcjaw;BTy
z&oHxeGKtIhL!uF7uDZ-vYj{WasN{dG@)|$4ix@9=i9}kQ${PJuN=_Awowu<j`jr;>
zwlR4(`|l>RmxdmEZL=zR$R$2x|B8LLgeR%J82YOYg$`PhrR%eegMR{snVMm6)qLxP
z7oL4x^*aex<AwA~u8NRyrQ?1m{l@wt8p>&flD5Vmvr(~J9~uxBF?&@bMT|7Ao0#F&
zuWSE>cod6)xD!*JnNX@-y`O4b7Sg54xhbez%uaWij)O&<kS&|D<8gp5!51xuh($SR
z^~zr1C-`>}aWdAx=+AH1^cZXv4caQbHm3rzl+pH_7t@<B9%0<RMJwL;%MN+p03d>#
zx6rmywg8TyDVIb?%aPV-M5XC%n2>MnVJ!)EUSzVWqxFx};V50y$RggGr2VHG$k04P
zu0v&=8PZ(8Z`{j2$WZuy7tgTDyL5<Y9Js2%<U)Ysc{`WrIe$NUI&ij#)OwIq|6pil
z83kM@6Rd#f;81E~=v3ny{HiknK3(m-4p&;9sZfMvZy^cDyPjk7J7n;q54~&@($frt
za^9uJvjc85&6#dJOYkAO1bHrt{PxrwL^{F*laQ%O<=K-zC#<4-ZU=NH7PTTultC!H
z0gNJ5L7jKWWZG}#;&emE4q@jt68Y7+$}zBa3=jl64ERoWCQw{w?)sF@g(5ctK6btv
zanv?=D-UQh(rA51rQ$E71nHqYArWGm+;x$qh50>mM@r_&)mNR2UrF;`$|Qj>TRv|z
zpKV7Gew$?}BaQP)TS+GdfzqU(Fep0vc*$bGep#i1D=EX8&~MM_P4QS#u-z{VBRT@?
z%EnW8!}!!cdsdd$#qRrQ%6HNHSWPhhTr@3tPau`Wt<_jYPVYA<=2?jV2WqTR&7RUv
zyI4|cY1PY&?0ob-4U2Fmn}0>-K9F3^I@Lf+CVOpT#wf&<=&LUyv{*Cq$wU(cV*e@j
z5k`ffOP4))MLp1RPlo6XE=tBlJCkAwOSrc>QCuSrCxNR$)?@D#umrwGu8;CpZ=Zh)
zUA3tciiw-p3z`sQ2sfir-JG94Ji(Q)Wsp4!2ax%P0L~dNnm_sKe=Q6rV~X<W|7bF+
zz=lmSr|y=2yq=#QcbrkoW8U>hX|Uq7UTjVq%|_;)J51eSy(AmmYPF#<IQG6Cj+*r$
z=P73=CZ%%NbewxHt52MH?A{Q`aIHVf!pP6o=!pE5AMum(oO+f9on%bCYNPK%AM1hN
zOv&nhUsEb%Ak<Cdx*`h&0iv-3me0vuJk*$j0qgOwgh4s5GvSx9$CDw)AETA$THQW{
zPUXV`@1MlzPe0)KN;wL|mhR?R`<_8nfk^_adgH5(gNr6!JO~~S5&5)7*fj;a_OH59
zdSNzGQW?i0rx>?ZZ{upT02L17iK03&1(bSzWiFEQXjm7C$u<u-mcc*kg)kMSGsgK6
z5|&dU5%N3mZMwbAL?r_Ee}D2=K1!!*>zsvLGzTF8u|@^O^(C(_lyeO~X>Hi&ZqQ9{
zRqL%P;aKsFgRUF>7v5=1x)eq>N}fkztqZ?zZ4fik=WQ%_i<)d5pO-76yYJFGFkapv
zq9GSy-UXX|o@p{pM9RpNc3Mb#trol<xo<xOwYFm16Kf@KtG@*DlNM=6w6N(Hwn98W
z6?}G|*&^CpvlXH{JG3ci;_kno<(Y~{!mfXU1(Xye;i4d_>DTtLicxAbPZ;3#aqsg_
z9#a1j_Sc^#S4pP-2C(kZ<={}x)Eys?5p}>~89gX5a?EEJA7WlN$mJ$FVb5U^g`7(W
zoT5LQb@xImfubeG*#c@QCCXulFc`!nzZ^s6-yctaZ1|Mtym7dF+2vgO8V3pa#ZJi?
z6{K^kln0>3Fl(4$r7+l}w>q@r{qCXko8KQmW6Y`ZS!)X%&SGsB!OF2Kk9XR|m))jo
z>XKO1WFK~y++GSbL9$?|?PHi(*L4W5p8H}pK(js^eu3h#Da}O|AQrFbTi1#W1iMnQ
zeL<h*UBIJ93Ef79fmjtS^wL)M`u_F&@?$)p#rKJkVH{mEV~$VW3HwWuQ<Q25xEU55
zL}gv-e;?HZqgZnzV`A$*y>TJ33jHHVD<TriC?+$)PKPsdf{S+a>vVk7Pfs4#i==f{
z{m?)4)DYQjM7nj(`8V>krCBb#{4}cTagyP(5!B*wPb0|XvzZhq-!2QhzB;6>q0iJf
z^w{44@7hd4BoFe>$KZ06LE+djT}w!Yxr~Ri1xkC%X9=q1cw0cwIN_UzY~BsabryBS
zQOqBLgW@(Wydg9usF;$VGxR)N^o1oz8r*XWNR{=S&chb2Z+;O>S1b%z7OzqktR3*U
zwTnR+)N&r#odB#;_TIDWTXmedtimw4hd*MGw1Jo>R&Ad`&Ju*Fj{dQ>stbDwU+gN`
zIeTx*VT#)>KkRsABb@3_@@m5#)VA@ou<gbj^yw2&FTe0>dbnJzqyrA&%^?u+7a4<I
zV%~ww6%A7(az7EMH0O+;j9?|7NH}$x4_jQ7XZCRqaz+@wloCZk#q!}1^U6Zx8OvM&
zA&Ky(1347otXY+-{3G)BO0#Xg>fFCKezEi@aN*Z#w>vlx;I4{>uN%{fFNOaY1OBd@
zuaM;YIBjp4=Qv-^kXm=EMt2b$ZQFu?Feetkoe(^lgYJy<yXu>afhlQB90g?QitQA>
zmDMPt!7V-`H(VV2vyC>kiieGte)-7^(r`^W;v-{t5<=r2bP9^Bmp`@N+L2O(t!E}M
z9|!Mu74N!-KhF<LD&SjmwDn76g~?&9MBL*96`m><nFFnpO!!f^_@G|$X70N!r3D1Z
z4sYY10^Z)J5?#TT54o6B!`1lpzthfZZjgdL;Hj<7_FR=y8|ifdt%vrtuUCqMI2_=8
zmiG-G{1R_UNKt&yS841loik&Y3GbrF0=0*jIW%bpV6Netu{=2@CaL}SFl<au*UKfc
zQBwKoyIo0hmIFs3z&E4HS+X=ygB4<~{P~6Yv7BAHnjBQaUb}A=wT{_yF%xWKHdFD<
zCKiuYtO{piKgfO7RH7Lu7NQh&Rs*|pU{Z6e2M_1uBy}*q5C6=d;bpG1e)5dXfusOE
z-y-hiS!{1EYJCGJpV*(LDR5@pTN>H55edlIP{n!4l-rOrzFENK5(|uEJwb<jFKHw_
z<--EM7g6fBPiKM14w-*Vd(;xlBy#Ia>@(pos{Lw%#O3mg|GgmxfeEgC1&?l|efnpC
z7<!#qOR>RxD465Ev0@GUQv*kLfcK^c+PWwJmPj7qJZf#N|LucMw!26${pJ`56NgVQ
z<CEqJGemTAzJ=nhLxlx}I6d}r>c|Q9z|L`)Vv>_UbrycUC}}JSmW7aMnkHv#-Di1y
z!;{j558pTQFdN~rTz13MeM~Y-T@CA?btLxlR6ucjm9$ZMB`0qA4~twb7k?$SuP$p#
zG+2hqL4Weh*2Bf~^}CLC<i%<wUILG~5*O4N6#q-VMsZ)x{qum|Jje7~E`+3V;lk4o
zkQQH{R4vOiU6_TGdx(|5qm<s!XwIbG14aN<<sC#A8@XF`V~*xQ-fK;5HhUeq;%e-Y
z)m2HkOsRp|Gl~8QQp^;>c8&Ub6rk&&JtsAXR|UDK->B~1?r=j-O+eL^##zh6ma#l`
zMym0E$?_Z{@Dtw-O`nd94-RGQUU~b+{Hs~bUIXP63tJ8=mHigA(G+>WFi=|d8TFZr
zv5pm1?)Zp8mogo`^6N2~4B2<jWX|TR$VV}<)KZtR8G3StI9Z6IMzR#YY;QL7RoGF0
zZbz~fw;#syud}e@d1|xd(j2+&-WR#=W9R|J<)g!Iez^!<ztqG$R6cUR<u&Kk8b6_;
zXH!-ZXlm9s16`EuIN6onw_JhGlg-*qi8t=JH=I&vX6^UW$kGykC*#hvY{}iY=Qe-S
z!B&dfhi#G>GX`Ikd&5ObnLu)CeM;p*Eq{qV*COaz++=O~!xcn=N^iAokmkL^$b!2#
zXF<a{nj772EOiL{=TvO?&GrdW`8%ZPm}?^saB?6dEDuUYq_Qx8p^Cpq;uO~E+r<%J
zDdp~roQGD%=nl3C_1W8fU>DIQu^RF&);!c1M=~5EOgK%jEqA*~)HCEj+lfitp`V%7
zOY@#4&7Bnr^FrW|RmO$GslBND1qB`vgzoMq3w3U|)umzo18=H6W4ZWe>-BAE-Y3s|
z9$qs61qnC_V92>Pq5~o01Q}dL7Ctr&PNfDml9NDwEBeDDL1h@s(cJ6d7h1;rt;Aj^
zJFBpn77kKh<oB9hD0P)|q=mCC#yK9;4n>}x@&20$9C6O$df+K?Lx{)SA2yrTvANz~
z4OL-iW^t1>dXWMAQT0Na^h2Tj$TTb2>6cGr?>YM*?c(pb%wJ_RMSMH|hl_LA5(PkT
zXxa8F+qP}nwr$(CZQHhO+qP@ES94}QB3a}nxzYk}j@dVRT`@pG!=&o!nMyxvY~jM#
z^Hp(gk^ZyicZ8hg)x92RMZ=w7l|(E*NY*BSepxL$te8Xn$xfZkpv@|MRWe|Zn`3&m
zx$m54sHh&frwXQ?r(XUqYr$|Yf&-zL!GgoCX(tDYC1oilmc)jh6x0wp1RNxZCW^{e
zjzsY~^!g-O%>&O)BU9N)9{Q%MJC3{7c|EhUv+tq79MrG2@DTU62pwP!ISspIHn6r`
zRj0IWD*M+CKBdG82#cYU8~M)yGUw7t(H=elh-p$Vto^%>d;u8Gol}^IASE>P5Nleh
zC}fv8&-`+%;4#pdx+3N54dLik-(|e~-_iYz5mm^B5cS0dU1`o|y|!DE48iFt8lgsZ
zR01<1rgP)FkE$>Y0p0@@N-Lr5k7vZ+9b`Jq!M3LNm1XMg!*#?-PndHz^4BPEhRcCM
zY^yifZxUCbEM@07ohgAw)$K4dV0RlGxjmytbW7RkMcy4|!_q;L4#D{Cu_&Fy+&=2T
zSIEU64F=m|fvFWE37y|rPWSpv?eb0TIUv8k3ch}VDm`pj@{Q_ji*geUa+{L|$U~~3
zFEIjZATOg{3xq_5P}+;^&&Uu|9aHsw$92jyTjgRkz|G0lRy8>($E4FX3dyk)$E>2s
ztbkR14_{s`b0GN<OhloZVP}3qmNOy0lJ&pOM}VGjuTCK*hTPLGV{K3gN;1Xw;J`T_
zHeR`Ku(z{>0Z0HsOn`xP75=vxl%lMCsBKBMx`;q}CXC*EYW0;qd^o_+sSr>6eO(}T
zrCz2sf8mElr73Y)jg~$|m{ZJ0k8PwNB9CNjjX2>_T}lWvf&yjl^W9WI1tBaAbp|QO
zFS8%|=#*j%k7=cB$Zf@wMmqs|kp+`|xNu7}8XQl0OUy`}Q6)-u6p=vwR`*(r!V&8F
zP#g6{Ep8=+D?LAhX_a>d&ketTtE+2vVp-N;a-Ge<=rS7{q7T@Oe=ZR}gez}USywxS
zG(r7%iTynE^+rZltfvE|Gfjw7_&3Wr-SH_OatU^Qp2KJN1{!gzS+7oF3s%{pwS0&1
zo({<gF<z!(p_#^~JeBp9jE%Rj)8!d~SnP#CzL*KJfsU$)UNRhlL^+NEjul|q6oQKQ
zbADCchj<Vu!=D*7V*=!Xw=xB|O-eDn#ivJS?(11)8)+ps4P&9&UM{%2$wsz2;Ve0I
znf_^t1`);>F5`H9kIm3oq*#eV{z|N&``+PtEI)=R^3kLLe1i#$yQZ!`nHJF7#fhU~
zEAvU_68;2fK%l{Z(WmtfA3B6Vac>tS>L<R5TW+XR)!epnH#*GCJ_Go(HazHg`&I*n
zXF!5ug4}hSc^g~*w^P^JWhJwwUh$Z`Db)m(mby1|Bjm0q4}8_XhgMRjX}o*B+#YMa
z!@Q?aJQ@i<2)kEee)-5yU#dL$eUwFy>HdAlR8EtrjxkvGSq#-koID@f<gq?wpu}Yo
zVptDsiSrQ?t%a7!OD6I*%#ic_)@w$*DVZ4x!<Cpep-H0qd1z}?kP)}SJs@2=ein2H
zzKAWYPl%jtEk%6#B<dsyLr0OrTw5ER(p%kfMJLSn-CH3Ep8&uo098z-r}I3Dw57{D
zYOu9gst)N1&rAQ%F#0+G0h{NL;nkBDex{CK&;}#}01lVDgfC8d{x$BSEIovD49E*;
zt%PE==YX{R%e;PwcD_C<%gTt^*d9f^Saw?Fc}2I>zNp5=$tTi7QnCP0;)(m{YD@)z
z=5f^Ix6NnvQR}iznyEBcY4s-&gtXrP9}c3^%0C1Dd@^B8K6f{~32>$Fa1Vn+fDA-c
zC$#vR&m#+ygdhGqH;6VJh{4cG2rzzYfcG{`A7sgH{|t*H;cDDs)K2;iIk(D{c&=L|
zPw%qfZr4JSR&4u|w`?uMl|J2Htq&5)Y*XJRGWM2YD4O5r65}P&W*vq2nm7R`uLMg!
zM$u&uXIpb8oSEqbDK_2fdgy)gtP|tX@~1_uT%uX3ev0PzqYB(n1;{7)ZFIBw%DPvq
zj|)9*vR!J=00LyD{Ns6FbF2Q0jK;9qPA$qFo&Ly5Nb=_wyjO!9549}NtCb+R#Ku6q
z$}Hg_R&5z~OS&XaV@Xs5XUm!z9dQ{iMvyijFz@AaLsHF;2?r7BOI#EYdBsxwZf|!G
z$TiB9LG?O1mJN}YH(Kuxa)_1wH~l=t_iCDd#qQ`jn#b#KlF!GH*%u`Ic6$~wrn@$d
zX6kIY1E;J)O3zOkMtCQK2sDWy(<6?2@YJwILRP}tr?kEpf=0J4RI(ew>qaR!i&LcS
zDzc8-8NaHVY5ab1T14Kb_Wa=``h$Y^2=NdONXr9pVoo1h_@n`;17Ad*zMsCG!gDu>
zFXeYic5ODp!74?K4BTH{jZO^N*pz#oUK+0<&toJ-GB^<;hHawfNK<iWJud5r$%=K0
zs4su+KEeM9NQ64Le6tPBn@U?HP=HztN#5SIshYmh+;;@EOx<@7Yn*&h;oPgB93IYA
z2a<$>@O<-eAL(EAFf#IrZ%m2?ej%~7&<gi!jlN|E30~iZ43OMDN~i>{+Hv9DxwAsO
zx{1MuypZ<Kbr#Om_J;T$OwfQRnfe!N9Js=nG;AL<y+7J-1_j!X1#bp+s+F7DRbmy}
z9!!rx`N~V*clqRM54FACg)R{=cST6~7CR5BRLh|<%Y7~Yf)dw&tm<Hb0Jl8WpWFH|
z@CI7h0_P!2bR+3Q$FZ@ozf*AfjjR%=&8MDKLzu{Tw2ry@pj>MIXEtYozVDKUT^YY!
zHI5$~24u<Cx`c9vLg8MPNsq==f)J}paJ_QL6~m|nS#V_XMKt_q^OfDLsMB#LdPfyP
zV7KjZnF=ULn|9wvmYu^Es2%GW4H2b5l<j>G#=>OQY;Z>?Os_A@dSkHkOn<*lD2bx}
zN$5#m!lvf=6~mEa!mE@LQV>?N|9lmV*S(D+0`2vpuB82NTG^M9Bbld@WU^?2xbgro
z<g35julmjjh4$%3&kI?!)?~plDZDryey8TZY6tJo-B?uI8DL-QKx<t2CgTn8PWf2@
z`w+64D)D|}Tm5RxLYy#IO|piI-A|mhpWV<n-3B7Pf)bkdvyU}az11Ni<n(N+FdWNr
zxBQu$2=+WTPtnfdC}nP?mv*%zt)Qa<J3>-N^11M*!$1O~b-O)^YM(>Nt&0szwC<$o
z$hy;xfJ|^4vEtXHWd#{-gJH|lY4$=eTxf|LxE=GZB~q?k$}rB=4&!UQ$TLX;7?{7j
z7MF;*UU9%QB0tEi-lX0$V{{>S<^5WS%T28Zrr;u4v7@JdXevH!+={I+;opU|!}y1X
ztDF62)tXZ{*aM2WQOWLRg<_|alyXF=NVWb~J`a>JyvnIT5_oN@o@h(=$g!72X6}AE
z7QF2uzJ@<@Ec$mPlc+araRFL`B=ly0?8!+jwGsrJ-hcdU27*2n$@QxKN$c#nGW+ln
zV5$AZZi1t|xa*w13ybPI4WY7v!Ih{99ku)im<hPBCfB)bvyT;>cy3N_3V(^!x^G4u
z4PMeFi|(gR%r)>B<X`SKZvL;$s=0k{xwBy`lw7_^f8$gdcVSb!+bMLj<J06mXWxp>
z&rs>|ISy{J*X+~TKYqeNPd81kQ2P|BD^;X$!fH9f#_lR4;sx<3Gc224b)^znJc3+&
ztZcgWcwtz`e|J!4rffqIvpbc8f{TVgT0dycJh}q!0A%l{gAe?f=3NHLoI$sEAP|Oc
z{mv5bU%bykXu9D_Bg+2oI&JFsPha@4izG=YAKz~sjkjeU!!N!^!Y)|->CA-s{cBjr
z7PU%xvE{oQVQX5rSUr}8+T0G~_J_~&EcU+=j>UF3_xtqa>|If>-KI_UV}0j{`1vWY
z;U|=obWem&!*gAkD5KS9J|4&R6vr@9S{ml)k4!D7ok_88Y6)2LF{>f?!)XssNQt_e
z0<B%qcYQM5>Q?pkECCKk#n*+}5x*yQf!5SjSD}_gfhIdIyf1CMnUQu4+Fiuq0Y&8|
z)-bxq8Rd%zATm(uL%ugy#I7PY^>?$$rPbEy!biGV2|~&4sjqOB725;`YLsV6#yMRE
z=ml+ZJ7PU@vObf18fKt=AM2Op2m?j;0#<ffP>*BM0h80qDy8~zEucFZA<kl(L|(=m
zYT868JX(>6B#-Ubw$Z5ZO|vGkZC>in^U_M^s{k<@yWPrnE6UO0=Ma|dv^2yzLZxy#
zF>Dzo!a5_cWVNRH{viN?nuR<@Ff=)g*Yq<Oy7#K>>dPDOoD{Rqit2IOTmnRo9=+X_
z#h;$a&6ne+;VhIqRK3#%-T(pf@-p6UOlPH+D`6w_8|BWjC(2ogpi?6@!<wrTsY_*d
z$<%ddez6DFr(1BTpScoD1?WzrZ$-%+5Sz5o2#@QXc%MG%td5G5KmXnb%A-0DH3Ka#
zcj7yALB1JFC*^|fq<b<|)Sus$7I)Vq&YcC+hr9Py#}QH%=iiOgfX8m|Vke~hNf(5U
z1ExKmt9Ig2Z*9m5KdyJHy|7ub4B(n^(+;sF#tu7AbRzwV7xsNF3YnDg@hRx)`@50C
zM}}2OIbwX2gtKo2cW<?Xu@s>IJha)+^V4L>+{&r4pyp%-`@^JIylZjTp%{wayA@i1
zB5gQ>-ckl-N@`J%hOX?1>D!U8GjDj7pwUtouW_!ld+j6wK#%ydDOp|bNx<2k3Rhd;
z{Wv%c+m$ylT?H@V5=D1^_U}o*Ik&>b5Zq@(%Rc&V+dthPhw$gRdHVBq07;dYg6FXC
zL7s47GAJi1+sC!eRt(p@4~Qc!CN4(1?*WWq4>!!J#W__56+3g%4-t1Z5k-J;*aqjC
zEkj#51lHs-kS%Qz<<GP@P_rryB7wk>1NV`nkcQC(TQf$8?8!6R(7!GDhS08lzd;ZM
zhN99%eJWWb+3QjS<%x-t|K=Fa0L{D;zq7kKp9&fnOQQ{n)TcfQ4zK8qzDp1fHTxU9
zgq-~*zr=Jg5O(gP0rqVHQ}#my3-x_4NCAuh*+8r8Jk1unQ&*Vm(jQN}^O4)`B}pwa
zJ58*#Lr@0V)q>b=DxiC(f>}%3e%?7d>wFB*QVeRS`4Vb=N}Kr2@{)-V8vx@B5zd;S
z1?G$f_1S=(iv~IqbZPKmwqJ5_(K*D!0hj1?D8JLKz$ikQjXQI5#ee-votsQN^!y%k
z>px_FqCyb*hR)hY|LXE&l2sSSHJV~3|8#p>t`f|yZNg_6B*=Akp~6@lOc^>A)V6k@
zGQssH{L)B2U;9V%S!8$j7_rqG|Cb8$Ifr~@H`n?gZm~)((`<x4hmbT1C>N_K%3h(p
zR&XfYmlSU=G{(7<ZL&bb!6O-8){nDDL`Co*?*L>E7q6eg7Zr@$#}7Yt&AEOaRDI2e
zCMw1!kr4pT*tSMIPjzv_Cw34@6OqP1lNc{lcs7l>b>Y~&lt7vD*dkDc7EO*FErL(0
zk*{jSmF|c*W~FH8EY7DN!VOhHIYR;3-?0uIgAH*5dns%ynoYstGRxv(Di?KF(6~VC
zr4w8a$z{NQ*UxJzz@nF=*~-LEoZgn)vw7>;kXuo`O`J-|K-Cl1mvV&?XwvdH?0bie
zO$ZnE6sEHow~K#?E`c1L>l-1bAHAgpztuydi^;F;VO)Kv#Q^{QgZt%Xxzq7*v&FI&
z;e0;FMuKSq$0yWmQl(lRw50jR#J^u63kK2CAiin*09d@AdgCv9;8Vo4yV3P-DdJr&
zy~?j#$-9c)_aTp8WejfHv7c_eSar4_%6A9l#fpiHyQu=pi=LfT>96A}QB0p_la}-n
z!;Z;-Dj#_aE?L@4(8MV%ne<^MV@Z;!dc3)ftJxdf?6mRSL8u))<$ccr$c+C`#W}Fh
zgZxzx(BmB~Sn?OIv2^Lcue`FA3582W49}6g&tVCDTjET4tV5k{w5V4)M6>La$GfSy
z%IrvP>?r(xMHh3os1?X2mo%;X4JRy;j92~h{4PJQ6S_{{!bw)V8rmL;O16d(4{;~w
z0su3ytG;-57t0e6!d8ex8v0H!6BP^lP6RMvcS9MrU99glFWm8^hTzry2Z;PV?@c4h
z?h$W|7!8k*0d{Bn0CpsR>8inWzuV8TZI2sw{GeY%xoZP?lk3vTAePc5{g{Qpe7*$#
z*K^>~lI|N%8iUUD#}?DG-`xNwvNUmk+M2LA4@n*;fC6w5r?jBuKEg#B>NFb%)`>FZ
z@Y6(Cy@<laT2jH~=!ahp=PJV`oNloU%TY$%21+T%1g32+5<@;}KlO(-DLlF!8_F7p
z;=xEsv1JKXk^FWl%Hn_u8a3T9P`^(qP~llLF-3)SKIG3Gt(J_pGiOHT6y2BgLO4lt
z!x<zZs0?%+q8BPGMn?#_caEH#fVvf5HtS%0FF3RR<w1xu_u*-UF@p&Lf{y=cW)00q
zF=Y=l`#lLehSMj^bgG<EqDiHu#jXb&8^=b3&obu#TjecaA~1mpz3LeTXXa^v!U5lB
zggEjmD0I^i!Z!8-Wfpd`<z$+(3Y>RowL8*Oj0XVGdjgxbIJ7zsahF#xwx}?K`7}Hs
z&Sbo|O45HHm$7k8f<>X!-Y=5qiJqcmph2@bYHUfH>(468Gmt4i&EyPCFygq@Kn|&u
z-L;u4SXQc*a0tUIMD?+)tTz}1z=v-40JTz;a1uWXCT3uq4v=Cb!OaannIAbOP>MA}
zv}=rGrDkULk;_!+PA2%TXoYTXet9E=$lGy$>QFZ4_K6`XsM|})x+!ocEVwJV@!3W%
z)2|4lXU@vgI;|~|1DZ1fqSE1Jp9vw+mvg_Euy<0NsliTblac>NtPF}48J$J09V)oV
z&Mz&Yl*UOVb16Xs%%v*cTQ0K;lUF)v33D2GVazlzvpG|SZ!{Dw6?6qqY`2MhgVkz`
z_7N-soKlG%vIUShlVD&uGt9NdSCuO4FA_+ppUJf+*%G^ckD0`8X`7*m5pHvRhP#>4
zUL>$og+aD)jzzQC*$zCsyW6eCJfGM3QwVGnD@~j^blhy`blAmAI6*|G!g>f)<=X*t
z^hkaZX4ACg^Ab3itNEv`guMf?bFVge#62feCZzHvKIYPsirj`oh(Mf_GwDw=vjz%!
z!*M#sPsSo783}iAWVvRdUcopwuQhEFH!uNjk6DpZtJ&mI)8P;#L>yddTei{WlQAN0
zX3elPjZJ$c@k7OK4r~m@*Q4O$n1`i9ZhLnj#>&1yYq;|~p-pd)PTB%5H<HQ5X=TkL
zuDXWH$v%k{v2;ik#I1S(j*~5n%vBGII0HIf)ndeiYg^$9Qonw7`TiHZW{SX7obG|w
z9|18rht9?0hrH`~DIeQBqpS;zDAtwbL7>Ws(c$e;82HapcP6uG=#xVs<57uzPi&4h
z!^a0P8|)X;26=3;DTsYjpWJ(aG|u2-^4OTzF`6vcDQdTri%)dJA)n||f-6NhM&atS
z?m*K#E+4}BU>$BB%RDm*UaAl*AE}whNZT@btDJzZFABq|<LGr<H}sgKJaj+aGmNdp
z^tII>Tr+$<Py`HSLd1|1qT>~COEKrS&|rb2E5MS=rnv6&cJfrbHB;u{QWi7W_mgSp
zC@d?|+T)kfGIeNP_i_ngq);1h#nSTf6Qg5C47cn+B6>gCJk3E&6uZToR77-IfAlPt
z(cLh`Mffa7t)xPQHDRqMUv<}6H5*lY4iVI$U0dL^tfI>|N^^t!ivn2kj>ov0Z}H0N
zAOz8q|9q0ATm<(y;&yH6w+0^BSRaUmoo@yDM6uY_^GurO->%)NTQO$=Jb^M$FdEcY
zfWJ6cT${5otG``02o6{r)c}4#fxoa7eLka6q*D19QpD^gRPe;!XhnE#^tCSvdyyQ}
zhREYHr2(7^nb~*x&VB40BiJ4wFAN`kD!3jU0WiY@Vdx=A>SQ*%F}e{;xb`LH5PcZ-
z1*7ASa;|Kr?``wi_uHG*8h1INi#<+ur!w)TIex9d55}!>hulYuVH~6`GNI39NH+4H
zwAK|nvn@{q$;2M}eo9-!h{JLqZnpj3h$*TQG&&m_pZD;gjfxb+B3ex(f^wCGm4qMq
zWJ^SOsKjg}Pay&XvHp4Aq!i*SCGaCi!`Q0qhY;^9dxtyLXKVh3F>j9SneA?3>{pal
zv%wNAUlT9;@dtoL6x853yYW(15W=_3A%|Z3YW9WYFN(u*$O~@wf}!2FZ;FhA0D;xZ
z%AFhIV7mj~3d9+P!XZHfB(O+yL$TjoI$Y@FH2AHwtQP(FRd96+s1l<W#O{#fK;hHR
zBg{L#6z2sC+=E(=Cy73rI!m%)o;8@x{;MBURn%p^vct-hKDel*rOn=$)VPAa;cv1d
zecG<*7Ibxw5MlYzW&7o_-PzSg*-YCElzM-SRSwBBAWAr@KjHL1zU{O2iaK~|j+=tN
zy_3>4Zx8BqtT7~PGg1>(M4^@e5L0}0^`5Uj69~8$a+X@)ae76G%XlDGGYbrVNqQ1d
zR78w8jNDfx2dB9?5BQPa?bb%l>qQ59V1`_Ewa)@>3U0#zmgPdt;LTMih{C>RB^Fau
zx|^yGk}hZz&rj9bQ>}DnhtQM<@py!WqF~?5LWC_ZDTx?NFzGP=n*nhIrvR63hMfJ7
zcE!2W;-ZH~c@Ejt78MOrgb|0k1uBYhLZ?agQ{+-LztX#_zL|o46iE2MdsDp)l|c~e
zRX35m1mHg&$fm@AqUu4BdAhL3Op+14#xY)s3Z(CL2;=mK?2&VD&VM_B<6vTg5pv=r
zAg6zRW|q$J=7Qenri}}E1~!&G*!FigemXz2UrQ}~EkyXN-Fab_ZuA?$Uv=@W`C#aN
zWKdGo-6VT=wZic}&h7!~F?j=rgPMe~mcy1-?5|6MxRX3Okw-<>(;7GY6`or^IKcUV
z9WI1;!tl{OkL@Sj-ZGuG0@rc|DECHcG8z6&Dm1rvOtJw;Y+>%Rdt;O+nBz@m$~v{d
z#ch@*JATr11FcjKz0hJvO@03c^f{8r9*BT{<s}EB8H7sJwv)U+pI*mZ&DrfGaVzU=
zb3_wBIsB26W^+FVxE0<8z|zdsoIeFVR6-4{zu+5jiF+hKcgq(1Rd3#Nn5qH&vXW>g
z2{Y|+K#hvL!m&f=FA}}{T)II5Gqe>>yE?KcD#B=~12_>-1wB{T=q$?*G)I5gW>`MI
zB@dVfZdJ;kAy}48IB!vf+xMDh7*OeM=0HZw23et|Q=&Z#u0~s?@l553a|1)HO(o=t
zF^2dSEDae{C;Xz3GH+wL_6P@@RqJ%eVJ*L*5f0Mlg-6U5`_-skX(mBV9q|^2H5@pp
z2QfEQkWv+!N(vF3h$YdHCz4(z*?U_W#oXSp+q;;YM0<k2zj7{22q;p}5I#+L@tn?j
z&;M^mUASn-HV9^BaD;?yd&$#Scb_F#G8IYiv~j$j#iprV-IZ$vBA&Vil$&QhZMFJ>
z9G^oz#jAPFJS&#<MdqM$Fp{zXB4pKG&hBsvuV`urKG(^ch$s{eF9X;otkXN|n!hzv
z_=As2_$yBm^$OT^){aHitbI-4S>A9s)5dRNEtW#h+%gG$8`CZUZT_oqMINco=yj+j
zHcXU-t7kwdxi*I2UB3#}^BX`1ac4EwkUaq|OF5aCp@Yd5Nu|S4o@n<2K(q$vLd8#K
zftB$Sw8wFqfTSbuUPgP+|4ne*#R<l5QHbDoyi`;5SY>6yyJgawIzTJOq<NR>5^`Ed
zD!xAZW#vaGopm3+AFcD7I!1I-Z|v6ZtD>KH+vHZ8mXJjpSn#0QZP5arh@?XcQh(qj
z=xb)}ywE_Pi(e)N06>3@i!wECb}El2R;q9HRpKu2`coL*=}H;FknVX^E?eeM=8Cx)
z^IQSx2H^?qO5CzYZ-st*$`b=ugOK^0a_|!qkGEcj|C{k26GhdwcsCUMhgCAyr3ryh
zv-FABQ8H&(p*+T|Q+$s7W#<iJ%e;%({k+_O0=U|joOaZ>+~9uHscn{26xRMvF)Ar}
zF+8!)o{5iqmL;vPv{fHuy`s%RTG6B~y+-pVzHU!MG^YlM@m^ylX@&B1s)}{w>?Bq9
zbK8_Ia+K~WCC2h51x_T_9U*=gISbyi287d`8-|mPQr(YgI-8FD;Ts>wAW8xn6_SY%
zjt?uRX2*mJ$}QP-fkCCNcY-bT5+0FXyS?W?v}3}nzGHy*2OXEZZg?Vom%;rU#>FzP
zV_Fp!%Ib|w0}`}iOy*6pL@I3J`nRr*&=j~dWFa8ymTj&fMZm82GBfmjGcQ|iJgSa_
zt4&4JTNlme9lljkd|>w9AhkE~A@xJ5fiGqP&g`TrW<R)ow=Q?3<#4qdrtgKI4v>*s
z0*MD9k^=ts^ybfmAbEL60tFL(haeO&S^k^&ix_DBCSG{v$nPUcoijA<&sD3k;3S5h
z9D?Wr5Mk6^WX|_EDy@#8LIYjD<EhY&xAFU?T~e8sHkyRGB|uCH_d3v$@A?5hb)qnI
zzj?Szu8l}pq&7l4ylQkyxx-+54RJYiAL&55*_u3$N|)ckpcWXH6Lh+Nr!5L&q$3D;
znBalvWJJipZw~91-X=R8Rd?q0GEc%EMU*(eBVMR}U;)vrY-z`)V(#a!aHpp|w%Bz<
z5{s<=g&xqJ2Op^2@Ol&mvxO)^(|`bzC&~>HGHdRPtf<oXTzLvaTYNK5gXesu?>+m=
z6%rR{(aArQM1Cb0#dbq2`NBUVX6ih0q<OQ-@4(ArqhaQCW$nBhqtz9?(YxV{{cy6L
zME}E^UL5U)xBU8Ru}h<fqgq?bKc9h_Nmx3TTB>^&0~p-yn{-PuROAtoHH&*~f2Hc3
zl0AUE++{;sZJgj-Y5<9AOjS?DvMDIw`nDLjgHnyp9j$ylAfG68ASl3oawU|eP-F#=
z4Oq-m+;>4zFb5WC#Njc(52*ElV)8{q2l7kW#<$>CkFSb$Fzy9N#&5r>rbhm$s6Z{U
z48wa21pIWI9m`Ja3<>u;i5<+cc`xc#xEf+pAZVm=ZBP;6H^TUhEM|OizZtZoQYx-P
zLx%J$b&P&rqAt2c#ETpuR@hhTCtNQXvhPn9nbd`sV-7Wq%UPeXAmEO2%@}E_#?4km
z0K<BmQt+FuNetB!xAkhiKymAV0{Je_S%9x^gdys3QF`KfMLkA~AKk*F*KNnUm#=v-
zUrQ`FYcFGT?+0W0PE!gMKwm`WZc%S^-~g!oU^p6b0^h1Ku!8~pcf}9#vJs+z&2wqZ
z+w8x?2NcW7Twi<usc<^&s1sZkK+`kD?KI4J>_#Al*kUg4A@nWb;!K|`N&B8A3%6Ba
z0v?oRni(?dRV#{<&J&&{4zQgg#GDb#EnmERfxFz}bq0L53iRAh{X>E{U%ga2D|SkK
zo~{@xO8dg05+=5ouyTZ((dKibkay%}E%r^C3_a}BXPurxHI(DUr3GuM;}|65+Q>WB
zl?@17Vy}W*v6`Q3GG}7DT)+n9H_74NiiaX%$xC!%-CuE|qp}g10yNMR{q?i5xoRdr
zlenbi^|0&_{hRy@*4D*dTNVAm{oWBG8gVmEDbmkfKL-l>y<|TvmcP3Se_T0|1qQ<U
zmd14g<uB*}5f=W*-M8gFS=qoDi`4ZNBA|`7W4{f?T9D3&M^YdMPjE0b4J3nL<dlKh
z8}ps4v+sIdn`wJ#b95uj#y9zO?1y%m!WZ&H@)6#DF^$V_R0^%ToT&d6^P6v!;iMpC
zpS*_XVYvwWa)SiO`5&5N^c7-<J_}!_#`wU91Qieal<|`<DxsGmq0APFHb__IXS*G9
z)<$JUZXC~Z0L;U+E*8u`;igt@|INy0bPrB?QmT@Z2dyUrvNf2bud-3f5Tc>h2xr^-
zx?c93`Ku%OmK3LVqa|<Ax$vCV!n2ZS1xFUZCP}G&3$`n&{se?%dx|!I{S;(la)E*T
z_p{YWQ&=6<=Tki<$&hbxhg^?1izvvq?l`^BkuidFA3(W5H4(GAJPd8%1V6~=I9Eax
z`p?h`FzDHrU!YhMyL=C!M#clL$gA!mkip*xefCcV5D*&>w{&ms0R5_V)u@pUwAyt>
z?@pq)ss&y^mcU`Jkg|%Uzo(Aml=!+ws=^@z3zcRR_(4go$*=J$=qvXc%)JX43VV3n
zQCn<@2VABn;^3OT8R9i8O%J4WN=r6+AUqG(7z$JtZ8h=T0eFz{frXT#GE%24MP9=2
zeV;NQE{-=Z>5H{I)ywA|%6p2=8lc8Mj?)o0bw7bj^R>**@q0G&pS3J?tnIoO6o@OL
z($<3~#S{R~bs_MyNqu+B77IWr7aJrYXMlD}(T7YWq_<${4MGdziQUYbx1|9fZyENx
zMs#%oSkn%7nR)SH;-(G~XOoOWn3miPnCpc)Lqn32A10(#(C)rYDPNLMUc(qBhSF@E
zwRC-yPg7^mdktm-!(QNhGqfiv{X}y_s0wXOV!6wwXnK)?g=B4$yymOz4hMIy5jnCM
z0-PzmPm#xd_>dQ1KM7&urS-7Exbia_rKrOY9HAB8#E>HEW1(a_-}drO)Ey#gU3WWF
z@0CfklKyuaNz(uz7l5P2i~BWjMPBQPeBY0XaQb2+{7_##t-GraPZ%V^fTTkz71Ow$
zGpplcN*P-<G0s%62+S*H!uA&-$d^{=tn)6232;e^zm+cX3mAo<jS`AKu3dCC6UA^!
z)jscA%cfGgrOyK7Ob)+`lK3d+JQ@0PG`Xf_CNg|~GSX;pCMkcQ1%n43og>fgtk*92
z=!_~aq;BXPKq2ONpd28ATtc+&pR*HE;|sZ@8m7LEjkTI*^Zqk#>PE@afLjj-H40o$
z4ON`-$Z<dp`#6I4<Eqv)!;u6ykt?jANSHukr=T4zk5%|J4vQ`WtgCJPDaJ12IU8Wq
zKQF!35BAnGH~$8ObaaPQUri!9Q_&`-FU$Of2BPmAU@RiQt~4sGzP0wn!l^f*0*Hk;
zxeWOmkhD@;-keTG2xJ3pq(7M#6ClgxGEi2?vyIid3nb&@PYbGl{4aXsinIo$lPK(3
zCQ1Sz4{<Wr_9NcQ*lod!TLTUB7Z(l9Aw!F}PI^92k0*cbckmXN7tS%nAzvX`#a%?9
zN2TclB(>JVFQT*YtWu{;+rh--7h8akIfY0OBoE)RH*VCi!<<ymtm7mW6~tnaTRgMd
zElwyhM0O3?Y5-A3yO2d`kgl_BJ54=y_CSfid=_gA?G?NCvPV^p!jYOncj@biLq8Py
zoLdiXjHO;VPgr)Ost7d-&sv*F@q?9rOP&j)e#0%CNgbBDQ$@C7TW-qP$u-z$E8{@>
ziQn|Zo#A{rLhFmi!ai4M`#Jc65u8c}ngW17oMmwLMxrp$D?0M<lY|WkcnexcciRzR
zjds!ux6Wis6h|vn1`ivq4pJ#8HQYo5j4+a+3GMXF$Ys|9L0RDk64Im>WG{|emF91x
zezPBd3)5H!XCWTd7Dob{jly!?NAwL9@yyh=9oH1d8Rf~ZX#>*y$n|?pyV?!XZQ+t(
zn!#76R59Txw=M<d#qS%DUzS8olm2~sr%4S?_guofQ`+;FtO?26=JdWUZUmtOHZgki
z9C|*oMd$KEXr5cQkVK;VAqCNa+$y?p*WX-+1-IiC*Ae?W468f9P1o0{@#g(PXa8^C
z{`>?32VZLHCprYtOCvE{YVx(8i*QFE6`96_LB0%=Yun(y%O2?*x^6%fEgI!$pGbLR
z1|bBm)=uMK4~>38J%dv;wZMgxhbh@P8PxV{!%|0rK9HCg%z0a&LMja2Gu6~RjB$y;
z&Fw#ICUR8TC>tbuRiT%=9^>z}y8^*sv^gDrQaO61w;(fKR%-4KaX7aGozY}VN4BUp
z>W-!$5wy}WAVC+aP)nw{wnOeewyow*e5YLQFW2O>1x~<<r`$HdbiqBFlTfG%z@(^A
z9+*hel6|`@36#7r07Py$vg=<GV%BU#yc--6a~7xFu8ntcSD-rV&<cgChhL<s`-!as
zxa|7M5Rd$QVj7bX@J%k92bM6fp25Z6#JPAYB;W11P&PFk>iyhL_oFWs_Rf6_&CA%8
z4=zLlr-wFyuT?UhHM@1I!EF}JL;L5yY*Zwm7T1w#MhdO*60=POz1As?xH_GZBg~pj
zr~#tZPu$W=;o=vyMq}}X$W<`=PcLiu;v$(gOlejNkoaNl{tpk13%0-1S+t>X`ZA80
z=EA81W39u~mFL{A0EX4e&00}%YmqHT9psN)JuMihV)G>EkN;+#Q0iF8C;!t&SprAc
zA!XLg<Sh}K?`dpN(ojr;r?LMmtF_V$7|l2huDJ{&sOL;%#!sjIWJxOh;w|ZP>PxJ|
z?^MIWD14(v=9dzwiUoNx4!$P+@Tee{kd96|#ef0}_&4ruTWDVDpKus=hF|VhMbst>
ztetQE(UO4wQyIg|=U-cDiP>(_;);`*J5fk)su&?bM*9Nbj0iLSW+uy>@u<9J8DBy_
zl^qs<c0Is({cG&Z9mg>PyIcWjk~Dn1uiaqiA*OMR3vEW()T}{{n_RM1bSZ?sLo%`1
z2Am{xN=Fq4U-Ji-vVAaLNWJaXVCmBFZIj9Xyn0y1ayGpwLeBGyz1vjxZs#&TpZAp9
zp}(YZ49!sH2USt6=S1s0%ZL`?Gf(L2v5aFRjXMnSafBMSN5cnMv8z+RP-wz)i|dh&
z1L4Enp2o8XJbTv-)|D~@WT0i$mT|sb$ra}iwO>66I?Rj77+ChAzEoEfp6_rVu#kac
zJsm_qa&g5mcppmf4H0>RkdMP3;S>4=rHWqrF}|ZJU!d;7O6M2mT=KsNnh06B-vu{s
zg0Po^LQ<GKV)m3|@y{=I+L)XBY>TzHhcCO3cUL2V9fU3(1me@t1NpRKkFGkw7R62?
z@au=PSewuDmx?9ILYuOHKQzW%HU-jj^9mOiD<3&Dxi1*0P0*tPH?iov>gAo!9pG%>
zRGPE3%@TUjORC5n8<P@Y;vTS*iz{@kg?r6i(C#ebiy&{4G{S3v;(+wV{(I?Q^8{oE
zDvNMmjGwDHHO%F3*IkU;B#Te!iCS9oDb%>ibAHNC&+P9r<o4z|FJcUJOX}M%l-$OA
zv^_R)$DlgkGZbpea%yVN>P*|GL{L2l-QslELi~Sze1p+t`C2k<)Z~9Ry3B8&2CoTy
z1hUI{)CVo<K%0-ByI?e3*)6}fnkHX%b|~%UtTMGm4kcLvyEpVmAV<1l@?X4e0#1wQ
zzHJV<R+he~KY#@360H(`rQ2}Y`1uo%!!ZLJR`>1*u)k>}(1rC=Lwd>htx$Tvu@r4;
z#@`AGLrQo@*$7GHIj`=cFCcE2QK_OKbs@n3ZnV6}D@x(|xwx6!r!&9urOHmiycNMp
zVDPmq^z4Bm>wN|*uv%LEC6@$9E+QzvMjLYQqMpk8!q2q*Xu>tU_$7$D;)~Qx3FL=6
zSSNu2Ch8^<bxn+HvywwA+7bz14qon`gzXl_6>dQMiZ<HDo6_?v($CezwyPm=o85EB
zFOmwIwhM4?rz7!-*{A(K^EVpoa7+?EXtU})Np5U6b|o<&sR}w+bg_1uifw{Vt1teW
zJ@mL&fu~7#1e?QzRJB|zZrCnIr*=V5s(;0RtYQ*D@3e!10xskPyn<;&at;bQ08&dE
z%>;Tb8aK?WwVA+uueg4q(6Sq%NeTOwJVF?g&#pmv!UW%0Fj8vkpKYAYE$CRs0Ts4r
zKU+-B&MTKW43^3L`D0dXZIo@e<DY>&*r=w%yR$|BgAsxKVh~uGo=ghdg7-?J7Wc2)
zF~xWdz4c>50^pZARS(OLir8T;D(RLq3-H#oIo|*xA3TeHa;qyh`6gZU1`KL7(-}D=
zt;bF_dt;ov5QR!S+n>)2N&mtKn6ezS)WeIr(!jD2>Hrv|=GXwK_vEvv-}I|U{A06=
z>B(Dkype^}I^!DUBzWdg5qA`CSMkjDDtDJxU4rH?1jjT8)8RW#)$r>_c!To?4TMWK
zl$e;gFkNJQ5kYm5^o`qQqU!SKxB#Ep-Mb_PZ09vr>C8P!%h&8}NJ%iseK6$MlV0WD
zamnZBh>%iWz2ZE^SA@~50=(7Zxl#ITj4uW?s%F*+Ve`)Oesp8#8D0rXha7H!@#T!T
zrrV)C2zx!PV3|ZQ#{-E@ILY`LA$zv=Kq8AsN`Ivi*@Gpyl9#u|9WpaGvAlQKg6Per
zd0rUz03qu2#B1XmpjUb}6ju&bwflqSkT{6*0=!Wsfl^XV8*ZPf@9*>oL63i4X23AK
z!ih4*{Mo(cb}H{9Lo4qgG}zEl0C;lQ?(XMoY2=#)f`elU9t9z>&(aM3y8JUES``P~
zpmFDyB?1<$@9L++IdvoN32TBlYnjNeR76=cB0t~tL18NO{-ZB>MV&v?)wm7ky}|$t
zirg*}!D`rfx$_knzk?5;NQ*ZE%=)-dLp6EqZDL<(h^pq>V5{O_tC-U%UP%1eVj1n@
zv(2LZ{f3q_o?SB3@|J}~*b0Lhb|@LhK717WMsc5e-qmc6z>6Sv5sMgzMH6X=ZNaDP
z6$Sk&m|>pe_zi>p4)B!fP@~K};N3GH4)L~kG<)hxQ4q{3czg9m>gc)Z>uC@dTS4$$
zeMQ3o5SwPLvn4gLTsxD#Bk+>Ef4+9OB7Ll*+GZNokyfW~HpTNi-d3dYCHSm_)ZG<K
z0Z3x?M>;`vXF3v<>GJ5ob7hGIdwL4lXTP-k9qwIRXFfKG8p3E2QSXOoXV+?l9lDC<
zC-R?9fec<dgW!!*d_^&}Ell>Xeg7PHoaSkNd}Seep4`KSLEr+_XGX2j8?5#F!=r1x
z7=aIi(eB%f&5q~IdT4x~_p_KSV(n&y2fYd`T5GgNssG?O7Ty2VNTFMbVbMw3<W{NG
zr=?Vvu|el0N44e;?)#-$-`*vO?(z~+77dNprZ5p>B5?-ou>o~FQ(;(M^D{lNn+1?c
z@;NGs!4oDjt`7UsrKO%Zd1&zpxw6OY2Omhx%vKdxTo>?hk1i$`B813|;Z*fCSL1}0
z9K$Udb?*}rpU|~OBdyU!>Ock}k*zmhSMU&OEG@f~W5dN&GS%WdC9nV&-jGWy0cD}n
zn6o!>39L@AI)3klA(uB@#WP)=+9I=7NYdC}21>VfCHfT*tjH}YbG5VJ-F$)IadaWR
z=(*ML#Z30dpY5rglNv=~SjSEC6bs4ZOWM(jPRmmrrF+Nzo-OiJKBgI*A1WyJ&h_cI
zNCbsbA6U^l7>qPnXAnwUmOmi}72vE&?jt1)!dtuxCG`V^&)aCm-cT&r(Bn!Mu2Wt5
zIL4NH<U*Hys;qp0cJ07!4wVeoJ#Y;nIQcZV(GH%-YBDmo+D>!EMXN%AzDl`Ohnpm#
zWcL0D=9rd+J!j;z!aJr)y#LH_R-8=CSDWnv<!>Tvjg<*jXzR~Erzc69$g~Vk32oAK
zf%^4Mcjv<S@~R^a*-(s@&``a|@}<RCf%ILnKgQIR+B=2h=BiWou2%D5%o99svfOzn
z#4y<$U~Lg}307?Q4TR+}`_f*F4jlhlnLI1HoZRf9Xf+YLB2hd^44Zb_l*GiVMVM1y
zxUut`qN!7JB<kS>qYqdG<Ww-K#0)0&c?BS@9O1}VK*ympj5rs23Kr}%cFd=r(CbKF
zsDwS|Cd0t?=Gj!3&aWQ*!3gW-7yM!j;^`j)L%+?^1ER>aY@wQP)fedX%FraJLY%GN
z%4dh6Paqm5S2oH)p)IX{-c5(j-SB%#q{7!)9$?riCZ@((jU8p&?xqK=wiqwet8reT
zZ?#<inh(B0nE!Dlo8#e?6eR(k0b8MUmC-c6+lCvMiknPJPI-=I4`RfK)d9(C;f$an
zA+B@*8T+7YicMQYZw7&tH7dyt*ZVix8R(sDvT7318Z>_zY{h?~uea6o1jCN#3apzN
z?XGN2ki_EPxIF}ysP+!ftN6pk3BpIrX5F-RPc&8v+MjINejuCR_md==@7JM=rsXiE
z)Boca3*&Ks>-<A)Whn<2ih!^X9HV?PgNIDgn19pnjDpg+ra=G2<u%7tC;6+`*x2+O
zCiavtU1<52ChY||RyvhP3}&FGbcQIlQuGebtjI<8Alg#Km>&3s(M<*t(li)Wza^uN
zxTMd}8}=CwmK7?3?K8t*Z4>N3j8`D#+t_o>^8rTBpiybl{#@mSg7e~?&^3r)OU%E;
zlwqGHHy%9jLzM8<?cfB3-b!lKWpK0Hw^p~pwAny;%Y#4>!Mh~hOSm+I4yD$lZMu-v
z2yWmJv_1l}Togk*TR%Nxbn#;HUWbs$`7HJ87@d_L9v4K+?1>4Ib`=0^yvc?E*ssnB
zr0;u$i3UR#6|<|;tsPEAwdJ=Qe<bypQ_Y-yOxTrBU)<Ekk=GLAdrig%0kjeR<k+jj
zq&j&&v8XR+v-F;B8NtBgLE;Rj%V~#)(D<@=!`}{K-Y$Z4o!>-?N0&rm*tVJ`lB8xQ
z)>gLwuBd($)i}QfP8?W%TlEUBhZs8*ygXqFbIu$ZCdQq!pT$G0Lf9o>vH|sZO!Bt6
zeXxDhQP13)Fq;YMh(?1^ulAg8-J+G7mIy$#AAm?aVeX=(I8MN!q<1efK!Gc#nJ`Ar
zK`7t=fxHO*(&G+4ie@`Fj8<X&^r+z#S+Bz)E6b6?>5+#xIP#M6D_3K^rC!|*KGRsE
zIsP2EKTb!Q%0Vi?klW9R)C5tuf^If6WI%BMI2GD~c7HD{I)u2>C+lGG`smuKeA~Jt
z+X{=QP&Qf}j$590OweH(28&OvTL>|T%_oG4DrrpiMKdYA(D|Yil%^9LDa$#ipP16r
zRD^@jv|M_mj*IwU56GUx3cnwGg8UX|=A1F84$Dp=)zdDZr!~(0uk1oWuziIX?ln_b
z%`o9{kmjFSYjn_Kw=G0!9R=ZyolQAc=T2lSeBM@S{8+c0gH6rmKu0TmPTEr_<6kEp
zrojfUS!DC-Ixx`nd;!^k!3)-zZFyA9)+}G_F7FcvCBp1pEh0VOiJ61;9GTFMPTI3}
zbrm?MQmH1=VYKAuF$8POFVbGM1H;id%ST(88{X^P=bWQLROVHWu;3?uq_5-Vck|%v
zKCFsjh@8(#BUv0TRufx}+0Euz4F&h1eWEzL=oj%?r#$>*m@*cNOcDxLgld)oH#4iE
z>%XcFSLx$6m$>8_@(87?)D15&6oQ$EQ=Z{D+1Dcr=`NGoY>h5|R%kDPigpW{g<L%3
zb(vDo$H;c3@!rGz*uoIQZ_y(eNnHBeU8DRx8vynhb<?(hsceGzO@3;V=P-HKQpCcN
z1JQ*+MvY*Cm#MX`v)0&3rNAHzDr5627iP;4@O+vvsg#JrPiP+V;EIMX)i4+Hu6SUd
z5jAc|EN7CIu~8cMe1X#C@@G<pN)3T0d`lOV=<LfnSdiOFeJ)k;uKX}jdW!bsTXux1
zxrR_$&*wirvvb)v!)bqfx}^Kq(o-KcZ__q0{>_zOPK`3ASDb#cSQUJ(0G8(L$PAU>
zuvdRW>I^i40;p?*T<ySwT7U9b9bp-MjV4$6qzVmh2s)=V<o`CpA%iVCq4{!8(z(N~
zOhrTYGTPfzkbq={=&77m19=zcN=BNxh8wNrc?i!u)trfk#;`|@!~fZSLfFvc4)deA
zAVU?tRr#K82pWM62cg8p!4Ja?xeR^56}Di5Qng@yW}^?1$D?iM@7ql(pv-P#yh0UP
zQ<V5(T2Du3F~`Pr3||oDBJj*;T1AtoyuMpcC+@xG<zltR9YXt#&F3K3u)#|7B1IUu
zUzTfA?4!(kcBDSdKpC$4OZc?50GQEA&A}iR{df}$Y=~anEBS};kAsSKh!9K1N8~v9
zO$W3xs`rY3&K1=xCIM{rCGB`&QQQZGY}y)o46kG0F<nVd;Dg-`9h_3R3+aS%82A&c
z3!ur-wn`Hf3A~{F$Y3T*<jt$hN!!J;Z=0?k;5+FxG1KV`fm4e2RvXckiATDrLmpc_
zRE68B&Ob6+-;FQfsuj%zQ5=#+zba%T;{xi{|I9_4^GW|1bH<#z;8qcv2*F08mPAFi
z&!H>S0zTsD8)Hu^`@P$bY);8okJ0QUB$H_>ng?7x2UCAUfS>N^f?3pz3m~-I&Y=B0
z&Dm^nP|I>-rkC@0Avws8Z@P29I~t*%AT8LU<H}dSn0|zqr9wO;)su5agP=KE1UJue
zTxvxyj-Zq_l6X_#r&7hp@W$ZegN_e7fCkRE#PVQNN|I#!cQ_L2fMPY@J$W)^<Ye(^
zic1!d+IS}5MO(Q+*C~N6QAz7zmpZ+d{q1~_ci>|@&V4L}Kriio@X@z&^PNgK9?D_!
zN%vFduQ74_K2OFr)w{!z!6jm0pol77o*n);TXvkEu(<lzS9p=##@<Z>!}JYAz%7L(
zpsm_*;=5Mk&8S(3vTnDe2Q98M61LeLDoF^dC@2s9b)`_k(bnUn`kSN;J+i$b{DyEI
zG#_#7IC^(oUVSO9cg%}Wo1NbV;d_APf|Q3MgE%0+N{4d<YCWr<215FtU{9Dx>stwP
zG5mps%xXV~?*{A&3fGyA%iwe+$iyouL^Y?dTC8j9@I$|%HCA>Uyrf{mjC*qHNaM|E
zqp3)R3z**(`XsYbT2W`}vJI6z76&!I=HdTsKOkg_#ub324)WjEj)+(%!3EW|L70CM
zY8U$9eBPhrbxxn>cjSinmNwc}R3p)qPk9&^PMbt>8pS5{@2Aa7;UC*-{nTY=mC*v%
zB+NTn3A8=^Es@4zQh?Q>_X|iKtTN&So*{G~dQI_Mq<S=eok)hPYCNBMcLE`NV>teh
zXL6CbMD_dcH1`)biNC<K%>1Pq{5G<251Ue_Qn_F;?Hyn5Hu?}(2;8^q3CrlAv^TKw
zG~x*fzilcM*{Kpt{2;)rJeAuTvQyE@i?He?2-mWY7FYzMBt$}YNnBdx`_-#s1wSLP
zKZ|Tvjocw9K$}2Sgb+F>8)Z=X-X$M&)`$zytOP+~<7YO|SRD02Q{hHY3qC!_mejK)
zg8f-N&FUs1?9q@9*X6ZiCbzk<?+#=lBQ4?+-S}VV;Cq5dIx>W(c)wUL%!Cr8zSdXc
zn35Qxz^#bH;z1dRW<#ecFjla7pIGu(0|ifCqy~w^JO)_8E^-I9gUts}_~Q?j>LLt)
z*f7mlTW}#agHfu{5H0#^xnheQLMTn6$cCCxUI6(v7j<*qG*`A^Drbo&%%Ue?axtPl
zt<7lV5Yw~Fup%7n5P@6O7K*~1GXkkMQ}3G4&e>?5&aqXgrev)<JxI`LT#%sJ$WZGV
zHm|rva~O|EcP|wuiy-B^tr?NDwLLrI_Gv!n+1-=WfnGl$7EyPQ+Ua=+$4+iBE0{~N
zz~4?L_^4Qo`>W{;ZTLiWeNB^-e)lJ%Gftg5D)_GBZnv;KF=Xl3v#*hk@0eXbFgP9J
z84)@VG6_9ilFtmE71;k`4ly-oVCdPzz|YMN?fG|Ky@|@t7}2c3?g`?PcyNTA)}Wk9
z=uVW=s#g)Buj`w(1SJc<ZO@y<`u0ZkSL;%$BK|lb8hX(+Ih}mO4N<~Y-IEqptcS6c
z_`cZn7bomuw~pwopx*);QzoIF&o+o^xMl)x{e-;i=qM&)wZIVrTh@S0F5=GyQk@pc
zk?-@<o%IIRpWK_6(dGTPL-u6iE`XW7y%CbgL~Oz!jkL91+;{>%=a7o?tfxB9)R3<#
zi@q#}NGU2}WGJy8`vS*Y|0g;F#r$drzh!5ut$?~&%~tG#1wdOBEee;n%=Tun*F;_k
zx?nCe{y^<k;y`^-+AHvZnw_n_^!={h|1Nn5TuYqf>l6cxsoS6I`Ov@14?P`67D<cv
z;+N;_C$LABKl%dOW%nmPM1h%H$)bt)xak3fxcvr#x38n`HzrV0P!sv+*j?^?pX>;`
z!i=pJWU1kzlPMq;YBYZnP0a%pax;wmGkQeLCzJ7z(Oi4WK{odSyn_E;;vbaj^!RyM
zAu?dBB}X-9-fc3<87$T*EQ-%E!pB-0&tOmE^tMPxlbN~Ag3AX$g3^JzDS4guq8_=9
zupb5-+Qb(h$FDw1^mcA9AqB0EY--h@x||D))Krp-;Ez#zuNEl+ijHeNiqD=naSudT
z>208-l+0vfFpjr_fHj;W8|#XXxb{ZNr<jKga9OCN(c5iCOAj-TpKFrWkt082W^4dB
z0Pl*Z*RZ&I3}(()S^2^f#Nn(fy2&%RX1mi0(|c#G#h>zkThQDevDUler)ROpi;S-;
z?NdUwD_qbTzV&bH)d+yHtp&_k$gF)wqBpVhe`WVTs*B%AaXd6S)eV|Gf=1F`bANyM
z#DWbTAd}Au2rK4ll;dc?DhR6pS9e!%BeMyGG@izkIA&A;*#=nd4kH9i@&3!xvX_NG
z?3H9aPP;Sz74~aPb@}y_?S?@?;9nA_AApuDi6*nN6lPtDHc~SC@1FmaHLffpu@ElI
z+lRIYVAOMs4AT!rZ?FZeZf`_wfW$!==siFkey-LZY`f10oKi5uXonl56GL|hZrHJm
z^5tdR8ZKvPbEcrEXlr79LLi<_UFQt&V$)CuWfVzlecbmZG*>6lUk+pg`pYsyS}uEW
zZC1BV={5fv(=cSJVHWshWZ21$W(3GRjfE1J%#`sDxI33Wq(?_~?jjNXtQn7q`xv?i
zNEtK830(*^4@9n9qcu|uN~q(HE=i!mFCq3sv9rQ-yV(3XnkvY9T%OFCBUn2|Q!5rL
z0HIG&J<O#fOzA3!%L}m69wU_xs3luv>cGArm^!CR_d%?<xBWeBaO*x|8QtDUL>?-R
zvk{vm^y#qVc@ASd7vCdTzsH~J3T19&b98cLVQmU!Ze(v_Y6>zjGcX`9Aa7!73Oqat
zFHB`_XLM*WAU88P3NK7$ZfA68G9WZLH#sm0FHB`_XLM*YATSCqOl59obZ8(oFflkV
zARr(hAPO%=X>4?5av(28Y+-a|L}g=dWMv9IJ_>Vma%Ev{3V7Olw_|Xo-PSD}8y!2j
zW81cEtHV3C)v;~cw(X9sj_sslo@eh}?>XPM>(r@Lb=B0EW6iPFk9Cm~E2+{8o7(?0
zld!jQp=V}b;sL1IyE>VeiQ1c*DLL8GtC(51+88?l{z<`c15AM?E`XnA7C<{V#(yCc
zjBU*TjAFtHvck%=|97CNnfZTerR~h^0bKuSVe0DeKW*I1oScF7b^xk>7^neGX2z!W
zb~c{>fhpL#08PvQRR8G=06GJV0WMC)re?OrPSyZ>bHKm14p8-Ub}_Sc21whP*gHAc
zI~lu}nF4@z02fO$fVv&<U!JPVKf!<CcBYK>P5}FVGEM*!dsjObC!m=#!+-ON+B<kU
z0WB<D0960SWMN`vqWjmE<^Ms;01&pZ0jT^7;0#bPb2f8w`zQF{{UwZTfi|B17Vm#%
z`<HBhe{}orQlPU0(8J7B3Fu;C`OoaGX8#FGyBOO5O@!?%Y|H>m{|TwvnVLD-0PW0_
z?45xw|B{rRnd!fCYL-9~YdbS%X8`+uQf79h|HUZ&&p4()JBxp5W@l>bWct4Y|4u6z
z1MOVYJRSZw@4wFW-#+vI@#T&G5!M5s%f!IM#Qaa;KkI)i{r{p<b#bz{Hq!)}x>){y
zJBo<ddjP!YnYmd3^la>G0M365W+o0M-~T_{#MQ~k%+BTCm+GHU{>L{5{-cYTnTMGP
z-1>^W32(4fdRB%@Jyzc6l$cgKF{C)F`$+e=Rk-6BkVY%$r@;y(i45FdOZs?tk=zHI
z5(+o3H(8&p&J3}GpR(~G#BkJg`zCOl6`t&XYGRaJL8WfES4+pE@&uwbcgruKTj-Ie
zb7&1FBl^Y|utqwJXXS3pSpvy^K1(ABxW2o)u)CcA_A-D_e41>MRtXGP_to$=|79PO
zR_<7dU$M<Ea8R}(NkCOEmudy{k5jPhA+LhN2#OD=S0iQ*?7k@xOm*DYzFAHjIKwM-
z!)PA99z@f-x5{IvWX=ZU&Vc8l64!=%AqCmmI<!kN)*R(@mOzSOq!zk<`y5aMLfDpV
zm86thb0V?O3kOx!L;mW<c$kQ&(b(T<po5K#b>&mfwlF8{Y`Ig~d6qHiuT(CzSfsnu
zZp#2d><1=&G}*jgOT9%lr;{er1~H<FSR0uGwTvq)1k#bhf|-of>h%mj={C^;0RA(2
zG2F}Hb$tL5jkF!}XDg7%tQw{#_cqDxJz{-)CYy;E5`~uhi>YmvTX(5UQ2B2>7M~?O
zP+f%iTjMU8hL^1Iwt1`Q1o$FNBNV{Pz<T9hQo7ieUn$NVVfdz9=4IUlTC9*caC85y
zju^^b@RUjrtntksGTb5J_gGneE-kS#8Kh>7L5C^T%E7^Ch%)O&#MX-}4I(>Mo(pQk
z$voxSa$bi#JYV*#Qs&c7SS(_Y3%)XCCwAjhPr5bJf8((JKt&O*nN*?gTc?AWcngjg
zs~?tDoJn>^_oA-ZjtuJ;iqfo)P#P;Mbdny_-R&r=be#OxpUa@E;_thKdMf&2PnzcH
z1<gn|WUHQZ>G3O1TYYpUU3n9$Jr3!NWE7pD<#3LE%1Q}BX;1|xBlnBm7xHmcfhPfR
z-s;^rOl&sI!)KWeHTd{WjP839$0J9u;emUo8NlJZhbR8gn{p!FjP(J1bP66{WGi#Q
zHTkwt3t9UtFg?&n)K<>=uudya@5l|HGz7m&&*CoLj{A-SzEU#gM8fIDUMxzdbbza)
zN7eP)m>&?$^l{(Z=a#zC+@#Bids4wUwIR*I_MWt-j`Xsy1*1bhqa+uBo<_rFMyc(c
z<ieI9j*DX&tpdSkL0*r3NtB|>tZd2eZ{gE%JEoKk(-{=zQ0uje)Te9jSIoqOot@Rp
ztaW67E?+Z@7ljsuU~`3Z9PScu^sHX9+nHbBtMB=u5cgC9SO_BMR%n<6{e@AP|Gl8>
zJ#95zmigr}&)BDCts$)jZnMAL{1{-Fq6_>%uv0CX?b#r2?9gk&gt{3Q@A0Q*&#x)%
zH-N_lO6I^(;5HPja>+HEu7kRyD1q9i(#x1E*G-R@1|VcX#eAZe>62q)7h0La2eY1f
zJjp6e!o_SwUo%m268v-i8@yZh!Rhsn!RuVJGwb7q`eAXkD4A;$j1kS*oyiK6{!Q||
zmB;c)3((!tU9xG3fs8#;??Vcw&U2_G3jX>~Yur~OMKzSDh+E0LMQwXHHt&lVrQ%aV
z*M-aA&LBWI^lBd!D>8Ms>c(K{-bqTUtU<gtByGQ9`sJrq5i$xMK8X>3rtzK{g#|wN
zPljR`%_4l2@{Du_NB$c!<qumbb>h=GzlUawRAsFUCQ1y9kDtXHN|%}Ul7%Va7SS~^
z4cUa8r%S_?eoJ#J;HzuFz)1F7Quq)Z?dYkSevl6dRoXY#p!c^4%3Hz|#J0hz?XR-s
zn*}s#kkamFsHoG&N58y#c?o1JFwK(%6pLwF2~Thau`B4>er{DJ$b$@pv%8V&fXWY7
zsr7yxbBJvbhDuKND*<{lIm))698<|;ASxWake|;HUyF)eD3K|Q@DKLuo-mB>p;K!&
z#TZI(MmKJ99fbG{hx@*PMTI6w@7WXujbyaMbzb{+ndWfLwFzk-kA3jQu3#QpsV9f{
z1|8mIka%4*MGx;M)eI0{79Fg#Q{;k-p;Za-3Pw|WfjG!-{ct&?AB&a@Bf%;b%$Dod
ze&-;&3BQzFmNlVguf=RW=eAdRE!5D*8o}n3M&J%f9%{TcD;1n|zB~*-i{2v(n=203
zI7JkJ5~Q?2lz)lZkmsxnKlps5Sb2q)NkZLiXiXPcne!<*LAQ3F=dL0V0DEumMs8MQ
zw`2LrEnQ!ZD;-oan{DnSuTm%HR(sV>+yT@{M@K-*AGgwpvFSIUze^SDL^}ppwplM`
zhw@0E%%B0^K@6|FVi=zvmi`Ka5nT5)<a_>E{?Xmr0Y3RAbq&Q3?+lpPVPc@kSWNd{
zDqq;fX@g}g*pLxybE29=0|!n_Af9Y}1)cC<C88TtNd~S41CVOqq`~$88Oy5x$v;?#
z5(Hh|YMcQ(zro?D1`fBeK}nAJx<bsxFWWiIl0e7RG$16i-Xxk!6s(m3n?vt{6O@<U
z3QfW2mfTcP^P2w(Kh@<FEH<;xYGWI`_g?`j-&FtX{(1=|)|fNfwuo6~uL2i2k-5<h
zdo6pPX0m+%7Q({St|!%=4(;B5X9CM9DfHH%$=8_XI}-Z4(hCF?xY)yBq(6%bS69ge
z@#YrLS#Koi5iNz@)GI!S=5sZq_D4-LZK<~OoL1W@H#i#{ojj@CotmUHTur6Ws*{>V
z60>Go04Ib_#Gb|30><1E`0u;Wz^01lmtRT7i%rK}u<Pn78`E2^qCq34W^foR?TQwV
z<RnBRanlFo*0SYh`vJ<wQc<73bWl|D;G0h6IjV5?RF8)6vF9Ot_7Z+}*Yxc8G8J=1
zpSMM2DR(UPn-|L*=k62YBR>xkQH&d96Job7aNjI#6T6EE1oZzgB&v^Po}_QRFURe?
z3u;NDJ!lY~aH2fG8l2BAwU4bG8Klj(_f%!qcmF*whaWV)a^hG5!3}eV2FTA@C0cC#
zk`s?5kip%_st@L04-0@3H1<@Dhoe-yJ4k+fCXUtX@Ub*Go5Zi6*Fj-iTsw4_G)j0X
zSilF}a74tvJ2Vcq|6|{1(wOq~PKNy<Fz~fO9rAHPUN_pS)aS=$X6>a!&}=DNMY{#1
zWuP<^H6To!qZK;~G7#h<;HQgTEOz6lBfvJ;rn}bMCe&9<>&RS%jL+vW7bxY>*BUJo
zV{YnxzDnr^&@RIRDK6c+zifap8gsS+WxT|d|2Y*pya%!+TuuQ*tX5^3TUlW*7t6C1
zhCeBlzWn+%l|vWT6It#9l7;h<E9kTO*&ejhQ0A)6{EC#eMX<cZ36P1!IZr$JcuL}&
z5XW?9p|{c-md0y(_OITk|1Nh>cJqD9o34$Bjc<FlG#Fx#PqCD;Sc1@D|GJ^OD?3zI
zyQ>0*)>cRyAX}LAHbayeQbhS2JG}t!&^!zCdl4&T)D9&cA&;qN9-bI1@)kvidLAX6
zjB4l>{6JluPBf%eQ&P9cdIb1%X*~qK2c3Ke%D>M-OlAekyGV(wxn|$j*3AHuSXSEx
zUR=UWQ8*YPYc@o=I1gz{Nh>DO@Ah!c@73qVyXmhfVRvg2T%a0oN^!Izka&bfm#urz
zKt<E+Q$)PhfR|jv>8a?I!D|0DR~>q(M(NfO>95gTLW*kW;|!Bv?1LP*zvSxytkQeN
zjZkwp)rH+sYI${%j>j1U0&-3F;56Zd*ba`8XB_&_Ys=mJ=nU4$kY(m%ue`jwvhR;>
z97zi@i8&ICnm!nHnDWxq%srtJhx6mI-izfxgES<#RPMwhe2$9JMAF8KX@PnRoRHc2
zod?j&{DgArxt-`qNzY?@O4O#iY=aq|xkZ*_1J^ifs)@9U^D7!`xUXjKu=2ay&ft&J
zHYhtTDnEOlfLYfwYj?bS9(|i_yc_TVA3fUW?oIO<Y+R7WWvGMcN->&m#e)x`*t)8u
z`Hg_bACH&CPL)Sn{)-7tIW>L;*1X<9Ep}+XeRQmS9GN4#gW3;LHUjk9**bzDb4cl&
zCk-0W)Bi5}xhrxgQMr+9)~eqMH_v^OSsmgw*iGWH$pLZ$W7n*NbM=1AJ%YZqkT;7a
zsvs8?0M!nM8OOl`f1C3i-C+xww{gnVXk-#Y?8J(0U^x2$cdmu&QYlWb=6?T%XiI{C
z4A%`?G=;BpEG$!GA2|(5AeOcNnS}9&*Qok~ajL}nk9Ta+Vx$fsczjl$-zZ}kfjkLx
zSRX^-{phw+cRi#sK45>OQaVMO(Ks|q==VI`36)uMB%G|JW&qHV8r_x<HKoquZVWE)
zL86sh=Q4rEOQ{_tS@=QRe^pG<Nl}!OA31rih?1?+%9yp><{L)|osPMt9>(xc>jljq
zCq3M1Z$98Rd}noO>Z4RV@u!t5(BCLXb9zmfhEZZ((iIwcd<ca$asOV%DEB5JzXf(b
zyr-iNVcKyB<u}P$9~)l)Z>d6Gi%4>lS8z7;>IHE5NEoDxEoKmmjAyA?=}C1=XmJ~V
zLuVi{PCL`|FI6x2JA4_+Sv8bAq!h6f;I`U?T3wMF)};}9VQhIzQoBx>0NG8$GI;Z&
z3q(Sa3wZZ}ZE)L-ykwhNF?u15tgL5@fwvAC!C?7CG1{?5G5J<)4!01j(tzd3{3O|f
zaWbw5o#CiwJTp8Z?0KAZi(1tGNu56wYczt0i1hxa&daiH_oBM(qM5U}8cQMEo;@EY
zVomsGjN4#%9`|pXrS?U`D`0EF`(g<`WFT!F4s28d(H6+G>%5Em2G;<cxur-FhzxK_
z(|%Vwl`?fRWE}FvxmD|^$%dE!*oxj6<)^Yo^H!?ag@_aH`;|DISeY8_aVF9yy8A|H
zgT>3GBtwv9M=*GbOtB0eA^ok|i7H7i56`JVzS%!J+_xqPoe?4TSsJqgtA{yoPA6~Z
zyk9ydqojw<%(Ty^2q*VN_SD;$BMQ#tYBMWlxmSD&iL|CpUPuTe!<N{CRV&F?K4^NP
z9j-P2U-R9}?sM!J^bl{V=v>&@LbPaP@E)^Rjg!Ekz#Zx^iX!1N&UsI=MebqXc2vOD
zCQ3J;D~vp&u^|3tyfgRz{fvV{x3xAUr31h27pqV86TvFMg;P%Oh0t#jN_r)d7<>^~
z(>$<NY0wZ-Ox8SAO%$jbKxbP(3lk^sBG`9=P@b*)I+!6b>O<4tqZ-=3o}_zAj<N`?
zU2aLv_9Jqg6V0ullT!1@^DCLYWum3SSAZ|t7CP44Y|}Q)JWlV3(f2oIKxPV@n@G`P
zQ!XRHEcK@aJs8@Rd7FeBS68FmmBO5Ly$O6PRD)X)A63nMNiX`VMJ~exxkS$-VF?UO
z;7zkG^<>eyCT&`JG}10(%zkG3w`u8>zMtVT0THpW>AU&~0n)KQVT^WS?(AvoF)=3w
z+)dfI#z$`qluf*5>^EI)B#}~fuJ=VAS>|g@`KV{5J-<DEn3gjdsYf^JGQ!eX?Ro$~
z^0p|&-(65nStWg_^7d<UVc=G?X?6S@J;&*Y6izQ_KbOO4$#M?ISN<L6HJq4E>l;od
z5w(x*@Z5~?aWx@{<ZFG;__>$O(}6cd_0DmKew=Ro#`Sd;XRci|5L;T3A6diNg=cKS
zYs@C9mR1}qi^y{N8}4nKJ-ow?Vv-BOXUpLH&6ZVX3H0th7UlEjCK9>Dpfs-+7#UL7
za-BjKA#*X-D2AgygGO*CiQb}h)f44NjylotFkj|4Ttw%=&~2I7chJ&e8C(`LNZ61h
z{@>A-KetpK03KEr1}-k@#3su>Fv50AP>>r~3~2pR$|7FA_Y&`Dtn)Up%&uWU<mdhe
zYGiF0N$o_LsIf}DY77dd35&l53DW4l-OZTek7(jx&ggp)q5}zlHB=IKiGNGR^cN;n
zhT_8&Grh{eNnzpLf3wIrgXE5x*hv&D;33e-|1>&=l{aFi#^mU_{$xFry79N8ok9=9
z$Mum?X49Z(2SvxlF^A(NSSMra$un`7QT)Ys(oH6^in{ImhZ|=m@AmKcxzE~olk^*A
zP5naHl_emx^^c{Yu3vj|eK)-bC%Wf5wR$LBUWZII={hr<7_Q=0W;VTjGnm3`07XE$
zzn~se7(xo-5+=^0K!#EBH6wfqj3_x}!E(>La^^;B@W(aT-G_RmJJM=Ry68!bJKPet
zFs)MGbv%2-#^2C+z8(QyQ6V~~RMkT6oueb+zSj<<`y<PLE`2{fb8b{W<UtE{LTTgA
zYPn^B<FEFbfh7d5V5!ib#+M@TZUXX_iX@)9VrNvNUI^^d8H6vdI#S8#L>vuDgpXvb
zGr`2NR;z)`9`#o-1*=;edn1iPwMA)Wo1q5TjODWWVcCXq156`K!QzXPxjqHZMweL5
zz@>|d!XyrpQR=|`HMd0mY>}V1e1z?@&WTan<zY`^tkAmfr{a}u&wE?0`-q@qH#4;1
z4KT$Z_Pq7u)A>q6&_;D)&Q_wu?muVVrN_bW6@5EFiHemssUvxk?bR{+TgirmC~?9k
zInIbRFD_k3FBh}uZG|eW7&TK3U~VyJLoIYlwz@95c}T{??sy-HtLJ#j8v@2`RFh+r
zE2{g2*AntyjB35nlp@9)>@*z}8Yq6?oE<A8oy}U*i1LV3;qxWy%0aj(J`roVVN&x(
zYh-82-a=^qF!=@ZuAD%gTFVD1wzT|!$*gw>JN8s3tHn6~4G4m2-{8ad{lzXfhfTe!
z8igx@re5nr9u7WpYh@MQs#NSHS?V{b?)@#jc^bjkrVnxh%*`gz`l67H)6kzRT#ulK
zg$_A>qRmqLEMxt>cIwe~op_ccM{e1qvA_CG$4;*65WU0S$}CA<m}LpmI*yz?i4%oU
z%pXW>Ge^Tc<bPr6-9&JZ&1qF;V@|Y3U#iSWpH63T?qf{Xg62F>)9CnXgjgYhlhwgj
za|4EU7~YPNY`5{q7(FJ;lo9m6Dc$yh$`^Me(%zUc+r)BW*1)J~Bff`}4=`edlX9y3
z4!easCo6&X#!5}h)+3JD43fEaKF9z06>8c4t$Xi%m|cjPdDhsE*V^$b2v`*%HJ~g}
z9MZ!A+m~aTU(L10M&(y<OB*v%Og!wUOv0ite@JcILkLD9FPd5QD>R5yM@Nxv;GEx7
zr-tMqd_fHKy#_8nJ>2pH4Y}#{3K_~6^(rNqmfhfbW|8+a#!g;1yEC`P3bfiR0scU1
zeYv#!8veNQZeS=OQYm%3U=S(B={?J!f>q94g}d&l<v7Q@{f-dP!*W)drvhY7_ynGL
zmx{0!(Z^u%BO3kFE@-oMR%$VRwc@@}vLc=IYQCjWCKA795gwX@&ZQJ?e96WQXAFh#
zM`unDj<D%0tHoWn+A2+Ftlse_pPzs45zQZz5Uk<jn6<3JF>^oYPQBj?+~{J#i6MrB
zhxEZE_2Ee^`x1IaKXm7-<aM5A|4^TKWzz>%qAPMDj>jTFJN`U=zYlu1WEsU%(np}k
zpp}*$)%AA4^kClbqG}loy(p&(Cr5$Y{rMdq7eu>1<67N4F8~3H1Is}>Yfr>g;^zO=
z*DuE5_jz@~X~FKpFpELeN1KCOzRSj^z~$Y)_jTvPO7s+#KyNe}V*U{l&s^sq5hH(i
z&Jo*6e%p`pH=ph$t8zV~9FS9O1;hen?1RGfqzpq-hFAvygF>^Pj$qPt$ktyuVK`;q
zu4`(mjjW@v!G(dFa@M^Ry%b)H-6>G^fR=pGV~-yL@nVHzWg+YAV=cM8N5p%_kqWl^
zhEOD!z2hj5EHn&UKS}j6A?=_Ju^0E9*Cl+`8Sm`fS{WXdev24S7_c5}km>V6F~8XQ
zdCF&k2s%WusgWn%#Q)f%$g5N$7<^3O{x?^#+^9nKI>j!wksh_At8FM^V)+(cgeWaz
ztgTrS!JeQibby{-^Vj$<B~BI=7w5|njI17Bi>C%xiO%WGp2qp6x1~AZ&f9f#pceFW
zoQSeW4ZXG;Dfs7%t#waY&k?jer67jd{EvFzYFGw#gtPt0WIwwuPTYEkjC4V%SaO(T
zGih;YzoK->5vBj=#W~5eYfU;$!?zzGnC+F+4T2s`dpUMvJgP~{Ml&R(OybweVccVr
zkjW9kQ|LYaMDK)-<jscOmXc8RpCV8T0z6rc+-j6_0Fm})K|!YGg4+=CQ?=@J7`HMA
zQQ5VW{!7wXd&Rs-yLaz~Y4;&|4D-_5T6LOeI+O5$tA+ER$h2PCi{I`!)?j28F>S4F
z<SpIqF3M@mr)rY>{l@sYq>eBa6`Y|=kV+)A_T!Ylj5W&1rYfeRT7eMN!rjTJsXn?6
z$g8>mEEa}?IL%lCkLQnyn~`PRIH(=Dr&mytlxp$E+8hR4FrNyC1zSW3;)vN@vmwSz
z)VNy~%vUAu=Q09PJ4}Q`3s*c#Z$y1_C&mZ8d5jxEmHLjiHO1kQ4$sT!Jp+0NeVutl
z8;4|rB2Ru8cevbiROY%lwY`f=NWQ#%I)O>wA-c8%$xeUI`v+>)R^G3{Z?x8UC%=C3
zhFs}i>ODT#P5u%2@m`Al;a(K$21Br?J!LV@g28uu8}MdqUyrwVIsM@%hKKx-6Kh-S
zWP%5lJCPv=3Ut@8&C1Y$?to%dgh(H1bzQHn1qMEj=JTpY&r;>NFA(A0V_ZrPI<hO&
z2l4Jbfvj&zu_*SA^*w2hny&EIVQp?u%(193=}vKNXVZ-kzPH&Q)=vq^PRqme!j&Ke
zaZzTooR4BPqS9$H76j%O;<yGxZd41k$Z5Tq7nF^mPJ}~*8+>hdL$rq<Xs-{Y#MsJ;
zIde1-E=6Hs6~W91u~?%c65uh{>h48bK~k#-1^lUf0>UuPMh&cJZ9S#!+}SK9FAOP!
zA)B#kI^Ud*Op+Nhgd)Nu6wXlc3cRUTBZ1%wTbw`t{Au1xfq7oI0l~<E-t;2e%PqPs
z&Gby$-?Jn-t<uLrhTD_+X7-&{SZb&Cx&JdL>3>%9j;8DnL6v(!%-u&EFb9z(@4oCQ
z#69@R&Kb4Tfb~i6D|{%pI*xIteFt;Oyjxq2$P%#FIRyf;A`#B7x6<{NQYgXVb}gf-
z`AL*~bEVAB(RzKy0%<Huh?)iNSF35WQeK6HR-}+dKjw$Zxf8$ox$WD&fk+wrR#Et8
z34P8gXJ)mGfu1Hghit20OTVH1QJ`XO8}V%tcUx*b-GGWhtK~Vo7sJum;3`@sOjN)U
zVk#oaR#UUmo#+|K_;4SK-nWkB+kE#-gz{T{uqY93DBs?dO5=FWozs%jJ9GnGQ>qQo
z87Sc_+pG7+2yyhW*GORGLNSw(Xp2STtV|lJe*%vQ^S!+5@spZ563@@+WW0RNOg+%o
zU-s8ZC93CwUSzuKJ5QP8L2#a<?D^yvJ*MW?b({NDIN1)`v|@CE*A@`nfTdNjj<hdE
zHdg2FZSFnf1vTnExb`Cc$S_Xd4o@NS54ylU)r-+Cszk}lo%-rWd;u%l<tu@<db~i&
zfORXqYa13I00(pVx#ovg7Tq_nRI=;rx8zkNevS0td-Y~oe!r0bF&jRUjGz<eN^s{X
z4`8~pyy1IKN_n^8kP|#F=v_PBp9y{fQX9$?>@{PR`~COfKRR>ayiLCl0_Q8KMGMsL
z)^NPNbq$~282Y1%bTgT54A_W{75-LB5{K?RPIYv)jE2F>R2NyM41w}rn$SRXf8Xo-
zid}cq(}}p!V4dC1TrI7Xt)0W>T#&ldiQUbk!WIkAFK-<h8TTVm$MR{xD;ydCGf9wK
z=-;o!67r5u-@?2i1H+t&av*%cJ~!XQ)ZxI_qdy={Gv$vVYCNT>Ey~NPr_xItNR&L1
zzsXAvB4&6oG8`cf^rs|b96CQXXJf*1AcZ6Pdeqg4ucRQCKRI2gXMZ5K^{wa8*DTir
z25n`}kK-j<R$cOJAaS_8^GNk@@2TrB7L9@x6+PY7oCR%YQjd5H=13M-3MKDGw}RDk
zsGn{L?{+X%48)zx^aHaR=vW_Nxk~3=t=lnB-<`4G{hoXcRJ6L_Ry`(pn82%_>=}K3
z$nhH8AyhH0^V|~bgNyt?QIcNp7@h)oZ;(TrEHu6DV9%o!bk?Za2hnKyBf@*LX`fjQ
zFysCGjSys6F(-obqN)0-Ao|M@T*tP1_fJa&Of0V-o}GJkBa<MOvB>0;fcsf-aO7)D
zo&-=%S@vaV>?;ykh|DVro=*KDFnp@dR|f9+2z|Eae(>UfHotd1Sn3E2PLYlXWc*Av
z9`UV#kHZftkuRGoBLAq1Qp2cFGM9Y3yx3?}i9up7a~{GIeGkzYoaUoPTXBs_D+OC`
z@-5Hqi;MB_z3b$<haSO{pWj9~&X&5lr<nvS!1!^UOlm~DNxZ=)*S#-H-yv)~uF*<z
z&2uGAC!$~-kK~tv*udJ#^YtznTgh^&={K>3KcQq~{B_of1ofC~h5PT@idLKP0iTgU
zB-iJ^k+#XRB2xkN>eO%C$f6mnUsmm5$IdZgx}dRqRB53NPY68$67HsH-ag6XlRu90
zT&$7gc<zfj{J>G}Q-3VkHh*;3(b7mrt};#|P`GcrR#7pxVrn@uqd(`!x8X}NZbaZt
z{!QMs+EeCALMAaU*`?Skl?J&kD*4Q5utJ}v#KLxW(oDFEcHg536I~a%NH@2(q%88C
zOHf_zFXK(bCbqzq|6^ESkNd?^EL$@(jsp&8WOp+tZ>Qz(v4liW3363vraiV^)z)m}
zBU!C$;apq4Eff;!*R5+-_#FyMUS<}H{tKV$8bhpor+JzggYjw>V#H#Q?N<&hcU_6+
zE0L-!MmZZtZ;h5N30JHiM`<i+0M`_{Yd~~~70rwafx<#U#)UJ#cAW2QRa9@``|r-K
z%);Dmnku|4>AQZj(4h_=L2CQ=-FZ>7MbF(6z8XEc`tJeihdwq~rc`y+Cd#$kztOeg
zeDs=iDJfg=yB-RgBg67rCsdp2C7E>dG3aCOY!zl_{M(N%x4gXbag1$dAp~K}&;{K~
zmFW*fuaBY{2bNa0EyX{>j2TgDI$H;ohVJv)tlyU;(fW*<c8mLp=hFw(4F4Xyh0Nk5
z{f%O*m;`U_n;G<J@q@UY%aWR(#s>(~cloPx_3VfA3*E<-xJbiF+uv|t4qZKT1?>3T
zI9BhLvjGjGt!D!)NS|`^s|gQ16LgSzd-xt7C_u|5UkLjC%(Xdz3kZ*)2fO)7Y9RA+
zLk`Q%Oq&dsmmbmRluHs&`Qt;hZr4#aWCwDb<z7Y*Et$rNMtOfV^nSd%eX=ME3?t4*
z!9NLr`~szPLB!)Bb+5Y+$$s{1g8apgtL^?)dHw}Qg;t~-^p{Kb+lQzm`p7a>E=>_;
z4PYA$_tQ{;+Am<d^^TfLq5+*GJ%sLVK`xRka8<B$YdnOpSPqIu_%p3qf@vL{EN`4a
z_E}di)$S|@r=wQJ4J;5;GPM#}cS4Fphq2cyK~CI+W11Y9Bn0w0wv8;oT@@x~`Xv(N
z@O1ypl=|fTNlbu>hw(S^wy?@m!Zi9Cv9(vAI8v9bFhJV=x+Snl20n71nJ-MOU8hZ6
zRzoV4$mK*=#iEV_M$N-?a5`CY9_iS8O#I%f%b&tVO_Sg@{>VeS&l`@59^$rlw|_l<
zor<8zq*!+-hF+BWCVQJVg>B?LTJG3E?@m@)?B1DwUfg}HVw_Y3BOS>^XYg8?l<%^p
zBM-Z$nXeI3rpPo0ZvrZr**~1O=kjU+S84&Ny#FG)t)7Hca?T5O<V#{qw@d(JCr)70
z$M)yOtVKL1;N!wd0w`wU=Q~J41Zvz0>W`(BB{qgQ8nrmA@$3_Fh;8eaiejF^N9XV+
zlHB^1QM!n<9r@BP!RC$vO-4|SKQxvJmiDyrSCkQ;vYlYo@|K=i`_340Wy14cTjv8X
zV^P*IFNc@2g+V7a1xd4I`DRXh98e(|?~G7zYK$O-n^G<XJU1C8Ugc8BO<S{x2;y7v
z(3Xq3t)7tc`>}b}Z&i6!HM|(>{kfduB2uzjkB{vry5Nd(#T7CrDI!@z+Q26#M=hWO
zvnk4x@`O+rnh6u0vw`3El~oSEodn~z_yP$#51W@1GV%Gd?x&OiSJG!8iiX6tdX1o1
z>8C=ied+?5Nae(WiV%jIxYPqmEa^2vv&(qw!r%|JmMYY*JdTwp72<^=cPa}g(`ndd
zkoA^*&{<l@{*<B;(pye<=mC7!yvmcnMk=alo#AL*=GK*TdK$2&L>fXH(4^bRAquav
zvplmxZTAcQa5Q-*J!MCS5z>9lUK4UB=SHz#ZpKP1Ua_2>^a=;BiJ}XzdFCmM6K!1u
zyAv6P8sN}Ul({G?l`Cgjxt+Lj*t7~O$31eDZEu2FPI`TGpSd8ia9e&fcOpBc-LYjd
z_HA7|bi&v~ONQ<qy?hTDVcm>w#v07lfeki!iW^5p@UyG?TH3&j&Lyng%A=Rl{htQc
zr*q5Rus3mkiL45Y^0rYz%~{(X%@-NbykW(yoQbgf?K7^MJFB5U6!qPs3*s}S9aXkr
z-@y_g%xgv53eY%QrIFZnO*Rsep{|o<Kj!A%ZydbiIv7ggR?JZ$jn$2q;$%03L8}nY
zd7=(l1A#%;xe8oEQnVF<;*=u5v=vS^0_T@*LyDSsOms^IL>Ct5H^h<ffEbkr7`+_9
z!Pq==wo^)Nz`YmcYGQD9-!46$slOE_knrmy?{B7>GW@=&10GBbX;I`2JGTTMb?<ZT
z2$_0y{C353P51X=qv!T?Vr`v4r4FtT_DIA0@e3^5J^72!`0qnCJ=bgh2WNC}hbcej
zIyBiV=o<RL-zqf8tO0t61$&fai(ZsALN?!{Gs7syV*>oIzi%(0b6421e`nJ=LZrJg
z;4}|$jDc^MeB$0N{O;#4Q8coQ5DG8efj(xIOee~Z6bh%eiP1?#O|MX)pMko)%u=>+
zcuYrl=A=KJe4)g)J_Q7fh1nD?3h^tW216%DoqU&C>O$Lk0C0YIt=MtFa<&%CX%SeT
zlxHgEI=tnLq~#}7I~I?YW@e2lz&%rMW^(z+la>`fitZNt5doCq&mrTrP>HckU{#eB
z)(h){50|%?@ueH=ym?SZfWu$G%dof`dK#%$?0D&RTvg&#w`q8dVd|~xMAG;PJh8<O
zKCC(a!dPX|erH4XO$l*!9{_vx8DIQam6K3bHGO9*SZYB(Iv<0p->1qU`)74aX>2>J
z829_fkTvLDlE4lAeP{{Q?rVOZ8KGb7Pi2=F9z7L?AsvmwHcPyP;^Q#+OavJZ7kJtV
zYluz8iF^^ZE6~wUJ<Zx!f|rsI2QJva%$B^sZZco2pY&L1<r#Gopny-mRsMFrw89|m
zzq5tK6Y4tzsgDa&GC~Z=i1+3Y#(YkQ$BT`;0t)3zb1$ODn13#%!uSYrK#q^6-M<u$
zcjZXWp)K*biF^(Qa?4o@Uk`6bn61IyoFF+{{Er3q9UoZ!`h8QaSUfX)C+p<$vXX3I
zmU_`EBF7P=n7ljl=_RUPe2Zjam6l*??nY$N=pFC6bI8;#w3RW<$|gV_?TCf9m(?Vq
z)Ul%-xuM@oc6((tLs%I7#G&|U`2KF&pm)kV<E{`+B*zJfq=U2ZM@J=_IA8<T?E?jn
z?61rIeNz!Y5M+3~8~noy;WHjuS^2!zS@q(xp}cRcJx~bc3xBt1tUL$)!tt=(<EjAA
zbL3opFv-#4+d{u&KkwvVwB@S-AyVMF5;@!icA#HpcG_*_MX!y8<n+KK3-3V__`dZd
zSF4MPj47DNLy}~uD`VG`uinT!6-iP&UcH>&Uyu|fL#m#sdDbl;0KVU(*s+pZ2)UIH
z^JT^l*>5c+TF{yS&ol~fyn$j<iwzL%Ixm#*LwUK`1D#S=!sNVMBj{IwCag?_rDM4U
z#jCL#xH0hK7W$$mix3Aj$hq#gAeT4c?KSZi3KQ$`8JXX49J6OcK)U+07=mS@1h-M|
zw+N?5e}OiZrK*FIj&!A?+a6+J<elX=&Sj*Ij)!M_%wt?K4>W%E`|h2jphh%$r&?WO
zui=8L(n(s|7(g&3TLag2^QLXHk?A^2KvCFSb!jWk1(O_fJya$0<w!H0tXa)E*d60f
z=6%cvpo5xUU4by&ZVwmKj44If7J6f~7qXeahX_Ntfk0ZbQ!XN1O0#(LXE^$7RffX(
zWsRou<ag4GgvTwoiT1(EZTEz7nWkf+nDZ8aMvJX)sD#dq^#~UV{Ft0_QLh{CXEuPF
znV^Y2WrNnL-5OLV<#Ov{Bc!c&_Zs#S0GqwcEk4@S=vo~<TE%UyEQ>x|<qv^f{N^Y4
zt|k#IoX08VgFhGOhf`PX9z2O{2`!Ig=eHBr=((`9p{*5*TXx{#S#+>dGsa189vE&K
z7-N5G*$zt^oP$Jg$kt&gw(R0BKT;NFGMH`W*{?Ih9Wt)7h9RNP%v1Rz60ii->Pkds
z4$iscZOQRN>{@<{O868_F=umA*d93D*!nVvOCHASHsT^tJ#6b=_4v4>x$gI%g?0Wd
z2tUc$oBqZ%W)q-EQDdl(yOR?r#+E<5*ub4oz}$o;>|8xq_`!*s##Vaq_$1z26=A9m
znO4sLt>1wIzN<*|lXKu3M&i(J;Zry_7cy&DEf;D=h7WM^BS#oJ@CjS8XevFoYmQ@p
z04Wm470@%f3kN0tQ(Ar}`aZL+iw}(UUE(GQ^AANjKG)RjW@N>5sj~Hu9XlTpy6}2&
zUhgtBOrasNE}KY=Tsv2g+Mm8*PVHwTPGa{SCPPmw<__>cJ++hZPGd#mz?$k`GvOd1
zaez28{!ZS)tu|yG#V;FJMtDqqtvYEG;F9rA;2CsKt<@@j^bYLzvI^GWpSx&kypyI)
ztkcg1vIe}Lz679+c_3IKKWRN~5QIBoQ1+RaUyiqNAl|q1bv4tjs+Fs=*ODLQMauJD
zO*zTuuCXU(&$m;l?`N;Q!&QdIZS5Dmi{TUBy#Ou<(T<q6($k&OjxQH~5&ad}S3$E8
z-w|UWGjlPucie(AE2~}qT3nc8!o?NUo-f`6sX|-H^NKQ61P;MfASFJpu&&-Com+Bx
zVhF}O3(yPCvIQ?Z>cn(Cu#K?)bklayE|WDBe3}*7R4q|?%*e#8gUkF=5X5~9Ej@_6
zYtodm?o?PG`^b64;vD2b$m0<;{5Q|lOoSI0YImD4(5#EQt}aP_(Y!8<QI5=t>#1Ib
zd)o(-Hb$MK(<B&5UJI{DnNlBUtHNSej^5L1f1UY?f0Pu3N=OyYe}2~AOu!(5<=wF;
z0Tn!m>uRlod(k#$`~;uc^p}KD!+5vtjff9>ivMH*vM18-jCyD+pdU-Y>Om8D^qQJ!
zx`1e+7q|Tf^&T(u{aZ#kqlI5QBP68(Pu~YWfr*tAq(f4ufu=;pCSOP_G{}x&rq?3z
zmZYLCncT1sY&FB%du?~63|;~qa@(cm8gQfPibi#i&<WpE9u8-A`U6MtKzp8sMtIan
z?R=Q$w>ofUM`FqIjsy2))Yo8%NNeD0U9#iGOXwV&kwk!Zbl^_J_%Hrv2+r)QnwE3+
zFV&jov0NBiQ(Q!n2vQ=1lB(zRi-V_<-A+;K9`e9!!7I_Ex>GL2o$PC-x1baO<Rq8h
zRS^>Aw&vyxR7ds{ZdSc9mo+^(-it{XIJs<&^D{_H15M~FI0VW#fhB7fSMC(GUoM%A
z%&YX5ff@xhDA#(PWmbBXif3<2Eu@{ng)=OVH#44fHp7Gyk@Ry#`fL}_>v2C)rZmx;
zI=QJc`Y*IBynnE{7YjbalcI?A%m+izbpd#jL1r%Ji+}bX0aI7fJ~wK#W(9JnWk(q|
z4e~e%&RBDqM8FqPpSolBZmQpFN1;ht^!{Q6ZlEJC8bQP54ijW_!2j-Pnv!XVOkZ$L
zuJbpfe7AefhqnmLaC(-kn(<0YwQ!Eju0pa1yQcj{q|x*cT?+(sMdfPm_38Ah_ojn|
zYi`S(RKX%8ZqY&4mI~i>sUNJdKmcCkCo0JAx4xzDsReuoy)f9~RGlqTyAoaN+q7M_
zCZI0fRwOR57V`HDTPB1N=*J4t%#^oj%1|>X*CcCe8P#H@aI@@_Z>6sq1LJ88<4|cR
zhHU+Xc>d_p9(n5!*{yQ=I8sz^eSozM*woSPEfz2yOU(^Jp#Cmf$}3$m%MDDiJAVbL
zSV0Vc2jNQ<c8z2&wdpj4OC0KsXOa$1Y3uJ3!|r(NUGw<dyHSIJ)++R;foT%ytpsqn
z1b8M@;R_utd=C3BZLz5`XYN=b61pTqv^+T2i*S1`A7Z7y0C;JRss}wbVNnT|mK3ur
zNpP1l5<cKrNy1Ed_98aw20nVHH97u%B+So#oaE7`kCm-s^Xb`0Q<qjnhm={~)Ciod
z6lJCOsHXdo<SC9M2FuLsmEhh{*)}T@HUnE`CDa>`KH(UZHq6pn#Sict<y&kIBbT!s
zQwhzhse&y^J!IOqK?zkHIQEAJH#-(z>!ebLl=SfBxAHO<==pJxzG%-Ct{|E+vVO@c
znZ;tlKH0d6N6PItGmA%cN4Kcq1v#5icKCn+ba(6I%F7<uz+Ti;-}V-m3GFT_N6?FD
z#XfXJOURGGBOPzS*b?DW?DCUVNG)T+ZF;4GLOsY)g`0sshsC3yhOJhR1fkmH3E$JZ
zDP~v?*~>T0?5sotDrCubBx!#`a;OVwYz7Ra)7#neJNU+IYE%b9lFyV^e%V>Ax#cu9
z&ArDd=#`awJ}W81n*VBZ4!#g0pw}B=fJ?1eQIvkLrkC9^cBvk+!>3kRK1LMieGk3T
z`Mu?DPV@w?*#;EcdU-%}jG-uPRjBaSVvj9#@U@1m{Q<Uo+TR3p`D!6Hlk1c*d;0ZG
z`_?UOm2l;yKjG!^g)s%52NSl(kC>`cL+%VA&CWQ@UIQ6jn1Xw|Jlg@$QRXtW(!5?C
z9dxMuY^_nd0JoBBHeo_^nX>GSx^UgRH;8E<ZGVSay@+wH|CvJF6W`8ZaG%Mu-k)6H
z8#f{dGL?jK@p2i}P9-)E4N=oLYr>Qd5tP&u&<0_48Omb`rIaQVH7BOptPTGlGHWtu
z#0l<FG|-t~Ko5Z4w&e~%@`k#=S?-cru1vf+)iu1lMHRXCZ5L&1#=wjudi9)NL>S!6
z!rCv*qoypff9&9O>AsHU%gAJ=sRjKSXXswh@VBi<4IdC6-a}^a%X#?;bB(kb8!A2!
zITUJ$J)i>g0XA9wh>W0cY#><V8vuJoCSbe)azsoW@NvcIK@ug}Pbm`*qld?S1<bhZ
zmYXSZ(4tDlw;#dl&3X?#9Q9f9XR;E5E_t7XbvN8gNN}MR%6IjYn58yRZT<y%2FULi
z|FzLw>ojf${uYEH6;avmQv)QQRl+ZZ7yC!gFBoFROB-B^MfHsUMlhQ*w(N@H7Uoho
zWd~Oi)-?F<>NI<T0wsLE#7<-P6_%*}@^K^t=>OP<l~KWgDQEscLPIXL($<^gDQCii
zg92$r<z6cb-azG4H5w5^bgyK@3Z(%uXF0GqoM%ea>)jKkQfx~-8f?y6o6Z)n8<4)v
zUZ2NNsIw;CugSf|XX(=z2sw&jP&unpNqf1n<`WsQlnD$igEvTKa$-B`;DI)3SMi|T
zMq$s5TG#u-CLp1)E&p4x<j@a!fd!T7Py4S{Dibco*GYHJM`$R{)A=>2(aR}&h8$$S
zBh3`gzP}i3U*ZAM;Fwn}Ydm5z<Q9F$Bf8Wur){FZCK9HF!mz%Oo!BkyX(FlPm;lHG
zjC;f5>Ft&TSJWv5$VLxC77^lhbFnm^zw;znmY+x;-)>Z(ydcxc*KOUSNrG;z{7F1Y
zIEwdR^23vT4jxuM;2E~5RtGGA;a9Y(Wm7wYkT)@y?;BzZRxic(Xf@{<=qXWF=%dz<
zADW;W;|LL$#FS-C{6bk!Yo7eO(Q)(gDvbb6C`9);HXqc4&bQnCKAUdM1xbhfd0}c+
zO&&#s6lanSa0QuTlvLryY{nT;C~c9ZunE|tB)VM0>2=-VK5`DUUYog})9t*b6B{I4
zf9eM2`J0_}12Crci{e<8e`4Bh2q=u)@s)%U@dhfE(-))KoSh2Hp{pq|72`X#2e$rA
z`yueDY8wd<C?m#Mkdn(e`oK@ydBT2>=``To6&H!t-{<$eo89q4D8}Zrv=H?E{d#yX
zTek8bjD$t8Sw;Kqv+q>h6k7^o<alZsfSbT`@u(7nuFVhjdq&u%K#D{O^cO_}r6R=o
zkTG}>*q(eJJ>S6=|MN$uTrE)<FLF=TIN(p5;g+Nydu^(pY%0SW)`1|rtBEYA=kq>h
z`SxuB%P7LggM_NEOG}-HZ{?Fr&mXX4iqsak0j_vBguu%1sRZukWC6hy#rl?x?s6-@
zyvR$zII_$0B(1I)un~p~b9O1_2-ujOlMkUzd@S^rA2>%rDPa&qQGI-S!4ZfBQ*S(6
zlj|~7#Jp`3ut`D4_F?$8hl=`|G@eTL3!psoJJJaPI!t6wVlpCR?<<E@ZNLC!;B_pi
z&<J>{!r6{Dadx^iI#oiCqmeM-dg|flF$z~gHiehS8PMiH-t3JWC;v!7FCdaes_3Y!
z4<fs`-b}0m)17-ySAL6(A!-nMCy>}XCR7*pkr?Jw*+^ZOlMLOYE|s<b<gF%%%h$P_
z3q_bL5+d)k4;^Pvx<dQ8z8H37(P8Fngu->{SVh2C4GWr6d0Y3ibG3OFoo!y>R+Y!U
z5Gf3Z?&2O>W(=mWG9pgOdBoibDHKK3hs=ta^t6Fdkg3F5MO4J#MkUESXuVNno=0QO
zNm<Rh<lz)U5Sl{qu3q91LL3*+9XLtAy`e*;9Dj1xpDFSGyq+CT7}0Kj(YwnkR&=4l
zf3{^&ba%AaYt(p2+Bdz7ia8_^^}wXdkUgiK)o5?Rp*ye4(EL=)nd}PrpsiT2JKj&!
zX2vV-xVp}!ul2Jfi|V^ks;o`hyODz8PrP{~{IXi1JGu1!U@$(GL_u-?+MzWP6(%jS
z`K`6l@x*@k&m-QnWw<xj?y#9HQ<+E63&)HPyXkEsv~fghT#W}G5@lWu<-kGd>o%VR
z1ZFBKyGT9ihyKE~8>V^AO5Zss6!$NOPj8F^SIR`l9I4{4+hf<E#EQFGrLY0|Wyz-G
zIdCu^8%?ymYfOn%=PAV_^Q}Tk|7{pH!I1P~oh5bH4{qa^guZ3-8!tRW#^(&27w)eG
z#})<A0SL{oAGKy<_fGb|*xNoLQNpv7u!@HzSRolow$OQ%NISbQHq%C-9w(UH0*bXl
zBBDL5ez4T|#NQeVY<F;-uAGN(@k<|LH0p=nJ{jSIjNYHvp%DG9=f6M1I4}=;2F^=6
zsz<4+XU0|#DX`T=IAFEB(;E$%_=j~RUU*)j@`@ApS$iHfsn(X7t&>jQIkls3D9&VP
zCO#m%u`EXHAW;o%akG(GGWqyCA34$7nR;qH(q_p3IYEa%wfLf!gW)mr?(HOE^uQ$)
z4R7)Qo@k&&VF5=y)n<__DXw<3)WZ9;efA?9y&oeG<e6oiG=~VOj8K~o>{c?%vtJH?
zW$u+Kaz~!3tM~Xi9Y>z@C>YjNUy{+{GeIQq>34X6{g^*lECghtp}~s9A6O%*JPMd!
z`_Qt2Io7H;sb!>0m)l<y7Qbq~e3*$8U)-vn=*XH035!Hd15pmMF}j*)9n;F%1%|lu
z3;tBi2=+Q<0AP)!Wl!bip*3I0vmlrV&2}Rt!YG>kzAl=PHk*}Olrf;K#%dg;yk(PK
z&yGKLy)o%=N#*fY#>lt?a?9hD0;p<cV(!Kcm`b~C%*|?=$7UhOo_=+(jsX~6n{Gn_
zqvY^(|7hlAy$O-63@&JG%Yw&rggeHLcbS?7R@QC{TtUwspm?|*6>NF*U4>!tje2f{
z3@{d+K_rKGAk2VAIJlKo<|0e#Wu5tUz<Z=^PqrRSgB8=L!3S8iW^${B1k#%Z@&tR5
zC*ppqZmD;_KD4h|Z~GZzHS{!u?McH2e)&m&6Q%|YVqjY(#@^9GA0o6uPpaVkZbdEC
zu1Z@<(4$!sx=afCJhH)d;f69AkQ0YcTa?B4j5ws;u93Lb>i|V&k=c`=Rj+j~KbKw;
z@+mn^CwRxvKVQ@jn3l)4JHr{00ORRmeOE`RmcH13o5*o*#iPx|#K1Ia=W^gZ_s_Xt
zM<E0WEDXnH{I~>ro86@Jj%sa)N=hv5pdSDH?38jxgfqC<xk~$cf#N2*c=-5^IKe)>
zc&U5j{yye=J=1lSh8e`^5bL-5Q0xx{;8i#{K{Y!$p|W5rq4SmK2P2!_8cOHXm_gL6
zKFAy;;@gbd|M1`4Tj-d4PBYcxEuF3w^@I@jr3iKhuB>wmLThYmuJ`5*k{EJRk&niC
z{-CTc=sgtYOsy)~K8n9NldCb96<{7{6sD2KWCJ%W9j5s%e01|DM&=jQWI>f7soPGp
zHdQu*o)&q-%we3P*v$+@EV?S-as;W0z^!<KDRlz_dq^83x7!wX``rVIqb#3ZY+6eB
za4+gD88_7?HmzQ`w{eUW;!$r|Bc*}J@ZCSZ853EC(x8e+kS+c@zWvGfORCRS7X`~K
zx)i?rK|X-Yh$+8;MG^4=nA<@dB??|$>N#*g9!#99Aa^i<UBp)7HC}}hy7mL|Dh<Nx
zm^~0<4kUs<H)@tvTei4DMT{{Q1r>T|JjBI>?7;JVI=@pNfmRN&Wh$N7CfVW5Y0k#C
zZ>C#Pon*(r7@Q!=B&8|l7MR<eU-)C>#C;Wf_3Q_^@99ry91mj@S78(g1)IPRC}=cV
z^KvNd`BP9_I(z(3Noo}CRbd;5Oc~>y;E%OR(X|ShEfe=iiEK4hjH9;k*blHrr_%XG
zE(EE#u(xtzGDjs>hGIIH*o;E_fjlk3eu{N<Za*j`WNFa1a}Os?mi=aFRH~X0E@h+u
zI7~jiM9a9?$!Cy~X0c4;Y1C{Y?U-nEgG$eE|Be}(YIC^#1NDAQl8R|^Ovi00dtXf#
zOPw)KvTEpzM8O`!zEAIy1y>nih2`5nbRW}VAUdymg>@gTXq}gDe{S#ktO85V0i4vk
z)fW8L<q^-}glH2il}OgxG(Kr@f?8+HNIasVF0MPTOVXQ;Ka@%S`PCBeXOfV5uPFQq
z^mW91c6^@NAU<+h2ayjc97jr+|6RKd)|QRrg{j_i*7RLTr>T#R3{g4GD^rtl&09v4
zVORgGf*}znFr#xlxfK5B$^L6H>dGo!ps*2H4*b1qZ@9i50b!0IC7i|x<%N|<M>rHg
z@ad*^S5%;Ky77E;hKLv9k1iOg^easHBgfx4B6^x;3TbNN?zZ;y|HaigbqNFL$hK|U
zwr$(CZQHhO+qP}H`)k`a=geLA<;-8mL#2|`+PQlUJ<jP_plQ`8LnU5fOu~C<>CI67
z@XTqjSfkx&Aut!n1QK7Y8ZZs(kK-mo)cK4C;&(O0VxI#}W77g4XnQCSvVCW<IUcT^
z@Im<wg;c5f4I3>k>2hBhX3)pKT!SLSEp?lBssLhVbRstz^>Ag3!%|p`4+Y6sK1%xM
zNv0q1_I}|ZG^n0`EOH=GdxrUp9@?H5=ak(2K>^tm{-c_EW`xkmZ!kN5`zX^TX?^2|
z(CH`v&<xhAL^b3Ho0QaFgo<YuS}~KxM}^eF7(qt<QOD;;o<BFsBO8708(nSi41p<e
z_BpKzwJtkeBw)@HN{TW&ugP}yIz~w1ek}55E8N2N=Pl#VUDOrgB2Zld;Z=xT<z8hz
z^y}DUAfPyW7UN4V9~Qf#zlW)*sU@aeg4NI<y~_y_jsNt)eqmsp&8aS^ngvbQM~z7i
zSLWd*hf9=F`P?$q)Q|#pvK8pQV#k)_uI7uR2O7=$Q@#FFCzEFTAP-yX7!8FN61?Y>
zrK|@wU4ZDI>PjlJ)|Y~&Y?<p8L!nL3LxzfL7Rkp*k<=J7iu3PDt$av)_HMB-w7Odo
zN06%?Q04k)mACDUsoai-AQ3Kz+ZY6l;OGRG<Bt2~Ba{pU;XawslqYoSCJ?PFQ@+^D
zD4IX4KBA2^aRUVqoTd<9#g<oC634($)&c7nmk*Iq7uFPPKKYhqj&VH})bCN0@2zA%
z_fcg<uKa^LKJ(fnGcCsOL1b?acDg$xC~eQnwxTXaj$){*_dS<GUP)AJ{LemXIVXSG
z0OoNx;TeuG_R7U0&W3`=-dYe$Fz&b()!|t*H39cXRBAc2OlmKt!0;ERZbX+~?kC=G
z*|pxC@SYURHNfnP&HPu)&t=934@y)NS#n;_sE#fWUH2IQ{LdatX~X($_;}1(l$9k9
z{;}%G5uUzmJ1GSm%)v{<$3KdKaSn)|A|X2Up~G`jPt|HB#AnOB=+;>o)<&4fSOEIf
z_EpA}rpvL_xoeUuU$YkJUco&0*{b44A_20)9s<NQe8Y$iiA-Ny9vu-reRI(twX^$5
zX&TJ7$<`&QbbCFnA2TkQBj~&cCUbS$?vG<hB+AHI9ZaKz!|C4@8XY8=#_<IgTN~b!
zjKfR>#1d~ehNIT=09eq)B+g|)7S3$@8VLV2KLdQ^(z+EZE^7Y#V3^VHAcG~j2(%CH
zp;yDIM|kcMo*2k1161IRH46J!EO>!j<2<1_t@QBc<OfO@I&uuzfs9_fOn^G-Hnhel
zO}b%?jNdua@0syseuDRTe|LC0z{9xb1u`NazU~L>iH5w$@9QPtG#LoMZBD}vR2-z<
z;oML{TEMYW<6)cFcI6B2fNSTGaE7HmlVkSVlD1YqdQl&yogT8;O|SN5$&!T=>mQ_R
z@-7#53S0UNeC|`HR46jweJD-=isZu6xwk{I<L%^>uUtfp-6R5(U_7=$?Zf8TE+)LI
z->LlJ*2-RrinLvxvv3F^V1w+Tw*{s<*>$F}2|;jZ(FoO~WULdH^^l(MYKuMYlES<2
zsJI<Jg{L@)lWnre2B`CPWW+IFd7u}>eu*ct>@09<>**nDX^hFna=-v|DW-7!2*{w6
z4LpnBoP1-_9l9Y$MK*7V*JUn2^RRFNB<lOyBeWD0n1L<xfOAeqJcxO0v=uMXt%e~;
zDx^G8Zwz4ApD4JY-R&kWgI9@34}5VShCv@FuQAZXd66;!5eK|hY+y)L9K;kQA9#;C
zV!h)@Z*E#Idz}~_vyXsD1O_@~)69PqKf2{*`!j)h@D*EN$RlWK^NnT`EaPNc*SBu`
zPcN}{!eW`481xvKtI3jxx0cEO1FLBAq)+$u`kaV)B4l4h%!2u)s`t^K$Kbq<o5G!`
z7)~rQ=~dD5Dz~o4d}*QQ>KBYZheTq-<5N&Ew!b6L=c=VPEQpB$%WNE2xQX}Ssj@XR
z;w#`N3g~KbcVeJ#4XEFv89>Ks?(4EXo5+u4z0_Q6MO$@)D~+BBgO}TW%<t6z{SAN2
zMZE_pSjJGSz=jDrtzxFhH@Tz8i*}aa$h?rJaf-N@IiAtD0hfK*7&#@Ojp0_dG7ULg
z4HnB81z_%y3{IhC@z{{ftz125T-T(FL3nMxlRESqxMXhc{JuXVKAAFyD!3uWCL#%L
zO1x463>WlOc%aW(uAodESz~axT7syfKbBa!X*aDBs)hI4Hd-lEALPe#EE=e*)n5Tm
zZm7(B$YIQn@E^RR$}r`wYI}mHWoE2P<2G_#`lZufo1P(8en^ZCwDzoNST1s`Hoyr5
zB&2(@HVO^=S01)1BU2rx;yt58{R<K@#0Ta}su9>qJR~5}&*Co$z0|ARB<po%y`54b
z1;qKNvXa5drT#B9`mm)2jimI+{W8G}TMLplb>{HR(Mk^$2K=*E(KZ<;BM_dy*(6!F
zP5+i6VMN>9SkEslIp*IqH)Xk@QmOI<OlXl6iH&{~5hWF!6NbY`st^POPjO_LiB`yG
zbsM@XI((})R(@u|d9rR7amTzB?}}~zxc55{&Ocy7(W_XYGOUX|f_GT9+(i$4nta=x
zmzJ9(r>qTfws}6?lf39a45qBTQiP(1zz=;24sOskBye;WX^o{y>*7Hhpu7o*h3t6^
zHCo?(^Ca?IjOnI5BXPn&Q_0;u4_vAACO0@R|NJdY3IWNr?5rNd$wyha$oM5mX$1%A
zUUDZVH3qObT_rO3`592oaG=2PS^18_+-8b>j*m*5V`xKRww(Uh4d*M@CLrFB0(eA^
z&CwEo6zdMtu4$kD=z|aY81Ok26k<OWZ@6%vD~NkW3kgJ;`nSVNwPgZ43zMOzyjhLf
zFB)20((Zov!V57YS)<9eDOcnbzJK}}O>YNo1~^Sv=+le$e@0bCt1G|HQ|AyfA7x81
ze=yVDst@EE6hCDB<t3j0wC3w(<tFI#zRGyBozJ6w$I>4LV0J=nNO%)70~%k&1k}h;
z#t>C1Yi4GnJNGt;>dZq0t?{>l4fY^)mnv?*%?&lZ9Ipw)?@qkZehV5g)?}&nn+vAD
zEwV5jId9%0wZH3x`KkopYLeIi%2C4sUW_EsnkTq65zp72Q7I%mej}aZNNpvis6pqv
zYrr4wTxH4T=LY%39Ks-(>A|a#bb*}hEps)%k^prdXgMrmM2$hTz+0pD!vqaG@+2v@
z-(RysQoD6QC~a=m{2r8wa<^HX0x|M!g%Y@geufm~ObZ|^WnP$RG?Rq2i6&}Ef0Zq>
z>Eb}3G$5*eO<Fj`wH!B^;O}>m1TQB^+G<oXpQuZ2UqJ-Kt6~Gdp}srR9=Jgr2{0KR
zyERhX?ZGsg$2~yEK>YF-QRv7?8slIwmcVo*^yzKUl6rHUWD%H~2lNxiFK~rUbwjZD
z3I|ZXdbQ;0WB1Jne<1*n1FQXo8uZFSJy^?Fby1JBatIHPYD|vDGY`)SNQJJ|OFV1E
z$V!e~i?nrWn6*Nutl2aDGUx9lOs@uwk0Beq=88^MZe1eY*|MxIW38`hO}Bsz!scq5
z<p_NMBo3<L?)^z7Ki!}EK~Cfro3nkYxhHpKYeL3j=oDm~o$>94dV1EgCP42+ly)Pu
z?Vk8etE*s04Q5^k+IQB=VzWYg8&{JFxW-fcD(szz{GNG$PYh_1?iP|aFAh1V7dr+S
z>xdV)*F)%<>C}G*e@bX_mJ>OYlAl1}+(W=01%1L+Cb@%_gi6k7$kek(O{RBJf~>)+
zdygwy*ClKn!Zfhf=PSsQ@U<b_sgujrhyK=V8W3iC+C_y&A4M<Nf>yB4@*A*=WGr8E
zQvkT(>KLud%Y~tPvOD&Xy7`kUw|f^oR@+~?;R(6Xcd_rH`B>kN0b>eQj6VxD%nhHJ
zPY9gzFSM<T8twZXR!5HSk!WvoIqV?>0A2GS*GLD=R;U-wxKZ;r%>#nVe&1$i94=2S
zl5uWjT~MMQ@d}~nAJA%e=oqlly>%!|9H)K|BVVGh(2R*_b_;oNY{gZYA8B!Se(FJo
zO(O2l%fdRsK8LR*BPUAnKIKjD)UK#-$i!SKHzjsTU@LBt(|6a(u`JVPA^pDgm70L^
za9Ot7c8LUb8-b$g=2RSnY|DUIgyvG~XM8Xn8HO;0!HtV5*>*Yo_41z)geF08q;Xpx
zGch<)KDbOKmi6Wl5(}3OCl}2Mlb`l^PD32!y>R+V7t5`qSGswzNH+}T{X31XTUVt+
z`;T>YU*&^Ezg3L>2iIq$;R@cckIs}gDg11l8<;`xjP&dS=(el4Xs{f8>1fDMrjo5F
z&?pI;;;L~4w^!==zXh7<!{_scKE$d<-6-+>+0vNjSS=wgi<srkz~4xM;tG?qfxt6%
zy=S{akG{bahZNhJrJbKu2be_GDf#FaZ-E;L@z)t)-*o)km{am*7On5Z*w`5L1-sVk
z1l-De>$XMm*Cth1+B{DE%#XJ+@puSuV9c;l2iaCf@A&l7BX^@f+<QuLq=b~KP|iX;
zii0-r)_c{V;u7feCRXtN^!7KkZhxPCbrBW88Jw_~oM^7`l}rqVM<pdZDo>D_6e2H-
zEP{0~GfI;Dun4r&J~lg$J(7rRRp}DCsQe#(`py@HMwK3_mtF2iAzlRO;io1w%nXWX
zBxLz!n7HyggwH`~vM43@90I=JP}K68x4FdQ`+{6IJYsjn?LT@QL`!llB~@zdKWWYj
zg4VTMTwNG@%<lDoyu|TOb2_j{kQ}?to+eiqOit!wq7Cw8UE1UC;7$|1jJ!K{kjp+u
z#wroIftOkQ;ql%)TDllxwNoLG+OD>wL_M&)Q&V&4k4K_|<t$Qt1B;v|?@Wen45#W8
z0ZO1o7Jv-vu?%HUpwa0}yvugwn`y|`<~ppVrLb)XtI)!8GUf5%%bpNglF$_ewefAU
zM}hi!a+k5>2}eG44^b>8xIc@K16z2^RQyZ4(*`x3S|t^$l8%HC{sY^QU-r}s`Y(hd
z02qE>>g;01{Nx{93N1?M0;`uEU#bK8(r+Ep>GNk?>Am*Z$(sU0K7%=W9IY3A84cnC
zp&g&%7?(;I1jq}~-<}B^i&Hn(Z7_1kl#u?(gFN@(=R!&#PZo&@q=nX4dv9@b&;?=b
zem#c!3N}|x&ImjM9A!Kp+v3IxvgXMJ$f9RmER!wBK&sJ-Ffi2cb^4Osb%7HG&><6F
zPU^<6B9r>tCjD#F@J2m!0i<^gq-cwd%}&5nP)wAb&>oCMxAYyX0fi+0ad-Q{((b^8
z*QaNiK&>dCHbrh?)#affYCj;eoEJF4^L~+Ld*7ND6QCt2C*8X_c2D~Agb;4%7#5>9
z+Duj@I9zezHr<uV2UejP9V3xKC2fk-iyPjyWB{{NP6#p&B{pNxjJc$P5xN(SeiYeN
zUZwtbwqGc1HGAvV($?sdPEQ0T-L%n?aP2@`*dP!ZSNuA|z+5rFD4BflcJcB26y&_0
zsa~XW`6M|JatGV`jvFH|<S-88Z3nshggUW-g7ATim{UUO#Xr}$#O0uy0vwuyVOHYl
zsxN6_{K<tP=x~t*eTC&k;XR}@=9bQ71hxV{AZR&3;oWsjBW4!~>=oQI2W5Ylui;*^
zPpKjI&Z)JN03ZdV8MWJW$h@JyMju<YBk45cg&89IyCxF@_U`J{3+)f@EM^kWtq=HI
z4?1rsmJ!{r{_k9BhGY7!6bXm0{XTJ`ib779`-i_ZLC838^>fPxmE_QG1ClSkvljZi
z4;maZlGyJtsEaA>bwj<4`0mBqX2%m?&jd!wLu*(`R*_p3vSt*0V*%`GMKSs$V6>lI
z03(%)oPR`gDS3qF5UeX)W{2`^(cSm4deJ&rRbVU0v?*IVh+!>d)O0V5xuEi$;bNOP
z&D=HRcR2F#@7*YHFcwN12<@SbhLy#(V@t))9jD97O-W8Iy-e`TxhZOgiTDr8?UUjQ
z(l&5s*Ld`Bd9>L9tATkC<qEAhb}-Hsqi5W6x1yE#sGObTV{>@l2|VHajNH_%vaP1G
zLVF3jqoA4{#KxvykbPNYFWe}Y*;}QZwHwRj5>n<FiuBDat3w|5p*-*f8%<Fh$=3C(
zY#x#~AUS*F7o|E9+`DQHnchy*#Wgti{^t1hWc&px=Fg{_#v1-E(Z<9!KH`ETNRY6m
zwz=8mlj(ADCDc6695<}Ufory)(FgwfkXp^Tan*ImlqV6jF>4l*Y!M?WLRy7}Rg#=;
zQyKflXp_nAW;aOMIs!}={#(&2e?|6JcZ1PFHVkq-nsU8Fg2pWJhHm`6cVV38U1PP3
z{vHQeQTXRUY7hq4ykODzGQXO7P6~)zp=534c05-Kqo7n+uH$S)98>n38I40s^XdQ0
z=5VkitQiO|8Ck7~9h>eZu!*`msnS((JUJc0+&s?&*@rz$JCQ|;G@=;gzvo+pgG4s@
z+Y2^;%kYZ19-XX`4W>}UMNXjpSxqrg0VH?sh7xIzugoa$U07xa(u4NFt$Dq(=(86z
z!9~9oDVN4_F2Qi9ZnMmlfy}-LKurKsK&-z&{yQ=F1;K`pFJdyBxZa@3VE7D&9U)I7
zx2H-Fp||t9`0q88K+_SGJ+JP?!+v6t1bA0Og~HPHzSUV#n(1j@XzjOcn+u6YXqHAQ
zxbmTFOV7Spo9T^Sn3LQ@&&3S5qQs<4Dg-Q5A;Rud+HQ;K3>g&CiVba-PCuAZSkimu
z1@=p?q=2;-*13(1!}nCb+z(Y;pQ;zroS^3y2EjMX`#h}xp=7^5U-Fs7U6?J%O4z<J
z9Ir6QmviILp2p_{jmss3h&Aqe<?;<7@J2y60+sOJ5?6#hfSg2S_^m^FYWfgqcWx0-
zE-4Zr&*5wu6TQRyaO%QBFU+E~_MsDw6VAE|$~W5Av{Ydj`odq6oHHr}TY0XdhEZ=k
zX?rqMj^&&zWj;~s!P4VzIH6k;%UY(~RX)64&p>rPaA+C@*)Z~R9M)@sR6r)Jp44ny
z%6v-&BIuX`0qUuUf=krZa_U>?H2juO5Cb^DMZmvN|14U8nk--Rl%Zxr0hGtqctHk@
z9lYp^*Yo{2fGW@aNo|PRKlPD1c`Y$_lNTS<PP_9csO%^7#U%gLg(7)ul#9F4=33~&
z2N#Yd?+%r~17(k*xY#IoM4&_T?K$Zhr?wAnuD!3;CAi$OPdTj%!MVz&-Mx(D{6dQ#
zql;gENR1)0{W|>C+yV%ziZ6rlz$l5YVc(0Wzu9kKxL}zZ`<n6^QZZ?fhB)#?fVQSB
zIMm!5LwrXk-njTOn-E{%+U`7*2~0O3)eIg6Fv5IsF(pPzbh>j6A&Lz6vtS$-7JQjU
zk$`tE{KBEM2M@qs-DI^?w5{;5)sK!*Cm12ylu;*O1vQuwVI5KJwLHo;y(T3edtH=<
zv?(C;_@Sfo&y=mo&;U%K*eRIO*=D^su_$b#iKP5*{7ayA1FzE1hV7tcvqdU@GQ@!R
zyx#;NbcIOo1UAU~9)o_gn<LiD=a#~1^8q^voBDmHO?cM{0Crhj!W`yYN%0u1XdC^s
z&i_6gi8B7{o(GP2!`;#H=$3e8aH2fuuG*hqPanJQ_W5T?)#}V4?IW)Yz4IHXo^X21
z%m*U2@zIS3QcP>%#e1w2lV>>j+IMD${}1rw*OYy)G)Ac_WkVaaoaZekd<@mg6k3uU
zo@R@R7|WA<@6dWObR!fo>JrUR$)8qdZkj!%d8R>)bPh<7Tg0Jb+5ncFqN{$ikEa?c
zo_EajZ6n7vbqk^pn?kEMpi=Ne0^+KOhkMrDA6<_?@Dlj&eS(;xAP_eYDDlJMZ`Ht-
zCN)$eRi?FI_sS0`0LFs+ZMB4rR@Ie<mj9y6nF3%C#`X_bNxizQU3d6P5ZFXlo8$nL
zElKFxKKI|RhwBMOv9^&~Q6p|vAT-CcR~kQ-3u^!|whgqeZO_UBuPLWO6{eo19a)>N
z+s*Enb$e<>$ta@IKMiA!4j+EiO0HtL10veyP$ftV#%2FGpq}g>r#aZN6l+JC)%YpM
z4v-NoNnEchXgC@lbhl8w>I8wnT8K8}yn&KF_(ts(b6lULVj`78sA#Sni0|JQ9g-RB
z)Yww8e^mPl7~Ga`UhGO6s?{BLENaAiian7R$_Izv;q3n%e@VXE?gfQFP)om~VU5Z|
zI6%M)*1)^Vp*5}sQH>E86btOy<ftaGLafKrq8zX<aa6aJ$R_HnPs6Q!an^%4&e@JC
zN^Ej-3yD$!dSjFhv5vLSPp4Tu)e5BlW&U(xkosdUyL9<JO;esvfgqLTV~Bj|i9}Q0
zvWa;SLA7vWfsukFYS;67#uAL09L0i_V#J<ziXfCU=6&w2kOK5#OWXHz&SGYgV@fsj
znSSyI%k4455^OE)ZRJW=s#L_M&zSrp(aP02E|6Br8GQ?z%!yR@C@+-%>90fGK1@pp
z4+)Ejh&?rv8b%}cg*hJ$(jV6JaDMrg95n!&epH&B)|Ttt635>yR5VD+AScjYf1W`+
zq1<Ncvh`igf>XSv#!tm(QUrdJwCc`j?BLN&aRxA?1Fji|L5rsVSh@^n)$Ki1C2g?8
z2Dr!>Ji0|S;<}EU3)vAA>C!9^qH4S5uTlqc&wo5<LT<?)<x{zbFvYFR&n?8)lV|%S
z3#2}V3v1<n3Id@I;9<c+PUDamV!>w0W+{X&VN{ORrQqy=DGC_T&oE4&-{fuRHc%X*
z>6){qBBH)^bK9t%YC<pV46IvFe*UD%h3s0{FCO4fR{k0#2cejF@hW`Zve@#MW^^(h
z;GJ^FHi$O&nim_u-ewPu?r?X)vgTgny?|dweDTJ$jZjttE}|*r*84HQtR<i+mCM#h
zlzxXp0X>?PiyG&GLIv}Bn}5I#t0*tznJJmv@2{~;{jozA5Ek7#ir+~~8wDw5;Bua;
zofXEL?D3_)$NVyxsB`^|TUrm#i3tEfz!~N9-%RXcF}KEYFyiH;X}Arkw`im<{@1^`
z+@&1O?<bWq+LqNb+qZ6?TCs^}cTT91we5?I@|<s0-{e7rbu<sWbT&5u8}Wvp{!?Is
zaV1QpY5x3fsyU9^HlG>(lz74J;Zgu|Fdf}1;E8)#uoBjP8*;}qpiYAs$cKZmi7Lwg
zaX!rX;1XpQt@;=K$dO$m(mg#fw@&e6N>(1MhP*&^6sE;;oDyAxKY+I={8kz!of>~I
zdd~@32&0*)yTiQI5qQxU?56GL=wqQQR+(jKMzVBo8i38^CG3Wwjx3+A4^slr{Zqy;
zkxV<Xg|HP*_y8e6Bu}i>9U3?tvX@K@PKyv2#-H<4B~wd(3^F0U$f6F$O(Y^QHV#hx
ziv!`@^Q8l(y0Z4|;79T@htQda(qjL<d6^H9=j5ZeZl#R%$kj@<A#QjDaRJWB_K17f
z=#~q@giUpyF)yQ~N&>`8X4gvE(Ek?(b}unOkOS{e)B_A^C<)wPkhNk@T=>jH2gI0N
z^k|C;l(>r`3m@>7?)b9P3E4c=ohNGZSc~ry4QSe}$sNwfL(R0azUlIkm6UTEWN!k|
zuN0(6?#2l4pw)tMnN=!={RrTZ)(cRIOPjrnUYAkHQf#z+>ywycLCKyWS(pMy2Ia!D
z0AE7QOU1ifJ?%l#p<4C#+P!j1zcj1c;g-B7MzK8Fg;Z<fg#lh0yw{`Z=mBLKl~b61
zQ4-jb6)fZ)K;jn#Zs8gxG?V)^#lY;wn|wz+H8>(QWqP-DWQ#`piX_JAErA&fE36J4
ztQZy)mklsKrEA73oh$)1h9dM;MC<3I?}>w#tY?kkibHaZ6KM#kfYK=3EVd;kij+tM
zC-8vb?<@0zcElszlzA4R3|_-yHNHwBrrQ;{GIILMXjiHB5g{e?dl?01_CBIoVU!yX
z@Go@EjI|npVx41O;2u22#XY&7`6ecX+gGe~P!Ny%sYXVKlP5^hVDCC{K|p8c<`n{4
zN*5)q6YS{f)Z_-@jk6+ESEx{#Nm$JMKC(GJgR*Z5&ulUbrgesPO?nUH9ml!@qml5E
z(;`}1_ixP+%jkVo@xL1mAMvxH=sinLRRd#~lxl<X7})hRVhGdM)$vPKjcC5f1G9*J
zhI%SiE(9liYr-a07PE{OkP<luM%Kp{&Gf1`!U!5!)-5C{0r9KlK|M$_a9hteLKdsX
z9jF#3hpL9n2rkP0Xj;$Nr3qefe2xQDwMh2K%$tV68LQKj4;7V=p6%14h<>(!W7;kF
zSNYk($#`hCfc;=mY}VYy=tb9W9&Yu*m-Y^i`o7}}wVS@cVnK4v4-W50g%+fQ$)Os$
zsSAM=lPu0ZGWrmUsyj};d@61)J|QJtN(k7PGKsTJ&O8}bcDrWc#%a6Nb<U#Z1ou<u
z7N@7FeXjmhZR~b_67<OD{^hwe`{e7+bs^A*Q^OCo2URCcsb%*$*d(BX^xw^>Fx}4K
z0bv3446lf)JEmvO_vuy@GrKknaTmt(#O!=6%})VCQhPwlwN-q!Z3NEM{RiI&49smX
zyHHfuonm=c;4(6u_&F~7lu$aE*5}3m!yGc<t82teBG6SxPo0_&G6n<<NjMGgN32^%
zly@yUG{jMqg~D-0%C3OV&K?{ah}hl)INO(ORIZjEhy1vh<jBjI#!60Z8IZ}9a`p@l
znlcLQ5)e!z#Zyx#3*!DDwkJ1gc9QP_J;p4BL=)0shy|Km+dNRR+*f+Ec9i`9s+-T<
z1#tDTNyvX{+8b*?>SL>jk*}kxgxMvfS#r<%u8<hRo@59mGA+8(_);UIE*eD3G^tre
zQ}a52gMkd58JuLi_I6f97O9RATzIS|CN@9;2B7huPkAL3Zg<Yyc;7hhs$jff+R0C(
zwIpUTdA`+p^QPj~thfS9<gv)=@<XHj8ICTK7ZKK&(U1*)->EN#2q~Me_5uxLo&=>$
zq{QZTn{RHzPbdDxy!b@YUe38|haUiJdeo|P@^f`0&42t3pZH%L`rBZB-8HH|z#Y<n
zBcYRfD*<3&0=PjeI|H8Hk3nj;kK>V;x~^t00SJ}la}8cSoGKsX)9xObkG6j#=P(5X
z<4P<)HtK@tDE6mxr^%GdJ!!Cb1A(mFkDV?H$*AUqIM#~WsB&%jeA2m}0U@JE0)B#)
z0p3vp#$Rz&j?LL#kK1a;UrJCPm5_9l_=f&?Q8Vq(7QIJ6&+7ckElq)PDhc|elMFM)
z45CcABqxGg_0h4OPM}v4AHygiDbU%L1@ccbnhBK|<qUVA+aT=wX~_!~#|b1`e0%_e
z16{@B@jBdfR$m<L>RW+oS#aSF{W$*^KAm+}S`O+hvWWH&vr99uu080?9_jRE`y`}c
z39~obq=|d68~X8`=(~hY_ISaJFbd_dl2Ol^`i1E#&Q=<U(;{OldJ@8w){-<nEwM72
z5O1rAef!E(r5$?B>ZX06y>GM&g4}Ij5Q38l0RP1N!o9h~y`?v2&CeC3RppHR#WP>9
z!^HJ@MyS3>Ad?b!F%Fb2Mz@;gyk7lx*xotMe(7FYvdl3?c7uQa!h}Qm7bMy*rw_K+
zBq#W7%2YeuPI$nQkw1PHmZSk8kndFS9znvAg&^@^8l$vRW}IAuQeUER(N?DlX7~pj
zOp*2TIZ&{prGXp6+cb-IS?D5u+3Sevx^o|S8&-#aX<-Ozj`R};4tjLYWCiT>GQKD~
zXOZB;!vp(s($J?psx-;B2G36#T!?d?3eARxyWS_nkKH9G^tj;r{7r!<Mhg>mXeyfH
z$@4Q{)n5*x`z53Y9^wj(0tMeA<1PI>YkS4bDddc~lsv*56&rDOJRZ`=2%H?fel(9$
zdo|**Hc?&`#+-bQ3u5>En^PzkM$&tFIh9c}jrJRGbU!6?m^*^gDImo+*Vc;9nt!wV
zYcIW9i%<4K#t_LUp=%*RSLbhk=#OFE>9L&2lNuf<Oqyt9^n(ugmMXo>Kxh{3zblqT
zLu!vhK{8d`i~Q!vwmm9x=Q`EiLAyZhK_(Revuy+(t4(Qsdk<m)D)KAnSu)^t=x3^3
zrQCk?-vA3GIQ&R+7_g>U7Q7+E3W$Eg$H$DQ;-w=?9+KCl?XSn6tRE(L_-7^{{NK4%
zr76K#D#U77pk)}IF9N@m&ikk~fMt&BW`3GNJ=;060=%uM;;~XHv7)8Vr_~STBtN9v
zF>>a3#qVR`uRPYtp<fJxrc)GtzxwaUr@;@#r%Wc@N$kXaU;^}wK=+|(J5Te+1ypa)
zfHpLu9xem9={lP#mTc(L_3RzKE>eN8|CGT!HfuOV_t99?T>d><r#p;NKOp-4s*Ri7
z{+oxeG=M9r3tHZ6W$#IO8q5tPWF3?>1KsFt*`XOAqN2=>Tir1W3ltUplVN`Ndr&_`
z{HJfNZX_JTg0`@yikcKOZvv%6dOYygO)Jl3VJqznDHzu)><@!%!=nD{0_rr?vbZtk
z?wxG-98U#7JU;U^WX;-7&eL9gD5sc`DiyEf#_E1%zrwkm_wlDii%tIWg&;XRDT}{Z
zvp*Y=QySFk1=$J4-~P{W^t=UoKn|2G{~?;C8x2zy7IgUFZii;RU)Yq;bLJ_1aI`H5
zJ(bvwJTXH(5bp;B=UPRw))(&6aI|Rn>_Mt*Ur_3T!bN_vBIkGAgHkt}jq2XQ1O*vo
zP9l{;d~`?zRBDdazi>-hno4}YqWD#n%lA{L4YneJ0{u1a6618=Oa)0Ty7|KYNv*Mm
zxqPAt6p^kyVwAuqkX!<KqeCOQ$;ANYG^8C#aT)<>%9Imb*Kj)R<;TT#03I(@Am<`{
zy|@N(-Od+9y_rz4%VxAx_XP4d2i~))3|EvY1*{pDB|5+9G9>jH?%E0DrgJ8$FoANg
za{Fk<>%?QA1yl;kk9vTZ!Vfg<d>ja;w!+|t<Q~wuq2qqmn#$gSLT5?Sy+E5YS&%wS
zp$Sh-kmPNWS(Yg&on#)7C<b;+SUlYWl_^l1y5h@l12}Pa%-$2B0z$3h1w_T`607i=
zK!`$1nBjzr*XoYcr4GsCr8N!nmaT9?yB(80yq#)(_umzIEmz?p4g~YsZO?R(>1Y5H
zDo1a_h39Be|EVk@)E84k^wfcV+w9A7RqhKpQnFxNj|e$aEJs|_m%Pb79BgCbfk5?#
zK)l|6hs+UJw|!MvV`oPc;RYN@2TUMDtdGlruu}|Qw54+S9isWpLNz_ukH>;2Bnf7t
zjFkE*P>}YSTQ!nZLTKJ@%kAiK{lzx4F2>=a%&pHSjp21kD3j&0UhS;7)RDO`Yd|_@
z)ctXUNM2uFXd<f>d8@dy_wxLF-PQDNS<4^3ql24ANq4%J3X)yY3F*6HbDjy94Vb{v
z=Q&N;x6QwgfaC>4t&Ak#`@*p!Es5QUmDT%;de;YW-I@;OTwbLBzpQWe=<FZ;WfR4H
z<H#Di#3ckHz5yFp@<7xZ#<4K>1}_?b*lnO$R~n39oV=&8hRyO1R!nSn5Uq6~&NNRO
z$D-zS5FaeUMzmFqFHQ4xc1UOPim~w@VBRkukkI~t><ryQ)#NrVq`#Gr(u4e@yCB+R
z{t-uXZlRbo|LszsmUT_t?6iQmDQAy+S0C<6{suf~gk>6d^3*i7ae=>d?)U-MgV&>1
zfph>pGy!tyvU+RMjac7MfWSW9AGU?7BCADO1TX^Bt0%uuUFkX(6Lw+R!>`$=1P_0Z
zEfW?uDcP5l@l8^cIXyK12=lrpfL$5m>9Idkz#tJ1?~p|aZt@+)lJK&QFm-~;zX2rw
z)P^hd@qwmBY8+U7QfR$|2?~o+e1v)LMpNT-n2o?cof0Cpti+QCCFabTJs51)MhT5F
zD6@Rln5ego(!K!1cm)dwKeB7>ETp3uV|ue=6BOLQr3BUpDS}1}QaNrNHda2oeetgK
zI`_%9s*#bG-lZ8Ok@_zcSXw&Td*7bc@U)gCr9u^T#UVx9bZ+|?_|DjHg)Fc8wCDfd
zsutDZb3nZ6kx+_3tt_c;@0M$Dnl-Q>Z*ZJoolD{0!dr)De4^BQ7DzBV++BsKS9lR^
z5bspg^HknCC(;Zoe(NSU{KI2`xM|pzz7l-R6-HGo)ygoM!ink9dF_LLb$F64`LkL~
zL%D+2-VmK&(G|nve~&6%vz~FMvNAf}j@yn0@nr3i9X5qm0Se#HU-jrhzu@McQx4ms
z8elXNt+HZAPS)dn?9sOOalW=4Rwr<xf0xYL`G1g2_xFfS4{bS6k3RdIHWUzuu=ZDD
zC|^=U-L<i;o~$I-s;p-A{TpH%sr9jp1=>Gs&1jRK^LJs0*fex;WIzY%s7k7uGF?pD
z5UYXhi%cFj?7N7>DcCgp`*gT}7V3>^`ulXGOnujHVaL^}*ROo9N7TyJd0tMu)EyCz
z{LHRCb#a;+(vO3ZgbzjJ2?YdUI@6L?t|n|3u|*fxZ9gCQye>MRQKGos)6ovQs2U~M
zcsEh-V<auI*Uoq9j6>;?V{~VKF*#`B>inC5q!*M7l^g1mIP-A%xg%UBQ{<+^x_oYd
zu6HLuv>Fn4`qE@m3S1=TK&AQJZPbbVg(;PuNx$Jl74&75W%}X<x}wMZ_Mu-v#75GJ
zkZ<k}7`cOt-U1gs9ghuvBy=gD;J*w_@wW3Zm;h^%-W=^lwr)45qPa$1d8UO-+?A+<
z@J@QN76%DBT*}j~89<k|=r1YAfKFRnt5C-C>E%FlIMu5d8GcBuG5+sEt1LI#z~_eS
z<GT(4$sybd0v+)IGzCOaY<7svhESVZjKX!*2K&2@kXc!C)1!jpGncIj@1`XsH)j3Q
z)85czSc?qM3I{^wc^$`b>={91i|aRZEAs#xAh20_d);RpcYx>?Z$ESr9wZj|2RE;d
zxhwn5c`2)wxofF&QFiH;#Y;|gw_9kFp>)kI(I2XSpYhM_yncRL89QM9M1YfNyVLBB
z5;ey?w7jNn+Wz~}R%VS%0_{TBa>CLMK&LW8jmt1VOQ@8xL>5@^`TpEX4y*^VV>*v{
zJFKuXJX>6AtZ?0mWvC$H!7|}M=~gpuV&;@pPqJA7Mpd*3`bGc&r<qO<(8z*(BisT9
zWJ9h-@I<tB?H*w8tOdh!td~5v+ORc)9?^M5Gb$?A;YJa1@~}a}{XcK7BeQHS<Hdn(
zf~py8BrPRIEOgt#?W#D{o0A(+Q>K}kFI4yjCI~<EX&s1HvGM_l*gcUg00MF=@eh=_
zkzpSg)(-AwE0gbFLXMHiXv(55N5d{Sm;;=@O;#i!$f;5`I6}?V<9}Q5a4IVAsxspw
zN*4;(g{ECJso_DS-6>X>C*yr!m440BS@n_)s>@`3&-}v+SO3CQQg|5L1Y8ugIeX6H
zV-MXmS_Vb^=IS8NE6Z9?Fi$lQIDXMXAxKIjgi7C0J3#a9xTww=uMLA1%>?s#Mj`WZ
z_A_%JkPcdG{?eb`ic0!0R^9Mw%P0rfpKmoz3%z;zvsG3P8~~;SQAo}*n7clKRtxoW
z>akt}0PPzpGes4OHIko0Nod<;;DcDh2GtR|dlfTLtQ?r+c>dgBvLH~eGF~2RKYu!@
zutnLhEJiS~dmb=-JZbTjN@GoZFEjm+c_(uPyixWMZCL0-sgCdW(<VxE{YHdME61I}
z53vJ^rr4IZ+WSq891e*YRR=#lfHRFL!h-|<&QJEiQZ-03`atj@x&548grDqs9;Ep5
zGp6&KdF&Vll3cLM-(dlKUOA{3WN5ke`$marCCt4f#nf78q;zydSib<a5d7ld;JVTp
zBQ)V*1>?V)gB&Q1kekOfV_b6CfKM-<GPnw@@P*WZsyu{<#)R>Zj(C|{t-C3*+lFj4
z3`GpGd93zuC1`)qqez*2lT-;V<Xu``Yf3K`eaGRD_0252d#*(aF`Vf0_Ku*k4fkV}
z>Zs1n8_UUOKA-|Xy4|7c0Tvn|Tk4@k{t(u>$PA4tSJAs#t~Q|8L51dLBDs_tOHZ1f
zd5@3z^W(spZziXrsR%t#kwO?d%aixlqxtyG)pfr~Um@w%iD%lHv-Q_pDPi$Zarw8P
zxx3~kE1MZ3=hoJM&r|3dK^jp|$=D)M6a8I<EzOSNB(QGig)CP<eUuz*;cZ69*nuX)
zGbHnjUduJFas{1uNXoGnckbUvb-G)yJl~);P1cdPAq()Gnc~98k%*-|p|SlPi43Vz
zp2PO=RgKBJ!?Ns=x%Dxj8$pLR(ub>Q<e+qgBjfg6=iWQgK-O3XbWP>&0lcj&-ngh1
zjb=(R)GMTAM8q>>833_tz;JxS`#y0bthorEMvM3V!c);H{$RNT2SQvLxB%JLBbf5R
zfuDlLSpI3r4c{5;>T=sz^#H_#O#_DFN>Y~<h&7b@u5DbCxzCnKI@W@Av4BAm7rq|N
z$XR-aQ8F~uaz<r*Iy*jq0+QV!f1&F<)T3G%Uo$qkq6jJZDLbje_MyvB&c4}A5lwM(
zLw&T#A|)X{TOu|y<@x;+7k|vFUD{O3AQD10B@jj`)<4O+LENi9lA`=N!(s3iDn^$Y
zvdL#aD}t-(eOL?raN)}}ui&ys(o&hQ=0bdYjn+ef-IpU+boiw|TmP5QqNg%jL)*Z0
z?H$jV8}6rVbNBXRQxD@P^V7?Z0r3W|fow;zN737Nz-Bd6L$mTC1yjqPl4?#(2BR-~
z%6sUlQ06B6qaom2gAiEIZA&gU=3jPQsIt9bWE=X_J;Qhdo74rCVPt>Hs1$7e46Nj4
z*<Fd!ZxPY2p7}SW<5?C43TBNT?Qum~`GRl;9{G?nkr&chllXyHf>1m=D9dn?l&%EH
z0eTci6KUNM4Hf<<Ai#iqbo<*ToXcJF#v*c;tb2jom+M8!N4um5(@xAu$Db)Gk09Zb
z;=$TU2$8^Kn+xr{A&5r>6|M6$YXuvdNZ|Y`>dHD|ZV(B(lj{Rv(2*;HfGB&3omwie
zf}O)syL(2`5UQM@1slR{kqt_DB--aCQkrc>eDB2yw8fI~U<}=n*^v8i1ccgC0zsXz
z0sh}P-rgO%a{7-py$;=8w*!SLV@=@bZ^PLx$5R2ZDHrwRYW2OE$p!qYET+dtE%2WE
z!ezYRgHM9eMe;ksSL7-c>xNEpwFcik%h{_bQ|1pdIY%;)1BN^%tYJ6Lee#!b7?#&I
zu24!jW|H?3nK3H4Bz$u*a8p3$hPa@bJzs8=Ja29Cz593x=rJ<JF`(ym{OGebVwI}{
z?Om0IZQ^-4k7edNOJ!6RWJ^^s3qP+!OU{`MP|Ew2JsS&_*{TIC>-jXp8V&2LyF`o#
znrB_5LmB3-mAb{X)<I6(!q494VX#<K+@oN46cgi1c?xA2%p_*`cHnLt!E?Sxg&i$r
z{N|y<cFPb!kzFIPluwQvb`I@g{tPc5W`><z;YsSUMt7?aiX%0sZ+H6h`ff^%tasdx
z?0R}Lj4QM)TSPypM-e?B=!)i}0hr$Ihx~EjhX;$+B0P1NDBZC(r2doDgc3{Y=T4o1
z2K=<fwN4u7j|$p-xnim=Ue>HCD1S~s*-<*tB;!C8%`RHYY?EUdCoPQ6YszIL>9p)n
ziOk*>y}8Qn^UU1*(y~dlFG`NeH%m&rWp`0+_j~RW3el2csJo!d`i8kK*f|i%+gyC#
zSpkEL`*ciGK5R{T30vEtJ-eK6#lr3987-txeZHxm!JACf(!?sBO+RFsdZBI*3oRby
ze0>B(F<_%5R1?>sp*LAU8F&$~6IN&noGKf@1-7cKT0>X_@Sz%ulzX0-QzOw)GsWG+
zCI_?FYE^K7H(C`&k!h!fu#zy2{uY`JaBR{Lb61y|D)2oj8SvF#<)D2pU?$u)xILgo
zu_+`dNgT$HxTj;|A5-6YSYsp<$=P@hZ+ks-xE^X?z-Q^Uu+LXp-NK(}GTsKWNhKGk
z@3-|@GiYP=Xzm^N@4`pIhdyX=q~?CiK?BYd@&{sQ$V_D(CL~auc1bmJ)6H=ofdf`<
zcwl8_eVqiNq%M<=niiq$jw=j_JO{5uZQxcU)%)&`;OG>_MRFobMk5G{&U|+UZLTWy
zWH#hQz0_yjDc@9u+8I2!t2w4Ja*xEkvrI?&Ie?Z1{pxU7e}tBmrs(?8=tCm5lZlSX
zo&Kl7j&p9&e-6B2H$The*U!0`W;#u^&jqs^E<VI^0oXTJ*fk}%bxY<wSPTc+S%*2m
zmBtQ|5XCUFgm@_c3GLBYJ#HaRGG?yRBKD0u3C?>H!Sz52AR5|*MhVaJ6v)Qyd$okN
zWH<KtU%Hk{I>_o!-=uA~Ku@K--$xR^^6~fIC)24?Sr;~=MZwJ#V+DJ0uevEQekBN6
zq;y>0SIX(`u_@*2fUuHmjK9#`kP|h-pfBrY@c+kp(g&uw)w}KF8!$F$i$Im!(-~KK
z4}H!z6tAnLd~~F_<*e;7NpjXxsGob2z35nRfh;}$3VX_~QL7F-ep)h-e923H?ERB;
z05@+)#jA1(;MAQAqRbfjag_*#$&nDNp8raa$-HOzO3Q?F^jf$ouZ}KMIx|qU{7Ss(
zNSYd!hvAKNi!weZAcNI41wf>*!jaF&08gGngXLTWsx$$yRO3Sd%<7qug7c)5Ql0O(
zKL8^c^Y(g<x00Q@Dty>(e6^s3B1vO4)VM%^uIvE8%vfI*tTn<AlWYVKiX^sLk*8g3
zGUq3#KUt8vbq_rq`ifih`2V2iKQJTu=ukmz;883+AYYBaaIpwLo0@HN3o&M3H;lB=
zd38q?t=wf3yk;{~YTMiFq?34R=32{ec*2nEM<K&f1IqRs{D1P|drpaTw6A~Kq>BA<
zZwJ;_H-Pg~TFyPVSEw;x4*n2@fvm80k3Gx$2A8@E+b8z#yE)ERTolF|)(+H8<{@88
zr|&8lWAP&<hwPe~fihJZs$SGEEC?~G5*5d3k-dd*J;2CODAba;O=@}}dHfP_e)?(l
z)*Ms@yxp<ZR1QT0NGWq^aV7XBXmi$8`1Ej!mJ`_4Oj80H-P}lbJh-QW2%g6U$PbE7
z>X|*Wf9qUAUWsIt<AyVg9B75icwW!^Dh{gN#GUJxnB>c~J4~*D;vjOYM7}zlDgD~%
zN87PqviUv_Db5ce>KMP82vLaU)V`sOZOTcpv!!zlCuglwG$V@N@HYu9KZs>Hb+wS)
z`0lbe7|O(b;29J1)-oeW$sxpy<2(Q|`m1J}I9`qeh7x01@9==P#c1~;e#|gQ-Jcoe
z%>$1;E8svTFrL(~7_)?6?%d*INN=L>9NjnAS1J&YK~~lOOo|xu<Be8}0?jN97U`OE
zGIH$pF@IrQQ$oDir1;r>2OM&3xZxVy8Zr(f!bn#4uwj#x>P6i|0Qz8J5K0G3vh;ga
zl{ai2cRGz9<>>Fsa^WXZL%Y_RI8fdt-5J;OU9K8&mP#0Z8G-9zYakzC*}t^t6G|N=
zD_9#ynTL6V=flR+Q3f(eNBa2K2QXMjnXsI=%X3Lmg~cXR2AB)GF9()_Bc>d?T1SSa
zy<kXnb0D-eITk^L=tdMo0|(u%d%pcc1-W9-EoI+*kZi;%Ce-4Oq*FxvDoKE9o&5&!
z5P^{N6mFbt^A3JTfZ3Zw(g_2cw=5}VV*x0Pu+Lab7a~X%GGruKnX(tN97$v`c<qKM
zDZ~7)FTu{35)QLqZn0EdLXV>W>`%x}k=h^VoZYeVUih;X?U(E<!SbEEHmxJGnzjaC
zK7eC)30qtXKRrV%Lic`hA^R`k(RFFg)l;atNgR5x^F?E+B8WW+Uk#_5=F+yeApJsX
zhG9E|g`l<}5Y!EDt2TRwFlP3{h7bxo2T;Awgld8|%ffryMbX|K7@giR#a(=Q{Op!o
zG>QPDW5rfCo|959^ymAR#!?;pPjrmyE^u)>cy@bWA;d@d>0|ZroR8blO?6bnSwE;>
zLx`%;I>v)RL24LRzb1=ceeIS4yDvY1$uINcwIhd@bF}7p#Zsx|TUcGw2}Nd7Y|m$<
zW(W4a$q_`2&~kBs+W#H2qFk;{VCXi?uV&8n!hXK6*-oj0t;w^u%Bk~*^-qml`}SSm
z;QA_mZtF$#{@V;MU)BWg<5Xhmr|w-;LL21YIUTEkWRR$H-qJ6?6E6kXo5b!HZS9M(
z7o6?HQhXF8dqg%il4f8$RhtQ(pY~{k8yY1nfecQdI3;yvcXomh(eZJl86BfHJUSM#
zGW}3=r0*t^@Ig+ph{0s?k}d41+%uc6;^WDCJeMsKn^7i%lUtVi@TLW<ch-|N22I_<
zLYJEbM7rMYn=ny(J^=J+KduWb@NFO=Zo>OSu$%A6XRI@@{}F0oT@I*tY0u69XRCwb
z=SD*+8fAV$(pAMsd+gtOwX`xG(Ss#BKg`5Fm??No6iuxvScDnoiHqy>GnthP_G<2p
z-7Pcu>Y#S4D@kS7WiHB$>n=4*BclDJt^*KByW}5*AZZ|EaMpcd&{J4y@@U@dQ|9P%
zG9x+@M}WOK5dSo+Le?LT7+9MEzcUqy4z$Okq{T2KyrBewnkGn`#M;!N2#x2QCm+!y
ze$Lu5V17UO9AT^iq)to$m5%{E!Dx`|qwP;KHhu}d{`J9M?7fuiSOJ7@Tjp7H&7>r~
z+A-SebUAdk(cA8V9I}~3f(-DvRO`<PzvAojgb)1=&!$e?9_^@e!D{GnD|Y$u^927j
zZWu@U?IeUJqOVQYiQKKW3`b7kBa8S2QbK&c5$%W}@Xc2NAh&Y`+Nw8CvC@(YM47CQ
zZKu?$&ayB>{a<W9LI5Zjr#VlS#w1=|bf#bk1s%N2D`5+c?S5_2@#^%Qz6GHlcW+;w
z37X?Oa)`1nne@rf-j||KSi;07t7-LL+XHVgatdMOAl(bJyg1hkn%=AG1Eg}5&k~&I
z?K(w83F`raVqVhfN4(FL2Xe$ye-VDpJen&`XHSHJ0poC-qcWt6B-iV1kLHRV40^iL
z?fizbg82rG%&_vlsDHtw;fh~<KLQPrmnmT2A@_6_Ls=Z)8Gs3~qiOYu;A2(7!6(TD
zQPKat*?HBF>wUZ3e8p;@0`8if_2WR3rbM0V-$`P~qOKT`uLa+8H5g;G)+g(S?x|tw
zUZ}*L`>8qSV>-o-_7?qWv$lVH`FJyXT=QuJ3wqxUwuzkU1(hKMRIPfPKB=qXI?laM
z+U)vTG=Ow=Qo5CsJsBV#WFJ3Zl+_4zLguvCm@IYcQ&{E^Xtl@rZ${}*?oL1FD-Qe2
z3^$&|Pz2#g<JIG=>TS8Zv#zou12Eo=Ik?4%#@K`xO<uG{)^CdR64?qF-)(~#IFpMe
zPvuwy44NzeptFCH$k1@7=$MYT`*?RO?oJ@Ntur`t(unuxZRsUqIq4u&j8u3|lRUX=
z&i;yXc5`<ghA!@2mbND3y3{LiTt7;m;sB!yJF<?(+!r@74_FKj2~EIGW02!F3B!0@
zpl#SPcpB&wHE6EqiPQGk4HP>pQ|;h`WRmJ|pcgkhT)Dg!ZiLpKZwB?`9dIoab3v=o
z=JyYA!f8>uAZ?2)+9xU*K}%PQd|Dzx{Y*|b6+=n8a#XRi>m6%hWEPi``^>YqZyDt#
z529GsMi?$M?w@{X=9-rjcQtKxnhXFQO-9~nfW4B%j<?`mbQxMp3JtoqiIT7R-Ih1y
zII##X`OX5rx@6Gly4h1Y^2t0z0#;pZaQ=$xt!bo|FCH%t2BJL@Kq3u*-}x#zjLn-=
zpPu_G-bm)vAJqcvbXykncK)wO*+dK_%_=Mb<9-z0uS(jS2Lv-r`^9F3F&W2O!7}S1
za;fG&ZIr0<M2o?3AfS8w0x?IzoZDa~$DLg8i2vd0otlJ!l5N|vZQHhO+qP}nwr$(C
zZQJ#g?Xx@L-iPi#v2ta`oMSxt9LtP%6;-c?RYbz#QTPEUUMb{@^J<qn*n8RTeml2h
zg6%iE47^Ly-(<W#9aAZ+V=5|qZOgd5hMi3dYr+{k51t^PJ#5q+;p75=j)7<tlLCna
zlP!D;5SH^ZVpg+*yh{!-+Nbsj-N(6aF$U(mS^lG8A`Phj5v;r==4*|3$PXMb805EO
z5{=rP#x?~f93&12W@7|N@}gHkWp(6q7QcujFw-!axAk%Dwuc24M+jzJpCb-^+Z#@H
z+EcgJpf~ua5xBOU?iW3n^9;bmD%MPGg^}j0;;?*dB4rvw-y196`VpB-o%`{}Mwb-x
z1eTXE`aB-%T3E80$wR3|hIYPtw$+f&T2mIIRshVQ^~Td~vj`ew5b!Y$b?jr5LM?O3
zMSMy-9)t->F*1;YJh4Kuuz8H=9|h(0V5v_$z=mf}e%Gd5t>3hHR$IVP=s>)>bJggp
z-Vg5}_d5%#x?{9YaE0p?W{iR_1Wl!0(Y#6^wIv8`T>Xk&wQD@QG{y=t;p&wF4j>e;
zOi74w5#oc98$EhVNf-Lmzf#1d?h%oA){~Um=W-n*D6=^KdMZOcZ6P!0xBu6>Tw}<M
zJoH?{opzY{HXb*XM+8X<^t;3mqhb<MZa<BIVT8$m{p#jjNfxRyESiK}0`0gqzXV5@
zZw?^XyXVq$C%(C66~j4uwP&4P+0dq(@l&@F+{v{*)$5_x_zJq*M1%b-o7I*=9oXBI
zZa)_ph5qQ+qRJY&j;MGpSyr*2EE>S`&&!i0Z*Lli>-wRSZp?3ly5rFO!siX<OeUAN
zf_97?3k87A&ZZ$Kh~~P|i>mkqE9hpp;iEjBtekxagZeP|qxBt!k602>B^Z>55Ib@A
zr?&zs+6CW4+eNR8s}P=01t<g!YCJGN0xX;~`nK&m)D7_o{A_eHdbPR28gc^pXx#Uo
zf@8QIQ3T6D<)YP1M4>9}Ci35RQLw2Xnmg~M_V<HF6-nrAJFLD5B~m_GuQu&K!4`|U
z^6#p*Wru_-(CmlJ#(fmne`hNP)ghfh-K)RB83q{w%|4Z|9@dDS5MwJsTQ)EuhXYO@
z@KhL7N5X>fJg@h!PM51!rJZXY-j?{2=f@TKqwj7~@8!s}vpob5FP(G^HE_09k5-#&
zz_evga@IJN|BYCzxdfI)P*#jGEa|uBuVE33K!gx$Yg`Y9t<LDTUoHqw>8m^X^O*O>
zJ%pcszj$$++Huh4E6i#Fj}s|Tm;Uk_5!G36>q(x0s(zl$xn7j$ZQYL48tNK?v}r0{
zFlWxH#$>X%NJHD=F7n(ywThe%uF1|-ySNn{z^k1{;e^w*kE@{G@@>PY$YO>x-6y=w
z_F)nMOQ{Wzc(Amndrlu%uOGW2(`*gSkSi!&0IaM0+a(c|U~zbidex_#hA@n5u)K*W
zK3SFSHQ1CF9HLfHu8XOPksmCBPcN~T`mxLV_RQT@&QdDS!T5-PBOVDTnuSL2X~QA{
zw>h;s88#jke7Kj%d<u7dam)d*h|s=VwD^RZUcJSfQ^@S>F*$q5Zy~Cbtz4lc<raaN
zSaXnw0H<zOu00{f(~L9F!oLBsxavAXm0EmdK2Z?WRU;r4)8k*cpImfm<U_drQ~iV(
zmwR-;kKP;nq<7}#r&)O(_pvb<c&6CwZAjGWN;;pV&w4br%}YPIq1v(_s5AV*xK-(t
z|A;jT|L$T_h<B#0sld0`l=jQ`cWb7EI&d7IwN89y)io9#qNp&lTi#W}P*|*o7Ec>R
z69Z=bdY!YmBR;YP^JtErP5cb18MPT1fG|o39LO&`$?EdqlqR8P%i2R~G&2?OC)jZ}
zdut(#Jm}am1>1p04F|Nnt8bgNz$qYdLYJM=D~CGjCN1Y{+IqkHbXN_ESwu$tnijeZ
zir?RhS|nn;d{7;1=6drAlDRmWV1?mY<Uk4n#u_CKq|8##=ExpBoLpuCi@mPf?1faK
zbly0_yFMSlGe0=_O~HXk7WPg!8yyGxP7E+<e?!dq*t1cWAY1YUCp2kL6pw*v2GpqT
zl5=1sVBue<YO#6{E$tX+=)A7;XH-jV)cSApqFlCZ@5Hmb-#ztKYc0ljzG84!5vJ~?
zr4N5Tv8kEPIrIhQ*#wsh2Z8vMrsx&syy9k^o_$qIy5!5ZA$gu0VWBFb*(kOW!xdDr
zA$?=H5-pJqr0_>IA17UYxYg$e^oXWk676GN6yvFxEHD4k#fnU>r>ZEMDnX~+!{~)O
z+5tKCLLX_&8xd(baE@seQaDEVgL{$crSEHvQ8qzwawetRz$qZKy<t8l;8-O&x0L-N
ziHIfE=R_Y1`_M%#^<fLd6X-=xX5QugtWuh`Ys2ztclQ@2tg!VqnwND+%#J=a{luzU
z-@ZfhiX=-gK(8jDGpKRw$NZfe5uoOaSVtGX*7$Jf_#o^8|5*=c838Dr&X8rHUI2Un
z;Dbd^S+}(XUJwzOwG7|wU{0M2`hrKRU323OBk+GL4V*RL{y~lX5ZO!7M+};VDz?CQ
zo_sK+rYL8sK%$<0^V_!_<vzt^SzV&XBO<%MD=ExEDnz-mG|2xASxk`(3+SH~HPG+X
zxHl~;Lg?S~UO8h)*?aBWWk|=nsNwJ#36PBxEnMjr2<;#KX#;`6{NhkG0_4(7`@i8}
z9~EE+h5|fN)4oYk5<`$m3cVbJ;rF$`K&D_rkx;j7QfhaoPHxgycdFj1EW<N!5Wb71
z=~6KwUVjaCQc`uflYi;iJV9g*_hU#P+Idv>Vsf7=I?kNlx_jhy4I7Qrq?X91v^w3+
zCRI3j%%W8P#tCJMg1R-j=en8vct=d0T6e!zrfvRYNj*T$R9J{u$L(X{MsaZ)gZd2&
z#@fg}6+qw{5#}Xax$?7j`?YtDBx0Td72In5eb<TYQJ(3SAWOCG4h`v+_P$0V@I>Pg
zOIp`7sZ*SU$+q$FF-|faK+LYkJwh8`=HBirQUXL%KH`)V;p4hHJGLmxnGljuU8~{%
z7{wC~#n%fpe*Et@umPznqy}V?fID2HqI^DOo@bqE$+ZhlZXiY(dwZXrm_?Kl^6|ks
zS}n8HN;%?V)`1E{yc=${!siJ##Qj&}lHCQo=`WYKF_~D!^L|6v4@v_sYss<RL<@)o
z;)(;t*@s!9mfw?S%B_m)JTDoTvD>CRXc+ejgx7B>i5LfWf1@tm0@kXS<vj)PSZBuo
zZ4qrZ{FEg_RhF@vlgvN(A4E%F3gSH?L(gI%b_iClX<3_7T!#9qO21-D@G1PHQJA^D
z^7mcdQ+Rz0y^6yn&?vQRE!QGl<@|7eg#)34YS4xCB<F}Ae4GJ*(mjfs730y+7jhBg
zPr~kD4Rb2AlVdikOdLTyY7XTqdN)se!$9?@9z;Ki8?2_18$K#H3!}O>il`ourL@5k
zvg0boOGQ9Oru1<yy)gJjKd$G?a@BZ`j6mu@Lm|d9R6?3aj6<Jf_o3s5?XWElCg5HY
z)UOpicWr$@Hdsw~T$x^d%S+r3^^eQh1w@-0gVP^v3<tJ&+gq+G^bj&mAK$iRLQK<O
zpn_`K^^A^5hObWog}cb0+!j9w<j!+I`Q-V(iRK3E?1(YV{{17lYXdb#IaA<rfBT+9
zDf@f=L<uBwDom?@d;KSBwE?r5t6~i0d@(ytL07+w#>`$+6O*xEoZ}I@Nh(=>9TC>Q
zlBj#YNd`p@)iR+%%dLje*_|+eEdyPdLR2M&gCw2vNsbnc5v@dR5jAUjfpu<u^L!Lx
zU9TBxwJD`jWO6w$PCAzKBmXGyZG-lFZ#>4_|AvoSoJ-bnXI=B}xj5Ahs*xWKNDk)r
zeB?|`PKYC#76l$(p+F_Rr{*N`6F-Fv3JY!&Qc2>u%XN8R25b-gefGGxo~Mk1TJ&jC
zL0Q54&++};qtapSjXWtmA0yuF{*f}>)K)nz1oSn>jB#SOSPvLP!v2YKKWqMetcqwb
zI++>~q;Ie(ABd3r6pl<tp?Z<S?9wy_m$Bl09Kysxe0#`({Ba*LXVx=^Bp5^Q3E}NV
z^E(cY7Yb#p(Ko16(u6-_O_&zx%B<@ILWkHH@`n})(pTHA<b@Nd*mB*5yIb(;YHWCf
zHb<LOF*!;4G!7OK%sQ7fc=;-O@Ibx;Kz0j<u;L$84&~^APH)EPsOTkqoO>%=z*LXs
zGC`&;g!3!-=(mWbivY5E2*!U`jpsAvTOPR@Nqxz|U+(4BV#s&Hv*w8@b0RDPVddI!
zjlBnEWzFx8=b;h0_TMw#G!N>LrU%D8jl}3LvoTyyYr52fYH&+*Jc&5RY%8GZ=cO^m
zL=?!mdSerJ(y4XD{bt$E@*gpkQK*x%{QxAcwtCw^b7#?sDeYB7BJ;9`M8l;+q)4LN
zV=Vo3r*;f1?YZJ3NoI2bvBGJ+g;UB~*P3_>vV1G6JqIgPQ_Hq#jb@^^;h#D8i3a$g
z!BkhE;z+r2Y)08G372HDUs!^o7xXLSNi;a%;RF2sS7qSTSq`McnGwa4oYHzah6yFf
z>JB!(1gJ?t!kvfwYf*RfG6;iE|HmzkzxqI}r)itEfH)!vB2NQySYEN@CE&JVQclY%
zF?y8*flhlCCJpB?t14<Amz1o@<fVN_TtVj=ps*Qq?NzcfN0gqT?<T9N(jcq%DH4+=
z=vF`mw_=Jpnt(83Qe;0K>#zJ`dymuw`@>Bul0ZlTb<DZ^eIbbR$T)o@I_lf~VRE81
zs<D}NcqD1(&ieH<;z<=#?=%KD9hI%^9m|y;3kuec95{=*iOj0Jhrf*NAJ4K*oExr7
zEbV!bCNI5t49&M<mzL-VGwHzq66xUm5xK`Dr+gr+rPg7H%B?Pld*ZIcP=eq{);n0s
zvR^^ImPv&3ev`_LW`|6KZ9$^R^$$-|#a(#@!q`+8sJW@S>1k~`AX>|0RqvVTqN7@k
zHp#@R+c`Bka^(L5E$YT!O#|JcUR}1Zv!-@1`pGH$3bda6##2JS(^pvj<4D$Z{1HWm
zN5$rpZR*>5)HH-%mI~kFr)1C+1XNngW^?K^O9;2!6G7X97v?L%%`dX}`<amiVYhZ+
z=<n@-!jb|1XWzhpX}E_zo=q3TA>)0n<K0Drw>PvHd?raJELVr4*bS;ci}%YaBE_jz
z*Kue&cdP8WPU%O+9`owVTFU&);~=clS4TB$Ym(}rC#xSgn~{3`0K!p_dIXvYc7&d}
z<Bl<oS9|Qmj-E;YphhJkTqU7vr7l^)n0lvO%p~1?a5ezE>#WDyS7CmoA+{Y9@FJQe
zEit_p4uE20G*|IVZaPX+pM%dIH2Vij7a&BKC{2r$Xt_utD3Gf`Qjm(6H&1zUl<b;@
zGCnCa;TI?ROzAS95ov0{5{=;FfnlzCU)sM^qiD&!BrY4&{&y3_L(rKx<Ai9jtL|uh
zJInH2>hg>6t!Gqa0ODJQ?bV>zP0+q?Mcz6gu!_q}F|Xt8iJhN;flqN&8{xOX0pZ(T
zgFm*ZmDq?4(Kk0VN9QTP%{Pqd?9%0Ud;!k<+9~(p+`FX0lwhoSgM-J-CE)fGkpb8R
zMfQp(fKilkBGoMS`u`_sWlrfQ>s+ge;17VeF`tNvjrF7n92B=lq_R4Ju5-?il;3w>
zfg<!G&i<G`NnQ6jym59L_hT#T<3(@#9h!qV0x=}Iwsb-Q1PuWwRgHJ6#V}uY-6}`*
zO=gB9pNukLG{vkQaGEycPw6Vn+Y`1=^k4jv9?A&ee2_LN>}=YocHNPaU-tL$RtPis
zoxEmezyNehFv>Ot7+b8YF~L@RNL|g{RK{;lXM``&PPD$G^1$12uwk(QBG@BwSyC~_
z_sBg$LsCmm>f&yQ128G=jbuEiK1YosCyxJa=27G&^3JW4)Jm+wQEBY?Alofq!aNR#
zci=Dh24iqtt)Sd4V;kV6QFL-@+F@nDjJ>3=-T$X8l1ziRJQ?=fNjfl`Y@)qXepW-%
zV-EtlOW*Ha>iVjH_*=;3i*CVACu^=mbae%S&bF7`$)gHO6eb|uro1p_=XHYymr__&
zY~yr3I)YSp^8(3Kqs0*l=G$7hhLUBBYBm_8EwYd);0_U>kS2HkgTz26L?0Kwmhe_N
z1QUCTwitl?*-t#1J?-5)>b~nFdb%*n^iG=5Kno6aldi~Tk9xJTQgWhp?ZNYf+%?ID
z`apMnI#UmeOeDK?$ye3su>AvB<z=!8$I(%m&bDm`y!1fPYfL?+tF^fLax8Tm^<h_r
z?#j+^_pn(DZGSST@Md@CnzVOH)14TwsfQnu`!?wV+4~uJ<)T`NVH5jlk;C3AEh7Me
zHqM+(z%b$iow?2Y0i7=uAVoJFZj|WzdNFV`0ocx)$XSMoEIdkibXkyL>TOxhr2x7>
zz~LJG!lvCJ5>a4Mozcg*;ikmO{g=z`jkzKl%a<b=e5)_6dQjCzJEt`~_diNAGlYWJ
z7wOv6g1Bp!scB07BVHJj#s9{{Vfd)nx9Ctui$yFsf+0E-md*<$FHCVrCmU}7E$=Z`
zjxbU`bZNTKIg0>jK$pKYjIT|cXb`f_3K=NGf#0a)SoN5nzJ=c6ML9wS+y_zX7TtKZ
z0cyDX&IgpYbf^Gu4Kl4Bkl?Gd3I7&6fk{T?K6&RN@?)9Y9<M<~t2&=alzH~bh@+w0
zd8zW4kvL~PD>v6BoBG<4JwdOf#g<rE1@*Ay_k_DIg)R@18TM<@cy=}dnO1PY+YTZD
zA@6JE@)+rD1&K8y7|s;?L+kPSrIw7e>i?kR>34l79HezgbVQv!g!G+Bu*{$O39VYk
z*u~9dmoXh9xKdnNDttzyL`aGlv1@aO)$Oj(!x!OdUWghvr)VRzn~0jI4Wz@rm8-`I
z(*Z+|ai<Y28juAoSh9Z^$KIfpyT3b(AR{N?S=<x#3M|Y%B7!)q=Z@hzklU;SDlKnk
z8c+$uv@Fe<Wi=0RTP49y@cPHTW*9T&xZ-BjEV2jj-0?kC_w=IAEJ<D-QWV#M6-Luo
zcY_=3{w2BXkF~LM5`^94Z15iVw8wFpVK-YSeX8{Ryk4KyK5HoVxF$#zZ4VdyD3e?V
zF?{TigxdtE?J7h7H`1#0Zc*P%zSKU>09vk@6jPHaQ|I15np3AjI(`(j$zxat=0hte
z)?9#-(Vk9oKwcM73tE|i1)Z(Ptwz?|X{bT&3x)11*m83ZI@V1UfNa{=SP>?!8`lq{
zM!ejrUjQ=9*@gyS)L<Zx)+f6IXj$9O_hL6N^N!Z2O$@v74TI1_;=uM>KHT2T>{vEw
zfy58)djM@a@mz<OV+vS8;UQ{3qxgBgGEnvc3w9D_-4a&S+`-OaF_KIUYHRcy+R`g5
zXD7UlaSWUoa8<Yk%W947&bbiJ`w6cmpAR-`$kl8!9?LPnkIckl_v>sgce2s|$j}C^
z!U|eY(iwiUv?QqjVB#>3_`c=80K~YF?XpV0%cnM`^d7^G`BKE*v8`@G;E{Njty-99
zENV8~V{j>m8fcQaepDarkc8{?!A5y7O*2`ZmhJ*%Z-~Hg(Li>vaT78=(XL^SVr-_7
z)~t!QmB{rX+h@>_33hDo?IhEL*3D<e&4>bwU(?;DaK%f(rJS=K;JkkuH+F}jLHlk7
zMoDOU^$|#%t|8cQ24G?(`E$N$&Xsu?dCbBY@Gn+&b+24MpScKKtKrJvj@&BvdluIR
zta6ni*?%6*Q3-@tRLwTBK;Cz_n0%3czT)Cy_<yR~VJ4~bGySt$pJZW=0V^c&2mHiD
z<1SX+GU777`M}8pFHYO;dkd|PO$hjhl)}9`%v3gR3kSnD-3%`b@@v6xpf*zHpfY)s
zTS<l8Bbe5f9sC$8gqiWa3p;cJ&!G`S{vfX8h!w}A2BbMnf-ZK6+?h+PY&<nW*huLm
zYaNc%+?N`VQ8#$fM_Yc%A$jcDym&BEEEj*mhdKWoRs7z9(uu6A0!qSGY~5{1apy&7
zhZ|Bdjf08N8Pg0Has4)8px2kR6>P&<_w?ruqCba}fTR{Bv)Y7}5$5(RU7ci_gGia@
zxd^3qryTAkdB}bct9st*5AYL{^}f?)v@6iba;>5Dfx`h1Pz}{<6mqcEcpN|df2##Z
z$sKZRu2HWZr4oXbqwF#dZs>ur_6OC--Wyrf9NRw=_k<mn8X~0d^Kj{|2#2PZFzzIz
zaM@7AUU3yZqORS9cfIUn8^z(;o*@WmywT^ZM^L;jD17`(i8qWoeh!W&h*qSb+H>1=
zaSiUiNJ&#Tz3TL1Ih3j6XRhsSCLL>vxcykI^GR9blhalCISj#UmRjp?WI@2?Yjvd=
z{=YLTq5rtk?3QF-(-8?sdC<%uQ9EyYb#y+$<ZrEecGkAum|4QNs&(`7U+;C(5DhP-
zB-zl|ZF*erF{mhs=uYFtfd!p2v#j#_>4m-vJA`QT<W-MbhPcAcRIqs4$_TXV>MH?_
zVn$7+0-O#>q1<4b(U^@?3C&_1HY~kA9v>FTC#1AzPFMHXTKAj2eZSpDR#}`aJ#qa$
z<=LXi2aVy=5UW0?Wxo^I)W5UomyCh3Em>wL*P1HF6I~E8Mh~xvRf8(FYSOw@9cjrD
z2^#L5nQ9;rze^?F9#npni!qH7l>_&aiKA*q8psy>MhfvAS*X&DvLJ}-lmz;e8w*}P
zMxf2|VJCHZ$5Zn+CHe{pdgUPBU7My%)WQL$K%6q*%%|9p2dF9bdRPyNLAQJ9!gmk!
zX>-4oA6fdMMDf5|jWn@qizPaUE*z&QimjP*1HZN(KTfMP&<_BgLi1ZYBSj5`E_OCr
z#tQ%5-<B76>vBzAk|YZvaNJ{D{;|>+`b{>5ld#dF-1!{oZle()T900vd>W}Vthde%
z3h*+?c33zaMYH1#x5$32gLL{=3#jMmGtm?JdrXVl712P7D8t<3xB)5CVp%JL-cxBg
z7j;dTZDZb#=QZO0-=M_Now$a2Cc0F`>75%<<u}=wtpV0!R~h$zSD>^!RabQwKrl3l
zFO7(a0+KE4l2xLJP-8--@8ymxUaAlEnlZ-9q<OhpZ9=4jNK6I#%Mu)@#UB}&==f0-
z$w&GPnL+=hWi6JRGx!)?5m95WJ}7FAqsO%%mE@q+s?mmD2dvEFbF&u1BI921+eOI!
z``nOfm=^#F#R8*JxM1?-;CfatiPE_7vIre_zgx^6d-Ws{{R^;DpkNUaArjhq8K6nL
zmtg7WRY3(HQY5J(dVoX`j%|wU^uG_o5M&UoOlIX))i%F=e2rM;%lIYSf;;x~XPmd8
zb;#XOIwc>;qL9Kjb02bA?Qr^Ti9e|@smc26!#J0^XRv?cakj5G;6s_5-LSFWS64gx
z7Ad9Eah+MR;X-m6bj5$Bl}uJ}o{)GEXxMh_2xA7YK3uO0-YX1Uq%Y2HlDIq?iji7i
z{i{dZ#{ZujUjCAOW?K~3eBD*XQSNzgAKT)`->Et;FuiIcdYM3f!W23Fo{7&OwgMZU
zDD4XLAfxrzZR?0{Sbnm-CtS2lgS`;>PFhf)Q6UWP5Dc&@CF^5V%-ZmKKjogAc>+Dd
zYj)9ksovRi_aX+uUa$aQ%vrF#P&x(~`gm?)lGCQi=KMl_FT<na<QJQfF$5pL7H{#s
z)bK$0+9z>(k%aPN^nKK!NNq>j$fil}M#NJ>j3t`^ZjY$aeL>JaTO9A&5qHs7=R7DH
z23o>sm(wc;ZbNWX7(Z}5IpxuNletW(_wgn0&!}Myr^H%)h6p>QhTMN9R$Zu*-9u`M
zVYWR5#p#&HEYj2vomf`12P(-Wdr?MUxWCF(=KGZF;=q>T&R@?`9+|oQuZ2`&OeK6k
zODGkC+Ie*Axd#e?ih)(S*PQTGe%L4+M1Mu@x#{dUlKUc!@Y>K}@EY=gbp=&ayeCFa
zn6pKMyj1f0*_rjEWa`7n<Jf?C<9v&`+=@0wKWi+e8JeW|m>)$}hVQ>)pjfn;k|@$6
z@TX;n-F2ccKEgoX1>R%Oh)xAr8fMt^xZE+lZpL`5hS(0eh9eaHmM3e?F#k|}ILYh4
zG+kbqFZG2HL>78P@+v%*KYc^B6+w4JfRn8<gK9y|gSsKlDwPyMpN~IoBM)rJEtYg$
zi{JfvYBU%r$=N~R2yH(&)VoE*$g`SEb)WHYLy)kT<D<oX;HYZdY$Uza1CCb#1oJ0O
zYsL>=gt7_<KMmfLsXP&BHs;hj#fN7wPGAqme!)KXv@k5pS=zJyjm7e-1}R%{cO7gZ
z*F1IugFm%D)K@kaRzbB=$vI5yNG>%rJtc;SMNOPs1BGx1y1c`}TBRb;Tk|NZ+sioa
zRd;Oubq-(Xym+Tvd3FssZ+2>33hSeK6Cu0SeAwm8z+jY6E=*msRhf(iu^|=4o_*EZ
zEV;J+`$)=@B=43YNd)R?{3xV$UJ0n>0#IhVAahYkg`oj9@#?DTmOcA#oV^2>HjVoG
z$TAdd;+;u$f+qqG+uDVmXVbrP-r~lOHlz0G(y=0Jkf|%)$vsm+C9iCy_ZwpV*)JZj
zO=m-V1bJcUD)Y<Brq(3%zK)xvII=8Ex-8ru2mv2ec(+{-j#%~e5LqnN9)XuiV-sOV
zj`nTiuO_Tdhit^kqxL`sZhfU^NV^|H4uVc2Do$Q8`MG}8YgUSg<Y=P%AC#P*x?=nM
z0a%!feBPVL$FqUBPuuRMz<h4(b3dhTH8;L`sz7@X@w4|HvD$q2M<(2)Y|yQhdy^b{
zfj~8G99*AYZw%mcaB>}Q+GW5t@hwH<aHx;@|2*TqmvB7+rFwrQDVl6qR}ro#imLO+
zQ{Ak+EVjOf-8~gEx<0U*zkGKuA;VMHW?pUsulMwqqlC+^`6Z1%rqf%)aZIy9KVqHz
z4%jlB<^mpYS;~?9EE5%OH@*jj?Np=D@8zXSbIU{aC)F|S!w>GpThlL`S4bS%?c7iv
z1zrDsYDTCq?2K4bLwgpffN0<KW)Bzf)X|5gC$k2!-cx7C6(7C5?W?{`0e+{=DV*t<
z4qLW60<Izm93DU5Md7>&nrIo5qN;Pc4$s{1K({Yg;b&tl86GZLO$*I!D@vjJGuZ<c
zbT(u-oIq`R?zl`ZNRDPRmcgEIKT~jTnHoJ&PDs(h3Q|kTiVaHQ{7%x$2VV2<&C(G5
zR_YG~9KR*2Q$0YM?}oGuPLRgWnI@h7n(`PLenhyg6BYF918aH`1rkNFlL7mJY5uCp
zAAbwHI`kMwUZX;$Lv(w}C7|nF5V|xwx5uIQy-PL`8a_+hk?F$&Q{735BcrZaYu+8~
z4}-YJc+|27T7#PYbG95c3uSp{IB?2!ljJgfL2uZc;uJ?WH*UtKlx-*gnCB)EjpB#w
z%^((EIxcIwZty~dbyUh-xNF9onG@c4cT|_-p8Q3<ZVTRFnwKc%a&2VELo0Lb7w-Y}
zF8z8AYmB)K;xN@iby}_0dV8)~QS`t=8dNJ7<egGCOWq`O`Us0v9xiwh`O~sr`sUi|
z57h}HsT-j2<#pzUrJ`l2S<}W5rk(;3)X`3Q%^w->+|yj9>sQxljZc?7SYpR(VWj%j
z@=6_&qG$`RIpa+E(A(ePWUu7{b2UQIA8!b=bCb2OV&^h|IAjkyud9q+>htrqS|oOu
zBY&|M&KTy?AtGQ1e+KRNBp>OLcm}y0(TTK~P4ZLQD%nodjDD@u`t!RIHQ|qu%bRlT
zMpEpN1yyV2r7v?un@2CdnCjTn8SXXf;A<bobmg9Mrl%wM8JUWqJ|+VlID?J>ahT;=
zv_-=q%s?~@<mQ^#bzCM6F*HtYVh<xKFI~(c>JoP=fA|eYep1ag849KQ(~m0N8ui4m
zf;G(iWis9ilO!z`Mqk4xla9(x6iR>Q^aUWtI@uR$4rDA&CBEhHCwjF`t{zXqqTh)B
ztrG+nvXr7b!_<4d;kMxMMp5-)TCP4mH-CIuZ7}-l>+hSPRW%gMvMk79$lWx4Xl8*x
z=dV-g_}n~x13R+c$r{`+Vv8C)+E`*DVkBiSPFCmnxLH2Oo}4w6RPfY=C|Dm)FZ+c8
z5tzb-dKQo(q{rz>`+B44<vGe|9l5x2iYJ@h8xzU8cCrlXdvLHT*XQJwh<2mtOWfvT
zl}<Mh!o6gk_2s4U8d~k$CY2at0P^e4wPe0nO=M;S3}KL4a_5XZ;C9D~ip&i*L!vJq
zSsNUH?rpP1!CVgICzVo{Hr_}Qj~)Ek5>Tn)TSM>vP#~x_0k`YCTrWH@xmZ&gMvqHM
z$KP((9;#Zh*wI6W0(^=Q|3D3AIent!HFxTc$tQG!mAzU;AESn(0suk$cR`o1OdE|L
zw5Q7!?891vIFKbbH?RXHY3{1>ouzeM4E2Y^t@pj!qdXcDdsGzCzN$jOdnF3v3*E!A
z94Ccsu%o0tF{V{+ho5AmT=*~?OX&!v!^B!sUn%SnOv$Ky<b1j;rurfx90-^akWips
zh6a)>-FjA|hJugmDjwS{Xe6ipaKFa=vlMICfQL;m4Cg~=r|P(JhiD`|TUH1buVqo@
zcpD9j{FxY)q;KY>b%g3>sFUw#Pp?G@W8=9F*xk?4?(sO_4p@q_q<&R_N9msKy^#DP
zgW>YmjNcRQyie@!O7sjcn7H+6O7A`-sdfx)h|p7<Ka%&yh+<!GmqqiMzq}<&Kb}RI
z&xR~-zrao7(uw{A%P+jFwycn<*}60X<RAR&_v_Qr#7jL0r*h;XiP6n%SfPhaLRazK
zU-onI$C6{r6wvFfGEWLV`ZaeH1qmpSLj)aVCVJHw2Mn{qNV*er`9lm8cyyZ$>w=f(
z13j@FGH96fuX{nigA8zUpK~Kk5SK#0aVg<RFIy{;l@|N?@A?XaQ7<>CT0Q8+`Y<gA
z9N<l+{dha*>Tu^~F^3M3J|ugGu&Yb5OL2G{l+rn=kok(&*D%<NGF0<uk?2`-QOT-w
z^mF9iuyY&vC-h3B8LT9$VzoMAU4c@o$g~P{Z;ab=18tL_#$BC6gIzHD-@HtjDeYtk
z+S%50;YVJ}_ISKanGfBmAQvum6i;T2y*Ai~E3Q9qqlk=&IIQ?`kOlTPB}PLSE}j$V
zMTb*UMqWXfzoYMN6D{ouN#=wvs?@2T2*6;Nj7xxxWzquA3a%K0L>q$5k%qoWnSyG1
zDGYg<dWzKwoF(_u!wgPNrV3U25-d}|rS=+d(1yEkeu81|nj!|0LKH6lDYxzGa;H1J
zx#-4xXh;akP%@3OEC9(}n7)AnxGFye)(bsx?RB_{pp3p>-XDOx2#9aEl_#V?i@MUF
z1^kR$soFQ-IEq7UJbFt>=krv0ZZZbty4*}fRc~2=x$r*zpna!SPrXytsUv^;*oI-X
zfxd=k#fKE^ptxB{vi^zJJ#5g<Z|iJPU)VO-Gm_3U1?ZvMuEd6aUn$Kx7CXrvrj-Z)
zERPT6-jWHp>Ox#FqJj*IbRI~rFHAF_o(gcg2&v0eDB}JZH7c|zY-}1=q>p9^6veup
zyK)AUrSh)>#o`BRt_chrs=V$i;t@+pQfy~;edcW<gJ(bAklmb40Q;WN#1C;~L>QWm
zwdf-VX_a*>P0+QSsJ6o54gzyIWQwG~`qGGku3_lAi&Wmob)`Jqy{BiS1(FE^kpgm*
zAN_7CBDl{{-DW_2b+^?>-F~r<yuNUOMQiIU&Mvg%<C=`;tNhBfq#ZHOXCN?yd2j=Q
zzq8TwcyKmVHI%W=J?29f)|LZ%3LJy&Xu+UwzA%ZYqbME<g=-#|^Q0Z{vcvI@awRKo
zv|4Qb^4CEb&3^bq&10w7JrBNvp*>lo9+mFU@sI<zzQCM_&ikcSD=*CLzN-lvQmZ2}
zySlSNiglgrFD}G_=INB6ep@;*Q~2aMO5ua|A^GJ_STNc#iV^FyFX4!)6io_F3xr8(
zH<pI#axWgV1_mftI!j(=U8>H$-Lj`+FkD+0j5f3zW%m{${ZtX5eu=ZfzEh%~6HzBF
zt7TK>U=qQ%hKQtYE-duUT1CUgxvMkNj_4xj82Z;0D8VS7`^EfH`-JjypO1*@&3G}!
z>Wnzp>xZ!ps{DvRWkFb>mdt@>h?76By#7DE^uuGI{enm}&lP=Iwr$=Uf`p<vUu9HQ
zyU?CxMwjW^Q09rF4Wl`g&@;27bTM$Bew6}ahzkLxRlIqvd=!oqBXNhCpRi}CV(y&r
zjS=)<W99K+qKke=a5RVV-IY5EED`{pDpgVGn(t7~8E%~5i5zxQ&J`s*?`Ft74GpTa
zYEEk((NW2t81R99YBa~fYJJxX8`K5P_{#wT=}8!j{MqxyYqCyAEFW!eg8wi`L2ov&
zwgxi5&o5C{8vgB%*Gw8M<l!e(15SWP`D5?DmXwn<b@i1hM*PHQ(7h+lgCI*(;u6Wv
z#G~@BTD|*y+vfvI>ON*YlU_7pnxaP=(Riz~@@aZwKt=!dysGRi=U<UYy3#<{e8<H1
z4KXv-WhD|{WwsUr_^6&ITfV1m>~&>I$FUlh>zM@?ly1JckW2HM42?+`aj#7?Lj>}w
z%I(RTMoy0;uKZd(fKy>mbxO|hR~go6eFrI=KTcm*`e(M`%(Kg;6Xqm%clUDq<cWHN
z{VDLz-WmhBt{OB#^a-xbaSIHBXHp=zLP+&Y7jNN+0%yb}U=rBqEPWdZHft94u^x>~
zf9iv<{&WW=RUZVx#pbwMq)ZKzd_&c}$WB;h|I)4gZA=Tim$<DEGG@HXxaQM3H{<V{
z;JV*Cb-&#w6m{(2?)cx&j&mKQoLvh<%BE?Eu0?)abS1k>#)9KC4@`g}>R#O#5LnIf
zS%Lo>n$uU&+`;OE)g1VNzY&I?S?*4S2*cM+J0{dlGp^!_%N4Kvzn?;M!^2Da^uLO3
zeNQlRin*YkPE7xkK_DFGXs|@4Kc=ASqqUU1lrVngc$Le~$^5cDLftA)ckO9#i}II5
z5C+5eM8s567q#xi-^|31L|)^b1ELLIo9NHdC<(d|z>B@rStqnzr31s(w!z`_N($9U
z!NFy8Gwm%&RSnpMV`A1mS}_~C<Nf6NQj+K?I0l+EKdUm^BvD3q1<~1uakuPHg4a1A
z7HOB>_QgWa#NM13&YmMXqAh6uh!(x8GW9{j0rLSNa(PNxVe~_N9&}Q2p_ZH--eDyX
zHfO1?slf%Jr17PUL)T2Kk~rYjQyH;~o`u(%!3E2>RtWQ~Bq@BdQR*KViGrU=PkT>}
zxd+y@=0~MjBbRY^m){^3z>J|yugI}f`z~sL$5e7%U_&?5sXkezd9J9(gg~afb~ls}
z<-rc}wF$CQQ3I;dq|0LhkJjb%hiv}j&OOiriE$4Z@k-LqgZn*s=G!P+*=!zZ#Li=?
z+Q}NTUEL*v%`~SJq>wydS_Y)=dNB8xNVxYTzTRx1wblBd{!L31(bEg*5QW+rh7f<s
z$b=Rgi@s2{T+#1g+Y=^#70CZ+brI)Qe^lQxCt*C`eme0x*RFSAjAU0e^r!7GC>3eh
zXeZZSXjEAz6MF?^r|=>&?DD^=7>kg@vl#qHdljR!IrB}l{80g;%oMiWho)e*wPfMP
zV5NbW?XF2-!#pwf@yzz?O_<T>S|F`ZiolpibnIftfo<aTr~Qi3uMrUv?F8j<s~qWE
zk1uNqJ|hg#-y&+JM1uC?0IZ<1a*L`akELwBT^tM@&3t6uLv#?q@q!58!Je!fS-eGa
zu`aQMT0LBl)yAMNgR%4<Ss70@#nWE)_IHQ~c8Awt%1hoVP)sIy#@{yG8I@>Mf2+%%
zROEm%FYm!FCCYu&a)!Gc3e>%iSFZKNIbs58{G@agBw!PoFWJOUEQ7+XY`#L<;;g?Q
zAyAV+>WTW6Ib<})McKypA6<sLO~)t{YSvCG7_RZ@<5DyKL061|a9<d}!G^~Y*f8Ij
z60Jt(g>Wz~o-4iIJt31mfExJKux1UdUl4AU_X#qXuc(<Or=y$_IM<CKKzl9ohu&in
zXfp(cw;kMOkbbFs#I!|BI?9}r!W3t(pbcfM)N=!hA>wyLmjB^1Yr>sXWg(G$5Km99
zaL8*K|4Q)PFmk9?wdk#riCJ37brx*K(`S~fDPQte&ZW6?edSs*5)f_mRnI-PfD$wK
zCH77Uef8d5H6T6wt>m+!@-(qT$kLiXFclaABPlQ$p4dE-^-qR>%zOO$^lk23dSvwW
zZ9y3SM`dH%j!VmNQCfZ@caF=PUd@FhYBrI1T)Gos9_!I|L2p1LJ-!Epwwlee+hZ1c
zz%n?CR{>sU=+^C>(q=WoN?`89K9(ZK_pg!JKNd7u{|0>xa>^nJ6p1AF?XU^?me}%S
z#`-Z%-C?livexul7yzZ#w`$v6#_%#Z9Y2HkumFj8O<XxyRIZ=hUdJ-89bdTW9>+_e
zQo^#4UrpTG*n;QvjD6xo4;yjYrf+KO$w7b*g2#6NVa=j(V@U{VnTu<rF~UFqw+T;0
zoL=0W3uU}<KW~=MYHBz5;l>Tp?tS`KuO8>D+z(hLBL^l)a6WB}BJ#Iu$oY;GD)&VO
z-hIagOj!ie)-B2^Igsa<tvy9jWD7CX>#=RX(R~RF2801BIoc=)x6Pn>=*su!PX341
z>&HQ73;L*4?HW7(CBG?tQZFU%(OgWsf1~I>m6-=C6y4>a;G_oCR%hT|*SA*iUS+uU
zGeVsgU$C<;&7ngJyhcUQ5!NOQ(ofTEY2G_N><XX1xPKmUPVl&;5^tt)%DV2Ius(>=
zPTTTpA8Rg<c{u+LrPZPTMo!I2`estyQa8WOesMsoWqerp%HO!}Y)qjDYRMt_&@uAK
z{r&%%Sx%srn<MeqEA?~1RAvIqJdt&$M?;jiNQfu|+gC~;FNwa)4vR^S73x>hk+MRD
z%|!YCw0wx_H5bUk`4NEals2Gd%*<4dJ|xw|G#<oCM1<SB9Dk;>YLrXOON)#o+lUYl
zypJYSvsO-2L>V@Y3;&_?Ns#Vs4Www)M>;kJIw@5-nYO=uJyy+MiqrhQJN@ai=zP$C
zQb#i?M6T%4C46fj{H%-?=Ete)k&d2~lAk9{7PWn#@^W!3hU;12pzwziN<0eovU|7~
ze}I}q>s$)ajvqEU=fn`f*Y7x#k4+gcw!n+k{8X0Au74YZ;P6lTIu>0v<SCBl7VeS5
zt>Tj**5?SeXxA7${4peXL_Ne8l|TdFy1B@ur7Ra<8?w)Y1IUUa+@l%TuxORgF7bc{
zWGD}5pI&hBEQH*-yI3J7;reg@SK9wmszJcPZ6>F(PA7zri46<xH>cDk$p~`m91`}8
z3L)$9xfoRYu1d;1Hkp>|UKoyuMkw7OD>kCIm;-s?RVVpf9!13RWxhSj_G4!zx?1fI
zb#*^jV*oWqP8E_bx@?+sk6)P&6^39YZBA}YUOTv^C3D>1U7&yup|E=5@@>$ME@uz8
zQ#t<%LSgAufF8F`U9HA@-13;3qKzgoJS=4}+2YXTD8iBKNtxhS+)*zup=NeLiSjO+
z0e`Yyw$h0_X;!E}51qpwBR?0lFr&BX+^>pMrL*{~LEF*O@0eXPn~8)Jy3Y7<j98~O
zsN@tf;s+Ps^<#rp*bji>>>)36^$hOoe$FQe9>Hi>hi&xcW!s<q6kpgA%R*gx2_)DO
zoUYY+OM;Sg_RFu0u7(BBH_rRFbC6k@4SsN936lZLZRcC-%$~xPcvdJm594Hj27(rr
z<tB0;Eoi<KEJG9la#s5)FCd20c=b_}jmyCMbuxX|v4Lzo*(8TfNf?Ah!!=E|YSOM-
z<)?rffuxl}Xs;Ee@*-bsNW2(x7p69hXf}UA`$l<aFbZctq1trRdyU4R$bDY8VNOrt
zINjUj<__hY+@v*5mvt>x_=qo>__PQ9UJ@=|@%wa;Ut<wHRh$eO1=7rTqJYrH6R5dt
zfh2p)7}>~dTCnV;+`;2-&Dx3ahBg<u-crqpQ>7Vz77JH^kZD1ml9tD24WCpswb*#U
z2aEn`y6)1{KkE`ujL&3fe*`<2n$J>HwEM}!EK6Kza|hUB)@g@qY<88b81O1EgjKJ+
z7T46`qFjN!5cQM8ju98H(`R^G^`}xmj*~$^YWmr$6ht2C#dB(b4!<Sio3?ufs86X~
zl`#GKEJj~xDusYBd{L>(A&nV5^p0rhpKsuk^ieeKm_UxI#sBGarIhDZsTUsHlBo}(
zhlnO5+jrU!QuA(($xQ_0#(hWNs-8&i&fgvznXV9es&*DCNAt7XittI07TA+1OA?Cj
zWk#<1sc-96S?Wt(f7@>H#?e7bqKKgF3?0aYGNv#!HL{h3qa;i{+XatVIvopkZS9Kb
zQ_*(=+XMLWuFww&7_f=_7FwinyG33QiRmNS$alpY9o)k&8Pt{$a9O`?;@ArhT>WQM
ze6-(mfn_}=m6a^w%MFDqcrgBQqNPMN{kB8A>~tX14;gSU13@}(O<lsPzUN(MHdjMF
zJ>*ltE_xDe8Y(C4hxa3+?*$pk335G)Re1t$WfHBvQS$6dm>^V4zhMuuUb}rs8yqup
zyB;{~wHu3qt?9|tN`sdx={@p>_~*@t*6%}b_ANq6Is|`f03zZ4%kTE1Tw-Y;!z}AP
z+aU8&+#*`cJ;iz{N`)`y%HYfG*^(W)z$bN$cf5IE#LqEU1>>>V`U`d))NRqn)^Zj=
z64tTza2Hs4nJtUdM*xp|Wqw|eWU3Co>D6q<2!qzo(Zt}vw8`+?aNKi@JF2q~jF8^_
z!1i*Usk-i!e@Dt@jJ!^D#WElNzL6tJhQQEAjBjiPP|qx?*ZP*%g-$)E$9wHTgm=(;
zJXOvzczglAugu-(IAsvVdE<UjHhui4CP_K-Qh4OR;axf5`eTT2D3!0^9o(Wv#?C%j
zm%$L|A6M?SjkxGX%*k5{AN%07;o*D$mhCAUEl0Ju!{Cs%`N>igp5VDW3?dajo;YRQ
z9V-2@J{vp)vLg(LR%U*F)>NinNAI`;`*y5#qN8Z3eRIPalg4SJVO3=}FcN^Hvkg~~
z3W_0;=K`gkK%RqP!GP9zUO@4HvH-N<Cc1#EqEM6=_Mp}}&C+5W^a7LLg)qA?beaur
zO4D*~&PlPzP$P+_&qJ{fRwmdmo+GkY=J`=Vc{oW|1V__j^6B$gSWFRTj6GKJ>EpJA
zG$^xz*g5V@R+uU#p$W{Ff65E35kX`T*Wm&H6g{w-i9&bFf=JDBMRMA9_*~T*EOlH1
zT-0pW)X$hPQ#d{_@vb^B+3?ef5nl++F6_faMiwH5Acs6T?>)T}WFIx|lr%<~dZP$9
z%wfS;wj;#_l~9Yw*1ZI9udP{N>|SOV><#<u`hY?*=oK}d#lXEujXx2`lB(Cg$Luw2
z4LLP`;z0SnXxl}0;AB{JWfdn!gXfjL<l7zNAcEqaM3jefRJ(03rnZV5>7xxe=#aNS
zb3MQwxpN>h)L~QJ!OtKLUMd8LpYHdAnfuPL)L3#EL|G7jD{9p?iQINJD{y}kCu3oa
zkQT*yVL5#wd0#J$gtr2O(R$OEyOcloWjrmo>Wdxa=(KB1Yk*lU1VRHy_i$d3DG`PI
zTds!cDRG|+dRcg2GiVAv#vPzZ_8!E7Wg<j_iq&}b<)45jtJidH8FK%bPkJhGr!t?j
zV0P3gqV=0u4oUoB2%P`(u>$swxdCyJg-ywi{joK9HNvU@vzvO|J@-oo>@fy>pDS@+
z$V|P5z@{7D5#rqTKpDU9v=;H}Xj-*(4Oa`~x!Yd=1=#GNcSFiGD{2N(6z|JZl?~)-
z+#mkye$$aPWN!iyBC7O`PPyd#{B&mIo#=GH#0<}y<R;;nkX!hQVg26{Bb)?8w2uyu
z1UVvMM1@aNO$kWuH!Z@Hj75Ksvh>X0Dm0y%^!-Zq5x<oNoEU4qD)V}n*thCE3p<4+
z_jwsM;HiGlFS9}_vzs&~1w0_dq4(vtJS!KQ?}Hk^Lw%j`L$kXToaJ|np;SHr&mYUC
zRe@HdG$7Jcwn-c<fPFt9q#a+nVgKz?I*iESz<KU^AjJ9p(eOR55}T%y^L>DviH=&+
zN*JC_%J(DQgMmyo?+7qaCBhkg`%0{jPr^mfasRs|P?*ZQU4H*vcK|49HE$xEw6dYO
zE}Aww!G1KuUdP(BtlI_u;M^2sjh@nyKTz?JkLs-q@%Z;RN3uQZ3&>AfodS0(l3vL>
zehPo9!gAy+f0)kT4N+w?aRWB%iaQ@uL_90xcC&@UFwXg{UIqf%PjIZGz#{*<+p3k4
zcB;gMJ?WQO@ut$ktqnW+AtIjr-+mfx!jn3!nVT69(r6jm;jYbEZP-`c2=I6Ykntr0
z+RvUV6X6HTHoZ-Oj6?EyIxKB!GkdjZdooJ60|_pQb(@Q!OUtLk(8cXTVmLEZ$#vl}
z*)0m-`IRH45e|KXd==^O83Y#sXdQP(Ie|xmO9M4nL+H8iItK|x^8D@reUHRpd0jpg
zk%1q6BR9LwIvitqXw`i8RXvdAOtRx24EWaX;|yE@_uck{l~O-8W6?}!)kKlL)rbk4
zR8>WnG>BdAJIChJNJT4axohW1I*T$`ozTl{6T}N^R33L&sp?&}mU;TR9ic_vjK)$l
z4T0cVEUTdfvvibB7;{o^OPnCehDj}%zF8i;Zq+~Wn&udPqDyQDL;A2o)*m`k>hDFU
zU13DwDw%8{xM_#Ww_5VJuqB&>M@$-s(br%tR1<RbljFk<yK@gMip$_Gsizp`By)G_
zC5ydLu?8ze%l>a7^s~Mtaz0`-Nm`7z_1}1HI~^+P)(dSkO*I|R9y|{VKFcCVO;ajG
z0Z6$C!=vhdarqbc!5P%R)cT-#%b3;sn+mF{2$|Xl57Utah?!A|SF?_qaJW7uVapOn
zeo)CDK6u5LUGycUWMCNW1=C-c$N-dw_u$W@B3T54X(R#SKdD_X6|~nXP?ocUL{`uF
z-u`k<MRnE{yJwLRw!p2u+506Tq%45Rgi?yo-!LeF^?qc3{c5wLy_&~`X~zd9w%%nx
z9)UZ_#L_E*^2CTlg`8YTdLmrU|C$Uu7PMQi(*kNQ9&+{`Gm<3Z^)F_rO+>VS0**T<
z1K3@GEr0ma&86lXCywksrcZiF7lhi;2ba%h0%&X3Hv-iEoK%Dpdit8_&;Q;SY!#eB
zNNxZre{o)d9ln2uB+miFhrMQ^Cq<Gz^e>OraF@_Wh36+$F1SPC5Pp)i605=!n=AeL
zg4yV2?w~mD9sUv*WZa){EzXnI-^;kWkV21?$-WW3D4wDU*O2pj43P}A_1XnTxx+?8
z4L5*29-hhCcgS(1qz_}Ivhk(fbW7+ju{srSv>DBQVfU)I<EL!0K<31<3`!z$C-7>A
zPN+5MS+FOG@#JH2uVBGvi@K;e(E{G1?Xy>S;qMb8*gW8R3PJTFfgnkcvN4<hoQj16
zdNxlYzlTu)8Gdb<2!sK7l#%bJZfM)jsfx5qfAv|!M}cjlGMKR?Kofv~`g1Vkm83xl
zlbZV3TYGTCV{3Tt5omA6-39MPhXp@KP4St9S<QAyfm&b#{B^iTtU)GBEZ}`3aKJD|
zJH{b4yU+Gyye`K;Wbt*~d;77vkqfv&Hpxek&c!G`-~6nT0QnB#>o!HfDmz065f#0U
z-9Nin{Xr=o<9OvR5yAQ_FV<vfE|H%c`;PF)1XL`J6GZbaOH19)i13+#YrUE!!F7*~
zxu`ThO9z(Z<%7)AUaKCqnSo_{>z)rbWvwp+dFijz(&EnKyx|snNvvV~tks)?89lhx
z8t`*k%}4xVB#ly#gg6me^09FvOylmWu+umInL(3`^q$v+1wH>?0H*?2{aKV9RE){Y
zyg15n7q!3w!58$yIOLw_Mr7im&JJXQo2iMLI@lt_pI5~IcGlJ01^ZY@79e+A{)A~{
zvTJqN5fA{BR_DL8W3zIVr%8@7M`mf8$i5BGgquh#humS@qHcT)UCskOYqA>wEqND+
z;kSNe9(fLKhLUz{1h12?2H)q*4_sn-{Lt02*afFWT|>UJYM(2D>rY4Aj%gzzGXf5C
z#o0zn+f%!k96YA*b%K)W8e4u0e|_Z>8Y~V=3X4vW*c|O{E-NYcQW)4xSYjY3MvBbb
zCF2MHJ06XmNP9#29Xgu)8$)p;gt`b}umNf`C`v{@8mS}B{PdPsC&|;ftkCSfN{Woc
zj1|K`#^!$R%=1{bMyg-7+&y;8pAE>8Au;$PH(_E54p7KY$1xBJp=--H*Kck3oT~(`
zKUzUfbpSP41%O&o4%<~<-?!nfAgj&V@%Yw$HV5|vlA($dhm(ZSvtNQY@E}bPB?5&M
z?seHeq#$B7qoi+j?5GLRmg*eQxHFXS*v@>!_eE~PbiLh=-F{^)uHCH47Ys{Xch^;i
zSYISkibB(v-Me(m(WIK2y;rmAn{j?J@O;Z~1Fq|$YkI;<1r?zvpQhltBuG4yX)05a
z?bX35-kCTV^MRC`nn%I`Q$|l=zbxzd)FVQ?rMG`jU(WZ5mB9h>$_nZlRM2!WS3rvN
zv(8!&L&h|4?li^8=J(8MeaWB#WjUb-#R6aBugtfrYQNisS?W+mYPyv1o)qf5R$b-p
zW`$F&r+-opl8}9;SQAMYs?L09eH=IIZd)A1EoNLIhzX-3hUThK&W%5i*(@mvpR#s=
zIsF^Z9I$Q9cdu_@%B(H~7@5ybt0d385Oa$AWKa08W(XcDGgKw+=$+aml{{7+=o){U
z3T19&b98cLVQmU!Ze(v_Y6>zjGcq7BAa7!73OqatFHB`_XLM*WATl*MF$ynCWo~D5
zXfhx%F*7hQ3NK7$ZfA68GaxVuFHB`_XLM*FF)}tYHy|J&ARr1aMrmwxWpW@dMr>hp
zWkh9TZ)9Z(K0XR_baG{3Z3=kWjkX0;l-u{mO-d-;QbS3{&?Ozx-5moA%`kK~(hbr`
zN;lFXDJ9)0AuS*w{o=j%ch&oU*LvT@TEjg1v-dvVv-dvd85#;@HD*zB2UC!ggFTcP
z$il`C5Lb}Y2C@Oz*f?0&*w|5NXw<<_ThRaPs5F`&h%?y1p8q!waR|r+`e>6dfj&Aa
zIM@SZU2FkB4gipwAIQVc#s*+#W8?duqXUE=AYtMPHU}uM0AwBPLC&Z&;tq};5U`~c
z^l_d4eFe~&(F1^dd^}8lg#$$GKoGE*i9JBU1ZoAcdtA}X#1^3DU<L+3J^n~RCujwQ
zI`Xr!y1BWrnAkb9I6y3g=$Qa+V5k*9732(pxPr_9Ker4}G_eEyvo#h}8i2YL*!gdu
z=3oJJGl75rj|N+?8OYxGF~r5*90UP8t`1O>kq0O_g6#ixmjBy<3Gh#G06-Stzv=#o
z{+S4D|5vbynVEx~qlvu-*xnLg0k#DJl%(WYpzcs6fQh~NPe&75XNSjl6IT<kt%>QQ
z!(Y-(08*kV0F%cZ{<A-4GYHrb>dfK{w*9$B)}Lt}%PeVcF79Aw2eOAcqyE$<0fvCg
z9_#MG`cHFh>>b?fz5e%P0k$`{__+;p7e`hNd$5xWNJioxpGOnwFPkL@3gBYnW9Q-K
z27sIZAa^q>)}ITgdpLsrg212V$36IXIXXB3EFQN3@&Q|b9{-_wIh(kG08oev$j9q<
z@aGdM5C|{_n?V7lAWN`4>aXmNW{}0-&yW3wfZYLlY>(py1hD=5`}do{<Aj+z*xP#i
za{sHvtO`nMucTfv{<GzOp_rJ1JHU&XofE*!&d1FL0CIA10(f}X0X}~wQZ@nqBjmSW
z8G8!{0N>ww9}D$A&946p0o^|*gdXr`E=7mO(FFnMeoeYQ8yB0|;|K8n_q_iV^8YsG
zKePOQPWpcpNx9hC{>AD3cKAQs#13rh@sG#j=(<23kAZ^2<6*G>ud5d5@3X4_G6%cZ
z{g=u>O&*VfsJ*4_zl8)lOM%@%=E`8GnbqGD^S52&=UKA_+k=!HoWVbD7XUMmjqN||
z@qn4xJl-JAk7M~40zICeKbMrWH*+xmdC=IoxB(^*h=~X4W1AnJxBy<j#}jD|a{p_P
z0jw<c4$#LCz~lCO02U4q)So@&W(TmELQKp+k8Y3PFB_2ke?}`P*wOs)oY;aapub^`
z|KPv7{EKt`#$6o$6U_Ao{4<o3?H~9vMht9e`7dGT{*9RaPPlk~C%<xW|DMb4@jCuD
z;P+He$A9e{|H!rhLI3aaf1t<pe`Wg*`Xzwt&wRfXn2YTX^1BAEf5~5!`9lQ9pX?76
z|7E|)_)p5;bpZcFzX|y{0v5Io4v_y-@1NcOi~LW-FU0rQ3M&(+8`vE5%f|Pw4f4zQ
zSb)b12V@B`u{8%fJKCCf{L6URf3Zigzm3jg-y9!bS@!>Q=T}|W{{+7X+i&3aashe&
z1iuJ3=kMgVY5INf-%7~&d#2y&2juz3{=YW<Qp@ol@|&#R$sc<8{zHBf%=HKQP4OQJ
zewX}5vfnj7{-j}rn3(^(h<{fM1akdu{~fUZ3;vJ`<ow<KJK+8We(UVde*9zpLl4it
z=Es)~^p7hF@c15bhd$ni&fveU+n?l*o6*b#^0-InUvIBRxBor=^?T&;wgZ`=&d)fQ
z2?Sf$2DkmF7A1CL-Wq*5NwcMuM$hau4{3L~Mn_7fugVNug<OaxfBJ~Iv>{1%F0w#=
z=e6Hlj}+FDsM31leQOl2I=<D4IyZ*j_qkwSw62#Fg@jpMWZV1B$y+nP2EG}#Q<f&z
z$%PkP`7O>5w;pNty57>o;n0~amF+5Sxu>_K6U@;XQThSJQ#5&|S<{4+PoT`C$PC!W
z?if?&7n9gIpC8C&;~7zX4x%{<y>!>uW3Hw>m&?`Joj(v!6X_C?!k=RgkH7E|+ewin
znDr_tT*;rJh_P^dkQUavP!E9GxMb3D;VV5;1^`b-^jIpzK8YnCh&iw&ji3!{Iv^iQ
z3k2k!y5;N>pO=d(n9CuNz=yuj*HQPPaxF)*`vhovBFV+mBJI1H_lz-+L$9|mB%LRU
zqc;nm|AmWUo`KjViYy50X~F2d>&-_x>@VSPJhCt{=u%3oh&H<P;&4Sa0g0YvZy0zA
zwOw4@(tFUNiswUrSiYEvEZ4X_!ex}G&=M`|U0a>1BReGB?4MT7au>*_F@R)Yi4?5C
zB0a?9VC6|8C0DNbA8z?tT+4C>@AJ-GdATM@DV)9i;rHW96J>^P`B%h*J-2k%@-O{U
zVfT2zhW-1VeHj>-XLhpMHqTLP5MpqrcYC>jcoNys;VXoFdP7<cj<y?ybcf#r@Fv^>
z*9Eoil(d&HQA+R~{dLj-!A}<P-SYZp*F_TH%~Idd%HWM3CS7-8zx8MesOhC&F?Ny3
z#}Z8*N3wJ{pg}sfPuYs4qPZocmshT;WXBJGZ(3}0KxmL^ugNrZeThR7)NZXeh4-p<
zs!lURA;H>|w{S$sZ_XEbt}C{cPWNL6yk2OL$=nvupr?!?T|nRE|KlZs5$d<-HBZ&?
zj&0gOa%)i`bbYBOGiO6Go|mse-c@^0r@p0zAU>-no@ZgLakn1f<;cK9I$u0m7Ygze
zdcNI`!FQ=Rp6SZGaO#@BMt`fWo-e^b+A1UOlhf!#QvIc5(bRob=<<Tp1>L7TuJXk$
z|LFcpTZ3t`gE!+2{7p^uVS5FViIElH#o|M~SW&~2&vzGS;r)84tQcF8sv`meC#t*A
z0!iqUecW`t@9N};g~cgGgaj<D=sTvLNt-l(L*cVmc}KqdK`>B+VEZ^Uiw=ieZ~-<v
zI)m7oF_hkCTDfm1443`IyBc${6el~g+f}L!8FZ`K#+~jhsSA+I4Rg3!9Rg|g{Ig(k
z`^~Sz-d=@f<lz<`uF;%ql{X*v4P2GNgin?ub+JFc9#o`~IjW?E)#qDs`*hPCqFf2@
z82kMmvBh|=9b9pMK)1<0q=-la>uFK=_OvQWv~4h6PdMK1YVg4))o+qVagn~48G24%
zU|1$8Lua^@prD7b3YchrzprjlqSvwCNj&7ayz7;_^b~?AKEqzwsq@|Ji+FEAKVo5T
zIavkVmBrQ_!IkTfeTDjwrMj5qyrD%)uo7;8R{7Hq%nlB@Wy`O1V3*^axkQi<@x_&5
zb4c85@(=9tBJxY865yN~%h1eQ#@IHtf!BV5>zBm_jl-_ll{0oox|c_Vxu2e8RFU7j
zY}eAX(!qG4dhb<1!S1)X)V4#ruL$8L{Ztcc8l7uXU=&2S@NRqqdAIr=KZoGMyXPb8
z`?McK!ij)=c<USuUg*aPa?5f4Mi)0m^F7ig%j%t?@TRbX5hHB6C1PHi1$k)ZGh6LX
z*kvzLDmR6006eOb$gfA`%6X#J5z=O3<B-Sa2)fiU1qP~j4k{S+IOD{=Gr*7}cOpA+
z`U>#m)=2=k;6t>8;nXf$U(9kOmyEBR=*e##UX@)O8`643-73>-4%hqUi7{mW*>nv^
z)o-Gw#o2K|0xv}-c~<PNtGH3UiE_@T$(EQTu>x`%t^*kdTWPAk8=;G#jkfU6(LW8n
z3311h<9?HL>-4c(<?a~AcA1}`5~b*qvD4v{$2V}!cO#|z4g{WE+%F1p6rEhvbk2K(
z7J`G4Y{;#z+*Wk=g)Hlfdq!S8Da2O+k}F2?QEUnDJ{4J5eZ%knx%|Z*b1a=i5F1<5
zamlfTjW{V1u3kV+Wy+E}QPeW+OZU(&Ex%W%@1BBvQJv9?Be3!A%H6`QqNDTclnUDY
zHTNpRL%<nRvZ3;8gp=1a3CToW(;mxda8fWtx0L*-m2*J7rTgx0Sr~d~_A|~QpzL)6
zW{EA9l+(0KGw2JW6!Bi_?2RlL1Btn`zOqu<x3R(Tqu(whq{Gyn^r%oz_&dT9mre(F
zB5?~oeEm`Fi*!sWvD;}moejiY!*`fMU}fa}UKy}1w4{^O9tihbJCm{U^VnW>#l=KO
z$Lo$RehY#!Z7hYZr33W5KsxMVU+5EG_o$woGqV!LQsY-<@tdx=l$nfU@D$sY8!_;t
zwwQzhG&Vb<&1h7CQR%J+lqiAkX%-dUeMAsV;GPLp^zsWCplL}qh$fV{sWleZ7LwU-
ztJ}`++p=t=$jlR01Mkdrr#V=R^;;4|*b=(%J8<NBq>ySHFTb5V);GyP#P1S7JfKZn
zdu>rBvYzqAb?!@f#J%oamGx6av08(mLfm5(kN4nnEspLR6fUpur~zt97#soO-C3Pm
zgR&UoTnaGo#Ot=UMmM<H9>e=POnI=T8k1`k_fqc{mVOgEYVQpi+<Bq#i7$ur>VD##
zUsI-RDD&ev$Z<CuYDE}SHD;M&X1K#Q1&OIRKXM`)N4ZP%9rwUG@Jr4>7s<R7g1M>q
z!%{;l(+%9du4JEQh>MnHC(~LR<h{JPk>d<UJR?1l6zzUlP&3Luq@Xlq(l^1Vsr^ZB
zqsWwLnbgi?o}oE{Fkyt7FI6|9+F-DtwJkizyULhs@Ut*3ek9cOL~7@)+uSIGoxX33
zl>9Z>4C^NI2E3AM4HtYbR+or}-@C{_K7c_B)z|~<?D)XdcUX7dP%oHo6evUao}UDD
zVa^?5SfHl;ViqmWk5wAie6t5XzOAO2C|3>l)_*1jmSa$)IriNrMKAL*ns{M4pKSpV
zB-w<(1yU26Sg6J&_V--ZE}%v{$;<BK1E4;oB9D3Bd<Ge_)!O&RZ)?$FBSO3r@UOnf
zt`mM0tw_-1wbp;Q4R1x9(<hU<L}d%E|F+{%mq;*=O#J}CSrx2@5@keywP8)Va}Ph`
zK4H%hysJw|0QE1X2l|h^j+NOA8Ap4!Hj5V(wPo;xbb+wCiwR!p`z3$%tGu~{8CxAR
zsgzf0en)s<l7aZ=5g&;7lcjNWHqP#XTTZ3xU_B3EUVCZj!l(iDHhUUTv?)ulg4GuW
z@fxoA%{$-MP=J~hvIA!is|KrF<q_1M77<}5cj{wp`o|G@u)^#vec#h`nsLzb)HVC!
z%jH9{^F|%NrMXVA0R>qyp)_XtsixzbxcyPP?`O=LDC<MH$zjpNn*@?!^;1z5d~wtG
zH7aZyF=W_2OssDylWMe<#%xX7o)zR^Tc{H8W)1QZX+{a-sVnFUXc%>E$*21D9z3-+
zaF`FVBz@Zhd2;l5f}z0sM`xQgO1fpzvjBk{Fe9CZ#Tw3$m97NLGnixZfaa7Iwqwe}
z?bc^5`B~;n@y^N%B^LCSM<QW?aMDV0p6AJ^{+rt$no-o=8XhN|Pb4srt4>?Q9W>rc
ze$>yXe8>aCR#UpZ(?}f>;NSYDr*fh?xy3g}VgqB|5qjnJA$ic?E7%QZhUg0q2~fHH
zu7=7yP0@UzVd#T!IZt4IwU~G6?Ow{asZW7!ZK5tN143Tvy%GjIEf4#jj?LE(yb`Y}
zkSB;Xw4ivWK?;sIp}F9N06LtEol|orP}{6yCllMr#P-BCpV+o-+qP}nwmq?JJNahs
zs<&$Is`ubKynn%3)!kQjmz1bC83>D*Ht2TA%ysKDnvF?|Xu|BTV;jCMf~VAwLgJo`
z?2M|j_rGD?s`1|o{GpJUM9d}-$VCpn^5y{Q8E_$wIjK5o12yT?xAER-O};a86<V-m
zH-zE-lUk7s+DA}y8)%mHVtqjRyq$=^6&jT4B;Qc|VhHUEz0x>H4-Ae%at}kZf;Tu#
z5PAXF*kt6o(=ek?*9NPFq?V*4AN!ow2vQIUdY4w{d-X{huTPVtDmCf$8Za}LEyz}M
z5m(JxXV(|51L1q|$-#E&;3fjtLX2Ug4~~I%Q}Fg4TatXDeba$B8J9q|;Bhl+xkvz*
zm+O3NOq}K2-bHXk;ybFu{_~rdoc3bl1;{wCyZPHA10=6mhF>ZN&zZyvG3d|G(FU7i
zDL+Mu#2;&OLSSyJZGo}~e@h$h7s|YSxF9NiH&AmKB^a>oNRvdT5f^(6KWP0cplp<u
z$?c0oby>33R4Vq+ROF<@+eR|4x>(K-nz2aHVTr@$WRk8H2-lOb2B|3zTfHp>QE*U0
zZ?<caLl;y)AfP*Z__%LVR8Q@7A#;A&r|7N#3$MWlB6`e?2I${w5?JT2JaE%~_HFqI
z6g0l+PsuMQqXK%VqtxW`UJXSZ1x>uRQU1yw+N2fJbZ~Kg)tzhhe359<MdHqbXXHI2
z+E(E-kyTqrm*iND+uv41I`UYixZskOMwn=Ipz0+K|5a)cZ0ZRJH@@oLo@zX-TX~mM
zcne2)zTluypR6a(Lh%Xhw0zT1O;XU^Lyeb<OEK=HovOdg@e&K-)5IZ3b(!nD?X#~A
zH;o?$SQ<uMoD2>%;xCt_vYW!$iZ|IwSm01y-Zd+ZSJo3=xTNqZV6x&B?kFzn>)s$$
zJnFIgej)F9&rv8;LWtf&3*w*J`CdOaVTG_O1qs9nwCOu$?sK!n9QBcm@~=^a@Y|bI
zS>c8;Umos!S>qH08oDcKRwF9buKxtHBI9}1!ahxje=KgEnplqV4h5??Ezg4o4RSng
zszFMOQKIHKSI};aSbx)gHV9!b16MzQ^}^u=rL+g5Jp19MIr%tbmv4U_T$K*mis};)
zRKt^`w~m@WUBX`S+$K3ayd|^Ulm@XHBuIWm9B^`QgBKB!T@<H54fEPpiJD<eSsEh$
zX+mg5C6&yiYKjkJy)VQdHlN*|u9r7!i_LyF@KdVRwdlg7O|o0BiKo_(af`niX!Q(a
z*iT}lA{y(3F)NE0(nw{uHr+Gha=(RT+vpN2QnzEL$n>W5lUJW#2K~8A%k!Q*tYef}
zCSifJR&e$A<?PsC<1INDVA!4TSwTU`HV8v=s6<i*W7l>XlGw2<UP{XIE7IF$4p{5T
z-gV`Bup$_9*O{&Ge&`rqE6RwAzK0TMvg+;1|G@=wZ`@AP!vUo)2}6p99p9VA)~#82
zJBPj7jckgzMB8R=%$SwUZ$lts7aCp`c*RC5Kn-5&akEzO^X@r32%7H?PD{)v*m27i
zT-|4s@h9+9HF`DnA;^*KX}hD0i!|I`XPtVOqB00D@<_F9r7UMM;7U$2g%Equ(mL6P
zzEp9__I;=WtZGp5p@=9BWJB15aj*^sB<DEwiD9eFoZ=PFH6l=rfW!`=+DtGKDJz@M
z_*BTP-?L<MuDRJB@Je*C(O2p%Xu3OS>ZoIpJ*jxpw-OR=I*4!3*j1G@X7~Dy`PcH%
zsP9}*wC%s<Pe+U--(;aj#5VmiR6;b>bjubpKZ#E@&im_KE9~qe!d0mqEcu%640J2r
z8kJ=EHTEG*l0hV@TNt6oHD>MUPOPZVpU9E<rIlNDkHYS;zj4lu2}TH^25o{>5_DZg
zzzcpPi$ly!OM3i`!Yi2*TwCk9qJ^vlq5#JIOFV75RPW9~p7&6AH<og{?`S#h@-M8l
zILI(kD-VW>z7tax*nf<K7*xgCbyM43ih;{7S7kUrX1K-5oNa|&yW9Zl{AvLHLnsTm
z;Ri}{hq;*xS`|qkEJ2`UO1g;(s9@+E5cjgPawGT+2@zF@O|XOzyJTLyEu~Vu25-Jg
z09jd|4Z~9wc<0XF?9(KB1agM>Es0cueviTuGmMI2{=Yi3s#J5Mn#-!B8E;c>yTOaq
zJ%Nf6Qc)$W1u%XPTI2dCaX>=WzTaP2ap<id#z~}TLhKbOCC81=TIsCD|3009KaJ5`
zZu#Y`e7B9}vVw|SNwI3+F1t9r)sOu#_UOh38ni9(_8QMO`BoduQVfMP>^3bl5P9Mo
z?1uh6p+%~bMct^$oaN;?Sj<U6t^$Bcwh9tk>{7ATG^&MY>lJh2MI$X~8Jnx88)b|E
zEvV^m6~uYb*L<k~nPo^$5evpucK0fc^_Ex3&id122yLC;EB0js7Bqmw!Dt;W4|40P
z?j_}k)6$O-H4oFS9~kiFVAyGP9>*-2sL@Rrcc2=_k|};iBsD%BG_SbvZ*)uP02?K#
z+DtSIs^|)E@d8VJuG^{U_z!=1R+R3=VC9K?DI^IhCv*~B-1Z7q%K$oM8hNbK=M-z>
z#i~Y|P$iaPPIJmzJtK*KvT&aAg0NoUdpe<!>@*_q7yJO%QZW_^*+B<K+bWNmdA=9;
zJ7hG0lzj_xj$Ho{n^=$26OJ)&@m6x&Kog6%xfF|Cp>RX=KHDiJA~l7pT9I`a=W*os
zkx11{CH0CAasyMuN7x51M)@pcF%E%#ui?`BS$yo{0$+0YHx2qE9QpeI%ZSv}8|Wn6
z{z8O`+KX7Nsya3PWlx&*qS!E6ZfA;`Ne%7+CpLi|J#Qn~vzN3XIXDg@=&;;SCqhRt
zDT(%nw<ylL^yhwIOhy}@xk2o;cHZvI(!;Bnz+}sHWTE@z4Bc@zD=D$cHW;eRYk+t5
z<V*}Ulg>j1rWNJ~+uepdfoX8F$51n);NT<Z-M_7Z%ThCWu!iC8w~|)fv&qO#pX9ah
zdT1UP1N$3}FgpgqB2m?*S)|<hhmz#dFDkhkR0-Hd@SShxqbWFQ&$B<Nk{8^>aKQS3
z&8p?0q!+x&V{JDM2q~a0r_s2)xZ+8eZ7Cnwb#F3i-bS(<O8timE~k`HEo-0;@ph?L
z@$hW#9ZWBoY<xO~_o1^{glL!<m;NuR?wKQwhSgu)83QyHRFLGO(TDmR48J!dmH{Sb
z0V(;O4U4)|EYL)J2BYL~Rdg6Ds`-^O6S%1eBg3wkn{U*(^>tI4#aCv0=VNQ@ClG@w
zZX6qjH)uV@HAui%L6{(<ML2Dry80G!U8jyd1ahZreRlUGtn_;ab0t7`+a=7}A}q-b
zf4w04CIp}*QHXFh)}t3>PMHuK{d)!md_5gfkb5vugPw0en(B|FkF|cQ&$SNgGi6QP
z9_zT%53+{m&y4`3Ux0}2hj?;nv~7B48ctirmzY;c-g`=bsy*xpT?abv7)dBQ+9x>p
zAf^{|>bwsx+8W+m&VJS<nPT`ADQSos)&ye}bWvw-x*z9+-!vZnlFM5VE76*0Rkx9*
zQNOBJN~xBc5G=SJEqXoS!|f+<+`jAF4Kvq?NQ<MeF2=%45PoN(l~(#lP{G!2@k5tN
zpi4^z_{QlsIq}K618!ug!GTefq6_pH#d1@-?kvo0rl|YWZ-v)~M{G=@4i+x2>Pq4B
zaV4Dd7YWdxHpt7F6&;zUtHS2__eu|{;TeVC_&qZYx`=mcVY{qmjA}^tH=WLEtm<U2
z<f5-2i(9Y5<hab^vzB`dQWh3dn&4zw#d7IU%3-4>QG78pIa~gm5U2_RHq7`q)QxrQ
z@2n(W;4QcmryAF;+r~8?*ML`fzPN%g!L%axfj>$l26L2c8*xDfy(4b7>K~DetcZdF
zEFHX^^hGZ{V${pJ-W#>A=?{4({-T!rW#qtJ$5o44@>342O|((lydYqSgQ5QI1@af)
z=5wSX6|BevPp50uq*=Se*gq&Fl$)DR(g{;^7i$<Mo}iZ1m$n8h<+nf@91a{!=&XHb
zYzw*<5>O0s^{-M?2cuhRM~h~J^WD;SK-0MD7Zo4&G}%L2;Lyk>Leqi$mfX-&4Z!sv
zOCQ3h@46<Q|61?-&tM;To{Vi_w<2Lv;G#UouBuQbpE<0++PlvJ!3wrI*m&({z?35n
zyV6ae&{zZL!Wqxu#w1A*a-|VAx6A{>XVht(fR3XX8=&D$Ta|$HjaJGy)Q1z`Oe85n
zVs9Z>IDw?}YJ?e8c@N~Mt|R#*VYJ#ZEovpuG0geaZmv9etn67$-EzVzHlp^000gM5
zJrDFFlqOX#(gMTP9e`ll4VEf$0*4_z&1G9G&1B%r$9SL*Ycne7?QXDE=gV;)V0!pk
zK*0o*LSJl^3cr`v0?MvMxS<^=iec{Bv%%R(xIDpI*TyqUBRZIlj=xr8DME)r^Bcyh
zF*6pl3*--h`rOF(G%2$@uHp5GG|Xc#7$)D`-NC|uDpe^oEh;PmytK~ms+f;;Rp}d`
z4Vm3xz(J-V&f0sxdZ_+1Qb0|=rsReI{JG~MjH)%U<_Hg}6!}H)FJ1{@x9z}0Lb}{q
zVlSxb)Us#;loX_8;iw$ZaD{3mFBj%+#JN(JDZ!NFQ(oy;)LOB8i#|;<X%>yPhQean
ze4mx8F;6(ON_*M~mK9D@RMqiMA)g!hm6GHYUU>4Dfwib(Vnm8}NEqACofa*MAmGte
zV|We;2p`}nXjOcw2+M@Usa`mZF`mlE?uE$;pFxA{l4>};4ruS8{dL_u6Ar$D7266H
zL+H}d)F>~AeeOWl$l`+wpt|*Qe{ftPeB8u2WIfBCMhZH99OlyRf6p(elXx1+Z3gRV
zJ;xz+vY|Y&QRx(EpzFIVtg$6)*3w>L%JW`=V;WX4QxL8->ZeXo>J=DRg3Gj0|E>rM
z5ENTTco$6c#}2_wIb3v2u20gB*VYqz`s;wC^3xv*x<Jd`_+gp<w}Rj3SCrzYRi$-q
zII_5fUbjy`x9vzr&}tlI8o76l-LBLqRP=okoqy;Z15+wyR%fQGS@^Z6+3+#$U7dZ(
zgXgv~OlePjslu<p5llwHZpQY*?K{wdCWYrs)y)SXow#W;gjBp4kpnZKW&*~V3L%9y
z(M0Z(wG}u|WRQ5ZVF_hiVQwYBdKd(1Zg>v!VVHhsm)Rh}-dS*bAZeVRaF1X4SA0l@
zVB(2EFD?oVEWt+yOwhIg>o1+tehiqUO<p?--ZX3MK={XN{qoGq2I4Uy<2$kMb&jNn
zRiFJblh2ml4^zZ5t?=plqQGXquin5?LWIV;!i=mL#{3JUz&5fxFqm@}=Zrb!+wtBm
z>FVcmw@(@-`?J^r^tU9{<K$X9>5mQANw&{;x1;8ejVl&hXKuBFpX)5B%@KY_LQ@Xs
z*ge>WOI$itG3Tbtfr7ett29RpoQ)CrM12<LBnUEkVoDQba9-(8S>kB=xkWwxh5P*z
zwLQMtV6iT?m9OJ)k5V<oO7=xYZjGF(Yv?UA0RKyZMWxrhGd})m1g8MF&Y+@PgE1z`
zVZL%~20Xomb9SGtb-vO~Z?iSKX<nC@2`I3dhr)eLjh9SrMA@6MXPuLt-`f6`8?gX!
z&<E#jfkm0Ts5Fo(NZA3FI@=iCuQI%)WhUC*V%S47PD1(Ma5_O!HHL-I`&st<ZD}la
z#kihdouiEj7H$PexDkfnS{B}Zi>a0@!xcg#XKO>U`sS?oK`*H^{?>yylj*Uq#c^b+
zApZrj;uufHI<fLo9cw689ML`5BHjtA2CUXm7gM)TL_Sy6llFiy2?vI(*Qlg)-UVBk
zU>7B~F@C=xBQ8JWA%R0H)>B}pmX&dy-~Ggc3cD66^&m8{RLv0C^&Ja?@>|2vxiJ3J
z!@0Q1l8!Mjc^vZBy@H((aD0#V@fPKNVq4eELua^<-2<)ei!85~F4D19d>UhVus;rW
zM5<1oE}y+5{%e9K-l21)E)u~`1LB%+gj?;m`PGbi46fXnI(#_(zCH4{`mooW#NY3h
zR)4T3nhJ}XY@6A{dh4PB)27qtTNkA71Zsa2csV2iwAL~ZMGi$n_58i20jY2y{K<t8
zSU+o{O{nj+i<AIa;>9@V9Aqe!+$@?P=<6Pts@ce=wb2yRIk%XiCaTHE`sH-x2XL9E
z1dajb@=oEL8{MoUZw$S~AvX=JE$XePLjdFw4O4%ezXX}>#dh3@igFi;U)O4BSx~aG
zR@&e3cI$q%E-yjZ=2bGiwP-}dtXqv<hDLUAMdrqfx3o9f{J3h{bc{7Zh9=EMRC3*$
zuYqU4b%v9-T`COSb1SSmYD14-pf}e7v&+lwxDTZqH<8V;)b>z@T4OeyF)+~w?r|w5
zq^XN1vT$|?Ur~u~)uDzGX3Y~Bez%p8py03jwJ*f%lBU<Q<Pq6D!SX#*B9#jr^d^yQ
z>QIyQXaW~o(>Si~D5<~e+m#PdBEW*Z3r_tRMIwvJ`q3VdTF>)m(CqC=vagpD3hIdr
z>{V?M?>tu9RYsVW_7v5cEINF`gm8DLpj?u+H~Xh3Aroq#cIr9%S>tJu>(ui#XBf!l
z(&Y<kO*I{VUY3)S;FBobS~IyTD=(lDdCG2%k~3;n$aPM{^zV~J_+}d=_LF1M4^!Bz
zGGxTGIygbW`^lsP8Fzf5F?3YuO-q8LUbIIwF9i}ylVz4VoWsgYU!+;KhBgL@9oygf
z5|P3)SU>0zJWqFj2Y^FTJg|BuXW}OkH<~S|=sFbKyCJCYx8fP);SWkVX=r$c+&Y~Y
z3qi%fEaq*$<`lq}yPC6D*muV|-4^6mja=m+-Efa3T($3<_Uo$_ifaR!Aum_zv}!PH
z{;19$myU3_mU#ILAcq(y;Um9q<;|L=wL1^V7d2R=+Zvrw+n~e~<q+^b$|C~~EK5fR
z<L41_d3IdWm|G3FT4eRWJc`Ri&e>LN$X!#tv{EXjN=?e`a0z?I58w_oUY>b*l=231
zDKM=4Rf+0<W3-IrLJPSeJ<5h57|Q2#$yd9+i4~$HW?qN%`H8x>MUzNR=b?#LT`Kbw
z!Fq6QA6%*lEt2>CuDEyCGl9$h<<HN1|HV{Y?2+J*>&pk33aLb$fO}KIOr)<6*R-(Y
zDMoUf4_4rm7R<pJKpDd?jTjx3AB<1sH~7@-|7X~J&}1vx_~S@njp*5=I>%ImbD-`(
zQ}V<;hrQ{SkRyxagBMW?{6Hxs-=i7X6L=D9=YkI+Z#)Aj@)UR9I1+16_DN>c;f&V5
zO;lOH{y~4$g@;2w?-Y>H^1oQzGq-MJ)hQ$RFRGnhIT{NwWmoSmsWA>7h5c4y(7{=R
ziU#rpdbJ>l9`fPG(a276Bpeg*z#ZQ><I$B{tGU@x-Kvf(bhD|-_tWGn#5}x+pw9d1
z73`}OtP@rPRacaq4bF*`ZA4wn*3`&Lys4Le<EF1JHo%IGvz#CaFDbv7kNNyaey{fU
zHQ@!ih=-zThHCG02EHLvJTN3UOpq&+n?N&?TNV}38!$h3A2>`kVy7R(K_qf|<hntn
z4q4H31ZT$RN!uB+g0%5)46<o*Yw^h26(=QK`4oxUnz>=fAvK0mF-o>M8o<EXrf<#H
z3jP_>X%W}Xcf8h$k2-R`i+2siMP#ory<FL)+>|1vCVU=id%Im5wk9riX55T9K$({H
zv3%G)SielJ)cYX9VH!6s%M88F;;eiPS}S);%c}MIZJt9x=}*qT&1Qx9dC%W&k~UG+
z-GJ?5L**>tnYZ}au#6Y>_nzd+hQTqERa|DdMSQ|-;M`kl6s+00L4hX1@tZwMu6YNa
zsmLVdYeqx%H_emLQ9TNwLN~YiF*6o#M1Mpcdzx-=Wn1y<y$Bk;$;XP}k|iDU?iQ+u
z&|pT7HrF#}vG2S1f#;1}VW{%*3W}dRRxN=udB@k9^1^seP<v`oO9Rud0HJjQ+Hp~H
zTR;GpWy)8AM`x&;cic;VPFDQ}SDFWRw3CiKxs>N?f&58(W>E+pmoxT(o40==*l`qc
zMD1k5XiB*h#F=Fk&G`oLnukbJt6YaJ|DOn0GhCLjuX2~(j(RZJz<=WEM>3yl^4ZkL
zl_-meDGEb+*CR0E)J#b?wtJ4`%j5n`*?8?XF_yp)u#w$!w3PnlOrc@Dm%Y$>F$&i9
z-9gvcjJbTA^~SgU>Tm5^o*1&3=M1I7gBU`*^X{Rz-lyu3lUZDcrYA6*0Fl8lA=tO?
z!PFW=vi?~JihADf5NfgeVVY77`>ObS``JOcaaXpuAS<QJjz@Q=X%)dKuOo&cbj4vT
z6<CQ>-Xj8%q0+vqUNq14GB7XisVjd}l0WYv_K&yEdj}^7C0{7l0mEVM3vka3Oz=ae
zZBbg=6vdG<gl4k8Z9!h11l|NIdzF`|y2(NqVO*eQ0!wTw`2ELj3qGS2M0_S^E{1aQ
zqqAZNMJ}R51l*`~V<mpFL_l7~U9VDBWW-gU%QmFgetj)ztJHTVqq8iyvXVu0Ga0rY
z*Fzm&%>sFtpE3uy{L4Nun$1L#rgB$IQ9%@cS#0;K6!LhaSzQaWc>mQeesEU_Wta4E
z3m8*nep3T{DU*23q&B;`55#!k0#EqhT4SCjyf*M5%e#!qo^ovSjKvMaYB+W`vQ$OJ
zEx*;5-364sZ4u_nrwyHWtV`}sw$ZE&&v&w{Ij;UD8Jzyb{F-RjJvp@Ae0VEor3!^m
zexsuyMxe|D*~Yq9THb5RU3G-GU7G4s2XH3U=z=Lm(8nB1{I2M}=C(}lDd!>#td49;
z2_E80H+P5Ag&x*XC9x!B|E;X+oqRacRyto?yOxM$Xy0JEaHr(#-2l;$*1mqeoXYVN
z2>o|gJL1FY65dl!yN=nqlcVa}Ye_Y;8X2r@2^h6UPf4uDKBWfrm^vFa=DT#{&vsm6
z7vW*x9Ukb&Bs;O6AG$W=HK2;?CgB#?6AX4#)<wKOZD5s5xzK6x%CIy-CUTDU2Ns=X
zoA~V;PSxKitnv!bGeM5)<n937^0>g+lt!|~?`YCO4sYRe9u1Kr{zZZl*n|A~x3E<5
zpMzYe!+6F9Hr2Q<={*R0BiKUCq#Lt(k>Yd2Y6^va>4`O$JV5}`e(Y<34QE&mP{nwK
zYb@O7EK`nSv!&#bpNzA+9w+PH{1~7vvh5;`^Dm<74j2iZ(ntEb!7GSU<^2m`#QX4}
zTq1m)_h3cHUQA(0*UNUN96ueu$IB2jk3e{d&TxH7QwD0PY)lK|A%m=nj?|i3sNr&=
z_QkRcW&B0z9}|6sDrgPAmGKhm#tc#m8?||=5{49!--Io{IZ$uuTT5ob((q(*4Z0Tu
zJDaoI*;C%`XNVYzKv%28TYgQa_0ivh1cPv)u#$yAbgU%E18!FTM`}Q!PO4`tZ;u4_
z<W8}Fx#(Y+Dg1yo#G}S{KfnB*|AYr-(5M7?hGF?`A_cfbgYWN7Y1<R?_sWG>2auyf
z_xRIKMPj*pOnOwxz_5p4)LvAqcNX|G<@IE_S`CD%^xvrYbDUzuE^Zk<r#;0&Xnl28
zi@fIC={rtoSIQTBC$T1)rHjs6^l_;j$BsjLZj*gPT#6104B-5rzJCNLSM3PIu(cg#
z1IeD68hv~-_LflyTu}S3t9pkma4?`z;?~j<XFJ!z$M+I_WE1R9DViVS%{wzmAErz#
z1UU1d*844vDz-j8Bc0LOJ-(*C?e3uq+{0<N&M=l*G1BhQAjD48`B+R&Aml}}2jk1$
zjQuTbuzn9GMbQqOxtppEeooWvf)<&Um4AyQH@b%)k@Kzd^XU*hR!Hj4f6qmtL9|7u
zQ<V{6de5hn__j5wE#kS-ZmKeZwM4Dc@Tu)Hhc0OB$3AS-55_K(Ceu;9n;`x&H@z*U
zisqUzjusj5QCUYu4!J8c6vnMgmVfGf*N}$13jW9m%FX6)apzoDIQ_UsNj<+H*MRTO
z&>E4~7Zt)MS4Rz=2uHj_S&%_J6r@+K>frXb4}Z3lp**q|>H^Y;XF|;yhCa}IbSDhK
zhe7PqFQL@L2s1&y+qJejH+06b(%<#xZP-~~3^Y%ugVBH^*AIW41LmtwW8ii`F6+u6
zZjY9_Osi<u9UwKm7F<3M5_)868@&N`?i+(-jIf>p`mT77o;3U7epj`;Vc+%wYDlpd
zIPOT^|J%$1qn~@i!roA5C5|=GRABl6d1nkRD~-BWWn0@Y$Qr*d$9oTq6*29uchg(l
zor&KI_LM=2>*ei$(lqU5m>#efzMuO_wo^8HHz?bjYof*sycnItA~BXBX45eA0?8}k
zb|gjoZVW|hGIfRFmO&f(Yvyjty8hXoKq_)g>D{V`*Bhsv-#*m0hDUAs;5mB{f;S2f
zsEDL_aPcSWBqx4o;j0fJe{%6YI8{(bQhfo%3s);F$|h)8m~#9buO&$-@x8zz0DsyB
ze!}(|RT);75=rf??!JF)*m>8-<R>T-@=@6zdH+H@zsTpGF!ymQn!x#sP-@e)-IdH{
zB$81DylhoB%%@cy$0BT~biw3dT0EaQb)Xw)9FfIhqj04ghPuHrNN0Jid^fNl@~R~u
zX~>j;=L9K9sdMX>dY-E}=j}bEG!1==(kSp*o=rYV+-meyKXNQ9X6T9Uz7Wg#?%^R~
z@MvfAYPqRG&yUZxC9#7rcab9V>FOTsg!@GZlTthvSt_!m|K<F_%aMLpT<(;3{47xF
z1L^y8`OVmrcS{h9!D$~xqeXcwD;Mlz5q)jkA+O#j&gRl1^Mczs1W8O?wr!f<ZFct<
z&~T3SvA@H!V1OV_BMe4xbur`4DrqEn9{bZM;wY(9e%-0nNaL#XiIsWlGN8(N!bC=E
zON}$`lZ~~#<Oy(^c-yV|OO8xGt>#JB$PJF%IO;Xn78z2yqgLZC2iBJW)>t*!zrMw)
zQ(%zrnc{asDeEIHS&CP?{f=nJ#bHN*-;n$lsTGEyS@g;DLwSd$<0PvYlm0z&D^+EV
zShF4R_rQfM(>qo6=k>db{!_Tit#{M>O{XTFeQN}5_gUgbhSt_JsHxpF_)ZVT0ASs|
zrBC9jP+r%PtkA6&`L=@iuiwd61gSktbZehrdhxU~?24c5A-AY^y$^puwV@S^KJM(?
zVu-qO%!t&#wJTI6Hbsmi;T=jP+8(Y^R@jxBQfF-zNwddLtUSI085$XQH&FN(wY57s
z7ZMf{jq<x(e)!=;vkBkV5Y14ux0x`AZI3u36*T_a>!Ara`d%Fp*CTH{Q?{bSx{kCr
zS#gSB4I(qO6r0nhNQ2ywA4sbWIi~-Lb+G&wtb>t-k?lVm2NNM111r;iuK$<nU}s?b
z^M9d^{|(jQ3aOO1vB0(vlm`&uRPYx<5<l&Jp~5Q;K_pBfC5=apAnca7Q7)(I4it~4
zA{0j|4n(D-lB5UGm7nc?+5FMH_)uAq+)j5ld7j#yZCu5f$#ji=MZ1c3VPGW%nf!$g
zBmtc4?9AFvNC-tr0|WvM%<(6BPfhjjjt@{NC0w*W2@>GPUtV;OP_0W2JYX2qIUXu+
z1yq$#4HF7I5hPN|pP!!yQT*PP5SbQS$=|K-%D)4IpOzFP1lW#>L}A;ktq=vhcV;J+
z--hBBA`K;_&xtCh{5*U}h!7D}9v#>PnB!nARB#(ebeuc~+{ITc@+On48jYl+{M_QA
zBG7&a5`O{Sk=r#8&z>529uXX<kZ0eeKL!ct6QNBVUm-LQhyRGI-PDaA(gN@Ww7{UB
zTPQUyjF8?f<uAl%1{TC#&N*>K$g<b~BA-BKAJA){A1eez8lZ30cD^-?;N39YaYPE#
zvycG7u01HRJkU)vh$YoUq&^S*8z3U64_Lw>w7{VgA#OrD)i6B#+e&9Bd6{`2sD8Ou
zQ(ib!crkrO5X`oV;iv!%<J*m@h&ol#j`n^;HMrdZJ~R$&h~b;g*4^AeX8}SUBwk-l
z1iPT_!C}njU-AKDP;DQeX(bZ`aNFRot+PlW;1L0&L&3j={TW;dr(n-X-Rk}7*<qfk
zp?d5NOm80qh!prCKQS5j+CbrcKyUl`x%K#I-hroPPxwH8nt=2e=I{}QL9X;-?7((+
zmhWuR25oVCzs=e(=0Ji<2{55LUOz9!VP+XTNe7lcc5ZdWXw01Cu4)n}cXH!@(3Ajz
z9>CtO!NGo?f(3+xn54wOsH1$ZUyU=`P@i%HKk=2}cKjgp5!#QDJ7v9|&#yut*L*O0
zz~7B&_{ehFm~kIF(yd_O0eP?Vw?FAOy?}3Dmfo`aUZ$^JE`aCm`8)N@7tZa^5Rr|Y
z>J1+hw=;p>9n+l1{;=PxZ5h^OtBbS#^DX2*Bw%?+@YX?Im}8%E`WHG9Q26KbAftsH
zecG2c2m1MACR^_p8AgC(Fo~XPJm%{#THq{*|2K7rD%KLXZ?xX+<|kf&-fhC*H|H2E
zP6);_--o;ifBZAXISMiwsNkGg7(0;ftD6++>A!#M68`a1<QBgGFc@;A9PqyBGbktF
zykBG>ItmC3r_QGFt92iw&j&xhz-lBJ9qx}B{5w$eNb5r+*{jNTi!U?*#6y4pa3YF$
znNyvkv)!bAF>xdL&qN&DxXOBdJK~SLKPy&?rt^{sV?PC#rQ}EZx)Fxj^m@CD-1rxR
z>GfdDtOLAfb?X=}U*~!TNvHPaKXSbIPB<4dMH3|{8cc2qko31GVtxzERlTrmef;PO
zuk&As)%&O7`ZMixTtu~itOjkNV0B$bJ)u}=me~CLD3PdTyk^!-#3^;Jc2mx#UO;L-
z1*zNvtF_5kBNhlc2TcjK`^%eoNxKS9!CwPER;9-=$e>)g_(M%em`*iqhP(Tjx%ICh
zbKuCRY_^L2V$a7gx7>t&^LESP{d38(%yp5AJ^2P{sM#&XCK5HPMTjgVe0FLtwmDxm
z%82aJ9CfOjALEsDsAX3U^~cb80UCP26%Xt7pN889f0v4qVLLg77rZU7wZ?@bdLQSk
zS_5ueLFmk%3h8_um7~Jh{k2<1D3A5Qy$qhlSv)2iG9#k+<0S(cRTdJgO<X-$S{koq
z&>R&T2bsES$D5-8&ktUX_{LJM>S{?X#cMQ$Fw6|IO~L6XThac+1KPUXObs&+<tC<?
zMQi(=)&a|T$<kGL^;6XlOM8^Bj$8kCdlmr9Sk8l0AuSU}h(O+-eU-IGfa)v0CZ&EB
zk3hK2cI3ep|9dOX7&?f)pO}Q+;~m+xQz8Rs86^MmzIe_~H=Q--`Z2}IM^k2iyv_a0
z+pqRxr7s{_BzBRx-ST|9FJTmV?OZwP9mV%!V{D1@VZ_(KWc{roV{dRtaz3AR^XpMn
zf_bb*@M`iQ(N+q&OSU7?yC`MGSl{ji05Og@8R6Td<H~VO9SQTIY|168OQZ_Uvty@Y
z(e%@eBuxqU!zfQI1l_7_=9a<Fgm{})ZuXl8a6@-YS8d3A><P3i)$V2k#=6aY)QYD%
zTM1EX>3-`=wiPE=&xg)^b4qQqvfjzZkwDDvRq>5JmueYxjaj0emTFnXVV7U-oZL6y
z)&jI`{4)|emDA8e-M{dFo6KdSMB$0#+QsqT=ksP>#n^Gls{G>;p>g=x8!Mq^oV-op
z_1hZfMl4W370b%@eh<Bs43TH{gJY9I{@cneF{KK|=E)g_`ZYSzaky0<(;_wY8iQj-
z94HSL6!w=Qp3ZtMVQwnx5e_e~^;NW_Y?MyHhq1B}{alRbU&N+CM;Ni$WBiHBYp(21
zi(V^}%GQnD48vvQBgP?DzM4+S4v<Na!gPlu&B2r@vix#qd-@tSyUZ^WkgE&Z6-e=o
zNbYHNx}q%9<c>zCO95A$3cN`!W}@j(oF1Pw?v4Nm*JW_G!ahB#?M6+6{cS_?7Lb^E
zZX4`{l6{!?&~8s(M0%W_y*VZNhb{GwlLJPCRMFM1KeCzH><H8qPUFA7Yx261ap(}U
z{X?ah#mq+rUHf@&K+Mxm5#5nucTUP=-u`0sVq4i0!hb!o)=rTNnmQ#?cXi{<+2+u;
zFX=YfdEHjq=*aTMxmdKXaNHRFBdPLDNB$O|5=POYtTcfmZj!~C=@=TH>E*rfzOB1W
z({s?nb#fYivy{E1xb?MgvAOl{{jJS5aV8BfWl{?GJ7zb<z4ieSEsb8HGWJJ(r^A@B
zEIp#}OR`GB#c?bkcGE3NPse<i*_i$C6UHp8ap?xy;yvWclz`j=sa-r{Y;q-sN@NF3
z52EvVwIhs;(C)*NWCa$++WFq7jM`4$Jms9f_?k>a<?lGO=FRmsT0y*~@%r48G7gZQ
z>h(!x?#LL@8i)@1`UpGd=r%ZTVyi+^5$+k85KUZx-?^Wa0%-on003?ce#)EaWl7Ad
zFZ4uSQD#Jvn9&BO>AHFl$|}s8pMUPsv3pvch?)Yop^d)5k*;(29X<PdVbHT<U_7H$
zNXWBcV%6d0pc_hVos(>9+G~%i%UL@fzrY7%7WQ|fjneTpa$wUEKsRtpbM`ip1lg19
zuoMsM-NaBsg71;({r)jeuU@V5hac=;M++rU*U8jVW|rwd)?3IFd9){EGS83>K1;3H
z7f8zT>mMdt<MsKi!<xbNGEB^M{bEax+G;Kg78m1Ad72Bx>&T?&_z2wMo0-+AEH3JM
zlXcFR5G?N63bGTrcN!i<HFLY=ARaX=<JM;tf%iJabbjQPnPQWEJaD&n=0l3>GI5_U
zA_y78H`eOnHIr8EL#C&)K;HDVO|F6Bgv|s<vAngwS4@olAcq>5`EF`kQK^{oZC}z~
zRr|TC?WP_-?OPZ~54e#6bAbY;9)FhYQ_ua*Gmp2fIPI&SLghRTWB}AJ%&C4I`Zc2?
z*fic@^UOC>bZHWDvu|?JVZ?XRDx^%pY2IV25qN6)$vs^YQI;@S%hV^5CXxVOTM(5p
zSTw{W(xhv=`K~6$st~Dt3JZ9KH;OVD@x%9ifvHJvqS)k39Qms=)hBTwp}LUN2e);)
zwcqvoC*4%GMXQ}2C-=7HI&aIOwF_Z1-4||%?1!0-(yRF9%ghaU@%QyFjGg>~_cW*W
z%z+5z%D^n;1h_2jrnI=2#?a2@r@^Cc9GX7d)TNS6zB)R!voZ@N<Gw#8EF_vUB`W0n
zfuZ2mbp)!T7K9zSKXYmQvZHeHHsVXc$FPI`RsJW6U(EI%^SOJi>BR|ro9-3{LbZ;n
zygfCoKOs2|Zx!*^-=424AC-y@<Ev$!A-@JxYpOmGkH2=kHmaC4jfD(oBMd#KT)9q6
z0yPf3e)NnEuomBwWDA<6h5Wh&B2vi2yd5P^x_Pf-;T~;5t}UA&-!#zSpuZ7Nl!Oyp
z0XkvJULE?NyfXw3FJOSKO|KCHr%+*hTxtSg(R$|mf1=^d)x?*X#0g-L;n^6UYd%E+
zce<6{^6)B(eiK#H{Y-8@>WuWDL^%@*cKQ+TtrS4USwKn6x}s1Cw@QdzWYs!GV-H?h
zTMi~R`Zqv1TSmgGp&yBBx1aE!!bmZDQ9;3$*;-et?{o%m?z&U%ad`w{`upPvQE|;-
zv*=c-^DZdGcd|mn*>Y>mQ-4D-wu>1R1kcDpWOeD%{kW&kblvmwZ}Vf*(?o*>k=FAE
z!Cikam_85|2_sG0Ota<6bFub_1*w1yv;Q&K7gQy@-1FCCwmlAJ@;NKX<+313*=cI)
zL?nztY}{6XQnwB#mF))VBJl-5j>`pFeGfA!=((`>w8QD01{z>b)}1A6tTHEKEF)xg
zh?@Z49X+~z{m&XQ%Hx?D5RaIO#86A2#a&cx*<;h*<aTp&UIPDSWm2-5(;w)gkpT*^
z`~^oCJi-34wpl>5<fCU_c>6Cc;Pa_Jg+jZK287g6)`W!m>}%|Wf8skMbVyx$(mcO9
z0Z){C%~JmcI<j(5QwN5(5>wZ$j%*YfVI0}OBYdlU%zAgYS7e1;LI)8=ui0u01eDMh
z@dNJ(UUj~C0s@x2$Q16B24s-f+h6`f{_RngdtY(r-~&Aq7nGp3=rQJd@*pW}@R)D*
z6Yh3xPmb+8_Ad5cPwfp?Xr`Vah<aIHr)%Yuf(9UN6SnE1;Mn<?0EWgrX-2Hx>`Qgb
zof++^bV_^;k8D^VBgWck^14m88c}ctAnDTsQ!&hV2xiS+{~&l71OD>d;x)R|{@M65
zbVJ#9&#EHiesHhygQVF($Zu>9{@Rec35jr_(UBmlMD<V<(C%#by7bcQCr8!9H1&k|
z$ibJ@ndWnz&!mpM)7nZCO_TV7Gb@kFUqhf2XE33GuzUh-fz4<G#{Iy43=<?FS^p41
zeJibsfLNM3Ijwl)R1P%$G|MLRN&ltw=`4!M>r3nH?aZ&;GWj07I0}!S8Zkx2(H^v=
z_L&bd@;;KE`rgj(MS*oh7Wge-Hu6<wJA)==;WWZbdsDN$=Vt}Zh$X9vZ-|><@c_%K
z(E-k{-8V+nvm9!qA6WS?1dLaKqwk`r;F>hK<;vMEC^v1+&icUYIkFfwUrSSZq08st
z_<c5n7}E44u4W~=vBN*^i-k{LJLce4%k7libiO8>h&9ej<>WOQ{AfNWpKE~6h4xaE
ze9d$KX(8!A)$x_ianaWh|MEVrT+=vs(`BsgB2{HyP%-Zx#Szme20B;km<NneD6Vu#
zJ$w@+J8T9gGFEA)iCo<pZc3*>D-(ELA$y=@dA}Mc==atHdo<NDz9t2epuoNumjc5X
zj(DI%EkrYnA*rP~tSjAx2T8V&|5-sIO-&R&mOA2+YqBAnN{zkl(}OH&k?eUVvN5UM
z)@ZjE4d0U1OykINoqZFw203FV|DAVPVbLWh;+W8v688ksQ6f*3nyD;QgWAI8>}uEY
zFT20LYj_h%-tS{$qJHtockYCN7HWs78I`T8CHr`2PF|wmKSPGJNGzU~(O87OjV@iD
zoA>@MQgP8*_J>@`S+i5>1CcE}APNHEy8VjR5=DO<<|S(VuDiMz&p1Joeuihg0Q)6t
zwhY?faM2;w6arO!6kA0ET#^A$t<pWUfGv1LR9kVRy<t(A@5C}`3g!y?%f)tJ?7o!!
zwSIS$06ZPc^We%zFcHO6R$CTK)$3SdRV*PcUC6zuzuwj}ETsokvdj@6mKUP2ZS~A@
z`W=TY#VeZCbL@u=Zp9S2?UfU^2qziy$x@8yz+AlTs~+ri*%I`a94gmhIqS+WRMozM
zliUz9v6DNp3NbT5P;Kqgr=1d0UoUAT@Nw`1Y(c6b3}xU8aJsGv|CcFOb@kdza(N6<
zkGL0?P)RVnxsTnJfrydbhNyq-KHw#1Ioh5kbZ}8Ce_Zct_ETuI@kjRj0exbQvsm9p
z{3Fnx1>)A;!!QD?gJu4{`6c<OXE+>L#8d&jWI77cEtRva@_>?}&WuUq+G=Zh{|AHs
zw)rCVrP%xVIVw{aUm>{Vq$h@U#j}G|wwZtZyrF&0xOaCs|E!2=14jv0$1GVd7y_HT
z=|kK4yX5Wk(6VHX#_;W`GFX4;%oJw}{V=)Q?*)1evUFR9EX1D<R@nZPJ)z?p>KMZo
zbVD%ETXX6r<|Mu6{8(pnI@+r3&8^F0vpn=@8t*e6xR_X8q88goGOsRvq$6tf&eX&<
ztj+-)V>PiX8P~d)S?5pIm&UYHxVQpeY9}}9!$^KaqUSLfYY|hmy;N8vX*2<vM1J>?
zE|z|46SbG(v;TQazXyN{>!l<j$|>5>Sk-s%?z8B4A|5d$=E?l)Gu)usu*_p$lc#W7
z+$?tPb@up|8Mgo2+V8uUiP-Ao<a_kO6{<5M(|LQ67Rn7L^_pwrO+l4L>oX6AW2>uX
zQhE3bjJJISRF0!v$Y5vb^=H<Oi=`IVujrXI<7U&@Bjs3?=5%J}c8gaR;`{j&T){(J
zbb<o#ZZi3SYOu`Q>vV2P=U>F-2fzP4-=Mj86o91P8RCzRE+hwT3Tf0dVyeopFWTA>
zSEz^(UR;3-A$Rj(SwT=lH+yq1@4Mq3d@ZS#>lXygGSPuHy>WUPQ8N;v<&w=F&!#6)
zbCrVT6v&QTQe4KYFDoPw%^udhy%J0(jF0OXQfjdz=l8@Ict~juRP8b=RR#TR@m$d7
zBta{0t}OGGXgcaJ)|)UiNe<*CH)oFbD>>CXiis)ht~nocho!!yH_wqF?)UZ<>)M6_
zP8;S(p2H~ysav8_wmw(m4VILp)HWDNz;~lwvP$ihRpbT#l98R=RJ9%-p+p9cgl{V&
z1^;R*H@CpR3%d+;X<>36wBe-;OaPp{ToQ%&WcO4|+`yoenkwMwrQkaWx;}CgUo5RP
z-!-WnB=JFcM1yIDGK#Lg^IQ>xLIFC$@~O+M4f$6tHC5G^ioc+VB)Ffc{*8G`@!@=)
zVV6g3DLAOuEdK5T&<Sx6WHwzcOB+0$h5oWfR@xUeLsd0n6$_f&zSj)?-Oy3cL=k4g
z|9X*x<J9_5soz6x-syst`*(q|M$dVGRJ4M>{A;gg&=9Bk-HQ9+`bMQ}DWU$5^1VIi
ztwCowczTOeK~a?teXYu6GKU<HeP65+buLvCMD-5O#9d#c_SjO=MRH+`vUN8QbZBh#
z@rlCd-N0a4T>cyS4e@PvmYRx`-VkbEod*BLqmopnMcaj$x}iN=4+H#<+}+b*{z{W}
z9?eCN5MfUo{uH$jaJ7};673$Z2zZUMZLq5D8DBIHGU;!4A}cuonCnpku{4S<iUil^
zSBNwB;fq$55NiZ@BNwbbe9SMu62b#1)<wD-5yh)%#!*V_*vf!Q?J4<v>g(;0LZ)u3
zNFCIM&QEpxy&=Bk(6djf8&>Z@fyGf*>X0}0-5|bAY4YQCDQ6}?$`o#4jy9*zuW=X^
zg%*4N<)|;EYM&-3zI1xFeWSev>&KY!-cqeNVH?+`B{kGRv6@qNZ7u@4TbMBbdag)L
zEmh0ZxkU&<A`#jjuBEG1{RSokxtAa6>B4{~;q5-0-7ds?x>7yHGkr*r40_++N}&-P
zZmlb6X`C-OGrh4H84MR$>L+QGN?u)c(%Cov<gprn)3TS*Mu-&&7X7MTUi~X9^iQb1
zee0qqhR3>AqOMe(ZmntUwa@Ztk@JH5QvB$B(#D_TO`G6UofQn6X)j3545)%Lh;}wB
z<&~b{u{1VM<fP{Su|$NNx`&3*3PuR)()&~=*;QNxz9Ax!z;_r^ou$92$O`4W(%fe!
z=S7~6@lB;|ySMqWiGJK=|GIm=7Lwwt@6iIp_OBTZqCnmausQ`{P+>H;7}zz!S=km_
z;f=Ar=!o-fhWC<ppVHq)4}=~>N4ge(#h|U7IWukZGBW*9#<aEK)#qONf!mbm%hh3b
z*z7A@WTZA`p$!dYXJ;9%^+$@UWn+it6md2<!?ge=FT|Z13kUvEF|)AePM1doBm+>*
z^rkw>97sO4Z`0y(lKT?8U%D|v7a|WvWcQBTlLsL#StR6&W2Yg0O|0$W^_5cE-QV2i
zhwspS7_%PJJR&<V+t_N4&Jy%{A&d@COjG9FX3C{zTH;0*TNh*cz-QM)wp{t5QLBGC
zTU6=paE|RK(e4rO5@S=m&QOryT`kl%ES{H95xwX1E$!HU^!AZruVPKhAt1V+E;jk-
z3BpJtm`GREFd(PF`3e@ZYxzSO6T&MV)+c&dl@@8+JLa(Lo_*%*&+Z^9w)8D>5Cxn>
z=cv{}OPc6u%_4p(wwfgj^FeQIoXBgHEox6rm0lwmyL7F-O4g1+^eC>?o;<u`O%_xB
z?q#FM93aHhr8N@WOpVvshNrRs<Znxs9Tm$IW6%PrS06u7nrGC@OIDTM!7p%IKZwU|
zD#Vi!g%wbO?CueDN&*>ZjL$uO=R7}cH_S`$q?o!}%ME(cps@k3BR3$w?0+xN1Qby8
zQqK=e67ydYO=JXO)g`WZY8e(ABEqci^Hi;VXMFnC3J4&O_3;IJ@uGLSluW73h)ZPM
z7m@ch=UP;n4lU%)$K7y-ijgyk6f<~FFgoR2PI29uk=Viy;QJgkM&N^Vk94xO>BdY_
zyK=&1a5q66Qk@^eSn}!O{hZOs<1hP;t+Rd9FkFhR%I}%fS|*gI-sadpKTv;Uak_Aj
z0k}vu)vDnDRM*|})Iq+t@NK0x+Q3^c<O-|F>4e7VTDE)61frFL+;kUjuT~DScbdZV
zN4KSSM{M&@glK9dQdvp*6$%9HAT_$$+hYlK7xw)8ht3#<A<AeWRd_{l4VGblNuo&8
zmxJ_5)n2+@%YhB0?2w#1+|Qztz*CTpUZDjf1V3h4oS516AwaPR+q@b?&NROoZ?9*w
zdGlSakFx5&SjB{Cf%vqPCY)QkU^Gb6$BA>d>>V=KK~G|7-Z?C!hbD2WgYIjFgQc*z
zB$1k!?6MrTT029_+qAG8eX5K9P9m4X>H7@NH;{6c8_G&cf+WSK48mG8#yb1DrAzac
zG(%5EHty%TBVJ9<(()$7Ta=#;`@7rVeitG16&dpG;kg4sF4gkn#m!%h@RQIzFHjD}
zBH@1L4tL<bR!v_iV8yw45JEoLOJPiOzt(&eyN0HGjOK!^aH;Tudo_y2$M3$cd4OW+
zi51oSK<Vey2Di1vYBip^)s!744_Jqv*4#}>u+iN*d<?t-#k-t7K?hrU46?699c)oZ
zjK6(Du2W>{{8x{U?Z0|_>>U4b_5OK$EDX&5x&A+okNFS7|Jvho0#{nvTHzH>rGQ5U
zO#m=Exw?iP0)a*$Ff+ixgCx)sfkZ4+GeAKFuq@>JIi)r&<T-g6cD-!BeE(Q_?_@Ol
zIL<U4T%KH-5+JC68v2Nk2Z0L11B=}Sg82b}jI~J+5rM#9V1fJvhj4Mp<Wb@7A<2#z
z0ttEYq1u*wIVebo^iUxbfeRUYI-ofCCt)Fgph5W!0EG6GgouP-V3OZgVf>Q7N`2e%
z<bE|k{II~GZpe-lB%5Dx@))Y9$7#K(KsI9dK;i&E){QC$|2%v^5NE(1kT$~v+J4BN
zc_cVcOcLCD#H1ZQfZb$Uf&4WfAb^*b7etR;4j61fBqsFNE~rR5NEBfXViEfg^lc&o
z|F27kw`pm=G*G>Ln5WNZ9Fi;rUOiGUKVUt1F{n)xls$0Q07BrjLEw=LP9SersL$5M
z8Cyu+zT8e=KQR6u$rti>Q(}rwRffOwNPEW!As+&JbU>yQa3FL4+&%ucf>$8DwnaHH
zhA9;Pa7fR<e1<yEkv-KTJ->_sBA_Uu*U+3&66Eukd(_*Y0UrSBXF7!AWQ85MaZn}>
zSj;`?w|XuNDzyEOR%h<+%?xi4iPtW$FWZcKTjGNpvaHT^K#5&hD2A}&Q7kM3@i+T1
zhybv#KYw0Z5d+XZu-_|Z2h6v$?fwbEn=#l|5F=aw$R^S?kSQFrIN}mkz#9p|Etsbe
z5JeFNDa9@y{Ldy*AU_a!841iTkiGvv_)j(-s8E*g4db+!pf|7zf7CNS(C+L`uL>Kc
z`yZlRhF;&!pYz(h`kwmmc%mEIk)H-iN+K^{;7?+Hpx!tNM4)#{Ci^Mk(_C}7zBxYQ
zPq8YpO)TKJPi@-cjPIJiAF)7fKX8!0c5~Yw5iyqqfi{28+gasNPs85)zGngdr>ITw
zb(ZEQbL^*+kd+?uGBy1p_3MW*a4V0Ymk$AR-Xos@G%vC*3iz{Edzz>>R0U%n?8)K7
zwvr4c3JXl64faxZ73xVT@LezqE{{TNb;sf@I`cb=2@@58k>5?wmzNsIN80jxG6aiu
zG&HSr5QSu1mki3p;oH54Y7=Si(Qzn9co;~J6!}st%ACoB90m$V-(((o|3hjBNC21v
zBp4c4?-U$33rT!G7cUtM7z{5?Z!{Ef>f$8o-TX5e_6PQ7tG^fCAPp!i6r)7Jm*k(y
z*kOo|^vZ`E=J%cV1BcOn=p5?q{tfcoKQ#Y8jGaStAkdPAW7|&0wr$%^I<{@wwr$(C
zZQFJ-eP;2_EZ$<)cUN~+_2YvMzZd;pzzR9p{nVoa2RVSYC(gkVW0wr&D}%A+cCU0B
z4fS>AcL_3F{vv_*?L%o8zq}&tcg%T+8igJs8b0l|ZZ%0&Y4;_UZqx>|T2Vn1UM9n4
z%>vh5Bf)~LW<`*LTX6EGT~qd29Ty%Nvn{AYb2~>gL;FA|h5HOd6N?E=9g+#;o<Pop
zbKZ5$K1=TzyH^=(c3`khfPSL)?K01DKD$(i(GHzr;AHhe;X~f%iQv(Dn8z0*LRrtP
z)o4O;4-afP2!;l}p@ZV#z<td1G$cG)7o3&#&5eOP1sB;xF!fsyjX=8==2zIaS}n58
zmL`y5#oUe5_|XEw7Cb+eFI0Dv@eFp|Rc10HIpz41bt#K}Ea@O(51WjeAd}nC)lZv(
zq(bn8FwiuARmHsOX0&qgwFi3GbadBg*+-sUd}$(XZQ19ZJgR*=O8av-hW<>}&4Qn*
z6*#3g={mvG@#c3^-k~`>H6vck)ZHgk_+i_WCd6EBb1&n`m)h0`EH(zhy`#oy2tb1h
z;QvZ;w^9Fz;VSG1^qLEjta^LSLKm}><M%GV7Nipzey+pi<yPOS#(zs$Ws!?Go}F2<
z@u7C(1(cfaL{j9z3`5h_sxC>Cr+775kk~}b*}gliOI}dE)5EE-f^GP{cH3WW)YBKE
zQEon{H18#}l3cC+^AmkBhgVz|1~*Pr8eLqNfJCBoQ47(y*sYjrR=h)|Cp|Ri_-nxm
zHEwgiB&VVl)qQ{A=3bROwtv$Sp1@YAB?Hvn9)7j%;5%6fjnCUu0tK$NpJY?|qLyCP
zN0BL!JmB)tN&v;|`&|Z!SYfrmw?9HS-Ny17L}X|Uc3PvOWE;KygBrMiYnYhW)%g>t
zevLf%Jz13-lClo`w|J#igwq0Wp}@YJy-rwmCHy$9X%JN7rY()?G;PCO34gf4d?ey&
zo_1_NRSv&qK_`Gz|4;H(UkX@WUK*95ZfRWVg$FmsWq*Qi$pEI5V6|0xsq-=SGJJX%
zlEx*Pl+_C;>bu#Zt#F_vy1{H?sQ0oX?EW^1>w>|VTHZ0v;cSCCSxA_3P@Nf7Rl|af
zuO2PTiGH+WOr4wbS(`mq3WBAENmOJ-DI<aLZrB39I!h``mR`jH>+1DzrND5e=Fu^b
zn-XsFgeI`MCOI>jsiz<Nk5WC<L1(^6;B_EPrr{x$p=$!XA!sDuj&i2my}lB%cgkL)
zEzUHax({vyeG;ZRzMFs`qtWkKw=^jeggTCRng`9|-LiFS#fcN6Vb7|wEiai@{*xty
zzVq{D>-j_X)5}BX0YxS=v5Pl{o9b*PJc$>qkublZbZA4_EE$&i;*rsULAi6YfVef?
zoZ9{)vyw42w;7g9GFU~EgBkJJSZs{@*F`)LQ|NQRs7;u&_jg8EPZ(RJO2>A+iTBRt
zw(Bwal8$9eJxm{HNdmA=wbHhg)Lv!}$SpuUXou<H3fZ+E_Lx=d4l}F1$GTybX*%g(
z$_7y5hSRQxyXL#uGM@#wUt#hsRk<k(A`{d5wB2sh{7Hlr`T}@NS~ofJBJ*F<&}?5G
zPo>FR=}xm{!B!5a^V?{ibCW+st2dc@C<k`#C-OJ=&Q;4|xf7?sZbhG>Wq~ukJ5&@b
zWERR-x|)OCYDU@CMkGS33HV?UXh&QR&AIdJ4|CohUSGo=vn1Fm6BK$~Uq_Ckua4|J
z<0W!R51K@tF^M_N4%Qz1eN8{}Z>GV5js$usu#oVoxTX6I$6R64VpM;a`rGhGG=d{;
zO*gChic}}HgMKMKU(WaJT_ME8&O`HcEpq7|^@6Mr&R#(I)3=MeN-oJ9YNNGDT5(yt
zG|MgKWCq4j$OaDZ`05WwX%-&=NLd;R>^O1T%I^K)O&NYq#x@_QoqrPv$8DV&c(f&`
zw!BlLGtg^1OjQm6sXkErGU)`^Id@_VZI;~TF4SnUc&s|KcDFcDET%>YrOu%ww@q-_
z{gmkLZ?e}*t&yCLNa={7Ie*P&gIMe_%Ak%V88RszFsRNUYR2=-U4$v-Cp047p-0o{
z&l~*YXo$C5FUe@`_AK5ctb*#*>FNhafTBndC0ez(q|MsZuBKjX*vWN4Ex`$$B(BIu
z#dOnlb=wT2tq32r=C|9>pee@7>#RA!3LDfRP@e=c+XK?oEk3Ddcc4ERc4c&2h*y`h
zF_dq+d^t>m4_{6?c77eRH0gN}L1DZ#2Ubc*)WG_)nUy5z)eK-X*$&+!xGcArfQwHj
zw_&q=yDfzUl%$l_mh#Yc|DC27$57VH7W#3a?s@vXMNCau{_3N7XSRIab?VD<q=)xE
z!{63rDH)83^$@>SnupLE)mc7hST~?g|193Id+MqK0Piyl-MM|0X9qA<r_l$F1OTpi
zj^*w&<!;<qyYQ*3ej_0~@;phpa*MbJ(Au^JdWsi0_2G2Hk~b~Mr<n{+NqaE1*7JEB
zkz$IZ_xSm&5GK_s2o@M+aadLIFb<Gtt!TO^_{&(|OKV*Z5wGrl85=ZeoFQYdO%+$M
zIBw{-G5f?cy8!c~Fusg`OdeCH4<QJDw%j7-2|J7rR@J>Y>k<4JUwqkj$thh-bb6Ed
z9a0qxSonGdDz_KSruOy7Bx`-06HiDR;C4!D&5?Ahs_AGW?ks_OieS>PUpSsYXP2IW
z=aP&h7m(@w;1|s_b6(8r<^K&YAlw$a9Xr~4*H_0o^>_|Hz7R%TT>?mXf7-LaFqOii
zSH`vzk;^89APJC#^+e<_VrZg3mU41^qq0DWF@`oo{)EI)KYm-Gna<~r&@DcbR&*Vn
z1yUt^A4wF1<DGTG1uQ!Vz~zAe&Za7q)H}wJAk(6xdMr#vGS{_ty#&J7^MINpaW>fG
zP*`sGS&ufdfEuytT<<`6zBr`w)Dvn;N0rcLNfP4v^Qi$z=PyLJlYLCl(kR4n&eQYv
zcc+2?&Olh`T`-1D7XGgI^l<Z#=LzNH56kW>S0OeX8UU@DP5AwCkBJXoMX4>kal`u7
zv#;>$L38J;0F?%*l99p1ypd@<fTF|$or{HAb+JP^<5?AOu!~}wr#7)a%bPr%1+(fW
zjmu1!lh0)9ypej@1|}QtxKY-Q{uGtf`7m9uteWk{Oly6jq^%D1mSt4@xkvJ+S9>C3
z7E<)#`h%omi6)P~w8k>;z?EgrrGb(yIC!)EgP$hBKwi0_`I(D4J=cdh@@8C}-C@1|
zQP9!Kx0~EJxfSi_v7$)>%~8{PdD2efKsDZMJpT5qg*wVz_MkYERr=H^A$zc}7kPyK
z53%J{q2NhpQZlb2qK?)_0^I6*rPW7m0|nnpX6vp}VO=`1m+@_A3{%ygh#PMc>$k4!
zm<^sM>ZKqPKA%BVc)KH`)LKh?RRr@X{Ri1vT`lhXthiNjXRk>pjWnNm!}r2Ohs_0G
zq$PERadAxyPi}KeW4o7ZH-6YnWft4ZJ4zc{VXi9UC~cyc$_7oi&z}4NfcBFm!U!VY
zV@{I7;fgS^uE553!ck{{90^T<-{k~LC$O!Tog(ktNgzz_jyOt))<C||`GxUf0)iu1
zg+Ddgv+vAlqS~6(qbPa*lok3$(@*+s{T|q#YsBf6Zdvyp*nOC`K4$dJ8m2wMi~&z`
zHB<c-9~wF0cdj!MPkNFCnR}1XM-COO)qTLPAE24;C9w5_!`K|Y7ld<VI4HnypF`rH
zs)-5edcdZbAfS`}*InCd##%u0ed0*C8${Y1FuLru_t2`?32@Ox4rik+zCFdIgh6r%
z;7hz>H3_RM$S0Sgu2W~cvzJ$@2ROrb1GVxay%z>bu7}RJu2(Z6f}d#H&}cr=f(@Fw
z{`V@RCqq!1Aqpl1Ct6;mDv+TdvV%$s;S?yJ5!Tojq6XWdl$-$s$wR$eV&wWFBacQP
zIW`M}r5qm8Y%TUp3RN8dEfGE!UWPkV$oa6be7G;szdi~iJcTpdvd)tU2{P*Eoo7q?
z!yDu1u$C?HNK{|Q@)P_QErn$};lvv{vf|p60_P8P8%hYKWGCIwudfDr7D7Fl<Q$a^
zB;h|s&a*Qc@)T%9E)v)blrv{cY`u(lW-Q)OuD7e;)?z0j^_;k(k}cArl7H7iWp-U%
zIFM7)6_X1(YA5Z{S6ocWTnr(|bD5J8)OI0hInll35$e}Ck|(j4ZGWvym4X+u7>dBV
z8EwoIMq_I0?>w@*;EOfnH^fo4m`Lf3$J=V@AxmimO12t4dq8txt%-;&S9;Y&DK7T!
zFXQ@^4D`kh*QXcPHbkb{<VjTcCJN3a<6>G(eYo0mimFjqdda}f-F7)A6?4?3s9&vs
zU;5{G7!vY{>D+=!NgljU%$@iB|8_C^Gw`67TW`acqoUNP4g&|<D)*4w4k)+#U*H#m
zUzD4|@Yrxu-1xlA1y$P6U8XptOew{3-HoHx%A-h&VJ2D410r0Tx(Qsxv(N7{Xk$?J
zB5|F;E3%$g>rb%(H1ipRJnhh<U63Kt&P&c`|6u&!jCAuUPRovI!=JNgqdGaRA}^zj
zvi1fIfjl(LO3wruGAl}=->ya{MY;`Vf_a}^soVbQ2FV#iVHkWT8N7R4Vy{qKlDClV
zmmOQoi?epKj)oZKTex4NJ-T;PEhy)fOlKn)dd)HH-Rz5hTd!t%tA@mL)|g5CERV)U
z#jE-JRq(6vFZQ@!%VTaNw&5;>wYf*ACP9<23JNe2U)UNzU<1CF5G}xwF!5BEsf`#3
zEbG#m%vln@80~4K7_tPic32ENJU(vE#qS}{3~?5BEJ@l;*7z47_XKdCIOJ|BILWoq
zTht-SZLK;FEB;&6pHAo)y_(h(2z?)7C?OwWQo&==q{~{#5ABK<Acd;&{e_iU{37O7
z-hiX==>2-9h{xr%(v9_d8JXu8cPA+MaZ;+bJ!y3@OwlT}+Meai6AB@!K|Liy7f2gZ
zVQE3bRdHx7At*}5j`!r+Xu*6ta$%{B{PM!5U(e-@Pt}d6!eNJy)mEf(neExgxMcD|
z%kYT;nBf&7wXVXe%&bDXm3EU*uK&5Xx#Assq}U-{z<ifkf#6f2Jp6YwZLF3)h{GkB
z_Ei9g7pd~(c^mUHT%gUkFA&pnpMMS3wuJH1gH}d;#`w+v^wMHk<R%s|uq9RtaQ5L&
zWm4~E0<JBP%}PR$Oo#0m|6t8@<*JCTGdC+fMbhnkx`7B@@dPf{6Ojv?!=vdjqUD3f
zpgXG(bCSE$aiPI^F{u`%6Ek_R5iEIueaw6%_AG6mwlVN5)W=wn$+G7GW2iLV^RT&o
zt-qEQ_1n02kdK;*Q0K6RMJ(XF<<$rpjcGmaw}EE+fUMY!F9v}pv3TQkpux!)Ei>{V
z+D2JZf{gi#XkdR-dhQq!Hh8@>e<pX3_FusXq*cjp+tUfuxeIdM-UXYkjgt9UX282n
z-CQ`?;!(Wj4~CkL+Q#`y^TvyJdq-+nwg+dw#52m-G6#nwy-m`Q1w)Umd90hlQ7x1_
z1^32{wg~MQZm#Pz<~8B@#Eg^bQn+<tU8Sc5pE;LyX^l5Ajb?z}-eUK<UuQx1)0zr3
zl8dTW*I#$+i)YdzYn$TlJaaQw%d`O`_e-k8iCCiYpf__e6QDxDEDrt#!VA`OQoaug
zeAHP*9t@p1t=o<#jRlLR=2e#bd(P}47bNeuYdGX<gF4X9h*&Ax2hVT%%3|!EdROj_
z!G~`<mUg8YZ2T>xW%uY6EBh)qC?{hZeYqz!jRNNHt(T?Je#V@C3(~ZAx&QWa%q(a&
zW*$8Fa@`(<_+++?m1iw-FgA>Tn{A7F*{EqDTg5_$Rq61}6@<w<Z1!g0s{@7ZbKJ0c
zGW^h11HR3T()5JwqrTbjCJ@su!Su6UAl{kZpgBbVo|jkWPNiKAek0Zr=9N@Wi{%eq
z>>p=TyyN#H5(J-HlfG$(w6AZ2-@&z0d<uhOi&c^qjjea*5O&XpS$mh}>tlcy6%ick
z7YcVSM|=;S`G@D7T&xbOYPz3}RAC+;^S67c_?$pTkXvcPM=v+DfM{}`i)IRi!r!_4
zNw!p<yNSe7T@?!sp>oFSBs5bW$L2G(K(?71mYp+ujt;c=R)hv~PfEq2_?6sLUi&zc
zP0q7?U;$;5F6M1sFJTYuv?OM7+ScJk6}z+wK;I1K!k#i^g{*5!Kgpu_s62!V#n#5p
z&+^lY%DBhSgbcLZ#H`0irVzW78a}<8M@8B02KjA|Y==s*z0H5%`hJ%dPfezg12(M^
zn)FFXOtw20{HZtwdpJ1?8?*R_lqwP5GRr?jX(e&r@dc@!6SOBf=2L$S+&^nOnVgI{
z>1ss4bZ@IBOAWnGI5#+L_+V8Fof(x@F;5S5O7Dw&3Vl@m?d9rPR|?B9o1Q_}SgR60
zCr?kjkJ!I<$p%S4<#!B0gfu*!LavGeH-NEG?UmmVHik`G&%(O5&xLex!`4iw)%AVo
z&(CN_0z$RP@eDjoI;8*ZOW4FSPN{6VCuBTK5D6oZ`q<`6yjbmAab^ihj!M^WQy$s-
z4{I*Wn4#f-p~9TWWh8f*i2TDHhYaZ<t7U7=`lrWxQZ>>-`*`5yyPeH)>Ot@tQJ+O~
z@HjTMvvx}bO-PU+$Gj&OdbdXO1k)!%n`hqEJ1hHa6Zh2DCatQQL4AyJ_QHtMtyL?I
z;LAm}_viL#_6mZLhE-HgFj_3|%%Y#Y@C{94nc~(|(iPVWHJ3Vx2pC#TGk{b^J9o@&
zv>CC7p{(KTtL5cOZKP8W(mhP={f3rNY(uj+fUl4pM?~R0wQ7};JY`Ye-Ct=#3B)@P
z@QT;FtR78dTkKXcUp`39-p<DFB1^pfs@L&49k!NU7%}Fmf2mVN6rEj_+d6>S&);Ym
zWRNUVmhPU{iQ1vdb(uOJvAZ1%Yl&~ofdb7$E2`IlNXsEzgJnv;J6}AzRhJmE>_1AN
z2z@Z9QM`{7ys>Bme%=O}=ntntORZ*u8BdU#N6VH976UtjosS!r=4(Lkwuyr=qjMk_
ze>j7Q4XkmT?}Ln67WfDI%djoa8ddtffqiHD!bSz&aXAfDMv@qU6$xg}7+ObGts@3H
zArC!Ahy`&Qp281RZ?h~&6X}{5t4w$2=~Ef;=Gbjy7K63362K391zGq2V4aJpYKUqL
z$|f~KCTh+;o4PjNnlymw!{I6jdHL8feLt83-;PB(^xcw|uezelD}zypZVt~e-Nkl@
zVuqm#th+-rNDLLm#}eyST-WZ;s9Rp{W1?JXu}Sr?PgSjOBc{0^?@imNZtH;$yzX|h
ztnTDThe+te))dd{38gk^zM+kwh&asS>55RZnUinVphZK$5Fhyf{zkrQ`3B&|z3Pk@
znMKlXUBWM)l;)-}(J||?gk94q30FD>JI^^y<;?eB<J>51qh{bN(IeOv)0pnGXfxa%
z*D7C9HCu-%TA(~?I9*P0k<O|do0o9t?5d%|*6#L^QDMhqx*F_p5uJPXc&QZG%i~`q
zxg6-3v3BA11SF-<hOhX<(v6^Fp`+GUt3uXEtR5gLx{ePxbIvd~$=~-dN1$4*Ruoz|
zUWYFruLJ9XJhzcP)MIG#Jv)c;MYM_*QYUOGF51i|%g;kdNj}5;3zH0$Za>C+o$;0q
z#!rbAH(7iSe60DEBIC`^7QPR8>&a1y1`^B+%)WZ@<miCWH5i00-KVSoy(i<oayoI;
zrozj-`#*9dn_<baMl=8OuwzUHkD2S{+Dm~9U5FV<+L$R%AH6~iV`+VVJ)-U<DxhLN
zvn1YuT%g07lnAzR-?icn8R*o4nI~uG#7Zp5FbQx3py*ur*AE8Z-*mDVq8H?P{;Ptq
zKHSs0*Zp|tbku4TP8gtN)1n;Lb~D^ckpGN`Dr#(T$gdX`nq)fCLw!_fjy?xHEcm8F
zuCY(otFhp6LAsoML=9)=3k8V*8LD9O;%>*bI_hp!k1FQrvpZ))2bcr1rZ3MiK8)7O
z@Mau*rf@AY(B>K~$<aioqA?eO-@;6dHY<A#tgA9@<!>=THmB=|ajfgIsg#{KN7^}x
zJBgG9E~d`}e~;sKV5uddalgMP-X4)@?kUd4l@Jo^r5zbzRH5$Dybx30EzO!du<w=s
z&_WMPEV2nOt`Q6Vqf1MRVjCUZAdpzmm@cu^U;$I!Aonm>l*--aW+mXTt!prJQQdc$
z{}_0whPm1$X=~}exN6ae1I4pMqXqGr$6Q8&t7k<d1AN=SkW%u7wdx-euT>cQJeX@}
z-I!6d$6E=GX?RnB4|Z%t_RBaxnWYE0f!tp9jo=#haqNVLBX;QkXb6>2Cv!|I8Xmf@
z7;q{&&JLfs!*s8_#YP)0J7hfOzCtbF4hyen2PT=S$N@2RRC!@FPhS)%=CHSnuvDA~
zO;w2b28W_LbUETu%`8#HpOchfYv;;IQ)1jf8&JKFZF+dIb2#f?K2fZG^nKyE61=X?
z+6B^1T0NS5_bN25sUmt&VvK#pEloGpQ@ri&7RDv-TOn|d;TxVkb<6Tzlh3YG=t)og
zHr`FmrY35LEs~s!Igy;GH1B^*8W35N06n2@KYsBGD8{YJ^j~FF_WvTQGSbukCoK7o
zh(*uB{2xHp|J79)7}yzD{#RK2e-N?Sz?pF_&{(Vf!8bn6x?Kr1pJ>C70vIri!I(G^
zyI*w+TnmJ63rIa62m<~+6ZDL8oaQ*re*ZT2*lI+s&(hxR-1@xunCk1#xz7-p!Zrg@
z2oTue0qXil_ywFe8Up$N@bx0l@%8?JgO23IxPkb9j~O(8a0vF#M;QNYkhcTX7(iTO
zl42i_M}`6{Yv%w02?78rNB}ZO`1%46@bL-%#SqjX0i31Y?q30r(e=YZ0Sef!Bj?ZV
z*6wd{u^Zy}eTVo9?f?KnOgz-~YX?5Y;ji7#F9pHBh-3rqAeaTw<OISU3e^9nZM9dh
zj;}s&Fq`E4jhma>jVAkVm%Q3$Db5xEJpC429(aJi_O`%^-!2Jo6LmJfZ}LdwAkbVZ
z@ZgU!cj!hR54DZ}0LUH$Sim3pph)Z?tRpD?EI7BC3cxuRFo7TDmA6Gd?7P)t01!9O
zAGp`l*IWL-WB4;E{C}3l*O3lDL)L(-`eOe0wIqPiZ6VLV)&;5mvNkyd2_XKY+@$;!
zlcEah$1m3cN;+}^ObLVft$+t}0_7Ui@1n0eyv2I=2^G+8BplKNyf^|0=+kHKee?u#
z0R=>Sy%~O0bHx244<9z`ON>oFv~l8!xneWp1%^-;i7o!TuEz=iaf&~M1_B;-ClwYE
z0Rg~I3lK*(7jydx_@O(K$Cu>W7l!ram6bmOfGv#qhp!G6;vRey+~gb%s845`--q|N
z`mHsH=np_081Oy-RU8dQz@wZK2u#DflIYVe>J=pYvDlm(K;Q4@>8+dzM2LnEL5%Ki
z+V8vpFN+9~3@A@cujG#w1s(h|$lD!+B!C;S@H-$75KsXC0|TPpZtvlldcg0*VXv18
z0js><Q0_V5%{d=M=kGA!OD|Xu@ZGJtraXwF0KkiH*iV99(7&pDlDl8&S3d2ZU-ciZ
ziQnY2-^}=g%)&!_j~71JA3hH$vbw@mJv4}uj($X?JV=o#pd()yJ-;6u9ho?mwdr%;
z<_bHhA?7gwE5xW1M1+%L^dDsK!KA*NKbE-kIA*V8mOn$VKC_ub7(fgD1AV=0Fawdf
zzrIdk1po4nEf6pQ(|pQ*hG4!wwM8HhEWSP0KtzNAQ)+5bblJmniHH&XVgU>&s|9j=
z<1qV1q5TICX#s?zeE_h_sAg_QA|e3aP<(`Z#SIDoc%yxZ?E!LE_!2<;<UVaBaPpZx
z#01Zxep86w0KG_kr2+^6c%@8D)yl}mhGFMVe)j|eAhwoKFZLlC{C@+XzJq@xa(neb
zpoE&hU>y19hHq@O0R-H5sig9Nr{<Af^yqufY}cN90qXexEU89+2M{pT+ckj%ekb_-
z^v>wsb}1nKT+|8HK|a9;nL*(XyuY%3lxhP4<X+#l0vvtZdWayFa%c|t!b9NH^Z&Gn
z4cQ4hz3q;K-Q06%a1QRx^%4YSBV!zZeU2gnws8#f&p=RqIpPlh@In|=YyKc#umLrE
zAEJKdY60DG5Jd!O5~aM|I#vIUtDM239w72t0kDn!c8Cvkb8>a`|73^L2f)nK{QWoB
zL~uebHK|KJu+Q5d#Pq(tF!xWYo!&4QCr4+IqWjSi0h~|#_gFFzkOvSz`xDcA{_?+{
zul=~I%??X=+}MT~lT<p0_angjgmx>n?z}EoY3UP6v=3-2w|KEd=Tpg{J~_|DC){0l
zuU_v2Tz@H4NYG=7*y~7rrxJh9Pp$%K!e+1}jkyR?(CS5<lJcaa{_01VOr$&+ywD0)
zshL%JG@kW9w-^)FgWCZmX(KTfDGnMQIdr_p7eryEsqGDp&_No8>&Pr_KdHcy-Y69G
z)vG<=&L^?yT9(vf4n{`eH-dvN1j#}#eOd_b1}DfTcvLbRMG?a(B#;<YvJA$lAB)hU
zPADB<QbcA8YNX2345j$UqQkV1!9cdNY*<i$a4-1qY#gb=qoya%oa=BFyEo?)`woV_
zwvjLl#B{PoYvdG<V4XX#f0v1lF|@)*XTkX?>^9%!alq2hbTY2+zptP~zjKO@c{x7Z
zH;YJ{1o-7?L^Kt?&BY43E8vQ8o+4(V@80AT0f?8Wi#sVRG-F(;ZHcC}xzj;`_a%!y
zNyI%~r^Mn8cI%JIp9pC-l>50A&P;XxYzLXB#81)*hh#+*+;KkF&fTct*mKw~{&=)=
zm~md~Oem5zAUP&R*@p;%ynhJmJ+rjLEi~6flL#H{NFyI4vg4#Is`^$U;Z7&QS;fmD
z`VRhj=s#e*r{DPC=x=WjZSP<^J-Sh?8uAKIJ3epGB6CG5(y+3$*>Sg-)BV_J?YL3y
zsav^FsQsk+|57D1tkOAJ8~BXqT4l509b=6zhi0;i61&R1*LG9l6kjuAMiookrf=O|
z8cD4`D4Tb63v-dyW2vU#7{JiFR9QtJn=7f~5J%F}<2EgwvD@}LT;n#IbN2S5(sw5y
z>B8Hw?%FJII{awt6aK>ROZTy&6B0DN$UeeW2Ek+5wB`RYbwS%3a8Hhp@b_y#&?RKc
zAkTCZMt!h71n?-Gl%<Fld}tY;9!~ORh$OUdV(~T;*zTcP*ykm@$+_a@-}1dhHHq+f
zj0<#L!s*>wr+3@s`=FJrM<CizcWHZ|Z3rs=#bIO8{99D$R%K7u_99vLJm0zEP$n4x
z9|I74ib8T;s?%}<$0oV!Lca5Sd<{+Gq1i1xe6*Ml7`4)7p7fYoHEwwx^CSs>MEA`S
zCjWE=QgNQ-P$et5BlBh-%yB@Woe3xzf$p(hA7H*^ZIUOT&59h`$npwzcKKa{V{GP^
zCAA+FdOP*~T){d`nI*Ui+3P12xJ=c=Co{}g#v9?o_x*FaatO^+pjW%MNLR$zAttvF
zx@EMUPPvy=SWLVa3-i$6?X^lv>-0?>%$hWTqoC9=GsH7rQN|EivKLNCcQoO#TP?ZF
z<i(_&@fBP;$106w1W_%gWoJ9NE|+v(Emx8%Ia8B}_x_W_<z`pi)Nn~+cr2qNNjhxW
z4eaE~6P(m;&}6l3>cXD%w%AbH<`^<2)&R+g8#ZqEkAcv7q+IJfr9gSX#C^gfAz7uu
zSMJ7D2V5Hz!0kJ-DB=UP=F4l9SEy-wSjel(3swl#&x6BMb3Vqmx9R$ez^0$E`{5W<
zBN8aWrRPE?L_*FS{x+Tfr&Xo7KNisIIRNwe#}T+NTcEsT%KbZMN-}}z)T%Cx>qiah
z926HYZ8w8aAi*kLwIJ-hUS2ZjN3^?5MvL5ULaLRRx=baLs12|-LeI*>F#?nigi9T>
z^^8?vwlnVNYRY7k%xWR)EPea+k?azb7ZbDcfr6;B(WjLs@R5UF0Gl<KskV4|LrrLT
zak;^SpYy(m6CybcS_|Qrw87&oyDqSWdAPq;a$Vz}e{J1W>_!B)%wP#=xCD426-L&J
zff`+T(G~{dQ>}IS&~bB-F8|v9L@zy4qmXSzb~<bbSLt9SQ@^^ZzaRhYKuv9$gAwzI
z?PFK{AUnTG8Fxl*r8W-SiPv;0HDXh?-H}Pm1J9p`%prq-l1`>ceh*G#Y))^`BNb7N
zZN4Al;k_bjC<@*%YIJv&?CWulY<=roxrRJLnx6N;n`6#l3-iHTzgqueG)fLMN7en$
zLOd??Rf)kVJGkG{hqFj!7`L5w;l$nM$D6!-tFe2@`g_Y13>)o4e>p1mE(67Oj6Czc
z0M}Ocs|rQP&Wu^N@Ry?>N5u?aV9g`yOQ_pLrK;L|WemLH@H1)o=62xajiVD1T$`?N
z{ZTY!xbBRi@+_~n4lwb$5*~z+oFv{QXr>B{iH18B)fjyP$VUKLBo9W*rQ)ctp=iXj
zQ%gEqrWOf4XRn>tMU;wB#wKlxN#(`nS;>jH^}siI*pL_xz+aB=XXwn^l9-%NS^rCB
zBt=69v1PqHe>&cIM_znj91hDYzl51WWf2qeG8A8PrC%SPm88LVAOG`&!DnYcv#VrA
z`W4-*%;0}ac}Yj+kDUOq7QDpVHH5>QjB9Ss0-J{hoS&{MB}D(`DT8c8Oe{#xID2{V
z-{ioRNYTR_;No_&%?F?7axS=MXz#R15Ads96VfTrQ+0@Ug`=CB*X)4J_<~VL7aZ^G
zfV@Tg*U4ZCXSR*89ne$HZjyVMh|d#QIj^B-&K~mA8a|DBM2yelZ*Gf%lf50a9@o^S
zBhR~bge5Lid6ToQ7{S4q&Xs+%!iF+@n{nP90u!4UjY9q&91$-$C&tuURhmza%KfQv
zhav}!<U;kl4%laC#aoiqEj93eP5YSMYm`A^g3-)?Bv;$9pNbcPvE<$FJ*#Zfo^s|-
z=^&piqTE6y*eE9iZ5sP(+;%n*MU3t~n+~LGO-QQ2sJg%Q!45SdHazXmLS$R@l{VVP
zyt^n8e0|8QH5&Zb1<3vDao|A7qsePYP6Y1kUeKP>uGNHn1R*MD;3iS&S6om=C#Bh+
zsB`k`rsTjkq@*!-yqie)ICs~yJ{I49Z;Nv=k>^&;&X}9hTLjM)IV~#brPJ+F?3h;<
zQ^KkDqj!4p@oVaiW>?{Tw^Qij@FH}fijTR)6T|SFveOizCqg$?F>uQC65KXUTkj1V
z4Wo1%H6WK}dieUUm8KoMO6-d)11M4KmWu{~{c~KiTpD=NnQyd50Pe4#?{p5uP9s2t
z9VsK|i#0;CQ3MQ?yr)noa~OXGRR&&KjMHFFse@VwWC}OMeD>!*Fzx77xnJNsDDl|x
zZt^WVF%Gk)DgMd^;E!=(WB#~wxz@s|4bmSxzmNVy)R1v2icQ*cGodYb>y98rzP&uC
zEq6@+#qzCiKFyi*SMSi5mR2W>2t%(`kjJ5%UQ%Zn9VeCX_xS--94~J6^RV$3GCnpd
z`Y!M_21H{C+RZEwY|n5B-XSUAu$gW63ci7hbiI2rY<{hF6cteh?H06Cq5Ju%5!7um
zF9`!&BrYM`$TO9W&;vrw&bCZqr<$a_NLkp9)ow1W9wjh{6B7|uDq}?w$YNRg6iq<a
z(UJ@0WF+AjO<^{(mLD+~`yfWrph3PaZZq>9+jt+W;~?!FC)s#eu5i2o!zh0(pvGXW
zcc0D7ik_JFIqVC;(Us+{$eq6xC%u>q)0@N48&TWj;@oEN1*7d`qv1#i;&K1%7O6|!
z4{{p+bG5<-O_2<FC^|&%Hha=2a8wg1>$khSomBfI<V1S#DP0@ZQY(u;!b+`lP?SWw
zV(U#5jvSgYxuS6c>q_$qZm#z!DS${AEaz{afzH$xE>_MGZSKe%^wY)VC2730l&dNB
z2Cca7u})<jPu>;9jQ2;Ii*mxoE1S(d?^%QFbWy08-Yn3{dfURHZ?N)Er=7w#e?dj3
zA=}2CbMSF>Y7{8~)cWZDWr1D6gwb!}x|J92jHMPrIH)Ef8P2-2^D=mM!qlqP^m;N`
z1*z7lqbrvkk04L)M&pBeD-2TtT@{7Tom{@!@bL}WWf}AJ>d{|&{;?U(DT8~pKyfi>
z82$#0&yEZ0(q}|I%k|S!mx77!U*|FUt?b>$acc}8mW&N;;T!LKF>^{hOjY%yG(?im
z*G(WSFeoF>skkX99oTLPxTtSXjvIF8)t8rY2^(Hru~Lg%q{=b{jcNT|IpUyLDazb<
z-BlYA$4?~q2!#-6^7UowfgTRI3DR7yn~tev`@ROL;SE}<Gu&MY3|@X;GgidsATH=G
zatr%dc}B3kyqPZY8ygSZnisunb||wV=*Zu{MnlteDfSto3veWL+knRL5UTjQ^{l=R
z-VNNm@<}esBCqKec`0xDt~~)%m|?#+GK+~{tGZUQpLJxIU|fJmGcdIs@{-P6Jvvp5
z$!5kBY&g|<rF_F;nB>@(^=y`ZLb4z+c~mA}Q__81?nE^&IYHVNp#sFP4-W<Pji-K+
z>~*soxhgIf(}dDf=UALr9<a{2_NFz0CN$}Ex0qs=S1cO~*xIs}?u|TG+<7CyjRuFk
z74QHblc+F-yQYjnsI**^@2;e5!A-!yO_=t(u^5{S^JJuctYP&#fi@zQ*B9d(a|Lbb
z7PHQAoh=^{oY<HM=c-gE#@@1K<FIER%$^dpJCb-K3%fs9h{TfCVAqM}zXRUo_loc?
z2QO%zjY_z{iTci+rctO_IR$p=Qgz%0MkIW04W2~tF&&_M1$mIK?IuD=lxWX{$K#n8
z^&NLK9pf4rH0`d3hk8^E`yZxD12XE8@`X}qGh=g6TdvkUY#K@Q##LAjv7C$Ayo{iJ
zzou2SvPnf-<K9!=yO}ALWBY2?89P@R1o`0vnew43l)ttyGcbwKWM%kJGk7bavJY@5
zKraW9ou1@sX0nr-fr?&U+a-mSgSPkk5HM)CkNMNDEmZc6Ei#$FH@IS?qV(2l&zQoY
z*e^`hv}_wAu<^|;TK1V&zz^GDIvQD*Ve}}dZ6VAFSj^M<E_j}tx4e+X^l`0`n^!qG
z6tuZ*cau-Cg^QOdbTG6p>02seLqi-?T%<8w{fESj-l*1<XzTk{mX^}7U=>^7(va}O
z)}vM24+?Q<D(K*fsZGXOSdU5Tm5H${9#x}M%&I{Wu+%nHLcBD&145prqWMe}4;&AV
z7^srHB~;v5eJ#XE4!t7IpE;_Nl}*D%BszHNT+QTpVGbRB81*!~Wk+j%npFeMp))S-
zD!1?0pwve{j&`J!Sl6`H<Vuth3vYqqXo3)H8w&cWu(U?MCkPvZbMaWg1pb_`{iYh*
zhN(7#;95aZ&*s@!7zo;aL%L$iZmYR;&VDbW>%JLT%F;oOdqqc5ta7t-#(N1+h)5pE
z;^PaZ#n?1a2IEo7<%Z!ZSO_DHRyGz;k|a*%>7l^wc`v~b?TU3X(`(hjNhbYRp}b$t
zdl!fc$sJ*@vSr+r>MpS4XITQOLY3*%PTzCx11~w$IKRhO$*cmCP`H|o(aqMLIPlGz
zf#h9*9$ymINA&N5kf+*A>?mHe4D8PEj?PK2blTJ3EJ<zq_~s;RT<&l<c)8}^b?BC8
z(UfJ=DTFI5`L(9}o}+<HaICsIZ=e75ksR4SpO{_n1SG4=E)M)7SC+9pu?iKU5X+@-
zEu!07mc@40jp&P5a0oQG9RJ`^Ls@WeQsK(u+GG&_3|7}Sb+P^+P+J|n|3fVOy=4X(
z&t?4Y(}1IV1mhJ>+UlCjRChPpMsCEbRw*JUP%j-1ILtfU-dQj*Zm=h>UQq59@|ukS
zcE*JPkB<IKT5VbJqZ>aBA0k}0gZv9Xc<r1GthtV^d6ej&B%aGxlf6i1s`st|wSI7c
z?8Hy)zly)lBjbd2Ej{@{XBnqv0S_;aYlj5Y#HSfSES86EhU6#1TcJOv*doS_iJD%j
zkpP>lQPW{+fGvB|I4ytspm=F>IeVgIWTW`sx{iFD^yp~^y!x~>c~3@6=-`bV*RD;I
z1b#{s5oawc9Bh@ToSqX8z2Io6m`f>^pyzO-$KG%8d*ZN%+0vEm8EJkMDr#Ms`wd2F
zg&qVv$R;5cVzuPGhBlELt%DeK%fU$Q`YvlCtomtg()INmKqH3m3{U?pDB+a&hB0N5
ztS3#}Q$(rcTG1Tso2ZLHd&Q*ZigvN^o<tXYH@gd>Fa5Tp)~IJm<}yxVkR%VWzc{Az
zfG=Yjf0qRl3Nh%N(PALMJUa~c@x%oRUtV;L(!jS$JR;f^jfJ1O#2Lwv8qd5OF;69f
z+;(!D;6ti_cL;1MWMreOggdz9J?dE)0r+ls3lPek*=n>xJR2nvD4AT<>&n-F_edmS
zdeuK$yiv9$V#M9=+2*X!FbZF`XM6dU{j@o*)afV7tsPYaZ=ZvB-DmDbZ7xAxcivC3
zc<-HPHjY$hMpgz+b-CLkiFUlf5XfhuE7yFlTH$@ow!or{hkcCXj&kHEvw`S|bF)ai
zP}_m|lht)&%ZVNk%A&&CbiAn+l=Ae>_^-k~2Y+Pt;0Vr}ktA~^tZ7mtoQ0IM@-!_l
zB8(%B=2z0d;C@tD!Yh18=a}Qh6Ry(V9S^1Qh+9FCOj8R!QwX1{j>>z<f{^o_9b`P*
z^Sp7}F$bM=$o{SD&?0s{!?HG{cyguJ)6BZ7y49&FF6wX33Yx}>2K>vJCktl&82)%(
z4>1Y@$ADSS{rkwuaVm&!cJ#rY&LpP7r&@jNJ+&QbaO1W9sDUxvK2745m+A2w(0h8-
zaz}#>xk-og3i4pVijQs}%3QH)Ei%QaiQB>1UV635pQ6vIJbPWclsUnh^c(;0!pb04
z;c{aN=xo$m7Q?#^gN4O<`CU>9XyyPw?blrD)u&yHCmDFZMhp`%|369MqQ}g@l%HwG
zl2Lza#|+y#Pos)5n{9yFGs9P@7G6JAq9jm*RkgM7n}nuMa4flf>%u*<YQnL))&5-h
zl*pDwOS}_Ikw$K-t_)6<4AC1Ftvap#KI=C5Elk5@bNFY)d7{3|Rk);lD#OYo{%PQ%
zjd{He@tIt>hWPy=EJs~5Mr$xnX<ZdQfMLQC&$H;)7bTw7l(X3l@ctwG4?}>peL1{%
zb9nb$+6u{{rQ47R3||dK)vE6MC6zBHVgYfM=<ziqwIf-z74N^yRCA4?)53i<sQ`r*
z>&&|$Si1RfBUQA+kR&tW^h?FsI;gM57*Yx?$oGzIK>ooAHEmjvi0?RMboa(=IoiBp
zF}nkWbdmhbis0t6wKG@%zlz((`5M=nEv<gt=IgVYlQ{g=Jist{@wT8<(M+qeY<W_A
zp!Nh$)QLD#V+@zkHr;^H(EVp6hdV#XX|!h_;_b0=+6QgJ4>-n}*R(tTviLm*YGgDZ
z+J^l+B2PHpvgYSqQ@wng0vi?0T*vZU<Rg}*(r)`z`E;~-7B}4S8k`pJK1q2e?IuY)
zNXk}KukkLS%(Hiia5@%dr$eigP;G1YNEa9rN)}PVQjuSRUECdL$xkSRFM_zV4CVxs
zua*0o7~$vH5Q#{j^(L~V>(BLP<SgVs*2WIsrkg*F=}X^{q~S{E@vWmHGO_(<=Yprf
z(D+C!6BcbPAKtCQWH^6_N@_Bp(p!V^jVKY5>fIEzeiy8!S1*dmhQi%G%I>X-V3Noi
z&o_UQDw4YSEQN+`LuH){&>Nd}r;l<)@<Mr+1V22x*u2MUgVu<4+BLoBX2qA%Cx|zE
zX3<Ada43_VU@kFB%>L&<CHR(9veU)J<=9Hv@Yq~Xqh1qlIq=c$mruD9S9)LmINw34
z%Ww!)X9}`|vtvj`=Sg*+T(4=S8G7=<F{l`+Mym$Zv=gvD`&3l&d8}8g#MRl7({}yQ
zdYNvINHso@F(&G@jxSx4PT`0@1-hBkeU4P`D|11?Y26c6Igu0l2x>#rum$htXE*Rn
zHQXWR`O%Km<1D9UR3SpB6y-=FRrw#45UC?Cn8lSzg-#U8(rNcmjau|ee8I+!;MWlS
z2$<AYztQW^ODRYNjeHi*l|YrH(?sJ9C8;aMzso3>{0*Fk8^FZ`!CB2w2BrNDGkq2&
z_vJ60Aw?lum(K!(%W{N)Y$DIRxHHloJOb3$FD`n+86%|iFnBry*P6Mg8fXt5%01>q
zMb^J~7Wm{7h!67`RZt_i#nmw#KHj;vl<x_Y{ydUD_gtSVKA2y*rF#aO)XHO<*0dTP
z5Y_cy(oI!oH{K&QIg27qL?D~$L*bF1kpYp!-iN4}iGw`3tAG(7Xt*kBDd*-b`9VQb
zolN@fWK_RH>s;oF#qkZYPoAp_fKXBJ0m;q$IEj98;^sEHFmbfbN2RfMMM9U1T-`?J
z36b!2&*yCQ>$!^U9O0pwFL2N+2$gER_9YJq(2r#o#ajP@8e@}a$SCnIT`0*a+T||b
z&gSmoLCf;)e1jd^RSEd!WX+AdeApcwNOLM1gL*PC%Ex@Xyxi}EQ4dz}jnx=pFGDmJ
z<|y-R!X8zuj}~~s>~i1WwRUA!42Vi+riZ!*=cE13i?rsr+GC(6)MWzDBJ}VFtGG`M
zh+n(pH+^Mf76Vu2s+)s?n^@M3r)s3Sf3tuA+Q8vUFURShN2&eWStl{Z2s;*A+lw``
zcn7v{PVhLh5nAJNX0aswW?f|%^owmf`78B0VKVg4zuCvv(sHF&0WiUrTI5ey4fN%s
zv9W|t9Hm#7$uez*DuXDsnc{tb%6dqK@YpRx6$2UGYKn6lPt&M!b8jpRc!x8-Q1}Ex
zUq1YMj3mE^@E<Ha3}P+#9nv=@xj(PIo}LI&;Z#Ohn5=uh$IOc=;%uX&Fq|CqNz1nK
zagPky5`;iT=JFbpkX1`^d6N*D)JbFYrCFlkWkn*Sr`1<KyrfSeyY?IkRU~gzz2Waw
zSt0HRH;U@xtmbTl8>hQ!eFwea31UkLrWvBA<sBd8rr2V+KT}~%Fh`DkP?_Ff{Zkf0
zvC}#wn8Y@GB_1BHq@-yFDvV=DX@kgdb6>~E=p)layYfz_)959pajAFX_uY<2uO)Od
zUOh&5*#Z*cl?kz`?Fhq895@p9w+)ey5^mrV2q`11j$PT8U8)!pMrr3uhro4F+oE2m
zcml)gsf^tJiOwA|6wI+0j}L`yIiEePIL1s7r#ph5Em0i=Jy!Z`-p|3G2b*dydV4;L
zn%P}yeO)Rm*h@Xji1G#%kaR9HvleN6_aWKDdRZDK(~~Z?Tb<w6P@Qjy7;7NUveUs)
zxiJ!()#3-6UhkV>z#*;0&lqeavu0BBPm`;O#y9(g2fG+G1q_!Xb=Bd^+X1&5tmtzh
zeXvjWeMC2O*RMi$MBM<}PlXU-5^HIcR?p8Kt{$KsPNDaGYq2BOnR9`3QNRmIJgU7Y
zppGpw#<cZ`cpw>6ZkIOlfo<vCbeAt~S6c}jGr}}iX3QM(hFLU)Z_%QX(K-Z956+-T
z7FEbUNBDe<?uvKzFFCpj=nE_)!I!MupDW~>IJFOkidBeYzE*(}U%YBM(Zu<FLEHli
zm4P-hiM;+C)5rA{sPj%TA@_b)gZeT#!2;(4R-~a}b4Z7uQFmQ;dDX{A(DNLLfx!v(
zRP5p)EHlHN$pN`NbCz^f2k1zshR!+Bq2FU{T^j+JKdlx2EH)ZxXH<P@`@vrO{-_Oa
zKj^4TMwTw8Ic%Fb;t&4QkKX?rxD9%fH<g*>*e_T|s3pfaJjJo)X87iXL0s4I9e1B|
z_#Bt#!8t}PZ97@etQ&t-^NJJs(54CL+IsnU+`(dE3Ox(7Hq?!%9Q>(p{YaNlZ>7dr
zfKT>I@@1!e2R4n5Wt_9wRu+lLr78c_D*J5e%2NGoQ($vddefIKPF#>Fvd~^OHcBo}
z#V3eej6dTuJxs_(%`$8~bFWc#DA7Y$y~L8Ldg%&pi)<A=W{~upW?wDlS&<l`yEMP$
z|9KT^s3I;SX0{+2D0H<i|2}CBwO^Kf1Txaf!IxjaxOayl$My`i<wS8i@;Q6736<;e
z&5;F7#3~$495u@pw(8~NHK?{sDc*MdzPO%?x&JzL(J2<KZKr4#Glu*<C!n`$53L7*
zio)XSxyc~Z$|t5`7P#dxUm@pudEKuV<K5U?4`L^%^++j4dgM2eIeFsG;8Oz4bO1cE
z9VbqQVn3ZV%ZR{kF7BZDeo!5;9kW=qR$B1fH<1gxc9GL~8XLKdpe5IRb*tIF=jB*7
zM7g8GDHbhnaGRx3V^6OZUO6CfH<D&?vDSf1oAgVUHbqcuJ!;y6;2LQ32XA$F`&I_0
zXc5VYH)(-!y^6d)(Jf?_yf0`H=@>ByZm}qSOFtjN9~WqjsNW675s<G|=;$5<kBYA+
zz-~}MbOiq<+y7+7PU(^{lvsBRZye-f?$(oKF=&`xV|wiz|EA8u>Z-MWARG72>h4v1
zo<TAvYg5nXvtf4(JYeO$qHWqKKeXw5i1Q~8&B(Oss;%9NE>bY(@~*riSdd#7ejFG2
z>)fnlvef5>W-mCY<nEgm)YnmSS$p{4)mt$A;U>1fW&}<)b*=I>FLz)qrmg1i)6<55
z8tX(6_iSRQahqevfa!b;fX1f0FtwN>;mhKCQxRtG<q*2sxe(-jNHv^rl#cw!{fOiD
zwa@Nu?leD_Re0^eanG_|KdEiJ*XUo^nPcPgN9g3JEQ^!<7sFPX1C?#Kg!lk<waT^#
z({Ur!059Bj=G&>55vc&WMqY~mpg9s|TM=nNj$agS^zo;bXZDIjTrTABAd~O)04wSg
z&w;0*&O>asEeutcGSSWUHW+;cr(|-n%RBa5c8h)eB@yu|Bag2woEDN=WY{D0H68rj
z@C?8~Wfe_N4|^S@XZHlvMAa^6;fJ<{tPZdiwOJ=U=@1RR(gS+>Wv}XK15FBrEiIWB
zUZz%l(shY|bDhI5xc;AA#&bg+!#gJD0mASXpr_Q==9L0?P_CIo<zje(C~_wE6%gLo
znD5C4GGRF(pG42O(@TKt!1Ca=dgoa!o1Sze@zpMcs$US>7TxUYY2a_lT?0a<pdo&_
znf_w<5w@Dczk839$Aoa;Hd<>pd}}0qo$n>|s=p?o1!NJ8vDrTwk$bXtc4LLV1`#8Y
zN}2#JB#B{>qO?}_?D<I8wWp8DuS{8W#=wFs_X9??j!fZNk<GWfToQ}imay4jP=klQ
zbu4J8QBwR5(~Vm7G=F3zO^D5TY5m5ucJ^BdVqh>AQZBJJS-Pw7QFwetdgmnv6iaI$
zCA)@K;dktTl^l05DO*@LQc!~%!Q$BnKW9-UlVT6-XZA=eLY*g<;)TBU;y!p88V82Y
z?tD|oRo|u(V&9#25lx59!TswXd_M7U5)3MHUZwqof1QZ$M+Dv2>$HwmYOt}r`7QKb
z(5F+Qmo|>9k6sWd0GbXN<~6>zPX2te?Uo%a7!hI0rK{SakZ@Mk8B!_)p8LR+b?Xpi
z9l+qg9}UfM@{rfW=yb4wuL&I*j$HZ-fGTdhskJn*{_2)PAYd|EJ>Q3Wc<-Fo3r#hS
z>wLSvpPNMv43}sojE{i;M4d~qz!pm~QV?<ZT3O_-Fr8eLpmDG}4TOX8ZUFM6nm+E!
zF>=GFw@CC#(yt(m91}g>+;W)jK$KCGy^eQ{3JwCKQc_Dz1AXLXLhmv$3oOj{4PfCp
zA&DOH&iS*n^GJ?g)VD_>C6zS?vm(@Q@ETpjX>>)h0`RhUY#(qWR~d&)hb@MrB%;C=
z_wIw*Ug}DAzK+3vr>(_5KI4Ibi8`@Yi<+oWWMmvq3_F?iYBph(#cWwsUCp#f(y!q0
zV7C3q;50NVC$;r*_Wy^nbBNNci?(bi!?tbPwryrO!?tbPw&M%iMuu(MR%JEb7_ad+
zs>W&E=4sur=UIC$3^YiZ3=lp?nabJ2m>`PO#)P?30=Gq{17z_aWIX}#y!YUkf(<D4
zU3<RRy!ZDlf#X4wFv(IFXY{-I2>p|eWjt@9c_NS7$o0xX;?bW-<)jg^&JH(`#|}<4
zdl~1e<Fwm~p%MAe-CyKMs&s!POs@aNasSzbl)bU^@TyQ9d2T`PJ~F4+@V0lFsUfoa
zC*EJd7hsH^V`_M6&r6;4)YKlMH|2S8IMeyPKC5Z1HSJ&a!rmo00?wXP40FG*FuxWP
zI5NIIPX6_HtDh5Y%Y7%z5l$g}OnzBI%lD0gf5b;6_OSteNWs1S%K)l_3?Ig!%>~54
zwOC9&@0!Cpd05VLibFlj@J@D{>p}++*IOH`ZmMFchc$NPL<2vCHXuyVwXGX$!-wVC
z#&n&CZ8OJ;XUC3*1RP6S@{w{*w70yZLdq5{JIh3yPf6cJX=?bgvk8Tw^{s392%Z#y
zEg=%0u{gsQwL?<U7uYP2-4umiD0Hz$v^$D;Uui3|@+%UC3fglFl7YK@ev#5&x-WGB
zS+CQb{&pgEL$mOdx&qzNG}m=*jKL7wi%D95Zgr}oPFWldwaQx+TRhbl(+D;PJuMY+
z1<0VI#DdTqz9{*HEvbt`%t4aa1@RA62{Wr>7@}X1U9p!59zl-M%2LoHJ&^FIiP90`
z;-{uVL!Cs-ByZC%Blz~pTn7=i&+B86(a3#X*Idm3W+Q&H57Frt+5ThXA38E}$idN`
z9XM0`+w99}84_z2l8Fr!kf@m1QYSW4PH?-d8ZtR(0!8Y|CkAlYyn87^(22*n#Bz83
zs|zFVu3xi@9+1ufzGHCn>f3xMe`d4pSNeJBr`SiW`$-V3S656TrjHu;vR0KH5=sG=
z@8e+;@y3_LBMqZX|5#5}rPh(7I_um)KYw5m{2VY?0fsL@nFlZ4!U48Z8(?_86bicB
zNUzHK@0F7;Rfqx}S@z+@Q2#+=xR9=V?O1*q8@I&n3kc8dR}g(p6<$O#hL*C!aTO?i
zd-yLhxMs6o*1UlbkSHfLZeAx-7w?NR+{ATY$HG(&^Xv<OpD&rw2boK}(V9R=HAD*8
zY<Ke?7PoA5ogK3$K+$p5Z4eC11&RK4&1exCd*>FB@C_0$Pv1?5hHvxC(@7oUO4tec
zU<oK`C)y{JX0uu=IVu;|N!0S;@^3v}I6H1DH3U=YfiX<V>z>Zi@bhkUubXj7Tj6S)
zD)_VnsdM7FMn10~NDomRK3{jH)3I^WL2(QV(LO>FTwGIRRL|xl2LZXe$M}$W0lLOh
zd(Hc8&qjbjO7TxqjjB`B#Ag!?dHbJjF?N!?KuAwG^tfZh2SnJik2Yi6?<WW|UEdgN
zxvj?8oI~2Ogz?3i5*_*IGLS!_%hqZ~V;_H1>wP{`0+wR7R>cKVGl%&O1INZ!0&?kt
zJgCeY#$AE}n`2L{T2KoD=Gt|DJk7LHmvM)nU5ev}F9(Or7RDzkW06DvOw6rETU8F1
zd}_z}WDVk)Zy;zz7zX2kS8#6;=;fG4phyQ2YPY)HR`niV@N~3?%?p812+Rv#A6IWC
zkx9caVj(p=aZSv}U#(~X0nJFvOM!(Qi_V3E%?1%%@cvfdh;m5A0sY-zQ_mmCrg2`<
z%f~Bbd@$JpfAjJoZup(T3mUbdt<KIUe){nTS6-@gdQc9KKTX(8DT5y}y#LkKR4$X@
zVbwQ{WAN!-;L%o0A8R+@R0jA3GQ{eY+ntBd9;kb|dgL^`t&)$xhTP@)OZQmZ4<GR(
z^>K`Ayo7bDHUKEDUJzL4CnQxA=xaZ@0dJ-R8piRsA0=qOX#Su4jlx6GuI%sf?|qM?
z6k@-=V0~sLjtY2!@|>c@b(xqhI?|OPuqDYwyRHQ7tv#L_gN~Cn@9fu7qNrJK1e+e*
zTB=9nu1}o2iT|7g09%NbA<IK_YXvR_B-Lr|bmIN`;B^1$xL3eMzT3&Ug<Ty2<Ulaq
zAn{5c7E7v{WG`_Al@d{8j+Is8#U&i;sNpH8js=&iG;zX(G6uunX3CczC?KYj^HEAN
zc@J}kz+n{3WTY=Saj^mVFTzPs#iTr&thk8iQ3L{S-Z5Wbtw7*0gCJ}tlnR6N0o;kI
z@fm2E3iU_2(bu-<U@-LQQQsW`eahHbj2A38kf5;FaA)NF=&$37io@GiIa!AWoZXxR
zs|{Yj{ByYr3SOC$hd)P!VueYb#FDTpjw7m?lkj5FBc|N<j(xR-t>V*rdM}t|+g)Ay
zw{QgOS|#f`QXItuQHC3X1FQ8UTvj|e1(&BcRIOM!r<8ENpwQ7`-GpiPN=k=Xg|z$+
zJL_6;jxd}N&=uoN=~rZBCe;sgJgv(veqX(1j!@QC@|4~htsg&}?P7GFgq5#U<n@kf
zUyt_Sq?s@hwmFmL%uBUu@ci?;H<8gQw<VS{Tp*@=3M6Uciu_kVot5!F0_sf6%>OH(
z&cej}pZRnaE_UYsSMQgPDio)x*AInNs+ESeGpbe?R>q}nXSH!zV{`^f^vD#e(@3VW
zD3KSj9vWFevSPS(MRf?N$*g)KL&E25+e^>O&-O>pjZ?1qjQ7gy%g39KuY@@Cfj?uP
zN|%rdG?s9Fa)LGxIF4|`h;UCRDeZMK$1j;d9A!y&pork2F%n*8W^q<T7$hnaV_Gmv
z;8IrN-!X9DiswMkB_YZy%1c1136g4@b8DGGqwjrOl&b-vAe5{R!$HEp?!@pQT|xxJ
z84@88-qEu8K$OipU+&;fvzkDXMT3(5n*^u#K+!Q}@lyRj_@uYxm`dgStc7t5U}NKZ
zhrq_fry{_Numsh}PT+^nb(Qx!Ro0Y<NDv^7KXvcD7=ymtq9mRygop@?WA5?bf<1)4
z%?`lQ15DyW$@<y0Nr~a>1G5s(><tVc3oR?k!N9?(KjN__P&4D8;X{c7z$Aymq?P{_
z`goZ3`NRJP=0+Pr+C%@12w%)Br3O+?IGrcCqn{-S;r&OG83n5lNR&Bts4s|-_mZNo
z8dw$kSM0|iQX5*Ra2T#VD>2%?8w35Tph65MHDgByu}5Xb(+`4IgtRBm4K#iYvj56s
zf|Ma$0EUDOOy}Gt^oDn)hQfh87^o>{-G7FJ<xaSK!014pszj)UN0!{Vo;3M&9YVpp
zlFtGo<%bwpqAb<F(VTo$?YD<Q#4SnudEhOJMVbh=OkkE+A?8mdzRSP{BFpt%1a_f7
z`s+vSQ33d;dml46<b;LrqYszJ&kgCUbchf!j!2l(z~7&&zsT+*C%sIbI8i8l|LW?W
z^DW|2XKF4joJY*eG2olz5aheI=7$SHa^%McEq&ukOf{P>CsqfC<~t3hJ;z#jMCn%6
zQf*EGm^LO9?l|BE%eo*y5w#kI7jtMal_9@Swp7p_O3+l$@b+!aJ$>b}oIKxXz}ZFz
zyV5`MtL5ni#qhU>U~K=QAJQ<c!-n{m=^)B{xKJMyskcJLrL}<Hm)j454mP%OpuZv*
zG~7c)M5q7(C7d8$1=jOT1t)k6c$DeK0ubP-teegj5f;ROm!Ol1^GdS!aSHZyt~M)%
zZiHfe_7s<^QAW+h`C&NmxZ!IS@+l(ieDxi{w#U(h?d2U?+qG7sb~sCid(MipKoE@)
zhQ4ObP@$sQ+sX0T-ZARQS%1{o$YQGviM3?aJ7I~DWv@U?Q0_p*Z^H89i7~GKq_uPt
z_oZFUfftYQ+}`cAU_&OiDu^}6apzpBX-X54PFANSAFRr_&SZQ(RvOVNNzXj$-Bc=$
zKo*#(!Yj5%rtPgw-;JG+;Ak6gFJf|3`O`&b7c+;zL(*+J!)v6`TeD<UwnePgqJWB9
z$FUcREAT#P3VSV6i3NLtS0gAkRm^jDNhX0=KhoLm`s4f_o~iNg-_6LiCgi1{!*N#+
z*W+X_>NURFYfPCuiN4m4Q|Vbd_Z?!0awnA_H!d6H@^w6#LIrL}GF?gPQg>=AZM4^A
z!sXF!Cwk+QxW=zsQhbtzrMgCx!<Dv}G=2v_N27&XO{l~>NXY#KQ+l@Dti%*!z7U{#
zp$5Cfzpo>`GdVEil=^3o<yR;yU^2%6n}$%%_HVmf8<5tBqho)Q7*m)`AJH&|!h%hv
zrQ?X#?HW%4FLQ#!;ug6bt{%!qWubxTjPA$&Hf56ThJz7Tiq)#pfy!JATt@e>{DHYx
zk7`9BNzbupnszP!GClM33Kf$}yzS<lZ;_@j3Io<DdhBo9SadS!69(Qi038w7u85<t
z+k2y`ZpAh2Nr4qcNfU(}c8B>*QH|K~rs8g=qiDVW+kyI5+lis!uy{|+XZJ$`hxSsf
zC`0#r*Vgr(5io6d4;dSEZ!~t5GmXFn(_=9=gK7VhbGxk`fPFd4D_CH3EX<~*yC)!T
zL*>}f*t0o&IQcFp1JqNh+t$XjD?(7ATGR2_9?j!Z1S#Ldl`^ZOfRn|QphMi_-QOZR
zg7^8ixZxc!0oz&YWB<Yy>X1aUE5p-#NWE7MnwzdRp}GR@nEUvb>s(+I-F0T2zSE@o
z#XCTmFD-BB%|`iC2K<k}oaN8~P+bU=Xmk%B4IwK`LNbZqTZ(eB?WFZORafxfX*<u*
z5fhd3gMyTT@+KVK+!n`MS}hs^@v|o_H2;`Jwxwx%E^6=NmX7t6j_8Qs=TP6*%9jO}
z35L{E&<1Xv%>~R5I7JT*>n#Q0@IK>T#2vTZhYCv*e&@?B$`6%^lSj9ei!G6R*DGcj
zpyG}=_Df0WucmLJaPOze=}!r0x6qhvxui136yC_~x~oikP2<R@8WWF5ZOxBt?bD@^
zX2)XL$6PD@IuEkE{4db+V+ABf+N=Nxngkr5J3hlkQ0teKBsh&s&eD$VlKjd!5V<?0
z$!g-4H4XR>_B3RCeuHM;b!%kH)|WfzoFbNugxtxCziRtCe^Wi2zH{3JUp1oi<~X*~
zHR{m79lt|o<-ZIwyw$xM<a29<AtTq=rm9R#h$z?)fpeN4PWt{J*tdw&2#j%XxmA*%
zW=P1_({r#M?k)FnV}Iz$$Yvb{K0;y8pl91REQEY3xejIdv-IZfF?e*gkA8Z`+Ss~B
zn$_>YLU0Hv9Gsztp`0zW7dnB4wU+i6jC|QYXC3-NvhoFQy1VfBHsq~MuQTyFzoqdV
zt^kv^Q*Ldv7&t;h=j3t7I}URur=t=#BUA^QtNChu*i<kzt9v$yPZKp4F)Hkj?Pr->
z{Q|s!D<ozu8m+ar9cG|rObW9*x`pDILeL<b-pl9k*C8Ss4j3Fb%@@;)cA0~AJ-~_>
z2#8FtEn*y8*L6iObLXsn7<IUlb})@tg~^Xm+hxV48vLHG0xpu`UKb<0sn_yNu>I-T
z?@BZCA+Rat1a$PIlL#PaY{*JxGY-Edibfm{+3MeM7(zW`1mgz=T7O>D!7rlz(s=dQ
z#hGoX0o-6jLd6w?hE+mL-540OQTmL1OQA2uSXhiPc7AakWd?sxB{-fOaxXBx6;_x&
z`B<>zZ6BM$G}^(}t#N61dMWRV%aTtOYnbpQFI^$ZM8JQ7&wFf^a~ln=-z(dW7fZwm
zR;)F~?nbgq+VuYE;!8C?jGRz!Cphm54a_6G-@U$xYO|Xg!RlP+0aU>2?(NtPd$UD3
zu1_M;VWUc5ZutxkY%zdzEWnb|ny@09-P0#QbiZcOQI%zzJ)3(s!(MCVW}y3Y476M#
zzj?>xb8!U-hR$?5sK@CLN>oOeB0&!nwn@7h0Rk~#H?BtAFzzwn59B!|erb02sd_ow
zzKfN-T1)EZhx`EjNDwVNPCSiC9KGn0M$1!E!J<{^ttKi$s*;Fp`IyVIS)J1zG`S+e
zMrM)beyWS&(H#%EE%7z%wm28?SH{<jFWgReT9@Ly=rLV^r<nzd7E1^}v3^vCH($;e
z;q<5Qs(xo*DKfh?+Bn3krH<*=7gblvxzinZgi_KJ&xnT`rZsq>dh_))RqYQQMz`3)
z86Dqh?O+vXFx)`s*(}1#VPRZJ3LN?=(JE8^)W3;XZN$$lfC#@Yen5!{r^RTXn=DVR
z>Sc#@ilHz|aMCT_=@*4M2M;*V)3bN7Z!M;x$B@Tno3BVy0m$HHftI@xBZQVq8+J;g
zFL(tOcCcz)0b4_nVxE#VgsV;zO(kaI=M)Rb*H60Wd9X9;KAvhxi0kg{S_L}pd^|Q6
zsgE1(u<Nqs54Gy=I6S+I@7p|k-agZniy>b%xH@U4V{R!qVpVx2{2?J{s9MaQi9{D+
z81FMZkvna+pO-Z>CG{Ft%NxGo`Nei#S=Z(p+e5RNIn*){m56NGE^RNo>_b-PY-*~3
zUB7{}bL~z@I+nnCG^-Hsk{UiD7aoOZn84y-2uvj5)z63Lm4Q0<{Sy~%+;>)<S<^66
zPdT1L3_4cf0?nuRG&=WFQ11_|`Cs{6x!Xx*cNCjUPS1tALqCV;!r8i)jei8$KJaO^
z%c9-q9o_$!%9XVKU3kdPKh-Q{6c04lWjuRz@bS2>`U}Se&@MXOptmr@`zBx78hsrR
zJ?5+KsM_R1@j)Nnfl?)g_3J-_k=?56vC6c)EJYhRS00PYH$Zu;?RaKY4bo-UN$11B
zCYJRm`~=k-wGqhb^m-ki&WN^zUQX@3{d4zkw{w??Y{9_`!@TA5mala727&Le=8VLG
z*NlKmeq2wtyQM#4=RLoLV&vQREQCf~bCOb4P0>b9wWwg)WhE?`g^6n@87d6vS6y)O
z5S`m$$*Ut<jGS(NO$q9_PIe%~6GyRBTy8*Flh`4bfJQkfCAlBgbvnJI35Et-E)s@O
z)`Z2=*P~3VD@NIH_n<2}t@(kM*&RaLDD{ZIWuFl?`@$o+l>voDvX7}nZcpO4Ul${0
zUi1`)=WhS#IC>@y>$u(PgZ#?L3AwQdk+Hztdvc&&?FaD${YVEYuE~oxvR8%OQK2Z(
zdqS*|WAoywPW$^y4Tb8bJ~t+&uS6%=Cw2V77&SRpd(dFATwP}VqsyqCd`|sQ$nmK`
zTUurAJ7vrJ*3VNpz_QAqWPxg%l}lEFy-+O7^sh(>UDJ89bx?vTjthVo#mk`OLDO)J
zXY>o|3?6w*NACETh~wHh6_Dz2WBi*0ktk~`rkZwC)5(ub>ix6+6xGK`LqmV&<F0^q
zb|2#&l*TRo^r)qsR)6VlTvQISKbi7-yKPeIZQO+L&^eL#O$d4>BhPd8{)-H)?}NZS
z>5LvmSJF2ZjSbzE|0o|g&Kc`z^{dJiSZI+`W~g6_!^QoDUEY4GR*StUa-6;{Q@Iz5
zU5)EhW&c$-Ze3Dg;Y_I&3t9K+;CIfSMGiQTqi3jroKNi9Olki8nOp^JZ0l`WPu<o9
zzE$v8X>Ei9X?#eG@;H=Hs~eu{%N!1l9fGV9Uh(AKu^dEF+y(N0cdC^u8VSf8Gb}8@
ztHt+a)l?9N+!GTj(_p8tA6FqrK)ZKImDdF*Hin`uCsR8s?H~$ZESG7mYT&P??Yb8C
zkp3_OVGc~FJAvdJSv-9cgIVP0T0_S^NSSEi^5_7kc<(WCF<!@!!m2bH_2Sf(Wo}<^
zdw9e9tR56?gZ3AxZaamL%M7)Y!K=RA*B(2J)#OZ2?X_@?Pvr!d(zJ*;L)CMhLx&fM
z0rAsT<!z!5Rfmn;bW|=iS3_yHDYrqa!R5n7k^XNEck<35XLIJ0>G4!H;3^Uw5I&Cr
z^P5z^hK?Bc?lS>Vh8bw)m`lVT0R;Zy-~T!CGym7f&%(_5KXPwoLRKzzw*UG1-z({C
zT#QWruOq+9zmb3OX1&EEMiLrHKw5Mv+GnySBwnvEnrZ6K6k`;b%tVjr&cEJyHX@mp
z%S1ZVwH3$^6NBEp?1!H`r=FY6;|{aq_Ph_R?GKObw7iwSnxezN#t17z`FF&KO7*zE
zA$`KzL5E3z7}HIHhR@#Kmg?;(@gsz)MDFal1W}^Et_8b+!dZa=#)OKGhD?B%7&W6s
zSvy7ggmNTtA)$bRfWr^R5%|DBI0+5i`|z)WMHhnPEw{vZjDW(21neOKFM7SN?P+6G
zp%6dKpKyRGT0p^L?m=*OV;>3GqN^~_&w~3v0|V=E{JD{@gMne8<o{+-@cAqB4vYTq
zND~A}4D9;v_KN-d^poO-jS^o$xwnA|3i9+d0O12eG(E}z({W<Qh7?LNJyVjx{v{z3
zJCO81a__3bg*@EbBRUTz4>zipC&a8kg}x4l;A$@vy~};YDfEYi_XpwB;DnWu27U?d
z*HaMfhH4vVpqJb{Ye?YigMtyuQynn|BE=3R=IoO?l4>K7V(UYTB6`!4#8;*kfrYn&
z9VFJf0|JM^th54&m0X0viG=u#$r|eni|g+|%7n+r7zmS%`olngA1QhO#Jh>&r#y-;
zp9wmH_UnmwiLAtNm{>uFa5w%>S#Q-2X~j9%Fn9iL+d>ybWxOL7p+al`50vPa3SxN`
z&BO&Oy3@pa6+d!ZP!aCWiZ?41WitO_Nb%wLGcd))OE0A&2I<6w-TcpRdrVj&x1tbF
zf-2AtsPH~(Lasf82q}Ij5N-jbvQt^x*|@HfBx^S5zQvC}BpB>BLmm_el$Sq0SIwb-
z-tn!UbsW-WgnyvXPZcod`d)7*`MB#k%V|q!Xk~26twHO>VogA(A+EeLa#Au#z9#Kr
zRIamyWGd+{91+|a@U=%HB_t4O<jAfRkv|RajSMz;>i6M{<Twza3nw%JUw{G<mIaW>
zu5jkOK{zWW5EgN~AB%|;`}K~s1CU9yv*TMUE+UXgp#P7KdU!G^kZ>Vh!XMJ@3kFqK
zVQ4bCV$cBRke|X_(Ii0)*k@6-`NAJD3=*{wBwbzkyhN2(fo5jXE7I+u$)~!toNzZw
zk?$6%b;E>Qf3);R&TQDr$A=fciysj`HMMApT{>c5aM2;jvXq2WU=y-p@1G<nlHU!*
zzxaG&C5=DRd^`(w?G-qqeccg3lMtNcr4)vHO(!iclv~fAln^n@QBbn@KI@6RKBQBO
zon-%@9AWMlc7}wsY$@DCN{|HVJvXfwU6vL_4FS>B-Ghg)+_ACYy{y!Zx@>8AuJxj-
zY_ryd8lYao5^KoC`4$~)|5ZN=T)UQgXwImf)^B3nYpgD+%vQW;W#?~W3)Ht7nDjjM
zQ$IX?FLCnqk!pn*&?P`hq4H1J7i1sNszx3KQzWay2)GTNRQ8$EBAaei--01p-f4lS
zw<hcOWar$-ceU4$b4Yq8yPr%C*p^UvhH(Km6(+aH`>jGgNFP{{q}u?O^lfflMFiZd
zEPKW^2TXmaKt~52CmTl8J6x_0|MfZ-_Z7LdP|tMhIm1EHTW~!ua3p68bS-c}_gp@G
z8gx!!f~EAcRaeV19X6;F1vdIc+U*9FU45CJzQ411GnZAw61ijC+@{F3*)rKnh^B$I
zJcV*S%BQzfL2eZ=v#2x$v5sD2Rl3Rn9@hF+0tDsO0_PwOOmkOMGt3m{Os3`&sS2fS
z2%5mGV_&^ntc4(h)E0R8-C>qyj}MpRz&il!b+NEd!2*E;PNI@5$IUsPqE|FHL~PbX
zU(ENVao^~I<ELxfyJQA3-4rp!$z+_D2Uv98z8P@LNjK@6?TUBYx<d9SG%Fb3kvp)m
zK{a@h;WEK%<D>l&gIp2OOjv<lvHAtK&DFufI~A8N60at)gwRzsI*xMSW$ah3ed1{7
z2J*lhj9EC8F@mfN$U3p913<Pi;bJ&`$Bv!;^H;3>Nt-@vW?BJW))FXM7u5+mv83Jk
zSd&dz>WwyNwfFL?s!+i{7v&jDFFO`(U*FgqkhGd4SWZ>;6Eau&dSbvIUPYS39|kb>
zwT4GIv@f`vz7Y}It6W&pYspx6f~A=Rog&AX6_3DC8dy4<5rD6(no}j&24N-`M(FdL
z-Mj1r2ArQ9>$yAXU}pc7{&<Unhch){7H$Asw0YH4>dt+Z_tb}_If#hOXEf^<4{h1{
z2$AZJhvRbgMoL70_t{yeh}gQ+kn`E^GUoa`;<j^*(yVz5*+fN7DgHK{HQJP<&q3X5
zdK>)gP;_s|qNRI28>#&o&PnlKj3|RaspF3&D=D1zn8AH;@~q)Tvvw_GyDN6YwRH+V
z(OuLq$zyWLN#Y9)iy}wnVuA}Qx_0nrD)8e(IQW_c5P=_JWpxx0H{zt@P48^^`Tagv
z!m#ONK{s%uqSN?o@~crhWf8mA7|x4o5fDFAmVbWIJZ%A)x`R>20dpk%arrRx*VrPN
zB7?oVw|#)O+iAU(;lhC`YU!%gTpRMnnH{tfg_;u2G}RnUy2TXZe&p0?!YA?iWIXth
z;pV_i+|z>Vz7($nePUlAl%mZq6!@EV*4j5r=?#5~F`4$`@w=(*oSProHvn>)(g6`V
zkJv0u_3*B-e8NTiOj?E{(DxdmM*lrQwR3bV|09|+n1$cBPsJ}JH@2u*MgMi3!l|w-
z$tm(BCyy)Z!NJiYL5cpK>K>JZS?_Gi56Gjy^ml{kb@XNWLZ@3rQiVq}`HB9AN*oc{
z^zdHwOsPz#&HO&n$!4!Lvj#t%8^OZ`$0nSg*}ZSht7f&OeEkKZ#@9LY{JT`8>qgOp
z;F*TJ?be867|4{f??Eb>>$F-0Z|?D8$Fz?jml_mN#I<liet%a+O+Ry4N#Hj!)^S?X
z^9+=>EI|}JS#ZK4gNe$kU+6g=VjbhozWw9!ICaiB8+OT#9`_2M%)$9LnR|w-D)2}S
z=47%-g)XYvo(xp13@)#G<ziz}^_{=>%wl=Zt)wO4fd|J@&L~2lI5`;3I{v<!X14kC
zK&ng|2+sM|r;Itp$#wfn&Qk&07AJOV<N2IJ8r1i28$+-S%=7*Sm4-Q`i^+YK5Y<PL
z-Y3?C#ZUs)^09ck*$_+lrgqppS_GpE`I7eqyX$HO1DvCXcg)3<8wEvam=V5Jq_hgy
z#a^hBO7CIR*))JfsuUA7xw!_A5s1#ZWx_vM3vqLP=s4Z(KP6CEd~qo+_Q<Br$HG?C
zm9OOOCcwpft%ode+s--n!Pz+n1?=)<hs!N@LT&TOPJiLOcq$AP-TOv3M!-Tnet}xf
zVv6L*AxmMoszJEQ(ny9so|2dFfGakqeGyjPP~(Cq5v42NFzm62v5j8e*iZ+zcj=R)
zUZ6DbOkxLOoX(_nCzB#zKK8;G{z_6F(rY<;B6vDrcT=XuJgtzY`$TVa(+tg%TB?~l
ztK<Vvg-g4O?u5fS%cYSi^me;b=A5OMe~g0oRibe+$#eN+S#BvhQXX|J<jtq*Rk4~$
zn9e1*0PpKCcg^)vMIFPJH=q_X8JC3WeaoBWoNefYcm53OpbkoG*ku7`n5Dcf*ziOk
zUA}1$$FK9fYdtfs251i>S!WaL(b9#VdWB!NRLqa>-pDMGbh}d;D>M0E^X%2??8L<i
z<5<TUVAitm=}=}zS=n&T+?!_bL_UM0f9j|^nU&ZXAF*h><Ml1)uDSEn91sx<$aBKq
z>Lf!z@=ydw&}L3Gx{!=*O4efARfph~r8#T?QML&G(P!Fl<hFFIuH%bvw?e*f^kU>f
zn;sd@tPZ-PQN=8BSf@uRMQ;Qwvor5lS0}D%3{`D6&AooN>s%@-e$!Q%P0TzWP*3OC
zMe-0Z*IjZ*B}ygOOi4J=`EE=*#6@YMi*F>jL=W+Wd6M)0(TV7H_2!JnUy`XVf2oRI
z{p;mr&$lJR>d2#WF$s}ex7)!|w9GZ0>Uv#c7{%J(e$zLH_Y)O2RR){nAUv1vG&pr&
zN?4+(wSKiLTzCL|5u!Gk5qHp2s1Per`UNVu-R```(9>`JAgE176GyVj^SOe1<I5o-
z*?^s6p8MyLT0NvO1_KtgB5MfXaKxuE8IQnjTMw&0b#s}(Ul-@qeS1D?e_ZysHnpas
zaxguC@33(+qz0p2`Zpvy!<2=o@N+$fR+{wgputEs&dauTS4cW;{P_KS;f*w{y@tYh
z4re&pbFZsHAoGv#BAU*+sHVpd5<6N5w+1Ngda@^f6E#5pMwXCfRyVEOq(_@sK(&kP
znQ>Lv@QLqNm{wR^O^{IUm;|5=6Fp}@hlF1z(yopC9HI=|+#H&li9puUbV$b?McP%|
zgPtddg<)tE)gTI~*-L?JcDJQwF-!gs2#vmSQj8TCEeNg?e5L}Rg+D!E4Kt%&B!O}v
z#fKP&gJ4y;^y=4{BG%AB&mih9mG#F0HqaT53CCbytK0;vBNo(Rzq}?bY&NSpqE%(5
z)TkFrk)CdVrOfju?h&j_BzwK=_Wsgibl2Ww8^&pJ@StBXHLI@2O?($OnbqdkKGBeA
zYxY@>-gKG>+I5$(2HbsMVWv(mINNu5&cVFv#<>M@V|+fni9Nr1U3*m_T%}(PEEauZ
z*o{E>U+yXFmVy(EwVvfA=Ubjm-bQMXEy2NpJM$WS*W9nyQ1_lg6F7x!j4NhM7V=-R
z2_6tPv$)tw#mzzwB>peVIv1TAu7{rMwxxeq48LY=4PFK8p0DaVV_w7P&HgGO@X?ws
zA(ZCbt>kXdvj<uzm%3zaef~}H7z4@k=k=wsLjAHoL9uQTe!ud44l!{7rFHVvc<D@~
z{v>Dw%Ki4T;MbEVNsN2*<~urUx-NmV+=QKXqVN9ZTK3MXZP>>8AY?7`Sj?xNlaE*<
zPM*jZ=j5d<-_owQBfMVCst&W*5T<zE^L<br*Ahb6iyF>WiMy!z0ynxFnR7HLK#<r2
zB>6P5Y#Y|wOC$+AY%P8)dWN=Rp+U_T(BeJn<Z885lpoUI-SVZ@vSMKksJM7pMFg&x
z3T4b=)JF3Ff3Ka{?oo9KXRnCc+7C&E(a(@AjiZ67*W?N<y4T|x7OfVtMrvZu8TJed
zCB_A8OfqbD->v7h{d}MQ^V4oBWcw04xg6E2O*Y?8R;D~_=nYxGE%CO2l3()jGrOb2
z=vwndl)#RgU0`gClkPAaoT8z>TX}<GZ|Dtn{;C`=I^Ah%KL%&?eflVrdrO|JyvRwB
zrmN408$uSKTkKV4FTfjoDw`|nITyR+(Jy?oI;3MwE=W2(v6G#9y8Q0KN+l&(Or=Tx
zySbOTDU7t|19s$_-0H(Z%G=m}`J~W#N@WKxBCaSEA|_{DJr*xOL18d-DIKKA`Bs03
zW8BiwB((N%3s;lYgoz$dKdvlWlQR7xGPmS9aZ2mjp2!n%D3UU8tRd@~W&)G-CH7!G
zIqy77va;6Gyz}#9`KmenV3|_E>(P%#RC=1u%zVeeE2Uv&JYRb714~k3gzg)aTpNwx
z*Lo$QV1Xgn#v&+pEiuKa)PazW1VyfD&JT7{s(p<?PS3B`M6=7HH>UY0HzCFR9pus1
zb^1D25x`up#$^}0F~Dvh)oQP~rY@tLme5FCDJ+@&R8~-@FCfF{vd{{hbJ<xJq!(r0
z7UR*m|4_w@(DfnA4?n7)U~Wh6as6OJflVF%PqTSz%WOu!(3?cJY)H1UXY{RQYZsYS
z#~PH(p)s(e&uGhctE-3$46e?wvPAi1wBTyagx%31jAQUO!Mr5gx_-5LQUD9}YlGo2
zCw7ivhc+_vxZ}*U?!2O6{$59-U~sNx>K75iyfLyhvE0)#8PQb0CUFFgpZi&^9I?Ur
zN^35WH)dHNe;aT`apR($fX<lkaCtq?mf!L$9GMjfbzbKSAPDh+a*(2}e(7kU>@SF#
z#aq|V&KC`t)!Tj39q1@N|NX*|Z-k?p8_@y*SrsiwpZXnYO(o#fA*P?PrbNcWsy<<G
z&m)KE>Lq>NDBlKV)jPBiNaA9s(~S`P4<oeo*_ugSP@+klna$HyQk6(*MwHH#X)KqS
zfcCyGlJt0~;<y?IoE0GPoM*7m^)->3@q~rCcp<|QA!~%yo7L{R8{~T(-Y2F6ip{VW
zlQZ=kgHA)=b7`fGhE>iylIbz}<qKhSwl*+31%^I1k37hm{!Pv_Zl~~KBK|Om-@NKE
zrQm0e#>e1uCQt}=b3LVlZ_7qDF%1eiLF^M{`ytn^0;@4YqFoGdOf9fp_@?v}!Sb(N
zSoC*6D$2#ws!mD=vk9zn{CaIV5xTf_E*2mBmdpM1?p}|Hk?f`Fbbrv=F>{t}@Kp2I
z%WI2nq>Lj=v?^0!g_e#o^G|!)R?xI>@r!I-B(3(KwoSnqI`aXdOI`PM?GHkO4iD0T
zQYmm~$0+m2S7mc04U+i+BB9BJ<)Pfb1bCjyJI%>~7b>&bHV8XaO~)G%n+%i#28nQ5
z@^Csi<!m&JzC~Mc<aNVhN1yzNQOSc7)B)hgEBF=O<MSwJQx#BQb>gH!7qIxjWMyCh
zi#zkeXO5!kSd`rOH!TlKj|(g;*TSz^D!QAz#5aR+os>x_8;UsYPr@7M^(OB7_u;3u
z<KKH(=i1Q?o0V^n6x^RxKDrRpI@V*a&I59{G)M6QYrCo(QBVtXhP5V?Vay&nu2Owo
zePaXlDi8>%U{V4@bh+|%Iei?~C$(pnv+LSW{zbM}<^xK(`Jc$j<u1oPDd^2ggJ$PV
z=R*Y-g+XYM2q9g$hQIADSZQrUph)aFP$DA<z~vD}KV%EB{qKQ~^0S9r$-CY8?>W4`
zW;mYe#5d_2x9sM&=Y7}8?u|K^ssb&uF_B(n_9smb?&$posvSGrGEHgaj@UzvVeDnX
zCGHs<re`couRwNXl&}g400GydeTPTFvWqSJDd@kQI=}aze~NVjMpBl>R{<k226vWQ
zi4{`PA3ofd1f|o?VGv}lsc&{m8JmbswVO_45<FFmf6|X>|CC`vlkIqm+oF+JG`+(*
zT#wvY@l$*-H%i+5wY3X(zvQ!hkAa`5QI%`2R~GzM0<*i(Hz@AdPn_fdKm3iE<Nj5c
z^{xE*yN`a7CRkUkKyF2y%#MZD?#Q#%yTm@D${OLKEo^yfYHXKz9QrSEA6H^yUE_56
ziC0u~|D}t{A{jz!JkmqKZv^w`G8VxRbS}BQ{jREPiDtR8e|zpNsFhi|7|y15>V?_S
z&JN#p>YuC!^1~x{a~n-$XfM!W{0FrARj(J`c|GmzkRQx?s)~#;ua2g$SBw8@CsjX3
zbZM|0@nf5d(h&?Q<$Yn`hAL{iz|kM;ChrXvPDn56dA-b=qdQg6TSixH75<`WiN<EI
zb)*)rXdPO=Q4xRoAXM-*`c&p(@9?eaL5;afs>0(odwGqXBq7H2-`o`!4xP8oFTCj`
z$a=T8=JlCB>Nnm5Tgo;g5WkNiga7fYu2&_{gi(9rp_-aWhE|7xu^7Bsy$q2Pq`Mku
za?FTO!C*z<k2up{<>F_+c0(wrN^cSBc~SDBWn8!1(^vle=LzI_wfp1XMe}+xabjfy
zm%`?_XEm8{>RRaw>T5|<&8vO>6XR#^b}a?XoQ5c5vUkZcaeOf98Y2ZpcdXOgqD$7F
z@3hi2ih!c^h^TWu0`+o045Yvw10b+^Sg+lBs}XXJ&)*Q3<&0Poq1{gsLha}(ETRx7
z=w3#8AlVHy`K^MZ&||uJr4<^rV4(i*@#q6)ZZBF3AdgJD%#jRd(%_!5nY$ck7?s|x
zVe52GHh(1K8m^@EpqLHY07R$=4-e4&AiG}SycO8sj!OSe3eWnV6rO|qf4Dm{Ash35
zi(%QBI9dOHD17qQN()7Dc=%yMlFdlZ%|{QJ2@ZfUI>REGO*plBJlAK3hIGwT#w214
zTX-2!OXV?U`{$$E=jO#{#YyjBs@?A-`{Uo)85}e;qOXP?<m~?^vFT#e1S5q=7yM)l
zMeK^Ir^gQz76wKP2MQXhaCcvHu!sDI;;jU4|Dkx8Ffdr69w1F@prE6^c}F8=e@sYH
zar-9L$47ro6z@JTATW@=L~{meMofRA0{1?s>u6Dxh+L)fLL^Ke+B*#RhoNtu<6om5
zi0MGzKlLy`V78e5Nc^wg_FyUH+~5~cBHW$3z`nlaK=}c|JjOsM3g+Q|wczr1dX0#G
zVwnea6BIf7ek2}$2xR;CZaIsSDDdGydFL28;rlxKfNhGh{Mm8gMuilEG0Rv&U?HXE
z@PoVP-FemJRTnolbcV35r~>=BqJ*JTDLFwH&ZEU5c71L^i2W-efPi`bJvk)CgIt4#
z7|)8Lfz;b)^UNQ(ZAc>YVL=GxF&!`gA;XDz4D5n=g2WP*ga3BG9BAj+3yz-vI|$$i
z+mF}e_3sBJ$JPb{iQs++-36uuzY_X02--&onac>fxPT`g{__EgAXLcr55z<GjU0ec
zO!b3C1->G?Lsiss9^N4H^2Pr-@mmdntmGE4XGGmRaNA~4gKYmj24Lr%0a5xv0t*~2
z^!6Jy^7Z=S5J0{GmLU3G@owT(q9h6D5gG@Ofz7lazFNWjq)`mvMcjXlzDzys@cW5X
z7Un=1K~6?W27U)7;A=|~a87|mgmnNxLPjc{*nJ+ah5uh3pMP?r4rIs(`B=vSgF(6o
z=d=8}b7tHBqfN@v2B3fr{LnzMXT97>K|%f0Yb{Ud>8OpV-0i*Duw#&8MFr(ik(X75
z_TRNfG@u?}Bi2mp(_=KeWNe#VJ8uIjB>n@NiwN?Y?+!TlZnu2KHldo85L(DkkI?+q
z?ywq72@^Nu2L%MQ(7qZ1IYcPvN6-{l$!*rp8jDE~6{dd=Y6Dt?7}1Y27q5-*2e{9?
zFRulRFw&|26Ic?Y5MR>#MZgkB2Ds3DjKKM~1IRAO4?vztSiiS7;isY$$}r##y+@N1
z=v`uahy9fxE1w?z40ID5sfW~@`_uRP&F;ldR#Fb`0CW>A6&!fLXnx+Gwm`$e;m-f(
z4C-lU4|3!4V?zw{ap(DF10w9BuRt})7_;2F{Nf;*)f-Jk-mQ1!%IzC^e_z(fD2k%)
z*~De-94RM`m|If7F}JIsao%CG?St7qvUevU@gw2DTQ>wV|4L`0R4{dQanY)CI_s}p
zoBq4+my*qBBe$<yJwvo69m#b@u7-|V4$tweo~QqpuG4j9^Dx^5`hADi(~Ik|09x7K
z4Lo<PynUjQs;?|h=bh!JkN!06B4faty|PBa6_mOGjlg|DWnL|?k`%4-paa)ZPNYHi
zM8E5TYJj2M-HeRV)D$(2_SgK{QC_{9Z!}o%+9t&jN;uEIG@@(43Db?c*RBJ_m3Gel
znCs%!uptu9Ib;^H@!4K+l+K;iR3?G&SJ|Y2yLy!e4x{r3$Y8AZ5rg}VOTeOm8BZ9D
z=5SzRQ}P_%@-!A&SHd7O#yhtCwCjo@Mg!?P-1@m12}T3c5L0?>Hn#Z&hWp&~0xHEy
z;}g&?gGK$vD95$;{U9MnuyY}FxRll6J&4ve+ROS)jN@jZQ8zKH?~{kVmJho@dPn?J
zS%2_)$Cx(hiHF3Ax@+@nbU)`TeiK)%Lbh6;vS5fQsvZm5qf!BL>U>k{knc8Q@*H!i
z`P4ss|2Reg8$ny?T7`Sg*)+OKaPFhpt;sISL)Lz8I2r)kBIhC6n4e~jXX!$bbkD+b
z^Yrls<5}oPSeIR4M9NCF=qS8-UChkKmhafl`SWe%Qz6LqV+x6+{`;*G3N4I0$>0l4
zI+@~6Wb+ZN^WX~nl$)qrP1EhPvtZJ9xVNnl@>6JXl`Ny*v!I%?6>G(s&61l>720eY
zGnrf4@8|jIcHPTHx(f@#Y<I1nIWrJBbhgNj%?nTpsZZ5*<d~x|go!Ea%p0jJc@3CB
z(X71eQati|a#WjT`dRx*Sc7g-p(X$2c-kSa<}1$CP*W#_%&_+h<0|o`7#`0)eeLJ&
zPZ-Od8#Q9$al9OordSoePg;W52IUcfS0oY4jiB!SzRA+l?<6Vgg!Z`?J@iOvX$1Qr
zf@-yp>sZ{U<L6A<u)}@zCVr<W&#9%^Bma@0(_?&T^rs`8Yc?8MnZ8$!mANMH5EQmn
zxy7Oh_T0&zXkz3p7M<h@1)DeB*e0~+@D~4`U=1k+-QQQufj!VggK!vAZnjicCBFBG
z3U=%I%dOOy!$J5SUi>k)eyeXOHQ$_SwrIc;x{uwgqm_78wc)}tdFn4Nfrs>sF>jk0
zo>-G<J!0GEFN>}!EXSR99@Dp?t8FHO9A0){Utwbxl)VcNy6HXGB=2lj>on*w;mQdr
z;KcM!94j#;p3u=zBIIb@#9s@j0siJq9*sdY4K8Y%8poIH)c4DlLfXnZsmlyq#10hu
zO(q1{7L5DI47So^kU(Hz7kSu%%ettRceGRvbGoD3UWGi-Y;`uEI_E6_tNl-7X{C&W
zq&o3nooN?o!~W7q4^yka<Q3};Y`9g{otsu&Txf(A{x>2}*a%BK)O|BX7GLX<m*o{`
z>Lw-*g9I<k!%yD~Ct6+>4)4;Xz*x%TWs8iHD~U?EbKX=M%#M&MYQP{1eYDg}91dNx
zRO1wdQ8GfgQKLIVgzJ=<*8wpN*8^tN5mga#`)t(52!ykAT%=5q`D1emmGXOR+1aE@
z=)dK5m8BZ>(ajhozwqTS{2I0agNiDF&mO-6$_9(b!Tr=8E2TVT!66;C4L4LzA_#!4
z&m@_aOmy@O4#S_|48}e<t$F6W##M74pcrSSO-rbCg6X$oWNv%d<|nH?N9*d^91bn{
zI_D}Ga;ck}JHO_6?v`q$LVZ{4{M=QWW0`7!i=}>{J9WHiS!D+fEu+0|sqm($%au3c
z*5W98_H`VMVa>1JjT^!C*eG4$zc=DC3<yW~jNxEY^`xE)=esKBvMY_cMs$YIkrwo9
z0Z1tI*N3|p_xgOs=zGi^1?3Z2YoMwik<{6c6>dz3^3Mvpf~}+~Z3Z~<djsUDW%f@h
z@0StJ&2#+i&iTpO#wFWJM4LdouU$q{cV9A@B||5q3nN>o)saPPO-+2!(a654FYpW=
zC4qFdLHte*5eRbW>|3-?07EzHsYXqKot9gtB~rvhu9BZaO$M&#vb{rLz1dB+Ynl8s
z2fCQQv9es|{kYfkUw^Mh<1j4zs6z(8{xa2O&J%FO%E4_m_F`)$)MS5_<4yGT=}{H~
zIeNicCClnKt($RVISCL2LHp%Lb649NonNVVOn~gqFjMe<x9C{LsO>GvTUB+M9Mu+c
zjvAoa{)uQ{J81y)dg0C9`Sa<Pef1}57hE0U^5I#iEu_s6gHxd1v}k4Ec`Ms27|kvD
zsKPTD8`h<cMnAet;i9bhkYM{+jsp}^+Xi@ovRKBV+SL9IaKp02^l<4tqow`4g`pd}
z`kB(yTcNN899(YxKFr7MwxxG!ceYk^(3vjT;mOyw#^jrsxp3a$%fX6uM+{_x-uGen
z9)7=THMk4fswFo?=3-mI8y=5P)fgPe-HYHvFx2C+Vwh(T3&i-Cc{A14c2v+UKgCW{
z?yg)`eUD9KiZ_-GxPyODj09Ob4qnYyjV$|IUpfr%VylUph9QY-xVOl<?})vpIm&*#
zlk@EQ^U45~2}iszP!lw8_^}(DLiid{gaV2qSJxs_BTG7|aChDuJT+DGk>fnjiCT*m
zTSxZSL_4nsJ>@JNAmPfh?Y%oZ-)}2$`9V7L`}b}F+6F9%CP)R*?BVq<jN3a2aRD6x
zT7qM>G0%9)i>ekW`3S|tiVrbg6gCu1EY9Y&AFPS-UulcX5v%ID;lJ^`tqwtBf8n=J
zlwRhlS3xPkvhmQ`RlmYUS_W_aJZrDCH7JFX%IzEGRAOros6ctx%9`!y`h1~8<joil
zZHXd{@{}}R27O-?s43oSbWM%J1yvJgfB;LA5((Lfk)r6|0|1e-kx6ohD(G?v88f*$
z^@qM@!=*L^o?`dnxK^I~DDxeOJWDhtER~)mX*L9nE`e+i%j(7Bb^Vr@brn?3(%NUL
z=k;_1m|9Yg&g|EZaeuEW=)KvOt|qD6+R<XJq$j4Zl3y%5&IfueYLdG8rOg(QRZBbc
zRc1s({b9QUGotshZq_&ElriH>HUVHO@N*N?43C{o&l@=7CsKHKtEU{2N8I~r*j!-F
zd0lroiEp@7>E1pm&Cv|4IvcXa8fN?k`(E2!KZL#YLeb!@p!&zU(je(4m9%l$gLq>n
zh-f#a7in?Q!qa$D-dQcUQ*5M+rH78mb#hz`<P)pV&eh9AQ_xL=H9X!E@);h~y=<|T
zWA_%6-kEvn)n6{TV9J!tR;fbNXxlwm_$*6&DhWOMn`=}Y8zu#+rIb@lhg(I3h}GM1
z^|L$ZDy&nd_u4D!?0^grZos<{44WbdM~qk>1^xL>bD@!d4;cOYW`VWESj2d)TDq2S
zxPB(nsGABShB-sl+?b4ZM%n@uLgX~du~ng0uC{KS??}#MQB||KM*WOeKI6gdeg1Mw
z)K*?un-#yk)MI?jX~FsZAnh5X1-JC1-g(qK!tz<_)zGyMyDS#JVSY=eqNC2H0EvRk
z)333nL_pUf<n)<*9a_Z#%WbWwxj@)DK^6OiR-a3qjOw=Zd8BxJTO$_6BD&k*Jk8(U
zTQsHRx<?E1;qaKDI~Jsb?lsW3Dg&`bGG`TDVf>%I%QU+KT$L@EY=j-`=}XpTQ^>K0
z?5MIgvM%CQ?Y6x>?!y+uek*Shoqrb9Zx}qvfE~s4`mdh&J+m<{#z92VvA=aU@7pWn
zJ#kzPcEM!ltEdzYLseL%fdZ*mI&F6_-W&i!7Im&otU$TBWz!1(fl$vfkGtN)bkxuD
zf~s~x2vG|;VQGYDU$>{M{&L9L_n@DYlNA%7i4iS=wUm*zXS#8BRRFE6bmEc73t5qf
zHa#;0`y$uK1M5U@o2m?O?p<sr@VfC#HkFFf`jsMo5(tMBet@T)9W9HK&Yz`pbSR`F
z+iKQuWB$X>r0t4?5=_h`B`wR$wNY{s1(BHF$+hY*?~ctrc6Khd6|g5V8*}b%xmY20
zr->()e{^}i0A?womVnCKEo%XsMn$Y=SC){80wO?U<W*GU$#`3sUmMDxG<%M&x@fBQ
zbP=af;(}J3Y2?Q6j<${K<ym7-U{m!mO9nqQ4+qq6*L7PCqxi`=EWE06I|ZlJu}|A@
z5sa~ESZ&&)4#k>mBYn9L8)t|02l^SBPvio2(8*lq3LFlcGk6$7MuPAdGA=Wj2)B0^
zs2e_~39U=E@zIOVM<E$EC1aSe5WTb^71;orclS!&+=J&?eu0~7?JpIWxV@LIviXp6
zY05d~4+1~|meykhcCnf9F8e0H)#Wrc=T_y2=RGUMFg|jShNGo3jvSGkA-S;4SKdK`
zS3z!x*h7O=?mh!5WasW)qCQ~z!MUPDO_6S2WZTrI_e%*~4^-8brr4}Gb7!&h+#aAK
z^fF_Gl-I4>>2*Z~NPds^`r=UsB%+&isNWj~{3)L~B18wUv0vvYHGPBpDt5rhdC6mh
zl^oiZmoqu*e!U305ci!wiMoFKwBl4zR<1C}yqon5$`ZaEz)utFKsAf7O+6JOmv>z1
zIrHF_<8M9Z?jNnMdm2M{d5uxt@GWh|bkHkjJ8M2{^WGai9S`<z2B+Vdp8!|1EU>QG
z1Y9hpE<S&=_ia42@VI{Mfz)YTfxX;cqop5SD~EBNLuQN*r)YP$iJQGJjljfyIx9oZ
zN~d5py5!*a%zw~PxGf^*wdJU@R+rL@{fv`C77??^3F*f#bz(R`o_k&_31}nNm5F(F
zhlU<T4FjzGA~uHG*Uc8y`hRDt_2-Gkv6MrwCNlo>?YTc`*sY3eO&^ZbbAUDlMB~76
z1VCOq41^lL6%G&#^v?fkx>f&GMKCz{1+Ix5PLXRYgc|-DSegdnn!9=2Ni@ba*x{6T
z-jG|L`d5fCSNrk$y9B<dKcIYbdZL```<YOa(?Pdq@QGw*Mh5zO%FYI!Cq$;TRk)Eg
zg403){+?;r^o+#FFzc3X#$5b>#P$9EFm?{xqA1I@owjY;wr$(CZQHhO+qTxUZQHoJ
z_A00F@&~e((OQley>m8as{m*x4Z7zzQXyOB@K7Y>_H5&KiQ8_`lfqsp%z3lw%x|0t
zW6qEmXSx)gbs0Zk&tuLJ+i<1^h`MTGojy-XpX9)kj`2`e;p-DE6sqC>FcbK3)kpQ8
zv|BcrZ%d)SNcmN-cw7j5UVIZf8au5QPHW`QZCzDh@yjI*-?!Z399+L>$6t0`p_@%s
znDVpZbLXWDa5aP@Pf|g(_7bG@+AN-=<%onE_f2t<TVTP{a`H!uEFHAWHzA-_ds~v^
z?udRTf%7`Su1~KheIzXdCeyrQoF`CgXpVDJeh3LII9Cy3kj8)OnNIaJ&nJ1DSb3qO
zi>G*P-nwWMR)bWcT)-xHcZqn(_ue;8V&&e{KFkF0t=qw6ulgLEu*aAr+e$dgmpR34
z6&g5QsH-<&@__qRP>wfWQNb<x)iQTqcl68k(4OF`GeCYiWm;Q4h;E`*7-(SuyX*;h
zcs7=oAKp{G`cgaMl>|}07&xLs0_nz^wrA{p2@o3Y;3lLsQ%IPRWY|sLxzcdhdT4!e
zR{VL6UpMtv&t(^5DY63Aooa^<5=T2KzIK8F{u2VscOd$>-iS+&$=;+o4=qqRVqS9e
zirJdc_nv;eLs8QG<I=Yx_@<0Y*JEBdo?`sz<+xEB9TWz372B!;Y^SbsYVV(&*Oxv`
zaM+506M`0-pU@b`eczR4?Z|08gOO#HFsZlU?YK_2>tElva~Up#Zk#S`UG*vl2p~Ex
zF29#gLL9`xy>ZC~KApl}Toy}TV<Xl&6gvUF<%1A=<)aXqPDnLv5lm~FR%7O54^8cM
zIV7R14if0fQq|Mc8x~ul^w$28K|vsrl}%fUmP8vHAMP{mQ@)$Dte1N8oQ&&zL$+x$
z6a8axk@1dbfScX9&DP3vLOK5M3o$FNEZr$DxwFhY@ZBC8nArd!REt|w?@J>2SM{#r
z!@l#qtJ4>e_Rz%HVi+vqNu9NnDvXRfD%8BF@n`B<B1^^TI4tQ^$Kk1;srCn`-bTsX
z*+vqqK$fp%o?VAJM$Ky0)XNRBvU?`?v@?nCGanXq72bYX<#T=MP|vGiD_O&_L{hTc
zm66WGpujjFt6Xt4mQidfGdnnE$h>t+!Z^LkJesL<9V;R}!()qq@sIm+fSK{tG<J-w
zLG2KTX@B@FRF+a+(D!`kEIg0=;;!I!aBU}J?um?q&jyBi^+DMp{)6e&QMGJ?yW47g
z;7LfrflMNLge9rU(@N>~z3jen7;d9ljOHI8{jiHSc0Qu?K}W&wZsfg@)k?G7cyGI1
zx6e~v-thQo{P5V;+0~bx;?q0z<4h)d*NQoMhRU4|S>o7nA5L@dxOF{z3no+cm8zl#
zGk(hRIt4<bg#8!0POlz|-LnsO$eZnMcG!ML$B2Zm_Fax=H)AeI^Vj?vhUI)~)Uixh
zKlAolk}ehgKGBqiAl^w^k+%=9TAm?}K)QOuNd-v!3%v8HBj;^P<xS?)5NrpBrlVx)
zi%r!CMuM*uO^558U#ltUbDvjbSZP3po4lJxuk~}T_UH6`v#Rd+{%2o#mesXp+>v%T
zZ^*3FK5$lw_h^z!79n3X{JA|p$}_E_bG=D2fLsYa`*&BcRAq!XlrN`Nu#RYMa^L!R
z^nJjBggI1`8nV4;zZ^Z6I&Y~V=ST!J0j}q}5C^pjE;p{woR=-_a!^r!dhR1uaDDFR
zx~JGUl4|_Ospn=K*f5isd*%JoP>nsmnEy-DHwmsxW+a?%JS~>K3*2Me!*ytgSpX4k
z=FtyTIX4fQ=ZCg(4?9&{UjD%7bKa#xu$)2mAs4yZ)Iq8WPkTR|S>sEsmiRQ;1D8kd
zgBeXnb%ptVn<6buBk04nZU^E+OZd>y+Wh#v#qW#tJzIQEgS%2T>jYpzSHe=IO!Itp
z%LebDUX^1~C*R>>rBowvnk?rx>_yvmTvzf=@FQ+=qjtDazv%W77rFm4pkVu70R_kZ
zD+c_hu483o|DQbpMmBa%&i_w!U5mS^hW1t$D_rPd9W09be~03B5b}0(C%RkO(Vg-(
zRB3x>N0+bbjrZ-^`S$K^GPx#3eH`m4*O}C0B#O#Kh^$Raz)?E^85fzCm|Z|kZWQ5o
z;=dJX7#8O?HWvVfC1$3kq2$C!4glC(8XLi)FgXD`fbe8afx-0)-WT4W#6;*k7y>{L
zK-}G2G5})t0Gx$Mg_Rw+36KJoZ)l25F0Ox=m|T6`fdtIZ2nm<Gq9dzwJA*?r+ZXb%
zXk1<VslO)BOTZ<tv9&h3xV1C@VFbtlEWJ3q0=ug(I~ITiEuerg0jFWBWdPdB0aOaW
z1!Y8KH9!dpY6{9Kg+>1VVf1Xw&9CnHbwpHDR!A%Wi-aDli~tI_Xa$61MPKI2sSuv$
z_F~Wo>OXJ)md<$Z<-2k+suHToiZGCe^SuEW0CfW3;?Vpzf6>_=-OU30mA&OzT^m|I
zssRHQw<jm#vI7TKR#JCoHvUEKt`5fR&VADRRh^>60jO((QxANt09wF)OJkTPOm^qU
zoeuOve`}%y)Pyo397B93lh}Q=FP-X6y_L9kFa05TqYquk%U<;l1I7vL`~Uaj<nm3e
zuBNDnW?*aq=mZo5BRF;!C>JI-7Y7jBZ*yl~06Dq`5(t234j#_Wd;ELi^uEXTS43y?
z<!8>dnx_|5_t$F_Mqn-;?{D#x-)?II?bPDn;`H53gkWH02>U&}`!JuS8SvXQ2{|bp
z8C_YiIN1k}7a=$Is*8)zo1DF*|IhytGIGiZm;-YIP)4T)VC;<$6*x0ueRI91*)8Oc
zIKN`FTTZSG?iS|HyslQDt&U*+&zijzAT#4vad%`RP7v0_@D3&g=_mi(9r?T8)=YpL
zfEd669-s>|oB7-FZLQ^-k@<(Qe|rDK(%c%HfwjdA#7i>+XpjCrFD{HuARsxId13k~
zKkiTDUIqsM8QQq~P0)Xj=H6d59Vl}XsN<KdpZO#G%3mztu|8JnrB7PcMsW2O0J(sY
zC%C6}=}SQ2_iw}eZ$)~0V^eE)0Zig!FX`7kr!Y3O^~+!6SA7J1AHOc;u?>zbZ1wM?
z*#V{90XU07ivu(JkM?@cjgox$Z37)BQxkX({e(VwGylfqhp+w4JbpC47+PR_Wa?MH
z?=6v~4G0hy7qG^^9B|L{oqp(U^RE!vAOTf58Ev)bgP-|HJp*H6eRTxL1kAw97#xF(
z1EXu0xep$Q85?^b`NFqMHo(qb7AhbR&*bVk52(S}?GY$rlWX8lQx*;mAb1;pX)gkK
zfZ(q`A{lm|;I%*Eig)qe9}$_DIAHJ?KLK;V;1m8m&*}mGU2p9ZKLL8c;K_dCSr6(N
zegedx;RoEO-sTN{0?NSQ9oz(*f#V0bFWKdn{lp#3Ki4j9tPO6Uf99{U@i3qHwEkxJ
z-TvR-+5Hc6H3#Rq7GS?dMi0NoFCvlo`LV%4+3Y*tm(1Pk+gg9$#jiMUlfV6zzX>$-
zR+r~uBNH>wxyEj(fsv`9m-*b-<qLl2n16n-|4sMX#XI`1GX!9O0DA%B_fSnOqkMs>
zld(f&yT-2wB?~6{k0U2qf0zr+ve2doH(#8^92ulS!GHP#_S~KUHMJz|HST}#DkXww
z5tct)n@I`}SDPY|4$}+TbJ+NjG8Qb_xq7qwx^}zm&)|?Nfi7-&r|zqh&#UWdL;;)U
zg%IlAXTm5?@7R~7NRe`uS59;-)(0%g$YH^(SzA0%df>m`BH$fMTfky6C5G_}KrbnR
zdJL--EHS$7<ezLFCWQk_K7iP-k$fy_F6Lx!)EkVdI;s<<_la`e%9&MGIj9(ieJlG-
z)RTr{{;1f6KfWZAaTsny9Z9<rlD|9kE%Ww|$&oWRKc3LOmt|Db5bzgRt_d#|fr#|;
z(-1Gk8`d%Kzt`-ZgFwpJeHz6hzt0rx<76GS*2{YtY;cjhj64Qf-29<1))mdK$=0&I
zUR*|pFT#ZZ(j8<<u+6n8tHaXaR?)i$wjBtW|6O%%Ct2mgvq(J&v#1o<pu0n>me17k
zK=*#_m$N*Ca-CYrHph?lAI{lt>&e7tAi$7tChUG+d-SaJU}ddY7hKhK)gzZ)+t#AU
z^29YT`a!ih1&dyar3<Zn_1pP8SmK+WlN^=m5!p9OMyiO~%rF-GXb%(&SuxkaVz-+u
zs5h92vXq1tRc%)2{Ef-Mi(ctB_a;iuBEy=(vR4zrB0tf5jo=fDc?v-zC;SNHCVFXm
z32*v?eGt(o>^yXfdoD734Tp>zgUh0v4W=4oDer&+8P&p$hLL@$>$@e5Ny2nr-p6E7
z#sB04lz-CH&WMdYgd}4128<`2OiqVlcuQNC>FOrau=V1c4P(H+R)BZBcHtC=&4@%s
z=~SAb+Maz)DR_TwMar?csM#8O-gICqQ|1IozhBKLQ$#pw47#dKJy_%kl!9nTlHu;!
z!H;8U#09`{1kz6xGjEI088l-BOim>|kJPWjR4uU;`de{Z;Lqb~5u5ZTfZ}{rbxmM3
z1UD6TnNVx^x<O*n`TXB-YJcLB7OSg+Iw#LuPD(LW_DRVzliCcGY&@NOrL`O#<o+AA
z$%#FWlowK8K#GZWAks0@-A=n*m8gQ}J1TP&vT02}vVpp0aN(m>VlxT>NTx4DN@S+-
zsk-r=fscq_AmW7~uCDbiMyJM|UciZ|+$>@5b{;Vmm;k<z+jZo+H01oVA6nLAXxfgm
zlI~bZ;SVeVRp-*+cM{gadQsQypsRvs8xpnN{73V5Lv!&3={I2xP-e!%TS(*7*$-{s
zD1p5zSfLh86^Af%Id0H@xPE_+0-i?UQ7tvQ54X=cH3!;zhvjcAVzAu}2}eRh`@Olv
zLW<&V)ZwF2xe303gLjuQx)LIVCu>mX{Te@~QOl?QnkVmks|G48+cbW)Bq%lM`wTaE
zUd<}yBAnyC#?I{+nO|yG0iW6#oO>Mpf47aEr)K>LUHnh)xHRDa*LhN<S-il#ahq-M
z%F3k@8wW8Y%`JAO(w*&BQ&|ffM)<?m&z`=T<s4$#vHGpG)(*I%?WZP@V{>=m3wl^-
z?4v^WRrAX~z?67ei1cz|NsF}Ob;Y{w4rfRD>rL=-km<^7zE57+YYScQV3sv}1NoGE
zK7;SI!g1riAxD|rGReVt?LswaOBm<xDI0Puc?R)pt0>vrNf$+^v==hUj_I(Sb4<uX
z<v42yG&)S_mQN1*oA1O$osn0<lvHlY9au1Quw+bsg?hJ(<xpRNHn_PmD0z4mIrKw7
zXzjoG2%Iw81xn@6KQ-O@A$XAwE#gvHS?U-FMc=s>bWL|Jom*l?QUeg`Fw&r)rg}A}
z9eP*VDROsPBta$l<2f?>qSMGE9NU&9aY`mr&#rRuWd{p)5^6!vxORkf0uDa&Gu*6&
zCxYhov_U1sE^YdLuZ7cWLvtD9sebY@9$uTu;WLw?w`fy71z9syNIz-99+|^#$C`w)
zLvSK%IZ_4^7es$d&5;P5K3M%sZcC1<4tSRg=b6T9vfT5R-LuQF3jGALoxPCK$1wRa
zy7b%ggUwzQ=B4$CG*z^Lc^zictdwrg632#X4cO=+e*O{S=L>@4sEsS#;M7O6OE=ru
zdtunq|GKxm&^giv18@`{YZ5Q7=Rs*w{|op_DqaR|#W@XdBX*eL+e9umSJYY17}0g8
z_7@~3$_<`6qd(2ExUvL^AAu0)syU!?_g4H|x4&_;1tUCUID7-#51R*@lLK#i-xmyK
z<9@Q*&jx)MBMtd+U4jYYDU=BrnVh9Aoa2;A)fQ|PWtw*2(g8w*RjIrt6pw_Z@#;v5
zf#Z+YaIA7WpMfNG_(=s-0u)$PJ_e$~_|Vb*nCMM4q`vNT4Nz!geoF0gtQSG&39H2E
zlG186op{|SJv~)#;Cdp$Ywi{uTvG(ZepOrxj@!2xc+_{){@BIe1lk-?sa}kd#TKkj
zWZD%cW+7q7i!kYItgc<oqUW*+(g+R_`rY32&e!qo!Wq2$_7P3-t-8A-GxhU7V~0(W
z0Q7>6rf4&35IY9!CD2H%AWLKQ<q~6clOS{WhtN>Dr+1Y7C6Z?ktM|4AQX3guywhjx
zl4JL)q6{s>*;YzGT6o@jXH=9Dgze;sJehE)JyA<5|Gd018?UgToF8?@;*Q9Sv7{j=
z?UjMi$7R~^SX*An#HHb}Pncmq;izl8A1Nkevj_Z0Yvg#~Yf?|yh!b?4P?n={5lzN*
zD0@~jl8Em!gEBrtj9E&6vB1TCGYBIJ2m)YICNDJ2rK8q!6OEH!;cD1NY8$pp6t7R!
zdVXSuW{_ovx|!m;*yRZ1(|-v|u_66*QN29%iF|L1)8EF!(+ug`JS3ZXe{!Csggm}t
zY+F)*y_0X*7xA&aF6+JcJ$#kn(X;1AE^S7a>i%^>^Lkf}PW8j$3$bfD;{pj|G1!Zd
z1^En;&lNv{GSsDQg9x^8hxVvQ793~fEcDiR%8-eCRHaMwJ5>xk(V@LY&C$N3ZgAD|
zS|PO4`{?IKG|@TL87vU$F<^=hW8Z=GTLrQ8tW<DG&&JSK%hJeyIH4#P+%6Q7KRqI%
z(eyuptK{`qdg7{JdC=T<S{UsVs~c5EMNNFQG7zo&H=f8pp3z1EaTF#SIe0jiZd5{X
zfQ^r7ehP**_aU5W^Z^`qDuY`u$R(BPxn@dbPa*8MY34g7F0vZF+J3C=?b)*GIZbSA
zghi=c8rEpJI9EFCPuMwQ@d<iIOS1_Zp-fy@Yi-N@x`s{|RWC_5M8M&2Xf2n>guTlt
z3deB0(*LrI7``Z8wO#*m2FN8iLKx~-*&m}f55ha-7cO7-YadhnyzjBetOEcjLb!Wc
zTwYbsEqJp$AsVGChtp|tbPH3=3OW5fCsb8XV%|-96?q0DlAUvenVrdE_hjRM&8@yi
zCz;)&XUYR@F^%cZ$>qXQL+4=y(8gH_13VSdsNuE>1KHE)Z?g?R>ap=@3W;##NX;MV
z_YusIUSY2qMZI8Yr<cwu;IWAhCL0%+Q`XS(_lm6PVo_H1&_<+}3Spw@j{?dEAA-KJ
zA2W`}cM?^`O*?vw+_?g$mSt6-D%h6}OqB8RGi%S-H6?WFB!K(~x3qJp+<io0i~8EI
z4>@};^gV}7-u`K@Sol3Mzh@GqDi0f?7Xx^k6;cJn<&-d@lMORkPKgC=BV}lG^O6aw
zelEm3FS!6u_2fr@`W{YS{`zSyJE6x6eNfjd0)SRP)QjkJ$7=VwoVWKVGXK@mtJKiK
zq`OToIF?eulJw;5bKp=#E<%pg`=C^P+3L~?97Y+CDiIQaws@8`0)VSeVBMOCHLr8>
zw7G-IRg4hQT@*GRqvtP$u2@xWg%ATTdmL)`Vw$|>THQ+sWKJa0G|CY%INQPP<3s~p
ziFUh~739(2$ipjhipVnQWJ1w5iXdu_@hs9QsA3diycX3G(!Nam?YMOxy~Qhna=<oL
zhODoBwj?r@U!TJjjTJHG#<bhYir+6MuhXzHnOpy7E82Cf=JTH4MnQ*X`wZ<^26vFb
z!PGhRsZ|2=5;sQ6pMfPx!tX%JlxtDrg%8n$YqDk5`OU8X?Z<iad0Z1W(~Dm;13GGD
znW529Bg1BQ(1REy6_OhCzCYfhdqm6-Hs6mzr=+-{FQ4zjpQDZ`y|+Z)FwZmR_XG{m
z_M9>8`&f*HwX0(o-ky5e)ZdaMjtFq(z<Y3L`_S?%jEbLSRbjMI^MxhNEpjPf!8~w+
z`FH$17y>cI#5=*wU30<|rkjG(D;dz?TYrWyL_Gy68-sh_-0i37JLp-lKgYFW7Km=k
zjckR9W*-I4I$*QK{Se1pk8shANwuM(nXnzD{<g&oS{5WicJ}Fr+#Lz#SBDkM$znB3
z?f-#Em<HUF(g-SElcCA4GgwL6b3406tmZCU3Vg^5ez?~eC~Em`3M_vvnop7QA8EiS
z{bZKHS)2!YDti{+FA8C#73A<7UOb8^c29fu>9T@=ALrUr>)rnF)&M}##lQ>ev_xY`
zXE1L}a;k0iEg7cEWMzE?b9bU!{N)VZfo3(jb0ucD$+gYhOd8VW=Y7VToUJHs<;Q;X
zAl{=w1~^K&D!khKDTOPtD2&|5p7-#^fT>}sR!J?p_vwcp{QY4Z6pbkMHPo`aaph<l
zBXE}AgH6zHEFE;a!4l~DecizB)54Kdi2UFNe0u05Va|UgayFV3<vrw#Lj+M2B)B(e
z<wLISrD-4+l8||cdx=BdOzLW#EAr*6LX#<QeLTq_^6tV(?`9F1WYYm`c_X?y1%j*e
zH5zUN%rtduis|k++M3zClon+n^HLrsNxmOLuty;9qbt9Bjqkr*uH9`OI|VdR-B#n)
zsZx{pTXGSx*6j&0>r~r5B%Dt<69xX=p-Fzl8PW5zj{%72en}c3pn!&NpgMNxZY_O8
z7fY9zE-*Etf9QoBj|KuKqpf1b=#FLu=RQ%Qrp&$2<dP^O+z~;6;Yq`UzRAAKVXc+7
z1WpOE<Te-UfM=f|&|i-alkSfn2XS-D@rRr4lYw|TV<TF9JIMcuz3VXYgT)4d777AS
zJT2>SBZ9z9KRmIrH|q96%PlX1O>b_*7_l>G{Xo+_wrk65gsH`ffrBv~^!lB=Z7FJ{
z#{G^7vRCEx1$(8tjc9=J7bS3oa9t&8Gsw4~bv6*$EhDncb$+{~BSNppClWp?*41VF
zw#C-)W}lTFOH!RdqC#Vpx>#f~OChV|6CR!&kEjAn`;Z*Ve{|EbRA94UEG%C<m)iHL
zf|mt!wGK7fLsEI?0qr()osB-pyHX4mhfQ;YRFjmNl2!Xj*dU{j4iNJo@nUIa(b;-)
zrUJpAu6{<-R{jm^MRO9NE6&p`hT*Hs;RSXWE!~>sDV7+0Rg8nao52$0@-tHMdv<X{
zcr2WD*#09EB~m%YL_n~C(GnAgmXZQH&S73&>qP9iKs2H{EtYT2)(r3$D1g|7jCqXu
zCROh@Ykkl@#txaa0aI^l#y6}c`#YLlFQgY&(kuA{v$;eoJ8q^n3B~)ZHl6>GNg%d*
z2tWFS$C9j=z7Ir-GNk^mF-2E7)43O7&^J8al?2_O(MCVEb!ZCnQQIfQZ$-BpRf83R
zu3+5RxzPX@ls?o5&V2C7cd{Yp+s~XYqdElC3Hj&ZBGq$9JK)l;of=I>{2yPVhK^7G
z@5z!ohgz(g+dS}>aK1y(mj@RI3n#OgP2q1Ut1;3%B?PCAN8S;JcA<hr_r5I64vI}H
zse90M<cqbs$!=4zX`Y(Sh@!xA9w3FWkrpaPHS603o9bTj-6Oi1v5+iTC|kzQ^t@Sy
zVc+{5<hayZkT&u?Uiz81$nj!sPG{EYF4M5j<SJ1zW8Es;(~z6iXO#wN%MUj5T1y1A
zGnSaduH5govNhXu&en<l+gI8Zq*4t$NRyAflnJz+2XHlYblrtohd<Msr;-G+bGcT)
zS<()9=KY(-(bEx6zKtmlf|*W|tqn(Y^&XY~B_TC!W^^HX_E$lDopA$1F!dt8WdJd~
zsx-^vXm6_T{;h~snMHm>HBd~NGC0s?BgIgQxa73AgJU}1%WSaU(DN3;8lTr%An?vt
ztdJ2r0lP#wZyxhB`6u8EV=vXhi#wlOQ2trt%TNq6B&FlNxHq5?$$*cm7_6Im=K5ke
zLe+=PFyPs?eOvGtOycThq`+sd=IXRMb)6?ldcR7vU8{Gc@s_CaMG~ndA<-Ks?QD*w
zu7fgf`^Bi0#sgQy&%JqK3EE(L_jpD4@wL@5e=GFDd67m^t<qSP(`m;Jj|Bkslrl^w
z**KyI331tQ<6kP5V9*bVIxWi|l$hxpOt&vK3u1>_I}XNI>pQV>KH2!}HeDXU(4&2C
z6n=Hi+*FUX{e%x<*$kV^G!VI@3AJrG>Z}>HtUZHWdV(Yt37V!1dxe)>l_$($q=!_x
z(Q$gdU#Ju9VOH@!d@(@FSCXU%X(kis+ibUr&$SJ~OT3%T%c|T2S4Of9rHfP3erQqY
z(p<lNy(^h)KK+qf+$@BXaS|TOz33@7`y50A*m*b!S%L%yD`}b4;U)P8fPHqz!h*DV
z3o*)mXNV9WiAUHQr-%hvTHv*~Pw4bQF6*)-d%xuYe{&D8&{sZDicb*S0wo^&=+ZXF
zZyJ7X4H!%oC*Jt08Dc3*@N`2xIx!t<?8o>aT_uUWd_w-9s91Qb7Qq%66Qd}t;9HbF
z4j9~l{ft^e{jIV=1#<jsn2$8tzgV!|2A)=F%7ThG=g?-DITo9K%Mtn+RX(*GME7Ag
zWzU5y4I_Q2^1AY5ICp!~^bgOhOZ5VwF`C1!l84rKaBdr=Rh*dg-a_5KLiJ?m%`D9^
z@C?p)Ef`P!5<EXx_J7vLllo`<QasgfHS4iAT8PSH;K7?}=RsGPczm(m!x0v1WC9G}
z&pixuOd63N!#^IG()oAT>nBOO#gK1N<OF@lxNiq+)ZvpL(zG_!{RSEmrJZHY#SE=6
zn~Kx-N4bb<<cD@0lwS})H&P<y_Ket>t8@3FQEpK~84ETlV#l@t0A^3uj_}z3WC{^V
zboSA&a?tL^Uj(;m!7P1{&qZ!aH5h7J<+{<DnX&d@kXaoh`O8<ouF|sSBXwIL2)$KO
z0Wtsm%j*Fz6sg{tAlMFYu54bfAT`=HwI_bb%(Z?hW&_GnZJuArN?KjigJA|^GXLJ_
zU9RSFQ)C;_`w<W$wNF8qdMelf6XpL9?N(+^wMoD68D>vUeU+G1^}xMXC-MCqR7Oy6
zAe#>WOuaEr^fn$<%Hu(x4nsGUJx6qt?b&Aa$UFL~mcJTI6}}tCN1zPuFOaYgc9v7q
zvYADE2YgOx4MQjqd(G70?(JqV_{Sqr$r$`;%2AAo5@$M0z3K5OsF1lZLT*Ev)Tc2G
z*aT#f-*ViE-Xx;B<<}n8!STAWARp+<cQ|Vx#HSg~GvnBc+_)Ev)fN1Vco1gK*hQEI
zAX1vps$$_8bIFi)Z3-GIw(St5Z@peGu1G2ac1=l__Q*n;$#6-|)C)KM#qi9KY!2I(
zXC)qbQ1`bh``7YO_$Q}EV^nGzWu*X<=kax5Jp*ehIv*@A;jhy(CIOkjpx)+@C-3qI
zZYmB*y-b~I(VynYGBdx(L8-@D%Gh$ntz2T(<B*We!n}v#;ferhijtCWv(Avp(vesq
zq8`6$hcfpcz2HC@`(Wm*d@S3GD_$+}QE14NG)i`8g?DH<p?*G^zUtEHPRzyOpBBdK
zB?U}_L6ljT7d^6HoLW4p7^uWk)ty<7YkjBa@D+Db9ggn-g}x|VUN3c%JUi`4dhfI-
z%<aZWr6qqgqW94Ishw{rY$NK!2%lFcH{&=3(dLsy)<&s7NQ-zFxF;3a6^*(RZi`E$
zGqW&|d_X1=)xTYZSa{c^?BIE3N+kO>wh0zTyR0sVpXwUlKbJVy06*6*7;!Fzyo6{<
zJ&Qvuz$cpm%us<Livtg`#>j=TG9%B_w*HHpSDE%@+P5ijJlkXT_!ZizxEektWlA>;
zBDiacQggKw5DCeTbFfn%FebKu^4yTd4{^%3iXXh^<}Iaea@hwOC&A6a>8fd|Hg>>U
zLD0jpBq3P1^ipxK{KMAUx3b59(h;bgZRxnu-ryH`oO4n%+syhwmgRnF&uh2Or36dJ
zTB`L}BR-Tj?4#!{svELufApT{n)2$C)ApuHLV8%(rrM!5k|W&|W8S1|_2b=9TEivN
zbS0@TGFtNIRy}^Xo?G>D+%Q0?F05u|Groh8X|lcV1Ythhwq1t1Uf{m*b+k-5I@A;E
z;KaioWcN^b+t96R)LyNS#qszN525-_G2)fImj~B|Sa6#W3RRmj{6b59C?R8L#Mv*b
zrUvr+59WnpXJNWQ<2Bx(aN(-fu#PTG>0891Y4`+!TKcX!K`*>dzt}28*v~?omU%d~
zVzBnO1q@C{8sDxtFNT!*I=2;da^<H<iqRB-3qeb~5EAQU&p8j<j%Y<ZAbC!>!Ocv|
ziF6kGNd3DtA$}k<);T9C<30U}#izvURcWPDjiG~L5h+12>h(vqrA3$ruVhYKrzP;a
zUsV3CK`FXXvRe)-6bhFN@;|ik8v*A&az)ZrBz@mOGAdKYiFd3-&vn|H2~!`xp*%;f
zsYU3X{3Pkh@)QgpbUqEh<dE?-V(D)nzQn=6)|^x+CVeq^&aY?!;u6}asHD=(q_Xl1
z-~GCyfK|FE-B(t5eX$LG>I5HLJWwCzq{<(h{>2~<#`>x{mvbw@l))I?0&a!lUnS<b
z5dB&fW(h_K?}V0)PKIs(7E^bB;!TWUhRQ?G@)|R)jts5wR&_4#taI9QQ~5`U0m3gu
zv(EAaDEJ2y)X_%{N4JF4qs3WVpD}l>*PF=}EmXdekfZPe<*TnfVlcI7WTnUJamuU!
zgx}YRdDS)aZAo=#bh?&a4XU)_g!_q9nXedOeW~!KE-5}e36e)XQvNQ@-N(8SCREb(
zXl#W{@9Ix4k*>wUd_$JHQ^ZppAP^?65WGIBqWE-7*Hx)nNp*e>-1}8sMNMuf7^yPf
zly^e;TN9|dppvh$LQRQv%aZC3>3u$(1?yI@$Jph$ZyH|R5)n(zzT!u{62eT^8bkBw
zm{Q2ZiG^$#=}lo0NS2G!0Xsq6;api91|8i3Be~LR#GsS!brDguQ=FNPOKZXiPYnBY
z<vlGpU^4C;?z`96b0*@Xh@FbKA0?}iRrnGRX63lG6m=kFoaiH@a6;%tdh!)2M2IQj
zGVkLT+o#9h1nK{pv~)tF@73zRa6-0hAelU1RO}cYatUUKbWPgMh5nV)|0#{UVO8bs
zx|&&qAfG5-EdXBOvO+?dA~IoluN4!86w0nya+?A030~5^QObtMrQvaKl$pD{Hngea
zb;aykPhMCIuhWvE1^|WV&fa}2I|h^}h$kJR!z?5J<1zhWNGGZlN+#7qsd(*!Nn42|
z_7Hh5Sannq8ffawjTh&-c_EP&w*@|rIr8cBy>`1e+Pn}RiNDV@-kK{r%)2Cv9y4)K
z+BqjAT80<ijisCZcIz_p@yB}0SL%xexaRU`4B&cJ7iJ|v&T_@_vCpEbnWdy=&RnXV
zTOw*xb6Kkk?GrDIi`Gt5=1F?GR=fuIUa?>dzJa)sFTp7_F<5Yg3W?HRv6>sB^ztoq
zrcqNYKrM}UzruM;wW1r0DEs(|WK%`2$B4|v-PBvCYdOBy#D~SVEy<y2U%xn(J0&_Q
z0=_&Wghmc8#A|X!M=uMP-xjVN>0K85Z9^0<z@@%{d<DoOk=IrUab6c1syp+aGGhJ}
z$>e=^gm5DMbVOW}2m0Bo+S({4r;(8Z#RMX^mMe>d)`<UAS<{&HpRF+UXWr7zPM0pJ
z45LFUyWCydm-ggFB?na1GsSSAQnl+_Moqhz>Yqg35I#MX)<>MGHnH^5ZDQ=Q>835L
ze$x@4d$;otoyFFJ1zhAhsl1l?+|M44pmTAoHXJQ{Ix_Ez@e|uA0GJASY~RaXU^dhk
z^PDVl^#B`kHLPc&fTRJzvB0dh42zQH1~+U4>Mxj<wI^~{_5^#?i_95eM%=Ra35sq#
z`L&D;heofPl02-+-?sLRnU4bJzb^RWk!9I!?gL?iayNWI;&V=pKGO7San1MO@1-IZ
zJtu2eCsKMZgdT=?Yr%}zu2$;iLv{GFg3g#-rO6xQd-*3qk~w#BrYx#&R#3COyD24B
z#P52c&MIda<3tGfxoC6D!UaA|3SWOKzl8Roi{Df4?ySAgO~+<T?B`xF(WxA5OvG=~
z70)hnk!BS|aD<`{iqK+}fR7Rz`~JshA>$WPVOuu@Xu1fyB;xfSf#e4)Z@LcBaby9<
zwKVe~u&_z-Od*2!XwF}`4(5gHz+HFTBqI|ch13u?z4h%(g5*N{hJYp?;TY(AkD8yJ
zpJn2}@VwJjZmzeHB=H#jm)W_(Q;SkgK8S*YbO!sFWQrsGijA~6OcRM2MO#O5Dpv_h
z3hP1Kf{s5$k+~UOrGWLr4W%(;`5$t!6A{*a3#~QU6p91{2p|0BfPB4J<zhH?am!QC
zMUb3V7mX^K9M2#uG$_4g_Py7SfVg>h3(fn>eiupSP_zDWPPaa-qh9xTG788&9a;H$
z825~#PT#`Es(fSRyWLvI%4C2duI<*l$h+35EyDnENs-&p-pw5XhU8k}$}$W$>%r~x
zkoQT*{myQ#MZmmnKItF+{<2Mkz0{A#_wF1q8Ap`{76`K`Md9qLxfWdUF#W^2g`B?4
zL7{_*XPcaB_kELe8O%pvo%I7Awmg+@bEAb_JuHGW+b7AI(FuhN^|mKjN5nEZu~<I4
zzo0E0Dt2@W-DOCl<p*+7zrm#|Pl}>2bc6!4Bx|dYC|Zc$A1BY=<KcE07A>mQd9@)D
z&t=cQcHRv9;@CaAf)wvQXEzNV4$l5AbyL7D%Q{5!$IlxXckR(8Wf#u21JsI9K<2_f
z*LNOUgfl2N{6H!d@4ivL37AU7o#JS8RpIAS68k{EXY-~V5H8M>?Fq3GaO*lEM^(4z
zHNMiN>M-iwzbi5sq4ye`Wu@@UfG_zWJfEm0@zg6RFXAxM6;cXQHEpS`hs#X4LzMOE
z<L&2mwM6Q4vl;BIY}DHg^VJn`!ZxtKTW2Y2wP8mLQFMfrx0a1ZDTdX8AUvNr8$=w^
zi6{E2E(|=BQqNPSvjgrxnx34Uq<Rx_=OYQ#XJEQqgDlw$>MF6?m}$xulDrz^eNse8
z4l&r;q>#=nX(q%J!6PvmVZ8UlN@hinT_!GHoZHj$Ym)DYJ>kO8c4u4@xsn3;f$Er+
zkf0*Bwd{!w(i=PJhc{UnK@16|lKDE)_*vuX9d4t>2-kA+%lm(_JjoEc5P@S}um&*t
zT(%3a%)Z1J@N~;+<y*_XdTC3*;UyC<3`AeuJ;-^pHUn?1Uz!9JLl6^`ib&f_eX6PI
zw>jNh7G!>E1mpIMBsHa<BD|G_zQvte{~5DIZ9C2q8t7AS5xS>Z=rg1dY<8-5HQ%n6
z;ZB^Y-*nr$Kz1p_;U{%UITDYk5+Z`F7Yb)k=PxMytQ{GE=o8IS);dfO#@p2cD?*ps
ztStZ^G4SkVtP8OH0hTDE>4SH^xfTDZ5$Yf#QQ7Bs$~?<U|LthfzK!Mr<Zl6K89`|1
zPM(GnkRbL&R_{cUybN4$QqylMfpxs6*mB>&X6Q2eXr~L&>LG^ZS#7pG%-2}e-Gt;3
z7J82QfdK)iyo6#cxklk25?ApJX!t#k6Q9Bp`ZQM1lJm6ujwB|%DWk48!d?A`8T}r?
zAq4Lyv6SyidL#>FqS970^CvO8K)19v>Q%I3JC~FpwTPs48YPDSYqGC)9~^I$t7DMW
z?074W3%@#GPrq*M)N(BF$%Q0U6#h@!1u6Y=swR)E3-LB5hFm9lB232PrU~obg~ga*
zLPqW#`Rv0MxVR*<zEU4~yy-nl8^*|o7geLF5WQI@h<ZSwlwp<-K4%rN)Cvgd!h9vj
zijJ7&YLu>#1Ma5w9Tg9cNQvFeUKu51OhanFdBnR#duG7Z*d%{FOIn0Q@eeH>NKeH!
zZ$U<&)!8qRh~7o1qKYUt19VOe^weYfaQdxf-H$FNsUcjfP3W9*8w!tRh3GF(t2>H(
z`s42A0i<QQc3NXAe}3KN2;(w?ztjhZk5w>dso7~nqbuh$<UMFFs-Z$=_f(U}GyuTT
z{oJx+T}<pxpat1%TPLm|0!gg_*pUJ=zM9P{2yjw-5raVJEG3pG(Z@O<5NslL#n(^d
z@?^x^v#2&9ID;qF%Z`YU&t%F*5y?BDO{&E0$jz4tYR$1lB#>fF(2Xr1=3lPzSzC%M
z>$=um7m+u!UND0-P-l-@(xY`AZEG?5VB0h(0(TVCo4eQY*8?>tbUH4@-|BZ6HL93#
zIpT@WPodjYjzo9(Zn|L&y*LNUp{E0w)r=Q0LY>=`a192aiVu$cHaRF1D3YZ)*Z{qD
z7z+MgYq%|%W{`6(OB?vt74d;-CWoA|_<aC)@Rm<}C}||uVEV1=KnAp6Q2k3Uam@_H
zl+ZzA_nA^(;E_4BJ*~PfLV(#5<#yGtf`G}Z2Tt7Q4t&lXD<TpeVPxkJ`|Xue*r?9T
z=<qE3-1F1w$YxKCy=_v!*c5e67WQ{eJ%=whj(z58IX4O;Jm#CwgR7X`mJ#ze-L+Ix
ziUI`XbM7yR1tdswP94r3sPwBWp4XYpwwKY3143keF_R#dRDWuS;62xQ535~%tN*FB
zz?6vVd3{C8NzS$0L|=qLIB^79Rvfx(l^YGqr7!f2n~XH)a+`KLUxN)Z=z!KBzMR7v
zxbD-e7S8tE=)1#H%I{{{km*dW9g7wDExyO=^00sOKcbD&N=TQ%J;_>ib$x!5Sb>(V
zOit&M9-aY=c2!y$tP8zekH`U*qT4=PDi(ESbY$m?4@>!$t)uf{Z}zV3I7IDVEym(~
zl9!ENXveZ{!o<fa;bHPDF;RjjqfuJz7cB3_{<dn`j)ure33gf`=WeToab&X%&&gxJ
z()5$X)oH?VC}$f`9A#Z)(NDnmHNG_Lk&Xj-`!quQJ___%hTL&4Pi-zLH_EYt3UKhi
zS@@PgHbe9q)vno`N*+--dLO7;XtRV)S@L^@4VNQgALHH2E#Bz|2w0p(Dzdw|8rApK
z937!_5&K#_g?tUny|eq+e`#V4%>N0a&Evtm+s?F(7wsNBfarRHa&UuYzuw@7(PLY2
z$ntd_oq}FtJJX^VM5+NDV}A87N^W1gdt*T+c8wXo6DGmS%b>c_+)Ja6{Zc!!l?tk@
z>juK{GgGc?X!Rg<g}JLD<rC3#(nAS6$rl7gk}zpXm1*<*pdwdai%i4@)pFs62yC!!
zPR#0@iVkDyDtxK_{Z>U>a8w!@(SkMF+(18spNM3n+)xKgGP-k=;rsF%DfVq=(!_d|
zwK0?VHL;H0D4*Y&W^keo6i8mx#iMiLz$Jzh;+lcX78@t$!Pj1ZUcP9q3g3lAv_~}R
zhC=<G3)w@h)4G|pdwyMDP%{~aF=Ezu1g_QKv|eu<cM^+EqVQZvFp^zl;xGqi@Ckv9
zmlb0OK}zF$vs)d;PA#+g1z5FMaBcJI)F;k~y%LU-UpRZJqe~|3UO5-uSTTM<v8W6`
zvozU%<vGTiOB>6$1ZYy}ee||pmc(ENKXbuJPI@__U7f<N$7!xNyA^W-7ZoW%ahVe^
z?!lx8uF3M38I48@|ISp`sFfyIB2y+fOsRwfBNTE#OaO2p&}-G&zuag3Nc|*2;mz;-
z1sU=~8M6(LUt{dk{3k)92ebah5<(-suYC^V+`a?Q@6REh%0Ocy41~RSm*bn_zzDCF
z`5GHv>@<aARX~)~TVVmwz@Xzijht?s^<V(3+p~fLlMm<-IG&^D7vy=Gx!KnnK>9)@
zsxw;XA}Kvx-GBDVBxf#W3S}<=0N*WUEI*uv1Od!JVp}CmjeKLOIb*0|j?__E9O5nX
z1#UEtNUOsc;nhKSxzncCh+EpLM2LtAt0uPdjUSk=s!Zzfd9{v=+9v|Kig=h}9Je$!
z$a$dLd6f<DDPdHzx4Bl*-LS%Y?DIivm@BUHpqL--OrJuJqoP5_2E9qx&FN0&^rxkQ
zjC@214rTfBJ^TPyj+{8oru5n;cGpzF=`24!D80tp`?2>mPf0m_7A#B4Rm~ab47io0
zfY}Q)(8`Upv`_j~nzj02MvXBF%Ah;NbE6ZCe@ScZ)j`^YN0l&Z3Q1yv%9;nt(CTR3
zp5zbAca;Hk$QYgHye9b?swRIvbOISWr}cyI-DOx`Db3d+JxoB5&@VXJ*mYi$tWLm3
zV2Ewd_AL14tdtJktrhLFA{r}~?*?^zCdb<dp-1Pg59E`KJ|(GtbmAH|_8+p!rAtcP
zlQ*s;t<R}iufq4bZR!I|JPmt}-Gv$Ne8E*)?k(5smpooal(2?Q_;}m6wGdr>+zXWr
zbmzkd+*SB6SNgPb@?2>kxWhlboEsIL4F-5D%y-_Niqn#hxG%;KeiH}*vrwTfF#h`e
zSyDU?l^ick5n)#*;{?LPR?6y|<;n8f9X1#@1gV@DJuO&2j*7xXZeCeJ`yHTynyBGn
zQJ!rfp)>+q*P@$(Hc;Xir@@;?6KgNRb=nMplo3tM?-rR?;F|0V)ii??E)2@W#h)J`
z{sQdIUBG?}OVk_16HIQwIKS#ulHVoW3Yw|GQM~%oE?78bO?6$Np$UU6rpli;!TmcA
zdJVbU+h)uHSKkz>^NrZC$+Z6j>;kn&gyQ6@eb$W~?C1le>9~H~$rDG$@Z9U3pX>Hy
zy|*IiT-7B+;kJn&JV?!cgUawheG6LQU@K{f5GfzaMavQa_o07Dty0jXktDLEU7QYg
zZDDTLF^+Rs7H4jQtFrd@)wBSg$qQ-<HM3fl^dMhdE$kedFc_wiY<RBW6~{JOYK%Zu
z`fCd3L{Ym|)sWsE^St14Xj-F9Tyhzbb~Bdy`H<GX`C{XQ_{F;3<Lik2YK_FQJD5b3
zofME_q46Db$r)6X^W<B6rwAGqjNz4qJ2sTLn5JUY7#*rmREqY>fgX#~#a1$pP=>cY
zYNh}+Dqn9>h4@x)b*A4+@vG?#>Z8Y}Hx_4%BRsO-W8fKjI{2QX$fW}DK&%t3_c^ul
z#9VW31o8c!*+XvpZ8P~Gf02A~vL2wkpG{dpr*l+-g_rq1wx(J>m-VA(2b-rD+)oEC
z(Bje3Q)ED(Huct!fy^vf>=vVzD4J=ai5Ya%)H&{`@mgXvEMhp$e*6%2<vci(r^GS}
zr+i{n$9neR3d;6(Un1(a-Kmcan<U$Wa{||<0&orChQve_G07@IQLVAbMvQq>cTnWx
zG1e8<DhRRlNuBe?AGx^~5}AR|lTzyBc?-rA*gGrgDd-P#ToXmmz0H$Zb#BVQUP2@_
zsA&5^TcK$gUYleH0OQ9&Sm!YD)=2r0$ebsRsG-nn?=_RoB(Ygftk0HgA>6UeqfLs(
z{2#SkOgjTKB0v5zJmzcTcX}lDUkxqyZW}o8Tn_tt3GI>kBZ-?`*AJdKie(aXihUpJ
z5&cz`oZ3j^0%OruQQ+uOueq!NRzi$^Xrw|CK_XSWFPA><2xN_b8uK+M;7C3gvdY^Y
zYMmi-=({94{FVDEhk{)9vd?cdu3PknZ|}DUVcNm1gkcDfEavWsqxzp#`MmqI`V)%o
z9~R>xP<YR+_ZaTsr1%x#e((x%{(<;#AeFWFhCyXyjK}Q2Ox)LeAg-S(Z_6=f4SG;^
ziB%UQY~r?v?ZP!&2d*~ASY3Z!=|v#}p5j8L3nHLGmBdeH|4P%v`G4IXFxK2k07c-C
zc%~=dRLu+J&%$o-3ZQQ%&_nzfV!3rhd4#^4I%^B!Kw`FZ8us>#3iIuzY137ZKQKH!
zr&lHl4Mk|=`kU=BM4TxK^raY^1`OA&`7+K^;)Ty@_{goZ$%Il6iKqG`m;Mf|z7*Kv
z5;(a-E1DbV-IA&PYI*`wH$R-REcTrq3!$GMH9D|a-`n*>3(`v`|D;_${VWtZv+PJ*
zF3|kv*r`L<j@t!Jm%*p(iy(yv`l2y?<cq!AEx~wo^QHPVeueC<*_X?V{7h%U)Zu&4
z6Ibt+5Q_;}ZO6qS#v;<_4QTs44-;ehOmhfUL{@%Z-=lokCTEhecjo?;66dIodVKy&
z7qz5gr`RIRJrTvWP+$MYg9CTx8;a)>)O7m7ocU@wGPC7}2VNQGUTD7Qf)}QRfn+{$
z;UcQ@+k<{*gdtK2)%`SdKyL|8+EzInzJ1;mxyRqFvd1K2u$ZskVShjR8Ih-6450qE
zptW#%?ib|0HK!;v{g()Thz6#5H%Eq{CwdSfa>|#Q3J-U=U$n?hTJUSMG}XmbK4a<2
zJmv9{EpY+hY>j+lcY@o`YwqPY9T?LYx}I%uoLDE#97N$yW%M=2zyI{R%q)K{{AD3*
z^JJs7;^TYiVE7-DcF6b()#x0AD+<LKFmol=))w4z+VmmY8_z!i8XfmvxIl~(q}Y8?
zvEX4Blv;kwZ(dJ5jirxAg@)ye1(H=C{rQ{WDVo_i0kX!Ac<uJ4#v%OeD#$K0QX6!W
zDCg#EucsL1cc!DpEh;T`b9hBKef0@(B^*@mL(-*pCp!>3dERrFo><)Fa|({9kEx@`
zN@R%x^Fsa<QNIZBEJM0UYef+*6IXBsMnh-!N+<7w&lo=yWERcXCvclXmYAbCm^|MZ
zvzSXE>c&neJ~`#{9oV|`@qbX@gGW5M@>LSvI6S|Gv0()m(g>M@gM>h$i=1X2$T_$v
zhG$SFG|1~k5+F89!8gj~*9z%Hy;8K**Wk!g6gk@JAlL<4QOJzkLyg8kx8+k9{M@(X
z*;$f42G_XV&hfQRpIhsTAmy|(8-)^7{bVs(`V`_pbkanlyk%jl0EJ#y;O#EfXG;m>
zRL4lD<zmdS-m1ep)B^<!&|jedQe31Fx>sz;iE^$Du`rF5p1_CVhNA*Lm4g6JB8c?_
zZ=hj8R|Ew_rO4|;X0h@!n08a%GF36|m_MPrIDDp9J4o2%0ifDbAE{!V&M^;Se3s}w
zu=-wOE70|DsDb2nHuS?el-tmXL<gib)ZibK1NNwaT5FYG&<|lrHtDCIpBeeL6~H0k
zLdy)KtOGN--1!I#>MZ`snp5+owl;jrv!SWg8x{9HxlY1kFM+dtZdb9vvzEFIZyDJk
zmGC2s3+r?lm&8S7AJ<}$A5#r;`twbv59ai@_axb5Sg6h%;G^WT%=R%|R>zBpfMlr|
zndwjRjd_IE{IYWe7qJQQ%z$cj@$9<28DKyDxxm=#76EJ-e(OSCj6ajrILxa>GwEdk
zb3}T`Z;(;)`VR<Me02LpbKXGMbP}|owX@omMqZgHQn`8ic1mV`Qa9w?>h*gh-f`f#
zb^CXxZYFPrls!oUXCE~R#O||kRP51%KFeJl!9{RXGuua533gw}N^wc>(w)Mspl#~I
zH%^p7vc`K#^UJE591xsS6v9O*)QOchoW=+kBMqz*j$@}vihCJ+rk~-Q@@_v%u_<*s
zpUk|@bx+YCtq=Mfjzp;W<^p|Z8`9HE10@<$Y}Nfri}rn%Nhg5e-q^TB?2s9pU!WRx
z^)|o0Evsh-Da$IKZ~>Q#G?P{WEan&Z-?)2@1+S+Db2o8rzWgjjsa=ZMOgckx<AR!Q
zs`WhmwA)K;Z$Ax#S>p7pCbm19w)~H;boH&O8JrZ(I2IW2E~Kj5X+?F9Lgu$YSMjX(
z7U?w?@Y+z{!yxfhlIS#7$UC8QwiHjPb4kC0csq=tnZ<;f;6gnjCm(_9cg=8gS@`}*
z4wMy-im2*cpjH%Vh_qM$34ct2^uIW}wEEAL+sC)~>`q&zcGJ41dQ+BK5KYT(SuYu)
z{AEDsf?^3)o!bvGrlf*s=j=9hj|&G%sn!rf-#ZD|BpZqT^-Wsk);s|)TzrK2xi?j0
zLD`NV9d6&TIStz2Zn`KtRE820Uh`*28EbTl-l3$}#35izju*#d+z?h-e&50oJA>l2
zJWPp@X4*LIyl}$7jAx6LFFMRF4Bx`c@{oYAiUP1hjOkGSpHtu$3m>fzzTWMV4?s|V
z{h>7C3tJe}F*P@q14_{U9{rutk}ZiopHThr0>j6kuj8L&#Am;(a#5HNCM?+LR=g1u
zsF~j|IE}{>bE+fm#NiOU%1McuL}y6t+xn%Ub|rB)kCE-%INqUBA`Dt{%4{3cc5bH_
zS)k$nW9%G)bWx%;Tefk^wr%^AZQHhO+c;(0wr$&X)nDv8>eque=t)Lo>{&)mGWWOF
zlDD&o%vDAE+qug@Pa~mLC^<k2{Zqn;D>R<nw2WJyHVnf7bJm-U>_HcYO8ulPCgd2J
zrdtVF^0?o%1448JiltdU>ki>tM(OwP&u%%z@WZY$1P$1a^39&TX?F$qa*6n3!N4A+
zYf|WZC#+%sQ7)oO%Mgj)DoG^!xx&;V(Ht$uVI_%-a(Yj+N;8zq?X$o>m5_9w&pw{8
z>BOb$jBHy3F^H~(s`%){^o@`g#p^p|KER>4wO16eJLPZR&UClL9|qLE@!yP4=uW{(
z(yY--W47rPmRAk!Cj-$(E4J3EH&nZ%^0R&8a+3U}f9&muHHu2_R@mq_Veu^(vJ%{S
zG8IY}4GuLM4Cj4=e`7WM49B=B^CMAqr%2FINlNV2!IL94VpDQ3TeIl(7g2<LN%T(5
z2CQ>xRZC%KyR_dHFEukKI%P{-=aQ`nF<Lp+vdcsR+O?X<qOM>CR2D$%tAiDhf!(uI
zcD=wj>2<vGWJgbF(HHAvPYey89<_CNgns9DU!Q{L&l&%D=hzPj*mF^}L9vlTD%wv-
zS4cjeD~;M;Bu1TAoy=$vbmLZ_s$7EV60>Oa&$%(%lVC*X(wbVM(H#Y^fEu8J6AeIl
zt@B_uc=eT1!D{FE^{1y>EB=%p@BW2QR6&?!bGDAkp;0)m$ge9Jd*{pk*?c5A%?iJN
zf*xUPcE|j>1ujBN9@a?_25ywHh;{~1H9Z>&=N@54kNyvCsKYY~ic?4V8LaTo8&`y6
z$<8)@l$YhG1@a*)ZR6?;t@Me<E68$>A`V@ZFzyFV-GSgcEO)saI@Ip5UA(h07K+sx
z<`B~ybgkQuH4Y;P<W~(P;7tl1>K$^bHoJrNb8W(wRd(W)q*1Lw9)WCuAV-<)MjY>_
zWtZ+LbG-BDC1~ihls3b%Lv-9s;2(qgo^s>4D({!9sb<8HWgdLPa`N#=#yZw7yt2Ej
zka^HKXaGOHtdWmKUn|ny5Ko7fp>ggJ2dXV&4p`98IV$YIogV*qcYSMb_zx_`hG2Lp
zs}bzc(43pkx!GUvciZ1l)0Q#&p^P6s{|MGU^MsRPA-%)uSpTDVRf}Q5**#j~jlkj_
z9RDLUOO&G>Q_W@~(v9h#Sg2an^oAj<9_@+RE-s(9<Vv`2{3RmE*Cg~AnsG;uQV<n)
zZck4~8#hY_EDf5A&lHJ}$HxeJ2t0X5i%(&9oyC(C^&DW(bolF@V9TrAQ~h9>CkUKE
z+Y02wux{EOSObv$hmL{fxbGIEq~2m^XR(KoK4>&v(fw{={_D&*3Yo^Sx!U=Lvq)L^
z1`SK4c|xYP162-(p&1b4qK{r5S>3o!(&Zo^i<fOhRP|J@%Pnp|=Z_keb!RMZ-I*-M
zA^?AAmMwIZ!skgQ4$Y&&^QX8WE4EavLIFNo`P0-JNqfnbteL%%H~LC)&wY2dJ8R<^
zn3^c`oAAl*`E(u)oxtkEZ+->tLF2R-FNRqRZ4#n|k-fCkPTN`|r(8%#%#CP_SYKR!
zMG*9iMo%l`&V8bkI5Y^si$%Mv3|X}y)635Wx*Zv?I@`V&Ij#xgqPdn#a-XsmQw&ql
zKg8T}o%B<f%%DT^)5Di{v<UMD;hx7XEbvGbj61UX(D+aGjhD7tagO$tW|>qoT3u_{
z)BfZIAS~0UUd$-SCc>AOny4EG^@U3`)NpO`A?+3!pW%d;F}@A#!#E_mHTPSwFiKZH
zT~8eI@}=buc-2$3ae-=K#`i%@Xw-0m2>8ogMg8nl$Yq7Nq$ZKuS{sswYupA(-stt+
z8rG$qxThUGo{cM`V+c_h+P?(tR0ee`)S~tv9>=8hHi%wxCI204*=V)^wIa|lAZucW
zd27Kxw&_i-J{w6fC*#g*OT`Jlk^D3V&`p=>=bXci>PyMN+<AOa7#%=50C=u_$>oRP
z4_@)seH}zUKXy==4mW^_1Tkl0shrX4kBC@ewUweb5)J{9npyIuwoIMl_S8(zE==Em
zSQI?Y#^mP`cKH40*;z_5>}uf2(I&N~qE%^Skz701^o=0yP37||SA@3?7l%3xukh<f
zrFsX=rKr`zomGvD-ELkT1=K*<7GcOnAq++iN-w`9HsrcO4?<+gvFEFnFN0rXG8}dG
zE=z=<SIBvfP-!pv-@fWGXwck8ziFUfSmq_1d2M*Khq3fIH`D_qe(gVf5>04*UpZ&n
zbSEM50HZL!Zyq5N;(7~y;L7Q4_Vz?+n+}Htb?3WX9cGdM_a?U-!=nO2KB9>V^}(Fl
z;6X&&v~<QS&<=q2glj!J-h~5a@x~e;ceLRYaby)Nr$&FP;8!0=1*qzGp?2nQU*~BR
z$9&rV3k3hkZWeq1_5B#f4^&0ybArjC)<q-5*k>A^B^)x!0K6hn?8-igJgP8_Ct@o(
zkhc>7`C4U4Uhz^%qN6qXrMbhKFr3GUVxX4ye+5Hu{%>Fic4n^sAy;#9F#TVf<p1lB
z<l<!d|Mo|=sJp79?DUpOcanwQ-X_UiKnP)G0EcA=VPY4HU|&E;UfjwF2}*H;r6?d3
z7kBKt&+?sp-tM{2WO<tZK5oqU`FYo*N6Kl(OaB3D0-+fwaG>Y6({l{y48updq@FB+
zZFO}9>jZl5@bm<4$jCU>?HA+(gHQz;95@6*ZpRax!v0ABAPr2>gpGhHK!^;`Bg!H5
zPr?=ehk?HI<P7)#sqN>7vQMMWnu5FdIt2e~MWW2OAriwsy*>eHXbBcrMcO~~s(#L(
z8-rH>)#B;r)xjeKB18jII9lC@uo&dUk3j;WfrF_B*#gG11!Bbc4;_SQP*7H01*x#Q
z!aTofeB?)2eRg^c=M2F3<H_d>YAP2HR9)Ij3}iD7t-wCN<o~q-DME0+<pjd|Lhu#u
ziw{Du#D?}_`Xbwm4)DN#1KI~|57yNoinsj@j0FH=?F-D~E7|H8$oXRfY255Pm^<R*
zBO)T&X>+J^a!ik;ZEZXt^84~c(DZ|z9s+rIbOzuJ_(wPf2jQQGYz6*O-+Pb)xkj*I
zUc&gvC<T3=xOf#j5?u@+{OTJZiuMy&bN`?tIDiNLu8ty{e^99FYHQ=|gM|nl7*HZv
zeIr3Z4({jxr}^6;>N7^1Fa#I`bZ19T86(E}cj^4J&Gh}78o+lmJ3T*we11QR1pa4N
zyZbvbv)5(|AKWRdb5QHA3I=LJ7`HF+TpMo=5%|+!Ic`-+O?7$UTp<93I9UShZ2&n5
z{}%Ea=~qor^&hbr=>#3(J`x(lgFv(r3VdQHCkJ1l%ZP8W49IgI8o~MD5KvZ)SHOr^
zKKIMCC0Ib1EwL+PH_Hp5Fiyi!Upd$pL?EN_+prOI0Hg=>jt+EfX*>3`ejTrdKjRaB
z3W@UNlVdmwgi}yNvyNy4>gQnqXb0ID0BVoMS)V)W!}HfSe24@RM6!Mlo+c0nzmj)V
zEr*9Ad`$-cPxR^jq5xgxGnsm4C?dgKo`VAg+SYrp8s0;4)&Km;YzTN!b#QT8f#mjA
z{~A2!8wyfH6s`F~{*d%TeMO|GqMjatxqO8q9#(=qfCV=bIaqh~S$t2`!a(w$>;191
zg7V``5+DyAQ*`o~zaHuV_3Um1((#Ee{+W3rl(2@RTwg&aB>Wl!_ai(Y@F2+W7lqwl
zU{hpOVLp5>Gx!g9AY5Ze2(|pDpMew(sIwiYr{04=<-ZNs{egD^3)D40f*c6njuP%C
z0mQ*A_!HR1fL<eBf{Geyf5D&C@Vlqu^=SWF5&h18{_9HoqiHp`s~@*t9BF>Y_Di1#
zTd*B=_zC<jnDib08ZqDhm*rO*wDh+e(ARi!Y>>yV6BINEL=Vg_1eA}oH>CF-0`~or
zX8pZ(axW2w*!=DN&6x?rxQ=QQ+rkOilm^uxZwur7tJPr82$w1u<J}kZoY|lSGHEnw
z0!h{(yi3lf(aBg(Mwr}zR9Z`H_ruMFq-8UWCVc48wwk2^xIil=X#)4DdB{RhS9~?D
zC2A8=C><3R)WqQ1%G0{#Nk$pCm~IVW#|{`6yL&Xk+ev5oLVW`pqk1+owWR@+<BK)<
z%nZ}Vh2+7$qvfW+Cg(=ne@Y*=TNvbD6Mx2>bTKZ*TUpHZ?1=K9Gu%LlfiqepEGv&Z
zbm+U4gwtUc=`xtlDY?a7i?s*lP0W?<t@*~X@INz1CEH^S9aq{pNoamm_-!BZ#ESa&
zCjCJt#KG8F2zwoxF&;5e(%jx3t!y#u=pF$jwzf6p;QhoY`B;NQO`Dm(|2I#L`;;na
zziB%Ng*L%sn(u2FMNz}f-lQdDIN4~J>tLcUNC}&WMQ(Mu30b+#`UL(G0q6Mh5rBbZ
zFd-<ROLgeniD9pYePHZ3r49ndhDIZtciZ5nE4fo%W6UO6ncO)}G&?QDM#sTS_O$QS
z3-nbl>YArPv}nd!y7a*`j`Y3MSzRz^5;ZAT*-atBuPy{w9B49scPfL&`_#~M<?e-=
z{G5{ByoEC7mq<qexy_o46_=@%SsXU<jRYkqiyOD6Plk!`$fZivawj8|<j4u_BOv(a
zoEa;&N_Lwp=%eeU1V01XnVvw68hfm-t%sAVFl*c1%``?v+N??;PMcvMwU?Q}HB9U<
z)MCW)gY-iOH;|TlsPbk0V`R=2^3JA!5+Qw_jgG0W;_s$;?|AOZ^rC!5vI~k!bJ5XA
z7fWU5KZs3=*1_KpK!^ysJUpoza^@AwA#a5>w;OO!D)isbLz;U1XYsA8+Fd)pgK{kG
zNZAl^zj`rQBQqq0OiF0#Cb7k2EMF>(@V`RvYv#s`U9ICMc+zAL)PujOikHdH@8Ll;
z%lJWw7xt&?)loM5mz(52qXe7@s?_T!I%@eziO;)WcC9DGGCY4DN{QCz<%b#^toKn^
zQB%9pk@WrLy&M#uo%<-6$-aI4Bh#W5JlUCXeR!W>UQU^|Mc#8(=A=!FbK8iIWVNNo
zL>Ocr)CHW^px+vv>r$V3SU)e(@mbzXo-6)|6FE5VmHOC~5oIRV+M7+SQKU{!&)aF@
zvC6Gbd`o%2V-yqe9wRF;zN-2ILR1;~z$kYs!f=70ivH;UcjauVp79mw=#O2SY-fnt
z5_K|5`mFY!{)9O^G-;Ogyi7<?1^3FgY;k8~v(QHQWl&Jde;Ma7A3>bobSM~%hWSmo
zi<b~-gW^;$F8xj7Qc56XJBNL4Rvqd*5#37)B3Y%&cnM?ldu5<peT?XU_v`bcArfpQ
zQ~OO(8zeiWSzQ;kaiVo9w33{3Raai5SM1A{x$1g=sX||yo+skS+MGh2RE1lq&fm?v
z20_|T*GvEI5k`TqBo}F!%h$3SR+#}eWohdP-Sj3PJJFAO875@rddHuy{_R$9$Wu^E
z6V-~Vr8$adPZZjYLkyyPqd*g+S5mLA>p?fX!-NBa;%8vPUBUKp?epiEd=YQpe*}m<
zwYd=0U}ka-(LY&sE-{=g7;0^>cz0l=37-ta$or@@x6N&j#Y&F*S)69uoP(W5cFvkD
zHwSfjNn0Z~|0gkW&8U5Vk+9p8phLlSf0HXC%XZTu(^;v2(FapSVv@k<MN4X&N4qCo
z$ikH@NsZ;Z_b@nY2asPDQ^UM6(?l#R2pGe=epX2Ajl>J8yxUu4`Brc3FNV%vx0AwD
z6>Ba1Gs#&dmql#?`IY|R2&ol47J764^mgxr7{8Kn<(C&FBiFtK*aksfO2*H;yj2fm
z=Dte?{A>Rxe0yR-e_sH+NkV~J=Nw(rzhACBEW>ws?18FnIeG7D0D92?WJ9+%TBJXo
zj@T<v95f&<pt|#b2=jjQ5cQ)oJ8wEGi0*uLBkT%;_m@>QN^DCtRZ9D9(=09=J&hkn
zS8KK#@7%Ec$QmiO8YYEWId!_$Vfq9zHwDEPxX&)?2`YBu_6tb`2Ip(eN5qom#bmA;
zV7=ypoDJlw170`(3V7fJ7JR#OL41L`L?QT`R4VSd$0myW*@$tdXPNVxR+O%%6oo#0
zx4)stHkw|K@+UC8Jy!bPEo=8~6T@}lF}`tc>J~y+5=buOenClMsgWMTBgt(hqJ%=Y
zcKTf4y6G0=_|vFd(Ol>l+G|RwLJw?*lsAu{Fo@jsW@DRZYKQ%?TNxH@bZp8CgGOX>
ztytf$Y^-NHr&6y>$QCzN$9%78e|$!5WiQpY_~lx|$!-)%cC|TWkt(UK5jV1$NIfJ=
z6g_igK#G7o^B?%AR~x0@6D_M3-=lAbN~NF%VIN|)tCXpzICZ|bH7jDF%${8Tu6IYJ
zC{WQCj4$O#UY@-Qvl_{BX7#qOPBdas<*!P@6Nc{uA^IerWvF6`(OEQJp8<5Vc)M?!
zcH2+=C(;(1-AvTBY1rCl3|^2&ia|b&!Ws)W7SWve0Up$1>(QPk_Wk69%5Nd$xg15#
z0kvu#3vs0sz1bSNM^J#=ZS2GFWMd-5=}}`J)eol5S(WO|^ZZlT$5?k}Dsvz&w|+b$
zJ};1<q8Njtu76#`0yl4;Wu2Mn=X{cnQ+!7wvUJ3(aLm`?Egbr=qOUvXZWSf45V0hY
zrW=|0kN_>KIvJ05kBt{fR<e2vK!`9cEs!46y2!|gtsw<S4izXX2;i(<RO=P)=0f3a
zd=AwZ>-D)fH>{@>2<OzBuU7NHTz>mI5bI23gv{^M=g}%dR?9Kk1!Zz~mRc7{zuUR`
z$oDYfTe;*M1NN<E7T_pq5r=0xS0W(klR=}<a!B_%(l5$iHJ$-gpnG&!N`*Xqpl(PL
zo98*jgRH|wNO?Bo(K0ABI+7m?B-~_ThdRq{oktkGI+K-t!?`hzls~Y{Jd_i$!lfzW
zbS6hZP>}DGzTU&Pzl<EGf1<o=$%#npAf>skQNJdTb6GNiuy6Mr$%j!G1M(1cEn8VO
zz9*-{l8LcCYog7didi|t*rv`{2JenhNor{DCDZA(0xlG|y6l4q74y}j_+WM6`z1~&
zwD2<Goz{Q%yueGEiq#@*N6EOb*byQry0NbV<|h)<pFdhATI}=WH{BYnac<<1+$@5c
zkZ&#H8N6@-ku-QA^=->?Ja9UB?yWJ-e8c~mYoFo@efgiN2Sy4tTSnRBVL76%K{UU4
z9B~~n_Wi`r7Y18JSsC8*rcZNn7RG%YNY%fab4x6i5IA!Qoc8&2b~1yWuZZ_>=k+BL
z8!<s)E)##{2W^i?MFewT_a#)(VCQWS&Cq?UCKp4w=Efr&p$V+{dV6W@Id5e@Q()Hx
zsi04e?nqEI9K4NLsuS>IB~)WZx`Rr9NUo%7-6#yT%G+^hq~g`)c=x^`JLh_Bg4FQK
za?E_%p3x#wW{+>@(thu1|BwE6Gb80HYrUB@j|8IAeY-xw9k0c6l9`14rKx7IoqPhh
zdu~(I_^1wRqY;jK_<da0?M*WYf*@;mLnz3RkBBBxFa7zMsup*sglVhIzNHXdTu=3^
z5FB_~jZp<F2B!&ClTOn+rndVtHTun>kR8g%c@HBR>jN*Cd3nB4@Za=)++kS%u0eO#
zu{~a;?g`$t*%OZHm~n_$0(*WTEosclonZ~bAw095Qy>s<KbJ0ct~$se-|lrL6b#$u
z{hA*)uV$-F+>9_NYrZ1K>!&Ojc*)%kNs|vz4-GsfqY~%8D;OejlV&ad?ZcUyY^t6C
zPTWs-8*Q40)v~Ly&HN0q=|(T(x&f}<;irRk4Oig4imJva=X<gohGCbvgn7zA?bvwz
z2PUtlP4g<-y}KEl#%6yjvIJb(6I<LbpM9?{?@5z=aH6WzhU)UZ*ZuUSoPeP-{OdOI
zWKjl=rhLf~28^_Ewqj|=StU=>ents-jZtHf1mvsrl1L*`v@+eMH}x?@BbMlt_<y^i
zTm+}6p(DVJk$s@Lx2Wa!kju&0F@W{j;GR1ty+2pXSDNjN%=vTW{`i>3%|TgpF9wPy
z9>I#DRhbPbRa({>le4(K#OQuDpUVP!0_}j2`fu-bJ2ZU+2D(79I{@|}ZwslvhOm_6
zZ}nYRm(!Z5Xe@~U*j#~t@FydRV@JpP$Y%n_y4noGj6uq{kJlui`26!9|7l;UQ=C=|
zGBpJ(BZT&_|9DUuwBwWZypl4-?p0G*`HbI6+LK?Gk1cd}xo({ejAPI2Z9(Q=FEP&D
zoCJZa8$0YONf7<PxZE4btS<;zsk&oo1E?+fB`v->FJF&HDrT3P*cs;Y0)@Nm&e*;t
zkF6PZ{#`aFIoL^o6%1U}b{8Sl|6B!-UebQ9hYrWpoMR2k`uO$Lg$}$(&#*cThF?kn
z#PpNTu-BB$Z=$KQgyoAH?E>NDT>~-w+K(IWl?6|Njkpkcn-t+|QIqvOgJjvc!5Dgi
z%8P$tVtGeYnQ(FXV#Pv?*+t|9VqEkN4rs`g2TrTBOn$Rm2n;>$pTqMsWeEp{PxVyy
z011oumTn;ziCL-9XQjOPX?m!WP!dH>7*hc7n+@wwOOIZ7Zr5|x8V3;76w0)18P5=3
zVJk#0gaaeV+gG|8t_W-6Dv+r#SyCM}`-_`57J;(rMj<u2kg`f@B@}I@4*2xAoDXu9
ze=PiyLJ*aZ#OhI5#!-DcWVwfWu57Uv5G!)0qOv&h@MQl?-OLyY*0*yFODWieBjb+z
z4^HBY50MOF+-*X~h@HbCg<qQ_XxO<O%s>QeO_eM_6=EMQxrK`l?_>*g7aTlZEiZ@{
z4qwE%>Mgod8z580;`@8x-w2SF=U~d9bFA7{f@8wsPek)(QT4YdS(cC^=Gs^eudCe`
zlXmK8j+(;hDjiZm;ZbxddQ%<6mw#uuHcwGm2cY577mN6N5D<{NJ#T}sL!!QN&gat|
z3A6qi*%pKvE1}we&Xy1e?T+Af?f*N`Zh*1v=YT2AHEV74A*m#3mfl&CBL2r^z{yC-
zB6h|sp)e2?DK}fvoRYThqHS#H-)O?je$$LfUNs%pb`n4qx`xHHm$5l3AmbXVz8}T)
z=myx#t7f#kQ%zY~apm9phz#cagf6kDzO9~VINFA*%VHNy>YSiz;eU_Mqu4BQhtY|%
zr4l%}QDiy9qE@kaPt{W!NV<$<L10A5z<2@j(8UsMeM%6>U;OCJ>_auVphwo9D!AGq
zT2pZ8?Xc_rdK+Pu5tmB3xo~Cg;}VS2Nh&IbW$@jOp`H<u<jjgb;kM@3-nF6}!YYl%
zD2Du95Ik1Kci`C<4j<<U2I4?>$mtE3vNk6w2&f`3q=6j<m?$VqW%bMg4cHk-v%U=M
zQQB}~B)v4A&gng7DSw{a^lgtF*gkVfAyar-Gjdd-c4uEak10e=Z+?c{BR11lG#EyW
z{)DT)9OvufT8I2oE_5AY$Nu0BhuVCPP$UYAOHOI{`1H9&mD5VcX^kQ6)20jb<p~<P
zm2UWpU3UmdXYKzHbclH)T}KyAlJmuGDB7iV{RM~71;J=;l-Jorjt2bgn{G1xuo>xs
zLaL|pbJ0AAOk94{dMlULW*W{uzlac=wWm>FMO9)>l_lU1d-C^eA3`8m<e3rs0tD*)
z&(Eic1inh)u&<cH)LP7&CXaS0V#OgwWb@6~DFq*@89ld-ydhHx-wQ{Gw2o+*?Pz!u
z`sgAL(NTh$lh`lOL;RmRR?{&m7jxAcS$7zJ96L4(AOWrt+|7yir#iy_NVT~`baoun
zwO(NXhO-{~<!H@qz4Ts&-+vs50AvFNCXS<YO#gBm*`-!raniV7zL<xykk0yFA7Elp
zDUi&Fy*MHXW4Buw(WvIR%3)ti#N}ET!K#3AK4po#D`)KMzId+sOQ~sb*NyS0im_i0
zT9quGH6M~=XOJEZ_HZz!mr(TH>n}NJa>f5*1&s2nP84U>*UC)BtKtLa;*Tg<l2srt
z$RlH$DZZt;81%J(uiec|%A3MCxdnJyEHSTxG>_Lm1Kf#WXE@x&TI##`7r93ntOK22
zQKMeT&he|yR1>Sew6JJ;Izl63v!%=kpH6}@#kY$sU4$Pn$~jJcqT8>rvf$$+TO!z&
zPeCeWG28~VMyz$$s>lC12Pvt=WRi_9V^>%|1NWSVNrOI1%)oFf(h~JGYfm_ijq=9)
zGa8kfA$VaOUx@DsD~D|3>B&dC4N7|t;%dYGE40Aox?Z-j$Gt}ZVoO`beQnI=1-cpg
z%)L3v!RdXg0fXl^ek&UintzlkfnKpq0*Uxn;$E|W+D=g%Hso~bnH)vVWrkfvBkdEC
zR~I+W1HWarH8)~bdgR-2#N?ckrS|-b`}ZmO!`x5uZ(|1m@A*cMI_arnwgx#K_Ox^%
zp8;E;pm8h5jMhyoHo<tagn%#=e0iYo50kWdwc`!Ty2biQqQ#kZLZ&5{2<*{I_3-K3
ze3VEg6F<gcsdMjS!Gg)}TIdQQ=)fEJGE-uo75+QD=T{f7+TWahSSHb!+oCl2fv5Lv
zoNzR2G&cZp9s7Q=FoZL48roliFFO0NnG9II(s~z4`A$?|2N^MolKv(;yYOliOs+wD
z6WErvW(g{#xAR!qM&y%-3~#xjRBP8qUV;S2Wq9{{Od^wj<wvm=2$nA^%L}^LDftp|
zitAykH`s6wErAX{-G~)*I^yfG4^!H_+qu00xx^=Yx}P}RNk<xBU&E<qH@^Q|^!svv
z-lu7AV0FRQ1Vcq*I7(_GqT8!uP7Lq5ihsMop{aMW6cMZ!zij!$NG2?WUP~yFZJ(ng
zz)m5fnNGxcomlOu#%n>;>)%n^yqv2FbUXqk3>*~=8q_7P&#qFb;6E6BaBV$HZ!d<0
z_RB;;%gNWF(5rP={)uS~p7-*5GnVXYv+u{z=j^fTaA(l~@Pxw9Cr;-|@sgVH@4rmb
zLduliX=wjad<)q-(@fq0fNCIFFPhx@x#dIHqYTb5<UInc3>-7~D}1X}5uQ?qQxxks
z%9CZ7&%KpSUkxmIZ@khBVBEtdV3*l-2wHmwnkWdSg+CBk6YHvfoyoV|7RzyK*I4);
zD~oWN32Pp2$Z5i2$fVw;d~u|1l?NX6CR4qVE4q+pMb3&R4bbKi@)olhsF%?ea%`bU
za8?OoK28h~NIee1rP=U&_fsS=^%5InJB@6j!Ty^zn$Js&&-K&1AW;30B0+uH=IOQI
z?vxllw?YLBj=odL`(+svH&nJFinbVYmDvy+y}M>@egK&vQ2AZdMIq$y47DUz-<EHP
z)kghib+9;~-_tPix^M}CUkW(^t6dA-oOf%8>Z(VIkdy<#%8A*T8s=`0m`^e-Sz`?9
zLAV%k^Es2@NN#~xETU@%BU!xJX}*NVGAS6Z5t`u~7Po(M3-OWB3E3kdfKsJn?%92P
zckRs4mW;k^ud9^4tKkyC)%1#qcw3I4;!J!&JGu#{w4A$!CU0QdRcr*E$$UV;!@m($
zyWUUt*^e-4p%{)HDXaYIsm|JIUj)3|*k&$x7WPFz`06?626lBpim9%h7TqnUWiB|T
z`uVTid}R1HVTmp#V6EU{cu=&-%sOQ@yL$kS46nBKm?Gb}V4A4}KwbB~VAQSq_1cnc
zl$rl`ESd*?`!w#@YgPBk!wmP5lDun!=h7{xqc@1hXCWBy2DV3+!@bl`w4Uq=m2uEK
z4^5cLN!Rx2_)hBqRbgHor0AOrv%z6XiiVlzEHtF9uNH~6JFAzJQ@O2+ddADXDYPf*
zUS+D=YP?k{t=%SKdZI^9RO>7>TWx&eSkWtJ(O?8qQm&b+3;^=rMih~7{(SM5Ec3ZH
z+Ej5O@Z1C0Z-Yi0R4vNIPT@GDn4gt7rzk8QdECaTxN#N^GnIXgIn$`T)@k9kEed6h
z8RsU8jY_+M5YzoN7fwW)`HA?9IF?vtn;q=x+p+{c#!Vw2UYgB}e(d<0Y;eF*kv(4G
zh2oeBDnZM82`$=bkuS6jfRX(sl!d2r*1*PglQVwKQ_O)Q@NNmKBkd8!arHZgp@F5)
zE~J9)DF;5&kujZPK9DJlK;-R<`nN+;0rd1j`^N~T{*w6XgIbsqhr}l?KPQX`zk5VT
zO?9q1KIEq*Vra%gj>;3XKc77QIT%+gbh8o$(B#MIh{+)6){>K-xN9vw-XW%5<IB)^
zy<e&8(bEp0>;?9pP<aJ&pGF}|vFe!n$fjztQvZak2=bTN?tz}gtu<xVoUD1R3Rsg(
z32sL9%QlJxaiG*Ru{P=8vyhoTj=M3TzRgTpR+9Gl>N%vqh4F_%QUW$|$bE6SeS@?7
z>Jnxt$)a~y{Sxp|Rc3~4-GU{Pn*MsXfvCsQw$`RDX$p}+j%OBbzUr&(kKS<DR(pVR
zJV{KOSTi`f)M2kMLbIdeshA7dn0~ND;POoKFddQ}#pp+R2Ye4MddE@f*ID;1tGv1-
zvXT}b+uzqkQc1ksp?%E=sohHIvm-?F6y_c&BivJ-F*eHqn_j|88hMGa)l{Ef*T_6o
zQ=Qt7%k+twe<3iilO3|-pz;D18KJ2Wg_}qM%Sy-JttuI^{m)BulDzA#5$9F|Wx=BB
zB%(|}mEBQ3C4i-%kSh_pGODDT#%!ooV3KA+chxlCe`EEu^pzx)f<hFm4X@x6&7w+K
zC8}H>C-uk0oJ9WaQ!xlz>=m`5@eRJwyfKlg${wq+MTFKd@6uM90D*jEQljG9K@%MT
zvkwr3bg_11!awHz3@1NKK5M)l9~GuEHdx)4OQ-*V*CK;GMVI&<@kM>&&Zc*;d>1m-
zy8ABXEv0Q4#^x7zrL%6?iW{v7&20AHcO!3}F@-K_TT#Th*|}`*&>2bK2%Vn`UwTV*
zO^c3Kwi=Hs0(B*XGATl{`_icgW;xrQBEr4A*-6YWywJkK%=u?^wztBXqv$ZM0)FpL
z8()jv&Pq(uP`b?}r>|_(k%#N~s%8m8zDY~;%6`>lUN#+m^Y0)RNq$<mpbgt{5B8jD
zTeWOWz!K$)1Q|1ZDD>HN#uF42+wk~0s!*kVg?BDYNspO;45#>+IFCoCw%#L0aylYv
zY!!q?Vv)|$u#Se`x=@FCNhtJ>327+Cmx-EA6r?j&85lxzvbtu#PEnH<GGd=;z68T!
z$5z!EHgugEyQOK7&mx{gZxhO_C3!I&xea4v71zk1#jhej=!Q*rAbW0#S+*n4b5_5U
zqxUuIJ@ws#j=`g<@73eRkCgr1cgPpmv_4e5OavM)ti9teV78CDh$|k2iaFDG>``ao
z@fw<``K<V-_tmL4kTr(-(Z`B}L&fwoZ<=6btsN=XBhNvZ(o%>u+Oj!6+$EJ`biyTU
zi|W67F6_vC5{8c>ihX<c%L?FOfn8X<`ZtM9&sNFWU5AX(v~f6hb?FmKV)-hZ;9#9y
z-CS&?0Q@E;-dCovYsqd}S(C!(i}iBAt8<=E)>gG&c`d!WMmV1y4Df{)9Mf<i0i3@_
z*VnOTOA{esBT~T6-dG|tU+_mJw>)aLpYEMPz4N0aB7NboU&|-TBYU0z*+%2Eto)Qe
z?p6=a&9HP={qQ!kk7JqbGV;z3ED0Zkj|~rBOi1yq?YHkzmg!|q6t<}zzo=pj*^bXz
zXpB+$rgpA!5v;vlhL2Ey$tU?7&lHt0-33ByH8{puFJ35T!q{;C`>B$<h<4$@Ev*}9
z3%4$=&S}VRH~u@<7lq>2>UlM%HoK1cb3$AO(*N`?=w#$i2e7QUS&mlTd1ty~z)LFD
zKXcO5Hn`I-d22MGgbSB1D&F!O@sZjS4LZR`ax0mb!3wVV+l`&qD%9Y%?>i$w{Yk(y
zxSm4!yd52ErKSY~8G+OhnO}i>4=KL+$jP)UY!h_r`-VSEMmI#6z306~c${ep!ka$6
zt8e<?**FNQCM0Y;+GqLeV`(#UbNIDEY?!r)L6s-d@+P)cMxHEdLiwvnfr{pi{E_HA
z$1-|uTJWMB#y#rAS_8#l?|8<Q6m5Yqr+m6XkIL>374kTR&-HuN6)M{)t_W-j%eW+I
zDLb`tT|;;0`A5&#Jj;Ym9s_A-1bFyT?cJlAu!nMZB$J)xq1jxm`g9%Sj(X(BfPlJo
z#`;BR3o)-WgU)T$Q1Hh?GdDlUA!*H@f=&3Vu$Bo}nmFAm*Ja#`qwKE14za9YJkI62
z3!W=iyJkeUGGcfPsf3{{w4MaRsIP%YtAAd)a9O&1NNFlQ6SKh&^#_Pve+;)FxSSRq
zSla7y{wOoefq$N=mAO7!*CLT5!fy`p@#3kvl{c`4pF8%^vgG8<t=uH%VgSn`EMhl<
z7LdGaG$w+z_9VO1v3jsga+s&;s#d#v4UDIjdcbu;yRoAm^dYw7y+05@3#>$0Nc1R6
zxmY|h@q~$Oftp+L78fHaDEQK^9R>qEyc$taxVJdvSR~{9+4|(kL$z7y&G;Z}>Z$Hb
zCFCD)Yre1=QLe44QtxO&mu5LG!5v!r<Rbf*WNYOob~Sx^r<=tKcQ$?P3{M)W4-&U2
zL;vhUa?)tB!ZX?rOm0DyJZko;_*Bu3HZv-XF+=BQjFg$8UVb)po%DLb3#FmC#9*ZW
zgH4qsmizi&%5weigk-w2%U_H>j%F>4qG8p2<aj_}5H@&lw-2A&9s}$3jjq?n+@ka{
znR-@2RC^#~AL)Y5!CvsT?rviIH7-mSg9Y~T6qzcgMHA&XFL&tK8ynfQl`|SgH6qc~
zt^Ucq)_DCtPk?&b?38AlOW(`rnlGpkyN9_xz}-kYJdUa(xYP+|pf%j{!lrKp`P)N&
zwq}IYXx#1u`e=ZpIBM#qca{lyFGXFh5?uc~=$kzg^=+hB1MkfBtUb~fgu82ZQ+T#j
zYo|AMgJ+N5X<>3~qGtxk1ZZgRj&c`kjUJTBs@VEu65z3|-x>||Nk8@Fz<0x}ygR<!
zbl<`;-|*d?<!7B#y7$b^!AJRs(-Irlq?Ioxg0(HQldYuq_ZWQ~uECyUkChTLHt94D
z{%_1-$vgSiPn8+a&&IW|85Ah)LyCwOO)x1!d!6@wF0x0Ga^{$dY~U<^Pk>0iX1}r3
z8L0kcuE&y8r3rM`8jot;icDmT5sfNMRsx}<e`w4}S*Az%05X$}VMe*i05U#4>j-MT
zNy%eFeqjjc_MRVRS5`&GoWL!`#gub9>bsvR3H6t<$Oelg7+kflWwlSu9OP#MdJF8`
zGdE4U)TaAeu#eN;3F0wxzAW+afeEt<>4$hZJF(mdO0lL>0S5H3FJWV=kN6hj!^@cK
zvfb7$9EW0ilRn?D16XyhegTcT6*7Ug{ef+q*Lc5i{4f2FT0qvVDZ!<}yoW#oZ?Cb&
zVEOLp)A-Gj3CzfRD4B~Nagw)=NvOINzHWI6L?J1*=w{Q;-uCmn__bp3Xb||&9c<S|
zpx94xU0DkQNz;qAt`iDuPx_xPmiug*Y~>TH02#B$&*;Bt%L$Q`%2c9T<xUq{O-DO|
zh=I&RgR3)_5t%7v4V7=;6GW|Fs(+1Q>FdUM<4N9Ho1WJF+0Q*J)v)N$ws2uFMVS|j
zz7C1?<_*_AcgbkVho{pYPr*x-^`dSsl82bY6$byd2<FGD_}~2k@%8}E{%=_}Hpc%g
z%f`XQ@jqEMCN8f3aUT7@VK!zaCPw!EU&&Fssfx}{7F`U4``h4-Zg5wZ_klutR~PDb
z5IFR$-9RwUK<<9+0m<D(ukXk9=Nk@Nr5hjrn~rl0b_<nc>I$aVv{0#_WI`|v2@NVN
zfuNoc@G;PTCKDBKt<H^1PSEuX4GqqHDyos3K=RsELP$qcbwYxGtreI8lhY5ykHmq&
zL0Ea{1Q2eZT!EldAZ3OCkogtSohjsTm;xrim;yvcJ9Ac6yH7|ULfR^V;AT7uQ$kUC
zQ%gze)gJzJOpHIzcg-JDknyaHjUZkfn;Kwpgi|0*a&Ynh;_;&)OZWm{E+CmeHFCE<
zKndW2c;)c(3dstppnof<Eh(!On0;B*oSj|3xPCGIFtNG9Q$Yl@M-+B~0hLbxB`hi^
z|9(~=xV-n<PXH;X{6_rdJ|(`l=?Dqyit0*e2BiY-_<%HkZw1XUuKJ9BL2*_fnfreQ
zKPg)rAU1wkfCP<CE-%N&rmqhU223ta4=2H#T1}jqKOyrwQ(IvBL62=<-`lxCZvuXi
zj!q4JgvzSW@Act-=wSqRH{s6rLB6y}p+D5;e}X6D3laOSet}<wLaz4m+<v7X+kpaq
ztz%LnrvzFaCnq0)yiu(o)eC|ngEu4!t0JqiVf)_+_Qr55<h~gK1Oqa%o1XTtV1DtO
zeguqva5^c^k7l25e7<COf1F2TaJO~3e`B+MeYeaAAu+5m)YJW4!9ZxMZvy-my1sN-
zti%0%V2GoblTfTKV;%_xoeflgeCv<;k1sDi<X`vvhz}nLl#Wk<>l+$?HTXrT4-Dub
z8ryyaZ$N(2`5mti^K^t}^`?K*Yi;oa@e%lc+x7-G2euCxo=)xJrm+!L<?IBp9)2f(
z90q>l=TOdJ41j@hfdu7QXioc%eiWzlIj;2IAmRP^;|L`JLL+lw=jS#--W>zq9GM&;
zL3DC=0{r-Y@!$9U@X~PfZ*NY&h@2B-ga72;G9&yU1iI_nCiwfq^Q@0j05C6F^t)Da
zP6*D$=*P7Ho~wp(`bGkq`4#xQ7BG|S+}NPXS_ifKNnZS&W@%_^efrSP{7w7^{uP}n
zj(lumZ}djMWQ`$dEnm>l*ix0bB0&2yMR#QNNaTwQDb2G#mi2YC_}-O$&2RGQ_^k2E
z*8vn$t+V}${Hai!Spzyh0^UFJDG&8NFzCPkUGT%81=Lg=7ZK9T`~6Fn>Ay&dJ1uy+
zijW4xE=L~_+0+z?!8;X>u+GT^%wJ8A#1858LxKjxk_<ZjQUtbtetHR655zU}t1F{t
z2h=p7zt%^K57cz=+n}MT0W^L3Yt{&o5%}K<nnCwFa}2_`^_2}+n7rMS(F1Dw{>`W+
zs1f|H3Y>xWUloXv=l8$r#tXrdKTGk~Tfn=-C+)AuhJS9(@48j(ETMKmS_@#TZnUq2
zGNYYa+xxP?2P*w#==b5{&K}~|8tCNry2HQf4?4K#2cwfiBM9b(K!Rga(~oJ`)SZ87
z*7p_Mr}>-@?T!DIKOzzk(m6D1u)Zc(=0o^qF-Lg%r!<<Zx}fqSbFP{ptmPEBP>qSK
zJIa4N^Xmj)0Sw3%2s35$;Nubi+}BbNu2c(^NW=AyO>>D~ekYqc@-Ewh`aOkU8WPsr
zn#m&6teOTw^*MPFU)s!XrSd3K_D<fi9Navn0P@kzwP%!ZwvR;Dk8r8}!cP^_W|2u=
z3;5Co!j3Z*dSA4GXD96T%QrL&(Q2{2NvPfIyEW%)`!&T_FTRTI$8+7~CPfcFR*iaU
z(eRUEb|>yV*EI+k+n{IV85miqrs&?Y(PLD7zizL%_SS}(49*mTI*t%?{D(?Hc?r8>
zO(kmu3YI`6GRajGIA#zYN4QTI-%=Y<i#g>TY%^Db>017u2+T84mVye!8%Y+J{c-FY
z^@+O)__l|en&U$xZw7>cxjN(lGF+#VJku2#q-ayp_zWGGYCOrZ3q`+N^=Vj7yM!bg
zItr$NH+XWFjVV`FO9ZQT2<tg;5mT?lz>wa^xqlHhN%l_NDP>O%BhM^}XimVr!hZ!~
zi^l-}vAR#!U<`AGfnc4{ufw==!3SDDlh|Kd#@!6IGzYkg#-3m*pRU`L<nVYiC=njl
z1Ic2m*W*bt$6AC6oHohsHV;@90^RK(?G}k|KPS3Q2xbzf8#kEj6ClOrN8HY!uG$6-
z2ZUxwY;`DjKhqC4?>3|S2c>ME(6%KF-}`aO-cmM<QTe1mgIOB{Fs^npNiw4&?c8`9
zjoL{?F?^P$agc-}2ZF4q9rc!0F>9K&<7Fn0sq^)A_|doz8K9#%w-S7gFzI(Yvxmzu
zbJYo#Bk4HGt@~DB#0!<!PEC!2PO31`c2H!ukFv3UbQp0|4}_^^zgaU~GrJXTd!>uO
zT4fPid)ZIZ*~XF)HR%~=F#5_5RLx<J(4qDayn%g^M*xzh<r{MzLpB%mb``h=o=|N8
zx^wr+U^OLCagK;2Nf}!Y=;H7>j72HC+H1<jk!tz|@?;_=7NOD{*>X~xye^**AeIHy
zKgl8~cCc*xL{@=EIm5=yUMxi0m$sE8!*#9s0v<Reb?`3J^oHL_JCFf?+9#!7DL2Gk
zOR(H$B1m`Sdh=ICo>NwP%3uKS8`A0Kg?}r~TFh~ljU0StsK<NB>xHJbtKr9dL;D`Y
z$oLw|hb&V3X-*UsYhvV^E+q2`u@OL#xwzVZ3YHU4bEYiJW%OH6S&CS^X(!kf^holz
z`1)Rk(v8RUVaa?=!wiKlvT0p)>-SF;{%%K6TkN?!=4F2`qY%OzIpB*LFB26oTJK4#
z!3V9P0UCg6@+=C)l@+DqPMO<eDBdHTS+%MzY)2XJvnv{8Oy<m}vdL(`_;?%@9Qaa~
zoz+$*v+hLQDIReUM3;nzAPCKfuhaLcq@OcSW3TbaNy)Oh<uZaPCn_}z2f6F+?R(?q
zzfoD=wfu%DwJpB+86*yz=ZyK!t!UXs?Ud$sO1Efvm7Mx<u~X%P3oz-^r%j+Wh&;T=
z8bCKXYF`~hlpAC@C)cq5lEiX)^|%ZhZWIN1WEBcx#l026YF;;QMZJ2LX@%V*A1I)!
z<MBWyb-UWvUOPzjT^l7Ldy$XbnsnvQ&c{d*MMe(e%oz+Kwp0JSgc|#>Xs4^x`7{w<
zJC*dI`@)v6I#bI&%eL;pR<bkOQ%QCj>KmPuhIV+_M(vy%3c^H=V~v%mh)Q3T3Rez!
zuB;xN^jRQ*fjJyJX4@536VtCDDI@=BJ%~j>pXoyiTiD@_yU*YI*6<%Ygqhc4v-6F_
z)wn2pQDfSSL^FpRd1mwL)U;QE`t;ogCr6#i88}E4ZIVQ!s^LC%;zlyt?+eyZi%_E_
zFKTx^&S*$wv&QX$tREq2S*~`%32&3lz?e@HS`F)H43d@Iu*IUkoV6nK6p<?BzhGCC
ztr_SgN&z?)2P9&VKq2GKo0I*uPChKj*jCgQ$KjC(Q_op<#MXO7BE6h?B;MBPrOFzi
z^S=?}R|!jH`H9G=ABY}Tg9`7J*k71i=27+zPE;KG?HE5(MfsVOYhw>#Ks8BS)xU#o
z(J5R_DEStX>(%HLKg_#h&&KOD<kNA0rN*%w>bQY9AP01xxIbg{XkEO}XSwIj92i9%
zCr6wh4bGE{#E85(p4q&rv#9Fn4-j-x3M`cea^^8<f#QOTo;!*ZhF6`a67PLl`;TCa
zo|~g1I9J7kBP(hC*K5{WhV*0M3^r}GcFD#;Z0{*27IvSfE4GvZB~ouDof(J&Y1m`I
z5$W-n$_${`I~L}P4(xHW619~s8$sTeeWE6f|8uyK)n;Z%xX%N5NR=V=0PuCxVzhoV
zB)p#iVJpGuMiV1~EVVGF`8)K6w#WT(k^|zdDE(Xm%^!l~r&+j+R^yO%=j>F4?`w`=
zM2IO`WPLKG8L@>z794fiqH(Cc&iu#A5Adb^*x0~E*)dc)O9w%!C(v%Mt8tW!wQpmp
zu+VgbOMsIWP<3Wzo8=n54G0(+jx7mVH1;vA;7x!FA{X8j(2haFA5^fJfurjQYC$tQ
z?K-dAp*+^z?@3}+_+e_gV70j^q>3x-a#}f!gQ3af7%}Z<XLS?$lBqnSLjShouAds%
zwpaegXYc({yU+&Uo>C`BA9Qi7G2i#7iph@YP8f!(%NP6Ge}z1pmS~_=v>_A$E$+rP
z-QD{e&AqZ<pHTs2Vn<ojsGX)L+IhP|v$GAT&=3+9t7&p|hh1p|9@G_Bjys-ZVba8i
zk26f&x6K^R_WZkeT+>!9q3>SmB4{)FcpieZt$$^?Lvv=nykEUAXNNetPw{xn&g|**
zQ5A?4BMW_@v2<Z=Ya1Ikq|$I$o*lBX$r$rT2pZa?z=<nhUhM{gg8||dNu7=+i-y!{
zZ~0;!yS2XOMJ0&wUC!T-RwQw4WhkjJtH4xW!lAsD&W6Yako-cuNiqP>a*oTCz1NfG
z78WPhoGzK5a4f47JkeG^@W%)WiVv@;4B6|?7IJ0)6lKQxluD8EqS`ZO0CdsLUos5(
zusWQ(FHKiyQyJN~hTsvMBl%^JpLQC@@@MnAtZjLFzL)IXGjCyESihFP^104F(haY%
zXzh&+TOpaBhiuigdX){YS{#*A4bK-ddYZ_e_9n!-wrdJ%YL@C$1?Ii$BR|-NBvkQS
z4+D2?D7@(?4i=-?Pb+bBfq1XGRm08iX;mF0zUR&$<>-p=nn7^suM!LRx8Z{s4lGyD
zKuyV0)P6jLd~DtQJXG<z7<OzOMX0Z-tie^pfy;%hY@a1X+gk)HDo(s3KsEbfrT7<q
z#ShaQA67IiP<5o|-9~mfzIDDEEG6>r>I{wfi3#9S{pWACvRQT|_Q6WFGve)V#z<s{
z__NmWHnUqkOZ(!MCk&%UbQ-C2ui3^Ms*NZmwHqHpR)%JbaKxNSHv(sX_G}jVkAxyO
zYG!Oq$9|y+FxGfUmuW`^IT&q1JkggLyeV2y0i~Y7z(i~WO2X>R+A78G2F~0JLLOPz
zR+olEll2z*L)SK`%%nB})(|+jip4srPc7*o>jYv&{FBBdF6v@=?KwW<G+QH7l|DS;
zr=f9a`k9miDz#`CKqi1@ZOI<%B8=p)Fbiz8vs0|a1z{E(kDmh*rLMx~hUWWtF+Kbb
z=7r2_>b4M;We!-X$@{L2qkNjY7bNn{3XPIL>{nXw40;`h3-d_m39nzL&WY5G>Ygfw
z9_aBFow<Ki-{<Q%kB*w8(^dVGvi;zcLCH{zSqGDof)k4l*WA`-X%r)XfAtqBo?GK9
zt-~Iy=!4n6M1L4`fP=EmA+Tdm5Yn4-A*92q&vY7F9rvGhcm4HGw8o(lsFF+ZvU!j6
zSQ$u#2M)ASWo0Re2qac7tUH^G`9zFgUIgL0b<A^Z(Mro1Q^`X#Iv*!jslVQ}i``7K
zSKv(^vI+ZkQhd6im|PeGL7>jB(qiM!h^VyLT$?u8YB#~oO*<M^%!M?JNb}1g?AsF(
z@RbqT^pbDL>Zd?8LL^jI*nwAu(09DbzJ007yueYvg$z?O=cX@)4rO-KF=U?bz>#>M
z4s)KH1qNuSIOvfpWJY`G#1jlE`>On-Mo-83@I~|0(nWE@qPIP?1(6P=7%R2|wF8<O
zeKLDG66+NuecjYKd&~fy`4mmNn87?KDQDjLb!#mfQ)cx}59sO#p2b(MTe(tBqCWdr
z^NJ>Pa+lXSE#Kc<@DhI>sRjs0F$(W6DnIWneBAx)YsG;wz2<6!#P6i)Okgu${PAj4
zBZ9Z@dnR8Ni^7gg$Q`-D7c$(q#w`cbNOP|()0Gg6R)UXO3pMX`jw3j8aVvS5U}?uE
zt^~+xOZw@R&1Mkz3`WVusCh_rEYrT+Y$P+gD*xBW?RCA!)`#=2$xgEBB>#6;jvaQ^
zNze*AimwMUv!q_g=isWLKlmfkkc$@psf9V&=ck$3k#tj!ul%?odJL?JGdyRM4*fca
zo^|t~!7PlV>DfxM!(;^c>BwjlrwM1uFhU?@TVc#JmKv4nS>NkB+9zOs+B=od)CBHn
zs#&(>D+-MGn(=p-zR5g0u9N55{JG<PJ3+Y5wo}QogUiUPspZ6Nz0*dJis!$6j?E9M
zG=%5GkdhgD+~7$pD?^dAE2Fw%-Ch7?s!bgmTI|Vw?@R$9NO)I?rLynyd~OgNS0;_}
zWNMHky1hA5ay--v3C2Y`CzjW@Gm5WS%7QslLZvoFLNXMcOt@+ffJ+}|-U6^rP>om|
zjamWB^V)Oug<X?>D2<#gUFk`sLEf@KQ9O5?FJ5XPS@%Sp8ojEUNSnBkN}$#F|6%K#
zdW7M^CEK=b+qP}nw(YlV+qP}nwr$&ezDzQixj2{mAM8{;wN`Nz@h5W1Lt*NyFW4Jx
z`>5(<{>Bd1EA5MP-NG${E;(aAWYDp$mAq{IvK{FJ4$|179k$!)N@YcBR5JKN_9JZN
zk=(=07ywcp5`q%h5U!0bpp{UIB`IVYG__9`*9#S6<mpDR=AXn;`qW@V*Yu*vH?<J`
zbDrvsiHy8)vVa%krVEB$&fz+r_qi2RvHMyb{N+D0nBe0xTFvKD5gegXOKCa~&`0tK
zdUF89Cqh3+&CyZ%qr)Ye2l)iaw=blH4yF|ny!4e&kyFl#X>A~{siWZmI?_sV)x5Tx
zp*AICukB5xMhWr9Og+W4DL(ntvtbTKs$5!~%+XTcfxbvIu8zR$^Z<fMp{~}s^Uoa2
zhWpsMNhodRe_{=ZVOI%@rseZ}_dyny;bh(fo9(-UG4Q`JW4f6;0r_lw!0`A)^`wy-
zeRAYwlK1EPW;e4@WF?BM<02rMX5R{Y?K<yPh~SCnn`#wZ1Jp;C%u)_0dG4%mb(vf-
zzZ#E-+p7$o2G~tHl`-<L;^BU%(lSg4xub!naYw1tr?gu3k1JkJKh?Cxo^x_y6Ox$I
zY2B~2sgE%>aOh*UUupawV-INexez71@LomO8~2e_A+v3J!=KOOGtw=7;WeF8hZzJF
z8RCrenvZc)K}TX5QQ(o_M%QJ%uy3<N=lWP|xHh~a9V1B=LqsGq#U4nHbb#`BUVC87
z767ipA$NN-{5z(Ki-qKG+xp2(zVz@zy6kG-#u2XIS$-F#+4g?fRc-N8%jHfDAF3xF
z^{k0$fTvQX)xP2h#=EbU*gHuQF{IF17fJhF8fk;&4T;D=!CpcR$!RV%r|i)5XUA0H
zj}Z-$_{9>;b~(w8j_^5<rrf8vr@jWe7Ac%Ewq|el@_1Jw)i374Y%;wOq?^v9hk*)L
z_At8-3@bS1$0p53Yx|{0sW7^w*m0a)(?6${CYYbg6~UX{?huxhS%365$j>`4aTJh`
zpUSBZaEODEz|wGTMs!?xMX0@H(C-!!cK}N-3g<{=P@T0pls7kQzKL66TNJm4C=pwS
z|M0yjYXX3mpG^UJUtpxIx+soYx#!pV^Jr~!M~)x5;_lArFt*E&r`brmM))`Cw8Eo8
z!-(PLt@U=fwqHF-Z^N}RkU`LFZDmVPQy-v)?cEwcATzT74T)S@2hjzauRkdVZxlUu
zyhM8_?+o(8SQGDf#+M5A-IBR(J4N|In2$szh?*-%L7N%TGc79Prr#c&n;k8RKjaWP
zA||wU(vR2`RF1g=%!lkUxdPL@hTC~nm@h5(OUSQ5eiB6GZWj{iB2rbRhN;?lHVdLU
zM9u3O-5=1~ZB)H|KhhL{A_!)uGclmFWG8C36cRt1SzrzazxXsy|MhKL1c?@%3E*m}
zvqK>y<d{J<xRvlM>w~O?WmE*7m1-WGi3nOx>k$U+wVb~|rzU2od+&^N90hqrjzUEy
z(fU*z9L7!NyA7%dlEzP5CdnK|VEEV(jR_HEcZKU|#SQk+=G*ixC9y?s*$VJ(#<STs
ziZ0EVPcYV)+YxPYJN0(Rp2S3uWYj(i+P>bIG{RJpr?Ux8TC_1-{P^)wA|LU<yve*@
zr~zLD7b8}5er|U`*(;8tr2VK;C^F3z8bs1UEL1wV=2?#jG&;AG3bBKHlNoDv4oe_n
zInhCC3=UFN+Di=5S~Ve>>!T#-B*(z!4>MzGp6()O2y4Y(ql$HJQq_)g=edWwTuNbB
z*YW5C0{7#Qgt@W1-Ww!MXWMw@4?L~B_gYCdDG)q4ojA}bDoW@BcBrVJ`LVZALAJOu
zRIMqaKacoQpK3cxbH=iZxC}WThDo%M47pw@(;PZktgG*zyLd`?Uti5ov=6(J&y9<s
z<!vVA8h4%NrkOr-wfj}WKg9dAckjp~H?pcb=4hB9pkZ{Q)>|fB8&DC^LFKzlq?4RZ
zmr9CS4z5w+KdlTdAqzT|VA>uoBnDr-F=9%^7beEUYY@$*ff#T4;cv$`7(%vWxBj<m
z<+UndUglTIsR$O0_{+wdj1yKSfk3;3&*xl|nUP+E{xoa@sOfV74<{|Ca5*`#?mubs
z-pk#Uj44Q`<t`^uz8<Bb%2F7p28q|mMi$6TXkOm?mD#A1bCZIH!*McBz(9Ou2}Frr
zc8-+-65fh}cKc048tzh3zGE|&y`a}J1b;HdsS?fu&TUql;SMTlqwGh6ie$$dkAiYZ
z1-JCFJPQ_po*Xl)B4teY<ZmucOlyr2h-_lLAtUz?<G9h_Mpt?GWt1I_8}c+8YjQZT
z54m9B%`p*;j2rrE)JAbt?}>6-SFrxjo?Zj#7L$)mu4v}+3>FSPKjj9Q#4t?wm8y5|
zaj1-$MrQoa5@TGc_X?81!arHeC`rDE#{}NV>iXf=D*xC;UI>xX!0TNn9eP2*9@6G6
zCUMwQF<O}{A5AbarM=E2Q9`iEQ6WyZb{eL^tXdwIn)m=H*&15lI<Wwmz~a#y2*f+s
zA`NjOp|<mey)0rue9bHN$(F0&p#+?FP&8)*e->~qax>}~gnQAFJd!!EN#Xt*y5w9M
z3j_8tvSe~D0XH(qQ_A@tL90}1a#-f?M}37$a({6kt+xD}lAd0eOZfq3oTMI`9zLaX
zNhlfJUV;AVd=w**Ti-`t3P2b{+c%OodshJ9)pF#Urw~-FxCkwqV)mkELQX5%&Y#vP
zFC#4T;hxzP$Jb2Ul#H-n7qK`6jFY%|yyiRHz2+QTN!CHu)j>bA(Sr4^pn`Xv%U_;g
zRt8{W%Q6nb;jHY5zL&Koo#H3KX*&bakgt9)JjN|3&Pb<6p;Xa;V*^LlUUIxM&aCr^
z>L9y#AT?shKbYsDi&Qls@Rk2n2P=)nVD^gpIw*<b<XzP`K<={7rOWiTI(S*EjY2NB
zvF)6=L^ff(!<$Eth<MfXllv;XBVxU5G@hpbH)Ipot-zu)v<*nPu<0Qsk<5GpBb&AC
zs|yh@6JzY1mep?VzZ%9Qh*LfbXgH_hUNflsPjH&}ndL5h=6-ZoMsd@Sdk7fW+S#3s
zEv-$F6!Fw%etW^?7Jl-LuFIZ+O<8T~^*Eeh2+exC|H*Yr#_Vq<7i8JA^@<L#C&mnT
zKk_`|vo;1d$e$lX&ZvN+tu?AcTT!PQY6~9SPZPglYa>m$2f*|1%dD*$y)<@LhfCS#
zhJ3}V4Q3M3vewmsS3+My0;g@2h`}>}9^=ON)G(t4WLQjNOcaYa1*R{$)O)CR3q>H`
zQzDXb`#nFi!JnxMquBm8d?LsbVbFvt;cg$Z3F*h`P|jT;fkd~!z=0&;;K+gMEh2qm
z@o5&xIm@OYgxN8kP|^3trv;taDl*8K0p{r!HKpNT3)IO*XEhvsB@b{o$4#UwLI}#{
zM#hB+(1V#9>dPw=d39m3TtOCV?HH*c7z57vmf)1^F=wU+%SU!1>3wM$mm?&!?&G%T
zbJ`I|sLj^{!t5IpEq++MMD^_piiS2%T>n2Ou$NVN0_^wQwtC4o&5Qvw(0IZc6SCyS
z=rd0I5(^Pc!Ei1k#_h1K)qY&Q0=Pa{WK8ErR7l!|Njbk@YLnd9r7veM^J{CSxQ%WW
zkGNBmwZw|m8kt)trNbuKXEd&OJ4Sn#bb)+Pn6QeWH^fhE2@UT7-jdhImq763sD(%P
z#D1|Z_l=P_JJxAy8y(na)b0pm7?!cqs2=)_5`p>iiDbjanohl-EG2>K4^^D$i-iI*
z5Y`?k$({8Z){H%@bGQ|pC9EW4E_IUY0U}}?o1(B*gdeTDujM`Zn?};O^r>|#2*2-~
zQq|G9$7#qo@dl3S9w2B_20y)&6f@?44r?rq#%63Sps~f|-6^dH76fdC?9!lMZr-N0
zB-&2U4SeT=eu(9*Pbv4>q*xQ64}rCn#qOKK=O=)n;f12Hw3oecwvWF~GkQSo=kJ5m
zw9pWH)lXnARDFRVEp1p}Pr?+nq>Ph~m2E18l>^co|Lf&32Xf7G#Or2CS$MkPPRQc(
zc8}HdVjfn#bg~_fdO?f(-OS<*9-7BA&}=|JBXt`f^wP%{m3pD&CR*AF5lRXde_&P?
zAq{Rj&7JqWtk-jpYpG1UiM~wfV5&n}HvU2wn?|CYl!oNuPLsTy^8=l51k6zwxm=|A
z(W-|cY(d#_j%B(2rzyR`-5O%(WFy9vo;UqgdGB$e$Owk%$4#JJWV)@T^*`$^6=Qux
z3`7v0%|HjKxoV@I)q?U9995|g*d-#aBhW!iz+dI!a$F+Bw)9}zoIwx#54?ovn6n*~
zB4*A*Dh;5wDI6ss7O`(TUhx^>upPZ_pcyh#Ks(F(AwMEF{e6#hY~qB*4SP<nQ3hA4
zeP!~!#}mYvA6r`+jZn3-TAf9{-Vgb(X{jf7xlgK9PHhY!&Z?R8Kz<%)&|&>K4x=+4
zCsu4Kqkg1V;lb|=M$C*#IA|M4N%kIhy5tu{h^DnV9p64aPQMM0tMXaQ;`N97Xr-jZ
zA6z!8)$L-q{y*L~ZLq<o#j$j{eTJ}Xw`dR#3}>3Gm7|(TJhJ|+SjHK(S60JLub+om
ztIWtYoY=r?#s89pJv!iU*f!)%$9K;&m|@TS$VhKm$02eq(bC~l?hOMc!6QV#MNq3}
zs{XL7WAv=a;4!?2Q#oejX*WK}8-|n<CK8K~u_e_^lowfi`8ulvP3bV`N@0!xVHuC)
zutTszq+fwqB$4fuI{Rqr@vm9~<dLW{g|EP7e%KPvlEm&97E_JFIJrz@-wb32AT1OV
zNbtM;>19PcaB5Cj?-W<^AP2p8J;{Sa>el01`jIpF`Cm&gy21u9N_1rv9qQPbNQHH4
zerfX}S2R|e!uw*dFILZwvno-O{uy{?9oeNP+~_3-awQ<99?@HSOUt?*`4lPms+ac?
z*m<RNFrbUss14y|Blyi3K9J!2nfZm!qa!d#Ahxp}E~->N0y68M(Cc78nH5OJLSuN^
zg$QR$75~#TzRM22K<Q_1t;;}CPMkx%+tmwR*N%%;X)19twC;BaLcc}B7V&KIjm6|`
zE3A1)8Qx>U3~MQSMpeo&d9g-$stzl1S)qx>QS*I!CSxWeeukAN+TJa;H(QZ?`at&4
z&n<jsyDS2$$Y7l;WU^V;(_?^Q5W$rlIm&S!{l4wYB7LvfQ6TP!UQW?Aiq5_u9Jv%v
z&4fC;D!KcRfs@`(nqab7sH)9Q8mjD~n8{Tr6ryJyMS|f>yV=SLhxMrp)%k&6JJjNc
z0)Jo*!a}A2mapIp`+u^GhYUl^9IGYA+NpD|_Igez#aQC_om(7fr(-tgO?+&TC8tnY
z3@GLuz;V>`^VgmPxezHu?od<hFsu3xhp;bdQjl>dqmZa}J)fRXvPt&pzBNcS2KtMo
zY4RX|1s50+S7-WzE0`q_3_5yWJmMDHAl+Bbj=sp|wi`GlmErF{<Pc>+5QOC&vTsNg
zbd2L$Eobn5ew->;vjQCfD2}dn;{hLw$HLi@JgAg7OaoKF@aL~-&teuT@MqH8#hOGT
z@jF-ZnhbRkZZ|>fEdwC~-e^$T|1!H(-D+sOG)3YjRh|gt%zxCQs0tQlyop*e_ABV;
zwFAOij)ud7Q-C;|^TuiH>Ont+YtwXJ#XveA%lCP3C$b(_^isC|$yjnaS|o`f?;bvP
zgqIWmtepY8LPg9blGL)nU3$1d=U17+oI#HyLR$tsP!T1F6%+_gtWVqHd2II7S2b@H
zOz?WdU$2TtZkQt6te`Gh%_o}B4y%dx4dGmj&|cxWmMU(^?I3|mfKa|?#(_b8vejw)
z?BcB`=(YQ@a|PKvA#LY`m=C2JrL}bxF~#7h(luyb>Sh?W5TP%&sD|iGcvU;xQ1o5d
zyZSb@F`Jd<<~cfMkH~){v+wF|s9p4wNkQH7^1t@6k9zj&SJZc6T>d-WB3={B__Rb3
z1lpL>_%2Ii9E9Ia75N&ZqH~wQw|BJ)UT4rx(``6`0<%hV$Fyhj;@%6-R6yb8H1hnC
zGRXqataVQEg<~pqj&-x`YE^eHPBz8vIltueN=F3dkw#k}Cpqm*+S#CjkrstJF^91X
z1e8c$b**=bfyKz=GE=4Tb|e{GEb(6%U@1H*Z*hP1Ui`DT|49b5U*CIhd$qH;P+ROq
zA25ud&anCCdU3!rCv@*)s6??M8fzt!5hqpX&XoD!#~YK&OYui%m#;Vy>|b6vDXe+s
z^ho>st~Z21$aInoxLz=XKOpZY{db8%&}uGuv8`rY;!~ZKrQR6$YN|h$u!0nA7p?FN
zd*xdH_E@{qbr*4w?n|1s3!=@hf!8<N0(=J{aEFIE0x0-KQ|yj-mLx@(TZ=DB{cH^%
z(=f1yWvY{}DKR8Et0ZBoT=DM3b-0hnQM(<*o28IYjAyinaFgxqT5qZ5@5^yQ^Uu9d
z?2Yytg$n>7v8=u!JW#2swtm1338RWZKHO=r>M@HDu8hA+hp7-{Fm2LmD-gFT!PE>S
zBH@jZ&qz<W7Ij@p$px9}PiU4{Auzn)$mIrq7K3M|lUe6y43lMS5bH&0PJUz4^zyey
zY~nzr1UF-kR7$-tZc34(g{MD!$3d0KVm)nvxo{wO)!W;Q-7Xl(Hx4&gj}FJqPPtqK
zCCnr%=-gYx0x%+QDmJ!nsFn4}rA6Jpz&_TJV7F^=MQmCcPMjePgiXP0leZ7kBNLh5
zK3xKn_jr|sK;3A*``SRZUtx8!ngd&c?nwud8-r#VS%5@7YWIFjy59|f@xAo=EmDnE
z@}|C^=rXu5aRr8FXtI^=DHBjxz6ky!>3hGfq7ify+4LVUM3$8X^`wdH5eAJ(-1o~i
zTJMvPP4!R=cx9v{TyS|qp*JYmlhopX!8qZ1gktEM!LXfL$Q#cTON$>XZ;5N0iFhqA
zmf<QxwiMx;1bua)Ci1x&@m4Cq!Fus5E5mTnS$~iSdVS1DBC2E+73WZirY!FyB3Ni)
z1>2h<jmk_CV;$}!N1UHC2KBRQCQA(Kza=KpJ8Jfqh5Y+@s!(^pPl*4%I?#sB{oR0h
z)k3q)5~+oE-32k@2F0BS692x#$Z^yi&{o0vioJO2fnowJCJnY$7!V9;{_z5M%L8zy
zpZHi_<E5dGG=hIsIO<gIme#yHr*?>7*U{Uum4h9CyR`)D?6-DxR;zfVegMh`EWnkd
zH<_&ik#94T+6GvGo-+H*bI*DwNt&UMw?fOGYz{4W&3yH<q|XD5uzaJaDB+Uo2Kz3{
z5G#kj>7odd0&Nlj<9xoS(C9H_xbu>+ACM0xBV4X4)HYO4fTO_69@xk+(co4UwjKZg
zEWu#mhn)TVoY&Dw9IXo^jMW;co!*)3@6o)dez**;R>CW1H{offYiIfOFo912XC1}X
z2h!)2SZFoy=x1O#F*#lLZ<8zox_VXj^cg1^(10(>Y2*H*C3o_jh+Gay`+*wtJ`Tqx
zim@V*hL6i_efz?zW+RFqDGbUG{?87JuQiwEC7Z3`5J@P;4&2g%uQ}}qhBz}V#IuQ3
zq*@)nyt$<z{Tkh_^QGHK-ZR75!H}JBaCH~4!MVBYqg1f^$T<e>u_i3uDpD?;y9Uce
z9|nVS^h8=Q{vLd$OV^yu52k-;Y$WWT6{d(L2+THV93re^tCXKr6_AROke3%LJV}Op
zFnNya72zB0v&!HW86%5$Sk`f|ZQb(5B`G8Dn&Qw?Z1nhJ58Nr$N8}#R2GOCFTK}JT
zin^NZA>5lixp+p5P`wiISjqv|{zY(UEPd}D!~x!Z0KB9@)A7DFLSod7M%{-L6I0=g
zCMEj-;mf733^^SY{4@yom22a8WSYP|x*t?B8-ya)cehajZ2Ev$fxs+CU7>;*0q4uk
zlAYmNOOIJPZ~AC$1-i?*5V~mcwX=I2PivSigt~Sv9BD82T&yl-hwNv&!{)`Jccj_v
z)$EZ2ms_;UwiiCCgtZ|>B!s2loN)6G&#=ubd1PHvir8F=&5CD+i2Mcl9wlmwq<YG^
zZ0asHpRf={es5VF!;Z#1Ab8Fk5s$RT(zLmn&E-us%}iFb>en*$9pYfP=UwaF!_`qz
z?!>AS(dewAAiHTF%k=`JD-rF&Hxb$bL(4e8PP!vYOS$w=#QEUxVJN`fM9w?d<j?Iz
zyOWB+Te`!S4)|!zb=Z(vQyEF_a(?u-iJCga{;A?E4L_pqi%f@u27;Jo0%qV;eaQ$x
znri7;KL~s3of3tBNa%=`mh2GcV}|qoUjqvm$pS--ah})rN5o}?+2H&>vz%Bz8o3hs
ze$1bWKrn4XmL_~^`)YLX{YHKLy|b^$M|dP$#)_Skd;rPwgmRey$7XE<`Ib?6My68t
z8{Y>S`cA`Go~9RY`IYcF#Y;3tXI+-no4D0bJRq)_buFnhW^If8oyJ4YkvCG&`L9Yy
zpw9zlFA~U2o_sxeRjMPar4WsQ;G_nV$5#>On~NOPNNlRAF};$)s@{Tc1?+nWI3+e0
zbf?XXliz#nluzDgKXr#Q){jlL@Z+{?^wXe&A<4!qH{Zjp3WDpl#6WLL;WNDb1Bw%`
z(&ncu3JvAc0EI9|q#E<$<=R9)seHoT(KK<wkg8FhXHeew8Z^~Mm}9P>gGsO*gJrk=
z5FK3u+-HWz%@nsGE3Lg0?1%G5LRTBaH9mP1GH#S_%l;k+KIEHi{g=>Us}%=}Ose+F
zlyYeA%c1m^%auEQ3848VQtS(>CN<j6_Ol!j8{KgQm^=^Fxo~CSOXCk27>)`!F^91n
z7`uwD3lsa{3|s-mnTb0OlQ-t3^WHXLfxUb#gPFLZYu?qC-haNgsiCs;S-`VsWe%bQ
zo^jGz{+@$4PJyaK1-<bLSkGmmi#oXDJ3_T0h?g>`om)NlsQa0#9wwd<n2G?(Qq9c0
z4y=*Q^!ma(+L^&&sIQk(TbuTZr$veNv0Y$b3R}$uMErR+iqT9EnUhp4v$3Pga2PQP
zVK2D~5lzMPHW!<1<w_j!ts5F;_bNVGfn4!*(u&I5_M1x62k?+Q<?_43U0I8RGQ+_Y
zJd?KM?8G>=`=L4**h0n_LxtDcbfLikP6ccsCkjML4<;0qRPpkp?h6MLPZm56f5jU0
zK2K3H{1EKd@=p3<$*eA|)2+n^KmCE=0ldCXLn^sfBLVyn)LymX_6hB+gn?@{*CD7k
zjhi0AB0k+VTnGY=S$QaiSB-_i-MYXAMrqYmuU&$qq_5_@;q}bJF#6nIDN%cQc~Gl^
z`L9j6iRD6%eE||w-Cn!C`C)xgJPC9P0N+xl9(-S0q18Tz&TDGrgm=k*jLft?d5r;Z
z<eBUFun>5<7gv~{TWyLtnSiNq|IPf~ChjMrzdMGhh7GNLu}2HxhOJUvU2Drnx31(0
z;pqRF^s1l{Z=La>Q+=IJSV}kJL>yBU<`LHSM%N4<bU1ca8#hJGXN(~@oo<X@ZRHUh
z>G!#nV3$9B^{QIF_^qn{FiFzJl^VWJ$kf@kNPew!t_*&l$(Z!awr|y@kxY|OM)fIa
zSc$)Q*Yw&}bv-*gbiAF3X6;*$hS;6f!i~5D$yoP|E2E|eN5Vkrp?_Na?9)6|mAA~j
zF!k}e&^sJ*SZ_$pYhY&S@C|%*`5&}(-Idacu9zxqHl3)W?JGHcIr%-lw!m=jU^U*N
z4$hbC?tiIJrL`ivJ@c9lOv1`M;vuae8*k=nZpD~}!ZcaP6jl7;om*9Wl6u%O#6CSK
zG=sNeD6GqO#vyBdJ!|6&_5?<4*L+*j+EYtR=Klxiz?AcntuPfHYwF(aI#r^TRNNvb
zt2u=*gF9kSDL*L_x|E#l@-+qCXFQODGJ3TvwHa%x*CZQoSOP>W8jNJ$3^<h+w*7)$
zg^@K*Wt-t*6z*MLk7HT`!vUp=Cvr^jQHQ*R!LDYWHeti0jbPhs`<@f=QUsPLL72B-
zqb1Dya<R}@rj|K9Ks?+^LPYZiG&pU7!XBR=EoUg~DUzh^wN@JZHmzfp<#THgnzx$W
z*!I4gdqPrr&i-3-{sR09!5ZCrIirZ8)Cpr47h*?>Vm42+-9f{4k!51oZGNUJWBY;M
zrNpIwALt7oFh3s^E@tgR+X|B*A?jJGDlU!2yUVKl=N|R?vN3m1=TI25ro-683X9ad
z&0Hy6-(=X&7>~oW?posZpej3x*+CXld%B=&6-mLpIp=NEc$IYn!!%5ZKAC%dkefEx
zev$RXMI}0?lI%|`!R_5Q7AM_j)BOGm3RS=Ltg#Lr{>+~lwK{rzNz2|ZKG=kKFH@JQ
zm(xiGLmSIyG9iQNVh*AL$J8rFDQ92x{b{N#z2=kP!0-kA*2Xie@LyX_$hdDns(|f_
zbe32J&ljYqkY-gq;V)8sV>g^F=p=qvPx*Px^Sa*aN!$1MbEbHtGnBsE9LrV>W50t#
ziKgQX+p|XK9-{-?5N~FoZgu`*l!f%U3#tE#6|<`efvO4qZL<%RTh!SkU!dko>ngoS
zZ4$ap7JZ>QSTv*TD>*dVHgu;*Q-rs>gz6zO(d<S0Z`Lg;Pbim@A{9AE$qUafvQF!&
zD!zQ;F^~4tex-*V!Tz5lA2aq+-F&EDZ@JB}5wgh)vzuI;QK0n`4pchUupV2*&Ire;
zV;hrrg}asf<7Bhpx@wRgzL@xzu6MdWOGb#evmw!U@I}`l@r&@fhQs%J#YcK%wYk}X
zboY;skTEEun@w@q<;!FG_vau_vO#l+V*+zth+oZc-SS0M46Lp@-M=iK_?FD0$4?r@
zY1|d*zau)~a_wpmwM5;X=_#es8_zKQUL>u|Lebk+okIH(51u=cqZK^EODQLl^{39-
zzS(vQkR(~neUI<A%`q#a&MY}P#*~{ju1>a`BgF<jgrEcYvkpQzJB4FchB-O%vFL&U
z;YGr>iWHuIS16TbAN|=pH^y)=CVZSiaR9{Q3Tkvz%TM8Q4q!4xs0ZE_u<Z^RKxWIQ
zkyUO8pWp46CMplS@2>0Y%&298_ELXxiHl+;%V>M~<047G;Jj_;Jl!Z0pPZ5=1e3j8
z@c9$H^<7q(WJzdnL@SG>S{Btonus3aviY_S%!5-8zbElRUo1#P=t&!%y1J>rx+Kka
zaV0e1WV;TNCm+9{6>ld;b4eO7dp8K@-Vw4snlbzQZ1igD^nB5%rNVrYohu{?vET>{
z1|SVYZ$*`iAMRdSuD41oSkqspBzdI#QL1g{Y3brcu#LpN)D-GToq9vLN1_*<>x-Nz
zLrRrMN+pD0V9Ws_&c%iF^UDCvEMMQ&{27^!AP+xb`kjnk`K7*-nn7VDEPB_17c~t;
z+<JU*+JbR5F9M0{jlVd;R=oy`t?gOi6%|!)%i3^kUVgU6pBz*)-}V*2-dGr#FSPPK
zpoMS74DjgE*)bPI@tvGtg0O!ZUsv-F6%sR_buzO({1j5)dg+@)YceZkseB<^gS0P!
zMP-9fRVL&IuD-skfIK5oXOc75>?A^G;6!TQeYuq9YsbWgDjS>hLDy6opw!xAANePn
z-J!@P<xzSeM`0B1is~WX!d;Qu(QwKdPNU}$@`gA&^<qa|pF2SYp{7{`Z>)<9CBQHy
zR7^dv7k)Fpl$ggFmQvj%P7y8*xci9(_?AnRc*{AD7Ie1szW3Fx^VohWa=C21A43vh
zYZALxb1z1lLB|{dlY+5F_;q5pd&?02#QJ!dAIK+k)#K+VRJJ5zqB(Zdm0rliGQC4N
zPSz=N-XkoA=@T6&_dhmmE8R#WORuM--JLHHNBy%Lw$}1gOZ#+~JQH66<&6u=BN_!v
zFHcOU*ivD}R)aKQb`Ud2?7X6HoTU514#}fxTl7rnyTtU!tlW*xR3m%%)&bj%-!;CZ
zzW8Hbm2*ufSiCz6TJEy>2S1JOLm^Z~+mCD=YjQXkKzGD19ZgsK9J_rcf}N{)=`<?(
z@0*r)lFJ$9t=6+2BF(cibtdCZhE+dBmR%#yM^Sz^pvLqkDt8{(%Hw|aQK9VBmHqG@
z3lH~8HVg{7<a#dB$LZrcHYC;yGr+M-Zk}kx3(hv|QCIlMDCt%LkVKQ#<JH}!54dFy
z<tut0Lw6aRYBx?Mbh~IqDX%9t*^aM?rPelB7^<(5A{vv2#x*LP%?anS6dA)uh8=@<
zMqMc3(0Sl@eSM>PQW1I4_Im-`>fD;;HeOQDM>d*7&kBC#NqBxyY@(aB!2rz<?Ui5n
zw<>Gk73VZT;WZc+42MCA!+laZsgOkUe^S+e^L!V2TLp<3pCDIH6_CB|6Pl@~n=St&
zg-p%q?FsO1@n^nHmR_RUQQ3G{`HN|n-%6KfuHVI*V(V}=n#dvtDl>>O2e%?zuZ-R~
zEnnLg)hq;=#r2m;SnGA*<4Vezo7-Fa&&JmxP07&hvpI9Jz+iK$_c>v`N(jge6P7nX
zb}7w&&!G%cn-aQAadB3PZ-ut{DIYrZDXF1<{6FvOH#isOwUe)>c{XeJ5uvGCs_jq7
zJt2aR3`;hr-EH6W=<k8ZO;yT86@;~KZQEY*HNL5?0=~N%e;L}|)=QJvPqtbV9&xy&
zSdWqgp~%;}R2O`Hu!1=h@C8ewY0V*_$RwCR1`#+ZJut-c8`>)Z5hD=qr)(JsRi76E
z2beID*qwK*tX;SEXAdAYQQ=Wj?~<sz>4{|Q`gNVIgKn)~=4cNHz<d4D?^QF2$Oq+@
zzxD2((PYQAR4C9gsF6x6Ov!N^VToJ4X~{BbVG4@PPwck>bevzyfqQ!`_n1EYATH85
zZs~~paf>ssC^zwO8A6=?!?2kshdDDr_XgVz7cYPoz2HU!o8BXV(TyvM&rnX2hl{)a
z@LYN>3Pxd{Lh2#OMK^v4U2i9C|3Hwx8q5r@V^muZ*S~yRWGk?g2&in$&h4Knqni}S
zro7vVSIKxUySB7~d(%Pg6bEf`A3xMFuPRR4!G_j9Q@|^gQs4Q`?TR_DGZG1JUHeV6
zRnrfDL_z(C&S`%nLqL`mypf>4fA%VdhM1R_PNF?4LJF{*42_&7s51+er=6oO7*>C?
z7zYE6cpKXZ>uY914%IGtIk737Lk$HIq4V^bDTdkWXO6FPq8i_|=UyKW@d>N-JEU$<
zSn#wWpD#jfP_h=6?+&s&bg8za<+p>0>X0ADgpQA%m^g2lSDEJZT)&v?a^OXlYCAGN
z<4webC~wuGl1XBP1!Ea8wn+%3^lX!#fOR8&jkxd<hdt-wKNO_f73NsbPHyvN;e=M*
z*$*2g`-FYr*&GK!5l-F<WmhO5^OD!9ARi((&t#+RdTJV4W=)RwSp#CT=Z7_Nim3-J
zvV&}WhO7%m8h=C20;G+oC#$rqNn7eL1-~9~0i7?_e;f=oxu+EHXD8W7{TFZ1sR{}|
z<UBiiX5`sry#l-4S#5{)7Ge1~wl5&1IBsCO<9M?dxEM=~>o46olKRY&(hN2cSVM>~
zJ!<#24}`+J#*-xN_}+)o130!MfVQJ)eS`=lES=ml85tsk@Yh(n0gHMXbZ<jr{J3dS
zZPejbOXzS8%<VkbAeM-^9B0L;@^n)|%M_c&BzHYv8o#4UJf6F+`({Cl!Cab2$i6e2
zgCXM1WW-4~J#F)A@9jYTc;whq8r3+S{=P&JnXEbSMsCPh{iQ%<?V8mBmEF<;UR@*s
z+=X)~`J%hHOuw?Mmy0ZXD4wjs$uQ`V!Zaj(nyoN)#WQc<-^BS68N|Mb$5X!mBG2w)
ziNu1(`02law$R=<Gk5s6TwJr5QYbKj(f81O+aRW44ZvKzsf>lTm8^OpKi2aY(EBp4
z5;!JRd^AEuZDR@Y*DuH*`zy5YOcq5!8$9Z8R)6P8E4mKCni6Y)md2#~RjJO{pdf4Z
zPW+NV<hrx3I{`EIQ2PEvVx<;8DXsHw#_8%;*M6|VS}gI3lGUsiWD`8NhGyEg0IlA4
zyNRm4ph@L7-h6#HIrVZnIOc|+<96MlSy4;bQ_@mp2lH!G_^^(l?(;9B1j$JrSZorA
z6_YVR{0=qiDMwT%VJ_1(!o&z)#B!yAZ)r2(xpP<I8`NH8VRIm7W7q$yqI5OO;%qKx
z#WU-|cG`5jc_wwrIRL<T9ADm}C-i1vYxKf2PX1V0!x4TH<bO`~_cZ|{>wA4cMrh4f
z*$;VxL~1L<jb=6fMuWzl6>qQTvr7Q`S6kyvCqd7RQ>{RErh(%&)~Md{vakIKkL8`<
zk&;h}UaP{Au`fL*(la2QEe<C0<({bzxeJ(<SBAo5cTPWutaT)|F&rEM7KTB2HOm4@
zaY)WPm19@sWI7)2fs?_N)GAwVMT&2)xc<*U^BOBd)`pML^_<@iELE?Bk#%qbk%Sf4
zmz2BNt0Et%D%!g$J9KB4)XERQ)Aal~0X(AQ5{vh5{#o7>E4sa90w+glM)}Ag!%;pc
z;DbLrFX_*7b{#8>$hoxR{)9du75@@^FS@-u5N{SjZ+KJ+-W0biUaiDwkz-Y{Xg7t=
z@In~xyQtueyKN@b{M%!;fOO7ZUHU+SMXizxXyiv$^iD6I&kC|To!GLTorG+53iA^v
zm$zFkSo#@yEVqkvBL_Gc$GyzgW2-fjwj;W{zT1Q<k9**4F<`Vcp*3(1TBoNKw2&#w
zMoIuj+a1K3ESEl~gjTf1-g`(j5l(TzCeAhYpPt;6A1MakN;mHSOcw-xHZ9@&A;%gv
zLQwV)ngt99JG|Ws+V^Gk9!HH>>+ldPoaW)tHINHE8fGDKaeSSh;%Q88X`_$h{v>L!
z)-|ekaC}+ESXmcF-~Q$M!nOfk+Z~7J#LtKSx$wY{hBr}bw}h#?)cMwJ^Q5b^X<K>(
z&jt$t$aZtZp%C`Q%%C~Rm2G2vVr+ZJ8)|(Q%p~bIGBx}L8=zc8Bb==kJn8|zgA&f#
zV<C()f5|J#I;%{CXj1H@Qg+7+Y-E{HD0h-G&;70UB1*U~p><61ainz3fO63`0Z_9=
z5>?^8@wFDf-tJB~zhX<?hY#MC&t%66S=bnwuU5C|5*dzx5exxpjNpxp6>^{#ohd=e
zC!b-(AvA9oWZiBsm2k|)L$Q|6L4YG8%&je2N-5)3zsA64Vh_;0y*`pV{ri6))_bZx
z#D4n+HqWuxFgMQ3+qBPY0eV%E8wc}d&R}INhqIz1naD<sQo?7s+cVq6`VipOKp*5F
z`(I~Je%v*h3$YVFF%}?LQz36@tc;re7_IRF)+jVcQ{2jlL5{h9EO6C0pA>jZ-_8~I
ze~+u?2^tUfmOVm-8V8-Xeuwo!Ji5Pjh9~0efm_%hUxM-{`cJ@cC0_>iYIGYQFOY+A
zgd^`}-e2QKW(FlUOqld4oFCwh-%f>D?8QA>&~ouq%+fXZ@!Y2U?9R(I;?;DgNjaI+
zvgW>61PiQ(VSUb|_+HNYkkh4nQD$e9Kr``e?SWF1*-+k`%I5({b6{*X0j6EPE-V`7
zSRoO@pPjzGzb0as3V^?<MdRT|C<93GTKgO}HkGr>h?9oK%bld<oj$cD1A&b+()J$k
zd@sn4pW8Wyu<}pn3LeJXGzW@|FP$bb<RHAyHH!VKnw?sj<{`{hzd%Pr&%JQ)7{hy_
z2C~oz(<XqS|9a^`;u?2OtZ;~*oM)HwA{id)%XnLG;Wp%aMq)aT7w;t2n64XO`)y)i
z=AFh@=GYl=3@L@MT;W9s(dQkcGsyD*qHd|E?Sb->uMD=(>1Phkw(xW$XVd;3yGOk3
zFn932G1-f9Z-6em#3dnI0e0{3OY>JNyFhEBjyC_41+UI<0A(R{3p@2#dRSDiee*u6
z$o`YfLsr}JW9Zv=xIQQNw1I{!<iic3T;0Au{Gn4KuHI4*U3T#yUki!`8tRI$SViGm
z6o**++`>dDp>8<?AYZ`I1V+-S{L00{QHbE$tvzFObu9{}%e?Mi*%zGjNInlJNdh=W
z`^G2?5Qp=l=IjSD$s@wtn*0tRoshP1dL&iyCgqkPTHGNkJf67ALCIf>!j@D%RJu+(
zDcZ+<+Z`Q52>e;L;x6|Jov!=F`DT%FA-zT|VU+3URMkycM#OS=67((AnbRt6U!zwJ
z&o<*mm@I#{9PcP%TjglMc=22Un5ZgE6wvkpP;AHkZ88OFvfWKO{<5cYFhlMd6Bm#w
zsvn);ym?HYe^pKrkiEpXBT)_(6(@+QkSV|kLOJ{^RYSOc$@CP75xJn#vcxyVgN9TK
zlnzJtE*F$-W&yau)<rH11p3)l3G~T@=x-ZY;g3`0(#6tGyMADs%R#7nl>XD+PT*Gc
zNF@??lo%8CttV_S^COqfkClp+x)N9oR`33=9)2+KZDh_N?R>v04~qlZZ%*IM#?nUw
z7H?}sLb28pjA6U!9gtB{>qv%&065(t1HeA`k+8*SL=KT@L3FJTpZ<EiZMO9uYHH*Y
znvee#r2T2^IHs8y!$^%A4inv=5GI-)?AgP#LDN(fT7RNTr4=OPv;L2Qg}&WOg;3Av
zM~DmC6`*3&(PZslkw6Fe)#k@cqiI3di2&cwG-@0Sg8biHCvDajr{Gx|Y-2*{b)PJS
zQxYiB+IQ_nEm<enB6Tc~8S6~XBQ;cPAJVNmKZ7o;1a^|H_wXwT(^(vwzE$Yz`eaaB
zd0#*Xd9*^xYxoiUs7kss+@bN&i3*DI>5q!uk+6PXQRhqswu=(cHmyN5(<G>(``x0!
z{XqvMpXCfjtJuOhI?SlnoX?gbda(#n=edKI-1&I`ngLliyzlHO16}uQJ_4kz1mZYC
z9mlivWQ5mT*ZbV_d%qDXyNgpkhXXY`*y*i;Af+hV1%qGk*q%nMyCwUy_U!^uO_F^_
zX-DYNTSrMX%#fE^QNk0)zEJ}osLDHq<#16Uh@E5i?x4gv8O{`wkq7@KrI_!2Vki-`
z2TDm9Et$s-t%@cWP2@11?h&~p`772=D?)IIMx;uP6#)m*cz1#n#(;%Wynh3r>-Cj>
z^N~PIeUH<c#6gKFvco$_$i6z@gxOvrC=q2-t(<?NPBR&%O3lB}4h)yxAeDEQ<5N5D
z#7OxH+VMC%X7Hr21iY@I=fg9>;0{wD>})EAyT|@uI}~8kK#Nmc*j+sq-<inLM`XDw
zOkVvso|)yq4t;vBxyX1_s-}yk`cSN@PaobbcDkd<2tu)!mJ=9zE<X_eMv5vXFADxz
z)c;O~E5tBN)F~N+_@d}Ykj;aTwbD^|+A8t@0MU|wgB!|{0*`mS&}=zC08$%5NAXnV
zjaq7hO2ahLl$B^4RFpD3@?8yWm>~mG9GVXBF?sc*_I*SRFFi<<fzlD!b>rT+iS>3a
z2p+}6_-1}L!GV_x2O~(uQVqRMZKvRC#I`rooD7{)SBD~YXO|3e6$V_+=zA~fu#oSS
z(c<Iu4vmqzHLQ}a4oEayN9S&6K{&=Dov$7&!p;nBw=NxgkO|^O29Q*K0d^1t;``Fp
z#RoGjg_m|;6$RlQ^}Q03=7`1sfqI`kAhiH|bN<U6XpcY0Uls?zXr}Z|X90|+T~JSs
zLlyj9=#aBS3Zwy}4#C|J;Rqw_zw<^pZh+0iO>FGIpRlysE6QMN+jiNz`LV}OBr=&y
z=B-JEu#J6@&s(!j5;3)iw3V>oWmYGU@(_m&0#O3jy%qJZM8PZzEm38xk=_E$-=~$&
zJ6E}x3F6p#pIRaSi=Zl8S}Y@Zji$iCU)t-^+F*@UK%|`nuHM6_PrUtr0Q^g=qrpnC
zk?JLlmIY~(T@pF21be>Eh2}oh-n)#EGbwMfJUVvFUf5KIRlx4UlB&AAp4XtRfAL_I
znz`-E>C_Fh$?p`h|HYea6lTyp`;juS*DkFaUA~q^bv;q{3pm;XBPID!DC+$j3Qxk}
zkw30+??cIJmVPB35OVTRg(@^0+|F`&IJ9<CEvwq?Xk&dDzf$g?1iubO<mm=cuIsjW
zK6vRiw407oPyE%l=K9BjHEy9rBce?8|H9RdRw*g$n4phGK=kq{(9e_hK)|VW-Rds9
z!sZ<N>WD~iKa#s~!_10*e-N4Nl1*IbTOiYsnbzdT*ii6+H(6I>5-#}s3Al`EO5Q@W
zvAYcv3+_+uF3ZOj*LTnAEbU%;nh5(e6ZnAakm(}KhF3hHtHUqRBh`CzduT#qe`O_h
zs6<XMZ1=+IqUlxKd)5OCTG7$!MNY2VC{ged{Y@VLB=iK&lQBEzK>q$X_A`;$zn%<_
z^bLnXRV+>)aU=+>ZjYCcAbY&#@fv(|QC?D(SU+`4>cuL<Spyse+H41@Rg`NZneGh)
zB$x6O9R0)`L}<H3v4L`z{0LI)%qN4)@XybGn9;Io1m;UJ7BEzKVIweDuy0od7q-CX
zcPu$<zjy;g(xWQA<bGGf7{ngrLSPxXf2ge_3lqpDYX;|V=_@ZppfyVCupv0se{djN
zO{z0s;}Ak*CB1_|qK8tXMV2&_ya_hEf4Fqwm61#T{iN;^7KhPz(V6fItz(<mT2Ldr
zBPd&CWNkKE@Rf6jWIN?<E2sM`4^4ZF7LwIF`i@n&QoM3UqJ;STLQunM3gjemw`RyF
zFaBW2rTI%e<PF=NR(NLy6p`f3fHrlluOM_kV`Zp6F)H*@vq;U|Weu%-d6ge%{X99b
z?0>`a+WfinYvr8tMb#K>p+f_Fa<+E78$h`YiN-VJL}vF|F1Egc$+@!&Qv<p}ElW7#
zOCHeU*Ct-jUSh%>dCY6OB_=k9h;Q@KOr`8+LQ>bW`4pJ~>8N70WDH$OX$#*ig)eTl
zXLZQ$Y40B-vkS2~6<Vu7m4pvG%v`>{3!_qrwVOSUX<GAgm`|Ow3gk!v&oPCYor=e1
z9Wfc1HLmbRi}Z<rnNS~88Sag44WwoPH<X4ZL#l8SEC5H#>l?}3-|}~<T|%Rizam%3
zj;3mi@coo~Go+GrY(qkzSQ2sN6Q&stHp>ecfY6YjwCgWc9w}DpVChvw54HHDC`3qX
zXyd#r(GiC>iX-QYvFG<6%`Xsp^RE*BqL03!of%Gul}fyuJD()xVmB<0_@>$kFw+c4
z=2$+^%gkc+3R-(-9c~pcZJIU^ae`B~zSO)pE3w7h#(?;91WZ@&bF=K}b#a937~6%S
z3&4V6lW)y-2=5w17E_^lc-XRx0aik=g)(A}Xd)lIu6q6Xn`?GL@fWHYxKNfo3HTxE
zGDw#faSXpTz1G1Zm&)S-T~y;g{u*uk<Hy}o-C9Cvgil1R4|zW9oH)eP?pwG#`z}7z
z46*ys=_rTAsZrh7F>8?v3oolAUrO)I9jZAY(Oo4sihE^1^J+~y1|{D8G17>O+l-(K
zZzQq7GiExn+s$Eo+MEidP>H*n8vBO)_-Te2rx}_@4?L3<-I1BG7N7BIwk^zH$sv+w
z%g6m3@ne00S7vbk`Lu24>E#F`Tdng{!~)?aT^9Eabgngjnqic9P?2OCm;UB^&PW4w
zEkBlrhmuEFW$#Z}DzWBetODHsAe8j)SK}(3N$|08$Hl!6WeC6KZPcQ&+sD>|vZt&u
zEA_^gSbaod!wto^Qdk4Tp4@1+{xWx(OrK2on@5!UMA8Qc(4DN1{mVbOjE<4ooM3ZI
z4pz|o>9x`WeF8erDZ<NRGlMF6_)xHd<75tWNs@WI$I-jP$1$s#BK&V(jj&;gX~EXf
z$bCb3^~rzVPMtoP*KybvR9vVo{hr9;d+S>d(sG03jAF)kd&>1q_)m>Yy)fu?Ovz`e
zGt~c1zNvxj*A@fiFN@DyeccXD#;EU}AzhDK_Jmuu@!4~P-tqh&x_qhCS!?0+IR0Kb
zX#O_gzRC@aW|k1w2)57p5-=7iwLfp*RGoWUtF70oIR+$QOUc^}NfR#(4=lVbYy{cZ
zlAysd5WN|+PDS&e+jp+b>!G!cGIZWYy=Su{F!IMyO-{XoGS|PAg<6GT*&0sgz^il#
zpuViXa!8QkemwR-(fob%Ay(?_w#S0wO5RM71JquFFpD8zUMTqH*7dymdnY$xY5k<m
zLJbaa&n-$q+z5aG&POq`LAT0}tH=Di`^4Htm`o3y-p(;rdi625#bbkcYy^ha=_~~~
z{J-fUC1EUB?eFi;)kGriQ5?!8&&QoTW)55ZN*RpWa=a{W^Cg=qPoYS962sIwf+;D)
zlrbx-g?0Q`^+)is(up01mmKYeO^Esf!7w%>u>Ty+!6qh+YGEYM>RL9vcVJ3bd|n<i
z;T0)kSeE~3M<zvu0;TbVDd!b;%>Y*;*N-U)FYSi!NDFu)L7k-mUf;jjmJ&N4?S~m)
zfumWK-%#GC0f2M_H2F>SXNcZ>3ILapZ*G)*ot~sj-1zr8^{1$~9(Qcz6e>O=Uww{#
z1G(h*b6Ikh>bpt#KBpX2-w|1#;PyR)rM)jEvM{kiS=%rBXdSqZ0{NGFtFQgf?{h|#
z6PVNFF8z_JYOyGasr&VNzJ!whz%-*1bfu7ITQ6Y}RS>I!S5=f9K|p@G|9xE$atj)%
zl0{a@lczgOE>(!6^{TYEH4jXS_MQ4^o9M?m1s^{H)Gz|8j`)NHZ^eC~=0=nu|GUmV
z@$t3-l$L2+>>hxIwC~rdjBEQNMR*pA#)or8?WXs4fI39uZJsR>T8!<`4IDZBxk&ie
zTs7ymF93FKHaHT()FK#12whznt&kSlk900CtHZ}7wwaL=A<a3B!XYVUHQ%8^_?GMg
zT@7J*A(a@){gdWFn$f2Wc*i)8FwNRiFP_@=h%HZxiYm7r*0Zk!H4tl?UPoP6vhv@&
z+Vt2)fH(~kGKy?b%7%LmxA(-(ski(ytQI(a>#!RWWLDkumtr6AqFRS;QQt^Y>E!Y8
zoG+he(1XY+I}v~-L2;vav^Apf$qin=NPEU@8>pi1-h#YFoDvR`Hp|QxjS=*T+p^+I
zfjH&bu&H8=ntqHd<YNV3;NIB+U<n@r)n%xoTZ!Xo!J10Y7aoO~Bh=C-#RGsmqD|<g
z^K)@k>Xq9ca-O4$snz<+p#88aCp`mes!S&3k!fY)adhkA7+FcoFRwhq{7xRXMCgfD
zf*!HS`-09Mxl^efpX#8Y-PU#Z5&=0NR*~?tH^|f8W@Z2K0~^|nrzK;HqVu?hR>`B-
zX>mpxjUE+;ia=PDBNdvT{$pgc`4e%OdzTpjOQgm7ChJyLG@*CiVxLHQl6)7zmX@f^
z&s?|%GT?xk=Mg!zqGV}o_zL1H`Zo}3wi7xOSD?Tj1IVdpn(pOJXdAir63ApJmzRan
zW`AH6y^IA}gi$3d+!EK5onib0(+Onfb}aoOUF^oGb7@6FoncUH3ac^{U^vXd*6j*7
zjkwn&)%M!asq<Sf#%gqmeK@=()qbzdOORk{-q)Rj4xOylBkw1oRQD*0jZo8864{{>
zeo*x3UBz_-`!J-l{4>AQQi(HHJzzZ!NroP1`(SEt?;o7O1@D{N0nLiy?bZ${qSvb#
z+Ih?qpvzqLE3Ic2Yb-`_Q5jLiM2$$@U0~{r0o8MYMdW{o=%|$Z3hZir5N^kIj_&KA
zQj{B^BdagSjGS1U+bkP9RnXlY>dGsZy}m6V^-FyjL4V0@6Ok&7zV4AMfxAV?VG<{V
zGkvU*6HJk#Q0j_ogc}?q)C`;Nt{gGsbCT58(Y)`nE<@B^*5G|jd7Q$ODG8iLPWT9i
z5~geBt!wAj_TT)5v%>$Cb;<O<urAq|+5R8ul9QQ{>Hog|f6^r<2Pez_0bLS6(TiEy
zxR^Q-(2Lm^x|oWX8rz$gLh<oIIlDNS8rnj6Y}$IbD(LJj(xPqeU_!__xQFfE@=fp8
z{ztsz262Z#+YSPQ#0u>CAI8pMTNGtmkju7h+tyyTZQHhO+qP}nwrv~lbgFk7)ypr)
zkt1i!h%TY{G%vrq-8&zZx>^36s;)QP9@n*vGC0_h=%S#k(lq4soQ#|h8Gw93;<Aa^
z(NQpcqXXg6f&^#4tPTxsF7b?x08jufeFOpEYY_c20Q>*6h{Q_(B)B=<+SRl*+W^m}
zFn?fxo35q=R$f-<{;fqhMPV%+r~>0dLqgL-D>)-0I>9G)31Ap7_Dw**|7jTe0VGgW
zcsWfp0D7t$(FD-00bCiIfXh39Z>nbiBUexL!POXm`v(_5_21%v8Qh!!H2$KJsHgfM
zIo$nqwttbD>>ZlHDI}~Sr^+d;fQm^qSV;j*;RyT>&-zZ_Y-@r1ng&i+ZT~KBk^67@
zQgb%^X!reU{zP>M{?uv4ude~(TRN}-AX=MU1x$a^Y#rgyVgYjf^y9mh`~>|VJAO~V
z^Z%*{fdN2T-?MIP&C0Az;lbvmxHh>vx&m|nBx4Ecqk#AIHoZF4|DY3`n7_7;{ivp|
zYi_!K`&j?jkMf)yz^D4Cn3|j#>K%T9w7MqMh589ptlIiEyE@u~{sLV3|FV;iFpO_(
z|K_mv`#!&U4dRP}a@PKKUhTC0c9T_9#3V&ULNEG&AMWa#gF1-EHwTdN4_?)@n9=9`
z0{p^MW~%=($N;wfXx{Ade*C;inc2~;Ao~4dPV|ljU+d2JQy2b>CB!)dz0*B50H(8f
ztN~nQec=G`$@BL946SZ&4}o0mzuJENv_9hg?mB>i2=$$g_o8A?0}I$N_XeqdN}~K!
zee^#$O)5v4$Kigg8qWp+rarq+$Bo~j-!+roqsXZ?oDt;RpPYqmJ3}UE0HZtqjEGX1
z7R?jh8=j3)O8j;<n=}uUgo+<Hv)K}QI+(UamQ%D=AMf;4Mj>mnzmK>)&KwPBG+x!0
zB4dBy)2||=oZ8?TVMLwRH$Lsk_iaiwr}I<|M9u^j#%lH$5cLGv6zI*vrK=;Ug0fwS
zpP{j3VW)6Kq&Si?P=px8S<kW<^Z1$4it873HvVfqT1sn@sgV(x#oRb^lZnWcbd38V
z)aPxhTw#qEVRCj^<T=VcX$MKsq>I@q<J2Y!uJ-;Wj4IGcW1?|%e_q`?x!)+~D+av!
zEV^I+lujOBR9+h7QcIU5TDvxAeAas7u{=oQC&#8Xd7wI{1QO{3T$IU}g(HS|-q++&
zkJW1g;JA6&`{u9xzA5PW;1Vm>B&>vej?{YXVH3PMRz6rVxeQ$_L}YyWO8M{Cpi3<p
zTcw>0o}`sS{}@WxM7UpCLmt$ke`T`%E4`cHjcUJJc<N(6|EJ^RtMS4Q&>K?N7%%&Q
ziK+TLCo8Bruzvi4{0<-yS+61m%b0`{X`Z3?G^#Y&(;uBM_Ab14BVL#?YgX*NDIt(L
z0Z9lmui)rMU~hL;oF2<Wx|aW1pW68s8n9fD-M9+3@uU$hu{GjTY88ri;KtawKoM5y
z;;gmL+Ifr-rCMOD%=$>bdpTrjNpTB~X_7EmlqcwZM?u6vu!yMu-0PZMkdzQtQAG<w
z0ahyG_idef45&SIE?5fn1}C20yq*&xAUTA+1@?PwhlI@P+%uy9yCejS6cqK)QqDB0
zt@jqLxBU#A%ex(3tZ+P?$%4$+EoV*&(!WFqXE|A1TJ>MZJl<9<F*FIO;4C89^;jt~
z+USCQcj4=8O7{~V*EE0Co3<3kWQ*M_`F^JNgH@J-Yekt(*ANzNELuR5Qh;BK*oN{`
zScL$N_=yj5D|7qse8^*=twcSVIS?f*W%(ml11s*hF51ZM-L5hkDv3U@@h2fi0;2NQ
z%FL{u#iBwhU}$s~YIPktOcmS*v1C%Yi(U$-p-L`}hq-YC!gfs{=<eGOp_laq0|5Ww
z!G};1<-7tEo=Qb`L2RN&f+Ji>;~<L0{dc(1e8!y;Uk&uTLPrC|z2aG$&$Y~0hv?Wn
z-#IpIa~+w{@dE&LP*TOF$N~?G5%7~M5t2#bu8`th^J)Bn1*pSSh)1ltU4gK>j@d4g
zc1)3r0?_Ua!Wt2{3qF~aZ08{nb#X<b7%OlO{!3j&2Bw(}#u9Dz&G)kAhK~os$a}<G
z;xuW!an`OZIw81*^67G~0J3BzM{*?>CI-k71^(o^HA>P&LUs9ADu+4B3FSjK>v@Re
z(>Gy;M}b|&GB$O>MS&PNLKeVj@zS*d#L`b*bzEv$aeeTXe-YOZh5PX9bFfiH?P<n%
zAD<kKL#BXq$hzdRuhCU8WPp$vIT^QSS`c*?nnYI|0irUg=y>*%w(72qL#XQ|i%ruT
zLjLyJgalgYx3^r<V;u<{r$Jn1G+U|=Yjx)18P$kRkg<vR#!TSo8w&kIX!~VY^Pmy$
z%V^WO+h59lrw*A!_r9Oo;1ak_!FM};VQ0~%&=gXU_K_NL%3w|L4+uW)F#hx=W_NjZ
zmG@E#IU)T4cEDB_T`Bl72YgI=@)`2i)p3+|4Ym@1=?>p#by(4?1M0SpB|GSQoBqr@
zG+14(po>;w%_jE~2PaCcOu<XpqwIn9@$^!{MDD*4#a^&QW0GOqNm}XzV-#23geXGh
zaRpzX;lu(Q5aGP6@uLuZw)PSWhLEr)iQp|_vXu%kBjaO;xS3>XNwu7+P$s9!F9w(g
zXmiQ=BKR%vgEQw8ki#}_5Dbr{H|p}i+Bdvl<WGwvwqek}D4a8hS<<=z95yY3Jzi>b
z<8ras)rkQd?w+iuw3NNyefSsC`=!Y-y9F#6@VBtB{lV%;Z(|RUz?|!ph$1b1hOJX}
z3$1D1MvozvT==dbzD)cO?!?cewUM2oTDw<YX9sHuj)k<9N#m8{tcJLbBo~0+aHQfI
z<8w;3$rtIv3b82gFeY9+7MK0N`m=o=!MN0`DCEAED{M^!v8Qa*=&wv(NR0&6U-)Fg
z5Kmdo?gp50jX<f3>}MG*)xJv&iC?^_FkH^N1Xt<R996L*Idh~;(s`^b`*$rD)eGVl
zFCDkmTK-MnITv%`@Xx8TShtjg!cA`Oz81U)*E3_Peq$kn>T`672UC*tXDqbQ<*(j<
zAykAc2DmKXvuf4nGey*_|40}OT@)_%X!gqW?(SJ{k>f-Ox^;2W{4(=x3#2BzUJ&LH
z6*o8=1$I#PZhGvpnhWwWEgd?*9upcATU>m1cOdbuQtfO};tX3sxy#L%$Wzr{B3^zK
zKt#@5*x6~O6syrIGsw&q0i}I+-~h;OzMTJ4+|nCoE<Bp;yIG}S^YoMhHw^L56|)St
zHng<Sh(>>DX9#(2Dk_u|{U`np?{Qu)q-kT371%ui=i?3K0Z>bqtjLCnyM3eADHbd?
zXyDK+D~>>W5|lawf|8~?DcKGRGI&Lj=xYNhjk$+9L;g(OOh+|!P%olrzO&itwsl~Y
zXB<L;l&P@2`Hr;~_)GCE)P7_Oa?h-a1l)LU9`u#uOH@kR>zazpUd63Dm=Zz4Q-$Mo
zg_h)kXU`P-Wt@%VUp8+hW}o6_-W{~%MiJ{SWEi{hn#ki`IYm<@Gx}xji$5owyBr0>
z;OH9~%rB?7${F_(#iJ{Edov=tKoWU?Jt#_dw5Wc{TS(x}cs{#3sfdunjc)ub(uy-m
zAL>omK9rfOa8fh9xba^AvU)undlkJfle#4CDd{5HYSY=MXS*&*wYBzAMR}J$CT+z4
z4U+P$prU<u!A|vfc>V9i9#`lDKsX5Rna4sY41ElbamsaPedyJMV_dOIYe!$#gZ5IG
zPI^WIZKwpT_B^+f6$EgSsF^KfLSvO~Equxj2JUZ>r3VtM8B8KR((>-!L+^Qk?YkL=
z2R)Zd3pTH;|B<Mn;;Xp&kG*@KF7ZQb<!O?ud#*RUJHPR8tD5E@_p^5+T5Cvf3pT+~
zvT$S0(YDC1N}OKXt4MSv2r5f+j(mnMeXe`u)#y_egRrPxKibMm(=B?IP*pgAYecb=
zBVp@6tM98@VYra>zC2OW%?F?F=47sJw%aUOnos=<(ckN~efS$V&VR^Q0kDqU2}0(3
zB0=7kUPQ4LZ!6`LfsdcZUrCRzA<vSM?V*{3c1W_9uc+#}W`J0CZAsA=X``3Ls4KhY
zxIhb&{s&_B>+@nN*Kf;;?kCzb7Z}b4?ps-0{-K}{g00ChE;ImN{)bO4AuM<u7$V!I
z5M%olV=789ycCe#ora+GYm#+=97V0jU~~M=dQs}J)v>Bnu)d+oRf)pzyfc>Mfk&U)
zQFi$qHGU6ic^p%Q9ea?>d=sc8q4+6BpG`kYkY$Iy378azMFGsHt!|92g!~`AYgktz
zGoL|d94a(%v!7j5=SBei;Xo33GbTT-KR>)<@=opYqKGmi^wZj|GqRhg6(AEr3_r%<
z2k+Hr34n;%+PO@*6CYT~VD3?<T!`XA*q~3FxP3H&S;PDFP~IjM<;gpCz$vyF;~e&X
z`(oBjtRgvW-7L1z^b|lMIdKHSG1TN7zxH+Ed=X%@+Ku~kn6imrs|N>g9#Q!fnYW{C
z;}A+!^BXunQ`3!4J8s9H8IUexnFSCTnzq$r)tYKDU&C$%<AHVHQcTqeyj=?KD4YR$
zl0feNEor~?W_J`cRmy6`#Kshz<<pT_5eA$jT|+YXFV^3oEU4gbY~JA`@$YZ8xAd&&
z^1{z4&&_FyyhA}CaYmM+H{VGXRIl^RA5*`aOPjJyTY;c$(^ARA!n9a28yL&a)S!X%
z8Si8g53J<InZE;>!+S-GHV`1hBmb>f706FbnFYLNjTJf?0WmwKYa<<90`$4h$o@B3
zsYMP+B886gl<uMZr};X@E2zReNa{;Uc=B9>=ylDcrB^p;y<P1hLaL#h1s(d!gmm0=
zY`5FB!=BB5BPFp3<0OH1x2ESz-lYu*a}gv4syDC0JAisFO||OMYt4qs#IHeI*klnZ
zj4dXNLDgXQp>Oa#i+dG2qA8iEjpXthZ=VC#%+M-#|D2#h&R6_5F#Co8;DL)!7_AE-
zQR4bXAq>XE)j!~mHp3%-Cp6*?3tE&736(Wbn?Iz>N<AdQD$<d?S?=)rCX<@ax^iu|
zXBz>q6hbhAa>%|+#ACSazpJYMnEYdaNhGwmN;T?9!>F`)Pa5;;K($Rsn&`Gq;jpCp
z%@X(R%k3lG^<=Q<J99XZV+=)U_5nOcK6jlW00ZmI7u1+&u~1S<;i7eZZ2Dkp<YZy6
zt1QA<e#73!Ntwj(B{20hNAvHM3L`U8o|X~2)nuR-pDmq=>G&3RtlqF&V}#(cvYy-r
z`@Q7lGyi4iL)WGV*!W5_#i=ry7QXJyjdyTK?UBS%4qSBEn-50Phe_e~W%T&znfa*Y
zsbeQb+tel7LGrc|@>ND-SgNh9H{%_Bt{>+zT)_?w7H@F#ro5K;lQXy0ZCB#<4E=+)
z_SdN<QcL)@V^r~OK^GJQw&ZbxD&$8$ykVb?F2LJ;b8XJ^Rp&wqwcnA8>5s~URuVY4
z&atIxnYL_lGXdqWVSp3yRhf^(D|2&!MKy!{j+};j@Q3v|#Zf@K5B-HC?|t9cy<&VD
z+tnaAyG3lNc{44lr*}*o?OgHQc+7%MM`)AY0KeJEHTP^sk{$ewmr`rku*Ni>R6Q|Z
z6sJx_%M#{u4oKz&A$6{=w7k--1`gV?*TP$+U?c}-Q;qOMScD$hy(Z{=j2dG@v4zkN
zM?JgrB22I-!KXgzq;H=p(0Q4o|Co4$EpQfHvA9|@feg8xHgc~lF$BoJ&oNF8!MZLd
z%Pj<5oM1@<=)a%RT(`MvJtw#R!ifj+QEikjp!oPx3ni$rQfVsV1G$(&gu@fuCbIu-
zgj>SZ=tgE79IU@Qn-|_X4@>Y@t<%v^a>)sPwLuZv74O#9Rjgnlla49Qp)25U9NEWb
zV%dqOA7t8QPHJHw#~6(O>p*xKXOP{NHB5}qsxPUdz_~^}TCnx;kT=eG6(Hwv=rs8f
zK$mR=WS(K>?%|_n5sNSjn{YN$d<hV~urcg8&f;)+UELHf1k&AEEln)^s}N%tLz~14
z3Y;H_OWE{?$a=ftwRvB!u0wYOW1YW9>N5pF9UilHDIw(B#p9@GW|LbFh_EM3JV6-e
z=o2_!k;zMGXk^M;bD98;{8nN-4GaMqVL&G?iO14<i~_2GqTeYdGd<XP(IpkXcs1fY
zfS}Q<?=V+zE4{tLPWV#_(Y+W>zlJ}GH1ZvgwW(4kyv;)ZFcx2C>o>Xx-!HGLvM-g^
z<u;dUmf4#RXsnA|?hQH&I?}_Rs$^R{Sf?$r5$JrHxPguBl3U-9tH?QM6+|GhS@>eo
zS7>O^Bx9YiBezkh2nrdLXbAH2`_rN@eu@|{*zeQ`8U++VVR6<Vx8KS~%x<1Q_Y<rk
z$IC>bCO!RQ#1nVrvX!@n!_hGx{Bx%$1Jayq#53ZH5ri|R8_g>A)+UA4pz*-svn@}p
z3BELa01<%7rOJ&$8oTOO5?v&uPr1#@CB{MD?s~0o9;!(37+n0&s5&r#bSqjTFtUwK
z?Z5$<Ba6p!4$;yUn0iocnc+@ImqgWJ%<w}KstAvYOux^#)z}`BOZ&HmKkef-ClV7r
z(z$?I;KnE(vP*Ff_qN=fV6B(OLD}ziK;h`zE_(#K&b;3mAC*&@oxo1RD%#iqbrl{k
zvS+keU@z>A9Us_^P*&b;1_7L6j*U%AS@Vj$>#$+wF5%QeIFzuUyf!K@KTmbN(wRXG
z3<&Od-cRmM%IgT}q+I0s!$QfN*DrY4+7^xyJtSp6z~^Y6&9c9O3JG)J#TMojI#Q@V
zsi2PaRD2+<&WmTM?g4-SPC~w)V}PF04BnWzn&oty3JK9qPu!G5Xd+%Nm}&XYmmJjQ
zvRP->6q@q1@7O8Lr|<w;_IE8K*?Fe<kIgoa9I{Nqor*h}TcO*-ieO+`6#X|)&gvlf
zuO-R)4J)gR*}11cHz}2eiTUWqLt;}5iEg_iW=LE+g&!D?3V;CdvfuymcbE~yFHJh+
zOhsS~lZPu~+6T5nzjY_u;BDs&mT4mVZ@#PsyA>elXmf|F3m1O{QMwtc`*JpyJ8;Ni
z?Ms~lfJz))5M$)Y8zG+qtFEU>PvU4-qPtE)N|;ei>Yy~u=<SnhmtZy{OlvPCLrNoj
z3m9#Eu$^(T1Gszl8Pt}gNhmK9D(YBt{@PJN!A@H{dhw}~>yT-jH}CXG7G$ZLFfo}f
zdLfg(VRW_!uT1LY2wMG`9)md<6|fG;-E)3GS}`e~U-J`yF*t7^G=rZz?)X@;6ux&w
zcTii-P??_E@>RXBorsxfy`f)jks_(_%c;%ZhBC_G8t$EQ9lh69U!BB%74TMSZVJ12
zAp#2cs~p-qmkDG^ucP0b%|XNu)JE9JFGG{V8Ys`mp~T5{>d1HMGqAMM(I<^Az?v&c
zSLGXjUy}z*JhsegME*2<uu;_Y$A?~aAp1P9u-c$ktOitvGz1P2W)14r`Kw&?wPDqt
z)7#;OP0E9Nv=ErKEF1@|p{DHP?X#Q#ejT{ZkX~4LpSAmFQRV)2e5((IniHTTBGL^j
zvR~ZHqfN!WJ#t^A_7}3hpeWfNlR;W?5KKX+YG!9SjGmslp#`0jZ8<fJj?dL9B|lhD
z|L!uJQy7wzC<$>;avA^|t>Qq34Rb12Bgvpo)qtFyCa1|2gL;YIq*zCm705lP!{nFu
z@J4*1d10W_tcGhW^n}T~^osFfJBcsUM8WOg`b*Ep#X~SJAitj#Ne3Xa{8K?Jw+Md&
z6$QJROeHn#!$j8h%ak>XR5G%j;~0YnyDYB&bW|xVxUnqd(Syo-B-VP}YfIPh9ALe>
zt5k>Q5!OnAG<JI05$aKNQw$j#S=?yP_>O4q$zijC4qYNdx?>i-D9w?aUHkjw(q$=A
z#-oAj7H|h@c_e8TSRplzy&w!tD{T^E60xC=UWE-{4a7eY-QLVTL+eMmn|BfCQ@JyL
zVM>Y{`>LX9$j{R59o#sBo9d!pEN3!iTq9fWdSll34na@kVni$~8KE-y*%L9PE+|xV
z#}1^x;q!a)k@v`?&!5U{OCxoS?w22CIEf+Hg%FfnY`%@?sge=W3<JpTVuG81!i(Rd
zQkrh-0MU4Aph(6W>%3c6v)Sp6%5EkXcc2ulXT?gTPrrGLtxh0cmVha3hs-=@3VF;0
z4usltVjZY`hL^zJWV;o-PF6PbSx|e6t@QWi54UCtDGH|P_`Wl-Hp%&ui&(R_m20C@
z-FshJfOadN%rSOjd+A#rUrv$EiA_RWj7>PKE^m{QXIjp9L%QOVxXp_sISv+@;ylbp
zkaDbenNH|s0ONm2cYpHNH9oYQJ}U+Y*k%_w5B1{6e720k4Hh#y6r^egmH5~A(Xm}*
z4dKe%hQu$=<Cau9eIpooW1op+DoRMSti%I9I4gLRkn%#!QfT&~9|F!55p#J2?y;8{
zo<>5l!HoaYZ<maiZ0)+QEa))LfEYU<A~JKXiCLjWk%ozHA!0r79YX|pg6+9}p+%m(
zzPO<$yW0-Jhkk+*ox4Wzd#26?J7a8+Pox|QuY!E#8Q~ct6-}py5;y^y@a|NW(2m|H
znLa*p)vb%E{c4VPh3o=ne6k`*SpJgA+%at!OY*D2RRziRZjB^2P1^;vw0iG{%ozu6
z{lNZE8I>WKZcDtXPw44>81n62Dm1`ZO+3cpWz&*mD9FLNu2nvI^f+k&N!U6*P_zjc
z8uooqJ9+>cz=b1rHDsF6ju#LXV3&eze7LC@S|}xl6RLf(qbZw5^WrxR14WneSC%jt
z09kuIb25^}ceJnHyfcBirF#~h@TLjhXMYLf7d})_QR$Lk&$RZ+_I@IGyvF+7i7+89
zirqFFWMo6}UsAML@uK!152K3)Z>h^{oUDG{mIBjLrGndOzkP}H+`41R`nLF0sS;7L
zttJA~^3)I8ENr8e;i}<a?Yvsb)=jlj()V`4g&EK{vbiou%Bu5bU__?fO!h;=I&ddm
z`cKIzQ<w=Kn!PGOOw3jjaN^xjE@qwrqd_+*FC7Ux|0uxZ0TIt9CXpL-z5bV=DPhS{
z+MaH(el>=GTuxH*9<oTHOISX(@fkgN+N^(SFZ8cVsuz>nOfpG%51HRR^h}aC637a$
zD449bb&Djw%P_LQ(eBJdgk<jBq(+044=5em`<%{PHf5mJ>6{=dL9WSUHB}c@1;t?G
zB`+TKTp*(M_m6cK<=ImNU>qy6DwcRADyNzoCff}4{RvvD{?GGdvz)8_tLPwdF+c4f
zfmXK=T8MYCEF&RL39ma0K`SAu@YYHexzJnnf}f|3fBx1no?3g4oxcg#imX<FhU-Ny
z=B9cywLC0c0<%=^q(mPcd&#4JZN<T8yKR{$zUZ3uyUybXCBq(;YyAK%?3-+vXK`vU
z&}XuT&<M$p>*0wnd#@!XAPlPkU1z+24WHx|hzo&y=>WuNU&4kjc~{Rp-~Jm4ACPvr
z6!PH|enECKxi>Gw9WNWuV9D_=B&0P$rzMi|9=eez&O=G<NJ3G0FjWy70z&xFf*U52
zqZ#UzzS|UqkZVX(2@5IRx1mYdxO!uyU1wPD3`Nc^+j#c^-px$&_{w1w^LBCwchQ!V
z#2~8uyy|sjDnh}y`VG}uG<vyvDOTyGm&y8bag#D_d~Lm1lFnnNqcHFdZS0%l*``nQ
zJ8L3wiIkY`^KAy;>P-+W0qr@Z;-AIn(x^jr=#dHrfq<m?oZ}!(H!IRVkouP%kf@#J
z4O|2mQ0J@r?3f@0^OYlP5JM7%E0S}R^TK)fD+|^6vXLLy?A6;y0wJ<ZdFe)@?1UU<
z4ghP)H2*f`998WueX<)nhqX%HqPnfDjA1}y@oLw?!?P|Q3=%wQapI#UL#C!9ZGp!@
zQv$T>0`hHHSFHa@S`6l3pQOdD%7W-35)bAk>a-H!mtfE^R*^B-03^Kmn)>B0dtrhi
z)5eQ!rzj|%?hBYnJ?os=I6`=qMf!4jTq9lD2XSQ>1pXCKrW%S7bzf2j$?G$Y6u2AR
zZ`bO{saL1^aJ3-(di>C7zU``0b-^6A=QQqnuQ;WMgoXsymxpO@kXMME)mX6lRQKf!
z;sJ{<4b&WJaZO_LO+3Uj9Xh6Kp$<5+PB>AO0*ri6{!_dPW8^)QIU|a~7JNo{)G^ol
z1TBW5sC`hg#xk@E_$~x?ziWs;l8&R6ua^UEyx`F4#}IbcNk$c5TkxV-vXzR&t%<wL
zk1iK8-xK=VO-rToaU?v+$zlxn2-`GlKvnArm|#rQl=~?t`pB-c0)#oPvueT8liFI@
z4g2iCVQGI8W&(FhRk7{p=QAq9{4p2_vdt&_#z1Q`wv()=wV~@e3Qut=m)4aeEf>|e
zWSr}fw$>hivUhl}%q<iX!2XC~0QTksaFc+fw_1UMU5~8Yu@({^Vz<3<Oru*irC&G6
zdNm`&H&Aq9vAbSs-SK<uPCy<XPJO;t$u@jCr!O@aa9(1!M;zv}E}SFM*$3zq?rkT@
zB+j?)oM_nG4HlgmEK_(c@W_4N1JZEJ-3t422}KF^pea@BYi>Xo;nF&|6LSRHon>4+
zMvwvLtO?Q$MC)+pyTO)kZ1|7N9}8kNrHK`K;Ur~l14|HR9Un&Pq?vZ0Pt1G7l~mxZ
zKwzIKI8Q<DqtT0Kly}JH46_Vr=XXC^5++ZyC;8B}Vr@2ZE#$Z2zqjVRy~QX$=!|KQ
za0X4H7D#}G?zULD)1Us*?(mcC<vpd7FaP|Z8@SxET41j<I?n+>T-@jjbD3ZMXmy36
z7GD<2r<-?y&+F=U_G3=+v>KsWe#O4wkrG!n>Hw#5%eJCci6l-29@Q;VRh$qdgPsnr
zkTT7wTyQk4@FH>rP0dkU6I}g~e0P2aej7uREEEF#Vjb6tY;Og~dT`2eYlz{DYu~CJ
znC8Lr_$Lj1=p^w0Vx>9C#Y5f&S{FR~WZUKt{si{h&9P9It-Vjn8zsZ2^Bsz9>#@_r
z;1+Uhk_1g8S%>WAzt$4=n|UYNX}iJENDtMzSloh;^_{8?zW1I>8m0;&xCj!SI=;DK
zEwN4)gq>wM^s8)34x&z7LsUD~;TKQy8on^Hma>7uc<mRhjO_=iZnE#(WBMP1(oX=h
z%p!X0IRW=TpJUvgis^{~b@s~XAf2my4f5rE#e&^<%9M!`?H^ZjV<Lnvzdrzm98Hz<
zn#O?tP_}{-B=1kmc^of;j=D_je*|J%)sDZVTES@76`+$LgNx&aT|beib($TKN38mv
zBLD2`n?o_YB@w?4+68J@U4iYB)%jyWwSR=HT8Ba(RHOfvXPGr1{P>xJ!&p%?&9tyf
z{KEHF*u~uq4J3BKLMpCaW(l_Y{u;2qA)gn>AWMsEL*3tPHs~B;Il+_+b<%_&O*mo^
z$gh(UBrXbI|3Mb``dC}`${vDgm&ff~ezVeP_5^I-qUfBH3zd#Dj?Hgmc-m=vXyWdl
z3#RO&^h`MyKjb=fdQH{nFstMO%YOL7l_;XuQO4|;RXQu|#@1AbNiXX7#g!<~NVI$f
zvdE}fr)Gg!HKj}(y_50A3HGfftmUN805%m>??2l$`)1VWMkI<@*RcwYfu=TFt^|_^
z`CPj$%6xI9QfT!UTpYZmZQz@T*1N1T*q~!#G#b>2t!HUbJ2x3K-&uVYqO|g-1y5<s
zbtwPm4GLyA?!)W&Bx5Vu_ifXUrVbzAC^Jy}>c4m_4LO{JnK%E09QZA+&i~=e8NBHi
z*lElKS6G@mjhIR$g?oBE3g57q@(1j-WAAYwUWG@G(u*A9A<!giwi>|1L`1$4#Yv4D
zJ(UlImIXC=3g@uhsUr#t4tlR~n&bN{qH&kGfgXt;*PDE*Lu?^JB$-)?<?;R}X!pjs
zpz{&y<4};gH)`LsE%h<#K3hr0G<2edyV|$QRj>AqsK{NKK)EPhRoDBX`GKq(&gsC>
zC(GMw$QUyqQb-tIU)S9>4|yE4yM&e;7RsUxVG`CNw`d+Bpl***_o0P56^~8EnM48&
zs}(nP&QV#xOP%`Jhh?Lrp)353?|UyMf3HG4_R;&x?f-Kw$-tWWNRei&kt7zfqA>sg
z%#?kwv)f3Mkx&#@1i3_iw0e`08&l|0rNr<CXr!`mzLYM5;?T(L)<`~Mu@*9Ah_!21
zdZDmonQ_N4lWFj@>l65;v{g+{nCElS)eE&ix})5W341@c$3TagcG48!9gARPr2?84
zyW{l7QU^JP!}=WOO4j7%UkkDPu*rX3e&@+fu96nf`gtnt|3D$jCQSLhEYmTx&v9_v
z+cS{d3RU7dE|h)2C-%K+`BJgS2ilG}d#%J6|LV0sy0y`K5m;DJ``z}LmV~055IzsV
zmSyP}TU#iG4$o8bxPB9|0a04e!0&vN{k%J>Tq*8QD4`=s*B;M>VBB+rjvRyVZ9q58
zHnKbtK6=U6CaWO7!hZI%jCzw`8yLRh_fd34qy+KO<1a!SrXkotj3e{hnV<6#$dZmo
z(H;=|Au>YZASP!%p#V!I`2A8*>{0@@w$eE8Q~A^e(p;bYIf+WG8(rH}irWZ}m$5#l
zk~;P6GHvb7%#1*$wbrc<V=AVH%K4b=`u&#9z`d9>1WX`L&EEVHt1=prDa80_LfSJA
zZC8EhG+7r3usom{@ry0hvDPy-m&Dg_(Pkn+S!4)G<)&mvF*^9tu>~5s31Eh|T|xa)
zIKI)2z1Afsj<_e$<iSygF5HT1*lVk5urU#Oi_4*@D6~_cjN(bF!dtqH+>G$@sKORb
z?R`GnuO#lEi6?djrH`zYE(2bcT<eSY!D9Ju*j^mi11$p021w$X4xDaF^?VG6$7d#Q
zbf8237wsDj+dcyWh*aD7m1%a=lIwKJnoUz*z0|_(HeYu*YS)dDjVsqJbylNnDd6M!
z&QccxbL+$No%#_}E@}X%p=AXcTt)gm&B3$+lFF-6UVE+yxf!jT(NI+k`6T74kT%=s
z{AE*z64~VdMb=F`<Vb-`x<TkNOGa8uQ!x~DzXn5PK(0-&1m%IF#NV~tY>=2FCEZu8
zfxXG+wltu@U7DvwZ;QYaWe|XG?4)(S^64IKy62ByXCDgtf6{R*|AUTWWMgFce?S})
z0TUy`|K{VE7&(~#e@oESo+k3fdr73Xx3UMfa4}t7T`2z@SZ-yHZf`f+*;WOPHrtol
zhhFdanYpo>o-Ew>l!v=AJmhkxliozT(M3c`3uj3z56r=m9YH!68yK4&fmv3XRitlj
zWN58%P}7HTZfF7gX0#!cl@I9zl-RMw;)5%&<MH!hn*qcDW&r;~&Di+(_&})q+l!Lg
zqGD27u>D`@!Ek|rkNh-$Apz4p{=&}gWYw^#5&S>zg23l@7bXB99KWC+{<;-#Awb$S
zfB{q4UIqNy8zXD06IuLKH6@S<U_k(Nq;CMJZcSg=P60r%oe0KJ#lh>HodMVXwt)0+
zAsXvH%2?D+`_v0-0hIi!kmt8g@7VIwIOi|$(TV#Z&kyyFE<fKe^^c(H9T`D8er^J?
zi2K2eZSQO!-~;+xrN;aa9UUFMu@COo-q58-7FK3fwns)MVDH(hB7>Lv9-N!lT)xQL
zs%G@q`xhSVibIMzzW7Ohrryv`>(*wL!H%Kqfjv8ZlBQrF>gyd`8rffdv%cullcHw&
z%-g#*{r`)G0uuLwnwwl28o)cd{6IgUf92hNo8|cb6?rO=lY8Cy-MD_ME&rYBADu!q
zw;7A-8~tT1ZQt10Qez_kP4O2IK{q;s)HnQQ?F{Z*@O$G4r1`Oizvg8Wd`S&14<Z`h
z!PWz464g_1cK*mf=KuI9WqkO_9{&MG{E315;wQZQ$!-5tF8n!!|M|^5`Gur9wza7-
z{RcQ$A@B3wV)z5?=SiLZL9w~iH@yAc4gOgbv)%7;{JK#z+LroZ6GAn9QiD-5U-e<y
zXqWs4o}f{&*`BTfNGW~w>3pW?AR7ZZH?#o){#Aa|i2|jstN%wY$_;IfZo@sxZFT(4
zfP!dzt6lnmexf^|@#^3z(Bg*veTn|4Wl!pz?67p_{3?FENuuD={=OaJqcAdx(fiXQ
zW21BXM+BwZ+y#71g~bI;{q;?}(-XA*X;Wu*cC5PtKL4fuxSjEj{C(Z=tA+l7iXs``
znm^1{mv?0Z()`A*{jq%?vAMZ9d{kZb$$sw@{u%w${R0Q;@jr#A`Y&<{RM5KD$h?DP
z|0GsM@(lj9K_E7iJ(<NS85wwq^99r=a(l0m)x8qN*fxFd=j=(J%hhu?wSXV}cC^0q
zTmnoF9q~<h4i-7jUeV=&0Vk*)J(<8f*KFid=SyiVbm7osiTj<asKQL;6-1bOacUpX
z<yj0}cuxT}Sm<pp2+1&KF4=$k&kAtJoPv+HNS4#rJ0CL_0~`0ED->^?+<08dzYK|C
z1Xi4zRLbRv&7s4J?a$x$VPCQ`SdHkB=rqT0T~P^+H(d3<RT=(OSg25jnI$w7vMOY*
zS`r#~>nF_o?2V%z5C45htFx!t?+G$xlo=P~ndbU4qyVB<0^&uWWeRH+er#SPxIdS<
zI1{vO3TPhf2>@gQIhU<lZSo;CX=%A!lXR({fy{mMBc>lE^xG*eMSFxn!Mx+zz~<!~
z2eQ+X*0uT>Rhe-NbYlMi265nfGB$y?1!T|;{6CF}Yb-ov-WrDdi%^8sPB=g<T$W{V
zi86cb*0}2uh#e0rrZG?aPMhb}`I5YXjY2=BwE)X{hmB-Nc>~%{b%}pmL$PrTE4Q82
zbeyQ}%+{D9+H-k>)d02%X-?p=N`4^if>QR;Rj4NrqPQYeB~e?1Kj8<xpD|>`)49cT
z_PGZ`vm=p6U$TO_d&1~AR=(Y+)&oSBj~jhd_QZ2hgYkwI%PeBk;A`}lb~hv?r>3Y=
z<LAs~5s1hCNhgyARZfBZxD4_hi_XoojDgzzg^&bI_B6gzrrJ#AiQE1gEyRIyKh<Km
z+GJ3Q7zEO>QB9AN=Qym62rRy5KyTlnQ@&_h2)<!NQha$bywMyOTr0D)aowAm{^ziy
z6<jTk$`d}A#jwFaPZqb;AQga=A|`)EveJR{`!w<oIvLTpydquIe*8K3PLs6;MTN-?
zs2P4Qwr9BnI(YS)P8-QY&$L_c<K;ymB`RP5*-EgRQIx~B4xPErLjj#}2qxXv>$kb6
zMbG3kDRZ*~4o3Bi&LrjQ7D<E<F{PXNUBk+Fluo$Z5r*`_9a>9V0h*dFk<dxfT7^K-
zI1)4B-|>9E=*Tm#fUu23u!5nvkffP#NcnI!-qFiJFP#7u;A<P{<D|epH+JN}oZvz)
zF(edK=?pmE*$X)CprJOW6`m2d%-JM{z^@E1nu}vkEl=gy_7nPH6k$WCiG>3h5cgMF
zNrmtM^U!<RtDr~oiBI4U(ZzdTuT?0`Zb(kj?Z@FyJ@(S&sKMafL`gaH+Qv_l)ddCK
z7y5pkJ((eE{+N#-kxY>Zbdqt0#}{?0p*LEgwM6)?xJxKrPwm&hmh04HZl{rs^0S+u
zUjK<MzYCYQ9kWSM%0~&)%;Z!-=wV_H%Q{H|`F~A>JU=l$B)&H^k3}0`eeAsfb67Mo
z(sm^#L))6>O|V_TWo6zzvJOB2bslwnULXnw)Ldj@elh%6Y#8kgh+0nWX?>af-Lrwz
zIZcuWZ;Ciki{n%YG$yElT4d<DkxwkjR=!<oGr8d8SKMyMJY}L>|6-b7rxTtu(C+9>
zRt5PL0;Liw=Y;Pe_BTJ%SS?+2^D3yFR<<>D?;1q0e}-DO8J=Psoqh-tlY(m}j6>Rk
z^E5I?>wpgyUqb@@$d>>ggIajT@|!}PjCQcBxF9Vx!!L(CMjx=uYzzml?qPeN(lkTa
zG@N@AKrNYyv6n_&VrR_talO-#xp6QHdgj~O5=njP&L(oNYN9nhM0wQw2U^<Ontff;
zneHN<q@gIUmzt8n8<Xx|663rATrYe_3wuV@M~-hS=`F_J8sqHq$Xd8M%veR6!o|e!
z@&f@oS40-=-G%N8t$duHbf40Y1ig)1Ta$8zQRkI9M6%qgq)S_-0SO6D@;7NH{XS&#
zds&Z9Hpbm?KZhXr1u8B9Y97V{ZXJP36z5QMPv#l~aB0D~Tut!)?O(n@4JXemdc6zG
zN7)KK4sn4;__rs0du9E=aD_acX385to)r!2V!saQJT3{MihkKTdItyRAH~Lak=U^q
zo2`MdhpIA)!k*boz?Mg8(7(5GngSAemlb;Cjnuh~KO#WNp$Ov+b;|sVO5vdYi_We+
z2`7qTJCjxlIpjU7AV2Mvm1TMnG|7F^)!G3W6a>Z`%CcwdEV?j=?@F?EgC1!&R)<WU
zq=<$Kv=jU82*vwCYYYy`L3|cx@|e%<{QzGbb+jXXkT2acJ(NT2FSxjhu<m%48QLI`
z1t*n<cF=wLB_AB-XzK7Tx?e9OCzQ->u^lRKb^O=hT%<*&G*+m7N3q2rP4{TCW%Gb?
zG%vRuPb@Ma6Jo)AVUdRxRw&I@Cd2(rE+O(%auOR~8pr^Z(ZXQlE>pm_6mMl?F_JB%
z7#w7^Y93k3EXlpJu8ZkN?2M5BMP!V3o$Y*$2JHo>GqS!Hi<creN~rlA*B!olLiOnL
zX%Z8<xlI`OaK`_nc177<%BGBw{B5eT#|GNUl5#PkMj7Ljt+#7qKn462Jy=IS-&za<
zd}x~APcxm`huoxVAf;82eY+T`6|1_#Li*J+@8F6B2Ee)>_$`*?p>#ab3+EW!MM<9<
zW;n5=!Pi)=kp1{o-W?00=DkX|Rq~)9?-g16e!0`S)B?6&nAO=LRnoB30N)U}rLG1!
z$2Facc>dR6AeUvsz-X?NvHXa~Das;4q4ab9-!BY>45Rm0FFwR7mU^(+k7i~6+1I{Q
ztb)i2YV&KP2qKGS#c+VxNt#aLexxwI)Q%q$!G7u?#lvQHPS(u>r!Wj7loXo!9*yK!
zVg&J%Zr$X{@IUH7D9tX^l0&R@Ar2uFnNWGqXveLr_cdLp5<hZ?aFd~yfLF^ap2PJh
z*#?qKDJ7dAZM0e;7sbjm!wez!FXpIB2~G=2%pl+0Wp%rc7XV!=JT7c0*=}EbIUo9y
zEW&6Vj-Dh7^^f;pp>ZlpxPncAc;xV(Fs(iF>;@jsWl1$n+?>!hVv5#Gp=sEU`jG>5
z&QUWz!@Wi6F*7BYbM5lmU*(K>3^|(o*u;IDry?BX&6icjot+4Z`TDzYL>2jA12>7N
ztGaKfRm>j_(M&7!2y1Gc5%Kam>37t&zSKZ)8n7FYGN`mOq+c?RsY#T=0;e=xPn->a
z$0{6QX5Vm6-?zvCx(TOIz_Di=I|e_zmBVxYqo5qRXGg_xL1eWkG?<PO3UALkWUsXK
z)OugV?`^=!)>yrG(Hmc*Pu1Lud^!sgxg_#wtP=0-FZUe})As82@iPwu8W17Rpje_q
zF+oN<eb<l>5{ZANbba76ZkNs5630fv>CpG)zmtG2mT9ns%de__f+b?>2}p~NWt&2*
z4Fq`tK;rI;CD+2hPBuWQ#vENmW{AY$Lu7xgv8lgqo|0CL#1J>|<=@5V6k}T0Dlyj}
zhx>H$y(ZL8nz#-liUW?HZ%vuWEEh$2!`c#<b*2a|7?fIJM0~Bg!|>msvDfQ@<j*7K
ziu}1Rw7l-xO8G@#1uznXLHLYVSVlmI?=3~%mS#R^N1a&TV6k*5xf_jdzT~hvJ|ppA
z8-nbb@$T2oZ>rF?&d9&apEmV}r~_L=M?Wgh_TjZ@yRl6wOaLT7BGpdQ!kDlCpgPgR
zo~MPL7P@fbeAvN|gw7M4LMI`+WUD6P&SK)l^=2W9mQ&u;h+@na>{YJfg_Cp+{{^B~
zEC^pW6SGo11cg&C4C2x%`eDSC5t+swchKJk@|=FsQ^9{YH-N01@{;!N>Z6u~7p!vj
zY&od1BDh|pb}@V8Y}5Lp);V4YgEC)ciUJAzW&89!#d&YZfEq!_^L~cg)mF;|DLs$Y
zru-?e%xjIq;GwDeZ*1|rl}A~GwY65I%i7Y|#$Ma>@1>(s%*HpO34HqxG*2kJiL>T{
zzBSu6sx?<nwUDdG8z{_v?4Qq#&JVv7B7kJY-Z~hB|J7Prjkp^!7fQE7E(I}oPzP>v
z@ICWYilx>Af7e=p)$SGboVX-3qKi07mIp;tooN68DdG9uFY{1$is~aJy3pJr)+=k8
z7cL%&XO8Fmp*|!H9YpxHTS`l?Fbcs#3pVcw5V70dF+|IS63`nYKPnmS;w6Oz)|K0!
z^qfn+|Kb$yZz6$xR}0qj-bWYwl_DicK6k%s*=!Kbf!}DeA}qQOs-v(z73TcIn89f%
z4zFBFQNn)mLbSquv(0y;LZ1dS=@(2<t)vkxclpH+BSRHWrgvy+n`im<U21Q$p_Z0x
zAg_H=%%(Y0kec*%;n;&cSH{NO7WdIraHL5*l)qc)YdMIw5>V12GnS`sj#Y6?K(pNR
z<(<ASFx8baUXs+kEJ@1aY%#aH-6F`&1?1ZPk0OBtHOFl4Y+~OeWJD;+EA~N*N*lZj
z$&HaH+$m<$5@VvV+Xk$N9D;1t^^zIYBtO@$CcV2^n1Q=vyUUedbCn5jZ6JI?WP;x_
zUu-r*rpkrRDCXLgJd*q*pLI@ySut{Y@m=tpY53#1g>XAM0RNfJ{r0~CC!{}pOF~3c
zJdQ0GmDu6Pepwi2Kvev(EZ1(@*=#=0RguivV0B5Hqm5?zNQ<_lGYQk6!}vIEHN~80
z0`eQru0dzzlL_t0%Lktvq6=3RG7%iQ&(~Yr0dXi7MIlST!c9FSJ=yG(8I++qUZ~*h
zto6^!gO(m$b#8}Nsj@4TXem*n&(pK^4FkZz%5>431Hy`2=ct-(<b0_XVd9&Yh#zz!
z*{usz`?H43h}K-(nCHIdA=kt_Q4V$*8kV^cbICM%NYCg|DxzgE+njOVUtcUi3N^*Y
zl>_Urn+?J%_T6Fl1pblb-TpLZB6v+v-zXM3z(`AOud;vnPb+<qFDV#`V+t9`d_DKt
zpP%Bk&6+(<nFXSVsH7JT<qFAfY$YCR1Nha-r9_``^wWB$HQV+ib)gH{C`rtnHk=G-
z&a`|{Ff|5=r)*3`@#4Du<T`?Fh;gv9J`@xF4+1hP>;x@Z3UhWjB(sBAH1W7FUcX-a
zU)3ouyu%)tC@Px`=%3wi*LoZaxdKDrhDMw)%V(}HAG%!$C8oz3!_r4CGf(Two<FI@
zz?f36V@wLJ_9={ESAAbIH<A@I0_8(6Ntg#MZ#3epY>0p1zQ;Ab|Av;S<Zk>!5s1qO
zo=kVmD$&xb4WpIPzvWmD(X3~?L7@>Z3=HV13yEFZI!hKJVmVE|4Oi1>Ckyly#f+V@
zjn9Qw**7?!pp57pc995h71Cbvco9gxb_`nu62lW9Me_8Vj${aH@B<jb4Ujq<DaP)u
z0n}HfiNsal@Z!Z^>dFMo(fbB4goXXYA_nd)Xsm0}cIyn+K4(ShT7S@##=V~1<^p#v
zXgrWtyhPH0L>AX7OkC6=Z^PKJ5&ChHS|be#+M-l0?_TvB?up-gmg%UW>A#VMB$7<D
zg8PQ;y31+Wiwibuhb|(p4?s5cmiXH<mLF~<7#X=hPFoDi1ULt=0!$4d1JVn7x(EVA
z;zH`4!4P=pdAHxDQ7h9};D>H)CFk}^|D)+ErAA+nadN5urrsH)8Ka0gMwkNm`U0mT
z+npwh56J=6_g=n%iM_*Hlgg*mt73dsrVw;oW}$sPo$?DuZ`wjj5rso^CUU~E@@`_M
zFGAKzt&SIOJobJ`?DxjW&l4P8_IAuv+lw&Nnymx(koA@c>|StK897YJbuVWDAc8A&
zSs%QZ17>jU#Q4uCOfI#IYke;hEd_cI`M13>qdVgLIespMp55UgqgLIWq!6gka6hX@
z^Ao!D*YsHZ?n(R<WyelG6c&Pv8J(W2VOMgAG)NuE)h3&pT;HS3hHw)3;pTr5YcnVM
zt(iH6lMiJ8OycGU72FPrLbF|pzW(?VD8*G8A%Qv_VA_?msD~UyXZD3Q$(t&UKy!P$
zyD`$Zgic;PK+rZ!qWLAayaa>3=x5JWLSqvD`E8W_SI?GwM(Q@3m%NaAT1CdvJz~=u
z2Tst;)L{YXtDCNl{?%+P4Sl89i@F)^peyM<bIrBf<+_?8*CH8ZNd)<sB3=iBUsc(L
zq_u@SlP-;`ae_+taqoN4?`$3P7)>hMf2`QA7UaI?-ty+Nh<3E9Rv^K1+G*}{VVbZl
zA@*gC*G&%RB8&rgm>c}=v+DPR*Xk@TFc0$Cu{*&&eN%fxoH-DiI-G!@{!{qX1-d?I
zH)Kb0b&ycXAmqqsWv06(WNIYRM&?Udl2+X1bTL#c!6J{kplHlr`cLC2!TYgb5JwE3
zMOi(gZ!?_Pd=S`|cZDjK+vC=$`tr1@S#c{t&EPvUrP+n7naNxu8fcHiBts&^wjccl
z&kTKZtmt=wufjnpF)CkX)#bt&Ay_Ma2+a3!3y(Z(ywMF!m850`W)D@3P-BSl&d^;Z
z;;a*9H0#m6$Ofu)m46pS9g2By-KuPbV%&}o8j&hO_WUsV%NU*}z>lSy6RwuGk%sd;
zb(15aVR9c4T?nlBMvBah`4)r^aw9pi_RSn_E=xGGXpgUsgjgN%V5N@T3!K+dKIwhi
zuTeQ}ZAjh}I@#ncC`fsXA0BEV!k_M!K^g=-SXVoBz1)y=u<ytVJydejZz-T|aB)%F
z^T;v?IVtHLSv{hM9xM%Ed##n$-7I_m3xk6#fi9FK$Y;t>v?{u<6}WGK>A&@#6;;LM
zhP7x2b3O`%Nx{$fBCCc{gC$4q(U|Ca=0eK}9SrI<96cJd_#sRi3>^!L8fG7EefA0p
zsIyBk{vix*PLNQXR(w@NNN<KwE*g5NNFwvh1d#h~5)f9iwGg`(<=4TQW`!)Ez-^<?
zyTvt^KR6r(YG)W&t^ad`y}?T`aa)0GvYBNH^7r}Qw<$uz_eq9V57MT!)k|2T^E*0E
zhR_wtojk41YKk_7U?7>$?tTLFgb{N6QUvqh!dy@fgQaSrKUDnuO)%y>(q4?(ETlq2
zwdJR#5Y;a!q}TA9tI+ta|KiElPdG2|u~1sN{dqlX!>9JRHP`b+%lSzA_5ogyY)j6n
zYy*x3NhJ97Qj-6fw&r1UHD1Ke`u15UhZQ;{{%Y8flK?L1)0P4zrQWSbTF6rP?eltv
zWDz8)Khud3DyfE+Me}{IdDC@UR|fucO;NU{L4Pp#p7Gflg;0CsSkg2C$4T=YRc&#`
z=>`8LsV~w`rpb1+@evj#?T!+52(janc9kHGMd9>wAoZy%*WGxqBdbCFX8hBurnhyw
zKk*?kKAJT(KwdbUktKjqRo@-8(qK-#j2wsH*C+AvhkP=FUc*8~nL`JojCUo6W<Ebr
zs&azIpTz*Nhtxfs08Ph@g?^VX?bHGku8NsMtkhYy%tmZoCKC+;n`Fwjct?cU$_Qn>
zJ4xxkX8L>;&CY}ypRx(dem&~!lcCzg^-Qaekp%WMg}dkr;A`Y>Y-!|aG7udFr+h4q
zxBK1i;fqfPb;f32To<QH#<U{d9KOD+og)MnE?8La)ps>OIORa9tRU~+?9%n7E3_n&
zEmvDl;=vk|kZbCH_kr<**t2(OrsN$BKpRcEM?SZjK9wppEsqh^7i(xz;vy*;EK74l
zIX@RyI*sdfa!eju1ff7p2A(mQjpLw|v1((WLl~dR0tmo6PAPkf#eMs$xlr}|#hmwX
z5?h^Ky#9Xrd&mCm4`SR@P*n!~s{KLs;kwqDz@3Q;1izQ>@RdW$<2KMpz8v_;xz0BK
z$JjXphXOTOICfra+qP}nwr$(ClNZ~zZQHh;%$vp3%;I0n@>bpLt-gKw^mkqq0D3#t
zbo$L>5)oNTa|cEU@6Z-}5O%<kI1~=qzWU)G%S`F=_qHX5*pYWc=eBV8ExzPy3|IO8
zbyt?imB7DQ8|L;Ldiwqx1l&->*``>Jo2GR%KG3-riXrZlkwopPsxKAl26}V({mu(u
zh;ML0rj$RZfK%_Vqw69aUumRxj?0sqvEovvyD|2zHF=+(O1rxJBVJm$1&9+yloYm~
z$J$w1MpymfE6Y~kTa|nyUFD3|eseWzuFX&s_C6AeFf7TsxSj&4E1z>!34E70ah2y$
zSf=BM(dEipD2=N<O_vPHbv7thM@Wl5I%a}jlq6Y7{Fn<&p1Z?wBGNfq_-%{N#%HB4
zxzs?E!~m*YrBnF&f4<D84zfoST<ySlgcjvK5P$jG6ULP8KX5L-<iWxm4`lLkYJif?
zL<bAvpBdT5XU3CU_|)^Iv@yR)ujDj>waja3s%Lf4J=8DAGGc4lf|A4ATO5G`7V#0H
z^)vXbg}myGWEn|ANZHxl2AB2}CNB0m9}uzP#)N|j&~ee1HC~$C)~wy%b4|*|rLvx~
zze8pE5I8(VNKZ%tnkLEMOc2YH%uaa9J(v2IDGeqM^<N(>bv29w?l~nWUD%ffcB1Iu
zPM`(5gG&>DTNd-sSK?RcfXsW@a96cvu)%_5KB*F}e0}fOMFfU@+1Axa&@>bs=U<aY
zv<%pA3D0^8q)Vt*m7l)q?=XB`nQ8bB#j=oSM=`q*NK`m<G<?$pASp3Y2ip$FRf@`L
zdpyp9R+NzDB}B4GD$<?C41qMBq!OSkxeDX;_{0=fcA?a*j;_XScF@O=uzoUx1B@x|
zP9hTb?KhW<XGZRykY>h-c48;EaI;q$r0w0%&$CG)&t^twQrj+X;;$n+$`|jC#z>0u
z0L%Rb6gkRK379!~6e7|+Pdq@#r$nTZs|>f=&>(<llbl(~2fLnZ5pxM*BC!ZBDNj19
zxfyT<^pcei?sR482Q|cZB~FB|qO_+R8|h9DGTg|tOU2HNx7bJwTFp0R($X#WRp{sN
z+u{*G>;wl5rhQ6{?2+$Y-4ROE|40I1sJ|VVF^vNm$@8oyAxf~one_4&RVpGa$F=~>
zthe~8K+ih%c=|4HJPhp=fiT~q47X!0^*D_m=Qz+&+fE{yOeUl3nwUcQ`c+@@>0a)s
zMIY;v-iH27bCr<&XoH;`2#XDfJH2oemlrwq<i>X*QK0VPehqRYv0T6^KrvWSnq3L?
zRpxO5_58x9SYeb(rqW?_xn&a$lrC`^m6q1;L&N845LP8>!%(*@vQe~y3AgI%ar;Ht
z50XXvvBGD72_lHS^~ztrN<yl9<X&>JCf6;z4mK?*by@m6qa6g3Xh2LLx>g-#H=C+$
z==hStBIfCSnCFITU*oeRoQWrQgt4U4t5T|+^pMOiQ>1W(>46(6F(U<JCIrUhN#nF0
z5@Vjjb$oNbT21$4F^Hb&lV(=&r%?AP3~3sN8pc?)AOca@Ux<GUtlN69x$wx{viKqU
z@_iCQ=jkQ;M%PE_O}^V+ae}qF%%x5`TPu=e!%O2}+1vcmwBk4owI{23D`vOs*w&EZ
zg+Vc~OE6D5zidWFdEW-_{H_B?rggSAeT>&xQ}udHol%0xlRG#gX7>ypYS-TWOlRYS
zk3bf1X=Ufl@8dAWJ!Vli>oq17tto9~{PIkdXJn%%NQ*j&gLf>mqKtS9)3N?BmKOJ@
zyE;k2oU+SK<8%{sq>vBNWP_XUi=@=2u#J70e<it&(!irSwiCY=aCl0oEgU&~ab2{O
z7iyTH>qi;_<^Q=U>Dy1NPgv%&9@b4w|Edj&9xJfWuTNX49RAqQMaR8W+0|Eqz~TSq
zXZ{M$lk4Qe3oTJB4e8IL@0%p-qG`4RFmhr#FAY4(m%VXb_sx^<PVAv*@9=lN`DwYr
zzZ?$P9W$)+VFLBugpKPIf=s{X8mHbxOZB>9`JRgt8KcP;BGvY%et&ZaE?}SomvE}q
z1lvB)a;Us+_Q)<rQ|*(Uq{3m|aOJAj6Wyn^#&QGBP_Gd<<8)F|`#Zo7ce=NEVR6M?
zx%`bR7Y<KYm{dzX<}zZ%4dy~gHteK;i?J}ck#^u^pB7U_XU0Dt4}uap_q?}d&`FGX
zjoi{5>Z>iZ`I@v<MtT2_|0>^w;iUX#lN(22>H40z)43<?^Qgfa$Wp?Epg%QsG!!%8
z1LJ(inA#jvv#<~YE=|i>zgxqgP23cVony97lQ1+NG4|cM?kZsiEHa|sr^^CQMvb&J
znA;}@4Ah`IaGfm@8u#koMk5b%*vNiu5H~SV@>o(p`;@gM)zrOzo|Z$EW{&W6m+;Dh
z8mfO67r@|KaiXI%2g%INNR3fgO}gAdj>nH9iy>7(*%sF-mu$2ej8l&BRMD4I$IvT9
zzx!Ft5_kgZgaQ5JIWN0`L-+>K5;x?md7CuocBx~zynID~N|@R68^h!%9@1uQVhPJ_
zymGVDXW3pcskUAquMyD0qS{|w+qVY=&V2Yj>aAe)F(`o=X9U%v$L2;}#>Hv3>adQ%
zzhTnRf;?9s0+ph))@FR0&#WIp)WH4>+>Z;puc%fk)nISaF{YzOe2+teMG$)OY>rN2
zc@hTaAca1=_>PZZhU{VtTF61U8`V_nA?+k3tO&jjkqMbi83E2(oeLQhoo<5Z@S6B8
z@nFtTbUFwyK)mxOZ`Ige0JH(LquEM2U@QKM2TF)<5h(N{0e&oba7@@Gtm&RT0Q9v^
zk395Hyxb*ycJeq7B!hlluxiyPv_;n5JA<!jr?IZ#`<#YQEbPYf4<WXI(XssH7xTiQ
z)yFaijm}Y*mRAayH;)i0V^Qo@m?=<VNL=?^41!E`Inkh9toKHorG`r?E@M3&_hlx9
z`mk=#<#?iIa!;p3xP){43CmJioVX3n-Sa8Fr_?3!9YYs9(&BtQboUp};43{RZ)4vJ
z3qO~uNgyg|EIB5N<0Q~iUYm)+)g?I_+)j=+h}%uoEEHSF4HckMi3o@mahMFPA{{g*
z5TL>9PF1PHhYrU)uv=9dM6p<nqQbBprvP|s)-H{dEFJV8EO$~6)_#!Zu5w}YS!Yck
z)OB`+3<e-s&M+N|RQ|H#cn-m{i98<U_?a?GzHoM`QNGiG$~(7Pg-iJj#!>lq3>p!G
zal8|0sN^1ahwVREE(Q}7FKfkJS;vLJEj!0cpDab0dG^iuP<jg~1kGvjRLr+9?t!G}
zf+h`bxi1Qh+a<nqvQ`3mhj?-vZYHC-iMa;gU+udKwGhyF61U&M8i1W<q0sGtF}cYH
zhd^{XTL;6NZrcTB?0nbK6Yg><%&Ta6Kh6mIA4_ySzpTpwN$J)R5uF4dE$Hp=UEFbh
z3oRCpE&*3D&n4a#qegYf1jAys!a5xexyQ730!`N!*yDMsL&)>#d1RHqF5fEFM`xg<
zCJ`@#r{1pzV~{H282PeYdB+G?L(Qqa2cO&y2k3{-M-)7)GQZAVOt}fY#H;SWYZ$KD
zUA`Zwh|oC37Tp-X;(ou~7OG~iJ~?&Qc-CeOxu>!NqOLLI^z6I_$Y{ZmHU<@&=QQKo
z`Pxa6ELd+kJ%lp+)Bf#UK=H0~E|=R^PUcxXM;GhUTv6?ZOu-2)n#PdMa*uZaUB(s2
z0fdH_a5mP2(>sRR3@>J~nnza%`xe>$xLh^<={jZEWYikPZbQ`_uMG}y@v2FNgLGCG
zb0WN`8(kywJXIL~sq}q?=c)*)g)>~X&q{(x==gR|rp@s~CtX}y?<H?w%&S;}X|ovU
zfkCza{9?j3|06|e#mD?O+m+lkpPJ?)ThgA@UcdyQdbC3ehYLIf!Jx3=ci5ktd8MEX
zN12QqjhVha8lt>q#TvcEyC!x^urSJf5LCCT0EYy}8n>IvT6<6#&p%?f`PExkM{K`n
z4{f?7n#s$Xd<D)7UXWIN#=VB>&ozPOBgR@K3NQE6WfPQy?S=c4$e6to+0$U<Nr6r<
z*8Z*NrMqCGQq<F%jdKTS6b0ehRSsJnt-E?U)TvXPC~|u%En4)|-;stjy8xRu&A^T7
zK)%FY<UFn}<}B1%;0WBm8DEq2(yg>OB^t&*CRUvO^!=qwACTy*(-YA=^v(^w^5X<6
zY?zR*%;t6>RZ=7W(<&B*&UH`9HT(mAH(lC3mlV=mVVSDVwq#moQK!KSe?=*kka^+O
zV_p({9)so`oYMbw_;PoZ&rGPT#_sX<Xy7NP{z)a04+H1OD+%I((FsZ*gU{@7R}@Kp
z@G%^t2n!5JX&bTvd)O(0v*svUoi&}Vk+~Uj^{55;bcIpUGN%(B+Brw{p?tT-Cc~z5
zj5|JO4=7flSU&IqU2PZAtLn?znEA+yNuM!FG=^2gtib7QfR`~#XAng5>qwg2khVt*
zB=kb@I}o~sz?K!R{H#=0EjvIQ_oh#Om{XeAmF#q?i-<Xak9_@Z%g-^7Y!-0WCy~W{
zBpMiqiAHqG$Vk1mlz3ZL)2~3RF6AG-=0?NF?Tpar*GvX-WR{j;+dMTVtTd;8se&Hl
zl<l$cV;Kgs^9YI?#-Zuqtm$bVL7~&}SP-?X*>C5vUOI9NWPup^4)^eNT^jH=q9PD`
z`9xaIQl5YGG-lP5LBC259?dKo?$L|K<_kaj^aYpG+sV=1K?09441V-nNs<*=0AiR*
zi<>_ycY1|eoZFw3E;g>*gA96d6XR>)AOYbSvdZzKxVvayS9Qngusng4E6w0tJuSpg
zFC6Qg8$c3mu>R5a-;O!M;nXZO5w&b3P{E`NhfyEXDqx*G)(3DdOxr=Db0FeOA@H&t
zkUBM*5UA}m=+UqGAZ!6WQTe((Z=2@ap-vyt7Bw|N=8<|1{d66e6>;rPj_Dohf+8)y
zOsN835R3FSPXxf5N*_6@XOa=tn`}~g5R&A5_3z+OdS#XU03h2nWD60ri;tFL>6T_R
z&rwI@z)q`u=;gFtk@fb=d`^}4jCRj#wByAIk$*6Rm<aN7R{hOJ^FvHb9e-dqMxtN7
zcX?vpGWUk@*ff9<X3RjLn*~mGQH7-BBjQlHQ!7$xpV;r8Ln70p(7;_#cf)KS$G_p8
zI@uHP%y;+B6RE{Ge|wBj)18%F0P6qr0FN!4r7U-}4)}XZ(7KcJBkp?MS4zHNpyq<y
zK{H}I5L@c9n@gAaz}V6#6>;I7ipk;vDbyF=2IUE~OyW;Z!e(c_N}|7cfZ;MzwmEql
zVy8N4^528P3TNjO(Od~PcjGK#x58HP;mDTVh@~Jc>D>8p_EAyb&Mh_UL|}kz+Kfe-
z^U&Nc7l2EtmPow9K!uG$=NNRVC$dn$cPi?L*c3=Nf)AF_OR79pvL!_9>+FSJScV4D
zf!D$`qLImhDz6bXQ?YGvYwC#;FWP4a#M1noE-eJ;aO7i8k|2pV(WX>f0-U8)j5agt
zy)h=6Z_7r^>N1Us{;30jsHDZAAvg-(7wKP1R10D%0zP#w{?2E&v}%Mh9c~Vli-pow
zHua_NkcO%eqseSpd&2A^4~F=`95_*);vu=Smt%zfd4yGo-9h*3<6HI!@Vc-zg)N?n
zL!epZ;rHtBa2!3Y6j!p?c5X7sK20#Xd)BUw;%Q%ZcySEq=35CGU6xSaToCKJ-3-NI
ziaqpLYW^gW9R-ScS<QFNKXHCR7f*KOYl7&BYa`(diorp7u=1(hd_J5Sq52R2k@%>X
zcLj`7F!mw8$7Bn@x!61nRto-l_o7+CgWE+yQewOXt<XKRRZ-cXfy2G$8?+n}NkE;{
zmMK_L8;(lg^B6=}cB9=D+tLbuRTza+$6wbpQj?PK?V2bN&4uh2o6OmniT`t!cncT8
zPBv<<?SeTwY&Q0o0);L+Tho(>!c$Ecfw~@3>-Par!IWb<T*igmG-ZC|ibx3x7b>uD
z{EuVvj3K66PUH?6PuNxNO6oz_La*&6B_8M*>RB9_m@3pV^x31si)KDg!ho72vCh0W
zQ`b9|fw`uS;utn-Y*kbC+L2m0w^piLc3#V1iPKkz$V$gqZvLkf@ul6ujNxA1tbUxN
z8R+4!XRFJ~279te4{d2p<_fh!?zZ1Y`QjBQ8W$n*kHz!+^;AP5V^NdzaeZJ+z|36Q
zU(GK^g%4pv1hTZ)V_z$|ssc4i)x<H6a22EgelFmC(R7G!7xR$1z@C1^<$ex1#ptuC
zMnFtO$ZwFQyKBvq(xTrs;62_*dDvYIP3T7GVQCDU0%UI81(|(R^oZ0$c;1SJ#hx#a
zw&DQ`puBK|3KwNq1EdHu(-do$itW@&e!Sga;7L<*5VLW0aecLbD<|!;>m^i$13CWL
zI0CXA-NINm6DpO&VF~UDQH+B-33Lk-9R~r)MN0?W@fW6jE#ZmSp6X-K@asIJszufi
z4!kp&BB`{HGoeE3Dpyuv>UJ<ZE*4L+e`C99>AL7ZV-^K{4#mdCt9_evNj}IQ(0j=e
znA|fXs`(`%V8<6O3T-Y+5q764$p|SruR=&8c>(Gu4J2TPr9C|pAW7pkvzpQD{%icU
zw0l08bV+(VyRGDexo{6^P7vs5$3ZY}`@bh~)I|gi(%T+rS4;(DkT&x)Xpc-X)8y5{
z!cq)bV)DHNM^A{|OVoVo)SD|gCZUj;(fZrj^-!4X*vU_yN-q@q*0b$25j;CYUn4g7
zzNLiT1+-P6OC}Ny21P1;nm}(dx)oJ+WLxt2Y$mJyfX8&a%tk>)Uptg`sGq(AzWt%$
zqM?emYz8FwatOM+G81|mnp3g`uluiUfA#Ov-`0P-xUhm-0`Q{36&?J|OOKOO*kiQc
zafZZGW<TnUTP4##=)hN-Tiu_bGe^h8CxT}<`H9BJ5;vYDLZv%hrJ~o=Rq~(`nlcz0
zM&K8y)(S}4R3O{YggW;@J%66gs*TmyV2(7(NnGAT`4=6tWJ)^RS5^wBmrQ)8`?v-2
z(rs}ff0{K1g`G30!bVjM)u&fhJ~MxrP`8v8FuN`8K`5S}OF!(?coWWdAZI|BqL0GZ
z{rlXjH4gh(CFJ9yJIsFw!`wih`E8RPu0+AU$I%17pLU|bdg@g$;~+6yT{~Sz0S{Vi
z>*~!NFEt2v-~~(K!v+t(9)^Um-kJ_4oM)7_Nva7&p+---rg0Z3$Y|8#m~=y}t@1)*
zfh^rqr-Vh43Agj{Z9s5GplzF#;pZApg)zu<#&v-nZ;RvCI#HE_?`UD=HYKcIr*sQ=
zJCQqn+;cD;xYy1KN2b)!n1PW@l$m0A#EJ-c0V6rjPkABDJ=}LY=cUuJ%4r%haC`SD
zBzc8H6PK!#!b_hhq2_N==&yxVPTt#Icce?uHa;GpXEZ2w;INybdzBxQ9zq&e-{?am
zExU|K8QE_>5s-)-y^9F5>QljdEyU`7kvZlW&{vSGDe{hJEuvbhTxp1w%=lG<FR;@^
z+w49F<2gr)64g8{pB4RN)YH{qyhw9_jy|8oV8q(yXE=1E>8S%Wbo|p)pFI>uneX$r
z`ewrUGw2#OWKJDAL}lYAx)Ic*bY3<o`TQ!gT>Bo&h_k$56A7LHs%#h(ypx&B6Hea{
z>{J^81U=(LjjE|1_aJX&_Bz}kIX{KP3^UWp47Cv~9CW07=2L4cLqr{z*Ey3RAXs2;
zHSNR>M^fnATr@?8D)-F0ZwNwmb;_OskezB0h~m3mJCZng(wHen*eCsF?_1J`V{7)M
zM>AiaU(yL&Li!GFy9QHV4_(zM5*s^~)gRa?)S=fe33s$bF7Pgnt+W`y*r_;hh0YBR
zG)EiwCY%1{1@`|Hzjl*L@?*-F34E&-z`9WrWF+Y#L(gKQD;W&tpzZL84!<>z*mQB6
z<ZEceL559IMGXidnFqW)Aj6Q7(jg!FOh(_#88L4U;ozjE(VS?8+7erKY{Cmyq=HxQ
zKb{H4-=Nek1#QYv@hcw7p}0pPpo&uvF_6M+p*n8NZun<C-p=fg+r69TJ9e)WZtUj{
zK*$f*No-!3^3vY9s=f2!F3MWXg&$d+nI6pAq!3mpT=+<|)Gz|;Y-@+Y-`X_ZHmWOQ
zL5@5|TkJ!76n*!dHi?w<$pVcH^nIfmeS{6WA;c^kn%ZVd79^F-S6Q#s8Zj53z^o9H
zS1T$ojfSW|ZpR4C9Pft8-F;?u*`6X;-~3U<Go(9drs<)ZsZIYz=_97T<awqMZ2gg}
zO7XCAV_~X}NtQNtSfNI!`T2$&Xg`q6h{rBlpg@8_s9#``j*G!2HVks4Y{p`Z#?aS=
z|5lveD=)O!WxZ7E%*{QZj8me~5Y$_+Sa!PbKbET|b`rmnP(4b`0&rk^xUy6ROvLUe
z(t1oJyfR0~#@haB)7&^P)9Bw1aw{`y+-5~*?2brjnWdZEBQ_I0UX9=$&XS<0S7uQd
zrPEFHe;81%Iq#ov3U2io42Z0Jx*kJYW!VYD3ab0ud7WgiA|Uf?IrVq%SJ=mr`X>?8
zn653s0+Jat?Kt=HU_DYVyuIob0|y=`s`~w{Hm8t_WAV!{kMpru5}TQ$j&!`Ny(mvi
znXZk^I&NXfvy7HI@4tpC)=R*(sk$$3G}Oat21)kKKc@m=<Mijs`UT`VEOSuTyFGcX
zT=e&KQ{mgLKxBzEm-hzk{=4Z(nI{VWjJd`Rs(4gelM&iQG3C<ShjkR-oA&&n_zoTs
zn~NT^4}Yi;&Bf|>9i~ZZQef%B<NZ`NY>$*p;V}TL^SZ&|S4n#qM@XGtlYb~PYV4KW
z@PSL{nQB*P>S6?UaK&k!sEeAbWX!N2-7S72*XI-C;Gq9j06MFguuw|pe+SvW7yyT0
zF6($qn5X$_Bt#PxK(?eW6i`1B9i}zaa61}Hg#K<TmYWo<ZUi79J!oLuFuo3G3;*oJ
zVf-}CJi=a}duqmYhzFIgLRpD@A(R!jTNvL@5;t%<G^{Gvkm8GRR^LIn?!nEyP)}K1
zJb>~@HOustnZY#L{_W2oIeJO~%l5X#pkpqEqlHsSKPyMz>6-muKRt)3{&+i#l78S|
z+3|-jM3l2x#=BOxm=cf+;v=rV&9!@1=7V%&W_N8~i^)o+Z8D^Iw$CFnR(H$gT~UJ8
zX1L=`Ec4)HxVJ-Z3DQA>h67ibF$ic}#$eWaJvaOY`rwW$<5!)XUg+R8LDo`<dcDkV
zRzx4^go;YH(&RXS&9oQ(MuD57rV<xSq#hmOJVfVr43ZjtQBJpg@_LyP1=BW8rhS9m
zw-39vi)7)+1_Z(hC9_|zNq{_cCeag?%VEESbBJ_WfXC<gDZ}dR>s4-M9(iDpQzwlL
zWOu12;#16O)fBS~*VQc|J;vZt$for^@MHlmtOGS``Sr!S;?K#8LX+$!KxNn1ND0}n
z&(61`TR3}!P6Wz(kEhws0t6q7h1$ke+bX?Uwb<OluYK<UDa%09FT~=Hibs130tez_
zzS9|`lrv}RDm!lcxN~?X$zo#ss(YEr%~%Z<N}<T$xgeGWJ1c)+GUx<z`i~F!bRG}x
z!I$?slK*~(zVQciDO5H(FLelNC3nm^(qdVlz@qT&t<)W71Wqzg8^iv9$~G^C+xnlt
z{P+oz$f9>M${DVOTWKi>h?z6Se?uuHF}SEBy4lenE1X6el(RXdy3c|;>=8UsTJJns
zNeZ^(WRL*Mh6Fa4v`=D|K<#0b+3frDRn;Nbj6ogtAzbQ}+MMI_W`O$HGR!-9U*QD%
zF$w6n$|{FkQ{7Eo<_v%H?d>Vn(@P40HVzvdc)lc!3X4qo6rnY<1cz1MJh7BFzem4H
zF;`$0hyt(1Mu;5`8EK4*>$TFzW*ib<Q&80yLW^i;iIe#yZqJaZkll!e>Gmj;1EuH^
z88y3E#X_sgL8jX})6n&YG>}Cce)<#B7Pxvj7|GU#g;eYMYRxM{te@SK`5JNZj9#=j
zT6bZRwY>O4U->Mx=-k2wvXi4>IK({0zh}ZV%AFn8VLA2~saU9V6EzgH<QB^6;gCm>
zBa-~wXx?%V*Ogi-jSahZe;M8r*VRpd_%5~m6RB}P3TxZ)9w*-xpmw2GI>#Ve_H+$D
z`-PtK^Z#Y|$pJ`|n>yTs2DUf%$IB*-I-n<z3<p2Uv2vtPZOub8|I=`}_Z#HfS>#1|
zEtoUTQV(^?pio;K(UnL$tI1pXvy?J8)@U-SPNYN@)N{~Hc*BFJwt$>Fqz%PLP*1#V
z98noOT=XaD?6F6x>mzBfk57pv51S+`cYln0@%6e}maG4?+eDsPHHAohp1QxMAxBc0
zaaE5a@lQcIp**dKXOID(DC1tt<hUyO0V9zNUs)0qa|{<TT%)=nl>ZDx{FOUZjs0Uz
z)s-gMKAo`J`vy1u;C>rUqb<jxem6mF!c`Wv4!BZU7s}{+*RrQ*tudP2)-G}MiU5(f
zkW@OBP9C-5o%obK5GVn&@!W>W_b~b8pWv?NpJXRyGg96r^2a24UKG>)Zv`8adtXBT
z0S4?%{??GqT#gQ~)QS6^dyGVCq^(HTW}c2g+OqovirACViHL93?3wpiVVtxe<a&zR
zqohr5(qe7==tO@+;3YaVPrW=FL-={fVod$sF;4C<uF;yJyV6pF`?<KUxV{#|2mC)K
zmZ}G-$`IJeTwR09ENZN4QKQ<<nZ0s-MLGMF?cfKIc=WKO2xlKTr;i!62<u)j7rRNw
zv;SuE!A><mLl`ow<m#o1Yk?jXAeLEb1Y+V=us*lSNlM91yhPqjP{&Z`;N}l?R`7bI
zR;WUAvXY@4i=KpfX|;w;J=!wAx2xCvBhP?jU*f;M@`H;i15f}sH``T<GY=334kXh=
zsT6P*t*uXrX%%k<x)|UpU%yD5FR(R`pny(nMLsfIXSPR3zrn}Uv0F%+ewmaUs<Zc{
zCBD42?)#&)qgANd)n@=)`{Gd}^HIQjRBHcbv}JLoEcZv6o0l)#FXJ|{H?P<47*oE}
zQ8Tl1E36HY$pkRDae=5#F47-U6)q~je+B2#;((t`Jb3g5)Y|>1eQMG>ITf!Xx;V{b
z#Uf};gRgm8|5(#olB8sP855mJbdH$t?6Z_ay|+kNY}Lt*{^?o!ueKP8%GU88KMYmq
zpt>C>z&|R@4-6p3vB-{NL@s4=mM4cKRz6fJJ7ew_b=wS`7w);ZsOBV~I4>?Bepf{(
zU*??46DYcVY=5dJI@#7g6Eoz?Mi7DGri(ss@k7%m*k8flBJ>hhZ39PHYQB`xpWE{h
zw<?=Ss;>WC8`<g_CDzDf1yU7hV&)2Ozh#q0p=Q*;?oSHR+<L(>Z%vpwSp|q{u<+Bb
z@VTSvu7*Q$18tn>yzf9-8Idz8LpimckJtxuB9I~Qvk1u9Y!2s;ml{)te&gSg*_;?2
zq0L!)y6UR?O4YPd2)thn=+*&+TatKTah^4Nt!O+MXA`v8b-<8-`&}_#e6OA)m|1=W
zuT-uvCy9xcv=((S-rkPd<F;;3)(c<cR3RyQP7WgI-C;V^g~Hk?8uZ~I2YSXtkdT+!
zCc4uR?1N6mQ~Ka@Y%#T_#|{p00zp@hfhU*5HUjIob?=!<TLKN3N#me>##uv{L}Io8
zS9B#`gj&%$|4TFF%_~T$#lrb^(g;o{-%r;<G{Ty{IC*`}VATOer_@fJ=df8#cCJX5
z{B56KKT$&!{H<(Z>tY9oSla;D$F2E<%P~SAXYoUIASW%qo7qDCSLK&Y(&G4aYAaLF
zXLX({{FrG#a8yXc2Ph&e(Gb><9<)uD9D=7h2W*OOl_|eX&u-8RVi9oEBy*~A6VRn%
zY7rIi!mCBadN8ziGOgLfJZY^UtoPrveoouVgvKbEJ;OZdL$0XW!k)ZqODBRacK9OJ
z{cS>xovD!&y8H8+yk7$&LL!}c+3a)S6V!M48vohHmQo-}XO`3qgscVB168hcf*A4A
zQi|FEUP-^i1!!dMiQvV<W|YOT8}=KpKk|#~@mOlJJN+r|1GH%sqq`Q<;PzFk;fEZa
zFG#|o^aSLI627Yo)M*~v+#us(eeJjGReV=(e+erDY4@yXn0Iv?Z584YqqdUz9{l8n
zZ2hQx#n88LpLN>#u(2R%38dGdMIoXBszw#Hqb*ui_>uckt#A^YSv>M8mOcgCUx@4`
zvYi@-gq;ARtjR)`+Q<`X<16rT%h;>H0GkB9%_C>%t$SVhmSt<s#*_=i(*9RWjxMT}
zqq=yvOqcr6*T-<(>Wqr>U`c@MY6p5SOZX4eD1bg%l0NhSA)Izw&yy@Ki|2~2qzHtl
z{u2eJQq{2th%Q-pjs9!Ez+=O<?HHigtVPVF(xXx1x$X&j$Wpfz;ISxEuq4Z*6m-!i
zZM`~#DQDRGV}*z~EBabju-zw)=Wb2dNb~%5`um4(3uK<LiCsZ~)hES>#@w-0g~da9
zjIng~pl%mz{pG!!=-c@u>>!w07_M&<-XLlVN>KK$tTT{iY4{TC>2^Gsj$KO7)i<Zd
z)XQQyc+gamE1It;?ppKV9F2|7{N8ASuRje8!ZD+=iEjt5svPjXFiqSB9H<>~Y=*gh
z*RtB})g>y%;Fu`uB=VQ)k$V{vLFeB~;=ry=rVs;1)kisqm}hxzl5eA}1!EV1Pgem2
zE5^w-SC8V74K#1```2S2Z1i-%`=-P6>ILT{B-c`&B`S=PQGd^nc2gH|^e=3h*jIW8
zU8F}dZPT8rAC|r)y7!X@H|4MEAq2!26a!J@UsIfQS4mUErtFo-<J0U=*f!nHG4-~I
z&97|UH;1huYxO@|vP`C5vVR`J^8NXHdJw(fuA^L2a^A{dFccA5RM6Zm>W6Mwub`nm
zsQ0pnD`V%p48hp0Z+x#(tgkZQ^(MrbL6;H)1@EvzU8EfRK4}`pEG(x1`~!}3q~i^S
z$YDb!RT(*IkICvZbydwkzG?rxHpS>6*ck5>t~Td+lAQ+W<o!~hTPippMoOavPKx+(
z8cFAuxg963z9>{`MXmSA7V!*rt{lKvUI%quyukrDGY9QEU0;Q3ijO7EvF1PC-h3N2
zZIV?t*P@nL8*tg$@2smR?dh|8=dVqRO*Rwnj8?>!)OUsqId;h^z-y%hpGjS8X9K#N
zED-Fx*5-Ko*HF(N42fFv;xWU3KYULj_nK9yos#=?ibR)x90)r8nZK0}`|`(k?}BH2
z?x*9kzcKo_gNq$V;igFZ)#2P5o#-0-v^$)e+wvl`r6|Uy2NIMVQ0K*7v<8H7*@>=h
zVPF!w*1GGUr9(0r`2Nx8+mVszh-vjf4hv0B5l;&9kt3S)oYWrfm%@W^|7xLj%JyoU
za(Lf(Qch8SbEs1&@ml<u<hEY;=>_6L?z=qo03w>TXD}v6J2?{l_jkp-S=9+K-E?F)
z@&?gWLkoQC4sA4!OE%TTVHy{r$A_Y|g<qCeA!I_tavSV;QLNXRk)YIcj+3U^anT|G
z1G0BoV_PbofW{WUqtA;P3<`L~5BII}CDVH%5(R#to}Wm>ta3zQ$v<@D!OsK0G3RQ2
z#3c__ZVv@ds<gX<0_@D!jiyPwckh0-2v?6GTRDxVlDM%?UlH4vN`v=$L>MSVh_IMe
zH*ywlAHQ@oHL|51b??)zft}5HkD2H7%c>{SNFu=qzlS*~>;VdSsy&Hg_S~vE6c^kD
zi2tbyjCtk~_W+nNYZ`yJElgT0t%MO_JQ9SyRl^utxrP?kvP@Hc@rWvIBiz$D$b*>6
zC*OHwPbT4V;vHs;WnYQ~kHCq76Sd_!(Uzlh1sXNTy0E}m6-liu{&St+6=d?H0rg6D
zfc#;kcJ6Xs011TT6CDizIAVJ*{ZEH_FWDe?iZr|}>-PphZ?{NmXuuq&FL(Kv;etKx
z8BeY#6?d?B1M0@bzj8#R+mKVVnY8mQHm1MPFB^2{nDBvk`5d+X^qRcDR^1;YWcl<!
z{4@EBk58^pHDS8&C>x~;g*jDIpA~@yIdlGr>-}m*-ld{xX&L=vi);#`%#^5k#(quL
zUh=$HuBG>&sKs^w;r#ct+T(<o88*RdR`>{%tx~#<ANN=?Kh!wxn&3(BKAjP>1CJmO
z=;<=k?=?t+R>cIQ?8V`*1g6v3CRfk4t(@gw@q#igC3F&@MNPTEqZH~bIbW*vQ#EHZ
zzGMR$W~?hwVc9%KvXCYKgIn*t+n1pIIXmDF=-*9~OGd8;6)c&SxcptzHUefQiiBc#
z8_;~+L!-mNvxDt<B(eyZL57!k+<vE&d{rd)J1S*7P&S*u0~bO+$B)`*cUSHv8JBiQ
z2z!{<2AXf}+cD-k1TiDmq4M!@kAPqO58(Jd*F@@_B29M&VjCQbta|_<_Y@kN*z_k<
zv1Jr_bA_3a1&h?@7bMZ&RD|1nL~JIuAn0zMtkJ?^YnsI(T?$ejzWzvIK9XtJ^X;Eu
z6KgiJdt8xBad{tK|8b|kV6eQZ1?d@`lj7Q-`w?B3KR4%GS^QHJ6N#Rw&M<{pk-t)z
zxXagnFyU|$(E+lz;P^wMikY?g6d<pYNzsC#=2S*eUU@-=*T^2+`|;BMj!NvJL!hd-
zVO6yl%5L5SGq6OI3$?}m6*AeEABM+<zJjgUq~?Ux5%-D?H__y9ME6?f+(dmHo~owX
zaieLW63!wK*dfN(T31P<RSHF^?P3WlthExuGRtdRo--X(FlRaGW|wX#L+|YCB4AJo
zjR{xEy`SvdEDf*QCej`~r$#{<od9P>0z9l%(^UtZg}w9)d|xe~js@a*)?^1<94$^l
zqGG>c-h%UjbEWGj<Jh^)E&(${*t3Ux{q%-#F(5}~va8Wq;!%2$F9O3<B$eXAMtS<6
zLg?V=)yvvNeo#4PpHwij8z-dp+{idB<RxvvttD8CK>+0xp{Q}=x;*;>5(`be6(4IF
z(72hY-gYjL2WH@zj9xKR8(5)qdY=n`iC{-{O(#2xv*i()^7+R9Dw+m{L;I6Ar7;+<
zN+!W?@v_}?Nbm5kYo5@!7`!D*Uaq}TW32XtZkm=>IMl)2YZwss&521GPF>BW=}J0b
z1aai8I+9HJZ@VYAfjm}DQxJv>8b%!83q_J_U1CkFF;d22aJHGODe7@$Vxp)tiJ|b@
zZ!YQXxUf4nk6>)q={nA5S-V!#*&d{WJ%*aikcxH&0JRlxSEtv?^IN}jk~#bGFfS}N
zQE*Q)Uf3!ZGWgK4;h(j!!p$*0*fi~D)^`C&sk3}v^wn=bB8h(#)5^Luw(ZTBWMmRo
zx%s}&VA5T`tmEv-Z)6{K5d*FCSY|U@wNsNS>n8W)K0<`&_+hiby=N*kzKVWvK-cBC
zfgc}*Az+TqRkRLPB!IvZOnBk%FI;7evPbIKtGr&3UUB1h|9&8Ccn(mAW3g7w;NaoJ
zxNzcAKU)<E8(MobzQ_yMz9NVzAxyCMMM8TNeOB%h{!q(1bN)Mtp;a>@4>17>?eM>;
zw|${nK>{;tO=Z?ymm)ihKQK99_*>EtFU4myUfin?qblUsFz?f8k|AWr_>rk*DEGMU
zL9M~QlAv6|4tB^ts)dA7t|$<Kf%DCwK(%>&;(X7H<Y@R=+zmvDAG>E*H#aM_J`*Lj
z_LjGFzq?0x0ZeKS09pK}?tH_xbp(ZQ;DUN(L0y`n`<YL@F8P`FJdbtdcmYb#V{~0c
zHo=!B5!&HuJn!5|OTDPoSFE*9%SF~HA+<o-^6u|H&SQ;FpfyvtGR<!-bxnb@tNmzO
zoRZr8`qFQK0rFw)pIG?(s+2D8KF(d%&*oPpl#}PKZq0)(#Lq3H!6E6nb?M4@^8izs
z{h|e8=ghGx*i_m(f-6dkYzRsg&9e)mr++7a{2|acbCH*jFdbt=Yp?R%rkRWvwO~L`
zVzFT|`rN#|qp&%pe`=c&;Vur02^23-6>1K&C^yoxRs}5L0hgGs_}GBM-UcJVE?6%w
z8ZE#w3d=eOGr(m-_->sLsL{QahT)Vp-G-^pBAKf&haKJB%2&L$<``Y4l=qcIfJl`#
zg3kqO-&~@$!Zsg7Nj*Jp)Se1`FHJOVdkwlobRkwq%F~Y{v_*+l*dC(QqjF>q=4+I$
z8GX+Ho8jVcM$`Dto|!%*Ppvj<j9j<%6<tS#L|PrR9p6$ih#$5^E^WrB!O!(hd8&7h
zB5^67E?#mkV{#{{7q9ZB<Qn`++j(2|dB$;d<!3Z{(-IOa=v}ws?w$3;3ai9?4GqeJ
zJK2quSEG=1*HSVZm6Yz*a~~0og3!IezNJP~K<scjL>jqS6!r<~3}E+#B>7va&2x$&
zRH!r5*M2;;SJu_P)++eo)C70_?gJz6RS&i5<!bQ$b}r7EWHcCH6h)li(wPy&ogCpr
zlD5sg%t^i(-x#V~BRlQdE6(lpSiQKm#Q1j0!r0l|CVj>%PdZGlu)~)#6v0HCj|#H&
zI#q|2-d4xk81}@89oE^z>ZJz0x+%}T#Nu!?W_fYX{()}>4w@d6wnM2KA+{^qKF;=+
z!>(tyUufede_<QJ3Ib~Rqe@kL_X`EpE4_DEz9NLRBNmqp-YK`oS5pI#OWPLMSi|Wh
zHUNNMlKIyeo7PbKdP9GWFU-68u*{W11g-n_%YGMhb&@FNF%D^HIz21uA)Xza1~uD*
zn|<jrtpHE`zJeW#As_0kjRn=CGWeHC>DQWe`X?bV!&EbGQRJsUegTRdi9%cjpp(2z
zipn)qUuJKX_zI%zpOzySf~~N*10lPFGr`g&;VrmJMTcgrY)$j9Q!86d2K?;ZXL<!N
ziAfml(Yl-P?#(HV^Yhy~DwpoRV>+wr%mH?sNt(p{)(TndC%GL$hPyRB!=>=SG$qJ%
z4<p?8x+FfTZe4lrFDlsLvp$>$2rx;yi0}aqz_V^Z3%#|Z{gl6gdA1fnR!v|{f=qKh
z#JK}E(LOle#Qcp}FWR)=%T?@wZDa<-#E)9Ge)nk^ULkWf9|zweDcPe-flPMNJNiBX
zWEI7@p9KPmPt(6-IqqlH#Rq|t8P7ZSmx#?S@Fjsb?%T9ZAF+@V#>#KFrcnxn->0{y
zf&KerR0K>J*c9}Gb3*jF+bZ^MgJb`17%9L_kWCd=DL7^j+}AOvoq@{SO(jG?knNV<
zFiW^aSrGNaoF0P*Qq`g2ue5H(CWolKd`i2^oKJX;U(7erd7<NcUoVB)o>0I4bqOrC
zXJ-n*UO_GKjn_VLev6JIrF@*Gn%FuG^RC>yx`b|SN)&1(7_by)1Z~Bh>blMB=RKfj
z93L;WjTmG#SR>@bDtI>$_^29KyOj*aC(2HIT|i&W_6z>EO<-krG+MefycUwb?KXdr
zQ@{CgRo4L5Wm?X21}vk0i?X4uMDNy8-{d8{j3d@lo&xo4i-`h*>(7?A4^>);p2xlc
z_@tOvU+lH|p;f#<0AIlOa}pOk;cQH8#&?;fRVVcRutPD3p^sOB=+#0ft?a)0UNaJQ
zz+}RQ-Xr`r45s45X|k)9ArW6eTI~bluSf`l@BvYg1h&I+Sk9c3+P`$xYKjC`kFf%@
z-_*@J50fHZr$}+K3mWCs6xIXMw%N(|zH19@O4SY*+~CwoaOfB~C45hJ0~2z$M=+$n
zC2^UEO)WTt<-MD}p9V{o=M@y<(HDlZIW!4!y_$qHgXO^(OFQ+*jrbQXvNn%@F$~j_
z1!T<hsMy$@xcU!?t(?O^4tM<F&rh?dCnQKfix)WnP#FvAxY65kz}*e|t*ut-T-tDj
zkm@5p^-o6NDZXy`0(vb_=@I<o=OPjfeU>{}XBj@B*@A(s*VUj21^Ks|QVi5M%hc?t
zT8uej<{8xH#tEVtllJ7V1Nf!YgoHj?ua<INjoR`HQE#~w{5+1;pK;y8peu_=MIv2v
z6}<4ZueD)HK2WDbg3==}6g}@qZJL8lZV`zI1^<wjx-`V9gS3%Ti!k;tL9SWS*I}7h
zO=&xjBbTQR@nQ`mF^u>^-6yHBJwhR&+zAT(`XDlR()19RKjnE*I_UC^_@IiDAYSMX
z7|Jund~mCaP~nR(ExfJc*fq4|BQ|5d4C^xt4jydsBr1pFB9~k`*&at7v8zj@CHb;G
zU`nn)&C=DWHla8?+K~@?-*?Q?-jYnsiw3vQ?nMiDm;idsH4_?;o*^g8qo>DR*h!=b
z&<`eA2mj@<T0v~5F9R?1-m$-3Zqu7JQePLjm0H?gioGO|jZd{<ytjRM7Mk-;9ZSgR
zGF0Ww9@S~uKQk_%E?C*UPhTP`qs&TvLY~Ykl7sqA4u_$`+p)uI{cZ=Dmk{*>gw!P|
z*j%k7m#qT+G5{`Zl~9o+4>Bx?CD{9ZqL_3D!X=xph)#Oxi^C=3pY4_eS*Vh=QzI1e
zz3hBa?ZJU?rW6@-j!q=c5}YPZ;(EiiT}01nosieO8kfCB;o9+RTWGAXAYix9({4vZ
z+yb-nJG=Nvs(OMuNMK_p3ZkM)qF=<X*G7p{Z(>S8m_h^CU=9r0v&ycu68E?dxJrUd
zAoY8^He5IW;n7(lXacGn7cr}^{@K3oVjoNFcp?(OhLN<qi{6s<4fbMH?%86QP`?HQ
zCq1u9dt7|r1^(!JP{RldsX<C(4728Ld>Lu468_-WHZQJqhV!4n=l9Q5QLh<#SZx+x
zci>f~iyWn5lgMH}W^sBB>x`?4y>in89*?zhP+8b*|5Bv&4$ZSp%-sE=!+t>z#ru`^
zG_i~LJ0EsZh^sC{J;QmnI9q)|F)|4hJn+~k=d8qtUY1!0(q*E0ju=^xYhhfgv&fn(
zsn*pgSyduxViZan)%W28uSwYlynoSl^DY*2;5uxruqYuBh+!J&0-WMCc4FP;<(Y#p
zB)6eJOjEgNoKnmxcN|Y&;NCMQ{15;&-(iUs87utr?HN}_<enosL8nZ0eUAVQdeb#X
z2gUS0QpQ-xzX~31iqw`Q?-%%2*23b=C3_p|UA7A#kZ`3eG6iY>n4Li;@&Kau$?AFk
z#u2GIj@nZNV(%Yf1SovOcgYf~O@E%+j|CYr7WM3AXlZELTl%Sh#L+fbzi>dP=qV{f
ztW9_(IHsJFw>WE}#L(}g12|nyh;ML0vj;n=Mut+N2w|2?uw+WTm6?I^Z@8NiKhk|>
zBN<5C&P3DcoC{_=mt8F~rvE4kzF}Hl8qy$2L_D7?t2g<pId+Q3HCS+SP9p94im#MC
zzQu;9ox~;d?!8F7QhZC3)_B_*yro#hRyZ0v6l8^OKjCn0;Zh$jFEW`UZv$;Fx5Fdt
z(@Zm~pI`!gVd+F764O*BrNwxbNTK*5H4UXXl}iyssd+q-=x0cvca}T83})p8yv1^!
zkF4mL+KDc&&e2K)Pv>lgoRCVQt*J2XzW^@x3J6y=W9JlB$o{Ad;E+E8Twq6*mXVzc
zmIh6Z{HYPeesezRj`nxJw4z#-+WBx2W3!AYqHGrLYtGWd(1zXp!L<SUd{&7E26=|$
zxDL~HASxBCjDPH2J76EsLZ3M`1y5@>1$z!!8N>RFp}lW>P{+d5Oz^|F4)7MDx0`my
znBi7ttd2<HQ1EM13D%r60a3z5toAA!M<KQ}pf2$C#sC5{tn4O@_n27$Ep@k63bL{*
z6^4ms=HadpD8WG}vXKi}ms`Axyr}L2+kTJQ1URzD$c@DwDQ0#RFbWp#bN$ltX}SY2
zR0!N%GenadsY$AmN+xFEo?@p3iMOLn3eb7wZWp`I8Z5UV)vC5FugYT~n!(ZX&sUd|
z&y#mpao+_<&8v#tu(mZD3zQV0A!NJFMcPe?F3Rv;gf!+RCwDbdlLVsLF1~8OL&h||
zNZCWoFr$1WQ8s1w>1DWX2f|WXR8UGgIo-NCwfaIztL}4?)9RkaGDToMmx8kXu|HVj
z1uZc{&rx7G8;;kkrLg;*u?%C%Ee<h<cOkow3~OjfNJQY>MiHJ_z0#+v0RQe`5{;$K
z4LY2Z`%<383~+rMqI?PSgS9ew7Nf9VZgPZr*b@X<YIqlQ1t_PYOVt>W8IGI=TNbcW
zSER93lCWAh;||!q4FYry<nEo6bu)B>QMRYFMb7a0tC!u@)@ZSvvq9A(qV_<Pr2i4%
z+o(|{GtXVtKr>!NL5?5YBoj3M07BcQ8|o_x$_hS|E(h5Wh)*xSw>;gl9Ful;226&W
zBlgBkCPAYC#V$v}uZp0O^+1VbOJ!7Kcx-T9gWCA3bvR34=8t?CKj#?N-t75xt`C8m
zn)9*P`B_n1W7m@)?TCH*{2wW<tpAbX%0&Oa6jwGTM)v<X{ojytb_N!v|8K~7wVH`~
z@^woI2?~Y4KtZPfx?)Esu?adDI<~34DY<|oCx`_|5h%sEs^od<O?+ZJ*m#hy*{x5_
z?Uml?6&}-S=SufV_k+g{hN`Fzur7yIFsv{Jpx7hS1Ee@W^plLU06qE`YUJEeiddiZ
z@*F<;9x5V<v55-5Aip^|Pz{8j5HSELyp}&92n;+T5Cl*Fzo4o=(7-wpWV0JTFuSb>
zdwaXbPw~4AP~H6(xZ`C+;Km@Jbt{}LJo>;oB#_vy;#VFHEIvHI1|U#f&;ULFD2|R2
ziwz1uSY!(VKPXTbzFpLuOLKrgKY$E5I*^zEJD@8BF~Cb2007Zm%iL~p!JBRSkvsrA
zz%dM<kli*uu73EVT{!q2i~y&ALfNJn05WKuSm55-CT>0gz`wrSsrfsnuX^XtJsE)+
z@xjW#Ke#)|qDcN6IX&V<G}xBtvWoH#m0S}r#5Wf!lqWPWylZP}3q;-JZQ_r9D|0J3
zCP+vjfO$ChhZ{8(L?3FTV`u=kozZS02T_6Fg1v~q{>$%#Ke&JJT2OooJp`1Cl+)yQ
zC)aNU(j8v~I0x4t_pdwAkNP(gI5{S0!1_G`6x8+zNTHq4rqF)ySMRC_tZf(o1k_h?
ztBcDI_z9T6j|KebZO(p|43Ke@{~$MD3p|&guMNltE3yCKFRMw<?^gUTFwsvO%o`u$
z&hM`MFK^+0gIjw~-*1;DYUtBh#~>d@A@Ab6!15vAeC7dscbVnyejfli1PpS2XD)A3
zN#lDJzhz$BuIOwz+P<%)-2=S)dYj~v8JB<!E1^Y#Z~$EK*KAsU&SLgj;EDek*|%h$
z<%9sc0)c@124QgW_fGG2uhG7t1N&@#b3d>1e~f2p3(fOuD@VR>?N<3(I$E{-S)ZIk
z0e-mRBTxrk@&@ckJv@TJ`=Zq3r2+b=yt88S{nP0I$f-Xbc7?x%h~MZmpo2vPc>qp;
zyn%oM{Wov@ynjeEequx+|A?;r>|g;iL)dzCTK+13h4$BoA6*CkX8&vs{1W_XctQ*T
z1CeH9iuscO<0L65&6&V4a(*CBZ)$?na|LEUQo%YkeoG=u_ly@r^$}i9#tC?<03)tG
z=Nj>4*Y;F8Rl7FJ_pfCk5p-%e(`9gIHQvtFvqU=7g57ByCly&|2z!2GFKhQPcv0(j
zbu5*Z+CBdu#`)$ephx^ZCNIXV(`?2C9uK|Oz_FJTW;R4LK?PM&pN^txZH=5mex}{b
z#MVxKwpH(1wC{=_)ky~8z7c=vM~iuB62(#WueP1q8kQ|YrgL47I&aP@uwSE!6S5@t
zxHu2-WRZ@C*icmM4MBq9DrsH&z?goSiL<}=#OjX#`i)Z?j#`C9Iw37*44O<QJbpQF
z24zCQ5#1dJ<r&SIIRs9uYfko+VV@5Kd(q622B6yfRd)iiBU@?~ctT(6U>yV}Mv-$%
zjGyL`ufz(Hp&Fn+p|ISHfk@hBYW_plU0nuc1^X!4<51C9sEUMJ^QnB&FpPo<p8KK6
zXv*jq8z(i(^br13&h6LyX9>BzXUDD5!a<)6O;CpLy<2M1=)c&yr{>(cKuyE3xni%_
zwr$(CZQHh;tR!!2+qP}n>aTWHclE(Ood00f7|*ybt@al215~y4^ikb0)e0Lo?oA_(
z)t$XKh9C*jtihhAj%g^El4urRe<Lx1?(BMuyV&rvprgMkoNqBafB&tfo{)D*2ARhD
zr#)XL41?$mld+4ZW~ZvJ%M5<{Hob2vRn&k!lj;6t;y!sUUeQSPaU;|hk+MtfxCWW`
z@F{D%3ddmNGC~7Z8Xj}3eFd0YeDXeNAde#i^|Od{bx#{DZ}QU1e-fTJuF^901Op5t
zgUS#|uQ)0%IS6CBLKaK3VZruG^|3%4ET99LXl_MPcZsp}AV1bR=x<G5)|4xs!n&l$
zDzpl;evXKNJz1t^{h+wqDta#I(+6Xy*9C(Vm#~a7?LCS3a-f@e)fEP}K>^Cn>-Lyz
z<*`<knagt1?^CqeKEwWygcR8xJ&V&I?48^q23h*AUmldd0)x58u1V&$ITvJ|3EvOW
zjqjt+!0SN-Jz5eiO|lDZsRp~|5Lp_U(reshy(AaP@VC80PK=2GH)Bs#X7@U*(OCA4
zYCx@cEmR>ECOQTPmoOy{nS6X`JMDc*x7bTo2NbXIe7;))nnR?VV<vLQShuMRkT9{*
zI+F@$TsG_<Pw~=5pe}mNM1wTiYij_k#TR)F=iA)fZB%uBD-XwXZja1@E>QB2<9{WO
z7S7wd3uOZ}#0p0YSn)8<2du@%E3Rn(%aXn4VYhdirsY_(8GvT`T{;uUf`E@->C%rJ
zQSX$LQ*%as%t=c86QvU??>5Wk)W=xk;;Pl<lKk{H7v59_0wSr*W|V>`ljzDy_zpPi
zK-3*lZ6FDEFXD3`-4ufU2AB6{<{!4=kdAFXO0kzHU9}Ae8ZcY7%DSM)m&#ALg>2;f
z*K=(|3pK=Z*dvQk#?@P}T|jMJTD^gr*%h9sBCKshm4hgM%<mn;z}L|qT|?nm)x@;p
zC5Dvkue67KJ%0FmGRKE(=duq=B!x;P3Ac(W(<tnejr)brPnUve?|L_KDF*KCbaoa1
zMBRJzLY(jGKdaeKGQELzB-eENrG(edDl~kNEgEQf<86$J@jrxM6{9FYcS+S*K}yy7
zGz1~{gj|U3kqhj`;J2}wI2haJD0%Q*4(k8-QYG^{wra+$(xb5fjh<^LBuWe`^ggOA
z<u7Z1W;u)hsuz)xN_<t@KVQb2zbo0F;ExEtrGBRD=6qfkvZJo}5S)@S?lP7!R>YB`
zC!gXoug0__>0Di>?m@l!5j)$@&h$lNncSzn`J9=4Nd$!X>4RsD9K4xyzT?#LR}yS=
zjY|kjthxmrMr=>Xo8sk0SeqFsUiN*u#T`3Rt*OELCkat+68OQ})BR|JyIFimi-hLF
zE#SiYed)lehcc<*RsCdopw1a#X|3<{RdI5;@1?V(u2Oa`nr7yjM@4>RIrnS4is0+q
z+LO};Xo>PP6}kKYwcv->s@MtV19dOEvjg+@<`EKq6B2)7c{S-oiO=5)K%v23HBCx(
zG4j^DL-5znkqv?BL~WF{&z?!gTfC-gW&4wj>fyX-=3LnQusOwfM-s}!U_{D7?+hAP
zHUftVYpZwq8`t>Uf@O9T+jffu5@W`2mD=?wQc60|a6VOL6iwa)#N$T6lgl}#u(v2N
z3K`cXehMWDpaUj4-&}E_9z7p@dD5!#_Y77UEQ?9?Rd*^VH{{Ks7X{W^kBwXPVizX1
zdx{W>OVhCK#RG{Jm=Vup^-Yb%;gyh`)sb3lIzlu^*9@n+WUhDmswkKv<N#1OhYE;2
zPd@9f6sGA*JCm;0DbBv7imMqnjCob{)?Ltyyr8jZjA=N7^y6gz9)(O}_M$DcK+@@Q
z#rE#O8-JV?uVruf%n;qtb`x81RElc}G}>sm?K9ly)U)aBMM)%`Bur`cX~wU&VRJ@1
z2G;<42v5+=o~fHCE|B<7djF;{S*29p{r=uoIDIaoV0r~4SszAy1lkAGRl7d<aBK0*
z`kZmt1(uRuC@?`R_I%nb&+lbrew)MH44=iYG^$;2e}vY#<Jf+WxXstd>0d&g3t=<w
ziG=um@`eCJOQPMIr-~v%m1>5zm|tv(%?akcRW#gU-FZl!r#j?xw(I6UT<XA~<pS{q
zL@b7=xkfK}MLkgNCe|9&MpYiBFxYD>VDjaGbsSVEAR>1(fY(X*t|XaS@HIo4o>QH*
zkq_Um>D{W-FU#OZy%4$#Zc2a&h+68Yq>2YMlu;+w+s9?d0pwIe0kgL$@Il{$Jb%P5
zNYxjTzs_s+T?tfp;&m5w%(@Uy#%-^lPxGA+sdz{V|B4d*S#Stz;{M<o={Y;+^51pB
zHTzK6zeajZ&`353Ob!RR6rTEr{_oB3t<q{i>|5`ogCQ)Kco&gR7ZAxu`rp%n67!+a
zRm@8A8lW$3t%$PP2qLL<&vrsf>3V2?&yUI*jBIT8c@UbGPpc=oHFfOgA&oj<t16<d
zx=zs8H^B{|wFRz04>C9n{7N#Pu+2i;`H(9zV-SWO7Pr5nRqluo$<?J_{Oi0Dn&^IU
zqnz*c#tUM8l1gh?irp#^y|dE%e{M8z{;0UrQ}<ZKx{eMn@w}MGJ<TJ1c(qcOxFWi#
zI-(KJ9m)uT@+SmaydnM(met<Mp55qPVLjxz!d;Qk?9T)lS#?>mp4!SC_l#;7mp@$Q
zny^#2O$5;^@<BBj_}-76_X|V{3<8RKAeLPkF?jkh>ljW^8e#*<CaPd7C=C-Z|5ry!
zA!8^Pl8neaoXci7lBXAJt>jr1%7W!GN>c_u7lgP=LU4a6VV6~bgNZO}>?yWx<mr>N
zL!diLJEdQoJ@1FGB;Br}6;-v<D)C9-xQ5GCI!?vfi9I=!4}P{bRl73|0iF%U;NPu%
ztVl!tCcH;V-e}Q?hxyyg^$ve@P}qS+^N_Q=nP}ilLR2iAw%E+y{4)P(f}$mb+M>y*
zhGr-8@~Nx#-ciub2*vGvvpk+#GcG<X*3{7bG5cwq8r5)SgB^TWLU;O+@TXp%+>*2y
z(ADFrNYnl$6#Ax)jDWWOcw2U=@g%Qq$=>p`PlDzot#y0=byA*|J-oeF>v3nEm#-sy
zn$|Rt`tFh>JIqQU?3uS~S`@}qd+ksD^37q7%$aOn#h5VMa!A)u%3m|em5Jt7Up&VX
zC)>C34yBnrz4Og_jdh{~^qq#Es?a&_@>VEQ**+GWXI1~@ecyKx8>T<0z2@Bm_NfO6
z-=yt|`{J^VZU@d*#F<)bKOa~#K@f0g3n4x@i)m~LGUdP59ksnxM%Oa_XZr%(K1YFT
z7{KtooN0oJp}}>NDW(mEXr{f(!(&Muw|eK42!^Q4a~I#$G5c4f-(OHviFo#zCE=lP
z<w@&vzJ-F4V!V+C_qcKhlj$c%x>L`@OA&6;K0E;HxpaO(>P;DjvnnR}p@s!C2`qbw
zu>HHta@looaI|>!T?CzKyLe<<`#kO1kwMF9dha$8bJv{Ni$OJ~;#LFdJV4j&3JmCY
zCnE+@mg6OhhQEhs<=$2sD;*IyTq`YuB(XT}xTBzdsOIeRvUZi-hK-4QDM>Q>aMo9%
z&NM_{nM6UoO7b(p2-8l)l7fD&?_4Q-KRS@5E_1=6(gVz;h6~WKC!KNJdHD$}?LvE5
zT7%KOa&hfPl4Lxe7Ug|hHy5wT*<fEvyr@o2wqr&2>jGRMJrrAp3CeDntRfnuEe83}
zIkVQzDq6GVFvt(k;H;LzPf6$HtDaF{$QU5fnRCA<u>x2ote8=ydyAwbUks3))2eQ$
z0zQO{Ewv^Ow~Z06$P2zR-sUFMX3SH6*rSb<O=hLDi$F(xqS-}u7=Cr;lgkQFa-0b*
zCU$cOHU@gBa&*hEO!t2OedFSi6Wu%9_#u)r(?LMx_hrWZm+6KY#G?89RB#ZpM{yfY
zZtqmOKZwW&2g?d?oZ9{&ubizQzVW_A{z*=iO|6ii{*jVr48G!uvIy`aMfO2@$oV3x
zJz4Zhru_-0OKVI<Rz|^WrI+Y(Uh}qTtJfxixzeLvA59T9%D!NCNDa(bq(=T>3qIhl
zWBwW+cx)<;lm(tOI|`5cAjhFDtQsgS?`M?3Ephj7(HQ#A(v@=Ad!G$H`JS#kdQtuJ
z>b~ND&o-oR8ej{#gKAWbQYCZ7b2q+a^<AH^_Uos<*ry2xu-^A56@+)7OGzjn*s*I)
zME{eT<)BbL{7u$6<uEOaY9~Ts?_DqIczAJ}QLtHZrDlHE=A_#^%ZI9Ox7)5u#2*rS
ztjBd;y+8{gk(xKZf19E0{a({+PW{yhHODKYn2<IY?wZVX>^JsrOVDs?!8FP#5n9)C
ze_EPVUA(WnPs;|Gx2G^sf7(b$wQx?S7g47ej>$zVw!L;5NEW(xsZZ3Aa=p*T*63<u
zAAcH6dWDXbx5d?r%W?qNz1U|v*O9%9EE;zqhvM;QasZtFwxH==iCtM+U{zYqtHBOv
zjD6#+@fhlGzR1+1PVS?1I)>SL{(xK%lgb$m6CpAxn_wh86Ff`yZMLf}SO2C2005Ut
zd*<%J!*gj2USbTRH4!nGB~YTRkNP;bFYp)jnfT@N#6&ZseY4}EGi}cC)p_p%01a&l
zud;p%<>0So*O%PEmyR0U%B*q<>@T9r-r{S0(1u4<imiAq+qnT1q)+qaB$g)!$J70D
zFcPDOS=0=$J8L$NfyDDl1(mv!Sykan`ubFug7LdM`T^RB`d0pv4frZ30P|y?R`haF
z9$Oza#)_*Us$7@*F|XTCK$bqP8zu9u+`LaBu_kE5oX*;jvF~LveymcUc2<=fI_FYl
zXL3UNEvZr2bS;%%tpkdLvb5uFtXV8}(;dp`L2UhfCVrs0TsCb%&FhhiK=fNhTAh-s
z3b3<OJU_3`95m*|;^!)(w?3B}jXx`}I4wVr&TOmksd)qJl7L@U9iiS3U$37evrn%!
z6hE@;`pI!FTt};SP<qwtL!OUpm*dFAjG_N``+MwWzPrE6TjJS#HtyFCPc{E6+fXk3
z`XhG|@hPHlXOruKiK4!aoEMMXpsW>#hgpk0n)?9f-^_!b+f37nJ9$q!{nd?z(X}aV
zzDHp<ZR_DA8ZhVwHW^Kp)kGnCc9?~oALMpkTpUaOv`|xWTdz%!=(g|&%gd8L4Y$~G
zRNY@Plgn2Qi7M`atL|D@wzo8eU(L6G84fF9aAXRJK$%D#mj&IL=6MPu2gwvYf|`fD
zYn(X6iCz7Lr2e+kniZ++?(CmWpunDlUxSJExriSf&yplx&rED7q5Ao<;^(3eZ0M_S
z6ohIS7~(NhMb!=29Wt15)AZO$w@(oDTrfq}1brb38>{aSM|5#cHJMrecIyE;zx8Y&
zv3zDHmk{!lYY5V$)C#MLvP6_5xTk(p6<e0ZtC=rxzOQQc_=9BA!ew1aFhHD<VC-`v
z*)Pv8xZ@tEWnsxIZ|E|YuzFk>`BxYC2t(C%_x9zwU}}*@=Y^|Hi>47VS(owE`KR5f
zN{zFe`YNRPKa8}({l!@U7{OSV=b%T9;=%fTQ71+Qe<Cq_k#Vo)YGy|&8*j|XBvke>
zeVOJrxNT^AVYbWmakx-Ew@unF1%}U(;YR&>a+EO<qg{s9_sdTPUMcRalKxDM@=79z
zva7vcSEuGqz%QkcMT@ZqtrX2{>AT?7hUvx#;BBrW?`?FsocX7d9vv;!^RkrUL=_A#
zi!D&zCh8qr^x|@E-3H;So}dY42?S@Cgn5%Myq;N_-4Sl)!@nG!o*d;+Ct>71cF6K)
zDmhsg_;{yYvr9*8(>zvtzk`6YDil0u@I?5zyl$N+W<;))@ix8?=g4T*pNjp|NqaeF
zDyc<{x1y$mD?UuXs&@kV>w`+YstxQYU$~ADpm;zvG3_6cXd-2%Kw#*2IchjpyX)%V
zB-C@?oNo{yZ`FXK7tu_=B34r~Fz{DebI(7?_Yv(gcR^$(w)li0J^5G$O|7xJWy;If
zRg{vS+w0i41nama?j&$gBsp%S7$F6?bY?k+LD_Ac^L~&^tg<2{v$!Q3(E;ZVEI+zs
zE%5l%)x=F|R8p_cVWxD|2&{=kaej}-;!ll^38MI@E1$Eo#l~uSpy*RQVsxO~G!8;V
z>#2DW`JZP`flwC+6hOgOb+rsS#s#YNCh7(d*Q0Qpg~<cQU$(Sg1azt1aqdK$I&R}O
zr{nVJ^YbXkdwgq;tHe8#8881#H>WQwHdD9?f8QrV#vRA(u;M$LEA?ZPp0o97gy5!e
z^hE|+DYmea9bp8qeA#3u#p}^qy_iTJDptyX_#uq`z&v!u^}t0zosTyz{SG>Z<gDSi
z<V%^S>de7Y;dBO^N<<!z0DIyK`XCDG1DSfplj;eYN3J)$VapMcwH^QvX`87RBr6D_
zvzQ*Cu<kY<!zwV@3D<Wl?nH;|PxFs)mB+O_J{CH85i130<4!xhaL|qqc2{)b3jTKI
zI0xf{?LxQAymDj^;Yy12M4>vwHm3mG@qk7uD=q(tOr10?XGIwIHBPdknqJpD$CC2E
zl!k7^7DuPF&hG*{%5PKk@Ay8VQ2J-6lPeHc*H^w0DH!R0#XKIZQFiqAw?l?PHa3m$
zHwka<)*MwLz#Hu0(!txHVXlck02m>-+L!*unJ;=khBB72_wN4gBxd_zOPyQ3hdXgW
z*}0n?hYwi)j*;HjT?0F<J!^B*fq}kPF#@Rp0WS;#L>`_d{OP^eQ6|;+yt;=nZ>7`g
zf+=^5BwOvk+3V0Q)3cfs221pQg(ma(p9SQZ$T?Sof6X0>;EZ*6=?>^gzM7q0jZ|!Y
z!sHExTf2IE6a!<=Ibo)vKAT~#{xg5C7s-D|Q9*HDR&&J%x272F?P|T)k*~)c`<?E|
zfZ*ZYA?j*|V$LV&j4z|MDWJ=@4Nil?;qCET$`vahxe;|if>Oxri*qLQT;@}d95ccd
zxxD;%p2sU3s_i4tGF~cSaXk*(WV)MP+9-ai1w!ebOygD*I{0HIk_TrRf0XuvKlg0K
zcPZZnNuWnPlRFGgC4)6L8XMrCXbXuUlC&Ux=rRV5Dm<U1;l*QPQOTC&9XXe$ysVVI
zAUK4GnM~RN=3ku%IM*mu+tz@|VNl!}w~`*QXdi%QlWCxXl$aNV3Dd8axJgrkVq_m-
z5ldh-?|MKA8@|gD_Eaf}Q$*oG=hsDc)*;>ncciJBi=p{+4RWy*e+GpP&n{Swa$*NM
zhAZ{}M3|0BoD8B8IuVgu64?Om!P>q%rbLC-YEFG)=2Oqh%>bk0r4t(IP)Si5d<OU(
zzDSqvPMH9HK|djC6^ak4wBz<a!g>p_7Iq?rL_MK$v?pX(hgg$Hy~y77>=iYlR#Ta&
z$2D`f^`|frvJX=XVLsW_^LY4VUOjw7gpYfwHKFD%GfAq>eyO}uceE?-+5`I5>ArEu
z8eJjNpfHHVV?W1AD%_m8f7}9alphYb9}l??f@oTREJ3M^FvF**=?SfM8&-cSJ9Nbp
zc4}D2_91y65kk6Vc*;J)NJq!nQ&H$5S^mZaA51Kg{WhNz$>oPWkJjqEcC+!nZ09nw
zM<tqMj?!F}v=94>wKTdFiu@WyT}!N<z%kt8L8AEuY`tP-yPQWCu~{+-(=)+R(Ye^n
zdG*aGHrzeGKj(c;r&>rn19!RXm$b;+ornOyd%`Lzux_0R8p*?4$cd$KTc+aK^N=K5
z!Aa<#K>ACx5T&>(({X`=C*<|4>D%&_Pq!q-%dX^NHB|10Id$526I{cnMD9u1W{&)d
zBwun(6%A?C;<&6q4I{GQoTf^wU?pmg6q-FYrhz<9QpNVEaQmVSjWpAKiBLuhCcYYe
z<ZFgWJeevAqr3)Mj1@iezFRH>&7r4eIa{eXswgUh4k9nKWa06O-}BYZ0$eq+#|1IV
z>v3TS<dj5X<x<{+KC+o(YXsM=c2I(&>c;%X<+~gD5fx;OOf>F!G(=JBd5xNm$ho2z
z-r5o|VHpg#R;=yCGl3F)Ato-vzl>$>qF61jDuZQXS~sDzbqAlY6BC<hTG>ap-za)w
zxR{%+$fzM>cn=ro=4sA<$YIhWHZgSt@2(T@iow;RGJ{`AZ+QBCX^?ro2Cu)4xs1MS
zem{of^uQIHKfZm-x5ekYPo>kA5?pyEA^Cw*qNNQdf)q0!P3%sR%yU5)*KXo<0+;5Y
zLgZ8#h_}go7P#vfd2`l;m4w{|#!f;OHb1<A%m@RToTGy1!*%6pBBQ>D%2}eRvXB(0
zZZ18%iu~3yVP=0w=^)ND^jLbNdor3`V94qZE!K|Y_7FwbO7f-_JwnT#^}41ks1T+3
zI^4b7vNRutWk@-)B&iHE5eeQj(iEot@_S4Hgy_g(JLkw8=dyp}zU%v#+rxhf?%V5;
z3G<pG=>}#~8jeFC;`tfR7R9lpsy|{c$7*%$7-yAg*q$!5czK{Y46&WxGm(aN#1U8a
z`L5|6soKJkt@ivlSY>%c0`0a`siW<s5oUphf!`7Q-$ceASr`gP4<U|?+t7}#ncl^q
z)c6YzRQW=DIxuyj8@q>w7xFW|PbeC6$g^^z9Xp0tTcEa@D<%V8)!h;|E#L|C@OAi*
zw}6+yz3$&3NqV%CXrm(@5)c21gCW;PnYS=p(nySz;Zze~oO`mg&YU{yhc7P>cvz?K
zIU~Qktj_peBwtu1l4ZwJ#?!J5XlL7Ay}XnBqbaN?oHfoyhhn&#r}S{^OBInIm->pN
zQFasc-jZR*EPgKcQV+`j`C*cOELFb&|8ElU6}<*Nh#EpTkyVGCK77qpfL#+di({}7
zb4_~Vj6s%T`(|*$sjLvIYSpj$=kegWcGK{`LD=mC5xMfsbnA4Gw@WhDHS<8-VXKN`
zrb;a+)(Btt(vf^To8^gj)Mm^k<s@sKl1>u5@?n2mk3_Woc+YLK(~5ai1WC2DhQWVg
z@Ta1c)XEoaP{;=Z<%^8z;Bbe`BAvvxgt)WSvnJf0xPO#gj!om^1W}7!4f1u6nwMBE
zG#&6ECU<2Ux>kKUywTVo7SOx2Zl>gOZ*<WW4$8XkP~ggA8vJ-geR7{$L$ca=q3La$
zkvNzcw-ZfuPZt~|(0{Iq$80{mO7jB}Vp+wJ?s(YXsP=+F+_^F{*{uF(H_n}%3;9!Z
zpoULQR0k)Una<PeBFz=+$#x_Meh<qJ04qk`?d|GX(zH*r`qYty##(@-rpX+F;lERh
zq!IgIF3D;!g3ZZsKp4q}*0@o@8IiW@oXBl25JsU*_W!yl;P_t`1&nO}cSG=>3(LyL
z`u{%v|49K8D<>z*|7;3i=*28;Tuhz*(u>&`x|oWX8rz$g!tn9IIJ-ER8rs5mYyet~
zUCeZwNOMYzHa(0s0c&hjCn%d01xA}K$x*hOEz?oY%bz<vzTR6m`rps-%p>d8)x4`Z
z)-~OuoJT`ZR8K-=ZenEzZY?uUQ*bgoK&+<lqQchJ!YBo~G~Erkv4nfInRsbAIz#hI
zb66JU)K+IU6QFYptiTaj*q|C3KvP>=TXjc3$lTtD;sx!Y%mEkl5UjMc+y0vWp|F^q
zU*-Ryu;c~pv;haV(Nz`Ax!ti9)sOlSzn8I8Wo1C-u*v~ga~r6CiwgdclF|bgBqgYW
zN@Q$gabatLqUy!~0!M%}3+gMXlPOvIre=^0z6oFh>T|21@A4ONFn^4oIf3DT<zC80
z`g%8Iwqs?le;wz6uVnFT1LWlVcYy`S1KNuVs~>mqIW~a<T1Hmr$bOODtl1ks(-XRj
zw|AkkdwY5fqTJ+W<VNBIyv_S|0Z~N<zx|8r19R{BTXKE9b^*ETyAmke#_9f2AO8`u
z|EaL+NEVl14s5ROKji6&lzm`lCib>x_w29w4Bp~<y(~I80WBxL!vP3;7FKp=7Dl!g
zM^6|h^e_C|@3maOztYcPFE1<jzc($P>=*t{KwR9JTG@<!OpRaDmRDcwt(<e{Y4`bx
zi4ClcpbX64;oC#QKm5KmWpTf35f}fdus_lRb3-#L8=wY8RuGdE_NQO$qyZPda;9Ow
z?nJ-*32(oVeZTTY-@oKH{;OyHoJ0S^!iJ-}*SDmxH!*&@|Hs1KVg}Is$HI=h_(!uf
z0c&}G<sbZdmxP`BCw%b_NO0o(*oG!lzw2{mU}pRbLb?2b?8wfd0LhJ|1u_d;@&A0P
z)tz1$TAIhUu>Me=_1A_jFgG*$u;){SEvQeu=8rZ1MNF@#9``@}Q7`h(WVVuks;C^F
z`?+<$=yi<pZ+2IEuKSg~_D>-)H-F&A_$`2f-E0T#I?Lqb0D7X)0r<z`!ZHVB$1b1#
zCfw`)J^9s|#Ny%x*#_qFpENW2X?^>>|5}xL?d^|P$KK5RVy2~<32PzqhrIH~{&B?V
z<ly>Qe%@vNZ9V(d`Q^SSE4wKhsjHH98|ey6tH83HY1ZHrMk(U*<(=9M3%ZiNP=$T`
zuDtH>RTg${qlVkD4#LDP^Wf9wMW4OmM<}sG5cguJrtD0Xt$n!jm8kH3RAi&v;Iu>p
z#e9}RT9{!a>aM{sY3*<3%veHnCr?Rvj@l)JF5A$=E3D1`_bs9O_&qm{e~s*qUA{DI
z;PRI}pkM_(Cs&yyhlf`QAqD{v-fdUt34d?iY_u=o5!>inhKpFv<%PlFOWFPl`u<`9
z|I-F_bgfWvCF{f!O;y~g=i2tgmK@9t*IZh(RE?w~SXJ|{@J>s5k^@<~)s6iu8RlhY
zQi$To!=A%~xvgnV<w?(R;X8)@YM$+2o1<g{Uv!1n&XnbdWD%Oh(7Vhv?)3IAdZWLt
z2967p{22(lTYRDlau(VTZ>e-3l60{LWnXvZ3=&&uW$YFDdS>k_&PJe3E`_g^OsO3<
z1OT=cOQ>SjN2L*Bh*US%jx4VvM~B}HF>pS%jpH14&`F#1do_DsT6L7H=&9b|XHKu6
zC|Hg?pER~^J=a<nBeopAi2dEPKc}Y20E`Qdvc#D1Jgc<&g0vyvazywsW@~os<8p3<
z%C>@;S%BGv;R9`dV;uSsrG$r&17`>+eD&WPu_V+HL)&it*+8d8k6$!EU8E+1pBjsL
z>D?8NqJo){!dIO2Mq`CG3}gApXyzyWu9Bo)|EkqG*6Tm~${2+fu?(&m5<)UUwAi~G
z0;PI9Ji_Iy_tgskTC%anPjFRxy@+U5DVZ9F$Er|N0{yg<atm0lN#&TyW<&$h<BjFU
zg^^nSR|MZBS=3Tj4bkyHu9Tz<t~Y$!RLwo9T(YTI?zgTo;2(d6W$Yt>GEJZ1+4<Cs
z?+*E!HjP6cWX*kob^^mcPG7kbj8ei9fj)II_q}$0gctIKJ4TWl47w`Z)|4<K=VP$I
z`od$XzKz#J8kqsCxgg?<w9_X;l`i8*n_w_xaXs;{b=lXNKd$1|DnGp;s+EXfM!?Zy
z#nvR;u4k9XJdv3VtgbIRQs26WJY$SrHb34PHx?)$FK7E}Z2s{!^tAZL|H^BT!SoYS
z@<#i=TRg`y&ozdi0+D}!Y3qCxW%7`ori^NLT<}aOdj+}2%b2~fczZVTgkyaIA8JB_
zZo1_Sg0I{TPqsF{v@9rSac{o96QynF;g{z$1r>?y0z4_uUJT8<qG9&@8B5y02XH;u
zyYkI5NR+wnEMyXA4j$Cq1s*^vkUqCHZ`A@)gwB5z@}>+$>%%u0c*5f%h2YDXzZ<1#
z!4>kOw`e{{n`fxQ)!>9O950&NFcPx5gA>#r6~&v?z6mb6E0kN%!!eg6;Y^9|HMCWd
zLLnJj9mq2WJ+k&neGudpN51EVM2J7mr1kU9iiD)dw0FW>*%-W#idQ&UCqnWs*}Pzc
zD`^#BOTuf2p!xTS%jj2p!4k=cC@^Q0e|1h=$6U2loFYHc{xF9t$Ubn8`|!KY(hj4a
zPStVBng~u7<5dL`^QnNzm0ZP8i_eYLq4;%#Lk#q3+rEueiVLnHR`Hsc+^EQp53cVT
zOPUzH?(b@VY%Z^S*Ir*251^&Sif>d+0vw^WEvF!O$@t({I(k*&^QTP5)-Gv7s!H+m
z31pu+3g9TDNB15%rCT6XDd@{mVH^k38%cPlXQ37(A&@DCnsfgPw^9s27-#2g7Vn@h
z;prf;i+aa$+>!bFTboc67@8XpY>q3TPH~$%LYTRLXKXzTQ<hpBCBa*PZ%8`7!ASn#
zYnlNXnf$3&-+cWtkrvN9j~VM9K=H5C>DjXeIM1Y=FA?Q9uD(p&H=sx39ifQ26R#!8
zo^jlR0<qPiMt$zu4Q{QhpeFl7qf!DdwSu+?1m}u?@wr81WC*F;y79()eSG@uYe%%?
z0(il|)QZ9kbyvc(@XTG0aO0{zE;mN&D|aY0c{lmx5$;X#!}r5b!Z+j}tbx!#CHZv@
zvn<g=xL3`LGgcgtjs-;e_CHg!SzEmSYo44JSn1=gCO2l`^@@NXWSm{n$4|)YXgLqq
zUq|!#c$H{%$8O$kU@>{2Sr|f}si;)m!IuTAsv%2Jl3Wl{jif^~sJ9}>#H1xyDM0iO
z>sf@@Yig|9L`b)@_MXZh#LDTEn6snx1NM2VZu<VYQZLWmKv*YG^A*=Jo7$QhY)_zy
zWr()-2h{dN;(P-(#tMsUE=>hctHf)fMBZ)ktyD@-cGJWvXp)HgQAiVwG8csbwM}kv
zQZY7pYlGATd8YAc=SMZ1#=<^&d=b(?1Yr$5Tx3f#6!++fYpee?x9RQK<Ap>yD5}xF
z_^@5-K4z+P(&c+OvALu>>#8J|vdcFQNfsbIP!tw@iw>CH5-RMsBBd5E*nZV;%ROrH
z|2sj5q(ceX%cLR=fE<ohyr{DM^-XgxNAVQ+dbV~+p0blVwKI|M1{EUthu<0+1CBiH
zQjbTlmSyC}SEi*m8M;;}wCyoK&{JdbdlMkC&0xEJrQVi)Dl#dk=W)3V<cLbNLp9=u
zhzHiTh~EY+pu^?}2l49&qYeS)N-{|*S~!+%!Ry!Nm;Cy8ffc!wJo>k&okOjSvznPC
zFkfTo;k#t|ho^}vS?bl0k`E{}+jlB5wcG-|EwT0iO*$%)Zq1Fs24*W<DR}mM#B_~r
zhv+v@+)icp-B&rN-sAW56n-TJE0fiJ*N6qJ#uFYoOFQ00!Gx3IU)|PEaP>u=#tSi;
zKcZx6Vk%P*D8Jy9*%BWc-qZK|XuDqP8se-o-9jubGsdL+vPu=!<R)#4C0{EfJ=gUh
zpw!q?d1YpaJ+&~3k#5mw@p3_im!jzsmPr+AO=q46GwEDHt(;s;gGAp*FkwGpCwdyz
z`_Rknh~2)@p;|bqzRekYIfN+F!9T^+1L^QhtNyLRvH=N-Bje#@Hy3QRY&LcR%w?J;
zN1i&1JLPMX1*Q!H84`usFyW5AG73BtE8Xr!(04@*qshQv=b=74+>~yJ=9=R*j40=N
z%l;?1)CIXUYutpkfwsQ<!BFIA4Uk?lJyLo}CUXBIc;_U#EF^m?MXqtt=(Eyuh3}Ce
zLyKZ0x7AWy`SPUTE+To4&VInZx&yj8uM_3FixQBta~0`%)*{u*z|LyNgHCR)(Estz
z*CoO-BX_1PBr|;%oR(ZqhsPraRpT<q2s5pJB5<V?BPZTSL86To9^%vD+4*fFJW&@K
zk3z$@F}n5UiD480?Vg*%(BMu|b=iM}aM*kBc02pUFG!3IXM*|oes62+5awtK+#Z`D
zLPH}k-=j`a*b2#eBh2@;&CR6#NEdjYn$YJa8oru%jPmCQ6<H!xGe_;)l&6MK9=yQm
zDN$9!%B4*qme5VwQ&h7IHV(vEplmV<fvz{ySG2!<P6jJ|S(+llN~Jr?buIXrZ)i*1
z7$~*y?1+WB?Gic$XlfUYTTc7A<XS?S!Z^fmcaBC7>*fraYQ+2tm1E*XITj>-Un^Jr
z#cxq5cCnnXg>-j)0qF6mGNbsmU?D4sxdb`8qUL^egXZ|{@#2wb3K<RaDo&9gwR1(l
zI{Nc{q`ZQ{ua!Zhv4^aoc5-a|S6SAIeBjdDUN+7}K^k>StWzi20N>Ei8!UDKJRJ4(
zk{aT}X?%yf4F6J71`qDlwe;6`MpVt`IeLOb<^I*3^SPgC;>#F`rtG)^sHf=Y4zC#g
z#kYV_sttIdHtUwd;E*$&7pD;mEDyPe$-sR7d<u4h>P0$vSM=ux?(mO<y~LlQWPg4f
z;QVvWR*vsRKmT0<&}1q5R4W#y>q=C8@Qb=SdDyUN44KU$eM7>ED7Z8nq-SP-@?P4%
zI}zQaeR;PJH<Cjn?8)cxhUmG$&>V`KnGi47tON?=ba3F@tm^EZl2*Jm9rZcOE}b8^
z%$UL9M!@vnL1hJL(5iyO`5U3@-DQEeJdQewg958!MF3dfV7d!i%Ib!gkCV}b*_gX{
zObH`2Qo6@Up7Jj|7XleQ0wnrGJuB$Q&&p8y&*FG#MP%CLbxs)e1=<FeiMd(6scmdA
zg!V};SNQKG$EQ>G*4xkM0$|b1N+IhkN0!+dk$%`p^hoPdQpqZpWn*ZmhGI*oAv|U<
z&r+(cQP{Z2yS4z(w#Rp4JP_nvYOJ@snicXzPX>$mF|ebzw*e-uX}_<RY37p1Eeshb
zYlfmnLVC{wCFt;kIECtl+--P%C$B!nj9?RjOM%ceP&iL}{J5z9eh0FWvGx1oi&SqV
zu6b9#o(=Vk<2Q}?5=}+x?cuwbIx^W52$M7G<+m`eZqV%<nTfv_)KvCUUYoArC4>uv
z&KN_CuXvbu4w!4SYeWT6VnmeDK4Q%fU)HFbFinZ?tnRdVu{N_#T(766hbD`5`%VZ3
z4vq71%Z>U`8>I%yGT+UbgIXQCc(Q#fqTX3w#4-*OYw!5YRfPY{A!@1#!Md2wf|yI2
z;KWIjWtaOkR|i{gO*)Y&s4r*t>g}}g?{X4nls?w>LP}4p_jw2fsH#r(&A!Ur^|w8A
zqW^7BBd<#_`4duI8mGILuSm@KphaqnC_LOFrmAiDZ;%4&&)TSk7`C!GgEnM~xL{zH
zxK8eIE}My}7alVFF}><&nDFGaG|M<_drv@?CB~!bgU>k1;uAka+?sn_#qc_RA1A8H
z<+w#bbb6W9jwr!HVTBe*Y~rN^(8)wC${`RlbOLj8-Z*z^rv5Yz_VQ}je-Whj=@Lkq
z!BY!*V}s%Fe3t?7VfDWTW{>6kv>HUl*1qP>Cr|*s_xc5ilL=}QS`2vbgvqPCSI8G5
ze_wOR`D46pxXH)JtpPQ%nWL;#*_ajS0?|H}YM3S{Un1d^?fJztmZ<h@EJ~bR-JW!6
zfm=AWnGIv*9J_gy1EA>F3K-4z?>8kheob9P-T+Iy8Vvt#Zeop$A*fM~vVd8b_zcMK
zaGY?+Y$&DqeCyC%80h`>Y54p0yKEal-%sdvsQ>!(UpVYtA=JTda=Y*@MfynaB{9yP
z;nd2;unG*JNW7PYGNb&BrW~E}etmXH!d?*_9bE~8alnB+(Yaa-`95=3<o2=h>4_{n
zQ4p95mt)^2@n|4v*?e$ASO~uH=x&BGlpU95oYvwjfsyUj<q~*;o5=wp)jXr$ke};8
z>!^JE7qf!Q65c2+wbnS3RMzGqm=Soyq});&r`?BIS4~_A0B$A!IYgVknjGw?+D0>%
zjJ4KUsCPmlEMEOyX>D0{tWap@$0m9BMOYB5Te7?dV6g_D{i3Z56cWBc9I58F!?8v-
zSjwjTktE`uA8(L<L^@cR7^3OUnb!JS8PL6ZR4C)#3%ft+2c*xud!__!(3Wo_pSW23
z;!APLREfTh@~0cSk3=ClI{Hx($PGfG+$ft;mK|8<o4r_R(3y687@_P=n~y621OMXZ
z*!uZPpw7kD{u$h^`*Euv`0r$oq32cx7fA=$4}Jn%OIHeWZN07O-Bl3x-xhjX_f1X<
zYq}ap@^*4k$;hWQt<!p7ju{3YAWEMzI=Z(xE_sH>c*RY;(!2zdcs<g0bRhP&Z{~)d
zrE+19t;;_7F<&mSheGcWeLA^}s{WorD{dXuk>!YXTy_iWhG39xa$t{5c7<bzu%UKb
z&`q#cpKi6|x~y{ryI&-VAz4OgJd4nZM6J)eApy(laEV%3ZO~&i2UeIWfipe0EU^q+
za|M4JDO!pI>E@5}+EcT9JwPCiRxed6qHg&7>y0w7<j^}9xq5W9*YXees+Tr>00}tC
z@6Du+%<IYB`krz0!ZN^~u+;II`5Rl22A}sh<X<ZN6-rcK1MTEibSeaY^3o<fue*v+
zSEimHcQ^T;TD`=L@40$g8g@lQ<9`>Y??QF3TIZ=~JL#okb#N(-W{*fN7fG?^<6l<=
zYn(CYQ(7OCNWyj>#aE)>UVa#SU9Nt`AUSUedy@dht0V(22ut9weU{cZ8S4Z`{m|Y1
zG=ZywM4-f0WAI%hq3bJdpk-ii@ZKZ2BK;p%xu%)}w~ZhHc{y|?6wA?yC&kkvdbUUN
z7R97IkS~CO1u2#u<=<qp*1b6Jg~jYI{q(8@`&i`clHMqZjE}__7Zo}u7FLVNqA()s
zpv(}jrosR8A+|4e1d&K0F?R-9NaL+k>Y@LZXmUm&8uWTifpqPyogg!u3MpingzAzd
zQQI5_7(`g#Ty(YNv}#+H7t=frZKL4o^uEqSd{~TNe4?}Hr0`7B0UerWH0|v{GDhG?
zedl4imH4B-w%Mq!96$U|OlG>ej@DfRDn;e<(h1BoeBs#V2?@!V=sZWL?I10lAW_{e
zA{m^AHq&2A_*|dave6LU3S`0RA)fK%-ohzKMf+^y_X}#D0V&?}oE=^lGtNyc@8fUG
zK<5(S<rv_iKpSr%1H#F}6s!1GDD_`cs@3=P;0ns7swk*K_|~-%gp-uhb5PrMi?0BU
zCZ`PRo`k7yt5GXn8jT-}tB_g*0HG<jOhN)Ia}M}8IOcvxYwv6Z>VCs09Ja?9+b)5!
zpxnNykyL1$GyRy{%8)vbODzE3>%eXP7vCVL61@PmuK42!GP!X1=*!N)kW9dsCPquS
zJTU}P)0UH^sewgn_aRq|;m}a}a=xA&uTW8nfv%G2Lux?w`!u9hQua-Kww{Eo==zI>
zb`zxqqftS#JYC7I(2j8IkBH_R$GFYNTTQ{e_4yKk5epMdE{9R=QuS(VU;%130!hoJ
zrlxfhuT|NzoIWP>9X`ZJ**v|`hxfvNG)R$};Mj1(=^=~BxU%5uAN}Y`Oqo+mdLD-z
z%J)kP$YxlxRZ5)D^zB2>T`7)2fjAbTANK=#x?{*Wq73W;J+H?5GBA=H9QePIvlTJY
zx5=K&dBogy9C<K`Yo>6pStQJl#SpDQ!>w$_*_d+N4&^;g6lT!tG=xUy@&rO1t(e=y
zl>#5slynbq$41a{^2x&<wgp~B+E`7|YL#~odaE1}q{l`sq4NY;{Np0%m(65<G&8>=
zD<cf=awtP;Fw)tUF=u7U%mmXxF;Su4UmbjC!x|z(vs5l@J?qcSqW)Pi(la^fcN$z4
zWGFwVgZYBiM)qj-G|92Ra*a7XDjiW`=$|xmGF25(J;TOg1KeHTh$>cnoEB*iDa^9^
z6?6}YWj2S0Fw{o!zQ`PBWg>&VXsU^(X%1JQ7pU)(F$Lwt&>)0BqzT>1<2#CkW9uj5
z(vtrK*EjJnU(Nii(RmgZqZO5amd^pWgY`kTbD;<$b!JsJ2ad;w8+L=_gMMW!m&-do
zH;=xYR<=~yStvrnabr@K+5`!-ZDkj!urz(@cvnhOB62NL4b^-jv13i40Yq1R_kxQG
zs3}*Td3&dM2qZ(Ryv2S-x)Uv#%f#}fyW4Yal@9CAqTeus)lK<6_fWUzuRAu-FkvBa
z;IrXxj6sE07`C=R8-Jc?ukJJmSosObrhDxYoXUf-vqxkpE)4ZaxUoheNhY49_I^S!
zmmsWEBHgr>IIC~dorPjgu0#}u90t7vV`<O&`VONSoOYhjy*mnz-5HoxelX4k4Qs5+
ztiC%Db)rH7eBt%ob;QUEO7O7myo9M4hLfYjGC#Gg*tAh9&YNAoBvGR%296o}N6Iga
z$7T*hSz>N#$_ds6ZF^WU7g5zi4-8KYE#N|t%NULm?ct6hDN}rbhphBB8k-lPye|h~
zMu<<@krwO`=QgJUr*pQuYWO#UCH*r@yp3{yltiP7gD^7YsJHTiomC7JkXe6767r4;
z122%>QjQ=(75}yE@n&rpcXd+~MR3@-gC`#u4lYkvFHa(3tB&DdhJULvh0x%=T$-YJ
z%0U}y>rc%z+=*@qg&^V|*RlHR5DpFW`mni#deo#xG&tWx5g={hDx=eW`Mh3OLk%R{
zQc+88q_SNI!=v|({A||N->tj?pt%~-_OU*b@;%faooJV0Lwwf>a&DPG`h`ttbv`h6
zFnZlL(y|I*cpzT;BgJ~%>n63A%AUmSob#(g$#KbQAp{SW3|rN$Bhw~37Y~HyJKOB7
z#&-xx0T)alMB^C{c9l&==ffwdnSp&#l`OBSq?ig&3Wg<ruFvF>xVFOxutejM3hD4E
zIDGG9J+q_UL5jZin+*``b%5>UwB?M_#wYDOI?aY;UK+l1i0~hUKoJ)d<NDpUEPT%;
z$Yp<a^1Tai7vqOLt^89a+J*h$XbClxrk#h*2Gm7N(`niNhKM};ExO17jgbp;SJSp$
zR6I4c8D}BO@q*nz;Tu3$@|s6+uVh5VD5S{Z!KCuo8g`z#9u!Y*Uf${IM^r^NwZmNl
zD{n>D=S!t?^|oC>YxQs=+xI3a00o9%$z<Wz*yXWr@%ufgf>^Y$jf<yh4iDCEe9%A>
z-73#PKt$ocCt33InGET8V-#a+>U;&uB}qm8Go(mgv{QItC_S1?iyGY$q|kI&or+s9
z7dT8+DIJ$azwRI`!g58=Rr=ToRzC%@b7Rpyi2`35YhFew-;;V&B1$F;7@j?!(1Cl;
z5}nj&jsZ?w$a~2|V^Tj7=G6$BAe#{U&P-NKN8_lSlL>OA_Gi;E{=;>-w2I|J*GuCz
z7BAZ>Vz?5x@Y4Rg9MqHD$q$8{4z1eMN8KLK*UQgYyAEnsZqTQMsjvshuaw$Ql)<Mw
zO_H8MPs0h6KQ*yuaa-+6%EwE2nZcpnCO5qqaP7P@X$H!Sv2g~v))xiTK!!J%P^O$H
zkMg=ct2lA!Ptpl*fXbs<aRgJ@UK<HC%3vUV43{mtw5ouaPmBmIMYm}nb+qU&IOXcW
zJOf2ATJ0Mt4&Wuw$PXyIgRDMKi_Bts>aR(N32UDpzqybHu?c=5&}vM$ZcKpTb@=@$
zZRbpg7OGG`i1PAzHvw|3I`@rT2u~Hko=x$7Zmz8*=7~UFDu0gZg)#OQe`KCC6vT!V
zPT%cu72{_uTt{prcxexvGgTiye}F5$Vbd2N=-OO9Tb_#=d<&{q$~~*ObpqJJ?FLH>
zF#?yDoK$S=)dLf`;>1S&nN~Uc%a5v<X(Hk^Ts`7(>XV?TNyq0x7)-}3##up+^!EOa
zz@*+nF`Zm?;Cagno^9C^iT%E3foelU$7dcS#-8qbt1P9`=4ypDfQLdB-`GgghdB2n
z;u7z-*n*TR?~YGgUn$gN>Mtc}K0eV#y$ee`O7Q^4*)#SxlTu;T*?LEKFY1`ZT#W7v
z@5wR5XkzrjS3H3gF_sg~qnO&H#%#WL!$k(JMLlmlegHI6ZUbHrk258mrGz1&BD@TR
zM(vbuSUS-J_C~W3Br`lHG+!!E1$tg-n?H?$>n*Zrij}G+Nugo63qluBD*0uE<Lm9%
z^DJUfAa-zq5=#p}NeE*CNBKe$=aC)pMyH{R^UJdH#VW%x@RlsocH{7b+H!ivBJ)aK
zX|Bu?r38>$<nhfXy|EWK{D)32WvI1H*`*3*sqRI7i&qYC+rTG;&jY4n&1SV5SfdO!
z6Kg~7Y7t7ocYlc%NcsqSVV3ScH`m0gsL!X^9+R<a{)_yP@5bGEFHgc)7=Caf-8;<l
z{B1qNEXUiZp<BVA<vmToe#xK{5=(~u*zm_YXxLeuJ=rBn75pwKj+<6vwi)-Q4(@on
zb@gMwr|9Gk(ApIbk0Q@k7=OEPM_eGjL8AeksB!C?G8P+B?m1guTw>kU_cB>g`6$v5
zoj%0fcxxch8sW1JS<T}#5>07$g-70Rr+&kteB^lwMWVbB_3nMTSP>3GW`tIu>H?Bu
zMDWVn14&!GE4CG3cfRB9s>a65pkBq6F~Ym4qv_)X{_mdy*)g>zS9Yxwuk|c5HVCn5
zP4PA<=`rt8)gaU8Bq#1mh$oxI+}1ref>@ax{5K1c(VslY6AA_rHKM^%7PZI#j{^b<
z@OjvfD{3e-&`De&j)5)$(-m7Id{tfs1(79#ON-ltg;mot=cI>99IO?+8cJOzFxvqc
z9?j}GoPH_lELP_F7K^`9FB#W5JSTSQ;w#nS&ZeeK&>XLS%%xVH7^ic@&$;5MXx$Cm
z<9dgM+W_RzEChd9D%c96$R;a#aNeC*eq-)u!kV;+M;R)Q$sXB{;JZ;+e8fx>O{saX
z5O&=~^3Ei*q+y4wZ9R3B#&2kK#0O?3Md3{JF~wVQMW9J@h)kMnx9)R0ga5dSu`Sy|
zi=PwzCSSs!J5dW|6s$bRl<K943`@XOGfOSvq?WK3K#PZBS<IN+Rm3k-d0Des6vTlk
z@SGuD%07TFlRKACc!-!<g3lm^UMuike&X}5`@6EVTUc5j5ks~e<OJIZun$~=wI=&{
z4yC{2TYI_u*T^l)cVA{dNGs9ch>*>*K4Y^JvQ+=;OO9SrR)jFS3XI_`4^~TvG`#`}
z<9%-ws*{X*Bm?#56@1AIXrrd70+pJ$vNh{%_ZGX#<3*o(&xT5$@mLQfP8~Aco<rQ&
zinn$1%b9g5&NNX^6&bLzJLd0-6X`F=D2a4(m1JVzAq(##kE?b^vbgE$YwM7bV-~~j
zZr@WNk*JQoG{k>hl>%4A8(oNmW}(3UJN+vJicFh+mfo(S4!{7ZI0jAU{d@d<>Sh@4
z5)j8w$+D)lUWfH;0=l89O15-D0maUCS8z<ZSs|1g=J<sP5cF@qX|pIH`L~0<;G&C<
zmB&3zm(4Zx--~GE4~&UY?r!kvEd0~vgwRI);R+#oBg&9aKY~LM1L`~^R%tEi@g@5{
zJ$Bd-ju>(z8=cj9<{yhG90tHk0*m3`Ic)MNb(bCqf}g!_`E+W04?UB9XdJ{k7ODdq
z#9m_yZ<n%l*rS@tB^w2)NUxD$!)&iNpep$00?L_fz96d9Pu`%SZb(L%%2j{PR-|vR
zoE0omo3b;KIYhd2v>=n_eLI05!^zzt`p#}DaD3k`7n!`rR!~}I2JKxZ8*J!bQm|W%
zTgnS==)r6po`mhA!rK5juhw{p2Rx0#+JngqqRms+8%W@sh8+v<G4(D&`#Ae*O0c7v
zL=jPx<by0Yzlz0YLXewe%wFD)k4pPw9$BMT|1=IKO5-s>$w)k;U=k4??Vl}sd-SyK
zeIlC<J5YOaSGlkC2<rN8V~FhR9nQG|1vnDdk)-*qFxtC9Z{g+j#<RVzt&eBDtZ;m@
zWU8;W+;|KXM#<(6rThL4DZfg|g~_>3wtfRRkSI+Hp7OkDLCte_Ih@X*rD5a8dfoZv
zudRbmwGz4S^b9z>$2Rpy=q#k-?l%?Oj4p#x_3V*&d3{MuNXSr1<&mJV>+zSV`oYX+
zV>*k-l9mvMN_GHy@wv1`54ul#w_@<?KQuQAC5I055+A?<E+>^i%7>cuM<0?7XkAjA
zU9f_LFs=~nwYbHmAq{1Rju!O37qwD!g#NLa<C{%w?yGo}$3l>uf&gpwi@4OkGawH|
z#1~N{0<Aoem(3Az>i57IcK{uG_S+f4GMfS7{{cNf!oQ^!1My`cq(UqiS4>?{GaVFx
z625y`T6vXoqUz*8j72Ce#q4@R4dHm+&?%gBEwDUFPy7dNTc_*>v`cdHDb?xRe9WJ0
z7bYbTUv-8t+OR%l=Bx0HBA@gn=iRBoJ+683*8#@`{ARA7t(FYdBq`^?*?%g@e5f2b
zPy#JR1OyYS-WEG#QM4<JY--GDBPe3~NjjhGY|4}Qt$)CQS~n<~Q3~Xg(_$;B&$i$O
z_%5<KGQ&plM5G*4J4KIrpUr+~YtvXe^qC7L8m^R7tUwu&PhDNQF5$51M_cU~Kjfw#
z&$_(TNs6*CL&iN)eHh1Lpv_~Msf$q~2?cOV3<gSxs}474ltAe4jq*7}tDk*p3}!FV
zzYuu3sx5ND2_Sb&W$l4eh^om(vl#w394(pw$s8&b4I({wGgL0M()ocP*umR#Z4mDC
z)2H%X%~YNBhD!S3rwnx_zk;t2X02~>V#I;5Pv@GUyng2a{!Nc$+JQfMRji{9rgJ>Z
zFd6T05YoKS>#AP1pTW$sj`NS%z710sb@q&le0|;C+TU2^-?B;nu>KP5%3=+aFY6ZW
z=5?t5GpDKqAqer~3H1fECQZ#0_xCtsi)S1*?68RXVpS-w98yniq0nTRe8*VaPPc$x
zok@>`$3&!5F}*6&T9FO`jzia6BS_wF>N`e)7oqOTWLuh%+RE!o>~XxR32{_9RV;PT
z9}?#bD+^*Vs6OKnd#w`VIIly3nTy%(Em-s?WC^atWgB2UIhtiQIRZahr7Z1IBtTj(
zvcU=POGvS!0&?nnU$rWWbtmZUmn@WAW=K4xbOf^Wo!xnH+bRk+s98>D_CvLDH6I40
z6xt$q^P-in<UZ8ZcY>4Hd2e_ac6MF1kf@aHx5GGM$o#wesA9WdRJ|wY$T?E3scfZY
zS61xIrdE;53ebhqKnR8GWs$-yMs;rm{i^eEz`B;v8-9u8#cXm5p7V6Nh8Lqh3qa=!
z*D1-8$+sH}vi;eB`Z1Cz07yc+t~9W_NM|e=JFNsu3w7n40WnRgmCh!4VDmY;nHp9x
z5^SfrsEJ2+^L4;keI%{ltQMKm!xY&E5kGO(AW`e%^2u3QIhPKJSsEZrpjJ9C_U6p>
zv^Q6zBLVB)=~a=ul;l%dKz8*yKM|{MJ^bKzM&0t0IYtVORE&`x@w=>PTH?YNcPUzj
z#8um9GBjwv9U*_)5r6;oHva<g4KdqG9W&-=`guVod!GS!mf~;_8@92ea#&AW?}Iaw
za_*k@EJ}eG*x`9@u%E9##Br4H)w>a*frdveABoc?(rd?FIA4<dyGYf_2;0=dD>Vg&
zJ%4{(@eooV8Hsalf(W^ZIUv9|#}ozHO(U1!lQ2^0=1(3iKvA!i7~y?)mMdm$qXzF}
z$Fx#Ws!S?7mx#W!NX`1Wz4fVUr#&x4ZA8W4UF?|^V^{72(bv-Xrd{uB1f-gx9L<Y&
zYufpOch3JEznYN_F{Cb=+;I9^gaD#svNzsF$(2Vks;znHrAbVutLOZ6mqfNS>ri|G
zdZOWdYvu6IV>sXpHi2Ol5*H*o0XBxQI^!d}AZq6~R#Owvg+VXE9x&Ecs2GhkclQ3G
zYUts8C$UMcnMl$3wl}5KTUgeLP>bUqNYYhL9($OuM@*(YH=N5huBiQp^v2Hhyn}h8
zm*1_MV(BTEjr9wR-aQzZo04+IDUL?@Stt9>=-k%Vq=tq^`~9aQ43g2&2+z`GRKr>0
z21#f5U9ky&;otagdJKEbklR=>{*+kSByq}hJsy=aJ~&*HbOlX0+MsCRFkN@h#bs;G
zKWRuc**Bg>?x7i)@xQjS3n4*>raF<@DYZ1l&wE_vxXmbz(1<)pf~j~mi0l8bSa6|W
ze*b2RLZ#Np<A2|;k;^B2jpavkSZ?XPS8r5UN)m_Q8%iLD*=uxuz$@|nf(Y!SNRb*E
zuHNGV5hevv+RbxpNn_hV`186FIgF=x^`KdkX7b51HEI=myPsS{ber$_vlsj3oeq*l
zfg&cs5*ee0Og9bw(Ft!Sd8*LR!=EJ5iDSnXiwu@wMc|KUrY!|^7a?UM>OH8e<G#8V
zb%pZFuL{tWCE-!wOj<uBvDn+<hx2vVQprdhG8(kX<0v-tat>FPHMOgt7&GEs$)l9!
z0|dqR7J+Mv+OnJ9?vm;u>Q!a9gg{FRnH}X!mAJF*_(j_t3U9l*yQAEg-_M5@u5FH!
zxf(jbc!_KUJ{<2NbDn7(803Ou*IW)bLf+!s*#u;3rkwJh^oyhRO87)2Ypk}<m}*Mo
z{3hzS?unz7Q#T}NdG~u^h$rl;@IjNBWstILuM`1qVa+M~TxK6{sWx>jcvdxtbDPgy
z%X=QQ;B|McqtLXFV=-Lh=wddu{j$ZVS{xEc_y=8%zu9Upko47rUG;hSN0ZG+liP&(
zcxOrO;C2z(=-|C}Tm;4GsN9B}HqDUsYQ$2HosAw$a0Ca12B}uvu;+&fNu2##a~@uz
zD^KAk^@c0^HoJam(#<V@C!~7IwXH_N)9o2l`k|7<(~>+g@r$@37$GOWKy?Re#fr8)
z<*~=;E|iQd@>VAeZ>@i^rVW=kU!xf1TJ)S5=>L1l6UZ+D+c`Q=6iMrUfeZSMrpG*O
zi=FD!UKPBR-8$OCGdjeAl6Ii(eGz26=K?mhYYWm^4T+OpKJpdNJVi^xChF@#>bj(<
z_^zj}PM@2c6$vUOZOh-uxK-G$SW(t*Qs4l*r?D&>qT6=2iN)|3Na>d+oaf;LoAg%>
z6|?@4*A3dfiS{|6mqU+ZgHz7ifM<{V(jC=T{fXiE$rD$-wQLR}o0m~2n}SLqTNaq6
zI_%h$T07hRiBVdywFbL-zvzt|QQ<+?U<?KIwo?J0Av!axP#D~UNvi1D+FFGO))D;f
zi=t$=qiqNX4SlJnXT>dD21VXzwox=T%mrea9<al6<-=ex*78{_ZDVuYif24w$Q^J^
zWPrCEOyvtg-MV-Leuf<B&~LrpQOU{cy_)n(NeYvcy&rFkG{%^KzZ_l?3vPgX=-XkD
zS7R>7#GJ?qpwi3TZdz2@8#!WgMW9hO(S04Qq}n&YJxZqs!}BJenb1XzjW+iQ>@dz|
z)=_%HRm}O8_d~@PXnkX+ZN)>ZHP8lG(mO9hr$}$A)M^M*^hm_PK?%hlqA*Mu=w{E6
zq=fW_S2_~I_c#ON(U7<-at}L7r?<z0z|(e))$U4SqI!(NfVnr`SwwYj>+N$XppyK!
z>U#cA(T$`>*NvauKk0#t;Ci{3c>yqDt=w-o=q<VzE++i4MXx>fhpJKvwq6&0z1vTj
z<k6UJQWq2Sr2cSKe68UlxX;7Vdm2)tH{15^`Vs|ut4+IUf9?1*lGVe5qR3KiU~0<H
zg;MGu`kHWJ`7ViHgzoe*<6gaBpg4t12Srz64sn77rn#P5srOZnYI6v052;Hd%ps?X
z-P)g&)}HrdJ7XO**aA+k_#y3(I8nZ1aS=62$myg!DU$;4M=5yYYck>oqlNk_1wxtC
ziFk!pYKx`X;GDhm97~vZ`_MHM^~YygV=lQb0wQ}--`$gAYa2n}uVBqD#eM$ty!&3;
zC2dyz0w<a^3gPcC{Y#ePhcLDZWMl2r2C`S56{p*egz(h_PLWw)<<J$cv}u#r?q|@$
zEnN@Xb-lSJ8{-w3q%lk;H?r2>Oxl&UIHQoH#kW=QpkvgQ&(B(=`UD(~oT$;Q$<{CD
z{Gu8az7c-q3*hqWNZ?nAf5Y-ZYtEn04mc3?{NR~9naRlGXmvFXj_BaZ4W#?rsNj3_
zI%?M1<AOu&F+IZF_aPO^N;VMP<eOlDL(n(fQ;U#i?H-f%pSt)?ds7-N_-lpORstE?
zXR7_61d%7sejtDawQ)`Ss37d2dLQMuUSz`O7;LYqgPpcXLbU|11Lc7N#`#1>FEyjp
z&U_RRew-{o?At<ns#%Fkz~ZPgveHegK2+t0yLg2{%QF!-Q#17GF+U*%p^SAm|Lt4Q
zp^4!uXW*JG;`z3q3+;<^YHZb4*9?3=1_{{qIrw=wkuCk~leMPu2v-O1y4;brYRY$%
zThJa0#{+y!R$9geeY|a{th&&qhtlgP;HRg5<8+!AKdN#c&O&!UYI|r(2Nc<N!4&4N
z4Hl4n21<Ia?qxx+$_b{);@un+unI#8yLuC@RynW-Bf$g5*uz2RV|-VOjyV%tD>6t?
zNI?`k>wyr4dmN1%X3E)>F)Yf;v)nyO;;dcHa>(%@3!`pEYNgHO-`Q|cXc`KALpIod
zzV2kP$C6LN`fcd=##*lIJ_rMzUKeU!u1g9_U0??IUeG_GbtUjDKq$Pwoi3n1fzCW8
zdqe>9DG88JEO&FrTQgNL$^%IWt!{Z*bSL6EBA&LyEVIC?U(RXbdRNkTMYd|A558Nj
zZbmKG>)!^(ev?P9_xCHkGDd57bdVFuU(k@dY~oSPj7_nLfvRac$Qa6fqMHoNtXvg8
z(F0?rL@ajqlwvGId%7(S9tENC(~oKQ(Jw=s$>ni<mJE35Moyu4dbf2_7wmD%s-}Cb
z`7Rs#sov44^h5x0eltw_1fjFyM<>BHKvqzmGx|LuuzzTrvE*3_&k50v9f9>JjsH0i
zBkU912BO%sain(7ulJ0PPy#3;?NrX$3**Cj%V|P#UCAB*#lK<)E@qBI>uUlD!k8~i
zr2ZJ7eO56Fs~#Sejfe$#!3hy+fS)z$Mlv8e&u`|TJ)t{Ja%{%Sr2q00(Y${Hqvt^0
z5BXtXC4KQDGn4UGP+MGA`D}4;lTANTyGCJ@g&YWkZY+#^bH=S!u?{E=L*+Z;5)WC(
z=yQJ$YCukLMoh?C?(Or$BhFiq(j`f;sL&Tss<awQg^Ks?w6@fcyQW%=-GJp&dhIgN
zW168tM*nUFSl9IA8_&y;?qm9g>4X(jRT?AtP3mWPiD!tD-}n&9CHfm9sPF9fPeRgO
zv20F17w*u=r+n-E`9?|!Zg9z8g%j3UL4ZO14rkr0RhFfRrqALw6nmfQJfeQIt>XdU
zYLO`K%1LMBw&CZUrk*6j{yrJ3WsSPIvZ=hkN#qWGwN-!&l#qcF%uo^icN?Y0)8-Rt
z-S&!vO}^U{H2AV}kC7vl7l8QgVZAL0S5W&;!U(E%_40^rjl#E=!8yZ1sb1`@2)yj|
z2R^mn<nQ%cC{No@3*N~Cj#O!oNFy&8eXoRxSuC$a1Uf>j2FKvF-+RWsa1V%j%XXwM
zUmB*Usiyf*Ngv-(3-rSpROq_`iNz~k(cCCbB2^*GB33<ScEkH|WTXAuA%|CyBFWIl
zFoDk5Xh+U5UbeDdnrXSA!U>({K!j1M;^?nVc#P28EEa*E&0?_m9Yb7vH!A$CqJFW>
zBY@2q{k2^O(%1DDbEmmyn;Ikm*B&GX&s=cF7ZtU7Ul&`Jw0iWptk+tziU|d~yT#Yv
zL~CKvS^|NrYFZL(RWo|?Tp^5f(m33<ql{(9C}3?iTZ@)>sFJqwwDpdu^lZLuvkzT$
zc|XH+5?nf_@-}t-^OGhmV>fz*Is6QJ(Ltp5SiyLJJCsM$2q}egYMDV;-pua3BU9fu
zsJS)4YwV<=lGmL%l}j2_WcrKsmfvOCXY?uzzApg7FwIqCKDeE#fryC<yzE*!t}_$e
zV)>@mZdBb_E)%N+DgwIxgf8}VwDwc48XV^S9-2u|g#0DdXhLG6d_EO6^YXEp$v|*&
z5OATPNrkg$!dRCh9R<uE7?RHLS~4<ar|98&qZ7GX@%vI%HL0DxFtXQcENEJ<;M&%p
znbh@zG(r2hKsAHa0e;g|;yabPRQ7V{1fbU5<okFfZ8%6d)`B=l0J&exmr$?3g$DIf
zJe;XRfqM8VFi~USBJYx)A0#p_*9FtSJJbc;xsIOv>2|Sg@Nm<_E~{IIQMr84`#!sr
z{Ad$p81M`>K+n8>3UIpZH;;ptv_Df}N4}uXR5QD1-N#$=-zp|oV$FAoEq+TXS(U(f
z1V6L*9V_!==xk?VIY<ZX5(0AY2T%6*iPdU1Zn+XsD+`5HXz4zW2MZ!d+WK&j4}ME^
z1n-%ytqJeTrg%a!<uHleLiR{eUR>p*cJ%^2edC9eNtTRSSesAvI)cE^Dt!iq`~$D*
zV1>m=uixK19AS-x5F1nc-!cGHM*WqqxP75qcWPNR;oV6mY@vR4sZ(Z0TEEUT?$0Je
zqM7-9ly>+5rF{Vu89v;+5u$9fo3LCO3g>p8p?@Y87%ucV2d^?Q^eR+Bh}wAOju4Hf
ztBFlHW1{=qPr<M0^$qk`f7k8~^DVH^NU{IQkCk)=4afwc=i2nUp93Qw>4NQ+ek-w3
zrfdK~?FgA??{FgC8npbQquxc*k950Ho$+j^tg^nS{HJohbu<G6BB9!0OzQJZt~Fb4
z283FpY_Gj$X2ZB$$)HdURaLzWYo!$#1r!{r@owyZCS%+nCTfsw(^|Q0Hcye#qls%f
zfiFq5GyMq##$0@O=cABAXa32ujF>EUWa5WU4#@&*qGQN8EE7STS@3-F1uT{4y{;x=
z3~yOS;w|$fDDtTI0&xRLUVRxWsjw?F16CR@wY^B)le%*`^sCx+TZiZokMAZW*9Dzp
z8$Ha^5QO7?m7~NKLAyvuw7>%;mOmb{`((ZKt5QO|Vv;R+CF_|!eRE25kbix)`eE(|
z&%U&#LL%9HjY;uM*_7wNn|r1U^BZ(9;lQ?(`XcL2t~r9i_%c5sUaILN;C0K3GsJej
zS-i>oceTy8jkRnH0?JIz*|2npqoeF~tPbcgJQ#WfKo@TQPbT-=@yf@DK)bUTSH$j#
zKbbzJqgxZpEJTujdLv75K8T&fSx4s{C=6;g!$C#24D3%1YD}*w3WJ~VKBlUSI^gAq
zM1pbFyz~|e*4&mnseZv$i=7S}xsjXGn@W$QvbNhA++y)B!f;-ZRI<op^!6{8m>dkb
z@rBQnB0DO|cO&dS0P^fEtOiAnI9cd?b@Fl?efqUMe6^BQa=#cAw4QXGOJ~<vUOIXj
z`Vg!35r><Bx$fPH=QLglzp4WfDLA2ydDE2tJUqN%h7>`YgR`5_MV?oCM5&5Pn|c3_
zPw-<Yvrs~GnqPj$3Q=+jbgmQY3O?-x@w`v)t1RA$szX$;n$N9F&-@Oho}b32=)euP
z@3(V`?8PXpw;D0%2&rVwf8J_x_b$h#YkN<ii+v+eZ!5t6q%-vw^i;n~dL|$KCb?Gh
z1^~|=J}J4zV4A2VObkiiGih5|(KFk&e1bM`2)_oMHKQ&FvA+pzB;`au*e=$)+@tDs
z*DfQE@J?H%W+oqH3*pXj(`}|iU#DVV*1UBD&UwvNECtW9<1JR?8O~Jz?{CG3-SAc)
zOjf{uHzadS>KmBI8~pLf37hR*uzX+=p?Cse_=O2AnL@Ea>o*k}l8p-asA&l9`I)s(
z*8C&|^6mLhbL^sF$);I1wY@<nZl@wjRKQEZ>{Z%pl+X7=%8g?m(BsJU%7fsH+aG!(
z1F}^N!R*<@>_Uj}9_kZ>L{m7BcM^sksKLth8=_oVR-S|-QNp`7;W2R0afsnSxesJ3
z2H)oLSu%#sf3ux%rt0S#5;3(8(l16@iAINKG22u_ee^+;K}R<7;%rgLD9!byRNtjw
zo}?fIy3JqFGXXdyP*8PZ4YTC$yuLr+o{Qv~pw&f?6yp62og<OA`pW$TUuZnVZT8h2
zR{2`0bguO|moi6(b7WOX;H|=l(YY0Wa=*4~im-Ad5gc5lzhr?xJ)jBdI^+SYswKH|
ze`2;2KwIL^lfDoCByJ={@U4Emp_fTZ#RwyqB-gQ5g~VC$5<KSr7(f4VLWK}MBeyUL
zZkYI$YzBC!1gMGKB9%ii%ADmJ!gkSH8C}r8ToTx3&k{ri?`n&j2vR(kd}9hFE|SCr
zwTDlmTj^PDgFv==u~E~1;FbSF3rm3Fn9INXDIyC2>5!C@XWKrEDQntTtSf|&lktS9
zjD0f-uU5a>rNY&bZe3F#Qh2iq!dyC=b|{s;2gMNRaFwHxp^3R0Nhl;Ui9#OYcsZKh
z^)+a5WsQ!q`LIJRVBZFE#q)N1fvP`ppxj8;nO}KSTt(4Qn_GcPkqe#-J~6__*7q#G
zeP8vl?>$)Boo^3*A6s_Z5!->24;8;))fwZJRMIA5LisJ8H9bCL`M{QjuhBC`n8Lm&
zH>5WN{OuP>iYLa%#U^-HQa2~rfumc()Bw92Ni^*`snLa%Qz#Y0n0{+*9qjbLZB;5g
zICh_%i)(WKEi3z^hd*hzK#g-eTusZt_3uN}CB<PI*q(i-*dp)ypG%c7cX=BD_{&6l
zOV|ihkLD(iA!#deFYkpHDSd2ud{ZAL(ki32kCPV&dq_q&JDYmcfnLom%eB+E09%z7
zd_?KoGgP(=+4!F7H63-b5Y*}#UI}elXAK{Zwfs+1S*B0<<?wrKVmXozEOG8mt{L*W
z!{2ynxEK9EnpEwW;p+_B$uFD>S$qY``C>-@UKuq%zmtnMAZrAPIk&xGFrj=+FJ<VC
zUum(ZUhOmcOF$N$sRWhcO5$sFy7c#vgp&On(vLvgeuMziFR-)Zm+|q%4|F(PKKDuc
zfeIRN7s#%kA7Ubd+eOK79*1|^!L?p}ei|g>;pM?cE+ytoATeUPj$GdOd%a`$sSY5n
z{FTQ%1C5h-e@2@b=dRUf;yX)1fscveGwbT?&_a*$f}E8&4VJ<3_Z}_l7cu5;cdPdh
zmDMxVCO=_=SYuT(nv~33TgT&&n7%)+t9{^2+kHSCQ&f6vCh|QwUEvn-b{IG2=GXX9
z!NUX3E!j#FZ_leMvBwV^Iw=C@TB#z=n_a2i67JZ(l8)<+3CqXO#O+TyCybOUF#7YA
z)$tO`WtH*IK`ShsW9TQkPNCv6C<r@~ILyKN)Tr%25@%uW{ro{<v(!3vEDD}R1?<$&
zdE?5qDs)^eLVURX+%BhVV3jV`5?_*PkFlV~oVrwf=zMznC%y0`rQOV_KxSEwvgMNr
z)z>nOTuQ9uy0$NV;a~0MBO>1EY=DB{>l#bXs+dQcc60XbU6+cT)5>ue-cX^Fkz|O0
z#36&lhmQX~K7DQhGch`RIonHU|Ez;6Q8*!nz$Kt3I&owGq|KUBcSdcyoqU^?N@EnR
zi<x!m-*x?1Y#*KTwPAFiA=-dbw@KV@N(z}0r)%7#Qg@JxZjr*yT3F@|K+)xNBA?Dc
z_(O&Z`4q_lorc=*0Ua^l#W@r{PbSn-XTLCMa(Fs>cIB@Am1mF^*lf@gM3`4L-^9b>
zjq{+rLwuk~JsEPM=9Twbs^~Qj<&-GhcfN;ph;NIDR8jyLtrubFp49%qJ_B6-7c*0T
zZ(YA4LPW;%MxI^NQJL>C9f@293xTbOe5fh(s!$r3D11vvapO+c(Y{dIkvIYVrxcJ*
z2e2MXirC2LIyiHWDD+hNrUyCgBvCMoGg!J338xUshTTvei5nm{dC+n)1H*WQy|O&@
z{{@94dfWC@r@W`!NdOq|RvzoZ`lBA4q1z(QR3UUmvg5yzNFr3{dKV+#0R#CrpWs_b
z##OL(oC+{H$LX#a?xiYsr;V+x!4A3GgCY?&26a|-lvfP58FN9YSAllAjZ8SPbhEW_
zZ#s(_42kp{T|tIX<FQ9}y?iOhrMCeZ>~Ww&l_LlZF$WmKBtK~jX7(=2m%Ph(IaJ#M
zOo7T#jg_=D)#2I&;J1E*ImPC;$Fq4D@A8Q8>gR?SRInS|2Z@X<i>XA$u`>dX)@w|2
z%H4v#2-3X>Y;pu+r_Fvaiuk?&*7F{a7+CZD>euL&d{@M6TmHdS$kc4=`!jidm)rXx
zNt;*y?L^(0GnTgUM)KUzPK><)_@5GTLeXbh+rFd5#DP{-8`ZFW9E1dTcIswC(AY6r
zX|@Xy6NnQ?22vm^>wPy7=Enbjaemc&V!AY}U~|pyK^XHAc-kpt2rHp@(_*^QfmX@O
z47y<<qg8ncQ4bykka12ENG!$5cY^x{nx>p{PRUujQCAY;n~=HBUm=B{qGqUkd^%Wc
zVwiE30GTB}tju!np0Ac(9E$h*_BS&5KJ+5OskHD&M+~)Gg7OyjhRzZ133h5xOVuyV
zW{r;@;lNE6POy*LBkgG5#qNw&*u;I~U^)T8<ruge)tYRB%^x3U%x9?pA%oB94K4Gb
zs~O$JYsoHXG6{*J4^+a*mCp0H5ezTWyvAb!JJx;J*-d$;GDT4(Wd#lrv<<GV?s}{l
zeE)QFuXE@1tK1r|*pugmTObaC1jJ%9MJ7MBoAB3dR$L3>jh8C`BZPM*@6Tg>zBdk+
z7jl*X|1tyhF5G2Dqokb@b~wmkr5Gp^_6&_)NMH>iAHrs|(M50L@QqFYVCl`uE>aSg
zW(S-=g~CBYVD3BP?!U0`AE2e2xnCs&STiN4=b7U*QZiBVyE!e^Av>v8;amP_9trIf
z?_cUmpMdhdZ})A~>eHC%cqvDO8fC;j<d5@{4{n#Cm+X@_{gU8db7${_;C^+N9U*DM
zpSqjIP}$WCx$EiUsNi1x1gj%~6ENPaSpS)%=qJG^-ZBtf<6)iW_Jdf?u#>zmzSoHF
zUy9m6nm_@|LVAOXxg)T`tAhQbERpS$_|t_qonH4mGBbs2--auihG}6uOXtNmGIG+w
z{;V9?x!))4YK~6!F)&6F3|&>Elf+^}Hwhmod^xv>sDT+G|C=Dszf42RFXQY6%5(qv
z*%5G1(9v$zH@o2|Udm&hv8Wy2Gy&v;F(d=dweaXsh(d>bqBA}szPi4U3I%J}Wn7uu
zqgAGlKm?@YzMf#Pc-ER%$hK1Lt6aT*zk^Wem*h`4cK3n&irL*+ITS$>Lj0%W(8tmA
ze>3_)<A5jQg8`|2HI$+3tlZE+T}CCRfuea!?d$m$HdQJ&?F=Lg5PjhO!5R~;WY!-y
z4$ZPhYVk9MT&VLW%muxmba7MckW%Toh-xuedebs)z2xLjOyS%m?bg1l91rpf#|Cfb
zNITf_{a;$B<Hr~0Hcx^H&_41z1~q#JmawJFJ7e(gi_d9{TCAd$;u!c#y(4n(;SCZ;
zYSrjGWc38A&d45E_|QNn3%i{zgI#_MV7<7o)n%?tT@zJFxbJhn8g2~Fg~jAW8E-qm
z7*15|q{mRb$1h?2){)pUCu?gP?5q?#k7-I2R~ditx;G7EsZ-d=7+%T@oS3myG!+JJ
znzmibLe6Uq+*Z~u^;*yl$i@PH3AVl<`wu|FemX7SO0}YC3_}^8g9YReK3y;mJt+9^
zd27Xgjd7qgVJQ3?8$Ms}I<0uk@{VBm=%$d`TZYOIyXf(=&Qvh$^7`4^{ga!KxY+(z
zT%LcPE&2gM+U77>*+3^25@&uNPm2B(?4f}SN`)nPY*wZn;Y6~HTgtMtm_-3!`}kU?
zuNpMVhjI$_zOWUgxNHwONm&*?9I=*atFR|Ts#V0Dl0q@;twYb8U-1z!lK&vyCW77o
z%cx!ZVX1q_Yf3UP5_9{-Y_3=X$(FFRw8xdhSB;|6{t(7*=j~GF0Un1bGSJ{^v<we$
zz7c%^2!2@3M;=8!JmWhI!~^wG^FqA(_7T~-aHz+XRf)F?^^<<!0|$QyBX4HWEFO?L
zlaugbaOW*0*%6#xBC}T&DgFNK13pfLRJ`Y#2n;Tm*SdF&o%Cib`r9&K)B1~b2|Pvf
zRf(G4ILuRaio+v=baAsap|^AWpO&XUy0yy1S085Wm|!iNvafCRI5mX}8;3YS#T3!T
z+0FOzCqMsHM$eK#Yen(2AX|SBo#oBD!E1^c>w|j7N&EJt#!?`o$A7=}xD;cBl|Q#!
zt}H0C%_O|v&8w#CZqG-Uqg$vG_}@$<#g6Q|aII9>BT)F&{_jZLeDtY$q6oCH2$M!1
zQCJ<Oaj4%MoS!OH1Jw@v!>Vqv3yp&cA|U3}j@Qp4C7*d``6COPgh!w5<#69qWh~A6
zIAU7k=*pmA3xwSDuJQ~HeF*i8OxjPFUB?ZTpR$fgs>*bM0#H>_J<?VKkE^+c8Y@f|
zM6ChiB^SG}%qVg#nACn^$fSeXfrtJvJ+&TxL{Fr8qlos#@rF~Nq{pONwUk5EnVhaN
zh=0sp1hQ~s7NNb}FWyu_by-_AwpS~h6iD~FH=>i0;)2OfY!dHF+E!`{-2NANRM^u{
z#II?0l!|jvn0_=uNL=#dmoM4Io4p!5z<*Aj;U@Or{DTRIv5`f&uBw#~ed1eXf{5M=
zdB*u$1;{`#ds{Dj=*5$l2ULo_--}eQ)OgvRc*0_dhqo3FR5*)R*gdDB99pV)Q7&2;
z#5w9i{DC5^Y^lE&S`VC8)k0Ahumc=YDj6(}PaE*vl$nS+KVB~#X?YMUPxh0MS+y+n
zb6`faWjdS&$R(B{N;Q%(p7A$%p|2tS6>--Nfds7nuOv4vd$@(}o|wA8?)<E%ga_$w
z1^9aDG<(_V=m;{IF68soaVTCfG$(IUF)fkSs5l84_MIJqUWbNV$_m-{iq@kITIT^j
zJU5b3KN*m(Ho<<xpn+YjlMs$N0?4Q>*1h6;W%(Cn_F?^@3Q@hc%xH`oi&<C1w>;0i
zF;t1VK(2JS*(BtSCu@2a=Fz@gS0Fh2&5ha=Muy!;<jkOukIlZ+hlE49Uhs$!&XFKG
zPQH`a4twn&^(tAZCSmt%nOLR13vjCYQ5M419&fpYPp=7W6a{Hb)-<nhT_((<qJ&Bb
zj1K!3tNR1rrUpiwzVkmQsalfM2Hb@dadcr&Ud@TPz&vP?mpb20T0%1AOqYMZ{}_QG
z|B#HTof14FxOY>vSU(7p7Xs`)xaRIX&3~&7y#%qv6FV?TdxA$sv_U$1&-29U+>vvO
zjL6F8uHo+`9#fDn@^orUqYw2`H?An53wXU~uSuVku&k!NDVYKwQ42Tvb><!l2q^0X
z(^r&Yq5_O7Wk~qR^@3nTbD4T(U@S8Z;;?dQpr}<}tKrUmNFA~0ba$8|{|KN&VsX<R
z`v^}#@WPbJ1qs9w9<c-zEDyY*r$+`M9B{lbSW8d5?iVZtI<+<2Gpk)R1|%}$@>mHM
zo=G8}PES7`U2ScS#UqIUJ$M`+JR-*)s=rI?Cka|C;o`tE(90O<ywSeN9*<&hoiPET
z<Ut|kdx$qZ7Wzi;BUL$c6MgBjrT938P}HuY8AOUa)$7kkV^~z}wWO+Xae4Bu9+FkW
z9Gba59A@t?63gwCUB)t{=25AOmKIn}gQvke){mjT0feuPa>{Rlw_O`2cPfB7wnGje
z&o1v12c9JVMmVNf<k8-|spr9u)R*iR2)3Layj7*DoKfys%VH}VQlbU)eILtO`sM?G
z$OT&OT0eAh4iS5&wX}@1v<H5NjvEjtE+(PGBXASkKcbo3Z>=_vo4oRfuoC_S7#+UL
z{$~`+v3m?5@kIG#%^Awme4Igu7VR+$M7hV!r0y76G+qZ<AQ(VGJf}R&Myaw5m<*H3
z;@gnK@B;ct8!6QlOGFYMW|MERzCt;nK?(e_?8F7DXodTXQQ@ohjwK?Hgy<o25_gw7
zTTWeEMdsp6(RlA$00#DoyUgih_?w|cAE|*WdVbXA>D+Z98*tN~k{qfZ-Zmd3sSL|o
zR6qYC=$qLdWd_8R+O5m@%Q$Rn3AKH?DtLhSJ~p@Ha#%xfGjp^&f{1v^^kxTvcP7WH
zAQm25hHGkVXepWi_g2`8rz9zg2V@o`(bG7fN1teb%ZtIymQ?>U+F-WQ$txHY#c&<5
z$S+DC54xv6G>q5{2Su8+1R<Tf6fPjHGJ(~)HP;tV&uIN)J)B`!ixPc=K-frz3U&I0
zjvP)iuRkzKr>Vzs*YGjIF$F$rRvi>lg-YmJ!Q*J6&#WnCOxySq4@WVNe2F}vtS_LH
zHyc+ZtOH61q*X_;@oo=sh32$%=@5LmBWGp_!2$u;IcF{o9VYkF#Rv~kf>X3KHh8A5
zx^@Gs-wQ0QUW=_gi$-i8vNo7k3@wOzWAP7K0}$jXbW26w1A#RB_^yKNKwJG}VXo$Z
zPi=C;lNUnyVlpMS$bKCp!%hbM-G0M7G5Hd#UG!k9S^%?s4@d?mu{JdoNo5x~t71V4
z9rLnc1uvk(S?Q9~>gn~FGMtH_SlMfY`bBDY3_cM-4~+_LIRGQ(b1UC=)X|`d`{u7i
z+h2&=fBo8!I2akR!g%-k)M%yOyNqNpbG&H^=~f~R<lm{y6zL*VjYuHglvud{lix(}
zM!d$B8s;c2wyD<|teRm~npMkIwIN}}sxhan3vu_dQc!bBh?q#DQ20Bm0})G~aelf^
zHV&nP-ho}t2R03qNOkDEg}`YOVtOC^87;s_U8Cl^X%mnD6(~{$xWSl;DI&roLToRc
zFd8vqzQD_MGDNj-B8Cg>wKwJgYU$Te_yWxSm*kF}b)2>$G!UjZGm7U<Cj3^0aowo+
zb_I48Jqal6-f9T2Zi5>ZpRqoIcaXVP{8MjauCj{bfzpFM*I@gCTmTH~h9^P$-WZjs
zTu013)@8)T$#>Mvl%ennf*YJQ=JSzB<oZT%AJX!PJ&jAz!>mz&MA+%AgH3JeU&aDs
z_X>Z&wAQ>aZ%t}VNLfXtML6=G80F>l1=OvGS#O1bcSud$8;zd>ZzjbZ3faf)w_2h`
zd3o&9n6Ht0Ba46AJx^LFe#x1BltG+zE0PYly}*S~+(M@tnVP#cC5E&3o@lhAXB)Re
zAErm0z%ZJLU8NV$<?vINvhT<$3QN{1cqS!KZ(5wcXgK_5uOmVCGBP~W*i4_1xW1z1
z>xF_AfgHQBrbPI4k;^v$6pfhQnmC&dc#U&h8$=TwxDi<ct{PPE&Hy53KnP;^L>_zX
zIlgmX!-9oOc%0LWq{k0F99JjCXn3z}j)=WZ;sN;e*_T&+O&ICW&A$*Sl?JZM2&={t
zC0$X%`UD-nH6Ph(&B~AdsL(Jg9PUqGsAaLo{(LEP&_ML+0H^|jI$ur7c?Qr?mv$e;
z9{$`17oP^LR3F==OdOEpFtxzhrKk(&Y_SJt8Z&%_l^gHX8a?rw?M<tEn|Mew6Z5uy
zlGtly%^Nn4<EeAhK}z2$Jn?N9G5cC$zKaORmZbkibb+nU{9^B)zM#~v#sd;|0{gGt
z9m6;3XS=i;lh>*RtbVK1h1vjPHW3J3d-}z{6Twdy+ll-A?X~E9sM<4)HwVU*Vx&++
zX$kIqQT|hSpeqIMnRm#F;4nnU!Z{P0fivFA_s{A65~@btJ2HRtxEQ543Juwjwzx&>
z7s+XSE_Ab;QX3-Swg~$?-^K3UPHtw&Gj$VkgYiNPl9TB2|Bm^uC}Q^*4mN0BFnA&6
zkae+%I+W3Nfrb--vdkS2ScgR3?X_qlI2$6A40Q(CjWc&Ei+|Z;!Bc5?ccP;69onu!
zhfEYOWZ-?VRsZh>WTXP-6{Tt+S>?}1Ag;}44km9db+K=sVGur|IZ19bE}e<GJ&K$T
z+A?UVgAJdXH1}JCWH?@i+bqEK7CMJpDbI*|+O5#$8(R_8v20}fOL@amoEiJy#h2`U
zCR^$yIUu&J(?=?pgZjo^xUZlviBcz<bc9!!fNX>o10}?s7M=wL*@=xf0NS*n&lcL-
zlIa*?V4shQr}{2=q|jl&t}4e9Ks&T45QbtL@msQ1M5HuHBNgIFWOK{Afy>(JFEFZs
zDL_CIz0Z@<zV2~BT)T>A%7Ru$yhX|m<>|d_vkLb^Uo*YBBWtye{RT_J=__clZ4WrM
zr^P!U7@sJy2hA}<^9eQcOi@vpvE(B)4l@aGSR>{VFbl#5sy4HG2Rj#1_oGS){=jJ*
zb~HyNF&+gak-iFvJ3FhM!tPjeyM!_Q%FVDKPdfs_)BrKPQR@K8ev(R-zK7w7uE9b*
zx5#u|ddbw@v3CP?jcan(ze8}b=u(mZJ*rm14pRqO_*<0B^H*eDjqP4e1)cn4Lyqi=
zTqT^r2H!{mNdVzkb7gI;l3fQ#OTs%~Y3a8SM=eIF=Kpl18obEItQzO;kl9m7W-ZG-
z+_@kKT*mlVQUC25Ibg_dn9%$zAf-xw8$w}{mFqVu^hrV5xI|np8_V_93V>VW*KC@v
zST~pv5x!j1-MN<^-p9mK36%Mn#o%=a6)sgWj|XT+X&s4tmP@?asQTY(@|dy>1jt#I
zMaqn@a1QPUT`A2h;NS2VsYhy##^4%k9rs&ZuY*TrS0(=TF;|;^rB;+2Sj;AFWH2CK
z0BZ}Hg|>DZVL@y(dp5h|qnYwItMd;6lLt6^tjfWYd(O%upOCV4w%R$~wnJ+8QjHG%
zuJ3#RRD=?GP>WI1w(+Fv9u7phm{4hFcC}gQ2?IZ)stxye*F+Y$5Z#mgpJ-ac6y<1_
zgx=rP_A1%aEi3om4)8=EHNjRCfSR(b8x5BK@=WBCi&n|l;|HCs=@7-33C1^hAb^NU
zOtrWW_V{KggT6f|bo^mchfrHpc0#|M8j#N9)MhjXxcc5A>zl~!?uV`g*afXUWg`p?
z86DF+mAtU-lq@rKyI#>Vk>#CJ{B2J{3wjKQ@!&wN3%L1$ELHIMsJg(RM40J`5b{JR
z)OzL1zwN|B+L_Y$Q*$rtU-M<$DNio9fi3jLg_3R5ba64;lXG_makutaa;FEI>!QPD
zqb;Qyy@s08;u8|WTHZE21VPJa<*A!}s4Rp?J2N0e^ZSMG=KY($=IrgweudGBJb~jx
zZL>AT|BFp@7{K&^q9x6fDPxvS<BATA$CT<9JT*aVwal8Ce5Fc(sU!t@M&SM2M;koP
z!eW;vw4IM*BywKp<993fM6duRT*(g44f~}L8b+gt8pgWR8K|}I9-}bIO->;Sct0U}
zq1~n=HTOK;dziL=A#1`ukow9-sLTV%wA4WK8DHWEc0XNDXVCtb#Dyz;qP&H_wx<N3
zL`3xaLEIi9LFlM_kZm!%H1Fc_`<JS1b#TLz;T4_}MT?~wKSkB=z51x#ufo+o0P*W|
zT!8Qr&P;FqW#_EK)u3l+S<;|9uwI6C6*p=YoYfzEmR{WUs*23JA4iI|m2}HuX)30^
zun&C>FU=2P2umA5O5zH;uxS`hP@P=yEdvkKvJFX*-ZRSSU%o?@Ya_FeGh#YC26RG1
z7@3(}t<(R0Jp>l}!-7QDkDX?9lb)RFtCwh5thPQ-qN(BHvpkg6rx2irET_&=V2Xus
z?Z45NDhVuf<cFzY7_XZSZjun;H_e*{@@=}Pp_H2ee+Vo$v|*k5T70mz_n|d>+s2(J
zG834I=EyceXd3}{TjliuY38LXm_-#~CL$}QCL7H3Aco;oXV{52wbm`P>sFNioy%A$
zW2{<;w8`X)kMcOB4ksIEL*2~36j-1~EaC;8XZ7K^qV-!Z3_YQI@8H|HZ<SY%pKr|{
zjKozr6+-=#ry8tKT;0W;je8;&`v93m58q_1qOR=aR!Dq!SU_Y~iuHi~NO1|zC+B?!
z>hv7ps!JN$D|aH>zK&37Y>{Ye0c>N>K>habQXR~}ATA*>0fk28<Q{bxswUD#p)bR2
z^|b!Y$vTsqBL$QNI}(Ie`HO)dp|y7T;_>Z=_1mO1`kH+kDTnGB)dwRQIy8BFFhLmW
zavU{*%YlM+?T6kZ!j{rbf6Ft-0Jl>__|~>ts!o&;;^F_B$m+?}7?(`^eDRN$#z8lW
z-)jY(sq_t$pe|7X{Py7WF={uC@Q5g@oLLHKB&L^ESoq>&c8vO6dm3KLOk(+<+yg$7
zut27;qa&|HSMd*pn#?ks204n0D;s#PqM#+7FD0K6sm4eE`AD+<RtQ}6zam0h%H|oY
zRuax;BeuPS<pvQRJHx$rARnU_7lKU)DxrT5h1`jqk5V|Iex4!^4cp)t=0<zr{5lEu
ztQ<wM13v@jo+norYBv@!aEB4rc`{C<shl=i;(&gRLoFdxm_it_+G+1<&@Q_D$kn0O
zwP54zyxQ1xcK9|CkjwCfl5O}#d-?D(?f*E1jmjD?<XSf&9FkP^x|;Q&>+%U#^K4q_
zPMEXfAX0JXEM@1JFkoJp%w(tB_4VWg+k;rd1+_B)eV(H{GhpMP{?W61r5Z$N408|O
znLk_%xH`P(q^djRG2(D!aY#S`BIZ70UDLX+LBch=NVexDVhl2#u*;7g+#;HC&nxa2
zB0aU5=YMz&UV^3`Y26X(ZW<KpvAYvsdK=6_`EW}`1s)icuK=-;04oz~HeK+Do9a$4
z{`$b>Wm2>OJl^X%r*6n=4F`Un@P@eGTVu;?&w)xu1i@PR9#ii^8Yf|$?Cf%5q0*2y
zemKR(;N5<ohs-_UEIT2OtbhUS9Kk5wl|M9;%b5;t`#-ZmNwf5ueU7GM#->A!xWu(}
zCTtmNK&#39Q2pw4abuAo{l-gG7Gt;0!IYA?NsWOYJL0)Na|Sw$mlcLy92HNt8F=dP
zlmDj86D<RPI`2^-OWyNkZQ+1lfT8f#jSDv)*q+=Xa_gK1N!%IfJ6+W3QEO*{xu&yt
z0twm=PjV5pwal82gJ0w~jllGKe8ZSVTnal+2D3dVKwL1nUJHy6XsW*rWM(VOdJ!L>
z7k^H%`O1T%XJblDDz$w8(yNrLh5y-`JeA7@{!q96-zwZFrd2)5+Q2LgW}^e&5Qzc>
z4@pVj7W<+9n>(R}f^KxlBNl2MW=0@SpEgG0T_$z5Sok{e*AhiYgg{#YN^w5FL5+dc
zAp<BJ9R<0Mc^Ytlkxv$c9KOb6=QT#ovAkcAdaN?(><mv!ijKKl#!_7DRv&4kcIPNn
z-X;G>l^hiOQ1w*`dnBmOnJlq0q!wXQqjnjNwx+nL+y$|*tVB*#n<zhtf7n<(1@I0Q
zR3>{vm&VPi+G6Egj_cn8L!9x++OVWxL|BNy^aOnQPf1V~MNo8BPk}vh#0?`++ly4e
z$-J_a!)et5-D@3!ioSBjlURD93cQ;7&RZPXcs3;R86K)xc*P`JsWYB)U=S<l;#DCy
zvNYM6B_hSZjx>;_2YvMX5OGetW+iig85)0Q`_?kHUSR3y1JNuCez}nU?oj9xV-w+G
z<Y9v}5Hdq*aQ;APW5pCdTc~pa#?j-@HnvC)Cq$i~7NvdnF|k6@4wu*U!lh*=bATj4
z5bW}$F|<3AKJhjNwaQ{2T5ab5<BpgPGU0AYRP*=}ZZX2IT9fb4caOEq2(3MK?J(}i
z-^9KMQ$+Giy<eZb4qpSu+nn^f*VX0Z@FG*!&t0RN*MKyvYw}S`>4DaPv+azI=WShq
zM9>A>fawD+db8r04$hYEZE?a(XrfUmn6+oHJPshsK1gb;pV#M?qa~T>*5QEC8U$Gm
zZcd?gO*mkDNWpOfNe%`9>s8bPmF`aaxUC4C;o%|@eKMzkxYuruofQQxE!_TE!48jf
zrg(wXTh((B^t(2~Cdk;-SP^$3#Zj%@QDY3hI>Wbrsci8XgKm{O_6igt=PGuRPr|Ur
z$lAL>EZshx!zAgRkIdn%4EWTVzsz>{Z9w|j%Q!Pcc%M7B6$@Mdnr-h6eWHk|Tng33
zU{%g#+gT2>*j?Z$$Gu6y0?Ds3nx74fjG!arpXF^nKvP6|DFI(dXKvP8!wy8t>4~l7
zeRN%759l;o`@`a_U&+x!6TV2(3ULnZ&K>DTD=ySRKH4&X9j+^0r1*d7VZBSzT#Z{u
zu#4&{Bo}%hJ7+aE%f{EC3%8KL-I}Bcr`B{hj3g27b(V<!Og#0i;r359_JgYDCPyeF
z-VHE-FKG#V7udt}c7syXx-tyaloCkqVV(L3iFuu!Bq$yk7O!KOavTneA9(yM>P?_U
z#_ne^RL2O1c4NJA=Y?vON!-hT!2XeiuyrqeFt*K1@bNDOH97M~QU_2^x_yM0vlUQ%
z;+(vor;|0A?ESInM8m--sI5COzgA+p%mpF<pvLb7sywWRqo%EE&}YaUPh860B?r%i
z>niKezmge|!%}(_O>kDL%z$KYU=XDpeAwlMrqJz()#C{8S|%EfUTCD<2l@XIvxfOi
z)-1?%v8wau(S;N+7tpZtTzVH=U8hdaOi0ITg20PD(6#dkh-ndn5ox1t9yCpjXryO$
zypeE4T_6+v{6|ZkEP`E>zM4=fk$5`wf9BFA;N;AWc3g3o$neh|WHtYTnwjZLiDBkS
z3#iS;6hI*%OzuKx4E^g@1UGs|!!WD}{JC3mueA%GIqOFg_(?#xKmH~X8aDv<eiKuG
zl{m>wlN!g%z<l78Z3-kb_=#VST-|UuD6xpRxKNv#!FBrybw&u#m^!L|n$xkiT&sh*
z`Z<;tU??{85Oc?&U)_oOgbUDftOIZT^$FpUoHQc4t*K4^`3fW~HEG+xq#=KXIX}RZ
z5_zst*yjhMlYM-8!YmJa*SFW+$4ur?>^qlP*)7<~`wLk=#%R|y;vF#7x0_?%labzk
zkif|fADChR6>S+Bxe)l+>Yz1!I4giwpGKfS7C&jJ$4yfsVFN7?&tP;zdjKcy>O+h8
zmKeYO1?c#X^|)A)80SS!XBa8l0cD=yG}!G1?n=M(l{g%O5lhUe2VZ2n_w)aH?6K*W
zs|}?dH3D-b3(}|NquUN=1y@`ggO~$=355AqfVi$almeVcbm^Q;`y{Yq3+<6o3xVNi
zfG7l(D@=dsu`@q^kkUckoz+8FPjLY}RQf7Xg9^Fmz9LWo@K2%G5PkLkgHEsi;3>wC
zd*g&#S45r-0qj}k&OA6iK#7_kHoPWB3^4K<fTh!t5lfr^8ZIl1LA8s>r(@&!tfUjW
z=@);&8dj|p(Y@~&&6%8k&uIpX+kdia4UA?3LAhyRHLxf;GiL({_`@v(FNg9)8><u`
ze9c3O{g7vegS;bk0(5<EOrg5PsMR>tz+J?{JJw>|?hS0Uyj#~<(U(`lm)&dtA<BjL
zJFe^`p`JJA^(2crFhB@Ve;umrGa$AJq)y+G{H*aG@u|gftaUxhHF~__cT?c@HtMDa
z!SYX|8v+U>Xo-g3uC;ulFmeCbQp_!h(2_B#3ddBB1$|rt<LwZ24pZhDbiNF*QO?SS
zYi1fyP~g{4Q-@Bn*tp}owRiZ4sFoW}B4h{U_k(^7jVaK&ZV8ymWl$RMOTNh(2ffQI
z7m4-MjjbCfU2|Qr7clHuvIxu=&$0;Jt0a4Li`~QCX<0nG3u1+XP2U0VONmc@dYH}A
zqN?fK;a;%iHoGb*%Ago?<8MOh*E6?)Ux?D{-_KtHd9OA1#~%mj;ac3je3IgO(19=m
zb)l_-3r(OcCN}W%Yx35nNR^LIf!OEK*2`&#BmN>zs3V?8|B|AIUt?f@G&epOBmfo#
zFvEi%+pACRL9<?8kyL|+ka}v6Uq2|2LRQc$z%af@fzr)qGwC=+rdo(TvA>`?7pzpm
z++mcu*0AvFILoM8Sj2nFkqTVm#pAdg(tv}_Gl3iJ+(yA`NBE5d(|T*UikE+a#S&u(
z)zTR26`i$PgFiKv<~4WjqWQ-gP`*}@Obk|X89{N7YFl*N$wrBgKS9Yk>4{j8{y48j
zbIid|ENzrAMuymc(yjUJy5F6%d-mWPg!pl3ma>D)0s-C!Kr-_^q_Z(Hz+JmZZc+Fl
z`;N^YF4m%8&Kdd!`0MPs0@sRyBt(~+r7d(J3JjCvZhctN03p}JVsWzy+Kw|*N-xy#
zH>vwG{>@9Q%`nCl*&vB#hLkGAqX}Y!@{jzGgCXD*9_uh-=8PRBpUTaQ((Xo#okOfJ
zTC}8(ZQHhO{KvL!+qP}nwr$(CZM{3`q&s<o9`5;0)=I6a`ik9A$?M>qN%l^^CNsSf
zm}VP|BE>|<pFcqQqy+}6eP=1^9a0Em?HA@Obyf!+E~SlNPQ{TGO*Fk=1Bg-ZQ1jH-
zDkhw;H7<rDcAZ+N{ySSfGIZo1kIR<V`1HFAmOlTwI~}?b7ziMs^ZsuP7&eCgl>x)d
z@P9F2m>4*j{?GIOVZbnPvM~SuGhp0Jb(&~#wp?%5Te(_o&SGu0F4kLTWVzV5FP*Ko
zPS@tXa(mu7UOD?_+x%)+t(0RbHHuEP*9r*~6ipD=oEQKlIo7+F>KT}xfR<HQQlf2a
zVrHswjxm&RuA>3{W^n9LPzq-SjMS*Xy%>$u2<Qcn&A||WGlA4K0BLx5c<fOC6u`yk
z&8VTJ)Bs-e;IO#3KK~lOkAN5+KH-n;ilb6Oc>N#mqPXW*2NvMW?>^HG|2XAQAwb$S
zfdEfcTLt{f3lm$L14;Zf6$Nn#Ae+Itu(yDcx23IXqyZgOO9i~p31IuT2Qc-&?m+vO
z2X?0~^J+R6eov5G0WpBJaRN$J{f7K=O9jUK2t56X=Qn02M^5gz`zJy553H^4KHI=+
z;Q*+;l(g8J{Gk2=z_`7u3C7OC`SD)=8v0En<7jK^Naxhx=m7dPQAJ36cgd<r8Ljy_
zcB%h_0K%R6@y_5#@BD5o_SN^X_$jCvQG22LQ_nBG<Y|)$_-CX<71cDX?QQ!_;N;Z&
zs)S}mrSy#WsR+Q^gK1)LXKeuKX#XyLpZu0_|Fy{V|FwD<iHT{|`H6CWPhIg_)jPTX
zZe}tMHq`&lSX_Rxw}qte;w<qO5!u+?0_Yq3w(U&MJn%EP<v;adiTun>&wtYnu1^o_
zu0rdBv4j~Z+S~tRAqo8al`<XuWsm;>BmKoeeDf3C{qZ~eb^ZI>BK-aH%KhVR3C^mi
z32eZ<*Aw*rYc~RL^!KB~{Q4W(TA!U<z4|--+v8;o_@jILlO(>e{BA=Xd`0s+(cd>P
zJ~KW2VAnq(GB|-`QEGH#Z2?Kg-uk^+>ASvsH5D+^qW0R?llfbb1ptO-e)1byrm2y(
zx&E`@8|x#?@7eIH`>T`wtIU;<l~vXfOS%1PSHo1c`I}86geG6>BToL64XnycBSyII
zTeO(uPX9SO#OFgq1ef!N2Zkmm_YX+Py14^1HUYe^Z~FMrbNq`6Kj~Y29Ic&I%>~dg
z{by>L+E@I?lkqq4lLzMi>sN%>#_q!KW#+2<pC&r<8@u%%6DB}rCp+h_^1BXu|L*nA
z=%;P~9Edld)SfyDuGG6=6>}q_7IyuESjn&x*oS7H*dX>~#*$Ptzy+=c5Hn$mXXW4B
zi$OH)Qg<G%uAeL}|Eq~P{OD)Bb%p1We_9BLuQAJyXgPLrj<z%SfHc@ic!pW#o=+{>
zaVvo{M+PgLYiY6y%;atXL>c!+wn1&KIncy*6cFP$K5oLd%u?1NW0$<9z&uu@+#E$x
zTpnIU=%|=jsFxkV*W1Iz<DmiNdvwE(lAMI1c29KnU6u?#q23NBQmzbAMkfn3H&YM2
zQPwz3tM|-68}UAw1G|a}w#ZTEx^k}x_4=C<;tZPn+~3UxijtFDCB&sSA&>1sc2Bj`
z4Uapar~HP>&;?pGcpJ)AQ+wvyOSftvif4mq;okG>!M9FP`_a|lWqp*WBYbd$S$A^f
zeYW^fGasoTP|e~9du|NPXt@O|E=t_pjwo0RgY2=@u*MQ{r3`buKJPm}K?*p{Xv?MG
zb|cHVS(11a1q`w)Vg`@+q+%aKP1|y#&~CJzWfxn#acpqpKUd+DY-g{J8Vk{ByPYgg
z^PZb{Z%Vqzvlo+KLx%H-{A{Ul)tZ|XI{Ds?I{xqUls7r^-bm}3l>|Ok@Na*}v33$o
zUOrQgKt+N;{V<$^XH0Zok{Exi5Qy$L>JFKG{N`yN(Zkq2y;<E-O3O>3nRFR3T#o9u
z;Z_@VzB(fWH&{;*d19IO!*&fDR{aOG^mI!<04BC;>=L4s*t=W|8JA*Qf(0X%+t&>+
zsXvi-Z~Jb?>CL<vVkJJuRgst!LKb<ojuw?Z#Z#WIfD(xN3#*+|W#w66WUB)y)TXd2
zhD9}w)YMG6S4yw4t99K>vaLy8$mt@q4=;7OjKsh#b76fyZsVI*4=&IaJMAw7UCw%%
zCS`C*ABDF&d}1@9R->QrZK-sLt$#F=VUjC5dv5&xbJ6FwWc89ggC8<0OU}?nGE}xA
zxm+*AFn;E2L0YZV1CjRA#xIMX(`Nko@t72C%=xxgWJMzIU1qEuhJ1}OEvhCN{Ie?Y
zuzT8KvmISKs54f0B^xt+p~Dg50)j{7ZM;s-FyGB&>Ft{gvU`ifwC+)VhlCRi;aH+5
ziP6bTUbyxKx1lV_h;45&#UyGs@Kb*eT+}S3;I?isDlabseK>%tcdBJig!46e=3(gO
zH6iKV?N??mMM8d(6;3t0NU)Op)>)Gj_o8rx&vRLNZ_%%9m7si5;bmFv;LP(qpNz)P
z?%vD(kwVvP`vBltz*&(H<3LsWI~)%GQE(?VJc%m#2WfA&hR)G9Wd-_bM@@hbzS^Bj
zW<U*S++#U^EGcMxjxwb}FNU&ac;jh&S}nRMORkR;ETAvO#PK<tzLL1kZ#9D<0m;k-
zf!0CD2i!yL=^D(YbhLt}<0a+VCE8jO@mc8a=}Id|zG1hK#>8$Fj$S7d+>XkTFr9^A
zI)Ja?Ns;}wJ=?aWpkl@kjfJ9zU-iR^>~4D<?}iLgDuABj<KfWmKU8Z*7z{#J3!1eH
zi>~wBDss$-U|(4)mb?}XohKl=RAUn_+#?><DR9{)%CyIQeUN&RB@{1!b}QnFAESJI
zHEo*@n8!`c$!PZCLg0G*9x@eM0NVX}WI0^C%7u6t%eP%|$|_Dken%bCFa$Qz945VI
z*0Ko#By>4ko^`d;fJY+gk)Tnaoi&~%>)3ktE5%PnZN%^T3->C>xVL?J9i>Mk^_R*j
z@AFwcW**zsNw5Qx0*j5XYiV0Cm7~nDixCDR2S*I~l$oPcyN#j@>H1A4b!x?=#0tL`
zyv`)CYsQbqpq@fFahKl$ITN|exG9NdWTyO5_O?qHf+_MRqRokvfZviVHr1D1?J(dj
z-gXi~H*As|Lh$>9r`!bYJa{uoytFBM`t#r8>R*)^maO0&I8$KVz$l<=b9$Z2Xcl+F
z_CPpW$Kl~NfY>dxac-*4y@U=TVL09iuUy(N;Oz*YT2>9xQlTU>cI$QK+IK+Vv5@Rf
z;`|+T1dR;0zDzNAG$gWJMb{+ohvbWTOra+?KFh7a;&X-Aa`g}~#Sa>;Ls#PWZqy;D
z8s(zh&-J9K<#wm)O>PWX*|-}+|Ltp4Tb^I6MFg{w60|J9dN*`6Ktb|<7vJ*dOb$AU
z?B0oDop_PRqG*u4U!X06(FJ-)l(us`PIIb~V<Mzj#d(y*DUAr1C4z#B!u?Z3SEbk+
z^eXWslm^@vxbJ3glM!oCHq@B!#>LCksSXe6WpzYQ_Sep+lm}XQ2`Y{G+Gn*r8rTh2
zNDCF>qyu-$g&h)*V2K-=m2-r#An8yRpQ>)de6&eMlkSLq8oH9<4a5z_JcGhlMiK3r
z6j00YRc_=#reMEoQ3o*Q-B)#Xw?{CUTfaqro8&K~(KTmr0|!XII}7D^>c_@>P#2ij
ztn?Qh-91($q|(H^6sbK_dES%7WjsT-S-3#!pY}`$62Rg}9lVf#?8Si$liEA6VU%sy
zs3b4GAV3R^!E&|CC=`sy$6Wr>{%4MY8SKDo#{AL2o!;xwe0f8*_?Kwwb~*rFg*eC8
zl6WRm-ToTbB?$yvg@j2uU1V6h9DCnL5MUpLojB1bY4urp=)LcjDr(B1DmY0Rd&epr
z7AU4u4v|tDyq$(a;g0WBiGfw0)N~kMcp2v<R-7b+#g`Fo1VQPE@GmEKr^wJr9QpDp
z(Z;V+#CS-e0yLK?j2)p-s=^g<TWr*l!*Kw(n$eQdIWfRV-OBv%fcn7L%o{ovTLm*^
zw!la;8zhVEx%y0ga6N)Pv19zOfn$k{_V#PmVopP;?GVz|uR$VzvPd!OE@%#A1HWaO
ze59hMY~<RFAcEYnUJ3SObBhMB=dcH}xtSRs1*c9BDUAEc<qQ&3E~!^m5)N5M$Vi3!
zK}-|oqh?iCxbKp6Kbxl^O<SXP0H?rmtjKpM&UkC3y+fRl$S;qx-_RgM@qnFIbvjF7
zbk#$pC$2x*m$i+UX{DEW@z~=88wJc<_H$T%+@6+-FX678j2qJMwjpKLq~}N%c~F6W
zz^q~xC-^VG?HjsX&_URE1|&}F-?Hu;k!abSuUui2=(v``2hbz_xRC2WCG`Povw$-#
zi)(_H;*c7mcn7{^=}?HUtj!GgJtGs0-+6O!CIBq5t4h5=&rBMIV_Sj+>^;Udx>+5E
z$VDLtcD;?tTf;oD7|XO+ftvkVsII->29-)nz2dKFrT}}E9S}i$OF*cDNjNM&=n{MT
zeq65u@xR>HI4!+as|l?*QC|w2fo{ro@oUFcDykcf_NImYM7*>Kb9xoG4zQ0Hznn#g
zvbHN5(Uu>PuMiqP8$k4<ewKv{1aH|%6f91BM)1N_cZ!F_WL~(ch{x%)Q<u*XQv!q@
zsDcVSRciivs@k?jlgQU_$?4$z8qs$nsC>`GQwt>T(z4XHSucXKT!7lz%XVS^Ze+hA
zAJF+%@?6u{*RDm?xnpo{Y>pfsLyg}HYA-*WPS>%|ay(AhM1%hM&_1DF#;I711a8bD
zlSMNbWfI)kURG#aHEWpP9N3j~r6c1%;{L=xxtFo;z({#Y0LvJ%o+#P;crk?!K*g8O
zCW34CYjf(71~QzTa*6~1y!#{`g?8ZPIIQ69H-q$2sO0=Pr0$IK5}`J{7t&YFmY_!0
ze3MPrq7$=|4S$cFU`{&NCzeh!&Hge(g>7q_(FunYNFDm*_n&4l%2lsMl42|$ic)d(
z2dtA#-Q{QPN`!?5<kr9nQbDfU-(F*#@&qsoKFV;hr-6DbT3XQ5eYA?`67<v1TweB!
z6`QUEbJH(op3aY3GDK~?Uw!!4w^^%fQmfp?UNfngxe}MM&-d^^HZT#gDYey8vc;%h
zDuf6z>3iYOF(=%k$0V7jS(d3%$;?4oA&5LF&H|k+uvy^=03?eP{DDU)YRpExK`+Q0
zOlPjbQQegxox0fxoTwb$f!|$rRGCc%5iumlpzdO>hN4~1SGeR~GOlJZf8JlHh=aXq
zkF1Fd@EGMNq&isCQ5a+pqpDq?`~nm}s-n{}koSY{CzMuWX$3msqAz>qiP$+PfWC*^
zg;E1{kFVOSUQ<UH)e_|tO`*M^URGZVK+P#KS)+POFiK_Bt~5z?O@1A-HP#?%i@%kX
zh0kY}(Oioli<Z5Jlf+eKQhZ7}2y91#5~0_PGnkCHX#?spzu6WA+&(e6fI*W%b64>C
zG#X6<PDc4aNUxVN)=;CDkzm!MVdPhlF^k7M=P!$S<<)4qXcmJlzZ_VOc+k(&xeQb3
zEQ+6NDH0XOAj=&{=8<c0CUAP7xpkDR&1iS$F;&1>N=Kq}EqsprSxkh1^Zz_S2hkQW
zCn<c<-i1KKZ7`H7V)}I8X(QMf?%Sbqdx`1Y&s*yB2<H)5kREo>#nI|w&N*ehbQJ0O
z$2Pv>E8_RLihZ~FNW+BL^=i&Ib?v{LJeHc<DH{mU)$mmq3$^d6^@I1x2YE={*|YTr
z9)*q`h_Lxgt4AeS>;-2nT_m%Go-iC;S~`%h9-#^a+-eP$f9oe?TosyaMcp4W+x9uY
zS=E@xfcUgOMhkOv@s4k1FA1*;r)SH%$m<AlhR<9%y<V<jeJaKtROCrjBlXszW3&^h
z&$|YCEm7+V1*kF^O64DUCk}E_MLs|LAEL{tKoCH-FbC6>_SDybTroWnst9m2Y)hW#
z1zyLQ6HDp6gsCK3-#tcqm3H)VKs=4556mSe0~eV`G<p%0CF>*AueuUK^q~gK7yo&E
zC#lSL!O(4iim_#+%8P2~uGuNuzs(mhR9=D=av%H02HEgAFY_PPz$s@=%KXS^$$Xr}
z&lTlSeFXDt{@v*ZY9EddBcOilV!aCI&_`FtcO4ayUD_76QmMj9g;f{oW!D`c>`l^k
zG<knkQXFtv*(|8kXZo?b@WwjbEX+agmec!R(ZTL<4E&tSU%-7Nxs884gq9_zw(gKA
zk%ECOsfyN{ELtA{{v)x6qwn^<yDp_udOQM1hkrZkWlBFRj%zAcU^{dDyEqiJ=Bu&R
z=pz-_FsI$lJBS%=1!V8-JKKh#<x-1l+%FJud(igfk2Ii~V@>f?RtxNd=X7ij+nX8t
z2IKv2rleXWBk${xy(=f=i05~SZZKL8c$Zq(hfS=gGF4FW9WFTOot{FvsNFFRD#4qM
z&};O~f58g<2x<WYb&rFa;$uA@Z*g+)Q6S9Zf{wfbWfw!Cet)i}sGbsAiCU``o*@J6
zy`E#zul}g5xovw5Un~0T?Ob<H*UkG^>Q#autU%DPEt=^KTQsiI<t;^+s%4Hx8}iP9
zU6&pUZDI`&pEXJ+P(iq&<3M~9wIMVI%~CZAYmiL#;sJukINDnGvZt9M+YZ{=TcW;?
zd*98N4rm#W=LjS6M|7re7TKEShYV8t3~AQ=?gv7|9HPP7U>2287JLHpLWI41)2IPQ
zva41?CSI)u_ORpuSE33s6YoBZgtY(b=+HYe(YNF*J2dO;#r#8i2m)TI3*L`lCvn}y
zf_OgI`**H|?%8zB*|CL2N|7Ft@ZqrxnH{mZ8SO3c;!5Xuzv44gzh<L%pM)~#Tduy+
zkOsK3GWFn)S+y?GJ$CDcC?g7HOEJ(2>49qw(iY7QgrMYH=DUS_N4_`hC(0raDQe_*
z5b%`i#SC^zA(+u+w<|Z8S@~7<h4_a_>W-5%)v~4>g`0nCrGr?O0JL=(-I>o3QFLMO
zEW`Du&zS<oPRVDE8@yx*#OfJdD}m96-8|W}gK>m8g7@FD9!^A(#NliWrDn?zHhrgm
zI9=95-g`VX{fl_;%nIm+iGNbLq>R2Nl(OWOsN7<JXUKaO8AiM!4YzZinN+>6BX2lt
zRIfYPuu_bj)Hzq+;6_M3w;|vQc2ANa6<^KlVpB+)tq2SJ*q}9FF$VhYx-v6ZLnD$`
zfdm=^#DIfL#eeJq9V%?siG18OTsGIhYB(`JG6^+k-S?4|M-QyQx@rKvMwv0TDy*u?
zko&o8AhjyO+q*XxMqKv!(|)Y5<*e%j=SnuCT*3UMHFYFFhK{<UrRiNr=0I73q=J7q
zsC$q8{#ojl2DQxl0ulWprqZZ-7cWI^DPv85`1m~VI(F)z_lwO=0*dQ`!7?u*8Vvmo
zftFRrgC^dl+NJ0g0Wag+_^?v!i>KsK{K+-l*<J3=QM*&Qn$MM6>r3lKI|fM-0#SCx
zEJ$W}z>!anf&BihHos7wm3V8&kDnL)VLrP5?0H90f$-;MT6U_S)Y>f@@v<3z@%_ua
z*Lxa<RqvFs)fH$8eBe@<P?`2oOmiRKZ~oy)h+VS&Ki9~d>1OePm~LGge<#VQ@oOkC
zZ~?C(SNIgl6=m`99K(k6=SDJ&B9tM@s<OaF0JM5Y0&ux)v0(?{9vn3<R4P;KXRnto
znEb@QaW;#f(v?G0I5f+kPzD_}>Z_+=sw2o~PG}_Tc^4gSS0I`uPdhjId0~15`qF&D
zC_K>2#3;?+OnF20@<m4CRG>FJ$^CDiY4VKe<V4iTw+jwk48UMg$&-sq9;&c!f7}ga
zyP8zVmgAZ6m-Q`;N#0Pclxu|RsY07hI9-mybhq4qqNiGl>VkvpP>XJud!IK6wIm<i
z_<o?hsE;U@a&Nhtq-!0DjTZqOy~kx#J`imKdL%Q)U9@h>`Q&9DQ1xL&p7TSW+ooZ<
zC#uE=F9ww5F|*_Fx%97YqqT)22KF@FyK4t=F067!W^#DDGHMxAJF)U8yUZm#Fs2oD
ze`rgxG=4w6vb}}342a0>jbV;>j|uP1cn9W8>U1}S#1TXX%AXGI2CeVT4~T!hUaaN=
zk0H5=R<nrPY~wS@Q(nCSQY<y!`S&qkG2+$IHaV{7Z^KQjMT}T88Y?9caj`>N7>|4q
zsVG<1*uGY0X+i%ix@pe@20a55bOg<Qb8@WqHaitxj$%s<=@|fjcsKhXCOqJwLnQ)q
zsu{Gx?W*CEIR4?ci3=|UNN{US7zK4(OldElTaHClg9XM|X+V>5=&9$gf!zTT1cc`S
zoo8x(N*0FV^jpy-DuZ%gJs*a+Ej}wSr++%LYj#((N+V^d&VZ)m5fJFGD;OHMDgIUr
zt1^H2v<C=F+Hgy>s0z{!#25BOttQYE!FsSRxVAWI>Ix2ctzIE{P)81em{spJCsho8
zrBP7kuSPN4;5~<qqVItF7UM1O{$x0gI;0oOGdMKt(aqFY>egJ+#p?05E?=9Ho+K4!
z1bb1Ps{*zX`dgYl-g|-wYi+<`14t!9gk=?g%Ec35Ks{Q=MY*%C^9{q`J^x(tc5+ju
zHy1F|_Qp)o+-O~5-+V2$Ilrt#Gbx&D0QUB*f1|l?`b2tQ@&75#)>7~aL=4Q2md`5H
z*N_c@)ddrAHZe6Mm#_wcj=yN0kOzz8yqp1V6e90SU7|S{#qYfGS(~Z}&9G}_{vuL8
z7Ojj9(8mGDk=*T2yUCzmDOcL<#vOB36v4qtm4k;;j1q&@_AWEHtFBEHanq%JUlb`(
z9%s>=?^6wUp(x~*C{PBKCff0!`3=$lTy+Nz8AfM96&`m82Y$*wyx#)KMl7L(?0fH*
zujW{U=s)aS>i)9dp;LQdZyM|jG-;#a?`f|AqH#5JNkh@CT`)xMm==$JVB1h5Q3Y5_
z=AG2XK?t|Sl02*Kb0;eVE4Wgy)#hNfHO&ZnWdO~^B=w)EXu)N>taFTWIStf;sHL&9
zin<2;c1AlGub#U27v}b3;+e^N*S|!zoUbI7cLcD>vp(Y02I)CRocUjywp}`_7gy#^
zp3X&JdS08+ilH0D#uq@Jw69yE--welk%deCVxet9uP#dRm~w=b!vebPvNFh5JTQN<
zzt<M_`@TajQPq-b4M-d7w9ro}Ho(4_QX%hy@aMfxp^6q6Y>^NgUnXg>l@xx%9jWd0
zjxmUTnA&6#jBtDdW=^BuuQ<GwXd(bv+J5incEGkE!@HCiMO_YZ$qL3nD;^J$d5sKE
zwz=@;v2uFj!*otmdfJcpZX^G<+HQ>^r~KI-W~g>78wp5lkbrpFCEw%7X^IJCKS{TZ
z9lMs88eZpq{fiit?aEv|ZV9lvm##Ap4o<{9emVf+>CJX48xG#pgx@qH(4s)>(?ZXo
z-t84eBv_Ax4d^zuW3bn~Kz}Sv>UAJ^fXl9$INml8`GkP+J;LRbUmk+>M<z~njXLT7
zHtXnM^|=6+c8H5Y0h=8HT*yK!8d~Rtd-{e-1c?idN-YDl#CBnM9{JCje_?hnh43R5
zHn`w1C7xK9H7sR4YF{Z0Y@FK!z_vH4C2@HxywKdf`>ggg91OzlWL&+qKu`9A(D~Tx
zNExrev0yiA#N^JO=XkFi_z0~UM@tHHI?2o2cy#$8rsu2Tap7@soe>JlCD*^2k|od}
z{&;1Q3e>A*Y!3d{U}Ab=1*fTDNg%PJmyO$uE1t!rAVte_8HkuU#rZ{W1tT8Y?r+MN
zHppD)SMnEx-EYAyjqtk>K}nP~xCzQ;yD)xML6m<y3_v^#hw{nH02@A;C;1t!@Jq7c
z97*cK87>`mF9M<V6#Z#inz$)_M})0F90e1Yuaq%4Uhxr3e5UNZww``{zYQft`?Sf!
z3vt2-E$QTAn;@p|DLs4h_;j&_ca?P*f3<au>?#ky{2V6JIEvHEKOO&KvKmo=ocvgi
zh}K~st~0?v7Jfd}T&9SfY|pfYaS@`}=KKh!A^X?ivHv<!Z2w_>w?=xO_WCZ7AAM&x
z7d@ETG8^rm8(uVQT;|{<{E9RI!_*VXI4zyoRAj9*!0FwqGT2xil23}dGJ7qSkbLxW
z7h6d!Lx3<PyU1+YNUC}|kPDLL5AeG2`7ipGhbO}s1r7t}ivQW|U9(qawY@U4uS<3=
zsreUyi_;{`c;av-n@s71LrQYP1GP=T3yQ%oeE7Jr>7eh_c))qcm!cxZVN7*CS;{AC
zX2yOlY$n3bqIl*y+PeQk=bI27#|?u_|F<*!_eD~tqPrkavkAjf(+N5JwqFV!6qE4~
z6&zcJ!s(;(Kk_<)*=ix8Ku)qk>fK9vKwOmc1z}Y)FFi_7DXg7eWN_FQG;d5YslZ>*
zgG~&o#ejZ1%-<lO^5~eGADyQdN5V7EjIABFpB}K?Ia}uLnIU9V20ccvwsOlv1|`rI
z)<Q=sJ`sN3`cRc`b=(L*syN@|v}MK`7IkRP!Rb`HIq4twd1}-x$_#<SuA37&?~T%J
zPi-b;Sbd|94EP<;ZxpiYEC7aJ{sTvx2cApSEkT&{8iD63gPe?;#BEOLuJnt=aGH<2
zI!AAiCCaEssE$y(H6gf22e00`Mxi0C($gKnUzk1%>T^+@oX+H+))bcQL)XzX0bRUL
z%-GY92XL52beEy;E!F`9JQy$|!>2@yiq@V)@J6qv(dTPn#6>!IJhQ|^x&2|jce>+P
zEzND+DWo!Y<Ga`Sk|QN1W9Sy9;jkmkLRT;pvQk8#`dtRx3x_(%ZvN<2-Pgw5`(C(p
zsx|!01i}hAAr3}xdt;P+Ex0daZ>(Y7nXC|(S{8V{57#MVwa&Sc<_WBc*ge8H@(6)O
zRao0j9)%eE>1wCW)Oth;3A0jjEXD@QcESLY+F3G(qleRaj@Dkj`-cSc`uvaL&m{)~
z>5z@0j#a6PAFv)}C<}*cNbz8CVX&IxgCgKgj}7u`pJdRfwyx}*YymDN0y96tzwEe8
z&n!Ei5*R1)R6(c|JShUbpI6vmpkTG~r{GkY?iplmdXJ=Nz{R4UaaVK3ny|Haprw9-
zD-|8{{)9l_sP-pm3ka%A`yn9j<$Pr0^pP~xGT|&tqFzjgN|7qJ8cA5D>t9iZ3}IRK
z4cAT~u93`qs&(Vu7n)VCheZU`WKrnlK~d@Hyuv%p1NSk>Ulr!Ek-#mR{Vd$;LH$}g
zPvNe1<jONKmMsqo#IwldxuKVZL{4z&jt3VDJVxA|FWB9>+D#`oTu!JAl=&>}G7j8>
zG6U2z<d49P7mLo;l~aI0NyW+0UQcd%6PX_Rc&(gLs(&15K+;8KayQJ?CK(psyAz-U
z@NpMfbB!y#?lqGHCrd)^y|-UfvcHU}_7K1imM@ui^x**pSAjy#|LBDy7Mud|)lI$V
z>6RSalsmVDK{|2F?f;wqM^fV?vObvell>O~&nIULcJq3!rZc{37aVK5Kmz%RLH(75
z%QN3UY@a&o<eKI@V-2PEm@!l=O1|J#o*s?2yxWct*{T*nbnhId?AOUV(}ue4RI2n}
zWq>`l!8pyq`&EjY+>G>rn`gqDqE6f}OfLh)(v*jXp&y_PK7Mt&LwMD&>YgQ{;uJ`H
zbf?bf0b87WR@L@K{_JpU2Qki9T&6mE{W+7&2JD`sDkveRktZWCP__Y7P_K#A$FdNk
zZ=BJvMYl^~HK$a8zCW1qNFetcVh5X-_22P7`)3I3*Pu!ztoXEo5Q@CFt~|j@_?xvH
zNy`Xl3E*m1C-RssQXa-Dua-xk^Ar7eH4`56MgwbyqQWAa9vIU|6GRi>19V)6)HBqI
zj}3+}NDZsX87CF<pd099(zCRaZvSDFAO_*n`lW2;Nf=)Oi-7A_W&#=k=KtP9r(yT2
z*xG$UvkeGkChImUGMZDcSzq{L91uxxOZHXdWUYiMV!xmov>*!mObjCq-q!Hv&PhRx
zHn;wZhhA9*0OtJjxc-kGPoJ;JK%*MH--)g<3Ga?L@mRQ!8i^l9in2IWB4#bthqx)>
zh0L*&-Xp(1v8fZ^IiBF7*2_`v4rzM761tjMn!&5wgp-bQwIKTd5wa*b*GVj5!i|V&
zxi>V#&55l@$8UUrSio9oiLj>dwG~Lb@8!s?<jA=KKOpz0y-NNy7H<^0s4L&d5r8cb
zHvM?voI0$$%z;kY(^Tx6i&_AT2f+O?JFa)Mk+K9Ms+*7G`9?_Dyv;u!f9R=LA9fjm
z!sD&6sy4*0<gilvFuFf%UNhuse_Y;Cea|5d^_a`>>eZG8#fkAPqD|{9zs@`PxI)h0
zPRc4eKQZ2GRgU&{swC$ux3zgJO&|kd4*so;$7|Mch2OZwIq7Ee&9Zk{uVly<DNgL1
zTQau7Joi*6=KTrL`c3&C@)_((#Mi+wJ{BP=g&D0CDf}#K<P7|fcvFs;(nGBapqKr(
z9RqWjRtmEb)q_4OZ|QE+uXn)86Jg>l)`(*->{`Xqr#d_vl0{?1W+d?Ahf}-JX_33=
zXHr(m?IiR+MT>mr_^G~I$lp$MR&hTfs+l>nB{<EiofS~!$tV4221FjU!;`2a6##Vd
z#aH6Ur7`W3VPUuN1`FkqIlSSUPEW@OH;$+yNKp_42Z8r=r)>n;l<*_)zjOV1hKV^;
z=+Oc4@}AJPrqBXMP&v^GUa|pm1~uP5?U31UGV$)kdY+|<uN_~;w*DK}UGfiw%{f+o
zwTlBFMl=HJ;rY<?(0cUs3lj+23p}3>YcJxh@EeR-Gmf_RDloAJ$l<aOlVii!5Uzj6
zAqK$zxb3sP3N8}EOM|g1<#LBQD|&ysAw`a8xRX{iF_i9hFlqXatK#oEcB+C6Ki`(!
z)}Kj0idc<(#+S*)s`ItWY!yQyR|5-!yvj`jm6+k4Pofd8Do6|3LO<v}&;tCKlc^}{
z)29E6gWPv4#e>#OUI+Z{z+3eZn^o07|M8*lvhIyT$13K!7>j{~z8wv1-Fxt(7xr-M
z4&|AC^Igg<$`D#^^a;Jw@-BQp6KT#C%h*U+!A)6Gc40xWyurLuHWn6WKt`sLa`EK!
zFErY;_bzU!mN;4+OEMs>K3%_qOsteiKDle52WUnGxo*$7@aDWLN82aDZlR=~hF}e}
zJ2?kQnMz4Z6ABkty3pwa1e3QA_y_v(UsTk{zsver@JX7!TvkAm8V{7=8fn<wUo7pE
z4jwMTNaTvhg+lavFbrAz5C$F+1683)Jbaa;a^bHV#WA+Vm=q-kul<hjwBS%f)8b}L
z=!6ai$<RwxfS?{X*}T23ZD@`Zz@G8ke$NVuX)d*vRGkvuR*mAX<y*k!1Zyw>;Df${
zh!3wYFrW#uG5k$+0XIA&_-Um+c6UYHwUt+`;N?L3_2Lh;m1bHH4hZ~ZNS6uY^Yse|
zsUR!<r!0Ws_68a&ulU=+j1nWSN*Q;12w_Asq0CX-Ja3|^uk@t5bF5br;!m5ND4>2-
zM-X@av}0SbcO2g<*Od<yACzkCt8>FRXRe%2>xbF!y_0d5+%-qbw1_%!q$!=u#nRHe
zqTu==*SkC6cC|75MB|dIT{FH?p=~>4yJhM+#H36DElBZ|Sqv#V0bnAc6O!%M+nk_9
zb(Iy$@rA7Z`@}_Qw9Q+4CKSJeQ-SvT=friDS>;QWX9CtJx)?Icw{V>=z_bVMf;OXk
zz<fe5hZ=HN{g5nFH2ie+%;>ZCP|n`1h5m*dpsdt)^QbLUYALD$X3L~4D%3EbN-{|M
zN{-*NFPCxj{+-u8+T~ce{jlr{v6fYS!DLL@3ADjQV#baD82nJ?SPPwil^Qwdb+#43
zmXukOoSys>-nO)(MJjXfnjgtZm_%GdC&<;E!WTM(#3=6M5$9SW7I0<Lp)NgDo5h?W
zP1>JFf{~^BG(}uev+ph%W7Vcfh7td!iWvO65*>Wc(rI9$K>Zl4osZfRJL0o0@pZb#
zge+scXhGxbZ70zG+_dRb+g*`Te;ct8I(*$Z-a*`gLzA*P%-qSAvS|cEI-Pp&$-eFC
zVj3}+a^Fv2_Q6QEIX8@eb(u`I)hM#b6MYn69B*d#q!YUE0~8~}Cv~H!GgupG-m$yi
zm1-eDl+S$WYIEmAMZTp)gjX1(?&!u>C1?$C#pI`)153P*Vvn7C<z1ZI*(7^R0RBdZ
zoNh~_!`np-Yl4FK3rTEp3Ng{AIfs@lVkN>n;@ipaZn|?d+HNM1X{c+mEAua4x>CLK
zt1D=1-x-+vCj3pO*FVki`WjNYd;y#}DXI#5qTmY2E`k+epwaWi->Ib43M)rc%Ue9$
zjAG15QB@}5c}}c}d~r}7k=Wh({Maqn-gs7CIWq6fo{Pt{G?J<h!jmDS;Y~u^MMR!=
zu{nwR^Lk_ET{Q^Mu)5`&*Y=s@hs-Rs%r!=`tI>saS?DRV+}<WIPXN(Ci2K4N?HBJ?
z;M@PqK@!XbaPOb2N#V6c7Yra#Mm^sVh*l|5wQAzpe`kLtvA11x(uH-wG%p?C;v1MO
zlPoBGGefs%^co2&qHVGckL2BN(j#+ZH+0aTz-D}fL4^U>YyZs$c&2+>`W2D&C;0H5
z3cbC+`zt<P1_-;bB?)}>A@=w1-(SnSsQd*leNIKs@b#F{4TTBUr)<^03!<wKn_(oM
ziy$pJu%r<7TRxOCCIuPznvi#V9j(yKzlmd4zDJV&amZX48*us1Ayo9#Lb{UIp|mWi
z%-#>m4(!{1wXDo<C*|X1`zT6hhIkB;{Ip_->Cb@6>p^lgIfv7iqNv|FUw2nVolzx1
zT3X8GTs064Aw0ND9}#@<yK-OWtQx(4LN$us@qLf*s2ZeUsR~=K%Rj>5(E`K6W`3_a
zxM^-39{f)+AxsAex^bn*!?#$#dzl>-)cFBYHg_rtYe?1?53Wql<j=cp*uXq4C4}b5
zF7iB^*Z0=h0uLs+o@ye$<1_3ag3*?ANfo~?LDrgAtr&WgE#JbBT?E@XbcF5VfWi&J
zc0)eoQKXgSX#Iwy5C8oUwgksYiqz<+bmyZ6S%A?wp3l)r?9-r!*Rdi-xL5QY6s{z!
zmAw6|BX<0#TEj7XnH_kuuJ<{toG-pSUu{hhhmkLhaNyBl<^sZK!^Ao#EGu@La);?9
zQ23gA!pm?+N#G=R;zVcixbR383=zcJxtN8o(lMsW&@W=I3Rk;mLYj!vc45|~D45OV
zK>I_o8F>-IL+s833P#;Yw-ISJ7Fe@Rc3Jzon;2Lm9BA)UX*H{E&_-CS*l~$v>e}s`
zUz$iw!hZD7r|&A!-1pb#DZxqsW}zq$=h;V73TN{;1vPe5A9tJC%m>~8qvA!pn15~)
zFJ;_|(3?T5AM4zJr3f^ep?Z$WzyR5tFIG8cUMps%or@y+Z?oP_uZpT*+=-VHuJtgF
zQ4|xvkvHWORP9rpJee}7en3Ph5SNs7{UlJtW^nKA8@FZns1!TtUe~*qK6j1T2tkp4
z;y8sKlPq><`9QkY<vW`U+OZ~`6`l*`y|9BIE?@9nPZW>h2P(pOlk{$Nb@VpgZZ}k1
zm#Tt?j(?65;rg#J_9_}IeLTmWEAQgD+JrQuG1>fQ>Faf6pPCFrZR)(_h|K?u1Ul+@
zT3uvC>RpR-@904SXZ^J=N@6ycgFUrR3Xwr{{{4-2?B|-pkqd9n99*TF^O!g^YiWCb
zT|P|817qcM^;QDDU&Ngbq0lg)fs5*J!S-2l#si-#t>T*-v63`NBU-|+@1bC2{A+V#
zbGv<^P~u2^hx-I%u`#XvO{Iu>ztr{Q(=fy6R1=dhI;rX12H)J-q}g8>glNhNv_7b3
zj!<Y>7~xqYogS%_qo^ph`0{S@1vglAu}p4XIDK|B%J<dgD!xjK9_vU%Tcbq32q)O^
zxqBT*|C_gOl08-`m4x|nnwlr+9%PmOqB}c(V*BbPS}h(UaA*~WDTs~H*rm0Mxq=}M
zbRN^k*xcEk7mzO+W<c3G>BA0qd8!wHw>jOxfE3JtAy_Stwzzi*mQ)-foL-LCIu9cZ
z3k=m8>-KQhg$Yrr`h~0QE*i4nK3Ewsrf1n>m2AfjPMZlVMwVj<Hc<ak2@=c;u)Z3b
zE8ohor^fHp-|JGm)Ot&+NZGT(_gHGF_<x|uA1!A`L>Fcu%M9>Tce4mGoS;XfT+V5w
z+tRuU5AG&n)V>yV;Wk>fUYs9WX$gn!GI?uqP7lSmv1b{4r_lnTJYRPWm<SZg2qzXL
zsXUOKAKyT2zerz<*>;;k>J}%vtuHAetitAX==Yrp-6tgtTBmvvn=LNze;hp%G?je*
z42JQ;cqSDzEF%7H8~*m2-Prj8MKSS96A(;GZ{gT+n|O&jgomdLqbm&m-3Yh!72wQn
z3_xowT5GXol$bXb9gqY24k=HI<tu?c=ztCnjR1K$y-s!p<2{9M2AzOAulxJMt=Wgs
zLX=pRNAqbl3wWx(rNVW_zw-hIt=~OX+|sSF+p3YH`QH3uLK*D7_(LbFnV+&$uIUt6
znyj}t$9xLX59AdkKg*`|c%+HYjNYOs1wmTf_`(g*hb1L*lw#f8?b*#*em%`?uvv4`
zX)H5-#SX6Z^fzEdU8E;PTA7r$3*el%2WXvXUBZ4?rdJClI<L%gfom&aveN$U<#In5
z)^CW?H?grSw&qDWSwEGJP{CEV8%)_54Q<Wq!F3t^V>MzPJD1N8scY?HQ2ewYd+@#2
zYJd|lpU`v?wX|?X^ZBAFtlGFdiT3I5>n6SrxQ5KyS`<Jz8B9Jz^uSKgM+j@B4_n(q
z`-0j7egt~@GDLdS+ELL<)tFYj6DrnZ?z(cd$hm^nra%sgTf)hrqcZrjCOr3b;6X?A
zq);m(Xks_l8O!P;K*Cxfi!Ag**#&0H<%xh1@u8furf;Pkp0ri-=0gFnW3xo8FnNQH
zHvwg!0%6YM22L1qQ~NqF<4Gg29VGq96O1&$Lr}1Upy&5BsAYT1ds98M%l9OPrjC@A
zZ=To1(?D%SqG~Ih1K+IbB^i|Ba^1(R>AyuF!RHSLL{B-Jz##&4>P94l8@YeDmYIcr
z#XnQ&_nU`m=j3VTFUDM|XIEvRzAHgPOu`C<$<V3_v7^(NTfi+6ZYKYCRPio(B^NCd
zc1v*zpCFF32R3bv0df?{1h<XU&Jz8r;ipjkvZt?m<Tc`@T65@9Yyn3Fv0NB%{b#xi
z%=&<^Suei_R!UM9z>y6wp+4lR1%I)4y#avvtDQhFn&W%-!mzKD1S07(N;2yqcKy}t
zxVe%5*2f@yVat4T61m0!Hav?~4JSxci6vfp)EO-Ei|YoKp>e?c8(3IqPw7xJUfy>a
zb(J>5CV0IYvW2o!?+->V1OEYCw^j?6U|enjo?W~fm#+wc*1!riFPMnZDWtwlyR$8$
zbsjt7G|roY&B_9-7B;7YawzAV3HQkCG+1r8!|=+|zp6v|NFESoXlboUKno)3H(^s#
zGob{IZFZB0fJ`OMK>eMy1&^}qJSB8=2IywwfSS(-)kT^Bhc<38ZM5IctVyP`1Pt*=
zt7T7|!Js}KC09wyB}`4dM38vfg6rsB6V+kbzk?!buj1>&x_EbyFDlsLMko?<#johp
z434IkobriwyN;aZNGda6)TL@c(_uSG8@RpcHehaA#g(%4?cW@W#ZsvnXp;>W6LN#1
zf^wv^os7IJ-M+3vQ*Oy+5*2Q%Z~*66N4%MtmPu$!2R|<8<fx;s6W?sm_n*4Qae}Ue
z;G%t5JOqyjV7fi0NrhC1ru)j_cNkCbjlXMA)jJ9#+CG-*&{qE7nV*q?y4UUD8$*k@
z>iDJ;G4fjA-JpB)UtIEm1-nqGT)%>y$^(tlWJcO5q4Hd3<OM2orj!iyMOT}TbUB>N
z-vM^CUKb|oB9kX){y!Shw8_(RAQ6U?BRFRHfB>A`{3zvQ0%p}XqYg$OFWE~uPanU5
zrsbaaR~vf1Ja&z;?n3FWdH{jKsE)s(3G546(2~U28JJ5rWmSr%PpSyqa~Hv$=*pPo
z=AZYSTCrhs$l>ol1T%0U=p2K@iUOem1#ms#B7(UgagPrz3s4KHE+CS*AR=_<<j4H(
z<&@t@AC=fiqqOFjD7VJz%r9K>`Q&~jNEk~r|4OwfFwtZfMl^YGmptA*O=34J;EU|^
z8Qvqn5%d>i7S{d6DTQdGd2y)h-s?qURWvH0U`t=hqawlFt<A&<pS>HAzmTea`gS)-
zF~%+YF5TiJ<p{;1)+R+?UdP^b1s2Z7lQo(lbKJN^_+64qQ@=2*C~DD{@ED(w(wrvB
z?AD2}(Vm+fg)XA0J*t);d#p#3h*sdi_f&%9-hFcDb;?eKd6bzt@z|T$yIsac%(MlM
zeRphY%pc{LyfTkQy_V~c<q8|kI~R936aK6X({Gg2oWYi?qxnj<yt~-&r0OC)KiJYP
z3VG*hHNf&)miu0gVd4K$j+QJjs;-iTGP>Lokaq_p(|toC{?hNtD*uHdDJ(BfAq#?^
zZS`9Po7{%pua=(oqGYIx=ONeg^J=Whn=9#!VQdm$Mgjx#P6T>cO9xOS-*Vd!j#ZD4
z^dK7IhbmlrUa@so-O?y+dZFO|dWEe+!L2mtcJ%WWm^;Gom&|@Q3CKgc*8HnIP$43a
z)5l-8yMjM^ZY$_3TC6Dkv+k*wIFyRo9QB0&(kqEnHfSQ7TH!;O^!0o8lsMhC;~KOj
zBCKDHSU9qj2llT$f^qR0miv@`3YoNz^mQ`2wX;UV4*%&up@2Cmow`?rhg;tqSx#PQ
zNEj9Jw0AHbPql9>?s3l}_<%*ckUw96=Jr#~XwE;b%sJ;o5R3#y>Bc=`z)sw0<+!H6
z^tL-qL<f*YMU;5=fi}@;(pM;l8*;?nCiExo`A}<aG=M7cc<gelY=58dmqvUI1luaD
zfq@h8Qxr-1-&+uEqmEhHXpSxcuCO8kI;d|2N#uK2x1}o0OJ}Q@<3OR8MqeQJcYNT<
zHW*>TnGd&PMD6%%3eOlAb~NKaat8h*UyLWzjN_79%UgC8_ZwGgqyWgqC9xfz31CyD
z93SLLY1kR;<?K$xJv{g%LY8-55-N{ZsiPJ94*_1oZ=<vD>LV~32uOJ+QbI5xo{JgB
z59GfO;v1Q;oj!<1%a#Y+R}!{OO|AUp*05f*<zp{te=M|OL%8jQI2cru<*PX0o!1a6
zr#_G+jzThAUvTg&5|^%AN~_f090W9id+KMVMQA50wQtF{fh*uMzd%kVPa#|7uOorN
z+G-rtl2%!`+4KX`mjXhrqwrQg)`n`PKoJn_&@a*4>LYvIMG=~By$kJz_I`4Ydpi5?
zF3WDG(MG-MA+Mg84sk5a&fU+jEt&_&V-B>y?nL-bm8o)2;6X!sNE6cxdMMbwg&>p+
zhub-|tF?_h(Tkw>E?hfvK<_G&!vUlr<emys=+uy~kGyOUK9X^^P)U(q=Gyx$(63YP
zn=D#o-_)ZMv64;lNbeAg!VgdXS>rND3lC8-?(mnHI`L*}2Vbll7jkfDDbSm$;C_PY
zSBkW6a3)UXoY<PbI6q(P6-)Vt#r6J!rA>I585&&)ge;Ea)`4E5LP)h!Bxg`u9-Zg8
za*dUSkrxC!=*w0+L-3+Sx{@jVLQ$=C1M!5!acL%}A^PED=`F+DPGwSaiWk%`p{%Mb
zqmTJ~J8NG$uPIIEzc$WB%7)Ub63h%h=Qmfjxsd);N;#U32dV9s#_|R3w(~MJi4^13
z+U2IW|2RsaU~|V%1>REQk%D><qGc^7Na$yzEms>rC%N!1j`oI48cdwSxaItxrbQGB
z7|sMszpGY&l!9lFzxA^7%FAPr3JT#^dtCxS<-!xO2jYc!JQWY9dL9+c1Zthn)hgb$
z(wg7X11sLKz)?-7L)(a7N9|n-phM#~_g|m`ZqR4(qsM57Cg`jCUHDSxt?IxOoQPy~
z*mZ4%0jk5n$X@kZFF{$=3?oXG<9P!&Wl;-5S_Maa>aYE;CT+IHc^X0%&2U@E!K<6Y
zeP|HPMej>>VDC1y2bvm4%xe40c_<w4Ua#lZMiPlhnEa+sGR>RBBxGyOu9jp*0hQE@
zIwwDoqbbNFfpMt~t8)$PFH3OcRsR(`>&EPlF_7}`<i@7YmG3z+Y!Zd^rtNod7(wZb
zS`W0cyDqv(FdD#giY@t4X0T7tV00Fb-;p4j(`z}N79p9%HRn|v+CQwfEsi3@nB}lG
zBd)q)yj~|q=hMfYVAb5*y8rGvL>{=cG$?YYqxkwU)SJWN2D>0vh1!EX5aaO~LAazE
z^`Lf%g2RR3m~<&O4vh=VU(ZZcdEAwzZoHe1KzyAi90lALw3>{WBtu=uiu}~=<rvik
ze$RRuG*--7gyKf|z?1E71yiD<P8w0bp}J3m&btaG2nSm8>eT<-1zUJiSwKZQn8}?!
zI<$)Do@JwatOAx#i=2=6mo-G=n1lj7jkq+yN8Y4PkPO?bL|}|ro(<taV>heM%6cM|
zC+%z)(Kp{bJ&B%${9cMzs+1JY%LltY3UNnu(6n}bXpCUD<Z4ttQ0XGbfX9WXq*|B1
zvlACpdKIe{=t)8=<*XXpZ>}#HyW+rcs_p}qW$d`~>aWO<?whU^KlDyrCiHnOu|kGU
z%9;oqCQJvNKl>rDQNl_)HR6FdM-a_^kER4kEKKmqwGI{Ce*UcQeSvGk6u7fTC<x74
zw<DzZf$=pOoPJ)$DDsWp_(u<DS5qMA^wGl^!JKX38C0?lXZLgq3bZ?4@a0hK!Z>)g
zaK~Jo^;>E<2h2(bo@#@>zM;gZZvgvnnzpdKuF447ZA<yw4PYg1vsiQ=n!Teo#Mqta
zAcHcxeF6JQ*g?&pt3D1wa{gvo-ra$<i%?XasL;C=LIQP%XK}0EN4|6MpLXqf<_K%t
z4vXgdlwPnCdT8VZaJ%6Sq6R{e!H?R*S0!M~NTKGP0642IOi&q;RMnzw=IK31u=uSj
zp*Am#w)wHDJGA_}?_lL7L0y{g*LomorYWR6)8|0TSeKAp-~Md;&NM`Q6H@wi{v@Qs
zz29)cJgoS}BoXm&L>|ohxHJN1Q_5I=y_p}tV3cNWTRKd#ANxdOda2li<AhDUv<;25
z>zcv8WOY&1zM7w!nP5Ted}5PmnuX^(TScr*^h>@`&Ps7%h*P;IG4MVX)<#QJ3s%yE
zcJWZc7LqNCrEvFsV$x}<zQ*Kw!&(M$<!<h}Rj=X=UX#m0b}f60#vFHXpy851fz8vT
z1{Va%pn(ag&$R_-Zhd9*KZY$}jVIfd?VP?ihW#nV9__x?tKJ~A#+#5pQq$}LMJ5M_
ze#;^>>b7TSx#li&m0UTl+D#aMZs7-DmAke_7mV?_BcvhTNkwF@Anlj$DTNC7ugd~V
zxj3Iw_WV5ZK>cv*7a@eyy|~fTD?R7=1l?m5JWK{1DF$0==)@C&y6_pmyDBvoz4~9S
zs2WSv<1C3Z{S^eA7N#wEx<1FEJ4CZHw2Bwf9O^xoa2|Y6XHhndP?UWL_^R&h1t^vI
zL1dw7U311W-4GnfP3G_>$c0xjlGPRqnvO2L#4U1)bc~Lnreen7B}?fB$htPs#q(5p
z5j)yDfU0LkA$lr-w@KpiN%r)@qzdw1#DwFpi+jg5Byqzo0W4U=4E10F%Y_T7p`N?x
z%qSJfMXz<rWh=?YZVS3)?vi%&MpP4Q^=pv(s99o^{NJi(yn?GPImB;Z=R^2cOOjG3
zzLqfl!;Dh|J|IZYqY#u3V<$i#`Vlde#7mC8K#FmSEnhg<tGMB{Xr7P`LP#VcY%?}Z
z1X2UIIOG}dQM;wBNI+Nd<0=bpN-}t#mwcyVcOwroywm?sb`K4@C{cHS+qP|;wsG3F
zZQHhe+P00iZQHhO+sI8a$>5vhH`JtR)!xtY-@DPFTPJ2n8ZVrepa^{e0#PsQwe-%{
z4vIO~^^rHAl82}FmG0dU-L=L0jj)k+`dC29P)C>uzdA&?q0O^ZhETGAVftl`hUHGB
z2^`HQ>}6hkm9i45LZeJn_gBRb#6d5M7Rn3h5bX$DI&kA+FI3v|BnK0B&$`GZ=k7`S
z0Sx#oPTi;SguOpXe`<bjIyo@mCvHObxP$L_4CPG_6Q?YL{r%E^D}NJP#MREmIeLT<
zFXE~TrcBSbd%3C7TsblCy)gz9xj4MXc(1+7`MQjwZ1SXss`dlSLjm{hyw6Km^t59f
zQ90J=nqT4oXb6%2#@Gm#mm=aZi{tH|{~1@5S234ySHkWtu+sH4=8H6;SQGUSGkKd4
z%dtrCgT6KRhQndq{q_Q7VVRtQ_luSb1$#+_jZKnUA>3&_YB&c1JeDsH(l9#6xF(?U
zHlb8_y81(J7c9hhD7fCv63P~)&$JV9F$D-a#z{!H1hy0Zxc!k6Sj2QDASy>HQJ~kB
zy6bAc-ddA&kU&#=Pt(lMu!1|$DcI%4k={EM$m*!T^F4_~2`FynroM*$m%VECmjJ4E
zwr#}!739sY#-#E6L<TNIe>(TPiDfRj)C*(!P*$j-K~DJ2x~~`v<b)w#U|{&cL&5%M
zM>vnDTP0CWx}+!(e|O=BNinBoj~T-Hl{GUR7YD4Y1i*B7LynuRh~>njm3=QXFJK+I
zXxjR*(Lzrab`J&q7m8ZbmPQ#KNrn)Tn-gVf$)%YG_%MCJ+U>UHLN{6?=J-qdQvoEB
zja?-jf9Jw6IP!!9XPcqqM}ZE`cdfs7i?H0;%H?@<u6z68QX$pRZHAuFP4Rb^8WFe5
ze&(-TsQ6$byK+5c-+Dn(yquGY`F6M*Lj0*9Tg?X1H?JQ-z?6(~pwEL5dIqMe8p)m!
zK8`5n=WCdM@c5;^!=4m9vd0)vIoBeFu<F22+nS>aRXQdW#eAhv{Wdo8pKS@!vX7u;
zlO18RhK<?3LXbIOD7k9B+@TXkS#`%;ZM_iA5yUfe;4-y0`6@i(0w<yq4Bq_`D5(VX
zX&l>|G0AA;Zt~&M1+SzcoUfgwoQyazhX+pE$rm;XOoM&j1_7Fq5BN`EiGlnP97$-k
zKvjvUYsf{R|J=oSLhkix6dxaxw_ynTo6+HsjX;%kp>^(uRnG%czqZ!2H`=e=(g`@4
zUlwu7fbl3nPS>!4C`Uea>Q$|PX7#+UX;e*M_TD;dgNY=$JXxJA5&F5vY<_kh`4+Dc
zy)SXRCSnsMWW_K}p#pr@iT>(3jRxE<_rHnsAE`Q#&9#AYBgPCb!pWU|{|SZNxueGT
zQ?U1Seg02=JI4si!o#d33(f|F_FvFsJ)*NUcK2$dJrmiPwhS$;!?TTONLCj_HA<AP
ztl>Z<p+UbQS*y{@rA?eyL03MmW1<OVIgVP>(xmW0QL$>9ko~?Y=_%Oi9!L}KJX-Tx
zJQk`NXY0(L_M#@02J4Ny9#D5ndvskeAL9{$+%2eb8UJop%8xsWze4I;s7}s*>mp79
zi^M%JZEw~8sFUz>almx&u<hnuv_e&dcWLfyo{b5q{wdb*cwBmbNtGfK^>J7t-Cq`B
z2{J&uhS|ZA=h*Ov+9-i4?FJ#Bm&v;<UVT{lQlYNs6~zdes!t~NWgS`43N=Qsb&6_F
zpvDmhZvqcBb_(I8%KXS=iDkfpG(T}0B@h!Q#?wGiu2-^qwq52^U9l}=<5GULeN+@+
z{$y<RZ9d*yWVT0)6&8k60gD%nPFvvT`tRrd%nM38NA0^tM1rYV=2#9K4xy<|-9rDY
z>2L0jost?gXD-Gkp0jTR$CE=$Z$tI%JgAr?dFKRwreB^qLR=Uxn61P1^qV|3%P9rZ
z70x7F*5Nh*mZLi$L`_8Nh3~$b8d7E@K(i=@G7y~@*%ZF?DVrQ#*H^gHLLiS#fC;Pz
znd8p0fRwkdcPZ#y+jBV)B$`bE%&JQ5l8t1S75{D+HVEhWj4;=qa)F(a-^BM=7qnrV
zSyL?E?hi3@(qRlkKMIn{xibZ43@|y0IRhL<oC`dzl<JZQ@;Il`QpWRjP8jo0H!5!L
zkqWcf^i<4@(_7#MLOy8^By6ma@SIw?SNIBzI<})HJao-%{yqND4kk<|-E;iiv6@`u
zU+aT+(!$f!>Wf2izRKOYfX`kMIe@}r9k*^06ptd_S%pjd;P)Z}ltZQ<=tyssw@A+<
z8-Us|lI6G<6ZDebp4$fpdxgSz*B~Qpr9&+(F@A}X#!-SXc4p=PX5E}OL3fn*sP`?r
z!ra#zSl9}qO71Ify`!dW=nPU<h+jZMO4H`CGGDdgUni2$ZKJn&qgr;i_B#XNd_erl
zjT`P0q4TYVDF)sxE~@b1l&?8+$CDKy`U*#CfEB(dqEnH1>8H-Mv`F|M5`NHGz?O{Y
z<3@#^pjDQq$2<CIDmms8<;AKc3C2)MMa)hW>Xh#y!v5{?u^>{H|COD|wo;9muA`v-
z20IXRK6?sYk)uX7+EiTVKgh`y;$ykjCWG0EeXs95Y@BThufhs@D{S$UD7x(>6|cPa
zOZ^K?6nLKq+M@<+XSgt-<8GABI|#i+Au@J*V^SNI46(tEMXX@m)B0Y`T<zwqzG#PU
z;3eCa@fFRbCeVHlJf-)%hxjw$ur3i4C(g87)<F_7sr!bO-~C^_lK#j->}E_l4$qGX
zPqLkpCSq}ogos;0$6&r<NgIi8TVwcg03U>q%PzDTUhiBgHU@Q+zJ={<x*ZPIe*rAZ
zzkk$F<dT1G6UjVbbfb+OnZSnfikes%0!vGdLG2%Z1~jYp0)hMaF&(Ea)xfe78D0LG
z->EWL0&j~sYdMtpkP^b@T%Omgt=pbOsJzqHbVv1$?nRylpGsG+zjuR55pWlR$6o%_
zD24EfT^;mRDXWBkavio3dno17S2Q<wx_WM-Bn|(JG5NQbiq*Pbo&-4Rrnv8Kw~~@Q
z&!c+4W=kw#dIe96S~k>K2WWr&;@G4ydzHZJ(mu>uH%KWpK=hv^k&+xo?O4(Jn6=a<
z;Xob5A~yR*9RPCIT(x?pSp1&~bnu(rA+(*T5a`C78t;;S<qa`EvT6h1Ztz&Y8cl+B
zL(e-IRu5iG1jH<Ba+SpMX!Xsm@1{vP?qtd~$C0khHJQe@1f!#vL*Qczlyt)$8`^5+
zdZDmbfF%EHu7uY#%n<m}hTl4MgMjnP94Oo}TP1D-isdcB^udE-Q4%dJ0X+GB5|f+=
zRK40)q$cb~KS+!AMxRkR1v%s?G+TF7TAi}W=pD0QzrUP&nQdly5U9#OXJ8&*?tSgD
z%J50;1*`q&cS2^^Kl%bqeOROJ>iyUxka!EQQXse#AEyuHMC@E(ryd7ljjd#NLf?82
zyeiZyx)T$QG;?8aM!?f?J#$mu+w=(f3^Q`eqw@9}9;q~f@Y>K~*FVtD{u$n~^A7OQ
zMA9vaAp8Nqt%?($^?>ZZqhK1N1-ad02Du!{I~n2(C#92*iB&Hrq~5#__r474r4&B-
z;oqcKUs5%4twyXFy?(P2cDYg@i#)m-Xba&u$Gc7eR9T?$s1>8&%_BgRPHzZ?7}RB(
z3YOq8CQ^c6a7|Qr9nfWVQ>uEyQWB0DHGvtK(#g|n`&Sc&8oEjTXrQD6a?CUcTmf|v
z(KEEoqd*;CTN`Z=`74fHp+_6G*cdQp?6(7sPGEokOX_X@_KUL_eQh(!G}y)r%er)J
zCKA8Q>=1-V`@wpj%P>$dS1Sf;>x8aLLLW&O`S(D+18RFZ*WhquiMsy#;T$TmcJ<-g
z^rPKK-lgt)N0>s3K%%^&eQ@E`8&!zHOk3!%`P-yi!9^{nkii7uba&yD42PTI2-RK8
z{^^RC?&ryMyiOU-eW2`+{?5W}P}k^TCKnpcN0A!9I8Z+@!2d#_T<rE4<>CCjebnAA
zRsnmur8zwQG?t1Owl-2bmb|Av?{8!An>!wbXN&Tm?Vi9*%M@z)AWTksD&Q{2um;XJ
zM^iBAi$<A{Y)jnr7eV9`{vlsFO!pR5tKojxLaT0m9B=+kv&a0bpyJ4W{^Y;iO#czW
zhz%*Yz)TnOV2_6rROn3CF)h<F&)Q&HopXa^ei>y=Xc2U_=!%X{+G3^gpfudF@_-hB
zmsfdOH05g|8N#h5dTiU2b>^))io}4Qg(Zn^@S+P)B!(o;p1Bd{<57J6FRnj4R+E@@
z#bkR)l;JeiLc1Va(V@~99zZ(Sy<S#wF{f{x1z~UzHD+;<_VDjMK$r00V*a&?BBmt1
za#ard1NX#wtxufQ=INQeYjuM$m8rlMZcla&{=k__e7tCtVA0`bH!kBoxS5p?C+7w=
z%A3WIdaxsR$>(Cpq9iF6<;}nas=fUG`5j9K|FB_er$>e}!w?&cFlZCNLXt!_gCAj;
zgvHdwRH}kF(Rq;|afDjt)L*WU-OT?9Dnf~{8M3`&JN0ytL^2v|M{Ph`&IwoC1?A@E
zO83)S^Jghq;4Wx0#J6b-pDK<QnsjR14BB`c!cEP{IUWVG0DaotP!evU|L~M|6csi_
z)v}SJ`b3Ys`efeTz7s=*iiDpqp}we4NiClKJQU7pS*uLEPrE_;)E4TbO?dk{n!>3i
zk;#5G!Md^i@p^Jhfhf+oKQVI!&$4Pogpol*b~a<7k@jO55?AnmWfx6~D>4AH)q{?Z
z^<9t8<eU>CBF$bR?+QPBOJ?2v5Ea&%ApMd9h_Jpj5w88$41)LoVDv7t`2baLglx5=
zd;-P+GuM{YR-$XGlPLDo%PJ_qCHtnrIATa<eR^c5hAQ4OPrZ|hFC~s)hV%3HWW%MQ
z7VToobgwRZedQzpozdR&(;0{Xe}%EUwR5U(+K$ri|0V{@s6*A^0pkgm^l~%Xz$`hC
zMKp5pKEL>}n{h|s90#J1>tN&zO@fmL%VVZ5u^Hy+mN9n<D=C0XA$L8trI@*fz+h8G
z<?*IpNWr(uDw#wvP?G&?uNsnr3DMjgi~e4@xj|&42+GN+@IkrktLD_OMjCk-BxmHf
zKk8j)WUWH?c|U;PTk;E}ix2KxrRjr<H41|3&Hl90%l9VOra;l}&EJARe$S%^KG04s
z2>nZ*Ovi8Q1ko}|Wj)zv^n;c!fh-ly?p=bnWtgK4WCUL=^*xv4u&&ir&Rqk1vBAO_
zrY}`9RrCj%tt&|3hi}Gp+D>iJo^jg#No`dtQBG=dCakvX&ceoF5i+;-X@;Bb&3E67
zeH}UxlsZ@>*kpfbyL`+yJ^Iv0P&y!Gb2j)rFo=jK<Z%C_@Z>u7>&=w?JKAIywa<DZ
z5@#5OzOz1!MB?s<hF9#EtS-fA2ApYl8{A?!L;a_1J$aRd{bM%eB<@%}7c#PWCf1P@
zP}juMe%lwe9D(44pZmH|lfbiUj!8n|FPIY<Iko>Px*!>Fh*&Lg@TpYiXuvvfA0{Uz
zl+<h34cgKYtQM>5U|sC96hbDeu%}RcDL{ctiHXs{lJujwHTF4k_L>PDu>vUHT?4Qq
z^SPlyDvpXnLWclcZC*D*e@k8ldb13({uaHa<F*(zk!c<4Ni$7D;vYvY+_D_g^d{Pe
zIm0M9q}KzF2K+`it4#9S!SxBffP9!Lb}mj7haGDkvzOfE&q_iG#__DmZrO;76<1wz
z*S#FMGqPiB)+Xc;genHO9Y^>+UNSNS5cuz0uUP(x8D5*)<K7LI$t>$G0;^OHGNzCQ
z>G{otCtKj*sJqSr{J~*le!`C0nWF!a40`OyX8D%1Ok?6fZ#G_XxIgz7wwazchn*i%
z2b#)YN7Coa<)et~?Y@On5g*_~_t*_VF5pGeYgRP4mF`&TaGUyl5GP-mJSd2h5TGYH
z*YnWm@VcBnsr7GDdFl0f*p)3ekLoNu(M}B)3q7~k#rjq48EHL-AJDht-1Y>^pwy3G
zy8*HM;kUvBZJW?CD9c3MR(X!QhEjPyfEfCY<&znsU3u=R<c7(kL3d|GVPoB#wCvCm
z2><t7KlZM)%ks*$y(zA7{){bWIqd1`9Y(0KnFzz0?iU{{S%2O0S(x+_A%lX;lNXP`
z5p>8PPatb7gEhC=O9@1T8}vIGFQuj<@F1G-<rv^Dy(V>sq=A}c9oT4)_^>rLwKuum
z4bo?PS(YyyP2N*fA?LDUKG;1An}Y`fHFwU#@Q<(`K)ihPj`HD#&KMRo{vTJlMJplA
zWx3F{;iU=g#3ITbu(^d48^HOeAw{gL+iIr%?<j}R9uE3xu9HksD`LBE7q1cli+#Yn
zh~vaxo5dC~(Li=oqE+!{49I_|TjC&MKs6QET*38M<GaQ*v0uhMl{)SHQwI9jJCIwr
zIZ*siqt}RQz(U8msk1Q;Jo-8uXVKhmVrLMjx@nQuL8(yEQ*jV=J@goAYG)%pc_|nG
z19Duuk>FseHML6nsvDHm3q$v;4ey`v(a46&GNUY%M5Q8MA;=b%f*8wRR>PDwr7O9{
zOjj>f^x%|DW6WyM=LDp7pKRvm6ADkS0fL1i5n10=l%YWLr_&q-e+yAt>>$?!O)bkL
zdaGyE4VGOFEqxzPYmSQ3qnET!eC<`It~P9IsB@7DY*9t<mxB|S)}zxS3&xp*N5X5`
zzpSTkq&#NrO4*RDt51VhUVWnugRJ_PLmvRz+jNDq@Cp5T<{u%5xT#j*dRD;dMP9Fa
z>C}8+d~<c#z$LX+E+ohaY2f?)mUYL>_0+icxq$@TggjJE)OHd^HSLoLo9O4<uQw;y
z-z_eI$3vAx5-${BPWYm{T&NvdHUtbBAF48=ZZxGq9)G(t4z7L5#-<}#3|%nKiZ)Ee
z>2}Sx5(IlVi7>bak1bE0dH9YZ^09&J>2T3?PuL%nZdFOkN4^qw<q*RBLEHNE;VhI=
z>~#l1lq_p*RxaVJ4$u=cL<ZX&9`v9VvloN2RZGXyVm~8Q%f1{__40=oU1Rr@mFVKx
zhJkCR$v0ue4Wz|7pg!B@CzG$xG8jSUeau4{Q-NwuFpvLcG2vs$@Q`4WM#EM0ru#=<
zLB2fIqT@HNVaRrtPQ6uTQq{d|+Yl@dOf9oKO#DR0=>p@`&D-~)RzH3Ql5{a6{T&xj
z6lzTC-KhGM^tb}~L(2KP$Sk8yeiZu5iaU(CH8Dz?2GZY;mxv5zJpz2F@k`@A*p=yn
zD@4Y|hoVZO!%%83p0Oyehh;Acc@>yP@~Z3ywr1EpO<L<MD~GRZQ}3csT1FU(bTYtg
zmCtUUe9}F-!xP88b|b1}d9nSnTr~{PH5(~i`!Vv3$J3b~xwikqQ=BIAPm(|26#JFj
zx;IrJo6JKlvt}8OrgU^zM^6(0$U!rCgqH_WR*Z+Pqa()G<1y0>J2Dm(C0i_frO4v3
zpSnp3*(iU-;+4{6372d1F>f=)DPZpo0@p?45;qd*xWou839_|=U5sY`{?|~RF<x|#
zNqoujC^^sRJAbpPB4un9Aj_~~jHvV;G^A}9D1hk`E*^A~DJy`BL&_*x+~qU;B=6Zf
zM=P}Z3paw;h2}TA+8sCX0-KG(7y3seQ;@t6*@xSCZ>(rL$O0C+F=v#<{EU6;*}gL(
z1*X#X_BwciM+nuz+YI_|2tkFRvOVxd9KV(BZ2%{?LCOR<RSsQj8M_66_k$VdDdp`Z
zUTG=6SoOBt^%a|Z*_(F6=E+d{sKx!Hy(=!O`iSKNP&nU1xYmrgeaKWA`KVQB3Y>~*
z89ta)oy5Z6J-$wzl+AXAyc<9pMfinq`dxOa<mX839-_b=>D)(IayFFgW@jB!TIF?C
z6_d%^n9$~_5ei3Id^_|Fbg4S8!8RIh3kY_A`47-3bCpFfQlsBzZmjRAW+ot;tuk)k
zk&x2ek^Cuj1ZYazPW;4>e#5<@xs|sj#4V+_7>e=6#$E_Xamu0Ft6q~2_wX~YQ(}kc
zkD(5`o;VcHfT%~ag#y?+V0|$`(}T&t6Tg^0%D_>TKN-URcTE#=_aP)#rW$P_*<&+^
z-7>;3>(@o@2Z?LbzT+WleZxsk3%&@EkEdlz)UNiTP5n^pkpA0JP7ekhSJo5i!<9Fq
zkS6%<C^Pclp~oaMvX@x`3Lti4%;40?bgYbnLk+;B>k0ai(-Sg1-^*6u;@0=6^?|q!
zHKeoQ|AFghhlpNW<Z@QL#jTk8_KQjc&WkI;gEiP3b^7zcHKc0Jz?BirVIO4*4OBF>
zM+h@2ciVDyhAQu$R8%)cG%8AjjIriRLv%IgV}Aao=q`c9K?;w$SN1%wvj$RNskHwB
z9uu_1*RAr`<VO!tTKaxt<@eKtx+j9BNUm<t&w?Hl4_l>HJd9C_lmnveXS=}9yPZ)z
zu{eK9GS95pw@9ICx(p<2C}o~JQ7C#pShB#iq5hY!x#Bm(*pMlw>#i$G+sespZcjWI
zJ@FDp<ZR|L4*?4v>s;*$h9%Mzk~XM(HI5@JL#&$L7A5|D-3Ka_tE<Lto-l{wEvxHR
zeb+m~i_yIZMn;fU$g0$s-Ty!+Rr%QkJPnSt!od-w1PTlXarBjg{1Jv@*Zl>b$CLhG
z@)19A?G3dXZ+=Kz3F93cW3PjZI!iBXk_5jqu$WX?7r|BfD&C>kGM&IJ2u^V=mC1vc
z>q#YL%mE&j6ISMy^oK&?3DwXyk@hT<SL5@*2!+{>)Ji*d-u!u{ZNKIt0NP*Vq+%46
zH(&Q47=TiZA33lxfJ9NPK-~hG_kJ(H!7>#n2!1adpm`JbKf=(=|04{|{QtnvOkC_N
zT>m2s&B)66{{ch0nCYg`<88KR*4u8f-;lXq{mW6YQMcKYqpn`7UTe8ue(ceA&znE}
z`dj9I#ru-kMp4tgqLG!Tq8TGNwAGJBbZ&4v(Kj$QC0$rhSz=>kZEb|$5^o;w$W~AO
zPID4aRE*{fk=_P^e>E(t1KdTX1>}t47=R5;Ljw(qh=_<s9+@6oRK_y5Fp&~+!T>>s
zkN@p2`NKra^!$Z5tuK$6U&0H$yo=}_00=I^@7sB182W84!T}OEumUHl2igFU7)nld
zH9;$puqZ}jlPCsy0Xpp!rRvPs0GXCJ3N!=q#ss2^3=L+9yza#$GedYe>GPLzHvEbr
zKM*n!Iyiu01pRRZ^+*TC`&FDz;snXQ4xp#^@0ARoCrh27c&YLC-n$S0Y2s*fW%;EH
z>~(?-_a{F(ySiH)xT(EiNN)&^s*1{BnVU~JvR5&69QQvxw=zBWkhj6y(`Sne9@dw3
zhGc2=m-<M%VVKefR6)<}gbtza?SIJADg#Aw%?;153_aRk^%}&^^!b<UpMW&Ae$~RA
zNoE75Qp}vR0AZ%}GWys5p3nF%OFs0f*dLr(T)s2dhx==On<EIQV+H`gN%3GWchk*k
z{Ef9iuJ_Y_u+g!-VIfg4PW&gQxAA_3PZ1n`)PpU4GO#y#CDYTn*EYQp(?MzX($VmB
ze#6KL-g}KR?)_treu4k}ibMSn7QFt(Z2iv6|29Pa>c2kvho`zUHmkDMgMLj$9`Vm$
zp?D_xBVgyR|C-quK{h>q@pFB-OVjoBHmZPYbgXA`{<=T+?u%a0{W1^q4UXO0|Ctj-
zIRj=?hv>-EOwdSQ``<M55zfmAkhyj~?;b<huN5Wq|DmB5E3|>9ZT>F}tqSD*_?!Eq
zpZSx*lN8&K9g$7F{p(lrGb#7Z@LO|{8w-^9%O8;%6#VW_|09Tu?s`XbZ)UJ>L}6%P
zeDc<L{bR>N;D)jLD?H>6T>rJI)jzb1eI{5qIyE*mCG_z6O*rchfcwFX4(ZMae&<Im
zI<Ydc|8XDtQND}Yn3`UHXK4N1c=eC^iF@uH!3VknO792IbEV&hYLxv>IRFM{@N<z?
zanFo_2@zarZB-bkp~@Z4p5+mDPbygbYT-?7G7r9OUUhg}zt+;e-558*O{KT9VA`fA
z@5sxp83ewCm&Zljk%mk({-ryXqF?(q6ZL`@OpaH3*0B^6T4_8Wh;lBEpTIdki{J{}
z=ph9N-k(JrnWwHGg|CM$fCZ_kdw7YZxO_cH@N%;MSEuO>y51WnpN0x4J#HR!o#rH!
zd39lV<+E$?0{wn5bIc0<yAhmahyFvLO732H8j9*t)IvT-A^7?xEp@EbK-SFNa80Wy
zt-Sx;wJigAyP1iHf?7xIZ;v;Ywx6j+&Q!fD$|8J!`a^((v`-XAnwuah?E#br+C!%l
zk9JX67Slb8KL5aFg;wqOCYue$r~g@iD@6{CjCZIv`o5#9Qb(D|5wkT7e%k`Z1w`M`
z%$z|Xllf+tYq(i@H(3-h3lD7LJ!Gd2ce&r0I?~Tbu=s(D8!;_A%v2TZUg<9W_6)BU
z5I7;&z$BGkaL3x%dn4IN1T!gWh_B+-!gS@Hv@iN)sYT5gJp0_=D$Bte2hgenD<Bb7
z>E)p&mc7TC`#u3WR>O%OkrQ)~hp<VKZf4h(nU0@ykvqGq=NBs8Uu&*#zAW{a7o12M
zQ>yoOxZKKI9z^&;uS#<!nM2O*QI)F&-RZ0(eDAfCLl<2)2StqTRK#Q%z!$O0sA5I(
z^-?6VOo0!J9-I1wqrzF#MtD07=jiUUgo0I_egA#n_uMYE@OL`#drta2i7m{ym%uF%
z#D`tj+nYK5heyR-fw<Uu_#FiQn>2vAAMNUAnkQig33ifD3ZNxxZ?^#5P-1*6SBr}T
z;{5_!<4~HeT=<}bb_)W5e*)2VF%c><)ueis(QZ-w7eA5*6n6}{MD^Eit7-OjohB!Y
z1qC>V`6B2Kaq&zzAr8MAzne9ayqn8u1CX|^_{}TF<{-c}HUGU9l>F17l}xVKsS7rd
zjvCB%RaqsdiM+C^9N%_evUnCC4_Rt1!OiLKGK~1S+Ds*j;iPGjZ#2*9{Z_#=w|X{|
zhl$+=t8u$EL{mAjGGXJH6X=nO&pVX+X<|9vFD$lS>_c_op_2rafb8`7C!cWqTroR!
zyVT%mU*Fzzp>6*K`D3$Zr|T6y3VTZ1zp#OsGq#PwOddV&L0@#x7EW7`63U`F&cE1{
zNwG~Oow>S`TfC9g$!Czq<|02Eow;w>dvGuo<XG<NA2+`jk@lD1xb{ghsIxTBEv#FK
z`u#N2hJf{P`$cFbS0FQ2RaAUh6C`W<brZ6kuqYv(w(NgzfO;+^?9p+K!@#U8nhf)4
zXgL_`QGM=F*xai3(kR?n+5#9&z<cU1!mNm99tJksYg$1%JrWM)kYZj@y~q;wD`pR5
zy-r@r)1qrH9cE{;{x>M*!Ich#9DFcS+US|UHe?BhVN53&tKV|?c_dX$N$!jB7*#9e
z^Xod(r(+>4II?h5VNApeAQrrr(bxz<it!GM%qxQ)|FK3@;cMeYQW=XSHCaSRc!_yx
z7Ni^sqz~i-R0q1bBnge;j1XKuGzPI-TMqWr>ZysLteSB_V@}#EJWTNIzc8~|k=bb9
zMK<<n1`2fjeM@bYl43ZI-_8NCH5S*F{_C?T$Jthe+^@pu2SuR+vpd{7*&aL-3>ABR
zS_HpgWKCLK;db3GY+EWY31H>ZI@+3IP^8NfV9=>_pZgwwg?u48PapNFy@x^c&n>F&
zNx?@U9C5m~ht+$}4A-fIuuHe?@u27+l58o*TTMvl6gU!%S;--9-rZ@ClW5mY7d-su
z;VKIu=YfaJ-iEs%RjSw6CUuT2Rz@~$<>f+#Yt!Ms8wI+dCBdPF80WeR>PyE{_D|1E
zsnh;W_ClO>DM21Wf1>oz-ztN%LmA$KJuV4R4Rn<xDAOU8E-0_dBdw{aveKc|0>-55
za#MWIqcfI`kza|kH`3k}#EI(EfO9QC9*w#T0>z*6?ztxEGM;1QoOY(Fcb@(c=%=c?
z)j&3TWK$>v!weB%RE)!@J!IPP+!NsCl_KF!?kn=0=5p|xvIK|4I-0OGN;k`2Z$?TY
zk4vt+hJR6%_89#%kdi<w`-;8ORMO+Lt}>&83;q_|Jt352?e$paFj9yD)3q)8Ui|T9
zZYdxvfL0g5-_R0*3?6Y9n+LgsGd5GI!lqoQL+{VWlT1PwbbV@xt{NdK^)nIG4ZmMn
zq|_>|3x7~%-Rwk|HSrx%hQNn+B{J~9JJX~Gq4K9AJ@TB>cH)qmiN$-Zmim_*zn*lD
z?7m5oj1{PfW3N5LTZ8>Aiiyi0R;T#`W+kU@(%OYeA9LZpN)I`+<fMZ2q=YoWz7BG!
zF{zK1<xAm@u{XsoW(Lpx^LQhkfGsJdLS+ycxpj#gss9Fo%2RS>xT@4U3M|PFjX0J8
zQKur4yVV5P6#b0(S5v6&j~cBp4o@z9zZ(t*`&THNBjfMRr)KJm&PHljI2OeyB4pBf
z^JA02!cI&!hkv*-NP%DI7$pf-srv4G$DdJHAIeWYB^bir=v#-1h7~2BSqicAzVZ;`
z#~uDfJ5M|+6|}`#k`%SFs1G~FrYlnkhvY2sZL-`dWmA(ksEQZ?{$n)-eOK#QUzy79
zRhyzU!_mwA@CnZFH!kV&LS(m!#i9Vq*r@+RFbo~15JQ7h75RqJqjVl_il)st8>#RK
z<0f}~X;tSOAI5k_cE9=uTd9qUpS9+(LUaVRAYwHwiw>$Z9Al!02T(FqWOjv;JmPw7
zoQpK*IftQ#LGKPpj$fNihhzGu_GQHpvRRDw|4KU*aAK4Po?KyJrC=}yyU{j<%hI&n
zjN!&eau3XuS5B!jKVxYl_&YqYY0|GGC-*9~;7IJ{gV2;uYfXbT=(v3C3$>bgE-u)!
zC*k)#nC&pvRbs9M2eI?7UK9Zkv$!A`NKFNsS1Si^yYuQDko+3Ks)p+9A8LR8%5aMk
zlEndhS|))8$6(Kf8PuUZIN2!{ZI@3LO6H#xBTkQ>A|qs$2G3zQ0!IUp|L`1zDc$%x
z)hTI8%S5F3R_bg3!c~&JLW(+Lt2?X@>>#DhSuFm(I;M)~T$4_}BPgqVGRfqP6Q>Nu
zA*1PiM~Co>v7EwGl^iSwxT}m%*Wsbram0nJ*s`$xVRkOn#;K7K$E-zzDLS1YyDnN=
znBDk8gx>tNxF#4+)CV}6zJz6rVhLzD|Ci){14nsOO*f*LNg-msIDJS?bhbY1H$I(_
zuRXH7vPu#vT0ASIFIM;u*hqlBeKy6Ao0yJGHb`6kg7;_i6`}eXF{0}D32|v6(y^Rn
z+sxE*l$tS;gX>R(SolvR?pK>`W#$pi_Dh|oH*GLSs9`ljmL<O=<RR^a6WnhDU-A|Q
z>@|J??}~U%#yZ}4d0Y@AZ~ck#vLESo)ilD%^o+R&u@P)K+hn{i7$W!xA#csm)#APq
zU9WGdo@;+rx-`Fig3^!`Y1F<uKf=69E-MnxV(!KKU~Z`gjJ=3d<%x)p5G^yjSo8x}
zpRK%Rfdm|WDb-`Ak^mWqEbcz+=b}XS`;!nLWGhlz+e#Blaz*V?cCt=DHg8UjNP?w&
z1(ZiuaPl!XooGl#k*MV1gMg%2@_~FBWN$P9hj4vrni|TiJmL6nHYk!GUdOH-(M08+
zDzAWt_Ng_&j5nW3mJW&~h*=(fTTF0_3Gyw3EIilewg7mO$G~=FPoZzC8v8U&3gFs*
z(kWa*>MymybXr8qN+Qwx=t(vUks#jR82JKG6R-Jd0%hjgWxie>Tz%f^FOMucKKFPT
z5YHy+00{6>%>##ddQ4>aCH#u__@Fq5r$wm@TNCq8L*zU&CsUaNiZ~XTo3X?VRjT>I
zRv>^j>qJBFjM)X&$K<`sb^xj+yD=sw4B-Wq>YMB|MlmV-(6b@qQP?`~rY$r_;Ymm_
z3LzmJAoW*?<l0>F#9XL^@~S4{4XxG%cQbOt4Yq2mTG8MJ7MlF*n<Zo8D7}*ox%YiM
zTz`0ASv_vn_{jLcWxF{f_A{2Si}nrs=vdNKi@;0Yv8h*Jn89s+-8#oku_O%0C;Xh{
z4iv($cy?A_9$j?ipu-jMPCnt5)bHw%T~GIB&=p7Z)3^X9Lpja{|CBkyx|s?7p0kZ>
zx_y^v8Ppj-qtOVrf$3Ab?-*m$+>>uvNB2krRQOQgm^6d!SbJ!^O`FwxO_`G^p)x1P
z8#gn=)R%{G{va`~R8DF)wZ}5P%keT|$6ZhYWiEtHB52PA5$aioEuw_n&OA__gDaBy
z(-%yr;_J8}A?UHF@BXvamX8%@>Nm4$C*_y;y$1=1I}UFP$JZSAEV<Z8t?9dLiS3kc
zU^r|$6nAm@qHaa4)a(xm?eY|j+{nSk(UMkBngUcVtpD{Mn1TASnk1%487rVNAVw{8
zJ^Al!W-=bE(Dj8LD5LEYlvB-=Sv}CtxJF*(kY+#uM-LB%Ah+|_&$(k^cfkx%yEj=?
zY3SBo<XQFe>K|ipmGli_$>W=hq%FzkQ5=fM8@%y@wa-XKaMw$%F!O&1vxN+Jl!7jP
zGHc9-vQB@HWor@2cT1ajqR=MgPEPYkH1t&H2TfUET=g9-ZwStvVh{edI(ijU5w`N|
zO2OF&jgb7NKIm4}C(q+-FdTZWE_P<{$vf+GX_D0X@5D@rAXTMdVG?+Dnul}z2vHfY
zrJ%{;kIs{=Pp!%bM2c?dCrvV~(ssHx{GonE)(j%@2S)~;FD3Ub96jC`naT{_(DDpp
z9&NIg4<>_5V-FrS?})!QR2;R|!$8slr8l!@!QdtKi;y;F^Up%8Y9Gbp`ZFkYQfAmf
zNG_Zp*BBXR(j8Dmrmg|1mMT}yzll&41_uD9<N@vwVQ`YdZBv^%e1Rb}i21XG^6a}P
zzi$3vQTdo$pRR+Ii6jPoRIbwBH)o~#uR&KAA;@p6{;3)rb~sA{QcUib8IfXq>h9+w
z;RBfbK@xb$;OGUyce@<QpS?7LW?D})(Its`_-D`jcW@(;H&=b7ubB%|BZ@RBn%f<Q
ziEMq3qJ9AB92<unb2ZqxU{m(eF~2!F8QtS8o8jd@cPE<Std(rS`zRwiB12{#bQGoB
zokmK(_I#+W<9hymZ1*h6ofDeze(=#{^$+Gglmw=MOvZWK6j&wfHtDo~W%hp2_?^m^
zZy8nbp-TLLvsqfWb+{bcEoO+{MiTk0OQVd#K5-t=GvJgNY&DIJyme1R#7n-tK#ys5
zMX4h)uFp~xTPAQ3<~KWH*Ak;DBxWN1TJ{{?ql%dio+$D75)1bSL+f=yJsyo7NAlva
zN*wi{>+8~26erGKEGe-jr>I!?LKf-YcS>~q7W934ue4CDX~cyD`qj1v6>8k`2e_~r
zZH>D3${UChI7z7jooIrGr8;3%S?^VR@}h?P#yvvh-1FiB<nX-tn8parxfPZLmvS=S
zUMi3UR~O6+xa?Bh+W!<_LIHh-xrwV__3B)*1OH_{LnVR2Uf)MnaM{L+&ju&toq(`(
zv};euNq~~l*}ca>1Wf!#_X<c!ewbwoZp!(!-Avt)Y^+l4@kSo9&Nf&^8p9ZhI}pn!
zEtW!G9~Ua7zMsNxmX2xbBpo}1LzN}uw<Mc~cTdxUr%Z)2erSdS3QkCMO#9Pi8@OG5
zlEqFSHDY0~csH>RiZ@LK_(I*3)bie0=jXF%w1o~U0cRA$u)9zH7Evxz)k>4?#lhfA
z0sq1Z{w*qbnL@Z|!DP!GX~f^yK|^yk9v)JNrLbghu*;`59anyv`{NjH+{S#WwcP&R
zGA(R_ij^4;IivQ8*h^Bi6iJo24;BDF3V{ZZ`w!W@MigZSGBXG<h$S3=7vpZ2^#>2T
zSrk0xC<vybpE1G$vwmh`Hmw@DQSr(jg>B^S^$I8@_$ybW7w^rl){;-HUt+Jq45RLu
zIS!rfn_fe9?4c$0ev>1rhz*Pd+J9>9Gnyb-$p`#$I;&v7Wi1IG{H+kuQ^=tz_>lBJ
zu`yP0-c6{a*m>rPJ_$Z3$&4GpQA8gvLz~oU>IBPIcSNd$q#fzIe~h*we;P$xfFF7|
zT}0rzT>V>Y8DwGpY_QhKAeeoFe85{0flPTO5!q#nOSMmMmo8ds{0LpbGw=@$?U5i%
zg-9|@4^Q-V%zz0UDm+QuGHdEq)3D10<>{pfL;m>*D+0(p-AkhPfk_aRbLcphg?Qp=
zz=gTrrOKudBWx@t3{Q!`fxAB!yYiA^?O55k68n8Gca|k_-7(u}W==eJaeclj#n5?8
z=Lyuw%JJN9z>(xy!L~X7i;m`yuV(PM=uCfg1nM*OGDJ7?qz~J)tfeU`{!Z3^oyLF5
zhUrNW9V=I`2ZV0(G`@(`{!6CRQz|An=|z^Me=;$kIHxa^o)KD6&?CuNqMt*YS}kk%
z$XlX(8Vx}kA2`cLRlxvWNAPMJ?2GHS0vfrbkE^SD14PXL@e`K5Tj}_iv)qhrRIm7w
z>lT6Rqbp%5B(adO*YT>fn+-TFG){y|eJ@Ekd?9`4o~Ukd%E}qa53MegZy*wp-Qb`}
zjMz{2f4NQ=u?J$r$kR9Wkq1|6)elW7yM%aDzwK7h#LH^o@2UMds)WU09e$UEL&qVx
zpeJsFGP9p|;e&lhJdetupv<oh@Rl8xvpM5U2#xTDBbhAg3IAEyz-pQ{^zUJk61tOY
zNET_6tSd7tBb62tWyA)!yXWw$J+V%<`Y-*4ofR{E&p!$9xe@qyikwzUHhmZ>P?=6^
zwinXM&R8e=IX7fBWM*_RWf%F_cN(GX0^r(vo?Wd#(bBA7udo{hpo5Uc>JxlA>vUC!
zzqT))wVGBDfo%YZsGNH}Ila$UY8ag}rmz6l13;I@q>~T*LK*tV>t&DGNxPtZdTKFq
znE3J^l5B>`fyGJdn0hw?b_t!~ookzDtijHMtDvoE_s3rxXj_%f2k?SxKD`R>gCy`A
zkqqkmvF-4GiqqqpV8U11^o)3JOXKLFK>UJO;YCkF`YaNTUbPQArO70e=gpsC=^agI
z#bt0sN&J!tNTPia@fp}pLW6NUhJt(|v=14UF6XNkN&78=c`hmoX@gj+Jk#?D6klWF
z0k)cEj)%_rPhKy~LSux<eZzK9O6%-fYwfe&6WGglwsgSd{obN#{&_?_<k5dMm8`Xy
z$~0bsVl^nQO^awnQ9WD$!+>NfyQCkWF=7>o#S7~Fc*2f85kvL(Rn4Ov=CdYE7J~dW
zWX7=zbdlmx8^MEBzYjswtVuUwQ!W=+a5MrP7#pu2gz%mrUfyt%R=IA?)P6G^_))X%
zT_7Ir+A`JI+H9&yEqn;n+jW-`4>KwIzS$#F8c!dg_8DoT-UiMoG20I*$7v%(3V<}{
z(d0!=|Ju63Fd}Kag^g<GZltUmJ~P%eExu~=7?wuuk+v&NfvIn~y0$HB&>?gTK{i^E
z!@aibeO}i54wHW50*_BdC6lMM@=walG5RpXrruEUI?Xq5c9ACux43>LT6}JLpYjOK
z1!dXNjbx_Y)~kU89D3@tbgFl?d(stF;%cJ=OP%RKwQuYse(}Qs(ec@(ACODX8Hcga
zV4m_;{>T1BK|N>6mkAQDSayFM)4MT6ZE9vC8y8S%jmiFNUH??|NJYsv(70c^SxIZG
zr_(Ty+DCxs;`}nKh((X?hMY|BCMKERBHR=rfupx2&%;Y*R7S<ExAmJc+$raXWsO==
zo17N7li<5GWQLIZXcuwg7Nt1k*T#W_oHgz2QPOMT&jW(H^xR8(Yk^H3%MV|h%&6rq
zYRlX463G060+=|8z#xyViu#H=D9a|m^N`LVgX-b>Jlr|~iENd?kZ6;<klo%eNt^E^
z5&^X$VT?orMtZQpDF%+aL(qMJB#JJJmU$T1%nL}mGi_d7Vw^9M9R06_fcDE?_aJNq
zXy96RL^h%p>|af(Jzgkk2JeBv7Wr_${al@fHhr#fjU@|JgMMpcll<Hy0|LD3i{-Md
zm?-f>upYi_>!ePswv!_RaV@B=2HUUb?A4I|eV{>MEH?R?_4l-eg7NTNE$+mdECF?L
zmv!k}|Dq6re#69PKx;f#05k7sOoWfdH+`Ank%(3Ub+-pOkH_oA;EMjeXMUT&lg5v*
zA(Y)ES|CQZgD*KE?Am0g{k~WZ0-}2$gjpDb@3~{#hZ4+S?p9jDPna&l-JmIkNCe%P
zOjbE)w%%O~W#k2v6X?Zcm4+J^f_mnQXGp+hYkrlpsC|#^kp(N=Hwg*(d9--}0U{vh
zslMMU%AMo<2lC2lrsY?r+UygNs>95@W=2fHdIpt0SEknRn8GVVgCUt-%P^G!Hg>Y-
ztI7`87=5E-IqH#h>61mlrQds5`}k#-3V~mlg(+bIb_hU;5uJ*bK>nwEaBw~A6vg@m
zSg+izjqd(fW%#hSO+^%Dn$W!dnAZ89Xauy3>Rc)j*DF4Mog#9zF&mVgu?>?;!$8F*
z(@f=4wi!L#G{B*XASI9E6d9-b4yONJzP4T{dheprlqV6y4F!wTYH<J7fn`}8!{xvi
zdgnW6AocjD-Y;M-vU&^uP&y;~iZh+y{-6Vfa*+a&<dKDP6RN;;W*P?VtFkvcSzS07
z_zzVj%a^f>zJN!g&Z~H+dJyS?o6DfqHTuMkQOk2^us1`uKMo17QwCiRQ*;3|b6a%U
zTlDNJ$BMVhOoGwaRx5^iLi!Y8NP_2uNgS3cyP1)vU$mo7NuAIAPu`zB=#S*2&z@^7
zo;oHf<bzQ2jwT|ct6tO?!!fHKE+Y@z<Cm@zJhF*`iV?cFhzt^&Mgv)VEH2^uFJ#UU
zxUE^pz+QTiBX)vXI5kpb8}Q?8eSq|l$o<GKSEuG=h)}p-DmG5P`I}+1b~Y;<W~wjB
z)g@S&r*go(GN24ko={veBrd<lkk(Khk>3jte_A4~Anl(AaEv@Dz^n@-HE&?x;g~tK
z2xQVVM~TdUs>RMr*_xCWU=2J#m=|V}P60RIC_N3l%R7fUm_X@QR}pm$Z%mvUClt7K
zw<m<xm>2q&;`m2JFC~F<{&sXS<pp)$FJp!zctFPiIy<U3_O5LGw3HN`_4}gNCBJ|A
z2;Z6dxl(o;NuEeuIIg3*rci~?Os-I%YZ$+<ReSEK-=Q(-IozZ0y({M!Z{#p$c+8VU
z!Vt1mWGw{-ie$ze6liugsEVEHkarv6yBqY#c3Jh+#IuElDFnBP!;wN?Y1fWXxmc|a
z&6}7Vez)t8T$a|HMpxMo2S>DT#%heunc+6mhem{-gU`(_7#7~c0?FK=+i%0%hrTkm
z$r}p^{i-kcNTO^ccFKG%hw9!522`H^IDXdYXysWq{zK!~lFmfbl`IClM*4M9UgS1R
zyI*Ovp@+9*<CsfLW5{AfLQ{69u{WWVizrCC0Ic!fcALl*kq8A=NndKhI!^Cs7OcsE
z1gVjD^Op1ql9{STpxI3$Z9@e2Sdo>x^9bVl(ZeW8Bp3Sv3nIQmNAZqG!LC9@Zm{Gl
zAlnr=XeqoUekMfvM-L|$H(#bQ6&;Vml?&12)jy3E@%<U>@-Rgg!t+m4w=?2szmVd5
zd`!@%ZWzV4!CxKBQqy*>yGKBvYN4=0EcW!4aH$bdG)hMbD7<&PTnWSV!W-jjtll){
z+~2H3$DQ3oOdmizPp>8wPlWIRFNUrr_7u$KkQbU7|6_HbreOx_$Mc@WzJB5!ec*3K
z_DtYT7mMN+@2IiSLVDMEgHBCoT8C3MP88}53UWKIe3`t`;xkBY73&DiU@0613A@k6
z{m|;}Qh``e(P*YLbGNX$U=-$lV?ZS|#WE`?%B)LD!jwW2B5VP1H>`v*K93FkmnnAH
z+~)Vkd_+VG{`4^=L8l%j*}=!cBYEJH-Tkl=->e7G`f>>c=|APisgRdII!lm&L|a>y
zGUGz7`*vlJPTWMuMJcy-bVbpsCJHxJRSV*#0vPD<xg_91^k9m3Zo>69t-eF|ebbI(
zOk!=Nn>7?gt#Q5o77VJ_oiJnjzQyPVT|+!Z7Bn!xZs?kDBs_&C3&SGT7Z-jqx2Hcz
zh8i0}tHz%s*wn;lmDEdP(lZ*2Tb$gN&l?teX{syfK_1FR`;n19jvG+cn=otAw$jZ0
zl|g5CjKfVPshw@U<UD?qIN7|V40M=W#*5mIt341;H!@~uJkxXFQ<`$VAb(jb<D?;A
z=Iii4ku%=umf?H9-y^tOQ3_Z(y6~Lq%3V(mWl!N}3AMUY!|}0r@C!W3Aa^`m-^BaC
z4f*8m%-#k`v_s@@-r%-9A?s)*V7tzFEsQHrF!L73vYkqPyP|#`nK!?38v!}U?t-pc
zeGiw{nEB1DZ&Ue{EDXEyKZRJzC5)B^jitAweg%KvW$}3h^{@yGza&qwFha|dXVAQ5
zwp?j@N*@c3Td@UpNTU}AA&DhFYeIX&A||K^Fv%bpzo(-MsrHUY(cBA!m1!r{sEs4k
zXFDsn_io7f4Pk}i>2L-}rdddA8yYqqXe$8!m#Rf|-Y})v)D;=J3H4{?$V^&c>xdIj
zJh*BVJSbY>{4*sXDB_<%De@qI6x_2LdQGwVK7YlCpf4NY+6Rc~(eTc5%Vq_nwH^~9
zySv2-68uOASG~A7V`GVog;wywkfFx2`gOYXGl7BeN3LmFa!G6xpe{wbw!2klydZw6
zy>x(i%YS_judSMk&24t9$ah1T__@o1cyg?HAb@@-FUmx?vw$t|f!bzgN?!PHqV$I!
zJU%``!f8q2HRS#tlrmVl^3fxPwEcFUq?MOzH*YGqf^zRCHrc_18NSnKFU6k~Cfo!Q
zf{(-QrE2Wuyc9sb%Ar|5&cm6-CsfO&6XV+`oU3(}=#*ccwaE6y12#39|CZL8Z<=t+
zWmsJ&Xgi!<iuqEu7KL%aL%A4tyUd@qE^Y4+RckIM4P2(<b6=XSUIZN&4AD4|48sbG
zz3Gq+r-xB@-?A@Ual8jYZ>5W-{K!TGjkl_5@=z9hI=P9a@UA?6@$!3{{vhI=P>=^=
z|NU0pbNgt@(S*JET)1^u!p61D{W%U2)(TwcHI^FsK*neB0Pdzti`#jOd~#Oj4>cpl
z&CRxg_GxqXC98s||1-Hfanupc?xxKEP^p~XV|feAY{^9m?G(Scc^w$mL#5SMd#Oaw
z_BL``q{!)`tY|||m_@-Yo{q4W;#vu|2ui)dZGC}@TW&K8!`ukS+(loD?abn}oA5Cg
z0#`>LqnKDzw&+f*hIgSO07=?4b*vLvRUgLKUl*-;H@6|0SEYgr)RR)xr+(LlQZa>*
zI(DC|!N(25D{uDcg%YYHO!~ARq1p_y%#8)sQzQeHx|Xq;@}Ol|v+MNn%Eof&7S$k+
z^OG92p&H_|72yMaqQ(ssTVsKUXsrq)Zr>n!hWf;h<6una3sN(*^&rf|h#FTc{Rp|R
zD`bE#`ZVb|)6Rgb>g#y@Z!ZBUeee6f#PZlZm@Ev5fT?}mtQG2F9AhO*N#;wDQr>6!
zGd_0K@%#W?r`WZ6!H*Pbcc!vyXVz@8K90XY&3wtXNAncNx)pA0F;mT>5xErf&?9C|
zts<2GN0dBn*mhc1EE5F*Tpx*Wy(*D>0Q}p1o8``MV~;k%=T<{oo9^i;YxQI1RBocX
zEDJ-<n+oCzT(CfC&bHZ^2((F4Ju^>)(rh;CjetWJ%Z2k`gkL*%y(pWw<+*ctH%>;}
z8ClyU&OD+7s22^Vd~<Cm&<%JI1+PTulIl20S*J4&<n78nErjiQF+|sm$Xjcq_{*iE
zFZm756mOe=p!2~X339iO`WEeFo9o!1<|SCjzfBBkNA?a-%8?tI=G0DpKEh5^OIORV
zga+Z8Sk)%B6}K;@c=YBVU_MANPTOw6H;gB`scW1Ue1bOqp4kryFJa>j$389`^+P1(
z10U3HuVcQCSkYeyRQ10oyQd&oy0u-{ZQHhcw(aiOvu)e9ZO^uC+qP}nwyoXkTYLYA
z^&YHU5gCy=>ad=x#;D4X_XB#8AyHHjp)-22lJLm4aZ;8+rU;{Q5NMXOD0$Y@cw(Vf
z_C-VlPX4N!8aw-9jGw3C5%SSlK&Ygdv+Ggma$G^1`nY9B8lk)p0QO=N2(6DsK;ZNs
z6{3*+(;-_V--zBv?MU{c$f7dMmOQX}6BN%HDXgGoR9kSWJ_8q78%6@m<z@}!fYbHX
z?g;VPAgwccD{}up)8zKboc+Z)2LBhF+A(pI(}Nz;6YHEP(juv*A{*<yTO!p`DAtL-
zi&V-kz#AMDuEz~DZUufTs9Q7w(wWuxcV7^a=*fx}KCU^|xIbsKGD-M{T1(f6i%LQu
zdn7!KaQAF)B2GujDvhOzpK%E{pOoK)`RK47w2{Ob<+3tJr^lZbr)Bz-8R98t-!j}7
z`^xuY&gAf1V){~~6G5S1L1jpjKToW=yNKGs0adYql%osO4kcS=QHGwXZ>8ms5h2ur
zvLrzwyTjA%yaWLYNt$~FBUXR!X0-*@kuX+)J57t=MU=Rz7TyRkTk9O#5iVD!ueUaH
zV~3NkEM@p3l!@ffY-9!nHf1^Xv##q>SS%VGomeS#mIbUzB<${JtMomtQXtTtY^R}3
zue}Y;$H>?<&+LAFk2zg4my{<4Nx@fj*zGlT#l5qaw7bDm<m?9V)?$tnkg~|fwrs%N
zX?gWXoNBvqJr-E#9nDtpJJ~v&RDz-Akms%!BM>ev=Hv=?Q^ks;MHYC<1O|o9=q{u?
zjn+eAwzN1AJRYU^o)Q`tUpGdxNK-S=VObVL0o7^U`$P{w|H*(EZ{q(Lgf->CnlaWF
z(>`zNG12==P#8L*fc=%UT{0w+@UDeE91?Efopi(~JaN8`T_$>y_DA7OQgl~_EZv(c
zSo0Qb%u0=|E2n|ogD9sZRHn_{&~h6@ShgzKNSNrIdxhyK_nc`#kXkl6R1h`5gek)<
zWFf-+7N({m%K63nTOJhV35|<Qf8hCk-rfXLulA5x7L$2|=kwGg#M4}?xk^Qw-d%HQ
z#y_IuS|4_35BP4MUNA3=?5fCB-*X&9L=5NFgTr~M{MP#=SeMhN=FYG^|40G=A*fGs
zFgyiWsKX{Py$Y=zC=d2xwCELN7AH#&!}3r4Dp$xm27;hwnaPy>Y@Aud<V#V#+r{`j
zGC+8R%}CaK87yP>gr}AttXR+ze>@7?A9xq@E9DQdi!O=R<1r}HP-`2wO6>VMiKV(=
zMG7HfwcxXp%tBJ?EH-tS_68<_kYw};2Celjg;A}2oZZ(8>{(j05t5yQ*sM0=u%s?C
zL4d=3o7Y&4E|(EFdz!na9CUNH=0MU)G>=6+Ozom@32-NK>uM_sVVK=N4gtFw6*Eu^
zNW|vi?kU<TMSJ|6>!jaFwbhLnGP?snY(KrPxYKxiexe?bkKdpnetESmGz2bLKgC)_
zFa~8lxMr$)vkGq&*w`settz3QboyvL3<tTz=3Zs4wll_tn~OPj6A@o-qnWYfdGNZ%
z>*@(Yh2?)fj%DHu%UlfSOLd@m31QFcX<VblS1WZDO>O@9h~&}yaC45KHcdE>i%Tr%
znZNzb<VCpM$4`#CAUfqg8LZ&-_Uv^h33N$|q0N;{)jLTUvjMu=ADX_1&#Y}%p9o2*
z?Xm54>8~~7`SM&NfG`!iYe;#?W>_Sx7Q)g0+{D<7ViWGX^Vh8aD#Q(lJ#qan=<{K#
zl#=d#=bk+qb6srTDyn&p`ms?bHC4qUx9zYh11#0Aq-(es2V!vp%qdqYNiN1Sp8e;D
zMvLe^bh#-`GD-&UqpqC7i?YbibmQUVMj`RlES7LXVOhEtO<9{5IhFJ|&tG5Q_K6xd
zG<M@(CqI#D0RqZ8hcSebbYDacP3s$tK@Fw}lcx}m)fd7jxEtnozGGn!Qb1LIDDOKv
zu8-HGUCNMES`_6vL*<V0uIgnyq70gCRdZ-B+hRsS<q#}ORFFbc2AVe168rf^Z+q2L
z0i*ZjPO<-H`?A6MM7hd2FH9FS4Ne;}`iD(0$svzEs#^nut@WOEJ@)rY0nNEw&~fOT
zMv=Rr4%Gc(Z3SQ+hR-dDKeR>&TK!|27`K=GGzw>8=Ln2aziuoK=Nddq5k%^b&r_Fz
z{}-ZV4mZu(vg|rT1EraOfo!$X9XFR5gviGyr6{^)*-3{dYs)>EfrD3r&|d6?MCXa0
zV;(;=6zRBwR&O-Ki%!6M0PC*_3kKAyn?sO=g)<t2H=5#}Kaa<j?H*6GXMT4X2Um++
zTU3d-gLqV36V0E~*Tb)Xi?X`bh5^P#@t+2Wz7Ih2vhS#%=r$dyxiRUal&q9?_hKTu
zDV&nt&PBUsbi!b6!!_erC6tHghZ=SYf(mmv5mLf(I(-&gpHny~m&*V}w%gn}A%~4e
z<GxF%p`WApi2j&$E7<uwMk8BMo)@hZy}n)<4WNKhCJ1CDXiE(;wy@`|_p5yZ0_%^U
zn<DK16t72Bl-zX!%?&u)8IX*3zhO`HzbuISd>sQ%f@xDo(=8^-%3tp*?B|s8aCW)W
zfxcsy4YD2sJ=~h)ni*!j9sKjftP^>^k_bln*HMmwW(wrD397IA`oWY-<H3H~npw#m
zkv$*-Wj`P1DgW&aaD@V%W+OhjNm=dP9y`$In1`@}_e9cg6La+as=mgrMZGp!0{fcU
z=2NV1*r54ba+g2Ma!mKwSfwANeFrwK4RB;oAh|3l)(J&hD5&?b%%Lk+Z#wlt-R>{S
zcS?)`Ke7&LkaADfO!{)XsO&HE@zfQg!>4eO^>kL~=+{XYmk?ui^NY04U@=>P{dMFP
zbxcgmhhdK`^ya%~U4jVe#d?cMDC^Z2?u%{@d(c=ad)JydLly5s?KH>j)BRqV`ed*i
zB02c?&OTnM53@mi1c}MCvHD|QGGzjX(R2b+mp#hJ2$WlsY1s1x#dkYi;1Oysoh`uq
zqO~QGZEe-rS;&WY>>`;IPru#$gr3NDbElB~1o+UeWNcY*DgYK@>2`l42ozox^J1=t
zE^&3Y!M2u0E`5&E87|{X_R?1R&$YZViO1hu(=rki8EzlsJNFkS?IF-4fU~SGb>`e@
zfbt|36|GsP(0ygHRKJ%Ooa~%F-Ub{ZHzX8i$@wOzkh_4le(*q72>;7sn<l(s%WQcH
zA9JA27$4nm{wL|09%3)}sF>S(%tve_?6<UVjd>@W`=CbBwT+>Xzr4illcxP_RpeQ$
zJ!lMiwW`kO>}DayUD0m%UVXk06I_hngl-&~==GxjMHF=!t@bV3>)t&LB5Mi6HbIvF
zMDn}WuX9owF%1qMgi8B=j%Y!$Yr-U@+jZ8^F9lmuvPwd3uKi`itKWBTI~2-9-*~xZ
zKwjT?t|Dmy;tO>b8O!?h$Q^V!$8f#Yee)YbDuVB0@VV|KcbucT%R|R|%p_kR9XUqj
z{@j($A#(5W3|p4ItmSP1)j?^;$;d}lAuaGb#=KdT)isB}i|kSkK`;J_YxISbd$l`b
zA5^g}si>R%^{|Oacg+C}lJk&`Uo64J-fMN2qC86mkQOUkzATjCZxQ=MdSTj&GqSKc
zM*oF^GcPNMPK>gFh!^gx2|W`?xM2RVqUftd?~ZlTTU3gg2cqRpeWKEsz%;nAz&@y!
z5mHXOQdICO2j7U`eaxlrzDAFvK@s31Dpn*h3;?FC39bX@-HX-;=(Y)TgQJfT->%|0
zT23MhB*fHsW(y%bvef`<Vm7$ts}Lx-cOqPg>imLUjHMXbMlW!@;lnt;SSxn!_UfWq
zzz{4sJ5K`>k~8V*JxQR0JDde-cQ4=6^y_Y3MPnC#i+H0BO$i8eDQ~`U7R}OVpGj9K
zTLd0@n*q}NDMoOB0k^tKds8EDoA4N#MHKwDu>=>B35xiPvV!Kh-D+crgNh@V&HcVM
zLb?D60BoGn(xBt7*h|<uNJBd}F0G#rr?mYCFL7doTm*$oWDY&&`*;hqi6~@=>kE3>
zc8+TH;<zn#k!1ZizY|k&8kG5mP<s4>Rv5<h0$H7KSOY7irkcUC;b>hN_xk=kzF{*v
zR_op6Y7wIq$&uV?5Q0=T0WSxqXnIRh@^wvj{9izAA`)z-^{KN!8N3m*H0%5uNNyc$
zo5*@ejM%9>PFBv1#^$<>KUFN!^lJ0AA>Mt3&6zmS*MW`H!CApFb4Jr{cQ6D6&&7%N
z(=_~|_LAgV{iLe3T8w4p+e6Usde_PFdwuNMfKL(1N5QiCVo7R2<|AZ>RG=t_^rSzD
zITP8p)Wc1r77qm_2YL|>Y}t%Wsyob(fCBVv3IKwHCE0krV@0OPS>;E+2G=oWI{nk<
zw)II@+pWr22~x^opor=0n%X}xM>3+U(ca>(j1l;kHMIgZ$?IZx1L8Hc)Nc8#a@21*
z8yg&RF~e5o7M+*bgfEQLd@XD+PCZ>F@!fGObVkBWF=I_V_;QvJtjQo+@JI;AyDbnO
zO<DWtvvhr0=P98Eit>8Qe9zKXyuC991f=5A*$0tF3EcIS>uNHOh!>~fH1nR4Q9hes
zQcVp(FScaX6_OGNSr4vRCe0a``pE*&Wq4CP5Pu^kiTXPu^h?yDRM6T|r|h7w6gU!F
zbvDW6%isw&?q$o?2o&8s>&wXtiZ6_$lFZ`xb><<I9XDG9QAzz-sDw#oT>h1rmdukR
z)Tb-zy#wn3#)k87j~oHpkTnP$DVKh;`S6{pFcY?ojmsyGo&iK0$F2z_614@lo5CD=
zRI<FzDz(0p+I9b1n`;MV{~$sTaMJckh>woV=mpa7efQMox#>6<3kBI9Ny10c3UXIW
zCHZWu#}s3B38LRqCUO3h`!`qJmvdp?3T=f^w95DHgzhAjM~LEM4V1zLO;DlBY}#p3
z;QV_0(so~jz<Q1qO^wRFXW>$3vGL`Dtb5FO83ju|01mZzWHq1)p7{&wn?PykN)<U-
z-wB*UP~k|YCZT78+g*w(gfL!tUl1qI?8Z9itc%*B!M1-r8;??3u!!aPWC6OColS6Y
z)4Z2L{})>5SmZpgUFG^I^T)}KNDjFEA<@242qG$lqj~;I5O>160xqyR^Gbb9u0ho#
z@&Y4s@O5E?eAKbtPJ49cyzqNlS;%Bws$}x~>NJ)ptUgY}D&<}%^};~d7>+}=w~)(N
zVm16{FR0mG$LjoS`F14oD9c4PWe{j<x?oVt6Q32X@Dzo*xv(Rb<jFh>8)j2|CrvTy
zDaWxexk!~KHReSD<xI^n*3knYGW?-T@YhCegEGLd73$e*sGzg5!03KD3(Q>^kLBWY
z%{1?T_;{yBH~a;CEgYC`U6{WWRB?+y-D?Pi%LIF~rvViw4H#Kt?pK+F7uDU=LLH(C
z$SH9+nqVX>myBZe(O+u~F(5&iuNuJE-N&phZ0=3bn+8Rk29=mmo{%-qNt<^~Nr<~r
zCgL6e2l^)?H#zGaTD^(Cuyw&q1&>0L$Vsyc@J!umwG541h(rKO@N?C6GhZ3V^9Ig&
z)4jS^pcrnb5kg-@NICC6iGwUzHw`t|txC)FGnWaj4<^Uy0V>%g0<`(2_&6r&6j5lT
ztP`??e_nvtDdK0&DlEazVk71SgyuyD>XD%IirsFbt|U?wjte&uc%qeNUA9x6X(|qV
zVIdz2SdrdpR*&~JkLMgCZ~@44Zctv9ym!n4yJhGJXsQaFsFZl)`TYr;2+Z5(&fL`O
z=<+|TTHx<juKO=3Hl9TiY6Ah0Q}3U_;OdNqPw-LV9P!JFKyj7WNq&hin2lb=)?%?W
zoOIvD`S>wJ(zDP{?8&!C44)Erj+f_#7dg}L?se9fddqi=tpD+Ax<BpRQ;y&(U`py`
z$B}o84RSFsFGJJ7tDEp9hst3}<UPYadom&SF1d->G&PsJ?T@y+kzdnZXO({L<e3m@
ziA7ZcN+LVPeK5pB0M}JH+}169)qkzM?^XCcMg_hr@P>%RymQoiLK$CqK|jTT3qQaA
z_ClDjCQywUu`8s+9CnS!mgv?8ndPO;0rmFqXjTzoPA9gtwj1ZhF5w269@lA==|;Or
z0$vz;cm4d^%dt~=c5YipPSXEZn4}Hv7=EC+2VXPsM+bG9yD%3$&Qs>Pudu(V*@%Qc
z@57FVybf%TW7LY!P|N{cBVCpA&pw`;g&V}Utx{$%@urboL9K-I6?8vz=)ii7+}sRB
zSf=99i?UC-q)B)QlQ{MGG#EKqJ6vQo-QNXyO4b|iWMRPfox7PXTI&iXLlCf;pDr=p
z#DO~lRY*_9D~{y*Ikr@9psn0mXJSb2^J_-c!yFXz2@V_g_C+-DZNbGtOq|b9G_^&N
zLMdszK`R}sZr%kZ4)kMiM_zFaq*cufdG(4^$U%7-ImrpD$Mbj-;c2$HTQg#fL%9;D
zS6*_M`35u4UhaLBRpOx-w`TQtDt)Xts65x?BX_OT<s*q~X6*TnK#`dm<x6-C?S|d?
zW`@36;U#d^8%8G`XxsoBGGvLZpM;{L%b)2?qD7u9%6A42nsfFNF0iAG%T_tf0e_qc
zemelAlndg*_4ZXv8|GlY`ICqu81ZkQcfCsD%EflhE4B2<5<<&3*DNWg1Az!tEe7bK
z%AA7SJj^Y+yJiEJ06USez1uxW1Rs_I7=K?cK2O)=KY9F#42@F=AbYxFd{hi$FO?>R
zM>85uzRtdr!cAm9e_KX)yv)dcX7DXPoyi-+W)3fwmXTmk6x5566a(m*`U?*ynQ>Z}
zP5o8RX$N6oIJnaqeZiRq0&pM_h?!XZyNNcF{Mpb_O4a8i7A)1C<oH(XWs<UkZK++?
zJHC`4z)iC+#ifU0gBo=vEIbWvDp0^5DojAh3@34?36`g+4wHHgJ8};>7jle9@PkdB
z^psPe9c27qc5Br>JEo(iQ9Ys_<6VV+Fo9jNyRJQ(YI<2yn%|PLuS$Q?I7N~OoOOBh
zW-#B3OCInyqIMOvtDI{!=W;-WYDCi=S|OvCW(59tD@}1qE=xxlX$;QYZ#T(fF8lpi
znVG@SO(^L!I#M^T--3tD)(nFWH>wlz={ZM3;T%&1JMZkNZS)BqZ@lK~aaXb}d;(H_
zJo>|{=kIqV&ebP(w>@B#LE;p}2rpwy^QqQBV(XTm(9oR-y`E1-sv7Z`NW4i?7>SwQ
zSEZ`{Kid?rq`((Xz$DlEsUOSW;Iacm$kw8{q(#ph3sB%1-~YCE^5lfUo9sAahQYn=
zh;vV@W^115pvY-}Sr>B!GuHbK4sN+p7$tz897)sH;yWFULF9gg_WQ33&>qM9?PNdW
z8~XS=@q9t@kV9X6tODeXJ-lW=O;K7EPI+T)2fy#nU)aDdJiTam-F8!GL8Y^?fOmO{
zu#F`UKwHAYn#(K+E|PvadqLFUnAo6C5YAv*J4+Oi>*Cux^2a)E1|45;#a#<>i0TSn
zKLKgM928L>J?C(7qL{5UC5lJ?;?PPe<}K#YVZS0uHESb)*G1bC^Cw(@)+~ruPH@H|
zMd8K8i$W?^!To5cG3*YbbLGUg;w}Z0e+cUr*jUtqLU&*iMR|j%20^RCSc`}2^w?R{
z<P7-~So9d51wEW{y%)Zf&#d;^9E{o~P(avp95-#UNhd_S(E_BF1<jcf|8~gF2D2Wa
zr<h{=zPVZY<0pXS(f6W1r#uru6eb$$%8vs>IqrCBo#B`d1BlYMei86r?Ze9+-BEzA
zzu%^P!4s!o3$_ag$;o`3@}Zz7@EKNsCAq10C{;a{zl*D^Wmm9unhX5;-+RD`p0wvF
z%sL#|gIeHIimLfXWxc$W{K??&T{e!JQCiU8^}vI7z()&tz1p~!f+(NX4N*PHMS?Sz
zbO3`MWpVeRO;ciImy}2BF5rQWvOG+64&`IXFjq44+y%_GNwApuS>da(k;;YzuMF~3
z(vRskqY5<fmtLTh<i>i%#oCP$OA_FKQApskJbWs7<QcOJaNujTZlp(N%@d*v2oT+o
ziAhH3PqQQkmy;!XpQ&d9jFdvYXscM(>kC77IV;{ui#E=HH1zK`c1y)TX~0VgkT>*#
zdmha!V;xIT<*^MMdM<@OxOx*_m+)5e20APx!kDGGMj=W?jAr~cX*>o8ww_AmnXVC}
z`Xlg=0F6wnpT`+>7vrHeqF}bMS@X1e+A2a2B>l~ip~=EpF|*`;Q}Cs6M}GK6IU<$t
zVAdnaSP88k^{cu*7mo8^pVyO^>{zKaVg&3okAdfFhr+O$MZt2v$G!$e`S|?_JOz*U
zr61I_&~(A>^w?&)Ua-{a-k&A8Aq;bM$POO={J7+S>sUP``lgZG0i8v4;u>SJda|q^
z35pd!seeth9)3D`UOQBUWQgWGWCCj>s_Kp8G0t=$-fqBZ;iAzJ!t=LejlQWv$zaov
zGgmf#kle5HUkHl=XbXz<YH|cLX!BM(mQeG4TYNhg^)t$^|B9a<%@}d(HeqXO=VoW=
zon)=Zt{~vg85C!4H2s)Q${^l^Wn-l7w{sj)#x9Zd>ov;!Wx40Ig-Rn)zl3tr4>b9;
z^Xm>yd{I8fzGT>!j<cCwt2(6Qqh~bLgVdstw5^%?(^oO0fIlbH0uI{c!W+_p8=l9y
zwnnd9eU$7%+#FdssjL2`&Pb+YsBP-oQ@fO>IIUU2V`~2~ReTx-9>4rhrcYRo@GhEU
z<~CzM4DY+97qJd(f$rR|cV|@YtM$O!`qR%$aFPz7UVsk0HIELH;bm%M^F(Qv-ITa1
zH9-#N4|$ufJZ(j(J~68(Igbc3bElmY;c?@X43XMTE5CDLd1N^$FW%h&V`J#YG~R2;
zaF<Vas596v6S!Dyw4h(zb1-JkrbJ>WgF>N9FOf*%i<w%d@oSZK%-!_|?B`qRUYD*w
zWu0T_Z-)o7Z}ga!gi*R3sL_(e*0e%*K?DN0&vh)P`x&z<n8WXmY<lgU&nZ+k)N}_L
z7{OWMSuu=~f~c2IR)WdgP?7U|k%LM+&C%+j(!`BP)*NEV{YxYY+6NmpUCAMkEZ%N5
z=C#Cg=B?SUd}10u!O-7u0Jlf5Xd@8;?+CRv6QDEmjC=ui-N6`KDU{-*VBd*V6>D^9
z>K?5k4zJbbqB5Twjj}?<N;Dq*;BzcQ1=q?R4HggWzNWy_#Hw<|HMO;f!<~S74!D&O
z+%d!Xt{~l>J=hxY32GbTjI9HO0a&*_1$pooDDZ_OOz)AdqYyMqqpVKNpufOd_V15!
zzB&adrOYrx6Nj}KWA4K~=Js++q}=g)pl!LC9jfO5Yw5-;-HieqGf)u%?+43}Ay3w-
zz-gmV7V9W0MVs}a9g5Nw48FuUA}pIuYmFb<*U0^2#X@cYXmTefM5)RDdZFxZmVp8u
zS}H4K&1<kE5-YLe0x&ujqnM=2sUh_tZPsXHS4y(P8T7kd^R-_J2rpZ}<vK-TUpX%s
z_ll4w__1!1jZ$KZJ^rrC1a7Ea54pLiXajB~j#hwyFy#=T(?j*$4qWPIWmxv$-{}ko
zwh%RWh2`&+RKrW0sv$BPq?r}d55l2n)?0~2KO+Jyfc}IaJ&{@}+eKxuYFQ?9YP(f0
z0DQKMWQ~S<CqyB2aF#I&#T%<N8yh$SB)XoTmf}!qJ%+oT&?ru|!W4wfVC52x7$?TY
z03%2ll^3{!y(#6>!e1o$CS^@uxE-(a!n+=XNie^TL8Nnh3l4-arf;`gri6~#St>k3
zpkeC;8Wg^ZcFp+a(H4mmq*_<V>#AKbq_s0)0UJ<F9G`qnqPmkRX>v1dslp!HvT)`f
zz#BxH-m!z5v9(O;w=~JVJ;m*?`}F?lbGEC!(Fal|Vo@I-01SS?b5C5zxw{a?*=O3l
z%QK3nEjy*w-5z5;yX*#Tz}YE=yuGLbNS`iR=pdI0Tcgl!mvLHW{~-O8DepPnmODUh
z4atEQ&k3)3!y{u2U!}nbCe}_S5*9g&MRFTOfj=1Wva2B|;<jb&UVG=_juQ{2>ZcK$
zv1fAUH>Z}kO}b)ko_y2~(sM2B)4h%8+oiF1Y7RjG5QE)0%X%@?krZoy0=M9Kjn})3
zR3Bz#4xFMK=RMKE@e{%A%T`OslaKDRKe>mw_S-qK$OStsUNEkY98L6#y>2v4Gy44@
z7fj9;6|27pr)Fc$$QO0F5vpM<b@_L;q#Ch}svGnEv1U3G!jTAEt1YeqQON#b45Sqt
zzt82Me9q`Dllfey4xJ{ocK-%ip@-;1K(<!HuxaK<y%qo)IzxG3nq@&;)5z?4KR~5n
z!OMu-F#YFyIxczhnDK<;7fWa)FaGmbvAakIG7a4Q%f2pZ5Ms%0nV=0XfOI_G9HPW5
zjVh486PZv+^l6gur;)1E;s%5xz5L<2xnXAHIczoa4#Y@3`cvumG@q4ND7&q9N$`S-
zWE%NL-)8;u4wPhYdOI*!CTpdj!N{K9`b<$RmV+7OoLCJHH^iy<LxlTe7x%D*DDGCS
z2+a=oS-Ba64L#m6Csjsw{vo5=2{l<@YM5*riqWLqGcL!aPo)hcy}GPbp5@wYY&-?~
z1$J1OyCutZ>8mBZu@BA+Bzs3SED$6*n5HK>TknM?eOtUwF63eU9?S~Imd)a%A`)u)
zr`J1WS!uCKno^xSBkI94@dOgLpZ-z{%v#LZM%4&G3^@S}DpV14Ejp_oO@(<RgIO%~
zUwXifU`0Gx=QIG1IrcT4)}&g^Wv8D{6KzL_L{eJSUgC}FiWWQ7(jR45V$iDZ;xG7S
zaRQJpoe0Uw)lvGg74aSjKBm8EI-Un8I_4`%*l~E{IpO@ZA^DiGx+*k00TZHUJjlxn
zfB<$=c#)uTCY=+Eo)(tvyd#`q2{8np{ASPb*K>#|!%};BP(%xQ#_8#c{US%KRcH_y
z50a#}GDs|T(qM@7>&({&)hHkiRFV1ux}f%B5+!T#L&tI;;%q_*Lno1Z4ikZo7s*`f
zo6~BfGDe*QENfFADpl>iyYifg^sRZgAgD;#ET43qK%$oT$S~Jpa4pFiBPI@QX<+n3
z3SDdKTUW`jjpAY$A8}#~RG)iqrj@HfWBLFTTn>nmZSakf8L*P)*ShRrR`=tHO!P+E
z;;$vTqQMHjVSU%WjtY&8eNDOKX@m{b!;o?PI|vo*ZU?{T;YWxm*)>UiOg~cQp+7nW
zBqp0)K?h2|216}VUR%nDnqqUuCQz5dTZ|5HvK1wXhKGdm9uUzh*|}ZtuT~WpaL}ZE
z>qgrc%aRFT!l*#2)0C+BTZTu5^OKKgSZ`@chfm0S=E%y~5N#DF`{N?$H2D0;vDhZ*
zhdzYk)b&KTez`!y8S2MN9y@<vD}VwIi<-ecc-16OnbamFxexko0sLav4D<XwmIwl%
z;g*#EJdZ0qACm?$b}$YyhA__ye$kWHanZ)=P(g`RHlFbt4BbHWAf3(7LRU{V#F8>O
zg>9wjETy%N`l8NT1{4q<JV>K8L6tpQUsIsj+*9dA;B*QX!gEKFRa+9bLd1zTpb5-h
z#^>*GM{a`=znJ?f{}N4nuR?EbH&ac~uLjf9fwyyb^#$X|TUbjo`|PZ^?!0MH$UKmg
zis0plpNEM}gN-Trw{W}g>i-;HOg4mHwf|w6x|P>ZAl4HHUU3KeyMgl>db-ATZrke5
z$mVDyU^JSBsC5YTE;SrNQQW`{Iv>EV%ir_lboXLV#_T}?p^@F}k|-{;94{Fu;Zm7G
zYrnt^7&TcZ)ScgssYkeiLO<2ALE+WOfskGaZ)%Q6U#fM@g}QSQ2syCvXnDuW#I|tG
z_~-k!Ivk-MRr(Oq?NWOglreyJdu`4^_!q>8pQ7m^!y`VPS$FRfv9;srl;D<~@IglB
zNj?FE>kwHp=Nq$L;lf?<T0@N$IW!M?fK?3X_%}zTo$U+#N<nkJlA~OwKMmDiKxF>f
z4=8h!<ge{-(K#bvh|08-Fi+U86RaxEA`DJvuRFf>C^{o(0k31rDJ)cDV06DLVk>2_
z)3u9`sAfrHg|<*gQES!^mc!Zx0(ureX1~Y#1Q)$V{gDS*+L+Mbo(K@Zy$sy^VG01V
zhLKIGs|@8w2(1WnVae85D7rLrZ=4fYT+eh-jgd2%!-p8*D>`;rCZ`NAQi_^O*h_Xs
z(orVn9&J$#UxSrBA2uCa7?fs7(<?x(d&2I8;q-d!+q+u9mtcownsENQQ?H6$ht_*c
zCs4m5;bx)~K1}y$X{_n0DO@$vj!;d;>$2qloDjsycb%3%#OkzU+m7oiCMzNw7`xw)
ztcBC8TPQ<xz7t7}sG3eGdK-)q{v7og1<ZE(h?<%&hlEDl4WC{Vl<go^F#Orz;6bMX
z1e#M(!QpesGYk&y#OlNp8h3o!oP+ISo)l;R@juIKKbnn#?*s~Xfr!W*pqr^)F>B$N
zef8G>cz28)TI3G)aJS2HJL)lo;AemQ$sLl*04h8XCl$GFEIA1p38Z|u4kDQ)0}rj>
z=P@J7N)92Rd(v0+#>ZNmmyN$yh&vgGO}4*<z*d!FU5~+E1;&008f`W`<rHQK7LVte
z4YbP5DM0!=zMC97xPUHVG3`0Xm$`j$S;MxHC}aQWA|O=D8>f<R1Oz-(Wa7(qSgH8A
ze`HaUjwi{1Gy?`wPti&gP0ENxCD2}U!QZU1r*iX~*>hY$D@W~qi4lvi7`N{;2{T%6
z(_d53-)-9EhG3Y(rGRHcbzwR{TB!9;9hG;ccP)@Z3{+JJ$zRC(<{Etoj5%;>4lj|;
z=+TDDe2Oe`3<h^y)Th`^{R<#u?Z3?5j^J3y+>GOFfKM+S`K0%m_262lCkw`!s|?y2
z+eO&~!%spD3iw+3H&~69knoQwlduayP@c1!(tu)5)ZnHzHl*^@Fy?1imEq{<R1qdI
zAnJbwFu{KB`B-hYa|_dh#{DV&#N#8Yz2=Eo!?!SWbGvX3G#0UrmNJihvYO&giD`7;
zhIWk*w~zC`^<=Ej0(0qP>?_+5ogA1iXY?opCX3xcS5RHX-Gyl9SWNmht$K=vEfM4G
z3b=2s4&=YK%AyK;doDfzACJV8Y~AxBRW2x7BjIo_tSMq;K{IMHav-+{=Ofv*zCk>&
zX2Do)f4a54v#CLpYpbuLRiBS5iechWO*B}Df|*ALc7Pop*Js6;$B{uuF}eD6443+{
z_BACb*r|V>c0$+KCXo4bF}de?vjdD=<NItbl+n}!Br<v!-D)sY43zR~kql{9m#^a}
zh02^kmxTw-p}9HEny4<iN$DY96x<?R1U0eQ`YD}D4#s6^bd{bA$?rP9zWNQ3F38R`
z2rySwe;M#<TTD@BGNf}&F+6odhDNZ<#JbMIWWR;7_~L8Jm`H~N=VIVw=vkC6$2E-N
zT!7!GMY!jLM>&omAE;Mf@YbfzUVMzS-&hn$FP8W{ZM~xgd%suBtJuSBePsynovq0*
za{fwApJwQ-QDGx|!Q==twg{U<GRJ4*HxirKAG$cAR}In(=8Tk*{VfQjDPvK_jA@CX
zgO54UzhME2TMY$|1j^xl(Q?uyWfW6B8_9jD_My>GeVw@&(&K4uVp}NZ`15$F)lJPo
z#pl6fDNi@1tk>sCsrwqT6grc=t;a4}eECHDJHa{jlu&h>cnOj2jrQ00bu<*9WF-2w
zT=v5NXUZarfT{D9hESy4j#Q2;0d3_HzJ15`bY|HmlhUDq>$Sk_unDZDM@g4WfM}ov
znvmNhf??KWH*N`M`ySCkKJy?#p19?C#(p2qRyYQsfgc`iVm=@wyPJfvQu^DY2z_Gf
zzwkr?@?#j3%uJ4MlXmVW>A6qF?BA?HsUQ@z-pup0w3I&00DMteZ@yk!25J1dqb$QQ
zJVTS5!q-s7xU$7u2yG>q200WidB-YN$rpk@_Hyvux#c&F{s1SvKju(y(dfCEm~~L?
z)V$;YjkdxM@$HVaTS8i$7I9Q4?onaYS;c-KbaA1bzjbnGAr^fM0{DAxqe&PzkweEV
z(itOgg_bgo3JX;sZ%DGGH;r@}xFsJQ1BVLg0z)Nlv63)|I-1f|_!sA|lL^(Z4I$tp
z$cNEFXx&H2;X(B*3jbORnpzcp>-v>;=A0_2kPxaQzT@_;fKvw+T>$=x9mO0dxTMIn
zzwadD9W6YsTx{oav{BO4{0Z@o;e*Wo5k5%3&d5ga?-vxEu$iT!kv##Ou%(`((H|oN
z8$%-~US22%M|&eZD=62sM5SKqb#{c#bLy55fKve$Ga)2Y3^37AAp4c3k^mu_AmZ6l
zKn+I2*O+d)?Fn}1Ur#Z3mfX=>6CKYa3NAig>qS-gI=|Drmd-p{B9=>`%|4&6GB!3@
z3|x4E^uCaO<G-yhj_1FzMsBiBoOOAdX9f3vP{w5kU%6iPM?QX&;A_M8ccWa$Jx}k7
zv=xf;JP|e(dUHRWdTtcgi_Ir*k`}aQ51(^S+LR8Z9QS)hT>7k(5@~rQRGH55>9yeu
z?aHN2tDmc`z<BXwA&il#{bguuQQO21_C|6!1Jy>ZSNzOPKScL(DNdnHYqm+|$jA!*
zOsP;dzva>Hn3%_YeM|V&CGC|0*I(ZDU2?M0(3v$o4B#hDe%^T2&wCZd;j^0Qul}K@
zf4w<vbDWd(uwIi)`=m3~PAl2I%R)k3xop1JbWQ^E=~IPi=C-=sl0LF%^`U-Np-xU6
zU|e6|Un(~uN6`{9*f$4|pW`%8t4lnkLA#<#(L-TYN83}<Q$mFz2`L;&13UefU5!PT
zd1KDEmJoSj#dp?!mXTOLrM<Oga8l{nf=vv-j%BTZ@B-ggPlE0oh@E(&C(oN8s$V<P
zp%ai6gKH<wLqd1lX-{Qi9f+%nSme@;&d$8bV<$L%Kr-Hr?kt$awJ$?4_ylHpqQ*td
zp6&=ejo~Ocj)z&}&N^@LFRX!Uv&7JiHPg?5k746G+D>TIq#N<tWGIO<zLwJ)CcP%X
zISI_HF^wS;UWqB#ZG|!K*%nVW%u`2Wa8jY_6V-`&KHXcb%C|Wm_INy_Q2XWYuI2-H
zje1TOI{gN{S3iZ6M!lp)bRXtqyfDtmbY_$pjA!uEOFcY3V|P|;ilgR05@!^%J>wwa
zctC7w<($Oi;FQE<Pju?$ip1o^C639Fa};wc^B`g)<3RGB^xs47qnI9@2QXd98NJVO
zOb@Q3sqUmEu0N-@q$YbBU&TmCKD-^Sqxjk#2hqnf5?{>qyBTI=-}^k1u^+u=t}0PD
z>b)h+6K3ihgeQM`1pelKeLh@P&7;5@yW`K4hTT8V2R^F|xpC)*^FCCyc2eD*Tcm%0
zt;1fe{v$?~f5Q0hd`|*`{|;k1KSJj{>bhXT0e^=D(62-zGeF|k8IrR6vgR>AC-YG-
z=zYPju6N_1QpLv^JWC#U8{;jHox`K~dot(Eo3s4`?egj`Zq1>k;l<=j%C^?hkSF_x
z&xaGQ<M#8#dy_|QxXU4#vnKWBb8o&)(7QGp_~a=aH6QZCFRShEBHidVuiNc?Sf#Su
z`&=4rnOnz|!T8Yq<;&IKb^MTVX6ftNp_t@PM~KPq#bO|}Pi(c>I2&*7>yWQ{#%J|&
zS=C1^797N3j<irxmiDxb%Xe!OoikKgVkVW%uFL~;ujb(x>NK|N<5B$OPwT&E&@As*
zv{}WsncSN~^~JUK`&~W?&EI4k4VJXU^Sy#ubF=oPoZZmrzq_K>^Hu7<$IkMf4?hnU
zGrX~%W`nQp&DP)S*1KO#b06ht(u^@*yWaQHVw09P6WgIQHv_9%rNfLG=uKH2s8XPU
zP=X?$#V96V>7mrYaiCat8g~6*gJ2>u$j)IRfFknvdk=IKSLE*_eDj51g(qoPFsK=O
zBT#>$ZDE;!f?La>)3vl3`I_@V5R~(!u=@^y=`By)5&_8rEmRla2kchL6{?ndFb%*n
z9Bkk$TctnzJ+O6$oeoMQwj?sb&Pi=2#q^8BTF{;PvBZBI`u{kTtZ~V5pgS`8ad`Z1
zhyIJc|8nT?V-Wd&99I92Lso-`S5}k%zr*!7hAcD<hG1_ErlMCza;Y#6L+qg$wVE#!
z59p<Af1Midwqnq8ncQOCPt3=<FPN>LQRM7LlL3=V4rcTUCXpl1bIFoqC$b-luslN<
z^#SPQ9R2z6849|isqAq~&K#qd)1mPhwaa2tBlBWY1F;#qn_^Ql*FP2=<Cr?QMloAC
zMpKU@{yp@I)Z~JO)Z|8D6!SygJ~&I>zDWB=tHfyP%a28xZw1J4K74KQABaE4v>xT_
z{6v~VQR+uDomtcYy!2$d{J{)_aHpR1<_M$!IS|}$oO^r|8xPE3+h6nTzS2q=#qV1_
z9kkN@kXjG9&l8WJp*k{B573yF=b--xk@cS-{@=}t{~w5*{|DmXOu_$#7-kp{bW7;7
z?ZvdOSmr+3yJYZTeAH>8YR&7lK2(LR<C)>Pbk@!nom~28_vqvjW^H{|+lli@&*#4V
zdi8ex8vZjSUM@|aHMy7kOjcdc2O+sUnw&mHpV^9U%ib>%y?9RV7d}H8Wr_yJf*K#2
z-P&;Nlv(iauB>(+bzJYR&*!$O#ji9=DGwXdA@sZZlT?l+@QxJf&;2sT-Z)(p_i7fn
zQ0cklL&S3&BG7gvXe!h1-X~<WrP{ijj1r5Fm%mgVHt_@}Ot-o+O?*5to(vk)OND0U
zvyV?BH#UXBi%ahn_&Ung;ZNYa*2hmBZ)Qpy8??)(uKU|!UiYs~7WejNJK<fux!f8H
zEc&C?==VI=8ok$Gt9jMUI+e?d)I_{bzFsFq$H%|MHbJVa`<CU&2RYR-nbS#8C4+v4
z{T>P;K}iC`0IdXu1I4yov+WK34JIrd?*t|cC@hzwYfn#QS?VU-Gg}6hf0Tq30h_8j
z1Qr1n5sV4gzp)rJSyi>R-7W?MQ9fG)yL*R+(ezL^=8N3N^6SiFkNskia@k@hq850n
zl?}W}lia(v6Siib(|&>UnpAr55vk>fgkhd!BfL!ymQ+ugO29GLglwgKq80s~X%ge!
zVH7Wo#*MAlyeF8UW1aK|B;?QW9KqIeuo=Ue&M@Se&O{bvWHqBJM1Dz{d4!i{HHzU9
zUWp;tU5PpG`SyeHdxSM2w_f*?@(Hbi542nN^CA1^Bq+=O(PUx#jmecx)koTMDH}1{
z>1al?W(p_uM`W({QmQQ7IkO|S!xWr$#sN&Uw7s<7&om}FwQ@*evOz9wKcr30K`cYg
zfh=Ml>_W~V*d3pd`(yK86Xd}WiAfsWzlZ7*GHRO>GJ5I$4bolw-?QS+<(|aUK=Vfd
zQWE+9gYo(Q#W;}6EOG~4e56HbZwivXMPF=v5SW7u2u=w1@eJJ3DXGAoep<<2Of@y{
zeREA0HShWx-RmUM!OLOqtTglspjrIU`X4c}{U=6NCWildiJGY>Ws^^j(s57S5h8x+
zH{?v=M->+M8uPP4H7{G-`ssi6OCAVfB=`M%`CdCtyu7m^BxcvhnEQHlYO*;wGJ9cj
z>FwTwQu}fxGkEOn^ttZY+Pbca6W5B1M@?h>?e6@ALOZx>ZAEOO=-hIXbv~20di~6{
z>@w)lC$9Fr=rczb*Y4fe=MB58K+7s~!l&_lJ{RZXwC288?)9q)r&8x@?F+NN>zTA-
zEA4dWS62?2G!ZW`V#5wRSA^6uFO(Bqp<_@-<^0uOz{stOje-=dwJjDao(wKjjfYE0
znzN#`DVv9iDC;={o2}tdyyeTP`y>el2fs8stgX^+E}JLr1!vQ>&X!L}Tk?&Jw=Z2p
zc~P52q{hwG$Bm8bv3%5eI?3y^o2M;}+*2((nZ$dH_qsa4)vL<p1bfmXq1ro3jqm$)
zf7JEOR4mXaES{E-B=xEeqdZ(`v@62G{wjq07+XSTP+*t*5?um@m=q4wqZo`A%qQCw
zRCyG8+Jc`aF9>9l|14hH+pfaa6OqTqh-G5#L@<_tq-6SY*b2DrY9@;#E$vVYdfFx!
zX8?=zPzl?jG6@^w*02H`om`FYCNO-kRG07OL*K#lWGk>%?%dZ9$lwAUUsU@-4~-zp
z^T5P15p`t{%b?uy_K@c}acB&5lHY{d53Pizyo<*W!mW@47ky5k8e@b^^up{ru(}Qf
z*ZiJh-{*V_8~r8};o03ne#o-DhGZu@>J-N7C-2WeH*kdubmJ|U{Jm$GV$$29njr&t
z$vH)7eeme7SgdF35>iEj>yGnW)VQ3!>B6J<klP(N-i^JL1$|A|ujobgkA=+spM}iK
z!TetrvQ_-Q3K=2z`UC2kP~JT!CE8O7B6L7tOmDp`RJoJvi9`qr3^RuH+nWbZbWsUT
zAlhULvl}>mwEDR-yZjBr`(t7iyurza<RyK3cqm=+56wH2<1<taf8^>@P-nXORj$GO
zr1XX3Dv#Q&^teT}t|AXw<*xjh4dhmT5%|l45vY&gRrcH6TH@L<1#iOUVi?|0Dr{Z0
zz){$jui3bIcg;<zgEg>5C96}AL4!8Es$Q?@m8Q877~e(URR_I(S;D5Ry^PxGv)|ym
z#=1b`{PCKE0-EQF;d*NN{mTKUjj$i(V~~`OdVQ(aHnR#pc7%u85K4xX2-o*{Ef5-a
z#%&eH29|CtSIB|H64creMK8`HXh*_*M4?cuTh5|SYwn&SUV*+|p}rYD_moMM{=}e5
z<p(DgU8L6Xp>w7h94@-U!QVZNgOok$gTNHx>AzZ+84sX${(BsEaqNGz5p7}|_AuO}
zF0R%PL`M8o32Z`<5B#rwXOHgSvyATegmS+Ox5xE<_uWBR<Wu~y;aIZA+x^vN*YTT=
z$Pp{*Q|OP~1*N`zh6>5RiQ2@8R6DPO{<4YCg}L1pMZ{76vaJn%H_6qemzbeV#3p`o
zhBb{Xp*ZaGr_~u)`NlNO*NaU5yvFV)Fo+~h$UoK($A8ujD+lX;Uq3%zgnRS|oyRJ!
z@QG`{gZLzVP=u?HvFw%_iRO(eqLOS#VzLk+U+<6D$Q8;w2vS+w6CJnE4$tRKT(e5e
z&B%`<tLZ#0$z2Wc7tdEHo12GPZf2HT+!`9I=G;7dwaZ=9&Ca^dt7O{^k2gCb=}p|)
zR9to5WSH-A&3^v87hjuSnF&$~(7dZUXZy0v$Z&TaJUF<g!wyIMPpF0o55rH#Z5G)n
zt_K-hx|_;s%eoN`%$?$T^Ku)x*5zMiA(y$dNO4L|9GcrFYOGhhF&&@W@J*`?A++kW
zs}Ga7Ss`EX%)Ec*le_A?Hoa@!tLHXmpX@ntX$<M{DwQkhz*S}^zdI~`rGK&SN8W$A
z)k)5^-?WWY)g8}QHGEcm*38D;qiUcHQq@^KUP{_$*YQPcEZDu`z!&Z(0>kXF01C77
z=cN7Ri2>jbyP+fluK~z`9AO7oKxhR=@$DOwcO`K0{nMt2V9p32nj8#JmOG>D*V_t2
zFX%3*^#aqUUt{1HD+YuMiBW1oqHF3n8&01wq$wP@8v#TQ-;Y^<jU4Q7h7%LA3uA7`
z4U;SeaK@U^{&?$da(~-4?$iEt2BQL9U^fyImrtFHmreYXhz52$F>Eog1q2IA8>(OC
zkQr@%iWXFm2D%G+gf?&-tbc74rTW$tD6!2p+$g%uCW3qVju!adWv!21%u5DmKDH9m
zc;J-bSKWb$^p<$bu%Vp#oVd5K54zVl94g~PvY2h>qjG|+Ku5~pq4|R0`WuJPw|GH<
z!c??0S#c)mC@Y{{ycBNQZ@StUXnEtjm&XF}V~+8qQ0Ycu+Fyqg&ij<)rEq>Hd8wB!
z6TCYD#yeH$h8mfz`(!+i4)WXbB6od`g3jyu36h3d9a$B@$-?ISVj(j|h)+Fm|7dVs
z0`U{v?lNHDpCqu+|EB~tCXWAJg3`Yvpmh8n3C^NGN`Z#Qf_pVqN@a`MgbbUaF*yiu
zpU>yF#ICDP09bs{*(Mjg8=tOPGK!m4lO&!87n2!n4mzn~PYw<+S66pcocNMl+}H}~
z?ioBhKN8B9jDI9h=`>E>=#G{*tz}y<)pe1Z-On~;`FpK=w!iX}!gcP%U$(BoOL6V)
zbg}hs*>|NIHw{^vPtd#?{lpqH%1CnCPg&)fo>y@7^i!s2JJy_*SS~j>`5^W;#h^iq
zo)uxy5<hF<Ug+uGzTGpY^p|Zq38dAXre|`)eikddAV})GtvyJ+_g>i~tfoG`x^7^X
z=yI#ls#!TV<vxCOaA2^0PVFS#eYr79w$&0SDXFB-+bN`a&wHm39D0t<)oFqiReq;S
zm<~F#$s<b~I(P<Ase_fm&?GTHZRl(C7?zaMHwV=;u;=*?3SdEpBOgMcft&e{k_Zyh
z&Ism)sB5~^wh4oddMYY@u0Eh&a%<{hiWpG3!Tc(P1LY>E@G}X7-udgi)5D;*d|X$r
z0i7u?`g2=vfPojo#Ji)1{dpf|e-Hm_gwn=h2l7}TkR2Kj{)q%xS4{H+c?x)$N%vU5
zIBK7J1V<fIE1-`sJn#(4RoVh~*loXce+C!jhA4v#vA3H*r*J>Mc)|{9PgM7A_bJie
z8vCuVz+%D=x$g#7UItWJlt<9^HzzSH&nzhrd;DK^M0Pm=?_5HvQFQmAjF*|114VE?
zLhmSE$Wup`QufmUH>zES{DSF7C@<@UiDO|QUN)+yR=LpWNO@kiu)4e)5v>j`@2mTQ
zuyyXzJ64&r?IW@9YPY#Ni!8^J`vMbQ8O}Smx!UP?Y0s>Ojbkx7&#a#;$GNp*v46ew
z>^2uP9Z&0(#dvj}0GVAZ`NnNjf1cT@)Ae4sZ`H@;?)6yau)PjG%hPm~+mF25aP)?U
z-%hOD8tpMM+zlI{J+?!4dtV~^k7yYF6AcGF{r?r3bxwqibCsH(zj}LF_(<ewVSoz^
zfjJ(4rEYGjq6Q8i{~8H4!M{F7*IT4p#R^6&+O~3W;RL&6d-G0tcKbvz-3x{3+_vz^
zXl1Z$H-+q7Rc*Q)YvIUj8(bLpXz_TX@)dhMW611vxtZb*?LKgj@<nW}w?;2lM92U3
z#2w1;=iU3F{Ng%y9(Ex$nLggVWR748Jt*FNNQ&dtM2AH2@%Lx_L@wU1!S9B86cSJM
zeOzIC#~orvyN5<)=#KNi9P)0vL7diAp`dxo<<japX<jGgsd>HqbPa0{xi!-Hm8}0o
zO@k*h&t><aIpHJel@WZawDB!QL&K8=AJPng^&71`@nm13Eam9CYg35f+jdA}YW|%N
zN@_dI&FP&N4}P6idi|CKO@dNIzjd`DjQz={(rj2GOQ)*UX)ICdZPs0`GOTcpQaRXn
z4Tu6z5o86)7jy@}wHt5;5M(P4Fel60EgZZCtcI`$aApA@u~8iZV4;U`2SApGs7T!q
zV}V|?KR~2ZUNg5jfs!@}m62*{rbLf0P^klS;8$?{!n){9O^z-#m^r{qKP^BD3jnzE
zR4xC>Ud_pjvIGtC4xno`=|jo}6MPzJjzEptysvxiR2tX<1g`~_EOAAB5#S?MDvK4p
zO1@26QGZvKd1Y1c#%FUTZSahGZj>U4Dga$6r8VUdu?k9Zpbf=QyiJe1`W7@MiA0_)
zb6k{q6OI0!W+g+>MveFZ<dH0q$Zf3Zf?S2a83G+Uc$>wjwr`)hT?D7PLqum9M}lEa
zTIKIo=qSo<2gz5OYf!e%BUitaX>q^w+^#v@YQ3z{PD<m#QIg=sI6#Kf7AV8nFoQBd
zbxgrm?tBooekSoTlY93G>z6N4jFFm4)ZGa+KcCSwj044tYQ3Pa>*}Fjs5<Rf88q3+
zz%u9TU~J=FxM%AUxC;#%$&!|>HWIgKchqAhmEse7v|T{pBI<6y_if?IH_--_%<L_3
z*B~7SD=Ln!T9cfz-n!Ad5jeNCN%rbdcC^-$3|hmm!#fJ!N0@Y_aNDwppVv-L-E6Dh
zT@zo=N>AO*tKVBj-|w97d?Od%Id$IwQ(y38?|ccX--L$WA!P5}3~ot99h0IzOF)*d
z)_uyhPV7y$w7K~I#ok-SMzW<%!e(eQGc((5X1Af;<~B1kGc&ZAnVFdx+RV(%%*^&z
z_s+b#qi<X9&Pe-brTvjoR@RZKq>S@qWW*Ea2wtLfpt_I*GHqnm_kQ?j-icao-(UAH
zMoC@sSE`)g|8Z9Tue(PPa<KgSnz1F|Q3!ac==jImF;N8Ec3CJf7L*+Ps*+{Hiuo65
zb~z9v(!WSy8rI5sDP-*7by^BL0;4iNewK0Lx%N73>-@=Z#Ji!%TKQWGFSxR@;-*Gq
zCHE%xVyn{%-sRlx6#L{wdF7~7!|}r*j?9L``1VX|$USd~jrHqAoFCuGOaFUji;)vY
zWZv`#0u0>4=!eHu`6cv-=zLOfZoRe@<|Dv<8LL=Wfn3m3e-M@Y3SlpscM#XP-f0dQ
zldna@$e3gHIIgO6w~E_A_QT7IVDh5**vYe6qx3m$6L+$gyQCfZ072+9(DSjdvdiSw
zeKKkGcq?1zSf@d*V)Z$8!~gYsy5#v$-z!>cLq_^nJzIJCu{wKZV+2m)dKyfkTGd=_
zrLyno3WNK1IW1=t=ku1<2mYMM8^vmtA8bV6k|6@&0zJM%kb<D$J`4D-0tdDl_Jp^-
zVBzu9r(of@;XI<h^L65n<ZR)h{0+gs_fw%^nq&K6nKPO(ni;u-Zx>SC^s^}58gmB&
zce2jZ6KT-;FeSo|f^iUQ=n8ibMv7?T*|&p8W2$e(+ski-@Age{HbFGv21T`Qb@owl
zY5^&qk(#9j+as};=MbhNi^jd5se$3ZEk<Q`+q{!@|8jh8DSTD&tSp&$A^7Mk`Hb?#
z=xMV-n#nN2gIE)09^ffiGk1xPSsYM5t<i<8s*7f-H$>8(n?w&^{Jqye#84Z+bYe`U
zzhl6|&tk9xeow|Q2GVH&u5U6SK5#`@hAHovy4mMKUYp9~a8S<weKY-jaO<!<D>Dah
zG8I?%5n!k4?YC^dkcFIePXQ2+saOunnfrMsRE>pO%9(RD7EF&CbhGG;YIc~7>UZp6
zY57n@J-##0@a?wL?=ZipU1NEvU8isWzHQoJzEi)(;;MKGU#@t9-7J5Syze(Y0$g@L
z2aE(9PrsgEy@>dPWQh3WD*>Zmu2Y_Nn6H`nu%Jf8{8hhP6V##Z@Mrc4z7?guRjQN@
zUy6Nm?DMvvzCh~@3f;90fDw&5LptavO+*QmCdKW0vhl#Z88^gydzR70;F`D1yySko
z2Ijh_Fqjt6t~@pmnHR!)19xgzQvT!QWd2_txy<Ze|LsSvr(CyH4<l^HAx%3dGOqaS
zR{=rA0=VKnT}I$kC%TYP;^}<xRZwq!iYe(-#Gvm^Z*Omzv)!YIL@sd$0o?{;x={_L
z6~A?!W0WlzRj{TLi__?Gg_jji!9CPS;Fo^ew5VFvn|%k{FtIfyb1ELpcIKf0kZyY&
zm%hvFp`u?R;b!)-wUZzhsgb2Eh)-1T#ZXaUTVkD43nNG1!uk`On&k}|sWZ!EsxC=T
zx)iNH)mGZ1IXLHOL%&f^uQ7Xv#y;$&A}qFc+vUOZ#)9?mqMZ(CbsQkUI4k8UzQFbl
zB17Aml%4oV!`mbl64nqJtxU2pAoUb&xSw6C5%uv!S#kY}QBiIYNIPY&4Y);(rIo-k
zd#)O=#w*%MM9zjO(iC~yOXN7{>vm<6QT?hak2|Tw*<Ux-AIDMX!uj}EN7bK_hBY|N
z)e+-{&c10`)kLXgKHQ;Q!@g@l8Sze|P-v(5aXY#AE=s>@c!r^*z4!3PyZ8r>oSF6C
zzl)Z!HCFxr-Nu=}-!V8I7EEY`Oz?;4%nw+pTtr39Zaz5)9rXcN*Vh&Le6s}L-L_Be
zx6SS4*^+QJjd{$9MAV6;vTnxk#>)iqx@GB_93fMLVBF{Y2XhXre(`$7P4qUwFKsq9
zM5^32>BWOnG+MvjMBX?VyS9wyjvC8Oi}8ice}xwZG4wRTyA}IO^lXv_<RY+i{Ir7`
zj+Hw>wvM${JX}2z<5BZ<`ML--Jbnaeb{R=ZWz&6z`76QE3cs_t{T}ceY@r5NeZ~|I
z3u}IZ;2}|gc}O>3Zo3xJ72wsHjx2%5T;XwX9z_;Eb50O<f_OJlMx{vdiKM+enEGoL
z(S+oOX=1}?aBJWBC@7!#Ez673kUXVvUutb(3al<^$mueEJ}6?v7O)6hh}0?~Ka`aa
zB7QLPr@9kZFb*_e9$=)qPYUjM_Nkd_qh(%M7mjY_?Phv_4jRvk{qc_d!ANEKPacYL
zqgG{%i0iL3Ya;m$-<S&X5r{|yk*UJ?=7g1<CC-QzkueQ)V!FEUapW3CKscD~jW}*q
z);#h3G#V+Ob|AlZWm}7q5K?Z8MAD+CDTTuwZQ}{}rRTlVyt7nr{1!;+_o`boYi`=>
zK)<KdHCiXH*e<_g2mcgG5`Dr^&|FX6Vkvd|$)z*ITbk3DuC8n5c?YQW@%%b&^C!yK
zyo>|o`elaGY)e8iJz|h>#)b(qGm?5o&;>EU@+EzYsuy(F+^~{-8ZW@99hGXag)DEi
z-Cm3sMOT>t1mFE+UnG+NTx3Z1tQGqwxFP9iali<AeMeUk+|-qlzw*fS4DkL$WIOe=
z_)(jj57j-^j}evcv+MLE+51(Pw^&6ZX^D(NUT`7a1w>SX#No>bVgv+q2k8}vRZ|3q
z@PHixkxA%zwA9lXC%e~3Z$6CsY92}W#dxF<_3`j+%TsvVKhUqk4t1;|Svf~x`a)(=
zh>Ur+Bl`PCPWw`EE{&aFj}rkS`H7r5_G)frT4WC>wz&p#w*9J35%ZXw<S)LgLD)Qq
z_ul_(ZS19eY=%j@#HxtGdq&6^d~Nz882(`9vHmC1SH9coZ-#`nA(Ve7pij6}7)tCb
zmXF(5k!sSg2u7}&uM_Pn^x?(JJf*128Nd^@+3w2RXw1(C@1@3Oe3fTR#g}6aq3%=V
zovCKcHkzha+<j*aL>515^+?TUA&;AlVc#+NSa_0!>uMdk`m?9cO>$&QEyWwPoa8|C
z>0En7-IC^$N1>_nYp3u{&53p7osD*L_7gs8{2h#iGV6gN*rc2|#!{*rJ8R>#YuI$C
zisS)(0@eAEX!>*pMnG<PsF%icQ}H?v<Fn1BTk+$;jLT_*ibv~sWs~#G(;nT1C3NFg
zWu{N7aZ9|w#tz^W#F<qKVQ7q^qZe=Pc|vpq_qQF~OS<6b1>jfeROHqN1s~oN8W(da
zcD;^OOCH93Z+9f)bOsVuXk1q4b<yvt)waH{psAoGFvUZ<Xz~K0`J9HLW^yzi>E%uE
zHI?xfliR^$dVLZ-+;iH*@Ions(v+FVN&(79JMlKs9*wmlxRGSHvi$HIAAZT^1<5#d
zv#6&7<yAs`#RL#U3!&N^&S5V79&=(UqBMPHoa)3z<9KG|<tZpelUNH<D$IEr3u8*b
zEf^7wap?y_%wmmW!?t~ENdujzj|n*h{fD`>@Mh&<T%o6CbISUfXCk(mQY{+nLSCUS
zIg$;%sy(-ZDHdY<;d3sf4xui_SApJ|gL>5q>k4iV4#3wZiV&Sz;;Z$^BK&T8vuOkK
zuM@8@&B~0oe?;D&oGZ5ffSdpS1Y0M;YMKC4<mmq?*vg6J{v}g!Tc1r?H6ROH4=~?u
zg!5YgIx5Ij=keD_JCk<$Ggc+ETg5)?H$~XlMbx#VE3xo1$~0IjGYR_aHxs$pgI7J&
zEz;bY^R_JIJGH6VzrXXF>~EYLFoW^zJj}mmI;_8T-=VyS^>r-ta#wh3BO8R3(Q(`*
zolZKg@<0LHDbm5#lA8yx^+OZv=6(kS`|`9>1du*pJK_Qn#U!EXC-qDdpdTBa>5$GV
zY=dlSw0(EK&?kjHyGJf^kL9&nI;}SP&K1#IoiS>0QK7Ktl5*`tslRvdUC2mFcCkBZ
zdc7Z|Pg&4Z6<W|ZUpy3?K8~k_q07<t>(!}#uPo`TD2cY?h^(RNAK~PqmX)w8_hmmz
z#@PEASY3NJ%_TLIBdcYfNh%31JjS-6I}}6RV^2>ikIEX%rZdggKM-Kq1E0|A3w27O
zAQ=hETAOZeP#ja?seGnlZ?BILg4ag<dFFls+#-j8{%*oRQoz4@T?D6hEJAuDIpKl7
zk~px7R(;O_s^uEaXE-4*i>)7fHhK7atV(Z4X{<{B!FyX;mqPn2TW0*@{#%~-`yCS}
z>tp&~-m<|*!XtChqDb41W`Pcl!?kL%@LsQPN$~cN-qgI`y$5z-vBP&#v2Phy26gZQ
z-R01+C6|7hFf(_p9?d?f9)v%W4r5|V4z!dHmbVjN!&<kc+DaW2A=zK7JPx(v3_v@q
z{Ql*F^ALfMaNlx<pphH}r%|jAiZPK+_qO)iFvUF4Ua}c}Yvc=m6LGo#pKY^=3vA_L
zQ~Mq7x`h|h)51>YLPQ%MwdN{O*ykH|#3f#Nk<=Z=BC@W)AA$Y{XN!&PKXA4bx~=|p
zCjMd3>Twe-p$UG0{(<ZCU(N&o=wR^guNmx4;?8Zz!jlYt0sVZ<!+k}nwv@M7%!8r7
z4rk?WpnLugpmToORNBoyIk_Opc&cBlsjeNY>aETla(~%i+g~lP)1UkFSGs7hswA31
zYGSxp%s<_LZ#bs7i-YngnY2c|+Rq>q)PU~p=GWw(v%Dpo>xhpHc9h>ghh7LW!QSh4
zU~%{h>G*xXk$)rouzJVXH5ZG}^^q@q-E<Xq)uaY-*4T|R{nh!w2}6qDb>!ahmfEVS
zF?d7M<XGq1kVy+_0=FFN$>ZS1Uf-RBHo>L#1~&yuRl=iuV**aA=kbf@?Md#hc!`=N
zi^Lf7Qtyrb33Yq`)QJ~<<B%VKBaI*`CE}|UK_g-B0~QhP5hm1Yybd5DRtz}Q_)$n{
zhG?0c>lqpWy&3?W;4ctxk3>yQEEGX|h>c}tV7`flVs2%1MqHz>ynpE_!%L`q(vwSY
zZkc~~=;)y-q~eAG%@xO-71UU8?E-GWJK>{@pC^-Ys}MhsOnUo_n|!YbyBd1$IQB+y
zj(WN`%k?B**Tw0gP?E(*UUZ;k50}oASl)HRWA3J19~J;J7iK>Jr{2SHhT(C{=zLT(
zqkm`2GqKf8Sy{q_kmm_|pxaDgxZNE9_luqRwk?7$BFk_hStd4oyW1dr)N5?y_DaWP
zbw1p@hd(0Oez8+@VVzNWYZij8Z?)Rmqquz1kk3SykIf_}5>Ac7zVXW5Khdy4b%JS|
zOMr&vZ7tLW<hgG<ciaBT)_YW9b9Vl~8e&szZL!ssAS<ErJi6ce13!X*w;Z2+njn8(
zU+RxY|C0yzf2O3P)vwS<NwWX9CB3utFQkWYHzjpC#JwHngjrdIHnyj0F!OLJx0q@0
z@&_Mw(m7ipHa#~kvEAE<cCQwCbSYSczj#lLbj*(%L%H;wZ~Rr#-K}4s-${2K?ysaz
zH<Q4Ycg$em-}hha*VzyR*CnWiYphD!dA~h!rb1XT;mzZo7M}P6U4P+>naM33B~p2#
zlVJfOz2uwl=eA7K$Du>ldl3vf%=y$Ybe_H|snXD9x88Xu4Q+3OQHB2!5S}O@+d_ht
zr$vB3K~cDw)=GcL&qcwqI#NcpI+D>Qbo?T8d+;wfw>GEFZ^s{8ir3r6B}gbH;{@Q`
zX$;1WQb<Mh#Uz7#Z;G;tQwEtFlGq66l5y_`+BPRfELAABFiy2emY#kOn{Lp^_CJ6w
z8i=hr!TNP>XrYtFsl277Caf<~Gr=34_ufaN1SZU@J}G7<>-=dJqUMem*@4`c55ZDm
z_xj6{TRLzR&-bjahb5E|Okxj*_Bsd3Oa?Rui%hRwHbvEO3{F!;MxQU)vISl*$?2)?
z6Rb7^DqlZGUQ=dIb2UjdvbECft(K4nbi1-8f?E9f697wO@7hAZeYcg0aK${=2RMGb
zR4{Pl7L@fVDVvserf;<N10JykuyEwN8VmYL8whZLO`7A)<ae{doX+R&1{-j@U|g14
zg6(jx13@D%8V`{4V}lU%b2Yv&#uMmXR$H}WtfQRd8jv=Izp~VkqzSV)G+1~b7Ejf+
zUNWuNx-dP=ZM9DaGztCzbgzhkNBD(V@)sBZ0o~|70iELy7TbSX(*KJ!7d%7{umvnB
zqJ{_%?^j#SE0lE<ZaKtY3q<3-0Jw`qs#BYsUnap{CEc>pi{}EsT{;0Jou)qyXT|=X
z+(lVgL_IUAl5?FK0CW%TwU9vo&@V_UZ!K#qPiEo3t<!GKHCmWXKEKgCs<9^_9tEEo
zKf2F2T6<dlg>;at(^(vF4|j@SL<~WjD~3e`$JI4xN16gsy>$gw_keq%R0vllJUF})
z0Ma3^K}7#Xy1dGD9nTDGbh|r&^hL8Jyk)&g@G(jUuJk*%ds|2$g6IBAt1H?fhPv?O
zM7={bp+UVy=m=gh_Pu)-KNr*c!s6iT;o}7KxcUBjyybqNm3FJwW_zm%(Y_olTbkis
z>PeO}0P45{e?uJs0CgnuEjW~0KPW>e8VQ7o*#M}Bo77uUU1CJ0RhJ&be<3eF|4Ci~
zM!pgDyg;F$#N>QWw2uHVn1;*H5LR|}r<kRhvYW}yBK&9?N8RbD2R6BvTdvM>BHE4+
zu#Dl9Nukvlmv)dweB<9VQPV|ZuGHfClCjSo@MAC3ftLL*Tt**pPN9xBrg(29%gXTh
ziYG-;!zC??8lcAEWT#a;lc^gR=fv9}_dDtKy@`zQui)BTV|XtmcbM;3QOqukq2^{|
zy(R?yu9z2MsxFMWLfhkGZ&rCSyT&QZ2XUnOE{s@*wFGrHJUo~iPOQ7BS283AYYmq%
zj_lzmPxaqmuG!9T2&hl(QxMbu+(oTe-yi*N?vib&`Va0h^)KAz3D;w7^JeYQq2W2L
zy9xcV!D?;LxWJs-;5w^IbXOFiEC@he;BWOa{)qHH_+tNeBHjG|A^ra$9f0)21pT%@
zBK;3u5$AtG)0L#GS2<7t4zLXZ@QlBCY%OUiBJ59X0r*vPsp_RN5++s98iSz+qE9c*
zOI`}gX#aE~)=OI>%8-`l)5G_Xs9SxN0_}zuH0fJIte(<mcV|f}s}<K~+#ZJy{2t7K
z)Ad>Qd;652R~x0KaZPvj^GR*uXDe(idi~~FK8pEmSN;z9>d014sL#s8-jjVBE-buH
zo$PL}r7UgYQ+vPAx|81W+*<D*yOqgVcEe7yid|v$h;a{}IQd$1Rx{7}$NH=R5z-P-
zD9LceT`!K&V6PydaIpTzsv>)iH7OZ;IYqN~nhl<eBxSj0GSyF&f$Fm&W5?HT>g|Ai
z`tY?u94Ip^n#^EjY7ED=UM~>mUhLDX^Me-ke<ltDCcxcdN)N6hCa&M%#)_X-l)iYS
z1(?3sOy+CcJUQ&6$CTlD$z@r_jT9oS;-YZvaQLQ~j-${^=X>^H$q1(_w1hNYu^`#X
zp*5?rnB>E`#UIHg>Li2kV*jxCN*%%>Eg(8N`aNBNZpl&<1&Km2C^k}pE8c<fxFugm
zEGjlrf;jNKOeXJs{^@GdDkj`PO)w+S%iIy(!WEy`aikeygYm+LITD80VIuh@KD-3o
zRz<&r!Z_iunbB%v(57(p(bnJg5$AK{_%j`JG8!_D*bWrIwl~_pX6zTA5WYhPFgZfv
z4(ncb<CGbPP*evC_@ukeU_&33D@Aax`g(U5ku|x$cW<ZLPF+zT2cA7G<7>8K#dgq_
zmx}$O0&sfFUgr^BUJj~9$!;;uJZ!hGt~1eM1L?5o<@#My(f4OCy!PWBCX{NKkX<@n
zVSM1H`5g}Fb%9>Sm?SdBSb1F%Sa}PUn7gXB@pVi?v-{yC#@)#rk&j3#xLu@s)3Osr
zpqG*M8-&J-j01eFiA?uN={1*?#;_S=r<2uI!!OKN5(_euVt|t)DphakhI0O($Jl#}
zKI3=JYgJ?5ak5j|tBLq><^r{a6gj$6mMcZ^iK`tz66h|LnNLj1jy!TNyrF$B-1xV|
zgvz1>AkhG5g#Ihh@K<~R5G4W(Yj+1E8eE4{UjBKhoXogfU1DNuF_!tMVGowKVNbFF
zFbeSTuV~Tj>k>+0R7uh#uIK!Y>;0f;WnzTyuo@c#f(adtP}nt$Jjq(BHgy=G*)paL
z_OrlmNp;4o>O71PwgyA5#GRhS6!RgI&0CnocY5?76!_UDf9qyi2_BT=zJfS-<2A1@
z)<HM6e7T|YT;o-r-*oQsWIiBK(NMkqSQY+YQgQw#uQW=dR=XU4zL38nfGPfiq=sU1
z0*nG3KDbUsjL_)rj-=$~go&`re0!Hk^Xt;&{<wPaYV+fzxugdNRauWT{1Z{S30?qU
zn`1o55UT9pa-FHA#q@C74r=z<9ZDaw%2{S}Zqg9{qE~gD&G~%2p5HrUug%n@C4772
zC0!xt@cmA@o9pC(=&el+zAUNtcyEWH8sb^k+bbkr=CVvhfgmh2^#gj;lwW2u{<*Xv
z@|P8M+A|PNkpZ3D9AKRcQ~rpsBiX2*Ntk5=?>31vFL>}w2rQdjnG@SK?7ZV*iY50|
zF1HR^T4K9N2JiY<y|r1H^XH<b*9-6aE*{X)IS#e!rA$2iA=96a>qN!(MfjT>2vvjG
zw3;{ztYi8dc+Ns9PP6tus4JREPhKsN3$LTx$@dcni)fmHHO{Z%UM$iS*dmq=D35~m
zf!SwMVd&9k_;$i!0JGeRn4SlXSh^I1Jq*IXzB?S5c?20&UgtY#h}3sFq(Kme`gV&x
zv3b9BCxu{+60UlOwV9j(A&E&kJT<rayESB{Gi^`oimHL2Pp2@0{8{7^mOJ$_dqB)D
zc#dip+oiV|bb^CEc#2a$RsL(CR-bM}K}xz}<}SkG_$zDr@xyy-x8vVMfq&sQ^Ims;
z@Y-*oHO4(1TKgP8jnD+7{QL`C;9RQ#J26d6=`%=mJ0hGlCvlKng4w#1al0_>jT>B9
zFDc7-Jj@{$=|_Sa5!6?*zWw~x)$mqd{~KKYN!qm#l43uFii65nK}d#N2K(!9a-46R
z;~55aZKkKLQDq7OoXDu-70f|XvV1ilLO952e&ig;bmNf7bkG*h<hCBjl!+$FiQ~_s
zIgxCH9v-smLXb0lEg^`Hw{O9Ab-{56XXgZ`f0kzsNLCDEy<WHhLQa4<(`jNkKWLZ1
z`(1i`Ilq5q?EN7iS7GnMpz8e%kgGt+azvlfsGR`0lpN-g2P6%t?rWt%wF~xO8MN#I
zMHx7E<M#dzR;Y|hOt?S*3WH<EUN}$3-wDJGK%Vd&Fie_(v$riNJti$F-L&%`mnPKg
z23i5dz>AL4UmH*-<RvENXgGV_={Qle?a2AMqIh+mlii-jrj_Y3l?eFk8Z2NX7iyVY
z1k9H))xx!tv9}lA<W#|AM+banhaf?WR{Wj=@H&3fm(bnTp?W`Y*&|?`yL)Z7Zualk
z^vJ$AUBD$7hKLepG&#&{hp)<Ppa0;S5OtU)$k<=>#<#iut0-`ZqtyJdDEz_m<oFMI
z)g2QgtO3npzaC%E&WHtCZKUK9h0y|mPrwbPAyc!D7Gjd;1Zc5p+lRUsCdW8B9v_^L
z4uiRbGM_0!*e^Rbs@@oR0~KGsm054|XddvDpUAw*+u6@bzVBlW?At>LVOBa%63j2U
zo?Q0kTAj8XSn<$a<~>`TY10cew|;CY{5tG?TJ>|BI~RJYNsTH?+O5UJ`|J-hR5g7f
zfFG5(dmqo;X@1eykUf56%tDxbk_EhGoN`P>3i0AiPjaeCkT-M|dT@pJ(ss|0=#dYU
zZcj1TiFCa)`F*>7KXrv3Y`ZoUZ&YviK)23JdQOP#lv&U|AQ}t0R$c#yq2K7}N|$B;
z7K@j5;${sf;xIM1cJ+oedv7w@v_PmT`l~QVFUkmUWIcE$KknN9XjHqLYdm!l!wmXd
zW0nK3g?serzuCfhYz3?ADlRD335QRbYA+}~{GJUePjs(rk8C0)BRDQA$Q%}DZX~Em
z!M0eQbR*oFlmw_O#VR@(T~a-zsqBocSXeAP8D5fb#6hJdW<{0dJ^y1J;*@F10GAY;
zYKD+HzUYsP+RAM9a`mHdUBZ9~)ma`g^V0j|S(Vf}Vokq*O4T^+Xri`*rA!{YW9o~^
zg`d)KdTWG5jZ|*~Uo{OFsb1Nq;%AWsg*Yw!c>+0IQ|q>D5zd<vdB|qTql_zutf@z$
ziRwzQ7@j%3r7{zF(Nk|{ThX7-2XYtmG|S+en9VsU%$OgHRoEBLzFBr>76uw2)Hsd_
z>(d#l<T)Fw>Psipey>=q9@ya2(}Q($<}^<QY|Ds0I2JQ|EtJMK)z@cHjS-n*F_l!N
z>fd6nGW*s8=Dx?ZFr;et*Bnph_wU+!t1luuoQy8od*ht9`@ro;O79}%Kb#z%<b*Mx
z=QQ>;Zr-z?dge^aX_8+e?!st18ii4BztdWZ6<yNvVz8)j)>_h|W3cF+f>FPLQolR>
zMfcTRH>|V0TBJ*B$u9w;k)^9z1TZkB4!KdK3Ayn~xd;72e2G|w%%VhTDfdx)DOP(a
zHfu8}vfg8AN`h=`8gN{nDZWK2Y$-LoS@^4yjnyd8U6R|efxn7z-{^N8$$!_Tf4#g^
zNxS9t<7|sOwEjFE#K13_S$&??KH$Wsb8wip_4VIj3~ZV`XUX^};_X1Ig<zoJc^s=v
z9CT~?$NBsRN1UDM-|v|A{5wSTUu|N(068a_I<&7SzPC|PzTzR(Z&t(%2$=J3vuYDo
zOsT9Cate@Vd^=bk8;?5703@i=?)?D?s<XcnRLhf|m+NbLnvPs6PHy#$jrXpO9*C1d
z%I9aKR&|!0mTAWmF-?~Dwx#DO_pDS6ms0KpIs~?_X5O#z^OEA?$I7DKJoiplfb<kH
zXVv)KcJt<RfI5nHB{cQ*D7^k`13<v!_JcvS%y1OFV{;2r%~DB6BS>w;H;UPTi{&j4
zSyz~S6S+J&k`%nTuung;UB&0J?m!R2XyTCWrJ(A(MVAq>3@Y@*Xws&32wyL`<fom3
zI<{oXZG==_a=mTlSF3rdyoR~IuWJ&VZPH6nR9>ywRCZcwx?E6~XhAxyU2|!y_*7T7
z9az;Zgvf@daThzTZeq*nH9FQ(GXq0rL>BS23Gw+(n1dRmnQg<NX`5~1i+V3-#Bm0K
z1p*HFb;eN%QN&neKI@&)Jn3uCf}HJlC|o;FP*_*FLr%u?vN|-z$C&uvRENm?`d(ks
z<teCbRZIt_%86aj_XCtsolW6!bkKyNLb38p6}33AQyq(u7R%*@>WpMq>P?q&8i><v
z)9segB0O0aVPkN7@<8EU{nG7a`6)Tz---Is<`JlzQ*s9PLwAuFJ)oL50SY_^_+ze*
z#DH(_H2;TaUkLFY_Aq{rOkQOsjwde+O7&;6I0q<7P1ls}l6<FrU=_v-(^*O126>B{
z@ysbW^B%^l)zcgD5gWL^i@{PGn0voRCG<%j(It#0)2Y~$H;;QjD$^<SPv!&ecsaZo
zd9Wig=@6NYzPd|}x?~nwn9#bYx78~w3Xa`?XNQO5b_|w$sw|o}DdYF8L!|wKt)%eL
z`M1R>ZNE=!=HsuNrL&C9IfDLSfnWndvn!((r)~JEcT+!kqKrIv6E?Kl;=_yoFhLjr
zU0IBT)_P`NczC`rD7e}f5;Dl>naMj?eqoR%WMX9eD}8Tg@9@_t`xgdDLn~tk6GB!d
zj(@tj|NC(MIgh#hYGdu@JIzxKFBT41ForQ#Mh!}v7z#z?InU}Co?5PD=82-%L|)g&
z8NC29sfemv>@2f07rq}56zxL?Omt!}T-45wKJt)eR25mIAQ);c<gfjq5dj6Frtv25
zXrp<g2`Wf20VzLYVnSJ!6!Ldn5U9y#pm}R!6ghY!NX(J!M)Fu#ReVMCb!#BX2GEuL
zA=7a*$fSS^Fkf77g(7Imp@o^3{Zj+fdTJ03s4c4tNitjcd}75@VYVF2Y9PiHfMU}J
zWkJXme2{hZ%E<&!P=CtDpi%{~HV<G69<ar7eRJUf7cY_Ni;+V&hJqI+d2B|EK@X_=
zL0rK)uV7H0jYa-dY1@oAf>y>?26w6f+jZPmohthm>mCG?FNMI5bd^4WM942>@Cmrd
z0kl5USJ7>z$c46swqzk1_1Ll?^gw=Vztz4JhLCn^fzJjIzoAN^KopFjNlJiyiKR(S
z90GIyc2G@xZ>E5<j|r@AC@x_W{Y%cT6r}8XZh%f|B-#nk6+w325DjIvcxZxObA2=m
z5TQ?N@{bw&Ev`8rh6>-5wJitzYpO^IV9O9@*q`-90&HtgQ|KZ2*wFQ(0ri*zn23n%
z5COenIg(%W;)eLcLh}UD*S>%i*N|URAsV5TQvp}AieLJw*`$HYLmx57ggErkpcOY2
zb7fmJ|18%PFAdQJfm9(J&NsniDn=7c^A*Vz&Y8(m1lG<;Tr!Y}eHL?}hKrQ1P837}
z{fSp197~n@!_-)jO^!+xB^k~p29+nWp11)5Wp;roH5k_UB<}3}%3Y5|<KU4aGD}xh
zhv_%*pq^iUu9@8VqsoBGWGQfO^{Xta+X5J~1smkN9sFICJO4XsO<3^z*N*h>XR>#_
z#00cq$9n=ppdb2ox06y14Am-ZNL03%6%>Z&a=C4DNa(YrDhf2-VfV2YZi8seY2?k;
zHJN9+`!uwf%E%sm{kjDh7@KCU^O_&XVd}I8XwBs%OmJ`{wrh$U;ibSLzlNlY--?k5
zw%<fWGrXe|HFWJ2(5A+FbtgOhoYo88lM2n&6^akGb4Lroy)rJ~tm>A(W__9`9AqdH
zR+g=J{8~>We@*i<d4Q`zaPRUBijcHGh#ew*0eVx?Lzt<hp9g;8EP!WuVWuAU?HO{@
z$yM-kdTnLk3{veO1~Uz)q8cI~{0&U1Ebf=jF%b<U#CjqYUbc<9NCQRi?TW11o|nAB
zcp*<S<ws<Jln?cf6d;NdNHY|F7Mz^o0s%3(X7003uQNuZ5F(?VFJH_r5bQyf>!P{0
z<b#=`Z4wH8;X;5o8W1iFehPX2aDFzv?1wY3VGt%6UpZX3*)RqnZpP7Nxmk&>M-nal
z7NuYU-s$kwxc~`LPo(6BUnT5oT)@k)S-g6GLP@u(0@tfu0rK5k0`OQh3$>L-v`6ms
zbDEfJo(9&0UE>-ZovCfp-UQg&cr5EHayyt7ocSdc-Eud10l}rNviVeKWqqH-&~^3p
zvUE6LpJd+|OHG<WXsYEZ(CzGpgte5`LqYo%)+UGk^5jP8i7*E@>{iqhfuDgMj`pj+
zAaOR*&4T2~iY0AgPW{OU_vfn%Kof%AaN^PARU^MZOKVWf4YaH27s5LmPhrd3Q)Q_4
z%0|EKzz)3~_ZVX@#@ujo#=U;#lNrm1Cn|g!0{b?ERZ^V(yHFW(NaHv=u*ZSdpp?G)
z%gH=+X**1w_!bn+w1ut)7KBYe5<b{b{2D^HttZQ(kaK14@M?FJhJSHl(CFY^=+EsA
zyzek;$v8{H`4G;$RGLcap`_IrNLRnJL4jJ#T6+sC>@tY%S)ET;$jOdvhF8XmEP>)f
z=R5Y@=;X~1VAiKbazG}{ObnuKNje;Le;530m4s~X6V9fS?x{}|qa#-pkHeYz$sS1X
z;}ThiJLD%}w~u4Gd5#=6N48k8Ar%+w(!&lR5=n6U*Vg!UUaG`|_|cYd2muoXE<un4
z<_0UKP?D9Bn)b!VAjvq?v{u96&ncd?z@vl3gM+uPX3X=9@;$Pj8lUH*wjTr^-V__H
z+2My)5TK~%AOSf#BFd5lk)wD(fhuQ6^YA&Z$`fcK^SwGomf~$;HW6TPKs*+AyVvw`
z9j-nK7lN4kBJ>D$&i;pW=_nV?^kg`pAeapjxL2GHMagoQ>JbHP-b#(O^btZ!D^rSg
zCGFfYJ2U9eRAJ!}j{;Qpi{0x=y8+r5i`27t>0$OqH#gT8?S8=eU2&d2Z_wt-GG_T3
zZ=0S4!?PBa%L<yo8+-Bssv4DNIP0m*7goxCH<6DV>+88#R(;(V(vry7oYx{NxM_$;
z#e5GaKUbu1K=-wVwUHu8=PgVoCr_WCbf8_=ROhms`5iZ-<+)&R!KxyN94osbZzBzD
zX6W)H2b&y$<8BxKv#D9a*NY4ZSU?F>Ud|&2ED|{;e7hykl6+#jl|_dK;?`;qnKO4=
zuX=Oe)A#d41nV351bb3v-c6ghvK`H7BWY24bW!tlkE;#Ct1hk?ZqZq3D56Rgi*Cd~
zi?KNiymqu}L|l!DFrddr5;`d0I)X%4XtzELbz%0fq^Rt!v%n};Wh7`xc7y?z*~PNJ
zF5gURe}qGpS{N<QV?8m==rjFA;<<s6(ox`jm-mCfHEQDJ<2Ta}TI^S8+yTy3NJ3h%
zE6eutbjN6HH#HNs331z~$e}HDbKSR=5>vW`Olh9P!W%)dQo}6TR~rivc#R$<H`xdI
zMwE1+5ketz?f8U#Q$w@_$LO_F7{|PL_gBI8p?vjfvAddb;p&sGBgP&d26^pMUOh1V
znxP$@NC*iio;IFKug68aPS1vkqGtwNs(d3D{;o{w(ueMIUYQZw!~4^#NrRUJI)mFq
z9!z+LCkB~^uNbqkOaZ=mXGTb3?7K%MFPtxy+9f7go$`!S_~9oA?q@bkLlhF(ftNKw
z%V)M)Yisn^*$z?LwNDn>VZ09pDG{OF+j~awqSyFsDhIoZPchy2?x%-7NWj6X7G5z7
zc?2p<EtwxjO8ithv0ETAXA_TJAFd=fUc8?7Z+GwhVaMwp(fsM~w{_1BpC()<L~!df
zoeG5Jcdj*&<|__nLMtq>YSRPq^P=hXthQh0iENW-^_4At2htipXE`qiYu26a=ucDv
zYdZB>EuSYXTk`3Gzp}sjM=Vil(=H`TE$vx(BurxTj|WIChmY5~@?q^SN+_r8>RGV0
zs#o(br_(LPuGh2;BO1XL-llsRWU2AYuU_5l9nFu9QqX1csXo}C#xK~Sb^)2%%mKE5
zD~<sc6)8`{T|aE!D}lQcS>Z4^CgRpZSSE2AlU`ytgLWUutdW06@6a(mOV>mkt?SRG
zOzbyF7II9P4C^N#3@meHdm_<B#_(Lfw*3t;JM93n?~)E)JHn6feYu&eHqA8?ct}cw
zJ!7w92ehwHh~c*gA3ZL5m~3M+xG%kRxA^kSY(0tELyD)acTLxg>#E%uo1B-!)7$a(
zXkz^2ZCaZwn>L+UCb{%nAm7Rt5xgM3NOPt^UZuL400RGTZ_6Lg1xdFw52+ox*2~-v
z_=bECUi7uYQ)1pf#~__|`J{lhx!JT0?Pq?q&xEc)8)wua9-RUfiZ5el**mQ-0*Y7`
z{1rV~K<yfIb*S0-y9y!utL8y(nXF>V2zrY-?U#5__4}JBpPAgJH&*fFzE#TY<FK$n
ziYNje9$v4}U3^HAZq<D3O_iJzELro6Gx;{pw-1|!12Gg6tsbFql4pOFW3M(^qu%2p
zu?j3hnF^vgM|N}gmNanCt7%zVd-<ly)C!%oX3T|%ecHM1F9Ft|(ckNGsxv`0OM#Kp
zh)7SM1X`Trx6O6tz)TkNjvcgQ1uKvdh3hSzOMuL8A36IT4ZWBrzxV1f;TUR{vH@u=
zh{LI{Vt+d=HDT*f0s_7&KN(fkAw@e%eQR}9tXGRLGBGE6i-ozhP=T#Q609myd@qG6
zbM#bQFoPnv<plYJdK;n;GESlkLta&x>u-yz%UjfuEs}Ym{Q`6W^Z1deHT{b-+CtbO
zjm3^|IdZACo|?q1qV&%7Xh$u;GT~UWy}6mk8#*B4aT;7*LI%U?!KtY128~Peltu{x
z>hRqir4b#nab(ns*USCo3YWv?x^hvqaGLK~{LN~4%lW(G3H~LynnNwGqzjZTmVYfm
zPu;S|%kp&A_~ERKF#kO6yCD?5@#v}(;=}=FvC`m9+cYKQK;#V*1b3B;RfcS%2D-lo
zXGV-y%Y-ePI&G1Kh9n_=<mLCRcv0;o(l%^Ei`Y#AOx*PudK+Q{pe@8SX2u>pA5O%O
z(Qcq9z6lBcUqGDnTdoqZ>ib$%S)Fd)FGt;~4~uuG$+j#YMjBGSknh|Wl_nT|u?@|6
z=ffQi0KKSlhVfO9tV+afY$7dSxMJryZg)3Nn?O0BqDDUL_XE=7q1z+v*ZjN~A-)mO
z(eMsT|LWys%kBwu@HR>-LaDa)@onfct}wBhH;~v2_)7?h_Q^tf1z6N8JW>Z0Yg;`U
z{<nwYUW!Wu8tF4`5GCe5LZR@G8BG2WaX+xZ&`wy<<cANciZvwy_1ft(^skC~(tPTu
z(Ra3th<CE1Ao_RrtZBe)M&N0i(%+_I^heL&Zv^&Mw13vC)FwcBJH4GO<~<^0W@s72
zfk!P`vt7}=y@PdC4}M;4X$*TgXO2XywLm+c$OWc%kN49As=l%A?U9K0+wSsOTUMy=
zS?_mFZsK+Ex$X`(J^n<<enff6`hxF`b5e0!@KGzWpwxM`x4wP7Yb>yE?C`ZixipkS
z6EW>)F%v?#c7f>+bXIdq482#(h#-b$x;9@a^w2q&da?4Ak*<smSBIOOz3%?uI0+ua
zbM`|pf1B7H7_hbmf0*1VGvUqV6cs~)TmOB(Ca9GWsSfYQhpYQl$f}9Wbi+I(;pXGI
zcgJr(vSnYC@lBBrFl=w1#)5x(s6A%Fkpue%ZaO1R@txZi0bRa-TRnl}yy#xcYE$IC
z9GA8*fjNI!gLhl+bX_-6wcO{|=4s{Ncb`c>``Rd~l^^QhW*wW@E-bIrs%urs-fS~B
zp1HRTUX3d#`uHYmxlg(IX6N#JVj4Gbd(3f~BWs={eWkMXnnudwqcEP{gd|{@B~kN^
zdf~#60C$nd2`WxFu7MhMRc{8*&Bqm&`glp@0nz?(He9hCl|3M&AUtGvtc1twPJ_Ab
zj?WrM>c$uZu9W#B=I|K;Ql6#4qB&)*^i_yt^(whk?<NpZ$*j!e@oc%*`I4cvwZ)rh
zgnaaKcg4e)u7d8GX3ilY_ThDMi_%`5UCn*$=Uth`;eoWgHCoMP1~0FN+Xm`V2YlSe
zBRWX>UQRFScs-3N7{>-h;r9ix=Qn5MxVTB@NE`KFQxSYS9xTbAtt%h<>O~;#4A>iy
zyEMWvQQ24*%`jxN=5$J?1kn;+W1k&rH9~beufdw(x?YqcT*w{CH32<b)jqx%+ZzVq
z#$9@*DhGiLG?7&ILp81$xk+7wM)`6mL7;%s@2EcOR!1{tqN-i`?c~Ej#DnooULq<X
zFf;TaM38a{cwlHt<|&R7Ej%IDy1)EjKcX5C&|?rs2Dyq?7&mm|!@x!s`4fABVt!D{
za-SpM981IoS~Kxk6zdaZG+s;RE49C54D+5~47)TyD_*8_Hr=#1f8UgP?*VOP#DLbW
znm@rTFv8%4MR*%gw(b6UcVNYQUD&#TU|Eb9JG%&p)1z!@W{c1lh&j2;`e7p?MRhP_
z5y`LHeZl5$1se)Z%r%l1aIr<L6fi}zgw5GPbU}&}#4vi@M7jjm*=6`iPqbU22Ws#0
zk%Oa4l)tV?R}zwecoX^C?P~Z1feE{N`5gix5W-1r3_}7oXzTKEscwv&BWUZn5R={-
z1L++I60jDAJwM|&&7@gn6vjZJUul^Yb6n}K{0>Jt=}wH&g5kg5nWWizRV<^fc^S@E
zO(@zU8jl9K)?v-x*>y7Vfmg(YxD_zWTR_a~co)u{XYp9#*ivo%Tc`z?*=THPS9j;U
z+7`RT8f<5X<eoy5BI2}s<y%3e$1&{j;<*Osh2j|_DrLpnC{tYA<5#RlxHt;TMS?_e
z8v?J<IIhoaPx#-h^eMP;CDJaY18C#$K!fAZFvHH7IxlCtBje5}3Hp|RY|RpacNwnh
z5u@bWK&}nk`!3x(7Ia+(w~wZ_m!`}BU0Z{AdNPkvs*q)o-qUP*g;4K0<x1HNLO#cy
zmh;_}BxiF0;ww68?c&wkV6m)4fE7jK)d*e$?*fh>V$k4M&k%(#IP1=J)|oux9*n;G
zSqDDA+Zv4{SwQYP;N|H$W9X$@1YX0-nv8SG;OJ*qRAR4L1=Oq){jhY1U+T`V&RMuN
zYW_^}Phs-xhV_=OLP1e7pmGY*S$KVD@V61oy3THZhtRb`98BA86M{YUu2^9b=-fcD
z$Foy$WYUUZSj8h&dt5hxoQ>Sxe!Ov{#<UXMsg#LSHqdht3FjWQy6MMv=jiBd+IOGI
z+$ExeeW-O+gVj}+j2IGG9y0EYK}d}*LxjxjyZ2r){Uz*5D6P9~wZ?mYKDWDdXVN5+
zzp5@gbG~r(=(o0>!v9_sNuuD<PkixO_(L3FEd=?(%l0s7tFluMY1bFCE7&;g&LXbF
zJJaizWc+OI57$Pji&ZNMu`c4>i&a@Y>@ZUNA3GQ^opkTS6T?&df#;pUHv;wAeRVQe
zn-6F5Rx6QVE{cj1th1l^Ep75p=MB>K9k0YIWu;D^Maai}pi&;{FNYJ9&KkmQ;T<Pt
z`>QJqh3|Ti5#01UipuI;!OEKFipt!dI4N{=8-Zb%&OJz2Riy(N(jw&7zj@oj%N3va
zAEzMU|Daehar|p5@_(aPa&iE~%YO=%Ujg#vKhmfFd%?0mRkMt=1<vzQdz&Az^4=6D
z@fYHdgdnouh9Ew=ohn44RB{U3a95NDnd9pGMP<e+LzBXMr9G*tsL~@{4^`2OogZuo
zG5J_>sZOqBNn#?0y_ss9fmno@TwPkCf?bXCnaub+Y~Y*#eRRWe#<q{AiFfrqe=5-M
zhtHBkpMer<Sbw69Htjnos(S8-k)pcXmnFSwvIb1>ko870T!Sn<s|RdQ(f}h|i%Oyt
z>)#MQun@m1NKsjUh+rhAt#ZtIKvwA5Dyn(woN?;m6X<LL!c?P+xvdh(KttL@LV|15
z)5o%{#B$;AQ&SC_j1q*+F07>Gd@rmlja2!@YaJ8z^%A!mCSuGRh1I_@CFv<SQ({QN
zu5401P{k4%OYm6Z8-&G{I2!yi2pKaHlPX`u4^37qvt3fMTmHGkR%b<79<my!r><M&
z#5Q3x-x(>6Oo=z?m)L1-^bI+-6W$z|<RUS7zSb7AfYO0SBG=^jl(t_kxy=w)kvsIf
zpgyw<zx05PK}jpMsa*xQw4z4JO3BD1CzVVr<g}kwa-<_Qm9xZxAh|#c?S*SJ-;5#E
zC}nQOyjYcd{$`DC_sErPl!GhYQmFy?B4JiR?%S$;-Bh_&vXmONfyCI1f$p87U<uT|
zRf+*>YmtTE6x^gBU4a9F<d*{R$KPhPEGPI&S!PsB@GxBm3!JpcUsg657pWi2td-;v
zioMNJ%3;?*M-BE@u4#1}PAAJPl#bgqeGinHIOfGYo-f))S2sT%FCXt7k_fy!KJ8pz
z*Sox)pUr2&*1f&h&1s&}KiphjdyBli#Jm}lN}bH7kG&MTl)Kn--&bDl36d_mK7m5^
z9YJeS*Q-_s6JhVnSE~pKC`z^nhrAq-?T}YUO2-(YtK)*>f;Y=H%i=CfUkSx*WoVw}
zFVi;JuTR;}L++b=^Y=X&)HZmne7x2lOOtk;3f@4V&oiAg&y*oCC~iy5U%rGKcDj}{
zVk<|Dl~p#}*dlXJJE0zBjKRU@wfJ5KT%l1@!}oR%U-#{Q=@`8npkIoTGyNS8M1P$>
z5a;H3V|fq3oS<9*2Z<TjibMq-%zE_&&E?#RWXK9JNOpbe%I}m$0r@ww!ihUL8K!uj
z>Me}|Yq45BjoeC<G<!O(uXc7%u}Pm!HZDXEl6iBx;C{t#*i9^jFC=~;k~^8h+ETxm
z6(cZnb8EiC(&gUdK~wI+D4@Oy$WEiCkoqyU(*|#&QNu_5>ItTZ5u0(5dqOMyK_~)J
zBEmTCUn9#Fu3WqnMTT?B8SdhT`Q2uQ2HJo>--lD<a4USxPJwE)35|PPflo69sWNFR
zoV<J5`59;R+?Sd<HEprE51L#y9YlJ*i!$TqwLWzSJ<T+-Ft4VZ3<X!J-xg;u9N9_A
zd$Nx8dOJp)r}H+(k3O8#W|h9Q(ih!Iue}r3Tdy9ucDJxBjDB~Qd%a36oqLZ7ub!DV
zVN0)(hNEmwZP8GU`Fph7bDU0s8rx>Eba&PjHZAqG&YpR1q$?xrcZvbR&?pY^l;b;&
zpb!c8FC8P~!ghS4+u_9|I}c?-kUc{Z;jfFPXKs{Q?z2^oAHN%DZcC~HdhOs>5+pz0
zMi;aA-q*C>--{--!&M)Y007)YaB6$+sad}mvh#eMUCi2e9o!)ITD9cAx6Z`yzIuM=
ze?`E0e+Z7YhPuY1LbMx3>B&4G5u^k4ODAmZ@^<^YTzpqfLnB1+aREs-{@oB{14?_&
zsYAs~SOFw~2Xqe>ZT!0(EFL%-ABfA(-9;EvZ%#k_i#=Jhg!}lq#$D>I=5^|81b&G_
zuvXMG_rrAp5nlo1hKPsjgY@X0=31+>jp#2>MKAJ@bLG+)&ORr@%v<D0PG<a+g7xGA
zT)#aw%kGc$8~AUKV0_<x4?n&Gb%El(`2Pm|EuhlIatE3D+DqqsfsA)Tt#S9(5*EYH
zC3ijb<7w#!$ZCm8i3C0u<SukiIklD(Jida+IVa(GhW<Gb+zOE=5GK^9(l`J3U}-i3
zB_#`&GlcPx8`Qp!E4<rm+L6f=S#l#aCvaRaf6?x(T3p+oP&f7j0>%V7k@%;mybFh(
z;E+_?vf|&+#zyVoAgR82`fN>H*smi2L;UjLiQ=2=N5w&&?&UEi<h!(Fq=bna8L30m
zQEg)ULDh5!<(i00TM!RPW%$>>kVv{AEO=88U+oeh7s!}RF{Wx6L{Qhn&ySCiE49?K
z3Y)N!sao+}7=iAg9e|Y(zyo>)v*B^acl!ZK$O8lmrb@hrBUmx{2v{kEUPg3!Tz`{5
zrrTf1*uqeU?@PsR4XpZ`ceXx3ZrGYfC|B_Tv*3mz8&DqD8fTwb#MhzTRYdc~OrRUs
z<RqW3Kr^5MqAGATTc27s0iqO=Pg4hYnSo2@$ofJ1KMMT4>Ur=`SSRxmdwj7D^d0cC
zPVsIDS$My|8Rt+{&gA>oqGpl|(XwcMk%G(-;{~3>vN_IyeL>;`&WHZFf@~8Vw*|dL
zOQAX<2El3T4nI_vZa_tWzt`sm!Z9kAXt(oCT9U%2uGOW49)GU?8E~ug^tFIzvWciE
ze7=IxPZDD1f1{YmeumB6#}f9JJBpD5(WQdJ&i92h)xhy@V?h9C$LJvIPufamydL`H
zZ+2CjYT5_(xp|G(Lx|FM2*nr_C6Tg)$TtV+<6G?j9he=-d`jed&#1hW(=Wq(I=>4k
zlt@PqYn!ZUtq>Dl0H$k#9F%eoPc$h7k~lmh6vr`E{^(;Lr{bO*L~e_#dkJ(%9#bW?
z^O-%`fA}Ys!uhY!!2gz|u>22}!o<PK`tP!oUn$d}wCwPKmxokuAViI%5^Mp#k(bKI
z0t~eYF%F3lkvTZq&UQsl=pTR3JI;^y8MnU1mERmbQ_LD!79_GAQ06UryPLu``hYMx
zFd21QSz~Dw*U(i@p{mc#U^x=Wxj3l*jM?CGm>v<>y`rpJc5hDN97caRr+&ELlQ3V8
zegL0J*4zvm|DrCN{*w6Z1qANY7j<g+OSIFd0ou<^S{<^f?0GXDaegik@G&cOQ2L+!
zTJ!8eB)i|{HT^nP@`03IRY)}9=CCW!2peVQWV2^MH(?l6CSQ!AODjUypovXrDz1}r
zBrBL&mP^sg7fN8MmdfR^6cDLBk4le6xX`e`W0sXV2$u^*1cI!0Q8GwsEq^*hQ_c@@
ziN)(0WRSwUqvpo-nd&2FkSZRj2~mhuV5{cjBMd4jKrMYQiX$b(i#s9>sa`3?I2kuP
zA}x_A%ZVfPI+JUXnK|ZknjtcS%OI7iRd$T+P})5vb<Y@(S6HlSoe^_78kiMx5`GPk
zu8r$#sgbUo-Gx0Nb;oE#JE1zg$)Vx3UNF#K7bLJ;5jW?_WH}{mQF2$L7>=SC=MuA8
z*P0jmntknp{@qeexxCCYYigQlcn(_w#`%4w-ayw2mwCO^3*VxE^}7hSo0@*LWA>on
zYv7^noF*26F%oZk<4$f}4ZK4cm(*%dq>2qTcs^7L1gudc)sP$Ruk+(NHYz=tEJuzT
zj*ap-)VRhPE5$-yPo;Kz_p-7$+}6z~0<Wj15B@IC*Y_}duXioF&yS}Mg2u|pdxkEr
z_nX0{qJv3mkN2CdL)NRO!%6Lr&+!fEDDRJ_JV<!YA21wu_{|UxP!?3tc+nd`SpoCg
zfc9*YnBT08EM_Jpc>SNxtBYNoEs>x1r`l0T2sbj`AIiF$!zFg_<^Nx0Zvj=;vaAc^
z7TiK0xcfpD?jGFT-Q6t<3+@^S?gW<*+=IJ21oz+$|6-rJ_j&(&=Z<&Z7<1P6y1J^X
zx@WD~YxHd4`!)AS*Ffho_lPdJ&k2LvgU1G&0NqWnLAK4<pb$86<kTHC4*#1=qkr6~
zGG?C3sfwXTZMWi9?XZO5;2`M~!Bf;S*CQSq0addqfR7At5|=5+z}ZO@P%3!*WYi2P
zk;ye1o+-al8Ne^_n!ic7&v38>H-Lfgp_<pqy8}|f%9~EkW6Jd>q(rFqp`r)sWGh^P
z7>T)%v-LzYz8Hq4N)B6yy=<-}QX%-<95Cv7Lpf})4*!2t^u}BNT8mJMzdr+V@|qut
z>U2x3YH@e-U1{id`*FJT%g4$Qvx{o-`&{$xG#la=K8Qf19!}Y&Q)_w`SS@aIftpcQ
zbdCbhhAst;YEA!v!>h3R3mpQhC2X{<<=o~}n;`#2;ELH;zwu^n;C>>M0<Wj>zjDq=
z*;;8em6uMYZ|O4G?5b-O{Tymxuf0zv{PcAj!BG@pYG51tjy5{V3*{qbfK{jo7s)yV
zEtCz?F=`lRC&{R_Cww6EoJ~#?!bkzeeUgNjJ%}Gl2gww^fU`4*gq$*mT=!L(9CS|G
z%l2}3=k;rzMccynI9X35nLoE(nlO2~lbR`Ei0G3-Fjo|NLPTUit>hFT9{E8F>M9Bt
zV#SdIA`5<#l^(}m!9>N<NB@Iw#qdA%i7PLXAD~9Xt4>ML{i~QxPlbYbRb7f?{)8;h
zd6pD;=}##-ABxogmIUd)6uJn2#uUkanF{1A>Xo9iqQH6o3H)rr5!I%G>6mhnR??s*
zv1R3)np8G2Y2hK2fx4$j{2uli<PL&46fbBI8vXsVKmCIYa+|WHdBbmlAIjWh$x(d&
z<)m2HXz@SxC(}6zd2DW^nxRep%j8>?0NRNX%1-gBH&WhE5;lK(&PmcAOlKv6ld5>1
zr08K-Ut~U4{VuB&ejE!RRgb?GunlsAF2@+4{iA|HG;i%VlKX8}(VaV~pH~Gpw<s7K
z(|$^de1k`1JWOv6ir4sO>b-~kCLKiICR1-sSU3$az?68A1#V|Ul;W>RdQ?07iCLho
zbPvKj=pQyD;|RyJ6IwlQVV<VlGlqZ4*=dPVJXc-!WfL4!n=|<Cb5~v{$CLQ-6$fMK
zW1O8d7{!=w)JpvXhIjY1VE5#sU*x032mEsY+(4<ywqKs|b8R)P`h<dQ`n#>I_Qg~3
zWJ{FV4Rgr}S)ZKe6kDha>%S=53}^*yp@4VMN|~?<eD=;lN89w)x5@gxWs*fd)o~w*
zmVb%pk*!_|jJop7!$R04=$>RK`UPTxyYs%agTDCji0>YIJ|0?3jJYurs8@mOfp(CI
zI!0OIfFM~H*Svi})0nZ^?|74by2LM*sxJGv(dWJHJ+`h!De%_ymZ|iDw$wg2O_f`&
z>^OlgJK^xhV=Lu*xqECm6I?(3Y+%RiZ<Y>;`O}{hLpqir96bjIEcNnh<rU}SYv`}?
z%LCG{D9r{mlmD^9{vQr2umITp-D86{*ZHs2{{Q!Jg-o60n2(JZe)Aev`SSsGYs`U0
zo#D{S=dfQsa-juaM5!~L2;~zRxvNJ!Uze)zAD@zCkdrds?7ISwW=ORd655vb_)5C)
z3csNL`OJ^!aNdJFM(|WE>WC<~634K7Crsbx5F%_td`^$LME8;mcM|9&(s7X!1Y0zl
zljwd|2%m|SK0WjECMsUApRtDm-~SN)dyz+vAtTm=Q#C2*n|#pdS^`=od@=tT(ArUb
z3nd?ZkL)#%B(SgL7b+8NugS^=Yrpl=MuWHy74n+tkH{d8t}P9G-|7h8G`_pE$t<;?
zamWfyei%8l{nQVEgJC@)r8(!r@#%0Cnu5V;3rxtQY${0f*5=YPG`f-kwnTX@)h^ay
z!aZXLq`!i1zI`xuH<W(;6w$_gh`KcqaXG{Cbsa_YV#Vf)M7+=4N@`H@LU`1o&~R@y
z4%okW9mjZ!+?Jvz<TLZiZCN6`nrO1Y`4m^T$qD5xc%g9U3M+g-ecfEEl}MnlTIu`i
z4q@hujtWM!*)o}YiL5{Ux8%E-3~ik`UalxgD)HB%y)31u>XB8274~TKeT)Hu!RRqk
z_w=8l8B%MqrUztRY_$2X7eVoe{qe0@J>)bAoJV~4k=vv4`4kD989#Nbo-B(szlLE{
zxKKIFM7eW7%59H&;f6j)P4A1m3va!*+Mv2{|DbjSg>pSm70*fhlljy8biNfma2ESB
zg@b^PY)&v@!A}J#c{#tpU#wp5%pa@(Jw@5u(Yv(E35`M(o77(KNOM<Z*DPsRrPoA;
zHiWKR3kILIhncf|&&Qwd0o{dp-~JpAzBdT^l<xmp%Nu=bkzRYn3N0A>%HK2hN3i1|
zLg&uw*4!8G<#~{^e|g-W|4e?5;1C`UIotR4!jo^EczkQ`!5nk}EVKV&s|=m!R#c<k
z^F>>ohvR2kSD0~jhv5D1{^yM6CNheO&sC*WSv>VSIxj>@c6X?qdN#xkfKzQg7JC9d
z=*G$zVb(-uUBsZI2u7{^jRKrG%Z=OvBIqtV+vS=l&S!IytWk|X0B7=Z10G3}Lm}B-
ze1b<v!Vw<947A;yN;c(cy$pYfZfp1Iy3{58N@=RfP(x_cWF^(r<K^8Ki;!>|kT*?P
zs|0(qa;q4b^642C<zAgxkCV3{#&^pZ$VnTR3!Th4MMJ5Hj9*QbNTMr+=GY@y7Immi
zBWI6liEum>U;jwC)0UIG%K!qS5X>|n;DQb(hB)gcIlhBZ=f9DGfB`Uut6x8KhFqVl
zo}PH|`3>n2(IIc{$mwsp+07|tWGh{947qQX6|8&+wCC0!`H-G)VLky-u0z48Sx;YP
zgKM))q(@|@IbpUnBltsh_yaP8O?N>Xuo4QZvsvyC+Qu#PPm+qp54~=RFSoT;$ig=C
z%>xCg4DC<{b_8^O@uGHNIQtO0M0zgdvMsiyRtAW|D52I2aVxnIh?hDV$1GiPeOPRK
zr9`9GG*W&kX}gq?Yp1jhnZHuApWXP*P90)ni!UwKPZj!mPDfv8?G&pwJm0tcsz$Wk
zs#@l=AorQ2sg*0SeR1b&!#n%T75#Xby@lu=bE2#OSCI7Gp>b$r4sGiQLqapHO#>&{
zfym+PX4piJA1vRxxGGqfhH7d$Gkx1XeUuuy<?|yZWA3xyXTP+N8~f?qmS;$gi0gCz
z>D|Sx%jahmX_hb1BIIN0VH)@cHdiNi@Q7M`At__TiDTtTB|eueTf?c_rIu>mY%_;X
zfA0Q((wcC?-6Ws5PS;!gTY003Ew)~6P`|9mvPW+|u0BU@J%-Ld5)&!(-##wSw|xGH
z(J}NCc!Jarb4gBc9C<T+I?FHa+WIQ<f$;QTc?%k4o*zXq>toI7Q?`4tMt^81|8VcS
zs@Tmk6tq0e+X02Gdskl{zGGq<e8I&T=hrJwF2{QR1WL8>&qQCZXG#e^?B{Q-Sl2Qy
zkCW^3w`V-k1AAi{w<1?J4T_(?^HN~+EtY(87>^n6pS};UAaJ7vFMEtFTX(<p;LY=c
zVVvWgM?7jgkKy2W<mP$4`q0*0bP8Q`dI0{n{Ysy4V_|{w2VUn$#>2}<exiv8ckBCB
zc4MrGm6F_tk}M(S+fc#vkHo)JYPRU6AGOowV;f_u`g<;)t;V)<PlKU%9teNuGa*d(
zR;ujx-1T2>ZT5W?HC<}{o*Y#@FJX`?c;<dv^>XR%(eBgK@ww|$B&WWBfqDP(D*c+!
z^7b4^q>BJXL@~jE*0wMiHx6>U<=pj-!GSYZH%oDRr&gs&2sO_11kwx=_BgT?RI*nK
zgpvlOKQI3FsX4;vefeJgj1_FXg5>W@u?Zt65c6jW2|oq@2#WN{kmu=Qbf~fiO-2zl
zPf`hMk48_FksA{*=7Kw9;=RqlF3$MPWsPbe`H$^9X_#LGmc1vpqV(P7G)$rR-q(YD
ziQZqsFojG3+`qax2F<YT<w}OJWGyKlXJCGb72|Ki?ND6yrqzqKYv@Pe+X9!H9nJ+a
z+_<#WM<c_nwHipCtF48XYOIz1b)yd2*TDq!RZAwC$<%74Fh<JPF=UMJ;?`PGHg|0z
zD<Yg?>$nvJzxGv&P6=}|N@lP{A&SJHA?{eZ_%2Yh#a{ZR(VTc1nDXXhBIEWDjSlG0
zvv}zm%Ubs~dl~KJX5b}dxm_Cbc;S-mqYir3u^z`=mFvV+YI1%J<U7I@f4D5y!F=A(
zYNW8LMQ=)^erE0SJDxIK-S*t_o}^)6go@Fo$Xk$x!DG)pMoNGb!JD4Me0kr4I%wF0
zD9kFvNuCO15{gPbEfzInfrCElF|$WP@AhnzV`H!}-meS~aC>LmD!xw@COO9#7LKXK
z$*oCtL<KKnN@{Eo6KZA@wd(r;%=`gnCaRdOnmR|6>)<{}2n1`eS_*(P2pP9(`F@>V
zNLBEBO~WGoIeeaSSFJ}?+zWr=TumWh+E~jkVAx2=m}>O3dD!H)ik=n>rNCS=H}!c_
z<;YyPS)vJLjjgk(>HfslPli}MLbmU?A=x8k;*-&>myOPvrs{I2k$ieGVQFkfal8zd
zwS2p$7ipv!DpP)@vQyiaj#_vY0$aGB;6sD0_>~oP>^|l*k6F~&1xlQgHYw%FjizCx
z88ILt^YE}dAZOcg5x~L`W-iO`jTo{VPnr^DP6UppeM_Cul4LHei!pW*yFVz1YYlH_
zfF+=KPwS>3BjDsHY}tDo&x9XPV-Jo8`%jjGjr8&bbUg&7sg1+P=GAyNmKQM^gXUhv
z4A!na9t`>vR6UL8f!nSN`INsTnTw`3y_why7f88J?8U>)irC1hYYwF3UGH3JX?8jF
z-Lz=rnZ0KZPR*E?SH(I%QObAfMWB;GVem=q-8Vanu*Nl@pgv#TB<XXYwz;8;x4c)u
z%z$IQ0WPq5#;?^FrNxd>!(7}o%q%eI%DWQi53A?4s0ihr23evXhAly4z%Z*z+G6<X
zE^pk3x>DT0w<kiSt_pj&fQD8gRY93uzG57tan$b!R+7UG0{Ml}5kK+<<}`*gJ_ek+
zL(F<kkSQlnr@7H}$&-W?MEQX<?W<7`jK4?93GL6o598*vKnvXyeseSmD0g9tcqL)2
z+si0uFr;}3qQO?Y39V*aP`>i8@lRS2HZ{P05{{xydNiKOTS6nl$)7{B8pTfbms$=b
z1J8$Kyi_uQk7rlfg0JC3^gV0E#kuywjub3=q;ZC~eza`#AdxX!XFOKYkgQS7FMBWP
zM01$GZ<}<bRDgtirS#E1=Uc45PcUPjZ)1uiDPdm`a)>XsUL>nPgd9wSr!P*U2uW<(
z63vFm+L)2d{aVGEQfuUg@Fo6meki(IiWqjqM(VcRkTU*oGf1G%p)KVe<Q(II5;L{a
zP^2}#R@Xa&$}o12G`|*x+^<{0BX`9@kCgL5sk71$DDH4d7+7yiYD}8(L`@vUZ`4km
z;ZuC!(!?QN5zFDmw8JfG+=ss@cKM_)C?cvx4E7KRO2~`Oi6p__cF&v;>PD^g!v-uf
zW{Ik78r+*cONMdPl6Qs?c}p^Dolo)Bo4ij}uy`2|wYxAb?z0Sfzq#PTAZY>fM-nl0
zm-yH6<%{R*!QM1~BgXz-IzK1LoQ6)2Zb1D<YN+nYTy8<tKNds2x;^YPQEQz>1`m$W
zTRkZW+jCs)c21>z=M~bAUNIEvZH_?XSTq}sJGc!-g9$Z8<R3%qfC0MY&OeJ;cB<vu
z@8*pmW|c`RHW0Yj9idi`vLN@6c8$!^NqeDfwhCmN6K<X%JjL)0;0>wPYXTNMtPSlc
zI02uR3-IyuCbGt)<)_+-L7PMfMNWZ42o0VoTS)Bz7#avDBg9^_bU#buU-$sE@GI(5
zT7rFhDss2HSwDV!9g8Whuq&$D!ZWss0pHQ>;Y9>JVlwTHj0jg&_0jKL<x36J0X)<X
zaKEm?asycSjaT1godq<sU<RZOyjU~eSwa4Z#$tfxVF%n;G#B=+pUh5^161<y@tEX3
z6L~gK1%Jol2B?_eQyW?z#)M;0u6N4PvI83RLNkga9xh1+!oOAC2pO;cP^YcK(&_8F
z9}LCsP0*=f_qOij1gzW(+*8bOYD||*sx;-`Q$OLD0Pm6O8r_^(N7muyJ6GHZL7DiU
z+90B2^KIABKrZPOb)lOD#z)6SD_8LqR))bk7LYwylXF^~j?UGS9?*_?-wYq^A<8A%
zpq#X;ctdn5=HJECoubFoo}$Y0l3q#ablqEzoljE_EU}~BJxt93@P6UL^r!~4o>dC3
zdEZ+4d_8$?pq#&i-)OeaZae`F@ZREY`>Y=Njm0;Qbr>ZK{1iC5mLA^KTQkC(s~5N{
zW19Kn3ay(zP3RbX!RT0~)DC}Yo)#Vp=;+QoyHq_1aJa&!y=R)4p)qh!&+Iw^ag+)d
z`xFaYHyodr9Xlng{t&nuqvHe^5R|~}W<alYPH=HVJI1;v$12}@GB#W;72EZMGEd?2
z_$HfMCo$s??A&Wl&VuKfJ_xw@kXKZpXP=T6LhIMLO#zX*5=xbVvuvZuuC}eQbBpPF
zFQe;EF%`-+tiwdBux*k$I;wu*&Li`D#}|*&nx_dISry?5&&{f`8)2hQv^#+qGS$dO
zV0{GVa?gNsYa}K+np$4i2#i)SKl`??0c)+T*H*(r6AS3<`DUi`isiV`botqLHlKhX
z-+d;u)tjCJ^_t9n#<OURs*qu-Z<f0l4SZMA;V~I|F`flQQ`2^QF4NiU4fuV^)eobR
zTw3<bFdX|onDyVWD(JiD_9@*i7w$dbyc;XL<ZvAh6twxO;c6tD+`E@(Ol8Mk9L&KV
zJE)7XfhoPWBN2RgdT-Qfi_yUNZaYdoJFNi7GrPTHRwvN}X<+|}ib3`|KN&Bg?nZ%c
zj!(Uauw7R-nDsr<7vo9>N7x=l6?sfiC3Cu>lELz@u=h_3xU2MV^uwr1`eM$dX-Hqr
zK*K~ehIi--OLi7kh0@x4v&1*d14Btz@+xCbvfMjtJG$dZ-%Mm5ac&<RvN8ngzWD`T
z_%jG9-^Vixx^CvWy)sjAz6^c8n+R!ODt9yTqxF=#-yhR`seB0fEH-<e=kKl+I6;Y^
z5(p5NRk8f=_}L$#W748X%v{<OOd`6$qn`Ic+1r&>1d(VuA~lrQU?Og$<Ei=Dr2sY0
zNmCcwy@WgWqxy%qvW6$ua~E?W$HM^5RArodNIm`UCXt>#G+)}9cIGU8?4e4<n~+-O
zsr&BkWwe#2-MIkfU_a91cSZ)Tmm$I9&l%+>TQ|R{u;+5CQA6BlmO?}*?_&8syf5{<
zQG-ID1Svuxi1%sUi9N?#82HndYX*i<hJlIrxaNNEos-W-Q~e9EUIF^Se;CAJ1N^UF
zT@;-iOx;Ytr#RC%nwqPcX)>}fu>+V`*=doOh22~&9h_-&**H1CH>WND08n`&u5ZNX
zjo7~tT>v}y1^~p~h}j!){|DK=A%{2O_(tsh$@ok2uY&$oSN$jNucGZ}nN_V^ZO#72
z=9T}^H~$AY|AYRrXYx<-Z=-T=oPSrv?VqGAEwhT7v9+0rD-A8PDEK^U*Z-|0Wd9NU
zx8s^`yuXY7s}ZdJk@}ncS3CUMqUoE!UkzkJ%PeE&<>BCL>Ouo<{J(5kIoOMVkJcuk
z5#wP6uyO!^EZi*Y04^Y#HdsQ=!Suh+RCP9TbTl&spUrJ#>tcq)tf(rk$t2-sYin#|
zZ~ykRRV}StNWj;>o;m}Gx|y>J*c=HP6AL>FkeihQ$i~RZ#PN4HfECO%No1|u%}BtW
zuriC88Cx0I(~{UY8o64!m>H3{|Cc@o6B`pac)-Vio7tPbEtzCx1J5e@?^jQ9{k?XQ
z^WPJZ*jWD`)*dIwIe>S;|8n+>!`*EpcR5R;tY(GZ$ddNUU4$PdOtDY|12OdZk<yCa
zuG8J;aG}berAXCAm1GDhP^ybuQOq;~AVwM9iII5liG)eHKrAo?g9206Of7kH&wXzq
zUAZ)HjQ}Y}QV|vcG6#Y|ByrQr$I`DPo2%v8?(>M6zWiiwhMdJEYw<MI_ag&8Y<Lu>
z3P`nM9bwpZH5y@ID3Ay28qjloyLq^LU7);``*C8&^oE04dxc{XP-YHcsQV5AkhkNO
zsKj*&#X)}IW7pY4l-W=G&~M!rJG-VM%NL*&1o`K+sDm6>$5c?5h!}$I_sXXgoE^of
z*D?Ob29Jp%-slCHnYrC2R#XZ#H-uZ0Q&;yPhLZVt>^GVPDL?F<xe2d^nx_-@Pd5z1
zka0IgtdYZJMBG0Yrm?YBh+#oh%vOnQ|LLGxHo*Tm*sG~*W{$-C$==k=lSBtV!pg=1
z&_iNYvGN9o1sEeSYmn%$kg$=kfVs*J4z6G>;BB8{<u@gBhqvV0%jYdAMxw*V$|}an
zCC16h&CLoBXXRoO;pCP8f{*Xy;NlWtXW?WQAo;(GfGz#K;Fg2+f4tz9GTq0OK+CUx
zkJ^et(3IK<QxZJG=X^PaINU_ii`(6pH~fUFW~=w@!m?~}Oc5bfq&e=+OJqW;NeYG6
zlwJ&BWGF28NVM}&Ta%`0wGjLeAs878VQ1uXcUa6YP*%tGGCm9Jd7L>=tbToL(qI!D
zbo#irn6~xheHUdly35CT77O2AQdXDa>O4HhQ74X)5Afk>D)C=oQ%4U@70F=fgeXg(
zW=T-2g%}qgI0EHzP;KFKI@J}?egq&dpyGF;R%0DQMdu*a2b#yRQ@0dAp1}FwZF4&c
zSU-Rlde{?`KqTiNsRp4KvMs>5gE9@@FF>aYzq5qR%z^(1n{W7$L4syJ2(>!9rar)Z
z0+L7g{XU$JFxoz>Uk>pxw097sBLs3FsUx~+7pWu6N7(*O02iu6Ck+>*d?1S>zAOmW
zCC1kv)&AQ%_kevsJ~32rmsT|dQ_xNtRtpX^&=6|@QCo-_4<b}}Tf2N4B&`TrCQQB$
z_iRY42{SLmZg!rGCl!ca4QuH?4*CRuE$Y$&K#xIEt!2Wx2o(d-m;z;lE)3uz4H0!A
zhd1bc!l`$<RU^%HqBvqT{d8@IoQukaL98tv`EajiwxzqkBC!s66G#Mr8Uwj(@cl$i
z5oQncWI@I3gl#Y+Sb$s-;@pQ1+Q7qu^EEwZPlIdqb0T60u-Ps58zGI52gFyvmGfl-
zPwI;N=IY1OZ?vWbY?dtK9?Px%JLXa5e~@{4@PT-@>;PHrk{z;D7xQGP!*g5CWJ_SO
zhUgyDwgG_$$41gz9UcgcYzvbi#J|xHf;)la*&)KYfCkrT1a4$_azpHI5XGHVj&QP2
z0*#<xfuj|R%tvr}Zl`r)esfLZJhxhV|B?A()M<RHbu19MYEkpZ;GW%gTDRbb+cIu^
zs}SHcVKgRwa`A$<#rxK-8+2j5?$f9xP<n>MTrl#W#0}&$n0Rjcm>?H$v~dL4kwCi#
zt{S54!?X#_gOZQy68zL_$E9zy1_kIB=f!7m$NmdjP(GHMX8$h&y)VJsk+Da<S_>~$
za~ovYyb!2?`G)u-AZ|xU29S+mw8J2UQ~;79B1ZsE=lvJVG{zo_;XjM4zCVz&qq_|_
zEbck+hv|CKwpv75`s7^HUng@9($rS+2Vk9CPEXX>Ag4$k5?XF!XHuaEgrm6|=Vmm9
zd9xvw`W2=cVQpFWaNu2~2*WA!kQc1So;}uMtIC(74O_U;kjttd>jS7<>>ZTX?I+$X
zje3cuKHL?xtim>T=}(}+31xFTc(ueV_lIVaE}`AIH~Dpb^We`vCRwy{3^B+3!|3XR
z_mO??xns&nk}H_Ga5=P-#S!WwXYoX(s6;0Wm(^JDT5&r|4qB}H4mJKvvB&23lbZr=
zgG#k^><NH**jUg9mv(<;tyLukEIo+2UEfPJX-g4iaEPaRu6;Wbs9-QiZ@hnK^~2F+
z*^?xf-mj<(vSfa#nlE(-8K2m90I9k7dx<038xM)?;EfJ^v$(!7+rrTp6?k2@&QsvF
z|FKC51n<#lr@(FX!%o$2UuAc#Sl4Qfy||rjZ3laz(T`Co!|g?P5nmCWBMrq@)bCDX
z4Q8TzWK_kgZZTegz9qS_S)!iMQA2GkE!s{Mw`mALW6g?#KUcuMacIRpIb+0}`8(R?
z(27fP#whOLOuwJzSDk9-jKx{l>jvm2dKVq_Q)oT&aW7e|D8@50A$qat6+_2)4T(wT
zj>R^?Ij6@?VaLwnLSof*lFo)&`v~;PZ@`ug9j(*XAxQ|%fE}Ycc#C8%R6tFX_YRi1
z!#CwXH!XAj;eAaaPM`fI>HQ^g(GS-{iRMSVNu7hNR4J3?d}w1|+Zx5SWEsYxR(<u^
zeLlBgrfEw2yhvCKJ-ey-G4v2+V@~`S{^0t_K2WVr;j^lr^9dZZ_6`vfY97jAX{rmk
z$1uaqA*DEzR6y_|r^IY3%*yU+8Q^h@bGE-?y+T{j?xyN<Svz|PbiA^lzpNiYmpXd*
zm2C9G4?q+!?<8mK_?0SY0gl!1y=ArLZz4)}&}z=4af8XB4OE3;Y~LYgS=EPnXnRqI
z2Cedg1Wdx5cUJ*zxJRDS;Y1U|AbWJDaqlCD{YK68*ZsKX4HCS1%^zxhq8kVU+Q+@C
z?V=l8T-G&%+Si28J%*T3n#a01z5jam0#K<sX&v#uLGJt6;C_Qv!EJdWY8%!9^z1ue
z>OrJ-c}1??#iM?OciF(ygaH58HR~5H8xMcd3ynk43#I7REkqV2x1s(Np(NL?TNr+k
z4bQ+Oo+?ejBr&fMe9BA5Q}6zU$&u`szpZ_YvaZar)J}I1fnK0$Qq$NbA(uqAsfcJL
zoxyO^pD7LQjvxY)4jFIKBosx`m7zNmriiLnjfvYIS+k6#7gW1bF;SD3gzEYPN+UIY
ze<+?A5;u*JE>Zp+UtIsbUgx;3qiU-ekFD6?5Y|%Zux3HyMRZYM4fI_H9|jutaVa&q
zvq%{iS?hTDOqrM?*)xrf5LQzeA2O8ThS_13R7$108dEyCx`5rfCg+pZ2yL#k)rhg6
zo(3@|q8HI`X(<&Rvl<Kk5*V`NdU1K}6|b0wA_$J;F>Ss=H?4F??`B?y^FztmP&dje
zT2c+f3~3%2Us}D|yOi+*u5$WKHAFw@VO)o&1Wb(&$)<4v^Nee2RlRg*VtnQ0Gz8%@
zk0t5m=h?PuNl=K9#E&*WFJ0|YC>&hkDs=<O4GinTrQg#Iss{MBPy`GDuVmMDrg(g=
z6}GT$xE)g^z2==;^U}~Ijdx4*$pfGhaMx^f?$eySfOrFvh<sgZdyAo*#owQ+h5Z<F
zu<Es*8nRUlczY(it^%1NJ6@)x|MWK2qFre{k@ELe{DJt%3h|}WKeEF_EgTv|-IL6)
zaqhU2(?`5od`#+y_YqRE3lU&@-haun&4mZMwNcRb`Wq-6#Uj9;oeia8*!?@(IeDkc
z#SHTsY@OlK@9Y+zaEne@M~o4zr$OrE4M;aL46*j7RnbEoi22Thz78KPKMhd5X3ZVi
z;jBj{<X{j1-~8d`9Oz+=;U$0kyFy4vhS(>W*4dd*f<9{yMjr|ZEA_ET3X~W|Mns8n
zZ_saW9nbuR<{(vRe+xTBQerZz<`LxxIkiw)7te-r_ep@@`^>9{F|R4jlnQ@lT$Ye@
zhZ*x)*7Tl!APnRdKZNf?m2vfx^3wx#i(oKDw52Q%(@ero<b~n(J*N!F(+1#*N73X7
z#V<0m4vH7GcurvUI7~X>DT&3A)fs%O=`HO~;zIP?7&w|z`b8tJX>6Zgi>GAzJ2Ejv
z^C|fOy&w#}?}JSFQ_usAFE(=kw+!E{(RHpk*b=sA#E7d?7v`<CIF3`kCn1B*Z_2tu
zll1{zQ-*Z%(U&sH)9;Gz(IOBo`mI*s^K!~414!nl4bu_2=#9NS2C8h3bV3vZkCkQ2
zuv0lxn8XhR*J=UULPn%I@YW3H=T$T^nWQZNxpFM*YMMU+M^ik1D-qVh+QMI_+zov$
zb0AoN#b-Fb_B&keY8k;gfs9VM8+(>!hm3f?XfE>vs1QKHUqqcS56=dcDENd+tiyT>
z7gPHC>1kOLEfQ6uU46~C%ffs%L_g-_e1^PM!W}g4ifj4`KjVR}NH<6XtNN)nbY%6?
z%Pbbsnnuf-W--xlW~E)tY)27g5(ba$_D^Io9z3>2E%;I~#8+$`$YS-lI&I2?P13J@
z-&x*NoTY4xs<=KdmciRlA^^52UTt_f;&<4BoYIxhR7b3|jw4Z6ss~#ZuLtyaF&ol{
z=A=pQ4Xey(`BRO_*_Gb(xkE>sx`4KK^tqxS-BYczK?D;(9D?LxpqOt<DO3tohJE*^
z+g#D)v?BW;4h3QqNmF^j`1HGfF3h-&msw3INFXGtsSyYnLYM=oNJGz_{<vEwouiOT
zc+7H3c&22)whx1+v=81LNk-aGK7dRo7i~kr2tm<>{wn=3z@IQw6gG;6MAT?vMX-Z+
zK$K)yN-eY}sGG$_kSB*hyamcd->~RiJbZRiCTu;;NT8L=>!Hn|$=Ewt4<pEtua6;Z
z($!K(G8?r`3c&BN2TTC75c*w#ib5StbG$Mb){YPymyR+@t)KEnGJV~a_^%v=?F1=L
z>albti0Ev$EhyH%APaNrZB9GnN@AUr(!Di#(wDLsO2NleTV&^#;1-Pf6)BD4e^Ig+
zyzwY&!cf$ac7P{fSiGj9qyZMv0t*{K6*)wTRx&D9GG|aIDIdd*aCjkR8(pz1;i6Oc
zyhXI=oUQ1b4>%@_2hTYuwT0mq-Yqe5U&2zBtQ59D?zFfPFQ9odH2o9}y>em2-zLGU
z-aed`{s=L%G9}p=M|pN1gQ?|+ha8lj;yb93%mpXWiN|F>m@1gxOOpov&#DEmADnO3
zS)4XRQk<8`hG$Fhowi6W1SQU2=Z~ajH57Pw_$IwfJu#&0TGX(+bziko5_5lN*y+47
zt42;?xHnSyjTFbM!FV4IKO?ya@PjVbz6f7eB3A4DhOt_|t#Oa<;(}_jXk7~bFw^^8
zcO+r`yE;~tvuCmk<NevXI<fwawNMCU5X-ev<Or!;<y?Z`WAwvSbdT3I*X#<F%!~GW
z>QGVc)*<oijJvRh!w+y|I}_CNhNbBSGywwK1%p8CvQwvPQe{*FXT}t3PZ7VngnZR+
zHjOi58sE|Uoc9xt;8#<u+j7Owgubi@$Xv&iI0%v}w&8tn4ocy9%E1(cU~~>;0KRx%
z=Sri7IwK_HQ)Tknm%1>k;~KoB#4>rUN(&jKHx%tZ8p@-(IF}^w!c#H2ZRN!DeETT9
zv)kn*vwCSxdAG<`<;_8~&7^&(e*({~XTdkXsC@`q`H%vV-Wpa~znt;F$q_$oF4TWu
zy{=H^rb2QekAkEU5c{HtbDfdmz0->7B%E9*pjk>q6nsp%3D+qbY9+&eM{9Ns(IXow
zYd6p-3M7C~A*c=O83^L!IKmn6A9(t5j7)loQOh4pgZ=};oX1oYQHjJBe#CWP3_VY1
z;0B%boRp1_^^w#^Olp$f6t`PIrtSHYABB(D07WQRwT}40MRZ^s<xOas)a)Vd05zVJ
z4>j!|?klPiiCAgeI+e&Sr>V6lHp16mjw{M%v(AY3lVqj|qCY~xBBjrvco<}&11T6|
zP#hs~(PUA(t=-cE)oj6&PwM5DCZ7g+Lwys=*&0X_KkDa~uA?ZDY=1S?V`D6zG#y5`
zj>uro-%7ZQDHIrTI2-grYCf6|GX@!;%~xSkf@EC2(jY3cSD~wcq|;@@)N;~y$&~Q_
znAQ1zEy1l;DrZKwgUPVeCo6-_u+pbkg8JE|nN3{=zig>bG6Xtd-wVTsX#9RuC&t^w
zl)e{^FN`(nBGv-gF;_K?4h+Igw+Po3CK$EKbDt1)i73HRG}p?eIE2IYnjc5MhLCnU
zDh_{uFi<!u4!eQm$ht<%gLDM#a!r!p<#9{6qC3vl`X}t8bID%9Oky}D?b1xL0AyW)
z_qp(LOQk{`J6Z%}3vRb)z4e%<Llq|YqzmFNT4tLCM2mr+E&=hCkb-LEHbTvY*tc_a
zV$IEaR@PVP5!~sbV(E=u={#z37nJGnVWu!cRY%QI#IzmRQk9>)^u)qxcOJtFYLn)1
z3}BVpQ}^NGE@TzZX<t*5@Dc)&)@6b(0+K95pp(!FeBWaX!7FDiqA3c*T@Z@V5KDQf
zBEDg}zi=zk8)p0qQy{-#pRhNIN08%7vXk)_-Y0y+9vGzxL-e_vaTmN{S60gUXL-4w
z?oY&c&}ge-XFkc?+le)cDC=)c1CuXw!qY%PLr11*6+%M}RL0iOv{h{VL10WKk`Csx
z8flX3aH8j$knFh0Ie`lViiHtLdC5V59c@oX*a3?#irfngX<+7p?E)LXKHp=%)k2@>
z%@I8^aJ5<sguxR1;cvdOB7;>N{;I-3*ahU!o&#7hXAc$zRZ&3b{s|wfTE=JOLz3hB
z!V|9_aB;u?bH;{jN3d>@ROJ2%n1=@zQp<w@Q&uHRun>|GEEr=d;eaVP%=Y7dvnnxy
zG1R04=;B^R%nf4a@_qOZ2!4q%tXb%!aRZVeh@|mx4B#8ZJb|If7F3u)P!#waW623t
zLsnt~Q>{F8urfHtz?3#mKF9chLg_8{Lmm;BGcD#Pghu)nV5|D_=qTWVVzE%5z!ZZJ
z#+!rxlPdz)FK~Q;MV0bQ!1l?-`oR^&IY3tfehI(wFH1f&dvHt|C#IZ|QWB^dpRmgH
zeR+$np*<jp9w2GlBHYytoH1!brzjvOR^TOi-zu+`7<F+EEyfD9bJ@%^B{_6PonWJ^
z|6a*h`re<clwrrMPa`1djItTBz!yz{EM}H3cF9~^V(7?Y0Wa=iRqVkcH_4)i?Gwok
zZ}{CkHTljYS-wG8As?_p7GKKi<IuhXdcIJsu&nf#mQCckWZ{>n7dR(mlFz$b6ID?U
zqv`N@$ZHm?5IbxrLX=$C1II8A)KEhf_>g6$iUOkN9QmI~IH1MG>$MLf9p-?Rbu$XY
z8Z-7dXydep)k8%FG>7#=#I1@-scN&?>N9p~&^|aP4P}>TG>0{1kJ{MR+OrvlIM-UU
zCTHq14r<vS)a%E~GGWCZE1L7QCP8W)dFqpp-~9_DCv((ZFZLq7Ftpj}P72F(97g4p
z7j`rh?Z?f!KDn^x`eYO5GDiW>?W};JDj2m@v5wnz7+<oZ9Dgg2$$pINBxbZ8^X2QF
z*&`eI7!6p)H)`kTF0>$H@E7d9n8v8YT75DLUVlj-if_8{I$`In*nRORQ{Y8O<*$;H
z*l|8#);(|ik)r<H%aOu6HPnBIo^R%|daQ>BZIdtf(s8s0h&IZ{ehC=q0iZSUZD0Hu
z5~W4U;1jz*AM7DT^W(cZpXe7Q{tIxy#;W`Q^qX7}<=^79_3!gboUwSvPWn7`?r+wG
zZV|2dl*7heZ`PRe&BgaaT`s)x`72#5Jn+3(Zq*!lL4#1nx6Xh>KA9fq)uUda5KG=A
z$3{Q;%l>H65`FJBx%OnBid`MyMO+x*^XI=>=;3?$u~FmaWdXsPD%r4XGJP=ybweLJ
z71^)8b53(oBjfcRf*?uqV9mtg0vD9khzrS^AW1uC!gf(|>2oYf^qP3S5XkqUyi@b&
z`DD^%vekqo_w-0>p|)c5`wVdIk#@%``$S0ff{OK-+6)-$OU83o^V#`@SR|0p?*IH4
d=HhDP?CR+Z9@GZ!U(C+RjYLT)t{{Q*{{dH`gyH}I

literal 0
HcmV?d00001

diff --git a/docs/2_39/reference-manual/analysis.html b/docs/2_39/reference-manual/analysis.html
new file mode 100644
index 000000000..70de8e8bb
--- /dev/null
+++ b/docs/2_39/reference-manual/analysis.html
@@ -0,0 +1,1359 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Posterior Analysis</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/optimization.html" rel="next">
+<link href="../reference-manual/mcmc.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Posterior Analysis">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/analysis.html">Posterior Analysis</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#analysis.chapter" id="toc-analysis.chapter" class="nav-link active" data-scroll-target="#analysis.chapter">Posterior Analysis</a>
+  <ul class="collapse">
+  <li><a href="#markov-chains" id="toc-markov-chains" class="nav-link" data-scroll-target="#markov-chains">Markov chains</a></li>
+  <li><a href="#convergence" id="toc-convergence" class="nav-link" data-scroll-target="#convergence">Convergence</a></li>
+  <li><a href="#notation-for-samples-chains-and-draws" id="toc-notation-for-samples-chains-and-draws" class="nav-link" data-scroll-target="#notation-for-samples-chains-and-draws">Notation for samples, chains, and draws</a>
+  <ul class="collapse">
+  <li><a href="#potential-scale-reduction" id="toc-potential-scale-reduction" class="nav-link" data-scroll-target="#potential-scale-reduction">Potential scale reduction</a></li>
+  <li><a href="#split-r-hat-for-detecting-non-stationarity" id="toc-split-r-hat-for-detecting-non-stationarity" class="nav-link" data-scroll-target="#split-r-hat-for-detecting-non-stationarity">Split R-hat for detecting non-stationarity</a></li>
+  <li><a href="#rank-normalization-helps-when-there-are-heavy-tails" id="toc-rank-normalization-helps-when-there-are-heavy-tails" class="nav-link" data-scroll-target="#rank-normalization-helps-when-there-are-heavy-tails">Rank normalization helps when there are heavy tails</a></li>
+  <li><a href="#convergence-is-global" id="toc-convergence-is-global" class="nav-link" data-scroll-target="#convergence-is-global">Convergence is global</a></li>
+  </ul></li>
+  <li><a href="#effective-sample-size.section" id="toc-effective-sample-size.section" class="nav-link" data-scroll-target="#effective-sample-size.section">Effective sample size</a>
+  <ul class="collapse">
+  <li><a href="#definition-of-effective-sample-size" id="toc-definition-of-effective-sample-size" class="nav-link" data-scroll-target="#definition-of-effective-sample-size">Definition of effective sample size</a></li>
+  <li><a href="#estimation-of-effective-sample-size" id="toc-estimation-of-effective-sample-size" class="nav-link" data-scroll-target="#estimation-of-effective-sample-size">Estimation of effective sample size</a></li>
+  <li><a href="#estimation-of-mcmc-standard-error" id="toc-estimation-of-mcmc-standard-error" class="nav-link" data-scroll-target="#estimation-of-mcmc-standard-error">Estimation of MCMC standard error</a></li>
+  <li><a href="#thinning" id="toc-thinning" class="nav-link" data-scroll-target="#thinning">Thinning</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/analysis.html">Posterior Analysis</a></li></ol></nav></header>
+
+
+
+
+<section id="analysis.chapter" class="level1">
+<h1>Posterior Analysis</h1>
+<p>Stan uses Markov chain Monte Carlo (MCMC) techniques to generate draws from the posterior distribution for full Bayesian inference. Markov chain Monte Carlo (MCMC) methods were developed for situations in which it is not straightforward to make independent draws <span class="citation" data-cites="Metropolis:1953">Metropolis et al. (<a href="#ref-Metropolis:1953" role="doc-biblioref">1953</a>)</span>.</p>
+<p>Stan’s variational inference algorithm provides draws from the variational approximation to the posterior which may be analyzed just as any other MCMC output, despite the fact that it is not actually a Markov chain.</p>
+<p>Stan’s Laplace algorithm produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood.</p>
+<section id="markov-chains" class="level2">
+<h2 class="anchored" data-anchor-id="markov-chains">Markov chains</h2>
+<p>A <em>Markov chain</em> is a sequence of random variables <span class="math inline">\(\theta^{(1)},
+\theta^{(2)},\ldots\)</span> where each variable is conditionally independent of all other variables given the value of the previous value. Thus if <span class="math inline">\(\theta = \theta^{(1)}, \theta^{(2)},\ldots, \theta^{(N)}\)</span>, then</p>
+<p><span class="math display">\[
+p(\theta) = p(\theta^{(1)}) \prod_{n=2}^N p(\theta^{(n)}|\theta^{(n-1)}).
+\]</span></p>
+<p>Stan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the <a href="../reference-manual/mcmc.html">Hamiltonian Monte Carlo chapter</a>.</p>
+<p>The Markov chains Stan and other MCMC samplers generate are <em>ergodic</em> in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of <span class="math inline">\(\theta\)</span> from another. The Markov chains are also <em>stationary</em>, meaning that the transition probabilities do not change at different positions in the chain, so that for <span class="math inline">\(n, n' \geq 0\)</span>, the probability function <span class="math inline">\(p(\theta^{(n+1)}|\theta^{(n)})\)</span> is the same as <span class="math inline">\(p(\theta^{(n'+1)}|\theta^{(n')})\)</span> (following the convention of overloading random and bound variables and picking out a probability function by its arguments).</p>
+<p>Stationary Markov chains have an <em>equilibrium distribution</em> on states in which each has the same marginal probability function, so that <span class="math inline">\(p(\theta^{(n)})\)</span> is the same probability function as <span class="math inline">\(p(\theta^{(n+1)})\)</span>. In Stan, this equilibrium distribution <span class="math inline">\(p(\theta^{(n)})\)</span> is the target density <span class="math inline">\(p(\theta)\)</span> defined by a Stan program, which is typically a proper Bayesian posterior density <span class="math inline">\(p(\theta | y)\)</span> defined on the log scale up to a constant.</p>
+<p>Using MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.</p>
+<p>Stan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.</p>
+</section>
+<section id="convergence" class="level2">
+<h2 class="anchored" data-anchor-id="convergence">Convergence</h2>
+<p>By definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when <span class="math inline">\(p(\theta^{(n)})\)</span> is the target density). The following point cannot be expressed strongly enough:</p>
+<ul>
+<li><p>In theory, <em>convergence is only guaranteed asymptotically</em> as the number of draws grows without bound.</p></li>
+<li><p>In practice, <em>diagnostics must be applied to monitor convergence</em> for the finite number of draws actually available.</p></li>
+</ul>
+</section>
+<section id="notation-for-samples-chains-and-draws" class="level2">
+<h2 class="anchored" data-anchor-id="notation-for-samples-chains-and-draws">Notation for samples, chains, and draws</h2>
+<p>To establish basic notation, suppose a target Bayesian posterior density <span class="math inline">\(p(\theta | y)\)</span> given real-valued vectors of parameters <span class="math inline">\(\theta\)</span> and real- and discrete-valued data <span class="math inline">\(y\)</span>.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>An MCMC <em>sample</em> consists of a set of a sequence of <span class="math inline">\(M\)</span> Markov chains, each consisting of an ordered sequence of <span class="math inline">\(N\)</span> <em>draws</em> from the posterior.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> The sample thus consists of <span class="math inline">\(M \times N\)</span> draws from the posterior.</p>
+<section id="potential-scale-reduction" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="potential-scale-reduction">Potential scale reduction</h3>
+<p>One way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the <span class="citation" data-cites="GelmanRubin:1992">Gelman and Rubin (<a href="#ref-GelmanRubin:1992" role="doc-biblioref">1992</a>)</span> potential scale reduction statistic, <span class="math inline">\(\hat{R}\)</span>. The <span class="math inline">\(\hat{R}\)</span> statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and <span class="math inline">\(\hat{R}\)</span> will be one. If the chains have not converged to a common distribution, the <span class="math inline">\(\hat{R}\)</span> statistic will be greater than one.</p>
+<p>Gelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for <span class="math inline">\(\hat{R}\)</span> are below some threshold. <span class="citation" data-cites="Vehtari+etal:2021:Rhat">Vehtari et al. (<a href="#ref-Vehtari+etal:2021:Rhat" role="doc-biblioref">2021</a>)</span> suggest in general to use a threshold <span class="math inline">\(1.01\)</span>, but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.</p>
+<p>The <span class="math inline">\(\hat{R}\)</span> statistic is defined for a set of <span class="math inline">\(M\)</span> Markov chains, <span class="math inline">\(\theta_m\)</span>, each of which has <span class="math inline">\(N\)</span> draws <span class="math inline">\(\theta^{(n)}_m\)</span>. The <em>between-chain variance</em> estimate is</p>
+<p><span class="math display">\[
+B
+=
+\frac{N}{M-1}
+\,
+\sum_{m=1}^M (\bar{\theta}^{(\bullet)}_{m}
+                - \bar{\theta}^{(\bullet)}_{\bullet})^2,
+\]</span></p>
+<p>where</p>
+<p><span class="math display">\[
+\bar{\theta}_m^{(\bullet)}
+= \frac{1}{N} \sum_{n = 1}^N \theta_m^{(n)}
+\]</span></p>
+<p>and</p>
+<p><span class="math display">\[
+\bar{\theta}^{(\bullet)}_{\bullet}
+= \frac{1}{M} \, \sum_{m=1}^M \bar{\theta}_m^{(\bullet)}.
+\]</span></p>
+<p>The <em>within-chain variance</em> is averaged over the chains,</p>
+<p><span class="math display">\[
+W = \frac{1}{M} \, \sum_{m=1}^M s_m^2,
+\]</span></p>
+<p>where</p>
+<p><span class="math display">\[
+s_m^2
+=
+\frac{1}{N-1}
+\, \sum_{n=1}^N (\theta^{(n)}_m - \bar{\theta}^{(\bullet)}_m)^2.
+\]</span></p>
+<p>The <em>variance estimator</em> is a mixture of the within-chain and cross-chain sample variances,</p>
+<p><span class="math display">\[
+\widehat{\mbox{var}}^{+}\!(\theta|y)
+= \frac{N-1}{N}\, W \, + \, \frac{1}{N} \, B.
+\]</span></p>
+<p>Finally, the <em>potential scale reduction statistic</em> is defined by</p>
+<p><span class="math display">\[
+\hat{R}
+\, = \,
+\sqrt{\frac{\widehat{\mbox{var}}^{+}\!(\theta|y)}{W}}.
+\]</span></p>
+</section>
+<section id="split-r-hat-for-detecting-non-stationarity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="split-r-hat-for-detecting-non-stationarity">Split R-hat for detecting non-stationarity</h3>
+<p>Before Stan calculating the potential-scale-reduction statistic <span class="math inline">\(\hat{R}\)</span>, each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have <span class="math inline">\(\hat{R}\)</span> values near unity. In this case, splitting each chain into two parts leads to <span class="math inline">\(\hat{R}\)</span> values substantially greater than 1 because the first half of each chain has not mixed with the second half.</p>
+</section>
+<section id="rank-normalization-helps-when-there-are-heavy-tails" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="rank-normalization-helps-when-there-are-heavy-tails">Rank normalization helps when there are heavy tails</h3>
+<p>Split R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following <span class="citation" data-cites="Vehtari+etal:2021:Rhat">Vehtari et al. (<a href="#ref-Vehtari+etal:2021:Rhat" role="doc-biblioref">2021</a>)</span>, we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.</p>
+<p>Rank normalization proceeds as follows:</p>
+<ul>
+<li><p>First, replace each value <span class="math inline">\(\theta^{(nm)}\)</span> by its rank <span class="math inline">\(r^{(nm)}\)</span> within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.</p></li>
+<li><p>Second, transform ranks to normal scores using the inverse normal transformation and a fractional offset:</p></li>
+</ul>
+<p><span class="math display">\[
+z_{(nm)} = \Phi^{-1} \left( \frac{r_{(nm)} - 3/8}{S - 1/4} \right)
+\]</span></p>
+<p>To further improve sensitivity to chains having different scales,</p>
+<p>rank normalized R-hat is computed also for the for the corresponding <em>folded</em> draws <span class="math inline">\(\zeta^{(mn)}\)</span>, absolute deviations from the median, <span class="math display">\[
+\label{zeta}
+\zeta^{(mn)} = \left|\theta^{(nm)}-{\rm median}(\theta)\right|.
+\]</span> The rank normalized split-<span class="math inline">\(\widehat{R}\)</span> measure computed on the <span class="math inline">\(\zeta^{(mn)}\)</span> values is called -<span class="math inline">\(\widehat{R}\)</span>. This measures convergence in the tails rather than in the bulk of the distribution.</p>
+<p>To obtain a single conservative <span class="math inline">\(\widehat{R}\)</span> estimate, we propose to report the maximum of rank normalized split-<span class="math inline">\(\widehat{R}\)</span> and rank normalized folded-split-<span class="math inline">\(\widehat{R}\)</span> for each parameter.</p>
+<p>Bulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See <a href="#effective-sample-size.section">Effective Sample Size section</a> for details on how ESS is estimated.</p>
+</section>
+<section id="convergence-is-global" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="convergence-is-global">Convergence is global</h3>
+<p>A question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.</p>
+<p>For example, consider the value <code>lp__</code>, which is the log posterior density (up to a constant).<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>It is thus a mistake to declare convergence in any practical sense if <code>lp__</code> has not converged, because different chains are really in different parts of the space. Yet measuring convergence for <code>lp__</code> is particularly tricky, as noted below.</p>
+<section id="asymptotics-and-transience-vs.-equilibrium" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="asymptotics-and-transience-vs.-equilibrium">Asymptotics and transience vs.&nbsp;equilibrium</h4>
+<p>Markov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.</p>
+</section>
+<section id="multivariate-convergence-of-functions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="multivariate-convergence-of-functions">Multivariate convergence of functions</h4>
+<p>The <span class="math inline">\(\hat{R}\)</span> statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.</p>
+<p>The transformation from unconstrained space to constrained space is just another function, so does not effect convergence.</p>
+<p>Different functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, <code>lp__</code> and other measured quantities should at least be consistent.</p>
+<p>The obvious difference in <code>lp__</code> is that it tends to vary quickly with position and is consequently susceptible to outliers.</p>
+</section>
+<section id="finite-numbers-of-states" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="finite-numbers-of-states">Finite numbers of states</h4>
+<p>The question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.</p>
+<p>So what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.</p>
+</section>
+<section id="why-inconsistent-r-hat" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="why-inconsistent-r-hat">Why inconsistent R-hat?</h4>
+<p>Firstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better <span class="math inline">\(\hat{R}\)</span>. Functions vulnerable to such outliers may show fragility.</p>
+<p>Secondly, use of the <span class="math inline">\(\hat{R}\)</span> statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (<code>lp__</code>) almost never looks Gaussian, instead it features long tails that can lead to large <span class="math inline">\(\hat{R}\)</span> even in the large <span class="math inline">\(N\)</span> limit. Tweaks to <span class="math inline">\(\hat{R}\)</span>, such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the <span class="math inline">\(\hat{R}\)</span> statistic more accurate.</p>
+</section>
+<section id="final-words-on-convergence-monitoring" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="final-words-on-convergence-monitoring">Final words on convergence monitoring</h4>
+<p>“Convergence” is a global property and holds for all integrable functions at once, but employing the <span class="math inline">\(\hat{R}\)</span> statistic requires additional assumptions and thus may not work for all functions equally well.</p>
+<p>Note that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.</p>
+</section>
+</section>
+</section>
+<section id="effective-sample-size.section" class="level2">
+<h2 class="anchored" data-anchor-id="effective-sample-size.section">Effective sample size</h2>
+<p>The second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see <span class="citation" data-cites="Geyer:2011">Charles J. Geyer (<a href="#ref-Geyer:2011" role="doc-biblioref">2011</a>)</span>.</p>
+<p>Stan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).</p>
+<p>Unlike most packages, the particular calculations used by Stan follow those for split-<span class="math inline">\(\hat{R}\)</span>, which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> and <span class="citation" data-cites="Vehtari+etal:2021:Rhat">Vehtari et al. (<a href="#ref-Vehtari+etal:2021:Rhat" role="doc-biblioref">2021</a>)</span>.</p>
+<section id="definition-of-effective-sample-size" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="definition-of-effective-sample-size">Definition of effective sample size</h3>
+<p>The amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size <span class="math inline">\(N\)</span>. Given dependent sample, the sample size is replaced with the effective sample size <span class="math inline">\(N_{\mathrm{eff}}\)</span>.<br>
+For example, Monte Carlo standard error (MCSE) is proportional to <span class="math inline">\(1 / \sqrt{N_{\mathrm{eff}}}\)</span> rather than <span class="math inline">\(1/\sqrt{N}\)</span>.</p>
+<p>The effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation <span class="math inline">\(\rho_t\)</span> at lag <span class="math inline">\(t \geq 0\)</span> for a chain with joint probability function <span class="math inline">\(p(\theta)\)</span> with mean <span class="math inline">\(\mu\)</span> and variance <span class="math inline">\(\sigma^2\)</span> is defined to be</p>
+<p><span class="math display">\[
+\rho_t
+=
+\frac{1}{\sigma^2} \, \int_{\Theta} (\theta^{(n)} - \mu)
+(\theta^{(n+t)} - \mu) \, p(\theta) \, d\theta.
+\]</span></p>
+<p>This is the correlation between the two chains offset by <span class="math inline">\(t\)</span> positions (i.e., a lag in time-series terminology). Because we know <span class="math inline">\(\theta^{(n)}\)</span> and <span class="math inline">\(\theta^{(n+t)}\)</span> have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields</p>
+<p><span class="math display">\[
+\rho_t
+= \frac{1}{\sigma^2}
+\, \int_{\Theta}
+      \theta^{(n)} \, \theta^{(n+t)} \, p(\theta)
+   \, d\theta - \frac{\mu^2}{\sigma^2}.
+\]</span></p>
+<p>The effective sample size of <span class="math inline">\(N\)</span> draws generated by a process with autocorrelations <span class="math inline">\(\rho_t\)</span> is defined by <span class="math display">\[
+N_{\mathrm{eff}}
+\ = \
+\frac{N}{\sum_{t = -\infty}^{\infty} \rho_t}
+\ = \
+\frac{N}{1 + 2 \sum_{t = 1}^{\infty} \rho_t}.
+\]</span></p>
+<p>For independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.</p>
+</section>
+<section id="estimation-of-effective-sample-size" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimation-of-effective-sample-size">Estimation of effective sample size</h3>
+<p>In practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-<span class="math inline">\(\hat{R}\)</span> based effective sample size estimator, based on multiple chains. As before, each chain <span class="math inline">\(\theta_m\)</span> will be assumed to be of length <span class="math inline">\(N\)</span>.</p>
+<p>Stan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see <span class="citation" data-cites="Geyer:2011">Charles J. Geyer (<a href="#ref-Geyer:2011" role="doc-biblioref">2011</a>)</span> for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates <span class="math inline">\(\hat{\rho}_{t,m}\)</span> at lag <span class="math inline">\(t\)</span> from multiple chains <span class="math inline">\(m \in
+(1,\ldots,M)\)</span> are combined with within-sample variance estimate <span class="math inline">\(W\)</span> and multi-chain variance estimate <span class="math inline">\(\widehat{\mbox{var}}^{+}\)</span> introduced in the previous section to compute the combined autocorrelation at lag <span class="math inline">\(t\)</span> as</p>
+<p><span class="math display">\[
+\hat{\rho}_t
+= 1 - \frac{\displaystyle W
+              - \textstyle \frac{1}{M}\sum_{m=1}^M s_m^2 \hat{\rho}_{t,m}}
+        {\widehat{\mbox{var}}^{+}}.
+\]</span></p>
+<p>If the chains have not converged, the variance estimator <span class="math inline">\(\widehat{\mbox{var}}^{+}\)</span> will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.</p>
+<p>Because of the noise in the correlation estimates <span class="math inline">\(\hat{\rho}_t\)</span> as <span class="math inline">\(t\)</span> increases, a typical truncated sum of <span class="math inline">\(\hat{\rho}_t\)</span> is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise <span class="citation" data-cites="Geyer:1992">Charles J. Geyer (<a href="#ref-Geyer:1992" role="doc-biblioref">1992</a>)</span>, <span class="citation" data-cites="Geyer:2011">Charles J. Geyer (<a href="#ref-Geyer:2011" role="doc-biblioref">2011</a>)</span>. Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as</p>
+<p><span class="math display">\[
+\hat{N}_{\mathrm{eff}} = \frac{M \cdot N}{\hat{\tau}},
+\]</span></p>
+<p>where</p>
+<p><span class="math display">\[
+\hat{\tau} = 1 + 2 \sum_{t=1}^{2m+1} \hat{\rho}_t = -1 + 2 \sum_{t'=0}^{m} \hat{P}_{t'},
+\]</span></p>
+<p>where <span class="math inline">\(\hat{P}_{t'}=\hat{\rho}_{2t'}+\hat{\rho}_{2t'+1}\)</span>. Initial positive sequence estimators is obtained by choosing the largest <span class="math inline">\(m\)</span> such that <span class="math inline">\(\hat{P}_{t'}&gt;0, \quad t' = 1,\ldots,m\)</span>. The initial monotone sequence is obtained by further reducing <span class="math inline">\(\hat{P}_{t'}\)</span> to the minimum of the preceding ones so that the estimated sequence is monotone.</p>
+</section>
+<section id="estimation-of-mcmc-standard-error" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimation-of-mcmc-standard-error">Estimation of MCMC standard error</h3>
+<p>The posterior standard deviation of a parameter <span class="math inline">\(\theta_n\)</span> conditioned on observed data <span class="math inline">\(y\)</span> is just the standard deviation of the posterior density <span class="math inline">\(p(\theta_n | y)\)</span>. This is estimated by the standard deviation of the combined posterior draws across chains,</p>
+<p><span class="math display">\[
+\hat{\sigma}_n = \mathrm{sd}(\theta^{(1)}_n, \ldots, \theta^{(m)}_n).
+\]</span></p>
+<p>The previous section showed how to estimate <span class="math inline">\(N_{\mathrm{eff}}\)</span> for a parameter <span class="math inline">\(\theta_n\)</span> based on multiple chains of posterior draws.</p>
+<p>The mean of the posterior draws of <span class="math inline">\(\theta_n\)</span> <span class="math display">\[
+\hat{\theta}_n
+= \mathrm{mean}(\theta^{(1)}_n, \ldots, \theta^{(m)}_n)
+\]</span></p>
+<p>is treated as an estimator of the true posterior mean,</p>
+<p><span class="math display">\[
+\mathbb{E}[\theta_n \mid y]
+\ = \
+\int_{-\infty}^{\infty}
+    \, \theta \, p(\theta | y)
+\, \mathrm{d}\theta_n,
+\]</span></p>
+<p>based the observed data <span class="math inline">\(y\)</span>.</p>
+<p>The standard error for the estimator <span class="math inline">\(\hat{\theta}_n\)</span> is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as <span class="math inline">\(\hat{\sigma}_n / \sqrt{N_{\mathrm{eff}}}\)</span>. The smaller the standard error, the closer the estimate <span class="math inline">\(\hat{\theta}_n\)</span> is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see <span class="citation" data-cites="Geyer:2011">Charles J. Geyer (<a href="#ref-Geyer:2011" role="doc-biblioref">2011</a>)</span> for more details of the MCMC central limit theorem.</p>
+</section>
+<section id="thinning" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="thinning">Thinning</h3>
+<p>In complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag <span class="math inline">\(t\)</span>, <span class="math inline">\(\rho_t\)</span>, decreases as the lag, <span class="math inline">\(t\)</span>, increases. In this situation, thinning the sample by keeping only every <span class="math inline">\(N\)</span>-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.</p>
+<p>For instance, consider generating one thousand posterior draws in one of the following two ways.</p>
+<ul>
+<li><p>Generate 1000 draws after convergence and save all of them.</p></li>
+<li><p>Generate 10,000 draws after convergence and save every tenth draw.</p></li>
+</ul>
+<p>Even though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation <span class="math inline">\(\rho_t\)</span> for the thinned sequence is equivalent to <span class="math inline">\(\rho_{10t}\)</span> in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.</p>
+<p>Now contrast the second approach above with the unthinned alternative,</p>
+<ul>
+<li>Generate 10,000 draws after convergence and save every draw.</li>
+</ul>
+<p>This will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, <em>the only reason to thin a sample is to reduce memory requirements</em>.</p>
+<p>If draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-GelmanRubin:1992" class="csl-entry" role="listitem">
+Gelman, Andrew, and Donald B. Rubin. 1992. <span>“Inference from Iterative Simulation Using Multiple Sequences.”</span> <em>Statistical Science</em> 7 (4): 457–72.
+</div>
+<div id="ref-Geyer:1992" class="csl-entry" role="listitem">
+Geyer, Charles J. 1992. <span>“Practical Markov Chain Monte Carlo.”</span> <em>Statistical Science</em>, 473–83.
+</div>
+<div id="ref-Geyer:2011" class="csl-entry" role="listitem">
+Geyer, Charles J. 2011. <span>“Introduction to <span>M</span>arkov Chain <span>M</span>onte <span>C</span>arlo.”</span> In <em>Handbook of <span>M</span>arkov Chain <span>M</span>onte <span>C</span>arlo</em>, edited by Steve Brooks, Andrew Gelman, Galin L. Jones, and Xiao-Li Meng, 3–48. Chapman; Hall/CRC.
+</div>
+<div id="ref-Metropolis:1953" class="csl-entry" role="listitem">
+Metropolis, N., A. Rosenbluth, M. Rosenbluth, M. Teller, and E. Teller. 1953. <span>“Equations of State Calculations by Fast Computing Machines.”</span> <em>Journal of Chemical Physics</em> 21: 1087–92.
+</div>
+<div id="ref-Vehtari+etal:2021:Rhat" class="csl-entry" role="listitem">
+Vehtari, Aki, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian Bürkner. 2021. <span>“Rank-Normalization, Folding, and Localization: An Improved <span class="math inline">\(\widehat{R}\)</span> for Assessing Convergence of <span>MCMC</span>.”</span> <em>Bayesian Analysis</em> 16: 667–718.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Using vectors simplifies high level exposition at the expense of collapsing structure.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>The structure is assumed to be rectangular; in the future, this needs to be generalized to ragged samples.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>The <code>lp__</code> value also represents the potential energy in the Hamiltonian system and is rate bounded by the randomly supplied kinetic energy each iteration, which follows a Chi-square distribution in the number of parameters.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/mcmc.html" class="pagination-link" aria-label="MCMC Sampling">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">MCMC Sampling</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/optimization.html" class="pagination-link" aria-label="Optimization">
+        <span class="nav-page-text">Optimization</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/blocks.html b/docs/2_39/reference-manual/blocks.html
new file mode 100644
index 000000000..15e8de1b5
--- /dev/null
+++ b/docs/2_39/reference-manual/blocks.html
@@ -0,0 +1,1556 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Program Blocks</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/user-functions.html" rel="next">
+<link href="../reference-manual/statements.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Program Blocks">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/blocks.html">Program Blocks</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#blocks.chapter" id="toc-blocks.chapter" class="nav-link active" data-scroll-target="#blocks.chapter">Program Blocks</a>
+  <ul class="collapse">
+  <li><a href="#overview-of-stans-program-blocks" id="toc-overview-of-stans-program-blocks" class="nav-link" data-scroll-target="#overview-of-stans-program-blocks">Overview of Stan’s program blocks</a>
+  <ul class="collapse">
+  <li><a href="#optionality-and-ordering" id="toc-optionality-and-ordering" class="nav-link" data-scroll-target="#optionality-and-ordering">Optionality and ordering</a></li>
+  <li><a href="#variable-scope" id="toc-variable-scope" class="nav-link" data-scroll-target="#variable-scope">Variable scope</a></li>
+  <li><a href="#function-scope" id="toc-function-scope" class="nav-link" data-scroll-target="#function-scope">Function scope</a></li>
+  <li><a href="#automatic-variable-definitions" id="toc-automatic-variable-definitions" class="nav-link" data-scroll-target="#automatic-variable-definitions">Automatic variable definitions</a></li>
+  <li><a href="#transformed-variables" id="toc-transformed-variables" class="nav-link" data-scroll-target="#transformed-variables">Transformed variables</a></li>
+  <li><a href="#generated-quantities" id="toc-generated-quantities" class="nav-link" data-scroll-target="#generated-quantities">Generated quantities</a></li>
+  <li><a href="#variable-read-write-and-definition-summary" id="toc-variable-read-write-and-definition-summary" class="nav-link" data-scroll-target="#variable-read-write-and-definition-summary">Variable read, write, and definition summary</a></li>
+  </ul></li>
+  <li><a href="#statistical-variable-taxonomy" id="toc-statistical-variable-taxonomy" class="nav-link" data-scroll-target="#statistical-variable-taxonomy">Statistical variable taxonomy</a></li>
+  <li><a href="#program-block-data" id="toc-program-block-data" class="nav-link" data-scroll-target="#program-block-data">Program block: <code>data</code></a>
+  <ul class="collapse">
+  <li><a href="#variable-reads-and-transformations" id="toc-variable-reads-and-transformations" class="nav-link" data-scroll-target="#variable-reads-and-transformations">Variable reads and transformations</a></li>
+  <li><a href="#statements" id="toc-statements" class="nav-link" data-scroll-target="#statements">Statements</a></li>
+  <li><a href="#variable-constraint-checking" id="toc-variable-constraint-checking" class="nav-link" data-scroll-target="#variable-constraint-checking">Variable constraint checking</a></li>
+  </ul></li>
+  <li><a href="#program-block-transformed-data" id="toc-program-block-transformed-data" class="nav-link" data-scroll-target="#program-block-transformed-data">Program block: <code>transformed data</code></a>
+  <ul class="collapse">
+  <li><a href="#variable-reads-and-transformations-1" id="toc-variable-reads-and-transformations-1" class="nav-link" data-scroll-target="#variable-reads-and-transformations-1">Variable reads and transformations</a></li>
+  <li><a href="#statements-1" id="toc-statements-1" class="nav-link" data-scroll-target="#statements-1">Statements</a></li>
+  <li><a href="#variable-constraint-checking-1" id="toc-variable-constraint-checking-1" class="nav-link" data-scroll-target="#variable-constraint-checking-1">Variable constraint checking</a></li>
+  </ul></li>
+  <li><a href="#program-block-parameters" id="toc-program-block-parameters" class="nav-link" data-scroll-target="#program-block-parameters">Program block: <code>parameters</code></a>
+  <ul class="collapse">
+  <li><a href="#constraining-inverse-transform" id="toc-constraining-inverse-transform" class="nav-link" data-scroll-target="#constraining-inverse-transform">Constraining inverse transform</a></li>
+  <li><a href="#gradient-calculation" id="toc-gradient-calculation" class="nav-link" data-scroll-target="#gradient-calculation">Gradient calculation</a></li>
+  <li><a href="#writing-draws" id="toc-writing-draws" class="nav-link" data-scroll-target="#writing-draws">Writing draws</a></li>
+  </ul></li>
+  <li><a href="#program-block-transformed-parameters" id="toc-program-block-transformed-parameters" class="nav-link" data-scroll-target="#program-block-transformed-parameters">Program block: <code>transformed parameters</code></a>
+  <ul class="collapse">
+  <li><a href="#constraints-are-for-error-checking" id="toc-constraints-are-for-error-checking" class="nav-link" data-scroll-target="#constraints-are-for-error-checking">Constraints are for error checking</a></li>
+  </ul></li>
+  <li><a href="#program-block-model" id="toc-program-block-model" class="nav-link" data-scroll-target="#program-block-model">Program block: <code>model</code></a></li>
+  <li><a href="#program-block-generated-quantities" id="toc-program-block-generated-quantities" class="nav-link" data-scroll-target="#program-block-generated-quantities">Program block: <code>generated quantities</code></a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/blocks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/blocks.html">Program Blocks</a></li></ol></nav></header>
+
+
+
+
+<section id="blocks.chapter" class="level1">
+<h1>Program Blocks</h1>
+<p>A Stan program is organized into a sequence of named blocks, the bodies of which consist of variable declarations, followed in the case of some blocks with statements.</p>
+<section id="overview-of-stans-program-blocks" class="level2">
+<h2 class="anchored" data-anchor-id="overview-of-stans-program-blocks">Overview of Stan’s program blocks</h2>
+<p>The full set of named program blocks is exemplified in the following skeletal Stan program.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... function declarations and definitions ...</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... declarations ...</span></span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... declarations ... statements ...</span></span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... declarations ...</span></span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... declarations ... statements ...</span></span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... declarations ... statements ...</span></span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... declarations ... statements ...</span></span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.</p>
+<section id="optionality-and-ordering" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="optionality-and-ordering">Optionality and ordering</h3>
+<p>All of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.</p>
+</section>
+<section id="variable-scope" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-scope">Variable scope</h3>
+<p>The variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.</p>
+<p>Variables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).</p>
+</section>
+<section id="function-scope" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="function-scope">Function scope</h3>
+<p>Functions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).</p>
+<p>Random-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in <code>_rng</code>; such functions are suffixed with <code>_rng</code>. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with <code>_lp</code>.</p>
+<p>Density functions defined in the program may be used in distribution statements.</p>
+</section>
+<section id="automatic-variable-definitions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="automatic-variable-definitions">Automatic variable definitions</h3>
+<p>The variables declared in the <code>data</code> and <code>parameters</code> block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.</p>
+<p>The variables in the <code>data</code> block are read from an external input source such as a file or a designated R data structure. The variables in the <code>parameters</code> block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.</p>
+</section>
+<section id="transformed-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="transformed-variables">Transformed variables</h3>
+<p>The <code>transformed data</code> and <code>transformed parameters</code> block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.</p>
+<p>Before generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>Transformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.</p>
+</section>
+<section id="generated-quantities" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="generated-quantities">Generated quantities</h3>
+<p>The generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.</p>
+</section>
+<section id="variable-read-write-and-definition-summary" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-read-write-and-definition-summary">Variable read, write, and definition summary</h3>
+<p>A table summarizing the point at which variables are read, written, and defined is given in the <a href="#block-actions-table">block actions table</a>.</p>
+<p><strong>Block Actions Table.</strong> <a id="block-actions-table"></a> <em>The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;">block</th>
+<th style="text-align: center;">statement</th>
+<th style="text-align: left;">action / period</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><code>data</code></td>
+<td style="text-align: center;">no</td>
+<td style="text-align: left;">read / chain</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>transformed data</code></td>
+<td style="text-align: center;">yes</td>
+<td style="text-align: left;">evaluate / chain</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>parameters</code></td>
+<td style="text-align: center;">no</td>
+<td style="text-align: left;">inv. transform, Jacobian / leapfrog</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">&nbsp;</td>
+<td style="text-align: center;">&nbsp;</td>
+<td style="text-align: left;">inv. transform, write / sample</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>transformed parameters</code></td>
+<td style="text-align: center;">yes</td>
+<td style="text-align: left;">evaluate / leapfrog</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">&nbsp;</td>
+<td style="text-align: center;">&nbsp;</td>
+<td style="text-align: left;">write / sample</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>model</code></td>
+<td style="text-align: center;">yes</td>
+<td style="text-align: left;">evaluate / leapfrog step</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>generated quantities</code></td>
+<td style="text-align: center;">yes</td>
+<td style="text-align: left;">eval / sample</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">&nbsp;</td>
+<td style="text-align: center;">&nbsp;</td>
+<td style="text-align: left;">write / sample</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><em><code>(initialization)</code></em></td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: left;">read, transform / chain</td>
+</tr>
+</tbody>
+</table>
+<p><strong>Variable Declaration Table.</strong> <a id="where-to-declare-table"></a> <em>This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (<span class="math inline">\(*\)</span>) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: center;">param depend</th>
+<th style="text-align: center;">in target</th>
+<th style="text-align: center;">save</th>
+<th style="text-align: left;">declare in</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: left;"><code>transformed parameters</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: left;"><code>model</code> (local)</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: left;"><code>generated quantities</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: left;"><code>generated quantities</code> (local)</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: left;"><code>transformed data</code> &nbsp; <em>and</em> <code>generated quantities</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: left;"><code>transformed data</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">+</td>
+<td style="text-align: left;"><code>generated quantities</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: center;">-</td>
+<td style="text-align: left;"><code>transformed data</code> (local)</td>
+</tr>
+</tbody>
+</table>
+<p>Another way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the <a href="#where-to-declare-table">variable declaration table</a>. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>The rest of this chapter provides full details on when and how the variables and statements in each block are executed.</p>
+</section>
+</section>
+<section id="statistical-variable-taxonomy" class="level2">
+<h2 class="anchored" data-anchor-id="statistical-variable-taxonomy">Statistical variable taxonomy</h2>
+<p><strong>Statistical Variable Taxonomy Table.</strong> <a id="variable-kinds-table"></a> <em>Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.</em></p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 44%">
+<col style="width: 55%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: right;">variable kind</th>
+<th style="text-align: left;">declaration block</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: right;">constants</td>
+<td style="text-align: left;"><code>data</code>, <code>transformed data</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;">unmodeled data</td>
+<td style="text-align: left;"><code>data</code>, <code>transformed data</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;">modeled data</td>
+<td style="text-align: left;"><code>data</code>, <code>transformed data</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;">missing data</td>
+<td style="text-align: left;"><code>parameters</code>, <code>transformed parameters</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;">modeled parameters</td>
+<td style="text-align: left;"><code>parameters</code>, <code>transformed parameters</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;">unmodeled parameters</td>
+<td style="text-align: left;"><code>data</code>, <code>transformed data</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;">derived quantities</td>
+<td style="text-align: left;"><code>transformed data</code>, <code>transformed parameters</code>, <code>generated quantities</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;">loop indices</td>
+<td style="text-align: left;">loop statement</td>
+</tr>
+</tbody>
+</table>
+<p>Page 366 of <span class="citation" data-cites="GelmanHill:2007">(<a href="#ref-GelmanHill:2007" role="doc-biblioref">Gelman and Hill 2007</a>)</span> provides a taxonomy of the kinds of variables used in Bayesian models. The <a href="#variable-kinds-table">table of kinds of variables</a> contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.</p>
+<p>Constants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the <code>data</code> block.</p>
+<p>The following program illustrates various variables kinds, listing the kind of each variable next to its declaration.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;           <span class="co">// unmodeled data</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;          <span class="co">// modeled data</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_mu;               <span class="co">// config. unmodeled param</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_mu;   <span class="co">// config. unmodeled param</span></span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;      <span class="co">// const. unmodeled param</span></span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; beta;       <span class="co">// const. unmodeled param</span></span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  alpha = <span class="fl">0.1</span>;</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  beta = <span class="fl">0.1</span>;</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_y;                <span class="co">// modeled param</span></span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; tau_y;      <span class="co">// modeled param</span></span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_y;    <span class="co">// derived quantity (param)</span></span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>  sigma_y = pow(tau_y, -<span class="fl">0.5</span>);</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>  tau_y ~ gamma(alpha, beta);</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>  mu_y ~ normal(mu_mu, sigma_mu);</span>
+<span id="cb2-24"><a href="#cb2-24" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-25"><a href="#cb2-25" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(mu_y, sigma_y);</span>
+<span id="cb2-26"><a href="#cb2-26" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-27"><a href="#cb2-27" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-28"><a href="#cb2-28" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb2-29"><a href="#cb2-29" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> variance_y;       <span class="co">// derived quantity (transform)</span></span>
+<span id="cb2-30"><a href="#cb2-30" aria-hidden="true" tabindex="-1"></a>  variance_y = sigma_y * sigma_y;</span>
+<span id="cb2-31"><a href="#cb2-31" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this example, <code>y</code> is an array of modeled data. Although it is specified in the <code>data</code> block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.</p>
+<p>The variable <code>N</code> is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.</p>
+<p>The other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the <code>data</code> block may be specified on a per-model execution basis as part of the data read. In the above model, <code>mu_mu</code> and <code>sigma_mu</code> are configurable unmodeled parameters.</p>
+<p>Unmodeled parameters that are hard coded in the model must be declared in the <code>transformed data</code> block. For example, the unmodeled parameters <code>alpha</code> and <code>beta</code> are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the <code>data</code> block, like the variables <code>mu_mu</code> and <code>sigma_mu</code>.</p>
+<p>This program declares two modeled parameters, <code>mu_y</code> and <code>tau_y</code>. These are the location and precision used in the normal model of the values in <code>y</code>. The heart of the model will be sampling the values of these parameters from their posterior distribution.</p>
+<p>The modeled parameter <code>tau_y</code> is transformed from a precision to a scale parameter and assigned to the variable <code>sigma_y</code> in the <code>transformed parameters</code> block. Thus the variable <code>sigma_y</code> is considered a derived quantity — its value is entirely determined by the values of other variables.</p>
+<p>The <code>generated quantities</code> block defines a value <code>variance_y</code>, which is defined as a transform of the scale or deviation parameter <code>sigma_y</code>. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).</p>
+<p>In later versions of Stan which have random number generators for the distributions, the <code>generated quantities</code> block will be usable to generate replicated data for model checking.</p>
+<p>Finally, the variable <code>n</code> is used as a loop index in the <code>model</code> block.</p>
+</section>
+<section id="program-block-data" class="level2">
+<h2 class="anchored" data-anchor-id="program-block-data">Program block: <code>data</code></h2>
+<p>The rest of this chapter will lay out the details of each block in order, starting with the <code>data</code> block in this section.</p>
+<section id="variable-reads-and-transformations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-reads-and-transformations">Variable reads and transformations</h3>
+<p>The <code>data</code> block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>Data variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.</p>
+</section>
+<section id="statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="statements">Statements</h3>
+<p>The <code>data</code> block does not allow statements.</p>
+</section>
+<section id="variable-constraint-checking" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-constraint-checking">Variable constraint checking</h3>
+<p>Each variable’s value is validated against its declaration as it is read. For example, if a variable <code>sigma</code> is declared as <code>real&lt;lower=0&gt;</code>, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.</p>
+</section>
+</section>
+<section id="program-block-transformed-data" class="level2">
+<h2 class="anchored" data-anchor-id="program-block-transformed-data">Program block: <code>transformed data</code></h2>
+<p>The <code>transformed data</code> block is for declaring and defining variables that do not need to be changed when running the program.</p>
+<section id="variable-reads-and-transformations-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-reads-and-transformations-1">Variable reads and transformations</h3>
+<p>For the <code>transformed data</code> block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.</p>
+<p>Variables declared in the <code>data</code> block may be used to declare transformed variables.</p>
+</section>
+<section id="statements-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="statements-1">Statements</h3>
+<p>The statements in a <code>transformed data</code> block are used to define (provide values for) variables declared in the <code>transformed data</code> block. Assignments are only allowed to variables declared in the <code>transformed data</code> block.</p>
+<p>These statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.</p>
+<p>Variables declared in the <code>data</code> block may be used in statements in the <code>transformed data</code> block.</p>
+<section id="restriction-on-operations-in-transformed-data" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="restriction-on-operations-in-transformed-data">Restriction on operations in <code>transformed data</code></h4>
+<p>The statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and <a href="../reference-manual/statements.html#distribution-statements.section">distribution statements</a> may not be used.</p>
+</section>
+</section>
+<section id="variable-constraint-checking-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-constraint-checking-1">Variable constraint checking</h3>
+<p>Any constraints on variables declared in the <code>transformed data</code> block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.</p>
+</section>
+</section>
+<section id="program-block-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="program-block-parameters">Program block: <code>parameters</code></h2>
+<p>The variables declared in the <code>parameters</code> program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block <em>are</em> the parameters being sampled by Stan.</p>
+<p>Variables declared as parameters cannot be directly assigned values. So there is no block of statements in the <code>parameters</code> program block. Variable quantities derived from parameters may be declared in the <code>transformed parameters</code> or <code>generated quantities</code> blocks, or may be defined as local variables in any statement blocks following their declaration.</p>
+<p>There is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.</p>
+<section id="constraining-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constraining-inverse-transform">Constraining inverse transform</h3>
+<p>Stan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of <span class="math inline">\(\mathbb{R}^n\)</span>. To do this, the parameters defined in the <code>parameters</code> block must be transformed so they are unconstrained.</p>
+<p>In practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s <code>parameters</code> program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.</p>
+<p>In some cases, the number of parameters is reduced in the unconstrained space. For instance, a <span class="math inline">\(K\)</span>-simplex only requires <span class="math inline">\(K-1\)</span> unconstrained parameters, and a <span class="math inline">\(K\)</span>-correlation matrix only requires <span class="math inline">\(\binom{K}{2}\)</span> unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the <code>parameters</code> program block.</p>
+<p>The probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the <code>parameters</code> program block. The details of the specific transformations are provided in the <a href="../reference-manual/transforms.html">variable transforms chapter</a>.</p>
+</section>
+<section id="gradient-calculation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="gradient-calculation">Gradient calculation</h3>
+<p>Hamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.</p>
+<p>Gradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.</p>
+<p>The amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.</p>
+</section>
+<section id="writing-draws" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="writing-draws">Writing draws</h3>
+<p>In the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the <code>parameters</code> block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.</p>
+</section>
+</section>
+<section id="program-block-transformed-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="program-block-transformed-parameters">Program block: <code>transformed parameters</code></h2>
+<p>The <code>transformed parameters</code> program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.</p>
+<p>Any variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.</p>
+<section id="constraints-are-for-error-checking" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constraints-are-for-error-checking">Constraints are for error checking</h3>
+<p>Like the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the <a href="../reference-manual/statements.html#reject-statements">section on reject statements</a> for further discussion of the behavior of reject statements.</p>
+</section>
+</section>
+<section id="program-block-model" class="level2">
+<h2 class="anchored" data-anchor-id="program-block-model">Program block: <code>model</code></h2>
+<p>The <code>model</code> program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.</p>
+<p>Local variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.</p>
+<p>The statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.</p>
+</section>
+<section id="program-block-generated-quantities" class="level2">
+<h2 class="anchored" data-anchor-id="program-block-generated-quantities">Program block: <code>generated quantities</code></h2>
+<p>The <code>generated quantities</code> program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.</p>
+<p>Among the applications of posterior inference that can be coded in the generated quantities block are</p>
+<ul>
+<li>forward sampling to generate simulated data for model testing,</li>
+<li>generating predictions for new data,</li>
+<li>calculating posterior event probabilities, including multiple comparisons, sign tests, etc.,</li>
+<li>calculating posterior expectations,</li>
+<li>transforming parameters for reporting,</li>
+<li>applying full Bayesian decision theory,</li>
+<li>calculating log likelihoods, deviances, etc. for model comparison.</li>
+</ul>
+<p>Parameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.</p>
+<p>Within the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.</p>
+<p>It is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.</p>
+<p>After the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.</p>
+<p>All variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> a; <span class="co">// added to the output</span></span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> b; <span class="co">// not added to the output</span></span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanHill:2007" class="csl-entry" role="listitem">
+Gelman, Andrew, and Jennifer Hill. 2007. <em>Data Analysis Using Regression and Multilevel-Hierarchical Models</em>. Cambridge, United Kingdom: Cambridge University Press.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>If the C++ code is configured for concurrent threads, the data and transformed data blocks can be executed once and reused for multiple chains.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>It is possible to print a variable every iteration that does not depend on parameters—just define it (or redefine it if it is transformed data) in the <code>generated quantities</code> block.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>With multiple threads, or even running chains sequentially in a single thread, data could be read only once per set of chains. Stan was designed to be thread safe and future versions will provide a multithreading option for Markov chains.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/statements.html" class="pagination-link" aria-label="Statements">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Statements</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/user-functions.html" class="pagination-link" aria-label="User-Defined Functions">
+        <span class="nav-page-text">User-Defined Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/blocks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/comments.html b/docs/2_39/reference-manual/comments.html
new file mode 100644
index 000000000..ac1c5115d
--- /dev/null
+++ b/docs/2_39/reference-manual/comments.html
@@ -0,0 +1,1091 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Comments</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/whitespace.html" rel="next">
+<link href="../reference-manual/includes.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Comments">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/comments.html">Comments</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#comments" id="toc-comments" class="nav-link active" data-scroll-target="#comments">Comments</a>
+  <ul class="collapse">
+  <li><a href="#line-based-comments" id="toc-line-based-comments" class="nav-link" data-scroll-target="#line-based-comments">Line-based comments</a></li>
+  <li><a href="#bracketed-comments" id="toc-bracketed-comments" class="nav-link" data-scroll-target="#bracketed-comments">Bracketed comments</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/comments.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/comments.html">Comments</a></li></ol></nav></header>
+
+
+
+
+<section id="comments" class="level1">
+<h1>Comments</h1>
+<p>Stan supports C++-style line-based and bracketed comments. Comments may be used anywhere whitespace is allowed in a Stan program.</p>
+<section id="line-based-comments" class="level2">
+<h2 class="anchored" data-anchor-id="line-based-comments">Line-based comments</h2>
+<p>Any characters on a line following two forward slashes (<code>//</code>) is ignored along with the slashes. These may be used, for example, to document variables,</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;  <span class="co">// number of observations</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;  <span class="co">// observations</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="bracketed-comments" class="level2">
+<h2 class="anchored" data-anchor-id="bracketed-comments">Bracketed comments</h2>
+<p>For bracketed comments, any text between a forward-slash and asterisk pair (<code>/*</code>) and an asterisk and forward-slash pair (<code>*/</code>) is ignored.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/includes.html" class="pagination-link" aria-label="Includes">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Includes</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/whitespace.html" class="pagination-link" aria-label="Whitespace">
+        <span class="nav-page-text">Whitespace</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/comments.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/deprecations.html b/docs/2_39/reference-manual/deprecations.html
new file mode 100644
index 000000000..eca6bfe04
--- /dev/null
+++ b/docs/2_39/reference-manual/deprecations.html
@@ -0,0 +1,1110 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Deprecated Features</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/removals.html" rel="next">
+<link href="../reference-manual/execution.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Deprecated Features">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/deprecations.html">Deprecated Features</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#deprecated-features" id="toc-deprecated-features" class="nav-link active" data-scroll-target="#deprecated-features">Deprecated Features</a>
+  <ul class="collapse">
+  <li><a href="#lkj_cov-distribution" id="toc-lkj_cov-distribution" class="nav-link" data-scroll-target="#lkj_cov-distribution"><code>lkj_cov</code> distribution</a></li>
+  <li><a href="#use-of-_lp-functions-in-transformed-parameters" id="toc-use-of-_lp-functions-in-transformed-parameters" class="nav-link" data-scroll-target="#use-of-_lp-functions-in-transformed-parameters">Use of <code>_lp</code> functions in <code>transformed parameters</code></a></li>
+  <li><a href="#deprecated-functions" id="toc-deprecated-functions" class="nav-link" data-scroll-target="#deprecated-functions">Deprecated Functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/deprecations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/deprecations.html">Deprecated Features</a></li></ol></nav></header>
+
+
+
+
+<section id="deprecated-features" class="level1">
+<h1>Deprecated Features</h1>
+<p>This appendix lists currently deprecated functionality along with how to replace it.</p>
+<p>Starting with Stan 2.29, minor (syntax-level) deprecations can be removed 3 versions after release; e.g., syntax deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in <a href="../reference-manual/removals.html">Removed Features</a>. The Stan compiler can <a href="https://mc-stan.org/docs/stan-users-guide/using-stanc.html#stanc-pretty-printing">automatically update</a> many of these on the behalf of the user for at least one version after they are removed.</p>
+<p>Any feature which changes semantic meaning (such as the upgraded ODE solver interface) will not be removed until a major version change (e.g., Stan 3.0).</p>
+<section id="lkj_cov-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="lkj_cov-distribution"><code>lkj_cov</code> distribution</h2>
+<p><em>Deprecated</em>:The distribution <code>lkj_cov</code> is deprecated.</p>
+<p><em>Replacement</em>: Replace <code>lkj_cov_lpdf(...)</code> with an <code>lkj_corr</code> distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>Sigma ~ lkj_cov(mu, tau, eta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">corr_matrix</span>[K] Omega;</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] sigma;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>Omega ~ lkj_corr(eta);</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>sigma ~ lognormal(mu, tau);</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a><span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>Sigma &lt;- quad_form_diag(Omega, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The variable <code>Sigma</code> may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.</p>
+<p><em>Scheduled Removal</em>: Stan 3.0 or later.</p>
+</section>
+<section id="use-of-_lp-functions-in-transformed-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="use-of-_lp-functions-in-transformed-parameters">Use of <code>_lp</code> functions in <code>transformed parameters</code></h2>
+<p><em>Deprecated</em>: Using <a href="../reference-manual/user-functions.html#log-probability-access-in-functions">functions that end in <code>_lp</code></a> in the <code>transformed parameters</code> block.</p>
+<p><em>Replacement</em>: Use <code>_jacobian</code> functions and the <code>jacobian +=</code> statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.</p>
+</section>
+<section id="deprecated-functions" class="level2">
+<h2 class="anchored" data-anchor-id="deprecated-functions">Deprecated Functions</h2>
+<p>Several built-in Stan functions have been deprecated. Consult the <a href="https://mc-stan.org/docs/functions-reference/deprecated-functions.html">functions reference</a> for more information.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/execution.html" class="pagination-link" aria-label="Program Execution">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Program Execution</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/removals.html" class="pagination-link" aria-label="Removed Features">
+        <span class="nav-page-text">Removed Features</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/deprecations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/diagnostics.html b/docs/2_39/reference-manual/diagnostics.html
new file mode 100644
index 000000000..d2888edbd
--- /dev/null
+++ b/docs/2_39/reference-manual/diagnostics.html
@@ -0,0 +1,1132 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Diagnostic Mode</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/reproducibility.html" rel="next">
+<link href="../reference-manual/laplace_embedded.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Diagnostic Mode">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/diagnostics.html">Diagnostic Mode</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#diagnostic-algorithms.chapter" id="toc-diagnostic-algorithms.chapter" class="nav-link active" data-scroll-target="#diagnostic-algorithms.chapter">Diagnostic Mode</a>
+  <ul class="collapse">
+  <li><a href="#diagnostic-mode-output" id="toc-diagnostic-mode-output" class="nav-link" data-scroll-target="#diagnostic-mode-output">Diagnostic mode output</a>
+  <ul class="collapse">
+  <li><a href="#unconstrained-scale" id="toc-unconstrained-scale" class="nav-link" data-scroll-target="#unconstrained-scale">Unconstrained scale</a></li>
+  <li><a href="#includes-jacobian" id="toc-includes-jacobian" class="nav-link" data-scroll-target="#includes-jacobian">Includes Jacobian</a></li>
+  </ul></li>
+  <li><a href="#configuration-options" id="toc-configuration-options" class="nav-link" data-scroll-target="#configuration-options">Configuration options</a></li>
+  <li><a href="#speed-warning-and-data-trimming" id="toc-speed-warning-and-data-trimming" class="nav-link" data-scroll-target="#speed-warning-and-data-trimming">Speed warning and data trimming</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/diagnostics.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/diagnostics.html">Diagnostic Mode</a></li></ol></nav></header>
+
+
+
+
+<section id="diagnostic-algorithms.chapter" class="level1">
+<h1>Diagnostic Mode</h1>
+<p>Stan’s diagnostic mode runs a Stan program with data, initializing parameters either randomly or with user-specified initial values, and then evaluates the log probability and its gradients. The gradients computed by the Stan program are compared to values calculated by finite differences.</p>
+<p>Diagnostic mode may be configured with two parameters.</p>
+<p><strong>Diagnostic Mode Configuration Table.</strong> <em>The diagnostic model configuration parameters, constraints, and default values.</em></p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 21%">
+<col style="width: 42%">
+<col style="width: 21%">
+<col style="width: 14%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: center;">parameter</th>
+<th style="text-align: left;">description</th>
+<th style="text-align: center;">constraints</th>
+<th style="text-align: center;">default</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;"><code>epsilon</code></td>
+<td style="text-align: left;">finite difference size</td>
+<td style="text-align: center;">(0, infty)</td>
+<td style="text-align: center;">1e–6</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>error</code></td>
+<td style="text-align: left;">error threshold for matching</td>
+<td style="text-align: center;">(0, infty)</td>
+<td style="text-align: center;">1e–6</td>
+</tr>
+</tbody>
+</table>
+<p>If the difference between the Stan program’s gradient value and that calculated by finite difference is higher than the specified threshold, the argument will be flagged.</p>
+<section id="diagnostic-mode-output" class="level2">
+<h2 class="anchored" data-anchor-id="diagnostic-mode-output">Diagnostic mode output</h2>
+<p>Diagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.</p>
+<section id="unconstrained-scale" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unconstrained-scale">Unconstrained scale</h3>
+<p>The output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an <span class="math inline">\(N \times N\)</span> correlation matrix, requires <span class="math inline">\(\binom{N}{2}\)</span> unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.</p>
+</section>
+<section id="includes-jacobian" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="includes-jacobian">Includes Jacobian</h3>
+<p>The log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.</p>
+</section>
+</section>
+<section id="configuration-options" class="level2">
+<h2 class="anchored" data-anchor-id="configuration-options">Configuration options</h2>
+<p>The general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.</p>
+</section>
+<section id="speed-warning-and-data-trimming" class="level2">
+<h2 class="anchored" data-anchor-id="speed-warning-and-data-trimming">Speed warning and data trimming</h2>
+<p>Due to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/laplace_embedded.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Embedded Laplace Approximation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/reproducibility.html" class="pagination-link" aria-label="Reproducibility">
+        <span class="nav-page-text">Reproducibility</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/diagnostics.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/encoding.html b/docs/2_39/reference-manual/encoding.html
new file mode 100644
index 000000000..9d12a9011
--- /dev/null
+++ b/docs/2_39/reference-manual/encoding.html
@@ -0,0 +1,1065 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Character Encoding</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/includes.html" rel="next">
+<link href="../reference-manual/index.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Character Encoding">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Character Encoding</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#character-encoding" id="toc-character-encoding" class="nav-link active" data-scroll-target="#character-encoding">Character Encoding</a>
+  <ul class="collapse">
+  <li><a href="#content-characters" id="toc-content-characters" class="nav-link" data-scroll-target="#content-characters">Content characters</a>
+  <ul class="collapse">
+  <li><a href="#compatibility-with-latin-1-and-utf-8" id="toc-compatibility-with-latin-1-and-utf-8" class="nav-link" data-scroll-target="#compatibility-with-latin-1-and-utf-8">Compatibility with Latin-1 and UTF-8</a></li>
+  </ul></li>
+  <li><a href="#comment-characters" id="toc-comment-characters" class="nav-link" data-scroll-target="#comment-characters">Comment characters</a></li>
+  <li><a href="#string-literals" id="toc-string-literals" class="nav-link" data-scroll-target="#string-literals">String literals</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/encoding.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Character Encoding</a></li></ol></nav></header>
+
+
+
+
+<section id="character-encoding" class="level1">
+<h1>Character Encoding</h1>
+<section id="content-characters" class="level2">
+<h2 class="anchored" data-anchor-id="content-characters">Content characters</h2>
+<p>The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.</p>
+<section id="compatibility-with-latin-1-and-utf-8" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="compatibility-with-latin-1-and-utf-8">Compatibility with Latin-1 and UTF-8</h3>
+<p>The UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.</p>
+</section>
+</section>
+<section id="comment-characters" class="level2">
+<h2 class="anchored" data-anchor-id="comment-characters">Comment characters</h2>
+<p>Any bytes on a line after a line-comment sequence (<code>//</code> or <code>#</code>) are ignored up until the ASCII newline character (<code>\n</code>). They may thus be written in any character encoding which is convenient.</p>
+<p>Any content after a block comment open sequence in ASCII (<code>/*</code>) up to the closing block comment (<code>*/</code>) is ignored, and thus may also be written in whatever character set is convenient.</p>
+</section>
+<section id="string-literals" class="level2">
+<h2 class="anchored" data-anchor-id="string-literals">String literals</h2>
+<p>The raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A <code>print</code>, <code>reject</code>, or <code>fatal_error</code> statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (<code>"</code>) when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.</p>
+<p>ASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/index.html" class="pagination-link" aria-label="Stan Reference Manual">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Stan Reference Manual</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/includes.html" class="pagination-link" aria-label="Includes">
+        <span class="nav-page-text">Includes</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/encoding.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/execution.html b/docs/2_39/reference-manual/execution.html
new file mode 100644
index 000000000..04dd08047
--- /dev/null
+++ b/docs/2_39/reference-manual/execution.html
@@ -0,0 +1,1263 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Program Execution</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/deprecations.html" rel="next">
+<link href="../reference-manual/syntax.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Program Execution">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/execution.html">Program Execution</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#program-execution" id="toc-program-execution" class="nav-link active" data-scroll-target="#program-execution">Program Execution</a>
+  <ul class="collapse">
+  <li><a href="#reading-and-transforming-data" id="toc-reading-and-transforming-data" class="nav-link" data-scroll-target="#reading-and-transforming-data">Reading and transforming data</a>
+  <ul class="collapse">
+  <li><a href="#read-data" id="toc-read-data" class="nav-link" data-scroll-target="#read-data">Read data</a></li>
+  <li><a href="#define-transformed-data" id="toc-define-transformed-data" class="nav-link" data-scroll-target="#define-transformed-data">Define transformed data</a></li>
+  </ul></li>
+  <li><a href="#initialization" id="toc-initialization" class="nav-link" data-scroll-target="#initialization">Initialization</a>
+  <ul class="collapse">
+  <li><a href="#user-supplied-initial-values" id="toc-user-supplied-initial-values" class="nav-link" data-scroll-target="#user-supplied-initial-values">User-supplied initial values</a></li>
+  <li><a href="#random-initial-values" id="toc-random-initial-values" class="nav-link" data-scroll-target="#random-initial-values">Random initial values</a></li>
+  <li><a href="#zero-initial-values" id="toc-zero-initial-values" class="nav-link" data-scroll-target="#zero-initial-values">Zero initial values</a></li>
+  </ul></li>
+  <li><a href="#sampling" id="toc-sampling" class="nav-link" data-scroll-target="#sampling">Sampling</a>
+  <ul class="collapse">
+  <li><a href="#leapfrog-steps" id="toc-leapfrog-steps" class="nav-link" data-scroll-target="#leapfrog-steps">Leapfrog steps</a></li>
+  <li><a href="#log-probability-and-gradient-calculation" id="toc-log-probability-and-gradient-calculation" class="nav-link" data-scroll-target="#log-probability-and-gradient-calculation">Log probability and gradient calculation</a></li>
+  <li><a href="#metropolis-acceptreject" id="toc-metropolis-acceptreject" class="nav-link" data-scroll-target="#metropolis-acceptreject">Metropolis accept/reject</a></li>
+  </ul></li>
+  <li><a href="#optimization" id="toc-optimization" class="nav-link" data-scroll-target="#optimization">Optimization</a></li>
+  <li><a href="#variational-inference" id="toc-variational-inference" class="nav-link" data-scroll-target="#variational-inference">Variational inference</a></li>
+  <li><a href="#model-diagnostics" id="toc-model-diagnostics" class="nav-link" data-scroll-target="#model-diagnostics">Model diagnostics</a></li>
+  <li><a href="#output" id="toc-output" class="nav-link" data-scroll-target="#output">Output</a>
+  <ul class="collapse">
+  <li><a href="#generated-quantities" id="toc-generated-quantities" class="nav-link" data-scroll-target="#generated-quantities">Generated quantities</a></li>
+  <li><a href="#write" id="toc-write" class="nav-link" data-scroll-target="#write">Write</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/execution.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/execution.html">Program Execution</a></li></ol></nav></header>
+
+
+
+
+<section id="program-execution" class="level1">
+<h1>Program Execution</h1>
+<p>This chapter provides a sketch of how a compiled Stan model is executed using sampling. Optimization shares the same data reading and initialization steps, but then does optimization rather than sampling.</p>
+<p>This sketch is elaborated in the following chapters of this part, which cover variable declarations, expressions, statements, and blocks in more detail.</p>
+<section id="reading-and-transforming-data" class="level2">
+<h2 class="anchored" data-anchor-id="reading-and-transforming-data">Reading and transforming data</h2>
+<p>The reading and transforming data steps are the same for sampling, optimization and diagnostics.</p>
+<section id="read-data" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="read-data">Read data</h3>
+<p>The first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>All of the variables declared in the <code>data</code> block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.</p>
+<p>After each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable <code>N</code> is declared as <code>int&lt;lower=0&gt;</code>, after <code>N</code> is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.</p>
+</section>
+<section id="define-transformed-data" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="define-transformed-data">Define transformed data</h3>
+<p>After data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.</p>
+<p>Transformed data variables are initialized with real values set to <code>NaN</code> and integer values set to the smallest integer (large absolute value negative number).</p>
+<p>After the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.</p>
+</section>
+</section>
+<section id="initialization" class="level2">
+<h2 class="anchored" data-anchor-id="initialization">Initialization</h2>
+<p>Initialization is the same for all of Stan’s algorithms.</p>
+<section id="user-supplied-initial-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="user-supplied-initial-values">User-supplied initial values</h3>
+<p>If there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.</p>
+<p>After being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.</p>
+<section id="boundary-values-are-problematic" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="boundary-values-are-problematic">Boundary values are problematic</h4>
+<p>Because of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>an initial value of 0 for <code>theta</code> leads to an unconstrained value of <span class="math inline">\(-\infty\)</span>, whereas a value of 1 leads to an unconstrained value of <span class="math inline">\(+\infty\)</span>. While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.</p>
+</section>
+</section>
+<section id="random-initial-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="random-initial-values">Random initial values</h3>
+<p>If there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval <span class="math inline">\((-2,2)\)</span>. The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.</p>
+<p>An unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.</p>
+<p>For parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to <span class="math inline">\(\exp(0) = 1\)</span> on the constrained scale. A value of -2 corresponds to <span class="math inline">\(\exp(-2) = .13\)</span> and a value of 2 corresponds to <span class="math inline">\(\exp(2) = 7.4\)</span>.</p>
+<p>For parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.</p>
+<p>Simplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is <span class="math inline">\(1/K\)</span> in a <span class="math inline">\(K\)</span>-simplex).</p>
+<p>Cholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.</p>
+<p>The initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the <a href="../reference-manual/transforms.html">chapter on variable transforms</a>.</p>
+</section>
+<section id="zero-initial-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="zero-initial-values">Zero initial values</h3>
+<p>The initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.</p>
+</section>
+</section>
+<section id="sampling" class="level2">
+<h2 class="anchored" data-anchor-id="sampling">Sampling</h2>
+<p>Sampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.</p>
+<section id="leapfrog-steps" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="leapfrog-steps">Leapfrog steps</h3>
+<p>For each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.</p>
+<p>For simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.</p>
+<p>If the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively <span class="citation" data-cites="Hoffman-Gelman:2014">(<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">Hoffman and Gelman 2014</a>)</span>.</p>
+</section>
+<section id="log-probability-and-gradient-calculation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="log-probability-and-gradient-calculation">Log probability and gradient calculation</h3>
+<p>During each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.</p>
+<p>The first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.</p>
+<p>Because the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.</p>
+<p>Next, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.</p>
+<p>The final step in the log probability function calculation is to execute the statements defined in the model block.</p>
+<p>As the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.</p>
+</section>
+<section id="metropolis-acceptreject" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="metropolis-acceptreject">Metropolis accept/reject</h3>
+<p>A standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.</p>
+<p>If step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).</p>
+<p>If the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.</p>
+</section>
+</section>
+<section id="optimization" class="level2">
+<h2 class="anchored" data-anchor-id="optimization">Optimization</h2>
+<p>Optimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.</p>
+</section>
+<section id="variational-inference" class="level2">
+<h2 class="anchored" data-anchor-id="variational-inference">Variational inference</h2>
+<p>Variational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.</p>
+</section>
+<section id="model-diagnostics" class="level2">
+<h2 class="anchored" data-anchor-id="model-diagnostics">Model diagnostics</h2>
+<p>Model diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.</p>
+</section>
+<section id="output" class="level2">
+<h2 class="anchored" data-anchor-id="output">Output</h2>
+<p>For each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.</p>
+<section id="generated-quantities" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="generated-quantities">Generated quantities</h3>
+<p>Before generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.</p>
+<p>After the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.</p>
+</section>
+<section id="write" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="write">Write</h3>
+<p>The final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.</p>
+<p>In the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Hoffman-Gelman:2014" class="csl-entry" role="listitem">
+Hoffman, Matthew D., and Andrew Gelman. 2014. <span>“<span>T</span>he <span>N</span>o-<span>U</span>-<span>T</span>urn <span>S</span>ampler: <span>A</span>daptively <span>S</span>etting <span>P</span>ath <span>L</span>engths in <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>Journal of Machine Learning Research</em> 15: 1593–623. <a href="http://jmlr.org/papers/v15/hoffman14a.html">http://jmlr.org/papers/v15/hoffman14a.html</a>.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The C++ code underlying Stan is flexible enough to allow data to be read from memory or file. Calls from R, for instance, can be configured to read data from file or directly from R’s memory.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>In the R version of Stan, the values may either be written to a CSV file or directly back to R’s memory.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/syntax.html" class="pagination-link" aria-label="Language Syntax">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Language Syntax</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/deprecations.html" class="pagination-link" aria-label="Deprecated Features">
+        <span class="nav-page-text">Deprecated Features</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/execution.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/expressions.html b/docs/2_39/reference-manual/expressions.html
new file mode 100644
index 000000000..08aab3a8f
--- /dev/null
+++ b/docs/2_39/reference-manual/expressions.html
@@ -0,0 +1,2293 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Expressions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/statements.html" rel="next">
+<link href="../reference-manual/types.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Expressions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/expressions.html">Expressions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#expressions" id="toc-expressions" class="nav-link active" data-scroll-target="#expressions">Expressions</a>
+  <ul class="collapse">
+  <li><a href="#numeric-literals" id="toc-numeric-literals" class="nav-link" data-scroll-target="#numeric-literals">Numeric literals</a>
+  <ul class="collapse">
+  <li><a href="#integer-literals" id="toc-integer-literals" class="nav-link" data-scroll-target="#integer-literals">Integer literals</a></li>
+  <li><a href="#real-literals" id="toc-real-literals" class="nav-link" data-scroll-target="#real-literals">Real literals</a></li>
+  <li><a href="#imaginary-literals" id="toc-imaginary-literals" class="nav-link" data-scroll-target="#imaginary-literals">Imaginary literals</a></li>
+  <li><a href="#complex-literals" id="toc-complex-literals" class="nav-link" data-scroll-target="#complex-literals">Complex literals</a></li>
+  </ul></li>
+  <li><a href="#variables.section" id="toc-variables.section" class="nav-link" data-scroll-target="#variables.section">Variables</a>
+  <ul class="collapse">
+  <li><a href="#reserved-names" id="toc-reserved-names" class="nav-link" data-scroll-target="#reserved-names">Reserved names</a></li>
+  <li><a href="#legal-characters" id="toc-legal-characters" class="nav-link" data-scroll-target="#legal-characters">Legal characters</a></li>
+  </ul></li>
+  <li><a href="#vector-matrix-array-expressions.section" id="toc-vector-matrix-array-expressions.section" class="nav-link" data-scroll-target="#vector-matrix-array-expressions.section">Container expressions</a>
+  <ul class="collapse">
+  <li><a href="#vector-expressions" id="toc-vector-expressions" class="nav-link" data-scroll-target="#vector-expressions">Vector expressions</a></li>
+  <li><a href="#matrix-expressions" id="toc-matrix-expressions" class="nav-link" data-scroll-target="#matrix-expressions">Matrix expressions</a></li>
+  <li><a href="#complex-vector-and-matrix-expressions" id="toc-complex-vector-and-matrix-expressions" class="nav-link" data-scroll-target="#complex-vector-and-matrix-expressions">Complex vector and matrix expressions</a></li>
+  <li><a href="#empty-vectors-and-matrices" id="toc-empty-vectors-and-matrices" class="nav-link" data-scroll-target="#empty-vectors-and-matrices">Empty vectors and matrices</a></li>
+  <li><a href="#array-expressions" id="toc-array-expressions" class="nav-link" data-scroll-target="#array-expressions">Array expressions</a></li>
+  <li><a href="#empty-arrays" id="toc-empty-arrays" class="nav-link" data-scroll-target="#empty-arrays">Empty arrays</a></li>
+  <li><a href="#array-expression-types" id="toc-array-expression-types" class="nav-link" data-scroll-target="#array-expression-types">Array expression types</a></li>
+  <li><a href="#tuple-expressions-and-types" id="toc-tuple-expressions-and-types" class="nav-link" data-scroll-target="#tuple-expressions-and-types">Tuple expressions and types</a></li>
+  <li><a href="#restrictions-on-values" id="toc-restrictions-on-values" class="nav-link" data-scroll-target="#restrictions-on-values">Restrictions on values</a></li>
+  </ul></li>
+  <li><a href="#parentheses-for-grouping" id="toc-parentheses-for-grouping" class="nav-link" data-scroll-target="#parentheses-for-grouping">Parentheses for grouping</a></li>
+  <li><a href="#arithmetic-expressions.section" id="toc-arithmetic-expressions.section" class="nav-link" data-scroll-target="#arithmetic-expressions.section">Arithmetic and matrix operations on expressions</a>
+  <ul class="collapse">
+  <li><a href="#operator-precedence-and-associativity" id="toc-operator-precedence-and-associativity" class="nav-link" data-scroll-target="#operator-precedence-and-associativity">Operator precedence and associativity</a></li>
+  </ul></li>
+  <li><a href="#conditional-operator.section" id="toc-conditional-operator.section" class="nav-link" data-scroll-target="#conditional-operator.section">Conditional operator</a>
+  <ul class="collapse">
+  <li><a href="#conditional-operator-syntax" id="toc-conditional-operator-syntax" class="nav-link" data-scroll-target="#conditional-operator-syntax">Conditional operator syntax</a></li>
+  <li><a href="#conditional-operator-semantics" id="toc-conditional-operator-semantics" class="nav-link" data-scroll-target="#conditional-operator-semantics">Conditional operator semantics</a></li>
+  </ul></li>
+  <li><a href="#language-indexing.section" id="toc-language-indexing.section" class="nav-link" data-scroll-target="#language-indexing.section">Indexing</a>
+  <ul class="collapse">
+  <li><a href="#accessing-subarrays" id="toc-accessing-subarrays" class="nav-link" data-scroll-target="#accessing-subarrays">Accessing subarrays</a></li>
+  <li><a href="#accessing-matrix-rows" id="toc-accessing-matrix-rows" class="nav-link" data-scroll-target="#accessing-matrix-rows">Accessing matrix rows</a></li>
+  <li><a href="#mixing-array-and-vectormatrix-indexes" id="toc-mixing-array-and-vectormatrix-indexes" class="nav-link" data-scroll-target="#mixing-array-and-vectormatrix-indexes">Mixing array and vector/matrix indexes</a></li>
+  </ul></li>
+  <li><a href="#language-multi-indexing.section" id="toc-language-multi-indexing.section" class="nav-link" data-scroll-target="#language-multi-indexing.section">Multiple indexing and range indexing</a>
+  <ul class="collapse">
+  <li><a href="#multiple-index-semantics" id="toc-multiple-index-semantics" class="nav-link" data-scroll-target="#multiple-index-semantics">Multiple index semantics</a></li>
+  </ul></li>
+  <li><a href="#function-application.section" id="toc-function-application.section" class="nav-link" data-scroll-target="#function-application.section">Function application</a>
+  <ul class="collapse">
+  <li><a href="#type-signatures-and-result-type-inference" id="toc-type-signatures-and-result-type-inference" class="nav-link" data-scroll-target="#type-signatures-and-result-type-inference">Type signatures and result type inference</a></li>
+  <li><a href="#constants" id="toc-constants" class="nav-link" data-scroll-target="#constants">Constants</a></li>
+  <li><a href="#type-promotion-and-function-resolution" id="toc-type-promotion-and-function-resolution" class="nav-link" data-scroll-target="#type-promotion-and-function-resolution">Type promotion and function resolution</a></li>
+  <li><a href="#random-number-generating-functions" id="toc-random-number-generating-functions" class="nav-link" data-scroll-target="#random-number-generating-functions">Random-number generating functions</a></li>
+  </ul></li>
+  <li><a href="#type-inference" id="toc-type-inference" class="nav-link" data-scroll-target="#type-inference">Type inference</a>
+  <ul class="collapse">
+  <li><a href="#implementation-types" id="toc-implementation-types" class="nav-link" data-scroll-target="#implementation-types">Implementation types</a></li>
+  <li><a href="#type-inference-rules" id="toc-type-inference-rules" class="nav-link" data-scroll-target="#type-inference-rules">Type inference rules</a></li>
+  <li><a href="#promotion" id="toc-promotion" class="nav-link" data-scroll-target="#promotion">Promotion</a></li>
+  </ul></li>
+  <li><a href="#higher-order-functions" id="toc-higher-order-functions" class="nav-link" data-scroll-target="#higher-order-functions">Higher-order functions</a>
+  <ul class="collapse">
+  <li><a href="#functions-passed-by-reference" id="toc-functions-passed-by-reference" class="nav-link" data-scroll-target="#functions-passed-by-reference">Functions passed by reference</a></li>
+  <li><a href="#data-restricted-arguments" id="toc-data-restricted-arguments" class="nav-link" data-scroll-target="#data-restricted-arguments">Data-restricted arguments</a></li>
+  </ul></li>
+  <li><a href="#chain-rule-and-derivatives" id="toc-chain-rule-and-derivatives" class="nav-link" data-scroll-target="#chain-rule-and-derivatives">Chain rule and derivatives</a>
+  <ul class="collapse">
+  <li><a href="#errors-due-to-chain-rule" id="toc-errors-due-to-chain-rule" class="nav-link" data-scroll-target="#errors-due-to-chain-rule">Errors due to chain rule</a></li>
+  <li><a href="#diagnosing-problems-with-derivatives" id="toc-diagnosing-problems-with-derivatives" class="nav-link" data-scroll-target="#diagnosing-problems-with-derivatives">Diagnosing problems with derivatives</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/expressions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/expressions.html">Expressions</a></li></ol></nav></header>
+
+
+
+
+<section id="expressions" class="level1">
+<h1>Expressions</h1>
+<p>An expression is the syntactic unit in a Stan program that denotes a value. Every expression in a well-formed Stan program has a type that is determined statically (at compile time), based only on the type of its variables and the types of the functions used in it. If an expressions type cannot be determined statically, the Stan compiler will report the location of the problem.</p>
+<p>This chapter covers the syntax, typing, and usage of the various forms of expressions in Stan.</p>
+<section id="numeric-literals" class="level2">
+<h2 class="anchored" data-anchor-id="numeric-literals">Numeric literals</h2>
+<p>The simplest form of expression is a literal that denotes a primitive numerical value.</p>
+<section id="integer-literals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="integer-literals">Integer literals</h3>
+<p>Integer literals represent integers of type <code>int</code>. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression <code>--1</code> is interpreted as the negation of the literal <code>-1</code>.)</p>
+<p>The following list contains well-formed integer literals.</p>
+<pre><code>0, 1, -1, 256, -127098, 24567898765</code></pre>
+<p>Integer literals must have values that fall within the bounds for integer values (see <a href="../reference-manual/types.html#numerical-data-types.section">the section on numerical data types</a>).</p>
+<p>Integer literals may not contain decimal points (<code>.</code>). Thus the expressions <code>1.</code> and <code>1.0</code> are of type <code>real</code> and may not be used where a value of type <code>int</code> is required.</p>
+</section>
+<section id="real-literals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="real-literals">Real literals</h3>
+<p>A number written with a period or with scientific notation is assigned to a the continuous numeric type <code>real</code>. Real literals are written in base 10 with a period (<code>.</code>) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.</p>
+<pre><code>0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.</code></pre>
+<p>The notation <code>e</code> or <code>E</code> followed by a positive or negative integer denotes a power of 10 to multiply. For instance, <code>2.7e3</code> and <code>2.7e+3</code> denote <span class="math inline">\(2.7 \times 10^3\)</span>, whereas <code>-2E-5</code> denotes <span class="math inline">\(-2 \times
+10^{-5}\)</span>.</p>
+</section>
+<section id="imaginary-literals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="imaginary-literals">Imaginary literals</h3>
+<p>A number followed by the character <code>i</code> denotes an imaginary number and is assigned to the numeric type <code>complex</code>. The number preceding <code>i</code> may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.</p>
+<pre><code>1i, 2i, -325.786i, 1e10i, 2.87e-10i.</code></pre>
+<p>Note that the character <code>i</code> by itself is <em>not</em> a well-formed imaginary literal. The unit imaginary number must be written as <code>1i</code>.</p>
+</section>
+<section id="complex-literals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-literals">Complex literals</h3>
+<p>Stan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.</p>
+<pre><code>1 + 2i, -3.2e9 + 1e10i</code></pre>
+<p>These will be assigned the type <code>complex</code>, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.</p>
+</section>
+</section>
+<section id="variables.section" class="level2">
+<h2 class="anchored" data-anchor-id="variables.section">Variables</h2>
+<p>A variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (<code>_</code>) character. Variables must start with a letter (<code>a--z</code> and <code>A--Z</code>) and may not end with two underscores (<code>__</code>).</p>
+<p>Examples of legal variable identifiers are as follows.</p>
+<pre><code>a, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable</code></pre>
+<p>Unlike in R and BUGS, variable identifiers in Stan may not contain a period character.</p>
+<section id="reserved-names" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reserved-names">Reserved names</h3>
+<p>Stan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the <code>stanc</code> translator halting with an error message indicating which reserved name was used and its location in the model code.</p>
+<section id="model-name" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="model-name">Model name</h4>
+<p>The name of the model cannot be used as a variable within the model. This is usually not a problem because the default in <code>bin/stanc</code> is to append <code>_model</code> to the name of the file containing the model specification. For example, if the model is in file <code>foo.stan</code>, it would not be legal to have a variable named <code>foo_model</code> when using the default model name through <code>bin/stanc</code>. With user-specified model names, variables cannot match the model.</p>
+</section>
+<section id="reserved-words-from-stan-language" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="reserved-words-from-stan-language">Reserved words from Stan language</h4>
+<p>The following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.</p>
+<pre><code>for, in, while, repeat, until, if, then, else,
+true, false, target, struct, typedef, export,
+auto, extern, var, static, lower, upper, offset,
+multiplier</code></pre>
+<p>Variables should not be named after types, either, and thus may not be any of the following.</p>
+<pre><code>int, real, complex, vector, simplex, unit_vector,
+sum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,
+row_vector, matrix, cholesky_factor_corr,
+column_stochastic_matrix,row_stochastic_matrix,
+cholesky_factor_cov, corr_matrix, cov_matrix, array</code></pre>
+<p>The following keywords or built-in functions are also reserved and cannot be used as variable names:</p>
+<pre><code>print, reject, profile, fatal_error, target, jacobian</code></pre>
+<p>The following block identifiers are reserved and cannot be used as variable names:</p>
+<pre><code>functions, model, data, parameters, quantities,
+transformed, generated</code></pre>
+</section>
+<section id="reserved-distribution-names" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="reserved-distribution-names">Reserved distribution names</h4>
+<p>Variable names will also conflict with the names of distributions suffixed with <code>_lpdf</code>, <code>_lpmf</code>, <code>_lcdf</code>, and <code>_lccdf</code>, <code>_cdf</code>, and <code>_ccdf</code>, such as <code>normal_lcdf_log</code>. No user-defined variable can take a name ending in <code>_lupdf</code> or <code>_lupmf</code> even if a corresponding <code>_lpdf</code> or <code>_lpmf</code> is not defined.</p>
+<p>Using any of these variable names causes the <code>stanc</code> translator to halt and report the name and location of the variable causing the conflict.</p>
+</section>
+<section id="reserved-names-backend-languages" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="reserved-names-backend-languages">Reserved names backend languages</h4>
+<p>Stan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed <code>_stan_NAME</code> (e.g.&nbsp;<code>_stan_public</code>) behind the scenes (in the generated C++ code).</p>
+<!-- corresponds to the list in stanc3/src/stan_math_backend/Mangle.ml -->
+<pre><code>alignas, alignof, and, and_eq, asm, bitand, bitor, bool,
+case, catch, char, char16_t, char32_t, class, compl, const,
+constexpr, const_cast, decltype, default, delete, do,
+double, dynamic_cast, enum, explicit, float, friend, goto,
+inline, long, mutable, namespace, new, noexcept, not, not_eq,
+nullptr, operator, or, or_eq, private, protected, public,
+register, reinterpret_cast, short, signed, sizeof,
+static_assert, static_cast, switch, template, this, thread_local,
+throw, try, typeid, typename, union, unsigned, using, virtual,
+volatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,
+STAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH</code></pre>
+</section>
+</section>
+<section id="legal-characters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="legal-characters">Legal characters</h3>
+<p>The legal characters for variable identifiers are given in the following table.</p>
+<section id="identifier-characters-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="identifier-characters-table">Identifier Characters Table</h5>
+<p><em>The alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: center;">characters</th>
+<th style="text-align: center;">ASCII code points</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;"><code>a -- z</code></td>
+<td style="text-align: center;">97 – 122</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>A -- Z</code></td>
+<td style="text-align: center;">65 – 90</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>0 -- 9</code></td>
+<td style="text-align: center;">48 – 57</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>_</code></td>
+<td style="text-align: center;">95</td>
+</tr>
+</tbody>
+</table>
+<p>Although not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.</p>
+</section>
+<section id="comments-allow-ascii-compatible-encoding" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="comments-allow-ascii-compatible-encoding">Comments allow ASCII-compatible encoding</h4>
+<p>Within comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.</p>
+</section>
+</section>
+</section>
+<section id="vector-matrix-array-expressions.section" class="level2">
+<h2 class="anchored" data-anchor-id="vector-matrix-array-expressions.section">Container expressions</h2>
+<p>Expressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.</p>
+<section id="vector-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vector-expressions">Vector expressions</h3>
+<p>Square brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression <code>[ 1, 10, 100 ]</code> denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">2</span>] rv2 =  [ <span class="dv">1</span>, <span class="dv">2</span> ];</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">3</span>] v3 = [ <span class="dv">3</span>, <span class="dv">4</span>, <span class="dv">5</span> ]';</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The vector expression values may be compound expressions or variable names, so it is legal to write <code>[ 2 * 3, 1 + 4]</code> or <code>[ x, y ]</code>, providing that <code>x</code> and <code>y</code> are primitive variables.</p>
+</section>
+<section id="matrix-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrix-expressions">Matrix expressions</h3>
+<p>A matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] m1 = [ [ <span class="dv">1</span>, <span class="dv">2</span> ], [ <span class="dv">3</span>, <span class="dv">4</span> ], [<span class="dv">5</span>, <span class="dv">6</span> ] ];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Any expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">2</span>] vX = [ <span class="dv">1</span>, <span class="dv">10</span> ]';</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">2</span>] vY = [ <span class="dv">100</span>, <span class="dv">1000</span> ];</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] m2 = [ vX', vY, [ <span class="dv">1</span>, <span class="dv">2</span> ]  ];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="complex-vector-and-matrix-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-vector-and-matrix-expressions">Complex vector and matrix expressions</h3>
+<p>Complex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_vector</span>[<span class="dv">3</span>] = [<span class="dv">1</span> + <span class="fl">2i</span>, <span class="dv">3</span> - <span class="fl">1.7i</span>, <span class="dv">0</span>]';</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_row_vector</span>[<span class="dv">2</span>] = [<span class="dv">12</span>, -<span class="fl">2i</span>];</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_matrix</span>[<span class="dv">2</span>, <span class="dv">3</span>] = [[<span class="dv">1</span> + <span class="fl">2i</span>, <span class="dv">3</span> - <span class="fl">1.7i</span>, <span class="dv">0</span>],</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>                        [<span class="fl">3.9</span> - <span class="fl">1.234i</span>, <span class="fl">176i</span>, <span class="dv">1</span> + <span class="fl">1i</span>]];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="no-empty-vector-or-matrix-expressions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="no-empty-vector-or-matrix-expressions">No empty vector or matrix expressions</h4>
+<p>The empty expression <code>[ ]</code> is ambiguous and therefore is not allowed and similarly expressions such as <code>[ [ ] ]</code> or <code>[ [ ], [ ] ]</code> are not allowed.</p>
+</section>
+</section>
+<section id="empty-vectors-and-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="empty-vectors-and-matrices">Empty vectors and matrices</h3>
+<p>If needed, it is possible to create an empty vector with</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a>rep_vector(e, <span class="dv">0</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the first expression <code>e</code> needs to scalar of type <code>real</code>.</p>
+<p>If needed, it is possible to create an empty matrix with</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a>rep_matrix(e, <span class="dv">0</span>, <span class="dv">0</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the first expression <code>e</code> needs to scalar of type <code>real</code>.</p>
+</section>
+<section id="array-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="array-expressions">Array expressions</h3>
+<p>Curly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression <code>{ 1, 10, 100 }</code> denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> a = { <span class="dv">1</span>, <span class="dv">10</span>, <span class="dv">100</span> };</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The values may be compound expressions, so it is legal to write <code>{ 2 * 3, 1 + 4 }</code>. It is also possible to write two dimensional arrays directly, as in the following example.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>] <span class="dt">int</span> b = { { <span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span> }, { <span class="dv">4</span>, <span class="dv">5</span>, <span class="dv">6</span> } };</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This way, <code>b[1]</code> is <code>{ 1, 2, 3 }</code> and <code>b[2]</code> is <code>{ 4, 5, 6 }</code>.</p>
+<p>Whitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>] <span class="dt">int</span> b = { { <span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span> },</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>                { <span class="dv">4</span>, <span class="dv">5</span>, <span class="dv">6</span> } };</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="empty-arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="empty-arrays">Empty arrays</h3>
+<p>The empty array expression (<code>{ }</code>) is not allowed. See more about restrictions on array expressions in subsection <a href="#array-expressions">Restrictions on values</a>.</p>
+<p>If needed, it is possible to create an empty array with</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a>rep_array(e, <span class="dv">0</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the first expression <code>e</code> determines the type of the array. For example, <code>rep_array(0.0, 0)</code> returns an empty real array of type <code>real[]</code>, whereas <code>rep_array({123}, 0)</code> returns an empty two dimensional integer array of type <code>int[ , ]</code>. Only the type of the first argument is used, so the integer arrays <code>{123}</code> and <code>{0}</code> produce equivalent values.</p>
+</section>
+<section id="array-expression-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="array-expression-types">Array expression types</h3>
+<p>Any type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">3</span>] b;</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">3</span>] c;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">vector</span>[<span class="dv">3</span>] d = { b, c };</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The elements may also be a mixture of <code>int</code> and <code>real</code> typed expressions, in which case the result is an array of real values.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">real</span> b = { <span class="dv">1</span>, <span class="fl">1.9</span> };</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="tuple-expressions-and-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="tuple-expressions-and-types">Tuple expressions and types</h3>
+<p>Stan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">vector</span>[<span class="dv">3</span>]) xy = (<span class="dv">42</span>, [<span class="dv">1</span>, <span class="fl">2.9</span>, -<span class="fl">1.3</span>]');</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The expression <code>42</code> is of type <code>int</code> and the expression <code>[1, 2.9, -1.3]</code> is of type <code>row_vector</code> so that <code>[1, 2.9, -1.3]'</code> is of type <code>vector</code> and of size 3. The whole tuple expression <code>(42, [1, 2.9, -1.3]')</code> thus has a sized type of <code>tuple(int, vector[3])</code> and an unsized type (e.g., for a function argument) of <code>tuple(int, vector)</code>.</p>
+<p>A tuple of one element can be created using the same style as languages like Python, with a trailing comma, e.g., <code>(3.14,)</code>. For longer tuples, Stan does <em>not</em> support trailing commas.</p>
+</section>
+<section id="restrictions-on-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="restrictions-on-values">Restrictions on values</h3>
+<p>There are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.</p>
+<section id="rectangular-array-expressions-only" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="rectangular-array-expressions-only">Rectangular array expressions only</h4>
+<p>Although it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a>{ { <span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span> }, { <span class="dv">4</span>, <span class="dv">5</span> } }  <span class="co">// compile time error: size mismatch</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This may appear to be OK, because it is creating a two-dimensional integer array (<code>array[,] int</code>) out of two one-dimensional array integer arrays (<code>array[] int</code>). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a>{ { <span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span> }, m }  <span class="co">// runtime error if m not size 3</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="no-empty-array-expressions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="no-empty-array-expressions">No empty array expressions</h4>
+<p>Because there is no way to infer the type of the result, the empty array expression (<code>{ }</code>) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.</p>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">int</span> a;   <span class="co">// a is fully defined as zero element array</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="no-zero-tuples" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="no-zero-tuples">No zero-tuples</h4>
+<p>There is no way to declare or construct a zero-tuple in Stan. Tuples must be at least one element long. The expression <code>()</code> does not pick out a zero-tuple—it is ill formed.</p>
+<p>One-tuples need a trailing comma, like <code>(1,)</code>. The expression <code>(1)</code>, without the comma, is of type <code>int</code> rather than a tuple.</p>
+</section>
+</section>
+</section>
+<section id="parentheses-for-grouping" class="level2">
+<h2 class="anchored" data-anchor-id="parentheses-for-grouping">Parentheses for grouping</h2>
+<p>Any expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, <code>(</code> and <code>)</code>, are allowed. The square brackets <code>[</code> and <code>]</code> are reserved for array indexing and the curly braces <code>{</code> and <code>}</code> for grouping statements.</p>
+<p>With parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression <code>1 + 2 * 3</code> has a subexpression <code>2 * 3</code> and evaluates to 7. With parentheses, this grouping may be made explicit with the expression <code>1 + (2 * 3)</code>. More importantly, the expression <code>(1 + 2) *   3</code> has <code>1 + 2</code> as a subexpression and evaluates to 9.</p>
+</section>
+<section id="arithmetic-expressions.section" class="level2">
+<h2 class="anchored" data-anchor-id="arithmetic-expressions.section">Arithmetic and matrix operations on expressions</h2>
+<p>For integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (<code>+</code>), subtraction (<code>-</code>), multiplication (<code>*</code>) and division (<code>/</code>) in the usual ways.</p>
+<p>For integer expressions, Stan supports the modulus (<code>%</code>) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming <code>n</code> and <code>m</code> are integer variables and <code>x</code> and <code>y</code> real variables, the following expressions are legal.</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a><span class="fl">3.0</span> + <span class="fl">0.14</span></span>
+<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a>-<span class="dv">15</span></span>
+<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a><span class="dv">2</span> * <span class="dv">3</span> + <span class="dv">1</span></span>
+<span id="cb27-4"><a href="#cb27-4" aria-hidden="true" tabindex="-1"></a>(x - y) / <span class="fl">2.0</span></span>
+<span id="cb27-5"><a href="#cb27-5" aria-hidden="true" tabindex="-1"></a>(n * (n + <span class="dv">1</span>)) / <span class="dv">2</span></span>
+<span id="cb27-6"><a href="#cb27-6" aria-hidden="true" tabindex="-1"></a>x / n</span>
+<span id="cb27-7"><a href="#cb27-7" aria-hidden="true" tabindex="-1"></a>m % n</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (<code>'</code>) is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.</p>
+<p>For example, if <code>y</code> and <code>mu</code> are variables of type <code>vector</code> and <code>Sigma</code> is a variable of type <code>matrix</code>, then <code>(y - mu)' * Sigma * (y - mu)</code> is a well-formed expression of type <code>real</code>. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) <code>y - mu</code> are of type <code>vector</code> because the variables <code>y</code> and <code>mu</code> are of type <code>vector</code>. The transpose of this expression, the subexpression <code>(y - mu)'</code> is of type <code>row_vector</code>. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form <code>(((y - mu)') * Sigma) * (y - mu)</code>.</p>
+<p>The type of subexpression <code>(y - mu)' * Sigma</code> is inferred to be <code>row_vector</code>, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a <code>real</code> value.</p>
+<p>Stan provides elementwise matrix multiplication (e.g., <code>a .* b</code>) and division (e.g., <code>a ./ b</code>) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,</p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] a;</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] b;</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] c;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>the assignment,</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>c = a .* b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>produces the same result with roughly the same efficiency as the loop</p>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a>  c[n] = a[n] * b[n];</span>
+<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Stan supports exponentiation (<code>^</code>) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming <code>n</code> and <code>m</code> are integer variables and <code>x</code> and <code>y</code> real variables, the following expressions are legal.</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="dv">3</span> ^ <span class="dv">2</span></span>
+<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a><span class="fl">3.0</span> ^ -<span class="dv">2</span></span>
+<span id="cb31-3"><a href="#cb31-3" aria-hidden="true" tabindex="-1"></a><span class="fl">3.0</span> ^ <span class="fl">0.14</span></span>
+<span id="cb31-4"><a href="#cb31-4" aria-hidden="true" tabindex="-1"></a>x ^ n</span>
+<span id="cb31-5"><a href="#cb31-5" aria-hidden="true" tabindex="-1"></a>n ^ x</span>
+<span id="cb31-6"><a href="#cb31-6" aria-hidden="true" tabindex="-1"></a>n ^ m</span>
+<span id="cb31-7"><a href="#cb31-7" aria-hidden="true" tabindex="-1"></a>x ^ y</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Exponentiation is right associative, so the expression <code>2 ^ 3 ^ 4</code> is equivalent to the fully specified form <code>2 ^ (3 ^ 4)</code>.</p>
+<section id="operator-precedence-and-associativity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="operator-precedence-and-associativity">Operator precedence and associativity</h3>
+<p>The precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.</p>
+<section id="operator-precedence-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="operator-precedence-table">Operator Precedence Table</h5>
+<p><em>Stan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e.,</em> <code>operator*(int, int):int</code> <em>indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.</em></p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 12%">
+<col style="width: 8%">
+<col style="width: 10%">
+<col style="width: 22%">
+<col style="width: 45%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: center;">Op.</th>
+<th style="text-align: right;">Prec.</th>
+<th style="text-align: center;">Assoc.</th>
+<th style="text-align: left;">Placement</th>
+<th style="text-align: left;">Description</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;"><code>? ~ :</code></td>
+<td style="text-align: right;">10</td>
+<td style="text-align: center;">right</td>
+<td style="text-align: left;">ternary infix</td>
+<td style="text-align: left;">conditional</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>||</code></td>
+<td style="text-align: right;">9</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">logical or</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>&amp;&amp;</code></td>
+<td style="text-align: right;">8</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">logical and</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>==</code></td>
+<td style="text-align: right;">7</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">equality</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>!=</code></td>
+<td style="text-align: right;">7</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">inequality</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>&lt;</code></td>
+<td style="text-align: right;">6</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">less than</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>&lt;=</code></td>
+<td style="text-align: right;">6</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">less than or equal</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>&gt;</code></td>
+<td style="text-align: right;">6</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">greater than</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>&gt;=</code></td>
+<td style="text-align: right;">6</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">greater than or equal</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>+</code></td>
+<td style="text-align: right;">5</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">addition</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>-</code></td>
+<td style="text-align: right;">5</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">subtraction</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>*</code></td>
+<td style="text-align: right;">4</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">multiplication</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>.*</code></td>
+<td style="text-align: right;">4</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">elementwise multiplication</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>/</code></td>
+<td style="text-align: right;">4</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">(right) division</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>./</code></td>
+<td style="text-align: right;">4</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">elementwise division</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>%</code></td>
+<td style="text-align: right;">4</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">modulus</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>\</code></td>
+<td style="text-align: right;">3</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">left division</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>%/%</code></td>
+<td style="text-align: right;">3</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">integer division</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>!</code></td>
+<td style="text-align: right;">2</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: left;">unary prefix</td>
+<td style="text-align: left;">logical negation</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>-</code></td>
+<td style="text-align: right;">2</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: left;">unary prefix</td>
+<td style="text-align: left;">negation</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>+</code></td>
+<td style="text-align: right;">2</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: left;">unary prefix</td>
+<td style="text-align: left;">promotion (no-op in Stan)</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>^</code></td>
+<td style="text-align: right;">1</td>
+<td style="text-align: center;">right</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">exponentiation</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>.^</code></td>
+<td style="text-align: right;">1</td>
+<td style="text-align: center;">right</td>
+<td style="text-align: left;">binary infix</td>
+<td style="text-align: left;">elementwise exponentiation</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>'</code></td>
+<td style="text-align: right;">0</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: left;">unary postfix</td>
+<td style="text-align: left;">transposition</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>()</code></td>
+<td style="text-align: right;">0</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: left;">prefix, wrap</td>
+<td style="text-align: left;">function application</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>[]</code></td>
+<td style="text-align: right;">0</td>
+<td style="text-align: center;">left</td>
+<td style="text-align: left;">prefix, wrap</td>
+<td style="text-align: left;">array, matrix indexing</td>
+</tr>
+</tbody>
+</table>
+<p>Other expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.</p>
+<p>The precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression <code>a + b + c</code> is interpreted as <code>(a + b) + c</code>. Similarly, <code>a / b * c</code> is interpreted as <code>(a / b) * c</code>.</p>
+<p>Because multiplication has higher precedence than addition, the expression <code>a * b + c</code> is interpreted as <code>(a * b) + c</code> and the expression <code>a + b * c</code> is interpreted as <code>a + (b * c)</code>. Similarly, <code>2 * x + 3 * - y</code> is interpreted as <code>(2 * x) + (3 * (-y))</code>.</p>
+<p>Transposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, <code>-u'</code> is interpreted as <code>-(u')</code>, <code>u * v'</code> as <code>u* (v')</code>, and <code>u' * v</code> as <code>(u') * v</code>. For integer and reals, <code>-n ^ 3</code> is interpreted as <code>-(n ^ 3)</code>.</p>
+</section>
+</section>
+</section>
+<section id="conditional-operator.section" class="level2">
+<h2 class="anchored" data-anchor-id="conditional-operator.section">Conditional operator</h2>
+<section id="conditional-operator-syntax" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="conditional-operator-syntax">Conditional operator syntax</h3>
+<p>The ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If <code>a</code> is an expression of type <code>int</code> and <code>b</code> and <code>c</code> are expressions that can be converted to one another (e.g., compared with <code>==</code>), then</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a>a ? b : c</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is an expression of the promoted type of <code>b</code> and <code>c</code>. The result of this expression is <code>b</code> if condition <code>a</code> is true, <code>c</code> otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g.&nbsp;if one argument is of type <code>int</code> and the other of type <code>real</code>, the conditional expression as a whole is of type <code>real</code>. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.</p>
+<section id="conditional-operator-precedence" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="conditional-operator-precedence">Conditional operator precedence</h4>
+<p>The conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,</p>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a>a &gt; <span class="dv">0</span> || b &lt; <span class="dv">0</span> ? c + d : e - f</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is equivalent to the explicitly grouped version</p>
+<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a>(a &gt; <span class="dv">0</span> || b &lt; <span class="dv">0</span>) ? (c + d) : (e - f)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The latter is easier to read even if the parentheses are not strictly necessary.</p>
+</section>
+<section id="conditional-operator-associativity" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="conditional-operator-associativity">Conditional operator associativity</h4>
+<p>The conditional operator is right associative, so that</p>
+<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a>a ? b : c ? d : e</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>parses as if explicitly grouped as</p>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a>a ? b : (c ? d : e)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Again, the explicitly grouped version is easier to read.</p>
+</section>
+</section>
+<section id="conditional-operator-semantics" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="conditional-operator-semantics">Conditional operator semantics</h3>
+<p>Stan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable <code>a</code> in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value <code>b</code> is returned if the condition evaluates to a non-zero value and <code>c</code> is returned if the condition evaluates to zero.</p>
+<section id="lazy-evaluation-of-results" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="lazy-evaluation-of-results">Lazy evaluation of results</h4>
+<p>The key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.</p>
+</section>
+<section id="promotion-to-parameter" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="promotion-to-parameter">Promotion to parameter</h4>
+<p>If one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full <code>if</code>-<code>then</code> conditional statement. For example,</p>
+<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb37-2"><a href="#cb37-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">10</span>] <span class="dt">real</span> x;</span>
+<span id="cb37-3"><a href="#cb37-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb37-4"><a href="#cb37-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb37-5"><a href="#cb37-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb37-6"><a href="#cb37-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">10</span>] <span class="dt">real</span> z;</span>
+<span id="cb37-7"><a href="#cb37-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb37-8"><a href="#cb37-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb37-9"><a href="#cb37-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb37-10"><a href="#cb37-10" aria-hidden="true" tabindex="-1"></a>  y ~ normal(cond ? x : z, sigma);</span>
+<span id="cb37-11"><a href="#cb37-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb37-12"><a href="#cb37-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>would be more efficiently (if not more transparently) coded as</p>
+<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (cond) {</span>
+<span id="cb38-2"><a href="#cb38-2" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x, sigma);</span>
+<span id="cb38-3"><a href="#cb38-3" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb38-4"><a href="#cb38-4" aria-hidden="true" tabindex="-1"></a>  y ~ normal(z, sigma);</span>
+<span id="cb38-5"><a href="#cb38-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable <code>x</code> will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.</p>
+</section>
+</section>
+</section>
+<section id="language-indexing.section" class="level2">
+<h2 class="anchored" data-anchor-id="language-indexing.section">Indexing</h2>
+<p>Stan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if <code>x</code> is a variable of type <code>array [] real</code> (a one-dimensional array of reals) then <code>x[1]</code> is the value of the first element of the array.</p>
+<p>Subscripting has higher precedence than any of the arithmetic operations. For example, <code>alpha * x[1]</code> is equivalent to <code>alpha * (x[1])</code>.</p>
+<p>Multiple subscripts may be provided within a single pair of square brackets. If <code>x</code> is of type <code>array[,] real</code>, a two-dimensional array, then <code>x[2, 501]</code> is of type <code>real</code>.</p>
+<section id="accessing-subarrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="accessing-subarrays">Accessing subarrays</h3>
+<p>The subscripting operator also returns subarrays of arrays. For example, if <code>x</code> is of type <code>array[,,] real</code>, then <code>x[2]</code> is of type <code>array[,] real</code>, and <code>x[2, 3]</code> is of type <code>array[] real</code>. As a result, the expressions <code>x[2, 3]</code> and <code>x[2][3]</code> have the same meaning.</p>
+</section>
+<section id="accessing-matrix-rows" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="accessing-matrix-rows">Accessing matrix rows</h3>
+<p>If <code>Sigma</code> is a variable of type <code>matrix</code>, then <code>Sigma[1]</code> denotes the first row of <code>Sigma</code> and has the type <code>row_vector</code>.</p>
+</section>
+<section id="mixing-array-and-vectormatrix-indexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mixing-array-and-vectormatrix-indexes">Mixing array and vector/matrix indexes</h3>
+<p>Stan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if <code>m</code> is of type <code>matrix[ , ]</code>, a two-dimensional array of matrices, then <code>m[1]</code> refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that <code>m[1, 2]</code> is of type <code>matrix</code> and denotes the matrix in the first row and second column of the array. Continuing to add indices, <code>m[1, 2, 3]</code> is of type <code>row_vector</code> and denotes the third row of the matrix denoted by <code>m[1, 2]</code>. Finally, <code>m[1, 2, 3, 4]</code> is of type <code>real</code> and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array <code>m</code>.</p>
+</section>
+</section>
+<section id="language-multi-indexing.section" class="level2">
+<h2 class="anchored" data-anchor-id="language-multi-indexing.section">Multiple indexing and range indexing</h2>
+<p>In addition to single integer indexes, as described in <a href="#language-indexing.section">the language indexing section</a>, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.</p>
+<section id="index-types-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="index-types-table">Indexing Options Table</h5>
+<p><em>Types of indexes and examples with one-dimensional containers of size <code>N</code> and an integer array <code>ii</code> of type <code>array [] real</code> size <code>K</code>.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: center;">index type</th>
+<th style="text-align: center;">example</th>
+<th style="text-align: center;">value</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;">integer</td>
+<td style="text-align: center;"><code>a[11]</code></td>
+<td style="text-align: center;">value of <code>a</code> at index 11</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">integer array</td>
+<td style="text-align: center;"><code>a[ii]</code></td>
+<td style="text-align: center;"><code>a[ii[1]]</code>, …, <code>a[ii[K]]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">lower bound</td>
+<td style="text-align: center;"><code>a[3:]</code></td>
+<td style="text-align: center;"><code>a[3]</code>, …, <code>a[N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">upper bound</td>
+<td style="text-align: center;"><code>a[:5]</code></td>
+<td style="text-align: center;"><code>a[1]</code>, …, <code>a[5]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">range</td>
+<td style="text-align: center;"><code>a[2:7]</code></td>
+<td style="text-align: center;"><code>a[2]</code>, …, <code>a[7]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">range</td>
+<td style="text-align: center;"><code>a[7:2]</code></td>
+<td style="text-align: center;">[]</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">range</td>
+<td style="text-align: center;"><code>a[5-3:5+2]</code></td>
+<td style="text-align: center;"><code>a[2]</code>, …, <code>a[7]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">all</td>
+<td style="text-align: center;"><code>a[:]</code></td>
+<td style="text-align: center;"><code>a[1]</code>, …, <code>a[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;">all</td>
+<td style="text-align: center;"><code>a[]</code></td>
+<td style="text-align: center;"><code>a[1]</code>, …, <code>a[N]</code></td>
+</tr>
+</tbody>
+</table>
+<p>The range indexing with <code>:</code> allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:</p>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">6</span>] <span class="dt">int</span> ii = reverse(linspaced_int_array(<span class="dv">6</span>, <span class="dv">2</span>, <span class="dv">7</span>));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then <code>a[ii]</code> evaluates to <code>a[7]</code>, …, <code>a[2]</code>.</p>
+</section>
+<section id="multiple-index-semantics" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multiple-index-semantics">Multiple index semantics</h3>
+<p>The fundamental semantic rule for dealing with multiple indexes is the following. If <code>idxs</code> is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.</p>
+<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a>a[idxs, ...][i, ...] = a[idxs[i], ...][...]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>On the other hand, if <code>idx</code> is a single index, it reduces the dimensionality of the output, so that</p>
+<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a>a[idx, ...] = a[idx][...]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.</p>
+<section id="matrix-indexing-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="matrix-indexing-table">Matrix Indexing Table</h5>
+<p><em>Special rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix <code>a</code>, with integer single indexes <code>i</code> and <code>j</code> and integer array multiple indexes <code>is</code> and <code>js</code>. The same typing rules apply for all multiple indexes.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: center;">example</th>
+<th style="text-align: center;">row index</th>
+<th style="text-align: center;">column index</th>
+<th style="text-align: center;">result type</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;"><code>a[i]</code></td>
+<td style="text-align: center;">single</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: center;">row vector</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>a[is]</code></td>
+<td style="text-align: center;">multiple</td>
+<td style="text-align: center;">n/a</td>
+<td style="text-align: center;">matrix</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>a[i, j]</code></td>
+<td style="text-align: center;">single</td>
+<td style="text-align: center;">single</td>
+<td style="text-align: center;">real</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>a[i, js]</code></td>
+<td style="text-align: center;">single</td>
+<td style="text-align: center;">multiple</td>
+<td style="text-align: center;">row vector</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>a[is, j]</code></td>
+<td style="text-align: center;">multiple</td>
+<td style="text-align: center;">single</td>
+<td style="text-align: center;">vector</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>a[is, js]</code></td>
+<td style="text-align: center;">multiple</td>
+<td style="text-align: center;">multiple</td>
+<td style="text-align: center;">matrix</td>
+</tr>
+</tbody>
+</table>
+<p>Evaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.</p>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a>m[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]</span>
+<span id="cb42-2"><a href="#cb42-2" aria-hidden="true" tabindex="-1"></a>m[idxs, idx][j] = m[idxs[j], idx]</span>
+<span id="cb42-3"><a href="#cb42-3" aria-hidden="true" tabindex="-1"></a>m[idx, idxs][j] = m[idx, idxs[j]]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Evaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.</p>
+</section>
+</section>
+</section>
+<section id="function-application.section" class="level2">
+<h2 class="anchored" data-anchor-id="function-application.section">Function application</h2>
+<p>Stan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.</p>
+<p>Expressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, <code>log(2.0)</code> is the expression of type <code>real</code> denoting the result of applying the natural logarithm to the value of the real literal <code>2.0</code>.</p>
+<p>Syntactically, function application has higher precedence than any of the other operators, so that <code>y + log(x)</code> is interpreted as <code>y + (log(x))</code>.</p>
+<section id="type-signatures-and-result-type-inference" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-signatures-and-result-type-inference">Type signatures and result type inference</h3>
+<p>Each function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as</p>
+<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> log(<span class="dt">real</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and the signature for the <code>lmultiply</code> function is</p>
+<p><code>real lmultiply(real, real);</code></p>
+<p>A function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.</p>
+<p><code>real mean(array [] real);</code></p>
+<p><code>real mean(vector);</code></p>
+<p>The first applies to a one-dimensional array of real values and the second to a vector.</p>
+<p>The identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.</p>
+</section>
+<section id="constants" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constants">Constants</h3>
+<p>Constants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants <span class="math inline">\(\pi\)</span> and <span class="math inline">\(e\)</span> are represented as nullary functions named <code>pi()</code> and <code>e()</code>. See the <a href="https://mc-stan.org/docs/functions-reference/real-valued_basic_functions.html#built-in-constants">Stan Functions Reference built-in constants section</a> for a list of built-in constants.</p>
+</section>
+<section id="type-promotion-and-function-resolution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-promotion-and-function-resolution">Type promotion and function resolution</h3>
+<p>Because of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.</p>
+<p>For example, consider a situation in which the following two function signatures have been registered for <code>foo</code>.</p>
+<div class="sourceCode" id="cb44"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb44-1"><a href="#cb44-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span>, <span class="dt">real</span>);</span>
+<span id="cb44-2"><a href="#cb44-2" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> foo(<span class="dt">int</span>, <span class="dt">int</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The use of <code>foo</code> in the expression <code>foo(1.0, 1.0)</code> resolves to <code>foo(real, real)</code>, and thus the expression <code>foo(1.0, 1.0)</code> itself is assigned a type of <code>real</code>.</p>
+<p>Because integers may be promoted to real values, the expression <code>foo(1, 1)</code> could potentially match either <code>foo(real, real)</code> or <code>foo(int, int)</code>. The former requires two type promotions and the latter requires none, so <code>foo(1, 1)</code> is resolved to function <code>foo(int, int)</code> and is thus assigned the type <code>int</code>.</p>
+<p>The expression <code>foo(1, 1.0)</code> has argument types <code>(int, real)</code> and thus does not explicitly match either function signature. By promoting the integer expression <code>1</code> to type <code>real</code>, it is able to match <code>foo(real, real)</code>, and hence the type of the function expression <code>foo(1, 1.0)</code> is <code>real</code>.</p>
+<p>In some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named <code>bar</code> with the following signatures.</p>
+<div class="sourceCode" id="cb45"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb45-1"><a href="#cb45-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> bar(<span class="dt">real</span>, <span class="dt">int</span>);</span>
+<span id="cb45-2"><a href="#cb45-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> bar(<span class="dt">int</span>, <span class="dt">real</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With these signatures, the expression <code>bar(1.0, 1)</code> and <code>bar(1, 1.0)</code> resolve to the first and second of the above functions, respectively. The expression <code>bar(1.0, 1.0)</code> is illegal because real values may not be demoted to integers. The expression <code>bar(1, 1)</code> is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name <code>bar</code> is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with <code>bar(1, 1)</code> given the two declarations above, the Stan compiler will flag the expression as illegal.</p>
+</section>
+<section id="random-number-generating-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="random-number-generating-functions">Random-number generating functions</h3>
+<p>For most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix <code>_rng</code>. For example, a univariate normal random number can be generated by <code>normal_rng(0, 1)</code>; only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.</p>
+<section id="random-number-generators-locations" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="random-number-generators-locations">Random-number generators locations</h4>
+<p>The use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in <code>_rng</code>.</p>
+<p>This allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.</p>
+</section>
+<section id="posterior-predictive-checking" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="posterior-predictive-checking">Posterior predictive checking</h4>
+<p>Posterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of <span class="citation" data-cites="GelmanEtAl:2013">(<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">Gelman et al. 2013</a>)</span> for more information on posterior predictive checks.</p>
+</section>
+</section>
+</section>
+<section id="type-inference" class="level2">
+<h2 class="anchored" data-anchor-id="type-inference">Type inference</h2>
+<p>Stan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.</p>
+<section id="implementation-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="implementation-types">Implementation types</h3>
+<p>The primitive implementation types for Stan are</p>
+<pre><code>int, real, complex, vector, row_vector,  matrix, complex_vector,
+complex_row_vector, complex_matrix</code></pre>
+<p>Every basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.</p>
+<section id="primitive-type-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="primitive-type-table">Primitive Type Table</h5>
+<p><em>The table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;">type</th>
+<th style="text-align: left;">primitive type</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><code>int</code></td>
+<td style="text-align: left;"><code>int</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>real</code></td>
+<td style="text-align: left;"><code>real</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>vector</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>simplex</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>unit_vector</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>sum_to_zero_vector</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>ordered</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>positive_ordered</code></td>
+<td style="text-align: left;"><code>vector</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>row_vector</code></td>
+<td style="text-align: left;"><code>row_vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>cov_matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>corr_matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>cholesky_factor_cov</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>cholesky_factor_corr</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>column_stochastic_matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>row_stochastic_matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>sum_to_zero_matrix</code></td>
+<td style="text-align: left;"><code>matrix</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_vector</code></td>
+<td style="text-align: left;"><code>complex_vector</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex_row_vector</code></td>
+<td style="text-align: left;"><code>complex_row_vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_matrix</code></td>
+<td style="text-align: left;"><code>complex_matrix</code></td>
+</tr>
+</tbody>
+</table>
+<p>A full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, <code>array [] real</code> has an array dimensionality of 1, <code>int</code> an array dimensionality of 0, and <code>array [,,] int</code> an array dimensionality of 3. The implementation type <code>matrix[ , , ]</code> has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.</p>
+<p>Recall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.</p>
+<div class="sourceCode" id="cb47"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb47-1"><a href="#cb47-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[I, J, K] <span class="dt">matrix</span>[M, N] a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The matrix <code>a</code> is indexed as <code>a[i, j, k, m, n]</code> with the array indices first, followed by the matrix indices, with <code>a[i, j, k]</code> being a matrix and <code>a[i, j, k, m]</code> being a row vector.</p>
+</section>
+</section>
+<section id="type-inference-rules" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-inference-rules">Type inference rules</h3>
+<p>Stan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.</p>
+</section>
+<section id="promotion" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="promotion">Promotion</h3>
+<p>There are two basic promotion rules,</p>
+<ol type="1">
+<li><code>int</code> types may be promoted to <code>real</code>, and</li>
+<li><code>real</code> types may be promoted to <code>complex</code>.</li>
+</ol>
+<p>Plus, promotion is transitive, so that</p>
+<ol start="3" type="1">
+<li>if type <code>U</code> can be promoted to type <code>V</code> and type <code>V</code> can be promoted to type <code>T</code>, then <code>U</code> can be promoted to <code>T</code>.</li>
+</ol>
+<p>The first rule means that expressions of type <code>int</code> may be used anywhere an expression of type <code>real</code> is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.</p>
+<p>The remaining rules have to do with covariant typing rules, which say that a container of type <code>U</code> may be promoted to a container of the same shape of type <code>T</code> if <code>U</code> can be promoted to <code>T</code>. For vector and matrix types, this induces three rules,</p>
+<ol start="4" type="1">
+<li><code>vector</code> may be promoted to <code>complex_vector</code>,</li>
+<li><code>row_vector</code> may be promoted to <code>complex_row_vector</code></li>
+<li><code>matrix</code> may be promoted to <code>complex_matrix</code>.</li>
+</ol>
+<p>For array types, there’s a single rule</p>
+<ol start="7" type="1">
+<li><code>array[...] U</code> may be promoted to <code>array[...] T</code> if <code>U</code> can be promoted to <code>T</code>.</li>
+</ol>
+<p>For example, this means <code>array[,] int</code> may be used where <code>array [,] real</code> or <code>array [,] complex</code> is required; as another example, <code>array[] real</code> may be used anywhere <code>array[] complex</code> is required.</p>
+<p>Tuples have the natural extension of the above rules, applied to all sub-types at once</p>
+<ol start="8" type="1">
+<li>A <code>tuple(U1, ..., UN)</code> may be promoted to a <code>tuple(T1, ..., TN)</code> if every <code>Un</code> can be promoted to <code>Tn</code> for <code>n</code> in <code>1:N</code></li>
+</ol>
+<section id="literals" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="literals">Literals</h4>
+<p>An integer literal expression such as <code>42</code> is of type <code>int</code>. Real literals such as <code>42.0</code> are of type <code>real</code>. Imaginary literals such as <code>-17i</code> are of type <code>complex</code>. the expression <code>7 - 2i</code> acts like a complex literal, but technically it combines a real literal <code>7</code> and an imaginary literal <code>2i</code> through subtraction.</p>
+</section>
+<section id="variables" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="variables">Variables</h4>
+<p>The type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is <code>int</code>.</p>
+<p>There is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+</section>
+<section id="indexing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="indexing">Indexing</h4>
+<p>If <code>x</code> is an expression of total dimensionality greater than or equal to <span class="math inline">\(N\)</span>, then the type of expression <code>e[i1, i2, ..., iN]</code> is the same as that of <code>e[i1][i2]...[iN]</code>, so it suffices to define the type of a singly-indexed function. Suppose <code>e</code> is an expression and <code>i</code> is an expression of primitive type <code>int</code>. Then</p>
+<ul>
+<li>if <code>e</code> is an expression of type <code>array[i1, i2, ..., iN] T</code> and <code>k,</code> <code>i1</code>, …, <code>iN</code> are expressions of type <code>int</code>, then <code>e[k]</code> is an expression of type <code>array[i2, ..., iN] T</code>,</li>
+<li>if <code>e</code> is an expression of type <code>array[i] T</code> with <code>i</code> and <code>k</code> expressions of type <code>int</code>, then <code>e[k]</code> is of type <code>T</code>,</li>
+<li>if <code>e</code> has implementation type <code>vector</code> or <code>row_vector</code>, dimensionality 0, then <code>e[i]</code> has implementation type <code>real</code>,</li>
+<li>if <code>e</code> has implementation type <code>matrix</code>, then <code>e[i]</code> has type <code>row_vector</code>,</li>
+<li>if <code>e</code> has implementation type <code>complex_vector</code> or <code>complex_row_vector</code> and <code>i</code> is an expression of type <code>int</code>, then <code>e[i]</code> is an expression of type <code>complex</code>, and</li>
+<li>if <code>e</code> has implementation type <code>complex_matrix</code>, and <code>i</code> is an expression of type <code>int</code>, then <code>e[i]</code> is an expression of type <code>complex_row_vector</code>.</li>
+</ul>
+</section>
+<section id="function-application" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="function-application">Function application</h4>
+<p>If <code>f</code> is the name of a function and <code>e1,...,eN</code> are expressions for <span class="math inline">\(N \geq 0\)</span>, then <code>f(e1,...,eN)</code> is an expression whose type is determined by the return type in the function signature for <code>f</code> given <code>e1</code> through <code>eN</code>. Recall that a function signature is a declaration of the argument types and the result type.</p>
+<p>In looking up functions, binary operators like <code>real * real</code> are defined as <code>operator*(real, real)</code> in the documentation and index.</p>
+<p>In matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type <code>int</code> may be promoted to type <code>real</code> or <code>complex</code> if necessary (see the subsection on type promotion in <a href="#function-application">the function application section</a>, a <code>real</code> argument will be promoted to <code>complex</code> if necessary, a <code>vector</code> will be promoted to <code>complex_vector</code> if necessary, and so on.</p>
+<p>In general, matrix operations return the lowest inferable type. For example, <code>row_vector * vector</code> returns a value of type <code>real</code>, which is declared in the function documentation and index as <code>real operator*(row_vector, vector)</code>.</p>
+</section>
+</section>
+</section>
+<section id="higher-order-functions" class="level2">
+<h2 class="anchored" data-anchor-id="higher-order-functions">Higher-order functions</h2>
+<p>There are several expression constructions in Stan that act as higher-order functions.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>The higher-order functions and the signature of their argument functions are listed in the following pair of tables.</p>
+<section id="higher-order-functions-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="higher-order-functions-table">Higher-order Functions Table</h5>
+<p><em>Higher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.</em></p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 22%">
+<col style="width: 34%">
+<col style="width: 28%">
+<col style="width: 14%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: right;">function</th>
+<th style="text-align: center;">parameter or data args</th>
+<th style="text-align: center;">data args</th>
+<th style="text-align: center;">return type</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: right;"><code>algebra_solver</code></td>
+<td style="text-align: center;"><code>vector, vector</code></td>
+<td style="text-align: center;"><code>array [] real, array [] real</code></td>
+<td style="text-align: center;"><code>vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;"><code>algebra_solver_newton</code></td>
+<td style="text-align: center;"><code>vector, vector</code></td>
+<td style="text-align: center;"><code>array [] real, array [] real</code></td>
+<td style="text-align: center;"><code>vector</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;"><code>integrate_1d</code>,</td>
+<td style="text-align: center;"><code>real, real, array [] real</code></td>
+<td style="text-align: center;"><code>array [] real, array [] real</code></td>
+<td style="text-align: center;"><code>real</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;"><code>integrate_ode_X</code>,</td>
+<td style="text-align: center;"><code>real, array [] real, array [] real</code></td>
+<td style="text-align: center;"><code>array [] real, array [] real</code></td>
+<td style="text-align: center;"><code>array [] real</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;"><code>map_rect</code></td>
+<td style="text-align: center;"><code>vector, vector</code></td>
+<td style="text-align: center;"><code>array [] real, array [] real</code></td>
+<td style="text-align: center;"><code>vector</code></td>
+</tr>
+</tbody>
+</table>
+<p>For example, the <code>integrate_ode_rk45</code> function can be used to integrate differential equations in Stan:</p>
+<div class="sourceCode" id="cb48"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb48-1"><a href="#cb48-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb48-2"><a href="#cb48-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span> [] <span class="dt">real</span> foo(<span class="dt">real</span> t,</span>
+<span id="cb48-3"><a href="#cb48-3" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span> [] <span class="dt">real</span> y,</span>
+<span id="cb48-4"><a href="#cb48-4" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span> [] <span class="dt">real</span> theta,</span>
+<span id="cb48-5"><a href="#cb48-5" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span> [] <span class="dt">real</span> x_r,</span>
+<span id="cb48-6"><a href="#cb48-6" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span> [] <span class="dt">real</span> x_i) {</span>
+<span id="cb48-7"><a href="#cb48-7" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ...</span></span>
+<span id="cb48-8"><a href="#cb48-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb48-9"><a href="#cb48-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb48-10"><a href="#cb48-10" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb48-11"><a href="#cb48-11" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;</span>
+<span id="cb48-12"><a href="#cb48-12" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">real</span> y0;</span>
+<span id="cb48-13"><a href="#cb48-13" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> t0;</span>
+<span id="cb48-14"><a href="#cb48-14" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[T] <span class="dt">real</span> ts;</span>
+<span id="cb48-15"><a href="#cb48-15" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">1</span>] <span class="dt">real</span> theta;</span>
+<span id="cb48-16"><a href="#cb48-16" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">real</span> x_r;</span>
+<span id="cb48-17"><a href="#cb48-17" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">int</span> x_i;</span>
+<span id="cb48-18"><a href="#cb48-18" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb48-19"><a href="#cb48-19" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[T, <span class="dv">2</span>] <span class="dt">real</span> y_hat = <span class="kw">integrate_ode_rk45</span>(foo, y0, t0,</span>
+<span id="cb48-20"><a href="#cb48-20" aria-hidden="true" tabindex="-1"></a>                                              ts, theta, x_r, x_i);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function argument is <code>foo</code>, the name of the user-defined function; as shown in the <a href="#higher-order-functions">higher-order functions table</a>, <code>integrate_ode_rk45</code> takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.</p>
+</section>
+<section id="variadic-higher-order-functions-table" class="level5 unnumbered">
+<h5 class="unnumbered anchored" data-anchor-id="variadic-higher-order-functions-table">Variadic Higher-order Functions Table</h5>
+<p><em>Variadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: right;">function</th>
+<th style="text-align: center;">restricted args</th>
+<th style="text-align: center;">return type</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: right;"><code>solve_X</code></td>
+<td style="text-align: center;"><code>vector</code></td>
+<td style="text-align: center;"><code>vector</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: right;"><code>ode_X</code>,</td>
+<td style="text-align: center;"><code>vector, real, array [] real</code></td>
+<td style="text-align: center;"><code>vector[]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: right;"><code>reduce_sum</code></td>
+<td style="text-align: center;"><code>array[] T, T1, T2</code></td>
+<td style="text-align: center;"><code>real</code></td>
+</tr>
+</tbody>
+</table>
+<p><code>T</code>, <code>T1</code>, and <code>T2</code> can be any Stan type.</p>
+<p>For example, the <code>ode_rk45</code> function can be used to integrate differential equations in Stan:</p>
+<div class="sourceCode" id="cb49"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb49-1"><a href="#cb49-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb49-2"><a href="#cb49-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> foo(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">real</span> theta, <span class="dt">vector</span> beta,</span>
+<span id="cb49-3"><a href="#cb49-3" aria-hidden="true" tabindex="-1"></a>            <span class="dt">array</span> [] <span class="dt">real</span> x_i, <span class="dt">int</span> index) {</span>
+<span id="cb49-4"><a href="#cb49-4" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ...</span></span>
+<span id="cb49-5"><a href="#cb49-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb49-6"><a href="#cb49-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb49-7"><a href="#cb49-7" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb49-8"><a href="#cb49-8" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;</span>
+<span id="cb49-9"><a href="#cb49-9" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">2</span>] y0;</span>
+<span id="cb49-10"><a href="#cb49-10" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> t0;</span>
+<span id="cb49-11"><a href="#cb49-11" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[T] <span class="dt">real</span> ts;</span>
+<span id="cb49-12"><a href="#cb49-12" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> theta;</span>
+<span id="cb49-13"><a href="#cb49-13" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">7</span>] beta;</span>
+<span id="cb49-14"><a href="#cb49-14" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">10</span>] <span class="dt">int</span> x_i;</span>
+<span id="cb49-15"><a href="#cb49-15" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> index;</span>
+<span id="cb49-16"><a href="#cb49-16" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb49-17"><a href="#cb49-17" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">2</span>] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,</span>
+<span id="cb49-18"><a href="#cb49-18" aria-hidden="true" tabindex="-1"></a>                              beta, x_i, index);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function argument is <code>foo</code>, the name of the user-defined function. As shown in the <a href="#variadic-higher-order-functions-table">variadic higher-order functions table</a>, <code>ode_rk45</code> takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of <code>foo</code> and returns an array of vectors.</p>
+</section>
+<section id="functions-passed-by-reference" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="functions-passed-by-reference">Functions passed by reference</h3>
+<p>The function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.</p>
+</section>
+<section id="data-restricted-arguments" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-restricted-arguments">Data-restricted arguments</h3>
+<p>Some of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.</p>
+<p>For user-defined functions the qualifier <code>data</code> may be prepended to the type to restrict the argument to data-only variables.</p>
+</section>
+</section>
+<section id="chain-rule-and-derivatives" class="level2">
+<h2 class="anchored" data-anchor-id="chain-rule-and-derivatives">Chain rule and derivatives</h2>
+<p>Derivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.</p>
+<section id="errors-due-to-chain-rule" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="errors-due-to-chain-rule">Errors due to chain rule</h3>
+<p>Unlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.</p>
+<div class="sourceCode" id="cb50"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb50-1"><a href="#cb50-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb50-2"><a href="#cb50-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb50-3"><a href="#cb50-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb50-4"><a href="#cb50-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb50-5"><a href="#cb50-5" aria-hidden="true" tabindex="-1"></a>  x ~ normal(sqrt(x - x), <span class="dv">1</span>);</span>
+<span id="cb50-6"><a href="#cb50-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Algebraically, <a href="../reference-manual/statements.html#distribution-statements.section">the distribution statement</a> in the model could be reduced to</p>
+<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a>  x ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and it would seem the model should produce unit normal draws for <code>x</code>. But rather than canceling, the expression <code>sqrt(x -   x)</code> causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,</p>
+<p><span class="math display">\[
+\begin{array}{rcl}
+\frac{d}{dx} \sqrt{x - x}
+&amp; = &amp;
+\frac{1}{2 \sqrt{x - x}} \times \frac{d}{dx} (x - x)
+\\[4pt]
+&amp; = &amp;
+\frac{1}{0} \times (1 - 1)
+\\[4pt]
+&amp; = &amp;
+\infty \times 0
+\\[4pt]
+&amp; = &amp; \mathrm{NaN}.
+\end{array}
+\]</span></p>
+<p>Rather than the <span class="math inline">\(x - x\)</span> canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.</p>
+<p>The only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like <code>sqrt(x - x)</code> for which the chain rule produces not-a-number values.</p>
+</section>
+<section id="diagnosing-problems-with-derivatives" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="diagnosing-problems-with-derivatives">Diagnosing problems with derivatives</h3>
+<p>The best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).</p>
+<p>For example, compiling the above model to an executable <code>sqrt-x-minus-x</code> in CmdStan, the test can be run as</p>
+<pre><code>&gt; ./sqrt-x-minus-x diagnose test=gradient</code></pre>
+<p>which produces</p>
+<pre><code>...
+TEST GRADIENT MODE
+
+ Log probability=-0.393734
+
+ param idx           value           model     finite diff           error
+         0       -0.887393             nan               0             nan</code></pre>
+<p>Even though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Languages such as C++ and R allow the declaration of a variable of a given name in a narrower scope to hide (take precedence over for evaluation) a variable defined in a containing scope.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Internally, they are implemented as their own expression types because Stan doesn’t have object-level functional types (yet).<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/types.html" class="pagination-link" aria-label="Data Types and Declarations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Data Types and Declarations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/statements.html" class="pagination-link" aria-label="Statements">
+        <span class="nav-page-text">Statements</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/expressions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/img/logo_tm.png b/docs/2_39/reference-manual/img/logo_tm.png
new file mode 100644
index 0000000000000000000000000000000000000000..48c9769c7de2c7c8585b52a5fdee686aaa47c8b5
GIT binary patch
literal 21065
zcmZtt1yr0r@F)zki%W59ad$6Xq`2$iwz#{yJH@TI7I$|RDJ|~qQrz7?e*b&ld%yRb
zclYdhCdo)9n<tsfOd^#OBtId2L4tyU`XntSrUC^8(ERU*0QXUYYqKl-@c=lhNQyvJ
zj}sn#6qHOgrOo8!q3Azg1SlW?6AJo2l#ee|xFr<K|G-dC)E`eMDCj)E|830!{y(h%
z%{=J;3;%~Wr3x+c!9>dvr0Jq5FUMzWZ_8|GVsB*1>|yKh9|F`j555o3*3`w2)Wg=s
z&Y90cfc$?j_&(tOs#(ZM{|CjzT7X<rUWruH-pQ1di<y;~m0S>sl$7+FlZhFhikQUz
zCjTf2kXyL8IPkHsxVyVEyK^wxJDIbv@$&Mru(GqTvon2QFgbhLxfpsd**R1EFDCyF
zA2CyBV<$@o7fX9P(*O7x8ri$L2#}NiC+Pot{#QR;EY1F}Bs=H-t=5NvEdRBzuraf;
z{2$&Qq~HFl<x_OBH2skLAAdo%Z~ue&|El}na=x+rC;b1fGyhA||Ec{@RS@YL%l{cR
zL8Q<6Tkud&!cfv;-$5RL(`<MTLh<^C3iBkVRK~x&`-D<(jIzH0ay&dqGyJU7;XB0c
z=l+m$k8iE;lK-IkG0p;ZP=4bF7n&?B?t%%-Wm8SH5>uQPSKp5~flJFNT%#xBC+hwi
zejWTXGkUMbOUoH&M^;cWVgU%~0CHiN2q+9v;QvP<4;%0c1OYK)T!i&n6SRDCks9A3
zLfu!YAY$*`l5297u1ONMcf2q`MRD1T;o>h!KilKYUbSSem2e#W-KMr5GOmf`!kP^#
z*{PJMX-Rc;JBaIQ>v((U-jl6dt0|rkBh`eFg1|+)gv?z_e1}}%M!WnP{FlCJ7tVHy
zioa?Y+NeM<;I>GLwDHN7E8a$750ZY<n^c(2sW36-NIh>geF>K|Fnr6$V>I2MmneEa
zDB8pG9&nYY^eI6i0NbYMwl(HdPfs)5r4x(Urh(TfV~n+U4(T<MF@6>dhefX{zot`a
z;i>&qGf{35+kP9T4(<+GdmvQ8Y|&w&;M&S`Ktfy`9U%XTKwfG9T9{3{x}LCVs1+Z}
zei@}kCed4fu$M1Gb`gSLzsTTk!zbcGH_vL8%73jdFv0RkAU@W0KJN512vcD$*CF`O
z)2c8xh<@)lG*nrQX?*aM%g?j$XSVYCHetkhdmd8$)o@}VWJ(CtO4SIg5raA}sdZ{E
zP;2r%czBTdGDz-2pZLIvy7n|=<QeUitdTFHK7R82nM(7cNu#KoGUfU|Zysd7BqNM1
zT~6oB;_ve%A*!#Yq-oUXM@rX&(Me$mP=@I%cP?EC_PN{-Qq`O4+KR`<T`Gd@V0k<K
zk|KtV=&&0Xvv3$MW!^InKjN|e-m45ARuKX`sg?D_nIv31zPx~(X<o|LQDkK@!7Abm
z_$GAeNS9H<L5PH+n4Z$_Jcm1)g?oR!ahd4vk3@!xEP+K}&NbgIWR_$rJ%$%^H3nU^
z)j#kFDa-sd;Ox0VSAO~wHMEEwYZU6Wel(_GGE)M^hws(l|0&9#x3)rJuPAs{E2y;e
zyM}ibN_1)?`AO6A0FcaLc#a|y5O%_2*`+Y?zCn?;PmG+$(+)C)IT$LbN4=EHauq+8
z^G3_?Ggt>MkXUU7OO-u|vfV5ojJU}m5t1YVYtTL~U;IhDYy>jhbYZ5$6oH|4@TM!?
z?=DjhX+~eRe4Bjq^=J-6=~2{}834K$n}i9&Zq)dvBWdqOZ$tIC3Zbw+py(F!@>5F+
z{B2h8bxfyNhS=r^P`QP}lKvI%0V=jt91!w7UB2k>%GvuIEJu&K4jL%cwNh$xe+)_o
zMCp164~Z*BZ#~c7$)L*FEa~S~KV`8L1-d=V{(#>}reEF3!aYmIvsO`)LG&pVY{iwn
z_f-<pLh@I>cXNLxl%}Mn@~d^ESPvs38@Vr_`?#IBm#uow3D^?#@<Eah5wH2bE;tz;
zW%TC4W;cjRckT;}<SDw!Ssgy%Z3q!BOVacc(6P-M6xvsb_sb*7PDw+`OLfFu%Db4x
z_wVK|s8XuHpacubsA0q+W8sX*&(p&wg<EI*7Fp_?UE)aUuzHt$!zqqU?Xzj_a~BK*
zd~DrcsmM2J0>y3w({m7yv5Zc+ztsi=O*}*@H;+g`hQp@sgXXJ5-_>+x4SK4#Wv!Qa
zne=Xn$n>@Xg@FnY^26ZVB7!1<vmVBhzoJ}xm8<}jS|n&UjhB&)Zktj*!R;DvtB3jw
zw<yHTSPY`1aA7BG9%}RYl*Ig<)V*jt_dlxiDcejRMC*x(WF-4puX}|-G;ZqsdNj)x
zF=}u{ppZ0}0sPZhMvt?5ynR6_<*v>UYEdqgT6W=^UlApObgf>Boc?vKh?re}6+I&Y
zFku{~M4lh~+Ts0Mc=oBNhD+ELsf3a4K*aU_-bXvO#<H2V=cm<ymyJg<Zk?b!5}1KF
zKOa8+R3Zw!+vgEyi(>*Id@pZAI!$}&$Mo~Prj6eq&CK<uBbbiHnc>KWxEI=vE0_Uh
z6fBM<4d|VsXky(g)e@f{2^!F^7dQvkL-W|*Ui5JMOEaizvGZibhvi8H!j{a;8YKO_
zFIR>B>bVuANg}EL=HDFrK|n)TI#acz_Nm{5F*S-D_??WEIa>|whok<}<X)q3wQjF8
zWLWHK3Jp1DRT*+_-J}#?{WD&Ljf@mL3!R%oR~4&ZuE9L<w=P^bq_#W}+a_S`@uXdT
z|GX+iov1Q=7Am@6Br?DX0UQ(DeEIs|+SB8){$&oK`1^M$$kUP~!+q7)R6=P_^AY9N
ztmf7`EONbHRzM?68x?9B!}hS~9PRod)w#ZM5RXc&pUWtgHTT52>X0DsG6JHXu0q<!
z`ijnK2tsk$4+^E&-2D90L3uiFl_CROS6n19=(dhyK=Zg{@1@&~;YZU`L~OlvfUp^1
zZ?YeX0TljMxB~y!_^CepXL+ByRh%ytVOC<4u*aqv-#YBp0=6Y~N)E_aXGdgd2{4od
z0EWe}^895WbJHEv7T+oklZ%`gvrZ;UR$ILt`TY-3LDL2cIIB<MpF5RizGKvM_KItP
z6!J!7aFEnA%OyylO$w%I`}bRpcN(09)I3#1nThEZ_1lHzOnh5diZ3dVcrsJl?toDW
zWf{WQXWZYkzFnEKr3c2ney*#JINpqTw--_G_vdIHAd`(W)I^}uM9Xd`_kS*}VPMQ9
zp}uE83puFGKy+zQDJf9lq-~O?HrDxR(!nn1?Dr$XGeqWNuG;;H{sHHn%)3!}4D)hc
zA_g<7*c;-0q<J%tlQOnh34p@SVYP^a=0{cV6W{J#tqE2ylb|Pu_12broj7_)Ye~H<
zR)^2${q)IK?eoWI7@+`opodh3N*FAPy%y~1!Y->rPF6L#Mu<PRNkzXbPS{uJq+N(s
z95%2pJcRr?@UkkNM?he3bqcFhQH(8tn~}NAzzO&NeSNU9*xO)f2o#}!jYiMWYw<j4
zp;^3}Hj6iZ*f{P;V8;YZe7qjcn)OP*0+LiwBzDjP6w2yA6Gjh?Vg7WVUD5_kcR`Pi
zj+%Oymq>S!w>o}xg|lsmZqA=7p0yi(NV8_UWM>4*xPGCZE216aDi+zE7I%bpa<goP
zWk(>M`XFg~d~bWi|28jOQE$?kQi|Y@AoYNZtzLd1nH_2eGM+p%AVf}@9j0n!XJc^p
zO>Z39Hw#hJcv`#(7g~MM-cojJ$bR-9Go|@3nGc-uF=f<Rl?LBB2|0Tc)jZp>9sM+3
z0y@*w=(77n#Iu6*#@kFhS^P)L_fC95msBprdd2geoj~~@$s#dge7KjT)|1s|)|~HR
zKfhuWgWtQ&<X@`i7X?}YTbqA62FnH^<de6HLZ7N;%1V`njKt_Mo>;b%E?7%tUT`wm
z`f=$P$$zcn(&}cql`=l{=6m%jGs}nx>jNXH;b!Vvt?7=iL6zJ5RNOH5mC^2E1{id9
zx)_dloBCRxH*zM47_FeP6!OGifHae)#bJlLpv?d;hQNNV=)a9T0={n6_Dg+Rx{C~p
z`$Ba1FRq*?bhg`<Ta$*fVUrjU+&L&2+bZR+0F{EixBWsw8`KR4JDqqQhws5oDIwjv
zwcVmQdTByuBS;U;oj3)eaT%18Ujon6%MnBbFwzRG`lk#CiX5A*ODI488#2$<sQ47=
zo5pT~bFwkNuehu~-d%37x9mE1*Cliq1@+%dALM%nJz1m1TjZvZ%<_}5n1SuYb9p(_
z$ry1?I7sY&c6Vj@e+SfadPu@)nOs7n$G&lhSAukRgpl+~`~FM`{rUP@T`gtKkwwD4
z+ll3umN$*uuK*EjsdE^;Gmh0$vkizwrNao!fe8T;1s^||qoMd&^yExWCXgT!Cf>Zf
zwEB6@ZHZA<VxMqWbE7@|${?b(4kVT6M=4n$^I0L!dKCw~FF^yw4<`D2Z<E#eomyt>
zMiEP1yZa;{%<o04*HS6h7j<S=xRXaUvUY9adDwN>x4MlWoF7LlEt?p6#P8EKlVc-)
z{#YWzqZsCc4c0C^hs|mD;s`BwrMD*OX5rouBlMKTJ3$ycri|lH;e(&fgix#W+o;y*
z_5OtYOK(~ZU~c2|fA6YwGF?UcmrMbF1SN{08ub;Zw^L?Rk<8&(kqp>9CB$ukqW>lD
zNR%4A^hX?*iCbz0=`M{y<A}YIj>_8#9G}V`XmbA}VA0HrPw**MU~_l0$DyhKI}Aqb
zB<RiAe%2`fU*7(RJGtcKAu-t+|E=DOOZp^={<rcV<=jsmX&kI|vNtba!~g|KwSC#D
zN|3xk_1r(}`JmTxw5ttT+E9Wfq@3Ne%m9y<+z@CX=kR!y(@gcnf1ACJV~EgK>$674
zoMm8=$J)*O7nQg4{xb1hN@shR0Toc){Cnqjo?lVNul?WfUF~TPbnyUiT#?t}j<45O
z9jKOjdd^69WlaPuYYuYv0Nd+=6rE9i-ocaiRS1exI27f~73e%qT)i6i?_m}Zx5GLy
z6Q_<U(p{(`Qi%<-vyUtRw$&l-HDkr1+ml>YX$CiTT18@Twz-nauV~>illbxJX}hKj
z<yyrmX1S;4b!EpabxYSZn+zYkxR?Fy9e>m_I=0vs5a&`j8SYc`!>=b^n-@&$h`oz)
zuQ3k{!c!L?@!`1&K{ZqAI<&bU!+@xiKxq#5Y<gvhb|v#B$`ZjJ=x38a1bNL`|ChYY
z?5g4yD<&wa&;Hb;zf}g)k6sK>>bLy_6>=klH(?fQFrOxTnJ%1?nZDm5-PNISZ`770
zcUjWJ0Zh+fe#)FXRcW)n><-f}pD0q-f<_NvMir>DGn2EgPY-EkZ}jH=TL(jp1qJLj
zR!q!V-NF+y5D~0h$xs1|_}_KPO51Y#q`5%)rdiT6SoO|Fv&+nL^o!;TRUEzswtULd
ze=G|{n$`tqH5L|+R>@EXFx6$mNYh~)R(=MBoIE|oFge0B=-VHwoVxU1;_j%cRIj@k
zJ6uPaR+t+j^hQepVt)T+_@Ds6Op1)!mdU+B&-Z)-vEQiI&@(Mc3a&c6|H9l)dN>!8
z7Kg!fh>JjWL;#b0f-Wu9kiw=6=&!S}^O;|@wJOZ#gN=Tx7HwAJjsC*dW>sa~<XSL&
z7128v9R{t-ml0G+JC-qg9|r~RmbLOJX|9<d=qOp!vBQ|4dR8S{_szAVxn?**yqGTv
z0t8im?Jhs<MsYv<d=V&T?u)zP*_E+ePLF)2SaD^*@(qr=3{i=4m>quys4ED!>7(mN
z6;z`{4%8;={oKLD=6%L*m7u|ibT{>JWU`~1-`6?O<zz1_P70zL3V=cB<qhoD!Y2`a
zXiy;sW>ECzEkTOt{%Y|STgB_)2o)|MBNrmY@mR^~1k68&Y*wQuX=wZ~u0~|6*A$%u
zfQm<97Goby+Ig)U)>H;H_qVi%B#U%SyN5by6$W55zhQ#^YG!m^0r5<yPym7*#pI{N
z*!N{h(S%OOcL`0IPA$n&k6dc)h~e$t$z=wbW!GWst#IMcDg`2b$h0$3^qd8Vn|f<o
zlh={buU7-99AvU&P2j;_q2boW_tI@}HSa^kfK}e$!1O-_@>Rh@*;Yu*fSdfdgp(7T
zaZi0(Zh4flAR?yXx1AYkh5XcDhx15zU`=N4COIKy9%3aaR0;j8FxYM+l4&8I^c^LI
zmCRev=O{^Dm5VT?yQIn>`TR+1393cWoYW2dV>2R;5f+$P8X7bouR#>=5H*7kTf_^J
z?Umc4MPJSp&nVcrR@umd_%M{10znS=Ibw$)R0KEJnA}%FzDgGl5N+X(s6-VQGc*_v
zM+`cR;s=$sWwL;0KZ6yY0<a-1CF&D8<Z7-W$t^pg#+G%)Nt23iwFtV^ar&>6hD@@e
zDJDQ36YN0Qy~dvNL)QfH*UR@7@QN*Bg;X2;%;2lT&)-<rAE0QO%-mN1YQIk54|mv(
zPv4usB>$N*spN<J(ZQ+x!)sGJj^Xng*<-?D-34MW5dk{Ds@Bu{Qn1}K5eT`tQsF5e
zxnd<2?Et*46!4xcI?PEL;o?Y%9^tM{^iDJy1}6^@tZ(1^_3G*)#Y%vO=*sllA?SX*
zpj0Daa$#2DGXkrdtm<CsA$#(!8U{qUb|g*q`RKv9Q8ZA)ciZ*XF*R_ND`no;^Ha=+
zL8WU>;rND9eStK5VWEV5auvBRV6ZNeP`}&Rkp{kDzyL&Jez<f5v3N>O#kNcZ4fcj$
zquxc*YV-&iI==lX>rUD+l9zEp*NostiSPg>e>i|HGJCfkPM#cP!oPx6gPPI{HR7Fa
zpQ#TVw|H)FHTixdTcfeM1Ry*}fI%aX$rU~w1NW;@Zz^d$x2ApfW452<afR>QWdbp!
zqTxK;1felVGym%W0E;$uBK0?WsxewCB^oU4m*`$An`eGH4YXa~Z;Wq<=Rc)~=Ok^C
zvk|gHdSf|GahRY?nFxn779)mh9yZwef0URcUPc??o?N96sB~A&Ei}g8(#yK*v7QtK
zr4+N$haiOT0%+)`kK6*L%|O2%_~v7fa5fmU0XE~g1aYz>j}O@1?<F+`4%_>MSH3Ev
zL4&+i+}ey;VQ?k~3>ZCXHpe272<RUc!Vy0p7P$gMtGd7+P9YD4%R>CZgbGg7?*Yo{
ztv2lT#?f9?_?pQ3^-RQynr7l)=ZYZ{20qE6C)D{xkAOdn3@PeK6A4)m?Id;V`KW0M
zv1swKA##jN79guZ#q~ASAW0QZQ(B-#eIUq|ts!l>GAr3yaovxo&8E>HZAx5BI0+`C
zuW~@rbAjhDF4fWzIv9>a@cuoy-T&<{&wymvhSc?lh-G`Q=cXb+pE+@j@o8_^S3=g-
z)H^AZuAtM~QdTqPOSHFzKn{a9cPJHuRdO@J<_<66P~FtDSy<dX%|_hLuBxr%#Y%#O
z(H*0&<aX2w?ud5BiHrkEg@;e3cw*P;CWVIN50kF)LK3N{UW5q8X^I~-L(*~im^is3
zQ+zl{%20aq77N~B*GFg9hR|cl;=wl(eFkeO(Ixr93WLA){Xg3^{l7>rWXT8*xci!7
z6v`th4vK7m<npjc&IjeXJUUsdOqh|IU}&-?Go^0nc)yH}&9y&NjqF`Vr_4>i^gm~w
zFRM!ot5x$Ksw@N4nUg`WKp3wU8S&8k2B}v|J7XF#d7{0Cnkp#&yL~VBg+JwBhLxiC
z63)`Iywa`oMKgZ)(i1d%D~m6G^Von!(`DFIN`@&>OV1Ra);qpiR+c#wq^yOrkNg6g
zKEn!iwr>W#6#5i4zctLRyr#tgT0QG+)(Y3d5>3j<HihAR*dUZX($UJ1?LIP>*N!fs
zQAHcKyrNt6k}M<7O0lsCV?kAiAwv{pSi1q|WheqibI}rsPPYT$vG5AQ56Mw#3X)Uo
zOyWJ$g%4V^(`OY~z#8#Ps>a4PMolg3O2!_eI>x(eE8pxTxccOhF}noCBi>*D@wX(U
zVacyfvPat9{!YbM;YMspaq5<8I=k$L6?+o8g8Y5>*^kqw`?EJGEj8Emnzg8tLNdcm
zEVJvmJub=jo^&)A2jryG*#nwowKAdN^nx0jtz}B}$(=Od!M%n{y6m`ZoSv&U`g>ke
z<yuK?;mg*jvq7|;LM#sa7%9XaH|6ME3Q`|QYu0$_h5uMYUxYSnCYz8uG$t?s#_a*W
zfluB@nvl24!k|!{XwM+I)2IidKVCl`bT+Y}N}f~N%`TLGMyV+!CsQcRKz7bv?^79a
z03Iyy+CcSM`Kk*CQs88h5Z0gHp|Bhc{%;Do<k7<SrAp#IN^0CN8G-D-$dP}cRN1!l
zlGLSt_&kL<=-g{`(v(~V79;$euouO{P`ik@vAP|^g!RaSu|XlzFd_UP-q;_hU)cy*
zaLzZqEDf8Irfp_H-TFfwRzbYaU70MER~pp;jEf<tJN-qvuj2kku9p%4on%G2ip2%5
z_Ie6WB9Ta@^xMf%t;IYEL6M9qocFbB5H0^COWxtuQ3GCRmD<*S>C`PRRQ}7Z(GpPU
zz@r~CNJTZ@X(sO}{3gW<kf14*79jqC;ty5eSFWFDdTnl#XgE_tgf}#VU_4+%C~OV0
z87UWMQIQ|8TG9u_VtTFR#sNCOVZ8)fA0{M-Fpc(y=cR@-+T#p+kL@q~h&Kv44gdK2
z&VQYdB3*A4<sppA2$mNa*#WB(k|CM6E&eJG70Gv6rqEgv<AUs-{+7&_vJrb&ou%K*
zzt|}MhjMzVfZ)|H{XmP9=rUS(!Q)W6N5#G2!4F3k&oWEcl>gqlhq=dZu18OcXL|%m
za_TR*?Mn%s(PhaE^g}MMpQ~sD)q>mYsl`tu*~Sj`AB}m`%V&^H86jqQ?#p5Vr@JTZ
z3Ikk5^E)9PUOHF<!;TUELyPT$*yyaGRe8b4pFT2Ty*j`tz6DlD)B3;j;r+Cr!SqNx
zHyq*fH!S@4Ua(i{`@&3$0OIC6iBbr`E>ZP9DRoYW4mK0n>EFoM2Z_9TXxvA)!N045
znTlwgnL<a;SPl_k@GlVGBTjb(^IC_2vY~^_dzb;IDT-1tw4L&j0<Qqkho1#vo)d9S
zPZD6U_OE+Q$NOh@d-S<Xnn|c<WP3-*M)#hBeL-!aYq^Dm0*Ddb3NYNfF;O7RUwtA1
zVH=?%*i_``D8RbJYr)EWGlQcmEeMxOedw;h=6{rU+liVl4F$V_0qIkCQ7-<cfAN1k
zA44ry60^@3lB2@_<;_y}tvLlhVT@|&m&VC7>G0Xa7@<ALwCng0O?%&pz#X!!p-MbR
zURw0@ob4(U!C!R2P)JNAsf27`(_42($ZyWdBI=xvUh?U-DdL_iC7mps&{FMvvFOR=
zfFjp!i$eDazfd-yoqgR-ATj7O68LFh3C2V8(!c{?eE8%ub~37<7&=o%X<)7EY6-fS
z(yu(J9i&6r%sA68*+|Uq^pKixgGF}=g{nfXeEcUr-(ordJK%~{B_OFSbw9tpebL;H
zlQLxG8+B(q@LJt4a91hIx#6W6`C7{#Uabxtn4vD%kH_0d9NhK1=GyOH`W1I(VgRpG
z;QDP}3uT;O_PSL)F2Yfz?#ofQRX`UHnJW}c;Q&3|$GJckF-F=gM$hs3C3wAY;+QOm
zhwca!br);r6G+ihg-v^Y30fd~?zdyUl58gLOTe>j4yb%>NU>=OE-?`SMub#Qo2CyB
zDXmH{X#uHuZrJ&m@dx{j%w|CXjVRV#hIA;+AjNyNRV+)u+JhC9I~~p)W+XNv@P}Dz
z?@D{8dhOgP;T$xTKO#cz0&IU5al&OwMnH}8wx|pQO>MGVHgKCOctoX`rvzYJ4D5>#
z_C$^e*&CtqxqXy6PYLSUx$PVE{egP5jIv4nlOIO7a$Uu^yz)HY-YpsOt0u0mT3`@t
zXARv&hg45J-n+m!`B{y=^^0MmoS|Z`4g70R%n}yYCaspik6RfC^&oHc9#6veze2!1
zyDp0&uSyMg37L(6iDwc&b);N>cl`1Z1_$H^)fKar^sR>${yeMm`*==ITod_pK;nAE
z%HC|y0RaYrD6k)^XrxZ-igppT!#}~KFTTP+J|hv4m1n9(*0iIvZVu%mKnb*NCJBOu
zj|&>xUK1`KTLqXJ*oKyqbcddlnTzx|stQjh6r@xcmy*h3Iyi)<Cq1be^m*UPLTLW{
z^~Dj}7r<&Y0a;Wo47a0tG(e>sH3vKFIw0~4LRIAcU0#AK&fOOs9G@Vw#?)@)@ez)i
zS4?;L`491{tISTPrC4sy%jJm1mxe^u(bh4CT+gmD*6AygAwXwiL?AY$7kn&dUvw^E
zND#~AR8K+uP+^bmi@akbD|mS+UeF83ifDd8ixG%G9&jn<A?sqk>fXY1c2KqCt6hty
zAZ2i4`jbNHD4$!o$@9!e7q_C(smg}6|JoX@cezhM$+Hs^ZzY>6uN;a`R)>S|(M2HH
z|DE?H*A3Ss<|j-{ABa{sBML+TKuP*LE+3<PLDUQ1^>?8p2sYeT(Du&)fyl(Z(I+TR
z(QmnEyK>IqIB$2ChMG;LLdm#9;@L;<JN~3j2;w?dhj68<ube1ZYy{I`J4sltMDRh$
z)pH1dGjO!hY%jJ=iC%`pCH%{=<JSF`xqE|j#n76^dP058E-bgxZz1o#M_Nf%c6I>;
zL2-FNrseM(6T!P=2V7%PGf%T^$y%x~%QgXZT3(92Z!ejnKAPv^lUnXD1GT;46sy^n
zubOHNIK&1`X!ABt#9yqfd{{^$G+e&ULm=`b!g4ZvDCvHvMW_T=1IuOf<mlgku0)lJ
zYl|I`5MG@N-R`b_g<$J$w-sLK&o5^-1~q(~@9p*sTYpnF`q4U2^O~%LZi^(k2hH!o
z134zb2Ts0!Nk|`{1g>K_Xv@*@`?*&SsR6sRYg>BMd8|n2G|nyEN9eXuACgX56E-%G
zWef+)?Hz2I7IVo#uJYDuuqOu3s0iZx7W0Hsf2KVe3Dl;j3`OKgsUYmIrnEPdk~K(N
zY*Tdz8Q$DZ0yu<!hv2sgf)Ee8F)U6S`>*^@1Mh_%;g-L!4fv;1<y+kQ-f&T(eF?-p
zwTY1LpTkRz2>6X?GT6|z2=k6zxkjBWi>2GW0f=@G6}5j)U3M_MK7X&nQun)o)#O|D
zt&ua<_?n6%x^qbp|IGpaCf5sBySddLm5Mx14Y<l%!9C^O1a?dvt(ZgVQbV`gFPP%A
zKHKcpSu=C&&D~f-JFi?v&1M?&H9%M3om6b8JBc`^3PXKT<V8+|ANfjwuaH&BR8BK#
zXI|)yRj&cNV&k~&oYm5-X+I}6-j)SHF;e>LC1ZrxTX;?s($m!d%a=(ZSJxg$9)^l)
z=RvZ~Uavfr6By-&U5!%cJ^qQFuSf(qW_IApTD!pSN#=#Vc}LQSIvpPC>jwxX<AFP7
zwq4ReyBF&EJS-dA&M!X{A3!IF0X-}i>StWiDngBkO`c~4Ohm61$Ln(Sj9tpkB4%3T
z`0(Ni56=;?PYewVr1I}+FWCg@<Se2Z_`m7>QP=fM)z&j-tQxC6&`as)cA;Auy`1UA
zPs?K_lm0?}K>=8J`YZOV^y1Y^gdK-;H|qC`j-O!wnu*FDQTO0C=kd8tEF?GQmGbJh
zR#)ghvfKzsoj{r9t8jaVx;~48@F=c8>Va>Q1^xMy@t={g>ijmaZLp171(-fUW&F&a
zxxapEc%PtPuF6K$y!wOt@Pv#4j}fxTus&SY<g<Dh(7ydzou)}djcxn%$1#s1y>T8p
zky{RN;ESHDnMS*HjwJH@;jRdS>Q!QIP(BJuHQ;l=xvch9xlm=;6BdrXzWh=r+2oWG
zvSf>WNIgr7$Z|v-X&9Q33P)#K@)bH@5aB@h>Ogu^8KRx#(`tknss+I(!2B1-vsCIX
z$P{cKbO+EJ%Ql>sx&1n1*dTlxs8^ka&-JGlkqzTIpt(_5kwiR^(rsLTS@EV6%wHu;
zys5uVWp2VD)lU}G<L7}-{JbK$2~tjGV9Cr|vSCu&Bh{A_T7jVxmH58=2QhxupH=*7
z+9u9Q(IJ}X$un`eFvzx{&g4t*ZOG~DPPBxB$9BKy#o{#y(0#h~w#+5ydd#p%<aDd6
zTEj&J>8z0z(EtmAPeecnC{2x^nzS<}Ueo<Xuh{@yK%Sb8_K76Ece8^;MybtDY<Ldd
zPR8ui#>%11i|E~qpYqcexL1dk(4z1CmP4>q`H~Jxw(7(`u$Ec{b8c&^ztJzcRm@0L
zv_5rPhY=8Ly$jUAr7(m+3X14~1&Uu}qjgz~U9zm-Yf&1l3KLb-sAv!iZGmnyQVk?4
z360Qsl8YiONN!sV$8xQ|?}m6CZMpTO59OuKtK7MFa2M_H9BH*UQ%)%Wbqv)P6*w#y
ze`Rfia=1iVz4M8_M<T#T3ES64`=1gc%~xSIEy`O0=J7Dn7DyNZuP7wR3#zte|J=3H
zF9;fKblrYah_!`tR(%K!M=5!m7SVAT-QuDS)IoN0;Uj6TQAZv_s)}hUUPg-`(hkb6
z(vyo(JAFkhzw<?tUJ@mB?)yquLG&GBCbGhmQ7J$-yjOlZc-YsWR?FtKj}F!H`}A-@
z^2S|<=CM6&@KXT8X$F04yFU4%+`o`Re0W3lDqQ5~@lOgA+j5C30tu`1nnW{(|A?`S
zAPFGRCj0`Uuar!YN1p%~Hvbm!bn>=4CUlj%Lh3nx|MqsK+lvGPC4M)oa&A=pJOV4k
z#WtKFMa~){E)nB7h|wq3TDGiNIm$=)zK=&M={~SUn6`f`d^EsB{Yrv^pR3xnt7W3+
zHX5YWIU17*xd@bn_QekG7#y?7eMwO6e2UCq+U5OaS3^=eafRo|P;&$LY5H-`td8R~
z?OG^QZIjK|dCMWnE&$vp&wgX!4JRXFgst6V$z8abeJ|wp{JXWhsU}y_mpu|dcyNav
z`@wq9UW^gM^RJfJx;?D`iu_QVmnW?03}0aN>p<cYS}1^HBJjZvyIJ(u_e<%^$J$0?
zvZ_-{C<XD7&0@p<>5V8xlXP6$@~4j~_u|{)^OlaqR5u#YE$7xWqg6NwOqRmGyRwa{
zAXliuJtjfn`OE_F>UM>t%{aW-R<fFnc(?c5v4h8TIN0v2E}bSTL&o+}EFf@Y3aSNZ
z{%@O(?MaypA8K`}{0>X^mqfUTQ%_oPLI7c56$>vyhkAEZ#Q@j#fVuYoS;Pfg$C~VN
zRnT7Vs#&Z<fG?MUjWN5~7Eblp)6=Nj=f1=M1XfjFFwM!+0#fPXBNY$PU9Alb=Me~)
z#dgl5>B5^w7Nq*<`T9ee58rJah~Hp=h;O{A!04riYdXses5=%}TF^1v?_lXz?X)o9
zI=w5_NOaB(LvZ*PrXmcJ2#*>pt?`1C7~Z;M#g&(MXvBn3+uYQSOMmiuz=ZzYj5d@8
z@Kt8*NUlW`1`}Gx!wA8y0%eMg)$DDLmHu3Xb&v|K#i?9g4Mw|y9z1cxtvA&n7XMjh
zOFV$>U$g33ECG0C%?y==5@IcXbio$S?vl)4l@w9WLQFy@P7J8aj*nQMbJ#^k;d4eU
zipxOvMDJ@H#N4s_q0)j|8JiH4@+=5Wgwet>?@eZ~hmg#RAw|+~gFmIvq^d_hz`+YX
za;NM3sgf1d+gAn=f?TaR+tg}X(*jpUiDgT$^s&$reoE}>BB<7A|GVM>H#&waob)jn
zt3E0(zOGwW=dojVDy(!dc;b^gRDd0OQ^OFj&+IYWK^_EhT~AypjYtjZ&X-2f_w_Z+
zee0a|Q{=Fm?jVwSw(+$u#dbK13Qkkhs=iPu$1Z6F*-mp+)e%p8D5RovR|t*jX~v{K
z7xJy%6;KA{>e8Fs;Pb{^tdrZ-Y9AJl)CsJsrnD2e>B&Ba<v0W6I^GoUzgnZhdd{ot
z|D~+xHw!IX;qD7`Ww4}5my-+DjL|8a4Gwah=IbH7le;g!`AvHubSszcHpY(6>&`Y%
zZ0;r`pGy%CfS<W`{boI>>&XbzCLosIO?)o#Yq8Ugkv+hW2egPT-5QB&F8{f#PIFMI
zS?@tUin=e1PFtZ#P>PaqWf1ONo<e;V^22*l6oiHw|CZaSfDL9ePe7${CQulw3L2c$
z8QQDN8b*o<h{$819)QdK3GEtw-!|VqgG(LAHa9TGgR4__ENf9A0HAPUaf8L44q3&Z
zrR0N4?yzoNAq&E72+omTokF@~>zdllK*)R<F@S`~)&!%}U6&>FNrbfOVApiFZR4N3
zZjv;zq;D6!*5u7mx6S?Ja~G%HUn}jeia2w0kBu!04xnQOsKCXAR7!J@adARtx$$di
zPg!PD=gm!hXP4dTmB&AJFsMyW{u-emlhWow&#VN*<@c_4>CeO|N!xLY5I%C451g_+
z37V-XIVYngJ-mtkI3^#PT3Ux?h<ZEX8?scPBRTfaDtF=g3pE}><NKIZTdU<^x+?vA
z!vL%6@1M!sYhzRNk9z*<4OFjuzCM#FFfJ#bRCfj*#h!#eQNTl%*URabq|$p4P2{r>
zMaMIcg#*ul;dG*i5)6vN@28#rW}>f3h4BP6z_u-pS6)Kd>GRz?@!yg8>bUa?aF7<}
zm-3r%<2#8FB0h14XLmrQREDFt@X*4rx*@Z4vS@_+y9vM@tfFEdlo`dWuG)S0hrS<@
z!!kff3`G7xEm&V@wJ~oDw-<C(XFJf@<OHn-E4h4_3Ou{xh7WDX(tqIsTT18)Q1Ckj
ze>mYh3eYd|K-*Ws;+~f&@l^`|Q5LSLKH+bBMRH#J7>X5U(ry|vm2|Mi3cvdzhABE0
zxK>@<)IXskCc`CmFWN}F{7vmt8S)!nlS=>VpP5h>-4z(h7*Epe*{HNNBR8$a+d=rr
zJ|^h88f>al^itjj=#PEKaPsrMwd4V7A~udqQ^`LMP^|<Jf0X|ZbFcRJ*l;vJ-Kod#
z1L^9tHLI}S(?*SUm!T`l$oDKVM>I-x$H=Tyl!hkFe~<9EMk&7@0$huqca<UAy^ntP
zX_He9)TBma5mL|FRLR!P;$_zSM`4BGmOrb9`QHv=<<V2pVIr4;)tD_kB8Re1@=9<$
zLAAEH)Z4#2+hb!Tp#ah~;kytDU6%beHw%&IW>g$otk|{bX6~a7z`&%!)IUzv!>i2R
z2xMhi*l^*7vB>sm->XtE8Ylmq)b9IMjy4Q2&M+4&-q(n<bad<A5sF6m`S~}jZGL43
zeziaO^E{4eSkL+@ET<WyUR759-g%V|7Y2<DTUC};$!#?*_F}dF5nw)a5z=ZFHJ|7=
z3}Lt6@j_3r#h}x{S;ZxH`+f__<{{=+gPjiPL7?Uk`d60H)lO@0rQ)Y!3_sx2o*s)8
zcm^|2T}ICCPeH%@R2a_6B@!3PwAL_V4vUCiP+)kKB0JSVstwUDJ-`>x<X%lOzY^;!
zgk@lsiVUI$)%FYM$qhvcy94_sxX7&Yjydtu9Nh|NT%ltZxQH1<2BY=;8*yaZb))>6
zfAVh)oYhRaTE`aLig|50O3Y6Z1ncA2WDyUX;L|OxBTVzg!4xFT<o9^U^UmP-NO+J(
zu&vBnq40@amk_u8buF;kQ1~Np*Y_zdksZcJzO%F?(Oa!HZ^vL;>M^tnM|eLXfE#8Y
z$9M@p+n{qmOiooRk^x=bc~Dtf{NE>Gp(_N%xb|(|cwWDD5{((Eb#$=8!><;fF8ak~
z+%>aRzJFWmNO{g;y&Ax<()3&(+f)4gM#QwE`9fEQ)Du1jL>8(?ga{(mkL19i>zl+Q
ze$BVTp?3dsHHdmPxZ`vC%+!poyUC4~0?;TetPx{Jm!$+6k1IJX@i{FfVC6_5-~kIO
zi|(&tUO%MgylPgz3t=EMtqU}%vH_txkhRA?%sPPynR$`EuMuUgfw88sbva?#b&Rh0
zHK+I1cf7}ULt2{$|Gfr!_2U>@Z~{ZYQga?q*EK6z5)p9n@ej63Lx;EP2e2uM%dJrB
zE|RYHjE=Q(qEgi$YOZK^j>~@TqKU^|yZ=|~`|<V)44Hysuq*FZ=}J~%+6NnP>?HeM
zE4p{l;n0O5a@Wpg$y#7Cm?sPi9c`)jSY;*zsyC0yYZPHeC||vxhB)=`CiyWCI00L4
zoZxnkF}Kr3Z38EC0Zde3bJZH=$b4LWliTPN_vBp~a9w|Fx6G+3#BXNuQTZzeka;>^
z3<e}gN=-%Nt}Xv@SRX$%TPmoxt<=b_%*UjqXQKkCg}ZjqWDw=v%=HS=w21*UI+Hha
zX8;F^!jT9T!NQWhKSm~+QLA*Eop-gH$!ni(j?c+ifIsb5EQng{u(5QBo1+fPs}+&e
zP2^(D;H~%`KI`F4@nb0Q2d2-Wpw|R8j6W~cGp&<2_9MJX{Giw??;Xp)L=JyesV>m2
z!LHQbUj&jS1gYf}mP+j*x{x&3)>H(w7rgR(cOyM%;V_pAs^gdlP}swe)cqt8nXhod
z?^z5U$#gm2aSGnJ)g#;^1s+kOXQm=<b3Tl=tu=FOgd5`aD}IvC;DzNYw0KdA<gfoV
z_87>$=T!_YWIuAxtVkrn<u5#xvM*I3zQsED_e4eRAGz1w#1ldq6cJ72-jpLZ#pVDB
zayASfI(=wd@0<Kvlh(Vm1b!GMK<v!e#e`kExDP6rHWTd26rn@In4||_kQQ%QBrr%t
zEyl?Ww9XlMJp_(`J8Zu;YB$k#wVRH|><wwGb4^qbwRF{k#hsH1J-=IW#0ROlZzQ<@
zFeg9D3BeZBxi+@pa?BDw0YWr_)TepP<wNSQe)qQ5G(PU<xs9{Ig(<^z$tJolL6!5n
z)_7&6U10<|kb)y3gDWFBz*ji$04(ON^(NTpW#bAnv*P}hcUM3q%SRM_R$y`Q@7tm6
zl0F&+_^X%O)CLlt4x_N4XJ@hOV_$4Q!~+#De1Ui!TE6wNFhgW)aB!$E7Ka-kM$>zG
zEDfP|y#7oTQ@T_Xz0&?|=tTZeU#GTjpMe7R&O?=Uj|{jaBRqYmmya{V+>U9c>z-yU
zi)*ZV5IlDH2~P8|N29mb53x3|Qe)7r=SI_QQfH4o<iZPOK1C~Wy&r0o13M`Pi5mEY
zoHPu5z>Q}q$*iq%`RJ8D|55C-b`zwD_=f~V=j)$6hu;r2wLDPGd4}@RpMjX>6+U!J
zG_c;SQLn*=TP%<+Umr`<Jt`oCrk4^?B|OftORV>EzQrJV(AxH#s);aAQc@BE|L#sT
zDzZmE-E1tJ0}a49ecg`YIv+9+mtK#9ylI}}u;9(kH(occ59T$9k=7&f*r>bQE~BM%
z^r$o9il~SX`rPcMu=;8KR;)c=8psdn+j|=j9Y>R<p1+yOZu{GveE5og5?B0dsE^u!
zKCe@Z6ch0PMdF4p9jIE}2`jOu2Xs9AS#N|d$ODaJ(7A=Z*WQCG)ix~iINOIQU;8>s
z;p=~G86k|lVAm`VEZTxi(rn+H9E`BJT8l4;f9%zi#e@3cZwZmm<;0b0|LRWu?R^s%
z!Ik}Vt*3)6Rpw%A0FzPq($syG<+r|122}1fWx;KncfV?#K`uQ1xV&*ApV#yYlfL_z
z5XaVW<7dZSL@XDif5d6?-=Jy-k-%deu_)a~`Vl<8eEbcVVhHfU(r^g*(({%hj%zT?
zf6FpC!tgxE+cN83AKf*ou=J4<a3Db{NI~$Mu@jiVFp~F@shD3}-npW6Pco4sh^hYV
zD@pzIYDY*F1NGifPS!WcDRI}Km7Xj4wy2C(mD;)N7wIe#3Rnik4P`nE8B#WHbV_}v
zY=&OB!+lt=-61c3N>&ES#A+HmUJgee1yv>%*f?@%#zh1Uc#7*WpKJ8;h1{X)$g3`|
zY7m9(4J5OG4j^}+2h@qkc@Tn5UI+_4m&o|2aZ{Oj35eUh=+jZU;OiJ@VLhU@QL~Uh
zzwM{I?3nB=Vm4ZQBIg#2UhBbn747;Ai}ObRkbSG-udo2*lSu)ZWLzKqW>)Zt2fm#E
z@ZEWhG9r%cX7u16g1lW@Ao)&y41TlijzBkq23e3l5-y94<Avnq5A4c3ssHgx^qK!{
z6NReIOoy?Yn(1ObB|SUUTovZf(TBdKFwR`+*#ao1O!(F6U3K|euixF^uTS4*uSb3)
zTg3HN+63$~L$%uDitZihd^YKN`aQC+ik~sLs#&tKdF-n9_}UB>2WUCkxZ1+6E~?>o
zeROzJAaJTRb7V{vUry@&E#r(g)Oz0B)TGPJocw-&7GMA3ZjIN?f8>)CM7`x&NT%0r
zKw9p<OFkabc|u;=dO`^hUg%fNiuJ14I7dUkW0xCrLGMro>e`iX)a^ucp3aYjt_=U8
zO4u-0ele1dZaRpP68YNs@y0@4@}c;8;P(}984JwTZZJxEneBK+TxmJs6e{&o!=ZWJ
zN&*-)m8<mkzd-pd`(qr)kfX1|use?J3{re%Bm{sFl<M3P=y^T@)zP6U=gKi%=oU)=
zhDFgIYz*5sM2eaxZ{CzJ%Lwf`2&D<VQiR5O?5U8?uZ~>2o92VLI<yjcLiCuh(+|fJ
zpxswWH|A1X7eNw>GQfMT-(~MoHZ&I4IWK7P_9M)M5YrFN@&A?vQN6~`;uO~3qsRC4
z4ifc5rqiH0PaE}UW_oYA&Va6>03)j-Os<-}e}aA$Il2<=2(DhB2N>q(EagI2hJXko
zYJVI5)g~6JE(HnucI-?z#mApQ_MUq7(+I7@@;|))rs#kw^oYShb?0#QRu@X)9@2p{
z{`rBe>K|Wu@^)Vnum_KX+I{WXHS4}u2z-R;MG%q_gkltE1czi%8UZc>7znU|7<4Gi
zebN=zZ~%O7G#t_j59>)zhE0R6O3Y9LLuCmMZPY#bKYz?1QBbV^nKm)w6K6YRp$0-q
z(d01!7f|R}>d0Xn>C-F^KkS<wT0NNpaP%5!=|oAj;nnS2i)^08Ufg)C;>Ut;z6k9#
z1!=IM-@iH`NPP6X3BA`g5j^D(fE)&(%-C&^_Oab)Hwqh0Ingj^drTCf3VYo#xF0Um
z74N#D(_}nRYP#Gi#UBxPKqD9kr#84y)+gU20$eWo<7!uU@^tfku2SP&tL#kLoYf`&
zcic|#Uu2#9oI7|RoAdO-n^!jF_59)A0#^<1v>2py?}U<`!2<@4YP)C-{I6aeOAb=e
zQj>8N2mJQRNw?U|1MB1`46O9uPwf;BOm=)~s{gRiF3P!Hv%UV-EuJMi`rNye9r9zu
zCj@w$<Z^^`RL<p>g6C2*MxoA$NcYN3dJtqAex<x?pYfTrfI|aXI<;y~9R!3F?F`Gl
zk&}LeN|C{QXGPiE*kCiRq>nr>-7D7=3S4Z)4Q7<IYKog`n7~n<ZtJ~iPXYHNlv?&B
zCQY)q!6RL2w&o3K|8{^InjSL^z)(J>e=?x2$%;#&9%(Nmh5PiW5oKq<4|&q^nwonL
zF+{jTG;z%f=Ze+{5*1|cCGcL6D3MMjs7emZr(|XPyy*if-~VJpO0Hcd5J?u;bN)|B
zv)oO(U$on=uJ*la<?1)X)E99^O)?8(f*#naMUp>JY6UB$qumGy3-tM*>o6!3DW%ca
z16h>Z@#u}=d1FHTB}DA7NgAg>(`N0H5g-HPdM;|+bD=x&eeF^`{a>3GpD4^nZb$Nw
z4N5gRc7>*8aWQBW>Y~q`FzO;(<pROvcl{cJ{YZ8J=tq=Rt;i;I$BqPCyTDiM3edfB
zn^x<a!3hc}cEx;h`GEji<ZaC>O7qS<%S|KR>a*4hEVmB0EU}6HjXTb-PG^ec&WDQN
zyQw5D0HarX^d&c1MomuJk`1vY;2sTV^SgHmwFFlu&-dr$#ABvE%b2Q5j+8R8y36;>
zPj;*ad9(kN>_p_B_2iijqI+v8;mpx?9mb*H@clI#q%|wg&*0ICfhKif)dW+rXc?n-
zIw`5bj>zg?!G}Uu8YGp$vMF{bDI|h01JkpY_^nSUF?C+l%vw0=73xX&<SSZ(TNE)n
zPrf*G>YO*mZiF`e?w0GGG9v}8unb=uTJ_zXV<4V};XjOTKGx1&0RUF@JKph|`4*Qm
zxreJt^0SXNFMWajDbKhe4k|3T(nN-yjMNAf)O2i*5XSSp4lFKe=VLpPB$=6oZOMS&
zRi4kMQaB~^s+_MNwnxA4D<EG?DRQUcIGWdn9$---4$MD6YH>9ml4V%$ay_%PQylhT
zrNym`f7JZ^TodBbT0AInX=HjPDvv|+HWj+_UddSZnKXRsKmn)G-xF5)=V_m|0Iy|J
zPil}6Lj+%5fg^m>T^RKWGyrG2fAD)(ON}4X`|l`fkAC<^T5EjtgSP`AX!G}(Nvlt3
zlrI|GdByFY)+#NeE1tYL1ya}Qj+$5Tf~hQ(a4IckIcu;9Y3=ungXV<GX}-e1ZFb{w
zTuk{(;3}vWS8%6C3AvPiWbqPs0kD&uN-b_nVA`p%VJ_lc7vV>o!iSrPVz_M)meIeb
zv7%O7<9S7%7CMCKiS_O1X_oD9ObX%P9Lf2-EL>IgbPz+T;=%CrbBa?c0%8?at7Iy-
zY=+Wq%f;5gWxBBOc{;TP(>m*;{`U?7x^#l7)6@XR9Tdo7nJNe1mnnN%zdtMYJ2G68
z`tgK}s{v7)25fMfit8qc-`jAp=XEmuhHBeK#%Cxg|5)~Wf@xGY0!^pGEe}-{I=Y>z
zQA3YqDX3(Vfr_5;eIjZbGoiA(eDgvJwioT7!K%b{J?aWoV(Y_};5BU4!-{EFsG%_F
ziLX$%X)mjn_j+{+Jm~O3Ku9ZdJU9`;VFDSJMJ>I@udyN-VT`Ev2!}qw(yQyDi!xMV
zn-E16G;9~~zjwKuZHOqlOZl@bbY{HEdm@h_i1(K%d?d0g9wz(Iv=rl}TE#$o1*=GK
zV&6AFKp(xoTTt|B?4YLg!m~Za=s;hR)l`1AMdKR1=@65Myo#$~xV7yA3v(@+s*58L
z%$s|Uoj4?%lrp7KZ3L-PG|IYiEBi6&(5G9(Re$MJz4O<EkZMF+Ko*w6SC=DZq!cHd
zQ<O4T2AI~j+o0!I&a09EbD3c*^tjhTRmIr+sJ~?x4~ry6Ol2unu0-Q6{?WKc3l?Q(
z)xWD^l>Vwj5dyU?UlRN`_|EV1Z5;o6%$t{{nt$i~>%bimL~`9@++k-a@_x>@)!PFh
zK>zbkk*C!HLQ9OhbJljHkDYmEM<_Uv+Ur6iwz*waAu^0z)T+PsK9v#iaHE{w&&n2e
zD99qX8b+h9RD{@dgU?rXfispq`H1pHUjA#OHR#e{x#dFeANtn2n;M7DR;B-^@Yvk>
zw<M=q4AVuE1=-D$#wC_F)`4L)D7CSgvm=~GG*BbNp#N7|@FIF0|Ixt5wT$f;AqAz|
z{>X9%`7vBnZz~1**>hGHFw>9XFl&wHwQ4`d>D&z+zI#PKJkp)ea^>0rd5tcW3tLl$
zTxR3vzWP$y#_Pet?hEnk7$h~Gmx^)<(;_<vDZMwF$FRbU^qSh3p5gl+t`rQ1bet68
zPePU2Vcs@uzoJ&e)Z%uVa0P|d|BPr?$w6l_z@LxLZVk-hXQg8fyC4Z(#eF1C`leff
zF!p*%1N#+xb~k>+y}=*zEIQbvU}79Z3&#Zhv>t7SwNkDY(Vu&jtMN3QDTq=(#8P{C
zy_qXS%E5!EgoM%)KMwL(xuic8B8BYXvx>*$pZ%76TRV`sz2+$o3Mf0Mr3k1Ct<m@O
zUhlB7N<YG`3~=7DhOHs8-R7=9OrEYs@CRXxz3-}k;6596Mrc-MVa7ch`Sx7MSDh+l
zFsG25!8{E%psjas-a9olsj7*>=sEChAK8z&6z!@t_4QU8Z{0D!C=VL<Xom4ICGD_-
za8ut(jhK2Oezv0xDc{c(s0_MCU6I(iJ6o&-Cv`SajmI8~GK4-at64&jl<|5xJ>%#4
zqWDkIxzL96&Q$dPs(p^1%N_m+1tGZ#X_q`ZzJTPErYL%zNl1u3Wd$e?5s`i_`M2Tw
zmeVVn)B`g|uQ@eap}+1l7U{OXqD>TsbEevdr&gJc2~in%Q{75kW-ZE9`lW|kUaL+A
zRBOGh_gYE9*z3G~<!AQz|J8Ax(Qr6X8&`uBMDHac#Ol$@#_B@!)uToiB^HZBThVI>
z5+$tOgJ311too1MiC&_Y)oXUWoA>kkex7^g%$#%YoSA3t^E0ldsjxr+Tuj@u?hMt)
zi5w&9!Sc_$RoFBApX$fym#L(6dNEfpiO{1A-uvOUqnF&Me^w4gv;J68B=~&1U!9KJ
ztnhKkuT-lrt?iI)*`0!a1zOrUWsbT9p|P`y?CEYYLl-d6A)|D`9%TL3u@T_JP0<Sv
z?}5}m7C>r2NsEY&<MYH#HAlmfE3{|!@x&ZRa$LV1uI3h)?4g^WZCgy?1kM(1AcGbA
zf==}MstrmUk81&PLKm5xxx6d+dp!s`m7Cae4DP!yi?o{sG2|9aO?<h5fad*3<E5(t
z2W-T%@=473wG2vikQ}d+C^Z|pF?g=<traEp(L*{3!F$hNi(A1n%4>Cd*4xZNfEfs!
z<-v_N|5eAU@Da=vHqdfP&^hxp&PSuX!fHL#UROoG^q?~M;|qKLP?nn_U5#R(pmDWV
zB6k40SFa0(;y&fndP%Sg!v5@O+``vCI;xHW%3Pof#oFDbKs|oLW`amJEmYiv4W@sy
z6ln1~*ykHlR0P$;8)?XD8j!c5`vfAyi+F)6xlcok>;D?};39P;c{J3cPjMHwi~j!N
z@7j%%_lEe*@EQ49ry)&8@r3w7_5}a|KXs`HW|v~Ou}G~a6Iw6n{fK4`84-RK34n7=
zT2H;IUAsGQP-0CYgn(JBb^|l-b2>@q+>za;zoAvrMg-%TyyKqud_ph$J!i>u>2Pb}
z&GnA&*>jNW7eIcaN6DVNo<(6#nSP`raG|(+dYmt<bDh^}>YujvH2kBBf7zm&d6DD>
zP}FrCP7famM2HHjIV--LL5*U4^He25>)8BwlZuB`wT<7OPbkw=oDf-*tcl&l1Pu1U
z5c;mM<qSc|4j<FN_H>nd-X~(ju#MHh@QYL@`YLcRDxOumK*@Armh9MSyQlEZA>rBM
zs!4Oql%mo`R#+z)pV$s0iq-%mG*Od*6YpXK379Tt?Kg*G^OoK!eK4`CaU~YWx49VX
zB^#@|IS{W8NUb%DWKFktRUFq`;<#-&Cwa`no2T_=`N>>@UZo;uswF=GFZ=9ys)RFU
z5vpiZk!1U^GLTq+>o`~-vHH2I`_Tm47JOPlS;xVddG&+iIG*LRJGa&ole%D>Q5;O9
zKr5=bz+4)k=6?7EIR+K>$RQ&q7SJ~}sY#4o7h^qI|Bg}@aXLs|+i%s+@08S1IOOa?
z)5R*zf|Sl(bHmIuQ<w%quUE$dD;t&=9)`|7F&q07^(dS6+(3{sTT<nnDK}ZD;jznd
z@r#}<A4r+KAnv^6BEZtxHYb}{(6u;D9YAeP+v)tQm`wmC`5xthF9;78;yC-_(cwI>
z<ZCsVLcI)EF366>KaoIR^%rk5x=a4_%TAz4i$<p|Dz3&w1S2tESL@N0{HXPSg?dtT
zpW^XCfy>~erQ643QX*aO6GVvEK2jia1N-=Be9~Ci3}_bKTrJn0H-VBOjE*gRgtpKH
z`ip=BO~9)yOe)2{|4uv~gK_5z;;&XFwPoBj`j{^j_D|Dgjdz)#0698+O?=w-_M&bm
zKW(r``?U}NC05E!xF75gy?j+8_P5O#xdCFW5$v=;#p$-3qz`3IBD+XT_$FMgGQ3;Q
z%d9@Uz>DBN?q>2|kJg6XF%PqKTBG8ITVD5+oXQW5Z+g#x#d5r~+$;lVN3~ZHlHu;t
zf&;g(0Hnm=LPOAA)qu@{!qmP)9@2uV0{PB23C9iDAZ7AhiJ6cUv4+kN?|rOP13lJw
zPdp!aTfcRxA?thOVT7C|w%Xdx#l<?3#mo}OZ4O$6UJ_?nV|U4mV+?Bv1`8Mg;yS5N
z=U~1*Apk!${=e4SWA*7-&!)WAbJR6Sgh8IgZcKA*o*!%zBAHcO=Q`#M7Rm+tJP7yZ
zO@YgIG-VeXd?<OTmZzQrr%zl83e`1IY4jWq3+TWT`)6v=j{{rWe`m4CEFZWRA<M+^
z?8LB}f;wlZ=kC}EmFEqcQk2bASToNVj%uR~2a^`{c34(c`Y(_UWNPehqKhC$ls}FW
zJKgmfW|t-A%}0?$RDbT3xW|UEkZEqO$0=E2Pce?)^~wy|Z2MIK7aDVnpccEQGdoBG
ztEs)HeAq5G!Kx~k1(nnNIKQ1J(U=XOMtchUK>c{XPul4$e!`<?h`7&R{>4A7<p_?a
z>qI}(babltb{noGQm>(!5tH-hWb#|yol{8SvqrnM4w(RRA_UG)6%z&{dTKwrTd>=E
zjC*QtsJZ=ta3}`1`1JHvW(nSe`pz}4mPVX~<D4x<q}y-aW)h|OqN9;p&qDjjZq~TL
zfTm2=)syxOs;&oOdDFglhkDxoTrh`3YYC$(c(*?jVM14Jw{LF#isQ3zUwu|f)~h`H
zPoTVVoPs%}qf2g0FF9%gC!dkv`=WsRwL4L1J_qb{=C!S-*E{=;BAYhzb`hpx+>FHW
z*Rg=$7zqMi(Civd+v{_y(@v*l!N<2LbX8Qok(sD4RZO3;$GZ}$OaojhGZ3W~<d>lc
z$o&`cOlcS>0LqVWfC2pL8&m0Jn<@NcOBXvh=h-$*HT45>5<{Kgt)j^ik|oRfZiM8u
z3cm=ync|Mk*?7qZUS5wXk7z^)gHk3n{uHgI1d5Ct_OfZG=Gvy*+Ke8cnf5ZozrSxl
z_(^IitNKN}b8;+KX9$A|F4~x8RNQ+_bvo^FBZZ3Q_nT!k2h`2TVX1(#V!C3LjYL(&
zK^VUxdl8i?xRA)Oa8=Vv{^d^$>L-T91Mb$%ODyopR4}l7WmhWuG+l}{KE2Ki7)MGJ
z8TP#IwV3)9T5@WhGw~0pnNKG!lut*{eb^}SIU!B1SElr-nyxT(6~acXAA?KT+p2QH
znuQavXQZ;RszY6oE&c9=njT|Lsq|5(q&b%*0$66o{XjHa^$~>7S^8_MJS?~)(f}Vs
zI?7{8#dE@D2D3RV<(xWX`ouAWj*$enC=?s=tAaTQm`p}fS%yNEElOFz{l*+6hguKG
zr`hR0CtDp|_0BdpEar_B?{UNvo;_vM>!HT~iffpjrlYkqcK^0ehxww`Xl<j(jKL+)
zw8(3$mBiNXPJGNOw!12qa9Qe>=%~=`Z6GsgKCI*N+$m(S^m3$-LeX($M~Aw|wiUqx
z{Sr9hkwsmaY2DSaMC&M=759h$MC#DpflwD%u<+Jfy5Qj32`j*na)5Ej%g)m;zZK+V
zErNdhNYq1v_;Mc;w9$g&8;d4}(c6x<3{EWaUPQ1?I=%=cww&k$|Gt{xwd#$w8Fka<
zN4WK|r*m<(9R2<aT`ySqP3zI1|M0)<uKFK5w2--jztVe$xtBPC!EWE5A!jyg)G46>
z83dt$@E&1F`pRQloAxFo?OY#4^`)kCeTkZjc#pj_|BuCAq|lbePL2o!*>#l<(M*3&
zkKbSxFIT;fpt#(uIV`pt6qQ30E41IY5t-*%%v<f~rM0@&DTJy+M?r}I=}air@^%%@
z*D^fTg2rv~X86UifX|mDA9c5c)4Cs&GRr%vZ{al5M_}**8*|<MX^{9|r-9b*d2j!_
zn6T?A3xpA@hi=>#{ve~#(;pR=M#*RUTE@9>(Q}TxXV3C}Ary^*>OqG$e^Yr0UzOwC
ztkB}$9N)YbX#PAB?4~lEZ-&X+9rUKXcOylN^smqvB2pF2=zcOgJ9`*lyBBdv74%3T
z37O<8cP-qYn>kOlq3Fq!ek&*HPirl(*5b4xKxspfBg;5X<$28}w1_;q4U}Tl?alZj
zz3!~_j3z(w-ZKJnr;CG25t-wQ@t=+ixxzy79N14T_KB#fyj?@ia3(@`?byinr-Tkw
zZ6L1?$bRF7?$wo<70X!6=zhDu*XIL$Ea|()`Nr81eMU@~hxA(>>Jy7=En^+R50?3_
zb;IbY(1Y-ho@sgL+)}F~dbP1}e!@5PG_}FyiVaC$s!5`Rjva;k$Fz=Fn}w;j`hNdz
zt%W^m`bk25*k5^R&T?<!C`)fF%x^1d=)k{=F<>bnxsj}<{_-6+b`-Xn0&h-A<s@=3
z{=hk^?EmXzV^}#XxfYwe`PatIshy|Ku#!DPGV7s$!&8HvIO-V5dXvJXYkz3zp$J-N
zh@YZ?8<MjP8w%=X6HOYvCErRR9rS;O^^k`4Vn?fMKuirTFOlf8F%le`CpfIQq{{M6
zycDo;{ViYvWF*sKQ*}x;#jZd?U!$bslPcIoJ*vI%fu`%3*Z%uED;s2R+D|(|b0MIi
z*mRIqrv^fa>savq%~QNS2iP8a6L3CRRyE|FgoGbJl(Q`bevY4E@ojTC`x~KSwun7P
zcQ2E0@|0|*<fm)Fmb9Hop!lzIaeOhmbdEk4GHdnwX43ZMa=VxR-iWg!rh%5nIXVAG
z1{)+f-5HugB)>hn6G=XURvSt{LD;whwBD8ck;h?n+Qf|x=(9t_ZB!3#q62cz|7?x9
zj%}*Fk=x2`%oUyA<r*;2Vdv{LHhM9<MoIE^%6_c%0&NN1rOQn2a&}3qdC*pqp%NLb
z`*p;}7D^pN`LSS1?nO=)B0_ug2RmT!`u9%YP793aKb--2-K=3r%toV_c}DX<K>-_5
zB75~kJbQt}B3B}SGx8yUk?4w6W#rvA)b7<KBbA^Fsi#qfcGQ@B6T-O><u7{8hS`p?
zDJ1yhB@aoNe;jQGSsHL@Y%=?f5xF;t?pb$#9ewJxC#Q#Q`&a0!B12o)jsEl>w2rPu
z^|U$KqjKgiXNEQ8Wzr;;Mbo9Y_~T~s-KTCUqyj1>d0ppW$f0ePrTeknCDhPpdB4=7
z`&ZUm4$|3FTjZRxMyVl#3*W|!_b3#e{64$juk~`D=V#uIpTSjsP$E!4d^p7St&eh!
zhn-===?`A<oGGWC#_f-Pc^^2xFEJy$t7%NbkW23L;T~(ow^pB&8RH%j@XfS_XH*>l
z1IVNC{Gakc3NJezu=#=rWo+QaU+dr8vU>+qo~SSa8jUD=QsKrbB4wKQR<Xs>zehS;
zE*bN`IZ#GH*c@SaIvL-09*HGox%f=0?-Yw<&*i#T<uBKtV>=eh7a+wQ2%Fveg{+zO
zDCm(Sy-kGompv|TS4Lo@<aAfI>AIK}J`Ij2#?|57{%-iB^NB`*N054sC!#wSIw-|7
z<>u(R&B|<WNtvX567Z-pUP^{){@_ThRn0<#x%g`jcZzpANm($KHl~a-_U_Ym{~-~>
zvzFmWPw@nC=AvKaWA7;M0Y1;!MLevY<}}6g8>vE#j0G)GJAX{t4MV#%)q5P6T(q_&
z9{l^Tu0n05{P&Ot43K(%)K%?<2(V=D#ctn93;ycnRb{?*NDz>l{AJDnTp45_*VB(!
ztyFQwJ^{JMjoaOa^u~Kgxl&<e_q%?z(PWRkJevZ=TXa@?m4OfsaT7``JgPTY&2}HR
zG(3>{0(`DO-z+XPMugD`Vk&zjNWPw2A{N_&C!A6%_t50)x*R0vUILz@vNXxBH3#0G
zQJ;MW6V|(!-g8z3v6~%GZi?=DZ&~-U>+<b&*YC0E`WH<68nBczi1l$6A_>l2>ev<X
zH^MMd<t#FPU<W7i{$D)4|DV+Nzaym7;dfNZPC%~4;k{8Dml5x|nyzY%vMv0709UFk
Au>b%7

literal 0
HcmV?d00001

diff --git a/docs/2_39/reference-manual/img/warmup-epochs.png b/docs/2_39/reference-manual/img/warmup-epochs.png
new file mode 100644
index 0000000000000000000000000000000000000000..7ccf22e0546ba69d66eb041549e34ad0e15e159d
GIT binary patch
literal 16881
zcmeHvc{r5q+rKsYQWRww9-$KIQNq|tl7!?bWSvwJ8rxv3P1&MovlK(iQz|i)ond6k
zB%0~5jctZ8YV3otufJ>R`+a^#@f^QD-sATk$NPT$!7=Xpp6kA@^E%JZd7hv1yj(wF
zWhNrHO^}0wL&W^(;gcL3yaWyo&a3<zz!l}!VNCFcs;7y`2`>{f6Sz0r-_{rAe*WM&
z_}TL(%?|C=)z#tP*m2w4&F$pT-8(+QHo3Wd{H(T9&_DQOeEf}*Zc*>6YZZtJH3}aT
zGSi)$dZfjB@EkmQwrvitsIBIItT}YGq2?YzsPyE?#Ji|Thyu|{W5+I-n_C0N+mw0>
zLE~Y01;_0>Wg1*K&Ps90@t=cpzG+e%mT=oIJ}vk9!7T+3g$3?gpVN3QXrYFCE#hAD
z8hbylR8Z(ozmt9MCC5Z%?4zeoV^pq3D)1jWsdmh`Fd*{ad7BHNN3LI0+Vkhh&HAi6
zQ-P8<goV|mrHyH#HcPtVvROCIdWfIpzapNY^sc%GcMpa7Dt4&h&omSaCCbgc@<(*~
z14Zs@R}8uFRl0RgP0ycjm%e{vt2F;OCxmp&p=XQx3Xq(D?@@;U4h{_^_77*sQ==db
z4r31U!w3He;T)j|+>k%N#5b117eiIIdvt=|Ovv=`uAIYHB{_9@-^X+E-Zs&Acf{0E
z_!y6b>C>wpH`?rbtngT2gUvxM&az!4Cz3vS?(}%`R5ElpDtoSiJ{;6ZyxgrjPxQFc
zi1f&s(KW=iJI{{pVO~ihs<-irD;Qtl;N;@@`GfH_o(<X;^HPtkagY7<CQe$Z5Nh@J
z*jKixD#*f}>eSQMz8y`^U-h~lPhQNt;*0fDE^+&HHa~8QCsl0td6xfv!9zc;n(^9{
zuB!rTuEtnrblaM%<UrL5@DpV0{DoB35&{QUq}M+P;D65|QSZ)g)SvCS2E#2)t9j|r
zB^Ac9R~A*VOVgd^B$HTS0e3=@y4$@fL|NcK<Jc!P%%AXtfP{d8DUCfI#LUU}{eL~d
zpYh5|sVX*kh~vUk&0n&Ad(M1fdZW^jXc|6lrbLaYWPHW=5f|&*XCa|2yN}k|yWiYj
z)8U12?hnyZ{g>eP9W-wJg4-1}Q_(GACXYO~LLGKim)M~$>xvEWvxiqs)qHtmf`8zZ
z-~N`j|G7TOq2Pk@HAG-w^lc@x*y|FisRw5Lk0g*=7Ux(^1u{N`Hqu%bj?Z^k1*2w#
zlI(Gu+#90D>4Z2$t3ys0H->-G#En)HL!$(DTT-{;NCb?D3{c9!9b*kzr0qt0bW+W1
zB35s);<y~T;rhSS?28nqgNbY5q5e;AdquK~2YHVP5!~NFLi-<$RWc(qk7sD-&8K7>
z%;vSyp+)<)pHGXid2%FXr|q+7894S0DWY&*B!EG*9ChoCn4#cDO1kBrS!c@|PCgPV
z3QaE_fcI3RU}?@l0>Ko7ozB8=0Uj17*Q)JRaBGl9#qxn<Z_)mMZw<bS<AgXZl72yX
zo%DS$F*R`=`VFl+QFigTkDV7Xd}{JasO48%tv3E!;rlk;+TUk0@0#IwxlDPui0q<B
zy?7zlrCN)Si$QH&(_XF^*8zLF^G?KUZRKwf;$Q#rPm)^~*eYONSCU`7*t(LeD+xgC
zb;DRU3=j{kyV##R=(=I75o@h0$-2i_H;i?|_-+O3V6YAb|C7N$I6qP_cSSf=!@0c6
zR9JeC!h0jju8Gn<Pg}bRYyhb<KYz%Zf&BWyGD~~J`<xF?g5*cmR7#HJ?SSS3-wDGV
z#}t^PQtU{=gPce}^qr7G>354cwmj4D(oXK>?b5AEsKqT$VesWc`7#?swiPyH!|<^&
zFKl!2d--!2oh<gN$Sn+X7&Qpt>;WaIS^G>BdU_~%DV_M!I17W(_escnQuUoeG`8VF
zM%8}xv1`_!vbe&5D(CN;i)?$Tjc_PH-cwSLbv0!uCF%TzU7jDJU02OIS~H$MS~V7<
z8I5Fpi@`08z@6i^O$E!R>H7>Oqy~>7J6jOw5q*q)QC-tB^vzc{jc;=ytEwgnclsyI
z|DYaC1^Sk@rj<C`u$nd5Ow@0;7z?o6j^^{9zu6=KTGW&%p=u?Z0k1-KIAYzyZ|-kB
z>Usd7nrxA!jhL#3&R|Cd%ZWi*c3B;Lo=wCftw$anMd`b`B0D5?%_1b&WDx2x&&2YL
z`p*34$iXzANQjLeSxc`^`3%0jz}j)zSciJ3d#`)#pOrV|pxPApyw&&F52pW@XU`^e
zV~VfwtuR@Z1q<r3XN!h+dEKUvG*rU$E3$oT$xFlG1q%j@41+WZvz<6J!TWaV<EV$n
z9fK?isrZfc72_42x30GBOWyEyI`}QK@SUr6Lr8vHcB&tpak7|p3c`6nF<|GujB>Rz
zQSD2^<pqR-g$SR!K5(1?Z!~?x!7lf=H7Q=yW`2Hx-4(jzcNP0LY4&$$T9^Z7Z!mP}
z)Gz!vkRSWXYc5XOte|h^Ka4#sDV|i~;hdN7ug8Cx!D$7cNt{WxYmfaU3y^TruX0cQ
zP38SPEB1$KL7r&3TD9LljNM@ib~Fo7wCmSo>&H~=QCED!d3F9TP59H;#$kf|gwBHe
z)23_qUNC@~xOftVgRpb$u`gP4(n5rW%0tF`3rmJ0Ltj6FIwTH9rYimS?YH`?p{}xk
zCMq?Aelrj^07W`Y6u!uHr<O5%?8t~>7Ivap#X2KM`>i&YsbJsS{HLa8A9V*azRN~Z
z`PPhw6E(>6vn>wn%=1L!EM{EZCEK+98D{J{!<KeprmmL$(qxk-{6KejK*$UwxQT2@
zsiWV?v2<RVAR53c%JU1*jFcR0WJh7a?V^o4PAZVdc3Ik{M4RsKoBePH3Ts54C|0|a
zHLnbkL1&(ca3k+i8Rf6q^%FJ&;XY<!BByfqgiiLqo@#ZBduoyPtP&lyJO_IpzY&&f
zFnu(6zvmC*-~V4Xels^;U3?*PNA_Z$s>!1Jkj5?Yc|_4r{rqY2{vgc7g~~w-kxuy3
z>BA30IbE-{kgJ8+p7^02s%lYovRk61rS2mSf3w*`&F%ws+Fspub}G%USfjG=)0KIt
zYiqt9ut^Q>e^8(GDiAQQ^-|Y*J#W2e`JY{x{f{U=uP<`FTVlN>WWDcYJ>psKU|Mg}
z`Va3#`yUhxoXD*1{g8<7gyWpMIMfFnbf~`_rToOA{GIw5^J6Nf&$I<LS>oG0veCoj
zmcu$O<WK2RYjRGfZBceENaak>ss_7jsV$I$4mvYS#x0+!NZ}`V*=>?olWZ~$<KJ!n
zTEQuV76a>jf1MVul*F%fq2c-GnrM2_&Y0_K&g`K9Z|ZJ`P5I@6FJIG>ps|(H9cqh<
z3?s4e(TQCpoN{Q#trDd()G#~|%9It=?Yp%nY<l?j*nLM5c`tl^kmB_E+6ECpr6V`^
z9pROTM6>>|;BngWc!{NwS&}fj+wZQbwdj*0HwVMaM34l~Ylm*Wyu3J92Ony62^)!h
z)%AO#WJnK+ujUFTcTG)<5I$N!qGooSo+xX@rb+rsszIZh#J7g~5*x(iWn)F<wiUNu
z$cUWzdh>6_Y(-PStG_d4>{|vrmH>&;_XbbEmZr$e&J4RJt%+OI1y^PXM*g2eX2)o-
z;kGxhhV;F#w%0K8TJdfj&&EHu8<Sehjbu``By|UON3fX{{*5Xd9u&7&KV&d%&Ptz-
zQ-@;88(xTNRDM?U+Du{%TGqX(5-iQ1-}TIZlk1egY~-mFtlxLim;)8GWsGURlzSNq
z+e;m8&v(RHEY2~qnXHLk%d+t!sk0uR?(L*K%eXuk0Es~2d#d)iq!d3fOBM|qEn~b4
zq*ZPfdS8)OzHAp~Pnzw_cv&)=n7Y8k8KG-WJNkSjj{o7_j-HM34fFx6eCH<|%ks&~
zDqmbtaga!zGB5qADEiF<qB56DVNn$mhJag<+%?t}KvE7)A{^|Eif^W^TvZEv_jkkP
zJp~{+i2vf1PoNvIBns{#LfyQQ{8>c1?IB4^?!aZIwy2dQJSU&5_%t<)X^6vN;5$Z$
z<5gD2!`9|)?3Z{ZV(Hv`qQcOTLfcKT5U{ndg^?1?DRaK36y2)%lrJ}a$mv(Lf7rrh
zn`0$n28jr4gX5NJLWlCmgQEd6s<I@@r;5Km{ezO6G*d9NUL3=ViMh@<rqzf-hreWN
zoLQEqh18oPj;>8x+PHz8r?t5!i!?p66w2@^cE~#ctYcSP-I~_nHYF9GY@s{VY>VHS
zlewwY!2^9p<0GGc>)YKbYZ9ZV7F)OR-Vd<99-G4YzdR~(0o-xx*S)%H%WKCWH@A)G
z!MlIM#Lpj^9gKBc7=!8TzSLEPFR;14DadL@pUPS59;l6Y016hJ>wftkZpVX}iJwMq
zTE%y>1!0T^>>MILscdcH*;o;fpoi=c|Iiwa*dm@pF4OyVWbHkO0d%_%tYzogt7>*|
z^F^zAtTF8o%{}h5H}*u#Pj1!-WN%|{qNWQ>Mm0>Tlol&f(_-g}&g_a@7{M2MHig3b
zcc5a~o!zH>>jAy`i(oC?06>UUj1)Vj8-xXw_T90l$<z&K+-Zf@+aKPi1QB4{P-p|d
zAbDip?Xg7H#ov%kpoLdamMN3n>a!c<YxLeT+kRbv8$vvN5A4uv7sD{^I0EPa4OpLL
zjvRhMy`<~+2udfBjb^pU5J;@>*2d0(VuRLZM2XKq^;dylvj+M$Z2xtp;`Uj_*hKw`
zsc;J7GEgd7-(73bk}iLX^!;9^{mFA1WW2v1%6(QwN#&qtm^}7XGfH>weSM^VXNSEs
zs!C9}zrSmKUS(trtCLh5Ysy~UCmS~W#*=K3rXj+;LHMX^@CX*L&1{a@SyM(L=>US_
z`DjG^;-l*7Ez?ic*3iS;?R$KCR(fAbC(LZQfobHR9tU-%6~T~K2^Oml2e=h&EM-<L
z)D(~mWwvNIwfPXE7U>Xt`d;@&cdxvn(>gs@xEsj^!C8^>1HOP%j{~r%id~$jr!6s~
zqT2ija`@2-raUO^HRmD2z3Z1|y1Pg0!#5gj-xqo7Vpnf?<XpY0N<^)g<WjE`E`2g3
zb9g+NARo573wYi-J(uwD53g&td2J`1>IR+FH3#|Rnuj!b=F3O*14j4eUDl*;yU>un
zQ9gX^-59`?ICf+CNO7meoJ3x%7=)S^K~BWdYa^o;I=f?4v%`NEmeJmQ{QmFCvWR9I
z%B8lJ9Grlpa?hJxrgh)TYd>4wLlT#rlb@;1w{-VAGao8y>F%fAIxhAQ)iBuK->dSH
zvW<)kHsd>nof+eEbZh<b!=Y7P40JG6fQ?jA&Ve><9@!B!v^a=7aM*VykpNd!32eW8
zS<y%MEpBCr87JiuqPWP)UKt+TB@3XJCKfeoQVJGOqiQ&QeXP+!^C?54YPQ$6_qysc
z47&y0m#E9wPS*$cX3l}D*QkgV3-94ptV&vXUo=|SWJLpC0kl6ZVxC&sL`Gn|i4;8&
zHa=jM7+fVm)Ej!$cX5#y|BeqLd)8haKODxaJm8HTuT?Z${)UGv`+h^O;40qwd62T?
z3|aR{+i|k}bKfivQ1&?meBNKDxid72^}u)|fRf+-m@@yxL_r}>6O|G}cScV$+B-=`
zE8#kmUrg>oAG>oz&3#SF3HH=<!RiQ<g0QM%5AfpR*Cwcj!jrd5L<Pvs!Op+mlupV9
zN>nE|xOdS7m<B!<Q}rNoXloL|w`_>p#C>cj$7pHi4VeS^Ub?jTA&eMgso(4_SwiLb
z(a|<}XplFMevZ-5s4heL{Y>$Po_RWCC8QKYNO#7T2Xir^adJWS9s8CK^@3{_6f5BG
ze3$1KIkBGjmC;Gr8MfVXO<kONVI8q+j<*_%WYv%&rwhiaxy80KJAik?D<$i{ZJ;a|
zEKENzR^wTo5zje2`AlpL1KYn@=j%P&J8t2V{eRV=>FtdBV)FVr$FmI$1VZU@jPcO{
zYO#88+iwQ4+)Va66sKh%$rIp^#`Z_>4mVw>(d;|1a$foy2P_n7*r<GY@B0;?80o_$
zy^wZY%yXO;QAbIj);)togkl``$S0d(LhKfRzejhKK^W2y<pQia^m5^K7*hBH-QBM@
zjLwItb=BxXGSINxyy9c!NN^kh507MJd_@fCpi)8E(DBm?tyK#_=?QlE*Q?h(vFPpU
zRXPDex$=g6bo!y*0t|f`2v-NW1xY&6&UipDyYZfo{N6182E@qFbG~Y(62ox3Lmt=t
z#y-<A;nC?+GE!t2-f)bZ8WC<mezUxcMzbd7<C9~r2PY7G(1=z6RdnD<j3r&KYQ6>Y
zj8Qd_f{6|}5C(S)>YDAE=$AB^0jDw$Uh-ZWvx}%j2#P6ns-v4HVUu{0lya`?>X~=&
zpW~+FLkq()oEC3=E>#CXc9$<+q<-IZNLG8%hsa(Esa@UB8X8+M?(hwK@>e!2&|kj3
zQ0e3BP6((M$_xwZ{*-i{fyEYkWFaqv4g%Z_v<EiO1fXsjsLG@jGlZ}hBjwUoU_PM-
z9c_ML+eE0HY}N7Fl>P295LR{1?;G}R_|@G2pnb{HtCjE3kbYYf!t8OOYM=XfYqI%C
zvf+c#nPLeDsWHGYY^2C!u_;?!Muor#ODp=o$n%Kkjq}a-?qmmd4fnO~UI1b5%%=yK
z$?!$vmi)XFoOJ(SQ#Vhjgm}`-vtH>{Z#C}Z*PrXV3#+@MJP_K8rm!=cdA$l2Gin7I
zp&LBf6^M&R#|TNOkff4fzFpbCQ*Fe*DkSNQJ~mBk49f6+YhI4SqP;WnZ-pa$Sh6I;
z6$9%qAO6Bapk6I#8nHSOYd)<I`O!V70+`>SY0>fFvd;&)_1_!$4TUa;OfSA1#f{g>
z&=S=0`N7#4hmhenggc|PH)W($j;D+FmhCFeq6LLd&wncR)>)oxvSwFOV^Q1&?~vy&
z6Ui1J#KMH8i>zK{BJVd+c6T1ZLZfW^+?Tq`d=?>oUsLR2!6Y=;C!-W&>zi_{#m2XO
z1?kVxBe<TcLNyGdmAOq98YAM<TqAW2Ubmj{nwp0_^}Mz+dIPjn7)*jBDjF1wwaz6!
zRUHw3{szQgwIBBGU#)Vdq}sqqo^r${{J`515P0mB4QSX~N4Ge;Cz50^6CFYbk^64+
z=p*4+3+%Os^9O^c$F)Fs<lwmw$N}ZKw7q8q@!V$b@tB_@A0XR@Q(Ww%tli@bfpFzs
z<p*eOMb^;L6UfqY?RO!8ZI^QM%Kh|L7FiURnd#m#MwW%xOn6jijVS5?IP>G>qR<46
za{V%>{b>c6AkDDih1;`jch~@LFgt@=F`#I=GeBz$zjN>z0y1js8f19pb%aAO^`JH|
zpFRG49}AI-98D=c?b<ctDW1Bnb|46HR&#gx*>3_Zj};0G4BTz!_IW?qX|#-Ll^Htt
zVt`h!iPN9ssGBr?2c7ATiqbo@_~p~A6EnppQA%^P$lR4o@lvgn)gOM1Z+tS|-mliN
zWw*U}Bsg96)Kb)7@=G9YGgR$mnA$v<52XR#oED8+9VP9Oiht)7<Va<`;{!3rh%VWn
zJ8eX1l<it*227;2zYcXumV08^yWQ{Yg4&=L$h!vylr2)Df!8#MsKlZ{H_%k*OUzIb
z`W)jvg(NyQ0*`_&xqTVJ;0&>ojTY@qlxc8aZg3B<o4z(+_;|%Q*~XD@a^kRk`%*ru
z9aq38SZLRt7AQOe%;S>H(&~_W(orE&`drY^oL<=F?1*V|S1JgU1|s3mYkXp&;||?>
zosv51RI_FCfv==KP$5u(=O6p1l)O8@5SmSPjf%L>s+4OB>h!^b%)7~!%{BlIL8_7L
zr@CR99a!K&n0X7>Y4M+qc%l^8!f<$&F2O7}e8kJuA~qp7kDv=$y7nQ_ON2ztD(R~6
zoAB*%kH3P%H`{>XV?-)-hKGz4k;{WjT*#9mLHv{K&J=4M_-akMh^_`i!saZn!DDH-
z#ix!rcFu@X=~^CE^RU$RyNt{LD`{<Sd;g0ag`TwhiqT_Sl#mi698z8eqP*?FKJc=y
zspe6Tg*HM>)xxNZM&)Fq?@(Sz>`;D%hLi6*UfO#JYvH}Sms?A!QbeVgeUShxYJ5s4
zsq%@WO*r5J?~av~D3s;WZYi47SVvCW%};{?_CayOH+E{e=_mw*hOiP!kPu)R!e{5n
zxU9mksTRAtW$&S--M}SIR{E(j%NO(*Akf#!45lI#4a>iRBa7c%Qb`4o9_L%@whSmB
z%Pd+X>+^xQ8Q|Y2-?q3;p6{AtF;wNFmWEuLB6|_y%=~29e1cQcfi^{u;1fM&*E*;B
zhsqZuAuzmjR&eFUR;_2`ImCt+a_HpgqutO;&CipFd&=n<-HR76BlLGf<KibQVt<mQ
zcKh}SSMn@6I6XKpvicms#!Rl|udU&#5l(@BnbsE$UCLgnVEHN3%V}lJf>^Ta&D^()
z;GtaB);>8txxf=<40K;z81&jw5>ae;xb+!}`*7Mo@cAL476XJd{`Vv4Y&N3Q9l+GE
zv|)G%0UX+dit3M5C!vEnUOZvpjiCUW$$m8gs)1NAbY)eECfy~d{XC`he$j9TbMeDL
z3|;5w!_8dWXI0Ts88?5y8g}2)B{en+HP-(mS%vgzRQMqtXu`JMIM;md#N@GfNcz^5
zERX`F@AEtv9~>%G^ua9kJ(NIqqEiNJpB*>Po<hrMQB!^ZY1#t7WNcT_WH6=uO=<~*
z*^w07V+2CYN31ka7|0NX(6rf!2T&`wBH#}(LCRXNRcQU)ci5jorL4?*pE4VZT>Alq
z6IuXeJ`0}J`xW1Q_?{D3;0;3Dl~%jI?-}`L(Cf8dEq>r1ZjXa7y7%VIRnYt+z26V!
zPJX5E%;Z<({&D*QK=#A~+FYxg;CI<K9|u0N;A*<)+P5V&0S!TI<Nt?MtuK0&%>N%d
reM_$55AY8H&awX&!2jP}UxZxPGG=MFsOt#c$6;=2bvXZ!+tvRA$LlzT

literal 0
HcmV?d00001

diff --git a/docs/2_39/reference-manual/includes.html b/docs/2_39/reference-manual/includes.html
new file mode 100644
index 000000000..d34a68824
--- /dev/null
+++ b/docs/2_39/reference-manual/includes.html
@@ -0,0 +1,1152 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Includes</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/comments.html" rel="next">
+<link href="../reference-manual/encoding.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Includes">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/includes.html">Includes</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#includes.section" id="toc-includes.section" class="nav-link active" data-scroll-target="#includes.section">Includes</a>
+  <ul class="collapse">
+  <li><a href="#space-before-includes" id="toc-space-before-includes" class="nav-link" data-scroll-target="#space-before-includes">Space before includes</a></li>
+  <li><a href="#comments-after-includes" id="toc-comments-after-includes" class="nav-link" data-scroll-target="#comments-after-includes">Comments after includes</a></li>
+  <li><a href="#recursive-includes" id="toc-recursive-includes" class="nav-link" data-scroll-target="#recursive-includes">Recursive includes</a></li>
+  <li><a href="#include-paths" id="toc-include-paths" class="nav-link" data-scroll-target="#include-paths">Include paths</a>
+  <ul class="collapse">
+  <li><a href="#slashes-in-include-paths" id="toc-slashes-in-include-paths" class="nav-link" data-scroll-target="#slashes-in-include-paths">Slashes in include paths</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/includes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/includes.html">Includes</a></li></ol></nav></header>
+
+
+
+
+<section id="includes.section" class="level1">
+<h1>Includes</h1>
+<p>Stan allows one file to be included within another file using a syntax similar to that from C++. For example, suppose the file <code>my-std-normal.stan</code> defines the standard normal log probability density function (up to an additive constant).</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> my_std_normal_lpdf(<span class="dt">vector</span> y) {</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> -<span class="fl">0.5</span> * y' * y;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Suppose we also have a file containing a Stan program with an include statement.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="co">#include my-std-normal.stan</span></span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  y ~ my_std_normal();</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This Stan program behaves as if the contents of the file <code>my-std-normal.stan</code> replace the line with the <code>#include</code> statement, behaving as if a single Stan program were provided.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> my_std_normal_lpdf(<span class="dt">vector</span> y) {</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> -<span class="fl">0.5</span> * y' * y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  y ~ my_std_normal();</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There are no restrictions on where include statements may be placed within a file or what the contents are of the replaced file.</p>
+<section id="space-before-includes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="space-before-includes">Space before includes</h3>
+<p>It is possible to use includes on a line non-initially. For example, the previous example could’ve included space before the <code>#</code> in the include line:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>    <span class="co">#include my-std-normal.stan</span></span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If there is initial space before an include, it will be discarded.</p>
+</section>
+<section id="comments-after-includes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="comments-after-includes">Comments after includes</h3>
+<p>It is also possible to include line-based comments after the include. For example, the previous example can be coded as:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="co">#include my-std-normal.stan  // definition of standard normal</span></span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Line comments are discarded when the entire line is replaced with the contents of the included file.</p>
+</section>
+<section id="recursive-includes" class="level2">
+<h2 class="anchored" data-anchor-id="recursive-includes">Recursive includes</h2>
+<p>Recursive includes will lead to a compiler error. For example, suppose <code>a.stan</code> contains</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="co">#include b.stan</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and <code>b.stan</code> contains</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="co">#include a.stan</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This will result in an error explaining the circular dependency:</p>
+<pre><code>Syntax error in './b.stan', line 1, column 0, included from
+'./a.stan', line 1, column 0, included from
+'./b.stan', line 1, column 0, included from
+'a.stan', line 1, column 0, include error:
+   -------------------------------------------------
+     1:  #include a.stan
+         ^
+   -------------------------------------------------
+
+File a.stan recursively included itself.</code></pre>
+</section>
+<section id="include-paths" class="level2">
+<h2 class="anchored" data-anchor-id="include-paths">Include paths</h2>
+<p>The Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.</p>
+<section id="slashes-in-include-paths" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="slashes-in-include-paths">Slashes in include paths</h3>
+<p>If there is not a final <code>/</code> or <code>\</code> in the path, a <code>/</code> will be appended between the path and the included file name.</p>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/encoding.html" class="pagination-link" aria-label="Character Encoding">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Character Encoding</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/comments.html" class="pagination-link" aria-label="Comments">
+        <span class="nav-page-text">Comments</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/includes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/index.html b/docs/2_39/reference-manual/index.html
new file mode 100644
index 000000000..c4a192f88
--- /dev/null
+++ b/docs/2_39/reference-manual/index.html
@@ -0,0 +1,1062 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Stan Reference Manual</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/encoding.html" rel="next">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Reference Manual – Stan Docs">
+<meta property="og:description" content="Version 2.39">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed slimcontent">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/index.html">Stan Reference Manual</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar zindex-bottom">
+    </div>
+<!-- main -->
+<main class="content page-columns page-full" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">Stan Reference Manual</h1>
+<p class="subtitle lead">Version 2.39</p>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    
+  
+    
+  </div>
+  
+
+
+</header>
+
+
+<div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
+<p>This is the official reference manual for <a href="https://mc-stan.org/">Stan</a>’s <em>programming language</em> for coding probability models, <em>inference algorithms</em> for fitting models and making predictions, and <em>posterior analysis</em> tools for evaluating the results. This manual applies to all Stan interfaces.</p>
+<p>The first part of the reference manual provides a full specification of the Stan programming language. The language is responsible for defining a log density function conditioned on data. Typically, this is a Bayesian posterior, but it may also be a penalized likelihood function. The second part of the manual specifies the inference algorithms and posterior inference tools. The third part provides auxiliary information about the use of Stan.</p>
+<p><a href="https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf">Download the pdf version of this manual</a>.</p>
+<section id="copyright-and-trademark" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
+<ul>
+<li><p>Copyright 2011–2025, Stan Development Team and their assignees.</p></li>
+<li><p>The Stan name and logo are registered trademarks of NumFOCUS.</p></li>
+</ul>
+</section>
+<section id="licensing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
+<ul>
+<li><p><em>Text content:</em> <a href="https://creativecommons.org/licenses/by-nd/4.0/legalcode">CC-BY ND 4.0 license</a></p></li>
+<li><p><em>Computer code:</em> <a href="https://opensource.org/licenses/BSD-3-Clause">BSD 3-clause license</a></p></li>
+<li><p><em>Logo:</em> <a href="https://mc-stan.org/about/#copyright-and-trademark">Stan logo usage guidelines</a></p></li>
+</ul>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/encoding.html" class="pagination-link" aria-label="Character Encoding">
+        <span class="nav-page-text">Character Encoding</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/laplace.html b/docs/2_39/reference-manual/laplace.html
new file mode 100644
index 000000000..14b05ecbd
--- /dev/null
+++ b/docs/2_39/reference-manual/laplace.html
@@ -0,0 +1,1078 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Laplace Approximation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/laplace_embedded.html" rel="next">
+<link href="../reference-manual/variational.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Laplace Approximation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace.html">Laplace Approximation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#laplace-approximation" id="toc-laplace-approximation" class="nav-link active" data-scroll-target="#laplace-approximation">Laplace Approximation</a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace.html">Laplace Approximation</a></li></ol></nav></header>
+
+
+
+
+<section id="laplace-approximation" class="level1">
+<h1>Laplace Approximation</h1>
+<p>Stan provides a Laplace approximation algorithm which can be used to obtain draws from an approximated posterior. The Laplace approximation works in the unconstrained space, so that if there are constrained parameters, the normal approximation is centered at the mode in the unconstrained space and then the implemented method transforms the normal approximation sample to the constrained space before outputting them.</p>
+<p>Given the estimate of the mode <span class="math inline">\(\widehat{\theta}\)</span>, the Hessian <span class="math inline">\(H(\widehat{\theta})\)</span> is computed using central finite differences of the model functor. Next the algorithm computes the Cholesky factor of the negative inverse Hessian:</p>
+<p><span class="math inline">\(R^{-1} = \textrm{chol}(-H(\widehat{\theta})) \backslash \mathbf{1}\)</span>.</p>
+<p>Each draw is generated on the unconstrained scale by sampling</p>
+<p><span class="math inline">\(\theta^{\textrm{std}(m)} \sim \textrm{normal}(0, \textrm{I})\)</span></p>
+<p>and defining draw <span class="math inline">\(m\)</span> to be</p>
+<p><span class="math inline">\(\theta^{(m)} = \widehat{\theta} + R^{-1} \cdot \theta^{\textrm{std}(m)}\)</span></p>
+<p>Finally, each <span class="math inline">\(\theta^{(m)}\)</span> is transformed back to the constrained scale.</p>
+<p>The one-time computation of the Cholesky factor incurs a high constant overhead of <span class="math inline">\(\mathcal{O}(N^3)\)</span> in <span class="math inline">\(N\)</span> dimensions. It also requires <span class="math inline">\(2N\)</span> gradient calculations to use as the basis, which scales at best as <span class="math inline">\(\mathcal{O}(N^2)\)</span> and is worse for models whose gradient calculation is super-linear in dimension. The algorithm also has a high per-draw overhead, requiring <span class="math inline">\(N\)</span> standard normal pseudorandom numbers and <span class="math inline">\(\mathcal{O}(N^2)\)</span> per draw (to multiply by the Cholesky factor). For <span class="math inline">\(M\)</span> draws, the total cost is proportional to <span class="math inline">\(\mathcal{O}(N^3 + M \cdot N^2)\)</span>.</p>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/variational.html" class="pagination-link" aria-label="Variational Inference">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Variational Inference</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/laplace_embedded.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <span class="nav-page-text">Embedded Laplace Approximation</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/laplace_embedded.html b/docs/2_39/reference-manual/laplace_embedded.html
new file mode 100644
index 000000000..9e3a54307
--- /dev/null
+++ b/docs/2_39/reference-manual/laplace_embedded.html
@@ -0,0 +1,1226 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Embedded Laplace Approximation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/diagnostics.html" rel="next">
+<link href="../reference-manual/laplace.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Embedded Laplace Approximation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace_embedded.html">Embedded Laplace Approximation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#embedded-laplace-approximation" id="toc-embedded-laplace-approximation" class="nav-link active" data-scroll-target="#embedded-laplace-approximation">Embedded Laplace Approximation</a>
+  <ul class="collapse">
+  <li><a href="#latent-gaussian-models" id="toc-latent-gaussian-models" class="nav-link" data-scroll-target="#latent-gaussian-models">Latent Gaussian models</a></li>
+  <li><a href="#approximation-of-the-conditional-posterior-and-marginal-likelihood" id="toc-approximation-of-the-conditional-posterior-and-marginal-likelihood" class="nav-link" data-scroll-target="#approximation-of-the-conditional-posterior-and-marginal-likelihood">Approximation of the conditional posterior and marginal likelihood</a></li>
+  <li><a href="#trade-offs-of-the-approximation" id="toc-trade-offs-of-the-approximation" class="nav-link" data-scroll-target="#trade-offs-of-the-approximation">Trade-offs of the approximation</a>
+  <ul class="collapse">
+  <li><a href="#when-the-approximation-is-appropriate" id="toc-when-the-approximation-is-appropriate" class="nav-link" data-scroll-target="#when-the-approximation-is-appropriate">When the approximation is appropriate</a></li>
+  </ul></li>
+  <li><a href="#details-of-the-approximation" id="toc-details-of-the-approximation" class="nav-link" data-scroll-target="#details-of-the-approximation">Details of the approximation</a>
+  <ul class="collapse">
+  <li><a href="#tuning-the-newton-solver" id="toc-tuning-the-newton-solver" class="nav-link" data-scroll-target="#tuning-the-newton-solver">Tuning the Newton solver</a></li>
+  <li><a href="#solver-strategies" id="toc-solver-strategies" class="nav-link" data-scroll-target="#solver-strategies">Solver Strategies</a></li>
+  <li><a href="#sparse-hessian-of-the-log-likelihood" id="toc-sparse-hessian-of-the-log-likelihood" class="nav-link" data-scroll-target="#sparse-hessian-of-the-log-likelihood">Sparse Hessian of the log likelihood</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace_embedded.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace_embedded.html">Embedded Laplace Approximation</a></li></ol></nav></header>
+
+
+
+
+<section id="embedded-laplace-approximation" class="level1">
+<h1>Embedded Laplace Approximation</h1>
+<p>The embedded Laplace approximation replaces explicit sampling of potentially high-dimensional Gaussian latent variables with a local Gaussian approximation, marginalizing them out so that inference proceeds over the remaining hyperparameters alone. This approach is often referred to as the integrated Laplace approximation, although the exact details of the method can vary. The details of Stan’s implementation can be found in references <span class="citation" data-cites="Margossian:2020 Margossian:2023">(<a href="#ref-Margossian:2020" role="doc-biblioref">Margossian et al. 2020</a>; <a href="#ref-Margossian:2023" role="doc-biblioref">Margossian 2023</a>)</span>.</p>
+<p>A standard approach to fit a latent Gaussian model would be to perform inference jointly over the latent Gaussian variables and the hyperparameters. Instead, the embedded Laplace approximation can be used to do <em>approximate</em> marginalization of the latent Gaussian variables; we can then use any inference over the remaining hyperparameters. By marginalizing out the latent variables, the sampler explores a lower-dimensional, better-behaved marginal posterior. Individual iterations are more expensive (each requires an inner optimization), but the sampler typically needs far fewer iterations to achieve the same effective sample size. How this trade-off resolves depends on the specific problem at hand.</p>
+<p>For complete function signatures and the built-in likelihood wrappers (Poisson, Negative Binomial, Bernoulli), see the <a href="../functions-reference/embedded_laplace.html">Embedded Laplace functions reference</a>. For worked examples with full data blocks, see the <a href="../stan-users-guide/gaussian-processes.html#poisson-gp-using-an-embedded-laplace-approximation">Gaussian Processes chapter</a>.</p>
+<section id="latent-gaussian-models" class="level2">
+<h2 class="anchored" data-anchor-id="latent-gaussian-models">Latent Gaussian models</h2>
+<p>The embedded Laplace approximation is used for latent Gaussian models. A latent Gaussian model is defined by three components:</p>
+<ul>
+<li><span class="math inline">\(\phi\)</span>: hyperparameters (e.g., GP kernel length-scale and magnitude, or variance components in a hierarchical model),</li>
+<li><span class="math inline">\(\theta\)</span>: latent Gaussian variables (the high-dimensional quantity to be marginalized out),</li>
+<li><span class="math inline">\(y\)</span>: observed data.</li>
+</ul>
+<p>These components are related through a hierarchical structure. The hyperparameters <span class="math inline">\(\phi\)</span> are given a prior <span class="math inline">\(p(\phi)\)</span>. The latent variables <span class="math inline">\(\theta\)</span> have a multivariate normal prior centered at 0 with covariance matrix <span class="math inline">\(K(\phi)\)</span>. An non-zero mean offset can be incorporated into the likelihood function. The observations <span class="math inline">\(y\)</span> have a data model <span class="math inline">\(p(y \mid \theta, \phi)\)</span>. The prior on <span class="math inline">\(\theta\)</span> is centered at zero; an offset can be incorporated into the data model if a non-zero mean is needed.</p>
+<p><span class="math display">\[\begin{eqnarray*}
+  \phi &amp; \sim &amp; p(\phi) \\
+  \theta &amp; \sim &amp; \text{Multi-Normal}(0, K(\phi)) \\
+  y &amp; \sim &amp; p(y \mid \theta, \phi).
+\end{eqnarray*}\]</span></p>
+<p>The generative model above defines a joint distribution over all three quantities, <span class="math inline">\(p(\phi, \theta, y) = p(\phi) \, p(\theta \mid \phi) \, p(y \mid \theta, \phi)\)</span>. After observing data <span class="math inline">\(y\)</span>, Bayes’ theorem gives the joint posterior <span class="math inline">\(p(\phi, \theta \mid y) \propto p(\phi) \, p(\theta \mid \phi) \, p(y \mid \theta, \phi)\)</span>, where <span class="math inline">\(p(y \mid \theta, \phi)\)</span> as function of <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span> is the joint likelihood function.</p>
+<p>Sampling directly from the joint posterior <span class="math inline">\(p(\phi, \theta \mid y)\)</span> of this model is often difficult. Challenging geometries (e.g., funnels) frustrate inference algorithms, including Hamiltonian Monte Carlo and variational inference. However, the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span> is often well-behaved and low-dimensional, making it much easier to sample. With an embedded Laplace approximation, we can obtain an approximation of the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span>. This is done via an intermediate approximation of the conditional posterior <span class="math inline">\(p(\theta \mid \phi, y)\)</span> by a normal distribution and this normal approximation is well-justified when the likelihood <span class="math inline">\(p(y \mid \theta, \phi)\)</span> as function of <span class="math inline">\(\theta\)</span> is log concave (given that we already have a normal prior <span class="math inline">\(p(\theta \mid \phi)\)</span>). Once we obtain (approximate) samples <span class="math inline">\(\phi^{(i)} \sim p(\phi \mid y)\)</span>, we can in turn generate posterior draws for <span class="math inline">\(\theta\)</span> using the normal approximation to <span class="math inline">\(p(\theta \mid y, \phi)\)</span>, evaluated at the posterior draws <span class="math inline">\(\phi^{(i)}\)</span>.</p>
+</section>
+<section id="approximation-of-the-conditional-posterior-and-marginal-likelihood" class="level2">
+<h2 class="anchored" data-anchor-id="approximation-of-the-conditional-posterior-and-marginal-likelihood">Approximation of the conditional posterior and marginal likelihood</h2>
+<p>The two-step inference strategy for using embedded Laplace in a latent Gaussian model requires approximations to both the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> and the marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span>. The Laplace approximation is the normal distribution that matches the mode and curvature of the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span>. The mode, defined as the value of <span class="math inline">\(\theta\)</span> that maximizes the conditional posterior, is estimated by a Newton solver, <span class="math display">\[
+  \theta^* = \underset{\theta}{\text{argmax}} \ p(\theta \mid y, \phi),
+\]</span></p>
+<p>Since the approximation is normal, the curvature is matched by setting the covariance to the negative Hessian of the log conditional posterior, evaluated at the mode,</p>
+<p><span class="math display">\[
+  \Sigma^* = -  \left . \frac{\partial^2}{\partial \theta^2}
+    \log p (\theta \mid \phi, y) \right |_{\theta =\theta^*}.
+\]</span></p>
+<p>The resulting Laplace approximation is a multivariate normal centered at the mode with covariance given by the inverse curvature,</p>
+<p><span class="math display">\[
+\hat p_\mathcal{L} (\theta \mid y, \phi) = \text{Multi-Normal}(\theta^*, \Sigma^*)
+\approx p(\theta \mid y, \phi).
+\]</span></p>
+<p>This approximation also yields an approximation to the marginal likelihood, obtained by evaluating the prior, likelihood, and approximate posterior at the mode <span class="math inline">\(\theta^*\)</span>,</p>
+<p><span class="math display">\[
+   \hat p_\mathcal{L}(y \mid \phi) := \frac{p(\theta^* \mid \phi) \
+   p(y \mid \theta^*, \phi) }{ \hat p_\mathcal{L} (\theta^* \mid \phi, y) }
+   \approx p(y \mid \phi).
+\]</span></p>
+<p>Hence, a strategy to approximate the posterior of the latent Gaussian model is to first estimate the marginal posterior <span class="math inline">\(\hat p_\mathcal{L}(\phi \mid y) \propto p(\phi) p_\mathcal{L} (y \mid \phi)\)</span> using any algorithm supported by Stan. Approximate posterior draws for the latent Gaussian variables are then obtained by first sampling <span class="math inline">\(\phi \sim \hat p_\mathcal{L}(\phi \mid y)\)</span> and then <span class="math inline">\(\theta \sim  \hat p_\mathcal{L}(\theta \mid \phi, y)\)</span>.</p>
+</section>
+<section id="trade-offs-of-the-approximation" class="level2">
+<h2 class="anchored" data-anchor-id="trade-offs-of-the-approximation">Trade-offs of the approximation</h2>
+<p>The embedded Laplace approximation presents several trade-offs with standard inference over the joint posterior <span class="math inline">\(p(\theta, \phi \mid y)\)</span>. The main advantage of the embedded Laplace approximation is that it side-steps the intricate geometry of hierarchical models. The marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span> can then be handled by Hamiltonian Monte Carlo sampling without extensive tuning or reparameterization, and the mixing time is faster, meaning we can run shorter chains to achieve a desired precision. One additional benefit is that approximate methods, e.g.&nbsp;variational inference, which work poorly on the joint <span class="math inline">\(p(\theta, \phi \mid y)\)</span> can work well on the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span>.</p>
+<p>On the other hand, the embedded Laplace approximation presents certain disadvantages. First, we need to perform a Laplace approximation each time the log marginal likelihood is evaluated, meaning each iteration can be expensive. Secondly, the approximation can introduce non-negligible error, especially with non-log-concave likelihoods (note the prior is always multivariate normal). How these trade-offs are resolved depends on the application; see <span class="citation" data-cites="Margossian:2020">Margossian et al. (<a href="#ref-Margossian:2020" role="doc-biblioref">2020</a>)</span> for some examples.</p>
+<section id="when-the-approximation-is-appropriate" class="level3">
+<h3 class="anchored" data-anchor-id="when-the-approximation-is-appropriate">When the approximation is appropriate</h3>
+<p>The quality of the Laplace approximation depends on how close the true conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> is to Gaussian.</p>
+<p><strong>Works well.</strong> Log-concave likelihoods, for example from a Poisson model with log link or negative binomial with log link. These produce unimodal conditional posteriors when combined with a Gaussian prior. The approximation error is typically negligible with these likelihoods, especially with moderate-to-large counts <span class="citation" data-cites="Kuss:2005 Vanhatalo:2010 Cseke:2011 Vehtari:2016">(<a href="#ref-Kuss:2005" role="doc-biblioref">Kuss and Rasmussen 2005</a>; <a href="#ref-Vanhatalo:2010" role="doc-biblioref">Vanhatalo, Pietiläinen, and Vehtari 2010</a>; <a href="#ref-Cseke:2011" role="doc-biblioref">Cseke and Heskes 2011</a>; <a href="#ref-Vehtari:2016" role="doc-biblioref">Vehtari et al. 2016</a>)</span>. If the likelihood is normal, there is no error in the approximation but in this case the marginalization can be worked analytically and the resulting implementation is much faster than using the embedded Laplace approximation.</p>
+<p><strong>Works adequately.</strong> Bernoulli model with logit link has technically log-concave likelihood, but the likelihood can be very skewed making the Gaussian approximation less accurate than for count data. The embedded Laplace is still useful when <span class="math inline">\(\theta\)</span> is high-dimensional and joint sampling is infeasible; see <span class="citation" data-cites="Vehtari:2016">Vehtari et al. (<a href="#ref-Vehtari:2016" role="doc-biblioref">2016</a>)</span> and <span class="citation" data-cites="Margossian:2020">Margossian et al. (<a href="#ref-Margossian:2020" role="doc-biblioref">2020</a>)</span> for discussion.</p>
+<p><strong>Not appropriate.</strong> For likelihoods that are not log-concave in <span class="math inline">\(\theta\)</span>, the conditional posterior may be multimodal and the Newton solver finds only a single mode or can fail completely. When <span class="math inline">\(\theta\)</span> is low-dimensional (a few dozen or fewer), the overhead of the inner optimization may not pay for itself and standard joint HMC sampling is often adequate.</p>
+</section>
+</section>
+<section id="details-of-the-approximation" class="level2">
+<h2 class="anchored" data-anchor-id="details-of-the-approximation">Details of the approximation</h2>
+<p>When the embedded Laplace approximation does not converge or produces unexpected results, the solver configuration may need adjustment. This section describes the internals of the Newton solver and the options available for tuning it.</p>
+<section id="tuning-the-newton-solver" class="level3">
+<h3 class="anchored" data-anchor-id="tuning-the-newton-solver">Tuning the Newton solver</h3>
+<p>A critical component of the embedded Laplace approximation is the Newton solver used to estimate the mode <span class="math inline">\(\theta^*\)</span> of <span class="math inline">\(p(\theta \mid \phi, y)\)</span>. The objective function being maximized is the log joint density of the prior and likelihood with respect to <span class="math inline">\(\theta\)</span>.</p>
+<p><span class="math display">\[
+\Psi(\theta) = \log p(\theta \mid \phi) + \log p(y \mid \theta, \phi),
+\]</span></p>
+<p>Convergence is declared when the change in the objective between successive iterations falls below a <em>tolerance</em> <span class="math inline">\(\Delta\)</span>.</p>
+<p><span class="math display">\[
+| \Psi (\theta^{(i + 1)}) - \Psi (\theta^{(i)}) | \le \Delta.
+\]</span></p>
+<p>The solver also stops after reaching a pre-specified <em>maximum number of steps</em>. In that case, Stan throws a warning, but still returns the last iteration’s parameters. If you see this warning you should check the diagnostics to understand why the solver failed to converge.</p>
+<p>To help with cases where the Newton step does not lead to a decrease in the objective function, the Newton iteration is augmented with a wolfe line-search to ensure that at each iteration the objective function <span class="math inline">\(\Psi\)</span> decreases. Specifically, suppose the objective increases after a Newton step, indicating the step overshot a region of improvement,</p>
+<p><span class="math display">\[
+\Psi (\theta^{(i + 1)}) &lt; \Psi (\theta^{(i)}).
+\]</span></p>
+<p>This can indicate that the Newton step <span class="math inline">\(\alpha\)</span> at iteration <span class="math inline">\(i\)</span> is too large and that we skipped a region where the objective function decreases. In that case, we can fallback to a Wolfe line search to find a step size which satisfies the Wolfe conditions. The wolfe line search attempts to find a search direction <span class="math inline">\(p_i\)</span> and step size <span class="math inline">\(\alpha_k\)</span> such that an accepted step both increases our objective while ensuring the slope of the accepted step is flatter than our previous position. Together these help push the algorithm towards a minimum.</p>
+<p><span class="math display">\[
+f(x_i + \alpha_k p_i) \le f(x_i) + c_1 \alpha_k \nabla f(x_i)^T p_i
+-p^T_i \Delta f(x_i + \alpha_k p_i) \le -c_2 p^T_i \Delta f(x_i)
+\]</span></p>
+<p><span class="math display">\[
+  \theta^{(i + 1)} \leftarrow \frac{\theta^{(i + 1)} + \theta^{(i)}}{2}.
+\]</span></p>
+<p>We repeat this halving of steps until <span class="math inline">\(\Psi (\theta^{(i + 1)}) \ge \Psi (\theta^{(i)})\)</span>, or until a maximum number of linesearch steps is reached. For certain problems, adding a linesearch can make the optimization more stable.</p>
+</section>
+<section id="solver-strategies" class="level3">
+<h3 class="anchored" data-anchor-id="solver-strategies">Solver Strategies</h3>
+<p>The embedded Laplace approximation uses a custom Newton solver, specialized to find the mode of <span class="math inline">\(p(\theta \mid \phi, y)\)</span>. A key step for efficient optimization is to ensure all matrix inversions are numerically stable. This can be done using the Woodbury-Sherman-Morrison formula and requires one of three matrix decompositions:</p>
+<ol type="1">
+<li><p>Cholesky decomposition of the Hessian of the negative log likelihood <span class="math inline">\(W = - \partial^2_\theta \log p(y \mid \theta, \phi)\)</span>.</p></li>
+<li><p>Cholesky decomposition of the prior covariance matrix <span class="math inline">\(K(\phi)\)</span>.</p></li>
+<li><p>LU-decomposition of <span class="math inline">\(I + KW\)</span>, where <span class="math inline">\(I\)</span> is the identity matrix.</p></li>
+</ol>
+<p>The first solver (1) should be used if the negative log likelihood is positive-definite. Otherwise the user should rely on (2). In rarer cases where it is not numerically safe to invert the covariance matrix <span class="math inline">\(K\)</span>, users can use the third solver as a last-resort option.</p>
+</section>
+<section id="sparse-hessian-of-the-log-likelihood" class="level3">
+<h3 class="anchored" data-anchor-id="sparse-hessian-of-the-log-likelihood">Sparse Hessian of the log likelihood</h3>
+<p>A key step to speed up computation is to take advantage of the sparsity of <span class="math inline">\(H\)</span>, the Hessian of the log likelihood with respect to the latent variables, <span class="math display">\[
+  H = \frac{\partial^2}{\partial \theta^2} \log p(y \mid \theta, \phi).
+\]</span> For example, if the observations <span class="math inline">\((y_1, \cdots, y_N)\)</span> are conditionally independent and each depends on only one component of <span class="math inline">\(\theta\)</span>, the log likelihood decomposes into a sum of per-observation terms, <span class="math display">\[
+  \log p(y \mid \theta, \phi) = \sum_{i = 1}^N \log p(y_i \mid \theta_i, \phi),
+\]</span> and the Hessian is diagonal. This leads to faster calculations of the Hessian and subsequently sparse matrix operations. This case is common in Gaussian process models, and certain hierarchical models.</p>
+<p>Stan’s suite of functions for the embedded Laplace approximation exploits block-diagonal structure in the Hessian, where the user specifies the block size B. The user can specify the size <span class="math inline">\(B\)</span> of these blocks. The user is responsible for working out what <span class="math inline">\(B\)</span> is. If the Hessian is dense, then we simply set <span class="math inline">\(B = N\)</span>. The diagonal case above corresponds to B = 1. Arbitrary sparsity patterns beyond block-diagonal structure are not currently supported.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Cseke:2011" class="csl-entry" role="listitem">
+Cseke, Botond, and Tom Heskes. 2011. <span>“Approximate Marginals in Latent <span>Gaussian</span> Models.”</span> <em>Journal of Machine Learning Research</em> 12.
+</div>
+<div id="ref-Kuss:2005" class="csl-entry" role="listitem">
+Kuss, Malte, and Carl E Rasmussen. 2005. <span>“Assessing Approximate Inference for Binary <span>Gaussian</span> Process Classification.”</span> <em>Journal of Machine Learning Research</em> 6: 1679–1704.
+</div>
+<div id="ref-Margossian:2023" class="csl-entry" role="listitem">
+Margossian, Charles C. 2023. <span>“General Adjoint-Differentiated Laplace Approximation.”</span> <em>arXiv:2306.14976</em>.
+</div>
+<div id="ref-Margossian:2020" class="csl-entry" role="listitem">
+Margossian, Charles C, Aki Vehtari, Daniel Simpson, and Raj Agrawal. 2020. <span>“Hamiltonian Monte Carlo Using an Adjoint-Differentiated Laplace Approximation: Bayesian Inference for Latent Gaussian Models and Beyond.”</span> <em>Advances in Neural Information Processing Systems</em> 34.
+</div>
+<div id="ref-Vanhatalo:2010" class="csl-entry" role="listitem">
+Vanhatalo, Jarno, Ville Pietiläinen, and Aki Vehtari. 2010. <span>“Approximate Inference for Disease Mapping with Sparse <span>Gaussian</span> Processes.”</span> <em>Statistics in Medicine</em> 29 (15): 1580–1607.
+</div>
+<div id="ref-Vehtari:2016" class="csl-entry" role="listitem">
+Vehtari, Aki, Tommi Mononen, Ville Tolvanen, Tuomas Sivula, and Ole Winther. 2016. <span>“Bayesian Leave-One-Out Cross-Validation Approximations for <span>Gaussian</span> Latent Variable Models.”</span> <em>Journal of Machine Learning Research</em> 17 (103): 1–38. <a href="http://jmlr.org/papers/v17/14-540.html">http://jmlr.org/papers/v17/14-540.html</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/laplace.html" class="pagination-link" aria-label="Laplace Approximation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Laplace Approximation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/diagnostics.html" class="pagination-link" aria-label="Diagnostic Mode">
+        <span class="nav-page-text">Diagnostic Mode</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace_embedded.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/licenses.html b/docs/2_39/reference-manual/licenses.html
new file mode 100644
index 000000000..0e58ea5e0
--- /dev/null
+++ b/docs/2_39/reference-manual/licenses.html
@@ -0,0 +1,1101 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Licenses and Dependencies</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/reproducibility.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Licenses and Dependencies">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Usage</a></li><li class="breadcrumb-item"><a href="../reference-manual/licenses.html">Licenses and Dependencies</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#licensing.appendix" id="toc-licensing.appendix" class="nav-link active" data-scroll-target="#licensing.appendix">Licenses and Dependencies</a>
+  <ul class="collapse">
+  <li><a href="#stan-license" id="toc-stan-license" class="nav-link" data-scroll-target="#stan-license">Stan license</a></li>
+  <li><a href="#boost-license" id="toc-boost-license" class="nav-link" data-scroll-target="#boost-license">Boost license</a></li>
+  <li><a href="#eigen-license" id="toc-eigen-license" class="nav-link" data-scroll-target="#eigen-license">Eigen license</a></li>
+  <li><a href="#sundials-license" id="toc-sundials-license" class="nav-link" data-scroll-target="#sundials-license">SUNDIALS license</a></li>
+  <li><a href="#threaded-building-blocks-tbb-license" id="toc-threaded-building-blocks-tbb-license" class="nav-link" data-scroll-target="#threaded-building-blocks-tbb-license">Threaded Building Blocks (TBB) License</a></li>
+  <li><a href="#google-test-license" id="toc-google-test-license" class="nav-link" data-scroll-target="#google-test-license">Google test license</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/licenses.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Usage</a></li><li class="breadcrumb-item"><a href="../reference-manual/licenses.html">Licenses and Dependencies</a></li></ol></nav></header>
+
+
+
+
+<section id="licensing.appendix" class="level1">
+<h1>Licenses and Dependencies</h1>
+<p>Stan and its dependent libraries, are distributed under generous, freedom-respecting licenses approved by the <a href="http://opensource.org">Open Source Initiative</a>.</p>
+<p>In particular, the licenses for Stan and its dependent libraries have no “copyleft” provisions requiring applications of Stan to be open source if they are redistributed.</p>
+<p>This chapter specifies the licenses for the libraries on which Stan’s math library, language, and algorithms depend. The last tool mentioned, Google Test, is only used for testing and is not needed to run Stan.</p>
+<section id="stan-license" class="level2">
+<h2 class="anchored" data-anchor-id="stan-license">Stan license</h2>
+<p>Stan is distributed under</p>
+<ul>
+<li><a href="http://www.opensource.org/licenses/BSD-3-Clause">BSD 3-clause license (BSD New)</a></li>
+</ul>
+<p>The copyright holder of each contribution is the developer or his or her assignee.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+</section>
+<section id="boost-license" class="level2">
+<h2 class="anchored" data-anchor-id="boost-license">Boost license</h2>
+<p>Stan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the</p>
+<ul>
+<li><a href="https://opensource.org/license/bsl1-0-html/">Boost Software License version 1.0</a></li>
+</ul>
+<p>The copyright for each Boost package is held by its developers or their assignees.</p>
+</section>
+<section id="eigen-license" class="level2">
+<h2 class="anchored" data-anchor-id="eigen-license">Eigen license</h2>
+<p>Stan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the</p>
+<ul>
+<li><a href="http://opensource.org/licenses/mpl-2.0">Mozilla Public License, version 2</a></li>
+</ul>
+<p>The copyright of Eigen is owned jointly by its developers or their assignees.</p>
+</section>
+<section id="sundials-license" class="level2">
+<h2 class="anchored" data-anchor-id="sundials-license">SUNDIALS license</h2>
+<p>Stan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the</p>
+<ul>
+<li><a href="http://www.opensource.org/licenses/BSD-3-Clause">BSD 3-clause license (BSD New)</a></li>
+</ul>
+<p>The copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.</p>
+</section>
+<section id="threaded-building-blocks-tbb-license" class="level2">
+<h2 class="anchored" data-anchor-id="threaded-building-blocks-tbb-license">Threaded Building Blocks (TBB) License</h2>
+<p>Stan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the</p>
+<ul>
+<li><a href="https://opensource.org/license/apache-2-0">Apache License, version 2</a></li>
+</ul>
+<p>The copyright of TBB is owned by Intel Corporation.</p>
+</section>
+<section id="google-test-license" class="level2">
+<h2 class="anchored" data-anchor-id="google-test-license">Google test license</h2>
+<p>Stan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the</p>
+<ul>
+<li><a href="http://www.opensource.org/licenses/BSD-3-Clause">BSD 3-clause license (BSD New)</a></li>
+</ul>
+<p>The copyright of Google Test is owned by Google, Inc.</p>
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Universities or companies often own the copyright of computer programs developed by their employees.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/reproducibility.html" class="pagination-link" aria-label="Reproducibility">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Reproducibility</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/licenses.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/mcmc.html b/docs/2_39/reference-manual/mcmc.html
new file mode 100644
index 000000000..0c9a17252
--- /dev/null
+++ b/docs/2_39/reference-manual/mcmc.html
@@ -0,0 +1,1551 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>MCMC Sampling</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/analysis.html" rel="next">
+<link href="../reference-manual/removals.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="MCMC Sampling">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/warmup-epochs.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">MCMC Sampling</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#hmc.chapter" id="toc-hmc.chapter" class="nav-link active" data-scroll-target="#hmc.chapter">MCMC Sampling</a>
+  <ul class="collapse">
+  <li><a href="#hamiltonian-monte-carlo" id="toc-hamiltonian-monte-carlo" class="nav-link" data-scroll-target="#hamiltonian-monte-carlo">Hamiltonian Monte Carlo</a>
+  <ul class="collapse">
+  <li><a href="#target-density" id="toc-target-density" class="nav-link" data-scroll-target="#target-density">Target density</a></li>
+  <li><a href="#auxiliary-momentum-variable" id="toc-auxiliary-momentum-variable" class="nav-link" data-scroll-target="#auxiliary-momentum-variable">Auxiliary momentum variable</a></li>
+  <li><a href="#the-hamiltonian" id="toc-the-hamiltonian" class="nav-link" data-scroll-target="#the-hamiltonian">The Hamiltonian</a></li>
+  <li><a href="#generating-transitions" id="toc-generating-transitions" class="nav-link" data-scroll-target="#generating-transitions">Generating transitions</a></li>
+  <li><a href="#leapfrog-integrator" id="toc-leapfrog-integrator" class="nav-link" data-scroll-target="#leapfrog-integrator">Leapfrog integrator</a></li>
+  <li><a href="#metropolis-accept-step" id="toc-metropolis-accept-step" class="nav-link" data-scroll-target="#metropolis-accept-step">Metropolis accept step</a></li>
+  <li><a href="#algorithm-summary" id="toc-algorithm-summary" class="nav-link" data-scroll-target="#algorithm-summary">Algorithm summary</a></li>
+  </ul></li>
+  <li><a href="#hmc-algorithm-parameters" id="toc-hmc-algorithm-parameters" class="nav-link" data-scroll-target="#hmc-algorithm-parameters">HMC algorithm parameters</a>
+  <ul class="collapse">
+  <li><a href="#integration-time" id="toc-integration-time" class="nav-link" data-scroll-target="#integration-time">Integration time</a></li>
+  <li><a href="#automatic-parameter-tuning" id="toc-automatic-parameter-tuning" class="nav-link" data-scroll-target="#automatic-parameter-tuning">Automatic parameter tuning</a></li>
+  <li><a href="#discretization-interval-adaptation-parameters" id="toc-discretization-interval-adaptation-parameters" class="nav-link" data-scroll-target="#discretization-interval-adaptation-parameters">Discretization-interval adaptation parameters</a></li>
+  <li><a href="#step-size-jitter" id="toc-step-size-jitter" class="nav-link" data-scroll-target="#step-size-jitter">Step-size jitter</a></li>
+  <li><a href="#euclidean-metric" id="toc-euclidean-metric" class="nav-link" data-scroll-target="#euclidean-metric">Euclidean metric</a></li>
+  <li><a href="#nuts-and-its-configuration" id="toc-nuts-and-its-configuration" class="nav-link" data-scroll-target="#nuts-and-its-configuration">NUTS and its configuration</a></li>
+  </ul></li>
+  <li><a href="#sampling-without-parameters" id="toc-sampling-without-parameters" class="nav-link" data-scroll-target="#sampling-without-parameters">Sampling without parameters</a></li>
+  <li><a href="#general-config.section" id="toc-general-config.section" class="nav-link" data-scroll-target="#general-config.section">General configuration options</a>
+  <ul class="collapse">
+  <li><a href="#random-number-generator" id="toc-random-number-generator" class="nav-link" data-scroll-target="#random-number-generator">Random number generator</a></li>
+  <li><a href="#initialization" id="toc-initialization" class="nav-link" data-scroll-target="#initialization">Initialization</a></li>
+  </ul></li>
+  <li><a href="#divergent-transitions" id="toc-divergent-transitions" class="nav-link" data-scroll-target="#divergent-transitions">Divergent transitions</a>
+  <ul class="collapse">
+  <li><a href="#diagnosing-and-eliminating-divergences" id="toc-diagnosing-and-eliminating-divergences" class="nav-link" data-scroll-target="#diagnosing-and-eliminating-divergences">Diagnosing and eliminating divergences</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/mcmc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">MCMC Sampling</a></li></ol></nav></header>
+
+
+
+
+<section id="hmc.chapter" class="level1">
+<h1>MCMC Sampling</h1>
+<p>This chapter presents the two Markov chain Monte Carlo (MCMC) algorithms used in Stan, the Hamiltonian Monte Carlo (HMC) algorithm and its adaptive variant the no-U-turn sampler (NUTS), along with details of their implementation and configuration.</p>
+<section id="hamiltonian-monte-carlo" class="level2">
+<h2 class="anchored" data-anchor-id="hamiltonian-monte-carlo">Hamiltonian Monte Carlo</h2>
+<p>Hamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., <span class="citation" data-cites="Betancourt-Girolami:2013">Betancourt and Girolami (<a href="#ref-Betancourt-Girolami:2013" role="doc-biblioref">2013</a>)</span>, <span class="citation" data-cites="Neal:2011">Neal (<a href="#ref-Neal:2011" role="doc-biblioref">2011</a>)</span> for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.</p>
+<p>This section translates the presentation of HMC by <span class="citation" data-cites="Betancourt-Girolami:2013">Betancourt and Girolami (<a href="#ref-Betancourt-Girolami:2013" role="doc-biblioref">2013</a>)</span> into the notation of <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span>.</p>
+<section id="target-density" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="target-density">Target density</h3>
+<p>The goal of sampling is to draw from a density <span class="math inline">\(p(\theta)\)</span> for parameters <span class="math inline">\(\theta\)</span>. This is typically a Bayesian posterior <span class="math inline">\(p(\theta|y)\)</span> given data <span class="math inline">\(y\)</span>, and in particular, a Bayesian posterior coded as a Stan program.</p>
+</section>
+<section id="auxiliary-momentum-variable" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="auxiliary-momentum-variable">Auxiliary momentum variable</h3>
+<p>HMC introduces auxiliary momentum variables <span class="math inline">\(\rho\)</span> and draws from a joint density</p>
+<p><span class="math display">\[
+p(\rho, \theta) = p(\rho | \theta) p(\theta).
+\]</span></p>
+<p>In most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters <span class="math inline">\(\theta\)</span>,</p>
+<p><span class="math display">\[
+\rho \sim \mathsf{MultiNormal}(0, M).
+\]</span></p>
+<p><span class="math inline">\(M\)</span> is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see <span class="citation" data-cites="Betancourt:2017">Betancourt (<a href="#ref-Betancourt:2017" role="doc-biblioref">2017</a>)</span> for details.</p>
+<p>By default Stan sets <span class="math inline">\(M^{-1}\)</span> equal to a diagonal estimate of the covariance computed during warmup.</p>
+</section>
+<section id="the-hamiltonian" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="the-hamiltonian">The Hamiltonian</h3>
+<p>The joint density <span class="math inline">\(p(\rho, \theta)\)</span> defines a Hamiltonian</p>
+<p><span class="math display">\[
+\begin{array}{rcl}
+H(\rho, \theta) &amp; = &amp; - \log p(\rho, \theta)
+\\[3pt]
+&amp; = &amp; - \log p(\rho | \theta) - \log p(\theta).
+\\[3pt]
+&amp; = &amp; T(\rho | \theta) + V(\theta),
+\end{array}
+\]</span></p>
+<p>where the term</p>
+<p><span class="math display">\[
+T(\rho | \theta) = - \log p(\rho | \theta)
+\]</span></p>
+<p>is called the “kinetic energy” and the term</p>
+<p><span class="math display">\[
+V(\theta) = - \log p(\theta)
+\]</span></p>
+<p>is called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.</p>
+</section>
+<section id="generating-transitions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="generating-transitions">Generating transitions</h3>
+<p>Starting from the current value of the parameters <span class="math inline">\(\theta\)</span>, a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.</p>
+<p>First, a value for the momentum is drawn independently of the current parameter values,</p>
+<p><span class="math display">\[
+\rho \sim \mathsf{MultiNormal}(0, M).
+\]</span></p>
+<p>Thus momentum does not persist across iterations.</p>
+<p>Next, the joint system <span class="math inline">\((\theta,\rho)\)</span> made up of the current parameter values <span class="math inline">\(\theta\)</span> and new momentum <span class="math inline">\(\rho\)</span> is evolved via Hamilton’s equations,</p>
+<p><span class="math display">\[
+\begin{array}{rcccl}
+\displaystyle
+\frac{d\theta}{dt}
+&amp; = &amp;
+\displaystyle
++ \frac{\partial H}{\partial \rho}
+&amp; = &amp;
+\displaystyle
++ \frac{\partial T}{\partial \rho}
+\\[12pt]
+\displaystyle
+\frac{d\rho}{dt}
+&amp; = &amp;
+\displaystyle
+- \frac{\partial H}{\partial \theta }
+&amp; = &amp;
+\displaystyle
+- \frac{\partial T}{\partial \theta}
+- \frac{\partial V}{\partial \theta}.
+\end{array}
+\]</span></p>
+<p>With the momentum density being independent of the target density, i.e., <span class="math inline">\(p(\rho | \theta) = p(\rho)\)</span>, the first term in the momentum time derivative, <span class="math inline">\({\partial T} / {\partial \theta}\)</span> is zero, yielding the pair time derivatives</p>
+<p><span class="math display">\[
+\begin{array}{rcl}
+\frac{d \theta}{d t} &amp; = &amp; +\frac{\partial T}{\partial \rho}
+\\[2pt]
+\frac{d \rho}{d t} &amp; = &amp; -\frac{\partial V}{\partial \theta}.
+\end{array}
+\]</span></p>
+</section>
+<section id="leapfrog-integrator" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="leapfrog-integrator">Leapfrog integrator</h3>
+<p>The last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.</p>
+<p>Like most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval <span class="math inline">\(\epsilon\)</span>. The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values <span class="math inline">\(\theta\)</span> or previous momentum value.</p>
+<p><span class="math display">\[
+\rho \sim \mathsf{MultiNormal}(0, M).
+\]</span> It then alternates half-step updates of the momentum and full-step updates of the position.</p>
+<p><span class="math display">\[
+\begin{array}{rcl}
+\rho &amp; \leftarrow
+     &amp; \rho \, - \, \frac{\epsilon}{2} \frac{\partial V}{\partial \theta}
+\\[6pt]
+\theta &amp; \leftarrow
+       &amp; \theta \, + \, \epsilon \, M^{-1} \, \rho
+\\[6pt]
+\rho &amp; \leftarrow
+     &amp; \rho \, - \, \frac{\epsilon}{2} \frac{\partial V}{\partial \theta}.
+\end{array}
+\]</span></p>
+<p>By applying <span class="math inline">\(L\)</span> leapfrog steps, a total of <span class="math inline">\(L \, \epsilon\)</span> time is simulated. The resulting state at the end of the simulation (<span class="math inline">\(L\)</span> repetitions of the above three steps) will be denoted <span class="math inline">\((\rho^{*}, \theta^{*})\)</span>.</p>
+<p>The leapfrog integrator’s error is on the order of <span class="math inline">\(\epsilon^3\)</span> per step and <span class="math inline">\(\epsilon^2\)</span> globally, where <span class="math inline">\(\epsilon\)</span> is the time interval (also known as the step size); <span class="citation" data-cites="LeimkuhlerReich:2004">Leimkuhler and Reich (<a href="#ref-LeimkuhlerReich:2004" role="doc-biblioref">2004</a>)</span> provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.</p>
+</section>
+<section id="metropolis-accept-step" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="metropolis-accept-step">Metropolis accept step</h3>
+<p>If the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal <span class="math inline">\((\rho^{*}, \theta^{*})\)</span> generated by transitioning from <span class="math inline">\((\rho, \theta)\)</span> is</p>
+<p><span class="math display">\[
+\min \!
+\left(
+1,
+\ \exp \! \left( H(\rho, \theta) - H(\rho^{*}, \theta^{*}) \right)
+\right).
+\]</span></p>
+<p>If the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.</p>
+</section>
+<section id="algorithm-summary" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="algorithm-summary">Algorithm summary</h3>
+<p>The Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters <span class="math inline">\(\theta\)</span>; in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter <span class="math inline">\(\theta\)</span> is updated using the leapfrog integrator with discretization time <span class="math inline">\(\epsilon\)</span> and number of steps <span class="math inline">\(L\)</span> according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state <span class="math inline">\((\theta^{*}, \rho^{*})\)</span> or keep the existing state.</p>
+</section>
+</section>
+<section id="hmc-algorithm-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="hmc-algorithm-parameters">HMC algorithm parameters</h2>
+<p>The Hamiltonian Monte Carlo algorithm has three parameters which must be set,</p>
+<ul>
+<li>discretization time <span class="math inline">\(\epsilon\)</span>,</li>
+<li>metric <span class="math inline">\(M\)</span>, and</li>
+<li>number of steps taken <span class="math inline">\(L\)</span>.</li>
+</ul>
+<p>In practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters <span class="citation" data-cites="Neal:2011">Neal (<a href="#ref-Neal:2011" role="doc-biblioref">2011</a>)</span>, <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>.</p>
+<p>If <span class="math inline">\(\epsilon\)</span> is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If <span class="math inline">\(\epsilon\)</span> is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.</p>
+<p>If <span class="math inline">\(L\)</span> is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If <span class="math inline">\(L\)</span> is too large, the algorithm will do too much work on each iteration.</p>
+<p>If the inverse metric <span class="math inline">\(M^{-1}\)</span> is a poor estimate of the posterior covariance, the step size <span class="math inline">\(\epsilon\)</span> must be kept small to maintain arithmetic precision. This would lead to a large <span class="math inline">\(L\)</span> to compensate.</p>
+<section id="integration-time" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="integration-time">Integration time</h3>
+<p>The actual integration time is <span class="math inline">\(L \, \epsilon\)</span>, a function of number of steps. Some interfaces to Stan set an approximate integration time <span class="math inline">\(t\)</span> and the discretization interval (step size) <span class="math inline">\(\epsilon\)</span>. In these cases, the number of steps will be rounded down as</p>
+<p><span class="math display">\[
+L = \left\lfloor \frac{t}{\epsilon} \right\rfloor.
+\]</span></p>
+<p>and the actual integration time will still be <span class="math inline">\(L \, \epsilon\)</span>.</p>
+</section>
+<section id="automatic-parameter-tuning" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="automatic-parameter-tuning">Automatic parameter tuning</h3>
+<p>Stan is able to automatically optimize <span class="math inline">\(\epsilon\)</span> to match an acceptance-rate target, able to estimate <span class="math inline">\(M\)</span> based on warmup sample iterations, and able to dynamically adapt <span class="math inline">\(L\)</span> on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>.</p>
+<p><strong>Warmup Epochs Figure.</strong> <a id="adaptation.figure"></a> <em>Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.</em></p>
+<p><img src="img/warmup-epochs.png" class="img-fluid"></p>
+<p>When adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the <a href="#adaptation.figure">warmup adaptation figure</a>, with two <em>fast</em> intervals surrounding a series of growing <em>slow</em> intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.</p>
+<p>The motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.</p>
+<ol type="1">
+<li><p>In the initial fast interval the chain is allowed to converge towards the typical set,<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> with only parameters that can learn from local information adapted.</p></li>
+<li><p>After this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.</p></li>
+<li><p>Lastly, the fast parameters are allowed to adapt to the final update of the slow parameters.</p></li>
+</ol>
+<p>These intervals may be controlled through the following configuration parameters, all of which must be positive integers:</p>
+<p><strong>Adaptation Parameters Table.</strong> <em>The parameters controlling adaptation and their default values.</em></p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 25%">
+<col style="width: 64%">
+<col style="width: 10%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: left;">parameter</th>
+<th style="text-align: left;">description</th>
+<th style="text-align: center;">default</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><em>initial buffer</em></td>
+<td style="text-align: left;">width of initial fast adaptation interval</td>
+<td style="text-align: center;">75</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><em>term buffer</em></td>
+<td style="text-align: left;">width of final fast adaptation interval</td>
+<td style="text-align: center;">50</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><em>window</em></td>
+<td style="text-align: left;">initial width of slow adaptation interval</td>
+<td style="text-align: center;">25</td>
+</tr>
+</tbody>
+</table>
+</section>
+<section id="discretization-interval-adaptation-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="discretization-interval-adaptation-parameters">Discretization-interval adaptation parameters</h3>
+<p>Stan’s HMC algorithms utilize dual averaging <span class="citation" data-cites="Nesterov:2009">Nesterov (<a href="#ref-Nesterov:2009" role="doc-biblioref">2009</a>)</span> to optimize the step size.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>This warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>. In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>.</p>
+<p>The full set of dual-averaging parameters are:</p>
+<p><strong>Step Size Adaptation Parameters Table</strong> <em>The parameters controlling step size adaptation, with constraints and default values.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: center;">parameter</th>
+<th style="text-align: left;">description</th>
+<th style="text-align: left;">constraint</th>
+<th style="text-align: center;">default</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;"><code>delta</code></td>
+<td style="text-align: left;">target Metropolis acceptance rate</td>
+<td style="text-align: left;">[0, 1]</td>
+<td style="text-align: center;">0.8</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>gamma</code></td>
+<td style="text-align: left;">adaptation regularization scale</td>
+<td style="text-align: left;">(0, infty)</td>
+<td style="text-align: center;">0.05</td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>kappa</code></td>
+<td style="text-align: left;">adaptation relaxation exponent</td>
+<td style="text-align: left;">(0, infty)</td>
+<td style="text-align: center;">0.75</td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>t_0</code></td>
+<td style="text-align: left;">adaptation iteration offset</td>
+<td style="text-align: left;">(0, infty)</td>
+<td style="text-align: center;">10</td>
+</tr>
+</tbody>
+</table>
+<p>By setting the target acceptance parameter <span class="math inline">\(\delta\)</span> to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of <span class="math inline">\(\delta\)</span> will also allow some models that would otherwise get stuck to overcome their blockages.</p>
+</section>
+<section id="step-size-jitter" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="step-size-jitter">Step-size jitter</h3>
+<p>All implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.</p>
+<p>Small step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See <span class="citation" data-cites="Neal:2011">Neal (<a href="#ref-Neal:2011" role="doc-biblioref">2011</a>)</span> for further discussion of step-size jittering.</p>
+</section>
+<section id="euclidean-metric" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="euclidean-metric">Euclidean metric</h3>
+<p>All HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a <em>mass matrix</em> or, more formally, a <em>metric</em> <span class="citation" data-cites="Betancourt:2017">Betancourt (<a href="#ref-Betancourt:2017" role="doc-biblioref">2017</a>)</span>.</p>
+<p>If the metric is constant then the resulting implementation is known as <em>Euclidean</em> HMC. Stan allows a choice among three Euclidean HMC implementations,</p>
+<ul>
+<li>a unit metric (diagonal matrix of ones),</li>
+<li>a diagonal metric (diagonal matrix with positive diagonal entries), and</li>
+<li>a dense metric (a dense, symmetric positive definite matrix)</li>
+</ul>
+<p>to be configured by the user.</p>
+<p>If the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the <a href="#adaptation.figure">warmup adaptation stages figure</a>). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.</p>
+<p>If the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.</p>
+<p>Variances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.</p>
+<section id="warmup-times-and-estimating-the-metric" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="warmup-times-and-estimating-the-metric">Warmup times and estimating the metric</h4>
+<p>The metric can compensate for linear (i.e.&nbsp;global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.</p>
+<p>In complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e.&nbsp;local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.</p>
+</section>
+<section id="nonlinearity" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="nonlinearity">Nonlinearity</h4>
+<p>The metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>One of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.</p>
+</section>
+<section id="dense-vs.-diagonal-metrics" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="dense-vs.-diagonal-metrics">Dense vs.&nbsp;diagonal metrics</h4>
+<p>Statistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.</p>
+</section>
+<section id="warmup-times-and-curvature" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="warmup-times-and-curvature">Warmup times and curvature</h4>
+<p>MCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.</p>
+<p>This only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.</p>
+<p>Poor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.</p>
+</section>
+</section>
+<section id="nuts-and-its-configuration" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="nuts-and-its-configuration">NUTS and its configuration</h3>
+<p>The no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., <span class="citation" data-cites="RobertsEtAl:1997">Roberts, Gelman, and Gilks (<a href="#ref-RobertsEtAl:1997" role="doc-biblioref">1997</a>)</span>) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>.</p>
+<p>NUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either</p>
+<ul>
+<li>the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or</li>
+<li>the depth of the completed tree hits the maximum depth allowed.</li>
+</ul>
+<p>Rather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory <span class="citation" data-cites="Betancourt:2016">Betancourt (<a href="#ref-Betancourt:2016" role="doc-biblioref">2016b</a>)</span>.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
+<p>Configuring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.</p>
+<p>Both the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as <code>treedepth__</code> and <code>n_leapfrog__</code>, respectively. Because the final subtree may only be partially constructed, these two will always satisfy</p>
+<p><span class="math display">\[
+2^{\mathrm{treedepth} - 1} - 1
+\ &lt; \
+N_{\mathrm{leapfrog}}
+\ \le \
+2^{\mathrm{treedepth} } - 1.
+\]</span></p>
+<p>Tree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.</p>
+</section>
+</section>
+<section id="sampling-without-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="sampling-without-parameters">Sampling without parameters</h2>
+<p>In some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.</p>
+<p>For example, to generate a sequence of <span class="math inline">\(N\)</span> draws from a binomial with trials <span class="math inline">\(K\)</span> and chance of success <span class="math inline">\(\theta\)</span>, the following program suffices.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=K&gt; y;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>    y[n] = binomial_rng(K, theta);</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.</p>
+<p>Most models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, <code>theta</code> in the example above could be declared as a parameter and initialized as a parameter.</p>
+</section>
+<section id="general-config.section" class="level2">
+<h2 class="anchored" data-anchor-id="general-config.section">General configuration options</h2>
+<p>Stan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the <a href="../reference-manual/optimization.html">optimization algorithms chapter</a>, and the <a href="../reference-manual/diagnostics.html">diagnostics chapter</a>.</p>
+<section id="random-number-generator" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="random-number-generator">Random number generator</h3>
+<p>The random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.</p>
+<p>Stan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.</p>
+<section id="replication" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="replication">Replication</h4>
+<p>Together, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.</p>
+</section>
+</section>
+<section id="initialization" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="initialization">Initialization</h3>
+<p>The initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.</p>
+<section id="user-defined-initialization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="user-defined-initialization">User-defined initialization</h4>
+<p>If the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).</p>
+</section>
+<section id="system-constant-zero-initialization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="system-constant-zero-initialization">System constant zero initialization</h4>
+<p>It is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.</p>
+</section>
+<section id="system-random-initialization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="system-random-initialization">System random initialization</h4>
+<p>Random initialization by default initializes the parameter values with values drawn at random from a <span class="math inline">\(\mathsf{Uniform}(-2, 2)\)</span> distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.</p>
+<p>Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in <span class="math inline">\((-2, 2)\)</span>, variables constrained to be positive are initialized roughly in <span class="math inline">\((0.14, 7.4)\)</span>, variables constrained to fall between 0 and 1 are initialized with values roughly in <span class="math inline">\((0.12, 0.88)\)</span>.</p>
+</section>
+</section>
+</section>
+<section id="divergent-transitions" class="level2">
+<h2 class="anchored" data-anchor-id="divergent-transitions">Divergent transitions</h2>
+<p>The Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.</p>
+<p>In Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. <span class="citation" data-cites="LeimkuhlerReich:2004">Leimkuhler and Reich (<a href="#ref-LeimkuhlerReich:2004" role="doc-biblioref">2004</a>)</span> provide details and error analysis for the leapfrog algorithm.</p>
+<p>A divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,<a href="#fn5" class="footnote-ref" id="fnref5" role="doc-noteref"><sup>5</sup></a> the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. <span class="citation" data-cites="Betancourt:2016b">Betancourt (<a href="#ref-Betancourt:2016b" role="doc-biblioref">2016a</a>)</span> provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.</p>
+<p>The Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.</p>
+<p>If the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.</p>
+<p>The primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.</p>
+<section id="diagnosing-and-eliminating-divergences" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="diagnosing-and-eliminating-divergences">Diagnosing and eliminating divergences</h3>
+<p>In some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.</p>
+<p>Before reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Betancourt:2016b" class="csl-entry" role="listitem">
+Betancourt, Michael. 2016a. <span>“Diagnosing Suboptimal Cotangent Disintegrations in <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>arXiv</em> 1604.00695. <a href="https://arxiv.org/abs/1604.00695">https://arxiv.org/abs/1604.00695</a>.
+</div>
+<div id="ref-Betancourt:2016" class="csl-entry" role="listitem">
+———. 2016b. <span>“Identifying the Optimal Integration Time in <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>arXiv</em> 1601.00225. <a href="https://arxiv.org/abs/1601.00225">https://arxiv.org/abs/1601.00225</a>.
+</div>
+<div id="ref-Betancourt:2017" class="csl-entry" role="listitem">
+———. 2017. <span>“A Conceptual Introduction to <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>arXiv</em> 1701.02434. <a href="https://arxiv.org/abs/1701.02434">https://arxiv.org/abs/1701.02434</a>.
+</div>
+<div id="ref-Betancourt-Girolami:2013" class="csl-entry" role="listitem">
+Betancourt, Michael, and Mark Girolami. 2013. <span>“<span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo for Hierarchical Models.”</span> <em>arXiv</em> 1312.0906. <a href="http://arxiv.org/abs/1312.0906">http://arxiv.org/abs/1312.0906</a>.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-Hoffman-Gelman:2014" class="csl-entry" role="listitem">
+Hoffman, Matthew D., and Andrew Gelman. 2014. <span>“<span>T</span>he <span>N</span>o-<span>U</span>-<span>T</span>urn <span>S</span>ampler: <span>A</span>daptively <span>S</span>etting <span>P</span>ath <span>L</span>engths in <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>Journal of Machine Learning Research</em> 15: 1593–623. <a href="http://jmlr.org/papers/v15/hoffman14a.html">http://jmlr.org/papers/v15/hoffman14a.html</a>.
+</div>
+<div id="ref-LeimkuhlerReich:2004" class="csl-entry" role="listitem">
+Leimkuhler, Benedict, and Sebastian Reich. 2004. <em>Simulating Hamiltonian Dynamics</em>. Cambridge: Cambridge University Press.
+</div>
+<div id="ref-Neal:2011" class="csl-entry" role="listitem">
+Neal, Radford. 2011. <span>“<span>MCMC</span> Using <span>H</span>amiltonian Dynamics.”</span> In <em>Handbook of <span>M</span>arkov Chain <span>M</span>onte <span>C</span>arlo</em>, edited by Steve Brooks, Andrew Gelman, Galin L. Jones, and Xiao-Li Meng, 116–62. Chapman; Hall/CRC.
+</div>
+<div id="ref-Nesterov:2009" class="csl-entry" role="listitem">
+Nesterov, Y. 2009. <span>“Primal-Dual Subgradient Methods for Convex Problems.”</span> <em>Mathematical Programming</em> 120 (1): 221–59.
+</div>
+<div id="ref-RobertsEtAl:1997" class="csl-entry" role="listitem">
+Roberts, G. O., Andrew Gelman, and Walter R. Gilks. 1997. <span>“Weak Convergence and Optimal Scaling of Random Walk <span>M</span>etropolis Algorithms.”</span> <em>Annals of Applied Probability</em> 7 (1): 110–20.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The typical set is a concept borrowed from information theory and refers to the neighborhood (or neighborhoods in multimodal models) of substantial posterior probability mass through which the Markov chain will travel in equilibrium.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>This optimization of step size during adaptation of the sampler should not be confused with running Stan’s optimization method.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>In Riemannian HMC the metric compensates for nonlinear correlations.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn4"><p>Stan previously used slice sampling along the trajectory, following the original NUTS paper of <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>.<a href="#fnref4" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn5"><p>The current default threshold is a factor of <span class="math inline">\(10^3\)</span>, whereas when the leapfrog integrator is working properly, the divergences will be around <span class="math inline">\(10^{-7}\)</span> and do not compound due to the symplectic nature of the leapfrog integrator.<a href="#fnref5" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/removals.html" class="pagination-link" aria-label="Removed Features">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Removed Features</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/analysis.html" class="pagination-link" aria-label="Posterior Analysis">
+        <span class="nav-page-text">Posterior Analysis</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/mcmc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/optimization.html b/docs/2_39/reference-manual/optimization.html
new file mode 100644
index 000000000..074423a20
--- /dev/null
+++ b/docs/2_39/reference-manual/optimization.html
@@ -0,0 +1,1213 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Optimization</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/pathfinder.html" rel="next">
+<link href="../reference-manual/analysis.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Optimization">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/optimization.html">Optimization</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#optimization.chapter" id="toc-optimization.chapter" class="nav-link active" data-scroll-target="#optimization.chapter">Optimization</a>
+  <ul class="collapse">
+  <li><a href="#general-configuration" id="toc-general-configuration" class="nav-link" data-scroll-target="#general-configuration">General configuration</a></li>
+  <li><a href="#bfgs-and-l-bfgs-configuration" id="toc-bfgs-and-l-bfgs-configuration" class="nav-link" data-scroll-target="#bfgs-and-l-bfgs-configuration">BFGS and L-BFGS configuration</a>
+  <ul class="collapse">
+  <li><a href="#convergence-monitoring" id="toc-convergence-monitoring" class="nav-link" data-scroll-target="#convergence-monitoring">Convergence monitoring</a></li>
+  <li><a href="#initial-step-size" id="toc-initial-step-size" class="nav-link" data-scroll-target="#initial-step-size">Initial step size</a></li>
+  <li><a href="#l-bfgs-history-size" id="toc-l-bfgs-history-size" class="nav-link" data-scroll-target="#l-bfgs-history-size">L-BFGS history size</a></li>
+  </ul></li>
+  <li><a href="#writing-models-for-optimization" id="toc-writing-models-for-optimization" class="nav-link" data-scroll-target="#writing-models-for-optimization">Writing models for optimization</a>
+  <ul class="collapse">
+  <li><a href="#constrained-vs.-unconstrained-parameters" id="toc-constrained-vs.-unconstrained-parameters" class="nav-link" data-scroll-target="#constrained-vs.-unconstrained-parameters">Constrained vs.&nbsp;unconstrained parameters</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/optimization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/optimization.html">Optimization</a></li></ol></nav></header>
+
+
+
+
+<section id="optimization.chapter" class="level1">
+<h1>Optimization</h1>
+<p>Stan provides optimization algorithms which find modes of the density specified by a Stan program. Such modes may be used as parameter estimates or as the basis of approximations to a Bayesian posterior.</p>
+<p>Stan provides three different optimizers, a Newton optimizer, and two related quasi-Newton algorithms, BFGS and L-BFGS; see <span class="citation" data-cites="NocedalWright:2006">Nocedal and Wright (<a href="#ref-NocedalWright:2006" role="doc-biblioref">2006</a>)</span> for thorough description and analysis of all of these algorithms. The L-BFGS algorithm is the default optimizer. Newton’s method is the least efficient of the three, but has the advantage of setting its own stepsize.</p>
+<section id="general-configuration" class="level2">
+<h2 class="anchored" data-anchor-id="general-configuration">General configuration</h2>
+<p>All of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.</p>
+<p>All of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.</p>
+<p>All of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.</p>
+</section>
+<section id="bfgs-and-l-bfgs-configuration" class="level2">
+<h2 class="anchored" data-anchor-id="bfgs-and-l-bfgs-configuration">BFGS and L-BFGS configuration</h2>
+<section id="convergence-monitoring" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="convergence-monitoring">Convergence monitoring</h3>
+<p>Convergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.</p>
+<section id="parameter-convergence" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="parameter-convergence">Parameter convergence</h4>
+<p>The parameters <span class="math inline">\(\theta_i\)</span> in iteration <span class="math inline">\(i\)</span> are considered to have converged with respect to tolerance <code>tol_param</code> if</p>
+<p><span class="math display">\[
+|| \theta_{i} - \theta_{i-1} || &lt; \mathtt{tol\_param}.
+\]</span></p>
+</section>
+<section id="density-convergence" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="density-convergence">Density convergence</h4>
+<p>The (unnormalized) log density <span class="math inline">\(\log p(\theta_{i}|y)\)</span> for the parameters <span class="math inline">\(\theta_i\)</span> in iteration <span class="math inline">\(i\)</span> given data <span class="math inline">\(y\)</span> is considered to have converged with respect to tolerance <code>tol_obj</code> if</p>
+<p><span class="math display">\[
+\left| \log p(\theta_{i}|y) - \log p(\theta_{i-1}|y) \right| &lt;
+\mathtt{tol\_obj}.
+\]</span></p>
+<p>The log density is considered to have converged to within relative tolerance <code>tol_rel_obj</code> if</p>
+<p><span class="math display">\[
+\frac{\left| \log p(\theta_{i}|y) - \log p(\theta_{i-1}|y) \right|}{\
+  \max\left(\left| \log p(\theta_{i}|y)\right|,\left| \log
+      p(\theta_{i-1}|y)\right|,1.0\right)}
+&lt; \mathtt{tol\_rel\_obj} * \epsilon.
+\]</span></p>
+</section>
+<section id="gradient-convergence" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="gradient-convergence">Gradient convergence</h4>
+<p>The gradient is considered to have converged to 0 relative to a specified tolerance <code>tol_grad</code> if</p>
+<p><span class="math display">\[
+|| g_{i} || &lt; \mathtt{tol\_grad},
+\]</span> where <span class="math inline">\(\nabla_{\theta}\)</span> is the gradient operator with respect to <span class="math inline">\(\theta\)</span> and <span class="math inline">\(g_{i} = \nabla_{\theta} \log p(\theta | y)\)</span> is the gradient at iteration <span class="math inline">\(i\)</span> evaluated at <span class="math inline">\(\theta^{(i)}\)</span>, the value on the <span class="math inline">\(i\)</span>-th posterior iteration.</p>
+<p>The gradient is considered to have converged to 0 relative to a specified relative tolerance <code>tol_rel_grad</code> if</p>
+<p><span class="math display">\[
+\frac{g_{i}^T \hat{H}_{i}^{-1} g_{i} }{ \max\left(\left|\log
+p(\theta_{i}|y)\right|,1.0\right) }
+\ &lt; \
+\mathtt{tol\_rel\_grad} * \epsilon,
+\]</span></p>
+<p>where <span class="math inline">\(\hat{H}_{i}\)</span> is the estimate of the Hessian at iteration <span class="math inline">\(i\)</span>, <span class="math inline">\(|u|\)</span> is the absolute value (L1 norm) of <span class="math inline">\(u\)</span>, <span class="math inline">\(||u||\)</span> is the vector length (L2 norm) of <span class="math inline">\(u\)</span>, and <span class="math inline">\(\epsilon \approx 2e-16\)</span> is machine precision.</p>
+</section>
+</section>
+<section id="initial-step-size" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="initial-step-size">Initial step size</h3>
+<p>The initial step size parameter <span class="math inline">\(\alpha\)</span> for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize <span class="math inline">\(\alpha\)</span> to be the roughly equal to the <span class="math inline">\(\alpha\)</span> used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune <span class="math inline">\(\alpha\)</span> can save some real time.</p>
+</section>
+<section id="l-bfgs-history-size" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="l-bfgs-history-size">L-BFGS history size</h3>
+<p>L-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.</p>
+<p>If L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.</p>
+</section>
+</section>
+<section id="writing-models-for-optimization" class="level2">
+<h2 class="anchored" data-anchor-id="writing-models-for-optimization">Writing models for optimization</h2>
+<section id="constrained-vs.-unconstrained-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constrained-vs.-unconstrained-parameters">Constrained vs.&nbsp;unconstrained parameters</h3>
+<p>For constrained optimization problems, for instance, with a standard deviation parameter <span class="math inline">\(\sigma\)</span> constrained so that <span class="math inline">\(\sigma &gt; 0\)</span>, it can be much more efficient to declare a parameter <code>sigma</code> with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward <span class="math inline">\(-\infty\)</span> on the <span class="math inline">\(\log \sigma\)</span> scale.</p>
+<p>With unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[M] sigma;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and using the default random initialization which is <span class="math inline">\(\mathsf{Uniform}(-2, 2)\)</span> on the unconstrained scale means that there is only a <span class="math inline">\(2^{-M}\)</span> chance that the initialization will be within support.</p>
+<p>For any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-NocedalWright:2006" class="csl-entry" role="listitem">
+Nocedal, Jorge, and Stephen J. Wright. 2006. <em>Numerical Optimization</em>. Second. Berlin: Springer-<span>V</span>erlag.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/analysis.html" class="pagination-link" aria-label="Posterior Analysis">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Posterior Analysis</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/pathfinder.html" class="pagination-link" aria-label="Pathfinder">
+        <span class="nav-page-text">Pathfinder</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/optimization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/pathfinder.html b/docs/2_39/reference-manual/pathfinder.html
new file mode 100644
index 000000000..e7ce67d5b
--- /dev/null
+++ b/docs/2_39/reference-manual/pathfinder.html
@@ -0,0 +1,1117 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Pathfinder</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/variational.html" rel="next">
+<link href="../reference-manual/optimization.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Pathfinder">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/pathfinder.html">Pathfinder</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#pathfinder" id="toc-pathfinder" class="nav-link active" data-scroll-target="#pathfinder">Pathfinder</a>
+  <ul class="collapse">
+  <li><a href="#diagnosing-pathfinder" id="toc-diagnosing-pathfinder" class="nav-link" data-scroll-target="#diagnosing-pathfinder">Diagnosing Pathfinder</a></li>
+  <li><a href="#using-pathfinder-for-initializing-mcmc" id="toc-using-pathfinder-for-initializing-mcmc" class="nav-link" data-scroll-target="#using-pathfinder-for-initializing-mcmc">Using Pathfinder for initializing MCMC</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/pathfinder.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/pathfinder.html">Pathfinder</a></li></ol></nav></header>
+
+
+
+
+<section id="pathfinder" class="level1">
+<h1>Pathfinder</h1>
+<p>Stan supports the Pathfinder algorithm <span class="citation" data-cites="zhang_pathfinder:2022">(<a href="#ref-zhang_pathfinder:2022" role="doc-biblioref">Zhang et al. 2022</a>)</span>. Pathfinder is a variational method for approximately sampling from differentiable log densities. Starting from a random initialization, Pathfinder locates normal approximations to the target density along a quasi-Newton optimization path, with local covariance estimated using the negative inverse Hessian estimates produced by the LBFGS optimizer. Pathfinder returns draws from the Gaussian approximation with the lowest estimated Kullback-Leibler (KL) divergence to the true posterior.</p>
+<p>Stan provides two versions of the Pathfinder algorithm: single-path Pathfinder and multi-path Pathfinder. Single-path Pathfinder generates a set of approximate draws from one run of the basic Pathfinder algorithm. Multi-path Pathfinder uses importance resampling over the draws from multiple runs of Pathfinder. This better matches non-normal target densities and also mitigates the problem of L-BFGS getting stuck at local optima or in saddle points on plateaus. Compared to ADVI and short dynamic HMC runs, Pathfinder requires one to two orders of magnitude fewer log density and gradient evaluations, with greater reductions for more challenging posteriors. While the evaluations by <span class="citation" data-cites="zhang_pathfinder:2022">Zhang et al. (<a href="#ref-zhang_pathfinder:2022" role="doc-biblioref">2022</a>)</span> found that single-path and multi-path Pathfinder outperform ADVI for most of the models in the PosteriorDB <span class="citation" data-cites="Magnusson+etal:2024:posteriordb">(<a href="#ref-Magnusson+etal:2024:posteriordb" role="doc-biblioref">Magnusson et al. 2024</a>)</span> evaluation set, we recognize the need for further experiments on a wider range of models.</p>
+<section id="diagnosing-pathfinder" class="level2">
+<h2 class="anchored" data-anchor-id="diagnosing-pathfinder">Diagnosing Pathfinder</h2>
+<p>Pathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-<span class="math inline">\(\hat{k}\)</span> diagnostic <span class="citation" data-cites="Vehtari+etal:2024:PSIS">(<a href="#ref-Vehtari+etal:2024:PSIS" role="doc-biblioref">Vehtari et al. 2024</a>)</span> to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably <span class="citation" data-cites="Vehtari+etal:2024:PSIS">(<a href="#ref-Vehtari+etal:2024:PSIS" role="doc-biblioref">Vehtari et al. 2024, sec. 3</a>)</span>, which is the first requirement for reliable resampling. If estimated Pareto-<span class="math inline">\(\hat{k}\)</span> for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest <span class="citation" data-cites="Vehtari+etal:2024:PSIS">(<a href="#ref-Vehtari+etal:2024:PSIS" role="doc-biblioref">Vehtari et al. 2024, sec. 2.2</a>)</span>. If estimated Pareto-<span class="math inline">\(\hat{k}\)</span> is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow <span class="citation" data-cites="Gelman:etal:2020:workflow">(<a href="#ref-Gelman:etal:2020:workflow" role="doc-biblioref">Gelman et al. 2020</a>)</span>.</p>
+</section>
+<section id="using-pathfinder-for-initializing-mcmc" class="level2">
+<h2 class="anchored" data-anchor-id="using-pathfinder-for-initializing-mcmc">Using Pathfinder for initializing MCMC</h2>
+<p>If estimated Pareto-<span class="math inline">\(\hat{k}\)</span> for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-<span class="math inline">\(\hat{k}\)</span> for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g.&nbsp;uniform from -2 to 2 used by Stan by default). If Pareto-<span class="math inline">\(\hat{k}\)</span> is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Gelman:etal:2020:workflow" class="csl-entry" role="listitem">
+Gelman, Andrew, Aki Vehtari, Daniel Simpson, Charles C Margossian, Bob Carpenter, Yuling Yao, Lauren Kennedy, Jonah Gabry, Paul-Christian Bürkner, and Martin Modrák. 2020. <span>“Bayesian Workflow.”</span> <em>arXiv Preprint arXiv:2011.01808</em>.
+</div>
+<div id="ref-Magnusson+etal:2024:posteriordb" class="csl-entry" role="listitem">
+Magnusson, Måns, Jakob Torgander, Paul-Christian Bürkner, Lu Zhang, Bob Carpenter, and Aki Vehtari. 2024. <span>“Posteriordb: Testing, Benchmarking and Developing <span>Bayesian</span> Inference Algorithms.”</span> <em>arXiv Preprint arXiv:2407.04967</em>.
+</div>
+<div id="ref-Vehtari+etal:2024:PSIS" class="csl-entry" role="listitem">
+Vehtari, Aki, Daniel Simpson, Andrew Gelman, Yuling Yao, and Jonah Gabry. 2024. <span>“Pareto Smoothed Importance Sampling.”</span> <em>Journal of Machine Learning Research</em> 25 (72): 1–58.
+</div>
+<div id="ref-zhang_pathfinder:2022" class="csl-entry" role="listitem">
+Zhang, Lu, Bob Carpenter, Andrew Gelman, and Aki Vehtari. 2022. <span>“Pathfinder: Parallel Quasi-Newton Variational Inference.”</span> <em>Journal of Machine Learning Research</em> 23 (306): 1–49. <a href="http://jmlr.org/papers/v23/21-0889.html">http://jmlr.org/papers/v23/21-0889.html</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/optimization.html" class="pagination-link" aria-label="Optimization">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Optimization</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/variational.html" class="pagination-link" aria-label="Variational Inference">
+        <span class="nav-page-text">Variational Inference</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/pathfinder.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/references.html b/docs/2_39/reference-manual/references.html
new file mode 100644
index 000000000..d08beeba7
--- /dev/null
+++ b/docs/2_39/reference-manual/references.html
@@ -0,0 +1,821 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>references</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Docs">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#references" id="toc-references" class="nav-link active" data-scroll-target="#references"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
+
+
+
+
+<section id="references" class="level1 unnumbered">
+<h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></h1>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/removals.html b/docs/2_39/reference-manual/removals.html
new file mode 100644
index 000000000..c4c380cba
--- /dev/null
+++ b/docs/2_39/reference-manual/removals.html
@@ -0,0 +1,1203 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Removed Features</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/mcmc.html" rel="next">
+<link href="../reference-manual/deprecations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Removed Features">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/removals.html">Removed Features</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#removed-features" id="toc-removed-features" class="nav-link active" data-scroll-target="#removed-features">Removed Features</a>
+  <ul class="collapse">
+  <li><a href="#lp__-variable" id="toc-lp__-variable" class="nav-link" data-scroll-target="#lp__-variable"><code>lp__</code> variable</a></li>
+  <li><a href="#assignment-with--" id="toc-assignment-with--" class="nav-link" data-scroll-target="#assignment-with--">Assignment with <code>&lt;-</code></a></li>
+  <li><a href="#increment_log_prob-statement" id="toc-increment_log_prob-statement" class="nav-link" data-scroll-target="#increment_log_prob-statement"><code>increment_log_prob</code> statement</a></li>
+  <li><a href="#get_lp-function" id="toc-get_lp-function" class="nav-link" data-scroll-target="#get_lp-function"><code>get_lp()</code> function</a></li>
+  <li><a href="#log-density-and-mass-functions" id="toc-log-density-and-mass-functions" class="nav-link" data-scroll-target="#log-density-and-mass-functions"><code>_log</code> density and mass functions</a></li>
+  <li><a href="#cdf_log-and-ccdf_log-cumulative-distribution-functions" id="toc-cdf_log-and-ccdf_log-cumulative-distribution-functions" class="nav-link" data-scroll-target="#cdf_log-and-ccdf_log-cumulative-distribution-functions"><code>cdf_log</code> and <code>ccdf_log</code> cumulative distribution functions</a></li>
+  <li><a href="#user-defined-function-with-_log-suffix" id="toc-user-defined-function-with-_log-suffix" class="nav-link" data-scroll-target="#user-defined-function-with-_log-suffix">User-defined function with <code>_log</code> suffix</a></li>
+  <li><a href="#if_else-function" id="toc-if_else-function" class="nav-link" data-scroll-target="#if_else-function"><code>if_else</code> function</a></li>
+  <li><a href="#character-as-comment-prefix" id="toc-character-as-comment-prefix" class="nav-link" data-scroll-target="#character-as-comment-prefix">Character <code>#</code> as comment prefix</a></li>
+  <li><a href="#postfix-brackets-array-syntax" id="toc-postfix-brackets-array-syntax" class="nav-link" data-scroll-target="#postfix-brackets-array-syntax">Postfix brackets array syntax</a></li>
+  <li><a href="#nested-multiple-indexing-in-assignments" id="toc-nested-multiple-indexing-in-assignments" class="nav-link" data-scroll-target="#nested-multiple-indexing-in-assignments">Nested multiple indexing in assignments</a></li>
+  <li><a href="#real-values-in-conditionals" id="toc-real-values-in-conditionals" class="nav-link" data-scroll-target="#real-values-in-conditionals">Real values in conditionals</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/removals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/removals.html">Removed Features</a></li></ol></nav></header>
+
+
+
+
+<section id="removed-features" class="level1">
+<h1>Removed Features</h1>
+<p>This chapter lists functionalities that were once present in the language but have since been removed, along with how to replace them.</p>
+<section id="lp__-variable" class="level2">
+<h2 class="anchored" data-anchor-id="lp__-variable"><code>lp__</code> variable</h2>
+<p><em>Removed</em>: The variable <code>lp__</code> is no longer available for direct access or manipulation.</p>
+<p><em>Replacement</em>: General manipulation of the value of the <code>lp__</code> variable is not allowed, but</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>lp__ &lt;- lp__ + e;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>can be replaced with</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> e;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The value of <code>lp__</code> is available through the no-argument function <code>target()</code>.</p>
+</section>
+<section id="assignment-with--" class="level2">
+<h2 class="anchored" data-anchor-id="assignment-with--">Assignment with <code>&lt;-</code></h2>
+<p><em>Removed</em>: The operator <code>&lt;-</code> for assignment, e.g.,</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>a &lt;- b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is no longer available.</p>
+<p><em>Replacement</em>: The new syntax uses the operator <code>=</code> for assignment, e.g.,</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>a = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="increment_log_prob-statement" class="level2">
+<h2 class="anchored" data-anchor-id="increment_log_prob-statement"><code>increment_log_prob</code> statement</h2>
+<p><em>Removed</em>: The <code>increment_log_prob(u)</code> statement for incrementing the log density accumulator by <code>u</code> is no longer available.</p>
+<p><em>Replacement</em>: Replace the above statement with</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="get_lp-function" class="level2">
+<h2 class="anchored" data-anchor-id="get_lp-function"><code>get_lp()</code> function</h2>
+<p><em>Removed</em>: The built-in no-argument function <code>get_lp()</code> is no longer available.</p>
+<p><em>Replacement</em>: Use the no-argument function <code>target()</code> instead.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="log-density-and-mass-functions" class="level2">
+<h2 class="anchored" data-anchor-id="log-density-and-mass-functions"><code>_log</code> density and mass functions</h2>
+<p><em>Removed</em>: Formerly, the probability function for the distribution <code>foo</code> would be applied to an outcome variable <code>y</code> and sequence of zero or more parameters <code>...</code> to produce the expression <code>foo_log(y, ...)</code>. This suffix is no longer a special value.</p>
+<p><em>Replacement</em>: If <code>y</code> can be a real value (including vectors or matrices), replace</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a>foo_log(y, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with the log probability density function notation</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a>foo_lpdf(y | ...).</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If <code>y</code> must be an integer (including arrays), instead replace</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a>foo_log(y, ...</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with the log probability mass function</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a>foo_lpmf(y | ...).</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="cdf_log-and-ccdf_log-cumulative-distribution-functions" class="level2">
+<h2 class="anchored" data-anchor-id="cdf_log-and-ccdf_log-cumulative-distribution-functions"><code>cdf_log</code> and <code>ccdf_log</code> cumulative distribution functions</h2>
+<p><em>Removed</em>: The log cumulative distribution and complementary cumulative distribution functions for a distribution <code>foo</code> were formerly written as <code>foo_cdf_log</code> and <code>foo_ccdf_log</code>.</p>
+<p><em>Replacement</em>:</p>
+<p>Replace <code>foo_cdf_log(y, ...)</code> with <code>foo_lcdf(y | ...)</code>.</p>
+<p>Replace <code>foo_ccdf_log(y, ...)</code> with <code>foo_lccdf(y | ...)</code>.</p>
+</section>
+<section id="user-defined-function-with-_log-suffix" class="level2">
+<h2 class="anchored" data-anchor-id="user-defined-function-with-_log-suffix">User-defined function with <code>_log</code> suffix</h2>
+<p><em>Removed</em>: A user-defined function ending in <code>_log</code> can be no longer be used in <a href="distribution statements">statements.qmd#distribution-statements.section</a>.</p>
+<p><em>Replacement</em>: Replace the <code>_log</code> suffix with <code>_lpdf</code> for density functions or <code>_lpmf</code> for mass functions in the user-defined function.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+<p>Note: Following Stan 2.33, users can stil define a function ending in <code>_log</code>, it simply no longer has a special meaning or is supported in the <code>~</code> syntax.</p>
+</section>
+<section id="if_else-function" class="level2">
+<h2 class="anchored" data-anchor-id="if_else-function"><code>if_else</code> function</h2>
+<p><em>Removed</em>: The function <code>if_else</code> is no longer available.</p>
+<p><em>Replacement</em>: Use the conditional operator which allows more flexibility in the types of <code>b</code> and <code>c</code> and is much more efficient in that it only evaluates whichever of <code>b</code> or <code>c</code> is returned.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>x = if_else(a, b, c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a>x = a ? b : c;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="character-as-comment-prefix" class="level2">
+<h2 class="anchored" data-anchor-id="character-as-comment-prefix">Character <code>#</code> as comment prefix</h2>
+<p><em>Removed</em>: The use of <code>#</code> for line-based comments is no longer permitted. <code>#</code> may only be used for <code>#include</code> statements.</p>
+<p><em>Replacement</em>: Use a pair of forward slashes, <code>//</code>, for line comments.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="postfix-brackets-array-syntax" class="level2">
+<h2 class="anchored" data-anchor-id="postfix-brackets-array-syntax">Postfix brackets array syntax</h2>
+<p>Before Stan 2.26, arrays were declared by writing syntax after the variable.</p>
+<p><em>Removed</em>: The use of array declarations like</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n[<span class="dv">5</span>];</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> a[<span class="dv">3</span>, <span class="dv">4</span>];</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; z[<span class="dv">5</span>, <span class="dv">4</span>, <span class="dv">2</span>];</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">7</span>] mu[<span class="dv">3</span>];</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">7</span>, <span class="dv">2</span>] mu[<span class="dv">15</span>, <span class="dv">12</span>];</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a><span class="dt">cholesky_factor_cov</span>[<span class="dv">5</span>, <span class="dv">6</span>] mu[<span class="dv">2</span>, <span class="dv">3</span>, <span class="dv">4</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><em>Replacement</em>: The use of the <code>array</code> keyword, which replaces the above examples with</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">int</span> n;</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">real</span> a;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">4</span>, <span class="dv">2</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; z;</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">vector</span>[<span class="dv">7</span>] mu;</span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">15</span>, <span class="dv">12</span>] <span class="dt">matrix</span>[<span class="dv">7</span>, <span class="dv">2</span>] mu;</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">cholesky_factor_cov</span>[<span class="dv">5</span>, <span class="dv">6</span>] mu;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="nested-multiple-indexing-in-assignments" class="level2">
+<h2 class="anchored" data-anchor-id="nested-multiple-indexing-in-assignments">Nested multiple indexing in assignments</h2>
+<p>Stan interprets nested indexing in assingments as flat indexing so that a statement like</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a>a[:][<span class="dv">1</span>] = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is the same as</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a>a[:,<span class="dv">1</span>] = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>However, this is inconsistent with <a href="https://mc-stan.org/docs/reference-manual/expressions.html#language-multi-indexing.section">multiple indexing rules</a>.</p>
+<p>To avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.</p>
+<p><em>Removed In</em>: Stan 2.33</p>
+</section>
+<section id="real-values-in-conditionals" class="level2">
+<h2 class="anchored" data-anchor-id="real-values-in-conditionals">Real values in conditionals</h2>
+<p><em>Removed</em>: Using a real value in a conditional is no longer permitted.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="fl">1.0</span>;</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (x) {</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The value was interpreted as true if it is nonzero.</p>
+<p><em>Replacement</em>: For the exact equivalent, use a comparison operator to make the intent clear.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="fl">1.0</span>;</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (x != <span class="dv">0</span>) {</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>However, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative <code>abs(x) &lt; machine_precision()</code> is appropriate for the use case.</p>
+<p><em>Removed In</em>: Stan 2.34</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/deprecations.html" class="pagination-link" aria-label="Deprecated Features">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Deprecated Features</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/mcmc.html" class="pagination-link" aria-label="MCMC Sampling">
+        <span class="nav-page-text">MCMC Sampling</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/removals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/reproducibility.html b/docs/2_39/reference-manual/reproducibility.html
new file mode 100644
index 000000000..127106597
--- /dev/null
+++ b/docs/2_39/reference-manual/reproducibility.html
@@ -0,0 +1,1094 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Reproducibility</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/licenses.html" rel="next">
+<link href="../reference-manual/diagnostics.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Reproducibility">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Usage</a></li><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Reproducibility</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#reproducibility.chapter" id="toc-reproducibility.chapter" class="nav-link active" data-scroll-target="#reproducibility.chapter">Reproducibility</a>
+  <ul class="collapse">
+  <li><a href="#notable-changes-across-versions" id="toc-notable-changes-across-versions" class="nav-link" data-scroll-target="#notable-changes-across-versions">Notable changes across versions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/reproducibility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Usage</a></li><li class="breadcrumb-item"><a href="../reference-manual/reproducibility.html">Reproducibility</a></li></ol></nav></header>
+
+
+
+
+<section id="reproducibility.chapter" class="level1">
+<h1>Reproducibility</h1>
+<p>Floating point operations on modern computers are notoriously difficult to replicate because the fundamental arithmetic operations, right down to the IEEE 754 encoding level, are not fully specified. The primary problem is that the precision of operations varies across different hardware platforms and software implementations.</p>
+<p>Stan is designed to allow full reproducibility. However, this is only possible up to the external constraints imposed by floating point arithmetic.</p>
+<p>Stan results will only be exactly reproducible if <em>all</em> of the following components are <em>identical</em>:</p>
+<ul>
+<li>Stan version</li>
+<li>Stan interface (RStan, PyStan, CmdStan) and version, plus version of interface language (R, Python, shell)</li>
+<li>versions of included libraries (Boost and Eigen)</li>
+<li>operating system version</li>
+<li>computer hardware including CPU, motherboard and memory</li>
+<li>C++ compiler, including version, compiler flags, and linked libraries</li>
+<li>same configuration of call to Stan, including random seed, chain ID, initialization and data</li>
+</ul>
+<p>It doesn’t matter if you use a stable release version of Stan or the version with a particular Git hash tag. The same goes for all of the interfaces, compilers, and so on. The point is that if any of these moving parts changes in some way, floating point results may change.</p>
+<p>Concretely, if you compile a single Stan program using the same CmdStan code base, but changed the optimization flag (<code>-O3</code> vs.&nbsp;<code>-O2</code> or <code>-O0</code>), the two programs may not return the identical stream of results. Thus it is very hard to guarantee reproducibility on externally managed hardware, like in a cluster or even a desktop managed by an IT department or with automatic updates turned on.</p>
+<p>If, however, you compiled a Stan program today using one set of flags, took the computer away from the internet and didn’t allow it to update anything, then came back in a decade and recompiled the Stan program in the same way, you should get the same results.</p>
+<p>The data needs to be the same down to the bit level. For example, if you are running in RStan, Rcpp handles the conversion between R’s floating point numbers and C++ doubles. If Rcpp changes the conversion process or use different types, the results are not guaranteed to be the same down to the bit level.</p>
+<p>The compiler and compiler settings can also be an issue. There is a nice discussion of the issues and how to control reproducibility in Intel’s proprietary compiler by <span class="citation" data-cites="CordenKreitzer:2014">Corden and Kreitzer (<a href="#ref-CordenKreitzer:2014" role="doc-biblioref">2014</a>)</span>.</p>
+<section id="notable-changes-across-versions" class="level2">
+<h2 class="anchored" data-anchor-id="notable-changes-across-versions">Notable changes across versions</h2>
+<p>As noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.</p>
+<p>However, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still <em>does not</em> guarantee exact reproducibility between it and other versions.</p>
+<ul>
+<li>Stan 2.28 changed the default chain ID for MCMC from <code>0</code> to <code>1</code>. Users who had set a seed but not a chain ID would observe completely different outputs.</li>
+<li>Stan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.</li>
+</ul>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-CordenKreitzer:2014" class="csl-entry" role="listitem">
+Corden, Martyn J., and David Kreitzer. 2014. <span>“Consistency of Floating-Point Results Using the <span>I</span>ntel Compiler or <span>W</span>hy Doesn’t My Application Always Give the Same Answer?”</span> Intel Corporation. <a href="https://software.intel.com/en-us/articles/consistency-of-floating-point-results-using-the-intel-compiler">https://software.intel.com/en-us/articles/consistency-of-floating-point-results-using-the-intel-compiler</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/diagnostics.html" class="pagination-link" aria-label="Diagnostic Mode">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Diagnostic Mode</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/licenses.html" class="pagination-link" aria-label="Licenses and Dependencies">
+        <span class="nav-page-text">Licenses and Dependencies</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/reproducibility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/statements.html b/docs/2_39/reference-manual/statements.html
new file mode 100644
index 000000000..6fd7ff97d
--- /dev/null
+++ b/docs/2_39/reference-manual/statements.html
@@ -0,0 +1,1951 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Statements</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/blocks.html" rel="next">
+<link href="../reference-manual/expressions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Statements">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/statements.html">Statements</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#statements" id="toc-statements" class="nav-link active" data-scroll-target="#statements">Statements</a>
+  <ul class="collapse">
+  <li><a href="#statement-block-contexts" id="toc-statement-block-contexts" class="nav-link" data-scroll-target="#statement-block-contexts">Statement block contexts</a></li>
+  <li><a href="#assignment-statement.section" id="toc-assignment-statement.section" class="nav-link" data-scroll-target="#assignment-statement.section">Assignment statements</a>
+  <ul class="collapse">
+  <li><a href="#promotion" id="toc-promotion" class="nav-link" data-scroll-target="#promotion">Promotion</a></li>
+  <li><a href="#lvalue-summary" id="toc-lvalue-summary" class="nav-link" data-scroll-target="#lvalue-summary">Lvalue summary</a></li>
+  <li><a href="#multiple-indexes" id="toc-multiple-indexes" class="nav-link" data-scroll-target="#multiple-indexes">Multiple indexes</a></li>
+  <li><a href="#compound-arithmetic-assignment.section" id="toc-compound-arithmetic-assignment.section" class="nav-link" data-scroll-target="#compound-arithmetic-assignment.section">Compound arithmetic and assignment statement</a></li>
+  </ul></li>
+  <li><a href="#increment-log-prob.section" id="toc-increment-log-prob.section" class="nav-link" data-scroll-target="#increment-log-prob.section">Increment log density</a></li>
+  <li><a href="#increment-log-density-with-a-change-of-variables-adjustment" id="toc-increment-log-density-with-a-change-of-variables-adjustment" class="nav-link" data-scroll-target="#increment-log-density-with-a-change-of-variables-adjustment">Increment log density with a change of variables adjustment</a>
+  <ul class="collapse">
+  <li><a href="#accessing-the-log-density" id="toc-accessing-the-log-density" class="nav-link" data-scroll-target="#accessing-the-log-density">Accessing the log density</a></li>
+  </ul></li>
+  <li><a href="#sampling-statements.section" id="toc-sampling-statements.section" class="nav-link" data-scroll-target="#sampling-statements.section">Sampling statements</a></li>
+  <li><a href="#distribution-statements.section" id="toc-distribution-statements.section" class="nav-link" data-scroll-target="#distribution-statements.section">Distribution statements</a>
+  <ul class="collapse">
+  <li><a href="#log-probability-increment-vs.-distribution-statement" id="toc-log-probability-increment-vs.-distribution-statement" class="nav-link" data-scroll-target="#log-probability-increment-vs.-distribution-statement">Log probability increment vs.&nbsp;distribution statement</a></li>
+  <li><a href="#user-transformed-variables" id="toc-user-transformed-variables" class="nav-link" data-scroll-target="#user-transformed-variables">User-transformed variables</a></li>
+  <li><a href="#truncated-distributions" id="toc-truncated-distributions" class="nav-link" data-scroll-target="#truncated-distributions">Truncated distributions</a></li>
+  </ul></li>
+  <li><a href="#for-loops" id="toc-for-loops" class="nav-link" data-scroll-target="#for-loops">For loops</a>
+  <ul class="collapse">
+  <li><a href="#loop-variable-typing-and-scope" id="toc-loop-variable-typing-and-scope" class="nav-link" data-scroll-target="#loop-variable-typing-and-scope">Loop variable typing and scope</a></li>
+  <li><a href="#order-sensitivity-and-repeated-variables" id="toc-order-sensitivity-and-repeated-variables" class="nav-link" data-scroll-target="#order-sensitivity-and-repeated-variables">Order sensitivity and repeated variables</a></li>
+  </ul></li>
+  <li><a href="#foreach-loops" id="toc-foreach-loops" class="nav-link" data-scroll-target="#foreach-loops">Foreach loops</a></li>
+  <li><a href="#conditional-statements" id="toc-conditional-statements" class="nav-link" data-scroll-target="#conditional-statements">Conditional statements</a></li>
+  <li><a href="#while-statements" id="toc-while-statements" class="nav-link" data-scroll-target="#while-statements">While statements</a></li>
+  <li><a href="#statement-blocks-and-local-variable-declarations" id="toc-statement-blocks-and-local-variable-declarations" class="nav-link" data-scroll-target="#statement-blocks-and-local-variable-declarations">Statement blocks and local variable declarations</a>
+  <ul class="collapse">
+  <li><a href="#blocks-in-for-loops" id="toc-blocks-in-for-loops" class="nav-link" data-scroll-target="#blocks-in-for-loops">Blocks in for loops</a></li>
+  <li><a href="#local-variable-declarations" id="toc-local-variable-declarations" class="nav-link" data-scroll-target="#local-variable-declarations">Local variable declarations</a></li>
+  <li><a href="#no-constraints-on-local-variables" id="toc-no-constraints-on-local-variables" class="nav-link" data-scroll-target="#no-constraints-on-local-variables">No constraints on local variables</a></li>
+  <li><a href="#blocks-within-blocks" id="toc-blocks-within-blocks" class="nav-link" data-scroll-target="#blocks-within-blocks">Blocks within blocks</a></li>
+  </ul></li>
+  <li><a href="#break-continue-statements" id="toc-break-continue-statements" class="nav-link" data-scroll-target="#break-continue-statements">Break and continue statements</a>
+  <ul class="collapse">
+  <li><a href="#break-statements" id="toc-break-statements" class="nav-link" data-scroll-target="#break-statements">Break statements</a></li>
+  <li><a href="#continue-statements" id="toc-continue-statements" class="nav-link" data-scroll-target="#continue-statements">Continue statements</a></li>
+  <li><a href="#breaking-and-continuing-nested-loops" id="toc-breaking-and-continuing-nested-loops" class="nav-link" data-scroll-target="#breaking-and-continuing-nested-loops">Breaking and continuing nested loops</a></li>
+  </ul></li>
+  <li><a href="#print-statements.section" id="toc-print-statements.section" class="nav-link" data-scroll-target="#print-statements.section">Print statements</a>
+  <ul class="collapse">
+  <li><a href="#print-content" id="toc-print-content" class="nav-link" data-scroll-target="#print-content">Print content</a></li>
+  <li><a href="#non-void-input" id="toc-non-void-input" class="nav-link" data-scroll-target="#non-void-input">Non-void input</a></li>
+  <li><a href="#print-frequency" id="toc-print-frequency" class="nav-link" data-scroll-target="#print-frequency">Print frequency</a></li>
+  <li><a href="#string-literals" id="toc-string-literals" class="nav-link" data-scroll-target="#string-literals">String literals</a></li>
+  <li><a href="#debug-by-print" id="toc-debug-by-print" class="nav-link" data-scroll-target="#debug-by-print">Debug by <code>print</code></a></li>
+  </ul></li>
+  <li><a href="#reject-statements.section" id="toc-reject-statements.section" class="nav-link" data-scroll-target="#reject-statements.section">Reject statements</a>
+  <ul class="collapse">
+  <li><a href="#behavior-of-reject-statements" id="toc-behavior-of-reject-statements" class="nav-link" data-scroll-target="#behavior-of-reject-statements">Behavior of reject statements</a></li>
+  <li><a href="#rejection-is-not-for-constraints" id="toc-rejection-is-not-for-constraints" class="nav-link" data-scroll-target="#rejection-is-not-for-constraints">Rejection is not for constraints</a></li>
+  </ul></li>
+  <li><a href="#fatal-error-statements" id="toc-fatal-error-statements" class="nav-link" data-scroll-target="#fatal-error-statements">Fatal error statements</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/statements.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/statements.html">Statements</a></li></ol></nav></header>
+
+
+
+
+<section id="statements" class="level1">
+<h1>Statements</h1>
+<p>The blocks of a Stan program are made up of variable declarations and statements; see the <a href="../reference-manual/blocks.html">blocks chapter</a> for details. Unlike programs in BUGS, the declarations and statements making up a Stan program are executed in the order in which they are written. Variables must be defined to have some value (as well as declared to have some type) before they are used — if they do not, the behavior is undefined.</p>
+<p>The basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters. Log probability functions can be constructed by using distribution statements and log probability increment statements. Statements may be grouped into sequences and into for-each loops. In addition, Stan allows local variables to be declared in blocks and also allows an empty statement consisting only of a semicolon.</p>
+<section id="statement-block-contexts" class="level2">
+<h2 class="anchored" data-anchor-id="statement-block-contexts">Statement block contexts</h2>
+<p>The data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the <a href="../reference-manual/blocks.html">blocks chapter</a> for more information about the block structure of Stan programs.</p>
+</section>
+<section id="assignment-statement.section" class="level2">
+<h2 class="anchored" data-anchor-id="assignment-statement.section">Assignment statements</h2>
+<p>An assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>n = <span class="dv">0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Executing this statement assigns the value of the expression <code>0</code>, which is the integer zero, to the variable <code>n</code>. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because <code>0</code> is an expression of type <code>int</code>, the variable <code>n</code> must be declared as being of type <code>int</code> or of type <code>real</code>. If the variable is of type <code>real</code>, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable <code>n</code> will have the value zero (either as an integer or a floating-point value, depending on its type).</p>
+<p>Syntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).</p>
+<p>Because the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>n = n + <span class="dv">1</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Such self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.</p>
+<p>The left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if <code>Sigma</code> is of type <code>matrix</code>, then</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>Sigma[<span class="dv">1</span>, <span class="dv">1</span>] = <span class="fl">1.0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>sets the value in the first column of the first row of <code>Sigma</code> to one.</p>
+<p>Assignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, <code>a</code> is an array of type <code>array[,] real</code> and <code>b</code> is an array of type <code>array[] real</code>, then the following two statements are both well-formed.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">3</span>] = b;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>b = a[<span class="dv">4</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Similarly, if <code>x</code> is a variable declared to have type <code>row_vector</code> and <code>Y</code> is a variable declared as type <code>matrix</code>, then the following sequence of statements to swap the first two rows of <code>Y</code> is well formed.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a>x = Y[<span class="dv">1</span>];</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>Y[<span class="dv">1</span>] = Y[<span class="dv">2</span>];</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>Y[<span class="dv">2</span>] = x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="promotion" class="level3">
+<h3 class="anchored" data-anchor-id="promotion">Promotion</h3>
+<p>Stan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type <code>int</code> to an lvalue of type <code>real</code>, and we can assign an expression of type <code>real</code> to an lvalue of type <code>complex</code>. Furthermore, promotion is transitive, so that we can assign an expression of type <code>int</code> to an lvalue of type <code>complex</code>.</p>
+<p>Promotion extends to containers, so that arrays of <code>int</code> can be promoted to arrays of <code>real</code> during assignment, and arrays of <code>real</code> can be assigned to an lvalue of type array of <code>complex</code>. Similarly, an expression of type <code>vector</code> may be assigned to an lvalue of type <code>complex_vector</code>, and similarly for row vectors and matrices.</p>
+</section>
+<section id="lvalue-summary" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lvalue-summary">Lvalue summary</h3>
+<p>The expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,</p>
+<ul>
+<li>a variable, or</li>
+<li>a variable with one or more indices, or</li>
+<li>a comma separated list of lvalues surrounded by <code>(</code> and <code>)</code></li>
+</ul>
+<p>To be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.</p>
+</section>
+<section id="multiple-indexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multiple-indexes">Multiple indexes</h3>
+<p>Multiple indexes, as described in the <a href="../reference-manual/expressions.html#language-multi-indexing.section">multi-indexing section</a>, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.</p>
+<section id="aliasing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="aliasing">Aliasing</h4>
+<p>All assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.</p>
+</section>
+</section>
+<section id="compound-arithmetic-assignment.section" class="level3">
+<h3 class="anchored" data-anchor-id="compound-arithmetic-assignment.section">Compound arithmetic and assignment statement</h3>
+<p>Stan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="dv">5</span>;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>x += <span class="dv">7</span>;  <span class="co">// value of x is now 12</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The compound arithmetic and assignment statement above is equivalent to the following long form.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a>x = x + <span class="dv">7</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In general, the compound form</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a>x op= y</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>will be equivalent to</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a>x = x op y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The compound statement will be legal whenever the long form is legal. This requires that the operation <code>x op y</code> must itself be well formed and that the result of the operation be assignable to <code>x</code>. For the expression <code>x</code> to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] x;</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>x[<span class="dv">3</span>] += <span class="dv">2</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As a further example, consider</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, M] x;</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[M] y;</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> z;</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>x *= x;  <span class="co">// OK, (x * x) is a matrix</span></span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>x *= z;  <span class="co">// OK, (x * z) is a matrix</span></span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>x *= y;  <span class="co">// BAD, (x * y) is a vector</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The supported compound arithmetic and assignment operations are listed in the <a href="#compound-arithmetic-assign-table">compound arithmetic/assignment table</a>; they are also listed in the index prefaced by <code>operator</code>, e.g., <code>operator+=</code>.</p>
+<p><strong>Compound Arithmetic/Assignment Table.</strong> <a id="compound-arithmetic-assign-table"></a> <em>Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.</em></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;">operation</th>
+<th style="text-align: left;">compound</th>
+<th style="text-align: left;">unfolded</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;">addition</td>
+<td style="text-align: left;"><code>x += y</code></td>
+<td style="text-align: left;"><code>x = x + y</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">subtraction</td>
+<td style="text-align: left;"><code>x -= y</code></td>
+<td style="text-align: left;"><code>x = x - y</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">multiplication</td>
+<td style="text-align: left;"><code>x *= y</code></td>
+<td style="text-align: left;"><code>x = x * y</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">division</td>
+<td style="text-align: left;"><code>x /= y</code></td>
+<td style="text-align: left;"><code>x = x / y</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">elementwise multiplication</td>
+<td style="text-align: left;"><code>x .*= y</code></td>
+<td style="text-align: left;"><code>x = x .* y</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">elementwise division</td>
+<td style="text-align: left;"><code>x ./= y</code></td>
+<td style="text-align: left;"><code>x = x ./ y</code></td>
+</tr>
+</tbody>
+</table>
+</section>
+</section>
+<section id="increment-log-prob.section" class="level2">
+<h2 class="anchored" data-anchor-id="increment-log-prob.section">Increment log density</h2>
+<p>The basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> -<span class="fl">0.5</span> * y * y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The keyword <code>target</code> here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).</p>
+<p>In this example, the unnormalized log probability of a unit normal variable <span class="math inline">\(y\)</span> is added to the total log probability. In the general case, the argument can be any expression.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>An entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -<span class="fl">0.5</span> * y * y;</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This model defines a log probability function</p>
+<p><span class="math display">\[
+\log p(y) = - \, \frac{y^2}{2} - \log Z
+\]</span></p>
+<p>where <span class="math inline">\(Z\)</span> is a normalizing constant that does not depend on <span class="math inline">\(y\)</span>. The constant <span class="math inline">\(Z\)</span> is conventionally written this way because on the linear scale, <span class="math display">\[
+p(y) = \frac{1}{Z} \exp\left(-\frac{y^2}{2}\right).
+\]</span> which is typically written without reference to <span class="math inline">\(Z\)</span> as <span class="math display">\[
+p(y) \propto \exp\left(-\frac{y^2}{2}\right).
+\]</span></p>
+<p>Stan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant <span class="math inline">\(Z\)</span> is either time-consuming to compute or intractable to evaluate.</p>
+<section id="built-in-distributions" class="level4">
+<h4 class="anchored" data-anchor-id="built-in-distributions">Built in distributions</h4>
+<p>The built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.</p>
+<p>For instance, the <code>normal_lpdf</code> function returns the log density of a normal distribution:</p>
+<p><span class="math display">\[
+\textsf{normal\_lpdf}(x | \mu, \sigma) =
+-\log \left( \sigma \sqrt{2 \pi} \right)
+-\frac{1}{2} \left( \frac{x - \mu}{\sigma} \right)^2
+\]</span></p>
+<p>The <code>normal_lupdf</code> function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant <code>sigma</code> term, <code>normal_lupdf</code> would be equivalent to:</p>
+<p><span class="math display">\[
+\textsf{normal\_lupdf}(x | \mu, \sigma) =
+-\frac{1}{2} \left( \frac{x - \mu}{\sigma} \right)^2
+\]</span></p>
+<p>All functions ending in <code>_lpdf</code> have a corresponding <code>_lupdf</code> version which evaluates and returns the unnormalized density. The same is true for <code>_lpmf</code> and <code>_lupmf</code>.</p>
+</section>
+<section id="relation-to-compound-addition-and-assignment" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="relation-to-compound-addition-and-assignment">Relation to compound addition and assignment</h4>
+<p>The increment log density statement looks syntactically like compound addition and assignment (see the <a href="#compound-arithmetic-assignment.section">compound arithmetic/assignment section</a>, it is treated as a primitive statement because <code>target</code> is not itself a variable. So, even though</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> lp;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is a legal statement, the corresponding long form is not legal.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a>target = target + lp;  <span class="co">// BAD, target is not a variable</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="vectorization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="vectorization">Vectorization</h4>
+<p>The <code>target += ...</code> statement accepts an argument in place of <code>...</code> for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.</p>
+</section>
+</section>
+<section id="increment-log-density-with-a-change-of-variables-adjustment" class="level2">
+<h2 class="anchored" data-anchor-id="increment-log-density-with-a-change-of-variables-adjustment">Increment log density with a change of variables adjustment</h2>
+<p>A variant of the <code>target +=</code> statement described above is the <code>jacobian +=</code> statement. This can be used in the transformed parameters block or in functions ending with <code>_jacobian</code> to mimic the log Jacobian adjustments accrued by built-in variable transforms.</p>
+<p>Similarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.</p>
+<p>For example, here is a program which recreates the existing <a href="../reference-manual/transforms.html#upper-bounded-scalar"><code>&lt;upper=x&gt;</code> transform</a> on real numbers:</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> my_upper_bound_jacobian(<span class="dt">real</span> x, <span class="dt">real</span> ub) {</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>    <span class="kw">jacobian +=</span> x;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> ub - exp(x);</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ub;</span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-10"><a href="#cb16-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb16-11"><a href="#cb16-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b_raw;</span>
+<span id="cb16-12"><a href="#cb16-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-13"><a href="#cb16-13" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb16-14"><a href="#cb16-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b = my_upper_bound_jacobian(b_raw, ub);</span>
+<span id="cb16-15"><a href="#cb16-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-16"><a href="#cb16-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-17"><a href="#cb16-17" aria-hidden="true" tabindex="-1"></a>  <span class="co">// use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters</span></span>
+<span id="cb16-18"><a href="#cb16-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// e.g.</span></span>
+<span id="cb16-19"><a href="#cb16-19" aria-hidden="true" tabindex="-1"></a>  <span class="co">// b ~ lognormal(0, 1);</span></span>
+<span id="cb16-20"><a href="#cb16-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="accessing-the-log-density" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="accessing-the-log-density">Accessing the log density</h3>
+<p>To access the accumulated log density up to the current execution point, the function <code>target()</code> may be used.</p>
+</section>
+</section>
+<section id="sampling-statements.section" class="level2">
+<h2 class="anchored" data-anchor-id="sampling-statements.section">Sampling statements</h2>
+<p>The term “sampling statement” has been replaced with <a href="#distribution-statements.section">distribution statement</a>.</p>
+</section>
+<section id="distribution-statements.section" class="level2">
+<h2 class="anchored" data-anchor-id="distribution-statements.section">Distribution statements</h2>
+<p>Stan supports writing probability statements also using distribution statements, for example</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma);</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>mu ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>sigma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The symbol <span class="math inline">\(\sim\)</span> is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.</p>
+<p>In general, we can read <span class="math inline">\(\sim\)</span> as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as <span class="math display">\[
+\begin{aligned}
+   p(y| \mu, \sigma) &amp; = \mathrm{normal}(y |  \mu, \sigma)\\
+   p(\mu) &amp; = \mathrm{normal}(\mu |  0, 10)\\
+   p(\sigma) &amp; = \mathrm{normal}^+(\sigma |  0, 1).
+\end{aligned}
+\]</span> A collection of distribution statements define a joint distribution as the product of component distributions <span class="math display">\[
+p(y,\mu,\sigma) = p(y| \mu, \sigma )p(\mu) p(\sigma).
+\]</span></p>
+<p>This works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This is translated to <span class="math display">\[
+    p(a) = \mathrm{normal}(a | 0, 1)\mathrm{normal}(a |  0, 1),
+\]</span> which in this case is <span class="math inline">\(\mathrm{normal}(a|0,1/\sqrt{2})\)</span>. One might expect that the above two lines of code would represent a redundant expression of a <span class="math inline">\(\mathrm{normal}(a|0,1)\)</span> prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.</p>
+<p>When the joint distribution is considered as a function of parameters (e.g.&nbsp;<span class="math inline">\(\mu\)</span>, <span class="math inline">\(\sigma\)</span>) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.</p>
+<p>Stan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each <span class="math inline">\(\sim\)</span> statement corresponds to a multiplicative factor in the unnormalized posterior density.</p>
+<p>Distribution statements (<code>~</code>) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).</p>
+<p>In Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(y | mu, sigma);</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(mu | <span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(sigma | <span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Stan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g.&nbsp;censored data model) or coding language limitations (e.g.&nbsp;mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.</p>
+<p>In general, a distribution statement of the form</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a>y ~ dist(theta1, ..., thetaN);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>involving subexpressions <code>y</code> and <code>theta1</code> through <code>thetaN</code> (including the case where <code>N</code> is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real <code>y</code> values, the log probability density function is used,</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> dist_lpdf(y | theta1, ..., thetaN);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For those restricted to integer <code>y</code> values, the log probability mass function is used,</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> dist_lpmf(y | theta1, ..., thetaN);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This will be well formed if and only if <code>dist_lpdf(y | theta1,   ..., thetaN)</code> or <code>dist_lpmf(y | theta1, ..., thetaN)</code> is a well-formed expression of type <code>real</code>. User defined distributions can be defined in functions block by using function names ending with <code>_lpdf</code>.</p>
+<section id="log-probability-increment-vs.-distribution-statement" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="log-probability-increment-vs.-distribution-statement">Log probability increment vs.&nbsp;distribution statement</h3>
+<p>Although both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and explicitly incrementing the log probability function, as in</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(y | mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to <code>normal_lpdf</code> adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, <code>y</code>, <code>mu</code>, or <code>sigma</code>, involve only constants, data variables, and transformed data variables. See the section <a href="#built-in-distributions">Built in distributions</a> above discussing <code>_lupdf</code> and <code>_lupmf</code> functions that also drops all the constant terms.</p>
+</section>
+<section id="user-transformed-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="user-transformed-variables">User-transformed variables</h3>
+<p>The left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; beta;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a>  log(beta) ~ normal(mu, sigma);</span>
+<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Unfortunately, this is not enough to properly model <code>beta</code> as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to <code>beta</code> here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure <code>beta</code> gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the <a href="#change-of-variables.section">change of variables section</a> for full definitions. For the case above, the following adjustment will account for the log transform.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> - log(abs(y));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="truncated-distributions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="truncated-distributions">Truncated distributions</h3>
+<p>Stan supports truncating distributions with lower bounds, upper bounds, or both.</p>
+<section id="truncating-with-lower-and-upper-bounds" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncating-with-lower-and-upper-bounds">Truncating with lower and upper bounds</h4>
+<p>A probability density function <span class="math inline">\(p(x)\)</span> for a continuous distribution may be truncated to an interval <span class="math inline">\([a, b]\)</span> to define a new density <span class="math inline">\(p_{[a, b]}(x)\)</span> with support <span class="math inline">\([a, b]\)</span> by setting</p>
+<p><span class="math display">\[
+p_{[a, b]}(x)
+= \frac{p(x)}
+       {\int_a^b p(u) \, du}.
+\]</span></p>
+<p>A probability mass function <span class="math inline">\(p(x)\)</span> for a discrete distribution may be truncated to the closed interval <span class="math inline">\([a, b]\)</span> by</p>
+<p><span class="math display">\[
+p_{[a, b]}(x) = \frac{p(x)}
+                  {\sum_{u = a}^b p(u)}.
+\]</span></p>
+</section>
+<section id="truncating-with-a-lower-bound" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncating-with-a-lower-bound">Truncating with a lower bound</h4>
+<p>A probability density function <span class="math inline">\(p(x)\)</span> can be truncated to <span class="math inline">\([a, \infty]\)</span> by defining</p>
+<p><span class="math display">\[
+p_{[a, \infty]}(x)
+= \frac{p(x)}
+       {\int_a^{\infty} p(u) \, du}.
+\]</span></p>
+<p>A probability mass function <span class="math inline">\(p(x)\)</span> is truncated to <span class="math inline">\([a, \infty]\)</span> by defining</p>
+<p><span class="math display">\[
+p_{[a, \infty]}(x) = \frac{p(x)}
+                  {\sum_{a &lt;= u} p(u)}.
+\]</span></p>
+</section>
+<section id="truncating-with-an-upper-bound" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncating-with-an-upper-bound">Truncating with an upper bound</h4>
+<p>A probability density function <span class="math inline">\(p(x)\)</span> can be truncated to <span class="math inline">\([-\infty, b]\)</span> by defining</p>
+<p><span class="math display">\[
+p_{[-\infty, b]}(x)
+= \frac{p(x)}
+       {\int_{-\infty}^b p(u) \, du}.
+\]</span></p>
+<p>A probability mass function <span class="math inline">\(p(x)\)</span> is truncated to <span class="math inline">\([-\infty, b]\)</span> by defining</p>
+<p><span class="math display">\[
+p_{[-\infty,b]}(x) = \frac{p(x)}
+                  {\sum_{u &lt;= b} p(u)}.
+\]</span></p>
+</section>
+<section id="cumulative-distribution-functions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="cumulative-distribution-functions">Cumulative distribution functions</h4>
+<p>Given a probability function <span class="math inline">\(p_X(x)\)</span> for a random variable <span class="math inline">\(X\)</span>, its cumulative distribution function (cdf) <span class="math inline">\(F_X(x)\)</span> is defined to be the probability that <span class="math inline">\(X \leq x\)</span>,</p>
+<p><span class="math display">\[
+F_X(x) = \Pr[X \leq x].
+\]</span></p>
+<p>The upper-case variable <span class="math inline">\(X\)</span> is the random variable whereas the lower-case variable <span class="math inline">\(x\)</span> is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to</p>
+<p><span class="math display">\[
+F_X(x) \ = \ \int_{-\infty}^{x} p_X(u) \, du,
+\]</span></p>
+<p>For discrete variables, the cdf is defined to include the upper bound given by the argument,</p>
+<p><span class="math display">\[
+F_X(x) = \sum_{u \leq x} p_X(u).
+\]</span></p>
+</section>
+<section id="complementary-cumulative-distribution-functions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="complementary-cumulative-distribution-functions">Complementary cumulative distribution functions</h4>
+<p>The complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by</p>
+<p><span class="math display">\[
+F^C_X(x)
+\ = \ \Pr[X &gt; x]
+\ = \ 1 - F_X(x).
+\]</span></p>
+<p>Unlike the cdf, the ccdf is exclusive of the bound, hence the event <span class="math inline">\(X &gt; x\)</span> rather than the cdf’s event <span class="math inline">\(X \leq x\)</span>.</p>
+<p>For continuous distributions, the ccdf works out to</p>
+<p><span class="math display">\[
+F^C_X(x)
+\ = \ 1 - \int_{-\infty}^x p_X(u) \, du
+\ = \ \int_x^{\infty} p_X(u) \, du.
+\]</span></p>
+<p>The lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over <span class="math inline">\(u &gt; x\)</span>,</p>
+<p><span class="math display">\[
+F^C_X(x)
+\ = \ 1 - \sum_{u \leq x} p_X(u)
+\ = \ \sum_{u &gt; x} p_X(u).
+\]</span></p>
+<p>Cumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by <span class="math display">\[
+\int_a^b p(u) \, du = F_X(b) - F_X(a).
+\]</span> This allows truncated distributions to be defined as <span class="math display">\[
+p_{[a,b]}(x) = \frac{p_X(x)}
+                  {F_X(b) - F_X(a)}.
+\]</span></p>
+<p>For discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from <span class="math inline">\(F_X(b) - F_X(a)\)</span>,</p>
+<p><span class="math display">\[
+p_{[a,b]}(x) = \frac{p_X(x)}
+                  {F_X(b) - F_X(a) + p_X(a)}.
+\]</span></p>
+</section>
+<section id="truncation-with-lower-and-upper-bounds-in-stan" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncation-with-lower-and-upper-bounds-in-stan">Truncation with lower and upper bounds in Stan</h4>
+<p>Stan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to <span class="math inline">\([-0.5, 2.1]\)</span> can be coded with the following distribution statement.</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>) <span class="kw">T</span>[-<span class="fl">0.5</span>, <span class="fl">2.1</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Truncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.</p>
+<p>In general, the truncation bounds and parameters may be parameters or local variables.</p>
+<p>Because the example above involves a continuous distribution, it behaves the same way as the following more verbose form.</p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y &lt; -<span class="fl">0.5</span> || y &gt; <span class="fl">2.1</span>) {</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> negative_infinity();</span>
+<span id="cb28-4"><a href="#cb28-4" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb28-5"><a href="#cb28-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -log_diff_exp(normal_lcdf(<span class="fl">2.1</span> | <span class="dv">0</span>, <span class="dv">1</span>),</span>
+<span id="cb28-6"><a href="#cb28-6" aria-hidden="true" tabindex="-1"></a>                          normal_lcdf(-<span class="fl">0.5</span> | <span class="dv">0</span>, <span class="dv">1</span>));</span>
+<span id="cb28-7"><a href="#cb28-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because a Stan program defines a log density function, all calculations are on the log scale. The function <code>normal_lcdf</code> is the log of the cumulative normal distribution function and the function <code>log_diff_exp(a, b)</code> is a more arithmetically stable form of <code>log(exp(a) - exp(b))</code>.</p>
+<p>For a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(<span class="fl">3.7</span>) <span class="kw">T</span>[<span class="dv">2</span>, <span class="dv">10</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>behaves in the same way as the following code.</p>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(<span class="fl">3.7</span>);</span>
+<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y &lt; <span class="dv">2</span> || y &gt; <span class="dv">10</span>) {</span>
+<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> negative_infinity();</span>
+<span id="cb30-4"><a href="#cb30-4" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb30-5"><a href="#cb30-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -log_sum_exp(poisson_lpmf(<span class="dv">2</span> | <span class="fl">3.7</span>),</span>
+<span id="cb30-6"><a href="#cb30-6" aria-hidden="true" tabindex="-1"></a>                         log_diff_exp(poisson_lcdf(<span class="dv">10</span> | <span class="fl">3.7</span>),</span>
+<span id="cb30-7"><a href="#cb30-7" aria-hidden="true" tabindex="-1"></a>                                      poisson_lcdf(<span class="dv">2</span> | <span class="fl">3.7</span>)));</span>
+<span id="cb30-8"><a href="#cb30-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Recall that <code>log_sum_exp(a, b)</code> is just the arithmetically stable form of <code>log(exp(a) + exp(b))</code>.</p>
+</section>
+<section id="truncation-with-lower-bounds-in-stan" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncation-with-lower-bounds-in-stan">Truncation with lower bounds in Stan</h4>
+<p>For truncating with only a lower bound, the upper limit is left blank.</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>) <span class="kw">T</span>[-<span class="fl">0.5</span>, ];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This truncated distribution statement has the same behavior as the following code.</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y &lt; -<span class="fl">0.5</span>) {</span>
+<span id="cb32-3"><a href="#cb32-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> negative_infinity();</span>
+<span id="cb32-4"><a href="#cb32-4" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb32-5"><a href="#cb32-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -normal_lccdf(-<span class="fl">0.5</span> | <span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb32-6"><a href="#cb32-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>normal_lccdf</code> function is the normal complementary cumulative distribution function.</p>
+<p>As with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus</p>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(<span class="fl">3.7</span>) <span class="kw">T</span>[<span class="dv">2</span>, ];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>behaves the same way as</p>
+<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(<span class="fl">3.7</span>);</span>
+<span id="cb34-2"><a href="#cb34-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y &lt; <span class="dv">2</span>) {</span>
+<span id="cb34-3"><a href="#cb34-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> negative_infinity();</span>
+<span id="cb34-4"><a href="#cb34-4" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb34-5"><a href="#cb34-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -log_sum_exp(poisson_lpmf(<span class="dv">2</span> | <span class="fl">3.7</span>),</span>
+<span id="cb34-6"><a href="#cb34-6" aria-hidden="true" tabindex="-1"></a>                         poisson_lccdf(<span class="dv">2</span> | <span class="fl">3.7</span>));</span>
+<span id="cb34-7"><a href="#cb34-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="truncation-with-upper-bounds-in-stan" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncation-with-upper-bounds-in-stan">Truncation with upper bounds in Stan</h4>
+<p>To truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement</p>
+<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>) <span class="kw">T</span>[ , <span class="fl">2.1</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>produces the same result as the following code.</p>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(y | <span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb36-2"><a href="#cb36-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y &gt; <span class="fl">2.1</span>) {</span>
+<span id="cb36-3"><a href="#cb36-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> negative_infinity();</span>
+<span id="cb36-4"><a href="#cb36-4" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb36-5"><a href="#cb36-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -normal_lcdf(<span class="fl">2.1</span> | <span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb36-6"><a href="#cb36-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement</p>
+<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(<span class="fl">3.7</span>) <span class="kw">T</span>[ , <span class="dv">10</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>behaves the same way as the following code.</p>
+<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(<span class="fl">3.7</span>);</span>
+<span id="cb38-2"><a href="#cb38-2" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y &gt; <span class="dv">10</span>) {</span>
+<span id="cb38-3"><a href="#cb38-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> negative_infinity();</span>
+<span id="cb38-4"><a href="#cb38-4" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb38-5"><a href="#cb38-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -poisson_lcdf(<span class="dv">10</span> | <span class="fl">3.7</span>);</span>
+<span id="cb38-6"><a href="#cb38-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="cumulative-distributions-must-be-defined" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="cumulative-distributions-must-be-defined">Cumulative distributions must be defined</h4>
+<p>In all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.</p>
+</section>
+<section id="type-constraints-on-bounds" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="type-constraints-on-bounds">Type constraints on bounds</h4>
+<p>For continuous distributions, truncation points must be expressions of type <code>int</code> or <code>real</code>. For discrete distributions, truncation points must be expressions of type <code>int</code>.</p>
+</section>
+<section id="variates-outside-of-truncation-bounds" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="variates-outside-of-truncation-bounds">Variates outside of truncation bounds</h4>
+<p>For a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds <span class="math inline">\(-\infty\)</span> to the total log probability, which in turn results in the sample being rejected.</p>
+</section>
+<section id="vectorizing-truncated-distributions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="vectorizing-truncated-distributions">Vectorizing truncated distributions</h4>
+<p>Vectorization of distribution functions with truncation is available if the underlying distribution, <code>lcdf</code>, and <code>lccdf</code> functions meet the required signatures.</p>
+<p>The equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):</p>
+<ol type="1">
+<li><p>If the variate <code>y</code> is the only non-scalar, the result is the same as described in the above sections, but the <code>lcdf</code>/<code>lccdf</code> calculation is multiplied by <code>size(y)</code>.</p></li>
+<li><p>If the other arguments to the distribution are non-scalars, then the vectorized version of the <code>lcdf</code>/<code>lccdf</code> is used. These functions return the sum of their terms, so no multiplication by the size is needed.</p></li>
+<li><p>The exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the <code>log_diff_exp</code> of two sums is not the same as the sum of the pairwise <code>log_diff_exp</code> operations.</p></li>
+</ol>
+<p>Note that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.</p>
+</section>
+</section>
+</section>
+<section id="for-loops" class="level2">
+<h2 class="anchored" data-anchor-id="for-loops">For loops</h2>
+<p>Suppose <code>N</code> is a variable of type <code>int</code>, <code>y</code> is a one-dimensional array of type <code>array[] real</code>, and <code>mu</code> and <code>sigma</code> are variables of type <code>real</code>. Furthermore, suppose that <code>n</code> has not been defined as a variable. Then the following is a well-formed for-loop statement.</p>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb39-2"><a href="#cb39-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu, sigma);</span>
+<span id="cb39-3"><a href="#cb39-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The loop variable is <code>n</code>, the loop bounds are the values in the range <code>1:N</code>, and the body is the statement following the loop bounds.</p>
+<section id="loop-variable-typing-and-scope" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="loop-variable-typing-and-scope">Loop variable typing and scope</h3>
+<p>The type of the loop variable is <code>int</code>. Unlike in C++ and similarly to R, this variable must not be declared explicitly.</p>
+<p>The bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range <code>L:H</code> will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to <code>L</code> and less than or equal to <code>H</code>. For example, the loop <code>for (n in 2:5)</code> will cause the body of the for loop to be executed with <code>n</code> equal to 2, 3, 4, and 5, in order. The variable and bound <code>for (n in   5:2)</code> will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.</p>
+<p>The scope of the loop variable is limited to the body of the loop.</p>
+</section>
+<section id="order-sensitivity-and-repeated-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="order-sensitivity-and-repeated-variables">Order sensitivity and repeated variables</h3>
+<p>Unlike in BUGS, Stan allows variables to be reassigned. For example, the variable <code>theta</code> in the following program is reassigned in each iteration of the loop.</p>
+<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb40-2"><a href="#cb40-2" aria-hidden="true" tabindex="-1"></a>  theta = inv_logit(alpha + x[n] * beta);</span>
+<span id="cb40-3"><a href="#cb40-3" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli(theta);</span>
+<span id="cb40-4"><a href="#cb40-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Such reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
+<p>In Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.</p>
+<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb41-2"><a href="#cb41-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli(theta);</span>
+<span id="cb41-3"><a href="#cb41-3" aria-hidden="true" tabindex="-1"></a>  theta = inv_logit(alpha + x[n] * beta);</span>
+<span id="cb41-4"><a href="#cb41-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this program, <code>theta</code> is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.</p>
+<p>Stan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.</p>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a>total = <span class="fl">0.0</span>;</span>
+<span id="cb42-2"><a href="#cb42-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb42-3"><a href="#cb42-3" aria-hidden="true" tabindex="-1"></a>  total = total + x[n];</span>
+<span id="cb42-4"><a href="#cb42-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After the for loop is executed, the variable <code>total</code> will hold the sum of the elements in the array <code>x</code>. This example was purely pedagogical; it is easier and more efficient to write</p>
+<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a>total = sum(x);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.</p>
+<div class="sourceCode" id="cb44"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb44-1"><a href="#cb44-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">100</span>) {</span>
+<span id="cb44-2"><a href="#cb44-2" aria-hidden="true" tabindex="-1"></a>  y += y * epsilon;</span>
+<span id="cb44-3"><a href="#cb44-3" aria-hidden="true" tabindex="-1"></a>  epsilon = <span class="fl">0.5</span> * epsilon;</span>
+<span id="cb44-4"><a href="#cb44-4" aria-hidden="true" tabindex="-1"></a>  y += y * epsilon;</span>
+<span id="cb44-5"><a href="#cb44-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="foreach-loops" class="level2">
+<h2 class="anchored" data-anchor-id="foreach-loops">Foreach loops</h2>
+<p>A second form of for loops allows iteration over elements of containers. If <code>ys</code> is an expression denoting a container (vector, row vector, matrix, or array) with elements of type <code>T</code>, then the following is a well-formed foreach statement.</p>
+<div class="sourceCode" id="cb45"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb45-1"><a href="#cb45-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (y <span class="cf">in</span> ys) {</span>
+<span id="cb45-2"><a href="#cb45-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... do something with y ...</span></span>
+<span id="cb45-3"><a href="#cb45-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The order in which elements of <code>ys</code> are visited is defined for container types as follows.</p>
+<ul>
+<li><p><code>vector</code>, <code>row_vector</code>: elements visited in order, <code>y</code> is of type <code>double</code></p></li>
+<li><p><code>matrix</code>: elements visited in column-major order, <code>y</code> is of type <code>double</code></p></li>
+<li><p><code>array[] T</code>: elements visited in order, <code>y</code> is of type <code>T</code>.</p></li>
+</ul>
+<p>Consequently, if <code>ys</code> is a two dimensional array <code>array[,] real</code>, <code>y</code> will be a one-dimensional array of real values (type <code>array[] real</code>). If ’ys<code>is a matrix, then</code>y<code>will be a real value (type</code>real`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,</p>
+<div class="sourceCode" id="cb46"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb46-1"><a href="#cb46-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>] <span class="dt">real</span> yss;</span>
+<span id="cb46-2"><a href="#cb46-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (ys <span class="cf">in</span> yss) {</span>
+<span id="cb46-3"><a href="#cb46-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (y <span class="cf">in</span> ys) {</span>
+<span id="cb46-4"><a href="#cb46-4" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ... do something with y ...</span></span>
+<span id="cb46-5"><a href="#cb46-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb46-6"><a href="#cb46-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>whereas a matrix can be looped over in one foreach statement</p>
+<div class="sourceCode" id="cb47"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb47-1"><a href="#cb47-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">2</span>, <span class="dv">3</span>] yss;</span>
+<span id="cb47-2"><a href="#cb47-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (y <span class="cf">in</span> yss) {</span>
+<span id="cb47-3"><a href="#cb47-3" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... do something with y...</span></span>
+<span id="cb47-4"><a href="#cb47-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In both cases, the loop variable <code>y</code> is of type <code>real.  The elements of the matrix are visited in column-major order (e.g.,</code>y[1, 1]<code>,</code>y[2, 1]<code>,</code>y[1, 2]<code>, ...,</code>y[2, 3]<code>), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,</code>y[1, 1]<code>,</code>y[1, 2]<code>,</code>y[1, 3]<code>,</code>y[2, 1]<code>, ...,</code>y[2, 3]`).</p>
+</section>
+<section id="conditional-statements" class="level2">
+<h2 class="anchored" data-anchor-id="conditional-statements">Conditional statements</h2>
+<p>Stan supports full conditional statements using the same if-then-else syntax as C++. The general format is</p>
+<div class="sourceCode" id="cb48"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb48-1"><a href="#cb48-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (condition1)</span>
+<span id="cb48-2"><a href="#cb48-2" aria-hidden="true" tabindex="-1"></a>  statement1</span>
+<span id="cb48-3"><a href="#cb48-3" aria-hidden="true" tabindex="-1"></a><span class="cf">else</span> <span class="cf">if</span> (condition2)</span>
+<span id="cb48-4"><a href="#cb48-4" aria-hidden="true" tabindex="-1"></a>  statement2</span>
+<span id="cb48-5"><a href="#cb48-5" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb48-6"><a href="#cb48-6" aria-hidden="true" tabindex="-1"></a><span class="cf">else</span> <span class="cf">if</span> (conditionN<span class="dv">-1</span>)</span>
+<span id="cb48-7"><a href="#cb48-7" aria-hidden="true" tabindex="-1"></a>  statementN<span class="dv">-1</span></span>
+<span id="cb48-8"><a href="#cb48-8" aria-hidden="true" tabindex="-1"></a><span class="cf">else</span></span>
+<span id="cb48-9"><a href="#cb48-9" aria-hidden="true" tabindex="-1"></a>  statementN</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There must be a single leading <code>if</code> clause, which may be followed by any number of <code>else if</code> clauses, all of which may be optionally followed by an <code>else</code> clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.</p>
+<p>The entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.</p>
+</section>
+<section id="while-statements" class="level2">
+<h2 class="anchored" data-anchor-id="while-statements">While statements</h2>
+<p>Stan supports standard while loops using the same syntax as C++. The general format is as follows.</p>
+<div class="sourceCode" id="cb49"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb49-1"><a href="#cb49-1" aria-hidden="true" tabindex="-1"></a><span class="cf">while</span> (condition)</span>
+<span id="cb49-2"><a href="#cb49-2" aria-hidden="true" tabindex="-1"></a>  body</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).</p>
+<p>Evaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.</p>
+<p>The rest of the body of a while loop may be skipped using a <code>continue</code>. The loop will be exited with a <code>break</code> statement. See the <a href="#break-continue-statements">section on continue and break statements</a> for more details.</p>
+</section>
+<section id="statement-blocks-and-local-variable-declarations" class="level2">
+<h2 class="anchored" data-anchor-id="statement-blocks-and-local-variable-declarations">Statement blocks and local variable declarations</h2>
+<p>Just as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.</p>
+<section id="blocks-in-for-loops" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="blocks-in-for-loops">Blocks in for loops</h3>
+<p>Blocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.</p>
+<div class="sourceCode" id="cb50"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb50-1"><a href="#cb50-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb50-2"><a href="#cb50-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu, sigma);</span>
+<span id="cb50-3"><a href="#cb50-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To put multiple statements inside the body of a for loop, a block is used, as in the following example.</p>
+<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb51-2"><a href="#cb51-2" aria-hidden="true" tabindex="-1"></a>  lambda[n] ~ gamma(alpha, beta);</span>
+<span id="cb51-3"><a href="#cb51-3" aria-hidden="true" tabindex="-1"></a>  y[n] ~ poisson(lambda[n]);</span>
+<span id="cb51-4"><a href="#cb51-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The open curly bracket (<code>{</code>) is the first character of the block and the close curly bracket (<code>}</code>) is the last character.</p>
+<p>Because whitespace is ignored in Stan, the following program will not compile.</p>
+<div class="sourceCode" id="cb52"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb52-1"><a href="#cb52-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N)</span>
+<span id="cb52-2"><a href="#cb52-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu, sigma);</span>
+<span id="cb52-3"><a href="#cb52-3" aria-hidden="true" tabindex="-1"></a>  z[n] ~ normal(mu, sigma); <span class="co">// ERROR!</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The problem is that the body of the for loop is taken to be the statement directly following it, which is <code>y[n] ~ normal(mu, sigma)</code>. This leaves the probability statement for <code>z[n]</code> hanging, as is clear from the following equivalent program.</p>
+<div class="sourceCode" id="cb53"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb53-1"><a href="#cb53-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb53-2"><a href="#cb53-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu, sigma);</span>
+<span id="cb53-3"><a href="#cb53-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb53-4"><a href="#cb53-4" aria-hidden="true" tabindex="-1"></a>z[n] ~ normal(mu, sigma); <span class="co">// ERROR!</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Neither of these programs will compile. If the loop variable <code>n</code> was defined before the for loop, the for-loop declaration will raise an error. If the loop variable <code>n</code> was not defined before the for loop, then the use of the expression <code>z[n]</code> will raise an error.</p>
+</section>
+<section id="local-variable-declarations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="local-variable-declarations">Local variable declarations</h3>
+<p>A for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.</p>
+<div class="sourceCode" id="cb54"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb54-1"><a href="#cb54-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb54-2"><a href="#cb54-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;</span>
+<span id="cb54-3"><a href="#cb54-3" aria-hidden="true" tabindex="-1"></a>  theta = inv_logit(alpha + x[n] * beta);</span>
+<span id="cb54-4"><a href="#cb54-4" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli(theta);</span>
+<span id="cb54-5"><a href="#cb54-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The local variable <code>theta</code> is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus <code>theta</code> is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable <code>theta</code> if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.</p>
+<div class="sourceCode" id="cb55"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb55-1"><a href="#cb55-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb55-2"><a href="#cb55-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;</span>
+<span id="cb55-3"><a href="#cb55-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb55-4"><a href="#cb55-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> theta; <span class="co">// ERROR!</span></span>
+<span id="cb55-5"><a href="#cb55-5" aria-hidden="true" tabindex="-1"></a>    theta = inv_logit(alpha + x[m, n] * beta);</span>
+<span id="cb55-6"><a href="#cb55-6" aria-hidden="true" tabindex="-1"></a>    y[m, n] ~ bernoulli(theta);</span>
+<span id="cb55-7"><a href="#cb55-7" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The compiler will flag the second declaration of <code>theta</code> with a message that it is already defined.</p>
+</section>
+<section id="no-constraints-on-local-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="no-constraints-on-local-variables">No constraints on local variables</h3>
+<p>Local variables may not have constraints on their declaration. The only types that may be used are listed in the <a href="../reference-manual/types.html#id:constrained-types.figure">types table</a> under “local”.</p>
+</section>
+<section id="blocks-within-blocks" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="blocks-within-blocks">Blocks within blocks</h3>
+<p>A block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following</p>
+<div class="sourceCode" id="cb56"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb56-1"><a href="#cb56-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb56-2"><a href="#cb56-2" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb56-3"><a href="#cb56-3" aria-hidden="true" tabindex="-1"></a>     <span class="dt">int</span> n = <span class="dv">2</span> * m;</span>
+<span id="cb56-4"><a href="#cb56-4" aria-hidden="true" tabindex="-1"></a>     sum += n;</span>
+<span id="cb56-5"><a href="#cb56-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb56-6"><a href="#cb56-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb56-7"><a href="#cb56-7" aria-hidden="true" tabindex="-1"></a>    sum += x[m, n];</span>
+<span id="cb56-8"><a href="#cb56-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb56-9"><a href="#cb56-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The variable declaration <code>int n;</code> is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of <code>n</code> are unrelated.</p>
+</section>
+</section>
+<section id="break-continue-statements" class="level2">
+<h2 class="anchored" data-anchor-id="break-continue-statements">Break and continue statements</h2>
+<p>The one-token statements <code>continue</code> and <code>break</code> may be used within loops to alter control flow; <code>continue</code> causes the next iteration of the loop to run immediately, whereas <code>break</code> terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both <code>break</code> and <code>continue</code> scope to the most deeply nested loop, but pass through non-loop statements.</p>
+<p>Although these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.</p>
+<section id="break-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="break-statements">Break statements</h3>
+<p>When a <code>break</code> statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:</p>
+<div class="sourceCode" id="cb57"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb57-1"><a href="#cb57-1" aria-hidden="true" tabindex="-1"></a><span class="cf">while</span> (<span class="dv">1</span>) {</span>
+<span id="cb57-2"><a href="#cb57-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (n &lt; <span class="dv">0</span>) {</span>
+<span id="cb57-3"><a href="#cb57-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">break</span>;</span>
+<span id="cb57-4"><a href="#cb57-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb57-5"><a href="#cb57-5" aria-hidden="true" tabindex="-1"></a>  foo(n);</span>
+<span id="cb57-6"><a href="#cb57-6" aria-hidden="true" tabindex="-1"></a>  n = n - <span class="dv">1</span>;</span>
+<span id="cb57-7"><a href="#cb57-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>while~(1)</code> loop is a “forever” loop, because <code>1</code> is the true value, so the test always succeeds. Within the loop, if the value of <code>n</code> is less than 0, the loop terminates, otherwise it executes <code>foo(n)</code> and then decrements <code>n</code>. The statement above does exactly the same thing as</p>
+<div class="sourceCode" id="cb58"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb58-1"><a href="#cb58-1" aria-hidden="true" tabindex="-1"></a><span class="cf">while</span> (n &gt;= <span class="dv">0</span>) {</span>
+<span id="cb58-2"><a href="#cb58-2" aria-hidden="true" tabindex="-1"></a>  foo(n);</span>
+<span id="cb58-3"><a href="#cb58-3" aria-hidden="true" tabindex="-1"></a>  n = n - <span class="dv">1</span>;</span>
+<span id="cb58-4"><a href="#cb58-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This case is simply illustrative of the behavior; it is not a case where a <code>break</code> simplifies the loop.</p>
+</section>
+<section id="continue-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="continue-statements">Continue statements</h3>
+<p>The <code>continue</code> statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array <code>x</code>,</p>
+<div class="sourceCode" id="cb59"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb59-1"><a href="#cb59-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum;</span>
+<span id="cb59-2"><a href="#cb59-2" aria-hidden="true" tabindex="-1"></a>sum = <span class="dv">0</span>;</span>
+<span id="cb59-3"><a href="#cb59-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:size(x)) {</span>
+<span id="cb59-4"><a href="#cb59-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (x[n] &lt;= <span class="dv">0</span>) {</span>
+<span id="cb59-5"><a href="#cb59-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">continue</span>;</span>
+<span id="cb59-6"><a href="#cb59-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb59-7"><a href="#cb59-7" aria-hidden="true" tabindex="-1"></a>  sum += x[n];</span>
+<span id="cb59-8"><a href="#cb59-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an <code>x[n]</code> less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,</p>
+<div class="sourceCode" id="cb60"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb60-1"><a href="#cb60-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum;</span>
+<span id="cb60-2"><a href="#cb60-2" aria-hidden="true" tabindex="-1"></a>sum = <span class="dv">0</span>;</span>
+<span id="cb60-3"><a href="#cb60-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:size(x)) {</span>
+<span id="cb60-4"><a href="#cb60-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (x[n] &gt; <span class="dv">0</span>) {</span>
+<span id="cb60-5"><a href="#cb60-5" aria-hidden="true" tabindex="-1"></a>    sum += x[n];</span>
+<span id="cb60-6"><a href="#cb60-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb60-7"><a href="#cb60-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>While the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.</p>
+</section>
+<section id="breaking-and-continuing-nested-loops" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="breaking-and-continuing-nested-loops">Breaking and continuing nested loops</h3>
+<p>If there is a loop nested within a loop, a <code>break</code> or <code>continue</code> statement only breaks out of the inner loop. So</p>
+<div class="sourceCode" id="cb61"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb61-1"><a href="#cb61-1" aria-hidden="true" tabindex="-1"></a><span class="cf">while</span> (cond1) {</span>
+<span id="cb61-2"><a href="#cb61-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb61-3"><a href="#cb61-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">while</span> (cond2) {</span>
+<span id="cb61-4"><a href="#cb61-4" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ...</span></span>
+<span id="cb61-5"><a href="#cb61-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (cond3) {</span>
+<span id="cb61-6"><a href="#cb61-6" aria-hidden="true" tabindex="-1"></a>      <span class="cf">break</span>;</span>
+<span id="cb61-7"><a href="#cb61-7" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb61-8"><a href="#cb61-8" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ...</span></span>
+<span id="cb61-9"><a href="#cb61-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb61-10"><a href="#cb61-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// execution continues here after break</span></span>
+<span id="cb61-11"><a href="#cb61-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb61-12"><a href="#cb61-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the break is triggered by <code>cond3</code> being true, execution will continue after the nested loop.</p>
+<p>As with break statements, continue statements go back to the top of the most deeply nested loop in which the <code>continue</code> appears.</p>
+<p>Although break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.</p>
+</section>
+</section>
+<section id="print-statements.section" class="level2">
+<h2 class="anchored" data-anchor-id="print-statements.section">Print statements</h2>
+<p>Stan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.</p>
+<div class="sourceCode" id="cb62"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb62-1"><a href="#cb62-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) { <span class="kw">print</span>(<span class="st">"loop iteration: "</span>, n); ... }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression <code>n</code>, followed by a new line.</p>
+<section id="print-content" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="print-content">Print content</h3>
+<p>The text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.</p>
+<p>Printing a simple <code>real</code> or <code>int</code> typed variable always prints the variable’s value.<a href="#fn5" class="footnote-ref" id="fnref5" role="doc-noteref"><sup>5</sup></a></p>
+<p>For array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as</p>
+<div class="sourceCode" id="cb63"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb63-1"><a href="#cb63-1" aria-hidden="true" tabindex="-1"></a>[<span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span>]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and a <span class="math inline">\(2 \times 3\)</span>-matrix as</p>
+<div class="sourceCode" id="cb64"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb64-1"><a href="#cb64-1" aria-hidden="true" tabindex="-1"></a>[[<span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span>], [<span class="dv">4</span>, <span class="dv">5</span>, <span class="dv">6</span>]]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex numbers print as pairs. For example, the pair of statements</p>
+<div class="sourceCode" id="cb65"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb65-1"><a href="#cb65-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = to_complex(<span class="fl">1.2</span>, -<span class="fl">3.5</span>);</span>
+<span id="cb65-2"><a href="#cb65-2" aria-hidden="true" tabindex="-1"></a><span class="kw">print</span>(z)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>will print as <code>(1.2,-3.5)</code>, with no space after the comma or within the parentheses.</p>
+<p>Printing a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.</p>
+<div class="sourceCode" id="cb66"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb66-1"><a href="#cb66-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb66-2"><a href="#cb66-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">2</span>, <span class="dv">2</span>] u;</span>
+<span id="cb66-3"><a href="#cb66-3" aria-hidden="true" tabindex="-1"></a>  u[<span class="dv">1</span>, <span class="dv">1</span>] = <span class="fl">1.0</span>;   u[<span class="dv">1</span>, <span class="dv">2</span>] = <span class="fl">4.0</span>;</span>
+<span id="cb66-4"><a href="#cb66-4" aria-hidden="true" tabindex="-1"></a>  u[<span class="dv">2</span>, <span class="dv">1</span>] = <span class="fl">9.0</span>;   u[<span class="dv">2</span>, <span class="dv">2</span>] = <span class="fl">16.0</span>;</span>
+<span id="cb66-5"><a href="#cb66-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">2</span>) {</span>
+<span id="cb66-6"><a href="#cb66-6" aria-hidden="true" tabindex="-1"></a>    <span class="kw">print</span>(<span class="st">"u["</span>, n, <span class="st">"] = "</span>, u[n]);</span>
+<span id="cb66-7"><a href="#cb66-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb66-8"><a href="#cb66-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This print statement executes twice, printing the following two lines of output.</p>
+<div class="sourceCode" id="cb67"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb67-1"><a href="#cb67-1" aria-hidden="true" tabindex="-1"></a>u[<span class="dv">1</span>] = [<span class="dv">1</span>, <span class="dv">4</span>]</span>
+<span id="cb67-2"><a href="#cb67-2" aria-hidden="true" tabindex="-1"></a>u[<span class="dv">2</span>] = [<span class="dv">9</span>, <span class="dv">16</span>]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="non-void-input" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="non-void-input">Non-void input</h3>
+<p>The input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.</p>
+</section>
+<section id="print-frequency" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="print-frequency">Print frequency</h3>
+<p>Printing for a print statement happens every time it is executed. The <code>transformed data</code> block is executed once per chain, the <code>transformed parameter</code> and <code>model</code> blocks once per leapfrog step, and the <code>generated quantities</code> block once per iteration.</p>
+</section>
+<section id="string-literals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="string-literals">String literals</h3>
+<p>String literals begin and end with a double quote character (<code>"</code>). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.</p>
+<p>The Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.</p>
+</section>
+<section id="debug-by-print" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="debug-by-print">Debug by <code>print</code></h3>
+<p>Because Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.</p>
+<p>It is particularly useful to print the value of the target log density accumulator (through the <code>target()</code> function), as in the following example.</p>
+<div class="sourceCode" id="cb68"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb68-1"><a href="#cb68-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">2</span>] y;</span>
+<span id="cb68-2"><a href="#cb68-2" aria-hidden="true" tabindex="-1"></a>y[<span class="dv">1</span>] = <span class="dv">1</span>;</span>
+<span id="cb68-3"><a href="#cb68-3" aria-hidden="true" tabindex="-1"></a><span class="kw">print</span>(<span class="st">"log density before ="</span>, target());</span>
+<span id="cb68-4"><a href="#cb68-4" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>,<span class="dv">1</span>);  <span class="co">// bug!  y[2] not defined</span></span>
+<span id="cb68-5"><a href="#cb68-5" aria-hidden="true" tabindex="-1"></a><span class="kw">print</span>(<span class="st">"log density after ="</span>, target());</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The example has a bug in that <code>y[2]</code> is not defined before the vector <code>y</code> is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).</p>
+<p>Note that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.</p>
+</section>
+</section>
+<section id="reject-statements.section" class="level2">
+<h2 class="anchored" data-anchor-id="reject-statements.section">Reject statements</h2>
+<p>The Stan <code>reject</code> statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.</p>
+<p>Like the <code>print</code> statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.</p>
+<p>Reject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable <code>x</code>’s value is negative.</p>
+<div class="sourceCode" id="cb69"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb69-1"><a href="#cb69-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (x &lt; <span class="dv">0</span>) {</span>
+<span id="cb69-2"><a href="#cb69-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">reject</span>(<span class="st">"x must not be negative; found x="</span>, x);</span>
+<span id="cb69-3"><a href="#cb69-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="behavior-of-reject-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="behavior-of-reject-statements">Behavior of reject statements</h3>
+<p>Reject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the <code>normal_lpdf</code> function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.</p>
+<p>In all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.</p>
+<section id="rejections-in-functions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="rejections-in-functions">Rejections in functions</h4>
+<p>Rejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.</p>
+</section>
+<section id="fatal-exception-contexts" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="fatal-exception-contexts">Fatal exception contexts</h4>
+<p>Rejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.</p>
+<p>Reject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.</p>
+<p>Fatal errors in other blocks may also be signaled by use of the <code>fatal_error</code> statement.</p>
+</section>
+<section id="recoverable-rejection-contexts" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="recoverable-rejection-contexts">Recoverable rejection contexts</h4>
+<p>Rejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a <span class="math inline">\(-\infty\)</span> log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.</p>
+<p>If the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.</p>
+</section>
+</section>
+<section id="rejection-is-not-for-constraints" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="rejection-is-not-for-constraints">Rejection is not for constraints</h3>
+<p>Rejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.</p>
+<div class="sourceCode" id="cb70"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb70-1"><a href="#cb70-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb70-2"><a href="#cb70-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb70-3"><a href="#cb70-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=a&gt; b;</span>
+<span id="cb70-4"><a href="#cb70-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=a, <span class="kw">upper</span>=b&gt; theta;</span>
+<span id="cb70-5"><a href="#cb70-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb70-6"><a href="#cb70-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb70-7"><a href="#cb70-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb70-8"><a href="#cb70-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// **wrong** needs explicit truncation</span></span>
+<span id="cb70-9"><a href="#cb70-9" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb70-10"><a href="#cb70-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb70-11"><a href="#cb70-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This program is wrong because its truncation bounds on <code>theta</code> depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.</p>
+<div class="sourceCode" id="cb71"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb71-1"><a href="#cb71-1" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>) <span class="kw">T</span>[a, b];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over <span class="math inline">\([a ,b]\)</span>; in these simple univariate cases, we can overcome that with the <code>T[ , ]</code> notation, which essentially divides by whatever the prior integrates to over <span class="math inline">\([a, b]\)</span>.</p>
+<p>This problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.</p>
+<div class="sourceCode" id="cb72"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb72-1"><a href="#cb72-1" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (theta &lt; a || theta &gt; b) {</span>
+<span id="cb72-2"><a href="#cb72-2" aria-hidden="true" tabindex="-1"></a>    <span class="kw">reject</span>(<span class="st">"theta not in (a, b)"</span>);</span>
+<span id="cb72-3"><a href="#cb72-3" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb72-4"><a href="#cb72-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// still **wrong**, needs T[a,b]</span></span>
+<span id="cb72-5"><a href="#cb72-5" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.</p>
+<p>Even if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.</p>
+</section>
+</section>
+<section id="fatal-error-statements" class="level2">
+<h2 class="anchored" data-anchor-id="fatal-error-statements">Fatal error statements</h2>
+<p>The Stan <code>fatal_error</code> statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.</p>
+<p>Like the <code>print</code> or <code>reject</code> statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.</p>
+<p>The fatal error may be used to signal an unrecoverable error in blocks where <code>reject</code> leads to the algorithm attempting to try again, such as the <code>model</code> block.</p>
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The current notation replaces two previous versions. Originally, a variable <code>lp__</code> was directly exposed and manipulated; this is no longer allowed. The original statement syntax for <code>target += u</code> was <code>increment_log_prob(u)</code>, but this form was removed in Stan 2.33<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Writing this model with the expression <code>-0.5 * y * y</code> is more efficient than with the equivalent expression <code>y * y / -2</code> because multiplication is more efficient than division; in both cases, the negation is rolled into the numeric literal (<code>-0.5</code> and <code>-2</code>). Writing <code>square(y)</code> instead of <code>y * y</code> would be even more efficient because the derivatives can be precomputed, reducing the memory and number of operations required for automatic differentiation.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>Because <span class="math inline">\(\log | \frac{d}{dy} \log y | = \log | 1/y | = - \log |y|\)</span>.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn4"><p>A programming idiom in BUGS code simulates a local variable by replacing <code>theta</code> in the above example with <code>theta[n]</code>, effectively creating <code>N</code> different variables, <code>theta[1]</code>, …, <code>theta[N]</code>. Of course, this is not a hack if the value of <code>theta[n]</code> is required for all <code>n</code>.<a href="#fnref4" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn5"><p>The adjoint component is always zero during execution for the algorithmic differentiation variables used to implement parameters, transformed parameters, and local variables in the model.<a href="#fnref5" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/expressions.html" class="pagination-link" aria-label="Expressions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Expressions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/blocks.html" class="pagination-link" aria-label="Program Blocks">
+        <span class="nav-page-text">Program Blocks</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/statements.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/syntax.html b/docs/2_39/reference-manual/syntax.html
new file mode 100644
index 000000000..4d89e58a0
--- /dev/null
+++ b/docs/2_39/reference-manual/syntax.html
@@ -0,0 +1,1480 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Language Syntax</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/execution.html" rel="next">
+<link href="../reference-manual/transforms.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Language Syntax">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/syntax.html">Language Syntax</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#language-syntax" id="toc-language-syntax" class="nav-link active" data-scroll-target="#language-syntax">Language Syntax</a>
+  <ul class="collapse">
+  <li><a href="#bnf-grammars" id="toc-bnf-grammars" class="nav-link" data-scroll-target="#bnf-grammars">BNF grammars</a>
+  <ul class="collapse">
+  <li><a href="#syntactic-conventions" id="toc-syntactic-conventions" class="nav-link" data-scroll-target="#syntactic-conventions">Syntactic conventions</a></li>
+  <li><a href="#programs" id="toc-programs" class="nav-link" data-scroll-target="#programs">Programs</a></li>
+  <li><a href="#function-declarations-and-definitions" id="toc-function-declarations-and-definitions" class="nav-link" data-scroll-target="#function-declarations-and-definitions">Function declarations and definitions</a></li>
+  <li><a href="#variable-declarations-and-compound-definitions" id="toc-variable-declarations-and-compound-definitions" class="nav-link" data-scroll-target="#variable-declarations-and-compound-definitions">Variable declarations and compound definitions</a></li>
+  <li><a href="#expressions" id="toc-expressions" class="nav-link" data-scroll-target="#expressions">Expressions</a></li>
+  <li><a href="#statements" id="toc-statements" class="nav-link" data-scroll-target="#statements">Statements</a></li>
+  </ul></li>
+  <li><a href="#tokenizing-rules" id="toc-tokenizing-rules" class="nav-link" data-scroll-target="#tokenizing-rules">Tokenizing rules</a></li>
+  <li><a href="#extra-grammatical-constraints" id="toc-extra-grammatical-constraints" class="nav-link" data-scroll-target="#extra-grammatical-constraints">Extra-grammatical constraints</a>
+  <ul class="collapse">
+  <li><a href="#type-constraints" id="toc-type-constraints" class="nav-link" data-scroll-target="#type-constraints">Type constraints</a></li>
+  <li><a href="#operator-precedence-and-associativity" id="toc-operator-precedence-and-associativity" class="nav-link" data-scroll-target="#operator-precedence-and-associativity">Operator precedence and associativity</a></li>
+  <li><a href="#typing-of-compound-declaration-and-definition" id="toc-typing-of-compound-declaration-and-definition" class="nav-link" data-scroll-target="#typing-of-compound-declaration-and-definition">Typing of compound declaration and definition</a></li>
+  <li><a href="#typing-of-array-expressions" id="toc-typing-of-array-expressions" class="nav-link" data-scroll-target="#typing-of-array-expressions">Typing of array expressions</a></li>
+  <li><a href="#forms-of-numbers" id="toc-forms-of-numbers" class="nav-link" data-scroll-target="#forms-of-numbers">Forms of numbers</a></li>
+  <li><a href="#conditional-arguments" id="toc-conditional-arguments" class="nav-link" data-scroll-target="#conditional-arguments">Conditional arguments</a></li>
+  <li><a href="#for-loop-containers" id="toc-for-loop-containers" class="nav-link" data-scroll-target="#for-loop-containers">For loop containers</a></li>
+  <li><a href="#print-arguments" id="toc-print-arguments" class="nav-link" data-scroll-target="#print-arguments">Print arguments</a></li>
+  <li><a href="#only-break-and-continue-in-loops" id="toc-only-break-and-continue-in-loops" class="nav-link" data-scroll-target="#only-break-and-continue-in-loops">Only break and continue in loops</a></li>
+  <li><a href="#block-specific-restrictions" id="toc-block-specific-restrictions" class="nav-link" data-scroll-target="#block-specific-restrictions">Block-specific restrictions</a></li>
+  <li><a href="#probability-function-naming" id="toc-probability-function-naming" class="nav-link" data-scroll-target="#probability-function-naming">Probability function naming</a></li>
+  <li><a href="#indexes" id="toc-indexes" class="nav-link" data-scroll-target="#indexes">Indexes</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/syntax.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/syntax.html">Language Syntax</a></li></ol></nav></header>
+
+
+
+
+<section id="language-syntax" class="level1">
+<h1>Language Syntax</h1>
+<p>This chapter defines the basic syntax of the Stan modeling language using a Backus-Naur form (BNF) grammar plus extra-grammatical constraints on function typing and operator precedence and associativity.</p>
+<section id="bnf-grammars" class="level2">
+<h2 class="anchored" data-anchor-id="bnf-grammars">BNF grammars</h2>
+<section id="syntactic-conventions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="syntactic-conventions">Syntactic conventions</h3>
+<p>In the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by <code>&lt;</code> and <code>&gt;</code>. A square brackets (<code>[A]</code>) indicates optionality of <code>A</code>. A postfixed Kleene star (<code>A*</code>) indicates zero or more occurrences of <code>A</code>. Parenthesis can be used to group symbols together in productions.</p>
+<p>Finally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library <a href="http://gallium.inria.fr/~fpottier/menhir/manual.html#sec30">Menhir</a>. A rule like <code>&lt;list(x)&gt; ::= x (COMMA x)*</code> declares a generic list rule, which can later be applied to others by the symbol <code>&lt;list(&lt;expression&gt;)&gt;</code>.</p>
+<p>The following representation is constructed directly from the OCaml <a href="https://github.com/stan-dev/stanc3/blob/master/src/frontend/parser.mly">reference parser</a> using a tool called <a href="https://github.com/Lelio-Brun/Obelisk">Obelisk</a>. The raw output is available <a href="https://raw.githubusercontent.com/stan-dev/docs/master/src/reference-manual/grammar.txt">here</a>.</p>
+<!-- This is the direct output of `obelisk -i parser.mly`
+     copied and pasted into reasonable section distinctions.
+     Additionally, items which exist only for error messaging
+     were removed, like allowing FUNCTIONBLOCK as a decl identifier
+-->
+</section>
+<section id="programs" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="programs">Programs</h3>
+<pre><code>&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]
+              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]
+              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF
+
+&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF
+
+&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE
+
+&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE
+
+&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE
+                             &lt;top_vardecl_or_statement&gt;* RBRACE
+
+&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*
+                       RBRACE
+
+&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE
+                                   &lt;top_vardecl_or_statement&gt;* RBRACE
+
+&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE
+
+&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE
+                                 &lt;top_vardecl_or_statement&gt;* RBRACE</code></pre>
+</section>
+<section id="function-declarations-and-definitions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="function-declarations-and-definitions">Function declarations and definitions</h3>
+<pre><code>&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA
+                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;
+
+&lt;return_type&gt; ::= VOID
+                | &lt;unsized_type&gt;
+
+&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;
+
+&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;
+                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;
+                 | &lt;basic_type&gt;
+                 | &lt;unsized_tuple_type&gt;
+
+&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA [&lt;unsized_type&gt;
+                         (COMMA &lt;unsized_type&gt;)*] RPAREN
+
+&lt;basic_type&gt; ::= INT
+               | REAL
+               | COMPLEX
+               | VECTOR
+               | ROWVECTOR
+               | MATRIX
+               | COMPLEXVECTOR
+               | COMPLEXROWVECTOR
+               | COMPLEXMATRIX
+
+&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK
+</code></pre>
+</section>
+<section id="variable-declarations-and-compound-definitions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variable-declarations-and-compound-definitions">Variable declarations and compound definitions</h3>
+<pre><code>&lt;identifier&gt; ::= IDENTIFIER
+               | TRUNCATE
+
+&lt;decl_identifier&gt; ::= &lt;identifier&gt;
+
+&lt;no_assign&gt; ::= UNREACHABLE
+
+&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]
+
+&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;
+                                      &lt;optional_assignment(rhs)&gt;
+
+&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;
+                           &lt;optional_assignment(rhs)&gt; SEMICOLON
+                         | &lt;higher_type(type_rule)&gt;
+                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA
+                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON
+
+&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;
+                           | &lt;tuple_type(type_rule)&gt;
+                           | type_rule
+
+&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule
+                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;
+
+&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA
+                            [&lt;higher_type(type_rule)&gt; (COMMA
+                            &lt;higher_type(type_rule)&gt;)*] RPAREN
+
+&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;
+
+&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;
+
+&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;
+                           | SEMICOLON
+
+&lt;sized_basic_type&gt; ::= INT
+                     | REAL
+                     | COMPLEX
+                     | VECTOR LBRACK &lt;expression&gt; RBRACK
+                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK
+                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK
+                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK
+                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK
+                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;
+                       RBRACK
+
+&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]
+                 | REAL &lt;type_constraint&gt;
+                 | COMPLEX &lt;type_constraint&gt;
+                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK
+                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK
+                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA
+                   &lt;expression&gt; RBRACK
+                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK
+                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;
+                   RBRACK
+                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA
+                   &lt;expression&gt; RBRACK
+                 | ORDERED LBRACK &lt;expression&gt; RBRACK
+                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK
+                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK
+                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK
+                 | SUMTOZEROVEC LBRACK &lt;expression&gt; RBRACK
+                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK
+                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]
+                   RBRACK
+                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK
+                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK
+                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK
+                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA
+                   &lt;expression&gt; RBRACK
+                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;
+                   RBRACK
+
+&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]
+                    | LABRACK &lt;offset_mult&gt; RABRACK
+
+&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN
+            &lt;constr_expression&gt;
+          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN
+            &lt;constr_expression&gt;
+          | LOWER ASSIGN &lt;constr_expression&gt;
+          | UPPER ASSIGN &lt;constr_expression&gt;
+
+&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN
+                  &lt;constr_expression&gt;
+                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN
+                  &lt;constr_expression&gt;
+                | OFFSET ASSIGN &lt;constr_expression&gt;
+                | MULTIPLIER ASSIGN &lt;constr_expression&gt;
+
+&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK
+</code></pre>
+</section>
+<section id="expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="expressions">Expressions</h3>
+<pre><code>&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;
+               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;
+               | &lt;prefixOp&gt; &lt;expression&gt;
+               | &lt;expression&gt; &lt;postfixOp&gt;
+               | &lt;common_expression&gt;
+
+&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;
+                        &lt;constr_expression&gt;
+                      | &lt;prefixOp&gt; &lt;constr_expression&gt;
+                      | &lt;constr_expression&gt; &lt;postfixOp&gt;
+                      | &lt;common_expression&gt;
+
+&lt;common_expression&gt; ::= &lt;identifier&gt;
+                      | INTNUMERAL
+                      | REALNUMERAL
+                      | DOTNUMERAL
+                      | IMAGNUMERAL
+                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE
+                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK
+                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA
+                        &lt;expression&gt;)*] RPAREN
+                      | TARGET LPAREN RPAREN
+                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;
+                        (COMMA &lt;expression&gt;)*] RPAREN
+                      | LPAREN &lt;expression&gt; COMMA [&lt;expression&gt; (COMMA
+                        &lt;expression&gt;)*] RPAREN
+                      | &lt;common_expression&gt; DOTNUMERAL
+                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK
+                      | LPAREN &lt;expression&gt; RPAREN
+
+&lt;prefixOp&gt; ::= BANG
+             | MINUS
+             | PLUS
+
+&lt;postfixOp&gt; ::= TRANSPOSE
+
+&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;
+            | &lt;logicalBinOp&gt;
+
+&lt;arithmeticBinOp&gt; ::= PLUS
+                    | MINUS
+                    | TIMES
+                    | DIVIDE
+                    | IDIVIDE
+                    | MODULO
+                    | LDIVIDE
+                    | ELTTIMES
+                    | ELTDIVIDE
+                    | HAT
+                    | ELTPOW
+
+&lt;logicalBinOp&gt; ::= OR
+                 | AND
+                 | EQUALS
+                 | NEQUALS
+                 | LABRACK
+                 | LEQ
+                 | RABRACK
+                 | GEQ
+
+&lt;indexes&gt; ::= epsilon
+            | COLON
+            | &lt;expression&gt;
+            | &lt;expression&gt; COLON
+            | COLON &lt;expression&gt;
+            | &lt;expression&gt; COLON &lt;expression&gt;
+            | &lt;indexes&gt; COMMA &lt;indexes&gt;
+
+&lt;printables&gt; ::= &lt;expression&gt;
+               | &lt;string_literal&gt;
+               | &lt;printables&gt; COMMA &lt;printables&gt;</code></pre>
+</section>
+<section id="statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="statements">Statements</h3>
+<pre><code>&lt;statement&gt; ::= &lt;atomic_statement&gt;
+              | &lt;nested_statement&gt;
+
+&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;
+                       SEMICOLON
+                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA
+                       &lt;expression&gt;)*] RPAREN SEMICOLON
+                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;
+                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON
+                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON
+                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON
+                     | BREAK SEMICOLON
+                     | CONTINUE SEMICOLON
+                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON
+                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON
+                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON
+                     | RETURN &lt;expression&gt; SEMICOLON
+                     | RETURN SEMICOLON
+                     | SEMICOLON
+
+&lt;assignment_op&gt; ::= ASSIGN
+                  | PLUSASSIGN
+                  | MINUSASSIGN
+                  | TIMESASSIGN
+                  | DIVIDEASSIGN
+                  | ELTTIMESASSIGN
+                  | ELTDIVIDEASSIGN
+
+&lt;string_literal&gt; ::= STRINGLITERAL
+
+&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK
+
+&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;
+                       ELSE &lt;vardecl_or_statement&gt;
+                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;
+                     | WHILE LPAREN &lt;expression&gt; RPAREN
+                       &lt;vardecl_or_statement&gt;
+                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON
+                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;
+                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN
+                       &lt;vardecl_or_statement&gt;
+                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE
+                       &lt;vardecl_or_statement&gt;* RBRACE
+                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE
+
+&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;
+                         | &lt;var_decl&gt;
+
+&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;
+                             | &lt;top_var_decl&gt;</code></pre>
+</section>
+</section>
+<section id="tokenizing-rules" class="level2">
+<h2 class="anchored" data-anchor-id="tokenizing-rules">Tokenizing rules</h2>
+<p>Many of the tokens used in the BNF grammars follow obviously from their names: <code>DATABLOCK</code> is the literal string ‘data’, <code>COMMA</code> is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the <a href="../reference-manual/expressions.html#operator-precedence-table">operator precedence table</a>.</p>
+<p>A few tokens are not so obvious, and are defined here in regular expressions:</p>
+<pre><code>IDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*
+
+STRINGLITERAL = ".*"
+
+INTNUMERAL = [0-9]+ (_ [0-9]+)*
+
+EXPLITERAL = [eE] [+-]? INTNUMERAL
+
+REALNUMERAL = INTNUMERAL \. INTNUMERAL? EXPLITERAL?
+            | \. INTNUMERAL EXPLITERAL
+            | INTNUMERAL EXPLITERAL
+
+IMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i
+
+DOTNUMERAL = \. INTNUMERAL</code></pre>
+</section>
+<section id="extra-grammatical-constraints" class="level2">
+<h2 class="anchored" data-anchor-id="extra-grammatical-constraints">Extra-grammatical constraints</h2>
+<section id="type-constraints" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-constraints">Type constraints</h3>
+<p>A well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.</p>
+</section>
+<section id="operator-precedence-and-associativity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="operator-precedence-and-associativity">Operator precedence and associativity</h3>
+<p>In the Stan grammar provided in this chapter, the expression <code>1 + 2 * 3</code> has two parses. As described in the <a href="../reference-manual/expressions.html#operator-precedence-table">operator precedence table</a>, Stan disambiguates between the meaning <span class="math inline">\(1
++ (2 \times 3)\)</span> and the meaning <span class="math inline">\((1 + 2) \times 3\)</span> based on operator precedences and associativities.</p>
+</section>
+<section id="typing-of-compound-declaration-and-definition" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="typing-of-compound-declaration-and-definition">Typing of compound declaration and definition</h3>
+<p>In a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.</p>
+</section>
+<section id="typing-of-array-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="typing-of-array-expressions">Typing of array expressions</h3>
+<p>The types of expressions used for elements in array expressions (<code>'{' expressions '}'</code>) must all be of the same type or a mixture of scalar (<code>int</code>, <code>real</code> and <code>complex</code>) types (in which case the result is promoted to be of the highest type on the <code>int -&gt; real -&gt; complex</code> hierarchy).</p>
+</section>
+<section id="forms-of-numbers" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="forms-of-numbers">Forms of numbers</h3>
+<p>Integer literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.</p>
+</section>
+<section id="conditional-arguments" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="conditional-arguments">Conditional arguments</h3>
+<p>Both the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.</p>
+</section>
+<section id="for-loop-containers" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="for-loop-containers">For loop containers</h3>
+<p>The for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).</p>
+</section>
+<section id="print-arguments" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="print-arguments">Print arguments</h3>
+<p>The arguments to a print statement cannot be void.</p>
+</section>
+<section id="only-break-and-continue-in-loops" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="only-break-and-continue-in-loops">Only break and continue in loops</h3>
+<p>The <code>break</code> and <code>continue</code> statements may only be used within the body of a for-loop or while-loop.</p>
+</section>
+<section id="block-specific-restrictions" class="level3">
+<h3 class="anchored" data-anchor-id="block-specific-restrictions">Block-specific restrictions</h3>
+<p>Some constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.</p>
+<section id="prng-functions" class="level4">
+<h4 class="anchored" data-anchor-id="prng-functions">PRNG functions</h4>
+<p>Functions ending in <code>_rng</code> may only be called in the <code>transformed data</code> and <code>generated quantities</code> block, and within the bodies of user-defined functions with names ending in <code>_rng</code>.</p>
+</section>
+<section id="unnormalized-distributions" class="level4">
+<h4 class="anchored" data-anchor-id="unnormalized-distributions">Unnormalized distributions</h4>
+<p>Unnormalized distributions (with suffixes <code>_lupmf</code> or <code>_lupdf</code>) may only be called in the <code>model</code> block, user-defined probability functions, or within the bodies of user defined functions which end in <code>_lp</code>.</p>
+</section>
+<section id="incrementing-and-accessing-target" class="level4">
+<h4 class="anchored" data-anchor-id="incrementing-and-accessing-target">Incrementing and accessing target</h4>
+<p><code>target +=</code> statements can only be used inside of the <code>model</code> block or user-defined functions which end in <code>_lp</code>.</p>
+<p>User defined functions which end in <code>_lp</code> and the <code>target()</code> function can only be used in the <code>model</code> block, <code>transformed parameters</code> block, and in the bodies of other user defined functions which end in <code>_lp</code>.</p>
+<p>Sampling statements (using <code>~</code>) can only be used in the <code>model</code> block or in the bodies of user-defined functions which end in <code>_lp</code>.</p>
+<p><code>jacobian +=</code> statements can only be used inside of the <code>transformed parameters</code> block or in functions that end with <code>_jacobian</code>.</p>
+</section>
+</section>
+<section id="probability-function-naming" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="probability-function-naming">Probability function naming</h3>
+<p>A probability function literal must have one of the following suffixes: <code>_lpdf</code>, <code>_lpmf</code>, <code>_lcdf</code>, or <code>_lccdf</code>.</p>
+</section>
+<section id="indexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="indexes">Indexes</h3>
+<p>Standalone expressions used as indexes must denote either an integer (<code>int</code>) or an integer array (<code>array[] int</code>). Expressions participating in range indexes (e.g., <code>a</code> and <code>b</code> in <code>a : b</code>) must denote integers (<code>int</code>).</p>
+<p>A second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type <code>array[ , , ] matrix</code>, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.</p>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/transforms.html" class="pagination-link" aria-label="Constraint Transforms">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Constraint Transforms</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/execution.html" class="pagination-link" aria-label="Program Execution">
+        <span class="nav-page-text">Program Execution</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/syntax.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/transforms.html b/docs/2_39/reference-manual/transforms.html
new file mode 100644
index 000000000..bdf7d1c0e
--- /dev/null
+++ b/docs/2_39/reference-manual/transforms.html
@@ -0,0 +1,2232 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Constraint Transforms</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/syntax.html" rel="next">
+<link href="../reference-manual/user-functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Constraint Transforms">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/transforms.html">Constraint Transforms</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#variable-transforms.chapter" id="toc-variable-transforms.chapter" class="nav-link active" data-scroll-target="#variable-transforms.chapter">Constraint Transforms</a>
+  <ul class="collapse">
+  <li><a href="#limitations-due-to-finite-accuracy-presentation" id="toc-limitations-due-to-finite-accuracy-presentation" class="nav-link" data-scroll-target="#limitations-due-to-finite-accuracy-presentation">Limitations due to finite accuracy presentation</a></li>
+  <li><a href="#change-of-variables.section" id="toc-change-of-variables.section" class="nav-link" data-scroll-target="#change-of-variables.section">Changes of variables</a>
+  <ul class="collapse">
+  <li><a href="#univariate-changes-of-variables" id="toc-univariate-changes-of-variables" class="nav-link" data-scroll-target="#univariate-changes-of-variables">Univariate changes of variables</a></li>
+  <li><a href="#multivariate-changes-of-variables" id="toc-multivariate-changes-of-variables" class="nav-link" data-scroll-target="#multivariate-changes-of-variables">Multivariate changes of variables</a></li>
+  </ul></li>
+  <li><a href="#lower-bound-transform.section" id="toc-lower-bound-transform.section" class="nav-link" data-scroll-target="#lower-bound-transform.section">Lower bounded scalar</a>
+  <ul class="collapse">
+  <li><a href="#lower-bound-transform" id="toc-lower-bound-transform" class="nav-link" data-scroll-target="#lower-bound-transform">Lower bound transform</a></li>
+  <li><a href="#lower-bound-inverse-transform" id="toc-lower-bound-inverse-transform" class="nav-link" data-scroll-target="#lower-bound-inverse-transform">Lower bound inverse transform</a></li>
+  <li><a href="#absolute-derivative-of-the-lower-bound-inverse-transform" id="toc-absolute-derivative-of-the-lower-bound-inverse-transform" class="nav-link" data-scroll-target="#absolute-derivative-of-the-lower-bound-inverse-transform">Absolute derivative of the lower bound inverse transform</a></li>
+  </ul></li>
+  <li><a href="#upper-bounded-scalar" id="toc-upper-bounded-scalar" class="nav-link" data-scroll-target="#upper-bounded-scalar">Upper bounded scalar</a>
+  <ul class="collapse">
+  <li><a href="#upper-bound-transform" id="toc-upper-bound-transform" class="nav-link" data-scroll-target="#upper-bound-transform">Upper bound transform</a></li>
+  <li><a href="#upper-bound-inverse-transform" id="toc-upper-bound-inverse-transform" class="nav-link" data-scroll-target="#upper-bound-inverse-transform">Upper bound inverse transform</a></li>
+  <li><a href="#absolute-derivative-of-the-upper-bound-inverse-transform" id="toc-absolute-derivative-of-the-upper-bound-inverse-transform" class="nav-link" data-scroll-target="#absolute-derivative-of-the-upper-bound-inverse-transform">Absolute derivative of the upper bound inverse transform</a></li>
+  </ul></li>
+  <li><a href="#logit-transform-jacobian.section" id="toc-logit-transform-jacobian.section" class="nav-link" data-scroll-target="#logit-transform-jacobian.section">Lower and upper bounded scalar</a>
+  <ul class="collapse">
+  <li><a href="#log-odds-and-the-logistic-sigmoid" id="toc-log-odds-and-the-logistic-sigmoid" class="nav-link" data-scroll-target="#log-odds-and-the-logistic-sigmoid">Log odds and the logistic sigmoid</a></li>
+  <li><a href="#lower-and-upper-bounds-transform" id="toc-lower-and-upper-bounds-transform" class="nav-link" data-scroll-target="#lower-and-upper-bounds-transform">Lower and upper bounds transform</a></li>
+  <li><a href="#lower-and-upper-bounds-inverse-transform" id="toc-lower-and-upper-bounds-inverse-transform" class="nav-link" data-scroll-target="#lower-and-upper-bounds-inverse-transform">Lower and upper bounds inverse transform</a></li>
+  <li><a href="#absolute-derivative-of-the-lower-and-upper-bounds-inverse-transform" id="toc-absolute-derivative-of-the-lower-and-upper-bounds-inverse-transform" class="nav-link" data-scroll-target="#absolute-derivative-of-the-lower-and-upper-bounds-inverse-transform">Absolute derivative of the lower and upper bounds inverse transform</a></li>
+  </ul></li>
+  <li><a href="#affinely-transformed-scalar" id="toc-affinely-transformed-scalar" class="nav-link" data-scroll-target="#affinely-transformed-scalar">Affinely transformed scalar</a>
+  <ul class="collapse">
+  <li><a href="#affine-transform" id="toc-affine-transform" class="nav-link" data-scroll-target="#affine-transform">Affine transform</a></li>
+  <li><a href="#affine-inverse-transform" id="toc-affine-inverse-transform" class="nav-link" data-scroll-target="#affine-inverse-transform">Affine inverse transform</a></li>
+  <li><a href="#absolute-derivative-of-the-affine-inverse-transform" id="toc-absolute-derivative-of-the-affine-inverse-transform" class="nav-link" data-scroll-target="#absolute-derivative-of-the-affine-inverse-transform">Absolute derivative of the affine inverse transform</a></li>
+  </ul></li>
+  <li><a href="#ordered-vector" id="toc-ordered-vector" class="nav-link" data-scroll-target="#ordered-vector">Ordered vector</a>
+  <ul class="collapse">
+  <li><a href="#ordered-transform" id="toc-ordered-transform" class="nav-link" data-scroll-target="#ordered-transform">Ordered transform</a></li>
+  <li><a href="#ordered-inverse-transform" id="toc-ordered-inverse-transform" class="nav-link" data-scroll-target="#ordered-inverse-transform">Ordered inverse transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-the-ordered-inverse-transform" id="toc-absolute-jacobian-determinant-of-the-ordered-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-the-ordered-inverse-transform">Absolute Jacobian determinant of the ordered inverse transform</a></li>
+  </ul></li>
+  <li><a href="#positive-ordered-vector" id="toc-positive-ordered-vector" class="nav-link" data-scroll-target="#positive-ordered-vector">Positive ordered vector</a></li>
+  <li><a href="#sum-to-zero-transforms" id="toc-sum-to-zero-transforms" class="nav-link" data-scroll-target="#sum-to-zero-transforms">Sum-to-zero transforms</a>
+  <ul class="collapse">
+  <li><a href="#sum-to-zero-vector" id="toc-sum-to-zero-vector" class="nav-link" data-scroll-target="#sum-to-zero-vector">Sum-to-zero vector</a></li>
+  <li><a href="#sum-to-zero-vector-transform" id="toc-sum-to-zero-vector-transform" class="nav-link" data-scroll-target="#sum-to-zero-vector-transform">Sum-to-zero vector transform</a></li>
+  <li><a href="#sum-to-zero-vector-inverse-transform" id="toc-sum-to-zero-vector-inverse-transform" class="nav-link" data-scroll-target="#sum-to-zero-vector-inverse-transform">Sum-to-zero vector inverse transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-the-zero-sum-inverse-transform" id="toc-absolute-jacobian-determinant-of-the-zero-sum-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-the-zero-sum-inverse-transform">Absolute Jacobian determinant of the zero sum inverse transform</a></li>
+  <li><a href="#sum-to-zero-matrix-transform" id="toc-sum-to-zero-matrix-transform" class="nav-link" data-scroll-target="#sum-to-zero-matrix-transform">Sum-to-zero matrix transform</a></li>
+  </ul></li>
+  <li><a href="#simplex-transform.section" id="toc-simplex-transform.section" class="nav-link" data-scroll-target="#simplex-transform.section">Unit simplex</a>
+  <ul class="collapse">
+  <li><a href="#unit-simplex-inverse-transform" id="toc-unit-simplex-inverse-transform" class="nav-link" data-scroll-target="#unit-simplex-inverse-transform">Unit simplex inverse transform</a></li>
+  <li><a href="#unit-simplex-transform" id="toc-unit-simplex-transform" class="nav-link" data-scroll-target="#unit-simplex-transform">Unit simplex transform</a></li>
+  </ul></li>
+  <li><a href="#stochastic-matrix-transform.section" id="toc-stochastic-matrix-transform.section" class="nav-link" data-scroll-target="#stochastic-matrix-transform.section">Stochastic Matrix</a>
+  <ul class="collapse">
+  <li><a href="#definition-of-a-stochastic-matrix" id="toc-definition-of-a-stochastic-matrix" class="nav-link" data-scroll-target="#definition-of-a-stochastic-matrix">Definition of a Stochastic Matrix</a></li>
+  <li><a href="#inverse-transform-for-stochastic-matrix" id="toc-inverse-transform-for-stochastic-matrix" class="nav-link" data-scroll-target="#inverse-transform-for-stochastic-matrix">Inverse Transform for Stochastic Matrix</a></li>
+  <li><a href="#absolute-jacobian-determinant-for-the-inverse-transform" id="toc-absolute-jacobian-determinant-for-the-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-for-the-inverse-transform">Absolute Jacobian Determinant for the Inverse Transform</a></li>
+  <li><a href="#transform-for-stochastic-matrix" id="toc-transform-for-stochastic-matrix" class="nav-link" data-scroll-target="#transform-for-stochastic-matrix">Transform for Stochastic Matrix</a></li>
+  </ul></li>
+  <li><a href="#unit-vector.section" id="toc-unit-vector.section" class="nav-link" data-scroll-target="#unit-vector.section">Unit vector</a>
+  <ul class="collapse">
+  <li><a href="#unit-vector-inverse-transform" id="toc-unit-vector-inverse-transform" class="nav-link" data-scroll-target="#unit-vector-inverse-transform">Unit vector inverse transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-the-unit-vector-inverse-transform" id="toc-absolute-jacobian-determinant-of-the-unit-vector-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-the-unit-vector-inverse-transform">Absolute Jacobian determinant of the unit vector inverse transform</a></li>
+  </ul></li>
+  <li><a href="#correlation-matrix-transform.section" id="toc-correlation-matrix-transform.section" class="nav-link" data-scroll-target="#correlation-matrix-transform.section">Correlation matrices</a>
+  <ul class="collapse">
+  <li><a href="#correlation-matrix-inverse-transform" id="toc-correlation-matrix-inverse-transform" class="nav-link" data-scroll-target="#correlation-matrix-inverse-transform">Correlation matrix inverse transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-the-correlation-matrix-inverse-transform" id="toc-absolute-jacobian-determinant-of-the-correlation-matrix-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-the-correlation-matrix-inverse-transform">Absolute Jacobian determinant of the correlation matrix inverse transform</a></li>
+  <li><a href="#correlation-matrix-transform" id="toc-correlation-matrix-transform" class="nav-link" data-scroll-target="#correlation-matrix-transform">Correlation matrix transform</a></li>
+  </ul></li>
+  <li><a href="#covariance-matrices" id="toc-covariance-matrices" class="nav-link" data-scroll-target="#covariance-matrices">Covariance matrices</a>
+  <ul class="collapse">
+  <li><a href="#covariance-matrix-transform" id="toc-covariance-matrix-transform" class="nav-link" data-scroll-target="#covariance-matrix-transform">Covariance matrix transform</a></li>
+  <li><a href="#covariance-matrix-inverse-transform" id="toc-covariance-matrix-inverse-transform" class="nav-link" data-scroll-target="#covariance-matrix-inverse-transform">Covariance matrix inverse transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-the-covariance-matrix-inverse-transform" id="toc-absolute-jacobian-determinant-of-the-covariance-matrix-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-the-covariance-matrix-inverse-transform">Absolute Jacobian determinant of the covariance matrix inverse transform</a></li>
+  </ul></li>
+  <li><a href="#cholesky-factors-of-covariance-matrices" id="toc-cholesky-factors-of-covariance-matrices" class="nav-link" data-scroll-target="#cholesky-factors-of-covariance-matrices">Cholesky factors of covariance matrices</a>
+  <ul class="collapse">
+  <li><a href="#cholesky-factor-of-covariance-matrix-transform" id="toc-cholesky-factor-of-covariance-matrix-transform" class="nav-link" data-scroll-target="#cholesky-factor-of-covariance-matrix-transform">Cholesky factor of covariance matrix transform</a></li>
+  <li><a href="#cholesky-factor-of-covariance-matrix-inverse-transform" id="toc-cholesky-factor-of-covariance-matrix-inverse-transform" class="nav-link" data-scroll-target="#cholesky-factor-of-covariance-matrix-inverse-transform">Cholesky factor of covariance matrix inverse transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-cholesky-factor-inverse-transform" id="toc-absolute-jacobian-determinant-of-cholesky-factor-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-cholesky-factor-inverse-transform">Absolute Jacobian determinant of Cholesky factor inverse transform</a></li>
+  </ul></li>
+  <li><a href="#cholesky-factors-of-correlation-matrices" id="toc-cholesky-factors-of-correlation-matrices" class="nav-link" data-scroll-target="#cholesky-factors-of-correlation-matrices">Cholesky factors of correlation matrices</a>
+  <ul class="collapse">
+  <li><a href="#cholesky-factor-of-correlation-matrix-inverse-transform" id="toc-cholesky-factor-of-correlation-matrix-inverse-transform" class="nav-link" data-scroll-target="#cholesky-factor-of-correlation-matrix-inverse-transform">Cholesky factor of correlation matrix inverse transform</a></li>
+  <li><a href="#cholesky-factor-of-correlation-matrix-transform" id="toc-cholesky-factor-of-correlation-matrix-transform" class="nav-link" data-scroll-target="#cholesky-factor-of-correlation-matrix-transform">Cholesky factor of correlation matrix transform</a></li>
+  <li><a href="#absolute-jacobian-determinant-of-inverse-transform" id="toc-absolute-jacobian-determinant-of-inverse-transform" class="nav-link" data-scroll-target="#absolute-jacobian-determinant-of-inverse-transform">Absolute Jacobian determinant of inverse transform</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/transforms.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/transforms.html">Constraint Transforms</a></li></ol></nav></header>
+
+
+
+
+<section id="variable-transforms.chapter" class="level1">
+<h1>Constraint Transforms</h1>
+<p>To avoid having to deal with constraints while simulating the Hamiltonian dynamics during sampling, every (multivariate) parameter in a Stan model is transformed to an unconstrained variable behind the scenes by the model compiler. The transform is based on the constraints, if any, in the parameter’s definition. Scalars or the scalar values in vectors, row vectors or matrices may be constrained with lower and/or upper bounds. Vectors may alternatively be constrained to be ordered, positive ordered, or simplexes. Matrices may be constrained to be correlation matrices or covariance matrices. This chapter provides a definition of the transforms used for each type of variable. For examples of how to declare and define these variables in a Stan program, see section <a href="../reference-manual/types.html#variable-declaration.section">Variable declaration</a>. To directly access the functional form of these transformations from inside the Stan language, see <a href="https://mc-stan.org/docs/functions-reference/transform_functions.html">Variable Transformation Functions</a> in the <em>Functions Reference</em>.</p>
+<p>Stan converts models to C++ classes which define probability functions with support on all of <span class="math inline">\(\mathbb{R}^K\)</span>, where <span class="math inline">\(K\)</span> is the number of unconstrained parameters needed to define the constrained parameters defined in the program. The C++ classes also include code to transform the parameters from unconstrained to constrained and apply the appropriate Jacobians.</p>
+<section id="limitations-due-to-finite-accuracy-presentation" class="level2">
+<h2 class="anchored" data-anchor-id="limitations-due-to-finite-accuracy-presentation">Limitations due to finite accuracy presentation</h2>
+<p>In this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.</p>
+<p>Stan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with <code>lower=0</code>. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about <a href="../stan-users-guide/floating-point.html">floating point arithmetic</a> in the <em>Stan User’s Guide</em>.</p>
+<p>These issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the <a href="../cmdstan-guide/command_line_options.html">CmdStan Command-Line Interface Overview</a>.</p>
+</section>
+<section id="change-of-variables.section" class="level2">
+<h2 class="anchored" data-anchor-id="change-of-variables.section">Changes of variables</h2>
+<p>The support of a random variable <span class="math inline">\(X\)</span> with density <span class="math inline">\(p_X(x)\)</span> is that subset of values for which it has non-zero density,</p>
+<p><span class="math display">\[
+\mathrm{supp}(X) = \{ x | p_X(x) &gt; 0 \}.
+\]</span></p>
+<p>If <span class="math inline">\(f\)</span> is a total function defined on the support of <span class="math inline">\(X\)</span>, then <span class="math inline">\(Y =
+f(X)\)</span> is a new random variable. This section shows how to compute the probability density function of <span class="math inline">\(Y\)</span> for well-behaved transforms <span class="math inline">\(f\)</span>. The rest of the chapter details the transforms used by Stan.</p>
+<section id="univariate-changes-of-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="univariate-changes-of-variables">Univariate changes of variables</h3>
+<p>Suppose <span class="math inline">\(X\)</span> is one dimensional and <span class="math inline">\(f: \mathrm{supp}(X) \rightarrow
+\mathbb{R}\)</span> is a one-to-one, monotonic function with a differentiable inverse <span class="math inline">\(f^{-1}\)</span>. Then the density of <span class="math inline">\(Y\)</span> is given by</p>
+<p><span class="math display">\[
+p_Y(y) = p_X(f^{-1}(y))
+         \,
+         \left| \, \frac{d}{dy} f^{-1}(y)\, \right|.
+\]</span></p>
+<p>The absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.</p>
+</section>
+<section id="multivariate-changes-of-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multivariate-changes-of-variables">Multivariate changes of variables</h3>
+<p>The multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.</p>
+<p>Suppose <span class="math inline">\(X\)</span> is a <span class="math inline">\(K\)</span>-dimensional random variable with probability density function <span class="math inline">\(p_X(x)\)</span>. A new random variable <span class="math inline">\(Y = f(X)\)</span> may be defined by transforming <span class="math inline">\(X\)</span> with a suitably well-behaved function <span class="math inline">\(f\)</span>. It suffices for what follows to note that if <span class="math inline">\(f\)</span> is one-to-one and its inverse <span class="math inline">\(f^{-1}\)</span> has a well-defined Jacobian, then the density of <span class="math inline">\(Y\)</span> is</p>
+<p><span class="math display">\[
+p_Y(y) = p_X(f^{-1}(y)) \, \left| \, \det \, J_{f^{-1}}(y) \, \right|,
+\]</span></p>
+<p>where <span class="math inline">\(\det{}\)</span> is the matrix determinant operation and <span class="math inline">\(J_{f^{-1}}(y)\)</span> is the Jacobian matrix of <span class="math inline">\(f^{-1}\)</span> evaluated at <span class="math inline">\(y\)</span>. Taking <span class="math inline">\(x =
+f^{-1}(y)\)</span>, the Jacobian matrix is defined by</p>
+<p><span class="math display">\[
+J_{f^{-1}}(y) =
+\left[
+\begin{array}{ccc}\displaystyle
+\frac{\partial x_1}{\partial y_1}
+&amp; \cdots
+&amp; \displaystyle \frac{\partial x_1}{\partial y_{K}}
+\\
+\vdots &amp; \vdots &amp; \vdots
+\\
+\displaystyle\frac{\partial x_{K}}{\partial y_1}
+&amp; \cdots
+&amp; \displaystyle\frac{\partial x_{K}}{\partial y_{K}}
+\end{array}
+\right].
+\]</span></p>
+<p>If the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,</p>
+<p><span class="math display">\[
+\det \, J_{f^{-1}}(y)
+= \prod_{k=1}^K \frac{\partial x_k}{\partial y_k}.
+\]</span></p>
+<p>Triangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.</p>
+</section>
+</section>
+<section id="lower-bound-transform.section" class="level2">
+<h2 class="anchored" data-anchor-id="lower-bound-transform.section">Lower bounded scalar</h2>
+<p>Stan uses a logarithmic transform for lower and upper bounds.</p>
+<section id="lower-bound-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lower-bound-transform">Lower bound transform</h3>
+<p>If a variable <span class="math inline">\(X\)</span> is declared to have lower bound <span class="math inline">\(a\)</span>, it is transformed to an unbounded variable <span class="math inline">\(Y\)</span>, where</p>
+<p><span class="math display">\[
+Y = \log(X - a).
+\]</span></p>
+</section>
+<section id="lower-bound-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lower-bound-inverse-transform">Lower bound inverse transform</h3>
+<p>The inverse of the lower-bound transform maps an unbounded variable <span class="math inline">\(Y\)</span> to a variable <span class="math inline">\(X\)</span> that is bounded below by <span class="math inline">\(a\)</span> by</p>
+<p><span class="math display">\[
+X = \exp(Y) + a.
+\]</span></p>
+</section>
+<section id="absolute-derivative-of-the-lower-bound-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-derivative-of-the-lower-bound-inverse-transform">Absolute derivative of the lower bound inverse transform</h3>
+<p>The absolute derivative of the inverse transform is</p>
+<p><span class="math display">\[
+\left| \,
+\frac{d}{dy} \left( \exp(y) + a \right)
+\, \right|
+= \exp(y).
+\]</span></p>
+<p>Therefore, given the density <span class="math inline">\(p_X\)</span> of <span class="math inline">\(X\)</span>, the density of <span class="math inline">\(Y\)</span> is</p>
+<p><span class="math display">\[
+p_Y(y)
+= p_X\!\left( \exp(y) + a \right) \cdot \exp(y).
+\]</span></p>
+</section>
+</section>
+<section id="upper-bounded-scalar" class="level2">
+<h2 class="anchored" data-anchor-id="upper-bounded-scalar">Upper bounded scalar</h2>
+<p>Stan uses a negated logarithmic transform for upper bounds.</p>
+<section id="upper-bound-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="upper-bound-transform">Upper bound transform</h3>
+<p>If a variable <span class="math inline">\(X\)</span> is declared to have an upper bound <span class="math inline">\(b\)</span>, it is transformed to the unbounded variable <span class="math inline">\(Y\)</span> by</p>
+<p><span class="math display">\[
+Y = \log(b - X).
+\]</span></p>
+</section>
+<section id="upper-bound-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="upper-bound-inverse-transform">Upper bound inverse transform</h3>
+<p>The inverse of the upper bound transform converts the unbounded variable <span class="math inline">\(Y\)</span> to the variable <span class="math inline">\(X\)</span> bounded above by <span class="math inline">\(b\)</span> through</p>
+<p><span class="math display">\[
+X = b - \exp(Y).
+\]</span></p>
+</section>
+<section id="absolute-derivative-of-the-upper-bound-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-derivative-of-the-upper-bound-inverse-transform">Absolute derivative of the upper bound inverse transform</h3>
+<p>The absolute derivative of the inverse of the upper bound transform is</p>
+<p><span class="math display">\[
+\left| \,
+\frac{d}{dy} \left( b - \exp(y) \right)
+\, \right|
+= \exp(y).
+\]</span></p>
+<p>Therefore, the density of the unconstrained variable <span class="math inline">\(Y\)</span> is defined in terms of the density of the variable <span class="math inline">\(X\)</span> with an upper bound of <span class="math inline">\(b\)</span> by</p>
+<p><span class="math display">\[
+p_Y(y)
+=   p_X \!\left( b - \exp(y) \right) \cdot \exp(y).
+\]</span></p>
+</section>
+</section>
+<section id="logit-transform-jacobian.section" class="level2">
+<h2 class="anchored" data-anchor-id="logit-transform-jacobian.section">Lower and upper bounded scalar</h2>
+<p>For lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.</p>
+<section id="log-odds-and-the-logistic-sigmoid" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="log-odds-and-the-logistic-sigmoid">Log odds and the logistic sigmoid</h3>
+<p>The log-odds function is defined for <span class="math inline">\(u \in (0,1)\)</span> by</p>
+<p><span class="math display">\[
+\mathrm{logit}(u) = \log \frac{u}{1 - u}.
+\]</span></p>
+<p>The inverse of the log odds function is the logistic sigmoid, defined for <span class="math inline">\(v \in (-\infty,\infty)\)</span> by</p>
+<p><span class="math display">\[
+\mathrm{logit}^{-1}(v) = \frac{1}{1 + \exp(-v)}.
+\]</span></p>
+<p>The derivative of the logistic sigmoid is</p>
+<p><span class="math display">\[
+\frac{d}{dy} \mathrm{logit}^{-1}(y)
+= \mathrm{logit}^{-1}(y) \cdot \left( 1 - \mathrm{logit}^{-1}(y) \right).
+\]</span></p>
+</section>
+<section id="lower-and-upper-bounds-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lower-and-upper-bounds-transform">Lower and upper bounds transform</h3>
+<p>For variables constrained to be in the open interval <span class="math inline">\((a, b)\)</span>, Stan uses a scaled and translated log-odds transform. If variable <span class="math inline">\(X\)</span> is declared to have lower bound <span class="math inline">\(a\)</span> and upper bound <span class="math inline">\(b\)</span>, then it is transformed to a new variable <span class="math inline">\(Y\)</span>, where</p>
+<p><span class="math display">\[
+Y = \mathrm{logit} \left( \frac{X - a}{b - a} \right).
+\]</span></p>
+</section>
+<section id="lower-and-upper-bounds-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lower-and-upper-bounds-inverse-transform">Lower and upper bounds inverse transform</h3>
+<p>The inverse of this transform is</p>
+<p><span class="math display">\[
+X = a + (b - a) \cdot \mathrm{logit}^{-1}(Y).
+\]</span></p>
+</section>
+<section id="absolute-derivative-of-the-lower-and-upper-bounds-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-derivative-of-the-lower-and-upper-bounds-inverse-transform">Absolute derivative of the lower and upper bounds inverse transform</h3>
+<p>The absolute derivative of the inverse transform is given by</p>
+<p><span class="math display">\[
+\left|
+  \frac{d}{dy}
+    \left(
+      a + (b - a) \cdot \mathrm{logit}^{-1}(y)
+    \right)
+  \right|
+= (b - a)
+    \cdot \mathrm{logit}^{-1}(y)
+    \cdot \left( 1 - \mathrm{logit}^{-1}(y) \right).
+\]</span></p>
+<p>Therefore, the density of the transformed variable <span class="math inline">\(Y\)</span> is</p>
+<p><span class="math display">\[
+p_Y(y)
+=
+p_X \! \left( a + (b - a) \cdot \mathrm{logit}^{-1}(y) \right)
+    \cdot (b - a)
+    \cdot \mathrm{logit}^{-1}(y)
+    \cdot \left( 1 - \mathrm{logit}^{-1}(y) \right).
+\]</span></p>
+<p>Despite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, <span class="math inline">\(\mathrm{logit}^{-1}(y)\)</span> only needs to be evaluated once, so there is only one call to <span class="math inline">\(\exp(-y)\)</span>.</p>
+</section>
+</section>
+<section id="affinely-transformed-scalar" class="level2">
+<h2 class="anchored" data-anchor-id="affinely-transformed-scalar">Affinely transformed scalar</h2>
+<p>Stan uses an affine transform to be able to specify parameters with a given offset and multiplier.</p>
+<section id="affine-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="affine-transform">Affine transform</h3>
+<p>For variables with expected offset <span class="math inline">\(\mu\)</span> and/or (positive) multiplier <span class="math inline">\(\sigma\)</span>, Stan uses an affine transform. Such a variable <span class="math inline">\(X\)</span> is transformed to a new variable <span class="math inline">\(Y\)</span>, where</p>
+<p><span class="math display">\[
+Y = \frac{X - \mu}{\sigma}.
+\]</span></p>
+<p>The default value for the offset <span class="math inline">\(\mu\)</span> is <span class="math inline">\(0\)</span> and for the multiplier <span class="math inline">\(\sigma\)</span> is <span class="math inline">\(1\)</span> in case not both are specified.</p>
+</section>
+<section id="affine-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="affine-inverse-transform">Affine inverse transform</h3>
+<p>The inverse of this transform is</p>
+<p><span class="math display">\[
+X =  \mu + \sigma \cdot Y.
+\]</span></p>
+</section>
+<section id="absolute-derivative-of-the-affine-inverse-transform" class="level3">
+<h3 class="anchored" data-anchor-id="absolute-derivative-of-the-affine-inverse-transform">Absolute derivative of the affine inverse transform</h3>
+<p>The absolute derivative of the affine inverse transform is</p>
+<p><span class="math display">\[
+\left|
+  \frac{d}{dy}
+    \left(
+      \mu + \sigma \cdot y
+    \right)
+  \right|
+= \sigma.
+\]</span></p>
+<p>Therefore, the density of the transformed variable <span class="math inline">\(Y\)</span> is</p>
+<p><span class="math display">\[
+p_Y(y)
+=
+p_X \! \left( \mu + \sigma \cdot y \right)
+    \cdot \sigma.
+\]</span></p>
+<p>For an example of how to code this in Stan, see section <a href="../reference-manual/types.html#affine-transform.section">Affinely Transformed Real</a>.</p>
+</section>
+</section>
+<section id="ordered-vector" class="level2">
+<h2 class="anchored" data-anchor-id="ordered-vector">Ordered vector</h2>
+<p>For some modeling tasks, a vector-valued random variable <span class="math inline">\(X\)</span> is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.</p>
+<p>In constraint terms, an ordered <span class="math inline">\(K\)</span>-vector <span class="math inline">\(x \in \mathbb{R}^K\)</span> satisfies</p>
+<p><span class="math display">\[
+x_k &lt; x_{k+1}
+\]</span></p>
+<p>for <span class="math inline">\(k \in \{ 1, \ldots, K-1 \}\)</span>.</p>
+<section id="ordered-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ordered-transform">Ordered transform</h3>
+<p>Stan’s transform follows the constraint directly. It maps an increasing vector <span class="math inline">\(x \in \mathbb{R}^{K}\)</span> to an unconstrained vector <span class="math inline">\(y \in
+\mathbb{R}^K\)</span> by setting</p>
+<p><span class="math display">\[
+y_k
+=
+\left\{
+\begin{array}{ll}
+x_1 &amp; \mbox{if } k = 1, \mbox{ and}
+\\
+\log \left( x_{k} - x_{k-1} \right) &amp; \mbox{if } 1 &lt; k \leq K.
+\end{array}
+\right.
+\]</span></p>
+</section>
+<section id="ordered-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ordered-inverse-transform">Ordered inverse transform</h3>
+<p>The inverse transform for an unconstrained <span class="math inline">\(y \in \mathbb{R}^K\)</span> to an ordered sequence <span class="math inline">\(x \in \mathbb{R}^K\)</span> is defined by the recursion</p>
+<p><span class="math display">\[
+x_k
+=
+\left\{
+\begin{array}{ll}
+y_1 &amp; \mbox{if } k = 1, \mbox{ and}
+\\
+x_{k-1} + \exp(y_k) &amp; \mbox{if } 1 &lt; k \leq K.
+\end{array}
+\right.
+\]</span></p>
+<p><span class="math inline">\(x_k\)</span> can also be expressed iteratively as</p>
+<p><span class="math display">\[
+x_k = y_1 + \sum_{k'=2}^k \exp(y_{k'}).
+\]</span></p>
+</section>
+<section id="absolute-jacobian-determinant-of-the-ordered-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-the-ordered-inverse-transform">Absolute Jacobian determinant of the ordered inverse transform</h3>
+<p>The Jacobian of the inverse transform <span class="math inline">\(f^{-1}\)</span> is lower triangular, with diagonal elements for <span class="math inline">\(1 \leq k \leq K\)</span> of</p>
+<p><span class="math display">\[
+J_{k,k} =
+\left\{
+\begin{array}{ll}
+1 &amp; \mbox{if } k = 1, \mbox{ and}
+\\
+\exp(y_k) &amp; \mbox{if } 1 &lt; k \leq K.
+\end{array}
+\right.
+\]</span></p>
+<p>Because <span class="math inline">\(J\)</span> is triangular, the absolute Jacobian determinant is</p>
+<p><span class="math display">\[
+\left| \, \det \, J \, \right|
+\ = \
+\left| \, \prod_{k=1}^K J_{k,k} \, \right|
+\ = \
+\prod_{k=2}^K \exp(y_k).
+\]</span></p>
+<p>Putting this all together, if <span class="math inline">\(p_X\)</span> is the density of <span class="math inline">\(X\)</span>, then the transformed variable <span class="math inline">\(Y\)</span> has density <span class="math inline">\(p_Y\)</span> given by</p>
+<p><span class="math display">\[
+p_Y(y)
+= p_X(f^{-1}(y))
+\
+\prod_{k=2}^K \exp(y_k).
+\]</span></p>
+</section>
+</section>
+<section id="positive-ordered-vector" class="level2">
+<h2 class="anchored" data-anchor-id="positive-ordered-vector">Positive ordered vector</h2>
+<p>Placeholder.</p>
+<p>The positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.</p>
+</section>
+<section id="sum-to-zero-transforms" class="level2">
+<h2 class="anchored" data-anchor-id="sum-to-zero-transforms">Sum-to-zero transforms</h2>
+<p>Stan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length <span class="math inline">\(N\)</span> with real values and the sum of the vector equals zero. The sum-to-zero matrix is an <span class="math inline">\(N \times M\)</span> matrix where both the rows and columns sum-to-zero.</p>
+<p>Stan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix <span class="math inline">\(H\)</span> such that <span class="math inline">\(H\in\mathbb R^{N\times (N-1)}\)</span> and <span class="math inline">\(H^{\mathsf T}H=I_{N-1},\;H^{\mathsf T}\mathbf 1=0\)</span>. The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,<span class="citation" data-cites="seyboldt:2024">(<a href="#ref-seyboldt:2024" role="doc-biblioref">Seyboldt 2024</a>)</span>. Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.</p>
+<p>In many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:</p>
+<p><span class="math display">\[
+  \sigma^2 \begin{pmatrix}
+  1-\tfrac{1}{N} &amp; -\tfrac{1}{N} &amp; \cdots&amp;-\tfrac{1}{N} \\
+  -\tfrac{1}{N} &amp; 1-\tfrac{1}{N} &amp; \cdots &amp; -\tfrac{1}{N} \\
+  \vdots &amp; \vdots &amp; \ddots &amp; \vdots \\
+  -\tfrac{1}{N} &amp; -\tfrac{1}{N} &amp; \cdots &amp;1-\tfrac{1}{N}
+  \end{pmatrix}.
+\]</span></p>
+<p>The marginal standard deviation no longer corresponds to <span class="math inline">\(\sigma\)</span> but is <span class="math inline">\(\sigma \sqrt{1-\tfrac{1}{N}}\)</span>. The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of <span class="math inline">\(\sqrt{1-\tfrac{1}{N}}\)</span> to adjust the variance to the intended <span class="math inline">\(\sigma\)</span>:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">sum_to_zero_vector</span>[N] x;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>x ~ normal(<span class="dv">0</span>, sqrt(N %/% (N - <span class="dv">1</span>)));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When <span class="math inline">\(\sigma\)</span> is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let <span class="math inline">\(y\)</span> be the unconstrained <span class="math inline">\(N - 1\)</span> vector and <span class="math inline">\(y\)</span> is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of <span class="math inline">\(\sigma\)</span> is given by <span class="math display">\[
+x = \underbrace{\sigma\sqrt{\frac{N}{N-1}}\;H}_{\text{ size of } N \times (N - 1)}\,y.
+\]</span> This is the classic non-centered model. The crucial detail from above is that the operation of multiplying <span class="math inline">\(\sigma\)</span> to <span class="math inline">\(y\)</span> is on <span class="math inline">\(N - 1\)</span> dimensions, not <span class="math inline">\(N\)</span> dimensions. When writing the centered model using <span class="math display">\[
+x \sim \mathcal N \bigg(0,\, \sigma\sqrt{\frac{N}{N-1}} \bigg),
+\]</span> we are incrementing the log density by an additional <span class="math inline">\(-\log(\sigma)\)</span> and so must increment the log density by</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log(sigma * sqrt(N * inv(N - <span class="dv">1</span>)));</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="co">// or</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="co">// because `sqrt(N * inv(N - 1))` is constant</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log(sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>which correctly adjusts for the <span class="math inline">\(N - 1\)</span> free variables.</p>
+<p>More details of the Helmert matrix are in <span class="citation" data-cites="lancaster:1965">(<a href="#ref-lancaster:1965" role="doc-biblioref">Lancaster 1965</a>)</span>, for the basic definitions of the isometric log ratio transform see <span class="citation" data-cites="egozcue+etal:2003">(<a href="#ref-egozcue+etal:2003" role="doc-biblioref">Egozcue et al. 2003</a>)</span> and Chapter 3 of <span class="citation" data-cites="filzmoser+etal:2018">(<a href="#ref-filzmoser+etal:2018" role="doc-biblioref">Filzmoser, Hron, and Templ 2018</a>)</span> for the pivot coordinate version used here.</p>
+<section id="sum-to-zero-vector" class="level3">
+<h3 class="anchored" data-anchor-id="sum-to-zero-vector">Sum-to-zero vector</h3>
+<p>Vectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).</p>
+<p>A sum-to-zero <span class="math inline">\(K\)</span>-vector <span class="math inline">\(x \in \mathbb{R}^K\)</span> satisfies the constraint <span class="math display">\[
+\sum_{k=1}^K x_k = 0.
+\]</span></p>
+</section>
+<section id="sum-to-zero-vector-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="sum-to-zero-vector-transform">Sum-to-zero vector transform</h3>
+<p>The transform is defined iteratively. Given an <span class="math inline">\(x \in \mathbb{R}^{N + 1}\)</span> that sums to zero (i.e., <span class="math inline">\(\sum_{n=1}^{N+1} x_n = 0\)</span>), the transform proceeds as follows to produce an unconstrained <span class="math inline">\(y \in \mathbb{R}^N\)</span>. This is mathematically equivalent to pre-multiplying the unconstrained <span class="math inline">\(y\)</span> by an orthogonal standard basis, e.g.&nbsp;constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.</p>
+<p>The transform is initialized by setting <span class="math display">\[
+S_N = 0
+\]</span> and <span class="math display">\[
+y_N = -x_{N + 1} \sqrt{1 + \frac{1}{N}}.
+\]</span> The for each <span class="math inline">\(n\)</span> from <span class="math inline">\(N - 1\)</span> down to <span class="math inline">\(1\)</span>, let <span class="math display">\[
+w_{n + 1} = \frac{y_{n + 1}}{\sqrt{(n + 1)(n + 2)}},
+\]</span> <span class="math display">\[
+S_n = S_{n + 1} + w_{n + 1},
+\]</span> and <span class="math display">\[
+y_n = (S_n - x_{n + 1}) \frac{\sqrt{n (n + 1)}}{n}.
+\]</span></p>
+<p>This transform is expressed in Stan code as:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> manual_sum_to_zero_jacobian(<span class="dt">vector</span> x) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N = size(x) - <span class="dv">1</span>;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[N] y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>    y[N] = -x[N+<span class="dv">1</span>] * sqrt(<span class="dv">1</span> + <span class="fl">1.</span> / N);</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> sum_w = <span class="dv">0</span>;</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:(N<span class="dv">-1</span>)) {</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> i = N - n;</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> i_p_1 = i + <span class="dv">1</span>;</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>      <span class="dt">real</span> w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + <span class="dv">1</span>));</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>      sum_w += w;</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> y;</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="sum-to-zero-vector-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="sum-to-zero-vector-inverse-transform">Sum-to-zero vector inverse transform</h3>
+<p>The inverse transform follows the isometric logratio tranform. It maps an unconstrained vector <span class="math inline">\(y \in \mathbb{R}^N\)</span> to a sum-to-zero vector <span class="math inline">\(x \in \mathbb{R}^{N + 1}\)</span> such that <span class="math display">\[
+\sum_{n=1}^{N + 1} x_n = 0.
+\]</span> The values are defined inductively, starting with <span class="math display">\[
+x_1 = \sum_{n=1}^N \frac{y_n}{\sqrt{n (n + 1)}}
+\]</span> and then setting <span class="math display">\[
+x_{n + 1} = \sum_{i = n + 1}^N \frac{y_i}{\sqrt{i (i + 1)}}
+- n \cdot \frac{y_n}{\sqrt{n (n + 1)}}.
+\]</span> for <span class="math inline">\(n \in 1{:}N\)</span>.</p>
+<p>The definition is such that <span class="math display">\[
+\sum_{n = 1}^{N + 1} x_n = 0
+\]</span> by construction, because each of the terms added to <span class="math inline">\(x_{n}\)</span> is then subtracted from <span class="math inline">\(x_{n + 1}\)</span> the number of times it shows up in earlier terms.</p>
+</section>
+<section id="absolute-jacobian-determinant-of-the-zero-sum-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-the-zero-sum-inverse-transform">Absolute Jacobian determinant of the zero sum inverse transform</h3>
+<p>The inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.</p>
+<p>The sum-to-zero inverse transform is expressed within Stan as:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> manual_sum_to_zero_inverse_jacobian(<span class="dt">vector</span> y) {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N = num_elements(y);</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[N + <span class="dv">1</span>] x = zeros_vector(N + <span class="dv">1</span>);</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> sum_w = <span class="dv">0</span>;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> i = N - n + <span class="dv">1</span>;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">real</span> w = y[i] * inv_sqrt(i * (i + <span class="dv">1</span>));</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>      sum_w += w;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>      x[i] += sum_w;</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>      x[i + <span class="dv">1</span>] -= w * i;</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> x;</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Note that there is no <code>target +=</code> increment because the Jacobian is zero.</p>
+</section>
+<section id="sum-to-zero-matrix-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="sum-to-zero-matrix-transform">Sum-to-zero matrix transform</h3>
+<p>The matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any <span class="math inline">\(N\)</span>-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.</p>
+<p>Let the unconstrained matrix be <span class="math display">\[
+\mathcal Y \in \mathbb R^{n_1 \times n_2}
+\]</span> and the zero sum vector transform as <span class="math display">\[
+\mathbf z = \mathcal C_n(\mathbf y)\;=\;
+\begin{bmatrix}H_n\\[2pt]-\mathbf 1_{1\times d}\end{bmatrix}\mathbf y
+\;\in\mathbb R^{n+1},
+\]</span> where <span class="math inline">\(H_n\in\mathbb R^{n \times n}\)</span> is the orthogonal Helmert matrix and satisfies <span class="math inline">\(\mathbf 1_{1\times d}A_d^{\!\top}=0\)</span>.</p>
+<p>Applying <span class="math inline">\(C_n\)</span> to each slice results in</p>
+<p><span class="math display">\[
+\mathcal Z \;=\;
+\mathcal X
+\times_1 \bigl[\mathcal C_{n_1}\bigr]
+\times_2 \bigl[\mathcal C_{n_2}\bigr]
+\]</span> where <span class="math display">\[
+\mathcal Z \in
+\mathbb R^{(n_1+1)\times\cdots\times(n_2+1)}.
+\]</span></p>
+<p>Because each <span class="math inline">\(\mathbb R^{d_1\times\cdots\times d_N}\)</span> is invertible on the <span class="math inline">\(\mathbf 1^\perp\)</span> subspace the composite map applied to <span class="math inline">\(\mathcal Z\)</span> is a linear bijection between <span class="math inline">\(\mathbb R^{n_1\times n_2}\)</span> and the codomain <span class="math inline">\(\mathbb R^{(n_1+1)\times\cdots\times(n_2+1)}\)</span>.</p>
+</section>
+</section>
+<section id="simplex-transform.section" class="level2">
+<h2 class="anchored" data-anchor-id="simplex-transform.section">Unit simplex</h2>
+<p>Variables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).</p>
+<p>The unit <span class="math inline">\(K\)</span>-simplex is the set of points <span class="math inline">\(x \in \mathbb{R}^K\)</span> such that for <span class="math inline">\(1 \leq k \leq K\)</span>,</p>
+<p><span class="math display">\[
+x_k &gt; 0,
+\]</span></p>
+<p>and</p>
+<p><span class="math display">\[
+\sum_{k=1}^K x_k = 1.
+\]</span></p>
+<p>An alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values <span class="math inline">\((0,1)\)</span>, and <span class="math inline">\((1,0)\)</span> and the unit 2-simplex is the line connecting these two points; values such as <span class="math inline">\((0.3,0.7)\)</span> and <span class="math inline">\((0.99,0.01)\)</span> lie on the line. In 3-dimensions, the basis is <span class="math inline">\((0,0,1)\)</span>, <span class="math inline">\((0,1,0)\)</span> and <span class="math inline">\((1,0,0)\)</span> and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include <span class="math inline">\((0.5,0.5,0)\)</span>, <span class="math inline">\((0.2,0.7,0.1)\)</span> and all other triplets of non-negative values summing to 1.</p>
+<p>As these examples illustrate, the simplex always picks out a subspace of <span class="math inline">\(K-1\)</span> dimensions from <span class="math inline">\(\mathbb{R}^K\)</span>. Therefore a point <span class="math inline">\(x\)</span> in the <span class="math inline">\(K\)</span>-simplex is fully determined by its first <span class="math inline">\(K-1\)</span> elements <span class="math inline">\(x_1, x_2,
+\ldots, x_{K-1}\)</span>, with</p>
+<p><span class="math display">\[
+x_K = 1 - \sum_{k=1}^{K-1} x_k.
+\]</span></p>
+<section id="unit-simplex-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unit-simplex-inverse-transform">Unit simplex inverse transform</h3>
+<p>The length-<span class="math inline">\(K\)</span> unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length <span class="math inline">\(K\)</span>.</p>
+<p>Let <span class="math inline">\(y\)</span> represent the unconstrained <span class="math inline">\(K - 1\)</span> values in <span class="math inline">\((-\infty, \infty)\)</span>. The intermediate sum-to-zero vector <span class="math inline">\(z = \text{sum\_to\_zero\_transform}(y)\)</span> is length <span class="math inline">\(K\)</span>. The unit simplex is then given by <span class="math display">\[
+x_i = \text{softmax}(z) = \frac{\exp(z_i)}{\sum_{i = 1}^K \exp(z_i)}
+\]</span></p>
+<p>The sum-to-zero vector transform is described in further detail at the <a href="#sum-to-zero-vector-transform">sum-to-zero vector section of the <em>Reference Manual</em></a>.</p>
+<div class="callout callout-style-default callout-note callout-titled">
+<div class="callout-header d-flex align-content-center">
+<div class="callout-icon-container">
+<i class="callout-icon"></i>
+</div>
+<div class="callout-title-container flex-fill">
+Note
+</div>
+</div>
+<div class="callout-body-container callout-body">
+<p>All versions of Stan pre-2.37 used the stick-breaking transform. This is documented at <a href="https://mc-stan.org/docs/2_36/reference-manual/transforms.html#simplex-transform.section">Stan 2.36 <em>Reference Manual: Simplex Transform</em></a>.</p>
+</div>
+</div>
+<section id="absolute-jacobian-determinant-of-the-unit-simplex-inverse-transform" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-the-unit-simplex-inverse-transform">Absolute Jacobian determinant of the unit-simplex inverse transform</h4>
+<p>The Jacobian <span class="math inline">\(J\)</span> of the inverse unit-simplex transform is found by restricting <span class="math inline">\(J\)</span> to the subspace spanned by the sum-to-zero vector <span class="math inline">\(z\)</span>. The Jacobian is given as the <span class="math inline">\((K - 1) \times (K - 1)\)</span> matrix <span class="math inline">\(J\)</span> where</p>
+<p><span class="math display">\[
+J_{ij} = \frac{\partial x_i}{\partial z_j} =
+\frac{\partial}{\partial z_i} \left( \frac{\exp(z_i)}{{\sum_{i = 1}^K \exp(z_i)}} \right)
+\]</span> and <span class="math inline">\(i,j \in 1, \ldots, K - 1\)</span>.</p>
+<p>The diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification</p>
+<p><span class="math display">\[
+J_{ij} =
+\begin{cases}
+x_i (1 - x_i), &amp; \text{if } i = j, \\
+-x_i x_j, &amp; \text{if } i \neq j.
+\end{cases}
+\]</span></p>
+<p>In matrix form this can be expressed as</p>
+<p><span class="math display">\[
+J = \text{diag}(x) - x x^\top
+\]</span></p>
+<p>The determinant of this matrix can be found using the Matrix Determinant Lemma:</p>
+<p><span class="math display">\[
+\det\bigl(A + u v^{\top}\bigr)
+=
+\det(A)\,\bigl(1 + v^{\top}A^{-1}u\bigr).
+\]</span></p>
+<p>Here,</p>
+<p><span class="math display">\[
+A \;=\; \operatorname{diag}(x_{1},\ldots, x_{K-1}),
+\quad
+u \;=\; -\bigl(x_1,\ldots, x_{K-1}\bigr)^{\!\top},
+\quad
+v \;=\; \bigl(x_{1}, \ldots, x_{K-1}\bigr)^{\!\top}.
+\]</span> Therefore,</p>
+<p><span class="math display">\[
+\begin{aligned}
+\det(J)
+&amp;=
+\bigg(\prod_{i=1}^{K-1} x_i \bigg)
+\bigg(1 + (x_{1},\ldots, x_{K-1})\,\mathrm{diag}\bigl(x_{1}^{-1},\ldots,x_{K-1}^{-1}\bigr)\,
+\big(-x_{1},\ldots,-x_{K-1}\big)^{\top}
+\bigg) \\
+&amp;=
+\bigg(\prod_{i=1}^{K-1} x_{i}\bigg)
+\bigg(1 - \sum_{i=1}^{K-1} x_{i}\bigg)
+=
+\bigg(\prod_{i=1}^{K-1} x_{i}\bigg) x_{K} \\
+&amp;=
+\prod_{i=1}^{K} x_{i}.
+\end{aligned}
+\]</span></p>
+</section>
+</section>
+<section id="unit-simplex-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unit-simplex-transform">Unit simplex transform</h3>
+<p>The transform <span class="math inline">\(Y = f(X)\)</span> can be derived by reversing the stages of the inverse transform,</p>
+<p><span class="math display">\[
+y_k
+= H^\top \bigg(\log(x_k)
+- \frac{1}{K}\sum_{i=1}^K\log(x_i) \bigg)
+.
+\]</span></p>
+<p>The matrix <span class="math inline">\(H\)</span> is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.</p>
+</section>
+</section>
+<section id="stochastic-matrix-transform.section" class="level2">
+<h2 class="anchored" data-anchor-id="stochastic-matrix-transform.section">Stochastic Matrix</h2>
+<p>The <code>column_stochastic_matrix[N, M]</code> and <code>row_stochastic_matrix[M, N]</code> type in Stan represents an <span class="math inline">\(N \times M\)</span> matrix where each column (row) is a unit simplex of dimension <span class="math inline">\(N\)</span>. In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.</p>
+<section id="definition-of-a-stochastic-matrix" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="definition-of-a-stochastic-matrix">Definition of a Stochastic Matrix</h3>
+<p>A column stochastic matrix <span class="math inline">\(X \in \mathbb{R}^{N \times M}\)</span> is defined such that each column is a simplex. For column <span class="math inline">\(m\)</span> (where <span class="math inline">\(1 \leq m \leq M\)</span>):</p>
+<p><span class="math display">\[
+X_{n, m} \geq 0 \quad \text{for } 1 \leq n \leq N,
+\]</span></p>
+<p>and</p>
+<p><span class="math display">\[
+\sum_{n=1}^N X_{n, m} = 1.
+\]</span></p>
+<p>A row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e.&nbsp;the rows of the matrix are simplexes)</p>
+<p><span class="math display">\[
+X_{n, m} \geq 0 \quad \text{for } 1 \leq n \leq N,
+\]</span></p>
+<p>and</p>
+<p><span class="math display">\[
+\sum_{m=1}^N X_{n, m} = 1.
+\]</span></p>
+<p>This definition ensures that each column (row) of the matrix <span class="math inline">\(X\)</span> lies on the <span class="math inline">\(N-1\)</span> dimensional unit simplex, similar to the <code>simplex[N]</code> type, but extended across multiple columns(rows).</p>
+</section>
+<section id="inverse-transform-for-stochastic-matrix" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="inverse-transform-for-stochastic-matrix">Inverse Transform for Stochastic Matrix</h3>
+<p>For the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).</p>
+</section>
+<section id="absolute-jacobian-determinant-for-the-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-for-the-inverse-transform">Absolute Jacobian Determinant for the Inverse Transform</h3>
+<p>The Jacobian determinant of the inverse transform for each column <span class="math inline">\(m\)</span> in the matrix is given by the product of the diagonal entries <span class="math inline">\(J_{n, m}\)</span> of the lower-triangular Jacobian matrix. This determinant is calculated as:</p>
+<p><span class="math display">\[
+\left| \det J_m \right| = \prod_{n=1}^{N-1} \left( z_{n, m} (1 - z_{n, m}) \left( 1 - \sum_{n'=1}^{n-1} X_{n', m} \right) \right).
+\]</span></p>
+<p>Thus, the overall Jacobian determinant for the entire <code>column_stochastic_matrix</code> and <code>row_stochastic_matrix</code> is the product of the determinants for each column (row):</p>
+<p><span class="math display">\[
+\left| \det J \right| = \prod_{m=1}^{M} \left| \det J_m \right|.
+\]</span></p>
+</section>
+<section id="transform-for-stochastic-matrix" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="transform-for-stochastic-matrix">Transform for Stochastic Matrix</h3>
+<p>For the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).</p>
+</section>
+</section>
+<section id="unit-vector.section" class="level2">
+<h2 class="anchored" data-anchor-id="unit-vector.section">Unit vector</h2>
+<p>An <span class="math inline">\(n\)</span>-dimensional vector <span class="math inline">\(x \in \mathbb{R}^n\)</span> is said to be a unit vector if it has unit Euclidean length, so that</p>
+<p><span class="math display">\[
+\Vert x \Vert
+\ = \ \sqrt{x^{\top}\,x}
+\ = \ \sqrt{x_1^2 + x_2^2 + \cdots + x_n^2}
+\ = \ 1\ .
+\]</span></p>
+<section id="unit-vector-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unit-vector-inverse-transform">Unit vector inverse transform</h3>
+<p>Stan divides an unconstrained vector <span class="math inline">\(y \in \mathbb{R}^{n}\)</span> by its norm, <span class="math inline">\(\Vert y \Vert = \sqrt{y^\top y}\)</span>, to obtain a unit vector <span class="math inline">\(x\)</span>,</p>
+<p><span class="math display">\[
+x = \frac{y}{\Vert y \Vert}.
+\]</span></p>
+<p>To generate a unit vector, Stan generates points at random in <span class="math inline">\(\mathbb{R}^n\)</span> with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. <span class="citation" data-cites="Muller:1959">Muller (<a href="#ref-Muller:1959" role="doc-biblioref">1959</a>)</span> showed this generates points uniformly at random on <span class="math inline">\(S^{n-1}\)</span>. That is, if we draw <span class="math inline">\(y_n \sim \mathsf{Normal}(0, 1)\)</span> for <span class="math inline">\(n \in 1{:}n\)</span>, then <span class="math inline">\(x = \frac{y}{\Vert y \Vert}\)</span> has a uniform distribution over <span class="math inline">\(S^{n-1}\)</span>. This allows us to use an <span class="math inline">\(n\)</span>-dimensional basis for <span class="math inline">\(S^{n-1}\)</span> that preserves local neighborhoods in that points that are close to each other in <span class="math inline">\(\mathbb{R}^n\)</span> map to points near each other in <span class="math inline">\(S^{n-1}\)</span>. The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in <span class="math inline">\(\mathbb{R}^n\)</span> that map to identical points in <span class="math inline">\(S^{n-1}\)</span>.</p>
+<section id="warning-undefined-at-zero" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="warning-undefined-at-zero">Warning: undefined at zero!</h4>
+<p>The above mapping from <span class="math inline">\(\mathbb{R}^n\)</span> to <span class="math inline">\(S^n\)</span> is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.</p>
+</section>
+</section>
+<section id="absolute-jacobian-determinant-of-the-unit-vector-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-the-unit-vector-inverse-transform">Absolute Jacobian determinant of the unit vector inverse transform</h3>
+<p>The Jacobian matrix relating the input vector <span class="math inline">\(y\)</span> to the output vector <span class="math inline">\(x\)</span> is singular because <span class="math inline">\(x^\top x = 1\)</span> for any non-zero input vector <span class="math inline">\(y\)</span>. Thus, there technically is no unique transformation from <span class="math inline">\(x\)</span> to <span class="math inline">\(y\)</span>. To circumvent this issue, let <span class="math inline">\(r = \sqrt{y^\top y}\)</span> so that <span class="math inline">\(y = r
+x\)</span>. The transformation from <span class="math inline">\(\left(r, x_{-n}\right)\)</span> to <span class="math inline">\(y\)</span> is well-defined but <span class="math inline">\(r\)</span> is arbitrary, so we set <span class="math inline">\(r = 1\)</span>. In this case, the determinant of the Jacobian is proportional to <span class="math inline">\(e^{-\frac{1}{2} y^\top y}\)</span>, which is the kernel of a standard multivariate normal distribution with <span class="math inline">\(n\)</span> independent dimensions.</p>
+</section>
+</section>
+<section id="correlation-matrix-transform.section" class="level2">
+<h2 class="anchored" data-anchor-id="correlation-matrix-transform.section">Correlation matrices</h2>
+<p>A <span class="math inline">\(K \times K\)</span> correlation matrix <span class="math inline">\(x\)</span> must be symmetric, so that</p>
+<p><span class="math display">\[
+x_{k,k'} = x_{k',k}
+\]</span></p>
+<p>for all <span class="math inline">\(k,k' \in \{ 1, \ldots, K \}\)</span>, it must have a unit diagonal, so that</p>
+<p><span class="math display">\[
+x_{k,k} = 1
+\]</span></p>
+<p>for all <span class="math inline">\(k \in \{ 1, \ldots, K \}\)</span>, and it must be positive definite, so that for every non-zero <span class="math inline">\(K\)</span>-vector <span class="math inline">\(a\)</span>,</p>
+<p><span class="math display">\[
+a^{\top} x a &gt; 0.
+\]</span></p>
+<p>The number of free parameters required to specify a <span class="math inline">\(K \times K\)</span> correlation matrix is <span class="math inline">\(\binom{K}{2}\)</span>.</p>
+<p>There is more than one way to map from <span class="math inline">\(\binom{K}{2}\)</span> unconstrained parameters to a <span class="math inline">\(K \times K\)</span> correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform <span class="citation" data-cites="LewandowskiKurowickaJoe:2009">Lewandowski, Kurowicka, and Joe (<a href="#ref-LewandowskiKurowickaJoe:2009" role="doc-biblioref">2009</a>)</span>.</p>
+<section id="correlation-matrix-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="correlation-matrix-inverse-transform">Correlation matrix inverse transform</h3>
+<p>It is easiest to specify the inverse, going from its <span class="math inline">\(\binom{K}{2}\)</span> parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose <span class="math inline">\(y\)</span> is a vector containing <span class="math inline">\(\binom{K}{2}\)</span> unconstrained values. These are first transformed via the bijective function <span class="math inline">\(\tanh : \mathbb{R} \rightarrow
+(-1, 1)\)</span></p>
+<p><span class="math display">\[
+\tanh y = \frac{\exp(2y) - 1}{\exp(2y) + 1}.
+\]</span></p>
+<p>Then, define a <span class="math inline">\(K \times K\)</span> matrix <span class="math inline">\(z\)</span>, the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the <span class="math inline">\(4 \times 4\)</span> case, there are <span class="math inline">\(\binom{4}{2}\)</span> values arranged as</p>
+<p><span class="math display">\[
+z
+=
+\left[
+\begin{array}{cccc}
+1 &amp; \tanh y_1 &amp; \tanh y_2 &amp; \tanh y_4
+\\
+0 &amp; 1 &amp; \tanh y_3 &amp; \tanh y_5
+\\
+0 &amp; 0 &amp; 1 &amp; \tanh y_6
+\\
+0 &amp; 0 &amp; 0 &amp; 1
+\end{array}
+\right]
+.
+\]</span></p>
+<p>Lewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array <span class="math inline">\(z\)</span> to a correlation matrix <span class="math inline">\(x\)</span>. The entry <span class="math inline">\(z_{i,j}\)</span> for <span class="math inline">\(i &lt;
+j\)</span> is interpreted as the canonical partial correlation (CPC) between <span class="math inline">\(i\)</span> and <span class="math inline">\(j\)</span>, which is the correlation between <span class="math inline">\(i\)</span>’s residuals and <span class="math inline">\(j\)</span>’s residuals when both <span class="math inline">\(i\)</span> and <span class="math inline">\(j\)</span> are regressed on all variables <span class="math inline">\(i'\)</span> such that <span class="math inline">\(i'&lt; i\)</span>. In the case of <span class="math inline">\(i=1\)</span>, there are no earlier variables, so <span class="math inline">\(z_{1,j}\)</span> is just the Pearson correlation between <span class="math inline">\(i\)</span> and <span class="math inline">\(j\)</span>.</p>
+<p>In Stan, the LKJ transform is reformulated in terms of a Cholesky factor <span class="math inline">\(w\)</span> of the final correlation matrix, defined for <span class="math inline">\(1 \leq i,j \leq K\)</span> by</p>
+<p><span class="math display">\[
+w_{i,j} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } i &gt; j,
+\\
+1 &amp; \mbox{if } 1 = i = j,
+\\
+\prod_{i'=1}^{i - 1} \left( 1 - z_{i'\!,\,j}^2 \right)^{1/2}
+&amp; \mbox{if } 1 &lt; i = j,
+\\
+z_{i,j} &amp; \mbox{if } 1 = i &lt; j, \mbox{ and}
+\\\
+z_{i,j} \, \prod_{i'=1}^{i-1} \left( 1 - z_{i'\!,\,j}^2 \right)^{1/2}
+&amp; \mbox{ if } 1 &lt; i &lt; j.
+\end{array}
+\right.
+\]</span></p>
+<p>This does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression</p>
+<p><span class="math display">\[
+w_{i,j} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } i &gt; j,
+\\
+1 &amp; \mbox{if } 1 = i = j,
+\\
+z_{i,j} &amp; \mbox{if } 1 = i &lt; j, \mbox{ and}
+\\
+\frac{z_{i,j}}{z_{i-1,j}} \ w_{i-1,j} \left( 1 - z_{i-1,j}^2 \right)^{1/2}
+&amp; \mbox{ if } 1 &lt; i \leq j.
+\end{array}
+\right.
+\]</span></p>
+<p>Given the upper-triangular Cholesky factor <span class="math inline">\(w\)</span>, the final correlation matrix is</p>
+<p><span class="math display">\[
+x = w^{\top} w.
+\]</span></p>
+<p><span class="citation" data-cites="LewandowskiKurowickaJoe:2009">Lewandowski, Kurowicka, and Joe (<a href="#ref-LewandowskiKurowickaJoe:2009" role="doc-biblioref">2009</a>)</span> show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as</p>
+<p><span class="math display">\[
+\mbox{det} \, x = \prod_{i=1}^{K-1} \ \prod_{j=i+1}^K \ (1 - z_{i,j}^2)
+= \prod_{1 \leq i &lt; j \leq K} (1 - z_{i,j}^2),
+\]</span></p>
+</section>
+<section id="absolute-jacobian-determinant-of-the-correlation-matrix-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-the-correlation-matrix-inverse-transform">Absolute Jacobian determinant of the correlation matrix inverse transform</h3>
+<p>From the inverse of equation 11 in <span class="citation" data-cites="LewandowskiKurowickaJoe:2009">(<a href="#ref-LewandowskiKurowickaJoe:2009" role="doc-biblioref">Lewandowski, Kurowicka, and Joe 2009</a>)</span>, the absolute Jacobian determinant is</p>
+<p><span class="math display">\[
+\sqrt{\prod_{i=1}^{K-1}\prod_{j=i+1}^K \left(1-z_{i,j}^2\right)^{K-i-1}} \
+\times \prod_{i=1}^{K-1}\prod_{j=i+1}^K
+\frac{\partial z_{i,j}}{\partial y_{i,j}}
+\]</span></p>
+</section>
+<section id="correlation-matrix-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="correlation-matrix-transform">Correlation matrix transform</h3>
+<p>The correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.</p>
+<p>Starting with a correlation matrix <span class="math inline">\(x\)</span>, the first step is to find the unique upper triangular <span class="math inline">\(w\)</span> such that <span class="math inline">\(x = w w^{\top}\)</span>. Because <span class="math inline">\(x\)</span> is positive definite, this can be done by applying the Cholesky decomposition,</p>
+<p><span class="math display">\[
+w = \mbox{chol}(x).
+\]</span></p>
+<p>The next step from the Cholesky factor <span class="math inline">\(w\)</span> back to the array <span class="math inline">\(z\)</span> of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of <span class="math inline">\(w\)</span>, which when inverted yields</p>
+<p><span class="math display">\[
+z_{i,j} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } i \leq j,
+\\
+w_{i,j} &amp; \mbox{if } 1 = i &lt; j, \mbox{ and}
+\\
+{w_{i,j}}
+\
+\prod_{i'=1}^{i-1} \left( 1 - z_{i'\!,j}^2 \right)^{-1/2}
+&amp; \mbox{if } 1 &lt; i &lt; j.
+\end{array}
+\right.
+\]</span></p>
+<p>The final stage of the transform reverses the hyperbolic tangent transform, which is defined by</p>
+<p><span class="math display">\[
+y = \tanh^{-1} z = \frac{1}{2} \log \left( \frac{1 + z}{1 - z} \right).
+\]</span></p>
+<p>The inverse hyperbolic tangent function, <span class="math inline">\(\tanh^{-1}\)</span>, is also called the Fisher transformation.</p>
+</section>
+</section>
+<section id="covariance-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="covariance-matrices">Covariance matrices</h2>
+<p>A <span class="math inline">\(K \times K\)</span> matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires <span class="math inline">\(K + \binom{K}{2}\)</span> free parameters to specify a <span class="math inline">\(K \times K\)</span> covariance matrix.</p>
+<section id="covariance-matrix-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="covariance-matrix-transform">Covariance matrix transform</h3>
+<p>Stan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>If <span class="math inline">\(x\)</span> is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix <span class="math inline">\(z =
+\mathrm{chol}(x)\)</span> with positive diagonal entries, called a Cholesky factor, such that</p>
+<p><span class="math display">\[
+x = z \, z^{\top}.
+\]</span></p>
+<p>The off-diagonal entries of the Cholesky factor <span class="math inline">\(z\)</span> are unconstrained, but the diagonal entries <span class="math inline">\(z_{k,k}\)</span> must be positive for <span class="math inline">\(1 \leq k
+\leq K\)</span>.</p>
+<p>To complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix <span class="math inline">\(y\)</span> defined by</p>
+<p><span class="math display">\[
+y_{m,n} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } m &lt; n,
+\\
+\log z_{m,m} &amp; \mbox{if } m = n, \mbox{ and}
+\\
+z_{m,n} &amp; \mbox{if } m &gt; n.
+\end{array}
+\right.
+\]</span></p>
+</section>
+<section id="covariance-matrix-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="covariance-matrix-inverse-transform">Covariance matrix inverse transform</h3>
+<p>The inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular <span class="math inline">\(K \times K\)</span> matrix <span class="math inline">\(y\)</span>, the first step is to recover the intermediate matrix <span class="math inline">\(z\)</span> by reversing the log transform,</p>
+<p><span class="math display">\[
+z_{m,n} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } m &lt; n,
+\\
+\exp(y_{m,m}) &amp; \mbox{if } m = n, \mbox{ and}
+\\
+y_{m,n} &amp; \mbox{if } m &gt; n.
+\end{array}
+\right.
+\]</span></p>
+<p>The covariance matrix <span class="math inline">\(x\)</span> is recovered from its Cholesky factor <span class="math inline">\(z\)</span> by taking</p>
+<p><span class="math display">\[
+x = z \, z^{\top}.
+\]</span></p>
+</section>
+<section id="absolute-jacobian-determinant-of-the-covariance-matrix-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-the-covariance-matrix-inverse-transform">Absolute Jacobian determinant of the covariance matrix inverse transform</h3>
+<p>The Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix <span class="math inline">\(y\)</span> to matrix <span class="math inline">\(z\)</span> with positive diagonals and the product transform from the Cholesky factor <span class="math inline">\(z\)</span> to <span class="math inline">\(x\)</span>.</p>
+<p>The transform from unconstrained <span class="math inline">\(y\)</span> to Cholesky factor <span class="math inline">\(z\)</span> has a diagonal Jacobian matrix, the absolute determinant of which is thus</p>
+<p><span class="math display">\[
+\prod_{k=1}^K  \frac{\partial}{\partial_{y_{k,k}}} \, \exp(y_{k,k})
+\ = \
+\prod_{k=1}^K \exp(y_{k,k})
+\ = \
+\prod_{k=1}^K z_{k,k}.
+\]</span></p>
+<p>The Jacobian matrix of the second transform from the Cholesky factor <span class="math inline">\(z\)</span> to the covariance matrix <span class="math inline">\(x\)</span> is also triangular, with diagonal entries corresponding to pairs <span class="math inline">\((m,n)\)</span> with <span class="math inline">\(m \geq n\)</span>, defined by</p>
+<p><span class="math display">\[
+\frac{\partial}{\partial z_{m,n}}
+\left( z \, z^{\top} \right)_{m,n}
+\ = \
+\frac{\partial}{\partial z_{m,n}}
+\left( \sum_{k=1}^K z_{m,k} \, z_{n,k} \right)
+\ = \
+\left\{
+\begin{array}{cl}
+2 \, z_{n,n} &amp; \mbox{if } m = n \mbox{ and }
+\\
+z_{n,n} &amp; \mbox{if } m &gt; n.
+\end{array}
+\right.
+\]</span></p>
+<p>The absolute Jacobian determinant of the second transform is thus</p>
+<p><span class="math display">\[
+2^{K} \ \prod_{m = 1}^{K} \ \prod_{n=1}^{m} z_{n,n}
+\ = \
+\prod_{n=1}^K \ \prod_{m=n}^K z_{n,n}
+\ = \
+2^{K} \ \prod_{k=1}^K z_{k,k}^{K - k + 1}.
+\]</span></p>
+<p>Finally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular <span class="math inline">\(y\)</span> to a symmetric, positive definite matrix <span class="math inline">\(x\)</span> is the product of the Jacobian determinants of the exponentiation and product transforms,</p>
+<p><span class="math display">\[
+\left( \prod_{k=1}^K z_{k,k} \right)
+\left(
+2^{K} \ \prod_{k=1}^K z_{k,k}^{K - k + 1}
+\right)
+\ = \
+2^K
+\, \prod_{k=1}^K z_{k,k}^{K-k+2}.
+\]</span></p>
+<p>Let <span class="math inline">\(f^{-1}\)</span> be the inverse transform from a <span class="math inline">\(K + \binom{K}{2}\)</span>-vector <span class="math inline">\(y\)</span> to the <span class="math inline">\(K \times K\)</span> covariance matrix <span class="math inline">\(x\)</span>. A density function <span class="math inline">\(p_X(x)\)</span> defined on <span class="math inline">\(K \times K\)</span> covariance matrices is transformed to the density <span class="math inline">\(p_Y(y)\)</span> over <span class="math inline">\(K + \binom{K}{2}\)</span> vectors <span class="math inline">\(y\)</span> by</p>
+<p><span class="math display">\[
+p_Y(y) = p_X(f^{-1}(y)) \ 2^K \ \prod_{k=1}^K z_{k,k}^{K-k+2}.
+\]</span></p>
+</section>
+</section>
+<section id="cholesky-factors-of-covariance-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="cholesky-factors-of-covariance-matrices">Cholesky factors of covariance matrices</h2>
+<p>An <span class="math inline">\(M \times M\)</span> covariance matrix <span class="math inline">\(\Sigma\)</span> can be Cholesky factored to a lower triangular matrix <span class="math inline">\(L\)</span> such that <span class="math inline">\(L\,L^{\top} = \Sigma\)</span>. If <span class="math inline">\(\Sigma\)</span> is positive definite, then <span class="math inline">\(L\)</span> will be <span class="math inline">\(M \times M\)</span>. If <span class="math inline">\(\Sigma\)</span> is only positive semi-definite, then <span class="math inline">\(L\)</span> will be <span class="math inline">\(M \times N\)</span>, with <span class="math inline">\(N &lt; M\)</span>.</p>
+<p>A matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and <span class="math inline">\(M \geq
+N\)</span>. A matrix satisfying these conditions ensures that <span class="math inline">\(L \,
+L^{\top}\)</span> is positive semi-definite if <span class="math inline">\(M &gt; N\)</span> and positive definite if <span class="math inline">\(M = N\)</span>.</p>
+<p>A Cholesky factor of a covariance matrix requires <span class="math inline">\(N + \binom{N}{2} +
+(M - N)N\)</span> unconstrained parameters.</p>
+<section id="cholesky-factor-of-covariance-matrix-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factor-of-covariance-matrix-transform">Cholesky factor of covariance matrix transform</h3>
+<p>Stan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose <span class="math inline">\(x\)</span> is an <span class="math inline">\(M \times N\)</span> Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus</p>
+<p><span class="math display">\[
+y_{m,n} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } m &lt; n,
+\\
+\log x_{m,m} &amp; \mbox{if } m = n, \mbox{ and}
+\\
+x_{m,n} &amp; \mbox{if } m &gt; n.
+\end{array}
+\right.
+\]</span></p>
+</section>
+<section id="cholesky-factor-of-covariance-matrix-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factor-of-covariance-matrix-inverse-transform">Cholesky factor of covariance matrix inverse transform</h3>
+<p>The inverse transform need only invert the logarithm with an exponentiation. If <span class="math inline">\(y\)</span> is the unconstrained matrix representation, then the elements of the constrained matrix <span class="math inline">\(x\)</span> is defined by</p>
+<p><span class="math display">\[
+x_{m,n} =
+\left\{
+\begin{array}{cl}
+0 &amp; \mbox{if } m &lt; n,
+\\
+\exp(y_{m,m}) &amp; \mbox{if } m = n, \mbox{ and}
+\\
+y_{m,n} &amp; \mbox{if } m &gt; n.
+\end{array}
+\right.
+\]</span></p>
+</section>
+<section id="absolute-jacobian-determinant-of-cholesky-factor-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-cholesky-factor-inverse-transform">Absolute Jacobian determinant of Cholesky factor inverse transform</h3>
+<p>The transform has a diagonal Jacobian matrix, the absolute determinant of which is</p>
+<p><span class="math display">\[
+\prod_{n=1}^N  \frac{\partial}{\partial_{y_{n,n}}} \, \exp(y_{n,n})
+\ = \
+\prod_{n=1}^N \exp(y_{n,n})
+\ = \
+\prod_{n=1}^N x_{n,n}.
+\]</span></p>
+<p>Let <span class="math inline">\(x = f^{-1}(y)\)</span> be the inverse transform from a <span class="math inline">\(N + \binom{N}{2}
++ (M - N)N\)</span> vector to an <span class="math inline">\(M \times N\)</span> Cholesky factor for a covariance matrix <span class="math inline">\(x\)</span> defined in the previous section. A density function <span class="math inline">\(p_X(x)\)</span> defined on <span class="math inline">\(M \times N\)</span> Cholesky factors of covariance matrices is transformed to the density <span class="math inline">\(p_Y(y)\)</span> over <span class="math inline">\(N + \binom{N}{2}
++ (M - N)N\)</span> vectors <span class="math inline">\(y\)</span> by</p>
+<p><span class="math display">\[
+p_Y(y) = p_X(f^{-1}(y)) \prod_{N=1}^N x_{n,n}.
+\]</span></p>
+</section>
+</section>
+<section id="cholesky-factors-of-correlation-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="cholesky-factors-of-correlation-matrices">Cholesky factors of correlation matrices</h2>
+<p>A <span class="math inline">\(K \times K\)</span> correlation matrix <span class="math inline">\(\Omega\)</span> is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a <span class="math inline">\(K \times K\)</span> lower-triangular matrix <span class="math inline">\(L\)</span> with positive diagonal elements such that <span class="math inline">\(\Omega = L\,L^{\top}\)</span>. Because the correlation matrix has a unit diagonal,</p>
+<p><span class="math display">\[
+\Omega_{k,k} = L_k\,L_k^{\top} = 1,
+\]</span></p>
+<p>each row vector <span class="math inline">\(L_k\)</span> of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of <span class="math inline">\(L\)</span> to be calculated from the off-diagonal elements, so that a Cholesky factor for a <span class="math inline">\(K \times K\)</span> correlation matrix requires only <span class="math inline">\(\binom{K}{2}\)</span> unconstrained parameters.</p>
+<section id="cholesky-factor-of-correlation-matrix-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factor-of-correlation-matrix-inverse-transform">Cholesky factor of correlation matrix inverse transform</h3>
+<p>It is easiest to start with the inverse transform from the <span class="math inline">\(\binom{K}{2}\)</span> unconstrained parameters <span class="math inline">\(y\)</span> to the <span class="math inline">\(K \times K\)</span> lower-triangular Cholesky factor <span class="math inline">\(x\)</span>. The inverse transform is based on the hyperbolic tangent function, <span class="math inline">\(\tanh\)</span>, which satisfies <span class="math inline">\(\tanh(x) \in (-1,1)\)</span>. Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the <a href="#correlation-matrix-transform.section">section on correlation matrix transforms</a> for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.</p>
+<p>Suppose <span class="math inline">\(y\)</span> is a vector of <span class="math inline">\(\binom{K}{2}\)</span> unconstrained values. Let <span class="math inline">\(z\)</span> be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the <span class="math inline">\(3 \times 3\)</span> case,</p>
+<p><span class="math display">\[
+z =
+\left[
+\begin{array}{ccc}
+0 &amp; 0 &amp; 0
+\\
+\tanh y_1 &amp; 0 &amp; 0
+\\
+\tanh y_2 &amp; \tanh y_3 &amp; 0
+\end{array}
+\right]
+\]</span></p>
+<p>The matrix <span class="math inline">\(z\)</span>, with entries in the range <span class="math inline">\((-1, 1)\)</span>, is then transformed to the Cholesky factor <span class="math inline">\(x\)</span>, by taking<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p><span class="math display">\[
+x_{i,j}
+=
+\left\{
+\begin{array}{lll}
+0 &amp; \mbox{ if } i &lt; j &amp; \mbox{ [above diagonal]}
+\\
+\sqrt{1 - \sum_{j' &lt; j} x_{i,j'}^2}
+  &amp; \mbox{ if } i = j &amp; \mbox{ [on diagonal]}
+\\
+z_{i,j} \ \sqrt{1 - \sum_{j' &lt; j} x_{i,j'}^2}
+  &amp; \mbox{ if } i &gt; j &amp; \mbox{ [below diagonal]}
+\end{array}
+\right.
+\]</span></p>
+<p>In the <span class="math inline">\(3 \times 3\)</span> case, this yields</p>
+<p><span class="math display">\[
+x =
+\left[
+\begin{array}{ccc}
+1 &amp; 0 &amp; 0
+\\
+z_{2,1} &amp; \sqrt{1 - x_{2,1}^2} &amp; 0
+\\
+z_{3,1} &amp; z_{3,2} \sqrt{1 - x_{3,1}^2}
+        &amp; \sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}
+\end{array}
+\right],
+\]</span></p>
+<p>where the <span class="math inline">\(z_{i,j} \in (-1,1)\)</span> are the <span class="math inline">\(\tanh\)</span>-transformed <span class="math inline">\(y\)</span>.</p>
+<p>The approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at <span class="math inline">\(j=1\)</span>, each below-diagonal entry <span class="math inline">\(x_{i,j}\)</span> is determined by the (signed) fraction <span class="math inline">\(z_{i,j}\)</span> of the remaining length for the row that it consumes. The diagonal entries <span class="math inline">\(x_{i,i}\)</span> get any leftover length from earlier entries in their row. The above-diagonal entries are zero.</p>
+</section>
+<section id="cholesky-factor-of-correlation-matrix-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factor-of-correlation-matrix-transform">Cholesky factor of correlation matrix transform</h3>
+<p>Suppose <span class="math inline">\(x\)</span> is a <span class="math inline">\(K \times K\)</span> Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values <span class="math inline">\(z\)</span> from <span class="math inline">\(x\)</span>,</p>
+<p><span class="math display">\[
+z_{i,j} = \frac{x_{i,j}}{\sqrt{1 - \sum_{j' &lt; j}x_{i,j'}^2}}.
+\]</span></p>
+<p>The mapping from the resulting <span class="math inline">\(z\)</span> to <span class="math inline">\(y\)</span> inverts <span class="math inline">\(\tanh\)</span>,</p>
+<p><span class="math display">\[
+y
+\ = \
+\tanh^{-1} z
+\ = \
+\frac{1}{2} \left( \log (1 + z) - \log (1 - z) \right).
+\]</span></p>
+</section>
+<section id="absolute-jacobian-determinant-of-inverse-transform" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of-inverse-transform">Absolute Jacobian determinant of inverse transform</h3>
+<p>The Jacobian of the full transform is the product of the Jacobians of its component transforms.</p>
+<p>First, for the inverse transform <span class="math inline">\(z = \tanh y\)</span>, the derivative is</p>
+<p><span class="math display">\[
+\frac{d}{dy} \tanh y = \frac{1}{(\cosh y)^2}.
+\]</span></p>
+<p>Second, for the inverse transform of <span class="math inline">\(z\)</span> to <span class="math inline">\(x\)</span>, the resulting Jacobian matrix <span class="math inline">\(J\)</span> is of dimension <span class="math inline">\(\binom{K}{2} \times
+\binom{K}{2}\)</span>, with indexes <span class="math inline">\((i,j)\)</span> for <span class="math inline">\((i &gt; j)\)</span>. The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each <span class="math inline">\((i,j)\)</span> pair,</p>
+<p><span class="math display">\[
+\left| \, \mbox{det} \, J \, \right|
+  \ = \ \prod_{i &gt; j} \left| \frac{d}{dz_{i,j}} x_{i,j} \right|,
+\]</span></p>
+<p>where</p>
+<p><span class="math display">\[
+\frac{d}{dz_{i,j}} x_{i,j}
+= \sqrt{1 - \sum_{j' &lt; j} x^2_{i,j'}}.
+\]</span></p>
+<p>So the combined density for unconstrained <span class="math inline">\(y\)</span> is</p>
+<p><span class="math display">\[
+p_Y(y)
+= p_X(f^{-1}(y))
+  \ \
+  \prod_{n &lt; \binom{K}{2}} \frac{1}{(\cosh y)^2}
+  \ \
+  \prod_{i &gt; j} \left( 1 - \sum_{j' &lt; j} x_{i,j'}^2
+  \right)^{1/2},
+\]</span></p>
+<p>where <span class="math inline">\(x = f^{-1}(y)\)</span> is used for notational convenience. The log Jacobian determinant of the complete inverse transform <span class="math inline">\(x = f^{-1}(y)\)</span> is given by</p>
+<p><span class="math display">\[
+\log \left| \, \det J \, \right|
+=
+-2 \sum_{n \leq \binom{K}{2}}
+\log \cosh y
+\
++
+\
+\frac{1}{2} \
+\sum_{i &gt; j}
+\log \left( 1 - \sum_{j' &lt; j} x_{i,j'}^2 \right)
+.
+\]</span></p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-egozcue+etal:2003" class="csl-entry" role="listitem">
+Egozcue, Juan José, Vera Pawlowsky-Glahn, Glòria Mateu-Figueras, and Carles Barcelo-Vidal. 2003. <span>“Isometric Logratio Transformations for Compositional Data Analysis.”</span> <em>Mathematical Geology</em> 35 (3): 279–300.
+</div>
+<div id="ref-filzmoser+etal:2018" class="csl-entry" role="listitem">
+Filzmoser, Peter, Karel Hron, and Matthias Templ. 2018. <em>Geometrical Properties of Compositional Data</em>. Springer.
+</div>
+<div id="ref-lancaster:1965" class="csl-entry" role="listitem">
+Lancaster, H. O. 1965. <span>“The <span>H</span>elmert Matrices.”</span> <em>The American Mathematical Monthly</em> 72 (1): 4–12. <a href="http://www.jstor.org/stable/2312989">http://www.jstor.org/stable/2312989</a>.
+</div>
+<div id="ref-LewandowskiKurowickaJoe:2009" class="csl-entry" role="listitem">
+Lewandowski, Daniel, Dorota Kurowicka, and Harry Joe. 2009. <span>“Generating Random Correlation Matrices Based on Vines and Extended Onion Method.”</span> <em>Journal of Multivariate Analysis</em> 100: 1989–2001.
+</div>
+<div id="ref-Muller:1959" class="csl-entry" role="listitem">
+Muller, Mervin E. 1959. <span>“A Note on a Method for Generating Points Uniformly on n-Dimensional Spheres.”</span> <em>Commun. ACM</em> 2 (4): 19–20. <a href="https://doi.org/10.1145/377939.377946">https://doi.org/10.1145/377939.377946</a>.
+</div>
+<div id="ref-seyboldt:2024" class="csl-entry" role="listitem">
+Seyboldt, Adrian. 2024. <span>“Add ZeroSumNormal Distribution.”</span> <a href="https://github.com/pyro-ppl/numpyro/pull/1751#issuecomment-1980569811">https://github.com/pyro-ppl/numpyro/pull/1751#issuecomment-1980569811</a>.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>An alternative to the transform in this section, which can be coded directly in Stan, is to parameterize a covariance matrix as a scaled correlation matrix. An arbitrary <span class="math inline">\(K \times K\)</span> covariance matrix <span class="math inline">\(\Sigma\)</span> can be expressed in terms of a <span class="math inline">\(K\)</span>-vector <span class="math inline">\(\sigma\)</span> and correlation matrix <span class="math inline">\(\Omega\)</span> as <span class="math display">\[\Sigma = \mbox{diag}(\sigma) \times \Omega \times \mbox{diag}(\sigma),\]</span> so that each entry is just a deviation-scaled correlation, <span class="math display">\[\Sigma_{m,n} = \sigma_m \times \sigma_n \times \Omega_{m,n}.\]</span><a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>For convenience, a summation with no terms, such as <span class="math inline">\(\sum_{j' &lt; 1} x_{i,j'}\)</span>, is defined to be 0. This implies <span class="math inline">\(x_{1,1} = 1\)</span> and that <span class="math inline">\(x_{i,1} = z_{i,1}\)</span> for <span class="math inline">\(i &gt; 1\)</span>.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/user-functions.html" class="pagination-link" aria-label="User-Defined Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">User-Defined Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/syntax.html" class="pagination-link" aria-label="Language Syntax">
+        <span class="nav-page-text">Language Syntax</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/transforms.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/types.html b/docs/2_39/reference-manual/types.html
new file mode 100644
index 000000000..b9fc1d933
--- /dev/null
+++ b/docs/2_39/reference-manual/types.html
@@ -0,0 +1,2326 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Data Types and Declarations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/expressions.html" rel="next">
+<link href="../reference-manual/whitespace.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Data Types and Declarations">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/types.html">Data Types and Declarations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#data-types.chapter" id="toc-data-types.chapter" class="nav-link active" data-scroll-target="#data-types.chapter">Data Types and Declarations</a>
+  <ul class="collapse">
+  <li><a href="#overview-of-data-types" id="toc-overview-of-data-types" class="nav-link" data-scroll-target="#overview-of-data-types">Overview of data types</a>
+  <ul class="collapse">
+  <li><a href="#primitive-types" id="toc-primitive-types" class="nav-link" data-scroll-target="#primitive-types">Primitive types</a></li>
+  <li><a href="#complex-types" id="toc-complex-types" class="nav-link" data-scroll-target="#complex-types">Complex types</a></li>
+  <li><a href="#vector-and-matrix-types" id="toc-vector-and-matrix-types" class="nav-link" data-scroll-target="#vector-and-matrix-types">Vector and matrix types</a></li>
+  <li><a href="#array-types" id="toc-array-types" class="nav-link" data-scroll-target="#array-types">Array types</a></li>
+  <li><a href="#tuple-types" id="toc-tuple-types" class="nav-link" data-scroll-target="#tuple-types">Tuple types</a></li>
+  <li><a href="#constrained-data-types" id="toc-constrained-data-types" class="nav-link" data-scroll-target="#constrained-data-types">Constrained data types</a></li>
+  </ul></li>
+  <li><a href="#numerical-data-types" id="toc-numerical-data-types" class="nav-link" data-scroll-target="#numerical-data-types">Primitive numerical data types</a>
+  <ul class="collapse">
+  <li><a href="#integers" id="toc-integers" class="nav-link" data-scroll-target="#integers">Integers</a></li>
+  <li><a href="#reals" id="toc-reals" class="nav-link" data-scroll-target="#reals">Reals</a></li>
+  <li><a href="#promoting-integers-to-reals" id="toc-promoting-integers-to-reals" class="nav-link" data-scroll-target="#promoting-integers-to-reals">Promoting integers to reals</a></li>
+  </ul></li>
+  <li><a href="#complex-numerical-data-type" id="toc-complex-numerical-data-type" class="nav-link" data-scroll-target="#complex-numerical-data-type">Complex numerical data type</a>
+  <ul class="collapse">
+  <li><a href="#constructing-and-accessing-complex-numbers" id="toc-constructing-and-accessing-complex-numbers" class="nav-link" data-scroll-target="#constructing-and-accessing-complex-numbers">Constructing and accessing complex numbers</a></li>
+  <li><a href="#promoting-real-to-complex" id="toc-promoting-real-to-complex" class="nav-link" data-scroll-target="#promoting-real-to-complex">Promoting real to complex</a></li>
+  </ul></li>
+  <li><a href="#scalar-data-types-and-variable-declarations" id="toc-scalar-data-types-and-variable-declarations" class="nav-link" data-scroll-target="#scalar-data-types-and-variable-declarations">Scalar data types and variable declarations</a>
+  <ul class="collapse">
+  <li><a href="#unconstrained-integer" id="toc-unconstrained-integer" class="nav-link" data-scroll-target="#unconstrained-integer">Unconstrained integer</a></li>
+  <li><a href="#constrained-integer" id="toc-constrained-integer" class="nav-link" data-scroll-target="#constrained-integer">Constrained integer</a></li>
+  <li><a href="#unconstrained-real" id="toc-unconstrained-real" class="nav-link" data-scroll-target="#unconstrained-real">Unconstrained real</a></li>
+  <li><a href="#unconstrained-complex" id="toc-unconstrained-complex" class="nav-link" data-scroll-target="#unconstrained-complex">Unconstrained complex</a></li>
+  <li><a href="#constrained-real" id="toc-constrained-real" class="nav-link" data-scroll-target="#constrained-real">Constrained real</a></li>
+  <li><a href="#affine-transform.section" id="toc-affine-transform.section" class="nav-link" data-scroll-target="#affine-transform.section">Affinely transformed real</a></li>
+  <li><a href="#expressions-as-bounds-and-offsetmultiplier" id="toc-expressions-as-bounds-and-offsetmultiplier" class="nav-link" data-scroll-target="#expressions-as-bounds-and-offsetmultiplier">Expressions as bounds and offset/multiplier</a></li>
+  <li><a href="#declaring-optional-variables" id="toc-declaring-optional-variables" class="nav-link" data-scroll-target="#declaring-optional-variables">Declaring optional variables</a></li>
+  </ul></li>
+  <li><a href="#vector-and-matrix-data-types" id="toc-vector-and-matrix-data-types" class="nav-link" data-scroll-target="#vector-and-matrix-data-types">Vector and matrix data types</a>
+  <ul class="collapse">
+  <li><a href="#indexing-from-1" id="toc-indexing-from-1" class="nav-link" data-scroll-target="#indexing-from-1">Indexing from 1</a></li>
+  <li><a href="#vectors" id="toc-vectors" class="nav-link" data-scroll-target="#vectors">Vectors</a></li>
+  <li><a href="#complex-vectors" id="toc-complex-vectors" class="nav-link" data-scroll-target="#complex-vectors">Complex vectors</a></li>
+  <li><a href="#unit-simplexes" id="toc-unit-simplexes" class="nav-link" data-scroll-target="#unit-simplexes">Unit simplexes</a></li>
+  <li><a href="#stochastic-matrices" id="toc-stochastic-matrices" class="nav-link" data-scroll-target="#stochastic-matrices">Stochastic Matrices</a></li>
+  <li><a href="#unit-vectors" id="toc-unit-vectors" class="nav-link" data-scroll-target="#unit-vectors">Unit vectors</a></li>
+  <li><a href="#vectors-that-sum-to-zero" id="toc-vectors-that-sum-to-zero" class="nav-link" data-scroll-target="#vectors-that-sum-to-zero">Vectors that sum to zero</a></li>
+  <li><a href="#ordered-vectors" id="toc-ordered-vectors" class="nav-link" data-scroll-target="#ordered-vectors">Ordered vectors</a></li>
+  <li><a href="#positive-ordered-vectors" id="toc-positive-ordered-vectors" class="nav-link" data-scroll-target="#positive-ordered-vectors">Positive, ordered vectors</a></li>
+  <li><a href="#row-vectors" id="toc-row-vectors" class="nav-link" data-scroll-target="#row-vectors">Row vectors</a></li>
+  <li><a href="#complex-row-vectors" id="toc-complex-row-vectors" class="nav-link" data-scroll-target="#complex-row-vectors">Complex row vectors</a></li>
+  <li><a href="#matrices" id="toc-matrices" class="nav-link" data-scroll-target="#matrices">Matrices</a></li>
+  <li><a href="#complex-matrices" id="toc-complex-matrices" class="nav-link" data-scroll-target="#complex-matrices">Complex matrices</a></li>
+  <li><a href="#covariance-matrices" id="toc-covariance-matrices" class="nav-link" data-scroll-target="#covariance-matrices">Covariance matrices</a></li>
+  <li><a href="#correlation-matrices" id="toc-correlation-matrices" class="nav-link" data-scroll-target="#correlation-matrices">Correlation matrices</a></li>
+  <li><a href="#cholesky-factors-of-covariance-matrices" id="toc-cholesky-factors-of-covariance-matrices" class="nav-link" data-scroll-target="#cholesky-factors-of-covariance-matrices">Cholesky factors of covariance matrices</a></li>
+  <li><a href="#cholesky-factors-of-correlation-matrices" id="toc-cholesky-factors-of-correlation-matrices" class="nav-link" data-scroll-target="#cholesky-factors-of-correlation-matrices">Cholesky factors of correlation matrices</a></li>
+  <li><a href="#matrices-that-sum-to-zero" id="toc-matrices-that-sum-to-zero" class="nav-link" data-scroll-target="#matrices-that-sum-to-zero">Matrices that sum to zero</a></li>
+  <li><a href="#assigning-constrained-variables" id="toc-assigning-constrained-variables" class="nav-link" data-scroll-target="#assigning-constrained-variables">Assigning constrained variables</a></li>
+  <li><a href="#promoting-real-to-complex-matrixes" id="toc-promoting-real-to-complex-matrixes" class="nav-link" data-scroll-target="#promoting-real-to-complex-matrixes">Promoting real to complex matrixes</a></li>
+  <li><a href="#expressions-as-size-declarations" id="toc-expressions-as-size-declarations" class="nav-link" data-scroll-target="#expressions-as-size-declarations">Expressions as size declarations</a></li>
+  <li><a href="#accessing-vector-and-matrix-elements" id="toc-accessing-vector-and-matrix-elements" class="nav-link" data-scroll-target="#accessing-vector-and-matrix-elements">Accessing vector and matrix elements</a></li>
+  <li><a href="#size-declaration-restrictions" id="toc-size-declaration-restrictions" class="nav-link" data-scroll-target="#size-declaration-restrictions">Size declaration restrictions</a></li>
+  </ul></li>
+  <li><a href="#array-data-types.section" id="toc-array-data-types.section" class="nav-link" data-scroll-target="#array-data-types.section">Array data types</a>
+  <ul class="collapse">
+  <li><a href="#declaring-array-variables" id="toc-declaring-array-variables" class="nav-link" data-scroll-target="#declaring-array-variables">Declaring array variables</a></li>
+  <li><a href="#accessing-array-elements-and-subarrays" id="toc-accessing-array-elements-and-subarrays" class="nav-link" data-scroll-target="#accessing-array-elements-and-subarrays">Accessing array elements and subarrays</a></li>
+  <li><a href="#assigning" id="toc-assigning" class="nav-link" data-scroll-target="#assigning">Assigning</a></li>
+  <li><a href="#arrays-of-matrices-and-vectors" id="toc-arrays-of-matrices-and-vectors" class="nav-link" data-scroll-target="#arrays-of-matrices-and-vectors">Arrays of matrices and vectors</a></li>
+  <li><a href="#partial-array-assignment" id="toc-partial-array-assignment" class="nav-link" data-scroll-target="#partial-array-assignment">Partial array assignment</a></li>
+  <li><a href="#mixing-array-vector-and-matrix-types" id="toc-mixing-array-vector-and-matrix-types" class="nav-link" data-scroll-target="#mixing-array-vector-and-matrix-types">Mixing array, vector, and matrix types</a></li>
+  <li><a href="#size-declaration-restrictions-1" id="toc-size-declaration-restrictions-1" class="nav-link" data-scroll-target="#size-declaration-restrictions-1">Size declaration restrictions</a></li>
+  <li><a href="#size-zero-arrays" id="toc-size-zero-arrays" class="nav-link" data-scroll-target="#size-zero-arrays">Size zero arrays</a></li>
+  </ul></li>
+  <li><a href="#tuple-data-type" id="toc-tuple-data-type" class="nav-link" data-scroll-target="#tuple-data-type">Tuple data type</a>
+  <ul class="collapse">
+  <li><a href="#declaring-tuple-variables" id="toc-declaring-tuple-variables" class="nav-link" data-scroll-target="#declaring-tuple-variables">Declaring tuple variables</a></li>
+  <li><a href="#accessing-tuple-elements" id="toc-accessing-tuple-elements" class="nav-link" data-scroll-target="#accessing-tuple-elements">Accessing tuple elements</a></li>
+  <li><a href="#assigning-tuple-elements" id="toc-assigning-tuple-elements" class="nav-link" data-scroll-target="#assigning-tuple-elements">Assigning tuple elements</a></li>
+  <li><a href="#unpacking-assignment-of-tuples" id="toc-unpacking-assignment-of-tuples" class="nav-link" data-scroll-target="#unpacking-assignment-of-tuples">Unpacking assignment of tuples</a></li>
+  </ul></li>
+  <li><a href="#variable-types-vs.-constraints-and-sizes" id="toc-variable-types-vs.-constraints-and-sizes" class="nav-link" data-scroll-target="#variable-types-vs.-constraints-and-sizes">Variable types vs.&nbsp;constraints and sizes</a>
+  <ul class="collapse">
+  <li><a href="#type-information-excludes-sizes" id="toc-type-information-excludes-sizes" class="nav-link" data-scroll-target="#type-information-excludes-sizes">Type information excludes sizes</a></li>
+  <li><a href="#type-information-excludes-constraints" id="toc-type-information-excludes-constraints" class="nav-link" data-scroll-target="#type-information-excludes-constraints">Type information excludes constraints</a></li>
+  <li><a href="#type-naming-notation" id="toc-type-naming-notation" class="nav-link" data-scroll-target="#type-naming-notation">Type naming notation</a></li>
+  </ul></li>
+  <li><a href="#variable-declaration.section" id="toc-variable-declaration.section" class="nav-link" data-scroll-target="#variable-declaration.section">Variable declaration</a></li>
+  <li><a href="#compound-variable-declaration-and-definition" id="toc-compound-variable-declaration-and-definition" class="nav-link" data-scroll-target="#compound-variable-declaration-and-definition">Compound variable declaration and definition</a>
+  <ul class="collapse">
+  <li><a href="#assignment-typing" id="toc-assignment-typing" class="nav-link" data-scroll-target="#assignment-typing">Assignment typing</a></li>
+  <li><a href="#rhs-expressions" id="toc-rhs-expressions" class="nav-link" data-scroll-target="#rhs-expressions">Right-hand side expressions</a></li>
+  <li><a href="#scope-within-expressions" id="toc-scope-within-expressions" class="nav-link" data-scroll-target="#scope-within-expressions">Scope within expressions</a></li>
+  </ul></li>
+  <li><a href="#declaring-multiple-variables-at-once" id="toc-declaring-multiple-variables-at-once" class="nav-link" data-scroll-target="#declaring-multiple-variables-at-once">Declaring multiple variables at once</a>
+  <ul class="collapse">
+  <li><a href="#types-for-multiple-declarations" id="toc-types-for-multiple-declarations" class="nav-link" data-scroll-target="#types-for-multiple-declarations">Types for multiple declarations</a></li>
+  <li><a href="#combining-with-other-features" id="toc-combining-with-other-features" class="nav-link" data-scroll-target="#combining-with-other-features">Combining with other features</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/types.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/types.html">Data Types and Declarations</a></li></ol></nav></header>
+
+
+
+
+<section id="data-types.chapter" class="level1">
+<h1>Data Types and Declarations</h1>
+<p>This chapter covers the data types for expressions in Stan. Every variable used in a Stan program must have a declared data type. Only values of that type will be assignable to the variable (except for temporary states of transformed data and transformed parameter values). This follows the convention of programming languages like C++, not the conventions of scripting languages like Python or statistical languages such as R or BUGS.</p>
+<p>The motivation for strong, static typing is threefold.</p>
+<ol type="1">
+<li>Strong typing forces the programmer’s intent to be declared with the variable, making programs easier to comprehend and hence easier to debug and maintain.</li>
+<li>Strong typing allows programming errors relative to the declared intent to be caught sooner (at compile time) rather than later (at run time). The Stan compiler (called through an interface such as CmdStan, RStan, or PyStan) will flag any type errors and indicate the offending expressions quickly when the program is compiled.</li>
+<li>Constrained types will catch runtime data, initialization, and intermediate value errors as soon as they occur rather than allowing them to propagate and potentially pollute final results.</li>
+</ol>
+<p>Strong typing disallows assigning the same variable to objects of different types at different points in the program or in different invocations of the program.</p>
+<section id="overview-of-data-types" class="level2">
+<h2 class="anchored" data-anchor-id="overview-of-data-types">Overview of data types</h2>
+<p>Arguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.</p>
+<p>Passing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.</p>
+<section id="primitive-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="primitive-types">Primitive types</h3>
+<p>Stan provides two primitive data types, <code>real</code> for continuous values and <code>int</code> for integer values. These are both considered scalar types.</p>
+</section>
+<section id="complex-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-types">Complex types</h3>
+<p>Stan provides a complex number data type <code>complex</code>, where a complex number contains both a real and an imaginary component, both of which are of type <code>real</code>. Complex types are considered scalar types.</p>
+</section>
+<section id="vector-and-matrix-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vector-and-matrix-types">Vector and matrix types</h3>
+<p>Stan provides three real-valued matrix data types, <code>vector</code> for column vectors, <code>row_vector</code> for row vectors, and <code>matrix</code> for matrices.</p>
+<p>Stan also provides three complex-valued matrix data types, <code>complex_vector</code> for column vectors, <code>complex_row_vector</code> for row vectors, and <code>complex_matrix</code> for matrices.</p>
+</section>
+<section id="array-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="array-types">Array types</h3>
+<p>Any type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">10</span>] <span class="dt">real</span> x;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">6</span>, <span class="dv">7</span>] <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">3</span>] m;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">12</span>, <span class="dv">8</span>, <span class="dv">15</span>] <span class="dt">complex</span> z;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares <code>x</code> to be a one-dimensional array of size 10 containing real values, declares <code>m</code> to be a two-dimensional array of size <span class="math inline">\(6 \times 7\)</span> containing values that are <span class="math inline">\(3 \times 3\)</span> matrices, and declares <code>z</code> to be a <span class="math inline">\(12 \times 8 \times 15\)</span> array of complex numbers.</p>
+<p>Prior to 2.26 Stan models used a different syntax which has since been removed. See the <a href="../reference-manual/removals.html">Removed Features</a> chapter for more details.</p>
+</section>
+<section id="tuple-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="tuple-types">Tuple types</h3>
+<p>For any sequence of types, Stan provides a tuple data type. For example,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">real</span>, <span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">int</span>) xi;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares <code>xi</code> to be a tuple holding two values, the first of which is of type type <code>real</code> and the second of which a 5-dimensional array of type <code>int</code>.</p>
+</section>
+<section id="constrained-data-types" class="level3">
+<h3 class="anchored" data-anchor-id="constrained-data-types">Constrained data types</h3>
+<p>Declarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.</p>
+<p>All of the basic data types other than <code>complex</code> may be given lower and upper bounds using syntax such as</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt; log_p;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[<span class="dv">3</span>] rho;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There are also special data types for structured vectors and matrices. There are five constrained vector data types, <code>simplex</code> for unit simplexes, <code>unit_vector</code> for unit-length vectors, <code>sum_to_zero_vector</code> for vectors that sum to zero, <code>ordered</code> for ordered vectors of scalars, and <code>positive_ordered</code> for vectors of positive ordered scalars. There are specialized matrix data types <code>corr_matrix</code> and <code>cov_matrix</code> for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type <code>cholesky_factor_cov</code> is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type <code>cholesky_factor_corr</code> is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type <code>sum_to_zero_matrix</code> is for matrices that sum to zero across rows and columns.</p>
+<p>Constraints provide error checking for variables defined in the <code>data</code>, <code>transformed data</code>, <code>transformed parameters</code>, and <code>generated quantities</code> blocks. Constraints are critical for variables declared in the <code>parameters</code> block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of <span class="math inline">\(\mathbb{R}^n\)</span>).</p>
+<p>It is worth calling out the most important aspect of constrained data types:</p>
+<div style="margin:0 2em 0 2em">
+<p><em>The model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.</em></p>
+</div>
+<p>If this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.</p>
+</section>
+</section>
+<section id="numerical-data-types" class="level2">
+<h2 class="anchored" data-anchor-id="numerical-data-types">Primitive numerical data types</h2>
+<p>Unfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.</p>
+<section id="integers" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="integers">Integers</h3>
+<p>Stan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is <span class="math inline">\(2^{31}-1\)</span>; the minimum value is <span class="math inline">\(-(2^{31})\)</span>.</p>
+<p>When integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.</p>
+<p>Integer arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see <a href="https://mc-stan.org/docs/functions-reference/integer-valued_basic_functions.html#int-arithmetic">the Stan Functions Reference integer-valued arithmetic operators section</a> for more information).</p>
+</section>
+<section id="reals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reals">Reals</h3>
+<p>Stan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly <span class="math inline">\(\pm 2^{1022}\)</span>, which is slightly larger than <span class="math inline">\(\pm 10^{307}\)</span>. It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.</p>
+<p>64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.</p>
+<p>There are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.</p>
+<section id="not-a-number" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="not-a-number">Not-a-number</h4>
+<p>The not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).</p>
+</section>
+<section id="infinite-values" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="infinite-values">Infinite values</h4>
+<p>Positive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.</p>
+</section>
+</section>
+<section id="promoting-integers-to-reals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="promoting-integers-to-reals">Promoting integers to reals</h3>
+<p>Stan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.</p>
+<p>Unlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.</p>
+</section>
+</section>
+<section id="complex-numerical-data-type" class="level2">
+<h2 class="anchored" data-anchor-id="complex-numerical-data-type">Complex numerical data type</h2>
+<p>The <code>complex</code> data type is a scalar, but unlike <code>real</code> and <code>int</code> types, it contains two components, a real and imaginary component, both of which are of type <code>real</code>. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.</p>
+<section id="constructing-and-accessing-complex-numbers" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constructing-and-accessing-complex-numbers">Constructing and accessing complex numbers</h3>
+<p>Imaginary literals are written in mathematical notation using a numeral followed by the suffix <code>i</code>. For example, the following example constructs a complex number <span class="math inline">\(2 - 1.3i\)</span> and assigns it to the variable <code>z</code>.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = <span class="dv">2</span> - <span class="fl">1.3i</span>;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> re = get_real(z);  <span class="co">// re has value 2.0</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> im = get_imag(z);  <span class="co">// im has value -1.3</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The getter functions then extract the real and imaginary components of <code>z</code> and assign them to <code>re</code> and <code>im</code> respectively.</p>
+<p>The function <code>to_complex</code> constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[K] re;</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[K] im;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex</span> z = to_complex(re[k], im[k]);</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="promoting-real-to-complex" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="promoting-real-to-complex">Promoting real to complex</h3>
+<p>Expressions of type <code>real</code> may be assigned to variables of type <code>complex</code>. For example, the following is a valid sequence of Stan statements.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="fl">5.0</span>;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = x;  <span class="co">// get_real(z) == 5.0, get_imag(z) == 0</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.</p>
+<p>Assignability is transitive, so that expressions of type <code>int</code> may also be assigned to variables of type <code>complex</code>, as in the following example.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n = <span class="dv">2</span>;</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = n;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Function arguments also support promotion of integer or real typed expressions to type <code>complex</code>.</p>
+</section>
+</section>
+<section id="scalar-data-types-and-variable-declarations" class="level2">
+<h2 class="anchored" data-anchor-id="scalar-data-types-and-variable-declarations">Scalar data types and variable declarations</h2>
+<p>All variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.</p>
+<section id="unconstrained-integer" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unconstrained-integer">Unconstrained integer</h3>
+<p>Unconstrained integers are declared using the <code>int</code> keyword. For example, the variable <code>N</code> is declared to be an integer as follows.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="constrained-integer" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constrained-integer">Constrained integer</h3>
+<p>Integer data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare <code>N</code> to be a positive integer, use the following.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This illustrates that the bounds are inclusive for integers.</p>
+<p>To declare an integer variable <code>cond</code> to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; cond;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="unconstrained-real" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unconstrained-real">Unconstrained real</h3>
+<p>Unconstrained real variables are declared using the keyword <code>real</code>. The following example declares <code>theta</code> to be an unconstrained continuous value.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="unconstrained-complex" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unconstrained-complex">Unconstrained complex</h3>
+<p>Unconstrained complex numbers are declared using the keyword <code>complex</code>. The following example declares <code>z</code> to be an unconstrained complex variable.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="constrained-real" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constrained-real">Constrained real</h3>
+<p>Real variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.</p>
+<p>The variable <code>sigma</code> may be declared to be non-negative as follows.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The following declares the variable <code>x</code> to be less than or equal to <span class="math inline">\(-1\)</span>.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">upper</span>=-<span class="dv">1</span>&gt; x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To ensure <code>rho</code> takes on values between <span class="math inline">\(-1\)</span> and <span class="math inline">\(1\)</span>, use the following declaration.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; rho;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="infinite-constraints" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="infinite-constraints">Infinite constraints</h4>
+<p>Lower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants <code>positive_infinity()</code> and <code>negative_infinity()</code> which may be used for this purpose, or they may be supplied as data.</p>
+</section>
+</section>
+<section id="affine-transform.section" class="level3">
+<h3 class="anchored" data-anchor-id="affine-transform.section">Affinely transformed real</h3>
+<p>Real variables may be declared on a space that has been transformed using an affine transformation <span class="math inline">\(x\mapsto \mu + \sigma * x\)</span> with offset <span class="math inline">\(\mu\)</span> and (positive) multiplier <span class="math inline">\(\sigma\)</span>, using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.</p>
+<p>The variable <code>x</code> may be declared to have offset <span class="math inline">\(1\)</span> as follows.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="dt">offset</span>=<span class="dv">1</span>&gt; x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Similarly, it can be declared to have multiplier <span class="math inline">\(2\)</span> as follows.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="dt">multiplier</span>=<span class="dv">2</span>&gt; x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Finally, we can combine both declarations to declare a variable with offset <span class="math inline">\(1\)</span> and multiplier <span class="math inline">\(2\)</span>.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="dt">offset</span>=<span class="dv">1</span>, <span class="dt">multiplier</span>=<span class="dv">2</span>&gt; x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As an example, we can give <code>x</code> a normal distribution with non-centered parameterization as follows.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="dt">offset</span>=mu, <span class="dt">multiplier</span>=sigma&gt; x;</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>  x ~ normal(mu, sigma);</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Recall that the centered parameterization is achieved with the code</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>  x ~ normal(mu, sigma);</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>or equivalently</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="dt">offset</span>=<span class="dv">0</span>, <span class="dt">multiplier</span>=<span class="dv">1</span>&gt; x;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>  x ~ normal(mu, sigma);</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="expressions-as-bounds-and-offsetmultiplier" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="expressions-as-bounds-and-offsetmultiplier">Expressions as bounds and offset/multiplier</h3>
+<p>Bounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.</p>
+<p>For example, it is acceptable to have the following declarations.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">real</span> lb;</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a>   <span class="dt">real</span>&lt;<span class="kw">lower</span>=lb&gt; phi;</span>
+<span id="cb22-6"><a href="#cb22-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This declares a real-valued parameter <code>phi</code> to take values greater than the value of the real-valued data variable <code>lb</code>. Constraints may be arbitrary expressions, but must be of type <code>int</code> for integer variables and of type <code>real</code> for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>   <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>   <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>   <span class="dt">real</span>&lt;<span class="kw">lower</span>=min(y), <span class="kw">upper</span>=max(y)&gt; phi;</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This declares a positive integer data variable <code>N</code>, an array <code>y</code> of real-valued data of length <code>N</code>, and then a parameter ranging between the minimum and maximum value of <code>y</code>. As shown in the example code, the functions <code>min()</code> and <code>max()</code> may be applied to containers such as arrays.</p>
+<p>A more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=a&gt; b;  <span class="co">// enforces a &lt; b</span></span>
+<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> c;</span>
+<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=c&gt; d;</span>
+<span id="cb24-8"><a href="#cb24-8" aria-hidden="true" tabindex="-1"></a>  c = a;</span>
+<span id="cb24-9"><a href="#cb24-9" aria-hidden="true" tabindex="-1"></a>  d = b;</span>
+<span id="cb24-10"><a href="#cb24-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though <code>c</code> isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.</p>
+</section>
+<section id="declaring-optional-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="declaring-optional-variables">Declaring optional variables</h3>
+<p>A variable may be declared with a size that depends on a boolean constant. For example, consider the definition of <code>alpha</code> in the following program fragment.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; include_alpha;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[include_alpha ? N : <span class="dv">0</span>] alpha;</span>
+<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If <code>include_alpha</code> is true, the model will include the vector <code>alpha</code>; if the flag is false, the model will not include <code>alpha</code> (technically, it will include <code>alpha</code> of size 0, which means it won’t contain any values and won’t be included in any output).</p>
+<p>This technique is not just useful for containers. If the value of <code>N</code> is set to 1, then the vector <code>alpha</code> will contain a single element and thus <code>alpha[1]</code> behaves like an optional scalar, the existence of which is controlled by <code>include_alpha</code>.</p>
+<p>This coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the <a href="https://mc-stan.org/rstanarm">RStanArm package</a>.</p>
+</section>
+</section>
+<section id="vector-and-matrix-data-types" class="level2">
+<h2 class="anchored" data-anchor-id="vector-and-matrix-data-types">Vector and matrix data types</h2>
+<p>Stan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A <span class="math inline">\(3 \times 4\)</span> matrix is a different kind of object in Stan than a <span class="math inline">\(3
+\times 4\)</span> array.</p>
+<p>The intention of using matrix types is to call out their usage in the code. There are three situations in Stan where <em>only</em> vectors and matrices may be used,</p>
+<ul>
+<li>matrix arithmetic operations (e.g., matrix multiplication)</li>
+<li>linear algebra functions (e.g., eigenvalues and determinants), and</li>
+<li>multivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).</li>
+</ul>
+<p>Vectors and matrices cannot be typed to return integer values. They are restricted to <code>real</code> and <code>complex</code> values.</p>
+<p>For constructing vectors and matrices in Stan, see <a href="../reference-manual/expressions.html#vector-matrix-array-expressions.section">Vector, Matrix, and Array Expressions</a>.</p>
+<section id="indexing-from-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="indexing-from-1">Indexing from 1</h3>
+<p>Vectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.</p>
+</section>
+<section id="vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectors">Vectors</h3>
+<p>Vectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword <code>vector</code>, as follows.</p>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">3</span>] u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Vectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[<span class="dv">3</span>] u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Similarly, they may be declared with a offset and/or multiplier, as in the following example</p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="dt">offset</span>=<span class="dv">42</span>, <span class="dt">multiplier</span>=<span class="dv">3</span>&gt;[<span class="dv">3</span>] u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="complex-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-vectors">Complex vectors</h3>
+<p>Like real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword <code>complex_vector</code>, as follows.</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_vector</span>[<span class="dv">3</span>] v;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex vector declarations do not support any constraints.</p>
+</section>
+<section id="unit-simplexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unit-simplexes">Unit simplexes</h3>
+<p>A unit simplex is a vector with non-negative values whose entries sum to 1. For instance, <span class="math inline">\([0.2,0.3,0.4,0.1]^{\top}\)</span> is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, <code>theta</code> is declared to be a unit <span class="math inline">\(5\)</span>-simplex by</p>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="dt">simplex</span>[<span class="dv">5</span>] theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Unit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold <span class="math inline">\(\epsilon\)</span> to account for errors arising from floating-point imprecision.</p>
+<p>In high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.</p>
+</section>
+<section id="stochastic-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stochastic-matrices">Stochastic Matrices</h3>
+<p>A stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a <span class="math inline">\(3 \times 4\)</span> column-stochastic matrix.</p>
+<p><span class="math display">\[
+\begin{bmatrix}
+0.2 &amp; 0.5 &amp; 0.1 &amp; 0.3 \\
+0.3 &amp; 0.3 &amp; 0.6 &amp; 0.4 \\
+0.5 &amp; 0.2 &amp; 0.3 &amp; 0.3
+\end{bmatrix}
+\]</span></p>
+<p>An example of a <span class="math inline">\(3 \times 4\)</span> row-stochastic matrix is the following.</p>
+<p><span class="math display">\[
+\begin{bmatrix}
+0.2 &amp; 0.5 &amp; 0.1 &amp; 0.2 \\
+0.2 &amp; 0.1 &amp; 0.6 &amp; 0.1 \\
+0.5 &amp; 0.2 &amp; 0.2 &amp; 0.1
+\end{bmatrix}
+\]</span></p>
+<p>In the examples above, each column (or row) sums to 1, making the matrices valid <code>column_stochastic_matrix</code> and <code>row_stochastic_matrix</code> types.</p>
+<p>Column-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.</p>
+<p>The <code>column_stochastic_matrix</code> and <code>row_stochastic_matrix</code> types are declared with row and column sizes. For instance, a matrix <code>theta</code> with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="dt">column_stochastic_matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A matrix <code>theta</code> with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a><span class="dt">row_stochastic_matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As with simplexes, <code>column_stochastic_matrix</code> and <code>row_stochastic_matrix</code> variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold <span class="math inline">\(\epsilon\)</span>.</p>
+<section id="stability-considerations" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="stability-considerations">Stability Considerations</h4>
+<p>In high-dimensional settings, <code>column_stochastic_matrix</code> and <code>row_stochastic_matrix</code> types may require careful tuning of the inference algorithms. To ensure stability:</p>
+<ul>
+<li><strong>Smaller Step Sizes:</strong> In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.</li>
+<li><strong>Higher Target Acceptance Rates:</strong> Setting higher target acceptance rates can improve the robustness of the sampling process.</li>
+<li><strong>Longer Warmup Periods:</strong> Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.</li>
+<li><strong>Tighter Optimization Tolerances:</strong> For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.</li>
+<li><strong>Custom Initialization:</strong> If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.</li>
+</ul>
+</section>
+</section>
+<section id="unit-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unit-vectors">Unit vectors</h3>
+<p>A unit vector is a vector with a norm of one. For instance, <span class="math inline">\([0.5,
+0.5, 0.5, 0.5]^{\top}\)</span> is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, <code>theta</code> is declared to be a unit <span class="math inline">\(5\)</span>-vector by</p>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a><span class="dt">unit_vector</span>[<span class="dv">5</span>] theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Unit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold <span class="math inline">\(\epsilon\)</span> to account for errors arising from floating-point imprecision.</p>
+</section>
+<section id="vectors-that-sum-to-zero" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectors-that-sum-to-zero">Vectors that sum to zero</h3>
+<p>A zero-sum vector is constrained such that the sum of its elements is always <span class="math inline">\(0\)</span>. These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only <span class="math inline">\(N - 1\)</span> degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, <code>beta</code> is declared to be a zero-sum <span class="math inline">\(5\)</span>-vector (4 DoF) by</p>
+<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a><span class="dt">sum_to_zero_vector</span>[<span class="dv">5</span>] beta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Zero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold <span class="math inline">\(\epsilon\)</span> to account for errors arising from floating-point imprecision.</p>
+</section>
+<section id="ordered-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ordered-vectors">Ordered vectors</h3>
+<p>An ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, <span class="math inline">\((-1.3,2.7,2.71)^{\top}\)</span> is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see <a href="https://mc-stan.org/docs/stan-users-guide/regression.html#ordered-logistic.section">section</a>).</p>
+<p>The variable <code>c</code> is declared as an ordered 5-vector by</p>
+<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a><span class="dt">ordered</span>[<span class="dv">5</span>] c;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.</p>
+</section>
+<section id="positive-ordered-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="positive-ordered-vectors">Positive, ordered vectors</h3>
+<p>There is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, <span class="math inline">\((2,3.7,4,12.9)\)</span> is a positive, ordered 4-vector.</p>
+<p>The variable <code>d</code> is declared as a positive, ordered 5-vector by</p>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a><span class="dt">positive_ordered</span>[<span class="dv">5</span>] d;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Like ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.</p>
+</section>
+<section id="row-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="row-vectors">Row vectors</h3>
+<p>Row vectors are declared with the keyword <code>row_vector</code>. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector <code>u</code> would be declared as</p>
+<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">1093</span>] u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Constraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.</p>
+<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[<span class="dv">10</span>] u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Offset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.</p>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>&lt;<span class="dt">offset</span>=-<span class="dv">42</span>, <span class="dt">multiplier</span>=<span class="dv">3</span>&gt;[<span class="dv">3</span>] u;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Row vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.</p>
+</section>
+<section id="complex-row-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-row-vectors">Complex row vectors</h3>
+<p>Complex row vectors are declared with the keyword <code>complex_row_vector</code> and given a size in basic declarations. For example, a 12-dimensional complex row vector <code>v</code> would be declared as</p>
+<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_row_vector</span>[<span class="dv">12</span>] v;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex row vectors do not allow constraints.</p>
+</section>
+<section id="matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrices">Matrices</h3>
+<p>Matrices are declared with the keyword <code>matrix</code> along with a number of rows and number of columns. For example,</p>
+<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">3</span>] A;</span>
+<span id="cb41-2"><a href="#cb41-2" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] B;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares <code>A</code> to be a <span class="math inline">\(3 \times 3\)</span> matrix and <code>B</code> to be a <span class="math inline">\(M
+\times N\)</span> matrix. For the second declaration to be well formed, the variables <code>M</code> and <code>N</code> must be declared as integers in either the data or transformed data block and before the matrix declaration.</p>
+<p>Matrices may also be declared with constraints, as in this (<span class="math inline">\(3 \times 4\)</span>) matrix of non-positive values.</p>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt;[<span class="dv">3</span>, <span class="dv">4</span>] B;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Similarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.</p>
+<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>&lt;<span class="dt">multiplier</span>=<span class="dv">5</span>&gt;[<span class="dv">3</span>, <span class="dv">4</span>] B;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="assigning-to-rows-of-a-matrix" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="assigning-to-rows-of-a-matrix">Assigning to rows of a matrix</h4>
+<p>Rows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.</p>
+<div class="sourceCode" id="cb44"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb44-1"><a href="#cb44-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] a;</span>
+<span id="cb44-2"><a href="#cb44-2" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[N] b;</span>
+<span id="cb44-3"><a href="#cb44-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb44-4"><a href="#cb44-4" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">1</span>] = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This copies the values from row vector <code>b</code> to <code>a[1]</code>, which is the first row of the matrix <code>a</code>. If the number of columns in <code>a</code> is not the same as the size of <code>b</code>, a run-time error is raised; the number of columns of <code>a</code> is <code>N</code>, which is also the number of columns of <code>b</code>.</p>
+<p>Assignment works by copying values in Stan. That means any subsequent assignment to <code>a[1]</code> does not affect <code>b</code>, nor does an assignment to <code>b</code> affect <code>a</code>.</p>
+</section>
+</section>
+<section id="complex-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-matrices">Complex matrices</h3>
+<p>Complex matrices are declared with the keyword <code>complex_matrix</code> and a number of rows and columns. For example,</p>
+<div class="sourceCode" id="cb45"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb45-1"><a href="#cb45-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_matrix</span>[<span class="dv">3</span>, <span class="dv">3</span>] C;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex matrices do not allow constraints.</p>
+</section>
+<section id="covariance-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="covariance-matrices">Covariance matrices</h3>
+<p>Matrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,</p>
+<div class="sourceCode" id="cb46"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb46-1"><a href="#cb46-1" aria-hidden="true" tabindex="-1"></a><span class="dt">cov_matrix</span>[K] Sigma;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares <code>Sigma</code> to be a <span class="math inline">\(K \times K\)</span> covariance matrix, where <span class="math inline">\(K\)</span> is the value of the data variable <code>K</code>.</p>
+</section>
+<section id="correlation-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="correlation-matrices">Correlation matrices</h3>
+<p>Matrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between <span class="math inline">\(-1\)</span> and <span class="math inline">\(1\)</span>, and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,</p>
+<div class="sourceCode" id="cb47"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb47-1"><a href="#cb47-1" aria-hidden="true" tabindex="-1"></a><span class="dt">corr_matrix</span>[<span class="dv">3</span>] Omega;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares <code>Omega</code> to be a <span class="math inline">\(3 \times 3\)</span> correlation matrix.</p>
+<p>Correlation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.</p>
+</section>
+<section id="cholesky-factors-of-covariance-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factors-of-covariance-matrices">Cholesky factors of covariance matrices</h3>
+<p>Matrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.</p>
+<p>A Cholesky factor <span class="math inline">\(L\)</span> is an <span class="math inline">\(M \times N\)</span> lower-triangular matrix (if <span class="math inline">\(m &lt; n\)</span> then <span class="math inline">\(L[m, n] =0\)</span>) with a strictly positive diagonal (<span class="math inline">\(L[k, k]
+&gt; 0\)</span>) and <span class="math inline">\(M \geq N\)</span>. If <span class="math inline">\(L\)</span> is a Cholesky factor, then <span class="math inline">\(\Sigma = L
+\, L^{\top}\)</span> is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.</p>
+<p>The typical case of a square Cholesky factor may be declared with a single dimension,</p>
+<div class="sourceCode" id="cb48"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb48-1"><a href="#cb48-1" aria-hidden="true" tabindex="-1"></a><span class="dt">cholesky_factor_cov</span>[<span class="dv">4</span>] L;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="cholesky-factors-of-positive-semi-definite-matrices" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="cholesky-factors-of-positive-semi-definite-matrices">Cholesky factors of positive semi-definite matrices</h4>
+<p>In general, two dimensions may be declared, with the above being equal to <code>cholesky_factor_cov[4, 4]</code>. The type <code>cholesky_factor_cov[M, N]</code> may be used for the general <span class="math inline">\(M \times N\)</span> case to produce positive semi-definite matrices of rank <span class="math inline">\(M\)</span>.</p>
+</section>
+</section>
+<section id="cholesky-factors-of-correlation-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factors-of-correlation-matrices">Cholesky factors of correlation matrices</h3>
+<p>Matrix variables may be constrained to represent the Cholesky factors of a correlation matrix.</p>
+<p>A Cholesky factor for a correlation matrix <span class="math inline">\(L\)</span> is a <span class="math inline">\(K \times K\)</span> lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., <span class="math inline">\(\sum_{n=1}^K L_{m,n}^2 = 1\)</span>). If <span class="math inline">\(L\)</span> is a Cholesky factor for a correlation matrix, then <span class="math inline">\(L\,L^{\top}\)</span> is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).</p>
+<p>To declare the variable <code>L</code> to be a <code>K</code> by <code>K</code> Cholesky factor of a correlation matrix, the following code may be used.</p>
+<div class="sourceCode" id="cb49"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb49-1"><a href="#cb49-1" aria-hidden="true" tabindex="-1"></a><span class="dt">cholesky_factor_corr</span>[K] L;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="matrices-that-sum-to-zero" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrices-that-sum-to-zero">Matrices that sum to zero</h3>
+<p>A sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always <span class="math inline">\(0\)</span>. These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only <span class="math inline">\((N - 1) \times (M - 1)\)</span> degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, <code>beta</code> is declared to be a sum-to-zero <span class="math inline">\(5 \times 4\)</span>-matrix (20 degrees of freedom) by</p>
+<div class="sourceCode" id="cb50"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb50-1"><a href="#cb50-1" aria-hidden="true" tabindex="-1"></a><span class="dt">sum_to_zero_matrix</span>[<span class="dv">5</span>, <span class="dv">4</span>] beta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Sum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold <span class="math inline">\(\epsilon\)</span> to account for errors arising from floating-point imprecision.</p>
+</section>
+<section id="assigning-constrained-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="assigning-constrained-variables">Assigning constrained variables</h3>
+<p>Constrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be <code>real&lt;lower=0, upper=1&gt;</code> could be assigned to a variable declared as <code>real</code> and vice-versa. Similarly, a variable declared as <code>matrix[3, 3]</code> may be assigned to a variable declared as <code>cov_matrix[3]</code> or <code>cholesky_factor_cov[3]</code>, and vice-versa.</p>
+<p>Checks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.</p>
+</section>
+<section id="promoting-real-to-complex-matrixes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="promoting-real-to-complex-matrixes">Promoting real to complex matrixes</h3>
+<p>Real-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.</p>
+<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] v = ...;</span>
+<span id="cb51-2"><a href="#cb51-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_vector</span>[N] u = <span class="dv">2</span> * v;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Row vectors and matrices work the same way.</p>
+</section>
+<section id="expressions-as-size-declarations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="expressions-as-size-declarations">Expressions as size declarations</h3>
+<p>Variables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.</p>
+<div class="sourceCode" id="cb52"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb52-1"><a href="#cb52-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb52-2"><a href="#cb52-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_observed, N_missing;</span>
+<span id="cb52-3"><a href="#cb52-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb52-4"><a href="#cb52-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb52-5"><a href="#cb52-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_observed + N_missing] y;</span>
+<span id="cb52-6"><a href="#cb52-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="accessing-vector-and-matrix-elements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="accessing-vector-and-matrix-elements">Accessing vector and matrix elements</h3>
+<p>If <code>v</code> is a column vector or row vector, then <code>v[2]</code> is the second element in the vector. If <code>m</code> is a matrix, then <code>m[2, 3]</code> is the value in the second row and third column.</p>
+<p>Providing a matrix with a single index returns the specified row. For instance, if <code>m</code> is a matrix, then <code>m[2]</code> is the second row. This allows Stan blocks such as</p>
+<div class="sourceCode" id="cb53"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb53-1"><a href="#cb53-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] m;</span>
+<span id="cb53-2"><a href="#cb53-2" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[N] v;</span>
+<span id="cb53-3"><a href="#cb53-3" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x;</span>
+<span id="cb53-4"><a href="#cb53-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb53-5"><a href="#cb53-5" aria-hidden="true" tabindex="-1"></a>v = m[<span class="dv">2</span>];</span>
+<span id="cb53-6"><a href="#cb53-6" aria-hidden="true" tabindex="-1"></a>x = v[<span class="dv">3</span>];   <span class="co">// x == m[2][3] == m[2, 3]</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The type of <code>m[2]</code> is <code>row_vector</code> because it is the second row of <code>m</code>. Thus it is possible to write <code>m[2][3]</code> instead of <code>m[2, 3]</code> to access the third element in the second row. When given a choice, the form <code>m[2, 3]</code> is preferred.</p>
+<p>Complex versions work the same way,</p>
+<div class="sourceCode" id="cb54"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb54-1"><a href="#cb54-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_matrix</span>[M, N] m = ...;</span>
+<span id="cb54-2"><a href="#cb54-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_row_vector</span>[N] u = m[<span class="dv">3</span>];</span>
+<span id="cb54-3"><a href="#cb54-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_vector</span>[M] v = m[ , <span class="dv">2</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="array-index-style" class="level4">
+<h4 class="anchored" data-anchor-id="array-index-style">Array index style</h4>
+<p>The form <code>m[2, 3]</code> is more efficient because it does not require the creation and use of an intermediate expression template for <code>m[2]</code>. In later versions, explicit calls to <code>m[2][3]</code> may be optimized to be as efficient as <code>m[2, 3]</code> by the Stan compiler.</p>
+</section>
+</section>
+<section id="size-declaration-restrictions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="size-declaration-restrictions">Size declaration restrictions</h3>
+<p>An integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size <code>M + N</code> using</p>
+<div class="sourceCode" id="cb55"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb55-1"><a href="#cb55-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[M + N] y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Any integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.</p>
+</section>
+</section>
+<section id="array-data-types.section" class="level2">
+<h2 class="anchored" data-anchor-id="array-data-types.section">Array data types</h2>
+<p>Stan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.</p>
+<p>A two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if <code>a</code> is a two-dimensional array, then <code>a[m, n]</code> is just a convenient shorthand for <code>a[m][n]</code>.</p>
+<p>Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.</p>
+<p>For constructing arrays in Stan, see <a href="../reference-manual/expressions.html#vector-matrix-array-expressions.section">Vector, Matrix, and Array Expressions</a>.</p>
+<section id="declaring-array-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="declaring-array-variables">Declaring array variables</h3>
+<p>Arrays are declared with the keyword <code>array</code> followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.</p>
+<p>The variable <code>n</code> is declared as an array of five integers as follows.</p>
+<div class="sourceCode" id="cb56"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb56-1"><a href="#cb56-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">int</span> n;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A two-dimensional array of complex values with three rows and four columns is declared as follows.</p>
+<div class="sourceCode" id="cb57"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb57-1"><a href="#cb57-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">complex</span> a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A three-dimensional array <code>z</code> of positive reals with five rows, four columns, and two shelves can be declared as follows.</p>
+<div class="sourceCode" id="cb58"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb58-1"><a href="#cb58-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">4</span>, <span class="dv">2</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; z;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Arrays may also be declared to contain vectors. For example,</p>
+<div class="sourceCode" id="cb59"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb59-1"><a href="#cb59-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">vector</span>[<span class="dv">7</span>] mu;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares <code>mu</code> to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example</p>
+<div class="sourceCode" id="cb60"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb60-1"><a href="#cb60-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">15</span>, <span class="dv">12</span>] <span class="dt">complex_matrix</span>[<span class="dv">7</span>, <span class="dv">2</span>] mu;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares a 15 by 12 array of <span class="math inline">\(7 \times 2\)</span> complex matrices. Any of the constrained types may also be used in arrays, as in the declaration</p>
+<div class="sourceCode" id="cb61"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb61-1"><a href="#cb61-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">cholesky_factor_cov</span>[<span class="dv">5</span>, <span class="dv">6</span>] mu;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>of a <span class="math inline">\(2 \times 3 \times 4\)</span> array of <span class="math inline">\(5 \times 6\)</span> Cholesky factors of covariance matrices.</p>
+</section>
+<section id="accessing-array-elements-and-subarrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="accessing-array-elements-and-subarrays">Accessing array elements and subarrays</h3>
+<p>If <code>x</code> is a 1-dimensional array of length 5, then <code>x[1]</code> is the first element in the array and <code>x[5]</code> is the last. For a <span class="math inline">\(3
+\times 4\)</span> array <code>y</code> of two dimensions, <code>y[1, 1]</code> is the first element and <code>y[3, 4]</code> the last element. For a three-dimensional array <code>z</code>, the first element is <code>z[1, 1, 1]</code>, and so on.</p>
+<p>Subarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose <code>y</code> is a two-dimensional array with three rows and four columns. Then <code>y[3]</code> is one-dimensional array of length four. This means that <code>y[3][1]</code> may be used instead of <code>y[3, 1]</code> to access the value of the first column of the third row of <code>y</code>. The form <code>y[3, 1]</code> is the preferred form (see <a href="#array-index-style">note</a> in this chapter).</p>
+</section>
+<section id="assigning" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="assigning">Assigning</h3>
+<p>Subarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as</p>
+<div class="sourceCode" id="cb62"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb62-1"><a href="#cb62-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">9</span>, <span class="dv">10</span>, <span class="dv">11</span>] <span class="dt">real</span> w;</span>
+<span id="cb62-2"><a href="#cb62-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">10</span>, <span class="dv">11</span>] <span class="dt">real</span> x;</span>
+<span id="cb62-3"><a href="#cb62-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">11</span>] <span class="dt">real</span> y;</span>
+<span id="cb62-4"><a href="#cb62-4" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> z;</span>
+<span id="cb62-5"><a href="#cb62-5" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb62-6"><a href="#cb62-6" aria-hidden="true" tabindex="-1"></a>x = w[<span class="dv">5</span>];</span>
+<span id="cb62-7"><a href="#cb62-7" aria-hidden="true" tabindex="-1"></a>y = x[<span class="dv">4</span>];  <span class="co">// y == w[5][4] == w[5, 4]</span></span>
+<span id="cb62-8"><a href="#cb62-8" aria-hidden="true" tabindex="-1"></a>z = y[<span class="dv">3</span>];  <span class="co">// z == w[5][4][3] == w[5, 4, 3]</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex-valued arrays work the same way.</p>
+</section>
+<section id="arrays-of-matrices-and-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="arrays-of-matrices-and-vectors">Arrays of matrices and vectors</h3>
+<p>Arrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.</p>
+<div class="sourceCode" id="cb63"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb63-1"><a href="#cb63-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">vector</span>[<span class="dv">5</span>] a;</span>
+<span id="cb63-2"><a href="#cb63-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">vector</span>[<span class="dv">5</span>] b;</span>
+<span id="cb63-3"><a href="#cb63-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">5</span>] c;</span>
+<span id="cb63-4"><a href="#cb63-4" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With these declarations, the following assignments are legal.</p>
+<div class="sourceCode" id="cb64"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb64-1"><a href="#cb64-1" aria-hidden="true" tabindex="-1"></a>b = a[<span class="dv">1</span>];       <span class="co">// result is array of vectors</span></span>
+<span id="cb64-2"><a href="#cb64-2" aria-hidden="true" tabindex="-1"></a>c = a[<span class="dv">1</span>, <span class="dv">3</span>];    <span class="co">// result is vector</span></span>
+<span id="cb64-3"><a href="#cb64-3" aria-hidden="true" tabindex="-1"></a>c = b[<span class="dv">3</span>];       <span class="co">//   same result as above</span></span>
+<span id="cb64-4"><a href="#cb64-4" aria-hidden="true" tabindex="-1"></a>x = a[<span class="dv">1</span>, <span class="dv">3</span>, <span class="dv">5</span>]; <span class="co">// result is scalar</span></span>
+<span id="cb64-5"><a href="#cb64-5" aria-hidden="true" tabindex="-1"></a>x = b[<span class="dv">3</span>, <span class="dv">5</span>];    <span class="co">//   same result as above</span></span>
+<span id="cb64-6"><a href="#cb64-6" aria-hidden="true" tabindex="-1"></a>x = c[<span class="dv">5</span>];       <span class="co">//   same result as above</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Row vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.</p>
+<p>Consider the following matrix, vector and scalar declarations.</p>
+<div class="sourceCode" id="cb65"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb65-1"><a href="#cb65-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">matrix</span>[<span class="dv">6</span>, <span class="dv">5</span>] d;</span>
+<span id="cb65-2"><a href="#cb65-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">matrix</span>[<span class="dv">6</span>, <span class="dv">5</span>] e;</span>
+<span id="cb65-3"><a href="#cb65-3" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">6</span>, <span class="dv">5</span>] f;</span>
+<span id="cb65-4"><a href="#cb65-4" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">5</span>] g;</span>
+<span id="cb65-5"><a href="#cb65-5" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With these declarations, the following definitions are legal.</p>
+<div class="sourceCode" id="cb66"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb66-1"><a href="#cb66-1" aria-hidden="true" tabindex="-1"></a>e = d[<span class="dv">1</span>];           <span class="co">// result is array of matrices</span></span>
+<span id="cb66-2"><a href="#cb66-2" aria-hidden="true" tabindex="-1"></a>f = d[<span class="dv">1</span>, <span class="dv">3</span>];        <span class="co">// result is matrix</span></span>
+<span id="cb66-3"><a href="#cb66-3" aria-hidden="true" tabindex="-1"></a>f = e[<span class="dv">3</span>];           <span class="co">//   same result as above</span></span>
+<span id="cb66-4"><a href="#cb66-4" aria-hidden="true" tabindex="-1"></a>g = d[<span class="dv">1</span>, <span class="dv">3</span>, <span class="dv">2</span>];     <span class="co">// result is row vector</span></span>
+<span id="cb66-5"><a href="#cb66-5" aria-hidden="true" tabindex="-1"></a>g = e[<span class="dv">3</span>, <span class="dv">2</span>];        <span class="co">//   same result as above</span></span>
+<span id="cb66-6"><a href="#cb66-6" aria-hidden="true" tabindex="-1"></a>g = f[<span class="dv">2</span>];           <span class="co">//   same result as above</span></span>
+<span id="cb66-7"><a href="#cb66-7" aria-hidden="true" tabindex="-1"></a>x = d[<span class="dv">1</span>, <span class="dv">3</span>, <span class="dv">5</span>, <span class="dv">2</span>];  <span class="co">// result is scalar</span></span>
+<span id="cb66-8"><a href="#cb66-8" aria-hidden="true" tabindex="-1"></a>x = e[<span class="dv">3</span>, <span class="dv">5</span>, <span class="dv">2</span>];     <span class="co">//   same result as above</span></span>
+<span id="cb66-9"><a href="#cb66-9" aria-hidden="true" tabindex="-1"></a>x = f[<span class="dv">5</span>, <span class="dv">2</span>];        <span class="co">//   same result as above</span></span>
+<span id="cb66-10"><a href="#cb66-10" aria-hidden="true" tabindex="-1"></a>x = g[<span class="dv">2</span>];           <span class="co">//   same result as above</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As shown, the result <code>f[2]</code> of supplying a single index to a matrix is the indexed row, here row 2 of matrix <code>f</code>.</p>
+</section>
+<section id="partial-array-assignment" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="partial-array-assignment">Partial array assignment</h3>
+<p>Subarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.</p>
+<div class="sourceCode" id="cb67"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb67-1"><a href="#cb67-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[I, J, K] <span class="dt">real</span> x;</span>
+<span id="cb67-2"><a href="#cb67-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[J, K] <span class="dt">real</span> y;</span>
+<span id="cb67-3"><a href="#cb67-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[K] <span class="dt">real</span> z;</span>
+<span id="cb67-4"><a href="#cb67-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb67-5"><a href="#cb67-5" aria-hidden="true" tabindex="-1"></a>x[<span class="dv">1</span>] = y;</span>
+<span id="cb67-6"><a href="#cb67-6" aria-hidden="true" tabindex="-1"></a>x[<span class="dv">1</span>, <span class="dv">1</span>] = z;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The sizes must match. Here, <code>x[1]</code> is a <code>J</code> by <code>K</code> array, as is <code>y</code>.</p>
+<p>Partial array assignment also works for arrays of matrices, vectors, and row vectors.</p>
+</section>
+<section id="mixing-array-vector-and-matrix-types" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mixing-array-vector-and-matrix-types">Mixing array, vector, and matrix types</h3>
+<p>Arrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).</p>
+<section id="mixing-vectors-and-arrays" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="mixing-vectors-and-arrays">Mixing vectors and arrays</h4>
+<p>For example, vectors cannot be assigned to arrays or vice-versa.</p>
+<div class="sourceCode" id="cb68"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb68-1"><a href="#cb68-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">real</span> a;</span>
+<span id="cb68-2"><a href="#cb68-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">4</span>] b;</span>
+<span id="cb68-3"><a href="#cb68-3" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">4</span>] c;</span>
+<span id="cb68-4"><a href="#cb68-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb68-5"><a href="#cb68-5" aria-hidden="true" tabindex="-1"></a>a = b; <span class="co">// illegal assignment of vector to array</span></span>
+<span id="cb68-6"><a href="#cb68-6" aria-hidden="true" tabindex="-1"></a>b = a; <span class="co">// illegal assignment of array to vector</span></span>
+<span id="cb68-7"><a href="#cb68-7" aria-hidden="true" tabindex="-1"></a>a = c; <span class="co">// illegal assignment of row vector to array</span></span>
+<span id="cb68-8"><a href="#cb68-8" aria-hidden="true" tabindex="-1"></a>c = a; <span class="co">// illegal assignment of array to row vector</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="mixing-row-and-column-vectors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="mixing-row-and-column-vectors">Mixing row and column vectors</h4>
+<p>It is not even legal to assign row vectors to column vectors or vice versa.</p>
+<div class="sourceCode" id="cb69"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb69-1"><a href="#cb69-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">4</span>] b;</span>
+<span id="cb69-2"><a href="#cb69-2" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">4</span>] c;</span>
+<span id="cb69-3"><a href="#cb69-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb69-4"><a href="#cb69-4" aria-hidden="true" tabindex="-1"></a>b = c; <span class="co">// illegal assignment of row vector to column vector</span></span>
+<span id="cb69-5"><a href="#cb69-5" aria-hidden="true" tabindex="-1"></a>c = b; <span class="co">// illegal assignment of column vector to row vector</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="mixing-matrices-and-arrays" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="mixing-matrices-and-arrays">Mixing matrices and arrays</h4>
+<p>The same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.</p>
+<div class="sourceCode" id="cb70"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb70-1"><a href="#cb70-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">real</span> a;</span>
+<span id="cb70-2"><a href="#cb70-2" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] b;</span>
+<span id="cb70-3"><a href="#cb70-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb70-4"><a href="#cb70-4" aria-hidden="true" tabindex="-1"></a>a = b;  <span class="co">// illegal assignment of matrix to array</span></span>
+<span id="cb70-5"><a href="#cb70-5" aria-hidden="true" tabindex="-1"></a>b = a;  <span class="co">// illegal assignment of array to matrix</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="mixing-matrices-and-vectors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="mixing-matrices-and-vectors">Mixing matrices and vectors</h4>
+<p>A <span class="math inline">\(1 \times N\)</span> matrix cannot be assigned a row vector or vice versa.</p>
+<div class="sourceCode" id="cb71"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb71-1"><a href="#cb71-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">1</span>, <span class="dv">4</span>] a;</span>
+<span id="cb71-2"><a href="#cb71-2" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">4</span>] b;</span>
+<span id="cb71-3"><a href="#cb71-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb71-4"><a href="#cb71-4" aria-hidden="true" tabindex="-1"></a>a = b;  <span class="co">// illegal assignment of row vector to matrix</span></span>
+<span id="cb71-5"><a href="#cb71-5" aria-hidden="true" tabindex="-1"></a>b = a;  <span class="co">// illegal assignment of matrix to row vector</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Similarly, an <span class="math inline">\(M \times 1\)</span> matrix may not be assigned to a column vector.</p>
+<div class="sourceCode" id="cb72"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb72-1"><a href="#cb72-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">4</span>, <span class="dv">1</span>] a;</span>
+<span id="cb72-2"><a href="#cb72-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">4</span>] b;</span>
+<span id="cb72-3"><a href="#cb72-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb72-4"><a href="#cb72-4" aria-hidden="true" tabindex="-1"></a>a = b;  <span class="co">// illegal assignment of column vector to matrix</span></span>
+<span id="cb72-5"><a href="#cb72-5" aria-hidden="true" tabindex="-1"></a>b = a;  <span class="co">// illegal assignment of matrix to column vector</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="size-declaration-restrictions-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="size-declaration-restrictions-1">Size declaration restrictions</h3>
+<p>An integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.</p>
+</section>
+<section id="size-zero-arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="size-zero-arrays">Size zero arrays</h3>
+<p>If any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare</p>
+<div class="sourceCode" id="cb73"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb73-1"><a href="#cb73-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">0</span>] <span class="dt">real</span> a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>then the resulting size of <code>a</code> is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, <code>a[1]</code> will be a size-zero one-dimensional array. For comparison, declaring</p>
+<div class="sourceCode" id="cb74"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb74-1"><a href="#cb74-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">0</span>, <span class="dv">3</span>] <span class="dt">real</span> b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>also produces an array with an overall size of zero, but in this case, there is no way to index legally into <code>b</code>, because <code>b[0]</code> is undefined. The array will behave at run time as if it’s a <span class="math inline">\(0 \times
+0\)</span> array. For example, the result of <code>to_matrix(b)</code> will be a <span class="math inline">\(0 \times 0\)</span> matrix, not a <span class="math inline">\(0 \times 3\)</span> matrix.</p>
+</section>
+</section>
+<section id="tuple-data-type" class="level2">
+<h2 class="anchored" data-anchor-id="tuple-data-type">Tuple data type</h2>
+<p>Stan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.</p>
+<section id="declaring-tuple-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="declaring-tuple-variables">Declaring tuple variables</h3>
+<p>Tuples are declared with the keyword <code>tuple</code> followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as</p>
+<div class="sourceCode" id="cb75"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb75-1"><a href="#cb75-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">vector</span>[<span class="dv">3</span>], <span class="dt">complex</span>) abc;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Tuples must have at least one entry, so the following declaration is illegal.</p>
+<div class="sourceCode" id="cb76"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb76-1"><a href="#cb76-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>() nil;  <span class="co">// ILLEGAL</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Tuples of length one must use a trailing comma, to align with the expression syntax.</p>
+<div class="sourceCode" id="cb77"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb77-1"><a href="#cb77-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>,) m; <span class="co">// CORRECT</span></span>
+<span id="cb77-2"><a href="#cb77-2" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>) n;  <span class="co">// ILLEGAL</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Tuples can be assigned as a whole if their elements can be assigned individually. For example, <code>a</code> can be assigned to <code>b</code> in the following example because <code>int</code> can be promoted to <code>complex</code>.</p>
+<div class="sourceCode" id="cb78"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb78-1"><a href="#cb78-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">real</span>) a;</span>
+<span id="cb78-2"><a href="#cb78-2" aria-hidden="true" tabindex="-1"></a>...</span>
+<span id="cb78-3"><a href="#cb78-3" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">complex</span>, <span class="dt">real</span>) b = a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Tuple types may have elements which are declared as tuples, such as the following example.</p>
+<div class="sourceCode" id="cb79"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb79-1"><a href="#cb79-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">tuple</span>(<span class="dt">real</span>, <span class="dt">complex</span>)) x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case, it would probably be simpler to use a 3-tuple type, <code>tuple(int, real, complex)</code>.</p>
+<p>Tuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration</p>
+<div class="sourceCode" id="cb80"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb80-1"><a href="#cb80-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb80-2"><a href="#cb80-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>it is legal to have a tuple with constraints such as</p>
+<div class="sourceCode" id="cb81"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb81-1"><a href="#cb81-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;, <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;) sigma_theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="accessing-tuple-elements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="accessing-tuple-elements">Accessing tuple elements</h3>
+<p>Tuple elements may be accessed directly. For example, with our declaration of <code>abc</code> from the last section, Stan uses <code>abc.1</code> for the first element, <code>abc.2</code> for the second, and <code>abc.3</code> for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that <code>abc.1</code> is of type <code>int</code>, <code>abc.2</code> of type <code>vector[3]</code> and <code>abc.3</code> of type <code>complex</code>.</p>
+</section>
+<section id="assigning-tuple-elements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="assigning-tuple-elements">Assigning tuple elements</h3>
+<p>Tuple elements can be assigned individually, allowing, e.g.,</p>
+<div class="sourceCode" id="cb82"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb82-1"><a href="#cb82-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">real</span>) ab;</span>
+<span id="cb82-2"><a href="#cb82-2" aria-hidden="true" tabindex="-1"></a>ab<span class="fl">.1</span> = <span class="dv">123</span>;</span>
+<span id="cb82-3"><a href="#cb82-3" aria-hidden="true" tabindex="-1"></a>ab<span class="fl">.2</span> = <span class="fl">12.9</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As with other assignments, promotions will happen if necessary (of <code>int</code> to <code>real</code> and of <code>real</code> to <code>complex</code>, along with the corresponding container type promotions).</p>
+</section>
+<section id="unpacking-assignment-of-tuples" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unpacking-assignment-of-tuples">Unpacking assignment of tuples</h3>
+<p>For convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.</p>
+<p>Given a tuple <code>t</code> of type <code>tuple(T1, ..., Tn)</code> and a sequence of <a href="#rhs-expressions">assignable expressions</a> of types <code>v1</code>, …, <code>vn</code>, where each <code>vi</code> has a <a href="../reference-manual/statements.html#promotion">type which is assignable</a> from type <code>Ti</code>, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement</p>
+<div class="sourceCode" id="cb83"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb83-1"><a href="#cb83-1" aria-hidden="true" tabindex="-1"></a>(v1, <span class="co">/*...*/</span>, vn) = t;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Note that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).</p>
+<p>These unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.</p>
+<p>For example, if <code>T</code> is a tuple of type <code>tuple(int, (real, real), complex)</code>, then the program</p>
+<div class="sourceCode" id="cb84"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb84-1"><a href="#cb84-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> i;</span>
+<span id="cb84-2"><a href="#cb84-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x, y;</span>
+<span id="cb84-3"><a href="#cb84-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z;</span>
+<span id="cb84-4"><a href="#cb84-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb84-5"><a href="#cb84-5" aria-hidden="true" tabindex="-1"></a>(i, (x, y), z) = T;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Assigns the result of <code>T.1</code> to <code>i</code>, the result of <code>T.2.1</code> to <code>x</code>, the result of <code>T.2.2</code> to <code>y</code>, and the result of <code>T.3</code> to <code>z</code>.</p>
+<p>The left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.</p>
+</section>
+</section>
+<section id="variable-types-vs.-constraints-and-sizes" class="level2">
+<h2 class="anchored" data-anchor-id="variable-types-vs.-constraints-and-sizes">Variable types vs.&nbsp;constraints and sizes</h2>
+<p>The type information associated with a variable only contains the underlying type and dimensionality of the variable.</p>
+<section id="type-information-excludes-sizes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-information-excludes-sizes">Type information excludes sizes</h3>
+<p>The size associated with a given variable is not part of its data type. For example, declaring a variable using</p>
+<div class="sourceCode" id="cb85"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb85-1"><a href="#cb85-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span> a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares the variable <code>a</code> to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.</p>
+<section id="when-are-sizes-checked" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="when-are-sizes-checked">When are sizes checked?</h4>
+<p>Sizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an <span class="math inline">\(N \times M\)</span> by a <span class="math inline">\(J \times K\)</span> matrix will raise a run-time error if <span class="math inline">\(M \neq J\)</span>.</p>
+</section>
+</section>
+<section id="type-information-excludes-constraints" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-information-excludes-constraints">Type information excludes constraints</h3>
+<p>Like sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.</p>
+<p>Similarly, a bounded real may be assigned to an unconstrained real and vice-versa.</p>
+<section id="when-are-function-argument-constraints-checked" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="when-are-function-argument-constraints-checked">When are function argument constraints checked?</h4>
+<p>For arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.</p>
+</section>
+<section id="when-are-declared-variable-constraints-checked" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="when-are-declared-variable-constraints-checked">When are declared variable constraints checked?</h4>
+<p>For data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.</p>
+<p>For parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.</p>
+<p>For all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to <code>NaN</code> and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).</p>
+<p>For generated quantities, constraints are enforced after the statements in the generated quantities block have executed.</p>
+</section>
+</section>
+<section id="type-naming-notation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-naming-notation">Type naming notation</h3>
+<p>In order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.</p>
+<p>Because size information is not part of a data type, data types will be written without size information. For instance, <code>array[] real</code> is the type of one-dimensional array of reals and <code>matrix</code> is the type of matrices. The three-dimensional integer array type is written as <code>array[,,] int</code>, indicating the number slots available for indexing. Similarly, <code>array[,] vector</code> is the type of a two-dimensional array of vectors.</p>
+</section>
+</section>
+<section id="variable-declaration.section" class="level2">
+<h2 class="anchored" data-anchor-id="variable-declaration.section">Variable declaration</h2>
+<p>Variables in Stan are declared by giving a type and a name. For example</p>
+<div class="sourceCode" id="cb86"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb86-1"><a href="#cb86-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N;</span>
+<span id="cb86-2"><a href="#cb86-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] y;</span>
+<span id="cb86-3"><a href="#cb86-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] A;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares a variable <code>N</code> that is an integer, a variable <code>y</code> that is a vector of length <code>N</code> (the previously declared variable), and a variable <code>A</code>, which is a length-5 array where each element is a 3 by 4 matrix.</p>
+<p>The size of top-level variables in the <code>parameters</code>, <code>transformed parameters</code>, and <code>generated quantities</code> must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.</p>
+<div class="sourceCode" id="cb87"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb87-1"><a href="#cb87-1" aria-hidden="true" tabindex="-1"></a><span class="co">// illegal and will be flagged by the compiler:</span></span>
+<span id="cb87-2"><a href="#cb87-2" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb87-3"><a href="#cb87-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N = <span class="dv">10</span>;</span>
+<span id="cb87-4"><a href="#cb87-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> foo;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Depending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.</p>
+<div class="sourceCode" id="cb88"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb88-1"><a href="#cb88-1" aria-hidden="true" tabindex="-1"></a><span class="co">// valid block variables, but not locals or function parameters</span></span>
+<span id="cb88-2"><a href="#cb88-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N] u;</span>
+<span id="cb88-3"><a href="#cb88-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb88-4"><a href="#cb88-4" aria-hidden="true" tabindex="-1"></a><span class="co">// valid as a block or local variable, but not a function parameter</span></span>
+<span id="cb88-5"><a href="#cb88-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> is;</span>
+<span id="cb88-6"><a href="#cb88-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb88-7"><a href="#cb88-7" aria-hidden="true" tabindex="-1"></a><span class="co">// function parameters exclude sizes and cannot be constrained</span></span>
+<span id="cb88-8"><a href="#cb88-8" aria-hidden="true" tabindex="-1"></a><span class="dt">void</span> pretty_print_tri_lower(<span class="dt">matrix</span> x) { ... }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Top-level variables can have <a href="#constrained-data-types">constraints</a> and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as <a href="../reference-manual/user-functions.html#argument-types-and-qualifiers">function parameters</a> are not constrained types and <em>exclude</em> sizes.</p>
+<p>In the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables <code>M</code> and <code>N</code> indicate number of columns and rows, respectively. The variable <code>K</code> is used for square matrices, i.e., <code>K</code> denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types <code>cov_matrix[K]</code>, <code>corr_matrix[K]</code>, <code>cholesky_factor_cov[K]</code>, and <code>cholesky_factor_corr[K]</code> are only assignable to matrices of dimensions <code>matrix[K, K]</code> types.</p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 22%">
+<col style="width: 27%">
+<col style="width: 48%">
+</colgroup>
+<thead>
+<tr class="header">
+<th style="text-align: left;">Function Argument (unsized)</th>
+<th style="text-align: left;"><pre><code>   Local</code></pre>
+<p>(unconstrained)</p></th>
+<th style="text-align: left;"><pre><code>              Block
+          (constrained)</code></pre></th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><code>int</code></td>
+<td style="text-align: left;"><code>int</code></td>
+<td style="text-align: left;"><code>int</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>int&lt;lower=L&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>int&lt;upper=U&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>int&lt;lower=L, upper=U&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>int&lt;offset=O&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>int&lt;multiplier=M&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>int&lt;offset=O, multiplier=M&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>real</code></td>
+<td style="text-align: left;"><code>real</code></td>
+<td style="text-align: left;"><code>real</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>real&lt;lower=L&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>real&lt;upper=U&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>real&lt;lower=L, upper=U&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>real&lt;offset=O&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>real&lt;multiplier=M&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>real&lt;offset=O, multiplier=M&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex</code></td>
+<td style="text-align: left;"><code>complex</code></td>
+<td style="text-align: left;"><code>complex</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>vector</code></td>
+<td style="text-align: left;"><code>vector[N]</code></td>
+<td style="text-align: left;"><code>vector[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>vector[N]&lt;lower=L&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>vector[N]&lt;upper=U&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>vector[N]&lt;lower=L, upper=U&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>vector[N]&lt;offset=O&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>vector[N]&lt;multiplier=M&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>vector[N]&lt;offset=O, multiplier=M&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>ordered[N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>positive_ordered[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>simplex[N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>unit_vector[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>sum_to_zero_vector[N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>row_vector</code></td>
+<td style="text-align: left;"><code>row_vector[N]</code></td>
+<td style="text-align: left;"><code>row_vector[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_vector[N]&lt;lower=L&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_vector[N]&lt;upper=U&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_vector[N]&lt;lower=L, upper=U&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_vector[N]&lt;offset=O&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_vector[N]&lt;multiplier=M&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_vector[N]&lt;offset=O, multiplier=M&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>matrix</code></td>
+<td style="text-align: left;"><code>matrix[M, N]</code></td>
+<td style="text-align: left;"><code>matrix[M, N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[M, N]&lt;lower=L&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[M, N]&lt;upper=U&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[M, N]&lt;lower=L, upper=U&gt;</code> |</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[M, N]&lt;offset=O&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[M, N]&lt;multiplier=M&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[M, N]&lt;offset=O, multiplier=M&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>column_stochastic_matrix[M, N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>row_stochastic_matrix[M, N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>sum_to_zero_matrix[M, N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[K, K]</code></td>
+<td style="text-align: left;"><code>corr_matrix[K]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[K, K]</code></td>
+<td style="text-align: left;"><code>cov_matrix[K]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[K, K]</code></td>
+<td style="text-align: left;"><code>cholesky_factor_corr[K]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>matrix[K, K]</code></td>
+<td style="text-align: left;"><code>cholesky_factor_cov[K]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex_vector</code></td>
+<td style="text-align: left;"><code>complex_vector[M]</code></td>
+<td style="text-align: left;"><code>complex_vector[M]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>complex_row_vector</code></td>
+<td style="text-align: left;"><code>complex_row_vector[N]</code></td>
+<td style="text-align: left;"><code>complex_row_vector[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><code>complex_matrix</code></td>
+<td style="text-align: left;"><code>complex_matrix[M, N]</code></td>
+<td style="text-align: left;"><code>complex_matrix[M,N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><code>array[] vector</code></td>
+<td style="text-align: left;"><code>array[M] vector[N]</code></td>
+<td style="text-align: left;"><code>array[M] vector[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] vector[N]&lt;lower=L&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] vector[N]&lt;upper=U&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] vector[N]&lt;lower=L, upper=U&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] vector[N]&lt;offset=O&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] vector[N]&lt;multiplier=M&gt;</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] vector[N]&lt;offset=O, multiplier=M&gt;</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] ordered[N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] positive_ordered[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] simplex[N]</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] unit_vector[N]</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"></td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><code>array[M] sum_to_zero_vector[N]</code></td>
+</tr>
+</tbody>
+</table>
+<p><a name="id:constrained-types.figure"></a></p>
+<p>Additional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g.&nbsp;<code>array[ , ]</code> is a 2-D array.</p>
+<p>For more on how function arguments and return types are declared, consult the <a href="https://mc-stan.org/docs/stan-users-guide/user-functions.html#basic-functions.section#type-declarations-for-functions">User’s Guide chapter on functions</a>.</p>
+</section>
+<section id="compound-variable-declaration-and-definition" class="level2">
+<h2 class="anchored" data-anchor-id="compound-variable-declaration-and-definition">Compound variable declaration and definition</h2>
+<p>Stan allows assignable variables to be declared and defined in a single statement. Assignable variables are</p>
+<ul>
+<li>local variables, and</li>
+<li>variables declared in the transformed data, transformed parameters, or generated quantities blocks.</li>
+</ul>
+<p>For example, the statement</p>
+<div class="sourceCode" id="cb91"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb91-1"><a href="#cb91-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N = <span class="dv">5</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares the variable <code>N</code> to be an integer scalar type and at the same time defines it to be the value of the expression <code>5</code>.</p>
+<section id="assignment-typing" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="assignment-typing">Assignment typing</h3>
+<p>The type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have</p>
+<div class="sourceCode" id="cb92"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb92-1"><a href="#cb92-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum = <span class="dv">0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>even though <code>0</code> is of type <code>int</code> and <code>sum</code> is of type <code>real</code>, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.</p>
+<p>Variables of any type may have values assigned to them. For example,</p>
+<div class="sourceCode" id="cb93"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb93-1"><a href="#cb93-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] a = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares a <span class="math inline">\(3 \times 2\)</span> matrix variable <code>a</code> and assigns a copy of the value of <code>b</code> to the variable <code>a</code>. The variable <code>b</code> must be of type <code>matrix</code> for the statement to be well formed. For the code to execute successfully, <code>b</code> must be the same shape as <code>a</code>, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to <code>a</code> do not affect <code>b</code> and subsequent changes to <code>b</code> do not affect <code>a</code>.</p>
+</section>
+<section id="rhs-expressions" class="level3">
+<h3 class="anchored" data-anchor-id="rhs-expressions">Right-hand side expressions</h3>
+<p>The right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,</p>
+<div class="sourceCode" id="cb94"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb94-1"><a href="#cb94-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] a = <span class="fl">0.5</span> * (b + c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>assigns the matrix variable <code>a</code> to half of the sum of <code>b</code> and <code>c</code>. The only requirement on <code>b</code> and <code>c</code> is that the expression <code>b + c</code> be of type <code>matrix</code>. For example, <code>b</code> could be of type <code>matrix</code> and <code>c</code> of type <code>real</code>, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.</p>
+<p>Similarly,</p>
+<div class="sourceCode" id="cb95"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb95-1"><a href="#cb95-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = <span class="dv">2</span> + <span class="fl">3i</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>assigns the the complex number <span class="math inline">\(2 + 3i\)</span> to the complex scalar <code>z</code>. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).</p>
+</section>
+<section id="scope-within-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="scope-within-expressions">Scope within expressions</h3>
+<p>Any variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.</p>
+</section>
+</section>
+<section id="declaring-multiple-variables-at-once" class="level2">
+<h2 class="anchored" data-anchor-id="declaring-multiple-variables-at-once">Declaring multiple variables at once</h2>
+<p>Stan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.</p>
+<section id="types-for-multiple-declarations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="types-for-multiple-declarations">Types for multiple declarations</h3>
+<p>The code:</p>
+<div class="sourceCode" id="cb96"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb96-1"><a href="#cb96-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x, y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is equivalent to</p>
+<div class="sourceCode" id="cb97"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb97-1"><a href="#cb97-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x;</span>
+<span id="cb97-2"><a href="#cb97-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As a result, all declarations on the same line must be of the same type.</p>
+</section>
+<section id="combining-with-other-features" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="combining-with-other-features">Combining with other features</h3>
+<p>The ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in <a href="#compound-variable-declaration-and-definition">the section introducing compound declarations and definitions</a> :</p>
+<div class="sourceCode" id="cb98"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb98-1"><a href="#cb98-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="dv">3</span>, y = <span class="fl">5.6</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><a href="#constrained-data-types">Constrained data types</a> can also be declared together, so long as the constraint for each variable is the same:</p>
+<div class="sourceCode" id="cb99"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb99-1"><a href="#cb99-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; x, y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Stan compiles integers to <code>int</code> and reals to <code>double</code> types in C++. Precise details of rounding will depend on the compiler and hardware architecture on which the code is run.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/whitespace.html" class="pagination-link" aria-label="Whitespace">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Whitespace</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/expressions.html" class="pagination-link" aria-label="Expressions">
+        <span class="nav-page-text">Expressions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/types.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/user-functions.html b/docs/2_39/reference-manual/user-functions.html
new file mode 100644
index 000000000..2c1e2cbc5
--- /dev/null
+++ b/docs/2_39/reference-manual/user-functions.html
@@ -0,0 +1,1351 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>User-Defined Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/transforms.html" rel="next">
+<link href="../reference-manual/blocks.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="User-Defined Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/user-functions.html">User-Defined Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#functions.chapter" id="toc-functions.chapter" class="nav-link active" data-scroll-target="#functions.chapter">User-Defined Functions</a>
+  <ul class="collapse">
+  <li><a href="#function-definition-block" id="toc-function-definition-block" class="nav-link" data-scroll-target="#function-definition-block">Function-definition block</a></li>
+  <li><a href="#function-names" id="toc-function-names" class="nav-link" data-scroll-target="#function-names">Function names</a>
+  <ul class="collapse">
+  <li><a href="#function-overloading" id="toc-function-overloading" class="nav-link" data-scroll-target="#function-overloading">Function overloading</a></li>
+  </ul></li>
+  <li><a href="#calling-functions" id="toc-calling-functions" class="nav-link" data-scroll-target="#calling-functions">Calling functions</a>
+  <ul class="collapse">
+  <li><a href="#functions-as-expressions" id="toc-functions-as-expressions" class="nav-link" data-scroll-target="#functions-as-expressions">Functions as expressions</a></li>
+  <li><a href="#functions-as-statements" id="toc-functions-as-statements" class="nav-link" data-scroll-target="#functions-as-statements">Functions as statements</a></li>
+  <li><a href="#resolving-overloads" id="toc-resolving-overloads" class="nav-link" data-scroll-target="#resolving-overloads">Resolving overloads</a></li>
+  <li><a href="#argument-promotion" id="toc-argument-promotion" class="nav-link" data-scroll-target="#argument-promotion">Argument promotion</a></li>
+  <li><a href="#probability-functions-in-distribution-statements" id="toc-probability-functions-in-distribution-statements" class="nav-link" data-scroll-target="#probability-functions-in-distribution-statements">Probability functions in distribution statements</a></li>
+  <li><a href="#restrictions-on-placement" id="toc-restrictions-on-placement" class="nav-link" data-scroll-target="#restrictions-on-placement">Restrictions on placement</a></li>
+  </ul></li>
+  <li><a href="#argument-types-and-qualifiers" id="toc-argument-types-and-qualifiers" class="nav-link" data-scroll-target="#argument-types-and-qualifiers">Argument types and qualifiers</a>
+  <ul class="collapse">
+  <li><a href="#base-variable-type-declaration" id="toc-base-variable-type-declaration" class="nav-link" data-scroll-target="#base-variable-type-declaration">Base variable type declaration</a></li>
+  <li><a href="#dimensionality-declaration" id="toc-dimensionality-declaration" class="nav-link" data-scroll-target="#dimensionality-declaration">Dimensionality declaration</a></li>
+  <li><a href="#dimensionality-checks-and-exceptions" id="toc-dimensionality-checks-and-exceptions" class="nav-link" data-scroll-target="#dimensionality-checks-and-exceptions">Dimensionality checks and exceptions</a></li>
+  <li><a href="#data-only-qualifiers" id="toc-data-only-qualifiers" class="nav-link" data-scroll-target="#data-only-qualifiers">Data-only qualifiers</a></li>
+  </ul></li>
+  <li><a href="#function-bodies.section" id="toc-function-bodies.section" class="nav-link" data-scroll-target="#function-bodies.section">Function bodies</a>
+  <ul class="collapse">
+  <li><a href="#random-number-generating-functions" id="toc-random-number-generating-functions" class="nav-link" data-scroll-target="#random-number-generating-functions">Random number generating functions</a></li>
+  <li><a href="#log-probability-access-in-functions" id="toc-log-probability-access-in-functions" class="nav-link" data-scroll-target="#log-probability-access-in-functions">Log probability access in functions</a></li>
+  <li><a href="#defining-probability-functions-for-distribution-statements" id="toc-defining-probability-functions-for-distribution-statements" class="nav-link" data-scroll-target="#defining-probability-functions-for-distribution-statements">Defining probability functions for distribution statements</a></li>
+  </ul></li>
+  <li><a href="#parameters-are-constant" id="toc-parameters-are-constant" class="nav-link" data-scroll-target="#parameters-are-constant">Parameters are constant</a></li>
+  <li><a href="#return-value" id="toc-return-value" class="nav-link" data-scroll-target="#return-value">Return value</a>
+  <ul class="collapse">
+  <li><a href="#return-guarantee-required" id="toc-return-guarantee-required" class="nav-link" data-scroll-target="#return-guarantee-required">Return guarantee required</a></li>
+  </ul></li>
+  <li><a href="#void-functions-as-statements" id="toc-void-functions-as-statements" class="nav-link" data-scroll-target="#void-functions-as-statements">Void Functions as Statements</a>
+  <ul class="collapse">
+  <li><a href="#void-functions" id="toc-void-functions" class="nav-link" data-scroll-target="#void-functions">Void functions</a></li>
+  <li><a href="#usage-as-statement" id="toc-usage-as-statement" class="nav-link" data-scroll-target="#usage-as-statement">Usage as statement</a></li>
+  <li><a href="#special-return-statements" id="toc-special-return-statements" class="nav-link" data-scroll-target="#special-return-statements">Special return statements</a></li>
+  </ul></li>
+  <li><a href="#forward-declarations.section" id="toc-forward-declarations.section" class="nav-link" data-scroll-target="#forward-declarations.section">Declarations</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/user-functions.html">User-Defined Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="functions.chapter" class="level1">
+<h1>User-Defined Functions</h1>
+<p>Stan allows users to define their own functions. The basic syntax is a simplified version of that used in C and C++. This chapter specifies how functions are declared, defined, and used in Stan.</p>
+<section id="function-definition-block" class="level2">
+<h2 class="anchored" data-anchor-id="function-definition-block">Function-definition block</h2>
+<p>User-defined functions appear in a special function-definition block before all of the other program blocks.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... function declarations and definitions ...</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Function definitions and declarations may appear in any order. <a href="#forward-declarations.section">Forward declarations</a> are allowed but not required.</p>
+</section>
+<section id="function-names" class="level2">
+<h2 class="anchored" data-anchor-id="function-names">Function names</h2>
+<p>The rules for function naming and function-argument naming are the same as for other variables; see the <a href="#variables.section">section on variables</a> for more information on valid identifiers. For example,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> mu, <span class="dt">real</span> sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares a function named <code>foo</code> with two argument variables of types <code>real</code> and <code>real</code>. The arguments are named <code>mu</code> and <code>sigma</code>, but that is not part of the declaration.</p>
+<section id="function-overloading" class="level3">
+<h3 class="anchored" data-anchor-id="function-overloading">Function overloading</h3>
+<p>Multiple user-defined functions may have the same name if they have different sequences of <a href="#argument-types-and-qualifiers">argument types</a>. This is known as function overloading.</p>
+<p>For example, the following two functions are both defined with the name <code>add_up</code></p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> add_up(<span class="dt">real</span> a, <span class="dt">real</span> b){</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> a + b;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> add_up(<span class="dt">real</span> a, <span class="dt">real</span> b, <span class="dt">real</span> c){</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> a + b + c;</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The return types of overloaded functions do not need to be the same. One could define an additional <code>add_up</code> function as follows</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> add_up(<span class="dt">int</span> a, <span class="dt">int</span> b){</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> a + b;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>That being said, functions may <strong>not</strong> use the same name if their signature <em>only</em> differs by the return type.</p>
+<p>For example, the following is not permitted</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="co">// illegal</span></span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> baz(<span class="dt">int</span> x);</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> baz(<span class="dt">int</span> x);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Function names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the <code>reduce_sum</code> family of functions and ODE integrators, which cannot be overloaded.</p>
+</section>
+</section>
+<section id="calling-functions" class="level2">
+<h2 class="anchored" data-anchor-id="calling-functions">Calling functions</h2>
+<p>All function arguments are mandatory—there are no default values.</p>
+<section id="functions-as-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="functions-as-expressions">Functions as expressions</h3>
+<p>Functions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.</p>
+</section>
+<section id="functions-as-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="functions-as-statements">Functions as statements</h3>
+<p>Functions with void return types may be applied to arguments and used as <a href="../reference-manual/statements.html">statements</a>. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.</p>
+</section>
+<section id="resolving-overloads" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="resolving-overloads">Resolving overloads</h3>
+<p><a href="#function-overloading">Overloaded functions</a> alongside <a href="#promotion">type promotion</a> can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.</p>
+<p>Consider the following two overloaded functions</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">int</span> a, <span class="dt">real</span> b);</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> a, <span class="dt">int</span> b);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These functions do <strong>not</strong> have a unique minimum when called with two integer arguments <code>foo(1,2)</code>, and therefore cannot be called as such.</p>
+<p>Promotion of integers to complex numbers is considered as two separate promotions, one from <code>int</code> to <code>real</code> and a second from <code>real</code> to <code>complex</code>. Consider the following functions with <code>real</code> and <code>complex</code> signatures</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> bar(<span class="dt">real</span> x);</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> bar(<span class="dt">complex</span> z);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A call <code>bar(5)</code> with an integer argument will be resolved to <code>bar(real)</code> because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.</p>
+</section>
+<section id="argument-promotion" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="argument-promotion">Argument promotion</h3>
+<p>The rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> x) { <span class="cf">return</span> ... };</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>...</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> a = <span class="dv">5</span>;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> b = foo(a); <span class="co">// a promoted to type real</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In addition to promoting <code>int</code> to <code>real</code>, Stan also promotes <code>real</code> to <code>complex</code>, and by transitivity, <code>int</code> to <code>complex</code>. This also works for containers, so an array of <code>int</code> may be assigned to an array of <code>real</code> of the same shape. And we can also promote <code>vector</code> to <code>complex_vector</code> and similarly for row vectors and matrices.</p>
+</section>
+<section id="probability-functions-in-distribution-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="probability-functions-in-distribution-statements">Probability functions in distribution statements</h3>
+<p>Functions whose name ends in <code>_lpdf</code> or <code>_lpmf</code> (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of <a href="distribution statements">statements.qmd#distribution-statements.section</a>.</p>
+</section>
+<section id="restrictions-on-placement" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="restrictions-on-placement">Restrictions on placement</h3>
+<p>Functions of certain types are restricted on scope of usage. Functions whose names end in <code>_lp</code> assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.</p>
+<p>Functions whose name end in <code>_jacobian</code> assume access to the log probability accumulator may only be used within the transformed parameters block.</p>
+<p>Functions whose names end in <code>_rng</code> assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in <code>_rng</code>.</p>
+<p>Functions whose names end in <code>_lpdf</code> and <code>_lpmf</code> can be used anywhere. However, <code>_lupdf</code> and <code>_lupmf</code> functions can only be used in the model block or user-defined probability functions.</p>
+<p>See the <a href="#function-bodies.section">section on function bodies</a> for more information on these special types of function.</p>
+</section>
+</section>
+<section id="argument-types-and-qualifiers" class="level2">
+<h2 class="anchored" data-anchor-id="argument-types-and-qualifiers">Argument types and qualifiers</h2>
+<p>Stan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.</p>
+<p>The type <code>void</code> may not be used as an argument type, only a return type for a function with side effects.</p>
+<section id="base-variable-type-declaration" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="base-variable-type-declaration">Base variable type declaration</h3>
+<p>The base variable types are <code>integer</code>, <code>real</code>, <code>complex</code>, <code>vector</code>, <code>row_vector</code>, and <code>matrix</code>. No lower-bound or upper-bound constraints are allowed (e.g., <code>real&lt;lower=0&gt;</code> is illegal). Specialized constrained types are also not allowed (e.g., <code>simplex</code> is illegal).</p>
+<p>Tuple types of the form <code>tuple(T1, ..., TN)</code> are also allowed, with all of the types <code>T1</code> to <code>TN</code> being function argument types (i.e., no constraints and no sizes).</p>
+</section>
+<section id="dimensionality-declaration" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="dimensionality-declaration">Dimensionality declaration</h3>
+<p>Arguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, <code>int</code> denotes a single integer argument or return, whereas <code>array[] real</code> indicates a one-dimensional array of reals, <code>array[,] real</code> a two-dimensional array and <code>array[,,] real</code> a three-dimensional array; whitespace is optional, as usual.</p>
+<p>The dimensions for vectors and matrices are not included, so that <code>matrix</code> is the type of a single matrix argument or return type. Thus if a variable is declared as <code>matrix a</code>, then <code>a</code> has two indexing dimensions, so that <code>a[1]</code> is a row vector and <code>a[1, 1]</code> a real value. Matrices implicitly have two indexing dimensions. The type declaration <code>matrix[ , ] b</code> specifies that <code>b</code> is a two-dimensional array of matrices, for a total of four indexing dimensions, with <code>b[1, 1, 1, 1]</code> picking out a real value.</p>
+</section>
+<section id="dimensionality-checks-and-exceptions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="dimensionality-checks-and-exceptions">Dimensionality checks and exceptions</h3>
+<p>Function argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.</p>
+<p>Dimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.</p>
+<p>Any errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.</p>
+</section>
+<section id="data-only-qualifiers" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-only-qualifiers">Data-only qualifiers</h3>
+<p>Some of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.</p>
+<p>In user-defined functions, the qualifier <code>data</code> may be placed before an argument type declaration to indicate that the argument must be data only. For example,</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="kw">data</span> <span class="dt">real</span> x) {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> x^<span class="dv">2</span>;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>requires the argument <code>x</code> to be data only.</p>
+<p>Declaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.</p>
+</section>
+</section>
+<section id="function-bodies.section" class="level2">
+<h2 class="anchored" data-anchor-id="function-bodies.section">Function bodies</h2>
+<p>The body of a function is between an open curly brace (<code>{</code>) and close curly brace (<code>}</code>). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.</p>
+<p>Any user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.</p>
+<p>The only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.</p>
+<section id="random-number-generating-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="random-number-generating-functions">Random number generating functions</h3>
+<p>Functions that call random number generating functions in their bodies must have a name that ends in <code>_rng</code>; attempts to use random-number generators in other functions lead to a compile-time error.</p>
+<p>Like other random number generating functions, user-defined functions with names that end in <code>_rng</code> may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in <code>_rng</code>. An attempt to use such a function elsewhere results in a compile-time error.</p>
+</section>
+<section id="log-probability-access-in-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="log-probability-access-in-functions">Log probability access in functions</h3>
+<p>Functions that include <a href="../reference-manual/statements.html#distribution-statements.section">distribution statements</a> or <a href="../reference-manual/statements.html#increment-log-prob.section">log probability increment statements</a> must have a name that ends in <code>_lp</code>. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.</p>
+<p>Like the target log density increment statement and distribution statements, user-defined functions with names that end in <code>_lp</code> may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.</p>
+</section>
+<section id="defining-probability-functions-for-distribution-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="defining-probability-functions-for-distribution-statements">Defining probability functions for distribution statements</h3>
+<p>Functions whose names end in <code>_lpdf</code> and <code>_lpmf</code> (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (<code>~</code>) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of <code>y</code> given parameter <code>theta</code> allows the use of the distribution statement is defined as follows.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo_lpdf(<span class="dt">real</span> y, <span class="dt">vector</span> theta) { ... }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Note that for function definitions, the comma is used rather than the vertical bar.</p>
+<p>For every custom <code>_lpdf</code> and <code>_lpmf</code> defined there is a corresponding <code>_lupdf</code> and <code>_lupmf</code> defined automatically. The <code>_lupdf</code> and <code>_lupmf</code> versions of the functions cannot be defined directly (to do so will produce an error). The difference in the <code>_lpdf</code> and <code>_lpmf</code> and the corresponding <code>_lupdf</code> and <code>_lupmf</code> functions is that if any other unnormalized density functions are used inside the user-defined function, the <code>_lpdf</code> and <code>_lpmf</code> forms of the user-defined function will change these densities to be normalized. The <code>_lupdf</code> and <code>_lupmf</code> forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.</p>
+<p>The distribution statement shorthand</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a>z ~ foo(phi);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>will have the same effect as incrementing the target with the log of the unnormalized density:</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> foo_lupdf(z | phi);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Other <code>_lupdf</code> and <code>_lupmf</code> functions used in the definition of <code>foo_lpdf</code> will drop additive constants when <code>foo_lupdf</code> is called and will not drop additive constants when <code>foo_lpdf</code> is called.</p>
+<p>If there are <code>_lupdf</code> and <code>_lupmf</code> functions used inside the following call to <code>foo_lpdf</code>, they will be forced to normalize (return the equivalent of their <code>_lpdf</code> and <code>_lpmf</code> forms):</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> foo_lpdf(z | phi);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If there are no <code>_lupdf</code> or <code>_lupmf</code> functions used in the definition of <code>foo_lpdf</code>, then there will be no difference between a <code>foo_lpdf</code> or <code>foo_lupdf</code> call.</p>
+<p>The unnormalized <code>_lupdf</code> and <code>_lupmf</code> functions can only be used in the model block or in user-defined probability functions (those ending in <code>_lpdf</code> or <code>_lpmf</code>).</p>
+<p>The same syntax and shorthand that works for <code>_lpdf</code> also works for log probability mass functions with suffixes <code>_lpmf</code>.</p>
+<p>A function that is going to be accessed as distributions must return the log of the density or mass function it defines.</p>
+</section>
+</section>
+<section id="parameters-are-constant" class="level2">
+<h2 class="anchored" data-anchor-id="parameters-are-constant">Parameters are constant</h2>
+<p>Within function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (<code>=</code>) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>Local variables may be declared at the top of the function block and scope as usual.</p>
+</section>
+<section id="return-value" class="level2">
+<h2 class="anchored" data-anchor-id="return-value">Return value</h2>
+<p>Non-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.</p>
+<p>Void functions may use <code>return</code> only without an argument, but return statements are not mandatory.</p>
+<section id="return-guarantee-required" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="return-guarantee-required">Return guarantee required</h3>
+<p>Unlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that</p>
+<ul>
+<li>a return statement qualifies,</li>
+</ul>
+<p>and the recursive cases are that</p>
+<ul>
+<li>a sequence of statements qualifies if its last statement qualifies,</li>
+<li>a for loop or while loop qualifies if its body qualifies, and</li>
+<li>a conditional statement qualifies if it has a default else clause and all of its body statements qualify.</li>
+</ul>
+<p>An exception is made for “obviously infinite” loops like <code>while (1)</code>, which contain a <code>return</code> statement and no <code>break</code> statements. The only way to exit such a loop is to return, so they are considered as returning statements.</p>
+<p>These rules disqualify</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> x) {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (x &gt; <span class="dv">2</span>) {</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> <span class="fl">1.0</span>;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> <span class="cf">if</span> (x &lt;= <span class="dv">2</span>) {</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> -<span class="fl">1.0</span>;</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>because there is no default <code>else</code> clause, and disqualify</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> x) {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  y = x;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">while</span> (x &lt; <span class="dv">10</span>) {</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (x &gt; <span class="dv">0</span>) {</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>      <span class="cf">return</span> x;</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>    y = x / <span class="dv">2</span>;</span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>because the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> log_fancy(<span class="dt">real</span> x) {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (x &lt; <span class="fl">1e-30</span>) {</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> x;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> <span class="cf">if</span> (x &lt; <span class="fl">1e-14</span>) {</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> x * x;</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> log(x);</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>because there’s a default else clause and each condition body has return as its final statement.</p>
+</section>
+</section>
+<section id="void-functions-as-statements" class="level2">
+<h2 class="anchored" data-anchor-id="void-functions-as-statements">Void Functions as Statements</h2>
+<section id="void-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="void-functions">Void functions</h3>
+<p>A function can be declared without a return value by using <code>void</code> in place of a return type. Note that the type <code>void</code> may only be used as a return type—arguments may not be declared to be of type <code>void</code>.</p>
+</section>
+<section id="usage-as-statement" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="usage-as-statement">Usage as statement</h3>
+<p>A void function may be used as a statement.</p>
+<p>Because there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.</p>
+</section>
+<section id="special-return-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="special-return-statements">Special return statements</h3>
+<p>In a return statement within a void function’s definition, the <code>return</code> keyword is followed immediately by a semicolon (<code>;</code>) rather than by the expression whose value is returned.</p>
+</section>
+</section>
+<section id="forward-declarations.section" class="level2">
+<h2 class="anchored" data-anchor-id="forward-declarations.section">Declarations</h2>
+<p>Stan supports forward declarations, which look like function definitions without bodies. For example,</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> unit_normal_lpdf(<span class="dt">real</span> y);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>declares a function named <code>unit_normal_lpdf</code> that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.</p>
+<p>A function definition with a body simultaneously declares and defines the named function, as in</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> unit_normal_lpdf(<span class="dt">real</span> y) {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> -<span class="fl">0.5</span> * square(y);</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the <a href="#function-overloading">section on function overloading</a>.</p>
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Despite being declared constant and appearing to have a pass-by-value syntax in Stan, the implementation of the language passes function arguments by constant reference in C++.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/blocks.html" class="pagination-link" aria-label="Program Blocks">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Program Blocks</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/transforms.html" class="pagination-link" aria-label="Constraint Transforms">
+        <span class="nav-page-text">Constraint Transforms</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/variational.html b/docs/2_39/reference-manual/variational.html
new file mode 100644
index 000000000..98e2fbaf2
--- /dev/null
+++ b/docs/2_39/reference-manual/variational.html
@@ -0,0 +1,1127 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Variational Inference</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/laplace.html" rel="next">
+<link href="../reference-manual/pathfinder.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Variational Inference">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/variational.html">Variational Inference</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#vi-algorithms.chapter" id="toc-vi-algorithms.chapter" class="nav-link active" data-scroll-target="#vi-algorithms.chapter">Variational Inference</a>
+  <ul class="collapse">
+  <li><a href="#stochastic-gradient-ascent" id="toc-stochastic-gradient-ascent" class="nav-link" data-scroll-target="#stochastic-gradient-ascent">Stochastic gradient ascent</a>
+  <ul class="collapse">
+  <li><a href="#monte-carlo-approximation-of-the-elbo" id="toc-monte-carlo-approximation-of-the-elbo" class="nav-link" data-scroll-target="#monte-carlo-approximation-of-the-elbo">Monte Carlo approximation of the ELBO</a></li>
+  <li><a href="#monte-carlo-approximation-of-the-gradients" id="toc-monte-carlo-approximation-of-the-gradients" class="nav-link" data-scroll-target="#monte-carlo-approximation-of-the-gradients">Monte Carlo approximation of the gradients</a></li>
+  <li><a href="#adaptive-stepsize-sequence" id="toc-adaptive-stepsize-sequence" class="nav-link" data-scroll-target="#adaptive-stepsize-sequence">Adaptive stepsize sequence</a></li>
+  <li><a href="#assessing-convergence" id="toc-assessing-convergence" class="nav-link" data-scroll-target="#assessing-convergence">Assessing convergence</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/variational.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/variational.html">Variational Inference</a></li></ol></nav></header>
+
+
+
+
+<section id="vi-algorithms.chapter" class="level1">
+<h1>Variational Inference</h1>
+<p>Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) <span class="citation" data-cites="kucukelbir:2017">Kucukelbir et al. (<a href="#ref-kucukelbir:2017" role="doc-biblioref">2017</a>)</span>. In this chapter, we describe the specifics of how ADVI maximizes the variational objective.</p>
+<section id="stochastic-gradient-ascent" class="level2">
+<h2 class="anchored" data-anchor-id="stochastic-gradient-ascent">Stochastic gradient ascent</h2>
+<p>ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.</p>
+<section id="monte-carlo-approximation-of-the-elbo" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="monte-carlo-approximation-of-the-elbo">Monte Carlo approximation of the ELBO</h3>
+<p>ADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by <code>elbo_samples</code>. We recommend a default value of <span class="math inline">\(100\)</span>, as we only evaluate the ELBO every <code>eval_elbo</code> iterations, which also defaults to <span class="math inline">\(100\)</span>.</p>
+</section>
+<section id="monte-carlo-approximation-of-the-gradients" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="monte-carlo-approximation-of-the-gradients">Monte Carlo approximation of the gradients</h3>
+<p>ADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by <code>grad_samples</code>. We recommend a default value of <span class="math inline">\(1\)</span>, as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.</p>
+</section>
+<section id="adaptive-stepsize-sequence" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="adaptive-stepsize-sequence">Adaptive stepsize sequence</h3>
+<p>ADVI uses a finite-memory version of adaGrad <span class="citation" data-cites="Duchi:2011">Duchi, Hazan, and Singer (<a href="#ref-Duchi:2011" role="doc-biblioref">2011</a>)</span>. This has a single parameter that we expose, denoted <code>eta</code>. We now have a warmup adaptation phase that selects a good value for <code>eta</code>. The procedure does a heuristic search over <code>eta</code> values that span 5 orders of magnitude.</p>
+</section>
+<section id="assessing-convergence" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="assessing-convergence">Assessing convergence</h3>
+<p>ADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by <code>tol_rel_obj</code>, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Duchi:2011" class="csl-entry" role="listitem">
+Duchi, John, Elad Hazan, and Yoram Singer. 2011. <span>“Adaptive Subgradient Methods for Online Learning and Stochastic Optimization.”</span> <em>The Journal of Machine Learning Research</em> 12: 2121–59.
+</div>
+<div id="ref-kucukelbir:2017" class="csl-entry" role="listitem">
+Kucukelbir, Alp, Dustin Tran, Rajesh Ranganath, Andrew Gelman, and David M Blei. 2017. <span>“Automatic Differentiation Variational Inference.”</span> <em>Journal of Machine Learning Research</em>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/pathfinder.html" class="pagination-link" aria-label="Pathfinder">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Pathfinder</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/laplace.html" class="pagination-link" aria-label="Laplace Approximation">
+        <span class="nav-page-text">Laplace Approximation</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/variational.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/reference-manual/whitespace.html b/docs/2_39/reference-manual/whitespace.html
new file mode 100644
index 000000000..4ba49ebf7
--- /dev/null
+++ b/docs/2_39/reference-manual/whitespace.html
@@ -0,0 +1,1058 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Whitespace</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/types.html" rel="next">
+<link href="../reference-manual/comments.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Whitespace">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/whitespace.html">Whitespace</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#whitespace" id="toc-whitespace" class="nav-link active" data-scroll-target="#whitespace">Whitespace</a>
+  <ul class="collapse">
+  <li><a href="#whitespace-characters" id="toc-whitespace-characters" class="nav-link" data-scroll-target="#whitespace-characters">Whitespace characters</a></li>
+  <li><a href="#whitespace-neutrality" id="toc-whitespace-neutrality" class="nav-link" data-scroll-target="#whitespace-neutrality">Whitespace neutrality</a></li>
+  <li><a href="#whitespace-location" id="toc-whitespace-location" class="nav-link" data-scroll-target="#whitespace-location">Whitespace location</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/whitespace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/encoding.html">Language</a></li><li class="breadcrumb-item"><a href="../reference-manual/whitespace.html">Whitespace</a></li></ol></nav></header>
+
+
+
+
+<section id="whitespace" class="level1">
+<h1>Whitespace</h1>
+<section id="whitespace-characters" class="level2">
+<h2 class="anchored" data-anchor-id="whitespace-characters">Whitespace characters</h2>
+<p>The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).</p>
+</section>
+<section id="whitespace-neutrality" class="level2">
+<h2 class="anchored" data-anchor-id="whitespace-neutrality">Whitespace neutrality</h2>
+<p>Stan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.</p>
+<p>Other than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.</p>
+</section>
+<section id="whitespace-location" class="level2">
+<h2 class="anchored" data-anchor-id="whitespace-location">Whitespace location</h2>
+<p>Zero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as <code>a * b</code>, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.</p>
+<p>Identifiers and literals may not be separated by whitespace. Thus it is not legal to write the number <code>10000</code> as <code>10 000</code> or to write the identifier <code>normal_lpdf</code> as <code>normal _ lpdf</code>.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/comments.html" class="pagination-link" aria-label="Comments">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Comments</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/types.html" class="pagination-link" aria-label="Data Types and Declarations">
+        <span class="nav-page-text">Data Types and Declarations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/whitespace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/search.json b/docs/2_39/search.json
new file mode 100644
index 000000000..4a90c7f3e
--- /dev/null
+++ b/docs/2_39/search.json
@@ -0,0 +1,9494 @@
+[
+  {
+    "objectID": "stan-users-guide/wiener_diffusion_model.html",
+    "href": "stan-users-guide/wiener_diffusion_model.html",
+    "title": "Wiener diffusion model",
+    "section": "",
+    "text": "Diffusion models, sometimes also called Wiener diffusion models, are among the most frequently used model families in modeling two-alternative forced-choice tasks (see Wagenmakers (2009), for a review). Diffusion models allow to model response times and responses jointly. The basic version of a diffusion model comprises four parameters: the boundary separation, \\(a\\), the relative starting point, \\(w\\), the drift rate, \\(v\\), and the non-decision time, \\(t0\\) (Ratcliff 1978). In the basic model, it is assumed that the four basic parameters are the same for the whole experiment. As this assumption is very strict and there are examples that suggest that the basic parameters can be different from trial to trial, so called inter-trial variabilities were introduced and the basic four- parameter model was extended to a seven-parameter model. In the seven-parameter extension of the diffusion model there are the following three parameters added: the inter-trial variability in relative starting point, \\(s_w\\), the inter-trial variability in drift rate, \\(s_v\\), and the inter-trial variability in non-decision time, \\(s_{t0}\\) Nicenboim, Schad, and Vasishth (2025).\nData for the diffusion model is two-dimensional: There is one vector for the reaction times, \\(y\\), and one vector for the given responses, \\(\\text{resp}\\). The reaction times shall be positive, continuous and in seconds, the responses shall be binary.\nAs a diffusion model describes the decision process for a decision with exactly two choices, there exist reaction time distributions for each response alternative. This means that the probability density function (\\(p\\)) splits into one part for one response alternative and one part for the other response alternative. In the following, we will refer to one alternative as the upper response boundary and to the other alternative as the lower response boundary. \\(p\\) of the lower response boundary can be obtained when inserting \\(-v\\) and \\(1-w\\) to \\(p\\) of the upper response boundary. Let’s call \\(p\\) for the lower response boundary \\(p_0\\) and \\(p\\) for the upper response boundary \\(p_1\\). Then:\n\\[\np_0(a,t0,v,w,sv,sw,st0) = p_1(a,t0,-v,1-w,sv,sw,st0)\n\\]\nUsually, a \\(PDF\\) integrates to 1. In the case of the diffusion model, only the sum of both parts, \\(p_0\\) and \\(p_1\\), integrates to 1. This is called defective.\n\n\n\n\n\n\nFigure 1: Figure 1: Realization of a Four-Parameter Diffusion Process Modeling the Binary Decision Process. Image from Henrich et al. (2024), distributed under the Creative Commons Attribution 4.0 International License. Note. The parameters are the boundary separation a for two response alternatives, the relative starting point w, the drift rate v, and the non-decision time t0. The decision process is illustrated as a jagged line between the two boundaries. The predicted distributions of the reaction times are depicted as curved lines below and above the response boundaries (blue).\n\n\n\nIn this model it is assumed that the decision process behaves like a random walk and we are interested in the first time that the random walk crosses one of the two decision boundaries. Hence, we are interested in the first-passage time of the decision process. The Stan function wiener_lpdf() returns the logarithm of the first-passage time density function for a diffusion model with up to seven parameters for upper boundary responses, \\(\\log(p_1)\\). As can be seen above, it suffices to implement the density for only one response boundary, as the other can be obtained by mirroring the starting point and drift rate. Any combination of fixed and estimated parameters can be specified. In other words, with this implementation it is not only possible to estimate parameters of the fullseven- parameter model, but also to estimate restricted models such as the basic four- parameter model, or a five or six-parameter model, or even a one-parameter model when fixing the other six parameters.\nFor example, it is possible to permit variability in just one or two parameters and to fix the other variabilities to 0, or even to estimate a three-parameter model when fixing more parameters (e.g., fixing the relative starting point at 0.5).\nIt is assumed that the reaction time data that correspond to the upper response boundary \\(y_\\text{upper}\\) is distributed according to wiener_lpdf():\n\\[\ny_\\text{lower} \\sim \\operatorname{wiener\\_lpdf}(a, t0, w, v, s_v, s_w, s_{t0})\n\\] and the reaction time data that correspond to the lower response boundary \\(y_\\text{lower}\\) is distributed according to wiener_lpdf() with mirrored starting point and drift rate:\n\\[\ny_\\text{upper} \\sim \\operatorname{wiener\\_lpdf}(a, t0, 1-w, -v, s_v, s_w, s_{t0})\n\\]\n\n\nThe following example demonstrates a diffusion model call in Stan:\ndata {\n  int &lt;lower=0&gt; N; // Number of trials\n  array[N] real rt; // response times (in seconds )\n  array[N] int &lt;lower=0, upper=1&gt; resp; // responses {0 ,1}\n}\ntransformed data{\n  real min_rt = min(rt);\n}\nparameters {\n  real &lt;lower=0&gt; a;                // boundary separation\n  real v;                          // drift\n  real &lt;lower=0, upper=1&gt; w;       // relative starting point\n  real &lt;lower=0, upper=min_rt&gt; t0; // non-decision time\n\n  real &lt;lower=0&gt; sv;               // variability in drift\n  // variability in starting point\n  real &lt;lower=0, upper=fmin(2 * w, 2 * (1 - w))&gt; sw; \n  real &lt;lower=0&gt; st0;             // variability in non-decision time\n}\ntransformed parameters{\n  real one_minus_w = 1 - w;\n  real neg_v = -v;\n}\nmodel {\n  // prior\n  a ~ normal(1, 1);\n  w ~ normal(0.5, 0.1);\n  v ~ normal(2, 3);\n  t0 ~ normal(0.435, 0.12);\n\n  sv ~ normal(1, 3);\n  st0 ~ normal(0.183, 0.09);\n  sw ~ beta(1, 3);\n\n  // likelihood (diffusion model)\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // upper boundary\n      target += wiener_full_lpdf(rt[i] | a, t0, w, v,\n                                         sv, sw, st0);\n    } else {\n      // lower boundary: mirror drift and starting point\n      target += wiener_full_lpdf(rt[i] | a, t0, one_minus_w,\n                                         neg_v, sv, sw, st0);\n    }\n  }\n}\n\n\nThe data should consist of at least three variables:\n\nThe number of trials N,\nthe response, coded as 0 = “lower bound” and 1 = “upper bound”, and\nthe reaction times in seconds (not milliseconds).\n\nNote that two different ways of coding responses are commonly used: First, in response coding, the boundaries correspond to the two response alternatives. Second, in accuracy coding, the boundaries correspond to correct (upper bound) and wrong (lower bound) responses. This means, depending on the coding you choose, the bounds mentioned in the second variable above differ and the response variable will have a different form.\nMost often, an experimenter wants to find out whether an experimental manipulation influences the model parameters. As there exists psychological interpretations for each diffusion model parameter, the experimenter can draw conclusions from differing parameters. Therefore, usually an own diffusion model is being computed for each experimental group to enable a comparison of the parameters between the groups. This can be manipulation between different subjects, like an experimental group and a control group (so called between-subject manipulations). However, this can also be manipulations within the same subject by presenting stimuli from different experimental groups (so called within-subject manipulations). Depending on the experimental design, one would typically also provide the number of conditions and the condition associated with each trial as a vector. Then, one model for each condition will be computed. This means that the parameters also have to be defined for each condition.\nIn a hierarchical setting, the data block would also specify the number of participants and the participant associated with each trial as a vector. It is also possible to hand over a precision value in the data block.\n\n\n\nThe model arguments of the wiener_lpdf() function that are not fixed to a certain value are defned as parameters in the parameters block. In this block, it is also possible to insert restrictions on the parameters. Note that the MCMC algorithm iteratively searches for the next parameter set. If the suggested sample falls outside the internally defined parameter ranges, the program will throw an error, which causes the algorithm to restart the current iteration. Since this slows down the sampling process, it is advisable to include the parameter ranges in the defnition of the parameters in the parameters block to improve the sampling process (see table below for the parameter ranges). In addition, the parameter space is further constrained by the following conditions:\n\nThe non-decision time \\(t_0\\) has to be smaller or equal to the observed reaction time: \\(t0 \\leq y\\).\nThe varying relative starting point \\(w\\) has to be in the interval (0,1) and thus,\n\n\\[\n\\begin{aligned}\n&w + \\frac{s_w}{2} &lt; 1 \\text{, and} \\\\\n&0 &lt; w-\\frac{s_w}{2}\n\\end{aligned}\n\\]\n\n\n\nParameter\nRange\n\nParameter\nRange\n\n\n\n\n\\(a\\)\n(0, \\(\\infty\\))\n\n\\(y\\)\n(0, \\(\\infty\\))\n\n\n\\(v\\)\n(-\\(\\infty\\), \\(\\infty\\))\n\n\\(s_v\\)\n[0, \\(\\infty\\))\n\n\n\\(w\\)\n(0,1)\n\n\\(s_w\\)\n[0,min(2w, 2(1-w)))\n\n\n\\(t_0\\)\n[0,\\(\\infty\\))\n\n\\(s_{t0}\\)\n[0,\\(\\infty\\))\n\n\n\n\n\n\nIn the model block, the priors and likelihood are defined for the upper and the lower response boundary. Different kinds of priors can be specifed here. Generally, the regularization induced by mildly informative priors can help both statistically and computationally.\nIn the second part of the model block, the data generating distribution is applied to all responses. The drift rate \\(v\\) and relative starting point \\(w\\) have to be mirrored for responses at the lower boundary.\nFor more details regarding the application of the diffusion model in Stan, see Henrich et al. (2024).\n\n\n\n\nTruncation and censoring frequently occur in psychological data collection. For reaction time data, truncated and censored data regularly arise in psychological studies as a consequence of using response windows or deadlines. These are sometimes introduced in the analysis of data to exclude reaction times that appear too short or too long, but they are also sometimes already built into the study procedures to push participants to respond within a specifc temporal window.\nDepending on the implementation of the response window, two different types of data arise: truncated data or censored data. Since the effects of truncation or censoring on summary statistics such as mean, median, standard deviation, and skewness is regularly too large to ignore (Ulrich and Miller 1994), data analysts are well advised to account for these effects.\nAs described in the Truncated or Censored Data chapter, the cumulative distribution function (\\(F\\)) and its complement (\\(\\text{CCDF}\\)) are needed to model truncated and censored data.\nAs explained above, \\(p\\) is defined defectively, meaning that only the sum of \\(p\\)s for both response alternatives integrates to 1. For the same reason, \\(F\\) and \\(\\text{CCDF}\\) are also implemented defectively. Analogously, only the sum of the \\(F\\)s and \\(\\text{CCDF}\\)s for both response alternatives asymptotes above at 1.\nIn the case of the diffusion model, \\(F\\) asymptotes above at the probability \\(PROB\\) to hit the corresponding response boundary: (for simplicity, we omit the inter-trial variabilities in the following)\n\\[\n\\begin{aligned}\nF_1(\\infty\\mid a,w,v) &= \\text{PROB}(a,w,v) \\text{ and} \\\\\nF_0(\\infty\\mid a,w,v) &= F_1(\\infty\\mid a,1-w,-v) = \\text{PROB}(a,1-w,-v)\n\\end{aligned}\n\\]\n\n\nData are called truncated when there is no information available for analysis from trials with values larger (or smaller) than a right (or left) reaction-time bound. In reaction time experiments, reaction time data are truncated if trials with reaction times outside the response window are excluded from the analysis. Not even a count of those omitted trials is kept.\nLet \\(L\\) denote the left reaction-time bound and \\(U\\) denote the right reaction-time bound of a response window.\nThen, the density of truncated data for both response boundaries 0 and 1, here denoted as \\(\\text{resp}\\in\\{0,1\\}\\), can be formulated as follows:\n\\[\n\\begin{aligned}\n&p_{\\text{resp}}(y \\mid L&lt;X\\leq U, a, w, v) = \\\\ &\\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\leq U\\}}}\n{\\bigl(F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)\\bigr) -\n\\bigl(F_0(L\\mid a, w, v)+F_1(L\\mid a, w, v)\\bigr)}\n\\end{aligned}\n\\]\nThe density of left truncated data can be formulated as follows. \\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid L&lt;X, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\}}}\n{1-\\bigl(F_0(L \\mid a, w, v)+F_1(L \\mid a, w, v)\\bigr)},\n\\end{aligned}\n\\]\nThe density of right truncated data can be formulated as follows.\n\\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid X\\leq U, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{y\\leq U\\}}}{F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)}\n\\end{aligned}\n\\]\nAs the functions are implemented defectively, a truncated diffusion model cannot be calculated with the truncation functor \\(T[,]\\) as it would usually be done in Stan. This means the function call: y ~ wiener(...)T[L,U] does not work the way it is supposed to. When the truncation functor is called in Stan, Stan searches for a CDF implementation internally. In the case of the diffusion model, Stan would find the CDF, but is not aware of its defective implementation and calculates the computations as if it were a non-defective CDF. This causes misleading and incorrect results.\nTo implement the truncated model, write out the function shown above on the log-scale with left_bound = L and right_bound = U, where wiener_lcdf_unnorm() calls the logarithmized CDF of the diffusion model at the response-1-boundary:\nmodel {\n  real log_denom = log_diff_exp(\n    log_sum_exp(\n      wiener_lcdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                       sv, sw, st)),\n    log_sum_exp(\n      wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                      sv, sw, st)));\n  // likelihood\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // response -1 boundary\n      target += wiener_lpdf (rt[i] | a, t0, w, v, sv, sw, st);\n    } else { \n      // response -0 boundary ( mirror v and w)\n      target += wiener_lpdf (rt[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st);\n    }\n  } // end for\n  target += -N * log_denom;\n}\nFor details of how to call a truncated model within the parallelization routine of reduce_sum or with truncation to only on side, see Henrich and Klauer (2026).\n\n\n\nData are censored when observations that are above or below a right or left boundary value are reported as occurrences of the event \\((y &gt;\nU)\\), for \\(U\\) the right bound, or as occurrences of the event \\((y \\leq\nL)\\), for \\(L\\) the left bound, respectively. Like for truncated data, the range of the possible values is restricted, but the number of observations that fall outside the boundaries is kept, whereas in truncation, no count would be kept.\nFor the censored model, we distinguish two cases. In the first case, the responses of the censored trials are known, but the reaction times are not known. In the second case, neither the responses nor the reaction times of the censored trials are known. Note that the second case differs from a truncated model in the fact that the number of censored trials is still known. Consider first the case where the response is known even for censored data.\nTo model such data in Stan, the left and right reaction time bounds, left_bound and right_bound, respectively, are handed over in the data block, as well as a vector censored that tracks whether a trial is censored (= 1) or not (= 0), and counts of trials censored at the left reaction time bound and counts of trials censored at the right reaction time bound for each response in {0,1}. There are four such count variables: N_cens_left_0, N_cens_left_1, N_cens_right_0, N_cens_right_1:\nmodel {\n  for (i in 1:N) {\n    if (censored[i] == 0) {\n      if (resp[i] == 1) {\n        y[i] ~ wiener(a, t0, w, v, sv, sw, st0);\n      } else if (resp[i] == 0) {\n        y[i] ~ wiener(a, t0, one_minus_w, neg_v, sv, sw, st0);\n      }\n    }\n  }\n\n  // likelihood (response = 0)\n  target += N_cens_left_0 \n         * wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                           sv, sw, st0);\n\n  target += N_cens_right_0 \n         * wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                             sv, sw, st0);\n\n  // likelihood (response = 1)\n  target += N_cens_left_1 \n         * wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st0);\n\n  target += N_cens_right_1 \n         * wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st0);\n}\nWhen data are censored at only one side, meaning that the reaction time constraint only exists for one of the two boundaries, omit the lines for the other side in the code. A both sided reaction time window would be, for example, when only reaction times are accepted that occur between 0.2 and 0.8 seconds. A one sided reaction time constraint would be, for example, when all reaction times below 0.8 seconds are accepted.\nWhen data consist of many conditions (as explained in the beginning), it is sometimes more convenient to loop over all trials instead of using count variables as described above, using the following notation and code. A vector containing the information whether a trial is censored or not, here censored, needs to be handed over in the data block. This vector splits the data into three bins: all trials \\(i\\) withcensored[i]=0 are censored below the left reaction time bound, all trials \\(i\\) with censored[i]=1 fall between the reaction time bounds, and all trials \\(i\\) with censored[i]=2 are censored above the right reaction time bound. For non-censored trials, the log-PDF is computed, for left censored trials, the log-CDF is computed, and for right censored trials, the log-CCDF is computed:\nmodel { \n  for (i in 1:N) { \n    // right censored at right_bound\n    if (resp [i] == 1) { \n      // upper response boundary\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, w, v, \n                                                  sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, w, v, sv, sw, st0);\n      } else if (censored[i] == 2) {\n         target += wiener_lccdf_unnorm(right_bound | a, t0, w, v,\n                                                     sv, sw, st0);\n      }\n    } else { \n      // lower response boundary (mirror drift and // starting point!)\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w,\n                                                  neg_v, sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st0);\n      } else if (censored[i] == 2) {\n        target += wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                                    neg_v, sv, sw, st0);\n      }\n    }\n  }\n}\nWhen the data are censored on only one side, omit the case that is not needed.\nNote that this block can be inserted in the defnition of the parallelization function, partial_sum_wiener(), as defined below.\nSometimes also the response is missing (i.e., it is known that the reaction time in a trial fell outside the response window, but which response was given is unknown). One method that has been used to model such data has involved inferring the numbers of missing responses of either kind from the observed relative frequencies of the two responses. This approach has the problem that quite specifc assumptions on the missing data have to be made (namely, that the proportions of the two kinds of responses are the same for responses within and outside the response window).\nThe following is a more principled approach that uses the cumulative distribution functions and their complements to provide the data-generating distribution of censored data. As before, let \\(L\\) be the left reaction time bound, and \\(U\\) the right reaction time bound, and consider decision times without inter-trial variabilities for the sake of simplicity. It follows that the likelihood contribution \\(\\textit{lik}_l\\) for a left-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_l(a,w,v) = F_0(L\\mid a,w,v) + F_1(L\\mid a,w,v),\n\\end{aligned}\n\\]\nwhereas the likelihood contribution \\(lik_r\\) due to a right-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_r(a,w,v) = \\text{CCDF}_0(U\\mid a,w,v) + \\text{CCDF}_1(U\\mid a,w,v).\n\\end{aligned}\n\\]\nSee the following code for an example of Stan code implementing this second case of censoring. This model call deals with the problem of unknown responses by computing the probability of choosing the response-1 or response-0 boundary outside the response window. Here, the CDF and/or the CCDF are required, depending upon whether there is only left-censoring, right-censoring, or censoring both to the left and to the right. The following code shows the functions block for a model that is right-censored using the function partial_sum_wiener() to parallelize the execution of a single Stan chain across multiple cores:\nfunctions {\n  real partial_sum_wiener(array[] real rt_slice, int start,\n                          int end, real a, real t0, real w,\n                          real v, real sv, real sw, real st,\n                          array[] int resp, real right_bound,\n                          array[] int censored) {\n    real ans = 0;\n    for (i in start:end) {\n      if (censored[i] == 1) {\n        // not censored\n        if (resp[i] == 1) {\n          // upper boundary\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, w, v,\n                                                     sv, sw, st);\n        } else {\n          // lower boundary(mirror v and w)\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, one_minus_w,\n                                                     neg_v, sv, sw, st);\n        }\n      } else { \n        // censored\n        ans += log_sum_exp (\n          wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n          wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                            neg_v, sv, sw, st);\n      }\n    }\n    return ans;\n  }\n}\nCombine this block with the model block in the example above by using the function reduce_sum().\n  target += reduce_sum(partial_sum_wiener, rt, 1,\n    a, t0, w, v, sv, sw, st, resp, right_bound, censored);\n}\nFor more details, see Henrich and Klauer (2026).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Wiener diffusion model"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/wiener_diffusion_model.html#function-call-example",
+    "href": "stan-users-guide/wiener_diffusion_model.html#function-call-example",
+    "title": "Wiener diffusion model",
+    "section": "",
+    "text": "The following example demonstrates a diffusion model call in Stan:\ndata {\n  int &lt;lower=0&gt; N; // Number of trials\n  array[N] real rt; // response times (in seconds )\n  array[N] int &lt;lower=0, upper=1&gt; resp; // responses {0 ,1}\n}\ntransformed data{\n  real min_rt = min(rt);\n}\nparameters {\n  real &lt;lower=0&gt; a;                // boundary separation\n  real v;                          // drift\n  real &lt;lower=0, upper=1&gt; w;       // relative starting point\n  real &lt;lower=0, upper=min_rt&gt; t0; // non-decision time\n\n  real &lt;lower=0&gt; sv;               // variability in drift\n  // variability in starting point\n  real &lt;lower=0, upper=fmin(2 * w, 2 * (1 - w))&gt; sw; \n  real &lt;lower=0&gt; st0;             // variability in non-decision time\n}\ntransformed parameters{\n  real one_minus_w = 1 - w;\n  real neg_v = -v;\n}\nmodel {\n  // prior\n  a ~ normal(1, 1);\n  w ~ normal(0.5, 0.1);\n  v ~ normal(2, 3);\n  t0 ~ normal(0.435, 0.12);\n\n  sv ~ normal(1, 3);\n  st0 ~ normal(0.183, 0.09);\n  sw ~ beta(1, 3);\n\n  // likelihood (diffusion model)\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // upper boundary\n      target += wiener_full_lpdf(rt[i] | a, t0, w, v,\n                                         sv, sw, st0);\n    } else {\n      // lower boundary: mirror drift and starting point\n      target += wiener_full_lpdf(rt[i] | a, t0, one_minus_w,\n                                         neg_v, sv, sw, st0);\n    }\n  }\n}\n\n\nThe data should consist of at least three variables:\n\nThe number of trials N,\nthe response, coded as 0 = “lower bound” and 1 = “upper bound”, and\nthe reaction times in seconds (not milliseconds).\n\nNote that two different ways of coding responses are commonly used: First, in response coding, the boundaries correspond to the two response alternatives. Second, in accuracy coding, the boundaries correspond to correct (upper bound) and wrong (lower bound) responses. This means, depending on the coding you choose, the bounds mentioned in the second variable above differ and the response variable will have a different form.\nMost often, an experimenter wants to find out whether an experimental manipulation influences the model parameters. As there exists psychological interpretations for each diffusion model parameter, the experimenter can draw conclusions from differing parameters. Therefore, usually an own diffusion model is being computed for each experimental group to enable a comparison of the parameters between the groups. This can be manipulation between different subjects, like an experimental group and a control group (so called between-subject manipulations). However, this can also be manipulations within the same subject by presenting stimuli from different experimental groups (so called within-subject manipulations). Depending on the experimental design, one would typically also provide the number of conditions and the condition associated with each trial as a vector. Then, one model for each condition will be computed. This means that the parameters also have to be defined for each condition.\nIn a hierarchical setting, the data block would also specify the number of participants and the participant associated with each trial as a vector. It is also possible to hand over a precision value in the data block.\n\n\n\nThe model arguments of the wiener_lpdf() function that are not fixed to a certain value are defned as parameters in the parameters block. In this block, it is also possible to insert restrictions on the parameters. Note that the MCMC algorithm iteratively searches for the next parameter set. If the suggested sample falls outside the internally defined parameter ranges, the program will throw an error, which causes the algorithm to restart the current iteration. Since this slows down the sampling process, it is advisable to include the parameter ranges in the defnition of the parameters in the parameters block to improve the sampling process (see table below for the parameter ranges). In addition, the parameter space is further constrained by the following conditions:\n\nThe non-decision time \\(t_0\\) has to be smaller or equal to the observed reaction time: \\(t0 \\leq y\\).\nThe varying relative starting point \\(w\\) has to be in the interval (0,1) and thus,\n\n\\[\n\\begin{aligned}\n&w + \\frac{s_w}{2} &lt; 1 \\text{, and} \\\\\n&0 &lt; w-\\frac{s_w}{2}\n\\end{aligned}\n\\]\n\n\n\nParameter\nRange\n\nParameter\nRange\n\n\n\n\n\\(a\\)\n(0, \\(\\infty\\))\n\n\\(y\\)\n(0, \\(\\infty\\))\n\n\n\\(v\\)\n(-\\(\\infty\\), \\(\\infty\\))\n\n\\(s_v\\)\n[0, \\(\\infty\\))\n\n\n\\(w\\)\n(0,1)\n\n\\(s_w\\)\n[0,min(2w, 2(1-w)))\n\n\n\\(t_0\\)\n[0,\\(\\infty\\))\n\n\\(s_{t0}\\)\n[0,\\(\\infty\\))\n\n\n\n\n\n\nIn the model block, the priors and likelihood are defined for the upper and the lower response boundary. Different kinds of priors can be specifed here. Generally, the regularization induced by mildly informative priors can help both statistically and computationally.\nIn the second part of the model block, the data generating distribution is applied to all responses. The drift rate \\(v\\) and relative starting point \\(w\\) have to be mirrored for responses at the lower boundary.\nFor more details regarding the application of the diffusion model in Stan, see Henrich et al. (2024).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Wiener diffusion model"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/wiener_diffusion_model.html#wiener-truncation",
+    "href": "stan-users-guide/wiener_diffusion_model.html#wiener-truncation",
+    "title": "Wiener diffusion model",
+    "section": "",
+    "text": "Truncation and censoring frequently occur in psychological data collection. For reaction time data, truncated and censored data regularly arise in psychological studies as a consequence of using response windows or deadlines. These are sometimes introduced in the analysis of data to exclude reaction times that appear too short or too long, but they are also sometimes already built into the study procedures to push participants to respond within a specifc temporal window.\nDepending on the implementation of the response window, two different types of data arise: truncated data or censored data. Since the effects of truncation or censoring on summary statistics such as mean, median, standard deviation, and skewness is regularly too large to ignore (Ulrich and Miller 1994), data analysts are well advised to account for these effects.\nAs described in the Truncated or Censored Data chapter, the cumulative distribution function (\\(F\\)) and its complement (\\(\\text{CCDF}\\)) are needed to model truncated and censored data.\nAs explained above, \\(p\\) is defined defectively, meaning that only the sum of \\(p\\)s for both response alternatives integrates to 1. For the same reason, \\(F\\) and \\(\\text{CCDF}\\) are also implemented defectively. Analogously, only the sum of the \\(F\\)s and \\(\\text{CCDF}\\)s for both response alternatives asymptotes above at 1.\nIn the case of the diffusion model, \\(F\\) asymptotes above at the probability \\(PROB\\) to hit the corresponding response boundary: (for simplicity, we omit the inter-trial variabilities in the following)\n\\[\n\\begin{aligned}\nF_1(\\infty\\mid a,w,v) &= \\text{PROB}(a,w,v) \\text{ and} \\\\\nF_0(\\infty\\mid a,w,v) &= F_1(\\infty\\mid a,1-w,-v) = \\text{PROB}(a,1-w,-v)\n\\end{aligned}\n\\]\n\n\nData are called truncated when there is no information available for analysis from trials with values larger (or smaller) than a right (or left) reaction-time bound. In reaction time experiments, reaction time data are truncated if trials with reaction times outside the response window are excluded from the analysis. Not even a count of those omitted trials is kept.\nLet \\(L\\) denote the left reaction-time bound and \\(U\\) denote the right reaction-time bound of a response window.\nThen, the density of truncated data for both response boundaries 0 and 1, here denoted as \\(\\text{resp}\\in\\{0,1\\}\\), can be formulated as follows:\n\\[\n\\begin{aligned}\n&p_{\\text{resp}}(y \\mid L&lt;X\\leq U, a, w, v) = \\\\ &\\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\leq U\\}}}\n{\\bigl(F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)\\bigr) -\n\\bigl(F_0(L\\mid a, w, v)+F_1(L\\mid a, w, v)\\bigr)}\n\\end{aligned}\n\\]\nThe density of left truncated data can be formulated as follows. \\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid L&lt;X, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\}}}\n{1-\\bigl(F_0(L \\mid a, w, v)+F_1(L \\mid a, w, v)\\bigr)},\n\\end{aligned}\n\\]\nThe density of right truncated data can be formulated as follows.\n\\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid X\\leq U, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{y\\leq U\\}}}{F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)}\n\\end{aligned}\n\\]\nAs the functions are implemented defectively, a truncated diffusion model cannot be calculated with the truncation functor \\(T[,]\\) as it would usually be done in Stan. This means the function call: y ~ wiener(...)T[L,U] does not work the way it is supposed to. When the truncation functor is called in Stan, Stan searches for a CDF implementation internally. In the case of the diffusion model, Stan would find the CDF, but is not aware of its defective implementation and calculates the computations as if it were a non-defective CDF. This causes misleading and incorrect results.\nTo implement the truncated model, write out the function shown above on the log-scale with left_bound = L and right_bound = U, where wiener_lcdf_unnorm() calls the logarithmized CDF of the diffusion model at the response-1-boundary:\nmodel {\n  real log_denom = log_diff_exp(\n    log_sum_exp(\n      wiener_lcdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                       sv, sw, st)),\n    log_sum_exp(\n      wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                      sv, sw, st)));\n  // likelihood\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // response -1 boundary\n      target += wiener_lpdf (rt[i] | a, t0, w, v, sv, sw, st);\n    } else { \n      // response -0 boundary ( mirror v and w)\n      target += wiener_lpdf (rt[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st);\n    }\n  } // end for\n  target += -N * log_denom;\n}\nFor details of how to call a truncated model within the parallelization routine of reduce_sum or with truncation to only on side, see Henrich and Klauer (2026).\n\n\n\nData are censored when observations that are above or below a right or left boundary value are reported as occurrences of the event \\((y &gt;\nU)\\), for \\(U\\) the right bound, or as occurrences of the event \\((y \\leq\nL)\\), for \\(L\\) the left bound, respectively. Like for truncated data, the range of the possible values is restricted, but the number of observations that fall outside the boundaries is kept, whereas in truncation, no count would be kept.\nFor the censored model, we distinguish two cases. In the first case, the responses of the censored trials are known, but the reaction times are not known. In the second case, neither the responses nor the reaction times of the censored trials are known. Note that the second case differs from a truncated model in the fact that the number of censored trials is still known. Consider first the case where the response is known even for censored data.\nTo model such data in Stan, the left and right reaction time bounds, left_bound and right_bound, respectively, are handed over in the data block, as well as a vector censored that tracks whether a trial is censored (= 1) or not (= 0), and counts of trials censored at the left reaction time bound and counts of trials censored at the right reaction time bound for each response in {0,1}. There are four such count variables: N_cens_left_0, N_cens_left_1, N_cens_right_0, N_cens_right_1:\nmodel {\n  for (i in 1:N) {\n    if (censored[i] == 0) {\n      if (resp[i] == 1) {\n        y[i] ~ wiener(a, t0, w, v, sv, sw, st0);\n      } else if (resp[i] == 0) {\n        y[i] ~ wiener(a, t0, one_minus_w, neg_v, sv, sw, st0);\n      }\n    }\n  }\n\n  // likelihood (response = 0)\n  target += N_cens_left_0 \n         * wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                           sv, sw, st0);\n\n  target += N_cens_right_0 \n         * wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                             sv, sw, st0);\n\n  // likelihood (response = 1)\n  target += N_cens_left_1 \n         * wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st0);\n\n  target += N_cens_right_1 \n         * wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st0);\n}\nWhen data are censored at only one side, meaning that the reaction time constraint only exists for one of the two boundaries, omit the lines for the other side in the code. A both sided reaction time window would be, for example, when only reaction times are accepted that occur between 0.2 and 0.8 seconds. A one sided reaction time constraint would be, for example, when all reaction times below 0.8 seconds are accepted.\nWhen data consist of many conditions (as explained in the beginning), it is sometimes more convenient to loop over all trials instead of using count variables as described above, using the following notation and code. A vector containing the information whether a trial is censored or not, here censored, needs to be handed over in the data block. This vector splits the data into three bins: all trials \\(i\\) withcensored[i]=0 are censored below the left reaction time bound, all trials \\(i\\) with censored[i]=1 fall between the reaction time bounds, and all trials \\(i\\) with censored[i]=2 are censored above the right reaction time bound. For non-censored trials, the log-PDF is computed, for left censored trials, the log-CDF is computed, and for right censored trials, the log-CCDF is computed:\nmodel { \n  for (i in 1:N) { \n    // right censored at right_bound\n    if (resp [i] == 1) { \n      // upper response boundary\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, w, v, \n                                                  sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, w, v, sv, sw, st0);\n      } else if (censored[i] == 2) {\n         target += wiener_lccdf_unnorm(right_bound | a, t0, w, v,\n                                                     sv, sw, st0);\n      }\n    } else { \n      // lower response boundary (mirror drift and // starting point!)\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w,\n                                                  neg_v, sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st0);\n      } else if (censored[i] == 2) {\n        target += wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                                    neg_v, sv, sw, st0);\n      }\n    }\n  }\n}\nWhen the data are censored on only one side, omit the case that is not needed.\nNote that this block can be inserted in the defnition of the parallelization function, partial_sum_wiener(), as defined below.\nSometimes also the response is missing (i.e., it is known that the reaction time in a trial fell outside the response window, but which response was given is unknown). One method that has been used to model such data has involved inferring the numbers of missing responses of either kind from the observed relative frequencies of the two responses. This approach has the problem that quite specifc assumptions on the missing data have to be made (namely, that the proportions of the two kinds of responses are the same for responses within and outside the response window).\nThe following is a more principled approach that uses the cumulative distribution functions and their complements to provide the data-generating distribution of censored data. As before, let \\(L\\) be the left reaction time bound, and \\(U\\) the right reaction time bound, and consider decision times without inter-trial variabilities for the sake of simplicity. It follows that the likelihood contribution \\(\\textit{lik}_l\\) for a left-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_l(a,w,v) = F_0(L\\mid a,w,v) + F_1(L\\mid a,w,v),\n\\end{aligned}\n\\]\nwhereas the likelihood contribution \\(lik_r\\) due to a right-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_r(a,w,v) = \\text{CCDF}_0(U\\mid a,w,v) + \\text{CCDF}_1(U\\mid a,w,v).\n\\end{aligned}\n\\]\nSee the following code for an example of Stan code implementing this second case of censoring. This model call deals with the problem of unknown responses by computing the probability of choosing the response-1 or response-0 boundary outside the response window. Here, the CDF and/or the CCDF are required, depending upon whether there is only left-censoring, right-censoring, or censoring both to the left and to the right. The following code shows the functions block for a model that is right-censored using the function partial_sum_wiener() to parallelize the execution of a single Stan chain across multiple cores:\nfunctions {\n  real partial_sum_wiener(array[] real rt_slice, int start,\n                          int end, real a, real t0, real w,\n                          real v, real sv, real sw, real st,\n                          array[] int resp, real right_bound,\n                          array[] int censored) {\n    real ans = 0;\n    for (i in start:end) {\n      if (censored[i] == 1) {\n        // not censored\n        if (resp[i] == 1) {\n          // upper boundary\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, w, v,\n                                                     sv, sw, st);\n        } else {\n          // lower boundary(mirror v and w)\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, one_minus_w,\n                                                     neg_v, sv, sw, st);\n        }\n      } else { \n        // censored\n        ans += log_sum_exp (\n          wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n          wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                            neg_v, sv, sw, st);\n      }\n    }\n    return ans;\n  }\n}\nCombine this block with the model block in the example above by using the function reduce_sum().\n  target += reduce_sum(partial_sum_wiener, rt, 1,\n    a, t0, w, v, sv, sw, st, resp, right_bound, censored);\n}\nFor more details, see Henrich and Klauer (2026).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Wiener diffusion model"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html",
+    "href": "stan-users-guide/user-functions.html",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "This chapter explains functions from a user perspective with examples; see the language reference for a full specification. User-defined functions allow computations to be encapsulated into a single named unit and invoked elsewhere by name. Similarly, functions allow complex procedures to be broken down into more understandable components. Writing modular code using descriptively named functions is easier to understand than a monolithic program, even if the latter is heavily commented.1\n\n\nHere’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.\nfunctions {\n  real relative_diff(real x, real y) {\n    real abs_diff;\n    real avg_scale;\n    abs_diff = abs(x - y);\n    avg_scale = (abs(x) + abs(y)) / 2;\n    return abs_diff / avg_scale;\n  }\n}\n// ...\ngenerated quantities {\n  real rdiff;\n  rdiff = relative_diff(alpha, beta);\n}\nThe function is named relative_diff, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.\n\n\nAll functions are defined in their own block, which is labeled functions and must appear before all other program blocks. The user-defined functions block is optional.\n\n\n\nThe body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.\n\n\n\nReturn statements, such as the one on the last line of the definition of relative_diff above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.\nIf an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a fatal_error statement.\n\n\nRejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).\nThe most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.\nreal dbl_sqrt(real x) {\n  if (!(x &gt;= 0)) {\n    reject(\"dbl_sqrt(x): x must be positive; found x = \", x);\n  }\n  return 2 * sqrt(x);\n}\nThe negation of the positive test is important, because it also catches the case where x is a not-a-number value. If the condition had been coded as (x &lt; 0) it would not catch the not-a-number case, though it could be written as (x &lt; 0 || is_nan(x)). The positive infinite case is allowed through, but could also be checked with the is_inf(x) function. The square root function does not itself reject, but some downstream consumer of dbl_sqrt(-2) would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.\nThe effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.2\nIn generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.\n\n\n\n\nFunction argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the reference manual for full details.\nFor example, here’s a function to compute the entropy of a categorical distribution with simplex parameter theta.\nreal entropy(vector theta) {\n  return sum(theta .* log(theta));\n}\nAlthough theta must be a simplex, only the type vector is used.3\nUpper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.\n\n\n\nArray arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.\narray[] real baz(array[,] real x);\nThe notation [ ] is used for one-dimensional arrays (as in the return above), [ , ] for two-dimensional arrays, [ , , ] for three-dimensional arrays, and so on.\nFunctions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.\n\n\n\nA function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier data. The following is an example of a data-only function argument.\nreal foo(real y, data real mu) {\n  return -0.5 * (y - mu)^2;\n}\nThis qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.\nUse of the data qualifier must be consistent between the forward declaration and the definition of a functions.\nThis qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the ODE solver chapter for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the data prefix qualifier. This allows for compile-time type checking and increases overall program understandability.\n\n\n\n\nIn some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.\nfunctions {\n  void pretty_print_tri_lower(matrix x) {\n    if (rows(x) == 0) {\n      print(\"empty matrix\");\n      return;\n    }\n    print(\"rows=\", rows(x), \" cols=\", cols(x));\n    for (m in 1:rows(x)) {\n      for (n in 1:m) {\n        print(\"[\", m, \",\", n, \"]=\", x[m, n]);\n      }\n    }\n  }\n}\nThe special symbol void is used as the return type. This is not a type itself in that there are no values of type void; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.\nVoid functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ... code to set Sigma ...\n  pretty_print_tri_lower(Sigma);\n  // ...\n}\n\n\n\nFunctions whose names end in _lp are allowed to use sampling statements and target += statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.\nHere is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the reparameterization section for more information on efficient non-centered parameterizations\nfunctions {\n  vector center_lp(vector beta_raw, real mu, real sigma) {\n    beta_raw ~ std_normal();\n    sigma ~ cauchy(0, 5);\n    mu ~ cauchy(0, 2.5);\n    return sigma * beta_raw + mu;\n  }\n  // ...\n}\nparameters {\n  vector[K] beta_raw;\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // ...\n  beta = center_lp(beta_raw, mu_beta, sigma_beta);\n  // ...\n}\n\n\n\nFunctions whose names end in _jacobian can use the jacobian += statement. This can be used to implement a custom change of variables for arbitrary parameters.\nFor example, this function recreates the built-in &lt;upper=x&gt; transform on real numbers:\nreal my_upper_bound_jacobian(real x, real ub) {\n  jacobian += x;\n  return ub - exp(x);\n}\nIt can be used as a replacement for real&lt;lower=ub&gt; as follows:\nfunctions {\n  // my_upper_bound_jacobian as above\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  b ~ lognormal(0, 1);\n  // ...\n}\n\n\n\nA user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in _rng. Giving a function a name that ends in _rng allows it to access built-in functions and user-defined functions that end in _rng, which includes all the built-in PRNG functions. Only functions ending in _rng are able access the built-in PRNG functions. The use of functions ending in _rng must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in _rng.\nFor example, the following function generates an \\(N \\times K\\) data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.\nmatrix predictors_rng(int N, int K) {\n  matrix[N, K] x;\n  for (n in 1:N) {\n    x[n, 1] = 1.0;  // intercept\n    for (k in 2:K) {\n      x[n, k] = normal_rng(0, 1);\n    }\n  }\n  return x;\n}\nThe following function defines a simulator for regression outcomes based on a data matrix x, coefficients beta, and noise scale sigma.\nvector regression_rng(vector beta, matrix x, real sigma) {\n  vector[rows(x)] y;\n  vector[rows(x)] mu;\n  mu = x * beta;\n  for (n in 1:rows(x)) {\n    y[n] = normal_rng(mu[n], sigma);\n  }\n  return y;\n}\nThese might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n  // ...\n}\ngenerated quantities {\n  matrix[N_sim, K] x_sim;\n  vector[N_sim] y_sim;\n  x_sim = predictors_rng(N_sim, K);\n  y_sim = regression_rng(beta, x_sim, sigma);\n}\nA more sophisticated simulation might fit a multivariate normal to the predictors x and use the resulting parameters to generate multivariate normal draws for x_sim.\n\n\n\nProbability functions are distinguished in Stan by names ending in _lpdf for density functions and _lpmf for mass functions; in both cases, they must have real return types.\nSuppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.\nfunctions {\n  real unit_normal_lpdf(real y) {\n    return normal_lpdf(y | 0, 1);\n  }\n}\n// ...\nmodel {\n  alpha ~ unit_normal();\n  beta ~ unit_normal();\n  // ...\n}\nThe ability to use the unit_normal function as a density is keyed off its name ending in _lpdf (names ending in _lpmf for probability mass functions work the same way).\nIn general, if foo_lpdf is defined to consume \\(N + 1\\) arguments, then\ny ~ foo(theta1, ..., thetaN);\ncan be used as shorthand for\ntarget += foo_lpdf(y | theta1, ..., thetaN);\nAs with the built-in functions, the suffix _lpdf is dropped and the first argument moves to the left of the tilde symbol (~) in the distribution statement.\nFunctions ending in _lpmf (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (_lpdf) must be continuous (not an integer or integer array), whereas the first argument of a mass function (_lpmf) must be discrete (integer or integer array).\n\n\n\nAs described in the reference manual function overloading is permitted in Stan, beginning in version 2.29.\nThis means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.\n\n\nOverloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.\nFor example, consider a function triple which has the following three signatures\nreal triple(real x);\ncomplex triple(complex x);\narray[] real triple(array[] real);\nOne should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect\n// The function does what it says\ntriple(3.0) == 9.0\n// It is defined reasonably for different types\ntriple(to_complex(3.0)) == to_complex(triple(3.0))\n// A container version of this function works by element\ntriple({3.0, 4.0})[0] == triple({3.0, 4.0}[0])\nNote that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.\n\n\n\nStan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.\nConsider the following function signatures\nreal foo(int a, real b);\nreal foo(real a, real b);\nGiven these, the function call foo(1.5, 2.5) is unambiguous - it must resolve to the second signature. But, the function call foo(1, 1.5) could be valid for either under Stan’s promotion rules, which allow integers to be promoted to real numbers.\nTo resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call foo(1, 1.5) would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).\nFurthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures\nreal bar(int x, real y);\nreal bar(real x, int y);\nThese signatures do not have a unique minimum number of promotions for the call bar(1, 2). Both signatures require one int to real promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.\nPromotion from integers to complex numbers is considered to be two separate promotions, first from int to real, then from real to complex. This means that integer arguments will “prefer” a signature with real types over complex types.\nFor example, consider the function signatures\nreal pop(real x);\nreal pop(complex x);\nStan will select the first signature when pop is called with an integer argument such as pop(0).\n\n\n\n\nFunctions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.\nFor example, here’s some documentation for the prediction matrix generator.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows corresponding to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\nThe comment begins with /**, ends with */, and has an asterisk (*) on each line. It uses @param followed by the argument’s identifier to document a function argument. The tag @return is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with /* and ending with */ to the Stan parser.\nFor functions that raise exceptions, exceptions can be documented using @throws.4\nFor example,\n /** ...\n * @param theta\n * @throws If any of the entries of theta is negative.\n */\nreal entropy(vector theta) {\n  // ...\n}\nUsually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.\n\n\n\nFunctions may have a void or non-void return type and they may or may not have one of the special suffixes, _lpdf, _lpmf, _lp, or _rng.\n\n\nOnly functions declared to return void may be used as statements. These are also the only functions that use return statements with no arguments.\nOnly functions declared to return non-void values may be used as expressions. These functions require return statements with arguments of a type that matches the declared return type.\n\n\n\nOnly functions ending in _lpmf or _lpdf and with return type real may be used as probability functions in distribution statements.\nOnly functions ending in _lp may access the log probability accumulator through distribution statements or target += statements. Such functions may only be used in the transformed parameters or model blocks.\nOnly functions ending in _rng may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in _rng.\n\n\n\n\nStan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, \\(A^n\\), which is defined for a square matrix \\(A\\) and positive integer \\(n\\) by \\[\nA^n\n=\n\\begin{cases}\n\\textrm{I} & \\quad\\text{if } n = 0, \\text{ and} \\\\\nA \\, A^{n-1} & \\quad\\text{if } n &gt; 0.\n\\end{cases}\n\\]\nwhere \\(\\textrm{I}\\) is the identity matrix. This definition can be directly translated to a recursive function definition.\nmatrix matrix_pow(matrix a, int n) {\n  if (n == 0) {\n    return diag_matrix(rep_vector(1, rows(a)));\n  } else {\n    return a *  matrix_pow(a, n - 1);\n  }\n}\nIt would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.\nelse if (n == 1) {\n  return a;\n}\n\n\n\n\n\nTo generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then run through the inverse cumulative distribution function, \\(y = \\textrm{logit}(u)\\). If this were not already built in as logistic_rng(0, 1), it could be coded in Stan directly as\nreal standard_logistic_rng() {\n  real u = uniform_rng(0, 1);\n  real y = logit(u);\n  return y;\n}\nFollowing the same pattern, a standard normal RNG could be coded as\nreal standard_normal_rng() {\n  real u = uniform_rng(0, 1);\n  real y = inv_Phi(u);\n  return y;\n}\nthat is, \\(y = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}\\) is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function inv_Phi.\nIn order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate \\(\\textsf{normal}(\\mu, \\sigma)\\) variates, it is enough to generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then convert it to a standard normal variate, \\(z = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}(\\cdot)\\) is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, \\(y = \\mu +\n\\sigma \\times z\\). In code,\nreal my_normal_rng(real mu, real sigma) {\n  real u = uniform_rng(0, 1);\n  real z = inv_Phi(u);\n  real y = mu + sigma * z;\n  return y;\n}\nA robust version of this function would test that the arguments are finite and that sigma is non-negative, e.g.,\n  if (is_nan(mu) || is_inf(mu)) {\n    reject(\"my_normal_rng: mu must be finite; \",\n           \"found mu = \", mu);\n  }\n  if (is_nan(sigma) || is_inf(sigma) || sigma &lt; 0) {\n    reject(\"my_normal_rng: sigma must be finite and non-negative; \",\n           \"found sigma = \", sigma);\n  }\n\n\n\nOften truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.\n\n\nFor example, the following code generates a \\(\\textsf{Weibull}(\\alpha, \\sigma)\\) variate truncated below at a time \\(t\\),5\nreal weibull_lb_rng(real alpha, real sigma, real t) {\n  real p = weibull_cdf(t | alpha, sigma);   // cdf for lb\n  real u = uniform_rng(p, 1);               // unif in bounds\n  real y = sigma * (-log1m(u))^inv(alpha);  // inverse cdf\n  return y;\n}\n\n\n\nIf there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a \\(\\textsf{normal}(\\mu, \\sigma)\\) truncated to a region \\((a, b)\\), the following code suffices,\nreal normal_lub_rng(real mu, real sigma, real lb, real ub) {\n  real p_lb = normal_cdf(lb | mu, sigma);\n  real p_ub = normal_cdf(ub | mu, sigma);\n  real u = uniform_rng(p_lb, p_ub);\n  real y = mu + sigma * inv_Phi(u);\n  return y;\n}\nTo make this more robust, all variables should be tested for finiteness, sigma should be tested for positiveness, and lb and ub should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement\n  return mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),\n                                          normal_cdf(ub | mu, sigma)));\nfor readability. The names like p indicate probabilities, and p_lb and p_ub indicate the probabilities of the bounds. The variable u is clearly named as a uniform variate, and y is used to denote the variate being generated itself.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#basic-functions.section",
+    "href": "stan-users-guide/user-functions.html#basic-functions.section",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Here’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.\nfunctions {\n  real relative_diff(real x, real y) {\n    real abs_diff;\n    real avg_scale;\n    abs_diff = abs(x - y);\n    avg_scale = (abs(x) + abs(y)) / 2;\n    return abs_diff / avg_scale;\n  }\n}\n// ...\ngenerated quantities {\n  real rdiff;\n  rdiff = relative_diff(alpha, beta);\n}\nThe function is named relative_diff, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.\n\n\nAll functions are defined in their own block, which is labeled functions and must appear before all other program blocks. The user-defined functions block is optional.\n\n\n\nThe body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.\n\n\n\nReturn statements, such as the one on the last line of the definition of relative_diff above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.\nIf an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a fatal_error statement.\n\n\nRejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).\nThe most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.\nreal dbl_sqrt(real x) {\n  if (!(x &gt;= 0)) {\n    reject(\"dbl_sqrt(x): x must be positive; found x = \", x);\n  }\n  return 2 * sqrt(x);\n}\nThe negation of the positive test is important, because it also catches the case where x is a not-a-number value. If the condition had been coded as (x &lt; 0) it would not catch the not-a-number case, though it could be written as (x &lt; 0 || is_nan(x)). The positive infinite case is allowed through, but could also be checked with the is_inf(x) function. The square root function does not itself reject, but some downstream consumer of dbl_sqrt(-2) would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.\nThe effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.2\nIn generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.\n\n\n\n\nFunction argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the reference manual for full details.\nFor example, here’s a function to compute the entropy of a categorical distribution with simplex parameter theta.\nreal entropy(vector theta) {\n  return sum(theta .* log(theta));\n}\nAlthough theta must be a simplex, only the type vector is used.3\nUpper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.\n\n\n\nArray arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.\narray[] real baz(array[,] real x);\nThe notation [ ] is used for one-dimensional arrays (as in the return above), [ , ] for two-dimensional arrays, [ , , ] for three-dimensional arrays, and so on.\nFunctions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.\n\n\n\nA function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier data. The following is an example of a data-only function argument.\nreal foo(real y, data real mu) {\n  return -0.5 * (y - mu)^2;\n}\nThis qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.\nUse of the data qualifier must be consistent between the forward declaration and the definition of a functions.\nThis qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the ODE solver chapter for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the data prefix qualifier. This allows for compile-time type checking and increases overall program understandability.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#functions-as-statements",
+    "href": "stan-users-guide/user-functions.html#functions-as-statements",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "In some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.\nfunctions {\n  void pretty_print_tri_lower(matrix x) {\n    if (rows(x) == 0) {\n      print(\"empty matrix\");\n      return;\n    }\n    print(\"rows=\", rows(x), \" cols=\", cols(x));\n    for (m in 1:rows(x)) {\n      for (n in 1:m) {\n        print(\"[\", m, \",\", n, \"]=\", x[m, n]);\n      }\n    }\n  }\n}\nThe special symbol void is used as the return type. This is not a type itself in that there are no values of type void; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.\nVoid functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ... code to set Sigma ...\n  pretty_print_tri_lower(Sigma);\n  // ...\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#functions-accessing-the-log-probability-accumulator",
+    "href": "stan-users-guide/user-functions.html#functions-accessing-the-log-probability-accumulator",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Functions whose names end in _lp are allowed to use sampling statements and target += statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.\nHere is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the reparameterization section for more information on efficient non-centered parameterizations\nfunctions {\n  vector center_lp(vector beta_raw, real mu, real sigma) {\n    beta_raw ~ std_normal();\n    sigma ~ cauchy(0, 5);\n    mu ~ cauchy(0, 2.5);\n    return sigma * beta_raw + mu;\n  }\n  // ...\n}\nparameters {\n  vector[K] beta_raw;\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // ...\n  beta = center_lp(beta_raw, mu_beta, sigma_beta);\n  // ...\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#functions-implementing-change-of-variable-adjustments",
+    "href": "stan-users-guide/user-functions.html#functions-implementing-change-of-variable-adjustments",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Functions whose names end in _jacobian can use the jacobian += statement. This can be used to implement a custom change of variables for arbitrary parameters.\nFor example, this function recreates the built-in &lt;upper=x&gt; transform on real numbers:\nreal my_upper_bound_jacobian(real x, real ub) {\n  jacobian += x;\n  return ub - exp(x);\n}\nIt can be used as a replacement for real&lt;lower=ub&gt; as follows:\nfunctions {\n  // my_upper_bound_jacobian as above\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  b ~ lognormal(0, 1);\n  // ...\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#functions-acting-as-random-number-generators",
+    "href": "stan-users-guide/user-functions.html#functions-acting-as-random-number-generators",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "A user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in _rng. Giving a function a name that ends in _rng allows it to access built-in functions and user-defined functions that end in _rng, which includes all the built-in PRNG functions. Only functions ending in _rng are able access the built-in PRNG functions. The use of functions ending in _rng must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in _rng.\nFor example, the following function generates an \\(N \\times K\\) data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.\nmatrix predictors_rng(int N, int K) {\n  matrix[N, K] x;\n  for (n in 1:N) {\n    x[n, 1] = 1.0;  // intercept\n    for (k in 2:K) {\n      x[n, k] = normal_rng(0, 1);\n    }\n  }\n  return x;\n}\nThe following function defines a simulator for regression outcomes based on a data matrix x, coefficients beta, and noise scale sigma.\nvector regression_rng(vector beta, matrix x, real sigma) {\n  vector[rows(x)] y;\n  vector[rows(x)] mu;\n  mu = x * beta;\n  for (n in 1:rows(x)) {\n    y[n] = normal_rng(mu[n], sigma);\n  }\n  return y;\n}\nThese might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n  // ...\n}\ngenerated quantities {\n  matrix[N_sim, K] x_sim;\n  vector[N_sim] y_sim;\n  x_sim = predictors_rng(N_sim, K);\n  y_sim = regression_rng(beta, x_sim, sigma);\n}\nA more sophisticated simulation might fit a multivariate normal to the predictors x and use the resulting parameters to generate multivariate normal draws for x_sim.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#user-defined-probability-functions",
+    "href": "stan-users-guide/user-functions.html#user-defined-probability-functions",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Probability functions are distinguished in Stan by names ending in _lpdf for density functions and _lpmf for mass functions; in both cases, they must have real return types.\nSuppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.\nfunctions {\n  real unit_normal_lpdf(real y) {\n    return normal_lpdf(y | 0, 1);\n  }\n}\n// ...\nmodel {\n  alpha ~ unit_normal();\n  beta ~ unit_normal();\n  // ...\n}\nThe ability to use the unit_normal function as a density is keyed off its name ending in _lpdf (names ending in _lpmf for probability mass functions work the same way).\nIn general, if foo_lpdf is defined to consume \\(N + 1\\) arguments, then\ny ~ foo(theta1, ..., thetaN);\ncan be used as shorthand for\ntarget += foo_lpdf(y | theta1, ..., thetaN);\nAs with the built-in functions, the suffix _lpdf is dropped and the first argument moves to the left of the tilde symbol (~) in the distribution statement.\nFunctions ending in _lpmf (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (_lpdf) must be continuous (not an integer or integer array), whereas the first argument of a mass function (_lpmf) must be discrete (integer or integer array).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#overloading-functions",
+    "href": "stan-users-guide/user-functions.html#overloading-functions",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "As described in the reference manual function overloading is permitted in Stan, beginning in version 2.29.\nThis means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.\n\n\nOverloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.\nFor example, consider a function triple which has the following three signatures\nreal triple(real x);\ncomplex triple(complex x);\narray[] real triple(array[] real);\nOne should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect\n// The function does what it says\ntriple(3.0) == 9.0\n// It is defined reasonably for different types\ntriple(to_complex(3.0)) == to_complex(triple(3.0))\n// A container version of this function works by element\ntriple({3.0, 4.0})[0] == triple({3.0, 4.0}[0])\nNote that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.\n\n\n\nStan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.\nConsider the following function signatures\nreal foo(int a, real b);\nreal foo(real a, real b);\nGiven these, the function call foo(1.5, 2.5) is unambiguous - it must resolve to the second signature. But, the function call foo(1, 1.5) could be valid for either under Stan’s promotion rules, which allow integers to be promoted to real numbers.\nTo resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call foo(1, 1.5) would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).\nFurthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures\nreal bar(int x, real y);\nreal bar(real x, int y);\nThese signatures do not have a unique minimum number of promotions for the call bar(1, 2). Both signatures require one int to real promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.\nPromotion from integers to complex numbers is considered to be two separate promotions, first from int to real, then from real to complex. This means that integer arguments will “prefer” a signature with real types over complex types.\nFor example, consider the function signatures\nreal pop(real x);\nreal pop(complex x);\nStan will select the first signature when pop is called with an integer argument such as pop(0).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#documenting-functions.section",
+    "href": "stan-users-guide/user-functions.html#documenting-functions.section",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Functions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.\nFor example, here’s some documentation for the prediction matrix generator.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows corresponding to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\nThe comment begins with /**, ends with */, and has an asterisk (*) on each line. It uses @param followed by the argument’s identifier to document a function argument. The tag @return is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with /* and ending with */ to the Stan parser.\nFor functions that raise exceptions, exceptions can be documented using @throws.4\nFor example,\n /** ...\n * @param theta\n * @throws If any of the entries of theta is negative.\n */\nreal entropy(vector theta) {\n  // ...\n}\nUsually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#summary-of-function-types",
+    "href": "stan-users-guide/user-functions.html#summary-of-function-types",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Functions may have a void or non-void return type and they may or may not have one of the special suffixes, _lpdf, _lpmf, _lp, or _rng.\n\n\nOnly functions declared to return void may be used as statements. These are also the only functions that use return statements with no arguments.\nOnly functions declared to return non-void values may be used as expressions. These functions require return statements with arguments of a type that matches the declared return type.\n\n\n\nOnly functions ending in _lpmf or _lpdf and with return type real may be used as probability functions in distribution statements.\nOnly functions ending in _lp may access the log probability accumulator through distribution statements or target += statements. Such functions may only be used in the transformed parameters or model blocks.\nOnly functions ending in _rng may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in _rng.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#recursive-functions",
+    "href": "stan-users-guide/user-functions.html#recursive-functions",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Stan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, \\(A^n\\), which is defined for a square matrix \\(A\\) and positive integer \\(n\\) by \\[\nA^n\n=\n\\begin{cases}\n\\textrm{I} & \\quad\\text{if } n = 0, \\text{ and} \\\\\nA \\, A^{n-1} & \\quad\\text{if } n &gt; 0.\n\\end{cases}\n\\]\nwhere \\(\\textrm{I}\\) is the identity matrix. This definition can be directly translated to a recursive function definition.\nmatrix matrix_pow(matrix a, int n) {\n  if (n == 0) {\n    return diag_matrix(rep_vector(1, rows(a)));\n  } else {\n    return a *  matrix_pow(a, n - 1);\n  }\n}\nIt would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.\nelse if (n == 1) {\n  return a;\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#truncated-random-number-generation",
+    "href": "stan-users-guide/user-functions.html#truncated-random-number-generation",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "To generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then run through the inverse cumulative distribution function, \\(y = \\textrm{logit}(u)\\). If this were not already built in as logistic_rng(0, 1), it could be coded in Stan directly as\nreal standard_logistic_rng() {\n  real u = uniform_rng(0, 1);\n  real y = logit(u);\n  return y;\n}\nFollowing the same pattern, a standard normal RNG could be coded as\nreal standard_normal_rng() {\n  real u = uniform_rng(0, 1);\n  real y = inv_Phi(u);\n  return y;\n}\nthat is, \\(y = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}\\) is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function inv_Phi.\nIn order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate \\(\\textsf{normal}(\\mu, \\sigma)\\) variates, it is enough to generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then convert it to a standard normal variate, \\(z = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}(\\cdot)\\) is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, \\(y = \\mu +\n\\sigma \\times z\\). In code,\nreal my_normal_rng(real mu, real sigma) {\n  real u = uniform_rng(0, 1);\n  real z = inv_Phi(u);\n  real y = mu + sigma * z;\n  return y;\n}\nA robust version of this function would test that the arguments are finite and that sigma is non-negative, e.g.,\n  if (is_nan(mu) || is_inf(mu)) {\n    reject(\"my_normal_rng: mu must be finite; \",\n           \"found mu = \", mu);\n  }\n  if (is_nan(sigma) || is_inf(sigma) || sigma &lt; 0) {\n    reject(\"my_normal_rng: sigma must be finite and non-negative; \",\n           \"found sigma = \", sigma);\n  }\n\n\n\nOften truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.\n\n\nFor example, the following code generates a \\(\\textsf{Weibull}(\\alpha, \\sigma)\\) variate truncated below at a time \\(t\\),5\nreal weibull_lb_rng(real alpha, real sigma, real t) {\n  real p = weibull_cdf(t | alpha, sigma);   // cdf for lb\n  real u = uniform_rng(p, 1);               // unif in bounds\n  real y = sigma * (-log1m(u))^inv(alpha);  // inverse cdf\n  return y;\n}\n\n\n\nIf there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a \\(\\textsf{normal}(\\mu, \\sigma)\\) truncated to a region \\((a, b)\\), the following code suffices,\nreal normal_lub_rng(real mu, real sigma, real lb, real ub) {\n  real p_lb = normal_cdf(lb | mu, sigma);\n  real p_ub = normal_cdf(ub | mu, sigma);\n  real u = uniform_rng(p_lb, p_ub);\n  real y = mu + sigma * inv_Phi(u);\n  return y;\n}\nTo make this more robust, all variables should be tested for finiteness, sigma should be tested for positiveness, and lb and ub should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement\n  return mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),\n                                          normal_cdf(ub | mu, sigma)));\nfor readability. The names like p indicate probabilities, and p_lb and p_ub indicate the probabilities of the bounds. The variable u is clearly named as a uniform variate, and y is used to denote the variate being generated itself.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/user-functions.html#footnotes",
+    "href": "stan-users-guide/user-functions.html#footnotes",
+    "title": "User-Defined Functions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe main problem with comments is that they can be misleading, either due to misunderstandings on the programmer’s part or because the program’s behavior is modified after the comment is written. The program always behaves the way the code is written, which is why refactoring complex code into understandable units is preferable to simply adding comments.↩︎\nJust because this makes it possible to code a rejection sampler does not make it a good idea. Rejections break differentiability and the smooth exploration of the posterior. In Hamiltonian Monte Carlo, it can cause the sampler to be reduced to a diffusive random walk.↩︎\nA range of built-in validation routines is coming to Stan soon! Alternatively, the reject statement can be used to check constraints on the simplex.↩︎\nAs of Stan 2.9.0, the only way a user-defined producer will raise an exception is if a function it calls (including distribution statements) raises an exception via the reject statement.↩︎\nThe original code and impetus for including this in the manual came from the Stan forums post; by user lcomm, who also explained truncation above and below.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html",
+    "href": "stan-users-guide/time-series.html",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "Times series data come arranged in temporal order. This chapter presents two kinds of time series models, regression-like models such as autoregressive and moving average models, and hidden Markov models.\nThe Gaussian processes chapter presents Gaussian processes, which may also be used for time-series (and spatial) data.\n\n\nA first-order autoregressive model (AR(1)) with normal noise takes each point \\(y_n\\) in a sequence \\(y\\) to be generated according to \\[\ny_n \\sim \\textsf{normal}(\\alpha + \\beta y_{n-1}, \\sigma).\n\\]\nThat is, the expected value of \\(y_n\\) is \\(\\alpha + \\beta y_{n-1}\\), with noise scaled as \\(\\sigma\\).\n\n\nWith improper flat priors on the regression coefficients \\(\\alpha\\) and \\(\\beta\\) and on the positively-constrained noise scale (\\(\\sigma\\)), the Stan program for the AR(1) model is as follows.1\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  for (n in 2:N) {\n    y[n] ~ normal(alpha + beta * y[n-1], sigma);\n  }\n}\nThe first observed data point, y[1], is not modeled here because there is nothing to condition on; instead, it acts to condition y[2]. This model also uses an improper prior for sigma, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in y over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of y is available.\n\n\nAlthough perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner\nmodel {\n  y[2:N] ~ normal(alpha + beta * y[1:(N - 1)], sigma);\n}\nThe left-hand side slicing operation pulls out the last \\(N-1\\) elements and the right-hand side version pulls out the first \\(N-1\\).\n\n\n\n\nProper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-\\(t\\) distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.\nTo enforce the estimation of a stationary AR(1) process, the slope coefficient beta may be constrained with bounds as follows.\nreal&lt;lower=-1, upper=1&gt; beta;\nIn practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of beta near zero.\n\n\n\nExtending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient gamma and the following model statement.\nfor (n in 3:N) {\n  y[n] ~ normal(alpha + beta*y[n-1] + gamma*y[n-2], sigma);\n}\n\n\n\nA general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] real y;\n}\nparameters {\n  real alpha;\n  array[K] real beta;\n  real sigma;\n}\nmodel {\n  for (n in (K+1):N) {\n    real mu = alpha;\n    for (k in 1:K) {\n      mu += beta[k] * y[n-k];\n    }\n    y[n] ~ normal(mu, sigma);\n  }\n}\n\n\n\nEconometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model (Engle 1982). Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.\nThe ARCH(1) model is typically presented as the following sequence of equations, where \\(r_t\\) is the observed return at time point \\(t\\) and \\(\\mu\\), \\(\\alpha_0\\), and \\(\\alpha_1\\) are unknown regression coefficient parameters.\n\\[\\begin{align*}\nr_t        &=  \\mu + a_t \\\\\na_t        &=  \\sigma_t \\epsilon_t \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\sigma^2_t &=  \\alpha_0 + \\alpha_1 a_{t-1}^2\n\\end{align*}\\]\nIn order to ensure the noise terms \\(\\sigma^2_t\\) are positive, the scale coefficients are constrained to be positive, \\(\\alpha_0, \\alpha_1\n&gt; 0\\). To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., \\(\\alpha_1 &lt; 1\\).2\nThe ARCH(1) model may be coded directly in Stan as follows.\ndata {\n  int&lt;lower=0&gt; T;                // number of time points\n  array[T] real r;               // return at time t\n}\nparameters {\n  real mu;                       // average return\n  real&lt;lower=0&gt; alpha0;          // noise intercept\n  real&lt;lower=0, upper=1&gt; alpha1; // noise slope\n}\nmodel {\n  for (t in 2:T) {\n    r[t] ~ normal(mu, sqrt(alpha0 + alpha1\n                                    * pow(r[t - 1] - mu,2)));\n  }\n}\nThe loop in the model is defined so that the return at time \\(t=1\\) is not modeled; the model in the next section shows how to model the return at \\(t=1\\). The model can be vectorized to be more efficient; the model in the next section provides an example.\n\n\n\n\nA set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.\n\n\nThe basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time \\(t-1\\) as a predictor of volatility at time \\(t\\), defining \\[\n\\sigma^2_t = \\alpha_0 + \\alpha_1 a^2_{t-1} + \\beta_1 \\sigma^2_{t-1}.\n\\]\nTo ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy \\(\\alpha_0, \\alpha_1,\n\\beta_1 &gt; 0\\) and the slopes \\(\\alpha_1 + \\beta_1 &lt; 1\\).\ndata {\n  int&lt;lower=0&gt; T;\n  array[T] real r;\n  real&lt;lower=0&gt; sigma1;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; alpha0;\n  real&lt;lower=0, upper=1&gt; alpha1;\n  real&lt;lower=0, upper=(1-alpha1)&gt; beta1;\n}\ntransformed parameters {\n  array[T] real&lt;lower=0&gt; sigma;\n  sigma[1] = sigma1;\n  for (t in 2:T) {\n    sigma[t] = sqrt(alpha0\n                     + alpha1 * pow(r[t - 1] - mu, 2)\n                     + beta1 * pow(sigma[t - 1], 2));\n  }\n}\nmodel {\n  r ~ normal(mu, sigma);\n}\nTo get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value sigma1 for the scale of the noise at \\(t = 1\\).\nThe constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on beta1 depends on the value of alpha1.\nA transformed parameter array of non-negative values sigma is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept alpha0, a slope alpha1 for the squared difference in return from the mean at the previous time, and a slope beta1 for the previous noise scale squared. Finally, the whole regression is inside the sqrt function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.\nWith the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because r and sigma are of length T, all of the data are modeled directly.\n\n\n\n\nA moving average model uses previous errors as predictors for future outcomes. For a moving average model of order \\(Q\\), \\(\\mbox{MA}(Q)\\), there is an overall mean parameter \\(\\mu\\) and regression coefficients \\(\\theta_q\\) for previous error terms. With \\(\\epsilon_t\\) being the noise at time \\(t\\), the model for outcome \\(y_t\\) is defined by \\[\ny_t = \\mu + \\theta_1 \\epsilon_{t-1} + \\dotsb + \\theta_Q \\epsilon_{t-Q}\n+ \\epsilon_t,\n\\] with the noise term \\(\\epsilon_t\\) for outcome \\(y_t\\) modeled as normal, \\[\n\\epsilon_t \\sim \\textsf{normal}(0,\\sigma).\n\\] In a proper Bayesian model, the parameters \\(\\mu\\), \\(\\theta\\), and \\(\\sigma\\) must all be given priors.\n\n\nAn \\(\\mbox{MA}(2)\\) model can be coded in Stan as follows.\ndata {\n  int&lt;lower=3&gt; T;          // number of observations\n  vector[T] y;             // observation at time T\n}\nparameters {\n  real mu;                 // mean\n  real&lt;lower=0&gt; sigma;     // error scale\n  vector[2] theta;         // lag coefficients\n}\ntransformed parameters {\n  vector[T] epsilon;       // error terms\n  epsilon[1] = y[1] - mu;\n  epsilon[2] = y[2] - mu - theta[1] * epsilon[1];\n  for (t in 3:T) {\n    epsilon[t] = ( y[t] - mu\n                    - theta[1] * epsilon[t - 1]\n                    - theta[2] * epsilon[t - 2] );\n  }\n}\nmodel {\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 3:T) {\n    y[t] ~ normal(mu\n                  + theta[1] * epsilon[t - 1]\n                  + theta[2] * epsilon[t - 2],\n                  sigma);\n  }\n}\nThe error terms \\(\\epsilon_t\\) are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to \\(y_n\\) for \\(n &gt; Q\\). In this example, the parameters are all given Cauchy (half-Cauchy for \\(\\sigma\\)) priors, although other priors can be used just as easily.\nThis model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the \\(\\epsilon_t\\) could also be sped up by using a dot product instead of a loop.\n\n\n\nA general \\(\\mbox{MA}(Q)\\) model with a vectorized distribution statement may be defined as follows.\ndata {\n  int&lt;lower=0&gt; Q;       // num previous noise terms\n  int&lt;lower=3&gt; T;       // num observations\n  vector[T] y;          // observation at time t\n}\nparameters {\n  real mu;              // mean\n  real&lt;lower=0&gt; sigma;  // error scale\n  vector[Q] theta;      // error coeff, lag -t\n}\ntransformed parameters {\n  vector[T] epsilon;    // error term at time t\n  for (t in 1:T) {\n    epsilon[t] = y[t] - mu;\n    for (q in 1:min(t - 1, Q)) {\n      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];\n    }\n  }\n}\nmodel {\n  vector[T] eta;\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 1:T) {\n    eta[t] = mu;\n    for (q in 1:min(t - 1, Q)) {\n      eta[t] = eta[t] + theta[q] * epsilon[t - q];\n    }\n  }\n  y ~ normal(eta, sigma);\n}\nHere all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).\n\n\n\n\nAutoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.\ndata {\n  int&lt;lower=1&gt; T;            // num observations\n  array[T] real y;                 // observed outputs\n}\nparameters {\n  real mu;                   // mean coeff\n  real phi;                  // autoregression coeff\n  real theta;                // moving avg coeff\n  real&lt;lower=0&gt; sigma;       // noise scale\n}\nmodel {\n  vector[T] nu;              // prediction for time t\n  vector[T] err;             // error for time t\n  nu[1] = mu + phi * mu;     // assume err[0] == 0\n  err[1] = y[1] - nu[1];\n  for (t in 2:T) {\n    nu[t] = mu + phi * y[t - 1] + theta * err[t - 1];\n    err[t] = y[t] - nu[t];\n  }\n  mu ~ normal(0, 10);        // priors\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err ~ normal(0, sigma);    // error model\n}\nThe data are declared in the same way as the other time-series regressions and the parameters are documented in the code.\nIn the model block, the local vector nu stores the predictions and err the errors. These are computed similarly to the errors in the moving average models described in the previous section.\nThe priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.\nOften in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring err as a transformed parameter, then defining it the same way as in the model above. The vector nu could still be a local variable, only now it will be in the transformed parameter block.\nWayne Folta suggested encoding the model without local vector variables as follows.\nmodel {\n  real err;\n  mu ~ normal(0, 10);\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err = y[1] - (mu + phi * mu);\n  err ~ normal(0, sigma);\n  for (t in 2:T) {\n    err = y[t] - (mu + phi * y[t - 1] + theta * err);\n    err ~ normal(0, sigma);\n  }\n}\nThis approach to ARMA models illustrates how local variables, such as err in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.\nBoth encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just err.\n\n\nMA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint3\nreal&lt;lower=-1, upper=1&gt; theta;\nWhen the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (theta, phi) outside the \\([-1,1]\\) interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).\nFurther, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.\nreal&lt;lower=-1, upper=1&gt; phi;\n\n\n\n\nStochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time (Kim, Shephard, and Chib 1998). The data consist of mean corrected (i.e., centered) returns \\(y_t\\) on an underlying asset at \\(T\\) equally spaced time points. Kim et al. formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter \\(h_t\\) for the log volatility, along with parameters \\(\\mu\\) for the mean log volatility, and \\(\\phi\\) for the persistence of the volatility term. The variable \\(\\epsilon_t\\) represents the white-noise shock (i.e., multiplicative error) on the asset return at time \\(t\\), whereas \\(\\delta_t\\) represents the shock on volatility at time \\(t\\). \\[\\begin{align*}\ny_t        &=    \\epsilon_t \\exp(h_t / 2) \\\\\nh_{t+1}    &=    \\mu + \\phi (h_t - \\mu) + \\delta_t \\sigma \\\\\nh_1        &\\sim \\textsf{normal}\\left( \\mu, \\frac{\\sigma}{\\sqrt{1 - \\phi^2}} \\right) \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\delta_t   &\\sim \\textsf{normal}(0,1)\n\\end{align*}\\]\nRearranging the first line, \\(\\epsilon_t = y_t \\exp(-h_t / 2)\\), allowing the distribution for \\(y_t\\) to be written as \\[\ny_t \\sim \\textsf{normal}(0,\\exp(h_t/2)).\n\\] The recurrence equation for \\(h_{t+1}\\) may be combined with the scaling of \\(\\delta_t\\) to yield the distribution \\[\nh_t \\sim \\mathsf{normal}(\\mu + \\phi(h_{t-1} - \\mu), \\sigma).\n\\] This formulation can be directly encoded, as shown in the following Stan model.\ndata {\n  int&lt;lower=0&gt; T;   // # time points (equally spaced)\n  vector[T] y;      // mean corrected return at time t\n}\nparameters {\n  real mu;                     // mean log volatility\n  real&lt;lower=-1, upper=1&gt; phi; // persistence of volatility\n  real&lt;lower=0&gt; sigma;         // white noise shock scale\n  vector[T] h;                 // log volatility at time t\n}\nmodel {\n  phi ~ uniform(-1, 1);\n  sigma ~ cauchy(0, 5);\n  mu ~ cauchy(0, 10);\n  h[1] ~ normal(mu, sigma / sqrt(1 - phi * phi));\n  for (t in 2:T) {\n    h[t] ~ normal(mu + phi * (h[t - 1] -  mu), sigma);\n  }\n  for (t in 1:T) {\n    y[t] ~ normal(0, exp(h[t] / 2));\n  }\n}\nCompared to the Kim et al. formulation, the Stan model adds priors for the parameters \\(\\phi\\), \\(\\sigma\\), and \\(\\mu\\). The shock terms \\(\\epsilon_t\\) and \\(\\delta_t\\) do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.\nThe posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with \\(\\mu = -1.02\\), \\(\\phi = 0.95\\), and \\(\\sigma = 0.25\\) leads to 95% posterior intervals for \\(\\mu\\) of \\((-1.23, -0.54)\\), for \\(\\phi\\) of \\((0.82, 0.98)\\), and for \\(\\sigma\\) of \\((0.16, 0.38)\\).\nThe NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in (Hoffman and Gelman 2014). Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of \\(T\\).\nIt is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return \\(y\\) is easily vectorized to\ny ~ normal(0, exp(h / 2));\nThis speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter h_std to be declared instead of h.\nparameters {\n  // ...\n  vector[T] h_std;  // std log volatility time t\n}\nThe original value of h is then defined in a transformed parameter block.\ntransformed parameters {\n  vector[T] h = h_std * sigma;  // now h ~ normal(0, sigma)\n  h[1] /= sqrt(1 - phi * phi);  // rescale h[1]\n  h += mu;\n  for (t in 2:T) {\n    h[t] += phi * (h[t - 1] - mu);\n  }\n}\nThe first assignment rescales h_std to have a \\(\\textsf{normal}(0,\\sigma)\\) distribution and temporarily assigns it to h. The second assignment rescales h[1] so that its prior differs from that of h[2] through h[T]. The next assignment supplies a mu offset, so that h[2] through h[T] are now distributed \\(\\textsf{normal}(\\mu,\\sigma)\\); note that this shift must be done after the rescaling of h[1]. The final loop adds in the moving average so that h[2] through h[T] are appropriately modeled relative to phi and mu.\nAs a final improvement, the distribution statements for h[1] to h[T] are replaced with a single vectorized standard normal distribution statement.\nmodel {\n  // ...\n  h_std ~ std_normal();\n}\nAlthough the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.\n\n\n\nA Hidden Markov model is a probabilistic model over \\(N\\) observations \\(y_{1:N}\\) and \\(N\\) hidden states \\(z_{1:N}\\). This models is defined by the conditional distributions \\(p(y_n \\mid z_n, \\phi)\\) and \\(p(z_n \\mid z_{n-1}, \\phi)\\). Here we make the dependency on additional model parameters \\(\\phi\\) explicit. (\\(\\phi\\) may be a vector of parameters.) The complete data likelihood is then \\[\np(y, z \\mid \\phi) = \\prod_n p(y_n \\mid z_n, \\phi) p(z_n \\mid z_{n - 1}, \\phi)\n\\] When \\(z_{1:N}\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(z\\) out.\nWhen each state \\(z\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can use Stan’s suite of HMM functions to marginalize out \\(z_{1:N}\\) and compute \\[\np(y_{1:N} \\mid \\phi) = \\int_{\\mathcal Z} p(y, z \\mid \\phi) \\text d z.\n\\] We start by defining the conditional observation distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid z_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n\\Gamma_{ij} = p(z_n = j \\mid z_{n - 1} = i, \\phi).\n\\] (This is a right-stochastic matrix.) Finally, we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n\\rho_k = p(z_0 = k \\mid \\phi).\n\\] It is common practice to set \\(\\rho\\) to be the stationary distribution of the HMM, that is \\(\\rho\\) is the first eigenvector of \\(\\Gamma\\) and solves \\(\\Gamma \\rho = \\rho\\).\nAs an example, consider a three-state model with \\(K=3\\). The observations are normally distributed conditional on the HMM states with \\[\n  y_n \\sim \\text{normal}(\\mu_k, \\sigma),\n\\] where \\(\\mu = (1, 5, 9)\\) and the standard deviation \\(\\sigma\\) is the same across all observations. The model is then\ndata {\n  int N;  // Number of observations\n  array[N] real y;\n}\n\nparameters {\n  // Rows of the transition matrix\n  array[3] simplex[3] gamma_arr;\n\n  // Initial state\n  simplex[3] rho;\n\n  // Parameters of measurement model\n  vector[3] mu;\n  real&lt;lower = 0.0&gt; sigma;\n}\n\ntransformed parameters {\n  // Build transition matrix\n  matrix[3, 3] gamma;\n  for (k in 1:3) gamma[k, ] = to_row_vector(gamma_arr[k]);\n\n  // Compute the log likelihoods in each possible state\n  matrix[3, N] log_omega;\n  for (n in 1:N) {\n    for (i in 1:3) {\n      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);\n    }\n  }\n}\n\nmodel {\n  // prior\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  \n  // no explicit prior on gamma_arr, meaning we default to a\n  // uniform prior over the simplexes.\n\n  // Increment target by log p(y | mu, sigma, Gamma, rho)\n  target += hmm_marginal(log_omega, gamma, rho);\n}\nThe last function hmm_marginal takes in all the ingredients of the HMM and computes the relevant log marginal distribution, \\(\\log p(y \\mid \\phi)\\).\nIf we desire draws from the posterior distribution of \\(z\\), we use the generated quantities block and draw, for each sample \\(\\phi\\), a sample from \\(p(z \\mid y, \\phi)\\). In effect, MCMC produces draws from \\(p(\\phi \\mid y)\\) and with the draws in generated quantities, we obtain draws from \\(p(\\phi \\mid y) p(z \\mid y, \\phi) = p(z, \\phi \\mid y)\\). It is also possible to compute the posterior probbability of each hidden state, that is \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). Averagging these probabilities over all MCMC draws, we obtain \\(\\text{Pr}(z_n = k \\mid y)\\).\ngenerated quantities {\n  array[N] int latent_states = hmm_latent_rng(log_omega, gamma, rho);\n  matrix[3, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);\n}\nhmm_hidden_state_prob returns the marginal probabilities of each state, \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). This function cannot be used to compute the joint probability \\(\\text{Pr}(z \\mid \\phi, y)\\), because such calculation requires accounting for the posterior correlation between the different components of \\(z\\). Therefore, hidden_probs should not be used to obtain posterior draws. Instead, users should rely on hmm_latent_rng.\ngenerated quantities {\n   array[N] int&lt;lower=1, upper=K&gt; z = hmm_latent_rng(...fill-in params here to match example...);\n}\nThe example in this section is derived from the more detailed case study by Ben Bales: https://mc-stan.org/users/documentation/case-studies/hmm-example.html.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#autoregressive.section",
+    "href": "stan-users-guide/time-series.html#autoregressive.section",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "A first-order autoregressive model (AR(1)) with normal noise takes each point \\(y_n\\) in a sequence \\(y\\) to be generated according to \\[\ny_n \\sim \\textsf{normal}(\\alpha + \\beta y_{n-1}, \\sigma).\n\\]\nThat is, the expected value of \\(y_n\\) is \\(\\alpha + \\beta y_{n-1}\\), with noise scaled as \\(\\sigma\\).\n\n\nWith improper flat priors on the regression coefficients \\(\\alpha\\) and \\(\\beta\\) and on the positively-constrained noise scale (\\(\\sigma\\)), the Stan program for the AR(1) model is as follows.1\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  for (n in 2:N) {\n    y[n] ~ normal(alpha + beta * y[n-1], sigma);\n  }\n}\nThe first observed data point, y[1], is not modeled here because there is nothing to condition on; instead, it acts to condition y[2]. This model also uses an improper prior for sigma, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in y over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of y is available.\n\n\nAlthough perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner\nmodel {\n  y[2:N] ~ normal(alpha + beta * y[1:(N - 1)], sigma);\n}\nThe left-hand side slicing operation pulls out the last \\(N-1\\) elements and the right-hand side version pulls out the first \\(N-1\\).\n\n\n\n\nProper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-\\(t\\) distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.\nTo enforce the estimation of a stationary AR(1) process, the slope coefficient beta may be constrained with bounds as follows.\nreal&lt;lower=-1, upper=1&gt; beta;\nIn practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of beta near zero.\n\n\n\nExtending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient gamma and the following model statement.\nfor (n in 3:N) {\n  y[n] ~ normal(alpha + beta*y[n-1] + gamma*y[n-2], sigma);\n}\n\n\n\nA general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] real y;\n}\nparameters {\n  real alpha;\n  array[K] real beta;\n  real sigma;\n}\nmodel {\n  for (n in (K+1):N) {\n    real mu = alpha;\n    for (k in 1:K) {\n      mu += beta[k] * y[n-k];\n    }\n    y[n] ~ normal(mu, sigma);\n  }\n}\n\n\n\nEconometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model (Engle 1982). Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.\nThe ARCH(1) model is typically presented as the following sequence of equations, where \\(r_t\\) is the observed return at time point \\(t\\) and \\(\\mu\\), \\(\\alpha_0\\), and \\(\\alpha_1\\) are unknown regression coefficient parameters.\n\\[\\begin{align*}\nr_t        &=  \\mu + a_t \\\\\na_t        &=  \\sigma_t \\epsilon_t \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\sigma^2_t &=  \\alpha_0 + \\alpha_1 a_{t-1}^2\n\\end{align*}\\]\nIn order to ensure the noise terms \\(\\sigma^2_t\\) are positive, the scale coefficients are constrained to be positive, \\(\\alpha_0, \\alpha_1\n&gt; 0\\). To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., \\(\\alpha_1 &lt; 1\\).2\nThe ARCH(1) model may be coded directly in Stan as follows.\ndata {\n  int&lt;lower=0&gt; T;                // number of time points\n  array[T] real r;               // return at time t\n}\nparameters {\n  real mu;                       // average return\n  real&lt;lower=0&gt; alpha0;          // noise intercept\n  real&lt;lower=0, upper=1&gt; alpha1; // noise slope\n}\nmodel {\n  for (t in 2:T) {\n    r[t] ~ normal(mu, sqrt(alpha0 + alpha1\n                                    * pow(r[t - 1] - mu,2)));\n  }\n}\nThe loop in the model is defined so that the return at time \\(t=1\\) is not modeled; the model in the next section shows how to model the return at \\(t=1\\). The model can be vectorized to be more efficient; the model in the next section provides an example.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#modeling-temporal-heteroscedasticity",
+    "href": "stan-users-guide/time-series.html#modeling-temporal-heteroscedasticity",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "A set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.\n\n\nThe basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time \\(t-1\\) as a predictor of volatility at time \\(t\\), defining \\[\n\\sigma^2_t = \\alpha_0 + \\alpha_1 a^2_{t-1} + \\beta_1 \\sigma^2_{t-1}.\n\\]\nTo ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy \\(\\alpha_0, \\alpha_1,\n\\beta_1 &gt; 0\\) and the slopes \\(\\alpha_1 + \\beta_1 &lt; 1\\).\ndata {\n  int&lt;lower=0&gt; T;\n  array[T] real r;\n  real&lt;lower=0&gt; sigma1;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; alpha0;\n  real&lt;lower=0, upper=1&gt; alpha1;\n  real&lt;lower=0, upper=(1-alpha1)&gt; beta1;\n}\ntransformed parameters {\n  array[T] real&lt;lower=0&gt; sigma;\n  sigma[1] = sigma1;\n  for (t in 2:T) {\n    sigma[t] = sqrt(alpha0\n                     + alpha1 * pow(r[t - 1] - mu, 2)\n                     + beta1 * pow(sigma[t - 1], 2));\n  }\n}\nmodel {\n  r ~ normal(mu, sigma);\n}\nTo get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value sigma1 for the scale of the noise at \\(t = 1\\).\nThe constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on beta1 depends on the value of alpha1.\nA transformed parameter array of non-negative values sigma is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept alpha0, a slope alpha1 for the squared difference in return from the mean at the previous time, and a slope beta1 for the previous noise scale squared. Finally, the whole regression is inside the sqrt function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.\nWith the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because r and sigma are of length T, all of the data are modeled directly.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#moving-average-models",
+    "href": "stan-users-guide/time-series.html#moving-average-models",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "A moving average model uses previous errors as predictors for future outcomes. For a moving average model of order \\(Q\\), \\(\\mbox{MA}(Q)\\), there is an overall mean parameter \\(\\mu\\) and regression coefficients \\(\\theta_q\\) for previous error terms. With \\(\\epsilon_t\\) being the noise at time \\(t\\), the model for outcome \\(y_t\\) is defined by \\[\ny_t = \\mu + \\theta_1 \\epsilon_{t-1} + \\dotsb + \\theta_Q \\epsilon_{t-Q}\n+ \\epsilon_t,\n\\] with the noise term \\(\\epsilon_t\\) for outcome \\(y_t\\) modeled as normal, \\[\n\\epsilon_t \\sim \\textsf{normal}(0,\\sigma).\n\\] In a proper Bayesian model, the parameters \\(\\mu\\), \\(\\theta\\), and \\(\\sigma\\) must all be given priors.\n\n\nAn \\(\\mbox{MA}(2)\\) model can be coded in Stan as follows.\ndata {\n  int&lt;lower=3&gt; T;          // number of observations\n  vector[T] y;             // observation at time T\n}\nparameters {\n  real mu;                 // mean\n  real&lt;lower=0&gt; sigma;     // error scale\n  vector[2] theta;         // lag coefficients\n}\ntransformed parameters {\n  vector[T] epsilon;       // error terms\n  epsilon[1] = y[1] - mu;\n  epsilon[2] = y[2] - mu - theta[1] * epsilon[1];\n  for (t in 3:T) {\n    epsilon[t] = ( y[t] - mu\n                    - theta[1] * epsilon[t - 1]\n                    - theta[2] * epsilon[t - 2] );\n  }\n}\nmodel {\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 3:T) {\n    y[t] ~ normal(mu\n                  + theta[1] * epsilon[t - 1]\n                  + theta[2] * epsilon[t - 2],\n                  sigma);\n  }\n}\nThe error terms \\(\\epsilon_t\\) are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to \\(y_n\\) for \\(n &gt; Q\\). In this example, the parameters are all given Cauchy (half-Cauchy for \\(\\sigma\\)) priors, although other priors can be used just as easily.\nThis model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the \\(\\epsilon_t\\) could also be sped up by using a dot product instead of a loop.\n\n\n\nA general \\(\\mbox{MA}(Q)\\) model with a vectorized distribution statement may be defined as follows.\ndata {\n  int&lt;lower=0&gt; Q;       // num previous noise terms\n  int&lt;lower=3&gt; T;       // num observations\n  vector[T] y;          // observation at time t\n}\nparameters {\n  real mu;              // mean\n  real&lt;lower=0&gt; sigma;  // error scale\n  vector[Q] theta;      // error coeff, lag -t\n}\ntransformed parameters {\n  vector[T] epsilon;    // error term at time t\n  for (t in 1:T) {\n    epsilon[t] = y[t] - mu;\n    for (q in 1:min(t - 1, Q)) {\n      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];\n    }\n  }\n}\nmodel {\n  vector[T] eta;\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 1:T) {\n    eta[t] = mu;\n    for (q in 1:min(t - 1, Q)) {\n      eta[t] = eta[t] + theta[q] * epsilon[t - q];\n    }\n  }\n  y ~ normal(eta, sigma);\n}\nHere all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#autoregressive-moving-average-models",
+    "href": "stan-users-guide/time-series.html#autoregressive-moving-average-models",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "Autoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.\ndata {\n  int&lt;lower=1&gt; T;            // num observations\n  array[T] real y;                 // observed outputs\n}\nparameters {\n  real mu;                   // mean coeff\n  real phi;                  // autoregression coeff\n  real theta;                // moving avg coeff\n  real&lt;lower=0&gt; sigma;       // noise scale\n}\nmodel {\n  vector[T] nu;              // prediction for time t\n  vector[T] err;             // error for time t\n  nu[1] = mu + phi * mu;     // assume err[0] == 0\n  err[1] = y[1] - nu[1];\n  for (t in 2:T) {\n    nu[t] = mu + phi * y[t - 1] + theta * err[t - 1];\n    err[t] = y[t] - nu[t];\n  }\n  mu ~ normal(0, 10);        // priors\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err ~ normal(0, sigma);    // error model\n}\nThe data are declared in the same way as the other time-series regressions and the parameters are documented in the code.\nIn the model block, the local vector nu stores the predictions and err the errors. These are computed similarly to the errors in the moving average models described in the previous section.\nThe priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.\nOften in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring err as a transformed parameter, then defining it the same way as in the model above. The vector nu could still be a local variable, only now it will be in the transformed parameter block.\nWayne Folta suggested encoding the model without local vector variables as follows.\nmodel {\n  real err;\n  mu ~ normal(0, 10);\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err = y[1] - (mu + phi * mu);\n  err ~ normal(0, sigma);\n  for (t in 2:T) {\n    err = y[t] - (mu + phi * y[t - 1] + theta * err);\n    err ~ normal(0, sigma);\n  }\n}\nThis approach to ARMA models illustrates how local variables, such as err in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.\nBoth encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just err.\n\n\nMA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint3\nreal&lt;lower=-1, upper=1&gt; theta;\nWhen the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (theta, phi) outside the \\([-1,1]\\) interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).\nFurther, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.\nreal&lt;lower=-1, upper=1&gt; phi;",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#stochastic-volatility-models",
+    "href": "stan-users-guide/time-series.html#stochastic-volatility-models",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "Stochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time (Kim, Shephard, and Chib 1998). The data consist of mean corrected (i.e., centered) returns \\(y_t\\) on an underlying asset at \\(T\\) equally spaced time points. Kim et al. formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter \\(h_t\\) for the log volatility, along with parameters \\(\\mu\\) for the mean log volatility, and \\(\\phi\\) for the persistence of the volatility term. The variable \\(\\epsilon_t\\) represents the white-noise shock (i.e., multiplicative error) on the asset return at time \\(t\\), whereas \\(\\delta_t\\) represents the shock on volatility at time \\(t\\). \\[\\begin{align*}\ny_t        &=    \\epsilon_t \\exp(h_t / 2) \\\\\nh_{t+1}    &=    \\mu + \\phi (h_t - \\mu) + \\delta_t \\sigma \\\\\nh_1        &\\sim \\textsf{normal}\\left( \\mu, \\frac{\\sigma}{\\sqrt{1 - \\phi^2}} \\right) \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\delta_t   &\\sim \\textsf{normal}(0,1)\n\\end{align*}\\]\nRearranging the first line, \\(\\epsilon_t = y_t \\exp(-h_t / 2)\\), allowing the distribution for \\(y_t\\) to be written as \\[\ny_t \\sim \\textsf{normal}(0,\\exp(h_t/2)).\n\\] The recurrence equation for \\(h_{t+1}\\) may be combined with the scaling of \\(\\delta_t\\) to yield the distribution \\[\nh_t \\sim \\mathsf{normal}(\\mu + \\phi(h_{t-1} - \\mu), \\sigma).\n\\] This formulation can be directly encoded, as shown in the following Stan model.\ndata {\n  int&lt;lower=0&gt; T;   // # time points (equally spaced)\n  vector[T] y;      // mean corrected return at time t\n}\nparameters {\n  real mu;                     // mean log volatility\n  real&lt;lower=-1, upper=1&gt; phi; // persistence of volatility\n  real&lt;lower=0&gt; sigma;         // white noise shock scale\n  vector[T] h;                 // log volatility at time t\n}\nmodel {\n  phi ~ uniform(-1, 1);\n  sigma ~ cauchy(0, 5);\n  mu ~ cauchy(0, 10);\n  h[1] ~ normal(mu, sigma / sqrt(1 - phi * phi));\n  for (t in 2:T) {\n    h[t] ~ normal(mu + phi * (h[t - 1] -  mu), sigma);\n  }\n  for (t in 1:T) {\n    y[t] ~ normal(0, exp(h[t] / 2));\n  }\n}\nCompared to the Kim et al. formulation, the Stan model adds priors for the parameters \\(\\phi\\), \\(\\sigma\\), and \\(\\mu\\). The shock terms \\(\\epsilon_t\\) and \\(\\delta_t\\) do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.\nThe posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with \\(\\mu = -1.02\\), \\(\\phi = 0.95\\), and \\(\\sigma = 0.25\\) leads to 95% posterior intervals for \\(\\mu\\) of \\((-1.23, -0.54)\\), for \\(\\phi\\) of \\((0.82, 0.98)\\), and for \\(\\sigma\\) of \\((0.16, 0.38)\\).\nThe NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in (Hoffman and Gelman 2014). Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of \\(T\\).\nIt is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return \\(y\\) is easily vectorized to\ny ~ normal(0, exp(h / 2));\nThis speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter h_std to be declared instead of h.\nparameters {\n  // ...\n  vector[T] h_std;  // std log volatility time t\n}\nThe original value of h is then defined in a transformed parameter block.\ntransformed parameters {\n  vector[T] h = h_std * sigma;  // now h ~ normal(0, sigma)\n  h[1] /= sqrt(1 - phi * phi);  // rescale h[1]\n  h += mu;\n  for (t in 2:T) {\n    h[t] += phi * (h[t - 1] - mu);\n  }\n}\nThe first assignment rescales h_std to have a \\(\\textsf{normal}(0,\\sigma)\\) distribution and temporarily assigns it to h. The second assignment rescales h[1] so that its prior differs from that of h[2] through h[T]. The next assignment supplies a mu offset, so that h[2] through h[T] are now distributed \\(\\textsf{normal}(\\mu,\\sigma)\\); note that this shift must be done after the rescaling of h[1]. The final loop adds in the moving average so that h[2] through h[T] are appropriately modeled relative to phi and mu.\nAs a final improvement, the distribution statements for h[1] to h[T] are replaced with a single vectorized standard normal distribution statement.\nmodel {\n  // ...\n  h_std ~ std_normal();\n}\nAlthough the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#hmms.section",
+    "href": "stan-users-guide/time-series.html#hmms.section",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "A Hidden Markov model is a probabilistic model over \\(N\\) observations \\(y_{1:N}\\) and \\(N\\) hidden states \\(z_{1:N}\\). This models is defined by the conditional distributions \\(p(y_n \\mid z_n, \\phi)\\) and \\(p(z_n \\mid z_{n-1}, \\phi)\\). Here we make the dependency on additional model parameters \\(\\phi\\) explicit. (\\(\\phi\\) may be a vector of parameters.) The complete data likelihood is then \\[\np(y, z \\mid \\phi) = \\prod_n p(y_n \\mid z_n, \\phi) p(z_n \\mid z_{n - 1}, \\phi)\n\\] When \\(z_{1:N}\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(z\\) out.\nWhen each state \\(z\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can use Stan’s suite of HMM functions to marginalize out \\(z_{1:N}\\) and compute \\[\np(y_{1:N} \\mid \\phi) = \\int_{\\mathcal Z} p(y, z \\mid \\phi) \\text d z.\n\\] We start by defining the conditional observation distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid z_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n\\Gamma_{ij} = p(z_n = j \\mid z_{n - 1} = i, \\phi).\n\\] (This is a right-stochastic matrix.) Finally, we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n\\rho_k = p(z_0 = k \\mid \\phi).\n\\] It is common practice to set \\(\\rho\\) to be the stationary distribution of the HMM, that is \\(\\rho\\) is the first eigenvector of \\(\\Gamma\\) and solves \\(\\Gamma \\rho = \\rho\\).\nAs an example, consider a three-state model with \\(K=3\\). The observations are normally distributed conditional on the HMM states with \\[\n  y_n \\sim \\text{normal}(\\mu_k, \\sigma),\n\\] where \\(\\mu = (1, 5, 9)\\) and the standard deviation \\(\\sigma\\) is the same across all observations. The model is then\ndata {\n  int N;  // Number of observations\n  array[N] real y;\n}\n\nparameters {\n  // Rows of the transition matrix\n  array[3] simplex[3] gamma_arr;\n\n  // Initial state\n  simplex[3] rho;\n\n  // Parameters of measurement model\n  vector[3] mu;\n  real&lt;lower = 0.0&gt; sigma;\n}\n\ntransformed parameters {\n  // Build transition matrix\n  matrix[3, 3] gamma;\n  for (k in 1:3) gamma[k, ] = to_row_vector(gamma_arr[k]);\n\n  // Compute the log likelihoods in each possible state\n  matrix[3, N] log_omega;\n  for (n in 1:N) {\n    for (i in 1:3) {\n      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);\n    }\n  }\n}\n\nmodel {\n  // prior\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  \n  // no explicit prior on gamma_arr, meaning we default to a\n  // uniform prior over the simplexes.\n\n  // Increment target by log p(y | mu, sigma, Gamma, rho)\n  target += hmm_marginal(log_omega, gamma, rho);\n}\nThe last function hmm_marginal takes in all the ingredients of the HMM and computes the relevant log marginal distribution, \\(\\log p(y \\mid \\phi)\\).\nIf we desire draws from the posterior distribution of \\(z\\), we use the generated quantities block and draw, for each sample \\(\\phi\\), a sample from \\(p(z \\mid y, \\phi)\\). In effect, MCMC produces draws from \\(p(\\phi \\mid y)\\) and with the draws in generated quantities, we obtain draws from \\(p(\\phi \\mid y) p(z \\mid y, \\phi) = p(z, \\phi \\mid y)\\). It is also possible to compute the posterior probbability of each hidden state, that is \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). Averagging these probabilities over all MCMC draws, we obtain \\(\\text{Pr}(z_n = k \\mid y)\\).\ngenerated quantities {\n  array[N] int latent_states = hmm_latent_rng(log_omega, gamma, rho);\n  matrix[3, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);\n}\nhmm_hidden_state_prob returns the marginal probabilities of each state, \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). This function cannot be used to compute the joint probability \\(\\text{Pr}(z \\mid \\phi, y)\\), because such calculation requires accounting for the posterior correlation between the different components of \\(z\\). Therefore, hidden_probs should not be used to obtain posterior draws. Instead, users should rely on hmm_latent_rng.\ngenerated quantities {\n   array[N] int&lt;lower=1, upper=K&gt; z = hmm_latent_rng(...fill-in params here to match example...);\n}\nThe example in this section is derived from the more detailed case study by Ben Bales: https://mc-stan.org/users/documentation/case-studies/hmm-example.html.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/time-series.html#footnotes",
+    "href": "stan-users-guide/time-series.html#footnotes",
+    "title": "Time-Series Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe intercept in this model is \\(\\alpha / (1 - \\beta)\\). An alternative parameterization in terms of an intercept \\(\\gamma\\) suggested Mark Scheuerell on GitHub is \\(y_n \\sim \\textsf{normal}\\left(\\gamma + \\beta \\cdot (y_{n-1} - \\gamma), \\sigma\\right)\\).↩︎\nIn practice, it can be useful to remove the constraint to test whether a non-stationary set of coefficients provides a better fit to the data. It can also be useful to add a trend term to the model, because an unfitted trend will manifest as non-stationarity.↩︎\nThis subsection is a lightly edited comment of Jonathan Gilligan’s on GitHub; see https://github.com/stan-dev/stan/issues/1617#issuecomment-160249142.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Time-Series Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html",
+    "href": "stan-users-guide/style-guide.html",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "This chapter describes the preferred style for laying out Stan models. These are not rules of the language, but simply recommendations for laying out programs in a text editor. Although these recommendations may seem arbitrary, they are similar to those of many teams for many programming languages. Like rules for typesetting text, the goal is to achieve readability without wasting white space either vertically or horizontally. This is the style used in the Stan documentation, and should align with the auto-formatting ability of stanc3.\n\n\nThe most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.\n\n\n\nLine lengths should not exceed 80 characters.1\nThis is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.\n\n\n\nThe recommended file extension for Stan model files is .stan. Files which contain only function definitions (intended for use with #include) should be given the .stanfunctions extension. A .stanfunctions file only includes the function definition and does not require the functions{} block wrapped around the function. A simple example of usage where the function is defined and saved in the file foo.stanfunctions:\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe function foo can be accessed in the Stan program by including the path to the foo.stanfunctions file as:\nfunctions {\n  #include foo.stanfunctions;\n}\n// ...body...\nFor Stan data dump files, the recommended extension is .R, or more informatively, .data.R. For JSON output, the recommended extension is .json.\n\n\n\nThe recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (_) used as a separator. Thus it is preferred to use sigma_y, rather than the run together sigmay, camel-case sigmaY, or capitalized camel-case SigmaY. An exception is often made for terms appearing in mathematical expressions with standard names, like A for a matrix.\nAnother exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an \\(M \\times N\\) matrix \\(a\\) would look as follows.\nfor (m in 1:M) {\n  for (n in 1:N) {\n     a[m, n] = ...\n  }\n}\n\n\n\nDeclaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.\nThe following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of mu in each iteration.\nmodel {\n  array[N] real mu;\n  for (n in 1:N) {\n    mu[n] = alpha * x[n] + beta;\n    y[n] ~ normal(mu[n],sigma);\n  }\n}\nBecause variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.\nmodel {\n  for (n in 1:N) {\n    real mu;\n    mu = alpha * x[n] + beta;\n    y[n] ~ normal(mu,sigma);\n  }\n}\nThe local variable can be eliminated altogether, as follows.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(alpha * x[n] + beta, sigma);\n  }\n}\nThere is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.\n\n\nIn the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.\nmodel {\n  vector[K] mu;\n  for (n in 1:N) {\n    for (k in 1:K) {\n      mu[k] = // ...\n    }\n    y[n] ~ multi_normal(mu,Sigma);\n}\nIn this case, the vector mu will be allocated outside of both loops, and used a total of N times.\n\n\n\n\n\n\nSingle-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThe use of loops and conditionals without brackets can be dangerous. For instance, consider this program.\nfor (n in 1:N)\n  z[n] ~ normal(nu,1);\n  y[n] ~ normal(mu,1);\nBecause Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.\nfor (n in 1:N) {\n  z[n] ~ normal(nu,1);\n}\ny[n] ~ normal(mu,1);\nTherefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the else branch contains exactly one conditional. Then, it is preferred to place the following if on the same line, as in the following.\nif (x) {\n  // ...\n} else if (y) {\n  // ...\n} else {\n  // ...\n}\n\n\n\nThe preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression a + b * c is preferred to the equivalent a + (b * c) or (a + (b * c)). The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.\nSimilarly, comparison operators can usually be written with minimal bracketing, with the form y[n] &gt; 0 || x[n] != 0 preferred to the bracketed form (y[n] &gt; 0) || (x[n] != 0).\n\n\n\nVertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThis also goes for parameters blocks, transformed data blocks, which should look as follows.\ntransformed parameters {\n  real sigma;\n  // ...\n}\nThe exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.\n\n\n\n\nWhile Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued x, one should use\nif (x != 0) { ...\nin place of\nif (x) { ...\nBeyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.\n\n\n\nFunctions are laid out the same way as in languages such as Java and C++. For example,\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.\nIf function names or argument lists are long, they can be written as\nmatrix\nfunction_to_do_some_hairy_algebra(matrix thingamabob,\n                                  vector doohickey2) {\n  // ...body...\n}\nThe function starts a new line, under the type. The arguments are aligned under each other.\nFunction documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the documenting functions section.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows correspond to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\n}\nThe open comment is /**, asterisks are aligned below the first asterisk of the open comment, and the end comment */ is also aligned on the asterisk. The tags @param and @return are used to label function arguments (i.e., parameters) and return values.\n\n\n\nStan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII 0x20), line feed (ASCII 0x0A), carriage return (0x0D), and tab (0x09). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.\n\n\nEach statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax\nreal mu, sigma;\n\n\n\nStan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.\n\n\n\nStan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.\n\n\n\nUse a space after ifs. For instance, use if (x &lt; y) {..., not if(x &lt; y){ ....\n\n\n\nThere should not be space between a function name and the arguments it applies to. For instance, use normal(0, 1), not normal (0,1).\n\n\n\nThere should be spaces around binary operators. For instance, use y[1] = x, not y[1]=x, use (x + y) * z not (x+y)*z.\nUnary operators are written without a space, such as in -x, !y.\n\n\n\nAnother exception to the above rule is when the assignment operator (=) is used inside a type constraint, such as\nreal&lt;lower=0&gt; x;\nSpaces should still be used in arithmetic and following commas, as in\nreal&lt;lower=0, upper=a * x + b&gt; x;\n\n\n\nSometimes expressions are too long to fit on a single line. In that case, the recommended form is to break before an operator,2 aligning the operator to a term above to indicate scoping. For example, use the following form\nvector[J] p_distance = Phi((distance_tolerance - overshot)\n                           ./ ((x + overshot) * sigma_distance))\n                       - Phi((-overshot)\n                             ./ ((x + overshot) * sigma_distance));\nHere, the elementwise division operator (./) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (Phi).\nFor functions with multiple arguments, break after a comma and line the next argument up underneath as follows.\ny[n] ~ normal(alpha + beta * x + gamma * y,\n              pow(tau,-0.5));\n\n\n\nCommas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.\nFor example,\nnormal(alpha * x[n] + beta, sigma);\nis preferred over\nnormal(alpha * x[n] + beta,sigma);\n\n\n\nWherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.\n\n\nNewlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point 0x0A). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point 0x0D) followed by a line-feed (LF) character.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#choose-a-consistent-style",
+    "href": "stan-users-guide/style-guide.html#choose-a-consistent-style",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "The most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#line-length",
+    "href": "stan-users-guide/style-guide.html#line-length",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "Line lengths should not exceed 80 characters.1\nThis is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#file-extensions",
+    "href": "stan-users-guide/style-guide.html#file-extensions",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "The recommended file extension for Stan model files is .stan. Files which contain only function definitions (intended for use with #include) should be given the .stanfunctions extension. A .stanfunctions file only includes the function definition and does not require the functions{} block wrapped around the function. A simple example of usage where the function is defined and saved in the file foo.stanfunctions:\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe function foo can be accessed in the Stan program by including the path to the foo.stanfunctions file as:\nfunctions {\n  #include foo.stanfunctions;\n}\n// ...body...\nFor Stan data dump files, the recommended extension is .R, or more informatively, .data.R. For JSON output, the recommended extension is .json.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#variable-naming",
+    "href": "stan-users-guide/style-guide.html#variable-naming",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "The recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (_) used as a separator. Thus it is preferred to use sigma_y, rather than the run together sigmay, camel-case sigmaY, or capitalized camel-case SigmaY. An exception is often made for terms appearing in mathematical expressions with standard names, like A for a matrix.\nAnother exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an \\(M \\times N\\) matrix \\(a\\) would look as follows.\nfor (m in 1:M) {\n  for (n in 1:N) {\n     a[m, n] = ...\n  }\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#local-variable-scope",
+    "href": "stan-users-guide/style-guide.html#local-variable-scope",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "Declaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.\nThe following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of mu in each iteration.\nmodel {\n  array[N] real mu;\n  for (n in 1:N) {\n    mu[n] = alpha * x[n] + beta;\n    y[n] ~ normal(mu[n],sigma);\n  }\n}\nBecause variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.\nmodel {\n  for (n in 1:N) {\n    real mu;\n    mu = alpha * x[n] + beta;\n    y[n] ~ normal(mu,sigma);\n  }\n}\nThe local variable can be eliminated altogether, as follows.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(alpha * x[n] + beta, sigma);\n  }\n}\nThere is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.\n\n\nIn the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.\nmodel {\n  vector[K] mu;\n  for (n in 1:N) {\n    for (k in 1:K) {\n      mu[k] = // ...\n    }\n    y[n] ~ multi_normal(mu,Sigma);\n}\nIn this case, the vector mu will be allocated outside of both loops, and used a total of N times.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#parentheses-and-brackets",
+    "href": "stan-users-guide/style-guide.html#parentheses-and-brackets",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "Single-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThe use of loops and conditionals without brackets can be dangerous. For instance, consider this program.\nfor (n in 1:N)\n  z[n] ~ normal(nu,1);\n  y[n] ~ normal(mu,1);\nBecause Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.\nfor (n in 1:N) {\n  z[n] ~ normal(nu,1);\n}\ny[n] ~ normal(mu,1);\nTherefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the else branch contains exactly one conditional. Then, it is preferred to place the following if on the same line, as in the following.\nif (x) {\n  // ...\n} else if (y) {\n  // ...\n} else {\n  // ...\n}\n\n\n\nThe preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression a + b * c is preferred to the equivalent a + (b * c) or (a + (b * c)). The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.\nSimilarly, comparison operators can usually be written with minimal bracketing, with the form y[n] &gt; 0 || x[n] != 0 preferred to the bracketed form (y[n] &gt; 0) || (x[n] != 0).\n\n\n\nVertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThis also goes for parameters blocks, transformed data blocks, which should look as follows.\ntransformed parameters {\n  real sigma;\n  // ...\n}\nThe exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#conditionals",
+    "href": "stan-users-guide/style-guide.html#conditionals",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "While Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued x, one should use\nif (x != 0) { ...\nin place of\nif (x) { ...\nBeyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#functions",
+    "href": "stan-users-guide/style-guide.html#functions",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "Functions are laid out the same way as in languages such as Java and C++. For example,\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.\nIf function names or argument lists are long, they can be written as\nmatrix\nfunction_to_do_some_hairy_algebra(matrix thingamabob,\n                                  vector doohickey2) {\n  // ...body...\n}\nThe function starts a new line, under the type. The arguments are aligned under each other.\nFunction documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the documenting functions section.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows correspond to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\n}\nThe open comment is /**, asterisks are aligned below the first asterisk of the open comment, and the end comment */ is also aligned on the asterisk. The tags @param and @return are used to label function arguments (i.e., parameters) and return values.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#white-space",
+    "href": "stan-users-guide/style-guide.html#white-space",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "Stan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII 0x20), line feed (ASCII 0x0A), carriage return (0x0D), and tab (0x09). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.\n\n\nEach statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax\nreal mu, sigma;\n\n\n\nStan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.\n\n\n\nStan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.\n\n\n\nUse a space after ifs. For instance, use if (x &lt; y) {..., not if(x &lt; y){ ....\n\n\n\nThere should not be space between a function name and the arguments it applies to. For instance, use normal(0, 1), not normal (0,1).\n\n\n\nThere should be spaces around binary operators. For instance, use y[1] = x, not y[1]=x, use (x + y) * z not (x+y)*z.\nUnary operators are written without a space, such as in -x, !y.\n\n\n\nAnother exception to the above rule is when the assignment operator (=) is used inside a type constraint, such as\nreal&lt;lower=0&gt; x;\nSpaces should still be used in arithmetic and following commas, as in\nreal&lt;lower=0, upper=a * x + b&gt; x;\n\n\n\nSometimes expressions are too long to fit on a single line. In that case, the recommended form is to break before an operator,2 aligning the operator to a term above to indicate scoping. For example, use the following form\nvector[J] p_distance = Phi((distance_tolerance - overshot)\n                           ./ ((x + overshot) * sigma_distance))\n                       - Phi((-overshot)\n                             ./ ((x + overshot) * sigma_distance));\nHere, the elementwise division operator (./) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (Phi).\nFor functions with multiple arguments, break after a comma and line the next argument up underneath as follows.\ny[n] ~ normal(alpha + beta * x + gamma * y,\n              pow(tau,-0.5));\n\n\n\nCommas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.\nFor example,\nnormal(alpha * x[n] + beta, sigma);\nis preferred over\nnormal(alpha * x[n] + beta,sigma);\n\n\n\nWherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.\n\n\nNewlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point 0x0A). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point 0x0D) followed by a line-feed (LF) character.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/style-guide.html#footnotes",
+    "href": "stan-users-guide/style-guide.html#footnotes",
+    "title": "Stan Program Style Guide",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nEven 80 characters may be too many for rendering in print; for instance, in this manual, the number of code characters that fit on a line is about 65.↩︎\nThis is the usual convention in both typesetting and other programming languages. Neither R nor BUGS allows breaks before an operator because they allow newlines to signal the end of an expression or statement.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Stan Program Style Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/simulation-based-calibration.html",
+    "href": "stan-users-guide/simulation-based-calibration.html",
+    "title": "Simulation-Based Calibration Checking",
+    "section": "",
+    "text": "A Bayesian posterior is calibrated if the posterior intervals have appropriate coverage. For example, 80% intervals are expected to contain the true parameter 80% of the time. If data is generated according to a model, Bayesian posterior inference with respect to that model is calibrated by construction. Simulation-based calibration checking (SBC) exploits this property of Bayesian inference to assess the soundness of a posterior sampler. Roughly, the way it works is by simulating parameters according to the prior, then simulating data conditioned on the simulated parameters, then testing posterior calibration of the inference algorithm over independently simulated data sets. This chapter follows Talts et al. (2018), which improves on the original approach developed by Cook, Gelman, and Rubin (2006). See also Modrák et al. (2023) for further improvements.\n\n\nSuppose a Bayesian model is given in the form of a prior density \\(p(\\theta)\\) and sampling density \\(p(y \\mid \\theta).\\) Now consider a process that first simulates parameters from the prior, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta),\n\\] and then simulates data given the parameters, \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}}).\n\\] By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta).\n\\] From Bayes’s rule, it follows that for any observed (fixed) data \\(y\\), \\[\np(\\theta \\mid y) \\propto p(y, \\theta).\n\\] Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, \\[\n\\theta^{(1)}, \\ldots, \\theta^{(M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Because \\(\\theta^{\\textrm{sim}}\\) is also distributed as a draw from the posterior, the rank statistics of \\(\\theta^{\\textrm{sim}}\\) with respect to \\(\\theta^{(1)}, \\ldots \\theta^{(M)}\\) should be uniform.\nThis is one way to define calibration, because it follows that posterior intervals will have appropriate coverage (Dawid 1982; Gneiting, Balabdaoui, and Raftery 2007). If the rank of \\(\\theta^{\\textrm{sim}}\\) is uniform among the draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)},\\) then for any 90% interval selected, the probability the true value \\(\\theta^{\\textrm{sim}}\\) falls in it will also be 90%. The same goes for any other posterior interval.\n\n\n\nSuppose the Bayesian model to test has joint density \\[\np(y, \\theta) = p(y \\mid \\theta) \\cdot p(\\theta),\n\\] with data \\(y\\) and parameters \\(\\theta\\) (both are typically multivariate). Simulation-based calibration checking works by generating \\(N\\) simulated parameter and data pairs according to the joint density, \\[\n(y^{\\textrm{sim}(1)}, \\theta^{\\textrm{sim}(1)}),\n\\ldots, (y^{\\textrm{sim}(N)}, \\theta^{\\textrm{sim}(N)}),\n\\sim p(y, \\theta).\n\\] For each simulated data set \\(y^{\\textrm{sim}(n)}\\), use the algorithm to be tested to generate \\(M\\) posterior draws, which if everything is working properly, will be distributed marginally as \\[\n\\theta^{(n, 1)}, \\ldots, \\theta^{(n, M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}(n)}).\n\\] For a simulation \\(n\\) and parameter \\(k\\), the rank of the simulated parameter among the posterior draws is \\[\\begin{eqnarray*}\nr_{n, k}\n& = &\n\\textrm{rank}(\\theta_k^{\\textrm{sim}(n)},\n              (\\theta^{(n, 1)}, \\ldots, \\theta^{(n,M)}))\n\\\\[4pt]\n& = &\n\\sum_{m = 1}^M\n  \\textrm{I}[\\theta_k^{(n,m)} &lt; \\theta_k^{\\textrm{sim}(n)}].\n\\end{eqnarray*}\\] That is, the rank is the number of posterior draws \\(\\theta^{(n,m)}_k\\) that are less than the simulated draw \\(\\theta^{\\textrm{sim}(n)}_k.\\)\nIf the algorithm generates posterior draws according to the posterior, the ranks should have uniform discrete distribution from \\(0\\) to \\(M\\), so that the ranks plus one are uniformly distributed from \\(1\\) to \\(M + 1\\), \\[\nr_{n, k} + 1\n\\sim\n\\textrm{categorical}\\! \\left(\\frac{1}{M + 1}, \\ldots, \\frac{1}{M + 1}\\right).\n\\] Simulation-based calibration checking uses this expected behavior to test the calibration of each parameter of a model on simulated data. Talts et al. (2018) suggest plotting binned counts of \\(r_{1:N,\nk}\\) for different parameters \\(k\\); Säilynoja, Bürkner, and Vehtari (2022) provide a graphical test for discrete uniformity testing. Before uniformity testing the Markov chains should be thinned to remove autocorrelation as these uniformity tests assume independence (Säilynoja, Bürkner, and Vehtari 2022).\n\n\n\nRunning simulation-based calibration checking in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.\n\n\nFor illustration, a very simple model will suffice. Suppose there are two parameters \\((\\mu, \\sigma)\\) with independent priors, \\[\n\\mu \\sim \\textrm{normal}(0, 1),\n\\] and \\[\n\\sigma \\sim \\textrm{lognormal}(0, 1).\n\\] The data \\(y = y_1, \\ldots, y_N\\) is drawn conditionally independently given the parameters, \\[\ny_n \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The joint prior density is thus \\[\np(\\mu, \\sigma)\n= \\textrm{normal}(\\mu \\mid 0, 1)\n  \\cdot \\textrm{lognormal}(\\sigma \\mid 0, 1),\n\\] and the data model is \\[\np(y \\mid \\mu, \\sigma)\n= \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, \\sigma).\n\\]\nFor example, suppose the following two parameter values are drawn from the prior in the first simulation, \\[\n(\\mu^{\\textrm{sim(1)}}, \\sigma^{\\textrm{sim(1)}}) = (1.01, 0.23).\n\\] Then data \\(y^{\\textrm{sim}(1)} \\sim p(y \\mid \\mu^{\\textrm{sim(1)}},\n\\sigma^{\\textrm{sim(1)}})\\) is drawn according to the data model. Next, \\(M = 4\\) draws are taken from the posterior \\(\\mu^{(1,m)}, \\sigma^{(1,m)} \\sim p(\\mu, \\sigma \\mid y^{\\textrm{sim}(1)})\\), \\[\n\\begin{array}{r|rr}\nm & \\mu^{(1,m)} & \\sigma^{(1,m)}\n\\\\ \\hline\n1 & 1.07 & 0.33\n\\\\\n2 & -0.32 & 0.14\n\\\\\n3 & -0.99 & 0.26\n\\\\\n4 & 1.51 & 0.31\n\\end{array}\n\\] Then the comparisons on which ranks are based look as follows, \\[\n\\begin{array}{r|cc}\nm & \\textrm{I}(\\mu^{(1,m)} &lt; \\mu^{\\textrm{sim}(1)})\n& \\textrm{I}(\\sigma^{(1,m)} &lt; \\sigma^{\\textrm{sim}(1)})\n\\\\ \\hline\n1 & 0 & 0\n\\\\\n2 & 1 & 1\n\\\\\n3 & 1 & 0\n\\\\\n4 & 0 & 0\n\\end{array}\n\\] The ranks are the column sums, \\(r_{1,1} = 2\\) and \\(r_{1,2} = 1\\). Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between \\(0\\) and \\(M\\), the number of posterior draws.\n\n\n\nTo code simulation-based calibration checking in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = normal_rng(mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nTo avoid confusion with the number of simulated data sets used for simulation-based calibration checking, J is used for the number of simulated data points.\nThe model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.\n\n\n\nThe entire simulation-based calibration checking process is as follows, where\n\np(theta) is the prior density\np(y | theta) is the sampling density\nK is the number of parameters\nN is the total number of simulated data sets and fits\nM is the number of posterior draws per simulated data set\n\nSBC(p(theta), p(y | theta), K, N, M)\n------------------------------------\nfor (n in 1:N) {\n    // simulate parameters and data\n    theta(sim(n)) ~ p(theta)\n    y(sim(n)) ~ p(y | theta(sim(n)))\n\n    // posterior draws given simulated data\n    for (m in 1:M) {\n        theta(n, m) ~ p(theta | y(sim(n)))\n    }\n    // calculate rank of sim among posterior draws\n    for (k in 1:K) {\n        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))\n    }\n}\n// test uniformity of each parameter\nfor (k in 1:K) {\n    test uniformity of rank(1:N, k)\n}\n\n\n\nThe draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws (Säilynoja, Bürkner, and Vehtari 2022). Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of M, then thinning down to M draws.\n\n\n\n\nA simple, though not very highly powered, \\(\\chi^2\\)-squared test for uniformity can be formulated by binning the ranks \\(0:M\\) into \\(J\\) bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, Säilynoja, Bürkner, and Vehtari (2022) use binomial model pointiwise for the empirical cumlative distribution function and adjust to obtain simulatenous envelope to be used as graphical uniformity test.\nThe bins don’t need to be exactly the same size. In general, if \\(b_j\\) is the number of ranks that fall into bin \\(j\\) and \\(e_j\\) is the number of ranks expected to fall into bin \\(j\\) (which will be proportional to its size under uniformity), the test statistic is \\[\nX^2 = \\sum_{j = 1}^J \\frac{(b_j - e_j)^2}{e_j}.\n\\] The terms are approximately square standard normal, so that under the null hypothesis of uniformity, \\[\nX^2 \\sim \\textrm{chiSquared}(J - 1),\n\\] with the corresponding \\(p\\)-value given by the complementary cumulative distribution function (CCDF) of \\(\\textrm{chiSquared}(J - 1)\\) applied to \\(X^2\\). Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., \\(e_j \\geq 5.\\)\n\n\nBecause there are \\(M + 1\\) possible ranks, with \\(J\\) bins, it is easiest to have \\(M + 1\\) be divisible by \\(J\\). For instance, if \\(J = 20\\) and \\(M = 999\\), then there are \\(1000\\) possible ranks and an expected count in each bin of \\(\\frac{M + 1}{J} = 50.\\)\nDistributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, \\[\n\\textrm{bin}(r_{n, m}, M, J)\n= 1 + \\left\\lfloor \\frac{r_{n, m}}{(M + 1) / J} \\right\\rfloor.\n\\] For example, with \\(M = 999\\) and \\(J = 20\\), \\((M + 1) / J = 50\\). The lowest rank checks out, \\[\n\\textrm{bin}(0, 999, 20) = 1 + \\lfloor 0 / 50 \\rfloor = 1,\n\\] as does the 50th rank, \\[\n\\textrm{bin}(49, 999, 20) = 1 + \\lfloor 49 / 50 \\rfloor = 1,\n\\] and the 51st is appropriately put in the second bin, \\[\n\\textrm{bin}(50, 999, 20) = 1 + \\lfloor 50 / 50 \\rfloor = 2.\n\\] The highest rank also checks out, with \\(\\textrm{bin}(1000, 999, 20) = 50.\\)\nTo summarize, the following pseudocode computes the \\(b_j\\) values for the \\(\\chi^2\\) test or for visualization in a histogram.\nInputs: M draws, J bins, N parameters, ranks r[n, m]\nb[1:J] = 0\nfor (m in 1:M) {\n  ++b[1 + floor(r[n, m] * J / (M + 1))]\n}\nwhere the ++b[n] notation is a common form of syntactic sugar for b[n] = b[n] + 1.\nIn general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.\n\n\n\n\nThis section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.\n\n\nConsider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 1)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1)\n\\\\[4pt]\ny_{1:10} & \\sim & \\textrm{normal}(\\mu, \\sigma).\n\\end{eqnarray*}\\] The Stan program for evaluating SBC for this model is\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = student_t_rng(4, mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; I_lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nAfter running this for enough iterations so that the effective sample size is larger than \\(M\\), then thinning to \\(M\\) draws (here \\(M = 999\\)), the ranks are computed and binned, and then plotted.\n\n\n\n\n\n\nFigure 1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.\n\n\n\n\n\n\nNow consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.\n\n\n\n\n\n\nFigure 2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.\n\n\n\n\n\n\nThe example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.\nThis can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means \\(y_j\\) and standard deviations \\(\\sigma_j\\) of differences in test score after the same intervention in \\(J = 8\\) different schools. Rubin (1981) applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by Talts et al. (2018); this section replicates their results.\nThe meta-analysis model has parameters for a population mean \\(\\mu\\) and standard deviation \\(\\tau &gt; 0\\) as well as the effect \\(\\theta_j\\) of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 5)\n\\\\[4pt]\n\\tau & \\sim & \\textrm{normal}_{+}(0, 5).\n\\end{eqnarray*}\\] School level effects are modeled as normal given the population parameters, \\[\n\\theta_j \\sim \\textrm{normal}(\\mu, \\tau).\n\\] The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, \\[\ny_j \\sim \\textrm{normal}(\\theta_j, \\sigma_j).\n\\]\nThis model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.\ntransformed data {\n  real mu_sim = normal_rng(0, 5);\n  real tau_sim = abs(normal_rng(0, 5));\n  int&lt;lower=0&gt; J = 8;\n  array[J] real theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);\n  array[J] real&lt;lower=0&gt; sigma = abs(normal_rng(rep_vector(0, J), 5));\n  array[J] real y = normal_rng(theta_sim, sigma);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; tau;\n  array[J] real theta;\n}\nmodel {\n  tau ~ normal(0, 5);\n  mu ~ normal(0, 5);\n  theta ~ normal(mu, tau);\n  y ~ normal(theta, sigma);\n}\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mu_lt_sim = mu &lt; mu_sim;\n  int&lt;lower=0, upper=1&gt; tau_lt_sim = tau &lt; tau_sim;\n  int&lt;lower=0, upper=1&gt; theta1_lt_sim = theta[1] &lt; theta_sim[1];\n}\nAs usual for simulation-based calibration checking, the transformed data encodes the data-generating process using random number generators. Here, the population parameters \\(\\mu\\) and \\(\\tau\\) are first simulated, then the school-level effects \\(\\theta\\), and then finally the observed data \\(\\sigma_j\\) and \\(y_j.\\) The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only \\(\\theta_1\\) because the schools are exchangeable in the simulation.\nWhen fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.\nSBC also diagnoses the problem. Here’s the rank plots for running \\(N =\n200\\) simulations with 1000 warmup iterations and \\(M = 999\\) draws per simulation used to compute the ranks.\n\n\n\n\n\n\n\n\n\n\n\n(a) \\(\\mu\\)\n\n\n\n\n\n\n\n\n\n\n\n(b) \\(\\tau\\)\n\n\n\n\n\n\n\n\n\n\n\n(c) \\({\\theta}_1\\)\n\n\n\n\n\n\n\nFigure 3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for \\(\\theta_1\\) (Figure 3 (c)).\n\n\n\nAlthough the population mean and standard deviation \\(\\mu\\) and \\(\\tau\\) appear well calibrated, \\(\\theta_1\\) tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as \\(\\tau\\) approaches zero and the \\(\\theta_j\\) become highly constrained. The chapter on reparameterization explains how to remedy this problem and fit this kind of hierarchical model with Stan.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Simulation-Based Calibration Checking"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/simulation-based-calibration.html#bayes-is-calibrated-by-construction",
+    "href": "stan-users-guide/simulation-based-calibration.html#bayes-is-calibrated-by-construction",
+    "title": "Simulation-Based Calibration Checking",
+    "section": "",
+    "text": "Suppose a Bayesian model is given in the form of a prior density \\(p(\\theta)\\) and sampling density \\(p(y \\mid \\theta).\\) Now consider a process that first simulates parameters from the prior, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta),\n\\] and then simulates data given the parameters, \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}}).\n\\] By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta).\n\\] From Bayes’s rule, it follows that for any observed (fixed) data \\(y\\), \\[\np(\\theta \\mid y) \\propto p(y, \\theta).\n\\] Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, \\[\n\\theta^{(1)}, \\ldots, \\theta^{(M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Because \\(\\theta^{\\textrm{sim}}\\) is also distributed as a draw from the posterior, the rank statistics of \\(\\theta^{\\textrm{sim}}\\) with respect to \\(\\theta^{(1)}, \\ldots \\theta^{(M)}\\) should be uniform.\nThis is one way to define calibration, because it follows that posterior intervals will have appropriate coverage (Dawid 1982; Gneiting, Balabdaoui, and Raftery 2007). If the rank of \\(\\theta^{\\textrm{sim}}\\) is uniform among the draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)},\\) then for any 90% interval selected, the probability the true value \\(\\theta^{\\textrm{sim}}\\) falls in it will also be 90%. The same goes for any other posterior interval.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Simulation-Based Calibration Checking"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/simulation-based-calibration.html#simulation-based-calibration-checking-1",
+    "href": "stan-users-guide/simulation-based-calibration.html#simulation-based-calibration-checking-1",
+    "title": "Simulation-Based Calibration Checking",
+    "section": "",
+    "text": "Suppose the Bayesian model to test has joint density \\[\np(y, \\theta) = p(y \\mid \\theta) \\cdot p(\\theta),\n\\] with data \\(y\\) and parameters \\(\\theta\\) (both are typically multivariate). Simulation-based calibration checking works by generating \\(N\\) simulated parameter and data pairs according to the joint density, \\[\n(y^{\\textrm{sim}(1)}, \\theta^{\\textrm{sim}(1)}),\n\\ldots, (y^{\\textrm{sim}(N)}, \\theta^{\\textrm{sim}(N)}),\n\\sim p(y, \\theta).\n\\] For each simulated data set \\(y^{\\textrm{sim}(n)}\\), use the algorithm to be tested to generate \\(M\\) posterior draws, which if everything is working properly, will be distributed marginally as \\[\n\\theta^{(n, 1)}, \\ldots, \\theta^{(n, M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}(n)}).\n\\] For a simulation \\(n\\) and parameter \\(k\\), the rank of the simulated parameter among the posterior draws is \\[\\begin{eqnarray*}\nr_{n, k}\n& = &\n\\textrm{rank}(\\theta_k^{\\textrm{sim}(n)},\n              (\\theta^{(n, 1)}, \\ldots, \\theta^{(n,M)}))\n\\\\[4pt]\n& = &\n\\sum_{m = 1}^M\n  \\textrm{I}[\\theta_k^{(n,m)} &lt; \\theta_k^{\\textrm{sim}(n)}].\n\\end{eqnarray*}\\] That is, the rank is the number of posterior draws \\(\\theta^{(n,m)}_k\\) that are less than the simulated draw \\(\\theta^{\\textrm{sim}(n)}_k.\\)\nIf the algorithm generates posterior draws according to the posterior, the ranks should have uniform discrete distribution from \\(0\\) to \\(M\\), so that the ranks plus one are uniformly distributed from \\(1\\) to \\(M + 1\\), \\[\nr_{n, k} + 1\n\\sim\n\\textrm{categorical}\\! \\left(\\frac{1}{M + 1}, \\ldots, \\frac{1}{M + 1}\\right).\n\\] Simulation-based calibration checking uses this expected behavior to test the calibration of each parameter of a model on simulated data. Talts et al. (2018) suggest plotting binned counts of \\(r_{1:N,\nk}\\) for different parameters \\(k\\); Säilynoja, Bürkner, and Vehtari (2022) provide a graphical test for discrete uniformity testing. Before uniformity testing the Markov chains should be thinned to remove autocorrelation as these uniformity tests assume independence (Säilynoja, Bürkner, and Vehtari 2022).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Simulation-Based Calibration Checking"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/simulation-based-calibration.html#sbc-in-stan",
+    "href": "stan-users-guide/simulation-based-calibration.html#sbc-in-stan",
+    "title": "Simulation-Based Calibration Checking",
+    "section": "",
+    "text": "Running simulation-based calibration checking in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.\n\n\nFor illustration, a very simple model will suffice. Suppose there are two parameters \\((\\mu, \\sigma)\\) with independent priors, \\[\n\\mu \\sim \\textrm{normal}(0, 1),\n\\] and \\[\n\\sigma \\sim \\textrm{lognormal}(0, 1).\n\\] The data \\(y = y_1, \\ldots, y_N\\) is drawn conditionally independently given the parameters, \\[\ny_n \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The joint prior density is thus \\[\np(\\mu, \\sigma)\n= \\textrm{normal}(\\mu \\mid 0, 1)\n  \\cdot \\textrm{lognormal}(\\sigma \\mid 0, 1),\n\\] and the data model is \\[\np(y \\mid \\mu, \\sigma)\n= \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, \\sigma).\n\\]\nFor example, suppose the following two parameter values are drawn from the prior in the first simulation, \\[\n(\\mu^{\\textrm{sim(1)}}, \\sigma^{\\textrm{sim(1)}}) = (1.01, 0.23).\n\\] Then data \\(y^{\\textrm{sim}(1)} \\sim p(y \\mid \\mu^{\\textrm{sim(1)}},\n\\sigma^{\\textrm{sim(1)}})\\) is drawn according to the data model. Next, \\(M = 4\\) draws are taken from the posterior \\(\\mu^{(1,m)}, \\sigma^{(1,m)} \\sim p(\\mu, \\sigma \\mid y^{\\textrm{sim}(1)})\\), \\[\n\\begin{array}{r|rr}\nm & \\mu^{(1,m)} & \\sigma^{(1,m)}\n\\\\ \\hline\n1 & 1.07 & 0.33\n\\\\\n2 & -0.32 & 0.14\n\\\\\n3 & -0.99 & 0.26\n\\\\\n4 & 1.51 & 0.31\n\\end{array}\n\\] Then the comparisons on which ranks are based look as follows, \\[\n\\begin{array}{r|cc}\nm & \\textrm{I}(\\mu^{(1,m)} &lt; \\mu^{\\textrm{sim}(1)})\n& \\textrm{I}(\\sigma^{(1,m)} &lt; \\sigma^{\\textrm{sim}(1)})\n\\\\ \\hline\n1 & 0 & 0\n\\\\\n2 & 1 & 1\n\\\\\n3 & 1 & 0\n\\\\\n4 & 0 & 0\n\\end{array}\n\\] The ranks are the column sums, \\(r_{1,1} = 2\\) and \\(r_{1,2} = 1\\). Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between \\(0\\) and \\(M\\), the number of posterior draws.\n\n\n\nTo code simulation-based calibration checking in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = normal_rng(mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nTo avoid confusion with the number of simulated data sets used for simulation-based calibration checking, J is used for the number of simulated data points.\nThe model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.\n\n\n\nThe entire simulation-based calibration checking process is as follows, where\n\np(theta) is the prior density\np(y | theta) is the sampling density\nK is the number of parameters\nN is the total number of simulated data sets and fits\nM is the number of posterior draws per simulated data set\n\nSBC(p(theta), p(y | theta), K, N, M)\n------------------------------------\nfor (n in 1:N) {\n    // simulate parameters and data\n    theta(sim(n)) ~ p(theta)\n    y(sim(n)) ~ p(y | theta(sim(n)))\n\n    // posterior draws given simulated data\n    for (m in 1:M) {\n        theta(n, m) ~ p(theta | y(sim(n)))\n    }\n    // calculate rank of sim among posterior draws\n    for (k in 1:K) {\n        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))\n    }\n}\n// test uniformity of each parameter\nfor (k in 1:K) {\n    test uniformity of rank(1:N, k)\n}\n\n\n\nThe draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws (Säilynoja, Bürkner, and Vehtari 2022). Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of M, then thinning down to M draws.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Simulation-Based Calibration Checking"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/simulation-based-calibration.html#testing-uniformity",
+    "href": "stan-users-guide/simulation-based-calibration.html#testing-uniformity",
+    "title": "Simulation-Based Calibration Checking",
+    "section": "",
+    "text": "A simple, though not very highly powered, \\(\\chi^2\\)-squared test for uniformity can be formulated by binning the ranks \\(0:M\\) into \\(J\\) bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, Säilynoja, Bürkner, and Vehtari (2022) use binomial model pointiwise for the empirical cumlative distribution function and adjust to obtain simulatenous envelope to be used as graphical uniformity test.\nThe bins don’t need to be exactly the same size. In general, if \\(b_j\\) is the number of ranks that fall into bin \\(j\\) and \\(e_j\\) is the number of ranks expected to fall into bin \\(j\\) (which will be proportional to its size under uniformity), the test statistic is \\[\nX^2 = \\sum_{j = 1}^J \\frac{(b_j - e_j)^2}{e_j}.\n\\] The terms are approximately square standard normal, so that under the null hypothesis of uniformity, \\[\nX^2 \\sim \\textrm{chiSquared}(J - 1),\n\\] with the corresponding \\(p\\)-value given by the complementary cumulative distribution function (CCDF) of \\(\\textrm{chiSquared}(J - 1)\\) applied to \\(X^2\\). Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., \\(e_j \\geq 5.\\)\n\n\nBecause there are \\(M + 1\\) possible ranks, with \\(J\\) bins, it is easiest to have \\(M + 1\\) be divisible by \\(J\\). For instance, if \\(J = 20\\) and \\(M = 999\\), then there are \\(1000\\) possible ranks and an expected count in each bin of \\(\\frac{M + 1}{J} = 50.\\)\nDistributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, \\[\n\\textrm{bin}(r_{n, m}, M, J)\n= 1 + \\left\\lfloor \\frac{r_{n, m}}{(M + 1) / J} \\right\\rfloor.\n\\] For example, with \\(M = 999\\) and \\(J = 20\\), \\((M + 1) / J = 50\\). The lowest rank checks out, \\[\n\\textrm{bin}(0, 999, 20) = 1 + \\lfloor 0 / 50 \\rfloor = 1,\n\\] as does the 50th rank, \\[\n\\textrm{bin}(49, 999, 20) = 1 + \\lfloor 49 / 50 \\rfloor = 1,\n\\] and the 51st is appropriately put in the second bin, \\[\n\\textrm{bin}(50, 999, 20) = 1 + \\lfloor 50 / 50 \\rfloor = 2.\n\\] The highest rank also checks out, with \\(\\textrm{bin}(1000, 999, 20) = 50.\\)\nTo summarize, the following pseudocode computes the \\(b_j\\) values for the \\(\\chi^2\\) test or for visualization in a histogram.\nInputs: M draws, J bins, N parameters, ranks r[n, m]\nb[1:J] = 0\nfor (m in 1:M) {\n  ++b[1 + floor(r[n, m] * J / (M + 1))]\n}\nwhere the ++b[n] notation is a common form of syntactic sugar for b[n] = b[n] + 1.\nIn general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Simulation-Based Calibration Checking"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/simulation-based-calibration.html#examples-of-simulation-based-calibration-checking",
+    "href": "stan-users-guide/simulation-based-calibration.html#examples-of-simulation-based-calibration-checking",
+    "title": "Simulation-Based Calibration Checking",
+    "section": "",
+    "text": "This section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.\n\n\nConsider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 1)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1)\n\\\\[4pt]\ny_{1:10} & \\sim & \\textrm{normal}(\\mu, \\sigma).\n\\end{eqnarray*}\\] The Stan program for evaluating SBC for this model is\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = student_t_rng(4, mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; I_lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nAfter running this for enough iterations so that the effective sample size is larger than \\(M\\), then thinning to \\(M\\) draws (here \\(M = 999\\)), the ranks are computed and binned, and then plotted.\n\n\n\n\n\n\nFigure 1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.\n\n\n\n\n\n\nNow consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.\n\n\n\n\n\n\nFigure 2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.\n\n\n\n\n\n\nThe example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.\nThis can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means \\(y_j\\) and standard deviations \\(\\sigma_j\\) of differences in test score after the same intervention in \\(J = 8\\) different schools. Rubin (1981) applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by Talts et al. (2018); this section replicates their results.\nThe meta-analysis model has parameters for a population mean \\(\\mu\\) and standard deviation \\(\\tau &gt; 0\\) as well as the effect \\(\\theta_j\\) of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 5)\n\\\\[4pt]\n\\tau & \\sim & \\textrm{normal}_{+}(0, 5).\n\\end{eqnarray*}\\] School level effects are modeled as normal given the population parameters, \\[\n\\theta_j \\sim \\textrm{normal}(\\mu, \\tau).\n\\] The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, \\[\ny_j \\sim \\textrm{normal}(\\theta_j, \\sigma_j).\n\\]\nThis model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.\ntransformed data {\n  real mu_sim = normal_rng(0, 5);\n  real tau_sim = abs(normal_rng(0, 5));\n  int&lt;lower=0&gt; J = 8;\n  array[J] real theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);\n  array[J] real&lt;lower=0&gt; sigma = abs(normal_rng(rep_vector(0, J), 5));\n  array[J] real y = normal_rng(theta_sim, sigma);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; tau;\n  array[J] real theta;\n}\nmodel {\n  tau ~ normal(0, 5);\n  mu ~ normal(0, 5);\n  theta ~ normal(mu, tau);\n  y ~ normal(theta, sigma);\n}\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mu_lt_sim = mu &lt; mu_sim;\n  int&lt;lower=0, upper=1&gt; tau_lt_sim = tau &lt; tau_sim;\n  int&lt;lower=0, upper=1&gt; theta1_lt_sim = theta[1] &lt; theta_sim[1];\n}\nAs usual for simulation-based calibration checking, the transformed data encodes the data-generating process using random number generators. Here, the population parameters \\(\\mu\\) and \\(\\tau\\) are first simulated, then the school-level effects \\(\\theta\\), and then finally the observed data \\(\\sigma_j\\) and \\(y_j.\\) The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only \\(\\theta_1\\) because the schools are exchangeable in the simulation.\nWhen fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.\nSBC also diagnoses the problem. Here’s the rank plots for running \\(N =\n200\\) simulations with 1000 warmup iterations and \\(M = 999\\) draws per simulation used to compute the ranks.\n\n\n\n\n\n\n\n\n\n\n\n(a) \\(\\mu\\)\n\n\n\n\n\n\n\n\n\n\n\n(b) \\(\\tau\\)\n\n\n\n\n\n\n\n\n\n\n\n(c) \\({\\theta}_1\\)\n\n\n\n\n\n\n\nFigure 3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for \\(\\theta_1\\) (Figure 3 (c)).\n\n\n\nAlthough the population mean and standard deviation \\(\\mu\\) and \\(\\tau\\) appear well calibrated, \\(\\theta_1\\) tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as \\(\\tau\\) approaches zero and the \\(\\theta_j\\) become highly constrained. The chapter on reparameterization explains how to remedy this problem and fit this kind of hierarchical model with Stan.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Simulation-Based Calibration Checking"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html",
+    "href": "stan-users-guide/regression.html",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Stan supports regression models from simple linear regressions to multilevel generalized linear models.\n\n\nThe simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\quad\\text{where}\\quad\n\\epsilon_n \\sim \\operatorname{normal}(0,\\sigma).\n\\]\nThis is equivalent to the following sampling involving the residual, \\[\ny_n - (\\alpha + \\beta X_n) \\sim \\operatorname{normal}(0,\\sigma),\n\\] and reducing still further, to \\[\ny_n \\sim \\operatorname{normal}(\\alpha + \\beta X_n, \\, \\sigma).\n\\]\nThis latter form of the model is coded in Stan as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\nThere are N observations and for each observation, \\(n \\in N\\), we have predictor x[n] and outcome y[n]. The intercept and slope parameters are alpha and beta. The model assumes a normally distributed noise term with scale sigma. This model has improper priors for the two regression coefficients.\n\n\nThe distribution statement in the previous model is vectorized, with\ny ~ normal(alpha + beta * x, sigma);\nproviding the same model as the unvectorized version,\nfor (n in 1:N) {\n  y[n] ~ normal(alpha + beta * x[n], sigma);\n}\nIn addition to being more concise, the vectorized form is much faster.1\nIn general, Stan allows the arguments to distributions such as normal to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.\nThe other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because x is of type vector and beta of type real, the expression beta * x is of type vector. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] beta;       // coefficients for predictors\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(x * beta + alpha, sigma);  // data model\n}\nThe constraint lower=0 in the declaration of sigma constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.\nIn the model above, x is an \\(N \\times K\\) matrix of predictors and beta a \\(K\\)-vector of coefficients, so x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. These predictions line up with the outcomes in the \\(N\\)-vector y, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix x to remove the alpha parameter.\nThe distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, sigma);\n  }\n}\nWith Stan’s matrix indexing scheme, x[n] picks out row n of the matrix x; because beta is a column vector, the product x[n] * beta is a scalar of type real.\n\n\nIn the model formulation\ny ~ normal(x * beta, sigma);\nthere is no longer an intercept coefficient alpha. Instead, we have assumed that the first column of the input matrix x is a column of 1 values. This way, beta[1] plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.\n\n\n\n\n\nIn the previous example, the linear predictor can be written as \\(\\eta\n= x \\beta\\), where \\(\\eta\\) is a \\(N\\)-vector of predictions, \\(x\\) is a \\(N\n\\times K\\) matrix, and \\(\\beta\\) is a \\(K\\)-vector of coefficients. Presuming \\(N \\geq K\\), we can exploit the fact that any design matrix \\(x\\) can be decomposed using the thin QR decomposition into an orthogonal matrix \\(Q\\) and an upper-triangular matrix \\(R\\), i.e. \\(x = Q\nR\\).\nThe functions qr_thin_Q and qr_thin_R implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using qr_Q and qr_R, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the qr_thin_Q and qr_thin_R functions). In practice, it is best to write \\(x = Q^\\ast\nR^\\ast\\) where \\(Q^\\ast = Q * \\sqrt{n - 1}\\) and \\(R^\\ast =\n\\frac{1}{\\sqrt{n - 1}} R\\). Thus, we can equivalently write \\(\\eta = x\n\\beta = Q R \\beta = Q^\\ast R^\\ast \\beta\\). If we let \\(\\theta = R^\\ast\n\\beta\\), then we have \\(\\eta = Q^\\ast \\theta\\) and \\(\\beta = R^{\\ast^{-1}}\n\\theta\\). In that case, the previous Stan program becomes\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\ntransformed data {\n  matrix[N, K] Q_ast;\n  matrix[K, K] R_ast;\n  matrix[K, K] R_ast_inverse;\n  // thin and scale the QR decomposition\n  Q_ast = qr_thin_Q(x) * sqrt(N - 1);\n  R_ast = qr_thin_R(x) / sqrt(N - 1);\n  R_ast_inverse = inverse(R_ast);\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] theta;      // coefficients on Q_ast\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(Q_ast * theta + alpha, sigma);  // data model\n}\ngenerated quantities {\n  vector[K] beta;\n  beta = R_ast_inverse * theta; // coefficients on x\n}\nSince this Stan program generates equivalent predictions for \\(y\\) and the same posterior distribution for \\(\\alpha\\), \\(\\beta\\), and \\(\\sigma\\) as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:\n\nThe columns of \\(Q^\\ast\\) are orthogonal whereas the columns of \\(x\\) generally are not. Thus, it is easier for a Markov Chain to move around in \\(\\theta\\)-space than in \\(\\beta\\)-space.\nThe columns of \\(Q^\\ast\\) have the same scale whereas the columns of \\(x\\) generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps\nSince the covariance matrix for the columns of \\(Q^\\ast\\) is an identity matrix, \\(\\theta\\) typically has a reasonable scale if the units of \\(y\\) are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.\n\nConsequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever \\(K &gt; 1\\) and you do not have an informative prior on the location of \\(\\beta\\). It can also be worthwhile to subtract the mean from each column of \\(x\\) before obtaining the QR decomposition, which does not affect the posterior distribution of \\(\\theta\\) or \\(\\beta\\) but does affect \\(\\alpha\\) and allows you to interpret \\(\\alpha\\) as the expectation of \\(y\\) in a linear model.\n\n\n\nSee our general discussion of priors for tips on priors for parameters in regression models.\nLater sections discuss univariate hierarchical priors and multivariate hierarchical priors, as well as priors used to identify models.\nHowever, as described in QR-reparameterization section, if you do not have an informative prior on the location of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.\n\n\n\nThe standard approach to linear regression is to model the noise term \\(\\epsilon\\) as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-\\(t\\) distribution. To code this in Stan, the distribution distribution is changed to the following.\ndata {\n  // ...\n  real&lt;lower=0&gt; nu;\n}\n// ...\nmodel {\n  y ~ student_t(nu, alpha + beta * x, sigma);\n}\nThe degrees of freedom constant nu is specified as data.\n\n\n\nFor binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in \\((-\\infty,\\infty)\\) to probability values in \\((0,1)\\). Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both S-shaped).\nA logistic regression model with one predictor and an intercept is coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta * x);\n}\nThe noise parameter is built into the Bernoulli formulation here rather than specified directly.\nLogistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by \\[\n\\operatorname{logit}(v) = \\log \\left( \\frac{v}{1-v} \\right).\n\\]\nThe inverse of the link function appears in the model: \\[\n\\operatorname{logit}^{-1}(u) = \\texttt{inv}\\mathtt{\\_}\\texttt{logit}(u) = \\frac{1}{1 + \\exp(-u)}.\n\\]\nThe model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha \\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe formulation is also vectorized in the sense that alpha and beta are scalars and x is a vector, so that alpha   + beta * x is a vector. The vectorized formulation is equivalent to the less efficient version\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(alpha + beta * x[n]);\n}\nExpanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable\nfor (n in 1:N) {\n  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));\n}\nOther link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as\n\\[\n\\Phi(x) = \\int_{-\\infty}^x \\textsf{normal}\\left(y \\mid 0,1 \\right) \\,\\textrm{d}y.\n\\]\nThe cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.\ny[n] ~ bernoulli(Phi(alpha + beta * x[n]));\nA fast approximation to the cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi_approx.2 The approximate probit regression model may be coded with the following.\ny[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));\n\n\n\nMultiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are \\(K\\) possible outcomes for each output variable \\(y_n\\). Also suppose that there is a \\(D\\)-dimensional vector \\(x_n\\) of predictors for \\(y_n\\). The multi-logit model with \\(\\textsf{normal}(0,5)\\) priors on the coefficients is coded as follows.\ndata {\n  int K;\n  int N;\n  int D;\n  array[N] int y;\n  matrix[N, D] x;\n}\nparameters {\n  matrix[D, K] beta;\n}\nmodel {\n  matrix[N, K] x_beta = x * beta;\n\n  to_vector(beta) ~ normal(0, 5);\n\n  for (n in 1:N) {\n    y[n] ~ categorical_logit(x_beta[n]');\n\n  }\n}\nwhere x_beta[n]' is the transpose of x_beta[n]. The prior on beta is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, \\[\n\\texttt{categorical}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{categorical}\\left(y \\mid \\texttt{softmax}(\\alpha)\\right),\n\\] where \\[\n\\texttt{softmax}(u) = \\exp(u) / \\operatorname{sum}\\left(\\exp(u)\\right).\n\\]\nThe categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing\ny[n] ~ categorical(softmax(x[n] * beta));\n\n\nThe data block in the above model is defined without constraints on sizes K, N, and D or on the outcome array y. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to\nint&lt;lower=2&gt; K;\nint&lt;lower=0&gt; N;\nint&lt;lower=1&gt; D;\narray[N] int&lt;lower=1, upper=K&gt; y;\nThese constraints arise because the number of categories, K, must be at least two in order for a categorical model to be useful. The number of data items, N, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of 1:N when N is equal to zero ensures the loop’s body will not be executed. The number of predictors, D, must be at least one in order for beta * x[n] to produce an appropriate argument for softmax(). The categorical outcomes y[n] must be between 1 and K in order for the discrete sampling to be well defined.\nConstraints on data declarations are optional. Constraints on parameters declared in the parameters block, on the other hand, are not optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.\n\n\n\nBecause softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.\nAn alternative is to use \\((K-1)\\)-vectors by fixing one of them to be zero. The partially known parameters section discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use \\((K - 1)\\)-vectors\nparameters {\n  matrix[D, K - 1] beta_raw;\n}\nand then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\nwhich can then be appended to beta_raw to produce the coefficient matrix beta,\ntransformed parameters {\n  matrix[D, K] beta = append_col(beta_raw, zeros);\n}\nThe rep_vector(0, D) call creates a column vector of size D with all entries set to zero. The derived matrix beta is then defined to be the result of appending the vector zeros as a new column at the end of beta_raw; the vector zeros is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.\nThis is not the same model as using \\(K\\)-vectors as parameters, because now the prior only applies to \\((K-1)\\)-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.\n\n\n\n\nWhen there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept \\(\\alpha\\) and then a varying effect \\(\\beta_k\\) for age group \\(k\\) to make a linear predictor \\(\\alpha +\n\\beta_k\\). With this predictor, we can add a constant to \\(\\alpha\\) and subtract from each \\(\\beta_k\\) and get exactly the same likelihood.\nThe traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., \\(\\beta_1 = 0\\). With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to \\(\\beta_1\\).\nIn a Bayesian setting, a proper prior on each of the \\(\\beta\\) is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the collinearity section.\nAn alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., \\(\\sum_{k=1}^K \\beta_k = 0.\\)\nA parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the multi-logit section for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the item-response model section for details).\n\n\nAs of Stan 2.36, there is a built in sum_to_zero_vector type, which can be used as follows.\nparameters {\n  sum_to_zero_vector[K] beta;\n  // ...\n}\nThis produces a vector of size K such that sum(beta) = 0. In the unconstrained representation requires only K - 1 values because the last is determined by the first K - 1.\nPlacing a prior on beta in this parameterization, for example,\n  beta ~ normal(0, 1);\nleads to a subtly different posterior than what you would get with the same prior on an unconstrained size-K vector. As explained below, the variance is reduced.\nThe sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., \\(\\beta_K\n= -\\sum_{k=1}^{K-1} \\beta_k.\\) But that leads to high correlation among the \\(\\beta_k\\).\nThe transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; Seyboldt (2024) provides an explanation. The Stan Reference Manual provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.\n\n\nOn the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of beta,\nmodel {\n  beta ~ normal(0, inv(sqrt(1 - inv(K))));\n  // ...\n}\nThe scale component can be multiplied by sigma to produce a normal(0, sigma) prior marginally.\nTo generate distributions with marginals other than standard normal, the resulting beta may be scaled by some factor sigma and translated to some new location mu.\n\n\n\n\nAdding a prior such as \\(\\beta \\sim \\textsf{normal}(0,\\epsilon)\\) for a small \\(\\epsilon\\) will provide a kind of soft centering of a parameter vector \\(\\beta\\) by preferring, all else being equal, that \\(\\sum_{k=1}^K\n\\beta_k = 0\\). This approach is only guaranteed to roughly center if \\(\\beta\\) and the elementwise addition \\(\\beta + c\\) for a scalar constant \\(c\\) produce the same likelihood (perhaps by another vector \\(\\alpha\\) being transformed to \\(\\alpha - c\\), as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an \\(\\epsilon\\). If \\(\\epsilon\\) is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.\n\n\n\n\nOrdered regression for an outcome \\(y_n \\in \\{ 1, \\dotsc, k \\}\\) with predictors \\(x_n \\in \\mathbb{R}^D\\) is determined by a single coefficient vector \\(\\beta \\in \\mathbb{R}^D\\) along with a sequence of cutpoints \\(c \\in\n\\mathbb{R}^{K-1}\\) sorted so that \\(c_d &lt; c_{d+1}\\). The discrete output is \\(k\\) if the linear predictor \\(x_n \\beta\\) falls between \\(c_{k-1}\\) and \\(c_k\\), assuming \\(c_0 = -\\infty\\) and \\(c_K = \\infty\\). The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.\n\n\nThe ordered logistic model can be coded in Stan using the ordered data type for the cutpoints and the built-in ordered_logistic distribution.\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  matrix[N, D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  y ~ ordered_logistic(x * beta, c);\n}\nThe vector of cutpoints c is declared as ordered[K - 1], which guarantees that c[k] is less than c[k + 1].\nIf the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.\nThe equivalent model can be written using ordered_logistic_glm distribution, which can provide more efficient computation in case of higher dimensional beta.\n  y ~ ordered_logistic_glm(x,  beta, c);\n\n\nAn ordered probit model can be coded in exactly the same way by using the built-in ordered_probit distribution.\nmodel {\n  ordered_probit(x * beta, c);\n}\nThere is not yet an ordered_probit_glm distribution in Stan.\n\n\n\n\n\nThe simplest multilevel model is a hierarchical model in which the data are grouped into \\(L\\) distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients \\(\\beta\\). At the other extreme, an approach with no pooling assigns each level \\(l\\) its own coefficient vector \\(\\beta_l\\) that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.\nSuppose each binary outcome \\(y_n \\in \\{ 0, 1 \\}\\) has an associated level, \\(ll_n \\in \\{ 1, \\dotsc, L \\}\\). Each outcome will also have an associated predictor vector \\(x_n \\in \\mathbb{R}^D\\). Each level \\(l\\) gets its own coefficient vector \\(\\beta_l \\in \\mathbb{R}^D\\). The hierarchical structure involves drawing the coefficients \\(\\beta_{l,d}\n\\in \\mathbb{R}\\) from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.\nThe following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.\ndata {\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; L;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  array[N] int&lt;lower=1, upper=L&gt; ll;\n  array[N] row_vector[D] x;\n}\nparameters {\n  array[D] real mu;\n  array[D] real&lt;lower=0&gt; sigma;\n  array[L] vector[D] beta;\n}\nmodel {\n  for (d in 1:D) {\n    mu[d] ~ normal(0, 100);\n    for (l in 1:L) {\n      beta[l, d] ~ normal(mu[d], sigma[d]);\n    }\n  }\n  for (n in 1:N) {\n    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));\n  }\n}\nThe standard deviation parameter sigma gets an implicit uniform prior on \\((0,\\infty)\\) because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the regression priors section for recommendations on priors for regression coefficients and scales.\n\n\nWhere possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.\nThe first optimization vectorizes the for-loop over D as\nmu ~ normal(0, 100);\nfor (l in 1:L) {\n  beta[l] ~ normal(mu, sigma);\n}\nThe declaration of beta as an array of vectors means that the expression beta[l] denotes a vector. Although beta could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the indexing efficiency section for more information on the efficiency tradeoffs among arrays, vectors, and matrices.\nThis model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,3\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);\n}\nUnlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.\n{\n  vector[N] x_beta_ll;\n  for (n in 1:N) {\n    x_beta_ll[n] = x[n] * beta[ll[n]];\n  }\n  y ~ bernoulli_logit(x_beta_ll);\n}\nThe brackets introduce a new scope for the local variable x_beta_ll; alternatively, the variable may be declared at the top of the model block.\nIn some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.\n\n\n\n\nPriors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.\n\n\nThe fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of \\(y_n \\in\n\\mathbb{R}\\) on \\(x_n \\in \\mathbb{R}^K\\), formulated as \\[\\begin{align*}\ny_n     & \\sim \\textsf{normal}(x_n \\beta, \\sigma) \\\\\n\\beta_k & \\sim \\textsf{normal}(0,\\tau) \\\\\n\\tau    & \\sim \\textsf{Cauchy}(0,2.5) \\\\\n\\end{align*}\\]\nIn this case, as \\(\\tau \\rightarrow 0\\) and \\(\\beta_k \\rightarrow 0\\), the posterior density \\[ p(\\beta,\\tau,\\sigma|y,x) \\propto p(y|x,\\beta,\\tau,\\sigma) \\] grows without bound. See the Neal’s funnel density, which has similar behavior.\nThere is obviously no MLE estimate for \\(\\beta,\\tau,\\sigma\\) in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by Chung et al. (2013) is to use a gamma distribution as a prior, such as \\[\n\\sigma \\sim \\textsf{Gamma}(2, 1/A),\n\\] for a reasonably large value of \\(A\\), such as \\(A = 10\\).\n\n\n\n\nItem-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see Gelman and Hill (2007, pps. 314–320) for a textbook introduction to hierarchical IRT models and Curtis (2010) for encodings of a range of IRT models in BUGS.\n\n\nThe data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.\ndata {\n  int&lt;lower=1&gt; J;                     // number of students\n  int&lt;lower=1&gt; K;                     // number of questions\n  int&lt;lower=1&gt; N;                     // number of observations\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // student for observation n\n  array[N] int&lt;lower=1, upper=K&gt; kk;  // question for observation n\n  array[N] int&lt;lower=0, upper=1&gt; y;   // correctness for observation n\n}\nThis declares a total of N student-question pairs in the data set, where each n in 1:N indexes a binary observation y[n] of the correctness of the answer of student jj[n] on question kk[n].\nThe prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.\n\n\n\nThe 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).\nThe model parameters are declared as follows.\nparameters {\n  real delta;            // mean student ability\n  array[J] real alpha;   // ability of student j - mean ability\n  array[K] real beta;    // difficulty of question k\n}\nThe parameter alpha[J] is the ability coefficient for student j and beta[k] is the difficulty coefficient for question k. The non-standard parameterization used here also includes an intercept term delta, which represents the average student’s response to the average question.4\nThe model itself is as follows.\nmodel {\n  alpha ~ std_normal();         // informative true prior\n  beta ~ std_normal();          // informative true prior\n  delta ~ normal(0.75, 1);      // informative true prior\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);\n  }\n}\nThis model uses the logit-parameterized Bernoulli distribution, where \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe key to understanding it is the term inside the bernoulli_logit distribution, from which it follows that \\[\n\\Pr[y_n = 1] = \\operatorname{logit}^{-1}\\left(\\alpha_{jj[n]} - \\beta_{kk[n]}\n+ \\delta\\right).\n\\]\nThe model suffers from additive identifiability issues without the priors. For example, adding a term \\(\\xi\\) to each \\(\\alpha_j\\) and \\(\\beta_k\\) results in the same predictions. The use of priors for \\(\\alpha\\) and \\(\\beta\\) located at 0 identifies the parameters; see Gelman and Hill (2007) for a discussion of identifiability issues and alternative approaches to identification.\nFor testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.\n\n\n\nThe simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.\nparameters {\n  real mu_beta;                // mean question difficulty\n  vector[J] alpha;             // ability for j - mean\n  vector[K] beta;              // difficulty for k\n  vector&lt;lower=0&gt;[K] gamma;    // discrimination of k\n  real&lt;lower=0&gt; sigma_beta;    // scale of difficulties\n  real&lt;lower=0&gt; sigma_gamma;   // scale of log discrimination\n}\nThe parameters should be clearer after the model definition.\nmodel {\n  alpha ~ std_normal();\n  beta ~ normal(0, sigma_beta);\n  gamma ~ lognormal(0, sigma_gamma);\n  mu_beta ~ cauchy(0, 5);\n  sigma_beta ~ cauchy(0, 5);\n  sigma_gamma ~ cauchy(0, 5);\n  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));\n}\nThe std_normal function is used here, defined by \\[\n\\texttt{std}\\mathtt{\\_}\\texttt{normal}(y)\n=\n\\textsf{normal}\\left(y \\mid 0, 1\\right).\n\\]\nThe distribution statement is also vectorized using elementwise multiplication; it is equivalent to\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(gamma[kk[n]]\n                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));\n}\nThe 2PL model is similar to the 1PL model, with the additional parameter gamma[k] modeling how discriminative question k is. If gamma[k] is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter gamma[k] is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.\nThe model is parameterized here with student abilities alpha being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the problematic posteriors chapter for further discussion of identifiability. The difficulty and discrimination parameters beta and gamma then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as\nbeta ~ normal(0, 5);\ngamma ~ lognormal(0, 2);\nThe point is that the alpha determines the scale and location and beta and gamma are allowed to float.\nThe beta parameter is here given a non-centered parameterization, with parameter mu_beta serving as the mean beta location. An alternative would’ve been to take:\nbeta ~ normal(mu_beta, sigma_beta);\nand\ny[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));\nNon-centered parameterizations tend to be more efficient in hierarchical models; see the reparameterization section for more information on non-centered reparameterizations.\nThe intercept term mu_beta can’t itself be modeled hierarchically, so it is given a weakly informative \\(\\textsf{Cauchy}(0,5)\\) prior. Similarly, the scale terms, sigma_beta, and sigma_gamma, are given half-Cauchy priors. As mentioned earlier, the scale and location for alpha are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to \\((0,\\infty)\\) by their declarations.\n\n\n\n\n\n\nOne application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the collinearity section for a simple example of the problems this poses for estimation.\nThe non-identifiability is resolved by providing a standard normal (i.e., \\(\\textsf{normal}(0,1)\\)) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.\n\n\n\nAnother case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.\n\n\n\nIn a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.\nSimilar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.\n\n\n\n\nIn hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.\nAs an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.\n\n\nGelman and Hill (2007, chap. 13, Chapter 17) provide a discussion of a hierarchical model with \\(N\\) individuals organized into \\(J\\) groups. Each individual has a predictor row vector \\(x_n\\) of size \\(K\\); to unify the notation, they assume that \\(x_{n,1} = 1\\) is a fixed “intercept” predictor. To encode group membership, they assume individual \\(n\\) belongs to group \\(jj[n] \\in \\{ 1, \\dotsc, J \\}\\). Each individual \\(n\\) also has an observed outcome \\(y_n\\) taking on real values.\n\n\nThe model is a linear regression with slope and intercept coefficients varying by group, so that \\(\\beta_j\\) is the coefficient \\(K\\)-vector for group \\(j\\). The data model for individual \\(n\\) is then just \\[\ny_n \\sim \\textsf{normal}(x_n \\, \\beta_{jj[n]}, \\, \\sigma)\n\\quad\\text{for}\\quad n \\in \\{ 1, \\dotsc, N \\}.\n\\]\n\n\n\nGelman and Hill model the coefficient vectors \\(\\beta_j\\) as being drawn from a multivariate distribution with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\), \\[\n\\beta_j \\sim \\textsf{multivariate normal}(\\mu_j, \\, \\Sigma)\n\\quad\\text{for}\\quad j \\in \\{ 1, \\dotsc, J \\}.\n\\]\nBelow, we discuss the full model of Gelman and Hill, which uses group-level predictors to model \\(\\mu\\); for now, we assume \\(\\mu\\) is a simple vector parameter.\n\n\n\nFor hierarchical modeling, the group-level mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\) must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as \\[\n\\mu_j \\sim \\textsf{normal}(0,5).\n\\] If more is known about the expected coefficient values \\(\\beta_{j, k}\\), this information can be incorporated into the prior for \\(\\mu_j\\).\nFor the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling\nIn Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define \\[\n\\Sigma = \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau) \\times \\Omega \\times \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau),\n\\] where \\(\\Omega\\) is a correlation matrix and \\(\\tau\\) is the vector of coefficient scales. This mapping from scale vector \\(\\tau\\) and correlation matrix \\(\\Omega\\) can be inverted, using \\[\n\\tau_k = \\sqrt{\\Sigma_{k,k}}\n\\quad\\textsf{and}\\quad\n\\Omega_{i, j} = \\frac{\\Sigma_{i, j}}{\\tau_i \\, \\tau_j}.\n\\]\nThe components of the scale vector \\(\\tau\\) can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as \\[\n\\tau_k \\sim \\textsf{Cauchy}(0, 2.5)\n\\quad\\text{for}\\quad k \\in \\{ 1, \\dotsc, K \\}\n\\quad\\text{constrained\\ by}\\quad \\tau_k &gt; 0.\n\\] As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for \\(\\tau\\). For large numbers of exchangeable coefficients, the components of \\(\\tau\\) itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.\nOur final recommendation is to give the correlation matrix \\(\\Omega\\) an LKJ prior with shape \\(\\eta \\geq 1\\),5\n\\[\n\\Omega \\sim \\textsf{LKJCorr}(\\eta).\n\\]\nThe LKJ correlation distribution is defined by \\[\n\\textsf{LKJCorr}\\left(\\Sigma \\mid \\eta\\right)\n\\propto\n\\operatorname{det}\\left(\\Sigma\\right)^{\\eta - 1}.\n\\]\nThe basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For \\(\\eta = 1\\), the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.\nFor \\(\\eta &gt; 1\\), the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For \\(\\eta &lt; 1\\), it increasingly concentrates mass in the other direction, i.e., favoring more correlation.\nThe LKJ prior may thus be used to control the expected amount of correlation among the parameters \\(\\beta_j\\). For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see Barnard, McCulloch, and Meng (2000).\n\n\n\nTo complete Gelman and Hill’s model, suppose each group \\(j \\in \\{ 1, \\dotsc, J \\}\\) is supplied with an \\(L\\)-dimensional row-vector of group-level predictors \\(u_j\\). The prior mean for the \\(\\beta_j\\) can then itself be modeled as a regression, using an \\(L\\)-dimensional coefficient vector \\(\\gamma\\). The prior for the group-level coefficients then becomes \\[\n\\beta_j \\sim \\textsf{multivariate normal}(u_j \\, \\gamma, \\Sigma)\n\\]\nThe group-level coefficients \\(\\gamma\\) may themselves be given independent weakly informative priors, such as \\[\n\\gamma_l \\sim \\textsf{normal}(0,5).\n\\] As usual, information about the group-level means should be incorporated into this prior.\n\n\n\nThe Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.\ndata {\n  int&lt;lower=0&gt; N;              // num individuals\n  int&lt;lower=1&gt; K;              // num ind predictors\n  int&lt;lower=1&gt; J;              // num groups\n  int&lt;lower=1&gt; L;              // num group predictors\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // group for individual\n  matrix[N, K] x;              // individual predictors\n  array[J] row_vector[L] u;    // group predictors\n  vector[N] y;                 // outcomes\n}\nparameters {\n  corr_matrix[K] Omega;        // prior correlation\n  vector&lt;lower=0&gt;[K] tau;      // prior scale\n  matrix[L, K] gamma;          // group coeffs\n  array[J] vector[K] beta;     // indiv coeffs by group\n  real&lt;lower=0&gt; sigma;         // prediction error scale\n}\nmodel {\n  tau ~ cauchy(0, 2.5);\n  Omega ~ lkj_corr(2);\n  to_vector(gamma) ~ normal(0, 5);\n  {\n    array[J] row_vector[K] u_gamma;\n    for (j in 1:J) {\n      u_gamma[j] = u[j] * gamma;\n    }\n    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));\n  }\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n  }\n}\nThe hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix Omega and scale vector tau are more natural to inspect in the output; to output Sigma, define it as a transformed parameter. The function quad_form_diag is defined so that quad_form_diag(Sigma, tau) is equivalent to diag_matrix(tau) * Sigma * diag_matrix(tau), where diag_matrix(tau) returns the matrix with tau on the diagonal and zeroes off diagonal; the version using quad_form_diag should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.\n\n\n\nThe code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:\nfor (n in 1:N) {\n  y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n}\nwith the vectorized distribution statement:\n{\n  vector[N] x_beta_jj;\n  for (n in 1:N) {\n    x_beta_jj[n] = x[n] * beta[jj[n]];\n  }\n  y ~ normal(x_beta_jj, sigma);\n}\nThe outer brackets create a local scope in which to define the variable x_beta_jj, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.\nAlthough it is tempting to redeclare beta and include a revised model block distribution statement,\nparameters {\n  matrix[J, K] beta;\n// ...\n}\nmodel {\n  y ~ normal(rows_dot_product(x, beta[jj]), sigma);\n  // ...\n}\nthis fails because it breaks the vectorization for beta,6\nbeta ~ multi_normal(...);\nwhich requires beta to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because rows_dot_product cannot do much optimization in and of itself because there are no shared computations.\nThe code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.\n{\n  matrix[K, K] Sigma_beta;\n  Sigma_beta = quad_form_diag(Omega, tau);\n  for (j in 1:J) {\n    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);\n  }\n}\nIn this example, the covariance matrix Sigma_beta is defined as a local variable so as not to have to repeat the quadratic form computation \\(J\\) times. This vectorization can be combined with the Cholesky-factor optimization in the next section.\n\n\n\nThe multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.\ndata {\n  matrix[L, J] u;              // group predictors transposed\n  // ...\n}\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  matrix[K, L] gamma;\n  // ...\n}\ntransformed parameters {\n  matrix[K, J] beta;\n  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  // ...\n}\nThe data variable u was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with gamma and beta, to minimize the number of transposition operations. The new parameter L_Omega is the Cholesky factor of the original correlation matrix Omega, so that\nOmega = L_Omega * L_Omega'\nThe prior scale vector tau is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,\nSigma_beta\n  = quad_form_diag(Omega, tau)\n  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'\nwhere the diagonal pre-multiply compound operation is defined by\ndiag_pre_multiply(a, b) = diag_matrix(a) * b\nThe new variable z is declared as a matrix, the entries of which are given independent standard normal priors; the to_vector operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of z being a \\(K\\)-variate normal random vector with the identity as covariance matrix. Therefore, multiplying z by the Cholesky factor of the covariance matrix and adding the mean (u * gamma)' produces a beta distributed as in the original model, where the variance is, letting \\(L = \\mathrm{diag}(\\tau)\\,\\Omega_L\\),\n\\[\n\\begin{aligned}\n\\mathbb{V}[\\beta] &= \\mathbb{E}\\big((L \\, z)(L \\, z)^\\top) \\\\\n&= \\mathbb{E}\\big((L \\, z \\, z^\\top \\, L^\\top) \\\\\n&= L \\, \\mathbb{E}(z \\, z^\\top) \\, L^\\top \\\\\n&= L \\, L^\\top =(\\mathrm{diag}(\\tau)\\,\\Omega_L)\\,(\\mathrm{diag}(\\tau)\\,\\Omega_L)^\\top \\\\\n&= \\mathrm{diag}(\\tau)\\,\\Omega\\,\\mathrm{diag}(\\tau) \\\\\n&= \\Sigma.\n\\end{aligned}\n\\] Where we have used the linearity of expectations (line 2 to 3), the definition of \\(\\Omega = \\Omega_L \\, \\Omega_L^\\top\\), and the fact that \\(\\mathbb{E}(z \\, z^\\top) = I\\) since \\(z \\sim \\mathcal{N}(0, I)\\).\nOmitting the remaining data declarations, which are the same as before with the exception of u, the optimized model is as follows.\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0, upper=pi() / 2&gt;[K] tau_unif;  // prior scale\n  matrix[K, L] gamma;                        // group coeffs\n  real&lt;lower=0&gt; sigma;                       // prediction error scale\n}\ntransformed parameters {\n  vector&lt;lower=0&gt;[K] tau = 2.5 * tan(tau_unif);\n  matrix[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  vector[N] mu;\n  for(n in 1:N) {\n    mu[n] = x[n, ] * beta[, jj[n]];\n  }\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  to_vector(gamma) ~ normal(0, 5);\n  y ~ normal(mu, sigma);\n}\nThis model also reparameterizes the prior scale tau to avoid potential problems with the heavy tails of the Cauchy distribution. The statement tau_unif ~ uniform(0, pi() / 2) can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.\n\n\n\n\n\nStan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).\n\n\nAs a simple example, the following linear regression provides the same setup for estimating the coefficients beta as in our very first example, using y for the N observations and x for the N predictor vectors. The model parameters and model for observations are exactly the same as before.\nTo make predictions, we need to be given the number of predictions, N_new, and their predictor matrix, x_new. The predictions themselves are modeled as a parameter y_new. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector y_new and prediction matrix x_new.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=0&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n\n  int&lt;lower=0&gt; N_new;\n  matrix[N_new, K] x_new;\n}\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n\n  vector[N_new] y_new;                  // predictions\n}\nmodel {\n  y ~ normal(x * beta, sigma);          // observed model\n\n  y_new ~ normal(x_new * beta, sigma);  // prediction model\n}\n\n\n\nWhere possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.\n// ...data as above...\n\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(x * beta, sigma);\n}\ngenerated quantities {\n  vector[N_new] y_new;\n  for (n in 1:N_new) {\n    y_new[n] = normal_rng(x_new[n] * beta, sigma);\n  }\n}\nNow the data are just as before, but the parameter y_new is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.\n\n\nIt is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of \\(2^{31}\\), must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by \\(2^{31} - 1\\) in Stan.\n\n\n\n\n\nMost regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.\nThis section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.\n\n\nThe first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors (Zellner 1962; Greene 2011).\nThe model is easy to write down as a regression, \\[\\begin{align*}\ny_n        &=  x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nwhere \\(x_n\\) is a \\(J\\)-row-vector of predictors (\\(x\\) is an \\((N \\times\nJ)\\) matrix), \\(y_n\\) is a \\(K\\)-vector of observations, \\(\\beta\\) is a \\((K\n\\times J)\\) matrix of regression coefficients (vector \\(\\beta_k\\) holds coefficients for outcome \\(k\\)), and \\(\\Sigma\\) is covariance matrix governing the error. As usual, the intercept can be rolled into \\(x\\) as a column of ones.\nThe basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; J;\n  int&lt;lower=0&gt; N;\n  array[N] vector[J] x;\n  array[N] vector[K] y;\n}\nparameters {\n  matrix[K, J] beta;\n  cov_matrix[K] Sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n  }\n  y ~ multi_normal(mu, Sigma);\n}\nFor efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.\nFollowing the advice in the multivariate hierarchical priors section, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.\n// ...\nparameters {\n  matrix[K, J] beta;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0&gt;[K] L_sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  matrix[K, K] L_Sigma;\n\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n\n  }\n\n  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);\n\n  to_vector(beta) ~ normal(0, 5);\n  L_Omega ~ lkj_corr_cholesky(4);\n  L_sigma ~ cauchy(0, 2.5);\n\n  y ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix beta to a vector.\nIf required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.\n\n\n\nThe multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.\nThe observations \\(y_n\\) are \\(D\\)-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations \\(y_n\\) are based on latent values \\(z_n\\) drawn from a seemingly unrelated regression model (see the previous section), \\[\\begin{align*}\nz_n        &=     x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim  \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nThese are then put through the step function to produce a \\(K\\)-vector \\(z_n\\) of boolean values with elements defined by \\[\ny_{n, k} = \\operatorname{I}\\left(z_{n, k} &gt; 0\\right),\n\\] where \\(\\operatorname{I}()\\) is the indicator function taking the value 1 if its argument is true and 0 otherwise.\nUnlike in the seemingly unrelated regressions case, here the covariance matrix \\(\\Sigma\\) has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see Greene (2011)).\nMultivariate probit regression can be coded in Stan using the trick introduced by Albert and Chib (1993), where the underlying continuous value vectors \\(y_n\\) are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector \\(z\\) in two parts, based on whether the corresponding value of \\(y\\) is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.\nFirst, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in \\(y\\).\nfunctions {\n  int sum2d(array[,] int a) {\n    int s = 0;\n    for (i in 1:size(a)) {\n      s += sum(a[i]);\n    }\n    return s;\n  }\n}\nThe function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.\nThe data declaration block is much like for the seemingly unrelated regressions, but the observations y are now integers constrained to be 0 or 1.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  array[N, D] int&lt;lower=0, upper=1&gt; y;\n  array[N] vector[K] x;\n}\nAfter declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array y into positive and negative components, keeping track of indexes so that z can be easily reassembled in the transformed parameters block.\ntransformed data {\n  int&lt;lower=0&gt; N_pos;\n  array[sum2d(y)] int&lt;lower=1, upper=N&gt; n_pos;\n  array[size(n_pos)] int&lt;lower=1, upper=D&gt; d_pos;\n  int&lt;lower=0&gt; N_neg;\n  array[(N * D) - size(n_pos)] int&lt;lower=1, upper=N&gt; n_neg;\n  array[size(n_neg)] int&lt;lower=1, upper=D&gt; d_neg;\n\n  N_pos = size(n_pos);\n  N_neg = size(n_neg);\n  {\n    int i;\n    int j;\n    i = 1;\n    j = 1;\n    for (n in 1:N) {\n      for (d in 1:D) {\n        if (y[n, d] == 1) {\n          n_pos[i] = n;\n          d_pos[i] = d;\n          i += 1;\n        } else {\n          n_neg[j] = n;\n          d_neg[j] = d;\n          j += 1;\n        }\n      }\n    }\n  }\n}\nThe variables N_pos and N_neg are set to the number of true (1) and number of false (0) observations in y. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.\nThe parameters are declared as follows.\nparameters {\n  matrix[D, K] beta;\n  cholesky_factor_corr[D] L_Omega;\n  vector&lt;lower=0&gt;[N_pos] z_pos;\n  vector&lt;upper=0&gt;[N_neg] z_neg;\n}\nThese include the regression coefficients beta and the Cholesky factor of the correlation matrix, L_Omega. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).\nThe critical part of the parameter declaration is that the latent real value \\(z\\) is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct \\(z\\).\ntransformed parameters {\n  array[N] vector[D] z;\n  for (n in 1:N_pos) {\n    z[n_pos[n], d_pos[n]] = z_pos[n];\n  }\n  for (n in 1:N_neg) {\n    z[n_neg[n], d_neg[n]] = z_neg[n];\n  }\n}\nAt this point, the model is simple, pretty much recreating the seemingly unrelated regression.\nmodel {\n  L_Omega ~ lkj_corr_cholesky(4);\n  to_vector(beta) ~ normal(0, 5);\n  {\n    array[N] vector[D] beta_x;\n    for (n in 1:N) {\n      beta_x[n] = beta * x[n];\n    }\n    z ~ multi_normal_cholesky(beta_x, L_Omega);\n  }\n}\nThis simple form of model is made possible by the Albert and Chib-style constraints on z.\nFinally, the correlation matrix itself can be put back together in the generated quantities block if desired.\ngenerated quantities {\n  corr_matrix[D] Omega;\n  Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\nThe same could be done for the seemingly unrelated regressions in the previous section.\n\n\n\n\nThe main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.\n\n\nConsider predicting unobserved outcomes using linear regression. Given predictors \\(x_1, \\dotsc, x_N\\) and observed outcomes \\(y_1, \\dotsc, y_N\\), and assuming a standard linear regression with intercept \\(\\alpha\\), slope \\(\\beta\\), and error scale \\(\\sigma\\), along with improper uniform priors, the posterior over the parameters given \\(x\\) and \\(y\\) is \\[\np\\left(\\alpha, \\beta, \\sigma \\mid x, y \\right)\n\\propto\n\\prod_{n=1}^N\n  \\textsf{normal}\\left(y_n \\mid \\alpha + \\beta x_n, \\sigma\\right).\n\\]\nFor this model, the posterior predictive inference for a new outcome \\(\\tilde{y}_m\\) given a predictor \\(\\tilde{x}_m\\), conditioned on the observed data \\(x\\) and \\(y\\), is \\[\np\\left(\\tilde{y}_n \\mid \\tilde{x}_n, x, y\\right)\n= \\int_{(\\alpha,\\beta,\\sigma)}\n  \\textsf{normal}\\left(\\tilde{y}_n \\mid \\alpha + \\beta \\tilde{x}_n, \\sigma\\right)\n  \\times\n  p\\left(\\alpha, \\beta, \\sigma \\mid x, y\\right)\n  \\,\\textrm{d}(\\alpha,\\beta,\\sigma).\n\\]\nTo code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde;\n  for (n in 1:N_tilde) {\n    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);\n  }\n}\nGiven observed predictors \\(x\\) and outcomes \\(y\\), y_tilde will be drawn according to \\(p\\left(\\tilde{y} \\mid \\tilde{x}, y, x\\right)\\). This means that, for example, the posterior mean for y_tilde is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).\n\n\n\nA good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors \\(x\\) are used in place of new predictors \\(\\tilde{x}\\) for unobserved outcomes. If the model fits the data well, the predictions for \\(\\tilde{y}\\) based on \\(x\\) should match the observed data \\(y\\).\nTo code posterior predictive checks in Stan requires only a slight modification of the prediction code to use \\(x\\) and \\(N\\) in place of \\(\\tilde{x}\\) and \\(\\tilde{N}\\),\ngenerated quantities {\n  vector[N] y_tilde;\n  for (n in 1:N) {\n    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);\n  }\n}\nGelman et al. (2013) recommend choosing several posterior draws \\(\\tilde{y}^{(1)}, \\dotsc, \\tilde{y}^{(M)}\\) and plotting each of them alongside the data \\(y\\) that was actually observed. If the model fits well, the simulated \\(\\tilde{y}\\) will look like the actual data \\(y\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#linear-regression",
+    "href": "stan-users-guide/regression.html#linear-regression",
+    "title": "Regression Models",
+    "section": "",
+    "text": "The simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\quad\\text{where}\\quad\n\\epsilon_n \\sim \\operatorname{normal}(0,\\sigma).\n\\]\nThis is equivalent to the following sampling involving the residual, \\[\ny_n - (\\alpha + \\beta X_n) \\sim \\operatorname{normal}(0,\\sigma),\n\\] and reducing still further, to \\[\ny_n \\sim \\operatorname{normal}(\\alpha + \\beta X_n, \\, \\sigma).\n\\]\nThis latter form of the model is coded in Stan as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\nThere are N observations and for each observation, \\(n \\in N\\), we have predictor x[n] and outcome y[n]. The intercept and slope parameters are alpha and beta. The model assumes a normally distributed noise term with scale sigma. This model has improper priors for the two regression coefficients.\n\n\nThe distribution statement in the previous model is vectorized, with\ny ~ normal(alpha + beta * x, sigma);\nproviding the same model as the unvectorized version,\nfor (n in 1:N) {\n  y[n] ~ normal(alpha + beta * x[n], sigma);\n}\nIn addition to being more concise, the vectorized form is much faster.1\nIn general, Stan allows the arguments to distributions such as normal to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.\nThe other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because x is of type vector and beta of type real, the expression beta * x is of type vector. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] beta;       // coefficients for predictors\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(x * beta + alpha, sigma);  // data model\n}\nThe constraint lower=0 in the declaration of sigma constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.\nIn the model above, x is an \\(N \\times K\\) matrix of predictors and beta a \\(K\\)-vector of coefficients, so x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. These predictions line up with the outcomes in the \\(N\\)-vector y, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix x to remove the alpha parameter.\nThe distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, sigma);\n  }\n}\nWith Stan’s matrix indexing scheme, x[n] picks out row n of the matrix x; because beta is a column vector, the product x[n] * beta is a scalar of type real.\n\n\nIn the model formulation\ny ~ normal(x * beta, sigma);\nthere is no longer an intercept coefficient alpha. Instead, we have assumed that the first column of the input matrix x is a column of 1 values. This way, beta[1] plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#QR-reparameterization.section",
+    "href": "stan-users-guide/regression.html#QR-reparameterization.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "In the previous example, the linear predictor can be written as \\(\\eta\n= x \\beta\\), where \\(\\eta\\) is a \\(N\\)-vector of predictions, \\(x\\) is a \\(N\n\\times K\\) matrix, and \\(\\beta\\) is a \\(K\\)-vector of coefficients. Presuming \\(N \\geq K\\), we can exploit the fact that any design matrix \\(x\\) can be decomposed using the thin QR decomposition into an orthogonal matrix \\(Q\\) and an upper-triangular matrix \\(R\\), i.e. \\(x = Q\nR\\).\nThe functions qr_thin_Q and qr_thin_R implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using qr_Q and qr_R, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the qr_thin_Q and qr_thin_R functions). In practice, it is best to write \\(x = Q^\\ast\nR^\\ast\\) where \\(Q^\\ast = Q * \\sqrt{n - 1}\\) and \\(R^\\ast =\n\\frac{1}{\\sqrt{n - 1}} R\\). Thus, we can equivalently write \\(\\eta = x\n\\beta = Q R \\beta = Q^\\ast R^\\ast \\beta\\). If we let \\(\\theta = R^\\ast\n\\beta\\), then we have \\(\\eta = Q^\\ast \\theta\\) and \\(\\beta = R^{\\ast^{-1}}\n\\theta\\). In that case, the previous Stan program becomes\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\ntransformed data {\n  matrix[N, K] Q_ast;\n  matrix[K, K] R_ast;\n  matrix[K, K] R_ast_inverse;\n  // thin and scale the QR decomposition\n  Q_ast = qr_thin_Q(x) * sqrt(N - 1);\n  R_ast = qr_thin_R(x) / sqrt(N - 1);\n  R_ast_inverse = inverse(R_ast);\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] theta;      // coefficients on Q_ast\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(Q_ast * theta + alpha, sigma);  // data model\n}\ngenerated quantities {\n  vector[K] beta;\n  beta = R_ast_inverse * theta; // coefficients on x\n}\nSince this Stan program generates equivalent predictions for \\(y\\) and the same posterior distribution for \\(\\alpha\\), \\(\\beta\\), and \\(\\sigma\\) as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:\n\nThe columns of \\(Q^\\ast\\) are orthogonal whereas the columns of \\(x\\) generally are not. Thus, it is easier for a Markov Chain to move around in \\(\\theta\\)-space than in \\(\\beta\\)-space.\nThe columns of \\(Q^\\ast\\) have the same scale whereas the columns of \\(x\\) generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps\nSince the covariance matrix for the columns of \\(Q^\\ast\\) is an identity matrix, \\(\\theta\\) typically has a reasonable scale if the units of \\(y\\) are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.\n\nConsequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever \\(K &gt; 1\\) and you do not have an informative prior on the location of \\(\\beta\\). It can also be worthwhile to subtract the mean from each column of \\(x\\) before obtaining the QR decomposition, which does not affect the posterior distribution of \\(\\theta\\) or \\(\\beta\\) but does affect \\(\\alpha\\) and allows you to interpret \\(\\alpha\\) as the expectation of \\(y\\) in a linear model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#regression-priors.section",
+    "href": "stan-users-guide/regression.html#regression-priors.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "See our general discussion of priors for tips on priors for parameters in regression models.\nLater sections discuss univariate hierarchical priors and multivariate hierarchical priors, as well as priors used to identify models.\nHowever, as described in QR-reparameterization section, if you do not have an informative prior on the location of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#robust-noise-models",
+    "href": "stan-users-guide/regression.html#robust-noise-models",
+    "title": "Regression Models",
+    "section": "",
+    "text": "The standard approach to linear regression is to model the noise term \\(\\epsilon\\) as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-\\(t\\) distribution. To code this in Stan, the distribution distribution is changed to the following.\ndata {\n  // ...\n  real&lt;lower=0&gt; nu;\n}\n// ...\nmodel {\n  y ~ student_t(nu, alpha + beta * x, sigma);\n}\nThe degrees of freedom constant nu is specified as data.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#logistic-probit-regression.section",
+    "href": "stan-users-guide/regression.html#logistic-probit-regression.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "For binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in \\((-\\infty,\\infty)\\) to probability values in \\((0,1)\\). Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both S-shaped).\nA logistic regression model with one predictor and an intercept is coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta * x);\n}\nThe noise parameter is built into the Bernoulli formulation here rather than specified directly.\nLogistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by \\[\n\\operatorname{logit}(v) = \\log \\left( \\frac{v}{1-v} \\right).\n\\]\nThe inverse of the link function appears in the model: \\[\n\\operatorname{logit}^{-1}(u) = \\texttt{inv}\\mathtt{\\_}\\texttt{logit}(u) = \\frac{1}{1 + \\exp(-u)}.\n\\]\nThe model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha \\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe formulation is also vectorized in the sense that alpha and beta are scalars and x is a vector, so that alpha   + beta * x is a vector. The vectorized formulation is equivalent to the less efficient version\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(alpha + beta * x[n]);\n}\nExpanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable\nfor (n in 1:N) {\n  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));\n}\nOther link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as\n\\[\n\\Phi(x) = \\int_{-\\infty}^x \\textsf{normal}\\left(y \\mid 0,1 \\right) \\,\\textrm{d}y.\n\\]\nThe cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.\ny[n] ~ bernoulli(Phi(alpha + beta * x[n]));\nA fast approximation to the cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi_approx.2 The approximate probit regression model may be coded with the following.\ny[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#multi-logit.section",
+    "href": "stan-users-guide/regression.html#multi-logit.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Multiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are \\(K\\) possible outcomes for each output variable \\(y_n\\). Also suppose that there is a \\(D\\)-dimensional vector \\(x_n\\) of predictors for \\(y_n\\). The multi-logit model with \\(\\textsf{normal}(0,5)\\) priors on the coefficients is coded as follows.\ndata {\n  int K;\n  int N;\n  int D;\n  array[N] int y;\n  matrix[N, D] x;\n}\nparameters {\n  matrix[D, K] beta;\n}\nmodel {\n  matrix[N, K] x_beta = x * beta;\n\n  to_vector(beta) ~ normal(0, 5);\n\n  for (n in 1:N) {\n    y[n] ~ categorical_logit(x_beta[n]');\n\n  }\n}\nwhere x_beta[n]' is the transpose of x_beta[n]. The prior on beta is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, \\[\n\\texttt{categorical}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{categorical}\\left(y \\mid \\texttt{softmax}(\\alpha)\\right),\n\\] where \\[\n\\texttt{softmax}(u) = \\exp(u) / \\operatorname{sum}\\left(\\exp(u)\\right).\n\\]\nThe categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing\ny[n] ~ categorical(softmax(x[n] * beta));\n\n\nThe data block in the above model is defined without constraints on sizes K, N, and D or on the outcome array y. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to\nint&lt;lower=2&gt; K;\nint&lt;lower=0&gt; N;\nint&lt;lower=1&gt; D;\narray[N] int&lt;lower=1, upper=K&gt; y;\nThese constraints arise because the number of categories, K, must be at least two in order for a categorical model to be useful. The number of data items, N, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of 1:N when N is equal to zero ensures the loop’s body will not be executed. The number of predictors, D, must be at least one in order for beta * x[n] to produce an appropriate argument for softmax(). The categorical outcomes y[n] must be between 1 and K in order for the discrete sampling to be well defined.\nConstraints on data declarations are optional. Constraints on parameters declared in the parameters block, on the other hand, are not optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.\n\n\n\nBecause softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.\nAn alternative is to use \\((K-1)\\)-vectors by fixing one of them to be zero. The partially known parameters section discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use \\((K - 1)\\)-vectors\nparameters {\n  matrix[D, K - 1] beta_raw;\n}\nand then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\nwhich can then be appended to beta_raw to produce the coefficient matrix beta,\ntransformed parameters {\n  matrix[D, K] beta = append_col(beta_raw, zeros);\n}\nThe rep_vector(0, D) call creates a column vector of size D with all entries set to zero. The derived matrix beta is then defined to be the result of appending the vector zeros as a new column at the end of beta_raw; the vector zeros is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.\nThis is not the same model as using \\(K\\)-vectors as parameters, because now the prior only applies to \\((K-1)\\)-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#parameterizing-centered-vectors",
+    "href": "stan-users-guide/regression.html#parameterizing-centered-vectors",
+    "title": "Regression Models",
+    "section": "",
+    "text": "When there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept \\(\\alpha\\) and then a varying effect \\(\\beta_k\\) for age group \\(k\\) to make a linear predictor \\(\\alpha +\n\\beta_k\\). With this predictor, we can add a constant to \\(\\alpha\\) and subtract from each \\(\\beta_k\\) and get exactly the same likelihood.\nThe traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., \\(\\beta_1 = 0\\). With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to \\(\\beta_1\\).\nIn a Bayesian setting, a proper prior on each of the \\(\\beta\\) is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the collinearity section.\nAn alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., \\(\\sum_{k=1}^K \\beta_k = 0.\\)\nA parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the multi-logit section for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the item-response model section for details).\n\n\nAs of Stan 2.36, there is a built in sum_to_zero_vector type, which can be used as follows.\nparameters {\n  sum_to_zero_vector[K] beta;\n  // ...\n}\nThis produces a vector of size K such that sum(beta) = 0. In the unconstrained representation requires only K - 1 values because the last is determined by the first K - 1.\nPlacing a prior on beta in this parameterization, for example,\n  beta ~ normal(0, 1);\nleads to a subtly different posterior than what you would get with the same prior on an unconstrained size-K vector. As explained below, the variance is reduced.\nThe sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., \\(\\beta_K\n= -\\sum_{k=1}^{K-1} \\beta_k.\\) But that leads to high correlation among the \\(\\beta_k\\).\nThe transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; Seyboldt (2024) provides an explanation. The Stan Reference Manual provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.\n\n\nOn the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of beta,\nmodel {\n  beta ~ normal(0, inv(sqrt(1 - inv(K))));\n  // ...\n}\nThe scale component can be multiplied by sigma to produce a normal(0, sigma) prior marginally.\nTo generate distributions with marginals other than standard normal, the resulting beta may be scaled by some factor sigma and translated to some new location mu.\n\n\n\n\nAdding a prior such as \\(\\beta \\sim \\textsf{normal}(0,\\epsilon)\\) for a small \\(\\epsilon\\) will provide a kind of soft centering of a parameter vector \\(\\beta\\) by preferring, all else being equal, that \\(\\sum_{k=1}^K\n\\beta_k = 0\\). This approach is only guaranteed to roughly center if \\(\\beta\\) and the elementwise addition \\(\\beta + c\\) for a scalar constant \\(c\\) produce the same likelihood (perhaps by another vector \\(\\alpha\\) being transformed to \\(\\alpha - c\\), as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an \\(\\epsilon\\). If \\(\\epsilon\\) is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#ordered-logistic.section",
+    "href": "stan-users-guide/regression.html#ordered-logistic.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Ordered regression for an outcome \\(y_n \\in \\{ 1, \\dotsc, k \\}\\) with predictors \\(x_n \\in \\mathbb{R}^D\\) is determined by a single coefficient vector \\(\\beta \\in \\mathbb{R}^D\\) along with a sequence of cutpoints \\(c \\in\n\\mathbb{R}^{K-1}\\) sorted so that \\(c_d &lt; c_{d+1}\\). The discrete output is \\(k\\) if the linear predictor \\(x_n \\beta\\) falls between \\(c_{k-1}\\) and \\(c_k\\), assuming \\(c_0 = -\\infty\\) and \\(c_K = \\infty\\). The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.\n\n\nThe ordered logistic model can be coded in Stan using the ordered data type for the cutpoints and the built-in ordered_logistic distribution.\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  matrix[N, D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  y ~ ordered_logistic(x * beta, c);\n}\nThe vector of cutpoints c is declared as ordered[K - 1], which guarantees that c[k] is less than c[k + 1].\nIf the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.\nThe equivalent model can be written using ordered_logistic_glm distribution, which can provide more efficient computation in case of higher dimensional beta.\n  y ~ ordered_logistic_glm(x,  beta, c);\n\n\nAn ordered probit model can be coded in exactly the same way by using the built-in ordered_probit distribution.\nmodel {\n  ordered_probit(x * beta, c);\n}\nThere is not yet an ordered_probit_glm distribution in Stan.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#hierarchical-regression",
+    "href": "stan-users-guide/regression.html#hierarchical-regression",
+    "title": "Regression Models",
+    "section": "",
+    "text": "The simplest multilevel model is a hierarchical model in which the data are grouped into \\(L\\) distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients \\(\\beta\\). At the other extreme, an approach with no pooling assigns each level \\(l\\) its own coefficient vector \\(\\beta_l\\) that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.\nSuppose each binary outcome \\(y_n \\in \\{ 0, 1 \\}\\) has an associated level, \\(ll_n \\in \\{ 1, \\dotsc, L \\}\\). Each outcome will also have an associated predictor vector \\(x_n \\in \\mathbb{R}^D\\). Each level \\(l\\) gets its own coefficient vector \\(\\beta_l \\in \\mathbb{R}^D\\). The hierarchical structure involves drawing the coefficients \\(\\beta_{l,d}\n\\in \\mathbb{R}\\) from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.\nThe following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.\ndata {\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; L;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  array[N] int&lt;lower=1, upper=L&gt; ll;\n  array[N] row_vector[D] x;\n}\nparameters {\n  array[D] real mu;\n  array[D] real&lt;lower=0&gt; sigma;\n  array[L] vector[D] beta;\n}\nmodel {\n  for (d in 1:D) {\n    mu[d] ~ normal(0, 100);\n    for (l in 1:L) {\n      beta[l, d] ~ normal(mu[d], sigma[d]);\n    }\n  }\n  for (n in 1:N) {\n    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));\n  }\n}\nThe standard deviation parameter sigma gets an implicit uniform prior on \\((0,\\infty)\\) because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the regression priors section for recommendations on priors for regression coefficients and scales.\n\n\nWhere possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.\nThe first optimization vectorizes the for-loop over D as\nmu ~ normal(0, 100);\nfor (l in 1:L) {\n  beta[l] ~ normal(mu, sigma);\n}\nThe declaration of beta as an array of vectors means that the expression beta[l] denotes a vector. Although beta could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the indexing efficiency section for more information on the efficiency tradeoffs among arrays, vectors, and matrices.\nThis model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,3\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);\n}\nUnlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.\n{\n  vector[N] x_beta_ll;\n  for (n in 1:N) {\n    x_beta_ll[n] = x[n] * beta[ll[n]];\n  }\n  y ~ bernoulli_logit(x_beta_ll);\n}\nThe brackets introduce a new scope for the local variable x_beta_ll; alternatively, the variable may be declared at the top of the model block.\nIn some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#hierarchical-priors.section",
+    "href": "stan-users-guide/regression.html#hierarchical-priors.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Priors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.\n\n\nThe fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of \\(y_n \\in\n\\mathbb{R}\\) on \\(x_n \\in \\mathbb{R}^K\\), formulated as \\[\\begin{align*}\ny_n     & \\sim \\textsf{normal}(x_n \\beta, \\sigma) \\\\\n\\beta_k & \\sim \\textsf{normal}(0,\\tau) \\\\\n\\tau    & \\sim \\textsf{Cauchy}(0,2.5) \\\\\n\\end{align*}\\]\nIn this case, as \\(\\tau \\rightarrow 0\\) and \\(\\beta_k \\rightarrow 0\\), the posterior density \\[ p(\\beta,\\tau,\\sigma|y,x) \\propto p(y|x,\\beta,\\tau,\\sigma) \\] grows without bound. See the Neal’s funnel density, which has similar behavior.\nThere is obviously no MLE estimate for \\(\\beta,\\tau,\\sigma\\) in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by Chung et al. (2013) is to use a gamma distribution as a prior, such as \\[\n\\sigma \\sim \\textsf{Gamma}(2, 1/A),\n\\] for a reasonably large value of \\(A\\), such as \\(A = 10\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#item-response-models.section",
+    "href": "stan-users-guide/regression.html#item-response-models.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Item-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see Gelman and Hill (2007, pps. 314–320) for a textbook introduction to hierarchical IRT models and Curtis (2010) for encodings of a range of IRT models in BUGS.\n\n\nThe data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.\ndata {\n  int&lt;lower=1&gt; J;                     // number of students\n  int&lt;lower=1&gt; K;                     // number of questions\n  int&lt;lower=1&gt; N;                     // number of observations\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // student for observation n\n  array[N] int&lt;lower=1, upper=K&gt; kk;  // question for observation n\n  array[N] int&lt;lower=0, upper=1&gt; y;   // correctness for observation n\n}\nThis declares a total of N student-question pairs in the data set, where each n in 1:N indexes a binary observation y[n] of the correctness of the answer of student jj[n] on question kk[n].\nThe prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.\n\n\n\nThe 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).\nThe model parameters are declared as follows.\nparameters {\n  real delta;            // mean student ability\n  array[J] real alpha;   // ability of student j - mean ability\n  array[K] real beta;    // difficulty of question k\n}\nThe parameter alpha[J] is the ability coefficient for student j and beta[k] is the difficulty coefficient for question k. The non-standard parameterization used here also includes an intercept term delta, which represents the average student’s response to the average question.4\nThe model itself is as follows.\nmodel {\n  alpha ~ std_normal();         // informative true prior\n  beta ~ std_normal();          // informative true prior\n  delta ~ normal(0.75, 1);      // informative true prior\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);\n  }\n}\nThis model uses the logit-parameterized Bernoulli distribution, where \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe key to understanding it is the term inside the bernoulli_logit distribution, from which it follows that \\[\n\\Pr[y_n = 1] = \\operatorname{logit}^{-1}\\left(\\alpha_{jj[n]} - \\beta_{kk[n]}\n+ \\delta\\right).\n\\]\nThe model suffers from additive identifiability issues without the priors. For example, adding a term \\(\\xi\\) to each \\(\\alpha_j\\) and \\(\\beta_k\\) results in the same predictions. The use of priors for \\(\\alpha\\) and \\(\\beta\\) located at 0 identifies the parameters; see Gelman and Hill (2007) for a discussion of identifiability issues and alternative approaches to identification.\nFor testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.\n\n\n\nThe simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.\nparameters {\n  real mu_beta;                // mean question difficulty\n  vector[J] alpha;             // ability for j - mean\n  vector[K] beta;              // difficulty for k\n  vector&lt;lower=0&gt;[K] gamma;    // discrimination of k\n  real&lt;lower=0&gt; sigma_beta;    // scale of difficulties\n  real&lt;lower=0&gt; sigma_gamma;   // scale of log discrimination\n}\nThe parameters should be clearer after the model definition.\nmodel {\n  alpha ~ std_normal();\n  beta ~ normal(0, sigma_beta);\n  gamma ~ lognormal(0, sigma_gamma);\n  mu_beta ~ cauchy(0, 5);\n  sigma_beta ~ cauchy(0, 5);\n  sigma_gamma ~ cauchy(0, 5);\n  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));\n}\nThe std_normal function is used here, defined by \\[\n\\texttt{std}\\mathtt{\\_}\\texttt{normal}(y)\n=\n\\textsf{normal}\\left(y \\mid 0, 1\\right).\n\\]\nThe distribution statement is also vectorized using elementwise multiplication; it is equivalent to\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(gamma[kk[n]]\n                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));\n}\nThe 2PL model is similar to the 1PL model, with the additional parameter gamma[k] modeling how discriminative question k is. If gamma[k] is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter gamma[k] is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.\nThe model is parameterized here with student abilities alpha being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the problematic posteriors chapter for further discussion of identifiability. The difficulty and discrimination parameters beta and gamma then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as\nbeta ~ normal(0, 5);\ngamma ~ lognormal(0, 2);\nThe point is that the alpha determines the scale and location and beta and gamma are allowed to float.\nThe beta parameter is here given a non-centered parameterization, with parameter mu_beta serving as the mean beta location. An alternative would’ve been to take:\nbeta ~ normal(mu_beta, sigma_beta);\nand\ny[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));\nNon-centered parameterizations tend to be more efficient in hierarchical models; see the reparameterization section for more information on non-centered reparameterizations.\nThe intercept term mu_beta can’t itself be modeled hierarchically, so it is given a weakly informative \\(\\textsf{Cauchy}(0,5)\\) prior. Similarly, the scale terms, sigma_beta, and sigma_gamma, are given half-Cauchy priors. As mentioned earlier, the scale and location for alpha are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to \\((0,\\infty)\\) by their declarations.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#priors-for-identification.section",
+    "href": "stan-users-guide/regression.html#priors-for-identification.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "One application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the collinearity section for a simple example of the problems this poses for estimation.\nThe non-identifiability is resolved by providing a standard normal (i.e., \\(\\textsf{normal}(0,1)\\)) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.\n\n\n\nAnother case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.\n\n\n\nIn a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.\nSimilar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#multivariate-hierarchical-priors.section",
+    "href": "stan-users-guide/regression.html#multivariate-hierarchical-priors.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "In hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.\nAs an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.\n\n\nGelman and Hill (2007, chap. 13, Chapter 17) provide a discussion of a hierarchical model with \\(N\\) individuals organized into \\(J\\) groups. Each individual has a predictor row vector \\(x_n\\) of size \\(K\\); to unify the notation, they assume that \\(x_{n,1} = 1\\) is a fixed “intercept” predictor. To encode group membership, they assume individual \\(n\\) belongs to group \\(jj[n] \\in \\{ 1, \\dotsc, J \\}\\). Each individual \\(n\\) also has an observed outcome \\(y_n\\) taking on real values.\n\n\nThe model is a linear regression with slope and intercept coefficients varying by group, so that \\(\\beta_j\\) is the coefficient \\(K\\)-vector for group \\(j\\). The data model for individual \\(n\\) is then just \\[\ny_n \\sim \\textsf{normal}(x_n \\, \\beta_{jj[n]}, \\, \\sigma)\n\\quad\\text{for}\\quad n \\in \\{ 1, \\dotsc, N \\}.\n\\]\n\n\n\nGelman and Hill model the coefficient vectors \\(\\beta_j\\) as being drawn from a multivariate distribution with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\), \\[\n\\beta_j \\sim \\textsf{multivariate normal}(\\mu_j, \\, \\Sigma)\n\\quad\\text{for}\\quad j \\in \\{ 1, \\dotsc, J \\}.\n\\]\nBelow, we discuss the full model of Gelman and Hill, which uses group-level predictors to model \\(\\mu\\); for now, we assume \\(\\mu\\) is a simple vector parameter.\n\n\n\nFor hierarchical modeling, the group-level mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\) must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as \\[\n\\mu_j \\sim \\textsf{normal}(0,5).\n\\] If more is known about the expected coefficient values \\(\\beta_{j, k}\\), this information can be incorporated into the prior for \\(\\mu_j\\).\nFor the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling\nIn Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define \\[\n\\Sigma = \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau) \\times \\Omega \\times \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau),\n\\] where \\(\\Omega\\) is a correlation matrix and \\(\\tau\\) is the vector of coefficient scales. This mapping from scale vector \\(\\tau\\) and correlation matrix \\(\\Omega\\) can be inverted, using \\[\n\\tau_k = \\sqrt{\\Sigma_{k,k}}\n\\quad\\textsf{and}\\quad\n\\Omega_{i, j} = \\frac{\\Sigma_{i, j}}{\\tau_i \\, \\tau_j}.\n\\]\nThe components of the scale vector \\(\\tau\\) can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as \\[\n\\tau_k \\sim \\textsf{Cauchy}(0, 2.5)\n\\quad\\text{for}\\quad k \\in \\{ 1, \\dotsc, K \\}\n\\quad\\text{constrained\\ by}\\quad \\tau_k &gt; 0.\n\\] As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for \\(\\tau\\). For large numbers of exchangeable coefficients, the components of \\(\\tau\\) itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.\nOur final recommendation is to give the correlation matrix \\(\\Omega\\) an LKJ prior with shape \\(\\eta \\geq 1\\),5\n\\[\n\\Omega \\sim \\textsf{LKJCorr}(\\eta).\n\\]\nThe LKJ correlation distribution is defined by \\[\n\\textsf{LKJCorr}\\left(\\Sigma \\mid \\eta\\right)\n\\propto\n\\operatorname{det}\\left(\\Sigma\\right)^{\\eta - 1}.\n\\]\nThe basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For \\(\\eta = 1\\), the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.\nFor \\(\\eta &gt; 1\\), the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For \\(\\eta &lt; 1\\), it increasingly concentrates mass in the other direction, i.e., favoring more correlation.\nThe LKJ prior may thus be used to control the expected amount of correlation among the parameters \\(\\beta_j\\). For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see Barnard, McCulloch, and Meng (2000).\n\n\n\nTo complete Gelman and Hill’s model, suppose each group \\(j \\in \\{ 1, \\dotsc, J \\}\\) is supplied with an \\(L\\)-dimensional row-vector of group-level predictors \\(u_j\\). The prior mean for the \\(\\beta_j\\) can then itself be modeled as a regression, using an \\(L\\)-dimensional coefficient vector \\(\\gamma\\). The prior for the group-level coefficients then becomes \\[\n\\beta_j \\sim \\textsf{multivariate normal}(u_j \\, \\gamma, \\Sigma)\n\\]\nThe group-level coefficients \\(\\gamma\\) may themselves be given independent weakly informative priors, such as \\[\n\\gamma_l \\sim \\textsf{normal}(0,5).\n\\] As usual, information about the group-level means should be incorporated into this prior.\n\n\n\nThe Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.\ndata {\n  int&lt;lower=0&gt; N;              // num individuals\n  int&lt;lower=1&gt; K;              // num ind predictors\n  int&lt;lower=1&gt; J;              // num groups\n  int&lt;lower=1&gt; L;              // num group predictors\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // group for individual\n  matrix[N, K] x;              // individual predictors\n  array[J] row_vector[L] u;    // group predictors\n  vector[N] y;                 // outcomes\n}\nparameters {\n  corr_matrix[K] Omega;        // prior correlation\n  vector&lt;lower=0&gt;[K] tau;      // prior scale\n  matrix[L, K] gamma;          // group coeffs\n  array[J] vector[K] beta;     // indiv coeffs by group\n  real&lt;lower=0&gt; sigma;         // prediction error scale\n}\nmodel {\n  tau ~ cauchy(0, 2.5);\n  Omega ~ lkj_corr(2);\n  to_vector(gamma) ~ normal(0, 5);\n  {\n    array[J] row_vector[K] u_gamma;\n    for (j in 1:J) {\n      u_gamma[j] = u[j] * gamma;\n    }\n    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));\n  }\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n  }\n}\nThe hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix Omega and scale vector tau are more natural to inspect in the output; to output Sigma, define it as a transformed parameter. The function quad_form_diag is defined so that quad_form_diag(Sigma, tau) is equivalent to diag_matrix(tau) * Sigma * diag_matrix(tau), where diag_matrix(tau) returns the matrix with tau on the diagonal and zeroes off diagonal; the version using quad_form_diag should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.\n\n\n\nThe code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:\nfor (n in 1:N) {\n  y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n}\nwith the vectorized distribution statement:\n{\n  vector[N] x_beta_jj;\n  for (n in 1:N) {\n    x_beta_jj[n] = x[n] * beta[jj[n]];\n  }\n  y ~ normal(x_beta_jj, sigma);\n}\nThe outer brackets create a local scope in which to define the variable x_beta_jj, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.\nAlthough it is tempting to redeclare beta and include a revised model block distribution statement,\nparameters {\n  matrix[J, K] beta;\n// ...\n}\nmodel {\n  y ~ normal(rows_dot_product(x, beta[jj]), sigma);\n  // ...\n}\nthis fails because it breaks the vectorization for beta,6\nbeta ~ multi_normal(...);\nwhich requires beta to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because rows_dot_product cannot do much optimization in and of itself because there are no shared computations.\nThe code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.\n{\n  matrix[K, K] Sigma_beta;\n  Sigma_beta = quad_form_diag(Omega, tau);\n  for (j in 1:J) {\n    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);\n  }\n}\nIn this example, the covariance matrix Sigma_beta is defined as a local variable so as not to have to repeat the quadratic form computation \\(J\\) times. This vectorization can be combined with the Cholesky-factor optimization in the next section.\n\n\n\nThe multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.\ndata {\n  matrix[L, J] u;              // group predictors transposed\n  // ...\n}\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  matrix[K, L] gamma;\n  // ...\n}\ntransformed parameters {\n  matrix[K, J] beta;\n  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  // ...\n}\nThe data variable u was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with gamma and beta, to minimize the number of transposition operations. The new parameter L_Omega is the Cholesky factor of the original correlation matrix Omega, so that\nOmega = L_Omega * L_Omega'\nThe prior scale vector tau is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,\nSigma_beta\n  = quad_form_diag(Omega, tau)\n  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'\nwhere the diagonal pre-multiply compound operation is defined by\ndiag_pre_multiply(a, b) = diag_matrix(a) * b\nThe new variable z is declared as a matrix, the entries of which are given independent standard normal priors; the to_vector operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of z being a \\(K\\)-variate normal random vector with the identity as covariance matrix. Therefore, multiplying z by the Cholesky factor of the covariance matrix and adding the mean (u * gamma)' produces a beta distributed as in the original model, where the variance is, letting \\(L = \\mathrm{diag}(\\tau)\\,\\Omega_L\\),\n\\[\n\\begin{aligned}\n\\mathbb{V}[\\beta] &= \\mathbb{E}\\big((L \\, z)(L \\, z)^\\top) \\\\\n&= \\mathbb{E}\\big((L \\, z \\, z^\\top \\, L^\\top) \\\\\n&= L \\, \\mathbb{E}(z \\, z^\\top) \\, L^\\top \\\\\n&= L \\, L^\\top =(\\mathrm{diag}(\\tau)\\,\\Omega_L)\\,(\\mathrm{diag}(\\tau)\\,\\Omega_L)^\\top \\\\\n&= \\mathrm{diag}(\\tau)\\,\\Omega\\,\\mathrm{diag}(\\tau) \\\\\n&= \\Sigma.\n\\end{aligned}\n\\] Where we have used the linearity of expectations (line 2 to 3), the definition of \\(\\Omega = \\Omega_L \\, \\Omega_L^\\top\\), and the fact that \\(\\mathbb{E}(z \\, z^\\top) = I\\) since \\(z \\sim \\mathcal{N}(0, I)\\).\nOmitting the remaining data declarations, which are the same as before with the exception of u, the optimized model is as follows.\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0, upper=pi() / 2&gt;[K] tau_unif;  // prior scale\n  matrix[K, L] gamma;                        // group coeffs\n  real&lt;lower=0&gt; sigma;                       // prediction error scale\n}\ntransformed parameters {\n  vector&lt;lower=0&gt;[K] tau = 2.5 * tan(tau_unif);\n  matrix[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  vector[N] mu;\n  for(n in 1:N) {\n    mu[n] = x[n, ] * beta[, jj[n]];\n  }\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  to_vector(gamma) ~ normal(0, 5);\n  y ~ normal(mu, sigma);\n}\nThis model also reparameterizes the prior scale tau to avoid potential problems with the heavy tails of the Cauchy distribution. The statement tau_unif ~ uniform(0, pi() / 2) can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#prediction-forecasting-and-backcasting",
+    "href": "stan-users-guide/regression.html#prediction-forecasting-and-backcasting",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Stan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).\n\n\nAs a simple example, the following linear regression provides the same setup for estimating the coefficients beta as in our very first example, using y for the N observations and x for the N predictor vectors. The model parameters and model for observations are exactly the same as before.\nTo make predictions, we need to be given the number of predictions, N_new, and their predictor matrix, x_new. The predictions themselves are modeled as a parameter y_new. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector y_new and prediction matrix x_new.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=0&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n\n  int&lt;lower=0&gt; N_new;\n  matrix[N_new, K] x_new;\n}\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n\n  vector[N_new] y_new;                  // predictions\n}\nmodel {\n  y ~ normal(x * beta, sigma);          // observed model\n\n  y_new ~ normal(x_new * beta, sigma);  // prediction model\n}\n\n\n\nWhere possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.\n// ...data as above...\n\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(x * beta, sigma);\n}\ngenerated quantities {\n  vector[N_new] y_new;\n  for (n in 1:N_new) {\n    y_new[n] = normal_rng(x_new[n] * beta, sigma);\n  }\n}\nNow the data are just as before, but the parameter y_new is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.\n\n\nIt is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of \\(2^{31}\\), must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by \\(2^{31} - 1\\) in Stan.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#multivariate-outcomes",
+    "href": "stan-users-guide/regression.html#multivariate-outcomes",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Most regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.\nThis section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.\n\n\nThe first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors (Zellner 1962; Greene 2011).\nThe model is easy to write down as a regression, \\[\\begin{align*}\ny_n        &=  x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nwhere \\(x_n\\) is a \\(J\\)-row-vector of predictors (\\(x\\) is an \\((N \\times\nJ)\\) matrix), \\(y_n\\) is a \\(K\\)-vector of observations, \\(\\beta\\) is a \\((K\n\\times J)\\) matrix of regression coefficients (vector \\(\\beta_k\\) holds coefficients for outcome \\(k\\)), and \\(\\Sigma\\) is covariance matrix governing the error. As usual, the intercept can be rolled into \\(x\\) as a column of ones.\nThe basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; J;\n  int&lt;lower=0&gt; N;\n  array[N] vector[J] x;\n  array[N] vector[K] y;\n}\nparameters {\n  matrix[K, J] beta;\n  cov_matrix[K] Sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n  }\n  y ~ multi_normal(mu, Sigma);\n}\nFor efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.\nFollowing the advice in the multivariate hierarchical priors section, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.\n// ...\nparameters {\n  matrix[K, J] beta;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0&gt;[K] L_sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  matrix[K, K] L_Sigma;\n\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n\n  }\n\n  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);\n\n  to_vector(beta) ~ normal(0, 5);\n  L_Omega ~ lkj_corr_cholesky(4);\n  L_sigma ~ cauchy(0, 2.5);\n\n  y ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix beta to a vector.\nIf required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.\n\n\n\nThe multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.\nThe observations \\(y_n\\) are \\(D\\)-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations \\(y_n\\) are based on latent values \\(z_n\\) drawn from a seemingly unrelated regression model (see the previous section), \\[\\begin{align*}\nz_n        &=     x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim  \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nThese are then put through the step function to produce a \\(K\\)-vector \\(z_n\\) of boolean values with elements defined by \\[\ny_{n, k} = \\operatorname{I}\\left(z_{n, k} &gt; 0\\right),\n\\] where \\(\\operatorname{I}()\\) is the indicator function taking the value 1 if its argument is true and 0 otherwise.\nUnlike in the seemingly unrelated regressions case, here the covariance matrix \\(\\Sigma\\) has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see Greene (2011)).\nMultivariate probit regression can be coded in Stan using the trick introduced by Albert and Chib (1993), where the underlying continuous value vectors \\(y_n\\) are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector \\(z\\) in two parts, based on whether the corresponding value of \\(y\\) is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.\nFirst, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in \\(y\\).\nfunctions {\n  int sum2d(array[,] int a) {\n    int s = 0;\n    for (i in 1:size(a)) {\n      s += sum(a[i]);\n    }\n    return s;\n  }\n}\nThe function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.\nThe data declaration block is much like for the seemingly unrelated regressions, but the observations y are now integers constrained to be 0 or 1.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  array[N, D] int&lt;lower=0, upper=1&gt; y;\n  array[N] vector[K] x;\n}\nAfter declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array y into positive and negative components, keeping track of indexes so that z can be easily reassembled in the transformed parameters block.\ntransformed data {\n  int&lt;lower=0&gt; N_pos;\n  array[sum2d(y)] int&lt;lower=1, upper=N&gt; n_pos;\n  array[size(n_pos)] int&lt;lower=1, upper=D&gt; d_pos;\n  int&lt;lower=0&gt; N_neg;\n  array[(N * D) - size(n_pos)] int&lt;lower=1, upper=N&gt; n_neg;\n  array[size(n_neg)] int&lt;lower=1, upper=D&gt; d_neg;\n\n  N_pos = size(n_pos);\n  N_neg = size(n_neg);\n  {\n    int i;\n    int j;\n    i = 1;\n    j = 1;\n    for (n in 1:N) {\n      for (d in 1:D) {\n        if (y[n, d] == 1) {\n          n_pos[i] = n;\n          d_pos[i] = d;\n          i += 1;\n        } else {\n          n_neg[j] = n;\n          d_neg[j] = d;\n          j += 1;\n        }\n      }\n    }\n  }\n}\nThe variables N_pos and N_neg are set to the number of true (1) and number of false (0) observations in y. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.\nThe parameters are declared as follows.\nparameters {\n  matrix[D, K] beta;\n  cholesky_factor_corr[D] L_Omega;\n  vector&lt;lower=0&gt;[N_pos] z_pos;\n  vector&lt;upper=0&gt;[N_neg] z_neg;\n}\nThese include the regression coefficients beta and the Cholesky factor of the correlation matrix, L_Omega. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).\nThe critical part of the parameter declaration is that the latent real value \\(z\\) is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct \\(z\\).\ntransformed parameters {\n  array[N] vector[D] z;\n  for (n in 1:N_pos) {\n    z[n_pos[n], d_pos[n]] = z_pos[n];\n  }\n  for (n in 1:N_neg) {\n    z[n_neg[n], d_neg[n]] = z_neg[n];\n  }\n}\nAt this point, the model is simple, pretty much recreating the seemingly unrelated regression.\nmodel {\n  L_Omega ~ lkj_corr_cholesky(4);\n  to_vector(beta) ~ normal(0, 5);\n  {\n    array[N] vector[D] beta_x;\n    for (n in 1:N) {\n      beta_x[n] = beta * x[n];\n    }\n    z ~ multi_normal_cholesky(beta_x, L_Omega);\n  }\n}\nThis simple form of model is made possible by the Albert and Chib-style constraints on z.\nFinally, the correlation matrix itself can be put back together in the generated quantities block if desired.\ngenerated quantities {\n  corr_matrix[D] Omega;\n  Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\nThe same could be done for the seemingly unrelated regressions in the previous section.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#applications-of-pseudorandom-number-generation",
+    "href": "stan-users-guide/regression.html#applications-of-pseudorandom-number-generation",
+    "title": "Regression Models",
+    "section": "",
+    "text": "The main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.\n\n\nConsider predicting unobserved outcomes using linear regression. Given predictors \\(x_1, \\dotsc, x_N\\) and observed outcomes \\(y_1, \\dotsc, y_N\\), and assuming a standard linear regression with intercept \\(\\alpha\\), slope \\(\\beta\\), and error scale \\(\\sigma\\), along with improper uniform priors, the posterior over the parameters given \\(x\\) and \\(y\\) is \\[\np\\left(\\alpha, \\beta, \\sigma \\mid x, y \\right)\n\\propto\n\\prod_{n=1}^N\n  \\textsf{normal}\\left(y_n \\mid \\alpha + \\beta x_n, \\sigma\\right).\n\\]\nFor this model, the posterior predictive inference for a new outcome \\(\\tilde{y}_m\\) given a predictor \\(\\tilde{x}_m\\), conditioned on the observed data \\(x\\) and \\(y\\), is \\[\np\\left(\\tilde{y}_n \\mid \\tilde{x}_n, x, y\\right)\n= \\int_{(\\alpha,\\beta,\\sigma)}\n  \\textsf{normal}\\left(\\tilde{y}_n \\mid \\alpha + \\beta \\tilde{x}_n, \\sigma\\right)\n  \\times\n  p\\left(\\alpha, \\beta, \\sigma \\mid x, y\\right)\n  \\,\\textrm{d}(\\alpha,\\beta,\\sigma).\n\\]\nTo code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde;\n  for (n in 1:N_tilde) {\n    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);\n  }\n}\nGiven observed predictors \\(x\\) and outcomes \\(y\\), y_tilde will be drawn according to \\(p\\left(\\tilde{y} \\mid \\tilde{x}, y, x\\right)\\). This means that, for example, the posterior mean for y_tilde is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).\n\n\n\nA good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors \\(x\\) are used in place of new predictors \\(\\tilde{x}\\) for unobserved outcomes. If the model fits the data well, the predictions for \\(\\tilde{y}\\) based on \\(x\\) should match the observed data \\(y\\).\nTo code posterior predictive checks in Stan requires only a slight modification of the prediction code to use \\(x\\) and \\(N\\) in place of \\(\\tilde{x}\\) and \\(\\tilde{N}\\),\ngenerated quantities {\n  vector[N] y_tilde;\n  for (n in 1:N) {\n    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);\n  }\n}\nGelman et al. (2013) recommend choosing several posterior draws \\(\\tilde{y}^{(1)}, \\dotsc, \\tilde{y}^{(M)}\\) and plotting each of them alongside the data \\(y\\) that was actually observed. If the model fits well, the simulated \\(\\tilde{y}\\) will look like the actual data \\(y\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/regression.html#footnotes",
+    "href": "stan-users-guide/regression.html#footnotes",
+    "title": "Regression Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nUnlike in Python and R, which are interpreted, Stan is translated to C++ and compiled, so loops and assignment statements are fast. Vectorized code is faster in Stan because (a) the expression tree used to compute derivatives can be simplified, leading to fewer virtual function calls, and (b) computations that would be repeated in the looping version, such as log(sigma) in the above model, will be computed once and reused.↩︎\nThe Phi_approx function is a rescaled version of the inverse logit function, so while the scale is roughly the same \\(\\Phi\\), the tails do not match.↩︎\nThe Bernoulli-logit distribution builds in the log link function, taking \\[\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right) = \\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\\]↩︎\nGelman and Hill (2007) treat the \\(\\delta\\) term equivalently as the location parameter in the distribution of student abilities.↩︎\nThe prior is named for Lewandowski, Kurowicka, and Joe, as it was derived by inverting the random correlation matrix generation strategy of Lewandowski, Kurowicka, and Joe (2009).↩︎\nThanks to Mike Lawrence for pointing this out in the GitHub issue for the manual.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Regression Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/proportionality-constants.html",
+    "href": "stan-users-guide/proportionality-constants.html",
+    "title": "Proportionality Constants",
+    "section": "",
+    "text": "When evaluating a likelihood or prior as part of the log density computation in MCMC, variational inference, or optimization, it is usually only necessary to compute the functions up to a proportionality constant (or similarly compute log densities up to an additive constant). In MCMC this comes from the fact that the distribution being sampled does not need to be normalized (and so it is the normalization constant that is ignored). Similarly the distribution does not need normalized to perform variational inference or do optimizations. The advantage of working with unnormalized distributions is they can make computation quite a bit cheaper.\nThere are three different syntaxes to build the model in Stan. The way to select between them is by determining if the proportionality constants are necessary. If performance is not a problem, it is always safe to use the normalized densities.\nThe distribution statement (~) and log density increment statement (target +=) with _lupdf() use unnormalized densities for \\(x\\) (dropping proportionality constants):\nx ~ normal(0, 1);\ntarget += normal_lupdf(x | 0, 1); // the 'u' is for unnormalized\nThe log density increment statement (target +=) with _lpdf() uses the full normalized density for \\(x\\) (dropping no constants):\ntarget += normal_lpdf(x | 0, 1);\nFor discrete distributions, the target += syntax is using _lupmf and _lpmf instead:\ny ~ bernoulli(0.5);\ntarget += bernoulli_lupmf(y | 0.5);\ntarget += bernoulli_lpmf(y | 0.5);\n\n\nIf a density \\(p(\\theta)\\) can be factored into \\(K g(\\theta)\\) where \\(K\\) are all the factors that are a not a function of \\(\\theta\\) and \\(g(\\theta)\\) are all the terms that are a function of \\(\\theta\\), then it is said that \\(g(\\theta)\\) is proportional to \\(p(\\theta)\\) up to a constant.\nThe advantage of all this is that sometimes \\(K\\) is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.\nStan takes advantage of the proportionality constant fact with the ~ syntax. Take for instance the normal data model:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nSyntactically, this is just shorthand for the equivalent model that replaces the ~ syntax with a target += statement and a normal_lupdf function call:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  target += normal_lupdf(x | mu, sigma);\n}\nThe function normal_lupdf is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\nNow because the density here is only a function of \\(x\\), the additive terms in the log density that are not a function of \\(x\\) can be dropped. In this case it is enough to know only the quadratic term:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\n\n\n\nIn the case that the proportionality constants were needed for a normal log density the function normal_lpdf can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the ~ or target += normal_lupdf syntaxes if it is absolutely clear that the proportionality constants are not necessary.\n\n\n\nWhen a custom _lpdf or _lpmf function is defined, the compiler will automatically make available a _lupdf or _lupmf version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.\nThe difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.\nThe following code demonstrates the different behaviors:\nfunctions {\n  real custom1_lpdf(x) {\n    return normal_lupdf(x | 0.0, 1.0);\n  }\n  real custom2_lpdf(x) {\n    return normal_lpdf(x | 0.0, 1.0);\n  }\n}\nparameters {\n  real mu;\n}\nmodel {\n  mu ~ custom1(); // Normalization constants dropped\n  target += custom1_lupdf(mu); // Normalization constants dropped\n  target += custom1_lpdf(mu);  // Normalization constants kept\n\n  mu ~ custom2();  // Normalization constants kept\n  target += custom2_lupdf(mu);  // Normalization constants kept\n  target += custom2_lpdf(mu);  // Normalization constants kept\n}\n\n\n\nTo avoid ambiguities in how the normalization constants work, functions ending in _lupdf and _lupmf can only be used in the model block or user-defined probability functions (functions ending in _lpdf or _lpmf).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Proportionality Constants"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/proportionality-constants.html#dropping-proportionality-constants",
+    "href": "stan-users-guide/proportionality-constants.html#dropping-proportionality-constants",
+    "title": "Proportionality Constants",
+    "section": "",
+    "text": "If a density \\(p(\\theta)\\) can be factored into \\(K g(\\theta)\\) where \\(K\\) are all the factors that are a not a function of \\(\\theta\\) and \\(g(\\theta)\\) are all the terms that are a function of \\(\\theta\\), then it is said that \\(g(\\theta)\\) is proportional to \\(p(\\theta)\\) up to a constant.\nThe advantage of all this is that sometimes \\(K\\) is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.\nStan takes advantage of the proportionality constant fact with the ~ syntax. Take for instance the normal data model:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nSyntactically, this is just shorthand for the equivalent model that replaces the ~ syntax with a target += statement and a normal_lupdf function call:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  target += normal_lupdf(x | mu, sigma);\n}\nThe function normal_lupdf is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\nNow because the density here is only a function of \\(x\\), the additive terms in the log density that are not a function of \\(x\\) can be dropped. In this case it is enough to know only the quadratic term:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Proportionality Constants"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/proportionality-constants.html#keeping-proportionality-constants",
+    "href": "stan-users-guide/proportionality-constants.html#keeping-proportionality-constants",
+    "title": "Proportionality Constants",
+    "section": "",
+    "text": "In the case that the proportionality constants were needed for a normal log density the function normal_lpdf can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the ~ or target += normal_lupdf syntaxes if it is absolutely clear that the proportionality constants are not necessary.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Proportionality Constants"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/proportionality-constants.html#user-defined-distributions",
+    "href": "stan-users-guide/proportionality-constants.html#user-defined-distributions",
+    "title": "Proportionality Constants",
+    "section": "",
+    "text": "When a custom _lpdf or _lpmf function is defined, the compiler will automatically make available a _lupdf or _lupmf version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.\nThe difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.\nThe following code demonstrates the different behaviors:\nfunctions {\n  real custom1_lpdf(x) {\n    return normal_lupdf(x | 0.0, 1.0);\n  }\n  real custom2_lpdf(x) {\n    return normal_lpdf(x | 0.0, 1.0);\n  }\n}\nparameters {\n  real mu;\n}\nmodel {\n  mu ~ custom1(); // Normalization constants dropped\n  target += custom1_lupdf(mu); // Normalization constants dropped\n  target += custom1_lpdf(mu);  // Normalization constants kept\n\n  mu ~ custom2();  // Normalization constants kept\n  target += custom2_lupdf(mu);  // Normalization constants kept\n  target += custom2_lpdf(mu);  // Normalization constants kept\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Proportionality Constants"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/proportionality-constants.html#limitations-on-using-_lupdf-and-_lupmf-functions",
+    "href": "stan-users-guide/proportionality-constants.html#limitations-on-using-_lupdf-and-_lupmf-functions",
+    "title": "Proportionality Constants",
+    "section": "",
+    "text": "To avoid ambiguities in how the normalization constants work, functions ending in _lupdf and _lupmf can only be used in the model block or user-defined probability functions (functions ending in _lpdf or _lpmf).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Proportionality Constants"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html",
+    "href": "stan-users-guide/poststratification.html",
+    "title": "Poststratification",
+    "section": "",
+    "text": "Stratification is a technique developed for survey sampling in which a population is partitioned into subgroups (i.e., stratified) and each group (i.e., stratum) is sampled independently. If the subgroups are more homogeneous (i.e., lower variance) than the population as a whole, this can reduce variance in the estimate of a quantity of interest at the population level.\nPoststratification is a technique for adjusting a non-representative sample (i.e., a convenience sample or other observational data) for which there are demographic predictors characterizing the strata. It is carried out after a model is fit to the observed data, hence the name poststratification (Little 1993). Poststratification can be fruitfully combined with regression modeling (or more general parametric modeling), which provides estimates based on combinations of predictors (or general parameters) rather than raw counts in each stratum. Multilevel modeling is useful in determining how much partial pooling to apply in the regressions, leading to the popularity of the combination of multilevel regression and poststratification (MRP) (Park, Gelman, and Bafumi 2004).\n\n\n\n\nStratification and poststratification can be applied to many applications beyond survey sampling (Kennedy and Gelman 2019). For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid (Paustian et al. 1997). Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both (Field et al. 1998).\n\n\n\nSuppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.\nThe proportion of support in the poll among students in each group provides a simple maximum likelihood estimate \\(\\theta^* = (0.25, 0.5,\n0.8)\\) of support in each group for a simple Bernoulli model where student \\(n\\)’s vote is modeled as \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:3\\) is the subgroup to which the \\(n\\)-th student belongs.\nAn estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting \\(N = (20\\,000,\\, 5\\,000,\\, 2\\,000)\\) be the subgroup sizes, the poststratified estimate of support in the population \\(\\phi^*\\) is estimated by \\[\n\\phi^*\n= \\frac{\\displaystyle \\sum_{j = 1}^3 \\theta_j^* \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^3 N_j}.\n\\] Plugging in our estimates and population counts yields \\[\\begin{eqnarray*}\n\\phi*\n& = & \\frac{0.25 \\cdot 20\\,000 + 0.4 \\cdot 5\\,000 + 0.8 \\cdot 2\\,000}\n           {20\\,000 + 5\\,000 + 2\\,000}\n\\\\[4pt] & = & \\frac{8\\,600}{27\\,000}\n\\\\[4pt] & \\approx & 0.32.\n\\end{eqnarray*}\\]\n\n\n\n\nConsidering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:J\\) is the group to which item \\(n\\) belongs and \\(\\theta_j\\) is the proportion of support in group \\(j\\).\nThis can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting \\(A_j\\) be the number of respondents in group \\(j\\) and \\(a_j\\) be the number of positive responses in group \\(j\\), the data model may be reduced to the form \\[\na_j \\sim \\textrm{binomial}(A_j, \\theta_j).\n\\] A simple uniform prior on the proportion of support in each group completes the model, \\[\n\\theta_j \\sim \\textrm{beta}(1, 1).\n\\] A more informative prior could be used if there is prior information available about support among the student body.\nUsing sampling, draws \\(\\theta^{(m)} \\sim p(\\theta \\mid y)\\) from the posterior may be combined with the population sizes \\(N\\) to estimate \\(\\phi\\), the proportion of support in the population, \\[\n\\phi^{(m)}\n= \\frac{\\displaystyle \\sum_{j = 1}^J \\theta_j^{(m)} \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^J N_j}.\n\\] The posterior draws for \\(\\phi^{(m)}\\) characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the \\(\\phi^{(m)}\\)), posterior intervals (quantiles of the \\(\\phi^{(m)}\\)), or to plot a histogram.\n\n\n\nThe maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where \\(A_j\\) is the number of voters from group \\(j\\), \\(a_j\\) is the number of positive responses from group \\(j\\), and \\(N_j\\) is the size of group \\(j\\) in the population.\ndata {\n  int&lt;lower=1&gt; J;\n  array[J] int&lt;lower=0&gt; A; \n  array[J] int&lt;lower=0&gt; a;\n  vector&lt;lower=0&gt;[J] N;\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[J] theta;\n}\nmodel {\n  a ~ binomial(A, theta);\n}\ngenerated quantities {t\n  real&lt;lower=0, upper=1&gt; phi = dot(N, theta) / sum(N);\n}\nThe binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the \\(J\\) terms. The prior is implicitly uniform on \\((0, 1),\\) the support of \\(\\theta.\\) The summation is computed using a dot product and the sum function, which is why N was declared as a vector rather than as an array of integers.\n\n\n\nIn applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.\nThe problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to \\(5 \\cdot 2 \\cdot 5 \\cdot\n50 = 2000\\) strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter \\(\\theta_j\\) for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.\nA practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only \\(4 + 2 + 5 + 50 =\n61\\) regression coefficients to estimate. Now suppose that item \\(n\\) has demographic features \\(\\textrm{age}_n \\in 1:5\\), \\(\\textrm{sex}_n \\in 1:2\\), \\(\\textrm{income}_n \\in 1:5,\\) and \\(\\textrm{state}_n \\in 1:50\\). A logistic regression may be formulated as \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)),\n\\] where \\(\\textrm{age}[n]\\) is the age of the \\(n\\)-th respondent, \\(\\textrm{sex}[n]\\) is their sex, \\(\\textrm{income}[n]\\) their income and \\(\\textrm{state}[n]\\) their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.\nTo poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as \\[\n\\sum_{i = 1}^5 \\sum_{j = 1}^2 \\sum_{k = 1}^5 \\sum_{m = 1}^{50}\n\\textrm{logit}^{-1}(\\alpha + \\beta_i + \\gamma_j + \\delta_k + \\eta_m)\n\\cdot \\textrm{pop}_{i, j, k, m},\n\\] where \\(\\textrm{pop}_{i, j, k, m}\\) is the size of the subpopulation with age \\(i\\), sex \\(j\\), income level \\(k\\), and state of residence \\(m\\).\nAs formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes \\(y\\) given predictors \\(x\\).\n\n\n\nWith large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.\nThe model introduced in the previous section had the data model \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)).\n\\] The overall intercept can be given a broad fixed prior, \\[\n\\alpha \\sim \\textrm{normal}(0, 5).\n\\] The other regression parameters can be given hierarchical priors, \\[\\begin{eqnarray*}\n\\beta_{1:4} & \\sim & \\textrm{normal}(0, \\sigma^{\\beta})\n\\\\[2pt]\n\\gamma_{1:2} & \\sim & \\textrm{normal}(0, \\sigma^{\\gamma})\n\\\\[2pt]\n\\delta_{1:5} & \\sim & \\textrm{normal}(0, \\sigma^{\\delta})\n\\\\[2pt]\n\\epsilon_{1:50} & \\sim & \\textrm{normal}(0, \\sigma^{\\epsilon}).\n\\end{eqnarray*}\\]\nThe hyperparameters for scale of variation within a group can be given simple standard hyperpriors, \\[\n\\sigma^{\\beta}, \\sigma^{\\gamma}, \\sigma^{\\delta}, \\sigma^{\\epsilon}\n\\sim \\textrm{normal}(0, 1).\n\\] The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).\n\n\nThe multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.\nThis is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, \\(\\gamma_2 =\n-\\gamma_1.\\)\nTo implement sum-to-zero constraints,\nparameters {\n  vector[K - 1] alpha_raw;\n// ...\n}\ntransformed parameters {\n  vector&lt;multiplier=sigma_alpha&gt;[K] alpha\n    = append_row(alpha_raw, -sum(alpha_raw));\n// ...    \n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n}\nThis prior is hard to interpret in that there are K normal distributions, but only K - 1 free parameters. An alternative is to put the prior only on alpha_raw, but that is also difficult to interpret.\nSoft constraints can be more computationally tractable. They are also simpler to implement.\nparameters {\n  vector&lt;multiplier=alpha&gt;[K] alpha;\n// ...\n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n  sum(alpha) ~ normal(0, 0.001);\n}\nThis leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.\nNote that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept \\(\\alpha.\\) Even with a wide prior on the intercept, the hyperprior on \\(\\sigma^{\\gamma}\\) may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.\nNone of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.\n\n\n\n\nMultilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the multiplier transform is used on each of the parameters. The combination of\nvector&lt;multiplier=s&gt;[K] a;\n// ...\na ~ normal(0, s);\nimplements a non-centered parameterization for a; a centered parameterization would drop the multiplier specification. The prior scale s is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the multiplier.\nHere is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable P.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; age;\n  array[N] int&lt;lower=1, upper=5&gt; income;\n  array[N] int&lt;lower=1, upper=50&gt; state;\n  array[N] int&lt;lower=0&gt; y;\n  array[4, 5, 50] int&lt;lower=0&gt; P;\n}\nparameters {\n  real alpha;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;multiplier=sigma_beta&gt;[4] beta;\n  real&lt;lower=0&gt; sigma_gamma;\n  vector&lt;multiplier=sigma_gamma&gt;[5] gamma;\n  real&lt;lower=0&gt; sigma_delta;\n  vector&lt;multiplier=sigma_delta&gt;[50] delta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, sigma_beta);\n  gamma ~ normal(0, sigma_gamma);\n  delta ~ normal(0, sigma_delta);\n  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(0, 1);\n}\ngenerated quantities {\n  real expect_pos = 0;\n  int total = 0;\n  for (b in 1:4) {\n    for (c in 1:5) {\n      for (d in 1:50) {\n        total += P[b, c, d];\n        expect_pos\n          += P[b, c, d]\n             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);\n      }\n    }\n  }\n  real&lt;lower=0, upper=1&gt; phi = expect_pos / total;\n}\nUnlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).\n\n\nIn some cases, it can be more efficient to break the data down by group. Suppose there are \\(4 \\times 5 \\times 2 \\times 50 = 2000\\) groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group\nint&lt;lower=0&gt; G;\narray[G] int&lt;lower=1, upper=4&gt; age;\narray[G] int&lt;lower=1, upper=5&gt; income;\narray[G] int&lt;lower=1, upper=50&gt; state;\nThen the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.\narray[G] int&lt;lower=0&gt; pos_votes;\narray[G] int&lt;lower=0&gt; total_votes;\nFinally, the data model is converted to binomial.\npos_votes ~ binomial_logit(total_votes,\n                           alpha + beta[age] + ...);\nThe predictors look the same because of the way the age and other data items are coded.\n\n\n\nIn this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the N data points.\n  array[N] int&lt;lower=1, upper=2&gt; sex;\nThen add a single regression coefficient as a parameter,\n  real epsilon;\nIn the log odds calculation, introduce a new term\n[epsilon, -epsilon][sex]';\nThat is, the data model will now look like\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]\n                      + [epsilon, -epsilon][sex]');\nFor data point n, the expression [epsilon, -epsilon][sex] takes on value [epsilon, -epsilon][sex][n], which with Stan’s multi-indexing reduces to [epsilon, -epsilon][sex[n]]. This term evaluates to epsilon if sex[n] is 1 and to -epsilon if sex[n] is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The ' at the end transposes [epsilon, -epsilon][sex] which is a row_vector into a vector that can be added to the other variables.\nFinally, a prior is needed for the coefficient in the model block,\nepsilon ~ normal(0, 2);\nAs with other priors in multilevel models, the posterior for epsilon should be investigated to make sure it is not unrealistically wide.\n\n\n\n\nIf there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable\narray[50] real&lt;lower=0&gt; income;\nthen a regression coefficient psi can be added for the effect of average state income,\nreal psi;\nwith a fixed prior,\npsi ~ normal(0, 2);\nThis prior assumes the income predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,\ny ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]\n                    + income[state] * psi);\nAnd finally, the formula in the generated quantities block is also updated,\nexpect_pos\n  += P[b, c, d]\n     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]\n             + income[d] * psi);\nHere d is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#some-examples",
+    "href": "stan-users-guide/poststratification.html#some-examples",
+    "title": "Poststratification",
+    "section": "",
+    "text": "Stratification and poststratification can be applied to many applications beyond survey sampling (Kennedy and Gelman 2019). For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid (Paustian et al. 1997). Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both (Field et al. 1998).\n\n\n\nSuppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.\nThe proportion of support in the poll among students in each group provides a simple maximum likelihood estimate \\(\\theta^* = (0.25, 0.5,\n0.8)\\) of support in each group for a simple Bernoulli model where student \\(n\\)’s vote is modeled as \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:3\\) is the subgroup to which the \\(n\\)-th student belongs.\nAn estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting \\(N = (20\\,000,\\, 5\\,000,\\, 2\\,000)\\) be the subgroup sizes, the poststratified estimate of support in the population \\(\\phi^*\\) is estimated by \\[\n\\phi^*\n= \\frac{\\displaystyle \\sum_{j = 1}^3 \\theta_j^* \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^3 N_j}.\n\\] Plugging in our estimates and population counts yields \\[\\begin{eqnarray*}\n\\phi*\n& = & \\frac{0.25 \\cdot 20\\,000 + 0.4 \\cdot 5\\,000 + 0.8 \\cdot 2\\,000}\n           {20\\,000 + 5\\,000 + 2\\,000}\n\\\\[4pt] & = & \\frac{8\\,600}{27\\,000}\n\\\\[4pt] & \\approx & 0.32.\n\\end{eqnarray*}\\]",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#bayesian-poststratification",
+    "href": "stan-users-guide/poststratification.html#bayesian-poststratification",
+    "title": "Poststratification",
+    "section": "",
+    "text": "Considering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:J\\) is the group to which item \\(n\\) belongs and \\(\\theta_j\\) is the proportion of support in group \\(j\\).\nThis can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting \\(A_j\\) be the number of respondents in group \\(j\\) and \\(a_j\\) be the number of positive responses in group \\(j\\), the data model may be reduced to the form \\[\na_j \\sim \\textrm{binomial}(A_j, \\theta_j).\n\\] A simple uniform prior on the proportion of support in each group completes the model, \\[\n\\theta_j \\sim \\textrm{beta}(1, 1).\n\\] A more informative prior could be used if there is prior information available about support among the student body.\nUsing sampling, draws \\(\\theta^{(m)} \\sim p(\\theta \\mid y)\\) from the posterior may be combined with the population sizes \\(N\\) to estimate \\(\\phi\\), the proportion of support in the population, \\[\n\\phi^{(m)}\n= \\frac{\\displaystyle \\sum_{j = 1}^J \\theta_j^{(m)} \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^J N_j}.\n\\] The posterior draws for \\(\\phi^{(m)}\\) characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the \\(\\phi^{(m)}\\)), posterior intervals (quantiles of the \\(\\phi^{(m)}\\)), or to plot a histogram.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#poststratification-in-stan",
+    "href": "stan-users-guide/poststratification.html#poststratification-in-stan",
+    "title": "Poststratification",
+    "section": "",
+    "text": "The maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where \\(A_j\\) is the number of voters from group \\(j\\), \\(a_j\\) is the number of positive responses from group \\(j\\), and \\(N_j\\) is the size of group \\(j\\) in the population.\ndata {\n  int&lt;lower=1&gt; J;\n  array[J] int&lt;lower=0&gt; A; \n  array[J] int&lt;lower=0&gt; a;\n  vector&lt;lower=0&gt;[J] N;\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[J] theta;\n}\nmodel {\n  a ~ binomial(A, theta);\n}\ngenerated quantities {t\n  real&lt;lower=0, upper=1&gt; phi = dot(N, theta) / sum(N);\n}\nThe binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the \\(J\\) terms. The prior is implicitly uniform on \\((0, 1),\\) the support of \\(\\theta.\\) The summation is computed using a dot product and the sum function, which is why N was declared as a vector rather than as an array of integers.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#regression-and-poststratification",
+    "href": "stan-users-guide/poststratification.html#regression-and-poststratification",
+    "title": "Poststratification",
+    "section": "",
+    "text": "In applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.\nThe problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to \\(5 \\cdot 2 \\cdot 5 \\cdot\n50 = 2000\\) strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter \\(\\theta_j\\) for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.\nA practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only \\(4 + 2 + 5 + 50 =\n61\\) regression coefficients to estimate. Now suppose that item \\(n\\) has demographic features \\(\\textrm{age}_n \\in 1:5\\), \\(\\textrm{sex}_n \\in 1:2\\), \\(\\textrm{income}_n \\in 1:5,\\) and \\(\\textrm{state}_n \\in 1:50\\). A logistic regression may be formulated as \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)),\n\\] where \\(\\textrm{age}[n]\\) is the age of the \\(n\\)-th respondent, \\(\\textrm{sex}[n]\\) is their sex, \\(\\textrm{income}[n]\\) their income and \\(\\textrm{state}[n]\\) their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.\nTo poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as \\[\n\\sum_{i = 1}^5 \\sum_{j = 1}^2 \\sum_{k = 1}^5 \\sum_{m = 1}^{50}\n\\textrm{logit}^{-1}(\\alpha + \\beta_i + \\gamma_j + \\delta_k + \\eta_m)\n\\cdot \\textrm{pop}_{i, j, k, m},\n\\] where \\(\\textrm{pop}_{i, j, k, m}\\) is the size of the subpopulation with age \\(i\\), sex \\(j\\), income level \\(k\\), and state of residence \\(m\\).\nAs formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes \\(y\\) given predictors \\(x\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#multilevel-regression-and-poststratification",
+    "href": "stan-users-guide/poststratification.html#multilevel-regression-and-poststratification",
+    "title": "Poststratification",
+    "section": "",
+    "text": "With large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.\nThe model introduced in the previous section had the data model \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)).\n\\] The overall intercept can be given a broad fixed prior, \\[\n\\alpha \\sim \\textrm{normal}(0, 5).\n\\] The other regression parameters can be given hierarchical priors, \\[\\begin{eqnarray*}\n\\beta_{1:4} & \\sim & \\textrm{normal}(0, \\sigma^{\\beta})\n\\\\[2pt]\n\\gamma_{1:2} & \\sim & \\textrm{normal}(0, \\sigma^{\\gamma})\n\\\\[2pt]\n\\delta_{1:5} & \\sim & \\textrm{normal}(0, \\sigma^{\\delta})\n\\\\[2pt]\n\\epsilon_{1:50} & \\sim & \\textrm{normal}(0, \\sigma^{\\epsilon}).\n\\end{eqnarray*}\\]\nThe hyperparameters for scale of variation within a group can be given simple standard hyperpriors, \\[\n\\sigma^{\\beta}, \\sigma^{\\gamma}, \\sigma^{\\delta}, \\sigma^{\\epsilon}\n\\sim \\textrm{normal}(0, 1).\n\\] The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).\n\n\nThe multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.\nThis is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, \\(\\gamma_2 =\n-\\gamma_1.\\)\nTo implement sum-to-zero constraints,\nparameters {\n  vector[K - 1] alpha_raw;\n// ...\n}\ntransformed parameters {\n  vector&lt;multiplier=sigma_alpha&gt;[K] alpha\n    = append_row(alpha_raw, -sum(alpha_raw));\n// ...    \n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n}\nThis prior is hard to interpret in that there are K normal distributions, but only K - 1 free parameters. An alternative is to put the prior only on alpha_raw, but that is also difficult to interpret.\nSoft constraints can be more computationally tractable. They are also simpler to implement.\nparameters {\n  vector&lt;multiplier=alpha&gt;[K] alpha;\n// ...\n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n  sum(alpha) ~ normal(0, 0.001);\n}\nThis leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.\nNote that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept \\(\\alpha.\\) Even with a wide prior on the intercept, the hyperprior on \\(\\sigma^{\\gamma}\\) may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.\nNone of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#coding-mrp-in-stan",
+    "href": "stan-users-guide/poststratification.html#coding-mrp-in-stan",
+    "title": "Poststratification",
+    "section": "",
+    "text": "Multilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the multiplier transform is used on each of the parameters. The combination of\nvector&lt;multiplier=s&gt;[K] a;\n// ...\na ~ normal(0, s);\nimplements a non-centered parameterization for a; a centered parameterization would drop the multiplier specification. The prior scale s is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the multiplier.\nHere is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable P.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; age;\n  array[N] int&lt;lower=1, upper=5&gt; income;\n  array[N] int&lt;lower=1, upper=50&gt; state;\n  array[N] int&lt;lower=0&gt; y;\n  array[4, 5, 50] int&lt;lower=0&gt; P;\n}\nparameters {\n  real alpha;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;multiplier=sigma_beta&gt;[4] beta;\n  real&lt;lower=0&gt; sigma_gamma;\n  vector&lt;multiplier=sigma_gamma&gt;[5] gamma;\n  real&lt;lower=0&gt; sigma_delta;\n  vector&lt;multiplier=sigma_delta&gt;[50] delta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, sigma_beta);\n  gamma ~ normal(0, sigma_gamma);\n  delta ~ normal(0, sigma_delta);\n  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(0, 1);\n}\ngenerated quantities {\n  real expect_pos = 0;\n  int total = 0;\n  for (b in 1:4) {\n    for (c in 1:5) {\n      for (d in 1:50) {\n        total += P[b, c, d];\n        expect_pos\n          += P[b, c, d]\n             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);\n      }\n    }\n  }\n  real&lt;lower=0, upper=1&gt; phi = expect_pos / total;\n}\nUnlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).\n\n\nIn some cases, it can be more efficient to break the data down by group. Suppose there are \\(4 \\times 5 \\times 2 \\times 50 = 2000\\) groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group\nint&lt;lower=0&gt; G;\narray[G] int&lt;lower=1, upper=4&gt; age;\narray[G] int&lt;lower=1, upper=5&gt; income;\narray[G] int&lt;lower=1, upper=50&gt; state;\nThen the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.\narray[G] int&lt;lower=0&gt; pos_votes;\narray[G] int&lt;lower=0&gt; total_votes;\nFinally, the data model is converted to binomial.\npos_votes ~ binomial_logit(total_votes,\n                           alpha + beta[age] + ...);\nThe predictors look the same because of the way the age and other data items are coded.\n\n\n\nIn this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the N data points.\n  array[N] int&lt;lower=1, upper=2&gt; sex;\nThen add a single regression coefficient as a parameter,\n  real epsilon;\nIn the log odds calculation, introduce a new term\n[epsilon, -epsilon][sex]';\nThat is, the data model will now look like\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]\n                      + [epsilon, -epsilon][sex]');\nFor data point n, the expression [epsilon, -epsilon][sex] takes on value [epsilon, -epsilon][sex][n], which with Stan’s multi-indexing reduces to [epsilon, -epsilon][sex[n]]. This term evaluates to epsilon if sex[n] is 1 and to -epsilon if sex[n] is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The ' at the end transposes [epsilon, -epsilon][sex] which is a row_vector into a vector that can be added to the other variables.\nFinally, a prior is needed for the coefficient in the model block,\nepsilon ~ normal(0, 2);\nAs with other priors in multilevel models, the posterior for epsilon should be investigated to make sure it is not unrealistically wide.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/poststratification.html#adding-group-level-predictors",
+    "href": "stan-users-guide/poststratification.html#adding-group-level-predictors",
+    "title": "Poststratification",
+    "section": "",
+    "text": "If there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable\narray[50] real&lt;lower=0&gt; income;\nthen a regression coefficient psi can be added for the effect of average state income,\nreal psi;\nwith a fixed prior,\npsi ~ normal(0, 2);\nThis prior assumes the income predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,\ny ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]\n                    + income[state] * psi);\nAnd finally, the formula in the generated quantities block is also updated,\nexpect_pos\n  += P[b, c, d]\n     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]\n             + income[d] * psi);\nHere d is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Poststratification"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html",
+    "href": "stan-users-guide/posterior-prediction.html",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "The goal of inference is often posterior prediction, that is evaluating or sampling from the posterior predictive distribution \\(p(\\tilde{y} \\mid y),\\) where \\(y\\) is observed data and \\(\\tilde{y}\\) is yet to be observed data. Often there are unmodeled predictors \\(x\\) and \\(\\tilde{x}\\) for the observed data \\(y\\) and unobserved data \\(\\tilde{y}\\). With predictors, the posterior predictive density is \\(p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\\) All of these variables may represent multivariate quantities.\nThis chapter explains how to sample from the posterior predictive distribution in Stan, including applications to posterior predictive simulation and calculating event probabilities. These techniques can be coded in Stan using random number generation in the generated quantities block. Further, a technique for fitting and performing inference in two stages is presented in a section on stand-alone generated quantities in Stan\n\n\nGiven a full Bayesian model \\(p(y, \\theta)\\), the posterior predictive density for new data \\(\\tilde{y}\\) given observed data \\(y\\) is \\[\np(\\tilde{y} \\mid y)\n=\n\\int p(\\tilde{y} \\mid \\theta) \\cdot p(\\theta \\mid y)\n\\, \\textrm{d}\\theta.\n\\] The product under the integral reduces to the joint posterior density \\(p(\\tilde{y}, \\theta \\mid y),\\) so that the integral is simply marginalizing out the parameters \\(\\theta,\\) leaving the predictive density \\(p(\\tilde{y} \\mid y)\\) of future observations given past observations.\n\n\n\nThe posterior predictive density (or mass) of a prediction \\(\\tilde{y}\\) given observed data \\(y\\) can be computed using \\(M\\) Monte Carlo draws\n\\[\n\\theta^{(m)} \\sim p(\\theta \\mid y)\n\\] from the posterior as \\[\np(\\tilde{y} \\mid y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\]\nComputing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, \\(\\log\np(\\tilde{y} \\mid y)\\) may be computed using the stable log sum of exponents function as \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\\\[4pt]\n& = &\n- \\log M\n+ \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\theta^{(m)}),\n\\end{eqnarray*}\\] where \\[\n\\textrm{log-sum-exp}_{m = 1}^M v_m\n= \\log \\sum_{m = 1}^M \\exp v_m\n\\] is used to maintain arithmetic precision. See the section on log sum of exponentials for more details.\n\n\n\nGiven draws from the posterior \\(\\theta^{(m)} \\sim p(\\theta \\mid y),\\) draws from the posterior predictive \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid y)\\) can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\theta^{(m)}).\n\\]\nRandomly drawing \\(\\tilde{y}\\) from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because \\(\\theta\\) is unknown and estimated based only on a finite sample of data \\(y\\). Aleatoric uncertainty arises because even a known value of \\(\\theta\\) leads to uncertainty about new \\(\\tilde{y}\\) as described by the data model \\(p(\\tilde{y} \\mid \\theta)\\). Both forms of uncertainty show up in the factored form of the posterior predictive distribution, \\[\np(\\tilde{y} \\mid y)\n=\n\\int\n\\underbrace{p(\\tilde{y} \\mid \\theta)}_{\\begin{array}{l}\n                                         \\textrm{aleatoric}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\cdot \\underbrace{p(\\theta \\mid y)}_{\\begin{array}{l}\n                                         \\textrm{epistemic}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\, \\textrm{d}\\theta.\n\\]\n\n\n\nPosterior predictive quantities can be coded in Stan using the generated quantities block.\n\n\nFor example, consider a simple Poisson model for count data with a rate parameter \\(\\lambda &gt; 0\\) having a gamma-distributed prior, \\[\n\\lambda \\sim \\textrm{gamma}(1, 1).\n\\] The \\(N\\) observations \\(y_1, \\ldots, y_N\\) are modeled as Poisson distributed, \\[\ny_n \\sim \\textrm{poisson}(\\lambda).\n\\]\n\n\n\nThe following Stan program defines a variable for \\(\\tilde{y}\\) by random number generation in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  lambda ~ gamma(1, 1);\n  y ~ poisson(lambda);\n}\ngenerated quantities {\n  int&lt;lower=0&gt; y_tilde = poisson_rng(lambda);\n}\nThe random draw from the data model for \\(\\tilde{y}\\) is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new \\(\\tilde{y}^{(m)} \\sim p(y \\mid\n\\lambda^{(m)})\\) for each posterior draw \\(\\lambda^{(m)} \\sim p(\\theta\n\\mid y).\\)\nThe posterior draws \\(\\tilde{y}^{(m)}\\) may be used to estimate the expected value of \\(\\tilde{y}\\) or any of its quantiles or posterior intervals, as well as event probabilities involving \\(\\tilde{y}\\). In general, \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\) may be evaluated as \\[\n\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\n\\approx \\frac{1}{M} \\sum_{m=1}^M f(\\tilde{y}^{(m)}, \\theta^{(m)}),\n\\] which is just the posterior mean of \\(f(\\tilde{y}, \\theta).\\) This quantity is computed by Stan if the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to a variable in the generated quantities block. That is, if we have\ngenerated quantities {\n  real f_val = f(y_tilde, theta);\n  // ...\n}\nwhere the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to variable f_val, then the posterior mean of f_val will be the expectation \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\).\n\n\n\nThe gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density \\(p(\\lambda \\mid y)\\) will also follow a gamma distribution.\nBecause the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive \\(p(\\tilde{y} \\mid\ny)\\) will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, \\(y \\sim\n\\textrm{negative-binomial}(\\alpha, \\beta)\\) if \\(\\lambda \\sim\n\\textrm{gamma}(\\alpha, \\beta)\\) and \\(y \\sim \\textrm{poisson}(\\lambda).\\) Rather than marginalizing out the rate parameter \\(\\lambda\\) analytically as can be done to define the negative binomial probability mass function, the rate \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is sampled from the posterior and then used to generate a draw of \\(\\tilde{y}^{(m)} \\sim p(y \\mid \\lambda^{(m)}).\\)\n\n\n\n\n\n\nConsider a regression with a single predictor \\(x_n\\) for the training outcome \\(y_n\\) and \\(\\tilde{x}_n\\) for the test outcome \\(\\tilde{y}_n.\\) Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in \\[\\begin{eqnarray}\np(\\tilde{y} \\mid \\tilde{x}, y, x)\n& = & \\int p(\\tilde{y} \\mid \\tilde{x}, \\theta) \\cdot p(\\theta \\mid y, x) \\,\n\\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m=1}^M \\, p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\end{eqnarray}\\] where \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y).\\)\n\n\n\nThe following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed \\(y\\) values are coded as data. The predictive quantities \\(\\tilde{y}\\) appear in the generated quantities block, where they are generated by random number generation.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0&gt; y;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ poisson_log(alpha + beta * x);\n  { alpha, beta } ~ normal(0, 1);\n}\ngenerated quantities {\n  array[N_tilde] int&lt;lower=0&gt; y_tilde\n    = poisson_log_rng(alpha + beta * x_tilde);\n}\nThe Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s poisson_log vs. poisson, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept alpha and slope beta, are given standard normal priors.\nIn the model block, the log rate for the Poisson is a linear function of the training data \\(x\\), whereas in the generated quantities block it is a function of the test data \\(\\tilde{x}\\). Because the generated quantities block does not affect the posterior draws, the model fits \\(\\alpha\\) and \\(\\beta\\) using only the training data, reserving \\(\\tilde{x}\\) to generate \\(\\tilde{y}.\\)\nThe result from running Stan is a predictive sample \\(\\tilde{y}^{(1)},\n\\ldots \\tilde{y}^{(M)}\\) where each \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid \\tilde{x}, x, y).\\)\nThe mean of the posterior predictive distribution is the expected value \\[\\begin{align}\n\\mathbb{E}[\\tilde{y} \\mid \\tilde{x}, x, y]\n& =\n\\int\n\\tilde{y}\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n\\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}^{(m)},\n\\end{align}\\] where the \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y)\\) are drawn from the posterior predictive distribution. Thus the posterior mean of y_tilde[n] after running Stan is the expected value of \\(\\tilde{y}_n\\) conditioned on the training data \\(x, y\\) and predictor \\(\\tilde{x}_n.\\) This is the Bayesian estimate for \\(\\tilde{y}\\) with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for \\(\\tilde{y}\\), as well as covariance of the \\(\\tilde{y_n}.\\) The posterior draws \\(\\tilde{y}^{(m)}\\) may also be used to estimate predictive event probabilities, such as \\(\\Pr[\\tilde{y}_1 &gt; 0]\\) or \\(\\Pr[\\prod_{n =\n1}^{\\tilde{N}}(\\tilde{y_n}) &gt; 1],\\) as expectations of indicator functions.\nAll of this can be carried out by running Stan only a single time to draw a single sample of \\(M\\) draws, \\[\n\\tilde{y}^{(1)}, \\ldots, \\tilde{y}^{(M)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\n\\] It’s only when moving to cross-validation where multiple runs are required.\n\n\n\n\nEvent probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate \\(\\Pr[\\lambda &gt; 5 \\mid y]\\) in the simple Poisson example with only a rate parameter \\(\\lambda\\), it suffices to define a generated quantity\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; lambda_gt_5 = lambda &gt; 5;\n  // ...\n}\nThe value of the expression lambda &gt; 5 is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = &\n\\int \\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}[\\lambda^{(m)} &gt; 5],\n\\end{eqnarray*}\\] where each \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter lambda_gt_5.\nIn general, event probabilities may be expressed as expectations of indicator functions. For example, \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = & \\mathbb{E}[\\textrm{I}[\\lambda &gt; 5] \\mid y]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}(\\lambda^{(m)} &gt; 5).\n\\end{eqnarray*}\\] The last line above is the posterior mean of the indicator function as coded in Stan.\nEvent probabilities involving posterior predictive quantities \\(\\tilde{y}\\) work exactly the same way as those for parameters. For example, if \\(\\tilde{y}_n\\) is the prediction for the \\(n\\)-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then \\[\\begin{eqnarray*}\n\\Pr[\\tilde{y}_3 &gt; \\tilde{y}_7 \\mid \\tilde{x}, x, y]\n& = &\n\\mathbb{E}\\!\\left[I[\\tilde{y}_3 &gt; \\tilde{y}_7] \\mid \\tilde{x}, x, y\\right]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\tilde{y}_3 &gt; \\tilde{y}_7)\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\, \\textrm{d}\\tilde{y}\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n\\textrm{I}(\\tilde{y}^{(m)}_3 &gt; \\tilde{y}^{(m)}_7),\n\\end{eqnarray*}\\] where \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y).\\)\n\n\n\nStan’s sampling algorithms take a Stan program representing a posterior \\(p(\\theta \\mid y, x)\\) along with actual data \\(x\\) and \\(y\\) to produce a set of draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)}\\) from the posterior. Posterior predictive draws \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y)\\) can be generated by drawing \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\tilde{x}, \\theta^{(m)})\n\\] from the data model. Note that drawing \\(\\tilde{y}^{(m)}\\) only depends on the new predictors \\(\\tilde{x}\\) and the posterior draws \\(\\theta^{(m)}\\). Most importantly, neither the original data or the model density is required.\nBy saving the posterior draws, predictions for new data items \\(\\tilde{x}\\) may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 5);\n  beta ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.5);\n}\nA second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde\n    = normal_rng(alpha + beta * x_tilde, sigma);\n}\nWhen running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for \\(\\tilde{y}\\) or derived quantities such as event probabilities.\nAny posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-prediction.html#posterior-predictive-distribution",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Given a full Bayesian model \\(p(y, \\theta)\\), the posterior predictive density for new data \\(\\tilde{y}\\) given observed data \\(y\\) is \\[\np(\\tilde{y} \\mid y)\n=\n\\int p(\\tilde{y} \\mid \\theta) \\cdot p(\\theta \\mid y)\n\\, \\textrm{d}\\theta.\n\\] The product under the integral reduces to the joint posterior density \\(p(\\tilde{y}, \\theta \\mid y),\\) so that the integral is simply marginalizing out the parameters \\(\\theta,\\) leaving the predictive density \\(p(\\tilde{y} \\mid y)\\) of future observations given past observations.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#computing-the-posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-prediction.html#computing-the-posterior-predictive-distribution",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "The posterior predictive density (or mass) of a prediction \\(\\tilde{y}\\) given observed data \\(y\\) can be computed using \\(M\\) Monte Carlo draws\n\\[\n\\theta^{(m)} \\sim p(\\theta \\mid y)\n\\] from the posterior as \\[\np(\\tilde{y} \\mid y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\]\nComputing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, \\(\\log\np(\\tilde{y} \\mid y)\\) may be computed using the stable log sum of exponents function as \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\\\[4pt]\n& = &\n- \\log M\n+ \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\theta^{(m)}),\n\\end{eqnarray*}\\] where \\[\n\\textrm{log-sum-exp}_{m = 1}^M v_m\n= \\log \\sum_{m = 1}^M \\exp v_m\n\\] is used to maintain arithmetic precision. See the section on log sum of exponentials for more details.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#sampling-from-the-posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-prediction.html#sampling-from-the-posterior-predictive-distribution",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Given draws from the posterior \\(\\theta^{(m)} \\sim p(\\theta \\mid y),\\) draws from the posterior predictive \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid y)\\) can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\theta^{(m)}).\n\\]\nRandomly drawing \\(\\tilde{y}\\) from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because \\(\\theta\\) is unknown and estimated based only on a finite sample of data \\(y\\). Aleatoric uncertainty arises because even a known value of \\(\\theta\\) leads to uncertainty about new \\(\\tilde{y}\\) as described by the data model \\(p(\\tilde{y} \\mid \\theta)\\). Both forms of uncertainty show up in the factored form of the posterior predictive distribution, \\[\np(\\tilde{y} \\mid y)\n=\n\\int\n\\underbrace{p(\\tilde{y} \\mid \\theta)}_{\\begin{array}{l}\n                                         \\textrm{aleatoric}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\cdot \\underbrace{p(\\theta \\mid y)}_{\\begin{array}{l}\n                                         \\textrm{epistemic}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\, \\textrm{d}\\theta.\n\\]",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan",
+    "href": "stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Posterior predictive quantities can be coded in Stan using the generated quantities block.\n\n\nFor example, consider a simple Poisson model for count data with a rate parameter \\(\\lambda &gt; 0\\) having a gamma-distributed prior, \\[\n\\lambda \\sim \\textrm{gamma}(1, 1).\n\\] The \\(N\\) observations \\(y_1, \\ldots, y_N\\) are modeled as Poisson distributed, \\[\ny_n \\sim \\textrm{poisson}(\\lambda).\n\\]\n\n\n\nThe following Stan program defines a variable for \\(\\tilde{y}\\) by random number generation in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  lambda ~ gamma(1, 1);\n  y ~ poisson(lambda);\n}\ngenerated quantities {\n  int&lt;lower=0&gt; y_tilde = poisson_rng(lambda);\n}\nThe random draw from the data model for \\(\\tilde{y}\\) is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new \\(\\tilde{y}^{(m)} \\sim p(y \\mid\n\\lambda^{(m)})\\) for each posterior draw \\(\\lambda^{(m)} \\sim p(\\theta\n\\mid y).\\)\nThe posterior draws \\(\\tilde{y}^{(m)}\\) may be used to estimate the expected value of \\(\\tilde{y}\\) or any of its quantiles or posterior intervals, as well as event probabilities involving \\(\\tilde{y}\\). In general, \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\) may be evaluated as \\[\n\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\n\\approx \\frac{1}{M} \\sum_{m=1}^M f(\\tilde{y}^{(m)}, \\theta^{(m)}),\n\\] which is just the posterior mean of \\(f(\\tilde{y}, \\theta).\\) This quantity is computed by Stan if the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to a variable in the generated quantities block. That is, if we have\ngenerated quantities {\n  real f_val = f(y_tilde, theta);\n  // ...\n}\nwhere the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to variable f_val, then the posterior mean of f_val will be the expectation \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\).\n\n\n\nThe gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density \\(p(\\lambda \\mid y)\\) will also follow a gamma distribution.\nBecause the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive \\(p(\\tilde{y} \\mid\ny)\\) will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, \\(y \\sim\n\\textrm{negative-binomial}(\\alpha, \\beta)\\) if \\(\\lambda \\sim\n\\textrm{gamma}(\\alpha, \\beta)\\) and \\(y \\sim \\textrm{poisson}(\\lambda).\\) Rather than marginalizing out the rate parameter \\(\\lambda\\) analytically as can be done to define the negative binomial probability mass function, the rate \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is sampled from the posterior and then used to generate a draw of \\(\\tilde{y}^{(m)} \\sim p(y \\mid \\lambda^{(m)}).\\)",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#posterior-prediction-for-regressions",
+    "href": "stan-users-guide/posterior-prediction.html#posterior-prediction-for-regressions",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Consider a regression with a single predictor \\(x_n\\) for the training outcome \\(y_n\\) and \\(\\tilde{x}_n\\) for the test outcome \\(\\tilde{y}_n.\\) Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in \\[\\begin{eqnarray}\np(\\tilde{y} \\mid \\tilde{x}, y, x)\n& = & \\int p(\\tilde{y} \\mid \\tilde{x}, \\theta) \\cdot p(\\theta \\mid y, x) \\,\n\\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m=1}^M \\, p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\end{eqnarray}\\] where \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y).\\)\n\n\n\nThe following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed \\(y\\) values are coded as data. The predictive quantities \\(\\tilde{y}\\) appear in the generated quantities block, where they are generated by random number generation.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0&gt; y;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ poisson_log(alpha + beta * x);\n  { alpha, beta } ~ normal(0, 1);\n}\ngenerated quantities {\n  array[N_tilde] int&lt;lower=0&gt; y_tilde\n    = poisson_log_rng(alpha + beta * x_tilde);\n}\nThe Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s poisson_log vs. poisson, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept alpha and slope beta, are given standard normal priors.\nIn the model block, the log rate for the Poisson is a linear function of the training data \\(x\\), whereas in the generated quantities block it is a function of the test data \\(\\tilde{x}\\). Because the generated quantities block does not affect the posterior draws, the model fits \\(\\alpha\\) and \\(\\beta\\) using only the training data, reserving \\(\\tilde{x}\\) to generate \\(\\tilde{y}.\\)\nThe result from running Stan is a predictive sample \\(\\tilde{y}^{(1)},\n\\ldots \\tilde{y}^{(M)}\\) where each \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid \\tilde{x}, x, y).\\)\nThe mean of the posterior predictive distribution is the expected value \\[\\begin{align}\n\\mathbb{E}[\\tilde{y} \\mid \\tilde{x}, x, y]\n& =\n\\int\n\\tilde{y}\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n\\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}^{(m)},\n\\end{align}\\] where the \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y)\\) are drawn from the posterior predictive distribution. Thus the posterior mean of y_tilde[n] after running Stan is the expected value of \\(\\tilde{y}_n\\) conditioned on the training data \\(x, y\\) and predictor \\(\\tilde{x}_n.\\) This is the Bayesian estimate for \\(\\tilde{y}\\) with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for \\(\\tilde{y}\\), as well as covariance of the \\(\\tilde{y_n}.\\) The posterior draws \\(\\tilde{y}^{(m)}\\) may also be used to estimate predictive event probabilities, such as \\(\\Pr[\\tilde{y}_1 &gt; 0]\\) or \\(\\Pr[\\prod_{n =\n1}^{\\tilde{N}}(\\tilde{y_n}) &gt; 1],\\) as expectations of indicator functions.\nAll of this can be carried out by running Stan only a single time to draw a single sample of \\(M\\) draws, \\[\n\\tilde{y}^{(1)}, \\ldots, \\tilde{y}^{(M)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\n\\] It’s only when moving to cross-validation where multiple runs are required.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#estimating-event-probabilities",
+    "href": "stan-users-guide/posterior-prediction.html#estimating-event-probabilities",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Event probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate \\(\\Pr[\\lambda &gt; 5 \\mid y]\\) in the simple Poisson example with only a rate parameter \\(\\lambda\\), it suffices to define a generated quantity\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; lambda_gt_5 = lambda &gt; 5;\n  // ...\n}\nThe value of the expression lambda &gt; 5 is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = &\n\\int \\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}[\\lambda^{(m)} &gt; 5],\n\\end{eqnarray*}\\] where each \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter lambda_gt_5.\nIn general, event probabilities may be expressed as expectations of indicator functions. For example, \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = & \\mathbb{E}[\\textrm{I}[\\lambda &gt; 5] \\mid y]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}(\\lambda^{(m)} &gt; 5).\n\\end{eqnarray*}\\] The last line above is the posterior mean of the indicator function as coded in Stan.\nEvent probabilities involving posterior predictive quantities \\(\\tilde{y}\\) work exactly the same way as those for parameters. For example, if \\(\\tilde{y}_n\\) is the prediction for the \\(n\\)-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then \\[\\begin{eqnarray*}\n\\Pr[\\tilde{y}_3 &gt; \\tilde{y}_7 \\mid \\tilde{x}, x, y]\n& = &\n\\mathbb{E}\\!\\left[I[\\tilde{y}_3 &gt; \\tilde{y}_7] \\mid \\tilde{x}, x, y\\right]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\tilde{y}_3 &gt; \\tilde{y}_7)\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\, \\textrm{d}\\tilde{y}\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n\\textrm{I}(\\tilde{y}^{(m)}_3 &gt; \\tilde{y}^{(m)}_7),\n\\end{eqnarray*}\\] where \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y).\\)",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction",
+    "href": "stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Stan’s sampling algorithms take a Stan program representing a posterior \\(p(\\theta \\mid y, x)\\) along with actual data \\(x\\) and \\(y\\) to produce a set of draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)}\\) from the posterior. Posterior predictive draws \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y)\\) can be generated by drawing \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\tilde{x}, \\theta^{(m)})\n\\] from the data model. Note that drawing \\(\\tilde{y}^{(m)}\\) only depends on the new predictors \\(\\tilde{x}\\) and the posterior draws \\(\\theta^{(m)}\\). Most importantly, neither the original data or the model density is required.\nBy saving the posterior draws, predictions for new data items \\(\\tilde{x}\\) may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 5);\n  beta ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.5);\n}\nA second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde\n    = normal_rng(alpha + beta * x_tilde, sigma);\n}\nWhen running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for \\(\\tilde{y}\\) or derived quantities such as event probabilities.\nAny posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/one-dimensional-integrals.html",
+    "href": "stan-users-guide/one-dimensional-integrals.html",
+    "title": "Computing One Dimensional Integrals",
+    "section": "",
+    "text": "Definite and indefinite one dimensional integrals can be performed in Stan using the integrate_1d function\nAs an example, the normalizing constant of a left-truncated normal distribution is\n\\[\n  \\int_a^\\infty \\frac{1}{\\sqrt{2 \\pi \\sigma^2}} e^{-\\frac{1}{2}\\frac{(x - \\mu)^2}{\\sigma^2}}\n\\]\nTo compute this integral in Stan, the integrand must first be defined as a Stan function (see the Stan Reference Manual chapter on User-Defined Functions for more information on coding user-defined functions).\nreal normal_density(real x,             // Function argument\n                    real xc,            // Complement of function argument\n                                        //  on the domain (defined later)\n                    array[] real theta, // parameters\n                    array[] real x_r,   // data (real)\n                    array[] int x_i) {  // data (integer)\n  real mu = theta[1];\n  real sigma = theta[2];\n\n  return 1 / (sqrt(2 * pi()) * sigma) * exp(-0.5 * ((x - mu) / sigma)^2);\n}\nThis function is expected to return the value of the integrand evaluated at point x. The argument xc is used in definite integrals to avoid loss of precision near the limits of integration and is set to NaN when either limit is infinite (see the section on precision/loss in the chapter on Higher-Order Functions of the Stan Functions Reference for details on how to use this). The argument theta is used to pass in arguments of the integral that are a function of the parameters in our model. The arguments x_r and x_i are used to pass in real and integer arguments of the integral that are not a function of our parameters.\nThe function defining the integrand must have exactly the argument types and return type of normal_density above, though argument naming is not important. Even if x_r and x_i are unused in the integrand, they must be included in the function signature. Even if the integral does not involve some of these, they must still be supplied some value. The most efficient will be a zero-length array or vector, which can be created with rep_array(0, 0) and rep_vector(0, 0), respectively. Other options include an uninitialized variable declared with size 0, which is equivalent to the above, or any easy value, such as size 1 array created with {0}.\n\n\nSuppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.\nSuch a model might look like (include the function defined at the beginning of this chapter to make this code compile):\ndata {\n  int N;\n  array[N] real y;\n}\n\ntransformed data {\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\nparameters {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n  real left_limit;\n}\n\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  left_limit ~ normal(0, 1);\n  target += normal_lpdf(y | mu, sigma);\n  target += -log(integrate_1d(normal_density,\n                              left_limit,\n                              positive_infinity(),\n                              { mu, sigma }, x_r, x_i));\n}\n\n\nThe limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls negative_infinity() and positive_infinity().\nIf both limits are either negative_infinity() or positive_infinity(), the integral and its gradients are set to zero.\n\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.\nThe endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).\n\n\n\n\nThe integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library (Agrawal et al. 2017). If the \\(n\\)th estimate of the integral is denoted \\(I_n\\) and the \\(n\\)th estimate of the norm of the integral is denoted \\(|I|_n\\), the iteration is terminated when\n\\[\n  \\frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \\text{relative tolerance}.\n\\]\nThe relative_tolerance parameter can be optionally specified as the last argument to integrate_1d. By default, integrate_1d follows the Boost library recommendation of setting relative_tolerance to the square root of the machine epsilon of double precision floating point numbers (about 1e-8). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If integrate_1d causes an exception in transformed parameters block or model block, the result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If integrate_1d causes an exception in generated quantities block, the returned output from integrate_1d is NaN. In these cases, a bigger relative_tolerance value can be specified.\n\n\nIntegrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). This is because the quadrature methods employed internally can have difficulty near zero.\nIn this case, each integral is separately integrated to the given relative_tolerance.\n\n\n\nIf care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.\nFor instance, in integrating the pdf of a beta distribution when the values of \\(\\alpha\\) and \\(\\beta\\) are small, most of the probability mass is lumped near zero and one.\nThe pdf of a beta distribution is proportional to\n\\[\np(x) \\propto x^{\\alpha - 1}(1 - x)^{\\beta - 1}\n\\]\nNormalizing this distribution requires computing the integral of \\(p(x)\\) from zero to one. In Stan code, the integrand might look like:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n\n  return x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n}\nThe issue is that there will be numerical breakdown in the precision of 1.0 - x as x gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of alpha and beta much less than one.\nThis is where xc is useful. It is defined, for definite integrals, as a high precision version of the distance from x to the nearest endpoint — a - x or b - x for a lower endpoint a and an upper endpoint b. To make use of this for the beta integral, the integrand can be re-coded:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n  real v;\n\n  if(x &gt; 0.5) {\n    v = x^(alpha - 1.0) * xc^(beta - 1.0);\n  } else {\n    v = x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n  }\n\n  return v;\n}\nIn this case, as we approach the upper limit of integration \\(a = 1\\), xc will take on the value of \\(a - x = 1 - x\\). This version of the integrand will converge for much smaller values of alpha and beta than otherwise possible.\nConsider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount \\(\\delta\\), and truncated at some value \\(b\\). If we were interested in calculating the expectation of a variable \\(X\\) distributed in this way, we would need to calculate \\[\n\\int_a^b xf(x)\\,dx = \\int_{\\delta}^b xf(x)\\,dx\n\\] in the numerator, where \\(f(x)\\) is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:\nreal shift_lognormal_pdf(real x,\n                         real mu,\n                         real sigma,\n                         real delta) {\n  real p;\n\n  p = (1.0 / ((x - delta) * sigma * sqrt(2 * pi()))) *\n    exp(-1 * (log(x - delta) - mu)^2 / (2 * sigma^2));\n\n  return p;\n}\nTherefore, the function that we want to integrate is:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  p = shift_lognormal_pdf(x, mu, sigma, delta);\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhat happens here is that, given that the log-normal distribution is shifted by \\(\\delta\\), when we then try to integrate the numerator, our x starts at values just above delta. This, in turn, causes the x - delta term to be near zero, leading to a breakdown.\nWe can use xc, and define the integrand as:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  if (x &lt; delta + 1) {\n    p = shift_lognormal_pdf(xc, mu, sigma, delta);\n  } else {\n    p = shift_lognormal_pdf(x, mu, sigma, delta);\n  }\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhy does this work? When our values of x are less than delta + 1 (so, when they’re near delta, given that our lower bound of integration is equal to \\(\\delta\\)), we pass xc as an argument to our shift_lognormal_pdf function. This way, instead of dealing with x - delta in shift_lognormal_pdf, we are working with xc - delta which is equal to delta - x - delta, as delta is the lower endpoint in that case. The delta terms cancel out, and we are left with a high-precision version of x. We don’t encounter the same problem at the upper limit \\(b\\) so we don’t adjust the code for that case.\nNote, xc is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, xc will be NaN.\nFor zero-crossing definite integrals (see section Zero Crossing) the integrals are broken into two pieces (\\((a, 0)\\) and \\((0, b)\\) for endpoints \\(a &lt; 0\\) and \\(b &gt; 0\\)) and xc is a high precision version of the distance to the limits of each of the two integrals separately. This means xc will be a high precision version of a - x, x, or b - x, depending on the value of x and the endpoints.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Computing One Dimensional Integrals"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/one-dimensional-integrals.html#calling-the-integrator",
+    "href": "stan-users-guide/one-dimensional-integrals.html#calling-the-integrator",
+    "title": "Computing One Dimensional Integrals",
+    "section": "",
+    "text": "Suppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.\nSuch a model might look like (include the function defined at the beginning of this chapter to make this code compile):\ndata {\n  int N;\n  array[N] real y;\n}\n\ntransformed data {\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\nparameters {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n  real left_limit;\n}\n\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  left_limit ~ normal(0, 1);\n  target += normal_lpdf(y | mu, sigma);\n  target += -log(integrate_1d(normal_density,\n                              left_limit,\n                              positive_infinity(),\n                              { mu, sigma }, x_r, x_i));\n}\n\n\nThe limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls negative_infinity() and positive_infinity().\nIf both limits are either negative_infinity() or positive_infinity(), the integral and its gradients are set to zero.\n\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.\nThe endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Computing One Dimensional Integrals"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/one-dimensional-integrals.html#integrator-convergence",
+    "href": "stan-users-guide/one-dimensional-integrals.html#integrator-convergence",
+    "title": "Computing One Dimensional Integrals",
+    "section": "",
+    "text": "The integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library (Agrawal et al. 2017). If the \\(n\\)th estimate of the integral is denoted \\(I_n\\) and the \\(n\\)th estimate of the norm of the integral is denoted \\(|I|_n\\), the iteration is terminated when\n\\[\n  \\frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \\text{relative tolerance}.\n\\]\nThe relative_tolerance parameter can be optionally specified as the last argument to integrate_1d. By default, integrate_1d follows the Boost library recommendation of setting relative_tolerance to the square root of the machine epsilon of double precision floating point numbers (about 1e-8). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If integrate_1d causes an exception in transformed parameters block or model block, the result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If integrate_1d causes an exception in generated quantities block, the returned output from integrate_1d is NaN. In these cases, a bigger relative_tolerance value can be specified.\n\n\nIntegrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). This is because the quadrature methods employed internally can have difficulty near zero.\nIn this case, each integral is separately integrated to the given relative_tolerance.\n\n\n\nIf care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.\nFor instance, in integrating the pdf of a beta distribution when the values of \\(\\alpha\\) and \\(\\beta\\) are small, most of the probability mass is lumped near zero and one.\nThe pdf of a beta distribution is proportional to\n\\[\np(x) \\propto x^{\\alpha - 1}(1 - x)^{\\beta - 1}\n\\]\nNormalizing this distribution requires computing the integral of \\(p(x)\\) from zero to one. In Stan code, the integrand might look like:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n\n  return x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n}\nThe issue is that there will be numerical breakdown in the precision of 1.0 - x as x gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of alpha and beta much less than one.\nThis is where xc is useful. It is defined, for definite integrals, as a high precision version of the distance from x to the nearest endpoint — a - x or b - x for a lower endpoint a and an upper endpoint b. To make use of this for the beta integral, the integrand can be re-coded:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n  real v;\n\n  if(x &gt; 0.5) {\n    v = x^(alpha - 1.0) * xc^(beta - 1.0);\n  } else {\n    v = x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n  }\n\n  return v;\n}\nIn this case, as we approach the upper limit of integration \\(a = 1\\), xc will take on the value of \\(a - x = 1 - x\\). This version of the integrand will converge for much smaller values of alpha and beta than otherwise possible.\nConsider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount \\(\\delta\\), and truncated at some value \\(b\\). If we were interested in calculating the expectation of a variable \\(X\\) distributed in this way, we would need to calculate \\[\n\\int_a^b xf(x)\\,dx = \\int_{\\delta}^b xf(x)\\,dx\n\\] in the numerator, where \\(f(x)\\) is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:\nreal shift_lognormal_pdf(real x,\n                         real mu,\n                         real sigma,\n                         real delta) {\n  real p;\n\n  p = (1.0 / ((x - delta) * sigma * sqrt(2 * pi()))) *\n    exp(-1 * (log(x - delta) - mu)^2 / (2 * sigma^2));\n\n  return p;\n}\nTherefore, the function that we want to integrate is:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  p = shift_lognormal_pdf(x, mu, sigma, delta);\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhat happens here is that, given that the log-normal distribution is shifted by \\(\\delta\\), when we then try to integrate the numerator, our x starts at values just above delta. This, in turn, causes the x - delta term to be near zero, leading to a breakdown.\nWe can use xc, and define the integrand as:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  if (x &lt; delta + 1) {\n    p = shift_lognormal_pdf(xc, mu, sigma, delta);\n  } else {\n    p = shift_lognormal_pdf(x, mu, sigma, delta);\n  }\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhy does this work? When our values of x are less than delta + 1 (so, when they’re near delta, given that our lower bound of integration is equal to \\(\\delta\\)), we pass xc as an argument to our shift_lognormal_pdf function. This way, instead of dealing with x - delta in shift_lognormal_pdf, we are working with xc - delta which is equal to delta - x - delta, as delta is the lower endpoint in that case. The delta terms cancel out, and we are left with a high-precision version of x. We don’t encounter the same problem at the upper limit \\(b\\) so we don’t adjust the code for that case.\nNote, xc is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, xc will be NaN.\nFor zero-crossing definite integrals (see section Zero Crossing) the integrals are broken into two pieces (\\((a, 0)\\) and \\((0, b)\\) for endpoints \\(a &lt; 0\\) and \\(b &gt; 0\\)) and xc is a high precision version of the distance to the limits of each of the two integrals separately. This means xc will be a high precision version of a - x, x, or b - x, depending on the value of x and the endpoints.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Computing One Dimensional Integrals"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html",
+    "href": "stan-users-guide/multiple-imputation.html",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "Missing data is common in applied data analysis. Ignoring this missingness can distort posterior inferences and reduce their precision, and modeling it can improve inference quality. There are several ways to model missing data. Chapter 18 in Gelman et al. (2013) offers an approximately Bayesian perspective, and the chapter Missing Data and Partially Known Parameters in the Reference Manual shows an approach that handles missing data as parameters.\nAnother way to model missing data is through multiple imputation, which replaces missing values with sampled values to obtain different versions of a complete data set. Then, the model of interest can be fitted to each of these complete versions separately. Combining the resulting draws from these different fits produces a sample that accounts for the uncertainty in the missing values.\n\n\nSuppose that we have a data set \\(x\\) with columns \\(x_{\\cdot, 1}, \\ldots, x_{\\cdot, K}\\) that make up covariates in a regression with quantities of interest \\(\\theta\\). With the completely observed data, \\(x^\\text{comp}\\), we could estimate the posterior distribution of \\(\\theta\\) as \\(p(\\theta \\mid x^\\text{comp})\\). But with missing data, our matrix is split into \\(x^{\\text{obs}}\\) (the observed values of \\(x\\)) and \\(x^{\\text{mis}}\\) (the missing values of \\(x\\)).\nFortunately, we can treat \\(x^{\\text{mis}}\\) as additional, nuisance parameters that we can estimate along with \\(\\theta\\) as \\(p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\\). So, we can express the marginal distribution of \\(\\theta\\) given only the observed values \\(x^{\\text{obs}}\\) as \\[\\begin{align*}\np(\\theta \\mid x^{\\text{obs}}) =&\n\\int\n    p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=&\n\\int\n    p(\\theta \\mid x^{\\text{obs}}, x^{\\text{mis}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=& \\int\n    p(\\theta \\mid x^{\\text{imp}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}},\n\\end{align*}\\] where \\(x^{\\text{imp}}\\) is a data set that includes imputed values of \\(x^{\\text{mis}}\\).\nThe equations above show that we do not need to describe \\(p(\\theta \\mid x^{\\text{obs}})\\) directly. Instead, we can first find a way to sample from \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), and then use these samples to fit the model \\(p(\\theta \\mid x^\\text{imp})\\), treating \\(x^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nNote that the model for \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\) needs to be explicit. In a typical regression setting with covariates \\(x\\) and outcome \\(y\\), this model for \\(x\\) is not needed because inferences for the regression parameters are independent of the model of \\(x\\) when \\(x\\) is fully observed. But with missing data and multiple imputation, we need to generate values for \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), which in turn needs an explicit model for \\(x^{\\text{mis}}\\).\nThus, the general outline for multiple imputation with a number \\(M\\) of imputations is:\n\nDraw \\(M\\) random samples \\(x^{\\text{mis}}_1, \\ldots,\nx^{\\text{mis}}_M\\) from the posterior predictive distribution \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\).\nUse these samples to get \\(M\\) different complete data sets \\(x_1^\\text{imp}, \\ldots, x_M^\\text{imp}\\).\nFor each of the \\(M\\) imputed data sets, sample from the model of interest \\(p(\\theta \\mid x_m^\\text{imp})\\), each time treating \\(x_m^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nCombine the draws for \\(\\theta\\) from all \\(M\\) fits.\n\n\n\nFollowing the notation above, we can express the outline with the pseudocode below:\ntheta_estimates = []\nfor (m in 1:M) {\n  x_mis &lt;- get_imputations(x_obs)\n  x_imp &lt;- add_imputations(x_obs, x_mis)\n  imputation_fits &lt;- my_model.fit(x_imp).draws()\n  theta_estimates &lt;- theta_estimates.append(imputation_fits)\n}\nEach of the functions in this pseudocode can encapsulate simple or complex procedures. get_imputations(), for example, may impute missing data by taking values directly from x_obs; or it may model a regression where an \\(x_{\\cdot, k}\\) with missing values is the outcome and the \\(x_{\\cdot, j}\\) without missing observations are predictors. Below we illustrate how to implement this second approach in Stan.\n\n\n\n\nImagine our data set is composed of two numerical, continuous variables \\(x\\) and \\(y\\), and one binary variable \\(z\\). We want to estimate the parameters that govern the conditional association \\(p(y \\mid x, z)\\), but several values of \\(x\\) are missing. So, before we fit this model for \\(y\\), we impute the missing values in \\(x\\).\nWe first find all the fully observed data points and identify their values as \\(x^{\\text{obs}}\\), \\(y^{\\text{obs}}\\), and \\(z^{\\text{obs}}\\). Then we use these data to fit the model \\[\nx^{\\text{obs}} \\sim \\text{normal}(\\gamma_0 + \\gamma_1 y^{\\text{obs}} +\n    \\gamma_2 z^{\\text{obs}}, \\lambda).\n\\]\nThis model can then give us samples for the missing values \\(x^{\\text{mis}}\\) conditional on the corresponding values of \\(y\\) and \\(z\\), which we call \\(y^{\\text{aux}}\\) and \\(z^{\\text{aux}}\\). The Stan code for this imputation is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  vector[N_obs] x_obs;\n  vector[N_obs] y_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;  \n  int&lt;lower=0&gt; N_mis;   \n  vector[N_mis] y_aux;  \n  array[N_mis] int&lt;lower=0, upper=1&gt; z_aux; \n}\nparameters {\n  vector[3] gamma;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  lambda ~ exponential(1);\n  x_obs ~ normal(gamma[1] + gamma[2] * y_obs + gamma[3] * z_obs,\n          lambda);\n}\ngenerated quantities {\n  array[N_mis] x_imp = normal_rng(gamma[1] + gamma[2] * y_aux[n]\n                      + gamma[3] * z_aux[n], lambda);\n}\nThe generated quantities block automatically samples \\(\\gamma_0\\), \\(\\gamma_1\\), \\(\\gamma_2\\), and \\(\\lambda\\) from their posterior distributions. So, the random draws of x_imp incorporate uncertainty from the estimated parameters and from the sampling variation in \\(x^{\\text{mis}}\\).\nMultiple posterior draws of \\(x^{\\text{mis}}\\) give us multiple imputed data sets. With these data sets we can model \\(p(y \\mid \\beta, \\sigma, x, z)\\) as \\[\ny \\sim \\text{normal}(\\beta_0 + \\beta_1 x^\\text{comp} + \\beta_2 z,\n        \\sigma),\n\\] where \\(x^\\text{comp}\\) contains observed and imputed values. The Stan code for this model is:\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  vector[N] z;\n}\nparameters {\n  vector[3] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  beta ~ normal(0, 1);\n  sigma ~ exponential(1);\n  y ~ normal(beta[1] + beta[2] * x + beta[3] * z, sigma);\n}\nWith multiple imputation, the model for \\(p(y \\mid \\beta, \\sigma, x, z)\\) does not need to distinguish between observed and imputed values of \\(x\\). Instead, this model can treat all imputed data sets as complete because we can combine all the posterior draws from multiple fits.\n\n\n\nA more general scenario involves an outcome variable \\(y\\) and several explanatory variables \\(x_1, \\ldots, x_K\\), each of which can have missing values that we want to impute.\nOne solution is to do multiple imputation through chained equations, a procedure often called “MICE”1. The MICE procedure in this scenario is:\n\nInitialize missing values in all \\(x_i\\). For each variable \\(x_i\\) with missing entries, fill its missing values with random samples from its observed values (or use another simple initialization rule).\nUpdate each \\(x_i\\) given \\(y\\) and the other \\(x\\)’s. For \\(i=1, \\ldots, K\\), fit a model for the observed \\(x_i\\) conditional on the current versions (with observed and imputed values) of \\(y\\) and \\(x_{-k} = x_1, \\ldots x_{k-1}, x_{k+1}, \\ldots,\nx_{K}\\). Use this model to draw impuations from the predictive distribution of the missing \\(x_i\\). Completing this step for all \\(i = 1, \\ldots, K\\) constitutes a single imputation cycle.\nWarmup period. Repeat the imputation cycle in step 1 several times as warmup to let the imputations stabilize.\nCreate M imputed datasets. After the warmup, record the current complete dataset as one imputed dataset. Then either restart from step 0 and repeat steps 1–3 until you have \\(M\\) imputed datasets; or do a single long run, i.e., continue iterating steps 1–2 and save the imputed dataset at \\(M\\) well-spaced iterations (e.g., every \\(S\\) cycles) to obtain \\(M\\) imputed datasets, without restarting from step 0 each time.\nFit the target model \\(p(y \\mid \\beta, \\sigma, x_1, \\ldots,\nx_K)\\) separately to each of the \\(M\\) imputed datasets and save the posterior draws for the parameters of interest.\nCombine the draws (or other summaries) from all \\(M\\) fits.\n\nNote that, as described here, the MICE procedure does not guarantee that the conditional distributions of all variables will be compatible. Compatibility means that there is a joint distribution for all the variables used in the imputation that can be decomposed as the conditional distributions we used. An incompatible imputation model will technically not sample from any well-defined probability distribution. But the consequences of this are not always serious. See section 6.4 in Carpenter et al. (2023) for a more detailed explanation of compatibility and of the related concept of congeniality.\n\n\nImagine that we again want to use numerical variables \\(x\\) and \\(y\\), and dichotomic variable \\(z\\) to estimate the parameters in \\(p(y \\mid \\beta, \\sigma, x, z)\\). But now both \\(x\\) and \\(z\\) have missing values that we want to impute.\nTo apply the MICE procedure in this example, we can reuse the models for \\(x\\) and \\(y\\) that we defined above. We also need a new model to impute \\(z^{\\text{mis}}\\), so we use the logistic regression \\[\nz^{\\text{obs}} \\sim \\text{Bernoulli}(\n  \\text{logit}^{-1}(\n    \\alpha_0 + \\alpha_1 y^{\\text{obs}} + \\alpha_2 x^{\\text{obs}})\n  ).\n\\] Here, \\(z^{\\text{obs}}\\) contains only the completely observed values from our original data set, while \\(x^{\\text{obs}}\\) and \\(y^{\\text{obs}}\\) contain all the values that correspond to \\(z^{\\text{obs}}\\). Thus, \\(x^{\\text{obs}}\\) can include observed and imputed values, and \\(y^{\\text{obs}}\\) need not be the same as in the model for \\(x^{\\text{obs}}\\).\nWith this model we can sample values for \\(z^{\\text{mis}}\\) conditional on its corresponding values \\(y^{\\text{aux}}\\) and \\(x^{\\text{aux}}\\).\nThe Stan code to impute \\(z^{\\text{mis}}\\) is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;\n  vector[N_obs] y_obs;\n  vector[N_obs] x_obs;\n  int&lt;lower=0&gt; N_mis;\n  vector[N_mis] y_aux;\n  vector[N_mis] x_aux;\n}\nparameters {\n  vector[3] gamma;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  z_obs ~ bernoulli_logit(gamma[1] + gamma[2] * y_obs \n                          + gamma[3] * x_obs);\n}\ngenerated quantities {\n  array[N_mis] int z_imp = bernoulli_logit_rng(gamma[1]\n                      + gamma[2] * y_aux\n                      + gamma[3] * x_aux);\n}\nThe pseudocode for this example of MICE, restarting the imputations after each imputation cycle, is shown below. Note that we do not need to initialize \\(x^{\\text{mis}}\\) in step 0 because we can impute it directly with the model for \\(x^{\\text{obs}}\\).\ncompleted_datasets &lt;- []\nfor (m in 1:M) {\n  data_m &lt;- copy(data_orig)\n  data_m.z[missing_idz] &lt;- random_sample(data_m.z[observed_idz])\n  for (cycle in 1:n_warm) {\n    stanmod_x &lt;- build_model(\"model_for_x.stan\")\n    fit_x &lt;- stanmod_x.fit(\n      obs_data=data_m[observed_idx],\n      mis_data=data_m[missing_idx]\n    )\n    data_m.x[missing_idx] &lt;- get_imputations(fit_x, \"x_imp\")\n    stanmod_z &lt;- build_model(\"model_for_z.stan\")\n    fit_z &lt;- stanmod_z.fit(\n      obs_data=data_m[observed_idz],\n      mis_data=data_m[missing_idz]\n    )\n    data_m.z[missing_idz] &lt;- get_imputations(fit_z, \"z_imp\")\n  }\n  completed_datasets.append(data_m)\n}\nall_draws &lt;- []\nfor (dataset in completed_datasets) {\n  stanmod_y &lt;- build_model(\"model_for_y.stan\")\n  fit_y &lt;- stanmod_y.fit(dataset)\n  all_draws.append(extract_draws(model_results))\n}\nall_results &lt;- combine_results(all_draws)\n\n\n\n\nWith Stan’s MCMC sampler, we can treat posterior draw chains from imputed data sets as if they were chains based on complete data. There is one important difference. Multiple imputation expresses uncertainty in the missing values as consistent differences in the estimates obtained from different imputated data sets. This means that chains obtained from the same imputed data set should converge, but chains obtained from different data sets do not have to. So, we need not worry if diagnostics2 signal that the chains from different imputed data sets are not converging properly. See an example in Bürkner (2025).\n\n\n\nA full Bayesian probability model includes a feedback flow of information between all parameters and all data. Cut models separate some parts of this feedback flow so that different subsets of data influence only some parameters in the model (see Plummer (2015)).\nMultiple imputation interrupts the flow of information from data to parameters. In our regression above, for example, the imputations influence the distribution of the parameters in \\(p(y \\mid x_1, \\ldots, x_K)\\), but these parameters do not influence the imputations. So, we could use multiple imputation to implement a cut model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html#motivation-and-outline",
+    "href": "stan-users-guide/multiple-imputation.html#motivation-and-outline",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "Suppose that we have a data set \\(x\\) with columns \\(x_{\\cdot, 1}, \\ldots, x_{\\cdot, K}\\) that make up covariates in a regression with quantities of interest \\(\\theta\\). With the completely observed data, \\(x^\\text{comp}\\), we could estimate the posterior distribution of \\(\\theta\\) as \\(p(\\theta \\mid x^\\text{comp})\\). But with missing data, our matrix is split into \\(x^{\\text{obs}}\\) (the observed values of \\(x\\)) and \\(x^{\\text{mis}}\\) (the missing values of \\(x\\)).\nFortunately, we can treat \\(x^{\\text{mis}}\\) as additional, nuisance parameters that we can estimate along with \\(\\theta\\) as \\(p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\\). So, we can express the marginal distribution of \\(\\theta\\) given only the observed values \\(x^{\\text{obs}}\\) as \\[\\begin{align*}\np(\\theta \\mid x^{\\text{obs}}) =&\n\\int\n    p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=&\n\\int\n    p(\\theta \\mid x^{\\text{obs}}, x^{\\text{mis}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=& \\int\n    p(\\theta \\mid x^{\\text{imp}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}},\n\\end{align*}\\] where \\(x^{\\text{imp}}\\) is a data set that includes imputed values of \\(x^{\\text{mis}}\\).\nThe equations above show that we do not need to describe \\(p(\\theta \\mid x^{\\text{obs}})\\) directly. Instead, we can first find a way to sample from \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), and then use these samples to fit the model \\(p(\\theta \\mid x^\\text{imp})\\), treating \\(x^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nNote that the model for \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\) needs to be explicit. In a typical regression setting with covariates \\(x\\) and outcome \\(y\\), this model for \\(x\\) is not needed because inferences for the regression parameters are independent of the model of \\(x\\) when \\(x\\) is fully observed. But with missing data and multiple imputation, we need to generate values for \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), which in turn needs an explicit model for \\(x^{\\text{mis}}\\).\nThus, the general outline for multiple imputation with a number \\(M\\) of imputations is:\n\nDraw \\(M\\) random samples \\(x^{\\text{mis}}_1, \\ldots,\nx^{\\text{mis}}_M\\) from the posterior predictive distribution \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\).\nUse these samples to get \\(M\\) different complete data sets \\(x_1^\\text{imp}, \\ldots, x_M^\\text{imp}\\).\nFor each of the \\(M\\) imputed data sets, sample from the model of interest \\(p(\\theta \\mid x_m^\\text{imp})\\), each time treating \\(x_m^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nCombine the draws for \\(\\theta\\) from all \\(M\\) fits.\n\n\n\nFollowing the notation above, we can express the outline with the pseudocode below:\ntheta_estimates = []\nfor (m in 1:M) {\n  x_mis &lt;- get_imputations(x_obs)\n  x_imp &lt;- add_imputations(x_obs, x_mis)\n  imputation_fits &lt;- my_model.fit(x_imp).draws()\n  theta_estimates &lt;- theta_estimates.append(imputation_fits)\n}\nEach of the functions in this pseudocode can encapsulate simple or complex procedures. get_imputations(), for example, may impute missing data by taking values directly from x_obs; or it may model a regression where an \\(x_{\\cdot, k}\\) with missing values is the outcome and the \\(x_{\\cdot, j}\\) without missing observations are predictors. Below we illustrate how to implement this second approach in Stan.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html#imputing-one-variable-in-stan",
+    "href": "stan-users-guide/multiple-imputation.html#imputing-one-variable-in-stan",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "Imagine our data set is composed of two numerical, continuous variables \\(x\\) and \\(y\\), and one binary variable \\(z\\). We want to estimate the parameters that govern the conditional association \\(p(y \\mid x, z)\\), but several values of \\(x\\) are missing. So, before we fit this model for \\(y\\), we impute the missing values in \\(x\\).\nWe first find all the fully observed data points and identify their values as \\(x^{\\text{obs}}\\), \\(y^{\\text{obs}}\\), and \\(z^{\\text{obs}}\\). Then we use these data to fit the model \\[\nx^{\\text{obs}} \\sim \\text{normal}(\\gamma_0 + \\gamma_1 y^{\\text{obs}} +\n    \\gamma_2 z^{\\text{obs}}, \\lambda).\n\\]\nThis model can then give us samples for the missing values \\(x^{\\text{mis}}\\) conditional on the corresponding values of \\(y\\) and \\(z\\), which we call \\(y^{\\text{aux}}\\) and \\(z^{\\text{aux}}\\). The Stan code for this imputation is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  vector[N_obs] x_obs;\n  vector[N_obs] y_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;  \n  int&lt;lower=0&gt; N_mis;   \n  vector[N_mis] y_aux;  \n  array[N_mis] int&lt;lower=0, upper=1&gt; z_aux; \n}\nparameters {\n  vector[3] gamma;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  lambda ~ exponential(1);\n  x_obs ~ normal(gamma[1] + gamma[2] * y_obs + gamma[3] * z_obs,\n          lambda);\n}\ngenerated quantities {\n  array[N_mis] x_imp = normal_rng(gamma[1] + gamma[2] * y_aux[n]\n                      + gamma[3] * z_aux[n], lambda);\n}\nThe generated quantities block automatically samples \\(\\gamma_0\\), \\(\\gamma_1\\), \\(\\gamma_2\\), and \\(\\lambda\\) from their posterior distributions. So, the random draws of x_imp incorporate uncertainty from the estimated parameters and from the sampling variation in \\(x^{\\text{mis}}\\).\nMultiple posterior draws of \\(x^{\\text{mis}}\\) give us multiple imputed data sets. With these data sets we can model \\(p(y \\mid \\beta, \\sigma, x, z)\\) as \\[\ny \\sim \\text{normal}(\\beta_0 + \\beta_1 x^\\text{comp} + \\beta_2 z,\n        \\sigma),\n\\] where \\(x^\\text{comp}\\) contains observed and imputed values. The Stan code for this model is:\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  vector[N] z;\n}\nparameters {\n  vector[3] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  beta ~ normal(0, 1);\n  sigma ~ exponential(1);\n  y ~ normal(beta[1] + beta[2] * x + beta[3] * z, sigma);\n}\nWith multiple imputation, the model for \\(p(y \\mid \\beta, \\sigma, x, z)\\) does not need to distinguish between observed and imputed values of \\(x\\). Instead, this model can treat all imputed data sets as complete because we can combine all the posterior draws from multiple fits.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html#imputing-two-or-more-variables",
+    "href": "stan-users-guide/multiple-imputation.html#imputing-two-or-more-variables",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "A more general scenario involves an outcome variable \\(y\\) and several explanatory variables \\(x_1, \\ldots, x_K\\), each of which can have missing values that we want to impute.\nOne solution is to do multiple imputation through chained equations, a procedure often called “MICE”1. The MICE procedure in this scenario is:\n\nInitialize missing values in all \\(x_i\\). For each variable \\(x_i\\) with missing entries, fill its missing values with random samples from its observed values (or use another simple initialization rule).\nUpdate each \\(x_i\\) given \\(y\\) and the other \\(x\\)’s. For \\(i=1, \\ldots, K\\), fit a model for the observed \\(x_i\\) conditional on the current versions (with observed and imputed values) of \\(y\\) and \\(x_{-k} = x_1, \\ldots x_{k-1}, x_{k+1}, \\ldots,\nx_{K}\\). Use this model to draw impuations from the predictive distribution of the missing \\(x_i\\). Completing this step for all \\(i = 1, \\ldots, K\\) constitutes a single imputation cycle.\nWarmup period. Repeat the imputation cycle in step 1 several times as warmup to let the imputations stabilize.\nCreate M imputed datasets. After the warmup, record the current complete dataset as one imputed dataset. Then either restart from step 0 and repeat steps 1–3 until you have \\(M\\) imputed datasets; or do a single long run, i.e., continue iterating steps 1–2 and save the imputed dataset at \\(M\\) well-spaced iterations (e.g., every \\(S\\) cycles) to obtain \\(M\\) imputed datasets, without restarting from step 0 each time.\nFit the target model \\(p(y \\mid \\beta, \\sigma, x_1, \\ldots,\nx_K)\\) separately to each of the \\(M\\) imputed datasets and save the posterior draws for the parameters of interest.\nCombine the draws (or other summaries) from all \\(M\\) fits.\n\nNote that, as described here, the MICE procedure does not guarantee that the conditional distributions of all variables will be compatible. Compatibility means that there is a joint distribution for all the variables used in the imputation that can be decomposed as the conditional distributions we used. An incompatible imputation model will technically not sample from any well-defined probability distribution. But the consequences of this are not always serious. See section 6.4 in Carpenter et al. (2023) for a more detailed explanation of compatibility and of the related concept of congeniality.\n\n\nImagine that we again want to use numerical variables \\(x\\) and \\(y\\), and dichotomic variable \\(z\\) to estimate the parameters in \\(p(y \\mid \\beta, \\sigma, x, z)\\). But now both \\(x\\) and \\(z\\) have missing values that we want to impute.\nTo apply the MICE procedure in this example, we can reuse the models for \\(x\\) and \\(y\\) that we defined above. We also need a new model to impute \\(z^{\\text{mis}}\\), so we use the logistic regression \\[\nz^{\\text{obs}} \\sim \\text{Bernoulli}(\n  \\text{logit}^{-1}(\n    \\alpha_0 + \\alpha_1 y^{\\text{obs}} + \\alpha_2 x^{\\text{obs}})\n  ).\n\\] Here, \\(z^{\\text{obs}}\\) contains only the completely observed values from our original data set, while \\(x^{\\text{obs}}\\) and \\(y^{\\text{obs}}\\) contain all the values that correspond to \\(z^{\\text{obs}}\\). Thus, \\(x^{\\text{obs}}\\) can include observed and imputed values, and \\(y^{\\text{obs}}\\) need not be the same as in the model for \\(x^{\\text{obs}}\\).\nWith this model we can sample values for \\(z^{\\text{mis}}\\) conditional on its corresponding values \\(y^{\\text{aux}}\\) and \\(x^{\\text{aux}}\\).\nThe Stan code to impute \\(z^{\\text{mis}}\\) is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;\n  vector[N_obs] y_obs;\n  vector[N_obs] x_obs;\n  int&lt;lower=0&gt; N_mis;\n  vector[N_mis] y_aux;\n  vector[N_mis] x_aux;\n}\nparameters {\n  vector[3] gamma;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  z_obs ~ bernoulli_logit(gamma[1] + gamma[2] * y_obs \n                          + gamma[3] * x_obs);\n}\ngenerated quantities {\n  array[N_mis] int z_imp = bernoulli_logit_rng(gamma[1]\n                      + gamma[2] * y_aux\n                      + gamma[3] * x_aux);\n}\nThe pseudocode for this example of MICE, restarting the imputations after each imputation cycle, is shown below. Note that we do not need to initialize \\(x^{\\text{mis}}\\) in step 0 because we can impute it directly with the model for \\(x^{\\text{obs}}\\).\ncompleted_datasets &lt;- []\nfor (m in 1:M) {\n  data_m &lt;- copy(data_orig)\n  data_m.z[missing_idz] &lt;- random_sample(data_m.z[observed_idz])\n  for (cycle in 1:n_warm) {\n    stanmod_x &lt;- build_model(\"model_for_x.stan\")\n    fit_x &lt;- stanmod_x.fit(\n      obs_data=data_m[observed_idx],\n      mis_data=data_m[missing_idx]\n    )\n    data_m.x[missing_idx] &lt;- get_imputations(fit_x, \"x_imp\")\n    stanmod_z &lt;- build_model(\"model_for_z.stan\")\n    fit_z &lt;- stanmod_z.fit(\n      obs_data=data_m[observed_idz],\n      mis_data=data_m[missing_idz]\n    )\n    data_m.z[missing_idz] &lt;- get_imputations(fit_z, \"z_imp\")\n  }\n  completed_datasets.append(data_m)\n}\nall_draws &lt;- []\nfor (dataset in completed_datasets) {\n  stanmod_y &lt;- build_model(\"model_for_y.stan\")\n  fit_y &lt;- stanmod_y.fit(dataset)\n  all_draws.append(extract_draws(model_results))\n}\nall_results &lt;- combine_results(all_draws)",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html#combining-posterior-draws",
+    "href": "stan-users-guide/multiple-imputation.html#combining-posterior-draws",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "With Stan’s MCMC sampler, we can treat posterior draw chains from imputed data sets as if they were chains based on complete data. There is one important difference. Multiple imputation expresses uncertainty in the missing values as consistent differences in the estimates obtained from different imputated data sets. This means that chains obtained from the same imputed data set should converge, but chains obtained from different data sets do not have to. So, we need not worry if diagnostics2 signal that the chains from different imputed data sets are not converging properly. See an example in Bürkner (2025).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html#cut-models",
+    "href": "stan-users-guide/multiple-imputation.html#cut-models",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "A full Bayesian probability model includes a feedback flow of information between all parameters and all data. Cut models separate some parts of this feedback flow so that different subsets of data influence only some parameters in the model (see Plummer (2015)).\nMultiple imputation interrupts the flow of information from data to parameters. In our regression above, for example, the imputations influence the distribution of the parameters in \\(p(y \\mid x_1, \\ldots, x_K)\\), but these parameters do not influence the imputations. So, we could use multiple imputation to implement a cut model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multiple-imputation.html#footnotes",
+    "href": "stan-users-guide/multiple-imputation.html#footnotes",
+    "title": "Multiple imputation",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nSection 5 of chapter 4 in van Buuren (2018) details the MICE procedure in a frequentist context.↩︎\nSuch as \\(\\hat{R}\\). See Split R-hat for detecting non-stationarity in the Reference Manual.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html",
+    "href": "stan-users-guide/missing-data.html",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "",
+    "text": "Bayesian inference supports a general approach to missing data in which any missing data item is represented as a parameter that is estimated in the posterior (Gelman et al. 2013). If the missing data are not explicitly modeled, as in the predictors for most regression models, then the result is an improper prior on the parameter representing the missing predictor.\nMixing arrays of observed and missing data can be difficult to include in Stan, partly because it can be tricky to model discrete unknowns in Stan and partly because unlike some other statistical languages (for example, R and Bugs), Stan requires observed and unknown quantities to be defined in separate places in the model. Thus it can be necessary to include code in a Stan program to splice together observed and missing parts of a data structure. Examples are provided later in the chapter.\n\n\nStan treats variables declared in the data and transformed data blocks as known and the variables in the parameters block as unknown.\nAn example involving missing normal observations could be coded as follows.1\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  array[N_mis] real y_mis;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_mis ~ normal(mu, sigma);\n}\nThe number of observed and missing data points are coded as data with non-negative integer variables N_obs and N_mis. The observed data are provided as an array data variable y_obs. The missing data are coded as an array parameter, y_mis. The ordinary parameters being estimated, the location mu and scale sigma, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.\n\n\n\nIn some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the transformed parameters block and assigning to it.\nThe following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] vector[2] y;\n  real&lt;lower=0&gt; var1;\n  real&lt;lower=0&gt; var2;\n}\ntransformed data {\n  real&lt;lower=0&gt; max_cov = sqrt(var1 * var2);\n  real&lt;upper=0&gt; min_cov = -max_cov;\n}\nparameters {\n  vector[2] mu;\n  real&lt;lower=min_cov, upper=max_cov&gt; cov;\n}\ntransformed parameters {\n  matrix[2, 2] Sigma;\n  Sigma[1, 1] = var1;\n  Sigma[1, 2] = cov;\n  Sigma[2, 1] = cov;\n  Sigma[2, 2] = var2;\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);\n}\nThe variances are defined as data in variables var1 and var2, whereas the covariance is defined as a parameter in variable cov. The \\(2 \\times 2\\) covariance matrix Sigma is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.\nThe constraint on the covariance declaration ensures that the resulting covariance matrix sigma is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.\nThe vectorization of the multivariate normal is critical for efficiency here. The transformed parameter Sigma could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.\n\n\n\nIf the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_obs;\n  array[N_mis] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_mis;\n  array[N_obs] real y_obs;\n}\ntransformed data {\n  int&lt;lower=0&gt; N = N_obs + N_mis;\n}\nparameters {\n  array[N_mis] real y_mis;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  array[N] real y;\n  y[ii_obs] = y_obs;\n  y[ii_mis] = y_mis;\n}\nmodel {\n  sigma ~ gamma(1, 1);\n  y[1] ~ normal(0, 100);\n  y[2:N] ~ normal(y[1:(N - 1)], sigma);\n}\nThe index arrays ii_obs and ii_mis contain the indexes into the final array y of the observed data (coded as a data vector y_obs) and the missing data (coded as a parameter vector y_mis). See the time series chapter for further discussion of time-series model and specifically the autoregression section for an explanation of the vectorization for y as well as an explanation of how to convert this example to a full AR(1) model. To ensure y[1] has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.\nAnother potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in\nx[N_obs_2, 2] = x_obs_2;\nx[N_mis_2, 2] = x_mis_2;\nwhere the relevant variables are all hard coded with index 2 because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the ragged data structures section for a general discussion of coding ragged data structures in Stan).\n\n\n\nRick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis (Aguilar and West 2000). This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.\ndata {\n  int&lt;lower=2&gt; K;\n}\ntransformed data {\n  int&lt;lower=1&gt; K_choose_2;\n  K_choose_2 = (K * (K - 1)) / 2;\n}\nparameters {\n  vector[K_choose_2] L_lower;\n}\ntransformed parameters {\n  cholesky_factor_cov[K] L;\n  for (k in 1:K) {\n    L[k, k] = 1;\n  }\n  {\n    int i;\n    for (m in 2:K) {\n      for (n in 1:(m - 1)) {\n        L[m, n] = L_lower[i];\n        L[n, m] = 0;\n        i += 1;\n      }\n    }\n  }\n}\nIt is most convenient to place a prior directly on L_lower. An alternative would be a prior for the full Cholesky factor L, because the transform from L_lower to L is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix L   * L', because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.\n\n\n\nIt’s often the case that one or more components of a multivariate outcome are missing.2\nAs an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations y and a boolean array y_observed to indicate which values were observed (others can have dummy values in the input).\narray[N] vector[2] y;\narray[N, 2] int&lt;lower=0, upper=1&gt; y_observed;\nIf both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    y[n] ~ multi_normal(mu, Sigma);\n  } else if (y_observed[n, 1]) {\n    y[n, 1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\n  } else if (y_observed[n, 2]) {\n    y[n, 2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\n  }\n}\nIt’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:\ntransformed data {\n  array[observed_12(y_observed)] int ns12;\n  array[observed_1(y_observed)] int ns1;\n  array[observed_2(y_observed)] int ns2;\n}\nYou will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.\nint n12 = 1;\nint n1 = 1;\nint n2 = 1;\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    ns12[n12] = n;\n    n12 += 1;\n  } else if (y_observed[n, 1]) {\n    ns1[n1] = n;\n    n1 += 1;\n  } else if (y_observed[n, 2]) {\n    ns2[n2] = n;\n    n2 += 1;\n  }\n}\nThen, in the model block, everything is vectorizable using those indexes constructed once in transformed data:\ny[ns12] ~ multi_normal(mu, Sigma);\ny[ns1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\ny[ns2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\nThe result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.\nThis approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html#missing-data",
+    "href": "stan-users-guide/missing-data.html#missing-data",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "",
+    "text": "Stan treats variables declared in the data and transformed data blocks as known and the variables in the parameters block as unknown.\nAn example involving missing normal observations could be coded as follows.1\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  array[N_mis] real y_mis;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_mis ~ normal(mu, sigma);\n}\nThe number of observed and missing data points are coded as data with non-negative integer variables N_obs and N_mis. The observed data are provided as an array data variable y_obs. The missing data are coded as an array parameter, y_mis. The ordinary parameters being estimated, the location mu and scale sigma, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html#partially-known-parameters.section",
+    "href": "stan-users-guide/missing-data.html#partially-known-parameters.section",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "",
+    "text": "In some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the transformed parameters block and assigning to it.\nThe following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] vector[2] y;\n  real&lt;lower=0&gt; var1;\n  real&lt;lower=0&gt; var2;\n}\ntransformed data {\n  real&lt;lower=0&gt; max_cov = sqrt(var1 * var2);\n  real&lt;upper=0&gt; min_cov = -max_cov;\n}\nparameters {\n  vector[2] mu;\n  real&lt;lower=min_cov, upper=max_cov&gt; cov;\n}\ntransformed parameters {\n  matrix[2, 2] Sigma;\n  Sigma[1, 1] = var1;\n  Sigma[1, 2] = cov;\n  Sigma[2, 1] = cov;\n  Sigma[2, 2] = var2;\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);\n}\nThe variances are defined as data in variables var1 and var2, whereas the covariance is defined as a parameter in variable cov. The \\(2 \\times 2\\) covariance matrix Sigma is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.\nThe constraint on the covariance declaration ensures that the resulting covariance matrix sigma is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.\nThe vectorization of the multivariate normal is critical for efficiency here. The transformed parameter Sigma could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html#sliced-missing-data",
+    "href": "stan-users-guide/missing-data.html#sliced-missing-data",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "",
+    "text": "If the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_obs;\n  array[N_mis] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_mis;\n  array[N_obs] real y_obs;\n}\ntransformed data {\n  int&lt;lower=0&gt; N = N_obs + N_mis;\n}\nparameters {\n  array[N_mis] real y_mis;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  array[N] real y;\n  y[ii_obs] = y_obs;\n  y[ii_mis] = y_mis;\n}\nmodel {\n  sigma ~ gamma(1, 1);\n  y[1] ~ normal(0, 100);\n  y[2:N] ~ normal(y[1:(N - 1)], sigma);\n}\nThe index arrays ii_obs and ii_mis contain the indexes into the final array y of the observed data (coded as a data vector y_obs) and the missing data (coded as a parameter vector y_mis). See the time series chapter for further discussion of time-series model and specifically the autoregression section for an explanation of the vectorization for y as well as an explanation of how to convert this example to a full AR(1) model. To ensure y[1] has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.\nAnother potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in\nx[N_obs_2, 2] = x_obs_2;\nx[N_mis_2, 2] = x_mis_2;\nwhere the relevant variables are all hard coded with index 2 because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the ragged data structures section for a general discussion of coding ragged data structures in Stan).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html#loading-matrix-for-factor-analysis",
+    "href": "stan-users-guide/missing-data.html#loading-matrix-for-factor-analysis",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "",
+    "text": "Rick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis (Aguilar and West 2000). This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.\ndata {\n  int&lt;lower=2&gt; K;\n}\ntransformed data {\n  int&lt;lower=1&gt; K_choose_2;\n  K_choose_2 = (K * (K - 1)) / 2;\n}\nparameters {\n  vector[K_choose_2] L_lower;\n}\ntransformed parameters {\n  cholesky_factor_cov[K] L;\n  for (k in 1:K) {\n    L[k, k] = 1;\n  }\n  {\n    int i;\n    for (m in 2:K) {\n      for (n in 1:(m - 1)) {\n        L[m, n] = L_lower[i];\n        L[n, m] = 0;\n        i += 1;\n      }\n    }\n  }\n}\nIt is most convenient to place a prior directly on L_lower. An alternative would be a prior for the full Cholesky factor L, because the transform from L_lower to L is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix L   * L', because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html#missing-multivariate-data",
+    "href": "stan-users-guide/missing-data.html#missing-multivariate-data",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "",
+    "text": "It’s often the case that one or more components of a multivariate outcome are missing.2\nAs an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations y and a boolean array y_observed to indicate which values were observed (others can have dummy values in the input).\narray[N] vector[2] y;\narray[N, 2] int&lt;lower=0, upper=1&gt; y_observed;\nIf both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    y[n] ~ multi_normal(mu, Sigma);\n  } else if (y_observed[n, 1]) {\n    y[n, 1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\n  } else if (y_observed[n, 2]) {\n    y[n, 2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\n  }\n}\nIt’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:\ntransformed data {\n  array[observed_12(y_observed)] int ns12;\n  array[observed_1(y_observed)] int ns1;\n  array[observed_2(y_observed)] int ns2;\n}\nYou will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.\nint n12 = 1;\nint n1 = 1;\nint n2 = 1;\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    ns12[n12] = n;\n    n12 += 1;\n  } else if (y_observed[n, 1]) {\n    ns1[n1] = n;\n    n1 += 1;\n  } else if (y_observed[n, 2]) {\n    ns2[n2] = n;\n    n2 += 1;\n  }\n}\nThen, in the model block, everything is vectorizable using those indexes constructed once in transformed data:\ny[ns12] ~ multi_normal(mu, Sigma);\ny[ns1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\ny[ns2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\nThe result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.\nThis approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/missing-data.html#footnotes",
+    "href": "stan-users-guide/missing-data.html#footnotes",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nA more meaningful estimation example would involve a regression of the observed and missing observations using predictors that were known for each and specified in the data block.↩︎\nThis is not the same as missing components of a multivariate predictor in a regression problem; in that case, you will need to represent the missing data as a parameter and impute missing values in order to feed them into the regression.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html",
+    "href": "stan-users-guide/matrices-arrays.html",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "This chapter provides pointers as to how to choose among the various container types (matrix, vector, array, and tuple) provided by Stan.\n\n\nStan provides three basic scalar types, int, real, and complex, as well as three basic linear algebra types, vector, row_vector, and matrix. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).\nThis leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.\narray[N] real a;\nvector[N] a;\nrow_vector[N] a;\nThese distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.\nThe following code fragment shows all four ways to declare a two-dimensional container of size \\(M \\times N\\).\narray[M, N] real b;          // b[m] : array[] real     (efficient)\narray[M] vector[N] b;        // b[m] : vector     (efficient)\narray[M] row_vector[N] b;    // b[m] : row_vector (efficient)\nmatrix[M, N] b;              // b[m] : row_vector (inefficient)\nThe main differences among these choices involve efficiency for various purposes and the type of b[m], which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.\nThe inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).\n\n\n\nArrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single complex typed entry, all the entries are complex).\nWith arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a complex_vector[3] holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.\nTuples provide a way to represent a sequence of values of heterogeneous types. For example, tuple(int, real) is the type of a pair consisting of an integer and a real number and tuple(array[5] int, vector[6]) is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.\n\n\nTuples are declared using the keyword tuple followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.\ntuple(int, vector[3]) ny = (5, [3, 2.9, 1.8]');\nThe elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using\nint n = ny.1;\nvector[3] y = ny.2;\nWe can also assign into the elements of a tuple.\ntuple(int, vector[3], complex) abc;\nabc.1 = 5;\nabc.2[1] = 3;\nabc.2[2] = 2.9;\nabc.2[3] = 1.4798;\nabc.3 = 2 + 1.9j;\nAs the cascaded indexing example shows, the result of abc.1 is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., abc.2[1] pulls out the first element of the vector value of the second element of the tuple.)\nThere are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.\nint a = 5;\nmatrix[10, 100] b = ...;\ntuple(int, matrix[10, 100]) ab = (a, b);  // COPIES b\nb[1,1] = 10.3;  // does NOT change ab\n\n\n\nTuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, array[N] tuple(int, real, vector[5]) is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as tuple(array[N] int, array[N] real, array[N] vector[5]).\nThe second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix \\(A = Q \\cdot R\\), where \\(Q\\) is orthonormal and \\(R\\) is upper triangular. In the past, this required two function calls.\nmatrix[M, N] A = ...;\nmatrix[M, M] Q = qr_Q(A);\nmatrix[M, N] R = qr_R(A);\nWith tuples, this can be simplified to the following,\ntuple(matrix[M, M], matrix[M, N]) QR = qr(A);\nwith QR.1 being Q and QR.2 giving R.\n\n\n\n\nStan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.\nWhen an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.\n\n\n\nThe underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.\nArrays, on the other hand, are implemented as instances of the C++\nstd::vector class (not to be confused with Eigen’s Eigen::Vector class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.\n\n\nIn Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.\nFirst, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.\nSecond, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).\nThird, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.\nmatrix[M, N] a;\n//...\nfor (n in 1:N) {\n  for (m in 1:M) {\n    // ... do something with a[m, n] ...\n  }\n}\nArrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  for (n in 1:N) {\n    // ... do something with a[m, n] ...\n  }\n}\nThe first use of a[m ,n] should bring a[m] into memory. Overall, traversing matrices is more efficient than traversing arrays.\nThis is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is\narray[I, J] matrix[M, N] b;\n// ...\nfor (i in 1:I) {\n  for (j in 1:J) {\n    for (n in 1:N) {\n      for (m in 1:M) {\n        // ... do something with b[i, j, m, n] ...\n      }\n    }\n  }\n}\nIf a is a matrix, the notation a[m] picks out row m of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this\narray[M] row_vector[N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nis much more efficient than the pure matrix version\nmatrix[M, N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nSimilarly, indexing an array of column vectors is more efficient than using the col function to pick out a column of a matrix.\nIn contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this\nmatrix[N, k] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\ny_hat = x * beta;\nthan it is to do this\narray[N] row_vector[K] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\nfor (n in 1:N) {\n  y_hat[n] = x[n] * beta;\n}\n\n\n\nFor use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The Eigen::Vector template specialization and the std::vector template class are implemented similarly as containers of double values (the type real in Stan). Only arrays in Stan are allowed to store integer values.\n\n\n\n\nThe key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.\n\n\nCPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is much slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.\n\n\n\nMatrices are stored internally in column-major order. That is, an \\(M\n\\times N\\) matrix stores its elements in the order \\[\n(1,1), (2, 1), \\dotsc, (M, 1), (1, 2), \\dotsc, (M, 2), \\dotsc, (1, N),\n\\dotsc, (M, N).\n\\]\nThis means that it’s much more efficient to write loops over matrices column by column, as in the following example.\nmatrix[M, N] a;\n// ...\nfor (n in 1:N) {\n  for (m in 1:M) {\n     // ... do something with a[m, n] ...\n  }\n}\nIt also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.\narray[M] row_vector[N] a;\n// ...\nfor (m in 1:M) {\n  // ... do something with row vector a[m] ...\n}\nEven if what is done involves a function call, the row vector a[m] will not have to be copied.\n\n\n\nArrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  // ... do something with a[m] ...\n}\nA difference with matrices is that the entries a[m] in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”\n\n\n\n\nThere is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the multi-indexing chapter for some reshaping operations involving multiple indexing and range indexing.\n\n\n\nStan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to u and the compound slicing assignment to x, where u and x start with the same values.\ntransformed data {\n  vector[4] x = [ 1, 2, 3, 4 ]';\n  vector[4] u = [ 1, 2, 3, 4 ]';\n\n  for (t in 2:4) {\n    u[t] = u[t - 1] * 3;\n  }\n\n  x[2:4] = x[1:3] * 3;\n\n  print(\"u = \", u);\n  print(\"x = \", x);\n}\nThe output it produces is,\nu = [1, 3, 9, 27]\nx = [1, 3, 6, 9]\nIn the loop version assigning to u, the values are updated before being used to define subsequent values; in the sliced expression assigning to x, the entire right-hand side is evaluated before assigning to the left-hand side.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#basic-motivation",
+    "href": "stan-users-guide/matrices-arrays.html#basic-motivation",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "Stan provides three basic scalar types, int, real, and complex, as well as three basic linear algebra types, vector, row_vector, and matrix. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).\nThis leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.\narray[N] real a;\nvector[N] a;\nrow_vector[N] a;\nThese distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.\nThe following code fragment shows all four ways to declare a two-dimensional container of size \\(M \\times N\\).\narray[M, N] real b;          // b[m] : array[] real     (efficient)\narray[M] vector[N] b;        // b[m] : vector     (efficient)\narray[M] row_vector[N] b;    // b[m] : row_vector (efficient)\nmatrix[M, N] b;              // b[m] : row_vector (inefficient)\nThe main differences among these choices involve efficiency for various purposes and the type of b[m], which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.\nThe inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#tuple-types",
+    "href": "stan-users-guide/matrices-arrays.html#tuple-types",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "Arrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single complex typed entry, all the entries are complex).\nWith arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a complex_vector[3] holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.\nTuples provide a way to represent a sequence of values of heterogeneous types. For example, tuple(int, real) is the type of a pair consisting of an integer and a real number and tuple(array[5] int, vector[6]) is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.\n\n\nTuples are declared using the keyword tuple followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.\ntuple(int, vector[3]) ny = (5, [3, 2.9, 1.8]');\nThe elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using\nint n = ny.1;\nvector[3] y = ny.2;\nWe can also assign into the elements of a tuple.\ntuple(int, vector[3], complex) abc;\nabc.1 = 5;\nabc.2[1] = 3;\nabc.2[2] = 2.9;\nabc.2[3] = 1.4798;\nabc.3 = 2 + 1.9j;\nAs the cascaded indexing example shows, the result of abc.1 is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., abc.2[1] pulls out the first element of the vector value of the second element of the tuple.)\nThere are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.\nint a = 5;\nmatrix[10, 100] b = ...;\ntuple(int, matrix[10, 100]) ab = (a, b);  // COPIES b\nb[1,1] = 10.3;  // does NOT change ab\n\n\n\nTuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, array[N] tuple(int, real, vector[5]) is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as tuple(array[N] int, array[N] real, array[N] vector[5]).\nThe second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix \\(A = Q \\cdot R\\), where \\(Q\\) is orthonormal and \\(R\\) is upper triangular. In the past, this required two function calls.\nmatrix[M, N] A = ...;\nmatrix[M, M] Q = qr_Q(A);\nmatrix[M, N] R = qr_R(A);\nWith tuples, this can be simplified to the following,\ntuple(matrix[M, M], matrix[M, N]) QR = qr(A);\nwith QR.1 being Q and QR.2 giving R.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#fixed-sizes-and-indexing-out-of-bounds",
+    "href": "stan-users-guide/matrices-arrays.html#fixed-sizes-and-indexing-out-of-bounds",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "Stan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.\nWhen an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#indexing-efficiency.section",
+    "href": "stan-users-guide/matrices-arrays.html#indexing-efficiency.section",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "The underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.\nArrays, on the other hand, are implemented as instances of the C++\nstd::vector class (not to be confused with Eigen’s Eigen::Vector class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.\n\n\nIn Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.\nFirst, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.\nSecond, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).\nThird, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.\nmatrix[M, N] a;\n//...\nfor (n in 1:N) {\n  for (m in 1:M) {\n    // ... do something with a[m, n] ...\n  }\n}\nArrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  for (n in 1:N) {\n    // ... do something with a[m, n] ...\n  }\n}\nThe first use of a[m ,n] should bring a[m] into memory. Overall, traversing matrices is more efficient than traversing arrays.\nThis is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is\narray[I, J] matrix[M, N] b;\n// ...\nfor (i in 1:I) {\n  for (j in 1:J) {\n    for (n in 1:N) {\n      for (m in 1:M) {\n        // ... do something with b[i, j, m, n] ...\n      }\n    }\n  }\n}\nIf a is a matrix, the notation a[m] picks out row m of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this\narray[M] row_vector[N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nis much more efficient than the pure matrix version\nmatrix[M, N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nSimilarly, indexing an array of column vectors is more efficient than using the col function to pick out a column of a matrix.\nIn contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this\nmatrix[N, k] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\ny_hat = x * beta;\nthan it is to do this\narray[N] row_vector[K] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\nfor (n in 1:N) {\n  y_hat[n] = x[n] * beta;\n}\n\n\n\nFor use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The Eigen::Vector template specialization and the std::vector template class are implemented similarly as containers of double values (the type real in Stan). Only arrays in Stan are allowed to store integer values.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#memory-locality",
+    "href": "stan-users-guide/matrices-arrays.html#memory-locality",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "The key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.\n\n\nCPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is much slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.\n\n\n\nMatrices are stored internally in column-major order. That is, an \\(M\n\\times N\\) matrix stores its elements in the order \\[\n(1,1), (2, 1), \\dotsc, (M, 1), (1, 2), \\dotsc, (M, 2), \\dotsc, (1, N),\n\\dotsc, (M, N).\n\\]\nThis means that it’s much more efficient to write loops over matrices column by column, as in the following example.\nmatrix[M, N] a;\n// ...\nfor (n in 1:N) {\n  for (m in 1:M) {\n     // ... do something with a[m, n] ...\n  }\n}\nIt also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.\narray[M] row_vector[N] a;\n// ...\nfor (m in 1:M) {\n  // ... do something with row vector a[m] ...\n}\nEven if what is done involves a function call, the row vector a[m] will not have to be copied.\n\n\n\nArrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  // ... do something with a[m] ...\n}\nA difference with matrices is that the entries a[m] in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#converting-among-matrix-vector-and-array-types",
+    "href": "stan-users-guide/matrices-arrays.html#converting-among-matrix-vector-and-array-types",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "There is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the multi-indexing chapter for some reshaping operations involving multiple indexing and range indexing.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html#aliasing-in-stan-containers",
+    "href": "stan-users-guide/matrices-arrays.html#aliasing-in-stan-containers",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "Stan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to u and the compound slicing assignment to x, where u and x start with the same values.\ntransformed data {\n  vector[4] x = [ 1, 2, 3, 4 ]';\n  vector[4] u = [ 1, 2, 3, 4 ]';\n\n  for (t in 2:4) {\n    u[t] = u[t - 1] * 3;\n  }\n\n  x[2:4] = x[1:3] * 3;\n\n  print(\"u = \", u);\n  print(\"x = \", x);\n}\nThe output it produces is,\nu = [1, 3, 9, 27]\nx = [1, 3, 6, 9]\nIn the loop version assigning to u, the values are updated before being used to define subsequent values; in the sliced expression assigning to x, the entire right-hand side is evaluated before assigning to the left-hand side.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/index.html",
+    "href": "stan-users-guide/index.html",
+    "title": "Stan User’s Guide",
+    "section": "",
+    "text": "This is the official user’s guide for Stan. It provides example models and programming techniques for coding statistical models in Stan.\n\nPart 1 gives Stan code and discussions for several important classes of models.\nPart 2 discusses various general Stan programming techniques that are not tied to any particular model.\nPart 3 introduces algorithms for calibration and model checking that require multiple runs of Stan.\nThe appendices provide an introduction to the stanc3 compiler used in the various interfaces to Stan, a style guide, and advice for users of BUGS and JAGS.\n\nWe recommend working through this guide using the textbooks Bayesian Data Analysis and Statistical Rethinking: A Bayesian Course with Examples in R and Stan as references on the concepts, and using the Stan Reference Manual when necessary to clarify programming issues.\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Stan Users Guide",
+      "Stan User's Guide"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/gaussian-processes.html",
+    "href": "stan-users-guide/gaussian-processes.html",
+    "title": "Gaussian Processes",
+    "section": "",
+    "text": "Gaussian processes are continuous stochastic processes and thus may be interpreted as providing a probability distribution over functions. A probability distribution over continuous functions may be viewed, roughly, as an uncountably infinite collection of random variables, one for each valid input. The generality of the supported functions makes Gaussian priors popular choices for priors in general multivariate (non-linear) regression problems.\nThe defining feature of a Gaussian process is that the joint distribution of the function’s value at a finite number of input points is a multivariate normal distribution. This makes it tractable to both fit models from finite amounts of observed data and make predictions for finitely many new data points.\nUnlike a simple multivariate normal distribution, which is parameterized by a mean vector and covariance matrix, a Gaussian process is parameterized by a mean function and covariance function. The mean and covariance functions apply to vectors of inputs and return a mean vector and covariance matrix which provide the mean and covariance of the outputs corresponding to those input points in the functions drawn from the process.\nGaussian processes can be encoded in Stan by implementing their mean and covariance functions or by using the specialized covariance functions outlined below, and plugging the result into the Gaussian model. This form of model is straightforward and may be used for simulation, model fitting, or posterior predictive inference. A more efficient Stan implementation for the GP with a normally distributed outcome marginalizes over the latent Gaussian process, and applies a Cholesky-factor reparameterization of the Gaussian to compute the likelihood and the posterior predictive distribution analytically.\nAfter defining Gaussian processes, this chapter covers the basic implementations for simulation, hyperparameter estimation, and posterior predictive inference for univariate regressions, multivariate regressions, and multivariate logistic regressions. Gaussian processes are general, and by necessity this chapter only touches on some basic models. For more information, see Rasmussen and Williams (2006).\nNote that fitting Gaussian processes as described below using exact inference by computing Cholesky of the covariance matrix scales cubicly with the size of data. Due to how Stan autodiff is implemented, Stan is also slower than Gaussian process specialized software. It is likely that Gaussian processes using exact inference by computing Cholesky of the covariance matrix with \\(N&gt;1000\\) are too slow for practical purposes in Stan. There are many approximations to speed-up Gaussian process computation, from which the basis function approaches for 1-3 dimensional \\(x\\) are easiest to implement in Stan (see, e.g., Riutort-Mayol et al. (2023)).\n\n\nThe data for a multivariate Gaussian process regression consists of a series of \\(N\\) inputs \\(x_1,\\dotsc,x_N \\in \\mathbb{R}^D\\) paired with outputs \\(y_1,\\dotsc,y_N \\in \\mathbb{R}\\). The defining feature of Gaussian processes is that the probability of a finite number of outputs \\(y\\) conditioned on their inputs \\(x\\) is Gaussian: \\[\ny \\sim \\textsf{multivariate normal}(m(x), K(x \\mid \\theta)),\n\\] where \\(m(x)\\) is an \\(N\\)-vector and \\(K(x \\mid \\theta)\\) is an \\(N \\times N\\) covariance matrix. The mean function \\(m : \\mathbb{R}^{N \\times D}\n\\rightarrow \\mathbb{R}^{N}\\) can be anything, but the covariance function \\(K : \\mathbb{R}^{N \\times D} \\rightarrow \\mathbb{R}^{N \\times N}\\) must produce a positive-definite matrix for any input \\(x\\).1\nA popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, \\[\n  K(x \\mid \\alpha, \\rho, \\sigma)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j} \\sigma^2,\n\\] where \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\) are hyperparameters defining the covariance function and where \\(\\delta_{i, j}\\) is the Kronecker delta function with value 1 if \\(i = j\\) and value 0 otherwise; this test is between the indexes \\(i\\) and \\(j\\), not between values \\(x_i\\) and \\(x_j\\). This kernel is obtained through a convolution of two independent Gaussian processes, \\(f_1\\) and \\(f_2\\), with kernels \\[\n  K_1(x \\mid \\alpha, \\rho)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n\\] and \\[\n  K_2(x \\mid \\sigma)_{i, j}\n=\n\\delta_{i, j} \\sigma^2.\n\\]\nThe addition of \\(\\sigma^2\\) on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs \\(x_i = x_j\\). In statistical terms, \\(\\sigma\\) is the scale of the noise term in the regression.\nThe hyperparameter \\(\\rho\\) is the length-scale, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of \\(\\rho\\) closer to zero lead the GP to represent high-frequency functions, whereas larger values of \\(\\rho\\) lead to low-frequency functions. The hyperparameter \\(\\alpha\\) is the marginal standard deviation. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP \\(f_1\\) prior at a single input \\(x\\) conditional on one value of \\(\\alpha\\) one would recover \\(\\alpha\\).\nThe only term in the squared exponential covariance function involving the inputs \\(x_i\\) and \\(x_j\\) is their vector difference, \\(x_i - x_j\\). This produces a process with stationary covariance in the sense that if an input vector \\(x\\) is translated by a vector \\(\\epsilon\\) to \\(x +\n\\epsilon\\), the covariance at any pair of outputs is unchanged, because \\(K(x \\mid \\theta) = K(x + \\epsilon \\mid \\theta)\\).\nThe summation involved is just the squared Euclidean distance between \\(x_i\\) and \\(x_j\\) (i.e., the squared \\(L_2\\) norm of their difference, \\(x_i -\nx_j\\)). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\).\nChanging the notion of distance from Euclidean to taxicab distance (i.e., an \\(L_1\\) norm) changes the support to functions which are continuous but not smooth.\n\n\n\nIt is simplest to start with a Stan model that does nothing more than simulate draws of functions \\(f\\) from a Gaussian process. In practical terms, the model will draw values \\(y_n = f(x_n)\\) for finitely many input points \\(x_n\\).\nThe Stan model defines the mean and covariance functions in a transformed data block and then samples outputs \\(y\\) in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to \\(\\alpha^2 = 1\\), \\(\\rho^2 = 1\\), and \\(\\sigma^2 = 0.1\\), and the mean function \\(m\\) is defined to always return the zero vector, \\(m(x) = \\textbf{0}\\). Consider the following implementation of a Gaussian process simulator.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K;\n  vector[N] mu = rep_vector(0, N);\n  for (i in 1:(N - 1)) {\n    K[i, i] = 1 + 0.1;\n    for (j in (i + 1):N) {\n      K[i, j] = exp(-0.5 * square(x[i] - x[j]));\n      K[j, i] = K[i, j];\n    }\n  }\n  K[N, N] = 1 + 0.1;\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K = gp_exp_quad_cov(x, 1.0, 1.0);\n  vector[N] mu = rep_vector(0, N);\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + 0.1;\n  }\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe input data are just the vector of inputs x and its size N. Such a model can be used with values of x evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.\n\n\nOnly the input data needs to change in moving from a univariate model to a multivariate model.\nThe only lines that change from the univariate model above are as follows.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n}\ntransformed data {\n  // ...\n}\nThe data are now declared as an array of vectors instead of an array of scalars; the dimensionality D is also declared.\nIn the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.\n\n\n\nA more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose \\(\\eta\\) is an an isotropic standard normal variate \\[\n\\eta \\sim \\textsf{normal}(\\textbf{0}, \\textrm{I}),\n\\] where \\(\\textbf{0}\\) is an \\(N\\)-vector of 0 values and \\(\\textrm{I}\\) is the \\(N\n\\times N\\) identity matrix. Let \\(L\\) be the Cholesky decomposition of \\(K(x \\mid \\theta)\\), i.e., the lower-triangular matrix \\(L\\) such that \\(LL^{\\top} =\nK(x \\mid \\theta)\\). Then the transformed variable \\(\\mu + L\\eta\\) has the intended target distribution, \\[\n  \\mu + L\\eta \\sim \\textsf{multivariate normal}(\\mu(x), K(x \\mid \\theta)).\n\\]\nThis transform can be applied directly to Gaussian process simulation.\nThis model has the same data declarations for N and x, and the same transformed data definitions of mu and K as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.\n// ...\ntransformed data {\n  matrix[N, N] L;\n  // ...\n  L = cholesky_decompose(K);\n}\nparameters {\n  vector[N] eta;\n}\nmodel {\n  eta ~ std_normal();\n}\ngenerated quantities {\n  vector[N] y;\n  y = mu + L * eta;\n}\nThe Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix K computed. The isotropic normal distribution for eta is specified as a vectorized univariate distribution for efficiency; this specifies that each eta[n] has an independent standard normal distribution. The sampled vector y is then defined as a generated quantity using a direct encoding of the transform described above.\n\n\n\n\n\n\nThe full generative model for a GP with a normal outcome, \\(y \\in \\mathbb{R}^N\\), with inputs \\(x \\in \\mathbb{R}^N\\), for a finite \\(N\\): \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\nf      &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i    &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\}\n\\end{align*}\\] With a normal outcome, it is possible to integrate out the Gaussian process \\(f\\), yielding the more parsimonious model: \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\ny      &\\sim \\textsf{multivariate normal}\n         \\left(0, K(x \\mid \\alpha, \\rho) + \\textbf{I}_N \\sigma^2\\right) \\\\\n\\end{align*}\\]\nIt can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.\nThe hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (\\(\\alpha\\)), the scale of the output noise (\\(\\sigma\\)), and the scale at which distances are measured among inputs (\\(\\rho\\)). See the Gaussian process priors section for more information about how to specify appropriate priors for the hyperparameters.\nThe Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix K in the model block, rather than the transformed data block.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  vector[N] mu = rep_vector(0, N);\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N, N] L_K;\n  matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n  real sq_sigma = square(sigma);\n\n  // diagonal elements\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + sq_sigma;\n  }\n\n  L_K = cholesky_decompose(K);\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y ~ multi_normal_cholesky(mu, L_K);\n}\nThe data block declares a vector y of observed values y[n] for inputs x[n]. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix K is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value y is known and the covariance matrix K is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.\nWe have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the cholesky_decompose function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (\\(N \\gtrsim 100\\)) the Cholesky decomposition version will be faster.\nHamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model (Neal 1997). If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.\n\n\nWe can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, \\(\\delta\\) to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\nTwo differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length \\(N\\) called eta. This is used in the model block to generate a multivariate normal vector called \\(f\\), corresponding to the latent GP. We put a \\(\\textsf{normal}(0,1)\\) prior on eta like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for \\(y\\) with one \\(N\\)-dimensional multivariate normal with an identity covariance matrix multiplied by \\(\\sigma^2\\), we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.\n\n\n\n\nGaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.\n\n\nIf we want to model count data, we can remove the \\(\\sigma\\) parameter, and use poisson_log, which implements a Poisson distribution with log link function, rather than normal. We can also add an overall mean parameter, \\(a\\), which will account for the marginal expected value for \\(y\\). We do this because we cannot center count data like we would for normally distributed data.\ndata {\n  // ...\n  array[N] int&lt;lower=0&gt; y;\n  // ...\n}\n// ...\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ poisson_log(a + f);\n}\n\n\n\nFor computational reasons, we may want to integrate out the latent variable \\(f\\), as was done in the normal output model. Unfortunately, exact marginalization over \\(f\\) is not possible when the outcome model is not normal. Instead, we may perform approximate marginalization with an embedded Laplace approximation (Rasmussen and Williams 2006; Rue, Martino, and Chopin 2009; Margossian et al. 2020). To do so, we first use the function laplace_marginal to approximate the marginal likelihood \\(p(y \\mid \\rho, \\alpha, a)\\) and sample the hyperparameters with Hamiltonian Monte Carlo sampling. Then, we recover the integrated out \\(f\\) in the generated quantities block using laplace_latent_rng.\nThe embedded Laplace approximation computes a Gaussian approximation of the conditional posterior, \\[\n  \\hat p_\\mathcal{L}(f \\mid \\rho, \\alpha, a, y) \\approx p(f \\mid \\rho, \\alpha, a, y),\n\\] where \\(\\hat p_\\mathcal{L}\\) is a Gaussian that matches the mode and curvature of \\(p(f \\mid \\rho, \\alpha, a, y)\\). We then obtain an approximation of the marginal likelihood as follows: \\[\n  \\hat p_\\mathcal{L}(y \\mid \\rho, \\alpha, a)\n    = \\frac{p(f^* \\mid \\alpha, \\rho) p(y \\mid f^*, a)}{\n    \\hat p_\\mathcal{L}(f^* \\mid \\rho, \\alpha, a, y)},\n\\] where \\(f^*\\) is the mode of \\(p(f \\mid \\rho, \\alpha, a, y)\\), obtained via numerical optimization.\nTo use Stan’s embedded Laplace approximation, we must define the prior covariance function and the log likelihood function in the functions block.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int y) {\n      return poisson_log_lpmf(y | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n\n}\nThe embedded Laplace relies on calculations of the log likelihood’s Hessian, \\(\\partial^2 \\log p(y \\mid f, a, \\rho, \\alpha) / \\partial f^2\\), and these calculations can be much faster when the Hessian is sparse. In particular, it is expected that the Hessian is block diagonal. In the transformed data block we can specify the block size of the Hessian.\ntransformed data {\n  int hessian_block_size = 1;\n}\nFor example, if \\(y_i\\) depends only on \\(f_i\\), then the Hessian of the log likelihood is diagonal and the block size is 1. On the other hand, if the Hessian is not sparse, then we set the hessian block size to \\(N\\), where \\(N\\) is the dimension of \\(f\\). Currently, Stan does not check the block size of the Hessian and so the user is responsible for correctly specifying the block size.\nFinally, we increment target in the model block with the approximation to \\(\\log p(y \\mid \\rho, \\alpha, a)\\).\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  target += laplace_marginal(ll_function, (a, y), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\nNotice that we do not need to construct \\(f\\) explicitly, since it is marginalized out. Instead, we can recover the latent variables in generated quantities:\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nUsers can set the control parameters of the embedded Laplace approximation, via laplace_marginal_tol and laplace_latent_tol_rng. When using these functions, the user must set all the control options and store them in a tuple. These control parameters mostly concern the numerical optimizer used to find the mode \\(f^*\\) of \\(p(f \\mid \\rho, \\alpha, a)\\).\ntransformed data {\n  tuple(vector[N], real, int, int, int, int) laplace_ops;\n  laplace_ops.1 = rep_vector(0, N);  // starting point for Laplace optimizer\n  laplace_ops.2 = 1.49e-8;           // tolerance for optimizer\n  laplace_ops.3 = 500;               // maximum number of steps for optimizer.\n  laplace_ops.4 = 1;              // solver type being used.\n  laplace_ops.5 = 1000;           // max number of steps for linesearch.\n  laplace_ops.6 = 1;              // allow_fallback (1: TRUE, 0: FALSE)\nIf users want to depart from the defaults for only some of the control parameters, a tuple with the default values (as above) can be created with the helper callable generate_laplace_options(), and the specific control parameter can then be modified,\ntransformed data {\n  tuple(vector[N], real, int, int, int, int, int) laplace_ops =\n    generate_laplace_options(N);\n\n  laplace_ops.2 = 1e-6; // make tolerance of the optimizer less strict.\n}\nThe tuple laplace_ops is then passed to laplace_marginal_tol and laplace_rng_tol.\nmodel {\n// ...\n\n  target += laplace_marginal_tol(ll_function, (a, y), hessian_block_size,\n                                 cov_function, (rho, alpha, x, N, delta),\n                                 laplace_ops);\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta),\n                                   laplace_ops);\n}\nStan also provides support for a limited menu of built-in functions, including the Poisson distribution with a log link and and prior mean \\(m\\). When using such a built-in function, the user does not need to specify a likelihood in the functions block. However, the user must strictly follow the signature of the likelihood: in this case, \\(m\\) must be a vector of length \\(N\\) (to allow for different offsets for each observation \\(y_i\\)) and we must indicate which element of \\(f\\) each component of \\(y\\) matches using the variable \\(y_\\text{index}\\). In our example, there is a simple pairing \\((y_i, f_i)\\), however we could imagine a scenario where multiple observations \\((y_{j1}, y_{j2}, ...)\\) are observed for a single \\(f_j\\).\ntransformed data {\n  // ...\n  array[n_obs] int y_index;\n  for (i in 1:n_obs) y_index[i] = i - 1;\n}\n\n// ...\n\ntransformed parameter {\n  vector[N] m = rep_vector(a, N);\n}\n\nmodel {\n  // ...\n  target += laplace_marginal_poisson_log_lpmf(y | y_index, m,\n                                       cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_poisson_log_rng(y, y_index, m,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nAs before, we could specify the control parameters for the embedded Laplace approximation using laplace_marginal_tol_poisson_log_lpmf and laplace_latent_tol_poisson_log_nrg.\nMarginalization with a Laplace approximation can lead to faster inference, however it also introduces an approximation error. In practice, this error is negligible when using a Poisson likelihood and the approximation works well for log concave likelihoods (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). Still, users should exercise caution, especially when trying unconventional likelihoods.\n\n\n\nFor binary classification problems, the observed outputs \\(z_n \\in\n\\{ 0,1 \\}\\) are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs \\(y_n\\) through the logistic link, \\[\nz_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(y_n)),\n\\] or in other words, \\[\n\\Pr[z_n = 1] = \\operatorname{logit}^{-1}(y_n).\n\\]\nWe can extend our latent variable GP Stan program to deal with classification problems. Below a is the bias term, which can help account for imbalanced classes in the training data:\ndata {\n  // ...\n  array[N] int&lt;lower=0, upper=1&gt; z;\n  // ...\n}\n// ...\nmodel {\n  // ...\n  z ~ bernoulli_logit(a + f);\n}\n\n\n\nAs with the Poisson GP, we cannot marginalize the latent variables exactly, however we can resort to an embedded Laplace approximation.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int z) {\n      return bernoulli_logit_lpmf(z | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n}\n\n// ...\n\ntransformed data {\n  int hessian_block_size = 1;\n}\n\n// ...\n\nmodel {\n  target += laplace_marginal(ll_function, (a, z), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, z), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nWhile marginalization with a Laplace approximation can lead to faster inference, it also introduces an approximation error. In practice, this error may not be negligible with a Bernoulli likelihood; for more discussion see, e.g. (Vehtari et al. 2016; Margossian et al. 2020).\n\n\n\n\nIf we have multivariate inputs \\(x \\in \\mathbb{R}^D\\), the squared exponential covariance function can be further generalized by fitting a scale parameter \\(\\rho_d\\) for each dimension \\(d\\), \\[\n  k(x \\mid \\alpha, \\vec{\\rho}, \\sigma)_{i, j} = \\alpha^2 \\exp\n\\left(-\\dfrac{1}{2}\n\\sum_{d=1}^D \\dfrac{1}{\\rho_d^2} (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j}\\sigma^2.\n\\] The estimation of \\(\\rho\\) was termed “automatic relevance determination” by Neal (1996), but this is misleading, because the magnitude of the scale of the posterior for each \\(\\rho_d\\) is dependent on the scaling of the input data along dimension \\(d\\). Moreover, the scale of the parameters \\(\\rho_d\\) measures non-linearity along the \\(d\\)-th dimension, rather than “relevance” (Piironen and Vehtari 2016).\nA priori, the closer \\(\\rho_d\\) is to zero, the more nonlinear the conditional mean in dimension \\(d\\) is. A posteriori, the actual dependencies between \\(x\\) and \\(y\\) play a role. With one covariate \\(x_1\\) having a linear effect and another covariate \\(x_2\\) having a nonlinear effect, it is possible that \\(\\rho_1 &gt; \\rho_2\\) even if the predictive relevance of \\(x_1\\) is higher (Rasmussen and Williams 2006, 80). The collection of \\(\\rho_d\\) (or \\(1/\\rho_d\\)) parameters can also be modeled hierarchically.\nThe implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying rho to be an array.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  array[D] real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\n\n\n\nFormulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.\nPerhaps most importantly, the parameters \\(\\rho\\) and \\(\\alpha\\) are weakly identified (Zhang 2004). The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.\n\n\nGPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.\nWe may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.\nIf our model consists of only the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\} \\\\\n    & x \\in \\mathbb{R}^{N \\times D}, \\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, inv_gamma_lpdf in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.2.\nIf we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\big(0, K(x \\mid \\alpha, \\rho)\\big) \\\\\ny_i &\\sim \\textsf{normal}\\left(\\beta_0 + x_i \\beta_{[1:D]} + f_i, \\sigma\\right) \\, \\forall i\n  \\in \\{1, \\dots, N\\} \\\\\n    & x_i^T, \\beta_{[1:D]} \\in \\mathbb{R}^D,\\quad\n      x \\in \\mathbb{R}^{N \\times D},\\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: \\[\\begin{align*}\nf(x \\mid a, b, p) &= \\dfrac{\\left(a/b\\right)^{p/2}}{2K_p\\left(\\sqrt{ab}\\right)} x^{p - 1}\\exp\\big(-(ax + b\n  / x)/2\\big) \\\\\n  & x, a, b \\in \\mathbb{R}^{+},\\quad\n    p \\in \\mathbb{Z}\n\\end{align*}\\]\nwhich has an inverse gamma left tail if \\(p \\leq 0\\) and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:\nfunctions {\n  real generalized_inverse_gaussian_lpdf(real x, int p,\n                                        real a, real b) {\n    return p * 0.5 * log(a / b)\n      - log(2 * modified_bessel_second_kind(p, sqrt(a * b)))\n      + (p - 1) * log(x)\n      - (a * x + b / x) * 0.5;\n }\n}\ndata {\n  // ...\n}\nIf we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of \\(\\rho\\) in a fixed-domain \\([0,T]\\), we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is \\(T / \\pi\n\\rho\\). You can see that as \\(\\rho\\) decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.\n\n\n\nThe parameter \\(\\alpha\\) corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-\\(t\\) prior on \\(\\alpha\\).\nA half-\\(t\\) or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.\n\n\n\n\nSuppose for a given sequence of inputs \\(x\\) that the corresponding outputs \\(y\\) are observed. Given a new sequence of inputs \\(\\tilde{x}\\), the posterior predictive distribution of their labels is computed by sampling outputs \\(\\tilde{y}\\) according to \\[\np\\left(\\tilde{y} \\mid \\tilde{x},x,y\\right)\n\\ = \\\n\\frac{p\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right)}\n     {p(y \\mid x)}\n\\ \\propto \\\np\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right).\n\\]\nA direct implementation in Stan defines a model in terms of the joint distribution of the observed \\(y\\) and unobserved \\(\\tilde{y}\\).\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y1 ~ normal(f[1:N1], sigma);\n}\ngenerated quantities {\n  vector[N2] y2;\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f[N1 + n2], sigma);\n  }\n}\nThe input vectors x1 and x2 are declared as data, as is the observed output vector y1. The unknown output vector y2, which corresponds to input vector x2, is declared in the generated quantities block and will be sampled when the model is executed.\nA transformed data block is used to combine the input vectors x1 and x2 into a single vector x.\nThe model block declares and defines a local variable for the combined output vector f, which consists of the concatenation of the conditional mean for known outputs y1 and unknown outputs y2. Thus the combined output vector f is aligned with the combined input vector x. All that is left is to define the univariate normal distribution statement for y.\nThe generated quantities block defines the quantity y2. We generate y2 by randomly generating N2 values from univariate normals with each mean corresponding to the appropriate element in f.\n\n\nWe can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.\nConsider the following full model for prediction using logistic Gaussian process regression.\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  array[N1] int&lt;lower=0, upper=1&gt; z1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  z1 ~ bernoulli_logit(a + f[1:N1]);\n}\ngenerated quantities {\n  array[N2] int z2;\n  for (n2 in 1:N2) {\n    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);\n  }\n}\n\n\n\nBayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.\nJumping straight to the result, \\[\np\\left(\\tilde{y} \\mid \\tilde{x},y,x\\right)\n=\n\\textsf{normal}\\left(K^{\\top}\\Sigma^{-1}y,\\\n                \\Omega - K^{\\top}\\Sigma^{-1}K\\right),\n\\] where \\(\\Sigma = K(x \\mid \\alpha, \\rho, \\sigma)\\) is the result of applying the covariance function to the inputs \\(x\\) with observed outputs \\(y\\), \\(\\Omega =\nK(\\tilde{x} \\mid \\alpha, \\rho)\\) is the result of applying the covariance function to the inputs \\(\\tilde{x}\\) for which predictions are to be inferred, and \\(K\\) is the matrix of covariances between inputs \\(x\\) and \\(\\tilde{x}\\), which in the case of the exponentiated quadratic covariance function would be \\[\nK(x \\mid \\alpha, \\rho)_{i, j} = \\alpha^2 \\exp\\left(-\\dfrac{1}{2 \\rho^2}\n\\sum_{d=1}^D \\left(x_{i,d} - \\tilde{x}_{j,d}\\right)^2\\right).\n\\]\nThere is no noise term including \\(\\sigma^2\\) because the indexes of elements in \\(x\\) and \\(\\tilde{x}\\) are never the same.\nThis Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called gp_pred_rng which will generate a draw from the posterior predictive mean conditioned on observed data y1. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of \\(p(\\tilde{y})\\).\nfunctions {\n  vector gp_pred_rng(array[] real x2,\n                     vector y1,\n                     array[] real x1,\n                     real alpha,\n                     real rho,\n                     real sigma,\n                     real delta) {\n    int N1 = rows(y1);\n    int N2 = size(x2);\n    vector[N2] f2;\n    {\n      matrix[N1, N1] L_K;\n      vector[N1] K_div_y1;\n      matrix[N1, N2] k_x1_x2;\n      matrix[N1, N2] v_pred;\n      vector[N2] f2_mu;\n      matrix[N2, N2] cov_f2;\n      matrix[N2, N2] diag_delta;\n      matrix[N1, N1] K;\n      K = gp_exp_quad_cov(x1, alpha, rho);\n      for (n in 1:N1) {\n        K[n, n] = K[n, n] + square(sigma);\n      }\n      L_K = cholesky_decompose(K);\n      K_div_y1 = mdivide_left_tri_low(L_K, y1);\n      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';\n      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);\n      f2_mu = (k_x1_x2' * K_div_y1);\n      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);\n      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;\n      diag_delta = diag_matrix(rep_vector(delta, N2));\n\n      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);\n    }\n    return f2;\n  }\n}\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  vector[N1] mu = rep_vector(0, N1);\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N1, N1] L_K;\n  {\n    matrix[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);\n    real sq_sigma = square(sigma);\n\n    // diagonal elements\n    for (n1 in 1:N1) {\n      K[n1, n1] = K[n1, n1] + sq_sigma;\n    }\n\n    L_K = cholesky_decompose(K);\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y1 ~ multi_normal_cholesky(mu, L_K);\n}\ngenerated quantities {\n  vector[N2] f2;\n  vector[N2] y2;\n\n  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f2[n2], sigma);\n  }\n}\n\n\n\n\nSuppose we have observations \\(y_i \\in \\mathbb{R}^M\\) observed at \\(x_i \\in \\mathbb{R}^K\\). One can model the data like so: \\[\\begin{align*}\ny_i  &\\sim \\textsf{multivariate normal}\\left(f(x_i), \\textbf{I}_M \\sigma^2\\right) \\\\\nf(x) &\\sim \\textsf{GP}\\big(m(x), K(x \\mid \\theta, \\phi)\\big) \\\\\n     & K(x \\mid \\theta) \\in \\mathbb{R}^{M \\times M}, \\quad\n       f(x), m(x) \\in \\mathbb{R}^M\n\\end{align*}\\] where the \\(K(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]}\\) entry defines the covariance between \\(f_m(x)\\) and \\(f_{m^\\prime}(x^\\prime)(x)\\). This construction of Gaussian processes allows us to learn the covariance between the output dimensions of \\(f(x)\\). If we parameterize our kernel \\(K\\): \\[\nK(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]} = k\\left(x, x^\\prime \\mid\n\\theta\\right) k\\left(m, m^\\prime \\mid \\phi\\right)\n\\] then our finite dimensional generative model for the above is: \\[\\begin{align*}\nf        &\\sim \\textsf{matrixnormal}\\big(m(x), K(x \\mid \\alpha, \\rho), C(\\phi)\\big) \\\\\ny_{i, m} &\\sim \\textsf{normal}(f_{i,m}, \\sigma) \\\\\nf        &\\in  \\mathbb{R}^{N \\times M}\n\\end{align*}\\] where \\(K(x \\mid \\alpha, \\rho)\\) is the exponentiated quadratic kernel we’ve used throughout this chapter, and \\(C(\\phi)\\) is a positive-definite matrix, parameterized by some vector \\(\\phi\\).\nThe matrix normal distribution has two covariance matrices: \\(K(x \\mid\n\\alpha, \\rho)\\) to encode column covariance, and \\(C(\\phi)\\) to define row covariance. The salient features of the matrix normal are that the rows of the matrix \\(f\\) are distributed: \\[\nf_{[n,]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[n,]}, K(x \\mid \\alpha,\n\\rho)_{[n,n]} C(\\phi)\\big)\n\\] and that the columns of the matrix \\(f\\) are distributed: \\[\nf_{[,m]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[,m]}, K(x\n  \\mid \\alpha, \\rho) C(\\phi)_{[m,m]}\\big)\n\\] This also means means that \\(\\mathbb{E}\\left[f^T f\\right]\\) is equal to \\(\\operatorname{trace}\\!\\big(K(x \\mid \\alpha, \\rho)\\big) \\times C\\), whereas \\(\\mathbb{E}\\left[ff^T\\right]\\) is \\(\\operatorname{trace}(C) \\times K(x \\mid \\alpha, \\rho)\\). We can derive this using properties of expectation and the matrix normal density.\nWe should set \\(\\alpha\\) to \\(1.0\\) because the parameter is not identified unless we constrain \\(\\operatorname{trace}(C) = 1\\). Otherwise, we can multiply \\(\\alpha\\) by a scalar \\(d\\) and \\(C\\) by \\(1/d\\) and our likelihood will not change.\nWe can generate a random variable \\(f\\) from a matrix normal density in \\(\\mathbb{R}^{N \\times M}\\) using the following algorithm: \\[\\begin{align*}\n\\eta_{i,j} &\\sim \\textsf{normal}(0, 1) \\, \\forall i,j \\\\\nf          &= L_{K(x \\mid 1.0, \\rho)} \\, \\eta \\, L_C(\\phi)^T \\\\\nf          &\\sim \\textsf{matrixnormal}\\big(0, K(x \\mid 1.0, \\rho), C(\\phi)\\big) \\\\\n\\eta       &\\in \\mathbb{R}^{N \\times M} \\\\\nL_C(\\phi)  &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(C(\\phi)\\big) \\\\\nL_{K(x \\mid 1.0, \\rho)} &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(K(x \\mid 1.0, \\rho)\\big)\n\\end{align*}\\]\nThis can be implemented in Stan using a latent-variable GP formulation. We’ve used \\(\\textsf{LKJCorr}\\) for \\(C(\\phi)\\), but any positive-definite matrix will do.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] real x;\n  matrix[N, D] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  vector&lt;lower=0&gt;[D] alpha;\n  real&lt;lower=0&gt; sigma;\n  cholesky_factor_corr[D] L_Omega;\n  matrix[N, D] eta;\n}\nmodel {\n  matrix[N, D] f;\n  {\n    matrix[N, N] K = gp_exp_quad_cov(x, 1.0, rho);\n    matrix[N, N] L_K;\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta\n        * diag_pre_multiply(alpha, L_Omega)';\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(3);\n  to_vector(eta) ~ std_normal();\n\n  to_vector(y) ~ normal(to_vector(f), sigma);\n}\ngenerated quantities {\n  matrix[D, D] Omega;\n  Omega = L_Omega * L_Omega';\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Gaussian Processes"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/gaussian-processes.html#gaussian-process-regression",
+    "href": "stan-users-guide/gaussian-processes.html#gaussian-process-regression",
+    "title": "Gaussian Processes",
+    "section": "",
+    "text": "The data for a multivariate Gaussian process regression consists of a series of \\(N\\) inputs \\(x_1,\\dotsc,x_N \\in \\mathbb{R}^D\\) paired with outputs \\(y_1,\\dotsc,y_N \\in \\mathbb{R}\\). The defining feature of Gaussian processes is that the probability of a finite number of outputs \\(y\\) conditioned on their inputs \\(x\\) is Gaussian: \\[\ny \\sim \\textsf{multivariate normal}(m(x), K(x \\mid \\theta)),\n\\] where \\(m(x)\\) is an \\(N\\)-vector and \\(K(x \\mid \\theta)\\) is an \\(N \\times N\\) covariance matrix. The mean function \\(m : \\mathbb{R}^{N \\times D}\n\\rightarrow \\mathbb{R}^{N}\\) can be anything, but the covariance function \\(K : \\mathbb{R}^{N \\times D} \\rightarrow \\mathbb{R}^{N \\times N}\\) must produce a positive-definite matrix for any input \\(x\\).1\nA popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, \\[\n  K(x \\mid \\alpha, \\rho, \\sigma)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j} \\sigma^2,\n\\] where \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\) are hyperparameters defining the covariance function and where \\(\\delta_{i, j}\\) is the Kronecker delta function with value 1 if \\(i = j\\) and value 0 otherwise; this test is between the indexes \\(i\\) and \\(j\\), not between values \\(x_i\\) and \\(x_j\\). This kernel is obtained through a convolution of two independent Gaussian processes, \\(f_1\\) and \\(f_2\\), with kernels \\[\n  K_1(x \\mid \\alpha, \\rho)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n\\] and \\[\n  K_2(x \\mid \\sigma)_{i, j}\n=\n\\delta_{i, j} \\sigma^2.\n\\]\nThe addition of \\(\\sigma^2\\) on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs \\(x_i = x_j\\). In statistical terms, \\(\\sigma\\) is the scale of the noise term in the regression.\nThe hyperparameter \\(\\rho\\) is the length-scale, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of \\(\\rho\\) closer to zero lead the GP to represent high-frequency functions, whereas larger values of \\(\\rho\\) lead to low-frequency functions. The hyperparameter \\(\\alpha\\) is the marginal standard deviation. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP \\(f_1\\) prior at a single input \\(x\\) conditional on one value of \\(\\alpha\\) one would recover \\(\\alpha\\).\nThe only term in the squared exponential covariance function involving the inputs \\(x_i\\) and \\(x_j\\) is their vector difference, \\(x_i - x_j\\). This produces a process with stationary covariance in the sense that if an input vector \\(x\\) is translated by a vector \\(\\epsilon\\) to \\(x +\n\\epsilon\\), the covariance at any pair of outputs is unchanged, because \\(K(x \\mid \\theta) = K(x + \\epsilon \\mid \\theta)\\).\nThe summation involved is just the squared Euclidean distance between \\(x_i\\) and \\(x_j\\) (i.e., the squared \\(L_2\\) norm of their difference, \\(x_i -\nx_j\\)). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\).\nChanging the notion of distance from Euclidean to taxicab distance (i.e., an \\(L_1\\) norm) changes the support to functions which are continuous but not smooth.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Gaussian Processes"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/gaussian-processes.html#simulating-from-a-gaussian-process",
+    "href": "stan-users-guide/gaussian-processes.html#simulating-from-a-gaussian-process",
+    "title": "Gaussian Processes",
+    "section": "",
+    "text": "It is simplest to start with a Stan model that does nothing more than simulate draws of functions \\(f\\) from a Gaussian process. In practical terms, the model will draw values \\(y_n = f(x_n)\\) for finitely many input points \\(x_n\\).\nThe Stan model defines the mean and covariance functions in a transformed data block and then samples outputs \\(y\\) in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to \\(\\alpha^2 = 1\\), \\(\\rho^2 = 1\\), and \\(\\sigma^2 = 0.1\\), and the mean function \\(m\\) is defined to always return the zero vector, \\(m(x) = \\textbf{0}\\). Consider the following implementation of a Gaussian process simulator.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K;\n  vector[N] mu = rep_vector(0, N);\n  for (i in 1:(N - 1)) {\n    K[i, i] = 1 + 0.1;\n    for (j in (i + 1):N) {\n      K[i, j] = exp(-0.5 * square(x[i] - x[j]));\n      K[j, i] = K[i, j];\n    }\n  }\n  K[N, N] = 1 + 0.1;\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K = gp_exp_quad_cov(x, 1.0, 1.0);\n  vector[N] mu = rep_vector(0, N);\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + 0.1;\n  }\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe input data are just the vector of inputs x and its size N. Such a model can be used with values of x evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.\n\n\nOnly the input data needs to change in moving from a univariate model to a multivariate model.\nThe only lines that change from the univariate model above are as follows.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n}\ntransformed data {\n  // ...\n}\nThe data are now declared as an array of vectors instead of an array of scalars; the dimensionality D is also declared.\nIn the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.\n\n\n\nA more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose \\(\\eta\\) is an an isotropic standard normal variate \\[\n\\eta \\sim \\textsf{normal}(\\textbf{0}, \\textrm{I}),\n\\] where \\(\\textbf{0}\\) is an \\(N\\)-vector of 0 values and \\(\\textrm{I}\\) is the \\(N\n\\times N\\) identity matrix. Let \\(L\\) be the Cholesky decomposition of \\(K(x \\mid \\theta)\\), i.e., the lower-triangular matrix \\(L\\) such that \\(LL^{\\top} =\nK(x \\mid \\theta)\\). Then the transformed variable \\(\\mu + L\\eta\\) has the intended target distribution, \\[\n  \\mu + L\\eta \\sim \\textsf{multivariate normal}(\\mu(x), K(x \\mid \\theta)).\n\\]\nThis transform can be applied directly to Gaussian process simulation.\nThis model has the same data declarations for N and x, and the same transformed data definitions of mu and K as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.\n// ...\ntransformed data {\n  matrix[N, N] L;\n  // ...\n  L = cholesky_decompose(K);\n}\nparameters {\n  vector[N] eta;\n}\nmodel {\n  eta ~ std_normal();\n}\ngenerated quantities {\n  vector[N] y;\n  y = mu + L * eta;\n}\nThe Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix K computed. The isotropic normal distribution for eta is specified as a vectorized univariate distribution for efficiency; this specifies that each eta[n] has an independent standard normal distribution. The sampled vector y is then defined as a generated quantity using a direct encoding of the transform described above.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Gaussian Processes"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/gaussian-processes.html#fit-gp.section",
+    "href": "stan-users-guide/gaussian-processes.html#fit-gp.section",
+    "title": "Gaussian Processes",
+    "section": "",
+    "text": "The full generative model for a GP with a normal outcome, \\(y \\in \\mathbb{R}^N\\), with inputs \\(x \\in \\mathbb{R}^N\\), for a finite \\(N\\): \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\nf      &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i    &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\}\n\\end{align*}\\] With a normal outcome, it is possible to integrate out the Gaussian process \\(f\\), yielding the more parsimonious model: \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\ny      &\\sim \\textsf{multivariate normal}\n         \\left(0, K(x \\mid \\alpha, \\rho) + \\textbf{I}_N \\sigma^2\\right) \\\\\n\\end{align*}\\]\nIt can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.\nThe hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (\\(\\alpha\\)), the scale of the output noise (\\(\\sigma\\)), and the scale at which distances are measured among inputs (\\(\\rho\\)). See the Gaussian process priors section for more information about how to specify appropriate priors for the hyperparameters.\nThe Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix K in the model block, rather than the transformed data block.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  vector[N] mu = rep_vector(0, N);\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N, N] L_K;\n  matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n  real sq_sigma = square(sigma);\n\n  // diagonal elements\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + sq_sigma;\n  }\n\n  L_K = cholesky_decompose(K);\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y ~ multi_normal_cholesky(mu, L_K);\n}\nThe data block declares a vector y of observed values y[n] for inputs x[n]. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix K is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value y is known and the covariance matrix K is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.\nWe have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the cholesky_decompose function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (\\(N \\gtrsim 100\\)) the Cholesky decomposition version will be faster.\nHamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model (Neal 1997). If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.\n\n\nWe can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, \\(\\delta\\) to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\nTwo differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length \\(N\\) called eta. This is used in the model block to generate a multivariate normal vector called \\(f\\), corresponding to the latent GP. We put a \\(\\textsf{normal}(0,1)\\) prior on eta like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for \\(y\\) with one \\(N\\)-dimensional multivariate normal with an identity covariance matrix multiplied by \\(\\sigma^2\\), we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.\n\n\n\n\nGaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.\n\n\nIf we want to model count data, we can remove the \\(\\sigma\\) parameter, and use poisson_log, which implements a Poisson distribution with log link function, rather than normal. We can also add an overall mean parameter, \\(a\\), which will account for the marginal expected value for \\(y\\). We do this because we cannot center count data like we would for normally distributed data.\ndata {\n  // ...\n  array[N] int&lt;lower=0&gt; y;\n  // ...\n}\n// ...\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ poisson_log(a + f);\n}\n\n\n\nFor computational reasons, we may want to integrate out the latent variable \\(f\\), as was done in the normal output model. Unfortunately, exact marginalization over \\(f\\) is not possible when the outcome model is not normal. Instead, we may perform approximate marginalization with an embedded Laplace approximation (Rasmussen and Williams 2006; Rue, Martino, and Chopin 2009; Margossian et al. 2020). To do so, we first use the function laplace_marginal to approximate the marginal likelihood \\(p(y \\mid \\rho, \\alpha, a)\\) and sample the hyperparameters with Hamiltonian Monte Carlo sampling. Then, we recover the integrated out \\(f\\) in the generated quantities block using laplace_latent_rng.\nThe embedded Laplace approximation computes a Gaussian approximation of the conditional posterior, \\[\n  \\hat p_\\mathcal{L}(f \\mid \\rho, \\alpha, a, y) \\approx p(f \\mid \\rho, \\alpha, a, y),\n\\] where \\(\\hat p_\\mathcal{L}\\) is a Gaussian that matches the mode and curvature of \\(p(f \\mid \\rho, \\alpha, a, y)\\). We then obtain an approximation of the marginal likelihood as follows: \\[\n  \\hat p_\\mathcal{L}(y \\mid \\rho, \\alpha, a)\n    = \\frac{p(f^* \\mid \\alpha, \\rho) p(y \\mid f^*, a)}{\n    \\hat p_\\mathcal{L}(f^* \\mid \\rho, \\alpha, a, y)},\n\\] where \\(f^*\\) is the mode of \\(p(f \\mid \\rho, \\alpha, a, y)\\), obtained via numerical optimization.\nTo use Stan’s embedded Laplace approximation, we must define the prior covariance function and the log likelihood function in the functions block.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int y) {\n      return poisson_log_lpmf(y | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n\n}\nThe embedded Laplace relies on calculations of the log likelihood’s Hessian, \\(\\partial^2 \\log p(y \\mid f, a, \\rho, \\alpha) / \\partial f^2\\), and these calculations can be much faster when the Hessian is sparse. In particular, it is expected that the Hessian is block diagonal. In the transformed data block we can specify the block size of the Hessian.\ntransformed data {\n  int hessian_block_size = 1;\n}\nFor example, if \\(y_i\\) depends only on \\(f_i\\), then the Hessian of the log likelihood is diagonal and the block size is 1. On the other hand, if the Hessian is not sparse, then we set the hessian block size to \\(N\\), where \\(N\\) is the dimension of \\(f\\). Currently, Stan does not check the block size of the Hessian and so the user is responsible for correctly specifying the block size.\nFinally, we increment target in the model block with the approximation to \\(\\log p(y \\mid \\rho, \\alpha, a)\\).\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  target += laplace_marginal(ll_function, (a, y), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\nNotice that we do not need to construct \\(f\\) explicitly, since it is marginalized out. Instead, we can recover the latent variables in generated quantities:\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nUsers can set the control parameters of the embedded Laplace approximation, via laplace_marginal_tol and laplace_latent_tol_rng. When using these functions, the user must set all the control options and store them in a tuple. These control parameters mostly concern the numerical optimizer used to find the mode \\(f^*\\) of \\(p(f \\mid \\rho, \\alpha, a)\\).\ntransformed data {\n  tuple(vector[N], real, int, int, int, int) laplace_ops;\n  laplace_ops.1 = rep_vector(0, N);  // starting point for Laplace optimizer\n  laplace_ops.2 = 1.49e-8;           // tolerance for optimizer\n  laplace_ops.3 = 500;               // maximum number of steps for optimizer.\n  laplace_ops.4 = 1;              // solver type being used.\n  laplace_ops.5 = 1000;           // max number of steps for linesearch.\n  laplace_ops.6 = 1;              // allow_fallback (1: TRUE, 0: FALSE)\nIf users want to depart from the defaults for only some of the control parameters, a tuple with the default values (as above) can be created with the helper callable generate_laplace_options(), and the specific control parameter can then be modified,\ntransformed data {\n  tuple(vector[N], real, int, int, int, int, int) laplace_ops =\n    generate_laplace_options(N);\n\n  laplace_ops.2 = 1e-6; // make tolerance of the optimizer less strict.\n}\nThe tuple laplace_ops is then passed to laplace_marginal_tol and laplace_rng_tol.\nmodel {\n// ...\n\n  target += laplace_marginal_tol(ll_function, (a, y), hessian_block_size,\n                                 cov_function, (rho, alpha, x, N, delta),\n                                 laplace_ops);\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta),\n                                   laplace_ops);\n}\nStan also provides support for a limited menu of built-in functions, including the Poisson distribution with a log link and and prior mean \\(m\\). When using such a built-in function, the user does not need to specify a likelihood in the functions block. However, the user must strictly follow the signature of the likelihood: in this case, \\(m\\) must be a vector of length \\(N\\) (to allow for different offsets for each observation \\(y_i\\)) and we must indicate which element of \\(f\\) each component of \\(y\\) matches using the variable \\(y_\\text{index}\\). In our example, there is a simple pairing \\((y_i, f_i)\\), however we could imagine a scenario where multiple observations \\((y_{j1}, y_{j2}, ...)\\) are observed for a single \\(f_j\\).\ntransformed data {\n  // ...\n  array[n_obs] int y_index;\n  for (i in 1:n_obs) y_index[i] = i - 1;\n}\n\n// ...\n\ntransformed parameter {\n  vector[N] m = rep_vector(a, N);\n}\n\nmodel {\n  // ...\n  target += laplace_marginal_poisson_log_lpmf(y | y_index, m,\n                                       cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_poisson_log_rng(y, y_index, m,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nAs before, we could specify the control parameters for the embedded Laplace approximation using laplace_marginal_tol_poisson_log_lpmf and laplace_latent_tol_poisson_log_nrg.\nMarginalization with a Laplace approximation can lead to faster inference, however it also introduces an approximation error. In practice, this error is negligible when using a Poisson likelihood and the approximation works well for log concave likelihoods (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). Still, users should exercise caution, especially when trying unconventional likelihoods.\n\n\n\nFor binary classification problems, the observed outputs \\(z_n \\in\n\\{ 0,1 \\}\\) are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs \\(y_n\\) through the logistic link, \\[\nz_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(y_n)),\n\\] or in other words, \\[\n\\Pr[z_n = 1] = \\operatorname{logit}^{-1}(y_n).\n\\]\nWe can extend our latent variable GP Stan program to deal with classification problems. Below a is the bias term, which can help account for imbalanced classes in the training data:\ndata {\n  // ...\n  array[N] int&lt;lower=0, upper=1&gt; z;\n  // ...\n}\n// ...\nmodel {\n  // ...\n  z ~ bernoulli_logit(a + f);\n}\n\n\n\nAs with the Poisson GP, we cannot marginalize the latent variables exactly, however we can resort to an embedded Laplace approximation.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int z) {\n      return bernoulli_logit_lpmf(z | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n}\n\n// ...\n\ntransformed data {\n  int hessian_block_size = 1;\n}\n\n// ...\n\nmodel {\n  target += laplace_marginal(ll_function, (a, z), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, z), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nWhile marginalization with a Laplace approximation can lead to faster inference, it also introduces an approximation error. In practice, this error may not be negligible with a Bernoulli likelihood; for more discussion see, e.g. (Vehtari et al. 2016; Margossian et al. 2020).\n\n\n\n\nIf we have multivariate inputs \\(x \\in \\mathbb{R}^D\\), the squared exponential covariance function can be further generalized by fitting a scale parameter \\(\\rho_d\\) for each dimension \\(d\\), \\[\n  k(x \\mid \\alpha, \\vec{\\rho}, \\sigma)_{i, j} = \\alpha^2 \\exp\n\\left(-\\dfrac{1}{2}\n\\sum_{d=1}^D \\dfrac{1}{\\rho_d^2} (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j}\\sigma^2.\n\\] The estimation of \\(\\rho\\) was termed “automatic relevance determination” by Neal (1996), but this is misleading, because the magnitude of the scale of the posterior for each \\(\\rho_d\\) is dependent on the scaling of the input data along dimension \\(d\\). Moreover, the scale of the parameters \\(\\rho_d\\) measures non-linearity along the \\(d\\)-th dimension, rather than “relevance” (Piironen and Vehtari 2016).\nA priori, the closer \\(\\rho_d\\) is to zero, the more nonlinear the conditional mean in dimension \\(d\\) is. A posteriori, the actual dependencies between \\(x\\) and \\(y\\) play a role. With one covariate \\(x_1\\) having a linear effect and another covariate \\(x_2\\) having a nonlinear effect, it is possible that \\(\\rho_1 &gt; \\rho_2\\) even if the predictive relevance of \\(x_1\\) is higher (Rasmussen and Williams 2006, 80). The collection of \\(\\rho_d\\) (or \\(1/\\rho_d\\)) parameters can also be modeled hierarchically.\nThe implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying rho to be an array.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  array[D] real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\n\n\n\nFormulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.\nPerhaps most importantly, the parameters \\(\\rho\\) and \\(\\alpha\\) are weakly identified (Zhang 2004). The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.\n\n\nGPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.\nWe may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.\nIf our model consists of only the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\} \\\\\n    & x \\in \\mathbb{R}^{N \\times D}, \\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, inv_gamma_lpdf in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.2.\nIf we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\big(0, K(x \\mid \\alpha, \\rho)\\big) \\\\\ny_i &\\sim \\textsf{normal}\\left(\\beta_0 + x_i \\beta_{[1:D]} + f_i, \\sigma\\right) \\, \\forall i\n  \\in \\{1, \\dots, N\\} \\\\\n    & x_i^T, \\beta_{[1:D]} \\in \\mathbb{R}^D,\\quad\n      x \\in \\mathbb{R}^{N \\times D},\\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: \\[\\begin{align*}\nf(x \\mid a, b, p) &= \\dfrac{\\left(a/b\\right)^{p/2}}{2K_p\\left(\\sqrt{ab}\\right)} x^{p - 1}\\exp\\big(-(ax + b\n  / x)/2\\big) \\\\\n  & x, a, b \\in \\mathbb{R}^{+},\\quad\n    p \\in \\mathbb{Z}\n\\end{align*}\\]\nwhich has an inverse gamma left tail if \\(p \\leq 0\\) and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:\nfunctions {\n  real generalized_inverse_gaussian_lpdf(real x, int p,\n                                        real a, real b) {\n    return p * 0.5 * log(a / b)\n      - log(2 * modified_bessel_second_kind(p, sqrt(a * b)))\n      + (p - 1) * log(x)\n      - (a * x + b / x) * 0.5;\n }\n}\ndata {\n  // ...\n}\nIf we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of \\(\\rho\\) in a fixed-domain \\([0,T]\\), we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is \\(T / \\pi\n\\rho\\). You can see that as \\(\\rho\\) decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.\n\n\n\nThe parameter \\(\\alpha\\) corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-\\(t\\) prior on \\(\\alpha\\).\nA half-\\(t\\) or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.\n\n\n\n\nSuppose for a given sequence of inputs \\(x\\) that the corresponding outputs \\(y\\) are observed. Given a new sequence of inputs \\(\\tilde{x}\\), the posterior predictive distribution of their labels is computed by sampling outputs \\(\\tilde{y}\\) according to \\[\np\\left(\\tilde{y} \\mid \\tilde{x},x,y\\right)\n\\ = \\\n\\frac{p\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right)}\n     {p(y \\mid x)}\n\\ \\propto \\\np\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right).\n\\]\nA direct implementation in Stan defines a model in terms of the joint distribution of the observed \\(y\\) and unobserved \\(\\tilde{y}\\).\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y1 ~ normal(f[1:N1], sigma);\n}\ngenerated quantities {\n  vector[N2] y2;\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f[N1 + n2], sigma);\n  }\n}\nThe input vectors x1 and x2 are declared as data, as is the observed output vector y1. The unknown output vector y2, which corresponds to input vector x2, is declared in the generated quantities block and will be sampled when the model is executed.\nA transformed data block is used to combine the input vectors x1 and x2 into a single vector x.\nThe model block declares and defines a local variable for the combined output vector f, which consists of the concatenation of the conditional mean for known outputs y1 and unknown outputs y2. Thus the combined output vector f is aligned with the combined input vector x. All that is left is to define the univariate normal distribution statement for y.\nThe generated quantities block defines the quantity y2. We generate y2 by randomly generating N2 values from univariate normals with each mean corresponding to the appropriate element in f.\n\n\nWe can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.\nConsider the following full model for prediction using logistic Gaussian process regression.\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  array[N1] int&lt;lower=0, upper=1&gt; z1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  z1 ~ bernoulli_logit(a + f[1:N1]);\n}\ngenerated quantities {\n  array[N2] int z2;\n  for (n2 in 1:N2) {\n    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);\n  }\n}\n\n\n\nBayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.\nJumping straight to the result, \\[\np\\left(\\tilde{y} \\mid \\tilde{x},y,x\\right)\n=\n\\textsf{normal}\\left(K^{\\top}\\Sigma^{-1}y,\\\n                \\Omega - K^{\\top}\\Sigma^{-1}K\\right),\n\\] where \\(\\Sigma = K(x \\mid \\alpha, \\rho, \\sigma)\\) is the result of applying the covariance function to the inputs \\(x\\) with observed outputs \\(y\\), \\(\\Omega =\nK(\\tilde{x} \\mid \\alpha, \\rho)\\) is the result of applying the covariance function to the inputs \\(\\tilde{x}\\) for which predictions are to be inferred, and \\(K\\) is the matrix of covariances between inputs \\(x\\) and \\(\\tilde{x}\\), which in the case of the exponentiated quadratic covariance function would be \\[\nK(x \\mid \\alpha, \\rho)_{i, j} = \\alpha^2 \\exp\\left(-\\dfrac{1}{2 \\rho^2}\n\\sum_{d=1}^D \\left(x_{i,d} - \\tilde{x}_{j,d}\\right)^2\\right).\n\\]\nThere is no noise term including \\(\\sigma^2\\) because the indexes of elements in \\(x\\) and \\(\\tilde{x}\\) are never the same.\nThis Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called gp_pred_rng which will generate a draw from the posterior predictive mean conditioned on observed data y1. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of \\(p(\\tilde{y})\\).\nfunctions {\n  vector gp_pred_rng(array[] real x2,\n                     vector y1,\n                     array[] real x1,\n                     real alpha,\n                     real rho,\n                     real sigma,\n                     real delta) {\n    int N1 = rows(y1);\n    int N2 = size(x2);\n    vector[N2] f2;\n    {\n      matrix[N1, N1] L_K;\n      vector[N1] K_div_y1;\n      matrix[N1, N2] k_x1_x2;\n      matrix[N1, N2] v_pred;\n      vector[N2] f2_mu;\n      matrix[N2, N2] cov_f2;\n      matrix[N2, N2] diag_delta;\n      matrix[N1, N1] K;\n      K = gp_exp_quad_cov(x1, alpha, rho);\n      for (n in 1:N1) {\n        K[n, n] = K[n, n] + square(sigma);\n      }\n      L_K = cholesky_decompose(K);\n      K_div_y1 = mdivide_left_tri_low(L_K, y1);\n      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';\n      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);\n      f2_mu = (k_x1_x2' * K_div_y1);\n      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);\n      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;\n      diag_delta = diag_matrix(rep_vector(delta, N2));\n\n      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);\n    }\n    return f2;\n  }\n}\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  vector[N1] mu = rep_vector(0, N1);\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N1, N1] L_K;\n  {\n    matrix[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);\n    real sq_sigma = square(sigma);\n\n    // diagonal elements\n    for (n1 in 1:N1) {\n      K[n1, n1] = K[n1, n1] + sq_sigma;\n    }\n\n    L_K = cholesky_decompose(K);\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y1 ~ multi_normal_cholesky(mu, L_K);\n}\ngenerated quantities {\n  vector[N2] f2;\n  vector[N2] y2;\n\n  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f2[n2], sigma);\n  }\n}\n\n\n\n\nSuppose we have observations \\(y_i \\in \\mathbb{R}^M\\) observed at \\(x_i \\in \\mathbb{R}^K\\). One can model the data like so: \\[\\begin{align*}\ny_i  &\\sim \\textsf{multivariate normal}\\left(f(x_i), \\textbf{I}_M \\sigma^2\\right) \\\\\nf(x) &\\sim \\textsf{GP}\\big(m(x), K(x \\mid \\theta, \\phi)\\big) \\\\\n     & K(x \\mid \\theta) \\in \\mathbb{R}^{M \\times M}, \\quad\n       f(x), m(x) \\in \\mathbb{R}^M\n\\end{align*}\\] where the \\(K(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]}\\) entry defines the covariance between \\(f_m(x)\\) and \\(f_{m^\\prime}(x^\\prime)(x)\\). This construction of Gaussian processes allows us to learn the covariance between the output dimensions of \\(f(x)\\). If we parameterize our kernel \\(K\\): \\[\nK(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]} = k\\left(x, x^\\prime \\mid\n\\theta\\right) k\\left(m, m^\\prime \\mid \\phi\\right)\n\\] then our finite dimensional generative model for the above is: \\[\\begin{align*}\nf        &\\sim \\textsf{matrixnormal}\\big(m(x), K(x \\mid \\alpha, \\rho), C(\\phi)\\big) \\\\\ny_{i, m} &\\sim \\textsf{normal}(f_{i,m}, \\sigma) \\\\\nf        &\\in  \\mathbb{R}^{N \\times M}\n\\end{align*}\\] where \\(K(x \\mid \\alpha, \\rho)\\) is the exponentiated quadratic kernel we’ve used throughout this chapter, and \\(C(\\phi)\\) is a positive-definite matrix, parameterized by some vector \\(\\phi\\).\nThe matrix normal distribution has two covariance matrices: \\(K(x \\mid\n\\alpha, \\rho)\\) to encode column covariance, and \\(C(\\phi)\\) to define row covariance. The salient features of the matrix normal are that the rows of the matrix \\(f\\) are distributed: \\[\nf_{[n,]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[n,]}, K(x \\mid \\alpha,\n\\rho)_{[n,n]} C(\\phi)\\big)\n\\] and that the columns of the matrix \\(f\\) are distributed: \\[\nf_{[,m]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[,m]}, K(x\n  \\mid \\alpha, \\rho) C(\\phi)_{[m,m]}\\big)\n\\] This also means means that \\(\\mathbb{E}\\left[f^T f\\right]\\) is equal to \\(\\operatorname{trace}\\!\\big(K(x \\mid \\alpha, \\rho)\\big) \\times C\\), whereas \\(\\mathbb{E}\\left[ff^T\\right]\\) is \\(\\operatorname{trace}(C) \\times K(x \\mid \\alpha, \\rho)\\). We can derive this using properties of expectation and the matrix normal density.\nWe should set \\(\\alpha\\) to \\(1.0\\) because the parameter is not identified unless we constrain \\(\\operatorname{trace}(C) = 1\\). Otherwise, we can multiply \\(\\alpha\\) by a scalar \\(d\\) and \\(C\\) by \\(1/d\\) and our likelihood will not change.\nWe can generate a random variable \\(f\\) from a matrix normal density in \\(\\mathbb{R}^{N \\times M}\\) using the following algorithm: \\[\\begin{align*}\n\\eta_{i,j} &\\sim \\textsf{normal}(0, 1) \\, \\forall i,j \\\\\nf          &= L_{K(x \\mid 1.0, \\rho)} \\, \\eta \\, L_C(\\phi)^T \\\\\nf          &\\sim \\textsf{matrixnormal}\\big(0, K(x \\mid 1.0, \\rho), C(\\phi)\\big) \\\\\n\\eta       &\\in \\mathbb{R}^{N \\times M} \\\\\nL_C(\\phi)  &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(C(\\phi)\\big) \\\\\nL_{K(x \\mid 1.0, \\rho)} &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(K(x \\mid 1.0, \\rho)\\big)\n\\end{align*}\\]\nThis can be implemented in Stan using a latent-variable GP formulation. We’ve used \\(\\textsf{LKJCorr}\\) for \\(C(\\phi)\\), but any positive-definite matrix will do.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] real x;\n  matrix[N, D] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  vector&lt;lower=0&gt;[D] alpha;\n  real&lt;lower=0&gt; sigma;\n  cholesky_factor_corr[D] L_Omega;\n  matrix[N, D] eta;\n}\nmodel {\n  matrix[N, D] f;\n  {\n    matrix[N, N] K = gp_exp_quad_cov(x, 1.0, rho);\n    matrix[N, N] L_K;\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta\n        * diag_pre_multiply(alpha, L_Omega)';\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(3);\n  to_vector(eta) ~ std_normal();\n\n  to_vector(y) ~ normal(to_vector(f), sigma);\n}\ngenerated quantities {\n  matrix[D, D] Omega;\n  Omega = L_Omega * L_Omega';\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Gaussian Processes"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/gaussian-processes.html#footnotes",
+    "href": "stan-users-guide/gaussian-processes.html#footnotes",
+    "title": "Gaussian Processes",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nGaussian processes can be extended to covariance functions producing positive semi-definite matrices, but Stan does not support inference in the resulting models because the resulting distribution does not have unconstrained support.↩︎\nA boundary-avoiding prior is just one where the limit of the density is zero at the boundary, the result of which is estimates that are pushed away from the boundary.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Gaussian Processes"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html",
+    "href": "stan-users-guide/floating-point.html",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "Computers approximate real values in \\(\\mathbb{R}\\) using a fixed number of bits. This chapter explains how this is done and why it is important for writing robust Stan (and other numerical) programs. The subfield of computer science devoted to studying how real arithmetic works on computers is called numerical analysis.\n\n\nStan’s arithmetic is implemented using double-precision arithmetic. The behavior of most1 modern computers follows the floating-point arithmetic, IEEE Standard for Floating-Point Arithmetic (IEEE 754).\n\n\nThe double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:\n\nsignificand (53 bits): base value representing significant digits\nexponent (11 bits): power of two multiplied by the base\n\nThe value of a finite floating point number is\n\\[\nv = (-1)^s \\times c \\, 2^q\n\\]\n\n\n\nA normal floating-point value does not use any leading zeros in its significand; subnormal numbers may use leading zeros. Not all I/O systems support subnormal numbers.\n\n\n\nThere are some reserved exponent values so that legal exponent values range between\\(-(2^{10}) + 2 = -1022\\) and \\(2^{10} - 1 = 1023\\). Legal significand values are between \\(-2^{52}\\) and \\(2^{52} - 1\\). Floating point allows the representation of both really big and really small values. Some extreme values are\n\nlargest normal finite number: \\(\\approx 1.8 \\times 10^{308}\\)\nlargest subnormal finite number: \\(\\approx 2.2 \\times 10^{308}\\)\nsmallest positive normal number: \\(\\approx 2.2 \\times 10^{-308}\\)\nsmallest positive subnormal number: \\(\\approx 4.9 \\times 10^{-324}\\)\n\n\n\n\nBecause of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., 0 == -0 evaluates to true).\n\n\n\nA specially chosen bit pattern is used for the not-a-number value (often written as NaN in programming language output, including Stan’s).\nStan provides a value function not_a_number() that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.\nStan also provides a test function is_nan(x) that returns 1 if x is not-a-number and 0 otherwise.\nNot-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as log and exp have the same behavior, propagating not-a-number values.\nMost of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.\nComparisons with not-a-number always return false, up to and including comparison with itself. That is, not_a_number() == not_a_number() somewhat confusingly returns false. That is why there is a built-in is_nan() function in Stan (and in C++). The only exception is negation, which remains coherent. This means not_a_number() != not_a_number() returns true.\nUndefined operations often return not-a-number values. For example, sqrt(-1) will evaluate to not-a-number.\n\n\n\nThere are also two special values representing positive infinity (\\(\\infty)\\) and negative infinity (\\(-\\infty\\)). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, log(0) evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that 0 == exp(log(0)). Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.\nThere are value functions positive_infinity() and negative_infinity() as well as a test function is_inf().\nPositive and negative infinity have the expected comparison behavior, so that negative_infinity() &lt; 0 evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.\nPositive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.\n\n\n\n\nIn programming languages such as Stan, numbers may be represented in standard decimal (base 10) notation. For example, 2.39 or -1567846.276452. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using scientific notation, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, 36.29e-3 represents the number \\(36.29 \\times\n10^{-3}\\), which is the same number as is represented by 0.03629.\n\n\n\nThe choice of significand provides \\(\\log_{10} 2^{53} \\approx 15.95\\) decimal (base 10) digits of arithmetic precision. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.\n\n\nIn practice, the finite amount of arithmetic precision leads to rounding, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,\n1 + 1e-20 == 1\nThe closest floating point number to \\(1 + 10^{-20}\\) turns out to be \\(1\\) itself. By contrast,\n0 + 1e-20 == 1e-20\nThis highlights the fact that precision depends on scale. Even though 1 + 1e-20 == 1, we have 1e-20 + 1e-20 == 2e-20, as expected.\nRounding also manifests itself in a lack of transitivity. In particular, it does not usually hold for three floating point numbers \\(a, b, c\\) that \\((a + b) + c = a + (b + c)\\).\nIn statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix \\(L\\) with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix \\(L \\, L^{\\top}\\). In practice, rounding and loss of precision may render the result \\(L \\, L^{\\top}\\) neither symmetric nor positive definite.\nIn practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding \\(L \\, L^{\\top}\\) with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.\n\n\n\nThe smallest number greater than zero is roughly \\(0 + 10^{-323}\\). The largest number less than one is roughly \\(1 - 10^{-15.95}\\). The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of \\(0.9999999999999999\\).\nFor this reason, the machine precision is said to be roughly \\(10^{-15.95}\\). This constant is available as machine_precision() in Stan.\n\n\n\nSpecial operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation log(1 + x) for positive x. When x is small (less than \\(10^{-16}\\) for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing \\(f(x) = \\log (1 + x)\\). In Stan (as in C++), this operation is written as log1p(x). Because x itself may be close to zero, the function log1p(x) can take the logarithm of values very close to one, the results of which are close to zero.\nSimilarly, the complementary cumulative distribution functions (CCDF), defined by \\(F^{\\complement}_Y(y) = 1 - F_Y(y)\\), where \\(F_Y\\) is the cumulative distribution function (CDF) for the random variable \\(Y\\). This allows values very close to one to be represented in complementary form.\n\n\n\nAnother downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider \\[\\begin{align*}\n  1&.23456789012345 \\\\\n- 1&.23456789012344 \\\\\n= 0&.00000000000001\n\\end{align*}\\] We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.\nCatastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses Welford’s algorithm for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.\n\n\n\nEven though 1e200 may be represented as a double precision floating point value, there is no finite value large enough to represent 1e200 * 1e200. The result of 1e200 * 1e200 is said to overflow. The IEEE 754 standard requires the result to be positive infinity.\nOverflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.\n\n\n\nWhen there is no number small enough to represent a result, it is said to underflow. For instance, 1e-200 may be represented, but 1e-200 * 1e-200 underflows so that the result is zero.\nUnderflow is a ubiquitous problem in likelihood calculations, For example, if \\(p(y_n \\mid \\theta) &lt; 0.1\\), then \\[\np(y \\mid \\theta) = \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\] will underflow as soon as \\(N &gt; 350\\) or so.\nTo deal with underflow, work on the log scale. Even though \\(p(y \\mid\n\\theta)\\) can’t be represented, there is no problem representing \\[\n\\begin{array}{rcl}\n\\log p(y \\mid \\theta)\n& = & \\log \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\\\[4pt]\n& = & \\sum_{n = 1}^N \\log p(y_n \\mid \\theta)\n\\end{array}\n\\]\nThis is why all of Stan’s probability functions operate on the log scale.\n\n\n\n\nWorking on the log scale, multiplication is converted to addition, \\[\n\\log (a \\cdot b) = \\log a + \\log b.\n\\] Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given \\(\\log a\\) and \\(\\log b\\), how do we get \\(\\log (a + b)\\)? Working out the algebra, \\[\n\\log (a + b)\n=\n\\log (\\exp(\\log a) + \\exp(\\log b)).\n\\]\n\n\nThe nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\log (\\exp(u) + \\exp(v)).\n\\] so that \\[\n\\log (a + b)\n=\n\\textrm{log-sum-exp}(\\log a, \\log b).\n\\]\nAlthough it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\max(u, v) + \\log\\big( \\exp(u - \\max(u, v)) + \\exp(v - \\max(u, v)) \\big).\n\\]\nBecause the terms inside the exponentiations are \\(u - \\max(u, v)\\) and \\(v - \\max(u, v)\\), one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that \\(u \\geq v\\), so that \\[\n\\textrm{log-sum-exp}(u, v) = u + \\log\\big(1 + \\exp(v - u)\\big).\n\\]\nAlthough the inner term may itself be evaluated using the built-in function log1p, there is only limited gain because \\(\\exp(v - u)\\) is only near zero when \\(u\\) is much larger than \\(v\\), meaning the final result is likely to round to \\(u\\) anyway.\nTo conclude, when evaluating \\(\\log (a + b)\\) given \\(\\log a\\) and \\(\\log\nb\\), and assuming \\(\\log a &gt; \\log b\\), return\n\\[\n\\log (a + b) =\n\\log a + \\textrm{log1p}\\big(\\exp(\\log b - \\log a)\\big).\n\\]\n\n\n\nThe log sum of exponentials function may be generalized to sequences in the obvious way, so that if \\(v = v_1, \\ldots, v_N\\), then \\[\\begin{eqnarray*}\n\\textrm{log-sum-exp}(v)\n& = & \\log \\sum_{n = 1}^N \\exp(v_n)\n\\\\[4pt]\n& = & \\max(v) + \\log \\sum_{n = 1}^N \\exp(v_n - \\max(v)).\n\\end{eqnarray*}\\] The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate \\(\\log (u_1 + \\cdots + u_N)\\) given only \\(\\log u_n\\).\n\n\n\nAn immediate application is to computing the mean of a vector \\(u\\) entirely on the log scale. That is, given \\(\\log u\\) and returning \\(\\log \\textrm{mean}(u)\\). \\[\\begin{eqnarray*}\n\\log \\left( \\frac{1}{N} \\sum_{n = 1}^N u_n \\right)\n& = & \\log \\frac{1}{N} + \\log \\sum_{n = 1}^N \\exp(\\log u_n)\n\\\\[4pt]\n& = & -\\log N + \\textrm{log-sum-exp}(\\log u).\n\\end{eqnarray*}\\] where \\(\\log u = (\\log u_1, \\ldots, \\log u_N)\\) is understood elementwise.\n\n\n\n\nBecause floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing x == y, an approximate test may be used given an absolute or relative tolerance.\nGiven a positive absolute tolerance of epsilon, x can be compared to y using the conditional\nabs(x - y) &lt;= epsilon.\nAbsolute tolerances work when the scale of x and y and the relevant comparison is known.\nGiven a positive relative tolerance of epsilon, a typical comparison is\n2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html#floating-point-representations",
+    "href": "stan-users-guide/floating-point.html#floating-point-representations",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "Stan’s arithmetic is implemented using double-precision arithmetic. The behavior of most1 modern computers follows the floating-point arithmetic, IEEE Standard for Floating-Point Arithmetic (IEEE 754).\n\n\nThe double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:\n\nsignificand (53 bits): base value representing significant digits\nexponent (11 bits): power of two multiplied by the base\n\nThe value of a finite floating point number is\n\\[\nv = (-1)^s \\times c \\, 2^q\n\\]\n\n\n\nA normal floating-point value does not use any leading zeros in its significand; subnormal numbers may use leading zeros. Not all I/O systems support subnormal numbers.\n\n\n\nThere are some reserved exponent values so that legal exponent values range between\\(-(2^{10}) + 2 = -1022\\) and \\(2^{10} - 1 = 1023\\). Legal significand values are between \\(-2^{52}\\) and \\(2^{52} - 1\\). Floating point allows the representation of both really big and really small values. Some extreme values are\n\nlargest normal finite number: \\(\\approx 1.8 \\times 10^{308}\\)\nlargest subnormal finite number: \\(\\approx 2.2 \\times 10^{308}\\)\nsmallest positive normal number: \\(\\approx 2.2 \\times 10^{-308}\\)\nsmallest positive subnormal number: \\(\\approx 4.9 \\times 10^{-324}\\)\n\n\n\n\nBecause of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., 0 == -0 evaluates to true).\n\n\n\nA specially chosen bit pattern is used for the not-a-number value (often written as NaN in programming language output, including Stan’s).\nStan provides a value function not_a_number() that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.\nStan also provides a test function is_nan(x) that returns 1 if x is not-a-number and 0 otherwise.\nNot-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as log and exp have the same behavior, propagating not-a-number values.\nMost of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.\nComparisons with not-a-number always return false, up to and including comparison with itself. That is, not_a_number() == not_a_number() somewhat confusingly returns false. That is why there is a built-in is_nan() function in Stan (and in C++). The only exception is negation, which remains coherent. This means not_a_number() != not_a_number() returns true.\nUndefined operations often return not-a-number values. For example, sqrt(-1) will evaluate to not-a-number.\n\n\n\nThere are also two special values representing positive infinity (\\(\\infty)\\) and negative infinity (\\(-\\infty\\)). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, log(0) evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that 0 == exp(log(0)). Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.\nThere are value functions positive_infinity() and negative_infinity() as well as a test function is_inf().\nPositive and negative infinity have the expected comparison behavior, so that negative_infinity() &lt; 0 evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.\nPositive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html#literals-decimal-and-scientific-notation",
+    "href": "stan-users-guide/floating-point.html#literals-decimal-and-scientific-notation",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "In programming languages such as Stan, numbers may be represented in standard decimal (base 10) notation. For example, 2.39 or -1567846.276452. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using scientific notation, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, 36.29e-3 represents the number \\(36.29 \\times\n10^{-3}\\), which is the same number as is represented by 0.03629.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html#arithmetic-precision",
+    "href": "stan-users-guide/floating-point.html#arithmetic-precision",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "The choice of significand provides \\(\\log_{10} 2^{53} \\approx 15.95\\) decimal (base 10) digits of arithmetic precision. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.\n\n\nIn practice, the finite amount of arithmetic precision leads to rounding, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,\n1 + 1e-20 == 1\nThe closest floating point number to \\(1 + 10^{-20}\\) turns out to be \\(1\\) itself. By contrast,\n0 + 1e-20 == 1e-20\nThis highlights the fact that precision depends on scale. Even though 1 + 1e-20 == 1, we have 1e-20 + 1e-20 == 2e-20, as expected.\nRounding also manifests itself in a lack of transitivity. In particular, it does not usually hold for three floating point numbers \\(a, b, c\\) that \\((a + b) + c = a + (b + c)\\).\nIn statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix \\(L\\) with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix \\(L \\, L^{\\top}\\). In practice, rounding and loss of precision may render the result \\(L \\, L^{\\top}\\) neither symmetric nor positive definite.\nIn practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding \\(L \\, L^{\\top}\\) with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.\n\n\n\nThe smallest number greater than zero is roughly \\(0 + 10^{-323}\\). The largest number less than one is roughly \\(1 - 10^{-15.95}\\). The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of \\(0.9999999999999999\\).\nFor this reason, the machine precision is said to be roughly \\(10^{-15.95}\\). This constant is available as machine_precision() in Stan.\n\n\n\nSpecial operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation log(1 + x) for positive x. When x is small (less than \\(10^{-16}\\) for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing \\(f(x) = \\log (1 + x)\\). In Stan (as in C++), this operation is written as log1p(x). Because x itself may be close to zero, the function log1p(x) can take the logarithm of values very close to one, the results of which are close to zero.\nSimilarly, the complementary cumulative distribution functions (CCDF), defined by \\(F^{\\complement}_Y(y) = 1 - F_Y(y)\\), where \\(F_Y\\) is the cumulative distribution function (CDF) for the random variable \\(Y\\). This allows values very close to one to be represented in complementary form.\n\n\n\nAnother downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider \\[\\begin{align*}\n  1&.23456789012345 \\\\\n- 1&.23456789012344 \\\\\n= 0&.00000000000001\n\\end{align*}\\] We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.\nCatastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses Welford’s algorithm for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.\n\n\n\nEven though 1e200 may be represented as a double precision floating point value, there is no finite value large enough to represent 1e200 * 1e200. The result of 1e200 * 1e200 is said to overflow. The IEEE 754 standard requires the result to be positive infinity.\nOverflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.\n\n\n\nWhen there is no number small enough to represent a result, it is said to underflow. For instance, 1e-200 may be represented, but 1e-200 * 1e-200 underflows so that the result is zero.\nUnderflow is a ubiquitous problem in likelihood calculations, For example, if \\(p(y_n \\mid \\theta) &lt; 0.1\\), then \\[\np(y \\mid \\theta) = \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\] will underflow as soon as \\(N &gt; 350\\) or so.\nTo deal with underflow, work on the log scale. Even though \\(p(y \\mid\n\\theta)\\) can’t be represented, there is no problem representing \\[\n\\begin{array}{rcl}\n\\log p(y \\mid \\theta)\n& = & \\log \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\\\[4pt]\n& = & \\sum_{n = 1}^N \\log p(y_n \\mid \\theta)\n\\end{array}\n\\]\nThis is why all of Stan’s probability functions operate on the log scale.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html#log-sum-of-exponentials",
+    "href": "stan-users-guide/floating-point.html#log-sum-of-exponentials",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "Working on the log scale, multiplication is converted to addition, \\[\n\\log (a \\cdot b) = \\log a + \\log b.\n\\] Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given \\(\\log a\\) and \\(\\log b\\), how do we get \\(\\log (a + b)\\)? Working out the algebra, \\[\n\\log (a + b)\n=\n\\log (\\exp(\\log a) + \\exp(\\log b)).\n\\]\n\n\nThe nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\log (\\exp(u) + \\exp(v)).\n\\] so that \\[\n\\log (a + b)\n=\n\\textrm{log-sum-exp}(\\log a, \\log b).\n\\]\nAlthough it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\max(u, v) + \\log\\big( \\exp(u - \\max(u, v)) + \\exp(v - \\max(u, v)) \\big).\n\\]\nBecause the terms inside the exponentiations are \\(u - \\max(u, v)\\) and \\(v - \\max(u, v)\\), one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that \\(u \\geq v\\), so that \\[\n\\textrm{log-sum-exp}(u, v) = u + \\log\\big(1 + \\exp(v - u)\\big).\n\\]\nAlthough the inner term may itself be evaluated using the built-in function log1p, there is only limited gain because \\(\\exp(v - u)\\) is only near zero when \\(u\\) is much larger than \\(v\\), meaning the final result is likely to round to \\(u\\) anyway.\nTo conclude, when evaluating \\(\\log (a + b)\\) given \\(\\log a\\) and \\(\\log\nb\\), and assuming \\(\\log a &gt; \\log b\\), return\n\\[\n\\log (a + b) =\n\\log a + \\textrm{log1p}\\big(\\exp(\\log b - \\log a)\\big).\n\\]\n\n\n\nThe log sum of exponentials function may be generalized to sequences in the obvious way, so that if \\(v = v_1, \\ldots, v_N\\), then \\[\\begin{eqnarray*}\n\\textrm{log-sum-exp}(v)\n& = & \\log \\sum_{n = 1}^N \\exp(v_n)\n\\\\[4pt]\n& = & \\max(v) + \\log \\sum_{n = 1}^N \\exp(v_n - \\max(v)).\n\\end{eqnarray*}\\] The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate \\(\\log (u_1 + \\cdots + u_N)\\) given only \\(\\log u_n\\).\n\n\n\nAn immediate application is to computing the mean of a vector \\(u\\) entirely on the log scale. That is, given \\(\\log u\\) and returning \\(\\log \\textrm{mean}(u)\\). \\[\\begin{eqnarray*}\n\\log \\left( \\frac{1}{N} \\sum_{n = 1}^N u_n \\right)\n& = & \\log \\frac{1}{N} + \\log \\sum_{n = 1}^N \\exp(\\log u_n)\n\\\\[4pt]\n& = & -\\log N + \\textrm{log-sum-exp}(\\log u).\n\\end{eqnarray*}\\] where \\(\\log u = (\\log u_1, \\ldots, \\log u_N)\\) is understood elementwise.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html#comparing-floating-point-numbers",
+    "href": "stan-users-guide/floating-point.html#comparing-floating-point-numbers",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "Because floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing x == y, an approximate test may be used given an absolute or relative tolerance.\nGiven a positive absolute tolerance of epsilon, x can be compared to y using the conditional\nabs(x - y) &lt;= epsilon.\nAbsolute tolerances work when the scale of x and y and the relevant comparison is known.\nGiven a positive relative tolerance of epsilon, a typical comparison is\n2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/floating-point.html#footnotes",
+    "href": "stan-users-guide/floating-point.html#footnotes",
+    "title": "Floating Point Arithmetic",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe notable exception is Intel’s optimizing compilers under certain optimization settings.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Floating Point Arithmetic"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html",
+    "href": "stan-users-guide/efficiency-tuning.html",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "This chapter provides a grab bag of techniques for optimizing Stan code, including vectorization, sufficient statistics, and conjugacy. At a coarse level, efficiency involves both the amount of time required for a computation and the amount of memory required. For practical applied statistical modeling, we are mainly concerned with reducing wall time (how long a program takes as measured by a clock on the wall) and keeping memory requirements within available bounds.\n\n\nThe standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).\n\n\n\nStan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.\nWith probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.\nA well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.\n\n\n\nThere is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.\nStatistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.\n\n\n\nBecause Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.\nA second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density \\(\\log\np(\\theta)\\) with respect to the parameter vector \\(\\theta\\), defined as \\[\nH(\\theta) = \\nabla \\, \\nabla \\, \\log p(\\theta \\mid y),\n\\] so that \\[\nH_{i, j}(\\theta) = \\frac{\\partial^2 \\log p(\\theta \\mid y)}\n                {\\partial \\theta_i \\ \\partial \\theta_j}.\n\\] For pure penalized maximum likelihood problems, the posterior log density \\(\\log p(\\theta \\mid y)\\) is replaced by the penalized likelihood function \\(\\mathcal{L}(\\theta) = \\log p(y \\mid \\theta) - \\lambda(\\theta)\\).\n\n\nA good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix \\(H\\), which is the ratio of the largest to the smallest eigenvalue of \\(H\\) (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.\n\n\n\nIdeally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.\n\n\n\nIn all but very simple models (such as multivariate normals), the Hessian will vary as \\(\\theta\\) varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.\nFor all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.\nWith sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.\n\n\n\nImproving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the change of variables chapter.\nOne has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.\n\n\n\n\nModel misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.\nCounterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as normal(0, 1000)). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as normal(mu,   sigma), where mu and sigma are new parameters with their own hyperpriors, can be beneficial.\n\n\n\nStan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ...\n}                               // first validation\nmodel {\n  y ~ multi_normal(mu, Sigma);  // second validation\n  // ...\n}\nBecause Sigma is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that Sigma is positive definite. This test is expensive, having cubic run time (i.e., \\(\\mathcal{O}(N^3)\\) for \\(N \\times N\\) matrices), so it should not be done twice.\nThe test may be avoided by simply declaring Sigma to be a simple unconstrained matrix.\ntransformed parameters {\n  matrix[K, K] Sigma;\n  // ...\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);  // only validation\n}\nNow the only validation is carried out by the multivariate normal.\n\n\n\nStan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.\n\n\nIn this section, we discuss a general transform from a centered to a non-centered parameterization (Papaspiliopoulos, Roberts, and Sköld 2007).1\nThis reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.\nNeal (2003) defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for \\(y \\in\n\\mathbb{R}\\) and \\(x \\in \\mathbb{R}^9\\) with density\n\\[\np(y,x) = \\textsf{normal}(y \\mid 0,3) \\times \\prod_{n=1}^9\n\\textsf{normal}(x_n \\mid 0,\\exp(y/2)).\n\\]\nThe probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to \\(y\\). A plot of the log marginal density of \\(y\\) and the first dimension \\(x_1\\) is shown in the following plot.\nThe funnel can be implemented directly in Stan as follows.\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y/2));\n}\nWhen the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where \\(y\\) is small and thus \\(x\\) is constrained to be near 0. This is due to the fact that the density’s scale changes with \\(y\\), so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.\n\n\n\n\n\n\n\n\n\n\n\n\n\nFigure 1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable \\(y\\) and one lower-level variable \\(x_1\\) (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of \\(x_1\\) and \\(y\\) values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density \\(p(y) = \\textsf{normal}(y \\mid 0,3)\\), which has mean 0 and standard deviation 3.\n\n\n\nIn this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y/2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y/2))\n}\nIn this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.\n\n\n\nSampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.\nThe practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.\nSuppose a random variable of interest \\(X\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), so that \\(X \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\). The variable \\(X\\) has a cumulative distribution function \\(F_X:\\mathbb{R} \\rightarrow (0,1)\\) defined by \\[\nF_X(x) = \\frac{1}{\\pi} \\arctan \\left( \\frac{x - \\mu}{\\tau} \\right) +\n\\frac{1}{2}.\n\\] The inverse of the cumulative distribution function, \\(F_X^{-1}:(0,1) \\rightarrow \\mathbb{R}\\), is thus\n\\[\nF^{-1}_X(y) = \\mu + \\tau \\tan \\left( \\pi \\left( y - \\frac{1}{2} \\right) \\right).\n\\] Thus if the random variable \\(Y\\) has a unit uniform distribution, \\(Y\n\\sim \\textsf{uniform}(0,1)\\), then \\(F^{-1}_X(Y)\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), i.e., \\(F^{-1}_X(Y) \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\).\nConsider a Stan program involving a Cauchy-distributed parameter beta.\nparameters {\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ cauchy(mu, tau);\n  // ...\n}\nThis declaration of beta as a parameter may be replaced with a transformed parameter beta defined in terms of a uniform-distributed parameter beta_unif.\nparameters {\n  real&lt;lower=-pi() / 2, upper=pi() / 2&gt; beta_unif;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = mu + tau * tan(beta_unif);  // beta ~ cauchy(mu, tau)\n}\nmodel {\n  beta_unif ~ uniform(-pi() / 2, pi() / 2);  // not necessary\n  // ...\n}\nIt is more convenient in Stan to transform a uniform variable on \\((-\\pi/2, \\pi/2)\\) than one on \\((0,1)\\). The Cauchy location and scale parameters, mu and tau, may be defined as data or may themselves be parameters. The variable beta could also be defined as a local variable if it does not need to be included in the sampler’s output.\nThe uniform distribution on beta_unif is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because \\(\\log\n\\textsf{uniform}(\\beta_{\\textsf{unif}} \\mid -\\pi/2,\\pi/2) =\n-\\log \\pi\\) is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that beta_unif is between -pi() / 2 and pi() / 2, but this condition is guaranteed by the constraints in the declaration of beta_unif.\n\n\n\nOne thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable \\(\\beta\\), \\[\n\\beta \\sim \\textsf{Student-t}(\\nu, 0, 1),\n\\] by first generating a gamma-distributed precision (inverse variance) \\(\\tau\\) according to \\[\n\\tau \\sim \\textsf{Gamma}(\\nu/2, \\nu/2),\n\\] and then generating \\(\\beta\\) from the normal distribution, \\[\n\\beta \\sim \\textsf{normal}\\left(0,\\tau^{-\\frac{1}{2}}\\right).\n\\]\nBecause \\(\\tau\\) is precision, \\(\\tau^{-\\frac{1}{2}}\\) is the scale (standard deviation), which is the parameterization used by Stan.\nThe marginal distribution of \\(\\beta\\) when you integrate out \\(\\tau\\) is \\(\\textsf{Student-t}(\\nu, 0, 1)\\), i.e., \\[\n\\textsf{Student-t}(\\beta \\mid \\nu, 0, 1)\n=\n\\int_0^{\\infty}\n\\,\n\\textsf{normal}\\left(\\beta \\middle| 0, \\tau^{-0.5}\\right)\n\\times\n\\textsf{Gamma}\\left(\\tau \\middle| \\nu/2, \\nu/2\\right)\n\\\n\\text{d} \\tau.\n\\]\nTo go one step further, instead of defining a \\(\\beta\\) drawn from a normal with precision \\(\\tau\\), define \\(\\alpha\\) to be drawn from a unit normal, \\[\n\\alpha \\sim \\textsf{normal}(0,1)\n\\] and rescale by defining \\[\n\\beta = \\alpha \\, \\tau^{-\\frac{1}{2}}.\n\\]\nNow suppose \\(\\mu = \\beta x\\) is the product of \\(\\beta\\) with a regression predictor \\(x\\). Then the reparameterization \\(\\mu = \\alpha\n\\tau^{-\\frac{1}{2}} x\\) has the same distribution, but in the original, direct parameterization, \\(\\beta\\) has (potentially) heavy tails, whereas in the second, neither \\(\\tau\\) nor \\(\\alpha\\) have heavy tails.\nTo translate into Stan notation, this reparameterization replaces\nparameters {\n  real&lt;lower=0&gt; nu;\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ student_t(nu, 0, 1);\n  // ...\n}\nwith\nparameters {\n  real&lt;lower=0&gt; nu;\n  real&lt;lower=0&gt; tau;\n  real alpha;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = alpha / sqrt(tau);\n  // ...\n}\nmodel {\n  real half_nu;\n  half_nu = 0.5 * nu;\n  tau ~ gamma(half_nu, half_nu);\n  alpha ~ std_normal();\n  // ...\n}\nAlthough set to 0 here, in most cases, the lower bound for the degrees of freedom parameter nu can be set to 1 or higher; when nu is 1, the result is a Cauchy distribution with fat tails and as nu approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter nu characterizes the heaviness of the tails of the model.\n\n\n\nUnfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.\n\n\nFor example, a vectorized hierarchical model might draw a vector of coefficients \\(\\beta\\) with definitions as follows. The so-called centered parameterization is as follows.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nAlthough not shown, a full model will have priors on both mu_beta and sigma_beta along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix x and binary outcome vector y would include a likelihood statement such as form y ~ bernoulli_logit(x * beta), leading to an analytically intractable posterior.\nA hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of beta, mu_beta and sigma_beta are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See Betancourt and Girolami (2013) for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.\n\n\n\n\nSometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see Betancourt and Girolami (2013)), and in more extreme cases will be necessary to achieve convergence.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta_raw;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // implies: beta ~ normal(mu_beta, sigma_beta)\n  beta = mu_beta + sigma_beta * beta_raw;\n}\nmodel {\n  beta_raw ~ std_normal();\n  // ...\n}\nAny priors defined for mu_beta and sigma_beta remain as defined in the original model.\nAlternatively, Stan’s affine transform can be used to decouple sigma and beta:\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;offset=mu_beta, multiplier=sigma_beta&gt;[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nReparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let \\(\\beta = l + s\\alpha\\) where \\(l\\) is a location parameter and \\(s\\) is a scale parameter. The parameter \\(l\\) need not be the mean, \\(s\\) need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If \\(\\alpha\\) and \\(\\beta\\) are from the same distributional family but \\(\\alpha\\) has location zero and unit scale, while \\(\\beta\\) has location \\(l\\) and scale \\(s\\), then that distribution is a location-scale distribution. Thus, if \\(\\alpha\\) were a parameter and \\(\\beta\\) were a transformed parameter, then a prior distribution from the location-scale family on \\(\\alpha\\) with location zero and unit scale implies a prior distribution on \\(\\beta\\) with location \\(l\\) and scale \\(s\\). Doing so would reduce the dependence between \\(\\alpha\\), \\(l\\), and \\(s\\).\nThere are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if \\(\\alpha\\) is distributed standard normal, then \\(\\beta\\) is distributed normal with mean \\(\\mu = l\\) and standard deviation \\(\\sigma = s\\). The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.\nAlso, if \\(z\\) is distributed standard normal, then \\(z^2\\) is distributed chi-squared with one degree of freedom. By summing the squares of \\(K\\) independent standard normal variates, one can obtain a single variate that is distributed chi-squared with \\(K\\) degrees of freedom. However, for large \\(K\\), the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying \\(K\\) primitive parameters just to obtain one transformed parameter to use in a model.\n\n\n\nThe benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for \\(\\beta\\) to be multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). Such a belief is reflected by the following code.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\nparameters {\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ multi_normal(mu, Sigma);\n  // ...\n}\nIn this case mu and Sigma are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of beta.\nIf \\(\\alpha\\) has the same dimensions as \\(\\beta\\) but the elements of \\(\\alpha\\) are independently and identically distributed standard normal such that \\(\\beta = \\mu + L\\alpha\\), where \\(LL^\\top = \\Sigma\\), then \\(\\beta\\) is distributed multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). One choice for \\(L\\) is the Cholesky factor of \\(\\Sigma\\). Thus, the model above could be reparameterized as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Sigma);\n}\nparameters {\n  vector[K] alpha;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  beta = mu + L * alpha;\n}\nmodel {\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu, Sigma)\n  // ...\n}\nThis reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of alpha and second, it avoids the need to invert Sigma every time multi_normal is evaluated.\nThe Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix Tau were fixed data but the vector of standard deviations sigma were unknown parameters, then a reparameterization of beta in terms of alpha could be implemented as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  corr_matrix[K] Tau;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Tau);\n}\nparameters {\n  vector[K] alpha;\n  vector&lt;lower=0&gt;[K] sigma;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // This equals mu + diag_matrix(sigma) * L * alpha;\n  beta = mu + sigma .* (L * alpha);\n}\nmodel {\n  sigma ~ cauchy(0, 5);\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu,\n  //  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))\n  // ...\n}\nThis reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.\nA Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let \\(L\\) be the Cholesky factor of a \\(K \\times K\\) positive definite scale matrix \\(S\\) and let \\(\\nu\\) be the degrees of freedom. If \\[\nA = \\begin{pmatrix}\n\\sqrt{c_{1}} & 0            & \\cdots                & 0 \\\\\nz_{21}       & \\sqrt{c_{2}} & \\ddots                & \\vdots \\\\\n\\vdots       & \\ddots       & \\ddots                & 0 \\\\\nz_{K1}       & \\cdots       & z_{K\\left(K-1\\right)} & \\sqrt{c_{K}}\n\\end{pmatrix},\n\\] where each \\(c_i\\) is distributed chi-squared with \\(\\nu - i + 1\\) degrees of freedom and each \\(z_{ij}\\) is distributed standard normal, then \\(W = LAA^{\\top}L^{\\top}\\) is distributed Wishart with scale matrix \\(S = LL^{\\top}\\) and degrees of freedom \\(\\nu\\). Such a reparameterization can be implemented by the following Stan code:\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  vector[K] mu;\n  matrix[N, K] y;\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  int count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal();\n  // implies: L * A * A' * L' ~ wishart(nu, L * L')\n  y ~ multi_normal_cholesky(mu, L * A);\n  // ...\n}\nThis reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of z and second, it avoids the need to invert the covariance matrix, \\(W\\) every time wishart is evaluated. Third, if \\(W\\) is to be used with a multivariate normal distribution, you can pass \\(L A\\) to the more efficient multi_normal_cholesky function, rather than passing \\(W\\) to multi_normal.\nIf \\(W\\) is distributed Wishart with scale matrix \\(S\\) and degrees of freedom \\(\\nu\\), then \\(W^{-1}\\) is distributed inverse Wishart with inverse scale matrix \\(S^{-1}\\) and degrees of freedom \\(\\nu\\). Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since \\(W = L A A^{\\top} L^{\\top}\\), \\(W^{-1} = L^{{\\top}^{-1}} A^{{\\top}^{-1}} A^{-1} L^{-1}\\), where all four inverses exist, but \\(L^{{-1}^{\\top}} = L^{{\\top}^{-1}}\\) and \\(A^{{-1}^{\\top}} = A^{{\\top}^{-1}}\\). We can slightly modify the above Stan code for this case:\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  // ...\n}\ntransformed data {\n  matrix[K, K] eye;\n  matrix[K, K] L_inv;\n  for (j in 1:K) {\n    for (i in 1:K) {\n      eye[i, j] = 0.0;\n    }\n    eye[j, j] = 1.0;\n  }\n  L_inv = mdivide_left_tri_low(L, eye);\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  matrix[K, K] A_inv_L_inv;\n  int count;\n  count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal(); // implies: crossprod(A_inv_L_inv) ~\n  // inv_wishart(nu, L_inv' * L_inv)\n  // ...\n}\nAnother candidate for reparameterization is the Dirichlet distribution with all \\(K\\) shape parameters equal. Zyczkowski and Sommers (2001) shows that if \\(\\theta_i\\) is equal to the sum of \\(\\beta\\) independent squared standard normal variates and \\(\\rho_i = \\frac{\\theta_i}{\\sum \\theta_i}\\), then the \\(K\\)-vector \\(\\rho\\) is distributed Dirichlet with all shape parameters equal to \\(\\frac{\\beta}{2}\\). In particular, if \\(\\beta = 2\\), then \\(\\rho\\) is uniformly distributed on the unit simplex. Thus, we can make \\(\\rho\\) be a transformed parameter to reduce dependence, as in:\ndata {\n  int&lt;lower=1&gt; beta;\n  // ...\n}\nparameters {\n  array[K] vector[beta] z;\n  // ...\n}\ntransformed parameters {\n  simplex[K] rho;\n  for (k in 1:K) {\n    rho[k] = dot_self(z[k]); // sum-of-squares\n  }\n  rho = rho / sum(rho);\n}\nmodel {\n  for (k in 1:K) {\n    z[k] ~ std_normal();\n  }\n  // implies: rho ~ dirichlet(0.5 * beta * ones)\n  // ...\n}\n\n\n\n\n\n\nStan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.2 This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.\n\n\n\nBecause of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the sum() operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where foo() is some operation that depends on n.\nfor (n in 1:N) {\n  total += foo(n,...);\n}\nThis code has to create intermediate representations for each of the N summands.\nA faster alternative is to copy the values into a vector, then apply the sum() operator, as in the following refactoring.\n{\n  vector[N] summands;\n  for (n in 1:N) {\n    summands[n] = foo(n,...);\n  }\n  total = sum(summands);\n}\nSyntactically, the replacement is a statement block delineated by curly brackets ({, }), starting with the definition of the local variable summands.\nEven though it involves extra work to allocate the summands vector and copy N values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing total within the loop.\n\n\n\nThe following program directly encodes a linear regression with fixed unit noise using a two-dimensional array x of predictors, an array y of outcomes, and an array beta of regression coefficients.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[K, N] real x;\n  array[N] real y;\n}\nparameters {\n  array[K] real beta;\n}\nmodel {\n  for (n in 1:N) {\n    real gamma = 0;\n    for (k in 1:K) {\n      gamma += x[n, k] * beta[k];\n    }\n    y[n] ~ normal(gamma, 1);\n  }\n}\nThe following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[N] vector[K] x;\n  array[N] real y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(dot_product(x[n], beta), 1);\n  }\n}\nAlthough it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in\ny[n] ~ normal(x[n]' * beta, 1);\nThe relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.3\nThe inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.\n// ...\ntransformed parameters {\n  row_vector[K] beta_t;\n  beta_t = beta';\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(beta_t * x[n], 1);\n  }\n}\nThe problem with transposition could be completely solved by directly encoding the x as a row vector, as in the following example.\ndata {\n  // ...\n  array[N] row_vector[K] x;\n  // ...\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, 1);\n  }\n}\nDeclaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.\nHaving said all this, the most efficient way to code this model is with direct matrix multiplication, as in\ndata {\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nIn general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.\n\n\n\nThe final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nThe variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication x * beta in the model block is a vector of the same length as y.\nThe probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, y is a vector of size N, x * beta is a vector of size N, and 1 is a scalar.\n\n\n\nSometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).\nJohn Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu[ii[n]], sigma);\n}\nThe brute force vectorization would build up a mean vector and then vectorize all at once.\n{\n  vector[N] mu_ii;\n  for (n in 1:N) {\n    mu_ii[n] = mu[ii[n]];\n  }\n  y ~ normal(mu_ii, sigma);\n}\nIf there aren’t many levels (values ii[n] can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector y, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have\ny_1 ~ normal(mu[1], sigma);\n// ...\ny_5 ~ normal(mu[5], sigma);\nThis way, both the mu and sigma parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.\n\n\n\n\nIn some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.\n\n\nConsider the following Bernoulli sampling model.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(alpha, beta);\n  for (n in 1:N) {\n    y[n] ~ bernoulli(theta);\n  }\n}\nIn this model, the sum of positive outcomes in y is a sufficient statistic for the chance of success theta. The model may be recoded using the binomial distribution as follows.\ntheta ~ beta(alpha, beta);\nsum(y) ~ binomial(N, theta);\nBecause truth is represented as one and falsehood as zero, the sum sum(y) of a binary vector y is equal to the number of positive outcomes out of a total of N trials.\nThis can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only \\(K\\) possible discrete outcomes, \\(z_1, \\dotsc, z_K\\), but there are \\(N\\) observations, where \\(N\\) is much larger than \\(K\\). If \\(f_k\\) is the frequency of outcome \\(z_k\\), then the entire likelihood with distribution foo can be coded as follows.\nfor (k in 1:K) {\n  target += f[k] * foo_lpmf(z[k] | ...);\n}\nwhere the ellipses are the parameters of the log probability mass function for distribution foo (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).\nThe resulting program looks like a “weighted” regression, but here the weights f[k] are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies f[k] should be counted once in the transformed data block and stored.\nThe same trick works for combining multiple binomial observations.\n\n\n\nConsider the following Stan model for fitting a normal distribution to data.\ndata {\n  int N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nWith the vectorized form used for y, Stan is clever enough to only evaluate log(sigma) once, but it still has to evaluate the normal for all of y[1] to y[N], which involves adding up all the squared differences from the mean and then dividing by sigma squared.\nAn equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.\ndata {\n  int N;\n  vector[N] y;\n}\ntransformed data {\n  real mean_y = mean(y);\n  real&lt;lower=0&gt; var_y = variance(y);\n  real nm1_over2 = 0.5 * (N - 1);\n  real sqrt_N = sqrt(N);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mean_y ~ normal(mu, sigma / sqrt_N);\n  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^2);\n}\nThe data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.\nThe original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on mu and sigma are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.\nThis transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.\n\n\n\nThe Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace\ny ~ poisson(lambda);\nwith\nsum(y) ~ poisson(size(y) * lambda);\nThis will work even if y is a parameter vector because no Jacobian adjustment is required for summation.\n\n\n\n\nIf an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using exp(theta) in multiple places, declare a local variable to store its value and reuse the local variable.\nAnother case that may not be so obvious is with two multilevel parameters, say a[ii[n]] + b[jj[n]]. If a and b are small (i.e., do not have many levels), then a table a_b of their sums can be created, with\nmatrix[size(a), size(b)] a_b;\nfor (i in 1:size(a)) {\n  for (j in 1:size(b)) {\n    a_b[i, j] = a[i] + b[j];\n  }\n}\nThen the sum can be replaced with a_b[ii[n], jj[n]].\n\n\n\nContinuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.\ntheta ~ beta(alpha + sum(y), beta + N - sum(y));\nTo make the model even more efficient, a transformed data variable defined to be sum(y) could be used in the place of sum(y).\n\n\n\nStandardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:\n\nIt helps in faster convergence of MCMC chains.\nIt makes the model less sensitive to the specifics of the parameterization.\nIt aids in the interpretation and comparison of the importance of coefficients across different predictors.\n\nWhen there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.\nSuppose that \\(y = (y_1,\\dotsc,y_N)\\) is a vector of \\(N\\) outcomes and \\(x = (x_1,\\dotsc,x_N)\\) the corresponding vector of \\(N\\) predictors. A simple linear regression involving an intercept coefficient \\(\\alpha\\) and slope coefficient \\(\\beta\\) can be expressed as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n,\n\\] where \\[\n\\epsilon_n \\sim \\textsf{normal}(0,\\sigma).\n\\]\nIf \\(x\\) has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values \\(x_n\\). First the elements of \\(x\\) are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.\nThe mean of \\(x\\) is given by:\n\\[\nmean_x = \\frac{1}{N} \\sum_{n=1}^{N} x_n\n\\]\nThe standard deviation of \\(x\\) is calculated as: \\[\nsd_x = {\\left({\\frac{1}{N} \\sum_{n=1}^{N} (x_n - mean_x)^2}\\right)}^{1/2}\n\\]\nWith these, we compute the \\(z\\), the standardized predictors\n\\[\nz_n = \\frac{x_n - mean_x}{sd_x}\n\\]\nwhere \\(z_n\\) is the standardized value corresponding to \\(x_n\\).\nThe inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.\n\\[\nx_n = z_n sd_x + mean_x\n\\]\nStandardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.\nConsider the following initial model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  // priors\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ normal(0, 5);\n  // likelihood\n  y ~ normal(x * beta + alpha, sigma);\n}\nThe data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\ntransformed data {\n  real mean_x = mean(x);\n  real sd_x = sd(x);\n  vector[N] x_std = (x - mean_x) / sd_x;\n}\nparameters {\n  real alpha_std;\n  real beta_std;\n  real&lt;lower=0&gt; sigma_std;\n}\nmodel {\n  alpha_std ~ normal(0, 10);\n  beta_std ~ normal(0, 10);\n  sigma_std ~ normal(0, 5);\n  y ~ normal(x_std * beta_std + alpha_std, sigma_std);\n}\nThe parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data x_std is defined in terms of variables mean_x and sd_x; by declaring these variables in the transformed data block, they will be available in all following blocks, and therefore can be used in the generated quantities block to record the “natural” parameters alpha and beta.\nThe fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.\nThe original regression \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\] has been transformed to a regression on the standardized data variable \\(z\\),\n\\[\ny_n = \\alpha' + \\beta' z_n + \\epsilon_n.\n\\]\nThe likelihood is specified in terms of the standardized parameters. The original slope \\(\\beta\\) is the standardized slope \\(\\beta'\\) scaled by the inverse of the standard deviation of \\(x\\). The original intercept \\(\\alpha\\) is the intercept from the standardized model \\(\\alpha'\\), corrected for the effect of scaling and centering \\(x\\). Thus, the formulas to retrieve \\(\\alpha\\) and \\(\\beta\\) from \\(\\alpha'\\) and \\(\\beta'\\) are:\n\\[\\begin{align*}\n\\beta = \\frac{\\beta'}{\\sigma_x} \\\\\n\\alpha = \\alpha' - \\beta' \\frac{\\mu_x}{\\sigma_x}\n\\end{align*}\\]\nThese recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,\ngenerated quantities {\n  real beta = beta_std / sd_x;\n  real alpha = alpha_std - beta_std * mean_x / sd_x;\n\n}\nWhen there are multiple real-valued predictors, i.e., when K is the number of predictors, x is an \\(N \\times K\\) matrix, and beta ia \\(K\\)-vector of coefficients, then x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. When \\(K \\ll N\\) the QR reparameterization is recommended for linear and generalized linear models unless there is an informative prior on the location of \\(\\beta\\).\n\n\nFor many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use\ny ~ std_normal();\nthan to use\ny ~ normal(0, 1);\nbecause the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.\n\n\n\n\nThe map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the map-reduce chapter for more information on implementing map-reduce operations.\nMap-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#what-is-efficiency",
+    "href": "stan-users-guide/efficiency-tuning.html#what-is-efficiency",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "The standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#efficiency-for-probabilistic-models-and-algorithms",
+    "href": "stan-users-guide/efficiency-tuning.html#efficiency-for-probabilistic-models-and-algorithms",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Stan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.\nWith probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.\nA well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#statistical-vs.-computational-efficiency",
+    "href": "stan-users-guide/efficiency-tuning.html#statistical-vs.-computational-efficiency",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "There is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.\nStatistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature",
+    "href": "stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Because Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.\nA second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density \\(\\log\np(\\theta)\\) with respect to the parameter vector \\(\\theta\\), defined as \\[\nH(\\theta) = \\nabla \\, \\nabla \\, \\log p(\\theta \\mid y),\n\\] so that \\[\nH_{i, j}(\\theta) = \\frac{\\partial^2 \\log p(\\theta \\mid y)}\n                {\\partial \\theta_i \\ \\partial \\theta_j}.\n\\] For pure penalized maximum likelihood problems, the posterior log density \\(\\log p(\\theta \\mid y)\\) is replaced by the penalized likelihood function \\(\\mathcal{L}(\\theta) = \\log p(y \\mid \\theta) - \\lambda(\\theta)\\).\n\n\nA good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix \\(H\\), which is the ratio of the largest to the smallest eigenvalue of \\(H\\) (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.\n\n\n\nIdeally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.\n\n\n\nIn all but very simple models (such as multivariate normals), the Hessian will vary as \\(\\theta\\) varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.\nFor all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.\nWith sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.\n\n\n\nImproving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the change of variables chapter.\nOne has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#well-specified-models",
+    "href": "stan-users-guide/efficiency-tuning.html#well-specified-models",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Model misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.\nCounterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as normal(0, 1000)). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as normal(mu,   sigma), where mu and sigma are new parameters with their own hyperpriors, can be beneficial.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#avoiding-validation",
+    "href": "stan-users-guide/efficiency-tuning.html#avoiding-validation",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Stan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ...\n}                               // first validation\nmodel {\n  y ~ multi_normal(mu, Sigma);  // second validation\n  // ...\n}\nBecause Sigma is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that Sigma is positive definite. This test is expensive, having cubic run time (i.e., \\(\\mathcal{O}(N^3)\\) for \\(N \\times N\\) matrices), so it should not be done twice.\nThe test may be avoided by simply declaring Sigma to be a simple unconstrained matrix.\ntransformed parameters {\n  matrix[K, K] Sigma;\n  // ...\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);  // only validation\n}\nNow the only validation is carried out by the multivariate normal.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#reparameterization.section",
+    "href": "stan-users-guide/efficiency-tuning.html#reparameterization.section",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Stan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.\n\n\nIn this section, we discuss a general transform from a centered to a non-centered parameterization (Papaspiliopoulos, Roberts, and Sköld 2007).1\nThis reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.\nNeal (2003) defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for \\(y \\in\n\\mathbb{R}\\) and \\(x \\in \\mathbb{R}^9\\) with density\n\\[\np(y,x) = \\textsf{normal}(y \\mid 0,3) \\times \\prod_{n=1}^9\n\\textsf{normal}(x_n \\mid 0,\\exp(y/2)).\n\\]\nThe probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to \\(y\\). A plot of the log marginal density of \\(y\\) and the first dimension \\(x_1\\) is shown in the following plot.\nThe funnel can be implemented directly in Stan as follows.\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y/2));\n}\nWhen the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where \\(y\\) is small and thus \\(x\\) is constrained to be near 0. This is due to the fact that the density’s scale changes with \\(y\\), so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.\n\n\n\n\n\n\n\n\n\n\n\n\n\nFigure 1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable \\(y\\) and one lower-level variable \\(x_1\\) (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of \\(x_1\\) and \\(y\\) values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density \\(p(y) = \\textsf{normal}(y \\mid 0,3)\\), which has mean 0 and standard deviation 3.\n\n\n\nIn this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y/2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y/2))\n}\nIn this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.\n\n\n\nSampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.\nThe practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.\nSuppose a random variable of interest \\(X\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), so that \\(X \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\). The variable \\(X\\) has a cumulative distribution function \\(F_X:\\mathbb{R} \\rightarrow (0,1)\\) defined by \\[\nF_X(x) = \\frac{1}{\\pi} \\arctan \\left( \\frac{x - \\mu}{\\tau} \\right) +\n\\frac{1}{2}.\n\\] The inverse of the cumulative distribution function, \\(F_X^{-1}:(0,1) \\rightarrow \\mathbb{R}\\), is thus\n\\[\nF^{-1}_X(y) = \\mu + \\tau \\tan \\left( \\pi \\left( y - \\frac{1}{2} \\right) \\right).\n\\] Thus if the random variable \\(Y\\) has a unit uniform distribution, \\(Y\n\\sim \\textsf{uniform}(0,1)\\), then \\(F^{-1}_X(Y)\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), i.e., \\(F^{-1}_X(Y) \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\).\nConsider a Stan program involving a Cauchy-distributed parameter beta.\nparameters {\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ cauchy(mu, tau);\n  // ...\n}\nThis declaration of beta as a parameter may be replaced with a transformed parameter beta defined in terms of a uniform-distributed parameter beta_unif.\nparameters {\n  real&lt;lower=-pi() / 2, upper=pi() / 2&gt; beta_unif;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = mu + tau * tan(beta_unif);  // beta ~ cauchy(mu, tau)\n}\nmodel {\n  beta_unif ~ uniform(-pi() / 2, pi() / 2);  // not necessary\n  // ...\n}\nIt is more convenient in Stan to transform a uniform variable on \\((-\\pi/2, \\pi/2)\\) than one on \\((0,1)\\). The Cauchy location and scale parameters, mu and tau, may be defined as data or may themselves be parameters. The variable beta could also be defined as a local variable if it does not need to be included in the sampler’s output.\nThe uniform distribution on beta_unif is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because \\(\\log\n\\textsf{uniform}(\\beta_{\\textsf{unif}} \\mid -\\pi/2,\\pi/2) =\n-\\log \\pi\\) is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that beta_unif is between -pi() / 2 and pi() / 2, but this condition is guaranteed by the constraints in the declaration of beta_unif.\n\n\n\nOne thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable \\(\\beta\\), \\[\n\\beta \\sim \\textsf{Student-t}(\\nu, 0, 1),\n\\] by first generating a gamma-distributed precision (inverse variance) \\(\\tau\\) according to \\[\n\\tau \\sim \\textsf{Gamma}(\\nu/2, \\nu/2),\n\\] and then generating \\(\\beta\\) from the normal distribution, \\[\n\\beta \\sim \\textsf{normal}\\left(0,\\tau^{-\\frac{1}{2}}\\right).\n\\]\nBecause \\(\\tau\\) is precision, \\(\\tau^{-\\frac{1}{2}}\\) is the scale (standard deviation), which is the parameterization used by Stan.\nThe marginal distribution of \\(\\beta\\) when you integrate out \\(\\tau\\) is \\(\\textsf{Student-t}(\\nu, 0, 1)\\), i.e., \\[\n\\textsf{Student-t}(\\beta \\mid \\nu, 0, 1)\n=\n\\int_0^{\\infty}\n\\,\n\\textsf{normal}\\left(\\beta \\middle| 0, \\tau^{-0.5}\\right)\n\\times\n\\textsf{Gamma}\\left(\\tau \\middle| \\nu/2, \\nu/2\\right)\n\\\n\\text{d} \\tau.\n\\]\nTo go one step further, instead of defining a \\(\\beta\\) drawn from a normal with precision \\(\\tau\\), define \\(\\alpha\\) to be drawn from a unit normal, \\[\n\\alpha \\sim \\textsf{normal}(0,1)\n\\] and rescale by defining \\[\n\\beta = \\alpha \\, \\tau^{-\\frac{1}{2}}.\n\\]\nNow suppose \\(\\mu = \\beta x\\) is the product of \\(\\beta\\) with a regression predictor \\(x\\). Then the reparameterization \\(\\mu = \\alpha\n\\tau^{-\\frac{1}{2}} x\\) has the same distribution, but in the original, direct parameterization, \\(\\beta\\) has (potentially) heavy tails, whereas in the second, neither \\(\\tau\\) nor \\(\\alpha\\) have heavy tails.\nTo translate into Stan notation, this reparameterization replaces\nparameters {\n  real&lt;lower=0&gt; nu;\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ student_t(nu, 0, 1);\n  // ...\n}\nwith\nparameters {\n  real&lt;lower=0&gt; nu;\n  real&lt;lower=0&gt; tau;\n  real alpha;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = alpha / sqrt(tau);\n  // ...\n}\nmodel {\n  real half_nu;\n  half_nu = 0.5 * nu;\n  tau ~ gamma(half_nu, half_nu);\n  alpha ~ std_normal();\n  // ...\n}\nAlthough set to 0 here, in most cases, the lower bound for the degrees of freedom parameter nu can be set to 1 or higher; when nu is 1, the result is a Cauchy distribution with fat tails and as nu approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter nu characterizes the heaviness of the tails of the model.\n\n\n\nUnfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.\n\n\nFor example, a vectorized hierarchical model might draw a vector of coefficients \\(\\beta\\) with definitions as follows. The so-called centered parameterization is as follows.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nAlthough not shown, a full model will have priors on both mu_beta and sigma_beta along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix x and binary outcome vector y would include a likelihood statement such as form y ~ bernoulli_logit(x * beta), leading to an analytically intractable posterior.\nA hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of beta, mu_beta and sigma_beta are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See Betancourt and Girolami (2013) for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.\n\n\n\n\nSometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see Betancourt and Girolami (2013)), and in more extreme cases will be necessary to achieve convergence.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta_raw;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // implies: beta ~ normal(mu_beta, sigma_beta)\n  beta = mu_beta + sigma_beta * beta_raw;\n}\nmodel {\n  beta_raw ~ std_normal();\n  // ...\n}\nAny priors defined for mu_beta and sigma_beta remain as defined in the original model.\nAlternatively, Stan’s affine transform can be used to decouple sigma and beta:\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;offset=mu_beta, multiplier=sigma_beta&gt;[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nReparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let \\(\\beta = l + s\\alpha\\) where \\(l\\) is a location parameter and \\(s\\) is a scale parameter. The parameter \\(l\\) need not be the mean, \\(s\\) need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If \\(\\alpha\\) and \\(\\beta\\) are from the same distributional family but \\(\\alpha\\) has location zero and unit scale, while \\(\\beta\\) has location \\(l\\) and scale \\(s\\), then that distribution is a location-scale distribution. Thus, if \\(\\alpha\\) were a parameter and \\(\\beta\\) were a transformed parameter, then a prior distribution from the location-scale family on \\(\\alpha\\) with location zero and unit scale implies a prior distribution on \\(\\beta\\) with location \\(l\\) and scale \\(s\\). Doing so would reduce the dependence between \\(\\alpha\\), \\(l\\), and \\(s\\).\nThere are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if \\(\\alpha\\) is distributed standard normal, then \\(\\beta\\) is distributed normal with mean \\(\\mu = l\\) and standard deviation \\(\\sigma = s\\). The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.\nAlso, if \\(z\\) is distributed standard normal, then \\(z^2\\) is distributed chi-squared with one degree of freedom. By summing the squares of \\(K\\) independent standard normal variates, one can obtain a single variate that is distributed chi-squared with \\(K\\) degrees of freedom. However, for large \\(K\\), the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying \\(K\\) primitive parameters just to obtain one transformed parameter to use in a model.\n\n\n\nThe benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for \\(\\beta\\) to be multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). Such a belief is reflected by the following code.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\nparameters {\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ multi_normal(mu, Sigma);\n  // ...\n}\nIn this case mu and Sigma are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of beta.\nIf \\(\\alpha\\) has the same dimensions as \\(\\beta\\) but the elements of \\(\\alpha\\) are independently and identically distributed standard normal such that \\(\\beta = \\mu + L\\alpha\\), where \\(LL^\\top = \\Sigma\\), then \\(\\beta\\) is distributed multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). One choice for \\(L\\) is the Cholesky factor of \\(\\Sigma\\). Thus, the model above could be reparameterized as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Sigma);\n}\nparameters {\n  vector[K] alpha;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  beta = mu + L * alpha;\n}\nmodel {\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu, Sigma)\n  // ...\n}\nThis reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of alpha and second, it avoids the need to invert Sigma every time multi_normal is evaluated.\nThe Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix Tau were fixed data but the vector of standard deviations sigma were unknown parameters, then a reparameterization of beta in terms of alpha could be implemented as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  corr_matrix[K] Tau;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Tau);\n}\nparameters {\n  vector[K] alpha;\n  vector&lt;lower=0&gt;[K] sigma;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // This equals mu + diag_matrix(sigma) * L * alpha;\n  beta = mu + sigma .* (L * alpha);\n}\nmodel {\n  sigma ~ cauchy(0, 5);\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu,\n  //  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))\n  // ...\n}\nThis reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.\nA Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let \\(L\\) be the Cholesky factor of a \\(K \\times K\\) positive definite scale matrix \\(S\\) and let \\(\\nu\\) be the degrees of freedom. If \\[\nA = \\begin{pmatrix}\n\\sqrt{c_{1}} & 0            & \\cdots                & 0 \\\\\nz_{21}       & \\sqrt{c_{2}} & \\ddots                & \\vdots \\\\\n\\vdots       & \\ddots       & \\ddots                & 0 \\\\\nz_{K1}       & \\cdots       & z_{K\\left(K-1\\right)} & \\sqrt{c_{K}}\n\\end{pmatrix},\n\\] where each \\(c_i\\) is distributed chi-squared with \\(\\nu - i + 1\\) degrees of freedom and each \\(z_{ij}\\) is distributed standard normal, then \\(W = LAA^{\\top}L^{\\top}\\) is distributed Wishart with scale matrix \\(S = LL^{\\top}\\) and degrees of freedom \\(\\nu\\). Such a reparameterization can be implemented by the following Stan code:\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  vector[K] mu;\n  matrix[N, K] y;\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  int count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal();\n  // implies: L * A * A' * L' ~ wishart(nu, L * L')\n  y ~ multi_normal_cholesky(mu, L * A);\n  // ...\n}\nThis reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of z and second, it avoids the need to invert the covariance matrix, \\(W\\) every time wishart is evaluated. Third, if \\(W\\) is to be used with a multivariate normal distribution, you can pass \\(L A\\) to the more efficient multi_normal_cholesky function, rather than passing \\(W\\) to multi_normal.\nIf \\(W\\) is distributed Wishart with scale matrix \\(S\\) and degrees of freedom \\(\\nu\\), then \\(W^{-1}\\) is distributed inverse Wishart with inverse scale matrix \\(S^{-1}\\) and degrees of freedom \\(\\nu\\). Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since \\(W = L A A^{\\top} L^{\\top}\\), \\(W^{-1} = L^{{\\top}^{-1}} A^{{\\top}^{-1}} A^{-1} L^{-1}\\), where all four inverses exist, but \\(L^{{-1}^{\\top}} = L^{{\\top}^{-1}}\\) and \\(A^{{-1}^{\\top}} = A^{{\\top}^{-1}}\\). We can slightly modify the above Stan code for this case:\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  // ...\n}\ntransformed data {\n  matrix[K, K] eye;\n  matrix[K, K] L_inv;\n  for (j in 1:K) {\n    for (i in 1:K) {\n      eye[i, j] = 0.0;\n    }\n    eye[j, j] = 1.0;\n  }\n  L_inv = mdivide_left_tri_low(L, eye);\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  matrix[K, K] A_inv_L_inv;\n  int count;\n  count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal(); // implies: crossprod(A_inv_L_inv) ~\n  // inv_wishart(nu, L_inv' * L_inv)\n  // ...\n}\nAnother candidate for reparameterization is the Dirichlet distribution with all \\(K\\) shape parameters equal. Zyczkowski and Sommers (2001) shows that if \\(\\theta_i\\) is equal to the sum of \\(\\beta\\) independent squared standard normal variates and \\(\\rho_i = \\frac{\\theta_i}{\\sum \\theta_i}\\), then the \\(K\\)-vector \\(\\rho\\) is distributed Dirichlet with all shape parameters equal to \\(\\frac{\\beta}{2}\\). In particular, if \\(\\beta = 2\\), then \\(\\rho\\) is uniformly distributed on the unit simplex. Thus, we can make \\(\\rho\\) be a transformed parameter to reduce dependence, as in:\ndata {\n  int&lt;lower=1&gt; beta;\n  // ...\n}\nparameters {\n  array[K] vector[beta] z;\n  // ...\n}\ntransformed parameters {\n  simplex[K] rho;\n  for (k in 1:K) {\n    rho[k] = dot_self(z[k]); // sum-of-squares\n  }\n  rho = rho / sum(rho);\n}\nmodel {\n  for (k in 1:K) {\n    z[k] ~ std_normal();\n  }\n  // implies: rho ~ dirichlet(0.5 * beta * ones)\n  // ...\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#vectorization",
+    "href": "stan-users-guide/efficiency-tuning.html#vectorization",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Stan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.2 This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.\n\n\n\nBecause of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the sum() operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where foo() is some operation that depends on n.\nfor (n in 1:N) {\n  total += foo(n,...);\n}\nThis code has to create intermediate representations for each of the N summands.\nA faster alternative is to copy the values into a vector, then apply the sum() operator, as in the following refactoring.\n{\n  vector[N] summands;\n  for (n in 1:N) {\n    summands[n] = foo(n,...);\n  }\n  total = sum(summands);\n}\nSyntactically, the replacement is a statement block delineated by curly brackets ({, }), starting with the definition of the local variable summands.\nEven though it involves extra work to allocate the summands vector and copy N values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing total within the loop.\n\n\n\nThe following program directly encodes a linear regression with fixed unit noise using a two-dimensional array x of predictors, an array y of outcomes, and an array beta of regression coefficients.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[K, N] real x;\n  array[N] real y;\n}\nparameters {\n  array[K] real beta;\n}\nmodel {\n  for (n in 1:N) {\n    real gamma = 0;\n    for (k in 1:K) {\n      gamma += x[n, k] * beta[k];\n    }\n    y[n] ~ normal(gamma, 1);\n  }\n}\nThe following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[N] vector[K] x;\n  array[N] real y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(dot_product(x[n], beta), 1);\n  }\n}\nAlthough it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in\ny[n] ~ normal(x[n]' * beta, 1);\nThe relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.3\nThe inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.\n// ...\ntransformed parameters {\n  row_vector[K] beta_t;\n  beta_t = beta';\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(beta_t * x[n], 1);\n  }\n}\nThe problem with transposition could be completely solved by directly encoding the x as a row vector, as in the following example.\ndata {\n  // ...\n  array[N] row_vector[K] x;\n  // ...\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, 1);\n  }\n}\nDeclaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.\nHaving said all this, the most efficient way to code this model is with direct matrix multiplication, as in\ndata {\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nIn general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.\n\n\n\nThe final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nThe variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication x * beta in the model block is a vector of the same length as y.\nThe probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, y is a vector of size N, x * beta is a vector of size N, and 1 is a scalar.\n\n\n\nSometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).\nJohn Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu[ii[n]], sigma);\n}\nThe brute force vectorization would build up a mean vector and then vectorize all at once.\n{\n  vector[N] mu_ii;\n  for (n in 1:N) {\n    mu_ii[n] = mu[ii[n]];\n  }\n  y ~ normal(mu_ii, sigma);\n}\nIf there aren’t many levels (values ii[n] can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector y, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have\ny_1 ~ normal(mu[1], sigma);\n// ...\ny_5 ~ normal(mu[5], sigma);\nThis way, both the mu and sigma parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#exploiting-sufficient-statistics",
+    "href": "stan-users-guide/efficiency-tuning.html#exploiting-sufficient-statistics",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "In some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.\n\n\nConsider the following Bernoulli sampling model.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(alpha, beta);\n  for (n in 1:N) {\n    y[n] ~ bernoulli(theta);\n  }\n}\nIn this model, the sum of positive outcomes in y is a sufficient statistic for the chance of success theta. The model may be recoded using the binomial distribution as follows.\ntheta ~ beta(alpha, beta);\nsum(y) ~ binomial(N, theta);\nBecause truth is represented as one and falsehood as zero, the sum sum(y) of a binary vector y is equal to the number of positive outcomes out of a total of N trials.\nThis can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only \\(K\\) possible discrete outcomes, \\(z_1, \\dotsc, z_K\\), but there are \\(N\\) observations, where \\(N\\) is much larger than \\(K\\). If \\(f_k\\) is the frequency of outcome \\(z_k\\), then the entire likelihood with distribution foo can be coded as follows.\nfor (k in 1:K) {\n  target += f[k] * foo_lpmf(z[k] | ...);\n}\nwhere the ellipses are the parameters of the log probability mass function for distribution foo (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).\nThe resulting program looks like a “weighted” regression, but here the weights f[k] are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies f[k] should be counted once in the transformed data block and stored.\nThe same trick works for combining multiple binomial observations.\n\n\n\nConsider the following Stan model for fitting a normal distribution to data.\ndata {\n  int N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nWith the vectorized form used for y, Stan is clever enough to only evaluate log(sigma) once, but it still has to evaluate the normal for all of y[1] to y[N], which involves adding up all the squared differences from the mean and then dividing by sigma squared.\nAn equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.\ndata {\n  int N;\n  vector[N] y;\n}\ntransformed data {\n  real mean_y = mean(y);\n  real&lt;lower=0&gt; var_y = variance(y);\n  real nm1_over2 = 0.5 * (N - 1);\n  real sqrt_N = sqrt(N);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mean_y ~ normal(mu, sigma / sqrt_N);\n  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^2);\n}\nThe data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.\nThe original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on mu and sigma are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.\nThis transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.\n\n\n\nThe Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace\ny ~ poisson(lambda);\nwith\nsum(y) ~ poisson(size(y) * lambda);\nThis will work even if y is a parameter vector because no Jacobian adjustment is required for summation.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#aggregating-common-subexpressions",
+    "href": "stan-users-guide/efficiency-tuning.html#aggregating-common-subexpressions",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "If an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using exp(theta) in multiple places, declare a local variable to store its value and reuse the local variable.\nAnother case that may not be so obvious is with two multilevel parameters, say a[ii[n]] + b[jj[n]]. If a and b are small (i.e., do not have many levels), then a table a_b of their sums can be created, with\nmatrix[size(a), size(b)] a_b;\nfor (i in 1:size(a)) {\n  for (j in 1:size(b)) {\n    a_b[i, j] = a[i] + b[j];\n  }\n}\nThen the sum can be replaced with a_b[ii[n], jj[n]].",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#exploiting-conjugacy",
+    "href": "stan-users-guide/efficiency-tuning.html#exploiting-conjugacy",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Continuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.\ntheta ~ beta(alpha + sum(y), beta + N - sum(y));\nTo make the model even more efficient, a transformed data variable defined to be sum(y) could be used in the place of sum(y).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#standardizing-predictors",
+    "href": "stan-users-guide/efficiency-tuning.html#standardizing-predictors",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "Standardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:\n\nIt helps in faster convergence of MCMC chains.\nIt makes the model less sensitive to the specifics of the parameterization.\nIt aids in the interpretation and comparison of the importance of coefficients across different predictors.\n\nWhen there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.\nSuppose that \\(y = (y_1,\\dotsc,y_N)\\) is a vector of \\(N\\) outcomes and \\(x = (x_1,\\dotsc,x_N)\\) the corresponding vector of \\(N\\) predictors. A simple linear regression involving an intercept coefficient \\(\\alpha\\) and slope coefficient \\(\\beta\\) can be expressed as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n,\n\\] where \\[\n\\epsilon_n \\sim \\textsf{normal}(0,\\sigma).\n\\]\nIf \\(x\\) has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values \\(x_n\\). First the elements of \\(x\\) are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.\nThe mean of \\(x\\) is given by:\n\\[\nmean_x = \\frac{1}{N} \\sum_{n=1}^{N} x_n\n\\]\nThe standard deviation of \\(x\\) is calculated as: \\[\nsd_x = {\\left({\\frac{1}{N} \\sum_{n=1}^{N} (x_n - mean_x)^2}\\right)}^{1/2}\n\\]\nWith these, we compute the \\(z\\), the standardized predictors\n\\[\nz_n = \\frac{x_n - mean_x}{sd_x}\n\\]\nwhere \\(z_n\\) is the standardized value corresponding to \\(x_n\\).\nThe inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.\n\\[\nx_n = z_n sd_x + mean_x\n\\]\nStandardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.\nConsider the following initial model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  // priors\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ normal(0, 5);\n  // likelihood\n  y ~ normal(x * beta + alpha, sigma);\n}\nThe data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\ntransformed data {\n  real mean_x = mean(x);\n  real sd_x = sd(x);\n  vector[N] x_std = (x - mean_x) / sd_x;\n}\nparameters {\n  real alpha_std;\n  real beta_std;\n  real&lt;lower=0&gt; sigma_std;\n}\nmodel {\n  alpha_std ~ normal(0, 10);\n  beta_std ~ normal(0, 10);\n  sigma_std ~ normal(0, 5);\n  y ~ normal(x_std * beta_std + alpha_std, sigma_std);\n}\nThe parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data x_std is defined in terms of variables mean_x and sd_x; by declaring these variables in the transformed data block, they will be available in all following blocks, and therefore can be used in the generated quantities block to record the “natural” parameters alpha and beta.\nThe fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.\nThe original regression \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\] has been transformed to a regression on the standardized data variable \\(z\\),\n\\[\ny_n = \\alpha' + \\beta' z_n + \\epsilon_n.\n\\]\nThe likelihood is specified in terms of the standardized parameters. The original slope \\(\\beta\\) is the standardized slope \\(\\beta'\\) scaled by the inverse of the standard deviation of \\(x\\). The original intercept \\(\\alpha\\) is the intercept from the standardized model \\(\\alpha'\\), corrected for the effect of scaling and centering \\(x\\). Thus, the formulas to retrieve \\(\\alpha\\) and \\(\\beta\\) from \\(\\alpha'\\) and \\(\\beta'\\) are:\n\\[\\begin{align*}\n\\beta = \\frac{\\beta'}{\\sigma_x} \\\\\n\\alpha = \\alpha' - \\beta' \\frac{\\mu_x}{\\sigma_x}\n\\end{align*}\\]\nThese recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,\ngenerated quantities {\n  real beta = beta_std / sd_x;\n  real alpha = alpha_std - beta_std * mean_x / sd_x;\n\n}\nWhen there are multiple real-valued predictors, i.e., when K is the number of predictors, x is an \\(N \\times K\\) matrix, and beta ia \\(K\\)-vector of coefficients, then x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. When \\(K \\ll N\\) the QR reparameterization is recommended for linear and generalized linear models unless there is an informative prior on the location of \\(\\beta\\).\n\n\nFor many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use\ny ~ std_normal();\nthan to use\ny ~ normal(0, 1);\nbecause the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#using-map-reduce",
+    "href": "stan-users-guide/efficiency-tuning.html#using-map-reduce",
+    "title": "Efficiency Tuning",
+    "section": "",
+    "text": "The map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the map-reduce chapter for more information on implementing map-reduce operations.\nMap-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/efficiency-tuning.html#footnotes",
+    "href": "stan-users-guide/efficiency-tuning.html#footnotes",
+    "title": "Efficiency Tuning",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis parameterization came to be known on our mailing lists as the “Matt trick” after Matt Hoffman, who independently came up with it while fitting hierarchical models in Stan.↩︎\nStan uses its own arena-based allocation, so allocation and deallocation are faster than with a raw call to new.↩︎\nFuture versions of Stan may remove this inefficiency by more fully exploiting expression templates inside the Eigen C++ matrix library. This will require enhancing Eigen to deal with mixed-type arguments, such as the type double used for constants and the algorithmic differentiation type stan::math::var used for variables.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html",
+    "href": "stan-users-guide/dae.html",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "Stan support solving systems of differential-algebraic equations (DAEs) of index 1 (Serban et al. 2021). The solver adaptively refines the solutions in order to satisfy given tolerances.\nOne can think a differential-algebraic system of equations as ODEs with additional algebraic constraints applied to some of the variables. In such a system, the variable derivatives may not be expressed explicitly with a right-hand-side as in ODEs, but implicitly constrained.\nSimilar to ODE solvers, the DAE solvers must not only provide the solution to the DAE itself, but also the gradient of the DAE solution with respect to parameters (the sensitivities). Stan’s DAE solver uses the forward sensitivity calculation to expand the base DAE system with additional DAE equations for the gradients of the solution. For each parameter, an additional full set of \\(N\\) sensitivity states are added meaning that the full DAE solved has \\(N \\, + N \\cdot M\\) states.\nTwo interfaces are provided for the forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. The same principles in the control parameters section apply here.\nInternally Stan’s DAE solver uses a variable-step, variable-order, backward-differentiation formula implementation (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005).\n\n\nA DAE is defined by a set of expressions for the residuals of differential equations and algebraic equations \\(r(y', y, t, \\theta)\\), and consistent initial conditions \\(y(t_0, \\theta) = y_0, y'(t_0, \\theta)=y'_0\\). The DAE is define by residual function as \\(r(y', y, t, \\theta)=0\\). The \\(\\theta\\) dependence is included in the notation to highlight that the solution \\(y(t)\\) is a function of any parameters used in the computation.\n\n\n\nAs an example of a system of DAEs, consider following chemical kinetics problem(Robertson 1966). The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants \\(y_1\\) and \\(y_2\\) to the product \\(y_3\\), and the algebraic equation describes the mass conservation. (Serban and Hindmarsh 2021).\n\\[\\begin{align*}\n\\frac{dy_1}{dt} + \\alpha y_1 - \\beta y_2 y_3 = 0 \\\\\n\\frac{dy_2}{dt} - \\alpha y_1 + \\beta y_2 y_3 + \\gamma y_2^2 = 0 \\\\\ny_1 + y_2 + y_3 - 1.0 = 0\n\\end{align*}\\]\nThe state equations implicitly defines the state \\((y_1(t), y_2(t), y_3(t))\\) at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients \\((\\alpha, \\beta, \\gamma)\\).\nUnlike solving ODEs, solving DAEs requires a consistent initial condition. That is, one must specify both \\(y(t_0)\\) and \\(y'(t_0)\\) so that residual function becomes zero at initial time \\(t_0\\) \\[\\begin{equation*}\nr(y'(t_0), y(t_0), t_0) = 0\n\\end{equation*}\\]\n\n\n\nThe index along a DAE solution \\(y(t)\\) is the minimum number of differentiations of some of the components of the system required to solve for \\(y'\\) uniquely in terms of \\(y\\) and \\(t\\), so that the DAE is converted into an ODE for \\(y\\). Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for \\(y_3\\).\nMost DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically (Ascher and Petzold 1998).\n\n\n\nThe first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the residual function is time, passed as a real; the second argument to the residual function is the system state \\(y\\), passed as a vector, the third argument to the residual function is the state derivative \\(y'\\), also passed as a vector. The residual function’s return value is a vector of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient \\(\\alpha\\), \\(\\beta\\), and \\(\\gamma\\)), data, or any quantities that are needed to define the DAE.\nThe above reaction be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector chem(real t, vector yy, vector yp,\n            real alpha, real beta, real gamma) {\n  vector[3] res;\n  res[1] = yp[1] + alpha * yy[1] - beta * yy[2] * yy[3];\n  res[2] = yp[2] - alpha * yy[1] + beta * yy[2] * yy[3] + gamma * yy[2] * yy[2];\n  res[3] = yy[1] + yy[2] + yy[3] - 1.0;\n  return res;\n}\nThe function takes in a time t (a real), the system state yy (a vector), state derivative yp (a vector), as well as parameter alpha (a real), beta (a real), and gamma (a real). The function returns a vector of the residuals at time t. The DAE coded here does not explicitly depend on t, however one still needs to specify t as an argument.\n\n\nThe types in the DAE residual function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, the third argument is the state derivative passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible DAE signatures:\nvector my_dae1(real t, vector y, vector yp, real a0);\nvector my_dae2(real t, vector y, vector yp, array[] int a0, vector a1);\nvector my_dae3(real t, vector y, vector yp, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector my_dae1(real t, array[] real y, vector yp);\n// Second argument is not a vector\narray[] real my_dae2(real t, vector y, vector yp);\n// Return type is not a vector\nvector my_dae3(real t, vector y);\n// First argument is not a real and missing the third argument\n\n\n\n\nStan provides a dae function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.\ndata {\n  int N;\n  vector[3] yy0;\n  vector[3] yp0;\n  real t0;\n  real alpha;\n  real beta;\n  array[N] real ts;\n  array[N] vector[3] y;\n}\nparameters {\n  real gamma;\n}\ntransformed parameters {\n  vector[3] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);\n}\nSince gamma is a parameter, the DAE solver is called in the transformed parameters block.\n\n\n\nUsing dae_tol one can specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters in order to control the DAE solution.\nvector[3] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,\n                             relative_tolerance,\n                             absolute_tolerance,\n                             max_num_steps,\n                             alpha, beta, gamma);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nThe default value of relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html#notation",
+    "href": "stan-users-guide/dae.html#notation",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "A DAE is defined by a set of expressions for the residuals of differential equations and algebraic equations \\(r(y', y, t, \\theta)\\), and consistent initial conditions \\(y(t_0, \\theta) = y_0, y'(t_0, \\theta)=y'_0\\). The DAE is define by residual function as \\(r(y', y, t, \\theta)=0\\). The \\(\\theta\\) dependence is included in the notation to highlight that the solution \\(y(t)\\) is a function of any parameters used in the computation.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html#example-chemical-kinetics",
+    "href": "stan-users-guide/dae.html#example-chemical-kinetics",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "As an example of a system of DAEs, consider following chemical kinetics problem(Robertson 1966). The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants \\(y_1\\) and \\(y_2\\) to the product \\(y_3\\), and the algebraic equation describes the mass conservation. (Serban and Hindmarsh 2021).\n\\[\\begin{align*}\n\\frac{dy_1}{dt} + \\alpha y_1 - \\beta y_2 y_3 = 0 \\\\\n\\frac{dy_2}{dt} - \\alpha y_1 + \\beta y_2 y_3 + \\gamma y_2^2 = 0 \\\\\ny_1 + y_2 + y_3 - 1.0 = 0\n\\end{align*}\\]\nThe state equations implicitly defines the state \\((y_1(t), y_2(t), y_3(t))\\) at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients \\((\\alpha, \\beta, \\gamma)\\).\nUnlike solving ODEs, solving DAEs requires a consistent initial condition. That is, one must specify both \\(y(t_0)\\) and \\(y'(t_0)\\) so that residual function becomes zero at initial time \\(t_0\\) \\[\\begin{equation*}\nr(y'(t_0), y(t_0), t_0) = 0\n\\end{equation*}\\]",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html#index-of-daes",
+    "href": "stan-users-guide/dae.html#index-of-daes",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "The index along a DAE solution \\(y(t)\\) is the minimum number of differentiations of some of the components of the system required to solve for \\(y'\\) uniquely in terms of \\(y\\) and \\(t\\), so that the DAE is converted into an ODE for \\(y\\). Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for \\(y_3\\).\nMost DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically (Ascher and Petzold 1998).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html#coding-the-dae-system-function",
+    "href": "stan-users-guide/dae.html#coding-the-dae-system-function",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "The first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the residual function is time, passed as a real; the second argument to the residual function is the system state \\(y\\), passed as a vector, the third argument to the residual function is the state derivative \\(y'\\), also passed as a vector. The residual function’s return value is a vector of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient \\(\\alpha\\), \\(\\beta\\), and \\(\\gamma\\)), data, or any quantities that are needed to define the DAE.\nThe above reaction be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector chem(real t, vector yy, vector yp,\n            real alpha, real beta, real gamma) {\n  vector[3] res;\n  res[1] = yp[1] + alpha * yy[1] - beta * yy[2] * yy[3];\n  res[2] = yp[2] - alpha * yy[1] + beta * yy[2] * yy[3] + gamma * yy[2] * yy[2];\n  res[3] = yy[1] + yy[2] + yy[3] - 1.0;\n  return res;\n}\nThe function takes in a time t (a real), the system state yy (a vector), state derivative yp (a vector), as well as parameter alpha (a real), beta (a real), and gamma (a real). The function returns a vector of the residuals at time t. The DAE coded here does not explicitly depend on t, however one still needs to specify t as an argument.\n\n\nThe types in the DAE residual function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, the third argument is the state derivative passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible DAE signatures:\nvector my_dae1(real t, vector y, vector yp, real a0);\nvector my_dae2(real t, vector y, vector yp, array[] int a0, vector a1);\nvector my_dae3(real t, vector y, vector yp, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector my_dae1(real t, array[] real y, vector yp);\n// Second argument is not a vector\narray[] real my_dae2(real t, vector y, vector yp);\n// Return type is not a vector\nvector my_dae3(real t, vector y);\n// First argument is not a real and missing the third argument",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html#solving-daes",
+    "href": "stan-users-guide/dae.html#solving-daes",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "Stan provides a dae function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.\ndata {\n  int N;\n  vector[3] yy0;\n  vector[3] yp0;\n  real t0;\n  real alpha;\n  real beta;\n  array[N] real ts;\n  array[N] vector[3] y;\n}\nparameters {\n  real gamma;\n}\ntransformed parameters {\n  vector[3] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);\n}\nSince gamma is a parameter, the DAE solver is called in the transformed parameters block.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/dae.html#control-dae.section",
+    "href": "stan-users-guide/dae.html#control-dae.section",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "Using dae_tol one can specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters in order to control the DAE solution.\nvector[3] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,\n                             relative_tolerance,\n                             absolute_tolerance,\n                             max_num_steps,\n                             alpha, beta, gamma);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nThe default value of relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/cross-validation.html",
+    "href": "stan-users-guide/cross-validation.html",
+    "title": "Held-Out Evaluation and Cross-Validation",
+    "section": "",
+    "text": "Held-out evaluation involves splitting a data set into two parts, a training data set and a test data set. The training data is used to estimate the model and the test data is used for evaluation. Held-out evaluation is commonly used to declare winners in predictive modeling competitions such as those run by Kaggle.\nCross-validation involves repeated held-out evaluations performed by partitioning a single data set in different ways. The training/test split can be done either by randomly selecting the test set, or by partitioning the data set into several equally-sized subsets and then using each subset in turn as the test data with the other folds as training data.\nHeld-out evaluation and cross-validation may involve any kind of predictive statistics, with common choices being the predictive log density on test data, squared error of parameter estimates, or accuracy in a classification task.\n\n\nGiven training data \\((x, y)\\) consisting of parallel sequences of predictors and observations and test data \\((\\tilde{x}, \\tilde{y})\\) of the same structure, the posterior predictive density is \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n=\n\\int\n  p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n  \\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta,\n\\]\nwhere \\(\\theta\\) is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors \\(\\tilde{x}\\) and the training data \\((x, y).\\)\nThis integral may be calculated with Monte Carlo methods as usual, \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\] where the \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y)\\) are draws from the posterior given only the training data \\((x, y).\\)\nTo avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid \\tilde{x}, x, y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M \\exp(\\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}))\n\\\\[4pt]\n& = & -\\log M + \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)})\n\\end{eqnarray*}\\] where the log sum of exponentials function is defined so as to make the above equation hold, \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\, \\mu_m\n= \\log \\sum_{m=1}^M \\exp(\\mu_m).\n\\] The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\mu_m\n= \\textrm{max}(\\mu)\n+ \\log \\sum_{m = 1}^M \\exp(\\mu_m - \\textrm{max}(\\mu)).\n\\] Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms \\(\\mu_m - \\textrm{max}(\\mu) \\leq 0\\), and thus will not overflow.\n\n\nTo evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of \\(\\log p(\\tilde{y} \\mid \\tilde{x},\n\\theta^{(m)}).\\)\nHere is the code for evaluating the log posterior predictive density in a simple linear regression of the test data \\(\\tilde{y}\\) given predictors \\(\\tilde{x}\\) and training data \\((x, y).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  vector[N_tilde] y_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  real log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);\n}\nOnly the training data x and y are used in the model block. The test data y_tilde and test predictors x_tilde appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use x_tilde in the model block—only the test observations y_tilde are unknown before they are predicted.\nGiven \\(M\\) posterior draws from Stan, the sequence log_p[1:M] will be available, so that the log posterior predictive density of the test data given training data and predictors is just log_sum_exp(log_p) - log(M).\n\n\n\n\n\n\nEstimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If \\(\\theta\\) is the true value and \\(\\hat{\\theta}\\) the estimate, then error is just the difference between the prediction and the true value, \\[\n\\textrm{err} = \\hat{\\theta} - \\theta.\n\\]\nIf the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. Absolute error is defined as \\[\n\\textrm{abs-err} = \\left| \\hat{\\theta} - \\theta \\right|\n\\] and squared error as \\[\n\\textrm{sq-err} = \\left( \\hat{\\theta} - \\theta \\right)^2.\n\\] Gneiting and Raftery (2007) provide a thorough overview of such scoring rules and their properties.\nBayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.\n\n\n\nIn addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, \\(z_1, \\ldots, z_N,\\) then it is common to report mean square error (MSE), \\[\n\\textrm{mse}\n= \\frac{1}{N} \\sum_{n = 1}^N \\left( \\hat{z}_n - z_n\\right)^2.\n\\] To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as root mean square error (RMSE), \\[\n\\textrm{rmse} = \\sqrt{\\textrm{mean-sq-err}}.\n\\]\n\n\n\nConsider a simple linear regression model, parameters for the intercept \\(\\alpha\\) and slope \\(\\beta\\), along with predictors \\(\\tilde{x}_n\\). The standard Bayesian estimate is the expected value of \\(\\tilde{y}\\) given the predictors and training data, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}[\\tilde{y}_n \\mid \\tilde{x}_n, x, y]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}_n^{(m)}\n\\end{eqnarray*}\\] where \\(\\tilde{y}_n^{(m)}\\) is drawn from the data model \\[\n\\tilde{y}_n^{(m)}\n\\sim p(\\tilde{y}_n \\mid \\tilde{x}_n, \\alpha^{(m)}, \\beta^{(m)}),\n\\] for parameters \\(\\alpha^{(m)}\\) and \\(\\beta^{(m)}\\) drawn from the posterior, \\[\n(\\alpha^{(m)}, \\beta^{(m)}) \\sim p(\\alpha, \\beta \\mid x, y).\n\\]\nIn the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws \\(\\tilde{y}_n^{(m)}\\), the same result is obtained by averaging linear predictions, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}\\left[\n          \\alpha + \\beta \\cdot \\tilde{x}_n\n          \\mid \\tilde{x}_n, x, y\n       \\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\] This is possible because \\[\n\\tilde{y}_n^{(m)} \\sim \\textrm{normal}(\\tilde{y}_n \\mid \\alpha^{(m)} +\n\\beta^{(m)} \\cdot \\tilde{x}_n, \\sigma^{(m)}),\n\\] and the normal distribution has symmetric error so that the expectation of \\(\\tilde{y}_n^{(m)}\\) is the same as \\(\\alpha^{(m)} + \\beta^{(m)} \\cdot\n\\tilde{x}_n\\). Replacing the sampled quantity \\(\\tilde{y}_n^{(m)}\\) with its expectation is a general variance reduction technique for Monte Carlo estimates known as Rao-Blackwellization (Rao 1945; Blackwell 1947).\nIn the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, \\[\\begin{eqnarray*}\n\\tilde{y}_n^{(m)}\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\left( \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n \\right)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\frac{1}{M} \\sum_{m = 1}^M (\\beta^{(m)} \\cdot \\tilde{x}_n)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\left( \\frac{1}{M} \\sum_{m = 1}^M \\beta^{(m)}\\right) \\cdot \\tilde{x}_n\n\\\\[4pt]\n& = & \\hat{\\alpha} + \\hat{\\beta} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\]\nIn Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  // ...\n}\n// ...\ngenerated quantities {\n  vector[N_tilde] tilde_y = alpha + beta * x_tilde;\n}\nThe posterior mean of tilde_y calculated by Stan is the Bayesian estimate \\(\\hat{\\tilde{y}}.\\) The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.\n\n\n\n\nCross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation (Vehtari, Gelman, and Gabry 2017).\nPartitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.\n\n\nFor the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.\nfunctions {\n  array[] int permutation_rng(int N) {\n    array[N] int y;\n    for (n in 1 : N) {\n      y[n] = n;\n    }\n    vector[N] theta = rep_vector(1.0 / N, N);\n    for (n in 1 : size(y)) {\n      int i = categorical_rng(theta);\n      int temp = y[n];\n      y[n] = y[i];\n      y[i] = temp;\n    }\n    return y;\n  }\n}\nThe name of the function must end in _rng because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers 1:N.1\nThe transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=N&gt; N_test;\n}\ntransformed data {\n  int N_train = N - N_test;\n  array[N] int permutation = permutation_rng(N);\n  vector[N_train] x_train = x[permutation[1 : N_train]];\n  vector[N_train] y_train = y[permutation[1 : N_train]];\n  vector[N_test] x_test = x[permutation[N_train + 1 : N]];\n  vector[N_test] y_test = y[permutation[N_train + 1 : N]];\n}\nRecall that in Stan, permutation[1:N_train] is an array of integers, so that x[permutation[1 : N_train]] is a vector defined for i in 1:N_train by\nx[permutation[1 : N_train]][i] = x[permutation[1:N_train][i]]\n                               = x[permutation[i]]\nGiven the test/train split, the rest of the model is straightforward.\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_train ~ normal(alpha + beta * x_train, sigma);\n  { alpha, beta, sigma } ~ normal(0, 1);\n}\ngenerated quantities {\n  vector[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);\n  vector[N] err = y_test_sim - y_hat;\n}\nThe prediction y_test_hat is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, \\[\\begin{eqnarray*}\n\\hat{y}^{\\textrm{test}}\n& = & \\mathbb{E}\\left[ y^{\\textrm{test}} \\mid x^{\\textrm{test}}, x^{\\textrm{train}} y^{\\textrm{train}} \\right]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)}.\n\\end{eqnarray*}\\]\nBecause the test set is constant and the expectation operator is linear, the posterior mean of err as defined in the Stan program will be the error of the posterior mean estimate, \\[\\begin{eqnarray*}\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n  - y^{\\textrm{test}}\n\\\\[4pt]\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)} - y^{\\textrm{test}},\n\\end{eqnarray*}\\] where \\[\n\\hat{y}^{\\textrm{test}(m)}\n\\sim p(y \\mid x^{\\textrm{test}}, x^{\\textrm{train}},\ny^{\\textrm{train}}).\n\\] This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should not be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.\nBecause the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.\n\n\n\nIt is straightforward to declare the variable permutation in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.\n\n\n\nCross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to mixed replication in posterior predictive checking, where there is a choice to simulate new elements of existing groups or generate entirely new groups.\nEducation testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.\n\n\n\nOften data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.\n\n\n\nVehtari, Gelman, and Gabry (2017) introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo (Gabry et al. 2019).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/cross-validation.html#evaluating-posterior-predictive.section",
+    "href": "stan-users-guide/cross-validation.html#evaluating-posterior-predictive.section",
+    "title": "Held-Out Evaluation and Cross-Validation",
+    "section": "",
+    "text": "Given training data \\((x, y)\\) consisting of parallel sequences of predictors and observations and test data \\((\\tilde{x}, \\tilde{y})\\) of the same structure, the posterior predictive density is \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n=\n\\int\n  p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n  \\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta,\n\\]\nwhere \\(\\theta\\) is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors \\(\\tilde{x}\\) and the training data \\((x, y).\\)\nThis integral may be calculated with Monte Carlo methods as usual, \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\] where the \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y)\\) are draws from the posterior given only the training data \\((x, y).\\)\nTo avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid \\tilde{x}, x, y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M \\exp(\\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}))\n\\\\[4pt]\n& = & -\\log M + \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)})\n\\end{eqnarray*}\\] where the log sum of exponentials function is defined so as to make the above equation hold, \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\, \\mu_m\n= \\log \\sum_{m=1}^M \\exp(\\mu_m).\n\\] The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\mu_m\n= \\textrm{max}(\\mu)\n+ \\log \\sum_{m = 1}^M \\exp(\\mu_m - \\textrm{max}(\\mu)).\n\\] Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms \\(\\mu_m - \\textrm{max}(\\mu) \\leq 0\\), and thus will not overflow.\n\n\nTo evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of \\(\\log p(\\tilde{y} \\mid \\tilde{x},\n\\theta^{(m)}).\\)\nHere is the code for evaluating the log posterior predictive density in a simple linear regression of the test data \\(\\tilde{y}\\) given predictors \\(\\tilde{x}\\) and training data \\((x, y).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  vector[N_tilde] y_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  real log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);\n}\nOnly the training data x and y are used in the model block. The test data y_tilde and test predictors x_tilde appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use x_tilde in the model block—only the test observations y_tilde are unknown before they are predicted.\nGiven \\(M\\) posterior draws from Stan, the sequence log_p[1:M] will be available, so that the log posterior predictive density of the test data given training data and predictors is just log_sum_exp(log_p) - log(M).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/cross-validation.html#estimation-error",
+    "href": "stan-users-guide/cross-validation.html#estimation-error",
+    "title": "Held-Out Evaluation and Cross-Validation",
+    "section": "",
+    "text": "Estimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If \\(\\theta\\) is the true value and \\(\\hat{\\theta}\\) the estimate, then error is just the difference between the prediction and the true value, \\[\n\\textrm{err} = \\hat{\\theta} - \\theta.\n\\]\nIf the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. Absolute error is defined as \\[\n\\textrm{abs-err} = \\left| \\hat{\\theta} - \\theta \\right|\n\\] and squared error as \\[\n\\textrm{sq-err} = \\left( \\hat{\\theta} - \\theta \\right)^2.\n\\] Gneiting and Raftery (2007) provide a thorough overview of such scoring rules and their properties.\nBayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.\n\n\n\nIn addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, \\(z_1, \\ldots, z_N,\\) then it is common to report mean square error (MSE), \\[\n\\textrm{mse}\n= \\frac{1}{N} \\sum_{n = 1}^N \\left( \\hat{z}_n - z_n\\right)^2.\n\\] To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as root mean square error (RMSE), \\[\n\\textrm{rmse} = \\sqrt{\\textrm{mean-sq-err}}.\n\\]\n\n\n\nConsider a simple linear regression model, parameters for the intercept \\(\\alpha\\) and slope \\(\\beta\\), along with predictors \\(\\tilde{x}_n\\). The standard Bayesian estimate is the expected value of \\(\\tilde{y}\\) given the predictors and training data, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}[\\tilde{y}_n \\mid \\tilde{x}_n, x, y]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}_n^{(m)}\n\\end{eqnarray*}\\] where \\(\\tilde{y}_n^{(m)}\\) is drawn from the data model \\[\n\\tilde{y}_n^{(m)}\n\\sim p(\\tilde{y}_n \\mid \\tilde{x}_n, \\alpha^{(m)}, \\beta^{(m)}),\n\\] for parameters \\(\\alpha^{(m)}\\) and \\(\\beta^{(m)}\\) drawn from the posterior, \\[\n(\\alpha^{(m)}, \\beta^{(m)}) \\sim p(\\alpha, \\beta \\mid x, y).\n\\]\nIn the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws \\(\\tilde{y}_n^{(m)}\\), the same result is obtained by averaging linear predictions, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}\\left[\n          \\alpha + \\beta \\cdot \\tilde{x}_n\n          \\mid \\tilde{x}_n, x, y\n       \\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\] This is possible because \\[\n\\tilde{y}_n^{(m)} \\sim \\textrm{normal}(\\tilde{y}_n \\mid \\alpha^{(m)} +\n\\beta^{(m)} \\cdot \\tilde{x}_n, \\sigma^{(m)}),\n\\] and the normal distribution has symmetric error so that the expectation of \\(\\tilde{y}_n^{(m)}\\) is the same as \\(\\alpha^{(m)} + \\beta^{(m)} \\cdot\n\\tilde{x}_n\\). Replacing the sampled quantity \\(\\tilde{y}_n^{(m)}\\) with its expectation is a general variance reduction technique for Monte Carlo estimates known as Rao-Blackwellization (Rao 1945; Blackwell 1947).\nIn the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, \\[\\begin{eqnarray*}\n\\tilde{y}_n^{(m)}\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\left( \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n \\right)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\frac{1}{M} \\sum_{m = 1}^M (\\beta^{(m)} \\cdot \\tilde{x}_n)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\left( \\frac{1}{M} \\sum_{m = 1}^M \\beta^{(m)}\\right) \\cdot \\tilde{x}_n\n\\\\[4pt]\n& = & \\hat{\\alpha} + \\hat{\\beta} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\]\nIn Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  // ...\n}\n// ...\ngenerated quantities {\n  vector[N_tilde] tilde_y = alpha + beta * x_tilde;\n}\nThe posterior mean of tilde_y calculated by Stan is the Bayesian estimate \\(\\hat{\\tilde{y}}.\\) The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/cross-validation.html#cross-validation",
+    "href": "stan-users-guide/cross-validation.html#cross-validation",
+    "title": "Held-Out Evaluation and Cross-Validation",
+    "section": "",
+    "text": "Cross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation (Vehtari, Gelman, and Gabry 2017).\nPartitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.\n\n\nFor the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.\nfunctions {\n  array[] int permutation_rng(int N) {\n    array[N] int y;\n    for (n in 1 : N) {\n      y[n] = n;\n    }\n    vector[N] theta = rep_vector(1.0 / N, N);\n    for (n in 1 : size(y)) {\n      int i = categorical_rng(theta);\n      int temp = y[n];\n      y[n] = y[i];\n      y[i] = temp;\n    }\n    return y;\n  }\n}\nThe name of the function must end in _rng because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers 1:N.1\nThe transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=N&gt; N_test;\n}\ntransformed data {\n  int N_train = N - N_test;\n  array[N] int permutation = permutation_rng(N);\n  vector[N_train] x_train = x[permutation[1 : N_train]];\n  vector[N_train] y_train = y[permutation[1 : N_train]];\n  vector[N_test] x_test = x[permutation[N_train + 1 : N]];\n  vector[N_test] y_test = y[permutation[N_train + 1 : N]];\n}\nRecall that in Stan, permutation[1:N_train] is an array of integers, so that x[permutation[1 : N_train]] is a vector defined for i in 1:N_train by\nx[permutation[1 : N_train]][i] = x[permutation[1:N_train][i]]\n                               = x[permutation[i]]\nGiven the test/train split, the rest of the model is straightforward.\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_train ~ normal(alpha + beta * x_train, sigma);\n  { alpha, beta, sigma } ~ normal(0, 1);\n}\ngenerated quantities {\n  vector[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);\n  vector[N] err = y_test_sim - y_hat;\n}\nThe prediction y_test_hat is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, \\[\\begin{eqnarray*}\n\\hat{y}^{\\textrm{test}}\n& = & \\mathbb{E}\\left[ y^{\\textrm{test}} \\mid x^{\\textrm{test}}, x^{\\textrm{train}} y^{\\textrm{train}} \\right]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)}.\n\\end{eqnarray*}\\]\nBecause the test set is constant and the expectation operator is linear, the posterior mean of err as defined in the Stan program will be the error of the posterior mean estimate, \\[\\begin{eqnarray*}\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n  - y^{\\textrm{test}}\n\\\\[4pt]\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)} - y^{\\textrm{test}},\n\\end{eqnarray*}\\] where \\[\n\\hat{y}^{\\textrm{test}(m)}\n\\sim p(y \\mid x^{\\textrm{test}}, x^{\\textrm{train}},\ny^{\\textrm{train}}).\n\\] This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should not be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.\nBecause the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.\n\n\n\nIt is straightforward to declare the variable permutation in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.\n\n\n\nCross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to mixed replication in posterior predictive checking, where there is a choice to simulate new elements of existing groups or generate entirely new groups.\nEducation testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.\n\n\n\nOften data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.\n\n\n\nVehtari, Gelman, and Gabry (2017) introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo (Gabry et al. 2019).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/cross-validation.html#footnotes",
+    "href": "stan-users-guide/cross-validation.html#footnotes",
+    "title": "Held-Out Evaluation and Cross-Validation",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe traditional approach is to walk through a vector and replace each item with a random element from the remaining elements, which is guaranteed to only move each item once. This was not done here as it’d require new categorical theta because Stan does not have a uniform discrete RNG built in.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/complex-numbers.html",
+    "href": "stan-users-guide/complex-numbers.html",
+    "title": "Complex Numbers",
+    "section": "",
+    "text": "Stan supports complex scalars, matrices, and vectors as well as real-based ones.\n\n\nThis section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.\n\n\nComplex numbers can be constructed using imaginary literals. For example,\ncomplex z = -1.1 + 2.3i;\nproduces the complex number \\(-1.1 + 2.3i\\). This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the to_complex() function may be used. For example, the following code will work if x and y are parameters, transformed data, or local variables in a function or model block.\nreal x = // ...\nreal y = // ...\ncomplex z = to_complex(x, y);\nThe real and imaginary parts of the complex number can be accessed with getters as follows.\nreal x = get_real(z);  // x = -1.1\nreal y = get_imag(z);  // y = 2.3\nComplex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.\ncomplex a = 3.2 + 2i;\ncomplex b = to_complex(3.2, 2);\nif (a == b) print(\"hello\");\n\n\n\nInteger- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.\ncomplex z1 = 3;  // int promoted to 3 + 0i\ncomplex z2 = 3.2;  // real promoted to 3.2 + 0.0i\n\n\n\nArrays of complex numbers work as usual and allow the usual curly bracket constructors.\ncomplex z1;  complex z2;  complex z3;\n// ...\narray[3] complex zs = { z1, z2, z3 };\nfor (z in zs) {\n  print(z);\n}\nComplex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.\n\n\n\nAll of the standard complex functions are available, including natural logarithm log(z), natural exponentiation exp(z), and powers pow(z1, z2), as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as sin(z), acos(z), tanh(z) and asinh(z).\nPromotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.\ncomplex times_i(complex z) {\n  complex i = to_complex(0, 1);\n  return i * z;\n}\nFor instance, times_i(1) evaluates to the imaginary base \\(i\\), as does times_i(1.0).\n\n\n\n\nThe simplest way to model a distribution over a complex random number \\(z = x + yi\\) is to consider its real part \\(x\\) and imaginary part \\(y\\) to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.\ncomplex z;\nvector[2] mu;\ncholesky_factor_cov[2] L_Sigma;\n// ...\n[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);\nFor example, if z is data, this can be used to estimate mu and the covariance Cholesky factor L_Sigma. Alternatively, if z is a parameter, mu and L_Sigma may constants defining a prior or further parameters defining a hierarchical model.\n\n\n\nStan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.\ncomplex_vector[3] v;\ncomplex_row_vector[2] rv;\ncomplex_matrix[3, 2] m;\nWe can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of rv above, we could assign it to a constructed row vector.\nrv =  [2 + 3i, 1.9 - 2.3i];\nComplex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.\nFurthermore, it is possible to convert back and forth between arrays and matrices using the to_array functions.\n\n\n\nComplex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take x to be predictors, y to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.\ndata {\n  int&lt;lower=0&gt; N;\n  complex_vector[N] x;\n  complex_vector[N] y;\n}\nparameters {\n  complex alpha;\n  complex beta;\n}\nmodel {\n  complex_vector[N] eps = y - (alpha + beta * x);\n  eps ~  // ...error distribution...\n}\nThe question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.\n\n\nThe simplest approach to error in complex regression is to give the real and imaginary parts of eps_n independent independent normal distributions, as follows.\nparameters {\n  // ...\n  vector[2] sigma;\n}\n// ...\nmodel {\n  // ...\n  get_real(eps) ~ normal(0, sigma[1]);\n  get_imag(eps) ~ normal(0, sigma[2]);\n  sigma ~ //...hyperprior...\n}\nA new error scale vector sigma is introduced, and it should itself get a prior based on the expected scale of error for the problem.\n\n\n\nThe next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.\nparameters {\n  cholesky_factor_corr[2] L_Omega;  // correlation matrix\n  vector[2] sigma;                  // real, imag error scales\n  // ...\n}\n// ...\nmodel {\n  array[N] vector[2] eps_arr;\n  for (n in 1:N) {\n    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };\n  }\n  eps_arr ~ multi_normal_cholesky([0, 0]',\n                                  diag_pre_multiply(sigma, L_Omega));\n  L_Omega ~ lkj_cholesky(4);  // shrink toward diagonal correlation\n  sigma ~ // ... hyperprior ...\n}\nHere, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector sigma and a Cholesky factor or a correlation matrix, L_Omega. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to multi_normal_cholesky, the vector of complex numbers needs to be converted to an array of size 2 vectors.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/complex-numbers.html#working-with-complex-numbers",
+    "href": "stan-users-guide/complex-numbers.html#working-with-complex-numbers",
+    "title": "Complex Numbers",
+    "section": "",
+    "text": "This section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.\n\n\nComplex numbers can be constructed using imaginary literals. For example,\ncomplex z = -1.1 + 2.3i;\nproduces the complex number \\(-1.1 + 2.3i\\). This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the to_complex() function may be used. For example, the following code will work if x and y are parameters, transformed data, or local variables in a function or model block.\nreal x = // ...\nreal y = // ...\ncomplex z = to_complex(x, y);\nThe real and imaginary parts of the complex number can be accessed with getters as follows.\nreal x = get_real(z);  // x = -1.1\nreal y = get_imag(z);  // y = 2.3\nComplex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.\ncomplex a = 3.2 + 2i;\ncomplex b = to_complex(3.2, 2);\nif (a == b) print(\"hello\");\n\n\n\nInteger- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.\ncomplex z1 = 3;  // int promoted to 3 + 0i\ncomplex z2 = 3.2;  // real promoted to 3.2 + 0.0i\n\n\n\nArrays of complex numbers work as usual and allow the usual curly bracket constructors.\ncomplex z1;  complex z2;  complex z3;\n// ...\narray[3] complex zs = { z1, z2, z3 };\nfor (z in zs) {\n  print(z);\n}\nComplex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.\n\n\n\nAll of the standard complex functions are available, including natural logarithm log(z), natural exponentiation exp(z), and powers pow(z1, z2), as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as sin(z), acos(z), tanh(z) and asinh(z).\nPromotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.\ncomplex times_i(complex z) {\n  complex i = to_complex(0, 1);\n  return i * z;\n}\nFor instance, times_i(1) evaluates to the imaginary base \\(i\\), as does times_i(1.0).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/complex-numbers.html#complex-random-variables",
+    "href": "stan-users-guide/complex-numbers.html#complex-random-variables",
+    "title": "Complex Numbers",
+    "section": "",
+    "text": "The simplest way to model a distribution over a complex random number \\(z = x + yi\\) is to consider its real part \\(x\\) and imaginary part \\(y\\) to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.\ncomplex z;\nvector[2] mu;\ncholesky_factor_cov[2] L_Sigma;\n// ...\n[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);\nFor example, if z is data, this can be used to estimate mu and the covariance Cholesky factor L_Sigma. Alternatively, if z is a parameter, mu and L_Sigma may constants defining a prior or further parameters defining a hierarchical model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/complex-numbers.html#complex-matrices-and-vectors",
+    "href": "stan-users-guide/complex-numbers.html#complex-matrices-and-vectors",
+    "title": "Complex Numbers",
+    "section": "",
+    "text": "Stan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.\ncomplex_vector[3] v;\ncomplex_row_vector[2] rv;\ncomplex_matrix[3, 2] m;\nWe can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of rv above, we could assign it to a constructed row vector.\nrv =  [2 + 3i, 1.9 - 2.3i];\nComplex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.\nFurthermore, it is possible to convert back and forth between arrays and matrices using the to_array functions.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/complex-numbers.html#complex-linear-regression",
+    "href": "stan-users-guide/complex-numbers.html#complex-linear-regression",
+    "title": "Complex Numbers",
+    "section": "",
+    "text": "Complex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take x to be predictors, y to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.\ndata {\n  int&lt;lower=0&gt; N;\n  complex_vector[N] x;\n  complex_vector[N] y;\n}\nparameters {\n  complex alpha;\n  complex beta;\n}\nmodel {\n  complex_vector[N] eps = y - (alpha + beta * x);\n  eps ~  // ...error distribution...\n}\nThe question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.\n\n\nThe simplest approach to error in complex regression is to give the real and imaginary parts of eps_n independent independent normal distributions, as follows.\nparameters {\n  // ...\n  vector[2] sigma;\n}\n// ...\nmodel {\n  // ...\n  get_real(eps) ~ normal(0, sigma[1]);\n  get_imag(eps) ~ normal(0, sigma[2]);\n  sigma ~ //...hyperprior...\n}\nA new error scale vector sigma is introduced, and it should itself get a prior based on the expected scale of error for the problem.\n\n\n\nThe next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.\nparameters {\n  cholesky_factor_corr[2] L_Omega;  // correlation matrix\n  vector[2] sigma;                  // real, imag error scales\n  // ...\n}\n// ...\nmodel {\n  array[N] vector[2] eps_arr;\n  for (n in 1:N) {\n    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };\n  }\n  eps_arr ~ multi_normal_cholesky([0, 0]',\n                                  diag_pre_multiply(sigma, L_Omega));\n  L_Omega ~ lkj_cholesky(4);  // shrink toward diagonal correlation\n  sigma ~ // ... hyperprior ...\n}\nHere, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector sigma and a Cholesky factor or a correlation matrix, L_Omega. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to multi_normal_cholesky, the vector of complex numbers needs to be converted to an array of size 2 vectors.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/bootstrap.html",
+    "href": "stan-users-guide/bootstrap.html",
+    "title": "The Bootstrap and Bagging",
+    "section": "",
+    "text": "The bootstrap is a technique for approximately sampling from the error distribution for an estimator. Thus it can be used as a Monte Carlo method to estimate standard errors and confidence intervals for point estimates (Efron and Tibshirani 1986; 1994). It works by subsampling the original data and computing sample estimates from the subsample. Like other Monte Carlo methods, the bootstrap is plug-and-play, allowing great flexibility in both model choice and estimator.\nBagging is a technique for combining bootstrapped estimators for model criticism and more robust inference (Breiman 1996; Huggins and Miller 2019).\n\n\n\n\nAn estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set \\(y_{1,\\ldots, N}\\) to a number by \\[\n\\textrm{mean}(y) = \\frac{1}{N} \\sum_{n=1}^N y_n,\n\\] and hence meets the definition of an estimator. Given the likelihood function \\[\np(y \\mid \\mu) = \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, 1),\n\\] the mean is the maximum likelihood estimator,\n\\[\n\\textrm{mean}(y) = \\textrm{arg max}_{\\mu} \\ p(y \\mid \\mu, 1)\n\\] A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.\nIn analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.\n\n\n\nThe bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.\nThe following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate \\(\\hat{\\theta}\\) that is a function of data \\(y\\).\nfor (m in 1:M) {\n  y_rep[m] &lt;- sample_uniform(y)\n  theta_hat[m] &lt;- estimate_theta(y_rep[m])\n}\nstd_error = sd(theta_hat)\nconf_95pct = [ quantile(theta_hat, 0.025),\n               quantile(theta_hat, 0.975) ]\nThe sample_uniform function works by independently assigning each element of y_rep an element of y drawn uniformly at random. This produces a sample with replacement. That is, some elements of y may show up more than once in y_rep and some may not appear at all.\n\n\n\n\nThe bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector x of predictors in addition to outcomes y. To allow a single program to fit both the original data and random subsamples, the variable resample is set to 1 to resample and 0 to use the original data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=1&gt; resample;\n}\ntransformed data {\n  simplex[N] uniform = rep_vector(1.0 / N, N);\n  array[N] int&lt;lower=1, upper=N&gt; boot_idxs;\n  for (n in 1:N) {\n    boot_idxs[n] = resample ? categorical_rng(uniform) : n;\n  }\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);\n}\nThe model accepts data in the usual form for a linear regression as a number of observations \\(N\\) with a size \\(N\\) vector \\(x\\) of predictors and a size \\(N\\) vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of boot_idxs from 1:N, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array boot_idxs is defined to be the sequence 1:N, because x == x[1:N] and y = y[1:N].\nFor example, when resample == 1, if \\(N = 4,\\) the value of boot_idxs might be {2, 1, 1, 3}, resulting in a bootstrap sample {y[2], y[1], y[1], y[3]} with the first element repeated twice and the fourth element not sampled at all.\nThe parameters are the usual regression coefficients for the intercept alpha, slope beta, and error scale sigma. The model uses the bootstrap index variable boot_idx to index the predictors as x[boot_idx] and outcomes as y[boot_idx]. This generates a new size-\\(N\\) vector whose entries are defined by x[boot_idx][n] = x[boot_idx[n]] and similarly for y. For example, if \\(N = 4\\) and boot_idxs = {2, 1, 1, 3}, then x[boot_idxs] = [x[2], x[1], x[1], x[3]]' and y[boot_idxs] = [y[2], y[1], y[1], y[3]]'. The predictor and outcome vectors remain aligned, with both elements of the pair x[1] and y[1] repeated twice.\nWith the model defined this way, if resample is 1, the model is fit to a bootstrap subsample of the data. If resample is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.\n\n\n\nRunning the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.\nTo estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data \\(y\\), it is usually not worth calculating bootstrap estimates to high precision.\n\n\nHere’s the result of calculating standard errors for the linear regression model above with \\(N = 50\\) data points, \\(\\alpha = 1.2, \\beta\n= -0.5,\\) and \\(\\sigma = 1.5.\\) With a total of \\(M = 100\\) bootstrap samples, there are 100 estimates of \\(\\alpha\\), 100 of \\(\\beta\\), and 100 of \\(\\sigma\\). These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for \\(\\alpha\\) provide the bootstrap estimate of the standard error in the estimate for \\(\\alpha\\). Here’s what it looks like for the above model with \\(M =\n100\\)\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.218\n      beta     -0.610      0.204\n     sigma      1.537      0.142\nWith the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.206\n     alpha      1.359      0.240\n     alpha      1.359      0.234\n     alpha      1.359      0.249\n     alpha      1.359      0.227\nIncreasing \\(M\\) will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample \\(y\\).\n\n\n\nAs usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are \\(M = 1000\\) estimates of \\(\\hat{\\alpha}\\) in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set \\(y\\). To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).\n\n\n\n\nWhen bootstrapping is carried through inference it is known as bootstrap aggregation, or bagging, in the machine-learning literature (Breiman 1996). In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate \\(\\hat{\\sigma}\\) from the original data set, bootstrapped data sets \\(y^{\\textrm{boot}(1)}, \\ldots,\ny^{\\textrm{boot}(N)}\\) are generated. Each is used to generate an estimate \\(\\hat{\\sigma}^{\\textrm{boot}(n)}.\\) The final estimate is \\[\n\\hat{\\sigma} = \\frac{1}{N} \\sum_{n = 1}^N \\hat{\\sigma}^{\\textrm{boot}(n)}.\n\\] The same would be done to estimate a predictive quantity \\(\\tilde{y}\\) for as yet unseen data. \\[\n\\hat{\\tilde{y}} = \\frac{1}{N} \\sum_{n = 1}^N\n\\hat{\\tilde{y}}^{\\textrm{boot}(n)}.\n\\] For discrete parameters, voting is used to select the outcome.\nOne way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.\n\n\n\nA Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.\n(Huggins and Miller 2019) use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/bootstrap.html#the-bootstrap",
+    "href": "stan-users-guide/bootstrap.html#the-bootstrap",
+    "title": "The Bootstrap and Bagging",
+    "section": "",
+    "text": "An estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set \\(y_{1,\\ldots, N}\\) to a number by \\[\n\\textrm{mean}(y) = \\frac{1}{N} \\sum_{n=1}^N y_n,\n\\] and hence meets the definition of an estimator. Given the likelihood function \\[\np(y \\mid \\mu) = \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, 1),\n\\] the mean is the maximum likelihood estimator,\n\\[\n\\textrm{mean}(y) = \\textrm{arg max}_{\\mu} \\ p(y \\mid \\mu, 1)\n\\] A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.\nIn analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.\n\n\n\nThe bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.\nThe following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate \\(\\hat{\\theta}\\) that is a function of data \\(y\\).\nfor (m in 1:M) {\n  y_rep[m] &lt;- sample_uniform(y)\n  theta_hat[m] &lt;- estimate_theta(y_rep[m])\n}\nstd_error = sd(theta_hat)\nconf_95pct = [ quantile(theta_hat, 0.025),\n               quantile(theta_hat, 0.975) ]\nThe sample_uniform function works by independently assigning each element of y_rep an element of y drawn uniformly at random. This produces a sample with replacement. That is, some elements of y may show up more than once in y_rep and some may not appear at all.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/bootstrap.html#coding-the-bootstrap-in-stan",
+    "href": "stan-users-guide/bootstrap.html#coding-the-bootstrap-in-stan",
+    "title": "The Bootstrap and Bagging",
+    "section": "",
+    "text": "The bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector x of predictors in addition to outcomes y. To allow a single program to fit both the original data and random subsamples, the variable resample is set to 1 to resample and 0 to use the original data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=1&gt; resample;\n}\ntransformed data {\n  simplex[N] uniform = rep_vector(1.0 / N, N);\n  array[N] int&lt;lower=1, upper=N&gt; boot_idxs;\n  for (n in 1:N) {\n    boot_idxs[n] = resample ? categorical_rng(uniform) : n;\n  }\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);\n}\nThe model accepts data in the usual form for a linear regression as a number of observations \\(N\\) with a size \\(N\\) vector \\(x\\) of predictors and a size \\(N\\) vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of boot_idxs from 1:N, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array boot_idxs is defined to be the sequence 1:N, because x == x[1:N] and y = y[1:N].\nFor example, when resample == 1, if \\(N = 4,\\) the value of boot_idxs might be {2, 1, 1, 3}, resulting in a bootstrap sample {y[2], y[1], y[1], y[3]} with the first element repeated twice and the fourth element not sampled at all.\nThe parameters are the usual regression coefficients for the intercept alpha, slope beta, and error scale sigma. The model uses the bootstrap index variable boot_idx to index the predictors as x[boot_idx] and outcomes as y[boot_idx]. This generates a new size-\\(N\\) vector whose entries are defined by x[boot_idx][n] = x[boot_idx[n]] and similarly for y. For example, if \\(N = 4\\) and boot_idxs = {2, 1, 1, 3}, then x[boot_idxs] = [x[2], x[1], x[1], x[3]]' and y[boot_idxs] = [y[2], y[1], y[1], y[3]]'. The predictor and outcome vectors remain aligned, with both elements of the pair x[1] and y[1] repeated twice.\nWith the model defined this way, if resample is 1, the model is fit to a bootstrap subsample of the data. If resample is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/bootstrap.html#error-statistics-from-the-bootstrap",
+    "href": "stan-users-guide/bootstrap.html#error-statistics-from-the-bootstrap",
+    "title": "The Bootstrap and Bagging",
+    "section": "",
+    "text": "Running the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.\nTo estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data \\(y\\), it is usually not worth calculating bootstrap estimates to high precision.\n\n\nHere’s the result of calculating standard errors for the linear regression model above with \\(N = 50\\) data points, \\(\\alpha = 1.2, \\beta\n= -0.5,\\) and \\(\\sigma = 1.5.\\) With a total of \\(M = 100\\) bootstrap samples, there are 100 estimates of \\(\\alpha\\), 100 of \\(\\beta\\), and 100 of \\(\\sigma\\). These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for \\(\\alpha\\) provide the bootstrap estimate of the standard error in the estimate for \\(\\alpha\\). Here’s what it looks like for the above model with \\(M =\n100\\)\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.218\n      beta     -0.610      0.204\n     sigma      1.537      0.142\nWith the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.206\n     alpha      1.359      0.240\n     alpha      1.359      0.234\n     alpha      1.359      0.249\n     alpha      1.359      0.227\nIncreasing \\(M\\) will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample \\(y\\).\n\n\n\nAs usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are \\(M = 1000\\) estimates of \\(\\hat{\\alpha}\\) in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set \\(y\\). To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/bootstrap.html#bagging",
+    "href": "stan-users-guide/bootstrap.html#bagging",
+    "title": "The Bootstrap and Bagging",
+    "section": "",
+    "text": "When bootstrapping is carried through inference it is known as bootstrap aggregation, or bagging, in the machine-learning literature (Breiman 1996). In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate \\(\\hat{\\sigma}\\) from the original data set, bootstrapped data sets \\(y^{\\textrm{boot}(1)}, \\ldots,\ny^{\\textrm{boot}(N)}\\) are generated. Each is used to generate an estimate \\(\\hat{\\sigma}^{\\textrm{boot}(n)}.\\) The final estimate is \\[\n\\hat{\\sigma} = \\frac{1}{N} \\sum_{n = 1}^N \\hat{\\sigma}^{\\textrm{boot}(n)}.\n\\] The same would be done to estimate a predictive quantity \\(\\tilde{y}\\) for as yet unseen data. \\[\n\\hat{\\tilde{y}} = \\frac{1}{N} \\sum_{n = 1}^N\n\\hat{\\tilde{y}}^{\\textrm{boot}(n)}.\n\\] For discrete parameters, voting is used to select the outcome.\nOne way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/bootstrap.html#bayesian-bootstrap-and-bagging",
+    "href": "stan-users-guide/bootstrap.html#bayesian-bootstrap-and-bagging",
+    "title": "The Bootstrap and Bagging",
+    "section": "",
+    "text": "A Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.\n(Huggins and Miller 2019) use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
+    ]
+  },
+  {
+    "objectID": "reference-manual/whitespace.html",
+    "href": "reference-manual/whitespace.html",
+    "title": "Whitespace",
+    "section": "",
+    "text": "The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).\n\n\n\nStan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.\nOther than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.\n\n\n\nZero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as a * b, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.\nIdentifiers and literals may not be separated by whitespace. Thus it is not legal to write the number 10000 as 10 000 or to write the identifier normal_lpdf as normal _ lpdf.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Whitespace"
+    ]
+  },
+  {
+    "objectID": "reference-manual/whitespace.html#whitespace-characters",
+    "href": "reference-manual/whitespace.html#whitespace-characters",
+    "title": "Whitespace",
+    "section": "",
+    "text": "The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Whitespace"
+    ]
+  },
+  {
+    "objectID": "reference-manual/whitespace.html#whitespace-neutrality",
+    "href": "reference-manual/whitespace.html#whitespace-neutrality",
+    "title": "Whitespace",
+    "section": "",
+    "text": "Stan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.\nOther than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Whitespace"
+    ]
+  },
+  {
+    "objectID": "reference-manual/whitespace.html#whitespace-location",
+    "href": "reference-manual/whitespace.html#whitespace-location",
+    "title": "Whitespace",
+    "section": "",
+    "text": "Zero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as a * b, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.\nIdentifiers and literals may not be separated by whitespace. Thus it is not legal to write the number 10000 as 10 000 or to write the identifier normal_lpdf as normal _ lpdf.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Whitespace"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html",
+    "href": "reference-manual/user-functions.html",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Stan allows users to define their own functions. The basic syntax is a simplified version of that used in C and C++. This chapter specifies how functions are declared, defined, and used in Stan.\n\n\nUser-defined functions appear in a special function-definition block before all of the other program blocks.\nfunctions {\n   // ... function declarations and definitions ...\n}\ndata {\n  // ...\nFunction definitions and declarations may appear in any order. Forward declarations are allowed but not required.\n\n\n\nThe rules for function naming and function-argument naming are the same as for other variables; see the section on variables for more information on valid identifiers. For example,\nreal foo(real mu, real sigma);\ndeclares a function named foo with two argument variables of types real and real. The arguments are named mu and sigma, but that is not part of the declaration.\n\n\nMultiple user-defined functions may have the same name if they have different sequences of argument types. This is known as function overloading.\nFor example, the following two functions are both defined with the name add_up\nreal add_up(real a, real b){\n  return a + b;\n}\n\nreal add_up(real a, real b, real c){\n  return a + b + c;\n}\nThe return types of overloaded functions do not need to be the same. One could define an additional add_up function as follows\nint add_up(int a, int b){\n  return a + b;\n}\nThat being said, functions may not use the same name if their signature only differs by the return type.\nFor example, the following is not permitted\n// illegal\nreal baz(int x);\nint baz(int x);\nFunction names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the reduce_sum family of functions and ODE integrators, which cannot be overloaded.\n\n\n\n\nAll function arguments are mandatory—there are no default values.\n\n\nFunctions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.\n\n\n\nFunctions with void return types may be applied to arguments and used as statements. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.\n\n\n\nOverloaded functions alongside type promotion can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.\nConsider the following two overloaded functions\nreal foo(int a, real b);\nreal foo(real a, int b);\nThese functions do not have a unique minimum when called with two integer arguments foo(1,2), and therefore cannot be called as such.\nPromotion of integers to complex numbers is considered as two separate promotions, one from int to real and a second from real to complex. Consider the following functions with real and complex signatures\nreal bar(real x);\nreal bar(complex z);\nA call bar(5) with an integer argument will be resolved to bar(real) because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.\n\n\n\nThe rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.\nreal foo(real x) { return ... };\n...\nint a = 5;\nreal b = foo(a); // a promoted to type real\nIn addition to promoting int to real, Stan also promotes real to complex, and by transitivity, int to complex. This also works for containers, so an array of int may be assigned to an array of real of the same shape. And we can also promote vector to complex_vector and similarly for row vectors and matrices.\n\n\n\nFunctions whose name ends in _lpdf or _lpmf (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of statements.qmd#distribution-statements.section.\n\n\n\nFunctions of certain types are restricted on scope of usage. Functions whose names end in _lp assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.\nFunctions whose name end in _jacobian assume access to the log probability accumulator may only be used within the transformed parameters block.\nFunctions whose names end in _rng assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in _rng.\nFunctions whose names end in _lpdf and _lpmf can be used anywhere. However, _lupdf and _lupmf functions can only be used in the model block or user-defined probability functions.\nSee the section on function bodies for more information on these special types of function.\n\n\n\n\nStan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.\nThe type void may not be used as an argument type, only a return type for a function with side effects.\n\n\nThe base variable types are integer, real, complex, vector, row_vector, and matrix. No lower-bound or upper-bound constraints are allowed (e.g., real&lt;lower=0&gt; is illegal). Specialized constrained types are also not allowed (e.g., simplex is illegal).\nTuple types of the form tuple(T1, ..., TN) are also allowed, with all of the types T1 to TN being function argument types (i.e., no constraints and no sizes).\n\n\n\nArguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, int denotes a single integer argument or return, whereas array[] real indicates a one-dimensional array of reals, array[,] real a two-dimensional array and array[,,] real a three-dimensional array; whitespace is optional, as usual.\nThe dimensions for vectors and matrices are not included, so that matrix is the type of a single matrix argument or return type. Thus if a variable is declared as matrix a, then a has two indexing dimensions, so that a[1] is a row vector and a[1, 1] a real value. Matrices implicitly have two indexing dimensions. The type declaration matrix[ , ] b specifies that b is a two-dimensional array of matrices, for a total of four indexing dimensions, with b[1, 1, 1, 1] picking out a real value.\n\n\n\nFunction argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.\nDimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.\nAny errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.\n\n\n\nSome of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.\nIn user-defined functions, the qualifier data may be placed before an argument type declaration to indicate that the argument must be data only. For example,\nreal foo(data real x) {\n  return x^2;\n}\nrequires the argument x to be data only.\nDeclaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.\n\n\n\n\nThe body of a function is between an open curly brace ({) and close curly brace (}). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.\nAny user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.\nThe only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.\n\n\nFunctions that call random number generating functions in their bodies must have a name that ends in _rng; attempts to use random-number generators in other functions lead to a compile-time error.\nLike other random number generating functions, user-defined functions with names that end in _rng may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in _rng. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions that include distribution statements or log probability increment statements must have a name that ends in _lp. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.\nLike the target log density increment statement and distribution statements, user-defined functions with names that end in _lp may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions whose names end in _lpdf and _lpmf (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (~) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of y given parameter theta allows the use of the distribution statement is defined as follows.\nreal foo_lpdf(real y, vector theta) { ... }\nNote that for function definitions, the comma is used rather than the vertical bar.\nFor every custom _lpdf and _lpmf defined there is a corresponding _lupdf and _lupmf defined automatically. The _lupdf and _lupmf versions of the functions cannot be defined directly (to do so will produce an error). The difference in the _lpdf and _lpmf and the corresponding _lupdf and _lupmf functions is that if any other unnormalized density functions are used inside the user-defined function, the _lpdf and _lpmf forms of the user-defined function will change these densities to be normalized. The _lupdf and _lupmf forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.\nThe distribution statement shorthand\nz ~ foo(phi);\nwill have the same effect as incrementing the target with the log of the unnormalized density:\ntarget += foo_lupdf(z | phi);\nOther _lupdf and _lupmf functions used in the definition of foo_lpdf will drop additive constants when foo_lupdf is called and will not drop additive constants when foo_lpdf is called.\nIf there are _lupdf and _lupmf functions used inside the following call to foo_lpdf, they will be forced to normalize (return the equivalent of their _lpdf and _lpmf forms):\ntarget += foo_lpdf(z | phi);\nIf there are no _lupdf or _lupmf functions used in the definition of foo_lpdf, then there will be no difference between a foo_lpdf or foo_lupdf call.\nThe unnormalized _lupdf and _lupmf functions can only be used in the model block or in user-defined probability functions (those ending in _lpdf or _lpmf).\nThe same syntax and shorthand that works for _lpdf also works for log probability mass functions with suffixes _lpmf.\nA function that is going to be accessed as distributions must return the log of the density or mass function it defines.\n\n\n\n\nWithin function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (=) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.1\nLocal variables may be declared at the top of the function block and scope as usual.\n\n\n\nNon-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.\nVoid functions may use return only without an argument, but return statements are not mandatory.\n\n\nUnlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that\n\na return statement qualifies,\n\nand the recursive cases are that\n\na sequence of statements qualifies if its last statement qualifies,\na for loop or while loop qualifies if its body qualifies, and\na conditional statement qualifies if it has a default else clause and all of its body statements qualify.\n\nAn exception is made for “obviously infinite” loops like while (1), which contain a return statement and no break statements. The only way to exit such a loop is to return, so they are considered as returning statements.\nThese rules disqualify\nreal foo(real x) {\n  if (x &gt; 2) {\n    return 1.0;\n  } else if (x &lt;= 2) {\n    return -1.0;\n  }\n}\nbecause there is no default else clause, and disqualify\nreal foo(real x) {\n  real y;\n  y = x;\n  while (x &lt; 10) {\n    if (x &gt; 0) {\n      return x;\n    }\n    y = x / 2;\n  }\n}\nbecause the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow\nreal log_fancy(real x) {\n  if (x &lt; 1e-30) {\n    return x;\n  } else if (x &lt; 1e-14) {\n    return x * x;\n  } else {\n    return log(x);\n  }\n}\nbecause there’s a default else clause and each condition body has return as its final statement.\n\n\n\n\n\n\nA function can be declared without a return value by using void in place of a return type. Note that the type void may only be used as a return type—arguments may not be declared to be of type void.\n\n\n\nA void function may be used as a statement.\nBecause there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.\n\n\n\nIn a return statement within a void function’s definition, the return keyword is followed immediately by a semicolon (;) rather than by the expression whose value is returned.\n\n\n\n\nStan supports forward declarations, which look like function definitions without bodies. For example,\nreal unit_normal_lpdf(real y);\ndeclares a function named unit_normal_lpdf that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.\nA function definition with a body simultaneously declares and defines the named function, as in\nreal unit_normal_lpdf(real y) {\n  return -0.5 * square(y);\n}\nA function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the section on function overloading.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#function-definition-block",
+    "href": "reference-manual/user-functions.html#function-definition-block",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "User-defined functions appear in a special function-definition block before all of the other program blocks.\nfunctions {\n   // ... function declarations and definitions ...\n}\ndata {\n  // ...\nFunction definitions and declarations may appear in any order. Forward declarations are allowed but not required.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#function-names",
+    "href": "reference-manual/user-functions.html#function-names",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "The rules for function naming and function-argument naming are the same as for other variables; see the section on variables for more information on valid identifiers. For example,\nreal foo(real mu, real sigma);\ndeclares a function named foo with two argument variables of types real and real. The arguments are named mu and sigma, but that is not part of the declaration.\n\n\nMultiple user-defined functions may have the same name if they have different sequences of argument types. This is known as function overloading.\nFor example, the following two functions are both defined with the name add_up\nreal add_up(real a, real b){\n  return a + b;\n}\n\nreal add_up(real a, real b, real c){\n  return a + b + c;\n}\nThe return types of overloaded functions do not need to be the same. One could define an additional add_up function as follows\nint add_up(int a, int b){\n  return a + b;\n}\nThat being said, functions may not use the same name if their signature only differs by the return type.\nFor example, the following is not permitted\n// illegal\nreal baz(int x);\nint baz(int x);\nFunction names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the reduce_sum family of functions and ODE integrators, which cannot be overloaded.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#calling-functions",
+    "href": "reference-manual/user-functions.html#calling-functions",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "All function arguments are mandatory—there are no default values.\n\n\nFunctions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.\n\n\n\nFunctions with void return types may be applied to arguments and used as statements. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.\n\n\n\nOverloaded functions alongside type promotion can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.\nConsider the following two overloaded functions\nreal foo(int a, real b);\nreal foo(real a, int b);\nThese functions do not have a unique minimum when called with two integer arguments foo(1,2), and therefore cannot be called as such.\nPromotion of integers to complex numbers is considered as two separate promotions, one from int to real and a second from real to complex. Consider the following functions with real and complex signatures\nreal bar(real x);\nreal bar(complex z);\nA call bar(5) with an integer argument will be resolved to bar(real) because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.\n\n\n\nThe rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.\nreal foo(real x) { return ... };\n...\nint a = 5;\nreal b = foo(a); // a promoted to type real\nIn addition to promoting int to real, Stan also promotes real to complex, and by transitivity, int to complex. This also works for containers, so an array of int may be assigned to an array of real of the same shape. And we can also promote vector to complex_vector and similarly for row vectors and matrices.\n\n\n\nFunctions whose name ends in _lpdf or _lpmf (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of statements.qmd#distribution-statements.section.\n\n\n\nFunctions of certain types are restricted on scope of usage. Functions whose names end in _lp assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.\nFunctions whose name end in _jacobian assume access to the log probability accumulator may only be used within the transformed parameters block.\nFunctions whose names end in _rng assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in _rng.\nFunctions whose names end in _lpdf and _lpmf can be used anywhere. However, _lupdf and _lupmf functions can only be used in the model block or user-defined probability functions.\nSee the section on function bodies for more information on these special types of function.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#argument-types-and-qualifiers",
+    "href": "reference-manual/user-functions.html#argument-types-and-qualifiers",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Stan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.\nThe type void may not be used as an argument type, only a return type for a function with side effects.\n\n\nThe base variable types are integer, real, complex, vector, row_vector, and matrix. No lower-bound or upper-bound constraints are allowed (e.g., real&lt;lower=0&gt; is illegal). Specialized constrained types are also not allowed (e.g., simplex is illegal).\nTuple types of the form tuple(T1, ..., TN) are also allowed, with all of the types T1 to TN being function argument types (i.e., no constraints and no sizes).\n\n\n\nArguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, int denotes a single integer argument or return, whereas array[] real indicates a one-dimensional array of reals, array[,] real a two-dimensional array and array[,,] real a three-dimensional array; whitespace is optional, as usual.\nThe dimensions for vectors and matrices are not included, so that matrix is the type of a single matrix argument or return type. Thus if a variable is declared as matrix a, then a has two indexing dimensions, so that a[1] is a row vector and a[1, 1] a real value. Matrices implicitly have two indexing dimensions. The type declaration matrix[ , ] b specifies that b is a two-dimensional array of matrices, for a total of four indexing dimensions, with b[1, 1, 1, 1] picking out a real value.\n\n\n\nFunction argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.\nDimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.\nAny errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.\n\n\n\nSome of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.\nIn user-defined functions, the qualifier data may be placed before an argument type declaration to indicate that the argument must be data only. For example,\nreal foo(data real x) {\n  return x^2;\n}\nrequires the argument x to be data only.\nDeclaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#function-bodies.section",
+    "href": "reference-manual/user-functions.html#function-bodies.section",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "The body of a function is between an open curly brace ({) and close curly brace (}). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.\nAny user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.\nThe only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.\n\n\nFunctions that call random number generating functions in their bodies must have a name that ends in _rng; attempts to use random-number generators in other functions lead to a compile-time error.\nLike other random number generating functions, user-defined functions with names that end in _rng may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in _rng. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions that include distribution statements or log probability increment statements must have a name that ends in _lp. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.\nLike the target log density increment statement and distribution statements, user-defined functions with names that end in _lp may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions whose names end in _lpdf and _lpmf (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (~) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of y given parameter theta allows the use of the distribution statement is defined as follows.\nreal foo_lpdf(real y, vector theta) { ... }\nNote that for function definitions, the comma is used rather than the vertical bar.\nFor every custom _lpdf and _lpmf defined there is a corresponding _lupdf and _lupmf defined automatically. The _lupdf and _lupmf versions of the functions cannot be defined directly (to do so will produce an error). The difference in the _lpdf and _lpmf and the corresponding _lupdf and _lupmf functions is that if any other unnormalized density functions are used inside the user-defined function, the _lpdf and _lpmf forms of the user-defined function will change these densities to be normalized. The _lupdf and _lupmf forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.\nThe distribution statement shorthand\nz ~ foo(phi);\nwill have the same effect as incrementing the target with the log of the unnormalized density:\ntarget += foo_lupdf(z | phi);\nOther _lupdf and _lupmf functions used in the definition of foo_lpdf will drop additive constants when foo_lupdf is called and will not drop additive constants when foo_lpdf is called.\nIf there are _lupdf and _lupmf functions used inside the following call to foo_lpdf, they will be forced to normalize (return the equivalent of their _lpdf and _lpmf forms):\ntarget += foo_lpdf(z | phi);\nIf there are no _lupdf or _lupmf functions used in the definition of foo_lpdf, then there will be no difference between a foo_lpdf or foo_lupdf call.\nThe unnormalized _lupdf and _lupmf functions can only be used in the model block or in user-defined probability functions (those ending in _lpdf or _lpmf).\nThe same syntax and shorthand that works for _lpdf also works for log probability mass functions with suffixes _lpmf.\nA function that is going to be accessed as distributions must return the log of the density or mass function it defines.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#parameters-are-constant",
+    "href": "reference-manual/user-functions.html#parameters-are-constant",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Within function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (=) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.1\nLocal variables may be declared at the top of the function block and scope as usual.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#return-value",
+    "href": "reference-manual/user-functions.html#return-value",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Non-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.\nVoid functions may use return only without an argument, but return statements are not mandatory.\n\n\nUnlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that\n\na return statement qualifies,\n\nand the recursive cases are that\n\na sequence of statements qualifies if its last statement qualifies,\na for loop or while loop qualifies if its body qualifies, and\na conditional statement qualifies if it has a default else clause and all of its body statements qualify.\n\nAn exception is made for “obviously infinite” loops like while (1), which contain a return statement and no break statements. The only way to exit such a loop is to return, so they are considered as returning statements.\nThese rules disqualify\nreal foo(real x) {\n  if (x &gt; 2) {\n    return 1.0;\n  } else if (x &lt;= 2) {\n    return -1.0;\n  }\n}\nbecause there is no default else clause, and disqualify\nreal foo(real x) {\n  real y;\n  y = x;\n  while (x &lt; 10) {\n    if (x &gt; 0) {\n      return x;\n    }\n    y = x / 2;\n  }\n}\nbecause the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow\nreal log_fancy(real x) {\n  if (x &lt; 1e-30) {\n    return x;\n  } else if (x &lt; 1e-14) {\n    return x * x;\n  } else {\n    return log(x);\n  }\n}\nbecause there’s a default else clause and each condition body has return as its final statement.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#void-functions-as-statements",
+    "href": "reference-manual/user-functions.html#void-functions-as-statements",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "A function can be declared without a return value by using void in place of a return type. Note that the type void may only be used as a return type—arguments may not be declared to be of type void.\n\n\n\nA void function may be used as a statement.\nBecause there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.\n\n\n\nIn a return statement within a void function’s definition, the return keyword is followed immediately by a semicolon (;) rather than by the expression whose value is returned.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#forward-declarations.section",
+    "href": "reference-manual/user-functions.html#forward-declarations.section",
+    "title": "User-Defined Functions",
+    "section": "",
+    "text": "Stan supports forward declarations, which look like function definitions without bodies. For example,\nreal unit_normal_lpdf(real y);\ndeclares a function named unit_normal_lpdf that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.\nA function definition with a body simultaneously declares and defines the named function, as in\nreal unit_normal_lpdf(real y) {\n  return -0.5 * square(y);\n}\nA function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the section on function overloading.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/user-functions.html#footnotes",
+    "href": "reference-manual/user-functions.html#footnotes",
+    "title": "User-Defined Functions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nDespite being declared constant and appearing to have a pass-by-value syntax in Stan, the implementation of the language passes function arguments by constant reference in C++.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "User-Defined Functions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html",
+    "href": "reference-manual/transforms.html",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "To avoid having to deal with constraints while simulating the Hamiltonian dynamics during sampling, every (multivariate) parameter in a Stan model is transformed to an unconstrained variable behind the scenes by the model compiler. The transform is based on the constraints, if any, in the parameter’s definition. Scalars or the scalar values in vectors, row vectors or matrices may be constrained with lower and/or upper bounds. Vectors may alternatively be constrained to be ordered, positive ordered, or simplexes. Matrices may be constrained to be correlation matrices or covariance matrices. This chapter provides a definition of the transforms used for each type of variable. For examples of how to declare and define these variables in a Stan program, see section Variable declaration. To directly access the functional form of these transformations from inside the Stan language, see Variable Transformation Functions in the Functions Reference.\nStan converts models to C++ classes which define probability functions with support on all of \\(\\mathbb{R}^K\\), where \\(K\\) is the number of unconstrained parameters needed to define the constrained parameters defined in the program. The C++ classes also include code to transform the parameters from unconstrained to constrained and apply the appropriate Jacobians.\n\n\nIn this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.\nStan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with lower=0. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about floating point arithmetic in the Stan User’s Guide.\nThese issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the CmdStan Command-Line Interface Overview.\n\n\n\nThe support of a random variable \\(X\\) with density \\(p_X(x)\\) is that subset of values for which it has non-zero density,\n\\[\n\\mathrm{supp}(X) = \\{ x | p_X(x) &gt; 0 \\}.\n\\]\nIf \\(f\\) is a total function defined on the support of \\(X\\), then \\(Y =\nf(X)\\) is a new random variable. This section shows how to compute the probability density function of \\(Y\\) for well-behaved transforms \\(f\\). The rest of the chapter details the transforms used by Stan.\n\n\nSuppose \\(X\\) is one dimensional and \\(f: \\mathrm{supp}(X) \\rightarrow\n\\mathbb{R}\\) is a one-to-one, monotonic function with a differentiable inverse \\(f^{-1}\\). Then the density of \\(Y\\) is given by\n\\[\np_Y(y) = p_X(f^{-1}(y))\n         \\,\n         \\left| \\, \\frac{d}{dy} f^{-1}(y)\\, \\right|.\n\\]\nThe absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.\n\n\n\nThe multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.\nSuppose \\(X\\) is a \\(K\\)-dimensional random variable with probability density function \\(p_X(x)\\). A new random variable \\(Y = f(X)\\) may be defined by transforming \\(X\\) with a suitably well-behaved function \\(f\\). It suffices for what follows to note that if \\(f\\) is one-to-one and its inverse \\(f^{-1}\\) has a well-defined Jacobian, then the density of \\(Y\\) is\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\, \\left| \\, \\det \\, J_{f^{-1}}(y) \\, \\right|,\n\\]\nwhere \\(\\det{}\\) is the matrix determinant operation and \\(J_{f^{-1}}(y)\\) is the Jacobian matrix of \\(f^{-1}\\) evaluated at \\(y\\). Taking \\(x =\nf^{-1}(y)\\), the Jacobian matrix is defined by\n\\[\nJ_{f^{-1}}(y) =\n\\left[\n\\begin{array}{ccc}\\displaystyle\n\\frac{\\partial x_1}{\\partial y_1}\n& \\cdots\n& \\displaystyle \\frac{\\partial x_1}{\\partial y_{K}}\n\\\\\n\\vdots & \\vdots & \\vdots\n\\\\\n\\displaystyle\\frac{\\partial x_{K}}{\\partial y_1}\n& \\cdots\n& \\displaystyle\\frac{\\partial x_{K}}{\\partial y_{K}}\n\\end{array}\n\\right].\n\\]\nIf the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,\n\\[\n\\det \\, J_{f^{-1}}(y)\n= \\prod_{k=1}^K \\frac{\\partial x_k}{\\partial y_k}.\n\\]\nTriangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.\n\n\n\n\nStan uses a logarithmic transform for lower and upper bounds.\n\n\nIf a variable \\(X\\) is declared to have lower bound \\(a\\), it is transformed to an unbounded variable \\(Y\\), where\n\\[\nY = \\log(X - a).\n\\]\n\n\n\nThe inverse of the lower-bound transform maps an unbounded variable \\(Y\\) to a variable \\(X\\) that is bounded below by \\(a\\) by\n\\[\nX = \\exp(Y) + a.\n\\]\n\n\n\nThe absolute derivative of the inverse transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( \\exp(y) + a \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, given the density \\(p_X\\) of \\(X\\), the density of \\(Y\\) is\n\\[\np_Y(y)\n= p_X\\!\\left( \\exp(y) + a \\right) \\cdot \\exp(y).\n\\]\n\n\n\n\nStan uses a negated logarithmic transform for upper bounds.\n\n\nIf a variable \\(X\\) is declared to have an upper bound \\(b\\), it is transformed to the unbounded variable \\(Y\\) by\n\\[\nY = \\log(b - X).\n\\]\n\n\n\nThe inverse of the upper bound transform converts the unbounded variable \\(Y\\) to the variable \\(X\\) bounded above by \\(b\\) through\n\\[\nX = b - \\exp(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse of the upper bound transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( b - \\exp(y) \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, the density of the unconstrained variable \\(Y\\) is defined in terms of the density of the variable \\(X\\) with an upper bound of \\(b\\) by\n\\[\np_Y(y)\n=   p_X \\!\\left( b - \\exp(y) \\right) \\cdot \\exp(y).\n\\]\n\n\n\n\nFor lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.\n\n\nThe log-odds function is defined for \\(u \\in (0,1)\\) by\n\\[\n\\mathrm{logit}(u) = \\log \\frac{u}{1 - u}.\n\\]\nThe inverse of the log odds function is the logistic sigmoid, defined for \\(v \\in (-\\infty,\\infty)\\) by\n\\[\n\\mathrm{logit}^{-1}(v) = \\frac{1}{1 + \\exp(-v)}.\n\\]\nThe derivative of the logistic sigmoid is\n\\[\n\\frac{d}{dy} \\mathrm{logit}^{-1}(y)\n= \\mathrm{logit}^{-1}(y) \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\n\n\n\nFor variables constrained to be in the open interval \\((a, b)\\), Stan uses a scaled and translated log-odds transform. If variable \\(X\\) is declared to have lower bound \\(a\\) and upper bound \\(b\\), then it is transformed to a new variable \\(Y\\), where\n\\[\nY = \\mathrm{logit} \\left( \\frac{X - a}{b - a} \\right).\n\\]\n\n\n\nThe inverse of this transform is\n\\[\nX = a + (b - a) \\cdot \\mathrm{logit}^{-1}(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse transform is given by\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      a + (b - a) \\cdot \\mathrm{logit}^{-1}(y)\n    \\right)\n  \\right|\n= (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( a + (b - a) \\cdot \\mathrm{logit}^{-1}(y) \\right)\n    \\cdot (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nDespite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, \\(\\mathrm{logit}^{-1}(y)\\) only needs to be evaluated once, so there is only one call to \\(\\exp(-y)\\).\n\n\n\n\nStan uses an affine transform to be able to specify parameters with a given offset and multiplier.\n\n\nFor variables with expected offset \\(\\mu\\) and/or (positive) multiplier \\(\\sigma\\), Stan uses an affine transform. Such a variable \\(X\\) is transformed to a new variable \\(Y\\), where\n\\[\nY = \\frac{X - \\mu}{\\sigma}.\n\\]\nThe default value for the offset \\(\\mu\\) is \\(0\\) and for the multiplier \\(\\sigma\\) is \\(1\\) in case not both are specified.\n\n\n\nThe inverse of this transform is\n\\[\nX =  \\mu + \\sigma \\cdot Y.\n\\]\n\n\n\nThe absolute derivative of the affine inverse transform is\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      \\mu + \\sigma \\cdot y\n    \\right)\n  \\right|\n= \\sigma.\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( \\mu + \\sigma \\cdot y \\right)\n    \\cdot \\sigma.\n\\]\nFor an example of how to code this in Stan, see section Affinely Transformed Real.\n\n\n\n\nFor some modeling tasks, a vector-valued random variable \\(X\\) is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.\nIn constraint terms, an ordered \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies\n\\[\nx_k &lt; x_{k+1}\n\\]\nfor \\(k \\in \\{ 1, \\ldots, K-1 \\}\\).\n\n\nStan’s transform follows the constraint directly. It maps an increasing vector \\(x \\in \\mathbb{R}^{K}\\) to an unconstrained vector \\(y \\in\n\\mathbb{R}^K\\) by setting\n\\[\ny_k\n=\n\\left\\{\n\\begin{array}{ll}\nx_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\log \\left( x_{k} - x_{k-1} \\right) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform for an unconstrained \\(y \\in \\mathbb{R}^K\\) to an ordered sequence \\(x \\in \\mathbb{R}^K\\) is defined by the recursion\n\\[\nx_k\n=\n\\left\\{\n\\begin{array}{ll}\ny_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\nx_{k-1} + \\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\\(x_k\\) can also be expressed iteratively as\n\\[\nx_k = y_1 + \\sum_{k'=2}^k \\exp(y_{k'}).\n\\]\n\n\n\nThe Jacobian of the inverse transform \\(f^{-1}\\) is lower triangular, with diagonal elements for \\(1 \\leq k \\leq K\\) of\n\\[\nJ_{k,k} =\n\\left\\{\n\\begin{array}{ll}\n1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\nBecause \\(J\\) is triangular, the absolute Jacobian determinant is\n\\[\n\\left| \\, \\det \\, J \\, \\right|\n\\ = \\\n\\left| \\, \\prod_{k=1}^K J_{k,k} \\, \\right|\n\\ = \\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\nPutting this all together, if \\(p_X\\) is the density of \\(X\\), then the transformed variable \\(Y\\) has density \\(p_Y\\) given by\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n\\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\n\n\n\n\nPlaceholder.\nThe positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.\n\n\n\nStan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length \\(N\\) with real values and the sum of the vector equals zero. The sum-to-zero matrix is an \\(N \\times M\\) matrix where both the rows and columns sum-to-zero.\nStan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix \\(H\\) such that \\(H\\in\\mathbb R^{N\\times (N-1)}\\) and \\(H^{\\mathsf T}H=I_{N-1},\\;H^{\\mathsf T}\\mathbf 1=0\\). The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,(Seyboldt 2024). Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.\nIn many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:\n\\[\n  \\sigma^2 \\begin{pmatrix}\n  1-\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots&-\\tfrac{1}{N} \\\\\n  -\\tfrac{1}{N} & 1-\\tfrac{1}{N} & \\cdots & -\\tfrac{1}{N} \\\\\n  \\vdots & \\vdots & \\ddots & \\vdots \\\\\n  -\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots &1-\\tfrac{1}{N}\n  \\end{pmatrix}.\n\\]\nThe marginal standard deviation no longer corresponds to \\(\\sigma\\) but is \\(\\sigma \\sqrt{1-\\tfrac{1}{N}}\\). The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of \\(\\sqrt{1-\\tfrac{1}{N}}\\) to adjust the variance to the intended \\(\\sigma\\):\nsum_to_zero_vector[N] x;\nx ~ normal(0, sqrt(N %/% (N - 1)));\nWhen \\(\\sigma\\) is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let \\(y\\) be the unconstrained \\(N - 1\\) vector and \\(y\\) is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of \\(\\sigma\\) is given by \\[\nx = \\underbrace{\\sigma\\sqrt{\\frac{N}{N-1}}\\;H}_{\\text{ size of } N \\times (N - 1)}\\,y.\n\\] This is the classic non-centered model. The crucial detail from above is that the operation of multiplying \\(\\sigma\\) to \\(y\\) is on \\(N - 1\\) dimensions, not \\(N\\) dimensions. When writing the centered model using \\[\nx \\sim \\mathcal N \\bigg(0,\\, \\sigma\\sqrt{\\frac{N}{N-1}} \\bigg),\n\\] we are incrementing the log density by an additional \\(-\\log(\\sigma)\\) and so must increment the log density by\ntarget += log(sigma * sqrt(N * inv(N - 1)));\n\n// or\n// because `sqrt(N * inv(N - 1))` is constant\n\ntarget += log(sigma);\nwhich correctly adjusts for the \\(N - 1\\) free variables.\nMore details of the Helmert matrix are in (Lancaster 1965), for the basic definitions of the isometric log ratio transform see (Egozcue et al. 2003) and Chapter 3 of (Filzmoser, Hron, and Templ 2018) for the pivot coordinate version used here.\n\n\nVectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).\nA sum-to-zero \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies the constraint \\[\n\\sum_{k=1}^K x_k = 0.\n\\]\n\n\n\nThe transform is defined iteratively. Given an \\(x \\in \\mathbb{R}^{N + 1}\\) that sums to zero (i.e., \\(\\sum_{n=1}^{N+1} x_n = 0\\)), the transform proceeds as follows to produce an unconstrained \\(y \\in \\mathbb{R}^N\\). This is mathematically equivalent to pre-multiplying the unconstrained \\(y\\) by an orthogonal standard basis, e.g. constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.\nThe transform is initialized by setting \\[\nS_N = 0\n\\] and \\[\ny_N = -x_{N + 1} \\sqrt{1 + \\frac{1}{N}}.\n\\] The for each \\(n\\) from \\(N - 1\\) down to \\(1\\), let \\[\nw_{n + 1} = \\frac{y_{n + 1}}{\\sqrt{(n + 1)(n + 2)}},\n\\] \\[\nS_n = S_{n + 1} + w_{n + 1},\n\\] and \\[\ny_n = (S_n - x_{n + 1}) \\frac{\\sqrt{n (n + 1)}}{n}.\n\\]\nThis transform is expressed in Stan code as:\n  vector manual_sum_to_zero_jacobian(vector x) {\n    int N = size(x) - 1;\n    vector[N] y;\n    y[N] = -x[N+1] * sqrt(1 + 1. / N);\n    real sum_w = 0;\n    for (n in 1:(N-1)) {\n      int i = N - n;\n      int i_p_1 = i + 1;\n      real w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + 1));\n      sum_w += w;\n      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;\n    }\n    return y;\n  }\n\n\n\nThe inverse transform follows the isometric logratio tranform. It maps an unconstrained vector \\(y \\in \\mathbb{R}^N\\) to a sum-to-zero vector \\(x \\in \\mathbb{R}^{N + 1}\\) such that \\[\n\\sum_{n=1}^{N + 1} x_n = 0.\n\\] The values are defined inductively, starting with \\[\nx_1 = \\sum_{n=1}^N \\frac{y_n}{\\sqrt{n (n + 1)}}\n\\] and then setting \\[\nx_{n + 1} = \\sum_{i = n + 1}^N \\frac{y_i}{\\sqrt{i (i + 1)}}\n- n \\cdot \\frac{y_n}{\\sqrt{n (n + 1)}}.\n\\] for \\(n \\in 1{:}N\\).\nThe definition is such that \\[\n\\sum_{n = 1}^{N + 1} x_n = 0\n\\] by construction, because each of the terms added to \\(x_{n}\\) is then subtracted from \\(x_{n + 1}\\) the number of times it shows up in earlier terms.\n\n\n\nThe inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.\nThe sum-to-zero inverse transform is expressed within Stan as:\n  vector manual_sum_to_zero_inverse_jacobian(vector y) {\n    int N = num_elements(y);\n    vector[N + 1] x = zeros_vector(N + 1);\n    real sum_w = 0;\n    for (n in 1:N) {\n      int i = N - n + 1;\n      real w = y[i] * inv_sqrt(i * (i + 1));\n      sum_w += w;\n      x[i] += sum_w;\n      x[i + 1] -= w * i;\n    }\n    return x;\n  }\nNote that there is no target += increment because the Jacobian is zero.\n\n\n\nThe matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any \\(N\\)-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.\nLet the unconstrained matrix be \\[\n\\mathcal Y \\in \\mathbb R^{n_1 \\times n_2}\n\\] and the zero sum vector transform as \\[\n\\mathbf z = \\mathcal C_n(\\mathbf y)\\;=\\;\n\\begin{bmatrix}H_n\\\\[2pt]-\\mathbf 1_{1\\times d}\\end{bmatrix}\\mathbf y\n\\;\\in\\mathbb R^{n+1},\n\\] where \\(H_n\\in\\mathbb R^{n \\times n}\\) is the orthogonal Helmert matrix and satisfies \\(\\mathbf 1_{1\\times d}A_d^{\\!\\top}=0\\).\nApplying \\(C_n\\) to each slice results in\n\\[\n\\mathcal Z \\;=\\;\n\\mathcal X\n\\times_1 \\bigl[\\mathcal C_{n_1}\\bigr]\n\\times_2 \\bigl[\\mathcal C_{n_2}\\bigr]\n\\] where \\[\n\\mathcal Z \\in\n\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}.\n\\]\nBecause each \\(\\mathbb R^{d_1\\times\\cdots\\times d_N}\\) is invertible on the \\(\\mathbf 1^\\perp\\) subspace the composite map applied to \\(\\mathcal Z\\) is a linear bijection between \\(\\mathbb R^{n_1\\times n_2}\\) and the codomain \\(\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}\\).\n\n\n\n\nVariables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).\nThe unit \\(K\\)-simplex is the set of points \\(x \\in \\mathbb{R}^K\\) such that for \\(1 \\leq k \\leq K\\),\n\\[\nx_k &gt; 0,\n\\]\nand\n\\[\n\\sum_{k=1}^K x_k = 1.\n\\]\nAn alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values \\((0,1)\\), and \\((1,0)\\) and the unit 2-simplex is the line connecting these two points; values such as \\((0.3,0.7)\\) and \\((0.99,0.01)\\) lie on the line. In 3-dimensions, the basis is \\((0,0,1)\\), \\((0,1,0)\\) and \\((1,0,0)\\) and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include \\((0.5,0.5,0)\\), \\((0.2,0.7,0.1)\\) and all other triplets of non-negative values summing to 1.\nAs these examples illustrate, the simplex always picks out a subspace of \\(K-1\\) dimensions from \\(\\mathbb{R}^K\\). Therefore a point \\(x\\) in the \\(K\\)-simplex is fully determined by its first \\(K-1\\) elements \\(x_1, x_2,\n\\ldots, x_{K-1}\\), with\n\\[\nx_K = 1 - \\sum_{k=1}^{K-1} x_k.\n\\]\n\n\nThe length-\\(K\\) unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length \\(K\\).\nLet \\(y\\) represent the unconstrained \\(K - 1\\) values in \\((-\\infty, \\infty)\\). The intermediate sum-to-zero vector \\(z = \\text{sum\\_to\\_zero\\_transform}(y)\\) is length \\(K\\). The unit simplex is then given by \\[\nx_i = \\text{softmax}(z) = \\frac{\\exp(z_i)}{\\sum_{i = 1}^K \\exp(z_i)}\n\\]\nThe sum-to-zero vector transform is described in further detail at the sum-to-zero vector section of the Reference Manual.\n\n\n\n\n\n\nNote\n\n\n\nAll versions of Stan pre-2.37 used the stick-breaking transform. This is documented at Stan 2.36 Reference Manual: Simplex Transform.\n\n\n\n\nThe Jacobian \\(J\\) of the inverse unit-simplex transform is found by restricting \\(J\\) to the subspace spanned by the sum-to-zero vector \\(z\\). The Jacobian is given as the \\((K - 1) \\times (K - 1)\\) matrix \\(J\\) where\n\\[\nJ_{ij} = \\frac{\\partial x_i}{\\partial z_j} =\n\\frac{\\partial}{\\partial z_i} \\left( \\frac{\\exp(z_i)}{{\\sum_{i = 1}^K \\exp(z_i)}} \\right)\n\\] and \\(i,j \\in 1, \\ldots, K - 1\\).\nThe diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification\n\\[\nJ_{ij} =\n\\begin{cases}\nx_i (1 - x_i), & \\text{if } i = j, \\\\\n-x_i x_j, & \\text{if } i \\neq j.\n\\end{cases}\n\\]\nIn matrix form this can be expressed as\n\\[\nJ = \\text{diag}(x) - x x^\\top\n\\]\nThe determinant of this matrix can be found using the Matrix Determinant Lemma:\n\\[\n\\det\\bigl(A + u v^{\\top}\\bigr)\n=\n\\det(A)\\,\\bigl(1 + v^{\\top}A^{-1}u\\bigr).\n\\]\nHere,\n\\[\nA \\;=\\; \\operatorname{diag}(x_{1},\\ldots, x_{K-1}),\n\\quad\nu \\;=\\; -\\bigl(x_1,\\ldots, x_{K-1}\\bigr)^{\\!\\top},\n\\quad\nv \\;=\\; \\bigl(x_{1}, \\ldots, x_{K-1}\\bigr)^{\\!\\top}.\n\\] Therefore,\n\\[\n\\begin{aligned}\n\\det(J)\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_i \\bigg)\n\\bigg(1 + (x_{1},\\ldots, x_{K-1})\\,\\mathrm{diag}\\bigl(x_{1}^{-1},\\ldots,x_{K-1}^{-1}\\bigr)\\,\n\\big(-x_{1},\\ldots,-x_{K-1}\\big)^{\\top}\n\\bigg) \\\\\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg)\n\\bigg(1 - \\sum_{i=1}^{K-1} x_{i}\\bigg)\n=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg) x_{K} \\\\\n&=\n\\prod_{i=1}^{K} x_{i}.\n\\end{aligned}\n\\]\n\n\n\n\nThe transform \\(Y = f(X)\\) can be derived by reversing the stages of the inverse transform,\n\\[\ny_k\n= H^\\top \\bigg(\\log(x_k)\n- \\frac{1}{K}\\sum_{i=1}^K\\log(x_i) \\bigg)\n.\n\\]\nThe matrix \\(H\\) is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.\n\n\n\n\nThe column_stochastic_matrix[N, M] and row_stochastic_matrix[M, N] type in Stan represents an \\(N \\times M\\) matrix where each column (row) is a unit simplex of dimension \\(N\\). In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.\n\n\nA column stochastic matrix \\(X \\in \\mathbb{R}^{N \\times M}\\) is defined such that each column is a simplex. For column \\(m\\) (where \\(1 \\leq m \\leq M\\)):\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{n=1}^N X_{n, m} = 1.\n\\]\nA row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e. the rows of the matrix are simplexes)\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{m=1}^N X_{n, m} = 1.\n\\]\nThis definition ensures that each column (row) of the matrix \\(X\\) lies on the \\(N-1\\) dimensional unit simplex, similar to the simplex[N] type, but extended across multiple columns(rows).\n\n\n\nFor the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).\n\n\n\nThe Jacobian determinant of the inverse transform for each column \\(m\\) in the matrix is given by the product of the diagonal entries \\(J_{n, m}\\) of the lower-triangular Jacobian matrix. This determinant is calculated as:\n\\[\n\\left| \\det J_m \\right| = \\prod_{n=1}^{N-1} \\left( z_{n, m} (1 - z_{n, m}) \\left( 1 - \\sum_{n'=1}^{n-1} X_{n', m} \\right) \\right).\n\\]\nThus, the overall Jacobian determinant for the entire column_stochastic_matrix and row_stochastic_matrix is the product of the determinants for each column (row):\n\\[\n\\left| \\det J \\right| = \\prod_{m=1}^{M} \\left| \\det J_m \\right|.\n\\]\n\n\n\nFor the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).\n\n\n\n\nAn \\(n\\)-dimensional vector \\(x \\in \\mathbb{R}^n\\) is said to be a unit vector if it has unit Euclidean length, so that\n\\[\n\\Vert x \\Vert\n\\ = \\ \\sqrt{x^{\\top}\\,x}\n\\ = \\ \\sqrt{x_1^2 + x_2^2 + \\cdots + x_n^2}\n\\ = \\ 1\\ .\n\\]\n\n\nStan divides an unconstrained vector \\(y \\in \\mathbb{R}^{n}\\) by its norm, \\(\\Vert y \\Vert = \\sqrt{y^\\top y}\\), to obtain a unit vector \\(x\\),\n\\[\nx = \\frac{y}{\\Vert y \\Vert}.\n\\]\nTo generate a unit vector, Stan generates points at random in \\(\\mathbb{R}^n\\) with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. Muller (1959) showed this generates points uniformly at random on \\(S^{n-1}\\). That is, if we draw \\(y_n \\sim \\mathsf{Normal}(0, 1)\\) for \\(n \\in 1{:}n\\), then \\(x = \\frac{y}{\\Vert y \\Vert}\\) has a uniform distribution over \\(S^{n-1}\\). This allows us to use an \\(n\\)-dimensional basis for \\(S^{n-1}\\) that preserves local neighborhoods in that points that are close to each other in \\(\\mathbb{R}^n\\) map to points near each other in \\(S^{n-1}\\). The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in \\(\\mathbb{R}^n\\) that map to identical points in \\(S^{n-1}\\).\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nThe Jacobian matrix relating the input vector \\(y\\) to the output vector \\(x\\) is singular because \\(x^\\top x = 1\\) for any non-zero input vector \\(y\\). Thus, there technically is no unique transformation from \\(x\\) to \\(y\\). To circumvent this issue, let \\(r = \\sqrt{y^\\top y}\\) so that \\(y = r\nx\\). The transformation from \\(\\left(r, x_{-n}\\right)\\) to \\(y\\) is well-defined but \\(r\\) is arbitrary, so we set \\(r = 1\\). In this case, the determinant of the Jacobian is proportional to \\(e^{-\\frac{1}{2} y^\\top y}\\), which is the kernel of a standard multivariate normal distribution with \\(n\\) independent dimensions.\n\n\n\n\nA \\(K \\times K\\) correlation matrix \\(x\\) must be symmetric, so that\n\\[\nx_{k,k'} = x_{k',k}\n\\]\nfor all \\(k,k' \\in \\{ 1, \\ldots, K \\}\\), it must have a unit diagonal, so that\n\\[\nx_{k,k} = 1\n\\]\nfor all \\(k \\in \\{ 1, \\ldots, K \\}\\), and it must be positive definite, so that for every non-zero \\(K\\)-vector \\(a\\),\n\\[\na^{\\top} x a &gt; 0.\n\\]\nThe number of free parameters required to specify a \\(K \\times K\\) correlation matrix is \\(\\binom{K}{2}\\).\nThere is more than one way to map from \\(\\binom{K}{2}\\) unconstrained parameters to a \\(K \\times K\\) correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform Lewandowski, Kurowicka, and Joe (2009).\n\n\nIt is easiest to specify the inverse, going from its \\(\\binom{K}{2}\\) parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose \\(y\\) is a vector containing \\(\\binom{K}{2}\\) unconstrained values. These are first transformed via the bijective function \\(\\tanh : \\mathbb{R} \\rightarrow\n(-1, 1)\\)\n\\[\n\\tanh y = \\frac{\\exp(2y) - 1}{\\exp(2y) + 1}.\n\\]\nThen, define a \\(K \\times K\\) matrix \\(z\\), the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the \\(4 \\times 4\\) case, there are \\(\\binom{4}{2}\\) values arranged as\n\\[\nz\n=\n\\left[\n\\begin{array}{cccc}\n1 & \\tanh y_1 & \\tanh y_2 & \\tanh y_4\n\\\\\n0 & 1 & \\tanh y_3 & \\tanh y_5\n\\\\\n0 & 0 & 1 & \\tanh y_6\n\\\\\n0 & 0 & 0 & 1\n\\end{array}\n\\right]\n.\n\\]\nLewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array \\(z\\) to a correlation matrix \\(x\\). The entry \\(z_{i,j}\\) for \\(i &lt;\nj\\) is interpreted as the canonical partial correlation (CPC) between \\(i\\) and \\(j\\), which is the correlation between \\(i\\)’s residuals and \\(j\\)’s residuals when both \\(i\\) and \\(j\\) are regressed on all variables \\(i'\\) such that \\(i'&lt; i\\). In the case of \\(i=1\\), there are no earlier variables, so \\(z_{1,j}\\) is just the Pearson correlation between \\(i\\) and \\(j\\).\nIn Stan, the LKJ transform is reformulated in terms of a Cholesky factor \\(w\\) of the final correlation matrix, defined for \\(1 \\leq i,j \\leq K\\) by\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\n\\prod_{i'=1}^{i - 1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{if } 1 &lt; i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\\\nz_{i,j} \\, \\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThis does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n\\frac{z_{i,j}}{z_{i-1,j}} \\ w_{i-1,j} \\left( 1 - z_{i-1,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i \\leq j.\n\\end{array}\n\\right.\n\\]\nGiven the upper-triangular Cholesky factor \\(w\\), the final correlation matrix is\n\\[\nx = w^{\\top} w.\n\\]\nLewandowski, Kurowicka, and Joe (2009) show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as\n\\[\n\\mbox{det} \\, x = \\prod_{i=1}^{K-1} \\ \\prod_{j=i+1}^K \\ (1 - z_{i,j}^2)\n= \\prod_{1 \\leq i &lt; j \\leq K} (1 - z_{i,j}^2),\n\\]\n\n\n\nFrom the inverse of equation 11 in (Lewandowski, Kurowicka, and Joe 2009), the absolute Jacobian determinant is\n\\[\n\\sqrt{\\prod_{i=1}^{K-1}\\prod_{j=i+1}^K \\left(1-z_{i,j}^2\\right)^{K-i-1}} \\\n\\times \\prod_{i=1}^{K-1}\\prod_{j=i+1}^K\n\\frac{\\partial z_{i,j}}{\\partial y_{i,j}}\n\\]\n\n\n\nThe correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.\nStarting with a correlation matrix \\(x\\), the first step is to find the unique upper triangular \\(w\\) such that \\(x = w w^{\\top}\\). Because \\(x\\) is positive definite, this can be done by applying the Cholesky decomposition,\n\\[\nw = \\mbox{chol}(x).\n\\]\nThe next step from the Cholesky factor \\(w\\) back to the array \\(z\\) of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of \\(w\\), which when inverted yields\n\\[\nz_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i \\leq j,\n\\\\\nw_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n{w_{i,j}}\n\\\n\\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,j}^2 \\right)^{-1/2}\n& \\mbox{if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThe final stage of the transform reverses the hyperbolic tangent transform, which is defined by\n\\[\ny = \\tanh^{-1} z = \\frac{1}{2} \\log \\left( \\frac{1 + z}{1 - z} \\right).\n\\]\nThe inverse hyperbolic tangent function, \\(\\tanh^{-1}\\), is also called the Fisher transformation.\n\n\n\n\nA \\(K \\times K\\) matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires \\(K + \\binom{K}{2}\\) free parameters to specify a \\(K \\times K\\) covariance matrix.\n\n\nStan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.1\nIf \\(x\\) is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix \\(z =\n\\mathrm{chol}(x)\\) with positive diagonal entries, called a Cholesky factor, such that\n\\[\nx = z \\, z^{\\top}.\n\\]\nThe off-diagonal entries of the Cholesky factor \\(z\\) are unconstrained, but the diagonal entries \\(z_{k,k}\\) must be positive for \\(1 \\leq k\n\\leq K\\).\nTo complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix \\(y\\) defined by\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log z_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nz_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular \\(K \\times K\\) matrix \\(y\\), the first step is to recover the intermediate matrix \\(z\\) by reversing the log transform,\n\\[\nz_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe covariance matrix \\(x\\) is recovered from its Cholesky factor \\(z\\) by taking\n\\[\nx = z \\, z^{\\top}.\n\\]\n\n\n\nThe Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix \\(y\\) to matrix \\(z\\) with positive diagonals and the product transform from the Cholesky factor \\(z\\) to \\(x\\).\nThe transform from unconstrained \\(y\\) to Cholesky factor \\(z\\) has a diagonal Jacobian matrix, the absolute determinant of which is thus\n\\[\n\\prod_{k=1}^K  \\frac{\\partial}{\\partial_{y_{k,k}}} \\, \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K z_{k,k}.\n\\]\nThe Jacobian matrix of the second transform from the Cholesky factor \\(z\\) to the covariance matrix \\(x\\) is also triangular, with diagonal entries corresponding to pairs \\((m,n)\\) with \\(m \\geq n\\), defined by\n\\[\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( z \\, z^{\\top} \\right)_{m,n}\n\\ = \\\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( \\sum_{k=1}^K z_{m,k} \\, z_{n,k} \\right)\n\\ = \\\n\\left\\{\n\\begin{array}{cl}\n2 \\, z_{n,n} & \\mbox{if } m = n \\mbox{ and }\n\\\\\nz_{n,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe absolute Jacobian determinant of the second transform is thus\n\\[\n2^{K} \\ \\prod_{m = 1}^{K} \\ \\prod_{n=1}^{m} z_{n,n}\n\\ = \\\n\\prod_{n=1}^K \\ \\prod_{m=n}^K z_{n,n}\n\\ = \\\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}.\n\\]\nFinally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular \\(y\\) to a symmetric, positive definite matrix \\(x\\) is the product of the Jacobian determinants of the exponentiation and product transforms,\n\\[\n\\left( \\prod_{k=1}^K z_{k,k} \\right)\n\\left(\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}\n\\right)\n\\ = \\\n2^K\n\\, \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\nLet \\(f^{-1}\\) be the inverse transform from a \\(K + \\binom{K}{2}\\)-vector \\(y\\) to the \\(K \\times K\\) covariance matrix \\(x\\). A density function \\(p_X(x)\\) defined on \\(K \\times K\\) covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(K + \\binom{K}{2}\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\ 2^K \\ \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\n\n\n\n\nAn \\(M \\times M\\) covariance matrix \\(\\Sigma\\) can be Cholesky factored to a lower triangular matrix \\(L\\) such that \\(L\\,L^{\\top} = \\Sigma\\). If \\(\\Sigma\\) is positive definite, then \\(L\\) will be \\(M \\times M\\). If \\(\\Sigma\\) is only positive semi-definite, then \\(L\\) will be \\(M \\times N\\), with \\(N &lt; M\\).\nA matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and \\(M \\geq\nN\\). A matrix satisfying these conditions ensures that \\(L \\,\nL^{\\top}\\) is positive semi-definite if \\(M &gt; N\\) and positive definite if \\(M = N\\).\nA Cholesky factor of a covariance matrix requires \\(N + \\binom{N}{2} +\n(M - N)N\\) unconstrained parameters.\n\n\nStan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose \\(x\\) is an \\(M \\times N\\) Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log x_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nx_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform need only invert the logarithm with an exponentiation. If \\(y\\) is the unconstrained matrix representation, then the elements of the constrained matrix \\(x\\) is defined by\n\\[\nx_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe transform has a diagonal Jacobian matrix, the absolute determinant of which is\n\\[\n\\prod_{n=1}^N  \\frac{\\partial}{\\partial_{y_{n,n}}} \\, \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N x_{n,n}.\n\\]\nLet \\(x = f^{-1}(y)\\) be the inverse transform from a \\(N + \\binom{N}{2}\n+ (M - N)N\\) vector to an \\(M \\times N\\) Cholesky factor for a covariance matrix \\(x\\) defined in the previous section. A density function \\(p_X(x)\\) defined on \\(M \\times N\\) Cholesky factors of covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(N + \\binom{N}{2}\n+ (M - N)N\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\prod_{N=1}^N x_{n,n}.\n\\]\n\n\n\n\nA \\(K \\times K\\) correlation matrix \\(\\Omega\\) is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a \\(K \\times K\\) lower-triangular matrix \\(L\\) with positive diagonal elements such that \\(\\Omega = L\\,L^{\\top}\\). Because the correlation matrix has a unit diagonal,\n\\[\n\\Omega_{k,k} = L_k\\,L_k^{\\top} = 1,\n\\]\neach row vector \\(L_k\\) of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of \\(L\\) to be calculated from the off-diagonal elements, so that a Cholesky factor for a \\(K \\times K\\) correlation matrix requires only \\(\\binom{K}{2}\\) unconstrained parameters.\n\n\nIt is easiest to start with the inverse transform from the \\(\\binom{K}{2}\\) unconstrained parameters \\(y\\) to the \\(K \\times K\\) lower-triangular Cholesky factor \\(x\\). The inverse transform is based on the hyperbolic tangent function, \\(\\tanh\\), which satisfies \\(\\tanh(x) \\in (-1,1)\\). Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the section on correlation matrix transforms for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.\nSuppose \\(y\\) is a vector of \\(\\binom{K}{2}\\) unconstrained values. Let \\(z\\) be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the \\(3 \\times 3\\) case,\n\\[\nz =\n\\left[\n\\begin{array}{ccc}\n0 & 0 & 0\n\\\\\n\\tanh y_1 & 0 & 0\n\\\\\n\\tanh y_2 & \\tanh y_3 & 0\n\\end{array}\n\\right]\n\\]\nThe matrix \\(z\\), with entries in the range \\((-1, 1)\\), is then transformed to the Cholesky factor \\(x\\), by taking2\n\\[\nx_{i,j}\n=\n\\left\\{\n\\begin{array}{lll}\n0 & \\mbox{ if } i &lt; j & \\mbox{ [above diagonal]}\n\\\\\n\\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i = j & \\mbox{ [on diagonal]}\n\\\\\nz_{i,j} \\ \\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i &gt; j & \\mbox{ [below diagonal]}\n\\end{array}\n\\right.\n\\]\nIn the \\(3 \\times 3\\) case, this yields\n\\[\nx =\n\\left[\n\\begin{array}{ccc}\n1 & 0 & 0\n\\\\\nz_{2,1} & \\sqrt{1 - x_{2,1}^2} & 0\n\\\\\nz_{3,1} & z_{3,2} \\sqrt{1 - x_{3,1}^2}\n        & \\sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}\n\\end{array}\n\\right],\n\\]\nwhere the \\(z_{i,j} \\in (-1,1)\\) are the \\(\\tanh\\)-transformed \\(y\\).\nThe approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at \\(j=1\\), each below-diagonal entry \\(x_{i,j}\\) is determined by the (signed) fraction \\(z_{i,j}\\) of the remaining length for the row that it consumes. The diagonal entries \\(x_{i,i}\\) get any leftover length from earlier entries in their row. The above-diagonal entries are zero.\n\n\n\nSuppose \\(x\\) is a \\(K \\times K\\) Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values \\(z\\) from \\(x\\),\n\\[\nz_{i,j} = \\frac{x_{i,j}}{\\sqrt{1 - \\sum_{j' &lt; j}x_{i,j'}^2}}.\n\\]\nThe mapping from the resulting \\(z\\) to \\(y\\) inverts \\(\\tanh\\),\n\\[\ny\n\\ = \\\n\\tanh^{-1} z\n\\ = \\\n\\frac{1}{2} \\left( \\log (1 + z) - \\log (1 - z) \\right).\n\\]\n\n\n\nThe Jacobian of the full transform is the product of the Jacobians of its component transforms.\nFirst, for the inverse transform \\(z = \\tanh y\\), the derivative is\n\\[\n\\frac{d}{dy} \\tanh y = \\frac{1}{(\\cosh y)^2}.\n\\]\nSecond, for the inverse transform of \\(z\\) to \\(x\\), the resulting Jacobian matrix \\(J\\) is of dimension \\(\\binom{K}{2} \\times\n\\binom{K}{2}\\), with indexes \\((i,j)\\) for \\((i &gt; j)\\). The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each \\((i,j)\\) pair,\n\\[\n\\left| \\, \\mbox{det} \\, J \\, \\right|\n  \\ = \\ \\prod_{i &gt; j} \\left| \\frac{d}{dz_{i,j}} x_{i,j} \\right|,\n\\]\nwhere\n\\[\n\\frac{d}{dz_{i,j}} x_{i,j}\n= \\sqrt{1 - \\sum_{j' &lt; j} x^2_{i,j'}}.\n\\]\nSo the combined density for unconstrained \\(y\\) is\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n  \\ \\\n  \\prod_{n &lt; \\binom{K}{2}} \\frac{1}{(\\cosh y)^2}\n  \\ \\\n  \\prod_{i &gt; j} \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2\n  \\right)^{1/2},\n\\]\nwhere \\(x = f^{-1}(y)\\) is used for notational convenience. The log Jacobian determinant of the complete inverse transform \\(x = f^{-1}(y)\\) is given by\n\\[\n\\log \\left| \\, \\det J \\, \\right|\n=\n-2 \\sum_{n \\leq \\binom{K}{2}}\n\\log \\cosh y\n\\\n+\n\\\n\\frac{1}{2} \\\n\\sum_{i &gt; j}\n\\log \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2 \\right)\n.\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#limitations-due-to-finite-accuracy-presentation",
+    "href": "reference-manual/transforms.html#limitations-due-to-finite-accuracy-presentation",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "In this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.\nStan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with lower=0. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about floating point arithmetic in the Stan User’s Guide.\nThese issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the CmdStan Command-Line Interface Overview.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#change-of-variables.section",
+    "href": "reference-manual/transforms.html#change-of-variables.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "The support of a random variable \\(X\\) with density \\(p_X(x)\\) is that subset of values for which it has non-zero density,\n\\[\n\\mathrm{supp}(X) = \\{ x | p_X(x) &gt; 0 \\}.\n\\]\nIf \\(f\\) is a total function defined on the support of \\(X\\), then \\(Y =\nf(X)\\) is a new random variable. This section shows how to compute the probability density function of \\(Y\\) for well-behaved transforms \\(f\\). The rest of the chapter details the transforms used by Stan.\n\n\nSuppose \\(X\\) is one dimensional and \\(f: \\mathrm{supp}(X) \\rightarrow\n\\mathbb{R}\\) is a one-to-one, monotonic function with a differentiable inverse \\(f^{-1}\\). Then the density of \\(Y\\) is given by\n\\[\np_Y(y) = p_X(f^{-1}(y))\n         \\,\n         \\left| \\, \\frac{d}{dy} f^{-1}(y)\\, \\right|.\n\\]\nThe absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.\n\n\n\nThe multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.\nSuppose \\(X\\) is a \\(K\\)-dimensional random variable with probability density function \\(p_X(x)\\). A new random variable \\(Y = f(X)\\) may be defined by transforming \\(X\\) with a suitably well-behaved function \\(f\\). It suffices for what follows to note that if \\(f\\) is one-to-one and its inverse \\(f^{-1}\\) has a well-defined Jacobian, then the density of \\(Y\\) is\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\, \\left| \\, \\det \\, J_{f^{-1}}(y) \\, \\right|,\n\\]\nwhere \\(\\det{}\\) is the matrix determinant operation and \\(J_{f^{-1}}(y)\\) is the Jacobian matrix of \\(f^{-1}\\) evaluated at \\(y\\). Taking \\(x =\nf^{-1}(y)\\), the Jacobian matrix is defined by\n\\[\nJ_{f^{-1}}(y) =\n\\left[\n\\begin{array}{ccc}\\displaystyle\n\\frac{\\partial x_1}{\\partial y_1}\n& \\cdots\n& \\displaystyle \\frac{\\partial x_1}{\\partial y_{K}}\n\\\\\n\\vdots & \\vdots & \\vdots\n\\\\\n\\displaystyle\\frac{\\partial x_{K}}{\\partial y_1}\n& \\cdots\n& \\displaystyle\\frac{\\partial x_{K}}{\\partial y_{K}}\n\\end{array}\n\\right].\n\\]\nIf the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,\n\\[\n\\det \\, J_{f^{-1}}(y)\n= \\prod_{k=1}^K \\frac{\\partial x_k}{\\partial y_k}.\n\\]\nTriangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#lower-bound-transform.section",
+    "href": "reference-manual/transforms.html#lower-bound-transform.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Stan uses a logarithmic transform for lower and upper bounds.\n\n\nIf a variable \\(X\\) is declared to have lower bound \\(a\\), it is transformed to an unbounded variable \\(Y\\), where\n\\[\nY = \\log(X - a).\n\\]\n\n\n\nThe inverse of the lower-bound transform maps an unbounded variable \\(Y\\) to a variable \\(X\\) that is bounded below by \\(a\\) by\n\\[\nX = \\exp(Y) + a.\n\\]\n\n\n\nThe absolute derivative of the inverse transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( \\exp(y) + a \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, given the density \\(p_X\\) of \\(X\\), the density of \\(Y\\) is\n\\[\np_Y(y)\n= p_X\\!\\left( \\exp(y) + a \\right) \\cdot \\exp(y).\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#upper-bounded-scalar",
+    "href": "reference-manual/transforms.html#upper-bounded-scalar",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Stan uses a negated logarithmic transform for upper bounds.\n\n\nIf a variable \\(X\\) is declared to have an upper bound \\(b\\), it is transformed to the unbounded variable \\(Y\\) by\n\\[\nY = \\log(b - X).\n\\]\n\n\n\nThe inverse of the upper bound transform converts the unbounded variable \\(Y\\) to the variable \\(X\\) bounded above by \\(b\\) through\n\\[\nX = b - \\exp(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse of the upper bound transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( b - \\exp(y) \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, the density of the unconstrained variable \\(Y\\) is defined in terms of the density of the variable \\(X\\) with an upper bound of \\(b\\) by\n\\[\np_Y(y)\n=   p_X \\!\\left( b - \\exp(y) \\right) \\cdot \\exp(y).\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#logit-transform-jacobian.section",
+    "href": "reference-manual/transforms.html#logit-transform-jacobian.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "For lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.\n\n\nThe log-odds function is defined for \\(u \\in (0,1)\\) by\n\\[\n\\mathrm{logit}(u) = \\log \\frac{u}{1 - u}.\n\\]\nThe inverse of the log odds function is the logistic sigmoid, defined for \\(v \\in (-\\infty,\\infty)\\) by\n\\[\n\\mathrm{logit}^{-1}(v) = \\frac{1}{1 + \\exp(-v)}.\n\\]\nThe derivative of the logistic sigmoid is\n\\[\n\\frac{d}{dy} \\mathrm{logit}^{-1}(y)\n= \\mathrm{logit}^{-1}(y) \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\n\n\n\nFor variables constrained to be in the open interval \\((a, b)\\), Stan uses a scaled and translated log-odds transform. If variable \\(X\\) is declared to have lower bound \\(a\\) and upper bound \\(b\\), then it is transformed to a new variable \\(Y\\), where\n\\[\nY = \\mathrm{logit} \\left( \\frac{X - a}{b - a} \\right).\n\\]\n\n\n\nThe inverse of this transform is\n\\[\nX = a + (b - a) \\cdot \\mathrm{logit}^{-1}(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse transform is given by\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      a + (b - a) \\cdot \\mathrm{logit}^{-1}(y)\n    \\right)\n  \\right|\n= (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( a + (b - a) \\cdot \\mathrm{logit}^{-1}(y) \\right)\n    \\cdot (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nDespite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, \\(\\mathrm{logit}^{-1}(y)\\) only needs to be evaluated once, so there is only one call to \\(\\exp(-y)\\).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#affinely-transformed-scalar",
+    "href": "reference-manual/transforms.html#affinely-transformed-scalar",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Stan uses an affine transform to be able to specify parameters with a given offset and multiplier.\n\n\nFor variables with expected offset \\(\\mu\\) and/or (positive) multiplier \\(\\sigma\\), Stan uses an affine transform. Such a variable \\(X\\) is transformed to a new variable \\(Y\\), where\n\\[\nY = \\frac{X - \\mu}{\\sigma}.\n\\]\nThe default value for the offset \\(\\mu\\) is \\(0\\) and for the multiplier \\(\\sigma\\) is \\(1\\) in case not both are specified.\n\n\n\nThe inverse of this transform is\n\\[\nX =  \\mu + \\sigma \\cdot Y.\n\\]\n\n\n\nThe absolute derivative of the affine inverse transform is\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      \\mu + \\sigma \\cdot y\n    \\right)\n  \\right|\n= \\sigma.\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( \\mu + \\sigma \\cdot y \\right)\n    \\cdot \\sigma.\n\\]\nFor an example of how to code this in Stan, see section Affinely Transformed Real.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#ordered-vector",
+    "href": "reference-manual/transforms.html#ordered-vector",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "For some modeling tasks, a vector-valued random variable \\(X\\) is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.\nIn constraint terms, an ordered \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies\n\\[\nx_k &lt; x_{k+1}\n\\]\nfor \\(k \\in \\{ 1, \\ldots, K-1 \\}\\).\n\n\nStan’s transform follows the constraint directly. It maps an increasing vector \\(x \\in \\mathbb{R}^{K}\\) to an unconstrained vector \\(y \\in\n\\mathbb{R}^K\\) by setting\n\\[\ny_k\n=\n\\left\\{\n\\begin{array}{ll}\nx_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\log \\left( x_{k} - x_{k-1} \\right) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform for an unconstrained \\(y \\in \\mathbb{R}^K\\) to an ordered sequence \\(x \\in \\mathbb{R}^K\\) is defined by the recursion\n\\[\nx_k\n=\n\\left\\{\n\\begin{array}{ll}\ny_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\nx_{k-1} + \\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\\(x_k\\) can also be expressed iteratively as\n\\[\nx_k = y_1 + \\sum_{k'=2}^k \\exp(y_{k'}).\n\\]\n\n\n\nThe Jacobian of the inverse transform \\(f^{-1}\\) is lower triangular, with diagonal elements for \\(1 \\leq k \\leq K\\) of\n\\[\nJ_{k,k} =\n\\left\\{\n\\begin{array}{ll}\n1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\nBecause \\(J\\) is triangular, the absolute Jacobian determinant is\n\\[\n\\left| \\, \\det \\, J \\, \\right|\n\\ = \\\n\\left| \\, \\prod_{k=1}^K J_{k,k} \\, \\right|\n\\ = \\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\nPutting this all together, if \\(p_X\\) is the density of \\(X\\), then the transformed variable \\(Y\\) has density \\(p_Y\\) given by\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n\\\n\\prod_{k=2}^K \\exp(y_k).\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#positive-ordered-vector",
+    "href": "reference-manual/transforms.html#positive-ordered-vector",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Placeholder.\nThe positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#sum-to-zero-transforms",
+    "href": "reference-manual/transforms.html#sum-to-zero-transforms",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Stan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length \\(N\\) with real values and the sum of the vector equals zero. The sum-to-zero matrix is an \\(N \\times M\\) matrix where both the rows and columns sum-to-zero.\nStan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix \\(H\\) such that \\(H\\in\\mathbb R^{N\\times (N-1)}\\) and \\(H^{\\mathsf T}H=I_{N-1},\\;H^{\\mathsf T}\\mathbf 1=0\\). The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,(Seyboldt 2024). Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.\nIn many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:\n\\[\n  \\sigma^2 \\begin{pmatrix}\n  1-\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots&-\\tfrac{1}{N} \\\\\n  -\\tfrac{1}{N} & 1-\\tfrac{1}{N} & \\cdots & -\\tfrac{1}{N} \\\\\n  \\vdots & \\vdots & \\ddots & \\vdots \\\\\n  -\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots &1-\\tfrac{1}{N}\n  \\end{pmatrix}.\n\\]\nThe marginal standard deviation no longer corresponds to \\(\\sigma\\) but is \\(\\sigma \\sqrt{1-\\tfrac{1}{N}}\\). The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of \\(\\sqrt{1-\\tfrac{1}{N}}\\) to adjust the variance to the intended \\(\\sigma\\):\nsum_to_zero_vector[N] x;\nx ~ normal(0, sqrt(N %/% (N - 1)));\nWhen \\(\\sigma\\) is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let \\(y\\) be the unconstrained \\(N - 1\\) vector and \\(y\\) is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of \\(\\sigma\\) is given by \\[\nx = \\underbrace{\\sigma\\sqrt{\\frac{N}{N-1}}\\;H}_{\\text{ size of } N \\times (N - 1)}\\,y.\n\\] This is the classic non-centered model. The crucial detail from above is that the operation of multiplying \\(\\sigma\\) to \\(y\\) is on \\(N - 1\\) dimensions, not \\(N\\) dimensions. When writing the centered model using \\[\nx \\sim \\mathcal N \\bigg(0,\\, \\sigma\\sqrt{\\frac{N}{N-1}} \\bigg),\n\\] we are incrementing the log density by an additional \\(-\\log(\\sigma)\\) and so must increment the log density by\ntarget += log(sigma * sqrt(N * inv(N - 1)));\n\n// or\n// because `sqrt(N * inv(N - 1))` is constant\n\ntarget += log(sigma);\nwhich correctly adjusts for the \\(N - 1\\) free variables.\nMore details of the Helmert matrix are in (Lancaster 1965), for the basic definitions of the isometric log ratio transform see (Egozcue et al. 2003) and Chapter 3 of (Filzmoser, Hron, and Templ 2018) for the pivot coordinate version used here.\n\n\nVectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).\nA sum-to-zero \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies the constraint \\[\n\\sum_{k=1}^K x_k = 0.\n\\]\n\n\n\nThe transform is defined iteratively. Given an \\(x \\in \\mathbb{R}^{N + 1}\\) that sums to zero (i.e., \\(\\sum_{n=1}^{N+1} x_n = 0\\)), the transform proceeds as follows to produce an unconstrained \\(y \\in \\mathbb{R}^N\\). This is mathematically equivalent to pre-multiplying the unconstrained \\(y\\) by an orthogonal standard basis, e.g. constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.\nThe transform is initialized by setting \\[\nS_N = 0\n\\] and \\[\ny_N = -x_{N + 1} \\sqrt{1 + \\frac{1}{N}}.\n\\] The for each \\(n\\) from \\(N - 1\\) down to \\(1\\), let \\[\nw_{n + 1} = \\frac{y_{n + 1}}{\\sqrt{(n + 1)(n + 2)}},\n\\] \\[\nS_n = S_{n + 1} + w_{n + 1},\n\\] and \\[\ny_n = (S_n - x_{n + 1}) \\frac{\\sqrt{n (n + 1)}}{n}.\n\\]\nThis transform is expressed in Stan code as:\n  vector manual_sum_to_zero_jacobian(vector x) {\n    int N = size(x) - 1;\n    vector[N] y;\n    y[N] = -x[N+1] * sqrt(1 + 1. / N);\n    real sum_w = 0;\n    for (n in 1:(N-1)) {\n      int i = N - n;\n      int i_p_1 = i + 1;\n      real w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + 1));\n      sum_w += w;\n      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;\n    }\n    return y;\n  }\n\n\n\nThe inverse transform follows the isometric logratio tranform. It maps an unconstrained vector \\(y \\in \\mathbb{R}^N\\) to a sum-to-zero vector \\(x \\in \\mathbb{R}^{N + 1}\\) such that \\[\n\\sum_{n=1}^{N + 1} x_n = 0.\n\\] The values are defined inductively, starting with \\[\nx_1 = \\sum_{n=1}^N \\frac{y_n}{\\sqrt{n (n + 1)}}\n\\] and then setting \\[\nx_{n + 1} = \\sum_{i = n + 1}^N \\frac{y_i}{\\sqrt{i (i + 1)}}\n- n \\cdot \\frac{y_n}{\\sqrt{n (n + 1)}}.\n\\] for \\(n \\in 1{:}N\\).\nThe definition is such that \\[\n\\sum_{n = 1}^{N + 1} x_n = 0\n\\] by construction, because each of the terms added to \\(x_{n}\\) is then subtracted from \\(x_{n + 1}\\) the number of times it shows up in earlier terms.\n\n\n\nThe inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.\nThe sum-to-zero inverse transform is expressed within Stan as:\n  vector manual_sum_to_zero_inverse_jacobian(vector y) {\n    int N = num_elements(y);\n    vector[N + 1] x = zeros_vector(N + 1);\n    real sum_w = 0;\n    for (n in 1:N) {\n      int i = N - n + 1;\n      real w = y[i] * inv_sqrt(i * (i + 1));\n      sum_w += w;\n      x[i] += sum_w;\n      x[i + 1] -= w * i;\n    }\n    return x;\n  }\nNote that there is no target += increment because the Jacobian is zero.\n\n\n\nThe matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any \\(N\\)-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.\nLet the unconstrained matrix be \\[\n\\mathcal Y \\in \\mathbb R^{n_1 \\times n_2}\n\\] and the zero sum vector transform as \\[\n\\mathbf z = \\mathcal C_n(\\mathbf y)\\;=\\;\n\\begin{bmatrix}H_n\\\\[2pt]-\\mathbf 1_{1\\times d}\\end{bmatrix}\\mathbf y\n\\;\\in\\mathbb R^{n+1},\n\\] where \\(H_n\\in\\mathbb R^{n \\times n}\\) is the orthogonal Helmert matrix and satisfies \\(\\mathbf 1_{1\\times d}A_d^{\\!\\top}=0\\).\nApplying \\(C_n\\) to each slice results in\n\\[\n\\mathcal Z \\;=\\;\n\\mathcal X\n\\times_1 \\bigl[\\mathcal C_{n_1}\\bigr]\n\\times_2 \\bigl[\\mathcal C_{n_2}\\bigr]\n\\] where \\[\n\\mathcal Z \\in\n\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}.\n\\]\nBecause each \\(\\mathbb R^{d_1\\times\\cdots\\times d_N}\\) is invertible on the \\(\\mathbf 1^\\perp\\) subspace the composite map applied to \\(\\mathcal Z\\) is a linear bijection between \\(\\mathbb R^{n_1\\times n_2}\\) and the codomain \\(\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}\\).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#simplex-transform.section",
+    "href": "reference-manual/transforms.html#simplex-transform.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Variables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).\nThe unit \\(K\\)-simplex is the set of points \\(x \\in \\mathbb{R}^K\\) such that for \\(1 \\leq k \\leq K\\),\n\\[\nx_k &gt; 0,\n\\]\nand\n\\[\n\\sum_{k=1}^K x_k = 1.\n\\]\nAn alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values \\((0,1)\\), and \\((1,0)\\) and the unit 2-simplex is the line connecting these two points; values such as \\((0.3,0.7)\\) and \\((0.99,0.01)\\) lie on the line. In 3-dimensions, the basis is \\((0,0,1)\\), \\((0,1,0)\\) and \\((1,0,0)\\) and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include \\((0.5,0.5,0)\\), \\((0.2,0.7,0.1)\\) and all other triplets of non-negative values summing to 1.\nAs these examples illustrate, the simplex always picks out a subspace of \\(K-1\\) dimensions from \\(\\mathbb{R}^K\\). Therefore a point \\(x\\) in the \\(K\\)-simplex is fully determined by its first \\(K-1\\) elements \\(x_1, x_2,\n\\ldots, x_{K-1}\\), with\n\\[\nx_K = 1 - \\sum_{k=1}^{K-1} x_k.\n\\]\n\n\nThe length-\\(K\\) unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length \\(K\\).\nLet \\(y\\) represent the unconstrained \\(K - 1\\) values in \\((-\\infty, \\infty)\\). The intermediate sum-to-zero vector \\(z = \\text{sum\\_to\\_zero\\_transform}(y)\\) is length \\(K\\). The unit simplex is then given by \\[\nx_i = \\text{softmax}(z) = \\frac{\\exp(z_i)}{\\sum_{i = 1}^K \\exp(z_i)}\n\\]\nThe sum-to-zero vector transform is described in further detail at the sum-to-zero vector section of the Reference Manual.\n\n\n\n\n\n\nNote\n\n\n\nAll versions of Stan pre-2.37 used the stick-breaking transform. This is documented at Stan 2.36 Reference Manual: Simplex Transform.\n\n\n\n\nThe Jacobian \\(J\\) of the inverse unit-simplex transform is found by restricting \\(J\\) to the subspace spanned by the sum-to-zero vector \\(z\\). The Jacobian is given as the \\((K - 1) \\times (K - 1)\\) matrix \\(J\\) where\n\\[\nJ_{ij} = \\frac{\\partial x_i}{\\partial z_j} =\n\\frac{\\partial}{\\partial z_i} \\left( \\frac{\\exp(z_i)}{{\\sum_{i = 1}^K \\exp(z_i)}} \\right)\n\\] and \\(i,j \\in 1, \\ldots, K - 1\\).\nThe diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification\n\\[\nJ_{ij} =\n\\begin{cases}\nx_i (1 - x_i), & \\text{if } i = j, \\\\\n-x_i x_j, & \\text{if } i \\neq j.\n\\end{cases}\n\\]\nIn matrix form this can be expressed as\n\\[\nJ = \\text{diag}(x) - x x^\\top\n\\]\nThe determinant of this matrix can be found using the Matrix Determinant Lemma:\n\\[\n\\det\\bigl(A + u v^{\\top}\\bigr)\n=\n\\det(A)\\,\\bigl(1 + v^{\\top}A^{-1}u\\bigr).\n\\]\nHere,\n\\[\nA \\;=\\; \\operatorname{diag}(x_{1},\\ldots, x_{K-1}),\n\\quad\nu \\;=\\; -\\bigl(x_1,\\ldots, x_{K-1}\\bigr)^{\\!\\top},\n\\quad\nv \\;=\\; \\bigl(x_{1}, \\ldots, x_{K-1}\\bigr)^{\\!\\top}.\n\\] Therefore,\n\\[\n\\begin{aligned}\n\\det(J)\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_i \\bigg)\n\\bigg(1 + (x_{1},\\ldots, x_{K-1})\\,\\mathrm{diag}\\bigl(x_{1}^{-1},\\ldots,x_{K-1}^{-1}\\bigr)\\,\n\\big(-x_{1},\\ldots,-x_{K-1}\\big)^{\\top}\n\\bigg) \\\\\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg)\n\\bigg(1 - \\sum_{i=1}^{K-1} x_{i}\\bigg)\n=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg) x_{K} \\\\\n&=\n\\prod_{i=1}^{K} x_{i}.\n\\end{aligned}\n\\]\n\n\n\n\nThe transform \\(Y = f(X)\\) can be derived by reversing the stages of the inverse transform,\n\\[\ny_k\n= H^\\top \\bigg(\\log(x_k)\n- \\frac{1}{K}\\sum_{i=1}^K\\log(x_i) \\bigg)\n.\n\\]\nThe matrix \\(H\\) is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#stochastic-matrix-transform.section",
+    "href": "reference-manual/transforms.html#stochastic-matrix-transform.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "The column_stochastic_matrix[N, M] and row_stochastic_matrix[M, N] type in Stan represents an \\(N \\times M\\) matrix where each column (row) is a unit simplex of dimension \\(N\\). In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.\n\n\nA column stochastic matrix \\(X \\in \\mathbb{R}^{N \\times M}\\) is defined such that each column is a simplex. For column \\(m\\) (where \\(1 \\leq m \\leq M\\)):\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{n=1}^N X_{n, m} = 1.\n\\]\nA row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e. the rows of the matrix are simplexes)\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{m=1}^N X_{n, m} = 1.\n\\]\nThis definition ensures that each column (row) of the matrix \\(X\\) lies on the \\(N-1\\) dimensional unit simplex, similar to the simplex[N] type, but extended across multiple columns(rows).\n\n\n\nFor the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).\n\n\n\nThe Jacobian determinant of the inverse transform for each column \\(m\\) in the matrix is given by the product of the diagonal entries \\(J_{n, m}\\) of the lower-triangular Jacobian matrix. This determinant is calculated as:\n\\[\n\\left| \\det J_m \\right| = \\prod_{n=1}^{N-1} \\left( z_{n, m} (1 - z_{n, m}) \\left( 1 - \\sum_{n'=1}^{n-1} X_{n', m} \\right) \\right).\n\\]\nThus, the overall Jacobian determinant for the entire column_stochastic_matrix and row_stochastic_matrix is the product of the determinants for each column (row):\n\\[\n\\left| \\det J \\right| = \\prod_{m=1}^{M} \\left| \\det J_m \\right|.\n\\]\n\n\n\nFor the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#unit-vector.section",
+    "href": "reference-manual/transforms.html#unit-vector.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "An \\(n\\)-dimensional vector \\(x \\in \\mathbb{R}^n\\) is said to be a unit vector if it has unit Euclidean length, so that\n\\[\n\\Vert x \\Vert\n\\ = \\ \\sqrt{x^{\\top}\\,x}\n\\ = \\ \\sqrt{x_1^2 + x_2^2 + \\cdots + x_n^2}\n\\ = \\ 1\\ .\n\\]\n\n\nStan divides an unconstrained vector \\(y \\in \\mathbb{R}^{n}\\) by its norm, \\(\\Vert y \\Vert = \\sqrt{y^\\top y}\\), to obtain a unit vector \\(x\\),\n\\[\nx = \\frac{y}{\\Vert y \\Vert}.\n\\]\nTo generate a unit vector, Stan generates points at random in \\(\\mathbb{R}^n\\) with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. Muller (1959) showed this generates points uniformly at random on \\(S^{n-1}\\). That is, if we draw \\(y_n \\sim \\mathsf{Normal}(0, 1)\\) for \\(n \\in 1{:}n\\), then \\(x = \\frac{y}{\\Vert y \\Vert}\\) has a uniform distribution over \\(S^{n-1}\\). This allows us to use an \\(n\\)-dimensional basis for \\(S^{n-1}\\) that preserves local neighborhoods in that points that are close to each other in \\(\\mathbb{R}^n\\) map to points near each other in \\(S^{n-1}\\). The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in \\(\\mathbb{R}^n\\) that map to identical points in \\(S^{n-1}\\).\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nThe Jacobian matrix relating the input vector \\(y\\) to the output vector \\(x\\) is singular because \\(x^\\top x = 1\\) for any non-zero input vector \\(y\\). Thus, there technically is no unique transformation from \\(x\\) to \\(y\\). To circumvent this issue, let \\(r = \\sqrt{y^\\top y}\\) so that \\(y = r\nx\\). The transformation from \\(\\left(r, x_{-n}\\right)\\) to \\(y\\) is well-defined but \\(r\\) is arbitrary, so we set \\(r = 1\\). In this case, the determinant of the Jacobian is proportional to \\(e^{-\\frac{1}{2} y^\\top y}\\), which is the kernel of a standard multivariate normal distribution with \\(n\\) independent dimensions.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#correlation-matrix-transform.section",
+    "href": "reference-manual/transforms.html#correlation-matrix-transform.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "A \\(K \\times K\\) correlation matrix \\(x\\) must be symmetric, so that\n\\[\nx_{k,k'} = x_{k',k}\n\\]\nfor all \\(k,k' \\in \\{ 1, \\ldots, K \\}\\), it must have a unit diagonal, so that\n\\[\nx_{k,k} = 1\n\\]\nfor all \\(k \\in \\{ 1, \\ldots, K \\}\\), and it must be positive definite, so that for every non-zero \\(K\\)-vector \\(a\\),\n\\[\na^{\\top} x a &gt; 0.\n\\]\nThe number of free parameters required to specify a \\(K \\times K\\) correlation matrix is \\(\\binom{K}{2}\\).\nThere is more than one way to map from \\(\\binom{K}{2}\\) unconstrained parameters to a \\(K \\times K\\) correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform Lewandowski, Kurowicka, and Joe (2009).\n\n\nIt is easiest to specify the inverse, going from its \\(\\binom{K}{2}\\) parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose \\(y\\) is a vector containing \\(\\binom{K}{2}\\) unconstrained values. These are first transformed via the bijective function \\(\\tanh : \\mathbb{R} \\rightarrow\n(-1, 1)\\)\n\\[\n\\tanh y = \\frac{\\exp(2y) - 1}{\\exp(2y) + 1}.\n\\]\nThen, define a \\(K \\times K\\) matrix \\(z\\), the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the \\(4 \\times 4\\) case, there are \\(\\binom{4}{2}\\) values arranged as\n\\[\nz\n=\n\\left[\n\\begin{array}{cccc}\n1 & \\tanh y_1 & \\tanh y_2 & \\tanh y_4\n\\\\\n0 & 1 & \\tanh y_3 & \\tanh y_5\n\\\\\n0 & 0 & 1 & \\tanh y_6\n\\\\\n0 & 0 & 0 & 1\n\\end{array}\n\\right]\n.\n\\]\nLewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array \\(z\\) to a correlation matrix \\(x\\). The entry \\(z_{i,j}\\) for \\(i &lt;\nj\\) is interpreted as the canonical partial correlation (CPC) between \\(i\\) and \\(j\\), which is the correlation between \\(i\\)’s residuals and \\(j\\)’s residuals when both \\(i\\) and \\(j\\) are regressed on all variables \\(i'\\) such that \\(i'&lt; i\\). In the case of \\(i=1\\), there are no earlier variables, so \\(z_{1,j}\\) is just the Pearson correlation between \\(i\\) and \\(j\\).\nIn Stan, the LKJ transform is reformulated in terms of a Cholesky factor \\(w\\) of the final correlation matrix, defined for \\(1 \\leq i,j \\leq K\\) by\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\n\\prod_{i'=1}^{i - 1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{if } 1 &lt; i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\\\nz_{i,j} \\, \\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThis does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n\\frac{z_{i,j}}{z_{i-1,j}} \\ w_{i-1,j} \\left( 1 - z_{i-1,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i \\leq j.\n\\end{array}\n\\right.\n\\]\nGiven the upper-triangular Cholesky factor \\(w\\), the final correlation matrix is\n\\[\nx = w^{\\top} w.\n\\]\nLewandowski, Kurowicka, and Joe (2009) show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as\n\\[\n\\mbox{det} \\, x = \\prod_{i=1}^{K-1} \\ \\prod_{j=i+1}^K \\ (1 - z_{i,j}^2)\n= \\prod_{1 \\leq i &lt; j \\leq K} (1 - z_{i,j}^2),\n\\]\n\n\n\nFrom the inverse of equation 11 in (Lewandowski, Kurowicka, and Joe 2009), the absolute Jacobian determinant is\n\\[\n\\sqrt{\\prod_{i=1}^{K-1}\\prod_{j=i+1}^K \\left(1-z_{i,j}^2\\right)^{K-i-1}} \\\n\\times \\prod_{i=1}^{K-1}\\prod_{j=i+1}^K\n\\frac{\\partial z_{i,j}}{\\partial y_{i,j}}\n\\]\n\n\n\nThe correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.\nStarting with a correlation matrix \\(x\\), the first step is to find the unique upper triangular \\(w\\) such that \\(x = w w^{\\top}\\). Because \\(x\\) is positive definite, this can be done by applying the Cholesky decomposition,\n\\[\nw = \\mbox{chol}(x).\n\\]\nThe next step from the Cholesky factor \\(w\\) back to the array \\(z\\) of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of \\(w\\), which when inverted yields\n\\[\nz_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i \\leq j,\n\\\\\nw_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n{w_{i,j}}\n\\\n\\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,j}^2 \\right)^{-1/2}\n& \\mbox{if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThe final stage of the transform reverses the hyperbolic tangent transform, which is defined by\n\\[\ny = \\tanh^{-1} z = \\frac{1}{2} \\log \\left( \\frac{1 + z}{1 - z} \\right).\n\\]\nThe inverse hyperbolic tangent function, \\(\\tanh^{-1}\\), is also called the Fisher transformation.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#covariance-matrices",
+    "href": "reference-manual/transforms.html#covariance-matrices",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "A \\(K \\times K\\) matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires \\(K + \\binom{K}{2}\\) free parameters to specify a \\(K \\times K\\) covariance matrix.\n\n\nStan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.1\nIf \\(x\\) is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix \\(z =\n\\mathrm{chol}(x)\\) with positive diagonal entries, called a Cholesky factor, such that\n\\[\nx = z \\, z^{\\top}.\n\\]\nThe off-diagonal entries of the Cholesky factor \\(z\\) are unconstrained, but the diagonal entries \\(z_{k,k}\\) must be positive for \\(1 \\leq k\n\\leq K\\).\nTo complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix \\(y\\) defined by\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log z_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nz_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular \\(K \\times K\\) matrix \\(y\\), the first step is to recover the intermediate matrix \\(z\\) by reversing the log transform,\n\\[\nz_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe covariance matrix \\(x\\) is recovered from its Cholesky factor \\(z\\) by taking\n\\[\nx = z \\, z^{\\top}.\n\\]\n\n\n\nThe Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix \\(y\\) to matrix \\(z\\) with positive diagonals and the product transform from the Cholesky factor \\(z\\) to \\(x\\).\nThe transform from unconstrained \\(y\\) to Cholesky factor \\(z\\) has a diagonal Jacobian matrix, the absolute determinant of which is thus\n\\[\n\\prod_{k=1}^K  \\frac{\\partial}{\\partial_{y_{k,k}}} \\, \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K z_{k,k}.\n\\]\nThe Jacobian matrix of the second transform from the Cholesky factor \\(z\\) to the covariance matrix \\(x\\) is also triangular, with diagonal entries corresponding to pairs \\((m,n)\\) with \\(m \\geq n\\), defined by\n\\[\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( z \\, z^{\\top} \\right)_{m,n}\n\\ = \\\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( \\sum_{k=1}^K z_{m,k} \\, z_{n,k} \\right)\n\\ = \\\n\\left\\{\n\\begin{array}{cl}\n2 \\, z_{n,n} & \\mbox{if } m = n \\mbox{ and }\n\\\\\nz_{n,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe absolute Jacobian determinant of the second transform is thus\n\\[\n2^{K} \\ \\prod_{m = 1}^{K} \\ \\prod_{n=1}^{m} z_{n,n}\n\\ = \\\n\\prod_{n=1}^K \\ \\prod_{m=n}^K z_{n,n}\n\\ = \\\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}.\n\\]\nFinally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular \\(y\\) to a symmetric, positive definite matrix \\(x\\) is the product of the Jacobian determinants of the exponentiation and product transforms,\n\\[\n\\left( \\prod_{k=1}^K z_{k,k} \\right)\n\\left(\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}\n\\right)\n\\ = \\\n2^K\n\\, \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\nLet \\(f^{-1}\\) be the inverse transform from a \\(K + \\binom{K}{2}\\)-vector \\(y\\) to the \\(K \\times K\\) covariance matrix \\(x\\). A density function \\(p_X(x)\\) defined on \\(K \\times K\\) covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(K + \\binom{K}{2}\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\ 2^K \\ \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#cholesky-factors-of-covariance-matrices",
+    "href": "reference-manual/transforms.html#cholesky-factors-of-covariance-matrices",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "An \\(M \\times M\\) covariance matrix \\(\\Sigma\\) can be Cholesky factored to a lower triangular matrix \\(L\\) such that \\(L\\,L^{\\top} = \\Sigma\\). If \\(\\Sigma\\) is positive definite, then \\(L\\) will be \\(M \\times M\\). If \\(\\Sigma\\) is only positive semi-definite, then \\(L\\) will be \\(M \\times N\\), with \\(N &lt; M\\).\nA matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and \\(M \\geq\nN\\). A matrix satisfying these conditions ensures that \\(L \\,\nL^{\\top}\\) is positive semi-definite if \\(M &gt; N\\) and positive definite if \\(M = N\\).\nA Cholesky factor of a covariance matrix requires \\(N + \\binom{N}{2} +\n(M - N)N\\) unconstrained parameters.\n\n\nStan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose \\(x\\) is an \\(M \\times N\\) Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log x_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nx_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform need only invert the logarithm with an exponentiation. If \\(y\\) is the unconstrained matrix representation, then the elements of the constrained matrix \\(x\\) is defined by\n\\[\nx_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe transform has a diagonal Jacobian matrix, the absolute determinant of which is\n\\[\n\\prod_{n=1}^N  \\frac{\\partial}{\\partial_{y_{n,n}}} \\, \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N x_{n,n}.\n\\]\nLet \\(x = f^{-1}(y)\\) be the inverse transform from a \\(N + \\binom{N}{2}\n+ (M - N)N\\) vector to an \\(M \\times N\\) Cholesky factor for a covariance matrix \\(x\\) defined in the previous section. A density function \\(p_X(x)\\) defined on \\(M \\times N\\) Cholesky factors of covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(N + \\binom{N}{2}\n+ (M - N)N\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\prod_{N=1}^N x_{n,n}.\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#cholesky-factors-of-correlation-matrices",
+    "href": "reference-manual/transforms.html#cholesky-factors-of-correlation-matrices",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "A \\(K \\times K\\) correlation matrix \\(\\Omega\\) is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a \\(K \\times K\\) lower-triangular matrix \\(L\\) with positive diagonal elements such that \\(\\Omega = L\\,L^{\\top}\\). Because the correlation matrix has a unit diagonal,\n\\[\n\\Omega_{k,k} = L_k\\,L_k^{\\top} = 1,\n\\]\neach row vector \\(L_k\\) of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of \\(L\\) to be calculated from the off-diagonal elements, so that a Cholesky factor for a \\(K \\times K\\) correlation matrix requires only \\(\\binom{K}{2}\\) unconstrained parameters.\n\n\nIt is easiest to start with the inverse transform from the \\(\\binom{K}{2}\\) unconstrained parameters \\(y\\) to the \\(K \\times K\\) lower-triangular Cholesky factor \\(x\\). The inverse transform is based on the hyperbolic tangent function, \\(\\tanh\\), which satisfies \\(\\tanh(x) \\in (-1,1)\\). Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the section on correlation matrix transforms for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.\nSuppose \\(y\\) is a vector of \\(\\binom{K}{2}\\) unconstrained values. Let \\(z\\) be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the \\(3 \\times 3\\) case,\n\\[\nz =\n\\left[\n\\begin{array}{ccc}\n0 & 0 & 0\n\\\\\n\\tanh y_1 & 0 & 0\n\\\\\n\\tanh y_2 & \\tanh y_3 & 0\n\\end{array}\n\\right]\n\\]\nThe matrix \\(z\\), with entries in the range \\((-1, 1)\\), is then transformed to the Cholesky factor \\(x\\), by taking2\n\\[\nx_{i,j}\n=\n\\left\\{\n\\begin{array}{lll}\n0 & \\mbox{ if } i &lt; j & \\mbox{ [above diagonal]}\n\\\\\n\\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i = j & \\mbox{ [on diagonal]}\n\\\\\nz_{i,j} \\ \\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i &gt; j & \\mbox{ [below diagonal]}\n\\end{array}\n\\right.\n\\]\nIn the \\(3 \\times 3\\) case, this yields\n\\[\nx =\n\\left[\n\\begin{array}{ccc}\n1 & 0 & 0\n\\\\\nz_{2,1} & \\sqrt{1 - x_{2,1}^2} & 0\n\\\\\nz_{3,1} & z_{3,2} \\sqrt{1 - x_{3,1}^2}\n        & \\sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}\n\\end{array}\n\\right],\n\\]\nwhere the \\(z_{i,j} \\in (-1,1)\\) are the \\(\\tanh\\)-transformed \\(y\\).\nThe approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at \\(j=1\\), each below-diagonal entry \\(x_{i,j}\\) is determined by the (signed) fraction \\(z_{i,j}\\) of the remaining length for the row that it consumes. The diagonal entries \\(x_{i,i}\\) get any leftover length from earlier entries in their row. The above-diagonal entries are zero.\n\n\n\nSuppose \\(x\\) is a \\(K \\times K\\) Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values \\(z\\) from \\(x\\),\n\\[\nz_{i,j} = \\frac{x_{i,j}}{\\sqrt{1 - \\sum_{j' &lt; j}x_{i,j'}^2}}.\n\\]\nThe mapping from the resulting \\(z\\) to \\(y\\) inverts \\(\\tanh\\),\n\\[\ny\n\\ = \\\n\\tanh^{-1} z\n\\ = \\\n\\frac{1}{2} \\left( \\log (1 + z) - \\log (1 - z) \\right).\n\\]\n\n\n\nThe Jacobian of the full transform is the product of the Jacobians of its component transforms.\nFirst, for the inverse transform \\(z = \\tanh y\\), the derivative is\n\\[\n\\frac{d}{dy} \\tanh y = \\frac{1}{(\\cosh y)^2}.\n\\]\nSecond, for the inverse transform of \\(z\\) to \\(x\\), the resulting Jacobian matrix \\(J\\) is of dimension \\(\\binom{K}{2} \\times\n\\binom{K}{2}\\), with indexes \\((i,j)\\) for \\((i &gt; j)\\). The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each \\((i,j)\\) pair,\n\\[\n\\left| \\, \\mbox{det} \\, J \\, \\right|\n  \\ = \\ \\prod_{i &gt; j} \\left| \\frac{d}{dz_{i,j}} x_{i,j} \\right|,\n\\]\nwhere\n\\[\n\\frac{d}{dz_{i,j}} x_{i,j}\n= \\sqrt{1 - \\sum_{j' &lt; j} x^2_{i,j'}}.\n\\]\nSo the combined density for unconstrained \\(y\\) is\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n  \\ \\\n  \\prod_{n &lt; \\binom{K}{2}} \\frac{1}{(\\cosh y)^2}\n  \\ \\\n  \\prod_{i &gt; j} \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2\n  \\right)^{1/2},\n\\]\nwhere \\(x = f^{-1}(y)\\) is used for notational convenience. The log Jacobian determinant of the complete inverse transform \\(x = f^{-1}(y)\\) is given by\n\\[\n\\log \\left| \\, \\det J \\, \\right|\n=\n-2 \\sum_{n \\leq \\binom{K}{2}}\n\\log \\cosh y\n\\\n+\n\\\n\\frac{1}{2} \\\n\\sum_{i &gt; j}\n\\log \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2 \\right)\n.\n\\]",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/transforms.html#footnotes",
+    "href": "reference-manual/transforms.html#footnotes",
+    "title": "Constraint Transforms",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nAn alternative to the transform in this section, which can be coded directly in Stan, is to parameterize a covariance matrix as a scaled correlation matrix. An arbitrary \\(K \\times K\\) covariance matrix \\(\\Sigma\\) can be expressed in terms of a \\(K\\)-vector \\(\\sigma\\) and correlation matrix \\(\\Omega\\) as \\[\\Sigma = \\mbox{diag}(\\sigma) \\times \\Omega \\times \\mbox{diag}(\\sigma),\\] so that each entry is just a deviation-scaled correlation, \\[\\Sigma_{m,n} = \\sigma_m \\times \\sigma_n \\times \\Omega_{m,n}.\\]↩︎\nFor convenience, a summation with no terms, such as \\(\\sum_{j' &lt; 1} x_{i,j'}\\), is defined to be 0. This implies \\(x_{1,1} = 1\\) and that \\(x_{i,1} = z_{i,1}\\) for \\(i &gt; 1\\).↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Constraint Transforms"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html",
+    "href": "reference-manual/statements.html",
+    "title": "Statements",
+    "section": "",
+    "text": "The blocks of a Stan program are made up of variable declarations and statements; see the blocks chapter for details. Unlike programs in BUGS, the declarations and statements making up a Stan program are executed in the order in which they are written. Variables must be defined to have some value (as well as declared to have some type) before they are used — if they do not, the behavior is undefined.\nThe basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters. Log probability functions can be constructed by using distribution statements and log probability increment statements. Statements may be grouped into sequences and into for-each loops. In addition, Stan allows local variables to be declared in blocks and also allows an empty statement consisting only of a semicolon.\n\n\nThe data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the blocks chapter for more information about the block structure of Stan programs.\n\n\n\nAn assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.\nn = 0;\nExecuting this statement assigns the value of the expression 0, which is the integer zero, to the variable n. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because 0 is an expression of type int, the variable n must be declared as being of type int or of type real. If the variable is of type real, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable n will have the value zero (either as an integer or a floating-point value, depending on its type).\nSyntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).\nBecause the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing\nn = n + 1;\nSuch self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.\nThe left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if Sigma is of type matrix, then\nSigma[1, 1] = 1.0;\nsets the value in the first column of the first row of Sigma to one.\nAssignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, a is an array of type array[,] real and b is an array of type array[] real, then the following two statements are both well-formed.\na[3] = b;\nb = a[4];\nSimilarly, if x is a variable declared to have type row_vector and Y is a variable declared as type matrix, then the following sequence of statements to swap the first two rows of Y is well formed.\nx = Y[1];\nY[1] = Y[2];\nY[2] = x;\n\n\nStan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type int to an lvalue of type real, and we can assign an expression of type real to an lvalue of type complex. Furthermore, promotion is transitive, so that we can assign an expression of type int to an lvalue of type complex.\nPromotion extends to containers, so that arrays of int can be promoted to arrays of real during assignment, and arrays of real can be assigned to an lvalue of type array of complex. Similarly, an expression of type vector may be assigned to an lvalue of type complex_vector, and similarly for row vectors and matrices.\n\n\n\nThe expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,\n\na variable, or\na variable with one or more indices, or\na comma separated list of lvalues surrounded by ( and )\n\nTo be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.\n\n\n\nMultiple indexes, as described in the multi-indexing section, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.\n\n\nAll assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.\n\n\n\n\nStan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.\nreal x = 5;\nx += 7;  // value of x is now 12\nThe compound arithmetic and assignment statement above is equivalent to the following long form.\nx = x + 7;\nIn general, the compound form\nx op= y\nwill be equivalent to\nx = x op y;\nThe compound statement will be legal whenever the long form is legal. This requires that the operation x op y must itself be well formed and that the result of the operation be assignable to x. For the expression x to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.\nvector[N] x;\nx[3] += 2;\nAs a further example, consider\nmatrix[M, M] x;\nvector[M] y;\nreal z;\nx *= x;  // OK, (x * x) is a matrix\nx *= z;  // OK, (x * z) is a matrix\nx *= y;  // BAD, (x * y) is a vector\nThe supported compound arithmetic and assignment operations are listed in the compound arithmetic/assignment table; they are also listed in the index prefaced by operator, e.g., operator+=.\nCompound Arithmetic/Assignment Table.  Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.\n\n\n\noperation\ncompound\nunfolded\n\n\n\n\naddition\nx += y\nx = x + y\n\n\nsubtraction\nx -= y\nx = x - y\n\n\nmultiplication\nx *= y\nx = x * y\n\n\ndivision\nx /= y\nx = x / y\n\n\nelementwise multiplication\nx .*= y\nx = x .* y\n\n\nelementwise division\nx ./= y\nx = x ./ y\n\n\n\n\n\n\n\nThe basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.1\ntarget += -0.5 * y * y;\nThe keyword target here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).\nIn this example, the unnormalized log probability of a unit normal variable \\(y\\) is added to the total log probability. In the general case, the argument can be any expression.2\nAn entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.\nparameters {\n  real y;\n}\nmodel {\n  target += -0.5 * y * y;\n}\nThis model defines a log probability function\n\\[\n\\log p(y) = - \\, \\frac{y^2}{2} - \\log Z\n\\]\nwhere \\(Z\\) is a normalizing constant that does not depend on \\(y\\). The constant \\(Z\\) is conventionally written this way because on the linear scale, \\[\np(y) = \\frac{1}{Z} \\exp\\left(-\\frac{y^2}{2}\\right).\n\\] which is typically written without reference to \\(Z\\) as \\[\np(y) \\propto \\exp\\left(-\\frac{y^2}{2}\\right).\n\\]\nStan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant \\(Z\\) is either time-consuming to compute or intractable to evaluate.\n\n\nThe built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.\nFor instance, the normal_lpdf function returns the log density of a normal distribution:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nThe normal_lupdf function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant sigma term, normal_lupdf would be equivalent to:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nAll functions ending in _lpdf have a corresponding _lupdf version which evaluates and returns the unnormalized density. The same is true for _lpmf and _lupmf.\n\n\n\nThe increment log density statement looks syntactically like compound addition and assignment (see the compound arithmetic/assignment section, it is treated as a primitive statement because target is not itself a variable. So, even though\ntarget += lp;\nis a legal statement, the corresponding long form is not legal.\ntarget = target + lp;  // BAD, target is not a variable\n\n\n\nThe target += ... statement accepts an argument in place of ... for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.\n\n\n\n\nA variant of the target += statement described above is the jacobian += statement. This can be used in the transformed parameters block or in functions ending with _jacobian to mimic the log Jacobian adjustments accrued by built-in variable transforms.\nSimilarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.\nFor example, here is a program which recreates the existing &lt;upper=x&gt; transform on real numbers:\nfunctions {\n  real my_upper_bound_jacobian(real x, real ub) {\n    jacobian += x;\n    return ub - exp(x);\n  }\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  // use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters\n  // e.g.\n  // b ~ lognormal(0, 1);\n}\n\n\nTo access the accumulated log density up to the current execution point, the function target() may be used.\n\n\n\n\nThe term “sampling statement” has been replaced with distribution statement.\n\n\n\nStan supports writing probability statements also using distribution statements, for example\ny ~ normal(mu, sigma);\nmu ~ normal(0, 10);\nsigma ~ normal(0, 1);\nThe symbol \\(\\sim\\) is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.\nIn general, we can read \\(\\sim\\) as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as \\[\n\\begin{aligned}\n   p(y| \\mu, \\sigma) & = \\mathrm{normal}(y |  \\mu, \\sigma)\\\\\n   p(\\mu) & = \\mathrm{normal}(\\mu |  0, 10)\\\\\n   p(\\sigma) & = \\mathrm{normal}^+(\\sigma |  0, 1).\n\\end{aligned}\n\\] A collection of distribution statements define a joint distribution as the product of component distributions \\[\np(y,\\mu,\\sigma) = p(y| \\mu, \\sigma )p(\\mu) p(\\sigma).\n\\]\nThis works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:\n  a ~ normal(0, 1);\n  a ~ normal(0, 1);\nThis is translated to \\[\n    p(a) = \\mathrm{normal}(a | 0, 1)\\mathrm{normal}(a |  0, 1),\n\\] which in this case is \\(\\mathrm{normal}(a|0,1/\\sqrt{2})\\). One might expect that the above two lines of code would represent a redundant expression of a \\(\\mathrm{normal}(a|0,1)\\) prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.\nWhen the joint distribution is considered as a function of parameters (e.g. \\(\\mu\\), \\(\\sigma\\)) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.\nStan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each \\(\\sim\\) statement corresponds to a multiplicative factor in the unnormalized posterior density.\nDistribution statements (~) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).\nIn Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as\ntarget += normal_lpdf(y | mu, sigma);\ntarget += normal_lpdf(mu | 0, 10);\ntarget += normal_lpdf(sigma | 0, 1);\nStan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g. censored data model) or coding language limitations (e.g. mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.\nIn general, a distribution statement of the form\ny ~ dist(theta1, ..., thetaN);\ninvolving subexpressions y and theta1 through thetaN (including the case where N is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real y values, the log probability density function is used,\ntarget += dist_lpdf(y | theta1, ..., thetaN);\nFor those restricted to integer y values, the log probability mass function is used,\ntarget += dist_lpmf(y | theta1, ..., thetaN);\nThis will be well formed if and only if dist_lpdf(y | theta1,   ..., thetaN) or dist_lpmf(y | theta1, ..., thetaN) is a well-formed expression of type real. User defined distributions can be defined in functions block by using function names ending with _lpdf.\n\n\nAlthough both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in\ny ~ normal(mu, sigma);\nand explicitly incrementing the log probability function, as in\ntarget += normal_lpdf(y | mu, sigma);\nThe distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to normal_lpdf adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, y, mu, or sigma, involve only constants, data variables, and transformed data variables. See the section Built in distributions above discussing _lupdf and _lupmf functions that also drops all the constant terms.\n\n\n\nThe left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write\nparameters {\n  real&lt;lower=0&gt; beta;\n}\n// ...\nmodel {\n  log(beta) ~ normal(mu, sigma);\n}\nUnfortunately, this is not enough to properly model beta as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to beta here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure beta gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the change of variables section for full definitions. For the case above, the following adjustment will account for the log transform.3\ntarget += - log(abs(y));\n\n\n\nStan supports truncating distributions with lower bounds, upper bounds, or both.\n\n\nA probability density function \\(p(x)\\) for a continuous distribution may be truncated to an interval \\([a, b]\\) to define a new density \\(p_{[a, b]}(x)\\) with support \\([a, b]\\) by setting\n\\[\np_{[a, b]}(x)\n= \\frac{p(x)}\n       {\\int_a^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) for a discrete distribution may be truncated to the closed interval \\([a, b]\\) by\n\\[\np_{[a, b]}(x) = \\frac{p(x)}\n                  {\\sum_{u = a}^b p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x)\n= \\frac{p(x)}\n       {\\int_a^{\\infty} p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x) = \\frac{p(x)}\n                  {\\sum_{a &lt;= u} p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty, b]}(x)\n= \\frac{p(x)}\n       {\\int_{-\\infty}^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty,b]}(x) = \\frac{p(x)}\n                  {\\sum_{u &lt;= b} p(u)}.\n\\]\n\n\n\nGiven a probability function \\(p_X(x)\\) for a random variable \\(X\\), its cumulative distribution function (cdf) \\(F_X(x)\\) is defined to be the probability that \\(X \\leq x\\),\n\\[\nF_X(x) = \\Pr[X \\leq x].\n\\]\nThe upper-case variable \\(X\\) is the random variable whereas the lower-case variable \\(x\\) is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to\n\\[\nF_X(x) \\ = \\ \\int_{-\\infty}^{x} p_X(u) \\, du,\n\\]\nFor discrete variables, the cdf is defined to include the upper bound given by the argument,\n\\[\nF_X(x) = \\sum_{u \\leq x} p_X(u).\n\\]\n\n\n\nThe complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by\n\\[\nF^C_X(x)\n\\ = \\ \\Pr[X &gt; x]\n\\ = \\ 1 - F_X(x).\n\\]\nUnlike the cdf, the ccdf is exclusive of the bound, hence the event \\(X &gt; x\\) rather than the cdf’s event \\(X \\leq x\\).\nFor continuous distributions, the ccdf works out to\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\int_{-\\infty}^x p_X(u) \\, du\n\\ = \\ \\int_x^{\\infty} p_X(u) \\, du.\n\\]\nThe lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over \\(u &gt; x\\),\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\sum_{u \\leq x} p_X(u)\n\\ = \\ \\sum_{u &gt; x} p_X(u).\n\\]\nCumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by \\[\n\\int_a^b p(u) \\, du = F_X(b) - F_X(a).\n\\] This allows truncated distributions to be defined as \\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a)}.\n\\]\nFor discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from \\(F_X(b) - F_X(a)\\),\n\\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a) + p_X(a)}.\n\\]\n\n\n\nStan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to \\([-0.5, 2.1]\\) can be coded with the following distribution statement.\ny ~ normal(0, 1) T[-0.5, 2.1];\nTruncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.\nIn general, the truncation bounds and parameters may be parameters or local variables.\nBecause the example above involves a continuous distribution, it behaves the same way as the following more verbose form.\ny ~ normal(0, 1);\nif (y &lt; -0.5 || y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -log_diff_exp(normal_lcdf(2.1 | 0, 1),\n                          normal_lcdf(-0.5 | 0, 1));\n}\nBecause a Stan program defines a log density function, all calculations are on the log scale. The function normal_lcdf is the log of the cumulative normal distribution function and the function log_diff_exp(a, b) is a more arithmetically stable form of log(exp(a) - exp(b)).\nFor a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution\ny ~ poisson(3.7) T[2, 10];\nbehaves in the same way as the following code.\ny ~ poisson(3.7);\nif (y &lt; 2 || y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         log_diff_exp(poisson_lcdf(10 | 3.7),\n                                      poisson_lcdf(2 | 3.7)));\n}\nRecall that log_sum_exp(a, b) is just the arithmetically stable form of log(exp(a) + exp(b)).\n\n\n\nFor truncating with only a lower bound, the upper limit is left blank.\ny ~ normal(0, 1) T[-0.5, ];\nThis truncated distribution statement has the same behavior as the following code.\ny ~ normal(0, 1);\nif (y &lt; -0.5) {\n  target += negative_infinity();\n} else {\n  target += -normal_lccdf(-0.5 | 0, 1);\n}\nThe normal_lccdf function is the normal complementary cumulative distribution function.\nAs with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus\ny ~ poisson(3.7) T[2, ];\nbehaves the same way as\ny ~ poisson(3.7);\nif (y &lt; 2) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         poisson_lccdf(2 | 3.7));\n}\n\n\n\nTo truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement\ny ~ normal(0, 1) T[ , 2.1];\nproduces the same result as the following code.\ntarget += normal_lpdf(y | 0, 1);\nif (y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -normal_lcdf(2.1 | 0, 1);\n}\nWith only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement\ny ~ poisson(3.7) T[ , 10];\nbehaves the same way as the following code.\ny ~ poisson(3.7);\nif (y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -poisson_lcdf(10 | 3.7);\n}\n\n\n\nIn all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.\n\n\n\nFor continuous distributions, truncation points must be expressions of type int or real. For discrete distributions, truncation points must be expressions of type int.\n\n\n\nFor a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds \\(-\\infty\\) to the total log probability, which in turn results in the sample being rejected.\n\n\n\nVectorization of distribution functions with truncation is available if the underlying distribution, lcdf, and lccdf functions meet the required signatures.\nThe equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):\n\nIf the variate y is the only non-scalar, the result is the same as described in the above sections, but the lcdf/lccdf calculation is multiplied by size(y).\nIf the other arguments to the distribution are non-scalars, then the vectorized version of the lcdf/lccdf is used. These functions return the sum of their terms, so no multiplication by the size is needed.\nThe exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the log_diff_exp of two sums is not the same as the sum of the pairwise log_diff_exp operations.\n\nNote that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.\n\n\n\n\n\nSuppose N is a variable of type int, y is a one-dimensional array of type array[] real, and mu and sigma are variables of type real. Furthermore, suppose that n has not been defined as a variable. Then the following is a well-formed for-loop statement.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nThe loop variable is n, the loop bounds are the values in the range 1:N, and the body is the statement following the loop bounds.\n\n\nThe type of the loop variable is int. Unlike in C++ and similarly to R, this variable must not be declared explicitly.\nThe bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range L:H will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to L and less than or equal to H. For example, the loop for (n in 2:5) will cause the body of the for loop to be executed with n equal to 2, 3, 4, and 5, in order. The variable and bound for (n in   5:2) will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.\nThe scope of the loop variable is limited to the body of the loop.\n\n\n\nUnlike in BUGS, Stan allows variables to be reassigned. For example, the variable theta in the following program is reassigned in each iteration of the loop.\nfor (n in 1:N) {\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nSuch reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.4\nIn Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.\nfor (n in 1:N) {\n  y[n] ~ bernoulli(theta);\n  theta = inv_logit(alpha + x[n] * beta);\n}\nIn this program, theta is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.\nStan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.\ntotal = 0.0;\nfor (n in 1:N) {\n  total = total + x[n];\n}\nAfter the for loop is executed, the variable total will hold the sum of the elements in the array x. This example was purely pedagogical; it is easier and more efficient to write\ntotal = sum(x);\nA variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.\nfor (n in 1:100) {\n  y += y * epsilon;\n  epsilon = 0.5 * epsilon;\n  y += y * epsilon;\n}\n\n\n\n\nA second form of for loops allows iteration over elements of containers. If ys is an expression denoting a container (vector, row vector, matrix, or array) with elements of type T, then the following is a well-formed foreach statement.\nfor (y in ys) {\n  // ... do something with y ...\n}\nThe order in which elements of ys are visited is defined for container types as follows.\n\nvector, row_vector: elements visited in order, y is of type double\nmatrix: elements visited in column-major order, y is of type double\narray[] T: elements visited in order, y is of type T.\n\nConsequently, if ys is a two dimensional array array[,] real, y will be a one-dimensional array of real values (type array[] real). If ’ysis a matrix, thenywill be a real value (typereal`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,\narray[2, 3] real yss;\nfor (ys in yss) {\n  for (y in ys) {\n    // ... do something with y ...\n  }\n}\nwhereas a matrix can be looped over in one foreach statement\nmatrix[2, 3] yss;\nfor (y in yss) {\n   // ... do something with y...\n}\nIn both cases, the loop variable y is of type real.  The elements of the matrix are visited in column-major order (e.g.,y[1, 1],y[2, 1],y[1, 2], ...,y[2, 3]), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,y[1, 1],y[1, 2],y[1, 3],y[2, 1], ...,y[2, 3]`).\n\n\n\nStan supports full conditional statements using the same if-then-else syntax as C++. The general format is\nif (condition1)\n  statement1\nelse if (condition2)\n  statement2\n// ...\nelse if (conditionN-1)\n  statementN-1\nelse\n  statementN\nThere must be a single leading if clause, which may be followed by any number of else if clauses, all of which may be optionally followed by an else clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.\nThe entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.\n\n\n\nStan supports standard while loops using the same syntax as C++. The general format is as follows.\nwhile (condition)\n  body\nThe condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).\nEvaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.\nThe rest of the body of a while loop may be skipped using a continue. The loop will be exited with a break statement. See the section on continue and break statements for more details.\n\n\n\nJust as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.\n\n\nBlocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nTo put multiple statements inside the body of a for loop, a block is used, as in the following example.\nfor (n in 1:N) {\n  lambda[n] ~ gamma(alpha, beta);\n  y[n] ~ poisson(lambda[n]);\n}\nThe open curly bracket ({) is the first character of the block and the close curly bracket (}) is the last character.\nBecause whitespace is ignored in Stan, the following program will not compile.\nfor (n in 1:N)\n  y[n] ~ normal(mu, sigma);\n  z[n] ~ normal(mu, sigma); // ERROR!\nThe problem is that the body of the for loop is taken to be the statement directly following it, which is y[n] ~ normal(mu, sigma). This leaves the probability statement for z[n] hanging, as is clear from the following equivalent program.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nz[n] ~ normal(mu, sigma); // ERROR!\nNeither of these programs will compile. If the loop variable n was defined before the for loop, the for-loop declaration will raise an error. If the loop variable n was not defined before the for loop, then the use of the expression z[n] will raise an error.\n\n\n\nA for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.\nfor (n in 1:N) {\n  real theta;\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nThe local variable theta is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus theta is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable theta if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.\nfor (m in 1:M) {\n  real theta;\n  for (n in 1:N) {\n    real theta; // ERROR!\n    theta = inv_logit(alpha + x[m, n] * beta);\n    y[m, n] ~ bernoulli(theta);\n// ...\nThe compiler will flag the second declaration of theta with a message that it is already defined.\n\n\n\nLocal variables may not have constraints on their declaration. The only types that may be used are listed in the types table under “local”.\n\n\n\nA block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following\nfor (m in 1:M) {\n  {\n     int n = 2 * m;\n     sum += n;\n  }\n  for (n in 1:N) {\n    sum += x[m, n];\n  }\n}\nThe variable declaration int n; is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of n are unrelated.\n\n\n\n\nThe one-token statements continue and break may be used within loops to alter control flow; continue causes the next iteration of the loop to run immediately, whereas break terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both break and continue scope to the most deeply nested loop, but pass through non-loop statements.\nAlthough these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.\n\n\nWhen a break statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:\nwhile (1) {\n  if (n &lt; 0) {\n    break;\n  }\n  foo(n);\n  n = n - 1;\n}\nThe while~(1) loop is a “forever” loop, because 1 is the true value, so the test always succeeds. Within the loop, if the value of n is less than 0, the loop terminates, otherwise it executes foo(n) and then decrements n. The statement above does exactly the same thing as\nwhile (n &gt;= 0) {\n  foo(n);\n  n = n - 1;\n}\nThis case is simply illustrative of the behavior; it is not a case where a break simplifies the loop.\n\n\n\nThe continue statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array x,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &lt;= 0) {\n    continue;\n  }\n  sum += x[n];\n}\nWhen the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an x[n] less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &gt; 0) {\n    sum += x[n];\n  }\n}\nWhile the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.\n\n\n\nIf there is a loop nested within a loop, a break or continue statement only breaks out of the inner loop. So\nwhile (cond1) {\n  // ...\n  while (cond2) {\n    // ...\n    if (cond3) {\n      break;\n    }\n    // ...\n  }\n  // execution continues here after break\n  // ...\n}\nIf the break is triggered by cond3 being true, execution will continue after the nested loop.\nAs with break statements, continue statements go back to the top of the most deeply nested loop in which the continue appears.\nAlthough break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.\n\n\n\n\nStan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.\nfor (n in 1:N) { print(\"loop iteration: \", n); ... }\nThe print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression n, followed by a new line.\n\n\nThe text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.\nPrinting a simple real or int typed variable always prints the variable’s value.5\nFor array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as\n[1, 2, 3]\nand a \\(2 \\times 3\\)-matrix as\n[[1, 2, 3], [4, 5, 6]]\nComplex numbers print as pairs. For example, the pair of statements\ncomplex z = to_complex(1.2, -3.5);\nprint(z)\nwill print as (1.2,-3.5), with no space after the comma or within the parentheses.\nPrinting a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.\ntransformed data {\n  matrix[2, 2] u;\n  u[1, 1] = 1.0;   u[1, 2] = 4.0;\n  u[2, 1] = 9.0;   u[2, 2] = 16.0;\n  for (n in 1:2) {\n    print(\"u[\", n, \"] = \", u[n]);\n  }\n}\nThis print statement executes twice, printing the following two lines of output.\nu[1] = [1, 4]\nu[2] = [9, 16]\n\n\n\nThe input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.\n\n\n\nPrinting for a print statement happens every time it is executed. The transformed data block is executed once per chain, the transformed parameter and model blocks once per leapfrog step, and the generated quantities block once per iteration.\n\n\n\nString literals begin and end with a double quote character (\"). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.\nThe Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.\n\n\n\nBecause Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.\nIt is particularly useful to print the value of the target log density accumulator (through the target() function), as in the following example.\nvector[2] y;\ny[1] = 1;\nprint(\"log density before =\", target());\ny ~ normal(0,1);  // bug!  y[2] not defined\nprint(\"log density after =\", target());\nThe example has a bug in that y[2] is not defined before the vector y is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).\nNote that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.\n\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.\nLike the print statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.\nReject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable x’s value is negative.\nif (x &lt; 0) {\n  reject(\"x must not be negative; found x=\", x);\n}\n\n\nReject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the normal_lpdf function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.\nIn all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.\n\n\nRejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.\n\n\n\nRejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.\nReject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.\nFatal errors in other blocks may also be signaled by use of the fatal_error statement.\n\n\n\nRejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.\nIf the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.\n\n\n\n\nRejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;\n  real&lt;lower=a, upper=b&gt; theta;\n  // ...\n}\nmodel {\n  // **wrong** needs explicit truncation\n  theta ~ normal(0, 1);\n  // ...\n}\nThis program is wrong because its truncation bounds on theta depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.\n  theta ~ normal(0, 1) T[a, b];\nThe conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over \\([a ,b]\\); in these simple univariate cases, we can overcome that with the T[ , ] notation, which essentially divides by whatever the prior integrates to over \\([a, b]\\).\nThis problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.\n  if (theta &lt; a || theta &gt; b) {\n    reject(\"theta not in (a, b)\");\n  }\n  // still **wrong**, needs T[a,b]\n  theta ~ normal(0, 1);\nIn this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.\nEven if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.\n\n\n\n\nThe Stan fatal_error statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.\nLike the print or reject statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.\nThe fatal error may be used to signal an unrecoverable error in blocks where reject leads to the algorithm attempting to try again, such as the model block.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#statement-block-contexts",
+    "href": "reference-manual/statements.html#statement-block-contexts",
+    "title": "Statements",
+    "section": "",
+    "text": "The data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the blocks chapter for more information about the block structure of Stan programs.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#assignment-statement.section",
+    "href": "reference-manual/statements.html#assignment-statement.section",
+    "title": "Statements",
+    "section": "",
+    "text": "An assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.\nn = 0;\nExecuting this statement assigns the value of the expression 0, which is the integer zero, to the variable n. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because 0 is an expression of type int, the variable n must be declared as being of type int or of type real. If the variable is of type real, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable n will have the value zero (either as an integer or a floating-point value, depending on its type).\nSyntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).\nBecause the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing\nn = n + 1;\nSuch self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.\nThe left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if Sigma is of type matrix, then\nSigma[1, 1] = 1.0;\nsets the value in the first column of the first row of Sigma to one.\nAssignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, a is an array of type array[,] real and b is an array of type array[] real, then the following two statements are both well-formed.\na[3] = b;\nb = a[4];\nSimilarly, if x is a variable declared to have type row_vector and Y is a variable declared as type matrix, then the following sequence of statements to swap the first two rows of Y is well formed.\nx = Y[1];\nY[1] = Y[2];\nY[2] = x;\n\n\nStan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type int to an lvalue of type real, and we can assign an expression of type real to an lvalue of type complex. Furthermore, promotion is transitive, so that we can assign an expression of type int to an lvalue of type complex.\nPromotion extends to containers, so that arrays of int can be promoted to arrays of real during assignment, and arrays of real can be assigned to an lvalue of type array of complex. Similarly, an expression of type vector may be assigned to an lvalue of type complex_vector, and similarly for row vectors and matrices.\n\n\n\nThe expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,\n\na variable, or\na variable with one or more indices, or\na comma separated list of lvalues surrounded by ( and )\n\nTo be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.\n\n\n\nMultiple indexes, as described in the multi-indexing section, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.\n\n\nAll assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.\n\n\n\n\nStan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.\nreal x = 5;\nx += 7;  // value of x is now 12\nThe compound arithmetic and assignment statement above is equivalent to the following long form.\nx = x + 7;\nIn general, the compound form\nx op= y\nwill be equivalent to\nx = x op y;\nThe compound statement will be legal whenever the long form is legal. This requires that the operation x op y must itself be well formed and that the result of the operation be assignable to x. For the expression x to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.\nvector[N] x;\nx[3] += 2;\nAs a further example, consider\nmatrix[M, M] x;\nvector[M] y;\nreal z;\nx *= x;  // OK, (x * x) is a matrix\nx *= z;  // OK, (x * z) is a matrix\nx *= y;  // BAD, (x * y) is a vector\nThe supported compound arithmetic and assignment operations are listed in the compound arithmetic/assignment table; they are also listed in the index prefaced by operator, e.g., operator+=.\nCompound Arithmetic/Assignment Table.  Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.\n\n\n\noperation\ncompound\nunfolded\n\n\n\n\naddition\nx += y\nx = x + y\n\n\nsubtraction\nx -= y\nx = x - y\n\n\nmultiplication\nx *= y\nx = x * y\n\n\ndivision\nx /= y\nx = x / y\n\n\nelementwise multiplication\nx .*= y\nx = x .* y\n\n\nelementwise division\nx ./= y\nx = x ./ y",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#increment-log-prob.section",
+    "href": "reference-manual/statements.html#increment-log-prob.section",
+    "title": "Statements",
+    "section": "",
+    "text": "The basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.1\ntarget += -0.5 * y * y;\nThe keyword target here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).\nIn this example, the unnormalized log probability of a unit normal variable \\(y\\) is added to the total log probability. In the general case, the argument can be any expression.2\nAn entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.\nparameters {\n  real y;\n}\nmodel {\n  target += -0.5 * y * y;\n}\nThis model defines a log probability function\n\\[\n\\log p(y) = - \\, \\frac{y^2}{2} - \\log Z\n\\]\nwhere \\(Z\\) is a normalizing constant that does not depend on \\(y\\). The constant \\(Z\\) is conventionally written this way because on the linear scale, \\[\np(y) = \\frac{1}{Z} \\exp\\left(-\\frac{y^2}{2}\\right).\n\\] which is typically written without reference to \\(Z\\) as \\[\np(y) \\propto \\exp\\left(-\\frac{y^2}{2}\\right).\n\\]\nStan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant \\(Z\\) is either time-consuming to compute or intractable to evaluate.\n\n\nThe built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.\nFor instance, the normal_lpdf function returns the log density of a normal distribution:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nThe normal_lupdf function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant sigma term, normal_lupdf would be equivalent to:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nAll functions ending in _lpdf have a corresponding _lupdf version which evaluates and returns the unnormalized density. The same is true for _lpmf and _lupmf.\n\n\n\nThe increment log density statement looks syntactically like compound addition and assignment (see the compound arithmetic/assignment section, it is treated as a primitive statement because target is not itself a variable. So, even though\ntarget += lp;\nis a legal statement, the corresponding long form is not legal.\ntarget = target + lp;  // BAD, target is not a variable\n\n\n\nThe target += ... statement accepts an argument in place of ... for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment",
+    "href": "reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment",
+    "title": "Statements",
+    "section": "",
+    "text": "A variant of the target += statement described above is the jacobian += statement. This can be used in the transformed parameters block or in functions ending with _jacobian to mimic the log Jacobian adjustments accrued by built-in variable transforms.\nSimilarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.\nFor example, here is a program which recreates the existing &lt;upper=x&gt; transform on real numbers:\nfunctions {\n  real my_upper_bound_jacobian(real x, real ub) {\n    jacobian += x;\n    return ub - exp(x);\n  }\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  // use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters\n  // e.g.\n  // b ~ lognormal(0, 1);\n}\n\n\nTo access the accumulated log density up to the current execution point, the function target() may be used.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#sampling-statements.section",
+    "href": "reference-manual/statements.html#sampling-statements.section",
+    "title": "Statements",
+    "section": "",
+    "text": "The term “sampling statement” has been replaced with distribution statement.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#distribution-statements.section",
+    "href": "reference-manual/statements.html#distribution-statements.section",
+    "title": "Statements",
+    "section": "",
+    "text": "Stan supports writing probability statements also using distribution statements, for example\ny ~ normal(mu, sigma);\nmu ~ normal(0, 10);\nsigma ~ normal(0, 1);\nThe symbol \\(\\sim\\) is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.\nIn general, we can read \\(\\sim\\) as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as \\[\n\\begin{aligned}\n   p(y| \\mu, \\sigma) & = \\mathrm{normal}(y |  \\mu, \\sigma)\\\\\n   p(\\mu) & = \\mathrm{normal}(\\mu |  0, 10)\\\\\n   p(\\sigma) & = \\mathrm{normal}^+(\\sigma |  0, 1).\n\\end{aligned}\n\\] A collection of distribution statements define a joint distribution as the product of component distributions \\[\np(y,\\mu,\\sigma) = p(y| \\mu, \\sigma )p(\\mu) p(\\sigma).\n\\]\nThis works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:\n  a ~ normal(0, 1);\n  a ~ normal(0, 1);\nThis is translated to \\[\n    p(a) = \\mathrm{normal}(a | 0, 1)\\mathrm{normal}(a |  0, 1),\n\\] which in this case is \\(\\mathrm{normal}(a|0,1/\\sqrt{2})\\). One might expect that the above two lines of code would represent a redundant expression of a \\(\\mathrm{normal}(a|0,1)\\) prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.\nWhen the joint distribution is considered as a function of parameters (e.g. \\(\\mu\\), \\(\\sigma\\)) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.\nStan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each \\(\\sim\\) statement corresponds to a multiplicative factor in the unnormalized posterior density.\nDistribution statements (~) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).\nIn Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as\ntarget += normal_lpdf(y | mu, sigma);\ntarget += normal_lpdf(mu | 0, 10);\ntarget += normal_lpdf(sigma | 0, 1);\nStan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g. censored data model) or coding language limitations (e.g. mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.\nIn general, a distribution statement of the form\ny ~ dist(theta1, ..., thetaN);\ninvolving subexpressions y and theta1 through thetaN (including the case where N is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real y values, the log probability density function is used,\ntarget += dist_lpdf(y | theta1, ..., thetaN);\nFor those restricted to integer y values, the log probability mass function is used,\ntarget += dist_lpmf(y | theta1, ..., thetaN);\nThis will be well formed if and only if dist_lpdf(y | theta1,   ..., thetaN) or dist_lpmf(y | theta1, ..., thetaN) is a well-formed expression of type real. User defined distributions can be defined in functions block by using function names ending with _lpdf.\n\n\nAlthough both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in\ny ~ normal(mu, sigma);\nand explicitly incrementing the log probability function, as in\ntarget += normal_lpdf(y | mu, sigma);\nThe distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to normal_lpdf adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, y, mu, or sigma, involve only constants, data variables, and transformed data variables. See the section Built in distributions above discussing _lupdf and _lupmf functions that also drops all the constant terms.\n\n\n\nThe left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write\nparameters {\n  real&lt;lower=0&gt; beta;\n}\n// ...\nmodel {\n  log(beta) ~ normal(mu, sigma);\n}\nUnfortunately, this is not enough to properly model beta as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to beta here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure beta gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the change of variables section for full definitions. For the case above, the following adjustment will account for the log transform.3\ntarget += - log(abs(y));\n\n\n\nStan supports truncating distributions with lower bounds, upper bounds, or both.\n\n\nA probability density function \\(p(x)\\) for a continuous distribution may be truncated to an interval \\([a, b]\\) to define a new density \\(p_{[a, b]}(x)\\) with support \\([a, b]\\) by setting\n\\[\np_{[a, b]}(x)\n= \\frac{p(x)}\n       {\\int_a^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) for a discrete distribution may be truncated to the closed interval \\([a, b]\\) by\n\\[\np_{[a, b]}(x) = \\frac{p(x)}\n                  {\\sum_{u = a}^b p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x)\n= \\frac{p(x)}\n       {\\int_a^{\\infty} p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x) = \\frac{p(x)}\n                  {\\sum_{a &lt;= u} p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty, b]}(x)\n= \\frac{p(x)}\n       {\\int_{-\\infty}^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty,b]}(x) = \\frac{p(x)}\n                  {\\sum_{u &lt;= b} p(u)}.\n\\]\n\n\n\nGiven a probability function \\(p_X(x)\\) for a random variable \\(X\\), its cumulative distribution function (cdf) \\(F_X(x)\\) is defined to be the probability that \\(X \\leq x\\),\n\\[\nF_X(x) = \\Pr[X \\leq x].\n\\]\nThe upper-case variable \\(X\\) is the random variable whereas the lower-case variable \\(x\\) is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to\n\\[\nF_X(x) \\ = \\ \\int_{-\\infty}^{x} p_X(u) \\, du,\n\\]\nFor discrete variables, the cdf is defined to include the upper bound given by the argument,\n\\[\nF_X(x) = \\sum_{u \\leq x} p_X(u).\n\\]\n\n\n\nThe complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by\n\\[\nF^C_X(x)\n\\ = \\ \\Pr[X &gt; x]\n\\ = \\ 1 - F_X(x).\n\\]\nUnlike the cdf, the ccdf is exclusive of the bound, hence the event \\(X &gt; x\\) rather than the cdf’s event \\(X \\leq x\\).\nFor continuous distributions, the ccdf works out to\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\int_{-\\infty}^x p_X(u) \\, du\n\\ = \\ \\int_x^{\\infty} p_X(u) \\, du.\n\\]\nThe lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over \\(u &gt; x\\),\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\sum_{u \\leq x} p_X(u)\n\\ = \\ \\sum_{u &gt; x} p_X(u).\n\\]\nCumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by \\[\n\\int_a^b p(u) \\, du = F_X(b) - F_X(a).\n\\] This allows truncated distributions to be defined as \\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a)}.\n\\]\nFor discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from \\(F_X(b) - F_X(a)\\),\n\\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a) + p_X(a)}.\n\\]\n\n\n\nStan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to \\([-0.5, 2.1]\\) can be coded with the following distribution statement.\ny ~ normal(0, 1) T[-0.5, 2.1];\nTruncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.\nIn general, the truncation bounds and parameters may be parameters or local variables.\nBecause the example above involves a continuous distribution, it behaves the same way as the following more verbose form.\ny ~ normal(0, 1);\nif (y &lt; -0.5 || y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -log_diff_exp(normal_lcdf(2.1 | 0, 1),\n                          normal_lcdf(-0.5 | 0, 1));\n}\nBecause a Stan program defines a log density function, all calculations are on the log scale. The function normal_lcdf is the log of the cumulative normal distribution function and the function log_diff_exp(a, b) is a more arithmetically stable form of log(exp(a) - exp(b)).\nFor a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution\ny ~ poisson(3.7) T[2, 10];\nbehaves in the same way as the following code.\ny ~ poisson(3.7);\nif (y &lt; 2 || y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         log_diff_exp(poisson_lcdf(10 | 3.7),\n                                      poisson_lcdf(2 | 3.7)));\n}\nRecall that log_sum_exp(a, b) is just the arithmetically stable form of log(exp(a) + exp(b)).\n\n\n\nFor truncating with only a lower bound, the upper limit is left blank.\ny ~ normal(0, 1) T[-0.5, ];\nThis truncated distribution statement has the same behavior as the following code.\ny ~ normal(0, 1);\nif (y &lt; -0.5) {\n  target += negative_infinity();\n} else {\n  target += -normal_lccdf(-0.5 | 0, 1);\n}\nThe normal_lccdf function is the normal complementary cumulative distribution function.\nAs with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus\ny ~ poisson(3.7) T[2, ];\nbehaves the same way as\ny ~ poisson(3.7);\nif (y &lt; 2) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         poisson_lccdf(2 | 3.7));\n}\n\n\n\nTo truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement\ny ~ normal(0, 1) T[ , 2.1];\nproduces the same result as the following code.\ntarget += normal_lpdf(y | 0, 1);\nif (y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -normal_lcdf(2.1 | 0, 1);\n}\nWith only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement\ny ~ poisson(3.7) T[ , 10];\nbehaves the same way as the following code.\ny ~ poisson(3.7);\nif (y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -poisson_lcdf(10 | 3.7);\n}\n\n\n\nIn all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.\n\n\n\nFor continuous distributions, truncation points must be expressions of type int or real. For discrete distributions, truncation points must be expressions of type int.\n\n\n\nFor a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds \\(-\\infty\\) to the total log probability, which in turn results in the sample being rejected.\n\n\n\nVectorization of distribution functions with truncation is available if the underlying distribution, lcdf, and lccdf functions meet the required signatures.\nThe equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):\n\nIf the variate y is the only non-scalar, the result is the same as described in the above sections, but the lcdf/lccdf calculation is multiplied by size(y).\nIf the other arguments to the distribution are non-scalars, then the vectorized version of the lcdf/lccdf is used. These functions return the sum of their terms, so no multiplication by the size is needed.\nThe exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the log_diff_exp of two sums is not the same as the sum of the pairwise log_diff_exp operations.\n\nNote that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#for-loops",
+    "href": "reference-manual/statements.html#for-loops",
+    "title": "Statements",
+    "section": "",
+    "text": "Suppose N is a variable of type int, y is a one-dimensional array of type array[] real, and mu and sigma are variables of type real. Furthermore, suppose that n has not been defined as a variable. Then the following is a well-formed for-loop statement.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nThe loop variable is n, the loop bounds are the values in the range 1:N, and the body is the statement following the loop bounds.\n\n\nThe type of the loop variable is int. Unlike in C++ and similarly to R, this variable must not be declared explicitly.\nThe bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range L:H will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to L and less than or equal to H. For example, the loop for (n in 2:5) will cause the body of the for loop to be executed with n equal to 2, 3, 4, and 5, in order. The variable and bound for (n in   5:2) will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.\nThe scope of the loop variable is limited to the body of the loop.\n\n\n\nUnlike in BUGS, Stan allows variables to be reassigned. For example, the variable theta in the following program is reassigned in each iteration of the loop.\nfor (n in 1:N) {\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nSuch reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.4\nIn Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.\nfor (n in 1:N) {\n  y[n] ~ bernoulli(theta);\n  theta = inv_logit(alpha + x[n] * beta);\n}\nIn this program, theta is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.\nStan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.\ntotal = 0.0;\nfor (n in 1:N) {\n  total = total + x[n];\n}\nAfter the for loop is executed, the variable total will hold the sum of the elements in the array x. This example was purely pedagogical; it is easier and more efficient to write\ntotal = sum(x);\nA variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.\nfor (n in 1:100) {\n  y += y * epsilon;\n  epsilon = 0.5 * epsilon;\n  y += y * epsilon;\n}",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#foreach-loops",
+    "href": "reference-manual/statements.html#foreach-loops",
+    "title": "Statements",
+    "section": "",
+    "text": "A second form of for loops allows iteration over elements of containers. If ys is an expression denoting a container (vector, row vector, matrix, or array) with elements of type T, then the following is a well-formed foreach statement.\nfor (y in ys) {\n  // ... do something with y ...\n}\nThe order in which elements of ys are visited is defined for container types as follows.\n\nvector, row_vector: elements visited in order, y is of type double\nmatrix: elements visited in column-major order, y is of type double\narray[] T: elements visited in order, y is of type T.\n\nConsequently, if ys is a two dimensional array array[,] real, y will be a one-dimensional array of real values (type array[] real). If ’ysis a matrix, thenywill be a real value (typereal`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,\narray[2, 3] real yss;\nfor (ys in yss) {\n  for (y in ys) {\n    // ... do something with y ...\n  }\n}\nwhereas a matrix can be looped over in one foreach statement\nmatrix[2, 3] yss;\nfor (y in yss) {\n   // ... do something with y...\n}\nIn both cases, the loop variable y is of type real.  The elements of the matrix are visited in column-major order (e.g.,y[1, 1],y[2, 1],y[1, 2], ...,y[2, 3]), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,y[1, 1],y[1, 2],y[1, 3],y[2, 1], ...,y[2, 3]`).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#conditional-statements",
+    "href": "reference-manual/statements.html#conditional-statements",
+    "title": "Statements",
+    "section": "",
+    "text": "Stan supports full conditional statements using the same if-then-else syntax as C++. The general format is\nif (condition1)\n  statement1\nelse if (condition2)\n  statement2\n// ...\nelse if (conditionN-1)\n  statementN-1\nelse\n  statementN\nThere must be a single leading if clause, which may be followed by any number of else if clauses, all of which may be optionally followed by an else clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.\nThe entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#while-statements",
+    "href": "reference-manual/statements.html#while-statements",
+    "title": "Statements",
+    "section": "",
+    "text": "Stan supports standard while loops using the same syntax as C++. The general format is as follows.\nwhile (condition)\n  body\nThe condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).\nEvaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.\nThe rest of the body of a while loop may be skipped using a continue. The loop will be exited with a break statement. See the section on continue and break statements for more details.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#statement-blocks-and-local-variable-declarations",
+    "href": "reference-manual/statements.html#statement-blocks-and-local-variable-declarations",
+    "title": "Statements",
+    "section": "",
+    "text": "Just as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.\n\n\nBlocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nTo put multiple statements inside the body of a for loop, a block is used, as in the following example.\nfor (n in 1:N) {\n  lambda[n] ~ gamma(alpha, beta);\n  y[n] ~ poisson(lambda[n]);\n}\nThe open curly bracket ({) is the first character of the block and the close curly bracket (}) is the last character.\nBecause whitespace is ignored in Stan, the following program will not compile.\nfor (n in 1:N)\n  y[n] ~ normal(mu, sigma);\n  z[n] ~ normal(mu, sigma); // ERROR!\nThe problem is that the body of the for loop is taken to be the statement directly following it, which is y[n] ~ normal(mu, sigma). This leaves the probability statement for z[n] hanging, as is clear from the following equivalent program.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nz[n] ~ normal(mu, sigma); // ERROR!\nNeither of these programs will compile. If the loop variable n was defined before the for loop, the for-loop declaration will raise an error. If the loop variable n was not defined before the for loop, then the use of the expression z[n] will raise an error.\n\n\n\nA for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.\nfor (n in 1:N) {\n  real theta;\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nThe local variable theta is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus theta is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable theta if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.\nfor (m in 1:M) {\n  real theta;\n  for (n in 1:N) {\n    real theta; // ERROR!\n    theta = inv_logit(alpha + x[m, n] * beta);\n    y[m, n] ~ bernoulli(theta);\n// ...\nThe compiler will flag the second declaration of theta with a message that it is already defined.\n\n\n\nLocal variables may not have constraints on their declaration. The only types that may be used are listed in the types table under “local”.\n\n\n\nA block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following\nfor (m in 1:M) {\n  {\n     int n = 2 * m;\n     sum += n;\n  }\n  for (n in 1:N) {\n    sum += x[m, n];\n  }\n}\nThe variable declaration int n; is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of n are unrelated.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#break-continue-statements",
+    "href": "reference-manual/statements.html#break-continue-statements",
+    "title": "Statements",
+    "section": "",
+    "text": "The one-token statements continue and break may be used within loops to alter control flow; continue causes the next iteration of the loop to run immediately, whereas break terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both break and continue scope to the most deeply nested loop, but pass through non-loop statements.\nAlthough these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.\n\n\nWhen a break statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:\nwhile (1) {\n  if (n &lt; 0) {\n    break;\n  }\n  foo(n);\n  n = n - 1;\n}\nThe while~(1) loop is a “forever” loop, because 1 is the true value, so the test always succeeds. Within the loop, if the value of n is less than 0, the loop terminates, otherwise it executes foo(n) and then decrements n. The statement above does exactly the same thing as\nwhile (n &gt;= 0) {\n  foo(n);\n  n = n - 1;\n}\nThis case is simply illustrative of the behavior; it is not a case where a break simplifies the loop.\n\n\n\nThe continue statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array x,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &lt;= 0) {\n    continue;\n  }\n  sum += x[n];\n}\nWhen the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an x[n] less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &gt; 0) {\n    sum += x[n];\n  }\n}\nWhile the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.\n\n\n\nIf there is a loop nested within a loop, a break or continue statement only breaks out of the inner loop. So\nwhile (cond1) {\n  // ...\n  while (cond2) {\n    // ...\n    if (cond3) {\n      break;\n    }\n    // ...\n  }\n  // execution continues here after break\n  // ...\n}\nIf the break is triggered by cond3 being true, execution will continue after the nested loop.\nAs with break statements, continue statements go back to the top of the most deeply nested loop in which the continue appears.\nAlthough break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#print-statements.section",
+    "href": "reference-manual/statements.html#print-statements.section",
+    "title": "Statements",
+    "section": "",
+    "text": "Stan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.\nfor (n in 1:N) { print(\"loop iteration: \", n); ... }\nThe print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression n, followed by a new line.\n\n\nThe text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.\nPrinting a simple real or int typed variable always prints the variable’s value.5\nFor array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as\n[1, 2, 3]\nand a \\(2 \\times 3\\)-matrix as\n[[1, 2, 3], [4, 5, 6]]\nComplex numbers print as pairs. For example, the pair of statements\ncomplex z = to_complex(1.2, -3.5);\nprint(z)\nwill print as (1.2,-3.5), with no space after the comma or within the parentheses.\nPrinting a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.\ntransformed data {\n  matrix[2, 2] u;\n  u[1, 1] = 1.0;   u[1, 2] = 4.0;\n  u[2, 1] = 9.0;   u[2, 2] = 16.0;\n  for (n in 1:2) {\n    print(\"u[\", n, \"] = \", u[n]);\n  }\n}\nThis print statement executes twice, printing the following two lines of output.\nu[1] = [1, 4]\nu[2] = [9, 16]\n\n\n\nThe input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.\n\n\n\nPrinting for a print statement happens every time it is executed. The transformed data block is executed once per chain, the transformed parameter and model blocks once per leapfrog step, and the generated quantities block once per iteration.\n\n\n\nString literals begin and end with a double quote character (\"). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.\nThe Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.\n\n\n\nBecause Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.\nIt is particularly useful to print the value of the target log density accumulator (through the target() function), as in the following example.\nvector[2] y;\ny[1] = 1;\nprint(\"log density before =\", target());\ny ~ normal(0,1);  // bug!  y[2] not defined\nprint(\"log density after =\", target());\nThe example has a bug in that y[2] is not defined before the vector y is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).\nNote that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#reject-statements.section",
+    "href": "reference-manual/statements.html#reject-statements.section",
+    "title": "Statements",
+    "section": "",
+    "text": "The Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.\nLike the print statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.\nReject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable x’s value is negative.\nif (x &lt; 0) {\n  reject(\"x must not be negative; found x=\", x);\n}\n\n\nReject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the normal_lpdf function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.\nIn all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.\n\n\nRejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.\n\n\n\nRejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.\nReject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.\nFatal errors in other blocks may also be signaled by use of the fatal_error statement.\n\n\n\nRejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.\nIf the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.\n\n\n\n\nRejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;\n  real&lt;lower=a, upper=b&gt; theta;\n  // ...\n}\nmodel {\n  // **wrong** needs explicit truncation\n  theta ~ normal(0, 1);\n  // ...\n}\nThis program is wrong because its truncation bounds on theta depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.\n  theta ~ normal(0, 1) T[a, b];\nThe conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over \\([a ,b]\\); in these simple univariate cases, we can overcome that with the T[ , ] notation, which essentially divides by whatever the prior integrates to over \\([a, b]\\).\nThis problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.\n  if (theta &lt; a || theta &gt; b) {\n    reject(\"theta not in (a, b)\");\n  }\n  // still **wrong**, needs T[a,b]\n  theta ~ normal(0, 1);\nIn this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.\nEven if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#fatal-error-statements",
+    "href": "reference-manual/statements.html#fatal-error-statements",
+    "title": "Statements",
+    "section": "",
+    "text": "The Stan fatal_error statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.\nLike the print or reject statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.\nThe fatal error may be used to signal an unrecoverable error in blocks where reject leads to the algorithm attempting to try again, such as the model block.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/statements.html#footnotes",
+    "href": "reference-manual/statements.html#footnotes",
+    "title": "Statements",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe current notation replaces two previous versions. Originally, a variable lp__ was directly exposed and manipulated; this is no longer allowed. The original statement syntax for target += u was increment_log_prob(u), but this form was removed in Stan 2.33↩︎\nWriting this model with the expression -0.5 * y * y is more efficient than with the equivalent expression y * y / -2 because multiplication is more efficient than division; in both cases, the negation is rolled into the numeric literal (-0.5 and -2). Writing square(y) instead of y * y would be even more efficient because the derivatives can be precomputed, reducing the memory and number of operations required for automatic differentiation.↩︎\nBecause \\(\\log | \\frac{d}{dy} \\log y | = \\log | 1/y | = - \\log |y|\\).↩︎\nA programming idiom in BUGS code simulates a local variable by replacing theta in the above example with theta[n], effectively creating N different variables, theta[1], …, theta[N]. Of course, this is not a hack if the value of theta[n] is required for all n.↩︎\nThe adjoint component is always zero during execution for the algorithmic differentiation variables used to implement parameters, transformed parameters, and local variables in the model.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Statements"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html",
+    "href": "reference-manual/removals.html",
+    "title": "Removed Features",
+    "section": "",
+    "text": "This chapter lists functionalities that were once present in the language but have since been removed, along with how to replace them.\n\n\nRemoved: The variable lp__ is no longer available for direct access or manipulation.\nReplacement: General manipulation of the value of the lp__ variable is not allowed, but\nlp__ &lt;- lp__ + e;\ncan be replaced with\ntarget += e;\nThe value of lp__ is available through the no-argument function target().\n\n\n\nRemoved: The operator &lt;- for assignment, e.g.,\na &lt;- b;\nis no longer available.\nReplacement: The new syntax uses the operator = for assignment, e.g.,\na = b;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The increment_log_prob(u) statement for incrementing the log density accumulator by u is no longer available.\nReplacement: Replace the above statement with\ntarget += u;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The built-in no-argument function get_lp() is no longer available.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33\n\n\n\nRemoved: Formerly, the probability function for the distribution foo would be applied to an outcome variable y and sequence of zero or more parameters ... to produce the expression foo_log(y, ...). This suffix is no longer a special value.\nReplacement: If y can be a real value (including vectors or matrices), replace\nfoo_log(y, ...)\nwith the log probability density function notation\nfoo_lpdf(y | ...).\nIf y must be an integer (including arrays), instead replace\nfoo_log(y, ...\nwith the log probability mass function\nfoo_lpmf(y | ...).\nRemoved In: Stan 2.33\n\n\n\nRemoved: The log cumulative distribution and complementary cumulative distribution functions for a distribution foo were formerly written as foo_cdf_log and foo_ccdf_log.\nReplacement:\nReplace foo_cdf_log(y, ...) with foo_lcdf(y | ...).\nReplace foo_ccdf_log(y, ...) with foo_lccdf(y | ...).\n\n\n\nRemoved: A user-defined function ending in _log can be no longer be used in statements.qmd#distribution-statements.section.\nReplacement: Replace the _log suffix with _lpdf for density functions or _lpmf for mass functions in the user-defined function.\nRemoved In: Stan 2.33\nNote: Following Stan 2.33, users can stil define a function ending in _log, it simply no longer has a special meaning or is supported in the ~ syntax.\n\n\n\nRemoved: The function if_else is no longer available.\nReplacement: Use the conditional operator which allows more flexibility in the types of b and c and is much more efficient in that it only evaluates whichever of b or c is returned.\nx = if_else(a, b, c);\nwith\nx = a ? b : c;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The use of # for line-based comments is no longer permitted. # may only be used for #include statements.\nReplacement: Use a pair of forward slashes, //, for line comments.\nRemoved In: Stan 2.33\n\n\n\nBefore Stan 2.26, arrays were declared by writing syntax after the variable.\nRemoved: The use of array declarations like\nint n[5];\nreal a[3, 4];\nreal&lt;lower=0&gt; z[5, 4, 2];\nvector[7] mu[3];\nmatrix[7, 2] mu[15, 12];\ncholesky_factor_cov[5, 6] mu[2, 3, 4];\nReplacement: The use of the array keyword, which replaces the above examples with\narray[5] int n;\narray[3, 4] real a;\narray[5, 4, 2] real&lt;lower=0&gt; z;\narray[3] vector[7] mu;\narray[15, 12] matrix[7, 2] mu;\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nRemoved In: Stan 2.33\n\n\n\nStan interprets nested indexing in assingments as flat indexing so that a statement like\na[:][1] = b;\nis the same as\na[:,1] = b;\nHowever, this is inconsistent with multiple indexing rules.\nTo avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.\nRemoved In: Stan 2.33\n\n\n\nRemoved: Using a real value in a conditional is no longer permitted.\nreal x = 1.0;\nif (x) {\nThe value was interpreted as true if it is nonzero.\nReplacement: For the exact equivalent, use a comparison operator to make the intent clear.\nreal x = 1.0;\nif (x != 0) {\nHowever, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative abs(x) &lt; machine_precision() is appropriate for the use case.\nRemoved In: Stan 2.34",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#lp__-variable",
+    "href": "reference-manual/removals.html#lp__-variable",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The variable lp__ is no longer available for direct access or manipulation.\nReplacement: General manipulation of the value of the lp__ variable is not allowed, but\nlp__ &lt;- lp__ + e;\ncan be replaced with\ntarget += e;\nThe value of lp__ is available through the no-argument function target().",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#assignment-with--",
+    "href": "reference-manual/removals.html#assignment-with--",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The operator &lt;- for assignment, e.g.,\na &lt;- b;\nis no longer available.\nReplacement: The new syntax uses the operator = for assignment, e.g.,\na = b;\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#increment_log_prob-statement",
+    "href": "reference-manual/removals.html#increment_log_prob-statement",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The increment_log_prob(u) statement for incrementing the log density accumulator by u is no longer available.\nReplacement: Replace the above statement with\ntarget += u;\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#get_lp-function",
+    "href": "reference-manual/removals.html#get_lp-function",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The built-in no-argument function get_lp() is no longer available.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#log-density-and-mass-functions",
+    "href": "reference-manual/removals.html#log-density-and-mass-functions",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: Formerly, the probability function for the distribution foo would be applied to an outcome variable y and sequence of zero or more parameters ... to produce the expression foo_log(y, ...). This suffix is no longer a special value.\nReplacement: If y can be a real value (including vectors or matrices), replace\nfoo_log(y, ...)\nwith the log probability density function notation\nfoo_lpdf(y | ...).\nIf y must be an integer (including arrays), instead replace\nfoo_log(y, ...\nwith the log probability mass function\nfoo_lpmf(y | ...).\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#cdf_log-and-ccdf_log-cumulative-distribution-functions",
+    "href": "reference-manual/removals.html#cdf_log-and-ccdf_log-cumulative-distribution-functions",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The log cumulative distribution and complementary cumulative distribution functions for a distribution foo were formerly written as foo_cdf_log and foo_ccdf_log.\nReplacement:\nReplace foo_cdf_log(y, ...) with foo_lcdf(y | ...).\nReplace foo_ccdf_log(y, ...) with foo_lccdf(y | ...).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#user-defined-function-with-_log-suffix",
+    "href": "reference-manual/removals.html#user-defined-function-with-_log-suffix",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: A user-defined function ending in _log can be no longer be used in statements.qmd#distribution-statements.section.\nReplacement: Replace the _log suffix with _lpdf for density functions or _lpmf for mass functions in the user-defined function.\nRemoved In: Stan 2.33\nNote: Following Stan 2.33, users can stil define a function ending in _log, it simply no longer has a special meaning or is supported in the ~ syntax.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#if_else-function",
+    "href": "reference-manual/removals.html#if_else-function",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The function if_else is no longer available.\nReplacement: Use the conditional operator which allows more flexibility in the types of b and c and is much more efficient in that it only evaluates whichever of b or c is returned.\nx = if_else(a, b, c);\nwith\nx = a ? b : c;\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#character-as-comment-prefix",
+    "href": "reference-manual/removals.html#character-as-comment-prefix",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The use of # for line-based comments is no longer permitted. # may only be used for #include statements.\nReplacement: Use a pair of forward slashes, //, for line comments.\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#postfix-brackets-array-syntax",
+    "href": "reference-manual/removals.html#postfix-brackets-array-syntax",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Before Stan 2.26, arrays were declared by writing syntax after the variable.\nRemoved: The use of array declarations like\nint n[5];\nreal a[3, 4];\nreal&lt;lower=0&gt; z[5, 4, 2];\nvector[7] mu[3];\nmatrix[7, 2] mu[15, 12];\ncholesky_factor_cov[5, 6] mu[2, 3, 4];\nReplacement: The use of the array keyword, which replaces the above examples with\narray[5] int n;\narray[3, 4] real a;\narray[5, 4, 2] real&lt;lower=0&gt; z;\narray[3] vector[7] mu;\narray[15, 12] matrix[7, 2] mu;\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#nested-multiple-indexing-in-assignments",
+    "href": "reference-manual/removals.html#nested-multiple-indexing-in-assignments",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Stan interprets nested indexing in assingments as flat indexing so that a statement like\na[:][1] = b;\nis the same as\na[:,1] = b;\nHowever, this is inconsistent with multiple indexing rules.\nTo avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/removals.html#real-values-in-conditionals",
+    "href": "reference-manual/removals.html#real-values-in-conditionals",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: Using a real value in a conditional is no longer permitted.\nreal x = 1.0;\nif (x) {\nThe value was interpreted as true if it is nonzero.\nReplacement: For the exact equivalent, use a comparison operator to make the intent clear.\nreal x = 1.0;\nif (x != 0) {\nHowever, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative abs(x) &lt; machine_precision() is appropriate for the use case.\nRemoved In: Stan 2.34",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Removed Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/pathfinder.html",
+    "href": "reference-manual/pathfinder.html",
+    "title": "Pathfinder",
+    "section": "",
+    "text": "Stan supports the Pathfinder algorithm (Zhang et al. 2022). Pathfinder is a variational method for approximately sampling from differentiable log densities. Starting from a random initialization, Pathfinder locates normal approximations to the target density along a quasi-Newton optimization path, with local covariance estimated using the negative inverse Hessian estimates produced by the LBFGS optimizer. Pathfinder returns draws from the Gaussian approximation with the lowest estimated Kullback-Leibler (KL) divergence to the true posterior.\nStan provides two versions of the Pathfinder algorithm: single-path Pathfinder and multi-path Pathfinder. Single-path Pathfinder generates a set of approximate draws from one run of the basic Pathfinder algorithm. Multi-path Pathfinder uses importance resampling over the draws from multiple runs of Pathfinder. This better matches non-normal target densities and also mitigates the problem of L-BFGS getting stuck at local optima or in saddle points on plateaus. Compared to ADVI and short dynamic HMC runs, Pathfinder requires one to two orders of magnitude fewer log density and gradient evaluations, with greater reductions for more challenging posteriors. While the evaluations by Zhang et al. (2022) found that single-path and multi-path Pathfinder outperform ADVI for most of the models in the PosteriorDB (Magnusson et al. 2024) evaluation set, we recognize the need for further experiments on a wider range of models.\n\n\nPathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-\\(\\hat{k}\\) diagnostic (Vehtari et al. 2024) to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably (Vehtari et al. 2024, sec. 3), which is the first requirement for reliable resampling. If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest (Vehtari et al. 2024, sec. 2.2). If estimated Pareto-\\(\\hat{k}\\) is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow (Gelman et al. 2020).\n\n\n\nIf estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-\\(\\hat{k}\\) for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g. uniform from -2 to 2 used by Stan by default). If Pareto-\\(\\hat{k}\\) is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Pathfinder"
+    ]
+  },
+  {
+    "objectID": "reference-manual/pathfinder.html#diagnosing-pathfinder",
+    "href": "reference-manual/pathfinder.html#diagnosing-pathfinder",
+    "title": "Pathfinder",
+    "section": "",
+    "text": "Pathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-\\(\\hat{k}\\) diagnostic (Vehtari et al. 2024) to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably (Vehtari et al. 2024, sec. 3), which is the first requirement for reliable resampling. If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest (Vehtari et al. 2024, sec. 2.2). If estimated Pareto-\\(\\hat{k}\\) is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow (Gelman et al. 2020).",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Pathfinder"
+    ]
+  },
+  {
+    "objectID": "reference-manual/pathfinder.html#using-pathfinder-for-initializing-mcmc",
+    "href": "reference-manual/pathfinder.html#using-pathfinder-for-initializing-mcmc",
+    "title": "Pathfinder",
+    "section": "",
+    "text": "If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-\\(\\hat{k}\\) for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g. uniform from -2 to 2 used by Stan by default). If Pareto-\\(\\hat{k}\\) is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Pathfinder"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html",
+    "href": "reference-manual/mcmc.html",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "This chapter presents the two Markov chain Monte Carlo (MCMC) algorithms used in Stan, the Hamiltonian Monte Carlo (HMC) algorithm and its adaptive variant the no-U-turn sampler (NUTS), along with details of their implementation and configuration.\n\n\nHamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., Betancourt and Girolami (2013), Neal (2011) for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.\nThis section translates the presentation of HMC by Betancourt and Girolami (2013) into the notation of Gelman et al. (2013).\n\n\nThe goal of sampling is to draw from a density \\(p(\\theta)\\) for parameters \\(\\theta\\). This is typically a Bayesian posterior \\(p(\\theta|y)\\) given data \\(y\\), and in particular, a Bayesian posterior coded as a Stan program.\n\n\n\nHMC introduces auxiliary momentum variables \\(\\rho\\) and draws from a joint density\n\\[\np(\\rho, \\theta) = p(\\rho | \\theta) p(\\theta).\n\\]\nIn most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters \\(\\theta\\),\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\n\\(M\\) is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see Betancourt (2017) for details.\nBy default Stan sets \\(M^{-1}\\) equal to a diagonal estimate of the covariance computed during warmup.\n\n\n\nThe joint density \\(p(\\rho, \\theta)\\) defines a Hamiltonian\n\\[\n\\begin{array}{rcl}\nH(\\rho, \\theta) & = & - \\log p(\\rho, \\theta)\n\\\\[3pt]\n& = & - \\log p(\\rho | \\theta) - \\log p(\\theta).\n\\\\[3pt]\n& = & T(\\rho | \\theta) + V(\\theta),\n\\end{array}\n\\]\nwhere the term\n\\[\nT(\\rho | \\theta) = - \\log p(\\rho | \\theta)\n\\]\nis called the “kinetic energy” and the term\n\\[\nV(\\theta) = - \\log p(\\theta)\n\\]\nis called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.\n\n\n\nStarting from the current value of the parameters \\(\\theta\\), a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.\nFirst, a value for the momentum is drawn independently of the current parameter values,\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\nThus momentum does not persist across iterations.\nNext, the joint system \\((\\theta,\\rho)\\) made up of the current parameter values \\(\\theta\\) and new momentum \\(\\rho\\) is evolved via Hamilton’s equations,\n\\[\n\\begin{array}{rcccl}\n\\displaystyle\n\\frac{d\\theta}{dt}\n& = &\n\\displaystyle\n+ \\frac{\\partial H}{\\partial \\rho}\n& = &\n\\displaystyle\n+ \\frac{\\partial T}{\\partial \\rho}\n\\\\[12pt]\n\\displaystyle\n\\frac{d\\rho}{dt}\n& = &\n\\displaystyle\n- \\frac{\\partial H}{\\partial \\theta }\n& = &\n\\displaystyle\n- \\frac{\\partial T}{\\partial \\theta}\n- \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nWith the momentum density being independent of the target density, i.e., \\(p(\\rho | \\theta) = p(\\rho)\\), the first term in the momentum time derivative, \\({\\partial T} / {\\partial \\theta}\\) is zero, yielding the pair time derivatives\n\\[\n\\begin{array}{rcl}\n\\frac{d \\theta}{d t} & = & +\\frac{\\partial T}{\\partial \\rho}\n\\\\[2pt]\n\\frac{d \\rho}{d t} & = & -\\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\n\n\n\nThe last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.\nLike most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval \\(\\epsilon\\). The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values \\(\\theta\\) or previous momentum value.\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\] It then alternates half-step updates of the momentum and full-step updates of the position.\n\\[\n\\begin{array}{rcl}\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}\n\\\\[6pt]\n\\theta & \\leftarrow\n       & \\theta \\, + \\, \\epsilon \\, M^{-1} \\, \\rho\n\\\\[6pt]\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nBy applying \\(L\\) leapfrog steps, a total of \\(L \\, \\epsilon\\) time is simulated. The resulting state at the end of the simulation (\\(L\\) repetitions of the above three steps) will be denoted \\((\\rho^{*}, \\theta^{*})\\).\nThe leapfrog integrator’s error is on the order of \\(\\epsilon^3\\) per step and \\(\\epsilon^2\\) globally, where \\(\\epsilon\\) is the time interval (also known as the step size); Leimkuhler and Reich (2004) provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.\n\n\n\nIf the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal \\((\\rho^{*}, \\theta^{*})\\) generated by transitioning from \\((\\rho, \\theta)\\) is\n\\[\n\\min \\!\n\\left(\n1,\n\\ \\exp \\! \\left( H(\\rho, \\theta) - H(\\rho^{*}, \\theta^{*}) \\right)\n\\right).\n\\]\nIf the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.\n\n\n\nThe Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters \\(\\theta\\); in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter \\(\\theta\\) is updated using the leapfrog integrator with discretization time \\(\\epsilon\\) and number of steps \\(L\\) according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state \\((\\theta^{*}, \\rho^{*})\\) or keep the existing state.\n\n\n\n\nThe Hamiltonian Monte Carlo algorithm has three parameters which must be set,\n\ndiscretization time \\(\\epsilon\\),\nmetric \\(M\\), and\nnumber of steps taken \\(L\\).\n\nIn practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters Neal (2011), Hoffman and Gelman (2014).\nIf \\(\\epsilon\\) is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If \\(\\epsilon\\) is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.\nIf \\(L\\) is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If \\(L\\) is too large, the algorithm will do too much work on each iteration.\nIf the inverse metric \\(M^{-1}\\) is a poor estimate of the posterior covariance, the step size \\(\\epsilon\\) must be kept small to maintain arithmetic precision. This would lead to a large \\(L\\) to compensate.\n\n\nThe actual integration time is \\(L \\, \\epsilon\\), a function of number of steps. Some interfaces to Stan set an approximate integration time \\(t\\) and the discretization interval (step size) \\(\\epsilon\\). In these cases, the number of steps will be rounded down as\n\\[\nL = \\left\\lfloor \\frac{t}{\\epsilon} \\right\\rfloor.\n\\]\nand the actual integration time will still be \\(L \\, \\epsilon\\).\n\n\n\nStan is able to automatically optimize \\(\\epsilon\\) to match an acceptance-rate target, able to estimate \\(M\\) based on warmup sample iterations, and able to dynamically adapt \\(L\\) on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm Hoffman and Gelman (2014).\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\nWhen adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the warmup adaptation figure, with two fast intervals surrounding a series of growing slow intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.\nThe motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.\n\nIn the initial fast interval the chain is allowed to converge towards the typical set,1 with only parameters that can learn from local information adapted.\nAfter this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.\nLastly, the fast parameters are allowed to adapt to the final update of the slow parameters.\n\nThese intervals may be controlled through the following configuration parameters, all of which must be positive integers:\nAdaptation Parameters Table. The parameters controlling adaptation and their default values.\n\n\n\n\n\n\n\n\nparameter\ndescription\ndefault\n\n\n\n\ninitial buffer\nwidth of initial fast adaptation interval\n75\n\n\nterm buffer\nwidth of final fast adaptation interval\n50\n\n\nwindow\ninitial width of slow adaptation interval\n25\n\n\n\n\n\n\nStan’s HMC algorithms utilize dual averaging Nesterov (2009) to optimize the step size.2\nThis warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of Hoffman and Gelman (2014). In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman and Gelman (2014).\nThe full set of dual-averaging parameters are:\nStep Size Adaptation Parameters Table The parameters controlling step size adaptation, with constraints and default values.\n\n\n\nparameter\ndescription\nconstraint\ndefault\n\n\n\n\ndelta\ntarget Metropolis acceptance rate\n[0, 1]\n0.8\n\n\ngamma\nadaptation regularization scale\n(0, infty)\n0.05\n\n\nkappa\nadaptation relaxation exponent\n(0, infty)\n0.75\n\n\nt_0\nadaptation iteration offset\n(0, infty)\n10\n\n\n\nBy setting the target acceptance parameter \\(\\delta\\) to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of \\(\\delta\\) will also allow some models that would otherwise get stuck to overcome their blockages.\n\n\n\nAll implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.\nSmall step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See Neal (2011) for further discussion of step-size jittering.\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nIf the metric is constant then the resulting implementation is known as Euclidean HMC. Stan allows a choice among three Euclidean HMC implementations,\n\na unit metric (diagonal matrix of ones),\na diagonal metric (diagonal matrix with positive diagonal entries), and\na dense metric (a dense, symmetric positive definite matrix)\n\nto be configured by the user.\nIf the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the warmup adaptation stages figure). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.\nIf the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.\nVariances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.\n\n\nThe metric can compensate for linear (i.e. global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.\nIn complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e. local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.\n\n\n\nThe metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.3\nOne of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.\n\n\n\nStatistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.\n\n\n\nMCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.\nThis only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.\nPoor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.\n\n\n\n\nThe no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., Roberts, Gelman, and Gilks (1997)) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see Hoffman and Gelman (2014).\nNUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either\n\nthe NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or\nthe depth of the completed tree hits the maximum depth allowed.\n\nRather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory Betancourt (2016b).4\nConfiguring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.\nBoth the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as treedepth__ and n_leapfrog__, respectively. Because the final subtree may only be partially constructed, these two will always satisfy\n\\[\n2^{\\mathrm{treedepth} - 1} - 1\n\\ &lt; \\\nN_{\\mathrm{leapfrog}}\n\\ \\le \\\n2^{\\mathrm{treedepth} } - 1.\n\\]\nTree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\n\n\n\n\nIn some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.\nFor example, to generate a sequence of \\(N\\) draws from a binomial with trials \\(K\\) and chance of success \\(\\theta\\), the following program suffices.\ndata {\n  real&lt;lower=0, upper=1&gt; theta;\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n}\ngenerated quantities {\n  array[N] int&lt;lower=0, upper=K&gt; y;\n  for (n in 1:N) {\n    y[n] = binomial_rng(K, theta);\n  }\n}\nFor this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.\nMost models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, theta in the example above could be declared as a parameter and initialized as a parameter.\n\n\n\nStan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the optimization algorithms chapter, and the diagnostics chapter.\n\n\nThe random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nStan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.\n\n\nTogether, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.\n\n\n\n\nThe initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.\n\n\nIf the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).\n\n\n\nIt is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.\n\n\n\nRandom initialization by default initializes the parameter values with values drawn at random from a \\(\\mathsf{Uniform}(-2, 2)\\) distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.\nBecause zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\n\n\n\n\n\nThe Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.\nIn Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. Leimkuhler and Reich (2004) provide details and error analysis for the leapfrog algorithm.\nA divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,5 the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. Betancourt (2016a) provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.\nThe Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.\nIf the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.\nThe primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.\n\n\nIn some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.\nBefore reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html#hamiltonian-monte-carlo",
+    "href": "reference-manual/mcmc.html#hamiltonian-monte-carlo",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "Hamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., Betancourt and Girolami (2013), Neal (2011) for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.\nThis section translates the presentation of HMC by Betancourt and Girolami (2013) into the notation of Gelman et al. (2013).\n\n\nThe goal of sampling is to draw from a density \\(p(\\theta)\\) for parameters \\(\\theta\\). This is typically a Bayesian posterior \\(p(\\theta|y)\\) given data \\(y\\), and in particular, a Bayesian posterior coded as a Stan program.\n\n\n\nHMC introduces auxiliary momentum variables \\(\\rho\\) and draws from a joint density\n\\[\np(\\rho, \\theta) = p(\\rho | \\theta) p(\\theta).\n\\]\nIn most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters \\(\\theta\\),\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\n\\(M\\) is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see Betancourt (2017) for details.\nBy default Stan sets \\(M^{-1}\\) equal to a diagonal estimate of the covariance computed during warmup.\n\n\n\nThe joint density \\(p(\\rho, \\theta)\\) defines a Hamiltonian\n\\[\n\\begin{array}{rcl}\nH(\\rho, \\theta) & = & - \\log p(\\rho, \\theta)\n\\\\[3pt]\n& = & - \\log p(\\rho | \\theta) - \\log p(\\theta).\n\\\\[3pt]\n& = & T(\\rho | \\theta) + V(\\theta),\n\\end{array}\n\\]\nwhere the term\n\\[\nT(\\rho | \\theta) = - \\log p(\\rho | \\theta)\n\\]\nis called the “kinetic energy” and the term\n\\[\nV(\\theta) = - \\log p(\\theta)\n\\]\nis called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.\n\n\n\nStarting from the current value of the parameters \\(\\theta\\), a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.\nFirst, a value for the momentum is drawn independently of the current parameter values,\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\nThus momentum does not persist across iterations.\nNext, the joint system \\((\\theta,\\rho)\\) made up of the current parameter values \\(\\theta\\) and new momentum \\(\\rho\\) is evolved via Hamilton’s equations,\n\\[\n\\begin{array}{rcccl}\n\\displaystyle\n\\frac{d\\theta}{dt}\n& = &\n\\displaystyle\n+ \\frac{\\partial H}{\\partial \\rho}\n& = &\n\\displaystyle\n+ \\frac{\\partial T}{\\partial \\rho}\n\\\\[12pt]\n\\displaystyle\n\\frac{d\\rho}{dt}\n& = &\n\\displaystyle\n- \\frac{\\partial H}{\\partial \\theta }\n& = &\n\\displaystyle\n- \\frac{\\partial T}{\\partial \\theta}\n- \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nWith the momentum density being independent of the target density, i.e., \\(p(\\rho | \\theta) = p(\\rho)\\), the first term in the momentum time derivative, \\({\\partial T} / {\\partial \\theta}\\) is zero, yielding the pair time derivatives\n\\[\n\\begin{array}{rcl}\n\\frac{d \\theta}{d t} & = & +\\frac{\\partial T}{\\partial \\rho}\n\\\\[2pt]\n\\frac{d \\rho}{d t} & = & -\\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\n\n\n\nThe last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.\nLike most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval \\(\\epsilon\\). The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values \\(\\theta\\) or previous momentum value.\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\] It then alternates half-step updates of the momentum and full-step updates of the position.\n\\[\n\\begin{array}{rcl}\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}\n\\\\[6pt]\n\\theta & \\leftarrow\n       & \\theta \\, + \\, \\epsilon \\, M^{-1} \\, \\rho\n\\\\[6pt]\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nBy applying \\(L\\) leapfrog steps, a total of \\(L \\, \\epsilon\\) time is simulated. The resulting state at the end of the simulation (\\(L\\) repetitions of the above three steps) will be denoted \\((\\rho^{*}, \\theta^{*})\\).\nThe leapfrog integrator’s error is on the order of \\(\\epsilon^3\\) per step and \\(\\epsilon^2\\) globally, where \\(\\epsilon\\) is the time interval (also known as the step size); Leimkuhler and Reich (2004) provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.\n\n\n\nIf the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal \\((\\rho^{*}, \\theta^{*})\\) generated by transitioning from \\((\\rho, \\theta)\\) is\n\\[\n\\min \\!\n\\left(\n1,\n\\ \\exp \\! \\left( H(\\rho, \\theta) - H(\\rho^{*}, \\theta^{*}) \\right)\n\\right).\n\\]\nIf the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.\n\n\n\nThe Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters \\(\\theta\\); in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter \\(\\theta\\) is updated using the leapfrog integrator with discretization time \\(\\epsilon\\) and number of steps \\(L\\) according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state \\((\\theta^{*}, \\rho^{*})\\) or keep the existing state.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html#hmc-algorithm-parameters",
+    "href": "reference-manual/mcmc.html#hmc-algorithm-parameters",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "The Hamiltonian Monte Carlo algorithm has three parameters which must be set,\n\ndiscretization time \\(\\epsilon\\),\nmetric \\(M\\), and\nnumber of steps taken \\(L\\).\n\nIn practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters Neal (2011), Hoffman and Gelman (2014).\nIf \\(\\epsilon\\) is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If \\(\\epsilon\\) is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.\nIf \\(L\\) is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If \\(L\\) is too large, the algorithm will do too much work on each iteration.\nIf the inverse metric \\(M^{-1}\\) is a poor estimate of the posterior covariance, the step size \\(\\epsilon\\) must be kept small to maintain arithmetic precision. This would lead to a large \\(L\\) to compensate.\n\n\nThe actual integration time is \\(L \\, \\epsilon\\), a function of number of steps. Some interfaces to Stan set an approximate integration time \\(t\\) and the discretization interval (step size) \\(\\epsilon\\). In these cases, the number of steps will be rounded down as\n\\[\nL = \\left\\lfloor \\frac{t}{\\epsilon} \\right\\rfloor.\n\\]\nand the actual integration time will still be \\(L \\, \\epsilon\\).\n\n\n\nStan is able to automatically optimize \\(\\epsilon\\) to match an acceptance-rate target, able to estimate \\(M\\) based on warmup sample iterations, and able to dynamically adapt \\(L\\) on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm Hoffman and Gelman (2014).\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\nWhen adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the warmup adaptation figure, with two fast intervals surrounding a series of growing slow intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.\nThe motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.\n\nIn the initial fast interval the chain is allowed to converge towards the typical set,1 with only parameters that can learn from local information adapted.\nAfter this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.\nLastly, the fast parameters are allowed to adapt to the final update of the slow parameters.\n\nThese intervals may be controlled through the following configuration parameters, all of which must be positive integers:\nAdaptation Parameters Table. The parameters controlling adaptation and their default values.\n\n\n\n\n\n\n\n\nparameter\ndescription\ndefault\n\n\n\n\ninitial buffer\nwidth of initial fast adaptation interval\n75\n\n\nterm buffer\nwidth of final fast adaptation interval\n50\n\n\nwindow\ninitial width of slow adaptation interval\n25\n\n\n\n\n\n\nStan’s HMC algorithms utilize dual averaging Nesterov (2009) to optimize the step size.2\nThis warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of Hoffman and Gelman (2014). In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman and Gelman (2014).\nThe full set of dual-averaging parameters are:\nStep Size Adaptation Parameters Table The parameters controlling step size adaptation, with constraints and default values.\n\n\n\nparameter\ndescription\nconstraint\ndefault\n\n\n\n\ndelta\ntarget Metropolis acceptance rate\n[0, 1]\n0.8\n\n\ngamma\nadaptation regularization scale\n(0, infty)\n0.05\n\n\nkappa\nadaptation relaxation exponent\n(0, infty)\n0.75\n\n\nt_0\nadaptation iteration offset\n(0, infty)\n10\n\n\n\nBy setting the target acceptance parameter \\(\\delta\\) to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of \\(\\delta\\) will also allow some models that would otherwise get stuck to overcome their blockages.\n\n\n\nAll implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.\nSmall step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See Neal (2011) for further discussion of step-size jittering.\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nIf the metric is constant then the resulting implementation is known as Euclidean HMC. Stan allows a choice among three Euclidean HMC implementations,\n\na unit metric (diagonal matrix of ones),\na diagonal metric (diagonal matrix with positive diagonal entries), and\na dense metric (a dense, symmetric positive definite matrix)\n\nto be configured by the user.\nIf the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the warmup adaptation stages figure). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.\nIf the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.\nVariances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.\n\n\nThe metric can compensate for linear (i.e. global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.\nIn complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e. local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.\n\n\n\nThe metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.3\nOne of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.\n\n\n\nStatistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.\n\n\n\nMCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.\nThis only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.\nPoor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.\n\n\n\n\nThe no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., Roberts, Gelman, and Gilks (1997)) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see Hoffman and Gelman (2014).\nNUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either\n\nthe NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or\nthe depth of the completed tree hits the maximum depth allowed.\n\nRather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory Betancourt (2016b).4\nConfiguring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.\nBoth the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as treedepth__ and n_leapfrog__, respectively. Because the final subtree may only be partially constructed, these two will always satisfy\n\\[\n2^{\\mathrm{treedepth} - 1} - 1\n\\ &lt; \\\nN_{\\mathrm{leapfrog}}\n\\ \\le \\\n2^{\\mathrm{treedepth} } - 1.\n\\]\nTree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html#sampling-without-parameters",
+    "href": "reference-manual/mcmc.html#sampling-without-parameters",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "In some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.\nFor example, to generate a sequence of \\(N\\) draws from a binomial with trials \\(K\\) and chance of success \\(\\theta\\), the following program suffices.\ndata {\n  real&lt;lower=0, upper=1&gt; theta;\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n}\ngenerated quantities {\n  array[N] int&lt;lower=0, upper=K&gt; y;\n  for (n in 1:N) {\n    y[n] = binomial_rng(K, theta);\n  }\n}\nFor this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.\nMost models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, theta in the example above could be declared as a parameter and initialized as a parameter.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html#general-config.section",
+    "href": "reference-manual/mcmc.html#general-config.section",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "Stan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the optimization algorithms chapter, and the diagnostics chapter.\n\n\nThe random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nStan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.\n\n\nTogether, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.\n\n\n\n\nThe initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.\n\n\nIf the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).\n\n\n\nIt is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.\n\n\n\nRandom initialization by default initializes the parameter values with values drawn at random from a \\(\\mathsf{Uniform}(-2, 2)\\) distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.\nBecause zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html#divergent-transitions",
+    "href": "reference-manual/mcmc.html#divergent-transitions",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "The Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.\nIn Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. Leimkuhler and Reich (2004) provide details and error analysis for the leapfrog algorithm.\nA divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,5 the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. Betancourt (2016a) provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.\nThe Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.\nIf the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.\nThe primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.\n\n\nIn some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.\nBefore reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/mcmc.html#footnotes",
+    "href": "reference-manual/mcmc.html#footnotes",
+    "title": "MCMC Sampling",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe typical set is a concept borrowed from information theory and refers to the neighborhood (or neighborhoods in multimodal models) of substantial posterior probability mass through which the Markov chain will travel in equilibrium.↩︎\nThis optimization of step size during adaptation of the sampler should not be confused with running Stan’s optimization method.↩︎\nIn Riemannian HMC the metric compensates for nonlinear correlations.↩︎\nStan previously used slice sampling along the trajectory, following the original NUTS paper of Hoffman and Gelman (2014).↩︎\nThe current default threshold is a factor of \\(10^3\\), whereas when the leapfrog integrator is working properly, the divergences will be around \\(10^{-7}\\) and do not compound due to the symplectic nature of the leapfrog integrator.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
+    ]
+  },
+  {
+    "objectID": "reference-manual/laplace_embedded.html",
+    "href": "reference-manual/laplace_embedded.html",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The embedded Laplace approximation replaces explicit sampling of potentially high-dimensional Gaussian latent variables with a local Gaussian approximation, marginalizing them out so that inference proceeds over the remaining hyperparameters alone. This approach is often referred to as the integrated Laplace approximation, although the exact details of the method can vary. The details of Stan’s implementation can be found in references (Margossian et al. 2020; Margossian 2023).\nA standard approach to fit a latent Gaussian model would be to perform inference jointly over the latent Gaussian variables and the hyperparameters. Instead, the embedded Laplace approximation can be used to do approximate marginalization of the latent Gaussian variables; we can then use any inference over the remaining hyperparameters. By marginalizing out the latent variables, the sampler explores a lower-dimensional, better-behaved marginal posterior. Individual iterations are more expensive (each requires an inner optimization), but the sampler typically needs far fewer iterations to achieve the same effective sample size. How this trade-off resolves depends on the specific problem at hand.\nFor complete function signatures and the built-in likelihood wrappers (Poisson, Negative Binomial, Bernoulli), see the Embedded Laplace functions reference. For worked examples with full data blocks, see the Gaussian Processes chapter.\n\n\nThe embedded Laplace approximation is used for latent Gaussian models. A latent Gaussian model is defined by three components:\n\n\\(\\phi\\): hyperparameters (e.g., GP kernel length-scale and magnitude, or variance components in a hierarchical model),\n\\(\\theta\\): latent Gaussian variables (the high-dimensional quantity to be marginalized out),\n\\(y\\): observed data.\n\nThese components are related through a hierarchical structure. The hyperparameters \\(\\phi\\) are given a prior \\(p(\\phi)\\). The latent variables \\(\\theta\\) have a multivariate normal prior centered at 0 with covariance matrix \\(K(\\phi)\\). An non-zero mean offset can be incorporated into the likelihood function. The observations \\(y\\) have a data model \\(p(y \\mid \\theta, \\phi)\\). The prior on \\(\\theta\\) is centered at zero; an offset can be incorporated into the data model if a non-zero mean is needed.\n\\[\\begin{eqnarray*}\n  \\phi & \\sim & p(\\phi) \\\\\n  \\theta & \\sim & \\text{Multi-Normal}(0, K(\\phi)) \\\\\n  y & \\sim & p(y \\mid \\theta, \\phi).\n\\end{eqnarray*}\\]\nThe generative model above defines a joint distribution over all three quantities, \\(p(\\phi, \\theta, y) = p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\). After observing data \\(y\\), Bayes’ theorem gives the joint posterior \\(p(\\phi, \\theta \\mid y) \\propto p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\), where \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) and \\(\\phi\\) is the joint likelihood function.\nSampling directly from the joint posterior \\(p(\\phi, \\theta \\mid y)\\) of this model is often difficult. Challenging geometries (e.g., funnels) frustrate inference algorithms, including Hamiltonian Monte Carlo and variational inference. However, the marginal posterior \\(p(\\phi \\mid y)\\) is often well-behaved and low-dimensional, making it much easier to sample. With an embedded Laplace approximation, we can obtain an approximation of the marginal posterior \\(p(\\phi \\mid y)\\). This is done via an intermediate approximation of the conditional posterior \\(p(\\theta \\mid \\phi, y)\\) by a normal distribution and this normal approximation is well-justified when the likelihood \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) is log concave (given that we already have a normal prior \\(p(\\theta \\mid \\phi)\\)). Once we obtain (approximate) samples \\(\\phi^{(i)} \\sim p(\\phi \\mid y)\\), we can in turn generate posterior draws for \\(\\theta\\) using the normal approximation to \\(p(\\theta \\mid y, \\phi)\\), evaluated at the posterior draws \\(\\phi^{(i)}\\).\n\n\n\nThe two-step inference strategy for using embedded Laplace in a latent Gaussian model requires approximations to both the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and the marginal likelihood \\(p(y \\mid \\phi)\\). The Laplace approximation is the normal distribution that matches the mode and curvature of the conditional posterior \\(p(\\theta \\mid y, \\phi)\\). The mode, defined as the value of \\(\\theta\\) that maximizes the conditional posterior, is estimated by a Newton solver, \\[\n  \\theta^* = \\underset{\\theta}{\\text{argmax}} \\ p(\\theta \\mid y, \\phi),\n\\]\nSince the approximation is normal, the curvature is matched by setting the covariance to the negative Hessian of the log conditional posterior, evaluated at the mode,\n\\[\n  \\Sigma^* = -  \\left . \\frac{\\partial^2}{\\partial \\theta^2}\n    \\log p (\\theta \\mid \\phi, y) \\right |_{\\theta =\\theta^*}.\n\\]\nThe resulting Laplace approximation is a multivariate normal centered at the mode with covariance given by the inverse curvature,\n\\[\n\\hat p_\\mathcal{L} (\\theta \\mid y, \\phi) = \\text{Multi-Normal}(\\theta^*, \\Sigma^*)\n\\approx p(\\theta \\mid y, \\phi).\n\\]\nThis approximation also yields an approximation to the marginal likelihood, obtained by evaluating the prior, likelihood, and approximate posterior at the mode \\(\\theta^*\\),\n\\[\n   \\hat p_\\mathcal{L}(y \\mid \\phi) := \\frac{p(\\theta^* \\mid \\phi) \\\n   p(y \\mid \\theta^*, \\phi) }{ \\hat p_\\mathcal{L} (\\theta^* \\mid \\phi, y) }\n   \\approx p(y \\mid \\phi).\n\\]\nHence, a strategy to approximate the posterior of the latent Gaussian model is to first estimate the marginal posterior \\(\\hat p_\\mathcal{L}(\\phi \\mid y) \\propto p(\\phi) p_\\mathcal{L} (y \\mid \\phi)\\) using any algorithm supported by Stan. Approximate posterior draws for the latent Gaussian variables are then obtained by first sampling \\(\\phi \\sim \\hat p_\\mathcal{L}(\\phi \\mid y)\\) and then \\(\\theta \\sim  \\hat p_\\mathcal{L}(\\theta \\mid \\phi, y)\\).\n\n\n\nThe embedded Laplace approximation presents several trade-offs with standard inference over the joint posterior \\(p(\\theta, \\phi \\mid y)\\). The main advantage of the embedded Laplace approximation is that it side-steps the intricate geometry of hierarchical models. The marginal posterior \\(p(\\phi \\mid y)\\) can then be handled by Hamiltonian Monte Carlo sampling without extensive tuning or reparameterization, and the mixing time is faster, meaning we can run shorter chains to achieve a desired precision. One additional benefit is that approximate methods, e.g. variational inference, which work poorly on the joint \\(p(\\theta, \\phi \\mid y)\\) can work well on the marginal posterior \\(p(\\phi \\mid y)\\).\nOn the other hand, the embedded Laplace approximation presents certain disadvantages. First, we need to perform a Laplace approximation each time the log marginal likelihood is evaluated, meaning each iteration can be expensive. Secondly, the approximation can introduce non-negligible error, especially with non-log-concave likelihoods (note the prior is always multivariate normal). How these trade-offs are resolved depends on the application; see Margossian et al. (2020) for some examples.\n\n\nThe quality of the Laplace approximation depends on how close the true conditional posterior \\(p(\\theta \\mid y, \\phi)\\) is to Gaussian.\nWorks well. Log-concave likelihoods, for example from a Poisson model with log link or negative binomial with log link. These produce unimodal conditional posteriors when combined with a Gaussian prior. The approximation error is typically negligible with these likelihoods, especially with moderate-to-large counts (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). If the likelihood is normal, there is no error in the approximation but in this case the marginalization can be worked analytically and the resulting implementation is much faster than using the embedded Laplace approximation.\nWorks adequately. Bernoulli model with logit link has technically log-concave likelihood, but the likelihood can be very skewed making the Gaussian approximation less accurate than for count data. The embedded Laplace is still useful when \\(\\theta\\) is high-dimensional and joint sampling is infeasible; see Vehtari et al. (2016) and Margossian et al. (2020) for discussion.\nNot appropriate. For likelihoods that are not log-concave in \\(\\theta\\), the conditional posterior may be multimodal and the Newton solver finds only a single mode or can fail completely. When \\(\\theta\\) is low-dimensional (a few dozen or fewer), the overhead of the inner optimization may not pay for itself and standard joint HMC sampling is often adequate.\n\n\n\n\nWhen the embedded Laplace approximation does not converge or produces unexpected results, the solver configuration may need adjustment. This section describes the internals of the Newton solver and the options available for tuning it.\n\n\nA critical component of the embedded Laplace approximation is the Newton solver used to estimate the mode \\(\\theta^*\\) of \\(p(\\theta \\mid \\phi, y)\\). The objective function being maximized is the log joint density of the prior and likelihood with respect to \\(\\theta\\).\n\\[\n\\Psi(\\theta) = \\log p(\\theta \\mid \\phi) + \\log p(y \\mid \\theta, \\phi),\n\\]\nConvergence is declared when the change in the objective between successive iterations falls below a tolerance \\(\\Delta\\).\n\\[\n| \\Psi (\\theta^{(i + 1)}) - \\Psi (\\theta^{(i)}) | \\le \\Delta.\n\\]\nThe solver also stops after reaching a pre-specified maximum number of steps. In that case, Stan throws a warning, but still returns the last iteration’s parameters. If you see this warning you should check the diagnostics to understand why the solver failed to converge.\nTo help with cases where the Newton step does not lead to a decrease in the objective function, the Newton iteration is augmented with a wolfe line-search to ensure that at each iteration the objective function \\(\\Psi\\) decreases. Specifically, suppose the objective increases after a Newton step, indicating the step overshot a region of improvement,\n\\[\n\\Psi (\\theta^{(i + 1)}) &lt; \\Psi (\\theta^{(i)}).\n\\]\nThis can indicate that the Newton step \\(\\alpha\\) at iteration \\(i\\) is too large and that we skipped a region where the objective function decreases. In that case, we can fallback to a Wolfe line search to find a step size which satisfies the Wolfe conditions. The wolfe line search attempts to find a search direction \\(p_i\\) and step size \\(\\alpha_k\\) such that an accepted step both increases our objective while ensuring the slope of the accepted step is flatter than our previous position. Together these help push the algorithm towards a minimum.\n\\[\nf(x_i + \\alpha_k p_i) \\le f(x_i) + c_1 \\alpha_k \\nabla f(x_i)^T p_i\n-p^T_i \\Delta f(x_i + \\alpha_k p_i) \\le -c_2 p^T_i \\Delta f(x_i)\n\\]\n\\[\n  \\theta^{(i + 1)} \\leftarrow \\frac{\\theta^{(i + 1)} + \\theta^{(i)}}{2}.\n\\]\nWe repeat this halving of steps until \\(\\Psi (\\theta^{(i + 1)}) \\ge \\Psi (\\theta^{(i)})\\), or until a maximum number of linesearch steps is reached. For certain problems, adding a linesearch can make the optimization more stable.\n\n\n\nThe embedded Laplace approximation uses a custom Newton solver, specialized to find the mode of \\(p(\\theta \\mid \\phi, y)\\). A key step for efficient optimization is to ensure all matrix inversions are numerically stable. This can be done using the Woodbury-Sherman-Morrison formula and requires one of three matrix decompositions:\n\nCholesky decomposition of the Hessian of the negative log likelihood \\(W = - \\partial^2_\\theta \\log p(y \\mid \\theta, \\phi)\\).\nCholesky decomposition of the prior covariance matrix \\(K(\\phi)\\).\nLU-decomposition of \\(I + KW\\), where \\(I\\) is the identity matrix.\n\nThe first solver (1) should be used if the negative log likelihood is positive-definite. Otherwise the user should rely on (2). In rarer cases where it is not numerically safe to invert the covariance matrix \\(K\\), users can use the third solver as a last-resort option.\n\n\n\nA key step to speed up computation is to take advantage of the sparsity of \\(H\\), the Hessian of the log likelihood with respect to the latent variables, \\[\n  H = \\frac{\\partial^2}{\\partial \\theta^2} \\log p(y \\mid \\theta, \\phi).\n\\] For example, if the observations \\((y_1, \\cdots, y_N)\\) are conditionally independent and each depends on only one component of \\(\\theta\\), the log likelihood decomposes into a sum of per-observation terms, \\[\n  \\log p(y \\mid \\theta, \\phi) = \\sum_{i = 1}^N \\log p(y_i \\mid \\theta_i, \\phi),\n\\] and the Hessian is diagonal. This leads to faster calculations of the Hessian and subsequently sparse matrix operations. This case is common in Gaussian process models, and certain hierarchical models.\nStan’s suite of functions for the embedded Laplace approximation exploits block-diagonal structure in the Hessian, where the user specifies the block size B. The user can specify the size \\(B\\) of these blocks. The user is responsible for working out what \\(B\\) is. If the Hessian is dense, then we simply set \\(B = N\\). The diagonal case above corresponds to B = 1. Arbitrary sparsity patterns beyond block-diagonal structure are not currently supported.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "reference-manual/laplace_embedded.html#latent-gaussian-models",
+    "href": "reference-manual/laplace_embedded.html#latent-gaussian-models",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The embedded Laplace approximation is used for latent Gaussian models. A latent Gaussian model is defined by three components:\n\n\\(\\phi\\): hyperparameters (e.g., GP kernel length-scale and magnitude, or variance components in a hierarchical model),\n\\(\\theta\\): latent Gaussian variables (the high-dimensional quantity to be marginalized out),\n\\(y\\): observed data.\n\nThese components are related through a hierarchical structure. The hyperparameters \\(\\phi\\) are given a prior \\(p(\\phi)\\). The latent variables \\(\\theta\\) have a multivariate normal prior centered at 0 with covariance matrix \\(K(\\phi)\\). An non-zero mean offset can be incorporated into the likelihood function. The observations \\(y\\) have a data model \\(p(y \\mid \\theta, \\phi)\\). The prior on \\(\\theta\\) is centered at zero; an offset can be incorporated into the data model if a non-zero mean is needed.\n\\[\\begin{eqnarray*}\n  \\phi & \\sim & p(\\phi) \\\\\n  \\theta & \\sim & \\text{Multi-Normal}(0, K(\\phi)) \\\\\n  y & \\sim & p(y \\mid \\theta, \\phi).\n\\end{eqnarray*}\\]\nThe generative model above defines a joint distribution over all three quantities, \\(p(\\phi, \\theta, y) = p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\). After observing data \\(y\\), Bayes’ theorem gives the joint posterior \\(p(\\phi, \\theta \\mid y) \\propto p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\), where \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) and \\(\\phi\\) is the joint likelihood function.\nSampling directly from the joint posterior \\(p(\\phi, \\theta \\mid y)\\) of this model is often difficult. Challenging geometries (e.g., funnels) frustrate inference algorithms, including Hamiltonian Monte Carlo and variational inference. However, the marginal posterior \\(p(\\phi \\mid y)\\) is often well-behaved and low-dimensional, making it much easier to sample. With an embedded Laplace approximation, we can obtain an approximation of the marginal posterior \\(p(\\phi \\mid y)\\). This is done via an intermediate approximation of the conditional posterior \\(p(\\theta \\mid \\phi, y)\\) by a normal distribution and this normal approximation is well-justified when the likelihood \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) is log concave (given that we already have a normal prior \\(p(\\theta \\mid \\phi)\\)). Once we obtain (approximate) samples \\(\\phi^{(i)} \\sim p(\\phi \\mid y)\\), we can in turn generate posterior draws for \\(\\theta\\) using the normal approximation to \\(p(\\theta \\mid y, \\phi)\\), evaluated at the posterior draws \\(\\phi^{(i)}\\).",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "reference-manual/laplace_embedded.html#approximation-of-the-conditional-posterior-and-marginal-likelihood",
+    "href": "reference-manual/laplace_embedded.html#approximation-of-the-conditional-posterior-and-marginal-likelihood",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The two-step inference strategy for using embedded Laplace in a latent Gaussian model requires approximations to both the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and the marginal likelihood \\(p(y \\mid \\phi)\\). The Laplace approximation is the normal distribution that matches the mode and curvature of the conditional posterior \\(p(\\theta \\mid y, \\phi)\\). The mode, defined as the value of \\(\\theta\\) that maximizes the conditional posterior, is estimated by a Newton solver, \\[\n  \\theta^* = \\underset{\\theta}{\\text{argmax}} \\ p(\\theta \\mid y, \\phi),\n\\]\nSince the approximation is normal, the curvature is matched by setting the covariance to the negative Hessian of the log conditional posterior, evaluated at the mode,\n\\[\n  \\Sigma^* = -  \\left . \\frac{\\partial^2}{\\partial \\theta^2}\n    \\log p (\\theta \\mid \\phi, y) \\right |_{\\theta =\\theta^*}.\n\\]\nThe resulting Laplace approximation is a multivariate normal centered at the mode with covariance given by the inverse curvature,\n\\[\n\\hat p_\\mathcal{L} (\\theta \\mid y, \\phi) = \\text{Multi-Normal}(\\theta^*, \\Sigma^*)\n\\approx p(\\theta \\mid y, \\phi).\n\\]\nThis approximation also yields an approximation to the marginal likelihood, obtained by evaluating the prior, likelihood, and approximate posterior at the mode \\(\\theta^*\\),\n\\[\n   \\hat p_\\mathcal{L}(y \\mid \\phi) := \\frac{p(\\theta^* \\mid \\phi) \\\n   p(y \\mid \\theta^*, \\phi) }{ \\hat p_\\mathcal{L} (\\theta^* \\mid \\phi, y) }\n   \\approx p(y \\mid \\phi).\n\\]\nHence, a strategy to approximate the posterior of the latent Gaussian model is to first estimate the marginal posterior \\(\\hat p_\\mathcal{L}(\\phi \\mid y) \\propto p(\\phi) p_\\mathcal{L} (y \\mid \\phi)\\) using any algorithm supported by Stan. Approximate posterior draws for the latent Gaussian variables are then obtained by first sampling \\(\\phi \\sim \\hat p_\\mathcal{L}(\\phi \\mid y)\\) and then \\(\\theta \\sim  \\hat p_\\mathcal{L}(\\theta \\mid \\phi, y)\\).",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "reference-manual/laplace_embedded.html#trade-offs-of-the-approximation",
+    "href": "reference-manual/laplace_embedded.html#trade-offs-of-the-approximation",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The embedded Laplace approximation presents several trade-offs with standard inference over the joint posterior \\(p(\\theta, \\phi \\mid y)\\). The main advantage of the embedded Laplace approximation is that it side-steps the intricate geometry of hierarchical models. The marginal posterior \\(p(\\phi \\mid y)\\) can then be handled by Hamiltonian Monte Carlo sampling without extensive tuning or reparameterization, and the mixing time is faster, meaning we can run shorter chains to achieve a desired precision. One additional benefit is that approximate methods, e.g. variational inference, which work poorly on the joint \\(p(\\theta, \\phi \\mid y)\\) can work well on the marginal posterior \\(p(\\phi \\mid y)\\).\nOn the other hand, the embedded Laplace approximation presents certain disadvantages. First, we need to perform a Laplace approximation each time the log marginal likelihood is evaluated, meaning each iteration can be expensive. Secondly, the approximation can introduce non-negligible error, especially with non-log-concave likelihoods (note the prior is always multivariate normal). How these trade-offs are resolved depends on the application; see Margossian et al. (2020) for some examples.\n\n\nThe quality of the Laplace approximation depends on how close the true conditional posterior \\(p(\\theta \\mid y, \\phi)\\) is to Gaussian.\nWorks well. Log-concave likelihoods, for example from a Poisson model with log link or negative binomial with log link. These produce unimodal conditional posteriors when combined with a Gaussian prior. The approximation error is typically negligible with these likelihoods, especially with moderate-to-large counts (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). If the likelihood is normal, there is no error in the approximation but in this case the marginalization can be worked analytically and the resulting implementation is much faster than using the embedded Laplace approximation.\nWorks adequately. Bernoulli model with logit link has technically log-concave likelihood, but the likelihood can be very skewed making the Gaussian approximation less accurate than for count data. The embedded Laplace is still useful when \\(\\theta\\) is high-dimensional and joint sampling is infeasible; see Vehtari et al. (2016) and Margossian et al. (2020) for discussion.\nNot appropriate. For likelihoods that are not log-concave in \\(\\theta\\), the conditional posterior may be multimodal and the Newton solver finds only a single mode or can fail completely. When \\(\\theta\\) is low-dimensional (a few dozen or fewer), the overhead of the inner optimization may not pay for itself and standard joint HMC sampling is often adequate.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "reference-manual/laplace_embedded.html#details-of-the-approximation",
+    "href": "reference-manual/laplace_embedded.html#details-of-the-approximation",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "When the embedded Laplace approximation does not converge or produces unexpected results, the solver configuration may need adjustment. This section describes the internals of the Newton solver and the options available for tuning it.\n\n\nA critical component of the embedded Laplace approximation is the Newton solver used to estimate the mode \\(\\theta^*\\) of \\(p(\\theta \\mid \\phi, y)\\). The objective function being maximized is the log joint density of the prior and likelihood with respect to \\(\\theta\\).\n\\[\n\\Psi(\\theta) = \\log p(\\theta \\mid \\phi) + \\log p(y \\mid \\theta, \\phi),\n\\]\nConvergence is declared when the change in the objective between successive iterations falls below a tolerance \\(\\Delta\\).\n\\[\n| \\Psi (\\theta^{(i + 1)}) - \\Psi (\\theta^{(i)}) | \\le \\Delta.\n\\]\nThe solver also stops after reaching a pre-specified maximum number of steps. In that case, Stan throws a warning, but still returns the last iteration’s parameters. If you see this warning you should check the diagnostics to understand why the solver failed to converge.\nTo help with cases where the Newton step does not lead to a decrease in the objective function, the Newton iteration is augmented with a wolfe line-search to ensure that at each iteration the objective function \\(\\Psi\\) decreases. Specifically, suppose the objective increases after a Newton step, indicating the step overshot a region of improvement,\n\\[\n\\Psi (\\theta^{(i + 1)}) &lt; \\Psi (\\theta^{(i)}).\n\\]\nThis can indicate that the Newton step \\(\\alpha\\) at iteration \\(i\\) is too large and that we skipped a region where the objective function decreases. In that case, we can fallback to a Wolfe line search to find a step size which satisfies the Wolfe conditions. The wolfe line search attempts to find a search direction \\(p_i\\) and step size \\(\\alpha_k\\) such that an accepted step both increases our objective while ensuring the slope of the accepted step is flatter than our previous position. Together these help push the algorithm towards a minimum.\n\\[\nf(x_i + \\alpha_k p_i) \\le f(x_i) + c_1 \\alpha_k \\nabla f(x_i)^T p_i\n-p^T_i \\Delta f(x_i + \\alpha_k p_i) \\le -c_2 p^T_i \\Delta f(x_i)\n\\]\n\\[\n  \\theta^{(i + 1)} \\leftarrow \\frac{\\theta^{(i + 1)} + \\theta^{(i)}}{2}.\n\\]\nWe repeat this halving of steps until \\(\\Psi (\\theta^{(i + 1)}) \\ge \\Psi (\\theta^{(i)})\\), or until a maximum number of linesearch steps is reached. For certain problems, adding a linesearch can make the optimization more stable.\n\n\n\nThe embedded Laplace approximation uses a custom Newton solver, specialized to find the mode of \\(p(\\theta \\mid \\phi, y)\\). A key step for efficient optimization is to ensure all matrix inversions are numerically stable. This can be done using the Woodbury-Sherman-Morrison formula and requires one of three matrix decompositions:\n\nCholesky decomposition of the Hessian of the negative log likelihood \\(W = - \\partial^2_\\theta \\log p(y \\mid \\theta, \\phi)\\).\nCholesky decomposition of the prior covariance matrix \\(K(\\phi)\\).\nLU-decomposition of \\(I + KW\\), where \\(I\\) is the identity matrix.\n\nThe first solver (1) should be used if the negative log likelihood is positive-definite. Otherwise the user should rely on (2). In rarer cases where it is not numerically safe to invert the covariance matrix \\(K\\), users can use the third solver as a last-resort option.\n\n\n\nA key step to speed up computation is to take advantage of the sparsity of \\(H\\), the Hessian of the log likelihood with respect to the latent variables, \\[\n  H = \\frac{\\partial^2}{\\partial \\theta^2} \\log p(y \\mid \\theta, \\phi).\n\\] For example, if the observations \\((y_1, \\cdots, y_N)\\) are conditionally independent and each depends on only one component of \\(\\theta\\), the log likelihood decomposes into a sum of per-observation terms, \\[\n  \\log p(y \\mid \\theta, \\phi) = \\sum_{i = 1}^N \\log p(y_i \\mid \\theta_i, \\phi),\n\\] and the Hessian is diagonal. This leads to faster calculations of the Hessian and subsequently sparse matrix operations. This case is common in Gaussian process models, and certain hierarchical models.\nStan’s suite of functions for the embedded Laplace approximation exploits block-diagonal structure in the Hessian, where the user specifies the block size B. The user can specify the size \\(B\\) of these blocks. The user is responsible for working out what \\(B\\) is. If the Hessian is dense, then we simply set \\(B = N\\). The diagonal case above corresponds to B = 1. Arbitrary sparsity patterns beyond block-diagonal structure are not currently supported.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "reference-manual/index.html",
+    "href": "reference-manual/index.html",
+    "title": "Stan Reference Manual",
+    "section": "",
+    "text": "This is the official reference manual for Stan’s programming language for coding probability models, inference algorithms for fitting models and making predictions, and posterior analysis tools for evaluating the results. This manual applies to all Stan interfaces.\nThe first part of the reference manual provides a full specification of the Stan programming language. The language is responsible for defining a log density function conditioned on data. Typically, this is a Bayesian posterior, but it may also be a penalized likelihood function. The second part of the manual specifies the inference algorithms and posterior inference tools. The third part provides auxiliary information about the use of Stan.\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Reference Manual",
+      "Stan Reference Manual"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html",
+    "href": "reference-manual/expressions.html",
+    "title": "Expressions",
+    "section": "",
+    "text": "An expression is the syntactic unit in a Stan program that denotes a value. Every expression in a well-formed Stan program has a type that is determined statically (at compile time), based only on the type of its variables and the types of the functions used in it. If an expressions type cannot be determined statically, the Stan compiler will report the location of the problem.\nThis chapter covers the syntax, typing, and usage of the various forms of expressions in Stan.\n\n\nThe simplest form of expression is a literal that denotes a primitive numerical value.\n\n\nInteger literals represent integers of type int. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression --1 is interpreted as the negation of the literal -1.)\nThe following list contains well-formed integer literals.\n0, 1, -1, 256, -127098, 24567898765\nInteger literals must have values that fall within the bounds for integer values (see the section on numerical data types).\nInteger literals may not contain decimal points (.). Thus the expressions 1. and 1.0 are of type real and may not be used where a value of type int is required.\n\n\n\nA number written with a period or with scientific notation is assigned to a the continuous numeric type real. Real literals are written in base 10 with a period (.) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.\n0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.\nThe notation e or E followed by a positive or negative integer denotes a power of 10 to multiply. For instance, 2.7e3 and 2.7e+3 denote \\(2.7 \\times 10^3\\), whereas -2E-5 denotes \\(-2 \\times\n10^{-5}\\).\n\n\n\nA number followed by the character i denotes an imaginary number and is assigned to the numeric type complex. The number preceding i may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.\n1i, 2i, -325.786i, 1e10i, 2.87e-10i.\nNote that the character i by itself is not a well-formed imaginary literal. The unit imaginary number must be written as 1i.\n\n\n\nStan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.\n1 + 2i, -3.2e9 + 1e10i\nThese will be assigned the type complex, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.\n\n\n\n\nA variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (_) character. Variables must start with a letter (a--z and A--Z) and may not end with two underscores (__).\nExamples of legal variable identifiers are as follows.\na, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable\nUnlike in R and BUGS, variable identifiers in Stan may not contain a period character.\n\n\nStan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the stanc translator halting with an error message indicating which reserved name was used and its location in the model code.\n\n\nThe name of the model cannot be used as a variable within the model. This is usually not a problem because the default in bin/stanc is to append _model to the name of the file containing the model specification. For example, if the model is in file foo.stan, it would not be legal to have a variable named foo_model when using the default model name through bin/stanc. With user-specified model names, variables cannot match the model.\n\n\n\nThe following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.\nfor, in, while, repeat, until, if, then, else,\ntrue, false, target, struct, typedef, export,\nauto, extern, var, static, lower, upper, offset,\nmultiplier\nVariables should not be named after types, either, and thus may not be any of the following.\nint, real, complex, vector, simplex, unit_vector,\nsum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,\nrow_vector, matrix, cholesky_factor_corr,\ncolumn_stochastic_matrix,row_stochastic_matrix,\ncholesky_factor_cov, corr_matrix, cov_matrix, array\nThe following keywords or built-in functions are also reserved and cannot be used as variable names:\nprint, reject, profile, fatal_error, target, jacobian\nThe following block identifiers are reserved and cannot be used as variable names:\nfunctions, model, data, parameters, quantities,\ntransformed, generated\n\n\n\nVariable names will also conflict with the names of distributions suffixed with _lpdf, _lpmf, _lcdf, and _lccdf, _cdf, and _ccdf, such as normal_lcdf_log. No user-defined variable can take a name ending in _lupdf or _lupmf even if a corresponding _lpdf or _lpmf is not defined.\nUsing any of these variable names causes the stanc translator to halt and report the name and location of the variable causing the conflict.\n\n\n\nStan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed _stan_NAME (e.g. _stan_public) behind the scenes (in the generated C++ code).\n\nalignas, alignof, and, and_eq, asm, bitand, bitor, bool,\ncase, catch, char, char16_t, char32_t, class, compl, const,\nconstexpr, const_cast, decltype, default, delete, do,\ndouble, dynamic_cast, enum, explicit, float, friend, goto,\ninline, long, mutable, namespace, new, noexcept, not, not_eq,\nnullptr, operator, or, or_eq, private, protected, public,\nregister, reinterpret_cast, short, signed, sizeof,\nstatic_assert, static_cast, switch, template, this, thread_local,\nthrow, try, typeid, typename, union, unsigned, using, virtual,\nvolatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,\nSTAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH\n\n\n\n\nThe legal characters for variable identifiers are given in the following table.\n\n\nThe alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.\n\n\n\ncharacters\nASCII code points\n\n\n\n\na -- z\n97 – 122\n\n\nA -- Z\n65 – 90\n\n\n0 -- 9\n48 – 57\n\n\n_\n95\n\n\n\nAlthough not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.\n\n\n\nWithin comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.\n\n\n\n\n\nExpressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.\n\n\nSquare brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression [ 1, 10, 100 ] denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.\nrow_vector[2] rv2 =  [ 1, 2 ];\nvector[3] v3 = [ 3, 4, 5 ]';\nThe vector expression values may be compound expressions or variable names, so it is legal to write [ 2 * 3, 1 + 4] or [ x, y ], providing that x and y are primitive variables.\n\n\n\nA matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.\nmatrix[3, 2] m1 = [ [ 1, 2 ], [ 3, 4 ], [5, 6 ] ];\nAny expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:\nvector[2] vX = [ 1, 10 ]';\nrow_vector[2] vY = [ 100, 1000 ];\nmatrix[3, 2] m2 = [ vX', vY, [ 1, 2 ]  ];\n\n\n\nComplex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.\ncomplex_vector[3] = [1 + 2i, 3 - 1.7i, 0]';\ncomplex_row_vector[2] = [12, -2i];\ncomplex_matrix[2, 3] = [[1 + 2i, 3 - 1.7i, 0],\n                        [3.9 - 1.234i, 176i, 1 + 1i]];\n\n\nThe empty expression [ ] is ambiguous and therefore is not allowed and similarly expressions such as [ [ ] ] or [ [ ], [ ] ] are not allowed.\n\n\n\n\nIf needed, it is possible to create an empty vector with\nrep_vector(e, 0)\nwhere the first expression e needs to scalar of type real.\nIf needed, it is possible to create an empty matrix with\nrep_matrix(e, 0, 0)\nwhere the first expression e needs to scalar of type real.\n\n\n\nCurly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression { 1, 10, 100 } denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.\narray[3] int a = { 1, 10, 100 };\nThe values may be compound expressions, so it is legal to write { 2 * 3, 1 + 4 }. It is also possible to write two dimensional arrays directly, as in the following example.\narray[2, 3] int b = { { 1, 2, 3 }, { 4, 5, 6 } };\nThis way, b[1] is { 1, 2, 3 } and b[2] is { 4, 5, 6 }.\nWhitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.\narray[2, 3] int b = { { 1, 2, 3 },\n                { 4, 5, 6 } };\n\n\n\nThe empty array expression ({ }) is not allowed. See more about restrictions on array expressions in subsection Restrictions on values.\nIf needed, it is possible to create an empty array with\nrep_array(e, 0)\nwhere the first expression e determines the type of the array. For example, rep_array(0.0, 0) returns an empty real array of type real[], whereas rep_array({123}, 0) returns an empty two dimensional integer array of type int[ , ]. Only the type of the first argument is used, so the integer arrays {123} and {0} produce equivalent values.\n\n\n\nAny type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.\nvector[3] b;\nvector[3] c;\n// ...\narray[2] vector[3] d = { b, c };\nThe elements may also be a mixture of int and real typed expressions, in which case the result is an array of real values.\narray[2] real b = { 1, 1.9 };\n\n\n\nStan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.\ntuple(int, vector[3]) xy = (42, [1, 2.9, -1.3]');\nThe expression 42 is of type int and the expression [1, 2.9, -1.3] is of type row_vector so that [1, 2.9, -1.3]' is of type vector and of size 3. The whole tuple expression (42, [1, 2.9, -1.3]') thus has a sized type of tuple(int, vector[3]) and an unsized type (e.g., for a function argument) of tuple(int, vector).\nA tuple of one element can be created using the same style as languages like Python, with a trailing comma, e.g., (3.14,). For longer tuples, Stan does not support trailing commas.\n\n\n\nThere are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.\n\n\nAlthough it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.\n{ { 1, 2, 3 }, { 4, 5 } }  // compile time error: size mismatch\nThis may appear to be OK, because it is creating a two-dimensional integer array (array[,] int) out of two one-dimensional array integer arrays (array[] int). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.\n{ { 1, 2, 3 }, m }  // runtime error if m not size 3\n\n\n\nBecause there is no way to infer the type of the result, the empty array expression ({ }) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.\narray[0] int a;   // a is fully defined as zero element array\n\n\n\nThere is no way to declare or construct a zero-tuple in Stan. Tuples must be at least one element long. The expression () does not pick out a zero-tuple—it is ill formed.\nOne-tuples need a trailing comma, like (1,). The expression (1), without the comma, is of type int rather than a tuple.\n\n\n\n\n\nAny expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, ( and ), are allowed. The square brackets [ and ] are reserved for array indexing and the curly braces { and } for grouping statements.\nWith parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression 1 + 2 * 3 has a subexpression 2 * 3 and evaluates to 7. With parentheses, this grouping may be made explicit with the expression 1 + (2 * 3). More importantly, the expression (1 + 2) *   3 has 1 + 2 as a subexpression and evaluates to 9.\n\n\n\nFor integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (+), subtraction (-), multiplication (*) and division (/) in the usual ways.\nFor integer expressions, Stan supports the modulus (%) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3.0 + 0.14\n-15\n2 * 3 + 1\n(x - y) / 2.0\n(n * (n + 1)) / 2\nx / n\nm % n\nThe negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (') is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.\nFor example, if y and mu are variables of type vector and Sigma is a variable of type matrix, then (y - mu)' * Sigma * (y - mu) is a well-formed expression of type real. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) y - mu are of type vector because the variables y and mu are of type vector. The transpose of this expression, the subexpression (y - mu)' is of type row_vector. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form (((y - mu)') * Sigma) * (y - mu).\nThe type of subexpression (y - mu)' * Sigma is inferred to be row_vector, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a real value.\nStan provides elementwise matrix multiplication (e.g., a .* b) and division (e.g., a ./ b) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,\nvector[N] a;\nvector[N] b;\nvector[N] c;\nthe assignment,\nc = a .* b;\nproduces the same result with roughly the same efficiency as the loop\nfor (n in 1:N) {\n  c[n] = a[n] * b[n];\n}\nStan supports exponentiation (^) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3 ^ 2\n3.0 ^ -2\n3.0 ^ 0.14\nx ^ n\nn ^ x\nn ^ m\nx ^ y\nExponentiation is right associative, so the expression 2 ^ 3 ^ 4 is equivalent to the fully specified form 2 ^ (3 ^ 4).\n\n\nThe precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.\n\n\nStan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e., operator*(int, int):int indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.\n\n\n\n\n\n\n\n\n\n\nOp.\nPrec.\nAssoc.\nPlacement\nDescription\n\n\n\n\n? ~ :\n10\nright\nternary infix\nconditional\n\n\n||\n9\nleft\nbinary infix\nlogical or\n\n\n&&\n8\nleft\nbinary infix\nlogical and\n\n\n==\n7\nleft\nbinary infix\nequality\n\n\n!=\n7\nleft\nbinary infix\ninequality\n\n\n&lt;\n6\nleft\nbinary infix\nless than\n\n\n&lt;=\n6\nleft\nbinary infix\nless than or equal\n\n\n&gt;\n6\nleft\nbinary infix\ngreater than\n\n\n&gt;=\n6\nleft\nbinary infix\ngreater than or equal\n\n\n+\n5\nleft\nbinary infix\naddition\n\n\n-\n5\nleft\nbinary infix\nsubtraction\n\n\n*\n4\nleft\nbinary infix\nmultiplication\n\n\n.*\n4\nleft\nbinary infix\nelementwise multiplication\n\n\n/\n4\nleft\nbinary infix\n(right) division\n\n\n./\n4\nleft\nbinary infix\nelementwise division\n\n\n%\n4\nleft\nbinary infix\nmodulus\n\n\n\\\n3\nleft\nbinary infix\nleft division\n\n\n%/%\n3\nleft\nbinary infix\ninteger division\n\n\n!\n2\nn/a\nunary prefix\nlogical negation\n\n\n-\n2\nn/a\nunary prefix\nnegation\n\n\n+\n2\nn/a\nunary prefix\npromotion (no-op in Stan)\n\n\n^\n1\nright\nbinary infix\nexponentiation\n\n\n.^\n1\nright\nbinary infix\nelementwise exponentiation\n\n\n'\n0\nn/a\nunary postfix\ntransposition\n\n\n()\n0\nn/a\nprefix, wrap\nfunction application\n\n\n[]\n0\nleft\nprefix, wrap\narray, matrix indexing\n\n\n\nOther expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.\nThe precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression a + b + c is interpreted as (a + b) + c. Similarly, a / b * c is interpreted as (a / b) * c.\nBecause multiplication has higher precedence than addition, the expression a * b + c is interpreted as (a * b) + c and the expression a + b * c is interpreted as a + (b * c). Similarly, 2 * x + 3 * - y is interpreted as (2 * x) + (3 * (-y)).\nTransposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, -u' is interpreted as -(u'), u * v' as u* (v'), and u' * v as (u') * v. For integer and reals, -n ^ 3 is interpreted as -(n ^ 3).\n\n\n\n\n\n\n\nThe ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If a is an expression of type int and b and c are expressions that can be converted to one another (e.g., compared with ==), then\na ? b : c\nis an expression of the promoted type of b and c. The result of this expression is b if condition a is true, c otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g. if one argument is of type int and the other of type real, the conditional expression as a whole is of type real. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.\n\n\nThe conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,\na &gt; 0 || b &lt; 0 ? c + d : e - f\nis equivalent to the explicitly grouped version\n(a &gt; 0 || b &lt; 0) ? (c + d) : (e - f)\nThe latter is easier to read even if the parentheses are not strictly necessary.\n\n\n\nThe conditional operator is right associative, so that\na ? b : c ? d : e\nparses as if explicitly grouped as\na ? b : (c ? d : e)\nAgain, the explicitly grouped version is easier to read.\n\n\n\n\nStan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable a in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value b is returned if the condition evaluates to a non-zero value and c is returned if the condition evaluates to zero.\n\n\nThe key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.\n\n\n\nIf one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full if-then conditional statement. For example,\ndata {\n  array[10] real x;\n  // ...\n}\nparameters {\n  array[10] real z;\n  // ...\n}\nmodel {\n  y ~ normal(cond ? x : z, sigma);\n  // ...\n}\nwould be more efficiently (if not more transparently) coded as\nif (cond) {\n  y ~ normal(x, sigma);\n} else {\n  y ~ normal(z, sigma);\n}\nThe conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable x will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.\n\n\n\n\n\nStan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if x is a variable of type array [] real (a one-dimensional array of reals) then x[1] is the value of the first element of the array.\nSubscripting has higher precedence than any of the arithmetic operations. For example, alpha * x[1] is equivalent to alpha * (x[1]).\nMultiple subscripts may be provided within a single pair of square brackets. If x is of type array[,] real, a two-dimensional array, then x[2, 501] is of type real.\n\n\nThe subscripting operator also returns subarrays of arrays. For example, if x is of type array[,,] real, then x[2] is of type array[,] real, and x[2, 3] is of type array[] real. As a result, the expressions x[2, 3] and x[2][3] have the same meaning.\n\n\n\nIf Sigma is a variable of type matrix, then Sigma[1] denotes the first row of Sigma and has the type row_vector.\n\n\n\nStan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if m is of type matrix[ , ], a two-dimensional array of matrices, then m[1] refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that m[1, 2] is of type matrix and denotes the matrix in the first row and second column of the array. Continuing to add indices, m[1, 2, 3] is of type row_vector and denotes the third row of the matrix denoted by m[1, 2]. Finally, m[1, 2, 3, 4] is of type real and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array m.\n\n\n\n\nIn addition to single integer indexes, as described in the language indexing section, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.\n\n\nTypes of indexes and examples with one-dimensional containers of size N and an integer array ii of type array [] real size K.\n\n\n\nindex type\nexample\nvalue\n\n\n\n\ninteger\na[11]\nvalue of a at index 11\n\n\ninteger array\na[ii]\na[ii[1]], …, a[ii[K]]\n\n\nlower bound\na[3:]\na[3], …, a[N]\n\n\nupper bound\na[:5]\na[1], …, a[5]\n\n\nrange\na[2:7]\na[2], …, a[7]\n\n\nrange\na[7:2]\n[]\n\n\nrange\na[5-3:5+2]\na[2], …, a[7]\n\n\nall\na[:]\na[1], …, a[N]\n\n\nall\na[]\na[1], …, a[N]\n\n\n\nThe range indexing with : allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:\n  array[6] int ii = reverse(linspaced_int_array(6, 2, 7));\nThen a[ii] evaluates to a[7], …, a[2].\n\n\n\nThe fundamental semantic rule for dealing with multiple indexes is the following. If idxs is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.\na[idxs, ...][i, ...] = a[idxs[i], ...][...]\nOn the other hand, if idx is a single index, it reduces the dimensionality of the output, so that\na[idx, ...] = a[idx][...]\nThe only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.\n\n\nSpecial rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix a, with integer single indexes i and j and integer array multiple indexes is and js. The same typing rules apply for all multiple indexes.\n\n\n\nexample\nrow index\ncolumn index\nresult type\n\n\n\n\na[i]\nsingle\nn/a\nrow vector\n\n\na[is]\nmultiple\nn/a\nmatrix\n\n\na[i, j]\nsingle\nsingle\nreal\n\n\na[i, js]\nsingle\nmultiple\nrow vector\n\n\na[is, j]\nmultiple\nsingle\nvector\n\n\na[is, js]\nmultiple\nmultiple\nmatrix\n\n\n\nEvaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.\nm[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]\nm[idxs, idx][j] = m[idxs[j], idx]\nm[idx, idxs][j] = m[idx, idxs[j]]\nEvaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.\n\n\n\n\n\nStan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.\nExpressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, log(2.0) is the expression of type real denoting the result of applying the natural logarithm to the value of the real literal 2.0.\nSyntactically, function application has higher precedence than any of the other operators, so that y + log(x) is interpreted as y + (log(x)).\n\n\nEach function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as\nreal log(real);\nand the signature for the lmultiply function is\nreal lmultiply(real, real);\nA function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.\nreal mean(array [] real);\nreal mean(vector);\nThe first applies to a one-dimensional array of real values and the second to a vector.\nThe identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.\n\n\n\nConstants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants \\(\\pi\\) and \\(e\\) are represented as nullary functions named pi() and e(). See the Stan Functions Reference built-in constants section for a list of built-in constants.\n\n\n\nBecause of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.\nFor example, consider a situation in which the following two function signatures have been registered for foo.\nreal foo(real, real);\nint foo(int, int);\nThe use of foo in the expression foo(1.0, 1.0) resolves to foo(real, real), and thus the expression foo(1.0, 1.0) itself is assigned a type of real.\nBecause integers may be promoted to real values, the expression foo(1, 1) could potentially match either foo(real, real) or foo(int, int). The former requires two type promotions and the latter requires none, so foo(1, 1) is resolved to function foo(int, int) and is thus assigned the type int.\nThe expression foo(1, 1.0) has argument types (int, real) and thus does not explicitly match either function signature. By promoting the integer expression 1 to type real, it is able to match foo(real, real), and hence the type of the function expression foo(1, 1.0) is real.\nIn some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named bar with the following signatures.\nreal bar(real, int);\nreal bar(int, real);\nWith these signatures, the expression bar(1.0, 1) and bar(1, 1.0) resolve to the first and second of the above functions, respectively. The expression bar(1.0, 1.0) is illegal because real values may not be demoted to integers. The expression bar(1, 1) is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name bar is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with bar(1, 1) given the two declarations above, the Stan compiler will flag the expression as illegal.\n\n\n\nFor most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix _rng. For example, a univariate normal random number can be generated by normal_rng(0, 1); only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.\n\n\nThe use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in _rng.\nThis allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.\n\n\n\nPosterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of (Gelman et al. 2013) for more information on posterior predictive checks.\n\n\n\n\n\nStan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.\n\n\nThe primitive implementation types for Stan are\nint, real, complex, vector, row_vector,  matrix, complex_vector,\ncomplex_row_vector, complex_matrix\nEvery basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.\n\n\nThe table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.\n\n\n\ntype\nprimitive type\n\n\n\n\nint\nint\n\n\nreal\nreal\n\n\nvector\nvector\n\n\nsimplex\nvector\n\n\nunit_vector\nvector\n\n\nsum_to_zero_vector\nvector\n\n\nordered\nvector\n\n\npositive_ordered\nvector\n\n\nrow_vector\nrow_vector\n\n\nmatrix\nmatrix\n\n\ncov_matrix\nmatrix\n\n\ncorr_matrix\nmatrix\n\n\ncholesky_factor_cov\nmatrix\n\n\ncholesky_factor_corr\nmatrix\n\n\ncolumn_stochastic_matrix\nmatrix\n\n\nrow_stochastic_matrix\nmatrix\n\n\nsum_to_zero_matrix\nmatrix\n\n\ncomplex_vector\ncomplex_vector\n\n\ncomplex_row_vector\ncomplex_row_vector\n\n\ncomplex_matrix\ncomplex_matrix\n\n\n\nA full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, array [] real has an array dimensionality of 1, int an array dimensionality of 0, and array [,,] int an array dimensionality of 3. The implementation type matrix[ , , ] has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.\nRecall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.\narray[I, J, K] matrix[M, N] a;\nThe matrix a is indexed as a[i, j, k, m, n] with the array indices first, followed by the matrix indices, with a[i, j, k] being a matrix and a[i, j, k, m] being a row vector.\n\n\n\n\nStan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.\n\n\n\nThere are two basic promotion rules,\n\nint types may be promoted to real, and\nreal types may be promoted to complex.\n\nPlus, promotion is transitive, so that\n\nif type U can be promoted to type V and type V can be promoted to type T, then U can be promoted to T.\n\nThe first rule means that expressions of type int may be used anywhere an expression of type real is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.\nThe remaining rules have to do with covariant typing rules, which say that a container of type U may be promoted to a container of the same shape of type T if U can be promoted to T. For vector and matrix types, this induces three rules,\n\nvector may be promoted to complex_vector,\nrow_vector may be promoted to complex_row_vector\nmatrix may be promoted to complex_matrix.\n\nFor array types, there’s a single rule\n\narray[...] U may be promoted to array[...] T if U can be promoted to T.\n\nFor example, this means array[,] int may be used where array [,] real or array [,] complex is required; as another example, array[] real may be used anywhere array[] complex is required.\nTuples have the natural extension of the above rules, applied to all sub-types at once\n\nA tuple(U1, ..., UN) may be promoted to a tuple(T1, ..., TN) if every Un can be promoted to Tn for n in 1:N\n\n\n\nAn integer literal expression such as 42 is of type int. Real literals such as 42.0 are of type real. Imaginary literals such as -17i are of type complex. the expression 7 - 2i acts like a complex literal, but technically it combines a real literal 7 and an imaginary literal 2i through subtraction.\n\n\n\nThe type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is int.\nThere is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.1\n\n\n\nIf x is an expression of total dimensionality greater than or equal to \\(N\\), then the type of expression e[i1, i2, ..., iN] is the same as that of e[i1][i2]...[iN], so it suffices to define the type of a singly-indexed function. Suppose e is an expression and i is an expression of primitive type int. Then\n\nif e is an expression of type array[i1, i2, ..., iN] T and k, i1, …, iN are expressions of type int, then e[k] is an expression of type array[i2, ..., iN] T,\nif e is an expression of type array[i] T with i and k expressions of type int, then e[k] is of type T,\nif e has implementation type vector or row_vector, dimensionality 0, then e[i] has implementation type real,\nif e has implementation type matrix, then e[i] has type row_vector,\nif e has implementation type complex_vector or complex_row_vector and i is an expression of type int, then e[i] is an expression of type complex, and\nif e has implementation type complex_matrix, and i is an expression of type int, then e[i] is an expression of type complex_row_vector.\n\n\n\n\nIf f is the name of a function and e1,...,eN are expressions for \\(N \\geq 0\\), then f(e1,...,eN) is an expression whose type is determined by the return type in the function signature for f given e1 through eN. Recall that a function signature is a declaration of the argument types and the result type.\nIn looking up functions, binary operators like real * real are defined as operator*(real, real) in the documentation and index.\nIn matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type int may be promoted to type real or complex if necessary (see the subsection on type promotion in the function application section, a real argument will be promoted to complex if necessary, a vector will be promoted to complex_vector if necessary, and so on.\nIn general, matrix operations return the lowest inferable type. For example, row_vector * vector returns a value of type real, which is declared in the function documentation and index as real operator*(row_vector, vector).\n\n\n\n\n\nThere are several expression constructions in Stan that act as higher-order functions.2\nThe higher-order functions and the signature of their argument functions are listed in the following pair of tables.\n\n\nHigher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.\n\n\n\n\n\n\n\n\n\nfunction\nparameter or data args\ndata args\nreturn type\n\n\n\n\nalgebra_solver\nvector, vector\narray [] real, array [] real\nvector\n\n\nalgebra_solver_newton\nvector, vector\narray [] real, array [] real\nvector\n\n\nintegrate_1d,\nreal, real, array [] real\narray [] real, array [] real\nreal\n\n\nintegrate_ode_X,\nreal, array [] real, array [] real\narray [] real, array [] real\narray [] real\n\n\nmap_rect\nvector, vector\narray [] real, array [] real\nvector\n\n\n\nFor example, the integrate_ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  array [] real foo(real t,\n                    array [] real y,\n                    array [] real theta,\n                    array [] real x_r,\n                    array [] real x_i) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\narray[2] real y0;\nreal t0;\narray[T] real ts;\narray[1] real theta;\narray[0] real x_r;\narray[0] int x_i;\n// ...\narray[T, 2] real y_hat = integrate_ode_rk45(foo, y0, t0,\n                                              ts, theta, x_r, x_i);\nThe function argument is foo, the name of the user-defined function; as shown in the higher-order functions table, integrate_ode_rk45 takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.\n\n\n\nVariadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.\n\n\n\nfunction\nrestricted args\nreturn type\n\n\n\n\nsolve_X\nvector\nvector\n\n\node_X,\nvector, real, array [] real\nvector[]\n\n\nreduce_sum\narray[] T, T1, T2\nreal\n\n\n\nT, T1, and T2 can be any Stan type.\nFor example, the ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  vector foo(real t, vector y, real theta, vector beta,\n            array [] real x_i, int index) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\nvector[2] y0;\nreal t0;\narray[T] real ts;\nreal theta;\nvector[7] beta;\narray[10] int x_i;\nint index;\n// ...\nvector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,\n                              beta, x_i, index);\nThe function argument is foo, the name of the user-defined function. As shown in the variadic higher-order functions table, ode_rk45 takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of foo and returns an array of vectors.\n\n\n\nThe function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.\n\n\n\nSome of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.\nFor user-defined functions the qualifier data may be prepended to the type to restrict the argument to data-only variables.\n\n\n\n\nDerivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.\n\n\nUnlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(sqrt(x - x), 1);\n}\nAlgebraically, the distribution statement in the model could be reduced to\n  x ~ normal(0, 1);\nand it would seem the model should produce unit normal draws for x. But rather than canceling, the expression sqrt(x -   x) causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,\n\\[\n\\begin{array}{rcl}\n\\frac{d}{dx} \\sqrt{x - x}\n& = &\n\\frac{1}{2 \\sqrt{x - x}} \\times \\frac{d}{dx} (x - x)\n\\\\[4pt]\n& = &\n\\frac{1}{0} \\times (1 - 1)\n\\\\[4pt]\n& = &\n\\infty \\times 0\n\\\\[4pt]\n& = & \\mathrm{NaN}.\n\\end{array}\n\\]\nRather than the \\(x - x\\) canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.\nThe only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like sqrt(x - x) for which the chain rule produces not-a-number values.\n\n\n\nThe best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).\nFor example, compiling the above model to an executable sqrt-x-minus-x in CmdStan, the test can be run as\n&gt; ./sqrt-x-minus-x diagnose test=gradient\nwhich produces\n...\nTEST GRADIENT MODE\n\n Log probability=-0.393734\n\n param idx           value           model     finite diff           error\n         0       -0.887393             nan               0             nan\nEven though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#numeric-literals",
+    "href": "reference-manual/expressions.html#numeric-literals",
+    "title": "Expressions",
+    "section": "",
+    "text": "The simplest form of expression is a literal that denotes a primitive numerical value.\n\n\nInteger literals represent integers of type int. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression --1 is interpreted as the negation of the literal -1.)\nThe following list contains well-formed integer literals.\n0, 1, -1, 256, -127098, 24567898765\nInteger literals must have values that fall within the bounds for integer values (see the section on numerical data types).\nInteger literals may not contain decimal points (.). Thus the expressions 1. and 1.0 are of type real and may not be used where a value of type int is required.\n\n\n\nA number written with a period or with scientific notation is assigned to a the continuous numeric type real. Real literals are written in base 10 with a period (.) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.\n0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.\nThe notation e or E followed by a positive or negative integer denotes a power of 10 to multiply. For instance, 2.7e3 and 2.7e+3 denote \\(2.7 \\times 10^3\\), whereas -2E-5 denotes \\(-2 \\times\n10^{-5}\\).\n\n\n\nA number followed by the character i denotes an imaginary number and is assigned to the numeric type complex. The number preceding i may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.\n1i, 2i, -325.786i, 1e10i, 2.87e-10i.\nNote that the character i by itself is not a well-formed imaginary literal. The unit imaginary number must be written as 1i.\n\n\n\nStan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.\n1 + 2i, -3.2e9 + 1e10i\nThese will be assigned the type complex, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#variables.section",
+    "href": "reference-manual/expressions.html#variables.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "A variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (_) character. Variables must start with a letter (a--z and A--Z) and may not end with two underscores (__).\nExamples of legal variable identifiers are as follows.\na, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable\nUnlike in R and BUGS, variable identifiers in Stan may not contain a period character.\n\n\nStan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the stanc translator halting with an error message indicating which reserved name was used and its location in the model code.\n\n\nThe name of the model cannot be used as a variable within the model. This is usually not a problem because the default in bin/stanc is to append _model to the name of the file containing the model specification. For example, if the model is in file foo.stan, it would not be legal to have a variable named foo_model when using the default model name through bin/stanc. With user-specified model names, variables cannot match the model.\n\n\n\nThe following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.\nfor, in, while, repeat, until, if, then, else,\ntrue, false, target, struct, typedef, export,\nauto, extern, var, static, lower, upper, offset,\nmultiplier\nVariables should not be named after types, either, and thus may not be any of the following.\nint, real, complex, vector, simplex, unit_vector,\nsum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,\nrow_vector, matrix, cholesky_factor_corr,\ncolumn_stochastic_matrix,row_stochastic_matrix,\ncholesky_factor_cov, corr_matrix, cov_matrix, array\nThe following keywords or built-in functions are also reserved and cannot be used as variable names:\nprint, reject, profile, fatal_error, target, jacobian\nThe following block identifiers are reserved and cannot be used as variable names:\nfunctions, model, data, parameters, quantities,\ntransformed, generated\n\n\n\nVariable names will also conflict with the names of distributions suffixed with _lpdf, _lpmf, _lcdf, and _lccdf, _cdf, and _ccdf, such as normal_lcdf_log. No user-defined variable can take a name ending in _lupdf or _lupmf even if a corresponding _lpdf or _lpmf is not defined.\nUsing any of these variable names causes the stanc translator to halt and report the name and location of the variable causing the conflict.\n\n\n\nStan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed _stan_NAME (e.g. _stan_public) behind the scenes (in the generated C++ code).\n\nalignas, alignof, and, and_eq, asm, bitand, bitor, bool,\ncase, catch, char, char16_t, char32_t, class, compl, const,\nconstexpr, const_cast, decltype, default, delete, do,\ndouble, dynamic_cast, enum, explicit, float, friend, goto,\ninline, long, mutable, namespace, new, noexcept, not, not_eq,\nnullptr, operator, or, or_eq, private, protected, public,\nregister, reinterpret_cast, short, signed, sizeof,\nstatic_assert, static_cast, switch, template, this, thread_local,\nthrow, try, typeid, typename, union, unsigned, using, virtual,\nvolatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,\nSTAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH\n\n\n\n\nThe legal characters for variable identifiers are given in the following table.\n\n\nThe alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.\n\n\n\ncharacters\nASCII code points\n\n\n\n\na -- z\n97 – 122\n\n\nA -- Z\n65 – 90\n\n\n0 -- 9\n48 – 57\n\n\n_\n95\n\n\n\nAlthough not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.\n\n\n\nWithin comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#vector-matrix-array-expressions.section",
+    "href": "reference-manual/expressions.html#vector-matrix-array-expressions.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "Expressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.\n\n\nSquare brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression [ 1, 10, 100 ] denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.\nrow_vector[2] rv2 =  [ 1, 2 ];\nvector[3] v3 = [ 3, 4, 5 ]';\nThe vector expression values may be compound expressions or variable names, so it is legal to write [ 2 * 3, 1 + 4] or [ x, y ], providing that x and y are primitive variables.\n\n\n\nA matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.\nmatrix[3, 2] m1 = [ [ 1, 2 ], [ 3, 4 ], [5, 6 ] ];\nAny expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:\nvector[2] vX = [ 1, 10 ]';\nrow_vector[2] vY = [ 100, 1000 ];\nmatrix[3, 2] m2 = [ vX', vY, [ 1, 2 ]  ];\n\n\n\nComplex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.\ncomplex_vector[3] = [1 + 2i, 3 - 1.7i, 0]';\ncomplex_row_vector[2] = [12, -2i];\ncomplex_matrix[2, 3] = [[1 + 2i, 3 - 1.7i, 0],\n                        [3.9 - 1.234i, 176i, 1 + 1i]];\n\n\nThe empty expression [ ] is ambiguous and therefore is not allowed and similarly expressions such as [ [ ] ] or [ [ ], [ ] ] are not allowed.\n\n\n\n\nIf needed, it is possible to create an empty vector with\nrep_vector(e, 0)\nwhere the first expression e needs to scalar of type real.\nIf needed, it is possible to create an empty matrix with\nrep_matrix(e, 0, 0)\nwhere the first expression e needs to scalar of type real.\n\n\n\nCurly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression { 1, 10, 100 } denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.\narray[3] int a = { 1, 10, 100 };\nThe values may be compound expressions, so it is legal to write { 2 * 3, 1 + 4 }. It is also possible to write two dimensional arrays directly, as in the following example.\narray[2, 3] int b = { { 1, 2, 3 }, { 4, 5, 6 } };\nThis way, b[1] is { 1, 2, 3 } and b[2] is { 4, 5, 6 }.\nWhitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.\narray[2, 3] int b = { { 1, 2, 3 },\n                { 4, 5, 6 } };\n\n\n\nThe empty array expression ({ }) is not allowed. See more about restrictions on array expressions in subsection Restrictions on values.\nIf needed, it is possible to create an empty array with\nrep_array(e, 0)\nwhere the first expression e determines the type of the array. For example, rep_array(0.0, 0) returns an empty real array of type real[], whereas rep_array({123}, 0) returns an empty two dimensional integer array of type int[ , ]. Only the type of the first argument is used, so the integer arrays {123} and {0} produce equivalent values.\n\n\n\nAny type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.\nvector[3] b;\nvector[3] c;\n// ...\narray[2] vector[3] d = { b, c };\nThe elements may also be a mixture of int and real typed expressions, in which case the result is an array of real values.\narray[2] real b = { 1, 1.9 };\n\n\n\nStan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.\ntuple(int, vector[3]) xy = (42, [1, 2.9, -1.3]');\nThe expression 42 is of type int and the expression [1, 2.9, -1.3] is of type row_vector so that [1, 2.9, -1.3]' is of type vector and of size 3. The whole tuple expression (42, [1, 2.9, -1.3]') thus has a sized type of tuple(int, vector[3]) and an unsized type (e.g., for a function argument) of tuple(int, vector).\nA tuple of one element can be created using the same style as languages like Python, with a trailing comma, e.g., (3.14,). For longer tuples, Stan does not support trailing commas.\n\n\n\nThere are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.\n\n\nAlthough it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.\n{ { 1, 2, 3 }, { 4, 5 } }  // compile time error: size mismatch\nThis may appear to be OK, because it is creating a two-dimensional integer array (array[,] int) out of two one-dimensional array integer arrays (array[] int). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.\n{ { 1, 2, 3 }, m }  // runtime error if m not size 3\n\n\n\nBecause there is no way to infer the type of the result, the empty array expression ({ }) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.\narray[0] int a;   // a is fully defined as zero element array\n\n\n\nThere is no way to declare or construct a zero-tuple in Stan. Tuples must be at least one element long. The expression () does not pick out a zero-tuple—it is ill formed.\nOne-tuples need a trailing comma, like (1,). The expression (1), without the comma, is of type int rather than a tuple.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#parentheses-for-grouping",
+    "href": "reference-manual/expressions.html#parentheses-for-grouping",
+    "title": "Expressions",
+    "section": "",
+    "text": "Any expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, ( and ), are allowed. The square brackets [ and ] are reserved for array indexing and the curly braces { and } for grouping statements.\nWith parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression 1 + 2 * 3 has a subexpression 2 * 3 and evaluates to 7. With parentheses, this grouping may be made explicit with the expression 1 + (2 * 3). More importantly, the expression (1 + 2) *   3 has 1 + 2 as a subexpression and evaluates to 9.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#arithmetic-expressions.section",
+    "href": "reference-manual/expressions.html#arithmetic-expressions.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "For integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (+), subtraction (-), multiplication (*) and division (/) in the usual ways.\nFor integer expressions, Stan supports the modulus (%) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3.0 + 0.14\n-15\n2 * 3 + 1\n(x - y) / 2.0\n(n * (n + 1)) / 2\nx / n\nm % n\nThe negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (') is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.\nFor example, if y and mu are variables of type vector and Sigma is a variable of type matrix, then (y - mu)' * Sigma * (y - mu) is a well-formed expression of type real. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) y - mu are of type vector because the variables y and mu are of type vector. The transpose of this expression, the subexpression (y - mu)' is of type row_vector. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form (((y - mu)') * Sigma) * (y - mu).\nThe type of subexpression (y - mu)' * Sigma is inferred to be row_vector, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a real value.\nStan provides elementwise matrix multiplication (e.g., a .* b) and division (e.g., a ./ b) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,\nvector[N] a;\nvector[N] b;\nvector[N] c;\nthe assignment,\nc = a .* b;\nproduces the same result with roughly the same efficiency as the loop\nfor (n in 1:N) {\n  c[n] = a[n] * b[n];\n}\nStan supports exponentiation (^) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3 ^ 2\n3.0 ^ -2\n3.0 ^ 0.14\nx ^ n\nn ^ x\nn ^ m\nx ^ y\nExponentiation is right associative, so the expression 2 ^ 3 ^ 4 is equivalent to the fully specified form 2 ^ (3 ^ 4).\n\n\nThe precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.\n\n\nStan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e., operator*(int, int):int indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.\n\n\n\n\n\n\n\n\n\n\nOp.\nPrec.\nAssoc.\nPlacement\nDescription\n\n\n\n\n? ~ :\n10\nright\nternary infix\nconditional\n\n\n||\n9\nleft\nbinary infix\nlogical or\n\n\n&&\n8\nleft\nbinary infix\nlogical and\n\n\n==\n7\nleft\nbinary infix\nequality\n\n\n!=\n7\nleft\nbinary infix\ninequality\n\n\n&lt;\n6\nleft\nbinary infix\nless than\n\n\n&lt;=\n6\nleft\nbinary infix\nless than or equal\n\n\n&gt;\n6\nleft\nbinary infix\ngreater than\n\n\n&gt;=\n6\nleft\nbinary infix\ngreater than or equal\n\n\n+\n5\nleft\nbinary infix\naddition\n\n\n-\n5\nleft\nbinary infix\nsubtraction\n\n\n*\n4\nleft\nbinary infix\nmultiplication\n\n\n.*\n4\nleft\nbinary infix\nelementwise multiplication\n\n\n/\n4\nleft\nbinary infix\n(right) division\n\n\n./\n4\nleft\nbinary infix\nelementwise division\n\n\n%\n4\nleft\nbinary infix\nmodulus\n\n\n\\\n3\nleft\nbinary infix\nleft division\n\n\n%/%\n3\nleft\nbinary infix\ninteger division\n\n\n!\n2\nn/a\nunary prefix\nlogical negation\n\n\n-\n2\nn/a\nunary prefix\nnegation\n\n\n+\n2\nn/a\nunary prefix\npromotion (no-op in Stan)\n\n\n^\n1\nright\nbinary infix\nexponentiation\n\n\n.^\n1\nright\nbinary infix\nelementwise exponentiation\n\n\n'\n0\nn/a\nunary postfix\ntransposition\n\n\n()\n0\nn/a\nprefix, wrap\nfunction application\n\n\n[]\n0\nleft\nprefix, wrap\narray, matrix indexing\n\n\n\nOther expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.\nThe precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression a + b + c is interpreted as (a + b) + c. Similarly, a / b * c is interpreted as (a / b) * c.\nBecause multiplication has higher precedence than addition, the expression a * b + c is interpreted as (a * b) + c and the expression a + b * c is interpreted as a + (b * c). Similarly, 2 * x + 3 * - y is interpreted as (2 * x) + (3 * (-y)).\nTransposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, -u' is interpreted as -(u'), u * v' as u* (v'), and u' * v as (u') * v. For integer and reals, -n ^ 3 is interpreted as -(n ^ 3).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#conditional-operator.section",
+    "href": "reference-manual/expressions.html#conditional-operator.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "The ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If a is an expression of type int and b and c are expressions that can be converted to one another (e.g., compared with ==), then\na ? b : c\nis an expression of the promoted type of b and c. The result of this expression is b if condition a is true, c otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g. if one argument is of type int and the other of type real, the conditional expression as a whole is of type real. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.\n\n\nThe conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,\na &gt; 0 || b &lt; 0 ? c + d : e - f\nis equivalent to the explicitly grouped version\n(a &gt; 0 || b &lt; 0) ? (c + d) : (e - f)\nThe latter is easier to read even if the parentheses are not strictly necessary.\n\n\n\nThe conditional operator is right associative, so that\na ? b : c ? d : e\nparses as if explicitly grouped as\na ? b : (c ? d : e)\nAgain, the explicitly grouped version is easier to read.\n\n\n\n\nStan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable a in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value b is returned if the condition evaluates to a non-zero value and c is returned if the condition evaluates to zero.\n\n\nThe key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.\n\n\n\nIf one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full if-then conditional statement. For example,\ndata {\n  array[10] real x;\n  // ...\n}\nparameters {\n  array[10] real z;\n  // ...\n}\nmodel {\n  y ~ normal(cond ? x : z, sigma);\n  // ...\n}\nwould be more efficiently (if not more transparently) coded as\nif (cond) {\n  y ~ normal(x, sigma);\n} else {\n  y ~ normal(z, sigma);\n}\nThe conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable x will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#language-indexing.section",
+    "href": "reference-manual/expressions.html#language-indexing.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "Stan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if x is a variable of type array [] real (a one-dimensional array of reals) then x[1] is the value of the first element of the array.\nSubscripting has higher precedence than any of the arithmetic operations. For example, alpha * x[1] is equivalent to alpha * (x[1]).\nMultiple subscripts may be provided within a single pair of square brackets. If x is of type array[,] real, a two-dimensional array, then x[2, 501] is of type real.\n\n\nThe subscripting operator also returns subarrays of arrays. For example, if x is of type array[,,] real, then x[2] is of type array[,] real, and x[2, 3] is of type array[] real. As a result, the expressions x[2, 3] and x[2][3] have the same meaning.\n\n\n\nIf Sigma is a variable of type matrix, then Sigma[1] denotes the first row of Sigma and has the type row_vector.\n\n\n\nStan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if m is of type matrix[ , ], a two-dimensional array of matrices, then m[1] refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that m[1, 2] is of type matrix and denotes the matrix in the first row and second column of the array. Continuing to add indices, m[1, 2, 3] is of type row_vector and denotes the third row of the matrix denoted by m[1, 2]. Finally, m[1, 2, 3, 4] is of type real and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array m.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#language-multi-indexing.section",
+    "href": "reference-manual/expressions.html#language-multi-indexing.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "In addition to single integer indexes, as described in the language indexing section, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.\n\n\nTypes of indexes and examples with one-dimensional containers of size N and an integer array ii of type array [] real size K.\n\n\n\nindex type\nexample\nvalue\n\n\n\n\ninteger\na[11]\nvalue of a at index 11\n\n\ninteger array\na[ii]\na[ii[1]], …, a[ii[K]]\n\n\nlower bound\na[3:]\na[3], …, a[N]\n\n\nupper bound\na[:5]\na[1], …, a[5]\n\n\nrange\na[2:7]\na[2], …, a[7]\n\n\nrange\na[7:2]\n[]\n\n\nrange\na[5-3:5+2]\na[2], …, a[7]\n\n\nall\na[:]\na[1], …, a[N]\n\n\nall\na[]\na[1], …, a[N]\n\n\n\nThe range indexing with : allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:\n  array[6] int ii = reverse(linspaced_int_array(6, 2, 7));\nThen a[ii] evaluates to a[7], …, a[2].\n\n\n\nThe fundamental semantic rule for dealing with multiple indexes is the following. If idxs is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.\na[idxs, ...][i, ...] = a[idxs[i], ...][...]\nOn the other hand, if idx is a single index, it reduces the dimensionality of the output, so that\na[idx, ...] = a[idx][...]\nThe only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.\n\n\nSpecial rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix a, with integer single indexes i and j and integer array multiple indexes is and js. The same typing rules apply for all multiple indexes.\n\n\n\nexample\nrow index\ncolumn index\nresult type\n\n\n\n\na[i]\nsingle\nn/a\nrow vector\n\n\na[is]\nmultiple\nn/a\nmatrix\n\n\na[i, j]\nsingle\nsingle\nreal\n\n\na[i, js]\nsingle\nmultiple\nrow vector\n\n\na[is, j]\nmultiple\nsingle\nvector\n\n\na[is, js]\nmultiple\nmultiple\nmatrix\n\n\n\nEvaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.\nm[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]\nm[idxs, idx][j] = m[idxs[j], idx]\nm[idx, idxs][j] = m[idx, idxs[j]]\nEvaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#function-application.section",
+    "href": "reference-manual/expressions.html#function-application.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "Stan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.\nExpressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, log(2.0) is the expression of type real denoting the result of applying the natural logarithm to the value of the real literal 2.0.\nSyntactically, function application has higher precedence than any of the other operators, so that y + log(x) is interpreted as y + (log(x)).\n\n\nEach function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as\nreal log(real);\nand the signature for the lmultiply function is\nreal lmultiply(real, real);\nA function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.\nreal mean(array [] real);\nreal mean(vector);\nThe first applies to a one-dimensional array of real values and the second to a vector.\nThe identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.\n\n\n\nConstants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants \\(\\pi\\) and \\(e\\) are represented as nullary functions named pi() and e(). See the Stan Functions Reference built-in constants section for a list of built-in constants.\n\n\n\nBecause of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.\nFor example, consider a situation in which the following two function signatures have been registered for foo.\nreal foo(real, real);\nint foo(int, int);\nThe use of foo in the expression foo(1.0, 1.0) resolves to foo(real, real), and thus the expression foo(1.0, 1.0) itself is assigned a type of real.\nBecause integers may be promoted to real values, the expression foo(1, 1) could potentially match either foo(real, real) or foo(int, int). The former requires two type promotions and the latter requires none, so foo(1, 1) is resolved to function foo(int, int) and is thus assigned the type int.\nThe expression foo(1, 1.0) has argument types (int, real) and thus does not explicitly match either function signature. By promoting the integer expression 1 to type real, it is able to match foo(real, real), and hence the type of the function expression foo(1, 1.0) is real.\nIn some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named bar with the following signatures.\nreal bar(real, int);\nreal bar(int, real);\nWith these signatures, the expression bar(1.0, 1) and bar(1, 1.0) resolve to the first and second of the above functions, respectively. The expression bar(1.0, 1.0) is illegal because real values may not be demoted to integers. The expression bar(1, 1) is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name bar is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with bar(1, 1) given the two declarations above, the Stan compiler will flag the expression as illegal.\n\n\n\nFor most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix _rng. For example, a univariate normal random number can be generated by normal_rng(0, 1); only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.\n\n\nThe use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in _rng.\nThis allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.\n\n\n\nPosterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of (Gelman et al. 2013) for more information on posterior predictive checks.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#type-inference",
+    "href": "reference-manual/expressions.html#type-inference",
+    "title": "Expressions",
+    "section": "",
+    "text": "Stan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.\n\n\nThe primitive implementation types for Stan are\nint, real, complex, vector, row_vector,  matrix, complex_vector,\ncomplex_row_vector, complex_matrix\nEvery basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.\n\n\nThe table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.\n\n\n\ntype\nprimitive type\n\n\n\n\nint\nint\n\n\nreal\nreal\n\n\nvector\nvector\n\n\nsimplex\nvector\n\n\nunit_vector\nvector\n\n\nsum_to_zero_vector\nvector\n\n\nordered\nvector\n\n\npositive_ordered\nvector\n\n\nrow_vector\nrow_vector\n\n\nmatrix\nmatrix\n\n\ncov_matrix\nmatrix\n\n\ncorr_matrix\nmatrix\n\n\ncholesky_factor_cov\nmatrix\n\n\ncholesky_factor_corr\nmatrix\n\n\ncolumn_stochastic_matrix\nmatrix\n\n\nrow_stochastic_matrix\nmatrix\n\n\nsum_to_zero_matrix\nmatrix\n\n\ncomplex_vector\ncomplex_vector\n\n\ncomplex_row_vector\ncomplex_row_vector\n\n\ncomplex_matrix\ncomplex_matrix\n\n\n\nA full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, array [] real has an array dimensionality of 1, int an array dimensionality of 0, and array [,,] int an array dimensionality of 3. The implementation type matrix[ , , ] has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.\nRecall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.\narray[I, J, K] matrix[M, N] a;\nThe matrix a is indexed as a[i, j, k, m, n] with the array indices first, followed by the matrix indices, with a[i, j, k] being a matrix and a[i, j, k, m] being a row vector.\n\n\n\n\nStan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.\n\n\n\nThere are two basic promotion rules,\n\nint types may be promoted to real, and\nreal types may be promoted to complex.\n\nPlus, promotion is transitive, so that\n\nif type U can be promoted to type V and type V can be promoted to type T, then U can be promoted to T.\n\nThe first rule means that expressions of type int may be used anywhere an expression of type real is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.\nThe remaining rules have to do with covariant typing rules, which say that a container of type U may be promoted to a container of the same shape of type T if U can be promoted to T. For vector and matrix types, this induces three rules,\n\nvector may be promoted to complex_vector,\nrow_vector may be promoted to complex_row_vector\nmatrix may be promoted to complex_matrix.\n\nFor array types, there’s a single rule\n\narray[...] U may be promoted to array[...] T if U can be promoted to T.\n\nFor example, this means array[,] int may be used where array [,] real or array [,] complex is required; as another example, array[] real may be used anywhere array[] complex is required.\nTuples have the natural extension of the above rules, applied to all sub-types at once\n\nA tuple(U1, ..., UN) may be promoted to a tuple(T1, ..., TN) if every Un can be promoted to Tn for n in 1:N\n\n\n\nAn integer literal expression such as 42 is of type int. Real literals such as 42.0 are of type real. Imaginary literals such as -17i are of type complex. the expression 7 - 2i acts like a complex literal, but technically it combines a real literal 7 and an imaginary literal 2i through subtraction.\n\n\n\nThe type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is int.\nThere is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.1\n\n\n\nIf x is an expression of total dimensionality greater than or equal to \\(N\\), then the type of expression e[i1, i2, ..., iN] is the same as that of e[i1][i2]...[iN], so it suffices to define the type of a singly-indexed function. Suppose e is an expression and i is an expression of primitive type int. Then\n\nif e is an expression of type array[i1, i2, ..., iN] T and k, i1, …, iN are expressions of type int, then e[k] is an expression of type array[i2, ..., iN] T,\nif e is an expression of type array[i] T with i and k expressions of type int, then e[k] is of type T,\nif e has implementation type vector or row_vector, dimensionality 0, then e[i] has implementation type real,\nif e has implementation type matrix, then e[i] has type row_vector,\nif e has implementation type complex_vector or complex_row_vector and i is an expression of type int, then e[i] is an expression of type complex, and\nif e has implementation type complex_matrix, and i is an expression of type int, then e[i] is an expression of type complex_row_vector.\n\n\n\n\nIf f is the name of a function and e1,...,eN are expressions for \\(N \\geq 0\\), then f(e1,...,eN) is an expression whose type is determined by the return type in the function signature for f given e1 through eN. Recall that a function signature is a declaration of the argument types and the result type.\nIn looking up functions, binary operators like real * real are defined as operator*(real, real) in the documentation and index.\nIn matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type int may be promoted to type real or complex if necessary (see the subsection on type promotion in the function application section, a real argument will be promoted to complex if necessary, a vector will be promoted to complex_vector if necessary, and so on.\nIn general, matrix operations return the lowest inferable type. For example, row_vector * vector returns a value of type real, which is declared in the function documentation and index as real operator*(row_vector, vector).",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#higher-order-functions",
+    "href": "reference-manual/expressions.html#higher-order-functions",
+    "title": "Expressions",
+    "section": "",
+    "text": "There are several expression constructions in Stan that act as higher-order functions.2\nThe higher-order functions and the signature of their argument functions are listed in the following pair of tables.\n\n\nHigher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.\n\n\n\n\n\n\n\n\n\nfunction\nparameter or data args\ndata args\nreturn type\n\n\n\n\nalgebra_solver\nvector, vector\narray [] real, array [] real\nvector\n\n\nalgebra_solver_newton\nvector, vector\narray [] real, array [] real\nvector\n\n\nintegrate_1d,\nreal, real, array [] real\narray [] real, array [] real\nreal\n\n\nintegrate_ode_X,\nreal, array [] real, array [] real\narray [] real, array [] real\narray [] real\n\n\nmap_rect\nvector, vector\narray [] real, array [] real\nvector\n\n\n\nFor example, the integrate_ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  array [] real foo(real t,\n                    array [] real y,\n                    array [] real theta,\n                    array [] real x_r,\n                    array [] real x_i) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\narray[2] real y0;\nreal t0;\narray[T] real ts;\narray[1] real theta;\narray[0] real x_r;\narray[0] int x_i;\n// ...\narray[T, 2] real y_hat = integrate_ode_rk45(foo, y0, t0,\n                                              ts, theta, x_r, x_i);\nThe function argument is foo, the name of the user-defined function; as shown in the higher-order functions table, integrate_ode_rk45 takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.\n\n\n\nVariadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.\n\n\n\nfunction\nrestricted args\nreturn type\n\n\n\n\nsolve_X\nvector\nvector\n\n\node_X,\nvector, real, array [] real\nvector[]\n\n\nreduce_sum\narray[] T, T1, T2\nreal\n\n\n\nT, T1, and T2 can be any Stan type.\nFor example, the ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  vector foo(real t, vector y, real theta, vector beta,\n            array [] real x_i, int index) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\nvector[2] y0;\nreal t0;\narray[T] real ts;\nreal theta;\nvector[7] beta;\narray[10] int x_i;\nint index;\n// ...\nvector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,\n                              beta, x_i, index);\nThe function argument is foo, the name of the user-defined function. As shown in the variadic higher-order functions table, ode_rk45 takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of foo and returns an array of vectors.\n\n\n\nThe function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.\n\n\n\nSome of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.\nFor user-defined functions the qualifier data may be prepended to the type to restrict the argument to data-only variables.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#chain-rule-and-derivatives",
+    "href": "reference-manual/expressions.html#chain-rule-and-derivatives",
+    "title": "Expressions",
+    "section": "",
+    "text": "Derivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.\n\n\nUnlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(sqrt(x - x), 1);\n}\nAlgebraically, the distribution statement in the model could be reduced to\n  x ~ normal(0, 1);\nand it would seem the model should produce unit normal draws for x. But rather than canceling, the expression sqrt(x -   x) causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,\n\\[\n\\begin{array}{rcl}\n\\frac{d}{dx} \\sqrt{x - x}\n& = &\n\\frac{1}{2 \\sqrt{x - x}} \\times \\frac{d}{dx} (x - x)\n\\\\[4pt]\n& = &\n\\frac{1}{0} \\times (1 - 1)\n\\\\[4pt]\n& = &\n\\infty \\times 0\n\\\\[4pt]\n& = & \\mathrm{NaN}.\n\\end{array}\n\\]\nRather than the \\(x - x\\) canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.\nThe only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like sqrt(x - x) for which the chain rule produces not-a-number values.\n\n\n\nThe best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).\nFor example, compiling the above model to an executable sqrt-x-minus-x in CmdStan, the test can be run as\n&gt; ./sqrt-x-minus-x diagnose test=gradient\nwhich produces\n...\nTEST GRADIENT MODE\n\n Log probability=-0.393734\n\n param idx           value           model     finite diff           error\n         0       -0.887393             nan               0             nan\nEven though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/expressions.html#footnotes",
+    "href": "reference-manual/expressions.html#footnotes",
+    "title": "Expressions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nLanguages such as C++ and R allow the declaration of a variable of a given name in a narrower scope to hide (take precedence over for evaluation) a variable defined in a containing scope.↩︎\nInternally, they are implemented as their own expression types because Stan doesn’t have object-level functional types (yet).↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Expressions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/encoding.html",
+    "href": "reference-manual/encoding.html",
+    "title": "Character Encoding",
+    "section": "",
+    "text": "The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.\n\n\nThe UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.\n\n\n\n\nAny bytes on a line after a line-comment sequence (// or #) are ignored up until the ASCII newline character (\\n). They may thus be written in any character encoding which is convenient.\nAny content after a block comment open sequence in ASCII (/*) up to the closing block comment (*/) is ignored, and thus may also be written in whatever character set is convenient.\n\n\n\nThe raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A print, reject, or fatal_error statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (\") when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.\nASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
+    ]
+  },
+  {
+    "objectID": "reference-manual/encoding.html#content-characters",
+    "href": "reference-manual/encoding.html#content-characters",
+    "title": "Character Encoding",
+    "section": "",
+    "text": "The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.\n\n\nThe UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
+    ]
+  },
+  {
+    "objectID": "reference-manual/encoding.html#comment-characters",
+    "href": "reference-manual/encoding.html#comment-characters",
+    "title": "Character Encoding",
+    "section": "",
+    "text": "Any bytes on a line after a line-comment sequence (// or #) are ignored up until the ASCII newline character (\\n). They may thus be written in any character encoding which is convenient.\nAny content after a block comment open sequence in ASCII (/*) up to the closing block comment (*/) is ignored, and thus may also be written in whatever character set is convenient.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
+    ]
+  },
+  {
+    "objectID": "reference-manual/encoding.html#string-literals",
+    "href": "reference-manual/encoding.html#string-literals",
+    "title": "Character Encoding",
+    "section": "",
+    "text": "The raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A print, reject, or fatal_error statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (\") when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.\nASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
+    ]
+  },
+  {
+    "objectID": "reference-manual/deprecations.html",
+    "href": "reference-manual/deprecations.html",
+    "title": "Deprecated Features",
+    "section": "",
+    "text": "This appendix lists currently deprecated functionality along with how to replace it.\nStarting with Stan 2.29, minor (syntax-level) deprecations can be removed 3 versions after release; e.g., syntax deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in Removed Features. The Stan compiler can automatically update many of these on the behalf of the user for at least one version after they are removed.\nAny feature which changes semantic meaning (such as the upgraded ODE solver interface) will not be removed until a major version change (e.g., Stan 3.0).\n\n\nDeprecated:The distribution lkj_cov is deprecated.\nReplacement: Replace lkj_cov_lpdf(...) with an lkj_corr distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace\ncov_matrix[K] Sigma;\n// ...\nSigma ~ lkj_cov(mu, tau, eta);\nwith\ncorr_matrix[K] Omega;\nvector&lt;lower=0&gt;[K] sigma;\n// ...\nOmega ~ lkj_corr(eta);\nsigma ~ lognormal(mu, tau);\n// ...\ncov_matrix[K] Sigma;\nSigma &lt;- quad_form_diag(Omega, sigma);\nThe variable Sigma may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.\nScheduled Removal: Stan 3.0 or later.\n\n\n\nDeprecated: Using functions that end in _lp in the transformed parameters block.\nReplacement: Use _jacobian functions and the jacobian += statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.\n\n\n\nSeveral built-in Stan functions have been deprecated. Consult the functions reference for more information.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/deprecations.html#lkj_cov-distribution",
+    "href": "reference-manual/deprecations.html#lkj_cov-distribution",
+    "title": "Deprecated Features",
+    "section": "",
+    "text": "Deprecated:The distribution lkj_cov is deprecated.\nReplacement: Replace lkj_cov_lpdf(...) with an lkj_corr distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace\ncov_matrix[K] Sigma;\n// ...\nSigma ~ lkj_cov(mu, tau, eta);\nwith\ncorr_matrix[K] Omega;\nvector&lt;lower=0&gt;[K] sigma;\n// ...\nOmega ~ lkj_corr(eta);\nsigma ~ lognormal(mu, tau);\n// ...\ncov_matrix[K] Sigma;\nSigma &lt;- quad_form_diag(Omega, sigma);\nThe variable Sigma may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.\nScheduled Removal: Stan 3.0 or later.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/deprecations.html#use-of-_lp-functions-in-transformed-parameters",
+    "href": "reference-manual/deprecations.html#use-of-_lp-functions-in-transformed-parameters",
+    "title": "Deprecated Features",
+    "section": "",
+    "text": "Deprecated: Using functions that end in _lp in the transformed parameters block.\nReplacement: Use _jacobian functions and the jacobian += statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/deprecations.html#deprecated-functions",
+    "href": "reference-manual/deprecations.html#deprecated-functions",
+    "title": "Deprecated Features",
+    "section": "",
+    "text": "Several built-in Stan functions have been deprecated. Consult the functions reference for more information.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html",
+    "href": "reference-manual/blocks.html",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "A Stan program is organized into a sequence of named blocks, the bodies of which consist of variable declarations, followed in the case of some blocks with statements.\n\n\nThe full set of named program blocks is exemplified in the following skeletal Stan program.\nfunctions {\n  // ... function declarations and definitions ...\n}\ndata {\n  // ... declarations ...\n}\ntransformed data {\n   // ... declarations ... statements ...\n}\nparameters {\n   // ... declarations ...\n}\ntransformed parameters {\n   // ... declarations ... statements ...\n}\nmodel {\n   // ... declarations ... statements ...\n}\ngenerated quantities {\n   // ... declarations ... statements ...\n}\nThe function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.\n\n\nAll of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.\n\n\n\nThe variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.\nVariables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).\n\n\n\nFunctions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).\nRandom-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in _rng; such functions are suffixed with _rng. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with _lp.\nDensity functions defined in the program may be used in distribution statements.\n\n\n\nThe variables declared in the data and parameters block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.\nThe variables in the data block are read from an external input source such as a file or a designated R data structure. The variables in the parameters block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.\n\n\n\nThe transformed data and transformed parameters block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.\nBefore generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.1\nTransformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.\n\n\n\nThe generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.\n\n\n\nA table summarizing the point at which variables are read, written, and defined is given in the block actions table.\nBlock Actions Table.  The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.\n\n\n\nblock\nstatement\naction / period\n\n\n\n\ndata\nno\nread / chain\n\n\ntransformed data\nyes\nevaluate / chain\n\n\nparameters\nno\ninv. transform, Jacobian / leapfrog\n\n\n \n \ninv. transform, write / sample\n\n\ntransformed parameters\nyes\nevaluate / leapfrog\n\n\n \n \nwrite / sample\n\n\nmodel\nyes\nevaluate / leapfrog step\n\n\ngenerated quantities\nyes\neval / sample\n\n\n \n \nwrite / sample\n\n\n(initialization)\nn/a\nread, transform / chain\n\n\n\nVariable Declaration Table.  This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (\\(*\\)) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.\n\n\n\nparam depend\nin target\nsave\ndeclare in\n\n\n\n\n+\n+\n+\ntransformed parameters\n\n\n+\n+\n-\nmodel (local)\n\n\n+\n-\n+\ngenerated quantities\n\n\n+\n-\n-\ngenerated quantities (local)\n\n\n-\n+\n+\ntransformed data   and generated quantities\n\n\n-\n+\n-\ntransformed data\n\n\n-\n-\n+\ngenerated quantities\n\n\n-\n-\n-\ntransformed data (local)\n\n\n\nAnother way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the variable declaration table. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.2\nThe rest of this chapter provides full details on when and how the variables and statements in each block are executed.\n\n\n\n\nStatistical Variable Taxonomy Table.  Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.\n\n\n\n\n\n\n\nvariable kind\ndeclaration block\n\n\n\n\nconstants\ndata, transformed data\n\n\nunmodeled data\ndata, transformed data\n\n\nmodeled data\ndata, transformed data\n\n\nmissing data\nparameters, transformed parameters\n\n\nmodeled parameters\nparameters, transformed parameters\n\n\nunmodeled parameters\ndata, transformed data\n\n\nderived quantities\ntransformed data, transformed parameters, generated quantities\n\n\nloop indices\nloop statement\n\n\n\nPage 366 of (Gelman and Hill 2007) provides a taxonomy of the kinds of variables used in Bayesian models. The table of kinds of variables contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.\nConstants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the data block.\nThe following program illustrates various variables kinds, listing the kind of each variable next to its declaration.\ndata {\n  int&lt;lower=0&gt; N;           // unmodeled data\n  array[N] real y;          // modeled data\n  real mu_mu;               // config. unmodeled param\n  real&lt;lower=0&gt; sigma_mu;   // config. unmodeled param\n}\ntransformed data {\n  real&lt;lower=0&gt; alpha;      // const. unmodeled param\n  real&lt;lower=0&gt; beta;       // const. unmodeled param\n  alpha = 0.1;\n  beta = 0.1;\n}\nparameters {\n  real mu_y;                // modeled param\n  real&lt;lower=0&gt; tau_y;      // modeled param\n}\ntransformed parameters {\n  real&lt;lower=0&gt; sigma_y;    // derived quantity (param)\n  sigma_y = pow(tau_y, -0.5);\n}\nmodel {\n  tau_y ~ gamma(alpha, beta);\n  mu_y ~ normal(mu_mu, sigma_mu);\n  for (n in 1:N) {\n    y[n] ~ normal(mu_y, sigma_y);\n  }\n}\ngenerated quantities {\n  real variance_y;       // derived quantity (transform)\n  variance_y = sigma_y * sigma_y;\n}\nIn this example, y is an array of modeled data. Although it is specified in the data block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.\nThe variable N is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.\nThe other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the data block may be specified on a per-model execution basis as part of the data read. In the above model, mu_mu and sigma_mu are configurable unmodeled parameters.\nUnmodeled parameters that are hard coded in the model must be declared in the transformed data block. For example, the unmodeled parameters alpha and beta are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the data block, like the variables mu_mu and sigma_mu.\nThis program declares two modeled parameters, mu_y and tau_y. These are the location and precision used in the normal model of the values in y. The heart of the model will be sampling the values of these parameters from their posterior distribution.\nThe modeled parameter tau_y is transformed from a precision to a scale parameter and assigned to the variable sigma_y in the transformed parameters block. Thus the variable sigma_y is considered a derived quantity — its value is entirely determined by the values of other variables.\nThe generated quantities block defines a value variance_y, which is defined as a transform of the scale or deviation parameter sigma_y. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).\nIn later versions of Stan which have random number generators for the distributions, the generated quantities block will be usable to generate replicated data for model checking.\nFinally, the variable n is used as a loop index in the model block.\n\n\n\nThe rest of this chapter will lay out the details of each block in order, starting with the data block in this section.\n\n\nThe data block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.3\nData variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.\n\n\n\nThe data block does not allow statements.\n\n\n\nEach variable’s value is validated against its declaration as it is read. For example, if a variable sigma is declared as real&lt;lower=0&gt;, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.\n\n\n\n\nThe transformed data block is for declaring and defining variables that do not need to be changed when running the program.\n\n\nFor the transformed data block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.\nVariables declared in the data block may be used to declare transformed variables.\n\n\n\nThe statements in a transformed data block are used to define (provide values for) variables declared in the transformed data block. Assignments are only allowed to variables declared in the transformed data block.\nThese statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.\nVariables declared in the data block may be used in statements in the transformed data block.\n\n\nThe statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and distribution statements may not be used.\n\n\n\n\nAny constraints on variables declared in the transformed data block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.\n\n\n\n\nThe variables declared in the parameters program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block are the parameters being sampled by Stan.\nVariables declared as parameters cannot be directly assigned values. So there is no block of statements in the parameters program block. Variable quantities derived from parameters may be declared in the transformed parameters or generated quantities blocks, or may be defined as local variables in any statement blocks following their declaration.\nThere is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.\n\n\nStan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of \\(\\mathbb{R}^n\\). To do this, the parameters defined in the parameters block must be transformed so they are unconstrained.\nIn practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s parameters program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.\nIn some cases, the number of parameters is reduced in the unconstrained space. For instance, a \\(K\\)-simplex only requires \\(K-1\\) unconstrained parameters, and a \\(K\\)-correlation matrix only requires \\(\\binom{K}{2}\\) unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the parameters program block.\nThe probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the parameters program block. The details of the specific transformations are provided in the variable transforms chapter.\n\n\n\nHamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.\nGradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.\nThe amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.\n\n\n\nIn the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the parameters block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.\n\n\n\n\nThe transformed parameters program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.\nAny variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.\n\n\nLike the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the section on reject statements for further discussion of the behavior of reject statements.\n\n\n\n\nThe model program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.\nLocal variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.\nThe statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.\n\n\n\nThe generated quantities program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.\nAmong the applications of posterior inference that can be coded in the generated quantities block are\n\nforward sampling to generate simulated data for model testing,\ngenerating predictions for new data,\ncalculating posterior event probabilities, including multiple comparisons, sign tests, etc.,\ncalculating posterior expectations,\ntransforming parameters for reporting,\napplying full Bayesian decision theory,\ncalculating log likelihoods, deviances, etc. for model comparison.\n\nParameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.\nWithin the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.\nIt is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.\nAfter the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.\nAll variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:\ngenerated quantities {\n  int a; // added to the output\n\n  {\n    int b; // not added to the output\n  }\n}",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#overview-of-stans-program-blocks",
+    "href": "reference-manual/blocks.html#overview-of-stans-program-blocks",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The full set of named program blocks is exemplified in the following skeletal Stan program.\nfunctions {\n  // ... function declarations and definitions ...\n}\ndata {\n  // ... declarations ...\n}\ntransformed data {\n   // ... declarations ... statements ...\n}\nparameters {\n   // ... declarations ...\n}\ntransformed parameters {\n   // ... declarations ... statements ...\n}\nmodel {\n   // ... declarations ... statements ...\n}\ngenerated quantities {\n   // ... declarations ... statements ...\n}\nThe function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.\n\n\nAll of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.\n\n\n\nThe variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.\nVariables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).\n\n\n\nFunctions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).\nRandom-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in _rng; such functions are suffixed with _rng. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with _lp.\nDensity functions defined in the program may be used in distribution statements.\n\n\n\nThe variables declared in the data and parameters block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.\nThe variables in the data block are read from an external input source such as a file or a designated R data structure. The variables in the parameters block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.\n\n\n\nThe transformed data and transformed parameters block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.\nBefore generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.1\nTransformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.\n\n\n\nThe generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.\n\n\n\nA table summarizing the point at which variables are read, written, and defined is given in the block actions table.\nBlock Actions Table.  The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.\n\n\n\nblock\nstatement\naction / period\n\n\n\n\ndata\nno\nread / chain\n\n\ntransformed data\nyes\nevaluate / chain\n\n\nparameters\nno\ninv. transform, Jacobian / leapfrog\n\n\n \n \ninv. transform, write / sample\n\n\ntransformed parameters\nyes\nevaluate / leapfrog\n\n\n \n \nwrite / sample\n\n\nmodel\nyes\nevaluate / leapfrog step\n\n\ngenerated quantities\nyes\neval / sample\n\n\n \n \nwrite / sample\n\n\n(initialization)\nn/a\nread, transform / chain\n\n\n\nVariable Declaration Table.  This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (\\(*\\)) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.\n\n\n\nparam depend\nin target\nsave\ndeclare in\n\n\n\n\n+\n+\n+\ntransformed parameters\n\n\n+\n+\n-\nmodel (local)\n\n\n+\n-\n+\ngenerated quantities\n\n\n+\n-\n-\ngenerated quantities (local)\n\n\n-\n+\n+\ntransformed data   and generated quantities\n\n\n-\n+\n-\ntransformed data\n\n\n-\n-\n+\ngenerated quantities\n\n\n-\n-\n-\ntransformed data (local)\n\n\n\nAnother way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the variable declaration table. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.2\nThe rest of this chapter provides full details on when and how the variables and statements in each block are executed.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#statistical-variable-taxonomy",
+    "href": "reference-manual/blocks.html#statistical-variable-taxonomy",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "Statistical Variable Taxonomy Table.  Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.\n\n\n\n\n\n\n\nvariable kind\ndeclaration block\n\n\n\n\nconstants\ndata, transformed data\n\n\nunmodeled data\ndata, transformed data\n\n\nmodeled data\ndata, transformed data\n\n\nmissing data\nparameters, transformed parameters\n\n\nmodeled parameters\nparameters, transformed parameters\n\n\nunmodeled parameters\ndata, transformed data\n\n\nderived quantities\ntransformed data, transformed parameters, generated quantities\n\n\nloop indices\nloop statement\n\n\n\nPage 366 of (Gelman and Hill 2007) provides a taxonomy of the kinds of variables used in Bayesian models. The table of kinds of variables contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.\nConstants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the data block.\nThe following program illustrates various variables kinds, listing the kind of each variable next to its declaration.\ndata {\n  int&lt;lower=0&gt; N;           // unmodeled data\n  array[N] real y;          // modeled data\n  real mu_mu;               // config. unmodeled param\n  real&lt;lower=0&gt; sigma_mu;   // config. unmodeled param\n}\ntransformed data {\n  real&lt;lower=0&gt; alpha;      // const. unmodeled param\n  real&lt;lower=0&gt; beta;       // const. unmodeled param\n  alpha = 0.1;\n  beta = 0.1;\n}\nparameters {\n  real mu_y;                // modeled param\n  real&lt;lower=0&gt; tau_y;      // modeled param\n}\ntransformed parameters {\n  real&lt;lower=0&gt; sigma_y;    // derived quantity (param)\n  sigma_y = pow(tau_y, -0.5);\n}\nmodel {\n  tau_y ~ gamma(alpha, beta);\n  mu_y ~ normal(mu_mu, sigma_mu);\n  for (n in 1:N) {\n    y[n] ~ normal(mu_y, sigma_y);\n  }\n}\ngenerated quantities {\n  real variance_y;       // derived quantity (transform)\n  variance_y = sigma_y * sigma_y;\n}\nIn this example, y is an array of modeled data. Although it is specified in the data block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.\nThe variable N is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.\nThe other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the data block may be specified on a per-model execution basis as part of the data read. In the above model, mu_mu and sigma_mu are configurable unmodeled parameters.\nUnmodeled parameters that are hard coded in the model must be declared in the transformed data block. For example, the unmodeled parameters alpha and beta are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the data block, like the variables mu_mu and sigma_mu.\nThis program declares two modeled parameters, mu_y and tau_y. These are the location and precision used in the normal model of the values in y. The heart of the model will be sampling the values of these parameters from their posterior distribution.\nThe modeled parameter tau_y is transformed from a precision to a scale parameter and assigned to the variable sigma_y in the transformed parameters block. Thus the variable sigma_y is considered a derived quantity — its value is entirely determined by the values of other variables.\nThe generated quantities block defines a value variance_y, which is defined as a transform of the scale or deviation parameter sigma_y. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).\nIn later versions of Stan which have random number generators for the distributions, the generated quantities block will be usable to generate replicated data for model checking.\nFinally, the variable n is used as a loop index in the model block.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#program-block-data",
+    "href": "reference-manual/blocks.html#program-block-data",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The rest of this chapter will lay out the details of each block in order, starting with the data block in this section.\n\n\nThe data block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.3\nData variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.\n\n\n\nThe data block does not allow statements.\n\n\n\nEach variable’s value is validated against its declaration as it is read. For example, if a variable sigma is declared as real&lt;lower=0&gt;, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#program-block-transformed-data",
+    "href": "reference-manual/blocks.html#program-block-transformed-data",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The transformed data block is for declaring and defining variables that do not need to be changed when running the program.\n\n\nFor the transformed data block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.\nVariables declared in the data block may be used to declare transformed variables.\n\n\n\nThe statements in a transformed data block are used to define (provide values for) variables declared in the transformed data block. Assignments are only allowed to variables declared in the transformed data block.\nThese statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.\nVariables declared in the data block may be used in statements in the transformed data block.\n\n\nThe statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and distribution statements may not be used.\n\n\n\n\nAny constraints on variables declared in the transformed data block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#program-block-parameters",
+    "href": "reference-manual/blocks.html#program-block-parameters",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The variables declared in the parameters program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block are the parameters being sampled by Stan.\nVariables declared as parameters cannot be directly assigned values. So there is no block of statements in the parameters program block. Variable quantities derived from parameters may be declared in the transformed parameters or generated quantities blocks, or may be defined as local variables in any statement blocks following their declaration.\nThere is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.\n\n\nStan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of \\(\\mathbb{R}^n\\). To do this, the parameters defined in the parameters block must be transformed so they are unconstrained.\nIn practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s parameters program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.\nIn some cases, the number of parameters is reduced in the unconstrained space. For instance, a \\(K\\)-simplex only requires \\(K-1\\) unconstrained parameters, and a \\(K\\)-correlation matrix only requires \\(\\binom{K}{2}\\) unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the parameters program block.\nThe probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the parameters program block. The details of the specific transformations are provided in the variable transforms chapter.\n\n\n\nHamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.\nGradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.\nThe amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.\n\n\n\nIn the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the parameters block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#program-block-transformed-parameters",
+    "href": "reference-manual/blocks.html#program-block-transformed-parameters",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The transformed parameters program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.\nAny variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.\n\n\nLike the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the section on reject statements for further discussion of the behavior of reject statements.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#program-block-model",
+    "href": "reference-manual/blocks.html#program-block-model",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The model program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.\nLocal variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.\nThe statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#program-block-generated-quantities",
+    "href": "reference-manual/blocks.html#program-block-generated-quantities",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The generated quantities program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.\nAmong the applications of posterior inference that can be coded in the generated quantities block are\n\nforward sampling to generate simulated data for model testing,\ngenerating predictions for new data,\ncalculating posterior event probabilities, including multiple comparisons, sign tests, etc.,\ncalculating posterior expectations,\ntransforming parameters for reporting,\napplying full Bayesian decision theory,\ncalculating log likelihoods, deviances, etc. for model comparison.\n\nParameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.\nWithin the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.\nIt is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.\nAfter the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.\nAll variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:\ngenerated quantities {\n  int a; // added to the output\n\n  {\n    int b; // not added to the output\n  }\n}",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "reference-manual/blocks.html#footnotes",
+    "href": "reference-manual/blocks.html#footnotes",
+    "title": "Program Blocks",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nIf the C++ code is configured for concurrent threads, the data and transformed data blocks can be executed once and reused for multiple chains.↩︎\nIt is possible to print a variable every iteration that does not depend on parameters—just define it (or redefine it if it is transformed data) in the generated quantities block.↩︎\nWith multiple threads, or even running chains sequentially in a single thread, data could be read only once per set of chains. Stan was designed to be thread safe and future versions will provide a multithreading option for Markov chains.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
+    ]
+  },
+  {
+    "objectID": "functions-reference/void_functions.html",
+    "href": "functions-reference/void_functions.html",
+    "title": "Void Functions",
+    "section": "",
+    "text": "Stan supports a few special statements for printing or for signaling an issue with the program.\nAlthough print, reject, and fatal_error appear to have the syntax of functions, they are actually special kinds of statements with slightly different form and behavior than other functions. First, they are the constructs that allow a variable number of arguments. Second, they are the the only constructs to accept string literals (e.g., \"hello world\") as arguments. Third, they have no effect on the log density function and operate solely through side effects.\nThe special keyword void is used for their return type because they behave like variadic functions with void return type, even though they are special kinds of statements built in to the language.\n\n\nPrinting has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).\n \n\nvoid print(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.1\n\n\n\nThe reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).\n \n\nvoid reject(T1 x1,..., TN xN) Reject the current iteration and print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.18\n\n\n\nThe fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.\n \n\nvoid fatal_error(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.35",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Void Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/void_functions.html#print-statement",
+    "href": "functions-reference/void_functions.html#print-statement",
+    "title": "Void Functions",
+    "section": "",
+    "text": "Printing has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).\n \n\nvoid print(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.1",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Void Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/void_functions.html#reject-statement",
+    "href": "functions-reference/void_functions.html#reject-statement",
+    "title": "Void Functions",
+    "section": "",
+    "text": "The reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).\n \n\nvoid reject(T1 x1,..., TN xN) Reject the current iteration and print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Void Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/void_functions.html#fatal-error-statement",
+    "href": "functions-reference/void_functions.html#fatal-error-statement",
+    "title": "Void Functions",
+    "section": "",
+    "text": "The fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.\n \n\nvoid fatal_error(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.35",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Void Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html",
+    "href": "functions-reference/unbounded_continuous_distributions.html",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "The unbounded univariate continuous probability distributions have support on all real numbers.\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Normal}(y|\\mu,\\sigma) =\n\\frac{1}{\\sqrt{2 \\pi} \\ \\sigma} \\exp\\left( - \\, \\frac{1}{2} \\left(\\frac{y - \\mu}{\\sigma} \\right)^2 \\right) \\!.\n\\end{equation*}\\]\n\n\n\ny ~ normal(mu, sigma)\nIncrement target log probability density with normal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal normal_lpdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal normal_lupdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_cdf(reals y | reals mu, reals sigma) The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 1 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; the function Phi_approx is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.0\n \n\nreal normal_lcdf(reals y | reals mu, reals sigma) The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.12\n \n\nreal normal_lccdf(reals y | reals mu, reals sigma) The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.15\n \n\nR normal_rng(reals mu, reals sigma) Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nThe standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. \\[\\begin{equation*}\n\\text{StdNormal}(y) \\ = \\ \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{1}{\\sqrt{2 \\pi}} \\, \\exp \\left( \\frac{-y^2}{2} \\right)\\!.\n\\end{equation*}\\] Up to a proportion on the log scale, where Stan computes, \\[\\begin{equation*}\n\\log \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{-y^2}{2} + \\text{const}.\n\\end{equation*}\\] With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location \\(0\\) and scale \\(1\\).\n\n\n\ny ~ std_normal()\nIncrement target log probability density with std_normal_lupdf(y).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal std_normal_lpdf(reals y) The standard normal (location zero, scale one) log probability density of y.\nAvailable since 2.18\n \n\nreal std_normal_lupdf(reals y) The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.\nAvailable since 2.25\n \n\nreal std_normal_cdf(reals y) The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for \\(y\\) below -37.5 and overflow to 1 for \\(y\\) above 8.25; the function Phi_approx is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lcdf(reals y) The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to \\(-\\infty\\) for \\(y\\) below -37.5 and overflow to 0 for \\(y\\) above 8.25; log(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lccdf(reals y) The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for \\(y\\) below -37.5 and underflow to \\(-\\infty\\) for \\(y\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nR std_normal_qf(T x) Returns the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The std_normal_qf is equivalent to the inv_Phi function.\nAvailable since 2.31\n \n\nR std_normal_log_qf(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) evaluated at the log of the specified quantile x. This function is equivalent to std_normal_qf(exp(x)) but is more numerically stable.\nAvailable since 2.31\n \n\nreal std_normal_rng() Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.21\n\n\n\n\nStan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e. a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\sigma\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^n\\), \\[\\begin{equation*} \\text{NormalIdGLM}(y|x, \\alpha, \\beta, \\sigma) = \\prod_{1\\leq i\n\\leq n}\\text{Normal}(y_i|\\alpha_i + x_i\\cdot \\beta, \\sigma). \\end{equation*}\\]\n\n\n\ny ~ normal_id_glm(x, alpha, beta, sigma)\nIncrement target log probability density with normal_id_glm_lupdf(y | x, alpha, beta, sigma).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n\n\n\n\nExponentially modified Gaussian describes the distribution of \\(Z = X + Y\\) when \\(X\\) and \\(Y\\) are independent and \\(X\\) is normally distributed (with mean \\(\\mu\\) and standard deviation \\(\\sigma\\)) and \\(Y\\) is exponentially distributed (with rate \\(\\lambda\\)).\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\), and \\(\\lambda \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{ExpModNormal}(y|\\mu,\\sigma,\\lambda) = \\frac{\\lambda}{2} \\ \\exp\n\\left(\\frac{\\lambda}{2} \\left(2\\mu + \\lambda \\sigma^2 -\n2y\\right)\\right) \\text{erfc}\\left(\\frac{\\mu + \\lambda\\sigma^2 -\ny}{\\sqrt{2}\\sigma}\\right) . \\end{equation*}\\]\n\n\n\ny ~ exp_mod_normal(mu, sigma, lambda)\nIncrement target log probability density with exp_mod_normal_lupdf(y | mu, sigma, lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exp_mod_normal_lpdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lupdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal exp_mod_normal_cdf(reals y | reals mu, reals sigma, reals lambda) The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.0\n \n\nreal exp_mod_normal_lcdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lccdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nR exp_mod_normal_rng(reals mu, reals sigma, reals lambda) Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\xi \\in \\mathbb{R}\\), \\(\\omega \\in \\mathbb{R}^+\\), and \\(\\alpha \\in\n\\mathbb{R}\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*} \\text{SkewNormal}(y \\mid\n\\xi, \\omega, \\alpha) = \\frac{1}{\\omega\\sqrt{2\\pi}} \\ \\exp\\left( - \\,\n\\frac{1}{2}            \\left(  \\frac{y - \\xi}{\\omega} \\right)^2\n\\right) \\ \\left(1 + \\text{erf}\\left( \\alpha\\left(\\frac{y -\n\\xi}{\\omega\\sqrt{2}}\\right)\\right)\\right) . \\end{equation*}\\]\n\n\n\ny ~ skew_normal(xi, omega, alpha)\nIncrement target log probability density with skew_normal_lupdf(y | xi, omega, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal skew_normal_lpdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lupdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal skew_normal_cdf(reals y | reals xi, reals omega, reals alpha) The skew normal distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lcdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nreal skew_normal_lccdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nR skew_normal_rng(reals xi, reals omega, real alpha) Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}\\), and \\(\\sigma \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{StudentT}(y|\\nu,\\mu,\\sigma) = \\frac{\\Gamma\\left((\\nu +\n1)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\nu \\pi} \\ \\sigma}\n\\ \\left( 1 + \\frac{1}{\\nu} \\left(\\frac{y - \\mu}{\\sigma}\\right)^2\n\\right)^{-(\\nu + 1)/2} \\! . \\end{equation*}\\]\n\n\n\ny ~ student_t(nu, mu, sigma)\nIncrement target log probability density with student_t_lupdf(y | nu, mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal student_t_lpdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lupdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal student_t_cdf(reals y | reals nu, reals mu, reals sigma) The Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.0\n \n\nreal student_t_lcdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lccdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nR student_t_rng(reals nu, reals mu, reals sigma) Generate a Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Cauchy}(y|\\mu,\\sigma) = \\frac{1}{\\pi \\sigma} \\ \\frac{1}{1 + \\left((y - \\mu)/\\sigma\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ cauchy(mu, sigma)\nIncrement target log probability density with cauchy_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal cauchy_lpdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lupdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal cauchy_cdf(reals y | reals mu, reals sigma) The Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal cauchy_lcdf(reals y | reals mu, reals sigma) The log of the Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lccdf(reals y | reals mu, reals sigma) The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR cauchy_rng(reals mu, reals sigma) Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{DoubleExponential}(y|\\mu,\\sigma) =\n\\frac{1}{2\\sigma}   \\exp \\left( - \\, \\frac{|y - \\mu|}{\\sigma} \\right)\n. \\end{equation*}\\] Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section exponential distribution), which is parameterized in terms of inverse scale.\nThe double-exponential distribution can be defined as a compound exponential-normal distribution (Ding and Blitzstein 2018). Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write \\[\\begin{equation*} \\alpha \\sim\n\\mathsf{Exponential}\\left( \\frac{1}{2 \\sigma^2} \\right) \\end{equation*}\\] and \\[\\begin{equation*} \\beta \\mid\n\\alpha \\sim \\mathsf{Normal}(\\mu, \\sqrt{\\alpha}), \\end{equation*}\\] then \\[\\begin{equation*} \\beta \\sim\n\\mathsf{DoubleExponential}(\\mu, \\sigma ). \\end{equation*}\\] This may be used to code a non-centered parameterization by taking \\[\\begin{equation*} \\beta^{\\text{raw}} \\sim\n\\mathsf{Normal}(0, 1) \\end{equation*}\\] and defining \\[\\begin{equation*} \\beta = \\mu + \\sqrt{\\alpha} \\,\n\\beta^{\\text{raw}}. \\end{equation*}\\]\n\n\n\ny ~ double_exponential(mu, sigma)\nIncrement target log probability density with double_exponential_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal double_exponential_lpdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lupdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal double_exponential_cdf(reals y | reals mu, reals sigma) The double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal double_exponential_lcdf(reals y | reals mu, reals sigma) The log of the double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lccdf(reals y | reals mu, reals sigma) The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR double_exponential_rng(reals mu, reals sigma) Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Logistic}(y|\\mu,\\sigma) = \\frac{1}{\\sigma} \\ \\exp\\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right)\n\\ \\left(1 + \\exp \\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right) \\right)^{\\!-2} \\! .\n\\end{equation*}\\]\n\n\n\ny ~ logistic(mu, sigma)\nIncrement target log probability density with logistic_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal logistic_lpdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lupdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal logistic_cdf(reals y | reals mu, reals sigma) The logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal logistic_lcdf(reals y | reals mu, reals sigma) The log of the logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lccdf(reals y | reals mu, reals sigma) The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR logistic_rng(reals mu, reals sigma) Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Gumbel}(y|\\mu,\\beta) = \\frac{1}{\\beta} \\ \\exp\\left(-\\frac{y-\\mu}{\\beta}-\\exp\\left(-\\frac{y-\\mu}{\\beta}\\right)\\right) .\n\\end{equation*}\\]\n\n\n\ny ~ gumbel(mu, beta)\nIncrement target log probability density with gumbel_lupdf(y | mu, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gumbel_lpdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lupdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gumbel_cdf(reals y | reals mu, reals beta) The gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.0\n \n\nreal gumbel_lcdf(reals y | reals mu, reals beta) The log of the gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lccdf(reals y | reals mu, reals beta) The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nR gumbel_rng(reals mu, reals beta) Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\) and \\(\\tau \\in [0, 1]\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{aligned}\n& \\text{SkewDoubleExponential}  (y|\\mu,\\sigma, \\tau) = \\\\\n& \\qquad \\qquad \\frac{2 \\tau (1 - \\tau) }{\\sigma} \\exp \\left[ - \\frac{2}{\\sigma} \\left[ \\left(1 - \\tau \\right) I(y &lt; \\mu)  (\\mu - y) + \\tau I(y &gt; \\mu)(y-\\mu)  \\right]    \\right]\n\\end{aligned}\\]\n\n\n\ny ~ skew_double_exponential(mu, sigma, tau)\nIncrement target log probability density with skew_double_exponential(y | mu, sigma, tau)\nAvailable since 2.28\n \n\n\n\n\n \n\nreal skew_double_exponential_lpdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lupdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms\nAvailable since 2.28\n \n\nreal skew_double_exponential_cdf(reals y | reals mu, reals sigma, reals tau) The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lcdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lccdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nR skew_double_exponential_rng(reals mu, reals sigma, reals tau) Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#normal-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#normal-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Normal}(y|\\mu,\\sigma) =\n\\frac{1}{\\sqrt{2 \\pi} \\ \\sigma} \\exp\\left( - \\, \\frac{1}{2} \\left(\\frac{y - \\mu}{\\sigma} \\right)^2 \\right) \\!.\n\\end{equation*}\\]\n\n\n\ny ~ normal(mu, sigma)\nIncrement target log probability density with normal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal normal_lpdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal normal_lupdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_cdf(reals y | reals mu, reals sigma) The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 1 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; the function Phi_approx is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.0\n \n\nreal normal_lcdf(reals y | reals mu, reals sigma) The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.12\n \n\nreal normal_lccdf(reals y | reals mu, reals sigma) The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.15\n \n\nR normal_rng(reals mu, reals sigma) Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nThe standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. \\[\\begin{equation*}\n\\text{StdNormal}(y) \\ = \\ \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{1}{\\sqrt{2 \\pi}} \\, \\exp \\left( \\frac{-y^2}{2} \\right)\\!.\n\\end{equation*}\\] Up to a proportion on the log scale, where Stan computes, \\[\\begin{equation*}\n\\log \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{-y^2}{2} + \\text{const}.\n\\end{equation*}\\] With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location \\(0\\) and scale \\(1\\).\n\n\n\ny ~ std_normal()\nIncrement target log probability density with std_normal_lupdf(y).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal std_normal_lpdf(reals y) The standard normal (location zero, scale one) log probability density of y.\nAvailable since 2.18\n \n\nreal std_normal_lupdf(reals y) The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.\nAvailable since 2.25\n \n\nreal std_normal_cdf(reals y) The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for \\(y\\) below -37.5 and overflow to 1 for \\(y\\) above 8.25; the function Phi_approx is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lcdf(reals y) The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to \\(-\\infty\\) for \\(y\\) below -37.5 and overflow to 0 for \\(y\\) above 8.25; log(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lccdf(reals y) The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for \\(y\\) below -37.5 and underflow to \\(-\\infty\\) for \\(y\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nR std_normal_qf(T x) Returns the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The std_normal_qf is equivalent to the inv_Phi function.\nAvailable since 2.31\n \n\nR std_normal_log_qf(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) evaluated at the log of the specified quantile x. This function is equivalent to std_normal_qf(exp(x)) but is more numerically stable.\nAvailable since 2.31\n \n\nreal std_normal_rng() Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.21",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#normal-id-glm",
+    "href": "functions-reference/unbounded_continuous_distributions.html#normal-id-glm",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "Stan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e. a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\sigma\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^n\\), \\[\\begin{equation*} \\text{NormalIdGLM}(y|x, \\alpha, \\beta, \\sigma) = \\prod_{1\\leq i\n\\leq n}\\text{Normal}(y_i|\\alpha_i + x_i\\cdot \\beta, \\sigma). \\end{equation*}\\]\n\n\n\ny ~ normal_id_glm(x, alpha, beta, sigma)\nIncrement target log probability density with normal_id_glm_lupdf(y | x, alpha, beta, sigma).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#exponentially-modified-normal-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#exponentially-modified-normal-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "Exponentially modified Gaussian describes the distribution of \\(Z = X + Y\\) when \\(X\\) and \\(Y\\) are independent and \\(X\\) is normally distributed (with mean \\(\\mu\\) and standard deviation \\(\\sigma\\)) and \\(Y\\) is exponentially distributed (with rate \\(\\lambda\\)).\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\), and \\(\\lambda \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{ExpModNormal}(y|\\mu,\\sigma,\\lambda) = \\frac{\\lambda}{2} \\ \\exp\n\\left(\\frac{\\lambda}{2} \\left(2\\mu + \\lambda \\sigma^2 -\n2y\\right)\\right) \\text{erfc}\\left(\\frac{\\mu + \\lambda\\sigma^2 -\ny}{\\sqrt{2}\\sigma}\\right) . \\end{equation*}\\]\n\n\n\ny ~ exp_mod_normal(mu, sigma, lambda)\nIncrement target log probability density with exp_mod_normal_lupdf(y | mu, sigma, lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exp_mod_normal_lpdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lupdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal exp_mod_normal_cdf(reals y | reals mu, reals sigma, reals lambda) The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.0\n \n\nreal exp_mod_normal_lcdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lccdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nR exp_mod_normal_rng(reals mu, reals sigma, reals lambda) Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#skew-normal-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#skew-normal-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\xi \\in \\mathbb{R}\\), \\(\\omega \\in \\mathbb{R}^+\\), and \\(\\alpha \\in\n\\mathbb{R}\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*} \\text{SkewNormal}(y \\mid\n\\xi, \\omega, \\alpha) = \\frac{1}{\\omega\\sqrt{2\\pi}} \\ \\exp\\left( - \\,\n\\frac{1}{2}            \\left(  \\frac{y - \\xi}{\\omega} \\right)^2\n\\right) \\ \\left(1 + \\text{erf}\\left( \\alpha\\left(\\frac{y -\n\\xi}{\\omega\\sqrt{2}}\\right)\\right)\\right) . \\end{equation*}\\]\n\n\n\ny ~ skew_normal(xi, omega, alpha)\nIncrement target log probability density with skew_normal_lupdf(y | xi, omega, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal skew_normal_lpdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lupdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal skew_normal_cdf(reals y | reals xi, reals omega, reals alpha) The skew normal distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lcdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nreal skew_normal_lccdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nR skew_normal_rng(reals xi, reals omega, real alpha) Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#student-t-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#student-t-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}\\), and \\(\\sigma \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{StudentT}(y|\\nu,\\mu,\\sigma) = \\frac{\\Gamma\\left((\\nu +\n1)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\nu \\pi} \\ \\sigma}\n\\ \\left( 1 + \\frac{1}{\\nu} \\left(\\frac{y - \\mu}{\\sigma}\\right)^2\n\\right)^{-(\\nu + 1)/2} \\! . \\end{equation*}\\]\n\n\n\ny ~ student_t(nu, mu, sigma)\nIncrement target log probability density with student_t_lupdf(y | nu, mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal student_t_lpdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lupdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal student_t_cdf(reals y | reals nu, reals mu, reals sigma) The Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.0\n \n\nreal student_t_lcdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lccdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nR student_t_rng(reals nu, reals mu, reals sigma) Generate a Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#cauchy-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#cauchy-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Cauchy}(y|\\mu,\\sigma) = \\frac{1}{\\pi \\sigma} \\ \\frac{1}{1 + \\left((y - \\mu)/\\sigma\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ cauchy(mu, sigma)\nIncrement target log probability density with cauchy_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal cauchy_lpdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lupdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal cauchy_cdf(reals y | reals mu, reals sigma) The Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal cauchy_lcdf(reals y | reals mu, reals sigma) The log of the Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lccdf(reals y | reals mu, reals sigma) The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR cauchy_rng(reals mu, reals sigma) Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#double-exponential-laplace-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#double-exponential-laplace-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{DoubleExponential}(y|\\mu,\\sigma) =\n\\frac{1}{2\\sigma}   \\exp \\left( - \\, \\frac{|y - \\mu|}{\\sigma} \\right)\n. \\end{equation*}\\] Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section exponential distribution), which is parameterized in terms of inverse scale.\nThe double-exponential distribution can be defined as a compound exponential-normal distribution (Ding and Blitzstein 2018). Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write \\[\\begin{equation*} \\alpha \\sim\n\\mathsf{Exponential}\\left( \\frac{1}{2 \\sigma^2} \\right) \\end{equation*}\\] and \\[\\begin{equation*} \\beta \\mid\n\\alpha \\sim \\mathsf{Normal}(\\mu, \\sqrt{\\alpha}), \\end{equation*}\\] then \\[\\begin{equation*} \\beta \\sim\n\\mathsf{DoubleExponential}(\\mu, \\sigma ). \\end{equation*}\\] This may be used to code a non-centered parameterization by taking \\[\\begin{equation*} \\beta^{\\text{raw}} \\sim\n\\mathsf{Normal}(0, 1) \\end{equation*}\\] and defining \\[\\begin{equation*} \\beta = \\mu + \\sqrt{\\alpha} \\,\n\\beta^{\\text{raw}}. \\end{equation*}\\]\n\n\n\ny ~ double_exponential(mu, sigma)\nIncrement target log probability density with double_exponential_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal double_exponential_lpdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lupdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal double_exponential_cdf(reals y | reals mu, reals sigma) The double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal double_exponential_lcdf(reals y | reals mu, reals sigma) The log of the double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lccdf(reals y | reals mu, reals sigma) The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR double_exponential_rng(reals mu, reals sigma) Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#logistic-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#logistic-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Logistic}(y|\\mu,\\sigma) = \\frac{1}{\\sigma} \\ \\exp\\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right)\n\\ \\left(1 + \\exp \\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right) \\right)^{\\!-2} \\! .\n\\end{equation*}\\]\n\n\n\ny ~ logistic(mu, sigma)\nIncrement target log probability density with logistic_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal logistic_lpdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lupdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal logistic_cdf(reals y | reals mu, reals sigma) The logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal logistic_lcdf(reals y | reals mu, reals sigma) The log of the logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lccdf(reals y | reals mu, reals sigma) The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR logistic_rng(reals mu, reals sigma) Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#gumbel-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#gumbel-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Gumbel}(y|\\mu,\\beta) = \\frac{1}{\\beta} \\ \\exp\\left(-\\frac{y-\\mu}{\\beta}-\\exp\\left(-\\frac{y-\\mu}{\\beta}\\right)\\right) .\n\\end{equation*}\\]\n\n\n\ny ~ gumbel(mu, beta)\nIncrement target log probability density with gumbel_lupdf(y | mu, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gumbel_lpdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lupdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gumbel_cdf(reals y | reals mu, reals beta) The gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.0\n \n\nreal gumbel_lcdf(reals y | reals mu, reals beta) The log of the gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lccdf(reals y | reals mu, reals beta) The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nR gumbel_rng(reals mu, reals beta) Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#skew-double-exponential-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#skew-double-exponential-distribution",
+    "title": "Unbounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\) and \\(\\tau \\in [0, 1]\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{aligned}\n& \\text{SkewDoubleExponential}  (y|\\mu,\\sigma, \\tau) = \\\\\n& \\qquad \\qquad \\frac{2 \\tau (1 - \\tau) }{\\sigma} \\exp \\left[ - \\frac{2}{\\sigma} \\left[ \\left(1 - \\tau \\right) I(y &lt; \\mu)  (\\mu - y) + \\tau I(y &gt; \\mu)(y-\\mu)  \\right]    \\right]\n\\end{aligned}\\]\n\n\n\ny ~ skew_double_exponential(mu, sigma, tau)\nIncrement target log probability density with skew_double_exponential(y | mu, sigma, tau)\nAvailable since 2.28\n \n\n\n\n\n \n\nreal skew_double_exponential_lpdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lupdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms\nAvailable since 2.28\n \n\nreal skew_double_exponential_cdf(reals y | reals mu, reals sigma, reals tau) The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lcdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lccdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nR skew_double_exponential_rng(reals mu, reals sigma, reals tau) Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/sparse_matrix_operations.html",
+    "href": "functions-reference/sparse_matrix_operations.html",
+    "title": "Sparse Matrix Operations",
+    "section": "",
+    "text": "For sparse matrices, for which many elements are zero, it is more efficient to use specialized representations to save memory and speed up matrix arithmetic (including derivative calculations). Given Stan’s implementation, there is substantial space (memory) savings by using sparse matrices. Because of the ease of optimizing dense matrix operations, speed improvements only arise at 90% or even greater sparsity; below that level, dense matrices are faster but use more memory.\nBecause of this speedup and space savings, it may even be useful to read in a dense matrix and convert it to a sparse matrix before multiplying it by a vector. This chapter covers a very specific form of sparsity consisting of a sparse matrix multiplied by a dense vector.\n\n\nSparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix \\[\\begin{equation*}\nA = \\begin{bmatrix} 19 & 27 & 0 & 0 \\\\ 0 & 0 & 0 & 0 \\\\ 0 & 0 & 0 & 52 \\\\ 81 & 0 & 95 & 33 \\end{bmatrix}\n\\end{equation*}\\] is translated into a vector of the non-zero real values, read by row from the matrix \\(A\\), \\[\\begin{equation*}\nw(A) = \\begin{bmatrix} 19 & 27 & 52 & 81 & 95 & 33 \\end{bmatrix}^{\\top} \\! \\! \\! ,\n\\end{equation*}\\] an array of integer column indices for the values, \\[\\begin{equation*}\nv(A) = \\begin{bmatrix} 1 & 2 & 4 & 1 & 3 & 4 \\end{bmatrix} \\! ,\n\\end{equation*}\\] and an array of integer indices indicating where in \\(w(A)\\) a given row’s values start, \\[\\begin{equation*} u(A) = \\begin{bmatrix} 1 & 3 & 3 & 4 & 7 \\end{bmatrix} \\! ,\n\\end{equation*}\\] with a padded value at the end to guarantee that \\[\\begin{equation*} u(A)[n+1] - u(A)[n] \\end{equation*}\\] is the number of non-zero elements in row \\(n\\) of the matrix (here \\(2\\), \\(0\\), \\(1\\), and \\(3\\)). Note that because the second row has no non-zero elements both the second and third elements of \\(u(A)\\) correspond to the third element of \\(w(A)\\), which is \\(52\\). The values \\((w(A), \\, v(A), \\, u(A))\\) are sufficient to reconstruct \\(A\\).\nThe values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly \\(12 K + M\\) bytes plus a small constant overhead, which is often considerably fewer bytes than the \\(M \\times\nN\\) required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.\n\n\n\nConversion functions between dense and sparse matrices are provided.\n\n\nConverting a dense matrix \\(m\\) to a sparse representation produces a vector \\(w\\) and two integer arrays, \\(u\\) and \\(v\\).\n \n\nvector csr_extract_w(matrix a) Return non-zero values in matrix a; see section compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_v(matrix a) Return column indices for values in csr_extract_w(a); see compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_u(matrix a) Return array of row starting indices for entries in csr_extract_w(a) followed by the size of csr_extract_w(a) plus one; see section compressed row storage.\nAvailable since 2.8\n \n\ntuple(vector, array[] int, array[] int) csr_extract(matrix a) Return all three components of the CSR representation of the matrix a; see section compressed row storage. This function is equivalent to (csr_extract_w(a), csr_extract_v(a), csr_extract_u(a)).\nAvailable since 2.33\n\n\n\nTo convert a sparse matrix representation to a dense matrix, there is a single function.\n \n\nmatrix csr_to_dense_matrix(int m, int n, vector w, array[] int v, array[] int u) Return dense \\(\\text{m} \\times \\text{n}\\) matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section compressed row storage for more details.\nAvailable since 2.10\n\n\n\n\n\n\nThe only supported operation is the multiplication of a sparse matrix \\(A\\) and a dense vector \\(b\\) to produce a dense vector \\(A\\,b\\). Multiplying a dense row vector \\(b\\) and a sparse matrix \\(A\\) can be coded using transposition as \\[\\begin{equation*}\nb \\, A = (A^{\\top} \\, b^{\\top})^{\\top},\n\\end{equation*}\\] but care must be taken to represent \\(A^{\\top}\\) rather than \\(A\\) as a sparse matrix.\n \n\nvector csr_matrix_times_vector(int m, int n, vector w, array[] int v, array[] int u, vector b) Multiply the \\(\\text{m} \\times \\text{n}\\) matrix represented by values w, column indices v, and row start indices u by the vector b; see compressed row storage.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Sparse Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/sparse_matrix_operations.html#CSR",
+    "href": "functions-reference/sparse_matrix_operations.html#CSR",
+    "title": "Sparse Matrix Operations",
+    "section": "",
+    "text": "Sparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix \\[\\begin{equation*}\nA = \\begin{bmatrix} 19 & 27 & 0 & 0 \\\\ 0 & 0 & 0 & 0 \\\\ 0 & 0 & 0 & 52 \\\\ 81 & 0 & 95 & 33 \\end{bmatrix}\n\\end{equation*}\\] is translated into a vector of the non-zero real values, read by row from the matrix \\(A\\), \\[\\begin{equation*}\nw(A) = \\begin{bmatrix} 19 & 27 & 52 & 81 & 95 & 33 \\end{bmatrix}^{\\top} \\! \\! \\! ,\n\\end{equation*}\\] an array of integer column indices for the values, \\[\\begin{equation*}\nv(A) = \\begin{bmatrix} 1 & 2 & 4 & 1 & 3 & 4 \\end{bmatrix} \\! ,\n\\end{equation*}\\] and an array of integer indices indicating where in \\(w(A)\\) a given row’s values start, \\[\\begin{equation*} u(A) = \\begin{bmatrix} 1 & 3 & 3 & 4 & 7 \\end{bmatrix} \\! ,\n\\end{equation*}\\] with a padded value at the end to guarantee that \\[\\begin{equation*} u(A)[n+1] - u(A)[n] \\end{equation*}\\] is the number of non-zero elements in row \\(n\\) of the matrix (here \\(2\\), \\(0\\), \\(1\\), and \\(3\\)). Note that because the second row has no non-zero elements both the second and third elements of \\(u(A)\\) correspond to the third element of \\(w(A)\\), which is \\(52\\). The values \\((w(A), \\, v(A), \\, u(A))\\) are sufficient to reconstruct \\(A\\).\nThe values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly \\(12 K + M\\) bytes plus a small constant overhead, which is often considerably fewer bytes than the \\(M \\times\nN\\) required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Sparse Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/sparse_matrix_operations.html#conversion-functions",
+    "href": "functions-reference/sparse_matrix_operations.html#conversion-functions",
+    "title": "Sparse Matrix Operations",
+    "section": "",
+    "text": "Conversion functions between dense and sparse matrices are provided.\n\n\nConverting a dense matrix \\(m\\) to a sparse representation produces a vector \\(w\\) and two integer arrays, \\(u\\) and \\(v\\).\n \n\nvector csr_extract_w(matrix a) Return non-zero values in matrix a; see section compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_v(matrix a) Return column indices for values in csr_extract_w(a); see compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_u(matrix a) Return array of row starting indices for entries in csr_extract_w(a) followed by the size of csr_extract_w(a) plus one; see section compressed row storage.\nAvailable since 2.8\n \n\ntuple(vector, array[] int, array[] int) csr_extract(matrix a) Return all three components of the CSR representation of the matrix a; see section compressed row storage. This function is equivalent to (csr_extract_w(a), csr_extract_v(a), csr_extract_u(a)).\nAvailable since 2.33\n\n\n\nTo convert a sparse matrix representation to a dense matrix, there is a single function.\n \n\nmatrix csr_to_dense_matrix(int m, int n, vector w, array[] int v, array[] int u) Return dense \\(\\text{m} \\times \\text{n}\\) matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section compressed row storage for more details.\nAvailable since 2.10",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Sparse Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/sparse_matrix_operations.html#sparse-matrix-arithmetic",
+    "href": "functions-reference/sparse_matrix_operations.html#sparse-matrix-arithmetic",
+    "title": "Sparse Matrix Operations",
+    "section": "",
+    "text": "The only supported operation is the multiplication of a sparse matrix \\(A\\) and a dense vector \\(b\\) to produce a dense vector \\(A\\,b\\). Multiplying a dense row vector \\(b\\) and a sparse matrix \\(A\\) can be coded using transposition as \\[\\begin{equation*}\nb \\, A = (A^{\\top} \\, b^{\\top})^{\\top},\n\\end{equation*}\\] but care must be taken to represent \\(A^{\\top}\\) rather than \\(A\\) as a sparse matrix.\n \n\nvector csr_matrix_times_vector(int m, int n, vector w, array[] int v, array[] int u, vector b) Multiply the \\(\\text{m} \\times \\text{n}\\) matrix represented by values w, column indices v, and row start indices u by the vector b; see compressed row storage.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Sparse Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/removed_functions.html",
+    "href": "functions-reference/removed_functions.html",
+    "title": "Removed Functions",
+    "section": "",
+    "text": "Functions which once existed in the Stan language and have since been replaced or removed are listed here.\n\n\nRemoved: Currently two non-conforming functions ending in suffix _log.\nReplacement: Replace multiply_log(...) with lmultiply(...). Replace binomial_coefficient_log(...) with lchoose(...).\nRemoved In: Stan 2.33\n\n\n\nRemoved: The built-in no-argument function get_lp() is deprecated.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33\n\n\n\nRemoved: The unary function fabs is deprecated.\nReplacement: Use the unary function abs instead. Note that the return type for abs is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.\nRemoved In: Stan 2.33\n\n\n\nThese covariance functions have been replaced by those described in Gaussian Process Covariance Functions\nWith magnitude \\(\\alpha\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(x_i, x_j) = \\alpha^2 \\exp \\left(-\\dfrac{1}{2\\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \\right)\n\\]\n \n\nmatrix cov_exp_quad(row_vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(row_vectors x1, row_vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x1, vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x1, array[] real x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n\n\n\nRemoved: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.\nReplacement: Explicit x != 0 comparison is preferred instead.\nRemoved In: Stan 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Removed Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/removed_functions.html#multiply_log-and-binomial_coefficient_log-functions",
+    "href": "functions-reference/removed_functions.html#multiply_log-and-binomial_coefficient_log-functions",
+    "title": "Removed Functions",
+    "section": "",
+    "text": "Removed: Currently two non-conforming functions ending in suffix _log.\nReplacement: Replace multiply_log(...) with lmultiply(...). Replace binomial_coefficient_log(...) with lchoose(...).\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Removed Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/removed_functions.html#get_lp-function",
+    "href": "functions-reference/removed_functions.html#get_lp-function",
+    "title": "Removed Functions",
+    "section": "",
+    "text": "Removed: The built-in no-argument function get_lp() is deprecated.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Removed Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/removed_functions.html#fabs-function",
+    "href": "functions-reference/removed_functions.html#fabs-function",
+    "title": "Removed Functions",
+    "section": "",
+    "text": "Removed: The unary function fabs is deprecated.\nReplacement: Use the unary function abs instead. Note that the return type for abs is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.\nRemoved In: Stan 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Removed Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/removed_functions.html#cov_exp_quad",
+    "href": "functions-reference/removed_functions.html#cov_exp_quad",
+    "title": "Removed Functions",
+    "section": "",
+    "text": "These covariance functions have been replaced by those described in Gaussian Process Covariance Functions\nWith magnitude \\(\\alpha\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(x_i, x_j) = \\alpha^2 \\exp \\left(-\\dfrac{1}{2\\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \\right)\n\\]\n \n\nmatrix cov_exp_quad(row_vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(row_vectors x1, row_vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x1, vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x1, array[] real x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Removed Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/removed_functions.html#real-arguments-to-logical-operators-operator-operator-and-operator",
+    "href": "functions-reference/removed_functions.html#real-arguments-to-logical-operators-operator-operator-and-operator",
+    "title": "Removed Functions",
+    "section": "",
+    "text": "Removed: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.\nReplacement: Explicit x != 0 comparison is preferred instead.\nRemoved In: Stan 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Removed Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html",
+    "href": "functions-reference/real-valued_basic_functions.html",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "This chapter describes built-in functions that take zero or more real or integer arguments and return real values.\n\n\nAlthough listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.\n\n\nMany of Stan’s unary functions can be applied to any argument type. For example, the exponential function, exp, can be applied to real arguments or arrays of real arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.\n\n\nWhen applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like exp() are defined elementwise. For example,\n // declare some variables for arguments\n real x0;\n array[5] real x1;\n array[4, 7] real x2;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = exp(x0);\n y1 = exp(x1);\n y2 = exp(x2);\nWhen exp is applied to an array, it applies elementwise. For example, the statement above,\n y2 = exp(x2);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = exp(x2[i, j]);\n  }\n}\n\n\n\nVectorized functions also apply elementwise to vectors and matrices. For example,\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = exp(xv);\n yrv = exp(xrv);\n ym = exp(xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = exp(u);\nAfter this has been executed, z[i, j, k] will be equal to exp(u[i, j, k]).\n\n\n\nInteger arguments are promoted to real values in vectorized unary functions. Thus if n is of type int, exp(n) is of type real. Arrays work the same way, so that if n2 is a one dimensional array of integers, then exp(n2) will be a one-dimensional array of reals with the same number of elements as n2. For example,\n array[23] int n1;\n array[23] real z1;\n z1 = exp(n1);\nIt would be illegal to try to assign exp(n1) to an array of integers; the return type is a real array.\n\n\n\n\nLike the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.\n\n\nWhen applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like pow() are defined elementwise. For example,\n // declare some variables for arguments\n real x00;\n real x01;\n array[5] real x10;\n array[5]real x11;\n array[4, 7] real x20;\n array[4, 7] real x21;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = pow(x00, x01);\n y1 = pow(x10, x11);\n y2 = pow(x20, x21);\nWhen pow is applied to two arrays, it applies elementwise. For example, the statement above,\n y2 = pow(x20, x21);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x21[i, j]);\n  }\n}\nAlternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:\ny2 = pow(x20, x00);\nproduces the same result for y2 as the explicit loop:\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x00);\n  }\n}\n\n\n\nVectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,\n real x00;\n vector[5] xv00;\n vector[5] xv01;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = pow(xv00, xv01);\n yrv = pow(xrv, x00);\n ym = pow(x00, xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = pow(u, x00);\nAfter this has been executed, z[i, j, k] will be equal to pow(u[i, j, k], x00).\n\n\n\nVectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv = pow(xv, xv)\n row_vector[7] yrv = pow(xrv, xrv)\n matrix[10, 20] = pow(xm, xm)\nBut the following statements are not:\n vector[5] xv;\n vector[7] xv2;\n row_vector[5] xrv;\n\n // Cannot mix different types\n vector[5] yv = pow(xv, xrv)\n\n // Cannot mix different sizes of the same type\n vector[5] yv = pow(xv, xv2)\nWhile the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., bessel_first_kind). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:\n vector[5] xv;\n matrix[5, 5] xm;\n array[5] int xi;\n array[5, 5] int xii;\n\n vector[5] yv = bessel_first_kind(xi, xv);\n matrix[5, 5] ym = bessel_first_kind(xii, xm);\nWhereas these are not:\n vector[5] xv;\n matrix[5, 5] xm;\n array[7] int xi;\n\n // Dimensions of containers do not match\n vector[5] yv = bessel_first_kind(xi, xv);\n\n // Function requires first argument be an integer type\n matrix[5, 5] ym = bessel_first_kind(xm, xm);\n\n\n\n\n\nConstants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant \\(\\pi\\) must be written in a Stan program as pi().\n \n\nreal pi() \\(\\pi\\), the ratio of a circle’s circumference to its diameter\nAvailable since 2.0\n \n\nreal e() \\(e\\), the base of the natural logarithm\nAvailable since 2.0\n \n\nreal sqrt2() The square root of 2\nAvailable since 2.0\n \n\nreal log2() The natural logarithm of 2\nAvailable since 2.0\n \n\nreal log10() The natural logarithm of 10\nAvailable since 2.0\n\n\n\n \n\nreal not_a_number() Not-a-number, a special non-finite real value returned to signal an error\nAvailable since 2.0\n \n\nreal positive_infinity() Positive infinity, a special non-finite real value larger than all finite numbers\nAvailable since 2.0\n \n\nreal negative_infinity() Negative infinity, a special non-finite real value smaller than all finite numbers\nAvailable since 2.0\n \n\nreal machine_precision() The smallest number \\(x\\) such that \\((x + 1) \\neq 1\\) in floating-point arithmetic on the current hardware platform\nAvailable since 2.0\n\n\n\nThe basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (~) and log probability increment statement (increment_log_prob) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.\nStan provides a special built-in function target() that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.\n \n\nreal target() Return the current value of the log probability accumulator.\nAvailable since 2.10\ntarget acts like a function ending in _lp, meaning that it may only may only be used in the model block.\n\n\n\nLike C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.\n\n\nAll comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.\n \n\nint operator&lt;(int x, int y)\n \n\nint operator&lt;(real x, real y) Return 1 if x is less than y and 0 otherwise. \\[\\begin{equation*} \\text{operator&lt;}(x,y)\n= \\begin{cases} 1 & \\text{if $x &lt; y$} \\\\ 0 & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&lt;=(int x, int y)\n \n\nint operator&lt;=(real x, real y) Return 1 if x is less than or equal y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&lt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\leq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;(int x, int y)\n \n\nint operator&gt;(real x, real y) Return 1 if x is greater than y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;}(x,y) = \\begin{cases} 1 & \\text{if $x &gt; y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;=(int x, int y)\n \n\nint operator&gt;=(real x, real y) Return 1 if x is greater than or equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\geq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator==(int x, int y)\n \n\nint operator==(real x, real y) Return 1 if x is equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator==}(x,y) = \\begin{cases} 1 & \\text{if $x = y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!=(int x, int y)\n \n\nint operator!=(real x, real y) Return 1 if x is not equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator!=}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq y$} \\\\ 0 &\n\\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\nBoolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as !a && b is interpreted as (!a) && b, and a &lt; b || c &gt;= d && e != f as (a &lt; b) || (((c &gt;= d) && (e != f))).\n \n\nint operator!(int x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0$} \\\\ 1 & \\text{if $x = 0$}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!(real x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0.0$} \\\\ 1 & \\text{if $x = 0.0$}\n\\end{cases} \\end{equation*}\\] deprecated; - use operator== instead.\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator&&(int x, int y)\nReturn 1 if x is unequal to 0 and y is unequal to 0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\text{ and } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&&(real x, real y) Return 1 if x is unequal to 0.0 and y is unequal to 0.0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\text{ and } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator||(int x, int y) Return 1 if x is unequal to 0 or y is unequal to 0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\textrm{ or } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator||(real x, real y) Return 1 if x is unequal to 0.0 or y is unequal to 0.0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\textrm{ or } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n\n\nLike in C++, the boolean operators && and || and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating a || b, if a evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression b. Similarly, evaluating a && b first evaluates a, and if the result is zero, returns 0 without evaluating b.\n\n\n\n\nThe logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.\n \n\nreal step(real x) Return 1 if x is positive and 0 otherwise. \\[\\begin{equation*} \\text{step}(x) =\n\\begin{cases} 0 & \\text{if } x &lt; 0 \\\\ 1 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nThe step function is often used in BUGS to perform conditional operations. For instance, step(a-b) evaluates to 1 if a is greater than b and evaluates to 0 otherwise. step is a step-like functions; see the warning in section step functions applied to expressions dependent on parameters.\nAvailable since 2.0\n \n\nint is_inf(real x) Return 1 if x is infinite (positive or negative) and 0 otherwise.\nAvailable since 2.5\n \n\nint is_nan(real x) Return 1 if x is NaN and 0 otherwise.\nAvailable since 2.5\nCare must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section step-like functions for details.\n\n\n\n\nThe arithmetic operators are presented using C++ notation. For instance operator+(x,y) refers to the binary addition operator and operator-(x) to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as x + y and -x, respectively.\n\n\n \n\nreal operator+(real x, real y) Return the sum of x and y. \\[\\begin{equation*} (x + y) = \\text{operator+}(x,y) = x+y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator-(real x, real y) Return the difference between x and y. \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x,y) = x - y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator*(real x, real y) Return the product of x and y. \\[\\begin{equation*} (x * y) = \\text{operator*}(x,y) = xy\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator/(real x, real y) Return the quotient of x and y. \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator^(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} (x^\\mathrm{\\wedge}y) =\n\\text{operator}^\\mathrm{\\wedge}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.5\n\n\n\n \n\nreal operator-(real x) Return the negation of the subtrahend x. \\[\\begin{equation*} \\text{operator-}(x) = (-x)\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of reals, -x is the same shape array where each individual number is negated.\nAvailable since 2.31\n \n\nreal operator+(real x) Return the value of x. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n\nWarning: These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is real vector, row_vector, matrix, or an array of those types, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nreal fdim(real x, real y) Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. \\[\\begin{equation*} \\text{fdim}(x,y) = \\begin{cases} x-y &\n\\text{if } x \\geq y \\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fdim(T1 x, T2 y) Vectorized implementation of the fdim function\nAvailable since 2.25\n\n\n\n \n\nreal fmin(real x, real y) Return the minimum of x and y; see warning above. \\[\\begin{equation*} \\text{fmin}(x,y) = \\begin{cases} x &\n\\text{if } x \\leq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmin(T1 x, T2 y) Vectorized implementation of the fmin function\nAvailable since 2.25\n \n\nreal fmax(real x, real y) Return the maximum of x and y; see warning above. \\[\\begin{equation*} \\text{fmax}(x,y) = \\begin{cases} x &\n\\text{if } x \\geq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmax(T1 x, T2 y) Vectorized implementation of the fmax function\nAvailable since 2.25\n\n\n\n \n\nreal fmod(real x, real y) Return the real value remainder after dividing x by y; see warning above. \\[\\begin{equation*} \\text{fmod}(x,y) = x - \\left\\lfloor \\frac{x}{y} \\right\\rfloor \\, y \\end{equation*}\\] The operator \\(\\lfloor u \\rfloor\\) is the floor operation; see below.\nAvailable since 2.0\n \n\nR fmod(T1 x, T2 y) Vectorized implementation of the fmod function\nAvailable since 2.25\n\n\n\nWarning: Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.\nThe rounding functions cannot be used as indices to arrays because they return real values. For operations over data or in the generated quantities block, the to_int() function can be used.\n \n\nR floor(T x) The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR ceil(T x) The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR round(T x) The nearest integer to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR trunc(T x) The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n\n\n\n\n \n\nR sqrt(T x) The square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR cbrt(T x) The cube root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR square(T x) The square of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp(T x) The natural exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp2(T x) The base-2 exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log(T x) The natural logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log2(T x) The base-2 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log10(T x) The base-10 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal pow(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} \\text{pow}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.0\n \n\nR pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.25\n \n\nR inv(T x) The inverse of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_sqrt(T x) The inverse of the square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_square(T x) The inverse of the square of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal hypot(real x, real y) Return the length of the hypotenuse of a right triangle with sides of length x and y. \\[\\begin{equation*} \\text{hypot}(x,y) = \\begin{cases} \\sqrt{x^2+y^2} &\n\\text{if } x,y\\geq 0 \\\\ \\textrm{NaN} & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR hypot(T1 x, T2 y) Vectorized implementation of the hypot function\nAvailable since 2.25\n \n\nR cos(T x) The cosine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sin(T x) The sine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tan(T x) The tangent of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acos(T x) The principal arc (inverse) cosine (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR asin(T x) The principal arc (inverse) sine (in radians) of x\nAvailable since 2.0\n \n\nR atan(T x) The principal arc (inverse) tangent (in radians) of x, with values from \\(-\\pi/2\\) to \\(\\pi/2\\)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atan2(T y, T x) Return the principal arc (inverse) tangent (in radians) of y divided by x, \\[\\begin{equation*} \\text{atan2}(y, x) = \\arctan\\left(\\frac{y}{x}\\right) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.34\n\n\n\n \n\nR cosh(T x) The hyperbolic cosine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sinh(T x) The hyperbolic sine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tanh(T x) The hyperbolic tangent of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acosh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR asinh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atanh(T x) The inverse hyperbolic tangent (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\nThe following functions are commonly used as link functions in generalized linear models. The function \\(\\Phi\\) is also commonly used as a link function (see section probability-related functions).\n \n\nR logit(T x) The log odds, or logit, function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_logit(T x) The logistic sigmoid function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_cloglog(T x) The inverse of the complementary log-log function applied to x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n\n\nThe error function erf is related to the standard normal cumulative distribution function \\(\\Phi\\) by scaling. See section normal distribution for the general normal cumulative distribution function (and its complement).\n \n\nR erf(T x) The error function, also known as the Gauss error function, of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR erfc(T x) The complementary error function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_erfc(T x) The inverse of the complementary error function of x\nAvailable since 2.29, vectorized in 2.29\n \n\nR Phi(T x) The standard normal cumulative distribution function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_Phi(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The details of the algorithm can be found in (Wichura 1988). Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.\nAvailable since 2.0, vectorized in 2.13\n \n\nR Phi_approx(T x) The fast approximation of the unit (may replace Phi for probit regression with maximum absolute error of 0.00014, see (Bowling et al. 2009) for details)\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal binary_log_loss(int y, real y_hat) Return the log loss function for for predicting \\(\\hat{y} \\in [0,1]\\) for boolean outcome \\(y \\in \\{0,1\\}\\). \\[\\begin{equation*}\n\\mathrm{binary\\_log\\_loss}(y,\\hat{y}) = \\begin{cases} -\\log \\hat{y} &\n\\text{if } y = 1\\\\ -\\log (1 - \\hat{y}) & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR binary_log_loss(T1 x, T2 y) Vectorized implementation of the binary_log_loss function\nAvailable since 2.25\n \n\nreal owens_t(real h, real a) Return the Owen’s T function for the probability of the event \\(X &gt; h\\) and \\(0&lt;Y&lt;aX\\) where X and Y are independent standard normal random variables. \\[\\begin{equation*} \\mathrm{owens\\_t}(h,a) = \\frac{1}{2\\pi} \\int_0^a\n\\frac{\\exp(-\\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \\end{equation*}\\]\nAvailable since 2.25\n \n\nR owens_t(T1 x, T2 y) Vectorized implementation of the owens_t function\nAvailable since 2.25\n\n\n\n\n \n\nreal beta(real alpha, real beta) Return the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.25\n \n\nR beta(T1 x, T2 y) Vectorized implementation of the beta function\nAvailable since 2.25\n \n\nreal inc_beta(real alpha, real beta, real x) Return the regularized incomplete beta function up to x applied to alpha and beta. See section appendix for a definition.\nAvailable since 2.10\n \n\nreal inv_inc_beta(real alpha, real beta, real p) Return the inverse of the regularized incomplete beta function. The return value x is the value that solves p = inc_beta(alpha, beta, x). See section appendix for a definition of the inc_beta.\nAvailable since 2.30\n \n\nreal lbeta(real alpha, real beta) Return the natural logarithm of the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). \\[\\begin{equation*}\n\\text{lbeta}(\\alpha,\\beta) = \\log \\Gamma(\\alpha) + \\log \\Gamma(\\beta) - \\log \\Gamma(\\alpha+\\beta)\n\\end{equation*}\\] See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.0\n \n\nR lbeta(T1 x, T2 y) Vectorized implementation of the lbeta function\nAvailable since 2.25\n \n\nR tgamma(T x) The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that \\(\\Gamma(n+1) = n!\\). See for a full definition of \\(\\Gamma(x)\\). The function is defined for positive numbers and non-integral negative numbers,\nAvailable since 2.0, vectorized in 2.13\n \n\nR lgamma(T x) The natural logarithm of the gamma function applied to x,\nAvailable since 2.0, vectorized in 2.15\n \n\nR digamma(T x) The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers\nAvailable since 2.0, vectorized in 2.13\n \n\nR trigamma(T x) The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function\nAvailable since 2.0, vectorized in 2.13\n \n\nreal lmgamma(int n, real x) Return the natural logarithm of the multivariate gamma function \\(\\Gamma_n\\) with n dimensions applied to x. \\[\\begin{equation*}\n\\text{lmgamma}(n,x) =\n\\begin{cases} \\frac{n(n-1)}{4} \\log \\pi + \\sum_{j=1}^n \\log \\Gamma\\left(x + \\frac{1 - j}{2}\\right)\n& \\text{if } x\\not\\in \\{\\dots,-3,-2,-1,0\\}\\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR lmgamma(T1 x, T2 y) Vectorized implementation of the lmgamma function\nAvailable since 2.25\n \n\nreal gamma_p(real a, real z) Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_p}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_0^zt^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_p(T1 x, T2 y) Vectorized implementation of the gamma_p function\nAvailable since 2.25\n \n\nreal gamma_q(real a, real z) Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_q}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_z^\\infty t^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\[6pt] \\textrm{error} & \\text{otherwise}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_q(T1 x, T2 y) Vectorized implementation of the gamma_q function\nAvailable since 2.25\n \n\nint choose(int x, int y) Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” In its the antilog of the lchoose function but returns an integer rather than a real number with no non-zero decimal places. For \\(0 \\leq y \\leq x\\), the binomial coefficient function can be defined via the factorial function \\[\\begin{equation*}\n\\text{choose}(x,y) = \\frac{x!}{\\left(y!\\right)\\left(x - y\\right)!}.\n\\end{equation*}\\]\nAvailable since 2.14\n \n\nR choose(T1 x, T2 y) Vectorized implementation of the choose function\nAvailable since 2.25\n \n\nreal bessel_first_kind(int v, real x) Return the Bessel function of the first kind with order v applied to x. \\[\\begin{equation*}\n\\mathrm{bessel\\_first\\_kind}(v,x) = J_v(x),\n\\end{equation*}\\] where \\[\\begin{equation*}\nJ_v(x)=\\left(\\frac{1}{2}x\\right)^v \\sum_{k=0}^\\infty\n\\frac{\\left(-\\frac{1}{4}x^2\\right)^k}{k!\\, \\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_first_kind(T1 x, T2 y) Vectorized implementation of the bessel_first_kind function\nAvailable since 2.25\n \n\nreal bessel_second_kind(int v, real x) Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For \\(x,v &gt; 0\\), \\[\\begin{equation*}\n\\mathrm{bessel\\_second\\_kind}(v,x) =\n\\begin{cases} Y_v(x) & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\nY_v(x)=\\frac{J_v(x)\\cos(v\\pi)-J_{-v}(x)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_second_kind(T1 x, T2 y) Vectorized implementation of the bessel_second_kind function\nAvailable since 2.25\n \n\nreal modified_bessel_first_kind(int v, real z) Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_first\\_kind}(v,z) = I_v(z)\n\\end{equation*}\\] where \\[\\begin{equation*}\n{I_v}(z) = \\left(\\frac{1}{2}z\\right)^v\\sum_{k=0}^\\infty \\frac{\\left(\\frac{1}{4}z^2\\right)^k}{k!\\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_first_kind function\nAvailable since 2.25\n \n\nreal log_modified_bessel_first_kind(real v, real z) Return the log of the modified Bessel function of the first kind. v does not have to be an integer.\nAvailable since 2.26\n \n\nR log_modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the log_modified_bessel_first_kind function\nAvailable since 2.26\n \n\nreal modified_bessel_second_kind(int v, real z) Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_second\\_kind}(v,z) =\n\\begin{cases} K_v(z) &  \\text{if } z &gt; 0 \\\\ \\textrm{error} & \\text{if } z \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} {K_v}(z) = \\frac{\\pi}{2}\\cdot\\frac{I_{-v}(z) - I_{v}(z)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_second_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_second_kind function\nAvailable since 2.25\n \n\nreal falling_factorial(real x, real n) Return the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*}\n\\mathrm{falling\\_factorial}(x,n) =\n\\begin{cases} (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\n(x)_n=\\frac{\\Gamma(x+1)}{\\Gamma(x-n+1)}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR falling_factorial(T1 x, T2 y) Vectorized implementation of the falling_factorial function\nAvailable since 2.25\n \n\nreal lchoose(real x, real y) Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For \\(0 \\leq y \\leq x\\), \\[\\begin{equation*} \\mathrm{binomial\\_coefficient\\_log}(x,y) =\n\\log\\Gamma(x+1) - \\log\\Gamma(y+1) - \\log\\Gamma(x-y+1). \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lchoose(T1 x, T2 y) Vectorized implementation of the lchoose function\nAvailable since 2.29\n \n\nreal log_falling_factorial(real x, real n) Return the log of the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_falling\\_factorial}(x,n) =\n\\begin{cases} \\log (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal rising_factorial(real x, int n) Return the rising factorial of x with power n defined for positive x and integer n. \\[\\begin{equation*}\n\\mathrm{rising\\_factorial}(x,n) = \\begin{cases} x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} x^{(n)}=\\frac{\\Gamma(x+n)}{\\Gamma(x)} \\end{equation*}\\]\nAvailable since 2.20\n \n\nR rising_factorial(T1 x, T2 y) Vectorized implementation of the rising_factorial function\nAvailable since 2.25\n \n\nreal log_rising_factorial(real x, real n) Return the log of the rising factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_rising\\_factorial}(x,n) =\n\\begin{cases} \\log x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR log_rising_factorial(T1 x, T2 y) Vectorized implementation of the log_rising_factorial function\nAvailable since 2.25\n\n\n\nThe functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.\n \n\nR expm1(T x) The natural exponential of x minus 1\nAvailable since 2.0, vectorized in 2.13\n \n\nreal fma(real x, real y, real z) Return z plus the result of x multiplied by y. \\[\\begin{equation*} \\text{fma}(x,y,z) =\n(x \\times y) + z \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal ldexp(real x, int y) Return the product of x and two raised to the y power. \\[\\begin{equation*}\n\\text{ldexp}(x,y) = x 2^y  \\end{equation*}\\]\nAvailable since 2.25\n \n\nR ldexp(T1 x, T2 y) Vectorized implementation of the ldexp function\nAvailable since 2.25\n \n\nreal lmultiply(real x, real y) Return the product of x and the natural logarithm of y. \\[\\begin{equation*}\n\\text{lmultiply}(x,y) = \\begin{cases} 0 & \\text{if } x = y = 0 \\\\ x\n\\log y & \\text{if } x, y \\neq 0 \\\\ \\text{NaN} & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lmultiply(T1 x, T2 y) Vectorized implementation of the lmultiply function\nAvailable since 2.25\n \n\nR log1p(T x) The natural logarithm of 1 plus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m(T x) The natural logarithm of 1 minus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1p_exp(T x) The natural logarithm of one plus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m_exp(T x) The logarithm of one minus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal log_diff_exp(real x, real y) Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_diff\\_exp}(x,y) = \\begin{cases} \\log(\\exp(x)-\\exp(y)) &\n\\text{if } +\\infty &gt; x \\ge y \\\\[6pt]\n\\textrm{NaN} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nWhen x is equal to y, log_diff_exp(x, y) returns \\(-\\infty\\), consistent with log(0) returning \\(-\\infty\\). This includes the case in which x and y are both equal to \\(-\\infty\\), which corresponds to log(0 - 0) because exp(negative_infinity()) returns 0.\nAvailable since 2.0\n \n\nR log_diff_exp(T1 x, T2 y) Vectorized implementation of the log_diff_exp function\nAvailable since 2.25\n \n\nreal log_mix(real theta, real lp1, real lp2) Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by \\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda_1, \\lambda_2) & = & \\log \\!\\left(\n\\theta \\exp(\\lambda_1) + \\left( 1 - \\theta \\right) \\exp(\\lambda_2)\n\\right) \\\\[3pt] & = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) +\n\\lambda_1, \\ \\log(1 - \\theta) + \\lambda_2\\right). \\end{eqnarray*}\\]\nAvailable since 2.6\n \n\nR log_mix(T1 thetas, T2 lps)\nCalculates the log mixture density given thetas, mixing proportions which should be between 0 and 1 and sum to 1, and lps, log densities. The lps variable must be either a 1-d container of the same length as thetas, or an array of such.\n\\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda)\n& = & \\log \\!\\left( \\sum_{n=1}^N \\theta_n * \\exp(\\lambda_n) \\right) \\\\[3pt]\n& = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) + \\lambda\\right).\n\\end{eqnarray*}\\]\nThis is a generalization of the above signature of three arguments to more than two densities. For example, log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2}).\nAvailable since 2.26\n \n\nR log_sum_exp(T1 x, T2 y) Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_sum\\_exp}(x,y) = \\log(\\exp(x)+\\exp(y)) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.33\n \n\nR log_inv_logit(T x) The natural logarithm of the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log_inv_logit_diff(T1 x, T2 y) The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y\nAvailable since 2.25\n \n\nR log1m_inv_logit(T x) The natural logarithm of 1 minus the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nR lambert_w0(T x) Implementation of the \\(W_0\\) branch of the Lambert W function, i.e., solution to the function \\(W_0(x) \\exp^{ W_0(x)} = x\\)\nAvailable since 2.25\n \n\nR lambert_wm1(T x) Implementation of the \\(W_{-1}\\) branch of the Lambert W function, i.e., solution to the function \\(W_{-1}(x) \\exp^{W_{-1}(x)} = x\\)\nAvailable since 2.25\n\n\n\nHypergeometric functions refer to a power series of the form \\[\\begin{equation*}\n_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \\sum_{n=0}^\\infty \\frac{(a_1)_n\\cdot\\cdot\\cdot(a_p)_n}{(b_1)_n\\cdot\\cdot\\cdot(b_q)_n} \\frac{z^n}{n!}\n\\end{equation*}\\] where \\((a)_n\\) is the Pochhammer symbol defined as \\((a)_n = \\frac{\\Gamma(a+n)}{\\Gamma(a)}\\).\nThe gradients of the hypergeometric function are given by: \\[\\begin{equation*}\n\\frac{\\partial }{\\partial a_1} =\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+a_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n    - \\psi\\left(a_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)\n\\end{equation*}\\] \\[\\begin{equation*}\n\\frac{\\partial }{\\partial b_1} =\n  \\psi\\left(b_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+b_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n\\end{equation*}\\] \\[\\begin{equation*}\n  \\frac{\\partial }{\\partial z} =\n  \\frac{\\prod_{j=1}^{p}a_j}{\\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)\n\\end{equation*}\\]\nStan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.\n \n\nreal hypergeometric_1F0(real a, real z) Special case of the hypergeometric function with \\(p=1\\) and \\(q=0\\).\nAvailable since 2.37\n \n\nreal hypergeometric_2F1(real a1, real a2, real b1, real z) Special case of the hypergeometric function with \\(p=2\\) and \\(q=1\\). If the function does not meet convergence criteria for given inputs, the function will attempt to apply Euler’s transformation to improve convergence: \\[\\begin{equation*}\n{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \\frac{z}{z-1})\\cdot(1-z)^{-a_2}\n\\end{equation*}\\]\nAvailable since 2.37\n \n\nreal hypergeometric_3F2(T1 a, T2 b, real z) Special case of the hypergeometric function with \\(p=3\\) and \\(q=2\\), where a and b are vectors of length 3 and 2, respectively.\nAvailable since 2.37\n \n\nreal hypergeometric_pFq(T1 a, T2 b, real z) Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#fun-vectorization",
+    "href": "functions-reference/real-valued_basic_functions.html#fun-vectorization",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "Although listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.\n\n\nMany of Stan’s unary functions can be applied to any argument type. For example, the exponential function, exp, can be applied to real arguments or arrays of real arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.\n\n\nWhen applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like exp() are defined elementwise. For example,\n // declare some variables for arguments\n real x0;\n array[5] real x1;\n array[4, 7] real x2;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = exp(x0);\n y1 = exp(x1);\n y2 = exp(x2);\nWhen exp is applied to an array, it applies elementwise. For example, the statement above,\n y2 = exp(x2);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = exp(x2[i, j]);\n  }\n}\n\n\n\nVectorized functions also apply elementwise to vectors and matrices. For example,\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = exp(xv);\n yrv = exp(xrv);\n ym = exp(xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = exp(u);\nAfter this has been executed, z[i, j, k] will be equal to exp(u[i, j, k]).\n\n\n\nInteger arguments are promoted to real values in vectorized unary functions. Thus if n is of type int, exp(n) is of type real. Arrays work the same way, so that if n2 is a one dimensional array of integers, then exp(n2) will be a one-dimensional array of reals with the same number of elements as n2. For example,\n array[23] int n1;\n array[23] real z1;\n z1 = exp(n1);\nIt would be illegal to try to assign exp(n1) to an array of integers; the return type is a real array.\n\n\n\n\nLike the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.\n\n\nWhen applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like pow() are defined elementwise. For example,\n // declare some variables for arguments\n real x00;\n real x01;\n array[5] real x10;\n array[5]real x11;\n array[4, 7] real x20;\n array[4, 7] real x21;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = pow(x00, x01);\n y1 = pow(x10, x11);\n y2 = pow(x20, x21);\nWhen pow is applied to two arrays, it applies elementwise. For example, the statement above,\n y2 = pow(x20, x21);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x21[i, j]);\n  }\n}\nAlternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:\ny2 = pow(x20, x00);\nproduces the same result for y2 as the explicit loop:\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x00);\n  }\n}\n\n\n\nVectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,\n real x00;\n vector[5] xv00;\n vector[5] xv01;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = pow(xv00, xv01);\n yrv = pow(xrv, x00);\n ym = pow(x00, xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = pow(u, x00);\nAfter this has been executed, z[i, j, k] will be equal to pow(u[i, j, k], x00).\n\n\n\nVectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv = pow(xv, xv)\n row_vector[7] yrv = pow(xrv, xrv)\n matrix[10, 20] = pow(xm, xm)\nBut the following statements are not:\n vector[5] xv;\n vector[7] xv2;\n row_vector[5] xrv;\n\n // Cannot mix different types\n vector[5] yv = pow(xv, xrv)\n\n // Cannot mix different sizes of the same type\n vector[5] yv = pow(xv, xv2)\nWhile the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., bessel_first_kind). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:\n vector[5] xv;\n matrix[5, 5] xm;\n array[5] int xi;\n array[5, 5] int xii;\n\n vector[5] yv = bessel_first_kind(xi, xv);\n matrix[5, 5] ym = bessel_first_kind(xii, xm);\nWhereas these are not:\n vector[5] xv;\n matrix[5, 5] xm;\n array[7] int xi;\n\n // Dimensions of containers do not match\n vector[5] yv = bessel_first_kind(xi, xv);\n\n // Function requires first argument be an integer type\n matrix[5, 5] ym = bessel_first_kind(xm, xm);",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#built-in-constants",
+    "href": "functions-reference/real-valued_basic_functions.html#built-in-constants",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "Constants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant \\(\\pi\\) must be written in a Stan program as pi().\n \n\nreal pi() \\(\\pi\\), the ratio of a circle’s circumference to its diameter\nAvailable since 2.0\n \n\nreal e() \\(e\\), the base of the natural logarithm\nAvailable since 2.0\n \n\nreal sqrt2() The square root of 2\nAvailable since 2.0\n \n\nreal log2() The natural logarithm of 2\nAvailable since 2.0\n \n\nreal log10() The natural logarithm of 10\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#special-values",
+    "href": "functions-reference/real-valued_basic_functions.html#special-values",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "real not_a_number() Not-a-number, a special non-finite real value returned to signal an error\nAvailable since 2.0\n \n\nreal positive_infinity() Positive infinity, a special non-finite real value larger than all finite numbers\nAvailable since 2.0\n \n\nreal negative_infinity() Negative infinity, a special non-finite real value smaller than all finite numbers\nAvailable since 2.0\n \n\nreal machine_precision() The smallest number \\(x\\) such that \\((x + 1) \\neq 1\\) in floating-point arithmetic on the current hardware platform\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#get-log-prob",
+    "href": "functions-reference/real-valued_basic_functions.html#get-log-prob",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "The basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (~) and log probability increment statement (increment_log_prob) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.\nStan provides a special built-in function target() that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.\n \n\nreal target() Return the current value of the log probability accumulator.\nAvailable since 2.10\ntarget acts like a function ending in _lp, meaning that it may only may only be used in the model block.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#logical-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#logical-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "Like C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.\n\n\nAll comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.\n \n\nint operator&lt;(int x, int y)\n \n\nint operator&lt;(real x, real y) Return 1 if x is less than y and 0 otherwise. \\[\\begin{equation*} \\text{operator&lt;}(x,y)\n= \\begin{cases} 1 & \\text{if $x &lt; y$} \\\\ 0 & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&lt;=(int x, int y)\n \n\nint operator&lt;=(real x, real y) Return 1 if x is less than or equal y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&lt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\leq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;(int x, int y)\n \n\nint operator&gt;(real x, real y) Return 1 if x is greater than y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;}(x,y) = \\begin{cases} 1 & \\text{if $x &gt; y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;=(int x, int y)\n \n\nint operator&gt;=(real x, real y) Return 1 if x is greater than or equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\geq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator==(int x, int y)\n \n\nint operator==(real x, real y) Return 1 if x is equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator==}(x,y) = \\begin{cases} 1 & \\text{if $x = y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!=(int x, int y)\n \n\nint operator!=(real x, real y) Return 1 if x is not equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator!=}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq y$} \\\\ 0 &\n\\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\nBoolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as !a && b is interpreted as (!a) && b, and a &lt; b || c &gt;= d && e != f as (a &lt; b) || (((c &gt;= d) && (e != f))).\n \n\nint operator!(int x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0$} \\\\ 1 & \\text{if $x = 0$}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!(real x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0.0$} \\\\ 1 & \\text{if $x = 0.0$}\n\\end{cases} \\end{equation*}\\] deprecated; - use operator== instead.\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator&&(int x, int y)\nReturn 1 if x is unequal to 0 and y is unequal to 0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\text{ and } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&&(real x, real y) Return 1 if x is unequal to 0.0 and y is unequal to 0.0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\text{ and } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator||(int x, int y) Return 1 if x is unequal to 0 or y is unequal to 0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\textrm{ or } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator||(real x, real y) Return 1 if x is unequal to 0.0 or y is unequal to 0.0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\textrm{ or } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n\n\nLike in C++, the boolean operators && and || and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating a || b, if a evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression b. Similarly, evaluating a && b first evaluates a, and if the result is zero, returns 0 without evaluating b.\n\n\n\n\nThe logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.\n \n\nreal step(real x) Return 1 if x is positive and 0 otherwise. \\[\\begin{equation*} \\text{step}(x) =\n\\begin{cases} 0 & \\text{if } x &lt; 0 \\\\ 1 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nThe step function is often used in BUGS to perform conditional operations. For instance, step(a-b) evaluates to 1 if a is greater than b and evaluates to 0 otherwise. step is a step-like functions; see the warning in section step functions applied to expressions dependent on parameters.\nAvailable since 2.0\n \n\nint is_inf(real x) Return 1 if x is infinite (positive or negative) and 0 otherwise.\nAvailable since 2.5\n \n\nint is_nan(real x) Return 1 if x is NaN and 0 otherwise.\nAvailable since 2.5\nCare must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section step-like functions for details.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#real-valued-arithmetic-operators",
+    "href": "functions-reference/real-valued_basic_functions.html#real-valued-arithmetic-operators",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "The arithmetic operators are presented using C++ notation. For instance operator+(x,y) refers to the binary addition operator and operator-(x) to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as x + y and -x, respectively.\n\n\n \n\nreal operator+(real x, real y) Return the sum of x and y. \\[\\begin{equation*} (x + y) = \\text{operator+}(x,y) = x+y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator-(real x, real y) Return the difference between x and y. \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x,y) = x - y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator*(real x, real y) Return the product of x and y. \\[\\begin{equation*} (x * y) = \\text{operator*}(x,y) = xy\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator/(real x, real y) Return the quotient of x and y. \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator^(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} (x^\\mathrm{\\wedge}y) =\n\\text{operator}^\\mathrm{\\wedge}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.5\n\n\n\n \n\nreal operator-(real x) Return the negation of the subtrahend x. \\[\\begin{equation*} \\text{operator-}(x) = (-x)\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of reals, -x is the same shape array where each individual number is negated.\nAvailable since 2.31\n \n\nreal operator+(real x) Return the value of x. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#step-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#step-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "Warning: These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is real vector, row_vector, matrix, or an array of those types, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nreal fdim(real x, real y) Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. \\[\\begin{equation*} \\text{fdim}(x,y) = \\begin{cases} x-y &\n\\text{if } x \\geq y \\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fdim(T1 x, T2 y) Vectorized implementation of the fdim function\nAvailable since 2.25\n\n\n\n \n\nreal fmin(real x, real y) Return the minimum of x and y; see warning above. \\[\\begin{equation*} \\text{fmin}(x,y) = \\begin{cases} x &\n\\text{if } x \\leq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmin(T1 x, T2 y) Vectorized implementation of the fmin function\nAvailable since 2.25\n \n\nreal fmax(real x, real y) Return the maximum of x and y; see warning above. \\[\\begin{equation*} \\text{fmax}(x,y) = \\begin{cases} x &\n\\text{if } x \\geq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmax(T1 x, T2 y) Vectorized implementation of the fmax function\nAvailable since 2.25\n\n\n\n \n\nreal fmod(real x, real y) Return the real value remainder after dividing x by y; see warning above. \\[\\begin{equation*} \\text{fmod}(x,y) = x - \\left\\lfloor \\frac{x}{y} \\right\\rfloor \\, y \\end{equation*}\\] The operator \\(\\lfloor u \\rfloor\\) is the floor operation; see below.\nAvailable since 2.0\n \n\nR fmod(T1 x, T2 y) Vectorized implementation of the fmod function\nAvailable since 2.25\n\n\n\nWarning: Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.\nThe rounding functions cannot be used as indices to arrays because they return real values. For operations over data or in the generated quantities block, the to_int() function can be used.\n \n\nR floor(T x) The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR ceil(T x) The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR round(T x) The nearest integer to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR trunc(T x) The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#power-and-logarithm-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#power-and-logarithm-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "R sqrt(T x) The square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR cbrt(T x) The cube root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR square(T x) The square of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp(T x) The natural exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp2(T x) The base-2 exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log(T x) The natural logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log2(T x) The base-2 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log10(T x) The base-10 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal pow(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} \\text{pow}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.0\n \n\nR pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.25\n \n\nR inv(T x) The inverse of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_sqrt(T x) The inverse of the square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_square(T x) The inverse of the square of x\nAvailable since 2.0, vectorized in 2.13",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#trigonometric-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#trigonometric-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "real hypot(real x, real y) Return the length of the hypotenuse of a right triangle with sides of length x and y. \\[\\begin{equation*} \\text{hypot}(x,y) = \\begin{cases} \\sqrt{x^2+y^2} &\n\\text{if } x,y\\geq 0 \\\\ \\textrm{NaN} & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR hypot(T1 x, T2 y) Vectorized implementation of the hypot function\nAvailable since 2.25\n \n\nR cos(T x) The cosine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sin(T x) The sine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tan(T x) The tangent of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acos(T x) The principal arc (inverse) cosine (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR asin(T x) The principal arc (inverse) sine (in radians) of x\nAvailable since 2.0\n \n\nR atan(T x) The principal arc (inverse) tangent (in radians) of x, with values from \\(-\\pi/2\\) to \\(\\pi/2\\)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atan2(T y, T x) Return the principal arc (inverse) tangent (in radians) of y divided by x, \\[\\begin{equation*} \\text{atan2}(y, x) = \\arctan\\left(\\frac{y}{x}\\right) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#hyperbolic-trigonometric-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#hyperbolic-trigonometric-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "R cosh(T x) The hyperbolic cosine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sinh(T x) The hyperbolic sine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tanh(T x) The hyperbolic tangent of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acosh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR asinh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atanh(T x) The inverse hyperbolic tangent (in radians) of x\nAvailable since 2.0, vectorized in 2.13",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#link-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#link-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "The following functions are commonly used as link functions in generalized linear models. The function \\(\\Phi\\) is also commonly used as a link function (see section probability-related functions).\n \n\nR logit(T x) The log odds, or logit, function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_logit(T x) The logistic sigmoid function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_cloglog(T x) The inverse of the complementary log-log function applied to x\nAvailable since 2.0, vectorized in 2.13",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#Phi-function",
+    "href": "functions-reference/real-valued_basic_functions.html#Phi-function",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "The error function erf is related to the standard normal cumulative distribution function \\(\\Phi\\) by scaling. See section normal distribution for the general normal cumulative distribution function (and its complement).\n \n\nR erf(T x) The error function, also known as the Gauss error function, of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR erfc(T x) The complementary error function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_erfc(T x) The inverse of the complementary error function of x\nAvailable since 2.29, vectorized in 2.29\n \n\nR Phi(T x) The standard normal cumulative distribution function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_Phi(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The details of the algorithm can be found in (Wichura 1988). Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.\nAvailable since 2.0, vectorized in 2.13\n \n\nR Phi_approx(T x) The fast approximation of the unit (may replace Phi for probit regression with maximum absolute error of 0.00014, see (Bowling et al. 2009) for details)\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal binary_log_loss(int y, real y_hat) Return the log loss function for for predicting \\(\\hat{y} \\in [0,1]\\) for boolean outcome \\(y \\in \\{0,1\\}\\). \\[\\begin{equation*}\n\\mathrm{binary\\_log\\_loss}(y,\\hat{y}) = \\begin{cases} -\\log \\hat{y} &\n\\text{if } y = 1\\\\ -\\log (1 - \\hat{y}) & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR binary_log_loss(T1 x, T2 y) Vectorized implementation of the binary_log_loss function\nAvailable since 2.25\n \n\nreal owens_t(real h, real a) Return the Owen’s T function for the probability of the event \\(X &gt; h\\) and \\(0&lt;Y&lt;aX\\) where X and Y are independent standard normal random variables. \\[\\begin{equation*} \\mathrm{owens\\_t}(h,a) = \\frac{1}{2\\pi} \\int_0^a\n\\frac{\\exp(-\\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \\end{equation*}\\]\nAvailable since 2.25\n \n\nR owens_t(T1 x, T2 y) Vectorized implementation of the owens_t function\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#betafun",
+    "href": "functions-reference/real-valued_basic_functions.html#betafun",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "real beta(real alpha, real beta) Return the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.25\n \n\nR beta(T1 x, T2 y) Vectorized implementation of the beta function\nAvailable since 2.25\n \n\nreal inc_beta(real alpha, real beta, real x) Return the regularized incomplete beta function up to x applied to alpha and beta. See section appendix for a definition.\nAvailable since 2.10\n \n\nreal inv_inc_beta(real alpha, real beta, real p) Return the inverse of the regularized incomplete beta function. The return value x is the value that solves p = inc_beta(alpha, beta, x). See section appendix for a definition of the inc_beta.\nAvailable since 2.30\n \n\nreal lbeta(real alpha, real beta) Return the natural logarithm of the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). \\[\\begin{equation*}\n\\text{lbeta}(\\alpha,\\beta) = \\log \\Gamma(\\alpha) + \\log \\Gamma(\\beta) - \\log \\Gamma(\\alpha+\\beta)\n\\end{equation*}\\] See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.0\n \n\nR lbeta(T1 x, T2 y) Vectorized implementation of the lbeta function\nAvailable since 2.25\n \n\nR tgamma(T x) The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that \\(\\Gamma(n+1) = n!\\). See for a full definition of \\(\\Gamma(x)\\). The function is defined for positive numbers and non-integral negative numbers,\nAvailable since 2.0, vectorized in 2.13\n \n\nR lgamma(T x) The natural logarithm of the gamma function applied to x,\nAvailable since 2.0, vectorized in 2.15\n \n\nR digamma(T x) The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers\nAvailable since 2.0, vectorized in 2.13\n \n\nR trigamma(T x) The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function\nAvailable since 2.0, vectorized in 2.13\n \n\nreal lmgamma(int n, real x) Return the natural logarithm of the multivariate gamma function \\(\\Gamma_n\\) with n dimensions applied to x. \\[\\begin{equation*}\n\\text{lmgamma}(n,x) =\n\\begin{cases} \\frac{n(n-1)}{4} \\log \\pi + \\sum_{j=1}^n \\log \\Gamma\\left(x + \\frac{1 - j}{2}\\right)\n& \\text{if } x\\not\\in \\{\\dots,-3,-2,-1,0\\}\\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR lmgamma(T1 x, T2 y) Vectorized implementation of the lmgamma function\nAvailable since 2.25\n \n\nreal gamma_p(real a, real z) Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_p}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_0^zt^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_p(T1 x, T2 y) Vectorized implementation of the gamma_p function\nAvailable since 2.25\n \n\nreal gamma_q(real a, real z) Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_q}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_z^\\infty t^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\[6pt] \\textrm{error} & \\text{otherwise}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_q(T1 x, T2 y) Vectorized implementation of the gamma_q function\nAvailable since 2.25\n \n\nint choose(int x, int y) Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” In its the antilog of the lchoose function but returns an integer rather than a real number with no non-zero decimal places. For \\(0 \\leq y \\leq x\\), the binomial coefficient function can be defined via the factorial function \\[\\begin{equation*}\n\\text{choose}(x,y) = \\frac{x!}{\\left(y!\\right)\\left(x - y\\right)!}.\n\\end{equation*}\\]\nAvailable since 2.14\n \n\nR choose(T1 x, T2 y) Vectorized implementation of the choose function\nAvailable since 2.25\n \n\nreal bessel_first_kind(int v, real x) Return the Bessel function of the first kind with order v applied to x. \\[\\begin{equation*}\n\\mathrm{bessel\\_first\\_kind}(v,x) = J_v(x),\n\\end{equation*}\\] where \\[\\begin{equation*}\nJ_v(x)=\\left(\\frac{1}{2}x\\right)^v \\sum_{k=0}^\\infty\n\\frac{\\left(-\\frac{1}{4}x^2\\right)^k}{k!\\, \\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_first_kind(T1 x, T2 y) Vectorized implementation of the bessel_first_kind function\nAvailable since 2.25\n \n\nreal bessel_second_kind(int v, real x) Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For \\(x,v &gt; 0\\), \\[\\begin{equation*}\n\\mathrm{bessel\\_second\\_kind}(v,x) =\n\\begin{cases} Y_v(x) & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\nY_v(x)=\\frac{J_v(x)\\cos(v\\pi)-J_{-v}(x)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_second_kind(T1 x, T2 y) Vectorized implementation of the bessel_second_kind function\nAvailable since 2.25\n \n\nreal modified_bessel_first_kind(int v, real z) Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_first\\_kind}(v,z) = I_v(z)\n\\end{equation*}\\] where \\[\\begin{equation*}\n{I_v}(z) = \\left(\\frac{1}{2}z\\right)^v\\sum_{k=0}^\\infty \\frac{\\left(\\frac{1}{4}z^2\\right)^k}{k!\\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_first_kind function\nAvailable since 2.25\n \n\nreal log_modified_bessel_first_kind(real v, real z) Return the log of the modified Bessel function of the first kind. v does not have to be an integer.\nAvailable since 2.26\n \n\nR log_modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the log_modified_bessel_first_kind function\nAvailable since 2.26\n \n\nreal modified_bessel_second_kind(int v, real z) Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_second\\_kind}(v,z) =\n\\begin{cases} K_v(z) &  \\text{if } z &gt; 0 \\\\ \\textrm{error} & \\text{if } z \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} {K_v}(z) = \\frac{\\pi}{2}\\cdot\\frac{I_{-v}(z) - I_{v}(z)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_second_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_second_kind function\nAvailable since 2.25\n \n\nreal falling_factorial(real x, real n) Return the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*}\n\\mathrm{falling\\_factorial}(x,n) =\n\\begin{cases} (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\n(x)_n=\\frac{\\Gamma(x+1)}{\\Gamma(x-n+1)}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR falling_factorial(T1 x, T2 y) Vectorized implementation of the falling_factorial function\nAvailable since 2.25\n \n\nreal lchoose(real x, real y) Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For \\(0 \\leq y \\leq x\\), \\[\\begin{equation*} \\mathrm{binomial\\_coefficient\\_log}(x,y) =\n\\log\\Gamma(x+1) - \\log\\Gamma(y+1) - \\log\\Gamma(x-y+1). \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lchoose(T1 x, T2 y) Vectorized implementation of the lchoose function\nAvailable since 2.29\n \n\nreal log_falling_factorial(real x, real n) Return the log of the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_falling\\_factorial}(x,n) =\n\\begin{cases} \\log (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal rising_factorial(real x, int n) Return the rising factorial of x with power n defined for positive x and integer n. \\[\\begin{equation*}\n\\mathrm{rising\\_factorial}(x,n) = \\begin{cases} x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} x^{(n)}=\\frac{\\Gamma(x+n)}{\\Gamma(x)} \\end{equation*}\\]\nAvailable since 2.20\n \n\nR rising_factorial(T1 x, T2 y) Vectorized implementation of the rising_factorial function\nAvailable since 2.25\n \n\nreal log_rising_factorial(real x, real n) Return the log of the rising factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_rising\\_factorial}(x,n) =\n\\begin{cases} \\log x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR log_rising_factorial(T1 x, T2 y) Vectorized implementation of the log_rising_factorial function\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#composed-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#composed-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "The functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.\n \n\nR expm1(T x) The natural exponential of x minus 1\nAvailable since 2.0, vectorized in 2.13\n \n\nreal fma(real x, real y, real z) Return z plus the result of x multiplied by y. \\[\\begin{equation*} \\text{fma}(x,y,z) =\n(x \\times y) + z \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal ldexp(real x, int y) Return the product of x and two raised to the y power. \\[\\begin{equation*}\n\\text{ldexp}(x,y) = x 2^y  \\end{equation*}\\]\nAvailable since 2.25\n \n\nR ldexp(T1 x, T2 y) Vectorized implementation of the ldexp function\nAvailable since 2.25\n \n\nreal lmultiply(real x, real y) Return the product of x and the natural logarithm of y. \\[\\begin{equation*}\n\\text{lmultiply}(x,y) = \\begin{cases} 0 & \\text{if } x = y = 0 \\\\ x\n\\log y & \\text{if } x, y \\neq 0 \\\\ \\text{NaN} & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lmultiply(T1 x, T2 y) Vectorized implementation of the lmultiply function\nAvailable since 2.25\n \n\nR log1p(T x) The natural logarithm of 1 plus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m(T x) The natural logarithm of 1 minus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1p_exp(T x) The natural logarithm of one plus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m_exp(T x) The logarithm of one minus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal log_diff_exp(real x, real y) Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_diff\\_exp}(x,y) = \\begin{cases} \\log(\\exp(x)-\\exp(y)) &\n\\text{if } +\\infty &gt; x \\ge y \\\\[6pt]\n\\textrm{NaN} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nWhen x is equal to y, log_diff_exp(x, y) returns \\(-\\infty\\), consistent with log(0) returning \\(-\\infty\\). This includes the case in which x and y are both equal to \\(-\\infty\\), which corresponds to log(0 - 0) because exp(negative_infinity()) returns 0.\nAvailable since 2.0\n \n\nR log_diff_exp(T1 x, T2 y) Vectorized implementation of the log_diff_exp function\nAvailable since 2.25\n \n\nreal log_mix(real theta, real lp1, real lp2) Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by \\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda_1, \\lambda_2) & = & \\log \\!\\left(\n\\theta \\exp(\\lambda_1) + \\left( 1 - \\theta \\right) \\exp(\\lambda_2)\n\\right) \\\\[3pt] & = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) +\n\\lambda_1, \\ \\log(1 - \\theta) + \\lambda_2\\right). \\end{eqnarray*}\\]\nAvailable since 2.6\n \n\nR log_mix(T1 thetas, T2 lps)\nCalculates the log mixture density given thetas, mixing proportions which should be between 0 and 1 and sum to 1, and lps, log densities. The lps variable must be either a 1-d container of the same length as thetas, or an array of such.\n\\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda)\n& = & \\log \\!\\left( \\sum_{n=1}^N \\theta_n * \\exp(\\lambda_n) \\right) \\\\[3pt]\n& = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) + \\lambda\\right).\n\\end{eqnarray*}\\]\nThis is a generalization of the above signature of three arguments to more than two densities. For example, log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2}).\nAvailable since 2.26\n \n\nR log_sum_exp(T1 x, T2 y) Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_sum\\_exp}(x,y) = \\log(\\exp(x)+\\exp(y)) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.33\n \n\nR log_inv_logit(T x) The natural logarithm of the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log_inv_logit_diff(T1 x, T2 y) The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y\nAvailable since 2.25\n \n\nR log1m_inv_logit(T x) The natural logarithm of 1 minus the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#special-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#special-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "R lambert_w0(T x) Implementation of the \\(W_0\\) branch of the Lambert W function, i.e., solution to the function \\(W_0(x) \\exp^{ W_0(x)} = x\\)\nAvailable since 2.25\n \n\nR lambert_wm1(T x) Implementation of the \\(W_{-1}\\) branch of the Lambert W function, i.e., solution to the function \\(W_{-1}(x) \\exp^{W_{-1}(x)} = x\\)\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/real-valued_basic_functions.html#hypergeometric-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#hypergeometric-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "Hypergeometric functions refer to a power series of the form \\[\\begin{equation*}\n_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \\sum_{n=0}^\\infty \\frac{(a_1)_n\\cdot\\cdot\\cdot(a_p)_n}{(b_1)_n\\cdot\\cdot\\cdot(b_q)_n} \\frac{z^n}{n!}\n\\end{equation*}\\] where \\((a)_n\\) is the Pochhammer symbol defined as \\((a)_n = \\frac{\\Gamma(a+n)}{\\Gamma(a)}\\).\nThe gradients of the hypergeometric function are given by: \\[\\begin{equation*}\n\\frac{\\partial }{\\partial a_1} =\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+a_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n    - \\psi\\left(a_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)\n\\end{equation*}\\] \\[\\begin{equation*}\n\\frac{\\partial }{\\partial b_1} =\n  \\psi\\left(b_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+b_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n\\end{equation*}\\] \\[\\begin{equation*}\n  \\frac{\\partial }{\\partial z} =\n  \\frac{\\prod_{j=1}^{p}a_j}{\\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)\n\\end{equation*}\\]\nStan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.\n \n\nreal hypergeometric_1F0(real a, real z) Special case of the hypergeometric function with \\(p=1\\) and \\(q=0\\).\nAvailable since 2.37\n \n\nreal hypergeometric_2F1(real a1, real a2, real b1, real z) Special case of the hypergeometric function with \\(p=2\\) and \\(q=1\\). If the function does not meet convergence criteria for given inputs, the function will attempt to apply Euler’s transformation to improve convergence: \\[\\begin{equation*}\n{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \\frac{z}{z-1})\\cdot(1-z)^{-a_2}\n\\end{equation*}\\]\nAvailable since 2.37\n \n\nreal hypergeometric_3F2(T1 a, T2 b, real z) Special case of the hypergeometric function with \\(p=3\\) and \\(q=2\\), where a and b are vectors of length 3 and 2, respectively.\nAvailable since 2.37\n \n\nreal hypergeometric_pFq(T1 a, T2 b, real z) Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html",
+    "href": "functions-reference/positive_continuous_distributions.html",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "The positive continuous probability functions have support on the positive real numbers.\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}^+\\), \\[\\begin{equation*} \\text{LogNormal}(y|\\mu,\\sigma) = \\frac{1}{\\sqrt{2\n\\pi} \\ \\sigma} \\, \\frac{1}{y} \\ \\exp \\! \\left(        - \\, \\frac{1}{2}\n\\, \\left( \\frac{\\log y - \\mu}{\\sigma} \\right)^2      \\right) . \\end{equation*}\\]\n\n\n\ny ~ lognormal(mu, sigma)\nIncrement target log probability density with lognormal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal lognormal_lpdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lupdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal lognormal_cdf(reals y | reals mu, reals sigma) The cumulative lognormal distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal lognormal_lcdf(reals y | reals mu, reals sigma) The log of the lognormal cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lccdf(reals y | reals mu, reals sigma) The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR lognormal_rng(reals mu, reals sigma) Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.22\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{ChiSquare}(y|\\nu) = \\frac{2^{-\\nu/2}}     {\\Gamma(\\nu / 2)} \\,\ny^{\\nu/2 - 1} \\, \\exp \\! \\left( -\\, \\frac{1}{2} \\, y \\right) . \\end{equation*}\\]\n\n\n\ny ~ chi_square(nu)\nIncrement target log probability density with chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal chi_square_lpdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lupdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal chi_square_cdf(reals y | reals nu) The Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal chi_square_lcdf(reals y | reals nu) The log of the Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lccdf(reals y | reals nu) The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR chi_square_rng(reals nu) Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{InvChiSquare}(y \\, | \\, \\nu) = \\frac{2^{-\\nu/2}}    {\\Gamma(\\nu\n/ 2)} \\, y^{-\\nu/2 - 1} \\, \\exp\\! \\left( \\! - \\, \\frac{1}{2} \\,\n\\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_chi_square(nu)\nIncrement target log probability density with inv_chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_chi_square_lpdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lupdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_chi_square_cdf(reals y | reals nu) The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal inv_chi_square_lcdf(reals y | reals nu) The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lccdf(reals y | reals nu) The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR inv_chi_square_rng(reals nu) Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{ScaledInvChiSquare}(y|\\nu,\\sigma) =\n\\frac{(\\nu / 2)^{\\nu/2}}      {\\Gamma(\\nu / 2)} \\, \\sigma^{\\nu} \\,\ny^{-(\\nu/2 + 1)} \\, \\exp \\! \\left( \\!    - \\, \\frac{1}{2} \\, \\nu \\,\n\\sigma^2 \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ scaled_inv_chi_square(nu, sigma)\nIncrement target log probability density with scaled_inv_chi_square_lupdf(y | nu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal scaled_inv_chi_square_lpdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lupdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal scaled_inv_chi_square_cdf(reals y | reals nu, reals sigma) The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.0\n \n\nreal scaled_inv_chi_square_lcdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lccdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nR scaled_inv_chi_square_rng(reals nu, reals sigma) Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf inverse scale (rate) \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Exponential}(y|\\beta) = \\beta \\, \\exp ( - \\beta \\, y ) . \\end{equation*}\\]\n\n\n\ny ~ exponential(beta)\nIncrement target log probability density with exponential_lupdf(y | beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exponential_lpdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lupdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal exponential_cdf(reals y | reals beta) The exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.0\n \n\nreal exponential_lcdf(reals y | reals beta) The log of the exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lccdf(reals y | reals beta) The log of the exponential complementary cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nR exponential_rng(reals beta) Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf the shape parameter \\(\\alpha \\in \\mathbb{R}^+\\) and the rate (or inverse scale) parameter \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Gamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\, y^{\\alpha - 1}\n\\exp(-\\beta \\, y) . \\end{equation*}\\]\nUnder the shape and rate formulation of the Gamma distribution, \\(\\mathbb{E}[y] = \\alpha / \\beta\\) and \\(\\textrm{var}[y] = \\alpha / \\beta^2\\).\n\n\n\ny ~ gamma(alpha, beta)\nIncrement target log probability density with gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gamma_lpdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lupdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gamma_cdf(reals y | reals alpha, reals beta) The cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal gamma_lcdf(reals y | reals alpha, reals beta) The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lccdf(reals y | reals alpha, reals beta) The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR gamma_rng(reals alpha, reals beta) Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{InvGamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\ y^{-(\\alpha + 1)} \\,\n\\exp \\! \\left( \\! - \\beta \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_gamma(alpha, beta)\nIncrement target log probability density with inv_gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_gamma_lpdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lupdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_gamma_cdf(reals y | reals alpha, reals beta) The inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.0\n \n\nreal inv_gamma_lcdf(reals y | reals alpha, reals beta) The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lccdf(reals y | reals alpha, reals beta) The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nR inv_gamma_rng(reals alpha, reals beta) Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*} \\text{Weibull}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Weibull}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Frechet}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ weibull(alpha, sigma)\nIncrement target log probability density with weibull_lupdf(y | alpha, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal weibull_lpdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lupdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal weibull_cdf(reals y | reals alpha, reals sigma) The Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.0\n \n\nreal weibull_lcdf(reals y | reals alpha, reals sigma) The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lccdf(reals y | reals alpha, reals sigma) The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR weibull_rng(reals alpha, reals sigma) Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Frechet}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Frechet}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Weibull}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ frechet(alpha, sigma)\nIncrement target log probability density with frechet_lupdf(y | alpha, sigma).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal frechet_lpdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lupdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal frechet_cdf(reals y | reals alpha, reals sigma) The Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.5\n \n\nreal frechet_lcdf(reals y | reals alpha, reals sigma) The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lccdf(reals y | reals alpha, reals sigma) The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR frechet_rng(reals alpha, reals sigma) Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*}\n\\text{Rayleigh}(y|\\sigma) = \\frac{y}{\\sigma^2} \\exp(-y^2 / 2\\sigma^2)\n\\!. \\end{equation*}\\]\n\n\n\ny ~ rayleigh(sigma)\nIncrement target log probability density with rayleigh_lupdf(y | sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal rayleigh_lpdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lupdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal rayleigh_cdf(real y | real sigma) The Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.0\n \n\nreal rayleigh_lcdf(real y | real sigma) The log of the Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lccdf(real y | real sigma) The log of the Rayleigh complementary cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nR rayleigh_rng(reals sigma) Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha, \\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Log-Logistic}(y|\\alpha,\\beta) =\n\\frac{\\ \\left(\\frac{\\beta}{\\alpha}\\right) \\left(\\frac{y}{\\alpha}\\right)^{\\beta-1}\\ }{\\left(1 + \\left(\\frac{y}{\\alpha}\\right)^\\beta\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ loglogistic(alpha, beta)\nIncrement target log probability density with unnormalized version of loglogistic_lpdf(y | alpha, beta)\nAvailable since 2.29\n \n\n\n\n\n \n\nreal loglogistic_lpdf(reals y | reals alpha, reals beta) The log of the log-logistic density of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nreal loglogistic_cdf(reals y | reals alpha, reals beta) The log-logistic cumulative distribution function of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nR loglogistic_rng(reals alpha, reals beta) Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.29",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#lognormal",
+    "href": "functions-reference/positive_continuous_distributions.html#lognormal",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}^+\\), \\[\\begin{equation*} \\text{LogNormal}(y|\\mu,\\sigma) = \\frac{1}{\\sqrt{2\n\\pi} \\ \\sigma} \\, \\frac{1}{y} \\ \\exp \\! \\left(        - \\, \\frac{1}{2}\n\\, \\left( \\frac{\\log y - \\mu}{\\sigma} \\right)^2      \\right) . \\end{equation*}\\]\n\n\n\ny ~ lognormal(mu, sigma)\nIncrement target log probability density with lognormal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal lognormal_lpdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lupdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal lognormal_cdf(reals y | reals mu, reals sigma) The cumulative lognormal distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal lognormal_lcdf(reals y | reals mu, reals sigma) The log of the lognormal cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lccdf(reals y | reals mu, reals sigma) The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR lognormal_rng(reals mu, reals sigma) Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.22",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#chi-square-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#chi-square-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{ChiSquare}(y|\\nu) = \\frac{2^{-\\nu/2}}     {\\Gamma(\\nu / 2)} \\,\ny^{\\nu/2 - 1} \\, \\exp \\! \\left( -\\, \\frac{1}{2} \\, y \\right) . \\end{equation*}\\]\n\n\n\ny ~ chi_square(nu)\nIncrement target log probability density with chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal chi_square_lpdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lupdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal chi_square_cdf(reals y | reals nu) The Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal chi_square_lcdf(reals y | reals nu) The log of the Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lccdf(reals y | reals nu) The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR chi_square_rng(reals nu) Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#inverse-chi-square-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#inverse-chi-square-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{InvChiSquare}(y \\, | \\, \\nu) = \\frac{2^{-\\nu/2}}    {\\Gamma(\\nu\n/ 2)} \\, y^{-\\nu/2 - 1} \\, \\exp\\! \\left( \\! - \\, \\frac{1}{2} \\,\n\\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_chi_square(nu)\nIncrement target log probability density with inv_chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_chi_square_lpdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lupdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_chi_square_cdf(reals y | reals nu) The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal inv_chi_square_lcdf(reals y | reals nu) The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lccdf(reals y | reals nu) The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR inv_chi_square_rng(reals nu) Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#scaled-inverse-chi-square-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#scaled-inverse-chi-square-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{ScaledInvChiSquare}(y|\\nu,\\sigma) =\n\\frac{(\\nu / 2)^{\\nu/2}}      {\\Gamma(\\nu / 2)} \\, \\sigma^{\\nu} \\,\ny^{-(\\nu/2 + 1)} \\, \\exp \\! \\left( \\!    - \\, \\frac{1}{2} \\, \\nu \\,\n\\sigma^2 \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ scaled_inv_chi_square(nu, sigma)\nIncrement target log probability density with scaled_inv_chi_square_lupdf(y | nu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal scaled_inv_chi_square_lpdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lupdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal scaled_inv_chi_square_cdf(reals y | reals nu, reals sigma) The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.0\n \n\nreal scaled_inv_chi_square_lcdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lccdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nR scaled_inv_chi_square_rng(reals nu, reals sigma) Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#exponential-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#exponential-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If inverse scale (rate) \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Exponential}(y|\\beta) = \\beta \\, \\exp ( - \\beta \\, y ) . \\end{equation*}\\]\n\n\n\ny ~ exponential(beta)\nIncrement target log probability density with exponential_lupdf(y | beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exponential_lpdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lupdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal exponential_cdf(reals y | reals beta) The exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.0\n \n\nreal exponential_lcdf(reals y | reals beta) The log of the exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lccdf(reals y | reals beta) The log of the exponential complementary cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nR exponential_rng(reals beta) Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#gamma-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#gamma-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If the shape parameter \\(\\alpha \\in \\mathbb{R}^+\\) and the rate (or inverse scale) parameter \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Gamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\, y^{\\alpha - 1}\n\\exp(-\\beta \\, y) . \\end{equation*}\\]\nUnder the shape and rate formulation of the Gamma distribution, \\(\\mathbb{E}[y] = \\alpha / \\beta\\) and \\(\\textrm{var}[y] = \\alpha / \\beta^2\\).\n\n\n\ny ~ gamma(alpha, beta)\nIncrement target log probability density with gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gamma_lpdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lupdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gamma_cdf(reals y | reals alpha, reals beta) The cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal gamma_lcdf(reals y | reals alpha, reals beta) The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lccdf(reals y | reals alpha, reals beta) The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR gamma_rng(reals alpha, reals beta) Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#inverse-gamma-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#inverse-gamma-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{InvGamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\ y^{-(\\alpha + 1)} \\,\n\\exp \\! \\left( \\! - \\beta \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_gamma(alpha, beta)\nIncrement target log probability density with inv_gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_gamma_lpdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lupdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_gamma_cdf(reals y | reals alpha, reals beta) The inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.0\n \n\nreal inv_gamma_lcdf(reals y | reals alpha, reals beta) The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lccdf(reals y | reals alpha, reals beta) The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nR inv_gamma_rng(reals alpha, reals beta) Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#weibull-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#weibull-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*} \\text{Weibull}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Weibull}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Frechet}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ weibull(alpha, sigma)\nIncrement target log probability density with weibull_lupdf(y | alpha, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal weibull_lpdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lupdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal weibull_cdf(reals y | reals alpha, reals sigma) The Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.0\n \n\nreal weibull_lcdf(reals y | reals alpha, reals sigma) The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lccdf(reals y | reals alpha, reals sigma) The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR weibull_rng(reals alpha, reals sigma) Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#frechet-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#frechet-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Frechet}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Frechet}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Weibull}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ frechet(alpha, sigma)\nIncrement target log probability density with frechet_lupdf(y | alpha, sigma).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal frechet_lpdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lupdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal frechet_cdf(reals y | reals alpha, reals sigma) The Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.5\n \n\nreal frechet_lcdf(reals y | reals alpha, reals sigma) The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lccdf(reals y | reals alpha, reals sigma) The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR frechet_rng(reals alpha, reals sigma) Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#rayleigh-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#rayleigh-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*}\n\\text{Rayleigh}(y|\\sigma) = \\frac{y}{\\sigma^2} \\exp(-y^2 / 2\\sigma^2)\n\\!. \\end{equation*}\\]\n\n\n\ny ~ rayleigh(sigma)\nIncrement target log probability density with rayleigh_lupdf(y | sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal rayleigh_lpdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lupdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal rayleigh_cdf(real y | real sigma) The Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.0\n \n\nreal rayleigh_lcdf(real y | real sigma) The log of the Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lccdf(real y | real sigma) The log of the Rayleigh complementary cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nR rayleigh_rng(reals sigma) Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_continuous_distributions.html#log-logistic-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#log-logistic-distribution",
+    "title": "Positive Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\alpha, \\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Log-Logistic}(y|\\alpha,\\beta) =\n\\frac{\\ \\left(\\frac{\\beta}{\\alpha}\\right) \\left(\\frac{y}{\\alpha}\\right)^{\\beta-1}\\ }{\\left(1 + \\left(\\frac{y}{\\alpha}\\right)^\\beta\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ loglogistic(alpha, beta)\nIncrement target log probability density with unnormalized version of loglogistic_lpdf(y | alpha, beta)\nAvailable since 2.29\n \n\n\n\n\n \n\nreal loglogistic_lpdf(reals y | reals alpha, reals beta) The log of the log-logistic density of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nreal loglogistic_cdf(reals y | reals alpha, reals beta) The log-logistic cumulative distribution function of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nR loglogistic_rng(reals alpha, reals beta) Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.29",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/mixed_operations.html",
+    "href": "functions-reference/mixed_operations.html",
+    "title": "Mixed Operations",
+    "section": "",
+    "text": "Mixed Operations\nThese functions perform conversions between Stan containers matrix, vector, row vector and arrays.\n \n\nmatrix to_matrix(matrix m) Return the matrix m itself.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_matrix m) Return the matrix m itself.\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v) Convert the column vector v to a size(v) by 1 matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_vector v) Convert the column vector v to a size(v) by 1 matrix.\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v) Convert the row vector v to a 1 by size(v) matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_row_vector v) Convert the row vector v to a 1 by size(v) matrix.\nAvailable since 2.30\n \n\nmatrix to_matrix(matrix M, int m, int n) Convert a matrix A to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_matrix M, int m, int n) Convert a matrix A to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v, int m, int n) Convert a vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_vector v, int m, int n) Convert a vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v, int m, int n) Convert a row_vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_row_vector v, int m, int n) Convert a row vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(matrix A, int m, int n, int col_major) Convert a matrix A to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_matrix A, int m, int n, int col_major) Convert a matrix A to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v, int m, int n, int col_major) Convert a vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_vector v, int m, int n, int col_major) Convert a vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v, int m, int n, int col_major) Convert a row vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_row_vector v, int m, int n, int col_major) Convert a row vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] real a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\nmatrix to_matrix(array[] int a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(array[] complex a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] real a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\nmatrix to_matrix(array[] int a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(array[] complex a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] row_vector vs) Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.\nAvailable since 2.28\n \n\ncomplex_matrix to_matrix(array[] complex_row_vector vs) Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.\nAvailable since 2.30\n \n\nmatrix to_matrix(array[,] real a) Convert the two dimensional array a to a matrix with the same dimensions and indexing order.\nAvailable since 2.3\n \n\nmatrix to_matrix(array[,] int a) Convert the two dimensional array a to a matrix with the same dimensions and indexing order. If any of the dimensions of a are zero, the result will be a \\(0 \\times 0\\) matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(array[,] complex a ) Convert the two dimensional array a to a matrix with the same dimensions and indexing order.\nAvailable since 2.30\n \n\nvector to_vector(matrix m) Convert the matrix m to a column vector in column-major order.\nAvailable since 2.0\n \n\ncomplex_vector to_vector(complex_matrix m) Convert the matrix m to a column vector in column-major order.\nAvailable since 2.30\n \n\nvector to_vector(vector v) Return the column vector v itself.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(complex_vector v) Return the column vector v itself.\nAvailable since 2.30\n \n\nvector to_vector(row_vector v) Convert the row vector v to a column vector.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(complex_row_vector v) Convert the row vector v to a column vector.\nAvailable since 2.30\n \n\nvector to_vector(array[] real a) Convert the one-dimensional array a to a column vector.\nAvailable since 2.3\n \n\nvector to_vector(array[] int a) Convert the one-dimensional integer array a to a column vector.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(array[] complex a) Convert the one-dimensional complex array a to a column vector.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(matrix m) Convert the matrix m to a row vector in column-major order.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_matrix m) Convert the matrix m to a row vector in column-major order.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(vector v) Convert the column vector v to a row vector.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_vector v) Convert the column vector v to a row vector.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(row_vector v) Return the row vector v itself.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_row_vector v) Return the row vector v itself.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(array[] real a) Convert the one-dimensional array a to a row vector.\nAvailable since 2.3\n \n\nrow_vector to_row_vector(array[] int a) Convert the one-dimensional array a to a row vector.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(array[] complex a) Convert the one-dimensional complex array a to a row vector.\nAvailable since 2.30\n \n\narray[,] real to_array_2d(matrix m) Convert the matrix m to a two dimensional array with the same dimensions and indexing order.\nAvailable since 2.3\n \n\narray[,] complex to_array_2d(complex_matrix m) Convert the matrix m to a two dimensional array with the same dimensions and indexing order.\nAvailable since 2.30\n \n\narray[] real to_array_1d(vector v) Convert the column vector v to a one-dimensional array.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(complex_vector v) Convert the column vector v to a one-dimensional array.\nAvailable since 2.30\n \n\narray[] real to_array_1d(row_vector v) Convert the row vector v to a one-dimensional array.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(complex_row_vector v) Convert the row vector v to a one-dimensional array.\nAvailable since 2.30\n \n\narray[] real to_array_1d(matrix m) Convert the matrix m to a one-dimensional array in column-major order.\nAvailable since 2.3\n \n\narray[] real to_array_1d(complex_matrix m) Convert the matrix m to a one-dimensional array in column-major order.\nAvailable since 2.30\n \n\narray[] real to_array_1d(array[...] real a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.3\n \n\narray[] int to_array_1d(array[...] int a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(array[...] complex a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.30\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Mixed Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/mathematical_functions.html",
+    "href": "functions-reference/mathematical_functions.html",
+    "title": "Mathematical Functions",
+    "section": "",
+    "text": "This appendix provides the definition of several mathematical functions used throughout the manual.\n\n\nThe beta function, \\(\\text{B}(a, b)\\), computes the normalizing constant for the beta distribution, and is defined for \\(a &gt; 0\\) and \\(b\n&gt; 0\\) by \\[\\begin{equation*} \\text{B}(a,b) \\ = \\ \\int_0^1 u^{a - 1} (1 - u)^{b - 1} \\,\ndu \\ = \\ \\frac{\\Gamma(a) \\, \\Gamma(b)}{\\Gamma(a+b)} \\, , \\end{equation*}\\] where \\(\\Gamma(x)\\) is the Gamma function.\n\n\n\nThe incomplete beta function, \\(\\text{B}(x; a, b)\\), is defined for \\(x\n\\in [0, 1]\\) and \\(a, b \\geq 0\\) such that \\(a + b \\neq 0\\) by \\[\\begin{equation*}\n\\text{B}(x; \\, a, b) \\ = \\ \\int_0^x u^{a - 1} \\, (1 - u)^{b - 1} \\,\ndu, \\end{equation*}\\] where \\(\\text{B}(a, b)\\) is the beta function defined in appendix. If \\(x = 1\\), the incomplete beta function reduces to the beta function, \\(\\text{B}(1; a, b) = \\text{B}(a, b)\\).\nThe regularized incomplete beta function divides the incomplete beta function by the beta function, \\[\\begin{equation*} I_x(a, b) \\ = \\ \\frac{\\text{B}(x; \\,\na, b)}{B(a, b)} \\, . \\end{equation*}\\]\n\n\n\nThe gamma function, \\(\\Gamma(x)\\), is the generalization of the factorial function to continuous variables, defined so that for positive integers \\(n\\), \\[\\begin{equation*} \\Gamma(n+1) = n! \\end{equation*}\\] Generalizing to all positive numbers and non-integer negative numbers, \\[\\begin{equation*} \\Gamma(x) =\n\\int_0^{\\infty} u^{x - 1} \\exp(-u) \\, du. \\end{equation*}\\]\n\n\n\nThe digamma function \\(\\Psi\\) is the derivative of the \\(\\log \\Gamma\\) function, \\[\\begin{equation*}\n\\Psi(u) \\ = \\ \\frac{d}{d u} \\log \\Gamma(u) \\ = \\ \\frac{1}{\\Gamma(u)} \\ \\frac{d}{d u} \\Gamma(u).\n\\end{equation*}\\]",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Mathematical Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/mathematical_functions.html#beta-appendix",
+    "href": "functions-reference/mathematical_functions.html#beta-appendix",
+    "title": "Mathematical Functions",
+    "section": "",
+    "text": "The beta function, \\(\\text{B}(a, b)\\), computes the normalizing constant for the beta distribution, and is defined for \\(a &gt; 0\\) and \\(b\n&gt; 0\\) by \\[\\begin{equation*} \\text{B}(a,b) \\ = \\ \\int_0^1 u^{a - 1} (1 - u)^{b - 1} \\,\ndu \\ = \\ \\frac{\\Gamma(a) \\, \\Gamma(b)}{\\Gamma(a+b)} \\, , \\end{equation*}\\] where \\(\\Gamma(x)\\) is the Gamma function.",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Mathematical Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/mathematical_functions.html#inc-beta-appendix",
+    "href": "functions-reference/mathematical_functions.html#inc-beta-appendix",
+    "title": "Mathematical Functions",
+    "section": "",
+    "text": "The incomplete beta function, \\(\\text{B}(x; a, b)\\), is defined for \\(x\n\\in [0, 1]\\) and \\(a, b \\geq 0\\) such that \\(a + b \\neq 0\\) by \\[\\begin{equation*}\n\\text{B}(x; \\, a, b) \\ = \\ \\int_0^x u^{a - 1} \\, (1 - u)^{b - 1} \\,\ndu, \\end{equation*}\\] where \\(\\text{B}(a, b)\\) is the beta function defined in appendix. If \\(x = 1\\), the incomplete beta function reduces to the beta function, \\(\\text{B}(1; a, b) = \\text{B}(a, b)\\).\nThe regularized incomplete beta function divides the incomplete beta function by the beta function, \\[\\begin{equation*} I_x(a, b) \\ = \\ \\frac{\\text{B}(x; \\,\na, b)}{B(a, b)} \\, . \\end{equation*}\\]",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Mathematical Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/mathematical_functions.html#gamma-appendix",
+    "href": "functions-reference/mathematical_functions.html#gamma-appendix",
+    "title": "Mathematical Functions",
+    "section": "",
+    "text": "The gamma function, \\(\\Gamma(x)\\), is the generalization of the factorial function to continuous variables, defined so that for positive integers \\(n\\), \\[\\begin{equation*} \\Gamma(n+1) = n! \\end{equation*}\\] Generalizing to all positive numbers and non-integer negative numbers, \\[\\begin{equation*} \\Gamma(x) =\n\\int_0^{\\infty} u^{x - 1} \\exp(-u) \\, du. \\end{equation*}\\]",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Mathematical Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/mathematical_functions.html#digamma-appendix",
+    "href": "functions-reference/mathematical_functions.html#digamma-appendix",
+    "title": "Mathematical Functions",
+    "section": "",
+    "text": "The digamma function \\(\\Psi\\) is the derivative of the \\(\\log \\Gamma\\) function, \\[\\begin{equation*}\n\\Psi(u) \\ = \\ \\frac{d}{d u} \\log \\Gamma(u) \\ = \\ \\frac{1}{\\Gamma(u)} \\ \\frac{d}{d u} \\Gamma(u).\n\\end{equation*}\\]",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Mathematical Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/index.html",
+    "href": "functions-reference/index.html",
+    "title": "Stan Functions Reference",
+    "section": "",
+    "text": "This is the reference for the functions defined in the Stan math library and available in the Stan programming language.\nFor more information the Stan language and inference engines and how to use Stan for Bayesian inference, see\n\nthe Stan User’s Guide. The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:\nthe Stan Reference Manual. Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.\n\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Functions Reference",
+      "Stan Functions Reference"
+    ]
+  },
+  {
+    "objectID": "functions-reference/hidden_markov_models.html",
+    "href": "functions-reference/hidden_markov_models.html",
+    "title": "Hidden Markov Models",
+    "section": "",
+    "text": "An elementary first-order Hidden Markov model is a probabilistic model over \\(N\\) observations, \\(y_n\\), and \\(N\\) hidden states, \\(x_n\\), which can be fully defined by the conditional distributions \\(p(y_n \\mid x_n, \\phi)\\) and \\(p(x_n \\mid x_{n - 1}, \\phi)\\). Here we make the dependency on additional model parameters, \\(\\phi\\), explicit. When \\(x\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(x\\) out.\nWhen each state \\(x\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can take advantage of the dependency structure to marginalize \\(x\\) and compute \\(p(y \\mid \\phi)\\). We start by defining the conditional observational distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid x_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n  \\Gamma_{ij} = p(x_n = j \\mid x_{n - 1} = i, \\phi).\n\\] Each row defines a probability distribution and must therefore be a simplex (i.e. its components must add to 1). Currently, Stan only supports stationary transitions where a single transition matrix is used for all transitions. Finally we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n  \\rho_k = p(x_0 = k \\mid \\phi).\n\\]\nThe Stan functions that support this type of model are special in that the user does not explicitly pass \\(y\\) and \\(\\phi\\) as arguments. Instead, the user passes \\(\\log \\omega\\), \\(\\Gamma\\), and \\(\\rho\\), which in turn depend on \\(y\\) and \\(\\phi\\).\n\n\n \n\nreal hmm_marginal(matrix log_omega, matrix Gamma, vector rho) Returns the log probability density of \\(y\\), with \\(x_n\\) integrated out at each iteration.\nAvailable since 2.24\nThe arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.\n\nlog_omega: \\(\\log \\omega_{kn} = \\log p(y_n \\mid x_n = k, \\phi)\\), log density of each output,\nGamma: \\(\\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \\phi)\\), the transition matrix,\nrho: \\(\\rho_k = p(x_0 = k \\mid \\phi)\\), the initial state probability.\n\n \n\narray[] int hmm_latent_rng(matrix log_omega, matrix Gamma, vector rho) Returns a length \\(N\\) array of integers over \\(\\{1, ..., K\\}\\), sampled from the joint posterior distribution of the hidden states, \\(p(x \\mid \\phi, y)\\). May be only used in transformed data and generated quantities.\nAvailable since 2.24\n \n\nmatrix hmm_hidden_state_prob(matrix log_omega, matrix Gamma, vector rho) Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a \\(K \\times N\\) matrix.    The \\(n^\\mathrm{th}\\) column is a simplex of probabilities for the \\(n^\\mathrm{th}\\) variable. Moreover, let \\(A\\) be the output. Then \\(A_{ij} = p(x_j = i \\mid \\phi, y)\\).  This function may only be used in transformed data and generated quantities.\nAvailable since 2.24",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Hidden Markov Models"
+    ]
+  },
+  {
+    "objectID": "functions-reference/hidden_markov_models.html#hmm-stan-functions",
+    "href": "functions-reference/hidden_markov_models.html#hmm-stan-functions",
+    "title": "Hidden Markov Models",
+    "section": "",
+    "text": "real hmm_marginal(matrix log_omega, matrix Gamma, vector rho) Returns the log probability density of \\(y\\), with \\(x_n\\) integrated out at each iteration.\nAvailable since 2.24\nThe arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.\n\nlog_omega: \\(\\log \\omega_{kn} = \\log p(y_n \\mid x_n = k, \\phi)\\), log density of each output,\nGamma: \\(\\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \\phi)\\), the transition matrix,\nrho: \\(\\rho_k = p(x_0 = k \\mid \\phi)\\), the initial state probability.\n\n \n\narray[] int hmm_latent_rng(matrix log_omega, matrix Gamma, vector rho) Returns a length \\(N\\) array of integers over \\(\\{1, ..., K\\}\\), sampled from the joint posterior distribution of the hidden states, \\(p(x \\mid \\phi, y)\\). May be only used in transformed data and generated quantities.\nAvailable since 2.24\n \n\nmatrix hmm_hidden_state_prob(matrix log_omega, matrix Gamma, vector rho) Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a \\(K \\times N\\) matrix.    The \\(n^\\mathrm{th}\\) column is a simplex of probabilities for the \\(n^\\mathrm{th}\\) variable. Moreover, let \\(A\\) be the output. Then \\(A_{ij} = p(x_j = i \\mid \\phi, y)\\).  This function may only be used in transformed data and generated quantities.\nAvailable since 2.24",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Hidden Markov Models"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html",
+    "href": "functions-reference/embedded_laplace.html",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The embedded Laplace approximation can be used to approximate certain marginal and conditional distributions that arise in latent Gaussian models. Embedded Laplace replaces explicit sampling of (high-dimensional) Gaussian latent variables with a local Gaussian approximation. In doing so, it marginalizes out the latent Gaussian variables. Inference can then be performed on the remaining, often low-dimensional, parameters. The embedded Laplace approximation in Stan is best suited for latent Gaussian models when jointly sampling over all model parameters is expensive and the conditional posterior of the Gaussian latent variables is reasonably close to Gaussian.\nFor observed data \\(y\\), latent Gaussian variables \\(\\theta\\), and hyperparameters \\(\\phi\\), a latent Gaussian model observes the following hierarchical structure: \\[\\begin{eqnarray}\n  \\phi &\\sim& p(\\phi), \\\\\n  \\theta &\\sim& \\text{MultiNormal}(0, K(\\phi)), \\\\\n  y &\\sim& p(y \\mid \\theta, \\phi).\n\\end{eqnarray}\\] In this formulation, \\(p(y \\mid \\theta, \\phi)\\) is the data model that specifies how observations are generated conditional on \\(\\theta\\) and \\(\\phi\\). \\(K(\\phi)\\) denotes the prior covariance matrix for the latent Gaussian variables \\(\\theta\\) and is parameterized by \\(\\phi\\). The prior on \\(\\theta\\) is centered at 0, however an offset can always be added when specifying the data model \\(p(y \\mid \\theta, \\phi)\\).\nConditioning on observations \\(y\\) we obtain the joint posterior \\(p(\\phi, \\theta \\mid y) \\propto p(y \\mid \\theta, \\phi) p(\\theta |\n\\phi) p(\\phi)\\), where \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) and \\(\\phi\\) is the likelihood function. To sample from the joint posterior, we can either use a standard method, such as Markov chain Monte Carlo, or we can follow a two-step procedure:\n\nsample from the marginal posterior \\(p(\\phi \\mid y)\\),\nsample from the conditional posterior \\(p(\\theta \\mid y, \\phi)\\).\n\nIn the above procedure, neither the marginal posterior nor the conditional posterior are typically available in closed form and so they must be approximated. The marginal posterior can be written as \\(p(\\phi \\mid y) \\propto p(y \\mid \\phi) p(\\phi)\\), where \\(p(y \\mid \\phi) = \\int p(y \\mid \\phi, \\theta) p(\\theta) \\text{d}\\theta\\) is called the marginal likelihood. The Laplace method approximates \\(p(y \\mid \\phi, \\theta) p(\\theta)\\) with a normal distribution centered at the mode, \\[\n  \\theta^* = \\underset{\\theta}{\\text{argmax}} \\ \\log p(\\theta \\mid y, \\phi),\n\\] and \\(\\theta^*\\) is obtained using a numerical optimizer. The resulting Gaussian integral can be evaluated analytically to obtain an approximation to the log marginal likelihood \\(\\log \\hat p(y \\mid \\phi) \\approx \\log p(y \\mid \\phi)\\). Specifically: \\[\n  \\hat p(y \\mid \\phi) = \\frac{p(\\theta^* \\mid \\phi) p(y \\mid \\theta^*, \\phi)}{\\hat p (\\theta^* \\mid \\phi, y)}.\n\\]\nCombining this marginal likelihood with the prior in the model block, we can then sample from the marginal posterior \\(p(\\phi \\mid y)\\) using one of Stan’s algorithms. The marginal posterior is lower dimensional and likely to have a simpler geometry leading to more efficient inference. On the other hand each marginal likelihood computation is more costly, and the combined change in efficiency depends on the application.\nTo obtain posterior draws for \\(\\theta\\), we sample from the normal approximation to \\(p(\\theta \\mid y, \\phi)\\) in generated quantities. The process of iteratively sampling from \\(p(\\phi \\mid y)\\) (say, with MCMC) and then \\(p(\\theta \\mid y, \\phi)\\) produces posterior draws from the joint posterior \\(p(\\theta, \\phi \\mid y)\\).\nThe Laplace approximation is especially useful if \\(p(y \\mid \\phi, \\theta)\\) as function of \\(\\theta\\) is log-concave, e.g., in case of Poisson, binomial, negative-binomial, and Bernoulli. (The likelihood of normal model is also log concave, however when the likelihood is normal, marginalization can be performed exactly and does not required an approximation.) Stan’s embedded Laplace approximation is restricted to the case where the prior \\(p(\\theta \\mid \\phi)\\) is multivariate normal. Furthermore, the likelihood \\(p(y \\mid \\phi, \\theta)\\) must be computed using only operations which support higher-order derivatives (see section specifying the likelihood function).\nThe Laplace approximation can also be useful in generated quantities to marginalize out latent variables even if the sampling had been done using the full joint posterior.\n\n\nIn the model block, we increment target with laplace_marginal, a function that approximates the log marginal likelihood \\(\\log p(y \\mid \\phi)\\). The signature of the function is:\n \n\nreal laplace_marginal(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nwhich returns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\).\nThe embedded Laplace functions accept two functors whose user defined arguments are passed in as tuples to laplace_marginal.\n\nlikelihood_function - user-specified log likelihood whose first argument is the vector of latent Gaussian variables \\(\\theta\\). The subsequent arguments are user defined.\n\n\nreal likelihood_function(vector theta, likelihood_arguments_1, likelihood_arguments_2, ...).\n\n\nlikelihood_arguments - A tuple of arguments whose internal members are be passed to the log likelihood function. This tuple does NOT include the latent variable \\(\\theta\\).\nhessian_block_size - the block size of the Hessian of the log likelihood, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\).\ncovariance_function - A function that returns the covariance matrix of the multivariate normal prior on \\(\\theta\\).\n\n\nmatrix covariance_function(covariance_argument_1, covariance_argument_2, ...).\n\n\ncovariance_arguments A tuple of the arguments whose internal members will be passed to the the covariance function.\n\nAvailable since 2.39\nBelow we go over each argument in more detail.\n\n\n\nThe first step to use the embedded Laplace approximation is to write down a function in the functions block which returns the log likelihood \\(\\log p(y \\mid \\theta, \\phi)\\).\nThere are a few constraints on this function:\n\nThe function return type must be real.\nThe first argument must be the latent Gaussian variable \\(\\theta\\) and must have type vector.\nThe operations in the function must support higher-order automatic differentiation (AD). Most functions in Stan support higher-order AD. The exceptions are functions with specialized calls for reverse-mode AD, and these are higher-order functions (algebraic solvers, differential equation solvers, and integrators), the marginalization function for hidden Markov models (HMM) function, and the embedded Laplace approximation itself.\n\nThe base signature of the function is\nreal likelihood_function(vector theta, ...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter.\nThe tuple after likelihood_function contains the arguments that get passed to likelihood_function excluding \\(\\theta\\). For instance, if a user defined likelihood uses a real and a matrix, the likelihood function’s signature would first have a vector and then a real and matrix argument.\nreal likelihood_fun(vector theta, real a, matrix X)\nThe call to the laplace marginal would start with this likelihood and tuple holding the other likelihood arguments. We do not need to pass theta, since it is marginalized out and therefore does not appear explicitly as a passed parameter.\nreal val = laplace_marginal(likelihood_fun, (a, X), hessian_block_size, ...);\nIf the likelihood_function has only one argument, the tuple syntax is (a, ).\nAs always, users should use parameter arguments only when necessary in order to speed up differentiation. In general, we recommend marking data only arguments with the keyword data, for example,\nreal likelihood_function(vector theta, data vector x, ...)\nIn addition to the likelihood function, users must specify the block size of the Hessian, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\). The Hessian is often block diagonal and this structure can be taken advantage of for fast computation. For example, if \\(y_i\\) only depends on \\(\\theta_i\\), then the Hessian is diagonal and hessian_block_size=1,\nreal val = laplace_marginal(likelihood_fun, (a, X), 1, ...);\nOn the other hand, if the Hessian is not block diagonal, we can always set hessian_block_size=n where \\(n\\) is the size of \\(\\theta\\).\n\n\n\nThe argument covariance_function returns the prior covariance matrix \\(K\\). The signature for this function is the same as a standard stan function. It’s return type must be a matrix of size \\(n \\times n\\) where \\(n\\) is the size of \\(\\theta\\).\nmatrix covariance_function(...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter. The variables \\(\\phi\\) is implicitly defined as the collection of all non-data arguments passed to likelihood_function (excluding \\(\\theta\\)) and covariance_function.\nThe tuple after covariance_function contains the arguments that get passed to covariance_function. For instance, if a user defined covariance function uses two vectors\nmatrix cov_fun(real b, matrix Z)\nthe call to the Laplace marginal would include the covariance function and a tuple holding the covariance function arguments.\nreal val = laplace_marginal(likelihood_fun, (a, X), cov_fun, (b, Z), ...);\nIf the covariance_function has only one argument, the tuple syntax is (b, ).\n\n\n\nIt also possible to specify control parameters, which can help improve the optimization that underlies the Laplace approximation, using laplace_marginal_tol with the following signature:\n \n\nreal laplace_marginal_tol(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...),  tuple(vector, real, int, int, int, int) tolerances)\nThe final argument, tolerances, is a tuple with the following elements\ntuple(vector theta_init, real tol, int max_steps, int solver,\n      int max_steps_linesearch, int allow_fallback)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) and allows the user to tune the control parameters of the approximation.\n\ntheta_init: the initial guess for a Newton solver when finding the mode of \\(p(\\theta \\mid y, \\phi)\\). By default, it is a zero-vector.\ntol: the tolerance \\(\\epsilon\\) of the optimizer. Specifically, the optimizer stops when \\(||\\nabla \\log p(\\theta \\mid y, \\phi)|| \\le \\epsilon\\). By default, the value is \\(\\epsilon \\approx 1.49 \\times 10^{-8}\\), which is the square-root of machine precision.\nmax_num_steps: the maximum number of steps taken by the optimizer before it gives up (in which case the Metropolis proposal gets rejected). The default is 500 steps.\nsolver: choice of Newton solver. The optimizer underlying the Laplace approximation does one of three matrix decompositions to compute a Newton step. The problem determines which decomposition is numerically stable. By default (solver=1), the solver attempts a Cholesky decomposition of the negative Hessian of the log likelihood, \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\). This operation is legal if the negative Hessian is positive-definite, which will always be true when the likelihood as function of \\(\\theta\\) is log concave. If solver=2, the solver makes a Cholesky decomposition of the covariance matrix \\(K(\\phi)\\). Since a covariance matrix is always positive-definite, computing its Cholesky decomposition is always a legal operation, at least in theory. In practice, we may not be able to compute the Cholesky decomposition of the negative Hessian nor of the covariance matrix, either because it does not exist or because of numerical issues. In that case, we can use solver=3 which uses a more expensive but less specialized approach to compute a Newton step.\nmax_steps_linesearch: maximum number of steps in linesearch. The linesearch adjusts to step size to ensure that a Newton step leads to an increase in the objective function (i.e., \\(f(\\theta) = p(\\theta \\mid \\phi, y)\\)). If a standard Newton step does not improve the objective function, the step is adjusted iteratively until the objective function increases or the maximum number of steps in the linesearch is reached. By default, max_steps_linesearch=1000. Setting max_steps_linesearch=0 results in no linesearch.\nallow_fallback: If user set solver fails, this flag determines whether to fallback to the next solver. For example, if the user specifies solver=1 but the Cholesky decomposition of the negative Hessian \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\) fails, the optimizer will try solver=2 instead. By default, allow_fallback = 1 (TRUE).\n\nAvailable since 2.39\nThe embedded Laplace approximation’s options have a helper callable generate_laplace_options(int theta_size) that will generate the tuple for the user. This can be useful for quickly setting up the control parameters in the transformed data block to reuse within the model.\ntuple(vector[theta_size], real, int, int, int, int, int) laplace_ops = generate_laplace_options(theta_size);\n// Modify solver type\nlaplace_ops.5 = 2;\n// Turn off fallthrough\nlaplace_ops.7 = 0;\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(int dimension)\nCreate a default laplace options tuple for a theta_init of size dimension.\nAvailable since 2.39\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(vector theta_init)\nCreate a default Laplace options tuple containing theta_init.\nAvailable since 2.39\n\n\n\nIn generated quantities, it is possible to sample from the Laplace approximation of \\(p(\\theta \\mid \\phi, y)\\) using laplace_latent_rng. The signature for laplace_latent_rng follows closely the signature for laplace_marginal:\n \n\nvector laplace_latent_rng(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nSamples from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\).\nAvailable since 2.39\nOnce again, it is possible to specify control parameters:\n \n\nvector laplace_latent_tol_rng(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...), tuple(vector, real, int, int, int, int) tolerances) Samples from the approximate conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nStan provides convenient wrappers for the embedded Laplace approximation when applied to latent Gaussian models with certain likelihoods arising from some common data models. With this wrapper, the likelihood is pre-specified and does not need to be specified by the user. The selection of supported likelihoods is currently narrow and expected to grow. The wrappers exist for the user’s convenience but are not more computationally efficient than specifying log likelihoods in the functions block.\n\n\nGiven count data, with each observed count \\(y_i\\) associated with a group \\(g(i)\\) and a corresponding latent variable \\(\\theta_{g(i)}\\), and a Poisson model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Poisson} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)})),\n\\] where \\(m_{g(i)}\\) acts as an offset for \\(\\theta_{g(i)}\\). This can also be interpreted as a prior mean on \\(\\theta_{g(i)}\\). The arguments required to compute this likelihood are:\n\ny: an array of counts.\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments.\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covaraince_arguments, tolerances).\nAvailable since 2.39\nThe signatures for the embedded Laplace approximation function with a Poisson likelihood are\n \n real laplace_marginal_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n real laplace_marginal_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n vector laplace_latent_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) Returns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nThe negative Binomial distribution generalizes the Poisson distribution by introducing the dispersion parameter \\(\\eta\\). The corresponding likelihood is then \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{NegBinomial2} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)}), \\eta).\n\\] Here we use the alternative parameterization implemented in Stan, meaning that \\[\n\\mathbb E(y_i) = \\exp (\\theta_{g(i)} + m_{g(i)}), \\\\\n\\text{Var}(y_i) = \\mathbb E(y_i) + \\frac{(\\mathbb E(y_i))^2}{\\eta}.\n\\] The arguments for the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\neta: the overdispersion parameter.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a negative Binomial likelihood are\n \n\nreal laplace_marginal_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n\nreal laplace_marginal_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, function covariance_function, data int hessian_block_size, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nGiven binary outcome \\(y_i \\in \\{0, 1\\}\\) and Bernoulli model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Bernoulli} (y_i \\mid \\text{logit}^{-1}(\\theta_{g(i)} + m_{g(i)})).\n\\] The arguments of the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a Bernoulli likelihood are\n \n\nreal laplace_marginal_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters.\nAvailable since 2.39\n \n\nreal laplace_marginal_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link, and lets the user tune the control parameters of the approximation.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html#approximating-the-log-marginal-likelihood-log-py-mid-phi",
+    "href": "functions-reference/embedded_laplace.html#approximating-the-log-marginal-likelihood-log-py-mid-phi",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "In the model block, we increment target with laplace_marginal, a function that approximates the log marginal likelihood \\(\\log p(y \\mid \\phi)\\). The signature of the function is:\n \n\nreal laplace_marginal(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nwhich returns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\).\nThe embedded Laplace functions accept two functors whose user defined arguments are passed in as tuples to laplace_marginal.\n\nlikelihood_function - user-specified log likelihood whose first argument is the vector of latent Gaussian variables \\(\\theta\\). The subsequent arguments are user defined.\n\n\nreal likelihood_function(vector theta, likelihood_arguments_1, likelihood_arguments_2, ...).\n\n\nlikelihood_arguments - A tuple of arguments whose internal members are be passed to the log likelihood function. This tuple does NOT include the latent variable \\(\\theta\\).\nhessian_block_size - the block size of the Hessian of the log likelihood, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\).\ncovariance_function - A function that returns the covariance matrix of the multivariate normal prior on \\(\\theta\\).\n\n\nmatrix covariance_function(covariance_argument_1, covariance_argument_2, ...).\n\n\ncovariance_arguments A tuple of the arguments whose internal members will be passed to the the covariance function.\n\nAvailable since 2.39\nBelow we go over each argument in more detail.",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html#laplace-likelihood_spec",
+    "href": "functions-reference/embedded_laplace.html#laplace-likelihood_spec",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The first step to use the embedded Laplace approximation is to write down a function in the functions block which returns the log likelihood \\(\\log p(y \\mid \\theta, \\phi)\\).\nThere are a few constraints on this function:\n\nThe function return type must be real.\nThe first argument must be the latent Gaussian variable \\(\\theta\\) and must have type vector.\nThe operations in the function must support higher-order automatic differentiation (AD). Most functions in Stan support higher-order AD. The exceptions are functions with specialized calls for reverse-mode AD, and these are higher-order functions (algebraic solvers, differential equation solvers, and integrators), the marginalization function for hidden Markov models (HMM) function, and the embedded Laplace approximation itself.\n\nThe base signature of the function is\nreal likelihood_function(vector theta, ...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter.\nThe tuple after likelihood_function contains the arguments that get passed to likelihood_function excluding \\(\\theta\\). For instance, if a user defined likelihood uses a real and a matrix, the likelihood function’s signature would first have a vector and then a real and matrix argument.\nreal likelihood_fun(vector theta, real a, matrix X)\nThe call to the laplace marginal would start with this likelihood and tuple holding the other likelihood arguments. We do not need to pass theta, since it is marginalized out and therefore does not appear explicitly as a passed parameter.\nreal val = laplace_marginal(likelihood_fun, (a, X), hessian_block_size, ...);\nIf the likelihood_function has only one argument, the tuple syntax is (a, ).\nAs always, users should use parameter arguments only when necessary in order to speed up differentiation. In general, we recommend marking data only arguments with the keyword data, for example,\nreal likelihood_function(vector theta, data vector x, ...)\nIn addition to the likelihood function, users must specify the block size of the Hessian, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\). The Hessian is often block diagonal and this structure can be taken advantage of for fast computation. For example, if \\(y_i\\) only depends on \\(\\theta_i\\), then the Hessian is diagonal and hessian_block_size=1,\nreal val = laplace_marginal(likelihood_fun, (a, X), 1, ...);\nOn the other hand, if the Hessian is not block diagonal, we can always set hessian_block_size=n where \\(n\\) is the size of \\(\\theta\\).",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html#specifying-the-covariance-function",
+    "href": "functions-reference/embedded_laplace.html#specifying-the-covariance-function",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "The argument covariance_function returns the prior covariance matrix \\(K\\). The signature for this function is the same as a standard stan function. It’s return type must be a matrix of size \\(n \\times n\\) where \\(n\\) is the size of \\(\\theta\\).\nmatrix covariance_function(...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter. The variables \\(\\phi\\) is implicitly defined as the collection of all non-data arguments passed to likelihood_function (excluding \\(\\theta\\)) and covariance_function.\nThe tuple after covariance_function contains the arguments that get passed to covariance_function. For instance, if a user defined covariance function uses two vectors\nmatrix cov_fun(real b, matrix Z)\nthe call to the Laplace marginal would include the covariance function and a tuple holding the covariance function arguments.\nreal val = laplace_marginal(likelihood_fun, (a, X), cov_fun, (b, Z), ...);\nIf the covariance_function has only one argument, the tuple syntax is (b, ).",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html#control-parameters",
+    "href": "functions-reference/embedded_laplace.html#control-parameters",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "It also possible to specify control parameters, which can help improve the optimization that underlies the Laplace approximation, using laplace_marginal_tol with the following signature:\n \n\nreal laplace_marginal_tol(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...),  tuple(vector, real, int, int, int, int) tolerances)\nThe final argument, tolerances, is a tuple with the following elements\ntuple(vector theta_init, real tol, int max_steps, int solver,\n      int max_steps_linesearch, int allow_fallback)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) and allows the user to tune the control parameters of the approximation.\n\ntheta_init: the initial guess for a Newton solver when finding the mode of \\(p(\\theta \\mid y, \\phi)\\). By default, it is a zero-vector.\ntol: the tolerance \\(\\epsilon\\) of the optimizer. Specifically, the optimizer stops when \\(||\\nabla \\log p(\\theta \\mid y, \\phi)|| \\le \\epsilon\\). By default, the value is \\(\\epsilon \\approx 1.49 \\times 10^{-8}\\), which is the square-root of machine precision.\nmax_num_steps: the maximum number of steps taken by the optimizer before it gives up (in which case the Metropolis proposal gets rejected). The default is 500 steps.\nsolver: choice of Newton solver. The optimizer underlying the Laplace approximation does one of three matrix decompositions to compute a Newton step. The problem determines which decomposition is numerically stable. By default (solver=1), the solver attempts a Cholesky decomposition of the negative Hessian of the log likelihood, \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\). This operation is legal if the negative Hessian is positive-definite, which will always be true when the likelihood as function of \\(\\theta\\) is log concave. If solver=2, the solver makes a Cholesky decomposition of the covariance matrix \\(K(\\phi)\\). Since a covariance matrix is always positive-definite, computing its Cholesky decomposition is always a legal operation, at least in theory. In practice, we may not be able to compute the Cholesky decomposition of the negative Hessian nor of the covariance matrix, either because it does not exist or because of numerical issues. In that case, we can use solver=3 which uses a more expensive but less specialized approach to compute a Newton step.\nmax_steps_linesearch: maximum number of steps in linesearch. The linesearch adjusts to step size to ensure that a Newton step leads to an increase in the objective function (i.e., \\(f(\\theta) = p(\\theta \\mid \\phi, y)\\)). If a standard Newton step does not improve the objective function, the step is adjusted iteratively until the objective function increases or the maximum number of steps in the linesearch is reached. By default, max_steps_linesearch=1000. Setting max_steps_linesearch=0 results in no linesearch.\nallow_fallback: If user set solver fails, this flag determines whether to fallback to the next solver. For example, if the user specifies solver=1 but the Cholesky decomposition of the negative Hessian \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\) fails, the optimizer will try solver=2 instead. By default, allow_fallback = 1 (TRUE).\n\nAvailable since 2.39\nThe embedded Laplace approximation’s options have a helper callable generate_laplace_options(int theta_size) that will generate the tuple for the user. This can be useful for quickly setting up the control parameters in the transformed data block to reuse within the model.\ntuple(vector[theta_size], real, int, int, int, int, int) laplace_ops = generate_laplace_options(theta_size);\n// Modify solver type\nlaplace_ops.5 = 2;\n// Turn off fallthrough\nlaplace_ops.7 = 0;\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(int dimension)\nCreate a default laplace options tuple for a theta_init of size dimension.\nAvailable since 2.39\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(vector theta_init)\nCreate a default Laplace options tuple containing theta_init.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html#sample-from-the-approximate-conditional-hatptheta-mid-y-phi",
+    "href": "functions-reference/embedded_laplace.html#sample-from-the-approximate-conditional-hatptheta-mid-y-phi",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "In generated quantities, it is possible to sample from the Laplace approximation of \\(p(\\theta \\mid \\phi, y)\\) using laplace_latent_rng. The signature for laplace_latent_rng follows closely the signature for laplace_marginal:\n \n\nvector laplace_latent_rng(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nSamples from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\).\nAvailable since 2.39\nOnce again, it is possible to specify control parameters:\n \n\nvector laplace_latent_tol_rng(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...), tuple(vector, real, int, int, int, int) tolerances) Samples from the approximate conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/embedded_laplace.html#built-in-laplace-marginal-likelihood-functions",
+    "href": "functions-reference/embedded_laplace.html#built-in-laplace-marginal-likelihood-functions",
+    "title": "Embedded Laplace Approximation",
+    "section": "",
+    "text": "Stan provides convenient wrappers for the embedded Laplace approximation when applied to latent Gaussian models with certain likelihoods arising from some common data models. With this wrapper, the likelihood is pre-specified and does not need to be specified by the user. The selection of supported likelihoods is currently narrow and expected to grow. The wrappers exist for the user’s convenience but are not more computationally efficient than specifying log likelihoods in the functions block.\n\n\nGiven count data, with each observed count \\(y_i\\) associated with a group \\(g(i)\\) and a corresponding latent variable \\(\\theta_{g(i)}\\), and a Poisson model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Poisson} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)})),\n\\] where \\(m_{g(i)}\\) acts as an offset for \\(\\theta_{g(i)}\\). This can also be interpreted as a prior mean on \\(\\theta_{g(i)}\\). The arguments required to compute this likelihood are:\n\ny: an array of counts.\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments.\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covaraince_arguments, tolerances).\nAvailable since 2.39\nThe signatures for the embedded Laplace approximation function with a Poisson likelihood are\n \n real laplace_marginal_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n real laplace_marginal_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n vector laplace_latent_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) Returns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nThe negative Binomial distribution generalizes the Poisson distribution by introducing the dispersion parameter \\(\\eta\\). The corresponding likelihood is then \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{NegBinomial2} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)}), \\eta).\n\\] Here we use the alternative parameterization implemented in Stan, meaning that \\[\n\\mathbb E(y_i) = \\exp (\\theta_{g(i)} + m_{g(i)}), \\\\\n\\text{Var}(y_i) = \\mathbb E(y_i) + \\frac{(\\mathbb E(y_i))^2}{\\eta}.\n\\] The arguments for the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\neta: the overdispersion parameter.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a negative Binomial likelihood are\n \n\nreal laplace_marginal_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n\nreal laplace_marginal_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, function covariance_function, data int hessian_block_size, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nGiven binary outcome \\(y_i \\in \\{0, 1\\}\\) and Bernoulli model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Bernoulli} (y_i \\mid \\text{logit}^{-1}(\\theta_{g(i)} + m_{g(i)})).\n\\] The arguments of the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a Bernoulli likelihood are\n \n\nreal laplace_marginal_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters.\nAvailable since 2.39\n \n\nreal laplace_marginal_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link, and lets the user tune the control parameters of the approximation.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "functions-reference/deprecated_functions.html",
+    "href": "functions-reference/deprecated_functions.html",
+    "title": "Deprecated Functions",
+    "section": "",
+    "text": "This appendix lists currently deprecated functionality along with how to replace it.\nStarting in Stan 2.29, deprecated functions with drop in replacements (such as the renaming of get_lp or multiply_log) will be removed 3 versions later e.g., functions deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in Removed Functions. The Stan compiler can automatically update these on the behalf of the user for the entire deprecation window and at least one version following the removal.\n\n\nDeprecated: Using / with two integer arguments is interpreted as integer floor division, such that\n\\[ 1 / 2 = 0 \\]\nThis is deprecated due to its confusion with real-valued division, where\n\\[ 1.0 / 2.0 = 0.5 \\]\nReplacement: Use the integer division operator operator%/% instead.\n\n\n\nThese ODE integrator functions have been replaced by those described in Ordinary Differential Equation (ODE) Solvers.\n\n\nA system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:\narray[] real ode(real time, array[] real state, array[] real theta,\n                 array[] real x_r, array[] int x_i);\nThe ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\ntheta, parameter values used to evaluate the ODE system\nx_r, data values used to evaluate the ODE system\nx_i, integer data values used to evaluate the ODE system.\n\nThe ODE system function separates parameter values, theta, from data values, x_r, for efficiency in computing the gradients of the ODE.\n\n\n\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.23, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.23, deprecated in 2.24\n\n\n\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.\n\node: function literal referring to a function specifying the system of differential equations with signature:\n\n(real, array[] real, array[] real, data array[] real, data array[] int):array[] real\nThe arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,\n\ninitial_state: initial state, type array[] real,\ninitial_time: initial time, type int or real,\ntimes: solution times, type array[] real,\ntheta: parameters, type array[] real,\ndata x_r: real data, type array[] real, data only, and\ndata x_i: integer data, type array[] int, data only.\n\nFor more fine-grained control of the ODE solvers, these parameters can also be provided:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\ndata max_num_steps: maximum number of steps to take in the ODE solver, type int, data only.\n\n\n\nThe return value for the ODE solvers is an array of type array[,] real, with values consisting of solutions at the specified times.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,\nsolution times and number of rows of the return value of the solver,\nparameters, real data and integer data passed to the solver will be passed to the system function\n\n\n\n\n\n\nThese algebraic solver functions have been replaced by those described in Algebraic Equation Solvers..\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:\n vector algebra_system(vector y, vector theta,\n                              data array[] real x_r, array[] int x_i)\nThe algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.\nThe argument of this function are:\n\ny, the unknowns we wish to solve for\ntheta, parameter values used to evaluate the algebraic system\nx_r, data values used to evaluate the algebraic system\nx_i, integer data used to evaluate the algebraic system\n\nThe algebraic system function separates parameter values, theta, from data values, x_r, for efficiency in propagating the derivatives through the algebraic system.\n\n\n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.\nAvailable since 2.17, deprecated in 2.31\n \n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.\nAvailable since 2.17, deprecated in 2.31\nNote: In future releases, the function algebra_solver will be deprecated and replaced with algebra_solver_powell.\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.24, deprecated in 2.31\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.24, deprecated in 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, vector, array[] real, array[] int):vector. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\ntheta: parameters only, type vector,\nx_r: real data only, type array[] real, and\nx_i: integer data only, type array[] int.\n\nFor more fine-grained control of the algebraic solver, these parameters can also be provided:\n\nrel_tol: relative tolerance for the algebraic solver, type real, data only,\nfunction_tol: function tolerance for the algebraic solver, type real, data only,\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only.\n\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0.\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\nThe parameters, real data, and integer data will be passed from the solver directly to the system function.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Deprecated Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/deprecated_functions.html#integer-division-with-operator",
+    "href": "functions-reference/deprecated_functions.html#integer-division-with-operator",
+    "title": "Deprecated Functions",
+    "section": "",
+    "text": "Deprecated: Using / with two integer arguments is interpreted as integer floor division, such that\n\\[ 1 / 2 = 0 \\]\nThis is deprecated due to its confusion with real-valued division, where\n\\[ 1.0 / 2.0 = 0.5 \\]\nReplacement: Use the integer division operator operator%/% instead.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Deprecated Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/deprecated_functions.html#functions-old-ode-solver",
+    "href": "functions-reference/deprecated_functions.html#functions-old-ode-solver",
+    "title": "Deprecated Functions",
+    "section": "",
+    "text": "These ODE integrator functions have been replaced by those described in Ordinary Differential Equation (ODE) Solvers.\n\n\nA system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:\narray[] real ode(real time, array[] real state, array[] real theta,\n                 array[] real x_r, array[] int x_i);\nThe ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\ntheta, parameter values used to evaluate the ODE system\nx_r, data values used to evaluate the ODE system\nx_i, integer data values used to evaluate the ODE system.\n\nThe ODE system function separates parameter values, theta, from data values, x_r, for efficiency in computing the gradients of the ODE.\n\n\n\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.23, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.23, deprecated in 2.24\n\n\n\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.\n\node: function literal referring to a function specifying the system of differential equations with signature:\n\n(real, array[] real, array[] real, data array[] real, data array[] int):array[] real\nThe arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,\n\ninitial_state: initial state, type array[] real,\ninitial_time: initial time, type int or real,\ntimes: solution times, type array[] real,\ntheta: parameters, type array[] real,\ndata x_r: real data, type array[] real, data only, and\ndata x_i: integer data, type array[] int, data only.\n\nFor more fine-grained control of the ODE solvers, these parameters can also be provided:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\ndata max_num_steps: maximum number of steps to take in the ODE solver, type int, data only.\n\n\n\nThe return value for the ODE solvers is an array of type array[,] real, with values consisting of solutions at the specified times.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,\nsolution times and number of rows of the return value of the solver,\nparameters, real data and integer data passed to the solver will be passed to the system function",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Deprecated Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/deprecated_functions.html#functions-old-algebra-solver",
+    "href": "functions-reference/deprecated_functions.html#functions-old-algebra-solver",
+    "title": "Deprecated Functions",
+    "section": "",
+    "text": "These algebraic solver functions have been replaced by those described in Algebraic Equation Solvers..\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:\n vector algebra_system(vector y, vector theta,\n                              data array[] real x_r, array[] int x_i)\nThe algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.\nThe argument of this function are:\n\ny, the unknowns we wish to solve for\ntheta, parameter values used to evaluate the algebraic system\nx_r, data values used to evaluate the algebraic system\nx_i, integer data used to evaluate the algebraic system\n\nThe algebraic system function separates parameter values, theta, from data values, x_r, for efficiency in propagating the derivatives through the algebraic system.\n\n\n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.\nAvailable since 2.17, deprecated in 2.31\n \n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.\nAvailable since 2.17, deprecated in 2.31\nNote: In future releases, the function algebra_solver will be deprecated and replaced with algebra_solver_powell.\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.24, deprecated in 2.31\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.24, deprecated in 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, vector, array[] real, array[] int):vector. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\ntheta: parameters only, type vector,\nx_r: real data only, type array[] real, and\nx_i: integer data only, type array[] int.\n\nFor more fine-grained control of the algebraic solver, these parameters can also be provided:\n\nrel_tol: relative tolerance for the algebraic solver, type real, data only,\nfunction_tol: function tolerance for the algebraic solver, type real, data only,\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only.\n\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0.\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\nThe parameters, real data, and integer data will be passed from the solver directly to the system function.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Deprecated Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/correlation_matrix_distributions.html",
+    "href": "functions-reference/correlation_matrix_distributions.html",
+    "title": "Correlation Matrix Distributions",
+    "section": "",
+    "text": "The correlation matrix distributions have support on the (Cholesky factors of) correlation matrices. A Cholesky factor \\(L\\) for a \\(K\n\\times K\\) correlation matrix \\(\\Sigma\\) of dimension \\(K\\) has rows of unit length so that the diagonal of \\(L L^{\\top}\\) is the unit \\(K\\)-vector. Even though models are usually conceptualized in terms of correlation matrices, it is better to operationalize them in terms of their Cholesky factors. If you are interested in the posterior distribution of the correlations, you can recover them in the generated quantities block via\n generated quantities {\n   corr_matrix[K] Sigma;\n   Sigma = multiply_lower_tri_self_transpose(L);\n }\n\n\n\n\nFor \\(\\eta &gt; 0\\), if \\(\\Sigma\\) a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*}\n\\text{LkjCorr}(\\Sigma|\\eta) \\propto \\det \\left( \\Sigma \\right)^{(\\eta\n- 1)}. \\end{equation*}\\] The expectation is the identity matrix for any positive value of the shape parameter \\(\\eta\\), which can be interpreted like the shape parameter of a symmetric beta distribution:\n\nif \\(\\eta = 1\\), then the density is uniform over correlation matrices of order \\(K\\);\nif \\(\\eta &gt; 1\\), the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger \\(\\eta\\); and\nfor \\(0 &lt; \\eta &lt; 1\\), the density has a trough at the identity matrix.\nif \\(\\eta\\) were an unknown parameter, the Jeffreys prior is proportional to \\(\\sqrt{2\\sum_{k=1}^{K-1}\\left(\n\\psi_1\\left(\\eta+\\frac{K-k-1}{2}\\right) -   2\\psi_1\\left(2\\eta+K-k-1\n\\right)\\right)}\\), where \\(\\psi_1()\\) is the trigamma function\n\nSee (Lewandowski, Kurowicka, and Joe 2009) for definitions. However, it is much better computationally to work directly with the Cholesky factor of \\(\\Sigma\\), so this distribution should never be explicitly used in practice.\n\n\n\ny ~ lkj_corr(eta)\nIncrement target log probability density with lkj_corr_lupdf(y | eta).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal lkj_corr_lpdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta. lkj_corr_cholesky_lpdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.12\n \n\nreal lkj_corr_lupdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. lkj_corr_cholesky_lupdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.25\n \n\nmatrix lkj_corr_rng(int K, real eta) Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n\n\n\n\nStan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if L is a Cholesky factor of a correlation matrix, then\n L ~ lkj_corr_cholesky(2.0); # implies L * L' ~ lkj_corr(2.0);\nBecause Stan requires models to have support on all valid constrained parameters, L will almost always1 be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,\n parameters {   cholesky_factor_corr[K] L;   # rather than corr_matrix[K] Sigma;   // ...\n\n\nFor \\(\\eta &gt; 0\\), if \\(L\\) is a \\(K \\times K\\) lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*} \\text{LkjCholesky}(L|\\eta)\n\\propto \\left|J\\right|\\det(L L^\\top)^{(\\eta - 1)} = \\prod_{k=2}^K\nL_{kk}^{K-k+2\\eta-2}. \\end{equation*}\\] See the previous section for details on interpreting the shape parameter \\(\\eta\\). Note that even if \\(\\eta=1\\), it is still essential to evaluate the density function because the density of \\(L\\) is not constant, regardless of the value of \\(\\eta\\), even though the density of \\(LL^\\top\\) is constant iff \\(\\eta=1\\).\nA lower triangular \\(L\\) is a Cholesky factor for a correlation matrix if and only if \\(L_{k,k} &gt; 0\\) for \\(k \\in 1{:}K\\) and each row \\(L_k\\) has unit Euclidean length.\n\n\n\nL ~ lkj_corr_cholesky(eta)\nIncrement target log probability density with lkj_corr_cholesky_lupdf(L | eta).\nAvailable since 2.4\n \n\n\n\n\n \n\nreal lkj_corr_cholesky_lpdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta\nAvailable since 2.12\n \n\nreal lkj_corr_cholesky_lupdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms\nAvailable since 2.25\n \n\nmatrix lkj_corr_cholesky_rng(int K, real eta) Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.4",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/correlation_matrix_distributions.html#lkj-correlation",
+    "href": "functions-reference/correlation_matrix_distributions.html#lkj-correlation",
+    "title": "Correlation Matrix Distributions",
+    "section": "",
+    "text": "For \\(\\eta &gt; 0\\), if \\(\\Sigma\\) a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*}\n\\text{LkjCorr}(\\Sigma|\\eta) \\propto \\det \\left( \\Sigma \\right)^{(\\eta\n- 1)}. \\end{equation*}\\] The expectation is the identity matrix for any positive value of the shape parameter \\(\\eta\\), which can be interpreted like the shape parameter of a symmetric beta distribution:\n\nif \\(\\eta = 1\\), then the density is uniform over correlation matrices of order \\(K\\);\nif \\(\\eta &gt; 1\\), the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger \\(\\eta\\); and\nfor \\(0 &lt; \\eta &lt; 1\\), the density has a trough at the identity matrix.\nif \\(\\eta\\) were an unknown parameter, the Jeffreys prior is proportional to \\(\\sqrt{2\\sum_{k=1}^{K-1}\\left(\n\\psi_1\\left(\\eta+\\frac{K-k-1}{2}\\right) -   2\\psi_1\\left(2\\eta+K-k-1\n\\right)\\right)}\\), where \\(\\psi_1()\\) is the trigamma function\n\nSee (Lewandowski, Kurowicka, and Joe 2009) for definitions. However, it is much better computationally to work directly with the Cholesky factor of \\(\\Sigma\\), so this distribution should never be explicitly used in practice.\n\n\n\ny ~ lkj_corr(eta)\nIncrement target log probability density with lkj_corr_lupdf(y | eta).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal lkj_corr_lpdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta. lkj_corr_cholesky_lpdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.12\n \n\nreal lkj_corr_lupdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. lkj_corr_cholesky_lupdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.25\n \n\nmatrix lkj_corr_rng(int K, real eta) Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/correlation_matrix_distributions.html#cholesky-lkj-correlation-distribution",
+    "href": "functions-reference/correlation_matrix_distributions.html#cholesky-lkj-correlation-distribution",
+    "title": "Correlation Matrix Distributions",
+    "section": "",
+    "text": "Stan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if L is a Cholesky factor of a correlation matrix, then\n L ~ lkj_corr_cholesky(2.0); # implies L * L' ~ lkj_corr(2.0);\nBecause Stan requires models to have support on all valid constrained parameters, L will almost always1 be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,\n parameters {   cholesky_factor_corr[K] L;   # rather than corr_matrix[K] Sigma;   // ...\n\n\nFor \\(\\eta &gt; 0\\), if \\(L\\) is a \\(K \\times K\\) lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*} \\text{LkjCholesky}(L|\\eta)\n\\propto \\left|J\\right|\\det(L L^\\top)^{(\\eta - 1)} = \\prod_{k=2}^K\nL_{kk}^{K-k+2\\eta-2}. \\end{equation*}\\] See the previous section for details on interpreting the shape parameter \\(\\eta\\). Note that even if \\(\\eta=1\\), it is still essential to evaluate the density function because the density of \\(L\\) is not constant, regardless of the value of \\(\\eta\\), even though the density of \\(LL^\\top\\) is constant iff \\(\\eta=1\\).\nA lower triangular \\(L\\) is a Cholesky factor for a correlation matrix if and only if \\(L_{k,k} &gt; 0\\) for \\(k \\in 1{:}K\\) and each row \\(L_k\\) has unit Euclidean length.\n\n\n\nL ~ lkj_corr_cholesky(eta)\nIncrement target log probability density with lkj_corr_cholesky_lupdf(L | eta).\nAvailable since 2.4\n \n\n\n\n\n \n\nreal lkj_corr_cholesky_lpdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta\nAvailable since 2.12\n \n\nreal lkj_corr_cholesky_lupdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms\nAvailable since 2.25\n \n\nmatrix lkj_corr_cholesky_rng(int K, real eta) Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.4",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/correlation_matrix_distributions.html#footnotes",
+    "href": "functions-reference/correlation_matrix_distributions.html#footnotes",
+    "title": "Correlation Matrix Distributions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nIt is possible to build up a valid L within Stan, but that would then require Jacobian adjustments to imply the intended posterior.↩︎",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/continuous_distributions_on_0_1.html",
+    "href": "functions-reference/continuous_distributions_on_0_1.html",
+    "title": "Continuous Distributions on [0, 1]",
+    "section": "",
+    "text": "The continuous distributions with outcomes in the interval \\([0,1]\\) are used to characterized bounded quantities, including probabilities.\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(\\theta \\in (0,1)\\), \\[\\begin{equation*} \\text{Beta}(\\theta|\\alpha,\\beta) =\n\\frac{1}{\\mathrm{B}(\\alpha,\\beta)} \\, \\theta^{\\alpha - 1} \\, (1 -\n\\theta)^{\\beta - 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, \\(\\alpha, \\beta &gt;\n0\\).\n\n\n\ntheta ~ beta(alpha, beta)\nIncrement target log probability density with beta_lupdf(theta | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_lpdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lupdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_cdf(reals theta | reals alpha, reals beta) The beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.0\n \n\nreal beta_lcdf(reals theta | reals alpha, reals beta) The log of the beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lccdf(reals theta | reals alpha, reals beta) The log of the beta complementary cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nR beta_rng(reals alpha, reals beta) Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in (0, 1)\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(\\theta\n\\in (0,1)\\), \\[\\begin{equation*} \\mathrm{Beta\\_Proportion}(\\theta|\\mu,\\kappa) =\n\\frac{1}{\\mathrm{B}(\\mu \\kappa, (1 - \\mu) \\kappa)} \\,\n\\theta^{\\mu\\kappa - 1} \\, (1 - \\theta)^{(1 - \\mu)\\kappa- 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires \\(\\mu \\in (0, 1)\\) and strictly positive parameter, \\(\\kappa &gt; 0\\).\n\n\n\ntheta ~ beta_proportion(mu, kappa)\nIncrement target log probability density with beta_proportion_lupdf(theta | mu, kappa).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal beta_proportion_lpdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.19\n \n\nreal beta_proportion_lupdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_proportion_lcdf(reals theta | reals mu, reals kappa) The log of the beta_proportion cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nreal beta_proportion_lccdf(reals theta | reals mu, reals kappa) The log of the beta_proportion complementary cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nR beta_proportion_rng(reals mu, reals kappa) Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Continuous Distributions on [0, 1]"
+    ]
+  },
+  {
+    "objectID": "functions-reference/continuous_distributions_on_0_1.html#beta-distribution",
+    "href": "functions-reference/continuous_distributions_on_0_1.html#beta-distribution",
+    "title": "Continuous Distributions on [0, 1]",
+    "section": "",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(\\theta \\in (0,1)\\), \\[\\begin{equation*} \\text{Beta}(\\theta|\\alpha,\\beta) =\n\\frac{1}{\\mathrm{B}(\\alpha,\\beta)} \\, \\theta^{\\alpha - 1} \\, (1 -\n\\theta)^{\\beta - 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, \\(\\alpha, \\beta &gt;\n0\\).\n\n\n\ntheta ~ beta(alpha, beta)\nIncrement target log probability density with beta_lupdf(theta | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_lpdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lupdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_cdf(reals theta | reals alpha, reals beta) The beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.0\n \n\nreal beta_lcdf(reals theta | reals alpha, reals beta) The log of the beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lccdf(reals theta | reals alpha, reals beta) The log of the beta complementary cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nR beta_rng(reals alpha, reals beta) Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Continuous Distributions on [0, 1]"
+    ]
+  },
+  {
+    "objectID": "functions-reference/continuous_distributions_on_0_1.html#beta-proportion-distribution",
+    "href": "functions-reference/continuous_distributions_on_0_1.html#beta-proportion-distribution",
+    "title": "Continuous Distributions on [0, 1]",
+    "section": "",
+    "text": "If \\(\\mu \\in (0, 1)\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(\\theta\n\\in (0,1)\\), \\[\\begin{equation*} \\mathrm{Beta\\_Proportion}(\\theta|\\mu,\\kappa) =\n\\frac{1}{\\mathrm{B}(\\mu \\kappa, (1 - \\mu) \\kappa)} \\,\n\\theta^{\\mu\\kappa - 1} \\, (1 - \\theta)^{(1 - \\mu)\\kappa- 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires \\(\\mu \\in (0, 1)\\) and strictly positive parameter, \\(\\kappa &gt; 0\\).\n\n\n\ntheta ~ beta_proportion(mu, kappa)\nIncrement target log probability density with beta_proportion_lupdf(theta | mu, kappa).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal beta_proportion_lpdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.19\n \n\nreal beta_proportion_lupdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_proportion_lcdf(reals theta | reals mu, reals kappa) The log of the beta_proportion cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nreal beta_proportion_lccdf(reals theta | reals mu, reals kappa) The log of the beta_proportion complementary cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nR beta_proportion_rng(reals mu, reals kappa) Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Continuous Distributions on [0, 1]"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html",
+    "href": "functions-reference/complex_matrix_operations.html",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.\n\n\nIf an expression e can be assigned to a variable of type T, then it can be used as an argument to a function that is specified to take arguments of type T. For instance, sqrt(real) is specified to take a real argument, but an integer expression such as 2 + 2 of type int can be passed to sqrt, so that sqrt(2 + 2) is well defined. This works by promoting the integer expression 2 + 2 to be of real type.\nThe rules for promotion in Stan are simple:\n\nint may be promoted to real,\nreal may be promoted to complex,\nvector can be promoted to complex_vector,\nrow_vector can be promoted to complex_row_vector,\nmatrix can be promoted to complex_matrix,\nif T can be promoted to U and U can be promoted to V, then T can be promoted to V (transitive), and\nif T can be promoted to U, then T[] can be promoted to U[] (covariant).\n\n\n\n\nWhen a function is called, the definition requiring the fewest number of promotions is used. For example, when calling vector + vector, the real-valued signature is used. When calling any of complex_vector + vector, vector + complex_vector, or complex_vector + complex_vector, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.\n\n\n\nComplex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, complex operator+(complex_vector, complex_vector). Through promotion, operator+ may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures operator+(complex_vector, vector) and operator+(vector, complex_vector).\n\n\n\nGeneric functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions append_array, dims, head, num_elements, rep_array, reverse, segment, size, and tail.\n\n\n\n\n \n\nint num_elements(complex_vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.30\n \n\nint num_elements(complex_row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.30\n \n\nint num_elements(complex_matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.30\n \n\nint rows(complex_vector x) The number of rows in the vector x\nAvailable since 2.30\n \n\nint rows(complex_row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.30\n \n\nint rows(complex_matrix x) The number of rows in the matrix x\nAvailable since 2.30\n \n\nint cols(complex_vector x) The number of columns in the vector x, namely 1\nAvailable since 2.30\n \n\nint cols(complex_row_vector x) The number of columns in the row vector x\nAvailable since 2.30\n \n\nint cols(complex_matrix x) The number of columns in the matrix x\nAvailable since 2.30\n \n\nint size(complex_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(complex_row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15.\nAvailable since 2.30\n\n\n\nStan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\ncomplex_vector operator-(complex_vector x) The negation of the vector x.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x) The negation of the row vector x.\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x) The negation of the matrix x.\nAvailable since 2.30\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex_vector y) The sum of the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex_row_vector y) The sum of the row vectors x and y.\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex_matrix y) The sum of the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex_vector y) The difference between the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex_row_vector y) The difference between the row vectors x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex_matrix y) The difference between the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex x, complex_vector y) The product of the scalar x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex x, complex_row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex x, complex_matrix y) The product of the scalar x and the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_vector x, complex y) The product of the scalar y and vector x\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_vector x, complex_row_vector y) The product of the vector x and row vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex y) The product of the scalar y and row vector x\nAvailable since 2.30\n \n\ncomplex operator*(complex_row_vector x, complex_vector y) The product of the row vector x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex_matrix y) The product of the row vector x and matrix y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex y) The product of the scalar y and matrix x\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_matrix x, complex_vector y) The product of the matrix x and vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex_matrix y) The product of the matrices x and y\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex y) The result of adding y to every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator+(complex x, complex_vector y) The result of adding x to every entry in the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex y) The result of adding y to every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex x, complex_row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex y) The result of adding y to every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex x, complex_matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex y) The result of subtracting y from every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex x, complex_vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex x, complex_row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex x, complex_matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator/(complex_vector x, complex y) The result of dividing each entry in the vector x by y\nAvailable since 2.30\n \n\ncomplex_row_vector operator/(complex_row_vector x, complex y) The result of dividing each entry in the row vector x by y\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix x, complex y) The result of dividing each entry in the matrix x by y\nAvailable since 2.30\n\n\n\n\nComplex complex_matrix transposition is represented using a postfix operator.\n \n\ncomplex_matrix operator'(complex_matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.30\n \n\ncomplex_row_vector operator'(complex_vector x) The transpose of the vector x, written as x'\nAvailable since 2.30\n \n\ncomplex_vector operator'(complex_row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.30\n\n\n\nAs in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.\n \n\ncomplex_vector operator.*(complex_vector x, complex_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator.*(complex_row_vector x, complex_row_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator.*(complex_matrix x, complex_matrix y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^(complex x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n\n\n\n \n\ncomplex dot_product(complex_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_vector x, complex_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_matrix x, complex_matrix y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_vector x, complex_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_matrix x, complex_matrix y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex dot_self(complex_vector x) The dot product of the vector x with itself\nAvailable since 2.30\n \n\ncomplex dot_self(complex_row_vector x) The dot product of the row vector x with itself\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_matrix x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_matrix x) The dot product of the rows of x with themselves\nAvailable since 2.30\n\n\n \n\ncomplex_matrix diag_pre_multiply(complex_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_pre_multiply(complex_row_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_row_vector v) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex sum(complex_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n\n\n\n\nMuch like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.\n\n\nThese functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a real type. For example, get_real(complex_matrix M) yields a matrix containing the real component of each value in M.\nThe following table contains examples of what this notation can mean:\n\n\n\nType T\nType T_demoted\n\n\n\n\ncomplex\nreal\n\n\ncomplex_vector\nvector\n\n\ncomplex_row_vector\nrow_vector\n\n\ncomplex_matrix\nmatrix\n\n\narray[] complex\narray[] real\n\n\narray[,,] complex\narray[,,] real\n\n\n\n\n\n\n \n\nT_demoted get_real(T x) Given an object of complex type T, return the same shape object but of type real by getting the real component of each element of x.\nAvailable since 2.30\n \n\nT_demoted get_imag(T x) Given an object of complex type T, return the same shape object but of type real by getting the imaginary component of each element of x.\nAvailable since 2.30\nFor example, given the Stan declaration\n  complex_vector[2] z = [3+4i, 5+6i]';\nA call get_real(z) will yield the vector [3, 5]', and a call get_imag(z) will yield the vector [4, 6]'.\n\n\n\n\nThe following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\ncomplex_vector rep_vector(complex z, int m) Return the size m (column) vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_row_vector rep_row_vector(complex z, int n) Return the size n row vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex z, int m, int n) Return the m by n matrix consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.30\n\n\n \n\ncomplex_matrix symmetrize_from_lower_tri(complex_matrix A) Construct a symmetric matrix from the lower triangle of A.\nAvailable since 2.30\n\n\n\n\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.30\n \n\ncomplex_vector diagonal(complex_matrix x) The diagonal of the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix diag_matrix(complex_vector x) The diagonal matrix with diagonal x\nAvailable since 2.30\n\n\n\nStan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\ncomplex_vector col(complex_matrix x, int n) The n-th column of matrix x\nAvailable since 2.30\n \n\ncomplex_row_vector row(complex_matrix x, int m) The m-th row of matrix x\nAvailable since 2.30\n\n\n\n\n\n \n\ncomplex_matrix block(complex_matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.30\n \n\ncomplex_vector sub_col(complex_matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.30\n \n\ncomplex_row_vector sub_row(complex_matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector head(complex_vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector head(complex_row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector tail(complex_vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector tail(complex_row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector segment(complex_vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n \n\ncomplex_row_vector segment(complex_row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n\n\n\n\n\n\n\n \n\ncomplex_matrix append_col(complex_matrix x, complex_matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_matrix x, complex_vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex_row_vector y) Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex x, complex_row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex y) Append y to the end of x, returning another row vector.\nAvailable since 2.30\n\n\n\n \n\ncomplex_matrix append_row(complex_matrix x, complex_matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_matrix x, complex_row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex_vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex x, complex_vector y) Append x to the top of y, returning another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex y) Append y to the bottom of x, returning another vector.\nAvailable since 2.30\n\n\n\n\n\n\nStan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors u and v,\n    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v\nand in the 2-dimensional case for complex matrices A and B,\n    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B\nAlthough the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.\n \n\ncomplex_vector fft(complex_vector v) Return the discrete Fourier transform of the specified complex vector v. If \\(v \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(u =\n\\textrm{fft}(v)\\), then \\[\\begin{equation*}\nu_n = \\sum_{m &lt; n}\nv_m \\cdot\n\\exp\\left(\\frac{-n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\]\nAvailable since 2.30\n \n\ncomplex_matrix fft2(complex_matrix m) Return the 2D discrete Fourier transform of the specified complex matrix m. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.\nAvailable since 2.30\n \n\ncomplex_vector inv_fft(complex_vector u) Return the inverse of the discrete Fourier transform of the specified complex vector u. The inverse FFT (this function) is scaled so that fft(inv_fft(u)) == u. If \\(u \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(v = \\textrm{fft}^{-1}(u)\\), then \\[\\begin{equation*}\nv_n =  \\frac{1}{N} \\sum_{m &lt; n}\nu_m \\cdot\n\\exp\\left(\\frac{n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\] This only differs from the FFT by the sign inside the exponential and the scaling. The \\(\\frac{1}{N}\\) scaling ensures that fft(inv_fft(u)) == u and inv_fft(fft(v)) == v for complex vectors u and v.\nAvailable since 2.30\n \n\ncomplex_matrix inv_fft2(complex_matrix m) Return the inverse of the 2D discrete Fourier transform of the specified complex matrix m. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures fft2(inv_fft2(A)) == A and inv_fft2(fft2(B)) == B.\nAvailable since 2.30\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] complex cumulative_sum(array[] complex x) The cumulative sum of x\nAvailable since 2.30\n \n\ncomplex_vector cumulative_sum(complex_vector v) The cumulative sum of v\nAvailable since 2.30\n \n\ncomplex_row_vector cumulative_sum(complex_row_vector rv) The cumulative sum of rv\nAvailable since 2.30\n\n\n\n\n\n\nIn general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.\n\n\n \n\ncomplex_row_vector operator/(complex_row_vector b, complex_matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix B, complex_matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex trace(complex_matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector eigenvalues(complex_matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.32\n \n\ncomplex_matrix eigenvectors(complex_matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.32\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\ncomplex_vector eigenvalues_sym(complex_matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors_sym(complex_matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose_sym(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(complex_matrix A) The singular values of A in descending order\nAvailable since 2.30\n \n\ncomplex_matrix svd_U(complex_matrix A) The left-singular vectors of A\nAvailable since 2.30\n \n\ncomplex_matrix svd_V(complex_matrix A) The right-singular vectors of A\nAvailable since 2.30\n \n\ntuple(complex_matrix, vector, complex_matrix) svd(complex_matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\nThe complex Schur decomposition of a square matrix \\(A\\) produces a complex unitary matrix \\(U\\) and a complex upper-triangular Schur form matrix \\(T\\) such that \\[A = U \\cdot T \\cdot U^{-1}\\]\nSince \\(U\\) is unitary, its inverse is also its conjugate transpose, \\(U^{-1} = U^*\\), \\(U^*(i, j) = \\mathrm{conj}(U(j, i))\\)\n \n\ncomplex_matrix complex_schur_decompose_t(matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_t(complex_matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(complex_matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results. This overload is equivalent to complex_schur_decompose(to_complex(A,0)) but is more efficient.\nAvailable since 2.33\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(complex_matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n\n\n\n\n\n \n\ncomplex_vector reverse(complex_vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.30\n \n\ncomplex_row_vector reverse(complex_row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-promotion",
+    "href": "functions-reference/complex_matrix_operations.html#complex-promotion",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.\n\n\nIf an expression e can be assigned to a variable of type T, then it can be used as an argument to a function that is specified to take arguments of type T. For instance, sqrt(real) is specified to take a real argument, but an integer expression such as 2 + 2 of type int can be passed to sqrt, so that sqrt(2 + 2) is well defined. This works by promoting the integer expression 2 + 2 to be of real type.\nThe rules for promotion in Stan are simple:\n\nint may be promoted to real,\nreal may be promoted to complex,\nvector can be promoted to complex_vector,\nrow_vector can be promoted to complex_row_vector,\nmatrix can be promoted to complex_matrix,\nif T can be promoted to U and U can be promoted to V, then T can be promoted to V (transitive), and\nif T can be promoted to U, then T[] can be promoted to U[] (covariant).\n\n\n\n\nWhen a function is called, the definition requiring the fewest number of promotions is used. For example, when calling vector + vector, the real-valued signature is used. When calling any of complex_vector + vector, vector + complex_vector, or complex_vector + complex_vector, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.\n\n\n\nComplex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, complex operator+(complex_vector, complex_vector). Through promotion, operator+ may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures operator+(complex_vector, vector) and operator+(vector, complex_vector).\n\n\n\nGeneric functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions append_array, dims, head, num_elements, rep_array, reverse, segment, size, and tail.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#integer-valued-complex-matrix-size-functions",
+    "href": "functions-reference/complex_matrix_operations.html#integer-valued-complex-matrix-size-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "int num_elements(complex_vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.30\n \n\nint num_elements(complex_row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.30\n \n\nint num_elements(complex_matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.30\n \n\nint rows(complex_vector x) The number of rows in the vector x\nAvailable since 2.30\n \n\nint rows(complex_row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.30\n \n\nint rows(complex_matrix x) The number of rows in the matrix x\nAvailable since 2.30\n \n\nint cols(complex_vector x) The number of columns in the vector x, namely 1\nAvailable since 2.30\n \n\nint cols(complex_row_vector x) The number of columns in the row vector x\nAvailable since 2.30\n \n\nint cols(complex_matrix x) The number of columns in the matrix x\nAvailable since 2.30\n \n\nint size(complex_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(complex_row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-matrix-arithmetic-operators",
+    "href": "functions-reference/complex_matrix_operations.html#complex-matrix-arithmetic-operators",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "Stan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\ncomplex_vector operator-(complex_vector x) The negation of the vector x.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x) The negation of the row vector x.\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x) The negation of the matrix x.\nAvailable since 2.30\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex_vector y) The sum of the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex_row_vector y) The sum of the row vectors x and y.\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex_matrix y) The sum of the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex_vector y) The difference between the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex_row_vector y) The difference between the row vectors x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex_matrix y) The difference between the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex x, complex_vector y) The product of the scalar x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex x, complex_row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex x, complex_matrix y) The product of the scalar x and the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_vector x, complex y) The product of the scalar y and vector x\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_vector x, complex_row_vector y) The product of the vector x and row vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex y) The product of the scalar y and row vector x\nAvailable since 2.30\n \n\ncomplex operator*(complex_row_vector x, complex_vector y) The product of the row vector x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex_matrix y) The product of the row vector x and matrix y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex y) The product of the scalar y and matrix x\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_matrix x, complex_vector y) The product of the matrix x and vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex_matrix y) The product of the matrices x and y\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex y) The result of adding y to every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator+(complex x, complex_vector y) The result of adding x to every entry in the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex y) The result of adding y to every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex x, complex_row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex y) The result of adding y to every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex x, complex_matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex y) The result of subtracting y from every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex x, complex_vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex x, complex_row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex x, complex_matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator/(complex_vector x, complex y) The result of dividing each entry in the vector x by y\nAvailable since 2.30\n \n\ncomplex_row_vector operator/(complex_row_vector x, complex y) The result of dividing each entry in the row vector x by y\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix x, complex y) The result of dividing each entry in the matrix x by y\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-transposition-operator",
+    "href": "functions-reference/complex_matrix_operations.html#complex-transposition-operator",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "Complex complex_matrix transposition is represented using a postfix operator.\n \n\ncomplex_matrix operator'(complex_matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.30\n \n\ncomplex_row_vector operator'(complex_vector x) The transpose of the vector x, written as x'\nAvailable since 2.30\n \n\ncomplex_vector operator'(complex_row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-elementwise-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-elementwise-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "As in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.\n \n\ncomplex_vector operator.*(complex_vector x, complex_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator.*(complex_row_vector x, complex_row_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator.*(complex_matrix x, complex_matrix y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^(complex x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#dot-products-and-specialized-products-for-complex-matrices",
+    "href": "functions-reference/complex_matrix_operations.html#dot-products-and-specialized-products-for-complex-matrices",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "complex dot_product(complex_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_vector x, complex_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_matrix x, complex_matrix y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_vector x, complex_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_matrix x, complex_matrix y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex dot_self(complex_vector x) The dot product of the vector x with itself\nAvailable since 2.30\n \n\ncomplex dot_self(complex_row_vector x) The dot product of the row vector x with itself\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_matrix x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_matrix x) The dot product of the rows of x with themselves\nAvailable since 2.30\n\n\n \n\ncomplex_matrix diag_pre_multiply(complex_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_pre_multiply(complex_row_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_row_vector v) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-reductions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-reductions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "complex sum(complex_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#vectorized-accessor-functions",
+    "href": "functions-reference/complex_matrix_operations.html#vectorized-accessor-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "Much like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.\n\n\nThese functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a real type. For example, get_real(complex_matrix M) yields a matrix containing the real component of each value in M.\nThe following table contains examples of what this notation can mean:\n\n\n\nType T\nType T_demoted\n\n\n\n\ncomplex\nreal\n\n\ncomplex_vector\nvector\n\n\ncomplex_row_vector\nrow_vector\n\n\ncomplex_matrix\nmatrix\n\n\narray[] complex\narray[] real\n\n\narray[,,] complex\narray[,,] real\n\n\n\n\n\n\n \n\nT_demoted get_real(T x) Given an object of complex type T, return the same shape object but of type real by getting the real component of each element of x.\nAvailable since 2.30\n \n\nT_demoted get_imag(T x) Given an object of complex type T, return the same shape object but of type real by getting the imaginary component of each element of x.\nAvailable since 2.30\nFor example, given the Stan declaration\n  complex_vector[2] z = [3+4i, 5+6i]';\nA call get_real(z) will yield the vector [3, 5]', and a call get_imag(z) will yield the vector [4, 6]'.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-broadcast-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-broadcast-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\ncomplex_vector rep_vector(complex z, int m) Return the size m (column) vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_row_vector rep_row_vector(complex z, int n) Return the size n row vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex z, int m, int n) Return the m by n matrix consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.30\n\n\n \n\ncomplex_matrix symmetrize_from_lower_tri(complex_matrix A) Construct a symmetric matrix from the lower triangle of A.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#diagonal-complex-matrix-functions",
+    "href": "functions-reference/complex_matrix_operations.html#diagonal-complex-matrix-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "complex_matrix add_diag(complex_matrix m, complex_row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.30\n \n\ncomplex_vector diagonal(complex_matrix x) The diagonal of the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix diag_matrix(complex_vector x) The diagonal matrix with diagonal x\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#slicing-and-blocking-functions-for-complex-matrices",
+    "href": "functions-reference/complex_matrix_operations.html#slicing-and-blocking-functions-for-complex-matrices",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "Stan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\ncomplex_vector col(complex_matrix x, int n) The n-th column of matrix x\nAvailable since 2.30\n \n\ncomplex_row_vector row(complex_matrix x, int m) The m-th row of matrix x\nAvailable since 2.30\n\n\n\n\n\n \n\ncomplex_matrix block(complex_matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.30\n \n\ncomplex_vector sub_col(complex_matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.30\n \n\ncomplex_row_vector sub_row(complex_matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector head(complex_vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector head(complex_row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector tail(complex_vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector tail(complex_row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector segment(complex_vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n \n\ncomplex_row_vector segment(complex_row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-matrix-concatenation",
+    "href": "functions-reference/complex_matrix_operations.html#complex-matrix-concatenation",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "complex_matrix append_col(complex_matrix x, complex_matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_matrix x, complex_vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex_row_vector y) Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex x, complex_row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex y) Append y to the end of x, returning another row vector.\nAvailable since 2.30\n\n\n\n \n\ncomplex_matrix append_row(complex_matrix x, complex_matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_matrix x, complex_row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex_vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex x, complex_vector y) Append x to the top of y, returning another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex y) Append y to the bottom of x, returning another vector.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-special-matrix-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-special-matrix-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "Stan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors u and v,\n    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v\nand in the 2-dimensional case for complex matrices A and B,\n    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B\nAlthough the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.\n \n\ncomplex_vector fft(complex_vector v) Return the discrete Fourier transform of the specified complex vector v. If \\(v \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(u =\n\\textrm{fft}(v)\\), then \\[\\begin{equation*}\nu_n = \\sum_{m &lt; n}\nv_m \\cdot\n\\exp\\left(\\frac{-n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\]\nAvailable since 2.30\n \n\ncomplex_matrix fft2(complex_matrix m) Return the 2D discrete Fourier transform of the specified complex matrix m. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.\nAvailable since 2.30\n \n\ncomplex_vector inv_fft(complex_vector u) Return the inverse of the discrete Fourier transform of the specified complex vector u. The inverse FFT (this function) is scaled so that fft(inv_fft(u)) == u. If \\(u \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(v = \\textrm{fft}^{-1}(u)\\), then \\[\\begin{equation*}\nv_n =  \\frac{1}{N} \\sum_{m &lt; n}\nu_m \\cdot\n\\exp\\left(\\frac{n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\] This only differs from the FFT by the sign inside the exponential and the scaling. The \\(\\frac{1}{N}\\) scaling ensures that fft(inv_fft(u)) == u and inv_fft(fft(v)) == v for complex vectors u and v.\nAvailable since 2.30\n \n\ncomplex_matrix inv_fft2(complex_matrix m) Return the inverse of the 2D discrete Fourier transform of the specified complex matrix m. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures fft2(inv_fft2(A)) == A and inv_fft2(fft2(B)) == B.\nAvailable since 2.30\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] complex cumulative_sum(array[] complex x) The cumulative sum of x\nAvailable since 2.30\n \n\ncomplex_vector cumulative_sum(complex_vector v) The cumulative sum of v\nAvailable since 2.30\n \n\ncomplex_row_vector cumulative_sum(complex_row_vector rv) The cumulative sum of rv\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-linear-algebra-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-linear-algebra-functions",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.\n\n\n \n\ncomplex_row_vector operator/(complex_row_vector b, complex_matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix B, complex_matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex trace(complex_matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector eigenvalues(complex_matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.32\n \n\ncomplex_matrix eigenvectors(complex_matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.32\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\ncomplex_vector eigenvalues_sym(complex_matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors_sym(complex_matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose_sym(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(complex_matrix A) The singular values of A in descending order\nAvailable since 2.30\n \n\ncomplex_matrix svd_U(complex_matrix A) The left-singular vectors of A\nAvailable since 2.30\n \n\ncomplex_matrix svd_V(complex_matrix A) The right-singular vectors of A\nAvailable since 2.30\n \n\ntuple(complex_matrix, vector, complex_matrix) svd(complex_matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\nThe complex Schur decomposition of a square matrix \\(A\\) produces a complex unitary matrix \\(U\\) and a complex upper-triangular Schur form matrix \\(T\\) such that \\[A = U \\cdot T \\cdot U^{-1}\\]\nSince \\(U\\) is unitary, its inverse is also its conjugate transpose, \\(U^{-1} = U^*\\), \\(U^*(i, j) = \\mathrm{conj}(U(j, i))\\)\n \n\ncomplex_matrix complex_schur_decompose_t(matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_t(complex_matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(complex_matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results. This overload is equivalent to complex_schur_decompose(to_complex(A,0)) but is more efficient.\nAvailable since 2.33\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(complex_matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex_matrix_operations.html#reverse-functions-for-complex-matrices",
+    "href": "functions-reference/complex_matrix_operations.html#reverse-functions-for-complex-matrices",
+    "title": "Complex Matrix Operations",
+    "section": "",
+    "text": "complex_vector reverse(complex_vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.30\n \n\ncomplex_row_vector reverse(complex_row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/circular_distributions.html",
+    "href": "functions-reference/circular_distributions.html",
+    "title": "Circular Distributions",
+    "section": "",
+    "text": "Circular distributions are defined for finite values y in any interval of length \\(2\\pi\\).\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{VonMises}(y|\\mu,\\kappa) =\n\\frac{\\exp(\\kappa\\cos(y-\\mu))}{2\\pi I_0(\\kappa)} \\!. \\end{equation*}\\] In order for this density to properly normalize, \\(y\\) must be restricted to some interval \\((c, c + 2\\pi)\\) of length \\(2 \\pi\\), because \\[\\begin{equation*} \\int_{c}^{c +\n2\\pi} \\text{VonMises}(y|\\mu,\\kappa) dy = 1. \\end{equation*}\\] Similarly, if \\(\\mu\\) is a parameter, it will typically be restricted to the same range as \\(y\\).\nIf \\(\\kappa &gt; 0\\), a von Mises distribution with its \\(2 \\pi\\) interval of support centered around its location \\(\\mu\\) will have a single mode at \\(\\mu\\); for example, restricting \\(y\\) to \\((-\\pi,\\pi)\\) and taking \\(\\mu = 0\\) leads to a single local optimum at the mode \\(\\mu\\). If the location \\(\\mu\\) is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean \\(\\mu\\).\nFor \\(\\kappa = 0\\), the Von Mises distribution corresponds to the circular uniform distribution with density \\(1 / (2 \\pi)\\) (independently of the values of \\(y\\) or \\(\\mu\\)).\n\n\n\ny ~ von_mises(mu, kappa)\nIncrement target log probability density with von_mises_lupdf(y | mu, kappa).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal von_mises_lpdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa.\nAvailable since 2.18\n \n\nreal von_mises_lupdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.\nAvailable since 2.25\n \n\nreal von_mises_cdf(reals y | reals mu, reals kappa) The von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lcdf(reals y | reals mu, reals kappa) The log of the von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lccdf(reals y | reals mu, reals kappa) The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nR von_mises_rng(reals mu, reals kappa) Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval \\([(\\mu \\mod 2\\pi)-\\pi,(\\mu \\mod\n2\\pi)+\\pi]\\)); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nEvaluating the Von Mises distribution for \\(\\kappa &gt; 100\\) is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as \\(\\kappa \\rightarrow \\infty\\), \\[\\begin{equation*}\n\\text{VonMises}(y|\\mu,\\kappa) \\rightarrow \\text{Normal}(\\mu, \\sqrt{1 /\n\\kappa}). \\end{equation*}\\] The workaround is to replace y ~ von_mises(mu,kappa) with\nif (kappa &lt; 100) {\n  y ~ von_mises(mu, kappa);\n} else {\n  y ~ normal(mu, sqrt(1 / kappa));\n}",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Circular Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/circular_distributions.html#von-mises-distribution",
+    "href": "functions-reference/circular_distributions.html#von-mises-distribution",
+    "title": "Circular Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{VonMises}(y|\\mu,\\kappa) =\n\\frac{\\exp(\\kappa\\cos(y-\\mu))}{2\\pi I_0(\\kappa)} \\!. \\end{equation*}\\] In order for this density to properly normalize, \\(y\\) must be restricted to some interval \\((c, c + 2\\pi)\\) of length \\(2 \\pi\\), because \\[\\begin{equation*} \\int_{c}^{c +\n2\\pi} \\text{VonMises}(y|\\mu,\\kappa) dy = 1. \\end{equation*}\\] Similarly, if \\(\\mu\\) is a parameter, it will typically be restricted to the same range as \\(y\\).\nIf \\(\\kappa &gt; 0\\), a von Mises distribution with its \\(2 \\pi\\) interval of support centered around its location \\(\\mu\\) will have a single mode at \\(\\mu\\); for example, restricting \\(y\\) to \\((-\\pi,\\pi)\\) and taking \\(\\mu = 0\\) leads to a single local optimum at the mode \\(\\mu\\). If the location \\(\\mu\\) is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean \\(\\mu\\).\nFor \\(\\kappa = 0\\), the Von Mises distribution corresponds to the circular uniform distribution with density \\(1 / (2 \\pi)\\) (independently of the values of \\(y\\) or \\(\\mu\\)).\n\n\n\ny ~ von_mises(mu, kappa)\nIncrement target log probability density with von_mises_lupdf(y | mu, kappa).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal von_mises_lpdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa.\nAvailable since 2.18\n \n\nreal von_mises_lupdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.\nAvailable since 2.25\n \n\nreal von_mises_cdf(reals y | reals mu, reals kappa) The von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lcdf(reals y | reals mu, reals kappa) The log of the von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lccdf(reals y | reals mu, reals kappa) The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nR von_mises_rng(reals mu, reals kappa) Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval \\([(\\mu \\mod 2\\pi)-\\pi,(\\mu \\mod\n2\\pi)+\\pi]\\)); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nEvaluating the Von Mises distribution for \\(\\kappa &gt; 100\\) is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as \\(\\kappa \\rightarrow \\infty\\), \\[\\begin{equation*}\n\\text{VonMises}(y|\\mu,\\kappa) \\rightarrow \\text{Normal}(\\mu, \\sqrt{1 /\n\\kappa}). \\end{equation*}\\] The workaround is to replace y ~ von_mises(mu,kappa) with\nif (kappa &lt; 100) {\n  y ~ von_mises(mu, kappa);\n} else {\n  y ~ normal(mu, sqrt(1 / kappa));\n}",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Circular Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_continuous_distributions.html",
+    "href": "functions-reference/bounded_continuous_distributions.html",
+    "title": "Bounded Continuous Distributions",
+    "section": "",
+    "text": "The bounded continuous probabilities have support on a finite interval of real numbers.\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in (\\alpha,\\infty)\\), then for \\(y \\in [\\alpha,\\beta]\\), \\[\\begin{equation*} \\text{Uniform}(y|\\alpha,\\beta) =\n\\frac{1}{\\beta - \\alpha} . \\end{equation*}\\]\n\n\n\ny ~ uniform(alpha, beta)\nIncrement target log probability density with uniform_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal uniform_lpdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lupdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal uniform_cdf(reals y | reals alpha, reals beta) The uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.0\n \n\nreal uniform_lcdf(reals y | reals alpha, reals beta) The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lccdf(reals y | reals alpha, reals beta) The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nR uniform_rng(reals alpha, reals beta) Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Bounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_continuous_distributions.html#uniform-distribution",
+    "href": "functions-reference/bounded_continuous_distributions.html#uniform-distribution",
+    "title": "Bounded Continuous Distributions",
+    "section": "",
+    "text": "If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in (\\alpha,\\infty)\\), then for \\(y \\in [\\alpha,\\beta]\\), \\[\\begin{equation*} \\text{Uniform}(y|\\alpha,\\beta) =\n\\frac{1}{\\beta - \\alpha} . \\end{equation*}\\]\n\n\n\ny ~ uniform(alpha, beta)\nIncrement target log probability density with uniform_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal uniform_lpdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lupdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal uniform_cdf(reals y | reals alpha, reals beta) The uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.0\n \n\nreal uniform_lcdf(reals y | reals alpha, reals beta) The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lccdf(reals y | reals alpha, reals beta) The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nR uniform_rng(reals alpha, reals beta) Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Bounded Continuous Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html",
+    "href": "functions-reference/array_operations.html",
+    "title": "Array Operations",
+    "section": "",
+    "text": "The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).\n\n\n \n\nreal min(array[] real x) The minimum value in x, or \\(+\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint min(array[] int x) The minimum value in x, or error if x is size 0.\nAvailable since 2.0\n \n\nreal max(array[] real x) The maximum value in x, or \\(-\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint max(array[] int x) The maximum value in x, or error if x is size 0.\nAvailable since 2.0\n\n\n\n \n\nint sum(array[] int x) The sum of the elements in x, or 0 if the array is empty.\nAvailable since 2.1\n \n\nreal sum(array[] real x) The sum of the elements in x; see definition above.\nAvailable since 2.0\n \n\ncomplex sum(array[] complex x) The sum of the elements in x; see definition above.\nAvailable since 2.30\n \n\nreal prod(array[] real x) The product of the elements in x, or 1 if x is size 0.\nAvailable since 2.0\n \n\nreal prod(array[] int x) The product of the elements in x, \\[\\begin{equation*}\n\\text{product}(x) = \\begin{cases}\n\\prod_{n=1}^N x_n & \\text{if} N &gt; 0 \\\\[4pt] 1 & \\text{if} N = 0\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal log_sum_exp(array[] real x) The natural logarithm of the sum of the exponentials of the elements in x, or \\(-\\infty\\) if the array is empty.\nAvailable since 2.0\n\n\n\nThe sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.1 The sample deviation is defined as the square root of the sample deviation, but is not unbiased.\n \n\nreal mean(array[] real x) The sample mean of the elements in x. For an array \\(x\\) of size \\(N &gt;\n0\\), \\[\\begin{equation*}\n\\text{mean}(x) \\ = \\ \\bar{x} \\ = \\ \\frac{1}{N} \\sum_{n=1}^N\nx_n.\n\\end{equation*}\\] It is an error to the call the mean function with an array of size \\(0\\).\nAvailable since 2.0\n \n\nreal variance(array[] real x) The sample variance of the elements in x. For \\(N &gt; 0\\), \\[\\begin{equation*}\n\\text{variance}(x) \\ = \\ \\begin{cases} \\frac{1}{N-1} \\sum_{n=1}^N (x_n\n- \\bar{x})^2 & \\text{if } N &gt; 1 \\\\[4pt] 0 & \\text{if } N = 1\n\\end{cases}\n\\end{equation*}\\] It is an error to call the variance function with an array of size 0.\nAvailable since 2.0\n \n\nreal sd(array[] real x) The sample standard deviation of elements in x. \\[\\begin{equation*}\n\\text{sd}(x) =\n\\begin{cases} \\sqrt{\\, \\text{variance}(x)} & \\text{if } N &gt; 1 \\\\[4pt]\n0 & \\text{if } N = 0 \\end{cases}\n\\end{equation*}\\] It is an error to call the sd function with an array of size 0.\nAvailable since 2.0\n\n\n\n \n\nreal norm1(vector x) The L1 norm of x, defined by \\[\\begin{equation*}\n\\text{norm1}(x) \\ = \\ \\textstyle \\sum_{n=1}^N (|x_n|)\n\\end{equation*}\\] where N is the size of x.\nAvailable since 2.30\n \n\nreal norm1(row_vector x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm1(array[] real x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm2(vector x) The L2 norm of x, defined by \\[\\begin{equation*}\n\\text{norm2}(x) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n)^2}\n\\end{equation*}\\] where N is the size of x\nAvailable since 2.30\n \n\nreal norm2(row_vector x) The L2 norm of x\nAvailable since 2.30\n \n\nreal norm2(array[] real x) The L2 norm of x\nAvailable since 2.30\n\n\n\n \n\nreal distance(vector x, vector y) The Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\text{distance}(x,y) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n - y_n)^2}\n\\end{equation*}\\] where N is the size of x and y. It is an error to call distance with arguments of unequal size.\nAvailable since 2.2\n \n\nreal distance(vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal squared_distance(vector x, vector y) The squared Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\mathrm{squared\\_distance}(x,y) \\ = \\ \\text{distance}(x,y)^2 \\ = \\ \\textstyle \\sum_{n=1}^N (x_n - y_n)^2,\n\\end{equation*}\\] where N is the size of x and y. It is an error to call squared_distance with arguments of unequal size.\nAvailable since 2.7\n \n\nreal squared_distance(vector x, row_vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.26\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data array[] real x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data array[] real x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n\n\n\n\nThe size of an array or matrix can be obtained using the dims() function. The dims() function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,\n array[7, 8, 9] real x;\n array[7] matrix[8, 9] y;\nthen calling dims(x) or dims(y) returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.\nThe size() function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as\n array[M, N] real a;\nthe size of a is M.\nThe function num_elements, on the other hand, measures all of the elements, so that the array a above has \\(M \\times N\\) elements.\nThe specialized functions rows() and cols() should be used to extract the dimensions of vectors and matrices.\n \n\narray[] int dims(T x) Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.\nAvailable since 2.0\n \n\nint num_elements(array[] T x) Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if x is of type array[4, 3] real then num_elements(x) is 12, and if y is declared as array[5] matrix[3, 4] y, then size(y) evaluates to 60.\nAvailable since 2.5\n \n\nint size(array[] T x) Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if x is of type array[4, 3] real then size(x) is 4.\nAvailable since 2.0\n\n\n\nThe following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.\n \n\narray[] T rep_array(T x, int n) Return the n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray [,] T rep_array(T x, int m, int n) Return the m by n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray[,,] T rep_array(T x, int k, int m, int n) Return the k by m by n array with every entry assigned to x.\nAvailable since 2.0\nFor example, rep_array(1.0,5) produces a real array (type array[] real) of size 5 with all values set to 1.0. On the other hand, rep_array(1,5) produces an integer array (type array[] int) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment\n array[5] real y;\n array[5] int x;\n\n x = rep_array(1, 5);     // ok\n y = rep_array(1.0, 5);   // ok\n\n x = rep_array(1.0, 5);   // illegal\n y = rep_array(1, 5);     // illegal\n\n x = y;                  // illegal\n y = x;                  // illegal\nIf the value being repeated v is a vector (i.e., T is vector), then rep_array(v, 27) is a size 27 array consisting of 27 copies of the vector v.\n vector[5] v;\n array[3] vector[5] a;\n\n a = rep_array(v, 3);  // fill a with copies of v\n a[2, 4] = 9.0;        // v[4], a[1, 4], a[3, 4] unchanged\nIf the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the rep_array functions is called. For instance, consider the following legal code snippet.\n array[5, 6] real a;\n array[3, 4, 5, 6] real b;\n\n b = rep_array(a, 3, 4); //  make (3 x 4) copies of a\n b[1, 1, 1, 1] = 27.9;    //  a[1, 1] unchanged\nAfter the assignment to b, the value for b[j, k, m, n] is equal to a[m, n] where it is defined, for j in 1:3, k in 1:4, m in 1:5, and n in 1:6.\n\n\n\n \n\nT append_array(T x, T y) Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.\nAvailable since 2.18\nFor example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.\n array[2, 1, 7] matrix[4, 6] x1;\n array[3, 1, 7] matrix[4, 6] x2;\n array[5, 1, 7] matrix[4, 6] x3;\n\n x3 = append_array(x1, x2);\n\n\n\nSorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1, -10.3,\n20.987),\n\\end{equation*}\\] then the various sort routines produce \\[\\begin{eqnarray*}\n\\mathrm{sort\\_asc(v)} & = &  (-10.3,1,20.987) \\\\[4pt]\n\\mathrm{sort\\_desc(v)} & = &  (20.987,1,-10.3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_asc(v)} & = &  (2,1,3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_desc(v)} & = &  (3,1,2)\n\\end{eqnarray*}\\]\n \n\narray[] real sort_asc(array[] real v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] int sort_asc(array[] int v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] real sort_desc(array[] real v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_desc(array[] int v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(array[] real v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(array[] int v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n\n\n\nStan provides functions to create a new array by reversing the order of elements in an existing array. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1,\\, -10.3,\\, 20.987),\n\\end{equation*}\\] then \\[\\begin{equation*}\n\\mathrm{reverse(v)} = (20.987,\\, -10.3,\\, 1).\n\\end{equation*}\\]\n \n\narray[] T reverse(array[] T v) Return a new array containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#array-reductions",
+    "href": "functions-reference/array_operations.html#array-reductions",
+    "title": "Array Operations",
+    "section": "",
+    "text": "The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).\n\n\n \n\nreal min(array[] real x) The minimum value in x, or \\(+\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint min(array[] int x) The minimum value in x, or error if x is size 0.\nAvailable since 2.0\n \n\nreal max(array[] real x) The maximum value in x, or \\(-\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint max(array[] int x) The maximum value in x, or error if x is size 0.\nAvailable since 2.0\n\n\n\n \n\nint sum(array[] int x) The sum of the elements in x, or 0 if the array is empty.\nAvailable since 2.1\n \n\nreal sum(array[] real x) The sum of the elements in x; see definition above.\nAvailable since 2.0\n \n\ncomplex sum(array[] complex x) The sum of the elements in x; see definition above.\nAvailable since 2.30\n \n\nreal prod(array[] real x) The product of the elements in x, or 1 if x is size 0.\nAvailable since 2.0\n \n\nreal prod(array[] int x) The product of the elements in x, \\[\\begin{equation*}\n\\text{product}(x) = \\begin{cases}\n\\prod_{n=1}^N x_n & \\text{if} N &gt; 0 \\\\[4pt] 1 & \\text{if} N = 0\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal log_sum_exp(array[] real x) The natural logarithm of the sum of the exponentials of the elements in x, or \\(-\\infty\\) if the array is empty.\nAvailable since 2.0\n\n\n\nThe sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.1 The sample deviation is defined as the square root of the sample deviation, but is not unbiased.\n \n\nreal mean(array[] real x) The sample mean of the elements in x. For an array \\(x\\) of size \\(N &gt;\n0\\), \\[\\begin{equation*}\n\\text{mean}(x) \\ = \\ \\bar{x} \\ = \\ \\frac{1}{N} \\sum_{n=1}^N\nx_n.\n\\end{equation*}\\] It is an error to the call the mean function with an array of size \\(0\\).\nAvailable since 2.0\n \n\nreal variance(array[] real x) The sample variance of the elements in x. For \\(N &gt; 0\\), \\[\\begin{equation*}\n\\text{variance}(x) \\ = \\ \\begin{cases} \\frac{1}{N-1} \\sum_{n=1}^N (x_n\n- \\bar{x})^2 & \\text{if } N &gt; 1 \\\\[4pt] 0 & \\text{if } N = 1\n\\end{cases}\n\\end{equation*}\\] It is an error to call the variance function with an array of size 0.\nAvailable since 2.0\n \n\nreal sd(array[] real x) The sample standard deviation of elements in x. \\[\\begin{equation*}\n\\text{sd}(x) =\n\\begin{cases} \\sqrt{\\, \\text{variance}(x)} & \\text{if } N &gt; 1 \\\\[4pt]\n0 & \\text{if } N = 0 \\end{cases}\n\\end{equation*}\\] It is an error to call the sd function with an array of size 0.\nAvailable since 2.0\n\n\n\n \n\nreal norm1(vector x) The L1 norm of x, defined by \\[\\begin{equation*}\n\\text{norm1}(x) \\ = \\ \\textstyle \\sum_{n=1}^N (|x_n|)\n\\end{equation*}\\] where N is the size of x.\nAvailable since 2.30\n \n\nreal norm1(row_vector x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm1(array[] real x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm2(vector x) The L2 norm of x, defined by \\[\\begin{equation*}\n\\text{norm2}(x) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n)^2}\n\\end{equation*}\\] where N is the size of x\nAvailable since 2.30\n \n\nreal norm2(row_vector x) The L2 norm of x\nAvailable since 2.30\n \n\nreal norm2(array[] real x) The L2 norm of x\nAvailable since 2.30\n\n\n\n \n\nreal distance(vector x, vector y) The Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\text{distance}(x,y) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n - y_n)^2}\n\\end{equation*}\\] where N is the size of x and y. It is an error to call distance with arguments of unequal size.\nAvailable since 2.2\n \n\nreal distance(vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal squared_distance(vector x, vector y) The squared Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\mathrm{squared\\_distance}(x,y) \\ = \\ \\text{distance}(x,y)^2 \\ = \\ \\textstyle \\sum_{n=1}^N (x_n - y_n)^2,\n\\end{equation*}\\] where N is the size of x and y. It is an error to call squared_distance with arguments of unequal size.\nAvailable since 2.7\n \n\nreal squared_distance(vector x, row_vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.26\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data array[] real x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data array[] real x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#array-size-and-dimension-function",
+    "href": "functions-reference/array_operations.html#array-size-and-dimension-function",
+    "title": "Array Operations",
+    "section": "",
+    "text": "The size of an array or matrix can be obtained using the dims() function. The dims() function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,\n array[7, 8, 9] real x;\n array[7] matrix[8, 9] y;\nthen calling dims(x) or dims(y) returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.\nThe size() function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as\n array[M, N] real a;\nthe size of a is M.\nThe function num_elements, on the other hand, measures all of the elements, so that the array a above has \\(M \\times N\\) elements.\nThe specialized functions rows() and cols() should be used to extract the dimensions of vectors and matrices.\n \n\narray[] int dims(T x) Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.\nAvailable since 2.0\n \n\nint num_elements(array[] T x) Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if x is of type array[4, 3] real then num_elements(x) is 12, and if y is declared as array[5] matrix[3, 4] y, then size(y) evaluates to 60.\nAvailable since 2.5\n \n\nint size(array[] T x) Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if x is of type array[4, 3] real then size(x) is 4.\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#array-broadcasting",
+    "href": "functions-reference/array_operations.html#array-broadcasting",
+    "title": "Array Operations",
+    "section": "",
+    "text": "The following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.\n \n\narray[] T rep_array(T x, int n) Return the n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray [,] T rep_array(T x, int m, int n) Return the m by n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray[,,] T rep_array(T x, int k, int m, int n) Return the k by m by n array with every entry assigned to x.\nAvailable since 2.0\nFor example, rep_array(1.0,5) produces a real array (type array[] real) of size 5 with all values set to 1.0. On the other hand, rep_array(1,5) produces an integer array (type array[] int) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment\n array[5] real y;\n array[5] int x;\n\n x = rep_array(1, 5);     // ok\n y = rep_array(1.0, 5);   // ok\n\n x = rep_array(1.0, 5);   // illegal\n y = rep_array(1, 5);     // illegal\n\n x = y;                  // illegal\n y = x;                  // illegal\nIf the value being repeated v is a vector (i.e., T is vector), then rep_array(v, 27) is a size 27 array consisting of 27 copies of the vector v.\n vector[5] v;\n array[3] vector[5] a;\n\n a = rep_array(v, 3);  // fill a with copies of v\n a[2, 4] = 9.0;        // v[4], a[1, 4], a[3, 4] unchanged\nIf the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the rep_array functions is called. For instance, consider the following legal code snippet.\n array[5, 6] real a;\n array[3, 4, 5, 6] real b;\n\n b = rep_array(a, 3, 4); //  make (3 x 4) copies of a\n b[1, 1, 1, 1] = 27.9;    //  a[1, 1] unchanged\nAfter the assignment to b, the value for b[j, k, m, n] is equal to a[m, n] where it is defined, for j in 1:3, k in 1:4, m in 1:5, and n in 1:6.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#array-concatenation",
+    "href": "functions-reference/array_operations.html#array-concatenation",
+    "title": "Array Operations",
+    "section": "",
+    "text": "T append_array(T x, T y) Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.\nAvailable since 2.18\nFor example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.\n array[2, 1, 7] matrix[4, 6] x1;\n array[3, 1, 7] matrix[4, 6] x2;\n array[5, 1, 7] matrix[4, 6] x3;\n\n x3 = append_array(x1, x2);",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#sorting-functions",
+    "href": "functions-reference/array_operations.html#sorting-functions",
+    "title": "Array Operations",
+    "section": "",
+    "text": "Sorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1, -10.3,\n20.987),\n\\end{equation*}\\] then the various sort routines produce \\[\\begin{eqnarray*}\n\\mathrm{sort\\_asc(v)} & = &  (-10.3,1,20.987) \\\\[4pt]\n\\mathrm{sort\\_desc(v)} & = &  (20.987,1,-10.3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_asc(v)} & = &  (2,1,3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_desc(v)} & = &  (3,1,2)\n\\end{eqnarray*}\\]\n \n\narray[] real sort_asc(array[] real v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] int sort_asc(array[] int v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] real sort_desc(array[] real v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_desc(array[] int v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(array[] real v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(array[] int v, int s) Number of components of v less than v[s]\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#reversing-functions",
+    "href": "functions-reference/array_operations.html#reversing-functions",
+    "title": "Array Operations",
+    "section": "",
+    "text": "Stan provides functions to create a new array by reversing the order of elements in an existing array. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1,\\, -10.3,\\, 20.987),\n\\end{equation*}\\] then \\[\\begin{equation*}\n\\mathrm{reverse(v)} = (20.987,\\, -10.3,\\, 1).\n\\end{equation*}\\]\n \n\narray[] T reverse(array[] T v) Return a new array containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/array_operations.html#footnotes",
+    "href": "functions-reference/array_operations.html#footnotes",
+    "title": "Array Operations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nDividing by \\(N\\) rather than \\((N-1)\\) produces a maximum likelihood estimate of variance, which is biased to underestimate variance.↩︎",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stansummary.html",
+    "href": "cmdstan-guide/stansummary.html",
+    "title": "stansummary: MCMC Output Analysis",
+    "section": "",
+    "text": "The CmdStan stansummary program reports statistics for one or more sampler chains over all sampler and model parameters and quantities of interest. The statistics reported include both summary statistics of the estimates and diagnostic statistics on the sampler chains, reported in the following order:\n\nMean - sample mean\nMCSE - Monte Carlo Standard Error, a measure of the amount of noise in the sample\nStdDev - sample standard deviation - the standard deviation around the sample mean.\nMAD - Median Absolute Deviation - the median absolute deviation around the sample median.\nQuantiles - default 5%, 50%, 95%\nESS_bulk\nESS_tail\nESS_bulk/s - Bulk ESS per second\nR_hat - \\(\\hat{R}\\) statistic, a MCMC convergence diagnostic\n\nWhen reviewing the stansummary output, it is important to check the final three output columns first - these are the diagnostic statistics on MCMC convergence and effective sample size. A \\(\\hat{R}\\) statistic of greater than \\(1\\) indicates potential convergence problems and that the sample is not presentative of the target posterior, thus the estimates of the mean and all other summary statistics are likely to be invalid. A value \\(1.01\\) can be used as generic threshold to decide whether more iterations or further convergence analysis is needed, but other thresholds can be used depending on the specific use case.\nEstimation by sampling produces an approximate value for the model parameters; the MCSE statistic indicates the amount of uncertainty in the estimate. Therefore MCSE column is placed next to the sample mean column, in order to make it easy to compare this sample with others.\nFor more information, see the Posterior Analysis chapter of the Stan Reference Manual which describes both the theory and practice of MCMC estimation techniques.\nThe statistics - Mean, StdDev, MAD, and Quantiles - are computed directly from all draws across all chains. The diagnostic statistics - ESS_bulk, ESS_tail, and R_hat are computed from the rank-normalized, folded, and splitted chains according to the definitions by Vehtari et al. (2021). the MCSE statistic is computed using split chain R_hat and autocorrelations. The summary statistics and the algorithms used to compute them are described in sections Notation for draws and Effective sample size.\n\n\nThe CmdStan makefile task build compiles the stansummary utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/stansummary\n\n\n\nThe stansummary utility processes one or more output files from a set of chains from one run of the HMC sampler. To run stansummary on the output file or files generated by a run of the sampler, on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nFor example, after running 4 chains to fit the example model eight_schools.stan to the supplied example data file, we run stansummary on the resulting Stan CSV output files to get the following report:\n&gt; bin/stansummary eight_*.csv\nInference for Stan model: eight_schools_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total\nSampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total\n\n                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0\naccept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0\nstepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan\ntreedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0\nn_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0\ndivergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan\nenergy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0\n\nmu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0\ntheta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0\ntheta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0\ntheta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0\ntheta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0\ntheta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0\ntheta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0\ntheta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0\ntheta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0\ntau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nThe console output information consists of\n\nModel, chains, and timing summaries\nSampler parameter statistics\nModel parameter statistics\nSampling algorithm - either nuts (shown here) or static HMC.\n\nThere is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g. if there are no divergent transitions.\n\n\nThe initial Stan CSV columns provide information on the sampler state for each draw:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause we ran the NUTS sampler, the above summary reports sampler parameters treedepth__, n_leapfrog__, and divergent__; the static HMC sampler would report int_time__ instead.\n\n\n\nThe remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The eight_schools.stan program parameters block contains the following parameter variable declarations:\n  real mu;\n  array[J] real theta;\n  real&lt;lower=0&gt; tau;\nIn the example data, J is \\(8\\); therefore the stansummary listing reports on theta[1] through theta[8].\n\n\n\n\nThe stansummary command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the -h or --help option, the program prints the usage message to the console and exits.\nReport statistics for one or more Stan CSV files from a HMC sampler run.\nExample:  stansummary model_chain_1.csv model_chain_2.csv\nOptions:\n  -a, --autocorr [n]          Display the chain autocorrelation for the n-th\n                              input file, in addition to statistics.\n  -c, --csv_filename [file]   Write statistics to a CSV file.\n  -h, --help                  Produce help message, then exit.\n  -p, --percentiles [values]  Percentiles to report as ordered set of\n                              comma-separated numbers from (0.1,99.9), inclusive.\n                              Default is 5,50,95.\n  -s, --sig_figs [n]          Significant figures reported. Default is 2.\n                              Must be an integer from (1, 18), inclusive.\n  -i, --include_param [name]  Include the named parameter in the summary output.\n                              By default, all parameters in the file are summarized,\n                              passing this argument one or more times will filter\n                              the output down to just the requested arguments.\nBoth short an long option names are allowed. Short names are specified as -&lt;o&gt; &lt;value&gt;; long option names can be specified either as --&lt;option&gt;=&lt;value&gt; or --&lt;option&gt; &lt;value&gt;.\nThe --percentiles argument can also be passed an empty string \"\", which results in no percentiles being displayed in the output of the command.\nThe amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument sig_figs. The default sampler output precision is 8. The --sig_figs argument to the stansummary program should not exceed the sig_figs argument to the sampler.",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stansummary.html#building-the-stansummary-command",
+    "href": "cmdstan-guide/stansummary.html#building-the-stansummary-command",
+    "title": "stansummary: MCMC Output Analysis",
+    "section": "",
+    "text": "The CmdStan makefile task build compiles the stansummary utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/stansummary",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stansummary.html#running-the-stansummary-program",
+    "href": "cmdstan-guide/stansummary.html#running-the-stansummary-program",
+    "title": "stansummary: MCMC Output Analysis",
+    "section": "",
+    "text": "The stansummary utility processes one or more output files from a set of chains from one run of the HMC sampler. To run stansummary on the output file or files generated by a run of the sampler, on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nFor example, after running 4 chains to fit the example model eight_schools.stan to the supplied example data file, we run stansummary on the resulting Stan CSV output files to get the following report:\n&gt; bin/stansummary eight_*.csv\nInference for Stan model: eight_schools_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total\nSampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total\n\n                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0\naccept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0\nstepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan\ntreedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0\nn_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0\ndivergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan\nenergy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0\n\nmu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0\ntheta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0\ntheta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0\ntheta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0\ntheta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0\ntheta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0\ntheta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0\ntheta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0\ntheta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0\ntau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nThe console output information consists of\n\nModel, chains, and timing summaries\nSampler parameter statistics\nModel parameter statistics\nSampling algorithm - either nuts (shown here) or static HMC.\n\nThere is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g. if there are no divergent transitions.\n\n\nThe initial Stan CSV columns provide information on the sampler state for each draw:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause we ran the NUTS sampler, the above summary reports sampler parameters treedepth__, n_leapfrog__, and divergent__; the static HMC sampler would report int_time__ instead.\n\n\n\nThe remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The eight_schools.stan program parameters block contains the following parameter variable declarations:\n  real mu;\n  array[J] real theta;\n  real&lt;lower=0&gt; tau;\nIn the example data, J is \\(8\\); therefore the stansummary listing reports on theta[1] through theta[8].",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stansummary.html#command-line-options",
+    "href": "cmdstan-guide/stansummary.html#command-line-options",
+    "title": "stansummary: MCMC Output Analysis",
+    "section": "",
+    "text": "The stansummary command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the -h or --help option, the program prints the usage message to the console and exits.\nReport statistics for one or more Stan CSV files from a HMC sampler run.\nExample:  stansummary model_chain_1.csv model_chain_2.csv\nOptions:\n  -a, --autocorr [n]          Display the chain autocorrelation for the n-th\n                              input file, in addition to statistics.\n  -c, --csv_filename [file]   Write statistics to a CSV file.\n  -h, --help                  Produce help message, then exit.\n  -p, --percentiles [values]  Percentiles to report as ordered set of\n                              comma-separated numbers from (0.1,99.9), inclusive.\n                              Default is 5,50,95.\n  -s, --sig_figs [n]          Significant figures reported. Default is 2.\n                              Must be an integer from (1, 18), inclusive.\n  -i, --include_param [name]  Include the named parameter in the summary output.\n                              By default, all parameters in the file are summarized,\n                              passing this argument one or more times will filter\n                              the output down to just the requested arguments.\nBoth short an long option names are allowed. Short names are specified as -&lt;o&gt; &lt;value&gt;; long option names can be specified either as --&lt;option&gt;=&lt;value&gt; or --&lt;option&gt; &lt;value&gt;.\nThe --percentiles argument can also be passed an empty string \"\", which results in no percentiles being displayed in the output of the command.\nThe amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument sig_figs. The default sampler output precision is 8. The --sig_figs argument to the stansummary program should not exceed the sig_figs argument to the sampler.",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html",
+    "href": "cmdstan-guide/stan_csv_apdx.html",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "The output from all CmdStan methods is in CSV format. A Stan CSV file is a data table where the columns are the method and model parameters and quantities of interest. Each row contains one record’s worth of data in plain-text format using the comma character (‘,’) as the field delimiter (hence the name).\nFor the Stan CSV files, data is strictly numerical, however, possible values include both positive and negative infinity and “Not-a-Number” which are represented as the strings NaN, inf, +inf, -inf. All other values are written in decimal notation by default with at most 8 digits of precision. The number of significant digits written can be controlled with argument sig_figs=&lt;int&gt;. See more in Output control arguments section.\nStan CSV files have a header row containing the column names. They also make extensive use of CSV comments, i.e., lines which begin with the # character. In addition to initial and final comment rows, some methods also put comment rows in the middle of the data table, which makes it difficult to use many of the commonly used CSV parser packages.\n\n\nThe data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.\nStan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is\n\nThe column name consists of the variable name followed by the element indices.\nIndices are delimited by periods (‘.’).\nIndexing is 1-based, i.e., given a dimension of size \\(N\\), the first element index is \\(1\\) and the last element index is \\(N\\).\nTuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).\n\nContainer variables are serialized in column major order, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.\nTo see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:\n array[2, 3, 4] real foo;\nThe Stan CSV file will require 24 columns to output the elements of foo. The first 6 columns will be labeled:\nfoo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1\nThe final 6 columns will be labeled:\nfoo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4\nTo see how a tuple would be laid out, consider the following variable:\ntuple(real, array[3] real) bar;\nThis will correspond to 4 columns in the CSV file, which are labeled\nbar:1,bar:2.1,bar:2.2,bar:2.3\n\n\n\nThe sample method produces both a Stan CSV output file and a diagnostic file which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.\nTo see how this works, we show snippets of the output file resulting from the following command:\n./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \\\n            data file=bernoulli.data.json \\\n            output file=bernoulli_samples.csv\n\n\nThe sampler output file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nComment rows for adaptation listing step size and metric used for sampling\nSampling draws\nComment rows giving timing information\n\nInitial comments rows: argument configuration\nAll configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as (Default).\nIn the above example the num_samples, num_warmup, and save_warmup arguments were specified, whereas subargument thin is left at its default value, as seen in the initial comment rows:\n# stan_version_major = 2\n# stan_version_minor = 24\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 100\n#     num_warmup = 200\n#     save_warmup = 1\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000000000000003 (Default)\n#       delta = 0.80000000000000004 (Default)\n#       kappa = 0.75 (Default)\n#       t0 = 10 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1 (Default)\n#         stepsize_jitter = 0 (Default)\n# id = 0 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2991989946 (Default)\n# output\n#   file = bernoulli_samples.csv\n#   diagnostic_file = bernoulli_diagnostics.csv\n#   refresh = 100 (Default)\nNote that when running multi-threaded programs which use reduce_sum for high-level parallelization, the number of threads used will also be included in this initial comment header.\nColumn headers\nThe CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the output file section of the chapter on MCMC Sampling. The example model bernoulli.stan only contains one parameter theta, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nAs a second example, we show the output of the eight_schools.stan model on run on example dataset. This model has 3 parameters: mu, theta a vector whose length is dependent on the input data, here N = 8, and tau. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:\nmu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau\nData rows containing warmup draws\nWhen run with option save_warmup=1, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is save_warmup=0, this section is usually not present in the output file.\nHere we specified num_warmup=200 and left thin at the default value \\(1\\), therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.\n-6.74827,1,1,1,1,0,6.75348,0.247195\n-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195\n-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195\n-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619\n-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517\nComment rows for adaptation\nDuring warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.\n# Adaptation terminated\n# Step size = 0.813694\n# Diagonal elements of inverse mass matrix:\n# 0.592879\nAs the example bernoulli model only contains a single parameter, and as the default metric is diag_e, the inverse mass matrix is a \\(1 \\times 1\\) matrix, and the length of the diagonal vector is also \\(1\\).\nIn contrast, if we run the eight schools example model with metric dense_e, the adaptation comments section lists both the stepsize and the full \\(10 \\times 10\\) inverse mass matrix:\n# Adaptation terminated\n# Step size = 0.211252\n# Elements of inverse mass matrix:\n# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757\n# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028\n# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353\n# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097\n# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245\n# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507\n# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946\n# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661\n# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104\n# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433\nNote that when the sampler is run with arguments algorithm=fixed_param, this section will be missing.\nData rows containing sampling draws\nThe output file contains the values for the thinned set draws during sampling. Here we specified num_sampling=100 and left thin at the default value \\(1\\), therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.\n-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093\n-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431\n-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341\nTiming information\nUpon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:\n#\n#  Elapsed Time: 0.005 seconds (Warm-up)\n#                0.002 seconds (Sampling)\n#                0.007 seconds (Total)\n#\n\n\n\nThe diagnostic file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nSampling draws\nComment rows giving timing information\n\nThe columns in this file contain, in order:\n\nall sampler parameters\nall model parameter estimates (on the unconstrained scale)\nthe latent Hamiltonian for each parameter\nthe gradient for each parameters\n\nThe labels for the latent Hamiltonian columns are the parameter column label with prefix p_ and the labels for the gradient columns are the parameter column label with prefix g_.\nThese are the column labels from the file bernoulli_diagnostic.csv:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta\n\n\n\nThe profiling information is stored in a plain CSV format with no meta information in the comments.\nEach row represents timing information collected in a profile statement for a given thread. It is possible that some profile statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).\nThe columns are as follows:\n\nname, The name of the profile statement that is being timed\nthread_id, The thread that executed the profile statement\ntotal_time, The combined time spent executing statements inside the profile which includes calculation with and without automatic differentiation\nforward_time, The time spent in the profile statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation\nreverse_time, The time spent in the profile statement during the reverse (backward) pass of reverse mode automatic differentiation\nchain_stack, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass\nno_chain_stack, The number of objects allocated on the non-chaining automatic differentiation stack\nautodiff_calls, The total number of times the profile statement was executed with automatic differentiation\nno_autodiff_calls - The total number of times the profile statement was executed without automatic differentiation\n\n\n\n\n\n\nConfig as comments\nHeader row\nPenalized maximum likelihood estimate\n\n\n\n\n\nConfig as comments\nHeader row\nAdaptation as comments\nVariational estimate\nSample draws from estimate of the posterior\n\n\n\n\n\nHeader row\nQuantities of interest\n\n\n\n\n\nHeader row\nGradients",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#csv-column-names-and-order",
+    "href": "cmdstan-guide/stan_csv_apdx.html#csv-column-names-and-order",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "The data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.\nStan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is\n\nThe column name consists of the variable name followed by the element indices.\nIndices are delimited by periods (‘.’).\nIndexing is 1-based, i.e., given a dimension of size \\(N\\), the first element index is \\(1\\) and the last element index is \\(N\\).\nTuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).\n\nContainer variables are serialized in column major order, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.\nTo see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:\n array[2, 3, 4] real foo;\nThe Stan CSV file will require 24 columns to output the elements of foo. The first 6 columns will be labeled:\nfoo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1\nThe final 6 columns will be labeled:\nfoo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4\nTo see how a tuple would be laid out, consider the following variable:\ntuple(real, array[3] real) bar;\nThis will correspond to 4 columns in the CSV file, which are labeled\nbar:1,bar:2.1,bar:2.2,bar:2.3",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#mcmc-sampler-csv-output",
+    "href": "cmdstan-guide/stan_csv_apdx.html#mcmc-sampler-csv-output",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "The sample method produces both a Stan CSV output file and a diagnostic file which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.\nTo see how this works, we show snippets of the output file resulting from the following command:\n./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \\\n            data file=bernoulli.data.json \\\n            output file=bernoulli_samples.csv\n\n\nThe sampler output file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nComment rows for adaptation listing step size and metric used for sampling\nSampling draws\nComment rows giving timing information\n\nInitial comments rows: argument configuration\nAll configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as (Default).\nIn the above example the num_samples, num_warmup, and save_warmup arguments were specified, whereas subargument thin is left at its default value, as seen in the initial comment rows:\n# stan_version_major = 2\n# stan_version_minor = 24\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 100\n#     num_warmup = 200\n#     save_warmup = 1\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000000000000003 (Default)\n#       delta = 0.80000000000000004 (Default)\n#       kappa = 0.75 (Default)\n#       t0 = 10 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1 (Default)\n#         stepsize_jitter = 0 (Default)\n# id = 0 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2991989946 (Default)\n# output\n#   file = bernoulli_samples.csv\n#   diagnostic_file = bernoulli_diagnostics.csv\n#   refresh = 100 (Default)\nNote that when running multi-threaded programs which use reduce_sum for high-level parallelization, the number of threads used will also be included in this initial comment header.\nColumn headers\nThe CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the output file section of the chapter on MCMC Sampling. The example model bernoulli.stan only contains one parameter theta, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nAs a second example, we show the output of the eight_schools.stan model on run on example dataset. This model has 3 parameters: mu, theta a vector whose length is dependent on the input data, here N = 8, and tau. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:\nmu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau\nData rows containing warmup draws\nWhen run with option save_warmup=1, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is save_warmup=0, this section is usually not present in the output file.\nHere we specified num_warmup=200 and left thin at the default value \\(1\\), therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.\n-6.74827,1,1,1,1,0,6.75348,0.247195\n-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195\n-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195\n-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619\n-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517\nComment rows for adaptation\nDuring warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.\n# Adaptation terminated\n# Step size = 0.813694\n# Diagonal elements of inverse mass matrix:\n# 0.592879\nAs the example bernoulli model only contains a single parameter, and as the default metric is diag_e, the inverse mass matrix is a \\(1 \\times 1\\) matrix, and the length of the diagonal vector is also \\(1\\).\nIn contrast, if we run the eight schools example model with metric dense_e, the adaptation comments section lists both the stepsize and the full \\(10 \\times 10\\) inverse mass matrix:\n# Adaptation terminated\n# Step size = 0.211252\n# Elements of inverse mass matrix:\n# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757\n# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028\n# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353\n# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097\n# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245\n# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507\n# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946\n# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661\n# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104\n# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433\nNote that when the sampler is run with arguments algorithm=fixed_param, this section will be missing.\nData rows containing sampling draws\nThe output file contains the values for the thinned set draws during sampling. Here we specified num_sampling=100 and left thin at the default value \\(1\\), therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.\n-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093\n-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431\n-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341\nTiming information\nUpon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:\n#\n#  Elapsed Time: 0.005 seconds (Warm-up)\n#                0.002 seconds (Sampling)\n#                0.007 seconds (Total)\n#\n\n\n\nThe diagnostic file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nSampling draws\nComment rows giving timing information\n\nThe columns in this file contain, in order:\n\nall sampler parameters\nall model parameter estimates (on the unconstrained scale)\nthe latent Hamiltonian for each parameter\nthe gradient for each parameters\n\nThe labels for the latent Hamiltonian columns are the parameter column label with prefix p_ and the labels for the gradient columns are the parameter column label with prefix g_.\nThese are the column labels from the file bernoulli_diagnostic.csv:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta\n\n\n\nThe profiling information is stored in a plain CSV format with no meta information in the comments.\nEach row represents timing information collected in a profile statement for a given thread. It is possible that some profile statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).\nThe columns are as follows:\n\nname, The name of the profile statement that is being timed\nthread_id, The thread that executed the profile statement\ntotal_time, The combined time spent executing statements inside the profile which includes calculation with and without automatic differentiation\nforward_time, The time spent in the profile statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation\nreverse_time, The time spent in the profile statement during the reverse (backward) pass of reverse mode automatic differentiation\nchain_stack, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass\nno_chain_stack, The number of objects allocated on the non-chaining automatic differentiation stack\nautodiff_calls, The total number of times the profile statement was executed with automatic differentiation\nno_autodiff_calls - The total number of times the profile statement was executed without automatic differentiation",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#optimization-output",
+    "href": "cmdstan-guide/stan_csv_apdx.html#optimization-output",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "Config as comments\nHeader row\nPenalized maximum likelihood estimate",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#variational-inference-output",
+    "href": "cmdstan-guide/stan_csv_apdx.html#variational-inference-output",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "Config as comments\nHeader row\nAdaptation as comments\nVariational estimate\nSample draws from estimate of the posterior",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#generate-quantities-outputs",
+    "href": "cmdstan-guide/stan_csv_apdx.html#generate-quantities-outputs",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "Header row\nQuantities of interest",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#diagnose-method-outputs",
+    "href": "cmdstan-guide/stan_csv_apdx.html#diagnose-method-outputs",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "Header row\nGradients",
+    "crumbs": [
+      "Appendices",
+      "Stan CSV File Format"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/print.html",
+    "href": "cmdstan-guide/print.html",
+    "title": "print (deprecated): MCMC Output Analysis",
+    "section": "",
+    "text": "print (deprecated): MCMC Output Analysis\nThe print utility is deprecated, but is still available until CmdStan v3.0. It has been replaced by the stansummary utility.\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Tools and Utilities",
+      "`print` (deprecated): MCMC Output Analysis"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/parallelization.html",
+    "href": "cmdstan-guide/parallelization.html",
+    "title": "Parallelization",
+    "section": "",
+    "text": "Stan provides three ways of parallelizing execution of a Stan model:\n\nmulti-threading with Intel Threading Building Blocks (TBB),\nmulti-processing with Message Passing Interface (MPI) and\nmanycore processing with OpenCL.\n\n\n\nIn order to exploit multi-threading in a Stan model, the models must be rewritten to use the reduce_sum and map_rect functions. For instructions on how to rewrite Stan models to use these functions see Stan’s User guide chapter on parallelization, the reduce_sum case study or the Multithreading and Map-Reduce tutorial.\n\n\nOnce a model is rewritten to use the above-mentioned functions, the model must be compiled with the STAN_THREADS makefile flag. The flag can be supplied in the make call but we recommend writing the flag to the make/local file. If the STAN_THREADS flag is defined/non-empty, threads will be enabled.\nAn example of the contents of make/local to enable threading with TBB:\nSTAN_THREADS=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nBefore running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the num_threads argument. Valid values for num_threads are positive integers and -1. If num_threads is set to -1, all available cores will be used.\nGenerally, this number should not exceed the number of available cores for best performance.\nExample:\n./model sample data file=data.json num_threads=4 ...\nWhen the model is compiled with STAN_THREADS we can sample with multiple chains with a single executable (see section running multiple chains for cases when this is available). When running multiple chains num_threads is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:\n./model sample num_chains=2 data file=data.json num_threads=8 ...\n\n\n\n\nIn order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the map_rect function. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.\n\n\nCompiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are MPICH and OpenMPI.\n\n\n\nOnce a model is rewritten to use map_rect, additional makefile flags must be written to the make/local. These are:\n\nSTAN_MPI: Enables the use of MPI with Stan if defined.\nCXX: The name of the MPI C++ compiler wrapper. Typically mpicxx.\nTBB_CXX_TYPE: The C++ compiler the MPI wrapper wraps. Typically gcc on Linux and clang on macOS.\n\nAn example of make/local on Linux:\nSTAN_MPI=true\nCXX=mpicxx\nTBB_CXX_TYPE=gcc\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_MPI is run using an MPI launcher. The MPI standard suggests using mpiexec, but a vendor wrapper for the launcher like mpirun can also be used. The launcher is supplied the path to the built executable and the number of processes to start: -n X for mpiexec or -np X for mpirun where X is replaced by the integer representing the number of processes.\nExample for running a model with six processes:\nmpiexec -n 6 path/to/model sample data file=data.json ...\n\n\n\n\n\n\nOpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.\nIn order to check if any OpenCL-enabled device and its runtime is already present use the clinfo tool. On Linux, clinfo can typically be installed with the default package manager (for example sudo apt-get install clinfo on Ubuntu). For Windows, pre-built clinfo binary can be found here.\nAlso use clinfo to verify successful installation of OpenCL runtimes.\n\n\n\nLinux:\nInstall the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:\nsudo apt update\nsudo apt install nvidia-driver-460 nvidia-cuda-toolkit\nReplace the driver version (460 in the above case) with the lastest number at the time of installation.\nWindows:\nInstall the NVIDIA GPU Driver and CUDA Toolkit.\n\n\n\n\n\nLinux:\nInstall Radeon Software for Linux available here.\nWindows:\nWe recommend installing the open source OCL-SDK.\n\n\n\n\nInstall the open source PoCL.\n\n\n\nFollow Intel’s install instructions given here (requires registration).\n\n\n\n\nIn order to enable the OpenCL backend the model must be compiled with the STAN_OPENCL makefile flag defined/non-empty. The flag can be supplied in the make call but we recommend writing the flag to the make/local file.\nAn example of the contents of make/local to enable parallelization with OpenCL:\nSTAN_OPENCL=true\nIf you are using OpenCL with an integrated GPU you also need to define the INTEGRATED_OPENCL flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:\nINTEGRATED_OPENCL=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_OPENCL can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the clinfo program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically 0. In that case you can also omit the OpenCL IDs as the default 0 IDs are used in that case.\nWe supply these IDs when starting the executable as shown below:\npath/to/model sample data file=data.json opencl platform=0 device=1",
+    "crumbs": [
+      "Getting Started",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/parallelization.html#multi-threading-with-tbb",
+    "href": "cmdstan-guide/parallelization.html#multi-threading-with-tbb",
+    "title": "Parallelization",
+    "section": "",
+    "text": "In order to exploit multi-threading in a Stan model, the models must be rewritten to use the reduce_sum and map_rect functions. For instructions on how to rewrite Stan models to use these functions see Stan’s User guide chapter on parallelization, the reduce_sum case study or the Multithreading and Map-Reduce tutorial.\n\n\nOnce a model is rewritten to use the above-mentioned functions, the model must be compiled with the STAN_THREADS makefile flag. The flag can be supplied in the make call but we recommend writing the flag to the make/local file. If the STAN_THREADS flag is defined/non-empty, threads will be enabled.\nAn example of the contents of make/local to enable threading with TBB:\nSTAN_THREADS=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nBefore running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the num_threads argument. Valid values for num_threads are positive integers and -1. If num_threads is set to -1, all available cores will be used.\nGenerally, this number should not exceed the number of available cores for best performance.\nExample:\n./model sample data file=data.json num_threads=4 ...\nWhen the model is compiled with STAN_THREADS we can sample with multiple chains with a single executable (see section running multiple chains for cases when this is available). When running multiple chains num_threads is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:\n./model sample num_chains=2 data file=data.json num_threads=8 ...",
+    "crumbs": [
+      "Getting Started",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/parallelization.html#multi-processing-with-mpi",
+    "href": "cmdstan-guide/parallelization.html#multi-processing-with-mpi",
+    "title": "Parallelization",
+    "section": "",
+    "text": "In order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the map_rect function. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.\n\n\nCompiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are MPICH and OpenMPI.\n\n\n\nOnce a model is rewritten to use map_rect, additional makefile flags must be written to the make/local. These are:\n\nSTAN_MPI: Enables the use of MPI with Stan if defined.\nCXX: The name of the MPI C++ compiler wrapper. Typically mpicxx.\nTBB_CXX_TYPE: The C++ compiler the MPI wrapper wraps. Typically gcc on Linux and clang on macOS.\n\nAn example of make/local on Linux:\nSTAN_MPI=true\nCXX=mpicxx\nTBB_CXX_TYPE=gcc\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_MPI is run using an MPI launcher. The MPI standard suggests using mpiexec, but a vendor wrapper for the launcher like mpirun can also be used. The launcher is supplied the path to the built executable and the number of processes to start: -n X for mpiexec or -np X for mpirun where X is replaced by the integer representing the number of processes.\nExample for running a model with six processes:\nmpiexec -n 6 path/to/model sample data file=data.json ...",
+    "crumbs": [
+      "Getting Started",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/parallelization.html#opencl",
+    "href": "cmdstan-guide/parallelization.html#opencl",
+    "title": "Parallelization",
+    "section": "",
+    "text": "OpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.\nIn order to check if any OpenCL-enabled device and its runtime is already present use the clinfo tool. On Linux, clinfo can typically be installed with the default package manager (for example sudo apt-get install clinfo on Ubuntu). For Windows, pre-built clinfo binary can be found here.\nAlso use clinfo to verify successful installation of OpenCL runtimes.\n\n\n\nLinux:\nInstall the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:\nsudo apt update\nsudo apt install nvidia-driver-460 nvidia-cuda-toolkit\nReplace the driver version (460 in the above case) with the lastest number at the time of installation.\nWindows:\nInstall the NVIDIA GPU Driver and CUDA Toolkit.\n\n\n\n\n\nLinux:\nInstall Radeon Software for Linux available here.\nWindows:\nWe recommend installing the open source OCL-SDK.\n\n\n\n\nInstall the open source PoCL.\n\n\n\nFollow Intel’s install instructions given here (requires registration).\n\n\n\n\nIn order to enable the OpenCL backend the model must be compiled with the STAN_OPENCL makefile flag defined/non-empty. The flag can be supplied in the make call but we recommend writing the flag to the make/local file.\nAn example of the contents of make/local to enable parallelization with OpenCL:\nSTAN_OPENCL=true\nIf you are using OpenCL with an integrated GPU you also need to define the INTEGRATED_OPENCL flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:\nINTEGRATED_OPENCL=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_OPENCL can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the clinfo program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically 0. In that case you can also omit the OpenCL IDs as the default 0 IDs are used in that case.\nWe supply these IDs when starting the executable as shown below:\npath/to/model sample data file=data.json opencl platform=0 device=1",
+    "crumbs": [
+      "Getting Started",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html",
+    "href": "cmdstan-guide/mcmc_config.html",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "The sample method provides Bayesian inference over the model conditioned on data using Hamiltonian Monte Carlo (HMC) sampling. By default, the inference engine used is the No-U-Turn sampler (NUTS), an adaptive form of Hamiltonian Monte Carlo sampling. For details on HMC and NUTS, see the Stan Reference Manual chapter on MCMC Sampling.\n\n\nTo generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument sample or method=sample together with the input data. The executable can be run from any directory.\nThe full set of configuration options available for the sample method is available by using the sample help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.\nHere, we run it in the directory which contains the Stan program and input data, &lt;cmdstan-home&gt;/examples/bernoulli:\n&gt; cd examples/bernoulli\n&gt; ls\n  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan\nTo execute sampling of the model under Linux or Mac, use:\n&gt; ./bernoulli sample data file=bernoulli.data.json\nIn Windows, the ./ prefix is not needed:\n&gt; bernoulli.exe sample data file=bernoulli.data.json\nThe output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:\nmethod = sample (Default)\n  sample\n    num_samples = 1000 (Default)\n    num_warmup = 1000 (Default)\n    save_warmup = false (Default)\n    thin = 1 (Default)\n    adapt\n      engaged = true (Default)\n      gamma = 0.050000000000000003 (Default)\n      delta = 0.80000000000000004 (Default)\n      kappa = 0.75 (Default)\n      t0 = 10 (Default)\n      init_buffer = 75 (Default)\n      term_buffer = 50 (Default)\n      window = 25 (Default)\n      save_metric = false (Default)\n    algorithm = hmc (Default)\n      hmc\n        engine = nuts (Default)\n          nuts\n            max_depth = 10 (Default)\n        metric = diag_e (Default)\n        metric_file =  (Default)\n        stepsize = 1 (Default)\n        stepsize_jitter = 0 (Default)\n    num_chains = 1 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 3252652196 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\nAfter the configuration has been displayed, a short timing message is given.\nGradient evaluation took 1.2e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.\nAdjust your expectations accordingly!\nNext, the sampler reports the iteration number, reporting the percentage complete.\nIteration:    1 / 2000 [  0%]  (Warmup)\n...\nIteration: 2000 / 2000 [100%]  (Sampling)\nFinally, the sampler reports timing information:\n Elapsed Time: 0.007 seconds (Warm-up)\n               0.017 seconds (Sampling)\n               0.024 seconds (Total)\n\n\n\nEach execution of the model results in draws from a single Markov chain being written to a file in comma-separated value (CSV) format. The default name of the output file is output.csv.\nThe first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (#)).\nWhen the example model bernoulli.stan is run via the command line with all default arguments, the following configuration is displayed:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 1000 (Default)\n#     num_warmup = 1000 (Default)\n#     save_warmup = false (Default)\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000 (Default)\n#       delta = 0.800000 (Default)\n#       kappa = 0.750000 (Default)\n#       t0 = 10.000000 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#       save_metric = false (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1.000000 (Default)\n#         stepsize_jitter = 0.000000 (Default)\n#     num_chains = 1 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\nThis is followed by a CSV header indicating the names of the values sampled.\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nThe first output columns report the HMC sampler information:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause the above header is from the NUTS sampler, it has columns treedepth__, n_leapfrog__, and divergent__ and doesn’t have column int_time__. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, theta.\nThe header line is written to the output file before warmup begins. If option save_warmup is set to true, the warmup draws are output directly after the header. The total number of warmup draws saved is num_warmup divided by thin, rounded up (i.e., ceiling).\nFollowing the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:\n# Adaptation terminated\n# Step size = 0.884484\n# Diagonal elements of inverse mass matrix:\n# 0.535006\nThe default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter theta.\nDraws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.\n-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853\n-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295\n-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299\n-6.88712,1,0.884484,1,1,0,7.02101,0.188229\n-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596\n...\nThe output ends with timing details:\n#  Elapsed Time: 0.007 seconds (Warm-up)\n#                0.017 seconds (Sampling)\n#                0.024 seconds (Total)\n\n\n\nAt every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters metric and stepsize in order to efficiently sample from typical set, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.\nThe following keyword-value arguments control the total number of iterations:\n\nnum_samples\nnum_warmup\nsave_warmup\nthin\n\nThe values for arguments num_samples and num_warmup must be a non-negative integer. The default value for both is \\(1000\\).\nFor well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.\nThe number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is \\(\\sqrt{N}\\); therefore every additional decimal place of accuracy increases this by a factor of 10.\nArgument save_warmup takes values false or true. The default value is false, i.e., warmup draws are not saved to the output file. When the value is true, the warmup draws are written to the CSV output file directly after the CSV header line.\nArgument thin controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce anticorrelated draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.\nThe value of argument thin must be a positive integer. When thin is set to value \\(N\\), every \\(N^{th}\\) iteration is written to the output file. Should the value of thin exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of thin equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When num_sampling=M and thin=N, the number of iterations written to the output CSV file will be ceiling(M/N). If save_warmup=true, thinning is applied to the warmup iterations as well.\n\n\n\nThe adapt keyword is used to specify non-default options for the sampler adaptation schedule and settings.\nAdaptation can be turned off by setting sub-argument engaged to value false. If engaged=false, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is engaged=1, this keyword-value pair can be omitted from the command.\nThere are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section Automatic Parameter Tuning.\nThe boolean sub-argument save_metric was added in Stan version 2.34. When save_metric=true, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix _metric.json, e.g., if using the default output filename output.csv, the saved metric file will be output_metric.json. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument metric_file.\n\n\nThe Stan User’s Guide section on model conditioning and curvature provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on HMC algorithm parameters explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.\nThe following keyword-value arguments control the settings used to optimize the step size:\n\ndelta - The target Metropolis acceptance rate. The default value is \\(0.8\\). Its value must be strictly between \\(0\\) and \\(1\\). Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of delta will also allow some models that would otherwise get stuck to overcome their blockages.  Models with difficult posterior geometries may required increasing the delta argument closer to \\(1\\); we recommend first trying to raise it to \\(0.9\\) or at most \\(0.95\\). Values about \\(0.95\\) are strong indication of bad geometry; the better solution is to change the model geometry through reparameterization which could yield both more efficient and faster sampling.\ngamma - Adaptation regularization scale. Must be a positive real number, default value is \\(0.05\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nkappa - Adaptation relaxation exponent. Must be a positive real number, default value is \\(0.75\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nt_0 - Adaptation iteration offset. Must be a positive real number, default value is \\(10\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\n\n\n\n\nWhen adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:\n\ninit_buffer - The number of iterations spent tuning the step size at the outset of adaptation.\nwindow - The initial number of iterations devoted to tune the metric, will be doubled successively.\nterm_buffer - The number of iterations used to re-tune the step size once the metric has been tuned.\n\nThe specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.\nThe following figure is taken from the Stan Reference Manual, where label “I” correspond to init_buffer, the initial “II” corresponds to window, and the final “III” corresponds to term_buffer:\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\n\n\n\n\nThe algorithm keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: hmc, which generates from an HMC-driven Markov chain; and fixed_param which generates a new sample without changing the state of the Markov chain. The default argument is algorithm=hmc.\n\n\nIf a model doesn’t specify any parameters, then argument algorithm=fixed_param is mandatory.\nThe fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.\n\n\n\nAll HMC algorithms have three parameters:\n\nstep size\nmetric\nintegration time - the number of steps taken along the Hamiltonian trajectory\n\nSee the Stan Reference Manual section on HMC algorithm parameters for further details.\n\n\nThe HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.\nThe following keyword-value arguments control the step size:\n\nstepsize - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is \\(1\\).\nstepsize_jitter - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between \\(0\\) and \\(1\\). The default value is \\(0\\). Setting stepsize_jitter to \\(1\\) causes step sizes to be selected in the range of \\(0\\) to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.\n\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nThe metric argument specifies the choice of Euclidean HMC implementations:\n\nmetric=unit specifies unit metric (diagonal matrix of ones).\nmetric=diag_e specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.\nmetric=dense_e specifies a dense metric (a dense, symmetric positive definite matrix).\n\nBy default, the metric is estimated during warmup. However, when metric=diag_e or metric=dense_e, an initial guess for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a single variable inv_metric. For a diag_e metric the inv_metric value must be a vector of positive values, one for each parameter in the system. For a dense_e metric, inv_metric value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.\nThe metric_file option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n\n\n\nThe total integration time is determined by the argument engine which take possible values:\n\nnuts - the No-U-Turn Sampler which dynamically determines the optimal integration time.\nstatic - an HMC sampler which uses a user-specified integration time.\n\nThe default argument is engine=nuts.\nThe NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.\nWhen engine=nuts, the subargument max_depth can be used to control the depth of the tree. The default argument is max_depth=10. In the case where a model has a difficult posterior from which to sample, max_depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\nWhen the argument engine=static is specified, the user must specify the integration time via keyword int_time which takes as a value a positive number. The default value is \\(2\\pi\\).\n\n\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.\n\n\n\nA Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.\nThe preferred way of using multiple chains is to run them all from the same executable using the num_chains argument. There is also the option to use the Unix or DOS shell to run multiple executables.\n\n\nThe num_chains argument can be used for all of Stan’s samplers with the exception of the static HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with STAN_THREADS=true, these will either run in parallel or one after the other.\nWhen num_chains is greather than 1 (the default), arguments related to filenames (e.g. output file=, init=) can accept a comma separated list of values, one per each chain.\nFor example, sample will specify the names of the three chain’s output files.\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv\nThis will write the output in output_1.csv, output_2.csv, output_3.csv.\nIf the model was not compiled with STAN_THREADS=true, the above command will run 3 chains sequentially.\nIf the model was compiled with STAN_THREADS=true, the chains can run in parallel, with the num_threads argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (map_rect or reduce_sum calls), the below command will run 3 chains in parallel, provided there are cores available:\n./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4\nIf the model uses within-chain parallelization (map_rect or reduce_sum calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the Threading Building Blocks scheduler.\n./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16\n\n\nIf a comma separated list is not used, the num_chains argument changes the normal meanings of filename arguments when it is greater than 1 (the default). They are now interpreted as a “template” which is used for each chain.\nFor example, when num_chains=2, the argument output file=foo.csv no longer produces a file foo.csv, but instead produces two files, foo_1.csv and foo_2.csv. If you also supply id=5, the files produced will be foo_5.csv and foo_6.csv – id=5 gives the id of the first chain, and the remaining chains are sequential from there.\nThis also applies to input files, like those used for initialization. For example, if num_chains=3 and init=bar.json will first look for bar_1.json. If it exists, it will use bar_1.json for the first chain, bar_2.json for the second, and so on. If bar_1.json does not exist, it falls back to looking for bar.json, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the id argument, which defaults to 1.\nFor example, this shorthand is equivalent to the example given above:\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv\nA suffix with the chain id is appended to the provided output filename (output.csv in the above command), so this will also produce files output_1.csv, output_2.csv, output_3.csv.\n\n\n\n\n\nThe stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\nTo run stansummary on the output files generated by the for loop above, by the above run of the bernoulli model on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary output_*.csv\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe output_*.csv\nThe stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable theta:\nInference for Stan model: bernoulli_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total\nSampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total\n\n                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0\naccept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0\nstepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan\ntreedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0\nn_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0\ndivergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan\nenergy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0\n\ntheta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nIn this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for theta reflect the uncertainty in our estimate, due to the small amount of data, given the prior of beta(1, 1)\n\n\n\nNote: Many of these examples can be simplified by using the num_chains argument.\nWhen the num_chains argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.\nTo run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell for loop to set up index variables needed to identify each chain and its outputs.\nOn MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:\nfor NAME [in LIST]; do COMMANDS; done\nThe list can be a simple sequence of numbers, or you can use the shell expansion syntax {1..N} which expands to the sequence from \\(1\\) to \\(N\\), e.g. {1..4} expands to 1 2 3 4. Note that the expression {1..N} cannot contain spaces.\nTo run 4 chains for the example bernoulli model on MacOS or Linux:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv\n    done\nThe backslash (\\) indicates a line continuation in Unix. The expression ${i} substitutes in the value of loop index variable i. To run chains in parallel, put an ampersand (&) at the end of the nested sampler command:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv &\n    done\nThis pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.\nOn Windows, the DOS for-loop syntax is one of:\nfor %i in (SET) do COMMAND COMMAND-ARGUMENTS\nfor /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS\nTo run 4 chains in parallel on Windows:\n&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^\n                                    data file=bernoulli.data.json my_data ^\n                                    output file=output_%i.csv\nThe caret (^) indicates a line continuation in DOS. The expression %i is the loop index.\nIn the following extended examples, we focus on just the nested sampler command for Unix.\n\n\nFor reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.\nThe RNG seed is specified by random seed=&lt;int&gt; and the offset is specified by id=&lt;loop index&gt;, so the call to the sampler is:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            random seed=12345 id=${i}\n\n\n\nThe warmup and sampling iteration keyword-value arguments must follow the sample keyword. The call to the sampler which overrides the default warmup and sampling iterations is:\n./my_model sample num_warmup=500 num_sampling=500 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nTo save warmup draws as part of the Stan CSV output file, use the keyword-value argument save_warmup=true. This must be grouped with the other sample keyword sub-arguments.\n./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nBy default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range \\([{-2}, 2]\\). To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument init=&lt;filepath&gt;:\n./my_model sample init=my_param_inits.json data file=my_model.data.json \\\n           output file=output_${i}.csv\nTo verify that the specified values will be used by the sampler, you can run the sampler with option algorithm=fixed_param, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting num_warmp=0 and num_samples=1 saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.\nFor example, if we run the example Bernoulli model with specified initial value for parameter “theta”:\n{ \"theta\" : 0.5 }\nvia command:\n./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \\\n            init=bernoulli.init.json data file=bernoulli.data.json\nThe resulting output CSV file contains a single draw:\nlp__,accept_stat__,theta\n0,0,0.5\n#\n#  Elapsed Time: 0 seconds (Warm-up)\n#                0 seconds (Sampling)\n#                0 seconds (Total)\n#\n\n\n\nAn initial estimate for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a variable inv_metric. The metric_file option can be used with and without adaptation enabled.\nBy default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file bernoulli.diag_e.json, contents\n{ \"inv_metric\" : [0.296291] }\ncan be used as the initial metric as follows:\n../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nThe resulting output CSV file will contain the following set of comment lines:\n# Adaptation terminated\n# Step size = 0.9\n# Diagonal elements of inverse mass matrix:\n# 0.296291\nAs of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument save_metric, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the stepsize argument as well as the metric_file arguments; the former is the value of the stepsize element in the saved metric file, and the later is the metric file path.\n\n\n\nThe keyword-value arguments for these settings are grouped together under the adapt keyword which itself is a sub-argument of the sample keyword.\nModels with difficult posterior geometries may required increasing the delta argument closer to \\(1\\).\n./my_model sample adapt delta=0.95 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nTo skip adaptation altogether, use the keyword-value argument engaged=false. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nEven with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the typical set.\nTo skip warmup altogether requires specifying both num_warmup=0 and adapt engaged=false.\n../my_model sample num_warmup=0 adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nModels with difficult posterior geometries may required increasing the max_depth argument from its default value \\(10\\). This requires specifying a series of keyword-argument pairs:\n./my_model sample adapt delta=0.95 \\\n            algorithm=hmc engine=nuts max_depth=15 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; write the sampler parameters and gradients of all model parameters for each draw to a CSV file:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            diagnostic_file=diagnostics_${i}.csv\n\n\n\nThe output keyword sub-argument refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is \\(100\\) iterations. The progress updates look like:\nIteration:    1 / 2000 [  0%]  (Warmup)\nIteration:  100 / 2000 [  5%]  (Warmup)\nIteration:  200 / 2000 [ 10%]  (Warmup)\nIteration:  300 / 2000 [ 15%]  (Warmup)\nFor simple models which fit quickly, such updates can be annoying; to suppress them altogether, set refresh=0. This only turns off the Iteration: messages; the configuration and timing information are still written to the terminal.\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            refresh=0\nFor complicated models which take a long time to fit, setting the refresh rate to a low number, e.g. \\(10\\) or even \\(1\\), provides a way to more closely monitor the sampler.\n\n\n\nThe CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:\n./bernoulli sample data file=bernoulli.data.json adapt delta=0.95\nadapt is either mistyped or misplaced.\nPerhaps you meant one of the following valid configurations?\n  method=sample sample adapt\n  method=variational variational adapt\nFailed to parse arguments, terminating Stan\nThe following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.\n./my_model sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100 \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           random seed=12345 id=${i}\nThe keywords sample, data, output, and random are the top-level argument groups. Within the sample config arguments, the keyword adapt groups the adaptation algorithm parameters and the keyword-value algorithm=hmc groups the NUTS-HMC parameters.\nThe top-level groups can be freely ordered with respect to one another. The following is also a valid command:\n./my_model random seed=12345 id=${i} \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#running-the-sampler",
+    "href": "cmdstan-guide/mcmc_config.html#running-the-sampler",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "To generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument sample or method=sample together with the input data. The executable can be run from any directory.\nThe full set of configuration options available for the sample method is available by using the sample help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.\nHere, we run it in the directory which contains the Stan program and input data, &lt;cmdstan-home&gt;/examples/bernoulli:\n&gt; cd examples/bernoulli\n&gt; ls\n  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan\nTo execute sampling of the model under Linux or Mac, use:\n&gt; ./bernoulli sample data file=bernoulli.data.json\nIn Windows, the ./ prefix is not needed:\n&gt; bernoulli.exe sample data file=bernoulli.data.json\nThe output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:\nmethod = sample (Default)\n  sample\n    num_samples = 1000 (Default)\n    num_warmup = 1000 (Default)\n    save_warmup = false (Default)\n    thin = 1 (Default)\n    adapt\n      engaged = true (Default)\n      gamma = 0.050000000000000003 (Default)\n      delta = 0.80000000000000004 (Default)\n      kappa = 0.75 (Default)\n      t0 = 10 (Default)\n      init_buffer = 75 (Default)\n      term_buffer = 50 (Default)\n      window = 25 (Default)\n      save_metric = false (Default)\n    algorithm = hmc (Default)\n      hmc\n        engine = nuts (Default)\n          nuts\n            max_depth = 10 (Default)\n        metric = diag_e (Default)\n        metric_file =  (Default)\n        stepsize = 1 (Default)\n        stepsize_jitter = 0 (Default)\n    num_chains = 1 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 3252652196 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\nAfter the configuration has been displayed, a short timing message is given.\nGradient evaluation took 1.2e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.\nAdjust your expectations accordingly!\nNext, the sampler reports the iteration number, reporting the percentage complete.\nIteration:    1 / 2000 [  0%]  (Warmup)\n...\nIteration: 2000 / 2000 [100%]  (Sampling)\nFinally, the sampler reports timing information:\n Elapsed Time: 0.007 seconds (Warm-up)\n               0.017 seconds (Sampling)\n               0.024 seconds (Total)",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#mcmc_output_csv",
+    "href": "cmdstan-guide/mcmc_config.html#mcmc_output_csv",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "Each execution of the model results in draws from a single Markov chain being written to a file in comma-separated value (CSV) format. The default name of the output file is output.csv.\nThe first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (#)).\nWhen the example model bernoulli.stan is run via the command line with all default arguments, the following configuration is displayed:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 1000 (Default)\n#     num_warmup = 1000 (Default)\n#     save_warmup = false (Default)\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000 (Default)\n#       delta = 0.800000 (Default)\n#       kappa = 0.750000 (Default)\n#       t0 = 10.000000 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#       save_metric = false (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1.000000 (Default)\n#         stepsize_jitter = 0.000000 (Default)\n#     num_chains = 1 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\nThis is followed by a CSV header indicating the names of the values sampled.\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nThe first output columns report the HMC sampler information:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause the above header is from the NUTS sampler, it has columns treedepth__, n_leapfrog__, and divergent__ and doesn’t have column int_time__. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, theta.\nThe header line is written to the output file before warmup begins. If option save_warmup is set to true, the warmup draws are output directly after the header. The total number of warmup draws saved is num_warmup divided by thin, rounded up (i.e., ceiling).\nFollowing the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:\n# Adaptation terminated\n# Step size = 0.884484\n# Diagonal elements of inverse mass matrix:\n# 0.535006\nThe default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter theta.\nDraws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.\n-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853\n-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295\n-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299\n-6.88712,1,0.884484,1,1,0,7.02101,0.188229\n-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596\n...\nThe output ends with timing details:\n#  Elapsed Time: 0.007 seconds (Warm-up)\n#                0.017 seconds (Sampling)\n#                0.024 seconds (Total)",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#iterations",
+    "href": "cmdstan-guide/mcmc_config.html#iterations",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "At every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters metric and stepsize in order to efficiently sample from typical set, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.\nThe following keyword-value arguments control the total number of iterations:\n\nnum_samples\nnum_warmup\nsave_warmup\nthin\n\nThe values for arguments num_samples and num_warmup must be a non-negative integer. The default value for both is \\(1000\\).\nFor well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.\nThe number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is \\(\\sqrt{N}\\); therefore every additional decimal place of accuracy increases this by a factor of 10.\nArgument save_warmup takes values false or true. The default value is false, i.e., warmup draws are not saved to the output file. When the value is true, the warmup draws are written to the CSV output file directly after the CSV header line.\nArgument thin controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce anticorrelated draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.\nThe value of argument thin must be a positive integer. When thin is set to value \\(N\\), every \\(N^{th}\\) iteration is written to the output file. Should the value of thin exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of thin equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When num_sampling=M and thin=N, the number of iterations written to the output CSV file will be ceiling(M/N). If save_warmup=true, thinning is applied to the warmup iterations as well.",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#adaptation",
+    "href": "cmdstan-guide/mcmc_config.html#adaptation",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "The adapt keyword is used to specify non-default options for the sampler adaptation schedule and settings.\nAdaptation can be turned off by setting sub-argument engaged to value false. If engaged=false, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is engaged=1, this keyword-value pair can be omitted from the command.\nThere are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section Automatic Parameter Tuning.\nThe boolean sub-argument save_metric was added in Stan version 2.34. When save_metric=true, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix _metric.json, e.g., if using the default output filename output.csv, the saved metric file will be output_metric.json. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument metric_file.\n\n\nThe Stan User’s Guide section on model conditioning and curvature provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on HMC algorithm parameters explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.\nThe following keyword-value arguments control the settings used to optimize the step size:\n\ndelta - The target Metropolis acceptance rate. The default value is \\(0.8\\). Its value must be strictly between \\(0\\) and \\(1\\). Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of delta will also allow some models that would otherwise get stuck to overcome their blockages.  Models with difficult posterior geometries may required increasing the delta argument closer to \\(1\\); we recommend first trying to raise it to \\(0.9\\) or at most \\(0.95\\). Values about \\(0.95\\) are strong indication of bad geometry; the better solution is to change the model geometry through reparameterization which could yield both more efficient and faster sampling.\ngamma - Adaptation regularization scale. Must be a positive real number, default value is \\(0.05\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nkappa - Adaptation relaxation exponent. Must be a positive real number, default value is \\(0.75\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nt_0 - Adaptation iteration offset. Must be a positive real number, default value is \\(10\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\n\n\n\n\nWhen adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:\n\ninit_buffer - The number of iterations spent tuning the step size at the outset of adaptation.\nwindow - The initial number of iterations devoted to tune the metric, will be doubled successively.\nterm_buffer - The number of iterations used to re-tune the step size once the metric has been tuned.\n\nThe specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.\nThe following figure is taken from the Stan Reference Manual, where label “I” correspond to init_buffer, the initial “II” corresponds to window, and the final “III” corresponds to term_buffer:\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#algorithm",
+    "href": "cmdstan-guide/mcmc_config.html#algorithm",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "The algorithm keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: hmc, which generates from an HMC-driven Markov chain; and fixed_param which generates a new sample without changing the state of the Markov chain. The default argument is algorithm=hmc.\n\n\nIf a model doesn’t specify any parameters, then argument algorithm=fixed_param is mandatory.\nThe fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.\n\n\n\nAll HMC algorithms have three parameters:\n\nstep size\nmetric\nintegration time - the number of steps taken along the Hamiltonian trajectory\n\nSee the Stan Reference Manual section on HMC algorithm parameters for further details.\n\n\nThe HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.\nThe following keyword-value arguments control the step size:\n\nstepsize - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is \\(1\\).\nstepsize_jitter - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between \\(0\\) and \\(1\\). The default value is \\(0\\). Setting stepsize_jitter to \\(1\\) causes step sizes to be selected in the range of \\(0\\) to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.\n\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nThe metric argument specifies the choice of Euclidean HMC implementations:\n\nmetric=unit specifies unit metric (diagonal matrix of ones).\nmetric=diag_e specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.\nmetric=dense_e specifies a dense metric (a dense, symmetric positive definite matrix).\n\nBy default, the metric is estimated during warmup. However, when metric=diag_e or metric=dense_e, an initial guess for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a single variable inv_metric. For a diag_e metric the inv_metric value must be a vector of positive values, one for each parameter in the system. For a dense_e metric, inv_metric value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.\nThe metric_file option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n\n\n\nThe total integration time is determined by the argument engine which take possible values:\n\nnuts - the No-U-Turn Sampler which dynamically determines the optimal integration time.\nstatic - an HMC sampler which uses a user-specified integration time.\n\nThe default argument is engine=nuts.\nThe NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.\nWhen engine=nuts, the subargument max_depth can be used to control the depth of the tree. The default argument is max_depth=10. In the case where a model has a difficult posterior from which to sample, max_depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\nWhen the argument engine=static is specified, the user must specify the integration time via keyword int_time which takes as a value a positive number. The default value is \\(2\\pi\\).",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#sampler-diag-file",
+    "href": "cmdstan-guide/mcmc_config.html#sampler-diag-file",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "The output keyword sub-argument diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#multi-chain-sampling",
+    "href": "cmdstan-guide/mcmc_config.html#multi-chain-sampling",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "A Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.\nThe preferred way of using multiple chains is to run them all from the same executable using the num_chains argument. There is also the option to use the Unix or DOS shell to run multiple executables.\n\n\nThe num_chains argument can be used for all of Stan’s samplers with the exception of the static HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with STAN_THREADS=true, these will either run in parallel or one after the other.\nWhen num_chains is greather than 1 (the default), arguments related to filenames (e.g. output file=, init=) can accept a comma separated list of values, one per each chain.\nFor example, sample will specify the names of the three chain’s output files.\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv\nThis will write the output in output_1.csv, output_2.csv, output_3.csv.\nIf the model was not compiled with STAN_THREADS=true, the above command will run 3 chains sequentially.\nIf the model was compiled with STAN_THREADS=true, the chains can run in parallel, with the num_threads argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (map_rect or reduce_sum calls), the below command will run 3 chains in parallel, provided there are cores available:\n./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4\nIf the model uses within-chain parallelization (map_rect or reduce_sum calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the Threading Building Blocks scheduler.\n./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16\n\n\nIf a comma separated list is not used, the num_chains argument changes the normal meanings of filename arguments when it is greater than 1 (the default). They are now interpreted as a “template” which is used for each chain.\nFor example, when num_chains=2, the argument output file=foo.csv no longer produces a file foo.csv, but instead produces two files, foo_1.csv and foo_2.csv. If you also supply id=5, the files produced will be foo_5.csv and foo_6.csv – id=5 gives the id of the first chain, and the remaining chains are sequential from there.\nThis also applies to input files, like those used for initialization. For example, if num_chains=3 and init=bar.json will first look for bar_1.json. If it exists, it will use bar_1.json for the first chain, bar_2.json for the second, and so on. If bar_1.json does not exist, it falls back to looking for bar.json, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the id argument, which defaults to 1.\nFor example, this shorthand is equivalent to the example given above:\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv\nA suffix with the chain id is appended to the provided output filename (output.csv in the above command), so this will also produce files output_1.csv, output_2.csv, output_3.csv.",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#summarizing-sampler-outputs-with-stansummary",
+    "href": "cmdstan-guide/mcmc_config.html#summarizing-sampler-outputs-with-stansummary",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\nTo run stansummary on the output files generated by the for loop above, by the above run of the bernoulli model on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary output_*.csv\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe output_*.csv\nThe stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable theta:\nInference for Stan model: bernoulli_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total\nSampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total\n\n                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0\naccept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0\nstepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan\ntreedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0\nn_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0\ndivergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan\nenergy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0\n\ntheta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nIn this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for theta reflect the uncertainty in our estimate, due to the small amount of data, given the prior of beta(1, 1)",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/mcmc_config.html#old-multichain",
+    "href": "cmdstan-guide/mcmc_config.html#old-multichain",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "section": "",
+    "text": "Note: Many of these examples can be simplified by using the num_chains argument.\nWhen the num_chains argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.\nTo run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell for loop to set up index variables needed to identify each chain and its outputs.\nOn MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:\nfor NAME [in LIST]; do COMMANDS; done\nThe list can be a simple sequence of numbers, or you can use the shell expansion syntax {1..N} which expands to the sequence from \\(1\\) to \\(N\\), e.g. {1..4} expands to 1 2 3 4. Note that the expression {1..N} cannot contain spaces.\nTo run 4 chains for the example bernoulli model on MacOS or Linux:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv\n    done\nThe backslash (\\) indicates a line continuation in Unix. The expression ${i} substitutes in the value of loop index variable i. To run chains in parallel, put an ampersand (&) at the end of the nested sampler command:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv &\n    done\nThis pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.\nOn Windows, the DOS for-loop syntax is one of:\nfor %i in (SET) do COMMAND COMMAND-ARGUMENTS\nfor /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS\nTo run 4 chains in parallel on Windows:\n&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^\n                                    data file=bernoulli.data.json my_data ^\n                                    output file=output_%i.csv\nThe caret (^) indicates a line continuation in DOS. The expression %i is the loop index.\nIn the following extended examples, we focus on just the nested sampler command for Unix.\n\n\nFor reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.\nThe RNG seed is specified by random seed=&lt;int&gt; and the offset is specified by id=&lt;loop index&gt;, so the call to the sampler is:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            random seed=12345 id=${i}\n\n\n\nThe warmup and sampling iteration keyword-value arguments must follow the sample keyword. The call to the sampler which overrides the default warmup and sampling iterations is:\n./my_model sample num_warmup=500 num_sampling=500 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nTo save warmup draws as part of the Stan CSV output file, use the keyword-value argument save_warmup=true. This must be grouped with the other sample keyword sub-arguments.\n./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nBy default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range \\([{-2}, 2]\\). To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument init=&lt;filepath&gt;:\n./my_model sample init=my_param_inits.json data file=my_model.data.json \\\n           output file=output_${i}.csv\nTo verify that the specified values will be used by the sampler, you can run the sampler with option algorithm=fixed_param, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting num_warmp=0 and num_samples=1 saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.\nFor example, if we run the example Bernoulli model with specified initial value for parameter “theta”:\n{ \"theta\" : 0.5 }\nvia command:\n./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \\\n            init=bernoulli.init.json data file=bernoulli.data.json\nThe resulting output CSV file contains a single draw:\nlp__,accept_stat__,theta\n0,0,0.5\n#\n#  Elapsed Time: 0 seconds (Warm-up)\n#                0 seconds (Sampling)\n#                0 seconds (Total)\n#\n\n\n\nAn initial estimate for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a variable inv_metric. The metric_file option can be used with and without adaptation enabled.\nBy default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file bernoulli.diag_e.json, contents\n{ \"inv_metric\" : [0.296291] }\ncan be used as the initial metric as follows:\n../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nThe resulting output CSV file will contain the following set of comment lines:\n# Adaptation terminated\n# Step size = 0.9\n# Diagonal elements of inverse mass matrix:\n# 0.296291\nAs of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument save_metric, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the stepsize argument as well as the metric_file arguments; the former is the value of the stepsize element in the saved metric file, and the later is the metric file path.\n\n\n\nThe keyword-value arguments for these settings are grouped together under the adapt keyword which itself is a sub-argument of the sample keyword.\nModels with difficult posterior geometries may required increasing the delta argument closer to \\(1\\).\n./my_model sample adapt delta=0.95 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nTo skip adaptation altogether, use the keyword-value argument engaged=false. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nEven with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the typical set.\nTo skip warmup altogether requires specifying both num_warmup=0 and adapt engaged=false.\n../my_model sample num_warmup=0 adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nModels with difficult posterior geometries may required increasing the max_depth argument from its default value \\(10\\). This requires specifying a series of keyword-argument pairs:\n./my_model sample adapt delta=0.95 \\\n            algorithm=hmc engine=nuts max_depth=15 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; write the sampler parameters and gradients of all model parameters for each draw to a CSV file:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            diagnostic_file=diagnostics_${i}.csv\n\n\n\nThe output keyword sub-argument refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is \\(100\\) iterations. The progress updates look like:\nIteration:    1 / 2000 [  0%]  (Warmup)\nIteration:  100 / 2000 [  5%]  (Warmup)\nIteration:  200 / 2000 [ 10%]  (Warmup)\nIteration:  300 / 2000 [ 15%]  (Warmup)\nFor simple models which fit quickly, such updates can be annoying; to suppress them altogether, set refresh=0. This only turns off the Iteration: messages; the configuration and timing information are still written to the terminal.\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            refresh=0\nFor complicated models which take a long time to fit, setting the refresh rate to a low number, e.g. \\(10\\) or even \\(1\\), provides a way to more closely monitor the sampler.\n\n\n\nThe CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:\n./bernoulli sample data file=bernoulli.data.json adapt delta=0.95\nadapt is either mistyped or misplaced.\nPerhaps you meant one of the following valid configurations?\n  method=sample sample adapt\n  method=variational variational adapt\nFailed to parse arguments, terminating Stan\nThe following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.\n./my_model sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100 \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           random seed=12345 id=${i}\nThe keywords sample, data, output, and random are the top-level argument groups. Within the sample config arguments, the keyword adapt groups the adaptation algorithm parameters and the keyword-value algorithm=hmc groups the NUTS-HMC parameters.\nThe top-level groups can be freely ordered with respect to one another. The following is also a valid command:\n./my_model random seed=12345 id=${i} \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/laplace_sample_config.html",
+    "href": "cmdstan-guide/laplace_sample_config.html",
+    "title": "Laplace sampling",
+    "section": "",
+    "text": "The laplace method produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood. In general, the posterior mode in the unconstrained space doesn’t correspond to the mean (nor mode) in the constrained space, and thus the sample is needed to infer the mean as well as the standard deviation. (See this case study for a visual illustration.)\nThis is computationally inexpensive compared to exact Bayesian inference with MCMC. The goodness of this estimate depends on both the estimate of the mode and how much the true posterior in the unconstrained space resembles a Gaussian.\n\n\nThis method takes several arguments:\n\nmode - Input file of parameters values on the constrained scale. When Stan’s optimize method is used to estimate the modal values, the value of boolean argument jacobian should be false if optimize was run with default settings, i.e., the input is the MLE estimate; if optimize was run with argument jacobian=true, then the laplace method default setting, jacobian=true, should be used.\njacobian - Whether or not the Jacobian adjustment should be included in the gradient. The default value is true (include adjustment). (Note: in optimization, the default value is false, for historical reasons.)\ndraws - How many total draws to return. The default is \\(1000\\).\ncalculate_lp - Whether to calculate the log probability of the model at each draw. If this is false, the log_p__ column of the output will be entirely nan. The default value is true.\n\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the laplace method is reported at the beginning of the output file as CSV comments.\nOutput columns log_p__ and log_q__, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.\nOutput columns for all model parameters on the constrained scale.\n\n\n\n\nIf requested with output diagnostic_file=, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.\n\n\n\nTo get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:\n\nfind the MAP estimate by running optimization with argument jacobian=true\nrun the Laplace estimator using the MAP estimate as the mode argument.\n\nBecause the default output file name from all methods is output.csv, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.\n./examples/bernoulli/bernoulli optimize jacobian=1 \\\n  data file=examples/bernoulli/bernoulli.data.json \\\n  output file=bernoulli_optimize_lbfgs.csv random seed=1234\n\n\n./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \\\n data file=examples/bernoulli/bernoulli.data.json random seed=1234\nThe header and first few data rows of the output sample are shown below.\n# method = laplace\n#   laplace\n#     mode = bernoulli_lbfgs.csv\n#     jacobian = true (Default)\n#     draws = 1000 (Default)\n#     calculate_lp = true (default)\n# id = 1 (Default)\n# data\n#   file = examples/bernoulli/bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 875960551 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlog_p__,log_q__,theta\n-9.4562,-2.33997,0.0498545\n-6.9144,-0.0117349,0.182898\n-7.18171,-0.746034,0.376428\n...",
+    "crumbs": [
+      "Running CmdStan",
+      "Laplace sampling"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/laplace_sample_config.html#configuration",
+    "href": "cmdstan-guide/laplace_sample_config.html#configuration",
+    "title": "Laplace sampling",
+    "section": "",
+    "text": "This method takes several arguments:\n\nmode - Input file of parameters values on the constrained scale. When Stan’s optimize method is used to estimate the modal values, the value of boolean argument jacobian should be false if optimize was run with default settings, i.e., the input is the MLE estimate; if optimize was run with argument jacobian=true, then the laplace method default setting, jacobian=true, should be used.\njacobian - Whether or not the Jacobian adjustment should be included in the gradient. The default value is true (include adjustment). (Note: in optimization, the default value is false, for historical reasons.)\ndraws - How many total draws to return. The default is \\(1000\\).\ncalculate_lp - Whether to calculate the log probability of the model at each draw. If this is false, the log_p__ column of the output will be entirely nan. The default value is true.",
+    "crumbs": [
+      "Running CmdStan",
+      "Laplace sampling"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/laplace_sample_config.html#csv-output",
+    "href": "cmdstan-guide/laplace_sample_config.html#csv-output",
+    "title": "Laplace sampling",
+    "section": "",
+    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the laplace method is reported at the beginning of the output file as CSV comments.\nOutput columns log_p__ and log_q__, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.\nOutput columns for all model parameters on the constrained scale.",
+    "crumbs": [
+      "Running CmdStan",
+      "Laplace sampling"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/laplace_sample_config.html#diagnostic-file-outputs",
+    "href": "cmdstan-guide/laplace_sample_config.html#diagnostic-file-outputs",
+    "title": "Laplace sampling",
+    "section": "",
+    "text": "If requested with output diagnostic_file=, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.",
+    "crumbs": [
+      "Running CmdStan",
+      "Laplace sampling"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/laplace_sample_config.html#example",
+    "href": "cmdstan-guide/laplace_sample_config.html#example",
+    "title": "Laplace sampling",
+    "section": "",
+    "text": "To get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:\n\nfind the MAP estimate by running optimization with argument jacobian=true\nrun the Laplace estimator using the MAP estimate as the mode argument.\n\nBecause the default output file name from all methods is output.csv, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.\n./examples/bernoulli/bernoulli optimize jacobian=1 \\\n  data file=examples/bernoulli/bernoulli.data.json \\\n  output file=bernoulli_optimize_lbfgs.csv random seed=1234\n\n\n./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \\\n data file=examples/bernoulli/bernoulli.data.json random seed=1234\nThe header and first few data rows of the output sample are shown below.\n# method = laplace\n#   laplace\n#     mode = bernoulli_lbfgs.csv\n#     jacobian = true (Default)\n#     draws = 1000 (Default)\n#     calculate_lp = true (default)\n# id = 1 (Default)\n# data\n#   file = examples/bernoulli/bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 875960551 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlog_p__,log_q__,theta\n-9.4562,-2.33997,0.0498545\n-6.9144,-0.0117349,0.182898\n-7.18171,-0.746034,0.376428\n...",
+    "crumbs": [
+      "Running CmdStan",
+      "Laplace sampling"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html",
+    "href": "cmdstan-guide/installation.html",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "There are a few ways that you can install CmdStan. Depending on your operating system and your level of expertise, you can either:\n\nUse the conda package management system to install a pre-built version of CmdStan along with the required dependencies. Recommended for Windows users.\nInstall the source code from GitHub CmdStan repository. This requires a modern C++ compiler and toolchain. See the C++ Toolchain section for further details.\n\n\n\nWith conda, you can install CmdStan from the conda-forge channel. This will install a pre-built version of CmdStan along with the required dependencies (i.e. a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command cmdstan_model to activate the CmdStan makefile from anywhere.\nNote: This requires that conda has been installed already on your machine. We recommend using the miniforge distribution.\nWe recommend installing CmdStan in a new conda environment:\n conda create -n stan -c conda-forge cmdstan\nThis command creates a new conda environment named stan and downloads and installs the cmdstan package as well as CmdStan and the required C++ toolchain.\nTo install into an existing conda environment, use the conda install command instead of create:\n conda install -c conda-forge cmdstan\nWhichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment stan, run the command\n conda activate stan\nBy default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and newer can be installed by specifying cmdstan==VERSION in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment\nconda install -c conda-forge cmdstan=2.27.0\n\n\nA Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command\n conda info -e\nThe shell environment variable CONDA_PREFIX points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the bin subdirectory of the conda environment directory, i.e., $CONDA_PREFIX/bin/cmdstan (Linux, MacOS), %CONDA_PREFIX%\\bin\\cmdstan (Windows).\nPlease report conda-specific install problems directly to the conda-forge issue tracker, here.\n\n\n\n\nInstallation from GitHub consists of the following steps:\n\nVerify that you have a modern C++ toolchain. See the C++ Toolchain section for details.\nDownload the CmdStan source code from GitHub\nBuild the CmdStan libraries and executables\nCheck the installation by compiling and running the CmdStan example model bernoulli.stan.\n\n\n\nThe GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the cmdstan module; the Stan inference engine algorithms and Stan math library functions are specified as submodules and stored in the GitHub repositories stan and math, respectively.\nA CmdStan release is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do not contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-”, e.g. “cmdstan-2.29.2”.\nBy cloning the CmdStan repository with argument --recursive, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named cmdstan:\n&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive\nThroughout this manual, we refer to this top-level CmdStan source directory as &lt;cmdstan-home&gt;. This directory contains the following subdirectories:\n\ndirectory cmdstan/stan contains the sub-module stan (https://github.com/stan-dev/stan)\ndirectory cmdstan/stan/lib/stan_math contains the sub-module math (https://github.com/stan-dev/math)\n\n\n\n\nBuilding CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:\n\nstanc: the Stan compiler (translates Stan language to C++).\nstansummary: a basic posterior analysis tool. The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\ndiagnose: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.\n\nCmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): bin/linux-stanc, bin/mac-stanc and bin/windows-stanc. The CmdStan makefile build task copies the appropriate binary to bin/stanc. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.\nSteps to build CmdStan:\n\nOpen a command-line terminal window and change directories to the CmdStan home directory.\nRun the makefile target build which instantiates the CmdStan utilities and compiles all necessary C++ libraries.\n\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make build\nIf your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the -j option. For example, to build on 4 cores, type:\n&gt; make -j4 build\nWhen make build is successful, the directory &lt;cmdstan-home&gt;/bin/ will contain the executables stanc, stansummary, and diagnose (on Windows, corresponding .exe files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:\n--- CmdStan v2.29.2 built ---\nWarning: The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.\nWindows only: CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory &lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb is part of the PATH environment variable. See these instructions for details on changing the PATH. To permanently make this setting for the current user, you may execute:\n&gt; make install-tbb\nAfter changing the PATH environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)\n\n\n\n\nTo check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.\nOn Linux and macOS:\n# compile the example\n&gt; make examples/bernoulli/bernoulli\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli sample\\\n  data file=examples/bernoulli/bernoulli.data.json\n\n# default output written to file `output.csv`,\n# default num_samples is 1000, output file should have approx. 1050 lines\n&gt; wc -l output.csv\n\n# run the `bin/stansummary utility to summarize parameter estimates\n&gt; bin/stansummary output.csv\nOn Windows:\n# compile the example\n&gt; make examples/bernoulli/bernoulli.exe\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json\n\n# run the `bin/stansummary.exe utility to summarize parameter estimates\n&gt; bin/stansummary.exe output.csv\nThe sample data in file bernoulli.json.data specifies 2 out of 10 successes, therefore the range mean(theta)\\(\\pm\\)sd(theta) should include 0.2.\n\n\n\nUpdates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the &lt;cmdstan-home&gt; directory:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make clean-all\n&gt; make build\n\n\nThis section contains solutions to problems reported on https://discourse.mc-stan.org\nCompiler error message about PCH file\nTo speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:\nerror: PCH file uses an older PCH format that is no longer supported\nIn this case, clean and rebuild CmdStan, as shown in the previous section.\nWindows: ‘g++’, ‘make’, or ‘cut’ is not recognized\nThe CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:\n'cut' is not recognized as an internal or external command,\noperable program or batch file.\nTo fix this, ensure you have followed the steps for adding the toolchain to your PATH and installing the additional utilities covered in the configuration instructions\nSpaces in paths to CmdStan or model\nmake can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the One Drive folder.\nUnfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:\nmake: *** INTERNAL: readdir: Invalid argument\nmake: *** [make/program:50: x.hpp] Error 2\nIf the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.\n\n\n\n\nCompiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.\n\n\nThe required C++ compiler is g++ 4.9 3. On most systems the GNU Make utility is pre-installed and is the default make utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:\ng++ --version\nmake --version\nIf these are at least at g++ version 4.9.3 or later and make version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler g++, because later versions are faster.\nTo install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:\nsudo apt install g++\nsudo apt install make\nIf you can’t run sudo, you will need to ask your sysadmin or cluster administrator to install these tools for you.\n\n\n\nTo check if you already already have an appropriate toolchain installed, open the Terminal application and enter:\nclang++ --version\nmake --version\nIf either of these commands prints the message command not found, you will need to install Xcode’s command line tools.\nOpen the Terminal application and enter:\nxcode-select --install\nSelect “Install” in the window that opens.\nAfter the installation completes, you can double check that installation was successful by reopening the Terminal and running:\nclang++ --version\nmake --version\nYou can read more about Xcode on its site: https://developer.apple.com/xcode/\nWe don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.\n\n\n\nThe Windows toolchain consists of programs g++, the C++ compiler, and make, the GNU Make utility. To check if these are present, open a command shell [^1] and type:\ng++ --version\nmake --version\nCmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your PATH variable, See these instructions for details on changing the PATH if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.\n\n\nAll required utilities (e.g., make, g++) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the R Project documentation.\nThe R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.\nNote that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a Microsoft update.\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45\\usr\\bin\nC:\\rtools45\\x86_64-w64-mingw32.static.posix\\bin\n\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45 - ARM64\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45-aarch64\\usr\\bin\nC:\\rtools45-aarch64\\aarch64-w64-mingw32.static.posix\\bin\n\n\n\n\n\n\nCmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.\nGNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed only when there are changes to those other files. Thus Make provides an efficient way to manage complex software.\nThe CmdStan Makefile is in the &lt;cmdstan-home&gt; directory and is named makefile. This is one of the default GNU Makefile names, which allows you to omit the -f makefile argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, Make only works properly when invoked from the &lt;cmdstan-home&gt; directory; attempts to use this Makefile from another directory by specifying the full path to the file makefile won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:\nmake -f ~/github/stan-dev/cmdstan/makefile\n/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory\nmake: *** No rule to make target `make/command'.  Stop.\nThe conda-forge cmdstan package provides a solution to this problem via cmdstan_model command which lets you run the CmdStan makefile from anywhere to compile a Stan model.\nMakefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix .stan; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program my_program.stan in directory ../my_dir/, the make target is ../my_dir/my_program or ../my_dir/my_program.exe (on Windows).\nTo call Make, you invoke the utility name, make, followed by, in order:\n\nzero or more Make program options, then specify any Make variables as a series of\nzero of more Make variables, described below\nzero or more target names; the set of names is determined by the Makefile rules.\n\nmake &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;\nMakefile Variables\nMake targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile ../my_dir/my_program.stan for an OpenCL (GPU) machine, set the makefile variable STAN_OPENCL to TRUE:\n&gt; make STAN_OPENCL=TRUE ../my_dir/my_program\nMakefile variables can also be set by creating a file named local in the CmdStan make subdirectory which contains a list of &lt;VARIABLE&gt;=&lt;VALUE&gt; pairs, one per line. For example, to get the same effect as the above command every time, you would put the line STAN_OPENCL=TRUE into the file &lt;cmdstan_home&gt;/make/local.\nThe complete set of Makefile variables can be found in file &lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags.\nMake Targets\nWhen invoked without any arguments at all, Make prints a help message:\n&gt; make\n--------------------------------------------------------------------------------\nCmdStan v2.33.1 help\n\n  Build CmdStan utilities:\n    &gt; make build\n\n    This target will:\n    1. Install the Stan compiler bin/stanc from stanc3 binaries.\n    2. Build the print utility bin/print (deprecated; will be removed in v3.0)\n    3. Build the stansummary utility bin/stansummary\n    4. Build the diagnose utility bin/diagnose\n    5. Build all libraries and object files compile and link an executable Stan program\n\n    Note: to build using multiple cores, use the -j option to make, e.g.,\n    for 4 cores:\n    &gt; make build -j4\n\n\n  Build a Stan program:\n\n    Given a Stan program at foo/bar.stan, build an executable by typing:\n    &gt; make foo/bar\n\n    This target will:\n    1. Install the Stan compiler (bin/stanc), as needed.\n    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.\n    3. Compile the C++ code using cc . to generate foo/bar\n\n  Additional make options:\n    STANCFLAGS: defaults to \"\". These are extra options passed to bin/stanc\n      when generating C++ code. If you want to allow undefined functions in the\n      Stan program, either add this to make/local or the command line:\n          STANCFLAGS = --allow_undefined\n    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the\n      header file that is included. This defaults to \"user_header.hpp\" in the\n      directory of the Stan program.\n    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads\n      the nightly version.\n    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.\n    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.\n\n\n  Example - bernoulli model: examples/bernoulli/bernoulli.stan\n\n    1. Build the model:\n       &gt; make examples/bernoulli/bernoulli\n    2. Run the sampling algorithm given the model and data:\n       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R\n    3. Look at the posterior sample:\n       &gt; bin/stansummary output.csv\n\n\n  Clean CmdStan:\n\n    Remove the built CmdStan tools:\n    &gt; make clean-all\n\n--------------------------------------------------------------------------------",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html#conda-install",
+    "href": "cmdstan-guide/installation.html#conda-install",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "With conda, you can install CmdStan from the conda-forge channel. This will install a pre-built version of CmdStan along with the required dependencies (i.e. a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command cmdstan_model to activate the CmdStan makefile from anywhere.\nNote: This requires that conda has been installed already on your machine. We recommend using the miniforge distribution.\nWe recommend installing CmdStan in a new conda environment:\n conda create -n stan -c conda-forge cmdstan\nThis command creates a new conda environment named stan and downloads and installs the cmdstan package as well as CmdStan and the required C++ toolchain.\nTo install into an existing conda environment, use the conda install command instead of create:\n conda install -c conda-forge cmdstan\nWhichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment stan, run the command\n conda activate stan\nBy default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and newer can be installed by specifying cmdstan==VERSION in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment\nconda install -c conda-forge cmdstan=2.27.0\n\n\nA Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command\n conda info -e\nThe shell environment variable CONDA_PREFIX points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the bin subdirectory of the conda environment directory, i.e., $CONDA_PREFIX/bin/cmdstan (Linux, MacOS), %CONDA_PREFIX%\\bin\\cmdstan (Windows).\nPlease report conda-specific install problems directly to the conda-forge issue tracker, here.",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html#installation-from-github",
+    "href": "cmdstan-guide/installation.html#installation-from-github",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "Installation from GitHub consists of the following steps:\n\nVerify that you have a modern C++ toolchain. See the C++ Toolchain section for details.\nDownload the CmdStan source code from GitHub\nBuild the CmdStan libraries and executables\nCheck the installation by compiling and running the CmdStan example model bernoulli.stan.\n\n\n\nThe GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the cmdstan module; the Stan inference engine algorithms and Stan math library functions are specified as submodules and stored in the GitHub repositories stan and math, respectively.\nA CmdStan release is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do not contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-”, e.g. “cmdstan-2.29.2”.\nBy cloning the CmdStan repository with argument --recursive, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named cmdstan:\n&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive\nThroughout this manual, we refer to this top-level CmdStan source directory as &lt;cmdstan-home&gt;. This directory contains the following subdirectories:\n\ndirectory cmdstan/stan contains the sub-module stan (https://github.com/stan-dev/stan)\ndirectory cmdstan/stan/lib/stan_math contains the sub-module math (https://github.com/stan-dev/math)\n\n\n\n\nBuilding CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:\n\nstanc: the Stan compiler (translates Stan language to C++).\nstansummary: a basic posterior analysis tool. The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\ndiagnose: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.\n\nCmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): bin/linux-stanc, bin/mac-stanc and bin/windows-stanc. The CmdStan makefile build task copies the appropriate binary to bin/stanc. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.\nSteps to build CmdStan:\n\nOpen a command-line terminal window and change directories to the CmdStan home directory.\nRun the makefile target build which instantiates the CmdStan utilities and compiles all necessary C++ libraries.\n\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make build\nIf your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the -j option. For example, to build on 4 cores, type:\n&gt; make -j4 build\nWhen make build is successful, the directory &lt;cmdstan-home&gt;/bin/ will contain the executables stanc, stansummary, and diagnose (on Windows, corresponding .exe files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:\n--- CmdStan v2.29.2 built ---\nWarning: The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.\nWindows only: CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory &lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb is part of the PATH environment variable. See these instructions for details on changing the PATH. To permanently make this setting for the current user, you may execute:\n&gt; make install-tbb\nAfter changing the PATH environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html#checking-the-stan-compiler",
+    "href": "cmdstan-guide/installation.html#checking-the-stan-compiler",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "To check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.\nOn Linux and macOS:\n# compile the example\n&gt; make examples/bernoulli/bernoulli\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli sample\\\n  data file=examples/bernoulli/bernoulli.data.json\n\n# default output written to file `output.csv`,\n# default num_samples is 1000, output file should have approx. 1050 lines\n&gt; wc -l output.csv\n\n# run the `bin/stansummary utility to summarize parameter estimates\n&gt; bin/stansummary output.csv\nOn Windows:\n# compile the example\n&gt; make examples/bernoulli/bernoulli.exe\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json\n\n# run the `bin/stansummary.exe utility to summarize parameter estimates\n&gt; bin/stansummary.exe output.csv\nThe sample data in file bernoulli.json.data specifies 2 out of 10 successes, therefore the range mean(theta)\\(\\pm\\)sd(theta) should include 0.2.",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html#troubleshooting-the-installation",
+    "href": "cmdstan-guide/installation.html#troubleshooting-the-installation",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "Updates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the &lt;cmdstan-home&gt; directory:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make clean-all\n&gt; make build\n\n\nThis section contains solutions to problems reported on https://discourse.mc-stan.org\nCompiler error message about PCH file\nTo speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:\nerror: PCH file uses an older PCH format that is no longer supported\nIn this case, clean and rebuild CmdStan, as shown in the previous section.\nWindows: ‘g++’, ‘make’, or ‘cut’ is not recognized\nThe CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:\n'cut' is not recognized as an internal or external command,\noperable program or batch file.\nTo fix this, ensure you have followed the steps for adding the toolchain to your PATH and installing the additional utilities covered in the configuration instructions\nSpaces in paths to CmdStan or model\nmake can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the One Drive folder.\nUnfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:\nmake: *** INTERNAL: readdir: Invalid argument\nmake: *** [make/program:50: x.hpp] Error 2\nIf the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html#cpp-toolchain",
+    "href": "cmdstan-guide/installation.html#cpp-toolchain",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "Compiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.\n\n\nThe required C++ compiler is g++ 4.9 3. On most systems the GNU Make utility is pre-installed and is the default make utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:\ng++ --version\nmake --version\nIf these are at least at g++ version 4.9.3 or later and make version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler g++, because later versions are faster.\nTo install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:\nsudo apt install g++\nsudo apt install make\nIf you can’t run sudo, you will need to ask your sysadmin or cluster administrator to install these tools for you.\n\n\n\nTo check if you already already have an appropriate toolchain installed, open the Terminal application and enter:\nclang++ --version\nmake --version\nIf either of these commands prints the message command not found, you will need to install Xcode’s command line tools.\nOpen the Terminal application and enter:\nxcode-select --install\nSelect “Install” in the window that opens.\nAfter the installation completes, you can double check that installation was successful by reopening the Terminal and running:\nclang++ --version\nmake --version\nYou can read more about Xcode on its site: https://developer.apple.com/xcode/\nWe don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.\n\n\n\nThe Windows toolchain consists of programs g++, the C++ compiler, and make, the GNU Make utility. To check if these are present, open a command shell [^1] and type:\ng++ --version\nmake --version\nCmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your PATH variable, See these instructions for details on changing the PATH if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.\n\n\nAll required utilities (e.g., make, g++) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the R Project documentation.\nThe R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.\nNote that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a Microsoft update.\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45\\usr\\bin\nC:\\rtools45\\x86_64-w64-mingw32.static.posix\\bin\n\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45 - ARM64\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45-aarch64\\usr\\bin\nC:\\rtools45-aarch64\\aarch64-w64-mingw32.static.posix\\bin",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/installation.html#gnu-make",
+    "href": "cmdstan-guide/installation.html#gnu-make",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "CmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.\nGNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed only when there are changes to those other files. Thus Make provides an efficient way to manage complex software.\nThe CmdStan Makefile is in the &lt;cmdstan-home&gt; directory and is named makefile. This is one of the default GNU Makefile names, which allows you to omit the -f makefile argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, Make only works properly when invoked from the &lt;cmdstan-home&gt; directory; attempts to use this Makefile from another directory by specifying the full path to the file makefile won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:\nmake -f ~/github/stan-dev/cmdstan/makefile\n/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory\nmake: *** No rule to make target `make/command'.  Stop.\nThe conda-forge cmdstan package provides a solution to this problem via cmdstan_model command which lets you run the CmdStan makefile from anywhere to compile a Stan model.\nMakefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix .stan; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program my_program.stan in directory ../my_dir/, the make target is ../my_dir/my_program or ../my_dir/my_program.exe (on Windows).\nTo call Make, you invoke the utility name, make, followed by, in order:\n\nzero or more Make program options, then specify any Make variables as a series of\nzero of more Make variables, described below\nzero or more target names; the set of names is determined by the Makefile rules.\n\nmake &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;\nMakefile Variables\nMake targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile ../my_dir/my_program.stan for an OpenCL (GPU) machine, set the makefile variable STAN_OPENCL to TRUE:\n&gt; make STAN_OPENCL=TRUE ../my_dir/my_program\nMakefile variables can also be set by creating a file named local in the CmdStan make subdirectory which contains a list of &lt;VARIABLE&gt;=&lt;VALUE&gt; pairs, one per line. For example, to get the same effect as the above command every time, you would put the line STAN_OPENCL=TRUE into the file &lt;cmdstan_home&gt;/make/local.\nThe complete set of Makefile variables can be found in file &lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags.\nMake Targets\nWhen invoked without any arguments at all, Make prints a help message:\n&gt; make\n--------------------------------------------------------------------------------\nCmdStan v2.33.1 help\n\n  Build CmdStan utilities:\n    &gt; make build\n\n    This target will:\n    1. Install the Stan compiler bin/stanc from stanc3 binaries.\n    2. Build the print utility bin/print (deprecated; will be removed in v3.0)\n    3. Build the stansummary utility bin/stansummary\n    4. Build the diagnose utility bin/diagnose\n    5. Build all libraries and object files compile and link an executable Stan program\n\n    Note: to build using multiple cores, use the -j option to make, e.g.,\n    for 4 cores:\n    &gt; make build -j4\n\n\n  Build a Stan program:\n\n    Given a Stan program at foo/bar.stan, build an executable by typing:\n    &gt; make foo/bar\n\n    This target will:\n    1. Install the Stan compiler (bin/stanc), as needed.\n    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.\n    3. Compile the C++ code using cc . to generate foo/bar\n\n  Additional make options:\n    STANCFLAGS: defaults to \"\". These are extra options passed to bin/stanc\n      when generating C++ code. If you want to allow undefined functions in the\n      Stan program, either add this to make/local or the command line:\n          STANCFLAGS = --allow_undefined\n    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the\n      header file that is included. This defaults to \"user_header.hpp\" in the\n      directory of the Stan program.\n    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads\n      the nightly version.\n    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.\n    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.\n\n\n  Example - bernoulli model: examples/bernoulli/bernoulli.stan\n\n    1. Build the model:\n       &gt; make examples/bernoulli/bernoulli\n    2. Run the sampling algorithm given the model and data:\n       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R\n    3. Look at the posterior sample:\n       &gt; bin/stansummary output.csv\n\n\n  Clean CmdStan:\n\n    Remove the built CmdStan tools:\n    &gt; make clean-all\n\n--------------------------------------------------------------------------------",
+    "crumbs": [
+      "Getting Started",
+      "CmdStan Installation"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/generate_quantities_config.html",
+    "href": "cmdstan-guide/generate_quantities_config.html",
+    "title": "Generating Quantities of Interest from a Fitted Model",
+    "section": "",
+    "text": "The generate_quantities method allows you to generate additional quantities of interest from a fitted model without re-running the sampler. Instead, you write a modified version of the original Stan program and add a generated quantities block or modify the existing one which specifies how to compute the new quantities of interest. Running the generate_quantities method on the new program together with sampler outputs (i.e., a set of draws) from the fitted model runs the generated quantities block of the new program using the the existing sample by plugging in the per-draw parameter estimates for the computations in the generated quantities block.\nThis method requires sub-argument fitted_params which takes as its value an existing Stan CSV file that contains a parameter values from an equivalent model, i.e., a model with the same parameters block, conditioned on the same data.\nThe generated quantities block computes quantities of interest (QOIs) based on the data, transformed data, parameters, and transformed parameters. It can be used to:\n\ngenerate simulated data for model testing by forward sampling\ngenerate predictions for new data\ncalculate posterior event probabilities, including multiple comparisons, sign tests, etc.\ncalculate posterior expectations\ntransform parameters for reporting\napply full Bayesian decision theory\ncalculate log likelihoods, deviances, etc. for model comparison\n\nFor an overview of the uses of this feature, see the Stan User’s Guide section on Stand-alone generated quantities and ongoing prediction.\n\n\nTo illustrate how this works we use the generate_quantities method to do posterior predictive checks using the estimate of theta given the example bernoulli model and data, following the posterior predictive simulation procedure in the Stan User’s Guide.\nWe write a program bernoulli_ppc.stan which contains the following generated quantities block, with comments to explain the procedure:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nThe rest of the program is the same as in bernoulli.stan.\nThe generate_method requires the sub-argument fitted_params which takes as its value the name of a Stan CSV file. The per-draw parameter values from the fitted_params file will be used to run the generated quantities block.\nIf we run the bernoulli.stan program for a single chain to generate a sample in file bernoulli_fit.csv:\n&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv\nThen we can run the bernoulli_ppc.stan to carry out the posterior predictive checks:\n&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \\\n                  data file=bernoulli.data.json \\\n                  output file=bernoulli_ppc.csv\nThe output file bernoulli_ppc.csv contains only the values for the variables declared in the generated quantities block, i.e., theta_rep and the elements of y_sim:\n# model = bernoulli_ppc_model\n# method = generate_quantities\n#   generate_quantities\n#     fitted_params = bernoulli_fit.csv\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2983956445 (Default)\n# output\n#   file = output.csv (Default)\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep\n1,1,1,0,0,0,1,1,0,1,0.6\n1,1,0,1,0,0,1,0,1,0,0.5\n1,0,1,1,1,1,1,1,0,1,0.8\n0,1,0,1,0,1,0,1,0,0,0.4\n1,0,0,0,0,0,0,0,0,0,0.1\n0,0,0,0,0,1,1,1,0,0,0.3\n0,0,1,0,1,0,0,0,0,0,0.2\n1,0,1,0,1,1,0,1,1,0,0.6\n...\nGiven the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  real&lt;lower=0, upper=1&gt; theta_cp = theta;\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nNow the output is slightly more interpretable: theta_cp is the same as the theta used to generate the values y_sim[1] through y_sim[1]. Comparing columns theta_cp and theta_rep allows us to see how the uncertainty in our estimate of theta is carried forward into our predictions:\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep\n0,1,1,0,1,0,0,1,1,0,0.545679,0.5\n1,1,1,1,1,1,0,1,1,0,0.527164,0.8\n1,1,1,1,0,1,1,1,1,0,0.529116,0.8\n1,0,1,1,1,1,0,0,1,0,0.478844,0.6\n0,1,0,0,0,0,1,0,1,0,0.238793,0.3\n0,0,0,0,0,1,1,0,0,0,0.258294,0.2\n1,1,1,0,0,0,0,0,0,0,0.258465,0.3\n\n\n\nThe fitted_params file must be a Stan CSV file; attempts to use a regular CSV file will result an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe fitted_params file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe parameter values of the fitted_params are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in bernoulli_fit.csv so that the value of theta is outside the declared bounds real&lt;lower=0, upper=1&gt;, the program will return the following error message:\nException: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \\\n(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)",
+    "crumbs": [
+      "Running CmdStan",
+      "Generating Quantities of Interest from a Fitted Model"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/generate_quantities_config.html#example",
+    "href": "cmdstan-guide/generate_quantities_config.html#example",
+    "title": "Generating Quantities of Interest from a Fitted Model",
+    "section": "",
+    "text": "To illustrate how this works we use the generate_quantities method to do posterior predictive checks using the estimate of theta given the example bernoulli model and data, following the posterior predictive simulation procedure in the Stan User’s Guide.\nWe write a program bernoulli_ppc.stan which contains the following generated quantities block, with comments to explain the procedure:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nThe rest of the program is the same as in bernoulli.stan.\nThe generate_method requires the sub-argument fitted_params which takes as its value the name of a Stan CSV file. The per-draw parameter values from the fitted_params file will be used to run the generated quantities block.\nIf we run the bernoulli.stan program for a single chain to generate a sample in file bernoulli_fit.csv:\n&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv\nThen we can run the bernoulli_ppc.stan to carry out the posterior predictive checks:\n&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \\\n                  data file=bernoulli.data.json \\\n                  output file=bernoulli_ppc.csv\nThe output file bernoulli_ppc.csv contains only the values for the variables declared in the generated quantities block, i.e., theta_rep and the elements of y_sim:\n# model = bernoulli_ppc_model\n# method = generate_quantities\n#   generate_quantities\n#     fitted_params = bernoulli_fit.csv\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2983956445 (Default)\n# output\n#   file = output.csv (Default)\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep\n1,1,1,0,0,0,1,1,0,1,0.6\n1,1,0,1,0,0,1,0,1,0,0.5\n1,0,1,1,1,1,1,1,0,1,0.8\n0,1,0,1,0,1,0,1,0,0,0.4\n1,0,0,0,0,0,0,0,0,0,0.1\n0,0,0,0,0,1,1,1,0,0,0.3\n0,0,1,0,1,0,0,0,0,0,0.2\n1,0,1,0,1,1,0,1,1,0,0.6\n...\nGiven the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  real&lt;lower=0, upper=1&gt; theta_cp = theta;\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nNow the output is slightly more interpretable: theta_cp is the same as the theta used to generate the values y_sim[1] through y_sim[1]. Comparing columns theta_cp and theta_rep allows us to see how the uncertainty in our estimate of theta is carried forward into our predictions:\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep\n0,1,1,0,1,0,0,1,1,0,0.545679,0.5\n1,1,1,1,1,1,0,1,1,0,0.527164,0.8\n1,1,1,1,0,1,1,1,1,0,0.529116,0.8\n1,0,1,1,1,1,0,0,1,0,0.478844,0.6\n0,1,0,0,0,0,1,0,1,0,0.238793,0.3\n0,0,0,0,0,1,1,0,0,0,0.258294,0.2\n1,1,1,0,0,0,0,0,0,0,0.258465,0.3",
+    "crumbs": [
+      "Running CmdStan",
+      "Generating Quantities of Interest from a Fitted Model"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/generate_quantities_config.html#errors",
+    "href": "cmdstan-guide/generate_quantities_config.html#errors",
+    "title": "Generating Quantities of Interest from a Fitted Model",
+    "section": "",
+    "text": "The fitted_params file must be a Stan CSV file; attempts to use a regular CSV file will result an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe fitted_params file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe parameter values of the fitted_params are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in bernoulli_fit.csv so that the value of theta is outside the declared bounds real&lt;lower=0, upper=1&gt;, the program will return the following error message:\nException: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \\\n(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)",
+    "crumbs": [
+      "Running CmdStan",
+      "Generating Quantities of Interest from a Fitted Model"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/example_model_data.html",
+    "href": "cmdstan-guide/example_model_data.html",
+    "title": "Example Model and Data",
+    "section": "",
+    "text": "The following is a simple, complete Stan program for a Bernoulli model of binary data.1 The model assumes the binary observed data y[1],...,y[N] are i.i.d. with Bernoulli chance-of-success theta.\ndata { \n  int&lt;lower=0&gt; N; \n  array[N] int&lt;lower=0, upper=1&gt; y;\n} \nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n} \nmodel {\n  theta ~ beta(1, 1);  // uniform prior on interval 0,1\n  y ~ bernoulli(theta);\n}\nThe input data file contains definitions for the two variables N and y which are specified in the data block of program bernoulli.stan (above).\nA data set of N=10 observations is included in the example Bernoulli model directory in both JSON notation and Rdump data format where 8 out of 10 trials had outcome 0 (failure) and 2 trials had outcome 1 (success). In JSON, this data is:\n{\n    \"N\" : 10,\n    \"y\" : [0,1,0,0,0,0,0,0,0,1]\n}",
+    "crumbs": [
+      "Getting Started",
+      "Example Model and Data"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/example_model_data.html#footnotes",
+    "href": "cmdstan-guide/example_model_data.html#footnotes",
+    "title": "Example Model and Data",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe model is available with the CmdStan distribution at the path &lt;cmdstan-home&gt;/examples/bernoulli/bernoulli.stan.↩︎",
+    "crumbs": [
+      "Getting Started",
+      "Example Model and Data"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/diagnose_config.html",
+    "href": "cmdstan-guide/diagnose_config.html",
+    "title": "Diagnosing HMC by Comparison of Gradients",
+    "section": "",
+    "text": "Diagnosing HMC by Comparison of Gradients\nCmdStan has a basic diagnostic feature that will calculate the gradients of the initial state and compare them with gradients calculated by finite differences. Discrepancies between the two indicate that there is a problem with the model or initial states or else there is a bug in Stan.\nTo allow for the possibility of adding other kinds of diagnostic tests, the diagnose method argument configuration has subargument test which currently only takes value gradient. There are two available gradient test configuration arguments:\n\nepsilon - The finite difference step size. Must be a positive real number. Default value is \\(1^{-6}\\)\nerror - The error threshold. Must be a positive real number. Default value is \\(1^{-6}\\)\n\nTo run on the different platforms with the default configuration, use one of the following.\nMac OS and Linux\n&gt; ./my_model diagnose data file=my_data\nWindows\n&gt; my_model diagnose data file=my_data\nTo relax the test threshold, specify the error argument as follows:\n&gt; ./my_model diagnose test=gradient error=0.0001 data file=my_data\nTo see how this works, we run diagnostics on the example bernoulli model:\n&gt; ./bernoulli diagnose data file=bernoulli.data.R\nExecuting this command prints output to the console and as a series of comment lines to the output csv file. The console output is:\nmethod = diagnose\n  diagnose\n    test = gradient (Default)\n      gradient\n        epsilon = 9.9999999999999995e-07 (Default)\n        error = 9.9999999999999995e-07 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2152196153 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n\nTEST GRADIENT MODE\n\n Log probability=-8.42814\n\n param idx           value           model     finite diff           error\n         0       0.0361376         -3.1084         -3.1084    -2.37554e-10\nThe same information is printed to the output file as csv comments, i.e., each line is prefixed with a pound sign #.\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Running CmdStan",
+      "Diagnosing HMC by Comparison of Gradients"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html",
+    "href": "cmdstan-guide/command_line_options.html",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "A CmdStan executable is built from the Stan model concept and the CmdStan command line parser. The command line argument syntax consists of sets of keywords and keyword-value pairs. Arguments are grouped by the following keywords:\n\nmethod - specifies the kind of inference done on the model. Each kind of inference requires further configuration via sub-arguments. The method argument is required. It can be specified overtly as the a keyword-value pair method=&lt;inference&gt; or implicitly as one of the following:\n\nsample - obtain a sample (set of draws) from the posterior using HMC\noptimize - penalized maximum likelihood estimation\npathfinder - quasi-Newton variational inference\nvariational - automatic differentation variational inference (ADVI)\nlaplace - sample from a normal approximation centered at the mode\ngenerate_quantities - run model’s generated quantities block on existing sample to obtain new quantities of interest.\nlog_prob - compute the log probability and gradient of the model for one set of parameters.\ndiagnose - compute and compare sampler gradient calculations to finite differences.\n\ndata - specifies the input data file, if any.\noutput - specifies program outputs, both disk files and terminal window outputs.\ninit - specifies initial values for the model parameters, if any.\nrandom - specifies the seed for the pseudo-random number.\n\nThe remainder of this chapter covers the general configuration options used for all processing. The following chapters cover the per-inference configuration options.\n\n\nThe values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:\ndata file=&lt;filepath&gt;\nThe keyword data must be followed directly by the keyword-value pair file=&lt;filepath&gt;. If the model doesn’t declare any data variables, this argument is ignored.\nThe input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model bernoulli.stan defines two data variables N and y. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.\nFor example if the input data file doesn’t include the definition for variable y, the executable exits with the following message:\nException: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)\n\n\n\nThe output keyword is used to specify non-default options for output files and messages written to the terminal window. The output keyword takes several keyword-value pair sub-arguments.\nThe keyword value pair file=&lt;filepath&gt; specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named output.csv in the current working directory.\nThe keyword value pair diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms sample and variational.\nThe keyword value pair refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.\nThe keyword value pair sig_figs=&lt;int&gt; specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8.   Note: increasing sig_figs above the default will increase the size of the output CSV files accordingly.\nThe keyword value pair profile_file=&lt;filepath&gt; specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and profile_file is unspecified, the profiling data is written to a file named profile.csv in the current working directory.\nThe keyword value pair save_cmdstan_config=&lt;boolean&gt; specifies whether to save the configuration options used to run the program to a file named &lt;output file&gt;_config.json alongside the other output files. The default value is false, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.\n\n\n\nInitialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range \\([-2, 2]\\). These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\nThe initialization argument is specified as keyword-value pair with keyword init. The value can be one of the following:\n\npositive real number \\(x\\). All parameters will be initialized to random draws from a uniform distribution over the range \\([-x, x]\\).\n\\(0\\) - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: \\(0\\) for unconstrained parameters; \\(1\\) for parameters constrained to be positive; \\(0.5\\) for variables to constrained to lie between \\(0\\) and \\(1\\); a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.\nfilepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range \\([-2, 2]\\).\n\n\n\n\nThe random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nThe syntax for the random seed argument is:\nrandom seed=&lt;int&gt;\nThe keyword random must be followed directly by the keyword-value pair seed=&lt;int&gt;.\n\n\n\nThe chain identifier argument is used in conjunction with the random seed argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.\nThe syntax for the random seed argument is:\nid=&lt;int&gt;\nThe default value is 1.\nWhen running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.\nFor complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the Stan Reference Manual Reproducibility chapter for further details.\n\n\n\nCmdStan provides a help and help-all mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword help:\n./bernoulli help\n\n\n\nCmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.\nAll program executables provide a return code between 0 and 255:\n\n0 - Program ran to termination as expected.\nvalue in range [1 : 125] - Method invoked could not run due to problems with model or data.\nvalue &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g. return code 139 results from termination signal 11 (segmentation violation).\n\nA non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#input-data-argument",
+    "href": "cmdstan-guide/command_line_options.html#input-data-argument",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "The values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:\ndata file=&lt;filepath&gt;\nThe keyword data must be followed directly by the keyword-value pair file=&lt;filepath&gt;. If the model doesn’t declare any data variables, this argument is ignored.\nThe input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model bernoulli.stan defines two data variables N and y. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.\nFor example if the input data file doesn’t include the definition for variable y, the executable exits with the following message:\nException: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#output-control.section",
+    "href": "cmdstan-guide/command_line_options.html#output-control.section",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "The output keyword is used to specify non-default options for output files and messages written to the terminal window. The output keyword takes several keyword-value pair sub-arguments.\nThe keyword value pair file=&lt;filepath&gt; specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named output.csv in the current working directory.\nThe keyword value pair diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms sample and variational.\nThe keyword value pair refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.\nThe keyword value pair sig_figs=&lt;int&gt; specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8.   Note: increasing sig_figs above the default will increase the size of the output CSV files accordingly.\nThe keyword value pair profile_file=&lt;filepath&gt; specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and profile_file is unspecified, the profiling data is written to a file named profile.csv in the current working directory.\nThe keyword value pair save_cmdstan_config=&lt;boolean&gt; specifies whether to save the configuration options used to run the program to a file named &lt;output file&gt;_config.json alongside the other output files. The default value is false, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#initialize-model-parameters-argument",
+    "href": "cmdstan-guide/command_line_options.html#initialize-model-parameters-argument",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "Initialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range \\([-2, 2]\\). These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\nThe initialization argument is specified as keyword-value pair with keyword init. The value can be one of the following:\n\npositive real number \\(x\\). All parameters will be initialized to random draws from a uniform distribution over the range \\([-x, x]\\).\n\\(0\\) - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: \\(0\\) for unconstrained parameters; \\(1\\) for parameters constrained to be positive; \\(0.5\\) for variables to constrained to lie between \\(0\\) and \\(1\\); a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.\nfilepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range \\([-2, 2]\\).",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#random-number-generator-arguments",
+    "href": "cmdstan-guide/command_line_options.html#random-number-generator-arguments",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "The random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nThe syntax for the random seed argument is:\nrandom seed=&lt;int&gt;\nThe keyword random must be followed directly by the keyword-value pair seed=&lt;int&gt;.",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#chain-identifier-argument-id",
+    "href": "cmdstan-guide/command_line_options.html#chain-identifier-argument-id",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "The chain identifier argument is used in conjunction with the random seed argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.\nThe syntax for the random seed argument is:\nid=&lt;int&gt;\nThe default value is 1.\nWhen running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.\nFor complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the Stan Reference Manual Reproducibility chapter for further details.",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#command-line-help",
+    "href": "cmdstan-guide/command_line_options.html#command-line-help",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "CmdStan provides a help and help-all mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword help:\n./bernoulli help",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/command_line_options.html#error-messages-and-return-codes",
+    "href": "cmdstan-guide/command_line_options.html#error-messages-and-return-codes",
+    "title": "Command-Line Interface Overview",
+    "section": "",
+    "text": "CmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.\nAll program executables provide a return code between 0 and 255:\n\n0 - Program ran to termination as expected.\nvalue in range [1 : 125] - Method invoked could not run due to problems with model or data.\nvalue &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g. return code 139 results from termination signal 11 (segmentation violation).\n\nA non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.",
+    "crumbs": [
+      "Running CmdStan",
+      "Command-Line Interface Overview"
+    ]
+  },
+  {
+    "objectID": "index.html",
+    "href": "index.html",
+    "title": "Stan Documentation",
+    "section": "",
+    "text": "This is the official documentation for Stan.\n\nThe Stan User’s Guide (pdf) provides example models and programming techniques for coding statistical models in Stan.\nThe Stan Reference Manual (pdf) specifies the Stan programming language and inference algorithms.\nThe Stan Functions Reference (pdf) specifies the functions built into the Stan programming language.\n\nThere are also separate installation and getting started guides for CmdStan (pdf), the command-line interface to the Stan inference engine, and the R, Python, and Julia interfaces.\n\nOlder Versions\nThis documentation is for Stan 2.39. Older versions of each of the documents linked above can be found in the table below:\n\n\n\nVersion\nStan Reference Manual\nStan Users Guide\nCmdStan Guide\nStan Functions Reference\n\n\n\n\n2.39\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.38\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.37\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.36\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.35\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.34\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.33\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.32\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.31\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.30\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.29\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.28\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.27\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.26\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.25\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.24\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.23\nhtml pdf\nhtml pdf\nhtml\nhtml pdf\n\n\n2.22\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.21\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.20\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.19\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.18\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n\nPrior to version 2.18, all documentation was part of a single document called the Stan User’s Guide and Reference Manual. These versions are still available for download as PDFs:\n\n\n\nVersion\nStan User’s Guide and Reference Manual\n\n\n\n\n2.17\npdf\n\n\n2.16\npdf\n\n\n2.15\npdf\n\n\n2.14\npdf\n\n\n2.13\npdf\n\n\n2.12\npdf\n\n\n2.11\npdf\n\n\n2.10\npdf\n\n\n2.9\npdf\n\n\n2.8\npdf\n\n\n2.7\npdf\n\n\n2.6\npdf\n\n\n2.5\npdf\n\n\n2.4\npdf\n\n\n2.3\npdf\n\n\n2.2\npdf\n\n\n2.1\npdf\n\n\n2.0\npdf\n\n\n1.3\npdf\n\n\n1.2\npdf\n\n\n1.1\npdf\n\n\n1.0\npdf\n\n\n\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "cmdstan-guide/bib.html",
+    "href": "cmdstan-guide/bib.html",
+    "title": "References",
+    "section": "",
+    "text": "References\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "cmdstan-guide/compiling_stan_programs.html",
+    "href": "cmdstan-guide/compiling_stan_programs.html",
+    "title": "Compiling a Stan Program",
+    "section": "",
+    "text": "A Stan program must be in a file with extension .stan. The CmdStan makefile rules specify all necessary steps to translate files with suffix .stan to a CmdStan executable program. This is a two-stage process:\n\nfirst the Stan program is translated to C++ by the stanc compiler\nthen the C++ compiler compiles all C++ sources and links them together with the CmdStan interface program and the Stan and math libraries.\n\n\n\nTo compile Stan programs, you must invoke the Make program from the &lt;cmdstan-home&gt; directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.\n&gt; cd &lt;cmdstan_home&gt;\nIn the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:\n&gt; make examples/bernoulli/bernoulli\nOn Windows, the command is the same with the addition of .exe at the end of the target (note the use of forward slashes):\n&gt; make examples/bernoulli/bernoulli.exe\nThe generated C++ code (bernoulli.hpp), object file (bernoulli.o) and the compiled executable will be placed in the same directory as the Stan program.\nThe compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.\n\n\n\nWhen executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target bernoulli twice in a row, without any editing bernoulli.stan or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:\n&gt; make examples/bernoulli/bernoulli\nmake: `examples/bernoulli/bernoulli' is up to date.\nIf the file containing the Stan program is updated, the next call to make will rebuild the CmdStan executable.\n\n\n\nThe Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.\nThe simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the bernoulli.stan program, we introduce a typo on line \\(9\\) by writing thata instead of theta, the Make command fails with the following\n--- Translating Stan model to C++ code ---\nbin/stanc  --o=bernoulli.hpp bernoulli.stan\n\nSemantic error in 'bernoulli.stan', line 9, column 2 to column 7:\n   -------------------------------------------------\n     7:  }\n     8:  model {\n     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1\n           ^\n    10:    y ~ bernoulli(theta);\n    11:  }\n   -------------------------------------------------\n\nIdentifier 'thata' not in scope.\n\nmake: *** [bernoulli.hpp] Error 1\nStan is a strongly-typed language; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program foo.stan contains an illegal assignment statement:\ndata {\n  real x;\n}\ntransformed data {\n  int y = x;\n}\nThe Make command fails with the following:\nSemantic error in 'foo.stan', line 5, column 2 to column 12:\n   -------------------------------------------------\n     3:  }\n     4:  transformed data {\n     5:    int y = x;\n           ^\n     6:  }\n   -------------------------------------------------\n\nIll-typed arguments supplied to assignment operator =:\n lhs has type int and rhs has type real\nThe Stan Reference Manual provides a complete specification of the Stan programming language. The Stan User’s Guide also contains a full description of the errors and warnings stanc can emit.\n\n\n\nIf the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.\nThe Make command prints the following message to the terminal at the point when it compiles and links the C++ file:\n--- Compiling, linking C++ code ---\nIf the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.\nIf this happens, please report the error, together with the Stan program on either the Stan Forums or on the Stan compiler GitHub issues tracker.\n\n\n\nUsers can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.\nThe CXXFLAGS and LDFLAGS makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the make/local file.\nFor example:\nCXXFLAGS = -O2\nA recommend a set of CXXFLAGS and LDFLAGS flags can be turned on by setting STAN_CPP_OPTIMS=true in the make/local file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.\n\n\nWhen assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.\nFor some models these checks can represent a significant part of the models execution time. By setting the STAN_NO_RANGE_CHECKS=true makefile flag in the make/local file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.",
+    "crumbs": [
+      "Getting Started",
+      "Compiling a Stan Program"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#invoking-the-make-utility",
+    "href": "cmdstan-guide/compiling_stan_programs.html#invoking-the-make-utility",
+    "title": "Compiling a Stan Program",
+    "section": "",
+    "text": "To compile Stan programs, you must invoke the Make program from the &lt;cmdstan-home&gt; directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.\n&gt; cd &lt;cmdstan_home&gt;\nIn the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:\n&gt; make examples/bernoulli/bernoulli\nOn Windows, the command is the same with the addition of .exe at the end of the target (note the use of forward slashes):\n&gt; make examples/bernoulli/bernoulli.exe\nThe generated C++ code (bernoulli.hpp), object file (bernoulli.o) and the compiled executable will be placed in the same directory as the Stan program.\nThe compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.",
+    "crumbs": [
+      "Getting Started",
+      "Compiling a Stan Program"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#dependencies",
+    "href": "cmdstan-guide/compiling_stan_programs.html#dependencies",
+    "title": "Compiling a Stan Program",
+    "section": "",
+    "text": "When executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target bernoulli twice in a row, without any editing bernoulli.stan or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:\n&gt; make examples/bernoulli/bernoulli\nmake: `examples/bernoulli/bernoulli' is up to date.\nIf the file containing the Stan program is updated, the next call to make will rebuild the CmdStan executable.",
+    "crumbs": [
+      "Getting Started",
+      "Compiling a Stan Program"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#compiler-errors",
+    "href": "cmdstan-guide/compiling_stan_programs.html#compiler-errors",
+    "title": "Compiling a Stan Program",
+    "section": "",
+    "text": "The Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.\nThe simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the bernoulli.stan program, we introduce a typo on line \\(9\\) by writing thata instead of theta, the Make command fails with the following\n--- Translating Stan model to C++ code ---\nbin/stanc  --o=bernoulli.hpp bernoulli.stan\n\nSemantic error in 'bernoulli.stan', line 9, column 2 to column 7:\n   -------------------------------------------------\n     7:  }\n     8:  model {\n     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1\n           ^\n    10:    y ~ bernoulli(theta);\n    11:  }\n   -------------------------------------------------\n\nIdentifier 'thata' not in scope.\n\nmake: *** [bernoulli.hpp] Error 1\nStan is a strongly-typed language; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program foo.stan contains an illegal assignment statement:\ndata {\n  real x;\n}\ntransformed data {\n  int y = x;\n}\nThe Make command fails with the following:\nSemantic error in 'foo.stan', line 5, column 2 to column 12:\n   -------------------------------------------------\n     3:  }\n     4:  transformed data {\n     5:    int y = x;\n           ^\n     6:  }\n   -------------------------------------------------\n\nIll-typed arguments supplied to assignment operator =:\n lhs has type int and rhs has type real\nThe Stan Reference Manual provides a complete specification of the Stan programming language. The Stan User’s Guide also contains a full description of the errors and warnings stanc can emit.",
+    "crumbs": [
+      "Getting Started",
+      "Compiling a Stan Program"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#troubleshooting-c-compiler-or-linker-errors",
+    "href": "cmdstan-guide/compiling_stan_programs.html#troubleshooting-c-compiler-or-linker-errors",
+    "title": "Compiling a Stan Program",
+    "section": "",
+    "text": "If the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.\nThe Make command prints the following message to the terminal at the point when it compiles and links the C++ file:\n--- Compiling, linking C++ code ---\nIf the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.\nIf this happens, please report the error, together with the Stan program on either the Stan Forums or on the Stan compiler GitHub issues tracker.",
+    "crumbs": [
+      "Getting Started",
+      "Compiling a Stan Program"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#c-compilation-and-linking-flags",
+    "href": "cmdstan-guide/compiling_stan_programs.html#c-compilation-and-linking-flags",
+    "title": "Compiling a Stan Program",
+    "section": "",
+    "text": "Users can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.\nThe CXXFLAGS and LDFLAGS makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the make/local file.\nFor example:\nCXXFLAGS = -O2\nA recommend a set of CXXFLAGS and LDFLAGS flags can be turned on by setting STAN_CPP_OPTIMS=true in the make/local file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.\n\n\nWhen assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.\nFor some models these checks can represent a significant part of the models execution time. By setting the STAN_NO_RANGE_CHECKS=true makefile flag in the make/local file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.",
+    "crumbs": [
+      "Getting Started",
+      "Compiling a Stan Program"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/diagnose_utility.html",
+    "href": "cmdstan-guide/diagnose_utility.html",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "section": "",
+    "text": "CmdStan is distributed with a utility that is able to read in and analyze the output of one or more Markov chains to check for the following potential problems:\n\nDivergent transitions\nTransitions that hit the maximum treedepth\nLow E-BFMI values\nLow effective sample sizes\nHigh \\(\\hat{R}\\) values\n\nThe meanings of several of these problems are discussed in https://arxiv.org/abs/1701.02434.\n\n\nThe CmdStan makefile task build compiles the diagnose utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/diagnose\n\n\n\nThe diagnose command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to diagnose, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.\nTo fully exercise the diagnose command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the Stan User’s Guide reparameterization section. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y / 2));\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan\nStan has trouble sampling from the region where y is small and thus x is constrained to be near 0. This is due to the fact that the density’s scale changes with y, so that a step size that works well when y is large is inefficient when y is small and vice-versa.\nRunning 4 chains produces output files output_1.csv, …, output_4.csv. We run diagnose command on this fileset:\n&gt; bin/diagnose output_*.csv\nThe output is printed to the terminal window:\nChecking sampler transitions treedepth.\n18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.\nTrajectories that are prematurely terminated due to this limit will result in slow exploration.\nFor optimal performance, increase this limit.\n\nChecking sampler transitions for divergences.\n11 of 4000 (0.28%) transitions ended with a divergence.\nThese divergent transitions indicate that HMC is not fully able to explore the posterior distribution.\nTry increasing adapt delta closer to 1.\nIf this doesn't remove all divergences, try to reparameterize the model.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nThe E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.\nIf possible, try to reparameterize the model.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nThe following parameters had rank-normalized split R-hat greater than 1.01:\n  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]\nSuch high values indicate incomplete mixing and biased estimation.\nYou should consider regularizing your model with additional prior information or a more effective parameterization.\n\nProcessing complete.\nIn this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y / 2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y / 2))\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan\nWe compile the program and run 4 chains, as before. Now the diagnose command doesn’t detect any problems:\nChecking sampler transitions treedepth.\nTreedepth satisfactory for all transitions.\n\nChecking sampler transitions for divergences.\nNo divergent transitions found.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nE-BFMI satisfactory.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nRank-normalized split R-hat values satisfactory for all parameters.\n\nProcessing complete, no problems detected.\n\n\n\n\n\nStan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a Hamiltonian system. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the step size controls the resolution of the sampler.\nUnfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as divergences which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.\nIf the divergent transitions cannot be eliminated by increasing the adapt_delta parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the reparameterization section of the Stan User’s Guide.\n\n\n\nWarnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a validity concern, hitting the maximum treedepth is an efficiency concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the Hamiltonian Monte Carlo Sampling chapter in the Stan Reference Manual). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.\nThis is controlled through the max_depth argument. If the number of transitions which exceed maximum treedepth is low, increasing max_depth may correct this problem.\n\n\n\nThe sampler csv output column energy__ is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of energy is much larger than \\(\\sqrt{D / 2}\\), where \\(D\\) is the number of unconstrained parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.\nThe warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.\n\n\n\nRoughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses \\(\\hat{R}\\) adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.\nBulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g. mean and median).\nOften quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.\n\n\n\n\\(\\hat{R}\\) (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), \\(\\hat{R}\\) is larger than 1. We recommend running at least four chains by default and only using the sample if \\(\\hat{R}\\) is less than 1.01. Stan reports \\(\\hat{R}\\) which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.\nThere is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.",
+    "crumbs": [
+      "Tools and Utilities",
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/diagnose_utility.html#building-the-diagnose-command",
+    "href": "cmdstan-guide/diagnose_utility.html#building-the-diagnose-command",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "section": "",
+    "text": "The CmdStan makefile task build compiles the diagnose utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/diagnose",
+    "crumbs": [
+      "Tools and Utilities",
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/diagnose_utility.html#running-the-diagnose-command",
+    "href": "cmdstan-guide/diagnose_utility.html#running-the-diagnose-command",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "section": "",
+    "text": "The diagnose command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to diagnose, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.\nTo fully exercise the diagnose command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the Stan User’s Guide reparameterization section. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y / 2));\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan\nStan has trouble sampling from the region where y is small and thus x is constrained to be near 0. This is due to the fact that the density’s scale changes with y, so that a step size that works well when y is large is inefficient when y is small and vice-versa.\nRunning 4 chains produces output files output_1.csv, …, output_4.csv. We run diagnose command on this fileset:\n&gt; bin/diagnose output_*.csv\nThe output is printed to the terminal window:\nChecking sampler transitions treedepth.\n18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.\nTrajectories that are prematurely terminated due to this limit will result in slow exploration.\nFor optimal performance, increase this limit.\n\nChecking sampler transitions for divergences.\n11 of 4000 (0.28%) transitions ended with a divergence.\nThese divergent transitions indicate that HMC is not fully able to explore the posterior distribution.\nTry increasing adapt delta closer to 1.\nIf this doesn't remove all divergences, try to reparameterize the model.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nThe E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.\nIf possible, try to reparameterize the model.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nThe following parameters had rank-normalized split R-hat greater than 1.01:\n  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]\nSuch high values indicate incomplete mixing and biased estimation.\nYou should consider regularizing your model with additional prior information or a more effective parameterization.\n\nProcessing complete.\nIn this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y / 2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y / 2))\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan\nWe compile the program and run 4 chains, as before. Now the diagnose command doesn’t detect any problems:\nChecking sampler transitions treedepth.\nTreedepth satisfactory for all transitions.\n\nChecking sampler transitions for divergences.\nNo divergent transitions found.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nE-BFMI satisfactory.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nRank-normalized split R-hat values satisfactory for all parameters.\n\nProcessing complete, no problems detected.",
+    "crumbs": [
+      "Tools and Utilities",
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/diagnose_utility.html#diagnose-warnings-and-recommendations",
+    "href": "cmdstan-guide/diagnose_utility.html#diagnose-warnings-and-recommendations",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "section": "",
+    "text": "Stan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a Hamiltonian system. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the step size controls the resolution of the sampler.\nUnfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as divergences which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.\nIf the divergent transitions cannot be eliminated by increasing the adapt_delta parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the reparameterization section of the Stan User’s Guide.\n\n\n\nWarnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a validity concern, hitting the maximum treedepth is an efficiency concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the Hamiltonian Monte Carlo Sampling chapter in the Stan Reference Manual). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.\nThis is controlled through the max_depth argument. If the number of transitions which exceed maximum treedepth is low, increasing max_depth may correct this problem.\n\n\n\nThe sampler csv output column energy__ is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of energy is much larger than \\(\\sqrt{D / 2}\\), where \\(D\\) is the number of unconstrained parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.\nThe warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.\n\n\n\nRoughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses \\(\\hat{R}\\) adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.\nBulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g. mean and median).\nOften quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.\n\n\n\n\\(\\hat{R}\\) (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), \\(\\hat{R}\\) is larger than 1. We recommend running at least four chains by default and only using the sample if \\(\\hat{R}\\) is less than 1.01. Stan reports \\(\\hat{R}\\) which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.\nThere is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.",
+    "crumbs": [
+      "Tools and Utilities",
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/external_code.html",
+    "href": "cmdstan-guide/external_code.html",
+    "title": "Using external C++ code",
+    "section": "",
+    "text": "The --allow-undefined flag can be passed to the call to stanc, which will allow undefined functions in the Stan language to be parsed without an error. We can then include a definition of the function in a C++ header file.\nThis requires specifying two makefile variables:\n\nSTANCFLAGS=--allow-undefined\nUSER_HEADER=&lt;header_file.hpp&gt;, where &lt;header_file.hpp&gt; is the name of a header file that defines a function with the same name and a compatible signature. This function can appear in the global namespace or in the model namespace, which is defined as the name of the model (either the file name, or the --name argument to stanc) followed by _namespace.\n\nThis is an advanced feature which is only recommended to users familiar with the internals of Stan’s Math library. Most existing C++ code will need to be modified to work with Stan, to varying degrees.\nAs an example, consider the following variant of the Bernoulli example\nfunctions {\n  real make_odds(data real theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(1, 1); // uniform prior on interval 0, 1\n  y ~ bernoulli(theta);\n}\ngenerated quantities {\n  real odds;\n  odds = make_odds(theta);\n}\nHere the make_odds function is declared but not defined, which would ordinarily result in a parser error. However, if you put STANCFLAGS = --allow-undefined into the make/local file or into the stanc call, then the stanc compiler will translate this program to C++, but the generated C++ code will not compile unless you write a file such as examples/bernoulli/make_odds.hpp with the following lines\n#include &lt;ostream&gt;\n\ndouble make_odds(const double& theta, std::ostream *pstream__) {\n  return theta / (1 - theta);\n}\nThe signature for this function needs to fulfill all the usages in the C++ class emitted by stanc. The pstream__ argument is mandatory in the signature but need not be used if your function does not print any output. Because make_odds was declared with a data argument and only used in generated quantites, a signature which accepts and returns double is acceptable. Functions which will have parameters passed as input in the transformed parameters or model blocks will require the ability to accept Stan’s autodiff types. If you wish to autodiff through this function, the simplest option is to make it a template, like\ntemplate &lt;typename T&gt;\nT make_odds(const T &theta, std::ostream *pstream__)\n{\n    return theta / (1 - theta);\n}\nGiven the above, the following make invocation should work\n&gt; make STANCFLAGS=--allow-undefined USER_HEADER=examples/bernoulli/make_odds.hpp examples/bernoulli/bernoulli # on Windows add .exe\nAlternatively, you could put STANCFLAGS and USER_HEADER into the make/local file instead of specifying them on the command-line.\nIf the function were more complicated and involved functions in the Stan Math Library, then you would need to add #include &lt;stan/model/model_header.hpp&gt; and prefix the function calls with stan::math::.\n\n\nExternal C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the Stan Math documentation.\nSuppose you have the following (nonsensical) model which relies on a function called my_dot_self. We will implement this as a copy of the built-in dot_self function.\nfunctions {\n  // both overloads end up using the same C++ template\n  real my_dot_self(vector theta);\n  real my_dot_self(row_vector theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] input_data;\n}\ntransformed data {\n  // no autodiff for data - will call using doubles\n  real ds = my_dot_self(input_data);\n}\nparameters {\n  row_vector[N] thetas;\n}\nmodel {\n  thetas ~ normal(0,1);\n  // autodiff - will call using stan::math::var types\n  input_data ~ normal(thetas, my_dot_self(thetas));\n}\nIf you wanted to autodiff through this function, the following header would suffice1:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr&gt;\ninline stan::value_type_t&lt;EigVec&gt; my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    const auto &x_ref = stan::math::to_ref(x);\n    stan::value_type_t&lt;EigVec&gt; sum_x = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum_x += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum_x;\n}\nHowever, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr,\n          stan::require_not_st_var&lt;EigVec&gt; * = nullptr&gt;\ninline double my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    auto x_ref = stan::math::to_ref(x);\n    double sum = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum;\n}\n\ntemplate &lt;typename EigVec, stan::require_eigen_vt&lt;stan::is_var, EigVec&gt; * = nullptr&gt;\ninline stan::math::var my_dot_self(const EigVec &v, std::ostream *pstream__)\n{\n    // (1) put v into our memory arena\n    stan::arena_t&lt;EigVec&gt; arena_v(v);\n    // (2) calculate forward pass using\n    // (3) the .val() method for matrices of var types\n    stan::math::var res = my_dot_self(arena_v.val(), pstream__);\n    // (4) Place a callback for the reverse pass on the callback stack.\n    stan::math::reverse_pass_callback(\n        [res, arena_v]() mutable\n        { arena_v.adj() += 2.0 * res.adj() * arena_v.val(); });\n    return res;\n}\nFor more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.\n\n\n\nSome functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.\n\nRNGs must end with _rng. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a stan::rng_t object (a type alias to boost::rng::mixmax).\nFunctions which edit the target directly must end with _lp and will be passed a reference to lp__ and a reference to a stan::math::accumulator object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter propto__ which controls whether or not constant terms can be dropped.\nProbability distributions must end with _lpdf or _lpmf and will be passed a boolean template parameter propto__ which controls whether or not constant terms can be dropped.",
+    "crumbs": [
+      "Appendices",
+      "Using external C++ code"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/external_code.html#derivative-specializations",
+    "href": "cmdstan-guide/external_code.html#derivative-specializations",
+    "title": "Using external C++ code",
+    "section": "",
+    "text": "External C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the Stan Math documentation.\nSuppose you have the following (nonsensical) model which relies on a function called my_dot_self. We will implement this as a copy of the built-in dot_self function.\nfunctions {\n  // both overloads end up using the same C++ template\n  real my_dot_self(vector theta);\n  real my_dot_self(row_vector theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] input_data;\n}\ntransformed data {\n  // no autodiff for data - will call using doubles\n  real ds = my_dot_self(input_data);\n}\nparameters {\n  row_vector[N] thetas;\n}\nmodel {\n  thetas ~ normal(0,1);\n  // autodiff - will call using stan::math::var types\n  input_data ~ normal(thetas, my_dot_self(thetas));\n}\nIf you wanted to autodiff through this function, the following header would suffice1:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr&gt;\ninline stan::value_type_t&lt;EigVec&gt; my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    const auto &x_ref = stan::math::to_ref(x);\n    stan::value_type_t&lt;EigVec&gt; sum_x = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum_x += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum_x;\n}\nHowever, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr,\n          stan::require_not_st_var&lt;EigVec&gt; * = nullptr&gt;\ninline double my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    auto x_ref = stan::math::to_ref(x);\n    double sum = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum;\n}\n\ntemplate &lt;typename EigVec, stan::require_eigen_vt&lt;stan::is_var, EigVec&gt; * = nullptr&gt;\ninline stan::math::var my_dot_self(const EigVec &v, std::ostream *pstream__)\n{\n    // (1) put v into our memory arena\n    stan::arena_t&lt;EigVec&gt; arena_v(v);\n    // (2) calculate forward pass using\n    // (3) the .val() method for matrices of var types\n    stan::math::var res = my_dot_self(arena_v.val(), pstream__);\n    // (4) Place a callback for the reverse pass on the callback stack.\n    stan::math::reverse_pass_callback(\n        [res, arena_v]() mutable\n        { arena_v.adj() += 2.0 * res.adj() * arena_v.val(); });\n    return res;\n}\nFor more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.",
+    "crumbs": [
+      "Appendices",
+      "Using external C++ code"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/external_code.html#special-functions-rngs-distributions-editing-target",
+    "href": "cmdstan-guide/external_code.html#special-functions-rngs-distributions-editing-target",
+    "title": "Using external C++ code",
+    "section": "",
+    "text": "Some functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.\n\nRNGs must end with _rng. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a stan::rng_t object (a type alias to boost::rng::mixmax).\nFunctions which edit the target directly must end with _lp and will be passed a reference to lp__ and a reference to a stan::math::accumulator object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter propto__ which controls whether or not constant terms can be dropped.\nProbability distributions must end with _lpdf or _lpmf and will be passed a boolean template parameter propto__ which controls whether or not constant terms can be dropped.",
+    "crumbs": [
+      "Appendices",
+      "Using external C++ code"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/external_code.html#footnotes",
+    "href": "cmdstan-guide/external_code.html#footnotes",
+    "title": "Using external C++ code",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nDetails of programming in the Stan Math style are omitted from this section, it is presented only as an example↩︎",
+    "crumbs": [
+      "Appendices",
+      "Using external C++ code"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/index.html",
+    "href": "cmdstan-guide/index.html",
+    "title": "CmdStan User’s Guide",
+    "section": "",
+    "text": "This document is a user’s guide for CmdStan, the command-line interface to the Stan statistical modeling language. CmdStan provides the programs and tools to compile Stan programs into C++ executables that can be run directly from the command line, together with a few utilities to check and summarize the resulting outputs.\nIn CmdStan, statistical models written in the Stan probabilistic programming language are translated into a C++ program which is then compiled together with the CmdStan routines that provide the logic needed to manage all user inputs and program outputs and the Stan inference algorithms and math library. The resulting command line executable program can be used to\n\ndo inference on data, producing an exact or approximate estimate of the posterior;\ngenerate new quantities of interest from an existing estimate;\ngenerate data from the model according to a given set of parameters.\n\nThe packages CmdStanR and CmdStanPy provide interfaces to CmdStan from R and Python, respectively, similarly, JuliaStan also interfaces with CmdStan.\nDownload the pdf version of this manual.\n\nBenefits of CmdStan\n\nWith every new Stan release, there is a corresponding CmdStan release, therefore CmdStan provides access to the latest version of Stan, and can be used to run the development version of Stan as well.\nOf the Stan interfaces, CmdStan has the lightest memory footprint, therefore it can fit larger and more complex models. It has has the fewest dependencies, which makes it easier to run in limited environments such as clusters.\nThe output generated is in CSV format and can be post-processed using other Stan interfaces or general tools.\n\n\n\nStan documentation\n\nStan User’s Guide The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:\nStan Reference Manual Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.\nStan Functions Reference The Stan Functions Reference provides definitions and examples for all the functions defined in the Stan math library and available in the Stan programming language, including all probability distributions.\n\n\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "crumbs": [
+      "CmdStan User's Guide"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/json_apdx.html",
+    "href": "cmdstan-guide/json_apdx.html",
+    "title": "JSON Format for CmdStan",
+    "section": "",
+    "text": "CmdStan can use JSON format for input data for both model data and parameters. Model data is read in by the model constructor. Model parameters are used to initialize the sampler and optimizer.\n\n\nYou can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the write_stan_json() function provided by the CmdStanR interface.\n\n\n\nJSON is a data interchange notation, defined by an EMCA standard. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:\n\nStructural tokens are the left and right curly bracket {}, left and right square bracket [], the semicolon ;, and the comma ,.\nLiteral tokens must always be in lowercase. There are three literal tokens: true, false, null.\nA primitive value is a single token which is either a literal, a string, or a number.\nA string consists of zero or more Unicode characters enclosed in double quotes, e.g. \"foo\". A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g. \"\\\\uHHHH\" where HHHH is the Unicode code point in hex.\nNumbers are represented using either decimal notation or scientific notation. The following are examples of numbers: 17, 17.2, -17.2, -17.2e8, 17.2e-8.  There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.\nThe special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string \"Inf\", the string \"Infinity\", or the atom Infinity. Negative infinity can be represented as the string \"-Inf\", the string \"-Infinity\", or the atom -Infinity. Not-a-number can be represented as the string \"NaN\" or the atom NaN. These values may be mixed with other numerical types.\nA complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number \\(2.3 - 1.83i\\) would be represented in JSON as the two-element array [2.3, -1.83].\nA JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: [], [1], [0.2, \"-inf\", true].\nVectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector \\([1 \\quad 2]^{\\top}\\) and the row vector \\([1 \\quad 2]\\) are represented by the JSON array [1, 2].\nComplex vectors are represented as arrays of two-element arrays. For example, the complex vector \\([2.3 - 1.83i \\quad -4.8 +\n2i]^{\\top}\\) is represented as [[2.3, -1.83], [-4.8, 2]] in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).\nMatrices are represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 & 2.7 & -9.8 \\\\\n4.2 & 1.8 & -7.3\n\\end{bmatrix}\n\\end{equation*}\\] is represented in JSON as [[1, 2.7, -9.8],   [4.2, 1.8, -7.3]].\nComplex matrices are also represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) complex matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 + 2i & 3 - 4.2i & 13.1 + 2.7i \\\\\n3.1 & -5i & 0\n\\end{bmatrix}\n\\end{equation*}\\] would be represented in JSON as [[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]].\nTuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple (1.5, 3.4) is represented in JSON as {\"1\": 1.5, \"2\": 3.4}.\nA name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.\nA JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: { }, {\"foo\": null}, {\"bar\" : 17, \"baz\" : [14,15,16.6] }.\n\n\n\n\nStan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:\n{ \"N\" : 10, \"y\" : [0,1,0,0,0,0,0,0,0,1] }\nMatrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:\ndata {\n  int d1;\n  int d2;\n  int d3;\n  array[d1, d2, d3] int ar;\n}\nthe following JSON input would be valid:\n{ \"d1\" : 2,\n  \"d2\" : 3,\n  \"d3\" : 4,\n  \"ar\" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],\n          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]\n}\nJSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.\nAll data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.\n\n\n\n\n\n\n\nStan declaration\nJSON encoding\n\n\n\n\nint i\n\"i\": 17\n\n\n\n\n\n\nreal a\n\"a\" : 17\n\n\n\n\"a\" : 17.2\n\n\n\n\"a\" : \"NaN\"\n\n\n\n\"a\" : \"+inf\"\n\n\n\n\"a\" : \"-inf\"\n\n\n\n\n\n\ncomplex z\n\"z\": [1, -2.3]\n\n\narray[5] int\n\"a\" : [1, 2, 3, 4, 5]\n\n\narray[5] real a\n\"a\" : [ 1, 2, 3.3, \"NaN\", 5 ]\n\n\narray[2] complex b\n\"b\" : [[1, -2.3], [4.9, 0]]\n\n\nvector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nrow_vector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nmatrix[2, 3] a\n\"a\" : [[ 1, 2, 3 ], [ 4, 5, 6]]\n\n\ncomplex_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_row_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_matrix[2, 3] d\n\"d\" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]\n\n\ntuple(real, array[2] int) t\n\"t\" : { \"1\": 1.4, \"2\": [1, 2]}\n\n\n\n\n\nJSON notation is not able to distinguish between multi-dimensional arrays where any dimension is \\(0\\), e.g., a 2-D array with dimensions \\((1,0)\\), i.e., an array which contains a single array which is empty, has JSON representation [ ]. To see how this works, consider the following Stan program data block:\ndata {\n  int d;\n  array[d] int ar_1d;\n  array[d, d] int ar_2d;\n  array[d, d, d] int ar_3d;\n}\nIn the case where variable d is 1, all arrays will contain a single value. If array variable ar_d1 contains value 7, 2-D array variable ar_d2 contains (an array which contains) value 8, and 3-D array variable ar_d3 contains (an array which contains an array which contains) value 9, the JSON representation is:\n{ \"ar_d1\" : [7],\n  \"ar_d2\" : [[8]],\n  \"ar_d3\" : [[[9]]]\n}\nHowever, in the case where variable d is 0, ar_d1 is empty, i.e., it contains no values, as is ar_d2, ar_d3, and the JSON representation is\n{ \"d\" : 0,\n  \"ar_d1\" : [ ],\n  \"ar_d2\" : [ ],\n  \"ar_d3\" : [ ]\n}",
+    "crumbs": [
+      "Appendices",
+      "JSON Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/json_apdx.html#creating-json-files",
+    "href": "cmdstan-guide/json_apdx.html#creating-json-files",
+    "title": "JSON Format for CmdStan",
+    "section": "",
+    "text": "You can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the write_stan_json() function provided by the CmdStanR interface.",
+    "crumbs": [
+      "Appendices",
+      "JSON Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/json_apdx.html#json-syntax-summary",
+    "href": "cmdstan-guide/json_apdx.html#json-syntax-summary",
+    "title": "JSON Format for CmdStan",
+    "section": "",
+    "text": "JSON is a data interchange notation, defined by an EMCA standard. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:\n\nStructural tokens are the left and right curly bracket {}, left and right square bracket [], the semicolon ;, and the comma ,.\nLiteral tokens must always be in lowercase. There are three literal tokens: true, false, null.\nA primitive value is a single token which is either a literal, a string, or a number.\nA string consists of zero or more Unicode characters enclosed in double quotes, e.g. \"foo\". A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g. \"\\\\uHHHH\" where HHHH is the Unicode code point in hex.\nNumbers are represented using either decimal notation or scientific notation. The following are examples of numbers: 17, 17.2, -17.2, -17.2e8, 17.2e-8.  There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.\nThe special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string \"Inf\", the string \"Infinity\", or the atom Infinity. Negative infinity can be represented as the string \"-Inf\", the string \"-Infinity\", or the atom -Infinity. Not-a-number can be represented as the string \"NaN\" or the atom NaN. These values may be mixed with other numerical types.\nA complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number \\(2.3 - 1.83i\\) would be represented in JSON as the two-element array [2.3, -1.83].\nA JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: [], [1], [0.2, \"-inf\", true].\nVectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector \\([1 \\quad 2]^{\\top}\\) and the row vector \\([1 \\quad 2]\\) are represented by the JSON array [1, 2].\nComplex vectors are represented as arrays of two-element arrays. For example, the complex vector \\([2.3 - 1.83i \\quad -4.8 +\n2i]^{\\top}\\) is represented as [[2.3, -1.83], [-4.8, 2]] in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).\nMatrices are represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 & 2.7 & -9.8 \\\\\n4.2 & 1.8 & -7.3\n\\end{bmatrix}\n\\end{equation*}\\] is represented in JSON as [[1, 2.7, -9.8],   [4.2, 1.8, -7.3]].\nComplex matrices are also represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) complex matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 + 2i & 3 - 4.2i & 13.1 + 2.7i \\\\\n3.1 & -5i & 0\n\\end{bmatrix}\n\\end{equation*}\\] would be represented in JSON as [[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]].\nTuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple (1.5, 3.4) is represented in JSON as {\"1\": 1.5, \"2\": 3.4}.\nA name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.\nA JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: { }, {\"foo\": null}, {\"bar\" : 17, \"baz\" : [14,15,16.6] }.",
+    "crumbs": [
+      "Appendices",
+      "JSON Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/json_apdx.html#stan-data-types-in-json-notation",
+    "href": "cmdstan-guide/json_apdx.html#stan-data-types-in-json-notation",
+    "title": "JSON Format for CmdStan",
+    "section": "",
+    "text": "Stan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:\n{ \"N\" : 10, \"y\" : [0,1,0,0,0,0,0,0,0,1] }\nMatrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:\ndata {\n  int d1;\n  int d2;\n  int d3;\n  array[d1, d2, d3] int ar;\n}\nthe following JSON input would be valid:\n{ \"d1\" : 2,\n  \"d2\" : 3,\n  \"d3\" : 4,\n  \"ar\" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],\n          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]\n}\nJSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.\nAll data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.\n\n\n\n\n\n\n\nStan declaration\nJSON encoding\n\n\n\n\nint i\n\"i\": 17\n\n\n\n\n\n\nreal a\n\"a\" : 17\n\n\n\n\"a\" : 17.2\n\n\n\n\"a\" : \"NaN\"\n\n\n\n\"a\" : \"+inf\"\n\n\n\n\"a\" : \"-inf\"\n\n\n\n\n\n\ncomplex z\n\"z\": [1, -2.3]\n\n\narray[5] int\n\"a\" : [1, 2, 3, 4, 5]\n\n\narray[5] real a\n\"a\" : [ 1, 2, 3.3, \"NaN\", 5 ]\n\n\narray[2] complex b\n\"b\" : [[1, -2.3], [4.9, 0]]\n\n\nvector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nrow_vector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nmatrix[2, 3] a\n\"a\" : [[ 1, 2, 3 ], [ 4, 5, 6]]\n\n\ncomplex_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_row_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_matrix[2, 3] d\n\"d\" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]\n\n\ntuple(real, array[2] int) t\n\"t\" : { \"1\": 1.4, \"2\": [1, 2]}\n\n\n\n\n\nJSON notation is not able to distinguish between multi-dimensional arrays where any dimension is \\(0\\), e.g., a 2-D array with dimensions \\((1,0)\\), i.e., an array which contains a single array which is empty, has JSON representation [ ]. To see how this works, consider the following Stan program data block:\ndata {\n  int d;\n  array[d] int ar_1d;\n  array[d, d] int ar_2d;\n  array[d, d, d] int ar_3d;\n}\nIn the case where variable d is 1, all arrays will contain a single value. If array variable ar_d1 contains value 7, 2-D array variable ar_d2 contains (an array which contains) value 8, and 3-D array variable ar_d3 contains (an array which contains an array which contains) value 9, the JSON representation is:\n{ \"ar_d1\" : [7],\n  \"ar_d2\" : [[8]],\n  \"ar_d3\" : [[[9]]]\n}\nHowever, in the case where variable d is 0, ar_d1 is empty, i.e., it contains no values, as is ar_d2, ar_d3, and the JSON representation is\n{ \"d\" : 0,\n  \"ar_d1\" : [ ],\n  \"ar_d2\" : [ ],\n  \"ar_d3\" : [ ]\n}",
+    "crumbs": [
+      "Appendices",
+      "JSON Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/log_prob_config.html",
+    "href": "cmdstan-guide/log_prob_config.html",
+    "title": "Extracting log probabilities and gradients for diagnostics",
+    "section": "",
+    "text": "CmdStan can return the computed log probability and the gradient with respect to a set of parameters.\nThis is similar to the diagnose subcommand, but the output format differs and the results here are not compared with those from finite differences.\nNote: Startup and data initialization costs mean that this method is not an efficient way to calculate these quantities. It is provided only for convenience and should not be used for serious computation.\n\n\nThis method takes 3 arguments:\n\njacobian - Whether or not the Jacobian adjustment for constrained parameters should be included in the gradient. Default value is true (include adjustment).\nconstrained_params - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using JSON format. Alternatively, the input file can be set of draws in StanCSV format.\nunconstrained_params - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called params_r, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.\n\nOnly one of constrained_params and unconstrained_params can be specified.\nFor more on the differences between constrained and unconstrained parameters, see the Stan reference manual section on variable transforms.\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the log_prob method is reported at the beginning of the output file as CSV comments.\nColumn headers, the first column is labelled lp__, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.\nValues which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).\n\nFor example, if we have a file called params.json:\n{\n    \"theta\" : 0.1\n}\nWe can run the example model:\n/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json\nThis yields\n# method = log_prob\n#   log_prob\n#     unconstrained_params =  (Default)\n#     constrained_params = params.json\n#     jacobian = true (Default)\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2390820139 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlp_,theta\n-7.856,1.8",
+    "crumbs": [
+      "Running CmdStan",
+      "Extracting log probabilities and gradients for diagnostics"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/log_prob_config.html#configuration",
+    "href": "cmdstan-guide/log_prob_config.html#configuration",
+    "title": "Extracting log probabilities and gradients for diagnostics",
+    "section": "",
+    "text": "This method takes 3 arguments:\n\njacobian - Whether or not the Jacobian adjustment for constrained parameters should be included in the gradient. Default value is true (include adjustment).\nconstrained_params - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using JSON format. Alternatively, the input file can be set of draws in StanCSV format.\nunconstrained_params - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called params_r, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.\n\nOnly one of constrained_params and unconstrained_params can be specified.\nFor more on the differences between constrained and unconstrained parameters, see the Stan reference manual section on variable transforms.",
+    "crumbs": [
+      "Running CmdStan",
+      "Extracting log probabilities and gradients for diagnostics"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/log_prob_config.html#csv-output",
+    "href": "cmdstan-guide/log_prob_config.html#csv-output",
+    "title": "Extracting log probabilities and gradients for diagnostics",
+    "section": "",
+    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the log_prob method is reported at the beginning of the output file as CSV comments.\nColumn headers, the first column is labelled lp__, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.\nValues which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).\n\nFor example, if we have a file called params.json:\n{\n    \"theta\" : 0.1\n}\nWe can run the example model:\n/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json\nThis yields\n# method = log_prob\n#   log_prob\n#     unconstrained_params =  (Default)\n#     constrained_params = params.json\n#     jacobian = true (Default)\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2390820139 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlp_,theta\n-7.856,1.8",
+    "crumbs": [
+      "Running CmdStan",
+      "Extracting log probabilities and gradients for diagnostics"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/optimize_config.html",
+    "href": "cmdstan-guide/optimize_config.html",
+    "title": "Optimization",
+    "section": "",
+    "text": "The CmdStan executable can run Stan’s optimization algorithms, which provide a deterministic method to find the posterior mode. If the posterior is not convex, there is no guarantee Stan will be able to find the global optimum as opposed to a local optimum of log probability.\nThe full set of configuration options available for the optimize method is available by using the optimize help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the optimizer console output and in the output CSV file’s comments.\nThe executable does not need to be recompiled in order to switch from sampling to optimization, and the data input format is the same. The following is a minimal call to Stan’s optimizer using defaults for everything but the location of the data file.\n&gt; ./bernoulli optimize data file=bernoulli.data.json\nExecuting this command prints both output to the console and to a CSV file.\nThe first part of the console output reports on the configuration used. The above command uses all default configurations, therefore the optimizer used is the L-BFGS optimizer and its default initial stepsize and tolerances for monitoring convergence:\n ./bernoulli optimize data file=bernoulli.data.json\nmethod = optimize\n  optimize\n    algorithm = lbfgs (Default)\n      lbfgs\n        init_alpha = 0.001 (Default)\n        tol_obj = 1e-12 (Default)\n        tol_rel_obj = 10000 (Default)\n        tol_grad = 1e-08 (Default)\n        tol_rel_grad = 1e+07 (Default)\n        tol_param = 1e-08 (Default)\n        history_size = 5 (Default)\n    jacobian = false (Default)\n    iter = 2000 (Default)\n    save_iterations = false (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 87122538 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nThe second part of the output indicates how well the algorithm fared, here converging and terminating normally. The numbers reported indicate that it took 5 iterations and 8 gradient evaluations. This is, not surprisingly, far fewer iterations than required for sampling; even fewer iterations would be used with less stringent user-specified convergence tolerances. The alpha value is for step size used. In the final state the change in parameters was roughly \\(0.002\\) and the length of the gradient roughly 3e-05 (\\(0.00003\\)).\nInitial log joint probability = -6.85653\n    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes\n       5      -5.00402    0.00184936   3.35074e-05           1           1        8\nOptimization terminated normally:\n  Convergence detected: relative gradient magnitude is below tolerance\nThe output from optimization is written into the file output.csv by default. The output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as comment lines:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = optimize\n#   optimize\n#     algorithm = lbfgs (Default)\n#       lbfgs\n#         init_alpha = 0.001 (Default)\n#         tol_obj = 1e-12 (Default)\n#         tol_rel_obj = 10000 (Default)\n#         tol_grad = 1e-08 (Default)\n#         tol_rel_grad = 1e+07 (Default)\n#         tol_param = 1e-08 (Default)\n#         history_size = 5 (Default)\n#     jacobian = false (Default)\n#     iter = 2000 (Default)\n#     save_iterations = false (Default)\nFollowing the config information are two lines of output, the CSV headers and the recorded values:\nlp__,converged__,theta\n-5.00402,31,0.200003\nNote that everything is a comment other than a line for the header, and a line for the values. Here, the header indicates the unnormalized log probability with lp__, algorithm status in converged__, and the model parameter theta. The maximum log probability is -5.0 and the posterior mode for theta is 0.20. The mode exactly matches what we would expect from the data. Because the prior was uniform, the result 0.20 represents the maximum likelihood estimate (MLE) for the very simple Bernoulli model. Note that no uncertainty is reported.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value is false, i.e., intermediate iterations are not saved to the output file. Running the optimizer with save_iterations=true writes both the initial log joint probability and values for all iterations to the output CSV file.\nRunning the example model with option save_iterations=true, i.e., the command\n&gt; ./bernoulli optimize save_iterations=1 data file=bernoulli.data.json\nproduces CSV file output rows:\nlp__,converged__,theta\n-6.85653,0,0.493689\n-6.10128,0,0.420936\n-5.02953,0,0.22956\n-5.00517,0,0.206107\n-5.00403,0,0.200299\n-5.00402,31,0.200003\n\n\nThe converged__ column is used to indicate the status of the algorithm. It can take on the following values.\n\n\n\n\n\n\n\nNumeric value\nMeaning\n\n\n\n\n-1\nLine search failed to achieve a sufficient decrease, no more progress can be made\n\n\n0\nSuccessful step completed\n\n\n10\nConvergence detected: absolute parameter change was below tolerance\n\n\n20\nConvergence detected: absolute change in objective function was below tolerance\n\n\n21\nConvergence detected: relative change in objective function was below tolerance\n\n\n30\nConvergence detected: gradient norm is below tolerance\n\n\n31\nConvergence detected: relative gradient magnitude is below tolerance\n\n\n40\nMaximum number of iterations hit, may not be at an optima\n\n\n\nNote that not all algorithms can produce all codes, and some, such as 0, will only be observed if save_iterations is used.\nA comment explaining the final value (with the same contents as the above table) is added at the end of the CSV file.\n\n\n\nIf there are constrained parameters, Stan makes a transformation to an unconstrained space and runs the optimization algorithm in the unconstrained space.\nThe jacobian argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of inverse parameter transforms.\nIf the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.\nIf the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\n\n\n\nThe algorithm argument specifies the optimization algorithm. This argument takes one of the following three values:\n\nlbfgs A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.\nbfgs A quasi-Newton optimizer.\nnewton A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.\n\nSee the Stan Reference Manual’s Optimization chapter for a description of these algorithms.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value isfalse`, i.e., intermediate iterations are not saved to the output file.\n\n\n\nFor both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the BFGS and L-BFGS configuration section for details on the convergence tests.\nBoth BFGS and L-BFGS have the following configuration arguments:\n\ninit_alpha - The initial step size parameter. Must be a positive real number. Default value is \\(0.001\\)\ntol_obj - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is \\(1^{-12}\\).\ntol_rel_obj - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is \\(1^{4}\\).\ntol_grad - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is \\(1^{-8}\\).\ntol_rel_grad - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is \\(1^{7}\\).\ntol_param - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is \\(1^{-8}\\).\n\nThe init_alpha argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set init_alpha to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.\nIn addition to the above, the L-BFGS algorithm has argument history_size which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (\\(5\\)-\\(10\\)) are sufficient; the default value is \\(5\\).\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.\n\n\n\nThere are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.",
+    "crumbs": [
+      "Running CmdStan",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/optimize_config.html#meaning-of-the-converged__-column",
+    "href": "cmdstan-guide/optimize_config.html#meaning-of-the-converged__-column",
+    "title": "Optimization",
+    "section": "",
+    "text": "The converged__ column is used to indicate the status of the algorithm. It can take on the following values.\n\n\n\n\n\n\n\nNumeric value\nMeaning\n\n\n\n\n-1\nLine search failed to achieve a sufficient decrease, no more progress can be made\n\n\n0\nSuccessful step completed\n\n\n10\nConvergence detected: absolute parameter change was below tolerance\n\n\n20\nConvergence detected: absolute change in objective function was below tolerance\n\n\n21\nConvergence detected: relative change in objective function was below tolerance\n\n\n30\nConvergence detected: gradient norm is below tolerance\n\n\n31\nConvergence detected: relative gradient magnitude is below tolerance\n\n\n40\nMaximum number of iterations hit, may not be at an optima\n\n\n\nNote that not all algorithms can produce all codes, and some, such as 0, will only be observed if save_iterations is used.\nA comment explaining the final value (with the same contents as the above table) is added at the end of the CSV file.",
+    "crumbs": [
+      "Running CmdStan",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/optimize_config.html#jacobian-adjustments",
+    "href": "cmdstan-guide/optimize_config.html#jacobian-adjustments",
+    "title": "Optimization",
+    "section": "",
+    "text": "If there are constrained parameters, Stan makes a transformation to an unconstrained space and runs the optimization algorithm in the unconstrained space.\nThe jacobian argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of inverse parameter transforms.\nIf the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.\nIf the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.",
+    "crumbs": [
+      "Running CmdStan",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/optimize_config.html#optimization-algorithms",
+    "href": "cmdstan-guide/optimize_config.html#optimization-algorithms",
+    "title": "Optimization",
+    "section": "",
+    "text": "The algorithm argument specifies the optimization algorithm. This argument takes one of the following three values:\n\nlbfgs A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.\nbfgs A quasi-Newton optimizer.\nnewton A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.\n\nSee the Stan Reference Manual’s Optimization chapter for a description of these algorithms.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value isfalse`, i.e., intermediate iterations are not saved to the output file.",
+    "crumbs": [
+      "Running CmdStan",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers",
+    "href": "cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers",
+    "title": "Optimization",
+    "section": "",
+    "text": "For both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the BFGS and L-BFGS configuration section for details on the convergence tests.\nBoth BFGS and L-BFGS have the following configuration arguments:\n\ninit_alpha - The initial step size parameter. Must be a positive real number. Default value is \\(0.001\\)\ntol_obj - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is \\(1^{-12}\\).\ntol_rel_obj - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is \\(1^{4}\\).\ntol_grad - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is \\(1^{-8}\\).\ntol_rel_grad - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is \\(1^{7}\\).\ntol_param - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is \\(1^{-8}\\).\n\nThe init_alpha argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set init_alpha to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.\nIn addition to the above, the L-BFGS algorithm has argument history_size which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (\\(5\\)-\\(10\\)) are sufficient; the default value is \\(5\\).\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.",
+    "crumbs": [
+      "Running CmdStan",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/optimize_config.html#the-newton-optimizer",
+    "href": "cmdstan-guide/optimize_config.html#the-newton-optimizer",
+    "title": "Optimization",
+    "section": "",
+    "text": "There are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.",
+    "crumbs": [
+      "Running CmdStan",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/pathfinder_config.html",
+    "href": "cmdstan-guide/pathfinder_config.html",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "section": "",
+    "text": "The CmdStan method pathfinder uses the Pathfinder algorithm of Zhang et al. (2022), which is further described in the Stan Reference Manual.\nA single run of the Pathfinder algorithm generates a set of approximate draws. Inference is improved by running multiple Pathfinder instances and using Pareto-smoothed importance resampling (PSIS) of the resulting sets of draws. This better matches non-normal target densities and also eliminates minor modes.\nThe pathfinder method runs multi-path Pathfinder by default, which returns a PSIS sample over the draws from several individual (“single-path”) Pathfinder runs. Argument num_paths specifies the number of single-path Pathfinders, the default is \\(4\\). If num_paths is set to 1, then only one individual Pathfinder is run without the PSIS reweighting of the sample.\nThe full set of configuration options available for the pathfinder method is available by using the pathfinder help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.\nThe following is a minimal call to the Pathfinder algorithm using defaults for everything but the location of the data file.\n&gt; ./bernoulli pathfinder data file=bernoulli.data.R\nExecuting this command prints output both to the console and to csv files.\nThe first part of the console output reports on the configuration used.\nmethod = pathfinder\n  pathfinder\n    init_alpha = 0.001 (Default)\n    tol_obj = 1e-12 (Default)\n    tol_rel_obj = 10000 (Default)\n    tol_grad = 1e-08 (Default)\n    tol_rel_grad = 1e+07 (Default)\n    tol_param = 1e-08 (Default)\n    history_size = 5 (Default)\n    num_psis_draws = 1000 (Default)\n    num_paths = 4 (Default)\n    save_single_paths = false (Default)\n    psis_resample = true (Default)\n    calculate_lp = true (Default)\n    max_lbfgs_iters = 1000 (Default)\n    num_draws = 1000 (Default)\n    num_elbo_draws = 25 (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2790476610 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nThe rest of the output describes the progression of the algorithm.\nBy default, the Pathfinder algorithm runs 4 single-path Pathfinders in parallel, then uses importance resampling on the set of returned draws to produce the specified number of draws.\nPath [1] :Initial log joint density = -11.543343\nPath [1] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      1.070e-03   1.707e-05    1.000e+00  1.000e+00       126 -6.220e+00 -6.220e+00\nPath [1] :Best Iter: [5] ELBO (-6.219833) evaluations: (126)\nPath [2] :Initial log joint density = -7.443345\nPath [2] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      9.936e-05   3.738e-07    1.000e+00  1.000e+00       126 -6.164e+00 -6.164e+00\nPath [2] :Best Iter: [5] ELBO (-6.164015) evaluations: (126)\nPath [3] :Initial log joint density = -18.986308\nPath [3] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      2.996e-04   4.018e-06    1.000e+00  1.000e+00       126 -6.201e+00 -6.201e+00\nPath [3] :Best Iter: [5] ELBO (-6.200559) evaluations: (126)\nPath [4] :Initial log joint density = -8.304453\nPath [4] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      2.814e-04   2.034e-06    1.000e+00  1.000e+00       126 -6.221e+00 -6.221e+00\nPath [4] :Best Iter: [3] ELBO (-6.161276) evaluations: (126)\nTotal log probability function evaluations:8404\n\n\n\nnum_psis_draws - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is \\(1000\\).\nnum_paths - Number of single pathfinders. Must be a positive integer. Default value is \\(4\\).\nsave_single_paths - When true, save outputs from single pathfinders. Valid values: [true, false]. Default is false.\nmax_lbfgs_iters - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is \\(1000\\).\nnum_draws - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is \\(1000\\). Can differ from num_psis_draws.\nnum_elbo_draws - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is \\(25\\).\npsis_resample - If true, perform PSIS resampling on draws returned from individual pathfinders. If false, returns all num_paths * num_draws draws from the individual pathfinders. Valid values: [true, false]. Default is true.\ncalculate_lp - If true, log probabilities of the approximate draws are calculated and returned with the output. If false, each pathfinder will only calculate the lp values needed for the ELBO calculation. If false, PSIS resampling cannot be performed and the algorithm returns num_paths * num_draws samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: [true, false]. Default is true.\n\n\n\n\nArguments init_alpha through history_size are the full set of arguments to the L-BFGS optimizer and have the same defaults for optimization.\n\n\n\nBy default, the pathfinder method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a StanCSV file.\nThe CSV files have the following structure:\nThe initial CSV comment rows contain the complete set of CmdStan configuration options.\n...\n# method = pathfinder\n#   pathfinder\n#     init_alpha = 0.001 (Default)\n#     tol_obj = 9.9999999999999998e-13 (Default)\n#     tol_rel_obj = 10000 (Default)\n#     tol_grad = 1e-08 (Default)\n#     tol_rel_grad = 10000000 (Default)\n#     tol_param = 1e-08 (Default)\n#     history_size = 5 (Default)\n#     num_psis_draws = 1000 (Default)\n#     num_paths = 4 (Default)\n#     psis_resample = 1 (Default)\n#     calculate_lp = 1 (Default)\n#     save_single_paths = 0 (Default)\n#     max_lbfgs_iters = 1000 (Default)\n#     num_draws = 1000 (Default)\n#     num_elbo_draws = 25 (Default)\n...\nNext is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs lp_approx__, the log density in the approximating distribution, and lp__, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.\nlp_approx__,lp__,theta\n-2.4973, -8.2951, 0.0811852\n-0.87445, -7.06526, 0.160207\n-0.812285, -7.07124, 0.35819\n...\nThe final lines are comment lines which give timing information.\n# Elapsed Time: 0.016000 seconds (Pathfinders)\n#               0.003000 seconds (PSIS)\n#               0.019000 seconds (Total)\nPathfinder provides option save_single_paths which will save output from the single-path Pathfinder runs.\n\n\n\nThe boolean option save_single_paths is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When save_single_paths is true, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name output.csv and default number of runs (4), the resulting CSV files are\noutput.csv\noutput_path_1.csv\noutput_path_1.json\noutput_path_2.csv\noutput_path_2.json\noutput_path_3.csv\noutput_path_3.json\noutput_path_4.csv\noutput_path_4.json\nThe individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.\nTo see how this works, we run Pathfinder on the centered-parameterization of the eight-schools model, where the posterior distribution has a funnel shape:\n&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json\nEach JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.\n{\n  \"0\": {\n    \"iter\": 0,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]\n  },\n  \"1\": {\n    \"iter\": 1,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],\n    \"history_size\": 1,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],\n    \"logDetCholHk\": -2.6839,\n    \"L_approx\": [[-0.0630456, -0.0187959], [0, 1.08328]],\n    \"Qk\": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],\n    \"alpha\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n    \"full\": false,\n    \"lbfgs_note\": \"\"\n  },\n  ...,\n  \"171\": {\n    \"iter\": 171,\n    \"unconstrained_parameters\": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],\n    \"grads\": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],\n    \"history_size\": 5,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],\n    \"logDetCholHk\": 299.023,\n    \"L_approx\": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],\n    \"Qk\": [],\n    \"alpha\": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],\n    \"full\": true,\n    \"lbfgs_note\": \"\"\n  },\n  \"172\": {\n    \"iter\": 172,\n    \"unconstrained_parameters\": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],\n    \"grads\": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],\n    \"history_size\": 5,\n    \"lbfgs_success\": false,\n    \"pathfinder_success\": false,\n    \"lbfgs_note\": \"\"\n  }\n}\nOption num_paths=1 runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments num_paths=1 save_single_paths=true creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.",
+    "crumbs": [
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/pathfinder_config.html#pathfinder-configuration",
+    "href": "cmdstan-guide/pathfinder_config.html#pathfinder-configuration",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "section": "",
+    "text": "num_psis_draws - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is \\(1000\\).\nnum_paths - Number of single pathfinders. Must be a positive integer. Default value is \\(4\\).\nsave_single_paths - When true, save outputs from single pathfinders. Valid values: [true, false]. Default is false.\nmax_lbfgs_iters - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is \\(1000\\).\nnum_draws - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is \\(1000\\). Can differ from num_psis_draws.\nnum_elbo_draws - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is \\(25\\).\npsis_resample - If true, perform PSIS resampling on draws returned from individual pathfinders. If false, returns all num_paths * num_draws draws from the individual pathfinders. Valid values: [true, false]. Default is true.\ncalculate_lp - If true, log probabilities of the approximate draws are calculated and returned with the output. If false, each pathfinder will only calculate the lp values needed for the ELBO calculation. If false, PSIS resampling cannot be performed and the algorithm returns num_paths * num_draws samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: [true, false]. Default is true.",
+    "crumbs": [
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/pathfinder_config.html#l-bfgs-configuration",
+    "href": "cmdstan-guide/pathfinder_config.html#l-bfgs-configuration",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "section": "",
+    "text": "Arguments init_alpha through history_size are the full set of arguments to the L-BFGS optimizer and have the same defaults for optimization.",
+    "crumbs": [
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/pathfinder_config.html#pathfinder_csv",
+    "href": "cmdstan-guide/pathfinder_config.html#pathfinder_csv",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "section": "",
+    "text": "By default, the pathfinder method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a StanCSV file.\nThe CSV files have the following structure:\nThe initial CSV comment rows contain the complete set of CmdStan configuration options.\n...\n# method = pathfinder\n#   pathfinder\n#     init_alpha = 0.001 (Default)\n#     tol_obj = 9.9999999999999998e-13 (Default)\n#     tol_rel_obj = 10000 (Default)\n#     tol_grad = 1e-08 (Default)\n#     tol_rel_grad = 10000000 (Default)\n#     tol_param = 1e-08 (Default)\n#     history_size = 5 (Default)\n#     num_psis_draws = 1000 (Default)\n#     num_paths = 4 (Default)\n#     psis_resample = 1 (Default)\n#     calculate_lp = 1 (Default)\n#     save_single_paths = 0 (Default)\n#     max_lbfgs_iters = 1000 (Default)\n#     num_draws = 1000 (Default)\n#     num_elbo_draws = 25 (Default)\n...\nNext is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs lp_approx__, the log density in the approximating distribution, and lp__, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.\nlp_approx__,lp__,theta\n-2.4973, -8.2951, 0.0811852\n-0.87445, -7.06526, 0.160207\n-0.812285, -7.07124, 0.35819\n...\nThe final lines are comment lines which give timing information.\n# Elapsed Time: 0.016000 seconds (Pathfinders)\n#               0.003000 seconds (PSIS)\n#               0.019000 seconds (Total)\nPathfinder provides option save_single_paths which will save output from the single-path Pathfinder runs.",
+    "crumbs": [
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/pathfinder_config.html#single-path-pathfinder-outputs",
+    "href": "cmdstan-guide/pathfinder_config.html#single-path-pathfinder-outputs",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "section": "",
+    "text": "The boolean option save_single_paths is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When save_single_paths is true, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name output.csv and default number of runs (4), the resulting CSV files are\noutput.csv\noutput_path_1.csv\noutput_path_1.json\noutput_path_2.csv\noutput_path_2.json\noutput_path_3.csv\noutput_path_3.json\noutput_path_4.csv\noutput_path_4.json\nThe individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.\nTo see how this works, we run Pathfinder on the centered-parameterization of the eight-schools model, where the posterior distribution has a funnel shape:\n&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json\nEach JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.\n{\n  \"0\": {\n    \"iter\": 0,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]\n  },\n  \"1\": {\n    \"iter\": 1,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],\n    \"history_size\": 1,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],\n    \"logDetCholHk\": -2.6839,\n    \"L_approx\": [[-0.0630456, -0.0187959], [0, 1.08328]],\n    \"Qk\": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],\n    \"alpha\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n    \"full\": false,\n    \"lbfgs_note\": \"\"\n  },\n  ...,\n  \"171\": {\n    \"iter\": 171,\n    \"unconstrained_parameters\": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],\n    \"grads\": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],\n    \"history_size\": 5,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],\n    \"logDetCholHk\": 299.023,\n    \"L_approx\": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],\n    \"Qk\": [],\n    \"alpha\": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],\n    \"full\": true,\n    \"lbfgs_note\": \"\"\n  },\n  \"172\": {\n    \"iter\": 172,\n    \"unconstrained_parameters\": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],\n    \"grads\": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],\n    \"history_size\": 5,\n    \"lbfgs_success\": false,\n    \"pathfinder_success\": false,\n    \"lbfgs_note\": \"\"\n  }\n}\nOption num_paths=1 runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments num_paths=1 save_single_paths=true creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.",
+    "crumbs": [
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html",
+    "href": "cmdstan-guide/rdump_apdx.html",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "NOTE: Although the RDump format is still supported, I/O with JSON is faster and recommended. See the chapter on JSON for more details.\nRDump format can be used to represent values for Stan variables. This format was introduced in SPLUS and is used in R, JAGS, and in BUGS (but with a different ordering).\nA dump file is structured as a sequence of variable definitions. Each variable is defined in terms of its dimensionality and its values. There are three kinds of variable declarations: - scalars - sequences - general arrays\n\n\nDump files can be created from R using RStan, via the rstan package function stan_rdump. Stan RDump files must be created via stan_rdump and not by R’s native dump function because R’s dump function uses a richer syntax than is supported by the underlying Stan i/o libraries.\n\n\n\nA simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable y with value \\(17.2\\):\ny &lt;- 17.2\n\n\n\nOne-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.\nn &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)\nArrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.\nSequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as\nn &lt;- 1:3\nThe sequence denoted by 1:3 is of length \\(3\\), running from \\(1\\) to \\(3\\) inclusive. The colon notation allows sequences going from high to low. The following are equivalent:\nn &lt;- 2:-2\nn &lt;- c(2,1,0,-1,-2)\nAs a special case, a sequence of zeros can also be represented in the dump format by integer(x) and double(x), for type int and double, respectively. Here x is a non-negative integer to specify the length. If x is \\(0\\), it can be omitted. The following are some examples.\nx1 &lt;- integer()\nx2 &lt;- integer(0)\nx3 &lt;- integer(2)\ny1 &lt;- double()\ny2 &lt;- double(0)\ny3 &lt;- double(2)\n\n\n\nFor more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a \\(2 \\times 3\\) array:\ny &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))\nData is stored column-major, thus the values for y will be:\ny[1, 1] = 1\ny[1, 2] = 3\ny[1, 3] = 5\ny[2, 1] = 2\ny[2, 2] = 4\ny[2, 3] = 6\nThe structure keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.\nIf the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.\ny &lt;- structure(1:6, .Dim = c(2,3))\nSequence notation can be used within any call to the generic c() function in R. In the above example, c(2,3) could be written as c(2:3).\nThe generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,\nz &lt;- structure(1:24, .Dim = c(2,3,4))\nproduces a three-dimensional int (assignable to real) array z with values:\nz[1, 1, 1] = 1\nz[2, 1, 1] = 2\nz[1, 2, 1] = 3\nz[2, 2, 1] = 4\nz[1, 3, 1] = 5\nz[2, 3, 1] = 6\nz[1, 1, 2] = 7\nz[2, 1, 2] = 8\nz[1, 2, 2] = 9\nz[2, 2, 2] = 10\nz[1, 3, 2] = 11\nz[2, 3, 2] = 12\nz[1, 1, 3] = 13\nz[2, 1, 3] = 14\nz[1, 2, 3] = 15\nz[2, 2, 3] = 16\nz[1, 3, 3] = 17\nz[2, 3, 3] = 18\nz[1, 1, 4] = 19\nz[2, 1, 4] = 20\nz[1, 2, 4] = 21\nz[2, 2, 4] = 22\nz[1, 3, 4] = 23\nz[2, 3, 4] = 24\nIf the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.\nThe sequence of values inside structure can also be integer(x) or double(x). In particular, if one or more dimensions is zero, integer() can be put inside structure. For instance, the following example is supported by the dump format.\ny &lt;- structure(integer(), .Dim = c(2,0))\n\n\n\nThe dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.\n\n\nThe following three declarations have the same dump format for their data.\narray[K] real a;\nvector[K] b;\nrow_vector[K] c;\n\n\n\nThe following declarations have the same dump format.\narray[M, N] real a;\nmatrix[M, N] b;\n\n\n\nThe key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.\nFor the dump data format, the following declarations have the same arrangement.\narray[M, N] real a;\nmatrix[M, N] b;\narray[M] vector[N] c;\narray[M] row_vector[N] d;\nSimilarly, the following also have the same dump format.\narray[P, M, N] real a;\narray[P] matrix[M, N] b;\narray[P, M] vector[N] c;\narray[P, M] row_vector[N] d;\n\n\n\n\nAt this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.\n\n\n\nThere is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., \\(132.3\\)) or uses scientific notation (e.g., \\(1.323e2\\)), Stan assumes that the values are real.\nFor a single value, if there is no decimal point, it may be assigned to an int or real variable in Stan. An array value may only be assigned to an int array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.\nThe following dump file declares an integer value for y.\ny &lt;- 2\nThis definition can be used for a Stan variable y declared as real or as int. Assigning an integer value to a real variable automatically promotes the integer value to a real value.\nInteger values may optionally be followed by L or l, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.\ny &lt;- 2L\nThe following dump file provides a real value for y.\ny &lt;- 2.0\nEven though this is a round value, the occurrence of the decimal point in the value, \\(2.0\\), causes Stan to infer that y is real valued. This dump file may only be used for variables y declared as real in Stan.\n\n\nNumbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as \\(1e+06\\).\n\n\n\nStan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.\n\n\n\nValue\nPreferred Form\nAlternative Forms\n\n\n\n\npositive infinity\nInf\nInfinity, infinity\n\n\nnegative infinity\n-Inf\n-Infinity, -infinity\n\n\nnot a number\nNaN\n\n\n\n\nThese strings are not case sensitive, so inf may also be used for positive infinity, or NAN for not-a-number.\n\n\n\n\nIn order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.\n\"y\" &lt;- c(1,2,3) \\end{Verbatim}\n\n\n\nThe line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.\ny &lt;- 2\ny &lt;-\n3\nAlso following R, breaking before the assignment arrow are not allowed, so the following is invalid.\ny\n&lt;- 2 # Syntax Error\nLines may also be broken in the middle of sequences declared using the c(...) notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a \\(2 \\times 2 \\times 3\\) array is valid.\ny &lt;-\nstructure(c(1,2,3,\n4,5,6,7,8,9,10,11,\n12), .Dim = c(2,2,\n3))\nBecause there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as int or real.\n\n\n\nA more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.\ndefinition ::= name &lt;- value optional_semicolon\n\nname ::= char*     | ''' char* '''     | '\"' char* '\"'\n\nvalue ::= value&lt;int&gt; | value&lt;double&gt;\n\nvalue&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |\n'structure' '(' seq&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'       | 'structure'\n'(' zero_array&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'\n\nseq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;\n\nzero_array&lt;int&gt; ::= \"integer\" '(' &lt;non-negative int&gt;? ')'\n\nzero_array&lt;real&gt; ::= \"double\" '(' &lt;non-negative int&gt;? ')'\n\nseq&lt;real&gt; ::= cseq&lt;real&gt;\n\ncseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'\n\nvseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;\nThe template parameters T will be set to either int or real. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#creating-dump-files",
+    "href": "cmdstan-guide/rdump_apdx.html#creating-dump-files",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "Dump files can be created from R using RStan, via the rstan package function stan_rdump. Stan RDump files must be created via stan_rdump and not by R’s native dump function because R’s dump function uses a richer syntax than is supported by the underlying Stan i/o libraries.",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#scalar-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#scalar-variables",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "A simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable y with value \\(17.2\\):\ny &lt;- 17.2",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#sequence-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#sequence-variables",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "One-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.\nn &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)\nArrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.\nSequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as\nn &lt;- 1:3\nThe sequence denoted by 1:3 is of length \\(3\\), running from \\(1\\) to \\(3\\) inclusive. The colon notation allows sequences going from high to low. The following are equivalent:\nn &lt;- 2:-2\nn &lt;- c(2,1,0,-1,-2)\nAs a special case, a sequence of zeros can also be represented in the dump format by integer(x) and double(x), for type int and double, respectively. Here x is a non-negative integer to specify the length. If x is \\(0\\), it can be omitted. The following are some examples.\nx1 &lt;- integer()\nx2 &lt;- integer(0)\nx3 &lt;- integer(2)\ny1 &lt;- double()\ny2 &lt;- double(0)\ny3 &lt;- double(2)",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#array-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#array-variables",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "For more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a \\(2 \\times 3\\) array:\ny &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))\nData is stored column-major, thus the values for y will be:\ny[1, 1] = 1\ny[1, 2] = 3\ny[1, 3] = 5\ny[2, 1] = 2\ny[2, 2] = 4\ny[2, 3] = 6\nThe structure keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.\nIf the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.\ny &lt;- structure(1:6, .Dim = c(2,3))\nSequence notation can be used within any call to the generic c() function in R. In the above example, c(2,3) could be written as c(2:3).\nThe generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,\nz &lt;- structure(1:24, .Dim = c(2,3,4))\nproduces a three-dimensional int (assignable to real) array z with values:\nz[1, 1, 1] = 1\nz[2, 1, 1] = 2\nz[1, 2, 1] = 3\nz[2, 2, 1] = 4\nz[1, 3, 1] = 5\nz[2, 3, 1] = 6\nz[1, 1, 2] = 7\nz[2, 1, 2] = 8\nz[1, 2, 2] = 9\nz[2, 2, 2] = 10\nz[1, 3, 2] = 11\nz[2, 3, 2] = 12\nz[1, 1, 3] = 13\nz[2, 1, 3] = 14\nz[1, 2, 3] = 15\nz[2, 2, 3] = 16\nz[1, 3, 3] = 17\nz[2, 3, 3] = 18\nz[1, 1, 4] = 19\nz[2, 1, 4] = 20\nz[1, 2, 4] = 21\nz[2, 2, 4] = 22\nz[1, 3, 4] = 23\nz[2, 3, 4] = 24\nIf the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.\nThe sequence of values inside structure can also be integer(x) or double(x). In particular, if one or more dimensions is zero, integer() can be put inside structure. For instance, the following example is supported by the dump format.\ny &lt;- structure(integer(), .Dim = c(2,0))",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#matrix--and-vector-valued-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#matrix--and-vector-valued-variables",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "The dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.\n\n\nThe following three declarations have the same dump format for their data.\narray[K] real a;\nvector[K] b;\nrow_vector[K] c;\n\n\n\nThe following declarations have the same dump format.\narray[M, N] real a;\nmatrix[M, N] b;\n\n\n\nThe key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.\nFor the dump data format, the following declarations have the same arrangement.\narray[M, N] real a;\nmatrix[M, N] b;\narray[M] vector[N] c;\narray[M] row_vector[N] d;\nSimilarly, the following also have the same dump format.\narray[P, M, N] real a;\narray[P] matrix[M, N] b;\narray[P, M] vector[N] c;\narray[P, M] row_vector[N] d;",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#complex-valued-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#complex-valued-variables",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "At this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#integer--and-real-valued-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#integer--and-real-valued-variables",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "There is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., \\(132.3\\)) or uses scientific notation (e.g., \\(1.323e2\\)), Stan assumes that the values are real.\nFor a single value, if there is no decimal point, it may be assigned to an int or real variable in Stan. An array value may only be assigned to an int array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.\nThe following dump file declares an integer value for y.\ny &lt;- 2\nThis definition can be used for a Stan variable y declared as real or as int. Assigning an integer value to a real variable automatically promotes the integer value to a real value.\nInteger values may optionally be followed by L or l, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.\ny &lt;- 2L\nThe following dump file provides a real value for y.\ny &lt;- 2.0\nEven though this is a round value, the occurrence of the decimal point in the value, \\(2.0\\), causes Stan to infer that y is real valued. This dump file may only be used for variables y declared as real in Stan.\n\n\nNumbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as \\(1e+06\\).\n\n\n\nStan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.\n\n\n\nValue\nPreferred Form\nAlternative Forms\n\n\n\n\npositive infinity\nInf\nInfinity, infinity\n\n\nnegative infinity\n-Inf\n-Infinity, -infinity\n\n\nnot a number\nNaN\n\n\n\n\nThese strings are not case sensitive, so inf may also be used for positive infinity, or NAN for not-a-number.",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#quoted-variable-names",
+    "href": "cmdstan-guide/rdump_apdx.html#quoted-variable-names",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "In order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.\n\"y\" &lt;- c(1,2,3) \\end{Verbatim}",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#line-breaks",
+    "href": "cmdstan-guide/rdump_apdx.html#line-breaks",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "The line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.\ny &lt;- 2\ny &lt;-\n3\nAlso following R, breaking before the assignment arrow are not allowed, so the following is invalid.\ny\n&lt;- 2 # Syntax Error\nLines may also be broken in the middle of sequences declared using the c(...) notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a \\(2 \\times 2 \\times 3\\) array is valid.\ny &lt;-\nstructure(c(1,2,3,\n4,5,6,7,8,9,10,11,\n12), .Dim = c(2,2,\n3))\nBecause there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as int or real.",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/rdump_apdx.html#bnf-grammar-for-dump-data",
+    "href": "cmdstan-guide/rdump_apdx.html#bnf-grammar-for-dump-data",
+    "title": "RDump Format for CmdStan",
+    "section": "",
+    "text": "A more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.\ndefinition ::= name &lt;- value optional_semicolon\n\nname ::= char*     | ''' char* '''     | '\"' char* '\"'\n\nvalue ::= value&lt;int&gt; | value&lt;double&gt;\n\nvalue&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |\n'structure' '(' seq&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'       | 'structure'\n'(' zero_array&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'\n\nseq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;\n\nzero_array&lt;int&gt; ::= \"integer\" '(' &lt;non-negative int&gt;? ')'\n\nzero_array&lt;real&gt; ::= \"double\" '(' &lt;non-negative int&gt;? ')'\n\nseq&lt;real&gt; ::= cseq&lt;real&gt;\n\ncseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'\n\nvseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;\nThe template parameters T will be set to either int or real. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.",
+    "crumbs": [
+      "Appendices",
+      "RDump Format for CmdStan"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stanc.html",
+    "href": "cmdstan-guide/stanc.html",
+    "title": "stanc: Translating Stan to C++",
+    "section": "",
+    "text": "CmdStan translates Stan programs to C++ using the Stan compiler program which is included in the CmdStan release bin directory as program stanc. One can view the complete stanc documentation in the Stan User’s Guide.\nAs of release 2.22, the CmdStan Stan to C++ compiler is written in OCaml. This compiler is called “stanc3” and has has its own repository https://github.com/stan-dev/stanc3, from which pre-built binaries for Linux, Mac, and Windows can be downloaded.\n\n\nBefore the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:\nmake bin/stanc\nFor Windows:\nmake bin/stanc.exe\nThis is also done as part of the make build command.\n\n\n\nThe Stan compiler program stanc converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model bernoulli.stan. For Linux and Mac:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nFor Windows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nThe base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (a--z or A--Z) and contain only alphanumeric characters (a--z, A--Z, and 0--9) and underscores (_) and should not conflict with any C++ reserved keyword.\nThe C++ code implementing the class is written to the file bernoulli.hpp in the current directory. The final argument, bernoulli.stan, is the file from which to read the Stan program.\nIn practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via make, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:\n&gt; make examples/bernoulli/bernoulli\nFor Windows:\n&gt; make examples/bernoulli/bernoulli.exe\nThe makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable STANCFLAGS can be used to to override the default arguments to stanc, e.g.,\n&gt; make STANCFLAGS=\"--include-paths=~/foo\" examples/bernoulli/bernoulli",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stanc`: Translating Stan to C++"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stanc.html#instantiating-the-stanc-binary",
+    "href": "cmdstan-guide/stanc.html#instantiating-the-stanc-binary",
+    "title": "stanc: Translating Stan to C++",
+    "section": "",
+    "text": "Before the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:\nmake bin/stanc\nFor Windows:\nmake bin/stanc.exe\nThis is also done as part of the make build command.",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stanc`: Translating Stan to C++"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/stanc.html#the-stan-compiler-program",
+    "href": "cmdstan-guide/stanc.html#the-stan-compiler-program",
+    "title": "stanc: Translating Stan to C++",
+    "section": "",
+    "text": "The Stan compiler program stanc converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model bernoulli.stan. For Linux and Mac:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nFor Windows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nThe base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (a--z or A--Z) and contain only alphanumeric characters (a--z, A--Z, and 0--9) and underscores (_) and should not conflict with any C++ reserved keyword.\nThe C++ code implementing the class is written to the file bernoulli.hpp in the current directory. The final argument, bernoulli.stan, is the file from which to read the Stan program.\nIn practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via make, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:\n&gt; make examples/bernoulli/bernoulli\nFor Windows:\n&gt; make examples/bernoulli/bernoulli.exe\nThe makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable STANCFLAGS can be used to to override the default arguments to stanc, e.g.,\n&gt; make STANCFLAGS=\"--include-paths=~/foo\" examples/bernoulli/bernoulli",
+    "crumbs": [
+      "Tools and Utilities",
+      "`stanc`: Translating Stan to C++"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/variational_config.html",
+    "href": "cmdstan-guide/variational_config.html",
+    "title": "Variational Inference using ADVI",
+    "section": "",
+    "text": "Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) Kucukelbir et al. (2017). ADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO (evidence lower bound). ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. The measures of convergence are similar to the relative tolerance scheme of Stan’s optimization algorithms.\nThe algorithm progression consists of an adaptation phase followed by a sampling phase. The adaptation phase finds a good value for the step size scaling parameter eta. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. The algorithm runs until the mean change in ELBO drops below the specified tolerance.\nThe full set of configuration options available for the variational method is available by using the variational help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.\nThe following is a minimal call to Stan’s variational inference algorithm using defaults for everything but the location of the data file.\n&gt; ./bernoulli variational data file=bernoulli.data.R\nExecuting this command prints both output to the console and to a csv file.\nThe first part of the console output reports on the configuration used: the default option algorithm=meanfield and the default tolerances for monitoring the algorithm’s convergence.\nmethod = variational\n  variational\n    algorithm = meanfield (Default)\n      meanfield\n    iter = 10000 (Default)\n    grad_samples = 1 (Default)\n    elbo_samples = 100 (Default)\n    eta = 1 (Default)\n    adapt\n      engaged = true (Default)\n      iter = 50 (Default)\n    tol_rel_obj = 0.01 (Default)\n    eval_elbo = 100 (Default)\n    output_samples = 1000 (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2790599354 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nAfter the configuration has been displayed, informational and timing messages are output:\n------------------------------------------------------------\nEXPERIMENTAL ALGORITHM:\n  This procedure has not been thoroughly tested and may be unstable\n  or buggy. The interface is subject to change.\n------------------------------------------------------------\n\nGradient evaluation took 2.1e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.21 seconds.\nAdjust your expectations accordingly!\nThe rest of the output describes the progression of the algorithm. An adaptation phase finds a good value for the step size scaling parameter eta. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. That is, while the algorithm appears to have converged in \\(\\sim\\) 250 iterations, the algorithm runs for another few thousand iterations until mean change in ELBO drops below the default tolerance of 0.01.\nBegin eta adaptation.\nIteration:   1 / 250 [  0%]  (Adaptation)\nIteration:  50 / 250 [ 20%]  (Adaptation)\nIteration: 100 / 250 [ 40%]  (Adaptation)\nIteration: 150 / 250 [ 60%]  (Adaptation)\nIteration: 200 / 250 [ 80%]  (Adaptation)\nSuccess! Found best value [eta = 1] earlier than expected.\n\nBegin stochastic gradient ascent.\n  iter             ELBO   delta_ELBO_mean   delta_ELBO_med   notes\n   100           -6.131             1.000            1.000\n   200           -6.458             0.525            1.000\n   300           -6.300             0.359            0.051\n   400           -6.137             0.276            0.051\n   500           -6.243             0.224            0.027\n   600           -6.305             0.188            0.027\n   700           -6.289             0.162            0.025\n   800           -6.402             0.144            0.025\n   900           -6.103             0.133            0.025\n  1000           -6.314             0.123            0.027\n  1100           -6.348             0.024            0.025\n  1200           -6.244             0.020            0.018\n  1300           -6.293             0.019            0.017\n  1400           -6.250             0.017            0.017\n  1500           -6.241             0.015            0.010   MEDIAN ELBO CONVERGED\n\nDrawing a sample of size 1000 from the approximate posterior...\nCOMPLETED.\n\n\nStan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option algorithm=meanfield. The algorithm argument specifies the variational algorithm.\n\nalgorithm=meanfield - Use a fully factorized Gaussian for the approximation. This is the default algorithm.\nalgorithm=fullrank Use a Gaussian with a full-rank covariance matrix for the approximation.\n\n\n\n\n\niter=&lt;int&gt; Maximum number of iterations. Must be \\(&gt; 0\\). Default is \\(10000\\).\ngrad_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of gradients. Must be \\(&gt; 0\\). Default is \\(1\\).\nelbo_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of ELBO (objective function). Must be \\(&gt; 0\\). Default is \\(100\\).\neta=&lt;double&gt; Stepsize weighting parameter for adaptive stepsize sequence. Must be \\(&gt; 0\\). Default is \\(1.0\\).\nadapt Warmup Adaptation keyword, takes sub-arguments:\n\nengaged=&lt;boolean&gt; Adaptation engaged? Valid values: [true, false]. Default is true.\niter=&lt;int&gt; Maximum number of adaptation iterations. Must be \\(&gt; 0\\). Default is \\(50\\).\n\ntol_rel_obj=&lt;double&gt; Convergence tolerance on the relative norm of the objective. Must be \\(&gt; 0\\). Default is \\(0.01\\).\neval_elbo=&lt;int&gt; Evaluate ELBO every Nth iteration. Must be \\(&gt; 0\\). Default is 100.\noutput_samples=&lt;int&gt; Number of posterior samples to draw and save. Must be \\(&gt; 0\\). Default is 1000.\n\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the variational method is reported at the beginning of the sampler output file as CSV comments.\nThe first three output columns are labelled lp__, log_p__, log_g__, the rest are the model parameters.\nThe stepsize adaptation information is output as CSV comments following column header row.\nThe following line contains the mean of the variational approximation.\nThe rest of the output contains output_samples number of draws sampled from the variational approximation.\n\nTo illustrate, we call Stan’s variational inference on the example model and data:\n&gt; ./bernoulli variational data file=bernoulli.data.R\nBy default, the output file is output.csv.\nThe output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = variational\n#   variational\n#     algorithm = meanfield (Default)\n#       meanfield\n#     iter = 10000 (Default)\n#     grad_samples = 1 (Default)\n#     elbo_samples = 100 (Default)\n#     eta = 1 (Default)\n#     adapt\n#       engaged = true (Default)\n#       iter = 50 (Default)\n#     tol_rel_obj = 0.01 (Default)\n#     eval_elbo = 100 (Default)\n#     output_samples = 1000 (Default)\n...\nNext, the column header row:\nlp__,log_p__,log_g__,theta\nAdditional comments provide stepsize adaptation information:\n# Stepsize adaptation complete.\n# eta = 1\nFollowed by the data rows. The first line is special — it is the mean of the variational approximation.\n0,0,0,0.214911\nThat is, the estimate for theta given the data is 0.2.\nThe rest of the output contains output_samples number of draws samples from the variational approximation.\nThe following is a sample based on this approximation:\n0,-14.0252,-5.21718,0.770397\n0,-7.05063,-0.10025,0.162061\n0,-6.75031,-0.0191099,0.241606\n...\nThe header indicates the unnormalized log probability with lp__. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.",
+    "crumbs": [
+      "Running CmdStan",
+      "Variational Inference using ADVI"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/variational_config.html#variational-algorithms",
+    "href": "cmdstan-guide/variational_config.html#variational-algorithms",
+    "title": "Variational Inference using ADVI",
+    "section": "",
+    "text": "Stan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option algorithm=meanfield. The algorithm argument specifies the variational algorithm.\n\nalgorithm=meanfield - Use a fully factorized Gaussian for the approximation. This is the default algorithm.\nalgorithm=fullrank Use a Gaussian with a full-rank covariance matrix for the approximation.",
+    "crumbs": [
+      "Running CmdStan",
+      "Variational Inference using ADVI"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/variational_config.html#configuration",
+    "href": "cmdstan-guide/variational_config.html#configuration",
+    "title": "Variational Inference using ADVI",
+    "section": "",
+    "text": "iter=&lt;int&gt; Maximum number of iterations. Must be \\(&gt; 0\\). Default is \\(10000\\).\ngrad_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of gradients. Must be \\(&gt; 0\\). Default is \\(1\\).\nelbo_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of ELBO (objective function). Must be \\(&gt; 0\\). Default is \\(100\\).\neta=&lt;double&gt; Stepsize weighting parameter for adaptive stepsize sequence. Must be \\(&gt; 0\\). Default is \\(1.0\\).\nadapt Warmup Adaptation keyword, takes sub-arguments:\n\nengaged=&lt;boolean&gt; Adaptation engaged? Valid values: [true, false]. Default is true.\niter=&lt;int&gt; Maximum number of adaptation iterations. Must be \\(&gt; 0\\). Default is \\(50\\).\n\ntol_rel_obj=&lt;double&gt; Convergence tolerance on the relative norm of the objective. Must be \\(&gt; 0\\). Default is \\(0.01\\).\neval_elbo=&lt;int&gt; Evaluate ELBO every Nth iteration. Must be \\(&gt; 0\\). Default is 100.\noutput_samples=&lt;int&gt; Number of posterior samples to draw and save. Must be \\(&gt; 0\\). Default is 1000.",
+    "crumbs": [
+      "Running CmdStan",
+      "Variational Inference using ADVI"
+    ]
+  },
+  {
+    "objectID": "cmdstan-guide/variational_config.html#csv-output",
+    "href": "cmdstan-guide/variational_config.html#csv-output",
+    "title": "Variational Inference using ADVI",
+    "section": "",
+    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the variational method is reported at the beginning of the sampler output file as CSV comments.\nThe first three output columns are labelled lp__, log_p__, log_g__, the rest are the model parameters.\nThe stepsize adaptation information is output as CSV comments following column header row.\nThe following line contains the mean of the variational approximation.\nThe rest of the output contains output_samples number of draws sampled from the variational approximation.\n\nTo illustrate, we call Stan’s variational inference on the example model and data:\n&gt; ./bernoulli variational data file=bernoulli.data.R\nBy default, the output file is output.csv.\nThe output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = variational\n#   variational\n#     algorithm = meanfield (Default)\n#       meanfield\n#     iter = 10000 (Default)\n#     grad_samples = 1 (Default)\n#     elbo_samples = 100 (Default)\n#     eta = 1 (Default)\n#     adapt\n#       engaged = true (Default)\n#       iter = 50 (Default)\n#     tol_rel_obj = 0.01 (Default)\n#     eval_elbo = 100 (Default)\n#     output_samples = 1000 (Default)\n...\nNext, the column header row:\nlp__,log_p__,log_g__,theta\nAdditional comments provide stepsize adaptation information:\n# Stepsize adaptation complete.\n# eta = 1\nFollowed by the data rows. The first line is special — it is the mean of the variational approximation.\n0,0,0,0.214911\nThat is, the estimate for theta given the data is 0.2.\nThe rest of the output contains output_samples number of draws samples from the variational approximation.\nThe following is a sample based on this approximation:\n0,-14.0252,-5.21718,0.770397\n0,-7.05063,-0.10025,0.162061\n0,-6.75031,-0.0191099,0.241606\n...\nThe header indicates the unnormalized log probability with lp__. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.",
+    "crumbs": [
+      "Running CmdStan",
+      "Variational Inference using ADVI"
+    ]
+  },
+  {
+    "objectID": "functions-reference/binary_distributions.html",
+    "href": "functions-reference/binary_distributions.html",
+    "title": "Binary Distributions",
+    "section": "",
+    "text": "Binary probability distributions have support on \\(\\{0,1\\}\\), where 1 represents the value true and 0 the value false.\n\n\n\n\nIf \\(\\theta \\in [0,1]\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{Bernoulli}(y~|~\\theta) = \\left\\{ \\begin{array}{ll} \\theta &\n\\text{if } y = 1, \\text{ and} \\\\ 1 - \\theta & \\text{if } y = 0.\n\\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli(theta)\nIncrement target log probability density with bernoulli_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_lpmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lupmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal bernoulli_cdf(ints y | reals theta) The Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.0\n \n\nreal bernoulli_lcdf(ints y | reals theta) The log of the Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lccdf(ints y | reals theta) The log of the Bernoulli complementary cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nints bernoulli_rng(reals theta) Generate a Bernoulli variate with chance of success theta or an array of Bernoulli variates given an array of thetas of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{BernoulliLogit}(y~|~\\alpha) = \\text{Bernoulli}(y |\n\\text{logit}^{-1}(\\alpha)) = \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha) & \\text{if } y = 1, \\text{ and} \\\\ 1 -\n\\text{logit}^{-1}(\\alpha) &  \\text{if } y = 0. \\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli_logit(alpha)\nIncrement target log probability density with bernoulli_logit_lupmf(y | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_logit_lpmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha)\nAvailable since 2.12\n \n\nreal bernoulli_logit_lupmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha) dropping constant additive terms\nAvailable since 2.25\n \n\nR bernoulli_logit_rng(reals alpha) Generate a Bernoulli variate with chance of success \\(\\text{logit}^{-1}(\\alpha)\\); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e. a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in {\\{0,1\\}}^n\\), \\[\\begin{align*}\n&\\text{BernoulliLogitGLM}(y~|~x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Bernoulli}(y_i~|~\\text{logit}^{-1}(\\alpha_i + x_i\\cdot\n\\beta))\\\\ &= \\prod_{1\\leq i \\leq n} \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j)\n& \\text{if } y_i = 1, \\text{ and} \\\\ 1 - \\text{logit}^{-1}(\\alpha_i +\n\\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) & \\text{if } y_i = 0.\n\\end{array} \\right. \\end{align*}\\]\n\n\n\ny ~ bernoulli_logit_glm(x, alpha, beta)\nIncrement target log probability density with bernoulli_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.25\n \n\n\n\n\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\narray[] int bernoulli_logit_glm_rng(matrix x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29\n \n\narray[] int bernoulli_logit_glm_rng(row_vector x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Binary Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/binary_distributions.html#bernoulli-distribution",
+    "href": "functions-reference/binary_distributions.html#bernoulli-distribution",
+    "title": "Binary Distributions",
+    "section": "",
+    "text": "If \\(\\theta \\in [0,1]\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{Bernoulli}(y~|~\\theta) = \\left\\{ \\begin{array}{ll} \\theta &\n\\text{if } y = 1, \\text{ and} \\\\ 1 - \\theta & \\text{if } y = 0.\n\\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli(theta)\nIncrement target log probability density with bernoulli_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_lpmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lupmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal bernoulli_cdf(ints y | reals theta) The Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.0\n \n\nreal bernoulli_lcdf(ints y | reals theta) The log of the Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lccdf(ints y | reals theta) The log of the Bernoulli complementary cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nints bernoulli_rng(reals theta) Generate a Bernoulli variate with chance of success theta or an array of Bernoulli variates given an array of thetas of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Binary Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/binary_distributions.html#bernoulli-logit-distribution",
+    "href": "functions-reference/binary_distributions.html#bernoulli-logit-distribution",
+    "title": "Binary Distributions",
+    "section": "",
+    "text": "Stan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{BernoulliLogit}(y~|~\\alpha) = \\text{Bernoulli}(y |\n\\text{logit}^{-1}(\\alpha)) = \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha) & \\text{if } y = 1, \\text{ and} \\\\ 1 -\n\\text{logit}^{-1}(\\alpha) &  \\text{if } y = 0. \\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli_logit(alpha)\nIncrement target log probability density with bernoulli_logit_lupmf(y | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_logit_lpmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha)\nAvailable since 2.12\n \n\nreal bernoulli_logit_lupmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha) dropping constant additive terms\nAvailable since 2.25\n \n\nR bernoulli_logit_rng(reals alpha) Generate a Bernoulli variate with chance of success \\(\\text{logit}^{-1}(\\alpha)\\); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Binary Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/binary_distributions.html#bernoulli-logit-glm",
+    "href": "functions-reference/binary_distributions.html#bernoulli-logit-glm",
+    "title": "Binary Distributions",
+    "section": "",
+    "text": "Stan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e. a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in {\\{0,1\\}}^n\\), \\[\\begin{align*}\n&\\text{BernoulliLogitGLM}(y~|~x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Bernoulli}(y_i~|~\\text{logit}^{-1}(\\alpha_i + x_i\\cdot\n\\beta))\\\\ &= \\prod_{1\\leq i \\leq n} \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j)\n& \\text{if } y_i = 1, \\text{ and} \\\\ 1 - \\text{logit}^{-1}(\\alpha_i +\n\\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) & \\text{if } y_i = 0.\n\\end{array} \\right. \\end{align*}\\]\n\n\n\ny ~ bernoulli_logit_glm(x, alpha, beta)\nIncrement target log probability density with bernoulli_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.25\n \n\n\n\n\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\narray[] int bernoulli_logit_glm_rng(matrix x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29\n \n\narray[] int bernoulli_logit_glm_rng(row_vector x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Binary Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html",
+    "href": "functions-reference/bounded_discrete_distributions.html",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "Bounded discrete probability functions have support on \\(\\{ 0, \\ldots,\nN \\}\\) for some upper bound \\(N\\).\n\n\n\n\nSuppose \\(N \\in \\mathbb{N}\\) and \\(\\theta \\in [0,1]\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). \\[\\begin{equation*} \\text{Binomial}(n~|~N,\\theta) = \\binom{N}{n}\n\\theta^n (1 - \\theta)^{N - n}. \\end{equation*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{Binomial}(n~|~N,\\theta) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt] & & {\n} + n \\log \\theta + (N - n) \\log (1 - \\theta), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\theta} \\log \\text{Binomial}(n~|~N,\\theta)\n= \\frac{n}{\\theta} - \\frac{N - n}{1 - \\theta} \\end{equation*}\\]\n\n\n\nn ~ binomial(N, theta)\nIncrement target log probability density with binomial_lupmf(n | N, theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_lpmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lupmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal binomial_cdf(ints n | ints N, reals theta) The binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.0\n \n\nreal binomial_lcdf(ints n | ints N, reals theta) The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lccdf(ints n | ints N, reals theta) The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nR binomial_rng(ints N, reals theta) Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{eqnarray*}\n\\text{BinomialLogit}(n~|~N,\\alpha) & = &\n\\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha)) \\\\[6pt] & = &\n\\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha) \\right)^{n}  \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right)^{N - n}.  \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{BinomialLogit}(n~|~N,\\alpha) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt]   & &\n{ } + n \\log \\text{logit}^{-1}(\\alpha) + (N - n) \\log \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\alpha} \\log\n\\text{BinomialLogit}(n~|~N,\\alpha) =\n\\frac{n}{\\text{logit}^{-1}(-\\alpha)} - \\frac{N -\nn}{\\text{logit}^{-1}(\\alpha)} \\end{equation*}\\]\n\n\n\nn ~ binomial_logit(N, alpha)\nIncrement target log probability density with binomial_logit_lupmf(n | N, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_logit_lpmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha\nAvailable since 2.12\n \n\nreal binomial_logit_lupmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms\nAvailable since 2.25\n\n\n\n\nStan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta \\in \\mathbb{R}^m\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{align*}\n  &\\text{BinomialLogitGLM}(n~|~N, x, \\alpha, \\beta) = \\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha_i + x_i \\cdot \\beta)) \\\\\n  &= \\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{n}  \\left( 1 - \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{N - n}.\n\\end{align*}\\]\n\n\n\nn ~ binomial_logit_glm(N, x, alpha, beta)\nIncrement target log probability density with binomial_logit_glm_lupmf(n | N, x, alpha, beta).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n\n\n\n\n\n\nIf \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in\n\\mathbb{R}^+\\), then for \\(n \\in {0,\\ldots,N}\\), \\[\\begin{equation*}\n\\text{BetaBinomial}(n~|~N,\\alpha,\\beta) = \\binom{N}{n}\n\\frac{\\mathrm{B}(n+\\alpha, N -n +   \\beta)}{\\mathrm{B}(\\alpha,\\beta)},\n\\end{equation*}\\] where the beta function \\(\\mathrm{B}(u,v)\\) is defined for \\(u \\in\n\\mathbb{R}^+\\) and \\(v \\in \\mathbb{R}^+\\) by \\[\\begin{equation*} \\mathrm{B}(u,v) =\n\\frac{\\Gamma(u) \\ \\Gamma(v)}{\\Gamma(u + v)}. \\end{equation*}\\]\n\n\n\nn ~ beta_binomial(N, alpha, beta)\nIncrement target log probability density with beta_binomial_lupmf(n | N, alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_binomial_lpmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lupmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_binomial_cdf(ints n | ints N, reals alpha, reals beta) The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.0\n \n\nreal beta_binomial_lcdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lccdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nR beta_binomial_rng(ints N, reals alpha, reals beta) Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(a \\in \\mathbb{N}\\), \\(b \\in \\mathbb{N}\\), and \\(N \\in\n\\{0,\\ldots,a+b\\}\\), then for \\(n \\in \\{\\max(0,N-b),\\ldots,\\min(a,N)\\}\\), \\[\\begin{equation*} \\text{Hypergeometric}(n~|~N,a,b) = \\frac{\\normalsize{\\binom{a}{n}\n\\binom{b}{N - n}}}      {\\normalsize{\\binom{a + b}{N}}}. \\end{equation*}\\]\n\n\n\nn ~ hypergeometric(N, a, b)\nIncrement target log probability density with hypergeometric_lupmf(n | N, a, b).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal hypergeometric_lpmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b\nAvailable since 2.12\n \n\nreal hypergeometric_lupmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms\nAvailable since 2.25\n \n\nint hypergeometric_rng(int N, int a, int b) Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(N \\in \\mathbb{N}\\), \\(N &gt; 0\\), and if \\(\\theta \\in \\mathbb{R}^N\\) forms an \\(N\\)-simplex (i.e., has nonnegative entries summing to one), then for \\(y \\in \\{1,\\ldots,N\\}\\), \\[\\begin{equation*} \\text{Categorical}(y~|~\\theta) =\n\\theta_y. \\end{equation*}\\] In addition, Stan provides a log-odds scaled categorical distribution, \\[\\begin{equation*} \\text{CategoricalLogit}(y~|~\\beta) =\n\\text{Categorical}(y~|~\\text{softmax}(\\beta)). \\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical(theta)\nIncrement target log probability density with categorical_lupmf(y | theta) dropping constant additive terms.\nAvailable since 2.0\n \n\n\n\n\ny ~ categorical_logit(beta)\nIncrement target log probability density with categorical_logit_lupmf(y | beta).\nAvailable since 2.4\n \n\n\n\n\nAll of the categorical distributions are vectorized so that the outcome y can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal categorical_lpmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.12\n \n\nreal categorical_lupmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.25\n \n\nreal categorical_logit_lpmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta.\nAvailable since 2.12\n \n\nreal categorical_logit_lupmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta dropping constant additive terms.\nAvailable since 2.25\n \n\nint categorical_rng(vector theta) Generate a categorical variate with \\(N\\)-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nint categorical_logit_rng(vector beta) Generate a categorical variate with outcome in range \\(1:N\\) from log-odds vector beta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.16\n\n\n\n\nStan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e. a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.\nNote that the implementation does not put any restrictions on the coefficient matrix \\(\\beta\\). It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the Stan User’s Guide.\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\), \\(N,M,K &gt; 0\\), and if \\(x\\in \\mathbb{R}^{M\\times K}, \\alpha \\in \\mathbb{R}^N, \\beta\\in \\mathbb{R}^{K\\cdot N}\\), then for \\(y \\in \\{1,\\ldots,N\\}^M\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{CategoricalLogitGLM}(y~|~x,\\alpha,\\beta)\n& =  \\prod_{1\\leq i \\leq M}\\text{CategoricalLogit}(y_i~|~\\alpha+x_i\\cdot\\beta)  \\\\[8pt]\n& =  \\prod_{1\\leq i \\leq M}\\text{Categorical}(y_i~|~softmax(\\alpha+x_i\\cdot\\beta)).\n\\end{split}\n\\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical_logit_glm(x, alpha, beta)\nIncrement target log probability density with categorical_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal categorical_logit_glm_lpmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(l, u \\in \\mathbb{Z}\\) are lower and upper bounds (\\(l \\le u\\)), then for any integer \\(y \\in \\{l,\\ldots,u\\}\\), \\[\\begin{equation*} \\text{DiscreteRange}(y ~|~ l, u) =\n\\frac{1}{u - l + 1}. \\end{equation*}\\]\n\n\n\ny ~ discrete_range(l, u)\nIncrement the target log probability density with discrete_range_lupmf(y | l, u) dropping constant additive terms.\nAvailable since 2.26\n \n\n\n\n\nAll of the discrete range distributions are vectorized so that the outcome y and the bounds l, u can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal discrete_range_lpmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\).\nAvailable since 2.26\n \n\nreal discrete_range_lupmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\) dropping constant additive terms.\nAvailable since 2.26\n \n\nreal discrete_range_cdf(ints y | ints l, ints u) The discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lcdf(ints y | ints l, ints u) The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lccdf(ints y | ints l, ints u) The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nints discrete_range_rng(ints l, ints u) Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.26\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedLogistic}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\text{logit}^{-1}(\\eta - c_1)  &  \\text{if } k = 1, \\\\[4pt]\n\\text{logit}^{-1}(\\eta - c_{k-1}) - \\text{logit}^{-1}(\\eta - c_{k})  &\n\\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt] \\text{logit}^{-1}(\\eta -\nc_{K-1}) - 0  &  \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\nk ~ ordered_logistic(eta, c)\nIncrement target log probability density with ordered_logistic_lupmf(k | eta, c).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal ordered_logistic_lpmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_logistic_lupmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nint ordered_logistic_rng(real eta, vector c) Generate an ordered logistic variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n\n\n\n\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\) with \\(N, M &gt; 0\\), \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(x\\in \\mathbb{R}^{N\\times M}, \\beta\\in \\mathbb{R}^M\\), then for \\(y \\in \\{1,\\ldots,K\\}^N\\), \\[\\begin{equation*}\n\\begin{split}\n\\\\\n& \\text{OrderedLogisticGLM}(y~|~x,\\beta,c) \\\\[8pt]\n& = \\prod_{1\\leq i \\leq N}\\text{OrderedLogistic}(y_i~|~x_i\\cdot \\beta,c) \\\\\n& = \\prod_{1\\leq i \\leq N}  \\left\\{ \\begin{array}{ll}\n1 - \\text{logit}^{-1}(x_i\\cdot \\beta - c_1)  &  \\text{if } y = 1, \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{y-1}) - \\text{logit}^{-1}(x_i\\cdot \\beta - c_{y}) & \\text{if } 1 &lt; y &lt; K, \\text{and} \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{K-1}) - 0  &  \\text{if } y = K.\n\\end{array} \\right.\n\\end{split}\n\\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(y=1\\) and \\(y=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\ny ~ ordered_logistic_glm(x, beta, c)\nIncrement target log probability density with ordered_logistic_lupmf(y | x, beta, c).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal ordered_logistic_glm_lpmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedProbit}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\Phi(\\eta - c_1) & \\text{if } k = 1, \\\\[4pt] \\Phi(\\eta - c_{k-1}) -\n\\Phi(\\eta - c_{k})  & \\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt]\n\\Phi(\\eta - c_{K-1}) - 0 & \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 =\n-\\infty\\) and \\(c_K = +\\infty\\) with \\(\\Phi(-\\infty) = 0\\) and \\(\\Phi(\\infty) = 1\\).\n\n\n\nk ~ ordered_probit(eta, c)\nIncrement target log probability density with ordered_probit_lupmf(k | eta, c).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal ordered_probit_lpmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_probit_lupmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nreal ordered_probit_lpmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.\nAvailable since 2.19\n \n\nreal ordered_probit_lupmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.19\n \n\nint ordered_probit_rng(real eta, vector c) Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#binomial-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "Suppose \\(N \\in \\mathbb{N}\\) and \\(\\theta \\in [0,1]\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). \\[\\begin{equation*} \\text{Binomial}(n~|~N,\\theta) = \\binom{N}{n}\n\\theta^n (1 - \\theta)^{N - n}. \\end{equation*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{Binomial}(n~|~N,\\theta) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt] & & {\n} + n \\log \\theta + (N - n) \\log (1 - \\theta), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\theta} \\log \\text{Binomial}(n~|~N,\\theta)\n= \\frac{n}{\\theta} - \\frac{N - n}{1 - \\theta} \\end{equation*}\\]\n\n\n\nn ~ binomial(N, theta)\nIncrement target log probability density with binomial_lupmf(n | N, theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_lpmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lupmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal binomial_cdf(ints n | ints N, reals theta) The binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.0\n \n\nreal binomial_lcdf(ints n | ints N, reals theta) The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lccdf(ints n | ints N, reals theta) The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nR binomial_rng(ints N, reals theta) Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-distribution-logit-parameterization",
+    "href": "functions-reference/bounded_discrete_distributions.html#binomial-distribution-logit-parameterization",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{eqnarray*}\n\\text{BinomialLogit}(n~|~N,\\alpha) & = &\n\\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha)) \\\\[6pt] & = &\n\\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha) \\right)^{n}  \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right)^{N - n}.  \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{BinomialLogit}(n~|~N,\\alpha) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt]   & &\n{ } + n \\log \\text{logit}^{-1}(\\alpha) + (N - n) \\log \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\alpha} \\log\n\\text{BinomialLogit}(n~|~N,\\alpha) =\n\\frac{n}{\\text{logit}^{-1}(-\\alpha)} - \\frac{N -\nn}{\\text{logit}^{-1}(\\alpha)} \\end{equation*}\\]\n\n\n\nn ~ binomial_logit(N, alpha)\nIncrement target log probability density with binomial_logit_lupmf(n | N, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_logit_lpmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha\nAvailable since 2.12\n \n\nreal binomial_logit_lupmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-logit-glm",
+    "href": "functions-reference/bounded_discrete_distributions.html#binomial-logit-glm",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta \\in \\mathbb{R}^m\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{align*}\n  &\\text{BinomialLogitGLM}(n~|~N, x, \\alpha, \\beta) = \\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha_i + x_i \\cdot \\beta)) \\\\\n  &= \\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{n}  \\left( 1 - \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{N - n}.\n\\end{align*}\\]\n\n\n\nn ~ binomial_logit_glm(N, x, alpha, beta)\nIncrement target log probability density with binomial_logit_glm_lupmf(n | N, x, alpha, beta).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#beta-binomial-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#beta-binomial-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in\n\\mathbb{R}^+\\), then for \\(n \\in {0,\\ldots,N}\\), \\[\\begin{equation*}\n\\text{BetaBinomial}(n~|~N,\\alpha,\\beta) = \\binom{N}{n}\n\\frac{\\mathrm{B}(n+\\alpha, N -n +   \\beta)}{\\mathrm{B}(\\alpha,\\beta)},\n\\end{equation*}\\] where the beta function \\(\\mathrm{B}(u,v)\\) is defined for \\(u \\in\n\\mathbb{R}^+\\) and \\(v \\in \\mathbb{R}^+\\) by \\[\\begin{equation*} \\mathrm{B}(u,v) =\n\\frac{\\Gamma(u) \\ \\Gamma(v)}{\\Gamma(u + v)}. \\end{equation*}\\]\n\n\n\nn ~ beta_binomial(N, alpha, beta)\nIncrement target log probability density with beta_binomial_lupmf(n | N, alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_binomial_lpmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lupmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_binomial_cdf(ints n | ints N, reals alpha, reals beta) The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.0\n \n\nreal beta_binomial_lcdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lccdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nR beta_binomial_rng(ints N, reals alpha, reals beta) Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#hypergeometric-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#hypergeometric-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(a \\in \\mathbb{N}\\), \\(b \\in \\mathbb{N}\\), and \\(N \\in\n\\{0,\\ldots,a+b\\}\\), then for \\(n \\in \\{\\max(0,N-b),\\ldots,\\min(a,N)\\}\\), \\[\\begin{equation*} \\text{Hypergeometric}(n~|~N,a,b) = \\frac{\\normalsize{\\binom{a}{n}\n\\binom{b}{N - n}}}      {\\normalsize{\\binom{a + b}{N}}}. \\end{equation*}\\]\n\n\n\nn ~ hypergeometric(N, a, b)\nIncrement target log probability density with hypergeometric_lupmf(n | N, a, b).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal hypergeometric_lpmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b\nAvailable since 2.12\n \n\nreal hypergeometric_lupmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms\nAvailable since 2.25\n \n\nint hypergeometric_rng(int N, int a, int b) Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#categorical-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#categorical-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(N \\in \\mathbb{N}\\), \\(N &gt; 0\\), and if \\(\\theta \\in \\mathbb{R}^N\\) forms an \\(N\\)-simplex (i.e., has nonnegative entries summing to one), then for \\(y \\in \\{1,\\ldots,N\\}\\), \\[\\begin{equation*} \\text{Categorical}(y~|~\\theta) =\n\\theta_y. \\end{equation*}\\] In addition, Stan provides a log-odds scaled categorical distribution, \\[\\begin{equation*} \\text{CategoricalLogit}(y~|~\\beta) =\n\\text{Categorical}(y~|~\\text{softmax}(\\beta)). \\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical(theta)\nIncrement target log probability density with categorical_lupmf(y | theta) dropping constant additive terms.\nAvailable since 2.0\n \n\n\n\n\ny ~ categorical_logit(beta)\nIncrement target log probability density with categorical_logit_lupmf(y | beta).\nAvailable since 2.4\n \n\n\n\n\nAll of the categorical distributions are vectorized so that the outcome y can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal categorical_lpmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.12\n \n\nreal categorical_lupmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.25\n \n\nreal categorical_logit_lpmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta.\nAvailable since 2.12\n \n\nreal categorical_logit_lupmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta dropping constant additive terms.\nAvailable since 2.25\n \n\nint categorical_rng(vector theta) Generate a categorical variate with \\(N\\)-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nint categorical_logit_rng(vector beta) Generate a categorical variate with outcome in range \\(1:N\\) from log-odds vector beta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.16",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#categorical-logit-glm",
+    "href": "functions-reference/bounded_discrete_distributions.html#categorical-logit-glm",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e. a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.\nNote that the implementation does not put any restrictions on the coefficient matrix \\(\\beta\\). It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the Stan User’s Guide.\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\), \\(N,M,K &gt; 0\\), and if \\(x\\in \\mathbb{R}^{M\\times K}, \\alpha \\in \\mathbb{R}^N, \\beta\\in \\mathbb{R}^{K\\cdot N}\\), then for \\(y \\in \\{1,\\ldots,N\\}^M\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{CategoricalLogitGLM}(y~|~x,\\alpha,\\beta)\n& =  \\prod_{1\\leq i \\leq M}\\text{CategoricalLogit}(y_i~|~\\alpha+x_i\\cdot\\beta)  \\\\[8pt]\n& =  \\prod_{1\\leq i \\leq M}\\text{Categorical}(y_i~|~softmax(\\alpha+x_i\\cdot\\beta)).\n\\end{split}\n\\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical_logit_glm(x, alpha, beta)\nIncrement target log probability density with categorical_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal categorical_logit_glm_lpmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#discrete-range-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#discrete-range-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(l, u \\in \\mathbb{Z}\\) are lower and upper bounds (\\(l \\le u\\)), then for any integer \\(y \\in \\{l,\\ldots,u\\}\\), \\[\\begin{equation*} \\text{DiscreteRange}(y ~|~ l, u) =\n\\frac{1}{u - l + 1}. \\end{equation*}\\]\n\n\n\ny ~ discrete_range(l, u)\nIncrement the target log probability density with discrete_range_lupmf(y | l, u) dropping constant additive terms.\nAvailable since 2.26\n \n\n\n\n\nAll of the discrete range distributions are vectorized so that the outcome y and the bounds l, u can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal discrete_range_lpmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\).\nAvailable since 2.26\n \n\nreal discrete_range_lupmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\) dropping constant additive terms.\nAvailable since 2.26\n \n\nreal discrete_range_cdf(ints y | ints l, ints u) The discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lcdf(ints y | ints l, ints u) The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lccdf(ints y | ints l, ints u) The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nints discrete_range_rng(ints l, ints u) Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.26",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedLogistic}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\text{logit}^{-1}(\\eta - c_1)  &  \\text{if } k = 1, \\\\[4pt]\n\\text{logit}^{-1}(\\eta - c_{k-1}) - \\text{logit}^{-1}(\\eta - c_{k})  &\n\\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt] \\text{logit}^{-1}(\\eta -\nc_{K-1}) - 0  &  \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\nk ~ ordered_logistic(eta, c)\nIncrement target log probability density with ordered_logistic_lupmf(k | eta, c).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal ordered_logistic_lpmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_logistic_lupmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nint ordered_logistic_rng(real eta, vector c) Generate an ordered logistic variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-generalized-linear-model-ordinal-regression",
+    "href": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-generalized-linear-model-ordinal-regression",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(N,M,K \\in \\mathbb{N}\\) with \\(N, M &gt; 0\\), \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(x\\in \\mathbb{R}^{N\\times M}, \\beta\\in \\mathbb{R}^M\\), then for \\(y \\in \\{1,\\ldots,K\\}^N\\), \\[\\begin{equation*}\n\\begin{split}\n\\\\\n& \\text{OrderedLogisticGLM}(y~|~x,\\beta,c) \\\\[8pt]\n& = \\prod_{1\\leq i \\leq N}\\text{OrderedLogistic}(y_i~|~x_i\\cdot \\beta,c) \\\\\n& = \\prod_{1\\leq i \\leq N}  \\left\\{ \\begin{array}{ll}\n1 - \\text{logit}^{-1}(x_i\\cdot \\beta - c_1)  &  \\text{if } y = 1, \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{y-1}) - \\text{logit}^{-1}(x_i\\cdot \\beta - c_{y}) & \\text{if } 1 &lt; y &lt; K, \\text{and} \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{K-1}) - 0  &  \\text{if } y = K.\n\\end{array} \\right.\n\\end{split}\n\\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(y=1\\) and \\(y=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\ny ~ ordered_logistic_glm(x, beta, c)\nIncrement target log probability density with ordered_logistic_lupmf(y | x, beta, c).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal ordered_logistic_glm_lpmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-probit-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#ordered-probit-distribution",
+    "title": "Bounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedProbit}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\Phi(\\eta - c_1) & \\text{if } k = 1, \\\\[4pt] \\Phi(\\eta - c_{k-1}) -\n\\Phi(\\eta - c_{k})  & \\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt]\n\\Phi(\\eta - c_{K-1}) - 0 & \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 =\n-\\infty\\) and \\(c_K = +\\infty\\) with \\(\\Phi(-\\infty) = 0\\) and \\(\\Phi(\\infty) = 1\\).\n\n\n\nk ~ ordered_probit(eta, c)\nIncrement target log probability density with ordered_probit_lupmf(k | eta, c).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal ordered_probit_lpmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_probit_lupmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nreal ordered_probit_lpmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.\nAvailable since 2.19\n \n\nreal ordered_probit_lupmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.19\n \n\nint ordered_probit_rng(real eta, vector c) Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html",
+    "href": "functions-reference/complex-valued_basic_functions.html",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "This chapter describes built-in functions that operate on complex numbers, either as an argument type or a return type. This includes the arithmetic operators generalized to complex numbers.\n\n\nJust as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.\nint n = 5;       // n = 5\nreal x = a;      // x = 5.0\ncomplex z1 = n;  // z = 5.0 + 0.0i\ncomplex z2 = x;  // z = 5.0 + 0.0i\n\n\nFunction arguments of type int or real may be promoted to type complex. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if z is complex, then pow(z, 2) will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where z + 2 will be legal and produce a complex result. Functions such as arg and conj that are only available for complex numbers can accept integer or real arguments, promoting them to complex before applying the function.\n\n\n\n\n\n\nVariables and constants of type complex are constructed from zero, one, or two real numbers.\ncomplex z1 = to_complex();        // z1 = 0.0 + 0.0i\nreal re = -2.9;\ncomplex z2 = to_complex(re);      // z2 = -2.9 + 0.0i\nreal im = 1.3;\ncomplex z3 = to_complex(re, im);  // z3 = -2.9 + 1.3i\n \n\ncomplex to_complex() Return complex number with real part 0.0 and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re) Return complex number with real part re and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re, real im) Return complex number with real part re and imaginary part im.\nAvailable since 2.28\n \n\nZ to_complex(T1 re, T2 im) Vectorized implementation of the to_complex function.\nT1 and T2 can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.\nAvailable since 2.30\n\n\n\nGiven a complex number, its real and imaginary parts can be extracted with the following functions.\n \n\nreal get_real(complex z) Return the real part of the complex number z.\nAvailable since 2.28\n \n\nreal get_imag(complex z) Return the imaginary part of the complex number z.\nAvailable since 2.28\n\n\n\n\nThe arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type complex. If there are two arguments and only one is of type complex, then the other will be promoted to type complex before performing the operation.\n\n\n \n\ncomplex operator+(complex z) Return the complex argument z, \\[\\begin{equation*} +z = z. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex z) Return the negation of the complex argument z, which for \\(z = x + yi\\) is \\[\\begin{equation*} -z = -x - yi. \\end{equation*}\\]\nAvailable since 2.28\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of complex numbers, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex operator+(complex x, complex y) Return the sum of x and y, \\[\\begin{equation*} (x + y) = \\text{operator+}(x, y) = x + y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex x, complex y) Return the difference between x and y, \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x, y) = x - y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator*(complex x, complex y) Return the product of x and y, \\[\\begin{equation*} (x \\, * \\, y) = \\text{operator*}(x, y) = x\n\\times y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator/(complex x, complex y) Return the quotient of x and y, \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator^(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n(x^\\mathrm{\\wedge}y)= \\text{operator}^\\mathrm{\\wedge}(x,y)\n= \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\n\nComplex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional\nz1 == z2\nis equivalent to\nget_real(z1) == get_real(z2) && get_imag(z1) == get_imag(z2)\nAs with other complex functions, if one of the arguments is of type real or int, it will be promoted to type complex before comparison. For example, if z is of type complex, then z == 0 will be true if z has real component equal to 0.0 and complex component equal to 0.0.\nWarning: As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.\n \n\nint operator==(complex x, complex y) Return 1 if x is equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{==}\\, y)\n\\ = \\ \\text{operator==}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x = y$}, \\ \\text{and} \\\\ 0 & \\text{otherwise.}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nint operator!=(complex x, complex y) Return 1 if x is not equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{!=}\\, y)\n\\ = \\ \\text{operator!=}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x \\neq y$}, \\ \\text{and} \\\\ 0 &\n\\text{otherwise.} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\nThe assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.\nAssignment of complex numbers works elementwise. If an expression of type int or real is assigned to a complex number, it will be promoted before assignment as if calling to_complex(), so that the imaginary component is 0.0.\n \n\nvoid operator=(complex x, complex y) y = x; assigns a (copy of) the value of y to x.\nAvailable since 2.28\n \n\nvoid operator+=(complex x, complex y) x += y; is equivalent to x = x + y;.\nAvailable since 2.28\n \n\nvoid operator-=(complex x, complex y) x -= y; is equivalent to x = x - y;.\nAvailable since 2.28\n \n\nvoid operator*=(complex x, complex y) x *= y; is equivalent to x = x * y;.\nAvailable since 2.28\n \n\nvoid operator/=(complex x, complex y) x /= y; is equivalent to x = x / y;.\nAvailable since 2.28\n\n\n\nThe following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.\n \n\nreal abs(complex z) Return the absolute value of z, also known as the modulus or magnitude, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{abs}(z) = \\sqrt{x^2 + y^2}.\n\\end{equation*}\\]\nThis function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a complex_vector[n] input will return a vector[n] output, with each element transformed by the above equation.\nAvailable since 2.28, vectorized in 2.30\n \n\nreal arg(complex z) Return the phase angle (in radians) of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{arg}(z) = \\textrm{atan2}(y, x) = \\textrm{atan}(y / x).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nreal norm(complex z) Return the Euclidean norm of z, which is its absolute value squared, and which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{norm}(z) = \\textrm{abs}^2(z) = x^2 + y^2.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex conj(complex z) Return the complex conjugate of z, which negates the imaginary component, so that if \\(z = x + yi\\), \\[\\begin{equation*}\n\\textrm{conj}(z) = x - yi.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ conj(Z z) Vectorized version of conj. This will apply the conj function to each element of a complex array, vector, or matrix.\nAvailable since 2.31\n \n\ncomplex proj(complex z) Return the projection of z onto the Riemann sphere, which for \\(z = x\n+ yi\\) is \\[\\begin{equation*}\n\\textrm{proj}(z)\n= \\begin{cases}\n    z & \\textrm{if} \\ z \\ \\textrm{is finite, and} \\\\\n    0 + \\textrm{sign}(y)i & \\textrm{otherwise,}\n\\end{cases}\n\\end{equation*}\\] where \\(\\textrm{sign}(y)\\) is -1 if \\(y\\) is negative and 1 otherwise.\nAvailable since 2.28\n \n\ncomplex polar(real r, real theta) Return the complex number with magnitude (absolute value) r and phase angle theta.\nAvailable since 2.28\n\n\n\nThe exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.\n \n\ncomplex exp(complex z) Return the complex natural exponential of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\exp z = \\exp(x) \\textrm{cis}(y) = \\exp(x) (\\cos(y) + i \\sin(y)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log(complex z) Return the complex natural logarithm of z, which for \\(z = \\textrm{polar}(r,\n\\theta)\\) is \\[\\begin{equation*}\n\\log z = \\log r + \\theta i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log10(complex z) Return the complex common logarithm of z, \\[\\begin{equation*}\n\\log_{10} z = \\frac{\\log z}{\\log 10}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex pow(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n\\text{pow}(x,y) = \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.30\n \n\ncomplex sqrt(complex x) Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.\nAvailable since 2.28\n\n\n\nThe standard trigonometric functions are supported for complex numbers.\n \n\ncomplex cos(complex z) Return the complex cosine of z, which is \\[\\begin{equation*}\n\\cos(z)\n= \\textrm{cosh}(z \\, i)\n= \\frac{\\displaystyle \\exp(z \\, i) + \\exp(-z \\, i)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sin(complex z) Return the complex sine of z, \\[\\begin{equation*}\n\\sin(z)\n= -\\textrm{sinh}(z \\, i) \\, i\n= \\frac{\\displaystyle \\exp(z \\, i) - \\exp(-z \\, i)}\n       {\\displaystyle 2 \\, i}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tan(complex z) Return the complex tangent of z, \\[\\begin{equation*}\n\\tan(z)\n= -\\textrm{tanh}(z \\, i) \\, i\n= \\frac{(\\exp(-z \\, i) - \\exp(z \\, i)) \\, i}\n       {\\exp(-z \\, i) + \\exp(z \\, i)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acos(complex z) Return the complex arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acos}(z)\n= \\frac{1}{2} \\pi + \\log (z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asin(complex z) Return the complex arc (inverse) sine of z, \\[\\begin{equation*}\n\\text{asin}(z)\n= -\\log(z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atan(complex z) Return the complex arc (inverse) tangent of z, \\[\\begin{equation*}\n\\text{atan}(z)\n= - \\frac{1}{2} (\\log(1 - z \\, i) - \\log(1 + z \\, i)) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\nThe standard hyperbolic trigonometric functions are supported for complex numbers.\n \n\ncomplex cosh(complex z) Return the complex hyperbolic cosine of z, \\[\\begin{equation*}\n\\textrm{cosh}(z)\n= \\frac{\\exp(z) + \\exp(-z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sinh(complex z) Return the complex hyperbolic sine of z, \\[\\begin{equation*}\n\\textrm{sinh}(z)\n= \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tanh(complex z) Return the complex hyperbolic tangent of z, \\[\\begin{equation*}\n\\textrm{tanh}(z)\n\\ = \\ \\frac{\\textrm{sinh}(z)}\n           {\\textrm{cosh}(z)}\n\\ = \\ \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n           {\\displaystyle \\exp(z) + \\exp(-z)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acosh(complex z) Return the complex hyperbolic arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acosh}(z)\n= \\log(z + \\sqrt{(z + 1)(z - 1)}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asinh(complex z) Return the complex hyperbolic arc (inverse) sine of z, \\[\\begin{equation*}\n\\textrm{asinh}(z)\n= \\log(z + \\sqrt{1 + z^2}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atanh(complex z) Return the complex hyperbolic arc (inverse) tangent of z, \\[\\begin{equation*}\n\\textrm{atanh}(z)\n= \\frac{\\log(1 + z) - \\log(1 - z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-assign-promote",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-assign-promote",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "Just as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.\nint n = 5;       // n = 5\nreal x = a;      // x = 5.0\ncomplex z1 = n;  // z = 5.0 + 0.0i\ncomplex z2 = x;  // z = 5.0 + 0.0i\n\n\nFunction arguments of type int or real may be promoted to type complex. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if z is complex, then pow(z, 2) will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where z + 2 will be legal and produce a complex result. Functions such as arg and conj that are only available for complex numbers can accept integer or real arguments, promoting them to complex before applying the function.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-constructors",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-constructors",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "Variables and constants of type complex are constructed from zero, one, or two real numbers.\ncomplex z1 = to_complex();        // z1 = 0.0 + 0.0i\nreal re = -2.9;\ncomplex z2 = to_complex(re);      // z2 = -2.9 + 0.0i\nreal im = 1.3;\ncomplex z3 = to_complex(re, im);  // z3 = -2.9 + 1.3i\n \n\ncomplex to_complex() Return complex number with real part 0.0 and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re) Return complex number with real part re and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re, real im) Return complex number with real part re and imaginary part im.\nAvailable since 2.28\n \n\nZ to_complex(T1 re, T2 im) Vectorized implementation of the to_complex function.\nT1 and T2 can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.\nAvailable since 2.30\n\n\n\nGiven a complex number, its real and imaginary parts can be extracted with the following functions.\n \n\nreal get_real(complex z) Return the real part of the complex number z.\nAvailable since 2.28\n \n\nreal get_imag(complex z) Return the imaginary part of the complex number z.\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-arithmetic",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-arithmetic",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "The arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type complex. If there are two arguments and only one is of type complex, then the other will be promoted to type complex before performing the operation.\n\n\n \n\ncomplex operator+(complex z) Return the complex argument z, \\[\\begin{equation*} +z = z. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex z) Return the negation of the complex argument z, which for \\(z = x + yi\\) is \\[\\begin{equation*} -z = -x - yi. \\end{equation*}\\]\nAvailable since 2.28\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of complex numbers, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex operator+(complex x, complex y) Return the sum of x and y, \\[\\begin{equation*} (x + y) = \\text{operator+}(x, y) = x + y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex x, complex y) Return the difference between x and y, \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x, y) = x - y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator*(complex x, complex y) Return the product of x and y, \\[\\begin{equation*} (x \\, * \\, y) = \\text{operator*}(x, y) = x\n\\times y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator/(complex x, complex y) Return the quotient of x and y, \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator^(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n(x^\\mathrm{\\wedge}y)= \\text{operator}^\\mathrm{\\wedge}(x,y)\n= \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-comparison",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-comparison",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "Complex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional\nz1 == z2\nis equivalent to\nget_real(z1) == get_real(z2) && get_imag(z1) == get_imag(z2)\nAs with other complex functions, if one of the arguments is of type real or int, it will be promoted to type complex before comparison. For example, if z is of type complex, then z == 0 will be true if z has real component equal to 0.0 and complex component equal to 0.0.\nWarning: As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.\n \n\nint operator==(complex x, complex y) Return 1 if x is equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{==}\\, y)\n\\ = \\ \\text{operator==}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x = y$}, \\ \\text{and} \\\\ 0 & \\text{otherwise.}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nint operator!=(complex x, complex y) Return 1 if x is not equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{!=}\\, y)\n\\ = \\ \\text{operator!=}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x \\neq y$}, \\ \\text{and} \\\\ 0 &\n\\text{otherwise.} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-assignment",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-assignment",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "The assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.\nAssignment of complex numbers works elementwise. If an expression of type int or real is assigned to a complex number, it will be promoted before assignment as if calling to_complex(), so that the imaginary component is 0.0.\n \n\nvoid operator=(complex x, complex y) y = x; assigns a (copy of) the value of y to x.\nAvailable since 2.28\n \n\nvoid operator+=(complex x, complex y) x += y; is equivalent to x = x + y;.\nAvailable since 2.28\n \n\nvoid operator-=(complex x, complex y) x -= y; is equivalent to x = x - y;.\nAvailable since 2.28\n \n\nvoid operator*=(complex x, complex y) x *= y; is equivalent to x = x * y;.\nAvailable since 2.28\n \n\nvoid operator/=(complex x, complex y) x /= y; is equivalent to x = x / y;.\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-special",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-special",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "The following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.\n \n\nreal abs(complex z) Return the absolute value of z, also known as the modulus or magnitude, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{abs}(z) = \\sqrt{x^2 + y^2}.\n\\end{equation*}\\]\nThis function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a complex_vector[n] input will return a vector[n] output, with each element transformed by the above equation.\nAvailable since 2.28, vectorized in 2.30\n \n\nreal arg(complex z) Return the phase angle (in radians) of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{arg}(z) = \\textrm{atan2}(y, x) = \\textrm{atan}(y / x).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nreal norm(complex z) Return the Euclidean norm of z, which is its absolute value squared, and which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{norm}(z) = \\textrm{abs}^2(z) = x^2 + y^2.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex conj(complex z) Return the complex conjugate of z, which negates the imaginary component, so that if \\(z = x + yi\\), \\[\\begin{equation*}\n\\textrm{conj}(z) = x - yi.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ conj(Z z) Vectorized version of conj. This will apply the conj function to each element of a complex array, vector, or matrix.\nAvailable since 2.31\n \n\ncomplex proj(complex z) Return the projection of z onto the Riemann sphere, which for \\(z = x\n+ yi\\) is \\[\\begin{equation*}\n\\textrm{proj}(z)\n= \\begin{cases}\n    z & \\textrm{if} \\ z \\ \\textrm{is finite, and} \\\\\n    0 + \\textrm{sign}(y)i & \\textrm{otherwise,}\n\\end{cases}\n\\end{equation*}\\] where \\(\\textrm{sign}(y)\\) is -1 if \\(y\\) is negative and 1 otherwise.\nAvailable since 2.28\n \n\ncomplex polar(real r, real theta) Return the complex number with magnitude (absolute value) r and phase angle theta.\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-exponential",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-exponential",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "The exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.\n \n\ncomplex exp(complex z) Return the complex natural exponential of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\exp z = \\exp(x) \\textrm{cis}(y) = \\exp(x) (\\cos(y) + i \\sin(y)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log(complex z) Return the complex natural logarithm of z, which for \\(z = \\textrm{polar}(r,\n\\theta)\\) is \\[\\begin{equation*}\n\\log z = \\log r + \\theta i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log10(complex z) Return the complex common logarithm of z, \\[\\begin{equation*}\n\\log_{10} z = \\frac{\\log z}{\\log 10}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex pow(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n\\text{pow}(x,y) = \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.30\n \n\ncomplex sqrt(complex x) Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-trig",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-trig",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "The standard trigonometric functions are supported for complex numbers.\n \n\ncomplex cos(complex z) Return the complex cosine of z, which is \\[\\begin{equation*}\n\\cos(z)\n= \\textrm{cosh}(z \\, i)\n= \\frac{\\displaystyle \\exp(z \\, i) + \\exp(-z \\, i)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sin(complex z) Return the complex sine of z, \\[\\begin{equation*}\n\\sin(z)\n= -\\textrm{sinh}(z \\, i) \\, i\n= \\frac{\\displaystyle \\exp(z \\, i) - \\exp(-z \\, i)}\n       {\\displaystyle 2 \\, i}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tan(complex z) Return the complex tangent of z, \\[\\begin{equation*}\n\\tan(z)\n= -\\textrm{tanh}(z \\, i) \\, i\n= \\frac{(\\exp(-z \\, i) - \\exp(z \\, i)) \\, i}\n       {\\exp(-z \\, i) + \\exp(z \\, i)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acos(complex z) Return the complex arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acos}(z)\n= \\frac{1}{2} \\pi + \\log (z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asin(complex z) Return the complex arc (inverse) sine of z, \\[\\begin{equation*}\n\\text{asin}(z)\n= -\\log(z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atan(complex z) Return the complex arc (inverse) tangent of z, \\[\\begin{equation*}\n\\text{atan}(z)\n= - \\frac{1}{2} (\\log(1 - z \\, i) - \\log(1 + z \\, i)) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-h-trig",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-h-trig",
+    "title": "Complex-Valued Basic Functions",
+    "section": "",
+    "text": "The standard hyperbolic trigonometric functions are supported for complex numbers.\n \n\ncomplex cosh(complex z) Return the complex hyperbolic cosine of z, \\[\\begin{equation*}\n\\textrm{cosh}(z)\n= \\frac{\\exp(z) + \\exp(-z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sinh(complex z) Return the complex hyperbolic sine of z, \\[\\begin{equation*}\n\\textrm{sinh}(z)\n= \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tanh(complex z) Return the complex hyperbolic tangent of z, \\[\\begin{equation*}\n\\textrm{tanh}(z)\n\\ = \\ \\frac{\\textrm{sinh}(z)}\n           {\\textrm{cosh}(z)}\n\\ = \\ \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n           {\\displaystyle \\exp(z) + \\exp(-z)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acosh(complex z) Return the complex hyperbolic arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acosh}(z)\n= \\log(z + \\sqrt{(z + 1)(z - 1)}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asinh(complex z) Return the complex hyperbolic arc (inverse) sine of z, \\[\\begin{equation*}\n\\textrm{asinh}(z)\n= \\log(z + \\sqrt{1 + z^2}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atanh(complex z) Return the complex hyperbolic arc (inverse) tangent of z, \\[\\begin{equation*}\n\\textrm{atanh}(z)\n= \\frac{\\log(1 + z) - \\log(1 - z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound arithmetic and assignment statements combine an arithmetic operation and assignment, replacing a statement such as\n x = x op y;\nwith the more compact compound form\n x op= y;\nFor example, x = x + 1; may be replaced with x += 1;. This works for all types that support arithmetic, including the scalar types int, real, complex, the real matrix types vector, row_vector, and matrix, and the complex matrix types, complex_vector, complex_row_vector, and complex_matrix.\n\n\nCompound addition and assignment works wherever the corresponding addition and assignment would be well formed.\n \n\nvoid operator+=(T x, U y) x += y is equivalent to x = x + y. Defined for all types T and U where T = T + U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.\n \n\nvoid operator-=(T x, U y) x -= y is equivalent to x = x - y. Defined for all types T and U where T = T - U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator*=(T x, U y) x *= y is equivalent to x = x * y. Defined for all types T and U where T = T * U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator/=(T x, U y) x /= y is equivalent to x = x / y. Defined for all types T and U where T = T / U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator.*=(T x, U y) x .*= y is equivalent to x = x .* y. Defined for all types T and U where T = T .* U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator./=(T x, U y) x ./= y is equivalent to x = x ./ y. Defined for all types T and U where T = T ./ U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-addition-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-addition-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound addition and assignment works wherever the corresponding addition and assignment would be well formed.\n \n\nvoid operator+=(T x, U y) x += y is equivalent to x = x + y. Defined for all types T and U where T = T + U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-subtraction-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-subtraction-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.\n \n\nvoid operator-=(T x, U y) x -= y is equivalent to x = x - y. Defined for all types T and U where T = T - U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-multiplication-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-multiplication-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator*=(T x, U y) x *= y is equivalent to x = x * y. Defined for all types T and U where T = T * U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-division-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-division-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator/=(T x, U y) x /= y is equivalent to x = x / y. Defined for all types T and U where T = T / U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-multiplication-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-multiplication-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator.*=(T x, U y) x .*= y is equivalent to x = x .* y. Defined for all types T and U where T = T .* U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-division-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-division-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
+    "section": "",
+    "text": "Compound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator./=(T x, U y) x ./= y is equivalent to x = x ./ y. Defined for all types T and U where T = T ./ U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html",
+    "href": "functions-reference/conventions_for_probability_functions.html",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "Functions associated with distributions are set up to follow the same naming conventions for both built-in distributions and for user-defined distributions.\n\n\nThe suffix is determined by the type of function according to the following table.\n\n\n\n\n\n\n\n\nfunction\noutcome\nsuffix\n\n\n\n\nlog probability mass function\ndiscrete\n_lpmf\n\n\nlog probability density function\ncontinuous\n_lpdf\n\n\nlog cumulative distribution function\nany\n_lcdf\n\n\nlog complementary cumulative distribution function\nany\n_lccdf\n\n\nrandom number generator\nany\n_rng\n\n\n\nFor example, normal_lpdf is the log of the normal probability density function (pdf) and bernoulli_lpmf is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, normal_lcdf and bernoulli_lcdf.\n\n\n\nEach probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, normal_lpdf(y | mu, sigma) returns the value of mathematical formula \\(\\log \\text{Normal}(y \\, | \\, \\mu, \\sigma)\\). Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., normal_lcdf(y_low | mu, sigma)\n\n\n\nThe notation\n y ~ normal(mu, sigma);\nprovides the same (proportional) contribution to the model log density as the explicit target density increment,\n target += normal_lpdf(y | mu, sigma);\nIn both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (~) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.\n\n\n\nAll of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.\n\n\n\nMany distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter \\(\\sigma &gt; 0\\). If \\(\\sigma = 0\\), the probability function will throw an exception.\nThis is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.\n\n\n\nFor most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix _rng. For example, the function normal_rng(real, real) accepts two real arguments, an unconstrained location \\(\\mu\\) and positive scale \\(\\sigma &gt; 0\\), and returns an unconstrained pseudorandom value drawn from \\(\\text{Normal}(\\mu,\\sigma)\\). There are also vectorized forms of random number generators which return more than one random variate at a time.\n\n\nUnlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.\n\n\n\nUnlike the probability functions, only some of the PRNG functions are vectorized.\n\n\n\n\nFor most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.\nFor a univariate random variable \\(Y\\) with probability function \\(p_Y(y \\, | \\, \\theta)\\), the cumulative distribution function (CDF) \\(F_Y\\) is defined by \\[\\begin{equation*}\nF_Y(y) \\ = \\ \\text{Pr}[Y \\le y] \\ = \\ \\int_{-\\infty}^y p(y\\, | \\, \\theta) \\ \\text{d}y.\n\\end{equation*}\\] The complementary cumulative distribution function (CCDF) is defined as \\[\\begin{equation*}\n\\text{Pr}[Y &gt; y] \\ = \\ 1 - F_Y(y).\n\\end{equation*}\\] The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly \\(10^{-300}\\)), but not numbers very close to 1 (the closest you can get is roughly \\(1 - 10^{-15}\\)).\nIn Stan, there is a cumulative distribution function for each probability function. For instance, normal_cdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\ \\text{d}y.\n\\end{equation*}\\] There are also log forms of the CDF and CCDF for most univariate distributions. For example, normal_lcdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right)\n\\end{equation*}\\] and normal_lccdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( 1 - \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right).\n\\end{equation*}\\]\n\n\n\nStan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section vectorized PRNG functions for more details.\nIn all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.\nStan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.\nStan also overloads some scalar functions, such as log and exp, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.\n\n\n\n\nThe normal probability function is specified with the signature\n normal_lpdf(reals | reals, reals);\nThe pseudotype reals is used to indicate that an argument position may be vectorized. Argument positions declared as reals may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use normal_lpdf(row_vector | vector, real) as long as the vector and row vector have the same size.\n\n\n\nThe multivariate normal distribution accepting vector or array of vector arguments is written as\n multi_normal_lpdf(vectors | vectors, matrix);\nThese arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.\n\n\n\nThe pseudotype ints is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.\n\n\n\n\nThe result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely real or int, is repeated. For instance, if y is a vector of size N, mu is a vector of size N, and sigma is a scalar, then\n ll = normal_lpdf(y | mu, sigma);\nis just a more efficient way to write\n ll = 0;\n for (n in 1:N) {\n   ll = ll + normal_lpdf(y[n] | mu[n], sigma);\n }\nWith the same arguments, the vectorized sampling statement\n y ~ normal(mu, sigma);\nhas the same effect on the total log probability as\n for (n in 1:N) {\n   y[n] ~ normal(mu[n], sigma);\n }\n\n\n\nSome PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes reals or ints. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.\n vector[3] mu = // ...\n array[3] real x = normal_rng(mu, 3);\n\n\nIn the case of PRNG functions, arguments marked ints may be integers or integer arrays, whereas arguments marked reals may be integers or reals, integer or real arrays, vectors, or row vectors.\n\n\n\n\n\n\n\npseudotype\nallowable PRNG arguments\n\n\n\n\nints\nint, array[] int\n\n\nreals\nint, array[] int, real, array[] real, vector, row_vector\n\n\n\n\n\n\nIn general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the normal_rng function may be called with one vector argument and one real array argument as long as they have the same number of elements.\n vector[3] mu = // ...\n array[3] real sigma = // ...\n array[3] real x = normal_rng(mu, sigma);\n\n\n\nThe result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (array[] real) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return ints and continuous distributions return reals, each of appropriate size. The symbol R denotes such a return type.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#suffix-marks-type-of-function",
+    "href": "functions-reference/conventions_for_probability_functions.html#suffix-marks-type-of-function",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "The suffix is determined by the type of function according to the following table.\n\n\n\n\n\n\n\n\nfunction\noutcome\nsuffix\n\n\n\n\nlog probability mass function\ndiscrete\n_lpmf\n\n\nlog probability density function\ncontinuous\n_lpdf\n\n\nlog cumulative distribution function\nany\n_lcdf\n\n\nlog complementary cumulative distribution function\nany\n_lccdf\n\n\nrandom number generator\nany\n_rng\n\n\n\nFor example, normal_lpdf is the log of the normal probability density function (pdf) and bernoulli_lpmf is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, normal_lcdf and bernoulli_lcdf.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#argument-order-and-the-vertical-bar",
+    "href": "functions-reference/conventions_for_probability_functions.html#argument-order-and-the-vertical-bar",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "Each probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, normal_lpdf(y | mu, sigma) returns the value of mathematical formula \\(\\log \\text{Normal}(y \\, | \\, \\mu, \\sigma)\\). Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., normal_lcdf(y_low | mu, sigma)",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#sampling-notation",
+    "href": "functions-reference/conventions_for_probability_functions.html#sampling-notation",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "The notation\n y ~ normal(mu, sigma);\nprovides the same (proportional) contribution to the model log density as the explicit target density increment,\n target += normal_lpdf(y | mu, sigma);\nIn both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (~) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#finite-inputs",
+    "href": "functions-reference/conventions_for_probability_functions.html#finite-inputs",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "All of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#boundary-conditions",
+    "href": "functions-reference/conventions_for_probability_functions.html#boundary-conditions",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "Many distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter \\(\\sigma &gt; 0\\). If \\(\\sigma = 0\\), the probability function will throw an exception.\nThis is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#distributions-prng",
+    "href": "functions-reference/conventions_for_probability_functions.html#distributions-prng",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "For most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix _rng. For example, the function normal_rng(real, real) accepts two real arguments, an unconstrained location \\(\\mu\\) and positive scale \\(\\sigma &gt; 0\\), and returns an unconstrained pseudorandom value drawn from \\(\\text{Normal}(\\mu,\\sigma)\\). There are also vectorized forms of random number generators which return more than one random variate at a time.\n\n\nUnlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.\n\n\n\nUnlike the probability functions, only some of the PRNG functions are vectorized.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#cumulative-distribution-functions",
+    "href": "functions-reference/conventions_for_probability_functions.html#cumulative-distribution-functions",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "For most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.\nFor a univariate random variable \\(Y\\) with probability function \\(p_Y(y \\, | \\, \\theta)\\), the cumulative distribution function (CDF) \\(F_Y\\) is defined by \\[\\begin{equation*}\nF_Y(y) \\ = \\ \\text{Pr}[Y \\le y] \\ = \\ \\int_{-\\infty}^y p(y\\, | \\, \\theta) \\ \\text{d}y.\n\\end{equation*}\\] The complementary cumulative distribution function (CCDF) is defined as \\[\\begin{equation*}\n\\text{Pr}[Y &gt; y] \\ = \\ 1 - F_Y(y).\n\\end{equation*}\\] The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly \\(10^{-300}\\)), but not numbers very close to 1 (the closest you can get is roughly \\(1 - 10^{-15}\\)).\nIn Stan, there is a cumulative distribution function for each probability function. For instance, normal_cdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\ \\text{d}y.\n\\end{equation*}\\] There are also log forms of the CDF and CCDF for most univariate distributions. For example, normal_lcdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right)\n\\end{equation*}\\] and normal_lccdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( 1 - \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right).\n\\end{equation*}\\]",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/conventions_for_probability_functions.html#vectorization",
+    "href": "functions-reference/conventions_for_probability_functions.html#vectorization",
+    "title": "Conventions for Probability Functions",
+    "section": "",
+    "text": "Stan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section vectorized PRNG functions for more details.\nIn all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.\nStan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.\nStan also overloads some scalar functions, such as log and exp, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.\n\n\n\n\nThe normal probability function is specified with the signature\n normal_lpdf(reals | reals, reals);\nThe pseudotype reals is used to indicate that an argument position may be vectorized. Argument positions declared as reals may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use normal_lpdf(row_vector | vector, real) as long as the vector and row vector have the same size.\n\n\n\nThe multivariate normal distribution accepting vector or array of vector arguments is written as\n multi_normal_lpdf(vectors | vectors, matrix);\nThese arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.\n\n\n\nThe pseudotype ints is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.\n\n\n\n\nThe result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely real or int, is repeated. For instance, if y is a vector of size N, mu is a vector of size N, and sigma is a scalar, then\n ll = normal_lpdf(y | mu, sigma);\nis just a more efficient way to write\n ll = 0;\n for (n in 1:N) {\n   ll = ll + normal_lpdf(y[n] | mu[n], sigma);\n }\nWith the same arguments, the vectorized sampling statement\n y ~ normal(mu, sigma);\nhas the same effect on the total log probability as\n for (n in 1:N) {\n   y[n] ~ normal(mu[n], sigma);\n }\n\n\n\nSome PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes reals or ints. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.\n vector[3] mu = // ...\n array[3] real x = normal_rng(mu, 3);\n\n\nIn the case of PRNG functions, arguments marked ints may be integers or integer arrays, whereas arguments marked reals may be integers or reals, integer or real arrays, vectors, or row vectors.\n\n\n\n\n\n\n\npseudotype\nallowable PRNG arguments\n\n\n\n\nints\nint, array[] int\n\n\nreals\nint, array[] int, real, array[] real, vector, row_vector\n\n\n\n\n\n\nIn general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the normal_rng function may be called with one vector argument and one real array argument as long as they have the same number of elements.\n vector[3] mu = // ...\n array[3] real sigma = // ...\n array[3] real x = normal_rng(mu, sigma);\n\n\n\nThe result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (array[] real) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return ints and continuous distributions return reals, each of appropriate size. The symbol R denotes such a return type.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Conventions for Probability Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/covariance_matrix_distributions.html",
+    "href": "functions-reference/covariance_matrix_distributions.html",
+    "title": "Covariance Matrix Distributions",
+    "section": "",
+    "text": "The covariance matrix distributions have support on symmetric, positive-definite \\(K \\times K\\) matrices or their Cholesky factors (square, lower triangular matrices with positive diagonal elements).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{Wishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}}\n\\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{-\\nu/2} \\ \\left| W \\right|^{(\\nu - K - 1)/2}\n\\ \\exp \\! \\left(- \\frac{1}{2} \\ \\text{tr}\\left( S^{-1} W \\right) \\right) \\! ,\n\\end{equation*}\\] where \\(\\text{tr}()\\) is the matrix trace function, and \\(\\Gamma_K()\\) is the multivariate Gamma function, \\[\\begin{equation*}\n\\Gamma_K(x) = \\frac{1}{\\pi^{K(K-1)/4}} \\ \\prod_{k=1}^K \\Gamma \\left( x + \\frac{1 - k}{2} \\right) \\!.\n\\end{equation*}\\]\n\n\n\nW ~ wishart(nu, Sigma)\nIncrement target log probability density with wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix wishart_rng(real nu, matrix Sigma) Generate a Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0\n\n\n\n\nThe Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{WishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{Wishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in \\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is \\[\\begin{equation*}\n\\text{WishartCholesky}(L_W \\mid \\nu,L_S)\n= \\text{Wishart}(L_W L_W^{\\top} \\mid \\nu,L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of a Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks\nAvailable since 2.30\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{InvWishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}} \\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{\\nu/2} \\ \\left| W \\right|^{-(\\nu + K + 1)/2}\n\\ \\exp \\! \\left( - \\frac{1}{2} \\ \\text{tr}(SW^{-1}) \\right) \\! .\n\\end{equation*}\\]\n\n\n\nW ~ inv_wishart(nu, Sigma)\nIncrement target log probability density with inv_wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal inv_wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix inv_wishart_rng(real nu, matrix Sigma) Generate an inverse Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0\n\n\n\n\nThe Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{InvWishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{InvWishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in\n\\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is \\[\\begin{equation*}\n\\text{InvWishartCholesky}(L_W \\mid \\nu,L_S) =\n\\text{InvWishart}(L_WL_W^{\\top} \\mid \\nu, L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal inv_wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal inv_wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix inv_wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/covariance_matrix_distributions.html#wishart-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#wishart-distribution",
+    "title": "Covariance Matrix Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{Wishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}}\n\\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{-\\nu/2} \\ \\left| W \\right|^{(\\nu - K - 1)/2}\n\\ \\exp \\! \\left(- \\frac{1}{2} \\ \\text{tr}\\left( S^{-1} W \\right) \\right) \\! ,\n\\end{equation*}\\] where \\(\\text{tr}()\\) is the matrix trace function, and \\(\\Gamma_K()\\) is the multivariate Gamma function, \\[\\begin{equation*}\n\\Gamma_K(x) = \\frac{1}{\\pi^{K(K-1)/4}} \\ \\prod_{k=1}^K \\Gamma \\left( x + \\frac{1 - k}{2} \\right) \\!.\n\\end{equation*}\\]\n\n\n\nW ~ wishart(nu, Sigma)\nIncrement target log probability density with wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix wishart_rng(real nu, matrix Sigma) Generate a Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/covariance_matrix_distributions.html#wishart-cholesky-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#wishart-cholesky-distribution",
+    "title": "Covariance Matrix Distributions",
+    "section": "",
+    "text": "The Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{WishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{Wishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in \\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is \\[\\begin{equation*}\n\\text{WishartCholesky}(L_W \\mid \\nu,L_S)\n= \\text{Wishart}(L_W L_W^{\\top} \\mid \\nu,L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of a Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-distribution",
+    "title": "Covariance Matrix Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{InvWishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}} \\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{\\nu/2} \\ \\left| W \\right|^{-(\\nu + K + 1)/2}\n\\ \\exp \\! \\left( - \\frac{1}{2} \\ \\text{tr}(SW^{-1}) \\right) \\! .\n\\end{equation*}\\]\n\n\n\nW ~ inv_wishart(nu, Sigma)\nIncrement target log probability density with inv_wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal inv_wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix inv_wishart_rng(real nu, matrix Sigma) Generate an inverse Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-cholesky-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-cholesky-distribution",
+    "title": "Covariance Matrix Distributions",
+    "section": "",
+    "text": "The Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{InvWishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{InvWishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in\n\\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is \\[\\begin{equation*}\n\\text{InvWishartCholesky}(L_W \\mid \\nu,L_S) =\n\\text{InvWishart}(L_WL_W^{\\top} \\mid \\nu, L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal inv_wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal inv_wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix inv_wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "The unbounded vector probability distributions have support on all of \\(\\mathbb{R}^K\\) for some fixed \\(K\\).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormal}(y|\\mu,\\Sigma) =\n\\frac{1}{\\left( 2 \\pi \\right)^{K/2}} \\ \\frac{1}{\\sqrt{|\\Sigma|}}\n\\  \\exp \\! \\left( \\! - \\frac{1}{2} (y - \\mu)^{\\top} \\, \\Sigma^{-1} \\, (y - \\mu) \\right) \\! ,\n\\end{equation*}\\] where \\(|\\Sigma|\\) is the absolute determinant of \\(\\Sigma\\).\n\n\n\ny ~ multi_normal(mu, Sigma)\nIncrement target log probability density with multi_normal_lupdf(y | mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\nThe multivariate normal probability function is overloaded to allow the variate vector \\(y\\) and location vector \\(\\mu\\) to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal multi_normal_lpdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\nAlthough there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call multi_normal_cholesky_rng; see section multi-variate normal, cholesky parameterization.\n \n\nvector multi_normal_rng(vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_normal_rng(row_vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(row_vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Omega \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\text{MultiNormalPrecision}(y|\\mu,\\Omega)\n= \\text{MultiNormal}(y|\\mu,\\Omega^{-1}) \\end{equation*}\\]\n\n\n\ny ~ multi_normal_prec(mu, Omega)\nIncrement target log probability density with multi_normal_prec_lupdf(y | mu, Omega).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_normal_prec_lpdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L \\in\n\\mathbb{R}^{K \\times K}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormalCholesky}(y|\\mu,L) =\n\\text{MultiNormal}(y|\\mu,LL^{\\top}). \\end{equation*}\\] If \\(L\\) is lower triangular and \\(LL^{top}\\) is a \\(K \\times K\\) positive definite matrix, then \\(L_{k,k}\\) must be strictly positive for \\(k \\in 1{:}K\\). If an \\(L\\) is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.\n\n\n\ny ~ multi_normal_cholesky(mu, L)\nIncrement target log probability density with multi_normal_cholesky_lupdf(y | mu, L).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_normal_cholesky_lpdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_normal_cholesky_rng(vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.3\n \n\nvector multi_normal_cholesky_rng(row_vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(row_vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(K,N \\in \\mathbb{N}\\), \\(\\Sigma \\in \\mathbb{R}^{N \\times N}\\) is symmetric, positive definite kernel matrix and \\(w \\in \\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y \\in \\mathbb{R}^{K\n\\times N}\\), \\[\\begin{equation*} \\text{MultiGP}(y|\\Sigma,w) = \\prod_{i=1}^{K}\n\\text{MultiNormal}(y_i|0,w_i^{-1} \\Sigma), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp(Sigma, w)\nIncrement target log probability density with multi_gp_lupdf(y | Sigma, w).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_gp_lpdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_lupdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K,N \\in \\mathbb{N}\\), \\(L \\in \\mathbb{R}^{N \\times N}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite kernel matrix (implying \\(L_{n,n} &gt; 0\\) for \\(n \\in 1{:}N\\)), and \\(w \\in\n\\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y\n\\in \\mathbb{R}^{K \\times N}\\), \\[\\begin{equation*} \\text{MultiGPCholesky}(y \\, | \\ L,w)\n= \\prod_{i=1}^{K} \\text{MultiNormal}(y_i|0,w_i^{-1} LL^{\\top}), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than \\(\\text{MultiGP}()\\). Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp_cholesky(L, w)\nIncrement target log probability density with multi_gp_cholesky_lupdf(y | L, w).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal multi_gp_cholesky_lpdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_cholesky_lupdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in \\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\begin{array}{l}\n\\text{MultiStudentT}(y\\,|\\,\\nu,\\,\\mu,\\,\\Sigma) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\left| \\Sigma\n\\right|}} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\n\\Sigma^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array} \\end{equation*}\\]\n\n\n\ny ~ multi_student_t(nu, mu, Sigma)\nIncrement target log probability density with multi_student_t_lupdf(y | nu, mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_student_t_lpdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_student_t_rng(real nu, vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_student_t_rng(real nu, row_vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, row_vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nLet \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L\\) a \\(K \\times K\\) lower-triangular matrix with strictly positive, finite diagonal then \\[\\begin{equation*}\n\\begin{array}{l}\n\\text{MultiStudentTCholesky}(y\\,\\mid \\nu,\\,\\mu,\\,L) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\left| L\n\\right|} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\nL^{-T}L^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array}\n\\end{equation*}\\]\n\n\n\ny ~ multi_student_t_cholesky(nu, mu, L)\nIncrement target log probability density with multi_student_t_cholesky_lupdf(y | nu, mu, L).\nAvailable since 2.30\n \n\n\n\n\n \n\nreal multi_student_t_cholesky_lpdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or array of vectors y given degrees of freedom nu, location vector or array of vectors mu, and Cholesky factor of the scale matrix L. For a definition of the arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nreal multi_student_t_cholesky_lupdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or vector array y given degrees of freedom nu, location vector or vector array mu, and Cholesky factor of the scale matrix L, dropping constant additive terms. For a definition of arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nvector multi_student_t_cholesky_rng(real nu, vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] row_vector mu, matrix L) Generate an array of multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n\n\n\n\nA Gaussian Dynamic Linear model is defined as follows, For \\(t \\in 1,\n\\dots, T\\), \\[\\begin{equation*}   \\begin{aligned}[t]     y_{t} &\\sim N(F' \\theta_{t}, V)\n\\\\     \\theta_{t} &\\sim N(G \\theta_{t - 1}, W) \\\\     \\theta_{0} &\\sim\nN(m_{0}, C_{0})   \\end{aligned} \\end{equation*}\\] where \\(y\\) is \\(n \\times T\\) matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (\\(p(y | F, G, V, W, m_{0}, C_{0})\\)). This log-density is a system that is calculated using the Kalman Filter. If \\(V\\) is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used (Durbin and Koopman 2001, sec. 6.4).\n\n\ny ~ gaussian_dlm_obs(F, G, V, W, m0, C0)\nIncrement target log probability density with gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0).\nAvailable since 2.0\n \n\n\n\n\nThe following two functions differ in the type of their V, the first taking a full observation covariance matrix V and the second a vector V representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormal}(y|\\mu,\\Sigma) =\n\\frac{1}{\\left( 2 \\pi \\right)^{K/2}} \\ \\frac{1}{\\sqrt{|\\Sigma|}}\n\\  \\exp \\! \\left( \\! - \\frac{1}{2} (y - \\mu)^{\\top} \\, \\Sigma^{-1} \\, (y - \\mu) \\right) \\! ,\n\\end{equation*}\\] where \\(|\\Sigma|\\) is the absolute determinant of \\(\\Sigma\\).\n\n\n\ny ~ multi_normal(mu, Sigma)\nIncrement target log probability density with multi_normal_lupdf(y | mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\nThe multivariate normal probability function is overloaded to allow the variate vector \\(y\\) and location vector \\(\\mu\\) to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal multi_normal_lpdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\nAlthough there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call multi_normal_cholesky_rng; see section multi-variate normal, cholesky parameterization.\n \n\nvector multi_normal_rng(vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_normal_rng(row_vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(row_vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution-precision-parameterization",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution-precision-parameterization",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Omega \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\text{MultiNormalPrecision}(y|\\mu,\\Omega)\n= \\text{MultiNormal}(y|\\mu,\\Omega^{-1}) \\end{equation*}\\]\n\n\n\ny ~ multi_normal_prec(mu, Omega)\nIncrement target log probability density with multi_normal_prec_lupdf(y | mu, Omega).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_normal_prec_lpdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multi-normal-cholesky-fun",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multi-normal-cholesky-fun",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L \\in\n\\mathbb{R}^{K \\times K}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormalCholesky}(y|\\mu,L) =\n\\text{MultiNormal}(y|\\mu,LL^{\\top}). \\end{equation*}\\] If \\(L\\) is lower triangular and \\(LL^{top}\\) is a \\(K \\times K\\) positive definite matrix, then \\(L_{k,k}\\) must be strictly positive for \\(k \\in 1{:}K\\). If an \\(L\\) is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.\n\n\n\ny ~ multi_normal_cholesky(mu, L)\nIncrement target log probability density with multi_normal_cholesky_lupdf(y | mu, L).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_normal_cholesky_lpdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_normal_cholesky_rng(vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.3\n \n\nvector multi_normal_cholesky_rng(row_vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(row_vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "If \\(K,N \\in \\mathbb{N}\\), \\(\\Sigma \\in \\mathbb{R}^{N \\times N}\\) is symmetric, positive definite kernel matrix and \\(w \\in \\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y \\in \\mathbb{R}^{K\n\\times N}\\), \\[\\begin{equation*} \\text{MultiGP}(y|\\Sigma,w) = \\prod_{i=1}^{K}\n\\text{MultiNormal}(y_i|0,w_i^{-1} \\Sigma), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp(Sigma, w)\nIncrement target log probability density with multi_gp_lupdf(y | Sigma, w).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_gp_lpdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_lupdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution-cholesky-parameterization",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution-cholesky-parameterization",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "If \\(K,N \\in \\mathbb{N}\\), \\(L \\in \\mathbb{R}^{N \\times N}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite kernel matrix (implying \\(L_{n,n} &gt; 0\\) for \\(n \\in 1{:}N\\)), and \\(w \\in\n\\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y\n\\in \\mathbb{R}^{K \\times N}\\), \\[\\begin{equation*} \\text{MultiGPCholesky}(y \\, | \\ L,w)\n= \\prod_{i=1}^{K} \\text{MultiNormal}(y_i|0,w_i^{-1} LL^{\\top}), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than \\(\\text{MultiGP}()\\). Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp_cholesky(L, w)\nIncrement target log probability density with multi_gp_cholesky_lupdf(y | L, w).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal multi_gp_cholesky_lpdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_cholesky_lupdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-student-t-distribution",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-student-t-distribution",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in \\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\begin{array}{l}\n\\text{MultiStudentT}(y\\,|\\,\\nu,\\,\\mu,\\,\\Sigma) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\left| \\Sigma\n\\right|}} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\n\\Sigma^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array} \\end{equation*}\\]\n\n\n\ny ~ multi_student_t(nu, mu, Sigma)\nIncrement target log probability density with multi_student_t_lupdf(y | nu, mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_student_t_lpdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_student_t_rng(real nu, vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_student_t_rng(real nu, row_vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, row_vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multi-student-t-cholesky-fun",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multi-student-t-cholesky-fun",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "Let \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L\\) a \\(K \\times K\\) lower-triangular matrix with strictly positive, finite diagonal then \\[\\begin{equation*}\n\\begin{array}{l}\n\\text{MultiStudentTCholesky}(y\\,\\mid \\nu,\\,\\mu,\\,L) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\left| L\n\\right|} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\nL^{-T}L^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array}\n\\end{equation*}\\]\n\n\n\ny ~ multi_student_t_cholesky(nu, mu, L)\nIncrement target log probability density with multi_student_t_cholesky_lupdf(y | nu, mu, L).\nAvailable since 2.30\n \n\n\n\n\n \n\nreal multi_student_t_cholesky_lpdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or array of vectors y given degrees of freedom nu, location vector or array of vectors mu, and Cholesky factor of the scale matrix L. For a definition of the arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nreal multi_student_t_cholesky_lupdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or vector array y given degrees of freedom nu, location vector or vector array mu, and Cholesky factor of the scale matrix L, dropping constant additive terms. For a definition of arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nvector multi_student_t_cholesky_rng(real nu, vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] row_vector mu, matrix L) Generate an array of multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#gaussian-dynamic-linear-models",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#gaussian-dynamic-linear-models",
+    "title": "Distributions over Unbounded Vectors",
+    "section": "",
+    "text": "A Gaussian Dynamic Linear model is defined as follows, For \\(t \\in 1,\n\\dots, T\\), \\[\\begin{equation*}   \\begin{aligned}[t]     y_{t} &\\sim N(F' \\theta_{t}, V)\n\\\\     \\theta_{t} &\\sim N(G \\theta_{t - 1}, W) \\\\     \\theta_{0} &\\sim\nN(m_{0}, C_{0})   \\end{aligned} \\end{equation*}\\] where \\(y\\) is \\(n \\times T\\) matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (\\(p(y | F, G, V, W, m_{0}, C_{0})\\)). This log-density is a system that is calculated using the Kalman Filter. If \\(V\\) is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used (Durbin and Koopman 2001, sec. 6.4).\n\n\ny ~ gaussian_dlm_obs(F, G, V, W, m0, C0)\nIncrement target log probability density with gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0).\nAvailable since 2.0\n \n\n\n\n\nThe following two functions differ in the type of their V, the first taking a full observation covariance matrix V and the second a vector V representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html",
+    "href": "functions-reference/functions_index.html",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "abs:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\nacos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nacosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nadd_diag:\n\n\n(complex_matrix m, complex_real d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_row_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, real d) : matrix (matrix_operations.html)\n\n\n(matrix m, row_vector d) : matrix (matrix_operations.html)\n\n\n(matrix m, vector d) : matrix (matrix_operations.html)\n\n\nalgebra_solver:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nalgebra_solver_newton:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector (deprecated_functions.html)\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nappend_array:\n\n\n(T x, T y) : T (array_operations.html)\n\n\nappend_col:\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(vector x, vector y) : matrix (matrix_operations.html)\n\n\nappend_row:\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, row_vector y) : matrix (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : matrix (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\narg:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nasin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nasinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan2:\n\n\n(T y, T x) : R (real-valued_basic_functions.html)\n\n\natanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\n\n\n\nbernoulli:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_cdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lccdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lcdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_logit:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_rng:\n\n\n(matrix x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\n(row_vector x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\nbernoulli_logit_lpmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_lupmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_rng:\n\n\n(reals alpha) : R (binary_distributions.html)\n\n\nbernoulli_lpmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lupmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_rng:\n\n\n(reals theta) : R (binary_distributions.html)\n\n\nbessel_first_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbessel_second_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_binomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbeta_binomial_cdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lccdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lcdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lpmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lupmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_rng:\n\n\n(ints N, reals alpha, reals beta) : R (bounded_discrete_distributions.html)\n\n\nbeta_cdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lccdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lcdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lpdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lupdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_cdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lccdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lcdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lpmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lupmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_rng:\n\n\n(reals r, reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nbeta_proportion:\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lccdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lcdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lpdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lupdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_rng:\n\n\n(reals mu, reals kappa) : R (continuous_distributions_on_0_1.html)\n\n\nbeta_rng:\n\n\n(reals alpha, reals beta) : R (continuous_distributions_on_0_1.html)\n\n\nbinary_log_loss:\n\n\n(int y, real y_hat) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbinomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_cdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lccdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lcdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lpmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lupmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lpmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lupmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lpmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lupmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_rng:\n\n\n(ints N, reals theta) : R (bounded_discrete_distributions.html)\n\n\nblock:\n\n\n(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows, int n_cols) : matrix (matrix_operations.html)\n\n\n\n\n\ncategorical:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lpmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lupmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_rng:\n\n\n(vector beta) : int (bounded_discrete_distributions.html)\n\n\ncategorical_lpmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_lupmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_rng:\n\n\n(vector theta) : int (bounded_discrete_distributions.html)\n\n\ncauchy:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ncauchy_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\ncbrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nceil:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nchi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nchi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\nchol2inv:\n\n\n(matrix L) : matrix (matrix_operations.html)\n\n\ncholesky_decompose:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ncholesky_factor_corr_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncholesky_factor_cov_constrain:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_jacobian:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\nchoose:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ncol:\n\n\n(complex_matrix x, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int n) : vector (matrix_operations.html)\n\n\ncols:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\ncolumns_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : row_vector (matrix_operations.html)\n\n\ncolumns_dot_self:\n\n\n(complex_matrix x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : row_vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\ncomplex_schur_decompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_t:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_u:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\nconj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(Z z) : Z (complex-valued_basic_functions.html)\n\n\ncorr_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncov_exp_quad:\n\n\n(array[] real x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(array[] real x1, array[] real x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x1, vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\ncov_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ncsr_extract:\n\n\n(matrix a) : tuple(vector, array[] int, array[] int) (sparse_matrix_operations.html)\n\n\ncsr_extract_u:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_v:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_w:\n\n\n(matrix a) : vector (sparse_matrix_operations.html)\n\n\ncsr_matrix_times_vector:\n\n\n(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector (sparse_matrix_operations.html)\n\n\ncsr_to_dense_matrix:\n\n\n(int m, int n, vector w, array[] int v, array[] int u) : matrix (sparse_matrix_operations.html)\n\n\ncumulative_sum:\n\n\n(array[] complex x) : array[] complex (complex_matrix_operations.html)\n\n\n(array[] int x) : array[] int (matrix_operations.html)\n\n\n(array[] real x) : array[] real (matrix_operations.html)\n\n\n(complex_row_vector rv) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\n\n\n\ndae:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\ndae_tol:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\ndeterminant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ndiag_matrix:\n\n\n(complex_vector x) : complex_matrix (complex_matrix_operations.html)\n\n\n(vector x) : matrix (matrix_operations.html)\n\n\ndiag_post_multiply:\n\n\n(complex_matrix m, complex_row_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\ndiag_pre_multiply:\n\n\n(complex_row_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(row_vector rv, matrix m) : matrix (matrix_operations.html)\n\n\n(vector v, matrix m) : matrix (matrix_operations.html)\n\n\ndiagonal:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\ndigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ndims:\n\n\n(T x) : array[] int (array_operations.html)\n\n\ndirichlet:\n\n\ndistribution statement (simplex_distributions.html)\n\n\ndirichlet_lpdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_lupdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_multinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lpmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lupmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_rng:\n\n\n(vector alpha, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\ndirichlet_rng:\n\n\n(vector alpha) : vector (simplex_distributions.html)\n\n\ndiscrete_range:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ndiscrete_range_cdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lccdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lcdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lpmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lupmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_rng:\n\n\n(ints l, ints u) : ints (bounded_discrete_distributions.html)\n\n\ndistance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\ndot_product:\n\n\n(complex_row_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(row_vector x, row_vector y) : real (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, row_vector y) : real (matrix_operations.html)\n\n\n(vector x, vector y) : real (matrix_operations.html)\n\n\ndot_self:\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\ndouble_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\n\n\n\ne:\n\n\n() : real (real-valued_basic_functions.html)\n\n\neigendecompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_vector) (matrix_operations.html)\n\n\neigendecompose_sym:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector) (matrix_operations.html)\n\n\neigenvalues:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : complex_vector (matrix_operations.html)\n\n\neigenvalues_sym:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\neigenvectors:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (matrix_operations.html)\n\n\neigenvectors_sym:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nerf:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nerfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp2:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp_mod_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_cdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_rng:\n\n\n(reals mu, reals sigma, reals lambda) : R (unbounded_continuous_distributions.html)\n\n\nexpm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexponential:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nexponential_cdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lccdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lcdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lpdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lupdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_rng:\n\n\n(reals beta) : R (positive_continuous_distributions.html)\n\n\n\n\n\nfalling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfatal_error:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nfdim:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfft:\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\nfft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\nfloor:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nfma:\n\n\n(real x, real y, real z) : real (real-valued_basic_functions.html)\n\n\nfmax:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmin:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmod:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfrechet:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nfrechet_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\n\n\n\ngamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ngamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_p:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_q:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ngaussian_dlm_obs:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lpdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lupdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngeneralized_inverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ngenerate_laplace_options:\n\n\n(int dimension) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\n(vector theta_init) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\nget_imag:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\nget_real:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\ngp_dot_prod_cov:\n\n\n(array[] real x, real sigma) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma) : matrix (matrix_operations.html)\n\n\ngp_exp_quad_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\ngp_exponential_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern23_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern52_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_periodic_cov:\n\n\n(array[] real x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\ngumbel:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ngumbel_cdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lccdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lcdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lpdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lupdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_rng:\n\n\n(reals mu, reals beta) : R (unbounded_continuous_distributions.html)\n\n\n\n\n\nhead:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\nhmm_hidden_state_prob:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : matrix (hidden_markov_models.html)\n\n\nhmm_latent_rng:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : array[] int (hidden_markov_models.html)\n\n\nhmm_marginal:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : real (hidden_markov_models.html)\n\n\nhypergeometric:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nhypergeometric_1F0:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_2F1:\n\n\n(real a1, real a2, real b1, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_3F2:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_lpmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_lupmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_pFq:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_rng:\n\n\n(int N, int a, int2 b) : int (bounded_discrete_distributions.html)\n\n\nhypot:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n\n\n\nidentity_matrix:\n\n\n(int k) : matrix (matrix_operations.html)\n\n\ninc_beta:\n\n\n(real alpha, real beta, real x) : real (real-valued_basic_functions.html)\n\n\nint_step:\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\nintegrate_1d:\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real (higher-order_functions.html)\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real (higher-order_functions.html)\n\n\nintegrate_ode:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_adams:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_bdf:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_rk45:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\ninv:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_chi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\ninv_cloglog:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_erfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_fft:\n\n\n(complex_vector u) : complex_vector (complex_matrix_operations.html)\n\n\ninv_fft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\ninv_gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_gamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ninv_inc_beta:\n\n\n(real alpha, real beta, real p) : real (real-valued_basic_functions.html)\n\n\ninv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_Phi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_sqrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_square:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_wishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\ninv_wishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\ninverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ninverse_spd:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nis_inf:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\nis_nan:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\n\n\n\nlambert_w0:\n\n\n(reals x) : R (real-valued_basic_functions.html)\n\n\nlambert_wm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlaplace_latent_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_marginal:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlchoose:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nldexp:\n\n\n(real x, int y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlinspaced_array:\n\n\n(int n, data real lower, data real upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_int_array:\n\n\n(int n, int lower, int upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_row_vector:\n\n\n(int n, data real lower, data real upper) : row_vector (matrix_operations.html)\n\n\nlinspaced_vector:\n\n\n(int n, data real lower, data real upper) : vector (matrix_operations.html)\n\n\nlkj_corr:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lpdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lupdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlkj_corr_lpdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_lupdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlmgamma:\n\n\n(int n, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlmultiply:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog10:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_determinant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\nlog_diff_exp:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\nlog_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_inv_logit_diff:\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_mix:\n\n\n(real theta, real lp1, real lp2) : real (real-valued_basic_functions.html)\n\n\n(T1 thetas, T2 lps) : R (real-valued_basic_functions.html)\n\n\nlog_modified_bessel_first_kind:\n\n\n(real v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_rising_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_softmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nlog_sum_exp:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nlogistic:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nlogistic_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nlogit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nloglogistic:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nloglogistic_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlognormal:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nlognormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlower_bound_constrain:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_jacobian:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_unconstrain:\n\n\n(reals x, reals lb) : reals (transform_functions.html)\n\n\nlower_upper_bound_constrain:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_jacobian:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_unconstrain:\n\n\n(reals x, reals lb, reals ub) : reals (transform_functions.html)\n\n\n\n\n\nmachine_precision:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nmap_rect:\n\n\n(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector (higher-order_functions.html)\n\n\nmatrix_exp:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nmatrix_exp_multiply:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nmatrix_power:\n\n\n(matrix A, int B) : matrix (matrix_operations.html)\n\n\nmax:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmdivide_left_spd:\n\n\n(matrix A, matrix B) : vector (matrix_operations.html)\n\n\n(matrix A, vector b) : matrix (matrix_operations.html)\n\n\nmdivide_left_tri_low:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\nmdivide_right_spd:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmdivide_right_tri_low:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmean:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmin:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmodified_bessel_first_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmodified_bessel_second_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmulti_gp:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lpdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lupdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lpdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lupdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_rng:\n\n\n(row_vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_rng:\n\n\n(row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_cholesky_t_rng:\n\n\n(real nu, vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lpdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lupdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_rng:\n\n\n(real nu, array[] row_vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, array[] vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lpdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lupdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_rng:\n\n\n(real nu, row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmultinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lpmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lupmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_rng:\n\n\n(vector gamma, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultinomial_lpmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_lupmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_rng:\n\n\n(vector theta, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultiply_lower_tri_self_transpose:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n\n\n\nneg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_cdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lccdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lcdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lpmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lupmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_rng:\n\n\n(reals eta, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lpmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lupmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_rng:\n\n\n(reals mu, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_cdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lccdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lcdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lpmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lupmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_rng:\n\n\n(reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nnegative_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnorm:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nnorm1:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnorm2:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnormal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lpdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lupdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nnot_a_number:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnum_elements:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\n\n\n\node_adams:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_adams_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_adjoint_tol_ctl:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector (higher-order_functions.html)\n\n\node_bdf:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_bdf_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\noffset_multiplier_constrain:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_jacobian:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_unconstrain:\n\n\n(reals x, reals offset, reals mult) : reals (transform_functions.html)\n\n\none_hot_array:\n\n\n(int n, int k) : array[] real (matrix_operations.html)\n\n\none_hot_int_array:\n\n\n(int n, int k) : array[] int (matrix_operations.html)\n\n\none_hot_row_vector:\n\n\n(int n, int k) : row_vector (matrix_operations.html)\n\n\none_hot_vector:\n\n\n(int K, int k) : vector (matrix_operations.html)\n\n\nones_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nones_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nones_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nones_vector:\n\n\n(int n) : vector (matrix_operations.html)\n\n\noperator!:\n\n\n(int x) : int (real-valued_basic_functions.html)\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\noperator!=:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator%/%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator&&:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\n&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;operator’:\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\noperator*:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : vector (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, row_vector y) : matrix (matrix_operations.html)\n\n\noperator*=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator+:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator+=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator-:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(T x) : T (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (matrix_operations.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator-=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.*:\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator.*=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator./:\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator./=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.^:\n\n\n( complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n( complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator/:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix B, complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector b, complex_matrix A) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\noperator/=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator&lt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&lt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\noperator==:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator\\:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\noperator^:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\noperator||:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\nordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_logistic:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lpmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lupmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lpmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lupmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_probit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_probit_lpmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_lupmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nowens_t:\n\n\n(real h, real a) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n\n\n\npareto:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_cdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lccdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lcdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lpdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lupdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_rng:\n\n\n(reals y_min, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\npareto_type_2:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_cdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lccdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lcdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lpdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lupdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_rng:\n\n\n(reals mu, reals lambda, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\nPhi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nPhi_approx:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\npi:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npoisson:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_cdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lccdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lcdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\npoisson_lpmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lupmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_rng:\n\n\n(reals lambda) : R (unbounded_discrete_distributions.html)\n\n\npolar:\n\n\n(real r, real theta) : complex (complex-valued_basic_functions.html)\n\n\npositive_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npositive_ordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\npow:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : Z (complex-valued_basic_functions.html)\n\n\nprint:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nprod:\n\n\n(array[] int x) : real (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nproj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n\n\n\nqr:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_thin_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nquad_form:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquad_form_diag:\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\nquad_form_sym:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquantile:\n\n\n(data array[] real x, data array[] real p) : array[] real (array_operations.html)\n\n\n(data array[] real x, data real p) : real (array_operations.html)\n\n\n(data row_vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data row_vector x, data real p) : real (matrix_operations.html)\n\n\n(data vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data vector x, data real p) : real (matrix_operations.html)\n\n\n\n\n\nrank:\n\n\n(array[] int v, int s) : int (array_operations.html)\n\n\n(array[] real v, int s) : int (array_operations.html)\n\n\n(row_vector v, int s) : int (matrix_operations.html)\n\n\n(vector v, int s) : int (matrix_operations.html)\n\n\nrayleigh:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nrayleigh_cdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lccdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lcdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lpdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lupdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_rng:\n\n\n(reals sigma) : R (positive_continuous_distributions.html)\n\n\nreduce_sum:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreduce_sum_static:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreject:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nrep_array:\n\n\n(T x, int k, int m, int n) : array[,,] T (array_operations.html)\n\n\n(T x, int m, int n) : array[,] T (array_operations.html)\n\n\n(T x, int n) : array[] T (array_operations.html)\n\n\nrep_matrix:\n\n\n(complex z, int m, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector rv, int m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(real x, int m, int n) : matrix (matrix_operations.html)\n\n\n(row_vector rv, int m) : matrix (matrix_operations.html)\n\n\n(vector v, int n) : matrix (matrix_operations.html)\n\n\nrep_row_vector:\n\n\n(complex z, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(real x, int n) : row_vector (matrix_operations.html)\n\n\nrep_vector:\n\n\n(complex z, int m) : complex_vector (complex_matrix_operations.html)\n\n\n(real x, int m) : vector (matrix_operations.html)\n\n\nreverse:\n\n\n(array[] T v) : array[] T (array_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nrising_factorial:\n\n\n(real x, int n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nround:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nrow:\n\n\n(complex_matrix x, int m) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int m) : row_vector (matrix_operations.html)\n\n\nrows:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nrows_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\nrows_dot_self:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n\n\n\nscale_matrix_exp_multiply:\n\n\n(real t, matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nscaled_inv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_cdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lccdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lcdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lpdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lupdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_rng:\n\n\n(reals nu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nsd:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsegment:\n\n\n(array[] T sv, int i, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int i, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int i, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int i, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int i, int n) : vector (matrix_operations.html)\n\n\nsimplex_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsingular_values:\n\n\n(complex_matrix A) : vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\nsinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsize:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nskew_double_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nskew_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_normal_cdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lccdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lcdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lpdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lupdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_rng:\n\n\n(reals xi, reals omega, real alpha) : R (unbounded_continuous_distributions.html)\n\n\nsoftmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nsolve_newton:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_newton_tol:\n\n\n(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell_tol:\n\n\n(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsort_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_indices_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsort_indices_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsqrt:\n\n\n(complex x) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsqrt2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nsquare:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsquared_distance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\nstd_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstd_normal_cdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lccdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lcdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_log_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_lpdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lupdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_rng:\n\n\n() : real (unbounded_continuous_distributions.html)\n\n\nstep:\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\nstochastic_column_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstochastic_row_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstudent_t:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstudent_t_cdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lccdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lcdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lpdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lupdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_rng:\n\n\n(reals nu, reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nsub_col:\n\n\n(complex_matrix x, int i, int j, int n_rows) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows) : vector (matrix_operations.html)\n\n\nsub_row:\n\n\n(complex_matrix x, int i, int j, int n_cols) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_cols) : row_vector (matrix_operations.html)\n\n\nsum:\n\n\n(array[] complex x) : complex (array_operations.html)\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsum_to_zero_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsvd:\n\n\n(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector, matrix) (matrix_operations.html)\n\n\nsvd_U:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsvd_V:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsymmetrize_from_lower_tri:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\n\n\n\ntail:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\ntan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntarget:\n\n\n() : real (real-valued_basic_functions.html)\n\n\ntcrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ntgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nto_array_1d:\n\n\n(array[...] complex a) : array[] complex (mixed_operations.html)\n\n\n(array[...] int a) : array[] int (mixed_operations.html)\n\n\n(array[...] real a) : array[] real (mixed_operations.html)\n\n\n(complex_matrix m) : array[] complex (mixed_operations.html)\n\n\n(complex_row_vector v) : array[] complex (mixed_operations.html)\n\n\n(complex_vector v) : array[] real (mixed_operations.html)\n\n\n(matrix m) : array[] real (mixed_operations.html)\n\n\n(row_vector v) : array[] real (mixed_operations.html)\n\n\n(vector v) : array[] real (mixed_operations.html)\n\n\nto_array_2d:\n\n\n(complex_matrix m) : array[,] real (mixed_operations.html)\n\n\n(matrix m) : array[,] real (mixed_operations.html)\n\n\nto_complex:\n\n\n() : complex (complex-valued_basic_functions.html)\n\n\n(real re) : complex (complex-valued_basic_functions.html)\n\n\n(real re, real im) : complex (complex-valued_basic_functions.html)\n\n\n(T1 re, T2 im) : Z (complex-valued_basic_functions.html)\n\n\nto_int:\n\n\n(data real x) : int (integer-valued_basic_functions.html)\n\n\nto_matrix:\n\n\n(array[,] complex a ) : complex_matrix (mixed_operations.html)\n\n\n(array[,] int a) : matrix (mixed_operations.html)\n\n\n(array[,] real a) : matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex_row_vector vs) : complex_matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] row_vector vs) : matrix (mixed_operations.html)\n\n\n(complex_matrix A, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix m) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix M, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(matrix A, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(matrix m) : matrix (mixed_operations.html)\n\n\n(matrix M, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(vector v) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\nto_row_vector:\n\n\n(array[] complex a) : complex_row_vector (mixed_operations.html)\n\n\n(array[] int a) : row_vector (mixed_operations.html)\n\n\n(array[] real a) : row_vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_row_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(matrix m) : row_vector (mixed_operations.html)\n\n\n(row_vector v) : row_vector (mixed_operations.html)\n\n\n(vector v) : row_vector (mixed_operations.html)\n\n\nto_vector:\n\n\n(array[] complex a) : complex_vector (mixed_operations.html)\n\n\n(array[] int a) : vector (mixed_operations.html)\n\n\n(array[] real a) : vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_vector (mixed_operations.html)\n\n\n(matrix m) : vector (mixed_operations.html)\n\n\n(row_vector v) : vector (mixed_operations.html)\n\n\n(vector v) : vector (mixed_operations.html)\n\n\ntrace:\n\n\n(complex_matrix A) : complex (complex_matrix_operations.html)\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ntrace_dot:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_gen_quad_form:\n\n\n(matrix D ,matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_quad_form:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\ntrigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntrunc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\n\n\n\nuniform:\n\n\ndistribution statement (bounded_continuous_distributions.html)\n\n\nuniform_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_rng:\n\n\n(reals alpha, reals beta) : R (bounded_continuous_distributions.html)\n\n\nuniform_simplex:\n\n\n(int n) : vector (matrix_operations.html)\n\n\nunit_vectors_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nupper_bound_constrain:\n\n\n(reals y, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_jacobian:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_unconstrain:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\n\n\n\nvariance:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nvon_mises:\n\n\ndistribution statement (circular_distributions.html)\n\n\nvon_mises_cdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lccdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lcdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lpdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lupdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_rng:\n\n\n(reals mu, reals kappa) : R (circular_distributions.html)\n\n\n\n\n\nweibull:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nweibull_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\nwiener:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\nwiener_lccdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lcdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lpdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lupdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\nwishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\n\n\n\nyule_simon:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nyule_simon_cdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lccdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lcdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\n\n\n\nzeros_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nzeros_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nzeros_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nzeros_vector:\n\n\n(int n) : vector (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#a",
+    "href": "functions-reference/functions_index.html#a",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "abs:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\nacos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nacosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nadd_diag:\n\n\n(complex_matrix m, complex_real d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_row_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, real d) : matrix (matrix_operations.html)\n\n\n(matrix m, row_vector d) : matrix (matrix_operations.html)\n\n\n(matrix m, vector d) : matrix (matrix_operations.html)\n\n\nalgebra_solver:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nalgebra_solver_newton:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector (deprecated_functions.html)\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nappend_array:\n\n\n(T x, T y) : T (array_operations.html)\n\n\nappend_col:\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(vector x, vector y) : matrix (matrix_operations.html)\n\n\nappend_row:\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, row_vector y) : matrix (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : matrix (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\narg:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nasin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nasinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan2:\n\n\n(T y, T x) : R (real-valued_basic_functions.html)\n\n\natanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#b",
+    "href": "functions-reference/functions_index.html#b",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "bernoulli:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_cdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lccdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lcdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_logit:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_rng:\n\n\n(matrix x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\n(row_vector x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\nbernoulli_logit_lpmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_lupmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_rng:\n\n\n(reals alpha) : R (binary_distributions.html)\n\n\nbernoulli_lpmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lupmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_rng:\n\n\n(reals theta) : R (binary_distributions.html)\n\n\nbessel_first_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbessel_second_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_binomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbeta_binomial_cdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lccdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lcdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lpmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lupmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_rng:\n\n\n(ints N, reals alpha, reals beta) : R (bounded_discrete_distributions.html)\n\n\nbeta_cdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lccdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lcdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lpdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lupdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_cdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lccdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lcdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lpmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lupmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_rng:\n\n\n(reals r, reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nbeta_proportion:\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lccdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lcdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lpdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lupdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_rng:\n\n\n(reals mu, reals kappa) : R (continuous_distributions_on_0_1.html)\n\n\nbeta_rng:\n\n\n(reals alpha, reals beta) : R (continuous_distributions_on_0_1.html)\n\n\nbinary_log_loss:\n\n\n(int y, real y_hat) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbinomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_cdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lccdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lcdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lpmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lupmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lpmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lupmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lpmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lupmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_rng:\n\n\n(ints N, reals theta) : R (bounded_discrete_distributions.html)\n\n\nblock:\n\n\n(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows, int n_cols) : matrix (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#c",
+    "href": "functions-reference/functions_index.html#c",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "categorical:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lpmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lupmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_rng:\n\n\n(vector beta) : int (bounded_discrete_distributions.html)\n\n\ncategorical_lpmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_lupmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_rng:\n\n\n(vector theta) : int (bounded_discrete_distributions.html)\n\n\ncauchy:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ncauchy_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\ncbrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nceil:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nchi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nchi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\nchol2inv:\n\n\n(matrix L) : matrix (matrix_operations.html)\n\n\ncholesky_decompose:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ncholesky_factor_corr_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncholesky_factor_cov_constrain:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_jacobian:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\nchoose:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ncol:\n\n\n(complex_matrix x, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int n) : vector (matrix_operations.html)\n\n\ncols:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\ncolumns_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : row_vector (matrix_operations.html)\n\n\ncolumns_dot_self:\n\n\n(complex_matrix x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : row_vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\ncomplex_schur_decompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_t:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_u:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\nconj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(Z z) : Z (complex-valued_basic_functions.html)\n\n\ncorr_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncov_exp_quad:\n\n\n(array[] real x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(array[] real x1, array[] real x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x1, vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\ncov_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ncsr_extract:\n\n\n(matrix a) : tuple(vector, array[] int, array[] int) (sparse_matrix_operations.html)\n\n\ncsr_extract_u:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_v:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_w:\n\n\n(matrix a) : vector (sparse_matrix_operations.html)\n\n\ncsr_matrix_times_vector:\n\n\n(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector (sparse_matrix_operations.html)\n\n\ncsr_to_dense_matrix:\n\n\n(int m, int n, vector w, array[] int v, array[] int u) : matrix (sparse_matrix_operations.html)\n\n\ncumulative_sum:\n\n\n(array[] complex x) : array[] complex (complex_matrix_operations.html)\n\n\n(array[] int x) : array[] int (matrix_operations.html)\n\n\n(array[] real x) : array[] real (matrix_operations.html)\n\n\n(complex_row_vector rv) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#d",
+    "href": "functions-reference/functions_index.html#d",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "dae:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\ndae_tol:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\ndeterminant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ndiag_matrix:\n\n\n(complex_vector x) : complex_matrix (complex_matrix_operations.html)\n\n\n(vector x) : matrix (matrix_operations.html)\n\n\ndiag_post_multiply:\n\n\n(complex_matrix m, complex_row_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\ndiag_pre_multiply:\n\n\n(complex_row_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(row_vector rv, matrix m) : matrix (matrix_operations.html)\n\n\n(vector v, matrix m) : matrix (matrix_operations.html)\n\n\ndiagonal:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\ndigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ndims:\n\n\n(T x) : array[] int (array_operations.html)\n\n\ndirichlet:\n\n\ndistribution statement (simplex_distributions.html)\n\n\ndirichlet_lpdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_lupdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_multinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lpmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lupmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_rng:\n\n\n(vector alpha, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\ndirichlet_rng:\n\n\n(vector alpha) : vector (simplex_distributions.html)\n\n\ndiscrete_range:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ndiscrete_range_cdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lccdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lcdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lpmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lupmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_rng:\n\n\n(ints l, ints u) : ints (bounded_discrete_distributions.html)\n\n\ndistance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\ndot_product:\n\n\n(complex_row_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(row_vector x, row_vector y) : real (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, row_vector y) : real (matrix_operations.html)\n\n\n(vector x, vector y) : real (matrix_operations.html)\n\n\ndot_self:\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\ndouble_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#e",
+    "href": "functions-reference/functions_index.html#e",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "e:\n\n\n() : real (real-valued_basic_functions.html)\n\n\neigendecompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_vector) (matrix_operations.html)\n\n\neigendecompose_sym:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector) (matrix_operations.html)\n\n\neigenvalues:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : complex_vector (matrix_operations.html)\n\n\neigenvalues_sym:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\neigenvectors:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (matrix_operations.html)\n\n\neigenvectors_sym:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nerf:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nerfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp2:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp_mod_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_cdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_rng:\n\n\n(reals mu, reals sigma, reals lambda) : R (unbounded_continuous_distributions.html)\n\n\nexpm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexponential:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nexponential_cdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lccdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lcdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lpdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lupdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_rng:\n\n\n(reals beta) : R (positive_continuous_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#f",
+    "href": "functions-reference/functions_index.html#f",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfatal_error:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nfdim:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfft:\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\nfft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\nfloor:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nfma:\n\n\n(real x, real y, real z) : real (real-valued_basic_functions.html)\n\n\nfmax:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmin:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmod:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfrechet:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nfrechet_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#g",
+    "href": "functions-reference/functions_index.html#g",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ngamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_p:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_q:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ngaussian_dlm_obs:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lpdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lupdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngeneralized_inverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ngenerate_laplace_options:\n\n\n(int dimension) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\n(vector theta_init) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\nget_imag:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\nget_real:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\ngp_dot_prod_cov:\n\n\n(array[] real x, real sigma) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma) : matrix (matrix_operations.html)\n\n\ngp_exp_quad_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\ngp_exponential_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern23_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern52_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_periodic_cov:\n\n\n(array[] real x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\ngumbel:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ngumbel_cdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lccdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lcdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lpdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lupdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_rng:\n\n\n(reals mu, reals beta) : R (unbounded_continuous_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#h",
+    "href": "functions-reference/functions_index.html#h",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "head:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\nhmm_hidden_state_prob:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : matrix (hidden_markov_models.html)\n\n\nhmm_latent_rng:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : array[] int (hidden_markov_models.html)\n\n\nhmm_marginal:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : real (hidden_markov_models.html)\n\n\nhypergeometric:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nhypergeometric_1F0:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_2F1:\n\n\n(real a1, real a2, real b1, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_3F2:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_lpmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_lupmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_pFq:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_rng:\n\n\n(int N, int a, int2 b) : int (bounded_discrete_distributions.html)\n\n\nhypot:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#i",
+    "href": "functions-reference/functions_index.html#i",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "identity_matrix:\n\n\n(int k) : matrix (matrix_operations.html)\n\n\ninc_beta:\n\n\n(real alpha, real beta, real x) : real (real-valued_basic_functions.html)\n\n\nint_step:\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\nintegrate_1d:\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real (higher-order_functions.html)\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real (higher-order_functions.html)\n\n\nintegrate_ode:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_adams:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_bdf:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_rk45:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\ninv:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_chi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\ninv_cloglog:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_erfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_fft:\n\n\n(complex_vector u) : complex_vector (complex_matrix_operations.html)\n\n\ninv_fft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\ninv_gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_gamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ninv_inc_beta:\n\n\n(real alpha, real beta, real p) : real (real-valued_basic_functions.html)\n\n\ninv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_Phi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_sqrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_square:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_wishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\ninv_wishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\ninverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ninverse_spd:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nis_inf:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\nis_nan:\n\n\n(real x) : int (real-valued_basic_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#l",
+    "href": "functions-reference/functions_index.html#l",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "lambert_w0:\n\n\n(reals x) : R (real-valued_basic_functions.html)\n\n\nlambert_wm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlaplace_latent_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_marginal:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlchoose:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nldexp:\n\n\n(real x, int y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlinspaced_array:\n\n\n(int n, data real lower, data real upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_int_array:\n\n\n(int n, int lower, int upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_row_vector:\n\n\n(int n, data real lower, data real upper) : row_vector (matrix_operations.html)\n\n\nlinspaced_vector:\n\n\n(int n, data real lower, data real upper) : vector (matrix_operations.html)\n\n\nlkj_corr:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lpdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lupdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlkj_corr_lpdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_lupdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlmgamma:\n\n\n(int n, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlmultiply:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog10:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_determinant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\nlog_diff_exp:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\nlog_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_inv_logit_diff:\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_mix:\n\n\n(real theta, real lp1, real lp2) : real (real-valued_basic_functions.html)\n\n\n(T1 thetas, T2 lps) : R (real-valued_basic_functions.html)\n\n\nlog_modified_bessel_first_kind:\n\n\n(real v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_rising_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_softmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nlog_sum_exp:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nlogistic:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nlogistic_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nlogit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nloglogistic:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nloglogistic_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlognormal:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nlognormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlower_bound_constrain:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_jacobian:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_unconstrain:\n\n\n(reals x, reals lb) : reals (transform_functions.html)\n\n\nlower_upper_bound_constrain:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_jacobian:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_unconstrain:\n\n\n(reals x, reals lb, reals ub) : reals (transform_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#m",
+    "href": "functions-reference/functions_index.html#m",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "machine_precision:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nmap_rect:\n\n\n(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector (higher-order_functions.html)\n\n\nmatrix_exp:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nmatrix_exp_multiply:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nmatrix_power:\n\n\n(matrix A, int B) : matrix (matrix_operations.html)\n\n\nmax:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmdivide_left_spd:\n\n\n(matrix A, matrix B) : vector (matrix_operations.html)\n\n\n(matrix A, vector b) : matrix (matrix_operations.html)\n\n\nmdivide_left_tri_low:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\nmdivide_right_spd:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmdivide_right_tri_low:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmean:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmin:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmodified_bessel_first_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmodified_bessel_second_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmulti_gp:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lpdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lupdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lpdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lupdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_rng:\n\n\n(row_vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_rng:\n\n\n(row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_cholesky_t_rng:\n\n\n(real nu, vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lpdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lupdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_rng:\n\n\n(real nu, array[] row_vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, array[] vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lpdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lupdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_rng:\n\n\n(real nu, row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmultinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lpmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lupmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_rng:\n\n\n(vector gamma, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultinomial_lpmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_lupmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_rng:\n\n\n(vector theta, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultiply_lower_tri_self_transpose:\n\n\n(matrix x) : matrix (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#n",
+    "href": "functions-reference/functions_index.html#n",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_cdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lccdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lcdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lpmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lupmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_rng:\n\n\n(reals eta, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lpmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lupmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_rng:\n\n\n(reals mu, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_cdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lccdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lcdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lpmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lupmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_rng:\n\n\n(reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nnegative_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnorm:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nnorm1:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnorm2:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnormal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lpdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lupdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nnot_a_number:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnum_elements:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#o",
+    "href": "functions-reference/functions_index.html#o",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "ode_adams:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_adams_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_adjoint_tol_ctl:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector (higher-order_functions.html)\n\n\node_bdf:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_bdf_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\noffset_multiplier_constrain:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_jacobian:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_unconstrain:\n\n\n(reals x, reals offset, reals mult) : reals (transform_functions.html)\n\n\none_hot_array:\n\n\n(int n, int k) : array[] real (matrix_operations.html)\n\n\none_hot_int_array:\n\n\n(int n, int k) : array[] int (matrix_operations.html)\n\n\none_hot_row_vector:\n\n\n(int n, int k) : row_vector (matrix_operations.html)\n\n\none_hot_vector:\n\n\n(int K, int k) : vector (matrix_operations.html)\n\n\nones_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nones_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nones_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nones_vector:\n\n\n(int n) : vector (matrix_operations.html)\n\n\noperator!:\n\n\n(int x) : int (real-valued_basic_functions.html)\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\noperator!=:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator%/%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator&&:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\n&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;operator’:\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\noperator*:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : vector (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, row_vector y) : matrix (matrix_operations.html)\n\n\noperator*=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator+:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator+=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator-:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(T x) : T (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (matrix_operations.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator-=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.*:\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator.*=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator./:\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator./=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.^:\n\n\n( complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n( complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator/:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix B, complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector b, complex_matrix A) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\noperator/=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator&lt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&lt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\noperator==:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator\\:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\noperator^:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\noperator||:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\nordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_logistic:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lpmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lupmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lpmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lupmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_probit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_probit_lpmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_lupmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nowens_t:\n\n\n(real h, real a) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#p",
+    "href": "functions-reference/functions_index.html#p",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "pareto:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_cdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lccdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lcdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lpdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lupdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_rng:\n\n\n(reals y_min, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\npareto_type_2:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_cdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lccdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lcdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lpdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lupdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_rng:\n\n\n(reals mu, reals lambda, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\nPhi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nPhi_approx:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\npi:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npoisson:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_cdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lccdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lcdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\npoisson_lpmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lupmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_rng:\n\n\n(reals lambda) : R (unbounded_discrete_distributions.html)\n\n\npolar:\n\n\n(real r, real theta) : complex (complex-valued_basic_functions.html)\n\n\npositive_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npositive_ordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\npow:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : Z (complex-valued_basic_functions.html)\n\n\nprint:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nprod:\n\n\n(array[] int x) : real (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nproj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#q",
+    "href": "functions-reference/functions_index.html#q",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "qr:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_thin_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nquad_form:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquad_form_diag:\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\nquad_form_sym:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquantile:\n\n\n(data array[] real x, data array[] real p) : array[] real (array_operations.html)\n\n\n(data array[] real x, data real p) : real (array_operations.html)\n\n\n(data row_vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data row_vector x, data real p) : real (matrix_operations.html)\n\n\n(data vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data vector x, data real p) : real (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#r",
+    "href": "functions-reference/functions_index.html#r",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "rank:\n\n\n(array[] int v, int s) : int (array_operations.html)\n\n\n(array[] real v, int s) : int (array_operations.html)\n\n\n(row_vector v, int s) : int (matrix_operations.html)\n\n\n(vector v, int s) : int (matrix_operations.html)\n\n\nrayleigh:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nrayleigh_cdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lccdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lcdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lpdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lupdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_rng:\n\n\n(reals sigma) : R (positive_continuous_distributions.html)\n\n\nreduce_sum:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreduce_sum_static:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreject:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nrep_array:\n\n\n(T x, int k, int m, int n) : array[,,] T (array_operations.html)\n\n\n(T x, int m, int n) : array[,] T (array_operations.html)\n\n\n(T x, int n) : array[] T (array_operations.html)\n\n\nrep_matrix:\n\n\n(complex z, int m, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector rv, int m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(real x, int m, int n) : matrix (matrix_operations.html)\n\n\n(row_vector rv, int m) : matrix (matrix_operations.html)\n\n\n(vector v, int n) : matrix (matrix_operations.html)\n\n\nrep_row_vector:\n\n\n(complex z, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(real x, int n) : row_vector (matrix_operations.html)\n\n\nrep_vector:\n\n\n(complex z, int m) : complex_vector (complex_matrix_operations.html)\n\n\n(real x, int m) : vector (matrix_operations.html)\n\n\nreverse:\n\n\n(array[] T v) : array[] T (array_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nrising_factorial:\n\n\n(real x, int n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nround:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nrow:\n\n\n(complex_matrix x, int m) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int m) : row_vector (matrix_operations.html)\n\n\nrows:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nrows_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\nrows_dot_self:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : vector (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#s",
+    "href": "functions-reference/functions_index.html#s",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "scale_matrix_exp_multiply:\n\n\n(real t, matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nscaled_inv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_cdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lccdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lcdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lpdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lupdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_rng:\n\n\n(reals nu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nsd:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsegment:\n\n\n(array[] T sv, int i, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int i, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int i, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int i, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int i, int n) : vector (matrix_operations.html)\n\n\nsimplex_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsingular_values:\n\n\n(complex_matrix A) : vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\nsinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsize:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nskew_double_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nskew_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_normal_cdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lccdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lcdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lpdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lupdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_rng:\n\n\n(reals xi, reals omega, real alpha) : R (unbounded_continuous_distributions.html)\n\n\nsoftmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nsolve_newton:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_newton_tol:\n\n\n(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell_tol:\n\n\n(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsort_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_indices_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsort_indices_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsqrt:\n\n\n(complex x) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsqrt2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nsquare:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsquared_distance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\nstd_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstd_normal_cdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lccdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lcdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_log_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_lpdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lupdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_rng:\n\n\n() : real (unbounded_continuous_distributions.html)\n\n\nstep:\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\nstochastic_column_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstochastic_row_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstudent_t:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstudent_t_cdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lccdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lcdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lpdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lupdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_rng:\n\n\n(reals nu, reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nsub_col:\n\n\n(complex_matrix x, int i, int j, int n_rows) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows) : vector (matrix_operations.html)\n\n\nsub_row:\n\n\n(complex_matrix x, int i, int j, int n_cols) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_cols) : row_vector (matrix_operations.html)\n\n\nsum:\n\n\n(array[] complex x) : complex (array_operations.html)\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsum_to_zero_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsvd:\n\n\n(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector, matrix) (matrix_operations.html)\n\n\nsvd_U:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsvd_V:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsymmetrize_from_lower_tri:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#t",
+    "href": "functions-reference/functions_index.html#t",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "tail:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\ntan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntarget:\n\n\n() : real (real-valued_basic_functions.html)\n\n\ntcrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ntgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nto_array_1d:\n\n\n(array[...] complex a) : array[] complex (mixed_operations.html)\n\n\n(array[...] int a) : array[] int (mixed_operations.html)\n\n\n(array[...] real a) : array[] real (mixed_operations.html)\n\n\n(complex_matrix m) : array[] complex (mixed_operations.html)\n\n\n(complex_row_vector v) : array[] complex (mixed_operations.html)\n\n\n(complex_vector v) : array[] real (mixed_operations.html)\n\n\n(matrix m) : array[] real (mixed_operations.html)\n\n\n(row_vector v) : array[] real (mixed_operations.html)\n\n\n(vector v) : array[] real (mixed_operations.html)\n\n\nto_array_2d:\n\n\n(complex_matrix m) : array[,] real (mixed_operations.html)\n\n\n(matrix m) : array[,] real (mixed_operations.html)\n\n\nto_complex:\n\n\n() : complex (complex-valued_basic_functions.html)\n\n\n(real re) : complex (complex-valued_basic_functions.html)\n\n\n(real re, real im) : complex (complex-valued_basic_functions.html)\n\n\n(T1 re, T2 im) : Z (complex-valued_basic_functions.html)\n\n\nto_int:\n\n\n(data real x) : int (integer-valued_basic_functions.html)\n\n\nto_matrix:\n\n\n(array[,] complex a ) : complex_matrix (mixed_operations.html)\n\n\n(array[,] int a) : matrix (mixed_operations.html)\n\n\n(array[,] real a) : matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex_row_vector vs) : complex_matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] row_vector vs) : matrix (mixed_operations.html)\n\n\n(complex_matrix A, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix m) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix M, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(matrix A, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(matrix m) : matrix (mixed_operations.html)\n\n\n(matrix M, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(vector v) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\nto_row_vector:\n\n\n(array[] complex a) : complex_row_vector (mixed_operations.html)\n\n\n(array[] int a) : row_vector (mixed_operations.html)\n\n\n(array[] real a) : row_vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_row_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(matrix m) : row_vector (mixed_operations.html)\n\n\n(row_vector v) : row_vector (mixed_operations.html)\n\n\n(vector v) : row_vector (mixed_operations.html)\n\n\nto_vector:\n\n\n(array[] complex a) : complex_vector (mixed_operations.html)\n\n\n(array[] int a) : vector (mixed_operations.html)\n\n\n(array[] real a) : vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_vector (mixed_operations.html)\n\n\n(matrix m) : vector (mixed_operations.html)\n\n\n(row_vector v) : vector (mixed_operations.html)\n\n\n(vector v) : vector (mixed_operations.html)\n\n\ntrace:\n\n\n(complex_matrix A) : complex (complex_matrix_operations.html)\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ntrace_dot:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_gen_quad_form:\n\n\n(matrix D ,matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_quad_form:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\ntrigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntrunc:\n\n\n(T x) : R (real-valued_basic_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#u",
+    "href": "functions-reference/functions_index.html#u",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "uniform:\n\n\ndistribution statement (bounded_continuous_distributions.html)\n\n\nuniform_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_rng:\n\n\n(reals alpha, reals beta) : R (bounded_continuous_distributions.html)\n\n\nuniform_simplex:\n\n\n(int n) : vector (matrix_operations.html)\n\n\nunit_vectors_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nupper_bound_constrain:\n\n\n(reals y, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_jacobian:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_unconstrain:\n\n\n(reals x, reals ub) : reals (transform_functions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#v",
+    "href": "functions-reference/functions_index.html#v",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "variance:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nvon_mises:\n\n\ndistribution statement (circular_distributions.html)\n\n\nvon_mises_cdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lccdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lcdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lpdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lupdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_rng:\n\n\n(reals mu, reals kappa) : R (circular_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#w",
+    "href": "functions-reference/functions_index.html#w",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "weibull:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nweibull_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\nwiener:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\nwiener_lccdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lcdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lpdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lupdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\nwishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#y",
+    "href": "functions-reference/functions_index.html#y",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "yule_simon:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nyule_simon_cdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lccdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lcdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/functions_index.html#z",
+    "href": "functions-reference/functions_index.html#z",
+    "title": "Alphabetical Index",
+    "section": "",
+    "text": "zeros_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nzeros_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nzeros_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nzeros_vector:\n\n\n(int n) : vector (matrix_operations.html)",
+    "crumbs": [
+      "Functions Reference",
+      "Appendix",
+      "Alphabetical Index"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html",
+    "href": "functions-reference/higher-order_functions.html",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides a few higher-order functions that act on other functions. In all cases, the function arguments to the higher-order functions are defined as functions within the Stan language and passed by name to the higher-order functions.\n\n\nStan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.\nAn algebraic solver is a higher-order function, i.e. it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section Ordinary Differential Equation (ODE) Solvers and Differential Algebraic Equation (DAE) solver). Ordinary Stan functions do not allow functions as arguments.\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The function must return a vector and takes in, as its first argument, the unknowns \\(y\\) we wish to solve for, also passed as a vector. This argument is followed by additional arguments as specified by the user; we call such arguments variadic arguments and denote them .... The signature of the algebraic system is then:\n vector algebra_system (vector y, ...)\nThere is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword data. In the below example, the last variadic argument, \\(x\\), is restricted to being data:\n  vector algebra_system (vector y, vector theta, data vector x)\nDistinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.\n\n\n\n \n vector solve_newton(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.31\n \n vector solve_newton_tol(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.31\n \n vector solve_powell(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method.\nAvailable since 2.31\n \n vector solve_powell_tol(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.\nAvailable since 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, ...):vector. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\n...: variadic arguments.\n\nThe algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:\n\nscaling_step: for the Newton solver only, the scaled-step stopping tolerance, type real, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is \\(10^{-3}\\).\nrel_tol: for the Powell solver only, the relative tolerance, type real, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is \\(10^{-10}\\).\nfunction_tol: function tolerance for the algebraic solver, type real, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is \\(10^{-6}\\).\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is \\(200\\).\n\nThe difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0 (approximately, within the specified function tolerance).\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\n\n\n\nStan offers two methods to solve algebraic equations. solve_newton and solve_newton_tol use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite (Hindmarsh et al. 2005). For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.\nsolve_powell and solve_powell_tol are based on the Powell hybrid method (Powell 1970), which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library (Guennebaud, Jacob, et al. 2010). This solver is in turn based on the algorithm developed for the package MINPACK-1 (Jorge J. More 1980).\nFor both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see (Gaebler 2021) and (Margossian and Betancourt 2022).\n\n\n\n\n\nStan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).\nSolving an initial value ODE means given a set of differential equations \\(y'(t, \\theta) = f(t, y, \\theta)\\) and initial conditions \\(y(t_0, \\theta)\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 &lt; t_2, \\cdots &lt; t_n\\). \\(f(t, y, \\theta)\\) is referred to here as the ODE system function.\n\\(f(t, y, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.\nTo make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as .... The types of the arguments represented by ... in the ODE solve function call must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector ode_rk45(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.24\n \n\narray[] vector ode_rk45_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.24\n \n\narray[] vector ode_ckrk(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.\nAvailable since 2.27\n \n\narray[] vector ode_ckrk_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.27\n \n\narray[] vector ode_adams(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.24\n \n\narray[] vector ode_adams_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_bdf(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.24\n \n\narray[] vector ode_bdf_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_adjoint_tol_ctl(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)\nSolves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.\nAvailable since 2.27\n\n\n\nThe first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a vector return type, and the first two arguments must be a real and vector in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:\n vector ode(real time, vector state, ...)\nThe ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\n..., sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the ... arguments of the ODE solve function call.\n\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see Arguments to the adjoint ODE solver.\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nFor the versions of the ode solver functions ending in _tol, these three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the ODE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\n\nThe arguments to the adjoint ODE solver are different from those for the other functions (for those see Arguments to the ODE solvers).\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\ndata rel_tol_forward: Relative tolerance for forward solve, type real, data only,\ndata abs_tol_forward: Absolute tolerance vector for each state for forward solve, type vector, data only,\ndata rel_tol_backward: Relative tolerance for backward solve, type real, data only,\ndata abs_tol_backward: Absolute tolerance vector for each state for backward solve, type vector, data only,\ndata rel_tol_quadrature: Relative tolerance for backward quadrature, type real, data only,\ndata abs_tol_quadrature: Absolute tolerance for backward quadrature, type real, data only,\ndata max_num_steps: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type int, data only,\nnum_steps_between_checkpoints: number of steps between checkpointing forward solution, type int, data only,\ninterpolation_polynomial: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type int, data only,\nsolver_forward: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only,\nsolver_backward: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only.\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe return value for the ODE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time in specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.\n\n\n\n\n\n\nStan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 (Serban et al. 2021).\nSolving an initial value DAE means given a set of residual functions \\(r(y'(t, \\theta), y(t, \\theta), t)\\) and initial conditions \\((y(t_0, \\theta), y'(t_0, \\theta))\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 \\leq t_2, \\cdots \\leq t_n\\). The residual function \\(r(y', y, t, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.\nSimilar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as ..., and the types of these arguments, also represented by ... in the DAE solver call, must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector dae(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) Solves the DAE system using the backward differentiation formula (BDF) method (Serban et al. 2021).\nAvailable since 2.29\n \n\narray[] vector dae_tol(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.29\n\n\n\nThe first argument to the DAE solver is the DAE residual function. The DAE residual function must have a vector return type, and the first three arguments must be a real, vector, and vector, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:\n  vector residual(real time, vector state, vector state_derivative, ...)\nThe DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the DAE system\nstate, the state of the DAE system at the time specified\nstate_derivative, the time derivatives of the state of the DAE system at the time specified\n..., sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the ... arguments of the DAE solve function call.\n\n\n\n\nThe arguments to the DAE solver are\n\nresidual: DAE residual function,\ninitial_state: initial state, type vector,\ninitial_state_derivative: time derivative of the initial state, type vector,\ninitial_time: initial time, type data real,\ntimes: solution times, type data array[] real,\n...: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the ... arguments of the DAE residual function.\n\nFor dae_tol, the following three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the DAE solver, type real, data only,\ndata abs_tol: absolute tolerance for the DAE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the DAE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe user is responsible to ensure the residual function becomes zero at the initial time, t0, when the arguments initial_state and initial_state_derivative are introduced as state and state_derivative, respectively.\n\n\n\nThe return value for the DAE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.\n\n\n\n\n\n\nStan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.\nIt operates similarly to the algebraic solver and the ordinary differential equations solver in that it allows as an argument a function.\nLike both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.\n\n\nPerforming a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:\nreal integrand(real x, real xc, array[] real theta,\n               array[] real x_r, array[] int x_i)\nThe function should return the value of the integrand evaluated at the point x.\nThe argument of this function are:\n\nx, the independent variable being integrated over\nxc, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section Precision Loss).\ntheta, parameter values used to evaluate the integral\nx_r, data values used to evaluate the integral\nx_i, integer data used to evaluate the integral\n\nLike algebraic solver and the differential equations solver, the 1D integrator separates parameter values, theta, from data values, x_r.\n\n\n\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) Integrates the integrand from a to b.\nAvailable since 2.23\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance) Integrates the integrand from a to b with the given relative tolerance.\nAvailable since 2.23\n\n\nThe arguments to the 1D integrator are as follows:\n\nintegrand: function literal referring to a function specifying the integrand with signature (real, real, array[] real, array[] real, array[] int):real The arguments represent\n\n\nwhere integrand is evaluated,\n\n\ndistance from evaluation point to integration limit for definite integrals,\n\n\nparameters,\n\n\nreal data\n\n\ninteger data, and the return value is the integrand evaluated at the given point,\n\n\na: left limit of integration, may be negative infinity, type real,\nb: right limit of integration, may be positive infinity, type real,\ntheta: parameters only, type array[] real,\nx_r: real data only, type array[] real,\nx_i: integer data only, type array[] int.\n\nA relative_tolerance argument can optionally be provided for more control over the algorithm:\n\nrelative_tolerance: relative tolerance for the 1d integrator, type real, data only.\n\n\n\n\nThe return value for the 1D integrator is a real, the value of the integral.\n\n\n\nFor numeric stability, integrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). Each integral is separately integrated to the given relative_tolerance.\n\n\n\nWhen integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, 1.0 - x). xc (as passed to the integrand) is a high-precision version of the distance between x and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, xc will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, xc is set to NaN.\n\n\n\nInternally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in (Takahasi and Mori 1974), (Mori 1978), (Bailey, Jeyabalan, and Li 2005), and (Tanaka et al. 2009).\nThe gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.\n\n\n\n\n\nStan provides a higher-order reduce function for summation. A function which returns a scalar g: U -&gt; real is mapped to every element of a list of type array[] U, { x1, x2, ... } and all the results are accumulated,\ng(x1) + g(x2) + ...\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g itself, but instead works through evaluating partial sums, f: array[] U -&gt; real, where:\nf({ x1 }) = g(x1)\nf({ x1, x2 }) = g(x1) + g(x2)\nf({ x1, x2, ... }) = g(x1) + g(x2) + ...\nMathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load). This should result in stable numerical evaluations.\n\n\n\nThe higher-order reduce function takes a partial sum function f, an array argument x (with one array element for each term in the sum), a recommended grainsize, and a set of shared arguments. This representation allows parallelization of the resultant sum.\n \n\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n \n real reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nReturns the equivalent of f(x, 1, size(x), s1, s2, ...), but computes the result in parallel by breaking the array x into independent partial sums. s1, s2, ... are shared between all terms in the sum.\nAvailable since 2.23\n\nf: function literal referring to a function specifying the partial sum operation. Refer to the partial sum function.\nx: array of T, one for each term of the reduction, T can be any type,\ngrainsize: For reduce_sum, grainsize is the recommended size of the partial sum (grainsize = 1 means pick totally automatically). For reduce_sum_static, grainsize determines the maximum size of the partial sums, type int,\ns1: first (optional) shared argument, type T1, where T1 can be any type\ns2: second (optional) shared argument, type T2, where T2 can be any type,\n...: remainder of shared arguments, each of which can be any type.\n\n\n\n\nThe partial sum function must have the following signature where the type T, and the types of all the shared arguments (T1, T2, …) match those of the original reduce_sum (reduce_sum_static) call.\n(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real\nThe partial sum function returns the sum of the start to end terms (inclusive) of the overall calculations. The arguments to the partial sum function are:\n\nx_subset, the subset of x a given partial sum is responsible for computing, type array[] T, where T matches the type of x in reduce_sum (reduce_sum_static)\nstart, the index of the first term of the partial sum, type int\nend, the index of the last term of the partial sum (inclusive), type int\ns1, first shared argument, type T1, matching type of s1 in reduce_sum (reduce_sum_static)\ns2, second shared argument, type T2, matching type of s2 in reduce_sum (reduce_sum_static)\n..., remainder of shared arguments, with types matching those in reduce_sum (reduce_sum_static)\n\n\n\n\n\nStan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.\n\n\nThe function being mapped must have a signature identical to that of the function f in the following declaration.\n vector f(vector phi, vector theta,\n          data array[] real x_r, data array[] int x_i);\nThe map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:\n\nphi, the sequence of parameters shared across shards\ntheta, the sequence of parameters specific to this shard\nx_r, sequence of real-valued data\nx_i, sequence of integer data\n\nAll input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.\n\n\n\nThe rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.\n \n\nvector map_rect(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) Return the concatenation of the results of applying the function f, of type (vector, vector, array[] real, array[] int):vector elementwise, i.e., f(phi, theta[n], x_r[n], x_i[n]) for each n in 1:N, where N is the size of the parallel arrays of job-specific/local parameters theta, real data x_r, and integer data x_r. The shared/global parameters phi are passed to each invocation of f.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html#functions-algebraic-solver",
+    "href": "functions-reference/higher-order_functions.html#functions-algebraic-solver",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.\nAn algebraic solver is a higher-order function, i.e. it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section Ordinary Differential Equation (ODE) Solvers and Differential Algebraic Equation (DAE) solver). Ordinary Stan functions do not allow functions as arguments.\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The function must return a vector and takes in, as its first argument, the unknowns \\(y\\) we wish to solve for, also passed as a vector. This argument is followed by additional arguments as specified by the user; we call such arguments variadic arguments and denote them .... The signature of the algebraic system is then:\n vector algebra_system (vector y, ...)\nThere is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword data. In the below example, the last variadic argument, \\(x\\), is restricted to being data:\n  vector algebra_system (vector y, vector theta, data vector x)\nDistinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.\n\n\n\n \n vector solve_newton(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.31\n \n vector solve_newton_tol(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.31\n \n vector solve_powell(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method.\nAvailable since 2.31\n \n vector solve_powell_tol(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.\nAvailable since 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, ...):vector. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\n...: variadic arguments.\n\nThe algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:\n\nscaling_step: for the Newton solver only, the scaled-step stopping tolerance, type real, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is \\(10^{-3}\\).\nrel_tol: for the Powell solver only, the relative tolerance, type real, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is \\(10^{-10}\\).\nfunction_tol: function tolerance for the algebraic solver, type real, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is \\(10^{-6}\\).\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is \\(200\\).\n\nThe difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0 (approximately, within the specified function tolerance).\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\n\n\n\nStan offers two methods to solve algebraic equations. solve_newton and solve_newton_tol use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite (Hindmarsh et al. 2005). For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.\nsolve_powell and solve_powell_tol are based on the Powell hybrid method (Powell 1970), which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library (Guennebaud, Jacob, et al. 2010). This solver is in turn based on the algorithm developed for the package MINPACK-1 (Jorge J. More 1980).\nFor both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see (Gaebler 2021) and (Margossian and Betancourt 2022).",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html#functions-ode-solver",
+    "href": "functions-reference/higher-order_functions.html#functions-ode-solver",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).\nSolving an initial value ODE means given a set of differential equations \\(y'(t, \\theta) = f(t, y, \\theta)\\) and initial conditions \\(y(t_0, \\theta)\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 &lt; t_2, \\cdots &lt; t_n\\). \\(f(t, y, \\theta)\\) is referred to here as the ODE system function.\n\\(f(t, y, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.\nTo make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as .... The types of the arguments represented by ... in the ODE solve function call must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector ode_rk45(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.24\n \n\narray[] vector ode_rk45_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.24\n \n\narray[] vector ode_ckrk(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.\nAvailable since 2.27\n \n\narray[] vector ode_ckrk_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.27\n \n\narray[] vector ode_adams(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.24\n \n\narray[] vector ode_adams_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_bdf(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.24\n \n\narray[] vector ode_bdf_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_adjoint_tol_ctl(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)\nSolves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.\nAvailable since 2.27\n\n\n\nThe first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a vector return type, and the first two arguments must be a real and vector in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:\n vector ode(real time, vector state, ...)\nThe ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\n..., sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the ... arguments of the ODE solve function call.\n\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see Arguments to the adjoint ODE solver.\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nFor the versions of the ode solver functions ending in _tol, these three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the ODE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\n\nThe arguments to the adjoint ODE solver are different from those for the other functions (for those see Arguments to the ODE solvers).\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\ndata rel_tol_forward: Relative tolerance for forward solve, type real, data only,\ndata abs_tol_forward: Absolute tolerance vector for each state for forward solve, type vector, data only,\ndata rel_tol_backward: Relative tolerance for backward solve, type real, data only,\ndata abs_tol_backward: Absolute tolerance vector for each state for backward solve, type vector, data only,\ndata rel_tol_quadrature: Relative tolerance for backward quadrature, type real, data only,\ndata abs_tol_quadrature: Absolute tolerance for backward quadrature, type real, data only,\ndata max_num_steps: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type int, data only,\nnum_steps_between_checkpoints: number of steps between checkpointing forward solution, type int, data only,\ninterpolation_polynomial: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type int, data only,\nsolver_forward: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only,\nsolver_backward: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only.\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe return value for the ODE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time in specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html#functions-dae-solver",
+    "href": "functions-reference/higher-order_functions.html#functions-dae-solver",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 (Serban et al. 2021).\nSolving an initial value DAE means given a set of residual functions \\(r(y'(t, \\theta), y(t, \\theta), t)\\) and initial conditions \\((y(t_0, \\theta), y'(t_0, \\theta))\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 \\leq t_2, \\cdots \\leq t_n\\). The residual function \\(r(y', y, t, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.\nSimilar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as ..., and the types of these arguments, also represented by ... in the DAE solver call, must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector dae(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) Solves the DAE system using the backward differentiation formula (BDF) method (Serban et al. 2021).\nAvailable since 2.29\n \n\narray[] vector dae_tol(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.29\n\n\n\nThe first argument to the DAE solver is the DAE residual function. The DAE residual function must have a vector return type, and the first three arguments must be a real, vector, and vector, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:\n  vector residual(real time, vector state, vector state_derivative, ...)\nThe DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the DAE system\nstate, the state of the DAE system at the time specified\nstate_derivative, the time derivatives of the state of the DAE system at the time specified\n..., sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the ... arguments of the DAE solve function call.\n\n\n\n\nThe arguments to the DAE solver are\n\nresidual: DAE residual function,\ninitial_state: initial state, type vector,\ninitial_state_derivative: time derivative of the initial state, type vector,\ninitial_time: initial time, type data real,\ntimes: solution times, type data array[] real,\n...: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the ... arguments of the DAE residual function.\n\nFor dae_tol, the following three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the DAE solver, type real, data only,\ndata abs_tol: absolute tolerance for the DAE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the DAE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe user is responsible to ensure the residual function becomes zero at the initial time, t0, when the arguments initial_state and initial_state_derivative are introduced as state and state_derivative, respectively.\n\n\n\nThe return value for the DAE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html#functions-1d-integrator",
+    "href": "functions-reference/higher-order_functions.html#functions-1d-integrator",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.\nIt operates similarly to the algebraic solver and the ordinary differential equations solver in that it allows as an argument a function.\nLike both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.\n\n\nPerforming a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:\nreal integrand(real x, real xc, array[] real theta,\n               array[] real x_r, array[] int x_i)\nThe function should return the value of the integrand evaluated at the point x.\nThe argument of this function are:\n\nx, the independent variable being integrated over\nxc, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section Precision Loss).\ntheta, parameter values used to evaluate the integral\nx_r, data values used to evaluate the integral\nx_i, integer data used to evaluate the integral\n\nLike algebraic solver and the differential equations solver, the 1D integrator separates parameter values, theta, from data values, x_r.\n\n\n\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) Integrates the integrand from a to b.\nAvailable since 2.23\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance) Integrates the integrand from a to b with the given relative tolerance.\nAvailable since 2.23\n\n\nThe arguments to the 1D integrator are as follows:\n\nintegrand: function literal referring to a function specifying the integrand with signature (real, real, array[] real, array[] real, array[] int):real The arguments represent\n\n\nwhere integrand is evaluated,\n\n\ndistance from evaluation point to integration limit for definite integrals,\n\n\nparameters,\n\n\nreal data\n\n\ninteger data, and the return value is the integrand evaluated at the given point,\n\n\na: left limit of integration, may be negative infinity, type real,\nb: right limit of integration, may be positive infinity, type real,\ntheta: parameters only, type array[] real,\nx_r: real data only, type array[] real,\nx_i: integer data only, type array[] int.\n\nA relative_tolerance argument can optionally be provided for more control over the algorithm:\n\nrelative_tolerance: relative tolerance for the 1d integrator, type real, data only.\n\n\n\n\nThe return value for the 1D integrator is a real, the value of the integral.\n\n\n\nFor numeric stability, integrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). Each integral is separately integrated to the given relative_tolerance.\n\n\n\nWhen integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, 1.0 - x). xc (as passed to the integrand) is a high-precision version of the distance between x and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, xc will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, xc is set to NaN.\n\n\n\nInternally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in (Takahasi and Mori 1974), (Mori 1978), (Bailey, Jeyabalan, and Li 2005), and (Tanaka et al. 2009).\nThe gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html#functions-reduce",
+    "href": "functions-reference/higher-order_functions.html#functions-reduce",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides a higher-order reduce function for summation. A function which returns a scalar g: U -&gt; real is mapped to every element of a list of type array[] U, { x1, x2, ... } and all the results are accumulated,\ng(x1) + g(x2) + ...\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g itself, but instead works through evaluating partial sums, f: array[] U -&gt; real, where:\nf({ x1 }) = g(x1)\nf({ x1, x2 }) = g(x1) + g(x2)\nf({ x1, x2, ... }) = g(x1) + g(x2) + ...\nMathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load). This should result in stable numerical evaluations.\n\n\n\nThe higher-order reduce function takes a partial sum function f, an array argument x (with one array element for each term in the sum), a recommended grainsize, and a set of shared arguments. This representation allows parallelization of the resultant sum.\n \n\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n \n real reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nReturns the equivalent of f(x, 1, size(x), s1, s2, ...), but computes the result in parallel by breaking the array x into independent partial sums. s1, s2, ... are shared between all terms in the sum.\nAvailable since 2.23\n\nf: function literal referring to a function specifying the partial sum operation. Refer to the partial sum function.\nx: array of T, one for each term of the reduction, T can be any type,\ngrainsize: For reduce_sum, grainsize is the recommended size of the partial sum (grainsize = 1 means pick totally automatically). For reduce_sum_static, grainsize determines the maximum size of the partial sums, type int,\ns1: first (optional) shared argument, type T1, where T1 can be any type\ns2: second (optional) shared argument, type T2, where T2 can be any type,\n...: remainder of shared arguments, each of which can be any type.\n\n\n\n\nThe partial sum function must have the following signature where the type T, and the types of all the shared arguments (T1, T2, …) match those of the original reduce_sum (reduce_sum_static) call.\n(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real\nThe partial sum function returns the sum of the start to end terms (inclusive) of the overall calculations. The arguments to the partial sum function are:\n\nx_subset, the subset of x a given partial sum is responsible for computing, type array[] T, where T matches the type of x in reduce_sum (reduce_sum_static)\nstart, the index of the first term of the partial sum, type int\nend, the index of the last term of the partial sum (inclusive), type int\ns1, first shared argument, type T1, matching type of s1 in reduce_sum (reduce_sum_static)\ns2, second shared argument, type T2, matching type of s2 in reduce_sum (reduce_sum_static)\n..., remainder of shared arguments, with types matching those in reduce_sum (reduce_sum_static)",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/higher-order_functions.html#functions-map",
+    "href": "functions-reference/higher-order_functions.html#functions-map",
+    "title": "Higher-Order Functions",
+    "section": "",
+    "text": "Stan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.\n\n\nThe function being mapped must have a signature identical to that of the function f in the following declaration.\n vector f(vector phi, vector theta,\n          data array[] real x_r, data array[] int x_i);\nThe map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:\n\nphi, the sequence of parameters shared across shards\ntheta, the sequence of parameters specific to this shard\nx_r, sequence of real-valued data\nx_i, sequence of integer data\n\nAll input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.\n\n\n\nThe rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.\n \n\nvector map_rect(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) Return the concatenation of the results of applying the function f, of type (vector, vector, array[] real, array[] int):vector elementwise, i.e., f(phi, theta[n], x_r[n], x_i[n]) for each n in 1:N, where N is the size of the parallel arrays of job-specific/local parameters theta, real data x_r, and integer data x_r. The shared/global parameters phi are passed to each invocation of f.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/integer-valued_basic_functions.html",
+    "href": "functions-reference/integer-valued_basic_functions.html",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "This chapter describes Stan’s built-in function that take various types of arguments and return integer values.\n\n\nStan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in 2 + 2, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in 2.0 + 2 and 2 + 2.0, then the integer is promoted to a floating point value and floating-point arithmetic is used.\nInteger arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.\nSecondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, 1 / 2 evaluates to 0 and 5 / 3 evaluates to 1.\nIf one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have 1 / 2 and -1 / 2 evaluate to 0, -7 / 2 evaluate to -3, and 7 / 2 evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.\nUnlike floating point division, where 1.0 / 0.0 produces the special positive infinite value, integer division by zero, as in 1 / 0, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.\n\n\nOperators are described using the C++ syntax. For instance, the binary operator of addition, written X + Y, would have the Stan signature int operator+(int, int) indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.\n \n\nint operator+(int x, int y) The sum of the addends x and y \\[\\begin{equation*} \\text{operator+}(x,y) = (x + y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator-(int x, int y) The difference between the minuend x and subtrahend y \\[\\begin{equation*}\n\\text{operator-}(x,y) = (x - y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator*(int x, int y) The product of the factors x and y \\[\\begin{equation*} \\text{operator*}(x,y) = (x\n\\times y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator/(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator/}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\] deprecated; - use operator%/% instead.\nAvailable since 2.0, deprecated in 2.24\n \n\nint operator%/%(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator\\%/\\%}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\]\nAvailable since 2.24\n \n\nint operator%(int x, int y) x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. \\[\\begin{equation*} \\mathrm{operator\\%}(x, y) \\ =\n\\ x \\ \\text{mod} \\ y \\ = \\ x - y * \\lfloor x / y \\rfloor \\end{equation*}\\]\nAvailable since 2.13\n\n\n\n \n\nint operator-(int x) The negation of the subtrahend x \\[\\begin{equation*} \\text{operator-}(x) = -x \\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of integers, -x is the same shape array where each individual integer is negated.\nAvailable since 2.31\n \n\nint operator+(int x) This is a no-op. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is int, or an array of ints, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nint int_step(int x)\n \n\nint int_step(real x) Return the step function of x as an integer, \\[\\begin{equation*} \\mathrm{int\\_step}(x)\n= \\begin{cases} 1 & \\text{if } x &gt; 0 \\\\ 0 & \\text{if } x \\leq 0 \\text{\nor } x \\text{ is } NaN \\end{cases} \\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nSee the warning in section step functions about the dangers of step functions applied to anything other than data.\nAvailable since 2.0\n\n\n\n \n\nint min(int x, int y) Return the minimum of x and y. \\[\\begin{equation*} \\text{min}(x, y) = \\begin{cases} x &\n\\text{if } x &lt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint max(int x, int y) Return the maximum of x and y. \\[\\begin{equation*} \\text{max}(x, y) = \\begin{cases} x &\n\\text{if } x &gt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n \n\nint size(int x)\n \n\nint size(real x)\nReturn the size of x which for scalar-valued x is 1\nAvailable since 2.26\n\n\n\nIt is possible to cast real numbers to integers as long as the real value is data. See data only qualifiers in the Stan Reference Manual.\n \n\nint to_int(data real x)\nReturn the value x truncated to an integer. This will throw an error if the value of x is too big to represent as a 32-bit signed integer.\nThis is similar to trunc (see Rounding functions) but the return type is of type int. For example, to_int(3.9) is 3, and to_int(-3.9) is -3.\nAvailable since 2.31\n \n\nI to_int(data T x)\nThe vectorized version of to_int. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.\nAvailable since 2.31",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/integer-valued_basic_functions.html#int-arithmetic",
+    "href": "functions-reference/integer-valued_basic_functions.html#int-arithmetic",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "Stan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in 2 + 2, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in 2.0 + 2 and 2 + 2.0, then the integer is promoted to a floating point value and floating-point arithmetic is used.\nInteger arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.\nSecondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, 1 / 2 evaluates to 0 and 5 / 3 evaluates to 1.\nIf one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have 1 / 2 and -1 / 2 evaluate to 0, -7 / 2 evaluate to -3, and 7 / 2 evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.\nUnlike floating point division, where 1.0 / 0.0 produces the special positive infinite value, integer division by zero, as in 1 / 0, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.\n\n\nOperators are described using the C++ syntax. For instance, the binary operator of addition, written X + Y, would have the Stan signature int operator+(int, int) indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.\n \n\nint operator+(int x, int y) The sum of the addends x and y \\[\\begin{equation*} \\text{operator+}(x,y) = (x + y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator-(int x, int y) The difference between the minuend x and subtrahend y \\[\\begin{equation*}\n\\text{operator-}(x,y) = (x - y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator*(int x, int y) The product of the factors x and y \\[\\begin{equation*} \\text{operator*}(x,y) = (x\n\\times y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator/(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator/}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\] deprecated; - use operator%/% instead.\nAvailable since 2.0, deprecated in 2.24\n \n\nint operator%/%(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator\\%/\\%}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\]\nAvailable since 2.24\n \n\nint operator%(int x, int y) x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. \\[\\begin{equation*} \\mathrm{operator\\%}(x, y) \\ =\n\\ x \\ \\text{mod} \\ y \\ = \\ x - y * \\lfloor x / y \\rfloor \\end{equation*}\\]\nAvailable since 2.13\n\n\n\n \n\nint operator-(int x) The negation of the subtrahend x \\[\\begin{equation*} \\text{operator-}(x) = -x \\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of integers, -x is the same shape array where each individual integer is negated.\nAvailable since 2.31\n \n\nint operator+(int x) This is a no-op. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/integer-valued_basic_functions.html#absolute-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#absolute-functions",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "T abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is int, or an array of ints, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nint int_step(int x)\n \n\nint int_step(real x) Return the step function of x as an integer, \\[\\begin{equation*} \\mathrm{int\\_step}(x)\n= \\begin{cases} 1 & \\text{if } x &gt; 0 \\\\ 0 & \\text{if } x \\leq 0 \\text{\nor } x \\text{ is } NaN \\end{cases} \\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nSee the warning in section step functions about the dangers of step functions applied to anything other than data.\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/integer-valued_basic_functions.html#bound-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#bound-functions",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "int min(int x, int y) Return the minimum of x and y. \\[\\begin{equation*} \\text{min}(x, y) = \\begin{cases} x &\n\\text{if } x &lt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint max(int x, int y) Return the maximum of x and y. \\[\\begin{equation*} \\text{max}(x, y) = \\begin{cases} x &\n\\text{if } x &gt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/integer-valued_basic_functions.html#size-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#size-functions",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "int size(int x)\n \n\nint size(real x)\nReturn the size of x which for scalar-valued x is 1\nAvailable since 2.26",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/integer-valued_basic_functions.html#casting-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#casting-functions",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "It is possible to cast real numbers to integers as long as the real value is data. See data only qualifiers in the Stan Reference Manual.\n \n\nint to_int(data real x)\nReturn the value x truncated to an integer. This will throw an error if the value of x is too big to represent as a 32-bit signed integer.\nThis is similar to trunc (see Rounding functions) but the return type is of type int. For example, to_int(3.9) is 3, and to_int(-3.9) is -3.\nAvailable since 2.31\n \n\nI to_int(data T x)\nThe vectorized version of to_int. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.\nAvailable since 2.31",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html",
+    "href": "functions-reference/matrix_operations.html",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "int num_elements(vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.5\n \n\nint num_elements(row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.5\n \n\nint num_elements(matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.5\n \n\nint rows(vector x) The number of rows in the vector x\nAvailable since 2.0\n \n\nint rows(row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.0\n \n\nint rows(matrix x) The number of rows in the matrix x\nAvailable since 2.0\n \n\nint cols(vector x) The number of columns in the vector x, namely 1\nAvailable since 2.0\n \n\nint cols(row_vector x) The number of columns in the row vector x\nAvailable since 2.0\n \n\nint cols(matrix x) The number of columns in the matrix x\nAvailable since 2.0\n \n\nint size(vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15\nAvailable since 2.26\n\n\n\nStan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\nvector operator-(vector x) The negation of the vector x.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x) The negation of the row vector x.\nAvailable since 2.0\n \n\nmatrix operator-(matrix x) The negation of the matrix x.\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\nvector operator+(vector x, vector y) The sum of the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, row_vector y) The sum of the row vectors x and y.\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, matrix y) The sum of the matrices x and y\nAvailable since 2.0\n \n\nvector operator-(vector x, vector y) The difference between the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, row_vector y) The difference between the row vectors x and y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, matrix y) The difference between the matrices x and y\nAvailable since 2.0\n \n\nvector operator*(real x, vector y) The product of the scalar x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(real x, row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.0\n \n\nmatrix operator*(real x, matrix y) The product of the scalar x and the matrix y\nAvailable since 2.0\n \n\nvector operator*(vector x, real y) The product of the scalar y and vector x\nAvailable since 2.0\n \n\nmatrix operator*(vector x, row_vector y) The product of the vector x and row vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, real y) The product of the scalar y and row vector x\nAvailable since 2.0\n \n\nreal operator*(row_vector x, vector y) The product of the row vector x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, matrix y) The product of the row vector x and matrix y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, real y) The product of the scalar y and matrix x\nAvailable since 2.0\n \n\nvector operator*(matrix x, vector y) The product of the matrix x and vector y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, matrix y) The product of the matrices x and y\nAvailable since 2.0\n\n\n\n \n\nvector operator+(vector x, real y) The result of adding y to every entry in the vector x\nAvailable since 2.0\n \n\nvector operator+(real x, vector y) The result of adding x to every entry in the vector y\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, real y) The result of adding y to every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator+(real x, row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, real y) The result of adding y to every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator+(real x, matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.0\n \n\nvector operator-(vector x, real y) The result of subtracting y from every entry in the vector x\nAvailable since 2.0\n \n\nvector operator-(real x, vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, real y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator-(real x, row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, real y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator-(real x, matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.0\n \n\nvector operator/(vector x, real y) The result of dividing each entry in the vector x by y\nAvailable since 2.0\n \n\nrow_vector operator/(row_vector x, real y) The result of dividing each entry in the row vector x by y\nAvailable since 2.0\n \n\nmatrix operator/(matrix x, real y) The result of dividing each entry in the matrix x by y\nAvailable since 2.0\n\n\n\n\nMatrix transposition is represented using a postfix operator.\n \n\nmatrix operator'(matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.0\n \n\nrow_vector operator'(vector x) The transpose of the vector x, written as x'\nAvailable since 2.0\n \n\nvector operator'(row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.0\n\n\n\nElementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section function vectorization for the general cases.\n \n\nvector operator.*(vector x, vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nrow_vector operator.*(row_vector x, row_vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nmatrix operator.*(matrix x, matrix y) The elementwise product of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator./(real x, vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(row_vector x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nrow_vector operator./(row_vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(real x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(matrix x, matrix y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nmatrix operator./(matrix x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(real x, matrix y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator.^(vector x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(real x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(real x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(real x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n\n\n\n \n\nreal dot_product(vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(vector x, vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(row_vector x, row_vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(matrix x, matrix y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(vector x, vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(row_vector x, row_vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(matrix x, matrix y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nreal dot_self(vector x) The dot product of the vector x with itself\nAvailable since 2.0\n \n\nreal dot_self(row_vector x) The dot product of the row vector x with itself\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(matrix x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(matrix x) The dot product of the rows of x with themselves\nAvailable since 2.0\n\n\n \n\nmatrix tcrossprod(matrix x) The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}\\,\\text{x}^{\\top}\\).\nAvailable since 2.0\n \n\nmatrix crossprod(matrix x) The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}^{\\top}\\,\\text{x}\\).\nAvailable since 2.0\nThe following functions all provide shorthand forms for common expressions, which are also much more efficient.\n \n\nmatrix quad_form(matrix A, matrix B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nreal quad_form(matrix A, vector B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nmatrix quad_form_diag(matrix m, vector v) The quadratic form using the column vector v as a diagonal matrix, i.e., diag_matrix(v) * m * diag_matrix(v).\nAvailable since 2.3\n \n\nmatrix quad_form_diag(matrix m, row_vector rv) The quadratic form using the row vector rv as a diagonal matrix, i.e., diag_matrix(rv) * m * diag_matrix(rv).\nAvailable since 2.3\n \n\nmatrix quad_form_sym(matrix A, matrix B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal quad_form_sym(matrix A, vector B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal trace_dot(matrix A, matrix B) The trace of the matrix product, i.e., trace(A * B).\nAvailable since 2.39\n \n\nreal trace_quad_form(matrix A, matrix B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_quad_form(matrix A, vector B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_gen_quad_form(matrix D, matrix A, matrix B) The trace of a generalized quadratic form, i.e., trace(D * B' * A * B).\nAvailable since 2.0\n \n\nmatrix multiply_lower_tri_self_transpose(matrix x) The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if L is a matrix of the same dimensions as x with L(m,n) equal to x(m,n) for \\(\\text{n}\n\\leq \\text{m}\\) and L(m,n) equal to 0 if \\(\\text{n} &gt; \\text{m}\\), the result is the symmetric matrix \\(\\text{L}\\,\\text{L}^{\\top}\\). This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(vector v, matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(row_vector rv, matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, row_vector rv) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.0\n\n\n\n\n\n\n \n\nreal log_sum_exp(vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(row_vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(matrix x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n\n\n\n \n\nreal min(vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(row_vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(matrix x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(row_vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(matrix x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n\n\n\n \n\nreal sum(vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal prod(vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n\n\n\nFull definitions are provided for sample moments in section array reductions.\n \n\nreal mean(vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(row_vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(matrix x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(row_vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(matrix x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(row_vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(matrix x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n \n\nreal quantile(data row_vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data row_vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n\n\n\n\nThe following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\nvector rep_vector(real x, int m) Return the size m (column) vector consisting of copies of x.\nAvailable since 2.0\n \n\nrow_vector rep_row_vector(real x, int n) Return the size n row vector consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(real x, int m, int n) Return the m by n matrix consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.0\n \n\nmatrix rep_matrix(row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.0\nUnlike the situation with array broadcasting (see section array broadcasting), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.\n vector[3] x;\n x = rep_vector(1, 3);\n x = rep_vector(1.0, 3);\nThere are no integer vector or matrix types, so integer values are automatically promoted.\n\n\n \n\nmatrix symmetrize_from_lower_tri(matrix A)\nConstruct a symmetric matrix from the lower triangle of A.\nAvailable since 2.26\n\n\n\n\n \n\nmatrix add_diag(matrix m, row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, real d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.21\n \n\nvector diagonal(matrix x) The diagonal of the matrix x\nAvailable since 2.0\n \n\nmatrix diag_matrix(vector x) The diagonal matrix with diagonal x\nAvailable since 2.0\nAlthough the diag_matrix function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,\n y ~ multi_normal(mu, diag_matrix(square(sigma)));\nit is much more efficient to just use a univariate normal, which produces the same density,\n y ~ normal(mu, sigma);\nRather than writing m * diag_matrix(v) where m is a matrix and v is a vector, it is much more efficient to write diag_post_multiply(m, v) (and similarly for pre-multiplication). By the same token, it is better to use quad_form_diag(m, v) rather than quad_form(m, diag_matrix(v)).\n \n\nmatrix identity_matrix(int k) Create an identity matrix of size \\(k \\times k\\)\nAvailable since 2.26\n\n\n\n \n\narray[] real linspaced_array(int n, data real lower, data real upper) Create a real array of length n of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int linspaced_int_array(int n, int lower, int upper) Create a regularly spaced, increasing integer array of length n between lower and upper, inclusively. If (upper - lower) / (n - 1) is less than one, repeat each output (n - 1) / (upper - lower) times. If neither (upper - lower) / (n - 1) or (n - 1) / (upper - lower) are integers, upper is reduced until one of these is true.\nAvailable since 2.26\n \n\nvector linspaced_vector(int n, data real lower, data real upper) Create an n-dimensional vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\nrow_vector linspaced_row_vector(int n, data real lower, data real upper) Create an n-dimensional row-vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int one_hot_int_array(int n, int k) Create a one-hot encoded int array of length n with array[k] = 1\nAvailable since 2.26\n \n\narray[] real one_hot_array(int n, int k) Create a one-hot encoded real array of length n with array[k] = 1\nAvailable since 2.24\n \n\nvector one_hot_vector(int n, int k) Create an n-dimensional one-hot encoded vector with vector[k] = 1\nAvailable since 2.24\n \n\nrow_vector one_hot_row_vector(int n, int k) Create an n-dimensional one-hot encoded row-vector with row_vector[k] = 1\nAvailable since 2.24\n \n\narray[] int ones_int_array(int n) Create an int array of length n of all ones\nAvailable since 2.26\n \n\narray[] real ones_array(int n) Create a real array of length n of all ones\nAvailable since 2.26\n \n\nvector ones_vector(int n) Create an n-dimensional vector of all ones\nAvailable since 2.26\n \n\nrow_vector ones_row_vector(int n) Create an n-dimensional row-vector of all ones\nAvailable since 2.26\n \n\narray[] int zeros_int_array(int n) Create an int array of length n of all zeros\nAvailable since 2.26\n \n\narray[] real zeros_array(int n) Create a real array of length n of all zeros\nAvailable since 2.24\n \n\nvector zeros_vector(int n) Create an n-dimensional vector of all zeros\nAvailable since 2.24\n \n\nrow_vector zeros_row_vector(int n) Create an n-dimensional row-vector of all zeros\nAvailable since 2.24\n \n\nvector uniform_simplex(int n) Create an n-dimensional simplex with elements vector[i] = 1 / n for all \\(i \\in 1, \\dots, n\\)\nAvailable since 2.24\n\n\n\nStan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\nvector col(matrix x, int n) The n-th column of matrix x\nAvailable since 2.0\n \n\nrow_vector row(matrix x, int m) The m-th row of matrix x\nAvailable since 2.0\nThe row function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation x[m] is just an alternative way of writing row(x,m). The col function may not, be used as an lvalue, nor is there an indexing based shorthand for it.\n\n\n\n\n\nBlock operations may be used to extract a sub-block of a matrix.\n \n\nmatrix block(matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.0\nThe sub-row and sub-column operations may be used to extract a slice of row or column from a matrix\n \n\nvector sub_col(matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.0\n \n\nrow_vector sub_row(matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.0\n\n\n\nThe head operation extracts the first \\(n\\) elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.\n \n\nvector head(vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.0\n \n\nrow_vector head(row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.0\n \n\narray[] T head(array[] T sv, int n) Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector tail(vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.0\n \n\nrow_vector tail(row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.0\n \n\narray[] T tail(array[] T sv, int n) Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector segment(vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.0\n \n\nrow_vector segment(row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.10\n \n\narray[] T segment(array[] T sv, int i, int n) Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n\n\n\n\n\nStan’s matrix and vector concatenation operations append_col and append_row are like the operations cbind and rbind in R.\n\n\n \n\nmatrix append_col(matrix x, matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(matrix x, vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.5\n \n\nrow_vector append_col(row_vector x, row_vector y) Combine row vectors x and y of any size into another row vector by appending y to the end of x.\nAvailable since 2.5\n \n\nrow_vector append_col(real x, row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.12\n \n\nrow_vector append_col(row_vector x, real y) Append y to the end of x, returning another row vector.\nAvailable since 2.12\n\n\n\n \n\nmatrix append_row(matrix x, matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(matrix x, row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.5\n \n\nvector append_row(vector x, vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.5\n \n\nvector append_row(real x, vector y) Append x to the top of y, returning another vector.\nAvailable since 2.12\n \n\nvector append_row(vector x, real y) Append y to the bottom of x, returning another vector.\nAvailable since 2.12\n\n\n\n\n\n\nThe softmax function maps1 \\(y \\in \\mathbb{R}^K\\) to the \\(K\\)-simplex by \\[\\begin{equation*} \\text{softmax}(y)  = \\frac{\\exp(y)}\n{\\sum_{k=1}^K \\exp(y_k)}, \\end{equation*}\\] where \\(\\exp(y)\\) is the componentwise exponentiation of \\(y\\). Softmax is usually calculated on the log scale, \\[\\begin{eqnarray*} \\log \\text{softmax}(y) & = & \\ y - \\log \\sum_{k=1}^K\n\\exp(y_k) \\\\[4pt] & = & y - \\mathrm{log\\_sum\\_exp}(y). \\end{eqnarray*}\\] where the vector \\(y\\) minus the scalar \\(\\mathrm{log\\_sum\\_exp}(y)\\) subtracts the scalar from each component of \\(y\\).\nStan provides the following functions for softmax and its log.\n \n\nvector softmax(vector x) The softmax of x\nAvailable since 2.0\n \n\nvector log_softmax(vector x) The natural logarithm of the softmax of x\nAvailable since 2.0\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] int cumulative_sum(array[] int x) The cumulative sum of x\nAvailable since 2.30\n \n\narray[] real cumulative_sum(array[] real x) The cumulative sum of x\nAvailable since 2.0\n \n\nvector cumulative_sum(vector v) The cumulative sum of v\nAvailable since 2.0\n \n\nrow_vector cumulative_sum(row_vector rv) The cumulative sum of rv\nAvailable since 2.0\n\n\n\n\nThe Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.\nFor one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by \\(K_{ij} = k(x_i, x_j)\\) (where \\(x_i\\) is the \\(i^{th}\\) element of the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(x_i, y_j)\\).\nFor multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{x}_j)\\) (where \\(\\mathbf{x}_i\\) is the \\(i^{th}\\) vector in the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{y}_j)\\).\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|^2}{2l^2} \\right)\n\\]\n \n\nmatrix gp_exp_quad_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith bias \\(\\sigma_0\\) the dot product kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma_0^2 + \\mathbf{x}_i^T \\mathbf{x}_j\n\\]\n \n\nmatrix gp_dot_prod_cov(array[] real x, real sigma)\nGaussian process covariance with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(array[] real x1, array[] real x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x, real sigma)\nGaussian process covariance with dot product kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x1, vectors x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in multiple dimensions.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponential kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_exponential_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 3/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right) \\exp \\left( -\\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern32_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 5/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{5}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} + \\frac{5 |\\mathbf{x}_i - \\mathbf{x}_j|^2}{3l^2} \\right)\n\\exp \\left( -\\frac{\\sqrt{5} |\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern52_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\), length scale \\(l\\), and period \\(p\\), the periodic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left(-\\frac{2 \\sin^2 \\left( \\pi \\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{p} \\right) }{l^2} \\right)\n\\]\n \n\nmatrix gp_periodic_cov(array[] real x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(array[] real x1, array[] real x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x1, vectors x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\n\n\n\nIn general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.\n\n\n \n\nrow_vector operator/(row_vector b, matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.0\n \n\nmatrix operator/(matrix B, matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.5\n \n\nvector operator\\(matrix A, vector b) The left division of A by b; equivalently inverse(A) * b\nAvailable since 2.18\n \n\nmatrix operator\\(matrix A, matrix B) The left division of A by B; equivalently inverse(A) * B\nAvailable since 2.18\n\n\n\nThere are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix \\(\\text{tri}(A)\\) is used in the definitions and defined by \\[\\begin{equation*} \\text{tri}(A)[m,n] = \\left\\{\n\\begin{array}{ll} A[m,n] & \\text{if } m \\geq n, \\text{ and} \\\\[4pt] 0\n& \\text{otherwise}. \\end{array} \\right. \\end{equation*}\\] When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.\n \n\nvector mdivide_left_tri_low(matrix A, vector b) The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * b, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_left_tri_low(matrix A, matrix B) The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * B, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n \n\nrow_vector mdivide_right_tri_low(row_vector b, matrix A) The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form b * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_right_tri_low(matrix B, matrix A) The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form B * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n\n\n\n\nThere are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.\n \n\nmatrix mdivide_left_spd(matrix A, vector b) The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * b.\nAvailable since 2.12\n \n\nvector mdivide_left_spd(matrix A, matrix B) The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * B.\nAvailable since 2.12\n \n\nrow_vector mdivide_right_spd(row_vector b, matrix A) The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form b *inverse(A).\nAvailable since 2.12\n \n\nmatrix mdivide_right_spd(matrix B, matrix A) The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form B * inverse(A).\nAvailable since 2.12\n\n\n\nThe exponential of the matrix \\(A\\) is formally defined by the convergent power series: \\[\\begin{equation*} e^A = \\sum_{n=0}^{\\infty} \\dfrac{A^n}{n!}\n\\end{equation*}\\]\n \n\nmatrix matrix_exp(matrix A) The matrix exponential of A\nAvailable since 2.13\n \n\nmatrix matrix_exp_multiply(matrix A, matrix B) The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form matrix_exp(A) * B.\nAvailable since 2.18\n \n\nmatrix scale_matrix_exp_multiply(real t, matrix A, matrix B) The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form matrix_exp(t * A) * B.\nAvailable since 2.18\n\n\n\nReturns the nth power of the specific matrix: \\[\\begin{equation*} M^n = M_1 * ... * M_n \\end{equation*}\\]\n \n\nmatrix matrix_power(matrix A, int B) Matrix A raised to the power B.\nAvailable since 2.24\n\n\n\n\n\n \n\nreal trace(matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.0\n\n\n\n \n\nreal determinant(matrix A) The determinant of A\nAvailable since 2.0\n \n\nreal log_determinant(matrix A) The log of the absolute value of the determinant of A\nAvailable since 2.0\nreal log_determinant_spd(matrix A) The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.\nAvailable since 2.30\n\n\n\nIt is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix m and post-multiplying by a vector or matrix a, as in inverse(m) * a, it is better to code this using matrix division, as in m \\ a. The pre-multiplication case is similar, with b * inverse(m) being more efficiently coded as as b / m. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.\nWarning: The function inv(m) is the elementwise inverse function, which returns 1 / m[i, j] for each element.\n \n\nmatrix inverse(matrix A) Compute the inverse of A\nAvailable since 2.0\n \n\nmatrix inverse_spd(matrix A) Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.\nAvailable since 2.0\n \n\nmatrix chol2inv(matrix L) Compute the inverse of the matrix whose cholesky factorization is L. That is, for \\(A = L L^T\\), return \\(A^{-1}\\).\nAvailable since 2.26\n\n\n\nThe generalized inverse \\(M^+\\) of a matrix \\(M\\) is a matrix that satisfies \\(M M^+ M = M\\). For an invertible, square matrix \\(M\\), \\(M^+\\) is equivalent to \\(M^{-1}\\). The dimensions of \\(M^+\\) are equivalent to the dimensions of \\(M^T\\). The generalized inverse exists for any matrix, so the \\(M\\) may be singular or less than full rank.\nEven though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank (Golub and Pereyra 1973), meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix \\(A\\) as a function of \\(\\epsilon\\):\n\\[\nA = \\left(\n    \\begin{array}{cccc}\n    1 + \\epsilon & 2 & 1 \\\\\n    2 & 4 & 2\n    \\end{array}\n    \\right)\n\\]\nWhen \\(\\epsilon = 0\\), \\(A\\) is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If \\(\\epsilon \\neq 0\\), the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at \\(\\epsilon = 0\\), and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.\n \n\nmatrix generalized_inverse(matrix A) The generalized inverse of A\nAvailable since 2.26\n\n\n\n \n\ncomplex_vector eigenvalues(matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors(matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nvector eigenvalues_sym(matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.0\n \n\nmatrix eigenvectors_sym(matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.0\n \n\ntuple(matrix, vector) eigendecompose_sym(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\n \n\nmatrix qr_thin_Q(matrix A) The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A\nAvailable since 2.18\n \n\nmatrix qr_thin_R(matrix A) The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A\nAvailable since 2.18\n \n\ntuple(matrix, matrix) qr_thin(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_thin_Q(A), qr_thin_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nmatrix qr_Q(matrix A) The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A\nAvailable since 2.3\n \n\nmatrix qr_R(matrix A) The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A\nAvailable since 2.3\n \n tuple(matrix, matrix) qr(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthogonal matrix in the fat QR decomposition and the second element (“R”) is upper trapezoidal. This function is equivalent to (qr_Q(A), qr_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nThe thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as \\[\\begin{equation*} A = Q \\, R. \\end{equation*}\\] Multiplying a column of an orthogonal matrix by \\(-1\\) still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by \\(-1\\) without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.\n\n\n\nEvery symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If \\(\\Sigma\\) is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector \\(L\\) such that \\[\\begin{equation*} \\Sigma = L \\, L^{\\top}. \\end{equation*}\\]\n \n\nmatrix cholesky_decompose(matrix A) The lower-triangular Cholesky factor of the symmetric positive-definite matrix A\nAvailable since 2.0\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(matrix A) The singular values of A in descending order\nAvailable since 2.0\n \n\nmatrix svd_U(matrix A) The left-singular vectors of A\nAvailable since 2.26\n \n\nmatrix svd_V(matrix A) The right-singular vectors of A\nAvailable since 2.26\n \n\ntuple(matrix, vector, matrix) svd(matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\n\n\nSee the sorting functions section for examples of how the functions work.\n \n\nvector sort_asc(vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nrow_vector sort_asc(row_vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nvector sort_desc(vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\nrow_vector sort_desc(row_vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(row_vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n\n\n\n \n\nvector reverse(vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.23\n \n\nrow_vector reverse(row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#integer-valued-matrix-size-functions",
+    "href": "functions-reference/matrix_operations.html#integer-valued-matrix-size-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "int num_elements(vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.5\n \n\nint num_elements(row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.5\n \n\nint num_elements(matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.5\n \n\nint rows(vector x) The number of rows in the vector x\nAvailable since 2.0\n \n\nint rows(row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.0\n \n\nint rows(matrix x) The number of rows in the matrix x\nAvailable since 2.0\n \n\nint cols(vector x) The number of columns in the vector x, namely 1\nAvailable since 2.0\n \n\nint cols(row_vector x) The number of columns in the row vector x\nAvailable since 2.0\n \n\nint cols(matrix x) The number of columns in the matrix x\nAvailable since 2.0\n \n\nint size(vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15\nAvailable since 2.26",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#matrix-arithmetic-operators",
+    "href": "functions-reference/matrix_operations.html#matrix-arithmetic-operators",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Stan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\nvector operator-(vector x) The negation of the vector x.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x) The negation of the row vector x.\nAvailable since 2.0\n \n\nmatrix operator-(matrix x) The negation of the matrix x.\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\nvector operator+(vector x, vector y) The sum of the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, row_vector y) The sum of the row vectors x and y.\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, matrix y) The sum of the matrices x and y\nAvailable since 2.0\n \n\nvector operator-(vector x, vector y) The difference between the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, row_vector y) The difference between the row vectors x and y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, matrix y) The difference between the matrices x and y\nAvailable since 2.0\n \n\nvector operator*(real x, vector y) The product of the scalar x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(real x, row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.0\n \n\nmatrix operator*(real x, matrix y) The product of the scalar x and the matrix y\nAvailable since 2.0\n \n\nvector operator*(vector x, real y) The product of the scalar y and vector x\nAvailable since 2.0\n \n\nmatrix operator*(vector x, row_vector y) The product of the vector x and row vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, real y) The product of the scalar y and row vector x\nAvailable since 2.0\n \n\nreal operator*(row_vector x, vector y) The product of the row vector x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, matrix y) The product of the row vector x and matrix y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, real y) The product of the scalar y and matrix x\nAvailable since 2.0\n \n\nvector operator*(matrix x, vector y) The product of the matrix x and vector y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, matrix y) The product of the matrices x and y\nAvailable since 2.0\n\n\n\n \n\nvector operator+(vector x, real y) The result of adding y to every entry in the vector x\nAvailable since 2.0\n \n\nvector operator+(real x, vector y) The result of adding x to every entry in the vector y\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, real y) The result of adding y to every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator+(real x, row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, real y) The result of adding y to every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator+(real x, matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.0\n \n\nvector operator-(vector x, real y) The result of subtracting y from every entry in the vector x\nAvailable since 2.0\n \n\nvector operator-(real x, vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, real y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator-(real x, row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, real y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator-(real x, matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.0\n \n\nvector operator/(vector x, real y) The result of dividing each entry in the vector x by y\nAvailable since 2.0\n \n\nrow_vector operator/(row_vector x, real y) The result of dividing each entry in the row vector x by y\nAvailable since 2.0\n \n\nmatrix operator/(matrix x, real y) The result of dividing each entry in the matrix x by y\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#transposition-operator",
+    "href": "functions-reference/matrix_operations.html#transposition-operator",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Matrix transposition is represented using a postfix operator.\n \n\nmatrix operator'(matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.0\n \n\nrow_vector operator'(vector x) The transpose of the vector x, written as x'\nAvailable since 2.0\n \n\nvector operator'(row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#elementwise-functions",
+    "href": "functions-reference/matrix_operations.html#elementwise-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Elementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section function vectorization for the general cases.\n \n\nvector operator.*(vector x, vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nrow_vector operator.*(row_vector x, row_vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nmatrix operator.*(matrix x, matrix y) The elementwise product of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator./(real x, vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(row_vector x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nrow_vector operator./(row_vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(real x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(matrix x, matrix y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nmatrix operator./(matrix x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(real x, matrix y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator.^(vector x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(real x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(real x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(real x, matrix y) The elementwise power of y and x\nAvailable since 2.24",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#dot-products-and-specialized-products",
+    "href": "functions-reference/matrix_operations.html#dot-products-and-specialized-products",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "real dot_product(vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(vector x, vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(row_vector x, row_vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(matrix x, matrix y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(vector x, vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(row_vector x, row_vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(matrix x, matrix y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nreal dot_self(vector x) The dot product of the vector x with itself\nAvailable since 2.0\n \n\nreal dot_self(row_vector x) The dot product of the row vector x with itself\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(matrix x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(matrix x) The dot product of the rows of x with themselves\nAvailable since 2.0\n\n\n \n\nmatrix tcrossprod(matrix x) The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}\\,\\text{x}^{\\top}\\).\nAvailable since 2.0\n \n\nmatrix crossprod(matrix x) The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}^{\\top}\\,\\text{x}\\).\nAvailable since 2.0\nThe following functions all provide shorthand forms for common expressions, which are also much more efficient.\n \n\nmatrix quad_form(matrix A, matrix B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nreal quad_form(matrix A, vector B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nmatrix quad_form_diag(matrix m, vector v) The quadratic form using the column vector v as a diagonal matrix, i.e., diag_matrix(v) * m * diag_matrix(v).\nAvailable since 2.3\n \n\nmatrix quad_form_diag(matrix m, row_vector rv) The quadratic form using the row vector rv as a diagonal matrix, i.e., diag_matrix(rv) * m * diag_matrix(rv).\nAvailable since 2.3\n \n\nmatrix quad_form_sym(matrix A, matrix B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal quad_form_sym(matrix A, vector B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal trace_dot(matrix A, matrix B) The trace of the matrix product, i.e., trace(A * B).\nAvailable since 2.39\n \n\nreal trace_quad_form(matrix A, matrix B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_quad_form(matrix A, vector B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_gen_quad_form(matrix D, matrix A, matrix B) The trace of a generalized quadratic form, i.e., trace(D * B' * A * B).\nAvailable since 2.0\n \n\nmatrix multiply_lower_tri_self_transpose(matrix x) The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if L is a matrix of the same dimensions as x with L(m,n) equal to x(m,n) for \\(\\text{n}\n\\leq \\text{m}\\) and L(m,n) equal to 0 if \\(\\text{n} &gt; \\text{m}\\), the result is the symmetric matrix \\(\\text{L}\\,\\text{L}^{\\top}\\). This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(vector v, matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(row_vector rv, matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, row_vector rv) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#reductions",
+    "href": "functions-reference/matrix_operations.html#reductions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "real log_sum_exp(vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(row_vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(matrix x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n\n\n\n \n\nreal min(vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(row_vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(matrix x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(row_vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(matrix x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n\n\n\n \n\nreal sum(vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal prod(vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n\n\n\nFull definitions are provided for sample moments in section array reductions.\n \n\nreal mean(vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(row_vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(matrix x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(row_vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(matrix x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(row_vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(matrix x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n \n\nreal quantile(data row_vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data row_vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#matrix-broadcast",
+    "href": "functions-reference/matrix_operations.html#matrix-broadcast",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\nvector rep_vector(real x, int m) Return the size m (column) vector consisting of copies of x.\nAvailable since 2.0\n \n\nrow_vector rep_row_vector(real x, int n) Return the size n row vector consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(real x, int m, int n) Return the m by n matrix consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.0\n \n\nmatrix rep_matrix(row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.0\nUnlike the situation with array broadcasting (see section array broadcasting), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.\n vector[3] x;\n x = rep_vector(1, 3);\n x = rep_vector(1.0, 3);\nThere are no integer vector or matrix types, so integer values are automatically promoted.\n\n\n \n\nmatrix symmetrize_from_lower_tri(matrix A)\nConstruct a symmetric matrix from the lower triangle of A.\nAvailable since 2.26",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#diagonal-matrix-functions",
+    "href": "functions-reference/matrix_operations.html#diagonal-matrix-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "matrix add_diag(matrix m, row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, real d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.21\n \n\nvector diagonal(matrix x) The diagonal of the matrix x\nAvailable since 2.0\n \n\nmatrix diag_matrix(vector x) The diagonal matrix with diagonal x\nAvailable since 2.0\nAlthough the diag_matrix function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,\n y ~ multi_normal(mu, diag_matrix(square(sigma)));\nit is much more efficient to just use a univariate normal, which produces the same density,\n y ~ normal(mu, sigma);\nRather than writing m * diag_matrix(v) where m is a matrix and v is a vector, it is much more efficient to write diag_post_multiply(m, v) (and similarly for pre-multiplication). By the same token, it is better to use quad_form_diag(m, v) rather than quad_form(m, diag_matrix(v)).\n \n\nmatrix identity_matrix(int k) Create an identity matrix of size \\(k \\times k\\)\nAvailable since 2.26",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#container-construction",
+    "href": "functions-reference/matrix_operations.html#container-construction",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "array[] real linspaced_array(int n, data real lower, data real upper) Create a real array of length n of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int linspaced_int_array(int n, int lower, int upper) Create a regularly spaced, increasing integer array of length n between lower and upper, inclusively. If (upper - lower) / (n - 1) is less than one, repeat each output (n - 1) / (upper - lower) times. If neither (upper - lower) / (n - 1) or (n - 1) / (upper - lower) are integers, upper is reduced until one of these is true.\nAvailable since 2.26\n \n\nvector linspaced_vector(int n, data real lower, data real upper) Create an n-dimensional vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\nrow_vector linspaced_row_vector(int n, data real lower, data real upper) Create an n-dimensional row-vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int one_hot_int_array(int n, int k) Create a one-hot encoded int array of length n with array[k] = 1\nAvailable since 2.26\n \n\narray[] real one_hot_array(int n, int k) Create a one-hot encoded real array of length n with array[k] = 1\nAvailable since 2.24\n \n\nvector one_hot_vector(int n, int k) Create an n-dimensional one-hot encoded vector with vector[k] = 1\nAvailable since 2.24\n \n\nrow_vector one_hot_row_vector(int n, int k) Create an n-dimensional one-hot encoded row-vector with row_vector[k] = 1\nAvailable since 2.24\n \n\narray[] int ones_int_array(int n) Create an int array of length n of all ones\nAvailable since 2.26\n \n\narray[] real ones_array(int n) Create a real array of length n of all ones\nAvailable since 2.26\n \n\nvector ones_vector(int n) Create an n-dimensional vector of all ones\nAvailable since 2.26\n \n\nrow_vector ones_row_vector(int n) Create an n-dimensional row-vector of all ones\nAvailable since 2.26\n \n\narray[] int zeros_int_array(int n) Create an int array of length n of all zeros\nAvailable since 2.26\n \n\narray[] real zeros_array(int n) Create a real array of length n of all zeros\nAvailable since 2.24\n \n\nvector zeros_vector(int n) Create an n-dimensional vector of all zeros\nAvailable since 2.24\n \n\nrow_vector zeros_row_vector(int n) Create an n-dimensional row-vector of all zeros\nAvailable since 2.24\n \n\nvector uniform_simplex(int n) Create an n-dimensional simplex with elements vector[i] = 1 / n for all \\(i \\in 1, \\dots, n\\)\nAvailable since 2.24",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#slicing-and-blocking-functions",
+    "href": "functions-reference/matrix_operations.html#slicing-and-blocking-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Stan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\nvector col(matrix x, int n) The n-th column of matrix x\nAvailable since 2.0\n \n\nrow_vector row(matrix x, int m) The m-th row of matrix x\nAvailable since 2.0\nThe row function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation x[m] is just an alternative way of writing row(x,m). The col function may not, be used as an lvalue, nor is there an indexing based shorthand for it.\n\n\n\n\n\nBlock operations may be used to extract a sub-block of a matrix.\n \n\nmatrix block(matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.0\nThe sub-row and sub-column operations may be used to extract a slice of row or column from a matrix\n \n\nvector sub_col(matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.0\n \n\nrow_vector sub_row(matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.0\n\n\n\nThe head operation extracts the first \\(n\\) elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.\n \n\nvector head(vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.0\n \n\nrow_vector head(row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.0\n \n\narray[] T head(array[] T sv, int n) Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector tail(vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.0\n \n\nrow_vector tail(row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.0\n \n\narray[] T tail(array[] T sv, int n) Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector segment(vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.0\n \n\nrow_vector segment(row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.10\n \n\narray[] T segment(array[] T sv, int i, int n) Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#matrix-concatenation",
+    "href": "functions-reference/matrix_operations.html#matrix-concatenation",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Stan’s matrix and vector concatenation operations append_col and append_row are like the operations cbind and rbind in R.\n\n\n \n\nmatrix append_col(matrix x, matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(matrix x, vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.5\n \n\nrow_vector append_col(row_vector x, row_vector y) Combine row vectors x and y of any size into another row vector by appending y to the end of x.\nAvailable since 2.5\n \n\nrow_vector append_col(real x, row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.12\n \n\nrow_vector append_col(row_vector x, real y) Append y to the end of x, returning another row vector.\nAvailable since 2.12\n\n\n\n \n\nmatrix append_row(matrix x, matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(matrix x, row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.5\n \n\nvector append_row(vector x, vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.5\n \n\nvector append_row(real x, vector y) Append x to the top of y, returning another vector.\nAvailable since 2.12\n \n\nvector append_row(vector x, real y) Append y to the bottom of x, returning another vector.\nAvailable since 2.12",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#softmax",
+    "href": "functions-reference/matrix_operations.html#softmax",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "The softmax function maps1 \\(y \\in \\mathbb{R}^K\\) to the \\(K\\)-simplex by \\[\\begin{equation*} \\text{softmax}(y)  = \\frac{\\exp(y)}\n{\\sum_{k=1}^K \\exp(y_k)}, \\end{equation*}\\] where \\(\\exp(y)\\) is the componentwise exponentiation of \\(y\\). Softmax is usually calculated on the log scale, \\[\\begin{eqnarray*} \\log \\text{softmax}(y) & = & \\ y - \\log \\sum_{k=1}^K\n\\exp(y_k) \\\\[4pt] & = & y - \\mathrm{log\\_sum\\_exp}(y). \\end{eqnarray*}\\] where the vector \\(y\\) minus the scalar \\(\\mathrm{log\\_sum\\_exp}(y)\\) subtracts the scalar from each component of \\(y\\).\nStan provides the following functions for softmax and its log.\n \n\nvector softmax(vector x) The softmax of x\nAvailable since 2.0\n \n\nvector log_softmax(vector x) The natural logarithm of the softmax of x\nAvailable since 2.0\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] int cumulative_sum(array[] int x) The cumulative sum of x\nAvailable since 2.30\n \n\narray[] real cumulative_sum(array[] real x) The cumulative sum of x\nAvailable since 2.0\n \n\nvector cumulative_sum(vector v) The cumulative sum of v\nAvailable since 2.0\n \n\nrow_vector cumulative_sum(row_vector rv) The cumulative sum of rv\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#gaussian-process-covariance-functions",
+    "href": "functions-reference/matrix_operations.html#gaussian-process-covariance-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "The Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.\nFor one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by \\(K_{ij} = k(x_i, x_j)\\) (where \\(x_i\\) is the \\(i^{th}\\) element of the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(x_i, y_j)\\).\nFor multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{x}_j)\\) (where \\(\\mathbf{x}_i\\) is the \\(i^{th}\\) vector in the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{y}_j)\\).\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|^2}{2l^2} \\right)\n\\]\n \n\nmatrix gp_exp_quad_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith bias \\(\\sigma_0\\) the dot product kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma_0^2 + \\mathbf{x}_i^T \\mathbf{x}_j\n\\]\n \n\nmatrix gp_dot_prod_cov(array[] real x, real sigma)\nGaussian process covariance with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(array[] real x1, array[] real x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x, real sigma)\nGaussian process covariance with dot product kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x1, vectors x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in multiple dimensions.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponential kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_exponential_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 3/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right) \\exp \\left( -\\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern32_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 5/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{5}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} + \\frac{5 |\\mathbf{x}_i - \\mathbf{x}_j|^2}{3l^2} \\right)\n\\exp \\left( -\\frac{\\sqrt{5} |\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern52_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\), length scale \\(l\\), and period \\(p\\), the periodic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left(-\\frac{2 \\sin^2 \\left( \\pi \\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{p} \\right) }{l^2} \\right)\n\\]\n \n\nmatrix gp_periodic_cov(array[] real x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(array[] real x1, array[] real x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x1, vectors x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#linear-algebra-functions-and-solvers",
+    "href": "functions-reference/matrix_operations.html#linear-algebra-functions-and-solvers",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.\n\n\n \n\nrow_vector operator/(row_vector b, matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.0\n \n\nmatrix operator/(matrix B, matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.5\n \n\nvector operator\\(matrix A, vector b) The left division of A by b; equivalently inverse(A) * b\nAvailable since 2.18\n \n\nmatrix operator\\(matrix A, matrix B) The left division of A by B; equivalently inverse(A) * B\nAvailable since 2.18\n\n\n\nThere are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix \\(\\text{tri}(A)\\) is used in the definitions and defined by \\[\\begin{equation*} \\text{tri}(A)[m,n] = \\left\\{\n\\begin{array}{ll} A[m,n] & \\text{if } m \\geq n, \\text{ and} \\\\[4pt] 0\n& \\text{otherwise}. \\end{array} \\right. \\end{equation*}\\] When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.\n \n\nvector mdivide_left_tri_low(matrix A, vector b) The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * b, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_left_tri_low(matrix A, matrix B) The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * B, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n \n\nrow_vector mdivide_right_tri_low(row_vector b, matrix A) The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form b * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_right_tri_low(matrix B, matrix A) The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form B * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n\n\n\n\nThere are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.\n \n\nmatrix mdivide_left_spd(matrix A, vector b) The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * b.\nAvailable since 2.12\n \n\nvector mdivide_left_spd(matrix A, matrix B) The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * B.\nAvailable since 2.12\n \n\nrow_vector mdivide_right_spd(row_vector b, matrix A) The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form b *inverse(A).\nAvailable since 2.12\n \n\nmatrix mdivide_right_spd(matrix B, matrix A) The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form B * inverse(A).\nAvailable since 2.12\n\n\n\nThe exponential of the matrix \\(A\\) is formally defined by the convergent power series: \\[\\begin{equation*} e^A = \\sum_{n=0}^{\\infty} \\dfrac{A^n}{n!}\n\\end{equation*}\\]\n \n\nmatrix matrix_exp(matrix A) The matrix exponential of A\nAvailable since 2.13\n \n\nmatrix matrix_exp_multiply(matrix A, matrix B) The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form matrix_exp(A) * B.\nAvailable since 2.18\n \n\nmatrix scale_matrix_exp_multiply(real t, matrix A, matrix B) The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form matrix_exp(t * A) * B.\nAvailable since 2.18\n\n\n\nReturns the nth power of the specific matrix: \\[\\begin{equation*} M^n = M_1 * ... * M_n \\end{equation*}\\]\n \n\nmatrix matrix_power(matrix A, int B) Matrix A raised to the power B.\nAvailable since 2.24\n\n\n\n\n\n \n\nreal trace(matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.0\n\n\n\n \n\nreal determinant(matrix A) The determinant of A\nAvailable since 2.0\n \n\nreal log_determinant(matrix A) The log of the absolute value of the determinant of A\nAvailable since 2.0\nreal log_determinant_spd(matrix A) The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.\nAvailable since 2.30\n\n\n\nIt is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix m and post-multiplying by a vector or matrix a, as in inverse(m) * a, it is better to code this using matrix division, as in m \\ a. The pre-multiplication case is similar, with b * inverse(m) being more efficiently coded as as b / m. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.\nWarning: The function inv(m) is the elementwise inverse function, which returns 1 / m[i, j] for each element.\n \n\nmatrix inverse(matrix A) Compute the inverse of A\nAvailable since 2.0\n \n\nmatrix inverse_spd(matrix A) Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.\nAvailable since 2.0\n \n\nmatrix chol2inv(matrix L) Compute the inverse of the matrix whose cholesky factorization is L. That is, for \\(A = L L^T\\), return \\(A^{-1}\\).\nAvailable since 2.26\n\n\n\nThe generalized inverse \\(M^+\\) of a matrix \\(M\\) is a matrix that satisfies \\(M M^+ M = M\\). For an invertible, square matrix \\(M\\), \\(M^+\\) is equivalent to \\(M^{-1}\\). The dimensions of \\(M^+\\) are equivalent to the dimensions of \\(M^T\\). The generalized inverse exists for any matrix, so the \\(M\\) may be singular or less than full rank.\nEven though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank (Golub and Pereyra 1973), meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix \\(A\\) as a function of \\(\\epsilon\\):\n\\[\nA = \\left(\n    \\begin{array}{cccc}\n    1 + \\epsilon & 2 & 1 \\\\\n    2 & 4 & 2\n    \\end{array}\n    \\right)\n\\]\nWhen \\(\\epsilon = 0\\), \\(A\\) is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If \\(\\epsilon \\neq 0\\), the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at \\(\\epsilon = 0\\), and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.\n \n\nmatrix generalized_inverse(matrix A) The generalized inverse of A\nAvailable since 2.26\n\n\n\n \n\ncomplex_vector eigenvalues(matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors(matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nvector eigenvalues_sym(matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.0\n \n\nmatrix eigenvectors_sym(matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.0\n \n\ntuple(matrix, vector) eigendecompose_sym(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\n \n\nmatrix qr_thin_Q(matrix A) The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A\nAvailable since 2.18\n \n\nmatrix qr_thin_R(matrix A) The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A\nAvailable since 2.18\n \n\ntuple(matrix, matrix) qr_thin(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_thin_Q(A), qr_thin_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nmatrix qr_Q(matrix A) The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A\nAvailable since 2.3\n \n\nmatrix qr_R(matrix A) The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A\nAvailable since 2.3\n \n tuple(matrix, matrix) qr(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthogonal matrix in the fat QR decomposition and the second element (“R”) is upper trapezoidal. This function is equivalent to (qr_Q(A), qr_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nThe thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as \\[\\begin{equation*} A = Q \\, R. \\end{equation*}\\] Multiplying a column of an orthogonal matrix by \\(-1\\) still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by \\(-1\\) without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.\n\n\n\nEvery symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If \\(\\Sigma\\) is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector \\(L\\) such that \\[\\begin{equation*} \\Sigma = L \\, L^{\\top}. \\end{equation*}\\]\n \n\nmatrix cholesky_decompose(matrix A) The lower-triangular Cholesky factor of the symmetric positive-definite matrix A\nAvailable since 2.0\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(matrix A) The singular values of A in descending order\nAvailable since 2.0\n \n\nmatrix svd_U(matrix A) The left-singular vectors of A\nAvailable since 2.26\n \n\nmatrix svd_V(matrix A) The right-singular vectors of A\nAvailable since 2.26\n \n\ntuple(matrix, vector, matrix) svd(matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#sort-functions",
+    "href": "functions-reference/matrix_operations.html#sort-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "See the sorting functions section for examples of how the functions work.\n \n\nvector sort_asc(vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nrow_vector sort_asc(row_vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nvector sort_desc(vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\nrow_vector sort_desc(row_vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(row_vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#reverse-functions",
+    "href": "functions-reference/matrix_operations.html#reverse-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "vector reverse(vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.23\n \n\nrow_vector reverse(row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#footnotes",
+    "href": "functions-reference/matrix_operations.html#footnotes",
+    "title": "Matrix Operations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe softmax function is so called because in the limit as \\(y_n \\rightarrow \\infty\\) with \\(y_m\\) for \\(m \\neq n\\) held constant, the result tends toward the “one-hot” vector \\(\\theta\\) with \\(\\theta_n\n= 1\\) and \\(\\theta_m = 0\\) for \\(m \\neq n\\), thus providing a “soft” version of the maximum function.↩︎",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html",
+    "href": "functions-reference/multivariate_discrete_distributions.html",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "The multivariate discrete distributions are over multiple integer values, which are expressed in Stan as arrays.\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\theta \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\text{Multinomial}(y|\\theta) = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K \\theta_k^{y_k},\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial(theta)\nIncrement target log probability density with multinomial_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multinomial_lpmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y)\nAvailable since 2.12\n \n\nreal multinomial_lupmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_rng(vector theta, int N) Generate a multinomial variate with simplex distribution parameter theta and total count \\(N\\); may only be used in transformed data and generated quantities blocks\nAvailable since 2.8\n\n\n\n\nStan also provides a version of the multinomial probability mass function distribution with the \\(\\text{$K$-simplex}\\) for the event count probabilities per category given on the unconstrained logistic scale.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\text{softmax}(\\theta) \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{MultinomialLogit}(y \\mid \\gamma) & = \\text{Multinomial}(y \\mid \\text{softmax}(\\gamma)) \\\\\n& = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K [\\text{softmax}(\\gamma_k)]^{y_k},\n\\end{split}\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial_logit(gamma)\nIncrement target log probability density with multinomial_logit_lupmf(y | gamma).\nAvailable since 2.24\n \n\n\n\n\n \n\nreal multinomial_logit_lpmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y)\nAvailable since 2.24\n \n\nreal multinomial_logit_lupmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_logit_rng(vector gamma, int N) Generate a variate from a multinomial distribution with probabilities softmax(gamma) and total count N; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.24\n\n\n\n\nStan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\alpha \\in\n\\mathbb{R}_{+}^K\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), the PMF of the Dirichlet-multinomial distribution is defined as \\[\\begin{equation*}\n\\text{DirMult}(y|\\theta) =\n\\frac{\\Gamma(\\alpha_0)\\Gamma(N+1)}{\\Gamma(N+\\alpha_0)} \\prod_{k=1}^K \\frac{\\Gamma(y_k + \\alpha_k)}{\\Gamma(\\alpha_k)\\Gamma(y_k+1)},\n\\end{equation*}\\] where \\(\\alpha_0\\) is defined as \\(\\alpha_0 = \\sum_{k=1}^K \\alpha_k\\).\n\n\n\ny ~ dirichlet_multinomial(alpha)\nIncrement target log probability density with dirichlet_multinomial_lupmf(y | alpha).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal dirichlet_multinomial_lpmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y).\nAvailable since 2.34\n \n\nreal dirichlet_multinomial_lupmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements, given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y) dropping constant additive terms.\nAvailable since 2.34\n \n\narray[] int dirichlet_multinomial_rng(vector alpha, int N) Generate a multinomial variate with positive vector distribution parameter alpha and total count N; may only be used in transformed data and generated quantities blocks. This is equivalent to multinomial_rng(dirichlet_rng(alpha), N).\nAvailable since 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution",
+    "href": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\theta \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\text{Multinomial}(y|\\theta) = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K \\theta_k^{y_k},\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial(theta)\nIncrement target log probability density with multinomial_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multinomial_lpmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y)\nAvailable since 2.12\n \n\nreal multinomial_lupmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_rng(vector theta, int N) Generate a multinomial variate with simplex distribution parameter theta and total count \\(N\\); may only be used in transformed data and generated quantities blocks\nAvailable since 2.8",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution-logit-parameterization",
+    "href": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution-logit-parameterization",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides a version of the multinomial probability mass function distribution with the \\(\\text{$K$-simplex}\\) for the event count probabilities per category given on the unconstrained logistic scale.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\text{softmax}(\\theta) \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{MultinomialLogit}(y \\mid \\gamma) & = \\text{Multinomial}(y \\mid \\text{softmax}(\\gamma)) \\\\\n& = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K [\\text{softmax}(\\gamma_k)]^{y_k},\n\\end{split}\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial_logit(gamma)\nIncrement target log probability density with multinomial_logit_lupmf(y | gamma).\nAvailable since 2.24\n \n\n\n\n\n \n\nreal multinomial_logit_lpmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y)\nAvailable since 2.24\n \n\nreal multinomial_logit_lupmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_logit_rng(vector gamma, int N) Generate a variate from a multinomial distribution with probabilities softmax(gamma) and total count N; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.24",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html#dirichlet-multinomial-distribution",
+    "href": "functions-reference/multivariate_discrete_distributions.html#dirichlet-multinomial-distribution",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\alpha \\in\n\\mathbb{R}_{+}^K\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), the PMF of the Dirichlet-multinomial distribution is defined as \\[\\begin{equation*}\n\\text{DirMult}(y|\\theta) =\n\\frac{\\Gamma(\\alpha_0)\\Gamma(N+1)}{\\Gamma(N+\\alpha_0)} \\prod_{k=1}^K \\frac{\\Gamma(y_k + \\alpha_k)}{\\Gamma(\\alpha_k)\\Gamma(y_k+1)},\n\\end{equation*}\\] where \\(\\alpha_0\\) is defined as \\(\\alpha_0 = \\sum_{k=1}^K \\alpha_k\\).\n\n\n\ny ~ dirichlet_multinomial(alpha)\nIncrement target log probability density with dirichlet_multinomial_lupmf(y | alpha).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal dirichlet_multinomial_lpmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y).\nAvailable since 2.34\n \n\nreal dirichlet_multinomial_lupmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements, given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y) dropping constant additive terms.\nAvailable since 2.34\n \n\narray[] int dirichlet_multinomial_rng(vector alpha, int N) Generate a multinomial variate with positive vector distribution parameter alpha and total count N; may only be used in transformed data and generated quantities blocks. This is equivalent to multinomial_rng(dirichlet_rng(alpha), N).\nAvailable since 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html",
+    "href": "functions-reference/positive_lower-bounded_distributions.html",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "The positive lower-bounded probabilities have support on real values above some positive minimum value.\n\n\n\n\nIf \\(y_{\\text{min}} \\in \\mathbb{R}^+\\) and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\) with \\(y \\geq y_{\\text{min}}\\), \\[\\begin{equation*}\n\\text{Pareto}(y|y_{\\text{min}},\\alpha) = \\frac{\\displaystyle \\alpha\\,y_{\\text{min}}^\\alpha}{\\displaystyle y^{\\alpha+1}}.\n\\end{equation*}\\]\n\n\n\ny ~ pareto(y_min, alpha)\nIncrement target log probability density with pareto_lupdf(y | y_min, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal pareto_lpdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lupdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_cdf(reals y | reals y_min, reals alpha) The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.0\n \n\nreal pareto_lcdf(reals y | reals y_min, reals alpha) The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lccdf(reals y | reals y_min, reals alpha) The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nR pareto_rng(reals y_min, reals alpha) Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\lambda \\in \\mathbb{R}^+\\), and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\geq \\mu\\), \\[\\begin{equation*}\n\\mathrm{Pareto\\_Type\\_2}(y|\\mu,\\lambda,\\alpha) =\n\\ \\frac{\\alpha}{\\lambda} \\, \\left( 1+\\frac{y-\\mu}{\\lambda} \\right)^{-(\\alpha+1)} \\! .\n\\end{equation*}\\]\nNote that the Lomax distribution is a Pareto Type 2 distribution with \\(\\mu=0\\).\n\n\n\ny ~ pareto_type_2(mu, lambda, alpha)\nIncrement target log probability density with pareto_type_2_lupdf(y | mu, lambda, alpha).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal pareto_type_2_lpdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lupdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_type_2_cdf(reals y | reals mu, reals lambda, reals alpha) The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.5\n \n\nreal pareto_type_2_lcdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lccdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nR pareto_type_2_rng(reals mu, reals lambda, reals alpha) Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nFor an extended explanation of how to use the wiener_lpdf and wiener_l[c]cdf_unnorm functions, see Henrich et al. (2024).\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), \\(\\tau \\in \\mathbb{R}^+\\), \\(\\beta \\in (0, 1)\\), \\(\\delta \\in \\mathbb{R}\\), \\(s_{\\delta} \\in \\mathbb{R}^{\\geq 0}\\), \\(s_{\\beta} \\in [0, 1)\\), and \\(s_{\\tau} \\in \\mathbb{R}^{\\geq 0}\\) then for \\(y &gt; \\tau\\),\n\\[\\begin{equation*}\n\\begin{split}\n&\\text{Wiener}(y\\mid \\alpha,\\tau,\\beta,\\delta,s_{\\delta},s_{\\beta},s_{\\tau}) =\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}}\\int_{-\\infty}^{\\infty} p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega)\n\\\\\n&\\times \\frac{1}{\\sqrt{2\\pi s_{\\delta}^2}}\\exp\\Bigl(-\\frac{(\\nu-\\delta)^2}{2s_{\\delta}^2}\\Bigr) \\,d\\nu \\,d\\omega \\,d{\\tau_0}=\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}} M\\times p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega) \\,d\\omega \\,d{\\tau_0},\n\\end{split}\n\\end{equation*}\\]\nwhere \\(p()\\) denotes the density function, and \\(M\\) and \\(p_3()\\) are defined, by using \\(t:=y-{\\tau_0}\\), as\n\\[\\begin{equation*}\nM \\coloneqq \\frac{1}{\\sqrt{1+s_{\\delta}^2t}}\\exp\\Bigl(\\alpha{\\delta}\\omega+\\frac{\\delta^2t}{2}+\\frac{s_{\\delta}^2\\alpha^2\\omega^2-2\\alpha{\\delta}\\omega-\\delta^2t}{2(1+s_{\\delta}^2t)}\\Bigr)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\np_3(t\\mid \\alpha,\\delta,\\beta) \\coloneqq \\frac{1}{\\alpha^2}\\exp\\Bigl(-\\alpha\\delta\\beta-\\frac{\\delta^2t}{2}\\Bigr)f(\\frac{t}{\\alpha^2}\\mid 0,1,\\beta),\n\\end{equation*}\\]\nwhere \\(f(t^*=\\frac{t}{\\alpha^2}\\mid0,1,\\beta)\\) can be specified in two ways:\n\\[\\begin{equation*}\nf_l(t^*\\mid 0,1,\\beta) = \\sum_{k=1}^\\infty k\\pi \\exp\\Bigl(-\\frac{k^2\\pi^2t^*}{2}\\Bigr)\\sin(k\\pi \\beta)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\nf_s(t^*\\mid0,1,\\beta) = \\sum_{k=-\\infty}^\\infty \\frac{1}{\\sqrt{2\\pi(t^*)^3}}(\\beta+2k) \\exp\\Bigl(-\\frac{(\\beta+2k)^2}{2t^*}\\Bigr).\n\\end{equation*}\\]\nWhich of these is used in the computations depends on which expression requires the smaller number of components \\(k\\) to guarantee a pre-specified precision\nIn the case where \\(s_{\\delta}\\), \\(s_{\\beta}\\), and \\(s_{\\tau}\\) are all \\(0\\), this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\alpha}{(y-\\tau)^{3/2}} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = - \\infty}^{\\infty} (2k +\n\\beta) \\phi \\! \\left(\\frac{(2k + \\beta)\\alpha }{\\sqrt{y - \\tau}}\\right),\n\\end{equation*}\\] where \\(\\phi(x)\\) denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\pi}{\\alpha^2} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = 1}^{\\infty} k \\exp \\! \\left(-\\frac{k^2\\pi^2(y-\\tau)};\n{2\\alpha^2}\\right) \\sin \\!(k\\pi\\beta)\n\\end{equation*}\\] see (Feller 1968), (Navarro and Fuss 2009).\n\n\n\nFor the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.\nLet \\(\\alpha\\), \\(\\tau\\), \\(\\beta\\), \\(\\delta\\), \\(s_{\\delta}\\), \\(s_{\\beta}\\), \\(s_{\\tau}\\) and \\(y\\) be as above.\nThe formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in \\(\\tau\\) for the time being) at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha, \\beta, \\delta) = P(\\alpha, \\beta, \\delta) -\n    \\exp\\left(\\delta\\alpha(1-\\beta)-\\frac{\\delta^2 y}{2}\\right)F_l(y|\\alpha,\\beta,\\delta),\n\\end{equation}\\] where \\(P(\\alpha,\\beta,\\delta)\\) is the probability to hit the upper boundary, defined as\n\\[\\begin{equation}\nP(\\alpha, \\beta, \\delta) =\n\\begin{cases}\n    \\frac{1-\\exp(2\\delta \\alpha \\beta)}{\\exp(-2\\delta \\alpha(1-\\beta)) - \\exp(2\\delta \\alpha \\beta)}, & \\text{for } \\delta\\neq 0 \\\\\n    \\beta, & \\text{for } \\delta=0,\n\\end{cases}\n\\end{equation}\\]\nand\n\\[\\begin{equation}\nF_l(y|\\alpha, \\beta, \\delta) =\n    \\frac{2\\pi}{\\alpha^2}\\sum_{k=1}^{\\infty}{\\frac{k\\sin{k\\pi(1-\\beta)}}{\\delta^2+(k\\pi)^2/\\alpha^2}}\\exp(-\\frac{k^2\\pi^2y}{2\\alpha^2}).\n\\end{equation}\\]\nThe formula for the small-time cdf at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha,\\beta,\\delta) = \\exp\\left(\\delta \\alpha(1-\\beta) -\\frac{\\delta^2y}{2}\\right)F_s(y|\\alpha, \\beta,\\delta),\n\\end{equation}\\] where\n\\[\\begin{equation}\nF_s(y|\\alpha,\\beta,\\delta) = \\sum_{k=0}^{\\infty}(-1)^k\\phi\\left(\\frac{\\alpha(k+\\beta^{*}_k)}\n    {\\sqrt{y}} \\right) \\times \\left( R \\left(\\frac{\\alpha(k+\\beta^{*}_k)+\\delta y}{\\sqrt{y}} \\right)\n        + R \\left(\\frac{\\alpha(k+\\beta^*_k)-\\delta y}{\\sqrt{y}} \\right)\\right),\n\\end{equation}\\]\nwhere \\(\\beta^*_k=(1-\\beta)\\) for \\(k\\) even, \\(\\beta^*_k=\\beta\\) for \\(k\\) odd, and \\(R\\) is Mill’s ratio.\nThe cdf for the lower boundary is \\(F(y|\\alpha,1-\\beta,-\\delta)\\)\n\n\n\ny ~ wiener(alpha, tau, beta, delta)\nIncrement target log probability density with wiener_lupdf(y | alpha, tau, beta, delta).\nAvailable since 2.7\ny ~ wiener(alpha, tau, beta, delta, var_delta) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta).\nAvailable since 2.35\ny ~ wiener(alpha, tau, beta, delta, var_delta, var_beta, var_tau) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau).\nAvailable since 2.35\n \n\n\n\n\n \n\nreal wiener_lpdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.18\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta, dropping constant additive terms\nAvailable since 2.25\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta, dropping constant additive terms.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau, dropping constant additive terms.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n\nNote: The lcdf and lccdf functions for the wiener distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper boundary.\nSimilarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical \\(1 - \\textrm{cdf}\\).\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n\n\n\nStan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use \\[\\begin{equation*}\n\\text{Wiener}(y | \\alpha, \\tau, 1 - \\beta, - \\delta)\n\\end{equation*}\\] For more details, see the appendix of Vandekerckhove and Wabersich (2014).\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions (listed above as recieving only real) are implemented in such a way where they can be fully vectorized, but currently only versions that accept all real and all vector arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by opening an issue.\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions can also accept an additional data real argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to 1e-4 for the density and 1e-8 for the cdf functions.",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html#pareto-distribution",
+    "href": "functions-reference/positive_lower-bounded_distributions.html#pareto-distribution",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "If \\(y_{\\text{min}} \\in \\mathbb{R}^+\\) and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\) with \\(y \\geq y_{\\text{min}}\\), \\[\\begin{equation*}\n\\text{Pareto}(y|y_{\\text{min}},\\alpha) = \\frac{\\displaystyle \\alpha\\,y_{\\text{min}}^\\alpha}{\\displaystyle y^{\\alpha+1}}.\n\\end{equation*}\\]\n\n\n\ny ~ pareto(y_min, alpha)\nIncrement target log probability density with pareto_lupdf(y | y_min, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal pareto_lpdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lupdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_cdf(reals y | reals y_min, reals alpha) The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.0\n \n\nreal pareto_lcdf(reals y | reals y_min, reals alpha) The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lccdf(reals y | reals y_min, reals alpha) The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nR pareto_rng(reals y_min, reals alpha) Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html#pareto-type-2-distribution",
+    "href": "functions-reference/positive_lower-bounded_distributions.html#pareto-type-2-distribution",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\), \\(\\lambda \\in \\mathbb{R}^+\\), and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\geq \\mu\\), \\[\\begin{equation*}\n\\mathrm{Pareto\\_Type\\_2}(y|\\mu,\\lambda,\\alpha) =\n\\ \\frac{\\alpha}{\\lambda} \\, \\left( 1+\\frac{y-\\mu}{\\lambda} \\right)^{-(\\alpha+1)} \\! .\n\\end{equation*}\\]\nNote that the Lomax distribution is a Pareto Type 2 distribution with \\(\\mu=0\\).\n\n\n\ny ~ pareto_type_2(mu, lambda, alpha)\nIncrement target log probability density with pareto_type_2_lupdf(y | mu, lambda, alpha).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal pareto_type_2_lpdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lupdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_type_2_cdf(reals y | reals mu, reals lambda, reals alpha) The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.5\n \n\nreal pareto_type_2_lcdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lccdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nR pareto_type_2_rng(reals mu, reals lambda, reals alpha) Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html#wiener-first-passage-time-distribution",
+    "href": "functions-reference/positive_lower-bounded_distributions.html#wiener-first-passage-time-distribution",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "For an extended explanation of how to use the wiener_lpdf and wiener_l[c]cdf_unnorm functions, see Henrich et al. (2024).\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), \\(\\tau \\in \\mathbb{R}^+\\), \\(\\beta \\in (0, 1)\\), \\(\\delta \\in \\mathbb{R}\\), \\(s_{\\delta} \\in \\mathbb{R}^{\\geq 0}\\), \\(s_{\\beta} \\in [0, 1)\\), and \\(s_{\\tau} \\in \\mathbb{R}^{\\geq 0}\\) then for \\(y &gt; \\tau\\),\n\\[\\begin{equation*}\n\\begin{split}\n&\\text{Wiener}(y\\mid \\alpha,\\tau,\\beta,\\delta,s_{\\delta},s_{\\beta},s_{\\tau}) =\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}}\\int_{-\\infty}^{\\infty} p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega)\n\\\\\n&\\times \\frac{1}{\\sqrt{2\\pi s_{\\delta}^2}}\\exp\\Bigl(-\\frac{(\\nu-\\delta)^2}{2s_{\\delta}^2}\\Bigr) \\,d\\nu \\,d\\omega \\,d{\\tau_0}=\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}} M\\times p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega) \\,d\\omega \\,d{\\tau_0},\n\\end{split}\n\\end{equation*}\\]\nwhere \\(p()\\) denotes the density function, and \\(M\\) and \\(p_3()\\) are defined, by using \\(t:=y-{\\tau_0}\\), as\n\\[\\begin{equation*}\nM \\coloneqq \\frac{1}{\\sqrt{1+s_{\\delta}^2t}}\\exp\\Bigl(\\alpha{\\delta}\\omega+\\frac{\\delta^2t}{2}+\\frac{s_{\\delta}^2\\alpha^2\\omega^2-2\\alpha{\\delta}\\omega-\\delta^2t}{2(1+s_{\\delta}^2t)}\\Bigr)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\np_3(t\\mid \\alpha,\\delta,\\beta) \\coloneqq \\frac{1}{\\alpha^2}\\exp\\Bigl(-\\alpha\\delta\\beta-\\frac{\\delta^2t}{2}\\Bigr)f(\\frac{t}{\\alpha^2}\\mid 0,1,\\beta),\n\\end{equation*}\\]\nwhere \\(f(t^*=\\frac{t}{\\alpha^2}\\mid0,1,\\beta)\\) can be specified in two ways:\n\\[\\begin{equation*}\nf_l(t^*\\mid 0,1,\\beta) = \\sum_{k=1}^\\infty k\\pi \\exp\\Bigl(-\\frac{k^2\\pi^2t^*}{2}\\Bigr)\\sin(k\\pi \\beta)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\nf_s(t^*\\mid0,1,\\beta) = \\sum_{k=-\\infty}^\\infty \\frac{1}{\\sqrt{2\\pi(t^*)^3}}(\\beta+2k) \\exp\\Bigl(-\\frac{(\\beta+2k)^2}{2t^*}\\Bigr).\n\\end{equation*}\\]\nWhich of these is used in the computations depends on which expression requires the smaller number of components \\(k\\) to guarantee a pre-specified precision\nIn the case where \\(s_{\\delta}\\), \\(s_{\\beta}\\), and \\(s_{\\tau}\\) are all \\(0\\), this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\alpha}{(y-\\tau)^{3/2}} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = - \\infty}^{\\infty} (2k +\n\\beta) \\phi \\! \\left(\\frac{(2k + \\beta)\\alpha }{\\sqrt{y - \\tau}}\\right),\n\\end{equation*}\\] where \\(\\phi(x)\\) denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\pi}{\\alpha^2} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = 1}^{\\infty} k \\exp \\! \\left(-\\frac{k^2\\pi^2(y-\\tau)};\n{2\\alpha^2}\\right) \\sin \\!(k\\pi\\beta)\n\\end{equation*}\\] see (Feller 1968), (Navarro and Fuss 2009).\n\n\n\nFor the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.\nLet \\(\\alpha\\), \\(\\tau\\), \\(\\beta\\), \\(\\delta\\), \\(s_{\\delta}\\), \\(s_{\\beta}\\), \\(s_{\\tau}\\) and \\(y\\) be as above.\nThe formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in \\(\\tau\\) for the time being) at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha, \\beta, \\delta) = P(\\alpha, \\beta, \\delta) -\n    \\exp\\left(\\delta\\alpha(1-\\beta)-\\frac{\\delta^2 y}{2}\\right)F_l(y|\\alpha,\\beta,\\delta),\n\\end{equation}\\] where \\(P(\\alpha,\\beta,\\delta)\\) is the probability to hit the upper boundary, defined as\n\\[\\begin{equation}\nP(\\alpha, \\beta, \\delta) =\n\\begin{cases}\n    \\frac{1-\\exp(2\\delta \\alpha \\beta)}{\\exp(-2\\delta \\alpha(1-\\beta)) - \\exp(2\\delta \\alpha \\beta)}, & \\text{for } \\delta\\neq 0 \\\\\n    \\beta, & \\text{for } \\delta=0,\n\\end{cases}\n\\end{equation}\\]\nand\n\\[\\begin{equation}\nF_l(y|\\alpha, \\beta, \\delta) =\n    \\frac{2\\pi}{\\alpha^2}\\sum_{k=1}^{\\infty}{\\frac{k\\sin{k\\pi(1-\\beta)}}{\\delta^2+(k\\pi)^2/\\alpha^2}}\\exp(-\\frac{k^2\\pi^2y}{2\\alpha^2}).\n\\end{equation}\\]\nThe formula for the small-time cdf at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha,\\beta,\\delta) = \\exp\\left(\\delta \\alpha(1-\\beta) -\\frac{\\delta^2y}{2}\\right)F_s(y|\\alpha, \\beta,\\delta),\n\\end{equation}\\] where\n\\[\\begin{equation}\nF_s(y|\\alpha,\\beta,\\delta) = \\sum_{k=0}^{\\infty}(-1)^k\\phi\\left(\\frac{\\alpha(k+\\beta^{*}_k)}\n    {\\sqrt{y}} \\right) \\times \\left( R \\left(\\frac{\\alpha(k+\\beta^{*}_k)+\\delta y}{\\sqrt{y}} \\right)\n        + R \\left(\\frac{\\alpha(k+\\beta^*_k)-\\delta y}{\\sqrt{y}} \\right)\\right),\n\\end{equation}\\]\nwhere \\(\\beta^*_k=(1-\\beta)\\) for \\(k\\) even, \\(\\beta^*_k=\\beta\\) for \\(k\\) odd, and \\(R\\) is Mill’s ratio.\nThe cdf for the lower boundary is \\(F(y|\\alpha,1-\\beta,-\\delta)\\)\n\n\n\ny ~ wiener(alpha, tau, beta, delta)\nIncrement target log probability density with wiener_lupdf(y | alpha, tau, beta, delta).\nAvailable since 2.7\ny ~ wiener(alpha, tau, beta, delta, var_delta) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta).\nAvailable since 2.35\ny ~ wiener(alpha, tau, beta, delta, var_delta, var_beta, var_tau) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau).\nAvailable since 2.35\n \n\n\n\n\n \n\nreal wiener_lpdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.18\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta, dropping constant additive terms\nAvailable since 2.25\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta, dropping constant additive terms.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau, dropping constant additive terms.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n\nNote: The lcdf and lccdf functions for the wiener distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper boundary.\nSimilarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical \\(1 - \\textrm{cdf}\\).\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n\n\n\nStan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use \\[\\begin{equation*}\n\\text{Wiener}(y | \\alpha, \\tau, 1 - \\beta, - \\delta)\n\\end{equation*}\\] For more details, see the appendix of Vandekerckhove and Wabersich (2014).\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions (listed above as recieving only real) are implemented in such a way where they can be fully vectorized, but currently only versions that accept all real and all vector arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by opening an issue.\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions can also accept an additional data real argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to 1e-4 for the density and 1e-8 for the cdf functions.",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/references.html",
+    "href": "functions-reference/references.html",
+    "title": "References",
+    "section": "",
+    "text": "References\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "functions-reference/simplex_distributions.html",
+    "href": "functions-reference/simplex_distributions.html",
+    "title": "Simplex Distributions",
+    "section": "",
+    "text": "The simplex probabilities have support on the unit \\(K\\)-simplex for a specified \\(K\\). A \\(K\\)-dimensional vector \\(\\theta\\) is a unit \\(K\\)-simplex if \\(\\theta_k \\geq 0\\) for \\(k \\in \\{1,\\ldots,K\\}\\) and \\(\\sum_{k = 1}^K \\theta_k = 1\\).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) and \\(\\alpha \\in (\\mathbb{R}^+)^{K}\\), then for \\(\\theta \\in \\text{$K$-simplex}\\),\n\\[\\begin{equation*}\n\\text{Dirichlet}(\\theta|\\alpha) =\n\\frac{\\Gamma \\! \\left( \\sum_{k=1}^K \\alpha_k \\right)}{\\prod_{k=1}^K \\Gamma(\\alpha_k)}  \\prod_{k=1}^K \\theta_k^{\\alpha_k -1}\n\\end{equation*}\\]\nWarning: If any of the components of \\(\\theta\\) satisfies \\(\\theta_i = 0\\) or \\(\\theta_i = 1\\), then the probability is \\(0\\) and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, with \\(\\alpha_i &gt; 0\\) for each \\(i\\).\n\n\n\nA symmetric Dirichlet prior is \\([\\alpha, \\ldots, \\alpha]^{\\top}\\). To code this in Stan,\n data {\n   int&lt;lower=1&gt; K;\n   real&lt;lower=0&gt; alpha;\n }\n generated quantities {\n   vector[K] theta = dirichlet_rng(rep_vector(alpha, K));\n }\nTaking \\(K = 10\\), here are the first five draws for \\(\\alpha = 1\\). For \\(\\alpha = 1\\), the distribution is uniform over simplexes.\n 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05\n 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06\n 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03\n 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18\n 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01\nThat does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) \\(1/K\\). When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on \\(10\\)-simplexes, the probability that a component is greater than the mean of \\(1/10\\) is only 39%. Most of the posterior marginal probability mass for each component is in the interval \\((0, 0.1)\\).\nWhen the \\(\\alpha\\) value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for \\(\\alpha = 0.001\\).\n 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279\n 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000\n 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112\n 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000\n 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000\nEach row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.\nAs \\(\\alpha\\) increases, the draws become increasingly uniform. For \\(\\alpha = 1000\\),\n 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10\n 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n\n\n\ntheta ~ dirichlet(alpha)\nIncrement target log probability density with dirichlet_lupdf(theta | alpha).\nAvailable since 2.0\n \n\n\n\n\nThe Dirichlet probability functions are overloaded to allow the simplex \\(\\theta\\) and prior counts (plus one) \\(\\alpha\\) to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal dirichlet_lpdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha\nAvailable since 2.12, vectorized in 2.21\n \n\nreal dirichlet_lupdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha dropping constant additive terms\nAvailable since 2.25\n \n\nvector dirichlet_rng(vector alpha) Generate a Dirichlet variate with prior counts (plus one) alpha; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Simplex Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/simplex_distributions.html#dirichlet-distribution",
+    "href": "functions-reference/simplex_distributions.html#dirichlet-distribution",
+    "title": "Simplex Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\) and \\(\\alpha \\in (\\mathbb{R}^+)^{K}\\), then for \\(\\theta \\in \\text{$K$-simplex}\\),\n\\[\\begin{equation*}\n\\text{Dirichlet}(\\theta|\\alpha) =\n\\frac{\\Gamma \\! \\left( \\sum_{k=1}^K \\alpha_k \\right)}{\\prod_{k=1}^K \\Gamma(\\alpha_k)}  \\prod_{k=1}^K \\theta_k^{\\alpha_k -1}\n\\end{equation*}\\]\nWarning: If any of the components of \\(\\theta\\) satisfies \\(\\theta_i = 0\\) or \\(\\theta_i = 1\\), then the probability is \\(0\\) and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, with \\(\\alpha_i &gt; 0\\) for each \\(i\\).\n\n\n\nA symmetric Dirichlet prior is \\([\\alpha, \\ldots, \\alpha]^{\\top}\\). To code this in Stan,\n data {\n   int&lt;lower=1&gt; K;\n   real&lt;lower=0&gt; alpha;\n }\n generated quantities {\n   vector[K] theta = dirichlet_rng(rep_vector(alpha, K));\n }\nTaking \\(K = 10\\), here are the first five draws for \\(\\alpha = 1\\). For \\(\\alpha = 1\\), the distribution is uniform over simplexes.\n 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05\n 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06\n 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03\n 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18\n 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01\nThat does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) \\(1/K\\). When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on \\(10\\)-simplexes, the probability that a component is greater than the mean of \\(1/10\\) is only 39%. Most of the posterior marginal probability mass for each component is in the interval \\((0, 0.1)\\).\nWhen the \\(\\alpha\\) value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for \\(\\alpha = 0.001\\).\n 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279\n 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000\n 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112\n 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000\n 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000\nEach row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.\nAs \\(\\alpha\\) increases, the draws become increasingly uniform. For \\(\\alpha = 1000\\),\n 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10\n 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n\n\n\ntheta ~ dirichlet(alpha)\nIncrement target log probability density with dirichlet_lupdf(theta | alpha).\nAvailable since 2.0\n \n\n\n\n\nThe Dirichlet probability functions are overloaded to allow the simplex \\(\\theta\\) and prior counts (plus one) \\(\\alpha\\) to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal dirichlet_lpdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha\nAvailable since 2.12, vectorized in 2.21\n \n\nreal dirichlet_lupdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha dropping constant additive terms\nAvailable since 2.25\n \n\nvector dirichlet_rng(vector alpha) Generate a Dirichlet variate with prior counts (plus one) alpha; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Simplex Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html",
+    "href": "functions-reference/transform_functions.html",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "Variable transformation functions provide implementations of the built-in constraining and unconstraining transforms defined in Stan Reference Manual.\nFor each of the built-in variable transforms there are three functions named after the transform with differing suffixes. A _unconstrain function that maps from the constrained space back to free variables (the “transform”), A _constrain function that maps from free variables to constrained variables (the “inverse transform”), and a _jacobian function, which computes the same value as the _constrain function while also incrementing the Jacobian accumulator with the log Jacobian determinant.\nFor this page, variables named y are unconstrained, while variables named x are in the constrained space. The unconstraining functions will reject if their input does not satisfy the declared constraint.\n\n\nThese transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.\nThe functions are all overloaded to apply to containers elementwise. If the y argument is a container, the others must be either scalars or containers of exactly the same size.\n\n\nThese functions perform the transform and inverse transform described in the Lower Bounded Scalar section.\n \n\nreals lower_bound_constrain(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_bound_jacobian(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_bound_unconstrain(reals x, reals lb) Takes a value x which is greater than lb and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Upper Bounded Scalar section.\n \n\nreals upper_bound_constrain(reals y, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals upper_bound_jacobian(reals x, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals upper_bound_unconstrain(reals x, reals ub) Takes a value x which is less than ub and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Lower and Upper Bounded Scalar section.\n \n\nreals lower_upper_bound_constrain(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_upper_bound_jacobian(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_upper_bound_unconstrain(reals x, reals lb, reals ub) Takes a value x which is bounded between lb and ub and returns returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Affinely Transformed Scalar section.\n \n\nreals offset_multiplier_constrain(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nAvailable since 2.37\n \n\nreals offset_multiplier_jacobian(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals offset_multiplier_unconstrain(reals x, reals offset, reals mult) Takes a value x, shift offset, and scale mult and a value which has been un-scaled and un-shifted.\nAvailable since 2.37\n\n\n\n\nThese functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.\nWhere vectors is used, this indicates that either a vector or a (possibly multidimensional) array of vectors may be provided. The array will be processed element by element.\n\n\nThese functions perform the transform and inverse transform described in the Ordered Vector section.\n \n\nvectors ordered_constrain(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nAvailable since 2.37\n \n\nvectors ordered_jacobian(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors ordered_unconstrain(vectors x) Takes an ordered vector x and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Positive Ordered Vector section.\n \n\nvectors positive_ordered_constrain(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nAvailable since 2.37\n \n\nvectors positive_ordered_jacobian(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors positive_ordered_unconstrain(vectors x) Takes an ordered vector x with positive entries and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Simplex section.\n \n\nvectors simplex_constrain(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors simplex_jacobian(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors simplex_unconstrain(vectors x) Takes a simplex x and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Zero Sum Vector section.\n \n\nvectors sum_to_zero_constrain(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors sum_to_zero_jacobian(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThe returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors sum_to_zero_unconstrain(vectors x) Takes a vector x with elements that sum to 0 and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Vector section.\n \n\nvectors unit_vectors_constrain(vectors y) Takes a free vector y and returns a vector with unit length, i.e., norm2(unit_vectors_constrain(y)) == 1 for any y that has a positive and finite norm itself (if y does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.\nAvailable since 2.37\n \n\nvectors unit_vectors_jacobian(vectors y) Takes a free vector y and returns a vector with unit length. This function rejects if given a vector of all zeros.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors unit_vectors_unconstrain(vectors x) Takes a vector x of unit length and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\n\nSimilarly to the above, vectors means a vector or array thereof, and matrices means a matrix or array thereof.\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Correlation Matrices section.\n \n\nmatrices cholesky_factor_corr_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.\nAvailable since 2.37\n \n Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment. matrices cholesky_factor_corr_jacobian(vectors y, int K)\nAvailable since 2.37\n \n\nvectors cholesky_factor_corr_unconstrain(matrices x) Takes x, a (\\(K \\times K\\)) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Covariance Matrices section.\n \n\nmatrices cholesky_factor_cov_constrain(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nAvailable since 2.37\n \n\nmatrices cholesky_factor_cov_jacobian(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cholesky_factor_cov_unconstrain(matrices x) Takes a \\(M \\times N\\) matrix x which is a Cholesky factor of a covariance matrix (a matrix \\(L\\) such that \\(L^T L\\) is positive definite) and returns the corresponding free vector of length \\(N + \\binom{N}{2} + (M - N)N\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Correlation Matrices section.\n \n\nmatrices corr_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nAvailable since 2.37\n \n\nmatrices corr_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors corr_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) matrix x which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size \\(\\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Covariance Matrices section.\n \n\nmatrices cov_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nAvailable since 2.37\n \n\nmatrices cov_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cov_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) positive definite matrix x and returns the corresponding free vector of size \\(K + \\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for column (left) stochastic matrices.\n \n\nmatrices stochastic_column_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nAvailable since 2.37\n \n\nmatrices stochastic_column_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_column_unconstrain(matrices x) Takes a left stochastic matrix x of size \\(N+1 \\times M\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for row (right) stochastic matrices.\n \n\nmatrices stochastic_row_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nAvailable since 2.37\n \n\nmatrices stochastic_row_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_row_unconstrain(matrices x) Takes a right stochastic matrix x of size \\(N \\times M+1\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThe sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the sum-to-zero matrix section of the Reference Manual.\n \n\nmatrices sum_to_zero_constrain(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_jacobian(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to sum_to_zero_constrain.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_unconstrain(matrices x) This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html#transforms-for-scalars",
+    "href": "functions-reference/transform_functions.html#transforms-for-scalars",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "These transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.\nThe functions are all overloaded to apply to containers elementwise. If the y argument is a container, the others must be either scalars or containers of exactly the same size.\n\n\nThese functions perform the transform and inverse transform described in the Lower Bounded Scalar section.\n \n\nreals lower_bound_constrain(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_bound_jacobian(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_bound_unconstrain(reals x, reals lb) Takes a value x which is greater than lb and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Upper Bounded Scalar section.\n \n\nreals upper_bound_constrain(reals y, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals upper_bound_jacobian(reals x, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals upper_bound_unconstrain(reals x, reals ub) Takes a value x which is less than ub and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Lower and Upper Bounded Scalar section.\n \n\nreals lower_upper_bound_constrain(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_upper_bound_jacobian(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_upper_bound_unconstrain(reals x, reals lb, reals ub) Takes a value x which is bounded between lb and ub and returns returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Affinely Transformed Scalar section.\n \n\nreals offset_multiplier_constrain(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nAvailable since 2.37\n \n\nreals offset_multiplier_jacobian(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals offset_multiplier_unconstrain(reals x, reals offset, reals mult) Takes a value x, shift offset, and scale mult and a value which has been un-scaled and un-shifted.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html#transforms-for-constrained-vectors",
+    "href": "functions-reference/transform_functions.html#transforms-for-constrained-vectors",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "These functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.\nWhere vectors is used, this indicates that either a vector or a (possibly multidimensional) array of vectors may be provided. The array will be processed element by element.\n\n\nThese functions perform the transform and inverse transform described in the Ordered Vector section.\n \n\nvectors ordered_constrain(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nAvailable since 2.37\n \n\nvectors ordered_jacobian(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors ordered_unconstrain(vectors x) Takes an ordered vector x and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Positive Ordered Vector section.\n \n\nvectors positive_ordered_constrain(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nAvailable since 2.37\n \n\nvectors positive_ordered_jacobian(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors positive_ordered_unconstrain(vectors x) Takes an ordered vector x with positive entries and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Simplex section.\n \n\nvectors simplex_constrain(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors simplex_jacobian(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors simplex_unconstrain(vectors x) Takes a simplex x and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Zero Sum Vector section.\n \n\nvectors sum_to_zero_constrain(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors sum_to_zero_jacobian(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThe returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors sum_to_zero_unconstrain(vectors x) Takes a vector x with elements that sum to 0 and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Vector section.\n \n\nvectors unit_vectors_constrain(vectors y) Takes a free vector y and returns a vector with unit length, i.e., norm2(unit_vectors_constrain(y)) == 1 for any y that has a positive and finite norm itself (if y does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.\nAvailable since 2.37\n \n\nvectors unit_vectors_jacobian(vectors y) Takes a free vector y and returns a vector with unit length. This function rejects if given a vector of all zeros.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors unit_vectors_unconstrain(vectors x) Takes a vector x of unit length and returns the corresponding free vector.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html#transforms-for-constrained-matrices",
+    "href": "functions-reference/transform_functions.html#transforms-for-constrained-matrices",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "Similarly to the above, vectors means a vector or array thereof, and matrices means a matrix or array thereof.\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Correlation Matrices section.\n \n\nmatrices cholesky_factor_corr_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.\nAvailable since 2.37\n \n Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment. matrices cholesky_factor_corr_jacobian(vectors y, int K)\nAvailable since 2.37\n \n\nvectors cholesky_factor_corr_unconstrain(matrices x) Takes x, a (\\(K \\times K\\)) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Covariance Matrices section.\n \n\nmatrices cholesky_factor_cov_constrain(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nAvailable since 2.37\n \n\nmatrices cholesky_factor_cov_jacobian(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cholesky_factor_cov_unconstrain(matrices x) Takes a \\(M \\times N\\) matrix x which is a Cholesky factor of a covariance matrix (a matrix \\(L\\) such that \\(L^T L\\) is positive definite) and returns the corresponding free vector of length \\(N + \\binom{N}{2} + (M - N)N\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Correlation Matrices section.\n \n\nmatrices corr_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nAvailable since 2.37\n \n\nmatrices corr_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors corr_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) matrix x which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size \\(\\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Covariance Matrices section.\n \n\nmatrices cov_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nAvailable since 2.37\n \n\nmatrices cov_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cov_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) positive definite matrix x and returns the corresponding free vector of size \\(K + \\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for column (left) stochastic matrices.\n \n\nmatrices stochastic_column_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nAvailable since 2.37\n \n\nmatrices stochastic_column_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_column_unconstrain(matrices x) Takes a left stochastic matrix x of size \\(N+1 \\times M\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for row (right) stochastic matrices.\n \n\nmatrices stochastic_row_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nAvailable since 2.37\n \n\nmatrices stochastic_row_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_row_unconstrain(matrices x) Takes a right stochastic matrix x of size \\(N \\times M+1\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThe sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the sum-to-zero matrix section of the Reference Manual.\n \n\nmatrices sum_to_zero_constrain(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_jacobian(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to sum_to_zero_constrain.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_unconstrain(matrices x) This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html",
+    "href": "functions-reference/unbounded_discrete_distributions.html",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "The unbounded discrete distributions have support over the natural numbers (i.e., the non-negative integers).\n\n\nFor the negative binomial distribution Stan uses the parameterization described in Gelman et al. (2013). For alternative parameterizations, see section negative binomial glm.\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n\n\\in \\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial}(n~|~\\alpha,\\beta) = \\binom{n +\n\\alpha - 1}{\\alpha - 1} \\, \\left( \\frac{\\beta}{\\beta+1}\n\\right)^{\\!\\alpha} \\, \\left( \\frac{1}{\\beta + 1} \\right)^{\\!n} \\!. \\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial}(\\alpha,\\beta)\\) are given by \\[\\begin{equation*} \\mathbb{E}[n] =\n\\frac{\\alpha}{\\beta} \\ \\ \\text{ and } \\ \\ \\text{Var}[n] =\n\\frac{\\alpha}{\\beta^2} (\\beta + 1). \\end{equation*}\\]\n\n\n\nn ~ neg_binomial(alpha, beta)\nIncrement target log probability density with neg_binomial_lupmf(n | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal neg_binomial_lpmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lupmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal neg_binomial_cdf(ints n | reals alpha, reals beta) The negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal neg_binomial_lcdf(ints n | reals alpha, reals beta) The log of the negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lccdf(ints n | reals alpha, reals beta) The log of the negative binomial complementary cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR neg_binomial_rng(reals alpha, reals beta) Generate a negative binomial variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) beta must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section combinatorial functions, below, provides a second alternative parameterization directly in terms of the log mean.\n\n\nThe first parameterization is for \\(\\mu \\in \\mathbb{R}^+\\) and \\(\\phi \\in\n\\mathbb{R}^+\\), which for \\(n \\in \\mathbb{N}\\) is defined as \\[\\begin{equation*}\n\\text{NegBinomial2}(n \\, | \\, \\mu, \\phi)\n= \\binom{n + \\phi - 1}{n} \\, \\left( \\frac{\\mu}{\\mu+\\phi} \\right)^{\\!n} \\, \\left( \\frac{\\phi}{\\mu+\\phi} \\right)^{\\!\\phi} \\!.\n\\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial2}(n~|~\\mu,\\phi)\\) are \\[\\begin{equation*}\n\\mathbb{E}[n] = \\mu \\ \\ \\ \\text{ and } \\ \\ \\ \\text{Var}[n] = \\mu + \\frac{\\mu^2}{\\phi}.\n\\end{equation*}\\] Recall that \\(\\text{Poisson}(\\mu)\\) has variance \\(\\mu\\), so \\(\\mu^2 / \\phi &gt; 0\\) is the additional variance of the negative binomial above that of the Poisson with mean \\(\\mu\\). So the inverse of parameter \\(\\phi\\) controls the overdispersion, scaled by the square of the mean, \\(\\mu^2\\).\n\n\n\nn ~ neg_binomial_2(mu, phi)\nIncrement target log probability density with neg_binomial_2_lupmf(n | mu, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_lpmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_lupmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_cdf(ints n | reals mu, reals phi) The negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.6\n \n\nreal neg_binomial_2_lcdf(ints n | reals mu, reals phi) The log of the negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nreal neg_binomial_2_lccdf(ints n | reals mu, reals phi) The log of the negative binomial complementary cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nR neg_binomial_2_rng(reals mu, reals phi) Generate a negative binomial variate with location mu and precision phi; may only be used in transformed data and generated quantities blocks. mu must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nRelated to the parameterization in section negative binomial, alternative parameterization, the following parameterization uses a log mean parameter \\(\\eta = \\log(\\mu)\\), defined for \\(\\eta \\in \\mathbb{R}\\), \\(\\phi \\in \\mathbb{R}^+\\), so that for \\(n \\in\n\\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial2Log}(n \\, | \\, \\eta, \\phi) =\n\\text{NegBinomial2}(n | \\exp(\\eta), \\phi). \\end{equation*}\\] This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.\n\n\nn ~ neg_binomial_2_log(eta, phi)\nIncrement target log probability density with neg_binomial_2_log_lupmf(n | eta, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_log_lpmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_log_lupmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nR neg_binomial_2_log_rng(reals eta, reals phi) Generate a negative binomial variate with log-location eta and inverse overdispersion control phi; may only be used in transformed data and generated quantities blocks. eta must be less than \\(29 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e. a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\phi\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{NegBinomial2LogGLM}(y~|~x, \\alpha, \\beta, \\phi) = \\prod_{1\\leq i\n\\leq n}\\text{NegBinomial2}(y_i~|~\\exp(\\alpha_i + x_i\\cdot \\beta),\n\\phi). \\end{equation*}\\]\n\n\n\ny ~ neg_binomial_2_log_glm(x, alpha, beta, phi)\nIncrement target log probability density with neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(\\lambda \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{Poisson}(n|\\lambda) = \\frac{1}{n!} \\, \\lambda^n \\,\n\\exp(-\\lambda). \\end{equation*}\\]\n\n\n\nn ~ poisson(lambda)\nIncrement target log probability density with poisson_lupmf(n | lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_lpmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lupmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal poisson_cdf(ints n | reals lambda) The Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.0\n \n\nreal poisson_lcdf(ints n | reals lambda) The log of the Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lccdf(ints n | reals lambda) The log of the Poisson complementary cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nR poisson_rng(reals lambda) Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than \\(2^{30}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also provides a parameterization of the Poisson using the log rate \\(\\alpha = \\log \\lambda\\) as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{PoissonLog}(n|\\alpha) = \\frac{1}{n!} \\, \\exp \\left(n\\alpha -\n\\exp(\\alpha) \\right). \\end{equation*}\\]\n\n\n\nn ~ poisson_log(alpha)\nIncrement target log probability density with poisson_log_lupmf(n | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_log_lpmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha\nAvailable since 2.12\n \n\nreal poisson_log_lupmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha dropping constant additive terms\nAvailable since 2.25\n \n\nR poisson_log_rng(reals alpha) Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than \\(30 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e. a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{PoissonLogGLM}(y|x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Poisson}(y_i|\\exp(\\alpha_i + x_i\\cdot \\beta)). \\end{equation*}\\]\n\n\n\ny ~ poisson_log_glm(x, alpha, beta)\nIncrement target log probability density with poisson_log_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(r \\in \\mathbb{R}^+\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{BetaNegBinomial}(n|r,\\alpha,\\beta) = \\frac {\\Gamma (n+r )}{n!\\;\\Gamma (r )}\n\\frac {\\mathrm {B} (\\beta+n,\\alpha +r )}{\\mathrm {B} (\\beta,\\alpha )}. \\end{equation*}\\]\n\n\n\nn ~ beta_neg_binomial(r,alpha,beta)\nIncrement target log probability density with beta_neg_binomial_lupmf(n | r, alpha, beta).\nAvailable since 2.36\n \n\n\n\n\n \n\nreal beta_neg_binomial_lpmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lupmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta dropping constant additive terms.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_cdf(ints n | reals r, reals alpha, reals beta) The beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lcdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lccdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial complementary cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nR beta_neg_binomial_rng(reals r, reals alpha, reals beta) Generate a beta negative binomial variate with parameters r, alpha and beta; may only be used in transformed data and generated quantities blocks. r \\(\\cdot\\) beta \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.36\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}^+=\\{1,2,...\\}\\), \\[\\begin{equation*}\n\\text{YuleSimon}(n|\\alpha) = \\alpha \\, \\mathrm{B}(\\alpha + 1, n) = \\alpha \\, \\frac{\\Gamma(n) \\, \\Gamma(\\alpha + 1)}{\\Gamma(n + \\alpha + 1)}.\n\\end{equation*}\\]\n\n\n\nn ~ yule_simon(alpha)\nIncrement target log probability density with yule_simon_lupmf(n | alpha).\nAvailable since 2.39\n \n\n\n\n\n \n\nreal yule_simon_lpmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lupmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha dropping constant additive terms.\nAvailable since 2.39\n \n\nreal yule_simon_cdf(ints n | reals alpha) The Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lcdf(ints n | reals alpha) The log of the Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lccdf(ints n | reals alpha) The log of the Yule Simon complementary cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nR yule_simon_rng(reals alpha) Generate a Yule Simon variate with parameter alpha; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#negative-binomial-distribution",
+    "href": "functions-reference/unbounded_discrete_distributions.html#negative-binomial-distribution",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "For the negative binomial distribution Stan uses the parameterization described in Gelman et al. (2013). For alternative parameterizations, see section negative binomial glm.\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n\n\\in \\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial}(n~|~\\alpha,\\beta) = \\binom{n +\n\\alpha - 1}{\\alpha - 1} \\, \\left( \\frac{\\beta}{\\beta+1}\n\\right)^{\\!\\alpha} \\, \\left( \\frac{1}{\\beta + 1} \\right)^{\\!n} \\!. \\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial}(\\alpha,\\beta)\\) are given by \\[\\begin{equation*} \\mathbb{E}[n] =\n\\frac{\\alpha}{\\beta} \\ \\ \\text{ and } \\ \\ \\text{Var}[n] =\n\\frac{\\alpha}{\\beta^2} (\\beta + 1). \\end{equation*}\\]\n\n\n\nn ~ neg_binomial(alpha, beta)\nIncrement target log probability density with neg_binomial_lupmf(n | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal neg_binomial_lpmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lupmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal neg_binomial_cdf(ints n | reals alpha, reals beta) The negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal neg_binomial_lcdf(ints n | reals alpha, reals beta) The log of the negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lccdf(ints n | reals alpha, reals beta) The log of the negative binomial complementary cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR neg_binomial_rng(reals alpha, reals beta) Generate a negative binomial variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) beta must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#nbalt",
+    "href": "functions-reference/unbounded_discrete_distributions.html#nbalt",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section combinatorial functions, below, provides a second alternative parameterization directly in terms of the log mean.\n\n\nThe first parameterization is for \\(\\mu \\in \\mathbb{R}^+\\) and \\(\\phi \\in\n\\mathbb{R}^+\\), which for \\(n \\in \\mathbb{N}\\) is defined as \\[\\begin{equation*}\n\\text{NegBinomial2}(n \\, | \\, \\mu, \\phi)\n= \\binom{n + \\phi - 1}{n} \\, \\left( \\frac{\\mu}{\\mu+\\phi} \\right)^{\\!n} \\, \\left( \\frac{\\phi}{\\mu+\\phi} \\right)^{\\!\\phi} \\!.\n\\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial2}(n~|~\\mu,\\phi)\\) are \\[\\begin{equation*}\n\\mathbb{E}[n] = \\mu \\ \\ \\ \\text{ and } \\ \\ \\ \\text{Var}[n] = \\mu + \\frac{\\mu^2}{\\phi}.\n\\end{equation*}\\] Recall that \\(\\text{Poisson}(\\mu)\\) has variance \\(\\mu\\), so \\(\\mu^2 / \\phi &gt; 0\\) is the additional variance of the negative binomial above that of the Poisson with mean \\(\\mu\\). So the inverse of parameter \\(\\phi\\) controls the overdispersion, scaled by the square of the mean, \\(\\mu^2\\).\n\n\n\nn ~ neg_binomial_2(mu, phi)\nIncrement target log probability density with neg_binomial_2_lupmf(n | mu, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_lpmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_lupmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_cdf(ints n | reals mu, reals phi) The negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.6\n \n\nreal neg_binomial_2_lcdf(ints n | reals mu, reals phi) The log of the negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nreal neg_binomial_2_lccdf(ints n | reals mu, reals phi) The log of the negative binomial complementary cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nR neg_binomial_2_rng(reals mu, reals phi) Generate a negative binomial variate with location mu and precision phi; may only be used in transformed data and generated quantities blocks. mu must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log",
+    "href": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "Related to the parameterization in section negative binomial, alternative parameterization, the following parameterization uses a log mean parameter \\(\\eta = \\log(\\mu)\\), defined for \\(\\eta \\in \\mathbb{R}\\), \\(\\phi \\in \\mathbb{R}^+\\), so that for \\(n \\in\n\\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial2Log}(n \\, | \\, \\eta, \\phi) =\n\\text{NegBinomial2}(n | \\exp(\\eta), \\phi). \\end{equation*}\\] This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.\n\n\nn ~ neg_binomial_2_log(eta, phi)\nIncrement target log probability density with neg_binomial_2_log_lupmf(n | eta, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_log_lpmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_log_lupmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nR neg_binomial_2_log_rng(reals eta, reals phi) Generate a negative binomial variate with log-location eta and inverse overdispersion control phi; may only be used in transformed data and generated quantities blocks. eta must be less than \\(29 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log-glm",
+    "href": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log-glm",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e. a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\phi\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{NegBinomial2LogGLM}(y~|~x, \\alpha, \\beta, \\phi) = \\prod_{1\\leq i\n\\leq n}\\text{NegBinomial2}(y_i~|~\\exp(\\alpha_i + x_i\\cdot \\beta),\n\\phi). \\end{equation*}\\]\n\n\n\ny ~ neg_binomial_2_log_glm(x, alpha, beta, phi)\nIncrement target log probability density with neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson",
+    "href": "functions-reference/unbounded_discrete_distributions.html#poisson",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(\\lambda \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{Poisson}(n|\\lambda) = \\frac{1}{n!} \\, \\lambda^n \\,\n\\exp(-\\lambda). \\end{equation*}\\]\n\n\n\nn ~ poisson(lambda)\nIncrement target log probability density with poisson_lupmf(n | lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_lpmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lupmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal poisson_cdf(ints n | reals lambda) The Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.0\n \n\nreal poisson_lcdf(ints n | reals lambda) The log of the Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lccdf(ints n | reals lambda) The log of the Poisson complementary cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nR poisson_rng(reals lambda) Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than \\(2^{30}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson-distribution-log-parameterization",
+    "href": "functions-reference/unbounded_discrete_distributions.html#poisson-distribution-log-parameterization",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides a parameterization of the Poisson using the log rate \\(\\alpha = \\log \\lambda\\) as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{PoissonLog}(n|\\alpha) = \\frac{1}{n!} \\, \\exp \\left(n\\alpha -\n\\exp(\\alpha) \\right). \\end{equation*}\\]\n\n\n\nn ~ poisson_log(alpha)\nIncrement target log probability density with poisson_log_lupmf(n | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_log_lpmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha\nAvailable since 2.12\n \n\nreal poisson_log_lupmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha dropping constant additive terms\nAvailable since 2.25\n \n\nR poisson_log_rng(reals alpha) Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than \\(30 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson-log-glm",
+    "href": "functions-reference/unbounded_discrete_distributions.html#poisson-log-glm",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "Stan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e. a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{PoissonLogGLM}(y|x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Poisson}(y_i|\\exp(\\alpha_i + x_i\\cdot \\beta)). \\end{equation*}\\]\n\n\n\ny ~ poisson_log_glm(x, alpha, beta)\nIncrement target log probability density with poisson_log_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#beta-neg-binomial",
+    "href": "functions-reference/unbounded_discrete_distributions.html#beta-neg-binomial",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(r \\in \\mathbb{R}^+\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{BetaNegBinomial}(n|r,\\alpha,\\beta) = \\frac {\\Gamma (n+r )}{n!\\;\\Gamma (r )}\n\\frac {\\mathrm {B} (\\beta+n,\\alpha +r )}{\\mathrm {B} (\\beta,\\alpha )}. \\end{equation*}\\]\n\n\n\nn ~ beta_neg_binomial(r,alpha,beta)\nIncrement target log probability density with beta_neg_binomial_lupmf(n | r, alpha, beta).\nAvailable since 2.36\n \n\n\n\n\n \n\nreal beta_neg_binomial_lpmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lupmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta dropping constant additive terms.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_cdf(ints n | reals r, reals alpha, reals beta) The beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lcdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lccdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial complementary cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nR beta_neg_binomial_rng(reals r, reals alpha, reals beta) Generate a beta negative binomial variate with parameters r, alpha and beta; may only be used in transformed data and generated quantities blocks. r \\(\\cdot\\) beta \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.36",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#yule-simon",
+    "href": "functions-reference/unbounded_discrete_distributions.html#yule-simon",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}^+=\\{1,2,...\\}\\), \\[\\begin{equation*}\n\\text{YuleSimon}(n|\\alpha) = \\alpha \\, \\mathrm{B}(\\alpha + 1, n) = \\alpha \\, \\frac{\\Gamma(n) \\, \\Gamma(\\alpha + 1)}{\\Gamma(n + \\alpha + 1)}.\n\\end{equation*}\\]\n\n\n\nn ~ yule_simon(alpha)\nIncrement target log probability density with yule_simon_lupmf(n | alpha).\nAvailable since 2.39\n \n\n\n\n\n \n\nreal yule_simon_lpmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lupmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha dropping constant additive terms.\nAvailable since 2.39\n \n\nreal yule_simon_cdf(ints n | reals alpha) The Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lcdf(ints n | reals alpha) The log of the Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lccdf(ints n | reals alpha) The log of the Yule Simon complementary cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nR yule_simon_rng(reals alpha) Generate a Yule Simon variate with parameter alpha; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "reference-manual/analysis.html",
+    "href": "reference-manual/analysis.html",
+    "title": "Posterior Analysis",
+    "section": "",
+    "text": "Stan uses Markov chain Monte Carlo (MCMC) techniques to generate draws from the posterior distribution for full Bayesian inference. Markov chain Monte Carlo (MCMC) methods were developed for situations in which it is not straightforward to make independent draws Metropolis et al. (1953).\nStan’s variational inference algorithm provides draws from the variational approximation to the posterior which may be analyzed just as any other MCMC output, despite the fact that it is not actually a Markov chain.\nStan’s Laplace algorithm produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood.\n\n\nA Markov chain is a sequence of random variables \\(\\theta^{(1)},\n\\theta^{(2)},\\ldots\\) where each variable is conditionally independent of all other variables given the value of the previous value. Thus if \\(\\theta = \\theta^{(1)}, \\theta^{(2)},\\ldots, \\theta^{(N)}\\), then\n\\[\np(\\theta) = p(\\theta^{(1)}) \\prod_{n=2}^N p(\\theta^{(n)}|\\theta^{(n-1)}).\n\\]\nStan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the Hamiltonian Monte Carlo chapter.\nThe Markov chains Stan and other MCMC samplers generate are ergodic in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of \\(\\theta\\) from another. The Markov chains are also stationary, meaning that the transition probabilities do not change at different positions in the chain, so that for \\(n, n' \\geq 0\\), the probability function \\(p(\\theta^{(n+1)}|\\theta^{(n)})\\) is the same as \\(p(\\theta^{(n'+1)}|\\theta^{(n')})\\) (following the convention of overloading random and bound variables and picking out a probability function by its arguments).\nStationary Markov chains have an equilibrium distribution on states in which each has the same marginal probability function, so that \\(p(\\theta^{(n)})\\) is the same probability function as \\(p(\\theta^{(n+1)})\\). In Stan, this equilibrium distribution \\(p(\\theta^{(n)})\\) is the target density \\(p(\\theta)\\) defined by a Stan program, which is typically a proper Bayesian posterior density \\(p(\\theta | y)\\) defined on the log scale up to a constant.\nUsing MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.\nStan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.\n\n\n\nBy definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when \\(p(\\theta^{(n)})\\) is the target density). The following point cannot be expressed strongly enough:\n\nIn theory, convergence is only guaranteed asymptotically as the number of draws grows without bound.\nIn practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available.\n\n\n\n\nTo establish basic notation, suppose a target Bayesian posterior density \\(p(\\theta | y)\\) given real-valued vectors of parameters \\(\\theta\\) and real- and discrete-valued data \\(y\\).1\nAn MCMC sample consists of a set of a sequence of \\(M\\) Markov chains, each consisting of an ordered sequence of \\(N\\) draws from the posterior.2 The sample thus consists of \\(M \\times N\\) draws from the posterior.\n\n\nOne way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the Gelman and Rubin (1992) potential scale reduction statistic, \\(\\hat{R}\\). The \\(\\hat{R}\\) statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and \\(\\hat{R}\\) will be one. If the chains have not converged to a common distribution, the \\(\\hat{R}\\) statistic will be greater than one.\nGelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for \\(\\hat{R}\\) are below some threshold. Vehtari et al. (2021) suggest in general to use a threshold \\(1.01\\), but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.\nThe \\(\\hat{R}\\) statistic is defined for a set of \\(M\\) Markov chains, \\(\\theta_m\\), each of which has \\(N\\) draws \\(\\theta^{(n)}_m\\). The between-chain variance estimate is\n\\[\nB\n=\n\\frac{N}{M-1}\n\\,\n\\sum_{m=1}^M (\\bar{\\theta}^{(\\bullet)}_{m}\n                - \\bar{\\theta}^{(\\bullet)}_{\\bullet})^2,\n\\]\nwhere\n\\[\n\\bar{\\theta}_m^{(\\bullet)}\n= \\frac{1}{N} \\sum_{n = 1}^N \\theta_m^{(n)}\n\\]\nand\n\\[\n\\bar{\\theta}^{(\\bullet)}_{\\bullet}\n= \\frac{1}{M} \\, \\sum_{m=1}^M \\bar{\\theta}_m^{(\\bullet)}.\n\\]\nThe within-chain variance is averaged over the chains,\n\\[\nW = \\frac{1}{M} \\, \\sum_{m=1}^M s_m^2,\n\\]\nwhere\n\\[\ns_m^2\n=\n\\frac{1}{N-1}\n\\, \\sum_{n=1}^N (\\theta^{(n)}_m - \\bar{\\theta}^{(\\bullet)}_m)^2.\n\\]\nThe variance estimator is a mixture of the within-chain and cross-chain sample variances,\n\\[\n\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)\n= \\frac{N-1}{N}\\, W \\, + \\, \\frac{1}{N} \\, B.\n\\]\nFinally, the potential scale reduction statistic is defined by\n\\[\n\\hat{R}\n\\, = \\,\n\\sqrt{\\frac{\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)}{W}}.\n\\]\n\n\n\nBefore Stan calculating the potential-scale-reduction statistic \\(\\hat{R}\\), each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have \\(\\hat{R}\\) values near unity. In this case, splitting each chain into two parts leads to \\(\\hat{R}\\) values substantially greater than 1 because the first half of each chain has not mixed with the second half.\n\n\n\nSplit R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following Vehtari et al. (2021), we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.\nRank normalization proceeds as follows:\n\nFirst, replace each value \\(\\theta^{(nm)}\\) by its rank \\(r^{(nm)}\\) within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.\nSecond, transform ranks to normal scores using the inverse normal transformation and a fractional offset:\n\n\\[\nz_{(nm)} = \\Phi^{-1} \\left( \\frac{r_{(nm)} - 3/8}{S - 1/4} \\right)\n\\]\nTo further improve sensitivity to chains having different scales,\nrank normalized R-hat is computed also for the for the corresponding folded draws \\(\\zeta^{(mn)}\\), absolute deviations from the median, \\[\n\\label{zeta}\n\\zeta^{(mn)} = \\left|\\theta^{(nm)}-{\\rm median}(\\theta)\\right|.\n\\] The rank normalized split-\\(\\widehat{R}\\) measure computed on the \\(\\zeta^{(mn)}\\) values is called -\\(\\widehat{R}\\). This measures convergence in the tails rather than in the bulk of the distribution.\nTo obtain a single conservative \\(\\widehat{R}\\) estimate, we propose to report the maximum of rank normalized split-\\(\\widehat{R}\\) and rank normalized folded-split-\\(\\widehat{R}\\) for each parameter.\nBulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See Effective Sample Size section for details on how ESS is estimated.\n\n\n\nA question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.\nFor example, consider the value lp__, which is the log posterior density (up to a constant).3\nIt is thus a mistake to declare convergence in any practical sense if lp__ has not converged, because different chains are really in different parts of the space. Yet measuring convergence for lp__ is particularly tricky, as noted below.\n\n\nMarkov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.\n\n\n\nThe \\(\\hat{R}\\) statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.\nThe transformation from unconstrained space to constrained space is just another function, so does not effect convergence.\nDifferent functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, lp__ and other measured quantities should at least be consistent.\nThe obvious difference in lp__ is that it tends to vary quickly with position and is consequently susceptible to outliers.\n\n\n\nThe question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.\nSo what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.\n\n\n\nFirstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better \\(\\hat{R}\\). Functions vulnerable to such outliers may show fragility.\nSecondly, use of the \\(\\hat{R}\\) statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (lp__) almost never looks Gaussian, instead it features long tails that can lead to large \\(\\hat{R}\\) even in the large \\(N\\) limit. Tweaks to \\(\\hat{R}\\), such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the \\(\\hat{R}\\) statistic more accurate.\n\n\n\n“Convergence” is a global property and holds for all integrable functions at once, but employing the \\(\\hat{R}\\) statistic requires additional assumptions and thus may not work for all functions equally well.\nNote that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.\n\n\n\n\n\nThe second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see Charles J. Geyer (2011).\nStan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).\nUnlike most packages, the particular calculations used by Stan follow those for split-\\(\\hat{R}\\), which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see Gelman et al. (2013) and Vehtari et al. (2021).\n\n\nThe amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size \\(N\\). Given dependent sample, the sample size is replaced with the effective sample size \\(N_{\\mathrm{eff}}\\).\nFor example, Monte Carlo standard error (MCSE) is proportional to \\(1 / \\sqrt{N_{\\mathrm{eff}}}\\) rather than \\(1/\\sqrt{N}\\).\nThe effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation \\(\\rho_t\\) at lag \\(t \\geq 0\\) for a chain with joint probability function \\(p(\\theta)\\) with mean \\(\\mu\\) and variance \\(\\sigma^2\\) is defined to be\n\\[\n\\rho_t\n=\n\\frac{1}{\\sigma^2} \\, \\int_{\\Theta} (\\theta^{(n)} - \\mu)\n(\\theta^{(n+t)} - \\mu) \\, p(\\theta) \\, d\\theta.\n\\]\nThis is the correlation between the two chains offset by \\(t\\) positions (i.e., a lag in time-series terminology). Because we know \\(\\theta^{(n)}\\) and \\(\\theta^{(n+t)}\\) have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields\n\\[\n\\rho_t\n= \\frac{1}{\\sigma^2}\n\\, \\int_{\\Theta}\n      \\theta^{(n)} \\, \\theta^{(n+t)} \\, p(\\theta)\n   \\, d\\theta - \\frac{\\mu^2}{\\sigma^2}.\n\\]\nThe effective sample size of \\(N\\) draws generated by a process with autocorrelations \\(\\rho_t\\) is defined by \\[\nN_{\\mathrm{eff}}\n\\ = \\\n\\frac{N}{\\sum_{t = -\\infty}^{\\infty} \\rho_t}\n\\ = \\\n\\frac{N}{1 + 2 \\sum_{t = 1}^{\\infty} \\rho_t}.\n\\]\nFor independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.\n\n\n\nIn practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-\\(\\hat{R}\\) based effective sample size estimator, based on multiple chains. As before, each chain \\(\\theta_m\\) will be assumed to be of length \\(N\\).\nStan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see Charles J. Geyer (2011) for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates \\(\\hat{\\rho}_{t,m}\\) at lag \\(t\\) from multiple chains \\(m \\in\n(1,\\ldots,M)\\) are combined with within-sample variance estimate \\(W\\) and multi-chain variance estimate \\(\\widehat{\\mbox{var}}^{+}\\) introduced in the previous section to compute the combined autocorrelation at lag \\(t\\) as\n\\[\n\\hat{\\rho}_t\n= 1 - \\frac{\\displaystyle W\n              - \\textstyle \\frac{1}{M}\\sum_{m=1}^M s_m^2 \\hat{\\rho}_{t,m}}\n        {\\widehat{\\mbox{var}}^{+}}.\n\\]\nIf the chains have not converged, the variance estimator \\(\\widehat{\\mbox{var}}^{+}\\) will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.\nBecause of the noise in the correlation estimates \\(\\hat{\\rho}_t\\) as \\(t\\) increases, a typical truncated sum of \\(\\hat{\\rho}_t\\) is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise Charles J. Geyer (1992), Charles J. Geyer (2011). Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as\n\\[\n\\hat{N}_{\\mathrm{eff}} = \\frac{M \\cdot N}{\\hat{\\tau}},\n\\]\nwhere\n\\[\n\\hat{\\tau} = 1 + 2 \\sum_{t=1}^{2m+1} \\hat{\\rho}_t = -1 + 2 \\sum_{t'=0}^{m} \\hat{P}_{t'},\n\\]\nwhere \\(\\hat{P}_{t'}=\\hat{\\rho}_{2t'}+\\hat{\\rho}_{2t'+1}\\). Initial positive sequence estimators is obtained by choosing the largest \\(m\\) such that \\(\\hat{P}_{t'}&gt;0, \\quad t' = 1,\\ldots,m\\). The initial monotone sequence is obtained by further reducing \\(\\hat{P}_{t'}\\) to the minimum of the preceding ones so that the estimated sequence is monotone.\n\n\n\nThe posterior standard deviation of a parameter \\(\\theta_n\\) conditioned on observed data \\(y\\) is just the standard deviation of the posterior density \\(p(\\theta_n | y)\\). This is estimated by the standard deviation of the combined posterior draws across chains,\n\\[\n\\hat{\\sigma}_n = \\mathrm{sd}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n).\n\\]\nThe previous section showed how to estimate \\(N_{\\mathrm{eff}}\\) for a parameter \\(\\theta_n\\) based on multiple chains of posterior draws.\nThe mean of the posterior draws of \\(\\theta_n\\) \\[\n\\hat{\\theta}_n\n= \\mathrm{mean}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n)\n\\]\nis treated as an estimator of the true posterior mean,\n\\[\n\\mathbb{E}[\\theta_n \\mid y]\n\\ = \\\n\\int_{-\\infty}^{\\infty}\n    \\, \\theta \\, p(\\theta | y)\n\\, \\mathrm{d}\\theta_n,\n\\]\nbased the observed data \\(y\\).\nThe standard error for the estimator \\(\\hat{\\theta}_n\\) is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as \\(\\hat{\\sigma}_n / \\sqrt{N_{\\mathrm{eff}}}\\). The smaller the standard error, the closer the estimate \\(\\hat{\\theta}_n\\) is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see Charles J. Geyer (2011) for more details of the MCMC central limit theorem.\n\n\n\nIn complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag \\(t\\), \\(\\rho_t\\), decreases as the lag, \\(t\\), increases. In this situation, thinning the sample by keeping only every \\(N\\)-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.\nFor instance, consider generating one thousand posterior draws in one of the following two ways.\n\nGenerate 1000 draws after convergence and save all of them.\nGenerate 10,000 draws after convergence and save every tenth draw.\n\nEven though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation \\(\\rho_t\\) for the thinned sequence is equivalent to \\(\\rho_{10t}\\) in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.\nNow contrast the second approach above with the unthinned alternative,\n\nGenerate 10,000 draws after convergence and save every draw.\n\nThis will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, the only reason to thin a sample is to reduce memory requirements.\nIf draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Posterior Analysis"
+    ]
+  },
+  {
+    "objectID": "reference-manual/analysis.html#markov-chains",
+    "href": "reference-manual/analysis.html#markov-chains",
+    "title": "Posterior Analysis",
+    "section": "",
+    "text": "A Markov chain is a sequence of random variables \\(\\theta^{(1)},\n\\theta^{(2)},\\ldots\\) where each variable is conditionally independent of all other variables given the value of the previous value. Thus if \\(\\theta = \\theta^{(1)}, \\theta^{(2)},\\ldots, \\theta^{(N)}\\), then\n\\[\np(\\theta) = p(\\theta^{(1)}) \\prod_{n=2}^N p(\\theta^{(n)}|\\theta^{(n-1)}).\n\\]\nStan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the Hamiltonian Monte Carlo chapter.\nThe Markov chains Stan and other MCMC samplers generate are ergodic in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of \\(\\theta\\) from another. The Markov chains are also stationary, meaning that the transition probabilities do not change at different positions in the chain, so that for \\(n, n' \\geq 0\\), the probability function \\(p(\\theta^{(n+1)}|\\theta^{(n)})\\) is the same as \\(p(\\theta^{(n'+1)}|\\theta^{(n')})\\) (following the convention of overloading random and bound variables and picking out a probability function by its arguments).\nStationary Markov chains have an equilibrium distribution on states in which each has the same marginal probability function, so that \\(p(\\theta^{(n)})\\) is the same probability function as \\(p(\\theta^{(n+1)})\\). In Stan, this equilibrium distribution \\(p(\\theta^{(n)})\\) is the target density \\(p(\\theta)\\) defined by a Stan program, which is typically a proper Bayesian posterior density \\(p(\\theta | y)\\) defined on the log scale up to a constant.\nUsing MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.\nStan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Posterior Analysis"
+    ]
+  },
+  {
+    "objectID": "reference-manual/analysis.html#convergence",
+    "href": "reference-manual/analysis.html#convergence",
+    "title": "Posterior Analysis",
+    "section": "",
+    "text": "By definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when \\(p(\\theta^{(n)})\\) is the target density). The following point cannot be expressed strongly enough:\n\nIn theory, convergence is only guaranteed asymptotically as the number of draws grows without bound.\nIn practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Posterior Analysis"
+    ]
+  },
+  {
+    "objectID": "reference-manual/analysis.html#notation-for-samples-chains-and-draws",
+    "href": "reference-manual/analysis.html#notation-for-samples-chains-and-draws",
+    "title": "Posterior Analysis",
+    "section": "",
+    "text": "To establish basic notation, suppose a target Bayesian posterior density \\(p(\\theta | y)\\) given real-valued vectors of parameters \\(\\theta\\) and real- and discrete-valued data \\(y\\).1\nAn MCMC sample consists of a set of a sequence of \\(M\\) Markov chains, each consisting of an ordered sequence of \\(N\\) draws from the posterior.2 The sample thus consists of \\(M \\times N\\) draws from the posterior.\n\n\nOne way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the Gelman and Rubin (1992) potential scale reduction statistic, \\(\\hat{R}\\). The \\(\\hat{R}\\) statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and \\(\\hat{R}\\) will be one. If the chains have not converged to a common distribution, the \\(\\hat{R}\\) statistic will be greater than one.\nGelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for \\(\\hat{R}\\) are below some threshold. Vehtari et al. (2021) suggest in general to use a threshold \\(1.01\\), but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.\nThe \\(\\hat{R}\\) statistic is defined for a set of \\(M\\) Markov chains, \\(\\theta_m\\), each of which has \\(N\\) draws \\(\\theta^{(n)}_m\\). The between-chain variance estimate is\n\\[\nB\n=\n\\frac{N}{M-1}\n\\,\n\\sum_{m=1}^M (\\bar{\\theta}^{(\\bullet)}_{m}\n                - \\bar{\\theta}^{(\\bullet)}_{\\bullet})^2,\n\\]\nwhere\n\\[\n\\bar{\\theta}_m^{(\\bullet)}\n= \\frac{1}{N} \\sum_{n = 1}^N \\theta_m^{(n)}\n\\]\nand\n\\[\n\\bar{\\theta}^{(\\bullet)}_{\\bullet}\n= \\frac{1}{M} \\, \\sum_{m=1}^M \\bar{\\theta}_m^{(\\bullet)}.\n\\]\nThe within-chain variance is averaged over the chains,\n\\[\nW = \\frac{1}{M} \\, \\sum_{m=1}^M s_m^2,\n\\]\nwhere\n\\[\ns_m^2\n=\n\\frac{1}{N-1}\n\\, \\sum_{n=1}^N (\\theta^{(n)}_m - \\bar{\\theta}^{(\\bullet)}_m)^2.\n\\]\nThe variance estimator is a mixture of the within-chain and cross-chain sample variances,\n\\[\n\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)\n= \\frac{N-1}{N}\\, W \\, + \\, \\frac{1}{N} \\, B.\n\\]\nFinally, the potential scale reduction statistic is defined by\n\\[\n\\hat{R}\n\\, = \\,\n\\sqrt{\\frac{\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)}{W}}.\n\\]\n\n\n\nBefore Stan calculating the potential-scale-reduction statistic \\(\\hat{R}\\), each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have \\(\\hat{R}\\) values near unity. In this case, splitting each chain into two parts leads to \\(\\hat{R}\\) values substantially greater than 1 because the first half of each chain has not mixed with the second half.\n\n\n\nSplit R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following Vehtari et al. (2021), we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.\nRank normalization proceeds as follows:\n\nFirst, replace each value \\(\\theta^{(nm)}\\) by its rank \\(r^{(nm)}\\) within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.\nSecond, transform ranks to normal scores using the inverse normal transformation and a fractional offset:\n\n\\[\nz_{(nm)} = \\Phi^{-1} \\left( \\frac{r_{(nm)} - 3/8}{S - 1/4} \\right)\n\\]\nTo further improve sensitivity to chains having different scales,\nrank normalized R-hat is computed also for the for the corresponding folded draws \\(\\zeta^{(mn)}\\), absolute deviations from the median, \\[\n\\label{zeta}\n\\zeta^{(mn)} = \\left|\\theta^{(nm)}-{\\rm median}(\\theta)\\right|.\n\\] The rank normalized split-\\(\\widehat{R}\\) measure computed on the \\(\\zeta^{(mn)}\\) values is called -\\(\\widehat{R}\\). This measures convergence in the tails rather than in the bulk of the distribution.\nTo obtain a single conservative \\(\\widehat{R}\\) estimate, we propose to report the maximum of rank normalized split-\\(\\widehat{R}\\) and rank normalized folded-split-\\(\\widehat{R}\\) for each parameter.\nBulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See Effective Sample Size section for details on how ESS is estimated.\n\n\n\nA question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.\nFor example, consider the value lp__, which is the log posterior density (up to a constant).3\nIt is thus a mistake to declare convergence in any practical sense if lp__ has not converged, because different chains are really in different parts of the space. Yet measuring convergence for lp__ is particularly tricky, as noted below.\n\n\nMarkov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.\n\n\n\nThe \\(\\hat{R}\\) statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.\nThe transformation from unconstrained space to constrained space is just another function, so does not effect convergence.\nDifferent functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, lp__ and other measured quantities should at least be consistent.\nThe obvious difference in lp__ is that it tends to vary quickly with position and is consequently susceptible to outliers.\n\n\n\nThe question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.\nSo what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.\n\n\n\nFirstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better \\(\\hat{R}\\). Functions vulnerable to such outliers may show fragility.\nSecondly, use of the \\(\\hat{R}\\) statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (lp__) almost never looks Gaussian, instead it features long tails that can lead to large \\(\\hat{R}\\) even in the large \\(N\\) limit. Tweaks to \\(\\hat{R}\\), such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the \\(\\hat{R}\\) statistic more accurate.\n\n\n\n“Convergence” is a global property and holds for all integrable functions at once, but employing the \\(\\hat{R}\\) statistic requires additional assumptions and thus may not work for all functions equally well.\nNote that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Posterior Analysis"
+    ]
+  },
+  {
+    "objectID": "reference-manual/analysis.html#effective-sample-size.section",
+    "href": "reference-manual/analysis.html#effective-sample-size.section",
+    "title": "Posterior Analysis",
+    "section": "",
+    "text": "The second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see Charles J. Geyer (2011).\nStan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).\nUnlike most packages, the particular calculations used by Stan follow those for split-\\(\\hat{R}\\), which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see Gelman et al. (2013) and Vehtari et al. (2021).\n\n\nThe amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size \\(N\\). Given dependent sample, the sample size is replaced with the effective sample size \\(N_{\\mathrm{eff}}\\).\nFor example, Monte Carlo standard error (MCSE) is proportional to \\(1 / \\sqrt{N_{\\mathrm{eff}}}\\) rather than \\(1/\\sqrt{N}\\).\nThe effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation \\(\\rho_t\\) at lag \\(t \\geq 0\\) for a chain with joint probability function \\(p(\\theta)\\) with mean \\(\\mu\\) and variance \\(\\sigma^2\\) is defined to be\n\\[\n\\rho_t\n=\n\\frac{1}{\\sigma^2} \\, \\int_{\\Theta} (\\theta^{(n)} - \\mu)\n(\\theta^{(n+t)} - \\mu) \\, p(\\theta) \\, d\\theta.\n\\]\nThis is the correlation between the two chains offset by \\(t\\) positions (i.e., a lag in time-series terminology). Because we know \\(\\theta^{(n)}\\) and \\(\\theta^{(n+t)}\\) have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields\n\\[\n\\rho_t\n= \\frac{1}{\\sigma^2}\n\\, \\int_{\\Theta}\n      \\theta^{(n)} \\, \\theta^{(n+t)} \\, p(\\theta)\n   \\, d\\theta - \\frac{\\mu^2}{\\sigma^2}.\n\\]\nThe effective sample size of \\(N\\) draws generated by a process with autocorrelations \\(\\rho_t\\) is defined by \\[\nN_{\\mathrm{eff}}\n\\ = \\\n\\frac{N}{\\sum_{t = -\\infty}^{\\infty} \\rho_t}\n\\ = \\\n\\frac{N}{1 + 2 \\sum_{t = 1}^{\\infty} \\rho_t}.\n\\]\nFor independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.\n\n\n\nIn practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-\\(\\hat{R}\\) based effective sample size estimator, based on multiple chains. As before, each chain \\(\\theta_m\\) will be assumed to be of length \\(N\\).\nStan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see Charles J. Geyer (2011) for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates \\(\\hat{\\rho}_{t,m}\\) at lag \\(t\\) from multiple chains \\(m \\in\n(1,\\ldots,M)\\) are combined with within-sample variance estimate \\(W\\) and multi-chain variance estimate \\(\\widehat{\\mbox{var}}^{+}\\) introduced in the previous section to compute the combined autocorrelation at lag \\(t\\) as\n\\[\n\\hat{\\rho}_t\n= 1 - \\frac{\\displaystyle W\n              - \\textstyle \\frac{1}{M}\\sum_{m=1}^M s_m^2 \\hat{\\rho}_{t,m}}\n        {\\widehat{\\mbox{var}}^{+}}.\n\\]\nIf the chains have not converged, the variance estimator \\(\\widehat{\\mbox{var}}^{+}\\) will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.\nBecause of the noise in the correlation estimates \\(\\hat{\\rho}_t\\) as \\(t\\) increases, a typical truncated sum of \\(\\hat{\\rho}_t\\) is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise Charles J. Geyer (1992), Charles J. Geyer (2011). Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as\n\\[\n\\hat{N}_{\\mathrm{eff}} = \\frac{M \\cdot N}{\\hat{\\tau}},\n\\]\nwhere\n\\[\n\\hat{\\tau} = 1 + 2 \\sum_{t=1}^{2m+1} \\hat{\\rho}_t = -1 + 2 \\sum_{t'=0}^{m} \\hat{P}_{t'},\n\\]\nwhere \\(\\hat{P}_{t'}=\\hat{\\rho}_{2t'}+\\hat{\\rho}_{2t'+1}\\). Initial positive sequence estimators is obtained by choosing the largest \\(m\\) such that \\(\\hat{P}_{t'}&gt;0, \\quad t' = 1,\\ldots,m\\). The initial monotone sequence is obtained by further reducing \\(\\hat{P}_{t'}\\) to the minimum of the preceding ones so that the estimated sequence is monotone.\n\n\n\nThe posterior standard deviation of a parameter \\(\\theta_n\\) conditioned on observed data \\(y\\) is just the standard deviation of the posterior density \\(p(\\theta_n | y)\\). This is estimated by the standard deviation of the combined posterior draws across chains,\n\\[\n\\hat{\\sigma}_n = \\mathrm{sd}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n).\n\\]\nThe previous section showed how to estimate \\(N_{\\mathrm{eff}}\\) for a parameter \\(\\theta_n\\) based on multiple chains of posterior draws.\nThe mean of the posterior draws of \\(\\theta_n\\) \\[\n\\hat{\\theta}_n\n= \\mathrm{mean}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n)\n\\]\nis treated as an estimator of the true posterior mean,\n\\[\n\\mathbb{E}[\\theta_n \\mid y]\n\\ = \\\n\\int_{-\\infty}^{\\infty}\n    \\, \\theta \\, p(\\theta | y)\n\\, \\mathrm{d}\\theta_n,\n\\]\nbased the observed data \\(y\\).\nThe standard error for the estimator \\(\\hat{\\theta}_n\\) is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as \\(\\hat{\\sigma}_n / \\sqrt{N_{\\mathrm{eff}}}\\). The smaller the standard error, the closer the estimate \\(\\hat{\\theta}_n\\) is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see Charles J. Geyer (2011) for more details of the MCMC central limit theorem.\n\n\n\nIn complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag \\(t\\), \\(\\rho_t\\), decreases as the lag, \\(t\\), increases. In this situation, thinning the sample by keeping only every \\(N\\)-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.\nFor instance, consider generating one thousand posterior draws in one of the following two ways.\n\nGenerate 1000 draws after convergence and save all of them.\nGenerate 10,000 draws after convergence and save every tenth draw.\n\nEven though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation \\(\\rho_t\\) for the thinned sequence is equivalent to \\(\\rho_{10t}\\) in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.\nNow contrast the second approach above with the unthinned alternative,\n\nGenerate 10,000 draws after convergence and save every draw.\n\nThis will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, the only reason to thin a sample is to reduce memory requirements.\nIf draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Posterior Analysis"
+    ]
+  },
+  {
+    "objectID": "reference-manual/analysis.html#footnotes",
+    "href": "reference-manual/analysis.html#footnotes",
+    "title": "Posterior Analysis",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nUsing vectors simplifies high level exposition at the expense of collapsing structure.↩︎\nThe structure is assumed to be rectangular; in the future, this needs to be generalized to ragged samples.↩︎\nThe lp__ value also represents the potential energy in the Hamiltonian system and is rate bounded by the randomly supplied kinetic energy each iteration, which follows a Chi-square distribution in the number of parameters.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Posterior Analysis"
+    ]
+  },
+  {
+    "objectID": "reference-manual/comments.html",
+    "href": "reference-manual/comments.html",
+    "title": "Comments",
+    "section": "",
+    "text": "Stan supports C++-style line-based and bracketed comments. Comments may be used anywhere whitespace is allowed in a Stan program.\n\n\nAny characters on a line following two forward slashes (//) is ignored along with the slashes. These may be used, for example, to document variables,\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  array[N] real y;  // observations\n}\n\n\n\nFor bracketed comments, any text between a forward-slash and asterisk pair (/*) and an asterisk and forward-slash pair (*/) is ignored.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Comments"
+    ]
+  },
+  {
+    "objectID": "reference-manual/comments.html#line-based-comments",
+    "href": "reference-manual/comments.html#line-based-comments",
+    "title": "Comments",
+    "section": "",
+    "text": "Any characters on a line following two forward slashes (//) is ignored along with the slashes. These may be used, for example, to document variables,\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  array[N] real y;  // observations\n}",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Comments"
+    ]
+  },
+  {
+    "objectID": "reference-manual/comments.html#bracketed-comments",
+    "href": "reference-manual/comments.html#bracketed-comments",
+    "title": "Comments",
+    "section": "",
+    "text": "For bracketed comments, any text between a forward-slash and asterisk pair (/*) and an asterisk and forward-slash pair (*/) is ignored.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Comments"
+    ]
+  },
+  {
+    "objectID": "reference-manual/diagnostics.html",
+    "href": "reference-manual/diagnostics.html",
+    "title": "Diagnostic Mode",
+    "section": "",
+    "text": "Stan’s diagnostic mode runs a Stan program with data, initializing parameters either randomly or with user-specified initial values, and then evaluates the log probability and its gradients. The gradients computed by the Stan program are compared to values calculated by finite differences.\nDiagnostic mode may be configured with two parameters.\nDiagnostic Mode Configuration Table. The diagnostic model configuration parameters, constraints, and default values.\n\n\n\n\n\n\n\n\n\nparameter\ndescription\nconstraints\ndefault\n\n\n\n\nepsilon\nfinite difference size\n(0, infty)\n1e–6\n\n\nerror\nerror threshold for matching\n(0, infty)\n1e–6\n\n\n\nIf the difference between the Stan program’s gradient value and that calculated by finite difference is higher than the specified threshold, the argument will be flagged.\n\n\nDiagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.\n\n\nThe output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an \\(N \\times N\\) correlation matrix, requires \\(\\binom{N}{2}\\) unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.\n\n\n\nThe log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.\n\n\n\n\nThe general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.\n\n\n\nDue to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Diagnostic Mode"
+    ]
+  },
+  {
+    "objectID": "reference-manual/diagnostics.html#diagnostic-mode-output",
+    "href": "reference-manual/diagnostics.html#diagnostic-mode-output",
+    "title": "Diagnostic Mode",
+    "section": "",
+    "text": "Diagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.\n\n\nThe output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an \\(N \\times N\\) correlation matrix, requires \\(\\binom{N}{2}\\) unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.\n\n\n\nThe log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Diagnostic Mode"
+    ]
+  },
+  {
+    "objectID": "reference-manual/diagnostics.html#configuration-options",
+    "href": "reference-manual/diagnostics.html#configuration-options",
+    "title": "Diagnostic Mode",
+    "section": "",
+    "text": "The general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Diagnostic Mode"
+    ]
+  },
+  {
+    "objectID": "reference-manual/diagnostics.html#speed-warning-and-data-trimming",
+    "href": "reference-manual/diagnostics.html#speed-warning-and-data-trimming",
+    "title": "Diagnostic Mode",
+    "section": "",
+    "text": "Due to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Diagnostic Mode"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html",
+    "href": "reference-manual/execution.html",
+    "title": "Program Execution",
+    "section": "",
+    "text": "This chapter provides a sketch of how a compiled Stan model is executed using sampling. Optimization shares the same data reading and initialization steps, but then does optimization rather than sampling.\nThis sketch is elaborated in the following chapters of this part, which cover variable declarations, expressions, statements, and blocks in more detail.\n\n\nThe reading and transforming data steps are the same for sampling, optimization and diagnostics.\n\n\nThe first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.1\nAll of the variables declared in the data block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.\nAfter each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable N is declared as int&lt;lower=0&gt;, after N is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.\n\n\n\nAfter data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.\nTransformed data variables are initialized with real values set to NaN and integer values set to the smallest integer (large absolute value negative number).\nAfter the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.\n\n\n\n\nInitialization is the same for all of Stan’s algorithms.\n\n\nIf there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.\nAfter being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.\n\n\nBecause of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nan initial value of 0 for theta leads to an unconstrained value of \\(-\\infty\\), whereas a value of 1 leads to an unconstrained value of \\(+\\infty\\). While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.\n\n\n\n\nIf there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval \\((-2,2)\\). The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.\nAn unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.\nFor parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to \\(\\exp(0) = 1\\) on the constrained scale. A value of -2 corresponds to \\(\\exp(-2) = .13\\) and a value of 2 corresponds to \\(\\exp(2) = 7.4\\).\nFor parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.\nSimplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is \\(1/K\\) in a \\(K\\)-simplex).\nCholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.\nThe initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the chapter on variable transforms.\n\n\n\nThe initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.\n\n\n\n\nSampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.\n\n\nFor each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.\nFor simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.\nIf the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively (Hoffman and Gelman 2014).\n\n\n\nDuring each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.\nThe first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.\nBecause the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.\nNext, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.\nThe final step in the log probability function calculation is to execute the statements defined in the model block.\nAs the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.\n\n\n\nA standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.\nIf step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).\nIf the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.\n\n\n\n\nOptimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.\n\n\n\nVariational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.\n\n\n\nModel diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.\n\n\n\nFor each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.\n\n\nBefore generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.\nAfter the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.\n\n\n\nThe final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.\nIn the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).2",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#reading-and-transforming-data",
+    "href": "reference-manual/execution.html#reading-and-transforming-data",
+    "title": "Program Execution",
+    "section": "",
+    "text": "The reading and transforming data steps are the same for sampling, optimization and diagnostics.\n\n\nThe first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.1\nAll of the variables declared in the data block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.\nAfter each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable N is declared as int&lt;lower=0&gt;, after N is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.\n\n\n\nAfter data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.\nTransformed data variables are initialized with real values set to NaN and integer values set to the smallest integer (large absolute value negative number).\nAfter the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#initialization",
+    "href": "reference-manual/execution.html#initialization",
+    "title": "Program Execution",
+    "section": "",
+    "text": "Initialization is the same for all of Stan’s algorithms.\n\n\nIf there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.\nAfter being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.\n\n\nBecause of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nan initial value of 0 for theta leads to an unconstrained value of \\(-\\infty\\), whereas a value of 1 leads to an unconstrained value of \\(+\\infty\\). While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.\n\n\n\n\nIf there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval \\((-2,2)\\). The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.\nAn unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.\nFor parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to \\(\\exp(0) = 1\\) on the constrained scale. A value of -2 corresponds to \\(\\exp(-2) = .13\\) and a value of 2 corresponds to \\(\\exp(2) = 7.4\\).\nFor parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.\nSimplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is \\(1/K\\) in a \\(K\\)-simplex).\nCholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.\nThe initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the chapter on variable transforms.\n\n\n\nThe initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#sampling",
+    "href": "reference-manual/execution.html#sampling",
+    "title": "Program Execution",
+    "section": "",
+    "text": "Sampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.\n\n\nFor each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.\nFor simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.\nIf the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively (Hoffman and Gelman 2014).\n\n\n\nDuring each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.\nThe first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.\nBecause the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.\nNext, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.\nThe final step in the log probability function calculation is to execute the statements defined in the model block.\nAs the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.\n\n\n\nA standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.\nIf step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).\nIf the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#optimization",
+    "href": "reference-manual/execution.html#optimization",
+    "title": "Program Execution",
+    "section": "",
+    "text": "Optimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#variational-inference",
+    "href": "reference-manual/execution.html#variational-inference",
+    "title": "Program Execution",
+    "section": "",
+    "text": "Variational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#model-diagnostics",
+    "href": "reference-manual/execution.html#model-diagnostics",
+    "title": "Program Execution",
+    "section": "",
+    "text": "Model diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#output",
+    "href": "reference-manual/execution.html#output",
+    "title": "Program Execution",
+    "section": "",
+    "text": "For each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.\n\n\nBefore generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.\nAfter the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.\n\n\n\nThe final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.\nIn the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).2",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/execution.html#footnotes",
+    "href": "reference-manual/execution.html#footnotes",
+    "title": "Program Execution",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe C++ code underlying Stan is flexible enough to allow data to be read from memory or file. Calls from R, for instance, can be configured to read data from file or directly from R’s memory.↩︎\nIn the R version of Stan, the values may either be written to a CSV file or directly back to R’s memory.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
+  },
+  {
+    "objectID": "reference-manual/includes.html",
+    "href": "reference-manual/includes.html",
+    "title": "Includes",
+    "section": "",
+    "text": "Stan allows one file to be included within another file using a syntax similar to that from C++. For example, suppose the file my-std-normal.stan defines the standard normal log probability density function (up to an additive constant).\nfunctions {\n  real my_std_normal_lpdf(vector y) {\n    return -0.5 * y' * y;\n  }\n}\nSuppose we also have a file containing a Stan program with an include statement.\n#include my-std-normal.stan\nparameters {\n  real y;\n}\nmodel {\n  y ~ my_std_normal();\n}\nThis Stan program behaves as if the contents of the file my-std-normal.stan replace the line with the #include statement, behaving as if a single Stan program were provided.\nfunctions {\n  real my_std_normal_lpdf(vector y) {\n    return -0.5 * y' * y;\n  }\n}\nparameters {\n  real y;\n}\nmodel {\n  y ~ my_std_normal();\n}\nThere are no restrictions on where include statements may be placed within a file or what the contents are of the replaced file.\n\n\nIt is possible to use includes on a line non-initially. For example, the previous example could’ve included space before the # in the include line:\n    #include my-std-normal.stan\nparameters {\n// ...\nIf there is initial space before an include, it will be discarded.\n\n\n\nIt is also possible to include line-based comments after the include. For example, the previous example can be coded as:\n#include my-std-normal.stan  // definition of standard normal\nparameters {\n// ...\nLine comments are discarded when the entire line is replaced with the contents of the included file.\n\n\n\nRecursive includes will lead to a compiler error. For example, suppose a.stan contains\n#include b.stan\nand b.stan contains\n#include a.stan\nThis will result in an error explaining the circular dependency:\nSyntax error in './b.stan', line 1, column 0, included from\n'./a.stan', line 1, column 0, included from\n'./b.stan', line 1, column 0, included from\n'a.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include a.stan\n         ^\n   -------------------------------------------------\n\nFile a.stan recursively included itself.\n\n\n\nThe Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.\n\n\nIf there is not a final / or \\ in the path, a / will be appended between the path and the included file name.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Includes"
+    ]
+  },
+  {
+    "objectID": "reference-manual/includes.html#recursive-includes",
+    "href": "reference-manual/includes.html#recursive-includes",
+    "title": "Includes",
+    "section": "",
+    "text": "Recursive includes will lead to a compiler error. For example, suppose a.stan contains\n#include b.stan\nand b.stan contains\n#include a.stan\nThis will result in an error explaining the circular dependency:\nSyntax error in './b.stan', line 1, column 0, included from\n'./a.stan', line 1, column 0, included from\n'./b.stan', line 1, column 0, included from\n'a.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include a.stan\n         ^\n   -------------------------------------------------\n\nFile a.stan recursively included itself.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Includes"
+    ]
+  },
+  {
+    "objectID": "reference-manual/includes.html#include-paths",
+    "href": "reference-manual/includes.html#include-paths",
+    "title": "Includes",
+    "section": "",
+    "text": "The Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.\n\n\nIf there is not a final / or \\ in the path, a / will be appended between the path and the included file name.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Includes"
+    ]
+  },
+  {
+    "objectID": "reference-manual/laplace.html",
+    "href": "reference-manual/laplace.html",
+    "title": "Laplace Approximation",
+    "section": "",
+    "text": "Laplace Approximation\nStan provides a Laplace approximation algorithm which can be used to obtain draws from an approximated posterior. The Laplace approximation works in the unconstrained space, so that if there are constrained parameters, the normal approximation is centered at the mode in the unconstrained space and then the implemented method transforms the normal approximation sample to the constrained space before outputting them.\nGiven the estimate of the mode \\(\\widehat{\\theta}\\), the Hessian \\(H(\\widehat{\\theta})\\) is computed using central finite differences of the model functor. Next the algorithm computes the Cholesky factor of the negative inverse Hessian:\n\\(R^{-1} = \\textrm{chol}(-H(\\widehat{\\theta})) \\backslash \\mathbf{1}\\).\nEach draw is generated on the unconstrained scale by sampling\n\\(\\theta^{\\textrm{std}(m)} \\sim \\textrm{normal}(0, \\textrm{I})\\)\nand defining draw \\(m\\) to be\n\\(\\theta^{(m)} = \\widehat{\\theta} + R^{-1} \\cdot \\theta^{\\textrm{std}(m)}\\)\nFinally, each \\(\\theta^{(m)}\\) is transformed back to the constrained scale.\nThe one-time computation of the Cholesky factor incurs a high constant overhead of \\(\\mathcal{O}(N^3)\\) in \\(N\\) dimensions. It also requires \\(2N\\) gradient calculations to use as the basis, which scales at best as \\(\\mathcal{O}(N^2)\\) and is worse for models whose gradient calculation is super-linear in dimension. The algorithm also has a high per-draw overhead, requiring \\(N\\) standard normal pseudorandom numbers and \\(\\mathcal{O}(N^2)\\) per draw (to multiply by the Cholesky factor). For \\(M\\) draws, the total cost is proportional to \\(\\mathcal{O}(N^3 + M \\cdot N^2)\\).\n\n\n\n\n Back to top",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Laplace Approximation"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html",
+    "href": "reference-manual/licenses.html",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan and its dependent libraries, are distributed under generous, freedom-respecting licenses approved by the Open Source Initiative.\nIn particular, the licenses for Stan and its dependent libraries have no “copyleft” provisions requiring applications of Stan to be open source if they are redistributed.\nThis chapter specifies the licenses for the libraries on which Stan’s math library, language, and algorithms depend. The last tool mentioned, Google Test, is only used for testing and is not needed to run Stan.\n\n\nStan is distributed under\n\nBSD 3-clause license (BSD New)\n\nThe copyright holder of each contribution is the developer or his or her assignee.1\n\n\n\nStan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the\n\nBoost Software License version 1.0\n\nThe copyright for each Boost package is held by its developers or their assignees.\n\n\n\nStan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the\n\nMozilla Public License, version 2\n\nThe copyright of Eigen is owned jointly by its developers or their assignees.\n\n\n\nStan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.\n\n\n\nStan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the\n\nApache License, version 2\n\nThe copyright of TBB is owned by Intel Corporation.\n\n\n\nStan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of Google Test is owned by Google, Inc.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#stan-license",
+    "href": "reference-manual/licenses.html#stan-license",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan is distributed under\n\nBSD 3-clause license (BSD New)\n\nThe copyright holder of each contribution is the developer or his or her assignee.1",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#boost-license",
+    "href": "reference-manual/licenses.html#boost-license",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the\n\nBoost Software License version 1.0\n\nThe copyright for each Boost package is held by its developers or their assignees.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#eigen-license",
+    "href": "reference-manual/licenses.html#eigen-license",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the\n\nMozilla Public License, version 2\n\nThe copyright of Eigen is owned jointly by its developers or their assignees.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#sundials-license",
+    "href": "reference-manual/licenses.html#sundials-license",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#threaded-building-blocks-tbb-license",
+    "href": "reference-manual/licenses.html#threaded-building-blocks-tbb-license",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the\n\nApache License, version 2\n\nThe copyright of TBB is owned by Intel Corporation.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#google-test-license",
+    "href": "reference-manual/licenses.html#google-test-license",
+    "title": "Licenses and Dependencies",
+    "section": "",
+    "text": "Stan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of Google Test is owned by Google, Inc.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#footnotes",
+    "href": "reference-manual/licenses.html#footnotes",
+    "title": "Licenses and Dependencies",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nUniversities or companies often own the copyright of computer programs developed by their employees.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/optimization.html",
+    "href": "reference-manual/optimization.html",
+    "title": "Optimization",
+    "section": "",
+    "text": "Stan provides optimization algorithms which find modes of the density specified by a Stan program. Such modes may be used as parameter estimates or as the basis of approximations to a Bayesian posterior.\nStan provides three different optimizers, a Newton optimizer, and two related quasi-Newton algorithms, BFGS and L-BFGS; see Nocedal and Wright (2006) for thorough description and analysis of all of these algorithms. The L-BFGS algorithm is the default optimizer. Newton’s method is the least efficient of the three, but has the advantage of setting its own stepsize.\n\n\nAll of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\nAll of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.\nAll of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.\n\n\n\n\n\nConvergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.\n\n\nThe parameters \\(\\theta_i\\) in iteration \\(i\\) are considered to have converged with respect to tolerance tol_param if\n\\[\n|| \\theta_{i} - \\theta_{i-1} || &lt; \\mathtt{tol\\_param}.\n\\]\n\n\n\nThe (unnormalized) log density \\(\\log p(\\theta_{i}|y)\\) for the parameters \\(\\theta_i\\) in iteration \\(i\\) given data \\(y\\) is considered to have converged with respect to tolerance tol_obj if\n\\[\n\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right| &lt;\n\\mathtt{tol\\_obj}.\n\\]\nThe log density is considered to have converged to within relative tolerance tol_rel_obj if\n\\[\n\\frac{\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right|}{\\\n  \\max\\left(\\left| \\log p(\\theta_{i}|y)\\right|,\\left| \\log\n      p(\\theta_{i-1}|y)\\right|,1.0\\right)}\n&lt; \\mathtt{tol\\_rel\\_obj} * \\epsilon.\n\\]\n\n\n\nThe gradient is considered to have converged to 0 relative to a specified tolerance tol_grad if\n\\[\n|| g_{i} || &lt; \\mathtt{tol\\_grad},\n\\] where \\(\\nabla_{\\theta}\\) is the gradient operator with respect to \\(\\theta\\) and \\(g_{i} = \\nabla_{\\theta} \\log p(\\theta | y)\\) is the gradient at iteration \\(i\\) evaluated at \\(\\theta^{(i)}\\), the value on the \\(i\\)-th posterior iteration.\nThe gradient is considered to have converged to 0 relative to a specified relative tolerance tol_rel_grad if\n\\[\n\\frac{g_{i}^T \\hat{H}_{i}^{-1} g_{i} }{ \\max\\left(\\left|\\log\np(\\theta_{i}|y)\\right|,1.0\\right) }\n\\ &lt; \\\n\\mathtt{tol\\_rel\\_grad} * \\epsilon,\n\\]\nwhere \\(\\hat{H}_{i}\\) is the estimate of the Hessian at iteration \\(i\\), \\(|u|\\) is the absolute value (L1 norm) of \\(u\\), \\(||u||\\) is the vector length (L2 norm) of \\(u\\), and \\(\\epsilon \\approx 2e-16\\) is machine precision.\n\n\n\n\nThe initial step size parameter \\(\\alpha\\) for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize \\(\\alpha\\) to be the roughly equal to the \\(\\alpha\\) used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune \\(\\alpha\\) can save some real time.\n\n\n\nL-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.\n\n\n\n\n\n\nFor constrained optimization problems, for instance, with a standard deviation parameter \\(\\sigma\\) constrained so that \\(\\sigma &gt; 0\\), it can be much more efficient to declare a parameter sigma with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward \\(-\\infty\\) on the \\(\\log \\sigma\\) scale.\nWith unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector\nvector[M] sigma;\nand using the default random initialization which is \\(\\mathsf{Uniform}(-2, 2)\\) on the unconstrained scale means that there is only a \\(2^{-M}\\) chance that the initialization will be within support.\nFor any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "reference-manual/optimization.html#general-configuration",
+    "href": "reference-manual/optimization.html#general-configuration",
+    "title": "Optimization",
+    "section": "",
+    "text": "All of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\nAll of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.\nAll of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "reference-manual/optimization.html#bfgs-and-l-bfgs-configuration",
+    "href": "reference-manual/optimization.html#bfgs-and-l-bfgs-configuration",
+    "title": "Optimization",
+    "section": "",
+    "text": "Convergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.\n\n\nThe parameters \\(\\theta_i\\) in iteration \\(i\\) are considered to have converged with respect to tolerance tol_param if\n\\[\n|| \\theta_{i} - \\theta_{i-1} || &lt; \\mathtt{tol\\_param}.\n\\]\n\n\n\nThe (unnormalized) log density \\(\\log p(\\theta_{i}|y)\\) for the parameters \\(\\theta_i\\) in iteration \\(i\\) given data \\(y\\) is considered to have converged with respect to tolerance tol_obj if\n\\[\n\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right| &lt;\n\\mathtt{tol\\_obj}.\n\\]\nThe log density is considered to have converged to within relative tolerance tol_rel_obj if\n\\[\n\\frac{\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right|}{\\\n  \\max\\left(\\left| \\log p(\\theta_{i}|y)\\right|,\\left| \\log\n      p(\\theta_{i-1}|y)\\right|,1.0\\right)}\n&lt; \\mathtt{tol\\_rel\\_obj} * \\epsilon.\n\\]\n\n\n\nThe gradient is considered to have converged to 0 relative to a specified tolerance tol_grad if\n\\[\n|| g_{i} || &lt; \\mathtt{tol\\_grad},\n\\] where \\(\\nabla_{\\theta}\\) is the gradient operator with respect to \\(\\theta\\) and \\(g_{i} = \\nabla_{\\theta} \\log p(\\theta | y)\\) is the gradient at iteration \\(i\\) evaluated at \\(\\theta^{(i)}\\), the value on the \\(i\\)-th posterior iteration.\nThe gradient is considered to have converged to 0 relative to a specified relative tolerance tol_rel_grad if\n\\[\n\\frac{g_{i}^T \\hat{H}_{i}^{-1} g_{i} }{ \\max\\left(\\left|\\log\np(\\theta_{i}|y)\\right|,1.0\\right) }\n\\ &lt; \\\n\\mathtt{tol\\_rel\\_grad} * \\epsilon,\n\\]\nwhere \\(\\hat{H}_{i}\\) is the estimate of the Hessian at iteration \\(i\\), \\(|u|\\) is the absolute value (L1 norm) of \\(u\\), \\(||u||\\) is the vector length (L2 norm) of \\(u\\), and \\(\\epsilon \\approx 2e-16\\) is machine precision.\n\n\n\n\nThe initial step size parameter \\(\\alpha\\) for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize \\(\\alpha\\) to be the roughly equal to the \\(\\alpha\\) used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune \\(\\alpha\\) can save some real time.\n\n\n\nL-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "reference-manual/optimization.html#writing-models-for-optimization",
+    "href": "reference-manual/optimization.html#writing-models-for-optimization",
+    "title": "Optimization",
+    "section": "",
+    "text": "For constrained optimization problems, for instance, with a standard deviation parameter \\(\\sigma\\) constrained so that \\(\\sigma &gt; 0\\), it can be much more efficient to declare a parameter sigma with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward \\(-\\infty\\) on the \\(\\log \\sigma\\) scale.\nWith unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector\nvector[M] sigma;\nand using the default random initialization which is \\(\\mathsf{Uniform}(-2, 2)\\) on the unconstrained scale means that there is only a \\(2^{-M}\\) chance that the initialization will be within support.\nFor any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Optimization"
+    ]
+  },
+  {
+    "objectID": "reference-manual/references.html",
+    "href": "reference-manual/references.html",
+    "title": "References",
+    "section": "",
+    "text": "References\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "reference-manual/reproducibility.html",
+    "href": "reference-manual/reproducibility.html",
+    "title": "Reproducibility",
+    "section": "",
+    "text": "Floating point operations on modern computers are notoriously difficult to replicate because the fundamental arithmetic operations, right down to the IEEE 754 encoding level, are not fully specified. The primary problem is that the precision of operations varies across different hardware platforms and software implementations.\nStan is designed to allow full reproducibility. However, this is only possible up to the external constraints imposed by floating point arithmetic.\nStan results will only be exactly reproducible if all of the following components are identical:\n\nStan version\nStan interface (RStan, PyStan, CmdStan) and version, plus version of interface language (R, Python, shell)\nversions of included libraries (Boost and Eigen)\noperating system version\ncomputer hardware including CPU, motherboard and memory\nC++ compiler, including version, compiler flags, and linked libraries\nsame configuration of call to Stan, including random seed, chain ID, initialization and data\n\nIt doesn’t matter if you use a stable release version of Stan or the version with a particular Git hash tag. The same goes for all of the interfaces, compilers, and so on. The point is that if any of these moving parts changes in some way, floating point results may change.\nConcretely, if you compile a single Stan program using the same CmdStan code base, but changed the optimization flag (-O3 vs. -O2 or -O0), the two programs may not return the identical stream of results. Thus it is very hard to guarantee reproducibility on externally managed hardware, like in a cluster or even a desktop managed by an IT department or with automatic updates turned on.\nIf, however, you compiled a Stan program today using one set of flags, took the computer away from the internet and didn’t allow it to update anything, then came back in a decade and recompiled the Stan program in the same way, you should get the same results.\nThe data needs to be the same down to the bit level. For example, if you are running in RStan, Rcpp handles the conversion between R’s floating point numbers and C++ doubles. If Rcpp changes the conversion process or use different types, the results are not guaranteed to be the same down to the bit level.\nThe compiler and compiler settings can also be an issue. There is a nice discussion of the issues and how to control reproducibility in Intel’s proprietary compiler by Corden and Kreitzer (2014).\n\n\nAs noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.\nHowever, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still does not guarantee exact reproducibility between it and other versions.\n\nStan 2.28 changed the default chain ID for MCMC from 0 to 1. Users who had set a seed but not a chain ID would observe completely different outputs.\nStan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Reproducibility"
+    ]
+  },
+  {
+    "objectID": "reference-manual/reproducibility.html#notable-changes-across-versions",
+    "href": "reference-manual/reproducibility.html#notable-changes-across-versions",
+    "title": "Reproducibility",
+    "section": "",
+    "text": "As noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.\nHowever, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still does not guarantee exact reproducibility between it and other versions.\n\nStan 2.28 changed the default chain ID for MCMC from 0 to 1. Users who had set a seed but not a chain ID would observe completely different outputs.\nStan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Reproducibility"
+    ]
+  },
+  {
+    "objectID": "reference-manual/syntax.html",
+    "href": "reference-manual/syntax.html",
+    "title": "Language Syntax",
+    "section": "",
+    "text": "This chapter defines the basic syntax of the Stan modeling language using a Backus-Naur form (BNF) grammar plus extra-grammatical constraints on function typing and operator precedence and associativity.\n\n\n\n\nIn the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by &lt; and &gt;. A square brackets ([A]) indicates optionality of A. A postfixed Kleene star (A*) indicates zero or more occurrences of A. Parenthesis can be used to group symbols together in productions.\nFinally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library Menhir. A rule like &lt;list(x)&gt; ::= x (COMMA x)* declares a generic list rule, which can later be applied to others by the symbol &lt;list(&lt;expression&gt;)&gt;.\nThe following representation is constructed directly from the OCaml reference parser using a tool called Obelisk. The raw output is available here.\n\n\n\n\n&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]\n              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]\n              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF\n\n&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF\n\n&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE\n\n&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE\n\n&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE\n                             &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*\n                       RBRACE\n\n&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE\n                                   &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE\n                                 &lt;top_vardecl_or_statement&gt;* RBRACE\n\n\n\n&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA\n                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;\n\n&lt;return_type&gt; ::= VOID\n                | &lt;unsized_type&gt;\n\n&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;\n\n&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;\n                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;\n                 | &lt;basic_type&gt;\n                 | &lt;unsized_tuple_type&gt;\n\n&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA [&lt;unsized_type&gt;\n                         (COMMA &lt;unsized_type&gt;)*] RPAREN\n\n&lt;basic_type&gt; ::= INT\n               | REAL\n               | COMPLEX\n               | VECTOR\n               | ROWVECTOR\n               | MATRIX\n               | COMPLEXVECTOR\n               | COMPLEXROWVECTOR\n               | COMPLEXMATRIX\n\n&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK\n\n\n\n\n&lt;identifier&gt; ::= IDENTIFIER\n               | TRUNCATE\n\n&lt;decl_identifier&gt; ::= &lt;identifier&gt;\n\n&lt;no_assign&gt; ::= UNREACHABLE\n\n&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]\n\n&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;\n                                      &lt;optional_assignment(rhs)&gt;\n\n&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;\n                           &lt;optional_assignment(rhs)&gt; SEMICOLON\n                         | &lt;higher_type(type_rule)&gt;\n                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA\n                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON\n\n&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;\n                           | &lt;tuple_type(type_rule)&gt;\n                           | type_rule\n\n&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule\n                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;\n\n&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA\n                            [&lt;higher_type(type_rule)&gt; (COMMA\n                            &lt;higher_type(type_rule)&gt;)*] RPAREN\n\n&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;\n                           | SEMICOLON\n\n&lt;sized_basic_type&gt; ::= INT\n                     | REAL\n                     | COMPLEX\n                     | VECTOR LBRACK &lt;expression&gt; RBRACK\n                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                       RBRACK\n\n&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]\n                 | REAL &lt;type_constraint&gt;\n                 | COMPLEX &lt;type_constraint&gt;\n                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;\n                   RBRACK\n                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | ORDERED LBRACK &lt;expression&gt; RBRACK\n                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK\n                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK\n                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROVEC LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]\n                   RBRACK\n                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                   RBRACK\n\n&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]\n                    | LABRACK &lt;offset_mult&gt; RABRACK\n\n&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN\n            &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN\n            &lt;constr_expression&gt;\n          | LOWER ASSIGN &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt;\n\n&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN\n                  &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN\n                  &lt;constr_expression&gt;\n                | OFFSET ASSIGN &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt;\n\n&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK\n\n\n\n\n&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;\n               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;\n               | &lt;prefixOp&gt; &lt;expression&gt;\n               | &lt;expression&gt; &lt;postfixOp&gt;\n               | &lt;common_expression&gt;\n\n&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;\n                        &lt;constr_expression&gt;\n                      | &lt;prefixOp&gt; &lt;constr_expression&gt;\n                      | &lt;constr_expression&gt; &lt;postfixOp&gt;\n                      | &lt;common_expression&gt;\n\n&lt;common_expression&gt; ::= &lt;identifier&gt;\n                      | INTNUMERAL\n                      | REALNUMERAL\n                      | DOTNUMERAL\n                      | IMAGNUMERAL\n                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE\n                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK\n                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | TARGET LPAREN RPAREN\n                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;\n                        (COMMA &lt;expression&gt;)*] RPAREN\n                      | LPAREN &lt;expression&gt; COMMA [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | &lt;common_expression&gt; DOTNUMERAL\n                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK\n                      | LPAREN &lt;expression&gt; RPAREN\n\n&lt;prefixOp&gt; ::= BANG\n             | MINUS\n             | PLUS\n\n&lt;postfixOp&gt; ::= TRANSPOSE\n\n&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;\n            | &lt;logicalBinOp&gt;\n\n&lt;arithmeticBinOp&gt; ::= PLUS\n                    | MINUS\n                    | TIMES\n                    | DIVIDE\n                    | IDIVIDE\n                    | MODULO\n                    | LDIVIDE\n                    | ELTTIMES\n                    | ELTDIVIDE\n                    | HAT\n                    | ELTPOW\n\n&lt;logicalBinOp&gt; ::= OR\n                 | AND\n                 | EQUALS\n                 | NEQUALS\n                 | LABRACK\n                 | LEQ\n                 | RABRACK\n                 | GEQ\n\n&lt;indexes&gt; ::= epsilon\n            | COLON\n            | &lt;expression&gt;\n            | &lt;expression&gt; COLON\n            | COLON &lt;expression&gt;\n            | &lt;expression&gt; COLON &lt;expression&gt;\n            | &lt;indexes&gt; COMMA &lt;indexes&gt;\n\n&lt;printables&gt; ::= &lt;expression&gt;\n               | &lt;string_literal&gt;\n               | &lt;printables&gt; COMMA &lt;printables&gt;\n\n\n\n&lt;statement&gt; ::= &lt;atomic_statement&gt;\n              | &lt;nested_statement&gt;\n\n&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;\n                       SEMICOLON\n                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                       &lt;expression&gt;)*] RPAREN SEMICOLON\n                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;\n                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON\n                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | BREAK SEMICOLON\n                     | CONTINUE SEMICOLON\n                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | RETURN &lt;expression&gt; SEMICOLON\n                     | RETURN SEMICOLON\n                     | SEMICOLON\n\n&lt;assignment_op&gt; ::= ASSIGN\n                  | PLUSASSIGN\n                  | MINUSASSIGN\n                  | TIMESASSIGN\n                  | DIVIDEASSIGN\n                  | ELTTIMESASSIGN\n                  | ELTDIVIDEASSIGN\n\n&lt;string_literal&gt; ::= STRINGLITERAL\n\n&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK\n\n&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                       ELSE &lt;vardecl_or_statement&gt;\n                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | WHILE LPAREN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON\n                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE\n                       &lt;vardecl_or_statement&gt;* RBRACE\n                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;\n                         | &lt;var_decl&gt;\n\n&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;\n                             | &lt;top_var_decl&gt;\n\n\n\n\nMany of the tokens used in the BNF grammars follow obviously from their names: DATABLOCK is the literal string ‘data’, COMMA is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the operator precedence table.\nA few tokens are not so obvious, and are defined here in regular expressions:\nIDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*\n\nSTRINGLITERAL = \".*\"\n\nINTNUMERAL = [0-9]+ (_ [0-9]+)*\n\nEXPLITERAL = [eE] [+-]? INTNUMERAL\n\nREALNUMERAL = INTNUMERAL \\. INTNUMERAL? EXPLITERAL?\n            | \\. INTNUMERAL EXPLITERAL\n            | INTNUMERAL EXPLITERAL\n\nIMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i\n\nDOTNUMERAL = \\. INTNUMERAL\n\n\n\n\n\nA well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.\n\n\n\nIn the Stan grammar provided in this chapter, the expression 1 + 2 * 3 has two parses. As described in the operator precedence table, Stan disambiguates between the meaning \\(1\n+ (2 \\times 3)\\) and the meaning \\((1 + 2) \\times 3\\) based on operator precedences and associativities.\n\n\n\nIn a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.\n\n\n\nThe types of expressions used for elements in array expressions ('{' expressions '}') must all be of the same type or a mixture of scalar (int, real and complex) types (in which case the result is promoted to be of the highest type on the int -&gt; real -&gt; complex hierarchy).\n\n\n\nInteger literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.\n\n\n\nBoth the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.\n\n\n\nThe for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).\n\n\n\nThe arguments to a print statement cannot be void.\n\n\n\nThe break and continue statements may only be used within the body of a for-loop or while-loop.\n\n\n\nSome constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.\n\n\nFunctions ending in _rng may only be called in the transformed data and generated quantities block, and within the bodies of user-defined functions with names ending in _rng.\n\n\n\nUnnormalized distributions (with suffixes _lupmf or _lupdf) may only be called in the model block, user-defined probability functions, or within the bodies of user defined functions which end in _lp.\n\n\n\ntarget += statements can only be used inside of the model block or user-defined functions which end in _lp.\nUser defined functions which end in _lp and the target() function can only be used in the model block, transformed parameters block, and in the bodies of other user defined functions which end in _lp.\nSampling statements (using ~) can only be used in the model block or in the bodies of user-defined functions which end in _lp.\njacobian += statements can only be used inside of the transformed parameters block or in functions that end with _jacobian.\n\n\n\n\nA probability function literal must have one of the following suffixes: _lpdf, _lpmf, _lcdf, or _lccdf.\n\n\n\nStandalone expressions used as indexes must denote either an integer (int) or an integer array (array[] int). Expressions participating in range indexes (e.g., a and b in a : b) must denote integers (int).\nA second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type array[ , , ] matrix, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
+    ]
+  },
+  {
+    "objectID": "reference-manual/syntax.html#bnf-grammars",
+    "href": "reference-manual/syntax.html#bnf-grammars",
+    "title": "Language Syntax",
+    "section": "",
+    "text": "In the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by &lt; and &gt;. A square brackets ([A]) indicates optionality of A. A postfixed Kleene star (A*) indicates zero or more occurrences of A. Parenthesis can be used to group symbols together in productions.\nFinally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library Menhir. A rule like &lt;list(x)&gt; ::= x (COMMA x)* declares a generic list rule, which can later be applied to others by the symbol &lt;list(&lt;expression&gt;)&gt;.\nThe following representation is constructed directly from the OCaml reference parser using a tool called Obelisk. The raw output is available here.\n\n\n\n\n&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]\n              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]\n              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF\n\n&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF\n\n&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE\n\n&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE\n\n&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE\n                             &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*\n                       RBRACE\n\n&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE\n                                   &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE\n                                 &lt;top_vardecl_or_statement&gt;* RBRACE\n\n\n\n&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA\n                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;\n\n&lt;return_type&gt; ::= VOID\n                | &lt;unsized_type&gt;\n\n&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;\n\n&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;\n                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;\n                 | &lt;basic_type&gt;\n                 | &lt;unsized_tuple_type&gt;\n\n&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA [&lt;unsized_type&gt;\n                         (COMMA &lt;unsized_type&gt;)*] RPAREN\n\n&lt;basic_type&gt; ::= INT\n               | REAL\n               | COMPLEX\n               | VECTOR\n               | ROWVECTOR\n               | MATRIX\n               | COMPLEXVECTOR\n               | COMPLEXROWVECTOR\n               | COMPLEXMATRIX\n\n&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK\n\n\n\n\n&lt;identifier&gt; ::= IDENTIFIER\n               | TRUNCATE\n\n&lt;decl_identifier&gt; ::= &lt;identifier&gt;\n\n&lt;no_assign&gt; ::= UNREACHABLE\n\n&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]\n\n&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;\n                                      &lt;optional_assignment(rhs)&gt;\n\n&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;\n                           &lt;optional_assignment(rhs)&gt; SEMICOLON\n                         | &lt;higher_type(type_rule)&gt;\n                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA\n                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON\n\n&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;\n                           | &lt;tuple_type(type_rule)&gt;\n                           | type_rule\n\n&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule\n                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;\n\n&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA\n                            [&lt;higher_type(type_rule)&gt; (COMMA\n                            &lt;higher_type(type_rule)&gt;)*] RPAREN\n\n&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;\n                           | SEMICOLON\n\n&lt;sized_basic_type&gt; ::= INT\n                     | REAL\n                     | COMPLEX\n                     | VECTOR LBRACK &lt;expression&gt; RBRACK\n                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                       RBRACK\n\n&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]\n                 | REAL &lt;type_constraint&gt;\n                 | COMPLEX &lt;type_constraint&gt;\n                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;\n                   RBRACK\n                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | ORDERED LBRACK &lt;expression&gt; RBRACK\n                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK\n                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK\n                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROVEC LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]\n                   RBRACK\n                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                   RBRACK\n\n&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]\n                    | LABRACK &lt;offset_mult&gt; RABRACK\n\n&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN\n            &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN\n            &lt;constr_expression&gt;\n          | LOWER ASSIGN &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt;\n\n&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN\n                  &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN\n                  &lt;constr_expression&gt;\n                | OFFSET ASSIGN &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt;\n\n&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK\n\n\n\n\n&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;\n               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;\n               | &lt;prefixOp&gt; &lt;expression&gt;\n               | &lt;expression&gt; &lt;postfixOp&gt;\n               | &lt;common_expression&gt;\n\n&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;\n                        &lt;constr_expression&gt;\n                      | &lt;prefixOp&gt; &lt;constr_expression&gt;\n                      | &lt;constr_expression&gt; &lt;postfixOp&gt;\n                      | &lt;common_expression&gt;\n\n&lt;common_expression&gt; ::= &lt;identifier&gt;\n                      | INTNUMERAL\n                      | REALNUMERAL\n                      | DOTNUMERAL\n                      | IMAGNUMERAL\n                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE\n                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK\n                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | TARGET LPAREN RPAREN\n                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;\n                        (COMMA &lt;expression&gt;)*] RPAREN\n                      | LPAREN &lt;expression&gt; COMMA [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | &lt;common_expression&gt; DOTNUMERAL\n                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK\n                      | LPAREN &lt;expression&gt; RPAREN\n\n&lt;prefixOp&gt; ::= BANG\n             | MINUS\n             | PLUS\n\n&lt;postfixOp&gt; ::= TRANSPOSE\n\n&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;\n            | &lt;logicalBinOp&gt;\n\n&lt;arithmeticBinOp&gt; ::= PLUS\n                    | MINUS\n                    | TIMES\n                    | DIVIDE\n                    | IDIVIDE\n                    | MODULO\n                    | LDIVIDE\n                    | ELTTIMES\n                    | ELTDIVIDE\n                    | HAT\n                    | ELTPOW\n\n&lt;logicalBinOp&gt; ::= OR\n                 | AND\n                 | EQUALS\n                 | NEQUALS\n                 | LABRACK\n                 | LEQ\n                 | RABRACK\n                 | GEQ\n\n&lt;indexes&gt; ::= epsilon\n            | COLON\n            | &lt;expression&gt;\n            | &lt;expression&gt; COLON\n            | COLON &lt;expression&gt;\n            | &lt;expression&gt; COLON &lt;expression&gt;\n            | &lt;indexes&gt; COMMA &lt;indexes&gt;\n\n&lt;printables&gt; ::= &lt;expression&gt;\n               | &lt;string_literal&gt;\n               | &lt;printables&gt; COMMA &lt;printables&gt;\n\n\n\n&lt;statement&gt; ::= &lt;atomic_statement&gt;\n              | &lt;nested_statement&gt;\n\n&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;\n                       SEMICOLON\n                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                       &lt;expression&gt;)*] RPAREN SEMICOLON\n                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;\n                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON\n                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | BREAK SEMICOLON\n                     | CONTINUE SEMICOLON\n                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | RETURN &lt;expression&gt; SEMICOLON\n                     | RETURN SEMICOLON\n                     | SEMICOLON\n\n&lt;assignment_op&gt; ::= ASSIGN\n                  | PLUSASSIGN\n                  | MINUSASSIGN\n                  | TIMESASSIGN\n                  | DIVIDEASSIGN\n                  | ELTTIMESASSIGN\n                  | ELTDIVIDEASSIGN\n\n&lt;string_literal&gt; ::= STRINGLITERAL\n\n&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK\n\n&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                       ELSE &lt;vardecl_or_statement&gt;\n                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | WHILE LPAREN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON\n                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE\n                       &lt;vardecl_or_statement&gt;* RBRACE\n                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;\n                         | &lt;var_decl&gt;\n\n&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;\n                             | &lt;top_var_decl&gt;",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
+    ]
+  },
+  {
+    "objectID": "reference-manual/syntax.html#tokenizing-rules",
+    "href": "reference-manual/syntax.html#tokenizing-rules",
+    "title": "Language Syntax",
+    "section": "",
+    "text": "Many of the tokens used in the BNF grammars follow obviously from their names: DATABLOCK is the literal string ‘data’, COMMA is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the operator precedence table.\nA few tokens are not so obvious, and are defined here in regular expressions:\nIDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*\n\nSTRINGLITERAL = \".*\"\n\nINTNUMERAL = [0-9]+ (_ [0-9]+)*\n\nEXPLITERAL = [eE] [+-]? INTNUMERAL\n\nREALNUMERAL = INTNUMERAL \\. INTNUMERAL? EXPLITERAL?\n            | \\. INTNUMERAL EXPLITERAL\n            | INTNUMERAL EXPLITERAL\n\nIMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i\n\nDOTNUMERAL = \\. INTNUMERAL",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
+    ]
+  },
+  {
+    "objectID": "reference-manual/syntax.html#extra-grammatical-constraints",
+    "href": "reference-manual/syntax.html#extra-grammatical-constraints",
+    "title": "Language Syntax",
+    "section": "",
+    "text": "A well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.\n\n\n\nIn the Stan grammar provided in this chapter, the expression 1 + 2 * 3 has two parses. As described in the operator precedence table, Stan disambiguates between the meaning \\(1\n+ (2 \\times 3)\\) and the meaning \\((1 + 2) \\times 3\\) based on operator precedences and associativities.\n\n\n\nIn a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.\n\n\n\nThe types of expressions used for elements in array expressions ('{' expressions '}') must all be of the same type or a mixture of scalar (int, real and complex) types (in which case the result is promoted to be of the highest type on the int -&gt; real -&gt; complex hierarchy).\n\n\n\nInteger literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.\n\n\n\nBoth the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.\n\n\n\nThe for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).\n\n\n\nThe arguments to a print statement cannot be void.\n\n\n\nThe break and continue statements may only be used within the body of a for-loop or while-loop.\n\n\n\nSome constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.\n\n\nFunctions ending in _rng may only be called in the transformed data and generated quantities block, and within the bodies of user-defined functions with names ending in _rng.\n\n\n\nUnnormalized distributions (with suffixes _lupmf or _lupdf) may only be called in the model block, user-defined probability functions, or within the bodies of user defined functions which end in _lp.\n\n\n\ntarget += statements can only be used inside of the model block or user-defined functions which end in _lp.\nUser defined functions which end in _lp and the target() function can only be used in the model block, transformed parameters block, and in the bodies of other user defined functions which end in _lp.\nSampling statements (using ~) can only be used in the model block or in the bodies of user-defined functions which end in _lp.\njacobian += statements can only be used inside of the transformed parameters block or in functions that end with _jacobian.\n\n\n\n\nA probability function literal must have one of the following suffixes: _lpdf, _lpmf, _lcdf, or _lccdf.\n\n\n\nStandalone expressions used as indexes must denote either an integer (int) or an integer array (array[] int). Expressions participating in range indexes (e.g., a and b in a : b) must denote integers (int).\nA second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type array[ , , ] matrix, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html",
+    "href": "reference-manual/types.html",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "This chapter covers the data types for expressions in Stan. Every variable used in a Stan program must have a declared data type. Only values of that type will be assignable to the variable (except for temporary states of transformed data and transformed parameter values). This follows the convention of programming languages like C++, not the conventions of scripting languages like Python or statistical languages such as R or BUGS.\nThe motivation for strong, static typing is threefold.\n\nStrong typing forces the programmer’s intent to be declared with the variable, making programs easier to comprehend and hence easier to debug and maintain.\nStrong typing allows programming errors relative to the declared intent to be caught sooner (at compile time) rather than later (at run time). The Stan compiler (called through an interface such as CmdStan, RStan, or PyStan) will flag any type errors and indicate the offending expressions quickly when the program is compiled.\nConstrained types will catch runtime data, initialization, and intermediate value errors as soon as they occur rather than allowing them to propagate and potentially pollute final results.\n\nStrong typing disallows assigning the same variable to objects of different types at different points in the program or in different invocations of the program.\n\n\nArguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.\nPassing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.\n\n\nStan provides two primitive data types, real for continuous values and int for integer values. These are both considered scalar types.\n\n\n\nStan provides a complex number data type complex, where a complex number contains both a real and an imaginary component, both of which are of type real. Complex types are considered scalar types.\n\n\n\nStan provides three real-valued matrix data types, vector for column vectors, row_vector for row vectors, and matrix for matrices.\nStan also provides three complex-valued matrix data types, complex_vector for column vectors, complex_row_vector for row vectors, and complex_matrix for matrices.\n\n\n\nAny type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,\narray[10] real x;\narray[6, 7] matrix[3, 3] m;\narray[12, 8, 15] complex z;\ndeclares x to be a one-dimensional array of size 10 containing real values, declares m to be a two-dimensional array of size \\(6 \\times 7\\) containing values that are \\(3 \\times 3\\) matrices, and declares z to be a \\(12 \\times 8 \\times 15\\) array of complex numbers.\nPrior to 2.26 Stan models used a different syntax which has since been removed. See the Removed Features chapter for more details.\n\n\n\nFor any sequence of types, Stan provides a tuple data type. For example,\ntuple(real, array[5] int) xi;\ndeclares xi to be a tuple holding two values, the first of which is of type type real and the second of which a 5-dimensional array of type int.\n\n\n\nDeclarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.\nAll of the basic data types other than complex may be given lower and upper bounds using syntax such as\nint&lt;lower=1&gt; N;\nreal&lt;upper=0&gt; log_p;\nvector&lt;lower=-1, upper=1&gt;[3] rho;\nThere are also special data types for structured vectors and matrices. There are five constrained vector data types, simplex for unit simplexes, unit_vector for unit-length vectors, sum_to_zero_vector for vectors that sum to zero, ordered for ordered vectors of scalars, and positive_ordered for vectors of positive ordered scalars. There are specialized matrix data types corr_matrix and cov_matrix for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type cholesky_factor_cov is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type cholesky_factor_corr is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type sum_to_zero_matrix is for matrices that sum to zero across rows and columns.\nConstraints provide error checking for variables defined in the data, transformed data, transformed parameters, and generated quantities blocks. Constraints are critical for variables declared in the parameters block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of \\(\\mathbb{R}^n\\)).\nIt is worth calling out the most important aspect of constrained data types:\n\nThe model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.\n\nIf this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.\n\n\n\n\nUnfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.\n\n\nStan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is \\(2^{31}-1\\); the minimum value is \\(-(2^{31})\\).\nWhen integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.\nInteger arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see the Stan Functions Reference integer-valued arithmetic operators section for more information).\n\n\n\nStan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly1 follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly \\(\\pm 2^{1022}\\), which is slightly larger than \\(\\pm 10^{307}\\). It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.\n64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.\nThere are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.\n\n\nThe not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).\n\n\n\nPositive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.\n\n\n\n\nStan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.\nUnlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.\n\n\n\n\nThe complex data type is a scalar, but unlike real and int types, it contains two components, a real and imaginary component, both of which are of type real. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.\n\n\nImaginary literals are written in mathematical notation using a numeral followed by the suffix i. For example, the following example constructs a complex number \\(2 - 1.3i\\) and assigns it to the variable z.\ncomplex z = 2 - 1.3i;\nreal re = get_real(z);  // re has value 2.0\nreal im = get_imag(z);  // im has value -1.3\nThe getter functions then extract the real and imaginary components of z and assign them to re and im respectively.\nThe function to_complex constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.\nvector[K] re;\nvector[K] im;\n// ...\nfor (k in 1:K) {\n  complex z = to_complex(re[k], im[k]);\n  // ...\n}\n\n\n\nExpressions of type real may be assigned to variables of type complex. For example, the following is a valid sequence of Stan statements.\nreal x = 5.0;\ncomplex z = x;  // get_real(z) == 5.0, get_imag(z) == 0\nThe real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.\nAssignability is transitive, so that expressions of type int may also be assigned to variables of type complex, as in the following example.\nint n = 2;\ncomplex z = n;\nFunction arguments also support promotion of integer or real typed expressions to type complex.\n\n\n\n\nAll variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.\n\n\nUnconstrained integers are declared using the int keyword. For example, the variable N is declared to be an integer as follows.\nint N;\n\n\n\nInteger data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare N to be a positive integer, use the following.\nint&lt;lower=1&gt; N;\nThis illustrates that the bounds are inclusive for integers.\nTo declare an integer variable cond to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.\nint&lt;lower=0, upper=1&gt; cond;\n\n\n\nUnconstrained real variables are declared using the keyword real. The following example declares theta to be an unconstrained continuous value.\nreal theta;\n\n\n\nUnconstrained complex numbers are declared using the keyword complex. The following example declares z to be an unconstrained complex variable.\ncomplex z;\n\n\n\nReal variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.\nThe variable sigma may be declared to be non-negative as follows.\nreal&lt;lower=0&gt; sigma;\nThe following declares the variable x to be less than or equal to \\(-1\\).\nreal&lt;upper=-1&gt; x;\nTo ensure rho takes on values between \\(-1\\) and \\(1\\), use the following declaration.\nreal&lt;lower=-1, upper=1&gt; rho;\n\n\nLower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants positive_infinity() and negative_infinity() which may be used for this purpose, or they may be supplied as data.\n\n\n\n\nReal variables may be declared on a space that has been transformed using an affine transformation \\(x\\mapsto \\mu + \\sigma * x\\) with offset \\(\\mu\\) and (positive) multiplier \\(\\sigma\\), using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.\nThe variable x may be declared to have offset \\(1\\) as follows.\nreal&lt;offset=1&gt; x;\nSimilarly, it can be declared to have multiplier \\(2\\) as follows.\nreal&lt;multiplier=2&gt; x;\nFinally, we can combine both declarations to declare a variable with offset \\(1\\) and multiplier \\(2\\).\nreal&lt;offset=1, multiplier=2&gt; x;\nAs an example, we can give x a normal distribution with non-centered parameterization as follows.\nparameters {\n  real&lt;offset=mu, multiplier=sigma&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nRecall that the centered parameterization is achieved with the code\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nor equivalently\nparameters {\n  real&lt;offset=0, multiplier=1&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\n\n\n\nBounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.\nFor example, it is acceptable to have the following declarations.\ndata {\n real lb;\n}\nparameters {\n   real&lt;lower=lb&gt; phi;\n}\nThis declares a real-valued parameter phi to take values greater than the value of the real-valued data variable lb. Constraints may be arbitrary expressions, but must be of type int for integer variables and of type real for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,\ndata {\n   int&lt;lower=1&gt; N;\n   array[N] real y;\n}\nparameters {\n   real&lt;lower=min(y), upper=max(y)&gt; phi;\n}\nThis declares a positive integer data variable N, an array y of real-valued data of length N, and then a parameter ranging between the minimum and maximum value of y. As shown in the example code, the functions min() and max() may be applied to containers such as arrays.\nA more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;  // enforces a &lt; b\n}\ntransformed parameters {\n  real c;\n  real&lt;lower=c&gt; d;\n  c = a;\n  d = b;\n}\nThe parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though c isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.\n\n\n\nA variable may be declared with a size that depends on a boolean constant. For example, consider the definition of alpha in the following program fragment.\ndata {\n  int&lt;lower=0, upper=1&gt; include_alpha;\n  // ...\n}\nparameters {\n  vector[include_alpha ? N : 0] alpha;\n  // ...\n}\nIf include_alpha is true, the model will include the vector alpha; if the flag is false, the model will not include alpha (technically, it will include alpha of size 0, which means it won’t contain any values and won’t be included in any output).\nThis technique is not just useful for containers. If the value of N is set to 1, then the vector alpha will contain a single element and thus alpha[1] behaves like an optional scalar, the existence of which is controlled by include_alpha.\nThis coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the RStanArm package.\n\n\n\n\nStan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A \\(3 \\times 4\\) matrix is a different kind of object in Stan than a \\(3\n\\times 4\\) array.\nThe intention of using matrix types is to call out their usage in the code. There are three situations in Stan where only vectors and matrices may be used,\n\nmatrix arithmetic operations (e.g., matrix multiplication)\nlinear algebra functions (e.g., eigenvalues and determinants), and\nmultivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).\n\nVectors and matrices cannot be typed to return integer values. They are restricted to real and complex values.\nFor constructing vectors and matrices in Stan, see Vector, Matrix, and Array Expressions.\n\n\nVectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.\n\n\n\nVectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword vector, as follows.\nvector[3] u;\nVectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.\nvector&lt;lower=0&gt;[3] u;\nSimilarly, they may be declared with a offset and/or multiplier, as in the following example\nvector&lt;offset=42, multiplier=3&gt;[3] u;\n\n\n\nLike real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword complex_vector, as follows.\ncomplex_vector[3] v;\nComplex vector declarations do not support any constraints.\n\n\n\nA unit simplex is a vector with non-negative values whose entries sum to 1. For instance, \\([0.2,0.3,0.4,0.1]^{\\top}\\) is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-simplex by\nsimplex[5] theta;\nUnit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\nIn high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.\n\n\n\nA stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a \\(3 \\times 4\\) column-stochastic matrix.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.3 \\\\\n0.3 & 0.3 & 0.6 & 0.4 \\\\\n0.5 & 0.2 & 0.3 & 0.3\n\\end{bmatrix}\n\\]\nAn example of a \\(3 \\times 4\\) row-stochastic matrix is the following.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.2 \\\\\n0.2 & 0.1 & 0.6 & 0.1 \\\\\n0.5 & 0.2 & 0.2 & 0.1\n\\end{bmatrix}\n\\]\nIn the examples above, each column (or row) sums to 1, making the matrices valid column_stochastic_matrix and row_stochastic_matrix types.\nColumn-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.\nThe column_stochastic_matrix and row_stochastic_matrix types are declared with row and column sizes. For instance, a matrix theta with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.\ncolumn_stochastic_matrix[3, 4] theta;\nA matrix theta with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.\nrow_stochastic_matrix[3, 4] theta;\nAs with simplexes, column_stochastic_matrix and row_stochastic_matrix variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold \\(\\epsilon\\).\n\n\nIn high-dimensional settings, column_stochastic_matrix and row_stochastic_matrix types may require careful tuning of the inference algorithms. To ensure stability:\n\nSmaller Step Sizes: In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.\nHigher Target Acceptance Rates: Setting higher target acceptance rates can improve the robustness of the sampling process.\nLonger Warmup Periods: Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.\nTighter Optimization Tolerances: For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.\nCustom Initialization: If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.\n\n\n\n\n\nA unit vector is a vector with a norm of one. For instance, \\([0.5,\n0.5, 0.5, 0.5]^{\\top}\\) is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-vector by\nunit_vector[5] theta;\nUnit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nA zero-sum vector is constrained such that the sum of its elements is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\(N - 1\\) degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, beta is declared to be a zero-sum \\(5\\)-vector (4 DoF) by\nsum_to_zero_vector[5] beta;\nZero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nAn ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, \\((-1.3,2.7,2.71)^{\\top}\\) is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see section).\nThe variable c is declared as an ordered 5-vector by\nordered[5] c;\nAfter their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nThere is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, \\((2,3.7,4,12.9)\\) is a positive, ordered 4-vector.\nThe variable d is declared as a positive, ordered 5-vector by\npositive_ordered[5] d;\nLike ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nRow vectors are declared with the keyword row_vector. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector u would be declared as\nrow_vector[1093] u;\nConstraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.\nrow_vector&lt;lower=-1, upper=1&gt;[10] u;\nOffset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.\nrow_vector&lt;offset=-42, multiplier=3&gt;[3] u;\nRow vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.\n\n\n\nComplex row vectors are declared with the keyword complex_row_vector and given a size in basic declarations. For example, a 12-dimensional complex row vector v would be declared as\ncomplex_row_vector[12] v;\nComplex row vectors do not allow constraints.\n\n\n\nMatrices are declared with the keyword matrix along with a number of rows and number of columns. For example,\nmatrix[3, 3] A;\nmatrix[M, N] B;\ndeclares A to be a \\(3 \\times 3\\) matrix and B to be a \\(M\n\\times N\\) matrix. For the second declaration to be well formed, the variables M and N must be declared as integers in either the data or transformed data block and before the matrix declaration.\nMatrices may also be declared with constraints, as in this (\\(3 \\times 4\\)) matrix of non-positive values.\nmatrix&lt;upper=0&gt;[3, 4] B;\nSimilarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.\nmatrix&lt;multiplier=5&gt;[3, 4] B;\n\n\nRows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.\nmatrix[M, N] a;\nrow_vector[N] b;\n// ...\na[1] = b;\nThis copies the values from row vector b to a[1], which is the first row of the matrix a. If the number of columns in a is not the same as the size of b, a run-time error is raised; the number of columns of a is N, which is also the number of columns of b.\nAssignment works by copying values in Stan. That means any subsequent assignment to a[1] does not affect b, nor does an assignment to b affect a.\n\n\n\n\nComplex matrices are declared with the keyword complex_matrix and a number of rows and columns. For example,\ncomplex_matrix[3, 3] C;\nComplex matrices do not allow constraints.\n\n\n\nMatrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,\ncov_matrix[K] Sigma;\ndeclares Sigma to be a \\(K \\times K\\) covariance matrix, where \\(K\\) is the value of the data variable K.\n\n\n\nMatrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between \\(-1\\) and \\(1\\), and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,\ncorr_matrix[3] Omega;\ndeclares Omega to be a \\(3 \\times 3\\) correlation matrix.\nCorrelation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.\nA Cholesky factor \\(L\\) is an \\(M \\times N\\) lower-triangular matrix (if \\(m &lt; n\\) then \\(L[m, n] =0\\)) with a strictly positive diagonal (\\(L[k, k]\n&gt; 0\\)) and \\(M \\geq N\\). If \\(L\\) is a Cholesky factor, then \\(\\Sigma = L\n\\, L^{\\top}\\) is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.\nThe typical case of a square Cholesky factor may be declared with a single dimension,\ncholesky_factor_cov[4] L;\n\n\nIn general, two dimensions may be declared, with the above being equal to cholesky_factor_cov[4, 4]. The type cholesky_factor_cov[M, N] may be used for the general \\(M \\times N\\) case to produce positive semi-definite matrices of rank \\(M\\).\n\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a correlation matrix.\nA Cholesky factor for a correlation matrix \\(L\\) is a \\(K \\times K\\) lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., \\(\\sum_{n=1}^K L_{m,n}^2 = 1\\)). If \\(L\\) is a Cholesky factor for a correlation matrix, then \\(L\\,L^{\\top}\\) is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).\nTo declare the variable L to be a K by K Cholesky factor of a correlation matrix, the following code may be used.\ncholesky_factor_corr[K] L;\n\n\n\nA sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\((N - 1) \\times (M - 1)\\) degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, beta is declared to be a sum-to-zero \\(5 \\times 4\\)-matrix (20 degrees of freedom) by\nsum_to_zero_matrix[5, 4] beta;\nSum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nConstrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be real&lt;lower=0, upper=1&gt; could be assigned to a variable declared as real and vice-versa. Similarly, a variable declared as matrix[3, 3] may be assigned to a variable declared as cov_matrix[3] or cholesky_factor_cov[3], and vice-versa.\nChecks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.\n\n\n\nReal-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.\nvector[N] v = ...;\ncomplex_vector[N] u = 2 * v;\nRow vectors and matrices work the same way.\n\n\n\nVariables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.\ndata {\n  int&lt;lower=0&gt; N_observed, N_missing;\n  // ...\ntransformed parameters {\n  vector[N_observed + N_missing] y;\n  // ...\n\n\n\nIf v is a column vector or row vector, then v[2] is the second element in the vector. If m is a matrix, then m[2, 3] is the value in the second row and third column.\nProviding a matrix with a single index returns the specified row. For instance, if m is a matrix, then m[2] is the second row. This allows Stan blocks such as\nmatrix[M, N] m;\nrow_vector[N] v;\nreal x;\n// ...\nv = m[2];\nx = v[3];   // x == m[2][3] == m[2, 3]\nThe type of m[2] is row_vector because it is the second row of m. Thus it is possible to write m[2][3] instead of m[2, 3] to access the third element in the second row. When given a choice, the form m[2, 3] is preferred.\nComplex versions work the same way,\ncomplex_matrix[M, N] m = ...;\ncomplex_row_vector[N] u = m[3];\ncomplex_vector[M] v = m[ , 2];\n\n\nThe form m[2, 3] is more efficient because it does not require the creation and use of an intermediate expression template for m[2]. In later versions, explicit calls to m[2][3] may be optimized to be as efficient as m[2, 3] by the Stan compiler.\n\n\n\n\nAn integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size M + N using\nvector[M + N] y;\nAny integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.\n\n\n\n\nStan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.\nA two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if a is a two-dimensional array, then a[m, n] is just a convenient shorthand for a[m][n].\nVectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.\nFor constructing arrays in Stan, see Vector, Matrix, and Array Expressions.\n\n\nArrays are declared with the keyword array followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.\nThe variable n is declared as an array of five integers as follows.\narray[5] int n;\nA two-dimensional array of complex values with three rows and four columns is declared as follows.\narray[3, 4] complex a;\nA three-dimensional array z of positive reals with five rows, four columns, and two shelves can be declared as follows.\narray[5, 4, 2] real&lt;lower=0&gt; z;\nArrays may also be declared to contain vectors. For example,\narray[3] vector[7] mu;\ndeclares mu to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example\narray[15, 12] complex_matrix[7, 2] mu;\ndeclares a 15 by 12 array of \\(7 \\times 2\\) complex matrices. Any of the constrained types may also be used in arrays, as in the declaration\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nof a \\(2 \\times 3 \\times 4\\) array of \\(5 \\times 6\\) Cholesky factors of covariance matrices.\n\n\n\nIf x is a 1-dimensional array of length 5, then x[1] is the first element in the array and x[5] is the last. For a \\(3\n\\times 4\\) array y of two dimensions, y[1, 1] is the first element and y[3, 4] the last element. For a three-dimensional array z, the first element is z[1, 1, 1], and so on.\nSubarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose y is a two-dimensional array with three rows and four columns. Then y[3] is one-dimensional array of length four. This means that y[3][1] may be used instead of y[3, 1] to access the value of the first column of the third row of y. The form y[3, 1] is the preferred form (see note in this chapter).\n\n\n\nSubarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as\narray[9, 10, 11] real w;\narray[10, 11] real x;\narray[11] real y;\nreal z;\n// ...\nx = w[5];\ny = x[4];  // y == w[5][4] == w[5, 4]\nz = y[3];  // z == w[5][4][3] == w[5, 4, 3]\nComplex-valued arrays work the same way.\n\n\n\nArrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.\narray[3, 4] vector[5] a;\narray[4] vector[5] b;\nvector[5] c;\nreal x;\nWith these declarations, the following assignments are legal.\nb = a[1];       // result is array of vectors\nc = a[1, 3];    // result is vector\nc = b[3];       //   same result as above\nx = a[1, 3, 5]; // result is scalar\nx = b[3, 5];    //   same result as above\nx = c[5];       //   same result as above\nRow vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.\nConsider the following matrix, vector and scalar declarations.\narray[3, 4] matrix[6, 5] d;\narray[4] matrix[6, 5] e;\nmatrix[6, 5] f;\nrow_vector[5] g;\nreal x;\nWith these declarations, the following definitions are legal.\ne = d[1];           // result is array of matrices\nf = d[1, 3];        // result is matrix\nf = e[3];           //   same result as above\ng = d[1, 3, 2];     // result is row vector\ng = e[3, 2];        //   same result as above\ng = f[2];           //   same result as above\nx = d[1, 3, 5, 2];  // result is scalar\nx = e[3, 5, 2];     //   same result as above\nx = f[5, 2];        //   same result as above\nx = g[2];           //   same result as above\nAs shown, the result f[2] of supplying a single index to a matrix is the indexed row, here row 2 of matrix f.\n\n\n\nSubarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.\narray[I, J, K] real x;\narray[J, K] real y;\narray[K] real z;\n// ...\nx[1] = y;\nx[1, 1] = z;\nThe sizes must match. Here, x[1] is a J by K array, as is y.\nPartial array assignment also works for arrays of matrices, vectors, and row vectors.\n\n\n\nArrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).\n\n\nFor example, vectors cannot be assigned to arrays or vice-versa.\narray[4] real a;\nvector[4] b;\nrow_vector[4] c;\n// ...\na = b; // illegal assignment of vector to array\nb = a; // illegal assignment of array to vector\na = c; // illegal assignment of row vector to array\nc = a; // illegal assignment of array to row vector\n\n\n\nIt is not even legal to assign row vectors to column vectors or vice versa.\nvector[4] b;\nrow_vector[4] c;\n// ...\nb = c; // illegal assignment of row vector to column vector\nc = b; // illegal assignment of column vector to row vector\n\n\n\nThe same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.\narray[3, 4] real a;\nmatrix[3, 4] b;\n// ...\na = b;  // illegal assignment of matrix to array\nb = a;  // illegal assignment of array to matrix\n\n\n\nA \\(1 \\times N\\) matrix cannot be assigned a row vector or vice versa.\nmatrix[1, 4] a;\nrow_vector[4] b;\n// ...\na = b;  // illegal assignment of row vector to matrix\nb = a;  // illegal assignment of matrix to row vector\nSimilarly, an \\(M \\times 1\\) matrix may not be assigned to a column vector.\nmatrix[4, 1] a;\nvector[4] b;\n// ...\na = b;  // illegal assignment of column vector to matrix\nb = a;  // illegal assignment of matrix to column vector\n\n\n\n\nAn integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.\n\n\n\nIf any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare\narray[3, 0] real a;\nthen the resulting size of a is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, a[1] will be a size-zero one-dimensional array. For comparison, declaring\narray[0, 3] real b;\nalso produces an array with an overall size of zero, but in this case, there is no way to index legally into b, because b[0] is undefined. The array will behave at run time as if it’s a \\(0 \\times\n0\\) array. For example, the result of to_matrix(b) will be a \\(0 \\times 0\\) matrix, not a \\(0 \\times 3\\) matrix.\n\n\n\n\nStan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.\n\n\nTuples are declared with the keyword tuple followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as\ntuple(int, vector[3], complex) abc;\nTuples must have at least one entry, so the following declaration is illegal.\ntuple() nil;  // ILLEGAL\nTuples of length one must use a trailing comma, to align with the expression syntax.\ntuple(int,) m; // CORRECT\ntuple(int) n;  // ILLEGAL\nTuples can be assigned as a whole if their elements can be assigned individually. For example, a can be assigned to b in the following example because int can be promoted to complex.\ntuple(int, real) a;\n...\ntuple(complex, real) b = a;\nTuple types may have elements which are declared as tuples, such as the following example.\ntuple(int, tuple(real, complex)) x;\nIn this case, it would probably be simpler to use a 3-tuple type, tuple(int, real, complex).\nTuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration\nreal&lt;lower=0&gt; sigma;\nreal&lt;lower=0, upper=1&gt; theta;\nit is legal to have a tuple with constraints such as\ntuple(real&lt;lower=0&gt;, real&lt;lower=0, upper=1&gt;) sigma_theta;\n\n\n\nTuple elements may be accessed directly. For example, with our declaration of abc from the last section, Stan uses abc.1 for the first element, abc.2 for the second, and abc.3 for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that abc.1 is of type int, abc.2 of type vector[3] and abc.3 of type complex.\n\n\n\nTuple elements can be assigned individually, allowing, e.g.,\ntuple(int, real) ab;\nab.1 = 123;\nab.2 = 12.9;\nAs with other assignments, promotions will happen if necessary (of int to real and of real to complex, along with the corresponding container type promotions).\n\n\n\nFor convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.\nGiven a tuple t of type tuple(T1, ..., Tn) and a sequence of assignable expressions of types v1, …, vn, where each vi has a type which is assignable from type Ti, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement\n(v1, /*...*/, vn) = t;\nNote that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).\nThese unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.\nFor example, if T is a tuple of type tuple(int, (real, real), complex), then the program\nint i;\nreal x, y;\ncomplex z;\n\n(i, (x, y), z) = T;\nAssigns the result of T.1 to i, the result of T.2.1 to x, the result of T.2.2 to y, and the result of T.3 to z.\nThe left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.\n\n\n\n\nThe type information associated with a variable only contains the underlying type and dimensionality of the variable.\n\n\nThe size associated with a given variable is not part of its data type. For example, declaring a variable using\narray[3] real a;\ndeclares the variable a to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.\n\n\nSizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an \\(N \\times M\\) by a \\(J \\times K\\) matrix will raise a run-time error if \\(M \\neq J\\).\n\n\n\n\nLike sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.\nSimilarly, a bounded real may be assigned to an unconstrained real and vice-versa.\n\n\nFor arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.\n\n\n\nFor data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.\nFor parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.\nFor all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to NaN and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).\nFor generated quantities, constraints are enforced after the statements in the generated quantities block have executed.\n\n\n\n\nIn order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.\nBecause size information is not part of a data type, data types will be written without size information. For instance, array[] real is the type of one-dimensional array of reals and matrix is the type of matrices. The three-dimensional integer array type is written as array[,,] int, indicating the number slots available for indexing. Similarly, array[,] vector is the type of a two-dimensional array of vectors.\n\n\n\n\nVariables in Stan are declared by giving a type and a name. For example\nint N;\nvector[N] y;\narray[5] matrix[3, 4] A;\ndeclares a variable N that is an integer, a variable y that is a vector of length N (the previously declared variable), and a variable A, which is a length-5 array where each element is a 3 by 4 matrix.\nThe size of top-level variables in the parameters, transformed parameters, and generated quantities must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.\n// illegal and will be flagged by the compiler:\ngenerated quantities {\n  int N = 10;\n  array[N] int foo;\nDepending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.\n// valid block variables, but not locals or function parameters\nvector&lt;lower=0&gt;[N] u;\n\n// valid as a block or local variable, but not a function parameter\narray[3] int is;\n\n// function parameters exclude sizes and cannot be constrained\nvoid pretty_print_tri_lower(matrix x) { ... }\nTop-level variables can have constraints and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as function parameters are not constrained types and exclude sizes.\nIn the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables M and N indicate number of columns and rows, respectively. The variable K is used for square matrices, i.e., K denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types cov_matrix[K], corr_matrix[K], cholesky_factor_cov[K], and cholesky_factor_corr[K] are only assignable to matrices of dimensions matrix[K, K] types.\n\n\n\n\n\n\n\n\nFunction Argument (unsized)\n   Local\n(unconstrained)\n              Block\n          (constrained)\n\n\n\n\nint\nint\nint\n\n\n\n\nint&lt;lower=L&gt;\n\n\n\n\nint&lt;upper=U&gt;\n\n\n\n\nint&lt;lower=L, upper=U&gt;\n\n\n\n\nint&lt;offset=O&gt;\n\n\n\n\nint&lt;multiplier=M&gt;\n\n\n\n\nint&lt;offset=O, multiplier=M&gt;\n\n\nreal\nreal\nreal\n\n\n\n\nreal&lt;lower=L&gt;\n\n\n\n\nreal&lt;upper=U&gt;\n\n\n\n\nreal&lt;lower=L, upper=U&gt;\n\n\n\n\nreal&lt;offset=O&gt;\n\n\n\n\nreal&lt;multiplier=M&gt;\n\n\n\n\nreal&lt;offset=O, multiplier=M&gt;\n\n\ncomplex\ncomplex\ncomplex\n\n\nvector\nvector[N]\nvector[N]\n\n\n\n\nvector[N]&lt;lower=L&gt;\n\n\n\n\nvector[N]&lt;upper=U&gt;\n\n\n\n\nvector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nvector[N]&lt;offset=O&gt;\n\n\n\n\nvector[N]&lt;multiplier=M&gt;\n\n\n\n\nvector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\nordered[N]\n\n\n\n\npositive_ordered[N]\n\n\n\n\nsimplex[N]\n\n\n\n\nunit_vector[N]\n\n\n\n\nsum_to_zero_vector[N]\n\n\nrow_vector\nrow_vector[N]\nrow_vector[N]\n\n\n\n\nrow_vector[N]&lt;lower=L&gt;\n\n\n\n\nrow_vector[N]&lt;upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O&gt;\n\n\n\n\nrow_vector[N]&lt;multiplier=M&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O, multiplier=M&gt;\n\n\nmatrix\nmatrix[M, N]\nmatrix[M, N]\n\n\n\n\nmatrix[M, N]&lt;lower=L&gt;\n\n\n\n\nmatrix[M, N]&lt;upper=U&gt;\n\n\n\n\nmatrix[M, N]&lt;lower=L, upper=U&gt; |\n\n\n\n\nmatrix[M, N]&lt;offset=O&gt;\n\n\n\n\nmatrix[M, N]&lt;multiplier=M&gt;\n\n\n\n\nmatrix[M, N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\ncolumn_stochastic_matrix[M, N]\n\n\n\n\nrow_stochastic_matrix[M, N]\n\n\n\n\nsum_to_zero_matrix[M, N]\n\n\n\nmatrix[K, K]\ncorr_matrix[K]\n\n\n\nmatrix[K, K]\ncov_matrix[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_corr[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_cov[K]\n\n\ncomplex_vector\ncomplex_vector[M]\ncomplex_vector[M]\n\n\ncomplex_row_vector\ncomplex_row_vector[N]\ncomplex_row_vector[N]\n\n\ncomplex_matrix\ncomplex_matrix[M, N]\ncomplex_matrix[M,N]\n\n\narray[] vector\narray[M] vector[N]\narray[M] vector[N]\n\n\n\n\narray[M] vector[N]&lt;lower=L&gt;\n\n\n\n\narray[M] vector[N]&lt;upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O&gt;\n\n\n\n\narray[M] vector[N]&lt;multiplier=M&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\narray[M] ordered[N]\n\n\n\n\narray[M] positive_ordered[N]\n\n\n\n\narray[M] simplex[N]\n\n\n\n\narray[M] unit_vector[N]\n\n\n\n\narray[M] sum_to_zero_vector[N]\n\n\n\n\nAdditional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g. array[ , ] is a 2-D array.\nFor more on how function arguments and return types are declared, consult the User’s Guide chapter on functions.\n\n\n\nStan allows assignable variables to be declared and defined in a single statement. Assignable variables are\n\nlocal variables, and\nvariables declared in the transformed data, transformed parameters, or generated quantities blocks.\n\nFor example, the statement\nint N = 5;\ndeclares the variable N to be an integer scalar type and at the same time defines it to be the value of the expression 5.\n\n\nThe type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have\nreal sum = 0;\neven though 0 is of type int and sum is of type real, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.\nVariables of any type may have values assigned to them. For example,\nmatrix[3, 2] a = b;\ndeclares a \\(3 \\times 2\\) matrix variable a and assigns a copy of the value of b to the variable a. The variable b must be of type matrix for the statement to be well formed. For the code to execute successfully, b must be the same shape as a, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to a do not affect b and subsequent changes to b do not affect a.\n\n\n\nThe right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,\nmatrix[3, 2] a = 0.5 * (b + c);\nassigns the matrix variable a to half of the sum of b and c. The only requirement on b and c is that the expression b + c be of type matrix. For example, b could be of type matrix and c of type real, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.\nSimilarly,\ncomplex z = 2 + 3i;\nassigns the the complex number \\(2 + 3i\\) to the complex scalar z. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).\n\n\n\nAny variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.\n\n\n\n\nStan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.\n\n\nThe code:\nreal x, y;\nis equivalent to\nreal x;\nreal y;\nAs a result, all declarations on the same line must be of the same type.\n\n\n\nThe ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in the section introducing compound declarations and definitions :\nreal x = 3, y = 5.6;\nConstrained data types can also be declared together, so long as the constraint for each variable is the same:\nreal&lt;lower=0&gt; x, y;",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#overview-of-data-types",
+    "href": "reference-manual/types.html#overview-of-data-types",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Arguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.\nPassing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.\n\n\nStan provides two primitive data types, real for continuous values and int for integer values. These are both considered scalar types.\n\n\n\nStan provides a complex number data type complex, where a complex number contains both a real and an imaginary component, both of which are of type real. Complex types are considered scalar types.\n\n\n\nStan provides three real-valued matrix data types, vector for column vectors, row_vector for row vectors, and matrix for matrices.\nStan also provides three complex-valued matrix data types, complex_vector for column vectors, complex_row_vector for row vectors, and complex_matrix for matrices.\n\n\n\nAny type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,\narray[10] real x;\narray[6, 7] matrix[3, 3] m;\narray[12, 8, 15] complex z;\ndeclares x to be a one-dimensional array of size 10 containing real values, declares m to be a two-dimensional array of size \\(6 \\times 7\\) containing values that are \\(3 \\times 3\\) matrices, and declares z to be a \\(12 \\times 8 \\times 15\\) array of complex numbers.\nPrior to 2.26 Stan models used a different syntax which has since been removed. See the Removed Features chapter for more details.\n\n\n\nFor any sequence of types, Stan provides a tuple data type. For example,\ntuple(real, array[5] int) xi;\ndeclares xi to be a tuple holding two values, the first of which is of type type real and the second of which a 5-dimensional array of type int.\n\n\n\nDeclarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.\nAll of the basic data types other than complex may be given lower and upper bounds using syntax such as\nint&lt;lower=1&gt; N;\nreal&lt;upper=0&gt; log_p;\nvector&lt;lower=-1, upper=1&gt;[3] rho;\nThere are also special data types for structured vectors and matrices. There are five constrained vector data types, simplex for unit simplexes, unit_vector for unit-length vectors, sum_to_zero_vector for vectors that sum to zero, ordered for ordered vectors of scalars, and positive_ordered for vectors of positive ordered scalars. There are specialized matrix data types corr_matrix and cov_matrix for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type cholesky_factor_cov is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type cholesky_factor_corr is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type sum_to_zero_matrix is for matrices that sum to zero across rows and columns.\nConstraints provide error checking for variables defined in the data, transformed data, transformed parameters, and generated quantities blocks. Constraints are critical for variables declared in the parameters block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of \\(\\mathbb{R}^n\\)).\nIt is worth calling out the most important aspect of constrained data types:\n\nThe model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.\n\nIf this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#numerical-data-types",
+    "href": "reference-manual/types.html#numerical-data-types",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Unfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.\n\n\nStan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is \\(2^{31}-1\\); the minimum value is \\(-(2^{31})\\).\nWhen integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.\nInteger arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see the Stan Functions Reference integer-valued arithmetic operators section for more information).\n\n\n\nStan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly1 follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly \\(\\pm 2^{1022}\\), which is slightly larger than \\(\\pm 10^{307}\\). It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.\n64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.\nThere are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.\n\n\nThe not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).\n\n\n\nPositive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.\n\n\n\n\nStan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.\nUnlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#complex-numerical-data-type",
+    "href": "reference-manual/types.html#complex-numerical-data-type",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "The complex data type is a scalar, but unlike real and int types, it contains two components, a real and imaginary component, both of which are of type real. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.\n\n\nImaginary literals are written in mathematical notation using a numeral followed by the suffix i. For example, the following example constructs a complex number \\(2 - 1.3i\\) and assigns it to the variable z.\ncomplex z = 2 - 1.3i;\nreal re = get_real(z);  // re has value 2.0\nreal im = get_imag(z);  // im has value -1.3\nThe getter functions then extract the real and imaginary components of z and assign them to re and im respectively.\nThe function to_complex constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.\nvector[K] re;\nvector[K] im;\n// ...\nfor (k in 1:K) {\n  complex z = to_complex(re[k], im[k]);\n  // ...\n}\n\n\n\nExpressions of type real may be assigned to variables of type complex. For example, the following is a valid sequence of Stan statements.\nreal x = 5.0;\ncomplex z = x;  // get_real(z) == 5.0, get_imag(z) == 0\nThe real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.\nAssignability is transitive, so that expressions of type int may also be assigned to variables of type complex, as in the following example.\nint n = 2;\ncomplex z = n;\nFunction arguments also support promotion of integer or real typed expressions to type complex.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#scalar-data-types-and-variable-declarations",
+    "href": "reference-manual/types.html#scalar-data-types-and-variable-declarations",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "All variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.\n\n\nUnconstrained integers are declared using the int keyword. For example, the variable N is declared to be an integer as follows.\nint N;\n\n\n\nInteger data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare N to be a positive integer, use the following.\nint&lt;lower=1&gt; N;\nThis illustrates that the bounds are inclusive for integers.\nTo declare an integer variable cond to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.\nint&lt;lower=0, upper=1&gt; cond;\n\n\n\nUnconstrained real variables are declared using the keyword real. The following example declares theta to be an unconstrained continuous value.\nreal theta;\n\n\n\nUnconstrained complex numbers are declared using the keyword complex. The following example declares z to be an unconstrained complex variable.\ncomplex z;\n\n\n\nReal variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.\nThe variable sigma may be declared to be non-negative as follows.\nreal&lt;lower=0&gt; sigma;\nThe following declares the variable x to be less than or equal to \\(-1\\).\nreal&lt;upper=-1&gt; x;\nTo ensure rho takes on values between \\(-1\\) and \\(1\\), use the following declaration.\nreal&lt;lower=-1, upper=1&gt; rho;\n\n\nLower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants positive_infinity() and negative_infinity() which may be used for this purpose, or they may be supplied as data.\n\n\n\n\nReal variables may be declared on a space that has been transformed using an affine transformation \\(x\\mapsto \\mu + \\sigma * x\\) with offset \\(\\mu\\) and (positive) multiplier \\(\\sigma\\), using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.\nThe variable x may be declared to have offset \\(1\\) as follows.\nreal&lt;offset=1&gt; x;\nSimilarly, it can be declared to have multiplier \\(2\\) as follows.\nreal&lt;multiplier=2&gt; x;\nFinally, we can combine both declarations to declare a variable with offset \\(1\\) and multiplier \\(2\\).\nreal&lt;offset=1, multiplier=2&gt; x;\nAs an example, we can give x a normal distribution with non-centered parameterization as follows.\nparameters {\n  real&lt;offset=mu, multiplier=sigma&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nRecall that the centered parameterization is achieved with the code\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nor equivalently\nparameters {\n  real&lt;offset=0, multiplier=1&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\n\n\n\nBounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.\nFor example, it is acceptable to have the following declarations.\ndata {\n real lb;\n}\nparameters {\n   real&lt;lower=lb&gt; phi;\n}\nThis declares a real-valued parameter phi to take values greater than the value of the real-valued data variable lb. Constraints may be arbitrary expressions, but must be of type int for integer variables and of type real for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,\ndata {\n   int&lt;lower=1&gt; N;\n   array[N] real y;\n}\nparameters {\n   real&lt;lower=min(y), upper=max(y)&gt; phi;\n}\nThis declares a positive integer data variable N, an array y of real-valued data of length N, and then a parameter ranging between the minimum and maximum value of y. As shown in the example code, the functions min() and max() may be applied to containers such as arrays.\nA more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;  // enforces a &lt; b\n}\ntransformed parameters {\n  real c;\n  real&lt;lower=c&gt; d;\n  c = a;\n  d = b;\n}\nThe parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though c isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.\n\n\n\nA variable may be declared with a size that depends on a boolean constant. For example, consider the definition of alpha in the following program fragment.\ndata {\n  int&lt;lower=0, upper=1&gt; include_alpha;\n  // ...\n}\nparameters {\n  vector[include_alpha ? N : 0] alpha;\n  // ...\n}\nIf include_alpha is true, the model will include the vector alpha; if the flag is false, the model will not include alpha (technically, it will include alpha of size 0, which means it won’t contain any values and won’t be included in any output).\nThis technique is not just useful for containers. If the value of N is set to 1, then the vector alpha will contain a single element and thus alpha[1] behaves like an optional scalar, the existence of which is controlled by include_alpha.\nThis coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the RStanArm package.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#vector-and-matrix-data-types",
+    "href": "reference-manual/types.html#vector-and-matrix-data-types",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Stan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A \\(3 \\times 4\\) matrix is a different kind of object in Stan than a \\(3\n\\times 4\\) array.\nThe intention of using matrix types is to call out their usage in the code. There are three situations in Stan where only vectors and matrices may be used,\n\nmatrix arithmetic operations (e.g., matrix multiplication)\nlinear algebra functions (e.g., eigenvalues and determinants), and\nmultivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).\n\nVectors and matrices cannot be typed to return integer values. They are restricted to real and complex values.\nFor constructing vectors and matrices in Stan, see Vector, Matrix, and Array Expressions.\n\n\nVectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.\n\n\n\nVectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword vector, as follows.\nvector[3] u;\nVectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.\nvector&lt;lower=0&gt;[3] u;\nSimilarly, they may be declared with a offset and/or multiplier, as in the following example\nvector&lt;offset=42, multiplier=3&gt;[3] u;\n\n\n\nLike real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword complex_vector, as follows.\ncomplex_vector[3] v;\nComplex vector declarations do not support any constraints.\n\n\n\nA unit simplex is a vector with non-negative values whose entries sum to 1. For instance, \\([0.2,0.3,0.4,0.1]^{\\top}\\) is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-simplex by\nsimplex[5] theta;\nUnit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\nIn high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.\n\n\n\nA stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a \\(3 \\times 4\\) column-stochastic matrix.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.3 \\\\\n0.3 & 0.3 & 0.6 & 0.4 \\\\\n0.5 & 0.2 & 0.3 & 0.3\n\\end{bmatrix}\n\\]\nAn example of a \\(3 \\times 4\\) row-stochastic matrix is the following.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.2 \\\\\n0.2 & 0.1 & 0.6 & 0.1 \\\\\n0.5 & 0.2 & 0.2 & 0.1\n\\end{bmatrix}\n\\]\nIn the examples above, each column (or row) sums to 1, making the matrices valid column_stochastic_matrix and row_stochastic_matrix types.\nColumn-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.\nThe column_stochastic_matrix and row_stochastic_matrix types are declared with row and column sizes. For instance, a matrix theta with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.\ncolumn_stochastic_matrix[3, 4] theta;\nA matrix theta with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.\nrow_stochastic_matrix[3, 4] theta;\nAs with simplexes, column_stochastic_matrix and row_stochastic_matrix variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold \\(\\epsilon\\).\n\n\nIn high-dimensional settings, column_stochastic_matrix and row_stochastic_matrix types may require careful tuning of the inference algorithms. To ensure stability:\n\nSmaller Step Sizes: In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.\nHigher Target Acceptance Rates: Setting higher target acceptance rates can improve the robustness of the sampling process.\nLonger Warmup Periods: Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.\nTighter Optimization Tolerances: For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.\nCustom Initialization: If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.\n\n\n\n\n\nA unit vector is a vector with a norm of one. For instance, \\([0.5,\n0.5, 0.5, 0.5]^{\\top}\\) is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-vector by\nunit_vector[5] theta;\nUnit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nA zero-sum vector is constrained such that the sum of its elements is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\(N - 1\\) degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, beta is declared to be a zero-sum \\(5\\)-vector (4 DoF) by\nsum_to_zero_vector[5] beta;\nZero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nAn ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, \\((-1.3,2.7,2.71)^{\\top}\\) is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see section).\nThe variable c is declared as an ordered 5-vector by\nordered[5] c;\nAfter their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nThere is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, \\((2,3.7,4,12.9)\\) is a positive, ordered 4-vector.\nThe variable d is declared as a positive, ordered 5-vector by\npositive_ordered[5] d;\nLike ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nRow vectors are declared with the keyword row_vector. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector u would be declared as\nrow_vector[1093] u;\nConstraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.\nrow_vector&lt;lower=-1, upper=1&gt;[10] u;\nOffset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.\nrow_vector&lt;offset=-42, multiplier=3&gt;[3] u;\nRow vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.\n\n\n\nComplex row vectors are declared with the keyword complex_row_vector and given a size in basic declarations. For example, a 12-dimensional complex row vector v would be declared as\ncomplex_row_vector[12] v;\nComplex row vectors do not allow constraints.\n\n\n\nMatrices are declared with the keyword matrix along with a number of rows and number of columns. For example,\nmatrix[3, 3] A;\nmatrix[M, N] B;\ndeclares A to be a \\(3 \\times 3\\) matrix and B to be a \\(M\n\\times N\\) matrix. For the second declaration to be well formed, the variables M and N must be declared as integers in either the data or transformed data block and before the matrix declaration.\nMatrices may also be declared with constraints, as in this (\\(3 \\times 4\\)) matrix of non-positive values.\nmatrix&lt;upper=0&gt;[3, 4] B;\nSimilarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.\nmatrix&lt;multiplier=5&gt;[3, 4] B;\n\n\nRows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.\nmatrix[M, N] a;\nrow_vector[N] b;\n// ...\na[1] = b;\nThis copies the values from row vector b to a[1], which is the first row of the matrix a. If the number of columns in a is not the same as the size of b, a run-time error is raised; the number of columns of a is N, which is also the number of columns of b.\nAssignment works by copying values in Stan. That means any subsequent assignment to a[1] does not affect b, nor does an assignment to b affect a.\n\n\n\n\nComplex matrices are declared with the keyword complex_matrix and a number of rows and columns. For example,\ncomplex_matrix[3, 3] C;\nComplex matrices do not allow constraints.\n\n\n\nMatrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,\ncov_matrix[K] Sigma;\ndeclares Sigma to be a \\(K \\times K\\) covariance matrix, where \\(K\\) is the value of the data variable K.\n\n\n\nMatrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between \\(-1\\) and \\(1\\), and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,\ncorr_matrix[3] Omega;\ndeclares Omega to be a \\(3 \\times 3\\) correlation matrix.\nCorrelation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.\nA Cholesky factor \\(L\\) is an \\(M \\times N\\) lower-triangular matrix (if \\(m &lt; n\\) then \\(L[m, n] =0\\)) with a strictly positive diagonal (\\(L[k, k]\n&gt; 0\\)) and \\(M \\geq N\\). If \\(L\\) is a Cholesky factor, then \\(\\Sigma = L\n\\, L^{\\top}\\) is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.\nThe typical case of a square Cholesky factor may be declared with a single dimension,\ncholesky_factor_cov[4] L;\n\n\nIn general, two dimensions may be declared, with the above being equal to cholesky_factor_cov[4, 4]. The type cholesky_factor_cov[M, N] may be used for the general \\(M \\times N\\) case to produce positive semi-definite matrices of rank \\(M\\).\n\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a correlation matrix.\nA Cholesky factor for a correlation matrix \\(L\\) is a \\(K \\times K\\) lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., \\(\\sum_{n=1}^K L_{m,n}^2 = 1\\)). If \\(L\\) is a Cholesky factor for a correlation matrix, then \\(L\\,L^{\\top}\\) is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).\nTo declare the variable L to be a K by K Cholesky factor of a correlation matrix, the following code may be used.\ncholesky_factor_corr[K] L;\n\n\n\nA sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\((N - 1) \\times (M - 1)\\) degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, beta is declared to be a sum-to-zero \\(5 \\times 4\\)-matrix (20 degrees of freedom) by\nsum_to_zero_matrix[5, 4] beta;\nSum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nConstrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be real&lt;lower=0, upper=1&gt; could be assigned to a variable declared as real and vice-versa. Similarly, a variable declared as matrix[3, 3] may be assigned to a variable declared as cov_matrix[3] or cholesky_factor_cov[3], and vice-versa.\nChecks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.\n\n\n\nReal-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.\nvector[N] v = ...;\ncomplex_vector[N] u = 2 * v;\nRow vectors and matrices work the same way.\n\n\n\nVariables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.\ndata {\n  int&lt;lower=0&gt; N_observed, N_missing;\n  // ...\ntransformed parameters {\n  vector[N_observed + N_missing] y;\n  // ...\n\n\n\nIf v is a column vector or row vector, then v[2] is the second element in the vector. If m is a matrix, then m[2, 3] is the value in the second row and third column.\nProviding a matrix with a single index returns the specified row. For instance, if m is a matrix, then m[2] is the second row. This allows Stan blocks such as\nmatrix[M, N] m;\nrow_vector[N] v;\nreal x;\n// ...\nv = m[2];\nx = v[3];   // x == m[2][3] == m[2, 3]\nThe type of m[2] is row_vector because it is the second row of m. Thus it is possible to write m[2][3] instead of m[2, 3] to access the third element in the second row. When given a choice, the form m[2, 3] is preferred.\nComplex versions work the same way,\ncomplex_matrix[M, N] m = ...;\ncomplex_row_vector[N] u = m[3];\ncomplex_vector[M] v = m[ , 2];\n\n\nThe form m[2, 3] is more efficient because it does not require the creation and use of an intermediate expression template for m[2]. In later versions, explicit calls to m[2][3] may be optimized to be as efficient as m[2, 3] by the Stan compiler.\n\n\n\n\nAn integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size M + N using\nvector[M + N] y;\nAny integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#array-data-types.section",
+    "href": "reference-manual/types.html#array-data-types.section",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Stan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.\nA two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if a is a two-dimensional array, then a[m, n] is just a convenient shorthand for a[m][n].\nVectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.\nFor constructing arrays in Stan, see Vector, Matrix, and Array Expressions.\n\n\nArrays are declared with the keyword array followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.\nThe variable n is declared as an array of five integers as follows.\narray[5] int n;\nA two-dimensional array of complex values with three rows and four columns is declared as follows.\narray[3, 4] complex a;\nA three-dimensional array z of positive reals with five rows, four columns, and two shelves can be declared as follows.\narray[5, 4, 2] real&lt;lower=0&gt; z;\nArrays may also be declared to contain vectors. For example,\narray[3] vector[7] mu;\ndeclares mu to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example\narray[15, 12] complex_matrix[7, 2] mu;\ndeclares a 15 by 12 array of \\(7 \\times 2\\) complex matrices. Any of the constrained types may also be used in arrays, as in the declaration\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nof a \\(2 \\times 3 \\times 4\\) array of \\(5 \\times 6\\) Cholesky factors of covariance matrices.\n\n\n\nIf x is a 1-dimensional array of length 5, then x[1] is the first element in the array and x[5] is the last. For a \\(3\n\\times 4\\) array y of two dimensions, y[1, 1] is the first element and y[3, 4] the last element. For a three-dimensional array z, the first element is z[1, 1, 1], and so on.\nSubarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose y is a two-dimensional array with three rows and four columns. Then y[3] is one-dimensional array of length four. This means that y[3][1] may be used instead of y[3, 1] to access the value of the first column of the third row of y. The form y[3, 1] is the preferred form (see note in this chapter).\n\n\n\nSubarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as\narray[9, 10, 11] real w;\narray[10, 11] real x;\narray[11] real y;\nreal z;\n// ...\nx = w[5];\ny = x[4];  // y == w[5][4] == w[5, 4]\nz = y[3];  // z == w[5][4][3] == w[5, 4, 3]\nComplex-valued arrays work the same way.\n\n\n\nArrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.\narray[3, 4] vector[5] a;\narray[4] vector[5] b;\nvector[5] c;\nreal x;\nWith these declarations, the following assignments are legal.\nb = a[1];       // result is array of vectors\nc = a[1, 3];    // result is vector\nc = b[3];       //   same result as above\nx = a[1, 3, 5]; // result is scalar\nx = b[3, 5];    //   same result as above\nx = c[5];       //   same result as above\nRow vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.\nConsider the following matrix, vector and scalar declarations.\narray[3, 4] matrix[6, 5] d;\narray[4] matrix[6, 5] e;\nmatrix[6, 5] f;\nrow_vector[5] g;\nreal x;\nWith these declarations, the following definitions are legal.\ne = d[1];           // result is array of matrices\nf = d[1, 3];        // result is matrix\nf = e[3];           //   same result as above\ng = d[1, 3, 2];     // result is row vector\ng = e[3, 2];        //   same result as above\ng = f[2];           //   same result as above\nx = d[1, 3, 5, 2];  // result is scalar\nx = e[3, 5, 2];     //   same result as above\nx = f[5, 2];        //   same result as above\nx = g[2];           //   same result as above\nAs shown, the result f[2] of supplying a single index to a matrix is the indexed row, here row 2 of matrix f.\n\n\n\nSubarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.\narray[I, J, K] real x;\narray[J, K] real y;\narray[K] real z;\n// ...\nx[1] = y;\nx[1, 1] = z;\nThe sizes must match. Here, x[1] is a J by K array, as is y.\nPartial array assignment also works for arrays of matrices, vectors, and row vectors.\n\n\n\nArrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).\n\n\nFor example, vectors cannot be assigned to arrays or vice-versa.\narray[4] real a;\nvector[4] b;\nrow_vector[4] c;\n// ...\na = b; // illegal assignment of vector to array\nb = a; // illegal assignment of array to vector\na = c; // illegal assignment of row vector to array\nc = a; // illegal assignment of array to row vector\n\n\n\nIt is not even legal to assign row vectors to column vectors or vice versa.\nvector[4] b;\nrow_vector[4] c;\n// ...\nb = c; // illegal assignment of row vector to column vector\nc = b; // illegal assignment of column vector to row vector\n\n\n\nThe same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.\narray[3, 4] real a;\nmatrix[3, 4] b;\n// ...\na = b;  // illegal assignment of matrix to array\nb = a;  // illegal assignment of array to matrix\n\n\n\nA \\(1 \\times N\\) matrix cannot be assigned a row vector or vice versa.\nmatrix[1, 4] a;\nrow_vector[4] b;\n// ...\na = b;  // illegal assignment of row vector to matrix\nb = a;  // illegal assignment of matrix to row vector\nSimilarly, an \\(M \\times 1\\) matrix may not be assigned to a column vector.\nmatrix[4, 1] a;\nvector[4] b;\n// ...\na = b;  // illegal assignment of column vector to matrix\nb = a;  // illegal assignment of matrix to column vector\n\n\n\n\nAn integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.\n\n\n\nIf any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare\narray[3, 0] real a;\nthen the resulting size of a is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, a[1] will be a size-zero one-dimensional array. For comparison, declaring\narray[0, 3] real b;\nalso produces an array with an overall size of zero, but in this case, there is no way to index legally into b, because b[0] is undefined. The array will behave at run time as if it’s a \\(0 \\times\n0\\) array. For example, the result of to_matrix(b) will be a \\(0 \\times 0\\) matrix, not a \\(0 \\times 3\\) matrix.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#tuple-data-type",
+    "href": "reference-manual/types.html#tuple-data-type",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Stan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.\n\n\nTuples are declared with the keyword tuple followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as\ntuple(int, vector[3], complex) abc;\nTuples must have at least one entry, so the following declaration is illegal.\ntuple() nil;  // ILLEGAL\nTuples of length one must use a trailing comma, to align with the expression syntax.\ntuple(int,) m; // CORRECT\ntuple(int) n;  // ILLEGAL\nTuples can be assigned as a whole if their elements can be assigned individually. For example, a can be assigned to b in the following example because int can be promoted to complex.\ntuple(int, real) a;\n...\ntuple(complex, real) b = a;\nTuple types may have elements which are declared as tuples, such as the following example.\ntuple(int, tuple(real, complex)) x;\nIn this case, it would probably be simpler to use a 3-tuple type, tuple(int, real, complex).\nTuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration\nreal&lt;lower=0&gt; sigma;\nreal&lt;lower=0, upper=1&gt; theta;\nit is legal to have a tuple with constraints such as\ntuple(real&lt;lower=0&gt;, real&lt;lower=0, upper=1&gt;) sigma_theta;\n\n\n\nTuple elements may be accessed directly. For example, with our declaration of abc from the last section, Stan uses abc.1 for the first element, abc.2 for the second, and abc.3 for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that abc.1 is of type int, abc.2 of type vector[3] and abc.3 of type complex.\n\n\n\nTuple elements can be assigned individually, allowing, e.g.,\ntuple(int, real) ab;\nab.1 = 123;\nab.2 = 12.9;\nAs with other assignments, promotions will happen if necessary (of int to real and of real to complex, along with the corresponding container type promotions).\n\n\n\nFor convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.\nGiven a tuple t of type tuple(T1, ..., Tn) and a sequence of assignable expressions of types v1, …, vn, where each vi has a type which is assignable from type Ti, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement\n(v1, /*...*/, vn) = t;\nNote that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).\nThese unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.\nFor example, if T is a tuple of type tuple(int, (real, real), complex), then the program\nint i;\nreal x, y;\ncomplex z;\n\n(i, (x, y), z) = T;\nAssigns the result of T.1 to i, the result of T.2.1 to x, the result of T.2.2 to y, and the result of T.3 to z.\nThe left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#variable-types-vs.-constraints-and-sizes",
+    "href": "reference-manual/types.html#variable-types-vs.-constraints-and-sizes",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "The type information associated with a variable only contains the underlying type and dimensionality of the variable.\n\n\nThe size associated with a given variable is not part of its data type. For example, declaring a variable using\narray[3] real a;\ndeclares the variable a to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.\n\n\nSizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an \\(N \\times M\\) by a \\(J \\times K\\) matrix will raise a run-time error if \\(M \\neq J\\).\n\n\n\n\nLike sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.\nSimilarly, a bounded real may be assigned to an unconstrained real and vice-versa.\n\n\nFor arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.\n\n\n\nFor data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.\nFor parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.\nFor all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to NaN and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).\nFor generated quantities, constraints are enforced after the statements in the generated quantities block have executed.\n\n\n\n\nIn order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.\nBecause size information is not part of a data type, data types will be written without size information. For instance, array[] real is the type of one-dimensional array of reals and matrix is the type of matrices. The three-dimensional integer array type is written as array[,,] int, indicating the number slots available for indexing. Similarly, array[,] vector is the type of a two-dimensional array of vectors.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#variable-declaration.section",
+    "href": "reference-manual/types.html#variable-declaration.section",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Variables in Stan are declared by giving a type and a name. For example\nint N;\nvector[N] y;\narray[5] matrix[3, 4] A;\ndeclares a variable N that is an integer, a variable y that is a vector of length N (the previously declared variable), and a variable A, which is a length-5 array where each element is a 3 by 4 matrix.\nThe size of top-level variables in the parameters, transformed parameters, and generated quantities must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.\n// illegal and will be flagged by the compiler:\ngenerated quantities {\n  int N = 10;\n  array[N] int foo;\nDepending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.\n// valid block variables, but not locals or function parameters\nvector&lt;lower=0&gt;[N] u;\n\n// valid as a block or local variable, but not a function parameter\narray[3] int is;\n\n// function parameters exclude sizes and cannot be constrained\nvoid pretty_print_tri_lower(matrix x) { ... }\nTop-level variables can have constraints and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as function parameters are not constrained types and exclude sizes.\nIn the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables M and N indicate number of columns and rows, respectively. The variable K is used for square matrices, i.e., K denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types cov_matrix[K], corr_matrix[K], cholesky_factor_cov[K], and cholesky_factor_corr[K] are only assignable to matrices of dimensions matrix[K, K] types.\n\n\n\n\n\n\n\n\nFunction Argument (unsized)\n   Local\n(unconstrained)\n              Block\n          (constrained)\n\n\n\n\nint\nint\nint\n\n\n\n\nint&lt;lower=L&gt;\n\n\n\n\nint&lt;upper=U&gt;\n\n\n\n\nint&lt;lower=L, upper=U&gt;\n\n\n\n\nint&lt;offset=O&gt;\n\n\n\n\nint&lt;multiplier=M&gt;\n\n\n\n\nint&lt;offset=O, multiplier=M&gt;\n\n\nreal\nreal\nreal\n\n\n\n\nreal&lt;lower=L&gt;\n\n\n\n\nreal&lt;upper=U&gt;\n\n\n\n\nreal&lt;lower=L, upper=U&gt;\n\n\n\n\nreal&lt;offset=O&gt;\n\n\n\n\nreal&lt;multiplier=M&gt;\n\n\n\n\nreal&lt;offset=O, multiplier=M&gt;\n\n\ncomplex\ncomplex\ncomplex\n\n\nvector\nvector[N]\nvector[N]\n\n\n\n\nvector[N]&lt;lower=L&gt;\n\n\n\n\nvector[N]&lt;upper=U&gt;\n\n\n\n\nvector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nvector[N]&lt;offset=O&gt;\n\n\n\n\nvector[N]&lt;multiplier=M&gt;\n\n\n\n\nvector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\nordered[N]\n\n\n\n\npositive_ordered[N]\n\n\n\n\nsimplex[N]\n\n\n\n\nunit_vector[N]\n\n\n\n\nsum_to_zero_vector[N]\n\n\nrow_vector\nrow_vector[N]\nrow_vector[N]\n\n\n\n\nrow_vector[N]&lt;lower=L&gt;\n\n\n\n\nrow_vector[N]&lt;upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O&gt;\n\n\n\n\nrow_vector[N]&lt;multiplier=M&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O, multiplier=M&gt;\n\n\nmatrix\nmatrix[M, N]\nmatrix[M, N]\n\n\n\n\nmatrix[M, N]&lt;lower=L&gt;\n\n\n\n\nmatrix[M, N]&lt;upper=U&gt;\n\n\n\n\nmatrix[M, N]&lt;lower=L, upper=U&gt; |\n\n\n\n\nmatrix[M, N]&lt;offset=O&gt;\n\n\n\n\nmatrix[M, N]&lt;multiplier=M&gt;\n\n\n\n\nmatrix[M, N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\ncolumn_stochastic_matrix[M, N]\n\n\n\n\nrow_stochastic_matrix[M, N]\n\n\n\n\nsum_to_zero_matrix[M, N]\n\n\n\nmatrix[K, K]\ncorr_matrix[K]\n\n\n\nmatrix[K, K]\ncov_matrix[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_corr[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_cov[K]\n\n\ncomplex_vector\ncomplex_vector[M]\ncomplex_vector[M]\n\n\ncomplex_row_vector\ncomplex_row_vector[N]\ncomplex_row_vector[N]\n\n\ncomplex_matrix\ncomplex_matrix[M, N]\ncomplex_matrix[M,N]\n\n\narray[] vector\narray[M] vector[N]\narray[M] vector[N]\n\n\n\n\narray[M] vector[N]&lt;lower=L&gt;\n\n\n\n\narray[M] vector[N]&lt;upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O&gt;\n\n\n\n\narray[M] vector[N]&lt;multiplier=M&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\narray[M] ordered[N]\n\n\n\n\narray[M] positive_ordered[N]\n\n\n\n\narray[M] simplex[N]\n\n\n\n\narray[M] unit_vector[N]\n\n\n\n\narray[M] sum_to_zero_vector[N]\n\n\n\n\nAdditional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g. array[ , ] is a 2-D array.\nFor more on how function arguments and return types are declared, consult the User’s Guide chapter on functions.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#compound-variable-declaration-and-definition",
+    "href": "reference-manual/types.html#compound-variable-declaration-and-definition",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Stan allows assignable variables to be declared and defined in a single statement. Assignable variables are\n\nlocal variables, and\nvariables declared in the transformed data, transformed parameters, or generated quantities blocks.\n\nFor example, the statement\nint N = 5;\ndeclares the variable N to be an integer scalar type and at the same time defines it to be the value of the expression 5.\n\n\nThe type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have\nreal sum = 0;\neven though 0 is of type int and sum is of type real, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.\nVariables of any type may have values assigned to them. For example,\nmatrix[3, 2] a = b;\ndeclares a \\(3 \\times 2\\) matrix variable a and assigns a copy of the value of b to the variable a. The variable b must be of type matrix for the statement to be well formed. For the code to execute successfully, b must be the same shape as a, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to a do not affect b and subsequent changes to b do not affect a.\n\n\n\nThe right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,\nmatrix[3, 2] a = 0.5 * (b + c);\nassigns the matrix variable a to half of the sum of b and c. The only requirement on b and c is that the expression b + c be of type matrix. For example, b could be of type matrix and c of type real, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.\nSimilarly,\ncomplex z = 2 + 3i;\nassigns the the complex number \\(2 + 3i\\) to the complex scalar z. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).\n\n\n\nAny variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#declaring-multiple-variables-at-once",
+    "href": "reference-manual/types.html#declaring-multiple-variables-at-once",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "Stan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.\n\n\nThe code:\nreal x, y;\nis equivalent to\nreal x;\nreal y;\nAs a result, all declarations on the same line must be of the same type.\n\n\n\nThe ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in the section introducing compound declarations and definitions :\nreal x = 3, y = 5.6;\nConstrained data types can also be declared together, so long as the constraint for each variable is the same:\nreal&lt;lower=0&gt; x, y;",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/types.html#footnotes",
+    "href": "reference-manual/types.html#footnotes",
+    "title": "Data Types and Declarations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nStan compiles integers to int and reals to double types in C++. Precise details of rounding will depend on the compiler and hardware architecture on which the code is run.↩︎",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
+    ]
+  },
+  {
+    "objectID": "reference-manual/variational.html",
+    "href": "reference-manual/variational.html",
+    "title": "Variational Inference",
+    "section": "",
+    "text": "Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) Kucukelbir et al. (2017). In this chapter, we describe the specifics of how ADVI maximizes the variational objective.\n\n\nADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.\n\n\nADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by elbo_samples. We recommend a default value of \\(100\\), as we only evaluate the ELBO every eval_elbo iterations, which also defaults to \\(100\\).\n\n\n\nADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by grad_samples. We recommend a default value of \\(1\\), as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.\n\n\n\nADVI uses a finite-memory version of adaGrad Duchi, Hazan, and Singer (2011). This has a single parameter that we expose, denoted eta. We now have a warmup adaptation phase that selects a good value for eta. The procedure does a heuristic search over eta values that span 5 orders of magnitude.\n\n\n\nADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by tol_rel_obj, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Variational Inference"
+    ]
+  },
+  {
+    "objectID": "reference-manual/variational.html#stochastic-gradient-ascent",
+    "href": "reference-manual/variational.html#stochastic-gradient-ascent",
+    "title": "Variational Inference",
+    "section": "",
+    "text": "ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.\n\n\nADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by elbo_samples. We recommend a default value of \\(100\\), as we only evaluate the ELBO every eval_elbo iterations, which also defaults to \\(100\\).\n\n\n\nADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by grad_samples. We recommend a default value of \\(1\\), as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.\n\n\n\nADVI uses a finite-memory version of adaGrad Duchi, Hazan, and Singer (2011). This has a single parameter that we expose, denoted eta. We now have a warmup adaptation phase that selects a good value for eta. The procedure does a heuristic search over eta values that span 5 orders of magnitude.\n\n\n\nADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by tol_rel_obj, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.",
+    "crumbs": [
+      "Reference Manual",
+      "Algorithms",
+      "Variational Inference"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/algebraic-equations.html",
+    "href": "stan-users-guide/algebraic-equations.html",
+    "title": "Solving Algebraic Equations",
+    "section": "",
+    "text": "Stan provides a built-in mechanism for specifying systems of algebraic equations. These systems can be solved either with the Newton method, as implemented in the Kinsol package (Hindmarsh et al. 2005), or with the Powell hybrid method (Powell 1970). The function signatures for Stan’s algebraic solvers are fully described in the algebraic solver section of the reference manual.\nSolving any system of algebraic equations can be translated into a root-finding problem, that is, given a function \\(f\\), we wish to find \\(y\\) such that \\(f(y) = 0\\).\n\n\nFor systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.\nAs an illustrative example, we consider the following nonlinear system of two equations with two unknowns: \\[\\begin{align*}\nz_1 &= y_1 - \\theta_1 \\\\\nz_2 &= y_1 y_2 + \\theta_2\n\\end{align*}\\]\nOur goal is to simultaneously solve all equations for \\(y_1\\) and \\(y_2\\), such that the vector \\(z\\) goes to 0.\n\n\n\nA system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the user-defined functions section for more information on coding user-defined functions).\nvector system(vector y,              // unknowns\n              vector theta,          // parameters\n              data array[] real x_r, // data (real)\n              array[] int x_i) {     // data (integer)\n  vector[2] z;\n  z[1] = y[1] - theta[1];\n  z[2] = y[1] * y[2] - theta[2];\n  return z;\n}\nThe function takes the unknowns we wish to solve for in y (a vector), the system parameters in theta (a vector), the real data in x_r (a real array) and the integer data in x_i (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.\nThe body of the system function here could also be coded using a row vector constructor and transposition,\nreturn [ y[1] - theta[1],\n         y[1] * y[2] - theta[2] ]';\nAs systems get more complicated, naming the intermediate expressions goes a long way toward readability.\n\n\nThe function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.\n\n\n\n\nLet’s suppose \\(\\theta = (3, 6)\\). To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.\nThe solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about algebraic solvers with control parameters.\nThe following code returns the solution to our nonlinear algebraic system:\ntransformed data {\n  vector[2] y_guess = [1, 1]';\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\ntransformed parameters {\n  vector[2] theta = [3, 6]';\n  vector[2] y;\n\n  y = solve_newton(system, y_guess, theta, x_r, x_i);\n}\nwhich returns \\(y = (3, -2)\\).\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, y_guess, which may be a data or a parameter vector.\n\n\n\nThe Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function \\(f\\) with respect to the unknowns \\(x\\) must be invertible. This requires the Jacobian to be square, meaning \\(f(y)\\) and \\(y\\) have the same length or, in other words the number of equations in the system is the same as the number of unknowns.\n\n\n\nCertain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.\nOn the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.\n\n\n\n\nThe call to the algebraic solver shown previously uses the default control settings. The _tol variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.\ny = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,\n                     theta, x_r, x_i);\nFor the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.\nThe default values for the control arguments are respectively scaling_step = 1e-3 (\\(10^{-3}\\)), rel_tol = 1e-10 (\\(10^{-10}\\)), f_tol = 1e-6 (\\(10^{-6}\\)), and max_steps = 200 (\\(200\\)).\n\n\nThe relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.\nSmaller relative tolerances produce more accurate solutions but require more computational time.\n\n\nThe tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by Timonen et al. (2023).\n\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Solving Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/algebraic-equations.html#example-system-of-nonlinear-algebraic-equations",
+    "href": "stan-users-guide/algebraic-equations.html#example-system-of-nonlinear-algebraic-equations",
+    "title": "Solving Algebraic Equations",
+    "section": "",
+    "text": "For systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.\nAs an illustrative example, we consider the following nonlinear system of two equations with two unknowns: \\[\\begin{align*}\nz_1 &= y_1 - \\theta_1 \\\\\nz_2 &= y_1 y_2 + \\theta_2\n\\end{align*}\\]\nOur goal is to simultaneously solve all equations for \\(y_1\\) and \\(y_2\\), such that the vector \\(z\\) goes to 0.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Solving Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/algebraic-equations.html#coding-an-algebraic-system",
+    "href": "stan-users-guide/algebraic-equations.html#coding-an-algebraic-system",
+    "title": "Solving Algebraic Equations",
+    "section": "",
+    "text": "A system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the user-defined functions section for more information on coding user-defined functions).\nvector system(vector y,              // unknowns\n              vector theta,          // parameters\n              data array[] real x_r, // data (real)\n              array[] int x_i) {     // data (integer)\n  vector[2] z;\n  z[1] = y[1] - theta[1];\n  z[2] = y[1] * y[2] - theta[2];\n  return z;\n}\nThe function takes the unknowns we wish to solve for in y (a vector), the system parameters in theta (a vector), the real data in x_r (a real array) and the integer data in x_i (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.\nThe body of the system function here could also be coded using a row vector constructor and transposition,\nreturn [ y[1] - theta[1],\n         y[1] * y[2] - theta[2] ]';\nAs systems get more complicated, naming the intermediate expressions goes a long way toward readability.\n\n\nThe function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Solving Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/algebraic-equations.html#calling-the-algebraic-solver",
+    "href": "stan-users-guide/algebraic-equations.html#calling-the-algebraic-solver",
+    "title": "Solving Algebraic Equations",
+    "section": "",
+    "text": "Let’s suppose \\(\\theta = (3, 6)\\). To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.\nThe solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about algebraic solvers with control parameters.\nThe following code returns the solution to our nonlinear algebraic system:\ntransformed data {\n  vector[2] y_guess = [1, 1]';\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\ntransformed parameters {\n  vector[2] theta = [3, 6]';\n  vector[2] y;\n\n  y = solve_newton(system, y_guess, theta, x_r, x_i);\n}\nwhich returns \\(y = (3, -2)\\).\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, y_guess, which may be a data or a parameter vector.\n\n\n\nThe Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function \\(f\\) with respect to the unknowns \\(x\\) must be invertible. This requires the Jacobian to be square, meaning \\(f(y)\\) and \\(y\\) have the same length or, in other words the number of equations in the system is the same as the number of unknowns.\n\n\n\nCertain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.\nOn the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Solving Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/algebraic-equations.html#algebra-control.section",
+    "href": "stan-users-guide/algebraic-equations.html#algebra-control.section",
+    "title": "Solving Algebraic Equations",
+    "section": "",
+    "text": "The call to the algebraic solver shown previously uses the default control settings. The _tol variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.\ny = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,\n                     theta, x_r, x_i);\nFor the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.\nThe default values for the control arguments are respectively scaling_step = 1e-3 (\\(10^{-3}\\)), rel_tol = 1e-10 (\\(10^{-10}\\)), f_tol = 1e-6 (\\(10^{-6}\\)), and max_steps = 200 (\\(200\\)).\n\n\nThe relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.\nSmaller relative tolerances produce more accurate solutions but require more computational time.\n\n\nThe tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by Timonen et al. (2023).\n\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Solving Algebraic Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html",
+    "href": "stan-users-guide/clustering.html",
+    "title": "Clustering Models",
+    "section": "",
+    "text": "Unsupervised methods for organizing data into groups are collectively referred to as clustering. This chapter describes the implementation in Stan of two widely used statistical clustering models, soft \\(K\\)-means and latent Dirichlet allocation (LDA). In addition, this chapter includes naive Bayesian classification, which can be viewed as a form of clustering which may be supervised. These models are typically expressed using discrete parameters for cluster assignments. Nevertheless, they can be implemented in Stan like any other mixture model by marginalizing out the discrete parameters (see the mixture modeling chapter).\n\n\nAs mentioned in the clustering section, clustering models and finite mixture models are really just two sides of the same coin. The “soft” \\(K\\)-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of \\(K\\)-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.\n\n\n\n\\(K\\)-means clustering is a method of clustering data represented as \\(D\\)-dimensional vectors. Specifically, there will be \\(N\\) items to be clustered, each represented as a vector \\(y_n \\in \\mathbb{R}^D\\). In the “soft” version of \\(K\\)-means, the assignments to clusters will be probabilistic.\n\n\n\\(K\\)-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters \\(K\\) and data vectors \\(y\\) as input.\n\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), randomly assign vector \\(y_n\\) to a cluster in \\(\\{1,\\dotsc,K\\}\\);\nRepeat\n\nFor each cluster \\(k\\) in \\(\\{1,\\dotsc,K\\}\\), compute the cluster centroid \\(\\mu_k\\) by averaging the vectors assigned to that cluster;\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), reassign \\(y_n\\) to the cluster \\(k\\) for which the (Euclidean) distance from \\(y_n\\) to \\(\\mu_k\\) is smallest;\nIf no vectors changed cluster, return the cluster assignments.\n\n\nThis algorithm is guaranteed to terminate.\n\n\n\nSoft \\(K\\)-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft \\(K\\)-means can be expressed (and coded in Stan) as a multivariate normal mixture model.\nIn the full generative model, each data point \\(n\\) in \\(\\{1,\\dotsc,N\\}\\) is assigned a cluster \\(z_n \\in \\{1,\\dotsc,K\\}\\) with symmetric uniform probability, \\[\nz_n \\sim \\textsf{categorical}(1/K),\n\\] where \\(1\\) is the unit vector of \\(K\\) dimensions, so that \\(1/K\\) is the symmetric \\(K\\)-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.\nThe data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment \\(z_n\\), \\[\ny_n \\sim  \\textsf{normal}(\\mu_{z[n]},\\Sigma_{z[n]})\n\\]\nThe sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters \\(k\\), \\[\n\\Sigma_k = \\mathrm{diag\\_matrix}({\\bf 1}),\n\\] so that the log multivariate normal can be implemented directly up to a proportion by \\[\n\\mathrm{normal}\\left( y_n \\mid \\mu_k, \\mathrm{diag\\_matrix}({\\bf 1}) \\right)\n\\propto \\exp \\left (- \\frac{1}{2} \\sum_{d=1}^D \\left( \\mu_{k,d} - y_{n,d}\n  \\right)^2 \\right).\n\\] The spatial perspective on \\(K\\)-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean \\(\\mu_k\\) to the data point \\(y_n\\).\n\n\n\nConsider the following Stan program for implementing \\(K\\)-means clustering.\ndata {\n  int&lt;lower=0&gt; N;        // number of data points\n  int&lt;lower=1&gt; D;        // number of dimensions\n  int&lt;lower=1&gt; K;        // number of clusters\n  array[N] vector[D] y;  // observations\n}\ntransformed data {\n  real&lt;upper=0&gt; neg_log_K;\n  neg_log_K = -log(K);\n}\nparameters {\n  array[K] vector[D] mu; // cluster means\n}\ntransformed parameters {\n  array[N, K] real&lt;upper=0&gt; soft_z; // log unnormalized clusters\n  for (n in 1:N) {\n    for (k in 1:K) {\n      soft_z[n, k] = neg_log_K\n                     - 0.5 * dot_self(mu[k] - y[n]);\n    }\n  }\n}\nmodel {\n  // prior\n  for (k in 1:K) {\n    mu[k] ~ std_normal();\n  }\n\n  // likelihood\n  for (n in 1:N) {\n    target += log_sum_exp(soft_z[n]);\n  }\n}\nThere is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.\nThe only parameter is mu, where mu[k] is the centroid for cluster \\(k\\). The transformed parameters soft_z[n] contain the log of the unnormalized cluster assignment probabilities. The vector soft_z[n] can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.\n\n\n\nThe multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., \\(L_2\\) distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on \\(L_1\\) distance (i.e., Manhattan or taxicab distance).\nWithin the multivariate normal version of \\(K\\)-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.\nAlthough there is no global spatial analog, it is common to see soft \\(K\\)-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.\n\n\n\n\nTwo problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the label switching section.\n\n\nCluster assignments are not identified—permuting the cluster mean vectors mu leads to a model with identical likelihoods. For instance, permuting the first two indexes in mu and the first two indexes in each soft_z[n] leads to an identical likelihood (and prior).\nThe lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft \\(K\\)-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.\n\n\n\nThe other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.\nBayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.\nIn light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.\n\n\n\n\nNaive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.1\nMultinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.\nNaive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.\nThe observed data consists of a sequence of \\(M\\) documents made up of bags of words drawn from a vocabulary of \\(V\\) distinct words. A document \\(m\\) has \\(N_m\\) words, which are indexed as \\(w_{m,1}, \\dotsc,\nw_{m,N[m]} \\in \\{1,\\dotsc,V\\}\\). Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) \\(K\\) is fixed.\nThe multinomial mixture model generates a single category \\(z_m \\in\n\\{1,\\dotsc,K\\}\\) for each document \\(m \\in \\{1,\\dotsc,M\\}\\) according to a categorical distribution, \\[\nz_m \\sim \\textsf{categorical}(\\theta).\n\\] The \\(K\\)-simplex parameter \\(\\theta\\) represents the prevalence of each category in the data.\nNext, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word \\(n\\) of document \\(m\\) being generated as \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m]}).\n\\] The parameter \\(\\phi_{z[m]}\\) is a \\(V\\)-simplex representing the probability of each word in the vocabulary in documents of category \\(z_m\\).\nThe parameters \\(\\theta\\) and \\(\\phi\\) are typically given symmetric Dirichlet priors. The prevalence \\(\\theta\\) is sometimes fixed to produce equal probabilities for each category \\(k \\in \\{1,\\dotsc,K\\}\\).\n\n\nThe specification for naive Bayes in the previous sections have used a ragged array notation for the words \\(w\\). Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.\n\\[\n\\begin{array}{lll}\n\\hline\n\\mathrm{n} \\qquad\\qquad\\qquad\\qquad & \\mathrm{w[n]} \\qquad & \\mathrm{doc[n]} \\\\\n\\hline\n1         & w_{1,1}    & 1 \\\\\n2         & w_{1,2}    & 1 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1     & w_{1,N[1]} & 1 \\\\\nN_1 + 1 & w_{2,1}    & 2 \\\\\nN_1 + 2 & w_{2,2}    & 2 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1 + N_2     & w_{2,N[2]} & 2 \\\\\nN_1 + N_2 + 1 & w_{3,1}    & 3 \\\\\n\\vdots       & \\vdots    & \\vdots \\\\\nN = \\sum_{m=1}^M N_m & w_{M,N[M]} & M \\\\\n\\hline\n\\end{array}\n\\]\nThe relevant variables for the program are N, the total number of words in all the documents, the word array w, and the document identity array doc.\n\n\n\nA naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows\ndata {\n  // training data\n  int&lt;lower=1&gt; K;               // num topics\n  int&lt;lower=1&gt; V;               // num words\n  int&lt;lower=0&gt; M;               // num docs\n  int&lt;lower=0&gt; N;               // total word instances\n  array[M] int&lt;lower=1, upper=K&gt; z;    // topic for doc m\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  // hyperparameters\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  simplex[K] theta;             // topic prevalence\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    z[m] ~ categorical(theta);\n  }\n  for (n in 1:N) {\n    w[n] ~ categorical(phi[z[doc[n]]]);\n  }\n}\nThe topic identifiers \\(z_m\\) are declared as data and the latent category assignments are included as part of the likelihood function.\n\n\n\nNaive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number \\(K\\) of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels z. Because z is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability\n\\[\\begin{align*}\n\\log\\, &p(w_{m,1},\\dotsc,w_{m,N_m} \\mid \\theta,\\phi) \\\\\n&= \\log \\sum_{k=1}^K\n    \\left( \\textsf{categorical}(k \\mid \\theta)\n           \\times \\prod_{n=1}^{N_m} \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right) \\\\\n&= \\log \\sum_{k=1}^K \\exp \\left(\n    \\log \\textsf{categorical}(k \\mid \\theta)\n     + \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right).\n\\end{align*}\\]\nThe last step shows how the log_sum_exp function can be used to stabilize the numerical calculation and return a result on the log scale.\nmodel {\n  array[M, K] real gamma;\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    for (k in 1:K) {\n      gamma[m, k] = categorical_lpmf(k | theta);\n    }\n  }\n  for (n in 1:N) {\n    for (k in 1:K) {\n      gamma[doc[n], k] = gamma[doc[n], k]\n                         + categorical_lpmf(w[n] | phi[k]);\n    }\n  }\n  for (m in 1:M) {\n    target += log_sum_exp(gamma[m]);\n  }\n}\nThe local variable gamma[m, k] represents the value \\[\n\\gamma_{m,k} = \\log \\textsf{categorical}(k \\mid \\theta)\n+ \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k).\n\\]\nGiven \\(\\gamma\\), the posterior probability that document \\(m\\) is assigned category \\(k\\) is \\[\n\\Pr[z_m = k \\mid w,\\alpha,\\beta]\n=\n\\exp \\left(\n\\gamma_{m,k}\n- \\log \\sum_{k=1}^K \\exp \\left( \\gamma_{m,k} \\right)\n\\right).\n\\]\nIf the variable gamma were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.\n\n\n\nFull Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see Gelman et al. (2013) and Gelman and Hill (2007) for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.\nTo specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables M2 N2, w2, and doc2. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.\n\n\n\nAn alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of gamma for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.\n\n\n\n\nLatent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model (Blei, Ng, and Jordan 2003) that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.\n\n\nThe basic model assumes each document is generated independently based on fixed hyperparameters. For document \\(m\\), the first step is to draw a topic distribution simplex \\(\\theta_m\\) over the \\(K\\) topics, \\[\n\\theta_m \\sim \\textsf{Dirichlet}(\\alpha).\n\\]\nThe prior hyperparameter \\(\\alpha\\) is fixed to a \\(K\\)-vector of positive values. Each word in the document is generated independently conditional on the distribution \\(\\theta_m\\). First, a topic \\(z_{m,n} \\in \\{1,\\dotsc,K\\}\\) is drawn for the word based on the document-specific topic-distribution, \\[\nz_{m,n} \\sim \\textsf{categorical}(\\theta_m).\n\\]\nFinally, the word \\(w_{m,n}\\) is drawn according to the word distribution for topic \\(z_{m,n}\\), \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m,n]}).\n\\] The distributions \\(\\phi_k\\) over words for topic \\(k\\) are also given a Dirichlet prior, \\[\n\\phi_k \\sim \\textsf{Dirichlet}(\\beta)\n\\]\nwhere \\(\\beta\\) is a fixed \\(V\\)-vector of positive values.\n\n\n\nAlthough Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is \\[\\begin{align*}\np(\\theta,\\phi \\mid w,\\alpha,\\beta)\n&\\propto p(\\theta \\mid \\alpha) \\, p(\\phi \\mid \\beta) \\, p(w \\mid \\theta,\\phi) \\\\\n&= \\prod_{m=1}^M p(\\theta_m \\mid \\alpha)\n    \\times \\prod_{k=1}^K p(\\phi_k \\mid \\beta)\n    \\times \\prod_{m=1}^M \\prod_{n=1}^{M[n]} p(w_{m,n} \\mid \\theta_m,\\phi).\n\\end{align*}\\]\nThe inner word-probability term is defined by summing out the topic assignments, \\[\\begin{align*}\np(w_{m,n} \\mid \\theta_m,\\phi)\n&= \\sum_{z=1}^K p(z,w_{m,n} \\mid \\theta_m,\\phi) \\\\\n&= \\sum_{z=1}^K p(z \\mid \\theta_m) \\, p(w_{m,n} \\mid \\phi_z).\n\\end{align*}\\]\nPlugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, \\[\\begin{align*}\n\\log\\, &p(\\theta,\\phi \\mid w,\\alpha,\\beta) \\\\\n&= \\sum_{m=1}^M \\log \\textsf{Dirichlet}(\\theta_m \\mid \\alpha)\n    + \\sum_{k=1}^K \\log \\textsf{Dirichlet}(\\phi_k \\mid \\beta) \\\\\n&\\qquad + \\sum_{m=1}^M \\sum_{n=1}^{N[m]} \\log \\left(\n    \\sum_{z=1}^K \\textsf{categorical}(z \\mid \\theta_m)\n    \\times \\textsf{categorical}(w_{m,n} \\mid \\phi_z)\n  \\right)\n\\end{align*}\\]\n\n\n\nApplying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.\ndata {\n  int&lt;lower=2&gt; K;               // num topics\n  int&lt;lower=2&gt; V;               // num words\n  int&lt;lower=1&gt; M;               // num docs\n  int&lt;lower=1&gt; N;               // total word instances\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  array[M] simplex[K] theta;    // topic dist for doc m\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  for (m in 1:M) {\n    theta[m] ~ dirichlet(alpha);  // prior\n  }\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);     // prior\n  }\n  for (n in 1:N) {\n    array[K] real gamma;\n    for (k in 1:K) {\n      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);\n    }\n    target += log_sum_exp(gamma);  // likelihood;\n  }\n}\nAs in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.\n\n\n\nTo account for correlations in the distribution of topics for documents, Blei and Lafferty (2007) introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.\nThe authors treat the prior as a fixed hyperparameter. They use an \\(L_1\\)-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.\n\n\nThe Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior alpha in the data declaration with the mean and covariance of the multivariate logistic normal prior.\ndata {\n  // ... data as before without alpha ...\n  vector[K] mu;          // topic mean\n  cov_matrix[K] Sigma;   // topic covariance\n}\nRather than drawing the simplex parameter theta from a Dirichlet, a parameter eta is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.\nparameters {\n  array[K] simplex[V] phi;     // word dist for topic k\n  array[M] vector[K] eta;      // topic dist for doc m\n}\ntransformed parameters {\n  array[M] simplex[K] theta;\n  for (m in 1:M) {\n    theta[m] = softmax(eta[m]);\n  }\n}\nmodel {\n  for (m in 1:M) {\n    eta[m] ~ multi_normal(mu, Sigma);\n  }\n  // ... model as before w/o prior for theta ...\n}\n\n\n\nBy adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean mu and covariance Sigma from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix Sigma may be encoded as follows.\n// ... data block as before, but without alpha ...\nparameters {\n  vector[K] mu;              // topic mean\n  corr_matrix[K] Omega;      // correlation matrix\n  vector&lt;lower=0&gt;[K] sigma;  // scales\n  array[M] vector[K] eta;    // logit topic dist for doc m\n  array[K] simplex[V] phi;   // word dist for topic k\n}\ntransformed parameters {\n  // ... eta as above ...\n  cov_matrix[K] Sigma;       // covariance matrix\n  for (m in 1:K) {\n    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];\n  }\n  for (m in 1:(K-1)) {\n    for (n in (m+1):K) {\n      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];\n      Sigma[n, m] = Sigma[m, n];\n    }\n  }\n}\nmodel {\n  mu ~ normal(0, 5);      // vectorized, diffuse\n  Omega ~ lkj_corr(2.0);  // regularize to unit correlation\n  sigma ~ cauchy(0, 5);   // half-Cauchy due to constraint\n  // ... words sampled as above ...\n}\nThe \\(\\textsf{LKJCorr}\\) distribution with shape \\(\\alpha &gt; 0\\) has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by \\[\n\\mathsf{LkjCorr}(\\Omega\\mid\\alpha) \\propto \\mathrm{det}(\\Omega)^{\\alpha - 1}\n\\] With a scale of \\(\\alpha = 2\\), the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix \\(\\Sigma\\) for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on sigma.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html#relation-to-finite-mixture-models",
+    "href": "stan-users-guide/clustering.html#relation-to-finite-mixture-models",
+    "title": "Clustering Models",
+    "section": "",
+    "text": "As mentioned in the clustering section, clustering models and finite mixture models are really just two sides of the same coin. The “soft” \\(K\\)-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of \\(K\\)-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html#soft-k-means",
+    "href": "stan-users-guide/clustering.html#soft-k-means",
+    "title": "Clustering Models",
+    "section": "",
+    "text": "\\(K\\)-means clustering is a method of clustering data represented as \\(D\\)-dimensional vectors. Specifically, there will be \\(N\\) items to be clustered, each represented as a vector \\(y_n \\in \\mathbb{R}^D\\). In the “soft” version of \\(K\\)-means, the assignments to clusters will be probabilistic.\n\n\n\\(K\\)-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters \\(K\\) and data vectors \\(y\\) as input.\n\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), randomly assign vector \\(y_n\\) to a cluster in \\(\\{1,\\dotsc,K\\}\\);\nRepeat\n\nFor each cluster \\(k\\) in \\(\\{1,\\dotsc,K\\}\\), compute the cluster centroid \\(\\mu_k\\) by averaging the vectors assigned to that cluster;\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), reassign \\(y_n\\) to the cluster \\(k\\) for which the (Euclidean) distance from \\(y_n\\) to \\(\\mu_k\\) is smallest;\nIf no vectors changed cluster, return the cluster assignments.\n\n\nThis algorithm is guaranteed to terminate.\n\n\n\nSoft \\(K\\)-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft \\(K\\)-means can be expressed (and coded in Stan) as a multivariate normal mixture model.\nIn the full generative model, each data point \\(n\\) in \\(\\{1,\\dotsc,N\\}\\) is assigned a cluster \\(z_n \\in \\{1,\\dotsc,K\\}\\) with symmetric uniform probability, \\[\nz_n \\sim \\textsf{categorical}(1/K),\n\\] where \\(1\\) is the unit vector of \\(K\\) dimensions, so that \\(1/K\\) is the symmetric \\(K\\)-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.\nThe data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment \\(z_n\\), \\[\ny_n \\sim  \\textsf{normal}(\\mu_{z[n]},\\Sigma_{z[n]})\n\\]\nThe sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters \\(k\\), \\[\n\\Sigma_k = \\mathrm{diag\\_matrix}({\\bf 1}),\n\\] so that the log multivariate normal can be implemented directly up to a proportion by \\[\n\\mathrm{normal}\\left( y_n \\mid \\mu_k, \\mathrm{diag\\_matrix}({\\bf 1}) \\right)\n\\propto \\exp \\left (- \\frac{1}{2} \\sum_{d=1}^D \\left( \\mu_{k,d} - y_{n,d}\n  \\right)^2 \\right).\n\\] The spatial perspective on \\(K\\)-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean \\(\\mu_k\\) to the data point \\(y_n\\).\n\n\n\nConsider the following Stan program for implementing \\(K\\)-means clustering.\ndata {\n  int&lt;lower=0&gt; N;        // number of data points\n  int&lt;lower=1&gt; D;        // number of dimensions\n  int&lt;lower=1&gt; K;        // number of clusters\n  array[N] vector[D] y;  // observations\n}\ntransformed data {\n  real&lt;upper=0&gt; neg_log_K;\n  neg_log_K = -log(K);\n}\nparameters {\n  array[K] vector[D] mu; // cluster means\n}\ntransformed parameters {\n  array[N, K] real&lt;upper=0&gt; soft_z; // log unnormalized clusters\n  for (n in 1:N) {\n    for (k in 1:K) {\n      soft_z[n, k] = neg_log_K\n                     - 0.5 * dot_self(mu[k] - y[n]);\n    }\n  }\n}\nmodel {\n  // prior\n  for (k in 1:K) {\n    mu[k] ~ std_normal();\n  }\n\n  // likelihood\n  for (n in 1:N) {\n    target += log_sum_exp(soft_z[n]);\n  }\n}\nThere is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.\nThe only parameter is mu, where mu[k] is the centroid for cluster \\(k\\). The transformed parameters soft_z[n] contain the log of the unnormalized cluster assignment probabilities. The vector soft_z[n] can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.\n\n\n\nThe multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., \\(L_2\\) distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on \\(L_1\\) distance (i.e., Manhattan or taxicab distance).\nWithin the multivariate normal version of \\(K\\)-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.\nAlthough there is no global spatial analog, it is common to see soft \\(K\\)-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html#the-difficulty-of-bayesian-inference-for-clustering",
+    "href": "stan-users-guide/clustering.html#the-difficulty-of-bayesian-inference-for-clustering",
+    "title": "Clustering Models",
+    "section": "",
+    "text": "Two problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the label switching section.\n\n\nCluster assignments are not identified—permuting the cluster mean vectors mu leads to a model with identical likelihoods. For instance, permuting the first two indexes in mu and the first two indexes in each soft_z[n] leads to an identical likelihood (and prior).\nThe lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft \\(K\\)-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.\n\n\n\nThe other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.\nBayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.\nIn light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html#naive-bayes-classification-and-clustering",
+    "href": "stan-users-guide/clustering.html#naive-bayes-classification-and-clustering",
+    "title": "Clustering Models",
+    "section": "",
+    "text": "Naive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.1\nMultinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.\nNaive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.\nThe observed data consists of a sequence of \\(M\\) documents made up of bags of words drawn from a vocabulary of \\(V\\) distinct words. A document \\(m\\) has \\(N_m\\) words, which are indexed as \\(w_{m,1}, \\dotsc,\nw_{m,N[m]} \\in \\{1,\\dotsc,V\\}\\). Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) \\(K\\) is fixed.\nThe multinomial mixture model generates a single category \\(z_m \\in\n\\{1,\\dotsc,K\\}\\) for each document \\(m \\in \\{1,\\dotsc,M\\}\\) according to a categorical distribution, \\[\nz_m \\sim \\textsf{categorical}(\\theta).\n\\] The \\(K\\)-simplex parameter \\(\\theta\\) represents the prevalence of each category in the data.\nNext, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word \\(n\\) of document \\(m\\) being generated as \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m]}).\n\\] The parameter \\(\\phi_{z[m]}\\) is a \\(V\\)-simplex representing the probability of each word in the vocabulary in documents of category \\(z_m\\).\nThe parameters \\(\\theta\\) and \\(\\phi\\) are typically given symmetric Dirichlet priors. The prevalence \\(\\theta\\) is sometimes fixed to produce equal probabilities for each category \\(k \\in \\{1,\\dotsc,K\\}\\).\n\n\nThe specification for naive Bayes in the previous sections have used a ragged array notation for the words \\(w\\). Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.\n\\[\n\\begin{array}{lll}\n\\hline\n\\mathrm{n} \\qquad\\qquad\\qquad\\qquad & \\mathrm{w[n]} \\qquad & \\mathrm{doc[n]} \\\\\n\\hline\n1         & w_{1,1}    & 1 \\\\\n2         & w_{1,2}    & 1 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1     & w_{1,N[1]} & 1 \\\\\nN_1 + 1 & w_{2,1}    & 2 \\\\\nN_1 + 2 & w_{2,2}    & 2 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1 + N_2     & w_{2,N[2]} & 2 \\\\\nN_1 + N_2 + 1 & w_{3,1}    & 3 \\\\\n\\vdots       & \\vdots    & \\vdots \\\\\nN = \\sum_{m=1}^M N_m & w_{M,N[M]} & M \\\\\n\\hline\n\\end{array}\n\\]\nThe relevant variables for the program are N, the total number of words in all the documents, the word array w, and the document identity array doc.\n\n\n\nA naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows\ndata {\n  // training data\n  int&lt;lower=1&gt; K;               // num topics\n  int&lt;lower=1&gt; V;               // num words\n  int&lt;lower=0&gt; M;               // num docs\n  int&lt;lower=0&gt; N;               // total word instances\n  array[M] int&lt;lower=1, upper=K&gt; z;    // topic for doc m\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  // hyperparameters\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  simplex[K] theta;             // topic prevalence\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    z[m] ~ categorical(theta);\n  }\n  for (n in 1:N) {\n    w[n] ~ categorical(phi[z[doc[n]]]);\n  }\n}\nThe topic identifiers \\(z_m\\) are declared as data and the latent category assignments are included as part of the likelihood function.\n\n\n\nNaive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number \\(K\\) of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels z. Because z is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability\n\\[\\begin{align*}\n\\log\\, &p(w_{m,1},\\dotsc,w_{m,N_m} \\mid \\theta,\\phi) \\\\\n&= \\log \\sum_{k=1}^K\n    \\left( \\textsf{categorical}(k \\mid \\theta)\n           \\times \\prod_{n=1}^{N_m} \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right) \\\\\n&= \\log \\sum_{k=1}^K \\exp \\left(\n    \\log \\textsf{categorical}(k \\mid \\theta)\n     + \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right).\n\\end{align*}\\]\nThe last step shows how the log_sum_exp function can be used to stabilize the numerical calculation and return a result on the log scale.\nmodel {\n  array[M, K] real gamma;\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    for (k in 1:K) {\n      gamma[m, k] = categorical_lpmf(k | theta);\n    }\n  }\n  for (n in 1:N) {\n    for (k in 1:K) {\n      gamma[doc[n], k] = gamma[doc[n], k]\n                         + categorical_lpmf(w[n] | phi[k]);\n    }\n  }\n  for (m in 1:M) {\n    target += log_sum_exp(gamma[m]);\n  }\n}\nThe local variable gamma[m, k] represents the value \\[\n\\gamma_{m,k} = \\log \\textsf{categorical}(k \\mid \\theta)\n+ \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k).\n\\]\nGiven \\(\\gamma\\), the posterior probability that document \\(m\\) is assigned category \\(k\\) is \\[\n\\Pr[z_m = k \\mid w,\\alpha,\\beta]\n=\n\\exp \\left(\n\\gamma_{m,k}\n- \\log \\sum_{k=1}^K \\exp \\left( \\gamma_{m,k} \\right)\n\\right).\n\\]\nIf the variable gamma were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.\n\n\n\nFull Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see Gelman et al. (2013) and Gelman and Hill (2007) for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.\nTo specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables M2 N2, w2, and doc2. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.\n\n\n\nAn alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of gamma for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html#latent-dirichlet-allocation",
+    "href": "stan-users-guide/clustering.html#latent-dirichlet-allocation",
+    "title": "Clustering Models",
+    "section": "",
+    "text": "Latent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model (Blei, Ng, and Jordan 2003) that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.\n\n\nThe basic model assumes each document is generated independently based on fixed hyperparameters. For document \\(m\\), the first step is to draw a topic distribution simplex \\(\\theta_m\\) over the \\(K\\) topics, \\[\n\\theta_m \\sim \\textsf{Dirichlet}(\\alpha).\n\\]\nThe prior hyperparameter \\(\\alpha\\) is fixed to a \\(K\\)-vector of positive values. Each word in the document is generated independently conditional on the distribution \\(\\theta_m\\). First, a topic \\(z_{m,n} \\in \\{1,\\dotsc,K\\}\\) is drawn for the word based on the document-specific topic-distribution, \\[\nz_{m,n} \\sim \\textsf{categorical}(\\theta_m).\n\\]\nFinally, the word \\(w_{m,n}\\) is drawn according to the word distribution for topic \\(z_{m,n}\\), \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m,n]}).\n\\] The distributions \\(\\phi_k\\) over words for topic \\(k\\) are also given a Dirichlet prior, \\[\n\\phi_k \\sim \\textsf{Dirichlet}(\\beta)\n\\]\nwhere \\(\\beta\\) is a fixed \\(V\\)-vector of positive values.\n\n\n\nAlthough Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is \\[\\begin{align*}\np(\\theta,\\phi \\mid w,\\alpha,\\beta)\n&\\propto p(\\theta \\mid \\alpha) \\, p(\\phi \\mid \\beta) \\, p(w \\mid \\theta,\\phi) \\\\\n&= \\prod_{m=1}^M p(\\theta_m \\mid \\alpha)\n    \\times \\prod_{k=1}^K p(\\phi_k \\mid \\beta)\n    \\times \\prod_{m=1}^M \\prod_{n=1}^{M[n]} p(w_{m,n} \\mid \\theta_m,\\phi).\n\\end{align*}\\]\nThe inner word-probability term is defined by summing out the topic assignments, \\[\\begin{align*}\np(w_{m,n} \\mid \\theta_m,\\phi)\n&= \\sum_{z=1}^K p(z,w_{m,n} \\mid \\theta_m,\\phi) \\\\\n&= \\sum_{z=1}^K p(z \\mid \\theta_m) \\, p(w_{m,n} \\mid \\phi_z).\n\\end{align*}\\]\nPlugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, \\[\\begin{align*}\n\\log\\, &p(\\theta,\\phi \\mid w,\\alpha,\\beta) \\\\\n&= \\sum_{m=1}^M \\log \\textsf{Dirichlet}(\\theta_m \\mid \\alpha)\n    + \\sum_{k=1}^K \\log \\textsf{Dirichlet}(\\phi_k \\mid \\beta) \\\\\n&\\qquad + \\sum_{m=1}^M \\sum_{n=1}^{N[m]} \\log \\left(\n    \\sum_{z=1}^K \\textsf{categorical}(z \\mid \\theta_m)\n    \\times \\textsf{categorical}(w_{m,n} \\mid \\phi_z)\n  \\right)\n\\end{align*}\\]\n\n\n\nApplying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.\ndata {\n  int&lt;lower=2&gt; K;               // num topics\n  int&lt;lower=2&gt; V;               // num words\n  int&lt;lower=1&gt; M;               // num docs\n  int&lt;lower=1&gt; N;               // total word instances\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  array[M] simplex[K] theta;    // topic dist for doc m\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  for (m in 1:M) {\n    theta[m] ~ dirichlet(alpha);  // prior\n  }\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);     // prior\n  }\n  for (n in 1:N) {\n    array[K] real gamma;\n    for (k in 1:K) {\n      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);\n    }\n    target += log_sum_exp(gamma);  // likelihood;\n  }\n}\nAs in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.\n\n\n\nTo account for correlations in the distribution of topics for documents, Blei and Lafferty (2007) introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.\nThe authors treat the prior as a fixed hyperparameter. They use an \\(L_1\\)-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.\n\n\nThe Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior alpha in the data declaration with the mean and covariance of the multivariate logistic normal prior.\ndata {\n  // ... data as before without alpha ...\n  vector[K] mu;          // topic mean\n  cov_matrix[K] Sigma;   // topic covariance\n}\nRather than drawing the simplex parameter theta from a Dirichlet, a parameter eta is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.\nparameters {\n  array[K] simplex[V] phi;     // word dist for topic k\n  array[M] vector[K] eta;      // topic dist for doc m\n}\ntransformed parameters {\n  array[M] simplex[K] theta;\n  for (m in 1:M) {\n    theta[m] = softmax(eta[m]);\n  }\n}\nmodel {\n  for (m in 1:M) {\n    eta[m] ~ multi_normal(mu, Sigma);\n  }\n  // ... model as before w/o prior for theta ...\n}\n\n\n\nBy adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean mu and covariance Sigma from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix Sigma may be encoded as follows.\n// ... data block as before, but without alpha ...\nparameters {\n  vector[K] mu;              // topic mean\n  corr_matrix[K] Omega;      // correlation matrix\n  vector&lt;lower=0&gt;[K] sigma;  // scales\n  array[M] vector[K] eta;    // logit topic dist for doc m\n  array[K] simplex[V] phi;   // word dist for topic k\n}\ntransformed parameters {\n  // ... eta as above ...\n  cov_matrix[K] Sigma;       // covariance matrix\n  for (m in 1:K) {\n    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];\n  }\n  for (m in 1:(K-1)) {\n    for (n in (m+1):K) {\n      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];\n      Sigma[n, m] = Sigma[m, n];\n    }\n  }\n}\nmodel {\n  mu ~ normal(0, 5);      // vectorized, diffuse\n  Omega ~ lkj_corr(2.0);  // regularize to unit correlation\n  sigma ~ cauchy(0, 5);   // half-Cauchy due to constraint\n  // ... words sampled as above ...\n}\nThe \\(\\textsf{LKJCorr}\\) distribution with shape \\(\\alpha &gt; 0\\) has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by \\[\n\\mathsf{LkjCorr}(\\Omega\\mid\\alpha) \\propto \\mathrm{det}(\\Omega)^{\\alpha - 1}\n\\] With a scale of \\(\\alpha = 2\\), the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix \\(\\Sigma\\) for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on sigma.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/clustering.html#footnotes",
+    "href": "stan-users-guide/clustering.html#footnotes",
+    "title": "Clustering Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nFor clustering, the non-identifiability problems for all mixture models present a problem, whereas there is no such problem for classification. Despite the difficulties with full Bayesian inference for clustering, researchers continue to use it, often in an exploratory data analysis setting rather than for predictive modeling.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Clustering Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html",
+    "href": "stan-users-guide/copulas.html",
+    "title": "Copulas",
+    "section": "",
+    "text": "Copulas provide a flexible way to model multivariate distributions by separating the marginal cumulative distribution functions from the dependence structure. This chapter introduces copulas in Stan, focusing on implementation techniques and practical examples. This chapter was derived from Brynjólfur Gauti Guðrúnar Jónsson’s A gentle introduction: the Gaussian copula.\n\n\nAccording to Sklar’s theorem (Sklar 1959), any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.\nFor a multivariate random variable \\(\\mathbf{X} = [X_1 \\cdots X_D]^\\top\\) with marginal cumulative distribution functions \\(F_i\\), the joint cumulative distribution function can be written as:\n\\[\nF_{\\mathbf{X}}(\\mathbf{x}) = C(F_1(x_1), \\ldots, F_D(x_D)) = \\Pr[X_1 \\leq x_1, \\ldots, X_D \\leq x_D]\n\\]\nwhere \\(C\\) is the copula function, \\(F_{\\mathbf{X}}\\) is the joint cumulative distribution function, and \\(F_i\\) are the marginal cumulative distribution functions. The copula function \\(C\\) must be a joint cumulative distribution function over the unit hypercube \\([0, 1]^D\\).\n\n\n\nThis section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.\nThe log density of a multivariate distribution using a copula can be written as:\n\\[\n\\log h(\\mathbf{x}) = \\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right) + \\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\n\\]\nwhere:\n\n\\(u_i = F_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) are the probability integral transforms of the data\n\\(\\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right)\\) is the log density of the copula\n\\(\\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) is the sum of the log densities of the marginals\n\\(\\boldsymbol{\\alpha}\\) represents the parameters describing the parametric form of the copula\n\\(\\boldsymbol{\\beta}_i\\) represents the parameters describing the parametric form of the \\(i\\)-th marginal distribution\n\nThe implementation of copulas in Stan has two key requirements:\n\nBoth the probability density functions and cumulative distribution functions of the marginal distributions must be available\nA function that computes the log density of the copula for the transformed data must be implemented\n\nMost copula implementations in Stan follow a three-step process:\n\nAccumulate marginal log likelihoods: Calculate and add the log density of each marginal distribution to the target log density\nTransform to uniform variables: Apply the marginal CDFs to transform the data to uniform variables on the unit interval\nCalculate copula density: Compute the log density of the copula based on these uniform variables and add it to the target log density\n\nThis process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.\nIn a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where \\(\\log c(\\mathbf{u}) = 0\\), resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\nMost parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).\n\n\n\nThe Gaussian copula is constructed using the multivariate normal distribution. For a \\(D\\)-dimensional random vector \\(\\mathbf{X}\\) with marginals \\(F_i\\), the log Gaussian copula density is given by:\n\\[\n\\begin{aligned}\n\\log c(\\mathbf{u}) &=\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top (\\boldsymbol{\\Omega}^{-1} - \\mathbf{I}) \\mathbf{z} \\\\\n& =\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top \\boldsymbol{\\Omega}^{-1} \\mathbf{z} + \\frac{1}{2} \\mathbf{z}^\\top \\mathbf{z} \\\\\n&= \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\boldsymbol{\\Omega}) - \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\mathbf{I})\n\\end{aligned}\n\\]\nwhere \\(\\mathbf{z} = [\\Phi^{-1}(u_1), \\ldots, \\Phi^{-1}(u_D)]^\\top\\) are the inverse normal CDF transforms of the uniform marginals, \\(\\boldsymbol{\\Omega}\\) is the correlation matrix, and \\(\\mathbf{I}\\) is the identity matrix. The joint log density is then:\n\\[\n\\log h(\\mathbf{x}) = \\log c(F_1(x_1), \\ldots, F_D(x_D)) + \\sum_{i=1}^D \\log f_i(x_i)\n\\]\nFollowing the three-step process for implementing copulas in Stan:\n\nAccumulate marginal log likelihoods: The exponential log densities are added to the target in the line target += exponential_lpdf(y[n] | lambda)\nTransform to uniform variables: The exponential CDF transforms the data to uniform variables: exponential_cdf(y[n, d] | lambda[d])\nCalculate copula density: The transformed variables are converted to normal scale using inv_Phi and the multivariate normal log density is computed: z ~ multi_normal_cholesky(zeros, L_Omega)\n\nThe following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  int&lt;lower=0&gt; D;  // number of dimensions\n  vector&lt;lower=0&gt;[D] y[N];  // data\n}\n\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\n\nparameters {\n  // Parameters for exponential marginal distributions\n  vector&lt;lower=0&gt;[D] lambda;  // rate parameters\n  \n  // Correlation matrix for Gaussian copula\n  cholesky_factor_corr[D] L_Omega;\n}\n\nmodel {\n  // Priors\n  lambda ~ gamma(2, 1);  // prior for rate parameters\n  L_Omega ~ lkj_corr_cholesky(2);\n  \n  // Likelihood using Gaussian copula with exponential marginals\n  for (n in 1:N) {\n    // Add exponential log density to target\n    target += exponential_lpdf(y[n] | lambda);\n    \n    vector[D] z;\n    for (d in 1:D) {\n      // Transform to uniform using exponential CDF\n      real u_d = exponential_cdf(y[n, d] | lambda[d]);\n      \n      // Transform to standard normal\n      z[d] = inv_Phi(u_d);\n    }\n    // Multivariate normal log density with correlation matrix\n    z ~ multi_normal_cholesky(zeros, L_Omega);\n  }\n}\n\ngenerated quantities {\n  // Optional: Recover correlation matrix from Cholesky factor\n  matrix[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\n\n\n\nCopulas offer several advantages in statistical modeling:\n\nFlexibility: They allow combining any marginal distributions with various dependence structures. For example:\n\nModeling financial returns with heavy-tailed marginals and complex dependence structures\nCombining different types of distributions (e.g., normal and gamma) in a single model\nCapturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic\nModeling different types of tail dependence in different parts of the distribution\n\nFactorability: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.\nFor example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.\nTail dependence: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.\nUniversal Framework: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\n\n\n\n\nWhen implementing copulas in Stan, several considerations should be kept in mind:\n\nComputational efficiency: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.\nParameter identifiability: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.\nModel selection: The choice of copula family should be guided by the specific dependence structure of the data. For example:\n\nThe Gaussian copula may underestimate the probability of joint extreme events in financial data\nThe Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications\nArchimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions\n\nNumerical stability: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.\nSymmetry considerations: Many copula families exhibit strong symmetries that may not match the data:\n\nRadial symmetry: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns\nExchangeability: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.\n\nTail dependence: Understanding and choosing appropriate tail dependence is crucial:\n\nThe upper (lower) tail dependence coefficient \\(\\lambda_U (\\lambda_L)\\) is the probability that one variable is extremely large (small) given that another is extremely large (small).\nDifferent copula families exhibit different tail dependence properties:\n\nSome copulas (like Gaussian) have zero tail dependence\nOthers can model symmetric tail dependence (\\(\\lambda_U = \\lambda_L\\))\nSome can capture asymmetric tail dependence (\\(\\lambda_U \\neq \\lambda_L\\))\nCertain copulas allow for tail dependence even with zero correlation\n\nThe choice of copula should be guided by the expected tail behavior in the application:\n\nFinancial data often requires modeling joint lower extreme events\nRisk management applications may need asymmetric tail dependence\nSome applications may require different tail behavior in different parts of the distribution\n\n\nHigh-dimensional modeling: As dimensionality increases:\n\nThe number of dependence parameters grows\nSome copula families become less flexible\nVine copulas or factor copulas may be more appropriate\n\n\n\n\n\nSeveral copula families are available for modeling different dependence structures in the correlation component:\n\nGaussian copula: Based on the multivariate normal distribution, offering symmetric dependence\nStudent-t copula: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula\nArchimedean copulas: A class of copulas defined through generator functions, including:\n\nClayton copula: Stronger lower tail dependence\nGumbel copula: Stronger upper tail dependence\nFrank copula: Symmetric dependence\n\nVine copulas: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas\n\n\n\n\n\nJónsson’s three part blog series, Copulas in Stan:\n\nPart I: If it bleeds, we can kill it\nPart II: A gentle introduction: the Gaussian copula\nPart III: It was the best of tails, it was the worst of tails: The T-Copula\n\nBrynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, Copulas in Stan: Modeling Spatial Dependence\nSean Pinkney’s Helpful Stan functions: copula functions",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#what-are-copulas",
+    "href": "stan-users-guide/copulas.html#what-are-copulas",
+    "title": "Copulas",
+    "section": "",
+    "text": "According to Sklar’s theorem (Sklar 1959), any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.\nFor a multivariate random variable \\(\\mathbf{X} = [X_1 \\cdots X_D]^\\top\\) with marginal cumulative distribution functions \\(F_i\\), the joint cumulative distribution function can be written as:\n\\[\nF_{\\mathbf{X}}(\\mathbf{x}) = C(F_1(x_1), \\ldots, F_D(x_D)) = \\Pr[X_1 \\leq x_1, \\ldots, X_D \\leq x_D]\n\\]\nwhere \\(C\\) is the copula function, \\(F_{\\mathbf{X}}\\) is the joint cumulative distribution function, and \\(F_i\\) are the marginal cumulative distribution functions. The copula function \\(C\\) must be a joint cumulative distribution function over the unit hypercube \\([0, 1]^D\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#general-structure-of-copula-models-in-stan",
+    "href": "stan-users-guide/copulas.html#general-structure-of-copula-models-in-stan",
+    "title": "Copulas",
+    "section": "",
+    "text": "This section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.\nThe log density of a multivariate distribution using a copula can be written as:\n\\[\n\\log h(\\mathbf{x}) = \\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right) + \\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\n\\]\nwhere:\n\n\\(u_i = F_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) are the probability integral transforms of the data\n\\(\\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right)\\) is the log density of the copula\n\\(\\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) is the sum of the log densities of the marginals\n\\(\\boldsymbol{\\alpha}\\) represents the parameters describing the parametric form of the copula\n\\(\\boldsymbol{\\beta}_i\\) represents the parameters describing the parametric form of the \\(i\\)-th marginal distribution\n\nThe implementation of copulas in Stan has two key requirements:\n\nBoth the probability density functions and cumulative distribution functions of the marginal distributions must be available\nA function that computes the log density of the copula for the transformed data must be implemented\n\nMost copula implementations in Stan follow a three-step process:\n\nAccumulate marginal log likelihoods: Calculate and add the log density of each marginal distribution to the target log density\nTransform to uniform variables: Apply the marginal CDFs to transform the data to uniform variables on the unit interval\nCalculate copula density: Compute the log density of the copula based on these uniform variables and add it to the target log density\n\nThis process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.\nIn a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where \\(\\log c(\\mathbf{u}) = 0\\), resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\nMost parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#gaussian-copula-example",
+    "href": "stan-users-guide/copulas.html#gaussian-copula-example",
+    "title": "Copulas",
+    "section": "",
+    "text": "The Gaussian copula is constructed using the multivariate normal distribution. For a \\(D\\)-dimensional random vector \\(\\mathbf{X}\\) with marginals \\(F_i\\), the log Gaussian copula density is given by:\n\\[\n\\begin{aligned}\n\\log c(\\mathbf{u}) &=\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top (\\boldsymbol{\\Omega}^{-1} - \\mathbf{I}) \\mathbf{z} \\\\\n& =\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top \\boldsymbol{\\Omega}^{-1} \\mathbf{z} + \\frac{1}{2} \\mathbf{z}^\\top \\mathbf{z} \\\\\n&= \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\boldsymbol{\\Omega}) - \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\mathbf{I})\n\\end{aligned}\n\\]\nwhere \\(\\mathbf{z} = [\\Phi^{-1}(u_1), \\ldots, \\Phi^{-1}(u_D)]^\\top\\) are the inverse normal CDF transforms of the uniform marginals, \\(\\boldsymbol{\\Omega}\\) is the correlation matrix, and \\(\\mathbf{I}\\) is the identity matrix. The joint log density is then:\n\\[\n\\log h(\\mathbf{x}) = \\log c(F_1(x_1), \\ldots, F_D(x_D)) + \\sum_{i=1}^D \\log f_i(x_i)\n\\]\nFollowing the three-step process for implementing copulas in Stan:\n\nAccumulate marginal log likelihoods: The exponential log densities are added to the target in the line target += exponential_lpdf(y[n] | lambda)\nTransform to uniform variables: The exponential CDF transforms the data to uniform variables: exponential_cdf(y[n, d] | lambda[d])\nCalculate copula density: The transformed variables are converted to normal scale using inv_Phi and the multivariate normal log density is computed: z ~ multi_normal_cholesky(zeros, L_Omega)\n\nThe following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  int&lt;lower=0&gt; D;  // number of dimensions\n  vector&lt;lower=0&gt;[D] y[N];  // data\n}\n\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\n\nparameters {\n  // Parameters for exponential marginal distributions\n  vector&lt;lower=0&gt;[D] lambda;  // rate parameters\n  \n  // Correlation matrix for Gaussian copula\n  cholesky_factor_corr[D] L_Omega;\n}\n\nmodel {\n  // Priors\n  lambda ~ gamma(2, 1);  // prior for rate parameters\n  L_Omega ~ lkj_corr_cholesky(2);\n  \n  // Likelihood using Gaussian copula with exponential marginals\n  for (n in 1:N) {\n    // Add exponential log density to target\n    target += exponential_lpdf(y[n] | lambda);\n    \n    vector[D] z;\n    for (d in 1:D) {\n      // Transform to uniform using exponential CDF\n      real u_d = exponential_cdf(y[n, d] | lambda[d]);\n      \n      // Transform to standard normal\n      z[d] = inv_Phi(u_d);\n    }\n    // Multivariate normal log density with correlation matrix\n    z ~ multi_normal_cholesky(zeros, L_Omega);\n  }\n}\n\ngenerated quantities {\n  // Optional: Recover correlation matrix from Cholesky factor\n  matrix[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#advantages-of-copulas",
+    "href": "stan-users-guide/copulas.html#advantages-of-copulas",
+    "title": "Copulas",
+    "section": "",
+    "text": "Copulas offer several advantages in statistical modeling:\n\nFlexibility: They allow combining any marginal distributions with various dependence structures. For example:\n\nModeling financial returns with heavy-tailed marginals and complex dependence structures\nCombining different types of distributions (e.g., normal and gamma) in a single model\nCapturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic\nModeling different types of tail dependence in different parts of the distribution\n\nFactorability: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.\nFor example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.\nTail dependence: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.\nUniversal Framework: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#common-pitfalls-and-considerations",
+    "href": "stan-users-guide/copulas.html#common-pitfalls-and-considerations",
+    "title": "Copulas",
+    "section": "",
+    "text": "When implementing copulas in Stan, several considerations should be kept in mind:\n\nComputational efficiency: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.\nParameter identifiability: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.\nModel selection: The choice of copula family should be guided by the specific dependence structure of the data. For example:\n\nThe Gaussian copula may underestimate the probability of joint extreme events in financial data\nThe Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications\nArchimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions\n\nNumerical stability: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.\nSymmetry considerations: Many copula families exhibit strong symmetries that may not match the data:\n\nRadial symmetry: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns\nExchangeability: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.\n\nTail dependence: Understanding and choosing appropriate tail dependence is crucial:\n\nThe upper (lower) tail dependence coefficient \\(\\lambda_U (\\lambda_L)\\) is the probability that one variable is extremely large (small) given that another is extremely large (small).\nDifferent copula families exhibit different tail dependence properties:\n\nSome copulas (like Gaussian) have zero tail dependence\nOthers can model symmetric tail dependence (\\(\\lambda_U = \\lambda_L\\))\nSome can capture asymmetric tail dependence (\\(\\lambda_U \\neq \\lambda_L\\))\nCertain copulas allow for tail dependence even with zero correlation\n\nThe choice of copula should be guided by the expected tail behavior in the application:\n\nFinancial data often requires modeling joint lower extreme events\nRisk management applications may need asymmetric tail dependence\nSome applications may require different tail behavior in different parts of the distribution\n\n\nHigh-dimensional modeling: As dimensionality increases:\n\nThe number of dependence parameters grows\nSome copula families become less flexible\nVine copulas or factor copulas may be more appropriate",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#common-copula-families",
+    "href": "stan-users-guide/copulas.html#common-copula-families",
+    "title": "Copulas",
+    "section": "",
+    "text": "Several copula families are available for modeling different dependence structures in the correlation component:\n\nGaussian copula: Based on the multivariate normal distribution, offering symmetric dependence\nStudent-t copula: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula\nArchimedean copulas: A class of copulas defined through generator functions, including:\n\nClayton copula: Stronger lower tail dependence\nGumbel copula: Stronger upper tail dependence\nFrank copula: Symmetric dependence\n\nVine copulas: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/copulas.html#further-readingviewing",
+    "href": "stan-users-guide/copulas.html#further-readingviewing",
+    "title": "Copulas",
+    "section": "",
+    "text": "Jónsson’s three part blog series, Copulas in Stan:\n\nPart I: If it bleeds, we can kill it\nPart II: A gentle introduction: the Gaussian copula\nPart III: It was the best of tails, it was the worst of tails: The T-Copula\n\nBrynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, Copulas in Stan: Modeling Spatial Dependence\nSean Pinkney’s Helpful Stan functions: copula functions",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Copulas"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/custom-probability.html",
+    "href": "stan-users-guide/custom-probability.html",
+    "title": "Custom Probability Functions",
+    "section": "",
+    "text": "Custom distributions may also be implemented directly within Stan’s programming language. The only thing that is needed is to increment the total log probability. The rest of the chapter provides examples.\n\n\n\n\nA simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}\\) are the bounds, with \\(\\alpha &lt; \\beta\\), then \\(y \\in (\\alpha,\\beta)\\) has a density defined as \\[\n\\textsf{triangle}(y \\mid \\alpha,\\beta)\n=\n\\frac{1}{(\\beta - \\alpha)^2}\n\\cdot\n\\textrm{min}(y - \\alpha, \\beta - y).\n\\]\nThe general form of triangle can be coded in Stan as follows.\ndata {\n  real alpha;\n  real&lt;lower=alpha&gt; beta;\n}\nparameters {\n  real&lt;lower=alpha, upper=beta&gt; y;\n}\nmodel {\n  target += -2 * log(beta - alpha)\n    + log(fmin(y - alpha, beta - y));\n}\nBecause the bounds are specified as data here, the term -2 * log(beta - alpha) could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.\nIf \\(\\alpha = -1\\), \\(\\beta = 1\\), and \\(y \\in (-1,1)\\), then fmin(y - alpha, beta - y) is fmin(y + 1, -1 - y), which is fmin(y + 1, -(y + 1)), which reduces to 1 - abs(y). Therfore, the density, dropping constants, reduces to \\[\n\\textsf{triangle}(y \\mid -1, 1) \\propto 1 - |y|.\n\\] Consider the following Stan implementation of \\(\\textsf{triangle}(-1,1)\\) for sampling.\nparameters {\n  real&lt;lower=-1, upper=1&gt; y;\n}\nmodel {\n  target += log1m(abs(y));\n}\nThe single scalar parameter y is declared as lying in the interval (-1, 1). The total log probability is incremented with the joint log probability of all parameters, i.e., \\(\\log \\mathsf{Triangle}(y \\mid -1, 1)\\). This value is coded in Stan as log1m(abs(y)). The function log1m is defined so that log1m(x) has the same value as \\(\\log(1 - x)\\), but the computation is faster, more accurate, and more stable.\nThe constrained type real&lt;lower=-1, upper=1&gt; declared for y is critical for correct sampling behavior. If the constraint on y is removed from the program, say by declaring y as having the unconstrained scalar type real, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of y outside of \\((-1,1)\\).\nNow suppose the log probability function were extended to all of \\(\\mathbb{R}\\) as follows by defining the probability to be log(0.0), i.e., \\(-\\infty\\), for values outside of \\((-1, 1)\\).\ntarget += log(fmax(0.0,1 - abs(y)));\nWith the constraint on y in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on y is removed, the model will compile and run without arithmetic errors, but will not sample properly.1\n\n\n\nIf Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where lambda is the inverse scale and y the sampled variate.\ntarget += log(lambda) - y * lambda;\nThis encoding will work for any lambda and y; they can be parameters, data, or one of each, or even local variables.\nThe assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.\ny ~ exponential(lambda);\nThere are two notable differences. First, the distribution statement will check the inputs to make sure both lambda is positive and y is non-negative (which includes checking that neither is the special not-a-number value).\nThe second difference is that if lambda is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the log(lambda) term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If lambda and y are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).\n\n\n\nFor another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation rho. That is, it computes \\[\n\\texttt{binormal}\\mathtt{\\_}\\texttt{cdf}(z_1, z_2, \\rho) = \\Pr[Z_1 \\leq z_1 \\text{ and } Z_2 \\leq z_2]\n\\] where the random 2-vector \\(Z\\) has the distribution \\[\nZ \\sim \\textsf{multivariate normal}\\left(\n\\begin{bmatrix}\n0 \\\\\n0\n\\end{bmatrix}, \\\n\\begin{bmatrix}\n1 & \\rho\n\\\\\n\\rho & 1\n\\end{bmatrix}\n\\right).\n\\]\nThe following Stan program implements this function,\nreal binormal_cdf(tuple(real, real) z, real rho) {\n  real z1 = z.1;\n  real z2 = z.2;\n  if (z1 == 0 && z2 == 0) {\n    return 0.25 + asin(rho) / (2 * pi());\n  }\n  real denom = sqrt((1 + rho) * (1 - rho));\n  real term1 = z1 == 0\n    ? (z2 &gt; 0 ? 0.25 : -0.25)\n    :  owens_t(z1, (z2 / z1 - rho) / denom);\n  real term2 = z2 == 0\n    ? (z1 &gt; 0 ? 0.25 : -0.25)\n    : owens_t(z2, (z1 / z2 - rho) / denom);\n  real z1z2 = z1 * z2;\n  real delta = z1z2 &lt; 0 || (z1z2 == 0 && (z1 + z2) &lt; 0);\n  return 0.5 * (Phi(z1) + Phi(z2) - delta) - term1 - term2;\n}\nIt is written using a tuple argument so that it may be called as binormal_cdf((z1, z2) | rho). The two ternary operators defining term1 and term2 are derived by taking the limit of the owens_t function when the second argument goes to infinity.\nOne way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.\ntransformed data {\n  for (zzr in {[0, 0, 0.5],\n               [0, 1, 0.5],\n               [1, 0, -0.2],\n               [1, 3, 0.9]}) {\n    real z1  = zzr[1];\n    real z2 = zzr[2];\n    real rho = zzr[3];\n    print(\"binomial_cdf((\",  z1, \", \", z2, \") | \", rho, \")\",\n          \"=\", binormal_cdf((z1, z2) | rho));\n  }\n}\nIn this case, we verified that the results match those of the pbivnorm package in R.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Custom Probability Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/custom-probability.html#examples",
+    "href": "stan-users-guide/custom-probability.html#examples",
+    "title": "Custom Probability Functions",
+    "section": "",
+    "text": "A simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}\\) are the bounds, with \\(\\alpha &lt; \\beta\\), then \\(y \\in (\\alpha,\\beta)\\) has a density defined as \\[\n\\textsf{triangle}(y \\mid \\alpha,\\beta)\n=\n\\frac{1}{(\\beta - \\alpha)^2}\n\\cdot\n\\textrm{min}(y - \\alpha, \\beta - y).\n\\]\nThe general form of triangle can be coded in Stan as follows.\ndata {\n  real alpha;\n  real&lt;lower=alpha&gt; beta;\n}\nparameters {\n  real&lt;lower=alpha, upper=beta&gt; y;\n}\nmodel {\n  target += -2 * log(beta - alpha)\n    + log(fmin(y - alpha, beta - y));\n}\nBecause the bounds are specified as data here, the term -2 * log(beta - alpha) could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.\nIf \\(\\alpha = -1\\), \\(\\beta = 1\\), and \\(y \\in (-1,1)\\), then fmin(y - alpha, beta - y) is fmin(y + 1, -1 - y), which is fmin(y + 1, -(y + 1)), which reduces to 1 - abs(y). Therfore, the density, dropping constants, reduces to \\[\n\\textsf{triangle}(y \\mid -1, 1) \\propto 1 - |y|.\n\\] Consider the following Stan implementation of \\(\\textsf{triangle}(-1,1)\\) for sampling.\nparameters {\n  real&lt;lower=-1, upper=1&gt; y;\n}\nmodel {\n  target += log1m(abs(y));\n}\nThe single scalar parameter y is declared as lying in the interval (-1, 1). The total log probability is incremented with the joint log probability of all parameters, i.e., \\(\\log \\mathsf{Triangle}(y \\mid -1, 1)\\). This value is coded in Stan as log1m(abs(y)). The function log1m is defined so that log1m(x) has the same value as \\(\\log(1 - x)\\), but the computation is faster, more accurate, and more stable.\nThe constrained type real&lt;lower=-1, upper=1&gt; declared for y is critical for correct sampling behavior. If the constraint on y is removed from the program, say by declaring y as having the unconstrained scalar type real, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of y outside of \\((-1,1)\\).\nNow suppose the log probability function were extended to all of \\(\\mathbb{R}\\) as follows by defining the probability to be log(0.0), i.e., \\(-\\infty\\), for values outside of \\((-1, 1)\\).\ntarget += log(fmax(0.0,1 - abs(y)));\nWith the constraint on y in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on y is removed, the model will compile and run without arithmetic errors, but will not sample properly.1\n\n\n\nIf Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where lambda is the inverse scale and y the sampled variate.\ntarget += log(lambda) - y * lambda;\nThis encoding will work for any lambda and y; they can be parameters, data, or one of each, or even local variables.\nThe assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.\ny ~ exponential(lambda);\nThere are two notable differences. First, the distribution statement will check the inputs to make sure both lambda is positive and y is non-negative (which includes checking that neither is the special not-a-number value).\nThe second difference is that if lambda is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the log(lambda) term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If lambda and y are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).\n\n\n\nFor another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation rho. That is, it computes \\[\n\\texttt{binormal}\\mathtt{\\_}\\texttt{cdf}(z_1, z_2, \\rho) = \\Pr[Z_1 \\leq z_1 \\text{ and } Z_2 \\leq z_2]\n\\] where the random 2-vector \\(Z\\) has the distribution \\[\nZ \\sim \\textsf{multivariate normal}\\left(\n\\begin{bmatrix}\n0 \\\\\n0\n\\end{bmatrix}, \\\n\\begin{bmatrix}\n1 & \\rho\n\\\\\n\\rho & 1\n\\end{bmatrix}\n\\right).\n\\]\nThe following Stan program implements this function,\nreal binormal_cdf(tuple(real, real) z, real rho) {\n  real z1 = z.1;\n  real z2 = z.2;\n  if (z1 == 0 && z2 == 0) {\n    return 0.25 + asin(rho) / (2 * pi());\n  }\n  real denom = sqrt((1 + rho) * (1 - rho));\n  real term1 = z1 == 0\n    ? (z2 &gt; 0 ? 0.25 : -0.25)\n    :  owens_t(z1, (z2 / z1 - rho) / denom);\n  real term2 = z2 == 0\n    ? (z1 &gt; 0 ? 0.25 : -0.25)\n    : owens_t(z2, (z1 / z2 - rho) / denom);\n  real z1z2 = z1 * z2;\n  real delta = z1z2 &lt; 0 || (z1z2 == 0 && (z1 + z2) &lt; 0);\n  return 0.5 * (Phi(z1) + Phi(z2) - delta) - term1 - term2;\n}\nIt is written using a tuple argument so that it may be called as binormal_cdf((z1, z2) | rho). The two ternary operators defining term1 and term2 are derived by taking the limit of the owens_t function when the second argument goes to infinity.\nOne way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.\ntransformed data {\n  for (zzr in {[0, 0, 0.5],\n               [0, 1, 0.5],\n               [1, 0, -0.2],\n               [1, 3, 0.9]}) {\n    real z1  = zzr[1];\n    real z2 = zzr[2];\n    real rho = zzr[3];\n    print(\"binomial_cdf((\",  z1, \", \", z2, \") | \", rho, \")\",\n          \"=\", binormal_cdf((z1, z2) | rho));\n  }\n}\nIn this case, we verified that the results match those of the pbivnorm package in R.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Custom Probability Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/custom-probability.html#footnotes",
+    "href": "stan-users-guide/custom-probability.html#footnotes",
+    "title": "Custom Probability Functions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe problem is the (extremely!) light tails of the triangle distribution. The standard HMC and NUTS samplers can’t get into the corners of the triangle properly. Because the Stan code declares y to be of type real&lt;lower=-1, upper=1&gt;, the inverse logit transform is applied to the unconstrained variable and its log absolute derivative added to the log probability. The resulting distribution on the logit-transformed y is well behaved.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Custom Probability Functions"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/decision-analysis.html",
+    "href": "stan-users-guide/decision-analysis.html",
+    "title": "Decision Analysis",
+    "section": "",
+    "text": "Statistical decision analysis is about making decisions under uncertainty. In order to make decisions, outcomes must have some notion of “utility” associated with them. The so-called “Bayes optimal” decision is the one that maximizes expected utility (or equivalently, minimizes expected loss). This chapter shows how Stan can be used to simultaneously estimate the distribution of outcomes based on decisions and compute the required expected utilities.\n\n\nFollowing Gelman et al. (2013), Bayesian decision analysis can be factored into the following four steps.\n\nDefine a set \\(X\\) of possible outcomes and a set \\(D\\) of possible decisions.\nDefine a probability distribution of outcomes conditional on decisions through a conditional density function \\(p(x \\mid d)\\) for \\(x \\in X\\) and \\(d \\in D.\\)\nDefine a utility function \\(U : X \\rightarrow \\mathbb{R}\\) mapping outcomes to their utility.\nChoose action \\(d^* \\in D\\) with highest expected utility, \\[\nd^* = \\textrm{arg max}_d \\ \\mathbb{E}[U(x) \\mid d].\n\\]\n\nThe outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).\n\n\n\nThis section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.\n\n\nA decision consists of the choice of commute mode and the outcome is a time and cost. More formally,\n\nthe set of decisions is \\(D = 1:4\\), corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and\nthe set of outcomes \\(X = \\mathbb{R} \\times \\mathbb{R}_+\\) contains pairs of numbers \\(x = (c, t)\\) consisting of a cost \\(c\\) and time \\(t \\geq 0\\).\n\n\n\n\nThe density required is \\(p(x \\mid d),\\) where \\(d \\in D\\) is a decision and \\(x = (c, t) \\in X\\) is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters \\(\\theta,\\) \\[\np(x \\mid d, x^{\\textrm{obs}}, d^{\\textrm{obs}})\n=\n\\int\n  p(x \\mid d, \\theta)\n  \\cdot p(\\theta \\mid x^{\\textrm{obs}}, d^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta.\n\\] The observed data for a year of commutes consists of choice of the chosen commute mode \\(d^{\\textrm{obs}}_n\\) and observed costs and times \\(x^{\\textrm{obs}}_n = (c^{\\textrm{obs}}_n, t^{\\textrm{obs}}_n)\\) for \\(n\n\\in 1:200.\\)\nFor simplicity, commute time \\(t_n\\) for trip \\(n\\) will be modeled as lognormal for a given choice of transportation \\(d_n \\in 1:4,\\) \\[\nt_n \\sim \\textrm{lognormal}(\\mu_{d[n]}, \\sigma_{d[n]}).\n\\] To understand the notation, \\(d_n\\), also written \\(d[n]\\), is the mode of transportation used for trip \\(n\\). For example if trip \\(n\\) was by bicycle, then \\(t_n \\sim \\textrm{lognormal}(\\mu_2, \\sigma_2),\\) where \\(\\mu_2\\) and \\(\\sigma_2\\) are the lognormal parameters for bicycling.\nSimple fixed priors are used for each mode of transportation \\(k \\in 1:4,\\) \\[\\begin{eqnarray*}\n\\mu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\sigma_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.\nCost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, \\[\nc_n \\sim \\textrm{lognormal}(\\nu_{d[n]}, \\tau_{d[n]}).\n\\] Again, the priors are fixed for the modes of transportation, \\[\\begin{eqnarray*}\n\\nu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\tau_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.\nThe full set of parameters that are marginalized in the posterior predictive distribution is \\[\n\\theta = (\\mu_{1:4}, \\sigma_{1:4}, \\nu_{1:4}, \\tau_{1:4}).\n\\]\n\n\n\nFor the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as\n\\[\nU(c, t) = -(c + 25 \\cdot t)\n\\]\nThe sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.\n\n\n\nAt this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.\nfunctions {\n  real U(real c, real t) {\n    return -(c + 25 * t);\n  }\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; d;\n  array[N] real c;\n  array[N] real&lt;lower=0&gt; t;\n}\nparameters {\n  vector[4] mu;\n  vector&lt;lower=0&gt;[4] sigma;\n  array[4] real nu;\n  array[4] real&lt;lower=0&gt; tau;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.25);\n  nu ~ normal(0, 20);\n  tau ~ lognormal(0, 0.25);\n  t ~ lognormal(mu[d], sigma[d]);\n  c ~ lognormal(nu[d], tau[d]);\n}\ngenerated quantities {\n  array[4] real util;\n  for (k in 1:4) {\n    util[k] = U(lognormal_rng(nu[k], tau[k]),\n                lognormal_rng(mu[k], sigma[k]));\n  }\n}\nThe generated quantities block defines an array variable util where util[k], which will hold the utility derived from a random commute for choice k generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.\nFor simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, nu and tau would be declared as univariate, and the RNG for cost would only be employed when k == 4.\nDefining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.\nAll that is left is to run Stan. The posterior mean for util[k] is the expected utility, which written out with full conditioning, is \\[\\begin{eqnarray*}\n\\mathbb{E}\\!\\left[U(x) \\mid d = k, d^{\\textrm{obs}}, x^{\\textrm{obs}}\\right]\n& = &\n\\int\n  U(x)\n  \\cdot p(x \\mid d = k, \\theta)\n  \\cdot p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M  U(x^{(m)} ),\n\\end{eqnarray*}\\] where \\[\nx^{(m)} \\sim p(x \\mid d = k, \\theta^{(m)} )\n\\] and \\[\n\\theta^{(m)}\n\\sim p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}}).\n\\]\nIn terms of Stan’s execution, the random generation of \\(x^{(m)}\\) is carried out with the lognormal_rng operations after \\(\\theta^{(m)}\\) is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.\nIt only remains to make the decision k with highest expected utility, which will correspond to the choice with the highest posterior mean for util[k]. This can be read off of the mean column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.\n\n\n\n\nMany choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/decision-analysis.html#outline-of-decision-analysis",
+    "href": "stan-users-guide/decision-analysis.html#outline-of-decision-analysis",
+    "title": "Decision Analysis",
+    "section": "",
+    "text": "Following Gelman et al. (2013), Bayesian decision analysis can be factored into the following four steps.\n\nDefine a set \\(X\\) of possible outcomes and a set \\(D\\) of possible decisions.\nDefine a probability distribution of outcomes conditional on decisions through a conditional density function \\(p(x \\mid d)\\) for \\(x \\in X\\) and \\(d \\in D.\\)\nDefine a utility function \\(U : X \\rightarrow \\mathbb{R}\\) mapping outcomes to their utility.\nChoose action \\(d^* \\in D\\) with highest expected utility, \\[\nd^* = \\textrm{arg max}_d \\ \\mathbb{E}[U(x) \\mid d].\n\\]\n\nThe outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/decision-analysis.html#example-decision-analysis",
+    "href": "stan-users-guide/decision-analysis.html#example-decision-analysis",
+    "title": "Decision Analysis",
+    "section": "",
+    "text": "This section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.\n\n\nA decision consists of the choice of commute mode and the outcome is a time and cost. More formally,\n\nthe set of decisions is \\(D = 1:4\\), corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and\nthe set of outcomes \\(X = \\mathbb{R} \\times \\mathbb{R}_+\\) contains pairs of numbers \\(x = (c, t)\\) consisting of a cost \\(c\\) and time \\(t \\geq 0\\).\n\n\n\n\nThe density required is \\(p(x \\mid d),\\) where \\(d \\in D\\) is a decision and \\(x = (c, t) \\in X\\) is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters \\(\\theta,\\) \\[\np(x \\mid d, x^{\\textrm{obs}}, d^{\\textrm{obs}})\n=\n\\int\n  p(x \\mid d, \\theta)\n  \\cdot p(\\theta \\mid x^{\\textrm{obs}}, d^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta.\n\\] The observed data for a year of commutes consists of choice of the chosen commute mode \\(d^{\\textrm{obs}}_n\\) and observed costs and times \\(x^{\\textrm{obs}}_n = (c^{\\textrm{obs}}_n, t^{\\textrm{obs}}_n)\\) for \\(n\n\\in 1:200.\\)\nFor simplicity, commute time \\(t_n\\) for trip \\(n\\) will be modeled as lognormal for a given choice of transportation \\(d_n \\in 1:4,\\) \\[\nt_n \\sim \\textrm{lognormal}(\\mu_{d[n]}, \\sigma_{d[n]}).\n\\] To understand the notation, \\(d_n\\), also written \\(d[n]\\), is the mode of transportation used for trip \\(n\\). For example if trip \\(n\\) was by bicycle, then \\(t_n \\sim \\textrm{lognormal}(\\mu_2, \\sigma_2),\\) where \\(\\mu_2\\) and \\(\\sigma_2\\) are the lognormal parameters for bicycling.\nSimple fixed priors are used for each mode of transportation \\(k \\in 1:4,\\) \\[\\begin{eqnarray*}\n\\mu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\sigma_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.\nCost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, \\[\nc_n \\sim \\textrm{lognormal}(\\nu_{d[n]}, \\tau_{d[n]}).\n\\] Again, the priors are fixed for the modes of transportation, \\[\\begin{eqnarray*}\n\\nu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\tau_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.\nThe full set of parameters that are marginalized in the posterior predictive distribution is \\[\n\\theta = (\\mu_{1:4}, \\sigma_{1:4}, \\nu_{1:4}, \\tau_{1:4}).\n\\]\n\n\n\nFor the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as\n\\[\nU(c, t) = -(c + 25 \\cdot t)\n\\]\nThe sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.\n\n\n\nAt this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.\nfunctions {\n  real U(real c, real t) {\n    return -(c + 25 * t);\n  }\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; d;\n  array[N] real c;\n  array[N] real&lt;lower=0&gt; t;\n}\nparameters {\n  vector[4] mu;\n  vector&lt;lower=0&gt;[4] sigma;\n  array[4] real nu;\n  array[4] real&lt;lower=0&gt; tau;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.25);\n  nu ~ normal(0, 20);\n  tau ~ lognormal(0, 0.25);\n  t ~ lognormal(mu[d], sigma[d]);\n  c ~ lognormal(nu[d], tau[d]);\n}\ngenerated quantities {\n  array[4] real util;\n  for (k in 1:4) {\n    util[k] = U(lognormal_rng(nu[k], tau[k]),\n                lognormal_rng(mu[k], sigma[k]));\n  }\n}\nThe generated quantities block defines an array variable util where util[k], which will hold the utility derived from a random commute for choice k generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.\nFor simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, nu and tau would be declared as univariate, and the RNG for cost would only be employed when k == 4.\nDefining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.\nAll that is left is to run Stan. The posterior mean for util[k] is the expected utility, which written out with full conditioning, is \\[\\begin{eqnarray*}\n\\mathbb{E}\\!\\left[U(x) \\mid d = k, d^{\\textrm{obs}}, x^{\\textrm{obs}}\\right]\n& = &\n\\int\n  U(x)\n  \\cdot p(x \\mid d = k, \\theta)\n  \\cdot p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M  U(x^{(m)} ),\n\\end{eqnarray*}\\] where \\[\nx^{(m)} \\sim p(x \\mid d = k, \\theta^{(m)} )\n\\] and \\[\n\\theta^{(m)}\n\\sim p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}}).\n\\]\nIn terms of Stan’s execution, the random generation of \\(x^{(m)}\\) is carried out with the lognormal_rng operations after \\(\\theta^{(m)}\\) is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.\nIt only remains to make the decision k with highest expected utility, which will correspond to the choice with the highest posterior mean for util[k]. This can be read off of the mean column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/decision-analysis.html#continuous-choices",
+    "href": "stan-users-guide/decision-analysis.html#continuous-choices",
+    "title": "Decision Analysis",
+    "section": "",
+    "text": "Many choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html",
+    "href": "stan-users-guide/finite-mixtures.html",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "Finite mixture models of an outcome assume that the outcome is drawn from one of several distributions, the identity of which is controlled by a categorical mixing distribution. Mixture models typically have multimodal densities with modes near the modes of the mixture components. Mixture models may be parameterized in several ways, as described in the following sections. Mixture models may be used directly for modeling data with multimodal distributions, or they may be used as priors for other parameters.\n\n\nClustering models, as discussed in the clustering chapter, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the \\(K\\)-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.\n\n\n\nOne way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider \\(K\\) normal distributions with locations \\(\\mu_k \\in \\mathbb{R}\\) and scales \\(\\sigma_k \\in (0,\\infty)\\). Now consider mixing them in proportion \\(\\lambda\\), where \\(\\lambda_k \\geq 0\\) and \\(\\sum_{k=1}^K \\lambda_k = 1\\) (i.e., \\(\\lambda\\) lies in the unit \\(K\\)-simplex). For each outcome \\(y_n\\) there is a latent variable \\(z_n\\) in \\(\\{ 1,\\dotsc,K \\}\\) with a categorical distribution parameterized by \\(\\lambda\\), \\[\nz_n \\sim \\textsf{categorical}(\\lambda).\n\\]\nThe variable \\(y_n\\) is distributed according to the parameters of the mixture component \\(z_n\\), \\[\ny_n \\sim \\textsf{normal}(\\mu_{z[n]},\\sigma_{z[n]}).\n\\]\nThis model is not directly supported by Stan because it involves discrete parameters \\(z_n\\), but Stan can sample \\(\\mu\\) and \\(\\sigma\\) by summing out the \\(z\\) parameter as described in the next section.\n\n\n\nTo implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If \\(Y\\) is a mixture of \\(K\\) normal distributions with locations \\(\\mu_k\\) and scales \\(\\sigma_k\\) with mixing proportions \\(\\lambda\\) in the unit \\(K\\)-simplex, then \\[\np_Y\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n=\n\\sum_{k=1}^K \\lambda_k \\, \\textsf{normal}\\left(y \\mid \\mu_k, \\sigma_k\\right).\n\\]\n\n\nThe log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b) = \\log \\left(\\exp(a) + \\exp(b)\\right).\n\\]\nIf \\(a\\) and \\(b\\) are probabilities on the log scale, then \\(\\exp(a) +\n\\exp(b)\\) is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in log_sum_exp function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as \\[\n\\log \\left( \\exp(a) + \\exp(b)\\right)\n= c + \\log \\left( \\exp(a - c) + \\exp(b - c) \\right),\n\\] where \\(c = \\max(a, b)\\). In this evaluation, one of the terms, \\(a - c\\) or \\(b - c\\), is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the \\(\\max(a, b)\\) out of the log-exp round trip.\nFor example, the mixture of \\(\\textsf{normal}(-1, 2)\\) with \\(\\textsf{normal}(3, 1)\\), with mixing proportion \\(\\lambda =\n[0.3,0.7]^{\\top}\\), can be implemented in Stan as follows.\nparameters {\n  real y;\n}\nmodel {\n  target += log_sum_exp(log(0.3) + normal_lpdf(y | -1, 2),\n                        log(0.7) + normal_lpdf(y | 3, 1));\n}\nThe log probability term is derived by taking \\[\\begin{align*}\n\\log\\, &p\\left(y \\mid \\lambda,\\mu,\\sigma \\right) \\\\\n&= \\log\\big( 0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\n            + 0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big) \\\\\n&= \\log\\bigg( \\exp\\Big(\\log\\big(0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\\big)\\Big)\n             + \\exp\\Big(\\log\\big(0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right)\\big)\\Big) \\bigg) \\\\\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}\\big(\n                         \\log(0.3) + \\log \\textsf{normal}\\left(y \\mid -1,2 \\right),\n                         \\log(0.7) + \\log \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big).\n\\end{align*}\\]\n\n\n\nIf a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because\nlog_half = log(0.5);\nfor (n in 1:N) {\n    target +=\n        log_sum_exp(log_half + normal_lpdf(y[n] | mu[1], sigma[1]),\n                    log_half + normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nthen the \\(\\log 0.5\\) term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version\nfor (n in 1:N) {\n  target += log_sum_exp(normal_lpdf(y[n] | mu[1], sigma[1]),\n                        normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nThe same result holds if there are \\(K\\) components and the mixing simplex \\(\\lambda\\) is symmetric, i.e., \\[\n\\lambda = \\left( \\frac{1}{K}, \\dotsc, \\frac{1}{K} \\right).\n\\]\nThe result follows from the identity \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(c + a, c + b)\n=\nc + \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b)\n\\] and the fact that adding a constant \\(c\\) to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.\n\n\n\nThe posterior \\(p(z_n \\mid y_n, \\mu, \\sigma)\\) over the mixture indicator \\(z_n\n\\in 1:K\\) is often of interest as \\(p(z_n = k \\mid y, \\mu, \\sigma)\\) is the posterior probability that that observation \\(y_n\\) was generated by mixture component \\(k\\). The posterior can be computed via Bayes’s rule, \\[\\begin{align*}\n\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right]\n  &\\propto p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right)\\, p\\left(z_n = k \\mid \\lambda\\right) \\\\\n  &= \\textsf{normal}\\left(y_n \\mid \\mu_k, \\sigma_k\\right) \\cdot \\lambda_k.\n\\end{align*}\\]\nThe normalization can be done via summation, because \\(z_n \\in 1{:}K\\) only takes on finitely many values. In detail, \\[\np\\left(z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right) =\n\\frac{p\\left(y_n \\mid z_n = k, \\mu, \\sigma \\right) \\cdot p\\left(z_n = k \\mid \\lambda \\right)}\n     {\\sum_{k' = 1}^K p\\left(y_n \\mid z_n = k', \\mu, \\sigma \\right)\n                    \\cdot p\\left(z_n = k' \\mid \\lambda \\right)}.\n\\]\nOn the log scale, the normalized probability is computed as \\[\\begin{align*}\n\\log\\,&\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda\\right] \\\\\n&= \\log p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right) + \\log \\Pr\\!\\left[z_n = k \\mid \\lambda\\right] \\\\\n&\\quad  - \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{k' = 1}^K \\big(\\log p\\left(y_n \\mid z_n = k', \\mu, \\sigma\\right)\n                                                 + \\log p\\left(z_n = k' \\mid \\lambda\\right)\\big).\n\\end{align*}\\] This can be coded up directly in Stan; the change-point model in the change point section provides an example.\n\n\n\nGiven the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.\ndata {\n  int&lt;lower=1&gt; K;          // number of mixture components\n  int&lt;lower=1&gt; N;          // number of data points\n  array[N] real y;         // observations\n}\nparameters {\n  simplex[K] theta;          // mixing proportions\n  ordered[K] mu;             // locations of mixture components\n  vector&lt;lower=0&gt;[K] sigma;  // scales of mixture components\n}\nmodel {\n  vector[K] log_theta = log(theta);  // cache log calculation\n  sigma ~ lognormal(0, 2);\n  mu ~ normal(0, 10);\n  for (n in 1:N) {\n    vector[K] lps = log_theta;\n    for (k in 1:K) {\n      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);\n    }\n    target += log_sum_exp(lps);\n  }\n}\nThe model involves K mixture components and N data points. The mixing proportion parameter theta is declared to be a unit \\(K\\)-simplex, whereas the component location parameter mu and scale parameter sigma are both defined to be K-vectors.\nThe location parameter mu is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components mu[k] is symmetric, as it is here (each component has an independent \\(\\textsf{normal}(0, 10)\\) prior). It would even be possible to include a hierarchical prior for the components.\nThe values in the scale array sigma are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.\nThe model declares a local array variable lps to be size K and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points n. For each such point, the log of \\(\\theta_k \\times\n\\textsf{normal}\\left(y_n \\mid \\mu_k,\\sigma_k\\right)\\) is calculated and added to the array lps. Then the log probability is incremented with the log sum of exponentials of those values.\n\n\n\n\nThere is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that lambda, y[n], mu[1], mu[2], and sigma[1], sigma[2] are all scalars and lambda is between 0 and 1.\nfor (n in 1:N) {\n  target += log_sum_exp(log(lambda)\n                          + normal_lpdf(y[n] | mu[1], sigma[1]),\n                        log1m(lambda)\n                          + normal_lpdf(y[n] | mu[2], sigma[2]));\nor equivalently\nfor (n in 1:N) {\n  target += log_mix(lambda,\n                    normal_lpdf(y[n] | mu[1], sigma[1]),\n                    normal_lpdf(y[n] | mu[2], sigma[2]))\n  };\nThis definition assumes that each observation \\(y_n\\) may have arisen from either of the mixture components. The density is \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma\\right)\n= \\prod_{n=1}^N \\big(\\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1 \\right)\n                 + (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2 \\right)\\big).\n\\]\nContrast the previous model with the following (erroneous) attempt to vectorize the model.\ntarget += log_sum_exp(log(lambda)\n                        + normal_lpdf(y | mu[1], sigma[1]),\n                      log1m(lambda)\n                        + normal_lpdf(y | mu[2], sigma[2]));\nor equivalently,\ntarget += log_mix(lambda,\n                  normal_lpdf(y | mu[1], sigma[1]),\n                  normal_lpdf(y | mu[2], sigma[2]));\nThis second definition implies that the entire sequence \\(y_1, \\dotsc, y_n\\) of observations comes form one component or the other, defining a different density, \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n= \\lambda \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right).\n\\]\n\n\n\nIn many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.\nWe have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs. female, or Democrat vs. Republican).\nIn other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.\n\n\nAs an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\), \\[\\begin{align*}\n\\mu_1, \\mu_2       &\\sim \\textsf{normal}(0, 10) \\\\\n\\sigma_1, \\sigma_2 &\\sim \\textsf{halfnormal}(0, 10) \\\\\n\\end{align*}\\]\nThe prior on the mixture ratio is uniform, \\[\n\\lambda \\sim \\textsf{uniform}(0, 1),\n\\] so that with the likelihood \\[\np\\left(y_n \\mid \\mu, \\sigma\\right)\n= \\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right),\n\\] the joint distribution \\(p(y, \\mu, \\sigma, \\lambda)\\) is exchangeable in the parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\) with \\(\\lambda\\) flipping to \\(1 - \\lambda\\).1\n\n\n\nIn cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.\n\n\nPosterior predictive distribution for a new observation \\(\\tilde{y}\\) given the complete parameter vector \\(\\theta\\) will be \\[\np(\\tilde{y} \\mid y)\n=\n\\int_{\\theta}\np(\\tilde{y} \\mid \\theta)\n\\, p(\\theta \\mid y)\n\\, \\textsf{d}\\theta.\n\\]\nThe normal mixture example from the previous section, with \\(\\theta =\n(\\mu, \\sigma, \\lambda)\\), shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing \\(p(\\tilde{y} \\mid y)\\), which is more efficient statistically in terms of effective sample size, or simulating draws of \\(\\tilde{y}\\), which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] y_tilde;\n  // ...\n}\ngenerated quantities {\n  vector[N_tilde] log_p_y_tilde;\n  for (n in 1:N_tilde) {\n    log_p_y_tilde[n]\n      = log_mix(lambda,\n                normal_lpdf(y_tilde[n] | mu[1], sigma[1])\n                normal_lpdf(y_tilde[n] | mu[2], sigma[2]));\n  }\n}\nIt is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply log_sum_exp of the posterior draws of log_p_y_tilde. The average log predictive density is then given by subtracting log(N_new).\n\n\n\nOften a mixture model will be applied to a clustering problem and there might be two data items \\(y_i\\) and \\(y_j\\) for which there is a question of whether they arose from the same mixture component. If we take \\(z_i\\) and \\(z_j\\) to be the component responsibility discrete variables, then the quantity of interest is \\(z_i = z_j\\), which can be summarized as an event probability \\[\n\\Pr[z_i = z_j \\mid y]\n=\n\\int_{\\theta}\n\\frac{\\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \\mid \\theta)}\n     {\\sum_{k=0}^1 \\sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \\mid \\theta)}\n\\,\np(\\theta \\mid y)\n\\,\n\\textsf{d}\\theta.\n\\]\nAs with other event probabilities, this can be calculated in the generated quantities block either by sampling \\(z_i\\) and \\(z_j\\) and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.\n\n\n\n\n\nZero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by Lambert (1992), add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.\nZero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.\n\n\nConsider the following example for zero-inflated Poisson distributions. There is a probability \\(\\theta\\) of observing a zero, and a probability \\(1 - \\theta\\) of observing a count with a \\(\\textsf{Poisson}(\\lambda)\\) distribution (now \\(\\theta\\) is being used for mixing proportions because \\(\\lambda\\) is the traditional notation for a Poisson mean parameter). Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as \\[\\begin{align*}\ny_n & = 0 & \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}(y_n \\mid \\lambda) & \\quad\\text{with probability } 1-\\theta.\n\\end{align*}\\]\nStan does not support conditional distribution statements (with ~) conditional on some parameter, and we need to consider the corresponding likelihood \\[\np(y_n \\mid \\theta,\\lambda)\n=\n\\begin{cases}\n\\theta + (1 - \\theta) \\times \\textsf{Poisson}(0 \\mid \\lambda) & \\quad\\text{if } y_n = 0, \\text{ and}\\\\\n(1-\\theta) \\times \\textsf{Poisson}(y_n \\mid \\lambda) &\\quad\\text{if } y_n &gt; 0.\n\\end{cases}\n\\] The log likelihood can be coded directly in Stan (with target +=) as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  for (n in 1:N) {\n    if (y[n] == 0) {\n      target += log_sum_exp(log(theta),\n                            log1m(theta)\n                              + poisson_lpmf(y[n] | lambda));\n    } else {\n      target += log1m(theta)\n                  + poisson_lpmf(y[n] | lambda);\n    }\n  }\n}\nThe log1m(theta) computes log(1-theta), but is more computationally stable. The log_sum_exp(lp1,lp2) function adds the log probabilities on the linear scale; it is defined to be equal to log(exp(lp1) + exp(lp2)), but is more computationally stable and faster.\n\n\nThe code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also poisson_lpmf(0 | lambda) every time the first condition body executes. The use of the redundant terms is conditioned on y, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.\nThe number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare y_nonzero, so it must be computed in a function.\nfunctions {\n  int num_zeros(array[] int y) {\n    int sum = 0;\n    for (n in 1:size(y)) {\n      sum += (y[n] == 0);\n    }\n    return sum;\n  }\n}\n// ...\ntransformed data {\n  int&lt;lower=0&gt; N_zero = num_zeros(y);\n  array[N - N_zero] int&lt;lower=1&gt; y_nonzero;\n  int N_nonzero = 0;\n  for (n in 1:N) {\n    if (y[n] == 0) continue;\n    N_nonzero += 1;\n    y_nonzero[N_nonzero] = y[n];\n  }\n}\n// ...\nmodel {\n  // ...\n   target\n     += N_zero\n          * log_sum_exp(log(theta),\n                        log1m(theta)\n                          + poisson_lpmf(0 | lambda));\n   target += N_nonzero * log1m(theta);\n   target += poisson_lpmf(y_nonzero | lambda);\n  // ...\n}\nThe boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if y_nonzero is empty, N_nonzero will be zero, and the last two target increment terms will add zeros.\n\n\n\n\nThe hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as [ \\[\\begin{align*}\ny_n & = 0 \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}_{x\\neq 0}(y_n \\mid \\lambda) \\quad\\text{with probability } 1-\\theta,\n\\end{align*}\\] ] Where \\(\\textsf{Poisson}_{x\\neq 0}\\) is a truncated Poisson distribution, truncated at \\(0\\).\nThe corresponding likelihood function for the hurdle model is defined by \\[\np(y\\mid\\theta,\\lambda)\n=\n\\begin{cases}\n\\theta &\\quad\\text{if } y = 0, \\text{ and}\\\\\n(1 - \\theta)\n   \\frac{\\displaystyle \\textsf{Poisson}(y \\mid \\lambda)}\n        {\\displaystyle  1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)}\n&\\quad\\text{if } y &gt; 0,\n\\end{cases}\n\\] where \\(\\textsf{PoissonCDF}\\) is the cumulative distribution function for the Poisson distribution and and \\(1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)\\) is the relative normalization term for the truncated Poisson (truncated at \\(0\\)).\nThe hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.\nif (y[n] == 0) {\n  target += log(theta);\n} else {\n  target += log1m(theta) + poisson_lpmf(y[n] | lambda)\n            - poisson_lccdf(0 | lambda));\n}\nJulian King pointed out that because \\[\\begin{align*}\n\\log \\left( 1 - \\textsf{PoissonCDF}(0 \\mid \\lambda) \\right)\n  &= \\log \\left( 1 - \\textsf{Poisson}(0 \\mid \\lambda) \\right) \\\\\n  &= \\log(1 - \\exp(-\\lambda))\n\\end{align*}\\] the CCDF in the else clause can be replaced with a simpler expression.\ntarget += log1m(theta) + poisson_lpmf(y[n] | lambda)\n          - log1m_exp(-lambda));\nThe resulting code is about 15% faster than the code with the CCDF.\nThis is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size \\(N=200\\) and parameters \\(\\theta=0.3\\) and \\(\\lambda = 8\\), the speedup is a factor of 10; it will be lower for smaller \\(N\\) and greater for larger \\(N\\); it will also be greater for larger \\(\\theta\\).\nTo achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,\nfunctions {\n  int num_zero(array[] int y) {\n    int nz = 0;\n    for (n in 1:size(y)) {\n      if (y[n] == 0) {\n        nz += 1;\n      }\n    }\n    return nz;\n  }\n}\nThen a transformed data block can be used to store the sufficient statistics,\ntransformed data {\n  int&lt;lower=0, upper=N&gt; N0 = num_zero(y);\n  int&lt;lower=0, upper=N&gt; Ngt0 = N - N0;\n  array[N - num_zero(y)] int&lt;lower=1&gt; y_nz;\n  {\n    int pos = 1;\n    for (n in 1:N) {\n      if (y[n] != 0) {\n        y_nz[pos] = y[n];\n        pos += 1;\n      }\n    }\n  }\n}\nThe model block is then reduced to three statements.\nmodel {\n  N0 ~ binomial(N, theta);\n  y_nz ~ poisson(lambda);\n  target += -Ngt0 * log1m_exp(-lambda);\n}\nThe first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant Ngt0; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.\n\n\n\n\nSuppose we have a two-component mixture model with mixing rate \\(\\lambda \\in (0, 1)\\). Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are \\(N\\) observations, the mixture components will be estimated with effective data sizes of \\(\\theta \\, N\\) and \\((1 - \\theta)\n\\, N\\) for the two components for some \\(\\theta \\in (0, 1)\\). The effective weighting size is determined by posterior responsibility, not simply by the mixing rate \\(\\lambda\\).\n\n\nIn contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data \\(y\\).\nIf different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see Hoeting et al. (1999) and Gelman et al. (2013) for theoretical and computational details).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#clustering-mixture.section",
+    "href": "stan-users-guide/finite-mixtures.html#clustering-mixture.section",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "Clustering models, as discussed in the clustering chapter, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the \\(K\\)-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#latent-discrete-parameterization",
+    "href": "stan-users-guide/finite-mixtures.html#latent-discrete-parameterization",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "One way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider \\(K\\) normal distributions with locations \\(\\mu_k \\in \\mathbb{R}\\) and scales \\(\\sigma_k \\in (0,\\infty)\\). Now consider mixing them in proportion \\(\\lambda\\), where \\(\\lambda_k \\geq 0\\) and \\(\\sum_{k=1}^K \\lambda_k = 1\\) (i.e., \\(\\lambda\\) lies in the unit \\(K\\)-simplex). For each outcome \\(y_n\\) there is a latent variable \\(z_n\\) in \\(\\{ 1,\\dotsc,K \\}\\) with a categorical distribution parameterized by \\(\\lambda\\), \\[\nz_n \\sim \\textsf{categorical}(\\lambda).\n\\]\nThe variable \\(y_n\\) is distributed according to the parameters of the mixture component \\(z_n\\), \\[\ny_n \\sim \\textsf{normal}(\\mu_{z[n]},\\sigma_{z[n]}).\n\\]\nThis model is not directly supported by Stan because it involves discrete parameters \\(z_n\\), but Stan can sample \\(\\mu\\) and \\(\\sigma\\) by summing out the \\(z\\) parameter as described in the next section.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#summing-out-the-responsibility-parameter",
+    "href": "stan-users-guide/finite-mixtures.html#summing-out-the-responsibility-parameter",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "To implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If \\(Y\\) is a mixture of \\(K\\) normal distributions with locations \\(\\mu_k\\) and scales \\(\\sigma_k\\) with mixing proportions \\(\\lambda\\) in the unit \\(K\\)-simplex, then \\[\np_Y\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n=\n\\sum_{k=1}^K \\lambda_k \\, \\textsf{normal}\\left(y \\mid \\mu_k, \\sigma_k\\right).\n\\]\n\n\nThe log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b) = \\log \\left(\\exp(a) + \\exp(b)\\right).\n\\]\nIf \\(a\\) and \\(b\\) are probabilities on the log scale, then \\(\\exp(a) +\n\\exp(b)\\) is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in log_sum_exp function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as \\[\n\\log \\left( \\exp(a) + \\exp(b)\\right)\n= c + \\log \\left( \\exp(a - c) + \\exp(b - c) \\right),\n\\] where \\(c = \\max(a, b)\\). In this evaluation, one of the terms, \\(a - c\\) or \\(b - c\\), is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the \\(\\max(a, b)\\) out of the log-exp round trip.\nFor example, the mixture of \\(\\textsf{normal}(-1, 2)\\) with \\(\\textsf{normal}(3, 1)\\), with mixing proportion \\(\\lambda =\n[0.3,0.7]^{\\top}\\), can be implemented in Stan as follows.\nparameters {\n  real y;\n}\nmodel {\n  target += log_sum_exp(log(0.3) + normal_lpdf(y | -1, 2),\n                        log(0.7) + normal_lpdf(y | 3, 1));\n}\nThe log probability term is derived by taking \\[\\begin{align*}\n\\log\\, &p\\left(y \\mid \\lambda,\\mu,\\sigma \\right) \\\\\n&= \\log\\big( 0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\n            + 0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big) \\\\\n&= \\log\\bigg( \\exp\\Big(\\log\\big(0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\\big)\\Big)\n             + \\exp\\Big(\\log\\big(0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right)\\big)\\Big) \\bigg) \\\\\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}\\big(\n                         \\log(0.3) + \\log \\textsf{normal}\\left(y \\mid -1,2 \\right),\n                         \\log(0.7) + \\log \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big).\n\\end{align*}\\]\n\n\n\nIf a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because\nlog_half = log(0.5);\nfor (n in 1:N) {\n    target +=\n        log_sum_exp(log_half + normal_lpdf(y[n] | mu[1], sigma[1]),\n                    log_half + normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nthen the \\(\\log 0.5\\) term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version\nfor (n in 1:N) {\n  target += log_sum_exp(normal_lpdf(y[n] | mu[1], sigma[1]),\n                        normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nThe same result holds if there are \\(K\\) components and the mixing simplex \\(\\lambda\\) is symmetric, i.e., \\[\n\\lambda = \\left( \\frac{1}{K}, \\dotsc, \\frac{1}{K} \\right).\n\\]\nThe result follows from the identity \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(c + a, c + b)\n=\nc + \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b)\n\\] and the fact that adding a constant \\(c\\) to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.\n\n\n\nThe posterior \\(p(z_n \\mid y_n, \\mu, \\sigma)\\) over the mixture indicator \\(z_n\n\\in 1:K\\) is often of interest as \\(p(z_n = k \\mid y, \\mu, \\sigma)\\) is the posterior probability that that observation \\(y_n\\) was generated by mixture component \\(k\\). The posterior can be computed via Bayes’s rule, \\[\\begin{align*}\n\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right]\n  &\\propto p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right)\\, p\\left(z_n = k \\mid \\lambda\\right) \\\\\n  &= \\textsf{normal}\\left(y_n \\mid \\mu_k, \\sigma_k\\right) \\cdot \\lambda_k.\n\\end{align*}\\]\nThe normalization can be done via summation, because \\(z_n \\in 1{:}K\\) only takes on finitely many values. In detail, \\[\np\\left(z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right) =\n\\frac{p\\left(y_n \\mid z_n = k, \\mu, \\sigma \\right) \\cdot p\\left(z_n = k \\mid \\lambda \\right)}\n     {\\sum_{k' = 1}^K p\\left(y_n \\mid z_n = k', \\mu, \\sigma \\right)\n                    \\cdot p\\left(z_n = k' \\mid \\lambda \\right)}.\n\\]\nOn the log scale, the normalized probability is computed as \\[\\begin{align*}\n\\log\\,&\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda\\right] \\\\\n&= \\log p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right) + \\log \\Pr\\!\\left[z_n = k \\mid \\lambda\\right] \\\\\n&\\quad  - \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{k' = 1}^K \\big(\\log p\\left(y_n \\mid z_n = k', \\mu, \\sigma\\right)\n                                                 + \\log p\\left(z_n = k' \\mid \\lambda\\right)\\big).\n\\end{align*}\\] This can be coded up directly in Stan; the change-point model in the change point section provides an example.\n\n\n\nGiven the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.\ndata {\n  int&lt;lower=1&gt; K;          // number of mixture components\n  int&lt;lower=1&gt; N;          // number of data points\n  array[N] real y;         // observations\n}\nparameters {\n  simplex[K] theta;          // mixing proportions\n  ordered[K] mu;             // locations of mixture components\n  vector&lt;lower=0&gt;[K] sigma;  // scales of mixture components\n}\nmodel {\n  vector[K] log_theta = log(theta);  // cache log calculation\n  sigma ~ lognormal(0, 2);\n  mu ~ normal(0, 10);\n  for (n in 1:N) {\n    vector[K] lps = log_theta;\n    for (k in 1:K) {\n      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);\n    }\n    target += log_sum_exp(lps);\n  }\n}\nThe model involves K mixture components and N data points. The mixing proportion parameter theta is declared to be a unit \\(K\\)-simplex, whereas the component location parameter mu and scale parameter sigma are both defined to be K-vectors.\nThe location parameter mu is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components mu[k] is symmetric, as it is here (each component has an independent \\(\\textsf{normal}(0, 10)\\) prior). It would even be possible to include a hierarchical prior for the components.\nThe values in the scale array sigma are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.\nThe model declares a local array variable lps to be size K and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points n. For each such point, the log of \\(\\theta_k \\times\n\\textsf{normal}\\left(y_n \\mid \\mu_k,\\sigma_k\\right)\\) is calculated and added to the array lps. Then the log probability is incremented with the log sum of exponentials of those values.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#vectorizing-mixtures",
+    "href": "stan-users-guide/finite-mixtures.html#vectorizing-mixtures",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "There is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that lambda, y[n], mu[1], mu[2], and sigma[1], sigma[2] are all scalars and lambda is between 0 and 1.\nfor (n in 1:N) {\n  target += log_sum_exp(log(lambda)\n                          + normal_lpdf(y[n] | mu[1], sigma[1]),\n                        log1m(lambda)\n                          + normal_lpdf(y[n] | mu[2], sigma[2]));\nor equivalently\nfor (n in 1:N) {\n  target += log_mix(lambda,\n                    normal_lpdf(y[n] | mu[1], sigma[1]),\n                    normal_lpdf(y[n] | mu[2], sigma[2]))\n  };\nThis definition assumes that each observation \\(y_n\\) may have arisen from either of the mixture components. The density is \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma\\right)\n= \\prod_{n=1}^N \\big(\\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1 \\right)\n                 + (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2 \\right)\\big).\n\\]\nContrast the previous model with the following (erroneous) attempt to vectorize the model.\ntarget += log_sum_exp(log(lambda)\n                        + normal_lpdf(y | mu[1], sigma[1]),\n                      log1m(lambda)\n                        + normal_lpdf(y | mu[2], sigma[2]));\nor equivalently,\ntarget += log_mix(lambda,\n                  normal_lpdf(y | mu[1], sigma[1]),\n                  normal_lpdf(y | mu[2], sigma[2]));\nThis second definition implies that the entire sequence \\(y_1, \\dotsc, y_n\\) of observations comes form one component or the other, defining a different density, \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n= \\lambda \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right).\n\\]",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#mixture-inference.section",
+    "href": "stan-users-guide/finite-mixtures.html#mixture-inference.section",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "In many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.\nWe have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs. female, or Democrat vs. Republican).\nIn other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.\n\n\nAs an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\), \\[\\begin{align*}\n\\mu_1, \\mu_2       &\\sim \\textsf{normal}(0, 10) \\\\\n\\sigma_1, \\sigma_2 &\\sim \\textsf{halfnormal}(0, 10) \\\\\n\\end{align*}\\]\nThe prior on the mixture ratio is uniform, \\[\n\\lambda \\sim \\textsf{uniform}(0, 1),\n\\] so that with the likelihood \\[\np\\left(y_n \\mid \\mu, \\sigma\\right)\n= \\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right),\n\\] the joint distribution \\(p(y, \\mu, \\sigma, \\lambda)\\) is exchangeable in the parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\) with \\(\\lambda\\) flipping to \\(1 - \\lambda\\).1\n\n\n\nIn cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.\n\n\nPosterior predictive distribution for a new observation \\(\\tilde{y}\\) given the complete parameter vector \\(\\theta\\) will be \\[\np(\\tilde{y} \\mid y)\n=\n\\int_{\\theta}\np(\\tilde{y} \\mid \\theta)\n\\, p(\\theta \\mid y)\n\\, \\textsf{d}\\theta.\n\\]\nThe normal mixture example from the previous section, with \\(\\theta =\n(\\mu, \\sigma, \\lambda)\\), shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing \\(p(\\tilde{y} \\mid y)\\), which is more efficient statistically in terms of effective sample size, or simulating draws of \\(\\tilde{y}\\), which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] y_tilde;\n  // ...\n}\ngenerated quantities {\n  vector[N_tilde] log_p_y_tilde;\n  for (n in 1:N_tilde) {\n    log_p_y_tilde[n]\n      = log_mix(lambda,\n                normal_lpdf(y_tilde[n] | mu[1], sigma[1])\n                normal_lpdf(y_tilde[n] | mu[2], sigma[2]));\n  }\n}\nIt is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply log_sum_exp of the posterior draws of log_p_y_tilde. The average log predictive density is then given by subtracting log(N_new).\n\n\n\nOften a mixture model will be applied to a clustering problem and there might be two data items \\(y_i\\) and \\(y_j\\) for which there is a question of whether they arose from the same mixture component. If we take \\(z_i\\) and \\(z_j\\) to be the component responsibility discrete variables, then the quantity of interest is \\(z_i = z_j\\), which can be summarized as an event probability \\[\n\\Pr[z_i = z_j \\mid y]\n=\n\\int_{\\theta}\n\\frac{\\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \\mid \\theta)}\n     {\\sum_{k=0}^1 \\sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \\mid \\theta)}\n\\,\np(\\theta \\mid y)\n\\,\n\\textsf{d}\\theta.\n\\]\nAs with other event probabilities, this can be calculated in the generated quantities block either by sampling \\(z_i\\) and \\(z_j\\) and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#zero-inflated.section",
+    "href": "stan-users-guide/finite-mixtures.html#zero-inflated.section",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "Zero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by Lambert (1992), add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.\nZero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.\n\n\nConsider the following example for zero-inflated Poisson distributions. There is a probability \\(\\theta\\) of observing a zero, and a probability \\(1 - \\theta\\) of observing a count with a \\(\\textsf{Poisson}(\\lambda)\\) distribution (now \\(\\theta\\) is being used for mixing proportions because \\(\\lambda\\) is the traditional notation for a Poisson mean parameter). Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as \\[\\begin{align*}\ny_n & = 0 & \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}(y_n \\mid \\lambda) & \\quad\\text{with probability } 1-\\theta.\n\\end{align*}\\]\nStan does not support conditional distribution statements (with ~) conditional on some parameter, and we need to consider the corresponding likelihood \\[\np(y_n \\mid \\theta,\\lambda)\n=\n\\begin{cases}\n\\theta + (1 - \\theta) \\times \\textsf{Poisson}(0 \\mid \\lambda) & \\quad\\text{if } y_n = 0, \\text{ and}\\\\\n(1-\\theta) \\times \\textsf{Poisson}(y_n \\mid \\lambda) &\\quad\\text{if } y_n &gt; 0.\n\\end{cases}\n\\] The log likelihood can be coded directly in Stan (with target +=) as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  for (n in 1:N) {\n    if (y[n] == 0) {\n      target += log_sum_exp(log(theta),\n                            log1m(theta)\n                              + poisson_lpmf(y[n] | lambda));\n    } else {\n      target += log1m(theta)\n                  + poisson_lpmf(y[n] | lambda);\n    }\n  }\n}\nThe log1m(theta) computes log(1-theta), but is more computationally stable. The log_sum_exp(lp1,lp2) function adds the log probabilities on the linear scale; it is defined to be equal to log(exp(lp1) + exp(lp2)), but is more computationally stable and faster.\n\n\nThe code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also poisson_lpmf(0 | lambda) every time the first condition body executes. The use of the redundant terms is conditioned on y, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.\nThe number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare y_nonzero, so it must be computed in a function.\nfunctions {\n  int num_zeros(array[] int y) {\n    int sum = 0;\n    for (n in 1:size(y)) {\n      sum += (y[n] == 0);\n    }\n    return sum;\n  }\n}\n// ...\ntransformed data {\n  int&lt;lower=0&gt; N_zero = num_zeros(y);\n  array[N - N_zero] int&lt;lower=1&gt; y_nonzero;\n  int N_nonzero = 0;\n  for (n in 1:N) {\n    if (y[n] == 0) continue;\n    N_nonzero += 1;\n    y_nonzero[N_nonzero] = y[n];\n  }\n}\n// ...\nmodel {\n  // ...\n   target\n     += N_zero\n          * log_sum_exp(log(theta),\n                        log1m(theta)\n                          + poisson_lpmf(0 | lambda));\n   target += N_nonzero * log1m(theta);\n   target += poisson_lpmf(y_nonzero | lambda);\n  // ...\n}\nThe boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if y_nonzero is empty, N_nonzero will be zero, and the last two target increment terms will add zeros.\n\n\n\n\nThe hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as [ \\[\\begin{align*}\ny_n & = 0 \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}_{x\\neq 0}(y_n \\mid \\lambda) \\quad\\text{with probability } 1-\\theta,\n\\end{align*}\\] ] Where \\(\\textsf{Poisson}_{x\\neq 0}\\) is a truncated Poisson distribution, truncated at \\(0\\).\nThe corresponding likelihood function for the hurdle model is defined by \\[\np(y\\mid\\theta,\\lambda)\n=\n\\begin{cases}\n\\theta &\\quad\\text{if } y = 0, \\text{ and}\\\\\n(1 - \\theta)\n   \\frac{\\displaystyle \\textsf{Poisson}(y \\mid \\lambda)}\n        {\\displaystyle  1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)}\n&\\quad\\text{if } y &gt; 0,\n\\end{cases}\n\\] where \\(\\textsf{PoissonCDF}\\) is the cumulative distribution function for the Poisson distribution and and \\(1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)\\) is the relative normalization term for the truncated Poisson (truncated at \\(0\\)).\nThe hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.\nif (y[n] == 0) {\n  target += log(theta);\n} else {\n  target += log1m(theta) + poisson_lpmf(y[n] | lambda)\n            - poisson_lccdf(0 | lambda));\n}\nJulian King pointed out that because \\[\\begin{align*}\n\\log \\left( 1 - \\textsf{PoissonCDF}(0 \\mid \\lambda) \\right)\n  &= \\log \\left( 1 - \\textsf{Poisson}(0 \\mid \\lambda) \\right) \\\\\n  &= \\log(1 - \\exp(-\\lambda))\n\\end{align*}\\] the CCDF in the else clause can be replaced with a simpler expression.\ntarget += log1m(theta) + poisson_lpmf(y[n] | lambda)\n          - log1m_exp(-lambda));\nThe resulting code is about 15% faster than the code with the CCDF.\nThis is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size \\(N=200\\) and parameters \\(\\theta=0.3\\) and \\(\\lambda = 8\\), the speedup is a factor of 10; it will be lower for smaller \\(N\\) and greater for larger \\(N\\); it will also be greater for larger \\(\\theta\\).\nTo achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,\nfunctions {\n  int num_zero(array[] int y) {\n    int nz = 0;\n    for (n in 1:size(y)) {\n      if (y[n] == 0) {\n        nz += 1;\n      }\n    }\n    return nz;\n  }\n}\nThen a transformed data block can be used to store the sufficient statistics,\ntransformed data {\n  int&lt;lower=0, upper=N&gt; N0 = num_zero(y);\n  int&lt;lower=0, upper=N&gt; Ngt0 = N - N0;\n  array[N - num_zero(y)] int&lt;lower=1&gt; y_nz;\n  {\n    int pos = 1;\n    for (n in 1:N) {\n      if (y[n] != 0) {\n        y_nz[pos] = y[n];\n        pos += 1;\n      }\n    }\n  }\n}\nThe model block is then reduced to three statements.\nmodel {\n  N0 ~ binomial(N, theta);\n  y_nz ~ poisson(lambda);\n  target += -Ngt0 * log1m_exp(-lambda);\n}\nThe first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant Ngt0; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#priors-and-effective-data-size-in-mixture-models",
+    "href": "stan-users-guide/finite-mixtures.html#priors-and-effective-data-size-in-mixture-models",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "Suppose we have a two-component mixture model with mixing rate \\(\\lambda \\in (0, 1)\\). Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are \\(N\\) observations, the mixture components will be estimated with effective data sizes of \\(\\theta \\, N\\) and \\((1 - \\theta)\n\\, N\\) for the two components for some \\(\\theta \\in (0, 1)\\). The effective weighting size is determined by posterior responsibility, not simply by the mixing rate \\(\\lambda\\).\n\n\nIn contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data \\(y\\).\nIf different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see Hoeting et al. (1999) and Gelman et al. (2013) for theoretical and computational details).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/finite-mixtures.html#footnotes",
+    "href": "stan-users-guide/finite-mixtures.html#footnotes",
+    "title": "Finite Mixtures",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nImposing a constraint such as \\(\\theta &lt; 0.5\\) will resolve the symmetry, but fundamentally changes the model and its posterior inferences.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Finite Mixtures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html",
+    "href": "stan-users-guide/for-bugs-users.html",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "From the outside, Stan and BUGS1 are similar—they use statistically-themed modeling languages (which are similar but with some differences; see below), they can be called from R, running some specified number of chains to some specified length, producing posterior simulations that can be assessed using standard convergence diagnostics. This is not a coincidence: in designing Stan: we wanted to keep many of the useful features of Bugs.\n\n\n\n\nStan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).\n\n\n\nBUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.\n\n\n\nBUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).\n\n\n\nThe BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).\n\n\n\nIn BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets mu before using it to sample y:\nmu = a + b * x;\ny ~ normal(mu, sigma);\ntranslates to the following C++ code:\nmu = a + b * x;\ntarget += normal_lpdf(y | mu, sigma);\nContrast this with the following Stan program:\ny ~ normal(mu, sigma);\nmu = a + b * x;\nThis program is well formed, but is almost certainly a coding error, because it attempts to use mu before it is set. The direct translation to C++ code highlights the potential error of using mu in the first statement:\ntarget += normal_lpdf(y | mu, sigma);\nmu = a + b * x;\nTo trap these kinds of errors, variables are initialized to the special not-a-number (NaN) value. If NaN is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.\n\n\n\nStan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.\n\n\n\nBoth BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (\\(\\hat{R}\\)) and the effective sample size to judge when to stop (stopping when \\(\\hat{R}\\) values do not counter-indicate convergence and when high enough effective sample size have been obtained).\n\n\n\nWinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.\n\n\n\nLike WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.\n\n\n\nLike OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.\n\n\n\n\nThe BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:\ny ~ normal(mu, sigma);\nand\nfor (i in 1:n) y[i] ~ normal(mu, sigma);\nOr, equivalently (recall that a line break is just another form of whitespace),\nfor (i in 1:n)\n  y[i] ~ normal(mu, sigma);\nand also equivalently,\nfor (i in 1:n) {\n  y[i] ~ normal(mu, sigma);\n}\nThere’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.\nIn Stan, variables can have names constructed using letters, numbers, and the underscore (_) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (.) symbol.\nIn Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in Bayesian Data Analysis) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is normal(1,2), not normal(1,4) or normal(1,0.25).\nSimilarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.\nThe distributions have slightly different names:\n\n\n\nBUGS\nStan\n\n\n\n\ndnorm\nnormal\n\n\ndbinom\nbinomial\n\n\ndpois\npoisson\n\n\n…\n…\n\n\n\nStan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.\n{\n  total = 0;\n  for (i in 1:n) {\n    theta[i] ~ normal(total, sigma);\n    total = total + theta[i];\n  }\n}\nIn BUGS, the above model would not be legal because the variable total is defined more than once. But in Stan, the loop is executed in order, so total is overwritten in each step.\nStan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.\nIn Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.\nStan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,2 allowing\nfor (i in 1:n) {\n  y[i] ~ normal(mu[i], sigma[i]);\n}\nto be expressed more compactly as\ny ~ normal(mu, sigma);\nThe vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.\nStan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of K parameters for each of J groups; this can be declared using\narray[J] vector[K] theta;\nThen theta[j] is an expression denoting a K-vector and may be used in the code just like any other vector variable.\nAn alternative encoding would be with a two-dimensional array, as in\narray[J, K] real theta;\nThe vector version can have some advantages, both in convenience and in computational speed for some operations.\nA third encoding would use a matrix:\nmatrix[J, K] theta;\nbut in this case, theta[j] is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in theta[j]', may be used to convert the row vector theta[j] to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.\nStan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the zero inflation section.\nStan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop\nmodel {\n    // ...\n    for (n in 1:N) {\n        // ... do something with n ....\n    }\n}\nmay be recoded using the following while loop.\nmodel {\n    int n;\n    // ...\n    n = 1;\n    while (n &lt;= N) {\n        // ... do something with n ...\n        n = n + 1;\n    }\n}\n\n\n\nStan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.\nStan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in Lewandowski, Kurowicka, and Joe (2009). In particular, the Lewandowski et al. prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.\nIn BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter p declared as\nreal&lt;lower=0, upper=1&gt; p;\nand then have no distribution statement for p in the model block, then you are implicitly assigning a uniform \\([0,1]\\) prior on p.\nOn the other hand, if you have a parameter theta declared with\nreal theta;\nand have no distribution statement for theta in the model block, then you are implicitly assigning an improper uniform prior on \\((-\\infty,\\infty)\\) to theta.\nBUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:\nparameters {\n  real theta;\n}\nmodel { }\nAlthough parameters in Stan models may have improper priors, we do not want improper posterior distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.\nEach statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:\ny ~ normal(0,1);\ny ~ normal(2,3);\nThe second line here does not simply overwrite the first; rather, both statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, \\(\\textsf{normal}(y \\mid 0,1) * \\textsf{normal}(y \\mid 2,3)\\), into the density function.\nFor a perhaps more confusing example, consider the following two lines in a Stan model:\nx ~ normal(0.8 * y, sigma);\ny ~ normal(0.8 * x, sigma);\nAt first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are not interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, \\(\\textsf{normal}(x \\mid 0.8y,\\sigma)\n\\times \\textsf{normal}(y \\mid 0.8x,\\sigma)\\), which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.\nWith censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in Gelman et al. (2013) and Gelman and Hill (2007) may be implemented in Stan directly as written.\nStan, like BUGS, can benefit from human intervention in the form of reparameterization.\n\n\n\nStan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.\nIn practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.\nWhen you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)\nWhen running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable x from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.\nAs in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using plot() and print(). You can access the simulations themselves using various extractor functions, as described in the RStan documentation.\nVarious information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.\n\n\n\nStan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the Stan web site for information on subscribing and posting and to look at archives.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-how-bugs-and-stan-work",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-how-bugs-and-stan-work",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "Stan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).\n\n\n\nBUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.\n\n\n\nBUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).\n\n\n\nThe BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).\n\n\n\nIn BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets mu before using it to sample y:\nmu = a + b * x;\ny ~ normal(mu, sigma);\ntranslates to the following C++ code:\nmu = a + b * x;\ntarget += normal_lpdf(y | mu, sigma);\nContrast this with the following Stan program:\ny ~ normal(mu, sigma);\nmu = a + b * x;\nThis program is well formed, but is almost certainly a coding error, because it attempts to use mu before it is set. The direct translation to C++ code highlights the potential error of using mu in the first statement:\ntarget += normal_lpdf(y | mu, sigma);\nmu = a + b * x;\nTo trap these kinds of errors, variables are initialized to the special not-a-number (NaN) value. If NaN is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.\n\n\n\nStan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.\n\n\n\nBoth BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (\\(\\hat{R}\\)) and the effective sample size to judge when to stop (stopping when \\(\\hat{R}\\) values do not counter-indicate convergence and when high enough effective sample size have been obtained).\n\n\n\nWinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.\n\n\n\nLike WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.\n\n\n\nLike OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-the-modeling-languages",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-the-modeling-languages",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "The BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:\ny ~ normal(mu, sigma);\nand\nfor (i in 1:n) y[i] ~ normal(mu, sigma);\nOr, equivalently (recall that a line break is just another form of whitespace),\nfor (i in 1:n)\n  y[i] ~ normal(mu, sigma);\nand also equivalently,\nfor (i in 1:n) {\n  y[i] ~ normal(mu, sigma);\n}\nThere’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.\nIn Stan, variables can have names constructed using letters, numbers, and the underscore (_) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (.) symbol.\nIn Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in Bayesian Data Analysis) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is normal(1,2), not normal(1,4) or normal(1,0.25).\nSimilarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.\nThe distributions have slightly different names:\n\n\n\nBUGS\nStan\n\n\n\n\ndnorm\nnormal\n\n\ndbinom\nbinomial\n\n\ndpois\npoisson\n\n\n…\n…\n\n\n\nStan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.\n{\n  total = 0;\n  for (i in 1:n) {\n    theta[i] ~ normal(total, sigma);\n    total = total + theta[i];\n  }\n}\nIn BUGS, the above model would not be legal because the variable total is defined more than once. But in Stan, the loop is executed in order, so total is overwritten in each step.\nStan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.\nIn Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.\nStan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,2 allowing\nfor (i in 1:n) {\n  y[i] ~ normal(mu[i], sigma[i]);\n}\nto be expressed more compactly as\ny ~ normal(mu, sigma);\nThe vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.\nStan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of K parameters for each of J groups; this can be declared using\narray[J] vector[K] theta;\nThen theta[j] is an expression denoting a K-vector and may be used in the code just like any other vector variable.\nAn alternative encoding would be with a two-dimensional array, as in\narray[J, K] real theta;\nThe vector version can have some advantages, both in convenience and in computational speed for some operations.\nA third encoding would use a matrix:\nmatrix[J, K] theta;\nbut in this case, theta[j] is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in theta[j]', may be used to convert the row vector theta[j] to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.\nStan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the zero inflation section.\nStan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop\nmodel {\n    // ...\n    for (n in 1:N) {\n        // ... do something with n ....\n    }\n}\nmay be recoded using the following while loop.\nmodel {\n    int n;\n    // ...\n    n = 1;\n    while (n &lt;= N) {\n        // ... do something with n ...\n        n = n + 1;\n    }\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-the-statistical-models-that-are-allowed",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-the-statistical-models-that-are-allowed",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "Stan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.\nStan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in Lewandowski, Kurowicka, and Joe (2009). In particular, the Lewandowski et al. prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.\nIn BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter p declared as\nreal&lt;lower=0, upper=1&gt; p;\nand then have no distribution statement for p in the model block, then you are implicitly assigning a uniform \\([0,1]\\) prior on p.\nOn the other hand, if you have a parameter theta declared with\nreal theta;\nand have no distribution statement for theta in the model block, then you are implicitly assigning an improper uniform prior on \\((-\\infty,\\infty)\\) to theta.\nBUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:\nparameters {\n  real theta;\n}\nmodel { }\nAlthough parameters in Stan models may have improper priors, we do not want improper posterior distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.\nEach statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:\ny ~ normal(0,1);\ny ~ normal(2,3);\nThe second line here does not simply overwrite the first; rather, both statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, \\(\\textsf{normal}(y \\mid 0,1) * \\textsf{normal}(y \\mid 2,3)\\), into the density function.\nFor a perhaps more confusing example, consider the following two lines in a Stan model:\nx ~ normal(0.8 * y, sigma);\ny ~ normal(0.8 * x, sigma);\nAt first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are not interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, \\(\\textsf{normal}(x \\mid 0.8y,\\sigma)\n\\times \\textsf{normal}(y \\mid 0.8x,\\sigma)\\), which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.\nWith censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in Gelman et al. (2013) and Gelman and Hill (2007) may be implemented in Stan directly as written.\nStan, like BUGS, can benefit from human intervention in the form of reparameterization.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-when-running-from-r",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-when-running-from-r",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "Stan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.\nIn practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.\nWhen you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)\nWhen running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable x from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.\nAs in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using plot() and print(). You can access the simulations themselves using various extractor functions, as described in the RStan documentation.\nVarious information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html#the-stan-community",
+    "href": "stan-users-guide/for-bugs-users.html#the-stan-community",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "Stan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the Stan web site for information on subscribing and posting and to look at archives.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/for-bugs-users.html#footnotes",
+    "href": "stan-users-guide/for-bugs-users.html#footnotes",
+    "title": "Transitioning from BUGS",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nExcept where otherwise noted, we use “BUGS” to refer to WinBUGS, OpenBUGS, and JAGS, indiscriminately.↩︎\nMost distributions have been vectorized, but currently the truncated versions may not exist.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Transitioning from BUGS"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/hyperspherical-models.html",
+    "href": "stan-users-guide/hyperspherical-models.html",
+    "title": "Directions, Rotations, and Hyperspheres",
+    "section": "",
+    "text": "Directional statistics involve data and/or parameters that are constrained to be directions. The set of directions forms a sphere, the geometry of which is not smoothly mappable to that of a Euclidean space because you can move around a sphere and come back to where you started. This is why it is impossible to make a map of the globe on a flat piece of paper where all points that are close to each other on the globe are close to each other on the flat map. The fundamental problem is easy to visualize in two dimensions, because as you move around a circle, you wind up back where you started. In other words, 0 degrees and 360 degrees (equivalently, 0 and \\(2 \\pi\\) radians) pick out the same point, and the distance between 359 degrees and 2 degrees is the same as the distance between 137 and 140 degrees.\nStan supports directional statistics by providing a unit-vector data type, the values of which determine points on a hypersphere (circle in two dimensions, sphere in three dimensions).\n\n\nThe length of a vector \\(x \\in \\mathbb{R}^K\\) is given by \\[\n\\Vert x \\Vert\n= \\sqrt{x^{\\top}\\,x}\n= \\sqrt{x_1^2 + x_2^2 + \\cdots + x_K^2}.\n\\] Unit vectors are defined to be vectors of unit length (i.e., length one).\nWith a variable declaration such as\nunit_vector[K] x;\nthe value of x will be constrained to be a vector of size K with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.\nWarning: An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (lp__ in some interfaces). The underlying source of the problem is that a unit vector of size \\(K\\) has only \\(K - 1\\) degrees of freedom. But there is no way to map those \\(K - 1\\) degrees of freedom continuously to \\(\\mathbb{R}^N\\)—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses \\(K\\) unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.\n\n\n\nAn \\(n\\)-sphere, written \\(S^{n}\\), is defined as the set of \\((n +\n1)\\)-dimensional unit vectors, \\[\nS^{n} = \\left\\{ x \\in \\mathbb{R}^{n+1} \\: : \\: \\Vert x \\Vert = 1 \\right\\}.\n\\]\nEven though \\(S^n\\) is made up of points in \\((n+1)\\) dimensions, it is only an \\(n\\)-dimensional manifold. For example, \\(S^2\\) is defined as a set of points in \\(\\mathbb{R}^3\\), but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by \\(S^2\\) in \\(\\mathbb{R}^3\\) behaves locally like a plane, i.e., \\(\\mathbb{R}^2\\). However, the overall shape of \\(S^2\\) is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (\\(S^2\\)) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.\nEven though \\(S^{n-1}\\) behaves locally like \\(\\mathbb{R}^{n-1}\\), there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at \\(2\\pi\\) radians in natural units), they do not produce a smooth map.\nLike a bounded interval \\((a, b)\\), in geometric terms, a sphere is compact in that the distance between any two points is bounded.\n\n\n\nStan (inverse) transforms arbitrary points in \\(\\mathbb{R}^{K+1}\\) to points in \\(S^K\\) using the auxiliary variable approach of Muller (1959). A point \\(y \\in \\mathbb{R}^K\\) is transformed to a point \\(x \\in S^{K-1}\\) by \\[\nx = \\frac{y}{\\sqrt{y^{\\top} y}}.\n\\]\nThe problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. Muller (1959) introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nUnit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle \\(\\theta\\). Given an angle \\(\\theta\\), a matrix can be defined, the pre-multiplication by which rotates a point by an angle of \\(\\theta\\). For angle \\(\\theta\\) (in two dimensions), the \\(2 \\times 2\\) rotation matrix is defined by \\[\nR_{\\theta}\n=\n\\begin{bmatrix}\n\\cos \\theta & -\\sin \\theta \\\\\n\\sin \\theta &  \\cos \\theta\n\\end{bmatrix}.\n\\] Given a two-dimensional vector \\(x\\), \\(R_{\\theta} \\, x\\) is the rotation of \\(x\\) (around the origin) by \\(\\theta\\) degrees.\n\n\nAngles can be calculated from unit vectors. For example, a random variable theta representing an angle in \\((-\\pi, \\pi)\\) radians can be declared as a two-dimensional unit vector then transformed to an angle.\nparameters {\n  unit_vector[2] xy;\n}\ntransformed parameters {\n  real&lt;lower=-pi(), upper=pi()&gt; theta = atan2(xy[2], xy[1]);\n}\nIf the distribution of \\((x, y)\\) is uniform over a circle, then the distribution of \\(\\arctan \\frac{y}{x}\\) is uniform over \\((-\\pi, \\pi)\\).\nIt might be tempting to try to just declare theta directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values \\(-\\pi\\) and \\(\\pi\\) are at \\(-\\infty\\) and \\(\\infty\\) on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.\nWith a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.\n\n\n\n\nA 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.\nIn human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/hyperspherical-models.html#unit-vectors",
+    "href": "stan-users-guide/hyperspherical-models.html#unit-vectors",
+    "title": "Directions, Rotations, and Hyperspheres",
+    "section": "",
+    "text": "The length of a vector \\(x \\in \\mathbb{R}^K\\) is given by \\[\n\\Vert x \\Vert\n= \\sqrt{x^{\\top}\\,x}\n= \\sqrt{x_1^2 + x_2^2 + \\cdots + x_K^2}.\n\\] Unit vectors are defined to be vectors of unit length (i.e., length one).\nWith a variable declaration such as\nunit_vector[K] x;\nthe value of x will be constrained to be a vector of size K with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.\nWarning: An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (lp__ in some interfaces). The underlying source of the problem is that a unit vector of size \\(K\\) has only \\(K - 1\\) degrees of freedom. But there is no way to map those \\(K - 1\\) degrees of freedom continuously to \\(\\mathbb{R}^N\\)—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses \\(K\\) unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/hyperspherical-models.html#circles-spheres-and-hyperspheres",
+    "href": "stan-users-guide/hyperspherical-models.html#circles-spheres-and-hyperspheres",
+    "title": "Directions, Rotations, and Hyperspheres",
+    "section": "",
+    "text": "An \\(n\\)-sphere, written \\(S^{n}\\), is defined as the set of \\((n +\n1)\\)-dimensional unit vectors, \\[\nS^{n} = \\left\\{ x \\in \\mathbb{R}^{n+1} \\: : \\: \\Vert x \\Vert = 1 \\right\\}.\n\\]\nEven though \\(S^n\\) is made up of points in \\((n+1)\\) dimensions, it is only an \\(n\\)-dimensional manifold. For example, \\(S^2\\) is defined as a set of points in \\(\\mathbb{R}^3\\), but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by \\(S^2\\) in \\(\\mathbb{R}^3\\) behaves locally like a plane, i.e., \\(\\mathbb{R}^2\\). However, the overall shape of \\(S^2\\) is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (\\(S^2\\)) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.\nEven though \\(S^{n-1}\\) behaves locally like \\(\\mathbb{R}^{n-1}\\), there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at \\(2\\pi\\) radians in natural units), they do not produce a smooth map.\nLike a bounded interval \\((a, b)\\), in geometric terms, a sphere is compact in that the distance between any two points is bounded.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/hyperspherical-models.html#transforming-to-unconstrained-parameters",
+    "href": "stan-users-guide/hyperspherical-models.html#transforming-to-unconstrained-parameters",
+    "title": "Directions, Rotations, and Hyperspheres",
+    "section": "",
+    "text": "Stan (inverse) transforms arbitrary points in \\(\\mathbb{R}^{K+1}\\) to points in \\(S^K\\) using the auxiliary variable approach of Muller (1959). A point \\(y \\in \\mathbb{R}^K\\) is transformed to a point \\(x \\in S^{K-1}\\) by \\[\nx = \\frac{y}{\\sqrt{y^{\\top} y}}.\n\\]\nThe problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. Muller (1959) introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/hyperspherical-models.html#unit-vectors-and-rotations",
+    "href": "stan-users-guide/hyperspherical-models.html#unit-vectors-and-rotations",
+    "title": "Directions, Rotations, and Hyperspheres",
+    "section": "",
+    "text": "Unit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle \\(\\theta\\). Given an angle \\(\\theta\\), a matrix can be defined, the pre-multiplication by which rotates a point by an angle of \\(\\theta\\). For angle \\(\\theta\\) (in two dimensions), the \\(2 \\times 2\\) rotation matrix is defined by \\[\nR_{\\theta}\n=\n\\begin{bmatrix}\n\\cos \\theta & -\\sin \\theta \\\\\n\\sin \\theta &  \\cos \\theta\n\\end{bmatrix}.\n\\] Given a two-dimensional vector \\(x\\), \\(R_{\\theta} \\, x\\) is the rotation of \\(x\\) (around the origin) by \\(\\theta\\) degrees.\n\n\nAngles can be calculated from unit vectors. For example, a random variable theta representing an angle in \\((-\\pi, \\pi)\\) radians can be declared as a two-dimensional unit vector then transformed to an angle.\nparameters {\n  unit_vector[2] xy;\n}\ntransformed parameters {\n  real&lt;lower=-pi(), upper=pi()&gt; theta = atan2(xy[2], xy[1]);\n}\nIf the distribution of \\((x, y)\\) is uniform over a circle, then the distribution of \\(\\arctan \\frac{y}{x}\\) is uniform over \\((-\\pi, \\pi)\\).\nIt might be tempting to try to just declare theta directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values \\(-\\pi\\) and \\(\\pi\\) are at \\(-\\infty\\) and \\(\\infty\\) on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.\nWith a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/hyperspherical-models.html#circular-representations-of-days-and-years",
+    "href": "stan-users-guide/hyperspherical-models.html#circular-representations-of-days-and-years",
+    "title": "Directions, Rotations, and Hyperspheres",
+    "section": "",
+    "text": "A 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.\nIn human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html",
+    "href": "stan-users-guide/latent-discrete.html",
+    "title": "Latent Discrete Parameters",
+    "section": "",
+    "text": "Stan does not support sampling discrete parameters. So it is not possible to directly translate BUGS or JAGS models with discrete parameters (i.e., discrete stochastic nodes). Nevertheless, it is possible to code many models that involve bounded discrete parameters by marginalizing out the discrete parameters.1\nThis chapter shows how to code several widely-used models involving latent discrete parameters. The next chapter, the clustering chapter, on clustering models, considers further models involving latent discrete parameters.\n\n\nAlthough it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.\nStandard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.\n\n\n\nThe first example is a model of coal mining disasters in the U.K. for the years 1851–1962.2\n\n\nFonnesbeck et al. (2013, sec. 3.1) provides a Poisson model of disaster \\(D_t\\) in year \\(t\\) with two rate parameters, an early rate (\\(e\\)) and late rate (\\(l\\)), that change at a given point in time \\(s\\). The full model expressed using a latent discrete parameter \\(s\\) is \\[\\begin{align*}\ne   &\\sim  \\textsf{exponential}(r_e) \\\\\nl   &\\sim  \\textsf{exponential}(r_l) \\\\\ns   &\\sim  \\textsf{uniform}(1, T) \\\\\nD_t &\\sim  \\textsf{Poisson}(t &lt; s \\; ? \\; e \\: : \\: l)\n\\end{align*}\\]\nThe last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.3\nIt uses a compact notation involving separating its three arguments by a question mark (?) and a colon (:). The conditional operator is defined by \\[\nc \\; ? \\; x_1 \\: : \\: x_2\n=\n\\begin{cases}\n\\ x_1 & \\quad\\text{if } c \\text{ is true (i.e., non-zero), and} \\\\\n\\ x_2 & \\quad\\text{if } c \\text{ is false (i.e., zero).}\n\\end{cases}\n\\]\n\n\n\nTo code this model in Stan, the discrete parameter \\(s\\) must be marginalized out to produce a model defining the log of the probability function \\(p(e,l,D_t)\\). The full joint probability factors as \\[\\begin{align*}\np(e,l,s,D) &=  p(e) \\, p(l) \\, p(s) \\, p(D \\mid s, e, l) \\\\\n&= \\textsf{exponential}(e \\mid r_e) \\ \\textsf{exponential}(l \\mid r_l) \\,\n    \\textsf{uniform}(s \\mid 1, T) \\\\\n& \\qquad \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nTo marginalize, an alternative factorization into prior and likelihood is used, \\[\np(e,l,D) = p(e,l) \\, p(D \\mid e,l),\n\\]\nwhere the likelihood is defined by marginalizing \\(s\\) as \\[\\begin{align*}\np(D \\mid e,l) &= \\sum_{s=1}^T p(s, D \\mid e,l) \\\\\n&= \\sum_{s=1}^T p(s) \\, p(D \\mid s,e,l) \\\\\n&= \\sum_{s=1}^T \\textsf{uniform}(s \\mid 1,T) \\,\n    \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nStan operates on the log scale and thus requires the log likelihood, \\[\\begin{align*}\n\\log p(D \\mid e,l)\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{s=1}^T\n    \\left( \\log \\textsf{uniform}(s \\mid 1, T) \\vphantom{\\sum_{t=1}^T}\\right. \\\\\n&\\qquad \\left.\n    + \\sum_{t=1}^T \\log \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l)\n\\right),\n\\end{align*}\\] where the log sum of exponents function is defined by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{n=1}^N \\, \\alpha_n =\n\\log \\sum_{n=1}^N \\exp(\\alpha_n).\n\\]\nThe log sum of exponents function allows the model to be coded directly in Stan using the built-in function log_sum_exp, which provides both arithmetic stability and efficiency for mixture model calculations.\n\n\n\nThe Stan program for the change point model is shown in the figure below. The transformed parameter lp[s] stores the quantity \\(\\log p(s, D \\mid e, l)\\).\ndata {\n  real&lt;lower=0&gt; r_e;\n  real&lt;lower=0&gt; r_l;\n\n  int&lt;lower=1&gt; T;\n  array[T] int&lt;lower=0&gt; D;\n}\ntransformed data {\n  real log_unif;\n  log_unif = -log(T);\n}\nparameters {\n  real&lt;lower=0&gt; e;\n  real&lt;lower=0&gt; l;\n}\ntransformed parameters {\n  vector[T] lp;\n  lp = rep_vector(log_unif, T);\n  for (s in 1:T) {\n    for (t in 1:T) {\n      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);\n    }\n  }\n}\nmodel {\n  e ~ exponential(r_e);\n  l ~ exponential(r_l);\n  target += log_sum_exp(lp);\n}\nA change point model in which disaster rates D[t] have one rate, e, before the change point and a different rate, l, after the change point. The change point itself, s, is marginalized out as described in the text.\nAlthough the change-point model is coded directly, the doubly nested loop used for s and t is quadratic in T. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.\ntransformed parameters {\n    vector[T] lp;\n    {\n      vector[T + 1] lp_e;\n      vector[T + 1] lp_l;\n      lp_e[1] = 0;\n      lp_l[1] = 0;\n      for (t in 1:T) {\n        lp_e[t + 1] = lp_e[t] + poisson_lpmf(D[t] | e);\n        lp_l[t + 1] = lp_l[t] + poisson_lpmf(D[t] | l);\n      }\n      lp = rep_vector(log_unif + lp_l[T + 1], T)\n           + head(lp_e, T) - head(lp_l, T);\n    }\n  }\nAs should be obvious from looking at it, it has linear complexity in T rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger T.\nThe key to understanding Wiklendt’s dynamic programming version is to see that head(lp_e) holds the forward values, whereas lp_l[T + 1] - head(lp_l, T) holds the backward values; the clever use of subtraction allows lp_l to be accumulated naturally in the forward direction.\n\n\n\nThis model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.\nThe value of lp for each iteration for each change point is available because it is declared as a transformed parameter. If the value of lp were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.\n\n\n\nThe value of lp[s] in a given iteration is given by \\(\\log\np(s,D \\mid e,l)\\) for the values of the early and late rates, \\(e\\) and \\(l\\), in the iteration. In each iteration after convergence, the early and late disaster rates, \\(e\\) and \\(l\\), are drawn from the posterior \\(p(e,l \\mid D)\\) by MCMC sampling and the associated lp calculated. The value of lp may be normalized to calculate \\(p(s \\mid e,l,D)\\) in each iteration, based on on the current values of \\(e\\) and \\(l\\). Averaging over iterations provides an unnormalized probability estimate of the change point being \\(s\\) (see below for the normalizing constant), \\[\\begin{align*}\np(s \\mid D) &\\propto q(s \\mid D) \\\\\n&= \\frac{1}{M} \\sum_{m=1}^{M} \\exp(\\texttt{lp}[m,s]).\n\\end{align*}\\] where \\(\\texttt{lp}[m,s]\\) represents the value of lp in posterior draw \\(m\\) for change point \\(s\\). By averaging over draws, \\(e\\) and \\(l\\) are themselves marginalized out, and the result has no dependence on a given iteration’s value for \\(e\\) and \\(l\\). A final normalization then produces the quantity of interest, the posterior probability of the change point being \\(s\\) conditioned on the data \\(D\\), \\[\np(s \\mid D) = \\frac{q(s \\mid D)}{\\sum_{s'=1}^T q(s' \\mid D)}.\n\\]\nA plot of the values of \\(\\log p(s \\mid D)\\) computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.\nLog probability of change point being in year, calculated analytically.\n\n\n\nAnalytical change-point posterior\n\n\nThe frequency of change points generated by sampling the discrete change points.\n\n\n\nSampled change-point posterior\n\n\nIn order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of \\(s\\) is roughly 1891.\n\n\n\nThe generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with lp defined as above, the following program draws a random value for s at every iteration.\ngenerated quantities {\n  int&lt;lower=1, upper=T&gt; s;\n  s = categorical_logit_rng(lp);\n}\nA posterior histogram of draws for \\(s\\) is shown on the second change point posterior figure above.\nCompared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for s given by softmax(lp).\n\n\n\nThe discrete sample generated for \\(s\\) can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using lp.\n\n\n\nThere is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, e, m, and l, and two loops for the change point and then one over time, with log densities being stored in a matrix.\nmatrix[T, T] lp;\nlp = rep_matrix(log_unif, T);\nfor (s1 in 1:T) {\n  for (s2 in 1:T) {\n    for (t in 1:T) {\n      lp[s1,s2] = lp[s1,s2]\n        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));\n    }\n  }\n}\nThe matrix can then be converted back to a vector using to_vector before being passed to log_sum_exp.\n\n\n\n\nA widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.\nThe first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.\n\n\nIn the simplest case, a one-stage mark-recapture study produces the following data\n\n\\(M\\) : number of animals marked in first capture,\n\\(C\\) : number animals in second capture, and\n\\(R\\) : number of marked animals in second capture.\n\nThe estimand of interest is\n\n\\(N\\) : number of animals in the population.\n\nDespite the notation, the model will take \\(N\\) to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.\nThe Lincoln-Petersen (Lincoln 1930; Petersen 1896) method for estimating population size is \\[\n\\hat{N} = \\frac{M C}{R}.\n\\]\nThis population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, \\[\nR \\sim \\textsf{binomial}(C, M / N)\n\\] given the total number of animals captured in the second round (\\(C\\)) with a recapture probability of \\(M/N\\), the fraction of the total population \\(N\\) marked in the first round.\ndata {\n  int&lt;lower=0&gt; M;\n  int&lt;lower=0&gt; C;\n  int&lt;lower=0, upper=min(M, C)&gt; R;\n}\nparameters {\n  real&lt;lower=(C - R + M)&gt; N;\n}\nmodel {\n  R ~ binomial(C, M / N);\n}\nA probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on \\(N\\) is necessary to efficiently eliminate impossible values.\nThe probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.\nTo ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for \\(N\\) is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.\nThe one tricky part of the model is the lower bound \\(C - R + M\\) placed on the population size \\(N\\). Values below this bound are impossible because it is otherwise not possible to draw \\(R\\) samples out of the \\(C\\) animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for \\(C\\) implies a variable transform \\(f : (C-R+M,\\infty) \\rightarrow (-\\infty,+\\infty)\\) defined by \\(f(N) = \\log(N - (C - R + M))\\); the reference manual contains full details of all constrained parameter transforms.\n\n\n\nThe Cormack-Jolly-Seber (CJS) model (Cormack 1964; Jolly 1965; Seber 1965) is an open-population model in which the population may change over time due to death; the presentation here draws heavily on Schofield (2007).\nThe basic data are\n\n\\(I\\): number of individuals,\n\\(T\\): number of capture periods, and\n\\(y_{i,t}\\): Boolean indicating if individual \\(i\\) was captured at time \\(t\\).\n\nEach individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.\nThere are two Bernoulli parameters in the model,\n\n\\(\\phi_t\\) : probability that animal alive at time \\(t\\) survives until \\(t + 1\\) and\n\\(p_t\\) : probability that animal alive at time \\(t\\) is captured at time \\(t\\).\n\nThese parameters will both be given uniform priors, but information should be used to tighten these priors in practice.\nThe CJS model also employs a latent discrete parameter \\(z_{i,t}\\) indicating for each individual \\(i\\) whether it is alive at time \\(t\\), distributed as \\[\nz_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t-1} \\; ? \\; 0 \\: : \\: \\phi_{t-1}).\n\\]\nThe conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on \\(z\\) as \\[\ny_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t} \\; ? \\; 0 \\: : \\: p_t).\n\\]\nThe conditional enforces the constraint that dead animals cannot be captured.\n\n\n\nThis subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.\nIn order to ease the marginalization of the latent discrete parameter \\(z_{i,t}\\), the Stan models rely on a derived quantity \\(\\chi_t\\) for the probability that an individual is never captured again if it is alive at time \\(t\\) (if it is dead, the recapture probability is zero). this quantity is defined recursively by \\[\n\\chi_t\n=\n\\begin{cases}\n1 & \\quad\\text{if } t = T \\\\\n(1 - \\phi_t) + \\phi_t (1 - p_{t+1}) \\chi_{t+1}\n  & \\quad\\text{if } t &lt; T\n\\end{cases}\n\\]\nThe base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining \\(\\chi_{t}\\) in terms of \\(\\chi_{t+1}\\) involves two possibilities: (1) not surviving to the next time period, with probability \\((1 - \\phi_t)\\), or (2) surviving to the next time period with probability \\(\\phi_t\\), not being captured in the next time period with probability \\((1 - p_{t+1})\\), and not being captured again after being alive in period \\(t+1\\) with probability \\(\\chi_{t+1}\\).\nWith three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.\n\\[\n\\begin{array}{crclc}\n\\hline\n& \\qquad\\qquad & captures & \\qquad\\qquad &  \\\\\n\\mathrm{profile} & 1 & 2 & 3 & \\mathrm{probability} \\\\\n\\hline\n0 & - & - & - & n/a \\\\\n1 & - & - & + & n/a \\\\\n2 & - & + & - & \\chi_2 \\\\\n3 & - & + & + & \\phi_2 \\, p_3 \\\\\n4 & + & - & - & \\chi_1 \\\\\n5 & + & - & + & \\phi_1 \\, (1 - p_2) \\, \\phi_2 \\, p_3 \\\\\n6 & + & + & - & \\phi_1 \\, p_2 \\, \\chi_2 \\\\\n7 & + & + & + & \\phi_1 \\, p_2 \\, \\phi_2 \\, p_3 \\\\\n\\hline\n\\end{array}\n\\]\nHistory 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.\nBy defining these probabilities in terms of \\(\\chi\\) directly, there is no need for a latent binary parameter indicating whether an animal is alive at time \\(t\\) or not. The definition of \\(\\chi\\) is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model (Schofield 2007).\nThe Stan model defines \\(\\chi\\) as a transformed parameter based on parameters \\(\\phi\\) and \\(p\\). In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using target += rather than being part of a built-in probability function.\nThe following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods\ndata {\n  array[7] int&lt;lower=0&gt; history;\n}\nparameters {\n  array[2] real&lt;lower=0, upper=1&gt; phi;\n  array[3] real&lt;lower=0, upper=1&gt; p;\n}\ntransformed parameters {\n  array[2] real&lt;lower=0, upper=1&gt; chi;\n  chi[2] = (1 - phi[2]) + phi[2] * (1 - p[3]);\n  chi[1] = (1 - phi[1]) + phi[1] * (1 - p[2]) * chi[2];\n}\nmodel {\n  target += history[2] * log(chi[2]);\n  target += history[3] * (log(phi[2]) + log(p[3]));\n  target += history[4] * (log(chi[1]));\n  target += history[5] * (log(phi[1]) + log1m(p[2])\n                            + log(phi[2]) + log(p[3]));\n  target += history[6] * (log(phi[1]) + log(p[2])\n                            + log(chi[2]));\n  target += history[7] * (log(phi[1]) + log(p[2])\n                            + log(phi[2]) + log(p[3]));\n}\ngenerated quantities {\n  real&lt;lower=0, upper=1&gt; beta3;\n  beta3 = phi[2] * p[3];\n}\n\n\nThe parameters \\(\\phi_2\\) and \\(p_3\\), the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, \\(\\beta_3 = \\phi_2 \\, p_3\\), is identified. The Stan model defines beta3 as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the \\(p\\) and \\(\\phi\\) parameters.\n\n\n\n\nThis section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number \\(T\\) of capture events, the number \\(I\\) of individuals, and a boolean flag \\(y_{i,t}\\) indicating if individual \\(i\\) was observed at time \\(t\\). In Stan,\ndata {\n  int&lt;lower=2&gt; T;\n  int&lt;lower=0&gt; I;\n  array[I, T] int&lt;lower=0, upper=1&gt; y;\n}\nThe advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding \\(2^T\\) observation histories for \\(T\\) capture times.\n\n\nThe individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.4\nfunctions {\n  int first_capture(array[] int y_i) {\n    for (k in 1:size(y_i)) {\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  int last_capture(array[] int y_i) {\n    for (k_rev in 0:(size(y_i) - 1)) {\n      int k;\n      k = size(y_i) - k_rev;\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  // ...\n}\nThese two functions are used to define the first and last capture time for each individual in the transformed data block.5\ntransformed data {\n  array[I] int&lt;lower=0, upper=T&gt; first;\n  array[I] int&lt;lower=0, upper=T&gt; last;\n  vector&lt;lower=0, upper=I&gt;[T] n_captured;\n  for (i in 1:I) {\n    first[i] = first_capture(y[i]);\n  }\n  for (i in 1:I) {\n    last[i] = last_capture(y[i]);\n  }\n  n_captured = rep_vector(0, T);\n  for (t in 1:T) {\n    for (i in 1:I) {\n      if (y[i, t]) {\n        n_captured[t] = n_captured[t] + 1;\n      }\n    }\n  }\n}\nThe transformed data block also defines n_captured[t], which is the total number of captures at time t. The variable n_captured is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.\nThe parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector chi, the probability that if an individual is alive at t that it will never be captured again.\nparameters {\n  vector&lt;lower=0, upper=1&gt;[T - 1] phi;\n  vector&lt;lower=0, upper=1&gt;[T] p;\n}\ntransformed parameters {\n  vector&lt;lower=0, upper=1&gt;[T] chi;\n  chi = prob_uncaptured(T, p, phi);\n}\nThe definition of prob_uncaptured, from the functions block, is\nfunctions {\n  // ...\n  vector prob_uncaptured(int T, vector p, vector phi) {\n    vector[T] chi;\n    chi[T] = 1.0;\n    for (t in 1:(T - 1)) {\n      int t_curr;\n      int t_next;\n      t_curr = T - t;\n      t_next = t_curr + 1;\n      chi[t_curr] = (1 - phi[t_curr])\n                     + phi[t_curr]\n                       * (1 - p[t_next])\n                       * chi[t_next];\n    }\n    return chi;\n  }\n}\nThe function definition directly follows the mathematical definition of \\(\\chi_t\\), unrolling the recursion into an iteration and defining the elements of chi from T down to 1.\n\n\n\nGiven the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations q given the parameters p and phi as well as the transformed parameter chi defined in terms of p and phi.\nmodel {\n  for (i in 1:I) {\n    if (first[i] &gt; 0) {\n      for (t in (first[i]+1):last[i]) {\n        1 ~ bernoulli(phi[t - 1]);\n        y[i, t] ~ bernoulli(p[t]);\n      }\n      1 ~ bernoulli(chi[last[i]]);\n    }\n  }\n}\nThe outer loop is over individuals, conditional skipping individuals i which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for first if an individual is never captured.\nThe inner loop for individual i first increments the log probability based on the survival of the individual with probability phi[t - 1]. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.\nIn the inner loop, the observed capture status y[i, t] for individual i at time t has a Bernoulli distribution based on the capture probability p[t] at time t.\nAfter the inner loop, the probability of an animal never being seen again after being observed at time last[i] is included, because last[i] was defined to be the last time period in which animal i was observed.\n\n\n\nAs with the collective model described in the previous subsection, this model does not identify phi[T - 1] and p[T], but does identify their product, beta. Thus beta is defined as a generated quantity to monitor convergence and report.\ngenerated quantities {\n  real beta;\n  // ...\n\n  beta = phi[T - 1] * p[T];\n  // ...\n}\nThe parameter p[1] is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case p[1] would be an unknown initial condition and both phi[T - 1] and p[T] could be identified.\n\n\n\nThe generated quantities also calculates an estimate of the population mean at each time t in the same way as in the simple mark-recapture model as the number of individuals captured at time t divided by the probability of capture at time t. This is done with the elementwise division operation for vectors (./) in the generated quantities block.\ngenerated quantities {\n  // ...\n  vector&lt;lower=0&gt;[T] pop;\n  // ...\n  pop = n_captured ./ p;\n  pop[1] = -1;\n}\n\n\n\nAll individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.\n\n\n\n\n\nAlthough seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.\n\n\nSuppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.\nThere are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.\n\n\n\nHumans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.\n\n\n\nIn this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.\nDawid and Skene (1979) introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.\n\n\nThe data for the model consists of \\(J\\) raters (diagnostic tests), \\(I\\) items (patients), and \\(K\\) categories (condition statuses) to annotate, with \\(y_{i, j} \\in \\{1, \\dotsc, K\\}\\) being the rating provided by rater \\(j\\) for item \\(i\\). In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often \\(K=2\\), with values signaling the presence or absence of the condition.6\nIt is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.\n\n\n\n\nThe model is based on three parameters, the first of which is discrete:\n\n\\(z_i\\) : a value in \\(\\{1, \\dotsc, K\\}\\) indicating the true category of item \\(i\\),\n\\(\\pi\\) : a \\(K\\)-simplex for the prevalence of the \\(K\\) categories in the population, and\n\\(\\theta_{j,k}\\) : a \\(K\\)-simplex for the response of annotator \\(j\\) to an item of true category \\(k\\).\n\n\n\n\nThe true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, \\[\nz_i \\sim \\textsf{categorical}(\\pi).\n\\]\nThe rating \\(y_{i, j}\\) provided for item \\(i\\) by rater \\(j\\) is modeled as a categorical response of rater \\(i\\) to an item of category \\(z_i\\),7 \\[\ny_{i, j} \\sim \\textsf{categorical}(\\theta_{j,\\pi_{z[i]}}).\n\\]\n\n\nDawid and Skene provided maximum likelihood estimates for \\(\\theta\\) and \\(\\pi\\), which allows them to generate probability estimates for each \\(z_i\\).\nTo mimic Dawid and Skene’s maximum likelihood model, the parameters \\(\\theta_{j,k}\\) and \\(\\pi\\) can be given uniform priors over \\(K\\)-simplexes. It is straightforward to generalize to Dirichlet priors, \\[\n\\pi \\sim \\textsf{Dirichlet}(\\alpha)\n\\] and \\[\n\\theta_{j,k} \\sim \\textsf{Dirichlet}(\\beta_k)\n\\] with fixed hyperparameters \\(\\alpha\\) (a vector) and \\(\\beta\\) (a matrix or array of vectors). The prior for \\(\\theta_{j,k}\\) must be allowed to vary in \\(k\\), so that, for instance, \\(\\beta_{k,k}\\) is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.\nBecause there are \\(J\\) coders, it would be natural to extend the model to include a hierarchical prior for \\(\\beta\\) and to partially pool the estimates of coder accuracy and bias.\n\n\n\nBecause the true category parameter \\(z\\) is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as \\[\np(y, \\theta, \\pi) = p(y \\mid \\theta,\\pi) \\, p(\\pi) \\, p(\\theta),\n\\] where \\(p(y \\mid \\theta,\\pi)\\) is derived by marginalizing \\(z\\) out of \\[\np(z, y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\left( \\textsf{categorical}(z_i \\mid \\pi)\n                     \\prod_{j=1}^J\n                     \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, z[i]})\n              \\right).\n\\]\nThis can be done item by item, with \\[\np(y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\sum_{k=1}^K\n  \\left( \\textsf{categorical}(k \\mid \\pi)\n         \\prod_{j=1}^J\n         \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n  \\right).\n\\]\nIn the missing data model, only the observed labels would be used in the inner product.\nDawid and Skene (1979) derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, \\[\\begin{align*}\n\\log p(y \\mid \\theta, \\pi)\n&= \\sum_{i=1}^I \\log \\left( \\sum_{k=1}^K \\exp\n    \\left(\\log \\textsf{categorical}(k \\mid \\pi) \\vphantom{\\sum_{j=1}^J}\\right.\\right.\n    \\left.\\left. + \\ \\sum_{j=1}^J\n           \\log \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n    \\right) \\right),\n\\end{align*}\\] which can be directly coded using Stan’s built-in log_sum_exp function.\n\n\n\n\nThe Stan program for the Dawid and Skene model is provided below (Dawid and Skene 1979).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=1&gt; I;\n  int&lt;lower=1&gt; J;\n\n  array[I, J] int&lt;lower=1, upper=K&gt; y;\n\n  vector&lt;lower=0&gt;[K] alpha;\n  vector&lt;lower=0&gt;[K] beta[K];\n}\nparameters {\n  simplex[K] pi;\n  array[J, K] simplex[K] theta;\n}\ntransformed parameters {\n  array[I] vector[K] log_q_z;\n  for (i in 1:I) {\n    log_q_z[i] = log(pi);\n    for (j in 1:J) {\n      for (k in 1:K) {\n        log_q_z[i, k] = log_q_z[i, k]\n                         + log(theta[j, k, y[i, j]]);\n      }\n    }\n  }\n}\nmodel {\n  pi ~ dirichlet(alpha);\n  for (j in 1:J) {\n    for (k in 1:K) {\n      theta[j, k] ~ dirichlet(beta[k]);\n    }\n  }\n\n  for (i in 1:I) {\n    target += log_sum_exp(log_q_z[i]);\n  }\n}\nThe model marginalizes out the discrete parameter \\(z\\), storing the unnormalized conditional probability \\(\\log q(z_i=k|\\theta,\\pi)\\) in log_q_z[i, k].\nThe Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are \\(\\alpha_k = 3\\) and \\(\\beta_{k,k} = 2.5 K\\) and \\(\\beta_{k,k'} = 1\\) if \\(k \\neq k'\\). Taking \\(\\alpha\\) and \\(\\beta_k\\) to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of Dawid and Skene (1979).\n\n\nThe quantity log_q_z[i] is defined as a transformed parameter. It encodes the (unnormalized) log of \\(p(z_i \\mid \\theta,\n\\pi)\\). Each iteration provides a value conditioned on that iteration’s values for \\(\\theta\\) and \\(\\pi\\). Applying the softmax function to log_q_z[i] provides a simplex corresponding to the probability mass function of \\(z_i\\) in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each \\(z_i\\).\n\n\n\n\n\n\n\nThis section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is not necessary to fit models with discrete parameters using Stan.\nThroughout, the model under consideration consists of both continuous parameters, \\(\\Theta\\), and discrete parameters, \\(Z\\). It is also assumed that \\(Z\\) can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.\n\n\n\nWhen performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating \\(\\mu\\), say, where \\(\\mu = \\mathbb{E}[g(\\Theta, Z)]\\) and \\(g(\\cdot)\\) is an arbitrary function. An example of such a quantity is \\(\\mathbb{E}[\\Theta]\\), the posterior mean of the continuous parameters, where we would take \\(g(\\theta, z) = \\theta\\). To estimate \\(\\mu\\) the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.\nIf both \\(\\Theta\\) and \\(Z\\) were continuous, Stan could be used to sample \\(M\\) draws from the joint posterior \\(p_{\\Theta, Z}(\\theta, z)\\) and then estimate \\(\\mu\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M {g(\\theta^{(i)}, z^{(i)})}.\n\\] Given \\(Z\\) is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out \\(Z\\) from the joint posterior to give the marginalized posterior \\(p_\\Theta(\\theta)\\). This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws \\(\\{\\theta^{(i)}\\}_{i = 1}^M\\) from the marginalized posterior.\nUsing only these draws, how can we estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\)? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function \\(g\\) of interest, we compute \\[\nh(\\Theta) = \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n\\] and then estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M h(\\theta^{(i)}).\n\\] This estimator is justified by the law of iterated expectation, the fact that \\[\n\\mathbb{E}[h(\\Theta)] = \\mathbb{E}[\\mathbb{E}[g(\\Theta, Z)] \\mid \\Theta] = \\mathbb{E}[g(\\Theta, Z)] = \\mu.\n\\] Using this marginalized estimator provides a way to estimate the expectation of any function \\(g(\\cdot)\\) for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\).\n\n\n\nFortunately, the discrete nature of \\(Z\\) makes evaluating \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\) easy. The function \\(h(\\Theta)\\) can be written as: \\[\nh(\\Theta)\n= \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n= \\sum_{k} g(\\Theta, k) \\Pr[Z = k \\mid \\Theta],\n\\] where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, \\(\\Pr[Z = k \\mid \\Theta]\\). More detail on how this quantity can be calculated is included below. Note that if \\(Z\\) takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.\nWhen \\(g(\\theta, z)\\) is a function of either \\(\\theta\\) or \\(z\\) only, the above formula simplifies further.\nIn the first case, where \\(g(\\theta, z) = g(\\theta)\\), we have: \\[\\begin{align*}\nh(\\Theta)\n&= \\sum_{k} g(\\Theta) \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta) \\sum_{k} \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta).\n\\end{align*}\\] This means that we can estimate \\(\\mathbb{E}[g(\\Theta)]\\) with the standard, seemingly unconditional, estimator: \\[\n\\frac{1}{M} \\sum_{i = 1}^M g(\\theta^{(i)}).\n\\] Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.\nIn the second case, where \\(g(\\theta, z) = g(z)\\), the conditional expectation instead simplifies as follows: \\[\nh(\\Theta) = \\sum_{k} g(k) \\Pr[Z = k \\mid \\Theta].\n\\] An important special case of this result is when \\(g(\\theta, z) = \\textrm{I}(z = k)\\), where \\(\\textrm{I}\\) is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable \\(Z\\), since \\(\\mathbb{E}[\\textrm{I}(Z = k)] = \\Pr[Z = k]\\). In this case, \\[\nh(\\Theta)\n= \\sum_{k} \\textrm{I}(z = k) \\Pr[Z = k \\mid \\Theta]\n= \\Pr[Z = k \\mid \\Theta].\n\\] The quantity \\(\\Pr[Z = k]\\) can therefore be estimated with: \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}].\n\\] When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, \\(Y\\). That is, we are really estimating \\(\\Pr[Z = k \\mid Y]\\) with \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y].\n\\] This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us \\[\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y]\n= \\frac{\\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}\n{\\sum_{k = 1}^K \\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}.\n\\] Here, \\(\\Pr[Y \\mid \\Theta = \\theta^{(i)}, Z = k]\\) is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.\nOther than the use of Bayes’s theorem, \\(\\Pr[Z = k \\mid \\theta = \\theta^{(i)}, Y]\\) can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the Dawid–Skene model).\nFor a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see Pullin, Gurrin, and Vukcevic (2021).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html#rao-blackwell.section",
+    "href": "stan-users-guide/latent-discrete.html#rao-blackwell.section",
+    "title": "Latent Discrete Parameters",
+    "section": "",
+    "text": "Although it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.\nStandard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html#change-point.section",
+    "href": "stan-users-guide/latent-discrete.html#change-point.section",
+    "title": "Latent Discrete Parameters",
+    "section": "",
+    "text": "The first example is a model of coal mining disasters in the U.K. for the years 1851–1962.2\n\n\nFonnesbeck et al. (2013, sec. 3.1) provides a Poisson model of disaster \\(D_t\\) in year \\(t\\) with two rate parameters, an early rate (\\(e\\)) and late rate (\\(l\\)), that change at a given point in time \\(s\\). The full model expressed using a latent discrete parameter \\(s\\) is \\[\\begin{align*}\ne   &\\sim  \\textsf{exponential}(r_e) \\\\\nl   &\\sim  \\textsf{exponential}(r_l) \\\\\ns   &\\sim  \\textsf{uniform}(1, T) \\\\\nD_t &\\sim  \\textsf{Poisson}(t &lt; s \\; ? \\; e \\: : \\: l)\n\\end{align*}\\]\nThe last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.3\nIt uses a compact notation involving separating its three arguments by a question mark (?) and a colon (:). The conditional operator is defined by \\[\nc \\; ? \\; x_1 \\: : \\: x_2\n=\n\\begin{cases}\n\\ x_1 & \\quad\\text{if } c \\text{ is true (i.e., non-zero), and} \\\\\n\\ x_2 & \\quad\\text{if } c \\text{ is false (i.e., zero).}\n\\end{cases}\n\\]\n\n\n\nTo code this model in Stan, the discrete parameter \\(s\\) must be marginalized out to produce a model defining the log of the probability function \\(p(e,l,D_t)\\). The full joint probability factors as \\[\\begin{align*}\np(e,l,s,D) &=  p(e) \\, p(l) \\, p(s) \\, p(D \\mid s, e, l) \\\\\n&= \\textsf{exponential}(e \\mid r_e) \\ \\textsf{exponential}(l \\mid r_l) \\,\n    \\textsf{uniform}(s \\mid 1, T) \\\\\n& \\qquad \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nTo marginalize, an alternative factorization into prior and likelihood is used, \\[\np(e,l,D) = p(e,l) \\, p(D \\mid e,l),\n\\]\nwhere the likelihood is defined by marginalizing \\(s\\) as \\[\\begin{align*}\np(D \\mid e,l) &= \\sum_{s=1}^T p(s, D \\mid e,l) \\\\\n&= \\sum_{s=1}^T p(s) \\, p(D \\mid s,e,l) \\\\\n&= \\sum_{s=1}^T \\textsf{uniform}(s \\mid 1,T) \\,\n    \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nStan operates on the log scale and thus requires the log likelihood, \\[\\begin{align*}\n\\log p(D \\mid e,l)\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{s=1}^T\n    \\left( \\log \\textsf{uniform}(s \\mid 1, T) \\vphantom{\\sum_{t=1}^T}\\right. \\\\\n&\\qquad \\left.\n    + \\sum_{t=1}^T \\log \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l)\n\\right),\n\\end{align*}\\] where the log sum of exponents function is defined by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{n=1}^N \\, \\alpha_n =\n\\log \\sum_{n=1}^N \\exp(\\alpha_n).\n\\]\nThe log sum of exponents function allows the model to be coded directly in Stan using the built-in function log_sum_exp, which provides both arithmetic stability and efficiency for mixture model calculations.\n\n\n\nThe Stan program for the change point model is shown in the figure below. The transformed parameter lp[s] stores the quantity \\(\\log p(s, D \\mid e, l)\\).\ndata {\n  real&lt;lower=0&gt; r_e;\n  real&lt;lower=0&gt; r_l;\n\n  int&lt;lower=1&gt; T;\n  array[T] int&lt;lower=0&gt; D;\n}\ntransformed data {\n  real log_unif;\n  log_unif = -log(T);\n}\nparameters {\n  real&lt;lower=0&gt; e;\n  real&lt;lower=0&gt; l;\n}\ntransformed parameters {\n  vector[T] lp;\n  lp = rep_vector(log_unif, T);\n  for (s in 1:T) {\n    for (t in 1:T) {\n      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);\n    }\n  }\n}\nmodel {\n  e ~ exponential(r_e);\n  l ~ exponential(r_l);\n  target += log_sum_exp(lp);\n}\nA change point model in which disaster rates D[t] have one rate, e, before the change point and a different rate, l, after the change point. The change point itself, s, is marginalized out as described in the text.\nAlthough the change-point model is coded directly, the doubly nested loop used for s and t is quadratic in T. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.\ntransformed parameters {\n    vector[T] lp;\n    {\n      vector[T + 1] lp_e;\n      vector[T + 1] lp_l;\n      lp_e[1] = 0;\n      lp_l[1] = 0;\n      for (t in 1:T) {\n        lp_e[t + 1] = lp_e[t] + poisson_lpmf(D[t] | e);\n        lp_l[t + 1] = lp_l[t] + poisson_lpmf(D[t] | l);\n      }\n      lp = rep_vector(log_unif + lp_l[T + 1], T)\n           + head(lp_e, T) - head(lp_l, T);\n    }\n  }\nAs should be obvious from looking at it, it has linear complexity in T rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger T.\nThe key to understanding Wiklendt’s dynamic programming version is to see that head(lp_e) holds the forward values, whereas lp_l[T + 1] - head(lp_l, T) holds the backward values; the clever use of subtraction allows lp_l to be accumulated naturally in the forward direction.\n\n\n\nThis model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.\nThe value of lp for each iteration for each change point is available because it is declared as a transformed parameter. If the value of lp were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.\n\n\n\nThe value of lp[s] in a given iteration is given by \\(\\log\np(s,D \\mid e,l)\\) for the values of the early and late rates, \\(e\\) and \\(l\\), in the iteration. In each iteration after convergence, the early and late disaster rates, \\(e\\) and \\(l\\), are drawn from the posterior \\(p(e,l \\mid D)\\) by MCMC sampling and the associated lp calculated. The value of lp may be normalized to calculate \\(p(s \\mid e,l,D)\\) in each iteration, based on on the current values of \\(e\\) and \\(l\\). Averaging over iterations provides an unnormalized probability estimate of the change point being \\(s\\) (see below for the normalizing constant), \\[\\begin{align*}\np(s \\mid D) &\\propto q(s \\mid D) \\\\\n&= \\frac{1}{M} \\sum_{m=1}^{M} \\exp(\\texttt{lp}[m,s]).\n\\end{align*}\\] where \\(\\texttt{lp}[m,s]\\) represents the value of lp in posterior draw \\(m\\) for change point \\(s\\). By averaging over draws, \\(e\\) and \\(l\\) are themselves marginalized out, and the result has no dependence on a given iteration’s value for \\(e\\) and \\(l\\). A final normalization then produces the quantity of interest, the posterior probability of the change point being \\(s\\) conditioned on the data \\(D\\), \\[\np(s \\mid D) = \\frac{q(s \\mid D)}{\\sum_{s'=1}^T q(s' \\mid D)}.\n\\]\nA plot of the values of \\(\\log p(s \\mid D)\\) computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.\nLog probability of change point being in year, calculated analytically.\n\n\n\nAnalytical change-point posterior\n\n\nThe frequency of change points generated by sampling the discrete change points.\n\n\n\nSampled change-point posterior\n\n\nIn order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of \\(s\\) is roughly 1891.\n\n\n\nThe generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with lp defined as above, the following program draws a random value for s at every iteration.\ngenerated quantities {\n  int&lt;lower=1, upper=T&gt; s;\n  s = categorical_logit_rng(lp);\n}\nA posterior histogram of draws for \\(s\\) is shown on the second change point posterior figure above.\nCompared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for s given by softmax(lp).\n\n\n\nThe discrete sample generated for \\(s\\) can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using lp.\n\n\n\nThere is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, e, m, and l, and two loops for the change point and then one over time, with log densities being stored in a matrix.\nmatrix[T, T] lp;\nlp = rep_matrix(log_unif, T);\nfor (s1 in 1:T) {\n  for (s2 in 1:T) {\n    for (t in 1:T) {\n      lp[s1,s2] = lp[s1,s2]\n        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));\n    }\n  }\n}\nThe matrix can then be converted back to a vector using to_vector before being passed to log_sum_exp.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html#mark-recapture-models",
+    "href": "stan-users-guide/latent-discrete.html#mark-recapture-models",
+    "title": "Latent Discrete Parameters",
+    "section": "",
+    "text": "A widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.\nThe first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.\n\n\nIn the simplest case, a one-stage mark-recapture study produces the following data\n\n\\(M\\) : number of animals marked in first capture,\n\\(C\\) : number animals in second capture, and\n\\(R\\) : number of marked animals in second capture.\n\nThe estimand of interest is\n\n\\(N\\) : number of animals in the population.\n\nDespite the notation, the model will take \\(N\\) to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.\nThe Lincoln-Petersen (Lincoln 1930; Petersen 1896) method for estimating population size is \\[\n\\hat{N} = \\frac{M C}{R}.\n\\]\nThis population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, \\[\nR \\sim \\textsf{binomial}(C, M / N)\n\\] given the total number of animals captured in the second round (\\(C\\)) with a recapture probability of \\(M/N\\), the fraction of the total population \\(N\\) marked in the first round.\ndata {\n  int&lt;lower=0&gt; M;\n  int&lt;lower=0&gt; C;\n  int&lt;lower=0, upper=min(M, C)&gt; R;\n}\nparameters {\n  real&lt;lower=(C - R + M)&gt; N;\n}\nmodel {\n  R ~ binomial(C, M / N);\n}\nA probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on \\(N\\) is necessary to efficiently eliminate impossible values.\nThe probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.\nTo ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for \\(N\\) is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.\nThe one tricky part of the model is the lower bound \\(C - R + M\\) placed on the population size \\(N\\). Values below this bound are impossible because it is otherwise not possible to draw \\(R\\) samples out of the \\(C\\) animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for \\(C\\) implies a variable transform \\(f : (C-R+M,\\infty) \\rightarrow (-\\infty,+\\infty)\\) defined by \\(f(N) = \\log(N - (C - R + M))\\); the reference manual contains full details of all constrained parameter transforms.\n\n\n\nThe Cormack-Jolly-Seber (CJS) model (Cormack 1964; Jolly 1965; Seber 1965) is an open-population model in which the population may change over time due to death; the presentation here draws heavily on Schofield (2007).\nThe basic data are\n\n\\(I\\): number of individuals,\n\\(T\\): number of capture periods, and\n\\(y_{i,t}\\): Boolean indicating if individual \\(i\\) was captured at time \\(t\\).\n\nEach individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.\nThere are two Bernoulli parameters in the model,\n\n\\(\\phi_t\\) : probability that animal alive at time \\(t\\) survives until \\(t + 1\\) and\n\\(p_t\\) : probability that animal alive at time \\(t\\) is captured at time \\(t\\).\n\nThese parameters will both be given uniform priors, but information should be used to tighten these priors in practice.\nThe CJS model also employs a latent discrete parameter \\(z_{i,t}\\) indicating for each individual \\(i\\) whether it is alive at time \\(t\\), distributed as \\[\nz_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t-1} \\; ? \\; 0 \\: : \\: \\phi_{t-1}).\n\\]\nThe conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on \\(z\\) as \\[\ny_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t} \\; ? \\; 0 \\: : \\: p_t).\n\\]\nThe conditional enforces the constraint that dead animals cannot be captured.\n\n\n\nThis subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.\nIn order to ease the marginalization of the latent discrete parameter \\(z_{i,t}\\), the Stan models rely on a derived quantity \\(\\chi_t\\) for the probability that an individual is never captured again if it is alive at time \\(t\\) (if it is dead, the recapture probability is zero). this quantity is defined recursively by \\[\n\\chi_t\n=\n\\begin{cases}\n1 & \\quad\\text{if } t = T \\\\\n(1 - \\phi_t) + \\phi_t (1 - p_{t+1}) \\chi_{t+1}\n  & \\quad\\text{if } t &lt; T\n\\end{cases}\n\\]\nThe base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining \\(\\chi_{t}\\) in terms of \\(\\chi_{t+1}\\) involves two possibilities: (1) not surviving to the next time period, with probability \\((1 - \\phi_t)\\), or (2) surviving to the next time period with probability \\(\\phi_t\\), not being captured in the next time period with probability \\((1 - p_{t+1})\\), and not being captured again after being alive in period \\(t+1\\) with probability \\(\\chi_{t+1}\\).\nWith three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.\n\\[\n\\begin{array}{crclc}\n\\hline\n& \\qquad\\qquad & captures & \\qquad\\qquad &  \\\\\n\\mathrm{profile} & 1 & 2 & 3 & \\mathrm{probability} \\\\\n\\hline\n0 & - & - & - & n/a \\\\\n1 & - & - & + & n/a \\\\\n2 & - & + & - & \\chi_2 \\\\\n3 & - & + & + & \\phi_2 \\, p_3 \\\\\n4 & + & - & - & \\chi_1 \\\\\n5 & + & - & + & \\phi_1 \\, (1 - p_2) \\, \\phi_2 \\, p_3 \\\\\n6 & + & + & - & \\phi_1 \\, p_2 \\, \\chi_2 \\\\\n7 & + & + & + & \\phi_1 \\, p_2 \\, \\phi_2 \\, p_3 \\\\\n\\hline\n\\end{array}\n\\]\nHistory 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.\nBy defining these probabilities in terms of \\(\\chi\\) directly, there is no need for a latent binary parameter indicating whether an animal is alive at time \\(t\\) or not. The definition of \\(\\chi\\) is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model (Schofield 2007).\nThe Stan model defines \\(\\chi\\) as a transformed parameter based on parameters \\(\\phi\\) and \\(p\\). In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using target += rather than being part of a built-in probability function.\nThe following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods\ndata {\n  array[7] int&lt;lower=0&gt; history;\n}\nparameters {\n  array[2] real&lt;lower=0, upper=1&gt; phi;\n  array[3] real&lt;lower=0, upper=1&gt; p;\n}\ntransformed parameters {\n  array[2] real&lt;lower=0, upper=1&gt; chi;\n  chi[2] = (1 - phi[2]) + phi[2] * (1 - p[3]);\n  chi[1] = (1 - phi[1]) + phi[1] * (1 - p[2]) * chi[2];\n}\nmodel {\n  target += history[2] * log(chi[2]);\n  target += history[3] * (log(phi[2]) + log(p[3]));\n  target += history[4] * (log(chi[1]));\n  target += history[5] * (log(phi[1]) + log1m(p[2])\n                            + log(phi[2]) + log(p[3]));\n  target += history[6] * (log(phi[1]) + log(p[2])\n                            + log(chi[2]));\n  target += history[7] * (log(phi[1]) + log(p[2])\n                            + log(phi[2]) + log(p[3]));\n}\ngenerated quantities {\n  real&lt;lower=0, upper=1&gt; beta3;\n  beta3 = phi[2] * p[3];\n}\n\n\nThe parameters \\(\\phi_2\\) and \\(p_3\\), the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, \\(\\beta_3 = \\phi_2 \\, p_3\\), is identified. The Stan model defines beta3 as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the \\(p\\) and \\(\\phi\\) parameters.\n\n\n\n\nThis section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number \\(T\\) of capture events, the number \\(I\\) of individuals, and a boolean flag \\(y_{i,t}\\) indicating if individual \\(i\\) was observed at time \\(t\\). In Stan,\ndata {\n  int&lt;lower=2&gt; T;\n  int&lt;lower=0&gt; I;\n  array[I, T] int&lt;lower=0, upper=1&gt; y;\n}\nThe advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding \\(2^T\\) observation histories for \\(T\\) capture times.\n\n\nThe individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.4\nfunctions {\n  int first_capture(array[] int y_i) {\n    for (k in 1:size(y_i)) {\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  int last_capture(array[] int y_i) {\n    for (k_rev in 0:(size(y_i) - 1)) {\n      int k;\n      k = size(y_i) - k_rev;\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  // ...\n}\nThese two functions are used to define the first and last capture time for each individual in the transformed data block.5\ntransformed data {\n  array[I] int&lt;lower=0, upper=T&gt; first;\n  array[I] int&lt;lower=0, upper=T&gt; last;\n  vector&lt;lower=0, upper=I&gt;[T] n_captured;\n  for (i in 1:I) {\n    first[i] = first_capture(y[i]);\n  }\n  for (i in 1:I) {\n    last[i] = last_capture(y[i]);\n  }\n  n_captured = rep_vector(0, T);\n  for (t in 1:T) {\n    for (i in 1:I) {\n      if (y[i, t]) {\n        n_captured[t] = n_captured[t] + 1;\n      }\n    }\n  }\n}\nThe transformed data block also defines n_captured[t], which is the total number of captures at time t. The variable n_captured is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.\nThe parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector chi, the probability that if an individual is alive at t that it will never be captured again.\nparameters {\n  vector&lt;lower=0, upper=1&gt;[T - 1] phi;\n  vector&lt;lower=0, upper=1&gt;[T] p;\n}\ntransformed parameters {\n  vector&lt;lower=0, upper=1&gt;[T] chi;\n  chi = prob_uncaptured(T, p, phi);\n}\nThe definition of prob_uncaptured, from the functions block, is\nfunctions {\n  // ...\n  vector prob_uncaptured(int T, vector p, vector phi) {\n    vector[T] chi;\n    chi[T] = 1.0;\n    for (t in 1:(T - 1)) {\n      int t_curr;\n      int t_next;\n      t_curr = T - t;\n      t_next = t_curr + 1;\n      chi[t_curr] = (1 - phi[t_curr])\n                     + phi[t_curr]\n                       * (1 - p[t_next])\n                       * chi[t_next];\n    }\n    return chi;\n  }\n}\nThe function definition directly follows the mathematical definition of \\(\\chi_t\\), unrolling the recursion into an iteration and defining the elements of chi from T down to 1.\n\n\n\nGiven the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations q given the parameters p and phi as well as the transformed parameter chi defined in terms of p and phi.\nmodel {\n  for (i in 1:I) {\n    if (first[i] &gt; 0) {\n      for (t in (first[i]+1):last[i]) {\n        1 ~ bernoulli(phi[t - 1]);\n        y[i, t] ~ bernoulli(p[t]);\n      }\n      1 ~ bernoulli(chi[last[i]]);\n    }\n  }\n}\nThe outer loop is over individuals, conditional skipping individuals i which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for first if an individual is never captured.\nThe inner loop for individual i first increments the log probability based on the survival of the individual with probability phi[t - 1]. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.\nIn the inner loop, the observed capture status y[i, t] for individual i at time t has a Bernoulli distribution based on the capture probability p[t] at time t.\nAfter the inner loop, the probability of an animal never being seen again after being observed at time last[i] is included, because last[i] was defined to be the last time period in which animal i was observed.\n\n\n\nAs with the collective model described in the previous subsection, this model does not identify phi[T - 1] and p[T], but does identify their product, beta. Thus beta is defined as a generated quantity to monitor convergence and report.\ngenerated quantities {\n  real beta;\n  // ...\n\n  beta = phi[T - 1] * p[T];\n  // ...\n}\nThe parameter p[1] is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case p[1] would be an unknown initial condition and both phi[T - 1] and p[T] could be identified.\n\n\n\nThe generated quantities also calculates an estimate of the population mean at each time t in the same way as in the simple mark-recapture model as the number of individuals captured at time t divided by the probability of capture at time t. This is done with the elementwise division operation for vectors (./) in the generated quantities block.\ngenerated quantities {\n  // ...\n  vector&lt;lower=0&gt;[T] pop;\n  // ...\n  pop = n_captured ./ p;\n  pop[1] = -1;\n}\n\n\n\nAll individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html#data-coding-models.section",
+    "href": "stan-users-guide/latent-discrete.html#data-coding-models.section",
+    "title": "Latent Discrete Parameters",
+    "section": "",
+    "text": "Although seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.\n\n\nSuppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.\nThere are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.\n\n\n\nHumans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.\n\n\n\nIn this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.\nDawid and Skene (1979) introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.\n\n\nThe data for the model consists of \\(J\\) raters (diagnostic tests), \\(I\\) items (patients), and \\(K\\) categories (condition statuses) to annotate, with \\(y_{i, j} \\in \\{1, \\dotsc, K\\}\\) being the rating provided by rater \\(j\\) for item \\(i\\). In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often \\(K=2\\), with values signaling the presence or absence of the condition.6\nIt is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.\n\n\n\n\nThe model is based on three parameters, the first of which is discrete:\n\n\\(z_i\\) : a value in \\(\\{1, \\dotsc, K\\}\\) indicating the true category of item \\(i\\),\n\\(\\pi\\) : a \\(K\\)-simplex for the prevalence of the \\(K\\) categories in the population, and\n\\(\\theta_{j,k}\\) : a \\(K\\)-simplex for the response of annotator \\(j\\) to an item of true category \\(k\\).\n\n\n\n\nThe true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, \\[\nz_i \\sim \\textsf{categorical}(\\pi).\n\\]\nThe rating \\(y_{i, j}\\) provided for item \\(i\\) by rater \\(j\\) is modeled as a categorical response of rater \\(i\\) to an item of category \\(z_i\\),7 \\[\ny_{i, j} \\sim \\textsf{categorical}(\\theta_{j,\\pi_{z[i]}}).\n\\]\n\n\nDawid and Skene provided maximum likelihood estimates for \\(\\theta\\) and \\(\\pi\\), which allows them to generate probability estimates for each \\(z_i\\).\nTo mimic Dawid and Skene’s maximum likelihood model, the parameters \\(\\theta_{j,k}\\) and \\(\\pi\\) can be given uniform priors over \\(K\\)-simplexes. It is straightforward to generalize to Dirichlet priors, \\[\n\\pi \\sim \\textsf{Dirichlet}(\\alpha)\n\\] and \\[\n\\theta_{j,k} \\sim \\textsf{Dirichlet}(\\beta_k)\n\\] with fixed hyperparameters \\(\\alpha\\) (a vector) and \\(\\beta\\) (a matrix or array of vectors). The prior for \\(\\theta_{j,k}\\) must be allowed to vary in \\(k\\), so that, for instance, \\(\\beta_{k,k}\\) is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.\nBecause there are \\(J\\) coders, it would be natural to extend the model to include a hierarchical prior for \\(\\beta\\) and to partially pool the estimates of coder accuracy and bias.\n\n\n\nBecause the true category parameter \\(z\\) is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as \\[\np(y, \\theta, \\pi) = p(y \\mid \\theta,\\pi) \\, p(\\pi) \\, p(\\theta),\n\\] where \\(p(y \\mid \\theta,\\pi)\\) is derived by marginalizing \\(z\\) out of \\[\np(z, y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\left( \\textsf{categorical}(z_i \\mid \\pi)\n                     \\prod_{j=1}^J\n                     \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, z[i]})\n              \\right).\n\\]\nThis can be done item by item, with \\[\np(y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\sum_{k=1}^K\n  \\left( \\textsf{categorical}(k \\mid \\pi)\n         \\prod_{j=1}^J\n         \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n  \\right).\n\\]\nIn the missing data model, only the observed labels would be used in the inner product.\nDawid and Skene (1979) derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, \\[\\begin{align*}\n\\log p(y \\mid \\theta, \\pi)\n&= \\sum_{i=1}^I \\log \\left( \\sum_{k=1}^K \\exp\n    \\left(\\log \\textsf{categorical}(k \\mid \\pi) \\vphantom{\\sum_{j=1}^J}\\right.\\right.\n    \\left.\\left. + \\ \\sum_{j=1}^J\n           \\log \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n    \\right) \\right),\n\\end{align*}\\] which can be directly coded using Stan’s built-in log_sum_exp function.\n\n\n\n\nThe Stan program for the Dawid and Skene model is provided below (Dawid and Skene 1979).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=1&gt; I;\n  int&lt;lower=1&gt; J;\n\n  array[I, J] int&lt;lower=1, upper=K&gt; y;\n\n  vector&lt;lower=0&gt;[K] alpha;\n  vector&lt;lower=0&gt;[K] beta[K];\n}\nparameters {\n  simplex[K] pi;\n  array[J, K] simplex[K] theta;\n}\ntransformed parameters {\n  array[I] vector[K] log_q_z;\n  for (i in 1:I) {\n    log_q_z[i] = log(pi);\n    for (j in 1:J) {\n      for (k in 1:K) {\n        log_q_z[i, k] = log_q_z[i, k]\n                         + log(theta[j, k, y[i, j]]);\n      }\n    }\n  }\n}\nmodel {\n  pi ~ dirichlet(alpha);\n  for (j in 1:J) {\n    for (k in 1:K) {\n      theta[j, k] ~ dirichlet(beta[k]);\n    }\n  }\n\n  for (i in 1:I) {\n    target += log_sum_exp(log_q_z[i]);\n  }\n}\nThe model marginalizes out the discrete parameter \\(z\\), storing the unnormalized conditional probability \\(\\log q(z_i=k|\\theta,\\pi)\\) in log_q_z[i, k].\nThe Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are \\(\\alpha_k = 3\\) and \\(\\beta_{k,k} = 2.5 K\\) and \\(\\beta_{k,k'} = 1\\) if \\(k \\neq k'\\). Taking \\(\\alpha\\) and \\(\\beta_k\\) to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of Dawid and Skene (1979).\n\n\nThe quantity log_q_z[i] is defined as a transformed parameter. It encodes the (unnormalized) log of \\(p(z_i \\mid \\theta,\n\\pi)\\). Each iteration provides a value conditioned on that iteration’s values for \\(\\theta\\) and \\(\\pi\\). Applying the softmax function to log_q_z[i] provides a simplex corresponding to the probability mass function of \\(z_i\\) in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each \\(z_i\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html#marginalization-mathematics.section",
+    "href": "stan-users-guide/latent-discrete.html#marginalization-mathematics.section",
+    "title": "Latent Discrete Parameters",
+    "section": "",
+    "text": "This section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is not necessary to fit models with discrete parameters using Stan.\nThroughout, the model under consideration consists of both continuous parameters, \\(\\Theta\\), and discrete parameters, \\(Z\\). It is also assumed that \\(Z\\) can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.\n\n\n\nWhen performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating \\(\\mu\\), say, where \\(\\mu = \\mathbb{E}[g(\\Theta, Z)]\\) and \\(g(\\cdot)\\) is an arbitrary function. An example of such a quantity is \\(\\mathbb{E}[\\Theta]\\), the posterior mean of the continuous parameters, where we would take \\(g(\\theta, z) = \\theta\\). To estimate \\(\\mu\\) the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.\nIf both \\(\\Theta\\) and \\(Z\\) were continuous, Stan could be used to sample \\(M\\) draws from the joint posterior \\(p_{\\Theta, Z}(\\theta, z)\\) and then estimate \\(\\mu\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M {g(\\theta^{(i)}, z^{(i)})}.\n\\] Given \\(Z\\) is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out \\(Z\\) from the joint posterior to give the marginalized posterior \\(p_\\Theta(\\theta)\\). This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws \\(\\{\\theta^{(i)}\\}_{i = 1}^M\\) from the marginalized posterior.\nUsing only these draws, how can we estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\)? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function \\(g\\) of interest, we compute \\[\nh(\\Theta) = \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n\\] and then estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M h(\\theta^{(i)}).\n\\] This estimator is justified by the law of iterated expectation, the fact that \\[\n\\mathbb{E}[h(\\Theta)] = \\mathbb{E}[\\mathbb{E}[g(\\Theta, Z)] \\mid \\Theta] = \\mathbb{E}[g(\\Theta, Z)] = \\mu.\n\\] Using this marginalized estimator provides a way to estimate the expectation of any function \\(g(\\cdot)\\) for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\).\n\n\n\nFortunately, the discrete nature of \\(Z\\) makes evaluating \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\) easy. The function \\(h(\\Theta)\\) can be written as: \\[\nh(\\Theta)\n= \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n= \\sum_{k} g(\\Theta, k) \\Pr[Z = k \\mid \\Theta],\n\\] where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, \\(\\Pr[Z = k \\mid \\Theta]\\). More detail on how this quantity can be calculated is included below. Note that if \\(Z\\) takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.\nWhen \\(g(\\theta, z)\\) is a function of either \\(\\theta\\) or \\(z\\) only, the above formula simplifies further.\nIn the first case, where \\(g(\\theta, z) = g(\\theta)\\), we have: \\[\\begin{align*}\nh(\\Theta)\n&= \\sum_{k} g(\\Theta) \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta) \\sum_{k} \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta).\n\\end{align*}\\] This means that we can estimate \\(\\mathbb{E}[g(\\Theta)]\\) with the standard, seemingly unconditional, estimator: \\[\n\\frac{1}{M} \\sum_{i = 1}^M g(\\theta^{(i)}).\n\\] Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.\nIn the second case, where \\(g(\\theta, z) = g(z)\\), the conditional expectation instead simplifies as follows: \\[\nh(\\Theta) = \\sum_{k} g(k) \\Pr[Z = k \\mid \\Theta].\n\\] An important special case of this result is when \\(g(\\theta, z) = \\textrm{I}(z = k)\\), where \\(\\textrm{I}\\) is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable \\(Z\\), since \\(\\mathbb{E}[\\textrm{I}(Z = k)] = \\Pr[Z = k]\\). In this case, \\[\nh(\\Theta)\n= \\sum_{k} \\textrm{I}(z = k) \\Pr[Z = k \\mid \\Theta]\n= \\Pr[Z = k \\mid \\Theta].\n\\] The quantity \\(\\Pr[Z = k]\\) can therefore be estimated with: \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}].\n\\] When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, \\(Y\\). That is, we are really estimating \\(\\Pr[Z = k \\mid Y]\\) with \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y].\n\\] This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us \\[\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y]\n= \\frac{\\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}\n{\\sum_{k = 1}^K \\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}.\n\\] Here, \\(\\Pr[Y \\mid \\Theta = \\theta^{(i)}, Z = k]\\) is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.\nOther than the use of Bayes’s theorem, \\(\\Pr[Z = k \\mid \\theta = \\theta^{(i)}, Y]\\) can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the Dawid–Skene model).\nFor a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see Pullin, Gurrin, and Vukcevic (2021).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/latent-discrete.html#footnotes",
+    "href": "stan-users-guide/latent-discrete.html#footnotes",
+    "title": "Latent Discrete Parameters",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe computations are similar to those involved in expectation maximization (EM) algorithms (Dempster, Laird, and Rubin 1977).↩︎\nThe source of the data is (Jarrett 1979), which itself is a note correcting an earlier data collection.↩︎\nThe R counterpart, ifelse, is slightly different in that it is typically used in a vectorized situation. The conditional operator is not (yet) vectorized in Stan.↩︎\nAn alternative would be to compute this on the outside and feed it into the Stan model as preprocessed data. Yet another alternative encoding would be a sparse one recording only the capture events along with their time and identifying the individual captured.↩︎\nBoth functions return 0 if the individual represented by the input array was never captured. Individuals with no captures are not relevant for estimating the model because all probability statements are conditional on earlier captures. Typically they would be removed from the data, but the program allows them to be included even though they make not contribution to the log probability function.↩︎\nDiagnostic procedures are often ordinal, as in stages of cancer in oncological diagnosis or the severity of a cavity in dental diagnosis. Dawid and Skene’s model may be used as is or naturally generalized for ordinal ratings using a latent continuous rating and cutpoints as in ordinal logistic regression.↩︎\nIn the subscript, \\(z_i\\) is written as \\(z[i]\\) to improve legibility.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Latent Discrete Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/measurement-error.html",
+    "href": "stan-users-guide/measurement-error.html",
+    "title": "Measurement Error and Meta-Analysis",
+    "section": "",
+    "text": "Most quantities used in statistical models arise from measurements. Most of these measurements are taken with some error. When the measurement error is small relative to the quantity being measured, its effect on a model is usually small. When measurement error is large relative to the quantity being measured, or when precise relations can be estimated being measured quantities, it is useful to introduce an explicit model of measurement error. One kind of measurement error is rounding.\nMeta-analysis plays out statistically much like measurement error models, where the inferences drawn from multiple data sets are combined to do inference over all of them. Inferences for each data set are treated as providing a kind of measurement error with respect to true parameter values.\n\n\nA Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data (Clayton 1992; Richardson and Gilks 1993). This requires a model of how the measurements are derived from the true values.\n\n\nBefore considering regression with measurement error, first consider a linear regression model where the observed data for \\(N\\) cases includes a predictor \\(x_n\\) and outcome \\(y_n\\). In Stan, a linear regression for \\(y\\) based on \\(x\\) with a slope and intercept is modeled as follows.\ndata {\n  int&lt;lower=0&gt; N;       // number of cases\n  vector[N] x;          // predictor (covariate)\n  vector[N] y;          // outcome (variate)\n}\nparameters {\n  real alpha;           // intercept\n  real beta;            // slope\n  real&lt;lower=0&gt; sigma;  // outcome noise\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ cauchy(0, 5);\n}\nNow suppose that the true values of the predictors \\(x_n\\) are not known, but for each \\(n\\), a measurement \\(x^{\\textrm{meas}}_n\\) of \\(x_n\\) is available. If the error in measurement can be modeled, the measured value \\(x^{\\textrm{meas}}_n\\) can be modeled in terms of the true value \\(x_n\\) plus measurement noise. The true value \\(x_n\\) is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation \\(\\tau\\). This leads to the following regression model with constant measurement error.\ndata {\n  // ...\n  array[N] real x_meas;   // measurement of x\n  real&lt;lower=0&gt; tau;     // measurement noise\n}\nparameters {\n  array[N] real x;    // unknown true value\n  real mu_x;          // prior location\n  real sigma_x;       // prior scale\n  // ...\n}\nmodel {\n  x ~ normal(mu_x, sigma_x);  // prior\n  x_meas ~ normal(x, tau);    // measurement model\n  y ~ normal(alpha + beta * x, sigma);\n  // ...\n}\nThe regression coefficients alpha and beta and regression noise scale sigma are the same as before, but now x is declared as a parameter rather than as data. The data are now x_meas, which is a measurement of the true x value with noise scale tau. The model then specifies that the measurement error for x_meas[n] given true value x[n] is normal with deviation tau. Furthermore, the true values x are given a hierarchical prior here.\nIn cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, Clayton (1992) introduces an exposure model for the unknown (but noisily measured) risk factors \\(x\\) in terms of known (without measurement error) risk factors \\(c\\). A simple model would regress \\(x_n\\) on the covariates \\(c_n\\) with noise term \\(\\upsilon\\), \\[\nx_n \\sim \\textsf{normal}(\\gamma^{\\top}c, \\upsilon).\n\\] This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.\n\n\n\nA common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.\nExercise 3.5(b) by Gelman et al. (2013) provides an example.\n\n3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on \\(\\mu\\) and \\(\\sigma^2\\).\n\nGive the correct posterior distribution for \\((\\mu, \\sigma^2)\\), treating the measurements as rounded.\n\n\nLetting \\(z_n\\) be the unrounded measurement for \\(y_n\\), the problem as stated assumes \\[\nz_n \\sim \\textsf{normal}(\\mu, \\sigma).\n\\]\nThe rounding process entails that \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\)1. The probability mass function for the discrete observation \\(y\\) is then given by marginalizing out the unrounded measurement, producing the likelihood \\[\\begin{align*}\np(y_n \\mid \\mu, \\sigma)\n&= \\int_{y_n - 0.5}^{y_n + 0.5} \\textsf{normal}(z_n \\mid \\mu, \\sigma) \\,\\textsf{d}z_n \\\\\n&= \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n   -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right).\n\\end{align*}\\] Gelman’s answer for this problem took the noninformative prior to be uniform in the variance \\(\\sigma^2\\) on the log scale, but we replace it with more recently recommended half-normal prior on \\(\\sigma\\) \\[\n\\sigma \\sim \\textsf{normal}^+(0, 1).\n\\] The posterior after observing \\(y = (10, 10, 12, 11, 9)\\) can be calculated by Bayes’s rule as \\[\\begin{align*}\np(\\mu, \\sigma \\mid y)\n&\\propto p(\\mu, \\sigma) \\ p(y \\mid \\mu, \\sigma) \\\\\n&\\propto \\textsf{normal}^+(\\sigma \\mid 0, 1)\\prod_{n=1}^5\n  \\left( \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n        -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right)\n  \\right).\n\\end{align*}\\]\nThe Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  for (n in 1:N) {\n    target += log_diff_exp(normal_lcdf(y[n] + 0.5 | mu, sigma),\n                           normal_lcdf(y[n] - 0.5 | mu, sigma));\n  }\n}\nwhere normal_lcdf(y[n]+0.5 | mu, sigma) is equal to log(Phi((y[n] + 0.5 - mu) / sigma)), and log_diff_exp(a, b) computes log(exp(a) - exp(b)) in numerically more stable way.\nAlternatively, the model may be defined with latent parameters for the unrounded measurements \\(z_n\\). The Stan code in this case uses a distribution statement for \\(z_n\\) directly while respecting the constraint \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;lower=y-0.5, upper=y+0.5&gt;[N] z;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  z ~ normal(mu, sigma);\n}\nThis explicit model for the unrounded measurements \\(z\\) produces the same posterior for \\(\\mu\\) and \\(\\sigma\\) as the previous model that marginalizes \\(z\\) out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.\n\n\n\n\nMeta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.\nThe Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.\n\n\nSuppose the data in question arise from a total of \\(M\\) studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen (Warn, Thompson, and Spiegelhalter 2002) or mortality after myocardial infarction under a treatment of beta blockers (Gelman et al. 2013, sec. 5.6).\n\n\nThe clinical data consists of \\(J\\) trials, each with \\(n^t\\) treatment cases, \\(n^c\\) control cases, \\(r^t\\) successful outcomes among those treated and \\(r^c\\) successful outcomes among those in the control group. This data can be declared in Stan as follows.2\ndata {\n  int&lt;lower=0&gt; J;\n  array[J] int&lt;lower=0&gt; n_t;  // num cases, treatment\n  array[J] int&lt;lower=0&gt; r_t;  // num successes, treatment\n  array[J] int&lt;lower=0&gt; n_c;  // num cases, control\n  array[J] int&lt;lower=0&gt; r_c;  // num successes, control\n}\n\n\n\nAlthough the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio \\[\\begin{align*}\ny_j &= \\log \\left( \\frac{r^t_j / (n^t_j - r^t_j)}\n                        {r^c_j / (n^c_j - r^c_j)}\n        \\right) \\\\\n&= \\log \\left( \\frac{r^t_j}{n^t_j - r^t_j} \\right)\n   -\\log \\left( \\frac{r^c_j}{n^c_j - r^c_j} \\right)\n\\end{align*}\\] and corresponding standard errors \\[\n\\sigma_j = \\sqrt{\n  \\frac{1}{r^T_j}\n+ \\frac{1}{n^T_j - r^T_j}\n+ \\frac{1}{r^C_j}\n+ \\frac{1}{n^C_j - r^C_j}\n}.\n\\]\nThe log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.3\ntransformed data {\n  array[J] real y;\n  array[J] real&lt;lower=0&gt; sigma;\n  for (j in 1:J) {\n    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])\n            - (log(r_c[j]) - log(n_c[j] - r_c[j]));\n  }\n  for (j in 1:J) {\n    sigma[j] = sqrt(1 / r_t[j] + 1 / (n_t[j] - r_t[j])\n                     + 1 / r_c[j] + 1 / (n_c[j] - r_c[j]));\n  }\n}\nThis definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.\n\n\n\nWith the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.\nparameters {\n  real theta;  // global treatment effect, log odds\n}\nmodel {\n  y ~ normal(theta, sigma);\n}\nThe distribution statement for y is vectorized; it has the same effect as the following.\n  for (j in 1:J) {\n    y[j] ~ normal(theta, sigma[j]);\n  }\nIt is common to include a prior for theta in this model, but it is not strictly necessary for the model to be proper because y is fixed and \\(\\textsf{normal}(y \\mid \\mu,\\sigma) =\n\\textsf{normal}(\\mu \\mid y,\\sigma)\\).\n\n\n\nTo model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.\nparameters {\n  array[J] real theta;  // per-trial treatment effect\n  real mu;              // mean treatment effect\n  real&lt;lower=0&gt; tau;    // deviation of treatment effects\n}\nmodel {\n  y ~ normal(theta, sigma);\n  theta ~ normal(mu, tau);\n  mu ~ normal(0, 10);\n  tau ~ cauchy(0, 5);\n}\nAlthough the vectorized distribution statement for y appears unchanged, the parameter theta is now a vector. The distribution statement for theta is also vectorized, with the hyperparameters mu and tau themselves being given wide priors compared to the scale of the data.\nRubin (1981) provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.\n\n\n\nSmith, Spiegelhalter, and Thomas (1995) and Gelman et al. (2013, sec. 19.4) provide meta-analyses based directly on binomial data. Warn, Thompson, and Spiegelhalter (2002) consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.\nIf trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects \\(\\theta_j\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/measurement-error.html#bayesian-measurement-error-model",
+    "href": "stan-users-guide/measurement-error.html#bayesian-measurement-error-model",
+    "title": "Measurement Error and Meta-Analysis",
+    "section": "",
+    "text": "A Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data (Clayton 1992; Richardson and Gilks 1993). This requires a model of how the measurements are derived from the true values.\n\n\nBefore considering regression with measurement error, first consider a linear regression model where the observed data for \\(N\\) cases includes a predictor \\(x_n\\) and outcome \\(y_n\\). In Stan, a linear regression for \\(y\\) based on \\(x\\) with a slope and intercept is modeled as follows.\ndata {\n  int&lt;lower=0&gt; N;       // number of cases\n  vector[N] x;          // predictor (covariate)\n  vector[N] y;          // outcome (variate)\n}\nparameters {\n  real alpha;           // intercept\n  real beta;            // slope\n  real&lt;lower=0&gt; sigma;  // outcome noise\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ cauchy(0, 5);\n}\nNow suppose that the true values of the predictors \\(x_n\\) are not known, but for each \\(n\\), a measurement \\(x^{\\textrm{meas}}_n\\) of \\(x_n\\) is available. If the error in measurement can be modeled, the measured value \\(x^{\\textrm{meas}}_n\\) can be modeled in terms of the true value \\(x_n\\) plus measurement noise. The true value \\(x_n\\) is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation \\(\\tau\\). This leads to the following regression model with constant measurement error.\ndata {\n  // ...\n  array[N] real x_meas;   // measurement of x\n  real&lt;lower=0&gt; tau;     // measurement noise\n}\nparameters {\n  array[N] real x;    // unknown true value\n  real mu_x;          // prior location\n  real sigma_x;       // prior scale\n  // ...\n}\nmodel {\n  x ~ normal(mu_x, sigma_x);  // prior\n  x_meas ~ normal(x, tau);    // measurement model\n  y ~ normal(alpha + beta * x, sigma);\n  // ...\n}\nThe regression coefficients alpha and beta and regression noise scale sigma are the same as before, but now x is declared as a parameter rather than as data. The data are now x_meas, which is a measurement of the true x value with noise scale tau. The model then specifies that the measurement error for x_meas[n] given true value x[n] is normal with deviation tau. Furthermore, the true values x are given a hierarchical prior here.\nIn cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, Clayton (1992) introduces an exposure model for the unknown (but noisily measured) risk factors \\(x\\) in terms of known (without measurement error) risk factors \\(c\\). A simple model would regress \\(x_n\\) on the covariates \\(c_n\\) with noise term \\(\\upsilon\\), \\[\nx_n \\sim \\textsf{normal}(\\gamma^{\\top}c, \\upsilon).\n\\] This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.\n\n\n\nA common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.\nExercise 3.5(b) by Gelman et al. (2013) provides an example.\n\n3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on \\(\\mu\\) and \\(\\sigma^2\\).\n\nGive the correct posterior distribution for \\((\\mu, \\sigma^2)\\), treating the measurements as rounded.\n\n\nLetting \\(z_n\\) be the unrounded measurement for \\(y_n\\), the problem as stated assumes \\[\nz_n \\sim \\textsf{normal}(\\mu, \\sigma).\n\\]\nThe rounding process entails that \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\)1. The probability mass function for the discrete observation \\(y\\) is then given by marginalizing out the unrounded measurement, producing the likelihood \\[\\begin{align*}\np(y_n \\mid \\mu, \\sigma)\n&= \\int_{y_n - 0.5}^{y_n + 0.5} \\textsf{normal}(z_n \\mid \\mu, \\sigma) \\,\\textsf{d}z_n \\\\\n&= \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n   -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right).\n\\end{align*}\\] Gelman’s answer for this problem took the noninformative prior to be uniform in the variance \\(\\sigma^2\\) on the log scale, but we replace it with more recently recommended half-normal prior on \\(\\sigma\\) \\[\n\\sigma \\sim \\textsf{normal}^+(0, 1).\n\\] The posterior after observing \\(y = (10, 10, 12, 11, 9)\\) can be calculated by Bayes’s rule as \\[\\begin{align*}\np(\\mu, \\sigma \\mid y)\n&\\propto p(\\mu, \\sigma) \\ p(y \\mid \\mu, \\sigma) \\\\\n&\\propto \\textsf{normal}^+(\\sigma \\mid 0, 1)\\prod_{n=1}^5\n  \\left( \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n        -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right)\n  \\right).\n\\end{align*}\\]\nThe Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  for (n in 1:N) {\n    target += log_diff_exp(normal_lcdf(y[n] + 0.5 | mu, sigma),\n                           normal_lcdf(y[n] - 0.5 | mu, sigma));\n  }\n}\nwhere normal_lcdf(y[n]+0.5 | mu, sigma) is equal to log(Phi((y[n] + 0.5 - mu) / sigma)), and log_diff_exp(a, b) computes log(exp(a) - exp(b)) in numerically more stable way.\nAlternatively, the model may be defined with latent parameters for the unrounded measurements \\(z_n\\). The Stan code in this case uses a distribution statement for \\(z_n\\) directly while respecting the constraint \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;lower=y-0.5, upper=y+0.5&gt;[N] z;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  z ~ normal(mu, sigma);\n}\nThis explicit model for the unrounded measurements \\(z\\) produces the same posterior for \\(\\mu\\) and \\(\\sigma\\) as the previous model that marginalizes \\(z\\) out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/measurement-error.html#meta-analysis",
+    "href": "stan-users-guide/measurement-error.html#meta-analysis",
+    "title": "Measurement Error and Meta-Analysis",
+    "section": "",
+    "text": "Meta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.\nThe Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.\n\n\nSuppose the data in question arise from a total of \\(M\\) studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen (Warn, Thompson, and Spiegelhalter 2002) or mortality after myocardial infarction under a treatment of beta blockers (Gelman et al. 2013, sec. 5.6).\n\n\nThe clinical data consists of \\(J\\) trials, each with \\(n^t\\) treatment cases, \\(n^c\\) control cases, \\(r^t\\) successful outcomes among those treated and \\(r^c\\) successful outcomes among those in the control group. This data can be declared in Stan as follows.2\ndata {\n  int&lt;lower=0&gt; J;\n  array[J] int&lt;lower=0&gt; n_t;  // num cases, treatment\n  array[J] int&lt;lower=0&gt; r_t;  // num successes, treatment\n  array[J] int&lt;lower=0&gt; n_c;  // num cases, control\n  array[J] int&lt;lower=0&gt; r_c;  // num successes, control\n}\n\n\n\nAlthough the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio \\[\\begin{align*}\ny_j &= \\log \\left( \\frac{r^t_j / (n^t_j - r^t_j)}\n                        {r^c_j / (n^c_j - r^c_j)}\n        \\right) \\\\\n&= \\log \\left( \\frac{r^t_j}{n^t_j - r^t_j} \\right)\n   -\\log \\left( \\frac{r^c_j}{n^c_j - r^c_j} \\right)\n\\end{align*}\\] and corresponding standard errors \\[\n\\sigma_j = \\sqrt{\n  \\frac{1}{r^T_j}\n+ \\frac{1}{n^T_j - r^T_j}\n+ \\frac{1}{r^C_j}\n+ \\frac{1}{n^C_j - r^C_j}\n}.\n\\]\nThe log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.3\ntransformed data {\n  array[J] real y;\n  array[J] real&lt;lower=0&gt; sigma;\n  for (j in 1:J) {\n    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])\n            - (log(r_c[j]) - log(n_c[j] - r_c[j]));\n  }\n  for (j in 1:J) {\n    sigma[j] = sqrt(1 / r_t[j] + 1 / (n_t[j] - r_t[j])\n                     + 1 / r_c[j] + 1 / (n_c[j] - r_c[j]));\n  }\n}\nThis definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.\n\n\n\nWith the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.\nparameters {\n  real theta;  // global treatment effect, log odds\n}\nmodel {\n  y ~ normal(theta, sigma);\n}\nThe distribution statement for y is vectorized; it has the same effect as the following.\n  for (j in 1:J) {\n    y[j] ~ normal(theta, sigma[j]);\n  }\nIt is common to include a prior for theta in this model, but it is not strictly necessary for the model to be proper because y is fixed and \\(\\textsf{normal}(y \\mid \\mu,\\sigma) =\n\\textsf{normal}(\\mu \\mid y,\\sigma)\\).\n\n\n\nTo model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.\nparameters {\n  array[J] real theta;  // per-trial treatment effect\n  real mu;              // mean treatment effect\n  real&lt;lower=0&gt; tau;    // deviation of treatment effects\n}\nmodel {\n  y ~ normal(theta, sigma);\n  theta ~ normal(mu, tau);\n  mu ~ normal(0, 10);\n  tau ~ cauchy(0, 5);\n}\nAlthough the vectorized distribution statement for y appears unchanged, the parameter theta is now a vector. The distribution statement for theta is also vectorized, with the hyperparameters mu and tau themselves being given wide priors compared to the scale of the data.\nRubin (1981) provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.\n\n\n\nSmith, Spiegelhalter, and Thomas (1995) and Gelman et al. (2013, sec. 19.4) provide meta-analyses based directly on binomial data. Warn, Thompson, and Spiegelhalter (2002) consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.\nIf trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects \\(\\theta_j\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/measurement-error.html#footnotes",
+    "href": "stan-users-guide/measurement-error.html#footnotes",
+    "title": "Measurement Error and Meta-Analysis",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThere are several different rounding rules (see, e.g., Wikipedia: Rounding), which affect which interval ends are open and which are closed, but these do not matter here as for continuous \\(z_n\\) \\(p(z_n=y_n-0.5)=p(z_n=y_n+0.5)=0\\).↩︎\nStan’s integer constraints are not powerful enough to express the constraint that \\(\\texttt{r}\\mathtt{\\_}\\texttt{t[j]} \\leq \\texttt{n}\\mathtt{\\_}\\texttt{t[j]}\\), but this constraint could be checked in the transformed data block.↩︎\nWhen dividing two integers, the result type is an integer and rounding will ensue if the result is not exact. See the discussion of primitive arithmetic types in the reference manual for more information.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multi-indexing.html",
+    "href": "stan-users-guide/multi-indexing.html",
+    "title": "Multiple Indexing and Range Indexing",
+    "section": "",
+    "text": "Stan allows multiple indexes to be provided for containers (i.e., arrays, vectors, and matrices) in a single position, using either an array of integer indexes or range bounds. In many cases, there are functions that provide similar behavior.\nAllowing multiple indexes supports inline vectorization of models. For instance, consider the data model for a varying-slope, varying-intercept hierarchical linear regression, which could be coded as\nfor (n in 1:N) {\n  y[n] ~ normal(alpha[ii[n]] + beta[ii[n]] * x[n], sigma);\n}\nWith multiple indexing, this can be coded in one line, leading to more efficient vectorized code.\ny ~ normal(alpha[ii] + rows_dot_product(beta[ii], x), sigma);\nThis latter version is faster than the loop version; it is equivalent in speed to the clunky assignment to a local variable.\n{\n  vector[N] mu;\n  for (n in 1:N) {\n    mu[n] = alpha[ii[n]] + beta[ii[n]] * x[n];\n  }\n  y ~ normal(mu, sigma);\n}\nThe boost in speed compared to the original version is because the single call to the normal log density in the distribution statement will be much more memory efficient than the original version.\n\n\nThe following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.\narray[3] int c;\n// ... define: c == (5, 9, 7)\narray[4] int idxs;\n// ... define: idxs == (3, 3, 1, 2)\narray[4] int d;\nd = c[idxs];    // result: d == (7, 7, 5, 9)\nIn general, the multiple indexed expression c[idxs] is defined as follows, assuming idxs is of size K.\nc[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )\nThus c[idxs] is of the same size as idxs, which is K in this example.\nMultiple indexing can also be used with multi-dimensional arrays. For example, consider the following.\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[4] int idxs;\n// ... define: idxs = (2, 2, 1, 2)\narray[4, 3] int d\nd = c[idxs];    // result: d = ((7, 11, 13), (7, 11, 13),\n                //              (1, 3, 5), (7, 11, 13))\nThat is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by c[idxs][j] ==   c[idxs[j]].\nMultiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.\narray[4] int e;\ne = c[2, idxs]; // result:  c[2] = (7, 11, 13)\n                // result:  e = (11, 11, 7, 11)\nThe single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, c[2,idxs] evaluates to the same value as c[2][idxs].\nMultiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[3] int idxs1;\n// ... define: idxs1 = (2, 2, 1)\narray[2] int idxs2;\n// ... define: idxs2 = (1, 3)\narray[3, 2] int d;\nd = c[idxs1, idxs2];  // result: d = ((7, 13), (7, 13), (1, 5))\nWith multiple indexes, we no longer have c[idxs1, idxs2] being the same as c[idxs1][idxs2]. Rather, the entry d[i, j] after executing the above is given by\nd[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]\nThis example illustrates the operation of multiple indexing in the general case: a multiple index like idxs1 converts an index i used on the result (here, c[idxs1, idxs2]) to index idxs1[i] in the variable being indexed (here, c). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.\n\n\n\nSlicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.\n\n\nFor instance, consider supplying an upper and lower bound for an index.\narray[7] int c;\n// ...\narray[4] int d;\nd = c[3:6];  // result: d == (c[3], c[4], c[5], c[6])\nThe range index 3:6 behaves semantically just like the multiple index (3, 4, 5, 6). In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.\n\n\n\nIt is also possible to supply just a lower bound, or just an upper bound. Writing c[3:] is just shorthand for c[3:size(c)]. Writing c[:5] is just shorthand for c[1:5].\n\n\n\nFinally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (:) as the index or leaving the index position empty. In both cases, c[] and c[:] are equal to c[1:size(c)], which in turn is just equal to c.\n\n\n\nStan provides head and tail functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,\nvector[M] a = ...;\nvector[N] b = head(a, N);\nassigns b to be a vector equivalent to the first N elements of the vector a. The function tail works the same way for suffixes, with\narray[M] a = ...;\narray[N] b = tail(a, N);\nFinally, there is a segment function, which specifies a first element and number of elements. For example,\narray[15] a = ...;\narray[3] b = segment(a, 5, 3);\nwill set b to be equal to { a[5], a[6], a[7] }, so that it starts at element 5 of a and includes a total of 3 elements.\n\n\n\n\nMultiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.\narray[3] int a;\narray[2] int c;\narray[2] int idxs;\n// ... define: a == (1, 2, 3);  c == (5, 9)\n               //         idxs = (3,2)\na[idxs] = c;   // result: a == (1, 9, 5)\nThe result above can be worked out by noting that the assignment sets a[idxs[1]] (a[3]) to c[1] (5) and a[idxs[2]] (a[2]) to c[2] (9).\nThe same principle applies when there are many multiple indexes, as in the following example.\narray[5, 7] int a;\narray[2, 2] int c;\n// ...\na[2:3, 5:6] = c;  // result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]\n                  //         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]\nAs in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.\nUsage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.\narray[10, 13] int a;\narray[2] int c;\n// ...\na[4, 2:3] = c;  // result:  a[4, 2] == c[1];  a[4, 3] == c[2]\n\n\nAliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array a in the following code fragment.\narray[3] int a;\n// ... define: a == (5, 6, 7)\na[2:3] = a[1:2];\n// ... result: a == (5, 5, 6)\nThe reason the value of a after the assignment is \\((5,5,6)\\) rather than \\((5,5,5)\\) is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.\narray[3] int a;\narray[3] int idxs;\n// ... define idxs = (2, 1, 3)\na[idxs] = a;\nIn this case, it is evident why the right-hand side needs to be copied before the assignment.\nIt is tempting (but wrong) to think of the assignment a[2:3] =   a[1:2] as executing the following assignments.\n// ... define: a = (5, 6, 7)\na[2] = a[1];      // result: a = (5, 5, 7)\na[3] = a[2];      // result: a = (5, 5, 5)!\nThis produces a different result than executing the assignment because a[2]’s value changes before it is used.\n\n\n\n\nMultiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.\n\n\nVectors and row vectors behave exactly the same way as arrays with multiple indexes. If v is a vector, then v[3] is a scalar real value, whereas v[2:4] is a vector of size 3 containing the elements v[2], v[3], and v[4].\nThe only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.\narray[3] vector[5] v;\narray[7] int idxs;\n// ...\nvector[7] u;\nu = v[2, idxs];\n\narray[7] real w;\nw = v[idxs, 2];\nThe key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because v is an array of vectors, v[2, idxs] reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, v[idxs, 2] does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for w is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.\n\n\n\nMatrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.\nmatrix[5, 7] m;\n// ...\nrow_vector[3] rv;\nrv = m[4, 3:5];    // result is 1 x 3\n// ...\nvector[4] v;\nv = m[2:5, 3];     // result is 3 x 1\n// ...\nmatrix[3, 4] m2;\nm2 = m[1:3, 2:5];  // result is 3 x 4\nThe key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.\n\n\n\nIf matrices receive a single multiple index, the result is a matrix. So if m is a matrix, so is m[2:4]. In contrast, supplying a single index, m[3], produces a row vector result. That is, m[3] produces the same result as m[3, ] or m[3, 1:cols(m)].\n\n\n\nWith arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.\narray[5, 7] matrix[3, 4] m;\n// ...\narray[2] matrix[3, 4] a;\na = m[1, 2:3];  // knock off first array dimension\na = m[3:4, 5];  // knock off second array dimension\nIn both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, a[i] ==   m[1, i + 1], whereas in the second case, a[i] == m[i + 2,   5].\nContinuing the previous example, consider the following.\n// ...\nvector[2] b;\nb = a[1, 3, 2:3, 2];\nHere, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, b[j] == a[1, 3, 1 + j, 2].\nThis last example illustrates an important point: if there is a lower-bounded index, such as 2:3, with lower bound 2, then the lower bound minus one is added to the index, as seen in the 1 + j expression above.\nContinuing further, consider continuing with the following.\n// ...\narray[2] row_vector[3] c;\nc = a[4:5, 3, 1, 2: ];\nHere, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by c[i, j] == a[i + 3, 3, 1, j + 1]\n\n\n\nMatrix slicing can also be performed using the block function. For example,\nmatrix[20, 20] a = ...;\nmatrix[3, 2] b = block(a, 5, 9, 3, 2);\nwill set b equal to the submatrix of a starting at index [5, 9] and extending 3 rows and 2 columns. Thus block(a, 5, 9, 3, 2) is equivalent to b[5:7, 9:10].\nThe sub_col function extracts a slice of a column of a matrix as a vector. For example,\nmatrix[10, 10] a = ...;\nvector b = sub_col(a, 2, 3, 5);\nwill set b equal to the vector a[2:6, 3], taking the element starting at [2, 3], then extending for a total of 5 rows. The function sub_row works the same way for extracting a slice of a row as a row vector. For example, sub_row(a, 2, 3, 5) is equal to the row vector a[2, 3:7], which also starts at position [2, 3] then extends for a total of 5 columns.\n\n\n\n\nSuppose you have a \\(3 \\times 3\\) matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.\nSuppose a \\(3 \\times 3\\) matrix is known to be zero at indexes \\([1,2]\\) and \\([1,3]\\). The indexes for parameters are included in a “melted” data-frame or database format.\ntransformed data {\n  array[7, 2] int&lt;lower=1, upper=3&gt; idxs\n    = { {1, 1},\n        {2, 1}, {2, 2}, {2, 3},\n        {3, 1}, {3, 2}, {3, 3} };\n  // ...\nThe seven remaining parameters are declared as a vector.\nparameters {\n  vector[7] A_raw;\n  // ...\n}\nThen the full matrix A is constructed in the model block as a local variable.\nmodel {\n  matrix[3, 3] A;\n  for (i in 1:7) {\n    A[idxs[i, 1], idxs[i, 2]] = A_raw[i];\n  }\n  A[1, 2] = 0;\n  A[1, 3] = 0;\n  // ...\n}\nThis may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the idxs array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexing",
+    "href": "stan-users-guide/multi-indexing.html#multiple-indexing",
+    "title": "Multiple Indexing and Range Indexing",
+    "section": "",
+    "text": "The following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.\narray[3] int c;\n// ... define: c == (5, 9, 7)\narray[4] int idxs;\n// ... define: idxs == (3, 3, 1, 2)\narray[4] int d;\nd = c[idxs];    // result: d == (7, 7, 5, 9)\nIn general, the multiple indexed expression c[idxs] is defined as follows, assuming idxs is of size K.\nc[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )\nThus c[idxs] is of the same size as idxs, which is K in this example.\nMultiple indexing can also be used with multi-dimensional arrays. For example, consider the following.\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[4] int idxs;\n// ... define: idxs = (2, 2, 1, 2)\narray[4, 3] int d\nd = c[idxs];    // result: d = ((7, 11, 13), (7, 11, 13),\n                //              (1, 3, 5), (7, 11, 13))\nThat is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by c[idxs][j] ==   c[idxs[j]].\nMultiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.\narray[4] int e;\ne = c[2, idxs]; // result:  c[2] = (7, 11, 13)\n                // result:  e = (11, 11, 7, 11)\nThe single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, c[2,idxs] evaluates to the same value as c[2][idxs].\nMultiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[3] int idxs1;\n// ... define: idxs1 = (2, 2, 1)\narray[2] int idxs2;\n// ... define: idxs2 = (1, 3)\narray[3, 2] int d;\nd = c[idxs1, idxs2];  // result: d = ((7, 13), (7, 13), (1, 5))\nWith multiple indexes, we no longer have c[idxs1, idxs2] being the same as c[idxs1][idxs2]. Rather, the entry d[i, j] after executing the above is given by\nd[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]\nThis example illustrates the operation of multiple indexing in the general case: a multiple index like idxs1 converts an index i used on the result (here, c[idxs1, idxs2]) to index idxs1[i] in the variable being indexed (here, c). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multi-indexing.html#slicing-with-range-indexes",
+    "href": "stan-users-guide/multi-indexing.html#slicing-with-range-indexes",
+    "title": "Multiple Indexing and Range Indexing",
+    "section": "",
+    "text": "Slicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.\n\n\nFor instance, consider supplying an upper and lower bound for an index.\narray[7] int c;\n// ...\narray[4] int d;\nd = c[3:6];  // result: d == (c[3], c[4], c[5], c[6])\nThe range index 3:6 behaves semantically just like the multiple index (3, 4, 5, 6). In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.\n\n\n\nIt is also possible to supply just a lower bound, or just an upper bound. Writing c[3:] is just shorthand for c[3:size(c)]. Writing c[:5] is just shorthand for c[1:5].\n\n\n\nFinally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (:) as the index or leaving the index position empty. In both cases, c[] and c[:] are equal to c[1:size(c)], which in turn is just equal to c.\n\n\n\nStan provides head and tail functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,\nvector[M] a = ...;\nvector[N] b = head(a, N);\nassigns b to be a vector equivalent to the first N elements of the vector a. The function tail works the same way for suffixes, with\narray[M] a = ...;\narray[N] b = tail(a, N);\nFinally, there is a segment function, which specifies a first element and number of elements. For example,\narray[15] a = ...;\narray[3] b = segment(a, 5, 3);\nwill set b to be equal to { a[5], a[6], a[7] }, so that it starts at element 5 of a and includes a total of 3 elements.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexing-on-the-left-of-assignments",
+    "href": "stan-users-guide/multi-indexing.html#multiple-indexing-on-the-left-of-assignments",
+    "title": "Multiple Indexing and Range Indexing",
+    "section": "",
+    "text": "Multiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.\narray[3] int a;\narray[2] int c;\narray[2] int idxs;\n// ... define: a == (1, 2, 3);  c == (5, 9)\n               //         idxs = (3,2)\na[idxs] = c;   // result: a == (1, 9, 5)\nThe result above can be worked out by noting that the assignment sets a[idxs[1]] (a[3]) to c[1] (5) and a[idxs[2]] (a[2]) to c[2] (9).\nThe same principle applies when there are many multiple indexes, as in the following example.\narray[5, 7] int a;\narray[2, 2] int c;\n// ...\na[2:3, 5:6] = c;  // result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]\n                  //         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]\nAs in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.\nUsage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.\narray[10, 13] int a;\narray[2] int c;\n// ...\na[4, 2:3] = c;  // result:  a[4, 2] == c[1];  a[4, 3] == c[2]\n\n\nAliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array a in the following code fragment.\narray[3] int a;\n// ... define: a == (5, 6, 7)\na[2:3] = a[1:2];\n// ... result: a == (5, 5, 6)\nThe reason the value of a after the assignment is \\((5,5,6)\\) rather than \\((5,5,5)\\) is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.\narray[3] int a;\narray[3] int idxs;\n// ... define idxs = (2, 1, 3)\na[idxs] = a;\nIn this case, it is evident why the right-hand side needs to be copied before the assignment.\nIt is tempting (but wrong) to think of the assignment a[2:3] =   a[1:2] as executing the following assignments.\n// ... define: a = (5, 6, 7)\na[2] = a[1];      // result: a = (5, 5, 7)\na[3] = a[2];      // result: a = (5, 5, 5)!\nThis produces a different result than executing the assignment because a[2]’s value changes before it is used.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexes-with-vectors-and-matrices",
+    "href": "stan-users-guide/multi-indexing.html#multiple-indexes-with-vectors-and-matrices",
+    "title": "Multiple Indexing and Range Indexing",
+    "section": "",
+    "text": "Multiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.\n\n\nVectors and row vectors behave exactly the same way as arrays with multiple indexes. If v is a vector, then v[3] is a scalar real value, whereas v[2:4] is a vector of size 3 containing the elements v[2], v[3], and v[4].\nThe only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.\narray[3] vector[5] v;\narray[7] int idxs;\n// ...\nvector[7] u;\nu = v[2, idxs];\n\narray[7] real w;\nw = v[idxs, 2];\nThe key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because v is an array of vectors, v[2, idxs] reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, v[idxs, 2] does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for w is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.\n\n\n\nMatrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.\nmatrix[5, 7] m;\n// ...\nrow_vector[3] rv;\nrv = m[4, 3:5];    // result is 1 x 3\n// ...\nvector[4] v;\nv = m[2:5, 3];     // result is 3 x 1\n// ...\nmatrix[3, 4] m2;\nm2 = m[1:3, 2:5];  // result is 3 x 4\nThe key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.\n\n\n\nIf matrices receive a single multiple index, the result is a matrix. So if m is a matrix, so is m[2:4]. In contrast, supplying a single index, m[3], produces a row vector result. That is, m[3] produces the same result as m[3, ] or m[3, 1:cols(m)].\n\n\n\nWith arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.\narray[5, 7] matrix[3, 4] m;\n// ...\narray[2] matrix[3, 4] a;\na = m[1, 2:3];  // knock off first array dimension\na = m[3:4, 5];  // knock off second array dimension\nIn both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, a[i] ==   m[1, i + 1], whereas in the second case, a[i] == m[i + 2,   5].\nContinuing the previous example, consider the following.\n// ...\nvector[2] b;\nb = a[1, 3, 2:3, 2];\nHere, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, b[j] == a[1, 3, 1 + j, 2].\nThis last example illustrates an important point: if there is a lower-bounded index, such as 2:3, with lower bound 2, then the lower bound minus one is added to the index, as seen in the 1 + j expression above.\nContinuing further, consider continuing with the following.\n// ...\narray[2] row_vector[3] c;\nc = a[4:5, 3, 1, 2: ];\nHere, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by c[i, j] == a[i + 3, 3, 1, j + 1]\n\n\n\nMatrix slicing can also be performed using the block function. For example,\nmatrix[20, 20] a = ...;\nmatrix[3, 2] b = block(a, 5, 9, 3, 2);\nwill set b equal to the submatrix of a starting at index [5, 9] and extending 3 rows and 2 columns. Thus block(a, 5, 9, 3, 2) is equivalent to b[5:7, 9:10].\nThe sub_col function extracts a slice of a column of a matrix as a vector. For example,\nmatrix[10, 10] a = ...;\nvector b = sub_col(a, 2, 3, 5);\nwill set b equal to the vector a[2:6, 3], taking the element starting at [2, 3], then extending for a total of 5 rows. The function sub_row works the same way for extracting a slice of a row as a row vector. For example, sub_row(a, 2, 3, 5) is equal to the row vector a[2, 3:7], which also starts at position [2, 3] then extends for a total of 5 columns.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/multi-indexing.html#matrices-with-parameters-and-constants",
+    "href": "stan-users-guide/multi-indexing.html#matrices-with-parameters-and-constants",
+    "title": "Multiple Indexing and Range Indexing",
+    "section": "",
+    "text": "Suppose you have a \\(3 \\times 3\\) matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.\nSuppose a \\(3 \\times 3\\) matrix is known to be zero at indexes \\([1,2]\\) and \\([1,3]\\). The indexes for parameters are included in a “melted” data-frame or database format.\ntransformed data {\n  array[7, 2] int&lt;lower=1, upper=3&gt; idxs\n    = { {1, 1},\n        {2, 1}, {2, 2}, {2, 3},\n        {3, 1}, {3, 2}, {3, 3} };\n  // ...\nThe seven remaining parameters are declared as a vector.\nparameters {\n  vector[7] A_raw;\n  // ...\n}\nThen the full matrix A is constructed in the model block as a local variable.\nmodel {\n  matrix[3, 3] A;\n  for (i in 1:7) {\n    A[idxs[i, 1], idxs[i, 2]] = A_raw[i];\n  }\n  A[1, 2] = 0;\n  A[1, 3] = 0;\n  // ...\n}\nThis may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the idxs array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html",
+    "href": "stan-users-guide/odes.html",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "Stan provides a number of different methods for solving systems of ordinary differential equations (ODEs). All of these methods adaptively refine their solutions in order to satisfy given tolerances, but internally they handle calculations quite a bit differently.\nBecause Stan’s algorithms requires gradients of the log density, the ODE solvers must not only provide the solution to the ODE itself, but also the gradient of the ODE solution with respect to parameters (the sensitivities). Two fundamentally different approaches are available in Stan to solve this problem, each having very different computational cost depending on the number of ODE states \\(N\\) and the number of parameters \\(M\\) being used:\n\nA forward sensitivity solver expands the base ODE system with additional ODE equations for the gradients of the solution. For each parameter, an additional full set of \\(N\\) sensitivity states are added meaning that the full ODE solved has \\(N \\, + N \\cdot M\\) states.\nAn adjoint sensitivity solver starts by solving the base ODE system forward in time to get the ODE solution and then solves another ODE system (the adjoint) backward in time to get the gradients. The forward and reverse solves both have \\(N\\) states each. There is additionally one quadrature problem solved for every parameter.\n\nThe adjoint sensitivity approach scales much better than the forward sensitivity approach. Whereas the computational cost of the forward approach scales multiplicatively in the number of ODE states \\(N\\) and parameters \\(M\\), the adjoint sensitivity approach scales linear in states \\(N\\) and parameters \\(M\\). However, the adjoint problem is harder to configure and the overhead for small problems actually makes it slower than solving the full forward sensitivity system. With that in mind, the rest of this introduction focuses on the forward sensitivity interfaces. For information on the adjoint sensitivity interface see the Adjoint ODE solver\nTwo interfaces are provided for each forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. For instance, if a solution component slowly varies between 3.0 and 5.0 and measurements of the ODE state are noisy, then perhaps the tolerances do not need to be as tight as for a situation where the solutions vary between 3.0 and 3.1 and very high precision measurements of the ODE state are available. It is also often useful to reduce the absolute tolerance when a component of the solution is expected to approach zero. For information on choosing tolerances, see the control parameters section.\nThe advantage of adaptive solvers is that as long as reasonable tolerances are provided and an ODE solver well-suited to the problem is chosen the technical details of solving the ODE can be abstracted away. The catch is that it is not always clear from the outset what reasonable tolerances are or which ODE solver is best suited to a problem. In addition, as changes are made to an ODE model, the optimal solver and tolerances may change.\nWith this in mind, the four forward solvers are rk45, bdf, adams, and ckrk. If no other information about the ODE is available, start with the rk45 solver. The list below has information on when each solver is useful.\nIf there is any uncertainty about which solver is the best, it can be useful to measure the performance of all the interesting solvers using profile statements. It is difficult to always know exactly what solver is the best in all situations, but a profile can provide a quick check.\n\nrk45: a fourth and fifth order Runge-Kutta method for non-stiff systems (Dormand and Prince 1980; Ahnert and Mulansky 2011). rk45 is the most generic solver and should be tried first.\nbdf: a variable-step, variable-order, backward-differentiation formula implementation for stiff systems (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). bdf is often useful for ODEs modeling chemical reactions.\nadams: a variable-step, variable-order, Adams-Moulton formula implementation for non-stiff systems (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). The method has order up to 12, hence is commonly used when high-accuracy is desired for a very smooth solution, such as in modeling celestial mechanics and orbital dynamics (Montenbruck and Gill 2000).\nckrk: a fourth and fifth order explicit Runge-Kutta method for non-stiff and semi-stiff systems (Cash and Karp 1990; Mazzia, Cash, and Soetaert 2012). The difference between ckrk and rk45 is that ckrk should perform better for systems that exhibit rapidly varying solutions. Often in those situations the derivatives become large or even nearly discontinuous, and ckrk is designed to address such problems.\n\nFor a discussion of stiff ODE systems, see the stiff ODE section. For information on the adjoint sensitivity interface see the Adjoint ODE solver section. The function signatures for Stan’s ODE solvers can be found in the function reference manual section on ODE solvers.\n\n\nAn ODE is defined by a set of differential equations, \\(y(t, \\theta)' = f(t, y, \\theta)\\), and initial conditions, \\(y(t_0, \\theta) = y_0\\). The function \\(f(t, y, \\theta)\\) is called the system function. The \\(\\theta\\) dependence is included in the notation for \\(y(t, \\theta)\\) and \\(f(t, y, \\theta)\\) as a reminder that the solution is a function of any parameters used in the computation.\n\n\n\nAs an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair \\(y = (y_1, y_2)\\) representing position and speed. The change in the system with respect to time is given by the following differential equations.1\n\\[\\begin{align*}\n&\\frac{d}{dt} y_1 = y_2  \\\\\n&\\frac{d}{dt} y_2 = -y_1 - \\theta y_2\n\\end{align*}\\]\nThe state equations implicitly defines the state at future times as a function of an initial state and the system parameters.\n\n\n\nThe first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the system function is time, passed as a real; the second argument to the system function is the system state, passed as a vector, and the return value from the system function are the current time derivatives of the state defined as a vector. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.\nThe simple harmonic oscillator can be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector sho(real t,        // time\n           vector y,      // state\n           real theta) {  // friction parameter\n  vector[2] dydt;\n  dydt[1] = y[2];\n  dydt[2] = -y[1] - theta * y[2];\n  return dydt;\n}\nThe function takes in a time t (a real), the system state y (a vector), and the parameter theta (a real). The function returns a vector of time derivatives of the system state at time t, state y, and parameter theta. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, t does not show up in the definition of dydt, however it is still required.\n\n\nThe types in the ODE system function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible ODE signatures:\nvector myode1(real t, vector y, real a0);\nvector myode2(real t, vector y, array[] int a0, vector a1);\nvector myode3(real t, vector y, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector myode1(real t, array[] real y, real a0);\n// Second argument is not a vector\narray[] real myode2(real t, vector y, real a0);\n// Return type is not a vector\nvector myode3(vector y, real a0);\n// First argument is not a real and second is not a vector\n\n\n\n\nNoisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.\n\n\nAs an example, suppose the simple harmonic oscillator has a parameter value of \\(\\theta = 0.15\\) and an initial state \\(y(t = 0, \\theta = 0.15) = (1, 0)\\). Assume the system is measured at 10 time points, \\(t = 1, 2, \\cdots, 10\\), where each measurement of \\(y(t, \\theta)\\) has independent \\(\\textsf{normal}(0, 0.1)\\) error in both dimensions (\\(y_1(t, \\theta)\\) and \\(y_2(t, \\theta)\\)).\nThe following model can be used to generate data like this:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  real t0;\n  array[T] real ts;\n  real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThe system parameters theta and initial state y0 are read in as data along with the initial time t0 and observation times ts. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations y_sim. Because the system is not stiff, the ode_rk45 solver is used.\nThis program illustrates the way in which the ODE solver is called in a Stan program,\narray[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\nthis returns the solution of the ODE initial value problem defined by system function sho, initial state y0, initial time t0, and parameter theta at the times ts. The call explicitly specifies the non-stiff RK45 solver.\nThe parameter theta is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, \\(\\theta\\) and \\(\\beta\\), the system function definition would look like:\nvector sho(real t, vector y, real theta, real beta) { ... }\nand the appropriate ODE solver call would be:\node_rk45(sho, y0, t0, ts, theta, beta);\nAny number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).\nBecause all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the T outputs with normal_rng.\n\n\n\nTypical realization of harmonic oscillator trajectory.\n\n\n\n\n\nThese ten noisy observations of the state can be used to estimate the friction parameter, \\(\\theta\\), the initial conditions, \\(y(t_0, \\theta)\\), and the scale of the noise in the problem. The full Stan model is:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  array[T] vector[2] y;\n  real t0;\n  array[T] real ts;\n}\nparameters {\n  vector[2] y0;\n  vector&lt;lower=0&gt;[2] sigma;\n  real theta;\n}\nmodel {\n  array[T] vector[2] mu = ode_rk45(sho, y0, t0, ts, theta);\n  sigma ~ normal(0, 2.5);\n  theta ~ std_normal();\n  y0 ~ std_normal();\n  for (t in 1:T) {\n    y[t] ~ normal(mu[t], sigma);\n  }\n}\nBecause the solves are now a function of model parameters, the ode_rk45 call is now made in the model block. There are half-normal priors on the measurement error scales sigma, and standard normal priors on theta and the initial state vector y0. The solutions to the ODE are assigned to mu, which is used as the location for the normal observation model.\nAs with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).\n\n\n\n\nStiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.\nStan provides a specialized solver for stiff ODEs (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the rk45 suffix is replaced with bdf, as in\node_bdf(sho, y0, t0, ts, theta);\nUsing the stiff (bdf) solver on a system that is not stiff may be much slower than using the non-stiff (rk45) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.\nIf it is not known for sure that an ODE system is stiff, run the model with both the rk45 and bdf solvers and see which is faster. If the rk45 solver is faster, then the problem is probably non-stiff, and then it makes sense to try the adams solver as well. The adams solver uses higher order methods which can take larger timesteps than the rk45 solver, though similar to the bdf solver each of these steps is more expensive to compute.\n\n\n\nFor additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters. These are the same as the regular function names but with _tol appended to the end. All three control arguments must be supplied with this signature (there are no defaults).\narray[T] vector[2] y_sim = ode_bdf_tol(sho, y0, t0, ts,\n                                 relative_tolerance,\n                                 absolute_tolerance,\n                                 max_num_steps,\n                                 theta);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nFor the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both \\(10^{-6}\\) and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million.\n\n\nIf there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.\nNonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times \\(0 &lt; t &lt; t'\\) and then positive for \\(t \\geq t'\\). In this example example, we would use code in the system such as\nif (t &lt; t_lag) {\n  return [0, 0]';\n} else {\n  // ... return non-zero vector...\n}\nIn general it is better to integrate up to t_lag in one solve and then integrate from t_lag onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.\nIf the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.\n\n\n\nThe relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns \\(y=(y_1,\\dots,y_n)^T\\), at every step the solver controls estimated local error \\(e=(e_1,\\dots,e_n)^T\\) through its weighted root-mean-square norm (Serban and Hindmarsh (2005), Hairer, Nørsett, and Wanner (1993))\n\\[\\begin{equation*}\n\\sqrt{\\sum_{i=1}^n{\\frac{1}{n}\\frac{e_i^2}{(\\text{RTOL}\\times y_i + \\text{ATOL})^2}}} &lt; 1\n\\end{equation*}\\] by reducing the stepsize when the inequality is not satisfied.\nTo understand the roles of the two tolerances it helps to assume \\(y\\) at opposite scales in the above expression: on one hand the absolute tolerance has little effect when \\(y_i \\gg 1\\), on the other the relative tolerance can not affect the norm when \\(y_i = 0\\). Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow Brenan, Campbell, and Petzold (1995) for a rule of thumb: let \\(m\\) be the number of significant digits required for \\(y\\), set \\(\\text{RTOL}=10^{-(m+1)}\\), and set ATOL at which \\(y\\) becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in bdf and adams solvers, and the same tolerances are used to control forward sensitivity calculation. See Serban and Hindmarsh (2005) for details.\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.\n\n\n\n\nThe adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with \\(N\\) additional states for each parameter for which gradients are needed. If there are \\(M\\) parameters for which sensitivities are required, then the augmented ODE system has a total of \\(N \\cdot (M +\n1)\\) states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.\nIn contrast, the adjoint ODE solver integrates forward in time a system of \\(N\\) equations to compute the ODE solution and then integrates backwards in time another system of \\(N\\) equations to get the sensitivities. Additionally, for \\(M\\) parameters there are \\(M\\) additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the bdf and adams forward sensitivity interfaces).\nThe solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.\nThe solver type (either bdf or adams) can be individually set for both the forward and backward solves.\nThe tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:\narray[T] vector[2] y_sim\n    = ode_adjoint_tol_ctl(sho, y0, t0, ts,\n                          relative_tolerance/9.0,                // forward tolerance\n                          rep_vector(absolute_tolerance/9.0, 2), // forward tolerance\n                          relative_tolerance/3.0,                // backward tolerance\n                          rep_vector(absolute_tolerance/3.0, 2), // backward tolerance\n                          relative_tolerance,                    // quadrature tolerance\n                          absolute_tolerance,                    // quadrature tolerance\n                          max_num_steps,\n                          150,                                   // number of steps between checkpoints\n                          1,                                     // interpolation polynomial: 1=Hermite, 2=polynomial\n                          2,                                     // solver for forward phase: 1=Adams, 2=BDF\n                          2,                                     // solver for backward phase: 1=Adams, 2=BDF\n                          theta);\nFor a detailed information on each argument please see the Stan function reference manual.\n\n\n\nLinear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.\nThe solution to \\(\\frac{d}{dt} y = ay\\) is \\(y = y_0e^{at}\\), where the constant \\(y_0\\) is determined by boundary conditions. We can extend this solution to the vector case: \\[\n\\frac{d}{dt}y = A \\, y\n\\]  where \\(y\\) is now a vector of length \\(n\\) and \\(A\\) is an \\(n\\) by \\(n\\) matrix. The solution is then given by: \\[\ny = e^{tA} \\, y_0\n\\]  where the matrix exponential is formally defined by the convergent power series: \\[\ne^{tA} = \\sum_{n=0}^{\\infty} \\dfrac{tA^n}{n!} = I + tA + \\frac{t^2A^2}{2!} + \\dotsb\n\\] \nWe can apply this technique to the simple harmonic oscillator example, by setting \\[\ny = \\begin{bmatrix} y_1 \\\\ y_2 \\end{bmatrix} \\qquad\nA = \\begin{bmatrix} 0 & 1 \\\\ -1 & -\\theta \\end{bmatrix}\n\\] \nThe Stan model to simulate noisy observations using a matrix exponential function is given below.\nIn general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  array[T] real ts;\n  array[1] real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim;\n  matrix[2, 2] A = [[ 0,  1],\n                    [-1, -theta[1]]]\n  for (t in 1:T) {\n    y_sim[t] = matrix_exp((t - 1) * A) * y0;\n  }\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThis Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters theta and initial state y0 are read in as data along observation times ts. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations y_sim. Because the ODEs are linear, we can use the matrix_exp function to solve the system.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#notation",
+    "href": "stan-users-guide/odes.html#notation",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "An ODE is defined by a set of differential equations, \\(y(t, \\theta)' = f(t, y, \\theta)\\), and initial conditions, \\(y(t_0, \\theta) = y_0\\). The function \\(f(t, y, \\theta)\\) is called the system function. The \\(\\theta\\) dependence is included in the notation for \\(y(t, \\theta)\\) and \\(f(t, y, \\theta)\\) as a reminder that the solution is a function of any parameters used in the computation.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#example-simple-harmonic-oscillator",
+    "href": "stan-users-guide/odes.html#example-simple-harmonic-oscillator",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "As an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair \\(y = (y_1, y_2)\\) representing position and speed. The change in the system with respect to time is given by the following differential equations.1\n\\[\\begin{align*}\n&\\frac{d}{dt} y_1 = y_2  \\\\\n&\\frac{d}{dt} y_2 = -y_1 - \\theta y_2\n\\end{align*}\\]\nThe state equations implicitly defines the state at future times as a function of an initial state and the system parameters.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#coding-the-ode-system-function",
+    "href": "stan-users-guide/odes.html#coding-the-ode-system-function",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "The first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the system function is time, passed as a real; the second argument to the system function is the system state, passed as a vector, and the return value from the system function are the current time derivatives of the state defined as a vector. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.\nThe simple harmonic oscillator can be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector sho(real t,        // time\n           vector y,      // state\n           real theta) {  // friction parameter\n  vector[2] dydt;\n  dydt[1] = y[2];\n  dydt[2] = -y[1] - theta * y[2];\n  return dydt;\n}\nThe function takes in a time t (a real), the system state y (a vector), and the parameter theta (a real). The function returns a vector of time derivatives of the system state at time t, state y, and parameter theta. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, t does not show up in the definition of dydt, however it is still required.\n\n\nThe types in the ODE system function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible ODE signatures:\nvector myode1(real t, vector y, real a0);\nvector myode2(real t, vector y, array[] int a0, vector a1);\nvector myode3(real t, vector y, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector myode1(real t, array[] real y, real a0);\n// Second argument is not a vector\narray[] real myode2(real t, vector y, real a0);\n// Return type is not a vector\nvector myode3(vector y, real a0);\n// First argument is not a real and second is not a vector",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#measurement-error-models",
+    "href": "stan-users-guide/odes.html#measurement-error-models",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "Noisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.\n\n\nAs an example, suppose the simple harmonic oscillator has a parameter value of \\(\\theta = 0.15\\) and an initial state \\(y(t = 0, \\theta = 0.15) = (1, 0)\\). Assume the system is measured at 10 time points, \\(t = 1, 2, \\cdots, 10\\), where each measurement of \\(y(t, \\theta)\\) has independent \\(\\textsf{normal}(0, 0.1)\\) error in both dimensions (\\(y_1(t, \\theta)\\) and \\(y_2(t, \\theta)\\)).\nThe following model can be used to generate data like this:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  real t0;\n  array[T] real ts;\n  real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThe system parameters theta and initial state y0 are read in as data along with the initial time t0 and observation times ts. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations y_sim. Because the system is not stiff, the ode_rk45 solver is used.\nThis program illustrates the way in which the ODE solver is called in a Stan program,\narray[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\nthis returns the solution of the ODE initial value problem defined by system function sho, initial state y0, initial time t0, and parameter theta at the times ts. The call explicitly specifies the non-stiff RK45 solver.\nThe parameter theta is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, \\(\\theta\\) and \\(\\beta\\), the system function definition would look like:\nvector sho(real t, vector y, real theta, real beta) { ... }\nand the appropriate ODE solver call would be:\node_rk45(sho, y0, t0, ts, theta, beta);\nAny number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).\nBecause all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the T outputs with normal_rng.\n\n\n\nTypical realization of harmonic oscillator trajectory.\n\n\n\n\n\nThese ten noisy observations of the state can be used to estimate the friction parameter, \\(\\theta\\), the initial conditions, \\(y(t_0, \\theta)\\), and the scale of the noise in the problem. The full Stan model is:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  array[T] vector[2] y;\n  real t0;\n  array[T] real ts;\n}\nparameters {\n  vector[2] y0;\n  vector&lt;lower=0&gt;[2] sigma;\n  real theta;\n}\nmodel {\n  array[T] vector[2] mu = ode_rk45(sho, y0, t0, ts, theta);\n  sigma ~ normal(0, 2.5);\n  theta ~ std_normal();\n  y0 ~ std_normal();\n  for (t in 1:T) {\n    y[t] ~ normal(mu[t], sigma);\n  }\n}\nBecause the solves are now a function of model parameters, the ode_rk45 call is now made in the model block. There are half-normal priors on the measurement error scales sigma, and standard normal priors on theta and the initial state vector y0. The solutions to the ODE are assigned to mu, which is used as the location for the normal observation model.\nAs with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#stiff-ode.section",
+    "href": "stan-users-guide/odes.html#stiff-ode.section",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "Stiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.\nStan provides a specialized solver for stiff ODEs (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the rk45 suffix is replaced with bdf, as in\node_bdf(sho, y0, t0, ts, theta);\nUsing the stiff (bdf) solver on a system that is not stiff may be much slower than using the non-stiff (rk45) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.\nIf it is not known for sure that an ODE system is stiff, run the model with both the rk45 and bdf solvers and see which is faster. If the rk45 solver is faster, then the problem is probably non-stiff, and then it makes sense to try the adams solver as well. The adams solver uses higher order methods which can take larger timesteps than the rk45 solver, though similar to the bdf solver each of these steps is more expensive to compute.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#control-ode.section",
+    "href": "stan-users-guide/odes.html#control-ode.section",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "For additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters. These are the same as the regular function names but with _tol appended to the end. All three control arguments must be supplied with this signature (there are no defaults).\narray[T] vector[2] y_sim = ode_bdf_tol(sho, y0, t0, ts,\n                                 relative_tolerance,\n                                 absolute_tolerance,\n                                 max_num_steps,\n                                 theta);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nFor the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both \\(10^{-6}\\) and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million.\n\n\nIf there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.\nNonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times \\(0 &lt; t &lt; t'\\) and then positive for \\(t \\geq t'\\). In this example example, we would use code in the system such as\nif (t &lt; t_lag) {\n  return [0, 0]';\n} else {\n  // ... return non-zero vector...\n}\nIn general it is better to integrate up to t_lag in one solve and then integrate from t_lag onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.\nIf the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.\n\n\n\nThe relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns \\(y=(y_1,\\dots,y_n)^T\\), at every step the solver controls estimated local error \\(e=(e_1,\\dots,e_n)^T\\) through its weighted root-mean-square norm (Serban and Hindmarsh (2005), Hairer, Nørsett, and Wanner (1993))\n\\[\\begin{equation*}\n\\sqrt{\\sum_{i=1}^n{\\frac{1}{n}\\frac{e_i^2}{(\\text{RTOL}\\times y_i + \\text{ATOL})^2}}} &lt; 1\n\\end{equation*}\\] by reducing the stepsize when the inequality is not satisfied.\nTo understand the roles of the two tolerances it helps to assume \\(y\\) at opposite scales in the above expression: on one hand the absolute tolerance has little effect when \\(y_i \\gg 1\\), on the other the relative tolerance can not affect the norm when \\(y_i = 0\\). Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow Brenan, Campbell, and Petzold (1995) for a rule of thumb: let \\(m\\) be the number of significant digits required for \\(y\\), set \\(\\text{RTOL}=10^{-(m+1)}\\), and set ATOL at which \\(y\\) becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in bdf and adams solvers, and the same tolerances are used to control forward sensitivity calculation. See Serban and Hindmarsh (2005) for details.\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#adjoint-ode.section",
+    "href": "stan-users-guide/odes.html#adjoint-ode.section",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "The adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with \\(N\\) additional states for each parameter for which gradients are needed. If there are \\(M\\) parameters for which sensitivities are required, then the augmented ODE system has a total of \\(N \\cdot (M +\n1)\\) states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.\nIn contrast, the adjoint ODE solver integrates forward in time a system of \\(N\\) equations to compute the ODE solution and then integrates backwards in time another system of \\(N\\) equations to get the sensitivities. Additionally, for \\(M\\) parameters there are \\(M\\) additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the bdf and adams forward sensitivity interfaces).\nThe solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.\nThe solver type (either bdf or adams) can be individually set for both the forward and backward solves.\nThe tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:\narray[T] vector[2] y_sim\n    = ode_adjoint_tol_ctl(sho, y0, t0, ts,\n                          relative_tolerance/9.0,                // forward tolerance\n                          rep_vector(absolute_tolerance/9.0, 2), // forward tolerance\n                          relative_tolerance/3.0,                // backward tolerance\n                          rep_vector(absolute_tolerance/3.0, 2), // backward tolerance\n                          relative_tolerance,                    // quadrature tolerance\n                          absolute_tolerance,                    // quadrature tolerance\n                          max_num_steps,\n                          150,                                   // number of steps between checkpoints\n                          1,                                     // interpolation polynomial: 1=Hermite, 2=polynomial\n                          2,                                     // solver for forward phase: 1=Adams, 2=BDF\n                          2,                                     // solver for backward phase: 1=Adams, 2=BDF\n                          theta);\nFor a detailed information on each argument please see the Stan function reference manual.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#solving-a-system-of-linear-odes-using-a-matrix-exponential",
+    "href": "stan-users-guide/odes.html#solving-a-system-of-linear-odes-using-a-matrix-exponential",
+    "title": "Ordinary Differential Equations",
+    "section": "",
+    "text": "Linear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.\nThe solution to \\(\\frac{d}{dt} y = ay\\) is \\(y = y_0e^{at}\\), where the constant \\(y_0\\) is determined by boundary conditions. We can extend this solution to the vector case: \\[\n\\frac{d}{dt}y = A \\, y\n\\]  where \\(y\\) is now a vector of length \\(n\\) and \\(A\\) is an \\(n\\) by \\(n\\) matrix. The solution is then given by: \\[\ny = e^{tA} \\, y_0\n\\]  where the matrix exponential is formally defined by the convergent power series: \\[\ne^{tA} = \\sum_{n=0}^{\\infty} \\dfrac{tA^n}{n!} = I + tA + \\frac{t^2A^2}{2!} + \\dotsb\n\\] \nWe can apply this technique to the simple harmonic oscillator example, by setting \\[\ny = \\begin{bmatrix} y_1 \\\\ y_2 \\end{bmatrix} \\qquad\nA = \\begin{bmatrix} 0 & 1 \\\\ -1 & -\\theta \\end{bmatrix}\n\\] \nThe Stan model to simulate noisy observations using a matrix exponential function is given below.\nIn general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  array[T] real ts;\n  array[1] real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim;\n  matrix[2, 2] A = [[ 0,  1],\n                    [-1, -theta[1]]]\n  for (t in 1:T) {\n    y_sim[t] = matrix_exp((t - 1) * A) * y0;\n  }\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThis Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters theta and initial state y0 are read in as data along observation times ts. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations y_sim. Because the ODEs are linear, we can use the matrix_exp function to solve the system.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/odes.html#footnotes",
+    "href": "stan-users-guide/odes.html#footnotes",
+    "title": "Ordinary Differential Equations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis example is drawn from the documentation for the Boost Numeric Odeint library (Ahnert and Mulansky 2011), which Stan uses to implement the rk45 and ckrk solver.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Ordinary Differential Equations"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/parallelization.html",
+    "href": "stan-users-guide/parallelization.html",
+    "title": "Parallelization",
+    "section": "",
+    "text": "Stan has support for different types of parallelization: multi-threading with Intel Threading Building Blocks (TBB), multi-processing with Message Passing Interface (MPI) and manycore processing with OpenCL.\nMulti-threading in Stan can be used with two mechanisms: reduce with summation and rectangular map. The latter can also be used with multi-processing.\nThe advantages of reduce with summation are:\n\nMore flexible argument interface, avoiding the packing and unpacking that is necessary with rectangular map.\nPartitions data for parallelization automatically (this is done manually in rectangular map).\nIs easier to use.\n\nThe advantages of rectangular map are:\n\nReturns a list of vectors, while the reduce summation returns only a scalar.\nCan be parallelized across multiple cores and multiple computers, while reduce summation can only parallelized across multiple cores on a single machine.\n\nThe actual speedup gained from using these functions will depend on many details. It is strongly recommended to only parallelize the computationally most expensive operations in a Stan program. Oftentimes this is the evaluation of the log likelihood for the observed data. When it is not clear which parts of the model is the most computationally expensive, we recommend using profiling, which is available in Stan 2.26 and newer.\nSince only portions of a Stan program will run in parallel, the maximal speedup one can achieve is capped, a phenomen described by Amdahl’s law.\n\n\nIt is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If g: U -&gt; real is the function and { x1, x2, ... } is an array of inputs, then that sum looks like:\ng(x1) + g(x2) + ...\nreduce_sum and reduce_sum_static are tools for parallelizing these calculations.\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g, but instead the partial sum function f: U[] -&gt; real, where f computes the partial sum corresponding to a slice of the sequence x passed in. Due to the associativity of the sum reduction it holds that:\ng(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })\n                      = f({ x1, x2 }) + f({ x3 })\n                      = f({ x1 }) + f({ x2, x3 })\n                      = f({ x1 }) + f({ x2 }) + f({ x3 })\nWith the partial sum function f: U[] -&gt; real reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load).\n\ngrainsize is the one tuning parameter. For reduce_sum, grainsize is a suggested partial sum size. A grainsize of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using reduce_sum unless time is spent carefully picking grainsize. For picking a grainsize, see details below.\nFor reduce_sum_static, grainsize specifies the maximal partial sum size. With reduce_sum_static it is more important to choose grainsize carefully since it entirely determines the partitioning of work. See details below.\nFor efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array { x1, x2, ... } and the shared arguments s1, s2, ... the effective sum (with individual terms) looks like:\ng(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...\nwhich can be written equivalently with partial sums to look like:\nf({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)\nwhere the particular slicing of the x array can change.\nGiven this, the signatures are:\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nreal reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n\nf - User defined function that computes partial sums\nx - Array to slice, each element corresponds to a term in the summation\ngrainsize - Target for size of slices\ns1, s2, ... - Arguments shared in every term\n\nThe user-defined partial sum functions have the signature:\nreal f(array[] T x_slice, int start, int end, T1 s1, T2 s2, ...)\nand take the arguments:\n\nx_slice - The subset of x (from reduce_sum / reduce_sum_static) for which this partial sum is responsible (x_slice = x[start:end])\nstart - An integer specifying the first term in the partial sum\nend - An integer specifying the last term in the partial sum (inclusive)\ns1, s2, ... - Arguments shared in every term (passed on without modification from the reduce_sum / reduce_sum_static call)\n\nThe user-provided function f is expected to compute the partial sum with the terms start through end of the overall sum. The user function is passed the subset x[start:end] as x_slice. start and end are passed so that f can index any of the tailing sM arguments as necessary. The trailing sM arguments are passed without modification to every call of f.\nA reduce_sum (or reduce_sum_static) call:\nreal sum = reduce_sum(f, x, grainsize, s1, s2, ...);\ncan be replaced by either:\nreal sum = f(x, 1, size(x), s1, s2, ...);\nor the code:\nreal sum = 0.0;\nfor(i in 1:size(x)) {\n  sum += f({ x[i] }, i, i, s1, s2, ...);\n}\n\n\nLogistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * x);\n}\nIn this model predictions are made about the N outputs y using the covariate x. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:\ny ~ bernoulli_logit(beta[1] + beta[2] * x);\ncan be rewritten (up to a proportionality constant) as:\nfor(n in 1:N) {\n  target += bernoulli_logit_lpmf(y[n] | beta[1] + beta[2] * x[n])\n}\nNow it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in Reduce-Sum, such a function can be written like:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\nThe likelihood statement in the model can now be written:\ntarget += partial_sum(y, 1, N, x, beta); // Sum terms 1 to N of the likelihood\nIn this example, y was chosen to be sliced over because there is one term in the summation per value of y. Technically x would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g. slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because x is a shared argument, it is subset accordingly with start:end. With this function, reduce summation can be used to automatically parallelize the likelihood:\nint grainsize = 1;\ntarget += reduce_sum(partial_sum, y,\n                     grainsize,\n                     x, beta);\nThe reduce summation facility automatically breaks the sum into pieces and computes them in parallel. grainsize = 1 specifies that the grainsize should be estimated automatically. The final model is:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  int grainsize = 1;\n  beta ~ std_normal();\n  target += reduce_sum(partial_sum, y,\n                       grainsize,\n                       x, beta);\n}\n\n\n\nThe rational for choosing a sensible grainsize is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.\nIn reduce_sum, grainsize is a recommendation on how to partition the work in the partial sum into smaller pieces. A grainsize of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.\nIn reduce_sum_static, grainsize is an upper limit on the worksize. Work will be split until all partial sums are just smaller than grainsize (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible grainsize.\nIn order to figure out an optimal grainsize, if there are N terms and M cores, run a quick test model with grainsize set roughly to N / M. Record the time, cut the grainsize in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable grainsize for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.\nFor instance, in a model with N=10000 and M = 4, start with grainsize = 2500, and sequentially try grainsize = 1250, grainsize = 625, etc.\nIt is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller grainsize that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a grainsize of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.\n\n\n\n\nMap-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).\nA map function is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, \\(f(x) = x^2\\), over the vector \\([3, 5, 10]\\) produces the vector \\([9, 25, 100]\\). In other words, map applies the square function elementwise.\nThe output of mapping a sequence is often fed into a reduction. A reduction function takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, map-reduce.\n\n\nIn order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.\n\n\n\nStan’s map function has the following signature\nvector map_rect((vector, vector, array[] real, array[] int):vector f,\n                vector phi, array[] vector thetas,\n                data array[,] real x_rs, data array[,] int x_is);\nThe arrays thetas of parameters, x_rs of real data, and x_is of integer data have the suffix “s” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function f. The value of phi is reused in each mapped operation.\nThe _rect suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.\nThe last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the data qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.\nThe notation (vector, vector, array[] real, array[] int):vector indicates that the function argument f must have the following signature.\nvector f(vector phi, vector theta,\n         data array[] real x_r, data array[] int x_i);\nAlthough f will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.\n\n\nStan’s map function applies the function f to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments theta, x_r, and x_i must be arrays of the same size. If the arrays are all size N, the result is defined as follows.\nmap_rect(f, phi, thetas, xs, ns)\n= f(phi, thetas[1], xs[1], ns[1]) . f(phi, thetas[2], xs[2], ns[2])\n  . ... . f(phi, thetas[N], xs[N], ns[N])\nThe dot operators in the notation above are meant to indicate concatenation (implemented as append_row in Stan). The output of each application of f is a vector, and the sequence of N vectors is concatenated together to return a single vector.\n\n\n\n\nAn example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.\n\n\nConsider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.\ndata {\n  array[12] int y;\n  array[12] real x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * to_vector(x));\n}\nThe program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The bernoulli_logit distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.\n\n\n\nThe unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.\nfunctions {\n  vector lr(vector beta, vector theta, array[] real x, array[] int y) {\n    real lp = bernoulli_logit_lpmf(y | beta[1]\n                                       + to_vector(x) * beta[2]);\n    return [lp]';\n  }\n}\ndata {\n  array[12] int y;\n  array[12] real x;\n}\ntransformed data {\n  // K = 3 shards\n  array[3, 4] int ys = { y[1:4], y[5:8], y[9:12] };\n  array[3, 4] real xs = { x[1:4], x[5:8], x[9:12] };\n  array[3] vector[0] theta;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  target += sum(map_rect(lr, beta, theta, xs, ys));\n}\nThe first piece of the code is the actual function to compute the logistic regression. The argument beta will contain the regression coefficients (intercept and slope), as before. The second argument theta of job-specific parameters is not used, but nevertheless must be present. The modeled data y is passed as an array of integers and the predictors x as an array of real values. The function body then computes the log probability mass of y and assigns it to the local variable lp. This variable is then used in [lp]' to construct a row vector and then transpose it to a vector to return.\nThe data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.1\nThe value 3 is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array ys contains the modeled data variables; each element of the array ys is an array of size four. The second array xs is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors x stand in a one-to-one relationship with the modeled data y. The final array theta is also of size three; its elements are empty vectors, because there are no shard-specific parameters.\nThe parameters and the prior are as before. The likelihood is now coded using map-reduce. The function lr to compute the log probability mass is mapped over the data xs and ys, which contain the original predictors and outcomes broken into shards. The parameters beta are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters theta contains only empty vectors.\n\n\n\n\nConsider a hierarchical model of American presidential voting behavior based on state of residence.2\nEach of the fifty states \\(k \\in \\{1,\\dotsc,50\\}\\) will have its own slope \\(\\beta_k\\) and intercept \\(\\alpha_k\\) to model the log odds of voting for the Republican candidate as a function of income. Suppose there are \\(N\\) voters and with voter \\(n \\in 1{:}N\\) being in state \\(s[n]\\) with income \\(x_n\\). The data model for the vote \\(y_n \\in \\{ 0, 1 \\}\\) is \\[\ny_n \\sim \\textsf{Bernoulli}\n\\Big(\n  \\operatorname{logit}^{-1}\\left( \\alpha_{s[n]} + \\beta_{s[n]} \\, x_n \\right)\n\\Big).\n\\]\nThe slopes and intercepts get hierarchical priors, \\[\\begin{align*}\n\\alpha_k &\\sim \\textsf{normal}(\\mu_{\\alpha}, \\sigma_{\\alpha}) \\\\\n\\beta_k  &\\sim \\textsf{normal}(\\mu_{\\beta}, \\sigma_{\\beta})\n\\end{align*}\\]\n\n\nThis model can be coded up in Stan directly as follows.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  matrix[K, 2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  for (i in 1:2) {\n    beta[ , i] ~ normal(mu[i], sigma[i]);\n  }\n  y ~ bernoulli_logit(beta[kk, 1] + beta[kk, 2] .* x);\n}\nFor this model the vector of predictors x is coded as a vector, corresponding to how it is used in the model. The priors for mu and sigma are vectorized. The priors on the two components of beta (intercept and slope, respectively) are stored in a \\(K \\times 2\\) matrix.\nThe distribution statement is also vectorized using multi-indexing with index kk for the states and elementwise multiplication (.*) for the income x. The vectorized distribution statement works out to the same thing as the following less efficient looped form.\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(beta[kk[n], 1] + beta[kk[n], 2] * x[n]);\n}\n\n\n\nThe mapped version of the model will map over the states K. This means the group-level parameters, real data, and integer-data must be arrays of the same size.\nThe mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name bernoulli_logit_glm derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).\nfunctions {\n vector bl_glm(vector mu_sigma, vector beta,\n               array[] real x, array[] int y) {\n   vector[2] mu = mu_sigma[1:2];\n   vector[2] sigma = mu_sigma[3:4];\n   real lp = normal_lpdf(beta | mu, sigma);\n   real ll = bernoulli_logit_lpmf(y | beta[1] + beta[2] * to_vector(x));\n   return [lp + ll]';\n }\n}\nThe shared parameter mu_sigma contains the locations (mu_sigma[1:2]) and scales (mu_sigma[3:4]) of the priors, which are extracted in the first two lines of the program. The variable lp is assigned the log density of the prior on beta. The vector beta is of size two, as are the vectors mu and sigma, so everything lines up for the vectorization. Next, the variable ll is assigned to the log likelihood contribution for the group. Here beta[1] is the intercept of the regression and beta[2] the slope. The predictor array x needs to be converted to a vector allow the multiplication.\nThe data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\ntransformed data {\n  int&lt;lower=0&gt; J = N / K;\n  array[K, J] real x_r;\n  array[K, J] int&lt;lower=0, upper=1&gt; x_i;\n  {\n    int pos = 1;\n    for (k in 1:K) {\n      int end = pos + J - 1;\n      x_r[k] = to_array_1d(x[pos:end]);\n      x_i[k] = to_array_1d(y[pos:end]);\n      pos += J;\n    }\n  }\n}\nThe integer J is set to the number of observations per group.3\nThe real data array x_r holds the predictors and the integer data array x_i holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector x (and converting it to an array) and slicing the outcome array y.\nGiven the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters mu and sigma, but moves the prior for beta and the likelihood to the mapped function.\nparameters {\n  array[K] vector[2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  target += sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));\n                         \n}\nThe model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.\n\n\n\n\nThe previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by map_rect.\n\n\nIf each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.\n\n\n\nThe output of each mapped function is concatenated in order of inputs to produce the output of map_rect. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the to_matrix function if they are rectangular.\nIf the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.\n\n\n\n\n\nOpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.\nStan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the STAN_OPENCL makefile flag. Setting this flag means that the Stan-to-C++ translator (stanc3) will be supplied the --use-opencl flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.\nIn Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:\n\nbernoulli_lpmf\nbernoulli_logit_lpmf\nbernoulli_logit_glm_lpmf*\nbeta_lpdf\nbeta_proportion_lpdf\nbinomial_lpmf\ncategorical_logit_glm_lpmf*\ncauchy_lpdf\nchi_square_lpdf\ndouble_exponential_lpdf\nexp_mod_normal_lpdf\nexponential_lpdf\nfrechet_lpdf\ngamma_lpdf\ngumbel_lpdf\ninv_chi_square_lpdf\ninv_gamma_lpdf\nlogistic_lpdf\nlognormal_lpdf\nneg_binomial_lpmf\nneg_binomial_2_lpmf\nneg_binomial_2_log_lpmf\nneg_binomial_2_log_glm_lpmf*\nnormal_lpdf\nnormal_id_glm_lpdf*\nordered_logistic_glm_lpmf*\npareto_lpdf\npareto_type_2_lpdf\npoisson_lpmf\npoisson_log_lpmf\npoisson_log_glm_lpmf*\nrayleigh_lpdf\nscaled_inv_chi_square_lpdf\nskew_normal_lpdf\nstd_normal_lpdf\nstudent_t_lpdf\nuniform_lpdf\nweibull_lpdf\n\n* OpenCL is not used when the covariate argument to the GLM functions is a row_vector.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/parallelization.html#reduce-sum",
+    "href": "stan-users-guide/parallelization.html#reduce-sum",
+    "title": "Parallelization",
+    "section": "",
+    "text": "It is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If g: U -&gt; real is the function and { x1, x2, ... } is an array of inputs, then that sum looks like:\ng(x1) + g(x2) + ...\nreduce_sum and reduce_sum_static are tools for parallelizing these calculations.\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g, but instead the partial sum function f: U[] -&gt; real, where f computes the partial sum corresponding to a slice of the sequence x passed in. Due to the associativity of the sum reduction it holds that:\ng(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })\n                      = f({ x1, x2 }) + f({ x3 })\n                      = f({ x1 }) + f({ x2, x3 })\n                      = f({ x1 }) + f({ x2 }) + f({ x3 })\nWith the partial sum function f: U[] -&gt; real reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load).\n\ngrainsize is the one tuning parameter. For reduce_sum, grainsize is a suggested partial sum size. A grainsize of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using reduce_sum unless time is spent carefully picking grainsize. For picking a grainsize, see details below.\nFor reduce_sum_static, grainsize specifies the maximal partial sum size. With reduce_sum_static it is more important to choose grainsize carefully since it entirely determines the partitioning of work. See details below.\nFor efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array { x1, x2, ... } and the shared arguments s1, s2, ... the effective sum (with individual terms) looks like:\ng(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...\nwhich can be written equivalently with partial sums to look like:\nf({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)\nwhere the particular slicing of the x array can change.\nGiven this, the signatures are:\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nreal reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n\nf - User defined function that computes partial sums\nx - Array to slice, each element corresponds to a term in the summation\ngrainsize - Target for size of slices\ns1, s2, ... - Arguments shared in every term\n\nThe user-defined partial sum functions have the signature:\nreal f(array[] T x_slice, int start, int end, T1 s1, T2 s2, ...)\nand take the arguments:\n\nx_slice - The subset of x (from reduce_sum / reduce_sum_static) for which this partial sum is responsible (x_slice = x[start:end])\nstart - An integer specifying the first term in the partial sum\nend - An integer specifying the last term in the partial sum (inclusive)\ns1, s2, ... - Arguments shared in every term (passed on without modification from the reduce_sum / reduce_sum_static call)\n\nThe user-provided function f is expected to compute the partial sum with the terms start through end of the overall sum. The user function is passed the subset x[start:end] as x_slice. start and end are passed so that f can index any of the tailing sM arguments as necessary. The trailing sM arguments are passed without modification to every call of f.\nA reduce_sum (or reduce_sum_static) call:\nreal sum = reduce_sum(f, x, grainsize, s1, s2, ...);\ncan be replaced by either:\nreal sum = f(x, 1, size(x), s1, s2, ...);\nor the code:\nreal sum = 0.0;\nfor(i in 1:size(x)) {\n  sum += f({ x[i] }, i, i, s1, s2, ...);\n}\n\n\nLogistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * x);\n}\nIn this model predictions are made about the N outputs y using the covariate x. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:\ny ~ bernoulli_logit(beta[1] + beta[2] * x);\ncan be rewritten (up to a proportionality constant) as:\nfor(n in 1:N) {\n  target += bernoulli_logit_lpmf(y[n] | beta[1] + beta[2] * x[n])\n}\nNow it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in Reduce-Sum, such a function can be written like:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\nThe likelihood statement in the model can now be written:\ntarget += partial_sum(y, 1, N, x, beta); // Sum terms 1 to N of the likelihood\nIn this example, y was chosen to be sliced over because there is one term in the summation per value of y. Technically x would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g. slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because x is a shared argument, it is subset accordingly with start:end. With this function, reduce summation can be used to automatically parallelize the likelihood:\nint grainsize = 1;\ntarget += reduce_sum(partial_sum, y,\n                     grainsize,\n                     x, beta);\nThe reduce summation facility automatically breaks the sum into pieces and computes them in parallel. grainsize = 1 specifies that the grainsize should be estimated automatically. The final model is:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  int grainsize = 1;\n  beta ~ std_normal();\n  target += reduce_sum(partial_sum, y,\n                       grainsize,\n                       x, beta);\n}\n\n\n\nThe rational for choosing a sensible grainsize is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.\nIn reduce_sum, grainsize is a recommendation on how to partition the work in the partial sum into smaller pieces. A grainsize of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.\nIn reduce_sum_static, grainsize is an upper limit on the worksize. Work will be split until all partial sums are just smaller than grainsize (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible grainsize.\nIn order to figure out an optimal grainsize, if there are N terms and M cores, run a quick test model with grainsize set roughly to N / M. Record the time, cut the grainsize in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable grainsize for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.\nFor instance, in a model with N=10000 and M = 4, start with grainsize = 2500, and sequentially try grainsize = 1250, grainsize = 625, etc.\nIt is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller grainsize that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a grainsize of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/parallelization.html#map-rect",
+    "href": "stan-users-guide/parallelization.html#map-rect",
+    "title": "Parallelization",
+    "section": "",
+    "text": "Map-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).\nA map function is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, \\(f(x) = x^2\\), over the vector \\([3, 5, 10]\\) produces the vector \\([9, 25, 100]\\). In other words, map applies the square function elementwise.\nThe output of mapping a sequence is often fed into a reduction. A reduction function takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, map-reduce.\n\n\nIn order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.\n\n\n\nStan’s map function has the following signature\nvector map_rect((vector, vector, array[] real, array[] int):vector f,\n                vector phi, array[] vector thetas,\n                data array[,] real x_rs, data array[,] int x_is);\nThe arrays thetas of parameters, x_rs of real data, and x_is of integer data have the suffix “s” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function f. The value of phi is reused in each mapped operation.\nThe _rect suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.\nThe last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the data qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.\nThe notation (vector, vector, array[] real, array[] int):vector indicates that the function argument f must have the following signature.\nvector f(vector phi, vector theta,\n         data array[] real x_r, data array[] int x_i);\nAlthough f will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.\n\n\nStan’s map function applies the function f to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments theta, x_r, and x_i must be arrays of the same size. If the arrays are all size N, the result is defined as follows.\nmap_rect(f, phi, thetas, xs, ns)\n= f(phi, thetas[1], xs[1], ns[1]) . f(phi, thetas[2], xs[2], ns[2])\n  . ... . f(phi, thetas[N], xs[N], ns[N])\nThe dot operators in the notation above are meant to indicate concatenation (implemented as append_row in Stan). The output of each application of f is a vector, and the sequence of N vectors is concatenated together to return a single vector.\n\n\n\n\nAn example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.\n\n\nConsider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.\ndata {\n  array[12] int y;\n  array[12] real x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * to_vector(x));\n}\nThe program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The bernoulli_logit distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.\n\n\n\nThe unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.\nfunctions {\n  vector lr(vector beta, vector theta, array[] real x, array[] int y) {\n    real lp = bernoulli_logit_lpmf(y | beta[1]\n                                       + to_vector(x) * beta[2]);\n    return [lp]';\n  }\n}\ndata {\n  array[12] int y;\n  array[12] real x;\n}\ntransformed data {\n  // K = 3 shards\n  array[3, 4] int ys = { y[1:4], y[5:8], y[9:12] };\n  array[3, 4] real xs = { x[1:4], x[5:8], x[9:12] };\n  array[3] vector[0] theta;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  target += sum(map_rect(lr, beta, theta, xs, ys));\n}\nThe first piece of the code is the actual function to compute the logistic regression. The argument beta will contain the regression coefficients (intercept and slope), as before. The second argument theta of job-specific parameters is not used, but nevertheless must be present. The modeled data y is passed as an array of integers and the predictors x as an array of real values. The function body then computes the log probability mass of y and assigns it to the local variable lp. This variable is then used in [lp]' to construct a row vector and then transpose it to a vector to return.\nThe data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.1\nThe value 3 is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array ys contains the modeled data variables; each element of the array ys is an array of size four. The second array xs is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors x stand in a one-to-one relationship with the modeled data y. The final array theta is also of size three; its elements are empty vectors, because there are no shard-specific parameters.\nThe parameters and the prior are as before. The likelihood is now coded using map-reduce. The function lr to compute the log probability mass is mapped over the data xs and ys, which contain the original predictors and outcomes broken into shards. The parameters beta are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters theta contains only empty vectors.\n\n\n\n\nConsider a hierarchical model of American presidential voting behavior based on state of residence.2\nEach of the fifty states \\(k \\in \\{1,\\dotsc,50\\}\\) will have its own slope \\(\\beta_k\\) and intercept \\(\\alpha_k\\) to model the log odds of voting for the Republican candidate as a function of income. Suppose there are \\(N\\) voters and with voter \\(n \\in 1{:}N\\) being in state \\(s[n]\\) with income \\(x_n\\). The data model for the vote \\(y_n \\in \\{ 0, 1 \\}\\) is \\[\ny_n \\sim \\textsf{Bernoulli}\n\\Big(\n  \\operatorname{logit}^{-1}\\left( \\alpha_{s[n]} + \\beta_{s[n]} \\, x_n \\right)\n\\Big).\n\\]\nThe slopes and intercepts get hierarchical priors, \\[\\begin{align*}\n\\alpha_k &\\sim \\textsf{normal}(\\mu_{\\alpha}, \\sigma_{\\alpha}) \\\\\n\\beta_k  &\\sim \\textsf{normal}(\\mu_{\\beta}, \\sigma_{\\beta})\n\\end{align*}\\]\n\n\nThis model can be coded up in Stan directly as follows.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  matrix[K, 2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  for (i in 1:2) {\n    beta[ , i] ~ normal(mu[i], sigma[i]);\n  }\n  y ~ bernoulli_logit(beta[kk, 1] + beta[kk, 2] .* x);\n}\nFor this model the vector of predictors x is coded as a vector, corresponding to how it is used in the model. The priors for mu and sigma are vectorized. The priors on the two components of beta (intercept and slope, respectively) are stored in a \\(K \\times 2\\) matrix.\nThe distribution statement is also vectorized using multi-indexing with index kk for the states and elementwise multiplication (.*) for the income x. The vectorized distribution statement works out to the same thing as the following less efficient looped form.\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(beta[kk[n], 1] + beta[kk[n], 2] * x[n]);\n}\n\n\n\nThe mapped version of the model will map over the states K. This means the group-level parameters, real data, and integer-data must be arrays of the same size.\nThe mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name bernoulli_logit_glm derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).\nfunctions {\n vector bl_glm(vector mu_sigma, vector beta,\n               array[] real x, array[] int y) {\n   vector[2] mu = mu_sigma[1:2];\n   vector[2] sigma = mu_sigma[3:4];\n   real lp = normal_lpdf(beta | mu, sigma);\n   real ll = bernoulli_logit_lpmf(y | beta[1] + beta[2] * to_vector(x));\n   return [lp + ll]';\n }\n}\nThe shared parameter mu_sigma contains the locations (mu_sigma[1:2]) and scales (mu_sigma[3:4]) of the priors, which are extracted in the first two lines of the program. The variable lp is assigned the log density of the prior on beta. The vector beta is of size two, as are the vectors mu and sigma, so everything lines up for the vectorization. Next, the variable ll is assigned to the log likelihood contribution for the group. Here beta[1] is the intercept of the regression and beta[2] the slope. The predictor array x needs to be converted to a vector allow the multiplication.\nThe data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\ntransformed data {\n  int&lt;lower=0&gt; J = N / K;\n  array[K, J] real x_r;\n  array[K, J] int&lt;lower=0, upper=1&gt; x_i;\n  {\n    int pos = 1;\n    for (k in 1:K) {\n      int end = pos + J - 1;\n      x_r[k] = to_array_1d(x[pos:end]);\n      x_i[k] = to_array_1d(y[pos:end]);\n      pos += J;\n    }\n  }\n}\nThe integer J is set to the number of observations per group.3\nThe real data array x_r holds the predictors and the integer data array x_i holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector x (and converting it to an array) and slicing the outcome array y.\nGiven the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters mu and sigma, but moves the prior for beta and the likelihood to the mapped function.\nparameters {\n  array[K] vector[2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  target += sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));\n                         \n}\nThe model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.\n\n\n\n\nThe previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by map_rect.\n\n\nIf each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.\n\n\n\nThe output of each mapped function is concatenated in order of inputs to produce the output of map_rect. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the to_matrix function if they are rectangular.\nIf the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/parallelization.html#opencl",
+    "href": "stan-users-guide/parallelization.html#opencl",
+    "title": "Parallelization",
+    "section": "",
+    "text": "OpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.\nStan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the STAN_OPENCL makefile flag. Setting this flag means that the Stan-to-C++ translator (stanc3) will be supplied the --use-opencl flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.\nIn Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:\n\nbernoulli_lpmf\nbernoulli_logit_lpmf\nbernoulli_logit_glm_lpmf*\nbeta_lpdf\nbeta_proportion_lpdf\nbinomial_lpmf\ncategorical_logit_glm_lpmf*\ncauchy_lpdf\nchi_square_lpdf\ndouble_exponential_lpdf\nexp_mod_normal_lpdf\nexponential_lpdf\nfrechet_lpdf\ngamma_lpdf\ngumbel_lpdf\ninv_chi_square_lpdf\ninv_gamma_lpdf\nlogistic_lpdf\nlognormal_lpdf\nneg_binomial_lpmf\nneg_binomial_2_lpmf\nneg_binomial_2_log_lpmf\nneg_binomial_2_log_glm_lpmf*\nnormal_lpdf\nnormal_id_glm_lpdf*\nordered_logistic_glm_lpmf*\npareto_lpdf\npareto_type_2_lpdf\npoisson_lpmf\npoisson_log_lpmf\npoisson_log_glm_lpmf*\nrayleigh_lpdf\nscaled_inv_chi_square_lpdf\nskew_normal_lpdf\nstd_normal_lpdf\nstudent_t_lpdf\nuniform_lpdf\nweibull_lpdf\n\n* OpenCL is not used when the covariate argument to the GLM functions is a row_vector.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/parallelization.html#footnotes",
+    "href": "stan-users-guide/parallelization.html#footnotes",
+    "title": "Parallelization",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe term “shard” is borrowed from databases, where it refers to a slice of the rows of a database. That is exactly what it is here if we think of rows of a dataframe. Stan’s shards are more general in that they need not correspond to rows of a dataframe.↩︎\nThis example is a simplified form of the model described in (Gelman and Hill 2007, sec. 14.2)↩︎\nThis makes the strong assumption that each group has the same number of observations!↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html",
+    "href": "stan-users-guide/posterior-predictive-checks.html",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "Posterior predictive checks are a way of measuring whether a model does a good job of capturing relevant aspects of the data, such as means, standard deviations, and quantiles (Rubin 1984; Andrew Gelman, Meng, and Stern 1996). Posterior predictive checking works by simulating new replicated data sets based on the fitted model parameters and then comparing statistics applied to the replicated data set with the same statistic applied to the original data set.\nPrior predictive checks evaluate the prior the same way. Specifically, they evaluate what data sets would be consistent with the prior. They will not be calibrated with actual data, but extreme values help diagnose priors that are either too strong, too weak, poorly shaped, or poorly located.\nPrior and posterior predictive checks are two cases of the general concept of predictive checks, just conditioning on different things (no data and the observed data, respectively). For hierarchical models, there are intermediate versions, as discussed in the section on hierarchical models and mixed replication.\n\n\nThe posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data \\(y\\).\nThe posterior predictive distribution for replications \\(y^{\\textrm{rep}}\\) of the original data set \\(y\\) given model parameters \\(\\theta\\) is defined by \\[\np(y^{\\textrm{rep}} \\mid y)\n= \\int p(y^{\\textrm{rep}} \\mid \\theta)\n       \\cdot p(\\theta \\mid y) \\, \\textrm{d}\\theta.\n\\]\nAs with other posterior predictive quantities, generating a replicated data set \\(y^{\\textrm{rep}}\\) from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters \\(\\theta = (\\alpha, \\beta, \\sigma).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  y ~ normal(alpha + beta * x, sigma);\n}\nTo generate a replicated data set y_rep for this simple model, the following generated quantities block suffices.\ngenerated quantities {\n  array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n}\nThe vectorized form of the normal random number generator is used with the original predictors x and the model parameters alpha, beta, and sigma. The replicated data variable y_rep is declared to be the same size as the original data y, but instead of a vector type, it is declared to be an array of reals to match the return type of the function normal_rng. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.\nThe posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in the chapter on posterior predictions in that the original predictors \\(x\\) are used. That is, the posterior predictions are for the original data.\n\n\n\nA standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.\nHere’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\ntransformed data {\n  real&lt;lower=0&gt; mean_y = mean(to_vector(y));\n  real&lt;lower=0&gt; sd_y = sd(to_vector(y));\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  y ~ poisson(lambda);\n  lambda ~ exponential(0.2);\n}\ngenerated quantities {\n  array[N] int&lt;lower=0&gt; y_rep = poisson_rng(rep_array(lambda, N));\n  real&lt;lower=0&gt; mean_y_rep = mean(to_vector(y_rep));\n  real&lt;lower=0&gt; sd_y_rep = sd(to_vector(y_rep));\n  int&lt;lower=0, upper=1&gt; mean_gte = (mean_y_rep &gt;= mean_y);\n  int&lt;lower=0, upper=1&gt; sd_gte = (sd_y_rep &gt;= sd_y);\n}\nThe generated quantities block creates a variable y_rep for the replicated data, variables mean_y_rep and sd_y_rep for the statistics of the replicated data, and indicator variables mean_gte and sd_gte for whether the replicated statistic is greater than or equal to the statistic applied to the original data.\nNow consider generating data \\(y \\sim \\textrm{Poisson}(5)\\). The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.\n\n\n\nPosterior predictive checks for Poisson data generating process and Poisson model.\n\n\nWith a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.\nNow consider generating over-dispersed data \\(y \\sim \\textrm{negative-binomial2}(5, 1).\\) This has the same mean as \\(\\textrm{Poisson}(5)\\), namely \\(5\\), but a standard deviation of \\(\\sqrt{5 + 5^2 /1} \\approx 5.5.\\) There is no way to fit this data with the Poisson model, because a variable distributed as \\(\\textrm{Poisson}(\\lambda)\\) has mean \\(\\lambda\\) and standard deviation \\(\\sqrt{\\lambda},\\) which is \\(\\sqrt{5}\\) for \\(\\textrm{Poisson}(5).\\) Here’s the resulting small multiples plot, again with original data in the upper left.\n\n\n\nPosterior predictive checks for negative binomial data generating process and Poisson model.\n\n\nThis time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.\n\n\n\nIf a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic \\(s(\\cdot)\\) in a replicated data set exceeds that in the original data, \\[\n\\Pr\\!\\left[ s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right]\n=\n\\int\n\\textrm{I}\\left( s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right)\n\\cdot p\\left( y^{\\textrm{rep}} \\mid y \\right)\n\\, \\textrm{d}{y^{\\textrm{rep}}}.\n\\] It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified (Bayarri and Berger 2000).\nNevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.\nTo calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mean_gt;\n  int&lt;lower=0, upper=1&gt; sd_gt;\n  {\n    array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n    mean_gt = mean(y_rep) &gt; mean(y);\n    sd_gt = sd(y_rep) &gt; sd(y);\n  }\n}\nThe indicator variable mean_gt will have value 1 if the mean of the simulated data y_rep is greater than or equal to the mean of he original data y. Because the values of y_rep are not needed for the posterior predictive checks, the program saves output space by using a local variable for y_rep. The statistics mean(u) and sd(y) could also be computed in the transformed data block and saved.\nFor the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.\n\n\n\nHistogram of means of replicated data sets; vertical red line at mean of original data.\n\n\nThe p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior \\(p\\)-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable mean_gt.\nThe standard deviation statistic tells a different story.\n\n\n\nScatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.\n\n\nHere, the original data has much higher standard deviation than any of the replicated data sets. The resulting \\(p\\)-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small \\(p\\)-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.\n\n\nAny statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.\nDespite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.\n\n\n\n\nPrior predictive checks generate data according to the prior in order to asses whether a prior is appropriate (Gabry et al. 2019). A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, \\[\ny^{\\textrm{sim}} \\sim p(y).\n\\] The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.\nThis is easy to carry out mechanically by simulating parameters \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta)\n\\] according to the priors, then simulating data \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}})\n\\] according to the data model given the simulated parameters. The result is a simulation from the joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta)\n\\] and thus \\[\ny^{\\textrm{sim}} \\sim p(y)\n\\] is a simulation from the prior predictive distribution.\n\n\nA prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.\nPrior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n}\ngenerated quantities {\n  real alpha = normal_rng(0, 1);\n  real beta = normal_rng(0, 1);\n  array[N] real y_sim = poisson_log_rng(alpha + beta * x);\n}\nRunning this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.\n\n\n\n\nSuppose we have a model for a football (aka soccer) league where there are \\(J\\) teams. Each team has a scoring rate \\(\\lambda_j\\) and in each game will be assumed to score \\(\\textrm{poisson}(\\lambda_j)\\) points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed \\[\n\\lambda_j \\sim \\textrm{gamma}(\\epsilon_1, \\epsilon_2).\n\\] This is not just a manufactured example; The BUGS Book recommends setting \\(\\epsilon = (0.5, 0.00001)\\), which corresponds to a Jeffreys prior for a Poisson rate parameter prior (Lunn et al. 2012, 85).\nSuppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.\ndata {\n  int&lt;lower=0&gt; J;\n  array[2] real&lt;lower=0&gt; epsilon;\n}\ngenerated quantities {\n  array[J] real&lt;lower=0&gt; lambda;\n  array[J, J] int y;\n  for (j in 1:J) lambda[j] = gamma_rng(epsilon[1], epsilon[2]);\n  for (i in 1:J) {\n    for (j in 1:J) {\n      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);\n    }\n  }\n}\nIn this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are \\(\\binom{J}{2} = \\frac{J \\cdot (J - 1)}{2}.\\)\nUsing the \\(\\textrm{gamma}(0.5, 0.00001)\\) reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, \\(y^{(1:20)}_{1, 2}\\).\n2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987\n7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66\nThat’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this \\(\\textrm{gamma}(0.5, 0.00001)\\) prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.\nThe posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values (A. Gelman 2006).\nJust as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.\nIn this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see (Gabry et al. 2019).\nPrior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.\n\n\n\nAndrew Gelman, Meng, and Stern (1996) discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.\nFor example, consider a simple varying intercept logistic regression, with intercepts \\(\\alpha_k\\) for \\(k \\in 1:K\\). Each data item \\(y_n \\in \\{ 0, 1 \\}\\) is assumed to correspond to group \\(kk_n \\in 1:K.\\) The data model is thus \\[\ny_n \\sim \\textrm{bernoulli}(\\textrm{logit}^{-1}(\\alpha_{kk[n]})).\n\\] The varying intercepts have a hierarchical normal prior, \\[\n\\alpha_k \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The hyperparameters are themselves given weakly informative priors, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 2)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\]\nLike in a posterior predictive check, the hyperparameters \\(\\mu\\) and \\(\\sigma\\) are drawn from the posterior, \\[\n\\mu^{(m)}, \\sigma^{(m)} \\sim p(\\mu, \\sigma \\mid y)\n\\] Like in a prior predictive check, replicated values of \\(\\alpha\\) are drawn from the hyperparameters, \\[\n\\alpha^{\\textrm{rep}(m)}_k \\sim \\textrm{normal}(\\alpha_k \\mid\n\\mu^{(m)}, \\sigma^{(m)}).\n\\] The data items are then each replicated using the replicated intercepts, \\[\ny^{\\textrm{rep}(m)}_n \\sim\n\\textrm{bernoulli}\n  (\\textrm{logit}^{-1}(\\alpha^{\\textrm{rep}(m)}_{kk[n]})).\n\\] Thus the \\(y^{\\textrm{rep}(m)}\\) can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original \\(K\\) groups.\nIn Stan, mixed predictive replications \\(y^{\\textrm{rep}(m)}\\) can be programmed directly.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;offset=mu, multiplier=sigma&gt;[K] alpha;\n}\nmodel {\n  mu ~ normal(0, 2);               // hyperprior\n  sigma ~ lognormal(0, 1);\n  alpha ~ normal(mu, sigma);       // hierarchical prior\n  y ~ bernoulli_logit(alpha[kk]);  // data model\n}\ngenerated quantities {\n  // alpha replicated;  mu and sigma not replicated\n  array[K] real alpha_rep\n    = normal_rng(rep_vector(mu, K), sigma);\n  array[N] int&lt;lower=0, upper=1&gt; y_rep\n    = bernoulli_logit_rng(alpha_rep[kk]);\n}\n\n\n\nFollowing Andrew Gelman, Meng, and Stern (1996), prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.\n\n\nFor example, posterior predictive replication may be formulated using distribution notation as follows. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] The heavily overloaded distribution notation is meant to indicate that both \\(y\\) and \\(y^{\\textrm{rep}}\\) are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities \\(p_{Y^{\\textrm{rep}} \\mid\n\\Theta}\\) and \\(p_{Y \\mid \\Theta}\\) are the same.\nThe joint density is \\[\np(\\theta, y, y^{\\textrm{rep}})\n= p(\\theta) \\cdot p(y \\mid \\theta) \\cdot p(y^{\\textrm{rep}} \\mid \\theta).\n\\] This again is assuming that the two distributions for \\(y\\) and \\(y^{\\textrm{rep}}\\) are identical.\nThe variable \\(y\\) is observed, with the predictive simulation \\(y^{\\textrm{rep}}\\) and parameter vector \\(\\theta\\) not observed. The posterior is \\(p(y^{\\textrm{rep}}, \\theta \\mid y)\\). Given draws from the posterior, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are retained.\n\n\n\nThe prior predictive model simply drops the data component of the posterior predictive model. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] This corresponds to the joint density \\[\np(\\theta, y^{\\textrm{rep}}) = p(\\theta) \\cdot p(y^{\\textrm{rep}} \\mid\n\\theta).\n\\]\nIt is typically straightforward to draw \\(\\theta\\) from the prior and \\(y^{\\textrm{rep}}\\) from the data model given \\(\\theta\\) efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.\n\n\n\nThe mixed replication corresponds to the model \\[\\begin{eqnarray*}\n\\phi & \\sim & p(\\phi)\n\\\\[2pt]\n\\alpha & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\alpha)\n\\\\[2pt]\n\\alpha^{\\textrm{rep}} & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\phi)\n\\end{eqnarray*}\\] The notation here is meant to indicate that \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\) have identical distributions, as do \\(y\\) and \\(y^{\\textrm{rep}}\\).\nThis corresponds to a joint model \\[\np(\\phi, \\alpha, \\alpha^{\\textrm{rep}}, y, y^{\\textrm{rep}})\n=\np(\\phi)\n\\cdot p(\\alpha \\mid \\phi)\n\\cdot p(y \\mid \\alpha)\n\\cdot p(\\alpha^{\\textrm{rep}} \\mid \\phi)\n\\cdot p(y^{\\textrm{rep}} \\mid \\alpha^{\\textrm{rep}}),\n\\] where \\(y\\) is the only observed variable, \\(\\alpha\\) contains the lower-level parameters and \\(\\phi\\) the hyperparameters. Note that \\(\\phi\\) is not replicated and instead appears in the distribution for both \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\).\nThe posterior is \\(p(\\phi, \\alpha, \\alpha^{\\textrm{rep}},\ny^{\\textrm{rep}} \\mid y)\\). From posterior draws, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are kept.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#simulating-from-the-posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-predictive-checks.html#simulating-from-the-posterior-predictive-distribution",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "The posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data \\(y\\).\nThe posterior predictive distribution for replications \\(y^{\\textrm{rep}}\\) of the original data set \\(y\\) given model parameters \\(\\theta\\) is defined by \\[\np(y^{\\textrm{rep}} \\mid y)\n= \\int p(y^{\\textrm{rep}} \\mid \\theta)\n       \\cdot p(\\theta \\mid y) \\, \\textrm{d}\\theta.\n\\]\nAs with other posterior predictive quantities, generating a replicated data set \\(y^{\\textrm{rep}}\\) from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters \\(\\theta = (\\alpha, \\beta, \\sigma).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  y ~ normal(alpha + beta * x, sigma);\n}\nTo generate a replicated data set y_rep for this simple model, the following generated quantities block suffices.\ngenerated quantities {\n  array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n}\nThe vectorized form of the normal random number generator is used with the original predictors x and the model parameters alpha, beta, and sigma. The replicated data variable y_rep is declared to be the same size as the original data y, but instead of a vector type, it is declared to be an array of reals to match the return type of the function normal_rng. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.\nThe posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in the chapter on posterior predictions in that the original predictors \\(x\\) are used. That is, the posterior predictions are for the original data.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#plotting-multiples",
+    "href": "stan-users-guide/posterior-predictive-checks.html#plotting-multiples",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "A standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.\nHere’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\ntransformed data {\n  real&lt;lower=0&gt; mean_y = mean(to_vector(y));\n  real&lt;lower=0&gt; sd_y = sd(to_vector(y));\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  y ~ poisson(lambda);\n  lambda ~ exponential(0.2);\n}\ngenerated quantities {\n  array[N] int&lt;lower=0&gt; y_rep = poisson_rng(rep_array(lambda, N));\n  real&lt;lower=0&gt; mean_y_rep = mean(to_vector(y_rep));\n  real&lt;lower=0&gt; sd_y_rep = sd(to_vector(y_rep));\n  int&lt;lower=0, upper=1&gt; mean_gte = (mean_y_rep &gt;= mean_y);\n  int&lt;lower=0, upper=1&gt; sd_gte = (sd_y_rep &gt;= sd_y);\n}\nThe generated quantities block creates a variable y_rep for the replicated data, variables mean_y_rep and sd_y_rep for the statistics of the replicated data, and indicator variables mean_gte and sd_gte for whether the replicated statistic is greater than or equal to the statistic applied to the original data.\nNow consider generating data \\(y \\sim \\textrm{Poisson}(5)\\). The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.\n\n\n\nPosterior predictive checks for Poisson data generating process and Poisson model.\n\n\nWith a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.\nNow consider generating over-dispersed data \\(y \\sim \\textrm{negative-binomial2}(5, 1).\\) This has the same mean as \\(\\textrm{Poisson}(5)\\), namely \\(5\\), but a standard deviation of \\(\\sqrt{5 + 5^2 /1} \\approx 5.5.\\) There is no way to fit this data with the Poisson model, because a variable distributed as \\(\\textrm{Poisson}(\\lambda)\\) has mean \\(\\lambda\\) and standard deviation \\(\\sqrt{\\lambda},\\) which is \\(\\sqrt{5}\\) for \\(\\textrm{Poisson}(5).\\) Here’s the resulting small multiples plot, again with original data in the upper left.\n\n\n\nPosterior predictive checks for negative binomial data generating process and Poisson model.\n\n\nThis time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#posterior-p-values",
+    "href": "stan-users-guide/posterior-predictive-checks.html#posterior-p-values",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "If a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic \\(s(\\cdot)\\) in a replicated data set exceeds that in the original data, \\[\n\\Pr\\!\\left[ s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right]\n=\n\\int\n\\textrm{I}\\left( s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right)\n\\cdot p\\left( y^{\\textrm{rep}} \\mid y \\right)\n\\, \\textrm{d}{y^{\\textrm{rep}}}.\n\\] It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified (Bayarri and Berger 2000).\nNevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.\nTo calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mean_gt;\n  int&lt;lower=0, upper=1&gt; sd_gt;\n  {\n    array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n    mean_gt = mean(y_rep) &gt; mean(y);\n    sd_gt = sd(y_rep) &gt; sd(y);\n  }\n}\nThe indicator variable mean_gt will have value 1 if the mean of the simulated data y_rep is greater than or equal to the mean of he original data y. Because the values of y_rep are not needed for the posterior predictive checks, the program saves output space by using a local variable for y_rep. The statistics mean(u) and sd(y) could also be computed in the transformed data block and saved.\nFor the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.\n\n\n\nHistogram of means of replicated data sets; vertical red line at mean of original data.\n\n\nThe p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior \\(p\\)-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable mean_gt.\nThe standard deviation statistic tells a different story.\n\n\n\nScatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.\n\n\nHere, the original data has much higher standard deviation than any of the replicated data sets. The resulting \\(p\\)-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small \\(p\\)-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.\n\n\nAny statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.\nDespite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#prior-predictive-checks",
+    "href": "stan-users-guide/posterior-predictive-checks.html#prior-predictive-checks",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "Prior predictive checks generate data according to the prior in order to asses whether a prior is appropriate (Gabry et al. 2019). A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, \\[\ny^{\\textrm{sim}} \\sim p(y).\n\\] The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.\nThis is easy to carry out mechanically by simulating parameters \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta)\n\\] according to the priors, then simulating data \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}})\n\\] according to the data model given the simulated parameters. The result is a simulation from the joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta)\n\\] and thus \\[\ny^{\\textrm{sim}} \\sim p(y)\n\\] is a simulation from the prior predictive distribution.\n\n\nA prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.\nPrior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n}\ngenerated quantities {\n  real alpha = normal_rng(0, 1);\n  real beta = normal_rng(0, 1);\n  array[N] real y_sim = poisson_log_rng(alpha + beta * x);\n}\nRunning this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#example-of-prior-predictive-checks",
+    "href": "stan-users-guide/posterior-predictive-checks.html#example-of-prior-predictive-checks",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "Suppose we have a model for a football (aka soccer) league where there are \\(J\\) teams. Each team has a scoring rate \\(\\lambda_j\\) and in each game will be assumed to score \\(\\textrm{poisson}(\\lambda_j)\\) points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed \\[\n\\lambda_j \\sim \\textrm{gamma}(\\epsilon_1, \\epsilon_2).\n\\] This is not just a manufactured example; The BUGS Book recommends setting \\(\\epsilon = (0.5, 0.00001)\\), which corresponds to a Jeffreys prior for a Poisson rate parameter prior (Lunn et al. 2012, 85).\nSuppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.\ndata {\n  int&lt;lower=0&gt; J;\n  array[2] real&lt;lower=0&gt; epsilon;\n}\ngenerated quantities {\n  array[J] real&lt;lower=0&gt; lambda;\n  array[J, J] int y;\n  for (j in 1:J) lambda[j] = gamma_rng(epsilon[1], epsilon[2]);\n  for (i in 1:J) {\n    for (j in 1:J) {\n      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);\n    }\n  }\n}\nIn this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are \\(\\binom{J}{2} = \\frac{J \\cdot (J - 1)}{2}.\\)\nUsing the \\(\\textrm{gamma}(0.5, 0.00001)\\) reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, \\(y^{(1:20)}_{1, 2}\\).\n2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987\n7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66\nThat’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this \\(\\textrm{gamma}(0.5, 0.00001)\\) prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.\nThe posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values (A. Gelman 2006).\nJust as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.\nIn this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see (Gabry et al. 2019).\nPrior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#mixed-replication",
+    "href": "stan-users-guide/posterior-predictive-checks.html#mixed-replication",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "Andrew Gelman, Meng, and Stern (1996) discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.\nFor example, consider a simple varying intercept logistic regression, with intercepts \\(\\alpha_k\\) for \\(k \\in 1:K\\). Each data item \\(y_n \\in \\{ 0, 1 \\}\\) is assumed to correspond to group \\(kk_n \\in 1:K.\\) The data model is thus \\[\ny_n \\sim \\textrm{bernoulli}(\\textrm{logit}^{-1}(\\alpha_{kk[n]})).\n\\] The varying intercepts have a hierarchical normal prior, \\[\n\\alpha_k \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The hyperparameters are themselves given weakly informative priors, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 2)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\]\nLike in a posterior predictive check, the hyperparameters \\(\\mu\\) and \\(\\sigma\\) are drawn from the posterior, \\[\n\\mu^{(m)}, \\sigma^{(m)} \\sim p(\\mu, \\sigma \\mid y)\n\\] Like in a prior predictive check, replicated values of \\(\\alpha\\) are drawn from the hyperparameters, \\[\n\\alpha^{\\textrm{rep}(m)}_k \\sim \\textrm{normal}(\\alpha_k \\mid\n\\mu^{(m)}, \\sigma^{(m)}).\n\\] The data items are then each replicated using the replicated intercepts, \\[\ny^{\\textrm{rep}(m)}_n \\sim\n\\textrm{bernoulli}\n  (\\textrm{logit}^{-1}(\\alpha^{\\textrm{rep}(m)}_{kk[n]})).\n\\] Thus the \\(y^{\\textrm{rep}(m)}\\) can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original \\(K\\) groups.\nIn Stan, mixed predictive replications \\(y^{\\textrm{rep}(m)}\\) can be programmed directly.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;offset=mu, multiplier=sigma&gt;[K] alpha;\n}\nmodel {\n  mu ~ normal(0, 2);               // hyperprior\n  sigma ~ lognormal(0, 1);\n  alpha ~ normal(mu, sigma);       // hierarchical prior\n  y ~ bernoulli_logit(alpha[kk]);  // data model\n}\ngenerated quantities {\n  // alpha replicated;  mu and sigma not replicated\n  array[K] real alpha_rep\n    = normal_rng(rep_vector(mu, K), sigma);\n  array[N] int&lt;lower=0, upper=1&gt; y_rep\n    = bernoulli_logit_rng(alpha_rep[kk]);\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#joint-model-representation",
+    "href": "stan-users-guide/posterior-predictive-checks.html#joint-model-representation",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "Following Andrew Gelman, Meng, and Stern (1996), prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.\n\n\nFor example, posterior predictive replication may be formulated using distribution notation as follows. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] The heavily overloaded distribution notation is meant to indicate that both \\(y\\) and \\(y^{\\textrm{rep}}\\) are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities \\(p_{Y^{\\textrm{rep}} \\mid\n\\Theta}\\) and \\(p_{Y \\mid \\Theta}\\) are the same.\nThe joint density is \\[\np(\\theta, y, y^{\\textrm{rep}})\n= p(\\theta) \\cdot p(y \\mid \\theta) \\cdot p(y^{\\textrm{rep}} \\mid \\theta).\n\\] This again is assuming that the two distributions for \\(y\\) and \\(y^{\\textrm{rep}}\\) are identical.\nThe variable \\(y\\) is observed, with the predictive simulation \\(y^{\\textrm{rep}}\\) and parameter vector \\(\\theta\\) not observed. The posterior is \\(p(y^{\\textrm{rep}}, \\theta \\mid y)\\). Given draws from the posterior, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are retained.\n\n\n\nThe prior predictive model simply drops the data component of the posterior predictive model. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] This corresponds to the joint density \\[\np(\\theta, y^{\\textrm{rep}}) = p(\\theta) \\cdot p(y^{\\textrm{rep}} \\mid\n\\theta).\n\\]\nIt is typically straightforward to draw \\(\\theta\\) from the prior and \\(y^{\\textrm{rep}}\\) from the data model given \\(\\theta\\) efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.\n\n\n\nThe mixed replication corresponds to the model \\[\\begin{eqnarray*}\n\\phi & \\sim & p(\\phi)\n\\\\[2pt]\n\\alpha & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\alpha)\n\\\\[2pt]\n\\alpha^{\\textrm{rep}} & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\phi)\n\\end{eqnarray*}\\] The notation here is meant to indicate that \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\) have identical distributions, as do \\(y\\) and \\(y^{\\textrm{rep}}\\).\nThis corresponds to a joint model \\[\np(\\phi, \\alpha, \\alpha^{\\textrm{rep}}, y, y^{\\textrm{rep}})\n=\np(\\phi)\n\\cdot p(\\alpha \\mid \\phi)\n\\cdot p(y \\mid \\alpha)\n\\cdot p(\\alpha^{\\textrm{rep}} \\mid \\phi)\n\\cdot p(y^{\\textrm{rep}} \\mid \\alpha^{\\textrm{rep}}),\n\\] where \\(y\\) is the only observed variable, \\(\\alpha\\) contains the lower-level parameters and \\(\\phi\\) the hyperparameters. Note that \\(\\phi\\) is not replicated and instead appears in the distribution for both \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\).\nThe posterior is \\(p(\\phi, \\alpha, \\alpha^{\\textrm{rep}},\ny^{\\textrm{rep}} \\mid y)\\). From posterior draws, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are kept.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html",
+    "href": "stan-users-guide/problematic-posteriors.html",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "Mathematically speaking, with a proper posterior, one can do Bayesian inference and that’s that. There is not even a need to require a finite variance or even a finite mean—all that’s needed is a finite integral. Nevertheless, modeling is a tricky business and even experienced modelers sometimes code models that lead to improper priors. Furthermore, some posteriors are mathematically sound, but ill-behaved in practice. This chapter discusses issues in models that create problematic posterior inferences, either in general for Bayesian inference or in practice for Stan.\n\n\nThis section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.\n\n\n\n\nThe first example of collinearity is an artificial example involving redundant intercept parameters.1\nSuppose there are observations \\(y_n\\) for \\(n \\in \\{1,\\dotsc,N\\}\\), two intercept parameters \\(\\lambda_1\\) and \\(\\lambda_2\\), a scale parameter \\(\\sigma &gt; 0\\), and the data model \\[\ny_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2, \\sigma).\n\\]\nFor any constant \\(q\\), the sampling density for \\(y\\) does not change if we add \\(q\\) to \\(\\lambda_1\\) and subtract it from \\(\\lambda_2\\), i.e., \\[\np(y \\mid \\lambda_1, \\lambda_2,\\sigma)\n=\np(y \\mid \\lambda_1 + q, \\lambda_2 - q, \\sigma).\n\\]\nThe consequence is that an improper uniform prior \\(p(\\mu,\\sigma)\n\\propto 1\\) leads to an improper posterior. This impropriety arises because the neighborhoods around \\(\\lambda_1 + q, \\lambda_2 - q\\) have the same mass no matter what \\(q\\) is. Therefore, a sampler would need to spend as much time in the neighborhood of \\(\\lambda_1=1\\,000\\,000\\,000\\) and \\(\\lambda_2=-1\\,000\\,000\\,000\\) as it does in the neighborhood of \\(\\lambda_1=0\\) and \\(\\lambda_2=0\\), and so on for ever more far-ranging values.\nThe marginal posterior \\(p(\\lambda_1,\\lambda_2 \\mid y)\\) for this model is thus improper.2\nThe impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where \\(\\lambda_2 = \\lambda_1 + c\\) for some constant \\(c\\).\nContrast this model with a simple regression with a single intercept parameter \\(\\mu\\) and data model \\[\ny_n \\sim \\textsf{normal}(\\mu,\\sigma).\n\\] Even with an improper prior, the posterior is proper as long as there are at least two data points \\(y_n\\) with distinct values.\n\n\n\nConsider an item-response theory model for students \\(j \\in 1{:}J\\) with abilities \\(\\alpha_j\\) and test items \\(i \\in 1{:}I\\) with difficulties \\(\\beta_i\\). The observed data are an \\(I \\times J\\) array with entries \\(y_{i, j} \\in \\{ 0, 1 \\}\\) coded such that \\(y_{i, j} = 1\\) indicates that student \\(j\\) answered question \\(i\\) correctly. The sampling distribution for the data is \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\alpha_j - \\beta_i)).\n\\]\nFor any constant \\(c\\), the probability of \\(y\\) is unchanged by adding a constant \\(c\\) to all the abilities and subtracting it from all the difficulties, i.e., \\[\np(y \\mid \\alpha, \\beta)\n=\np(y \\mid \\alpha + c, \\beta - c).\n\\]\nThis leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.\n\n\n\nThe general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model \\[\ny_n \\sim \\textsf{normal}(x_n \\beta, \\sigma)\n\\] for an \\(N\\)-dimensional observation vector \\(y\\), an \\(N \\times K\\) predictor matrix \\(x\\), and a \\(K\\)-dimensional coefficient vector \\(\\beta\\).\nNow suppose that column \\(k\\) of the predictor matrix is a multiple of column \\(k'\\), i.e., there is some constant \\(c\\) such that \\(x_{n,k} = c\n\\, x_{n,k'}\\) for all \\(n\\). In this case, the coefficients \\(\\beta_k\\) and \\(\\beta_{k'}\\) can covary without changing the predictions, so that for any \\(d \\neq 0\\), \\[\np(y \\mid \\ldots, \\beta_k, \\dotsc, \\beta_{k'}, \\dotsc, \\sigma)\n=\np(y \\mid \\ldots, d \\beta_k, \\dotsc, \\frac{d}{c} \\, \\beta_{k'}, \\dotsc,\n\\sigma).\n\\]\nEven if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.\n\n\n\nConsider adding a discrimination parameter \\(\\delta_i\\) for each question in an IRT model, with data model \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\delta_i(\\alpha_j - \\beta_i))).\n\\] For any constant \\(c \\neq 0\\), multiplying \\(\\delta\\) by \\(c\\) and dividing \\(\\alpha\\) and \\(\\beta\\) by \\(c\\) produces the same likelihood, \\[\np(y \\mid \\delta,\\alpha,\\beta)\n= p(y \\mid c \\delta, \\frac{1}{c}\\alpha, \\frac{1}{c}\\beta).\n\\] If \\(c &lt; 0\\), this switches the signs of every component in \\(\\alpha\\), \\(\\beta\\), and \\(\\delta\\) without changing the density.\n\n\n\nIn order to parameterize a \\(K\\)-simplex (i.e., a \\(K\\)-vector with non-negative values that sum to one), only \\(K - 1\\) parameters are necessary because the \\(K\\)th is just one minus the sum of the first \\(K\n- 1\\) parameters, so that if \\(\\theta\\) is a \\(K\\)-simplex, \\[\n\\theta_K = 1 - \\sum_{k=1}^{K-1} \\theta_k.\n\\]\nThe softmax function maps a \\(K\\)-vector \\(\\alpha\\) of linear predictors to a \\(K\\)-simplex \\(\\theta = \\texttt{softmax}(\\alpha)\\) by defining \\[\n\\theta_k = \\frac{\\exp(\\alpha_k)}{\\sum_{k'=1}^K \\exp(\\alpha_{k'})}.\n\\]\nThe softmax function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters \\(\\alpha\\). In particular, adding or subtracting a constant from each \\(\\alpha_k\\) produces the same simplex \\(\\theta\\).\n\n\n\n\nAll of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.\n\n\nIn the case of the multiple intercepts, \\(\\lambda_1\\) and \\(\\lambda_2\\), the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter \\(\\mu\\) and sampling distribution \\(y_n \\sim \\textsf{normal}(\\mu, \\sigma)\\). The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix \\(x\\).\n\n\n\nThe IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability \\(\\alpha_1\\) can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.\nThis solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters \\(\\delta_i\\). To solve this problem, one of the discrimination parameters, say \\(\\delta_1\\), must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.\nThe many-to-one nature of \\(\\texttt{softmax}(\\alpha)\\) is typically mitigated by pinning a component of \\(\\alpha\\), for instance fixing \\(\\alpha_K = 0\\). The resulting mapping is one-to-one from \\(K-1\\) unconstrained parameters to a \\(K\\)-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a \\(K-1\\)-vector can be written as\nvector softmax_id(vector alpha) {\n  vector[num_elements(alpha) + 1] alphac1;\n  for (k in 1:num_elements(alpha)) {\n    alphac1[k] = alpha[k];\n  }\n  alphac1[num_elements(alphac1)] = 0;\n  return softmax(alphac1);\n}\n\n\n\nSo far, the models have been discussed as if the priors on the parameters were improper uniform priors.\nA more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.\nFor example, normal priors on the multiple intercepts, \\[\n\\lambda_1, \\lambda_2 \\sim \\textsf{normal}(0,\\tau),\n\\] with a constant scale \\(\\tau\\), ensure that the posterior mode is located at a point where \\(\\lambda_1 = \\lambda_2\\), because this minimizes \\(\\log \\textsf{normal}(\\lambda_1 \\mid 0,\\tau) + \\log\n\\textsf{normal}(\\lambda_2 \\mid 0,\\tau)\\).3\nThe following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.\nThe two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.\n\n\n\nTwo intercepts with improper prior\n\n\nAdding a standard normal prior for the intercepts results in a proper posterior.\n\n\n\nTwo intercepts with proper prior\n\n\nThe single intercept parameterization with no prior also has a proper posterior.\n\n\n\nSingle intercepts with improper prior\n\n\nThe addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.\nAn alternative strategy for identifying a \\(K\\)-simplex parameterization \\(\\theta = \\texttt{softmax}(\\alpha)\\) in terms of an unconstrained \\(K\\)-vector \\(\\alpha\\) is to place a prior on the components of \\(\\alpha\\) with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning \\(\\alpha_K =\n0\\), the prior-based approach models the \\(K\\) outcomes symmetrically rather than modeling \\(K-1\\) outcomes relative to the \\(K\\)-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.\n\n\n\nCare must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.\nIdeally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.\nWhen there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.\nPriors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters \\(\\alpha\\) to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.\n\n\n\n\n\nWhere collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.\n\n\nConsider a normal mixture model with two location parameters \\(\\mu_1\\) and \\(\\mu_2\\), a shared scale \\(\\sigma &gt; 0\\), a mixture ratio \\(\\theta \\in\n[0,1]\\), and data model \\[\np(y \\mid \\theta,\\mu_1,\\mu_2,\\sigma)\n= \\prod_{n=1}^N \\big( \\theta \\, \\textsf{normal}(y_n \\mid \\mu_1,\\sigma)\n                       + (1 - \\theta) \\, \\textsf{normal}(y_n \\mid \\mu_2,\\sigma) \\big).\n\\] The issue here is exchangeability of the mixture components, because \\[\np(\\theta,\\mu_1,\\mu_2,\\sigma \\mid y) = p\\big((1-\\theta),\\mu_2,\\mu_1,\\sigma \\mid y\\big).\n\\] The problem is exacerbated as the number of mixture components \\(K\\) grows, as in clustering models, leading to \\(K!\\) identical posterior maxima.\n\n\n\nThe analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the \\(\\hat{R}\\) convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for \\(\\mu_1\\) that is equal to that of \\(\\mu_2\\), and a posterior mean for \\(\\theta\\) that is always 1/2, no matter what the data are.\n\n\n\nIn some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for \\(\\theta\\) in the two component mixture model will always be 1/2.\n\n\n\nTheoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.\nBeing able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for \\(\\mu_1\\) to move to a new mode when sampled conditioned on the current values of \\(\\mu_2\\) and \\(\\theta\\). For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.\nEven with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.\n\n\n\nSeveral hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.\n\n\nOne common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining \\(\\mu_1 &lt; \\mu_2\\) in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering \\(\\mu_1 &gt; \\mu_2\\). In these cases, the posteriors are affected by the constraint and true posterior uncertainty in \\(\\mu_1\\) and \\(\\mu_2\\) is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use \\(M\\) posterior draws to estimate \\(\\Pr[\\mu_1 &gt; \\mu_2]\\), will fail, because the estimator \\[\n\\Pr[\\mu_1 &gt; \\mu_2]\n\\approx\n\\sum_{m=1}^M \\textrm{I}\\left(\\mu_1^{(m)} &gt; \\mu_2^{(m)}\\right)\n\\] will result in an estimate of 0 because the posterior respects the constraint in the model.\n\n\n\nAnother common approach is to run a single chain or to initialize the parameters near realistic values.4\nThis can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.\n\n\n\n\n\nIt is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of \\(K\\) normals might devolve to have \\(\\mu_i =\n\\mu_j\\) and \\(\\sigma_i = \\sigma_j\\) for \\(i \\neq j\\).\nThis will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw \\((m)\\), it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.\nIt may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.\nIn general, it is difficult to recover exactly the right \\(K\\) mixture components in a mixture model as \\(K\\) increases beyond one (yes, even a two-component mixture can have this problem).\n\n\n\nIn some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.\n\n\nOne such example is a binary mixture model with scales varying by component, \\(\\sigma_1\\) and \\(\\sigma_2\\) for locations \\(\\mu_1\\) and \\(\\mu_2\\). In this situation, the density grows without bound as \\(\\sigma_1 \\rightarrow 0\\) and \\(\\mu_1 \\rightarrow y_n\\) for some \\(n\\); that is, one of the mixture components concentrates all of its mass around a single data item \\(y_n\\).\n\n\n\nAnother example of unbounded densities arises with a posterior such as \\(\\textsf{beta}(\\phi \\mid 0.5,0.5)\\), which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as \\(\\phi \\rightarrow 0\\) and \\(\\phi \\rightarrow 1\\). Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior \\[\\begin{align*}\np(\\phi \\mid y)\n&\\propto\n   \\textsf{beta}(\\phi \\mid 0.5,0.5) \\times \\prod_{n=1}^N \\textsf{Bernoulli}(y_n \\mid \\phi) \\\\\n&=\n   \\textsf{beta}\\left(\\phi \\,\\middle|\\, 0.5 + \\sum_{n=1}^N y_n, 0.5 + N - \\sum_{n=1}^N y_n\\right).\n\\end{align*}\\]\nIf \\(N = 9\\) and each \\(y_n = 1\\), the posterior is \\(\\textsf{beta}(\\phi \\mid 9.5,0.5)\\). This posterior is unbounded as \\(\\phi\n\\rightarrow 1\\). Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.\n\n\nStan does not sample directly on the constrained \\((0,1)\\) space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values \\(\\phi\\) are logit-transformed to \\((-\\infty,\\infty)\\). The boundaries at 0 and 1 are pushed out to \\(-\\infty\\) and \\(\\infty\\) respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of \\((0,1)\\) is \\(\\log\n\\operatorname{logit}^{-1}(\\phi) + \\log \\operatorname{logit}(1 - \\phi)\\).\nThere are two problems that still arise, though. The first is that if the posterior mass for \\(\\phi\\) is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.\n\n\n\n\n\nIn some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.\n\n\nConsider a logistic regression model with \\(N\\) observed outcomes \\(y_n\n\\in \\{ 0, 1 \\}\\), an \\(N \\times K\\) matrix \\(x\\) of predictors, a \\(K\\)-dimensional coefficient vector \\(\\beta\\), and data model \\[\ny_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(x_n \\beta)).\n\\] Now suppose that column \\(k\\) of the predictor matrix is such that \\(x_{n,k} &gt; 0\\) if and only if \\(y_n = 1\\), a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as \\(\\beta_k \\rightarrow \\infty\\), because for cases with \\(y_n = 1\\), \\(x_n \\beta \\rightarrow \\infty\\) and hence \\(\\operatorname{logit}^{-1}(x_n \\beta) \\rightarrow 1\\).\nWith separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for \\(\\beta_k\\) is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for \\(\\beta\\), which ensures a proper posterior.\n\n\n\n\nSuppose your model includes a parameter \\(\\psi\\) that is defined on \\([0,1]\\) and is given a flat prior \\(\\textsf{uniform}(\\psi \\mid 0,1)\\). Now if the data don’t tell us anything about \\(\\psi\\), the posterior is also \\(\\textsf{uniform}(\\psi \\mid 0,1)\\).\nAlthough there is no maximum likelihood estimate for \\(\\psi\\), the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on \\([0,1]\\), the posterior mean for \\(\\psi\\) is well-defined with value \\(1/2\\). Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for \\(\\psi\\) at all points in \\([0,1]\\).\n\n\n\nWith an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the collinearity section and illustrated in the non-identifiable density plot.\n\n\nGibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.\nConsider what happens with initial values \\(\\lambda_1^{(0)}, \\lambda_2^{(0)}\\). Gibbs sampling proceeds in iteration \\(m\\) by drawing \\[\\begin{align*}\n\\lambda_1^{(m)} &\\sim p(\\lambda_1 \\mid \\lambda_2^{(m-1)}, \\sigma^{(m-1)},  y) \\\\\n\\lambda_2^{(m)} &\\sim p(\\lambda_2 \\mid \\lambda_1^{(m)},   \\sigma^{(m-1)},  y) \\\\\n\\sigma^{(m)}    &\\sim p(\\sigma    \\mid \\lambda_1^{(m)},   \\lambda_2^{(m)}, y).\n\\end{align*}\\]\nNow consider the draw for \\(\\lambda_1\\) (the draw for \\(\\lambda_2\\) is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next \\(\\lambda_1\\) can be drawn is highly constrained by the current values of \\(\\lambda_2\\) and \\(\\sigma\\). Gibbs will run quickly and provide seemingly reasonable inferences for \\(\\lambda_1 + \\lambda_2\\). But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.\n\n\n\nHamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for \\(\\lambda_1\\) and \\(\\lambda_2\\), the gradient of the log posterior is going to adjust for the correlation and the simulation will run \\(\\lambda_1\\) and \\(\\lambda_2\\) in opposite directions along the valley corresponding to the ridge in the posterior log density.\n\n\n\nStan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior (see Hoffman and Gelman 2014). NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.\nHere are results of Stan runs with default parameters fit to \\(N=100\\) data points generated from \\(y_n \\sim \\textsf{normal}(0,1)\\):\nTwo Scale Parameters, Improper Prior\nInference for Stan model: improper_stan\nWarmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total\nSampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total\n\n                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat\nlp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0\nn_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0\nlambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2\nlambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2\nsigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1\nmu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0\nTwo Scale Parameters, Weak Prior\nWarmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total\nSampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total\n\n                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat\nlp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0\nn_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0\nlambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nlambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nsigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0\nmu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0\nOne Scale Parameter, Improper Prior\nWarmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total\nSampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total\n\n                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat\nlp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0\nn_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0\nmu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0\nsigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0\nOn the top is the non-identified model with improper uniform priors and data model \\(y_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2,\n\\sigma)\\).\nIn the middle is the same data model as in top plus priors \\(\\lambda_k \\sim \\textsf{normal}(0,10)\\).\nOn the bottom is an identified model with an improper prior, with data model \\(y_n \\sim \\textsf{normal}(\\mu,\\sigma)\\). All models estimate \\(\\mu\\) at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value \\(\\mu=0\\) is within the 90% posterior intervals in all three models.\n\n\n\nTo illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the collinearity section.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real lambda1;\n  real lambda2;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  real mu;\n  mu = lambda1 + lambda2;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nThe second adds priors to the model block for lambda1 and lambda2 to the previous model.\nlambda1 ~ normal(0, 10);\nlambda2 ~ normal(0, 10);\nThe third involves a single location parameter, but no priors.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nAll three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.\n\nAs indicated by R_hat column, all parameters have converged other than \\(\\lambda_1\\) and \\(\\lambda_2\\) in the non-identified model.\nThe average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.\nThe effective sample size per second for \\(\\mu\\) is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for \\(\\sigma\\).\nIn the non-identified model, the 95% interval for \\(\\lambda_1\\) is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.\nIn all three models, the simulated value of \\(\\mu=0\\) and \\(\\sigma=1\\) are well within the posterior 90% intervals.\n\nThe first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#collinearity.section",
+    "href": "stan-users-guide/problematic-posteriors.html#collinearity.section",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "This section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.\n\n\n\n\nThe first example of collinearity is an artificial example involving redundant intercept parameters.1\nSuppose there are observations \\(y_n\\) for \\(n \\in \\{1,\\dotsc,N\\}\\), two intercept parameters \\(\\lambda_1\\) and \\(\\lambda_2\\), a scale parameter \\(\\sigma &gt; 0\\), and the data model \\[\ny_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2, \\sigma).\n\\]\nFor any constant \\(q\\), the sampling density for \\(y\\) does not change if we add \\(q\\) to \\(\\lambda_1\\) and subtract it from \\(\\lambda_2\\), i.e., \\[\np(y \\mid \\lambda_1, \\lambda_2,\\sigma)\n=\np(y \\mid \\lambda_1 + q, \\lambda_2 - q, \\sigma).\n\\]\nThe consequence is that an improper uniform prior \\(p(\\mu,\\sigma)\n\\propto 1\\) leads to an improper posterior. This impropriety arises because the neighborhoods around \\(\\lambda_1 + q, \\lambda_2 - q\\) have the same mass no matter what \\(q\\) is. Therefore, a sampler would need to spend as much time in the neighborhood of \\(\\lambda_1=1\\,000\\,000\\,000\\) and \\(\\lambda_2=-1\\,000\\,000\\,000\\) as it does in the neighborhood of \\(\\lambda_1=0\\) and \\(\\lambda_2=0\\), and so on for ever more far-ranging values.\nThe marginal posterior \\(p(\\lambda_1,\\lambda_2 \\mid y)\\) for this model is thus improper.2\nThe impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where \\(\\lambda_2 = \\lambda_1 + c\\) for some constant \\(c\\).\nContrast this model with a simple regression with a single intercept parameter \\(\\mu\\) and data model \\[\ny_n \\sim \\textsf{normal}(\\mu,\\sigma).\n\\] Even with an improper prior, the posterior is proper as long as there are at least two data points \\(y_n\\) with distinct values.\n\n\n\nConsider an item-response theory model for students \\(j \\in 1{:}J\\) with abilities \\(\\alpha_j\\) and test items \\(i \\in 1{:}I\\) with difficulties \\(\\beta_i\\). The observed data are an \\(I \\times J\\) array with entries \\(y_{i, j} \\in \\{ 0, 1 \\}\\) coded such that \\(y_{i, j} = 1\\) indicates that student \\(j\\) answered question \\(i\\) correctly. The sampling distribution for the data is \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\alpha_j - \\beta_i)).\n\\]\nFor any constant \\(c\\), the probability of \\(y\\) is unchanged by adding a constant \\(c\\) to all the abilities and subtracting it from all the difficulties, i.e., \\[\np(y \\mid \\alpha, \\beta)\n=\np(y \\mid \\alpha + c, \\beta - c).\n\\]\nThis leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.\n\n\n\nThe general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model \\[\ny_n \\sim \\textsf{normal}(x_n \\beta, \\sigma)\n\\] for an \\(N\\)-dimensional observation vector \\(y\\), an \\(N \\times K\\) predictor matrix \\(x\\), and a \\(K\\)-dimensional coefficient vector \\(\\beta\\).\nNow suppose that column \\(k\\) of the predictor matrix is a multiple of column \\(k'\\), i.e., there is some constant \\(c\\) such that \\(x_{n,k} = c\n\\, x_{n,k'}\\) for all \\(n\\). In this case, the coefficients \\(\\beta_k\\) and \\(\\beta_{k'}\\) can covary without changing the predictions, so that for any \\(d \\neq 0\\), \\[\np(y \\mid \\ldots, \\beta_k, \\dotsc, \\beta_{k'}, \\dotsc, \\sigma)\n=\np(y \\mid \\ldots, d \\beta_k, \\dotsc, \\frac{d}{c} \\, \\beta_{k'}, \\dotsc,\n\\sigma).\n\\]\nEven if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.\n\n\n\nConsider adding a discrimination parameter \\(\\delta_i\\) for each question in an IRT model, with data model \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\delta_i(\\alpha_j - \\beta_i))).\n\\] For any constant \\(c \\neq 0\\), multiplying \\(\\delta\\) by \\(c\\) and dividing \\(\\alpha\\) and \\(\\beta\\) by \\(c\\) produces the same likelihood, \\[\np(y \\mid \\delta,\\alpha,\\beta)\n= p(y \\mid c \\delta, \\frac{1}{c}\\alpha, \\frac{1}{c}\\beta).\n\\] If \\(c &lt; 0\\), this switches the signs of every component in \\(\\alpha\\), \\(\\beta\\), and \\(\\delta\\) without changing the density.\n\n\n\nIn order to parameterize a \\(K\\)-simplex (i.e., a \\(K\\)-vector with non-negative values that sum to one), only \\(K - 1\\) parameters are necessary because the \\(K\\)th is just one minus the sum of the first \\(K\n- 1\\) parameters, so that if \\(\\theta\\) is a \\(K\\)-simplex, \\[\n\\theta_K = 1 - \\sum_{k=1}^{K-1} \\theta_k.\n\\]\nThe softmax function maps a \\(K\\)-vector \\(\\alpha\\) of linear predictors to a \\(K\\)-simplex \\(\\theta = \\texttt{softmax}(\\alpha)\\) by defining \\[\n\\theta_k = \\frac{\\exp(\\alpha_k)}{\\sum_{k'=1}^K \\exp(\\alpha_{k'})}.\n\\]\nThe softmax function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters \\(\\alpha\\). In particular, adding or subtracting a constant from each \\(\\alpha_k\\) produces the same simplex \\(\\theta\\).\n\n\n\n\nAll of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.\n\n\nIn the case of the multiple intercepts, \\(\\lambda_1\\) and \\(\\lambda_2\\), the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter \\(\\mu\\) and sampling distribution \\(y_n \\sim \\textsf{normal}(\\mu, \\sigma)\\). The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix \\(x\\).\n\n\n\nThe IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability \\(\\alpha_1\\) can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.\nThis solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters \\(\\delta_i\\). To solve this problem, one of the discrimination parameters, say \\(\\delta_1\\), must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.\nThe many-to-one nature of \\(\\texttt{softmax}(\\alpha)\\) is typically mitigated by pinning a component of \\(\\alpha\\), for instance fixing \\(\\alpha_K = 0\\). The resulting mapping is one-to-one from \\(K-1\\) unconstrained parameters to a \\(K\\)-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a \\(K-1\\)-vector can be written as\nvector softmax_id(vector alpha) {\n  vector[num_elements(alpha) + 1] alphac1;\n  for (k in 1:num_elements(alpha)) {\n    alphac1[k] = alpha[k];\n  }\n  alphac1[num_elements(alphac1)] = 0;\n  return softmax(alphac1);\n}\n\n\n\nSo far, the models have been discussed as if the priors on the parameters were improper uniform priors.\nA more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.\nFor example, normal priors on the multiple intercepts, \\[\n\\lambda_1, \\lambda_2 \\sim \\textsf{normal}(0,\\tau),\n\\] with a constant scale \\(\\tau\\), ensure that the posterior mode is located at a point where \\(\\lambda_1 = \\lambda_2\\), because this minimizes \\(\\log \\textsf{normal}(\\lambda_1 \\mid 0,\\tau) + \\log\n\\textsf{normal}(\\lambda_2 \\mid 0,\\tau)\\).3\nThe following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.\nThe two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.\n\n\n\nTwo intercepts with improper prior\n\n\nAdding a standard normal prior for the intercepts results in a proper posterior.\n\n\n\nTwo intercepts with proper prior\n\n\nThe single intercept parameterization with no prior also has a proper posterior.\n\n\n\nSingle intercepts with improper prior\n\n\nThe addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.\nAn alternative strategy for identifying a \\(K\\)-simplex parameterization \\(\\theta = \\texttt{softmax}(\\alpha)\\) in terms of an unconstrained \\(K\\)-vector \\(\\alpha\\) is to place a prior on the components of \\(\\alpha\\) with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning \\(\\alpha_K =\n0\\), the prior-based approach models the \\(K\\) outcomes symmetrically rather than modeling \\(K-1\\) outcomes relative to the \\(K\\)-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.\n\n\n\nCare must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.\nIdeally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.\nWhen there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.\nPriors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters \\(\\alpha\\) to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#label-switching-problematic.section",
+    "href": "stan-users-guide/problematic-posteriors.html#label-switching-problematic.section",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "Where collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.\n\n\nConsider a normal mixture model with two location parameters \\(\\mu_1\\) and \\(\\mu_2\\), a shared scale \\(\\sigma &gt; 0\\), a mixture ratio \\(\\theta \\in\n[0,1]\\), and data model \\[\np(y \\mid \\theta,\\mu_1,\\mu_2,\\sigma)\n= \\prod_{n=1}^N \\big( \\theta \\, \\textsf{normal}(y_n \\mid \\mu_1,\\sigma)\n                       + (1 - \\theta) \\, \\textsf{normal}(y_n \\mid \\mu_2,\\sigma) \\big).\n\\] The issue here is exchangeability of the mixture components, because \\[\np(\\theta,\\mu_1,\\mu_2,\\sigma \\mid y) = p\\big((1-\\theta),\\mu_2,\\mu_1,\\sigma \\mid y\\big).\n\\] The problem is exacerbated as the number of mixture components \\(K\\) grows, as in clustering models, leading to \\(K!\\) identical posterior maxima.\n\n\n\nThe analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the \\(\\hat{R}\\) convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for \\(\\mu_1\\) that is equal to that of \\(\\mu_2\\), and a posterior mean for \\(\\theta\\) that is always 1/2, no matter what the data are.\n\n\n\nIn some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for \\(\\theta\\) in the two component mixture model will always be 1/2.\n\n\n\nTheoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.\nBeing able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for \\(\\mu_1\\) to move to a new mode when sampled conditioned on the current values of \\(\\mu_2\\) and \\(\\theta\\). For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.\nEven with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.\n\n\n\nSeveral hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.\n\n\nOne common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining \\(\\mu_1 &lt; \\mu_2\\) in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering \\(\\mu_1 &gt; \\mu_2\\). In these cases, the posteriors are affected by the constraint and true posterior uncertainty in \\(\\mu_1\\) and \\(\\mu_2\\) is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use \\(M\\) posterior draws to estimate \\(\\Pr[\\mu_1 &gt; \\mu_2]\\), will fail, because the estimator \\[\n\\Pr[\\mu_1 &gt; \\mu_2]\n\\approx\n\\sum_{m=1}^M \\textrm{I}\\left(\\mu_1^{(m)} &gt; \\mu_2^{(m)}\\right)\n\\] will result in an estimate of 0 because the posterior respects the constraint in the model.\n\n\n\nAnother common approach is to run a single chain or to initialize the parameters near realistic values.4\nThis can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#component-collapsing-in-mixture-models",
+    "href": "stan-users-guide/problematic-posteriors.html#component-collapsing-in-mixture-models",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "It is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of \\(K\\) normals might devolve to have \\(\\mu_i =\n\\mu_j\\) and \\(\\sigma_i = \\sigma_j\\) for \\(i \\neq j\\).\nThis will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw \\((m)\\), it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.\nIt may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.\nIn general, it is difficult to recover exactly the right \\(K\\) mixture components in a mixture model as \\(K\\) increases beyond one (yes, even a two-component mixture can have this problem).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-densities",
+    "href": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-densities",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "In some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.\n\n\nOne such example is a binary mixture model with scales varying by component, \\(\\sigma_1\\) and \\(\\sigma_2\\) for locations \\(\\mu_1\\) and \\(\\mu_2\\). In this situation, the density grows without bound as \\(\\sigma_1 \\rightarrow 0\\) and \\(\\mu_1 \\rightarrow y_n\\) for some \\(n\\); that is, one of the mixture components concentrates all of its mass around a single data item \\(y_n\\).\n\n\n\nAnother example of unbounded densities arises with a posterior such as \\(\\textsf{beta}(\\phi \\mid 0.5,0.5)\\), which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as \\(\\phi \\rightarrow 0\\) and \\(\\phi \\rightarrow 1\\). Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior \\[\\begin{align*}\np(\\phi \\mid y)\n&\\propto\n   \\textsf{beta}(\\phi \\mid 0.5,0.5) \\times \\prod_{n=1}^N \\textsf{Bernoulli}(y_n \\mid \\phi) \\\\\n&=\n   \\textsf{beta}\\left(\\phi \\,\\middle|\\, 0.5 + \\sum_{n=1}^N y_n, 0.5 + N - \\sum_{n=1}^N y_n\\right).\n\\end{align*}\\]\nIf \\(N = 9\\) and each \\(y_n = 1\\), the posterior is \\(\\textsf{beta}(\\phi \\mid 9.5,0.5)\\). This posterior is unbounded as \\(\\phi\n\\rightarrow 1\\). Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.\n\n\nStan does not sample directly on the constrained \\((0,1)\\) space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values \\(\\phi\\) are logit-transformed to \\((-\\infty,\\infty)\\). The boundaries at 0 and 1 are pushed out to \\(-\\infty\\) and \\(\\infty\\) respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of \\((0,1)\\) is \\(\\log\n\\operatorname{logit}^{-1}(\\phi) + \\log \\operatorname{logit}(1 - \\phi)\\).\nThere are two problems that still arise, though. The first is that if the posterior mass for \\(\\phi\\) is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-parameters",
+    "href": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-parameters",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "In some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.\n\n\nConsider a logistic regression model with \\(N\\) observed outcomes \\(y_n\n\\in \\{ 0, 1 \\}\\), an \\(N \\times K\\) matrix \\(x\\) of predictors, a \\(K\\)-dimensional coefficient vector \\(\\beta\\), and data model \\[\ny_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(x_n \\beta)).\n\\] Now suppose that column \\(k\\) of the predictor matrix is such that \\(x_{n,k} &gt; 0\\) if and only if \\(y_n = 1\\), a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as \\(\\beta_k \\rightarrow \\infty\\), because for cases with \\(y_n = 1\\), \\(x_n \\beta \\rightarrow \\infty\\) and hence \\(\\operatorname{logit}^{-1}(x_n \\beta) \\rightarrow 1\\).\nWith separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for \\(\\beta_k\\) is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for \\(\\beta\\), which ensures a proper posterior.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#uniform-posteriors",
+    "href": "stan-users-guide/problematic-posteriors.html#uniform-posteriors",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "Suppose your model includes a parameter \\(\\psi\\) that is defined on \\([0,1]\\) and is given a flat prior \\(\\textsf{uniform}(\\psi \\mid 0,1)\\). Now if the data don’t tell us anything about \\(\\psi\\), the posterior is also \\(\\textsf{uniform}(\\psi \\mid 0,1)\\).\nAlthough there is no maximum likelihood estimate for \\(\\psi\\), the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on \\([0,1]\\), the posterior mean for \\(\\psi\\) is well-defined with value \\(1/2\\). Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for \\(\\psi\\) at all points in \\([0,1]\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#sampling-difficulties-with-problematic-priors",
+    "href": "stan-users-guide/problematic-posteriors.html#sampling-difficulties-with-problematic-priors",
+    "title": "Problematic Posteriors",
+    "section": "",
+    "text": "With an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the collinearity section and illustrated in the non-identifiable density plot.\n\n\nGibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.\nConsider what happens with initial values \\(\\lambda_1^{(0)}, \\lambda_2^{(0)}\\). Gibbs sampling proceeds in iteration \\(m\\) by drawing \\[\\begin{align*}\n\\lambda_1^{(m)} &\\sim p(\\lambda_1 \\mid \\lambda_2^{(m-1)}, \\sigma^{(m-1)},  y) \\\\\n\\lambda_2^{(m)} &\\sim p(\\lambda_2 \\mid \\lambda_1^{(m)},   \\sigma^{(m-1)},  y) \\\\\n\\sigma^{(m)}    &\\sim p(\\sigma    \\mid \\lambda_1^{(m)},   \\lambda_2^{(m)}, y).\n\\end{align*}\\]\nNow consider the draw for \\(\\lambda_1\\) (the draw for \\(\\lambda_2\\) is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next \\(\\lambda_1\\) can be drawn is highly constrained by the current values of \\(\\lambda_2\\) and \\(\\sigma\\). Gibbs will run quickly and provide seemingly reasonable inferences for \\(\\lambda_1 + \\lambda_2\\). But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.\n\n\n\nHamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for \\(\\lambda_1\\) and \\(\\lambda_2\\), the gradient of the log posterior is going to adjust for the correlation and the simulation will run \\(\\lambda_1\\) and \\(\\lambda_2\\) in opposite directions along the valley corresponding to the ridge in the posterior log density.\n\n\n\nStan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior (see Hoffman and Gelman 2014). NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.\nHere are results of Stan runs with default parameters fit to \\(N=100\\) data points generated from \\(y_n \\sim \\textsf{normal}(0,1)\\):\nTwo Scale Parameters, Improper Prior\nInference for Stan model: improper_stan\nWarmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total\nSampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total\n\n                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat\nlp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0\nn_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0\nlambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2\nlambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2\nsigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1\nmu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0\nTwo Scale Parameters, Weak Prior\nWarmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total\nSampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total\n\n                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat\nlp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0\nn_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0\nlambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nlambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nsigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0\nmu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0\nOne Scale Parameter, Improper Prior\nWarmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total\nSampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total\n\n                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat\nlp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0\nn_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0\nmu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0\nsigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0\nOn the top is the non-identified model with improper uniform priors and data model \\(y_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2,\n\\sigma)\\).\nIn the middle is the same data model as in top plus priors \\(\\lambda_k \\sim \\textsf{normal}(0,10)\\).\nOn the bottom is an identified model with an improper prior, with data model \\(y_n \\sim \\textsf{normal}(\\mu,\\sigma)\\). All models estimate \\(\\mu\\) at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value \\(\\mu=0\\) is within the 90% posterior intervals in all three models.\n\n\n\nTo illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the collinearity section.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real lambda1;\n  real lambda2;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  real mu;\n  mu = lambda1 + lambda2;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nThe second adds priors to the model block for lambda1 and lambda2 to the previous model.\nlambda1 ~ normal(0, 10);\nlambda2 ~ normal(0, 10);\nThe third involves a single location parameter, but no priors.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nAll three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.\n\nAs indicated by R_hat column, all parameters have converged other than \\(\\lambda_1\\) and \\(\\lambda_2\\) in the non-identified model.\nThe average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.\nThe effective sample size per second for \\(\\mu\\) is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for \\(\\sigma\\).\nIn the non-identified model, the 95% interval for \\(\\lambda_1\\) is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.\nIn all three models, the simulated value of \\(\\mu=0\\) and \\(\\sigma=1\\) are well within the posterior 90% intervals.\n\nThe first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/problematic-posteriors.html#footnotes",
+    "href": "stan-users-guide/problematic-posteriors.html#footnotes",
+    "title": "Problematic Posteriors",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis example was raised by Richard McElreath on the Stan users group in a query about the difference in behavior between Gibbs sampling as used in BUGS and JAGS and the Hamiltonian Monte Carlo (HMC) and no-U-turn samplers (NUTS) used by Stan.↩︎\nThe marginal posterior \\(p(\\sigma \\mid y)\\) for \\(\\sigma\\) is proper here as long as there are at least two distinct data points.↩︎\nA Laplace prior (or an L1 regularizer for penalized maximum likelihood estimation) is not sufficient to remove this additive invariance. It provides shrinkage, but does not in and of itself identify the parameters because adding a constant to \\(\\lambda_1\\) and subtracting it from \\(\\lambda_2\\) results in the same value for the prior density.↩︎\nTempering methods may be viewed as automated ways to carry out such a search for modes, though most MCMC tempering methods continue to search for modes on an ongoing basis; see (Swendsen and Wang 1986; Neal 1996).↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Problematic Posteriors"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/references.html",
+    "href": "stan-users-guide/references.html",
+    "title": "References",
+    "section": "",
+    "text": "References\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html",
+    "href": "stan-users-guide/reparameterization.html",
+    "title": "Reparameterization and Change of Variables",
+    "section": "",
+    "text": "Stan supports a direct encoding of reparameterizations. Stan also supports changes of variables by directly incrementing the log probability accumulator with the log Jacobian of the transform.\n\n\nA Bayesian posterior is technically a probability measure, which is a parameterization-invariant, abstract mathematical object.1\nStan’s modeling language, on the other hand, defines a probability density, which is a non-unique, parameterization-dependent function in \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^{+}\\). In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.\nAs pointed out by Gelman (2004) in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not just that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.\n\n\n\nReparameterizations may be implemented directly using the transformed parameters block or just in the model block.\n\n\nThe beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.\n\n\nFor example, the Beta distribution is parameterized by two positive count parameters \\(\\alpha, \\beta &gt; 0\\). The following example illustrates a hierarchical Stan model with a vector of parameters theta are drawn i.i.d. for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n  // ...\n}\nmodel {\n  alpha ~ ...\n  beta ~ ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nIt is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter \\[\n\\phi = \\alpha / (\\alpha + \\beta)\n\\] and total count parameter \\[\n\\lambda = \\alpha + \\beta.\n\\] Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with \\(p(\\lambda) \\propto \\lambda^{-2.5}\\).\nparameters {\n  real&lt;lower=0, upper=1&gt; phi;\n  real&lt;lower=0.1&gt; lambda;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0&gt; alpha = lambda * phi;\n  real&lt;lower=0&gt; beta = lambda * (1 - phi);\n  // ...\n}\nmodel {\n  phi ~ beta(1, 1); // uniform on phi, could drop\n  lambda ~ pareto(0.1, 1.5);\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nThe new parameters, phi and lambda, are declared in the parameters block and the parameters for the Beta distribution, alpha and beta, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.\nmodel {\n  real alpha = lambda * phi;\n  real beta = lambda * (1 - phi);\n  // ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nWith vectorization, this could be expressed more compactly and efficiently as follows.\nmodel {\n  theta ~ beta(lambda * phi, lambda * (1 - phi));\n  // ...\n}\nIf the variables alpha and beta are of interest, they can be defined in the transformed parameter block and then used in the model.\n\n\n\nBecause the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, alpha and beta have the correct posterior distribution.\n\n\n\nThe same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are \\(K &gt; 0\\) dimensions being considered (\\(K=1\\) is trivial and \\(K=2\\) reduces to the beta distribution case). The traditional prior is\nparameters {\n  vector[K] alpha;\n  array[N] simplex[K] theta;\n  // ...\n}\nmodel {\n  alpha ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n}\nThis provides essentially \\(K\\) degrees of freedom, one for each dimension of alpha, and it is not obvious how to specify a reasonable prior for alpha.\nAn alternative coding is to use the mean, which is a simplex, and a total count.\nparameters {\n  simplex[K] phi;\n  real&lt;lower=0&gt; kappa;\n  array[N] simplex[K] theta;\n  // ...\n}\ntransformed parameters {\n  vector[K] alpha = kappa * phi;\n  // ...\n}\nmodel {\n  phi ~ // ...\n  kappa ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n  // ...\n}\nNow it is much easier to formulate priors, because phi is the expected value of theta and kappa (minus K) is the strength of the prior mean measured in number of prior observations.\n\n\n\n\nIf the variable \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\operatorname{logit}(u)\\) is distributed as \\(\\textsf{logistic}(0, 1)\\). This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in \\((0, 1)\\) to a logistically-distributed quantity.\nThings work the same way for the probit case: if \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\Phi^{-1}(u)\\) has a \\(\\textsf{normal}(0, 1)\\) distribution. The other way around, if \\(v\\) has a \\(\\textsf{normal}(0, 1)\\) distribution, then \\(\\Phi(v)\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution.\nIn order to use the probit and logistic as priors on variables constrained to \\((0, 1)\\), create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a \\((0, 1)\\)-constrained parameter theta and gives it a beta prior, then uses it as a parameter in a distribution (here using foo as a placeholder).\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nmodel {\n  theta ~ beta(a, b);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIf the variables a and b are one, then this imposes a uniform distribution theta. If a and b are both less than one, then the density on theta has a U shape, whereas if they are both greater than one, the density of theta has an inverted-U or more bell-like shape.\nRoughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,\nparameters {\n  real theta_raw;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0, upper=1&gt; theta = inv_logit(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ logistic(mu, sigma);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIn this model, an unconstrained parameter theta_raw gets a logistic prior, and then the transformed parameter theta is defined to be the inverse logit of theta_raw. In this parameterization, inv_logit(mu) is the mean of the implied prior on theta. The prior distribution on theta will be flat if sigma is one and mu is zero, and will be U-shaped if sigma is larger than one and bell shaped if sigma is less than one.\nWhen moving from a variable in \\((0, 1)\\) to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.\nparameters {\n  simplex[K] theta;\n  // ...\n}\nmodel {\n  theta ~ dirichlet(a);\n  // ...\n  y ~ foo(theta);\n}\nNow a is a vector with K rows, but it has the same shape properties as the pair a and b for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector \\([a b]^{\\top}\\). To formulate an unconstrained prior, the exact same strategy works as for the beta.\nparameters {\n  vector[K] theta_raw;\n  // ...\n}\ntransformed parameters {\n  simplex[K] theta = softmax(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by softmax(mu), but we have additional control of covariance through L_Sigma at the expense of having on the order of \\(K^2\\) parameters in the prior rather than order \\(K\\). If no covariance is desired, the number of parameters can be reduced back to \\(K\\) using a vectorized normal distribution as follows.\ntheta_raw ~ normal(mu, sigma);\nwhere either or both of mu and sigma can be vectors.\n\n\n\n\nChanges of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable \\(y &gt; 0\\) whose logarithm \\(\\log y\\) has a normal distribution. The distribution is being assigned to \\(\\log y\\).\nThe change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^N\\).\nThe probability must be scaled by a Jacobian adjustment equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.\nIn the case of log normals, if \\(y\\)’s logarithm is normal with mean \\(\\mu\\) and deviation \\(\\sigma\\), then the distribution of \\(y\\) is given by \\[\np(y)\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\left| \\frac{d}{dy} \\log y \\right|\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\frac{1}{y}.\n\\] Stan works on the log scale to prevent underflow, where \\[\n\\log p(y)\n=\n\\log \\textsf{normal}(\\log y \\mid \\mu, \\sigma) - \\log y.\n\\]\nIn Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.2\nparameters {\n  real&lt;lower=0&gt; y;\n  // ...\n}\nmodel {\n  log(y) ~ normal(mu, sigma);\n  target += -log(y);\n  // ...\n}\nIt is important, as always, to declare appropriate constraints on parameters; here y is constrained to be positive.\nIt would be slightly more efficient to define a local variable for the logarithm, as follows.\nmodel {\n  real log_y;\n  log_y = log(y);\n  log_y ~ normal(mu, sigma);\n  target += -log_y;\n  // ...\n}\nIf y were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.\n\n\nThis section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.\nIt does not matter whether the probability function is expressed using a distribution statement, such as\nlog(y) ~ normal(mu, sigma);\nor as an increment to the log probability function, as in\ntarget += normal_lpdf(log(y) | mu, sigma);\n\n\nLike the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.\nThe transform based approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y_inv;\n  y_inv = 1 / y;\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe change-of-variables approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y_inv;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y;\n  y = 1 / y_inv;  // change variables\n  jacobian += -2 * log(y_inv); // Jacobian adjustment\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is\n\\[\n\\log \\left| \\frac{d}{du} \\left( \\frac{1}{u} \\right) \\right|\n= \\log \\left| - u^{-2} \\right|\n= \\log u^{-2}\n=  -2 \\log u.\n\\]\n\n\n\n\nIn the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.\nparameters {\n  vector[K] u;      // multivariate parameter\n   // ...\n}\ntransformed parameters {\n  vector[K] v;     // transformed parameter\n  matrix[K, K] J;   // Jacobian matrix of transform\n  // ... compute v as a function of u ...\n  // ... compute J[m, n] = d.v[m] / d.u[n] ...\n  jacobian += log(abs(determinant(J)));\n  // ...\n}\nmodel {\n  v ~ // ...\n  // ...\n}\nIf the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.\nIn many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element v[k] of the transformed parameter vector only depends on elements u[1], …, u[k] of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.\ntransformed parameters {\n  // ...\n  vector[K] J_diag;  // diagonals of Jacobian matrix\n  // ...\n  // ... compute J[k, k] = d.v[k] / d.u[k] ...\n  jacobian += sum(log(J_diag));\n  // ...\n}\n\n\n\n\nStan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.\n\n\nFor example, suppose there is a vector parameter \\(\\alpha\\) with a vector \\(L\\) of lower bounds. The simplest way to deal with this if \\(L\\) is a constant is to shift a lower-bounded parameter.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L&gt;[N] alpha_raw;\n  // ...\n}\nThe above is equivalent to manually calculating the vector bounds by the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + alpha_raw;\n  // ...\n}\nThe Jacobian for adding a constant is one, so its log drops out of the log density.\nEven if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from \\((L, \\alpha_{\\textrm{raw}})\\) to \\((L + \\alpha_{\\textrm{raw}}, \\alpha_{\\textrm{raw}})\\) produces a Jacobian derivative matrix with a unit determinant.\nIt’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  vector&lt;lower=L + alpha_raw&gt;[N] alpha;\n  // ...\n}\nThis is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + exp(alpha_raw);\n  jacobian += sum(alpha_raw); // log Jacobian\n  // ...\n}\nmodel {\n  // ...\n}\nThe adjustment in the log Jacobian determinant of the transform mapping \\(\\alpha_{\\textrm{raw}}\\) to \\(\\alpha = L + \\exp(\\alpha_{\\textrm{raw}})\\). The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here \\(L\\) can even be a vector containing parameters that don’t depend on \\(\\alpha_{\\textrm{raw}}\\); if the bounds do depend on \\(\\alpha_{\\textrm{raw}}\\) then a revised Jacobian needs to be calculated taking into account the dependencies.\n\n\n\nSuppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to \\((0, 1)\\). This is easily accomplished as the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L, upper=U&gt;[N] alpha;\n  // ...\n}\nThe same may be accomplished by manually constructing the transform as follows.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + (U - L) .* alpha_raw;\n}\nThe expression U - L is multiplied by alpha_raw elementwise to produce a vector of variables in \\((0, U-L)\\), then adding \\(L\\) results in a variable ranging between \\((L, U)\\).\nIn this case, it is important that \\(L\\) and \\(U\\) are constants, otherwise a Jacobian would be required when multiplying by \\(U - L\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html#theoretical-and-practical-background",
+    "href": "stan-users-guide/reparameterization.html#theoretical-and-practical-background",
+    "title": "Reparameterization and Change of Variables",
+    "section": "",
+    "text": "A Bayesian posterior is technically a probability measure, which is a parameterization-invariant, abstract mathematical object.1\nStan’s modeling language, on the other hand, defines a probability density, which is a non-unique, parameterization-dependent function in \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^{+}\\). In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.\nAs pointed out by Gelman (2004) in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not just that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html#reparameterizations",
+    "href": "stan-users-guide/reparameterization.html#reparameterizations",
+    "title": "Reparameterization and Change of Variables",
+    "section": "",
+    "text": "Reparameterizations may be implemented directly using the transformed parameters block or just in the model block.\n\n\nThe beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.\n\n\nFor example, the Beta distribution is parameterized by two positive count parameters \\(\\alpha, \\beta &gt; 0\\). The following example illustrates a hierarchical Stan model with a vector of parameters theta are drawn i.i.d. for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n  // ...\n}\nmodel {\n  alpha ~ ...\n  beta ~ ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nIt is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter \\[\n\\phi = \\alpha / (\\alpha + \\beta)\n\\] and total count parameter \\[\n\\lambda = \\alpha + \\beta.\n\\] Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with \\(p(\\lambda) \\propto \\lambda^{-2.5}\\).\nparameters {\n  real&lt;lower=0, upper=1&gt; phi;\n  real&lt;lower=0.1&gt; lambda;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0&gt; alpha = lambda * phi;\n  real&lt;lower=0&gt; beta = lambda * (1 - phi);\n  // ...\n}\nmodel {\n  phi ~ beta(1, 1); // uniform on phi, could drop\n  lambda ~ pareto(0.1, 1.5);\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nThe new parameters, phi and lambda, are declared in the parameters block and the parameters for the Beta distribution, alpha and beta, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.\nmodel {\n  real alpha = lambda * phi;\n  real beta = lambda * (1 - phi);\n  // ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nWith vectorization, this could be expressed more compactly and efficiently as follows.\nmodel {\n  theta ~ beta(lambda * phi, lambda * (1 - phi));\n  // ...\n}\nIf the variables alpha and beta are of interest, they can be defined in the transformed parameter block and then used in the model.\n\n\n\nBecause the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, alpha and beta have the correct posterior distribution.\n\n\n\nThe same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are \\(K &gt; 0\\) dimensions being considered (\\(K=1\\) is trivial and \\(K=2\\) reduces to the beta distribution case). The traditional prior is\nparameters {\n  vector[K] alpha;\n  array[N] simplex[K] theta;\n  // ...\n}\nmodel {\n  alpha ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n}\nThis provides essentially \\(K\\) degrees of freedom, one for each dimension of alpha, and it is not obvious how to specify a reasonable prior for alpha.\nAn alternative coding is to use the mean, which is a simplex, and a total count.\nparameters {\n  simplex[K] phi;\n  real&lt;lower=0&gt; kappa;\n  array[N] simplex[K] theta;\n  // ...\n}\ntransformed parameters {\n  vector[K] alpha = kappa * phi;\n  // ...\n}\nmodel {\n  phi ~ // ...\n  kappa ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n  // ...\n}\nNow it is much easier to formulate priors, because phi is the expected value of theta and kappa (minus K) is the strength of the prior mean measured in number of prior observations.\n\n\n\n\nIf the variable \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\operatorname{logit}(u)\\) is distributed as \\(\\textsf{logistic}(0, 1)\\). This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in \\((0, 1)\\) to a logistically-distributed quantity.\nThings work the same way for the probit case: if \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\Phi^{-1}(u)\\) has a \\(\\textsf{normal}(0, 1)\\) distribution. The other way around, if \\(v\\) has a \\(\\textsf{normal}(0, 1)\\) distribution, then \\(\\Phi(v)\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution.\nIn order to use the probit and logistic as priors on variables constrained to \\((0, 1)\\), create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a \\((0, 1)\\)-constrained parameter theta and gives it a beta prior, then uses it as a parameter in a distribution (here using foo as a placeholder).\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nmodel {\n  theta ~ beta(a, b);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIf the variables a and b are one, then this imposes a uniform distribution theta. If a and b are both less than one, then the density on theta has a U shape, whereas if they are both greater than one, the density of theta has an inverted-U or more bell-like shape.\nRoughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,\nparameters {\n  real theta_raw;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0, upper=1&gt; theta = inv_logit(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ logistic(mu, sigma);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIn this model, an unconstrained parameter theta_raw gets a logistic prior, and then the transformed parameter theta is defined to be the inverse logit of theta_raw. In this parameterization, inv_logit(mu) is the mean of the implied prior on theta. The prior distribution on theta will be flat if sigma is one and mu is zero, and will be U-shaped if sigma is larger than one and bell shaped if sigma is less than one.\nWhen moving from a variable in \\((0, 1)\\) to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.\nparameters {\n  simplex[K] theta;\n  // ...\n}\nmodel {\n  theta ~ dirichlet(a);\n  // ...\n  y ~ foo(theta);\n}\nNow a is a vector with K rows, but it has the same shape properties as the pair a and b for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector \\([a b]^{\\top}\\). To formulate an unconstrained prior, the exact same strategy works as for the beta.\nparameters {\n  vector[K] theta_raw;\n  // ...\n}\ntransformed parameters {\n  simplex[K] theta = softmax(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by softmax(mu), but we have additional control of covariance through L_Sigma at the expense of having on the order of \\(K^2\\) parameters in the prior rather than order \\(K\\). If no covariance is desired, the number of parameters can be reduced back to \\(K\\) using a vectorized normal distribution as follows.\ntheta_raw ~ normal(mu, sigma);\nwhere either or both of mu and sigma can be vectors.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html#changes-of-variables",
+    "href": "stan-users-guide/reparameterization.html#changes-of-variables",
+    "title": "Reparameterization and Change of Variables",
+    "section": "",
+    "text": "Changes of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable \\(y &gt; 0\\) whose logarithm \\(\\log y\\) has a normal distribution. The distribution is being assigned to \\(\\log y\\).\nThe change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^N\\).\nThe probability must be scaled by a Jacobian adjustment equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.\nIn the case of log normals, if \\(y\\)’s logarithm is normal with mean \\(\\mu\\) and deviation \\(\\sigma\\), then the distribution of \\(y\\) is given by \\[\np(y)\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\left| \\frac{d}{dy} \\log y \\right|\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\frac{1}{y}.\n\\] Stan works on the log scale to prevent underflow, where \\[\n\\log p(y)\n=\n\\log \\textsf{normal}(\\log y \\mid \\mu, \\sigma) - \\log y.\n\\]\nIn Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.2\nparameters {\n  real&lt;lower=0&gt; y;\n  // ...\n}\nmodel {\n  log(y) ~ normal(mu, sigma);\n  target += -log(y);\n  // ...\n}\nIt is important, as always, to declare appropriate constraints on parameters; here y is constrained to be positive.\nIt would be slightly more efficient to define a local variable for the logarithm, as follows.\nmodel {\n  real log_y;\n  log_y = log(y);\n  log_y ~ normal(mu, sigma);\n  target += -log_y;\n  // ...\n}\nIf y were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.\n\n\nThis section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.\nIt does not matter whether the probability function is expressed using a distribution statement, such as\nlog(y) ~ normal(mu, sigma);\nor as an increment to the log probability function, as in\ntarget += normal_lpdf(log(y) | mu, sigma);\n\n\nLike the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.\nThe transform based approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y_inv;\n  y_inv = 1 / y;\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe change-of-variables approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y_inv;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y;\n  y = 1 / y_inv;  // change variables\n  jacobian += -2 * log(y_inv); // Jacobian adjustment\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is\n\\[\n\\log \\left| \\frac{d}{du} \\left( \\frac{1}{u} \\right) \\right|\n= \\log \\left| - u^{-2} \\right|\n= \\log u^{-2}\n=  -2 \\log u.\n\\]\n\n\n\n\nIn the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.\nparameters {\n  vector[K] u;      // multivariate parameter\n   // ...\n}\ntransformed parameters {\n  vector[K] v;     // transformed parameter\n  matrix[K, K] J;   // Jacobian matrix of transform\n  // ... compute v as a function of u ...\n  // ... compute J[m, n] = d.v[m] / d.u[n] ...\n  jacobian += log(abs(determinant(J)));\n  // ...\n}\nmodel {\n  v ~ // ...\n  // ...\n}\nIf the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.\nIn many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element v[k] of the transformed parameter vector only depends on elements u[1], …, u[k] of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.\ntransformed parameters {\n  // ...\n  vector[K] J_diag;  // diagonals of Jacobian matrix\n  // ...\n  // ... compute J[k, k] = d.v[k] / d.u[k] ...\n  jacobian += sum(log(J_diag));\n  // ...\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html#vectors-with-varying-bounds",
+    "href": "stan-users-guide/reparameterization.html#vectors-with-varying-bounds",
+    "title": "Reparameterization and Change of Variables",
+    "section": "",
+    "text": "Stan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.\n\n\nFor example, suppose there is a vector parameter \\(\\alpha\\) with a vector \\(L\\) of lower bounds. The simplest way to deal with this if \\(L\\) is a constant is to shift a lower-bounded parameter.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L&gt;[N] alpha_raw;\n  // ...\n}\nThe above is equivalent to manually calculating the vector bounds by the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + alpha_raw;\n  // ...\n}\nThe Jacobian for adding a constant is one, so its log drops out of the log density.\nEven if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from \\((L, \\alpha_{\\textrm{raw}})\\) to \\((L + \\alpha_{\\textrm{raw}}, \\alpha_{\\textrm{raw}})\\) produces a Jacobian derivative matrix with a unit determinant.\nIt’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  vector&lt;lower=L + alpha_raw&gt;[N] alpha;\n  // ...\n}\nThis is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + exp(alpha_raw);\n  jacobian += sum(alpha_raw); // log Jacobian\n  // ...\n}\nmodel {\n  // ...\n}\nThe adjustment in the log Jacobian determinant of the transform mapping \\(\\alpha_{\\textrm{raw}}\\) to \\(\\alpha = L + \\exp(\\alpha_{\\textrm{raw}})\\). The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here \\(L\\) can even be a vector containing parameters that don’t depend on \\(\\alpha_{\\textrm{raw}}\\); if the bounds do depend on \\(\\alpha_{\\textrm{raw}}\\) then a revised Jacobian needs to be calculated taking into account the dependencies.\n\n\n\nSuppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to \\((0, 1)\\). This is easily accomplished as the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L, upper=U&gt;[N] alpha;\n  // ...\n}\nThe same may be accomplished by manually constructing the transform as follows.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + (U - L) .* alpha_raw;\n}\nThe expression U - L is multiplied by alpha_raw elementwise to produce a vector of variables in \\((0, U-L)\\), then adding \\(L\\) results in a variable ranging between \\((L, U)\\).\nIn this case, it is important that \\(L\\) and \\(U\\) are constants, otherwise a Jacobian would be required when multiplying by \\(U - L\\).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html#footnotes",
+    "href": "stan-users-guide/reparameterization.html#footnotes",
+    "title": "Reparameterization and Change of Variables",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis is in contrast to (penalized) maximum likelihood estimates, which are not parameterization invariant.↩︎\nThis example is for illustrative purposes only; the recommended way to implement the lognormal distribution in Stan is with the built-in lognormal probability function; see the functions reference manual for details.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/sparse-ragged.html",
+    "href": "stan-users-guide/sparse-ragged.html",
+    "title": "Sparse and Ragged Data Structures",
+    "section": "",
+    "text": "Stan does not directly support either sparse or ragged data structures, though both can be accommodated with some programming effort. The sparse matrices chapter introduces a special-purpose sparse matrix times dense vector multiplication, which should be used where applicable; this chapter covers more general data structures.\n\n\nCoding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the item-response model section. There are \\(J\\) students and \\(K\\) questions, and if every student answers every question, then it is practical to declare the data as a \\(J \\times K\\) array of answers.\ndata {\n  int&lt;lower=1&gt; J;\n  int&lt;lower=1&gt; K;\n  array[J, K] int&lt;lower=0, upper=1&gt; y;\n  // ...\nmodel {\n  for (j in 1:J) {\n    for (k in 1:K) {\n      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));\n    }\n  }\n  // ...\n}\nWhen not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.\nThe following missing data example shows an example with \\(J=3\\) and \\(K=4\\), with missing responses shown as NA, as in R.\n\\[\\begin{equation*}\ny\n=\n\\left[\n\\begin{array}{cccc}\n0 & 1 & \\mbox{NA} & 1\n\\\\\n0 & \\mbox{NA} & \\mbox{NA} & 1\n\\\\\n\\mbox{NA} & 0 & \\mbox{NA} & \\mbox{NA}\n\\end{array}\n\\right]\n\\end{equation*}\\]\nThere is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns \\(jj\\) and \\(kk\\) used for the indexes (following Gelman and Hill (2007)), the 2-D array \\(y\\) is recoded as a table. The number of rows in the table equals the number of defined array elements, here \\(y_{1,1} = 0\\), \\(y_{1,2} = 1\\), up to \\(y_{3,2} = 1\\). As the array becomes larger and sparser, the long form becomes the more economical encoding.\n\n\n\njj\nkk\ny\n\n\n\n\n1\n1\n0\n\n\n1\n2\n1\n\n\n1\n4\n1\n\n\n2\n1\n0\n\n\n2\n4\n1\n\n\n3\n2\n0\n\n\n\nLetting \\(N\\) be the number of \\(y\\) that are defined, here \\(N=6\\), the data and model can be formulated as follows.\ndata {\n  // ...\n  int&lt;lower=1&gt; N;\n  array[N] int&lt;lower=1, upper=J&gt; jj;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  // ...\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(delta[kk[n]]\n                           * (alpha[jj[n]] - beta[kk[n]]));\n  }\n  // ...\n}\nIn the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.\n\n\n\nRagged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.\nA general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.\nFor example, consider a data structure for three groups, each of which has a different number of observations.\n\n\n\n\n\n\n\\(y_1 =  \\left[1.3 \\ \\ 2.4 \\ \\ 0.9\\right]\\\\\\) \\(y_2 = \\left[-1.8 \\ \\ -0.1\\right]\\\\\\) \\(y_3 = \\left[12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7\\right]\\)\n\n\n\\(z = [1.3 \\ \\ 2.4 \\ \\ 0.9 \\ \\ -1.8 \\ \\ -0.1 \\ \\ 12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7]\\\\\\) \\(s  =  \\{ 3 \\ \\ 2 \\ \\ 4 \\}\\)\n\n\n\nOn the left is the definition of a ragged data structure \\(y\\) with three rows of different sizes (\\(y_1\\) is size 3, \\(y_2\\) size 2, and \\(y_3\\) size 4). On the right is an example of how to code the data in Stan, using a single vector \\(z\\) to hold all the values and a separate array of integers \\(s\\) to hold the group row sizes. In this example, \\(y_1 = z_{1:3}\\), \\(y_2 = z_{4:5}\\), and \\(y_3 = z_{6:9}\\).\nSuppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood \\[\\begin{equation*}\n\\sum_{n=1}^3 \\log \\textsf{normal}(y_n \\mid \\mu_n, \\sigma).\n\\end{equation*}\\] There’s no direct way to encode this in Stan.\nA full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.\nThe model can then be coded up using slicing operations as follows.\ndata {\n  int&lt;lower=0&gt; N;   // # observations\n  int&lt;lower=0&gt; K;   // # of groups\n  vector[N] y;      // observations\n  array[K] int s;   // group sizes\n  // ...\n}\nmodel {\n  int pos;\n  pos = 1;\n  for (k in 1:K) {\n    segment(y, pos, s[k]) ~ normal(mu[k], sigma);\n    pos = pos + s[k];\n  }\nThis coding allows for efficient vectorization, which is worth the copy cost entailed by the segment() vector slicing operation.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Sparse and Ragged Data Structures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/sparse-ragged.html#sparse-data-structures",
+    "href": "stan-users-guide/sparse-ragged.html#sparse-data-structures",
+    "title": "Sparse and Ragged Data Structures",
+    "section": "",
+    "text": "Coding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the item-response model section. There are \\(J\\) students and \\(K\\) questions, and if every student answers every question, then it is practical to declare the data as a \\(J \\times K\\) array of answers.\ndata {\n  int&lt;lower=1&gt; J;\n  int&lt;lower=1&gt; K;\n  array[J, K] int&lt;lower=0, upper=1&gt; y;\n  // ...\nmodel {\n  for (j in 1:J) {\n    for (k in 1:K) {\n      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));\n    }\n  }\n  // ...\n}\nWhen not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.\nThe following missing data example shows an example with \\(J=3\\) and \\(K=4\\), with missing responses shown as NA, as in R.\n\\[\\begin{equation*}\ny\n=\n\\left[\n\\begin{array}{cccc}\n0 & 1 & \\mbox{NA} & 1\n\\\\\n0 & \\mbox{NA} & \\mbox{NA} & 1\n\\\\\n\\mbox{NA} & 0 & \\mbox{NA} & \\mbox{NA}\n\\end{array}\n\\right]\n\\end{equation*}\\]\nThere is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns \\(jj\\) and \\(kk\\) used for the indexes (following Gelman and Hill (2007)), the 2-D array \\(y\\) is recoded as a table. The number of rows in the table equals the number of defined array elements, here \\(y_{1,1} = 0\\), \\(y_{1,2} = 1\\), up to \\(y_{3,2} = 1\\). As the array becomes larger and sparser, the long form becomes the more economical encoding.\n\n\n\njj\nkk\ny\n\n\n\n\n1\n1\n0\n\n\n1\n2\n1\n\n\n1\n4\n1\n\n\n2\n1\n0\n\n\n2\n4\n1\n\n\n3\n2\n0\n\n\n\nLetting \\(N\\) be the number of \\(y\\) that are defined, here \\(N=6\\), the data and model can be formulated as follows.\ndata {\n  // ...\n  int&lt;lower=1&gt; N;\n  array[N] int&lt;lower=1, upper=J&gt; jj;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  // ...\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(delta[kk[n]]\n                           * (alpha[jj[n]] - beta[kk[n]]));\n  }\n  // ...\n}\nIn the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Sparse and Ragged Data Structures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/sparse-ragged.html#ragged-data-structs.section",
+    "href": "stan-users-guide/sparse-ragged.html#ragged-data-structs.section",
+    "title": "Sparse and Ragged Data Structures",
+    "section": "",
+    "text": "Ragged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.\nA general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.\nFor example, consider a data structure for three groups, each of which has a different number of observations.\n\n\n\n\n\n\n\\(y_1 =  \\left[1.3 \\ \\ 2.4 \\ \\ 0.9\\right]\\\\\\) \\(y_2 = \\left[-1.8 \\ \\ -0.1\\right]\\\\\\) \\(y_3 = \\left[12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7\\right]\\)\n\n\n\\(z = [1.3 \\ \\ 2.4 \\ \\ 0.9 \\ \\ -1.8 \\ \\ -0.1 \\ \\ 12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7]\\\\\\) \\(s  =  \\{ 3 \\ \\ 2 \\ \\ 4 \\}\\)\n\n\n\nOn the left is the definition of a ragged data structure \\(y\\) with three rows of different sizes (\\(y_1\\) is size 3, \\(y_2\\) size 2, and \\(y_3\\) size 4). On the right is an example of how to code the data in Stan, using a single vector \\(z\\) to hold all the values and a separate array of integers \\(s\\) to hold the group row sizes. In this example, \\(y_1 = z_{1:3}\\), \\(y_2 = z_{4:5}\\), and \\(y_3 = z_{6:9}\\).\nSuppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood \\[\\begin{equation*}\n\\sum_{n=1}^3 \\log \\textsf{normal}(y_n \\mid \\mu_n, \\sigma).\n\\end{equation*}\\] There’s no direct way to encode this in Stan.\nA full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.\nThe model can then be coded up using slicing operations as follows.\ndata {\n  int&lt;lower=0&gt; N;   // # observations\n  int&lt;lower=0&gt; K;   // # of groups\n  vector[N] y;      // observations\n  array[K] int s;   // group sizes\n  // ...\n}\nmodel {\n  int pos;\n  pos = 1;\n  for (k in 1:K) {\n    segment(y, pos, s[k]) ~ normal(mu[k], sigma);\n    pos = pos + s[k];\n  }\nThis coding allows for efficient vectorization, which is worth the copy cost entailed by the segment() vector slicing operation.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Sparse and Ragged Data Structures"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html",
+    "href": "stan-users-guide/survival.html",
+    "title": "Survival Models",
+    "section": "",
+    "text": "Survival models apply to animals and plants as well as inanimate objects such as machine parts or electrical components. Survival models arise when there is an event of interest for a group of subjects, machine component, or other item that is\n\ncertain to occur after some amount of time,\nbut only measured for a fixed period of time, during which the event may not have occurred for all subjects.\n\nFor example, one might wish to estimate the the distribution of time to failure for solid state drives in a data center, but only measure drives for a two year period, after which some number will have failed and some will still be in service.\nSurvival models are often used comparatively, such as comparing time to death of patients diagnosed with stage one liver cancer under a new treatment and a standard treatment (pure controls are not allowed when there is an effective existing treatment for a serious condition). During a two year trial, some patients will die and others will survive.\nSurvival models may involve covariates, such as the factory at which a component is manufactured, the day on which it is manufactured, and the amount of usage it gets. A clinical trial might be adjusted for the sex and age of a cancer patient or the hospital at which treatment is received.\nSurvival models come in two main flavors, parametric and semi-parametric. In a parametric model, the survival time of a subject is modeled explicitly using a parametric probability distribution. There is a great deal of flexibility in how the parametric probability distribution is constructed. The sections below consider exponential and Weibull distributed survival times.\nRather than explicitly modeling a parametric survival probability, semi-parametric survival models instead model the relative effect on survival of covariates. The final sections of this chapter consider the proportional hazards survival model.\n\n\nThe exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if \\(T \\sim \\textrm{exponential}(\\lambda)\\) for some rate \\(\\lambda &gt; 0,\\) then \\[\\begin{equation*}\n\\Pr[T &gt; t] = \\Pr[T &gt; t + t' \\mid T &gt; t'].\n\\end{equation*}\\] If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.\nThe exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable \\(T\\) is \\[\\begin{equation*}\n\\mathbb{E}[T \\mid \\lambda] = \\frac{1}{\\lambda}.\n\\end{equation*}\\] The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) \\(\\beta = 1 / \\lambda\\).\nThe data for a survival model consists of two components. First, there is a vector \\(t \\in (0, \\infty)^N\\) of \\(N\\) observed failure times. Second, there is a censoring time \\(t^{\\textrm{cens}}\\) such that failure times greater than \\(t^{\\textrm{cens}}\\) are not observed. The censoring time assumption imposes a constraint which requires \\(t_n &lt; t^{\\textrm{cens}}\\) for all \\(n \\in 1{:}N.\\) For the censored subjects, the only thing required in the model is their total count, \\(N^\\textrm{cens}\\) (their covariates are also required for models with covariates).\nThe model for the observed failure times is exponential, so that for \\(n \\in 1{:}N,\\) \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda).\n\\end{equation*}\\]\nThe model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of \\[\\begin{equation*}\n\\Pr[T &gt; t^{\\textrm{cens}}] = 1 - F_T(t^{\\textrm{cens}}),\n\\end{equation*}\\] where \\(F_T\\) is the cumulative distribution function (cdf) of survival time \\(T\\) (\\(F_X(x) = \\Pr[X \\leq x]\\) is standard notation for the cdf of a random variable \\(X\\)). The function \\(1 - F_T(t)\\) is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood \\[\\begin{eqnarray*}\np(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\prod_{n=1}^{N^{\\textrm{cens}}}\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)\n\\\\\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)^{N^{\\textrm{cens}}}.\n\\end{eqnarray*}\\]\nOn the log scale, that’s \\[\\begin{eqnarray*}\n\\log p(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\sum_{n=1}^N \\log \\textrm{exponential}(t_n \\mid \\lambda)\n\\\\\n& & { } + N^{\\textrm{cens}} \\cdot \\log \\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda).\n\\end{eqnarray*}\\]\nThe model can be completed with a standard lognormal prior on \\(\\lambda,\\) \\[\\begin{equation*}\n\\lambda \\sim \\textrm{lognormal}(0, 1),\n\\end{equation*}\\] which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed \\(\\textrm{lognormal}(0, 1)\\). In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.\n\n\nThe data for a simple survival analysis without covariates can be coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n}\nIn this program, N is the number of uncensored observations and t contains the times of the uncensored observations. There are a further N_cens items that are right censored at time t_cens. Right censoring means that if the time to failure is greater than\nt_cens, it is only observed that the part survived until time t_cens. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.\nThere is a single rate parameter, the inverse of which is the expected time to failure.\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nThe exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.\nmodel {\n  t ~ exponential(lambda);\n  target += N_cens * exponential_lccdf(t_cens | lambda);\n\n  lambda ~ lognormal(0, 1);\n}\nThe likelihood for rate lambda is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector t as a having an exponential distribution with rate lambda. This data model could have been written as\nfor (n in 1:N) {\n  t[n] ~ exponential(lambda);\n}\nThe log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate lambda. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,\nfor (n in 1:N)\n  target += exponential_lccdf(t_cens | lambda);\nto define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.\n\n\n\n\nThe Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.\nThe Weibull distribution is parameterized by a shape \\(\\alpha &gt; 0\\) and scale \\(\\sigma &gt; 0.\\) For an outcome \\(t \\geq 0\\), the Weibull distribution’s probability density function is \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid \\alpha, \\sigma)\n= \\frac{\\alpha}{\\sigma}\n  \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n  \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^{\\alpha}\\right).\n\\end{equation*}\\] In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter \\(\\beta &gt; 0\\) with probability density function \\[\\begin{equation*}\n\\textrm{exponential}(t \\mid \\beta) =\n\\beta\n\\cdot\n\\exp(-\\beta \\cdot t).\n\\end{equation*}\\] When \\(\\alpha = 1,\\) the Weibull distribution reduces to an exponential distribution, \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid 1, \\sigma)\n=\n\\textrm{exponential}\\!\\left(t \\,\\bigg|\\, \\frac{1}{\\sigma}\\right).\n\\end{equation*}\\] In other words, the Weibull is a continuous expansion of the exponential distribution.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then the expected survival time is \\[\\begin{equation*}\n\\mathbb{E}[T] = \\sigma \\cdot \\Gamma\\!\\left(1 + \\frac{1}{\\alpha}\\right),\n\\end{equation*}\\] where the \\(\\Gamma\\) function is the continuous completion of the factorial function (i.e., \\(\\Gamma(1 + n) = n!\\ \\) for \\(n \\in\n\\mathbb{N}\\)). As \\(\\alpha \\rightarrow 0\\) for a fixed \\(\\sigma\\) or as \\(\\sigma \\rightarrow \\infty\\) for a fixed \\(\\alpha\\), the expected survival time goes to infinity.\nThere are three regimes of the Weibull distribution.\n\n\\(\\alpha &lt; 1.\\) A subject is more likely to fail early. When \\(\\alpha\n&lt; 1,\\) the Weibull density approaches infinity as \\(t \\rightarrow 0.\\)\n\\(\\alpha = 1.\\) The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When \\(\\alpha = 1,\\) the Weibull distribution approaches \\(\\sigma\\) as \\(t\n\\rightarrow 0.\\)\n\\(\\alpha &gt; 1.\\) Subjects are less likely to fail early. When \\(\\alpha &lt; 1,\\) the Weibull density approaches zero as \\(t \\rightarrow 0.\\)\n\nWith \\(\\alpha \\leq 1,\\) the mode is zero (\\(t = 0\\)), whereas with \\(\\alpha &gt; 1,\\) the mode is nonzero (\\(t &gt; 0\\)).\n\n\nWith Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.\nparameters {\n  real&lt;lower=0&gt; beta;\n}\nmodel {\n  t ~ exponential(beta);\n  target += N_cens * exponential_lccdf(t_cens | beta);\n\n  beta ~ lognormal(0, 1);\n}\nThe Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (alpha) and scale (sigma) parameters.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  t ~ weibull(alpha, sigma);\n  target += N_cens * weibull_lccdf(t_cens | alpha, sigma);\n\n  alpha ~ lognormal(0, 1);\n  sigma ~ lognormal(0, 1);\n}\nAs usual, if more is known about expected survival times, alpha and sigma should be given more informative priors.\n\n\n\n\nSuppose that for each of \\(n \\in 1{:}N\\) items observed, both censored and uncensored, there is a covariate (row) vector \\(x_n \\in\n\\mathbb{R}^K.\\) For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.\nSurvival with covariates replaces what is essentially a simple regression with only an intercept \\(\\lambda\\) with a generalized linear model with a log link, where the rate for item \\(n\\) is \\[\\begin{equation*}\n\\lambda_n = \\exp(x_n \\cdot \\beta),\n\\end{equation*}\\] where \\(\\beta \\in \\mathbb{R}^K\\) is a \\(K\\)-vector of regression coefficients. Thus \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda_n).\n\\end{equation*}\\] The censored items have probability \\[\\begin{equation*}\n\\Pr[n\\textrm{-th censored}] =\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid x^{\\textrm{cens}}_n\n\\cdot \\beta).\n\\end{equation*}\\]\nThe covariates form an \\(N \\times K\\) data matrix, \\(x \\in\n\\mathbb{R}^{N \\times K}\\). An intercept can be introduced by adding a column of 1 values to \\(x\\).\nA Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (t and t_cens) corresponds to the ordering of items in the covariate matrices (x and x_cens).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n  int&lt;lower=0&gt; K;\n  matrix[N, K] x;\n  matrix[N_cens, K] x_cens;\n}\nparameters {\n  vector[K] gamma;\n}\nmodel {\n  gamma ~ normal(0, 2);\n\n  t ~ exponential(exp(x * gamma));\n  target += exponential_lccdf(t_cens | exp(x_cens * gamma));\n}\nBoth the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.\n\n\n\nSuppose \\(T\\) is a random variable representing a survival time, with a smooth cumulative distribution function \\[\\begin{equation*}\nF_T(t) = \\Pr[T \\leq t],\n\\end{equation*}\\] so that its probability density function is \\[\\begin{equation*}\np_T(t) = \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t).\n\\end{equation*}\\]\nThe survival function \\(S(t)\\) is the probability of surviving until at least time \\(t\\), which is just the complementary cumulative distribution function (ccdf) of the survival random variable \\(T\\), \\[\\begin{equation*}\nS(t) = 1 - F_T(t).\n\\end{equation*}\\] The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).\nThe hazard function \\(h(t)\\) is the instantaneous risk of not surviving past time \\(t\\) assuming survival until time \\(t\\), which is given by \\[\\begin{equation*}\nh(t) = \\frac{p_T(t)}{S(t)} = \\frac{p_T(t)}{1 - F_T(t)}.\n\\end{equation*}\\] The cumulative hazard function \\(H(t)\\) is defined to be the accumulated hazard over time, \\[\\begin{equation*}\nH(t) = \\int_0^t h(u) \\, \\textrm{d}u.\n\\end{equation*}\\]\nThe hazard function and survival function are related through the differential equation \\[\\begin{eqnarray*}\nh(t) & = & -\\frac{\\textrm{d}}{\\textrm{d}t} \\log S(t).\n\\\\[4pt]\n& = & -\\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} S(t)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} -(1 - F_T(t))\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} (F_T(t) - 1)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t)\n\\\\[4pt]\n& = & \\frac{p_T(t)}{S(t)}.\n\\end{eqnarray*}\\]\nIf \\(T \\sim \\textrm{exponential}(\\beta)\\) has an exponential distribution, then its hazard function is constant, \\[\\begin{eqnarray*}\nh(t \\mid \\beta)\n& = & \\frac{p_T(t \\mid \\beta)}{S(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\textrm{exponential}(t \\mid \\beta)}{1 - \\textrm{exponentialCCDF}(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {1 - (1 - \\exp(-\\beta \\cdot t))}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {\\exp(-\\beta \\cdot t)}\n\\\\[4pt]\n& = & \\beta.\n\\end{eqnarray*}\\] The exponential distribution is the only distribution of survival times with a constant hazard function.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then its hazard function is \\[\\begin{eqnarray*}\nh(t \\mid \\alpha, \\sigma)\n& = & \\frac{p_T(t \\mid \\alpha, \\sigma)}{S(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = & \\frac{\\textrm{Weibull}(t \\mid \\alpha, \\sigma)}{1 - \\textrm{WeibullCCDF}(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = &\n\\frac{\\frac{\\alpha}{\\sigma} \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n      \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma} \\right)^\\alpha\\right)}\n     {1 - \\left(1 -\n       \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^\\alpha\n     \\right)\\right)}      \n\\\\[4pt]\n& = & \\frac{\\alpha}{\\sigma}\n  \\cdot\n  \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}.\n\\end{eqnarray*}\\]\nIf \\(\\alpha = 1\\) the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when \\(\\alpha = 1\\)). When \\(\\alpha &gt; 1,\\) the hazard grows as time passes, whereas when \\(\\alpha &lt; 1,\\) it decreases as time passes.\n\n\n\nThe exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. Cox (1972) introduced a semi-parametric survival model specified directly in terms of a hazard function \\(h(t)\\) rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.\nLet \\(x_n \\in \\mathbb{R}^K\\) be a (row) vector of covariates for subject \\(n\\) so that the full covariate data matrix is \\(x \\in \\mathbb{R}^{N \\times\nK}\\). In Cox’s model, the hazard function for subject \\(n\\) is defined conditionally in terms of their covariates \\(x_n\\) and the parameter vector \\(\\gamma \\in \\mathbb{R}^K\\) as \\[\\begin{equation*}\nh(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n \\cdot \\gamma),\n\\end{equation*}\\] where \\(h_0(t)\\) is a shared baseline hazard function and \\(x_n \\cdot\n\\gamma = \\sum_{k=1}^K x_{n, k} \\cdot \\beta_k\\) is a row vector-vector product.\nIn the semi-parametric, proportional hazards model, the baseline hazard function \\(h_0(t)\\) is not modeled. This is why it is called “semi-parametric.” Only the factor \\(\\exp(x_n \\cdot \\gamma),\\) which determines how individual \\(n\\) varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”\nCox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function \\(h_0(t)\\) is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function1.\n\n\nCox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function \\(h_0(t)\\) is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.\nThe hazard function \\(h(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n\n\\cdot \\beta)\\) for subject \\(n\\) represents the instantaneous probability that subject \\(n\\) fails at time \\(t\\) given that it has survived until time \\(t.\\) The probability that subject \\(n\\) is the first to fail among \\(N\\) subjects is thus proportional to subject \\(n\\)’s hazard function, \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail at time } t]\n\\propto h(t \\mid x_n, \\beta).\n\\end{equation*}\\] Normalizing yields \\[\\begin{eqnarray*}\n\\Pr[n \\textrm{ first to fail at time } t]\n& = &  \\frac{h(t \\mid x_n, \\beta)}\n            {\\sum_{n' = 1}^N h(t \\mid x_{n'}, \\beta)}\n\\\\[4pt]\n& = &  \\frac{h_0(t) \\cdot \\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N h_0(t) \\cdot \\exp(x_{n'} \\cdot \\beta)}\n\\\\[4pt]\n& = &  \\frac{\\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{eqnarray*}\\]\nSuppose there are \\(N\\) subjects with strictly ordered survival times \\(t_1 &lt;\nt_2 &lt; \\cdots &lt; t_N\\) and covariate (row) vectors \\(x_1, \\ldots, x_N\\). Let \\(t^{\\textrm{cens}}\\) be the (right) censoring time and let \\(N^{\\textrm{obs}}\\) be the largest value of \\(n\\) such that \\(t_n \\leq\nt^{\\textrm{cens}}\\). This means \\(N^{\\textrm{obs}}\\) is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.\nWith failure times sorted in decreasing order, the partial likelihood for each observed subject \\(n \\in 1{:}N^{\\textrm{obs}}\\) can be expressed as \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail among } n, n + 1, \\ldots N]\n= \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.\nThe partial likelihood, defined in this form by Breslow (1975), is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). \\[\\begin{equation*}\n\\Pr[\\textrm{observed failures ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n= \\prod_{n = 1}^{N^{\\textrm{obs}}}\n  \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] On the log scale, \\[\\begin{eqnarray*}\n\\log \\Pr[\\textrm{obs.\\ fail ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n& = &\n\\sum_{n = 1}^{N^{\\textrm{obs}}}\n  \\log \\left(\n          \\frac{\\exp(x_n \\cdot \\beta)}\n               {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}\n       \\right)\n\\\\[4pt]\n& = & x_n \\cdot \\beta - \\log \\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)\n\\\\\n& = & x_n \\cdot \\beta - \\textrm{logSumExp}_{n' = n}^N \\ x_{n'} \\cdot \\beta,\n\\end{eqnarray*}\\] where \\[\\begin{equation*}\n\\textrm{logSumExp}_{n = a}^b \\ x_n\n= \\log \\sum_{n = a}^b \\exp(x_n)\n\\end{equation*}\\] is implemented so as to preserve numerical precision.\nThis likelihood follows the same approach to ranking as that developed by Plackett (1975) for estimating the probability of the order of the first few finishers in a horse race.\nA simple normal prior on the components of \\(\\beta\\) completes the model, \\[\\begin{equation*}\n\\beta \\sim \\textrm{normal}(0, 2).\n\\end{equation*}\\] This should be scaled based on knowledge of the predictors.\n\n\n\nTo simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.\ndata {\n  int&lt;lower=0&gt; K;          // num covariates\n\n  int&lt;lower=0&gt; N;          // num uncensored obs\n  vector[N] t;             // event time (non-strict decreasing)\n  matrix[N, K] x;          // covariates for uncensored obs\n\n  int N_c;                 // num censored obs\n  real&lt;lower=t[N]&gt; t_c;    // censoring time\n  matrix[N_c, K] x_c;      // covariates for censored obs\n}\nThe parameters are just the coefficients.\nparameters {\n  vector[K] beta;          // slopes (no intercept)\n}\nThe prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.\nmodel {\n  beta ~ normal(0, 2);\n  ...\nThe log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\nThese vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.\n  real log_denom = log_sum_exp(log_theta_c);\nThen, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.\n  for (n in 1:N) {\n    log_denom = log_sum_exp(log_denom, log_theta[n]);\n    target += log_theta[n] - log_denom;   // log likelihood\n  }\nThe running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.\ntransformed data {\n  for (n in 2:N) {\n    if (!(t[n] &lt; t[n - 1])) {\n      reject(\"times must be strictly decreasing, but found\"\n             \"!(t[\", n, \"] &lt; t[, \", (n - 1), \"])\");\n    }   \n  }\n}\n\n\n\nTechnically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all \\(N!\\) permutations for a set of \\(N\\) subjects with tied survival times is not tractable. As an alternative, Efron (1977) introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.\nIn the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.\nfunctions {\n  int num_unique_starts(vector t) {\n    if (size(t) == 0) return 0;\n    int us = 1;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) us += 1;\n    }\n    return us;\n  }\nThis is then used to compute the value J to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.\n  array[] int unique_starts(vector t, int J) {\n    array[J + 1] int starts;\n    if (J == 0) return starts;\n    starts[1] = 1;\n    int pos = 2;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) {\n    starts[pos] = n;\n    pos += 1;\n      }\n    }\n    starts[J + 1] = size(t) + 1;\n    return starts;\n  }\n}\nThe data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.\ntransformed data {\n  int&lt;lower=0&gt; J = num_unique_starts(t);\n  array[J + 1] int&lt;lower=0&gt; starts = unique_starts(t, J);\n}\nFor each unique survival time j in 1:J, the subjects indexed from starts[j] to starts[j + 1] - 1 (inclusive) share the same survival time. The number of elements with survival time j is thus (starts[j + 1] - 1) - starts[j] + 1, or just starts[j + 1] - starts[j].\nThe parameters and prior are also the same—just a vector beta of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables J and starts.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\n  real log_denom_lhs = log_sum_exp(log_theta_c);\n  for (j in 1:J) {\n    int start = starts[j];\n    int end = starts[j + 1] - 1;\n    int len = end - start + 1;\n    real log_len = log(len);\n    real numerator = sum(log_theta[start:end]);\n    log_denom_lhs = log_sum_exp(log_denom_lhs,\n                                log_sum_exp(log_theta[start:end]));\n    vector[len] diff;\n    for (ell in 1:len) {\n      diff[ell] = log_diff_exp(log_denom_lhs,\n                               log(ell - 1) - log_len\n                               + log_sum_exp(log_theta[start:end]));\n    }\n    target += numerator - sum(diff);\n  }\nThe special function log_diff_exp is defined as\n\\[\\begin{equation*}\n\\textrm{logDiffExp}(u, v) = \\log(\\exp(u) - \\exp(v)).\n\\end{equation*}\\]\nBecause of how J and starts are constructed, the length len will always be strictly positive so that the log is well defined.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html#exponential-survival-model",
+    "href": "stan-users-guide/survival.html#exponential-survival-model",
+    "title": "Survival Models",
+    "section": "",
+    "text": "The exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if \\(T \\sim \\textrm{exponential}(\\lambda)\\) for some rate \\(\\lambda &gt; 0,\\) then \\[\\begin{equation*}\n\\Pr[T &gt; t] = \\Pr[T &gt; t + t' \\mid T &gt; t'].\n\\end{equation*}\\] If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.\nThe exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable \\(T\\) is \\[\\begin{equation*}\n\\mathbb{E}[T \\mid \\lambda] = \\frac{1}{\\lambda}.\n\\end{equation*}\\] The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) \\(\\beta = 1 / \\lambda\\).\nThe data for a survival model consists of two components. First, there is a vector \\(t \\in (0, \\infty)^N\\) of \\(N\\) observed failure times. Second, there is a censoring time \\(t^{\\textrm{cens}}\\) such that failure times greater than \\(t^{\\textrm{cens}}\\) are not observed. The censoring time assumption imposes a constraint which requires \\(t_n &lt; t^{\\textrm{cens}}\\) for all \\(n \\in 1{:}N.\\) For the censored subjects, the only thing required in the model is their total count, \\(N^\\textrm{cens}\\) (their covariates are also required for models with covariates).\nThe model for the observed failure times is exponential, so that for \\(n \\in 1{:}N,\\) \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda).\n\\end{equation*}\\]\nThe model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of \\[\\begin{equation*}\n\\Pr[T &gt; t^{\\textrm{cens}}] = 1 - F_T(t^{\\textrm{cens}}),\n\\end{equation*}\\] where \\(F_T\\) is the cumulative distribution function (cdf) of survival time \\(T\\) (\\(F_X(x) = \\Pr[X \\leq x]\\) is standard notation for the cdf of a random variable \\(X\\)). The function \\(1 - F_T(t)\\) is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood \\[\\begin{eqnarray*}\np(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\prod_{n=1}^{N^{\\textrm{cens}}}\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)\n\\\\\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)^{N^{\\textrm{cens}}}.\n\\end{eqnarray*}\\]\nOn the log scale, that’s \\[\\begin{eqnarray*}\n\\log p(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\sum_{n=1}^N \\log \\textrm{exponential}(t_n \\mid \\lambda)\n\\\\\n& & { } + N^{\\textrm{cens}} \\cdot \\log \\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda).\n\\end{eqnarray*}\\]\nThe model can be completed with a standard lognormal prior on \\(\\lambda,\\) \\[\\begin{equation*}\n\\lambda \\sim \\textrm{lognormal}(0, 1),\n\\end{equation*}\\] which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed \\(\\textrm{lognormal}(0, 1)\\). In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.\n\n\nThe data for a simple survival analysis without covariates can be coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n}\nIn this program, N is the number of uncensored observations and t contains the times of the uncensored observations. There are a further N_cens items that are right censored at time t_cens. Right censoring means that if the time to failure is greater than\nt_cens, it is only observed that the part survived until time t_cens. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.\nThere is a single rate parameter, the inverse of which is the expected time to failure.\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nThe exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.\nmodel {\n  t ~ exponential(lambda);\n  target += N_cens * exponential_lccdf(t_cens | lambda);\n\n  lambda ~ lognormal(0, 1);\n}\nThe likelihood for rate lambda is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector t as a having an exponential distribution with rate lambda. This data model could have been written as\nfor (n in 1:N) {\n  t[n] ~ exponential(lambda);\n}\nThe log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate lambda. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,\nfor (n in 1:N)\n  target += exponential_lccdf(t_cens | lambda);\nto define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html#weibull-survival-model",
+    "href": "stan-users-guide/survival.html#weibull-survival-model",
+    "title": "Survival Models",
+    "section": "",
+    "text": "The Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.\nThe Weibull distribution is parameterized by a shape \\(\\alpha &gt; 0\\) and scale \\(\\sigma &gt; 0.\\) For an outcome \\(t \\geq 0\\), the Weibull distribution’s probability density function is \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid \\alpha, \\sigma)\n= \\frac{\\alpha}{\\sigma}\n  \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n  \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^{\\alpha}\\right).\n\\end{equation*}\\] In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter \\(\\beta &gt; 0\\) with probability density function \\[\\begin{equation*}\n\\textrm{exponential}(t \\mid \\beta) =\n\\beta\n\\cdot\n\\exp(-\\beta \\cdot t).\n\\end{equation*}\\] When \\(\\alpha = 1,\\) the Weibull distribution reduces to an exponential distribution, \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid 1, \\sigma)\n=\n\\textrm{exponential}\\!\\left(t \\,\\bigg|\\, \\frac{1}{\\sigma}\\right).\n\\end{equation*}\\] In other words, the Weibull is a continuous expansion of the exponential distribution.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then the expected survival time is \\[\\begin{equation*}\n\\mathbb{E}[T] = \\sigma \\cdot \\Gamma\\!\\left(1 + \\frac{1}{\\alpha}\\right),\n\\end{equation*}\\] where the \\(\\Gamma\\) function is the continuous completion of the factorial function (i.e., \\(\\Gamma(1 + n) = n!\\ \\) for \\(n \\in\n\\mathbb{N}\\)). As \\(\\alpha \\rightarrow 0\\) for a fixed \\(\\sigma\\) or as \\(\\sigma \\rightarrow \\infty\\) for a fixed \\(\\alpha\\), the expected survival time goes to infinity.\nThere are three regimes of the Weibull distribution.\n\n\\(\\alpha &lt; 1.\\) A subject is more likely to fail early. When \\(\\alpha\n&lt; 1,\\) the Weibull density approaches infinity as \\(t \\rightarrow 0.\\)\n\\(\\alpha = 1.\\) The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When \\(\\alpha = 1,\\) the Weibull distribution approaches \\(\\sigma\\) as \\(t\n\\rightarrow 0.\\)\n\\(\\alpha &gt; 1.\\) Subjects are less likely to fail early. When \\(\\alpha &lt; 1,\\) the Weibull density approaches zero as \\(t \\rightarrow 0.\\)\n\nWith \\(\\alpha \\leq 1,\\) the mode is zero (\\(t = 0\\)), whereas with \\(\\alpha &gt; 1,\\) the mode is nonzero (\\(t &gt; 0\\)).\n\n\nWith Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.\nparameters {\n  real&lt;lower=0&gt; beta;\n}\nmodel {\n  t ~ exponential(beta);\n  target += N_cens * exponential_lccdf(t_cens | beta);\n\n  beta ~ lognormal(0, 1);\n}\nThe Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (alpha) and scale (sigma) parameters.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  t ~ weibull(alpha, sigma);\n  target += N_cens * weibull_lccdf(t_cens | alpha, sigma);\n\n  alpha ~ lognormal(0, 1);\n  sigma ~ lognormal(0, 1);\n}\nAs usual, if more is known about expected survival times, alpha and sigma should be given more informative priors.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html#survival-with-covariates",
+    "href": "stan-users-guide/survival.html#survival-with-covariates",
+    "title": "Survival Models",
+    "section": "",
+    "text": "Suppose that for each of \\(n \\in 1{:}N\\) items observed, both censored and uncensored, there is a covariate (row) vector \\(x_n \\in\n\\mathbb{R}^K.\\) For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.\nSurvival with covariates replaces what is essentially a simple regression with only an intercept \\(\\lambda\\) with a generalized linear model with a log link, where the rate for item \\(n\\) is \\[\\begin{equation*}\n\\lambda_n = \\exp(x_n \\cdot \\beta),\n\\end{equation*}\\] where \\(\\beta \\in \\mathbb{R}^K\\) is a \\(K\\)-vector of regression coefficients. Thus \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda_n).\n\\end{equation*}\\] The censored items have probability \\[\\begin{equation*}\n\\Pr[n\\textrm{-th censored}] =\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid x^{\\textrm{cens}}_n\n\\cdot \\beta).\n\\end{equation*}\\]\nThe covariates form an \\(N \\times K\\) data matrix, \\(x \\in\n\\mathbb{R}^{N \\times K}\\). An intercept can be introduced by adding a column of 1 values to \\(x\\).\nA Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (t and t_cens) corresponds to the ordering of items in the covariate matrices (x and x_cens).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n  int&lt;lower=0&gt; K;\n  matrix[N, K] x;\n  matrix[N_cens, K] x_cens;\n}\nparameters {\n  vector[K] gamma;\n}\nmodel {\n  gamma ~ normal(0, 2);\n\n  t ~ exponential(exp(x * gamma));\n  target += exponential_lccdf(t_cens | exp(x_cens * gamma));\n}\nBoth the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html#hazard-and-survival-functions",
+    "href": "stan-users-guide/survival.html#hazard-and-survival-functions",
+    "title": "Survival Models",
+    "section": "",
+    "text": "Suppose \\(T\\) is a random variable representing a survival time, with a smooth cumulative distribution function \\[\\begin{equation*}\nF_T(t) = \\Pr[T \\leq t],\n\\end{equation*}\\] so that its probability density function is \\[\\begin{equation*}\np_T(t) = \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t).\n\\end{equation*}\\]\nThe survival function \\(S(t)\\) is the probability of surviving until at least time \\(t\\), which is just the complementary cumulative distribution function (ccdf) of the survival random variable \\(T\\), \\[\\begin{equation*}\nS(t) = 1 - F_T(t).\n\\end{equation*}\\] The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).\nThe hazard function \\(h(t)\\) is the instantaneous risk of not surviving past time \\(t\\) assuming survival until time \\(t\\), which is given by \\[\\begin{equation*}\nh(t) = \\frac{p_T(t)}{S(t)} = \\frac{p_T(t)}{1 - F_T(t)}.\n\\end{equation*}\\] The cumulative hazard function \\(H(t)\\) is defined to be the accumulated hazard over time, \\[\\begin{equation*}\nH(t) = \\int_0^t h(u) \\, \\textrm{d}u.\n\\end{equation*}\\]\nThe hazard function and survival function are related through the differential equation \\[\\begin{eqnarray*}\nh(t) & = & -\\frac{\\textrm{d}}{\\textrm{d}t} \\log S(t).\n\\\\[4pt]\n& = & -\\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} S(t)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} -(1 - F_T(t))\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} (F_T(t) - 1)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t)\n\\\\[4pt]\n& = & \\frac{p_T(t)}{S(t)}.\n\\end{eqnarray*}\\]\nIf \\(T \\sim \\textrm{exponential}(\\beta)\\) has an exponential distribution, then its hazard function is constant, \\[\\begin{eqnarray*}\nh(t \\mid \\beta)\n& = & \\frac{p_T(t \\mid \\beta)}{S(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\textrm{exponential}(t \\mid \\beta)}{1 - \\textrm{exponentialCCDF}(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {1 - (1 - \\exp(-\\beta \\cdot t))}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {\\exp(-\\beta \\cdot t)}\n\\\\[4pt]\n& = & \\beta.\n\\end{eqnarray*}\\] The exponential distribution is the only distribution of survival times with a constant hazard function.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then its hazard function is \\[\\begin{eqnarray*}\nh(t \\mid \\alpha, \\sigma)\n& = & \\frac{p_T(t \\mid \\alpha, \\sigma)}{S(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = & \\frac{\\textrm{Weibull}(t \\mid \\alpha, \\sigma)}{1 - \\textrm{WeibullCCDF}(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = &\n\\frac{\\frac{\\alpha}{\\sigma} \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n      \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma} \\right)^\\alpha\\right)}\n     {1 - \\left(1 -\n       \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^\\alpha\n     \\right)\\right)}      \n\\\\[4pt]\n& = & \\frac{\\alpha}{\\sigma}\n  \\cdot\n  \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}.\n\\end{eqnarray*}\\]\nIf \\(\\alpha = 1\\) the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when \\(\\alpha = 1\\)). When \\(\\alpha &gt; 1,\\) the hazard grows as time passes, whereas when \\(\\alpha &lt; 1,\\) it decreases as time passes.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html#proportional-hazards-model",
+    "href": "stan-users-guide/survival.html#proportional-hazards-model",
+    "title": "Survival Models",
+    "section": "",
+    "text": "The exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. Cox (1972) introduced a semi-parametric survival model specified directly in terms of a hazard function \\(h(t)\\) rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.\nLet \\(x_n \\in \\mathbb{R}^K\\) be a (row) vector of covariates for subject \\(n\\) so that the full covariate data matrix is \\(x \\in \\mathbb{R}^{N \\times\nK}\\). In Cox’s model, the hazard function for subject \\(n\\) is defined conditionally in terms of their covariates \\(x_n\\) and the parameter vector \\(\\gamma \\in \\mathbb{R}^K\\) as \\[\\begin{equation*}\nh(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n \\cdot \\gamma),\n\\end{equation*}\\] where \\(h_0(t)\\) is a shared baseline hazard function and \\(x_n \\cdot\n\\gamma = \\sum_{k=1}^K x_{n, k} \\cdot \\beta_k\\) is a row vector-vector product.\nIn the semi-parametric, proportional hazards model, the baseline hazard function \\(h_0(t)\\) is not modeled. This is why it is called “semi-parametric.” Only the factor \\(\\exp(x_n \\cdot \\gamma),\\) which determines how individual \\(n\\) varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”\nCox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function \\(h_0(t)\\) is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function1.\n\n\nCox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function \\(h_0(t)\\) is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.\nThe hazard function \\(h(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n\n\\cdot \\beta)\\) for subject \\(n\\) represents the instantaneous probability that subject \\(n\\) fails at time \\(t\\) given that it has survived until time \\(t.\\) The probability that subject \\(n\\) is the first to fail among \\(N\\) subjects is thus proportional to subject \\(n\\)’s hazard function, \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail at time } t]\n\\propto h(t \\mid x_n, \\beta).\n\\end{equation*}\\] Normalizing yields \\[\\begin{eqnarray*}\n\\Pr[n \\textrm{ first to fail at time } t]\n& = &  \\frac{h(t \\mid x_n, \\beta)}\n            {\\sum_{n' = 1}^N h(t \\mid x_{n'}, \\beta)}\n\\\\[4pt]\n& = &  \\frac{h_0(t) \\cdot \\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N h_0(t) \\cdot \\exp(x_{n'} \\cdot \\beta)}\n\\\\[4pt]\n& = &  \\frac{\\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{eqnarray*}\\]\nSuppose there are \\(N\\) subjects with strictly ordered survival times \\(t_1 &lt;\nt_2 &lt; \\cdots &lt; t_N\\) and covariate (row) vectors \\(x_1, \\ldots, x_N\\). Let \\(t^{\\textrm{cens}}\\) be the (right) censoring time and let \\(N^{\\textrm{obs}}\\) be the largest value of \\(n\\) such that \\(t_n \\leq\nt^{\\textrm{cens}}\\). This means \\(N^{\\textrm{obs}}\\) is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.\nWith failure times sorted in decreasing order, the partial likelihood for each observed subject \\(n \\in 1{:}N^{\\textrm{obs}}\\) can be expressed as \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail among } n, n + 1, \\ldots N]\n= \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.\nThe partial likelihood, defined in this form by Breslow (1975), is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). \\[\\begin{equation*}\n\\Pr[\\textrm{observed failures ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n= \\prod_{n = 1}^{N^{\\textrm{obs}}}\n  \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] On the log scale, \\[\\begin{eqnarray*}\n\\log \\Pr[\\textrm{obs.\\ fail ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n& = &\n\\sum_{n = 1}^{N^{\\textrm{obs}}}\n  \\log \\left(\n          \\frac{\\exp(x_n \\cdot \\beta)}\n               {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}\n       \\right)\n\\\\[4pt]\n& = & x_n \\cdot \\beta - \\log \\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)\n\\\\\n& = & x_n \\cdot \\beta - \\textrm{logSumExp}_{n' = n}^N \\ x_{n'} \\cdot \\beta,\n\\end{eqnarray*}\\] where \\[\\begin{equation*}\n\\textrm{logSumExp}_{n = a}^b \\ x_n\n= \\log \\sum_{n = a}^b \\exp(x_n)\n\\end{equation*}\\] is implemented so as to preserve numerical precision.\nThis likelihood follows the same approach to ranking as that developed by Plackett (1975) for estimating the probability of the order of the first few finishers in a horse race.\nA simple normal prior on the components of \\(\\beta\\) completes the model, \\[\\begin{equation*}\n\\beta \\sim \\textrm{normal}(0, 2).\n\\end{equation*}\\] This should be scaled based on knowledge of the predictors.\n\n\n\nTo simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.\ndata {\n  int&lt;lower=0&gt; K;          // num covariates\n\n  int&lt;lower=0&gt; N;          // num uncensored obs\n  vector[N] t;             // event time (non-strict decreasing)\n  matrix[N, K] x;          // covariates for uncensored obs\n\n  int N_c;                 // num censored obs\n  real&lt;lower=t[N]&gt; t_c;    // censoring time\n  matrix[N_c, K] x_c;      // covariates for censored obs\n}\nThe parameters are just the coefficients.\nparameters {\n  vector[K] beta;          // slopes (no intercept)\n}\nThe prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.\nmodel {\n  beta ~ normal(0, 2);\n  ...\nThe log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\nThese vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.\n  real log_denom = log_sum_exp(log_theta_c);\nThen, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.\n  for (n in 1:N) {\n    log_denom = log_sum_exp(log_denom, log_theta[n]);\n    target += log_theta[n] - log_denom;   // log likelihood\n  }\nThe running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.\ntransformed data {\n  for (n in 2:N) {\n    if (!(t[n] &lt; t[n - 1])) {\n      reject(\"times must be strictly decreasing, but found\"\n             \"!(t[\", n, \"] &lt; t[, \", (n - 1), \"])\");\n    }   \n  }\n}\n\n\n\nTechnically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all \\(N!\\) permutations for a set of \\(N\\) subjects with tied survival times is not tractable. As an alternative, Efron (1977) introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.\nIn the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.\nfunctions {\n  int num_unique_starts(vector t) {\n    if (size(t) == 0) return 0;\n    int us = 1;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) us += 1;\n    }\n    return us;\n  }\nThis is then used to compute the value J to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.\n  array[] int unique_starts(vector t, int J) {\n    array[J + 1] int starts;\n    if (J == 0) return starts;\n    starts[1] = 1;\n    int pos = 2;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) {\n    starts[pos] = n;\n    pos += 1;\n      }\n    }\n    starts[J + 1] = size(t) + 1;\n    return starts;\n  }\n}\nThe data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.\ntransformed data {\n  int&lt;lower=0&gt; J = num_unique_starts(t);\n  array[J + 1] int&lt;lower=0&gt; starts = unique_starts(t, J);\n}\nFor each unique survival time j in 1:J, the subjects indexed from starts[j] to starts[j + 1] - 1 (inclusive) share the same survival time. The number of elements with survival time j is thus (starts[j + 1] - 1) - starts[j] + 1, or just starts[j + 1] - starts[j].\nThe parameters and prior are also the same—just a vector beta of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables J and starts.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\n  real log_denom_lhs = log_sum_exp(log_theta_c);\n  for (j in 1:J) {\n    int start = starts[j];\n    int end = starts[j + 1] - 1;\n    int len = end - start + 1;\n    real log_len = log(len);\n    real numerator = sum(log_theta[start:end]);\n    log_denom_lhs = log_sum_exp(log_denom_lhs,\n                                log_sum_exp(log_theta[start:end]));\n    vector[len] diff;\n    for (ell in 1:len) {\n      diff[ell] = log_diff_exp(log_denom_lhs,\n                               log(ell - 1) - log_len\n                               + log_sum_exp(log_theta[start:end]));\n    }\n    target += numerator - sum(diff);\n  }\nThe special function log_diff_exp is defined as\n\\[\\begin{equation*}\n\\textrm{logDiffExp}(u, v) = \\log(\\exp(u) - \\exp(v)).\n\\end{equation*}\\]\nBecause of how J and starts are constructed, the length len will always be strictly positive so that the log is well defined.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/survival.html#footnotes",
+    "href": "stan-users-guide/survival.html#footnotes",
+    "title": "Survival Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nCox mentioned in his seminal paper that modeling the baseline hazard function would improve statistical efficiency, but he did not do it for computational reasons.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Survival Models"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/truncation-censoring.html",
+    "href": "stan-users-guide/truncation-censoring.html",
+    "title": "Truncated or Censored Data",
+    "section": "",
+    "text": "Data in which measurements have been truncated or censored can be coded in Stan following their respective probability models.\n\n\nTruncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.\n\n\n\nTruncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.\nTruncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are \\(y_n\\) with an upper truncation point of \\(U = 300\\) so that \\(y_n &lt; 300\\). In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  real U;\n  array[N] real&lt;upper=U&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma) T[ , U];\n}\nThe model declares an upper bound U as data and constrains the data for y to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.\nSee the Stan Reference Manual’s Statements chapter for how to use truncated distributions with the log probability increment statements (target += ...).\nThis model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.\n\n\nIf the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to \\(-\\infty\\). For instance, if variate y is sampled with the statement.\ny ~ normal(mu, sigma) T[L, U];\nthen if any value inside y is less than the value of L or greater than the value of U, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.\nTo avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if y is a parameter in the above model, the declaration should constrain it to fall between the values of L and U.\nparameters {\n  array[N] real&lt;lower=L, upper=U&gt; y;\n  // ...\n}\nIf in the above model, L or U is a parameter and y is data, then L and U must be appropriately constrained so that all data are in range and the value of L is less than that of U (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.\nparameters {\n  real&lt;upper=min(y)&gt; L;           // L &lt; y[n]\n  real&lt;lower=fmax(L, max(y))&gt; U;  // L &lt; U; y[n] &lt; U\nFor pairs of real numbers, the function fmax is used rather than max.\n\n\n\nIf the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real y;\n}\nparameters {\n  real&lt;upper=min(y)&gt; L;\n  real&lt;lower=max(y)&gt; U;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ // ...\n  U ~ // ...\n  y ~ normal(mu, sigma) T[L, U];\n}\nHere there is a lower truncation point L which is declared to be less than or equal to the minimum value of y. The upper truncation point U is declared to be larger than the maximum value of y. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With N declared as type int&lt;lower=1&gt;, there must be at least one data point. The constraint that L is less than U is enforced indirectly, based on the non-empty data.\nThe ellipses where the priors for the bounds L and U should go should be filled in with a an informative prior in order for this model to not concentrate L strongly around min(y) and U strongly around max(y).\n\n\n\n\nCensoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.\n\n\nOne way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  array[N_cens] real&lt;lower=U&gt; y_cens;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_cens ~ normal(mu, sigma);\n}\nBecause the censored data array y_cens is declared to be a parameter, it will be sampled along with the location and scale parameters mu and sigma. Because the censored data array y_cens is declared to have values of type real&lt;lower=U&gt;, all imputed values for censored data will be greater than U. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.\n\n\n\nAlthough it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of \\[\\begin{align*}\n\\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\int_U^{\\infty} \\textsf{normal}\\left(y_{\\mathrm{cens},m} \\mid \\mu,\\sigma \\right) \\,\\textsf{d}y_{\\mathrm{cens},m} \\\\\n  &= 1 - \\Phi\\left(\\frac{U - \\mu}{\\sigma}\\right),\n\\end{align*}\\]\nwhere \\(\\Phi()\\) is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that \\(y_{\\mathrm{cens},m}&gt;U\\). With \\(M\\) censored observations, the likelihood on the log scale is \\[\\begin{align*}\n\\log \\prod_{m=1}^M \\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\log \\left( 1 - \\Phi\\left(\\left(\\frac{U - \\mu}{\\sigma}\\right)\\right)^{M}\\right) \\\\\n  &= M \\times \\texttt{normal}\\mathtt{\\_}\\texttt{lccdf}\\left(U \\mid \\mu, \\sigma \\right),\n\\end{align*}\\]\nwhere normal_lccdf is the log of complementary CDF (Stan provides &lt;distr&gt;_lccdf for each distribution implemented in Stan).\nThe following right-censored model assumes that the censoring point is known, so it is declared as data.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lccdf(U | mu, sigma);\n}\nFor the observed values in y_obs, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.\nFor the left-censored data the CDF (normal_lcdf) has to be used instead of complementary CDF. If the censoring point variable (L) is unknown, its declaration should be moved from the data to the parameters block.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real&lt;upper=min(y_obs)&gt; L;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ normal(mu, sigma);\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lcdf(L | mu, sigma);\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Truncated or Censored Data"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/truncation-censoring.html#truncation.section",
+    "href": "stan-users-guide/truncation-censoring.html#truncation.section",
+    "title": "Truncated or Censored Data",
+    "section": "",
+    "text": "Truncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Truncated or Censored Data"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/truncation-censoring.html#truncated-data.section",
+    "href": "stan-users-guide/truncation-censoring.html#truncated-data.section",
+    "title": "Truncated or Censored Data",
+    "section": "",
+    "text": "Truncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.\nTruncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are \\(y_n\\) with an upper truncation point of \\(U = 300\\) so that \\(y_n &lt; 300\\). In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  real U;\n  array[N] real&lt;upper=U&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma) T[ , U];\n}\nThe model declares an upper bound U as data and constrains the data for y to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.\nSee the Stan Reference Manual’s Statements chapter for how to use truncated distributions with the log probability increment statements (target += ...).\nThis model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.\n\n\nIf the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to \\(-\\infty\\). For instance, if variate y is sampled with the statement.\ny ~ normal(mu, sigma) T[L, U];\nthen if any value inside y is less than the value of L or greater than the value of U, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.\nTo avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if y is a parameter in the above model, the declaration should constrain it to fall between the values of L and U.\nparameters {\n  array[N] real&lt;lower=L, upper=U&gt; y;\n  // ...\n}\nIf in the above model, L or U is a parameter and y is data, then L and U must be appropriately constrained so that all data are in range and the value of L is less than that of U (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.\nparameters {\n  real&lt;upper=min(y)&gt; L;           // L &lt; y[n]\n  real&lt;lower=fmax(L, max(y))&gt; U;  // L &lt; U; y[n] &lt; U\nFor pairs of real numbers, the function fmax is used rather than max.\n\n\n\nIf the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real y;\n}\nparameters {\n  real&lt;upper=min(y)&gt; L;\n  real&lt;lower=max(y)&gt; U;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ // ...\n  U ~ // ...\n  y ~ normal(mu, sigma) T[L, U];\n}\nHere there is a lower truncation point L which is declared to be less than or equal to the minimum value of y. The upper truncation point U is declared to be larger than the maximum value of y. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With N declared as type int&lt;lower=1&gt;, there must be at least one data point. The constraint that L is less than U is enforced indirectly, based on the non-empty data.\nThe ellipses where the priors for the bounds L and U should go should be filled in with a an informative prior in order for this model to not concentrate L strongly around min(y) and U strongly around max(y).",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Truncated or Censored Data"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/truncation-censoring.html#censored.section",
+    "href": "stan-users-guide/truncation-censoring.html#censored.section",
+    "title": "Truncated or Censored Data",
+    "section": "",
+    "text": "Censoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.\n\n\nOne way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  array[N_cens] real&lt;lower=U&gt; y_cens;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_cens ~ normal(mu, sigma);\n}\nBecause the censored data array y_cens is declared to be a parameter, it will be sampled along with the location and scale parameters mu and sigma. Because the censored data array y_cens is declared to have values of type real&lt;lower=U&gt;, all imputed values for censored data will be greater than U. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.\n\n\n\nAlthough it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of \\[\\begin{align*}\n\\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\int_U^{\\infty} \\textsf{normal}\\left(y_{\\mathrm{cens},m} \\mid \\mu,\\sigma \\right) \\,\\textsf{d}y_{\\mathrm{cens},m} \\\\\n  &= 1 - \\Phi\\left(\\frac{U - \\mu}{\\sigma}\\right),\n\\end{align*}\\]\nwhere \\(\\Phi()\\) is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that \\(y_{\\mathrm{cens},m}&gt;U\\). With \\(M\\) censored observations, the likelihood on the log scale is \\[\\begin{align*}\n\\log \\prod_{m=1}^M \\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\log \\left( 1 - \\Phi\\left(\\left(\\frac{U - \\mu}{\\sigma}\\right)\\right)^{M}\\right) \\\\\n  &= M \\times \\texttt{normal}\\mathtt{\\_}\\texttt{lccdf}\\left(U \\mid \\mu, \\sigma \\right),\n\\end{align*}\\]\nwhere normal_lccdf is the log of complementary CDF (Stan provides &lt;distr&gt;_lccdf for each distribution implemented in Stan).\nThe following right-censored model assumes that the censoring point is known, so it is declared as data.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lccdf(U | mu, sigma);\n}\nFor the observed values in y_obs, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.\nFor the left-censored data the CDF (normal_lcdf) has to be used instead of complementary CDF. If the censoring point variable (L) is unknown, its declaration should be moved from the data to the parameters block.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real&lt;upper=min(y_obs)&gt; L;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ normal(mu, sigma);\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lcdf(L | mu, sigma);\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Truncated or Censored Data"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/using-stanc.html",
+    "href": "stan-users-guide/using-stanc.html",
+    "title": "Using the Stan Compiler",
+    "section": "",
+    "text": "Stan is used in most of our interfaces through the Stan compiler stanc. Since version 2.22, the Stan compiler has been implemented in OCaml and is referred to as stanc3. The binary name is still simply stanc, so this document uses both stanc and stanc3 interchangeably.\n\n\nThe stanc3 compiler has the following command-line syntax:\n&gt; stanc (options) &lt;model_file&gt;\nwhere &lt;model_file&gt; is either a path to a file ending in .stan or .stanfunctions (which automatically sets --standalone-functions), or '-' to read from standard input.\nThe stanc3 options are:\n\n--help / -? - Displays the complete list of stanc3 options, then exits.\n--version - Display stanc version number\n--info - Print information about the model, such as the type information for variables and the list of used distributions.\n--name=&lt;model_name&gt; - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.\n--o=&lt;file_name&gt; / -o=&lt;filename&gt; / --output=&lt;filename&gt; - Specify a path to an output file for generated C++ code (default = .hpp) or auto-formatting output (default: no file/print to stdout)\n--auto-format - Pretty prints the program to the console. See more on auto formatting.\n--allow-undefined - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.\n--canonicalize - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.\n--include_paths=&lt;dir1,...dirN&gt; - Takes a comma-separated list of directories that may contain a file in an #include directive.\n--max-line-length=&lt;number&gt; - Set the column number at which formatting with --auto-format attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.\n--print-canonical - Synonymous with --auto-format --canonicalize=deprecations,includes,parentheses,braces.\n--print-cpp - If set, output the generated C++ Stan model class to stdout.\n--filename-in-msg=&lt;name&gt; - Sets the filename used in compiler and runtime errors. If absent, the &lt;model_file&gt; argument is used.\n--standalone-functions - If set, only generate the code for the functions defined in the file. This is the default behavior for .stanfunctions files.\n--O0 (Default) Do not apply optimizations to the Stan code.\n--O1 Apply level 1 compiler optimizations (only basic optimizations).\n--Oexperimental WARNING: This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance! Allow the compiler to apply all optimizations to the Stan code.\n--O Synonym for --O1 as of Stan 2.37. In earlier versions this was a synonym for --Oexperimental.\n--use-opencl - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.\n--warn-pedantic - Emit warnings in Pedantic mode which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.\n--warn-uninitialized - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.\n--color - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are auto (the default), always, never. Can also be controlled by the STANC_COLOR environment variable.\n\nThe compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the --help option to see the full set.\n\n\n\nDuring model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).\n\n\nEven without the optional --warn-pedantic and --warn-uninitialized command line flags, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations\n\nA completely blank Stan program will produce the following warning message\nWarning in 'empty.stan', line 1, column 0 to line 2, column 0:\n    Empty model detected; this is a valid Stan model but likely unintended!\nThe use of any deprecated features will lead to warnings which will look as follows\n Warning in 'deprecated.stan', line 2, column 10 to column 17:\n     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with\n     an independent lognormal distribution on the scales, see:\n     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution\nA single Stan program can produce many warnings during compilation.\n\n\n\n\nErrors differ from warnings in their severity and format. In particular, errors are fatal and stop compilation, so at most one error is displayed per run of stanc.\nThere are five kinds of errors emitted by stanc3\n\nFile errors occur when the file passed to stanc is either missing or cannot be opened (i.e. has permissions issues). They look like\nError: file 'notfound.stan' not found or cannot be opened\nSyntactic errors occur whenever a program violates the Stan language’s syntax requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the #include directive fails, and “parsing” errors which result when the structure of the program is incorrect.\n\nThe lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of $ in a variable name will look like the following.\nSyntax error in 'char.stan', line 2, column 7 to column 8, lexing error:\n-------------------------------------------------\n  1:  data {\n  2:     int $ome_variable;\n             ^\n  3:  }\n-------------------------------------------------\nInvalid character found.\nWhen an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).\nSyntax error in './incl.stan', line 1, column 0, included from\n'./incl.stan', line 1, column 0, included from\n'incl.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include &lt;incl.stan&gt;\n         ^\n   -------------------------------------------------\nFile incl.stan recursively included itself.\nIt is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.\nSyntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:\n   -------------------------------------------------\n     1:  data {\n     2:     int&lt;lower=0&gt; N;\n     3:     vector x;\n                   ^\n     4:  }\n   -------------------------------------------------\nIll-formed type. Expected \"[\" expression \"]\" for vector size.\n\nSemantic errors (also known as type errors) occur when a program is structured correctly but features an error in the type rules imposed by the language. An example of this is assigning a real value to a variable defined as an integer.\nSemantic error in 'type.stan', line 2, column 3 to column 15:\n   -------------------------------------------------\n     1:  transformed data {\n     2:     int x = 1.5;\n            ^\n     3:  }\n   -------------------------------------------------\nIll-typed arguments supplied to assignment operator =:\nThe left hand side has type\n  int\nand the right hand side has type\n  real\nThe compiler will raise an error for use of any removed features for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.\nFinally, the compiler can raise an internal error. These are caused by bugs in the compiler, not your model, and we would appreciate it if you report them on the stanc3 repo with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.\n\n\n\n\n\nPedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.\nFor example, consider the following program.\ndata {\n  int N;\n  array[N] real x;\n}\nparameters {\n  real sigma;\n}\nmodel {\n  real mu;\n  x ~ normal(mu, sigma);\n}\nWhen pedantic mode is turned on, the compiler will produce the following warnings.\nWarning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:\n    The parameter sigma has no priors. This means either no prior is\n    provided, or the prior(s) depend on data variables. In the later case,\n    this may be a false positive.\nWarning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:\n    The variable mu may not have been assigned a value before its use.\nWarning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:\n    A normal distribution is given parameter sigma as a scale parameter\n    (argument 2), but sigma was not constrained to be strictly positive.\nHere are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):\n\nDistribution usages issues. Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.\nUnused parameter. A parameter is defined but doesn’t contribute to target.\nLarge or small constant in a distribution. Very large or very small constants are used as distribution arguments.\nControl flow depends on a parameter. Branching control flow (like if/else) depends on a parameter value .\nParameter has multiple tildes. A parameter is on the left-hand side of multiple tildes.\nParameter has zero or multiple priors. A parameter has zero or more than one prior distribution.\nVariable is used before assignment. A variable is used before being assigned a value.\nStrict or nonsensical parameter bounds. A parameter is given questionable bounds.\nNonlinear transformations. When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.\n\nSome important limitations of pedantic mode are listed at the end of this chapter.\n\n\nWhen an argument to a built-in distribution certainly does not match that distribution’s specification in the Stan Functions Reference, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g. simplex) is incompatible with the domain of the distribution. x\nFor example, consider the following program.\nparameters {\n  real unb_p;\n  real&lt;lower=0&gt; pos_p;\n}\nmodel {\n  1 ~ poisson(unb_p);\n  1 ~ poisson(pos_p);\n}\nThe parameter of poisson should be strictly positive, but unb_p is not constrained to be positive.\nPedantic mode produces the following warning.\nWarning in 'ex-dist-args.stan', line 6, column 14 to column 19:\n  A poisson distribution is given parameter unb_p as a rate parameter\n  (argument 1), but unb_p was not constrained to be strictly positive.\n\n\n\nPedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:\n\n\nAny use of uniform distribution generates a warning, except when the variate parameter’s declared upper and lower bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.\nFor example, consider the following program.\nparameters {\n  real a;\n  real&lt;lower=0, upper=1&gt; b;\n}\nmodel {\n  a ~ uniform(0, 1);\n  b ~ uniform(0, 1);\n}\na is assigned a uniform distribution that doesn’t match its constraints.\nPedantic mode produces the following warning.\nWarning in 'uniform-warn.stan', line 6, column 2 to column 20:\n  Parameter a is given a uniform distribution. The uniform distribution is\n  not recommended, for two reasons: (a) Except when there are logical or\n  physical constraints, it is very unusual for you to be sure that a\n  parameter will fall inside a specified range, and (b) The infinite gradient\n  induced by a uniform density can cause difficulties for Stan's sampling\n  algorithm. As a consequence, we recommend soft constraints rather than hard\n  constraints; for example, instead of giving an elasticity parameter a\n  uniform(0, 1) distribution, try normal(0.5, 0.5).\n\n\n\nGamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.\n\n\n\nAny use of the lkj_corr distribution generates a warning that suggests using the Cholesky variant instead. See the LKJ correlation distribution section of the Stan Functions Reference for details.\n\n\n\n\nA warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the target variable depends in any way on each of the parameters.\nFor example, consider the following program.\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(1, 1);\n}\na participates in the density function but b does not.\nPedantic mode produces the following warning.\nWarning in 'unused.stan', line 3, column 2 to column 9:\n    The parameter b was declared but was not used in the density calculation.\n\n\n\nWhen numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See the efficiency tuning section of the Stan User’s guide for a discussion of scaling parameters.\nFor example, consider the following program.\nparameters {\n  real x;\n  real y;\n}\nmodel {\n  x ~ normal(-100, 100);\n  y ~ normal(0, 1);\n}\nThe constants -100 and 100 suggest that x is not unit scaled.\nPedantic mode produces the following warning.\nWarning in 'constants-warn.stan', line 6, column 13 to column 17:\n    Argument -100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\nWarning in 'constants-warn.stan', line 6, column 19 to column 22:\n    Argument 100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\n\n\n\nControl flow statements, such as if, for and while should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.\nFor example, consider the following program.\nparameters {\n  real a;\n}\nmodel {\n  // x depends on parameter a\n  real x = a * a;\n\n  int m;\n\n  // the if-then-else depends on x which depends on a\n  if(x &gt; 0) {\n    //now m depends on x which depends on a\n    m = 1;\n  } else {\n    m = 2;\n  }\n\n  // for loop depends on m -&gt; x -&gt; a\n  for (i in 0:m) {\n    a ~ normal(i, 1);\n  }\n}\nThe if and for statements are control flow that depend (indirectly) on the value of the parameter m.\nPedantic mode produces the following warning.\nWarning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:\n  A control flow statement depends on parameter(s): a.\nWarning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:\n  A control flow statement depends on parameter(s): a.\n\n\n\nA warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent target += conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.\nPedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(0, 1);\n  a ~ normal(x, 1);\n\n  b ~ normal(1, 1);\n}\nPedantic mode produces the following warning.\nWarning in 'multi-tildes.stan', line 9, column 2 to column 19:\n  The parameter a is on the left-hand side of more than one tildes\n  statement.\n\n\n\nA warning is generated when a parameter appears to have greater than or less than one prior distribution factor.\nThis analysis depends on a factor graph representation of a Stan program. A factor F that depends on a parameter P is called a prior factor for P if there is no path in the factor graph from F to any data variable except through P.\nOne limitation of this approach is that the compiler cannot distinguish between modeled data variables and other convenient uses of data variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for int variables) are modeled data, which may cause extra warnings.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n  real c;\n  real d;\n}\nmodel\n{\n  a ~ normal(0, 1); // this is a prior\n  x ~ normal(a, 1); // this is not a prior, since data is involved\n\n  b ~ normal(x, 1); // this is also not a prior, since data is involved\n\n  // this is not a prior for c, since data is involved through b\n  // but it is a prior for b, since the data is only involved through b\n  c ~ normal(b, 1);\n\n  //these are multiple priors:\n  d ~ normal(0, 1);\n  1 ~ normal(d, 1);\n}\nOne prior is found for a and for b, while c only has a factor that touches a data variable and d has multiple priors.\nPedantic mode produces the following warning.\nWarning in 'priors.stan', line 7, column 2 to column 9:\n    The parameter c has no priors. This means either no prior is provided, or\n    the prior(s) depend on data variables. In the later case, this may be a\n    false positive.\nWarning in 'priors.stan', line 8, column 2 to column 9:\n    The parameter d has 2 priors.\n\n\n\n\nA warning is generated when any variable is used before it has been assigned a value.\nFor example, consider the following program.\ntransformed data {\n  real x;\n  if (1 &gt; 2) {\n    x = 1;\n  } else {\n    print(\"oops\");\n  }\n  print(x);\n}\nSince x is only assigned in one of the branches of the if statement, it might get to print(x) without having been assigned to.\nPedantic mode produces the following warning.\nWarning in 'uninit-warn.stan', line 7, column 8 to column 9:\n  The variable x may not have been assigned a value before its use.\n\n\n\nExcept when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds &lt;lower=.., upper=..&gt; except for &lt;lower=0, upper=1&gt; or &lt;lower=-1, upper=1&gt;.\nIn addition, a warning is generated when a parameter bound is found to have lower &gt;= upper.\nFor example, consider the following program.\nparameters {\n  real&lt;lower=0, upper=1&gt; a;\n  real&lt;lower=-1, upper=1&gt; b;\n  real&lt;lower=-2, upper=1012&gt; c;\n}\nmodel {\n  c ~ normal(b, a);\n}\nPedantic mode produces the following warning.\nWarning in 'hard-constraint.stan', line 4, column 2 to column 31:\n    Your Stan program has a parameter c with a lower and upper bound in its\n    declaration. These hard constraints are not recommended, for two reasons:\n    (a) Except when there are logical or physical constraints, it is very\n    unusual for you to be sure that a parameter will fall inside a specified\n    range, and (b) The infinite gradient induced by a hard constraint can\n    cause difficulties for Stan's sampling algorithm. As a consequence, we\n    recommend soft constraints rather than hard constraints; for example,\n    instead of constraining an elasticity parameter to fall between 0, and 1,\n    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.\n\n\n\nWhen a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the Changes of variables section.\nThis portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.\nFor example, consider the following program.\nparameters {\n  real y;\n}\nmodel {\n  log(y) ~ normal(0,1);\n}\nPedantic mode produces the following warning.\nWarning in 'jacobian.stan', line 5, column 2 to column 23:\n    Left-hand side of distribution statement (~) may contain a non-linear\n    transform of a parameter or local variable. If it does, you need to\n    include a target += statement with the log absolute determinant of the\n    Jacobian of the transform. You could also consider defining a transformed\n    parameter and using jacobian += in the transformed parameters block.\n\n\n\n\nConstant values are sometimes uncomputable\nPedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code normal(x, 1 - 2), the expression 1 - 2 will be evaluated to -1, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.\nContainer types\nCurrently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).\nData variables\nThe declaration information for data variables is currently not considered, so using data as incompatible arguments to distributions may not generate the appropriate warnings.\nControl flow dependent on parameters in nested functions\nIf a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.\n\n\n\n\n\nIn addition to compiling Stan programs, stanc3 features several flags which can be used to format Stan programs and update them to the most recent Stan syntax by removing any deprecation features which can be automatically replaced.\nThese flags work for both .stan model files and .stanfunctions function files. They can be combined with --o to redirect the formatted output to a new file.\n\n\nInvoking stanc --auto-format &lt;model_file&gt; will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the Stan Program Style Guide. This means spacing, indentation, and line length are all regularized. Some deprecated features, like the use of # for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.\nBy default, this will try to split lines at or before column 78. This number can be changed using --max-line-length.\n\n\n\nIn addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of if statements and for and while loops.\nThis can be done by using stanc --auto-format --canonicalize=... where ... is a comma-separated list of options. Currently these options are:\n\ndeprecations\nRemoves deprecated syntax such as replacing deprecated functions with their drop-in replacements.\nparentheses\nRemoves unnecessary extra parentheses, such as converting y = ((x-1)) to y = x - 1\nbraces\nPlaces braces around all blocks. For example, the following statement\nif (cond)\n  //result\nwill be formatted as\nif (cond) {\n  //result\n}\nand similarly for both kinds of loops containing a single statement.\nincludes\nThis will pretty-print code from other files included with #include as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same #include directives as the input program.\n\nInvoking stanc --print-canonical &lt;model_file&gt; is synonymous with running stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes\n\n\n\nThe formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:\n\nOddly placed comments\nIf your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string ^^^: to indicate they originally appeared higher in the program.\nWe hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.\nFailure to recreate strange #include structure\nPrinting without include inlining (--canonicalize=includes) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.\n\n\n\n\n\nThe stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.\nThis section introduces the available optimization options and describes their effect.\nTo print out a representation of the optimized Stan program, use the stanc3 command-line flag --debug-optimized-mir-pretty. To print an analogous representation of the Stan program prior to optimization, use the flag --debug-transformed-mir-pretty.\n\n\nTo turn optimizations on, the user specifies the desired optimization level. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.\nOptimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:\n\nO0 No optimizations are applied.\nO1 Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.\nOexperimental All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.\n\nO0 is the default setting.\nThe levels include these optimizations:\n\nO0 includes no optimizations.\nO1 includes:\n\nDead code elimination\nCopy propagation\nConstant propagation\nPartial evaluation\nFunction inlining\nMatrix memory layout optimization\n\nOexperimental includes optimizations specified by O1 and also:\n\nAutomatic-differentiation level optimization\nOne step loop unrolling\nExpression propagation\nLazy code motion\nStatic loop unrolling\n\n\nIn addition, Oexperimental will apply more repetitions of the optimizations, which may increase compile times.\n\n\n\n\n\nDead code is code that does not affect the behavior of the program. Code is not dead if it affects target, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.\nExample Stan program:\nmodel {\n  int i;\n  i = 5;\n  for (j in 1:10);\n  if (0) {\n    print(\"Dead code\");\n  } else {\n    print(\"Hi!\");\n  }\n}\nCompiler representation of program before dead code elimination (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  int i = 5;\n  for(j in 1:10) {\n    ;\n  }\n  if(0) {\n    FnPrint__(\"Dead code\");\n  } else {\n    FnPrint__(\"Hi!\");\n  }\n}\nCompiler representation of program after dead code elimination (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i;\n  FnPrint__(\"Hi!\");\n}\n\n\n\nConstant propagation replaces uses of a variable which is known to have a constant value C with that constant C. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).\nExample Stan program:\ntransformed data {\n  int n = 100;\n  int a[n];\n  for (i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before constant propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, n] a;\n  for(i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after constant propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, 100] a;\n  for(i in 1:100) {\n    a[i] = i;\n  }\n}\n\n\n\nCopy propagation is similar to expression propagation, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.\nExample Stan program:\nmodel {\n  int i = 1;\n  int j = i;\n  int k = i + j;\n}\nCompiler representation of program before copy propagation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n    int i = 1;\n    int j = i;\n    int k = (i + j);\n}\nCompiler representation of program after copy propagation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i = 1;\n  int j = i;\n  int k = (i + i);\n}\n\n\n\nPartial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.\nExample Stan program:\nmodel {\n  real a = 1 + 1;\n  real b = log(1 - a);\n  real c = a + b * 5;\n}\nCompiler representation of program before partial evaluation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real a = (1 + 1);\n  real b = log((1 - a));\n  real c = (a + (b * 5));\n}\nCompiler representation of program after partial evaluation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  real a = 2;\n  real b = log1m(a);\n  real c = fma(b, 5, a);\n}\n\n\n\nFunction inlining replaces each function call to each user-defined function f with the body of f. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return x + y;\n  }\n}\ntransformed data {\n  int a = 2;\n  int b = incr(a);\n}\nCompiler representation of program before function inlining (simplified from the output of --debug-transformed-mir-pretty):\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return (x + y);\n  }\n}\n\nprepare_data {\n  data int a = 2;\n  data int b = incr(a);\n}\nCompiler representation of program after function inlining (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int a;\n  a = 2;\n  data int b;\n  data int inline_sym1__;\n  data int inline_sym3__;\n  inline_sym3__ = 0;\n  for(inline_sym4__ in 1:1) {\n    int inline_sym2__;\n    inline_sym2__ = 1;\n    inline_sym3__ = 1;\n    inline_sym1__ = (a + inline_sym2__);\n    break;\n  }\n  b = inline_sym1__;\n}\nIn this code, the for loop and break is used to simulate the behavior of a return statement. The value to be returned is held in inline_sym1__. The flag variable inline_sym3__ indicates whether a return has occurred and is necessary to handle return statements nested inside loops within the function body.\n\n\n\nMatrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.\nThe first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.\nThe second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.\nMore general reading on AoS vs SoA can be found on Wikipedia\nThis optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:\n\nAny Stan Math Library functions the matrix is passed to must be able to support it.\nThe matrix should not be accessed/assigned elementwise in a loop.\n\nThe debug flag --debug-mem-patterns will list each variable and whether it is using the AoS representation or the SoA representation.\n\n\n\n\n\n\nStan variables can have two auto-differentiation (AD) levels: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.\nExample Stan program:\ndata {\n  real y;\n}\nmodel {\n  real x = y + 1;\n}\nCompiler representation of program before AD-level optimization (simplified from the output of --debug-transformed-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  real x = (y + 1);\n}\nCompiler representation of program after AD-level optimization (simplified from the output of --debug-optimized-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  data real x = (y + 1);\n}\n\n\n\nOne step loop unrolling is similar to static loop unrolling. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.\nExample Stan program:\ndata {\n  int n;\n}\ntransformed data {\n  int x = 0;\n  for (i in 1:n) {\n    x += i;\n  }\n}\nCompiler representation of program before one step static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  data int x = 0;\n  for(i in 1:n) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after one step static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  int x = 0;\n  if((n &gt;= 1)) {\n    x = (x + 1);\n    for(i in (1 + 1):n) {\n      x = (x + i);\n    }\n  }\n}\n\n\n\nConstant propagation replaces the uses of a variable which is known to have a constant value E with that constant E. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by lazy code motion to avoid unnecessarily recomputing expressions.\nExample Stan program:\ndata {\n  int m;\n}\ntransformed data {\n  int n = m+1;\n  int a[n];\n  for (i in 1:n-1) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before expression propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, n] a;\n  for(i in 1:(n - 1)) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after expression propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, (m + 1)] a;\n  for(i in 1:((m + 1) - 1)) {\n    a[i] = i;\n  }\n}\n\n\n\nLazy code motion rearranges the statements and expressions in a program with the goals of:\n\nAvoiding computing expressions more than once, and\nComputing expressions as late as possible (to minimize the strain on the working memory set).\n\nTo accomplish these goals, lazy code motion will perform optimizations such as:\n\nMoving a repeatedly calculated expression to its own variable (also referred to as common-subexpression elimination)\nMoving an expression outside of a loop if it does not need to be in the loop (also referred to as loop-invariant code motion)\n\nLazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.\nAs currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.\nExample Stan program:\nmodel {\n  real x;\n  real y;\n  real z;\n\n  for (i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program before lazy code motion (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real x;\n  real y;\n  real z;\n  for(i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program after lazy code motion (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  data real lcm_sym4__;\n  data real lcm_sym3__;\n  real x;\n  real y;\n  lcm_sym4__ = sqrt(10);\n  real z;\n  for(i in 1:10) {\n    x = lcm_sym4__;\n    y = sqrt(i);\n  }\n  z = lcm_sym4__;\n}\n\n\n\nStatic loop unrolling takes a loop with a predictable number of iterations X and replaces it by writing out the loop body X times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\ntransformed data {\n  int x = 0;\n  for (i in 1:4) {\n    x += i;\n  }\n}\nCompiler representation of program before static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int x = 0;\n  for(i in 1:4) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int x;\n  x = 0;\n  x = (x + 1);\n  x = (x + 2);\n  x = (x + 3);\n  x = (x + 4);\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Using the Stan Compiler"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/using-stanc.html#stanc-args",
+    "href": "stan-users-guide/using-stanc.html#stanc-args",
+    "title": "Using the Stan Compiler",
+    "section": "",
+    "text": "The stanc3 compiler has the following command-line syntax:\n&gt; stanc (options) &lt;model_file&gt;\nwhere &lt;model_file&gt; is either a path to a file ending in .stan or .stanfunctions (which automatically sets --standalone-functions), or '-' to read from standard input.\nThe stanc3 options are:\n\n--help / -? - Displays the complete list of stanc3 options, then exits.\n--version - Display stanc version number\n--info - Print information about the model, such as the type information for variables and the list of used distributions.\n--name=&lt;model_name&gt; - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.\n--o=&lt;file_name&gt; / -o=&lt;filename&gt; / --output=&lt;filename&gt; - Specify a path to an output file for generated C++ code (default = .hpp) or auto-formatting output (default: no file/print to stdout)\n--auto-format - Pretty prints the program to the console. See more on auto formatting.\n--allow-undefined - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.\n--canonicalize - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.\n--include_paths=&lt;dir1,...dirN&gt; - Takes a comma-separated list of directories that may contain a file in an #include directive.\n--max-line-length=&lt;number&gt; - Set the column number at which formatting with --auto-format attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.\n--print-canonical - Synonymous with --auto-format --canonicalize=deprecations,includes,parentheses,braces.\n--print-cpp - If set, output the generated C++ Stan model class to stdout.\n--filename-in-msg=&lt;name&gt; - Sets the filename used in compiler and runtime errors. If absent, the &lt;model_file&gt; argument is used.\n--standalone-functions - If set, only generate the code for the functions defined in the file. This is the default behavior for .stanfunctions files.\n--O0 (Default) Do not apply optimizations to the Stan code.\n--O1 Apply level 1 compiler optimizations (only basic optimizations).\n--Oexperimental WARNING: This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance! Allow the compiler to apply all optimizations to the Stan code.\n--O Synonym for --O1 as of Stan 2.37. In earlier versions this was a synonym for --Oexperimental.\n--use-opencl - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.\n--warn-pedantic - Emit warnings in Pedantic mode which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.\n--warn-uninitialized - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.\n--color - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are auto (the default), always, never. Can also be controlled by the STANC_COLOR environment variable.\n\nThe compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the --help option to see the full set.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Using the Stan Compiler"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/using-stanc.html#understanding-stanc3-errors-and-warnings",
+    "href": "stan-users-guide/using-stanc.html#understanding-stanc3-errors-and-warnings",
+    "title": "Using the Stan Compiler",
+    "section": "",
+    "text": "During model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).\n\n\nEven without the optional --warn-pedantic and --warn-uninitialized command line flags, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations\n\nA completely blank Stan program will produce the following warning message\nWarning in 'empty.stan', line 1, column 0 to line 2, column 0:\n    Empty model detected; this is a valid Stan model but likely unintended!\nThe use of any deprecated features will lead to warnings which will look as follows\n Warning in 'deprecated.stan', line 2, column 10 to column 17:\n     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with\n     an independent lognormal distribution on the scales, see:\n     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution\nA single Stan program can produce many warnings during compilation.\n\n\n\n\nErrors differ from warnings in their severity and format. In particular, errors are fatal and stop compilation, so at most one error is displayed per run of stanc.\nThere are five kinds of errors emitted by stanc3\n\nFile errors occur when the file passed to stanc is either missing or cannot be opened (i.e. has permissions issues). They look like\nError: file 'notfound.stan' not found or cannot be opened\nSyntactic errors occur whenever a program violates the Stan language’s syntax requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the #include directive fails, and “parsing” errors which result when the structure of the program is incorrect.\n\nThe lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of $ in a variable name will look like the following.\nSyntax error in 'char.stan', line 2, column 7 to column 8, lexing error:\n-------------------------------------------------\n  1:  data {\n  2:     int $ome_variable;\n             ^\n  3:  }\n-------------------------------------------------\nInvalid character found.\nWhen an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).\nSyntax error in './incl.stan', line 1, column 0, included from\n'./incl.stan', line 1, column 0, included from\n'incl.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include &lt;incl.stan&gt;\n         ^\n   -------------------------------------------------\nFile incl.stan recursively included itself.\nIt is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.\nSyntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:\n   -------------------------------------------------\n     1:  data {\n     2:     int&lt;lower=0&gt; N;\n     3:     vector x;\n                   ^\n     4:  }\n   -------------------------------------------------\nIll-formed type. Expected \"[\" expression \"]\" for vector size.\n\nSemantic errors (also known as type errors) occur when a program is structured correctly but features an error in the type rules imposed by the language. An example of this is assigning a real value to a variable defined as an integer.\nSemantic error in 'type.stan', line 2, column 3 to column 15:\n   -------------------------------------------------\n     1:  transformed data {\n     2:     int x = 1.5;\n            ^\n     3:  }\n   -------------------------------------------------\nIll-typed arguments supplied to assignment operator =:\nThe left hand side has type\n  int\nand the right hand side has type\n  real\nThe compiler will raise an error for use of any removed features for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.\nFinally, the compiler can raise an internal error. These are caused by bugs in the compiler, not your model, and we would appreciate it if you report them on the stanc3 repo with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Using the Stan Compiler"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/using-stanc.html#pedantic-mode",
+    "href": "stan-users-guide/using-stanc.html#pedantic-mode",
+    "title": "Using the Stan Compiler",
+    "section": "",
+    "text": "Pedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.\nFor example, consider the following program.\ndata {\n  int N;\n  array[N] real x;\n}\nparameters {\n  real sigma;\n}\nmodel {\n  real mu;\n  x ~ normal(mu, sigma);\n}\nWhen pedantic mode is turned on, the compiler will produce the following warnings.\nWarning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:\n    The parameter sigma has no priors. This means either no prior is\n    provided, or the prior(s) depend on data variables. In the later case,\n    this may be a false positive.\nWarning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:\n    The variable mu may not have been assigned a value before its use.\nWarning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:\n    A normal distribution is given parameter sigma as a scale parameter\n    (argument 2), but sigma was not constrained to be strictly positive.\nHere are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):\n\nDistribution usages issues. Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.\nUnused parameter. A parameter is defined but doesn’t contribute to target.\nLarge or small constant in a distribution. Very large or very small constants are used as distribution arguments.\nControl flow depends on a parameter. Branching control flow (like if/else) depends on a parameter value .\nParameter has multiple tildes. A parameter is on the left-hand side of multiple tildes.\nParameter has zero or multiple priors. A parameter has zero or more than one prior distribution.\nVariable is used before assignment. A variable is used before being assigned a value.\nStrict or nonsensical parameter bounds. A parameter is given questionable bounds.\nNonlinear transformations. When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.\n\nSome important limitations of pedantic mode are listed at the end of this chapter.\n\n\nWhen an argument to a built-in distribution certainly does not match that distribution’s specification in the Stan Functions Reference, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g. simplex) is incompatible with the domain of the distribution. x\nFor example, consider the following program.\nparameters {\n  real unb_p;\n  real&lt;lower=0&gt; pos_p;\n}\nmodel {\n  1 ~ poisson(unb_p);\n  1 ~ poisson(pos_p);\n}\nThe parameter of poisson should be strictly positive, but unb_p is not constrained to be positive.\nPedantic mode produces the following warning.\nWarning in 'ex-dist-args.stan', line 6, column 14 to column 19:\n  A poisson distribution is given parameter unb_p as a rate parameter\n  (argument 1), but unb_p was not constrained to be strictly positive.\n\n\n\nPedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:\n\n\nAny use of uniform distribution generates a warning, except when the variate parameter’s declared upper and lower bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.\nFor example, consider the following program.\nparameters {\n  real a;\n  real&lt;lower=0, upper=1&gt; b;\n}\nmodel {\n  a ~ uniform(0, 1);\n  b ~ uniform(0, 1);\n}\na is assigned a uniform distribution that doesn’t match its constraints.\nPedantic mode produces the following warning.\nWarning in 'uniform-warn.stan', line 6, column 2 to column 20:\n  Parameter a is given a uniform distribution. The uniform distribution is\n  not recommended, for two reasons: (a) Except when there are logical or\n  physical constraints, it is very unusual for you to be sure that a\n  parameter will fall inside a specified range, and (b) The infinite gradient\n  induced by a uniform density can cause difficulties for Stan's sampling\n  algorithm. As a consequence, we recommend soft constraints rather than hard\n  constraints; for example, instead of giving an elasticity parameter a\n  uniform(0, 1) distribution, try normal(0.5, 0.5).\n\n\n\nGamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.\n\n\n\nAny use of the lkj_corr distribution generates a warning that suggests using the Cholesky variant instead. See the LKJ correlation distribution section of the Stan Functions Reference for details.\n\n\n\n\nA warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the target variable depends in any way on each of the parameters.\nFor example, consider the following program.\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(1, 1);\n}\na participates in the density function but b does not.\nPedantic mode produces the following warning.\nWarning in 'unused.stan', line 3, column 2 to column 9:\n    The parameter b was declared but was not used in the density calculation.\n\n\n\nWhen numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See the efficiency tuning section of the Stan User’s guide for a discussion of scaling parameters.\nFor example, consider the following program.\nparameters {\n  real x;\n  real y;\n}\nmodel {\n  x ~ normal(-100, 100);\n  y ~ normal(0, 1);\n}\nThe constants -100 and 100 suggest that x is not unit scaled.\nPedantic mode produces the following warning.\nWarning in 'constants-warn.stan', line 6, column 13 to column 17:\n    Argument -100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\nWarning in 'constants-warn.stan', line 6, column 19 to column 22:\n    Argument 100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\n\n\n\nControl flow statements, such as if, for and while should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.\nFor example, consider the following program.\nparameters {\n  real a;\n}\nmodel {\n  // x depends on parameter a\n  real x = a * a;\n\n  int m;\n\n  // the if-then-else depends on x which depends on a\n  if(x &gt; 0) {\n    //now m depends on x which depends on a\n    m = 1;\n  } else {\n    m = 2;\n  }\n\n  // for loop depends on m -&gt; x -&gt; a\n  for (i in 0:m) {\n    a ~ normal(i, 1);\n  }\n}\nThe if and for statements are control flow that depend (indirectly) on the value of the parameter m.\nPedantic mode produces the following warning.\nWarning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:\n  A control flow statement depends on parameter(s): a.\nWarning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:\n  A control flow statement depends on parameter(s): a.\n\n\n\nA warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent target += conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.\nPedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(0, 1);\n  a ~ normal(x, 1);\n\n  b ~ normal(1, 1);\n}\nPedantic mode produces the following warning.\nWarning in 'multi-tildes.stan', line 9, column 2 to column 19:\n  The parameter a is on the left-hand side of more than one tildes\n  statement.\n\n\n\nA warning is generated when a parameter appears to have greater than or less than one prior distribution factor.\nThis analysis depends on a factor graph representation of a Stan program. A factor F that depends on a parameter P is called a prior factor for P if there is no path in the factor graph from F to any data variable except through P.\nOne limitation of this approach is that the compiler cannot distinguish between modeled data variables and other convenient uses of data variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for int variables) are modeled data, which may cause extra warnings.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n  real c;\n  real d;\n}\nmodel\n{\n  a ~ normal(0, 1); // this is a prior\n  x ~ normal(a, 1); // this is not a prior, since data is involved\n\n  b ~ normal(x, 1); // this is also not a prior, since data is involved\n\n  // this is not a prior for c, since data is involved through b\n  // but it is a prior for b, since the data is only involved through b\n  c ~ normal(b, 1);\n\n  //these are multiple priors:\n  d ~ normal(0, 1);\n  1 ~ normal(d, 1);\n}\nOne prior is found for a and for b, while c only has a factor that touches a data variable and d has multiple priors.\nPedantic mode produces the following warning.\nWarning in 'priors.stan', line 7, column 2 to column 9:\n    The parameter c has no priors. This means either no prior is provided, or\n    the prior(s) depend on data variables. In the later case, this may be a\n    false positive.\nWarning in 'priors.stan', line 8, column 2 to column 9:\n    The parameter d has 2 priors.\n\n\n\n\nA warning is generated when any variable is used before it has been assigned a value.\nFor example, consider the following program.\ntransformed data {\n  real x;\n  if (1 &gt; 2) {\n    x = 1;\n  } else {\n    print(\"oops\");\n  }\n  print(x);\n}\nSince x is only assigned in one of the branches of the if statement, it might get to print(x) without having been assigned to.\nPedantic mode produces the following warning.\nWarning in 'uninit-warn.stan', line 7, column 8 to column 9:\n  The variable x may not have been assigned a value before its use.\n\n\n\nExcept when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds &lt;lower=.., upper=..&gt; except for &lt;lower=0, upper=1&gt; or &lt;lower=-1, upper=1&gt;.\nIn addition, a warning is generated when a parameter bound is found to have lower &gt;= upper.\nFor example, consider the following program.\nparameters {\n  real&lt;lower=0, upper=1&gt; a;\n  real&lt;lower=-1, upper=1&gt; b;\n  real&lt;lower=-2, upper=1012&gt; c;\n}\nmodel {\n  c ~ normal(b, a);\n}\nPedantic mode produces the following warning.\nWarning in 'hard-constraint.stan', line 4, column 2 to column 31:\n    Your Stan program has a parameter c with a lower and upper bound in its\n    declaration. These hard constraints are not recommended, for two reasons:\n    (a) Except when there are logical or physical constraints, it is very\n    unusual for you to be sure that a parameter will fall inside a specified\n    range, and (b) The infinite gradient induced by a hard constraint can\n    cause difficulties for Stan's sampling algorithm. As a consequence, we\n    recommend soft constraints rather than hard constraints; for example,\n    instead of constraining an elasticity parameter to fall between 0, and 1,\n    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.\n\n\n\nWhen a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the Changes of variables section.\nThis portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.\nFor example, consider the following program.\nparameters {\n  real y;\n}\nmodel {\n  log(y) ~ normal(0,1);\n}\nPedantic mode produces the following warning.\nWarning in 'jacobian.stan', line 5, column 2 to column 23:\n    Left-hand side of distribution statement (~) may contain a non-linear\n    transform of a parameter or local variable. If it does, you need to\n    include a target += statement with the log absolute determinant of the\n    Jacobian of the transform. You could also consider defining a transformed\n    parameter and using jacobian += in the transformed parameters block.\n\n\n\n\nConstant values are sometimes uncomputable\nPedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code normal(x, 1 - 2), the expression 1 - 2 will be evaluated to -1, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.\nContainer types\nCurrently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).\nData variables\nThe declaration information for data variables is currently not considered, so using data as incompatible arguments to distributions may not generate the appropriate warnings.\nControl flow dependent on parameters in nested functions\nIf a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Using the Stan Compiler"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/using-stanc.html#stanc-pretty-printing",
+    "href": "stan-users-guide/using-stanc.html#stanc-pretty-printing",
+    "title": "Using the Stan Compiler",
+    "section": "",
+    "text": "In addition to compiling Stan programs, stanc3 features several flags which can be used to format Stan programs and update them to the most recent Stan syntax by removing any deprecation features which can be automatically replaced.\nThese flags work for both .stan model files and .stanfunctions function files. They can be combined with --o to redirect the formatted output to a new file.\n\n\nInvoking stanc --auto-format &lt;model_file&gt; will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the Stan Program Style Guide. This means spacing, indentation, and line length are all regularized. Some deprecated features, like the use of # for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.\nBy default, this will try to split lines at or before column 78. This number can be changed using --max-line-length.\n\n\n\nIn addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of if statements and for and while loops.\nThis can be done by using stanc --auto-format --canonicalize=... where ... is a comma-separated list of options. Currently these options are:\n\ndeprecations\nRemoves deprecated syntax such as replacing deprecated functions with their drop-in replacements.\nparentheses\nRemoves unnecessary extra parentheses, such as converting y = ((x-1)) to y = x - 1\nbraces\nPlaces braces around all blocks. For example, the following statement\nif (cond)\n  //result\nwill be formatted as\nif (cond) {\n  //result\n}\nand similarly for both kinds of loops containing a single statement.\nincludes\nThis will pretty-print code from other files included with #include as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same #include directives as the input program.\n\nInvoking stanc --print-canonical &lt;model_file&gt; is synonymous with running stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes\n\n\n\nThe formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:\n\nOddly placed comments\nIf your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string ^^^: to indicate they originally appeared higher in the program.\nWe hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.\nFailure to recreate strange #include structure\nPrinting without include inlining (--canonicalize=includes) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Using the Stan Compiler"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/using-stanc.html#optimization",
+    "href": "stan-users-guide/using-stanc.html#optimization",
+    "title": "Using the Stan Compiler",
+    "section": "",
+    "text": "The stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.\nThis section introduces the available optimization options and describes their effect.\nTo print out a representation of the optimized Stan program, use the stanc3 command-line flag --debug-optimized-mir-pretty. To print an analogous representation of the Stan program prior to optimization, use the flag --debug-transformed-mir-pretty.\n\n\nTo turn optimizations on, the user specifies the desired optimization level. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.\nOptimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:\n\nO0 No optimizations are applied.\nO1 Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.\nOexperimental All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.\n\nO0 is the default setting.\nThe levels include these optimizations:\n\nO0 includes no optimizations.\nO1 includes:\n\nDead code elimination\nCopy propagation\nConstant propagation\nPartial evaluation\nFunction inlining\nMatrix memory layout optimization\n\nOexperimental includes optimizations specified by O1 and also:\n\nAutomatic-differentiation level optimization\nOne step loop unrolling\nExpression propagation\nLazy code motion\nStatic loop unrolling\n\n\nIn addition, Oexperimental will apply more repetitions of the optimizations, which may increase compile times.\n\n\n\n\n\nDead code is code that does not affect the behavior of the program. Code is not dead if it affects target, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.\nExample Stan program:\nmodel {\n  int i;\n  i = 5;\n  for (j in 1:10);\n  if (0) {\n    print(\"Dead code\");\n  } else {\n    print(\"Hi!\");\n  }\n}\nCompiler representation of program before dead code elimination (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  int i = 5;\n  for(j in 1:10) {\n    ;\n  }\n  if(0) {\n    FnPrint__(\"Dead code\");\n  } else {\n    FnPrint__(\"Hi!\");\n  }\n}\nCompiler representation of program after dead code elimination (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i;\n  FnPrint__(\"Hi!\");\n}\n\n\n\nConstant propagation replaces uses of a variable which is known to have a constant value C with that constant C. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).\nExample Stan program:\ntransformed data {\n  int n = 100;\n  int a[n];\n  for (i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before constant propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, n] a;\n  for(i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after constant propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, 100] a;\n  for(i in 1:100) {\n    a[i] = i;\n  }\n}\n\n\n\nCopy propagation is similar to expression propagation, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.\nExample Stan program:\nmodel {\n  int i = 1;\n  int j = i;\n  int k = i + j;\n}\nCompiler representation of program before copy propagation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n    int i = 1;\n    int j = i;\n    int k = (i + j);\n}\nCompiler representation of program after copy propagation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i = 1;\n  int j = i;\n  int k = (i + i);\n}\n\n\n\nPartial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.\nExample Stan program:\nmodel {\n  real a = 1 + 1;\n  real b = log(1 - a);\n  real c = a + b * 5;\n}\nCompiler representation of program before partial evaluation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real a = (1 + 1);\n  real b = log((1 - a));\n  real c = (a + (b * 5));\n}\nCompiler representation of program after partial evaluation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  real a = 2;\n  real b = log1m(a);\n  real c = fma(b, 5, a);\n}\n\n\n\nFunction inlining replaces each function call to each user-defined function f with the body of f. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return x + y;\n  }\n}\ntransformed data {\n  int a = 2;\n  int b = incr(a);\n}\nCompiler representation of program before function inlining (simplified from the output of --debug-transformed-mir-pretty):\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return (x + y);\n  }\n}\n\nprepare_data {\n  data int a = 2;\n  data int b = incr(a);\n}\nCompiler representation of program after function inlining (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int a;\n  a = 2;\n  data int b;\n  data int inline_sym1__;\n  data int inline_sym3__;\n  inline_sym3__ = 0;\n  for(inline_sym4__ in 1:1) {\n    int inline_sym2__;\n    inline_sym2__ = 1;\n    inline_sym3__ = 1;\n    inline_sym1__ = (a + inline_sym2__);\n    break;\n  }\n  b = inline_sym1__;\n}\nIn this code, the for loop and break is used to simulate the behavior of a return statement. The value to be returned is held in inline_sym1__. The flag variable inline_sym3__ indicates whether a return has occurred and is necessary to handle return statements nested inside loops within the function body.\n\n\n\nMatrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.\nThe first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.\nThe second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.\nMore general reading on AoS vs SoA can be found on Wikipedia\nThis optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:\n\nAny Stan Math Library functions the matrix is passed to must be able to support it.\nThe matrix should not be accessed/assigned elementwise in a loop.\n\nThe debug flag --debug-mem-patterns will list each variable and whether it is using the AoS representation or the SoA representation.\n\n\n\n\n\n\nStan variables can have two auto-differentiation (AD) levels: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.\nExample Stan program:\ndata {\n  real y;\n}\nmodel {\n  real x = y + 1;\n}\nCompiler representation of program before AD-level optimization (simplified from the output of --debug-transformed-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  real x = (y + 1);\n}\nCompiler representation of program after AD-level optimization (simplified from the output of --debug-optimized-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  data real x = (y + 1);\n}\n\n\n\nOne step loop unrolling is similar to static loop unrolling. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.\nExample Stan program:\ndata {\n  int n;\n}\ntransformed data {\n  int x = 0;\n  for (i in 1:n) {\n    x += i;\n  }\n}\nCompiler representation of program before one step static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  data int x = 0;\n  for(i in 1:n) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after one step static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  int x = 0;\n  if((n &gt;= 1)) {\n    x = (x + 1);\n    for(i in (1 + 1):n) {\n      x = (x + i);\n    }\n  }\n}\n\n\n\nConstant propagation replaces the uses of a variable which is known to have a constant value E with that constant E. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by lazy code motion to avoid unnecessarily recomputing expressions.\nExample Stan program:\ndata {\n  int m;\n}\ntransformed data {\n  int n = m+1;\n  int a[n];\n  for (i in 1:n-1) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before expression propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, n] a;\n  for(i in 1:(n - 1)) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after expression propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, (m + 1)] a;\n  for(i in 1:((m + 1) - 1)) {\n    a[i] = i;\n  }\n}\n\n\n\nLazy code motion rearranges the statements and expressions in a program with the goals of:\n\nAvoiding computing expressions more than once, and\nComputing expressions as late as possible (to minimize the strain on the working memory set).\n\nTo accomplish these goals, lazy code motion will perform optimizations such as:\n\nMoving a repeatedly calculated expression to its own variable (also referred to as common-subexpression elimination)\nMoving an expression outside of a loop if it does not need to be in the loop (also referred to as loop-invariant code motion)\n\nLazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.\nAs currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.\nExample Stan program:\nmodel {\n  real x;\n  real y;\n  real z;\n\n  for (i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program before lazy code motion (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real x;\n  real y;\n  real z;\n  for(i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program after lazy code motion (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  data real lcm_sym4__;\n  data real lcm_sym3__;\n  real x;\n  real y;\n  lcm_sym4__ = sqrt(10);\n  real z;\n  for(i in 1:10) {\n    x = lcm_sym4__;\n    y = sqrt(i);\n  }\n  z = lcm_sym4__;\n}\n\n\n\nStatic loop unrolling takes a loop with a predictable number of iterations X and replaces it by writing out the loop body X times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\ntransformed data {\n  int x = 0;\n  for (i in 1:4) {\n    x += i;\n  }\n}\nCompiler representation of program before static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int x = 0;\n  for(i in 1:4) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int x;\n  x = 0;\n  x = (x + 1);\n  x = (x + 2);\n  x = (x + 3);\n  x = (x + 4);\n}",
+    "crumbs": [
+      "Stan Users Guide",
+      "Appendices",
+      "Using the Stan Compiler"
+    ]
+  }
+]
\ No newline at end of file
diff --git a/docs/2_39/site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css b/docs/2_39/site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css
new file mode 100644
index 000000000..a9fb78da9
--- /dev/null
+++ b/docs/2_39/site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css
@@ -0,0 +1,12 @@
+@import"https://fonts.googleapis.com/css2?family=Source+Sans+Pro:wght@300;400;700&display=swap";:root{--stan-bg: #FEFCF9;--stan-highlight: #F2E5BD;--stan-secondary: #3E8EBC;--stan-dark: #052744;--stan-hero: #F2E5BD;--stan-hero-bg: #052744}/*!
+ * Bootstrap  v5.3.1 (https://getbootstrap.com/)
+ * Copyright 2011-2023 The Bootstrap Authors
+ * Licensed under MIT (https://github.com/twbs/bootstrap/blob/main/LICENSE)
+ */:root,[data-bs-theme=light]{--bs-blue: #2780e3;--bs-indigo: #6610f2;--bs-purple: #613d7c;--bs-pink: #e83e8c;--bs-red: #ff0039;--bs-orange: #f0ad4e;--bs-yellow: #ff7518;--bs-green: #3fb618;--bs-teal: #20c997;--bs-cyan: #9954bb;--bs-black: #000;--bs-white: #fff;--bs-gray: #6c757d;--bs-gray-dark: #343a40;--bs-gray-100: #f8f9fa;--bs-gray-200: #e9ecef;--bs-gray-300: #dee2e6;--bs-gray-400: #ced4da;--bs-gray-500: #adb5bd;--bs-gray-600: #6c757d;--bs-gray-700: #495057;--bs-gray-800: #343a40;--bs-gray-900: #212529;--bs-default: #343a40;--bs-primary: #2780e3;--bs-secondary: #343a40;--bs-success: #3fb618;--bs-info: #9954bb;--bs-warning: #ff7518;--bs-danger: #ff0039;--bs-light: #f8f9fa;--bs-dark: #343a40;--bs-default-rgb: 52, 58, 64;--bs-primary-rgb: 39, 128, 227;--bs-secondary-rgb: 52, 58, 64;--bs-success-rgb: 63, 182, 24;--bs-info-rgb: 153, 84, 187;--bs-warning-rgb: 255, 117, 24;--bs-danger-rgb: 255, 0, 57;--bs-light-rgb: 248, 249, 250;--bs-dark-rgb: 52, 58, 64;--bs-primary-text-emphasis: #10335b;--bs-secondary-text-emphasis: #15171a;--bs-success-text-emphasis: #19490a;--bs-info-text-emphasis: #3d224b;--bs-warning-text-emphasis: #662f0a;--bs-danger-text-emphasis: #660017;--bs-light-text-emphasis: #495057;--bs-dark-text-emphasis: #495057;--bs-primary-bg-subtle: #d4e6f9;--bs-secondary-bg-subtle: #d6d8d9;--bs-success-bg-subtle: #d9f0d1;--bs-info-bg-subtle: #ebddf1;--bs-warning-bg-subtle: #ffe3d1;--bs-danger-bg-subtle: #ffccd7;--bs-light-bg-subtle: #fcfcfd;--bs-dark-bg-subtle: #ced4da;--bs-primary-border-subtle: #a9ccf4;--bs-secondary-border-subtle: #aeb0b3;--bs-success-border-subtle: #b2e2a3;--bs-info-border-subtle: #d6bbe4;--bs-warning-border-subtle: #ffc8a3;--bs-danger-border-subtle: #ff99b0;--bs-light-border-subtle: #e9ecef;--bs-dark-border-subtle: #adb5bd;--bs-white-rgb: 255, 255, 255;--bs-black-rgb: 0, 0, 0;--bs-font-sans-serif: "Source Sans Pro", -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol";--bs-font-monospace: SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", "Courier New", monospace;--bs-gradient: linear-gradient(180deg, rgba(255, 255, 255, 0.15), rgba(255, 255, 255, 0));--bs-root-font-size: 17px;--bs-body-font-family: "Source Sans Pro", -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol";--bs-body-font-size:1rem;--bs-body-font-weight: 400;--bs-body-line-height: 1.5;--bs-body-color: #343a40;--bs-body-color-rgb: 52, 58, 64;--bs-body-bg: #FEFBF2;--bs-body-bg-rgb: 254, 251, 242;--bs-emphasis-color: #000;--bs-emphasis-color-rgb: 0, 0, 0;--bs-secondary-color: rgba(52, 58, 64, 0.75);--bs-secondary-color-rgb: 52, 58, 64;--bs-secondary-bg: #e9ecef;--bs-secondary-bg-rgb: 233, 236, 239;--bs-tertiary-color: rgba(52, 58, 64, 0.5);--bs-tertiary-color-rgb: 52, 58, 64;--bs-tertiary-bg: #f8f9fa;--bs-tertiary-bg-rgb: 248, 249, 250;--bs-heading-color: inherit;--bs-link-color: #2761e3;--bs-link-color-rgb: 39, 97, 227;--bs-link-decoration: underline;--bs-link-hover-color: #1f4eb6;--bs-link-hover-color-rgb: 31, 78, 182;--bs-code-color: #7d12ba;--bs-highlight-bg: #ffe3d1;--bs-border-width: 1px;--bs-border-style: solid;--bs-border-color: #dee2e6;--bs-border-color-translucent: rgba(0, 0, 0, 0.175);--bs-border-radius: 0.25rem;--bs-border-radius-sm: 0.2em;--bs-border-radius-lg: 0.5rem;--bs-border-radius-xl: 1rem;--bs-border-radius-xxl: 2rem;--bs-border-radius-2xl: var(--bs-border-radius-xxl);--bs-border-radius-pill: 50rem;--bs-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-box-shadow-sm: 0 0.125rem 0.25rem rgba(0, 0, 0, 0.075);--bs-box-shadow-lg: 0 1rem 3rem rgba(0, 0, 0, 0.175);--bs-box-shadow-inset: inset 0 1px 2px rgba(0, 0, 0, 0.075);--bs-focus-ring-width: 0.25rem;--bs-focus-ring-opacity: 0.25;--bs-focus-ring-color: rgba(39, 128, 227, 0.25);--bs-form-valid-color: #3fb618;--bs-form-valid-border-color: #3fb618;--bs-form-invalid-color: #ff0039;--bs-form-invalid-border-color: #ff0039}[data-bs-theme=dark]{color-scheme:dark;--bs-body-color: #dee2e6;--bs-body-color-rgb: 222, 226, 230;--bs-body-bg: #212529;--bs-body-bg-rgb: 33, 37, 41;--bs-emphasis-color: #fff;--bs-emphasis-color-rgb: 255, 255, 255;--bs-secondary-color: rgba(222, 226, 230, 0.75);--bs-secondary-color-rgb: 222, 226, 230;--bs-secondary-bg: #343a40;--bs-secondary-bg-rgb: 52, 58, 64;--bs-tertiary-color: rgba(222, 226, 230, 0.5);--bs-tertiary-color-rgb: 222, 226, 230;--bs-tertiary-bg: #2b3035;--bs-tertiary-bg-rgb: 43, 48, 53;--bs-primary-text-emphasis: #7db3ee;--bs-secondary-text-emphasis: #85898c;--bs-success-text-emphasis: #8cd374;--bs-info-text-emphasis: #c298d6;--bs-warning-text-emphasis: #ffac74;--bs-danger-text-emphasis: #ff6688;--bs-light-text-emphasis: #f8f9fa;--bs-dark-text-emphasis: #dee2e6;--bs-primary-bg-subtle: #081a2d;--bs-secondary-bg-subtle: #0a0c0d;--bs-success-bg-subtle: #0d2405;--bs-info-bg-subtle: #1f1125;--bs-warning-bg-subtle: #331705;--bs-danger-bg-subtle: #33000b;--bs-light-bg-subtle: #343a40;--bs-dark-bg-subtle: #1a1d20;--bs-primary-border-subtle: #174d88;--bs-secondary-border-subtle: #1f2326;--bs-success-border-subtle: #266d0e;--bs-info-border-subtle: #5c3270;--bs-warning-border-subtle: #99460e;--bs-danger-border-subtle: #990022;--bs-light-border-subtle: #495057;--bs-dark-border-subtle: #343a40;--bs-heading-color: inherit;--bs-link-color: #7db3ee;--bs-link-hover-color: #97c2f1;--bs-link-color-rgb: 125, 179, 238;--bs-link-hover-color-rgb: 151, 194, 241;--bs-code-color: white;--bs-border-color: #495057;--bs-border-color-translucent: rgba(255, 255, 255, 0.15);--bs-form-valid-color: #8cd374;--bs-form-valid-border-color: #8cd374;--bs-form-invalid-color: #ff6688;--bs-form-invalid-border-color: #ff6688}*,*::before,*::after{box-sizing:border-box}:root{font-size:var(--bs-root-font-size)}body{margin:0;font-family:var(--bs-body-font-family);font-size:var(--bs-body-font-size);font-weight:var(--bs-body-font-weight);line-height:var(--bs-body-line-height);color:var(--bs-body-color);text-align:var(--bs-body-text-align);background-color:var(--bs-body-bg);-webkit-text-size-adjust:100%;-webkit-tap-highlight-color:rgba(0,0,0,0)}hr{margin:1rem 0;color:inherit;border:0;border-top:1px solid;opacity:.25}h6,.h6,h5,.h5,h4,.h4,h3,.h3,h2,.h2,h1,.h1{margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2;color:var(--bs-heading-color)}h1,.h1{font-size:calc(1.325rem + 0.9vw)}@media(min-width: 1200px){h1,.h1{font-size:2rem}}h2,.h2{font-size:calc(1.29rem + 0.48vw)}@media(min-width: 1200px){h2,.h2{font-size:1.65rem}}h3,.h3{font-size:calc(1.27rem + 0.24vw)}@media(min-width: 1200px){h3,.h3{font-size:1.45rem}}h4,.h4{font-size:1.25rem}h5,.h5{font-size:1.1rem}h6,.h6{font-size:1rem}p{margin-top:0;margin-bottom:1rem}abbr[title]{text-decoration:underline dotted;-webkit-text-decoration:underline dotted;-moz-text-decoration:underline dotted;-ms-text-decoration:underline dotted;-o-text-decoration:underline dotted;cursor:help;text-decoration-skip-ink:none}address{margin-bottom:1rem;font-style:normal;line-height:inherit}ol,ul{padding-left:2rem}ol,ul,dl{margin-top:0;margin-bottom:1rem}ol ol,ul ul,ol ul,ul ol{margin-bottom:0}dt{font-weight:700}dd{margin-bottom:.5rem;margin-left:0}blockquote{margin:0 0 1rem;padding:.625rem 1.25rem;border-left:.25rem solid #e9ecef}blockquote p:last-child,blockquote ul:last-child,blockquote ol:last-child{margin-bottom:0}b,strong{font-weight:bolder}small,.small{font-size:0.875em}mark,.mark{padding:.1875em;background-color:var(--bs-highlight-bg)}sub,sup{position:relative;font-size:0.75em;line-height:0;vertical-align:baseline}sub{bottom:-0.25em}sup{top:-0.5em}a{color:rgba(var(--bs-link-color-rgb), var(--bs-link-opacity, 1));text-decoration:underline;-webkit-text-decoration:underline;-moz-text-decoration:underline;-ms-text-decoration:underline;-o-text-decoration:underline}a:hover{--bs-link-color-rgb: var(--bs-link-hover-color-rgb)}a:not([href]):not([class]),a:not([href]):not([class]):hover{color:inherit;text-decoration:none}pre,code,kbd,samp{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;font-size:1em}pre{display:block;margin-top:0;margin-bottom:1rem;overflow:auto;font-size:0.875em;color:#000;background-color:#f8f9fa;line-height:1.5;padding:.5rem;border:1px solid var(--bs-border-color, #dee2e6)}pre code{background-color:rgba(0,0,0,0);font-size:inherit;color:inherit;word-break:normal}code{font-size:0.875em;color:var(--bs-code-color);background-color:#f8f9fa;padding:.125rem .25rem;word-wrap:break-word}a>code{color:inherit}kbd{padding:.4rem .4rem;font-size:0.875em;color:#fefbf2;background-color:#343a40}kbd kbd{padding:0;font-size:1em}figure{margin:0 0 1rem}img,svg{vertical-align:middle}table{caption-side:bottom;border-collapse:collapse}caption{padding-top:.5rem;padding-bottom:.5rem;color:rgba(52,58,64,.75);text-align:left}th{text-align:inherit;text-align:-webkit-match-parent}thead,tbody,tfoot,tr,td,th{border-color:inherit;border-style:solid;border-width:0}label{display:inline-block}button{border-radius:0}button:focus:not(:focus-visible){outline:0}input,button,select,optgroup,textarea{margin:0;font-family:inherit;font-size:inherit;line-height:inherit}button,select{text-transform:none}[role=button]{cursor:pointer}select{word-wrap:normal}select:disabled{opacity:1}[list]:not([type=date]):not([type=datetime-local]):not([type=month]):not([type=week]):not([type=time])::-webkit-calendar-picker-indicator{display:none !important}button,[type=button],[type=reset],[type=submit]{-webkit-appearance:button}button:not(:disabled),[type=button]:not(:disabled),[type=reset]:not(:disabled),[type=submit]:not(:disabled){cursor:pointer}::-moz-focus-inner{padding:0;border-style:none}textarea{resize:vertical}fieldset{min-width:0;padding:0;margin:0;border:0}legend{float:left;width:100%;padding:0;margin-bottom:.5rem;font-size:calc(1.275rem + 0.3vw);line-height:inherit}@media(min-width: 1200px){legend{font-size:1.5rem}}legend+*{clear:left}::-webkit-datetime-edit-fields-wrapper,::-webkit-datetime-edit-text,::-webkit-datetime-edit-minute,::-webkit-datetime-edit-hour-field,::-webkit-datetime-edit-day-field,::-webkit-datetime-edit-month-field,::-webkit-datetime-edit-year-field{padding:0}::-webkit-inner-spin-button{height:auto}[type=search]{-webkit-appearance:textfield;outline-offset:-2px}::-webkit-search-decoration{-webkit-appearance:none}::-webkit-color-swatch-wrapper{padding:0}::file-selector-button{font:inherit;-webkit-appearance:button}output{display:inline-block}iframe{border:0}summary{display:list-item;cursor:pointer}progress{vertical-align:baseline}[hidden]{display:none !important}.lead{font-size:1.25rem;font-weight:300}.display-1{font-size:calc(1.625rem + 4.5vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-1{font-size:5rem}}.display-2{font-size:calc(1.575rem + 3.9vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-2{font-size:4.5rem}}.display-3{font-size:calc(1.525rem + 3.3vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-3{font-size:4rem}}.display-4{font-size:calc(1.475rem + 2.7vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-4{font-size:3.5rem}}.display-5{font-size:calc(1.425rem + 2.1vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-5{font-size:3rem}}.display-6{font-size:calc(1.375rem + 1.5vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-6{font-size:2.5rem}}.list-unstyled{padding-left:0;list-style:none}.list-inline{padding-left:0;list-style:none}.list-inline-item{display:inline-block}.list-inline-item:not(:last-child){margin-right:.5rem}.initialism{font-size:0.875em;text-transform:uppercase}.blockquote{margin-bottom:1rem;font-size:1.25rem}.blockquote>:last-child{margin-bottom:0}.blockquote-footer{margin-top:-1rem;margin-bottom:1rem;font-size:0.875em;color:#6c757d}.blockquote-footer::before{content:"— "}.img-fluid{max-width:100%;height:auto}.img-thumbnail{padding:.25rem;background-color:#fefbf2;border:1px solid #dee2e6;max-width:100%;height:auto}.figure{display:inline-block}.figure-img{margin-bottom:.5rem;line-height:1}.figure-caption{font-size:0.875em;color:rgba(52,58,64,.75)}.container,.container-fluid,.container-xxl,.container-xl,.container-lg,.container-md,.container-sm{--bs-gutter-x: 1.5rem;--bs-gutter-y: 0;width:100%;padding-right:calc(var(--bs-gutter-x)*.5);padding-left:calc(var(--bs-gutter-x)*.5);margin-right:auto;margin-left:auto}@media(min-width: 576px){.container-sm,.container{max-width:540px}}@media(min-width: 768px){.container-md,.container-sm,.container{max-width:720px}}@media(min-width: 992px){.container-lg,.container-md,.container-sm,.container{max-width:960px}}@media(min-width: 1200px){.container-xl,.container-lg,.container-md,.container-sm,.container{max-width:1140px}}@media(min-width: 1400px){.container-xxl,.container-xl,.container-lg,.container-md,.container-sm,.container{max-width:1320px}}:root{--bs-breakpoint-xs: 0;--bs-breakpoint-sm: 576px;--bs-breakpoint-md: 768px;--bs-breakpoint-lg: 992px;--bs-breakpoint-xl: 1200px;--bs-breakpoint-xxl: 1400px}.grid{display:grid;grid-template-rows:repeat(var(--bs-rows, 1), 1fr);grid-template-columns:repeat(var(--bs-columns, 12), 1fr);gap:var(--bs-gap, 1.5rem)}.grid .g-col-1{grid-column:auto/span 1}.grid .g-col-2{grid-column:auto/span 2}.grid .g-col-3{grid-column:auto/span 3}.grid .g-col-4{grid-column:auto/span 4}.grid .g-col-5{grid-column:auto/span 5}.grid .g-col-6{grid-column:auto/span 6}.grid .g-col-7{grid-column:auto/span 7}.grid .g-col-8{grid-column:auto/span 8}.grid .g-col-9{grid-column:auto/span 9}.grid .g-col-10{grid-column:auto/span 10}.grid .g-col-11{grid-column:auto/span 11}.grid .g-col-12{grid-column:auto/span 12}.grid .g-start-1{grid-column-start:1}.grid .g-start-2{grid-column-start:2}.grid .g-start-3{grid-column-start:3}.grid .g-start-4{grid-column-start:4}.grid .g-start-5{grid-column-start:5}.grid .g-start-6{grid-column-start:6}.grid .g-start-7{grid-column-start:7}.grid .g-start-8{grid-column-start:8}.grid .g-start-9{grid-column-start:9}.grid .g-start-10{grid-column-start:10}.grid .g-start-11{grid-column-start:11}@media(min-width: 576px){.grid .g-col-sm-1{grid-column:auto/span 1}.grid .g-col-sm-2{grid-column:auto/span 2}.grid .g-col-sm-3{grid-column:auto/span 3}.grid .g-col-sm-4{grid-column:auto/span 4}.grid .g-col-sm-5{grid-column:auto/span 5}.grid .g-col-sm-6{grid-column:auto/span 6}.grid .g-col-sm-7{grid-column:auto/span 7}.grid .g-col-sm-8{grid-column:auto/span 8}.grid .g-col-sm-9{grid-column:auto/span 9}.grid .g-col-sm-10{grid-column:auto/span 10}.grid .g-col-sm-11{grid-column:auto/span 11}.grid .g-col-sm-12{grid-column:auto/span 12}.grid .g-start-sm-1{grid-column-start:1}.grid .g-start-sm-2{grid-column-start:2}.grid .g-start-sm-3{grid-column-start:3}.grid .g-start-sm-4{grid-column-start:4}.grid .g-start-sm-5{grid-column-start:5}.grid .g-start-sm-6{grid-column-start:6}.grid .g-start-sm-7{grid-column-start:7}.grid .g-start-sm-8{grid-column-start:8}.grid .g-start-sm-9{grid-column-start:9}.grid .g-start-sm-10{grid-column-start:10}.grid .g-start-sm-11{grid-column-start:11}}@media(min-width: 768px){.grid .g-col-md-1{grid-column:auto/span 1}.grid .g-col-md-2{grid-column:auto/span 2}.grid .g-col-md-3{grid-column:auto/span 3}.grid .g-col-md-4{grid-column:auto/span 4}.grid .g-col-md-5{grid-column:auto/span 5}.grid .g-col-md-6{grid-column:auto/span 6}.grid .g-col-md-7{grid-column:auto/span 7}.grid .g-col-md-8{grid-column:auto/span 8}.grid .g-col-md-9{grid-column:auto/span 9}.grid .g-col-md-10{grid-column:auto/span 10}.grid .g-col-md-11{grid-column:auto/span 11}.grid .g-col-md-12{grid-column:auto/span 12}.grid .g-start-md-1{grid-column-start:1}.grid .g-start-md-2{grid-column-start:2}.grid .g-start-md-3{grid-column-start:3}.grid .g-start-md-4{grid-column-start:4}.grid .g-start-md-5{grid-column-start:5}.grid .g-start-md-6{grid-column-start:6}.grid .g-start-md-7{grid-column-start:7}.grid .g-start-md-8{grid-column-start:8}.grid .g-start-md-9{grid-column-start:9}.grid .g-start-md-10{grid-column-start:10}.grid .g-start-md-11{grid-column-start:11}}@media(min-width: 992px){.grid .g-col-lg-1{grid-column:auto/span 1}.grid .g-col-lg-2{grid-column:auto/span 2}.grid .g-col-lg-3{grid-column:auto/span 3}.grid .g-col-lg-4{grid-column:auto/span 4}.grid .g-col-lg-5{grid-column:auto/span 5}.grid .g-col-lg-6{grid-column:auto/span 6}.grid .g-col-lg-7{grid-column:auto/span 7}.grid .g-col-lg-8{grid-column:auto/span 8}.grid .g-col-lg-9{grid-column:auto/span 9}.grid .g-col-lg-10{grid-column:auto/span 10}.grid .g-col-lg-11{grid-column:auto/span 11}.grid .g-col-lg-12{grid-column:auto/span 12}.grid .g-start-lg-1{grid-column-start:1}.grid .g-start-lg-2{grid-column-start:2}.grid .g-start-lg-3{grid-column-start:3}.grid .g-start-lg-4{grid-column-start:4}.grid .g-start-lg-5{grid-column-start:5}.grid .g-start-lg-6{grid-column-start:6}.grid .g-start-lg-7{grid-column-start:7}.grid .g-start-lg-8{grid-column-start:8}.grid .g-start-lg-9{grid-column-start:9}.grid .g-start-lg-10{grid-column-start:10}.grid .g-start-lg-11{grid-column-start:11}}@media(min-width: 1200px){.grid .g-col-xl-1{grid-column:auto/span 1}.grid .g-col-xl-2{grid-column:auto/span 2}.grid .g-col-xl-3{grid-column:auto/span 3}.grid .g-col-xl-4{grid-column:auto/span 4}.grid .g-col-xl-5{grid-column:auto/span 5}.grid .g-col-xl-6{grid-column:auto/span 6}.grid .g-col-xl-7{grid-column:auto/span 7}.grid .g-col-xl-8{grid-column:auto/span 8}.grid .g-col-xl-9{grid-column:auto/span 9}.grid .g-col-xl-10{grid-column:auto/span 10}.grid .g-col-xl-11{grid-column:auto/span 11}.grid .g-col-xl-12{grid-column:auto/span 12}.grid .g-start-xl-1{grid-column-start:1}.grid .g-start-xl-2{grid-column-start:2}.grid .g-start-xl-3{grid-column-start:3}.grid .g-start-xl-4{grid-column-start:4}.grid .g-start-xl-5{grid-column-start:5}.grid .g-start-xl-6{grid-column-start:6}.grid .g-start-xl-7{grid-column-start:7}.grid .g-start-xl-8{grid-column-start:8}.grid .g-start-xl-9{grid-column-start:9}.grid .g-start-xl-10{grid-column-start:10}.grid .g-start-xl-11{grid-column-start:11}}@media(min-width: 1400px){.grid .g-col-xxl-1{grid-column:auto/span 1}.grid .g-col-xxl-2{grid-column:auto/span 2}.grid .g-col-xxl-3{grid-column:auto/span 3}.grid .g-col-xxl-4{grid-column:auto/span 4}.grid .g-col-xxl-5{grid-column:auto/span 5}.grid .g-col-xxl-6{grid-column:auto/span 6}.grid .g-col-xxl-7{grid-column:auto/span 7}.grid .g-col-xxl-8{grid-column:auto/span 8}.grid .g-col-xxl-9{grid-column:auto/span 9}.grid .g-col-xxl-10{grid-column:auto/span 10}.grid .g-col-xxl-11{grid-column:auto/span 11}.grid .g-col-xxl-12{grid-column:auto/span 12}.grid .g-start-xxl-1{grid-column-start:1}.grid .g-start-xxl-2{grid-column-start:2}.grid .g-start-xxl-3{grid-column-start:3}.grid .g-start-xxl-4{grid-column-start:4}.grid .g-start-xxl-5{grid-column-start:5}.grid .g-start-xxl-6{grid-column-start:6}.grid .g-start-xxl-7{grid-column-start:7}.grid .g-start-xxl-8{grid-column-start:8}.grid .g-start-xxl-9{grid-column-start:9}.grid .g-start-xxl-10{grid-column-start:10}.grid .g-start-xxl-11{grid-column-start:11}}.table{--bs-table-color-type: initial;--bs-table-bg-type: initial;--bs-table-color-state: initial;--bs-table-bg-state: initial;--bs-table-color: #343a40;--bs-table-bg: #FEFBF2;--bs-table-border-color: #dee2e6;--bs-table-accent-bg: transparent;--bs-table-striped-color: #343a40;--bs-table-striped-bg: rgba(0, 0, 0, 0.05);--bs-table-active-color: #343a40;--bs-table-active-bg: rgba(0, 0, 0, 0.1);--bs-table-hover-color: #343a40;--bs-table-hover-bg: rgba(0, 0, 0, 0.075);width:100%;margin-bottom:1rem;vertical-align:top;border-color:var(--bs-table-border-color)}.table>:not(caption)>*>*{padding:.5rem .5rem;color:var(--bs-table-color-state, var(--bs-table-color-type, var(--bs-table-color)));background-color:var(--bs-table-bg);border-bottom-width:1px;box-shadow:inset 0 0 0 9999px var(--bs-table-bg-state, var(--bs-table-bg-type, var(--bs-table-accent-bg)))}.table>tbody{vertical-align:inherit}.table>thead{vertical-align:bottom}.table-group-divider{border-top:calc(1px*2) solid #999b99}.caption-top{caption-side:top}.table-sm>:not(caption)>*>*{padding:.25rem .25rem}.table-bordered>:not(caption)>*{border-width:1px 0}.table-bordered>:not(caption)>*>*{border-width:0 1px}.table-borderless>:not(caption)>*>*{border-bottom-width:0}.table-borderless>:not(:first-child){border-top-width:0}.table-striped>tbody>tr:nth-of-type(odd)>*{--bs-table-color-type: var(--bs-table-striped-color);--bs-table-bg-type: var(--bs-table-striped-bg)}.table-striped-columns>:not(caption)>tr>:nth-child(even){--bs-table-color-type: var(--bs-table-striped-color);--bs-table-bg-type: var(--bs-table-striped-bg)}.table-active{--bs-table-color-state: var(--bs-table-active-color);--bs-table-bg-state: var(--bs-table-active-bg)}.table-hover>tbody>tr:hover>*{--bs-table-color-state: var(--bs-table-hover-color);--bs-table-bg-state: var(--bs-table-hover-bg)}.table-primary{--bs-table-color: #000;--bs-table-bg: #d4e6f9;--bs-table-border-color: #bfcfe0;--bs-table-striped-bg: #c9dbed;--bs-table-striped-color: #000;--bs-table-active-bg: #bfcfe0;--bs-table-active-color: #000;--bs-table-hover-bg: #c4d5e6;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-secondary{--bs-table-color: #000;--bs-table-bg: #d6d8d9;--bs-table-border-color: #c1c2c3;--bs-table-striped-bg: #cbcdce;--bs-table-striped-color: #000;--bs-table-active-bg: #c1c2c3;--bs-table-active-color: #000;--bs-table-hover-bg: #c6c8c9;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-success{--bs-table-color: #000;--bs-table-bg: #d9f0d1;--bs-table-border-color: #c3d8bc;--bs-table-striped-bg: #cee4c7;--bs-table-striped-color: #000;--bs-table-active-bg: #c3d8bc;--bs-table-active-color: #000;--bs-table-hover-bg: #c9dec1;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-info{--bs-table-color: #000;--bs-table-bg: #ebddf1;--bs-table-border-color: #d4c7d9;--bs-table-striped-bg: #dfd2e5;--bs-table-striped-color: #000;--bs-table-active-bg: #d4c7d9;--bs-table-active-color: #000;--bs-table-hover-bg: #d9ccdf;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-warning{--bs-table-color: #000;--bs-table-bg: #ffe3d1;--bs-table-border-color: #e6ccbc;--bs-table-striped-bg: #f2d8c7;--bs-table-striped-color: #000;--bs-table-active-bg: #e6ccbc;--bs-table-active-color: #000;--bs-table-hover-bg: #ecd2c1;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-danger{--bs-table-color: #000;--bs-table-bg: #ffccd7;--bs-table-border-color: #e6b8c2;--bs-table-striped-bg: #f2c2cc;--bs-table-striped-color: #000;--bs-table-active-bg: #e6b8c2;--bs-table-active-color: #000;--bs-table-hover-bg: #ecbdc7;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-light{--bs-table-color: #000;--bs-table-bg: #f8f9fa;--bs-table-border-color: #dfe0e1;--bs-table-striped-bg: #ecedee;--bs-table-striped-color: #000;--bs-table-active-bg: #dfe0e1;--bs-table-active-color: #000;--bs-table-hover-bg: #e5e6e7;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-dark{--bs-table-color: #fff;--bs-table-bg: #343a40;--bs-table-border-color: #484e53;--bs-table-striped-bg: #3e444a;--bs-table-striped-color: #fff;--bs-table-active-bg: #484e53;--bs-table-active-color: #fff;--bs-table-hover-bg: #43494e;--bs-table-hover-color: #fff;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-responsive{overflow-x:auto;-webkit-overflow-scrolling:touch}@media(max-width: 575.98px){.table-responsive-sm{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 767.98px){.table-responsive-md{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 991.98px){.table-responsive-lg{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 1199.98px){.table-responsive-xl{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 1399.98px){.table-responsive-xxl{overflow-x:auto;-webkit-overflow-scrolling:touch}}.form-label,.shiny-input-container .control-label{margin-bottom:.5rem}.col-form-label{padding-top:calc(0.375rem + 1px);padding-bottom:calc(0.375rem + 1px);margin-bottom:0;font-size:inherit;line-height:1.5}.col-form-label-lg{padding-top:calc(0.5rem + 1px);padding-bottom:calc(0.5rem + 1px);font-size:1.25rem}.col-form-label-sm{padding-top:calc(0.25rem + 1px);padding-bottom:calc(0.25rem + 1px);font-size:0.875rem}.form-text{margin-top:.25rem;font-size:0.875em;color:rgba(52,58,64,.75)}.form-control{display:block;width:100%;padding:.375rem .75rem;font-size:1rem;font-weight:400;line-height:1.5;color:#343a40;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#fefbf2;background-clip:padding-box;border:1px solid #dee2e6;border-radius:0;transition:border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-control{transition:none}}.form-control[type=file]{overflow:hidden}.form-control[type=file]:not(:disabled):not([readonly]){cursor:pointer}.form-control:focus{color:#343a40;background-color:#fefbf2;border-color:#93c0f1;outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.form-control::-webkit-date-and-time-value{min-width:85px;height:1.5em;margin:0}.form-control::-webkit-datetime-edit{display:block;padding:0}.form-control::placeholder{color:rgba(52,58,64,.75);opacity:1}.form-control:disabled{background-color:#e9ecef;opacity:1}.form-control::file-selector-button{padding:.375rem .75rem;margin:-0.375rem -0.75rem;margin-inline-end:.75rem;color:#343a40;background-color:#f8f9fa;pointer-events:none;border-color:inherit;border-style:solid;border-width:0;border-inline-end-width:1px;border-radius:0;transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-control::file-selector-button{transition:none}}.form-control:hover:not(:disabled):not([readonly])::file-selector-button{background-color:#e9ecef}.form-control-plaintext{display:block;width:100%;padding:.375rem 0;margin-bottom:0;line-height:1.5;color:#343a40;background-color:rgba(0,0,0,0);border:solid rgba(0,0,0,0);border-width:1px 0}.form-control-plaintext:focus{outline:0}.form-control-plaintext.form-control-sm,.form-control-plaintext.form-control-lg{padding-right:0;padding-left:0}.form-control-sm{min-height:calc(1.5em + 0.5rem + calc(1px * 2));padding:.25rem .5rem;font-size:0.875rem}.form-control-sm::file-selector-button{padding:.25rem .5rem;margin:-0.25rem -0.5rem;margin-inline-end:.5rem}.form-control-lg{min-height:calc(1.5em + 1rem + calc(1px * 2));padding:.5rem 1rem;font-size:1.25rem}.form-control-lg::file-selector-button{padding:.5rem 1rem;margin:-0.5rem -1rem;margin-inline-end:1rem}textarea.form-control{min-height:calc(1.5em + 0.75rem + calc(1px * 2))}textarea.form-control-sm{min-height:calc(1.5em + 0.5rem + calc(1px * 2))}textarea.form-control-lg{min-height:calc(1.5em + 1rem + calc(1px * 2))}.form-control-color{width:3rem;height:calc(1.5em + 0.75rem + calc(1px * 2));padding:.375rem}.form-control-color:not(:disabled):not([readonly]){cursor:pointer}.form-control-color::-moz-color-swatch{border:0 !important}.form-control-color::-webkit-color-swatch{border:0 !important}.form-control-color.form-control-sm{height:calc(1.5em + 0.5rem + calc(1px * 2))}.form-control-color.form-control-lg{height:calc(1.5em + 1rem + calc(1px * 2))}.form-select{--bs-form-select-bg-img: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16'%3e%3cpath fill='none' stroke='%23343a40' stroke-linecap='round' stroke-linejoin='round' stroke-width='2' d='m2 5 6 6 6-6'/%3e%3c/svg%3e");display:block;width:100%;padding:.375rem 2.25rem .375rem .75rem;font-size:1rem;font-weight:400;line-height:1.5;color:#343a40;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#fefbf2;background-image:var(--bs-form-select-bg-img),var(--bs-form-select-bg-icon, none);background-repeat:no-repeat;background-position:right .75rem center;background-size:16px 12px;border:1px solid #dee2e6;border-radius:0;transition:border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-select{transition:none}}.form-select:focus{border-color:#93c0f1;outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.form-select[multiple],.form-select[size]:not([size="1"]){padding-right:.75rem;background-image:none}.form-select:disabled{background-color:#e9ecef}.form-select:-moz-focusring{color:rgba(0,0,0,0);text-shadow:0 0 0 #343a40}.form-select-sm{padding-top:.25rem;padding-bottom:.25rem;padding-left:.5rem;font-size:0.875rem}.form-select-lg{padding-top:.5rem;padding-bottom:.5rem;padding-left:1rem;font-size:1.25rem}[data-bs-theme=dark] .form-select{--bs-form-select-bg-img: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16'%3e%3cpath fill='none' stroke='%23dee2e6' stroke-linecap='round' stroke-linejoin='round' stroke-width='2' d='m2 5 6 6 6-6'/%3e%3c/svg%3e")}.form-check,.shiny-input-container .checkbox,.shiny-input-container .radio{display:block;min-height:1.5rem;padding-left:0;margin-bottom:.125rem}.form-check .form-check-input,.form-check .shiny-input-container .checkbox input,.form-check .shiny-input-container .radio input,.shiny-input-container .checkbox .form-check-input,.shiny-input-container .checkbox .shiny-input-container .checkbox input,.shiny-input-container .checkbox .shiny-input-container .radio input,.shiny-input-container .radio .form-check-input,.shiny-input-container .radio .shiny-input-container .checkbox input,.shiny-input-container .radio .shiny-input-container .radio input{float:left;margin-left:0}.form-check-reverse{padding-right:0;padding-left:0;text-align:right}.form-check-reverse .form-check-input{float:right;margin-right:0;margin-left:0}.form-check-input,.shiny-input-container .checkbox input,.shiny-input-container .checkbox-inline input,.shiny-input-container .radio input,.shiny-input-container .radio-inline input{--bs-form-check-bg: #FEFBF2;width:1em;height:1em;margin-top:.25em;vertical-align:top;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:var(--bs-form-check-bg);background-image:var(--bs-form-check-bg-image);background-repeat:no-repeat;background-position:center;background-size:contain;border:1px solid #dee2e6;print-color-adjust:exact}.form-check-input[type=radio],.shiny-input-container .checkbox input[type=radio],.shiny-input-container .checkbox-inline input[type=radio],.shiny-input-container .radio input[type=radio],.shiny-input-container .radio-inline input[type=radio]{border-radius:50%}.form-check-input:active,.shiny-input-container .checkbox input:active,.shiny-input-container .checkbox-inline input:active,.shiny-input-container .radio input:active,.shiny-input-container .radio-inline input:active{filter:brightness(90%)}.form-check-input:focus,.shiny-input-container .checkbox input:focus,.shiny-input-container .checkbox-inline input:focus,.shiny-input-container .radio input:focus,.shiny-input-container .radio-inline input:focus{border-color:#93c0f1;outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.form-check-input:checked,.shiny-input-container .checkbox input:checked,.shiny-input-container .checkbox-inline input:checked,.shiny-input-container .radio input:checked,.shiny-input-container .radio-inline input:checked{background-color:#2780e3;border-color:#2780e3}.form-check-input:checked[type=checkbox],.shiny-input-container .checkbox input:checked[type=checkbox],.shiny-input-container .checkbox-inline input:checked[type=checkbox],.shiny-input-container .radio input:checked[type=checkbox],.shiny-input-container .radio-inline input:checked[type=checkbox]{--bs-form-check-bg-image: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 20 20'%3e%3cpath fill='none' stroke='%23fff' stroke-linecap='round' stroke-linejoin='round' stroke-width='3' d='m6 10 3 3 6-6'/%3e%3c/svg%3e")}.form-check-input:checked[type=radio],.shiny-input-container .checkbox input:checked[type=radio],.shiny-input-container .checkbox-inline input:checked[type=radio],.shiny-input-container .radio input:checked[type=radio],.shiny-input-container .radio-inline input:checked[type=radio]{--bs-form-check-bg-image: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='2' fill='%23fff'/%3e%3c/svg%3e")}.form-check-input[type=checkbox]:indeterminate,.shiny-input-container .checkbox input[type=checkbox]:indeterminate,.shiny-input-container .checkbox-inline input[type=checkbox]:indeterminate,.shiny-input-container .radio input[type=checkbox]:indeterminate,.shiny-input-container .radio-inline input[type=checkbox]:indeterminate{background-color:#2780e3;border-color:#2780e3;--bs-form-check-bg-image: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 20 20'%3e%3cpath fill='none' stroke='%23fff' stroke-linecap='round' stroke-linejoin='round' stroke-width='3' d='M6 10h8'/%3e%3c/svg%3e")}.form-check-input:disabled,.shiny-input-container .checkbox input:disabled,.shiny-input-container .checkbox-inline input:disabled,.shiny-input-container .radio input:disabled,.shiny-input-container .radio-inline input:disabled{pointer-events:none;filter:none;opacity:.5}.form-check-input[disabled]~.form-check-label,.form-check-input[disabled]~span,.form-check-input:disabled~.form-check-label,.form-check-input:disabled~span,.shiny-input-container .checkbox input[disabled]~.form-check-label,.shiny-input-container .checkbox input[disabled]~span,.shiny-input-container .checkbox input:disabled~.form-check-label,.shiny-input-container .checkbox input:disabled~span,.shiny-input-container .checkbox-inline input[disabled]~.form-check-label,.shiny-input-container .checkbox-inline input[disabled]~span,.shiny-input-container .checkbox-inline input:disabled~.form-check-label,.shiny-input-container .checkbox-inline input:disabled~span,.shiny-input-container .radio input[disabled]~.form-check-label,.shiny-input-container .radio input[disabled]~span,.shiny-input-container .radio input:disabled~.form-check-label,.shiny-input-container .radio input:disabled~span,.shiny-input-container .radio-inline input[disabled]~.form-check-label,.shiny-input-container .radio-inline input[disabled]~span,.shiny-input-container .radio-inline input:disabled~.form-check-label,.shiny-input-container .radio-inline input:disabled~span{cursor:default;opacity:.5}.form-check-label,.shiny-input-container .checkbox label,.shiny-input-container .checkbox-inline label,.shiny-input-container .radio label,.shiny-input-container .radio-inline label{cursor:pointer}.form-switch{padding-left:2.5em}.form-switch .form-check-input{--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='rgba%280, 0, 0, 0.25%29'/%3e%3c/svg%3e");width:2em;margin-left:-2.5em;background-image:var(--bs-form-switch-bg);background-position:left center;transition:background-position .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-switch .form-check-input{transition:none}}.form-switch .form-check-input:focus{--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='%2393c0f1'/%3e%3c/svg%3e")}.form-switch .form-check-input:checked{background-position:right center;--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='%23fff'/%3e%3c/svg%3e")}.form-switch.form-check-reverse{padding-right:2.5em;padding-left:0}.form-switch.form-check-reverse .form-check-input{margin-right:-2.5em;margin-left:0}.form-check-inline{display:inline-block;margin-right:1rem}.btn-check{position:absolute;clip:rect(0, 0, 0, 0);pointer-events:none}.btn-check[disabled]+.btn,.btn-check:disabled+.btn{pointer-events:none;filter:none;opacity:.65}[data-bs-theme=dark] .form-switch .form-check-input:not(:checked):not(:focus){--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='rgba%28255, 255, 255, 0.25%29'/%3e%3c/svg%3e")}.form-range{width:100%;height:1.5rem;padding:0;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:rgba(0,0,0,0)}.form-range:focus{outline:0}.form-range:focus::-webkit-slider-thumb{box-shadow:0 0 0 1px #fefbf2,0 0 0 .25rem rgba(39,128,227,.25)}.form-range:focus::-moz-range-thumb{box-shadow:0 0 0 1px #fefbf2,0 0 0 .25rem rgba(39,128,227,.25)}.form-range::-moz-focus-outer{border:0}.form-range::-webkit-slider-thumb{width:1rem;height:1rem;margin-top:-0.25rem;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#2780e3;border:0;transition:background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-range::-webkit-slider-thumb{transition:none}}.form-range::-webkit-slider-thumb:active{background-color:#bed9f7}.form-range::-webkit-slider-runnable-track{width:100%;height:.5rem;color:rgba(0,0,0,0);cursor:pointer;background-color:#f8f9fa;border-color:rgba(0,0,0,0)}.form-range::-moz-range-thumb{width:1rem;height:1rem;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#2780e3;border:0;transition:background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-range::-moz-range-thumb{transition:none}}.form-range::-moz-range-thumb:active{background-color:#bed9f7}.form-range::-moz-range-track{width:100%;height:.5rem;color:rgba(0,0,0,0);cursor:pointer;background-color:#f8f9fa;border-color:rgba(0,0,0,0)}.form-range:disabled{pointer-events:none}.form-range:disabled::-webkit-slider-thumb{background-color:rgba(52,58,64,.75)}.form-range:disabled::-moz-range-thumb{background-color:rgba(52,58,64,.75)}.form-floating{position:relative}.form-floating>.form-control,.form-floating>.form-control-plaintext,.form-floating>.form-select{height:calc(3.5rem + calc(1px * 2));min-height:calc(3.5rem + calc(1px * 2));line-height:1.25}.form-floating>label{position:absolute;top:0;left:0;z-index:2;height:100%;padding:1rem .75rem;overflow:hidden;text-align:start;text-overflow:ellipsis;white-space:nowrap;pointer-events:none;border:1px solid rgba(0,0,0,0);transform-origin:0 0;transition:opacity .1s ease-in-out,transform .1s ease-in-out}@media(prefers-reduced-motion: reduce){.form-floating>label{transition:none}}.form-floating>.form-control,.form-floating>.form-control-plaintext{padding:1rem .75rem}.form-floating>.form-control::placeholder,.form-floating>.form-control-plaintext::placeholder{color:rgba(0,0,0,0)}.form-floating>.form-control:focus,.form-floating>.form-control:not(:placeholder-shown),.form-floating>.form-control-plaintext:focus,.form-floating>.form-control-plaintext:not(:placeholder-shown){padding-top:1.625rem;padding-bottom:.625rem}.form-floating>.form-control:-webkit-autofill,.form-floating>.form-control-plaintext:-webkit-autofill{padding-top:1.625rem;padding-bottom:.625rem}.form-floating>.form-select{padding-top:1.625rem;padding-bottom:.625rem}.form-floating>.form-control:focus~label,.form-floating>.form-control:not(:placeholder-shown)~label,.form-floating>.form-control-plaintext~label,.form-floating>.form-select~label{color:rgba(var(--bs-body-color-rgb), 0.65);transform:scale(0.85) translateY(-0.5rem) translateX(0.15rem)}.form-floating>.form-control:focus~label::after,.form-floating>.form-control:not(:placeholder-shown)~label::after,.form-floating>.form-control-plaintext~label::after,.form-floating>.form-select~label::after{position:absolute;inset:1rem .375rem;z-index:-1;height:1.5em;content:"";background-color:#fefbf2}.form-floating>.form-control:-webkit-autofill~label{color:rgba(var(--bs-body-color-rgb), 0.65);transform:scale(0.85) translateY(-0.5rem) translateX(0.15rem)}.form-floating>.form-control-plaintext~label{border-width:1px 0}.form-floating>:disabled~label,.form-floating>.form-control:disabled~label{color:#6c757d}.form-floating>:disabled~label::after,.form-floating>.form-control:disabled~label::after{background-color:#e9ecef}.input-group{position:relative;display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;align-items:stretch;-webkit-align-items:stretch;width:100%}.input-group>.form-control,.input-group>.form-select,.input-group>.form-floating{position:relative;flex:1 1 auto;-webkit-flex:1 1 auto;width:1%;min-width:0}.input-group>.form-control:focus,.input-group>.form-select:focus,.input-group>.form-floating:focus-within{z-index:5}.input-group .btn{position:relative;z-index:2}.input-group .btn:focus{z-index:5}.input-group-text{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;padding:.375rem .75rem;font-size:1rem;font-weight:400;line-height:1.5;color:#343a40;text-align:center;white-space:nowrap;background-color:#f8f9fa;border:1px solid #dee2e6}.input-group-lg>.form-control,.input-group-lg>.form-select,.input-group-lg>.input-group-text,.input-group-lg>.btn{padding:.5rem 1rem;font-size:1.25rem}.input-group-sm>.form-control,.input-group-sm>.form-select,.input-group-sm>.input-group-text,.input-group-sm>.btn{padding:.25rem .5rem;font-size:0.875rem}.input-group-lg>.form-select,.input-group-sm>.form-select{padding-right:3rem}.input-group>:not(:first-child):not(.dropdown-menu):not(.valid-tooltip):not(.valid-feedback):not(.invalid-tooltip):not(.invalid-feedback){margin-left:calc(1px*-1)}.valid-feedback{display:none;width:100%;margin-top:.25rem;font-size:0.875em;color:#3fb618}.valid-tooltip{position:absolute;top:100%;z-index:5;display:none;max-width:100%;padding:.25rem .5rem;margin-top:.1rem;font-size:0.875rem;color:#fff;background-color:#3fb618}.was-validated :valid~.valid-feedback,.was-validated :valid~.valid-tooltip,.is-valid~.valid-feedback,.is-valid~.valid-tooltip{display:block}.was-validated .form-control:valid,.form-control.is-valid{border-color:#3fb618;padding-right:calc(1.5em + 0.75rem);background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 8 8'%3e%3cpath fill='%233fb618' d='M2.3 6.73.6 4.53c-.4-1.04.46-1.4 1.1-.8l1.1 1.4 3.4-3.8c.6-.63 1.6-.27 1.2.7l-4 4.6c-.43.5-.8.4-1.1.1z'/%3e%3c/svg%3e");background-repeat:no-repeat;background-position:right calc(0.375em + 0.1875rem) center;background-size:calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-control:valid:focus,.form-control.is-valid:focus{border-color:#3fb618;box-shadow:0 0 0 .25rem rgba(63,182,24,.25)}.was-validated textarea.form-control:valid,textarea.form-control.is-valid{padding-right:calc(1.5em + 0.75rem);background-position:top calc(0.375em + 0.1875rem) right calc(0.375em + 0.1875rem)}.was-validated .form-select:valid,.form-select.is-valid{border-color:#3fb618}.was-validated .form-select:valid:not([multiple]):not([size]),.was-validated .form-select:valid:not([multiple])[size="1"],.form-select.is-valid:not([multiple]):not([size]),.form-select.is-valid:not([multiple])[size="1"]{--bs-form-select-bg-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 8 8'%3e%3cpath fill='%233fb618' d='M2.3 6.73.6 4.53c-.4-1.04.46-1.4 1.1-.8l1.1 1.4 3.4-3.8c.6-.63 1.6-.27 1.2.7l-4 4.6c-.43.5-.8.4-1.1.1z'/%3e%3c/svg%3e");padding-right:4.125rem;background-position:right .75rem center,center right 2.25rem;background-size:16px 12px,calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-select:valid:focus,.form-select.is-valid:focus{border-color:#3fb618;box-shadow:0 0 0 .25rem rgba(63,182,24,.25)}.was-validated .form-control-color:valid,.form-control-color.is-valid{width:calc(3rem + calc(1.5em + 0.75rem))}.was-validated .form-check-input:valid,.form-check-input.is-valid{border-color:#3fb618}.was-validated .form-check-input:valid:checked,.form-check-input.is-valid:checked{background-color:#3fb618}.was-validated .form-check-input:valid:focus,.form-check-input.is-valid:focus{box-shadow:0 0 0 .25rem rgba(63,182,24,.25)}.was-validated .form-check-input:valid~.form-check-label,.form-check-input.is-valid~.form-check-label{color:#3fb618}.form-check-inline .form-check-input~.valid-feedback{margin-left:.5em}.was-validated .input-group>.form-control:not(:focus):valid,.input-group>.form-control:not(:focus).is-valid,.was-validated .input-group>.form-select:not(:focus):valid,.input-group>.form-select:not(:focus).is-valid,.was-validated .input-group>.form-floating:not(:focus-within):valid,.input-group>.form-floating:not(:focus-within).is-valid{z-index:3}.invalid-feedback{display:none;width:100%;margin-top:.25rem;font-size:0.875em;color:#ff0039}.invalid-tooltip{position:absolute;top:100%;z-index:5;display:none;max-width:100%;padding:.25rem .5rem;margin-top:.1rem;font-size:0.875rem;color:#fff;background-color:#ff0039}.was-validated :invalid~.invalid-feedback,.was-validated :invalid~.invalid-tooltip,.is-invalid~.invalid-feedback,.is-invalid~.invalid-tooltip{display:block}.was-validated .form-control:invalid,.form-control.is-invalid{border-color:#ff0039;padding-right:calc(1.5em + 0.75rem);background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 12 12' width='12' height='12' fill='none' stroke='%23ff0039'%3e%3ccircle cx='6' cy='6' r='4.5'/%3e%3cpath stroke-linejoin='round' d='M5.8 3.6h.4L6 6.5z'/%3e%3ccircle cx='6' cy='8.2' r='.6' fill='%23ff0039' stroke='none'/%3e%3c/svg%3e");background-repeat:no-repeat;background-position:right calc(0.375em + 0.1875rem) center;background-size:calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-control:invalid:focus,.form-control.is-invalid:focus{border-color:#ff0039;box-shadow:0 0 0 .25rem rgba(255,0,57,.25)}.was-validated textarea.form-control:invalid,textarea.form-control.is-invalid{padding-right:calc(1.5em + 0.75rem);background-position:top calc(0.375em + 0.1875rem) right calc(0.375em + 0.1875rem)}.was-validated .form-select:invalid,.form-select.is-invalid{border-color:#ff0039}.was-validated .form-select:invalid:not([multiple]):not([size]),.was-validated .form-select:invalid:not([multiple])[size="1"],.form-select.is-invalid:not([multiple]):not([size]),.form-select.is-invalid:not([multiple])[size="1"]{--bs-form-select-bg-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 12 12' width='12' height='12' fill='none' stroke='%23ff0039'%3e%3ccircle cx='6' cy='6' r='4.5'/%3e%3cpath stroke-linejoin='round' d='M5.8 3.6h.4L6 6.5z'/%3e%3ccircle cx='6' cy='8.2' r='.6' fill='%23ff0039' stroke='none'/%3e%3c/svg%3e");padding-right:4.125rem;background-position:right .75rem center,center right 2.25rem;background-size:16px 12px,calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-select:invalid:focus,.form-select.is-invalid:focus{border-color:#ff0039;box-shadow:0 0 0 .25rem rgba(255,0,57,.25)}.was-validated .form-control-color:invalid,.form-control-color.is-invalid{width:calc(3rem + calc(1.5em + 0.75rem))}.was-validated .form-check-input:invalid,.form-check-input.is-invalid{border-color:#ff0039}.was-validated .form-check-input:invalid:checked,.form-check-input.is-invalid:checked{background-color:#ff0039}.was-validated .form-check-input:invalid:focus,.form-check-input.is-invalid:focus{box-shadow:0 0 0 .25rem rgba(255,0,57,.25)}.was-validated .form-check-input:invalid~.form-check-label,.form-check-input.is-invalid~.form-check-label{color:#ff0039}.form-check-inline .form-check-input~.invalid-feedback{margin-left:.5em}.was-validated .input-group>.form-control:not(:focus):invalid,.input-group>.form-control:not(:focus).is-invalid,.was-validated .input-group>.form-select:not(:focus):invalid,.input-group>.form-select:not(:focus).is-invalid,.was-validated .input-group>.form-floating:not(:focus-within):invalid,.input-group>.form-floating:not(:focus-within).is-invalid{z-index:4}.btn{--bs-btn-padding-x: 0.75rem;--bs-btn-padding-y: 0.375rem;--bs-btn-font-family: ;--bs-btn-font-size:1rem;--bs-btn-font-weight: 400;--bs-btn-line-height: 1.5;--bs-btn-color: #343a40;--bs-btn-bg: transparent;--bs-btn-border-width: 1px;--bs-btn-border-color: transparent;--bs-btn-border-radius: 0.25rem;--bs-btn-hover-border-color: transparent;--bs-btn-box-shadow: inset 0 1px 0 rgba(255, 255, 255, 0.15), 0 1px 1px rgba(0, 0, 0, 0.075);--bs-btn-disabled-opacity: 0.65;--bs-btn-focus-box-shadow: 0 0 0 0.25rem rgba(var(--bs-btn-focus-shadow-rgb), .5);display:inline-block;padding:var(--bs-btn-padding-y) var(--bs-btn-padding-x);font-family:var(--bs-btn-font-family);font-size:var(--bs-btn-font-size);font-weight:var(--bs-btn-font-weight);line-height:var(--bs-btn-line-height);color:var(--bs-btn-color);text-align:center;text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;vertical-align:middle;cursor:pointer;user-select:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;-o-user-select:none;border:var(--bs-btn-border-width) solid var(--bs-btn-border-color);background-color:var(--bs-btn-bg);transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.btn{transition:none}}.btn:hover{color:var(--bs-btn-hover-color);background-color:var(--bs-btn-hover-bg);border-color:var(--bs-btn-hover-border-color)}.btn-check+.btn:hover{color:var(--bs-btn-color);background-color:var(--bs-btn-bg);border-color:var(--bs-btn-border-color)}.btn:focus-visible{color:var(--bs-btn-hover-color);background-color:var(--bs-btn-hover-bg);border-color:var(--bs-btn-hover-border-color);outline:0;box-shadow:var(--bs-btn-focus-box-shadow)}.btn-check:focus-visible+.btn{border-color:var(--bs-btn-hover-border-color);outline:0;box-shadow:var(--bs-btn-focus-box-shadow)}.btn-check:checked+.btn,:not(.btn-check)+.btn:active,.btn:first-child:active,.btn.active,.btn.show{color:var(--bs-btn-active-color);background-color:var(--bs-btn-active-bg);border-color:var(--bs-btn-active-border-color)}.btn-check:checked+.btn:focus-visible,:not(.btn-check)+.btn:active:focus-visible,.btn:first-child:active:focus-visible,.btn.active:focus-visible,.btn.show:focus-visible{box-shadow:var(--bs-btn-focus-box-shadow)}.btn:disabled,.btn.disabled,fieldset:disabled .btn{color:var(--bs-btn-disabled-color);pointer-events:none;background-color:var(--bs-btn-disabled-bg);border-color:var(--bs-btn-disabled-border-color);opacity:var(--bs-btn-disabled-opacity)}.btn-default{--bs-btn-color: #fff;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #2c3136;--bs-btn-hover-border-color: #2a2e33;--bs-btn-focus-shadow-rgb: 82, 88, 93;--bs-btn-active-color: #fff;--bs-btn-active-bg: #2a2e33;--bs-btn-active-border-color: #272c30;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}.btn-primary{--bs-btn-color: #fff;--bs-btn-bg: #2780e3;--bs-btn-border-color: #2780e3;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #216dc1;--bs-btn-hover-border-color: #1f66b6;--bs-btn-focus-shadow-rgb: 71, 147, 231;--bs-btn-active-color: #fff;--bs-btn-active-bg: #1f66b6;--bs-btn-active-border-color: #1d60aa;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #2780e3;--bs-btn-disabled-border-color: #2780e3}.btn-secondary{--bs-btn-color: #fff;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #2c3136;--bs-btn-hover-border-color: #2a2e33;--bs-btn-focus-shadow-rgb: 82, 88, 93;--bs-btn-active-color: #fff;--bs-btn-active-bg: #2a2e33;--bs-btn-active-border-color: #272c30;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}.btn-success{--bs-btn-color: #fff;--bs-btn-bg: #3fb618;--bs-btn-border-color: #3fb618;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #369b14;--bs-btn-hover-border-color: #329213;--bs-btn-focus-shadow-rgb: 92, 193, 59;--bs-btn-active-color: #fff;--bs-btn-active-bg: #329213;--bs-btn-active-border-color: #2f8912;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #3fb618;--bs-btn-disabled-border-color: #3fb618}.btn-info{--bs-btn-color: #fff;--bs-btn-bg: #9954bb;--bs-btn-border-color: #9954bb;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #82479f;--bs-btn-hover-border-color: #7a4396;--bs-btn-focus-shadow-rgb: 168, 110, 197;--bs-btn-active-color: #fff;--bs-btn-active-bg: #7a4396;--bs-btn-active-border-color: #733f8c;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #9954bb;--bs-btn-disabled-border-color: #9954bb}.btn-warning{--bs-btn-color: #fff;--bs-btn-bg: #ff7518;--bs-btn-border-color: #ff7518;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #d96314;--bs-btn-hover-border-color: #cc5e13;--bs-btn-focus-shadow-rgb: 255, 138, 59;--bs-btn-active-color: #fff;--bs-btn-active-bg: #cc5e13;--bs-btn-active-border-color: #bf5812;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #ff7518;--bs-btn-disabled-border-color: #ff7518}.btn-danger{--bs-btn-color: #fff;--bs-btn-bg: #ff0039;--bs-btn-border-color: #ff0039;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #d90030;--bs-btn-hover-border-color: #cc002e;--bs-btn-focus-shadow-rgb: 255, 38, 87;--bs-btn-active-color: #fff;--bs-btn-active-bg: #cc002e;--bs-btn-active-border-color: #bf002b;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #ff0039;--bs-btn-disabled-border-color: #ff0039}.btn-light{--bs-btn-color: #000;--bs-btn-bg: #f8f9fa;--bs-btn-border-color: #f8f9fa;--bs-btn-hover-color: #000;--bs-btn-hover-bg: #d3d4d5;--bs-btn-hover-border-color: #c6c7c8;--bs-btn-focus-shadow-rgb: 211, 212, 213;--bs-btn-active-color: #000;--bs-btn-active-bg: #c6c7c8;--bs-btn-active-border-color: #babbbc;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #000;--bs-btn-disabled-bg: #f8f9fa;--bs-btn-disabled-border-color: #f8f9fa}.btn-dark{--bs-btn-color: #fff;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #52585d;--bs-btn-hover-border-color: #484e53;--bs-btn-focus-shadow-rgb: 82, 88, 93;--bs-btn-active-color: #fff;--bs-btn-active-bg: #5d6166;--bs-btn-active-border-color: #484e53;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}.btn-outline-default{--bs-btn-color: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #343a40;--bs-btn-hover-border-color: #343a40;--bs-btn-focus-shadow-rgb: 52, 58, 64;--bs-btn-active-color: #fff;--bs-btn-active-bg: #343a40;--bs-btn-active-border-color: #343a40;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #343a40;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #343a40;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-primary{--bs-btn-color: #2780e3;--bs-btn-border-color: #2780e3;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #2780e3;--bs-btn-hover-border-color: #2780e3;--bs-btn-focus-shadow-rgb: 39, 128, 227;--bs-btn-active-color: #fff;--bs-btn-active-bg: #2780e3;--bs-btn-active-border-color: #2780e3;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #2780e3;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #2780e3;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-secondary{--bs-btn-color: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #343a40;--bs-btn-hover-border-color: #343a40;--bs-btn-focus-shadow-rgb: 52, 58, 64;--bs-btn-active-color: #fff;--bs-btn-active-bg: #343a40;--bs-btn-active-border-color: #343a40;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #343a40;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #343a40;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-success{--bs-btn-color: #3fb618;--bs-btn-border-color: #3fb618;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #3fb618;--bs-btn-hover-border-color: #3fb618;--bs-btn-focus-shadow-rgb: 63, 182, 24;--bs-btn-active-color: #fff;--bs-btn-active-bg: #3fb618;--bs-btn-active-border-color: #3fb618;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #3fb618;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #3fb618;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-info{--bs-btn-color: #9954bb;--bs-btn-border-color: #9954bb;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #9954bb;--bs-btn-hover-border-color: #9954bb;--bs-btn-focus-shadow-rgb: 153, 84, 187;--bs-btn-active-color: #fff;--bs-btn-active-bg: #9954bb;--bs-btn-active-border-color: #9954bb;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #9954bb;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #9954bb;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-warning{--bs-btn-color: #ff7518;--bs-btn-border-color: #ff7518;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #ff7518;--bs-btn-hover-border-color: #ff7518;--bs-btn-focus-shadow-rgb: 255, 117, 24;--bs-btn-active-color: #fff;--bs-btn-active-bg: #ff7518;--bs-btn-active-border-color: #ff7518;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #ff7518;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #ff7518;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-danger{--bs-btn-color: #ff0039;--bs-btn-border-color: #ff0039;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #ff0039;--bs-btn-hover-border-color: #ff0039;--bs-btn-focus-shadow-rgb: 255, 0, 57;--bs-btn-active-color: #fff;--bs-btn-active-bg: #ff0039;--bs-btn-active-border-color: #ff0039;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #ff0039;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #ff0039;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-light{--bs-btn-color: #f8f9fa;--bs-btn-border-color: #f8f9fa;--bs-btn-hover-color: #000;--bs-btn-hover-bg: #f8f9fa;--bs-btn-hover-border-color: #f8f9fa;--bs-btn-focus-shadow-rgb: 248, 249, 250;--bs-btn-active-color: #000;--bs-btn-active-bg: #f8f9fa;--bs-btn-active-border-color: #f8f9fa;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #f8f9fa;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #f8f9fa;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-dark{--bs-btn-color: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #343a40;--bs-btn-hover-border-color: #343a40;--bs-btn-focus-shadow-rgb: 52, 58, 64;--bs-btn-active-color: #fff;--bs-btn-active-bg: #343a40;--bs-btn-active-border-color: #343a40;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #343a40;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #343a40;--bs-btn-bg: transparent;--bs-gradient: none}.btn-link{--bs-btn-font-weight: 400;--bs-btn-color: #2761e3;--bs-btn-bg: transparent;--bs-btn-border-color: transparent;--bs-btn-hover-color: #1f4eb6;--bs-btn-hover-border-color: transparent;--bs-btn-active-color: #1f4eb6;--bs-btn-active-border-color: transparent;--bs-btn-disabled-color: #6c757d;--bs-btn-disabled-border-color: transparent;--bs-btn-box-shadow: 0 0 0 #000;--bs-btn-focus-shadow-rgb: 71, 121, 231;text-decoration:underline;-webkit-text-decoration:underline;-moz-text-decoration:underline;-ms-text-decoration:underline;-o-text-decoration:underline}.btn-link:focus-visible{color:var(--bs-btn-color)}.btn-link:hover{color:var(--bs-btn-hover-color)}.btn-lg,.btn-group-lg>.btn{--bs-btn-padding-y: 0.5rem;--bs-btn-padding-x: 1rem;--bs-btn-font-size:1.25rem;--bs-btn-border-radius: 0.5rem}.btn-sm,.btn-group-sm>.btn{--bs-btn-padding-y: 0.25rem;--bs-btn-padding-x: 0.5rem;--bs-btn-font-size:0.875rem;--bs-btn-border-radius: 0.2em}.fade{transition:opacity .15s linear}@media(prefers-reduced-motion: reduce){.fade{transition:none}}.fade:not(.show){opacity:0}.collapse:not(.show){display:none}.collapsing{height:0;overflow:hidden;transition:height .2s ease}@media(prefers-reduced-motion: reduce){.collapsing{transition:none}}.collapsing.collapse-horizontal{width:0;height:auto;transition:width .35s ease}@media(prefers-reduced-motion: reduce){.collapsing.collapse-horizontal{transition:none}}.dropup,.dropend,.dropdown,.dropstart,.dropup-center,.dropdown-center{position:relative}.dropdown-toggle{white-space:nowrap}.dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:"";border-top:.3em solid;border-right:.3em solid rgba(0,0,0,0);border-bottom:0;border-left:.3em solid rgba(0,0,0,0)}.dropdown-toggle:empty::after{margin-left:0}.dropdown-menu{--bs-dropdown-zindex: 1000;--bs-dropdown-min-width: 10rem;--bs-dropdown-padding-x: 0;--bs-dropdown-padding-y: 0.5rem;--bs-dropdown-spacer: 0.125rem;--bs-dropdown-font-size:1rem;--bs-dropdown-color: #343a40;--bs-dropdown-bg: #FEFBF2;--bs-dropdown-border-color: rgba(0, 0, 0, 0.175);--bs-dropdown-border-radius: 0.25rem;--bs-dropdown-border-width: 1px;--bs-dropdown-inner-border-radius: calc(0.25rem - 1px);--bs-dropdown-divider-bg: rgba(0, 0, 0, 0.175);--bs-dropdown-divider-margin-y: 0.5rem;--bs-dropdown-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-dropdown-link-color: #343a40;--bs-dropdown-link-hover-color: #343a40;--bs-dropdown-link-hover-bg: #f8f9fa;--bs-dropdown-link-active-color: #fff;--bs-dropdown-link-active-bg: #2780e3;--bs-dropdown-link-disabled-color: rgba(52, 58, 64, 0.5);--bs-dropdown-item-padding-x: 1rem;--bs-dropdown-item-padding-y: 0.25rem;--bs-dropdown-header-color: #6c757d;--bs-dropdown-header-padding-x: 1rem;--bs-dropdown-header-padding-y: 0.5rem;position:absolute;z-index:var(--bs-dropdown-zindex);display:none;min-width:var(--bs-dropdown-min-width);padding:var(--bs-dropdown-padding-y) var(--bs-dropdown-padding-x);margin:0;font-size:var(--bs-dropdown-font-size);color:var(--bs-dropdown-color);text-align:left;list-style:none;background-color:var(--bs-dropdown-bg);background-clip:padding-box;border:var(--bs-dropdown-border-width) solid var(--bs-dropdown-border-color)}.dropdown-menu[data-bs-popper]{top:100%;left:0;margin-top:var(--bs-dropdown-spacer)}.dropdown-menu-start{--bs-position: start}.dropdown-menu-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-end{--bs-position: end}.dropdown-menu-end[data-bs-popper]{right:0;left:auto}@media(min-width: 576px){.dropdown-menu-sm-start{--bs-position: start}.dropdown-menu-sm-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-sm-end{--bs-position: end}.dropdown-menu-sm-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 768px){.dropdown-menu-md-start{--bs-position: start}.dropdown-menu-md-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-md-end{--bs-position: end}.dropdown-menu-md-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 992px){.dropdown-menu-lg-start{--bs-position: start}.dropdown-menu-lg-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-lg-end{--bs-position: end}.dropdown-menu-lg-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 1200px){.dropdown-menu-xl-start{--bs-position: start}.dropdown-menu-xl-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-xl-end{--bs-position: end}.dropdown-menu-xl-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 1400px){.dropdown-menu-xxl-start{--bs-position: start}.dropdown-menu-xxl-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-xxl-end{--bs-position: end}.dropdown-menu-xxl-end[data-bs-popper]{right:0;left:auto}}.dropup .dropdown-menu[data-bs-popper]{top:auto;bottom:100%;margin-top:0;margin-bottom:var(--bs-dropdown-spacer)}.dropup .dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:"";border-top:0;border-right:.3em solid rgba(0,0,0,0);border-bottom:.3em solid;border-left:.3em solid rgba(0,0,0,0)}.dropup .dropdown-toggle:empty::after{margin-left:0}.dropend .dropdown-menu[data-bs-popper]{top:0;right:auto;left:100%;margin-top:0;margin-left:var(--bs-dropdown-spacer)}.dropend .dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:"";border-top:.3em solid rgba(0,0,0,0);border-right:0;border-bottom:.3em solid rgba(0,0,0,0);border-left:.3em solid}.dropend .dropdown-toggle:empty::after{margin-left:0}.dropend .dropdown-toggle::after{vertical-align:0}.dropstart .dropdown-menu[data-bs-popper]{top:0;right:100%;left:auto;margin-top:0;margin-right:var(--bs-dropdown-spacer)}.dropstart .dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:""}.dropstart .dropdown-toggle::after{display:none}.dropstart .dropdown-toggle::before{display:inline-block;margin-right:.255em;vertical-align:.255em;content:"";border-top:.3em solid rgba(0,0,0,0);border-right:.3em solid;border-bottom:.3em solid rgba(0,0,0,0)}.dropstart .dropdown-toggle:empty::after{margin-left:0}.dropstart .dropdown-toggle::before{vertical-align:0}.dropdown-divider{height:0;margin:var(--bs-dropdown-divider-margin-y) 0;overflow:hidden;border-top:1px solid var(--bs-dropdown-divider-bg);opacity:1}.dropdown-item{display:block;width:100%;padding:var(--bs-dropdown-item-padding-y) var(--bs-dropdown-item-padding-x);clear:both;font-weight:400;color:var(--bs-dropdown-link-color);text-align:inherit;text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;white-space:nowrap;background-color:rgba(0,0,0,0);border:0}.dropdown-item:hover,.dropdown-item:focus{color:var(--bs-dropdown-link-hover-color);background-color:var(--bs-dropdown-link-hover-bg)}.dropdown-item.active,.dropdown-item:active{color:var(--bs-dropdown-link-active-color);text-decoration:none;background-color:var(--bs-dropdown-link-active-bg)}.dropdown-item.disabled,.dropdown-item:disabled{color:var(--bs-dropdown-link-disabled-color);pointer-events:none;background-color:rgba(0,0,0,0)}.dropdown-menu.show{display:block}.dropdown-header{display:block;padding:var(--bs-dropdown-header-padding-y) var(--bs-dropdown-header-padding-x);margin-bottom:0;font-size:0.875rem;color:var(--bs-dropdown-header-color);white-space:nowrap}.dropdown-item-text{display:block;padding:var(--bs-dropdown-item-padding-y) var(--bs-dropdown-item-padding-x);color:var(--bs-dropdown-link-color)}.dropdown-menu-dark{--bs-dropdown-color: #dee2e6;--bs-dropdown-bg: #343a40;--bs-dropdown-border-color: rgba(0, 0, 0, 0.175);--bs-dropdown-box-shadow: ;--bs-dropdown-link-color: #dee2e6;--bs-dropdown-link-hover-color: #fff;--bs-dropdown-divider-bg: rgba(0, 0, 0, 0.175);--bs-dropdown-link-hover-bg: rgba(255, 255, 255, 0.15);--bs-dropdown-link-active-color: #fff;--bs-dropdown-link-active-bg: #2780e3;--bs-dropdown-link-disabled-color: #adb5bd;--bs-dropdown-header-color: #adb5bd}.btn-group,.btn-group-vertical{position:relative;display:inline-flex;vertical-align:middle}.btn-group>.btn,.btn-group-vertical>.btn{position:relative;flex:1 1 auto;-webkit-flex:1 1 auto}.btn-group>.btn-check:checked+.btn,.btn-group>.btn-check:focus+.btn,.btn-group>.btn:hover,.btn-group>.btn:focus,.btn-group>.btn:active,.btn-group>.btn.active,.btn-group-vertical>.btn-check:checked+.btn,.btn-group-vertical>.btn-check:focus+.btn,.btn-group-vertical>.btn:hover,.btn-group-vertical>.btn:focus,.btn-group-vertical>.btn:active,.btn-group-vertical>.btn.active{z-index:1}.btn-toolbar{display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;justify-content:flex-start;-webkit-justify-content:flex-start}.btn-toolbar .input-group{width:auto}.btn-group>:not(.btn-check:first-child)+.btn,.btn-group>.btn-group:not(:first-child){margin-left:calc(1px*-1)}.dropdown-toggle-split{padding-right:.5625rem;padding-left:.5625rem}.dropdown-toggle-split::after,.dropup .dropdown-toggle-split::after,.dropend .dropdown-toggle-split::after{margin-left:0}.dropstart .dropdown-toggle-split::before{margin-right:0}.btn-sm+.dropdown-toggle-split,.btn-group-sm>.btn+.dropdown-toggle-split{padding-right:.375rem;padding-left:.375rem}.btn-lg+.dropdown-toggle-split,.btn-group-lg>.btn+.dropdown-toggle-split{padding-right:.75rem;padding-left:.75rem}.btn-group-vertical{flex-direction:column;-webkit-flex-direction:column;align-items:flex-start;-webkit-align-items:flex-start;justify-content:center;-webkit-justify-content:center}.btn-group-vertical>.btn,.btn-group-vertical>.btn-group{width:100%}.btn-group-vertical>.btn:not(:first-child),.btn-group-vertical>.btn-group:not(:first-child){margin-top:calc(1px*-1)}.nav{--bs-nav-link-padding-x: 1rem;--bs-nav-link-padding-y: 0.5rem;--bs-nav-link-font-weight: ;--bs-nav-link-color: #2761e3;--bs-nav-link-hover-color: #1f4eb6;--bs-nav-link-disabled-color: rgba(52, 58, 64, 0.75);display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;padding-left:0;margin-bottom:0;list-style:none}.nav-link{display:block;padding:var(--bs-nav-link-padding-y) var(--bs-nav-link-padding-x);font-size:var(--bs-nav-link-font-size);font-weight:var(--bs-nav-link-font-weight);color:var(--bs-nav-link-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;background:none;border:0;transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out}@media(prefers-reduced-motion: reduce){.nav-link{transition:none}}.nav-link:hover,.nav-link:focus{color:var(--bs-nav-link-hover-color)}.nav-link:focus-visible{outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.nav-link.disabled,.nav-link:disabled{color:var(--bs-nav-link-disabled-color);pointer-events:none;cursor:default}.nav-tabs{--bs-nav-tabs-border-width: 1px;--bs-nav-tabs-border-color: #dee2e6;--bs-nav-tabs-border-radius: 0.25rem;--bs-nav-tabs-link-hover-border-color: #e9ecef #e9ecef #dee2e6;--bs-nav-tabs-link-active-color: #000;--bs-nav-tabs-link-active-bg: #FEFBF2;--bs-nav-tabs-link-active-border-color: #dee2e6 #dee2e6 #FEFBF2;border-bottom:var(--bs-nav-tabs-border-width) solid var(--bs-nav-tabs-border-color)}.nav-tabs .nav-link{margin-bottom:calc(-1*var(--bs-nav-tabs-border-width));border:var(--bs-nav-tabs-border-width) solid rgba(0,0,0,0)}.nav-tabs .nav-link:hover,.nav-tabs .nav-link:focus{isolation:isolate;border-color:var(--bs-nav-tabs-link-hover-border-color)}.nav-tabs .nav-link.active,.nav-tabs .nav-item.show .nav-link{color:var(--bs-nav-tabs-link-active-color);background-color:var(--bs-nav-tabs-link-active-bg);border-color:var(--bs-nav-tabs-link-active-border-color)}.nav-tabs .dropdown-menu{margin-top:calc(-1*var(--bs-nav-tabs-border-width))}.nav-pills{--bs-nav-pills-border-radius: 0.25rem;--bs-nav-pills-link-active-color: #fff;--bs-nav-pills-link-active-bg: #2780e3}.nav-pills .nav-link.active,.nav-pills .show>.nav-link{color:var(--bs-nav-pills-link-active-color);background-color:var(--bs-nav-pills-link-active-bg)}.nav-underline{--bs-nav-underline-gap: 1rem;--bs-nav-underline-border-width: 0.125rem;--bs-nav-underline-link-active-color: #000;gap:var(--bs-nav-underline-gap)}.nav-underline .nav-link{padding-right:0;padding-left:0;border-bottom:var(--bs-nav-underline-border-width) solid rgba(0,0,0,0)}.nav-underline .nav-link:hover,.nav-underline .nav-link:focus{border-bottom-color:currentcolor}.nav-underline .nav-link.active,.nav-underline .show>.nav-link{font-weight:700;color:var(--bs-nav-underline-link-active-color);border-bottom-color:currentcolor}.nav-fill>.nav-link,.nav-fill .nav-item{flex:1 1 auto;-webkit-flex:1 1 auto;text-align:center}.nav-justified>.nav-link,.nav-justified .nav-item{flex-basis:0;-webkit-flex-basis:0;flex-grow:1;-webkit-flex-grow:1;text-align:center}.nav-fill .nav-item .nav-link,.nav-justified .nav-item .nav-link{width:100%}.tab-content>.tab-pane{display:none}.tab-content>.active{display:block}.navbar{--bs-navbar-padding-x: 0;--bs-navbar-padding-y: 0.5rem;--bs-navbar-color: #F2E5BD;--bs-navbar-hover-color: rgba(147, 176, 241, 0.8);--bs-navbar-disabled-color: rgba(242, 229, 189, 0.75);--bs-navbar-active-color: #93b0f1;--bs-navbar-brand-padding-y: 0.3125rem;--bs-navbar-brand-margin-end: 1rem;--bs-navbar-brand-font-size: 1.25rem;--bs-navbar-brand-color: #F2E5BD;--bs-navbar-brand-hover-color: #93b0f1;--bs-navbar-nav-link-padding-x: 0.5rem;--bs-navbar-toggler-padding-y: 0.25;--bs-navbar-toggler-padding-x: 0;--bs-navbar-toggler-font-size: 1.25rem;--bs-navbar-toggler-icon-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 30 30'%3e%3cpath stroke='%23F2E5BD' stroke-linecap='round' stroke-miterlimit='10' stroke-width='2' d='M4 7h22M4 15h22M4 23h22'/%3e%3c/svg%3e");--bs-navbar-toggler-border-color: rgba(242, 229, 189, 0);--bs-navbar-toggler-border-radius: 0.25rem;--bs-navbar-toggler-focus-width: 0.25rem;--bs-navbar-toggler-transition: box-shadow 0.15s ease-in-out;position:relative;display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between;padding:var(--bs-navbar-padding-y) var(--bs-navbar-padding-x)}.navbar>.container,.navbar>.container-fluid,.navbar>.container-sm,.navbar>.container-md,.navbar>.container-lg,.navbar>.container-xl,.navbar>.container-xxl{display:flex;display:-webkit-flex;flex-wrap:inherit;-webkit-flex-wrap:inherit;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between}.navbar-brand{padding-top:var(--bs-navbar-brand-padding-y);padding-bottom:var(--bs-navbar-brand-padding-y);margin-right:var(--bs-navbar-brand-margin-end);font-size:var(--bs-navbar-brand-font-size);color:var(--bs-navbar-brand-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;white-space:nowrap}.navbar-brand:hover,.navbar-brand:focus{color:var(--bs-navbar-brand-hover-color)}.navbar-nav{--bs-nav-link-padding-x: 0;--bs-nav-link-padding-y: 0.5rem;--bs-nav-link-font-weight: ;--bs-nav-link-color: var(--bs-navbar-color);--bs-nav-link-hover-color: var(--bs-navbar-hover-color);--bs-nav-link-disabled-color: var(--bs-navbar-disabled-color);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;padding-left:0;margin-bottom:0;list-style:none}.navbar-nav .nav-link.active,.navbar-nav .nav-link.show{color:var(--bs-navbar-active-color)}.navbar-nav .dropdown-menu{position:static}.navbar-text{padding-top:.5rem;padding-bottom:.5rem;color:var(--bs-navbar-color)}.navbar-text a,.navbar-text a:hover,.navbar-text a:focus{color:var(--bs-navbar-active-color)}.navbar-collapse{flex-basis:100%;-webkit-flex-basis:100%;flex-grow:1;-webkit-flex-grow:1;align-items:center;-webkit-align-items:center}.navbar-toggler{padding:var(--bs-navbar-toggler-padding-y) var(--bs-navbar-toggler-padding-x);font-size:var(--bs-navbar-toggler-font-size);line-height:1;color:var(--bs-navbar-color);background-color:rgba(0,0,0,0);border:var(--bs-border-width) solid var(--bs-navbar-toggler-border-color);transition:var(--bs-navbar-toggler-transition)}@media(prefers-reduced-motion: reduce){.navbar-toggler{transition:none}}.navbar-toggler:hover{text-decoration:none}.navbar-toggler:focus{text-decoration:none;outline:0;box-shadow:0 0 0 var(--bs-navbar-toggler-focus-width)}.navbar-toggler-icon{display:inline-block;width:1.5em;height:1.5em;vertical-align:middle;background-image:var(--bs-navbar-toggler-icon-bg);background-repeat:no-repeat;background-position:center;background-size:100%}.navbar-nav-scroll{max-height:var(--bs-scroll-height, 75vh);overflow-y:auto}@media(min-width: 576px){.navbar-expand-sm{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-sm .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-sm .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-sm .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-sm .navbar-nav-scroll{overflow:visible}.navbar-expand-sm .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-sm .navbar-toggler{display:none}.navbar-expand-sm .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-sm .offcanvas .offcanvas-header{display:none}.navbar-expand-sm .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 768px){.navbar-expand-md{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-md .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-md .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-md .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-md .navbar-nav-scroll{overflow:visible}.navbar-expand-md .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-md .navbar-toggler{display:none}.navbar-expand-md .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-md .offcanvas .offcanvas-header{display:none}.navbar-expand-md .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 992px){.navbar-expand-lg{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-lg .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-lg .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-lg .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-lg .navbar-nav-scroll{overflow:visible}.navbar-expand-lg .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-lg .navbar-toggler{display:none}.navbar-expand-lg .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-lg .offcanvas .offcanvas-header{display:none}.navbar-expand-lg .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 1200px){.navbar-expand-xl{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-xl .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-xl .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-xl .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-xl .navbar-nav-scroll{overflow:visible}.navbar-expand-xl .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-xl .navbar-toggler{display:none}.navbar-expand-xl .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-xl .offcanvas .offcanvas-header{display:none}.navbar-expand-xl .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 1400px){.navbar-expand-xxl{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-xxl .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-xxl .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-xxl .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-xxl .navbar-nav-scroll{overflow:visible}.navbar-expand-xxl .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-xxl .navbar-toggler{display:none}.navbar-expand-xxl .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-xxl .offcanvas .offcanvas-header{display:none}.navbar-expand-xxl .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}.navbar-expand{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand .navbar-nav .dropdown-menu{position:absolute}.navbar-expand .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand .navbar-nav-scroll{overflow:visible}.navbar-expand .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand .navbar-toggler{display:none}.navbar-expand .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand .offcanvas .offcanvas-header{display:none}.navbar-expand .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}.navbar-dark,.navbar[data-bs-theme=dark]{--bs-navbar-color: #F2E5BD;--bs-navbar-hover-color: rgba(147, 176, 241, 0.8);--bs-navbar-disabled-color: rgba(242, 229, 189, 0.75);--bs-navbar-active-color: #93b0f1;--bs-navbar-brand-color: #F2E5BD;--bs-navbar-brand-hover-color: #93b0f1;--bs-navbar-toggler-border-color: rgba(242, 229, 189, 0);--bs-navbar-toggler-icon-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 30 30'%3e%3cpath stroke='%23F2E5BD' stroke-linecap='round' stroke-miterlimit='10' stroke-width='2' d='M4 7h22M4 15h22M4 23h22'/%3e%3c/svg%3e")}[data-bs-theme=dark] .navbar-toggler-icon{--bs-navbar-toggler-icon-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 30 30'%3e%3cpath stroke='%23F2E5BD' stroke-linecap='round' stroke-miterlimit='10' stroke-width='2' d='M4 7h22M4 15h22M4 23h22'/%3e%3c/svg%3e")}.card{--bs-card-spacer-y: 1rem;--bs-card-spacer-x: 1rem;--bs-card-title-spacer-y: 0.5rem;--bs-card-title-color: ;--bs-card-subtitle-color: ;--bs-card-border-width: 1px;--bs-card-border-color: rgba(0, 0, 0, 0.175);--bs-card-border-radius: 0.25rem;--bs-card-box-shadow: ;--bs-card-inner-border-radius: calc(0.25rem - 1px);--bs-card-cap-padding-y: 0.5rem;--bs-card-cap-padding-x: 1rem;--bs-card-cap-bg: rgba(52, 58, 64, 0.25);--bs-card-cap-color: ;--bs-card-height: ;--bs-card-color: ;--bs-card-bg: #FEFBF2;--bs-card-img-overlay-padding: 1rem;--bs-card-group-margin: 0.75rem;position:relative;display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;min-width:0;height:var(--bs-card-height);color:var(--bs-body-color);word-wrap:break-word;background-color:var(--bs-card-bg);background-clip:border-box;border:var(--bs-card-border-width) solid var(--bs-card-border-color)}.card>hr{margin-right:0;margin-left:0}.card>.list-group{border-top:inherit;border-bottom:inherit}.card>.list-group:first-child{border-top-width:0}.card>.list-group:last-child{border-bottom-width:0}.card>.card-header+.list-group,.card>.list-group+.card-footer{border-top:0}.card-body{flex:1 1 auto;-webkit-flex:1 1 auto;padding:var(--bs-card-spacer-y) var(--bs-card-spacer-x);color:var(--bs-card-color)}.card-title{margin-bottom:var(--bs-card-title-spacer-y);color:var(--bs-card-title-color)}.card-subtitle{margin-top:calc(-0.5*var(--bs-card-title-spacer-y));margin-bottom:0;color:var(--bs-card-subtitle-color)}.card-text:last-child{margin-bottom:0}.card-link+.card-link{margin-left:var(--bs-card-spacer-x)}.card-header{padding:var(--bs-card-cap-padding-y) var(--bs-card-cap-padding-x);margin-bottom:0;color:var(--bs-card-cap-color);background-color:var(--bs-card-cap-bg);border-bottom:var(--bs-card-border-width) solid var(--bs-card-border-color)}.card-footer{padding:var(--bs-card-cap-padding-y) var(--bs-card-cap-padding-x);color:var(--bs-card-cap-color);background-color:var(--bs-card-cap-bg);border-top:var(--bs-card-border-width) solid var(--bs-card-border-color)}.card-header-tabs{margin-right:calc(-0.5*var(--bs-card-cap-padding-x));margin-bottom:calc(-1*var(--bs-card-cap-padding-y));margin-left:calc(-0.5*var(--bs-card-cap-padding-x));border-bottom:0}.card-header-tabs .nav-link.active{background-color:var(--bs-card-bg);border-bottom-color:var(--bs-card-bg)}.card-header-pills{margin-right:calc(-0.5*var(--bs-card-cap-padding-x));margin-left:calc(-0.5*var(--bs-card-cap-padding-x))}.card-img-overlay{position:absolute;top:0;right:0;bottom:0;left:0;padding:var(--bs-card-img-overlay-padding)}.card-img,.card-img-top,.card-img-bottom{width:100%}.card-group>.card{margin-bottom:var(--bs-card-group-margin)}@media(min-width: 576px){.card-group{display:flex;display:-webkit-flex;flex-flow:row wrap;-webkit-flex-flow:row wrap}.card-group>.card{flex:1 0 0%;-webkit-flex:1 0 0%;margin-bottom:0}.card-group>.card+.card{margin-left:0;border-left:0}}.accordion{--bs-accordion-color: #343a40;--bs-accordion-bg: #FEFBF2;--bs-accordion-transition: color 0.15s ease-in-out, background-color 0.15s ease-in-out, border-color 0.15s ease-in-out, box-shadow 0.15s ease-in-out, border-radius 0.15s ease;--bs-accordion-border-color: #dee2e6;--bs-accordion-border-width: 1px;--bs-accordion-border-radius: 0.25rem;--bs-accordion-inner-border-radius: calc(0.25rem - 1px);--bs-accordion-btn-padding-x: 1.25rem;--bs-accordion-btn-padding-y: 1rem;--bs-accordion-btn-color: #343a40;--bs-accordion-btn-bg: #FEFBF2;--bs-accordion-btn-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23343a40'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e");--bs-accordion-btn-icon-width: 1.25rem;--bs-accordion-btn-icon-transform: rotate(-180deg);--bs-accordion-btn-icon-transition: transform 0.2s ease-in-out;--bs-accordion-btn-active-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%2310335b'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e");--bs-accordion-btn-focus-border-color: #93c0f1;--bs-accordion-btn-focus-box-shadow: 0 0 0 0.25rem rgba(39, 128, 227, 0.25);--bs-accordion-body-padding-x: 1.25rem;--bs-accordion-body-padding-y: 1rem;--bs-accordion-active-color: #10335b;--bs-accordion-active-bg: #d4e6f9}.accordion-button{position:relative;display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;width:100%;padding:var(--bs-accordion-btn-padding-y) var(--bs-accordion-btn-padding-x);font-size:1rem;color:var(--bs-accordion-btn-color);text-align:left;background-color:var(--bs-accordion-btn-bg);border:0;overflow-anchor:none;transition:var(--bs-accordion-transition)}@media(prefers-reduced-motion: reduce){.accordion-button{transition:none}}.accordion-button:not(.collapsed){color:var(--bs-accordion-active-color);background-color:var(--bs-accordion-active-bg);box-shadow:inset 0 calc(-1*var(--bs-accordion-border-width)) 0 var(--bs-accordion-border-color)}.accordion-button:not(.collapsed)::after{background-image:var(--bs-accordion-btn-active-icon);transform:var(--bs-accordion-btn-icon-transform)}.accordion-button::after{flex-shrink:0;-webkit-flex-shrink:0;width:var(--bs-accordion-btn-icon-width);height:var(--bs-accordion-btn-icon-width);margin-left:auto;content:"";background-image:var(--bs-accordion-btn-icon);background-repeat:no-repeat;background-size:var(--bs-accordion-btn-icon-width);transition:var(--bs-accordion-btn-icon-transition)}@media(prefers-reduced-motion: reduce){.accordion-button::after{transition:none}}.accordion-button:hover{z-index:2}.accordion-button:focus{z-index:3;border-color:var(--bs-accordion-btn-focus-border-color);outline:0;box-shadow:var(--bs-accordion-btn-focus-box-shadow)}.accordion-header{margin-bottom:0}.accordion-item{color:var(--bs-accordion-color);background-color:var(--bs-accordion-bg);border:var(--bs-accordion-border-width) solid var(--bs-accordion-border-color)}.accordion-item:not(:first-of-type){border-top:0}.accordion-body{padding:var(--bs-accordion-body-padding-y) var(--bs-accordion-body-padding-x)}.accordion-flush .accordion-collapse{border-width:0}.accordion-flush .accordion-item{border-right:0;border-left:0}.accordion-flush .accordion-item:first-child{border-top:0}.accordion-flush .accordion-item:last-child{border-bottom:0}[data-bs-theme=dark] .accordion-button::after{--bs-accordion-btn-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%237db3ee'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e");--bs-accordion-btn-active-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%237db3ee'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e")}.breadcrumb{--bs-breadcrumb-padding-x: 0;--bs-breadcrumb-padding-y: 0;--bs-breadcrumb-margin-bottom: 1rem;--bs-breadcrumb-bg: ;--bs-breadcrumb-border-radius: ;--bs-breadcrumb-divider-color: rgba(52, 58, 64, 0.75);--bs-breadcrumb-item-padding-x: 0.5rem;--bs-breadcrumb-item-active-color: rgba(52, 58, 64, 0.75);display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;padding:var(--bs-breadcrumb-padding-y) var(--bs-breadcrumb-padding-x);margin-bottom:var(--bs-breadcrumb-margin-bottom);font-size:var(--bs-breadcrumb-font-size);list-style:none;background-color:var(--bs-breadcrumb-bg)}.breadcrumb-item+.breadcrumb-item{padding-left:var(--bs-breadcrumb-item-padding-x)}.breadcrumb-item+.breadcrumb-item::before{float:left;padding-right:var(--bs-breadcrumb-item-padding-x);color:var(--bs-breadcrumb-divider-color);content:var(--bs-breadcrumb-divider, ">") /* rtl: var(--bs-breadcrumb-divider, ">") */}.breadcrumb-item.active{color:var(--bs-breadcrumb-item-active-color)}.pagination{--bs-pagination-padding-x: 0.75rem;--bs-pagination-padding-y: 0.375rem;--bs-pagination-font-size:1rem;--bs-pagination-color: #2761e3;--bs-pagination-bg: #FEFBF2;--bs-pagination-border-width: 1px;--bs-pagination-border-color: #dee2e6;--bs-pagination-border-radius: 0.25rem;--bs-pagination-hover-color: #1f4eb6;--bs-pagination-hover-bg: #f8f9fa;--bs-pagination-hover-border-color: #dee2e6;--bs-pagination-focus-color: #1f4eb6;--bs-pagination-focus-bg: #e9ecef;--bs-pagination-focus-box-shadow: 0 0 0 0.25rem rgba(39, 128, 227, 0.25);--bs-pagination-active-color: #fff;--bs-pagination-active-bg: #2780e3;--bs-pagination-active-border-color: #2780e3;--bs-pagination-disabled-color: rgba(52, 58, 64, 0.75);--bs-pagination-disabled-bg: #e9ecef;--bs-pagination-disabled-border-color: #dee2e6;display:flex;display:-webkit-flex;padding-left:0;list-style:none}.page-link{position:relative;display:block;padding:var(--bs-pagination-padding-y) var(--bs-pagination-padding-x);font-size:var(--bs-pagination-font-size);color:var(--bs-pagination-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;background-color:var(--bs-pagination-bg);border:var(--bs-pagination-border-width) solid var(--bs-pagination-border-color);transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.page-link{transition:none}}.page-link:hover{z-index:2;color:var(--bs-pagination-hover-color);background-color:var(--bs-pagination-hover-bg);border-color:var(--bs-pagination-hover-border-color)}.page-link:focus{z-index:3;color:var(--bs-pagination-focus-color);background-color:var(--bs-pagination-focus-bg);outline:0;box-shadow:var(--bs-pagination-focus-box-shadow)}.page-link.active,.active>.page-link{z-index:3;color:var(--bs-pagination-active-color);background-color:var(--bs-pagination-active-bg);border-color:var(--bs-pagination-active-border-color)}.page-link.disabled,.disabled>.page-link{color:var(--bs-pagination-disabled-color);pointer-events:none;background-color:var(--bs-pagination-disabled-bg);border-color:var(--bs-pagination-disabled-border-color)}.page-item:not(:first-child) .page-link{margin-left:calc(1px*-1)}.pagination-lg{--bs-pagination-padding-x: 1.5rem;--bs-pagination-padding-y: 0.75rem;--bs-pagination-font-size:1.25rem;--bs-pagination-border-radius: 0.5rem}.pagination-sm{--bs-pagination-padding-x: 0.5rem;--bs-pagination-padding-y: 0.25rem;--bs-pagination-font-size:0.875rem;--bs-pagination-border-radius: 0.2em}.badge{--bs-badge-padding-x: 0.65em;--bs-badge-padding-y: 0.35em;--bs-badge-font-size:0.75em;--bs-badge-font-weight: 700;--bs-badge-color: #fff;--bs-badge-border-radius: 0.25rem;display:inline-block;padding:var(--bs-badge-padding-y) var(--bs-badge-padding-x);font-size:var(--bs-badge-font-size);font-weight:var(--bs-badge-font-weight);line-height:1;color:var(--bs-badge-color);text-align:center;white-space:nowrap;vertical-align:baseline}.badge:empty{display:none}.btn .badge{position:relative;top:-1px}.alert{--bs-alert-bg: transparent;--bs-alert-padding-x: 1rem;--bs-alert-padding-y: 1rem;--bs-alert-margin-bottom: 1rem;--bs-alert-color: inherit;--bs-alert-border-color: transparent;--bs-alert-border: 0 solid var(--bs-alert-border-color);--bs-alert-border-radius: 0.25rem;--bs-alert-link-color: inherit;position:relative;padding:var(--bs-alert-padding-y) var(--bs-alert-padding-x);margin-bottom:var(--bs-alert-margin-bottom);color:var(--bs-alert-color);background-color:var(--bs-alert-bg);border:var(--bs-alert-border)}.alert-heading{color:inherit}.alert-link{font-weight:700;color:var(--bs-alert-link-color)}.alert-dismissible{padding-right:3rem}.alert-dismissible .btn-close{position:absolute;top:0;right:0;z-index:2;padding:1.25rem 1rem}.alert-default{--bs-alert-color: var(--bs-default-text-emphasis);--bs-alert-bg: var(--bs-default-bg-subtle);--bs-alert-border-color: var(--bs-default-border-subtle);--bs-alert-link-color: var(--bs-default-text-emphasis)}.alert-primary{--bs-alert-color: var(--bs-primary-text-emphasis);--bs-alert-bg: var(--bs-primary-bg-subtle);--bs-alert-border-color: var(--bs-primary-border-subtle);--bs-alert-link-color: var(--bs-primary-text-emphasis)}.alert-secondary{--bs-alert-color: var(--bs-secondary-text-emphasis);--bs-alert-bg: var(--bs-secondary-bg-subtle);--bs-alert-border-color: var(--bs-secondary-border-subtle);--bs-alert-link-color: var(--bs-secondary-text-emphasis)}.alert-success{--bs-alert-color: var(--bs-success-text-emphasis);--bs-alert-bg: var(--bs-success-bg-subtle);--bs-alert-border-color: var(--bs-success-border-subtle);--bs-alert-link-color: var(--bs-success-text-emphasis)}.alert-info{--bs-alert-color: var(--bs-info-text-emphasis);--bs-alert-bg: var(--bs-info-bg-subtle);--bs-alert-border-color: var(--bs-info-border-subtle);--bs-alert-link-color: var(--bs-info-text-emphasis)}.alert-warning{--bs-alert-color: var(--bs-warning-text-emphasis);--bs-alert-bg: var(--bs-warning-bg-subtle);--bs-alert-border-color: var(--bs-warning-border-subtle);--bs-alert-link-color: var(--bs-warning-text-emphasis)}.alert-danger{--bs-alert-color: var(--bs-danger-text-emphasis);--bs-alert-bg: var(--bs-danger-bg-subtle);--bs-alert-border-color: var(--bs-danger-border-subtle);--bs-alert-link-color: var(--bs-danger-text-emphasis)}.alert-light{--bs-alert-color: var(--bs-light-text-emphasis);--bs-alert-bg: var(--bs-light-bg-subtle);--bs-alert-border-color: var(--bs-light-border-subtle);--bs-alert-link-color: var(--bs-light-text-emphasis)}.alert-dark{--bs-alert-color: var(--bs-dark-text-emphasis);--bs-alert-bg: var(--bs-dark-bg-subtle);--bs-alert-border-color: var(--bs-dark-border-subtle);--bs-alert-link-color: var(--bs-dark-text-emphasis)}@keyframes progress-bar-stripes{0%{background-position-x:.5rem}}.progress,.progress-stacked{--bs-progress-height: 0.5rem;--bs-progress-font-size:0.75rem;--bs-progress-bg: #e9ecef;--bs-progress-border-radius: 0.25rem;--bs-progress-box-shadow: inset 0 1px 2px rgba(0, 0, 0, 0.075);--bs-progress-bar-color: #fff;--bs-progress-bar-bg: #2780e3;--bs-progress-bar-transition: width 0.6s ease;display:flex;display:-webkit-flex;height:var(--bs-progress-height);overflow:hidden;font-size:var(--bs-progress-font-size);background-color:var(--bs-progress-bg)}.progress-bar{display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;justify-content:center;-webkit-justify-content:center;overflow:hidden;color:var(--bs-progress-bar-color);text-align:center;white-space:nowrap;background-color:var(--bs-progress-bar-bg);transition:var(--bs-progress-bar-transition)}@media(prefers-reduced-motion: reduce){.progress-bar{transition:none}}.progress-bar-striped{background-image:linear-gradient(45deg, rgba(255, 255, 255, 0.15) 25%, transparent 25%, transparent 50%, rgba(255, 255, 255, 0.15) 50%, rgba(255, 255, 255, 0.15) 75%, transparent 75%, transparent);background-size:var(--bs-progress-height) var(--bs-progress-height)}.progress-stacked>.progress{overflow:visible}.progress-stacked>.progress>.progress-bar{width:100%}.progress-bar-animated{animation:1s linear infinite progress-bar-stripes}@media(prefers-reduced-motion: reduce){.progress-bar-animated{animation:none}}.list-group{--bs-list-group-color: #343a40;--bs-list-group-bg: #FEFBF2;--bs-list-group-border-color: #dee2e6;--bs-list-group-border-width: 1px;--bs-list-group-border-radius: 0.25rem;--bs-list-group-item-padding-x: 1rem;--bs-list-group-item-padding-y: 0.5rem;--bs-list-group-action-color: rgba(52, 58, 64, 0.75);--bs-list-group-action-hover-color: #000;--bs-list-group-action-hover-bg: #f8f9fa;--bs-list-group-action-active-color: #343a40;--bs-list-group-action-active-bg: #e9ecef;--bs-list-group-disabled-color: rgba(52, 58, 64, 0.75);--bs-list-group-disabled-bg: #FEFBF2;--bs-list-group-active-color: #fff;--bs-list-group-active-bg: #2780e3;--bs-list-group-active-border-color: #2780e3;display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;padding-left:0;margin-bottom:0}.list-group-numbered{list-style-type:none;counter-reset:section}.list-group-numbered>.list-group-item::before{content:counters(section, ".") ". ";counter-increment:section}.list-group-item-action{width:100%;color:var(--bs-list-group-action-color);text-align:inherit}.list-group-item-action:hover,.list-group-item-action:focus{z-index:1;color:var(--bs-list-group-action-hover-color);text-decoration:none;background-color:var(--bs-list-group-action-hover-bg)}.list-group-item-action:active{color:var(--bs-list-group-action-active-color);background-color:var(--bs-list-group-action-active-bg)}.list-group-item{position:relative;display:block;padding:var(--bs-list-group-item-padding-y) var(--bs-list-group-item-padding-x);color:var(--bs-list-group-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;background-color:var(--bs-list-group-bg);border:var(--bs-list-group-border-width) solid var(--bs-list-group-border-color)}.list-group-item.disabled,.list-group-item:disabled{color:var(--bs-list-group-disabled-color);pointer-events:none;background-color:var(--bs-list-group-disabled-bg)}.list-group-item.active{z-index:2;color:var(--bs-list-group-active-color);background-color:var(--bs-list-group-active-bg);border-color:var(--bs-list-group-active-border-color)}.list-group-item+.list-group-item{border-top-width:0}.list-group-item+.list-group-item.active{margin-top:calc(-1*var(--bs-list-group-border-width));border-top-width:var(--bs-list-group-border-width)}.list-group-horizontal{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal>.list-group-item.active{margin-top:0}.list-group-horizontal>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}@media(min-width: 576px){.list-group-horizontal-sm{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-sm>.list-group-item.active{margin-top:0}.list-group-horizontal-sm>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-sm>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 768px){.list-group-horizontal-md{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-md>.list-group-item.active{margin-top:0}.list-group-horizontal-md>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-md>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 992px){.list-group-horizontal-lg{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-lg>.list-group-item.active{margin-top:0}.list-group-horizontal-lg>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-lg>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 1200px){.list-group-horizontal-xl{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-xl>.list-group-item.active{margin-top:0}.list-group-horizontal-xl>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-xl>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 1400px){.list-group-horizontal-xxl{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-xxl>.list-group-item.active{margin-top:0}.list-group-horizontal-xxl>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-xxl>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}.list-group-flush>.list-group-item{border-width:0 0 var(--bs-list-group-border-width)}.list-group-flush>.list-group-item:last-child{border-bottom-width:0}.list-group-item-default{--bs-list-group-color: var(--bs-default-text-emphasis);--bs-list-group-bg: var(--bs-default-bg-subtle);--bs-list-group-border-color: var(--bs-default-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-default-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-default-border-subtle);--bs-list-group-active-color: var(--bs-default-bg-subtle);--bs-list-group-active-bg: var(--bs-default-text-emphasis);--bs-list-group-active-border-color: var(--bs-default-text-emphasis)}.list-group-item-primary{--bs-list-group-color: var(--bs-primary-text-emphasis);--bs-list-group-bg: var(--bs-primary-bg-subtle);--bs-list-group-border-color: var(--bs-primary-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-primary-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-primary-border-subtle);--bs-list-group-active-color: var(--bs-primary-bg-subtle);--bs-list-group-active-bg: var(--bs-primary-text-emphasis);--bs-list-group-active-border-color: var(--bs-primary-text-emphasis)}.list-group-item-secondary{--bs-list-group-color: var(--bs-secondary-text-emphasis);--bs-list-group-bg: var(--bs-secondary-bg-subtle);--bs-list-group-border-color: var(--bs-secondary-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-secondary-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-secondary-border-subtle);--bs-list-group-active-color: var(--bs-secondary-bg-subtle);--bs-list-group-active-bg: var(--bs-secondary-text-emphasis);--bs-list-group-active-border-color: var(--bs-secondary-text-emphasis)}.list-group-item-success{--bs-list-group-color: var(--bs-success-text-emphasis);--bs-list-group-bg: var(--bs-success-bg-subtle);--bs-list-group-border-color: var(--bs-success-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-success-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-success-border-subtle);--bs-list-group-active-color: var(--bs-success-bg-subtle);--bs-list-group-active-bg: var(--bs-success-text-emphasis);--bs-list-group-active-border-color: var(--bs-success-text-emphasis)}.list-group-item-info{--bs-list-group-color: var(--bs-info-text-emphasis);--bs-list-group-bg: var(--bs-info-bg-subtle);--bs-list-group-border-color: var(--bs-info-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-info-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-info-border-subtle);--bs-list-group-active-color: var(--bs-info-bg-subtle);--bs-list-group-active-bg: var(--bs-info-text-emphasis);--bs-list-group-active-border-color: var(--bs-info-text-emphasis)}.list-group-item-warning{--bs-list-group-color: var(--bs-warning-text-emphasis);--bs-list-group-bg: var(--bs-warning-bg-subtle);--bs-list-group-border-color: var(--bs-warning-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-warning-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-warning-border-subtle);--bs-list-group-active-color: var(--bs-warning-bg-subtle);--bs-list-group-active-bg: var(--bs-warning-text-emphasis);--bs-list-group-active-border-color: var(--bs-warning-text-emphasis)}.list-group-item-danger{--bs-list-group-color: var(--bs-danger-text-emphasis);--bs-list-group-bg: var(--bs-danger-bg-subtle);--bs-list-group-border-color: var(--bs-danger-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-danger-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-danger-border-subtle);--bs-list-group-active-color: var(--bs-danger-bg-subtle);--bs-list-group-active-bg: var(--bs-danger-text-emphasis);--bs-list-group-active-border-color: var(--bs-danger-text-emphasis)}.list-group-item-light{--bs-list-group-color: var(--bs-light-text-emphasis);--bs-list-group-bg: var(--bs-light-bg-subtle);--bs-list-group-border-color: var(--bs-light-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-light-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-light-border-subtle);--bs-list-group-active-color: var(--bs-light-bg-subtle);--bs-list-group-active-bg: var(--bs-light-text-emphasis);--bs-list-group-active-border-color: var(--bs-light-text-emphasis)}.list-group-item-dark{--bs-list-group-color: var(--bs-dark-text-emphasis);--bs-list-group-bg: var(--bs-dark-bg-subtle);--bs-list-group-border-color: var(--bs-dark-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-dark-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-dark-border-subtle);--bs-list-group-active-color: var(--bs-dark-bg-subtle);--bs-list-group-active-bg: var(--bs-dark-text-emphasis);--bs-list-group-active-border-color: var(--bs-dark-text-emphasis)}.btn-close{--bs-btn-close-color: #000;--bs-btn-close-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23000'%3e%3cpath d='M.293.293a1 1 0 0 1 1.414 0L8 6.586 14.293.293a1 1 0 1 1 1.414 1.414L9.414 8l6.293 6.293a1 1 0 0 1-1.414 1.414L8 9.414l-6.293 6.293a1 1 0 0 1-1.414-1.414L6.586 8 .293 1.707a1 1 0 0 1 0-1.414z'/%3e%3c/svg%3e");--bs-btn-close-opacity: 0.5;--bs-btn-close-hover-opacity: 0.75;--bs-btn-close-focus-shadow: 0 0 0 0.25rem rgba(39, 128, 227, 0.25);--bs-btn-close-focus-opacity: 1;--bs-btn-close-disabled-opacity: 0.25;--bs-btn-close-white-filter: invert(1) grayscale(100%) brightness(200%);box-sizing:content-box;width:1em;height:1em;padding:.25em .25em;color:var(--bs-btn-close-color);background:rgba(0,0,0,0) var(--bs-btn-close-bg) center/1em auto no-repeat;border:0;opacity:var(--bs-btn-close-opacity)}.btn-close:hover{color:var(--bs-btn-close-color);text-decoration:none;opacity:var(--bs-btn-close-hover-opacity)}.btn-close:focus{outline:0;box-shadow:var(--bs-btn-close-focus-shadow);opacity:var(--bs-btn-close-focus-opacity)}.btn-close:disabled,.btn-close.disabled{pointer-events:none;user-select:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;-o-user-select:none;opacity:var(--bs-btn-close-disabled-opacity)}.btn-close-white{filter:var(--bs-btn-close-white-filter)}[data-bs-theme=dark] .btn-close{filter:var(--bs-btn-close-white-filter)}.toast{--bs-toast-zindex: 1090;--bs-toast-padding-x: 0.75rem;--bs-toast-padding-y: 0.5rem;--bs-toast-spacing: 1.5rem;--bs-toast-max-width: 350px;--bs-toast-font-size:0.875rem;--bs-toast-color: ;--bs-toast-bg: rgba(254, 251, 242, 0.85);--bs-toast-border-width: 1px;--bs-toast-border-color: rgba(0, 0, 0, 0.175);--bs-toast-border-radius: 0.25rem;--bs-toast-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-toast-header-color: rgba(52, 58, 64, 0.75);--bs-toast-header-bg: rgba(254, 251, 242, 0.85);--bs-toast-header-border-color: rgba(0, 0, 0, 0.175);width:var(--bs-toast-max-width);max-width:100%;font-size:var(--bs-toast-font-size);color:var(--bs-toast-color);pointer-events:auto;background-color:var(--bs-toast-bg);background-clip:padding-box;border:var(--bs-toast-border-width) solid var(--bs-toast-border-color);box-shadow:var(--bs-toast-box-shadow)}.toast.showing{opacity:0}.toast:not(.show){display:none}.toast-container{--bs-toast-zindex: 1090;position:absolute;z-index:var(--bs-toast-zindex);width:max-content;width:-webkit-max-content;width:-moz-max-content;width:-ms-max-content;width:-o-max-content;max-width:100%;pointer-events:none}.toast-container>:not(:last-child){margin-bottom:var(--bs-toast-spacing)}.toast-header{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;padding:var(--bs-toast-padding-y) var(--bs-toast-padding-x);color:var(--bs-toast-header-color);background-color:var(--bs-toast-header-bg);background-clip:padding-box;border-bottom:var(--bs-toast-border-width) solid var(--bs-toast-header-border-color)}.toast-header .btn-close{margin-right:calc(-0.5*var(--bs-toast-padding-x));margin-left:var(--bs-toast-padding-x)}.toast-body{padding:var(--bs-toast-padding-x);word-wrap:break-word}.modal{--bs-modal-zindex: 1055;--bs-modal-width: 500px;--bs-modal-padding: 1rem;--bs-modal-margin: 0.5rem;--bs-modal-color: ;--bs-modal-bg: #FEFBF2;--bs-modal-border-color: rgba(0, 0, 0, 0.175);--bs-modal-border-width: 1px;--bs-modal-border-radius: 0.5rem;--bs-modal-box-shadow: 0 0.125rem 0.25rem rgba(0, 0, 0, 0.075);--bs-modal-inner-border-radius: calc(0.5rem - 1px);--bs-modal-header-padding-x: 1rem;--bs-modal-header-padding-y: 1rem;--bs-modal-header-padding: 1rem 1rem;--bs-modal-header-border-color: #dee2e6;--bs-modal-header-border-width: 1px;--bs-modal-title-line-height: 1.5;--bs-modal-footer-gap: 0.5rem;--bs-modal-footer-bg: ;--bs-modal-footer-border-color: #dee2e6;--bs-modal-footer-border-width: 1px;position:fixed;top:0;left:0;z-index:var(--bs-modal-zindex);display:none;width:100%;height:100%;overflow-x:hidden;overflow-y:auto;outline:0}.modal-dialog{position:relative;width:auto;margin:var(--bs-modal-margin);pointer-events:none}.modal.fade .modal-dialog{transition:transform .3s ease-out;transform:translate(0, -50px)}@media(prefers-reduced-motion: reduce){.modal.fade .modal-dialog{transition:none}}.modal.show .modal-dialog{transform:none}.modal.modal-static .modal-dialog{transform:scale(1.02)}.modal-dialog-scrollable{height:calc(100% - var(--bs-modal-margin)*2)}.modal-dialog-scrollable .modal-content{max-height:100%;overflow:hidden}.modal-dialog-scrollable .modal-body{overflow-y:auto}.modal-dialog-centered{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;min-height:calc(100% - var(--bs-modal-margin)*2)}.modal-content{position:relative;display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;width:100%;color:var(--bs-modal-color);pointer-events:auto;background-color:var(--bs-modal-bg);background-clip:padding-box;border:var(--bs-modal-border-width) solid var(--bs-modal-border-color);outline:0}.modal-backdrop{--bs-backdrop-zindex: 1050;--bs-backdrop-bg: #000;--bs-backdrop-opacity: 0.5;position:fixed;top:0;left:0;z-index:var(--bs-backdrop-zindex);width:100vw;height:100vh;background-color:var(--bs-backdrop-bg)}.modal-backdrop.fade{opacity:0}.modal-backdrop.show{opacity:var(--bs-backdrop-opacity)}.modal-header{display:flex;display:-webkit-flex;flex-shrink:0;-webkit-flex-shrink:0;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between;padding:var(--bs-modal-header-padding);border-bottom:var(--bs-modal-header-border-width) solid var(--bs-modal-header-border-color)}.modal-header .btn-close{padding:calc(var(--bs-modal-header-padding-y)*.5) calc(var(--bs-modal-header-padding-x)*.5);margin:calc(-0.5*var(--bs-modal-header-padding-y)) calc(-0.5*var(--bs-modal-header-padding-x)) calc(-0.5*var(--bs-modal-header-padding-y)) auto}.modal-title{margin-bottom:0;line-height:var(--bs-modal-title-line-height)}.modal-body{position:relative;flex:1 1 auto;-webkit-flex:1 1 auto;padding:var(--bs-modal-padding)}.modal-footer{display:flex;display:-webkit-flex;flex-shrink:0;-webkit-flex-shrink:0;flex-wrap:wrap;-webkit-flex-wrap:wrap;align-items:center;-webkit-align-items:center;justify-content:flex-end;-webkit-justify-content:flex-end;padding:calc(var(--bs-modal-padding) - var(--bs-modal-footer-gap)*.5);background-color:var(--bs-modal-footer-bg);border-top:var(--bs-modal-footer-border-width) solid var(--bs-modal-footer-border-color)}.modal-footer>*{margin:calc(var(--bs-modal-footer-gap)*.5)}@media(min-width: 576px){.modal{--bs-modal-margin: 1.75rem;--bs-modal-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15)}.modal-dialog{max-width:var(--bs-modal-width);margin-right:auto;margin-left:auto}.modal-sm{--bs-modal-width: 300px}}@media(min-width: 992px){.modal-lg,.modal-xl{--bs-modal-width: 800px}}@media(min-width: 1200px){.modal-xl{--bs-modal-width: 1140px}}.modal-fullscreen{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen .modal-content{height:100%;border:0}.modal-fullscreen .modal-body{overflow-y:auto}@media(max-width: 575.98px){.modal-fullscreen-sm-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-sm-down .modal-content{height:100%;border:0}.modal-fullscreen-sm-down .modal-body{overflow-y:auto}}@media(max-width: 767.98px){.modal-fullscreen-md-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-md-down .modal-content{height:100%;border:0}.modal-fullscreen-md-down .modal-body{overflow-y:auto}}@media(max-width: 991.98px){.modal-fullscreen-lg-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-lg-down .modal-content{height:100%;border:0}.modal-fullscreen-lg-down .modal-body{overflow-y:auto}}@media(max-width: 1199.98px){.modal-fullscreen-xl-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-xl-down .modal-content{height:100%;border:0}.modal-fullscreen-xl-down .modal-body{overflow-y:auto}}@media(max-width: 1399.98px){.modal-fullscreen-xxl-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-xxl-down .modal-content{height:100%;border:0}.modal-fullscreen-xxl-down .modal-body{overflow-y:auto}}.tooltip{--bs-tooltip-zindex: 1080;--bs-tooltip-max-width: 200px;--bs-tooltip-padding-x: 0.5rem;--bs-tooltip-padding-y: 0.25rem;--bs-tooltip-margin: ;--bs-tooltip-font-size:0.875rem;--bs-tooltip-color: #FEFBF2;--bs-tooltip-bg: #000;--bs-tooltip-border-radius: 0.25rem;--bs-tooltip-opacity: 0.9;--bs-tooltip-arrow-width: 0.8rem;--bs-tooltip-arrow-height: 0.4rem;z-index:var(--bs-tooltip-zindex);display:block;margin:var(--bs-tooltip-margin);font-family:"Source Sans Pro",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,sans-serif,"Apple Color Emoji","Segoe UI Emoji","Segoe UI Symbol";font-style:normal;font-weight:400;line-height:1.5;text-align:left;text-align:start;text-decoration:none;text-shadow:none;text-transform:none;letter-spacing:normal;word-break:normal;white-space:normal;word-spacing:normal;line-break:auto;font-size:var(--bs-tooltip-font-size);word-wrap:break-word;opacity:0}.tooltip.show{opacity:var(--bs-tooltip-opacity)}.tooltip .tooltip-arrow{display:block;width:var(--bs-tooltip-arrow-width);height:var(--bs-tooltip-arrow-height)}.tooltip .tooltip-arrow::before{position:absolute;content:"";border-color:rgba(0,0,0,0);border-style:solid}.bs-tooltip-top .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=top] .tooltip-arrow{bottom:calc(-1*var(--bs-tooltip-arrow-height))}.bs-tooltip-top .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=top] .tooltip-arrow::before{top:-1px;border-width:var(--bs-tooltip-arrow-height) calc(var(--bs-tooltip-arrow-width)*.5) 0;border-top-color:var(--bs-tooltip-bg)}.bs-tooltip-end .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=right] .tooltip-arrow{left:calc(-1*var(--bs-tooltip-arrow-height));width:var(--bs-tooltip-arrow-height);height:var(--bs-tooltip-arrow-width)}.bs-tooltip-end .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=right] .tooltip-arrow::before{right:-1px;border-width:calc(var(--bs-tooltip-arrow-width)*.5) var(--bs-tooltip-arrow-height) calc(var(--bs-tooltip-arrow-width)*.5) 0;border-right-color:var(--bs-tooltip-bg)}.bs-tooltip-bottom .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=bottom] .tooltip-arrow{top:calc(-1*var(--bs-tooltip-arrow-height))}.bs-tooltip-bottom .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=bottom] .tooltip-arrow::before{bottom:-1px;border-width:0 calc(var(--bs-tooltip-arrow-width)*.5) var(--bs-tooltip-arrow-height);border-bottom-color:var(--bs-tooltip-bg)}.bs-tooltip-start .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=left] .tooltip-arrow{right:calc(-1*var(--bs-tooltip-arrow-height));width:var(--bs-tooltip-arrow-height);height:var(--bs-tooltip-arrow-width)}.bs-tooltip-start .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=left] .tooltip-arrow::before{left:-1px;border-width:calc(var(--bs-tooltip-arrow-width)*.5) 0 calc(var(--bs-tooltip-arrow-width)*.5) var(--bs-tooltip-arrow-height);border-left-color:var(--bs-tooltip-bg)}.tooltip-inner{max-width:var(--bs-tooltip-max-width);padding:var(--bs-tooltip-padding-y) var(--bs-tooltip-padding-x);color:var(--bs-tooltip-color);text-align:center;background-color:var(--bs-tooltip-bg)}.popover{--bs-popover-zindex: 1070;--bs-popover-max-width: 276px;--bs-popover-font-size:0.875rem;--bs-popover-bg: #FEFBF2;--bs-popover-border-width: 1px;--bs-popover-border-color: rgba(0, 0, 0, 0.175);--bs-popover-border-radius: 0.5rem;--bs-popover-inner-border-radius: calc(0.5rem - 1px);--bs-popover-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-popover-header-padding-x: 1rem;--bs-popover-header-padding-y: 0.5rem;--bs-popover-header-font-size:1rem;--bs-popover-header-color: inherit;--bs-popover-header-bg: #e9ecef;--bs-popover-body-padding-x: 1rem;--bs-popover-body-padding-y: 1rem;--bs-popover-body-color: #343a40;--bs-popover-arrow-width: 1rem;--bs-popover-arrow-height: 0.5rem;--bs-popover-arrow-border: var(--bs-popover-border-color);z-index:var(--bs-popover-zindex);display:block;max-width:var(--bs-popover-max-width);font-family:"Source Sans Pro",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,sans-serif,"Apple Color Emoji","Segoe UI Emoji","Segoe UI Symbol";font-style:normal;font-weight:400;line-height:1.5;text-align:left;text-align:start;text-decoration:none;text-shadow:none;text-transform:none;letter-spacing:normal;word-break:normal;white-space:normal;word-spacing:normal;line-break:auto;font-size:var(--bs-popover-font-size);word-wrap:break-word;background-color:var(--bs-popover-bg);background-clip:padding-box;border:var(--bs-popover-border-width) solid var(--bs-popover-border-color)}.popover .popover-arrow{display:block;width:var(--bs-popover-arrow-width);height:var(--bs-popover-arrow-height)}.popover .popover-arrow::before,.popover .popover-arrow::after{position:absolute;display:block;content:"";border-color:rgba(0,0,0,0);border-style:solid;border-width:0}.bs-popover-top>.popover-arrow,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow{bottom:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width))}.bs-popover-top>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::before,.bs-popover-top>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::after{border-width:var(--bs-popover-arrow-height) calc(var(--bs-popover-arrow-width)*.5) 0}.bs-popover-top>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::before{bottom:0;border-top-color:var(--bs-popover-arrow-border)}.bs-popover-top>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::after{bottom:var(--bs-popover-border-width);border-top-color:var(--bs-popover-bg)}.bs-popover-end>.popover-arrow,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow{left:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width));width:var(--bs-popover-arrow-height);height:var(--bs-popover-arrow-width)}.bs-popover-end>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::before,.bs-popover-end>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::after{border-width:calc(var(--bs-popover-arrow-width)*.5) var(--bs-popover-arrow-height) calc(var(--bs-popover-arrow-width)*.5) 0}.bs-popover-end>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::before{left:0;border-right-color:var(--bs-popover-arrow-border)}.bs-popover-end>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::after{left:var(--bs-popover-border-width);border-right-color:var(--bs-popover-bg)}.bs-popover-bottom>.popover-arrow,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow{top:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width))}.bs-popover-bottom>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::before,.bs-popover-bottom>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::after{border-width:0 calc(var(--bs-popover-arrow-width)*.5) var(--bs-popover-arrow-height)}.bs-popover-bottom>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::before{top:0;border-bottom-color:var(--bs-popover-arrow-border)}.bs-popover-bottom>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::after{top:var(--bs-popover-border-width);border-bottom-color:var(--bs-popover-bg)}.bs-popover-bottom .popover-header::before,.bs-popover-auto[data-popper-placement^=bottom] .popover-header::before{position:absolute;top:0;left:50%;display:block;width:var(--bs-popover-arrow-width);margin-left:calc(-0.5*var(--bs-popover-arrow-width));content:"";border-bottom:var(--bs-popover-border-width) solid var(--bs-popover-header-bg)}.bs-popover-start>.popover-arrow,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow{right:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width));width:var(--bs-popover-arrow-height);height:var(--bs-popover-arrow-width)}.bs-popover-start>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::before,.bs-popover-start>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::after{border-width:calc(var(--bs-popover-arrow-width)*.5) 0 calc(var(--bs-popover-arrow-width)*.5) var(--bs-popover-arrow-height)}.bs-popover-start>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::before{right:0;border-left-color:var(--bs-popover-arrow-border)}.bs-popover-start>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::after{right:var(--bs-popover-border-width);border-left-color:var(--bs-popover-bg)}.popover-header{padding:var(--bs-popover-header-padding-y) var(--bs-popover-header-padding-x);margin-bottom:0;font-size:var(--bs-popover-header-font-size);color:var(--bs-popover-header-color);background-color:var(--bs-popover-header-bg);border-bottom:var(--bs-popover-border-width) solid var(--bs-popover-border-color)}.popover-header:empty{display:none}.popover-body{padding:var(--bs-popover-body-padding-y) var(--bs-popover-body-padding-x);color:var(--bs-popover-body-color)}.carousel{position:relative}.carousel.pointer-event{touch-action:pan-y;-webkit-touch-action:pan-y;-moz-touch-action:pan-y;-ms-touch-action:pan-y;-o-touch-action:pan-y}.carousel-inner{position:relative;width:100%;overflow:hidden}.carousel-inner::after{display:block;clear:both;content:""}.carousel-item{position:relative;display:none;float:left;width:100%;margin-right:-100%;backface-visibility:hidden;-webkit-backface-visibility:hidden;-moz-backface-visibility:hidden;-ms-backface-visibility:hidden;-o-backface-visibility:hidden;transition:transform .6s ease-in-out}@media(prefers-reduced-motion: reduce){.carousel-item{transition:none}}.carousel-item.active,.carousel-item-next,.carousel-item-prev{display:block}.carousel-item-next:not(.carousel-item-start),.active.carousel-item-end{transform:translateX(100%)}.carousel-item-prev:not(.carousel-item-end),.active.carousel-item-start{transform:translateX(-100%)}.carousel-fade .carousel-item{opacity:0;transition-property:opacity;transform:none}.carousel-fade .carousel-item.active,.carousel-fade .carousel-item-next.carousel-item-start,.carousel-fade .carousel-item-prev.carousel-item-end{z-index:1;opacity:1}.carousel-fade .active.carousel-item-start,.carousel-fade .active.carousel-item-end{z-index:0;opacity:0;transition:opacity 0s .6s}@media(prefers-reduced-motion: reduce){.carousel-fade .active.carousel-item-start,.carousel-fade .active.carousel-item-end{transition:none}}.carousel-control-prev,.carousel-control-next{position:absolute;top:0;bottom:0;z-index:1;display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;justify-content:center;-webkit-justify-content:center;width:15%;padding:0;color:#fff;text-align:center;background:none;border:0;opacity:.5;transition:opacity .15s ease}@media(prefers-reduced-motion: reduce){.carousel-control-prev,.carousel-control-next{transition:none}}.carousel-control-prev:hover,.carousel-control-prev:focus,.carousel-control-next:hover,.carousel-control-next:focus{color:#fff;text-decoration:none;outline:0;opacity:.9}.carousel-control-prev{left:0}.carousel-control-next{right:0}.carousel-control-prev-icon,.carousel-control-next-icon{display:inline-block;width:2rem;height:2rem;background-repeat:no-repeat;background-position:50%;background-size:100% 100%}.carousel-control-prev-icon{background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23fff'%3e%3cpath d='M11.354 1.646a.5.5 0 0 1 0 .708L5.707 8l5.647 5.646a.5.5 0 0 1-.708.708l-6-6a.5.5 0 0 1 0-.708l6-6a.5.5 0 0 1 .708 0z'/%3e%3c/svg%3e")}.carousel-control-next-icon{background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23fff'%3e%3cpath d='M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e")}.carousel-indicators{position:absolute;right:0;bottom:0;left:0;z-index:2;display:flex;display:-webkit-flex;justify-content:center;-webkit-justify-content:center;padding:0;margin-right:15%;margin-bottom:1rem;margin-left:15%}.carousel-indicators [data-bs-target]{box-sizing:content-box;flex:0 1 auto;-webkit-flex:0 1 auto;width:30px;height:3px;padding:0;margin-right:3px;margin-left:3px;text-indent:-999px;cursor:pointer;background-color:#fff;background-clip:padding-box;border:0;border-top:10px solid rgba(0,0,0,0);border-bottom:10px solid rgba(0,0,0,0);opacity:.5;transition:opacity .6s ease}@media(prefers-reduced-motion: reduce){.carousel-indicators [data-bs-target]{transition:none}}.carousel-indicators .active{opacity:1}.carousel-caption{position:absolute;right:15%;bottom:1.25rem;left:15%;padding-top:1.25rem;padding-bottom:1.25rem;color:#fff;text-align:center}.carousel-dark .carousel-control-prev-icon,.carousel-dark .carousel-control-next-icon{filter:invert(1) grayscale(100)}.carousel-dark .carousel-indicators [data-bs-target]{background-color:#000}.carousel-dark .carousel-caption{color:#000}[data-bs-theme=dark] .carousel .carousel-control-prev-icon,[data-bs-theme=dark] .carousel .carousel-control-next-icon,[data-bs-theme=dark].carousel .carousel-control-prev-icon,[data-bs-theme=dark].carousel .carousel-control-next-icon{filter:invert(1) grayscale(100)}[data-bs-theme=dark] .carousel .carousel-indicators [data-bs-target],[data-bs-theme=dark].carousel .carousel-indicators [data-bs-target]{background-color:#000}[data-bs-theme=dark] .carousel .carousel-caption,[data-bs-theme=dark].carousel .carousel-caption{color:#000}.spinner-grow,.spinner-border{display:inline-block;width:var(--bs-spinner-width);height:var(--bs-spinner-height);vertical-align:var(--bs-spinner-vertical-align);border-radius:50%;animation:var(--bs-spinner-animation-speed) linear infinite var(--bs-spinner-animation-name)}@keyframes spinner-border{to{transform:rotate(360deg) /* rtl:ignore */}}.spinner-border{--bs-spinner-width: 2rem;--bs-spinner-height: 2rem;--bs-spinner-vertical-align: -0.125em;--bs-spinner-border-width: 0.25em;--bs-spinner-animation-speed: 0.75s;--bs-spinner-animation-name: spinner-border;border:var(--bs-spinner-border-width) solid currentcolor;border-right-color:rgba(0,0,0,0)}.spinner-border-sm{--bs-spinner-width: 1rem;--bs-spinner-height: 1rem;--bs-spinner-border-width: 0.2em}@keyframes spinner-grow{0%{transform:scale(0)}50%{opacity:1;transform:none}}.spinner-grow{--bs-spinner-width: 2rem;--bs-spinner-height: 2rem;--bs-spinner-vertical-align: -0.125em;--bs-spinner-animation-speed: 0.75s;--bs-spinner-animation-name: spinner-grow;background-color:currentcolor;opacity:0}.spinner-grow-sm{--bs-spinner-width: 1rem;--bs-spinner-height: 1rem}@media(prefers-reduced-motion: reduce){.spinner-border,.spinner-grow{--bs-spinner-animation-speed: 1.5s}}.offcanvas,.offcanvas-xxl,.offcanvas-xl,.offcanvas-lg,.offcanvas-md,.offcanvas-sm{--bs-offcanvas-zindex: 1045;--bs-offcanvas-width: 400px;--bs-offcanvas-height: 30vh;--bs-offcanvas-padding-x: 1rem;--bs-offcanvas-padding-y: 1rem;--bs-offcanvas-color: #343a40;--bs-offcanvas-bg: #FEFBF2;--bs-offcanvas-border-width: 1px;--bs-offcanvas-border-color: rgba(0, 0, 0, 0.175);--bs-offcanvas-box-shadow: 0 0.125rem 0.25rem rgba(0, 0, 0, 0.075);--bs-offcanvas-transition: transform 0.3s ease-in-out;--bs-offcanvas-title-line-height: 1.5}@media(max-width: 575.98px){.offcanvas-sm{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 575.98px)and (prefers-reduced-motion: reduce){.offcanvas-sm{transition:none}}@media(max-width: 575.98px){.offcanvas-sm.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-sm.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-sm.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-sm.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-sm.showing,.offcanvas-sm.show:not(.hiding){transform:none}.offcanvas-sm.showing,.offcanvas-sm.hiding,.offcanvas-sm.show{visibility:visible}}@media(min-width: 576px){.offcanvas-sm{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-sm .offcanvas-header{display:none}.offcanvas-sm .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 767.98px){.offcanvas-md{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 767.98px)and (prefers-reduced-motion: reduce){.offcanvas-md{transition:none}}@media(max-width: 767.98px){.offcanvas-md.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-md.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-md.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-md.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-md.showing,.offcanvas-md.show:not(.hiding){transform:none}.offcanvas-md.showing,.offcanvas-md.hiding,.offcanvas-md.show{visibility:visible}}@media(min-width: 768px){.offcanvas-md{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-md .offcanvas-header{display:none}.offcanvas-md .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 991.98px){.offcanvas-lg{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 991.98px)and (prefers-reduced-motion: reduce){.offcanvas-lg{transition:none}}@media(max-width: 991.98px){.offcanvas-lg.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-lg.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-lg.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-lg.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-lg.showing,.offcanvas-lg.show:not(.hiding){transform:none}.offcanvas-lg.showing,.offcanvas-lg.hiding,.offcanvas-lg.show{visibility:visible}}@media(min-width: 992px){.offcanvas-lg{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-lg .offcanvas-header{display:none}.offcanvas-lg .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 1199.98px){.offcanvas-xl{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 1199.98px)and (prefers-reduced-motion: reduce){.offcanvas-xl{transition:none}}@media(max-width: 1199.98px){.offcanvas-xl.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-xl.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-xl.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-xl.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-xl.showing,.offcanvas-xl.show:not(.hiding){transform:none}.offcanvas-xl.showing,.offcanvas-xl.hiding,.offcanvas-xl.show{visibility:visible}}@media(min-width: 1200px){.offcanvas-xl{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-xl .offcanvas-header{display:none}.offcanvas-xl .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 1399.98px){.offcanvas-xxl{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 1399.98px)and (prefers-reduced-motion: reduce){.offcanvas-xxl{transition:none}}@media(max-width: 1399.98px){.offcanvas-xxl.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-xxl.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-xxl.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-xxl.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-xxl.showing,.offcanvas-xxl.show:not(.hiding){transform:none}.offcanvas-xxl.showing,.offcanvas-xxl.hiding,.offcanvas-xxl.show{visibility:visible}}@media(min-width: 1400px){.offcanvas-xxl{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-xxl .offcanvas-header{display:none}.offcanvas-xxl .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}.offcanvas{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}@media(prefers-reduced-motion: reduce){.offcanvas{transition:none}}.offcanvas.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas.showing,.offcanvas.show:not(.hiding){transform:none}.offcanvas.showing,.offcanvas.hiding,.offcanvas.show{visibility:visible}.offcanvas-backdrop{position:fixed;top:0;left:0;z-index:1040;width:100vw;height:100vh;background-color:#000}.offcanvas-backdrop.fade{opacity:0}.offcanvas-backdrop.show{opacity:.5}.offcanvas-header{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between;padding:var(--bs-offcanvas-padding-y) var(--bs-offcanvas-padding-x)}.offcanvas-header .btn-close{padding:calc(var(--bs-offcanvas-padding-y)*.5) calc(var(--bs-offcanvas-padding-x)*.5);margin-top:calc(-0.5*var(--bs-offcanvas-padding-y));margin-right:calc(-0.5*var(--bs-offcanvas-padding-x));margin-bottom:calc(-0.5*var(--bs-offcanvas-padding-y))}.offcanvas-title{margin-bottom:0;line-height:var(--bs-offcanvas-title-line-height)}.offcanvas-body{flex-grow:1;-webkit-flex-grow:1;padding:var(--bs-offcanvas-padding-y) var(--bs-offcanvas-padding-x);overflow-y:auto}.placeholder{display:inline-block;min-height:1em;vertical-align:middle;cursor:wait;background-color:currentcolor;opacity:.5}.placeholder.btn::before{display:inline-block;content:""}.placeholder-xs{min-height:.6em}.placeholder-sm{min-height:.8em}.placeholder-lg{min-height:1.2em}.placeholder-glow .placeholder{animation:placeholder-glow 2s ease-in-out infinite}@keyframes placeholder-glow{50%{opacity:.2}}.placeholder-wave{mask-image:linear-gradient(130deg, #000 55%, rgba(0, 0, 0, 0.8) 75%, #000 95%);-webkit-mask-image:linear-gradient(130deg, #000 55%, rgba(0, 0, 0, 0.8) 75%, #000 95%);mask-size:200% 100%;-webkit-mask-size:200% 100%;animation:placeholder-wave 2s linear infinite}@keyframes placeholder-wave{100%{mask-position:-200% 0%;-webkit-mask-position:-200% 0%}}.clearfix::after{display:block;clear:both;content:""}.text-bg-default{color:#fff !important;background-color:RGBA(var(--bs-default-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-primary{color:#fff !important;background-color:RGBA(var(--bs-primary-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-secondary{color:#fff !important;background-color:RGBA(var(--bs-secondary-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-success{color:#fff !important;background-color:RGBA(var(--bs-success-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-info{color:#fff !important;background-color:RGBA(var(--bs-info-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-warning{color:#fff !important;background-color:RGBA(var(--bs-warning-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-danger{color:#fff !important;background-color:RGBA(var(--bs-danger-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-light{color:#000 !important;background-color:RGBA(var(--bs-light-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-dark{color:#fff !important;background-color:RGBA(var(--bs-dark-rgb), var(--bs-bg-opacity, 1)) !important}.link-default{color:RGBA(var(--bs-default-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-default-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-default:hover,.link-default:focus{color:RGBA(42, 46, 51, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(42, 46, 51, var(--bs-link-underline-opacity, 1)) !important}.link-primary{color:RGBA(var(--bs-primary-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-primary-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-primary:hover,.link-primary:focus{color:RGBA(31, 102, 182, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(31, 102, 182, var(--bs-link-underline-opacity, 1)) !important}.link-secondary{color:RGBA(var(--bs-secondary-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-secondary-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-secondary:hover,.link-secondary:focus{color:RGBA(42, 46, 51, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(42, 46, 51, var(--bs-link-underline-opacity, 1)) !important}.link-success{color:RGBA(var(--bs-success-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-success-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-success:hover,.link-success:focus{color:RGBA(50, 146, 19, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(50, 146, 19, var(--bs-link-underline-opacity, 1)) !important}.link-info{color:RGBA(var(--bs-info-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-info-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-info:hover,.link-info:focus{color:RGBA(122, 67, 150, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(122, 67, 150, var(--bs-link-underline-opacity, 1)) !important}.link-warning{color:RGBA(var(--bs-warning-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-warning-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-warning:hover,.link-warning:focus{color:RGBA(204, 94, 19, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(204, 94, 19, var(--bs-link-underline-opacity, 1)) !important}.link-danger{color:RGBA(var(--bs-danger-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-danger-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-danger:hover,.link-danger:focus{color:RGBA(204, 0, 46, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(204, 0, 46, var(--bs-link-underline-opacity, 1)) !important}.link-light{color:RGBA(var(--bs-light-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-light-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-light:hover,.link-light:focus{color:RGBA(249, 250, 251, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(249, 250, 251, var(--bs-link-underline-opacity, 1)) !important}.link-dark{color:RGBA(var(--bs-dark-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-dark-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-dark:hover,.link-dark:focus{color:RGBA(42, 46, 51, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(42, 46, 51, var(--bs-link-underline-opacity, 1)) !important}.link-body-emphasis{color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-body-emphasis:hover,.link-body-emphasis:focus{color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-opacity, 0.75)) !important;text-decoration-color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-underline-opacity, 0.75)) !important}.focus-ring:focus{outline:0;box-shadow:var(--bs-focus-ring-x, 0) var(--bs-focus-ring-y, 0) var(--bs-focus-ring-blur, 0) var(--bs-focus-ring-width) var(--bs-focus-ring-color)}.icon-link{display:inline-flex;gap:.375rem;align-items:center;-webkit-align-items:center;text-decoration-color:rgba(var(--bs-link-color-rgb), var(--bs-link-opacity, 0.5));text-underline-offset:.25em;backface-visibility:hidden;-webkit-backface-visibility:hidden;-moz-backface-visibility:hidden;-ms-backface-visibility:hidden;-o-backface-visibility:hidden}.icon-link>.bi{flex-shrink:0;-webkit-flex-shrink:0;width:1em;height:1em;fill:currentcolor;transition:.2s ease-in-out transform}@media(prefers-reduced-motion: reduce){.icon-link>.bi{transition:none}}.icon-link-hover:hover>.bi,.icon-link-hover:focus-visible>.bi{transform:var(--bs-icon-link-transform, translate3d(0.25em, 0, 0))}.ratio{position:relative;width:100%}.ratio::before{display:block;padding-top:var(--bs-aspect-ratio);content:""}.ratio>*{position:absolute;top:0;left:0;width:100%;height:100%}.ratio-1x1{--bs-aspect-ratio: 100%}.ratio-4x3{--bs-aspect-ratio: 75%}.ratio-16x9{--bs-aspect-ratio: 56.25%}.ratio-21x9{--bs-aspect-ratio: 42.8571428571%}.fixed-top{position:fixed;top:0;right:0;left:0;z-index:1030}.fixed-bottom{position:fixed;right:0;bottom:0;left:0;z-index:1030}.sticky-top{position:sticky;top:0;z-index:1020}.sticky-bottom{position:sticky;bottom:0;z-index:1020}@media(min-width: 576px){.sticky-sm-top{position:sticky;top:0;z-index:1020}.sticky-sm-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 768px){.sticky-md-top{position:sticky;top:0;z-index:1020}.sticky-md-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 992px){.sticky-lg-top{position:sticky;top:0;z-index:1020}.sticky-lg-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 1200px){.sticky-xl-top{position:sticky;top:0;z-index:1020}.sticky-xl-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 1400px){.sticky-xxl-top{position:sticky;top:0;z-index:1020}.sticky-xxl-bottom{position:sticky;bottom:0;z-index:1020}}.hstack{display:flex;display:-webkit-flex;flex-direction:row;-webkit-flex-direction:row;align-items:center;-webkit-align-items:center;align-self:stretch;-webkit-align-self:stretch}.vstack{display:flex;display:-webkit-flex;flex:1 1 auto;-webkit-flex:1 1 auto;flex-direction:column;-webkit-flex-direction:column;align-self:stretch;-webkit-align-self:stretch}.visually-hidden,.visually-hidden-focusable:not(:focus):not(:focus-within){width:1px !important;height:1px !important;padding:0 !important;margin:-1px !important;overflow:hidden !important;clip:rect(0, 0, 0, 0) !important;white-space:nowrap !important;border:0 !important}.visually-hidden:not(caption),.visually-hidden-focusable:not(:focus):not(:focus-within):not(caption){position:absolute !important}.stretched-link::after{position:absolute;top:0;right:0;bottom:0;left:0;z-index:1;content:""}.text-truncate{overflow:hidden;text-overflow:ellipsis;white-space:nowrap}.vr{display:inline-block;align-self:stretch;-webkit-align-self:stretch;width:1px;min-height:1em;background-color:currentcolor;opacity:.25}.align-baseline{vertical-align:baseline !important}.align-top{vertical-align:top !important}.align-middle{vertical-align:middle !important}.align-bottom{vertical-align:bottom !important}.align-text-bottom{vertical-align:text-bottom !important}.align-text-top{vertical-align:text-top !important}.float-start{float:left !important}.float-end{float:right !important}.float-none{float:none !important}.object-fit-contain{object-fit:contain !important}.object-fit-cover{object-fit:cover !important}.object-fit-fill{object-fit:fill !important}.object-fit-scale{object-fit:scale-down !important}.object-fit-none{object-fit:none !important}.opacity-0{opacity:0 !important}.opacity-25{opacity:.25 !important}.opacity-50{opacity:.5 !important}.opacity-75{opacity:.75 !important}.opacity-100{opacity:1 !important}.overflow-auto{overflow:auto !important}.overflow-hidden{overflow:hidden !important}.overflow-visible{overflow:visible !important}.overflow-scroll{overflow:scroll !important}.overflow-x-auto{overflow-x:auto !important}.overflow-x-hidden{overflow-x:hidden !important}.overflow-x-visible{overflow-x:visible !important}.overflow-x-scroll{overflow-x:scroll !important}.overflow-y-auto{overflow-y:auto !important}.overflow-y-hidden{overflow-y:hidden !important}.overflow-y-visible{overflow-y:visible !important}.overflow-y-scroll{overflow-y:scroll !important}.d-inline{display:inline !important}.d-inline-block{display:inline-block !important}.d-block{display:block !important}.d-grid{display:grid !important}.d-inline-grid{display:inline-grid !important}.d-table{display:table !important}.d-table-row{display:table-row !important}.d-table-cell{display:table-cell !important}.d-flex{display:flex !important}.d-inline-flex{display:inline-flex !important}.d-none{display:none !important}.shadow{box-shadow:0 .5rem 1rem rgba(0,0,0,.15) !important}.shadow-sm{box-shadow:0 .125rem .25rem rgba(0,0,0,.075) !important}.shadow-lg{box-shadow:0 1rem 3rem rgba(0,0,0,.175) !important}.shadow-none{box-shadow:none !important}.focus-ring-default{--bs-focus-ring-color: rgba(var(--bs-default-rgb), var(--bs-focus-ring-opacity))}.focus-ring-primary{--bs-focus-ring-color: rgba(var(--bs-primary-rgb), var(--bs-focus-ring-opacity))}.focus-ring-secondary{--bs-focus-ring-color: rgba(var(--bs-secondary-rgb), var(--bs-focus-ring-opacity))}.focus-ring-success{--bs-focus-ring-color: rgba(var(--bs-success-rgb), var(--bs-focus-ring-opacity))}.focus-ring-info{--bs-focus-ring-color: rgba(var(--bs-info-rgb), var(--bs-focus-ring-opacity))}.focus-ring-warning{--bs-focus-ring-color: rgba(var(--bs-warning-rgb), var(--bs-focus-ring-opacity))}.focus-ring-danger{--bs-focus-ring-color: rgba(var(--bs-danger-rgb), var(--bs-focus-ring-opacity))}.focus-ring-light{--bs-focus-ring-color: rgba(var(--bs-light-rgb), var(--bs-focus-ring-opacity))}.focus-ring-dark{--bs-focus-ring-color: rgba(var(--bs-dark-rgb), var(--bs-focus-ring-opacity))}.position-static{position:static !important}.position-relative{position:relative !important}.position-absolute{position:absolute !important}.position-fixed{position:fixed !important}.position-sticky{position:sticky !important}.top-0{top:0 !important}.top-50{top:50% !important}.top-100{top:100% !important}.bottom-0{bottom:0 !important}.bottom-50{bottom:50% !important}.bottom-100{bottom:100% !important}.start-0{left:0 !important}.start-50{left:50% !important}.start-100{left:100% !important}.end-0{right:0 !important}.end-50{right:50% !important}.end-100{right:100% !important}.translate-middle{transform:translate(-50%, -50%) !important}.translate-middle-x{transform:translateX(-50%) !important}.translate-middle-y{transform:translateY(-50%) !important}.border{border:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-0{border:0 !important}.border-top{border-top:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-top-0{border-top:0 !important}.border-end{border-right:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-end-0{border-right:0 !important}.border-bottom{border-bottom:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-bottom-0{border-bottom:0 !important}.border-start{border-left:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-start-0{border-left:0 !important}.border-default{--bs-border-opacity: 1;border-color:rgba(var(--bs-default-rgb), var(--bs-border-opacity)) !important}.border-primary{--bs-border-opacity: 1;border-color:rgba(var(--bs-primary-rgb), var(--bs-border-opacity)) !important}.border-secondary{--bs-border-opacity: 1;border-color:rgba(var(--bs-secondary-rgb), var(--bs-border-opacity)) !important}.border-success{--bs-border-opacity: 1;border-color:rgba(var(--bs-success-rgb), var(--bs-border-opacity)) !important}.border-info{--bs-border-opacity: 1;border-color:rgba(var(--bs-info-rgb), var(--bs-border-opacity)) !important}.border-warning{--bs-border-opacity: 1;border-color:rgba(var(--bs-warning-rgb), var(--bs-border-opacity)) !important}.border-danger{--bs-border-opacity: 1;border-color:rgba(var(--bs-danger-rgb), var(--bs-border-opacity)) !important}.border-light{--bs-border-opacity: 1;border-color:rgba(var(--bs-light-rgb), var(--bs-border-opacity)) !important}.border-dark{--bs-border-opacity: 1;border-color:rgba(var(--bs-dark-rgb), var(--bs-border-opacity)) !important}.border-black{--bs-border-opacity: 1;border-color:rgba(var(--bs-black-rgb), var(--bs-border-opacity)) !important}.border-white{--bs-border-opacity: 1;border-color:rgba(var(--bs-white-rgb), var(--bs-border-opacity)) !important}.border-primary-subtle{border-color:var(--bs-primary-border-subtle) !important}.border-secondary-subtle{border-color:var(--bs-secondary-border-subtle) !important}.border-success-subtle{border-color:var(--bs-success-border-subtle) !important}.border-info-subtle{border-color:var(--bs-info-border-subtle) !important}.border-warning-subtle{border-color:var(--bs-warning-border-subtle) !important}.border-danger-subtle{border-color:var(--bs-danger-border-subtle) !important}.border-light-subtle{border-color:var(--bs-light-border-subtle) !important}.border-dark-subtle{border-color:var(--bs-dark-border-subtle) !important}.border-1{border-width:1px !important}.border-2{border-width:2px !important}.border-3{border-width:3px !important}.border-4{border-width:4px !important}.border-5{border-width:5px !important}.border-opacity-10{--bs-border-opacity: 0.1}.border-opacity-25{--bs-border-opacity: 0.25}.border-opacity-50{--bs-border-opacity: 0.5}.border-opacity-75{--bs-border-opacity: 0.75}.border-opacity-100{--bs-border-opacity: 1}.w-25{width:25% !important}.w-50{width:50% !important}.w-75{width:75% !important}.w-100{width:100% !important}.w-auto{width:auto !important}.mw-100{max-width:100% !important}.vw-100{width:100vw !important}.min-vw-100{min-width:100vw !important}.h-25{height:25% !important}.h-50{height:50% !important}.h-75{height:75% !important}.h-100{height:100% !important}.h-auto{height:auto !important}.mh-100{max-height:100% !important}.vh-100{height:100vh !important}.min-vh-100{min-height:100vh !important}.flex-fill{flex:1 1 auto !important}.flex-row{flex-direction:row !important}.flex-column{flex-direction:column !important}.flex-row-reverse{flex-direction:row-reverse !important}.flex-column-reverse{flex-direction:column-reverse !important}.flex-grow-0{flex-grow:0 !important}.flex-grow-1{flex-grow:1 !important}.flex-shrink-0{flex-shrink:0 !important}.flex-shrink-1{flex-shrink:1 !important}.flex-wrap{flex-wrap:wrap !important}.flex-nowrap{flex-wrap:nowrap !important}.flex-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-start{justify-content:flex-start !important}.justify-content-end{justify-content:flex-end !important}.justify-content-center{justify-content:center !important}.justify-content-between{justify-content:space-between !important}.justify-content-around{justify-content:space-around !important}.justify-content-evenly{justify-content:space-evenly !important}.align-items-start{align-items:flex-start !important}.align-items-end{align-items:flex-end !important}.align-items-center{align-items:center !important}.align-items-baseline{align-items:baseline !important}.align-items-stretch{align-items:stretch !important}.align-content-start{align-content:flex-start !important}.align-content-end{align-content:flex-end !important}.align-content-center{align-content:center !important}.align-content-between{align-content:space-between !important}.align-content-around{align-content:space-around !important}.align-content-stretch{align-content:stretch !important}.align-self-auto{align-self:auto !important}.align-self-start{align-self:flex-start !important}.align-self-end{align-self:flex-end !important}.align-self-center{align-self:center !important}.align-self-baseline{align-self:baseline !important}.align-self-stretch{align-self:stretch !important}.order-first{order:-1 !important}.order-0{order:0 !important}.order-1{order:1 !important}.order-2{order:2 !important}.order-3{order:3 !important}.order-4{order:4 !important}.order-5{order:5 !important}.order-last{order:6 !important}.m-0{margin:0 !important}.m-1{margin:.25rem !important}.m-2{margin:.5rem !important}.m-3{margin:1rem !important}.m-4{margin:1.5rem !important}.m-5{margin:3rem !important}.m-auto{margin:auto !important}.mx-0{margin-right:0 !important;margin-left:0 !important}.mx-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-3{margin-right:1rem !important;margin-left:1rem !important}.mx-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-5{margin-right:3rem !important;margin-left:3rem !important}.mx-auto{margin-right:auto !important;margin-left:auto !important}.my-0{margin-top:0 !important;margin-bottom:0 !important}.my-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-0{margin-top:0 !important}.mt-1{margin-top:.25rem !important}.mt-2{margin-top:.5rem !important}.mt-3{margin-top:1rem !important}.mt-4{margin-top:1.5rem !important}.mt-5{margin-top:3rem !important}.mt-auto{margin-top:auto !important}.me-0{margin-right:0 !important}.me-1{margin-right:.25rem !important}.me-2{margin-right:.5rem !important}.me-3{margin-right:1rem !important}.me-4{margin-right:1.5rem !important}.me-5{margin-right:3rem !important}.me-auto{margin-right:auto !important}.mb-0{margin-bottom:0 !important}.mb-1{margin-bottom:.25rem !important}.mb-2{margin-bottom:.5rem !important}.mb-3{margin-bottom:1rem !important}.mb-4{margin-bottom:1.5rem !important}.mb-5{margin-bottom:3rem !important}.mb-auto{margin-bottom:auto !important}.ms-0{margin-left:0 !important}.ms-1{margin-left:.25rem !important}.ms-2{margin-left:.5rem !important}.ms-3{margin-left:1rem !important}.ms-4{margin-left:1.5rem !important}.ms-5{margin-left:3rem !important}.ms-auto{margin-left:auto !important}.p-0{padding:0 !important}.p-1{padding:.25rem !important}.p-2{padding:.5rem !important}.p-3{padding:1rem !important}.p-4{padding:1.5rem !important}.p-5{padding:3rem !important}.px-0{padding-right:0 !important;padding-left:0 !important}.px-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-3{padding-right:1rem !important;padding-left:1rem !important}.px-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-5{padding-right:3rem !important;padding-left:3rem !important}.py-0{padding-top:0 !important;padding-bottom:0 !important}.py-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-0{padding-top:0 !important}.pt-1{padding-top:.25rem !important}.pt-2{padding-top:.5rem !important}.pt-3{padding-top:1rem !important}.pt-4{padding-top:1.5rem !important}.pt-5{padding-top:3rem !important}.pe-0{padding-right:0 !important}.pe-1{padding-right:.25rem !important}.pe-2{padding-right:.5rem !important}.pe-3{padding-right:1rem !important}.pe-4{padding-right:1.5rem !important}.pe-5{padding-right:3rem !important}.pb-0{padding-bottom:0 !important}.pb-1{padding-bottom:.25rem !important}.pb-2{padding-bottom:.5rem !important}.pb-3{padding-bottom:1rem !important}.pb-4{padding-bottom:1.5rem !important}.pb-5{padding-bottom:3rem !important}.ps-0{padding-left:0 !important}.ps-1{padding-left:.25rem !important}.ps-2{padding-left:.5rem !important}.ps-3{padding-left:1rem !important}.ps-4{padding-left:1.5rem !important}.ps-5{padding-left:3rem !important}.gap-0{gap:0 !important}.gap-1{gap:.25rem !important}.gap-2{gap:.5rem !important}.gap-3{gap:1rem !important}.gap-4{gap:1.5rem !important}.gap-5{gap:3rem !important}.row-gap-0{row-gap:0 !important}.row-gap-1{row-gap:.25rem !important}.row-gap-2{row-gap:.5rem !important}.row-gap-3{row-gap:1rem !important}.row-gap-4{row-gap:1.5rem !important}.row-gap-5{row-gap:3rem !important}.column-gap-0{column-gap:0 !important}.column-gap-1{column-gap:.25rem !important}.column-gap-2{column-gap:.5rem !important}.column-gap-3{column-gap:1rem !important}.column-gap-4{column-gap:1.5rem !important}.column-gap-5{column-gap:3rem !important}.font-monospace{font-family:var(--bs-font-monospace) !important}.fs-1{font-size:calc(1.325rem + 0.9vw) !important}.fs-2{font-size:calc(1.29rem + 0.48vw) !important}.fs-3{font-size:calc(1.27rem + 0.24vw) !important}.fs-4{font-size:1.25rem !important}.fs-5{font-size:1.1rem !important}.fs-6{font-size:1rem !important}.fst-italic{font-style:italic !important}.fst-normal{font-style:normal !important}.fw-lighter{font-weight:lighter !important}.fw-light{font-weight:300 !important}.fw-normal{font-weight:400 !important}.fw-medium{font-weight:500 !important}.fw-semibold{font-weight:600 !important}.fw-bold{font-weight:700 !important}.fw-bolder{font-weight:bolder !important}.lh-1{line-height:1 !important}.lh-sm{line-height:1.25 !important}.lh-base{line-height:1.5 !important}.lh-lg{line-height:2 !important}.text-start{text-align:left !important}.text-end{text-align:right !important}.text-center{text-align:center !important}.text-decoration-none{text-decoration:none !important}.text-decoration-underline{text-decoration:underline !important}.text-decoration-line-through{text-decoration:line-through !important}.text-lowercase{text-transform:lowercase !important}.text-uppercase{text-transform:uppercase !important}.text-capitalize{text-transform:capitalize !important}.text-wrap{white-space:normal !important}.text-nowrap{white-space:nowrap !important}.text-break{word-wrap:break-word !important;word-break:break-word !important}.text-default{--bs-text-opacity: 1;color:rgba(var(--bs-default-rgb), var(--bs-text-opacity)) !important}.text-primary{--bs-text-opacity: 1;color:rgba(var(--bs-primary-rgb), var(--bs-text-opacity)) !important}.text-secondary{--bs-text-opacity: 1;color:rgba(var(--bs-secondary-rgb), var(--bs-text-opacity)) !important}.text-success{--bs-text-opacity: 1;color:rgba(var(--bs-success-rgb), var(--bs-text-opacity)) !important}.text-info{--bs-text-opacity: 1;color:rgba(var(--bs-info-rgb), var(--bs-text-opacity)) !important}.text-warning{--bs-text-opacity: 1;color:rgba(var(--bs-warning-rgb), var(--bs-text-opacity)) !important}.text-danger{--bs-text-opacity: 1;color:rgba(var(--bs-danger-rgb), var(--bs-text-opacity)) !important}.text-light{--bs-text-opacity: 1;color:rgba(var(--bs-light-rgb), var(--bs-text-opacity)) !important}.text-dark{--bs-text-opacity: 1;color:rgba(var(--bs-dark-rgb), var(--bs-text-opacity)) !important}.text-black{--bs-text-opacity: 1;color:rgba(var(--bs-black-rgb), var(--bs-text-opacity)) !important}.text-white{--bs-text-opacity: 1;color:rgba(var(--bs-white-rgb), var(--bs-text-opacity)) !important}.text-body{--bs-text-opacity: 1;color:rgba(var(--bs-body-color-rgb), var(--bs-text-opacity)) !important}.text-muted{--bs-text-opacity: 1;color:var(--bs-secondary-color) !important}.text-black-50{--bs-text-opacity: 1;color:rgba(0,0,0,.5) !important}.text-white-50{--bs-text-opacity: 1;color:rgba(255,255,255,.5) !important}.text-body-secondary{--bs-text-opacity: 1;color:var(--bs-secondary-color) !important}.text-body-tertiary{--bs-text-opacity: 1;color:var(--bs-tertiary-color) !important}.text-body-emphasis{--bs-text-opacity: 1;color:var(--bs-emphasis-color) !important}.text-reset{--bs-text-opacity: 1;color:inherit !important}.text-opacity-25{--bs-text-opacity: 0.25}.text-opacity-50{--bs-text-opacity: 0.5}.text-opacity-75{--bs-text-opacity: 0.75}.text-opacity-100{--bs-text-opacity: 1}.text-primary-emphasis{color:var(--bs-primary-text-emphasis) !important}.text-secondary-emphasis{color:var(--bs-secondary-text-emphasis) !important}.text-success-emphasis{color:var(--bs-success-text-emphasis) !important}.text-info-emphasis{color:var(--bs-info-text-emphasis) !important}.text-warning-emphasis{color:var(--bs-warning-text-emphasis) !important}.text-danger-emphasis{color:var(--bs-danger-text-emphasis) !important}.text-light-emphasis{color:var(--bs-light-text-emphasis) !important}.text-dark-emphasis{color:var(--bs-dark-text-emphasis) !important}.link-opacity-10{--bs-link-opacity: 0.1}.link-opacity-10-hover:hover{--bs-link-opacity: 0.1}.link-opacity-25{--bs-link-opacity: 0.25}.link-opacity-25-hover:hover{--bs-link-opacity: 0.25}.link-opacity-50{--bs-link-opacity: 0.5}.link-opacity-50-hover:hover{--bs-link-opacity: 0.5}.link-opacity-75{--bs-link-opacity: 0.75}.link-opacity-75-hover:hover{--bs-link-opacity: 0.75}.link-opacity-100{--bs-link-opacity: 1}.link-opacity-100-hover:hover{--bs-link-opacity: 1}.link-offset-1{text-underline-offset:.125em !important}.link-offset-1-hover:hover{text-underline-offset:.125em !important}.link-offset-2{text-underline-offset:.25em !important}.link-offset-2-hover:hover{text-underline-offset:.25em !important}.link-offset-3{text-underline-offset:.375em !important}.link-offset-3-hover:hover{text-underline-offset:.375em !important}.link-underline-default{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-default-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-primary{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-primary-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-secondary{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-secondary-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-success{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-success-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-info{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-info-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-warning{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-warning-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-danger{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-danger-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-light{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-light-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-dark{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-dark-rgb), var(--bs-link-underline-opacity)) !important}.link-underline{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-link-color-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-underline-opacity-0{--bs-link-underline-opacity: 0}.link-underline-opacity-0-hover:hover{--bs-link-underline-opacity: 0}.link-underline-opacity-10{--bs-link-underline-opacity: 0.1}.link-underline-opacity-10-hover:hover{--bs-link-underline-opacity: 0.1}.link-underline-opacity-25{--bs-link-underline-opacity: 0.25}.link-underline-opacity-25-hover:hover{--bs-link-underline-opacity: 0.25}.link-underline-opacity-50{--bs-link-underline-opacity: 0.5}.link-underline-opacity-50-hover:hover{--bs-link-underline-opacity: 0.5}.link-underline-opacity-75{--bs-link-underline-opacity: 0.75}.link-underline-opacity-75-hover:hover{--bs-link-underline-opacity: 0.75}.link-underline-opacity-100{--bs-link-underline-opacity: 1}.link-underline-opacity-100-hover:hover{--bs-link-underline-opacity: 1}.bg-default{--bs-bg-opacity: 1;background-color:rgba(var(--bs-default-rgb), var(--bs-bg-opacity)) !important}.bg-primary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-primary-rgb), var(--bs-bg-opacity)) !important}.bg-secondary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-secondary-rgb), var(--bs-bg-opacity)) !important}.bg-success{--bs-bg-opacity: 1;background-color:rgba(var(--bs-success-rgb), var(--bs-bg-opacity)) !important}.bg-info{--bs-bg-opacity: 1;background-color:rgba(var(--bs-info-rgb), var(--bs-bg-opacity)) !important}.bg-warning{--bs-bg-opacity: 1;background-color:rgba(var(--bs-warning-rgb), var(--bs-bg-opacity)) !important}.bg-danger{--bs-bg-opacity: 1;background-color:rgba(var(--bs-danger-rgb), var(--bs-bg-opacity)) !important}.bg-light{--bs-bg-opacity: 1;background-color:rgba(var(--bs-light-rgb), var(--bs-bg-opacity)) !important}.bg-dark{--bs-bg-opacity: 1;background-color:rgba(var(--bs-dark-rgb), var(--bs-bg-opacity)) !important}.bg-black{--bs-bg-opacity: 1;background-color:rgba(var(--bs-black-rgb), var(--bs-bg-opacity)) !important}.bg-white{--bs-bg-opacity: 1;background-color:rgba(var(--bs-white-rgb), var(--bs-bg-opacity)) !important}.bg-body{--bs-bg-opacity: 1;background-color:rgba(var(--bs-body-bg-rgb), var(--bs-bg-opacity)) !important}.bg-transparent{--bs-bg-opacity: 1;background-color:rgba(0,0,0,0) !important}.bg-body-secondary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-secondary-bg-rgb), var(--bs-bg-opacity)) !important}.bg-body-tertiary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-tertiary-bg-rgb), var(--bs-bg-opacity)) !important}.bg-opacity-10{--bs-bg-opacity: 0.1}.bg-opacity-25{--bs-bg-opacity: 0.25}.bg-opacity-50{--bs-bg-opacity: 0.5}.bg-opacity-75{--bs-bg-opacity: 0.75}.bg-opacity-100{--bs-bg-opacity: 1}.bg-primary-subtle{background-color:var(--bs-primary-bg-subtle) !important}.bg-secondary-subtle{background-color:var(--bs-secondary-bg-subtle) !important}.bg-success-subtle{background-color:var(--bs-success-bg-subtle) !important}.bg-info-subtle{background-color:var(--bs-info-bg-subtle) !important}.bg-warning-subtle{background-color:var(--bs-warning-bg-subtle) !important}.bg-danger-subtle{background-color:var(--bs-danger-bg-subtle) !important}.bg-light-subtle{background-color:var(--bs-light-bg-subtle) !important}.bg-dark-subtle{background-color:var(--bs-dark-bg-subtle) !important}.bg-gradient{background-image:var(--bs-gradient) !important}.user-select-all{user-select:all !important}.user-select-auto{user-select:auto !important}.user-select-none{user-select:none !important}.pe-none{pointer-events:none !important}.pe-auto{pointer-events:auto !important}.rounded{border-radius:var(--bs-border-radius) !important}.rounded-0{border-radius:0 !important}.rounded-1{border-radius:var(--bs-border-radius-sm) !important}.rounded-2{border-radius:var(--bs-border-radius) !important}.rounded-3{border-radius:var(--bs-border-radius-lg) !important}.rounded-4{border-radius:var(--bs-border-radius-xl) !important}.rounded-5{border-radius:var(--bs-border-radius-xxl) !important}.rounded-circle{border-radius:50% !important}.rounded-pill{border-radius:var(--bs-border-radius-pill) !important}.rounded-top{border-top-left-radius:var(--bs-border-radius) !important;border-top-right-radius:var(--bs-border-radius) !important}.rounded-top-0{border-top-left-radius:0 !important;border-top-right-radius:0 !important}.rounded-top-1{border-top-left-radius:var(--bs-border-radius-sm) !important;border-top-right-radius:var(--bs-border-radius-sm) !important}.rounded-top-2{border-top-left-radius:var(--bs-border-radius) !important;border-top-right-radius:var(--bs-border-radius) !important}.rounded-top-3{border-top-left-radius:var(--bs-border-radius-lg) !important;border-top-right-radius:var(--bs-border-radius-lg) !important}.rounded-top-4{border-top-left-radius:var(--bs-border-radius-xl) !important;border-top-right-radius:var(--bs-border-radius-xl) !important}.rounded-top-5{border-top-left-radius:var(--bs-border-radius-xxl) !important;border-top-right-radius:var(--bs-border-radius-xxl) !important}.rounded-top-circle{border-top-left-radius:50% !important;border-top-right-radius:50% !important}.rounded-top-pill{border-top-left-radius:var(--bs-border-radius-pill) !important;border-top-right-radius:var(--bs-border-radius-pill) !important}.rounded-end{border-top-right-radius:var(--bs-border-radius) !important;border-bottom-right-radius:var(--bs-border-radius) !important}.rounded-end-0{border-top-right-radius:0 !important;border-bottom-right-radius:0 !important}.rounded-end-1{border-top-right-radius:var(--bs-border-radius-sm) !important;border-bottom-right-radius:var(--bs-border-radius-sm) !important}.rounded-end-2{border-top-right-radius:var(--bs-border-radius) !important;border-bottom-right-radius:var(--bs-border-radius) !important}.rounded-end-3{border-top-right-radius:var(--bs-border-radius-lg) !important;border-bottom-right-radius:var(--bs-border-radius-lg) !important}.rounded-end-4{border-top-right-radius:var(--bs-border-radius-xl) !important;border-bottom-right-radius:var(--bs-border-radius-xl) !important}.rounded-end-5{border-top-right-radius:var(--bs-border-radius-xxl) !important;border-bottom-right-radius:var(--bs-border-radius-xxl) !important}.rounded-end-circle{border-top-right-radius:50% !important;border-bottom-right-radius:50% !important}.rounded-end-pill{border-top-right-radius:var(--bs-border-radius-pill) !important;border-bottom-right-radius:var(--bs-border-radius-pill) !important}.rounded-bottom{border-bottom-right-radius:var(--bs-border-radius) !important;border-bottom-left-radius:var(--bs-border-radius) !important}.rounded-bottom-0{border-bottom-right-radius:0 !important;border-bottom-left-radius:0 !important}.rounded-bottom-1{border-bottom-right-radius:var(--bs-border-radius-sm) !important;border-bottom-left-radius:var(--bs-border-radius-sm) !important}.rounded-bottom-2{border-bottom-right-radius:var(--bs-border-radius) !important;border-bottom-left-radius:var(--bs-border-radius) !important}.rounded-bottom-3{border-bottom-right-radius:var(--bs-border-radius-lg) !important;border-bottom-left-radius:var(--bs-border-radius-lg) !important}.rounded-bottom-4{border-bottom-right-radius:var(--bs-border-radius-xl) !important;border-bottom-left-radius:var(--bs-border-radius-xl) !important}.rounded-bottom-5{border-bottom-right-radius:var(--bs-border-radius-xxl) !important;border-bottom-left-radius:var(--bs-border-radius-xxl) !important}.rounded-bottom-circle{border-bottom-right-radius:50% !important;border-bottom-left-radius:50% !important}.rounded-bottom-pill{border-bottom-right-radius:var(--bs-border-radius-pill) !important;border-bottom-left-radius:var(--bs-border-radius-pill) !important}.rounded-start{border-bottom-left-radius:var(--bs-border-radius) !important;border-top-left-radius:var(--bs-border-radius) !important}.rounded-start-0{border-bottom-left-radius:0 !important;border-top-left-radius:0 !important}.rounded-start-1{border-bottom-left-radius:var(--bs-border-radius-sm) !important;border-top-left-radius:var(--bs-border-radius-sm) !important}.rounded-start-2{border-bottom-left-radius:var(--bs-border-radius) !important;border-top-left-radius:var(--bs-border-radius) !important}.rounded-start-3{border-bottom-left-radius:var(--bs-border-radius-lg) !important;border-top-left-radius:var(--bs-border-radius-lg) !important}.rounded-start-4{border-bottom-left-radius:var(--bs-border-radius-xl) !important;border-top-left-radius:var(--bs-border-radius-xl) !important}.rounded-start-5{border-bottom-left-radius:var(--bs-border-radius-xxl) !important;border-top-left-radius:var(--bs-border-radius-xxl) !important}.rounded-start-circle{border-bottom-left-radius:50% !important;border-top-left-radius:50% !important}.rounded-start-pill{border-bottom-left-radius:var(--bs-border-radius-pill) !important;border-top-left-radius:var(--bs-border-radius-pill) !important}.visible{visibility:visible !important}.invisible{visibility:hidden !important}.z-n1{z-index:-1 !important}.z-0{z-index:0 !important}.z-1{z-index:1 !important}.z-2{z-index:2 !important}.z-3{z-index:3 !important}@media(min-width: 576px){.float-sm-start{float:left !important}.float-sm-end{float:right !important}.float-sm-none{float:none !important}.object-fit-sm-contain{object-fit:contain !important}.object-fit-sm-cover{object-fit:cover !important}.object-fit-sm-fill{object-fit:fill !important}.object-fit-sm-scale{object-fit:scale-down !important}.object-fit-sm-none{object-fit:none !important}.d-sm-inline{display:inline !important}.d-sm-inline-block{display:inline-block !important}.d-sm-block{display:block !important}.d-sm-grid{display:grid !important}.d-sm-inline-grid{display:inline-grid !important}.d-sm-table{display:table !important}.d-sm-table-row{display:table-row !important}.d-sm-table-cell{display:table-cell !important}.d-sm-flex{display:flex !important}.d-sm-inline-flex{display:inline-flex !important}.d-sm-none{display:none !important}.flex-sm-fill{flex:1 1 auto !important}.flex-sm-row{flex-direction:row !important}.flex-sm-column{flex-direction:column !important}.flex-sm-row-reverse{flex-direction:row-reverse !important}.flex-sm-column-reverse{flex-direction:column-reverse !important}.flex-sm-grow-0{flex-grow:0 !important}.flex-sm-grow-1{flex-grow:1 !important}.flex-sm-shrink-0{flex-shrink:0 !important}.flex-sm-shrink-1{flex-shrink:1 !important}.flex-sm-wrap{flex-wrap:wrap !important}.flex-sm-nowrap{flex-wrap:nowrap !important}.flex-sm-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-sm-start{justify-content:flex-start !important}.justify-content-sm-end{justify-content:flex-end !important}.justify-content-sm-center{justify-content:center !important}.justify-content-sm-between{justify-content:space-between !important}.justify-content-sm-around{justify-content:space-around !important}.justify-content-sm-evenly{justify-content:space-evenly !important}.align-items-sm-start{align-items:flex-start !important}.align-items-sm-end{align-items:flex-end !important}.align-items-sm-center{align-items:center !important}.align-items-sm-baseline{align-items:baseline !important}.align-items-sm-stretch{align-items:stretch !important}.align-content-sm-start{align-content:flex-start !important}.align-content-sm-end{align-content:flex-end !important}.align-content-sm-center{align-content:center !important}.align-content-sm-between{align-content:space-between !important}.align-content-sm-around{align-content:space-around !important}.align-content-sm-stretch{align-content:stretch !important}.align-self-sm-auto{align-self:auto !important}.align-self-sm-start{align-self:flex-start !important}.align-self-sm-end{align-self:flex-end !important}.align-self-sm-center{align-self:center !important}.align-self-sm-baseline{align-self:baseline !important}.align-self-sm-stretch{align-self:stretch !important}.order-sm-first{order:-1 !important}.order-sm-0{order:0 !important}.order-sm-1{order:1 !important}.order-sm-2{order:2 !important}.order-sm-3{order:3 !important}.order-sm-4{order:4 !important}.order-sm-5{order:5 !important}.order-sm-last{order:6 !important}.m-sm-0{margin:0 !important}.m-sm-1{margin:.25rem !important}.m-sm-2{margin:.5rem !important}.m-sm-3{margin:1rem !important}.m-sm-4{margin:1.5rem !important}.m-sm-5{margin:3rem !important}.m-sm-auto{margin:auto !important}.mx-sm-0{margin-right:0 !important;margin-left:0 !important}.mx-sm-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-sm-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-sm-3{margin-right:1rem !important;margin-left:1rem !important}.mx-sm-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-sm-5{margin-right:3rem !important;margin-left:3rem !important}.mx-sm-auto{margin-right:auto !important;margin-left:auto !important}.my-sm-0{margin-top:0 !important;margin-bottom:0 !important}.my-sm-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-sm-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-sm-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-sm-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-sm-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-sm-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-sm-0{margin-top:0 !important}.mt-sm-1{margin-top:.25rem !important}.mt-sm-2{margin-top:.5rem !important}.mt-sm-3{margin-top:1rem !important}.mt-sm-4{margin-top:1.5rem !important}.mt-sm-5{margin-top:3rem !important}.mt-sm-auto{margin-top:auto !important}.me-sm-0{margin-right:0 !important}.me-sm-1{margin-right:.25rem !important}.me-sm-2{margin-right:.5rem !important}.me-sm-3{margin-right:1rem !important}.me-sm-4{margin-right:1.5rem !important}.me-sm-5{margin-right:3rem !important}.me-sm-auto{margin-right:auto !important}.mb-sm-0{margin-bottom:0 !important}.mb-sm-1{margin-bottom:.25rem !important}.mb-sm-2{margin-bottom:.5rem !important}.mb-sm-3{margin-bottom:1rem !important}.mb-sm-4{margin-bottom:1.5rem !important}.mb-sm-5{margin-bottom:3rem !important}.mb-sm-auto{margin-bottom:auto !important}.ms-sm-0{margin-left:0 !important}.ms-sm-1{margin-left:.25rem !important}.ms-sm-2{margin-left:.5rem !important}.ms-sm-3{margin-left:1rem !important}.ms-sm-4{margin-left:1.5rem !important}.ms-sm-5{margin-left:3rem !important}.ms-sm-auto{margin-left:auto !important}.p-sm-0{padding:0 !important}.p-sm-1{padding:.25rem !important}.p-sm-2{padding:.5rem !important}.p-sm-3{padding:1rem !important}.p-sm-4{padding:1.5rem !important}.p-sm-5{padding:3rem !important}.px-sm-0{padding-right:0 !important;padding-left:0 !important}.px-sm-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-sm-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-sm-3{padding-right:1rem !important;padding-left:1rem !important}.px-sm-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-sm-5{padding-right:3rem !important;padding-left:3rem !important}.py-sm-0{padding-top:0 !important;padding-bottom:0 !important}.py-sm-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-sm-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-sm-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-sm-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-sm-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-sm-0{padding-top:0 !important}.pt-sm-1{padding-top:.25rem !important}.pt-sm-2{padding-top:.5rem !important}.pt-sm-3{padding-top:1rem !important}.pt-sm-4{padding-top:1.5rem !important}.pt-sm-5{padding-top:3rem !important}.pe-sm-0{padding-right:0 !important}.pe-sm-1{padding-right:.25rem !important}.pe-sm-2{padding-right:.5rem !important}.pe-sm-3{padding-right:1rem !important}.pe-sm-4{padding-right:1.5rem !important}.pe-sm-5{padding-right:3rem !important}.pb-sm-0{padding-bottom:0 !important}.pb-sm-1{padding-bottom:.25rem !important}.pb-sm-2{padding-bottom:.5rem !important}.pb-sm-3{padding-bottom:1rem !important}.pb-sm-4{padding-bottom:1.5rem !important}.pb-sm-5{padding-bottom:3rem !important}.ps-sm-0{padding-left:0 !important}.ps-sm-1{padding-left:.25rem !important}.ps-sm-2{padding-left:.5rem !important}.ps-sm-3{padding-left:1rem !important}.ps-sm-4{padding-left:1.5rem !important}.ps-sm-5{padding-left:3rem !important}.gap-sm-0{gap:0 !important}.gap-sm-1{gap:.25rem !important}.gap-sm-2{gap:.5rem !important}.gap-sm-3{gap:1rem !important}.gap-sm-4{gap:1.5rem !important}.gap-sm-5{gap:3rem !important}.row-gap-sm-0{row-gap:0 !important}.row-gap-sm-1{row-gap:.25rem !important}.row-gap-sm-2{row-gap:.5rem !important}.row-gap-sm-3{row-gap:1rem !important}.row-gap-sm-4{row-gap:1.5rem !important}.row-gap-sm-5{row-gap:3rem !important}.column-gap-sm-0{column-gap:0 !important}.column-gap-sm-1{column-gap:.25rem !important}.column-gap-sm-2{column-gap:.5rem !important}.column-gap-sm-3{column-gap:1rem !important}.column-gap-sm-4{column-gap:1.5rem !important}.column-gap-sm-5{column-gap:3rem !important}.text-sm-start{text-align:left !important}.text-sm-end{text-align:right !important}.text-sm-center{text-align:center !important}}@media(min-width: 768px){.float-md-start{float:left !important}.float-md-end{float:right !important}.float-md-none{float:none !important}.object-fit-md-contain{object-fit:contain !important}.object-fit-md-cover{object-fit:cover !important}.object-fit-md-fill{object-fit:fill !important}.object-fit-md-scale{object-fit:scale-down !important}.object-fit-md-none{object-fit:none !important}.d-md-inline{display:inline !important}.d-md-inline-block{display:inline-block !important}.d-md-block{display:block !important}.d-md-grid{display:grid !important}.d-md-inline-grid{display:inline-grid !important}.d-md-table{display:table !important}.d-md-table-row{display:table-row !important}.d-md-table-cell{display:table-cell !important}.d-md-flex{display:flex !important}.d-md-inline-flex{display:inline-flex !important}.d-md-none{display:none !important}.flex-md-fill{flex:1 1 auto !important}.flex-md-row{flex-direction:row !important}.flex-md-column{flex-direction:column !important}.flex-md-row-reverse{flex-direction:row-reverse !important}.flex-md-column-reverse{flex-direction:column-reverse !important}.flex-md-grow-0{flex-grow:0 !important}.flex-md-grow-1{flex-grow:1 !important}.flex-md-shrink-0{flex-shrink:0 !important}.flex-md-shrink-1{flex-shrink:1 !important}.flex-md-wrap{flex-wrap:wrap !important}.flex-md-nowrap{flex-wrap:nowrap !important}.flex-md-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-md-start{justify-content:flex-start !important}.justify-content-md-end{justify-content:flex-end !important}.justify-content-md-center{justify-content:center !important}.justify-content-md-between{justify-content:space-between !important}.justify-content-md-around{justify-content:space-around !important}.justify-content-md-evenly{justify-content:space-evenly !important}.align-items-md-start{align-items:flex-start !important}.align-items-md-end{align-items:flex-end !important}.align-items-md-center{align-items:center !important}.align-items-md-baseline{align-items:baseline !important}.align-items-md-stretch{align-items:stretch !important}.align-content-md-start{align-content:flex-start !important}.align-content-md-end{align-content:flex-end !important}.align-content-md-center{align-content:center !important}.align-content-md-between{align-content:space-between !important}.align-content-md-around{align-content:space-around !important}.align-content-md-stretch{align-content:stretch !important}.align-self-md-auto{align-self:auto !important}.align-self-md-start{align-self:flex-start !important}.align-self-md-end{align-self:flex-end !important}.align-self-md-center{align-self:center !important}.align-self-md-baseline{align-self:baseline !important}.align-self-md-stretch{align-self:stretch !important}.order-md-first{order:-1 !important}.order-md-0{order:0 !important}.order-md-1{order:1 !important}.order-md-2{order:2 !important}.order-md-3{order:3 !important}.order-md-4{order:4 !important}.order-md-5{order:5 !important}.order-md-last{order:6 !important}.m-md-0{margin:0 !important}.m-md-1{margin:.25rem !important}.m-md-2{margin:.5rem !important}.m-md-3{margin:1rem !important}.m-md-4{margin:1.5rem !important}.m-md-5{margin:3rem !important}.m-md-auto{margin:auto !important}.mx-md-0{margin-right:0 !important;margin-left:0 !important}.mx-md-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-md-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-md-3{margin-right:1rem !important;margin-left:1rem !important}.mx-md-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-md-5{margin-right:3rem !important;margin-left:3rem !important}.mx-md-auto{margin-right:auto !important;margin-left:auto !important}.my-md-0{margin-top:0 !important;margin-bottom:0 !important}.my-md-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-md-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-md-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-md-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-md-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-md-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-md-0{margin-top:0 !important}.mt-md-1{margin-top:.25rem !important}.mt-md-2{margin-top:.5rem !important}.mt-md-3{margin-top:1rem !important}.mt-md-4{margin-top:1.5rem !important}.mt-md-5{margin-top:3rem !important}.mt-md-auto{margin-top:auto !important}.me-md-0{margin-right:0 !important}.me-md-1{margin-right:.25rem !important}.me-md-2{margin-right:.5rem !important}.me-md-3{margin-right:1rem !important}.me-md-4{margin-right:1.5rem !important}.me-md-5{margin-right:3rem !important}.me-md-auto{margin-right:auto !important}.mb-md-0{margin-bottom:0 !important}.mb-md-1{margin-bottom:.25rem !important}.mb-md-2{margin-bottom:.5rem !important}.mb-md-3{margin-bottom:1rem !important}.mb-md-4{margin-bottom:1.5rem !important}.mb-md-5{margin-bottom:3rem !important}.mb-md-auto{margin-bottom:auto !important}.ms-md-0{margin-left:0 !important}.ms-md-1{margin-left:.25rem !important}.ms-md-2{margin-left:.5rem !important}.ms-md-3{margin-left:1rem !important}.ms-md-4{margin-left:1.5rem !important}.ms-md-5{margin-left:3rem !important}.ms-md-auto{margin-left:auto !important}.p-md-0{padding:0 !important}.p-md-1{padding:.25rem !important}.p-md-2{padding:.5rem !important}.p-md-3{padding:1rem !important}.p-md-4{padding:1.5rem !important}.p-md-5{padding:3rem !important}.px-md-0{padding-right:0 !important;padding-left:0 !important}.px-md-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-md-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-md-3{padding-right:1rem !important;padding-left:1rem !important}.px-md-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-md-5{padding-right:3rem !important;padding-left:3rem !important}.py-md-0{padding-top:0 !important;padding-bottom:0 !important}.py-md-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-md-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-md-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-md-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-md-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-md-0{padding-top:0 !important}.pt-md-1{padding-top:.25rem !important}.pt-md-2{padding-top:.5rem !important}.pt-md-3{padding-top:1rem !important}.pt-md-4{padding-top:1.5rem !important}.pt-md-5{padding-top:3rem !important}.pe-md-0{padding-right:0 !important}.pe-md-1{padding-right:.25rem !important}.pe-md-2{padding-right:.5rem !important}.pe-md-3{padding-right:1rem !important}.pe-md-4{padding-right:1.5rem !important}.pe-md-5{padding-right:3rem !important}.pb-md-0{padding-bottom:0 !important}.pb-md-1{padding-bottom:.25rem !important}.pb-md-2{padding-bottom:.5rem !important}.pb-md-3{padding-bottom:1rem !important}.pb-md-4{padding-bottom:1.5rem !important}.pb-md-5{padding-bottom:3rem !important}.ps-md-0{padding-left:0 !important}.ps-md-1{padding-left:.25rem !important}.ps-md-2{padding-left:.5rem !important}.ps-md-3{padding-left:1rem !important}.ps-md-4{padding-left:1.5rem !important}.ps-md-5{padding-left:3rem !important}.gap-md-0{gap:0 !important}.gap-md-1{gap:.25rem !important}.gap-md-2{gap:.5rem !important}.gap-md-3{gap:1rem !important}.gap-md-4{gap:1.5rem !important}.gap-md-5{gap:3rem !important}.row-gap-md-0{row-gap:0 !important}.row-gap-md-1{row-gap:.25rem !important}.row-gap-md-2{row-gap:.5rem !important}.row-gap-md-3{row-gap:1rem !important}.row-gap-md-4{row-gap:1.5rem !important}.row-gap-md-5{row-gap:3rem !important}.column-gap-md-0{column-gap:0 !important}.column-gap-md-1{column-gap:.25rem !important}.column-gap-md-2{column-gap:.5rem !important}.column-gap-md-3{column-gap:1rem !important}.column-gap-md-4{column-gap:1.5rem !important}.column-gap-md-5{column-gap:3rem !important}.text-md-start{text-align:left !important}.text-md-end{text-align:right !important}.text-md-center{text-align:center !important}}@media(min-width: 992px){.float-lg-start{float:left !important}.float-lg-end{float:right !important}.float-lg-none{float:none !important}.object-fit-lg-contain{object-fit:contain !important}.object-fit-lg-cover{object-fit:cover !important}.object-fit-lg-fill{object-fit:fill !important}.object-fit-lg-scale{object-fit:scale-down !important}.object-fit-lg-none{object-fit:none !important}.d-lg-inline{display:inline !important}.d-lg-inline-block{display:inline-block !important}.d-lg-block{display:block !important}.d-lg-grid{display:grid !important}.d-lg-inline-grid{display:inline-grid !important}.d-lg-table{display:table !important}.d-lg-table-row{display:table-row !important}.d-lg-table-cell{display:table-cell !important}.d-lg-flex{display:flex !important}.d-lg-inline-flex{display:inline-flex !important}.d-lg-none{display:none !important}.flex-lg-fill{flex:1 1 auto !important}.flex-lg-row{flex-direction:row !important}.flex-lg-column{flex-direction:column !important}.flex-lg-row-reverse{flex-direction:row-reverse !important}.flex-lg-column-reverse{flex-direction:column-reverse !important}.flex-lg-grow-0{flex-grow:0 !important}.flex-lg-grow-1{flex-grow:1 !important}.flex-lg-shrink-0{flex-shrink:0 !important}.flex-lg-shrink-1{flex-shrink:1 !important}.flex-lg-wrap{flex-wrap:wrap !important}.flex-lg-nowrap{flex-wrap:nowrap !important}.flex-lg-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-lg-start{justify-content:flex-start !important}.justify-content-lg-end{justify-content:flex-end !important}.justify-content-lg-center{justify-content:center !important}.justify-content-lg-between{justify-content:space-between !important}.justify-content-lg-around{justify-content:space-around !important}.justify-content-lg-evenly{justify-content:space-evenly !important}.align-items-lg-start{align-items:flex-start !important}.align-items-lg-end{align-items:flex-end !important}.align-items-lg-center{align-items:center !important}.align-items-lg-baseline{align-items:baseline !important}.align-items-lg-stretch{align-items:stretch !important}.align-content-lg-start{align-content:flex-start !important}.align-content-lg-end{align-content:flex-end !important}.align-content-lg-center{align-content:center !important}.align-content-lg-between{align-content:space-between !important}.align-content-lg-around{align-content:space-around !important}.align-content-lg-stretch{align-content:stretch !important}.align-self-lg-auto{align-self:auto !important}.align-self-lg-start{align-self:flex-start !important}.align-self-lg-end{align-self:flex-end !important}.align-self-lg-center{align-self:center !important}.align-self-lg-baseline{align-self:baseline !important}.align-self-lg-stretch{align-self:stretch !important}.order-lg-first{order:-1 !important}.order-lg-0{order:0 !important}.order-lg-1{order:1 !important}.order-lg-2{order:2 !important}.order-lg-3{order:3 !important}.order-lg-4{order:4 !important}.order-lg-5{order:5 !important}.order-lg-last{order:6 !important}.m-lg-0{margin:0 !important}.m-lg-1{margin:.25rem !important}.m-lg-2{margin:.5rem !important}.m-lg-3{margin:1rem !important}.m-lg-4{margin:1.5rem !important}.m-lg-5{margin:3rem !important}.m-lg-auto{margin:auto !important}.mx-lg-0{margin-right:0 !important;margin-left:0 !important}.mx-lg-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-lg-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-lg-3{margin-right:1rem !important;margin-left:1rem !important}.mx-lg-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-lg-5{margin-right:3rem !important;margin-left:3rem !important}.mx-lg-auto{margin-right:auto !important;margin-left:auto !important}.my-lg-0{margin-top:0 !important;margin-bottom:0 !important}.my-lg-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-lg-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-lg-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-lg-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-lg-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-lg-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-lg-0{margin-top:0 !important}.mt-lg-1{margin-top:.25rem !important}.mt-lg-2{margin-top:.5rem !important}.mt-lg-3{margin-top:1rem !important}.mt-lg-4{margin-top:1.5rem !important}.mt-lg-5{margin-top:3rem !important}.mt-lg-auto{margin-top:auto !important}.me-lg-0{margin-right:0 !important}.me-lg-1{margin-right:.25rem !important}.me-lg-2{margin-right:.5rem !important}.me-lg-3{margin-right:1rem !important}.me-lg-4{margin-right:1.5rem !important}.me-lg-5{margin-right:3rem !important}.me-lg-auto{margin-right:auto !important}.mb-lg-0{margin-bottom:0 !important}.mb-lg-1{margin-bottom:.25rem !important}.mb-lg-2{margin-bottom:.5rem !important}.mb-lg-3{margin-bottom:1rem !important}.mb-lg-4{margin-bottom:1.5rem !important}.mb-lg-5{margin-bottom:3rem !important}.mb-lg-auto{margin-bottom:auto !important}.ms-lg-0{margin-left:0 !important}.ms-lg-1{margin-left:.25rem !important}.ms-lg-2{margin-left:.5rem !important}.ms-lg-3{margin-left:1rem !important}.ms-lg-4{margin-left:1.5rem !important}.ms-lg-5{margin-left:3rem !important}.ms-lg-auto{margin-left:auto !important}.p-lg-0{padding:0 !important}.p-lg-1{padding:.25rem !important}.p-lg-2{padding:.5rem !important}.p-lg-3{padding:1rem !important}.p-lg-4{padding:1.5rem !important}.p-lg-5{padding:3rem !important}.px-lg-0{padding-right:0 !important;padding-left:0 !important}.px-lg-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-lg-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-lg-3{padding-right:1rem !important;padding-left:1rem !important}.px-lg-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-lg-5{padding-right:3rem !important;padding-left:3rem !important}.py-lg-0{padding-top:0 !important;padding-bottom:0 !important}.py-lg-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-lg-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-lg-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-lg-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-lg-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-lg-0{padding-top:0 !important}.pt-lg-1{padding-top:.25rem !important}.pt-lg-2{padding-top:.5rem !important}.pt-lg-3{padding-top:1rem !important}.pt-lg-4{padding-top:1.5rem !important}.pt-lg-5{padding-top:3rem !important}.pe-lg-0{padding-right:0 !important}.pe-lg-1{padding-right:.25rem !important}.pe-lg-2{padding-right:.5rem !important}.pe-lg-3{padding-right:1rem !important}.pe-lg-4{padding-right:1.5rem !important}.pe-lg-5{padding-right:3rem !important}.pb-lg-0{padding-bottom:0 !important}.pb-lg-1{padding-bottom:.25rem !important}.pb-lg-2{padding-bottom:.5rem !important}.pb-lg-3{padding-bottom:1rem !important}.pb-lg-4{padding-bottom:1.5rem !important}.pb-lg-5{padding-bottom:3rem !important}.ps-lg-0{padding-left:0 !important}.ps-lg-1{padding-left:.25rem !important}.ps-lg-2{padding-left:.5rem !important}.ps-lg-3{padding-left:1rem !important}.ps-lg-4{padding-left:1.5rem !important}.ps-lg-5{padding-left:3rem !important}.gap-lg-0{gap:0 !important}.gap-lg-1{gap:.25rem !important}.gap-lg-2{gap:.5rem !important}.gap-lg-3{gap:1rem !important}.gap-lg-4{gap:1.5rem !important}.gap-lg-5{gap:3rem !important}.row-gap-lg-0{row-gap:0 !important}.row-gap-lg-1{row-gap:.25rem !important}.row-gap-lg-2{row-gap:.5rem !important}.row-gap-lg-3{row-gap:1rem !important}.row-gap-lg-4{row-gap:1.5rem !important}.row-gap-lg-5{row-gap:3rem !important}.column-gap-lg-0{column-gap:0 !important}.column-gap-lg-1{column-gap:.25rem !important}.column-gap-lg-2{column-gap:.5rem !important}.column-gap-lg-3{column-gap:1rem !important}.column-gap-lg-4{column-gap:1.5rem !important}.column-gap-lg-5{column-gap:3rem !important}.text-lg-start{text-align:left !important}.text-lg-end{text-align:right !important}.text-lg-center{text-align:center !important}}@media(min-width: 1200px){.float-xl-start{float:left !important}.float-xl-end{float:right !important}.float-xl-none{float:none !important}.object-fit-xl-contain{object-fit:contain !important}.object-fit-xl-cover{object-fit:cover !important}.object-fit-xl-fill{object-fit:fill !important}.object-fit-xl-scale{object-fit:scale-down !important}.object-fit-xl-none{object-fit:none !important}.d-xl-inline{display:inline !important}.d-xl-inline-block{display:inline-block !important}.d-xl-block{display:block !important}.d-xl-grid{display:grid !important}.d-xl-inline-grid{display:inline-grid !important}.d-xl-table{display:table !important}.d-xl-table-row{display:table-row !important}.d-xl-table-cell{display:table-cell !important}.d-xl-flex{display:flex !important}.d-xl-inline-flex{display:inline-flex !important}.d-xl-none{display:none !important}.flex-xl-fill{flex:1 1 auto !important}.flex-xl-row{flex-direction:row !important}.flex-xl-column{flex-direction:column !important}.flex-xl-row-reverse{flex-direction:row-reverse !important}.flex-xl-column-reverse{flex-direction:column-reverse !important}.flex-xl-grow-0{flex-grow:0 !important}.flex-xl-grow-1{flex-grow:1 !important}.flex-xl-shrink-0{flex-shrink:0 !important}.flex-xl-shrink-1{flex-shrink:1 !important}.flex-xl-wrap{flex-wrap:wrap !important}.flex-xl-nowrap{flex-wrap:nowrap !important}.flex-xl-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-xl-start{justify-content:flex-start !important}.justify-content-xl-end{justify-content:flex-end !important}.justify-content-xl-center{justify-content:center !important}.justify-content-xl-between{justify-content:space-between !important}.justify-content-xl-around{justify-content:space-around !important}.justify-content-xl-evenly{justify-content:space-evenly !important}.align-items-xl-start{align-items:flex-start !important}.align-items-xl-end{align-items:flex-end !important}.align-items-xl-center{align-items:center !important}.align-items-xl-baseline{align-items:baseline !important}.align-items-xl-stretch{align-items:stretch !important}.align-content-xl-start{align-content:flex-start !important}.align-content-xl-end{align-content:flex-end !important}.align-content-xl-center{align-content:center !important}.align-content-xl-between{align-content:space-between !important}.align-content-xl-around{align-content:space-around !important}.align-content-xl-stretch{align-content:stretch !important}.align-self-xl-auto{align-self:auto !important}.align-self-xl-start{align-self:flex-start !important}.align-self-xl-end{align-self:flex-end !important}.align-self-xl-center{align-self:center !important}.align-self-xl-baseline{align-self:baseline !important}.align-self-xl-stretch{align-self:stretch !important}.order-xl-first{order:-1 !important}.order-xl-0{order:0 !important}.order-xl-1{order:1 !important}.order-xl-2{order:2 !important}.order-xl-3{order:3 !important}.order-xl-4{order:4 !important}.order-xl-5{order:5 !important}.order-xl-last{order:6 !important}.m-xl-0{margin:0 !important}.m-xl-1{margin:.25rem !important}.m-xl-2{margin:.5rem !important}.m-xl-3{margin:1rem !important}.m-xl-4{margin:1.5rem !important}.m-xl-5{margin:3rem !important}.m-xl-auto{margin:auto !important}.mx-xl-0{margin-right:0 !important;margin-left:0 !important}.mx-xl-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-xl-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-xl-3{margin-right:1rem !important;margin-left:1rem !important}.mx-xl-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-xl-5{margin-right:3rem !important;margin-left:3rem !important}.mx-xl-auto{margin-right:auto !important;margin-left:auto !important}.my-xl-0{margin-top:0 !important;margin-bottom:0 !important}.my-xl-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-xl-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-xl-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-xl-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-xl-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-xl-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-xl-0{margin-top:0 !important}.mt-xl-1{margin-top:.25rem !important}.mt-xl-2{margin-top:.5rem !important}.mt-xl-3{margin-top:1rem !important}.mt-xl-4{margin-top:1.5rem !important}.mt-xl-5{margin-top:3rem !important}.mt-xl-auto{margin-top:auto !important}.me-xl-0{margin-right:0 !important}.me-xl-1{margin-right:.25rem !important}.me-xl-2{margin-right:.5rem !important}.me-xl-3{margin-right:1rem !important}.me-xl-4{margin-right:1.5rem !important}.me-xl-5{margin-right:3rem !important}.me-xl-auto{margin-right:auto !important}.mb-xl-0{margin-bottom:0 !important}.mb-xl-1{margin-bottom:.25rem !important}.mb-xl-2{margin-bottom:.5rem !important}.mb-xl-3{margin-bottom:1rem !important}.mb-xl-4{margin-bottom:1.5rem !important}.mb-xl-5{margin-bottom:3rem !important}.mb-xl-auto{margin-bottom:auto !important}.ms-xl-0{margin-left:0 !important}.ms-xl-1{margin-left:.25rem !important}.ms-xl-2{margin-left:.5rem !important}.ms-xl-3{margin-left:1rem !important}.ms-xl-4{margin-left:1.5rem !important}.ms-xl-5{margin-left:3rem !important}.ms-xl-auto{margin-left:auto !important}.p-xl-0{padding:0 !important}.p-xl-1{padding:.25rem !important}.p-xl-2{padding:.5rem !important}.p-xl-3{padding:1rem !important}.p-xl-4{padding:1.5rem !important}.p-xl-5{padding:3rem !important}.px-xl-0{padding-right:0 !important;padding-left:0 !important}.px-xl-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-xl-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-xl-3{padding-right:1rem !important;padding-left:1rem !important}.px-xl-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-xl-5{padding-right:3rem !important;padding-left:3rem !important}.py-xl-0{padding-top:0 !important;padding-bottom:0 !important}.py-xl-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-xl-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-xl-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-xl-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-xl-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-xl-0{padding-top:0 !important}.pt-xl-1{padding-top:.25rem !important}.pt-xl-2{padding-top:.5rem !important}.pt-xl-3{padding-top:1rem !important}.pt-xl-4{padding-top:1.5rem !important}.pt-xl-5{padding-top:3rem !important}.pe-xl-0{padding-right:0 !important}.pe-xl-1{padding-right:.25rem !important}.pe-xl-2{padding-right:.5rem !important}.pe-xl-3{padding-right:1rem !important}.pe-xl-4{padding-right:1.5rem !important}.pe-xl-5{padding-right:3rem !important}.pb-xl-0{padding-bottom:0 !important}.pb-xl-1{padding-bottom:.25rem !important}.pb-xl-2{padding-bottom:.5rem !important}.pb-xl-3{padding-bottom:1rem !important}.pb-xl-4{padding-bottom:1.5rem !important}.pb-xl-5{padding-bottom:3rem !important}.ps-xl-0{padding-left:0 !important}.ps-xl-1{padding-left:.25rem !important}.ps-xl-2{padding-left:.5rem !important}.ps-xl-3{padding-left:1rem !important}.ps-xl-4{padding-left:1.5rem !important}.ps-xl-5{padding-left:3rem !important}.gap-xl-0{gap:0 !important}.gap-xl-1{gap:.25rem !important}.gap-xl-2{gap:.5rem !important}.gap-xl-3{gap:1rem !important}.gap-xl-4{gap:1.5rem !important}.gap-xl-5{gap:3rem !important}.row-gap-xl-0{row-gap:0 !important}.row-gap-xl-1{row-gap:.25rem !important}.row-gap-xl-2{row-gap:.5rem !important}.row-gap-xl-3{row-gap:1rem !important}.row-gap-xl-4{row-gap:1.5rem !important}.row-gap-xl-5{row-gap:3rem !important}.column-gap-xl-0{column-gap:0 !important}.column-gap-xl-1{column-gap:.25rem !important}.column-gap-xl-2{column-gap:.5rem !important}.column-gap-xl-3{column-gap:1rem !important}.column-gap-xl-4{column-gap:1.5rem !important}.column-gap-xl-5{column-gap:3rem !important}.text-xl-start{text-align:left !important}.text-xl-end{text-align:right !important}.text-xl-center{text-align:center !important}}@media(min-width: 1400px){.float-xxl-start{float:left !important}.float-xxl-end{float:right !important}.float-xxl-none{float:none !important}.object-fit-xxl-contain{object-fit:contain !important}.object-fit-xxl-cover{object-fit:cover !important}.object-fit-xxl-fill{object-fit:fill !important}.object-fit-xxl-scale{object-fit:scale-down !important}.object-fit-xxl-none{object-fit:none !important}.d-xxl-inline{display:inline !important}.d-xxl-inline-block{display:inline-block !important}.d-xxl-block{display:block !important}.d-xxl-grid{display:grid !important}.d-xxl-inline-grid{display:inline-grid !important}.d-xxl-table{display:table !important}.d-xxl-table-row{display:table-row !important}.d-xxl-table-cell{display:table-cell !important}.d-xxl-flex{display:flex !important}.d-xxl-inline-flex{display:inline-flex !important}.d-xxl-none{display:none !important}.flex-xxl-fill{flex:1 1 auto !important}.flex-xxl-row{flex-direction:row !important}.flex-xxl-column{flex-direction:column !important}.flex-xxl-row-reverse{flex-direction:row-reverse !important}.flex-xxl-column-reverse{flex-direction:column-reverse !important}.flex-xxl-grow-0{flex-grow:0 !important}.flex-xxl-grow-1{flex-grow:1 !important}.flex-xxl-shrink-0{flex-shrink:0 !important}.flex-xxl-shrink-1{flex-shrink:1 !important}.flex-xxl-wrap{flex-wrap:wrap !important}.flex-xxl-nowrap{flex-wrap:nowrap !important}.flex-xxl-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-xxl-start{justify-content:flex-start !important}.justify-content-xxl-end{justify-content:flex-end !important}.justify-content-xxl-center{justify-content:center !important}.justify-content-xxl-between{justify-content:space-between !important}.justify-content-xxl-around{justify-content:space-around !important}.justify-content-xxl-evenly{justify-content:space-evenly !important}.align-items-xxl-start{align-items:flex-start !important}.align-items-xxl-end{align-items:flex-end !important}.align-items-xxl-center{align-items:center !important}.align-items-xxl-baseline{align-items:baseline !important}.align-items-xxl-stretch{align-items:stretch !important}.align-content-xxl-start{align-content:flex-start !important}.align-content-xxl-end{align-content:flex-end !important}.align-content-xxl-center{align-content:center !important}.align-content-xxl-between{align-content:space-between !important}.align-content-xxl-around{align-content:space-around !important}.align-content-xxl-stretch{align-content:stretch !important}.align-self-xxl-auto{align-self:auto !important}.align-self-xxl-start{align-self:flex-start !important}.align-self-xxl-end{align-self:flex-end !important}.align-self-xxl-center{align-self:center !important}.align-self-xxl-baseline{align-self:baseline !important}.align-self-xxl-stretch{align-self:stretch !important}.order-xxl-first{order:-1 !important}.order-xxl-0{order:0 !important}.order-xxl-1{order:1 !important}.order-xxl-2{order:2 !important}.order-xxl-3{order:3 !important}.order-xxl-4{order:4 !important}.order-xxl-5{order:5 !important}.order-xxl-last{order:6 !important}.m-xxl-0{margin:0 !important}.m-xxl-1{margin:.25rem !important}.m-xxl-2{margin:.5rem !important}.m-xxl-3{margin:1rem !important}.m-xxl-4{margin:1.5rem !important}.m-xxl-5{margin:3rem !important}.m-xxl-auto{margin:auto !important}.mx-xxl-0{margin-right:0 !important;margin-left:0 !important}.mx-xxl-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-xxl-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-xxl-3{margin-right:1rem !important;margin-left:1rem !important}.mx-xxl-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-xxl-5{margin-right:3rem !important;margin-left:3rem !important}.mx-xxl-auto{margin-right:auto !important;margin-left:auto !important}.my-xxl-0{margin-top:0 !important;margin-bottom:0 !important}.my-xxl-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-xxl-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-xxl-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-xxl-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-xxl-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-xxl-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-xxl-0{margin-top:0 !important}.mt-xxl-1{margin-top:.25rem !important}.mt-xxl-2{margin-top:.5rem !important}.mt-xxl-3{margin-top:1rem !important}.mt-xxl-4{margin-top:1.5rem !important}.mt-xxl-5{margin-top:3rem !important}.mt-xxl-auto{margin-top:auto !important}.me-xxl-0{margin-right:0 !important}.me-xxl-1{margin-right:.25rem !important}.me-xxl-2{margin-right:.5rem !important}.me-xxl-3{margin-right:1rem !important}.me-xxl-4{margin-right:1.5rem !important}.me-xxl-5{margin-right:3rem !important}.me-xxl-auto{margin-right:auto !important}.mb-xxl-0{margin-bottom:0 !important}.mb-xxl-1{margin-bottom:.25rem !important}.mb-xxl-2{margin-bottom:.5rem !important}.mb-xxl-3{margin-bottom:1rem !important}.mb-xxl-4{margin-bottom:1.5rem !important}.mb-xxl-5{margin-bottom:3rem !important}.mb-xxl-auto{margin-bottom:auto !important}.ms-xxl-0{margin-left:0 !important}.ms-xxl-1{margin-left:.25rem !important}.ms-xxl-2{margin-left:.5rem !important}.ms-xxl-3{margin-left:1rem !important}.ms-xxl-4{margin-left:1.5rem !important}.ms-xxl-5{margin-left:3rem !important}.ms-xxl-auto{margin-left:auto !important}.p-xxl-0{padding:0 !important}.p-xxl-1{padding:.25rem !important}.p-xxl-2{padding:.5rem !important}.p-xxl-3{padding:1rem !important}.p-xxl-4{padding:1.5rem !important}.p-xxl-5{padding:3rem !important}.px-xxl-0{padding-right:0 !important;padding-left:0 !important}.px-xxl-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-xxl-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-xxl-3{padding-right:1rem !important;padding-left:1rem !important}.px-xxl-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-xxl-5{padding-right:3rem !important;padding-left:3rem !important}.py-xxl-0{padding-top:0 !important;padding-bottom:0 !important}.py-xxl-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-xxl-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-xxl-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-xxl-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-xxl-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-xxl-0{padding-top:0 !important}.pt-xxl-1{padding-top:.25rem !important}.pt-xxl-2{padding-top:.5rem !important}.pt-xxl-3{padding-top:1rem !important}.pt-xxl-4{padding-top:1.5rem !important}.pt-xxl-5{padding-top:3rem !important}.pe-xxl-0{padding-right:0 !important}.pe-xxl-1{padding-right:.25rem !important}.pe-xxl-2{padding-right:.5rem !important}.pe-xxl-3{padding-right:1rem !important}.pe-xxl-4{padding-right:1.5rem !important}.pe-xxl-5{padding-right:3rem !important}.pb-xxl-0{padding-bottom:0 !important}.pb-xxl-1{padding-bottom:.25rem !important}.pb-xxl-2{padding-bottom:.5rem !important}.pb-xxl-3{padding-bottom:1rem !important}.pb-xxl-4{padding-bottom:1.5rem !important}.pb-xxl-5{padding-bottom:3rem !important}.ps-xxl-0{padding-left:0 !important}.ps-xxl-1{padding-left:.25rem !important}.ps-xxl-2{padding-left:.5rem !important}.ps-xxl-3{padding-left:1rem !important}.ps-xxl-4{padding-left:1.5rem !important}.ps-xxl-5{padding-left:3rem !important}.gap-xxl-0{gap:0 !important}.gap-xxl-1{gap:.25rem !important}.gap-xxl-2{gap:.5rem !important}.gap-xxl-3{gap:1rem !important}.gap-xxl-4{gap:1.5rem !important}.gap-xxl-5{gap:3rem !important}.row-gap-xxl-0{row-gap:0 !important}.row-gap-xxl-1{row-gap:.25rem !important}.row-gap-xxl-2{row-gap:.5rem !important}.row-gap-xxl-3{row-gap:1rem !important}.row-gap-xxl-4{row-gap:1.5rem !important}.row-gap-xxl-5{row-gap:3rem !important}.column-gap-xxl-0{column-gap:0 !important}.column-gap-xxl-1{column-gap:.25rem !important}.column-gap-xxl-2{column-gap:.5rem !important}.column-gap-xxl-3{column-gap:1rem !important}.column-gap-xxl-4{column-gap:1.5rem !important}.column-gap-xxl-5{column-gap:3rem !important}.text-xxl-start{text-align:left !important}.text-xxl-end{text-align:right !important}.text-xxl-center{text-align:center !important}}.bg-default{color:#fff}.bg-primary{color:#fff}.bg-secondary{color:#fff}.bg-success{color:#fff}.bg-info{color:#fff}.bg-warning{color:#fff}.bg-danger{color:#fff}.bg-light{color:#000}.bg-dark{color:#fff}@media(min-width: 1200px){.fs-1{font-size:2rem !important}.fs-2{font-size:1.65rem !important}.fs-3{font-size:1.45rem !important}}@media print{.d-print-inline{display:inline !important}.d-print-inline-block{display:inline-block !important}.d-print-block{display:block !important}.d-print-grid{display:grid !important}.d-print-inline-grid{display:inline-grid !important}.d-print-table{display:table !important}.d-print-table-row{display:table-row !important}.d-print-table-cell{display:table-cell !important}.d-print-flex{display:flex !important}.d-print-inline-flex{display:inline-flex !important}.d-print-none{display:none !important}}:root{--bslib-spacer: 1rem;--bslib-mb-spacer: var(--bslib-spacer, 1rem)}.bslib-mb-spacing{margin-bottom:var(--bslib-mb-spacer)}.bslib-gap-spacing{gap:var(--bslib-mb-spacer)}.bslib-gap-spacing>.bslib-mb-spacing,.bslib-gap-spacing>.form-group,.bslib-gap-spacing>p,.bslib-gap-spacing>pre{margin-bottom:0}.html-fill-container>.html-fill-item.bslib-mb-spacing{margin-bottom:0}.tab-content>.tab-pane.html-fill-container{display:none}.tab-content>.active.html-fill-container{display:flex}.tab-content.html-fill-container{padding:0}.bg-blue{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-blue{--bslib-color-fg: #2780e3;color:var(--bslib-color-fg)}.bg-indigo{--bslib-color-bg: #6610f2;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-indigo{--bslib-color-fg: #6610f2;color:var(--bslib-color-fg)}.bg-purple{--bslib-color-bg: #613d7c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-purple{--bslib-color-fg: #613d7c;color:var(--bslib-color-fg)}.bg-pink{--bslib-color-bg: #e83e8c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-pink{--bslib-color-fg: #e83e8c;color:var(--bslib-color-fg)}.bg-red{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-red{--bslib-color-fg: #ff0039;color:var(--bslib-color-fg)}.bg-orange{--bslib-color-bg: #f0ad4e;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-orange{--bslib-color-fg: #f0ad4e;color:var(--bslib-color-fg)}.bg-yellow{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-yellow{--bslib-color-fg: #ff7518;color:var(--bslib-color-fg)}.bg-green{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-green{--bslib-color-fg: #3fb618;color:var(--bslib-color-fg)}.bg-teal{--bslib-color-bg: #20c997;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-teal{--bslib-color-fg: #20c997;color:var(--bslib-color-fg)}.bg-cyan{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-cyan{--bslib-color-fg: #9954bb;color:var(--bslib-color-fg)}.text-default{--bslib-color-fg: #343a40}.bg-default{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-primary{--bslib-color-fg: #2780e3}.bg-primary{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff}.text-secondary{--bslib-color-fg: #343a40}.bg-secondary{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-success{--bslib-color-fg: #3fb618}.bg-success{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff}.text-info{--bslib-color-fg: #9954bb}.bg-info{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff}.text-warning{--bslib-color-fg: #ff7518}.bg-warning{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff}.text-danger{--bslib-color-fg: #ff0039}.bg-danger{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff}.text-light{--bslib-color-fg: #f8f9fa}.bg-light{--bslib-color-bg: #f8f9fa;--bslib-color-fg: #000}.text-dark{--bslib-color-fg: #343a40}.bg-dark{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.bg-gradient-blue-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4053e9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4053e9;color:#fff}.bg-gradient-blue-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3e65ba;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3e65ba;color:#fff}.bg-gradient-blue-pink{--bslib-color-fg: #fff;--bslib-color-bg: #7466c0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #7466c0;color:#fff}.bg-gradient-blue-red{--bslib-color-fg: #fff;--bslib-color-bg: #7d4d9f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #7d4d9f;color:#fff}.bg-gradient-blue-orange{--bslib-color-fg: #fff;--bslib-color-bg: #7792a7;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #7792a7;color:#fff}.bg-gradient-blue-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #7d7c92;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #7d7c92;color:#fff}.bg-gradient-blue-green{--bslib-color-fg: #fff;--bslib-color-bg: #319692;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #319692;color:#fff}.bg-gradient-blue-teal{--bslib-color-fg: #fff;--bslib-color-bg: #249dc5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #249dc5;color:#fff}.bg-gradient-blue-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #556ed3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #556ed3;color:#fff}.bg-gradient-indigo-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4d3dec;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4d3dec;color:#fff}.bg-gradient-indigo-purple{--bslib-color-fg: #fff;--bslib-color-bg: #6422c3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #6422c3;color:#fff}.bg-gradient-indigo-pink{--bslib-color-fg: #fff;--bslib-color-bg: #9a22c9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #9a22c9;color:#fff}.bg-gradient-indigo-red{--bslib-color-fg: #fff;--bslib-color-bg: #a30aa8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a30aa8;color:#fff}.bg-gradient-indigo-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9d4fb0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9d4fb0;color:#fff}.bg-gradient-indigo-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a3389b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a3389b;color:#fff}.bg-gradient-indigo-green{--bslib-color-fg: #fff;--bslib-color-bg: #56529b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #56529b;color:#fff}.bg-gradient-indigo-teal{--bslib-color-fg: #fff;--bslib-color-bg: #4a5ace;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #4a5ace;color:#fff}.bg-gradient-indigo-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #7a2bdc;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #7a2bdc;color:#fff}.bg-gradient-purple-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4a58a5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4a58a5;color:#fff}.bg-gradient-purple-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #632bab;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #632bab;color:#fff}.bg-gradient-purple-pink{--bslib-color-fg: #fff;--bslib-color-bg: #973d82;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #973d82;color:#fff}.bg-gradient-purple-red{--bslib-color-fg: #fff;--bslib-color-bg: #a02561;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a02561;color:#fff}.bg-gradient-purple-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9a6a6a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9a6a6a;color:#fff}.bg-gradient-purple-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a05354;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a05354;color:#fff}.bg-gradient-purple-green{--bslib-color-fg: #fff;--bslib-color-bg: #536d54;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #536d54;color:#fff}.bg-gradient-purple-teal{--bslib-color-fg: #fff;--bslib-color-bg: #477587;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #477587;color:#fff}.bg-gradient-purple-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #774695;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #774695;color:#fff}.bg-gradient-pink-blue{--bslib-color-fg: #fff;--bslib-color-bg: #9b58af;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #9b58af;color:#fff}.bg-gradient-pink-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b42cb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b42cb5;color:#fff}.bg-gradient-pink-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b23e86;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b23e86;color:#fff}.bg-gradient-pink-red{--bslib-color-fg: #fff;--bslib-color-bg: #f1256b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f1256b;color:#fff}.bg-gradient-pink-orange{--bslib-color-fg: #fff;--bslib-color-bg: #eb6a73;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #eb6a73;color:#fff}.bg-gradient-pink-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #f1545e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f1545e;color:#fff}.bg-gradient-pink-green{--bslib-color-fg: #fff;--bslib-color-bg: #a46e5e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a46e5e;color:#fff}.bg-gradient-pink-teal{--bslib-color-fg: #fff;--bslib-color-bg: #987690;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #987690;color:#fff}.bg-gradient-pink-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #c8479f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #c8479f;color:#fff}.bg-gradient-red-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a9337d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a9337d;color:#fff}.bg-gradient-red-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c20683;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c20683;color:#fff}.bg-gradient-red-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c01854;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c01854;color:#fff}.bg-gradient-red-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f6195a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f6195a;color:#fff}.bg-gradient-red-orange{--bslib-color-fg: #fff;--bslib-color-bg: #f94541;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f94541;color:#fff}.bg-gradient-red-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #ff2f2c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #ff2f2c;color:#fff}.bg-gradient-red-green{--bslib-color-fg: #fff;--bslib-color-bg: #b2492c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b2492c;color:#fff}.bg-gradient-red-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6505f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6505f;color:#fff}.bg-gradient-red-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d6226d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d6226d;color:#fff}.bg-gradient-orange-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a09b8a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a09b8a;color:#fff}.bg-gradient-orange-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b96e90;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b96e90;color:#fff}.bg-gradient-orange-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b78060;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b78060;color:#fff}.bg-gradient-orange-pink{--bslib-color-fg: #fff;--bslib-color-bg: #ed8167;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #ed8167;color:#fff}.bg-gradient-orange-red{--bslib-color-fg: #fff;--bslib-color-bg: #f66846;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f66846;color:#fff}.bg-gradient-orange-yellow{--bslib-color-fg: #000;--bslib-color-bg: #f69738;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f69738;color:#000}.bg-gradient-orange-green{--bslib-color-fg: #000;--bslib-color-bg: #a9b138;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a9b138;color:#000}.bg-gradient-orange-teal{--bslib-color-fg: #000;--bslib-color-bg: #9db86b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #9db86b;color:#000}.bg-gradient-orange-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #cd897a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #cd897a;color:#fff}.bg-gradient-yellow-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a97969;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a97969;color:#fff}.bg-gradient-yellow-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c24d6f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c24d6f;color:#fff}.bg-gradient-yellow-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c05f40;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c05f40;color:#fff}.bg-gradient-yellow-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f65f46;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f65f46;color:#fff}.bg-gradient-yellow-red{--bslib-color-fg: #fff;--bslib-color-bg: #ff4625;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #ff4625;color:#fff}.bg-gradient-yellow-orange{--bslib-color-fg: #000;--bslib-color-bg: #f98b2e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f98b2e;color:#000}.bg-gradient-yellow-green{--bslib-color-fg: #fff;--bslib-color-bg: #b28f18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b28f18;color:#fff}.bg-gradient-yellow-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6974b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6974b;color:#fff}.bg-gradient-yellow-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d66859;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d66859;color:#fff}.bg-gradient-green-blue{--bslib-color-fg: #fff;--bslib-color-bg: #35a069;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #35a069;color:#fff}.bg-gradient-green-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4f746f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4f746f;color:#fff}.bg-gradient-green-purple{--bslib-color-fg: #fff;--bslib-color-bg: #4d8640;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #4d8640;color:#fff}.bg-gradient-green-pink{--bslib-color-fg: #fff;--bslib-color-bg: #838646;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #838646;color:#fff}.bg-gradient-green-red{--bslib-color-fg: #fff;--bslib-color-bg: #8c6d25;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #8c6d25;color:#fff}.bg-gradient-green-orange{--bslib-color-fg: #000;--bslib-color-bg: #86b22e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #86b22e;color:#000}.bg-gradient-green-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #8c9c18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #8c9c18;color:#fff}.bg-gradient-green-teal{--bslib-color-fg: #000;--bslib-color-bg: #33be4b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #33be4b;color:#000}.bg-gradient-green-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #638f59;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #638f59;color:#fff}.bg-gradient-teal-blue{--bslib-color-fg: #fff;--bslib-color-bg: #23acb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #23acb5;color:#fff}.bg-gradient-teal-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #3c7fbb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #3c7fbb;color:#fff}.bg-gradient-teal-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3a918c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3a918c;color:#fff}.bg-gradient-teal-pink{--bslib-color-fg: #fff;--bslib-color-bg: #709193;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #709193;color:#fff}.bg-gradient-teal-red{--bslib-color-fg: #fff;--bslib-color-bg: #797971;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #797971;color:#fff}.bg-gradient-teal-orange{--bslib-color-fg: #000;--bslib-color-bg: #73be7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #73be7a;color:#000}.bg-gradient-teal-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #79a764;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #79a764;color:#fff}.bg-gradient-teal-green{--bslib-color-fg: #000;--bslib-color-bg: #2cc164;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #2cc164;color:#000}.bg-gradient-teal-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #509aa5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #509aa5;color:#fff}.bg-gradient-cyan-blue{--bslib-color-fg: #fff;--bslib-color-bg: #6b66cb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #6b66cb;color:#fff}.bg-gradient-cyan-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #8539d1;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #8539d1;color:#fff}.bg-gradient-cyan-purple{--bslib-color-fg: #fff;--bslib-color-bg: #834ba2;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #834ba2;color:#fff}.bg-gradient-cyan-pink{--bslib-color-fg: #fff;--bslib-color-bg: #b94ba8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #b94ba8;color:#fff}.bg-gradient-cyan-red{--bslib-color-fg: #fff;--bslib-color-bg: #c23287;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #c23287;color:#fff}.bg-gradient-cyan-orange{--bslib-color-fg: #fff;--bslib-color-bg: #bc788f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #bc788f;color:#fff}.bg-gradient-cyan-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #c2617a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #c2617a;color:#fff}.bg-gradient-cyan-green{--bslib-color-fg: #fff;--bslib-color-bg: #757b7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #757b7a;color:#fff}.bg-gradient-cyan-teal{--bslib-color-fg: #fff;--bslib-color-bg: #6983ad;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #6983ad;color:#fff}.bg-blue{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-blue{--bslib-color-fg: #2780e3;color:var(--bslib-color-fg)}.bg-indigo{--bslib-color-bg: #6610f2;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-indigo{--bslib-color-fg: #6610f2;color:var(--bslib-color-fg)}.bg-purple{--bslib-color-bg: #613d7c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-purple{--bslib-color-fg: #613d7c;color:var(--bslib-color-fg)}.bg-pink{--bslib-color-bg: #e83e8c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-pink{--bslib-color-fg: #e83e8c;color:var(--bslib-color-fg)}.bg-red{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-red{--bslib-color-fg: #ff0039;color:var(--bslib-color-fg)}.bg-orange{--bslib-color-bg: #f0ad4e;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-orange{--bslib-color-fg: #f0ad4e;color:var(--bslib-color-fg)}.bg-yellow{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-yellow{--bslib-color-fg: #ff7518;color:var(--bslib-color-fg)}.bg-green{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-green{--bslib-color-fg: #3fb618;color:var(--bslib-color-fg)}.bg-teal{--bslib-color-bg: #20c997;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-teal{--bslib-color-fg: #20c997;color:var(--bslib-color-fg)}.bg-cyan{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-cyan{--bslib-color-fg: #9954bb;color:var(--bslib-color-fg)}.text-default{--bslib-color-fg: #343a40}.bg-default{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-primary{--bslib-color-fg: #2780e3}.bg-primary{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff}.text-secondary{--bslib-color-fg: #343a40}.bg-secondary{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-success{--bslib-color-fg: #3fb618}.bg-success{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff}.text-info{--bslib-color-fg: #9954bb}.bg-info{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff}.text-warning{--bslib-color-fg: #ff7518}.bg-warning{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff}.text-danger{--bslib-color-fg: #ff0039}.bg-danger{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff}.text-light{--bslib-color-fg: #f8f9fa}.bg-light{--bslib-color-bg: #f8f9fa;--bslib-color-fg: #000}.text-dark{--bslib-color-fg: #343a40}.bg-dark{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.bg-gradient-blue-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4053e9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4053e9;color:#fff}.bg-gradient-blue-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3e65ba;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3e65ba;color:#fff}.bg-gradient-blue-pink{--bslib-color-fg: #fff;--bslib-color-bg: #7466c0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #7466c0;color:#fff}.bg-gradient-blue-red{--bslib-color-fg: #fff;--bslib-color-bg: #7d4d9f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #7d4d9f;color:#fff}.bg-gradient-blue-orange{--bslib-color-fg: #fff;--bslib-color-bg: #7792a7;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #7792a7;color:#fff}.bg-gradient-blue-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #7d7c92;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #7d7c92;color:#fff}.bg-gradient-blue-green{--bslib-color-fg: #fff;--bslib-color-bg: #319692;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #319692;color:#fff}.bg-gradient-blue-teal{--bslib-color-fg: #fff;--bslib-color-bg: #249dc5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #249dc5;color:#fff}.bg-gradient-blue-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #556ed3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #556ed3;color:#fff}.bg-gradient-indigo-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4d3dec;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4d3dec;color:#fff}.bg-gradient-indigo-purple{--bslib-color-fg: #fff;--bslib-color-bg: #6422c3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #6422c3;color:#fff}.bg-gradient-indigo-pink{--bslib-color-fg: #fff;--bslib-color-bg: #9a22c9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #9a22c9;color:#fff}.bg-gradient-indigo-red{--bslib-color-fg: #fff;--bslib-color-bg: #a30aa8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a30aa8;color:#fff}.bg-gradient-indigo-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9d4fb0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9d4fb0;color:#fff}.bg-gradient-indigo-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a3389b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a3389b;color:#fff}.bg-gradient-indigo-green{--bslib-color-fg: #fff;--bslib-color-bg: #56529b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #56529b;color:#fff}.bg-gradient-indigo-teal{--bslib-color-fg: #fff;--bslib-color-bg: #4a5ace;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #4a5ace;color:#fff}.bg-gradient-indigo-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #7a2bdc;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #7a2bdc;color:#fff}.bg-gradient-purple-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4a58a5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4a58a5;color:#fff}.bg-gradient-purple-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #632bab;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #632bab;color:#fff}.bg-gradient-purple-pink{--bslib-color-fg: #fff;--bslib-color-bg: #973d82;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #973d82;color:#fff}.bg-gradient-purple-red{--bslib-color-fg: #fff;--bslib-color-bg: #a02561;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a02561;color:#fff}.bg-gradient-purple-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9a6a6a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9a6a6a;color:#fff}.bg-gradient-purple-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a05354;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a05354;color:#fff}.bg-gradient-purple-green{--bslib-color-fg: #fff;--bslib-color-bg: #536d54;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #536d54;color:#fff}.bg-gradient-purple-teal{--bslib-color-fg: #fff;--bslib-color-bg: #477587;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #477587;color:#fff}.bg-gradient-purple-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #774695;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #774695;color:#fff}.bg-gradient-pink-blue{--bslib-color-fg: #fff;--bslib-color-bg: #9b58af;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #9b58af;color:#fff}.bg-gradient-pink-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b42cb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b42cb5;color:#fff}.bg-gradient-pink-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b23e86;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b23e86;color:#fff}.bg-gradient-pink-red{--bslib-color-fg: #fff;--bslib-color-bg: #f1256b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f1256b;color:#fff}.bg-gradient-pink-orange{--bslib-color-fg: #fff;--bslib-color-bg: #eb6a73;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #eb6a73;color:#fff}.bg-gradient-pink-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #f1545e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f1545e;color:#fff}.bg-gradient-pink-green{--bslib-color-fg: #fff;--bslib-color-bg: #a46e5e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a46e5e;color:#fff}.bg-gradient-pink-teal{--bslib-color-fg: #fff;--bslib-color-bg: #987690;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #987690;color:#fff}.bg-gradient-pink-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #c8479f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #c8479f;color:#fff}.bg-gradient-red-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a9337d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a9337d;color:#fff}.bg-gradient-red-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c20683;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c20683;color:#fff}.bg-gradient-red-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c01854;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c01854;color:#fff}.bg-gradient-red-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f6195a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f6195a;color:#fff}.bg-gradient-red-orange{--bslib-color-fg: #fff;--bslib-color-bg: #f94541;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f94541;color:#fff}.bg-gradient-red-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #ff2f2c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #ff2f2c;color:#fff}.bg-gradient-red-green{--bslib-color-fg: #fff;--bslib-color-bg: #b2492c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b2492c;color:#fff}.bg-gradient-red-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6505f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6505f;color:#fff}.bg-gradient-red-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d6226d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d6226d;color:#fff}.bg-gradient-orange-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a09b8a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a09b8a;color:#fff}.bg-gradient-orange-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b96e90;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b96e90;color:#fff}.bg-gradient-orange-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b78060;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b78060;color:#fff}.bg-gradient-orange-pink{--bslib-color-fg: #fff;--bslib-color-bg: #ed8167;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #ed8167;color:#fff}.bg-gradient-orange-red{--bslib-color-fg: #fff;--bslib-color-bg: #f66846;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f66846;color:#fff}.bg-gradient-orange-yellow{--bslib-color-fg: #000;--bslib-color-bg: #f69738;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f69738;color:#000}.bg-gradient-orange-green{--bslib-color-fg: #000;--bslib-color-bg: #a9b138;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a9b138;color:#000}.bg-gradient-orange-teal{--bslib-color-fg: #000;--bslib-color-bg: #9db86b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #9db86b;color:#000}.bg-gradient-orange-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #cd897a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #cd897a;color:#fff}.bg-gradient-yellow-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a97969;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a97969;color:#fff}.bg-gradient-yellow-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c24d6f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c24d6f;color:#fff}.bg-gradient-yellow-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c05f40;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c05f40;color:#fff}.bg-gradient-yellow-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f65f46;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f65f46;color:#fff}.bg-gradient-yellow-red{--bslib-color-fg: #fff;--bslib-color-bg: #ff4625;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #ff4625;color:#fff}.bg-gradient-yellow-orange{--bslib-color-fg: #000;--bslib-color-bg: #f98b2e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f98b2e;color:#000}.bg-gradient-yellow-green{--bslib-color-fg: #fff;--bslib-color-bg: #b28f18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b28f18;color:#fff}.bg-gradient-yellow-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6974b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6974b;color:#fff}.bg-gradient-yellow-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d66859;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d66859;color:#fff}.bg-gradient-green-blue{--bslib-color-fg: #fff;--bslib-color-bg: #35a069;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #35a069;color:#fff}.bg-gradient-green-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4f746f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4f746f;color:#fff}.bg-gradient-green-purple{--bslib-color-fg: #fff;--bslib-color-bg: #4d8640;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #4d8640;color:#fff}.bg-gradient-green-pink{--bslib-color-fg: #fff;--bslib-color-bg: #838646;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #838646;color:#fff}.bg-gradient-green-red{--bslib-color-fg: #fff;--bslib-color-bg: #8c6d25;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #8c6d25;color:#fff}.bg-gradient-green-orange{--bslib-color-fg: #000;--bslib-color-bg: #86b22e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #86b22e;color:#000}.bg-gradient-green-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #8c9c18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #8c9c18;color:#fff}.bg-gradient-green-teal{--bslib-color-fg: #000;--bslib-color-bg: #33be4b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #33be4b;color:#000}.bg-gradient-green-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #638f59;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #638f59;color:#fff}.bg-gradient-teal-blue{--bslib-color-fg: #fff;--bslib-color-bg: #23acb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #23acb5;color:#fff}.bg-gradient-teal-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #3c7fbb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #3c7fbb;color:#fff}.bg-gradient-teal-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3a918c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3a918c;color:#fff}.bg-gradient-teal-pink{--bslib-color-fg: #fff;--bslib-color-bg: #709193;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #709193;color:#fff}.bg-gradient-teal-red{--bslib-color-fg: #fff;--bslib-color-bg: #797971;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #797971;color:#fff}.bg-gradient-teal-orange{--bslib-color-fg: #000;--bslib-color-bg: #73be7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #73be7a;color:#000}.bg-gradient-teal-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #79a764;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #79a764;color:#fff}.bg-gradient-teal-green{--bslib-color-fg: #000;--bslib-color-bg: #2cc164;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #2cc164;color:#000}.bg-gradient-teal-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #509aa5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #509aa5;color:#fff}.bg-gradient-cyan-blue{--bslib-color-fg: #fff;--bslib-color-bg: #6b66cb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #6b66cb;color:#fff}.bg-gradient-cyan-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #8539d1;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #8539d1;color:#fff}.bg-gradient-cyan-purple{--bslib-color-fg: #fff;--bslib-color-bg: #834ba2;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #834ba2;color:#fff}.bg-gradient-cyan-pink{--bslib-color-fg: #fff;--bslib-color-bg: #b94ba8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #b94ba8;color:#fff}.bg-gradient-cyan-red{--bslib-color-fg: #fff;--bslib-color-bg: #c23287;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #c23287;color:#fff}.bg-gradient-cyan-orange{--bslib-color-fg: #fff;--bslib-color-bg: #bc788f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #bc788f;color:#fff}.bg-gradient-cyan-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #c2617a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #c2617a;color:#fff}.bg-gradient-cyan-green{--bslib-color-fg: #fff;--bslib-color-bg: #757b7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #757b7a;color:#fff}.bg-gradient-cyan-teal{--bslib-color-fg: #fff;--bslib-color-bg: #6983ad;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #6983ad;color:#fff}.tab-content>.tab-pane.html-fill-container{display:none}.tab-content>.active.html-fill-container{display:flex}.tab-content.html-fill-container{padding:0}:root{--bslib-spacer: 1rem;--bslib-mb-spacer: var(--bslib-spacer, 1rem)}.bslib-mb-spacing{margin-bottom:var(--bslib-mb-spacer)}.bslib-gap-spacing{gap:var(--bslib-mb-spacer)}.bslib-gap-spacing>.bslib-mb-spacing,.bslib-gap-spacing>.form-group,.bslib-gap-spacing>p,.bslib-gap-spacing>pre{margin-bottom:0}.html-fill-container>.html-fill-item.bslib-mb-spacing{margin-bottom:0}.bslib-grid{display:grid !important;gap:var(--bslib-spacer, 1rem);height:var(--bslib-grid-height)}.bslib-grid.grid{grid-template-columns:repeat(var(--bs-columns, 12), minmax(0, 1fr));grid-template-rows:unset;grid-auto-rows:var(--bslib-grid--row-heights);--bslib-grid--row-heights--xs: unset;--bslib-grid--row-heights--sm: unset;--bslib-grid--row-heights--md: unset;--bslib-grid--row-heights--lg: unset;--bslib-grid--row-heights--xl: unset;--bslib-grid--row-heights--xxl: unset}.bslib-grid.grid.bslib-grid--row-heights--xs{--bslib-grid--row-heights: var(--bslib-grid--row-heights--xs)}@media(min-width: 576px){.bslib-grid.grid.bslib-grid--row-heights--sm{--bslib-grid--row-heights: var(--bslib-grid--row-heights--sm)}}@media(min-width: 768px){.bslib-grid.grid.bslib-grid--row-heights--md{--bslib-grid--row-heights: var(--bslib-grid--row-heights--md)}}@media(min-width: 992px){.bslib-grid.grid.bslib-grid--row-heights--lg{--bslib-grid--row-heights: var(--bslib-grid--row-heights--lg)}}@media(min-width: 1200px){.bslib-grid.grid.bslib-grid--row-heights--xl{--bslib-grid--row-heights: var(--bslib-grid--row-heights--xl)}}@media(min-width: 1400px){.bslib-grid.grid.bslib-grid--row-heights--xxl{--bslib-grid--row-heights: var(--bslib-grid--row-heights--xxl)}}.bslib-grid>*>.shiny-input-container{width:100%}.bslib-grid-item{grid-column:auto/span 1}@media(max-width: 767.98px){.bslib-grid-item{grid-column:1/-1}}@media(max-width: 575.98px){.bslib-grid{grid-template-columns:1fr !important;height:var(--bslib-grid-height-mobile)}.bslib-grid.grid{height:unset !important;grid-auto-rows:var(--bslib-grid--row-heights--xs, auto)}}.bslib-card{overflow:auto}.bslib-card .card-body+.card-body{padding-top:0}.bslib-card .card-body{overflow:auto}.bslib-card .card-body p{margin-top:0}.bslib-card .card-body p:last-child{margin-bottom:0}.bslib-card .card-body{max-height:var(--bslib-card-body-max-height, none)}.bslib-card[data-full-screen=true]>.card-body{max-height:var(--bslib-card-body-max-height-full-screen, none)}.bslib-card .card-header .form-group{margin-bottom:0}.bslib-card .card-header .selectize-control{margin-bottom:0}.bslib-card .card-header .selectize-control .item{margin-right:1.15rem}.bslib-card .card-footer{margin-top:auto}.bslib-card .bslib-navs-card-title{display:flex;flex-wrap:wrap;justify-content:space-between;align-items:center}.bslib-card .bslib-navs-card-title .nav{margin-left:auto}.bslib-card .bslib-sidebar-layout:not([data-bslib-sidebar-border=true]){border:none}.bslib-card .bslib-sidebar-layout:not([data-bslib-sidebar-border-radius=true]){border-top-left-radius:0;border-top-right-radius:0}[data-full-screen=true]{position:fixed;inset:3.5rem 1rem 1rem;height:auto !important;max-height:none !important;width:auto !important;z-index:1070}.bslib-full-screen-enter{display:none;position:absolute;bottom:var(--bslib-full-screen-enter-bottom, 0.2rem);right:var(--bslib-full-screen-enter-right, 0);top:var(--bslib-full-screen-enter-top);left:var(--bslib-full-screen-enter-left);color:var(--bslib-color-fg, var(--bs-card-color));background-color:var(--bslib-color-bg, var(--bs-card-bg, var(--bs-body-bg)));border:var(--bs-card-border-width) solid var(--bslib-color-fg, var(--bs-card-border-color));box-shadow:0 2px 4px rgba(0,0,0,.15);margin:.2rem .4rem;padding:.55rem !important;font-size:.8rem;cursor:pointer;opacity:.7;z-index:1070}.bslib-full-screen-enter:hover{opacity:1}.card[data-full-screen=false]:hover>*>.bslib-full-screen-enter{display:block}.bslib-has-full-screen .card:hover>*>.bslib-full-screen-enter{display:none}@media(max-width: 575.98px){.bslib-full-screen-enter{display:none !important}}.bslib-full-screen-exit{position:relative;top:1.35rem;font-size:.9rem;cursor:pointer;text-decoration:none;display:flex;float:right;margin-right:2.15rem;align-items:center;color:rgba(var(--bs-body-bg-rgb), 0.8)}.bslib-full-screen-exit:hover{color:rgba(var(--bs-body-bg-rgb), 1)}.bslib-full-screen-exit svg{margin-left:.5rem;font-size:1.5rem}#bslib-full-screen-overlay{position:fixed;inset:0;background-color:rgba(var(--bs-body-color-rgb), 0.6);backdrop-filter:blur(2px);-webkit-backdrop-filter:blur(2px);z-index:1069;animation:bslib-full-screen-overlay-enter 400ms cubic-bezier(0.6, 0.02, 0.65, 1) forwards}@keyframes bslib-full-screen-overlay-enter{0%{opacity:0}100%{opacity:1}}:root{--bslib-value-box-shadow: none;--bslib-value-box-border-width-auto-yes: var(--bslib-value-box-border-width-baseline);--bslib-value-box-border-width-auto-no: 0;--bslib-value-box-border-width-baseline: 1px}.bslib-value-box{border-width:var(--bslib-value-box-border-width-auto-no, var(--bslib-value-box-border-width-baseline));container-name:bslib-value-box;container-type:inline-size}.bslib-value-box.card{box-shadow:var(--bslib-value-box-shadow)}.bslib-value-box.border-auto{border-width:var(--bslib-value-box-border-width-auto-yes, var(--bslib-value-box-border-width-baseline))}.bslib-value-box.default{--bslib-value-box-bg-default: var(--bs-card-bg, #FEFBF2);--bslib-value-box-border-color-default: var(--bs-card-border-color, rgba(0, 0, 0, 0.175));color:var(--bslib-value-box-color);background-color:var(--bslib-value-box-bg, var(--bslib-value-box-bg-default));border-color:var(--bslib-value-box-border-color, var(--bslib-value-box-border-color-default))}.bslib-value-box .value-box-grid{display:grid;grid-template-areas:"left right";align-items:center;overflow:hidden}.bslib-value-box .value-box-showcase{height:100%;max-height:var(---bslib-value-box-showcase-max-h, 100%)}.bslib-value-box .value-box-showcase,.bslib-value-box .value-box-showcase>.html-fill-item{width:100%}.bslib-value-box[data-full-screen=true] .value-box-showcase{max-height:var(---bslib-value-box-showcase-max-h-fs, 100%)}@media screen and (min-width: 575.98px){@container bslib-value-box (max-width: 300px){.bslib-value-box:not(.showcase-bottom) .value-box-grid{grid-template-columns:1fr !important;grid-template-rows:auto auto;grid-template-areas:"top" "bottom"}.bslib-value-box:not(.showcase-bottom) .value-box-grid .value-box-showcase{grid-area:top !important}.bslib-value-box:not(.showcase-bottom) .value-box-grid .value-box-area{grid-area:bottom !important;justify-content:end}}}.bslib-value-box .value-box-area{justify-content:center;padding:1.5rem 1rem;font-size:.9rem;font-weight:500}.bslib-value-box .value-box-area *{margin-bottom:0;margin-top:0}.bslib-value-box .value-box-title{font-size:1rem;margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2}.bslib-value-box .value-box-title:empty::after{content:" "}.bslib-value-box .value-box-value{font-size:calc(1.29rem + 0.48vw);margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2}@media(min-width: 1200px){.bslib-value-box .value-box-value{font-size:1.65rem}}.bslib-value-box .value-box-value:empty::after{content:" "}.bslib-value-box .value-box-showcase{align-items:center;justify-content:center;margin-top:auto;margin-bottom:auto;padding:1rem}.bslib-value-box .value-box-showcase .bi,.bslib-value-box .value-box-showcase .fa,.bslib-value-box .value-box-showcase .fab,.bslib-value-box .value-box-showcase .fas,.bslib-value-box .value-box-showcase .far{opacity:.85;min-width:50px;max-width:125%}.bslib-value-box .value-box-showcase .bi,.bslib-value-box .value-box-showcase .fa,.bslib-value-box .value-box-showcase .fab,.bslib-value-box .value-box-showcase .fas,.bslib-value-box .value-box-showcase .far{font-size:4rem}.bslib-value-box.showcase-top-right .value-box-grid{grid-template-columns:1fr var(---bslib-value-box-showcase-w, 50%)}.bslib-value-box.showcase-top-right .value-box-grid .value-box-showcase{grid-area:right;margin-left:auto;align-self:start;align-items:end;padding-left:0;padding-bottom:0}.bslib-value-box.showcase-top-right .value-box-grid .value-box-area{grid-area:left;align-self:end}.bslib-value-box.showcase-top-right[data-full-screen=true] .value-box-grid{grid-template-columns:auto var(---bslib-value-box-showcase-w-fs, 1fr)}.bslib-value-box.showcase-top-right[data-full-screen=true] .value-box-grid>div{align-self:center}.bslib-value-box.showcase-top-right:not([data-full-screen=true]) .value-box-showcase{margin-top:0}@container bslib-value-box (max-width: 300px){.bslib-value-box.showcase-top-right:not([data-full-screen=true]) .value-box-grid .value-box-showcase{padding-left:1rem}}.bslib-value-box.showcase-left-center .value-box-grid{grid-template-columns:var(---bslib-value-box-showcase-w, 30%) auto}.bslib-value-box.showcase-left-center[data-full-screen=true] .value-box-grid{grid-template-columns:var(---bslib-value-box-showcase-w-fs, 1fr) auto}.bslib-value-box.showcase-left-center:not([data-fill-screen=true]) .value-box-grid .value-box-showcase{grid-area:left}.bslib-value-box.showcase-left-center:not([data-fill-screen=true]) .value-box-grid .value-box-area{grid-area:right}.bslib-value-box.showcase-bottom .value-box-grid{grid-template-columns:1fr;grid-template-rows:1fr var(---bslib-value-box-showcase-h, auto);grid-template-areas:"top" "bottom";overflow:hidden}.bslib-value-box.showcase-bottom .value-box-grid .value-box-showcase{grid-area:bottom;padding:0;margin:0}.bslib-value-box.showcase-bottom .value-box-grid .value-box-area{grid-area:top}.bslib-value-box.showcase-bottom[data-full-screen=true] .value-box-grid{grid-template-rows:1fr var(---bslib-value-box-showcase-h-fs, 2fr)}.bslib-value-box.showcase-bottom[data-full-screen=true] .value-box-grid .value-box-showcase{padding:1rem}[data-bs-theme=dark] .bslib-value-box{--bslib-value-box-shadow: 0 0.5rem 1rem rgb(0 0 0 / 50%)}:root{--bslib-page-sidebar-title-bg: #052744;--bslib-page-sidebar-title-color: #fff}.bslib-page-title{background-color:var(--bslib-page-sidebar-title-bg);color:var(--bslib-page-sidebar-title-color);font-size:1.25rem;font-weight:300;padding:var(--bslib-spacer, 1rem);padding-left:1.5rem;margin-bottom:0;border-bottom:1px solid #dee2e6}html{height:100%}.bslib-page-fill{width:100%;height:100%;margin:0;padding:var(--bslib-spacer, 1rem);gap:var(--bslib-spacer, 1rem)}@media(max-width: 575.98px){.bslib-page-fill{height:var(--bslib-page-fill-mobile-height, auto)}}@media(min-width: 576px){.nav:not(.nav-hidden){display:flex !important;display:-webkit-flex !important}.nav:not(.nav-hidden):not(.nav-stacked):not(.flex-column){float:none !important}.nav:not(.nav-hidden):not(.nav-stacked):not(.flex-column)>.bslib-nav-spacer{margin-left:auto !important}.nav:not(.nav-hidden):not(.nav-stacked):not(.flex-column)>.form-inline{margin-top:auto;margin-bottom:auto}.nav:not(.nav-hidden).nav-stacked{flex-direction:column;-webkit-flex-direction:column;height:100%}.nav:not(.nav-hidden).nav-stacked>.bslib-nav-spacer{margin-top:auto !important}}.accordion .accordion-header{font-size:calc(1.29rem + 0.48vw);margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2;color:var(--bs-heading-color);margin-bottom:0}@media(min-width: 1200px){.accordion .accordion-header{font-size:1.65rem}}.accordion .accordion-icon:not(:empty){margin-right:.75rem;display:flex}.accordion .accordion-button:not(.collapsed){box-shadow:none}.accordion .accordion-button:not(.collapsed):focus{box-shadow:var(--bs-accordion-btn-focus-box-shadow)}.bslib-sidebar-layout{--bslib-sidebar-transition-duration: 500ms;--bslib-sidebar-transition-easing-x: cubic-bezier(0.8, 0.78, 0.22, 1.07);--bslib-sidebar-border: var(--bs-card-border-width, 1px) solid var(--bs-card-border-color, rgba(0, 0, 0, 0.175));--bslib-sidebar-border-radius: var(--bs-border-radius);--bslib-sidebar-vert-border: var(--bs-card-border-width, 1px) solid var(--bs-card-border-color, rgba(0, 0, 0, 0.175));--bslib-sidebar-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.05);--bslib-sidebar-fg: var(--bs-emphasis-color, black);--bslib-sidebar-main-fg: var(--bs-card-color, var(--bs-body-color));--bslib-sidebar-main-bg: var(--bs-card-bg, var(--bs-body-bg));--bslib-sidebar-toggle-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.1);--bslib-sidebar-padding: calc(var(--bslib-spacer) * 1.5);--bslib-sidebar-icon-size: var(--bslib-spacer, 1rem);--bslib-sidebar-icon-button-size: calc(var(--bslib-sidebar-icon-size, 1rem) * 2);--bslib-sidebar-padding-icon: calc(var(--bslib-sidebar-icon-button-size, 2rem) * 1.5);--bslib-collapse-toggle-border-radius: var(--bs-border-radius, 0.25rem);--bslib-collapse-toggle-transform: 0deg;--bslib-sidebar-toggle-transition-easing: cubic-bezier(1, 0, 0, 1);--bslib-collapse-toggle-right-transform: 180deg;--bslib-sidebar-column-main: minmax(0, 1fr);display:grid !important;grid-template-columns:min(100% - var(--bslib-sidebar-icon-size),var(--bslib-sidebar-width, 250px)) var(--bslib-sidebar-column-main);position:relative;transition:grid-template-columns ease-in-out var(--bslib-sidebar-transition-duration);border:var(--bslib-sidebar-border);border-radius:var(--bslib-sidebar-border-radius)}@media(prefers-reduced-motion: reduce){.bslib-sidebar-layout{transition:none}}.bslib-sidebar-layout[data-bslib-sidebar-border=false]{border:none}.bslib-sidebar-layout[data-bslib-sidebar-border-radius=false]{border-radius:initial}.bslib-sidebar-layout>.main,.bslib-sidebar-layout>.sidebar{grid-row:1/2;border-radius:inherit;overflow:auto}.bslib-sidebar-layout>.main{grid-column:2/3;border-top-left-radius:0;border-bottom-left-radius:0;padding:var(--bslib-sidebar-padding);transition:padding var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration);color:var(--bslib-sidebar-main-fg);background-color:var(--bslib-sidebar-main-bg)}.bslib-sidebar-layout>.sidebar{grid-column:1/2;width:100%;height:100%;border-right:var(--bslib-sidebar-vert-border);border-top-right-radius:0;border-bottom-right-radius:0;color:var(--bslib-sidebar-fg);background-color:var(--bslib-sidebar-bg);backdrop-filter:blur(5px)}.bslib-sidebar-layout>.sidebar>.sidebar-content{display:flex;flex-direction:column;gap:var(--bslib-spacer, 1rem);padding:var(--bslib-sidebar-padding);padding-top:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout>.sidebar>.sidebar-content>:last-child:not(.sidebar-title){margin-bottom:0}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion{margin-left:calc(-1*var(--bslib-sidebar-padding));margin-right:calc(-1*var(--bslib-sidebar-padding))}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:last-child{margin-bottom:calc(-1*var(--bslib-sidebar-padding))}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:not(:last-child){margin-bottom:1rem}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion .accordion-body{display:flex;flex-direction:column}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:not(:first-child) .accordion-item:first-child{border-top:var(--bs-accordion-border-width) solid var(--bs-accordion-border-color)}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:not(:last-child) .accordion-item:last-child{border-bottom:var(--bs-accordion-border-width) solid var(--bs-accordion-border-color)}.bslib-sidebar-layout>.sidebar>.sidebar-content.has-accordion>.sidebar-title{border-bottom:none;padding-bottom:0}.bslib-sidebar-layout>.sidebar .shiny-input-container{width:100%}.bslib-sidebar-layout[data-bslib-sidebar-open=always]>.sidebar>.sidebar-content{padding-top:var(--bslib-sidebar-padding)}.bslib-sidebar-layout>.collapse-toggle{grid-row:1/2;grid-column:1/2;display:inline-flex;align-items:center;position:absolute;right:calc(var(--bslib-sidebar-icon-size));top:calc(var(--bslib-sidebar-icon-size, 1rem)/2);border:none;border-radius:var(--bslib-collapse-toggle-border-radius);height:var(--bslib-sidebar-icon-button-size, 2rem);width:var(--bslib-sidebar-icon-button-size, 2rem);display:flex;align-items:center;justify-content:center;padding:0;color:var(--bslib-sidebar-fg);background-color:unset;transition:color var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration),top var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration),right var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration),left var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration)}.bslib-sidebar-layout>.collapse-toggle:hover{background-color:var(--bslib-sidebar-toggle-bg)}.bslib-sidebar-layout>.collapse-toggle>.collapse-icon{opacity:.8;width:var(--bslib-sidebar-icon-size);height:var(--bslib-sidebar-icon-size);transform:rotateY(var(--bslib-collapse-toggle-transform));transition:transform var(--bslib-sidebar-toggle-transition-easing) var(--bslib-sidebar-transition-duration)}.bslib-sidebar-layout>.collapse-toggle:hover>.collapse-icon{opacity:1}.bslib-sidebar-layout .sidebar-title{font-size:1.25rem;line-height:1.25;margin-top:0;margin-bottom:1rem;padding-bottom:1rem;border-bottom:var(--bslib-sidebar-border)}.bslib-sidebar-layout.sidebar-right{grid-template-columns:var(--bslib-sidebar-column-main) min(100% - var(--bslib-sidebar-icon-size),var(--bslib-sidebar-width, 250px))}.bslib-sidebar-layout.sidebar-right>.main{grid-column:1/2;border-top-right-radius:0;border-bottom-right-radius:0;border-top-left-radius:inherit;border-bottom-left-radius:inherit}.bslib-sidebar-layout.sidebar-right>.sidebar{grid-column:2/3;border-right:none;border-left:var(--bslib-sidebar-vert-border);border-top-left-radius:0;border-bottom-left-radius:0}.bslib-sidebar-layout.sidebar-right>.collapse-toggle{grid-column:2/3;left:var(--bslib-sidebar-icon-size);right:unset;border:var(--bslib-collapse-toggle-border)}.bslib-sidebar-layout.sidebar-right>.collapse-toggle>.collapse-icon{transform:rotateY(var(--bslib-collapse-toggle-right-transform))}.bslib-sidebar-layout.sidebar-collapsed{--bslib-collapse-toggle-transform: 180deg;--bslib-collapse-toggle-right-transform: 0deg;--bslib-sidebar-vert-border: none;grid-template-columns:0 minmax(0, 1fr)}.bslib-sidebar-layout.sidebar-collapsed.sidebar-right{grid-template-columns:minmax(0, 1fr) 0}.bslib-sidebar-layout.sidebar-collapsed:not(.transitioning)>.sidebar>*{display:none}.bslib-sidebar-layout.sidebar-collapsed>.main{border-radius:inherit}.bslib-sidebar-layout.sidebar-collapsed:not(.sidebar-right)>.main{padding-left:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout.sidebar-collapsed.sidebar-right>.main{padding-right:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout.sidebar-collapsed>.collapse-toggle{color:var(--bslib-sidebar-main-fg);top:calc(var(--bslib-sidebar-overlap-counter, 0)*(var(--bslib-sidebar-icon-size) + var(--bslib-sidebar-padding)) + var(--bslib-sidebar-icon-size, 1rem)/2);right:calc(-2.5*var(--bslib-sidebar-icon-size) - var(--bs-card-border-width, 1px))}.bslib-sidebar-layout.sidebar-collapsed.sidebar-right>.collapse-toggle{left:calc(-2.5*var(--bslib-sidebar-icon-size) - var(--bs-card-border-width, 1px));right:unset}@media(min-width: 576px){.bslib-sidebar-layout.transitioning>.sidebar>.sidebar-content{display:none}}@media(max-width: 575.98px){.bslib-sidebar-layout[data-bslib-sidebar-open=desktop]{--bslib-sidebar-js-init-collapsed: true}.bslib-sidebar-layout>.sidebar,.bslib-sidebar-layout.sidebar-right>.sidebar{border:none}.bslib-sidebar-layout>.main,.bslib-sidebar-layout.sidebar-right>.main{grid-column:1/3}.bslib-sidebar-layout[data-bslib-sidebar-open=always]{display:block !important}.bslib-sidebar-layout[data-bslib-sidebar-open=always]>.sidebar{max-height:var(--bslib-sidebar-max-height-mobile);overflow-y:auto;border-top:var(--bslib-sidebar-vert-border)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]){grid-template-columns:100% 0}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]):not(.sidebar-collapsed)>.sidebar{z-index:1}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]):not(.sidebar-collapsed)>.collapse-toggle{z-index:1}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-right{grid-template-columns:0 100%}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-collapsed{grid-template-columns:0 100%}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-collapsed.sidebar-right{grid-template-columns:100% 0}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]):not(.sidebar-right)>.main{padding-left:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-right>.main{padding-right:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always])>.main{opacity:0;transition:opacity var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-collapsed>.main{opacity:1}}.navbar+.container-fluid:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-sm:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-md:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-lg:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-xl:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-xxl:has(>.tab-content>.tab-pane.active.html-fill-container){padding-left:0;padding-right:0}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container{padding:var(--bslib-spacer, 1rem);gap:var(--bslib-spacer, 1rem)}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child){padding:0}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]){border-left:none;border-right:none;border-bottom:none}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]){border-radius:0}.navbar+div>.bslib-sidebar-layout{border-top:var(--bslib-sidebar-border)}.html-fill-container{display:flex;flex-direction:column;min-height:0;min-width:0}.html-fill-container>.html-fill-item{flex:1 1 auto;min-height:0;min-width:0}.html-fill-container>:not(.html-fill-item){flex:0 0 auto}.quarto-container{min-height:calc(100vh - 132px)}body.hypothesis-enabled #quarto-header{margin-right:16px}footer.footer .nav-footer,#quarto-header>nav{padding-left:1em;padding-right:1em}footer.footer div.nav-footer p:first-child{margin-top:0}footer.footer div.nav-footer p:last-child{margin-bottom:0}#quarto-content>*{padding-top:14px}#quarto-content>#quarto-sidebar-glass{padding-top:0px}@media(max-width: 991.98px){#quarto-content>*{padding-top:0}#quarto-content .subtitle{padding-top:14px}#quarto-content section:first-of-type h2:first-of-type,#quarto-content section:first-of-type .h2:first-of-type{margin-top:1rem}}.headroom-target,header.headroom{will-change:transform;transition:position 200ms linear;transition:all 200ms linear}header.headroom--pinned{transform:translateY(0%)}header.headroom--unpinned{transform:translateY(-100%)}.navbar-container{width:100%}.navbar-brand{overflow:hidden;text-overflow:ellipsis}.navbar-brand-container{max-width:calc(100% - 115px);min-width:0;display:flex;align-items:center}@media(min-width: 992px){.navbar-brand-container{margin-right:1em}}.navbar-brand.navbar-brand-logo{margin-right:4px;display:inline-flex}.navbar-toggler{flex-basis:content;flex-shrink:0}.navbar .navbar-brand-container{order:2}.navbar .navbar-toggler{order:1}.navbar .navbar-container>.navbar-nav{order:20}.navbar .navbar-container>.navbar-brand-container{margin-left:0 !important;margin-right:0 !important}.navbar .navbar-collapse{order:20}.navbar #quarto-search{order:4;margin-left:auto}.navbar .navbar-toggler{margin-right:.5em}.navbar-collapse .quarto-navbar-tools{margin-left:.5em}.navbar-logo{max-height:24px;width:auto;padding-right:4px}nav .nav-item:not(.compact){padding-top:1px}nav .nav-link i,nav .dropdown-item i{padding-right:1px}.navbar-expand-lg .navbar-nav .nav-link{padding-left:.6rem;padding-right:.6rem}nav .nav-item.compact .nav-link{padding-left:.5rem;padding-right:.5rem;font-size:1.1rem}.navbar .quarto-navbar-tools{order:3}.navbar .quarto-navbar-tools div.dropdown{display:inline-block}.navbar .quarto-navbar-tools .quarto-navigation-tool{color:#f2e5bd}.navbar .quarto-navbar-tools .quarto-navigation-tool:hover{color:#93b0f1}.navbar-nav .dropdown-menu{min-width:220px;font-size:.9rem}.navbar .navbar-nav .nav-link.dropdown-toggle::after{opacity:.75;vertical-align:.175em}.navbar ul.dropdown-menu{padding-top:0;padding-bottom:0}.navbar .dropdown-header{text-transform:uppercase;font-size:.8rem;padding:0 .5rem}.navbar .dropdown-item{padding:.4rem .5rem}.navbar .dropdown-item>i.bi{margin-left:.1rem;margin-right:.25em}.sidebar #quarto-search{margin-top:-1px}.sidebar #quarto-search svg.aa-SubmitIcon{width:16px;height:16px}.sidebar-navigation a{color:inherit}.sidebar-title{margin-top:.25rem;padding-bottom:.5rem;font-size:1.3rem;line-height:1.6rem;visibility:visible}.sidebar-title>a{font-size:inherit;text-decoration:none}.sidebar-title .sidebar-tools-main{margin-top:-6px}@media(max-width: 991.98px){#quarto-sidebar div.sidebar-header{padding-top:.2em}}.sidebar-header-stacked .sidebar-title{margin-top:.6rem}.sidebar-logo{max-width:90%;padding-bottom:.5rem}.sidebar-logo-link{text-decoration:none}.sidebar-navigation li a{text-decoration:none}.sidebar-navigation .quarto-navigation-tool{opacity:.7;font-size:.875rem}#quarto-sidebar>nav>.sidebar-tools-main{margin-left:14px}.sidebar-tools-main{display:inline-flex;margin-left:0px;order:2}.sidebar-tools-main:not(.tools-wide){vertical-align:middle}.sidebar-navigation .quarto-navigation-tool.dropdown-toggle::after{display:none}.sidebar.sidebar-navigation>*{padding-top:1em}.sidebar-item{margin-bottom:.2em;line-height:1rem;margin-top:.4rem}.sidebar-section{padding-left:.5em;padding-bottom:.2em}.sidebar-item .sidebar-item-container{display:flex;justify-content:space-between;cursor:pointer}.sidebar-item-toggle:hover{cursor:pointer}.sidebar-item .sidebar-item-toggle .bi{font-size:.7rem;text-align:center}.sidebar-item .sidebar-item-toggle .bi-chevron-right::before{transition:transform 200ms ease}.sidebar-item .sidebar-item-toggle[aria-expanded=false] .bi-chevron-right::before{transform:none}.sidebar-item .sidebar-item-toggle[aria-expanded=true] .bi-chevron-right::before{transform:rotate(90deg)}.sidebar-item-text{width:100%}.sidebar-navigation .sidebar-divider{margin-left:0;margin-right:0;margin-top:.5rem;margin-bottom:.5rem}@media(max-width: 991.98px){.quarto-secondary-nav{display:block}.quarto-secondary-nav button.quarto-search-button{padding-right:0em;padding-left:2em}.quarto-secondary-nav button.quarto-btn-toggle{margin-left:-0.75rem;margin-right:.15rem}.quarto-secondary-nav nav.quarto-title-breadcrumbs{display:none}.quarto-secondary-nav nav.quarto-page-breadcrumbs{display:flex;align-items:center;padding-right:1em;margin-left:-0.25em}.quarto-secondary-nav nav.quarto-page-breadcrumbs a{text-decoration:none}.quarto-secondary-nav nav.quarto-page-breadcrumbs ol.breadcrumb{margin-bottom:0}}@media(min-width: 992px){.quarto-secondary-nav{display:none}}.quarto-title-breadcrumbs .breadcrumb{margin-bottom:.5em;font-size:.9rem}.quarto-title-breadcrumbs .breadcrumb li:last-of-type a{color:#6c757d}.quarto-secondary-nav .quarto-btn-toggle{color:#052744}.quarto-secondary-nav[aria-expanded=false] .quarto-btn-toggle .bi-chevron-right::before{transform:none}.quarto-secondary-nav[aria-expanded=true] .quarto-btn-toggle .bi-chevron-right::before{transform:rotate(90deg)}.quarto-secondary-nav .quarto-btn-toggle .bi-chevron-right::before{transition:transform 200ms ease}.quarto-secondary-nav{cursor:pointer}.no-decor{text-decoration:none}.quarto-secondary-nav-title{margin-top:.3em;color:#052744;padding-top:4px}.quarto-secondary-nav nav.quarto-page-breadcrumbs{color:#052744}.quarto-secondary-nav nav.quarto-page-breadcrumbs a{color:#052744}.quarto-secondary-nav nav.quarto-page-breadcrumbs a:hover{color:rgba(32,80,186,.8)}.quarto-secondary-nav nav.quarto-page-breadcrumbs .breadcrumb-item::before{color:#0c5da3}.breadcrumb-item{line-height:1.2rem}div.sidebar-item-container{color:#052744}div.sidebar-item-container:hover,div.sidebar-item-container:focus{color:rgba(32,80,186,.8)}div.sidebar-item-container.disabled{color:rgba(5,39,68,.75)}div.sidebar-item-container .active,div.sidebar-item-container .show>.nav-link,div.sidebar-item-container .sidebar-link>code{color:#2050ba}div.sidebar.sidebar-navigation.rollup.quarto-sidebar-toggle-contents,nav.sidebar.sidebar-navigation:not(.rollup){background-color:#fefcf9}@media(max-width: 991.98px){.sidebar-navigation .sidebar-item a,.nav-page .nav-page-text,.sidebar-navigation{font-size:1rem}.sidebar-navigation ul.sidebar-section.depth1 .sidebar-section-item{font-size:1.1rem}.sidebar-logo{display:none}.sidebar.sidebar-navigation{position:static;border-bottom:1px solid #dee2e6}.sidebar.sidebar-navigation.collapsing{position:fixed;z-index:1000}.sidebar.sidebar-navigation.show{position:fixed;z-index:1000}.sidebar.sidebar-navigation{min-height:100%}nav.quarto-secondary-nav{background-color:#fefcf9;border-bottom:1px solid #dee2e6}.quarto-banner nav.quarto-secondary-nav{background-color:#052744;color:#f2e5bd;border-top:1px solid #dee2e6}.sidebar .sidebar-footer{visibility:visible;padding-top:1rem;position:inherit}.sidebar-tools-collapse{display:block}}#quarto-sidebar{transition:width .15s ease-in}#quarto-sidebar>*{padding-right:1em}@media(max-width: 991.98px){#quarto-sidebar .sidebar-menu-container{white-space:nowrap;min-width:225px}#quarto-sidebar.show{transition:width .15s ease-out}}@media(min-width: 992px){#quarto-sidebar{display:flex;flex-direction:column}.nav-page .nav-page-text,.sidebar-navigation .sidebar-section .sidebar-item{font-size:.875rem}.sidebar-navigation .sidebar-item{font-size:.925rem}.sidebar.sidebar-navigation{display:block;position:sticky}.sidebar-search{width:100%}.sidebar .sidebar-footer{visibility:visible}}@media(min-width: 992px){#quarto-sidebar-glass{display:none}}@media(max-width: 991.98px){#quarto-sidebar-glass{position:fixed;top:0;bottom:0;left:0;right:0;background-color:rgba(255,255,255,0);transition:background-color .15s ease-in;z-index:-1}#quarto-sidebar-glass.collapsing{z-index:1000}#quarto-sidebar-glass.show{transition:background-color .15s ease-out;background-color:rgba(102,102,102,.4);z-index:1000}}.sidebar .sidebar-footer{padding:.5rem 1rem;align-self:flex-end;color:#6c757d;width:100%}.quarto-page-breadcrumbs .breadcrumb-item+.breadcrumb-item,.quarto-page-breadcrumbs .breadcrumb-item{padding-right:.33em;padding-left:0}.quarto-page-breadcrumbs .breadcrumb-item::before{padding-right:.33em}.quarto-sidebar-footer{font-size:.875em}.sidebar-section .bi-chevron-right{vertical-align:middle}.sidebar-section .bi-chevron-right::before{font-size:.9em}.notransition{-webkit-transition:none !important;-moz-transition:none !important;-o-transition:none !important;transition:none !important}.btn:focus:not(:focus-visible){box-shadow:none}.page-navigation{display:flex;justify-content:space-between}.nav-page{padding-bottom:.75em}.nav-page .bi{font-size:1.8rem;vertical-align:middle}.nav-page .nav-page-text{padding-left:.25em;padding-right:.25em}.nav-page a{color:#6c757d;text-decoration:none;display:flex;align-items:center}.nav-page a:hover{color:#1f4eb6}.nav-footer .toc-actions{padding-bottom:.5em;padding-top:.5em}.nav-footer .toc-actions a,.nav-footer .toc-actions a:hover{text-decoration:none}.nav-footer .toc-actions ul{display:flex;list-style:none}.nav-footer .toc-actions ul :first-child{margin-left:auto}.nav-footer .toc-actions ul :last-child{margin-right:auto}.nav-footer .toc-actions ul li{padding-right:1.5em}.nav-footer .toc-actions ul li i.bi{padding-right:.4em}.nav-footer .toc-actions ul li:last-of-type{padding-right:0}.nav-footer{display:flex;flex-direction:row;flex-wrap:wrap;justify-content:space-between;align-items:baseline;text-align:center;padding-top:.5rem;padding-bottom:.5rem;background-color:#fefbf2}body.nav-fixed{padding-top:64px}.nav-footer-contents{color:#6c757d;margin-top:.25rem}.nav-footer{min-height:3.5em;color:#75736f}.nav-footer a{color:#75736f}.nav-footer .nav-footer-left{font-size:.825em}.nav-footer .nav-footer-center{font-size:.825em}.nav-footer .nav-footer-right{font-size:.825em}.nav-footer-left .footer-items,.nav-footer-center .footer-items,.nav-footer-right .footer-items{display:inline-flex;padding-top:.3em;padding-bottom:.3em;margin-bottom:0em}.nav-footer-left .footer-items .nav-link,.nav-footer-center .footer-items .nav-link,.nav-footer-right .footer-items .nav-link{padding-left:.6em;padding-right:.6em}@media(min-width: 768px){.nav-footer-left{flex:1 1 0px;text-align:left}}@media(max-width: 575.98px){.nav-footer-left{margin-bottom:1em;flex:100%}}@media(min-width: 768px){.nav-footer-right{flex:1 1 0px;text-align:right}}@media(max-width: 575.98px){.nav-footer-right{margin-bottom:1em;flex:100%}}.nav-footer-center{text-align:center;min-height:3em}@media(min-width: 768px){.nav-footer-center{flex:1 1 0px}}.nav-footer-center .footer-items{justify-content:center}@media(max-width: 767.98px){.nav-footer-center{margin-bottom:1em;flex:100%}}@media(max-width: 767.98px){.nav-footer-center{margin-top:3em;order:10}}.navbar .quarto-reader-toggle.reader .quarto-reader-toggle-btn{background-color:#f2e5bd;border-radius:3px}@media(max-width: 991.98px){.quarto-reader-toggle{display:none}}.quarto-reader-toggle.reader.quarto-navigation-tool .quarto-reader-toggle-btn{background-color:#052744;border-radius:3px}.quarto-reader-toggle .quarto-reader-toggle-btn{display:inline-flex;padding-left:.2em;padding-right:.2em;margin-left:-0.2em;margin-right:-0.2em;text-align:center}.navbar .quarto-reader-toggle:not(.reader) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(242, 229, 189, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}.navbar .quarto-reader-toggle.reader .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(5, 39, 68, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}.sidebar-navigation .quarto-reader-toggle:not(.reader) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(5, 39, 68, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}.sidebar-navigation .quarto-reader-toggle.reader .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(254, 252, 249, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}#quarto-back-to-top{display:none;position:fixed;bottom:50px;background-color:#fefbf2;border-radius:.25rem;box-shadow:0 .2rem .5rem #6c757d,0 0 .05rem #6c757d;color:#6c757d;text-decoration:none;font-size:.9em;text-align:center;left:50%;padding:.4rem .8rem;transform:translate(-50%, 0)}#quarto-announcement{padding:.5em;display:flex;justify-content:space-between;margin-bottom:0;font-size:.9em}#quarto-announcement .quarto-announcement-content{margin-right:auto}#quarto-announcement .quarto-announcement-content p{margin-bottom:0}#quarto-announcement .quarto-announcement-icon{margin-right:.5em;font-size:1.2em;margin-top:-0.15em}#quarto-announcement .quarto-announcement-action{cursor:pointer}.aa-DetachedSearchButtonQuery{display:none}.aa-DetachedOverlay ul.aa-List,#quarto-search-results ul.aa-List{list-style:none;padding-left:0}.aa-DetachedOverlay .aa-Panel,#quarto-search-results .aa-Panel{background-color:#fefbf2;position:absolute;z-index:2000}#quarto-search-results .aa-Panel{max-width:400px}#quarto-search input{font-size:.925rem}@media(min-width: 992px){.navbar #quarto-search{margin-left:.25rem;order:999}}.navbar.navbar-expand-sm #quarto-search,.navbar.navbar-expand-md #quarto-search{order:999}@media(min-width: 992px){.navbar .quarto-navbar-tools{order:900}}@media(min-width: 992px){.navbar .quarto-navbar-tools.tools-end{margin-left:auto !important}}@media(max-width: 991.98px){#quarto-sidebar .sidebar-search{display:none}}#quarto-sidebar .sidebar-search .aa-Autocomplete{width:100%}.navbar .aa-Autocomplete .aa-Form{width:180px}.navbar #quarto-search.type-overlay .aa-Autocomplete{width:40px}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form{background-color:inherit;border:none}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form:focus-within{box-shadow:none;outline:none}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-InputWrapper{display:none}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-InputWrapper:focus-within{display:inherit}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-Label svg,.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-LoadingIndicator svg{width:26px;height:26px;color:#f2e5bd;opacity:1}.navbar #quarto-search.type-overlay .aa-Autocomplete svg.aa-SubmitIcon{width:26px;height:26px;color:#f2e5bd;opacity:1}.aa-Autocomplete .aa-Form,.aa-DetachedFormContainer .aa-Form{align-items:center;background-color:#fefbf2;border:1px solid #dee2e6;border-radius:.25rem;color:#343a40;display:flex;line-height:1em;margin:0;position:relative;width:100%}.aa-Autocomplete .aa-Form:focus-within,.aa-DetachedFormContainer .aa-Form:focus-within{box-shadow:rgba(39,128,227,.6) 0 0 0 1px;outline:currentColor none medium}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix{align-items:center;display:flex;flex-shrink:0;order:1}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-Label,.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-Label,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator{cursor:initial;flex-shrink:0;padding:0;text-align:left}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-Label svg,.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator svg,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-Label svg,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator svg{color:#343a40;opacity:.5}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-SubmitButton,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-SubmitButton{appearance:none;background:none;border:0;margin:0}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator{align-items:center;display:flex;justify-content:center}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator[hidden],.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator[hidden]{display:none}.aa-Autocomplete .aa-Form .aa-InputWrapper,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper{order:3;position:relative;width:100%}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input{appearance:none;background:none;border:0;color:#343a40;font:inherit;height:calc(1.5em + .1rem + 2px);padding:0;width:100%}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::placeholder,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::placeholder{color:#343a40;opacity:.8}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input:focus,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input:focus{border-color:none;box-shadow:none;outline:none}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-decoration,.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-cancel-button,.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-button,.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-decoration,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-decoration,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-cancel-button,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-button,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-decoration{display:none}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix{align-items:center;display:flex;order:4}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton{align-items:center;background:none;border:0;color:#343a40;opacity:.8;cursor:pointer;display:flex;margin:0;width:calc(1.5em + .1rem + 2px)}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:hover,.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:focus,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:hover,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:focus{color:#343a40;opacity:.8}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton[hidden],.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton[hidden]{display:none}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton svg,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton svg{width:calc(1.5em + 0.75rem + calc(1px * 2))}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton{border:none;align-items:center;background:none;color:#343a40;opacity:.4;font-size:.7rem;cursor:pointer;display:none;margin:0;width:calc(1em + .1rem + 2px)}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:hover,.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:focus,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:hover,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:focus{color:#343a40;opacity:.8}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton[hidden],.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton[hidden]{display:none}.aa-PanelLayout:empty{display:none}.quarto-search-no-results.no-query{display:none}.aa-Source:has(.no-query){display:none}#quarto-search-results .aa-Panel{border:solid #dee2e6 1px}#quarto-search-results .aa-SourceNoResults{width:398px}.aa-DetachedOverlay .aa-Panel,#quarto-search-results .aa-Panel{max-height:65vh;overflow-y:auto;font-size:.925rem}.aa-DetachedOverlay .aa-SourceNoResults,#quarto-search-results .aa-SourceNoResults{height:60px;display:flex;justify-content:center;align-items:center}.aa-DetachedOverlay .search-error,#quarto-search-results .search-error{padding-top:10px;padding-left:20px;padding-right:20px;cursor:default}.aa-DetachedOverlay .search-error .search-error-title,#quarto-search-results .search-error .search-error-title{font-size:1.1rem;margin-bottom:.5rem}.aa-DetachedOverlay .search-error .search-error-title .search-error-icon,#quarto-search-results .search-error .search-error-title .search-error-icon{margin-right:8px}.aa-DetachedOverlay .search-error .search-error-text,#quarto-search-results .search-error .search-error-text{font-weight:300}.aa-DetachedOverlay .search-result-text,#quarto-search-results .search-result-text{font-weight:300;overflow:hidden;text-overflow:ellipsis;display:-webkit-box;-webkit-line-clamp:2;-webkit-box-orient:vertical;line-height:1.2rem;max-height:2.4rem}.aa-DetachedOverlay .aa-SourceHeader .search-result-header,#quarto-search-results .aa-SourceHeader .search-result-header{font-size:.875rem;background-color:#fcf4da;padding-left:14px;padding-bottom:4px;padding-top:4px}.aa-DetachedOverlay .aa-SourceHeader .search-result-header-no-results,#quarto-search-results .aa-SourceHeader .search-result-header-no-results{display:none}.aa-DetachedOverlay .aa-SourceFooter .algolia-search-logo,#quarto-search-results .aa-SourceFooter .algolia-search-logo{width:110px;opacity:.85;margin:8px;float:right}.aa-DetachedOverlay .search-result-section,#quarto-search-results .search-result-section{font-size:.925em}.aa-DetachedOverlay a.search-result-link,#quarto-search-results a.search-result-link{color:inherit;text-decoration:none}.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item,#quarto-search-results li.aa-Item[aria-selected=true] .search-item{background-color:#2780e3}.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item.search-result-more,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-section,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-text,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-title-container,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-text-container,#quarto-search-results li.aa-Item[aria-selected=true] .search-item.search-result-more,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-section,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-text,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-title-container,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-text-container{color:#fff;background-color:#2780e3}.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item mark.search-match,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-match.mark,#quarto-search-results li.aa-Item[aria-selected=true] .search-item mark.search-match,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-match.mark{color:#fff;background-color:#4b95e8}.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item,#quarto-search-results li.aa-Item[aria-selected=false] .search-item{background-color:#fefbf2}.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item.search-result-more,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-section,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-text,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-title-container,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-text-container,#quarto-search-results li.aa-Item[aria-selected=false] .search-item.search-result-more,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-section,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-text,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-title-container,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-text-container{color:#343a40}.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item mark.search-match,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-match.mark,#quarto-search-results li.aa-Item[aria-selected=false] .search-item mark.search-match,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-match.mark{color:inherit;background-color:#e5effc}.aa-DetachedOverlay .aa-Item .search-result-doc:not(.document-selectable) .search-result-title-container,#quarto-search-results .aa-Item .search-result-doc:not(.document-selectable) .search-result-title-container{background-color:#fefbf2;color:#343a40}.aa-DetachedOverlay .aa-Item .search-result-doc:not(.document-selectable) .search-result-text-container,#quarto-search-results .aa-Item .search-result-doc:not(.document-selectable) .search-result-text-container{padding-top:0px}.aa-DetachedOverlay li.aa-Item .search-result-doc.document-selectable .search-result-text-container,#quarto-search-results li.aa-Item .search-result-doc.document-selectable .search-result-text-container{margin-top:-4px}.aa-DetachedOverlay .aa-Item,#quarto-search-results .aa-Item{cursor:pointer}.aa-DetachedOverlay .aa-Item .search-item,#quarto-search-results .aa-Item .search-item{border-left:none;border-right:none;border-top:none;background-color:#fefbf2;border-color:#dee2e6;color:#343a40}.aa-DetachedOverlay .aa-Item .search-item p,#quarto-search-results .aa-Item .search-item p{margin-top:0;margin-bottom:0}.aa-DetachedOverlay .aa-Item .search-item i.bi,#quarto-search-results .aa-Item .search-item i.bi{padding-left:8px;padding-right:8px;font-size:1.3em}.aa-DetachedOverlay .aa-Item .search-item .search-result-title,#quarto-search-results .aa-Item .search-item .search-result-title{margin-top:.3em;margin-bottom:0em}.aa-DetachedOverlay .aa-Item .search-item .search-result-crumbs,#quarto-search-results .aa-Item .search-item .search-result-crumbs{white-space:nowrap;text-overflow:ellipsis;font-size:.8em;font-weight:300;margin-right:1em}.aa-DetachedOverlay .aa-Item .search-item .search-result-crumbs:not(.search-result-crumbs-wrap),#quarto-search-results .aa-Item .search-item .search-result-crumbs:not(.search-result-crumbs-wrap){max-width:30%;margin-left:auto;margin-top:.5em;margin-bottom:.1rem}.aa-DetachedOverlay .aa-Item .search-item .search-result-crumbs.search-result-crumbs-wrap,#quarto-search-results .aa-Item .search-item .search-result-crumbs.search-result-crumbs-wrap{flex-basis:100%;margin-top:0em;margin-bottom:.2em;margin-left:37px}.aa-DetachedOverlay .aa-Item .search-result-title-container,#quarto-search-results .aa-Item .search-result-title-container{font-size:1em;display:flex;flex-wrap:wrap;padding:6px 4px 6px 4px}.aa-DetachedOverlay .aa-Item .search-result-text-container,#quarto-search-results .aa-Item .search-result-text-container{padding-bottom:8px;padding-right:8px;margin-left:42px}.aa-DetachedOverlay .aa-Item .search-result-doc-section,.aa-DetachedOverlay .aa-Item .search-result-more,#quarto-search-results .aa-Item .search-result-doc-section,#quarto-search-results .aa-Item .search-result-more{padding-top:8px;padding-bottom:8px;padding-left:44px}.aa-DetachedOverlay .aa-Item .search-result-more,#quarto-search-results .aa-Item .search-result-more{font-size:.8em;font-weight:400}.aa-DetachedOverlay .aa-Item .search-result-doc,#quarto-search-results .aa-Item .search-result-doc{border-top:1px solid #dee2e6}.aa-DetachedSearchButton{background:none;border:none}.aa-DetachedSearchButton .aa-DetachedSearchButtonPlaceholder{display:none}.navbar .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon{color:#f2e5bd}.sidebar-tools-collapse #quarto-search,.sidebar-tools-main #quarto-search{display:inline}.sidebar-tools-collapse #quarto-search .aa-Autocomplete,.sidebar-tools-main #quarto-search .aa-Autocomplete{display:inline}.sidebar-tools-collapse #quarto-search .aa-DetachedSearchButton,.sidebar-tools-main #quarto-search .aa-DetachedSearchButton{padding-left:4px;padding-right:4px}.sidebar-tools-collapse #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon,.sidebar-tools-main #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon{color:#052744}.sidebar-tools-collapse #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon .aa-SubmitIcon,.sidebar-tools-main #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon .aa-SubmitIcon{margin-top:-3px}.aa-DetachedContainer{background:rgba(254,251,242,.65);width:90%;bottom:0;box-shadow:rgba(222,226,230,.6) 0 0 0 1px;outline:currentColor none medium;display:flex;flex-direction:column;left:0;margin:0;overflow:hidden;padding:0;position:fixed;right:0;top:0;z-index:1101}.aa-DetachedContainer::after{height:32px}.aa-DetachedContainer .aa-SourceHeader{margin:var(--aa-spacing-half) 0 var(--aa-spacing-half) 2px}.aa-DetachedContainer .aa-Panel{background-color:#fefbf2;border-radius:0;box-shadow:none;flex-grow:1;margin:0;padding:0;position:relative}.aa-DetachedContainer .aa-PanelLayout{bottom:0;box-shadow:none;left:0;margin:0;max-height:none;overflow-y:auto;position:absolute;right:0;top:0;width:100%}.aa-DetachedFormContainer{background-color:#fefbf2;border-bottom:1px solid #dee2e6;display:flex;flex-direction:row;justify-content:space-between;margin:0;padding:.5em}.aa-DetachedCancelButton{background:none;font-size:.8em;border:0;border-radius:3px;color:#343a40;cursor:pointer;margin:0 0 0 .5em;padding:0 .5em}.aa-DetachedCancelButton:hover,.aa-DetachedCancelButton:focus{box-shadow:rgba(39,128,227,.6) 0 0 0 1px;outline:currentColor none medium}.aa-DetachedContainer--modal{bottom:inherit;height:auto;margin:0 auto;position:absolute;top:100px;border-radius:6px;max-width:850px}@media(max-width: 575.98px){.aa-DetachedContainer--modal{width:100%;top:0px;border-radius:0px;border:none}}.aa-DetachedContainer--modal .aa-PanelLayout{max-height:var(--aa-detached-modal-max-height);padding-bottom:var(--aa-spacing-half);position:static}.aa-Detached{height:100vh;overflow:hidden}.aa-DetachedOverlay{background-color:rgba(52,58,64,.4);position:fixed;left:0;right:0;top:0;margin:0;padding:0;height:100vh;z-index:1100}.quarto-dashboard.nav-fixed.dashboard-sidebar #quarto-content.quarto-dashboard-content{padding:0em}.quarto-dashboard #quarto-content.quarto-dashboard-content{padding:1em}.quarto-dashboard #quarto-content.quarto-dashboard-content>*{padding-top:0}@media(min-width: 576px){.quarto-dashboard{height:100%}}.quarto-dashboard .card.valuebox.bslib-card.bg-primary{background-color:#5397e9 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-secondary{background-color:#343a40 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-success{background-color:#3aa716 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-info{background-color:rgba(153,84,187,.7019607843) !important}.quarto-dashboard .card.valuebox.bslib-card.bg-warning{background-color:#fa6400 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-danger{background-color:rgba(255,0,57,.7019607843) !important}.quarto-dashboard .card.valuebox.bslib-card.bg-light{background-color:#f8f9fa !important}.quarto-dashboard .card.valuebox.bslib-card.bg-dark{background-color:#343a40 !important}.quarto-dashboard.dashboard-fill{display:flex;flex-direction:column}.quarto-dashboard #quarto-appendix{display:none}.quarto-dashboard #quarto-header #quarto-dashboard-header{border-top:solid 1px #084274;border-bottom:solid 1px #084274}.quarto-dashboard #quarto-header #quarto-dashboard-header>nav{padding-left:1em;padding-right:1em}.quarto-dashboard #quarto-header #quarto-dashboard-header>nav .navbar-brand-container{padding-left:0}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-toggler{margin-right:0}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-toggler-icon{height:1em;width:1em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(242, 229, 189)" class="bi bi-menu-button-wide" viewBox="0 0 16 16"><path d="M0 1.5A1.5 1.5 0 0 1 1.5 0h13A1.5 1.5 0 0 1 16 1.5v2A1.5 1.5 0 0 1 14.5 5h-13A1.5 1.5 0 0 1 0 3.5v-2zM1.5 1a.5.5 0 0 0-.5.5v2a.5.5 0 0 0 .5.5h13a.5.5 0 0 0 .5-.5v-2a.5.5 0 0 0-.5-.5h-13z"/><path d="M2 2.5a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5zm10.823.323-.396-.396A.25.25 0 0 1 12.604 2h.792a.25.25 0 0 1 .177.427l-.396.396a.25.25 0 0 1-.354 0zM0 8a2 2 0 0 1 2-2h12a2 2 0 0 1 2 2v5a2 2 0 0 1-2 2H2a2 2 0 0 1-2-2V8zm1 3v2a1 1 0 0 0 1 1h12a1 1 0 0 0 1-1v-2H1zm14-1V8a1 1 0 0 0-1-1H2a1 1 0 0 0-1 1v2h14zM2 8.5a.5.5 0 0 1 .5-.5h9a.5.5 0 0 1 0 1h-9a.5.5 0 0 1-.5-.5zm0 4a.5.5 0 0 1 .5-.5h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5z"/></svg>')}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-brand-container{padding-right:1em}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-title{font-size:1.1em}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-nav{font-size:.9em}.quarto-dashboard #quarto-dashboard-header .navbar{padding:0}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-container{padding-left:1em}.quarto-dashboard #quarto-dashboard-header .navbar.slim .navbar-brand-container .nav-link,.quarto-dashboard #quarto-dashboard-header .navbar.slim .navbar-nav .nav-link{padding:.7em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-color-scheme-toggle{order:9}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-toggler{margin-left:.5em;order:10}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-nav .nav-link{padding:.5em;height:100%;display:flex;align-items:center}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-nav .active{background-color:#083d6a}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-brand-container{padding:.5em .5em .5em 0;display:flex;flex-direction:row;margin-right:2em;align-items:center}@media(max-width: 767.98px){.quarto-dashboard #quarto-dashboard-header .navbar .navbar-brand-container{margin-right:auto}}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse{align-self:stretch}@media(min-width: 768px){.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse{order:8}}@media(max-width: 767.98px){.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse{order:1000;padding-bottom:.5em}}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse .navbar-nav{align-self:stretch}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-title{font-size:1.25em;line-height:1.1em;display:flex;flex-direction:row;flex-wrap:wrap;align-items:baseline}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-title .navbar-title-text{margin-right:.4em}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-title a{text-decoration:none;color:inherit}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-subtitle,.quarto-dashboard #quarto-dashboard-header .navbar .navbar-author{font-size:.9rem;margin-right:.5em}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-author{margin-left:auto}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-logo{max-height:48px;min-height:30px;object-fit:cover;margin-right:1em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-dashboard-links{order:9;padding-right:1em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-dashboard-link-text{margin-left:.25em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-dashboard-link{padding-right:0em;padding-left:.7em;text-decoration:none;color:#f2e5bd}.quarto-dashboard .page-layout-custom .tab-content{padding:0;border:none}.quarto-dashboard-img-contain{height:100%;width:100%;object-fit:contain}@media(max-width: 575.98px){.quarto-dashboard .bslib-grid{grid-template-rows:minmax(1em, max-content) !important}.quarto-dashboard .sidebar-content{height:inherit}.quarto-dashboard .page-layout-custom{min-height:100vh}}.quarto-dashboard.dashboard-toolbar>.page-layout-custom,.quarto-dashboard.dashboard-sidebar>.page-layout-custom{padding:0}.quarto-dashboard .quarto-dashboard-content.quarto-dashboard-pages{padding:0}.quarto-dashboard .callout{margin-bottom:0;margin-top:0}.quarto-dashboard .html-fill-container figure{overflow:hidden}.quarto-dashboard bslib-tooltip .rounded-pill{border:solid #6c757d 1px}.quarto-dashboard bslib-tooltip .rounded-pill .svg{fill:#343a40}.quarto-dashboard .tabset .dashboard-card-no-title .nav-tabs{margin-left:0;margin-right:auto}.quarto-dashboard .tabset .tab-content{border:none}.quarto-dashboard .tabset .card-header .nav-link[role=tab]{margin-top:-6px;padding-top:6px;padding-bottom:6px}.quarto-dashboard .card.valuebox,.quarto-dashboard .card.bslib-value-box{min-height:3rem}.quarto-dashboard .card.valuebox .card-body,.quarto-dashboard .card.bslib-value-box .card-body{padding:0}.quarto-dashboard .bslib-value-box .value-box-value{font-size:clamp(.1em,15cqw,5em)}.quarto-dashboard .bslib-value-box .value-box-showcase .bi{font-size:clamp(.1em,max(18cqw,5.2cqh),5em);text-align:center;height:1em}.quarto-dashboard .bslib-value-box .value-box-showcase .bi::before{vertical-align:1em}.quarto-dashboard .bslib-value-box .value-box-area{margin-top:auto;margin-bottom:auto}.quarto-dashboard .card figure.quarto-float{display:flex;flex-direction:column;align-items:center}.quarto-dashboard .dashboard-scrolling{padding:1em}.quarto-dashboard .full-height{height:100%}.quarto-dashboard .showcase-bottom .value-box-grid{display:grid;grid-template-columns:1fr;grid-template-rows:1fr auto;grid-template-areas:"top" "bottom"}.quarto-dashboard .showcase-bottom .value-box-grid .value-box-showcase{grid-area:bottom;padding:0;margin:0}.quarto-dashboard .showcase-bottom .value-box-grid .value-box-showcase i.bi{font-size:4rem}.quarto-dashboard .showcase-bottom .value-box-grid .value-box-area{grid-area:top}.quarto-dashboard .tab-content{margin-bottom:0}.quarto-dashboard .bslib-card .bslib-navs-card-title{justify-content:stretch;align-items:end}.quarto-dashboard .card-header{display:flex;flex-wrap:wrap;justify-content:space-between}.quarto-dashboard .card-header .card-title{display:flex;flex-direction:column;justify-content:center;margin-bottom:0}.quarto-dashboard .tabset .card-toolbar{margin-bottom:1em}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout{border:none;gap:var(--bslib-spacer, 1rem)}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.main{padding:0}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.sidebar{border-radius:.25rem;border:1px solid rgba(0,0,0,.175)}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.collapse-toggle{display:none}@media(max-width: 767.98px){.quarto-dashboard .bslib-grid>.bslib-sidebar-layout{grid-template-columns:1fr;grid-template-rows:max-content 1fr}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.main{grid-column:1;grid-row:2}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout .sidebar{grid-column:1;grid-row:1}}.quarto-dashboard .sidebar-right .sidebar{padding-left:2.5em}.quarto-dashboard .sidebar-right .collapse-toggle{left:2px}.quarto-dashboard .quarto-dashboard .sidebar-right button.collapse-toggle:not(.transitioning){left:unset}.quarto-dashboard aside.sidebar{padding-left:1em;padding-right:1em;background-color:rgba(52,58,64,.25);color:#343a40}.quarto-dashboard .bslib-sidebar-layout>div.main{padding:.7em}.quarto-dashboard .bslib-sidebar-layout button.collapse-toggle{margin-top:.3em}.quarto-dashboard .bslib-sidebar-layout .collapse-toggle{top:0}.quarto-dashboard .bslib-sidebar-layout.sidebar-collapsed:not(.transitioning):not(.sidebar-right) .collapse-toggle{left:2px}.quarto-dashboard .sidebar>section>.h3:first-of-type{margin-top:0em}.quarto-dashboard .sidebar .h3,.quarto-dashboard .sidebar .h4,.quarto-dashboard .sidebar .h5,.quarto-dashboard .sidebar .h6{margin-top:.5em}.quarto-dashboard .sidebar form{flex-direction:column;align-items:start;margin-bottom:1em}.quarto-dashboard .sidebar form div[class*=oi-][class$=-input]{flex-direction:column}.quarto-dashboard .sidebar form[class*=oi-][class$=-toggle]{flex-direction:row-reverse;align-items:center;justify-content:start}.quarto-dashboard .sidebar form input[type=range]{margin-top:.5em;margin-right:.8em;margin-left:1em}.quarto-dashboard .sidebar label{width:fit-content}.quarto-dashboard .sidebar .card-body{margin-bottom:2em}.quarto-dashboard .sidebar .shiny-input-container{margin-bottom:1em}.quarto-dashboard .sidebar .shiny-options-group{margin-top:0}.quarto-dashboard .sidebar .control-label{margin-bottom:.3em}.quarto-dashboard .card .card-body .quarto-layout-row{align-items:stretch}.quarto-dashboard .toolbar{font-size:.9em;display:flex;flex-direction:row;border-top:solid 1px #bcbfc0;padding:1em;flex-wrap:wrap;background-color:rgba(52,58,64,.25)}.quarto-dashboard .toolbar .cell-output-display{display:flex}.quarto-dashboard .toolbar .shiny-input-container{padding-bottom:.5em;margin-bottom:.5em;width:inherit}.quarto-dashboard .toolbar .shiny-input-container>.checkbox:first-child{margin-top:6px}.quarto-dashboard .toolbar>*:last-child{margin-right:0}.quarto-dashboard .toolbar>*>*{margin-right:1em;align-items:baseline}.quarto-dashboard .toolbar>*>*>a{text-decoration:none;margin-top:auto;margin-bottom:auto}.quarto-dashboard .toolbar .shiny-input-container{padding-bottom:0;margin-bottom:0}.quarto-dashboard .toolbar .shiny-input-container>*{flex-shrink:0;flex-grow:0}.quarto-dashboard .toolbar .form-group.shiny-input-container:not([role=group])>label{margin-bottom:0}.quarto-dashboard .toolbar .shiny-input-container.no-baseline{align-items:start;padding-top:6px}.quarto-dashboard .toolbar .shiny-input-container{display:flex;align-items:baseline}.quarto-dashboard .toolbar .shiny-input-container label{padding-right:.4em}.quarto-dashboard .toolbar .shiny-input-container .bslib-input-switch{margin-top:6px}.quarto-dashboard .toolbar input[type=text]{line-height:1;width:inherit}.quarto-dashboard .toolbar .input-daterange{width:inherit}.quarto-dashboard .toolbar .input-daterange input[type=text]{height:2.4em;width:10em}.quarto-dashboard .toolbar .input-daterange .input-group-addon{height:auto;padding:0;margin-left:-5px !important;margin-right:-5px}.quarto-dashboard .toolbar .input-daterange .input-group-addon .input-group-text{padding-top:0;padding-bottom:0;height:100%}.quarto-dashboard .toolbar span.irs.irs--shiny{width:10em}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-line{top:9px}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-min,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-max,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-from,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-to,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-single{top:20px}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-bar{top:8px}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-handle{top:0px}.quarto-dashboard .toolbar .shiny-input-checkboxgroup>label{margin-top:6px}.quarto-dashboard .toolbar .shiny-input-checkboxgroup>.shiny-options-group{margin-top:0;align-items:baseline}.quarto-dashboard .toolbar .shiny-input-radiogroup>label{margin-top:6px}.quarto-dashboard .toolbar .shiny-input-radiogroup>.shiny-options-group{align-items:baseline;margin-top:0}.quarto-dashboard .toolbar .shiny-input-radiogroup>.shiny-options-group>.radio{margin-right:.3em}.quarto-dashboard .toolbar .form-select{padding-top:.2em;padding-bottom:.2em}.quarto-dashboard .toolbar .shiny-input-select{min-width:6em}.quarto-dashboard .toolbar div.checkbox{margin-bottom:0px}.quarto-dashboard .toolbar>.checkbox:first-child{margin-top:6px}.quarto-dashboard .toolbar form{width:fit-content}.quarto-dashboard .toolbar form label{padding-top:.2em;padding-bottom:.2em;width:fit-content}.quarto-dashboard .toolbar form input[type=date]{width:fit-content}.quarto-dashboard .toolbar form input[type=color]{width:3em}.quarto-dashboard .toolbar form button{padding:.4em}.quarto-dashboard .toolbar form select{width:fit-content}.quarto-dashboard .toolbar>*{font-size:.9em;flex-grow:0}.quarto-dashboard .toolbar .shiny-input-container label{margin-bottom:1px}.quarto-dashboard .toolbar-bottom{margin-top:1em;margin-bottom:0 !important;order:2}.quarto-dashboard .quarto-dashboard-content>.dashboard-toolbar-container>.toolbar-content>.tab-content>.tab-pane>*:not(.bslib-sidebar-layout){padding:1em}.quarto-dashboard .quarto-dashboard-content>.dashboard-toolbar-container>.toolbar-content>*:not(.tab-content){padding:1em}.quarto-dashboard .quarto-dashboard-content>.tab-content>.dashboard-page>.dashboard-toolbar-container>.toolbar-content,.quarto-dashboard .quarto-dashboard-content>.tab-content>.dashboard-page:not(.dashboard-sidebar-container)>*:not(.dashboard-toolbar-container){padding:1em}.quarto-dashboard .toolbar-content{padding:0}.quarto-dashboard .quarto-dashboard-content.quarto-dashboard-pages .tab-pane>.dashboard-toolbar-container .toolbar{border-radius:0;margin-bottom:0}.quarto-dashboard .dashboard-toolbar-container.toolbar-toplevel .toolbar{border-bottom:1px solid rgba(0,0,0,.175)}.quarto-dashboard .dashboard-toolbar-container.toolbar-toplevel .toolbar-bottom{margin-top:0}.quarto-dashboard .dashboard-toolbar-container:not(.toolbar-toplevel) .toolbar{margin-bottom:1em;border-top:none;border-radius:.25rem;border:1px solid rgba(0,0,0,.175)}.quarto-dashboard .vega-embed.has-actions details{width:1.7em;height:2em;position:absolute !important;top:0;right:0}.quarto-dashboard .dashboard-toolbar-container{padding:0}.quarto-dashboard .card .card-header p:last-child,.quarto-dashboard .card .card-footer p:last-child{margin-bottom:0}.quarto-dashboard .card .card-body>.h4:first-child{margin-top:0}.quarto-dashboard .card .card-body{z-index:4}@media(max-width: 767.98px){.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_length,.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_info,.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_paginate{text-align:initial}.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_filter{text-align:right}.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_paginate ul.pagination{justify-content:initial}}.quarto-dashboard .card .card-body .itables .dataTables_wrapper{display:flex;flex-wrap:wrap;justify-content:space-between;align-items:center;padding-top:0}.quarto-dashboard .card .card-body .itables .dataTables_wrapper table{flex-shrink:0}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons{margin-bottom:.5em;margin-left:auto;width:fit-content;float:right}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons.btn-group{background:#fefbf2;border:none}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons .btn-secondary{background-color:#fefbf2;background-image:none;border:solid #dee2e6 1px;padding:.2em .7em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons .btn span{font-size:.8em;color:#343a40}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_info{margin-left:.5em;margin-bottom:.5em;padding-top:0}@media(min-width: 768px){.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_info{font-size:.875em}}@media(max-width: 767.98px){.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_info{font-size:.8em}}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_filter{margin-bottom:.5em;font-size:.875em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_filter input[type=search]{padding:1px 5px 1px 5px;font-size:.875em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_length{flex-basis:1 1 50%;margin-bottom:.5em;font-size:.875em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_length select{padding:.4em 3em .4em .5em;font-size:.875em;margin-left:.2em;margin-right:.2em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_paginate{flex-shrink:0}@media(min-width: 768px){.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_paginate{margin-left:auto}}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_paginate ul.pagination .paginate_button .page-link{font-size:.8em}.quarto-dashboard .card .card-footer{font-size:.9em}.quarto-dashboard .card .card-toolbar{display:flex;flex-grow:1;flex-direction:row;width:100%;flex-wrap:wrap}.quarto-dashboard .card .card-toolbar>*{font-size:.8em;flex-grow:0}.quarto-dashboard .card .card-toolbar>.card-title{font-size:1em;flex-grow:1;align-self:flex-start;margin-top:.1em}.quarto-dashboard .card .card-toolbar .cell-output-display{display:flex}.quarto-dashboard .card .card-toolbar .shiny-input-container{padding-bottom:.5em;margin-bottom:.5em;width:inherit}.quarto-dashboard .card .card-toolbar .shiny-input-container>.checkbox:first-child{margin-top:6px}.quarto-dashboard .card .card-toolbar>*:last-child{margin-right:0}.quarto-dashboard .card .card-toolbar>*>*{margin-right:1em;align-items:baseline}.quarto-dashboard .card .card-toolbar>*>*>a{text-decoration:none;margin-top:auto;margin-bottom:auto}.quarto-dashboard .card .card-toolbar form{width:fit-content}.quarto-dashboard .card .card-toolbar form label{padding-top:.2em;padding-bottom:.2em;width:fit-content}.quarto-dashboard .card .card-toolbar form input[type=date]{width:fit-content}.quarto-dashboard .card .card-toolbar form input[type=color]{width:3em}.quarto-dashboard .card .card-toolbar form button{padding:.4em}.quarto-dashboard .card .card-toolbar form select{width:fit-content}.quarto-dashboard .card .card-toolbar .cell-output-display{display:flex}.quarto-dashboard .card .card-toolbar .shiny-input-container{padding-bottom:.5em;margin-bottom:.5em;width:inherit}.quarto-dashboard .card .card-toolbar .shiny-input-container>.checkbox:first-child{margin-top:6px}.quarto-dashboard .card .card-toolbar>*:last-child{margin-right:0}.quarto-dashboard .card .card-toolbar>*>*{margin-right:1em;align-items:baseline}.quarto-dashboard .card .card-toolbar>*>*>a{text-decoration:none;margin-top:auto;margin-bottom:auto}.quarto-dashboard .card .card-toolbar .shiny-input-container{padding-bottom:0;margin-bottom:0}.quarto-dashboard .card .card-toolbar .shiny-input-container>*{flex-shrink:0;flex-grow:0}.quarto-dashboard .card .card-toolbar .form-group.shiny-input-container:not([role=group])>label{margin-bottom:0}.quarto-dashboard .card .card-toolbar .shiny-input-container.no-baseline{align-items:start;padding-top:6px}.quarto-dashboard .card .card-toolbar .shiny-input-container{display:flex;align-items:baseline}.quarto-dashboard .card .card-toolbar .shiny-input-container label{padding-right:.4em}.quarto-dashboard .card .card-toolbar .shiny-input-container .bslib-input-switch{margin-top:6px}.quarto-dashboard .card .card-toolbar input[type=text]{line-height:1;width:inherit}.quarto-dashboard .card .card-toolbar .input-daterange{width:inherit}.quarto-dashboard .card .card-toolbar .input-daterange input[type=text]{height:2.4em;width:10em}.quarto-dashboard .card .card-toolbar .input-daterange .input-group-addon{height:auto;padding:0;margin-left:-5px !important;margin-right:-5px}.quarto-dashboard .card .card-toolbar .input-daterange .input-group-addon .input-group-text{padding-top:0;padding-bottom:0;height:100%}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny{width:10em}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-line{top:9px}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-min,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-max,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-from,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-to,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-single{top:20px}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-bar{top:8px}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-handle{top:0px}.quarto-dashboard .card .card-toolbar .shiny-input-checkboxgroup>label{margin-top:6px}.quarto-dashboard .card .card-toolbar .shiny-input-checkboxgroup>.shiny-options-group{margin-top:0;align-items:baseline}.quarto-dashboard .card .card-toolbar .shiny-input-radiogroup>label{margin-top:6px}.quarto-dashboard .card .card-toolbar .shiny-input-radiogroup>.shiny-options-group{align-items:baseline;margin-top:0}.quarto-dashboard .card .card-toolbar .shiny-input-radiogroup>.shiny-options-group>.radio{margin-right:.3em}.quarto-dashboard .card .card-toolbar .form-select{padding-top:.2em;padding-bottom:.2em}.quarto-dashboard .card .card-toolbar .shiny-input-select{min-width:6em}.quarto-dashboard .card .card-toolbar div.checkbox{margin-bottom:0px}.quarto-dashboard .card .card-toolbar>.checkbox:first-child{margin-top:6px}.quarto-dashboard .card-body>table>thead{border-top:none}.quarto-dashboard .card-body>.table>:not(caption)>*>*{background-color:#fefbf2}.tableFloatingHeaderOriginal{background-color:#fefbf2;position:sticky !important;top:0 !important}.dashboard-data-table{margin-top:-1px}div.value-box-area span.observablehq--number{font-size:calc(clamp(.1em,15cqw,5em)*1.25);line-height:1.2;color:inherit;font-family:var(--bs-body-font-family)}.quarto-listing{padding-bottom:1em}.listing-pagination{padding-top:.5em}ul.pagination{float:right;padding-left:8px;padding-top:.5em}ul.pagination li{padding-right:.75em}ul.pagination li.disabled a,ul.pagination li.active a{color:#fff;text-decoration:none}ul.pagination li:last-of-type{padding-right:0}.listing-actions-group{display:flex}.quarto-listing-filter{margin-bottom:1em;width:200px;margin-left:auto}.quarto-listing-sort{margin-bottom:1em;margin-right:auto;width:auto}.quarto-listing-sort .input-group-text{font-size:.8em}.input-group-text{border-right:none}.quarto-listing-sort select.form-select{font-size:.8em}.listing-no-matching{text-align:center;padding-top:2em;padding-bottom:3em;font-size:1em}#quarto-margin-sidebar .quarto-listing-category{padding-top:0;font-size:1rem}#quarto-margin-sidebar .quarto-listing-category-title{cursor:pointer;font-weight:600;font-size:1rem}.quarto-listing-category .category{cursor:pointer}.quarto-listing-category .category.active{font-weight:600}.quarto-listing-category.category-cloud{display:flex;flex-wrap:wrap;align-items:baseline}.quarto-listing-category.category-cloud .category{padding-right:5px}.quarto-listing-category.category-cloud .category-cloud-1{font-size:.75em}.quarto-listing-category.category-cloud .category-cloud-2{font-size:.95em}.quarto-listing-category.category-cloud .category-cloud-3{font-size:1.15em}.quarto-listing-category.category-cloud .category-cloud-4{font-size:1.35em}.quarto-listing-category.category-cloud .category-cloud-5{font-size:1.55em}.quarto-listing-category.category-cloud .category-cloud-6{font-size:1.75em}.quarto-listing-category.category-cloud .category-cloud-7{font-size:1.95em}.quarto-listing-category.category-cloud .category-cloud-8{font-size:2.15em}.quarto-listing-category.category-cloud .category-cloud-9{font-size:2.35em}.quarto-listing-category.category-cloud .category-cloud-10{font-size:2.55em}.quarto-listing-cols-1{grid-template-columns:repeat(1, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-1{grid-template-columns:repeat(1, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-1{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-2{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-2{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-2{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-3{grid-template-columns:repeat(3, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-3{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-3{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-4{grid-template-columns:repeat(4, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-4{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-4{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-5{grid-template-columns:repeat(5, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-5{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-5{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-6{grid-template-columns:repeat(6, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-6{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-6{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-7{grid-template-columns:repeat(7, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-7{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-7{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-8{grid-template-columns:repeat(8, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-8{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-8{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-9{grid-template-columns:repeat(9, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-9{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-9{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-10{grid-template-columns:repeat(10, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-10{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-10{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-11{grid-template-columns:repeat(11, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-11{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-11{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-12{grid-template-columns:repeat(12, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-12{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-12{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-grid{gap:1.5em}.quarto-grid-item.borderless{border:none}.quarto-grid-item.borderless .listing-categories .listing-category:last-of-type,.quarto-grid-item.borderless .listing-categories .listing-category:first-of-type{padding-left:0}.quarto-grid-item.borderless .listing-categories .listing-category{border:0}.quarto-grid-link{text-decoration:none;color:inherit}.quarto-grid-link:hover{text-decoration:none;color:inherit}.quarto-grid-item h5.title,.quarto-grid-item .title.h5{margin-top:0;margin-bottom:0}.quarto-grid-item .card-footer{display:flex;justify-content:space-between;font-size:.8em}.quarto-grid-item .card-footer p{margin-bottom:0}.quarto-grid-item p.card-img-top{margin-bottom:0}.quarto-grid-item p.card-img-top>img{object-fit:cover}.quarto-grid-item .card-other-values{margin-top:.5em;font-size:.8em}.quarto-grid-item .card-other-values tr{margin-bottom:.5em}.quarto-grid-item .card-other-values tr>td:first-of-type{font-weight:600;padding-right:1em;padding-left:1em;vertical-align:top}.quarto-grid-item div.post-contents{display:flex;flex-direction:column;text-decoration:none;height:100%}.quarto-grid-item .listing-item-img-placeholder{background-color:rgba(52,58,64,.25);flex-shrink:0}.quarto-grid-item .card-attribution{padding-top:1em;display:flex;gap:1em;text-transform:uppercase;color:#6c757d;font-weight:500;flex-grow:10;align-items:flex-end}.quarto-grid-item .description{padding-bottom:1em}.quarto-grid-item .card-attribution .date{align-self:flex-end}.quarto-grid-item .card-attribution.justify{justify-content:space-between}.quarto-grid-item .card-attribution.start{justify-content:flex-start}.quarto-grid-item .card-attribution.end{justify-content:flex-end}.quarto-grid-item .card-title{margin-bottom:.1em}.quarto-grid-item .card-subtitle{padding-top:.25em}.quarto-grid-item .card-text{font-size:.9em}.quarto-grid-item .listing-reading-time{padding-bottom:.25em}.quarto-grid-item .card-text-small{font-size:.8em}.quarto-grid-item .card-subtitle.subtitle{font-size:.9em;font-weight:600;padding-bottom:.5em}.quarto-grid-item .listing-categories{display:flex;flex-wrap:wrap;padding-bottom:5px}.quarto-grid-item .listing-categories .listing-category{color:#6c757d;border:solid 1px #dee2e6;border-radius:.25rem;text-transform:uppercase;font-size:.65em;padding-left:.5em;padding-right:.5em;padding-top:.15em;padding-bottom:.15em;cursor:pointer;margin-right:4px;margin-bottom:4px}.quarto-grid-item.card-right{text-align:right}.quarto-grid-item.card-right .listing-categories{justify-content:flex-end}.quarto-grid-item.card-left{text-align:left}.quarto-grid-item.card-center{text-align:center}.quarto-grid-item.card-center .listing-description{text-align:justify}.quarto-grid-item.card-center .listing-categories{justify-content:center}table.quarto-listing-table td.image{padding:0px}table.quarto-listing-table td.image img{width:100%;max-width:50px;object-fit:contain}table.quarto-listing-table a{text-decoration:none;word-break:keep-all}table.quarto-listing-table th a{color:inherit}table.quarto-listing-table th a.asc:after{margin-bottom:-2px;margin-left:5px;display:inline-block;height:1rem;width:1rem;background-repeat:no-repeat;background-size:1rem 1rem;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-sort-up" viewBox="0 0 16 16"><path d="M3.5 12.5a.5.5 0 0 1-1 0V3.707L1.354 4.854a.5.5 0 1 1-.708-.708l2-1.999.007-.007a.498.498 0 0 1 .7.006l2 2a.5.5 0 1 1-.707.708L3.5 3.707V12.5zm3.5-9a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5zM7.5 6a.5.5 0 0 0 0 1h5a.5.5 0 0 0 0-1h-5zm0 3a.5.5 0 0 0 0 1h3a.5.5 0 0 0 0-1h-3zm0 3a.5.5 0 0 0 0 1h1a.5.5 0 0 0 0-1h-1z"/></svg>');content:""}table.quarto-listing-table th a.desc:after{margin-bottom:-2px;margin-left:5px;display:inline-block;height:1rem;width:1rem;background-repeat:no-repeat;background-size:1rem 1rem;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-sort-down" viewBox="0 0 16 16"><path d="M3.5 2.5a.5.5 0 0 0-1 0v8.793l-1.146-1.147a.5.5 0 0 0-.708.708l2 1.999.007.007a.497.497 0 0 0 .7-.006l2-2a.5.5 0 0 0-.707-.708L3.5 11.293V2.5zm3.5 1a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5zM7.5 6a.5.5 0 0 0 0 1h5a.5.5 0 0 0 0-1h-5zm0 3a.5.5 0 0 0 0 1h3a.5.5 0 0 0 0-1h-3zm0 3a.5.5 0 0 0 0 1h1a.5.5 0 0 0 0-1h-1z"/></svg>');content:""}table.quarto-listing-table.table-hover td{cursor:pointer}.quarto-post.image-left{flex-direction:row}.quarto-post.image-right{flex-direction:row-reverse}@media(max-width: 767.98px){.quarto-post.image-right,.quarto-post.image-left{gap:0em;flex-direction:column}.quarto-post .metadata{padding-bottom:1em;order:2}.quarto-post .body{order:1}.quarto-post .thumbnail{order:3}}.list.quarto-listing-default div:last-of-type{border-bottom:none}@media(min-width: 992px){.quarto-listing-container-default{margin-right:2em}}div.quarto-post{display:flex;gap:2em;margin-bottom:1.5em;border-bottom:1px solid #dee2e6}@media(max-width: 767.98px){div.quarto-post{padding-bottom:1em}}div.quarto-post .metadata{flex-basis:20%;flex-grow:0;margin-top:.2em;flex-shrink:10}div.quarto-post .thumbnail{flex-basis:30%;flex-grow:0;flex-shrink:0}div.quarto-post .thumbnail img{margin-top:.4em;width:100%;object-fit:cover}div.quarto-post .body{flex-basis:45%;flex-grow:1;flex-shrink:0}div.quarto-post .body h3.listing-title,div.quarto-post .body .listing-title.h3{margin-top:0px;margin-bottom:0px;border-bottom:none}div.quarto-post .body .listing-subtitle{font-size:.875em;margin-bottom:.5em;margin-top:.2em}div.quarto-post .body .description{font-size:.9em}div.quarto-post .body pre code{white-space:pre-wrap}div.quarto-post a{color:#343a40;text-decoration:none}div.quarto-post .metadata{display:flex;flex-direction:column;font-size:.8em;font-family:"Source Sans Pro",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,sans-serif,"Apple Color Emoji","Segoe UI Emoji","Segoe UI Symbol";flex-basis:33%}div.quarto-post .listing-categories{display:flex;flex-wrap:wrap;padding-bottom:5px}div.quarto-post .listing-categories .listing-category{color:#6c757d;border:solid 1px #dee2e6;border-radius:.25rem;text-transform:uppercase;font-size:.65em;padding-left:.5em;padding-right:.5em;padding-top:.15em;padding-bottom:.15em;cursor:pointer;margin-right:4px;margin-bottom:4px}div.quarto-post .listing-description{margin-bottom:.5em}div.quarto-about-jolla{display:flex !important;flex-direction:column;align-items:center;margin-top:10%;padding-bottom:1em}div.quarto-about-jolla .about-image{object-fit:cover;margin-left:auto;margin-right:auto;margin-bottom:1.5em}div.quarto-about-jolla img.round{border-radius:50%}div.quarto-about-jolla img.rounded{border-radius:10px}div.quarto-about-jolla .quarto-title h1.title,div.quarto-about-jolla .quarto-title .title.h1{text-align:center}div.quarto-about-jolla .quarto-title .description{text-align:center}div.quarto-about-jolla h2,div.quarto-about-jolla .h2{border-bottom:none}div.quarto-about-jolla .about-sep{width:60%}div.quarto-about-jolla main{text-align:center}div.quarto-about-jolla .about-links{display:flex}@media(min-width: 992px){div.quarto-about-jolla .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-jolla .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-jolla .about-link{color:#626d78;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-jolla .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-jolla .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-jolla .about-link:hover{color:#2761e3}div.quarto-about-jolla .about-link i.bi{margin-right:.15em}div.quarto-about-solana{display:flex !important;flex-direction:column;padding-top:3em !important;padding-bottom:1em}div.quarto-about-solana .about-entity{display:flex !important;align-items:start;justify-content:space-between}@media(min-width: 992px){div.quarto-about-solana .about-entity{flex-direction:row}}@media(max-width: 991.98px){div.quarto-about-solana .about-entity{flex-direction:column-reverse;align-items:center;text-align:center}}div.quarto-about-solana .about-entity .entity-contents{display:flex;flex-direction:column}@media(max-width: 767.98px){div.quarto-about-solana .about-entity .entity-contents{width:100%}}div.quarto-about-solana .about-entity .about-image{object-fit:cover}@media(max-width: 991.98px){div.quarto-about-solana .about-entity .about-image{margin-bottom:1.5em}}div.quarto-about-solana .about-entity img.round{border-radius:50%}div.quarto-about-solana .about-entity img.rounded{border-radius:10px}div.quarto-about-solana .about-entity .about-links{display:flex;justify-content:left;padding-bottom:1.2em}@media(min-width: 992px){div.quarto-about-solana .about-entity .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-solana .about-entity .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-solana .about-entity .about-link{color:#626d78;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-solana .about-entity .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-solana .about-entity .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-solana .about-entity .about-link:hover{color:#2761e3}div.quarto-about-solana .about-entity .about-link i.bi{margin-right:.15em}div.quarto-about-solana .about-contents{padding-right:1.5em;flex-basis:0;flex-grow:1}div.quarto-about-solana .about-contents main.content{margin-top:0}div.quarto-about-solana .about-contents h2,div.quarto-about-solana .about-contents .h2{border-bottom:none}div.quarto-about-trestles{display:flex !important;flex-direction:row;padding-top:3em !important;padding-bottom:1em}@media(max-width: 991.98px){div.quarto-about-trestles{flex-direction:column;padding-top:0em !important}}div.quarto-about-trestles .about-entity{display:flex !important;flex-direction:column;align-items:center;text-align:center;padding-right:1em}@media(min-width: 992px){div.quarto-about-trestles .about-entity{flex:0 0 42%}}div.quarto-about-trestles .about-entity .about-image{object-fit:cover;margin-bottom:1.5em}div.quarto-about-trestles .about-entity img.round{border-radius:50%}div.quarto-about-trestles .about-entity img.rounded{border-radius:10px}div.quarto-about-trestles .about-entity .about-links{display:flex;justify-content:center}@media(min-width: 992px){div.quarto-about-trestles .about-entity .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-trestles .about-entity .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-trestles .about-entity .about-link{color:#626d78;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-trestles .about-entity .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-trestles .about-entity .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-trestles .about-entity .about-link:hover{color:#2761e3}div.quarto-about-trestles .about-entity .about-link i.bi{margin-right:.15em}div.quarto-about-trestles .about-contents{flex-basis:0;flex-grow:1}div.quarto-about-trestles .about-contents h2,div.quarto-about-trestles .about-contents .h2{border-bottom:none}@media(min-width: 992px){div.quarto-about-trestles .about-contents{border-left:solid 1px #dee2e6;padding-left:1.5em}}div.quarto-about-trestles .about-contents main.content{margin-top:0}div.quarto-about-marquee{padding-bottom:1em}div.quarto-about-marquee .about-contents{display:flex;flex-direction:column}div.quarto-about-marquee .about-image{max-height:550px;margin-bottom:1.5em;object-fit:cover}div.quarto-about-marquee img.round{border-radius:50%}div.quarto-about-marquee img.rounded{border-radius:10px}div.quarto-about-marquee h2,div.quarto-about-marquee .h2{border-bottom:none}div.quarto-about-marquee .about-links{display:flex;justify-content:center;padding-top:1.5em}@media(min-width: 992px){div.quarto-about-marquee .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-marquee .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-marquee .about-link{color:#626d78;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-marquee .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-marquee .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-marquee .about-link:hover{color:#2761e3}div.quarto-about-marquee .about-link i.bi{margin-right:.15em}@media(min-width: 992px){div.quarto-about-marquee .about-link{border:none}}div.quarto-about-broadside{display:flex;flex-direction:column;padding-bottom:1em}div.quarto-about-broadside .about-main{display:flex !important;padding-top:0 !important}@media(min-width: 992px){div.quarto-about-broadside .about-main{flex-direction:row;align-items:flex-start}}@media(max-width: 991.98px){div.quarto-about-broadside .about-main{flex-direction:column}}@media(max-width: 991.98px){div.quarto-about-broadside .about-main .about-entity{flex-shrink:0;width:100%;height:450px;margin-bottom:1.5em;background-size:cover;background-repeat:no-repeat}}@media(min-width: 992px){div.quarto-about-broadside .about-main .about-entity{flex:0 10 50%;margin-right:1.5em;width:100%;height:100%;background-size:100%;background-repeat:no-repeat}}div.quarto-about-broadside .about-main .about-contents{padding-top:14px;flex:0 0 50%}div.quarto-about-broadside h2,div.quarto-about-broadside .h2{border-bottom:none}div.quarto-about-broadside .about-sep{margin-top:1.5em;width:60%;align-self:center}div.quarto-about-broadside .about-links{display:flex;justify-content:center;column-gap:20px;padding-top:1.5em}@media(min-width: 992px){div.quarto-about-broadside .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-broadside .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-broadside .about-link{color:#626d78;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-broadside .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-broadside .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-broadside .about-link:hover{color:#2761e3}div.quarto-about-broadside .about-link i.bi{margin-right:.15em}@media(min-width: 992px){div.quarto-about-broadside .about-link{border:none}}.tippy-box[data-theme~=quarto]{background-color:#fefbf2;border:solid 1px #dee2e6;border-radius:.25rem;color:#343a40;font-size:.875rem}.tippy-box[data-theme~=quarto]>.tippy-backdrop{background-color:#fefbf2}.tippy-box[data-theme~=quarto]>.tippy-arrow:after,.tippy-box[data-theme~=quarto]>.tippy-svg-arrow:after{content:"";position:absolute;z-index:-1}.tippy-box[data-theme~=quarto]>.tippy-arrow:after{border-color:rgba(0,0,0,0);border-style:solid}.tippy-box[data-placement^=top]>.tippy-arrow:before{bottom:-6px}.tippy-box[data-placement^=bottom]>.tippy-arrow:before{top:-6px}.tippy-box[data-placement^=right]>.tippy-arrow:before{left:-6px}.tippy-box[data-placement^=left]>.tippy-arrow:before{right:-6px}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-arrow:before{border-top-color:#fefbf2}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-arrow:after{border-top-color:#dee2e6;border-width:7px 7px 0;top:17px;left:1px}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-svg-arrow>svg{top:16px}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-svg-arrow:after{top:17px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-arrow:before{border-bottom-color:#fefbf2;bottom:16px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-arrow:after{border-bottom-color:#dee2e6;border-width:0 7px 7px;bottom:17px;left:1px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-svg-arrow>svg{bottom:15px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-svg-arrow:after{bottom:17px}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-arrow:before{border-left-color:#fefbf2}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-arrow:after{border-left-color:#dee2e6;border-width:7px 0 7px 7px;left:17px;top:1px}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-svg-arrow>svg{left:11px}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-svg-arrow:after{left:12px}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-arrow:before{border-right-color:#fefbf2;right:16px}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-arrow:after{border-width:7px 7px 7px 0;right:17px;top:1px;border-right-color:#dee2e6}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-svg-arrow>svg{right:11px}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-svg-arrow:after{right:12px}.tippy-box[data-theme~=quarto]>.tippy-svg-arrow{fill:#343a40}.tippy-box[data-theme~=quarto]>.tippy-svg-arrow:after{background-image:url(data:image/svg+xml;base64,PHN2ZyB3aWR0aD0iMTYiIGhlaWdodD0iNiIgeG1sbnM9Imh0dHA6Ly93d3cudzMub3JnLzIwMDAvc3ZnIj48cGF0aCBkPSJNMCA2czEuNzk2LS4wMTMgNC42Ny0zLjYxNUM1Ljg1MS45IDYuOTMuMDA2IDggMGMxLjA3LS4wMDYgMi4xNDguODg3IDMuMzQzIDIuMzg1QzE0LjIzMyA2LjAwNSAxNiA2IDE2IDZIMHoiIGZpbGw9InJnYmEoMCwgOCwgMTYsIDAuMikiLz48L3N2Zz4=);background-size:16px 6px;width:16px;height:6px}.top-right{position:absolute;top:1em;right:1em}.visually-hidden{border:0;clip:rect(0 0 0 0);height:auto;margin:0;overflow:hidden;padding:0;position:absolute;width:1px;white-space:nowrap}.hidden{display:none !important}.zindex-bottom{z-index:-1 !important}figure.figure{display:block}.quarto-layout-panel{margin-bottom:1em}.quarto-layout-panel>figure{width:100%}.quarto-layout-panel>figure>figcaption,.quarto-layout-panel>.panel-caption{margin-top:10pt}.quarto-layout-panel>.table-caption{margin-top:0px}.table-caption p{margin-bottom:.5em}.quarto-layout-row{display:flex;flex-direction:row;align-items:flex-start}.quarto-layout-valign-top{align-items:flex-start}.quarto-layout-valign-bottom{align-items:flex-end}.quarto-layout-valign-center{align-items:center}.quarto-layout-cell{position:relative;margin-right:20px}.quarto-layout-cell:last-child{margin-right:0}.quarto-layout-cell figure,.quarto-layout-cell>p{margin:.2em}.quarto-layout-cell img{max-width:100%}.quarto-layout-cell .html-widget{width:100% !important}.quarto-layout-cell div figure p{margin:0}.quarto-layout-cell figure{display:block;margin-inline-start:0;margin-inline-end:0}.quarto-layout-cell table{display:inline-table}.quarto-layout-cell-subref figcaption,figure .quarto-layout-row figure figcaption{text-align:center;font-style:italic}.quarto-figure{position:relative;margin-bottom:1em}.quarto-figure>figure{width:100%;margin-bottom:0}.quarto-figure-left>figure>p,.quarto-figure-left>figure>div{text-align:left}.quarto-figure-center>figure>p,.quarto-figure-center>figure>div{text-align:center}.quarto-figure-right>figure>p,.quarto-figure-right>figure>div{text-align:right}.quarto-figure>figure>div.cell-annotation,.quarto-figure>figure>div code{text-align:left}figure>p:empty{display:none}figure>p:first-child{margin-top:0;margin-bottom:0}figure>figcaption.quarto-float-caption-bottom{margin-bottom:.5em}figure>figcaption.quarto-float-caption-top{margin-top:.5em}div[id^=tbl-]{position:relative}.quarto-figure>.anchorjs-link{position:absolute;top:.6em;right:.5em}div[id^=tbl-]>.anchorjs-link{position:absolute;top:.7em;right:.3em}.quarto-figure:hover>.anchorjs-link,div[id^=tbl-]:hover>.anchorjs-link,h2:hover>.anchorjs-link,.h2:hover>.anchorjs-link,h3:hover>.anchorjs-link,.h3:hover>.anchorjs-link,h4:hover>.anchorjs-link,.h4:hover>.anchorjs-link,h5:hover>.anchorjs-link,.h5:hover>.anchorjs-link,h6:hover>.anchorjs-link,.h6:hover>.anchorjs-link,.reveal-anchorjs-link>.anchorjs-link{opacity:1}#title-block-header{margin-block-end:1rem;position:relative;margin-top:-1px}#title-block-header .abstract{margin-block-start:1rem}#title-block-header .abstract .abstract-title{font-weight:600}#title-block-header a{text-decoration:none}#title-block-header .author,#title-block-header .date,#title-block-header .doi{margin-block-end:.2rem}#title-block-header .quarto-title-block>div{display:flex}#title-block-header .quarto-title-block>div>h1,#title-block-header .quarto-title-block>div>.h1{flex-grow:1}#title-block-header .quarto-title-block>div>button{flex-shrink:0;height:2.25rem;margin-top:0}@media(min-width: 992px){#title-block-header .quarto-title-block>div>button{margin-top:5px}}tr.header>th>p:last-of-type{margin-bottom:0px}table,table.table{margin-top:.5rem;margin-bottom:.5rem}caption,.table-caption{padding-top:.5rem;padding-bottom:.5rem;text-align:center}figure.quarto-float-tbl figcaption.quarto-float-caption-top{margin-top:.5rem;margin-bottom:.25rem;text-align:center}figure.quarto-float-tbl figcaption.quarto-float-caption-bottom{padding-top:.25rem;margin-bottom:.5rem;text-align:center}.utterances{max-width:none;margin-left:-8px}iframe{margin-bottom:1em}details{margin-bottom:1em}details[show]{margin-bottom:0}details>summary{color:#6c757d}details>summary>p:only-child{display:inline}pre.sourceCode,code.sourceCode{position:relative}dd code:not(.sourceCode),p code:not(.sourceCode){white-space:pre-wrap}code{white-space:pre}@media print{code{white-space:pre-wrap}}pre>code{display:block}pre>code.sourceCode{white-space:pre-wrap}pre>code.sourceCode>span>a:first-child::before{text-decoration:none}pre.code-overflow-wrap>code.sourceCode{white-space:pre-wrap}pre.code-overflow-scroll>code.sourceCode{white-space:pre}code a:any-link{color:inherit;text-decoration:none}code a:hover{color:inherit;text-decoration:underline}ul.task-list{padding-left:1em}[data-tippy-root]{display:inline-block}.tippy-content .footnote-back{display:none}.footnote-back{margin-left:.2em}.tippy-content{overflow-x:auto}.quarto-embedded-source-code{display:none}.quarto-unresolved-ref{font-weight:600}.quarto-cover-image{max-width:35%;float:right;margin-left:30px}.cell-output-display .widget-subarea{margin-bottom:1em}.cell-output-display:not(.no-overflow-x),.knitsql-table:not(.no-overflow-x){overflow-x:auto}.panel-input{margin-bottom:1em}.panel-input>div,.panel-input>div>div{display:inline-block;vertical-align:top;padding-right:12px}.panel-input>p:last-child{margin-bottom:0}.layout-sidebar{margin-bottom:1em}.layout-sidebar .tab-content{border:none}.tab-content>.page-columns.active{display:grid}div.sourceCode>iframe{width:100%;height:300px;margin-bottom:-0.5em}a{text-underline-offset:3px}.callout pre.sourceCode{padding-left:0}div.ansi-escaped-output{font-family:monospace;display:block}/*!
+*
+* ansi colors from IPython notebook's
+*
+* we also add `bright-[color]-` synonyms for the `-[color]-intense` classes since
+* that seems to be what ansi_up emits
+*
+*/.ansi-black-fg{color:#3e424d}.ansi-black-bg{background-color:#3e424d}.ansi-black-intense-black,.ansi-bright-black-fg{color:#282c36}.ansi-black-intense-black,.ansi-bright-black-bg{background-color:#282c36}.ansi-red-fg{color:#e75c58}.ansi-red-bg{background-color:#e75c58}.ansi-red-intense-red,.ansi-bright-red-fg{color:#b22b31}.ansi-red-intense-red,.ansi-bright-red-bg{background-color:#b22b31}.ansi-green-fg{color:#00a250}.ansi-green-bg{background-color:#00a250}.ansi-green-intense-green,.ansi-bright-green-fg{color:#007427}.ansi-green-intense-green,.ansi-bright-green-bg{background-color:#007427}.ansi-yellow-fg{color:#ddb62b}.ansi-yellow-bg{background-color:#ddb62b}.ansi-yellow-intense-yellow,.ansi-bright-yellow-fg{color:#b27d12}.ansi-yellow-intense-yellow,.ansi-bright-yellow-bg{background-color:#b27d12}.ansi-blue-fg{color:#208ffb}.ansi-blue-bg{background-color:#208ffb}.ansi-blue-intense-blue,.ansi-bright-blue-fg{color:#0065ca}.ansi-blue-intense-blue,.ansi-bright-blue-bg{background-color:#0065ca}.ansi-magenta-fg{color:#d160c4}.ansi-magenta-bg{background-color:#d160c4}.ansi-magenta-intense-magenta,.ansi-bright-magenta-fg{color:#a03196}.ansi-magenta-intense-magenta,.ansi-bright-magenta-bg{background-color:#a03196}.ansi-cyan-fg{color:#60c6c8}.ansi-cyan-bg{background-color:#60c6c8}.ansi-cyan-intense-cyan,.ansi-bright-cyan-fg{color:#258f8f}.ansi-cyan-intense-cyan,.ansi-bright-cyan-bg{background-color:#258f8f}.ansi-white-fg{color:#c5c1b4}.ansi-white-bg{background-color:#c5c1b4}.ansi-white-intense-white,.ansi-bright-white-fg{color:#a1a6b2}.ansi-white-intense-white,.ansi-bright-white-bg{background-color:#a1a6b2}.ansi-default-inverse-fg{color:#fff}.ansi-default-inverse-bg{background-color:#000}.ansi-bold{font-weight:bold}.ansi-underline{text-decoration:underline}:root{--quarto-body-bg: #FEFBF2;--quarto-body-color: #343a40;--quarto-text-muted: #6c757d;--quarto-border-color: #dee2e6;--quarto-border-width: 1px}table.gt_table{color:var(--quarto-body-color);font-size:1em;width:100%;background-color:rgba(0,0,0,0);border-top-width:inherit;border-bottom-width:inherit;border-color:var(--quarto-border-color)}table.gt_table th.gt_column_spanner_outer{color:var(--quarto-body-color);background-color:rgba(0,0,0,0);border-top-width:inherit;border-bottom-width:inherit;border-color:var(--quarto-border-color)}table.gt_table th.gt_col_heading{color:var(--quarto-body-color);font-weight:bold;background-color:rgba(0,0,0,0)}table.gt_table thead.gt_col_headings{border-bottom:1px solid currentColor;border-top-width:inherit;border-top-color:var(--quarto-border-color)}table.gt_table thead.gt_col_headings:not(:first-child){border-top-width:1px;border-top-color:var(--quarto-border-color)}table.gt_table td.gt_row{border-bottom-width:1px;border-bottom-color:var(--quarto-border-color);border-top-width:0px}table.gt_table tbody.gt_table_body{border-top-width:1px;border-bottom-width:1px;border-bottom-color:var(--quarto-border-color);border-top-color:currentColor}div.columns{display:initial;gap:initial}div.column{display:inline-block;overflow-x:initial;vertical-align:top;width:50%}.code-annotation-tip-content{word-wrap:break-word}.code-annotation-container-hidden{display:none !important}dl.code-annotation-container-grid{display:grid;grid-template-columns:min-content auto}dl.code-annotation-container-grid dt{grid-column:1}dl.code-annotation-container-grid dd{grid-column:2}pre.sourceCode.code-annotation-code{padding-right:0}code.sourceCode .code-annotation-anchor{z-index:100;position:relative;float:right;background-color:rgba(0,0,0,0)}input[type=checkbox]{margin-right:.5ch}:root{--mermaid-bg-color: #FEFBF2;--mermaid-edge-color: #343a40;--mermaid-node-fg-color: #343a40;--mermaid-fg-color: #343a40;--mermaid-fg-color--lighter: #4b545c;--mermaid-fg-color--lightest: #626d78;--mermaid-font-family: Source Sans Pro, -apple-system, BlinkMacSystemFont, Segoe UI, Roboto, Helvetica Neue, Arial, sans-serif, Apple Color Emoji, Segoe UI Emoji, Segoe UI Symbol;--mermaid-label-bg-color: #FEFBF2;--mermaid-label-fg-color: #2780e3;--mermaid-node-bg-color: rgba(39, 128, 227, 0.1);--mermaid-node-fg-color: #343a40}@media print{:root{font-size:11pt}#quarto-sidebar,#TOC,.nav-page{display:none}.page-columns .content{grid-column-start:page-start}.fixed-top{position:relative}.panel-caption,.figure-caption,figcaption{color:#666}}.code-copy-button{position:absolute;top:0;right:0;border:0;margin-top:5px;margin-right:5px;background-color:rgba(0,0,0,0);z-index:3}.code-copy-button:focus{outline:none}.code-copy-button-tooltip{font-size:.75em}.code-copy-button>.bi::before{display:inline-block;height:1rem;width:1rem;content:"";vertical-align:-0.125em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(143, 89, 2)" viewBox="0 0 16 16"><path d="M4 1.5H3a2 2 0 0 0-2 2V14a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V3.5a2 2 0 0 0-2-2h-1v1h1a1 1 0 0 1 1 1V14a1 1 0 0 1-1 1H3a1 1 0 0 1-1-1V3.5a1 1 0 0 1 1-1h1v-1z"/><path d="M9.5 1a.5.5 0 0 1 .5.5v1a.5.5 0 0 1-.5.5h-3a.5.5 0 0 1-.5-.5v-1a.5.5 0 0 1 .5-.5h3zm-3-1A1.5 1.5 0 0 0 5 1.5v1A1.5 1.5 0 0 0 6.5 4h3A1.5 1.5 0 0 0 11 2.5v-1A1.5 1.5 0 0 0 9.5 0h-3z"/></svg>');background-repeat:no-repeat;background-size:1rem 1rem}.code-copy-button-checked>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(143, 89, 2)" viewBox="0 0 16 16"><path d="M13.854 3.646a.5.5 0 0 1 0 .708l-7 7a.5.5 0 0 1-.708 0l-3.5-3.5a.5.5 0 1 1 .708-.708L6.5 10.293l6.646-6.647a.5.5 0 0 1 .708 0z"/></svg>')}.code-copy-button:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(32, 74, 135)" viewBox="0 0 16 16"><path d="M4 1.5H3a2 2 0 0 0-2 2V14a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V3.5a2 2 0 0 0-2-2h-1v1h1a1 1 0 0 1 1 1V14a1 1 0 0 1-1 1H3a1 1 0 0 1-1-1V3.5a1 1 0 0 1 1-1h1v-1z"/><path d="M9.5 1a.5.5 0 0 1 .5.5v1a.5.5 0 0 1-.5.5h-3a.5.5 0 0 1-.5-.5v-1a.5.5 0 0 1 .5-.5h3zm-3-1A1.5 1.5 0 0 0 5 1.5v1A1.5 1.5 0 0 0 6.5 4h3A1.5 1.5 0 0 0 11 2.5v-1A1.5 1.5 0 0 0 9.5 0h-3z"/></svg>')}.code-copy-button-checked:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(32, 74, 135)"  viewBox="0 0 16 16"><path d="M13.854 3.646a.5.5 0 0 1 0 .708l-7 7a.5.5 0 0 1-.708 0l-3.5-3.5a.5.5 0 1 1 .708-.708L6.5 10.293l6.646-6.647a.5.5 0 0 1 .708 0z"/></svg>')}main ol ol,main ul ul,main ol ul,main ul ol{margin-bottom:1em}ul>li:not(:has(>p))>ul,ol>li:not(:has(>p))>ul,ul>li:not(:has(>p))>ol,ol>li:not(:has(>p))>ol{margin-bottom:0}ul>li:not(:has(>p))>ul>li:has(>p),ol>li:not(:has(>p))>ul>li:has(>p),ul>li:not(:has(>p))>ol>li:has(>p),ol>li:not(:has(>p))>ol>li:has(>p){margin-top:1rem}body{margin:0}main.page-columns>header>h1.title,main.page-columns>header>.title.h1{margin-bottom:0}@media(min-width: 992px){body .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset] 35px [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.fullcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset] 35px [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] 42px [page-end-inset page-end] 5fr [screen-end-inset] 1.5em}body.slimcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset] 35px [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 35px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.listing:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 3em [body-end] 50px [body-end-outset] minmax(0px, 300px) [page-end-inset] minmax(50px, 100px) [page-end] 1fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 175px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 175px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] minmax(25px, 50px) [page-start-inset] minmax(50px, 150px) [body-start-outset] minmax(25px, 50px) [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] minmax(30px, 60px) [body-end-outset] minmax(60px, 180px) [page-end-inset] minmax(30px, 60px) [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(60px, 120px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 50px [page-start-inset] minmax(50px, 150px) [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(450px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 50px [page-start-inset] minmax(50px, 150px) [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(450px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(60px, 180px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] minmax(25px, 50px) [page-start-inset] minmax(50px, 150px) [body-start-outset] minmax(25px, 50px) [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] minmax(30px, 60px) [body-end-outset] minmax(60px, 180px) [page-end-inset] minmax(30px, 60px) [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}}@media(max-width: 991.98px){body .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.fullcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.slimcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.listing:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(1350px - 3em)) [body-content-end body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 145px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 145px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1.5em [body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(90px, 180px) [page-end-inset] 30px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(30px, 60px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 4fr [screen-end-inset] 1.5em [screen-end]}body.docked.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(30px, 60px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(30px, 60px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1em [body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 4fr [screen-end-inset] 1.5em [screen-end]}body.floating.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1em [body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(90px, 180px) [page-end-inset] 30px [page-end] 4fr [screen-end-inset] 1.5em [screen-end]}}@media(max-width: 767.98px){body .page-columns,body.fullcontent:not(.floating):not(.docked) .page-columns,body.slimcontent:not(.floating):not(.docked) .page-columns,body.docked .page-columns,body.docked.slimcontent .page-columns,body.docked.fullcontent .page-columns,body.floating .page-columns,body.floating.slimcontent .page-columns,body.floating.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(0px, 1fr) [body-content-end body-end body-end-outset page-end-inset page-end screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(0px, 1fr) [body-content-end body-end body-end-outset page-end-inset page-end screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(0px, 1fr) [body-content-end body-end body-end-outset page-end-inset page-end screen-end-inset] 1.5em [screen-end]}nav[role=doc-toc]{display:none}}body,.page-row-navigation{grid-template-rows:[page-top] max-content [contents-top] max-content [contents-bottom] max-content [page-bottom]}.page-rows-contents{grid-template-rows:[content-top] minmax(max-content, 1fr) [content-bottom] minmax(60px, max-content) [page-bottom]}.page-full{grid-column:screen-start/screen-end !important}.page-columns>*{grid-column:body-content-start/body-content-end}.page-columns.column-page>*{grid-column:page-start/page-end}.page-columns.column-page-left .page-columns.page-full>*,.page-columns.column-page-left>*{grid-column:page-start/body-content-end}.page-columns.column-page-right .page-columns.page-full>*,.page-columns.column-page-right>*{grid-column:body-content-start/page-end}.page-rows{grid-auto-rows:auto}.header{grid-column:screen-start/screen-end;grid-row:page-top/contents-top}#quarto-content{padding:0;grid-column:screen-start/screen-end;grid-row:contents-top/contents-bottom}body.floating .sidebar.sidebar-navigation{grid-column:page-start/body-start;grid-row:content-top/page-bottom}body.docked .sidebar.sidebar-navigation{grid-column:screen-start/body-start;grid-row:content-top/page-bottom}.sidebar.toc-left{grid-column:page-start/body-start;grid-row:content-top/page-bottom}.sidebar.margin-sidebar{grid-column:body-end/page-end;grid-row:content-top/page-bottom}.page-columns .content{grid-column:body-content-start/body-content-end;grid-row:content-top/content-bottom;align-content:flex-start}.page-columns .page-navigation{grid-column:body-content-start/body-content-end;grid-row:content-bottom/page-bottom}.page-columns .footer{grid-column:screen-start/screen-end;grid-row:contents-bottom/page-bottom}.page-columns .column-body{grid-column:body-content-start/body-content-end}.page-columns .column-body-fullbleed{grid-column:body-start/body-end}.page-columns .column-body-outset{grid-column:body-start-outset/body-end-outset;z-index:998;opacity:.999}.page-columns .column-body-outset table{background:#fefbf2}.page-columns .column-body-outset-left{grid-column:body-start-outset/body-content-end;z-index:998;opacity:.999}.page-columns .column-body-outset-left table{background:#fefbf2}.page-columns .column-body-outset-right{grid-column:body-content-start/body-end-outset;z-index:998;opacity:.999}.page-columns .column-body-outset-right table{background:#fefbf2}.page-columns .column-page{grid-column:page-start/page-end;z-index:998;opacity:.999}.page-columns .column-page table{background:#fefbf2}.page-columns .column-page-inset{grid-column:page-start-inset/page-end-inset;z-index:998;opacity:.999}.page-columns .column-page-inset table{background:#fefbf2}.page-columns .column-page-inset-left{grid-column:page-start-inset/body-content-end;z-index:998;opacity:.999}.page-columns .column-page-inset-left table{background:#fefbf2}.page-columns .column-page-inset-right{grid-column:body-content-start/page-end-inset;z-index:998;opacity:.999}.page-columns .column-page-inset-right figcaption table{background:#fefbf2}.page-columns .column-page-left{grid-column:page-start/body-content-end;z-index:998;opacity:.999}.page-columns .column-page-left table{background:#fefbf2}.page-columns .column-page-right{grid-column:body-content-start/page-end;z-index:998;opacity:.999}.page-columns .column-page-right figcaption table{background:#fefbf2}#quarto-content.page-columns #quarto-margin-sidebar,#quarto-content.page-columns #quarto-sidebar{z-index:1}@media(max-width: 991.98px){#quarto-content.page-columns #quarto-margin-sidebar.collapse,#quarto-content.page-columns #quarto-sidebar.collapse,#quarto-content.page-columns #quarto-margin-sidebar.collapsing,#quarto-content.page-columns #quarto-sidebar.collapsing{z-index:1055}}#quarto-content.page-columns main.column-page,#quarto-content.page-columns main.column-page-right,#quarto-content.page-columns main.column-page-left{z-index:0}.page-columns .column-screen-inset{grid-column:screen-start-inset/screen-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset table{background:#fefbf2}.page-columns .column-screen-inset-left{grid-column:screen-start-inset/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-inset-left table{background:#fefbf2}.page-columns .column-screen-inset-right{grid-column:body-content-start/screen-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset-right table{background:#fefbf2}.page-columns .column-screen{grid-column:screen-start/screen-end;z-index:998;opacity:.999}.page-columns .column-screen table{background:#fefbf2}.page-columns .column-screen-left{grid-column:screen-start/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-left table{background:#fefbf2}.page-columns .column-screen-right{grid-column:body-content-start/screen-end;z-index:998;opacity:.999}.page-columns .column-screen-right table{background:#fefbf2}.page-columns .column-screen-inset-shaded{grid-column:screen-start/screen-end;padding:1em;background:#f8f9fa;z-index:998;opacity:.999;margin-bottom:1em}.zindex-content{z-index:998;opacity:.999}.zindex-modal{z-index:1055;opacity:.999}.zindex-over-content{z-index:999;opacity:.999}img.img-fluid.column-screen,img.img-fluid.column-screen-inset-shaded,img.img-fluid.column-screen-inset,img.img-fluid.column-screen-inset-left,img.img-fluid.column-screen-inset-right,img.img-fluid.column-screen-left,img.img-fluid.column-screen-right{width:100%}@media(min-width: 992px){.margin-caption,div.aside,aside:not(.footnotes):not(.sidebar),.column-margin{grid-column:body-end/page-end !important;z-index:998}.column-sidebar{grid-column:page-start/body-start !important;z-index:998}.column-leftmargin{grid-column:screen-start-inset/body-start !important;z-index:998}.no-row-height{height:1em;overflow:visible}}@media(max-width: 991.98px){.margin-caption,div.aside,aside:not(.footnotes):not(.sidebar),.column-margin{grid-column:body-end/page-end !important;z-index:998}.no-row-height{height:1em;overflow:visible}.page-columns.page-full{overflow:visible}.page-columns.toc-left .margin-caption,.page-columns.toc-left div.aside,.page-columns.toc-left aside:not(.footnotes):not(.sidebar),.page-columns.toc-left .column-margin{grid-column:body-content-start/body-content-end !important;z-index:998;opacity:.999}.page-columns.toc-left .no-row-height{height:initial;overflow:initial}}@media(max-width: 767.98px){.margin-caption,div.aside,aside:not(.footnotes):not(.sidebar),.column-margin{grid-column:body-content-start/body-content-end !important;z-index:998;opacity:.999}.no-row-height{height:initial;overflow:initial}#quarto-margin-sidebar{display:none}#quarto-sidebar-toc-left{display:none}.hidden-sm{display:none}}.panel-grid{display:grid;grid-template-rows:repeat(1, 1fr);grid-template-columns:repeat(24, 1fr);gap:1em}.panel-grid .g-col-1{grid-column:auto/span 1}.panel-grid .g-col-2{grid-column:auto/span 2}.panel-grid .g-col-3{grid-column:auto/span 3}.panel-grid .g-col-4{grid-column:auto/span 4}.panel-grid .g-col-5{grid-column:auto/span 5}.panel-grid .g-col-6{grid-column:auto/span 6}.panel-grid .g-col-7{grid-column:auto/span 7}.panel-grid .g-col-8{grid-column:auto/span 8}.panel-grid .g-col-9{grid-column:auto/span 9}.panel-grid .g-col-10{grid-column:auto/span 10}.panel-grid .g-col-11{grid-column:auto/span 11}.panel-grid .g-col-12{grid-column:auto/span 12}.panel-grid .g-col-13{grid-column:auto/span 13}.panel-grid .g-col-14{grid-column:auto/span 14}.panel-grid .g-col-15{grid-column:auto/span 15}.panel-grid .g-col-16{grid-column:auto/span 16}.panel-grid .g-col-17{grid-column:auto/span 17}.panel-grid .g-col-18{grid-column:auto/span 18}.panel-grid .g-col-19{grid-column:auto/span 19}.panel-grid .g-col-20{grid-column:auto/span 20}.panel-grid .g-col-21{grid-column:auto/span 21}.panel-grid .g-col-22{grid-column:auto/span 22}.panel-grid .g-col-23{grid-column:auto/span 23}.panel-grid .g-col-24{grid-column:auto/span 24}.panel-grid .g-start-1{grid-column-start:1}.panel-grid .g-start-2{grid-column-start:2}.panel-grid .g-start-3{grid-column-start:3}.panel-grid .g-start-4{grid-column-start:4}.panel-grid .g-start-5{grid-column-start:5}.panel-grid .g-start-6{grid-column-start:6}.panel-grid .g-start-7{grid-column-start:7}.panel-grid .g-start-8{grid-column-start:8}.panel-grid .g-start-9{grid-column-start:9}.panel-grid .g-start-10{grid-column-start:10}.panel-grid .g-start-11{grid-column-start:11}.panel-grid .g-start-12{grid-column-start:12}.panel-grid .g-start-13{grid-column-start:13}.panel-grid .g-start-14{grid-column-start:14}.panel-grid .g-start-15{grid-column-start:15}.panel-grid .g-start-16{grid-column-start:16}.panel-grid .g-start-17{grid-column-start:17}.panel-grid .g-start-18{grid-column-start:18}.panel-grid .g-start-19{grid-column-start:19}.panel-grid .g-start-20{grid-column-start:20}.panel-grid .g-start-21{grid-column-start:21}.panel-grid .g-start-22{grid-column-start:22}.panel-grid .g-start-23{grid-column-start:23}@media(min-width: 576px){.panel-grid .g-col-sm-1{grid-column:auto/span 1}.panel-grid .g-col-sm-2{grid-column:auto/span 2}.panel-grid .g-col-sm-3{grid-column:auto/span 3}.panel-grid .g-col-sm-4{grid-column:auto/span 4}.panel-grid .g-col-sm-5{grid-column:auto/span 5}.panel-grid .g-col-sm-6{grid-column:auto/span 6}.panel-grid .g-col-sm-7{grid-column:auto/span 7}.panel-grid .g-col-sm-8{grid-column:auto/span 8}.panel-grid .g-col-sm-9{grid-column:auto/span 9}.panel-grid .g-col-sm-10{grid-column:auto/span 10}.panel-grid .g-col-sm-11{grid-column:auto/span 11}.panel-grid .g-col-sm-12{grid-column:auto/span 12}.panel-grid .g-col-sm-13{grid-column:auto/span 13}.panel-grid .g-col-sm-14{grid-column:auto/span 14}.panel-grid .g-col-sm-15{grid-column:auto/span 15}.panel-grid .g-col-sm-16{grid-column:auto/span 16}.panel-grid .g-col-sm-17{grid-column:auto/span 17}.panel-grid .g-col-sm-18{grid-column:auto/span 18}.panel-grid .g-col-sm-19{grid-column:auto/span 19}.panel-grid .g-col-sm-20{grid-column:auto/span 20}.panel-grid .g-col-sm-21{grid-column:auto/span 21}.panel-grid .g-col-sm-22{grid-column:auto/span 22}.panel-grid .g-col-sm-23{grid-column:auto/span 23}.panel-grid .g-col-sm-24{grid-column:auto/span 24}.panel-grid .g-start-sm-1{grid-column-start:1}.panel-grid .g-start-sm-2{grid-column-start:2}.panel-grid .g-start-sm-3{grid-column-start:3}.panel-grid .g-start-sm-4{grid-column-start:4}.panel-grid .g-start-sm-5{grid-column-start:5}.panel-grid .g-start-sm-6{grid-column-start:6}.panel-grid .g-start-sm-7{grid-column-start:7}.panel-grid .g-start-sm-8{grid-column-start:8}.panel-grid .g-start-sm-9{grid-column-start:9}.panel-grid .g-start-sm-10{grid-column-start:10}.panel-grid .g-start-sm-11{grid-column-start:11}.panel-grid .g-start-sm-12{grid-column-start:12}.panel-grid .g-start-sm-13{grid-column-start:13}.panel-grid .g-start-sm-14{grid-column-start:14}.panel-grid .g-start-sm-15{grid-column-start:15}.panel-grid .g-start-sm-16{grid-column-start:16}.panel-grid .g-start-sm-17{grid-column-start:17}.panel-grid .g-start-sm-18{grid-column-start:18}.panel-grid .g-start-sm-19{grid-column-start:19}.panel-grid .g-start-sm-20{grid-column-start:20}.panel-grid .g-start-sm-21{grid-column-start:21}.panel-grid .g-start-sm-22{grid-column-start:22}.panel-grid .g-start-sm-23{grid-column-start:23}}@media(min-width: 768px){.panel-grid .g-col-md-1{grid-column:auto/span 1}.panel-grid .g-col-md-2{grid-column:auto/span 2}.panel-grid .g-col-md-3{grid-column:auto/span 3}.panel-grid .g-col-md-4{grid-column:auto/span 4}.panel-grid .g-col-md-5{grid-column:auto/span 5}.panel-grid .g-col-md-6{grid-column:auto/span 6}.panel-grid .g-col-md-7{grid-column:auto/span 7}.panel-grid .g-col-md-8{grid-column:auto/span 8}.panel-grid .g-col-md-9{grid-column:auto/span 9}.panel-grid .g-col-md-10{grid-column:auto/span 10}.panel-grid .g-col-md-11{grid-column:auto/span 11}.panel-grid .g-col-md-12{grid-column:auto/span 12}.panel-grid .g-col-md-13{grid-column:auto/span 13}.panel-grid .g-col-md-14{grid-column:auto/span 14}.panel-grid .g-col-md-15{grid-column:auto/span 15}.panel-grid .g-col-md-16{grid-column:auto/span 16}.panel-grid .g-col-md-17{grid-column:auto/span 17}.panel-grid .g-col-md-18{grid-column:auto/span 18}.panel-grid .g-col-md-19{grid-column:auto/span 19}.panel-grid .g-col-md-20{grid-column:auto/span 20}.panel-grid .g-col-md-21{grid-column:auto/span 21}.panel-grid .g-col-md-22{grid-column:auto/span 22}.panel-grid .g-col-md-23{grid-column:auto/span 23}.panel-grid .g-col-md-24{grid-column:auto/span 24}.panel-grid .g-start-md-1{grid-column-start:1}.panel-grid .g-start-md-2{grid-column-start:2}.panel-grid .g-start-md-3{grid-column-start:3}.panel-grid .g-start-md-4{grid-column-start:4}.panel-grid .g-start-md-5{grid-column-start:5}.panel-grid .g-start-md-6{grid-column-start:6}.panel-grid .g-start-md-7{grid-column-start:7}.panel-grid .g-start-md-8{grid-column-start:8}.panel-grid .g-start-md-9{grid-column-start:9}.panel-grid .g-start-md-10{grid-column-start:10}.panel-grid .g-start-md-11{grid-column-start:11}.panel-grid .g-start-md-12{grid-column-start:12}.panel-grid .g-start-md-13{grid-column-start:13}.panel-grid .g-start-md-14{grid-column-start:14}.panel-grid .g-start-md-15{grid-column-start:15}.panel-grid .g-start-md-16{grid-column-start:16}.panel-grid .g-start-md-17{grid-column-start:17}.panel-grid .g-start-md-18{grid-column-start:18}.panel-grid .g-start-md-19{grid-column-start:19}.panel-grid .g-start-md-20{grid-column-start:20}.panel-grid .g-start-md-21{grid-column-start:21}.panel-grid .g-start-md-22{grid-column-start:22}.panel-grid .g-start-md-23{grid-column-start:23}}@media(min-width: 992px){.panel-grid .g-col-lg-1{grid-column:auto/span 1}.panel-grid .g-col-lg-2{grid-column:auto/span 2}.panel-grid .g-col-lg-3{grid-column:auto/span 3}.panel-grid .g-col-lg-4{grid-column:auto/span 4}.panel-grid .g-col-lg-5{grid-column:auto/span 5}.panel-grid .g-col-lg-6{grid-column:auto/span 6}.panel-grid .g-col-lg-7{grid-column:auto/span 7}.panel-grid .g-col-lg-8{grid-column:auto/span 8}.panel-grid .g-col-lg-9{grid-column:auto/span 9}.panel-grid .g-col-lg-10{grid-column:auto/span 10}.panel-grid .g-col-lg-11{grid-column:auto/span 11}.panel-grid .g-col-lg-12{grid-column:auto/span 12}.panel-grid .g-col-lg-13{grid-column:auto/span 13}.panel-grid .g-col-lg-14{grid-column:auto/span 14}.panel-grid .g-col-lg-15{grid-column:auto/span 15}.panel-grid .g-col-lg-16{grid-column:auto/span 16}.panel-grid .g-col-lg-17{grid-column:auto/span 17}.panel-grid .g-col-lg-18{grid-column:auto/span 18}.panel-grid .g-col-lg-19{grid-column:auto/span 19}.panel-grid .g-col-lg-20{grid-column:auto/span 20}.panel-grid .g-col-lg-21{grid-column:auto/span 21}.panel-grid .g-col-lg-22{grid-column:auto/span 22}.panel-grid .g-col-lg-23{grid-column:auto/span 23}.panel-grid .g-col-lg-24{grid-column:auto/span 24}.panel-grid .g-start-lg-1{grid-column-start:1}.panel-grid .g-start-lg-2{grid-column-start:2}.panel-grid .g-start-lg-3{grid-column-start:3}.panel-grid .g-start-lg-4{grid-column-start:4}.panel-grid .g-start-lg-5{grid-column-start:5}.panel-grid .g-start-lg-6{grid-column-start:6}.panel-grid .g-start-lg-7{grid-column-start:7}.panel-grid .g-start-lg-8{grid-column-start:8}.panel-grid .g-start-lg-9{grid-column-start:9}.panel-grid .g-start-lg-10{grid-column-start:10}.panel-grid .g-start-lg-11{grid-column-start:11}.panel-grid .g-start-lg-12{grid-column-start:12}.panel-grid .g-start-lg-13{grid-column-start:13}.panel-grid .g-start-lg-14{grid-column-start:14}.panel-grid .g-start-lg-15{grid-column-start:15}.panel-grid .g-start-lg-16{grid-column-start:16}.panel-grid .g-start-lg-17{grid-column-start:17}.panel-grid .g-start-lg-18{grid-column-start:18}.panel-grid .g-start-lg-19{grid-column-start:19}.panel-grid .g-start-lg-20{grid-column-start:20}.panel-grid .g-start-lg-21{grid-column-start:21}.panel-grid .g-start-lg-22{grid-column-start:22}.panel-grid .g-start-lg-23{grid-column-start:23}}@media(min-width: 1200px){.panel-grid .g-col-xl-1{grid-column:auto/span 1}.panel-grid .g-col-xl-2{grid-column:auto/span 2}.panel-grid .g-col-xl-3{grid-column:auto/span 3}.panel-grid .g-col-xl-4{grid-column:auto/span 4}.panel-grid .g-col-xl-5{grid-column:auto/span 5}.panel-grid .g-col-xl-6{grid-column:auto/span 6}.panel-grid .g-col-xl-7{grid-column:auto/span 7}.panel-grid .g-col-xl-8{grid-column:auto/span 8}.panel-grid .g-col-xl-9{grid-column:auto/span 9}.panel-grid .g-col-xl-10{grid-column:auto/span 10}.panel-grid .g-col-xl-11{grid-column:auto/span 11}.panel-grid .g-col-xl-12{grid-column:auto/span 12}.panel-grid .g-col-xl-13{grid-column:auto/span 13}.panel-grid .g-col-xl-14{grid-column:auto/span 14}.panel-grid .g-col-xl-15{grid-column:auto/span 15}.panel-grid .g-col-xl-16{grid-column:auto/span 16}.panel-grid .g-col-xl-17{grid-column:auto/span 17}.panel-grid .g-col-xl-18{grid-column:auto/span 18}.panel-grid .g-col-xl-19{grid-column:auto/span 19}.panel-grid .g-col-xl-20{grid-column:auto/span 20}.panel-grid .g-col-xl-21{grid-column:auto/span 21}.panel-grid .g-col-xl-22{grid-column:auto/span 22}.panel-grid .g-col-xl-23{grid-column:auto/span 23}.panel-grid .g-col-xl-24{grid-column:auto/span 24}.panel-grid .g-start-xl-1{grid-column-start:1}.panel-grid .g-start-xl-2{grid-column-start:2}.panel-grid .g-start-xl-3{grid-column-start:3}.panel-grid .g-start-xl-4{grid-column-start:4}.panel-grid .g-start-xl-5{grid-column-start:5}.panel-grid .g-start-xl-6{grid-column-start:6}.panel-grid .g-start-xl-7{grid-column-start:7}.panel-grid .g-start-xl-8{grid-column-start:8}.panel-grid .g-start-xl-9{grid-column-start:9}.panel-grid .g-start-xl-10{grid-column-start:10}.panel-grid .g-start-xl-11{grid-column-start:11}.panel-grid .g-start-xl-12{grid-column-start:12}.panel-grid .g-start-xl-13{grid-column-start:13}.panel-grid .g-start-xl-14{grid-column-start:14}.panel-grid .g-start-xl-15{grid-column-start:15}.panel-grid .g-start-xl-16{grid-column-start:16}.panel-grid .g-start-xl-17{grid-column-start:17}.panel-grid .g-start-xl-18{grid-column-start:18}.panel-grid .g-start-xl-19{grid-column-start:19}.panel-grid .g-start-xl-20{grid-column-start:20}.panel-grid .g-start-xl-21{grid-column-start:21}.panel-grid .g-start-xl-22{grid-column-start:22}.panel-grid .g-start-xl-23{grid-column-start:23}}@media(min-width: 1400px){.panel-grid .g-col-xxl-1{grid-column:auto/span 1}.panel-grid .g-col-xxl-2{grid-column:auto/span 2}.panel-grid .g-col-xxl-3{grid-column:auto/span 3}.panel-grid .g-col-xxl-4{grid-column:auto/span 4}.panel-grid .g-col-xxl-5{grid-column:auto/span 5}.panel-grid .g-col-xxl-6{grid-column:auto/span 6}.panel-grid .g-col-xxl-7{grid-column:auto/span 7}.panel-grid .g-col-xxl-8{grid-column:auto/span 8}.panel-grid .g-col-xxl-9{grid-column:auto/span 9}.panel-grid .g-col-xxl-10{grid-column:auto/span 10}.panel-grid .g-col-xxl-11{grid-column:auto/span 11}.panel-grid .g-col-xxl-12{grid-column:auto/span 12}.panel-grid .g-col-xxl-13{grid-column:auto/span 13}.panel-grid .g-col-xxl-14{grid-column:auto/span 14}.panel-grid .g-col-xxl-15{grid-column:auto/span 15}.panel-grid .g-col-xxl-16{grid-column:auto/span 16}.panel-grid .g-col-xxl-17{grid-column:auto/span 17}.panel-grid .g-col-xxl-18{grid-column:auto/span 18}.panel-grid .g-col-xxl-19{grid-column:auto/span 19}.panel-grid .g-col-xxl-20{grid-column:auto/span 20}.panel-grid .g-col-xxl-21{grid-column:auto/span 21}.panel-grid .g-col-xxl-22{grid-column:auto/span 22}.panel-grid .g-col-xxl-23{grid-column:auto/span 23}.panel-grid .g-col-xxl-24{grid-column:auto/span 24}.panel-grid .g-start-xxl-1{grid-column-start:1}.panel-grid .g-start-xxl-2{grid-column-start:2}.panel-grid .g-start-xxl-3{grid-column-start:3}.panel-grid .g-start-xxl-4{grid-column-start:4}.panel-grid .g-start-xxl-5{grid-column-start:5}.panel-grid .g-start-xxl-6{grid-column-start:6}.panel-grid .g-start-xxl-7{grid-column-start:7}.panel-grid .g-start-xxl-8{grid-column-start:8}.panel-grid .g-start-xxl-9{grid-column-start:9}.panel-grid .g-start-xxl-10{grid-column-start:10}.panel-grid .g-start-xxl-11{grid-column-start:11}.panel-grid .g-start-xxl-12{grid-column-start:12}.panel-grid .g-start-xxl-13{grid-column-start:13}.panel-grid .g-start-xxl-14{grid-column-start:14}.panel-grid .g-start-xxl-15{grid-column-start:15}.panel-grid .g-start-xxl-16{grid-column-start:16}.panel-grid .g-start-xxl-17{grid-column-start:17}.panel-grid .g-start-xxl-18{grid-column-start:18}.panel-grid .g-start-xxl-19{grid-column-start:19}.panel-grid .g-start-xxl-20{grid-column-start:20}.panel-grid .g-start-xxl-21{grid-column-start:21}.panel-grid .g-start-xxl-22{grid-column-start:22}.panel-grid .g-start-xxl-23{grid-column-start:23}}main{margin-top:1em;margin-bottom:1em}h1,.h1,h2,.h2{color:inherit;margin-top:2rem;margin-bottom:1rem;font-weight:600}h1.title,.title.h1{margin-top:0}main.content>section:first-of-type>h2:first-child,main.content>section:first-of-type>.h2:first-child{margin-top:0}h2,.h2{border-bottom:1px solid #dee2e6;padding-bottom:.5rem}h3,.h3{font-weight:600}h3,.h3,h4,.h4{opacity:.9;margin-top:1.5rem}h5,.h5,h6,.h6{opacity:.9}.header-section-number{color:#6d7a86}.nav-link.active .header-section-number{color:inherit}mark,.mark{padding:0em}.panel-caption,.figure-caption,.subfigure-caption,.table-caption,figcaption,caption{font-size:.9rem;color:#6d7a86}.quarto-layout-cell[data-ref-parent] caption{color:#6d7a86}.column-margin figcaption,.margin-caption,div.aside,aside,.column-margin{color:#6d7a86;font-size:.825rem}.panel-caption.margin-caption{text-align:inherit}.column-margin.column-container p{margin-bottom:0}.column-margin.column-container>*:not(.collapse):first-child{padding-bottom:.5em;display:block}.column-margin.column-container>*:not(.collapse):not(:first-child){padding-top:.5em;padding-bottom:.5em;display:block}.column-margin.column-container>*.collapse:not(.show){display:none}@media(min-width: 768px){.column-margin.column-container .callout-margin-content:first-child{margin-top:4.5em}.column-margin.column-container .callout-margin-content-simple:first-child{margin-top:3.5em}}.margin-caption>*{padding-top:.5em;padding-bottom:.5em}@media(max-width: 767.98px){.quarto-layout-row{flex-direction:column}}.nav-tabs .nav-item{margin-top:1px;cursor:pointer}.tab-content{margin-top:0px;border-left:#dee2e6 1px solid;border-right:#dee2e6 1px solid;border-bottom:#dee2e6 1px solid;margin-left:0;padding:1em;margin-bottom:1em}@media(max-width: 767.98px){.layout-sidebar{margin-left:0;margin-right:0}}.panel-sidebar,.panel-sidebar .form-control,.panel-input,.panel-input .form-control,.selectize-dropdown{font-size:.9rem}.panel-sidebar .form-control,.panel-input .form-control{padding-top:.1rem}.tab-pane div.sourceCode{margin-top:0px}.tab-pane>p{padding-top:0}.tab-pane>p:nth-child(1){padding-top:0}.tab-pane>p:last-child{margin-bottom:0}.tab-pane>pre:last-child{margin-bottom:0}.tab-content>.tab-pane:not(.active){display:none !important}div.sourceCode{background-color:rgba(233,236,239,.65);border:1px solid rgba(233,236,239,.65)}pre.sourceCode{background-color:rgba(0,0,0,0)}pre.sourceCode{border:none;font-size:.875em;overflow:visible !important;padding:.4em}div.sourceCode{overflow-y:hidden}.callout div.sourceCode{margin-left:initial}.blockquote{font-size:inherit;padding-left:1rem;padding-right:1.5rem;color:#6d7a86}.blockquote h1:first-child,.blockquote .h1:first-child,.blockquote h2:first-child,.blockquote .h2:first-child,.blockquote h3:first-child,.blockquote .h3:first-child,.blockquote h4:first-child,.blockquote .h4:first-child,.blockquote h5:first-child,.blockquote .h5:first-child{margin-top:0}pre{background-color:initial;padding:initial;border:initial}p pre code:not(.sourceCode),li pre code:not(.sourceCode),pre code:not(.sourceCode){background-color:initial}p code:not(.sourceCode),li code:not(.sourceCode),td code:not(.sourceCode){background-color:#f8f9fa;padding:.2em}nav p code:not(.sourceCode),nav li code:not(.sourceCode),nav td code:not(.sourceCode){background-color:rgba(0,0,0,0);padding:0}td code:not(.sourceCode){white-space:pre-wrap}#quarto-embedded-source-code-modal>.modal-dialog{max-width:1000px;padding-left:1.75rem;padding-right:1.75rem}#quarto-embedded-source-code-modal>.modal-dialog>.modal-content>.modal-body{padding:0}#quarto-embedded-source-code-modal>.modal-dialog>.modal-content>.modal-body div.sourceCode{margin:0;padding:.2rem .2rem;border-radius:0px;border:none}#quarto-embedded-source-code-modal>.modal-dialog>.modal-content>.modal-header{padding:.7rem}.code-tools-button{font-size:1rem;padding:.15rem .15rem;margin-left:5px;color:#6c757d;background-color:rgba(0,0,0,0);transition:initial;cursor:pointer}.code-tools-button>.bi::before{display:inline-block;height:1rem;width:1rem;content:"";vertical-align:-0.125em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>');background-repeat:no-repeat;background-size:1rem 1rem}.code-tools-button:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(52, 58, 64)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>')}#quarto-embedded-source-code-modal .code-copy-button>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" viewBox="0 0 16 16"><path d="M4 1.5H3a2 2 0 0 0-2 2V14a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V3.5a2 2 0 0 0-2-2h-1v1h1a1 1 0 0 1 1 1V14a1 1 0 0 1-1 1H3a1 1 0 0 1-1-1V3.5a1 1 0 0 1 1-1h1v-1z"/><path d="M9.5 1a.5.5 0 0 1 .5.5v1a.5.5 0 0 1-.5.5h-3a.5.5 0 0 1-.5-.5v-1a.5.5 0 0 1 .5-.5h3zm-3-1A1.5 1.5 0 0 0 5 1.5v1A1.5 1.5 0 0 0 6.5 4h3A1.5 1.5 0 0 0 11 2.5v-1A1.5 1.5 0 0 0 9.5 0h-3z"/></svg>')}#quarto-embedded-source-code-modal .code-copy-button-checked>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" viewBox="0 0 16 16"><path d="M13.854 3.646a.5.5 0 0 1 0 .708l-7 7a.5.5 0 0 1-.708 0l-3.5-3.5a.5.5 0 1 1 .708-.708L6.5 10.293l6.646-6.647a.5.5 0 0 1 .708 0z"/></svg>')}.sidebar{will-change:top;transition:top 200ms linear;position:sticky;overflow-y:auto;padding-top:1.2em;max-height:100vh}.sidebar.toc-left,.sidebar.margin-sidebar{top:0px;padding-top:1em}.sidebar.quarto-banner-title-block-sidebar>*{padding-top:1.65em}figure .quarto-notebook-link{margin-top:.5em}.quarto-notebook-link{font-size:.75em;color:#6c757d;margin-bottom:1em;text-decoration:none;display:block}.quarto-notebook-link:hover{text-decoration:underline;color:#2761e3}.quarto-notebook-link::before{display:inline-block;height:.75rem;width:.75rem;margin-bottom:0em;margin-right:.25em;content:"";vertical-align:-0.125em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" class="bi bi-journal-code" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M8.646 5.646a.5.5 0 0 1 .708 0l2 2a.5.5 0 0 1 0 .708l-2 2a.5.5 0 0 1-.708-.708L10.293 8 8.646 6.354a.5.5 0 0 1 0-.708zm-1.292 0a.5.5 0 0 0-.708 0l-2 2a.5.5 0 0 0 0 .708l2 2a.5.5 0 0 0 .708-.708L5.707 8l1.647-1.646a.5.5 0 0 0 0-.708z"/><path d="M3 0h10a2 2 0 0 1 2 2v12a2 2 0 0 1-2 2H3a2 2 0 0 1-2-2v-1h1v1a1 1 0 0 0 1 1h10a1 1 0 0 0 1-1V2a1 1 0 0 0-1-1H3a1 1 0 0 0-1 1v1H1V2a2 2 0 0 1 2-2z"/><path d="M1 5v-.5a.5.5 0 0 1 1 0V5h.5a.5.5 0 0 1 0 1h-2a.5.5 0 0 1 0-1H1zm0 3v-.5a.5.5 0 0 1 1 0V8h.5a.5.5 0 0 1 0 1h-2a.5.5 0 0 1 0-1H1zm0 3v-.5a.5.5 0 0 1 1 0v.5h.5a.5.5 0 0 1 0 1h-2a.5.5 0 0 1 0-1H1z"/></svg>');background-repeat:no-repeat;background-size:.75rem .75rem}.toc-actions i.bi,.quarto-code-links i.bi,.quarto-other-links i.bi,.quarto-alternate-notebooks i.bi,.quarto-alternate-formats i.bi{margin-right:.4em;font-size:.8rem}.quarto-other-links-text-target .quarto-code-links i.bi,.quarto-other-links-text-target .quarto-other-links i.bi{margin-right:.2em}.quarto-other-formats-text-target .quarto-alternate-formats i.bi{margin-right:.1em}.toc-actions i.bi.empty,.quarto-code-links i.bi.empty,.quarto-other-links i.bi.empty,.quarto-alternate-notebooks i.bi.empty,.quarto-alternate-formats i.bi.empty{padding-left:1em}.quarto-notebook h2,.quarto-notebook .h2{border-bottom:none}.quarto-notebook .cell-container{display:flex}.quarto-notebook .cell-container .cell{flex-grow:4}.quarto-notebook .cell-container .cell-decorator{padding-top:1.5em;padding-right:1em;text-align:right}.quarto-notebook .cell-container.code-fold .cell-decorator{padding-top:3em}.quarto-notebook .cell-code code{white-space:pre-wrap}.quarto-notebook .cell .cell-output-stderr pre code,.quarto-notebook .cell .cell-output-stdout pre code{white-space:pre-wrap;overflow-wrap:anywhere}.toc-actions,.quarto-alternate-formats,.quarto-other-links,.quarto-code-links,.quarto-alternate-notebooks{padding-left:0em}.sidebar .toc-actions a,.sidebar .quarto-alternate-formats a,.sidebar .quarto-other-links a,.sidebar .quarto-code-links a,.sidebar .quarto-alternate-notebooks a,.sidebar nav[role=doc-toc] a{text-decoration:none}.sidebar .toc-actions a:hover,.sidebar .quarto-other-links a:hover,.sidebar .quarto-code-links a:hover,.sidebar .quarto-alternate-formats a:hover,.sidebar .quarto-alternate-notebooks a:hover{color:#2761e3}.sidebar .toc-actions h2,.sidebar .toc-actions .h2,.sidebar .quarto-code-links h2,.sidebar .quarto-code-links .h2,.sidebar .quarto-other-links h2,.sidebar .quarto-other-links .h2,.sidebar .quarto-alternate-notebooks h2,.sidebar .quarto-alternate-notebooks .h2,.sidebar .quarto-alternate-formats h2,.sidebar .quarto-alternate-formats .h2,.sidebar nav[role=doc-toc]>h2,.sidebar nav[role=doc-toc]>.h2{font-weight:500;margin-bottom:.2rem;margin-top:.3rem;font-family:inherit;border-bottom:0;padding-bottom:0;padding-top:0px}.sidebar .toc-actions>h2,.sidebar .toc-actions>.h2,.sidebar .quarto-code-links>h2,.sidebar .quarto-code-links>.h2,.sidebar .quarto-other-links>h2,.sidebar .quarto-other-links>.h2,.sidebar .quarto-alternate-notebooks>h2,.sidebar .quarto-alternate-notebooks>.h2,.sidebar .quarto-alternate-formats>h2,.sidebar .quarto-alternate-formats>.h2{font-size:.8rem}.sidebar nav[role=doc-toc]>h2,.sidebar nav[role=doc-toc]>.h2{font-size:.875rem}.sidebar nav[role=doc-toc]>ul a{border-left:1px solid #e9ecef;padding-left:.6rem}.sidebar .toc-actions h2>ul a,.sidebar .toc-actions .h2>ul a,.sidebar .quarto-code-links h2>ul a,.sidebar .quarto-code-links .h2>ul a,.sidebar .quarto-other-links h2>ul a,.sidebar .quarto-other-links .h2>ul a,.sidebar .quarto-alternate-notebooks h2>ul a,.sidebar .quarto-alternate-notebooks .h2>ul a,.sidebar .quarto-alternate-formats h2>ul a,.sidebar .quarto-alternate-formats .h2>ul a{border-left:none;padding-left:.6rem}.sidebar .toc-actions ul a:empty,.sidebar .quarto-code-links ul a:empty,.sidebar .quarto-other-links ul a:empty,.sidebar .quarto-alternate-notebooks ul a:empty,.sidebar .quarto-alternate-formats ul a:empty,.sidebar nav[role=doc-toc]>ul a:empty{display:none}.sidebar .toc-actions ul,.sidebar .quarto-code-links ul,.sidebar .quarto-other-links ul,.sidebar .quarto-alternate-notebooks ul,.sidebar .quarto-alternate-formats ul{padding-left:0;list-style:none}.sidebar nav[role=doc-toc] ul{list-style:none;padding-left:0;list-style:none}.sidebar nav[role=doc-toc]>ul{margin-left:.45em}.quarto-margin-sidebar nav[role=doc-toc]{padding-left:.5em}.sidebar .toc-actions>ul,.sidebar .quarto-code-links>ul,.sidebar .quarto-other-links>ul,.sidebar .quarto-alternate-notebooks>ul,.sidebar .quarto-alternate-formats>ul{font-size:.8rem}.sidebar nav[role=doc-toc]>ul{font-size:.875rem}.sidebar .toc-actions ul li a,.sidebar .quarto-code-links ul li a,.sidebar .quarto-other-links ul li a,.sidebar .quarto-alternate-notebooks ul li a,.sidebar .quarto-alternate-formats ul li a,.sidebar nav[role=doc-toc]>ul li a{line-height:1.1rem;padding-bottom:.2rem;padding-top:.2rem;color:inherit}.sidebar nav[role=doc-toc] ul>li>ul>li>a{padding-left:1.2em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>a{padding-left:2.4em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>ul>li>a{padding-left:3.6em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>ul>li>ul>li>a{padding-left:4.8em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>ul>li>ul>li>ul>li>a{padding-left:6em}.sidebar nav[role=doc-toc] ul>li>a.active,.sidebar nav[role=doc-toc] ul>li>ul>li>a.active{border-left:1px solid #2761e3;color:#2761e3 !important}.sidebar nav[role=doc-toc] ul>li>a:hover,.sidebar nav[role=doc-toc] ul>li>ul>li>a:hover{color:#2761e3 !important}kbd,.kbd{color:#343a40;background-color:#f8f9fa;border:1px solid;border-radius:5px;border-color:#dee2e6}.quarto-appendix-contents div.hanging-indent{margin-left:0em}.quarto-appendix-contents div.hanging-indent div.csl-entry{margin-left:1em;text-indent:-1em}.citation a,.footnote-ref{text-decoration:none}.footnotes ol{padding-left:1em}.tippy-content>*{margin-bottom:.7em}.tippy-content>*:last-child{margin-bottom:0}.callout{margin-top:1.25rem;margin-bottom:1.25rem;border-radius:.25rem;overflow-wrap:break-word}.callout .callout-title-container{overflow-wrap:anywhere}.callout.callout-style-simple{padding:.4em .7em;border-left:5px solid;border-right:1px solid #dee2e6;border-top:1px solid #dee2e6;border-bottom:1px solid #dee2e6}.callout.callout-style-default{border-left:5px solid;border-right:1px solid #dee2e6;border-top:1px solid #dee2e6;border-bottom:1px solid #dee2e6}.callout .callout-body-container{flex-grow:1}.callout.callout-style-simple .callout-body{font-size:.9rem;font-weight:400}.callout.callout-style-default .callout-body{font-size:.9rem;font-weight:400}.callout:not(.no-icon).callout-titled.callout-style-simple .callout-body{padding-left:1.6em}.callout.callout-titled>.callout-header{padding-top:.2em;margin-bottom:-0.2em}.callout.callout-style-simple>div.callout-header{border-bottom:none;font-size:.9rem;font-weight:600;opacity:75%}.callout.callout-style-default>div.callout-header{border-bottom:none;font-weight:600;opacity:85%;font-size:.9rem;padding-left:.5em;padding-right:.5em}.callout.callout-style-default .callout-body{padding-left:.5em;padding-right:.5em}.callout.callout-style-default .callout-body>:first-child{padding-top:.5rem;margin-top:0}.callout>div.callout-header[data-bs-toggle=collapse]{cursor:pointer}.callout.callout-style-default .callout-header[aria-expanded=false],.callout.callout-style-default .callout-header[aria-expanded=true]{padding-top:0px;margin-bottom:0px;align-items:center}.callout.callout-titled .callout-body>:last-child:not(.sourceCode),.callout.callout-titled .callout-body>div>:last-child:not(.sourceCode){padding-bottom:.5rem;margin-bottom:0}.callout:not(.callout-titled) .callout-body>:first-child,.callout:not(.callout-titled) .callout-body>div>:first-child{margin-top:.25rem}.callout:not(.callout-titled) .callout-body>:last-child,.callout:not(.callout-titled) .callout-body>div>:last-child{margin-bottom:.2rem}.callout.callout-style-simple .callout-icon::before,.callout.callout-style-simple .callout-toggle::before{height:1rem;width:1rem;display:inline-block;content:"";background-repeat:no-repeat;background-size:1rem 1rem}.callout.callout-style-default .callout-icon::before,.callout.callout-style-default .callout-toggle::before{height:.9rem;width:.9rem;display:inline-block;content:"";background-repeat:no-repeat;background-size:.9rem .9rem}.callout.callout-style-default .callout-toggle::before{margin-top:5px}.callout .callout-btn-toggle .callout-toggle::before{transition:transform .2s linear}.callout .callout-header[aria-expanded=false] .callout-toggle::before{transform:rotate(-90deg)}.callout .callout-header[aria-expanded=true] .callout-toggle::before{transform:none}.callout.callout-style-simple:not(.no-icon) div.callout-icon-container{padding-top:.2em;padding-right:.55em}.callout.callout-style-default:not(.no-icon) div.callout-icon-container{padding-top:.1em;padding-right:.35em}.callout.callout-style-default:not(.no-icon) div.callout-title-container{margin-top:-1px}.callout.callout-style-default.callout-caution:not(.no-icon) div.callout-icon-container{padding-top:.3em;padding-right:.35em}.callout>.callout-body>.callout-icon-container>.no-icon,.callout>.callout-header>.callout-icon-container>.no-icon{display:none}div.callout.callout{border-left-color:#6c757d}div.callout.callout-style-default>.callout-header{background-color:#6c757d}div.callout-note.callout{border-left-color:#2780e3}div.callout-note.callout-style-default>.callout-header{background-color:#e9f2fc}div.callout-note:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %232373cc" class="bi bi-info-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="m8.93 6.588-2.29.287-.082.38.45.083c.294.07.352.176.288.469l-.738 3.468c-.194.897.105 1.319.808 1.319.545 0 1.178-.252 1.465-.598l.088-.416c-.2.176-.492.246-.686.246-.275 0-.375-.193-.304-.533L8.93 6.588zM9 4.5a1 1 0 1 1-2 0 1 1 0 0 1 2 0z"/></svg>');}div.callout-note.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %232373cc" class="bi bi-info-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="m8.93 6.588-2.29.287-.082.38.45.083c.294.07.352.176.288.469l-.738 3.468c-.194.897.105 1.319.808 1.319.545 0 1.178-.252 1.465-.598l.088-.416c-.2.176-.492.246-.686.246-.275 0-.375-.193-.304-.533L8.93 6.588zM9 4.5a1 1 0 1 1-2 0 1 1 0 0 1 2 0z"/></svg>');}div.callout-note .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(52, 58, 64)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-tip.callout{border-left-color:#3fb618}div.callout-tip.callout-style-default>.callout-header{background-color:#ecf8e8}div.callout-tip:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %2339a416" class="bi bi-lightbulb" viewBox="0 0 16 16"><path d="M2 6a6 6 0 1 1 10.174 4.31c-.203.196-.359.4-.453.619l-.762 1.769A.5.5 0 0 1 10.5 13a.5.5 0 0 1 0 1 .5.5 0 0 1 0 1l-.224.447a1 1 0 0 1-.894.553H6.618a1 1 0 0 1-.894-.553L5.5 15a.5.5 0 0 1 0-1 .5.5 0 0 1 0-1 .5.5 0 0 1-.46-.302l-.761-1.77a1.964 1.964 0 0 0-.453-.618A5.984 5.984 0 0 1 2 6zm6-5a5 5 0 0 0-3.479 8.592c.263.254.514.564.676.941L5.83 12h4.342l.632-1.467c.162-.377.413-.687.676-.941A5 5 0 0 0 8 1z"/></svg>');}div.callout-tip.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %2339a416" class="bi bi-lightbulb" viewBox="0 0 16 16"><path d="M2 6a6 6 0 1 1 10.174 4.31c-.203.196-.359.4-.453.619l-.762 1.769A.5.5 0 0 1 10.5 13a.5.5 0 0 1 0 1 .5.5 0 0 1 0 1l-.224.447a1 1 0 0 1-.894.553H6.618a1 1 0 0 1-.894-.553L5.5 15a.5.5 0 0 1 0-1 .5.5 0 0 1 0-1 .5.5 0 0 1-.46-.302l-.761-1.77a1.964 1.964 0 0 0-.453-.618A5.984 5.984 0 0 1 2 6zm6-5a5 5 0 0 0-3.479 8.592c.263.254.514.564.676.941L5.83 12h4.342l.632-1.467c.162-.377.413-.687.676-.941A5 5 0 0 0 8 1z"/></svg>');}div.callout-tip .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(52, 58, 64)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-warning.callout{border-left-color:#ff7518}div.callout-warning.callout-style-default>.callout-header{background-color:#fff1e8}div.callout-warning:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e66916" class="bi bi-exclamation-triangle" viewBox="0 0 16 16"><path d="M7.938 2.016A.13.13 0 0 1 8.002 2a.13.13 0 0 1 .063.016.146.146 0 0 1 .054.057l6.857 11.667c.036.06.035.124.002.183a.163.163 0 0 1-.054.06.116.116 0 0 1-.066.017H1.146a.115.115 0 0 1-.066-.017.163.163 0 0 1-.054-.06.176.176 0 0 1 .002-.183L7.884 2.073a.147.147 0 0 1 .054-.057zm1.044-.45a1.13 1.13 0 0 0-1.96 0L.165 13.233c-.457.778.091 1.767.98 1.767h13.713c.889 0 1.438-.99.98-1.767L8.982 1.566z"/><path d="M7.002 12a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 5.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 5.995z"/></svg>');}div.callout-warning.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e66916" class="bi bi-exclamation-triangle" viewBox="0 0 16 16"><path d="M7.938 2.016A.13.13 0 0 1 8.002 2a.13.13 0 0 1 .063.016.146.146 0 0 1 .054.057l6.857 11.667c.036.06.035.124.002.183a.163.163 0 0 1-.054.06.116.116 0 0 1-.066.017H1.146a.115.115 0 0 1-.066-.017.163.163 0 0 1-.054-.06.176.176 0 0 1 .002-.183L7.884 2.073a.147.147 0 0 1 .054-.057zm1.044-.45a1.13 1.13 0 0 0-1.96 0L.165 13.233c-.457.778.091 1.767.98 1.767h13.713c.889 0 1.438-.99.98-1.767L8.982 1.566z"/><path d="M7.002 12a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 5.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 5.995z"/></svg>');}div.callout-warning .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(52, 58, 64)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-caution.callout{border-left-color:#f0ad4e}div.callout-caution.callout-style-default>.callout-header{background-color:#fef7ed}div.callout-caution:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23d89c46" class="bi bi-cone-striped" viewBox="0 0 16 16"><path d="M9.97 4.88l.953 3.811C10.158 8.878 9.14 9 8 9c-1.14 0-2.159-.122-2.923-.309L6.03 4.88C6.635 4.957 7.3 5 8 5s1.365-.043 1.97-.12zm-.245-.978L8.97.88C8.718-.13 7.282-.13 7.03.88L6.274 3.9C6.8 3.965 7.382 4 8 4c.618 0 1.2-.036 1.725-.098zm4.396 8.613a.5.5 0 0 1 .037.96l-6 2a.5.5 0 0 1-.316 0l-6-2a.5.5 0 0 1 .037-.96l2.391-.598.565-2.257c.862.212 1.964.339 3.165.339s2.303-.127 3.165-.339l.565 2.257 2.391.598z"/></svg>');}div.callout-caution.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23d89c46" class="bi bi-cone-striped" viewBox="0 0 16 16"><path d="M9.97 4.88l.953 3.811C10.158 8.878 9.14 9 8 9c-1.14 0-2.159-.122-2.923-.309L6.03 4.88C6.635 4.957 7.3 5 8 5s1.365-.043 1.97-.12zm-.245-.978L8.97.88C8.718-.13 7.282-.13 7.03.88L6.274 3.9C6.8 3.965 7.382 4 8 4c.618 0 1.2-.036 1.725-.098zm4.396 8.613a.5.5 0 0 1 .037.96l-6 2a.5.5 0 0 1-.316 0l-6-2a.5.5 0 0 1 .037-.96l2.391-.598.565-2.257c.862.212 1.964.339 3.165.339s2.303-.127 3.165-.339l.565 2.257 2.391.598z"/></svg>');}div.callout-caution .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(52, 58, 64)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-important.callout{border-left-color:#ff0039}div.callout-important.callout-style-default>.callout-header{background-color:#ffe6eb}div.callout-important:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e60033" class="bi bi-exclamation-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="M7.002 11a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 4.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 4.995z"/></svg>');}div.callout-important.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e60033" class="bi bi-exclamation-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="M7.002 11a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 4.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 4.995z"/></svg>');}div.callout-important .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(52, 58, 64)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}.quarto-toggle-container{display:flex;align-items:center}.quarto-reader-toggle .bi::before,.quarto-color-scheme-toggle .bi::before{display:inline-block;height:1rem;width:1rem;content:"";background-repeat:no-repeat;background-size:1rem 1rem}.sidebar-navigation{padding-left:20px}.navbar{background-color:#052744;color:#f2e5bd}.navbar .quarto-color-scheme-toggle:not(.alternate) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(242, 229, 189, 1)" class="bi bi-toggle-off" viewBox="0 0 16 16"><path d="M11 4a4 4 0 0 1 0 8H8a4.992 4.992 0 0 0 2-4 4.992 4.992 0 0 0-2-4h3zm-6 8a4 4 0 1 1 0-8 4 4 0 0 1 0 8zM0 8a5 5 0 0 0 5 5h6a5 5 0 0 0 0-10H5a5 5 0 0 0-5 5z"/></svg>')}.navbar .quarto-color-scheme-toggle.alternate .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(242, 229, 189, 1)" class="bi bi-toggle-on" viewBox="0 0 16 16"><path d="M5 3a5 5 0 0 0 0 10h6a5 5 0 0 0 0-10H5zm6 9a4 4 0 1 1 0-8 4 4 0 0 1 0 8z"/></svg>')}.sidebar-navigation .quarto-color-scheme-toggle:not(.alternate) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(75, 84, 92, 1)" class="bi bi-toggle-off" viewBox="0 0 16 16"><path d="M11 4a4 4 0 0 1 0 8H8a4.992 4.992 0 0 0 2-4 4.992 4.992 0 0 0-2-4h3zm-6 8a4 4 0 1 1 0-8 4 4 0 0 1 0 8zM0 8a5 5 0 0 0 5 5h6a5 5 0 0 0 0-10H5a5 5 0 0 0-5 5z"/></svg>')}.sidebar-navigation .quarto-color-scheme-toggle.alternate .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(75, 84, 92, 1)" class="bi bi-toggle-on" viewBox="0 0 16 16"><path d="M5 3a5 5 0 0 0 0 10h6a5 5 0 0 0 0-10H5zm6 9a4 4 0 1 1 0-8 4 4 0 0 1 0 8z"/></svg>')}.quarto-sidebar-toggle{border-color:#dee2e6;border-bottom-left-radius:.25rem;border-bottom-right-radius:.25rem;border-style:solid;border-width:1px;overflow:hidden;border-top-width:0px;padding-top:0px !important}.quarto-sidebar-toggle-title{cursor:pointer;padding-bottom:2px;margin-left:.25em;text-align:center;font-weight:400;font-size:.775em}#quarto-content .quarto-sidebar-toggle{background:#fdf8e9}#quarto-content .quarto-sidebar-toggle-title{color:#343a40}.quarto-sidebar-toggle-icon{color:#dee2e6;margin-right:.5em;float:right;transition:transform .2s ease}.quarto-sidebar-toggle-icon::before{padding-top:5px}.quarto-sidebar-toggle.expanded .quarto-sidebar-toggle-icon{transform:rotate(-180deg)}.quarto-sidebar-toggle.expanded .quarto-sidebar-toggle-title{border-bottom:solid #dee2e6 1px}.quarto-sidebar-toggle-contents{background-color:#fefbf2;padding-right:10px;padding-left:10px;margin-top:0px !important;transition:max-height .5s ease}.quarto-sidebar-toggle.expanded .quarto-sidebar-toggle-contents{padding-top:1em;padding-bottom:10px}@media(max-width: 767.98px){.sidebar-menu-container{padding-bottom:5em}}.quarto-sidebar-toggle:not(.expanded) .quarto-sidebar-toggle-contents{padding-top:0px !important;padding-bottom:0px}nav[role=doc-toc]{z-index:1020}#quarto-sidebar>*,nav[role=doc-toc]>*{transition:opacity .1s ease,border .1s ease}#quarto-sidebar.slow>*,nav[role=doc-toc].slow>*{transition:opacity .4s ease,border .4s ease}.quarto-color-scheme-toggle:not(.alternate).top-right .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(136, 147, 158, 1)" class="bi bi-toggle-off" viewBox="0 0 16 16"><path d="M11 4a4 4 0 0 1 0 8H8a4.992 4.992 0 0 0 2-4 4.992 4.992 0 0 0-2-4h3zm-6 8a4 4 0 1 1 0-8 4 4 0 0 1 0 8zM0 8a5 5 0 0 0 5 5h6a5 5 0 0 0 0-10H5a5 5 0 0 0-5 5z"/></svg>')}.quarto-color-scheme-toggle.alternate.top-right .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(98, 109, 120, 1)" class="bi bi-toggle-on" viewBox="0 0 16 16"><path d="M5 3a5 5 0 0 0 0 10h6a5 5 0 0 0 0-10H5zm6 9a4 4 0 1 1 0-8 4 4 0 0 1 0 8z"/></svg>')}#quarto-appendix.default{border-top:1px solid #dee2e6}#quarto-appendix.default{background-color:#fefbf2;padding-top:1.5em;margin-top:2em;z-index:998}#quarto-appendix.default .quarto-appendix-heading{margin-top:0;line-height:1.4em;font-weight:600;opacity:.9;border-bottom:none;margin-bottom:0}#quarto-appendix.default .footnotes ol,#quarto-appendix.default .footnotes ol li>p:last-of-type,#quarto-appendix.default .quarto-appendix-contents>p:last-of-type{margin-bottom:0}#quarto-appendix.default .footnotes ol{margin-left:.5em}#quarto-appendix.default .quarto-appendix-secondary-label{margin-bottom:.4em}#quarto-appendix.default .quarto-appendix-bibtex{font-size:.7em;padding:1em;border:solid 1px #dee2e6;margin-bottom:1em}#quarto-appendix.default .quarto-appendix-bibtex code.sourceCode{white-space:pre-wrap}#quarto-appendix.default .quarto-appendix-citeas{font-size:.9em;padding:1em;border:solid 1px #dee2e6;margin-bottom:1em}#quarto-appendix.default .quarto-appendix-heading{font-size:1em !important}#quarto-appendix.default *[role=doc-endnotes]>ol,#quarto-appendix.default .quarto-appendix-contents>*:not(h2):not(.h2){font-size:.9em}#quarto-appendix.default section{padding-bottom:1.5em}#quarto-appendix.default section *[role=doc-endnotes],#quarto-appendix.default section>*:not(a){opacity:.9;word-wrap:break-word}.btn.btn-quarto,div.cell-output-display .btn-quarto{--bs-btn-color: #cacccd;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #cacccd;--bs-btn-hover-bg: #52585d;--bs-btn-hover-border-color: #484e53;--bs-btn-focus-shadow-rgb: 75, 80, 85;--bs-btn-active-color: #fff;--bs-btn-active-bg: #5d6166;--bs-btn-active-border-color: #484e53;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}nav.quarto-secondary-nav.color-navbar{background-color:#052744;color:#f2e5bd}nav.quarto-secondary-nav.color-navbar h1,nav.quarto-secondary-nav.color-navbar .h1,nav.quarto-secondary-nav.color-navbar .quarto-btn-toggle{color:#f2e5bd}@media(max-width: 991.98px){body.nav-sidebar .quarto-title-banner{margin-bottom:0;padding-bottom:1em}body.nav-sidebar #title-block-header{margin-block-end:0}}p.subtitle{margin-top:.25em;margin-bottom:.5em}code a:any-link{color:inherit;text-decoration-color:#6c757d}/*! light */div.observablehq table thead tr th{background-color:var(--bs-body-bg)}input,button,select,optgroup,textarea{background-color:var(--bs-body-bg)}.code-annotated .code-copy-button{margin-right:1.25em;margin-top:0;padding-bottom:0;padding-top:3px}.code-annotation-gutter-bg{background-color:#fefbf2}.code-annotation-gutter{background-color:rgba(233,236,239,.65)}.code-annotation-gutter,.code-annotation-gutter-bg{height:100%;width:calc(20px + .5em);position:absolute;top:0;right:0}dl.code-annotation-container-grid dt{margin-right:1em;margin-top:.25rem}dl.code-annotation-container-grid dt{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;color:#4b545c;border:solid #4b545c 1px;border-radius:50%;height:22px;width:22px;line-height:22px;font-size:11px;text-align:center;vertical-align:middle;text-decoration:none}dl.code-annotation-container-grid dt[data-target-cell]{cursor:pointer}dl.code-annotation-container-grid dt[data-target-cell].code-annotation-active{color:#fefbf2;border:solid #aaa 1px;background-color:#aaa}pre.code-annotation-code{padding-top:0;padding-bottom:0}pre.code-annotation-code code{z-index:3}#code-annotation-line-highlight-gutter{width:100%;border-top:solid rgba(170,170,170,.2666666667) 1px;border-bottom:solid rgba(170,170,170,.2666666667) 1px;z-index:2;background-color:rgba(170,170,170,.1333333333)}#code-annotation-line-highlight{margin-left:-4em;width:calc(100% + 4em);border-top:solid rgba(170,170,170,.2666666667) 1px;border-bottom:solid rgba(170,170,170,.2666666667) 1px;z-index:2;background-color:rgba(170,170,170,.1333333333)}code.sourceCode .code-annotation-anchor.code-annotation-active{background-color:var(--quarto-hl-normal-color, #aaaaaa);border:solid var(--quarto-hl-normal-color, #aaaaaa) 1px;color:#e9ecef;font-weight:bolder}code.sourceCode .code-annotation-anchor{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;color:var(--quarto-hl-co-color);border:solid var(--quarto-hl-co-color) 1px;border-radius:50%;height:18px;width:18px;font-size:9px;margin-top:2px}code.sourceCode button.code-annotation-anchor{padding:2px;user-select:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;-o-user-select:none}code.sourceCode a.code-annotation-anchor{line-height:18px;text-align:center;vertical-align:middle;cursor:default;text-decoration:none}@media print{.page-columns .column-screen-inset{grid-column:page-start-inset/page-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset table{background:#fefbf2}.page-columns .column-screen-inset-left{grid-column:page-start-inset/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-inset-left table{background:#fefbf2}.page-columns .column-screen-inset-right{grid-column:body-content-start/page-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset-right table{background:#fefbf2}.page-columns .column-screen{grid-column:page-start/page-end;z-index:998;opacity:.999}.page-columns .column-screen table{background:#fefbf2}.page-columns .column-screen-left{grid-column:page-start/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-left table{background:#fefbf2}.page-columns .column-screen-right{grid-column:body-content-start/page-end;z-index:998;opacity:.999}.page-columns .column-screen-right table{background:#fefbf2}.page-columns .column-screen-inset-shaded{grid-column:page-start-inset/page-end-inset;padding:1em;background:#f8f9fa;z-index:998;opacity:.999;margin-bottom:1em}}.quarto-video{margin-bottom:1em}.table{border-top:1px solid #d6d4ce;border-bottom:1px solid #d6d4ce}.table>thead{border-top-width:0;border-bottom:1px solid #999b99}.table a{word-break:break-word}.table>:not(caption)>*>*{background-color:unset;color:unset}#quarto-document-content .crosstalk-input .checkbox input[type=checkbox],#quarto-document-content .crosstalk-input .checkbox-inline input[type=checkbox]{position:unset;margin-top:unset;margin-left:unset}#quarto-document-content .row{margin-left:unset;margin-right:unset}.quarto-xref{white-space:nowrap}#quarto-draft-alert{margin-top:0px;margin-bottom:0px;padding:.3em;text-align:center;font-size:.9em}#quarto-draft-alert i{margin-right:.3em}#quarto-back-to-top{z-index:1000}pre{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;font-size:0.875em;font-weight:400}pre code{font-family:inherit;font-size:inherit;font-weight:inherit}code{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;font-size:0.875em;font-weight:400}a{background-color:rgba(0,0,0,0);font-weight:400;text-decoration:underline}a.external:after{content:"";background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(39, 97, 227)" class="bi bi-box-arrow-up-right" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M8.636 3.5a.5.5 0 0 0-.5-.5H1.5A1.5 1.5 0 0 0 0 4.5v10A1.5 1.5 0 0 0 1.5 16h10a1.5 1.5 0 0 0 1.5-1.5V7.864a.5.5 0 0 0-1 0V14.5a.5.5 0 0 1-.5.5h-10a.5.5 0 0 1-.5-.5v-10a.5.5 0 0 1 .5-.5h6.636a.5.5 0 0 0 .5-.5z"/><path fill-rule="evenodd" d="M16 .5a.5.5 0 0 0-.5-.5h-5a.5.5 0 0 0 0 1h3.793L6.146 9.146a.5.5 0 1 0 .708.708L15 1.707V5.5a.5.5 0 0 0 1 0v-5z"/></svg>');background-size:contain;background-repeat:no-repeat;background-position:center center;margin-left:.2em;padding-right:.75em}div.sourceCode code a.external:after{content:none}a.external:after:hover{cursor:pointer}.quarto-ext-icon{display:inline-block;font-size:.75em;padding-left:.3em}.code-with-filename .code-with-filename-file{margin-bottom:0;padding-bottom:2px;padding-top:2px;padding-left:.7em;border:var(--quarto-border-width) solid var(--quarto-border-color);border-radius:var(--quarto-border-radius);border-bottom:0;border-bottom-left-radius:0%;border-bottom-right-radius:0%}.code-with-filename div.sourceCode,.reveal .code-with-filename div.sourceCode{margin-top:0;border-top-left-radius:0%;border-top-right-radius:0%}.code-with-filename .code-with-filename-file pre{margin-bottom:0}.code-with-filename .code-with-filename-file{background-color:rgba(219,219,219,.8)}.quarto-dark .code-with-filename .code-with-filename-file{background-color:#555}.code-with-filename .code-with-filename-file strong{font-weight:400}.quarto-title-banner{margin-bottom:1em;color:#f2e5bd;background:#052744}.quarto-title-banner a{color:#f2e5bd}.quarto-title-banner h1,.quarto-title-banner .h1,.quarto-title-banner h2,.quarto-title-banner .h2{color:#f2e5bd}.quarto-title-banner .code-tools-button{color:#e1c368}.quarto-title-banner .code-tools-button:hover{color:#f2e5bd}.quarto-title-banner .code-tools-button>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(225, 195, 104)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>')}.quarto-title-banner .code-tools-button:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(242, 229, 189)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>')}.quarto-title-banner .quarto-title .title{font-weight:600}.quarto-title-banner .quarto-categories{margin-top:.75em}@media(min-width: 992px){.quarto-title-banner{padding-top:2.5em;padding-bottom:2.5em}}@media(max-width: 991.98px){.quarto-title-banner{padding-top:1em;padding-bottom:1em}}@media(max-width: 767.98px){body.hypothesis-enabled #title-block-header>*{padding-right:20px}}main.quarto-banner-title-block>section:first-child>h2,main.quarto-banner-title-block>section:first-child>.h2,main.quarto-banner-title-block>section:first-child>h3,main.quarto-banner-title-block>section:first-child>.h3,main.quarto-banner-title-block>section:first-child>h4,main.quarto-banner-title-block>section:first-child>.h4{margin-top:0}.quarto-title .quarto-categories{display:flex;flex-wrap:wrap;row-gap:.5em;column-gap:.4em;padding-bottom:.5em;margin-top:.75em}.quarto-title .quarto-categories .quarto-category{padding:.25em .75em;font-size:.65em;text-transform:uppercase;border:solid 1px;border-radius:.25rem;opacity:.6}.quarto-title .quarto-categories .quarto-category a{color:inherit}.quarto-title-meta-container{display:grid;grid-template-columns:1fr auto}.quarto-title-meta-column-end{display:flex;flex-direction:column;padding-left:1em}.quarto-title-meta-column-end a .bi{margin-right:.3em}#title-block-header.quarto-title-block.default .quarto-title-meta{display:grid;grid-template-columns:repeat(2, 1fr);grid-column-gap:1em}#title-block-header.quarto-title-block.default .quarto-title .title{margin-bottom:0}#title-block-header.quarto-title-block.default .quarto-title-author-orcid img{margin-top:-0.2em;height:.8em;width:.8em}#title-block-header.quarto-title-block.default .quarto-title-author-email{opacity:.7}#title-block-header.quarto-title-block.default .quarto-description p:last-of-type{margin-bottom:0}#title-block-header.quarto-title-block.default .quarto-title-meta-contents p,#title-block-header.quarto-title-block.default .quarto-title-authors p,#title-block-header.quarto-title-block.default .quarto-title-affiliations p{margin-bottom:.1em}#title-block-header.quarto-title-block.default .quarto-title-meta-heading{text-transform:uppercase;margin-top:1em;font-size:.8em;opacity:.8;font-weight:400}#title-block-header.quarto-title-block.default .quarto-title-meta-contents{font-size:.9em}#title-block-header.quarto-title-block.default .quarto-title-meta-contents p.affiliation:last-of-type{margin-bottom:.1em}#title-block-header.quarto-title-block.default p.affiliation{margin-bottom:.1em}#title-block-header.quarto-title-block.default .keywords,#title-block-header.quarto-title-block.default .description,#title-block-header.quarto-title-block.default .abstract{margin-top:0}#title-block-header.quarto-title-block.default .keywords>p,#title-block-header.quarto-title-block.default .description>p,#title-block-header.quarto-title-block.default .abstract>p{font-size:.9em}#title-block-header.quarto-title-block.default .keywords>p:last-of-type,#title-block-header.quarto-title-block.default .description>p:last-of-type,#title-block-header.quarto-title-block.default .abstract>p:last-of-type{margin-bottom:0}#title-block-header.quarto-title-block.default .keywords .block-title,#title-block-header.quarto-title-block.default .description .block-title,#title-block-header.quarto-title-block.default .abstract .block-title{margin-top:1em;text-transform:uppercase;font-size:.8em;opacity:.8;font-weight:400}#title-block-header.quarto-title-block.default .quarto-title-meta-author{display:grid;grid-template-columns:minmax(max-content, 1fr) 1fr;grid-column-gap:1em}.quarto-title-tools-only{display:flex;justify-content:right}body{-webkit-font-smoothing:antialiased}.badge.bg-light{color:#343a40}.progress .progress-bar{font-size:8px;line-height:8px}:root{--quarto-scss-export-gray-300: #dee2e6;--quarto-scss-export-gray-500: #adb5bd;--quarto-scss-export-gray-600: #6c757d;--quarto-scss-export-gray-800: #343a40;--quarto-scss-export-card-cap-bg: rgba(52, 58, 64, 0.25);--quarto-scss-export-border-color: #dee2e6;--quarto-scss-export-text-muted: #6c757d;--quarto-scss-export-old-quarto-body-bg: #FEFBF2;--quarto-scss-export-body-bg: #FEFBF2;--quarto-scss-export-white: #fff;--quarto-scss-export-gray-100: #f8f9fa;--quarto-scss-export-gray-200: #e9ecef;--quarto-scss-export-gray-400: #ced4da;--quarto-scss-export-gray-700: #495057;--quarto-scss-export-gray-900: #212529;--quarto-scss-export-black: #000;--quarto-scss-export-blue: #2780e3;--quarto-scss-export-indigo: #6610f2;--quarto-scss-export-purple: #613d7c;--quarto-scss-export-pink: #e83e8c;--quarto-scss-export-red: #ff0039;--quarto-scss-export-orange: #f0ad4e;--quarto-scss-export-yellow: #ff7518;--quarto-scss-export-green: #3fb618;--quarto-scss-export-teal: #20c997;--quarto-scss-export-cyan: #9954bb;--quarto-scss-export-primary: #2780e3;--quarto-scss-export-secondary: #343a40;--quarto-scss-export-success: #3fb618;--quarto-scss-export-info: #9954bb;--quarto-scss-export-warning: #ff7518;--quarto-scss-export-danger: #ff0039;--quarto-scss-export-light: #f8f9fa;--quarto-scss-export-dark: #343a40;--quarto-scss-export-body-color: #343a40;--quarto-scss-export-title-banner-color: ;--quarto-scss-export-title-banner-bg: ;--quarto-scss-export-btn-code-copy-color: #8f5902;--quarto-scss-export-btn-code-copy-color-active: #204a87;--quarto-scss-export-link-color: #2761e3;--quarto-scss-export-link-color-bg: transparent;--quarto-scss-export-code-color: #7d12ba;--quarto-scss-export-code-bg: #f8f9fa;--quarto-scss-export-toc-color: #2761e3;--quarto-scss-export-toc-active-border: #2761e3;--quarto-scss-export-toc-inactive-border: #e9ecef;--quarto-scss-export-navbar-default: #2780e3;--quarto-scss-export-navbar-hl-override: #93b0f1;--quarto-scss-export-btn-bg: #343a40;--quarto-scss-export-btn-fg: #cacccd;--quarto-scss-export-body-contrast-bg: #FEFBF2;--quarto-scss-export-body-contrast-color: #343a40;--quarto-scss-export-navbar-hl: #93b0f1;--quarto-scss-export-navbar-brand-hl: #93b0f1;--quarto-scss-export-navbar-toggler-border-color: rgba(242, 229, 189, 0);--quarto-scss-export-navbar-hover-color: rgba(147, 176, 241, 0.8);--quarto-scss-export-navbar-disabled-color: rgba(242, 229, 189, 0.75);--quarto-scss-export-title-block-color: #343a40;--quarto-scss-export-title-block-contast-color: #FEFBF2;--quarto-scss-export-footer-bg: #FEFBF2;--quarto-scss-export-footer-fg: #75736f;--quarto-scss-export-popover-bg: #FEFBF2;--quarto-scss-export-input-bg: #FEFBF2;--quarto-scss-export-input-border-color: #dee2e6;--quarto-scss-export-code-annotation-higlight-color: rgba(170, 170, 170, 0.2666666667);--quarto-scss-export-code-annotation-higlight-bg: rgba(170, 170, 170, 0.1333333333);--quarto-scss-export-table-group-separator-color: #999b99;--quarto-scss-export-table-group-separator-color-lighter: #d6d4ce;--quarto-scss-export-link-decoration: underline;--quarto-scss-export-table-border-color: #dee2e6;--quarto-scss-export-sidebar-glass-bg: rgba(102, 102, 102, 0.4);--quarto-scss-export-color-contrast-dark: #000;--quarto-scss-export-color-contrast-light: #fff;--quarto-scss-export-blue-100: #d4e6f9;--quarto-scss-export-blue-200: #a9ccf4;--quarto-scss-export-blue-300: #7db3ee;--quarto-scss-export-blue-400: #5299e9;--quarto-scss-export-blue-500: #2780e3;--quarto-scss-export-blue-600: #1f66b6;--quarto-scss-export-blue-700: #174d88;--quarto-scss-export-blue-800: #10335b;--quarto-scss-export-blue-900: #081a2d;--quarto-scss-export-indigo-100: #e0cffc;--quarto-scss-export-indigo-200: #c29ffa;--quarto-scss-export-indigo-300: #a370f7;--quarto-scss-export-indigo-400: #8540f5;--quarto-scss-export-indigo-500: #6610f2;--quarto-scss-export-indigo-600: #520dc2;--quarto-scss-export-indigo-700: #3d0a91;--quarto-scss-export-indigo-800: #290661;--quarto-scss-export-indigo-900: #140330;--quarto-scss-export-purple-100: #dfd8e5;--quarto-scss-export-purple-200: #c0b1cb;--quarto-scss-export-purple-300: #a08bb0;--quarto-scss-export-purple-400: #816496;--quarto-scss-export-purple-500: #613d7c;--quarto-scss-export-purple-600: #4e3163;--quarto-scss-export-purple-700: #3a254a;--quarto-scss-export-purple-800: #271832;--quarto-scss-export-purple-900: #130c19;--quarto-scss-export-pink-100: #fad8e8;--quarto-scss-export-pink-200: #f6b2d1;--quarto-scss-export-pink-300: #f18bba;--quarto-scss-export-pink-400: #ed65a3;--quarto-scss-export-pink-500: #e83e8c;--quarto-scss-export-pink-600: #ba3270;--quarto-scss-export-pink-700: #8b2554;--quarto-scss-export-pink-800: #5d1938;--quarto-scss-export-pink-900: #2e0c1c;--quarto-scss-export-red-100: #ffccd7;--quarto-scss-export-red-200: #ff99b0;--quarto-scss-export-red-300: #ff6688;--quarto-scss-export-red-400: #ff3361;--quarto-scss-export-red-500: #ff0039;--quarto-scss-export-red-600: #cc002e;--quarto-scss-export-red-700: #990022;--quarto-scss-export-red-800: #660017;--quarto-scss-export-red-900: #33000b;--quarto-scss-export-orange-100: #fcefdc;--quarto-scss-export-orange-200: #f9deb8;--quarto-scss-export-orange-300: #f6ce95;--quarto-scss-export-orange-400: #f3bd71;--quarto-scss-export-orange-500: #f0ad4e;--quarto-scss-export-orange-600: #c08a3e;--quarto-scss-export-orange-700: #90682f;--quarto-scss-export-orange-800: #60451f;--quarto-scss-export-orange-900: #302310;--quarto-scss-export-yellow-100: #ffe3d1;--quarto-scss-export-yellow-200: #ffc8a3;--quarto-scss-export-yellow-300: #ffac74;--quarto-scss-export-yellow-400: #ff9146;--quarto-scss-export-yellow-500: #ff7518;--quarto-scss-export-yellow-600: #cc5e13;--quarto-scss-export-yellow-700: #99460e;--quarto-scss-export-yellow-800: #662f0a;--quarto-scss-export-yellow-900: #331705;--quarto-scss-export-green-100: #d9f0d1;--quarto-scss-export-green-200: #b2e2a3;--quarto-scss-export-green-300: #8cd374;--quarto-scss-export-green-400: #65c546;--quarto-scss-export-green-500: #3fb618;--quarto-scss-export-green-600: #329213;--quarto-scss-export-green-700: #266d0e;--quarto-scss-export-green-800: #19490a;--quarto-scss-export-green-900: #0d2405;--quarto-scss-export-teal-100: #d2f4ea;--quarto-scss-export-teal-200: #a6e9d5;--quarto-scss-export-teal-300: #79dfc1;--quarto-scss-export-teal-400: #4dd4ac;--quarto-scss-export-teal-500: #20c997;--quarto-scss-export-teal-600: #1aa179;--quarto-scss-export-teal-700: #13795b;--quarto-scss-export-teal-800: #0d503c;--quarto-scss-export-teal-900: #06281e;--quarto-scss-export-cyan-100: #ebddf1;--quarto-scss-export-cyan-200: #d6bbe4;--quarto-scss-export-cyan-300: #c298d6;--quarto-scss-export-cyan-400: #ad76c9;--quarto-scss-export-cyan-500: #9954bb;--quarto-scss-export-cyan-600: #7a4396;--quarto-scss-export-cyan-700: #5c3270;--quarto-scss-export-cyan-800: #3d224b;--quarto-scss-export-cyan-900: #1f1125;--quarto-scss-export-default: #343a40;--quarto-scss-export-primary-text-emphasis: #10335b;--quarto-scss-export-secondary-text-emphasis: #15171a;--quarto-scss-export-success-text-emphasis: #19490a;--quarto-scss-export-info-text-emphasis: #3d224b;--quarto-scss-export-warning-text-emphasis: #662f0a;--quarto-scss-export-danger-text-emphasis: #660017;--quarto-scss-export-light-text-emphasis: #495057;--quarto-scss-export-dark-text-emphasis: #495057;--quarto-scss-export-primary-bg-subtle: #d4e6f9;--quarto-scss-export-secondary-bg-subtle: #d6d8d9;--quarto-scss-export-success-bg-subtle: #d9f0d1;--quarto-scss-export-info-bg-subtle: #ebddf1;--quarto-scss-export-warning-bg-subtle: #ffe3d1;--quarto-scss-export-danger-bg-subtle: #ffccd7;--quarto-scss-export-light-bg-subtle: #fcfcfd;--quarto-scss-export-dark-bg-subtle: #ced4da;--quarto-scss-export-primary-border-subtle: #a9ccf4;--quarto-scss-export-secondary-border-subtle: #aeb0b3;--quarto-scss-export-success-border-subtle: #b2e2a3;--quarto-scss-export-info-border-subtle: #d6bbe4;--quarto-scss-export-warning-border-subtle: #ffc8a3;--quarto-scss-export-danger-border-subtle: #ff99b0;--quarto-scss-export-light-border-subtle: #e9ecef;--quarto-scss-export-dark-border-subtle: #adb5bd;--quarto-scss-export-body-text-align: ;--quarto-scss-export-body-secondary-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-body-secondary-bg: #e9ecef;--quarto-scss-export-body-tertiary-color: rgba(52, 58, 64, 0.5);--quarto-scss-export-body-tertiary-bg: #f8f9fa;--quarto-scss-export-body-emphasis-color: #000;--quarto-scss-export-link-hover-color: #1f4eb6;--quarto-scss-export-link-hover-decoration: ;--quarto-scss-export-border-color-translucent: rgba(0, 0, 0, 0.175);--quarto-scss-export-component-active-bg: #2780e3;--quarto-scss-export-component-active-color: #fff;--quarto-scss-export-focus-ring-color: rgba(39, 128, 227, 0.25);--quarto-scss-export-headings-font-family: ;--quarto-scss-export-headings-font-style: ;--quarto-scss-export-display-font-family: ;--quarto-scss-export-display-font-style: ;--quarto-scss-export-blockquote-footer-color: #6c757d;--quarto-scss-export-blockquote-border-color: #e9ecef;--quarto-scss-export-hr-bg-color: ;--quarto-scss-export-hr-height: ;--quarto-scss-export-hr-border-color: ;--quarto-scss-export-legend-font-weight: ;--quarto-scss-export-mark-bg: #ffe3d1;--quarto-scss-export-table-color: #343a40;--quarto-scss-export-table-bg: #FEFBF2;--quarto-scss-export-table-accent-bg: transparent;--quarto-scss-export-table-th-font-weight: ;--quarto-scss-export-table-striped-color: #343a40;--quarto-scss-export-table-striped-bg: rgba(0, 0, 0, 0.05);--quarto-scss-export-table-active-color: #343a40;--quarto-scss-export-table-active-bg: rgba(0, 0, 0, 0.1);--quarto-scss-export-table-hover-color: #343a40;--quarto-scss-export-table-hover-bg: rgba(0, 0, 0, 0.075);--quarto-scss-export-table-caption-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-input-btn-font-family: ;--quarto-scss-export-input-btn-focus-color: rgba(39, 128, 227, 0.25);--quarto-scss-export-btn-color: #343a40;--quarto-scss-export-btn-font-family: ;--quarto-scss-export-btn-white-space: ;--quarto-scss-export-btn-link-color: #2761e3;--quarto-scss-export-btn-link-hover-color: #1f4eb6;--quarto-scss-export-btn-link-disabled-color: #6c757d;--quarto-scss-export-form-text-font-style: ;--quarto-scss-export-form-text-font-weight: ;--quarto-scss-export-form-text-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-form-label-font-size: ;--quarto-scss-export-form-label-font-style: ;--quarto-scss-export-form-label-font-weight: ;--quarto-scss-export-form-label-color: ;--quarto-scss-export-input-font-family: ;--quarto-scss-export-input-disabled-color: ;--quarto-scss-export-input-disabled-bg: #e9ecef;--quarto-scss-export-input-disabled-border-color: ;--quarto-scss-export-input-color: #343a40;--quarto-scss-export-input-focus-bg: #FEFBF2;--quarto-scss-export-input-focus-border-color: #93c0f1;--quarto-scss-export-input-focus-color: #343a40;--quarto-scss-export-input-placeholder-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-input-plaintext-color: #343a40;--quarto-scss-export-form-check-label-color: ;--quarto-scss-export-form-check-transition: ;--quarto-scss-export-form-check-input-bg: #FEFBF2;--quarto-scss-export-form-check-input-focus-border: #93c0f1;--quarto-scss-export-form-check-input-checked-color: #fff;--quarto-scss-export-form-check-input-checked-bg-color: #2780e3;--quarto-scss-export-form-check-input-checked-border-color: #2780e3;--quarto-scss-export-form-check-input-indeterminate-color: #fff;--quarto-scss-export-form-check-input-indeterminate-bg-color: #2780e3;--quarto-scss-export-form-check-input-indeterminate-border-color: #2780e3;--quarto-scss-export-form-switch-color: rgba(0, 0, 0, 0.25);--quarto-scss-export-form-switch-focus-color: #93c0f1;--quarto-scss-export-form-switch-checked-color: #fff;--quarto-scss-export-input-group-addon-color: #343a40;--quarto-scss-export-input-group-addon-bg: #f8f9fa;--quarto-scss-export-input-group-addon-border-color: #dee2e6;--quarto-scss-export-form-select-font-family: ;--quarto-scss-export-form-select-color: #343a40;--quarto-scss-export-form-select-bg: #FEFBF2;--quarto-scss-export-form-select-disabled-color: ;--quarto-scss-export-form-select-disabled-bg: #e9ecef;--quarto-scss-export-form-select-disabled-border-color: ;--quarto-scss-export-form-select-indicator-color: #343a40;--quarto-scss-export-form-select-border-color: #dee2e6;--quarto-scss-export-form-select-focus-border-color: #93c0f1;--quarto-scss-export-form-range-track-bg: #f8f9fa;--quarto-scss-export-form-range-thumb-bg: #2780e3;--quarto-scss-export-form-range-thumb-active-bg: #bed9f7;--quarto-scss-export-form-range-thumb-disabled-bg: rgba(52, 58, 64, 0.75);--quarto-scss-export-form-file-button-color: #343a40;--quarto-scss-export-form-file-button-bg: #f8f9fa;--quarto-scss-export-form-file-button-hover-bg: #e9ecef;--quarto-scss-export-form-floating-label-disabled-color: #6c757d;--quarto-scss-export-form-feedback-font-style: ;--quarto-scss-export-form-feedback-valid-color: #3fb618;--quarto-scss-export-form-feedback-invalid-color: #ff0039;--quarto-scss-export-form-feedback-icon-valid-color: #3fb618;--quarto-scss-export-form-feedback-icon-invalid-color: #ff0039;--quarto-scss-export-form-valid-color: #3fb618;--quarto-scss-export-form-valid-border-color: #3fb618;--quarto-scss-export-form-invalid-color: #ff0039;--quarto-scss-export-form-invalid-border-color: #ff0039;--quarto-scss-export-nav-link-font-size: ;--quarto-scss-export-nav-link-font-weight: ;--quarto-scss-export-nav-link-color: #2761e3;--quarto-scss-export-nav-link-hover-color: #1f4eb6;--quarto-scss-export-nav-link-disabled-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-nav-tabs-border-color: #dee2e6;--quarto-scss-export-nav-tabs-link-hover-border-color: #e9ecef #e9ecef #dee2e6;--quarto-scss-export-nav-tabs-link-active-color: #000;--quarto-scss-export-nav-tabs-link-active-bg: #FEFBF2;--quarto-scss-export-nav-pills-link-active-bg: #2780e3;--quarto-scss-export-nav-pills-link-active-color: #fff;--quarto-scss-export-nav-underline-link-active-color: #000;--quarto-scss-export-navbar-padding-x: ;--quarto-scss-export-navbar-light-contrast: #fff;--quarto-scss-export-navbar-dark-contrast: #fff;--quarto-scss-export-navbar-light-icon-color: rgba(255, 255, 255, 0.75);--quarto-scss-export-navbar-dark-icon-color: rgba(255, 255, 255, 0.75);--quarto-scss-export-dropdown-color: #343a40;--quarto-scss-export-dropdown-bg: #FEFBF2;--quarto-scss-export-dropdown-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-divider-bg: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-link-color: #343a40;--quarto-scss-export-dropdown-link-hover-color: #343a40;--quarto-scss-export-dropdown-link-hover-bg: #f8f9fa;--quarto-scss-export-dropdown-link-active-bg: #2780e3;--quarto-scss-export-dropdown-link-active-color: #fff;--quarto-scss-export-dropdown-link-disabled-color: rgba(52, 58, 64, 0.5);--quarto-scss-export-dropdown-header-color: #6c757d;--quarto-scss-export-dropdown-dark-color: #dee2e6;--quarto-scss-export-dropdown-dark-bg: #343a40;--quarto-scss-export-dropdown-dark-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-dark-divider-bg: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-dark-box-shadow: ;--quarto-scss-export-dropdown-dark-link-color: #dee2e6;--quarto-scss-export-dropdown-dark-link-hover-color: #fff;--quarto-scss-export-dropdown-dark-link-hover-bg: rgba(255, 255, 255, 0.15);--quarto-scss-export-dropdown-dark-link-active-color: #fff;--quarto-scss-export-dropdown-dark-link-active-bg: #2780e3;--quarto-scss-export-dropdown-dark-link-disabled-color: #adb5bd;--quarto-scss-export-dropdown-dark-header-color: #adb5bd;--quarto-scss-export-pagination-color: #2761e3;--quarto-scss-export-pagination-bg: #FEFBF2;--quarto-scss-export-pagination-border-color: #dee2e6;--quarto-scss-export-pagination-focus-color: #1f4eb6;--quarto-scss-export-pagination-focus-bg: #e9ecef;--quarto-scss-export-pagination-hover-color: #1f4eb6;--quarto-scss-export-pagination-hover-bg: #f8f9fa;--quarto-scss-export-pagination-hover-border-color: #dee2e6;--quarto-scss-export-pagination-active-color: #fff;--quarto-scss-export-pagination-active-bg: #2780e3;--quarto-scss-export-pagination-active-border-color: #2780e3;--quarto-scss-export-pagination-disabled-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-pagination-disabled-bg: #e9ecef;--quarto-scss-export-pagination-disabled-border-color: #dee2e6;--quarto-scss-export-card-title-color: ;--quarto-scss-export-card-subtitle-color: ;--quarto-scss-export-card-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-card-box-shadow: ;--quarto-scss-export-card-cap-color: ;--quarto-scss-export-card-height: ;--quarto-scss-export-card-color: ;--quarto-scss-export-card-bg: #FEFBF2;--quarto-scss-export-accordion-color: #343a40;--quarto-scss-export-accordion-bg: #FEFBF2;--quarto-scss-export-accordion-border-color: #dee2e6;--quarto-scss-export-accordion-button-color: #343a40;--quarto-scss-export-accordion-button-bg: #FEFBF2;--quarto-scss-export-accordion-button-active-bg: #d4e6f9;--quarto-scss-export-accordion-button-active-color: #10335b;--quarto-scss-export-accordion-button-focus-border-color: #93c0f1;--quarto-scss-export-accordion-icon-color: #343a40;--quarto-scss-export-accordion-icon-active-color: #10335b;--quarto-scss-export-tooltip-color: #FEFBF2;--quarto-scss-export-tooltip-bg: #000;--quarto-scss-export-tooltip-margin: ;--quarto-scss-export-tooltip-arrow-color: ;--quarto-scss-export-form-feedback-tooltip-line-height: ;--quarto-scss-export-popover-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-popover-header-bg: #e9ecef;--quarto-scss-export-popover-body-color: #343a40;--quarto-scss-export-popover-arrow-color: #FEFBF2;--quarto-scss-export-popover-arrow-outer-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-toast-color: ;--quarto-scss-export-toast-background-color: rgba(254, 251, 242, 0.85);--quarto-scss-export-toast-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-toast-header-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-toast-header-background-color: rgba(254, 251, 242, 0.85);--quarto-scss-export-toast-header-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-badge-color: #fff;--quarto-scss-export-modal-content-color: ;--quarto-scss-export-modal-content-bg: #FEFBF2;--quarto-scss-export-modal-content-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-modal-backdrop-bg: #000;--quarto-scss-export-modal-header-border-color: #dee2e6;--quarto-scss-export-modal-footer-bg: ;--quarto-scss-export-modal-footer-border-color: #dee2e6;--quarto-scss-export-progress-bg: #e9ecef;--quarto-scss-export-progress-bar-color: #fff;--quarto-scss-export-progress-bar-bg: #2780e3;--quarto-scss-export-list-group-color: #343a40;--quarto-scss-export-list-group-bg: #FEFBF2;--quarto-scss-export-list-group-border-color: #dee2e6;--quarto-scss-export-list-group-hover-bg: #f8f9fa;--quarto-scss-export-list-group-active-bg: #2780e3;--quarto-scss-export-list-group-active-color: #fff;--quarto-scss-export-list-group-active-border-color: #2780e3;--quarto-scss-export-list-group-disabled-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-list-group-disabled-bg: #FEFBF2;--quarto-scss-export-list-group-action-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-list-group-action-hover-color: #000;--quarto-scss-export-list-group-action-active-color: #343a40;--quarto-scss-export-list-group-action-active-bg: #e9ecef;--quarto-scss-export-thumbnail-bg: #FEFBF2;--quarto-scss-export-thumbnail-border-color: #dee2e6;--quarto-scss-export-figure-caption-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-breadcrumb-font-size: ;--quarto-scss-export-breadcrumb-bg: ;--quarto-scss-export-breadcrumb-divider-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-breadcrumb-active-color: rgba(52, 58, 64, 0.75);--quarto-scss-export-breadcrumb-border-radius: ;--quarto-scss-export-carousel-control-color: #fff;--quarto-scss-export-carousel-indicator-active-bg: #fff;--quarto-scss-export-carousel-caption-color: #fff;--quarto-scss-export-carousel-dark-indicator-active-bg: #000;--quarto-scss-export-carousel-dark-caption-color: #000;--quarto-scss-export-btn-close-color: #000;--quarto-scss-export-offcanvas-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-offcanvas-bg-color: #FEFBF2;--quarto-scss-export-offcanvas-color: #343a40;--quarto-scss-export-offcanvas-backdrop-bg: #000;--quarto-scss-export-code-color-dark: white;--quarto-scss-export-kbd-color: #FEFBF2;--quarto-scss-export-kbd-bg: #343a40;--quarto-scss-export-nested-kbd-font-weight: ;--quarto-scss-export-pre-bg: #f8f9fa;--quarto-scss-export-pre-color: #000;--quarto-scss-export-bslib-page-sidebar-title-bg: #052744;--quarto-scss-export-bslib-page-sidebar-title-color: #fff;--quarto-scss-export-bslib-sidebar-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.05);--quarto-scss-export-bslib-sidebar-toggle-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.1);--quarto-scss-export-sidebar-color: #052744;--quarto-scss-export-sidebar-hover-color: rgba(32, 80, 186, 0.8);--quarto-scss-export-sidebar-disabled-color: rgba(5, 39, 68, 0.75);--quarto-scss-export-valuebox-bg-primary: #5397e9;--quarto-scss-export-valuebox-bg-secondary: #343a40;--quarto-scss-export-valuebox-bg-success: #3aa716;--quarto-scss-export-valuebox-bg-info: rgba(153, 84, 187, 0.7019607843);--quarto-scss-export-valuebox-bg-warning: #fa6400;--quarto-scss-export-valuebox-bg-danger: rgba(255, 0, 57, 0.7019607843);--quarto-scss-export-valuebox-bg-light: #f8f9fa;--quarto-scss-export-valuebox-bg-dark: #343a40;--quarto-scss-export-mermaid-bg-color: #FEFBF2;--quarto-scss-export-mermaid-edge-color: #343a40;--quarto-scss-export-mermaid-node-fg-color: #343a40;--quarto-scss-export-mermaid-fg-color: #343a40;--quarto-scss-export-mermaid-fg-color--lighter: #4b545c;--quarto-scss-export-mermaid-fg-color--lightest: #626d78;--quarto-scss-export-mermaid-label-bg-color: #FEFBF2;--quarto-scss-export-mermaid-label-fg-color: #2780e3;--quarto-scss-export-mermaid-node-bg-color: rgba(39, 128, 227, 0.1);--quarto-scss-export-code-block-border-left-color: #dee2e6;--quarto-scss-export-callout-color-note: #2780e3;--quarto-scss-export-callout-color-tip: #3fb618;--quarto-scss-export-callout-color-important: #ff0039;--quarto-scss-export-callout-color-caution: #f0ad4e;--quarto-scss-export-callout-color-warning: #ff7518}
\ No newline at end of file
diff --git a/docs/2_39/site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css b/docs/2_39/site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css
new file mode 100644
index 000000000..05acf2182
--- /dev/null
+++ b/docs/2_39/site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css
@@ -0,0 +1,12 @@
+@import"https://fonts.googleapis.com/css2?family=Source+Sans+Pro:wght@300;400;700&display=swap";:root{--stan-bg: #242424;--stan-highlight: #053A42;--stan-secondary: #AED1E4;--stan-dark: #3E8EBC;--stan-hero: #cee9f7;--stan-hero-bg: #052744}/*!
+ * Bootstrap  v5.3.1 (https://getbootstrap.com/)
+ * Copyright 2011-2023 The Bootstrap Authors
+ * Licensed under MIT (https://github.com/twbs/bootstrap/blob/main/LICENSE)
+ */:root,[data-bs-theme=light]{--bs-blue: #2780e3;--bs-indigo: #6610f2;--bs-purple: #613d7c;--bs-pink: #e83e8c;--bs-red: #ff0039;--bs-orange: #f0ad4e;--bs-yellow: #ff7518;--bs-green: #3fb618;--bs-teal: #20c997;--bs-cyan: #9954bb;--bs-black: #000;--bs-white: #fff;--bs-gray: #6c757d;--bs-gray-dark: #343a40;--bs-gray-100: #f8f9fa;--bs-gray-200: #e9ecef;--bs-gray-300: #dee2e6;--bs-gray-400: #ced4da;--bs-gray-500: #adb5bd;--bs-gray-600: #6c757d;--bs-gray-700: #495057;--bs-gray-800: #343a40;--bs-gray-900: #212529;--bs-default: #343a40;--bs-primary: #2780e3;--bs-secondary: #343a40;--bs-success: #3fb618;--bs-info: #9954bb;--bs-warning: #ff7518;--bs-danger: #ff0039;--bs-light: #525252;--bs-dark: #343a40;--bs-default-rgb: 52, 58, 64;--bs-primary-rgb: 39, 128, 227;--bs-secondary-rgb: 52, 58, 64;--bs-success-rgb: 63, 182, 24;--bs-info-rgb: 153, 84, 187;--bs-warning-rgb: 255, 117, 24;--bs-danger-rgb: 255, 0, 57;--bs-light-rgb: 82, 82, 82;--bs-dark-rgb: 52, 58, 64;--bs-primary-text-emphasis: #10335b;--bs-secondary-text-emphasis: #15171a;--bs-success-text-emphasis: #19490a;--bs-info-text-emphasis: #3d224b;--bs-warning-text-emphasis: #662f0a;--bs-danger-text-emphasis: #660017;--bs-light-text-emphasis: #495057;--bs-dark-text-emphasis: #495057;--bs-primary-bg-subtle: #d4e6f9;--bs-secondary-bg-subtle: #d6d8d9;--bs-success-bg-subtle: #d9f0d1;--bs-info-bg-subtle: #ebddf1;--bs-warning-bg-subtle: #ffe3d1;--bs-danger-bg-subtle: #ffccd7;--bs-light-bg-subtle: #fcfcfd;--bs-dark-bg-subtle: #ced4da;--bs-primary-border-subtle: #a9ccf4;--bs-secondary-border-subtle: #aeb0b3;--bs-success-border-subtle: #b2e2a3;--bs-info-border-subtle: #d6bbe4;--bs-warning-border-subtle: #ffc8a3;--bs-danger-border-subtle: #ff99b0;--bs-light-border-subtle: #e9ecef;--bs-dark-border-subtle: #adb5bd;--bs-white-rgb: 255, 255, 255;--bs-black-rgb: 0, 0, 0;--bs-font-sans-serif: "Source Sans Pro", -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol";--bs-font-monospace: SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", "Courier New", monospace;--bs-gradient: linear-gradient(180deg, rgba(255, 255, 255, 0.15), rgba(255, 255, 255, 0));--bs-root-font-size: 17px;--bs-body-font-family: "Source Sans Pro", -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol";--bs-body-font-size:1rem;--bs-body-font-weight: 400;--bs-body-line-height: 1.5;--bs-body-color: #FAF1E4;--bs-body-color-rgb: 250, 241, 228;--bs-body-bg: #181818;--bs-body-bg-rgb: 24, 24, 24;--bs-emphasis-color: #000;--bs-emphasis-color-rgb: 0, 0, 0;--bs-secondary-color: rgba(250, 241, 228, 0.75);--bs-secondary-color-rgb: 250, 241, 228;--bs-secondary-bg: #e9ecef;--bs-secondary-bg-rgb: 233, 236, 239;--bs-tertiary-color: rgba(250, 241, 228, 0.5);--bs-tertiary-color-rgb: 250, 241, 228;--bs-tertiary-bg: #f8f9fa;--bs-tertiary-bg-rgb: 248, 249, 250;--bs-heading-color: inherit;--bs-link-color: #AED1E4;--bs-link-color-rgb: 174, 209, 228;--bs-link-decoration: underline;--bs-link-hover-color: #8ba7b6;--bs-link-hover-color-rgb: 139, 167, 182;--bs-code-color: #7d12ba;--bs-highlight-bg: #ffe3d1;--bs-border-width: 1px;--bs-border-style: solid;--bs-border-color: #dee2e6;--bs-border-color-translucent: rgba(0, 0, 0, 0.175);--bs-border-radius: 0.25rem;--bs-border-radius-sm: 0.2em;--bs-border-radius-lg: 0.5rem;--bs-border-radius-xl: 1rem;--bs-border-radius-xxl: 2rem;--bs-border-radius-2xl: var(--bs-border-radius-xxl);--bs-border-radius-pill: 50rem;--bs-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-box-shadow-sm: 0 0.125rem 0.25rem rgba(0, 0, 0, 0.075);--bs-box-shadow-lg: 0 1rem 3rem rgba(0, 0, 0, 0.175);--bs-box-shadow-inset: inset 0 1px 2px rgba(0, 0, 0, 0.075);--bs-focus-ring-width: 0.25rem;--bs-focus-ring-opacity: 0.25;--bs-focus-ring-color: rgba(39, 128, 227, 0.25);--bs-form-valid-color: #3fb618;--bs-form-valid-border-color: #3fb618;--bs-form-invalid-color: #ff0039;--bs-form-invalid-border-color: #ff0039}[data-bs-theme=dark]{color-scheme:dark;--bs-body-color: #dee2e6;--bs-body-color-rgb: 222, 226, 230;--bs-body-bg: #212529;--bs-body-bg-rgb: 33, 37, 41;--bs-emphasis-color: #fff;--bs-emphasis-color-rgb: 255, 255, 255;--bs-secondary-color: rgba(222, 226, 230, 0.75);--bs-secondary-color-rgb: 222, 226, 230;--bs-secondary-bg: #343a40;--bs-secondary-bg-rgb: 52, 58, 64;--bs-tertiary-color: rgba(222, 226, 230, 0.5);--bs-tertiary-color-rgb: 222, 226, 230;--bs-tertiary-bg: #2b3035;--bs-tertiary-bg-rgb: 43, 48, 53;--bs-primary-text-emphasis: #7db3ee;--bs-secondary-text-emphasis: #85898c;--bs-success-text-emphasis: #8cd374;--bs-info-text-emphasis: #c298d6;--bs-warning-text-emphasis: #ffac74;--bs-danger-text-emphasis: #ff6688;--bs-light-text-emphasis: #f8f9fa;--bs-dark-text-emphasis: #dee2e6;--bs-primary-bg-subtle: #081a2d;--bs-secondary-bg-subtle: #0a0c0d;--bs-success-bg-subtle: #0d2405;--bs-info-bg-subtle: #1f1125;--bs-warning-bg-subtle: #331705;--bs-danger-bg-subtle: #33000b;--bs-light-bg-subtle: #343a40;--bs-dark-bg-subtle: #1a1d20;--bs-primary-border-subtle: #174d88;--bs-secondary-border-subtle: #1f2326;--bs-success-border-subtle: #266d0e;--bs-info-border-subtle: #5c3270;--bs-warning-border-subtle: #99460e;--bs-danger-border-subtle: #990022;--bs-light-border-subtle: #495057;--bs-dark-border-subtle: #343a40;--bs-heading-color: inherit;--bs-link-color: #7db3ee;--bs-link-hover-color: #97c2f1;--bs-link-color-rgb: 125, 179, 238;--bs-link-hover-color-rgb: 151, 194, 241;--bs-code-color: white;--bs-border-color: #495057;--bs-border-color-translucent: rgba(255, 255, 255, 0.15);--bs-form-valid-color: #8cd374;--bs-form-valid-border-color: #8cd374;--bs-form-invalid-color: #ff6688;--bs-form-invalid-border-color: #ff6688}*,*::before,*::after{box-sizing:border-box}:root{font-size:var(--bs-root-font-size)}body{margin:0;font-family:var(--bs-body-font-family);font-size:var(--bs-body-font-size);font-weight:var(--bs-body-font-weight);line-height:var(--bs-body-line-height);color:var(--bs-body-color);text-align:var(--bs-body-text-align);background-color:var(--bs-body-bg);-webkit-text-size-adjust:100%;-webkit-tap-highlight-color:rgba(0,0,0,0)}hr{margin:1rem 0;color:inherit;border:0;border-top:1px solid;opacity:.25}h6,.h6,h5,.h5,h4,.h4,h3,.h3,h2,.h2,h1,.h1{margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2;color:var(--bs-heading-color)}h1,.h1{font-size:calc(1.325rem + 0.9vw)}@media(min-width: 1200px){h1,.h1{font-size:2rem}}h2,.h2{font-size:calc(1.29rem + 0.48vw)}@media(min-width: 1200px){h2,.h2{font-size:1.65rem}}h3,.h3{font-size:calc(1.27rem + 0.24vw)}@media(min-width: 1200px){h3,.h3{font-size:1.45rem}}h4,.h4{font-size:1.25rem}h5,.h5{font-size:1.1rem}h6,.h6{font-size:1rem}p{margin-top:0;margin-bottom:1rem}abbr[title]{text-decoration:underline dotted;-webkit-text-decoration:underline dotted;-moz-text-decoration:underline dotted;-ms-text-decoration:underline dotted;-o-text-decoration:underline dotted;cursor:help;text-decoration-skip-ink:none}address{margin-bottom:1rem;font-style:normal;line-height:inherit}ol,ul{padding-left:2rem}ol,ul,dl{margin-top:0;margin-bottom:1rem}ol ol,ul ul,ol ul,ul ol{margin-bottom:0}dt{font-weight:700}dd{margin-bottom:.5rem;margin-left:0}blockquote{margin:0 0 1rem;padding:.625rem 1.25rem;border-left:.25rem solid #e9ecef}blockquote p:last-child,blockquote ul:last-child,blockquote ol:last-child{margin-bottom:0}b,strong{font-weight:bolder}small,.small{font-size:0.875em}mark,.mark{padding:.1875em;background-color:var(--bs-highlight-bg)}sub,sup{position:relative;font-size:0.75em;line-height:0;vertical-align:baseline}sub{bottom:-0.25em}sup{top:-0.5em}a{color:rgba(var(--bs-link-color-rgb), var(--bs-link-opacity, 1));text-decoration:underline;-webkit-text-decoration:underline;-moz-text-decoration:underline;-ms-text-decoration:underline;-o-text-decoration:underline}a:hover{--bs-link-color-rgb: var(--bs-link-hover-color-rgb)}a:not([href]):not([class]),a:not([href]):not([class]):hover{color:inherit;text-decoration:none}pre,code,kbd,samp{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;font-size:1em}pre{display:block;margin-top:0;margin-bottom:1rem;overflow:auto;font-size:0.875em;color:#000;background-color:#f8f9fa;line-height:1.5;padding:.5rem;border:1px solid var(--bs-border-color, #dee2e6)}pre code{background-color:rgba(0,0,0,0);font-size:inherit;color:inherit;word-break:normal}code{font-size:0.875em;color:var(--bs-code-color);background-color:#f8f9fa;padding:.125rem .25rem;word-wrap:break-word}a>code{color:inherit}kbd{padding:.4rem .4rem;font-size:0.875em;color:#181818;background-color:#faf1e4}kbd kbd{padding:0;font-size:1em}figure{margin:0 0 1rem}img,svg{vertical-align:middle}table{caption-side:bottom;border-collapse:collapse}caption{padding-top:.5rem;padding-bottom:.5rem;color:rgba(250,241,228,.75);text-align:left}th{text-align:inherit;text-align:-webkit-match-parent}thead,tbody,tfoot,tr,td,th{border-color:inherit;border-style:solid;border-width:0}label{display:inline-block}button{border-radius:0}button:focus:not(:focus-visible){outline:0}input,button,select,optgroup,textarea{margin:0;font-family:inherit;font-size:inherit;line-height:inherit}button,select{text-transform:none}[role=button]{cursor:pointer}select{word-wrap:normal}select:disabled{opacity:1}[list]:not([type=date]):not([type=datetime-local]):not([type=month]):not([type=week]):not([type=time])::-webkit-calendar-picker-indicator{display:none !important}button,[type=button],[type=reset],[type=submit]{-webkit-appearance:button}button:not(:disabled),[type=button]:not(:disabled),[type=reset]:not(:disabled),[type=submit]:not(:disabled){cursor:pointer}::-moz-focus-inner{padding:0;border-style:none}textarea{resize:vertical}fieldset{min-width:0;padding:0;margin:0;border:0}legend{float:left;width:100%;padding:0;margin-bottom:.5rem;font-size:calc(1.275rem + 0.3vw);line-height:inherit}@media(min-width: 1200px){legend{font-size:1.5rem}}legend+*{clear:left}::-webkit-datetime-edit-fields-wrapper,::-webkit-datetime-edit-text,::-webkit-datetime-edit-minute,::-webkit-datetime-edit-hour-field,::-webkit-datetime-edit-day-field,::-webkit-datetime-edit-month-field,::-webkit-datetime-edit-year-field{padding:0}::-webkit-inner-spin-button{height:auto}[type=search]{-webkit-appearance:textfield;outline-offset:-2px}::-webkit-search-decoration{-webkit-appearance:none}::-webkit-color-swatch-wrapper{padding:0}::file-selector-button{font:inherit;-webkit-appearance:button}output{display:inline-block}iframe{border:0}summary{display:list-item;cursor:pointer}progress{vertical-align:baseline}[hidden]{display:none !important}.lead{font-size:1.25rem;font-weight:300}.display-1{font-size:calc(1.625rem + 4.5vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-1{font-size:5rem}}.display-2{font-size:calc(1.575rem + 3.9vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-2{font-size:4.5rem}}.display-3{font-size:calc(1.525rem + 3.3vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-3{font-size:4rem}}.display-4{font-size:calc(1.475rem + 2.7vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-4{font-size:3.5rem}}.display-5{font-size:calc(1.425rem + 2.1vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-5{font-size:3rem}}.display-6{font-size:calc(1.375rem + 1.5vw);font-weight:300;line-height:1.2}@media(min-width: 1200px){.display-6{font-size:2.5rem}}.list-unstyled{padding-left:0;list-style:none}.list-inline{padding-left:0;list-style:none}.list-inline-item{display:inline-block}.list-inline-item:not(:last-child){margin-right:.5rem}.initialism{font-size:0.875em;text-transform:uppercase}.blockquote{margin-bottom:1rem;font-size:1.25rem}.blockquote>:last-child{margin-bottom:0}.blockquote-footer{margin-top:-1rem;margin-bottom:1rem;font-size:0.875em;color:#6c757d}.blockquote-footer::before{content:"— "}.img-fluid{max-width:100%;height:auto}.img-thumbnail{padding:.25rem;background-color:#181818;border:1px solid #dee2e6;max-width:100%;height:auto}.figure{display:inline-block}.figure-img{margin-bottom:.5rem;line-height:1}.figure-caption{font-size:0.875em;color:rgba(250,241,228,.75)}.container,.container-fluid,.container-xxl,.container-xl,.container-lg,.container-md,.container-sm{--bs-gutter-x: 1.5rem;--bs-gutter-y: 0;width:100%;padding-right:calc(var(--bs-gutter-x)*.5);padding-left:calc(var(--bs-gutter-x)*.5);margin-right:auto;margin-left:auto}@media(min-width: 576px){.container-sm,.container{max-width:540px}}@media(min-width: 768px){.container-md,.container-sm,.container{max-width:720px}}@media(min-width: 992px){.container-lg,.container-md,.container-sm,.container{max-width:960px}}@media(min-width: 1200px){.container-xl,.container-lg,.container-md,.container-sm,.container{max-width:1140px}}@media(min-width: 1400px){.container-xxl,.container-xl,.container-lg,.container-md,.container-sm,.container{max-width:1320px}}:root{--bs-breakpoint-xs: 0;--bs-breakpoint-sm: 576px;--bs-breakpoint-md: 768px;--bs-breakpoint-lg: 992px;--bs-breakpoint-xl: 1200px;--bs-breakpoint-xxl: 1400px}.grid{display:grid;grid-template-rows:repeat(var(--bs-rows, 1), 1fr);grid-template-columns:repeat(var(--bs-columns, 12), 1fr);gap:var(--bs-gap, 1.5rem)}.grid .g-col-1{grid-column:auto/span 1}.grid .g-col-2{grid-column:auto/span 2}.grid .g-col-3{grid-column:auto/span 3}.grid .g-col-4{grid-column:auto/span 4}.grid .g-col-5{grid-column:auto/span 5}.grid .g-col-6{grid-column:auto/span 6}.grid .g-col-7{grid-column:auto/span 7}.grid .g-col-8{grid-column:auto/span 8}.grid .g-col-9{grid-column:auto/span 9}.grid .g-col-10{grid-column:auto/span 10}.grid .g-col-11{grid-column:auto/span 11}.grid .g-col-12{grid-column:auto/span 12}.grid .g-start-1{grid-column-start:1}.grid .g-start-2{grid-column-start:2}.grid .g-start-3{grid-column-start:3}.grid .g-start-4{grid-column-start:4}.grid .g-start-5{grid-column-start:5}.grid .g-start-6{grid-column-start:6}.grid .g-start-7{grid-column-start:7}.grid .g-start-8{grid-column-start:8}.grid .g-start-9{grid-column-start:9}.grid .g-start-10{grid-column-start:10}.grid .g-start-11{grid-column-start:11}@media(min-width: 576px){.grid .g-col-sm-1{grid-column:auto/span 1}.grid .g-col-sm-2{grid-column:auto/span 2}.grid .g-col-sm-3{grid-column:auto/span 3}.grid .g-col-sm-4{grid-column:auto/span 4}.grid .g-col-sm-5{grid-column:auto/span 5}.grid .g-col-sm-6{grid-column:auto/span 6}.grid .g-col-sm-7{grid-column:auto/span 7}.grid .g-col-sm-8{grid-column:auto/span 8}.grid .g-col-sm-9{grid-column:auto/span 9}.grid .g-col-sm-10{grid-column:auto/span 10}.grid .g-col-sm-11{grid-column:auto/span 11}.grid .g-col-sm-12{grid-column:auto/span 12}.grid .g-start-sm-1{grid-column-start:1}.grid .g-start-sm-2{grid-column-start:2}.grid .g-start-sm-3{grid-column-start:3}.grid .g-start-sm-4{grid-column-start:4}.grid .g-start-sm-5{grid-column-start:5}.grid .g-start-sm-6{grid-column-start:6}.grid .g-start-sm-7{grid-column-start:7}.grid .g-start-sm-8{grid-column-start:8}.grid .g-start-sm-9{grid-column-start:9}.grid .g-start-sm-10{grid-column-start:10}.grid .g-start-sm-11{grid-column-start:11}}@media(min-width: 768px){.grid .g-col-md-1{grid-column:auto/span 1}.grid .g-col-md-2{grid-column:auto/span 2}.grid .g-col-md-3{grid-column:auto/span 3}.grid .g-col-md-4{grid-column:auto/span 4}.grid .g-col-md-5{grid-column:auto/span 5}.grid .g-col-md-6{grid-column:auto/span 6}.grid .g-col-md-7{grid-column:auto/span 7}.grid .g-col-md-8{grid-column:auto/span 8}.grid .g-col-md-9{grid-column:auto/span 9}.grid .g-col-md-10{grid-column:auto/span 10}.grid .g-col-md-11{grid-column:auto/span 11}.grid .g-col-md-12{grid-column:auto/span 12}.grid .g-start-md-1{grid-column-start:1}.grid .g-start-md-2{grid-column-start:2}.grid .g-start-md-3{grid-column-start:3}.grid .g-start-md-4{grid-column-start:4}.grid .g-start-md-5{grid-column-start:5}.grid .g-start-md-6{grid-column-start:6}.grid .g-start-md-7{grid-column-start:7}.grid .g-start-md-8{grid-column-start:8}.grid .g-start-md-9{grid-column-start:9}.grid .g-start-md-10{grid-column-start:10}.grid .g-start-md-11{grid-column-start:11}}@media(min-width: 992px){.grid .g-col-lg-1{grid-column:auto/span 1}.grid .g-col-lg-2{grid-column:auto/span 2}.grid .g-col-lg-3{grid-column:auto/span 3}.grid .g-col-lg-4{grid-column:auto/span 4}.grid .g-col-lg-5{grid-column:auto/span 5}.grid .g-col-lg-6{grid-column:auto/span 6}.grid .g-col-lg-7{grid-column:auto/span 7}.grid .g-col-lg-8{grid-column:auto/span 8}.grid .g-col-lg-9{grid-column:auto/span 9}.grid .g-col-lg-10{grid-column:auto/span 10}.grid .g-col-lg-11{grid-column:auto/span 11}.grid .g-col-lg-12{grid-column:auto/span 12}.grid .g-start-lg-1{grid-column-start:1}.grid .g-start-lg-2{grid-column-start:2}.grid .g-start-lg-3{grid-column-start:3}.grid .g-start-lg-4{grid-column-start:4}.grid .g-start-lg-5{grid-column-start:5}.grid .g-start-lg-6{grid-column-start:6}.grid .g-start-lg-7{grid-column-start:7}.grid .g-start-lg-8{grid-column-start:8}.grid .g-start-lg-9{grid-column-start:9}.grid .g-start-lg-10{grid-column-start:10}.grid .g-start-lg-11{grid-column-start:11}}@media(min-width: 1200px){.grid .g-col-xl-1{grid-column:auto/span 1}.grid .g-col-xl-2{grid-column:auto/span 2}.grid .g-col-xl-3{grid-column:auto/span 3}.grid .g-col-xl-4{grid-column:auto/span 4}.grid .g-col-xl-5{grid-column:auto/span 5}.grid .g-col-xl-6{grid-column:auto/span 6}.grid .g-col-xl-7{grid-column:auto/span 7}.grid .g-col-xl-8{grid-column:auto/span 8}.grid .g-col-xl-9{grid-column:auto/span 9}.grid .g-col-xl-10{grid-column:auto/span 10}.grid .g-col-xl-11{grid-column:auto/span 11}.grid .g-col-xl-12{grid-column:auto/span 12}.grid .g-start-xl-1{grid-column-start:1}.grid .g-start-xl-2{grid-column-start:2}.grid .g-start-xl-3{grid-column-start:3}.grid .g-start-xl-4{grid-column-start:4}.grid .g-start-xl-5{grid-column-start:5}.grid .g-start-xl-6{grid-column-start:6}.grid .g-start-xl-7{grid-column-start:7}.grid .g-start-xl-8{grid-column-start:8}.grid .g-start-xl-9{grid-column-start:9}.grid .g-start-xl-10{grid-column-start:10}.grid .g-start-xl-11{grid-column-start:11}}@media(min-width: 1400px){.grid .g-col-xxl-1{grid-column:auto/span 1}.grid .g-col-xxl-2{grid-column:auto/span 2}.grid .g-col-xxl-3{grid-column:auto/span 3}.grid .g-col-xxl-4{grid-column:auto/span 4}.grid .g-col-xxl-5{grid-column:auto/span 5}.grid .g-col-xxl-6{grid-column:auto/span 6}.grid .g-col-xxl-7{grid-column:auto/span 7}.grid .g-col-xxl-8{grid-column:auto/span 8}.grid .g-col-xxl-9{grid-column:auto/span 9}.grid .g-col-xxl-10{grid-column:auto/span 10}.grid .g-col-xxl-11{grid-column:auto/span 11}.grid .g-col-xxl-12{grid-column:auto/span 12}.grid .g-start-xxl-1{grid-column-start:1}.grid .g-start-xxl-2{grid-column-start:2}.grid .g-start-xxl-3{grid-column-start:3}.grid .g-start-xxl-4{grid-column-start:4}.grid .g-start-xxl-5{grid-column-start:5}.grid .g-start-xxl-6{grid-column-start:6}.grid .g-start-xxl-7{grid-column-start:7}.grid .g-start-xxl-8{grid-column-start:8}.grid .g-start-xxl-9{grid-column-start:9}.grid .g-start-xxl-10{grid-column-start:10}.grid .g-start-xxl-11{grid-column-start:11}}.table{--bs-table-color-type: initial;--bs-table-bg-type: initial;--bs-table-color-state: initial;--bs-table-bg-state: initial;--bs-table-color: #FAF1E4;--bs-table-bg: #181818;--bs-table-border-color: #dee2e6;--bs-table-accent-bg: transparent;--bs-table-striped-color: #FAF1E4;--bs-table-striped-bg: rgba(0, 0, 0, 0.05);--bs-table-active-color: #FAF1E4;--bs-table-active-bg: rgba(0, 0, 0, 0.1);--bs-table-hover-color: #FAF1E4;--bs-table-hover-bg: rgba(0, 0, 0, 0.075);width:100%;margin-bottom:1rem;vertical-align:top;border-color:var(--bs-table-border-color)}.table>:not(caption)>*>*{padding:.5rem .5rem;color:var(--bs-table-color-state, var(--bs-table-color-type, var(--bs-table-color)));background-color:var(--bs-table-bg);border-bottom-width:1px;box-shadow:inset 0 0 0 9999px var(--bs-table-bg-state, var(--bs-table-bg-type, var(--bs-table-accent-bg)))}.table>tbody{vertical-align:inherit}.table>thead{vertical-align:bottom}.table-group-divider{border-top:calc(1px*2) solid #89857e}.caption-top{caption-side:top}.table-sm>:not(caption)>*>*{padding:.25rem .25rem}.table-bordered>:not(caption)>*{border-width:1px 0}.table-bordered>:not(caption)>*>*{border-width:0 1px}.table-borderless>:not(caption)>*>*{border-bottom-width:0}.table-borderless>:not(:first-child){border-top-width:0}.table-striped>tbody>tr:nth-of-type(odd)>*{--bs-table-color-type: var(--bs-table-striped-color);--bs-table-bg-type: var(--bs-table-striped-bg)}.table-striped-columns>:not(caption)>tr>:nth-child(even){--bs-table-color-type: var(--bs-table-striped-color);--bs-table-bg-type: var(--bs-table-striped-bg)}.table-active{--bs-table-color-state: var(--bs-table-active-color);--bs-table-bg-state: var(--bs-table-active-bg)}.table-hover>tbody>tr:hover>*{--bs-table-color-state: var(--bs-table-hover-color);--bs-table-bg-state: var(--bs-table-hover-bg)}.table-primary{--bs-table-color: #000;--bs-table-bg: #d4e6f9;--bs-table-border-color: #bfcfe0;--bs-table-striped-bg: #c9dbed;--bs-table-striped-color: #000;--bs-table-active-bg: #bfcfe0;--bs-table-active-color: #000;--bs-table-hover-bg: #c4d5e6;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-secondary{--bs-table-color: #000;--bs-table-bg: #d6d8d9;--bs-table-border-color: #c1c2c3;--bs-table-striped-bg: #cbcdce;--bs-table-striped-color: #000;--bs-table-active-bg: #c1c2c3;--bs-table-active-color: #000;--bs-table-hover-bg: #c6c8c9;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-success{--bs-table-color: #000;--bs-table-bg: #d9f0d1;--bs-table-border-color: #c3d8bc;--bs-table-striped-bg: #cee4c7;--bs-table-striped-color: #000;--bs-table-active-bg: #c3d8bc;--bs-table-active-color: #000;--bs-table-hover-bg: #c9dec1;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-info{--bs-table-color: #000;--bs-table-bg: #ebddf1;--bs-table-border-color: #d4c7d9;--bs-table-striped-bg: #dfd2e5;--bs-table-striped-color: #000;--bs-table-active-bg: #d4c7d9;--bs-table-active-color: #000;--bs-table-hover-bg: #d9ccdf;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-warning{--bs-table-color: #000;--bs-table-bg: #ffe3d1;--bs-table-border-color: #e6ccbc;--bs-table-striped-bg: #f2d8c7;--bs-table-striped-color: #000;--bs-table-active-bg: #e6ccbc;--bs-table-active-color: #000;--bs-table-hover-bg: #ecd2c1;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-danger{--bs-table-color: #000;--bs-table-bg: #ffccd7;--bs-table-border-color: #e6b8c2;--bs-table-striped-bg: #f2c2cc;--bs-table-striped-color: #000;--bs-table-active-bg: #e6b8c2;--bs-table-active-color: #000;--bs-table-hover-bg: #ecbdc7;--bs-table-hover-color: #000;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-light{--bs-table-color: #fff;--bs-table-bg: #525252;--bs-table-border-color: #636363;--bs-table-striped-bg: #5b5b5b;--bs-table-striped-color: #fff;--bs-table-active-bg: #636363;--bs-table-active-color: #fff;--bs-table-hover-bg: #5f5f5f;--bs-table-hover-color: #fff;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-dark{--bs-table-color: #fff;--bs-table-bg: #343a40;--bs-table-border-color: #484e53;--bs-table-striped-bg: #3e444a;--bs-table-striped-color: #fff;--bs-table-active-bg: #484e53;--bs-table-active-color: #fff;--bs-table-hover-bg: #43494e;--bs-table-hover-color: #fff;color:var(--bs-table-color);border-color:var(--bs-table-border-color)}.table-responsive{overflow-x:auto;-webkit-overflow-scrolling:touch}@media(max-width: 575.98px){.table-responsive-sm{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 767.98px){.table-responsive-md{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 991.98px){.table-responsive-lg{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 1199.98px){.table-responsive-xl{overflow-x:auto;-webkit-overflow-scrolling:touch}}@media(max-width: 1399.98px){.table-responsive-xxl{overflow-x:auto;-webkit-overflow-scrolling:touch}}.form-label,.shiny-input-container .control-label{margin-bottom:.5rem}.col-form-label{padding-top:calc(0.375rem + 1px);padding-bottom:calc(0.375rem + 1px);margin-bottom:0;font-size:inherit;line-height:1.5}.col-form-label-lg{padding-top:calc(0.5rem + 1px);padding-bottom:calc(0.5rem + 1px);font-size:1.25rem}.col-form-label-sm{padding-top:calc(0.25rem + 1px);padding-bottom:calc(0.25rem + 1px);font-size:0.875rem}.form-text{margin-top:.25rem;font-size:0.875em;color:rgba(250,241,228,.75)}.form-control{display:block;width:100%;padding:.375rem .75rem;font-size:1rem;font-weight:400;line-height:1.5;color:#faf1e4;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#242424;background-clip:padding-box;border:1px solid #dee2e6;border-radius:0;transition:border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-control{transition:none}}.form-control[type=file]{overflow:hidden}.form-control[type=file]:not(:disabled):not([readonly]){cursor:pointer}.form-control:focus{color:#faf1e4;background-color:#242424;border-color:#93c0f1;outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.form-control::-webkit-date-and-time-value{min-width:85px;height:1.5em;margin:0}.form-control::-webkit-datetime-edit{display:block;padding:0}.form-control::placeholder{color:rgba(250,241,228,.75);opacity:1}.form-control:disabled{background-color:#e9ecef;opacity:1}.form-control::file-selector-button{padding:.375rem .75rem;margin:-0.375rem -0.75rem;margin-inline-end:.75rem;color:#faf1e4;background-color:#f8f9fa;pointer-events:none;border-color:inherit;border-style:solid;border-width:0;border-inline-end-width:1px;border-radius:0;transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-control::file-selector-button{transition:none}}.form-control:hover:not(:disabled):not([readonly])::file-selector-button{background-color:#e9ecef}.form-control-plaintext{display:block;width:100%;padding:.375rem 0;margin-bottom:0;line-height:1.5;color:#faf1e4;background-color:rgba(0,0,0,0);border:solid rgba(0,0,0,0);border-width:1px 0}.form-control-plaintext:focus{outline:0}.form-control-plaintext.form-control-sm,.form-control-plaintext.form-control-lg{padding-right:0;padding-left:0}.form-control-sm{min-height:calc(1.5em + 0.5rem + calc(1px * 2));padding:.25rem .5rem;font-size:0.875rem}.form-control-sm::file-selector-button{padding:.25rem .5rem;margin:-0.25rem -0.5rem;margin-inline-end:.5rem}.form-control-lg{min-height:calc(1.5em + 1rem + calc(1px * 2));padding:.5rem 1rem;font-size:1.25rem}.form-control-lg::file-selector-button{padding:.5rem 1rem;margin:-0.5rem -1rem;margin-inline-end:1rem}textarea.form-control{min-height:calc(1.5em + 0.75rem + calc(1px * 2))}textarea.form-control-sm{min-height:calc(1.5em + 0.5rem + calc(1px * 2))}textarea.form-control-lg{min-height:calc(1.5em + 1rem + calc(1px * 2))}.form-control-color{width:3rem;height:calc(1.5em + 0.75rem + calc(1px * 2));padding:.375rem}.form-control-color:not(:disabled):not([readonly]){cursor:pointer}.form-control-color::-moz-color-swatch{border:0 !important}.form-control-color::-webkit-color-swatch{border:0 !important}.form-control-color.form-control-sm{height:calc(1.5em + 0.5rem + calc(1px * 2))}.form-control-color.form-control-lg{height:calc(1.5em + 1rem + calc(1px * 2))}.form-select{--bs-form-select-bg-img: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16'%3e%3cpath fill='none' stroke='%23343a40' stroke-linecap='round' stroke-linejoin='round' stroke-width='2' d='m2 5 6 6 6-6'/%3e%3c/svg%3e");display:block;width:100%;padding:.375rem 2.25rem .375rem .75rem;font-size:1rem;font-weight:400;line-height:1.5;color:#faf1e4;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#242424;background-image:var(--bs-form-select-bg-img),var(--bs-form-select-bg-icon, none);background-repeat:no-repeat;background-position:right .75rem center;background-size:16px 12px;border:1px solid #dee2e6;border-radius:0;transition:border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-select{transition:none}}.form-select:focus{border-color:#93c0f1;outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.form-select[multiple],.form-select[size]:not([size="1"]){padding-right:.75rem;background-image:none}.form-select:disabled{background-color:#e9ecef}.form-select:-moz-focusring{color:rgba(0,0,0,0);text-shadow:0 0 0 #faf1e4}.form-select-sm{padding-top:.25rem;padding-bottom:.25rem;padding-left:.5rem;font-size:0.875rem}.form-select-lg{padding-top:.5rem;padding-bottom:.5rem;padding-left:1rem;font-size:1.25rem}[data-bs-theme=dark] .form-select{--bs-form-select-bg-img: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16'%3e%3cpath fill='none' stroke='%23dee2e6' stroke-linecap='round' stroke-linejoin='round' stroke-width='2' d='m2 5 6 6 6-6'/%3e%3c/svg%3e")}.form-check,.shiny-input-container .checkbox,.shiny-input-container .radio{display:block;min-height:1.5rem;padding-left:0;margin-bottom:.125rem}.form-check .form-check-input,.form-check .shiny-input-container .checkbox input,.form-check .shiny-input-container .radio input,.shiny-input-container .checkbox .form-check-input,.shiny-input-container .checkbox .shiny-input-container .checkbox input,.shiny-input-container .checkbox .shiny-input-container .radio input,.shiny-input-container .radio .form-check-input,.shiny-input-container .radio .shiny-input-container .checkbox input,.shiny-input-container .radio .shiny-input-container .radio input{float:left;margin-left:0}.form-check-reverse{padding-right:0;padding-left:0;text-align:right}.form-check-reverse .form-check-input{float:right;margin-right:0;margin-left:0}.form-check-input,.shiny-input-container .checkbox input,.shiny-input-container .checkbox-inline input,.shiny-input-container .radio input,.shiny-input-container .radio-inline input{--bs-form-check-bg: #242424;width:1em;height:1em;margin-top:.25em;vertical-align:top;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:var(--bs-form-check-bg);background-image:var(--bs-form-check-bg-image);background-repeat:no-repeat;background-position:center;background-size:contain;border:1px solid #dee2e6;print-color-adjust:exact}.form-check-input[type=radio],.shiny-input-container .checkbox input[type=radio],.shiny-input-container .checkbox-inline input[type=radio],.shiny-input-container .radio input[type=radio],.shiny-input-container .radio-inline input[type=radio]{border-radius:50%}.form-check-input:active,.shiny-input-container .checkbox input:active,.shiny-input-container .checkbox-inline input:active,.shiny-input-container .radio input:active,.shiny-input-container .radio-inline input:active{filter:brightness(90%)}.form-check-input:focus,.shiny-input-container .checkbox input:focus,.shiny-input-container .checkbox-inline input:focus,.shiny-input-container .radio input:focus,.shiny-input-container .radio-inline input:focus{border-color:#93c0f1;outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.form-check-input:checked,.shiny-input-container .checkbox input:checked,.shiny-input-container .checkbox-inline input:checked,.shiny-input-container .radio input:checked,.shiny-input-container .radio-inline input:checked{background-color:#2780e3;border-color:#2780e3}.form-check-input:checked[type=checkbox],.shiny-input-container .checkbox input:checked[type=checkbox],.shiny-input-container .checkbox-inline input:checked[type=checkbox],.shiny-input-container .radio input:checked[type=checkbox],.shiny-input-container .radio-inline input:checked[type=checkbox]{--bs-form-check-bg-image: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 20 20'%3e%3cpath fill='none' stroke='%23fff' stroke-linecap='round' stroke-linejoin='round' stroke-width='3' d='m6 10 3 3 6-6'/%3e%3c/svg%3e")}.form-check-input:checked[type=radio],.shiny-input-container .checkbox input:checked[type=radio],.shiny-input-container .checkbox-inline input:checked[type=radio],.shiny-input-container .radio input:checked[type=radio],.shiny-input-container .radio-inline input:checked[type=radio]{--bs-form-check-bg-image: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='2' fill='%23fff'/%3e%3c/svg%3e")}.form-check-input[type=checkbox]:indeterminate,.shiny-input-container .checkbox input[type=checkbox]:indeterminate,.shiny-input-container .checkbox-inline input[type=checkbox]:indeterminate,.shiny-input-container .radio input[type=checkbox]:indeterminate,.shiny-input-container .radio-inline input[type=checkbox]:indeterminate{background-color:#2780e3;border-color:#2780e3;--bs-form-check-bg-image: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 20 20'%3e%3cpath fill='none' stroke='%23fff' stroke-linecap='round' stroke-linejoin='round' stroke-width='3' d='M6 10h8'/%3e%3c/svg%3e")}.form-check-input:disabled,.shiny-input-container .checkbox input:disabled,.shiny-input-container .checkbox-inline input:disabled,.shiny-input-container .radio input:disabled,.shiny-input-container .radio-inline input:disabled{pointer-events:none;filter:none;opacity:.5}.form-check-input[disabled]~.form-check-label,.form-check-input[disabled]~span,.form-check-input:disabled~.form-check-label,.form-check-input:disabled~span,.shiny-input-container .checkbox input[disabled]~.form-check-label,.shiny-input-container .checkbox input[disabled]~span,.shiny-input-container .checkbox input:disabled~.form-check-label,.shiny-input-container .checkbox input:disabled~span,.shiny-input-container .checkbox-inline input[disabled]~.form-check-label,.shiny-input-container .checkbox-inline input[disabled]~span,.shiny-input-container .checkbox-inline input:disabled~.form-check-label,.shiny-input-container .checkbox-inline input:disabled~span,.shiny-input-container .radio input[disabled]~.form-check-label,.shiny-input-container .radio input[disabled]~span,.shiny-input-container .radio input:disabled~.form-check-label,.shiny-input-container .radio input:disabled~span,.shiny-input-container .radio-inline input[disabled]~.form-check-label,.shiny-input-container .radio-inline input[disabled]~span,.shiny-input-container .radio-inline input:disabled~.form-check-label,.shiny-input-container .radio-inline input:disabled~span{cursor:default;opacity:.5}.form-check-label,.shiny-input-container .checkbox label,.shiny-input-container .checkbox-inline label,.shiny-input-container .radio label,.shiny-input-container .radio-inline label{cursor:pointer}.form-switch{padding-left:2.5em}.form-switch .form-check-input{--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='rgba%280, 0, 0, 0.25%29'/%3e%3c/svg%3e");width:2em;margin-left:-2.5em;background-image:var(--bs-form-switch-bg);background-position:left center;transition:background-position .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-switch .form-check-input{transition:none}}.form-switch .form-check-input:focus{--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='%2393c0f1'/%3e%3c/svg%3e")}.form-switch .form-check-input:checked{background-position:right center;--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='%23fff'/%3e%3c/svg%3e")}.form-switch.form-check-reverse{padding-right:2.5em;padding-left:0}.form-switch.form-check-reverse .form-check-input{margin-right:-2.5em;margin-left:0}.form-check-inline{display:inline-block;margin-right:1rem}.btn-check{position:absolute;clip:rect(0, 0, 0, 0);pointer-events:none}.btn-check[disabled]+.btn,.btn-check:disabled+.btn{pointer-events:none;filter:none;opacity:.65}[data-bs-theme=dark] .form-switch .form-check-input:not(:checked):not(:focus){--bs-form-switch-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='-4 -4 8 8'%3e%3ccircle r='3' fill='rgba%28255, 255, 255, 0.25%29'/%3e%3c/svg%3e")}.form-range{width:100%;height:1.5rem;padding:0;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:rgba(0,0,0,0)}.form-range:focus{outline:0}.form-range:focus::-webkit-slider-thumb{box-shadow:0 0 0 1px #181818,0 0 0 .25rem rgba(39,128,227,.25)}.form-range:focus::-moz-range-thumb{box-shadow:0 0 0 1px #181818,0 0 0 .25rem rgba(39,128,227,.25)}.form-range::-moz-focus-outer{border:0}.form-range::-webkit-slider-thumb{width:1rem;height:1rem;margin-top:-0.25rem;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#2780e3;border:0;transition:background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-range::-webkit-slider-thumb{transition:none}}.form-range::-webkit-slider-thumb:active{background-color:#bed9f7}.form-range::-webkit-slider-runnable-track{width:100%;height:.5rem;color:rgba(0,0,0,0);cursor:pointer;background-color:#f8f9fa;border-color:rgba(0,0,0,0)}.form-range::-moz-range-thumb{width:1rem;height:1rem;appearance:none;-webkit-appearance:none;-moz-appearance:none;-ms-appearance:none;-o-appearance:none;background-color:#2780e3;border:0;transition:background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.form-range::-moz-range-thumb{transition:none}}.form-range::-moz-range-thumb:active{background-color:#bed9f7}.form-range::-moz-range-track{width:100%;height:.5rem;color:rgba(0,0,0,0);cursor:pointer;background-color:#f8f9fa;border-color:rgba(0,0,0,0)}.form-range:disabled{pointer-events:none}.form-range:disabled::-webkit-slider-thumb{background-color:rgba(250,241,228,.75)}.form-range:disabled::-moz-range-thumb{background-color:rgba(250,241,228,.75)}.form-floating{position:relative}.form-floating>.form-control,.form-floating>.form-control-plaintext,.form-floating>.form-select{height:calc(3.5rem + calc(1px * 2));min-height:calc(3.5rem + calc(1px * 2));line-height:1.25}.form-floating>label{position:absolute;top:0;left:0;z-index:2;height:100%;padding:1rem .75rem;overflow:hidden;text-align:start;text-overflow:ellipsis;white-space:nowrap;pointer-events:none;border:1px solid rgba(0,0,0,0);transform-origin:0 0;transition:opacity .1s ease-in-out,transform .1s ease-in-out}@media(prefers-reduced-motion: reduce){.form-floating>label{transition:none}}.form-floating>.form-control,.form-floating>.form-control-plaintext{padding:1rem .75rem}.form-floating>.form-control::placeholder,.form-floating>.form-control-plaintext::placeholder{color:rgba(0,0,0,0)}.form-floating>.form-control:focus,.form-floating>.form-control:not(:placeholder-shown),.form-floating>.form-control-plaintext:focus,.form-floating>.form-control-plaintext:not(:placeholder-shown){padding-top:1.625rem;padding-bottom:.625rem}.form-floating>.form-control:-webkit-autofill,.form-floating>.form-control-plaintext:-webkit-autofill{padding-top:1.625rem;padding-bottom:.625rem}.form-floating>.form-select{padding-top:1.625rem;padding-bottom:.625rem}.form-floating>.form-control:focus~label,.form-floating>.form-control:not(:placeholder-shown)~label,.form-floating>.form-control-plaintext~label,.form-floating>.form-select~label{color:rgba(var(--bs-body-color-rgb), 0.65);transform:scale(0.85) translateY(-0.5rem) translateX(0.15rem)}.form-floating>.form-control:focus~label::after,.form-floating>.form-control:not(:placeholder-shown)~label::after,.form-floating>.form-control-plaintext~label::after,.form-floating>.form-select~label::after{position:absolute;inset:1rem .375rem;z-index:-1;height:1.5em;content:"";background-color:#242424}.form-floating>.form-control:-webkit-autofill~label{color:rgba(var(--bs-body-color-rgb), 0.65);transform:scale(0.85) translateY(-0.5rem) translateX(0.15rem)}.form-floating>.form-control-plaintext~label{border-width:1px 0}.form-floating>:disabled~label,.form-floating>.form-control:disabled~label{color:#6c757d}.form-floating>:disabled~label::after,.form-floating>.form-control:disabled~label::after{background-color:#e9ecef}.input-group{position:relative;display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;align-items:stretch;-webkit-align-items:stretch;width:100%}.input-group>.form-control,.input-group>.form-select,.input-group>.form-floating{position:relative;flex:1 1 auto;-webkit-flex:1 1 auto;width:1%;min-width:0}.input-group>.form-control:focus,.input-group>.form-select:focus,.input-group>.form-floating:focus-within{z-index:5}.input-group .btn{position:relative;z-index:2}.input-group .btn:focus{z-index:5}.input-group-text{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;padding:.375rem .75rem;font-size:1rem;font-weight:400;line-height:1.5;color:#faf1e4;text-align:center;white-space:nowrap;background-color:#f8f9fa;border:1px solid #dee2e6}.input-group-lg>.form-control,.input-group-lg>.form-select,.input-group-lg>.input-group-text,.input-group-lg>.btn{padding:.5rem 1rem;font-size:1.25rem}.input-group-sm>.form-control,.input-group-sm>.form-select,.input-group-sm>.input-group-text,.input-group-sm>.btn{padding:.25rem .5rem;font-size:0.875rem}.input-group-lg>.form-select,.input-group-sm>.form-select{padding-right:3rem}.input-group>:not(:first-child):not(.dropdown-menu):not(.valid-tooltip):not(.valid-feedback):not(.invalid-tooltip):not(.invalid-feedback){margin-left:calc(1px*-1)}.valid-feedback{display:none;width:100%;margin-top:.25rem;font-size:0.875em;color:#3fb618}.valid-tooltip{position:absolute;top:100%;z-index:5;display:none;max-width:100%;padding:.25rem .5rem;margin-top:.1rem;font-size:0.875rem;color:#fff;background-color:#3fb618}.was-validated :valid~.valid-feedback,.was-validated :valid~.valid-tooltip,.is-valid~.valid-feedback,.is-valid~.valid-tooltip{display:block}.was-validated .form-control:valid,.form-control.is-valid{border-color:#3fb618;padding-right:calc(1.5em + 0.75rem);background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 8 8'%3e%3cpath fill='%233fb618' d='M2.3 6.73.6 4.53c-.4-1.04.46-1.4 1.1-.8l1.1 1.4 3.4-3.8c.6-.63 1.6-.27 1.2.7l-4 4.6c-.43.5-.8.4-1.1.1z'/%3e%3c/svg%3e");background-repeat:no-repeat;background-position:right calc(0.375em + 0.1875rem) center;background-size:calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-control:valid:focus,.form-control.is-valid:focus{border-color:#3fb618;box-shadow:0 0 0 .25rem rgba(63,182,24,.25)}.was-validated textarea.form-control:valid,textarea.form-control.is-valid{padding-right:calc(1.5em + 0.75rem);background-position:top calc(0.375em + 0.1875rem) right calc(0.375em + 0.1875rem)}.was-validated .form-select:valid,.form-select.is-valid{border-color:#3fb618}.was-validated .form-select:valid:not([multiple]):not([size]),.was-validated .form-select:valid:not([multiple])[size="1"],.form-select.is-valid:not([multiple]):not([size]),.form-select.is-valid:not([multiple])[size="1"]{--bs-form-select-bg-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 8 8'%3e%3cpath fill='%233fb618' d='M2.3 6.73.6 4.53c-.4-1.04.46-1.4 1.1-.8l1.1 1.4 3.4-3.8c.6-.63 1.6-.27 1.2.7l-4 4.6c-.43.5-.8.4-1.1.1z'/%3e%3c/svg%3e");padding-right:4.125rem;background-position:right .75rem center,center right 2.25rem;background-size:16px 12px,calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-select:valid:focus,.form-select.is-valid:focus{border-color:#3fb618;box-shadow:0 0 0 .25rem rgba(63,182,24,.25)}.was-validated .form-control-color:valid,.form-control-color.is-valid{width:calc(3rem + calc(1.5em + 0.75rem))}.was-validated .form-check-input:valid,.form-check-input.is-valid{border-color:#3fb618}.was-validated .form-check-input:valid:checked,.form-check-input.is-valid:checked{background-color:#3fb618}.was-validated .form-check-input:valid:focus,.form-check-input.is-valid:focus{box-shadow:0 0 0 .25rem rgba(63,182,24,.25)}.was-validated .form-check-input:valid~.form-check-label,.form-check-input.is-valid~.form-check-label{color:#3fb618}.form-check-inline .form-check-input~.valid-feedback{margin-left:.5em}.was-validated .input-group>.form-control:not(:focus):valid,.input-group>.form-control:not(:focus).is-valid,.was-validated .input-group>.form-select:not(:focus):valid,.input-group>.form-select:not(:focus).is-valid,.was-validated .input-group>.form-floating:not(:focus-within):valid,.input-group>.form-floating:not(:focus-within).is-valid{z-index:3}.invalid-feedback{display:none;width:100%;margin-top:.25rem;font-size:0.875em;color:#ff0039}.invalid-tooltip{position:absolute;top:100%;z-index:5;display:none;max-width:100%;padding:.25rem .5rem;margin-top:.1rem;font-size:0.875rem;color:#fff;background-color:#ff0039}.was-validated :invalid~.invalid-feedback,.was-validated :invalid~.invalid-tooltip,.is-invalid~.invalid-feedback,.is-invalid~.invalid-tooltip{display:block}.was-validated .form-control:invalid,.form-control.is-invalid{border-color:#ff0039;padding-right:calc(1.5em + 0.75rem);background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 12 12' width='12' height='12' fill='none' stroke='%23ff0039'%3e%3ccircle cx='6' cy='6' r='4.5'/%3e%3cpath stroke-linejoin='round' d='M5.8 3.6h.4L6 6.5z'/%3e%3ccircle cx='6' cy='8.2' r='.6' fill='%23ff0039' stroke='none'/%3e%3c/svg%3e");background-repeat:no-repeat;background-position:right calc(0.375em + 0.1875rem) center;background-size:calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-control:invalid:focus,.form-control.is-invalid:focus{border-color:#ff0039;box-shadow:0 0 0 .25rem rgba(255,0,57,.25)}.was-validated textarea.form-control:invalid,textarea.form-control.is-invalid{padding-right:calc(1.5em + 0.75rem);background-position:top calc(0.375em + 0.1875rem) right calc(0.375em + 0.1875rem)}.was-validated .form-select:invalid,.form-select.is-invalid{border-color:#ff0039}.was-validated .form-select:invalid:not([multiple]):not([size]),.was-validated .form-select:invalid:not([multiple])[size="1"],.form-select.is-invalid:not([multiple]):not([size]),.form-select.is-invalid:not([multiple])[size="1"]{--bs-form-select-bg-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 12 12' width='12' height='12' fill='none' stroke='%23ff0039'%3e%3ccircle cx='6' cy='6' r='4.5'/%3e%3cpath stroke-linejoin='round' d='M5.8 3.6h.4L6 6.5z'/%3e%3ccircle cx='6' cy='8.2' r='.6' fill='%23ff0039' stroke='none'/%3e%3c/svg%3e");padding-right:4.125rem;background-position:right .75rem center,center right 2.25rem;background-size:16px 12px,calc(0.75em + 0.375rem) calc(0.75em + 0.375rem)}.was-validated .form-select:invalid:focus,.form-select.is-invalid:focus{border-color:#ff0039;box-shadow:0 0 0 .25rem rgba(255,0,57,.25)}.was-validated .form-control-color:invalid,.form-control-color.is-invalid{width:calc(3rem + calc(1.5em + 0.75rem))}.was-validated .form-check-input:invalid,.form-check-input.is-invalid{border-color:#ff0039}.was-validated .form-check-input:invalid:checked,.form-check-input.is-invalid:checked{background-color:#ff0039}.was-validated .form-check-input:invalid:focus,.form-check-input.is-invalid:focus{box-shadow:0 0 0 .25rem rgba(255,0,57,.25)}.was-validated .form-check-input:invalid~.form-check-label,.form-check-input.is-invalid~.form-check-label{color:#ff0039}.form-check-inline .form-check-input~.invalid-feedback{margin-left:.5em}.was-validated .input-group>.form-control:not(:focus):invalid,.input-group>.form-control:not(:focus).is-invalid,.was-validated .input-group>.form-select:not(:focus):invalid,.input-group>.form-select:not(:focus).is-invalid,.was-validated .input-group>.form-floating:not(:focus-within):invalid,.input-group>.form-floating:not(:focus-within).is-invalid{z-index:4}.btn{--bs-btn-padding-x: 0.75rem;--bs-btn-padding-y: 0.375rem;--bs-btn-font-family: ;--bs-btn-font-size:1rem;--bs-btn-font-weight: 400;--bs-btn-line-height: 1.5;--bs-btn-color: #FAF1E4;--bs-btn-bg: transparent;--bs-btn-border-width: 1px;--bs-btn-border-color: transparent;--bs-btn-border-radius: 0.25rem;--bs-btn-hover-border-color: transparent;--bs-btn-box-shadow: inset 0 1px 0 rgba(255, 255, 255, 0.15), 0 1px 1px rgba(0, 0, 0, 0.075);--bs-btn-disabled-opacity: 0.65;--bs-btn-focus-box-shadow: 0 0 0 0.25rem rgba(var(--bs-btn-focus-shadow-rgb), .5);display:inline-block;padding:var(--bs-btn-padding-y) var(--bs-btn-padding-x);font-family:var(--bs-btn-font-family);font-size:var(--bs-btn-font-size);font-weight:var(--bs-btn-font-weight);line-height:var(--bs-btn-line-height);color:var(--bs-btn-color);text-align:center;text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;vertical-align:middle;cursor:pointer;user-select:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;-o-user-select:none;border:var(--bs-btn-border-width) solid var(--bs-btn-border-color);background-color:var(--bs-btn-bg);transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.btn{transition:none}}.btn:hover{color:var(--bs-btn-hover-color);background-color:var(--bs-btn-hover-bg);border-color:var(--bs-btn-hover-border-color)}.btn-check+.btn:hover{color:var(--bs-btn-color);background-color:var(--bs-btn-bg);border-color:var(--bs-btn-border-color)}.btn:focus-visible{color:var(--bs-btn-hover-color);background-color:var(--bs-btn-hover-bg);border-color:var(--bs-btn-hover-border-color);outline:0;box-shadow:var(--bs-btn-focus-box-shadow)}.btn-check:focus-visible+.btn{border-color:var(--bs-btn-hover-border-color);outline:0;box-shadow:var(--bs-btn-focus-box-shadow)}.btn-check:checked+.btn,:not(.btn-check)+.btn:active,.btn:first-child:active,.btn.active,.btn.show{color:var(--bs-btn-active-color);background-color:var(--bs-btn-active-bg);border-color:var(--bs-btn-active-border-color)}.btn-check:checked+.btn:focus-visible,:not(.btn-check)+.btn:active:focus-visible,.btn:first-child:active:focus-visible,.btn.active:focus-visible,.btn.show:focus-visible{box-shadow:var(--bs-btn-focus-box-shadow)}.btn:disabled,.btn.disabled,fieldset:disabled .btn{color:var(--bs-btn-disabled-color);pointer-events:none;background-color:var(--bs-btn-disabled-bg);border-color:var(--bs-btn-disabled-border-color);opacity:var(--bs-btn-disabled-opacity)}.btn-default{--bs-btn-color: #fff;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #2c3136;--bs-btn-hover-border-color: #2a2e33;--bs-btn-focus-shadow-rgb: 82, 88, 93;--bs-btn-active-color: #fff;--bs-btn-active-bg: #2a2e33;--bs-btn-active-border-color: #272c30;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}.btn-primary{--bs-btn-color: #fff;--bs-btn-bg: #2780e3;--bs-btn-border-color: #2780e3;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #216dc1;--bs-btn-hover-border-color: #1f66b6;--bs-btn-focus-shadow-rgb: 71, 147, 231;--bs-btn-active-color: #fff;--bs-btn-active-bg: #1f66b6;--bs-btn-active-border-color: #1d60aa;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #2780e3;--bs-btn-disabled-border-color: #2780e3}.btn-secondary{--bs-btn-color: #fff;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #2c3136;--bs-btn-hover-border-color: #2a2e33;--bs-btn-focus-shadow-rgb: 82, 88, 93;--bs-btn-active-color: #fff;--bs-btn-active-bg: #2a2e33;--bs-btn-active-border-color: #272c30;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}.btn-success{--bs-btn-color: #fff;--bs-btn-bg: #3fb618;--bs-btn-border-color: #3fb618;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #369b14;--bs-btn-hover-border-color: #329213;--bs-btn-focus-shadow-rgb: 92, 193, 59;--bs-btn-active-color: #fff;--bs-btn-active-bg: #329213;--bs-btn-active-border-color: #2f8912;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #3fb618;--bs-btn-disabled-border-color: #3fb618}.btn-info{--bs-btn-color: #fff;--bs-btn-bg: #9954bb;--bs-btn-border-color: #9954bb;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #82479f;--bs-btn-hover-border-color: #7a4396;--bs-btn-focus-shadow-rgb: 168, 110, 197;--bs-btn-active-color: #fff;--bs-btn-active-bg: #7a4396;--bs-btn-active-border-color: #733f8c;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #9954bb;--bs-btn-disabled-border-color: #9954bb}.btn-warning{--bs-btn-color: #fff;--bs-btn-bg: #ff7518;--bs-btn-border-color: #ff7518;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #d96314;--bs-btn-hover-border-color: #cc5e13;--bs-btn-focus-shadow-rgb: 255, 138, 59;--bs-btn-active-color: #fff;--bs-btn-active-bg: #cc5e13;--bs-btn-active-border-color: #bf5812;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #ff7518;--bs-btn-disabled-border-color: #ff7518}.btn-danger{--bs-btn-color: #fff;--bs-btn-bg: #ff0039;--bs-btn-border-color: #ff0039;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #d90030;--bs-btn-hover-border-color: #cc002e;--bs-btn-focus-shadow-rgb: 255, 38, 87;--bs-btn-active-color: #fff;--bs-btn-active-bg: #cc002e;--bs-btn-active-border-color: #bf002b;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #ff0039;--bs-btn-disabled-border-color: #ff0039}.btn-light{--bs-btn-color: #fff;--bs-btn-bg: #525252;--bs-btn-border-color: #525252;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #464646;--bs-btn-hover-border-color: #424242;--bs-btn-focus-shadow-rgb: 108, 108, 108;--bs-btn-active-color: #fff;--bs-btn-active-bg: #424242;--bs-btn-active-border-color: #3e3e3e;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #525252;--bs-btn-disabled-border-color: #525252}.btn-dark{--bs-btn-color: #fff;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #52585d;--bs-btn-hover-border-color: #484e53;--bs-btn-focus-shadow-rgb: 82, 88, 93;--bs-btn-active-color: #fff;--bs-btn-active-bg: #5d6166;--bs-btn-active-border-color: #484e53;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}.btn-outline-default{--bs-btn-color: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #343a40;--bs-btn-hover-border-color: #343a40;--bs-btn-focus-shadow-rgb: 52, 58, 64;--bs-btn-active-color: #fff;--bs-btn-active-bg: #343a40;--bs-btn-active-border-color: #343a40;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #343a40;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #343a40;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-primary{--bs-btn-color: #2780e3;--bs-btn-border-color: #2780e3;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #2780e3;--bs-btn-hover-border-color: #2780e3;--bs-btn-focus-shadow-rgb: 39, 128, 227;--bs-btn-active-color: #fff;--bs-btn-active-bg: #2780e3;--bs-btn-active-border-color: #2780e3;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #2780e3;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #2780e3;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-secondary{--bs-btn-color: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #343a40;--bs-btn-hover-border-color: #343a40;--bs-btn-focus-shadow-rgb: 52, 58, 64;--bs-btn-active-color: #fff;--bs-btn-active-bg: #343a40;--bs-btn-active-border-color: #343a40;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #343a40;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #343a40;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-success{--bs-btn-color: #3fb618;--bs-btn-border-color: #3fb618;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #3fb618;--bs-btn-hover-border-color: #3fb618;--bs-btn-focus-shadow-rgb: 63, 182, 24;--bs-btn-active-color: #fff;--bs-btn-active-bg: #3fb618;--bs-btn-active-border-color: #3fb618;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #3fb618;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #3fb618;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-info{--bs-btn-color: #9954bb;--bs-btn-border-color: #9954bb;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #9954bb;--bs-btn-hover-border-color: #9954bb;--bs-btn-focus-shadow-rgb: 153, 84, 187;--bs-btn-active-color: #fff;--bs-btn-active-bg: #9954bb;--bs-btn-active-border-color: #9954bb;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #9954bb;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #9954bb;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-warning{--bs-btn-color: #ff7518;--bs-btn-border-color: #ff7518;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #ff7518;--bs-btn-hover-border-color: #ff7518;--bs-btn-focus-shadow-rgb: 255, 117, 24;--bs-btn-active-color: #fff;--bs-btn-active-bg: #ff7518;--bs-btn-active-border-color: #ff7518;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #ff7518;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #ff7518;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-danger{--bs-btn-color: #ff0039;--bs-btn-border-color: #ff0039;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #ff0039;--bs-btn-hover-border-color: #ff0039;--bs-btn-focus-shadow-rgb: 255, 0, 57;--bs-btn-active-color: #fff;--bs-btn-active-bg: #ff0039;--bs-btn-active-border-color: #ff0039;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #ff0039;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #ff0039;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-light{--bs-btn-color: #525252;--bs-btn-border-color: #525252;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #525252;--bs-btn-hover-border-color: #525252;--bs-btn-focus-shadow-rgb: 82, 82, 82;--bs-btn-active-color: #fff;--bs-btn-active-bg: #525252;--bs-btn-active-border-color: #525252;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #525252;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #525252;--bs-btn-bg: transparent;--bs-gradient: none}.btn-outline-dark{--bs-btn-color: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #fff;--bs-btn-hover-bg: #343a40;--bs-btn-hover-border-color: #343a40;--bs-btn-focus-shadow-rgb: 52, 58, 64;--bs-btn-active-color: #fff;--bs-btn-active-bg: #343a40;--bs-btn-active-border-color: #343a40;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #343a40;--bs-btn-disabled-bg: transparent;--bs-btn-disabled-border-color: #343a40;--bs-btn-bg: transparent;--bs-gradient: none}.btn-link{--bs-btn-font-weight: 400;--bs-btn-color: #AED1E4;--bs-btn-bg: transparent;--bs-btn-border-color: transparent;--bs-btn-hover-color: #8ba7b6;--bs-btn-hover-border-color: transparent;--bs-btn-active-color: #8ba7b6;--bs-btn-active-border-color: transparent;--bs-btn-disabled-color: #6c757d;--bs-btn-disabled-border-color: transparent;--bs-btn-box-shadow: 0 0 0 #000;--bs-btn-focus-shadow-rgb: 148, 178, 194;text-decoration:underline;-webkit-text-decoration:underline;-moz-text-decoration:underline;-ms-text-decoration:underline;-o-text-decoration:underline}.btn-link:focus-visible{color:var(--bs-btn-color)}.btn-link:hover{color:var(--bs-btn-hover-color)}.btn-lg,.btn-group-lg>.btn{--bs-btn-padding-y: 0.5rem;--bs-btn-padding-x: 1rem;--bs-btn-font-size:1.25rem;--bs-btn-border-radius: 0.5rem}.btn-sm,.btn-group-sm>.btn{--bs-btn-padding-y: 0.25rem;--bs-btn-padding-x: 0.5rem;--bs-btn-font-size:0.875rem;--bs-btn-border-radius: 0.2em}.fade{transition:opacity .15s linear}@media(prefers-reduced-motion: reduce){.fade{transition:none}}.fade:not(.show){opacity:0}.collapse:not(.show){display:none}.collapsing{height:0;overflow:hidden;transition:height .2s ease}@media(prefers-reduced-motion: reduce){.collapsing{transition:none}}.collapsing.collapse-horizontal{width:0;height:auto;transition:width .35s ease}@media(prefers-reduced-motion: reduce){.collapsing.collapse-horizontal{transition:none}}.dropup,.dropend,.dropdown,.dropstart,.dropup-center,.dropdown-center{position:relative}.dropdown-toggle{white-space:nowrap}.dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:"";border-top:.3em solid;border-right:.3em solid rgba(0,0,0,0);border-bottom:0;border-left:.3em solid rgba(0,0,0,0)}.dropdown-toggle:empty::after{margin-left:0}.dropdown-menu{--bs-dropdown-zindex: 1000;--bs-dropdown-min-width: 10rem;--bs-dropdown-padding-x: 0;--bs-dropdown-padding-y: 0.5rem;--bs-dropdown-spacer: 0.125rem;--bs-dropdown-font-size:1rem;--bs-dropdown-color: #FAF1E4;--bs-dropdown-bg: #181818;--bs-dropdown-border-color: rgba(0, 0, 0, 0.175);--bs-dropdown-border-radius: 0.25rem;--bs-dropdown-border-width: 1px;--bs-dropdown-inner-border-radius: calc(0.25rem - 1px);--bs-dropdown-divider-bg: rgba(0, 0, 0, 0.175);--bs-dropdown-divider-margin-y: 0.5rem;--bs-dropdown-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-dropdown-link-color: #FAF1E4;--bs-dropdown-link-hover-color: #FAF1E4;--bs-dropdown-link-hover-bg: #6c757d;--bs-dropdown-link-active-color: #fff;--bs-dropdown-link-active-bg: #2780e3;--bs-dropdown-link-disabled-color: rgba(250, 241, 228, 0.5);--bs-dropdown-item-padding-x: 1rem;--bs-dropdown-item-padding-y: 0.25rem;--bs-dropdown-header-color: #6c757d;--bs-dropdown-header-padding-x: 1rem;--bs-dropdown-header-padding-y: 0.5rem;position:absolute;z-index:var(--bs-dropdown-zindex);display:none;min-width:var(--bs-dropdown-min-width);padding:var(--bs-dropdown-padding-y) var(--bs-dropdown-padding-x);margin:0;font-size:var(--bs-dropdown-font-size);color:var(--bs-dropdown-color);text-align:left;list-style:none;background-color:var(--bs-dropdown-bg);background-clip:padding-box;border:var(--bs-dropdown-border-width) solid var(--bs-dropdown-border-color)}.dropdown-menu[data-bs-popper]{top:100%;left:0;margin-top:var(--bs-dropdown-spacer)}.dropdown-menu-start{--bs-position: start}.dropdown-menu-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-end{--bs-position: end}.dropdown-menu-end[data-bs-popper]{right:0;left:auto}@media(min-width: 576px){.dropdown-menu-sm-start{--bs-position: start}.dropdown-menu-sm-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-sm-end{--bs-position: end}.dropdown-menu-sm-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 768px){.dropdown-menu-md-start{--bs-position: start}.dropdown-menu-md-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-md-end{--bs-position: end}.dropdown-menu-md-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 992px){.dropdown-menu-lg-start{--bs-position: start}.dropdown-menu-lg-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-lg-end{--bs-position: end}.dropdown-menu-lg-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 1200px){.dropdown-menu-xl-start{--bs-position: start}.dropdown-menu-xl-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-xl-end{--bs-position: end}.dropdown-menu-xl-end[data-bs-popper]{right:0;left:auto}}@media(min-width: 1400px){.dropdown-menu-xxl-start{--bs-position: start}.dropdown-menu-xxl-start[data-bs-popper]{right:auto;left:0}.dropdown-menu-xxl-end{--bs-position: end}.dropdown-menu-xxl-end[data-bs-popper]{right:0;left:auto}}.dropup .dropdown-menu[data-bs-popper]{top:auto;bottom:100%;margin-top:0;margin-bottom:var(--bs-dropdown-spacer)}.dropup .dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:"";border-top:0;border-right:.3em solid rgba(0,0,0,0);border-bottom:.3em solid;border-left:.3em solid rgba(0,0,0,0)}.dropup .dropdown-toggle:empty::after{margin-left:0}.dropend .dropdown-menu[data-bs-popper]{top:0;right:auto;left:100%;margin-top:0;margin-left:var(--bs-dropdown-spacer)}.dropend .dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:"";border-top:.3em solid rgba(0,0,0,0);border-right:0;border-bottom:.3em solid rgba(0,0,0,0);border-left:.3em solid}.dropend .dropdown-toggle:empty::after{margin-left:0}.dropend .dropdown-toggle::after{vertical-align:0}.dropstart .dropdown-menu[data-bs-popper]{top:0;right:100%;left:auto;margin-top:0;margin-right:var(--bs-dropdown-spacer)}.dropstart .dropdown-toggle::after{display:inline-block;margin-left:.255em;vertical-align:.255em;content:""}.dropstart .dropdown-toggle::after{display:none}.dropstart .dropdown-toggle::before{display:inline-block;margin-right:.255em;vertical-align:.255em;content:"";border-top:.3em solid rgba(0,0,0,0);border-right:.3em solid;border-bottom:.3em solid rgba(0,0,0,0)}.dropstart .dropdown-toggle:empty::after{margin-left:0}.dropstart .dropdown-toggle::before{vertical-align:0}.dropdown-divider{height:0;margin:var(--bs-dropdown-divider-margin-y) 0;overflow:hidden;border-top:1px solid var(--bs-dropdown-divider-bg);opacity:1}.dropdown-item{display:block;width:100%;padding:var(--bs-dropdown-item-padding-y) var(--bs-dropdown-item-padding-x);clear:both;font-weight:400;color:var(--bs-dropdown-link-color);text-align:inherit;text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;white-space:nowrap;background-color:rgba(0,0,0,0);border:0}.dropdown-item:hover,.dropdown-item:focus{color:var(--bs-dropdown-link-hover-color);background-color:var(--bs-dropdown-link-hover-bg)}.dropdown-item.active,.dropdown-item:active{color:var(--bs-dropdown-link-active-color);text-decoration:none;background-color:var(--bs-dropdown-link-active-bg)}.dropdown-item.disabled,.dropdown-item:disabled{color:var(--bs-dropdown-link-disabled-color);pointer-events:none;background-color:rgba(0,0,0,0)}.dropdown-menu.show{display:block}.dropdown-header{display:block;padding:var(--bs-dropdown-header-padding-y) var(--bs-dropdown-header-padding-x);margin-bottom:0;font-size:0.875rem;color:var(--bs-dropdown-header-color);white-space:nowrap}.dropdown-item-text{display:block;padding:var(--bs-dropdown-item-padding-y) var(--bs-dropdown-item-padding-x);color:var(--bs-dropdown-link-color)}.dropdown-menu-dark{--bs-dropdown-color: #dee2e6;--bs-dropdown-bg: #343a40;--bs-dropdown-border-color: rgba(0, 0, 0, 0.175);--bs-dropdown-box-shadow: ;--bs-dropdown-link-color: #dee2e6;--bs-dropdown-link-hover-color: #fff;--bs-dropdown-divider-bg: rgba(0, 0, 0, 0.175);--bs-dropdown-link-hover-bg: rgba(255, 255, 255, 0.15);--bs-dropdown-link-active-color: #fff;--bs-dropdown-link-active-bg: #2780e3;--bs-dropdown-link-disabled-color: #adb5bd;--bs-dropdown-header-color: #adb5bd}.btn-group,.btn-group-vertical{position:relative;display:inline-flex;vertical-align:middle}.btn-group>.btn,.btn-group-vertical>.btn{position:relative;flex:1 1 auto;-webkit-flex:1 1 auto}.btn-group>.btn-check:checked+.btn,.btn-group>.btn-check:focus+.btn,.btn-group>.btn:hover,.btn-group>.btn:focus,.btn-group>.btn:active,.btn-group>.btn.active,.btn-group-vertical>.btn-check:checked+.btn,.btn-group-vertical>.btn-check:focus+.btn,.btn-group-vertical>.btn:hover,.btn-group-vertical>.btn:focus,.btn-group-vertical>.btn:active,.btn-group-vertical>.btn.active{z-index:1}.btn-toolbar{display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;justify-content:flex-start;-webkit-justify-content:flex-start}.btn-toolbar .input-group{width:auto}.btn-group>:not(.btn-check:first-child)+.btn,.btn-group>.btn-group:not(:first-child){margin-left:calc(1px*-1)}.dropdown-toggle-split{padding-right:.5625rem;padding-left:.5625rem}.dropdown-toggle-split::after,.dropup .dropdown-toggle-split::after,.dropend .dropdown-toggle-split::after{margin-left:0}.dropstart .dropdown-toggle-split::before{margin-right:0}.btn-sm+.dropdown-toggle-split,.btn-group-sm>.btn+.dropdown-toggle-split{padding-right:.375rem;padding-left:.375rem}.btn-lg+.dropdown-toggle-split,.btn-group-lg>.btn+.dropdown-toggle-split{padding-right:.75rem;padding-left:.75rem}.btn-group-vertical{flex-direction:column;-webkit-flex-direction:column;align-items:flex-start;-webkit-align-items:flex-start;justify-content:center;-webkit-justify-content:center}.btn-group-vertical>.btn,.btn-group-vertical>.btn-group{width:100%}.btn-group-vertical>.btn:not(:first-child),.btn-group-vertical>.btn-group:not(:first-child){margin-top:calc(1px*-1)}.nav{--bs-nav-link-padding-x: 1rem;--bs-nav-link-padding-y: 0.5rem;--bs-nav-link-font-weight: ;--bs-nav-link-color: #AED1E4;--bs-nav-link-hover-color: #8ba7b6;--bs-nav-link-disabled-color: rgba(250, 241, 228, 0.75);display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;padding-left:0;margin-bottom:0;list-style:none}.nav-link{display:block;padding:var(--bs-nav-link-padding-y) var(--bs-nav-link-padding-x);font-size:var(--bs-nav-link-font-size);font-weight:var(--bs-nav-link-font-weight);color:var(--bs-nav-link-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;background:none;border:0;transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out}@media(prefers-reduced-motion: reduce){.nav-link{transition:none}}.nav-link:hover,.nav-link:focus{color:var(--bs-nav-link-hover-color)}.nav-link:focus-visible{outline:0;box-shadow:0 0 0 .25rem rgba(39,128,227,.25)}.nav-link.disabled,.nav-link:disabled{color:var(--bs-nav-link-disabled-color);pointer-events:none;cursor:default}.nav-tabs{--bs-nav-tabs-border-width: 1px;--bs-nav-tabs-border-color: #dee2e6;--bs-nav-tabs-border-radius: 0.25rem;--bs-nav-tabs-link-hover-border-color: #e9ecef #e9ecef #dee2e6;--bs-nav-tabs-link-active-color: #000;--bs-nav-tabs-link-active-bg: #181818;--bs-nav-tabs-link-active-border-color: #dee2e6 #dee2e6 #181818;border-bottom:var(--bs-nav-tabs-border-width) solid var(--bs-nav-tabs-border-color)}.nav-tabs .nav-link{margin-bottom:calc(-1*var(--bs-nav-tabs-border-width));border:var(--bs-nav-tabs-border-width) solid rgba(0,0,0,0)}.nav-tabs .nav-link:hover,.nav-tabs .nav-link:focus{isolation:isolate;border-color:var(--bs-nav-tabs-link-hover-border-color)}.nav-tabs .nav-link.active,.nav-tabs .nav-item.show .nav-link{color:var(--bs-nav-tabs-link-active-color);background-color:var(--bs-nav-tabs-link-active-bg);border-color:var(--bs-nav-tabs-link-active-border-color)}.nav-tabs .dropdown-menu{margin-top:calc(-1*var(--bs-nav-tabs-border-width))}.nav-pills{--bs-nav-pills-border-radius: 0.25rem;--bs-nav-pills-link-active-color: #fff;--bs-nav-pills-link-active-bg: #2780e3}.nav-pills .nav-link.active,.nav-pills .show>.nav-link{color:var(--bs-nav-pills-link-active-color);background-color:var(--bs-nav-pills-link-active-bg)}.nav-underline{--bs-nav-underline-gap: 1rem;--bs-nav-underline-border-width: 0.125rem;--bs-nav-underline-link-active-color: #000;gap:var(--bs-nav-underline-gap)}.nav-underline .nav-link{padding-right:0;padding-left:0;border-bottom:var(--bs-nav-underline-border-width) solid rgba(0,0,0,0)}.nav-underline .nav-link:hover,.nav-underline .nav-link:focus{border-bottom-color:currentcolor}.nav-underline .nav-link.active,.nav-underline .show>.nav-link{font-weight:700;color:var(--bs-nav-underline-link-active-color);border-bottom-color:currentcolor}.nav-fill>.nav-link,.nav-fill .nav-item{flex:1 1 auto;-webkit-flex:1 1 auto;text-align:center}.nav-justified>.nav-link,.nav-justified .nav-item{flex-basis:0;-webkit-flex-basis:0;flex-grow:1;-webkit-flex-grow:1;text-align:center}.nav-fill .nav-item .nav-link,.nav-justified .nav-item .nav-link{width:100%}.tab-content>.tab-pane{display:none}.tab-content>.active{display:block}.navbar{--bs-navbar-padding-x: 0;--bs-navbar-padding-y: 0.5rem;--bs-navbar-color: #bdbdbd;--bs-navbar-hover-color: rgba(174, 209, 228, 0.8);--bs-navbar-disabled-color: rgba(189, 189, 189, 0.75);--bs-navbar-active-color: #aed1e4;--bs-navbar-brand-padding-y: 0.3125rem;--bs-navbar-brand-margin-end: 1rem;--bs-navbar-brand-font-size: 1.25rem;--bs-navbar-brand-color: #bdbdbd;--bs-navbar-brand-hover-color: #aed1e4;--bs-navbar-nav-link-padding-x: 0.5rem;--bs-navbar-toggler-padding-y: 0.25;--bs-navbar-toggler-padding-x: 0;--bs-navbar-toggler-font-size: 1.25rem;--bs-navbar-toggler-icon-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 30 30'%3e%3cpath stroke='%23bdbdbd' stroke-linecap='round' stroke-miterlimit='10' stroke-width='2' d='M4 7h22M4 15h22M4 23h22'/%3e%3c/svg%3e");--bs-navbar-toggler-border-color: rgba(189, 189, 189, 0);--bs-navbar-toggler-border-radius: 0.25rem;--bs-navbar-toggler-focus-width: 0.25rem;--bs-navbar-toggler-transition: box-shadow 0.15s ease-in-out;position:relative;display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between;padding:var(--bs-navbar-padding-y) var(--bs-navbar-padding-x)}.navbar>.container,.navbar>.container-fluid,.navbar>.container-sm,.navbar>.container-md,.navbar>.container-lg,.navbar>.container-xl,.navbar>.container-xxl{display:flex;display:-webkit-flex;flex-wrap:inherit;-webkit-flex-wrap:inherit;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between}.navbar-brand{padding-top:var(--bs-navbar-brand-padding-y);padding-bottom:var(--bs-navbar-brand-padding-y);margin-right:var(--bs-navbar-brand-margin-end);font-size:var(--bs-navbar-brand-font-size);color:var(--bs-navbar-brand-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;white-space:nowrap}.navbar-brand:hover,.navbar-brand:focus{color:var(--bs-navbar-brand-hover-color)}.navbar-nav{--bs-nav-link-padding-x: 0;--bs-nav-link-padding-y: 0.5rem;--bs-nav-link-font-weight: ;--bs-nav-link-color: var(--bs-navbar-color);--bs-nav-link-hover-color: var(--bs-navbar-hover-color);--bs-nav-link-disabled-color: var(--bs-navbar-disabled-color);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;padding-left:0;margin-bottom:0;list-style:none}.navbar-nav .nav-link.active,.navbar-nav .nav-link.show{color:var(--bs-navbar-active-color)}.navbar-nav .dropdown-menu{position:static}.navbar-text{padding-top:.5rem;padding-bottom:.5rem;color:var(--bs-navbar-color)}.navbar-text a,.navbar-text a:hover,.navbar-text a:focus{color:var(--bs-navbar-active-color)}.navbar-collapse{flex-basis:100%;-webkit-flex-basis:100%;flex-grow:1;-webkit-flex-grow:1;align-items:center;-webkit-align-items:center}.navbar-toggler{padding:var(--bs-navbar-toggler-padding-y) var(--bs-navbar-toggler-padding-x);font-size:var(--bs-navbar-toggler-font-size);line-height:1;color:var(--bs-navbar-color);background-color:rgba(0,0,0,0);border:var(--bs-border-width) solid var(--bs-navbar-toggler-border-color);transition:var(--bs-navbar-toggler-transition)}@media(prefers-reduced-motion: reduce){.navbar-toggler{transition:none}}.navbar-toggler:hover{text-decoration:none}.navbar-toggler:focus{text-decoration:none;outline:0;box-shadow:0 0 0 var(--bs-navbar-toggler-focus-width)}.navbar-toggler-icon{display:inline-block;width:1.5em;height:1.5em;vertical-align:middle;background-image:var(--bs-navbar-toggler-icon-bg);background-repeat:no-repeat;background-position:center;background-size:100%}.navbar-nav-scroll{max-height:var(--bs-scroll-height, 75vh);overflow-y:auto}@media(min-width: 576px){.navbar-expand-sm{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-sm .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-sm .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-sm .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-sm .navbar-nav-scroll{overflow:visible}.navbar-expand-sm .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-sm .navbar-toggler{display:none}.navbar-expand-sm .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-sm .offcanvas .offcanvas-header{display:none}.navbar-expand-sm .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 768px){.navbar-expand-md{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-md .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-md .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-md .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-md .navbar-nav-scroll{overflow:visible}.navbar-expand-md .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-md .navbar-toggler{display:none}.navbar-expand-md .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-md .offcanvas .offcanvas-header{display:none}.navbar-expand-md .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 992px){.navbar-expand-lg{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-lg .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-lg .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-lg .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-lg .navbar-nav-scroll{overflow:visible}.navbar-expand-lg .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-lg .navbar-toggler{display:none}.navbar-expand-lg .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-lg .offcanvas .offcanvas-header{display:none}.navbar-expand-lg .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 1200px){.navbar-expand-xl{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-xl .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-xl .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-xl .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-xl .navbar-nav-scroll{overflow:visible}.navbar-expand-xl .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-xl .navbar-toggler{display:none}.navbar-expand-xl .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-xl .offcanvas .offcanvas-header{display:none}.navbar-expand-xl .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}@media(min-width: 1400px){.navbar-expand-xxl{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand-xxl .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand-xxl .navbar-nav .dropdown-menu{position:absolute}.navbar-expand-xxl .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand-xxl .navbar-nav-scroll{overflow:visible}.navbar-expand-xxl .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand-xxl .navbar-toggler{display:none}.navbar-expand-xxl .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand-xxl .offcanvas .offcanvas-header{display:none}.navbar-expand-xxl .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}}.navbar-expand{flex-wrap:nowrap;-webkit-flex-wrap:nowrap;justify-content:flex-start;-webkit-justify-content:flex-start}.navbar-expand .navbar-nav{flex-direction:row;-webkit-flex-direction:row}.navbar-expand .navbar-nav .dropdown-menu{position:absolute}.navbar-expand .navbar-nav .nav-link{padding-right:var(--bs-navbar-nav-link-padding-x);padding-left:var(--bs-navbar-nav-link-padding-x)}.navbar-expand .navbar-nav-scroll{overflow:visible}.navbar-expand .navbar-collapse{display:flex !important;display:-webkit-flex !important;flex-basis:auto;-webkit-flex-basis:auto}.navbar-expand .navbar-toggler{display:none}.navbar-expand .offcanvas{position:static;z-index:auto;flex-grow:1;-webkit-flex-grow:1;width:auto !important;height:auto !important;visibility:visible !important;background-color:rgba(0,0,0,0) !important;border:0 !important;transform:none !important;transition:none}.navbar-expand .offcanvas .offcanvas-header{display:none}.navbar-expand .offcanvas .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible}.navbar-dark,.navbar[data-bs-theme=dark]{--bs-navbar-color: #bdbdbd;--bs-navbar-hover-color: rgba(174, 209, 228, 0.8);--bs-navbar-disabled-color: rgba(189, 189, 189, 0.75);--bs-navbar-active-color: #aed1e4;--bs-navbar-brand-color: #bdbdbd;--bs-navbar-brand-hover-color: #aed1e4;--bs-navbar-toggler-border-color: rgba(189, 189, 189, 0);--bs-navbar-toggler-icon-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 30 30'%3e%3cpath stroke='%23bdbdbd' stroke-linecap='round' stroke-miterlimit='10' stroke-width='2' d='M4 7h22M4 15h22M4 23h22'/%3e%3c/svg%3e")}[data-bs-theme=dark] .navbar-toggler-icon{--bs-navbar-toggler-icon-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 30 30'%3e%3cpath stroke='%23bdbdbd' stroke-linecap='round' stroke-miterlimit='10' stroke-width='2' d='M4 7h22M4 15h22M4 23h22'/%3e%3c/svg%3e")}.card{--bs-card-spacer-y: 1rem;--bs-card-spacer-x: 1rem;--bs-card-title-spacer-y: 0.5rem;--bs-card-title-color: ;--bs-card-subtitle-color: ;--bs-card-border-width: 1px;--bs-card-border-color: rgba(0, 0, 0, 0.175);--bs-card-border-radius: 0.25rem;--bs-card-box-shadow: ;--bs-card-inner-border-radius: calc(0.25rem - 1px);--bs-card-cap-padding-y: 0.5rem;--bs-card-cap-padding-x: 1rem;--bs-card-cap-bg: rgba(52, 58, 64, 0.25);--bs-card-cap-color: ;--bs-card-height: ;--bs-card-color: ;--bs-card-bg: #181818;--bs-card-img-overlay-padding: 1rem;--bs-card-group-margin: 0.75rem;position:relative;display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;min-width:0;height:var(--bs-card-height);color:var(--bs-body-color);word-wrap:break-word;background-color:var(--bs-card-bg);background-clip:border-box;border:var(--bs-card-border-width) solid var(--bs-card-border-color)}.card>hr{margin-right:0;margin-left:0}.card>.list-group{border-top:inherit;border-bottom:inherit}.card>.list-group:first-child{border-top-width:0}.card>.list-group:last-child{border-bottom-width:0}.card>.card-header+.list-group,.card>.list-group+.card-footer{border-top:0}.card-body{flex:1 1 auto;-webkit-flex:1 1 auto;padding:var(--bs-card-spacer-y) var(--bs-card-spacer-x);color:var(--bs-card-color)}.card-title{margin-bottom:var(--bs-card-title-spacer-y);color:var(--bs-card-title-color)}.card-subtitle{margin-top:calc(-0.5*var(--bs-card-title-spacer-y));margin-bottom:0;color:var(--bs-card-subtitle-color)}.card-text:last-child{margin-bottom:0}.card-link+.card-link{margin-left:var(--bs-card-spacer-x)}.card-header{padding:var(--bs-card-cap-padding-y) var(--bs-card-cap-padding-x);margin-bottom:0;color:var(--bs-card-cap-color);background-color:var(--bs-card-cap-bg);border-bottom:var(--bs-card-border-width) solid var(--bs-card-border-color)}.card-footer{padding:var(--bs-card-cap-padding-y) var(--bs-card-cap-padding-x);color:var(--bs-card-cap-color);background-color:var(--bs-card-cap-bg);border-top:var(--bs-card-border-width) solid var(--bs-card-border-color)}.card-header-tabs{margin-right:calc(-0.5*var(--bs-card-cap-padding-x));margin-bottom:calc(-1*var(--bs-card-cap-padding-y));margin-left:calc(-0.5*var(--bs-card-cap-padding-x));border-bottom:0}.card-header-tabs .nav-link.active{background-color:var(--bs-card-bg);border-bottom-color:var(--bs-card-bg)}.card-header-pills{margin-right:calc(-0.5*var(--bs-card-cap-padding-x));margin-left:calc(-0.5*var(--bs-card-cap-padding-x))}.card-img-overlay{position:absolute;top:0;right:0;bottom:0;left:0;padding:var(--bs-card-img-overlay-padding)}.card-img,.card-img-top,.card-img-bottom{width:100%}.card-group>.card{margin-bottom:var(--bs-card-group-margin)}@media(min-width: 576px){.card-group{display:flex;display:-webkit-flex;flex-flow:row wrap;-webkit-flex-flow:row wrap}.card-group>.card{flex:1 0 0%;-webkit-flex:1 0 0%;margin-bottom:0}.card-group>.card+.card{margin-left:0;border-left:0}}.accordion{--bs-accordion-color: #FAF1E4;--bs-accordion-bg: #181818;--bs-accordion-transition: color 0.15s ease-in-out, background-color 0.15s ease-in-out, border-color 0.15s ease-in-out, box-shadow 0.15s ease-in-out, border-radius 0.15s ease;--bs-accordion-border-color: #dee2e6;--bs-accordion-border-width: 1px;--bs-accordion-border-radius: 0.25rem;--bs-accordion-inner-border-radius: calc(0.25rem - 1px);--bs-accordion-btn-padding-x: 1.25rem;--bs-accordion-btn-padding-y: 1rem;--bs-accordion-btn-color: #FAF1E4;--bs-accordion-btn-bg: #181818;--bs-accordion-btn-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23FAF1E4'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e");--bs-accordion-btn-icon-width: 1.25rem;--bs-accordion-btn-icon-transform: rotate(-180deg);--bs-accordion-btn-icon-transition: transform 0.2s ease-in-out;--bs-accordion-btn-active-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%2310335b'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e");--bs-accordion-btn-focus-border-color: #93c0f1;--bs-accordion-btn-focus-box-shadow: 0 0 0 0.25rem rgba(39, 128, 227, 0.25);--bs-accordion-body-padding-x: 1.25rem;--bs-accordion-body-padding-y: 1rem;--bs-accordion-active-color: #10335b;--bs-accordion-active-bg: #d4e6f9}.accordion-button{position:relative;display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;width:100%;padding:var(--bs-accordion-btn-padding-y) var(--bs-accordion-btn-padding-x);font-size:1rem;color:var(--bs-accordion-btn-color);text-align:left;background-color:var(--bs-accordion-btn-bg);border:0;overflow-anchor:none;transition:var(--bs-accordion-transition)}@media(prefers-reduced-motion: reduce){.accordion-button{transition:none}}.accordion-button:not(.collapsed){color:var(--bs-accordion-active-color);background-color:var(--bs-accordion-active-bg);box-shadow:inset 0 calc(-1*var(--bs-accordion-border-width)) 0 var(--bs-accordion-border-color)}.accordion-button:not(.collapsed)::after{background-image:var(--bs-accordion-btn-active-icon);transform:var(--bs-accordion-btn-icon-transform)}.accordion-button::after{flex-shrink:0;-webkit-flex-shrink:0;width:var(--bs-accordion-btn-icon-width);height:var(--bs-accordion-btn-icon-width);margin-left:auto;content:"";background-image:var(--bs-accordion-btn-icon);background-repeat:no-repeat;background-size:var(--bs-accordion-btn-icon-width);transition:var(--bs-accordion-btn-icon-transition)}@media(prefers-reduced-motion: reduce){.accordion-button::after{transition:none}}.accordion-button:hover{z-index:2}.accordion-button:focus{z-index:3;border-color:var(--bs-accordion-btn-focus-border-color);outline:0;box-shadow:var(--bs-accordion-btn-focus-box-shadow)}.accordion-header{margin-bottom:0}.accordion-item{color:var(--bs-accordion-color);background-color:var(--bs-accordion-bg);border:var(--bs-accordion-border-width) solid var(--bs-accordion-border-color)}.accordion-item:not(:first-of-type){border-top:0}.accordion-body{padding:var(--bs-accordion-body-padding-y) var(--bs-accordion-body-padding-x)}.accordion-flush .accordion-collapse{border-width:0}.accordion-flush .accordion-item{border-right:0;border-left:0}.accordion-flush .accordion-item:first-child{border-top:0}.accordion-flush .accordion-item:last-child{border-bottom:0}[data-bs-theme=dark] .accordion-button::after{--bs-accordion-btn-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%237db3ee'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e");--bs-accordion-btn-active-icon: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%237db3ee'%3e%3cpath fill-rule='evenodd' d='M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e")}.breadcrumb{--bs-breadcrumb-padding-x: 0;--bs-breadcrumb-padding-y: 0;--bs-breadcrumb-margin-bottom: 1rem;--bs-breadcrumb-bg: ;--bs-breadcrumb-border-radius: ;--bs-breadcrumb-divider-color: rgba(250, 241, 228, 0.75);--bs-breadcrumb-item-padding-x: 0.5rem;--bs-breadcrumb-item-active-color: rgba(250, 241, 228, 0.75);display:flex;display:-webkit-flex;flex-wrap:wrap;-webkit-flex-wrap:wrap;padding:var(--bs-breadcrumb-padding-y) var(--bs-breadcrumb-padding-x);margin-bottom:var(--bs-breadcrumb-margin-bottom);font-size:var(--bs-breadcrumb-font-size);list-style:none;background-color:var(--bs-breadcrumb-bg)}.breadcrumb-item+.breadcrumb-item{padding-left:var(--bs-breadcrumb-item-padding-x)}.breadcrumb-item+.breadcrumb-item::before{float:left;padding-right:var(--bs-breadcrumb-item-padding-x);color:var(--bs-breadcrumb-divider-color);content:var(--bs-breadcrumb-divider, ">") /* rtl: var(--bs-breadcrumb-divider, ">") */}.breadcrumb-item.active{color:var(--bs-breadcrumb-item-active-color)}.pagination{--bs-pagination-padding-x: 0.75rem;--bs-pagination-padding-y: 0.375rem;--bs-pagination-font-size:1rem;--bs-pagination-color: #AED1E4;--bs-pagination-bg: #181818;--bs-pagination-border-width: 1px;--bs-pagination-border-color: #dee2e6;--bs-pagination-border-radius: 0.25rem;--bs-pagination-hover-color: #8ba7b6;--bs-pagination-hover-bg: #f8f9fa;--bs-pagination-hover-border-color: #dee2e6;--bs-pagination-focus-color: #8ba7b6;--bs-pagination-focus-bg: #e9ecef;--bs-pagination-focus-box-shadow: 0 0 0 0.25rem rgba(39, 128, 227, 0.25);--bs-pagination-active-color: #fff;--bs-pagination-active-bg: #2780e3;--bs-pagination-active-border-color: #2780e3;--bs-pagination-disabled-color: rgba(250, 241, 228, 0.75);--bs-pagination-disabled-bg: #e9ecef;--bs-pagination-disabled-border-color: #dee2e6;display:flex;display:-webkit-flex;padding-left:0;list-style:none}.page-link{position:relative;display:block;padding:var(--bs-pagination-padding-y) var(--bs-pagination-padding-x);font-size:var(--bs-pagination-font-size);color:var(--bs-pagination-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;background-color:var(--bs-pagination-bg);border:var(--bs-pagination-border-width) solid var(--bs-pagination-border-color);transition:color .15s ease-in-out,background-color .15s ease-in-out,border-color .15s ease-in-out,box-shadow .15s ease-in-out}@media(prefers-reduced-motion: reduce){.page-link{transition:none}}.page-link:hover{z-index:2;color:var(--bs-pagination-hover-color);background-color:var(--bs-pagination-hover-bg);border-color:var(--bs-pagination-hover-border-color)}.page-link:focus{z-index:3;color:var(--bs-pagination-focus-color);background-color:var(--bs-pagination-focus-bg);outline:0;box-shadow:var(--bs-pagination-focus-box-shadow)}.page-link.active,.active>.page-link{z-index:3;color:var(--bs-pagination-active-color);background-color:var(--bs-pagination-active-bg);border-color:var(--bs-pagination-active-border-color)}.page-link.disabled,.disabled>.page-link{color:var(--bs-pagination-disabled-color);pointer-events:none;background-color:var(--bs-pagination-disabled-bg);border-color:var(--bs-pagination-disabled-border-color)}.page-item:not(:first-child) .page-link{margin-left:calc(1px*-1)}.pagination-lg{--bs-pagination-padding-x: 1.5rem;--bs-pagination-padding-y: 0.75rem;--bs-pagination-font-size:1.25rem;--bs-pagination-border-radius: 0.5rem}.pagination-sm{--bs-pagination-padding-x: 0.5rem;--bs-pagination-padding-y: 0.25rem;--bs-pagination-font-size:0.875rem;--bs-pagination-border-radius: 0.2em}.badge{--bs-badge-padding-x: 0.65em;--bs-badge-padding-y: 0.35em;--bs-badge-font-size:0.75em;--bs-badge-font-weight: 700;--bs-badge-color: #fff;--bs-badge-border-radius: 0.25rem;display:inline-block;padding:var(--bs-badge-padding-y) var(--bs-badge-padding-x);font-size:var(--bs-badge-font-size);font-weight:var(--bs-badge-font-weight);line-height:1;color:var(--bs-badge-color);text-align:center;white-space:nowrap;vertical-align:baseline}.badge:empty{display:none}.btn .badge{position:relative;top:-1px}.alert{--bs-alert-bg: transparent;--bs-alert-padding-x: 1rem;--bs-alert-padding-y: 1rem;--bs-alert-margin-bottom: 1rem;--bs-alert-color: inherit;--bs-alert-border-color: transparent;--bs-alert-border: 0 solid var(--bs-alert-border-color);--bs-alert-border-radius: 0.25rem;--bs-alert-link-color: inherit;position:relative;padding:var(--bs-alert-padding-y) var(--bs-alert-padding-x);margin-bottom:var(--bs-alert-margin-bottom);color:var(--bs-alert-color);background-color:var(--bs-alert-bg);border:var(--bs-alert-border)}.alert-heading{color:inherit}.alert-link{font-weight:700;color:var(--bs-alert-link-color)}.alert-dismissible{padding-right:3rem}.alert-dismissible .btn-close{position:absolute;top:0;right:0;z-index:2;padding:1.25rem 1rem}.alert-default{--bs-alert-color: var(--bs-default-text-emphasis);--bs-alert-bg: var(--bs-default-bg-subtle);--bs-alert-border-color: var(--bs-default-border-subtle);--bs-alert-link-color: var(--bs-default-text-emphasis)}.alert-primary{--bs-alert-color: var(--bs-primary-text-emphasis);--bs-alert-bg: var(--bs-primary-bg-subtle);--bs-alert-border-color: var(--bs-primary-border-subtle);--bs-alert-link-color: var(--bs-primary-text-emphasis)}.alert-secondary{--bs-alert-color: var(--bs-secondary-text-emphasis);--bs-alert-bg: var(--bs-secondary-bg-subtle);--bs-alert-border-color: var(--bs-secondary-border-subtle);--bs-alert-link-color: var(--bs-secondary-text-emphasis)}.alert-success{--bs-alert-color: var(--bs-success-text-emphasis);--bs-alert-bg: var(--bs-success-bg-subtle);--bs-alert-border-color: var(--bs-success-border-subtle);--bs-alert-link-color: var(--bs-success-text-emphasis)}.alert-info{--bs-alert-color: var(--bs-info-text-emphasis);--bs-alert-bg: var(--bs-info-bg-subtle);--bs-alert-border-color: var(--bs-info-border-subtle);--bs-alert-link-color: var(--bs-info-text-emphasis)}.alert-warning{--bs-alert-color: var(--bs-warning-text-emphasis);--bs-alert-bg: var(--bs-warning-bg-subtle);--bs-alert-border-color: var(--bs-warning-border-subtle);--bs-alert-link-color: var(--bs-warning-text-emphasis)}.alert-danger{--bs-alert-color: var(--bs-danger-text-emphasis);--bs-alert-bg: var(--bs-danger-bg-subtle);--bs-alert-border-color: var(--bs-danger-border-subtle);--bs-alert-link-color: var(--bs-danger-text-emphasis)}.alert-light{--bs-alert-color: var(--bs-light-text-emphasis);--bs-alert-bg: var(--bs-light-bg-subtle);--bs-alert-border-color: var(--bs-light-border-subtle);--bs-alert-link-color: var(--bs-light-text-emphasis)}.alert-dark{--bs-alert-color: var(--bs-dark-text-emphasis);--bs-alert-bg: var(--bs-dark-bg-subtle);--bs-alert-border-color: var(--bs-dark-border-subtle);--bs-alert-link-color: var(--bs-dark-text-emphasis)}@keyframes progress-bar-stripes{0%{background-position-x:.5rem}}.progress,.progress-stacked{--bs-progress-height: 0.5rem;--bs-progress-font-size:0.75rem;--bs-progress-bg: #e9ecef;--bs-progress-border-radius: 0.25rem;--bs-progress-box-shadow: inset 0 1px 2px rgba(0, 0, 0, 0.075);--bs-progress-bar-color: #fff;--bs-progress-bar-bg: #2780e3;--bs-progress-bar-transition: width 0.6s ease;display:flex;display:-webkit-flex;height:var(--bs-progress-height);overflow:hidden;font-size:var(--bs-progress-font-size);background-color:var(--bs-progress-bg)}.progress-bar{display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;justify-content:center;-webkit-justify-content:center;overflow:hidden;color:var(--bs-progress-bar-color);text-align:center;white-space:nowrap;background-color:var(--bs-progress-bar-bg);transition:var(--bs-progress-bar-transition)}@media(prefers-reduced-motion: reduce){.progress-bar{transition:none}}.progress-bar-striped{background-image:linear-gradient(45deg, rgba(255, 255, 255, 0.15) 25%, transparent 25%, transparent 50%, rgba(255, 255, 255, 0.15) 50%, rgba(255, 255, 255, 0.15) 75%, transparent 75%, transparent);background-size:var(--bs-progress-height) var(--bs-progress-height)}.progress-stacked>.progress{overflow:visible}.progress-stacked>.progress>.progress-bar{width:100%}.progress-bar-animated{animation:1s linear infinite progress-bar-stripes}@media(prefers-reduced-motion: reduce){.progress-bar-animated{animation:none}}.list-group{--bs-list-group-color: #FAF1E4;--bs-list-group-bg: #181818;--bs-list-group-border-color: #dee2e6;--bs-list-group-border-width: 1px;--bs-list-group-border-radius: 0.25rem;--bs-list-group-item-padding-x: 1rem;--bs-list-group-item-padding-y: 0.5rem;--bs-list-group-action-color: rgba(250, 241, 228, 0.75);--bs-list-group-action-hover-color: #000;--bs-list-group-action-hover-bg: #f8f9fa;--bs-list-group-action-active-color: #FAF1E4;--bs-list-group-action-active-bg: #e9ecef;--bs-list-group-disabled-color: rgba(250, 241, 228, 0.75);--bs-list-group-disabled-bg: #181818;--bs-list-group-active-color: #fff;--bs-list-group-active-bg: #2780e3;--bs-list-group-active-border-color: #2780e3;display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;padding-left:0;margin-bottom:0}.list-group-numbered{list-style-type:none;counter-reset:section}.list-group-numbered>.list-group-item::before{content:counters(section, ".") ". ";counter-increment:section}.list-group-item-action{width:100%;color:var(--bs-list-group-action-color);text-align:inherit}.list-group-item-action:hover,.list-group-item-action:focus{z-index:1;color:var(--bs-list-group-action-hover-color);text-decoration:none;background-color:var(--bs-list-group-action-hover-bg)}.list-group-item-action:active{color:var(--bs-list-group-action-active-color);background-color:var(--bs-list-group-action-active-bg)}.list-group-item{position:relative;display:block;padding:var(--bs-list-group-item-padding-y) var(--bs-list-group-item-padding-x);color:var(--bs-list-group-color);text-decoration:none;-webkit-text-decoration:none;-moz-text-decoration:none;-ms-text-decoration:none;-o-text-decoration:none;background-color:var(--bs-list-group-bg);border:var(--bs-list-group-border-width) solid var(--bs-list-group-border-color)}.list-group-item.disabled,.list-group-item:disabled{color:var(--bs-list-group-disabled-color);pointer-events:none;background-color:var(--bs-list-group-disabled-bg)}.list-group-item.active{z-index:2;color:var(--bs-list-group-active-color);background-color:var(--bs-list-group-active-bg);border-color:var(--bs-list-group-active-border-color)}.list-group-item+.list-group-item{border-top-width:0}.list-group-item+.list-group-item.active{margin-top:calc(-1*var(--bs-list-group-border-width));border-top-width:var(--bs-list-group-border-width)}.list-group-horizontal{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal>.list-group-item.active{margin-top:0}.list-group-horizontal>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}@media(min-width: 576px){.list-group-horizontal-sm{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-sm>.list-group-item.active{margin-top:0}.list-group-horizontal-sm>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-sm>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 768px){.list-group-horizontal-md{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-md>.list-group-item.active{margin-top:0}.list-group-horizontal-md>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-md>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 992px){.list-group-horizontal-lg{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-lg>.list-group-item.active{margin-top:0}.list-group-horizontal-lg>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-lg>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 1200px){.list-group-horizontal-xl{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-xl>.list-group-item.active{margin-top:0}.list-group-horizontal-xl>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-xl>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}@media(min-width: 1400px){.list-group-horizontal-xxl{flex-direction:row;-webkit-flex-direction:row}.list-group-horizontal-xxl>.list-group-item.active{margin-top:0}.list-group-horizontal-xxl>.list-group-item+.list-group-item{border-top-width:var(--bs-list-group-border-width);border-left-width:0}.list-group-horizontal-xxl>.list-group-item+.list-group-item.active{margin-left:calc(-1*var(--bs-list-group-border-width));border-left-width:var(--bs-list-group-border-width)}}.list-group-flush>.list-group-item{border-width:0 0 var(--bs-list-group-border-width)}.list-group-flush>.list-group-item:last-child{border-bottom-width:0}.list-group-item-default{--bs-list-group-color: var(--bs-default-text-emphasis);--bs-list-group-bg: var(--bs-default-bg-subtle);--bs-list-group-border-color: var(--bs-default-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-default-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-default-border-subtle);--bs-list-group-active-color: var(--bs-default-bg-subtle);--bs-list-group-active-bg: var(--bs-default-text-emphasis);--bs-list-group-active-border-color: var(--bs-default-text-emphasis)}.list-group-item-primary{--bs-list-group-color: var(--bs-primary-text-emphasis);--bs-list-group-bg: var(--bs-primary-bg-subtle);--bs-list-group-border-color: var(--bs-primary-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-primary-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-primary-border-subtle);--bs-list-group-active-color: var(--bs-primary-bg-subtle);--bs-list-group-active-bg: var(--bs-primary-text-emphasis);--bs-list-group-active-border-color: var(--bs-primary-text-emphasis)}.list-group-item-secondary{--bs-list-group-color: var(--bs-secondary-text-emphasis);--bs-list-group-bg: var(--bs-secondary-bg-subtle);--bs-list-group-border-color: var(--bs-secondary-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-secondary-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-secondary-border-subtle);--bs-list-group-active-color: var(--bs-secondary-bg-subtle);--bs-list-group-active-bg: var(--bs-secondary-text-emphasis);--bs-list-group-active-border-color: var(--bs-secondary-text-emphasis)}.list-group-item-success{--bs-list-group-color: var(--bs-success-text-emphasis);--bs-list-group-bg: var(--bs-success-bg-subtle);--bs-list-group-border-color: var(--bs-success-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-success-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-success-border-subtle);--bs-list-group-active-color: var(--bs-success-bg-subtle);--bs-list-group-active-bg: var(--bs-success-text-emphasis);--bs-list-group-active-border-color: var(--bs-success-text-emphasis)}.list-group-item-info{--bs-list-group-color: var(--bs-info-text-emphasis);--bs-list-group-bg: var(--bs-info-bg-subtle);--bs-list-group-border-color: var(--bs-info-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-info-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-info-border-subtle);--bs-list-group-active-color: var(--bs-info-bg-subtle);--bs-list-group-active-bg: var(--bs-info-text-emphasis);--bs-list-group-active-border-color: var(--bs-info-text-emphasis)}.list-group-item-warning{--bs-list-group-color: var(--bs-warning-text-emphasis);--bs-list-group-bg: var(--bs-warning-bg-subtle);--bs-list-group-border-color: var(--bs-warning-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-warning-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-warning-border-subtle);--bs-list-group-active-color: var(--bs-warning-bg-subtle);--bs-list-group-active-bg: var(--bs-warning-text-emphasis);--bs-list-group-active-border-color: var(--bs-warning-text-emphasis)}.list-group-item-danger{--bs-list-group-color: var(--bs-danger-text-emphasis);--bs-list-group-bg: var(--bs-danger-bg-subtle);--bs-list-group-border-color: var(--bs-danger-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-danger-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-danger-border-subtle);--bs-list-group-active-color: var(--bs-danger-bg-subtle);--bs-list-group-active-bg: var(--bs-danger-text-emphasis);--bs-list-group-active-border-color: var(--bs-danger-text-emphasis)}.list-group-item-light{--bs-list-group-color: var(--bs-light-text-emphasis);--bs-list-group-bg: var(--bs-light-bg-subtle);--bs-list-group-border-color: var(--bs-light-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-light-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-light-border-subtle);--bs-list-group-active-color: var(--bs-light-bg-subtle);--bs-list-group-active-bg: var(--bs-light-text-emphasis);--bs-list-group-active-border-color: var(--bs-light-text-emphasis)}.list-group-item-dark{--bs-list-group-color: var(--bs-dark-text-emphasis);--bs-list-group-bg: var(--bs-dark-bg-subtle);--bs-list-group-border-color: var(--bs-dark-border-subtle);--bs-list-group-action-hover-color: var(--bs-emphasis-color);--bs-list-group-action-hover-bg: var(--bs-dark-border-subtle);--bs-list-group-action-active-color: var(--bs-emphasis-color);--bs-list-group-action-active-bg: var(--bs-dark-border-subtle);--bs-list-group-active-color: var(--bs-dark-bg-subtle);--bs-list-group-active-bg: var(--bs-dark-text-emphasis);--bs-list-group-active-border-color: var(--bs-dark-text-emphasis)}.btn-close{--bs-btn-close-color: #000;--bs-btn-close-bg: url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23000'%3e%3cpath d='M.293.293a1 1 0 0 1 1.414 0L8 6.586 14.293.293a1 1 0 1 1 1.414 1.414L9.414 8l6.293 6.293a1 1 0 0 1-1.414 1.414L8 9.414l-6.293 6.293a1 1 0 0 1-1.414-1.414L6.586 8 .293 1.707a1 1 0 0 1 0-1.414z'/%3e%3c/svg%3e");--bs-btn-close-opacity: 0.5;--bs-btn-close-hover-opacity: 0.75;--bs-btn-close-focus-shadow: 0 0 0 0.25rem rgba(39, 128, 227, 0.25);--bs-btn-close-focus-opacity: 1;--bs-btn-close-disabled-opacity: 0.25;--bs-btn-close-white-filter: invert(1) grayscale(100%) brightness(200%);box-sizing:content-box;width:1em;height:1em;padding:.25em .25em;color:var(--bs-btn-close-color);background:rgba(0,0,0,0) var(--bs-btn-close-bg) center/1em auto no-repeat;border:0;opacity:var(--bs-btn-close-opacity)}.btn-close:hover{color:var(--bs-btn-close-color);text-decoration:none;opacity:var(--bs-btn-close-hover-opacity)}.btn-close:focus{outline:0;box-shadow:var(--bs-btn-close-focus-shadow);opacity:var(--bs-btn-close-focus-opacity)}.btn-close:disabled,.btn-close.disabled{pointer-events:none;user-select:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;-o-user-select:none;opacity:var(--bs-btn-close-disabled-opacity)}.btn-close-white{filter:var(--bs-btn-close-white-filter)}[data-bs-theme=dark] .btn-close{filter:var(--bs-btn-close-white-filter)}.toast{--bs-toast-zindex: 1090;--bs-toast-padding-x: 0.75rem;--bs-toast-padding-y: 0.5rem;--bs-toast-spacing: 1.5rem;--bs-toast-max-width: 350px;--bs-toast-font-size:0.875rem;--bs-toast-color: ;--bs-toast-bg: rgba(24, 24, 24, 0.85);--bs-toast-border-width: 1px;--bs-toast-border-color: rgba(0, 0, 0, 0.175);--bs-toast-border-radius: 0.25rem;--bs-toast-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-toast-header-color: rgba(250, 241, 228, 0.75);--bs-toast-header-bg: rgba(24, 24, 24, 0.85);--bs-toast-header-border-color: rgba(0, 0, 0, 0.175);width:var(--bs-toast-max-width);max-width:100%;font-size:var(--bs-toast-font-size);color:var(--bs-toast-color);pointer-events:auto;background-color:var(--bs-toast-bg);background-clip:padding-box;border:var(--bs-toast-border-width) solid var(--bs-toast-border-color);box-shadow:var(--bs-toast-box-shadow)}.toast.showing{opacity:0}.toast:not(.show){display:none}.toast-container{--bs-toast-zindex: 1090;position:absolute;z-index:var(--bs-toast-zindex);width:max-content;width:-webkit-max-content;width:-moz-max-content;width:-ms-max-content;width:-o-max-content;max-width:100%;pointer-events:none}.toast-container>:not(:last-child){margin-bottom:var(--bs-toast-spacing)}.toast-header{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;padding:var(--bs-toast-padding-y) var(--bs-toast-padding-x);color:var(--bs-toast-header-color);background-color:var(--bs-toast-header-bg);background-clip:padding-box;border-bottom:var(--bs-toast-border-width) solid var(--bs-toast-header-border-color)}.toast-header .btn-close{margin-right:calc(-0.5*var(--bs-toast-padding-x));margin-left:var(--bs-toast-padding-x)}.toast-body{padding:var(--bs-toast-padding-x);word-wrap:break-word}.modal{--bs-modal-zindex: 1055;--bs-modal-width: 500px;--bs-modal-padding: 1rem;--bs-modal-margin: 0.5rem;--bs-modal-color: ;--bs-modal-bg: #181818;--bs-modal-border-color: rgba(0, 0, 0, 0.175);--bs-modal-border-width: 1px;--bs-modal-border-radius: 0.5rem;--bs-modal-box-shadow: 0 0.125rem 0.25rem rgba(0, 0, 0, 0.075);--bs-modal-inner-border-radius: calc(0.5rem - 1px);--bs-modal-header-padding-x: 1rem;--bs-modal-header-padding-y: 1rem;--bs-modal-header-padding: 1rem 1rem;--bs-modal-header-border-color: #dee2e6;--bs-modal-header-border-width: 1px;--bs-modal-title-line-height: 1.5;--bs-modal-footer-gap: 0.5rem;--bs-modal-footer-bg: ;--bs-modal-footer-border-color: #dee2e6;--bs-modal-footer-border-width: 1px;position:fixed;top:0;left:0;z-index:var(--bs-modal-zindex);display:none;width:100%;height:100%;overflow-x:hidden;overflow-y:auto;outline:0}.modal-dialog{position:relative;width:auto;margin:var(--bs-modal-margin);pointer-events:none}.modal.fade .modal-dialog{transition:transform .3s ease-out;transform:translate(0, -50px)}@media(prefers-reduced-motion: reduce){.modal.fade .modal-dialog{transition:none}}.modal.show .modal-dialog{transform:none}.modal.modal-static .modal-dialog{transform:scale(1.02)}.modal-dialog-scrollable{height:calc(100% - var(--bs-modal-margin)*2)}.modal-dialog-scrollable .modal-content{max-height:100%;overflow:hidden}.modal-dialog-scrollable .modal-body{overflow-y:auto}.modal-dialog-centered{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;min-height:calc(100% - var(--bs-modal-margin)*2)}.modal-content{position:relative;display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;width:100%;color:var(--bs-modal-color);pointer-events:auto;background-color:var(--bs-modal-bg);background-clip:padding-box;border:var(--bs-modal-border-width) solid var(--bs-modal-border-color);outline:0}.modal-backdrop{--bs-backdrop-zindex: 1050;--bs-backdrop-bg: #000;--bs-backdrop-opacity: 0.5;position:fixed;top:0;left:0;z-index:var(--bs-backdrop-zindex);width:100vw;height:100vh;background-color:var(--bs-backdrop-bg)}.modal-backdrop.fade{opacity:0}.modal-backdrop.show{opacity:var(--bs-backdrop-opacity)}.modal-header{display:flex;display:-webkit-flex;flex-shrink:0;-webkit-flex-shrink:0;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between;padding:var(--bs-modal-header-padding);border-bottom:var(--bs-modal-header-border-width) solid var(--bs-modal-header-border-color)}.modal-header .btn-close{padding:calc(var(--bs-modal-header-padding-y)*.5) calc(var(--bs-modal-header-padding-x)*.5);margin:calc(-0.5*var(--bs-modal-header-padding-y)) calc(-0.5*var(--bs-modal-header-padding-x)) calc(-0.5*var(--bs-modal-header-padding-y)) auto}.modal-title{margin-bottom:0;line-height:var(--bs-modal-title-line-height)}.modal-body{position:relative;flex:1 1 auto;-webkit-flex:1 1 auto;padding:var(--bs-modal-padding)}.modal-footer{display:flex;display:-webkit-flex;flex-shrink:0;-webkit-flex-shrink:0;flex-wrap:wrap;-webkit-flex-wrap:wrap;align-items:center;-webkit-align-items:center;justify-content:flex-end;-webkit-justify-content:flex-end;padding:calc(var(--bs-modal-padding) - var(--bs-modal-footer-gap)*.5);background-color:var(--bs-modal-footer-bg);border-top:var(--bs-modal-footer-border-width) solid var(--bs-modal-footer-border-color)}.modal-footer>*{margin:calc(var(--bs-modal-footer-gap)*.5)}@media(min-width: 576px){.modal{--bs-modal-margin: 1.75rem;--bs-modal-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15)}.modal-dialog{max-width:var(--bs-modal-width);margin-right:auto;margin-left:auto}.modal-sm{--bs-modal-width: 300px}}@media(min-width: 992px){.modal-lg,.modal-xl{--bs-modal-width: 800px}}@media(min-width: 1200px){.modal-xl{--bs-modal-width: 1140px}}.modal-fullscreen{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen .modal-content{height:100%;border:0}.modal-fullscreen .modal-body{overflow-y:auto}@media(max-width: 575.98px){.modal-fullscreen-sm-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-sm-down .modal-content{height:100%;border:0}.modal-fullscreen-sm-down .modal-body{overflow-y:auto}}@media(max-width: 767.98px){.modal-fullscreen-md-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-md-down .modal-content{height:100%;border:0}.modal-fullscreen-md-down .modal-body{overflow-y:auto}}@media(max-width: 991.98px){.modal-fullscreen-lg-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-lg-down .modal-content{height:100%;border:0}.modal-fullscreen-lg-down .modal-body{overflow-y:auto}}@media(max-width: 1199.98px){.modal-fullscreen-xl-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-xl-down .modal-content{height:100%;border:0}.modal-fullscreen-xl-down .modal-body{overflow-y:auto}}@media(max-width: 1399.98px){.modal-fullscreen-xxl-down{width:100vw;max-width:none;height:100%;margin:0}.modal-fullscreen-xxl-down .modal-content{height:100%;border:0}.modal-fullscreen-xxl-down .modal-body{overflow-y:auto}}.tooltip{--bs-tooltip-zindex: 1080;--bs-tooltip-max-width: 200px;--bs-tooltip-padding-x: 0.5rem;--bs-tooltip-padding-y: 0.25rem;--bs-tooltip-margin: ;--bs-tooltip-font-size:0.875rem;--bs-tooltip-color: #181818;--bs-tooltip-bg: #000;--bs-tooltip-border-radius: 0.25rem;--bs-tooltip-opacity: 0.9;--bs-tooltip-arrow-width: 0.8rem;--bs-tooltip-arrow-height: 0.4rem;z-index:var(--bs-tooltip-zindex);display:block;margin:var(--bs-tooltip-margin);font-family:"Source Sans Pro",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,sans-serif,"Apple Color Emoji","Segoe UI Emoji","Segoe UI Symbol";font-style:normal;font-weight:400;line-height:1.5;text-align:left;text-align:start;text-decoration:none;text-shadow:none;text-transform:none;letter-spacing:normal;word-break:normal;white-space:normal;word-spacing:normal;line-break:auto;font-size:var(--bs-tooltip-font-size);word-wrap:break-word;opacity:0}.tooltip.show{opacity:var(--bs-tooltip-opacity)}.tooltip .tooltip-arrow{display:block;width:var(--bs-tooltip-arrow-width);height:var(--bs-tooltip-arrow-height)}.tooltip .tooltip-arrow::before{position:absolute;content:"";border-color:rgba(0,0,0,0);border-style:solid}.bs-tooltip-top .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=top] .tooltip-arrow{bottom:calc(-1*var(--bs-tooltip-arrow-height))}.bs-tooltip-top .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=top] .tooltip-arrow::before{top:-1px;border-width:var(--bs-tooltip-arrow-height) calc(var(--bs-tooltip-arrow-width)*.5) 0;border-top-color:var(--bs-tooltip-bg)}.bs-tooltip-end .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=right] .tooltip-arrow{left:calc(-1*var(--bs-tooltip-arrow-height));width:var(--bs-tooltip-arrow-height);height:var(--bs-tooltip-arrow-width)}.bs-tooltip-end .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=right] .tooltip-arrow::before{right:-1px;border-width:calc(var(--bs-tooltip-arrow-width)*.5) var(--bs-tooltip-arrow-height) calc(var(--bs-tooltip-arrow-width)*.5) 0;border-right-color:var(--bs-tooltip-bg)}.bs-tooltip-bottom .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=bottom] .tooltip-arrow{top:calc(-1*var(--bs-tooltip-arrow-height))}.bs-tooltip-bottom .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=bottom] .tooltip-arrow::before{bottom:-1px;border-width:0 calc(var(--bs-tooltip-arrow-width)*.5) var(--bs-tooltip-arrow-height);border-bottom-color:var(--bs-tooltip-bg)}.bs-tooltip-start .tooltip-arrow,.bs-tooltip-auto[data-popper-placement^=left] .tooltip-arrow{right:calc(-1*var(--bs-tooltip-arrow-height));width:var(--bs-tooltip-arrow-height);height:var(--bs-tooltip-arrow-width)}.bs-tooltip-start .tooltip-arrow::before,.bs-tooltip-auto[data-popper-placement^=left] .tooltip-arrow::before{left:-1px;border-width:calc(var(--bs-tooltip-arrow-width)*.5) 0 calc(var(--bs-tooltip-arrow-width)*.5) var(--bs-tooltip-arrow-height);border-left-color:var(--bs-tooltip-bg)}.tooltip-inner{max-width:var(--bs-tooltip-max-width);padding:var(--bs-tooltip-padding-y) var(--bs-tooltip-padding-x);color:var(--bs-tooltip-color);text-align:center;background-color:var(--bs-tooltip-bg)}.popover{--bs-popover-zindex: 1070;--bs-popover-max-width: 276px;--bs-popover-font-size:0.875rem;--bs-popover-bg: #242424;--bs-popover-border-width: 1px;--bs-popover-border-color: rgba(0, 0, 0, 0.175);--bs-popover-border-radius: 0.5rem;--bs-popover-inner-border-radius: calc(0.5rem - 1px);--bs-popover-box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);--bs-popover-header-padding-x: 1rem;--bs-popover-header-padding-y: 0.5rem;--bs-popover-header-font-size:1rem;--bs-popover-header-color: inherit;--bs-popover-header-bg: #e9ecef;--bs-popover-body-padding-x: 1rem;--bs-popover-body-padding-y: 1rem;--bs-popover-body-color: #FAF1E4;--bs-popover-arrow-width: 1rem;--bs-popover-arrow-height: 0.5rem;--bs-popover-arrow-border: var(--bs-popover-border-color);z-index:var(--bs-popover-zindex);display:block;max-width:var(--bs-popover-max-width);font-family:"Source Sans Pro",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,sans-serif,"Apple Color Emoji","Segoe UI Emoji","Segoe UI Symbol";font-style:normal;font-weight:400;line-height:1.5;text-align:left;text-align:start;text-decoration:none;text-shadow:none;text-transform:none;letter-spacing:normal;word-break:normal;white-space:normal;word-spacing:normal;line-break:auto;font-size:var(--bs-popover-font-size);word-wrap:break-word;background-color:var(--bs-popover-bg);background-clip:padding-box;border:var(--bs-popover-border-width) solid var(--bs-popover-border-color)}.popover .popover-arrow{display:block;width:var(--bs-popover-arrow-width);height:var(--bs-popover-arrow-height)}.popover .popover-arrow::before,.popover .popover-arrow::after{position:absolute;display:block;content:"";border-color:rgba(0,0,0,0);border-style:solid;border-width:0}.bs-popover-top>.popover-arrow,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow{bottom:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width))}.bs-popover-top>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::before,.bs-popover-top>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::after{border-width:var(--bs-popover-arrow-height) calc(var(--bs-popover-arrow-width)*.5) 0}.bs-popover-top>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::before{bottom:0;border-top-color:var(--bs-popover-arrow-border)}.bs-popover-top>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=top]>.popover-arrow::after{bottom:var(--bs-popover-border-width);border-top-color:var(--bs-popover-bg)}.bs-popover-end>.popover-arrow,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow{left:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width));width:var(--bs-popover-arrow-height);height:var(--bs-popover-arrow-width)}.bs-popover-end>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::before,.bs-popover-end>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::after{border-width:calc(var(--bs-popover-arrow-width)*.5) var(--bs-popover-arrow-height) calc(var(--bs-popover-arrow-width)*.5) 0}.bs-popover-end>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::before{left:0;border-right-color:var(--bs-popover-arrow-border)}.bs-popover-end>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=right]>.popover-arrow::after{left:var(--bs-popover-border-width);border-right-color:var(--bs-popover-bg)}.bs-popover-bottom>.popover-arrow,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow{top:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width))}.bs-popover-bottom>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::before,.bs-popover-bottom>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::after{border-width:0 calc(var(--bs-popover-arrow-width)*.5) var(--bs-popover-arrow-height)}.bs-popover-bottom>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::before{top:0;border-bottom-color:var(--bs-popover-arrow-border)}.bs-popover-bottom>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=bottom]>.popover-arrow::after{top:var(--bs-popover-border-width);border-bottom-color:var(--bs-popover-bg)}.bs-popover-bottom .popover-header::before,.bs-popover-auto[data-popper-placement^=bottom] .popover-header::before{position:absolute;top:0;left:50%;display:block;width:var(--bs-popover-arrow-width);margin-left:calc(-0.5*var(--bs-popover-arrow-width));content:"";border-bottom:var(--bs-popover-border-width) solid var(--bs-popover-header-bg)}.bs-popover-start>.popover-arrow,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow{right:calc(-1*(var(--bs-popover-arrow-height)) - var(--bs-popover-border-width));width:var(--bs-popover-arrow-height);height:var(--bs-popover-arrow-width)}.bs-popover-start>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::before,.bs-popover-start>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::after{border-width:calc(var(--bs-popover-arrow-width)*.5) 0 calc(var(--bs-popover-arrow-width)*.5) var(--bs-popover-arrow-height)}.bs-popover-start>.popover-arrow::before,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::before{right:0;border-left-color:var(--bs-popover-arrow-border)}.bs-popover-start>.popover-arrow::after,.bs-popover-auto[data-popper-placement^=left]>.popover-arrow::after{right:var(--bs-popover-border-width);border-left-color:var(--bs-popover-bg)}.popover-header{padding:var(--bs-popover-header-padding-y) var(--bs-popover-header-padding-x);margin-bottom:0;font-size:var(--bs-popover-header-font-size);color:var(--bs-popover-header-color);background-color:var(--bs-popover-header-bg);border-bottom:var(--bs-popover-border-width) solid var(--bs-popover-border-color)}.popover-header:empty{display:none}.popover-body{padding:var(--bs-popover-body-padding-y) var(--bs-popover-body-padding-x);color:var(--bs-popover-body-color)}.carousel{position:relative}.carousel.pointer-event{touch-action:pan-y;-webkit-touch-action:pan-y;-moz-touch-action:pan-y;-ms-touch-action:pan-y;-o-touch-action:pan-y}.carousel-inner{position:relative;width:100%;overflow:hidden}.carousel-inner::after{display:block;clear:both;content:""}.carousel-item{position:relative;display:none;float:left;width:100%;margin-right:-100%;backface-visibility:hidden;-webkit-backface-visibility:hidden;-moz-backface-visibility:hidden;-ms-backface-visibility:hidden;-o-backface-visibility:hidden;transition:transform .6s ease-in-out}@media(prefers-reduced-motion: reduce){.carousel-item{transition:none}}.carousel-item.active,.carousel-item-next,.carousel-item-prev{display:block}.carousel-item-next:not(.carousel-item-start),.active.carousel-item-end{transform:translateX(100%)}.carousel-item-prev:not(.carousel-item-end),.active.carousel-item-start{transform:translateX(-100%)}.carousel-fade .carousel-item{opacity:0;transition-property:opacity;transform:none}.carousel-fade .carousel-item.active,.carousel-fade .carousel-item-next.carousel-item-start,.carousel-fade .carousel-item-prev.carousel-item-end{z-index:1;opacity:1}.carousel-fade .active.carousel-item-start,.carousel-fade .active.carousel-item-end{z-index:0;opacity:0;transition:opacity 0s .6s}@media(prefers-reduced-motion: reduce){.carousel-fade .active.carousel-item-start,.carousel-fade .active.carousel-item-end{transition:none}}.carousel-control-prev,.carousel-control-next{position:absolute;top:0;bottom:0;z-index:1;display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;justify-content:center;-webkit-justify-content:center;width:15%;padding:0;color:#fff;text-align:center;background:none;border:0;opacity:.5;transition:opacity .15s ease}@media(prefers-reduced-motion: reduce){.carousel-control-prev,.carousel-control-next{transition:none}}.carousel-control-prev:hover,.carousel-control-prev:focus,.carousel-control-next:hover,.carousel-control-next:focus{color:#fff;text-decoration:none;outline:0;opacity:.9}.carousel-control-prev{left:0}.carousel-control-next{right:0}.carousel-control-prev-icon,.carousel-control-next-icon{display:inline-block;width:2rem;height:2rem;background-repeat:no-repeat;background-position:50%;background-size:100% 100%}.carousel-control-prev-icon{background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23fff'%3e%3cpath d='M11.354 1.646a.5.5 0 0 1 0 .708L5.707 8l5.647 5.646a.5.5 0 0 1-.708.708l-6-6a.5.5 0 0 1 0-.708l6-6a.5.5 0 0 1 .708 0z'/%3e%3c/svg%3e")}.carousel-control-next-icon{background-image:url("data:image/svg+xml,%3csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 16 16' fill='%23fff'%3e%3cpath d='M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z'/%3e%3c/svg%3e")}.carousel-indicators{position:absolute;right:0;bottom:0;left:0;z-index:2;display:flex;display:-webkit-flex;justify-content:center;-webkit-justify-content:center;padding:0;margin-right:15%;margin-bottom:1rem;margin-left:15%}.carousel-indicators [data-bs-target]{box-sizing:content-box;flex:0 1 auto;-webkit-flex:0 1 auto;width:30px;height:3px;padding:0;margin-right:3px;margin-left:3px;text-indent:-999px;cursor:pointer;background-color:#fff;background-clip:padding-box;border:0;border-top:10px solid rgba(0,0,0,0);border-bottom:10px solid rgba(0,0,0,0);opacity:.5;transition:opacity .6s ease}@media(prefers-reduced-motion: reduce){.carousel-indicators [data-bs-target]{transition:none}}.carousel-indicators .active{opacity:1}.carousel-caption{position:absolute;right:15%;bottom:1.25rem;left:15%;padding-top:1.25rem;padding-bottom:1.25rem;color:#fff;text-align:center}.carousel-dark .carousel-control-prev-icon,.carousel-dark .carousel-control-next-icon{filter:invert(1) grayscale(100)}.carousel-dark .carousel-indicators [data-bs-target]{background-color:#000}.carousel-dark .carousel-caption{color:#000}[data-bs-theme=dark] .carousel .carousel-control-prev-icon,[data-bs-theme=dark] .carousel .carousel-control-next-icon,[data-bs-theme=dark].carousel .carousel-control-prev-icon,[data-bs-theme=dark].carousel .carousel-control-next-icon{filter:invert(1) grayscale(100)}[data-bs-theme=dark] .carousel .carousel-indicators [data-bs-target],[data-bs-theme=dark].carousel .carousel-indicators [data-bs-target]{background-color:#000}[data-bs-theme=dark] .carousel .carousel-caption,[data-bs-theme=dark].carousel .carousel-caption{color:#000}.spinner-grow,.spinner-border{display:inline-block;width:var(--bs-spinner-width);height:var(--bs-spinner-height);vertical-align:var(--bs-spinner-vertical-align);border-radius:50%;animation:var(--bs-spinner-animation-speed) linear infinite var(--bs-spinner-animation-name)}@keyframes spinner-border{to{transform:rotate(360deg) /* rtl:ignore */}}.spinner-border{--bs-spinner-width: 2rem;--bs-spinner-height: 2rem;--bs-spinner-vertical-align: -0.125em;--bs-spinner-border-width: 0.25em;--bs-spinner-animation-speed: 0.75s;--bs-spinner-animation-name: spinner-border;border:var(--bs-spinner-border-width) solid currentcolor;border-right-color:rgba(0,0,0,0)}.spinner-border-sm{--bs-spinner-width: 1rem;--bs-spinner-height: 1rem;--bs-spinner-border-width: 0.2em}@keyframes spinner-grow{0%{transform:scale(0)}50%{opacity:1;transform:none}}.spinner-grow{--bs-spinner-width: 2rem;--bs-spinner-height: 2rem;--bs-spinner-vertical-align: -0.125em;--bs-spinner-animation-speed: 0.75s;--bs-spinner-animation-name: spinner-grow;background-color:currentcolor;opacity:0}.spinner-grow-sm{--bs-spinner-width: 1rem;--bs-spinner-height: 1rem}@media(prefers-reduced-motion: reduce){.spinner-border,.spinner-grow{--bs-spinner-animation-speed: 1.5s}}.offcanvas,.offcanvas-xxl,.offcanvas-xl,.offcanvas-lg,.offcanvas-md,.offcanvas-sm{--bs-offcanvas-zindex: 1045;--bs-offcanvas-width: 400px;--bs-offcanvas-height: 30vh;--bs-offcanvas-padding-x: 1rem;--bs-offcanvas-padding-y: 1rem;--bs-offcanvas-color: #FAF1E4;--bs-offcanvas-bg: #181818;--bs-offcanvas-border-width: 1px;--bs-offcanvas-border-color: rgba(0, 0, 0, 0.175);--bs-offcanvas-box-shadow: 0 0.125rem 0.25rem rgba(0, 0, 0, 0.075);--bs-offcanvas-transition: transform 0.3s ease-in-out;--bs-offcanvas-title-line-height: 1.5}@media(max-width: 575.98px){.offcanvas-sm{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 575.98px)and (prefers-reduced-motion: reduce){.offcanvas-sm{transition:none}}@media(max-width: 575.98px){.offcanvas-sm.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-sm.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-sm.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-sm.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-sm.showing,.offcanvas-sm.show:not(.hiding){transform:none}.offcanvas-sm.showing,.offcanvas-sm.hiding,.offcanvas-sm.show{visibility:visible}}@media(min-width: 576px){.offcanvas-sm{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-sm .offcanvas-header{display:none}.offcanvas-sm .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 767.98px){.offcanvas-md{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 767.98px)and (prefers-reduced-motion: reduce){.offcanvas-md{transition:none}}@media(max-width: 767.98px){.offcanvas-md.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-md.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-md.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-md.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-md.showing,.offcanvas-md.show:not(.hiding){transform:none}.offcanvas-md.showing,.offcanvas-md.hiding,.offcanvas-md.show{visibility:visible}}@media(min-width: 768px){.offcanvas-md{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-md .offcanvas-header{display:none}.offcanvas-md .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 991.98px){.offcanvas-lg{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 991.98px)and (prefers-reduced-motion: reduce){.offcanvas-lg{transition:none}}@media(max-width: 991.98px){.offcanvas-lg.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-lg.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-lg.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-lg.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-lg.showing,.offcanvas-lg.show:not(.hiding){transform:none}.offcanvas-lg.showing,.offcanvas-lg.hiding,.offcanvas-lg.show{visibility:visible}}@media(min-width: 992px){.offcanvas-lg{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-lg .offcanvas-header{display:none}.offcanvas-lg .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 1199.98px){.offcanvas-xl{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 1199.98px)and (prefers-reduced-motion: reduce){.offcanvas-xl{transition:none}}@media(max-width: 1199.98px){.offcanvas-xl.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-xl.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-xl.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-xl.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-xl.showing,.offcanvas-xl.show:not(.hiding){transform:none}.offcanvas-xl.showing,.offcanvas-xl.hiding,.offcanvas-xl.show{visibility:visible}}@media(min-width: 1200px){.offcanvas-xl{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-xl .offcanvas-header{display:none}.offcanvas-xl .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}@media(max-width: 1399.98px){.offcanvas-xxl{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}}@media(max-width: 1399.98px)and (prefers-reduced-motion: reduce){.offcanvas-xxl{transition:none}}@media(max-width: 1399.98px){.offcanvas-xxl.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas-xxl.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas-xxl.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas-xxl.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas-xxl.showing,.offcanvas-xxl.show:not(.hiding){transform:none}.offcanvas-xxl.showing,.offcanvas-xxl.hiding,.offcanvas-xxl.show{visibility:visible}}@media(min-width: 1400px){.offcanvas-xxl{--bs-offcanvas-height: auto;--bs-offcanvas-border-width: 0;background-color:rgba(0,0,0,0) !important}.offcanvas-xxl .offcanvas-header{display:none}.offcanvas-xxl .offcanvas-body{display:flex;display:-webkit-flex;flex-grow:0;-webkit-flex-grow:0;padding:0;overflow-y:visible;background-color:rgba(0,0,0,0) !important}}.offcanvas{position:fixed;bottom:0;z-index:var(--bs-offcanvas-zindex);display:flex;display:-webkit-flex;flex-direction:column;-webkit-flex-direction:column;max-width:100%;color:var(--bs-offcanvas-color);visibility:hidden;background-color:var(--bs-offcanvas-bg);background-clip:padding-box;outline:0;transition:var(--bs-offcanvas-transition)}@media(prefers-reduced-motion: reduce){.offcanvas{transition:none}}.offcanvas.offcanvas-start{top:0;left:0;width:var(--bs-offcanvas-width);border-right:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(-100%)}.offcanvas.offcanvas-end{top:0;right:0;width:var(--bs-offcanvas-width);border-left:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateX(100%)}.offcanvas.offcanvas-top{top:0;right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-bottom:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(-100%)}.offcanvas.offcanvas-bottom{right:0;left:0;height:var(--bs-offcanvas-height);max-height:100%;border-top:var(--bs-offcanvas-border-width) solid var(--bs-offcanvas-border-color);transform:translateY(100%)}.offcanvas.showing,.offcanvas.show:not(.hiding){transform:none}.offcanvas.showing,.offcanvas.hiding,.offcanvas.show{visibility:visible}.offcanvas-backdrop{position:fixed;top:0;left:0;z-index:1040;width:100vw;height:100vh;background-color:#000}.offcanvas-backdrop.fade{opacity:0}.offcanvas-backdrop.show{opacity:.5}.offcanvas-header{display:flex;display:-webkit-flex;align-items:center;-webkit-align-items:center;justify-content:space-between;-webkit-justify-content:space-between;padding:var(--bs-offcanvas-padding-y) var(--bs-offcanvas-padding-x)}.offcanvas-header .btn-close{padding:calc(var(--bs-offcanvas-padding-y)*.5) calc(var(--bs-offcanvas-padding-x)*.5);margin-top:calc(-0.5*var(--bs-offcanvas-padding-y));margin-right:calc(-0.5*var(--bs-offcanvas-padding-x));margin-bottom:calc(-0.5*var(--bs-offcanvas-padding-y))}.offcanvas-title{margin-bottom:0;line-height:var(--bs-offcanvas-title-line-height)}.offcanvas-body{flex-grow:1;-webkit-flex-grow:1;padding:var(--bs-offcanvas-padding-y) var(--bs-offcanvas-padding-x);overflow-y:auto}.placeholder{display:inline-block;min-height:1em;vertical-align:middle;cursor:wait;background-color:currentcolor;opacity:.5}.placeholder.btn::before{display:inline-block;content:""}.placeholder-xs{min-height:.6em}.placeholder-sm{min-height:.8em}.placeholder-lg{min-height:1.2em}.placeholder-glow .placeholder{animation:placeholder-glow 2s ease-in-out infinite}@keyframes placeholder-glow{50%{opacity:.2}}.placeholder-wave{mask-image:linear-gradient(130deg, #000 55%, rgba(0, 0, 0, 0.8) 75%, #000 95%);-webkit-mask-image:linear-gradient(130deg, #000 55%, rgba(0, 0, 0, 0.8) 75%, #000 95%);mask-size:200% 100%;-webkit-mask-size:200% 100%;animation:placeholder-wave 2s linear infinite}@keyframes placeholder-wave{100%{mask-position:-200% 0%;-webkit-mask-position:-200% 0%}}.clearfix::after{display:block;clear:both;content:""}.text-bg-default{color:#fff !important;background-color:RGBA(var(--bs-default-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-primary{color:#fff !important;background-color:RGBA(var(--bs-primary-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-secondary{color:#fff !important;background-color:RGBA(var(--bs-secondary-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-success{color:#fff !important;background-color:RGBA(var(--bs-success-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-info{color:#fff !important;background-color:RGBA(var(--bs-info-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-warning{color:#fff !important;background-color:RGBA(var(--bs-warning-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-danger{color:#fff !important;background-color:RGBA(var(--bs-danger-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-light{color:#fff !important;background-color:RGBA(var(--bs-light-rgb), var(--bs-bg-opacity, 1)) !important}.text-bg-dark{color:#fff !important;background-color:RGBA(var(--bs-dark-rgb), var(--bs-bg-opacity, 1)) !important}.link-default{color:RGBA(var(--bs-default-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-default-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-default:hover,.link-default:focus{color:RGBA(42, 46, 51, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(42, 46, 51, var(--bs-link-underline-opacity, 1)) !important}.link-primary{color:RGBA(var(--bs-primary-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-primary-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-primary:hover,.link-primary:focus{color:RGBA(31, 102, 182, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(31, 102, 182, var(--bs-link-underline-opacity, 1)) !important}.link-secondary{color:RGBA(var(--bs-secondary-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-secondary-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-secondary:hover,.link-secondary:focus{color:RGBA(42, 46, 51, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(42, 46, 51, var(--bs-link-underline-opacity, 1)) !important}.link-success{color:RGBA(var(--bs-success-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-success-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-success:hover,.link-success:focus{color:RGBA(50, 146, 19, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(50, 146, 19, var(--bs-link-underline-opacity, 1)) !important}.link-info{color:RGBA(var(--bs-info-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-info-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-info:hover,.link-info:focus{color:RGBA(122, 67, 150, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(122, 67, 150, var(--bs-link-underline-opacity, 1)) !important}.link-warning{color:RGBA(var(--bs-warning-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-warning-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-warning:hover,.link-warning:focus{color:RGBA(204, 94, 19, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(204, 94, 19, var(--bs-link-underline-opacity, 1)) !important}.link-danger{color:RGBA(var(--bs-danger-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-danger-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-danger:hover,.link-danger:focus{color:RGBA(204, 0, 46, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(204, 0, 46, var(--bs-link-underline-opacity, 1)) !important}.link-light{color:RGBA(var(--bs-light-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-light-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-light:hover,.link-light:focus{color:RGBA(66, 66, 66, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(66, 66, 66, var(--bs-link-underline-opacity, 1)) !important}.link-dark{color:RGBA(var(--bs-dark-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-dark-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-dark:hover,.link-dark:focus{color:RGBA(42, 46, 51, var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(42, 46, 51, var(--bs-link-underline-opacity, 1)) !important}.link-body-emphasis{color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-opacity, 1)) !important;text-decoration-color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-body-emphasis:hover,.link-body-emphasis:focus{color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-opacity, 0.75)) !important;text-decoration-color:RGBA(var(--bs-emphasis-color-rgb), var(--bs-link-underline-opacity, 0.75)) !important}.focus-ring:focus{outline:0;box-shadow:var(--bs-focus-ring-x, 0) var(--bs-focus-ring-y, 0) var(--bs-focus-ring-blur, 0) var(--bs-focus-ring-width) var(--bs-focus-ring-color)}.icon-link{display:inline-flex;gap:.375rem;align-items:center;-webkit-align-items:center;text-decoration-color:rgba(var(--bs-link-color-rgb), var(--bs-link-opacity, 0.5));text-underline-offset:.25em;backface-visibility:hidden;-webkit-backface-visibility:hidden;-moz-backface-visibility:hidden;-ms-backface-visibility:hidden;-o-backface-visibility:hidden}.icon-link>.bi{flex-shrink:0;-webkit-flex-shrink:0;width:1em;height:1em;fill:currentcolor;transition:.2s ease-in-out transform}@media(prefers-reduced-motion: reduce){.icon-link>.bi{transition:none}}.icon-link-hover:hover>.bi,.icon-link-hover:focus-visible>.bi{transform:var(--bs-icon-link-transform, translate3d(0.25em, 0, 0))}.ratio{position:relative;width:100%}.ratio::before{display:block;padding-top:var(--bs-aspect-ratio);content:""}.ratio>*{position:absolute;top:0;left:0;width:100%;height:100%}.ratio-1x1{--bs-aspect-ratio: 100%}.ratio-4x3{--bs-aspect-ratio: 75%}.ratio-16x9{--bs-aspect-ratio: 56.25%}.ratio-21x9{--bs-aspect-ratio: 42.8571428571%}.fixed-top{position:fixed;top:0;right:0;left:0;z-index:1030}.fixed-bottom{position:fixed;right:0;bottom:0;left:0;z-index:1030}.sticky-top{position:sticky;top:0;z-index:1020}.sticky-bottom{position:sticky;bottom:0;z-index:1020}@media(min-width: 576px){.sticky-sm-top{position:sticky;top:0;z-index:1020}.sticky-sm-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 768px){.sticky-md-top{position:sticky;top:0;z-index:1020}.sticky-md-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 992px){.sticky-lg-top{position:sticky;top:0;z-index:1020}.sticky-lg-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 1200px){.sticky-xl-top{position:sticky;top:0;z-index:1020}.sticky-xl-bottom{position:sticky;bottom:0;z-index:1020}}@media(min-width: 1400px){.sticky-xxl-top{position:sticky;top:0;z-index:1020}.sticky-xxl-bottom{position:sticky;bottom:0;z-index:1020}}.hstack{display:flex;display:-webkit-flex;flex-direction:row;-webkit-flex-direction:row;align-items:center;-webkit-align-items:center;align-self:stretch;-webkit-align-self:stretch}.vstack{display:flex;display:-webkit-flex;flex:1 1 auto;-webkit-flex:1 1 auto;flex-direction:column;-webkit-flex-direction:column;align-self:stretch;-webkit-align-self:stretch}.visually-hidden,.visually-hidden-focusable:not(:focus):not(:focus-within){width:1px !important;height:1px !important;padding:0 !important;margin:-1px !important;overflow:hidden !important;clip:rect(0, 0, 0, 0) !important;white-space:nowrap !important;border:0 !important}.visually-hidden:not(caption),.visually-hidden-focusable:not(:focus):not(:focus-within):not(caption){position:absolute !important}.stretched-link::after{position:absolute;top:0;right:0;bottom:0;left:0;z-index:1;content:""}.text-truncate{overflow:hidden;text-overflow:ellipsis;white-space:nowrap}.vr{display:inline-block;align-self:stretch;-webkit-align-self:stretch;width:1px;min-height:1em;background-color:currentcolor;opacity:.25}.align-baseline{vertical-align:baseline !important}.align-top{vertical-align:top !important}.align-middle{vertical-align:middle !important}.align-bottom{vertical-align:bottom !important}.align-text-bottom{vertical-align:text-bottom !important}.align-text-top{vertical-align:text-top !important}.float-start{float:left !important}.float-end{float:right !important}.float-none{float:none !important}.object-fit-contain{object-fit:contain !important}.object-fit-cover{object-fit:cover !important}.object-fit-fill{object-fit:fill !important}.object-fit-scale{object-fit:scale-down !important}.object-fit-none{object-fit:none !important}.opacity-0{opacity:0 !important}.opacity-25{opacity:.25 !important}.opacity-50{opacity:.5 !important}.opacity-75{opacity:.75 !important}.opacity-100{opacity:1 !important}.overflow-auto{overflow:auto !important}.overflow-hidden{overflow:hidden !important}.overflow-visible{overflow:visible !important}.overflow-scroll{overflow:scroll !important}.overflow-x-auto{overflow-x:auto !important}.overflow-x-hidden{overflow-x:hidden !important}.overflow-x-visible{overflow-x:visible !important}.overflow-x-scroll{overflow-x:scroll !important}.overflow-y-auto{overflow-y:auto !important}.overflow-y-hidden{overflow-y:hidden !important}.overflow-y-visible{overflow-y:visible !important}.overflow-y-scroll{overflow-y:scroll !important}.d-inline{display:inline !important}.d-inline-block{display:inline-block !important}.d-block{display:block !important}.d-grid{display:grid !important}.d-inline-grid{display:inline-grid !important}.d-table{display:table !important}.d-table-row{display:table-row !important}.d-table-cell{display:table-cell !important}.d-flex{display:flex !important}.d-inline-flex{display:inline-flex !important}.d-none{display:none !important}.shadow{box-shadow:0 .5rem 1rem rgba(0,0,0,.15) !important}.shadow-sm{box-shadow:0 .125rem .25rem rgba(0,0,0,.075) !important}.shadow-lg{box-shadow:0 1rem 3rem rgba(0,0,0,.175) !important}.shadow-none{box-shadow:none !important}.focus-ring-default{--bs-focus-ring-color: rgba(var(--bs-default-rgb), var(--bs-focus-ring-opacity))}.focus-ring-primary{--bs-focus-ring-color: rgba(var(--bs-primary-rgb), var(--bs-focus-ring-opacity))}.focus-ring-secondary{--bs-focus-ring-color: rgba(var(--bs-secondary-rgb), var(--bs-focus-ring-opacity))}.focus-ring-success{--bs-focus-ring-color: rgba(var(--bs-success-rgb), var(--bs-focus-ring-opacity))}.focus-ring-info{--bs-focus-ring-color: rgba(var(--bs-info-rgb), var(--bs-focus-ring-opacity))}.focus-ring-warning{--bs-focus-ring-color: rgba(var(--bs-warning-rgb), var(--bs-focus-ring-opacity))}.focus-ring-danger{--bs-focus-ring-color: rgba(var(--bs-danger-rgb), var(--bs-focus-ring-opacity))}.focus-ring-light{--bs-focus-ring-color: rgba(var(--bs-light-rgb), var(--bs-focus-ring-opacity))}.focus-ring-dark{--bs-focus-ring-color: rgba(var(--bs-dark-rgb), var(--bs-focus-ring-opacity))}.position-static{position:static !important}.position-relative{position:relative !important}.position-absolute{position:absolute !important}.position-fixed{position:fixed !important}.position-sticky{position:sticky !important}.top-0{top:0 !important}.top-50{top:50% !important}.top-100{top:100% !important}.bottom-0{bottom:0 !important}.bottom-50{bottom:50% !important}.bottom-100{bottom:100% !important}.start-0{left:0 !important}.start-50{left:50% !important}.start-100{left:100% !important}.end-0{right:0 !important}.end-50{right:50% !important}.end-100{right:100% !important}.translate-middle{transform:translate(-50%, -50%) !important}.translate-middle-x{transform:translateX(-50%) !important}.translate-middle-y{transform:translateY(-50%) !important}.border{border:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-0{border:0 !important}.border-top{border-top:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-top-0{border-top:0 !important}.border-end{border-right:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-end-0{border-right:0 !important}.border-bottom{border-bottom:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-bottom-0{border-bottom:0 !important}.border-start{border-left:var(--bs-border-width) var(--bs-border-style) var(--bs-border-color) !important}.border-start-0{border-left:0 !important}.border-default{--bs-border-opacity: 1;border-color:rgba(var(--bs-default-rgb), var(--bs-border-opacity)) !important}.border-primary{--bs-border-opacity: 1;border-color:rgba(var(--bs-primary-rgb), var(--bs-border-opacity)) !important}.border-secondary{--bs-border-opacity: 1;border-color:rgba(var(--bs-secondary-rgb), var(--bs-border-opacity)) !important}.border-success{--bs-border-opacity: 1;border-color:rgba(var(--bs-success-rgb), var(--bs-border-opacity)) !important}.border-info{--bs-border-opacity: 1;border-color:rgba(var(--bs-info-rgb), var(--bs-border-opacity)) !important}.border-warning{--bs-border-opacity: 1;border-color:rgba(var(--bs-warning-rgb), var(--bs-border-opacity)) !important}.border-danger{--bs-border-opacity: 1;border-color:rgba(var(--bs-danger-rgb), var(--bs-border-opacity)) !important}.border-light{--bs-border-opacity: 1;border-color:rgba(var(--bs-light-rgb), var(--bs-border-opacity)) !important}.border-dark{--bs-border-opacity: 1;border-color:rgba(var(--bs-dark-rgb), var(--bs-border-opacity)) !important}.border-black{--bs-border-opacity: 1;border-color:rgba(var(--bs-black-rgb), var(--bs-border-opacity)) !important}.border-white{--bs-border-opacity: 1;border-color:rgba(var(--bs-white-rgb), var(--bs-border-opacity)) !important}.border-primary-subtle{border-color:var(--bs-primary-border-subtle) !important}.border-secondary-subtle{border-color:var(--bs-secondary-border-subtle) !important}.border-success-subtle{border-color:var(--bs-success-border-subtle) !important}.border-info-subtle{border-color:var(--bs-info-border-subtle) !important}.border-warning-subtle{border-color:var(--bs-warning-border-subtle) !important}.border-danger-subtle{border-color:var(--bs-danger-border-subtle) !important}.border-light-subtle{border-color:var(--bs-light-border-subtle) !important}.border-dark-subtle{border-color:var(--bs-dark-border-subtle) !important}.border-1{border-width:1px !important}.border-2{border-width:2px !important}.border-3{border-width:3px !important}.border-4{border-width:4px !important}.border-5{border-width:5px !important}.border-opacity-10{--bs-border-opacity: 0.1}.border-opacity-25{--bs-border-opacity: 0.25}.border-opacity-50{--bs-border-opacity: 0.5}.border-opacity-75{--bs-border-opacity: 0.75}.border-opacity-100{--bs-border-opacity: 1}.w-25{width:25% !important}.w-50{width:50% !important}.w-75{width:75% !important}.w-100{width:100% !important}.w-auto{width:auto !important}.mw-100{max-width:100% !important}.vw-100{width:100vw !important}.min-vw-100{min-width:100vw !important}.h-25{height:25% !important}.h-50{height:50% !important}.h-75{height:75% !important}.h-100{height:100% !important}.h-auto{height:auto !important}.mh-100{max-height:100% !important}.vh-100{height:100vh !important}.min-vh-100{min-height:100vh !important}.flex-fill{flex:1 1 auto !important}.flex-row{flex-direction:row !important}.flex-column{flex-direction:column !important}.flex-row-reverse{flex-direction:row-reverse !important}.flex-column-reverse{flex-direction:column-reverse !important}.flex-grow-0{flex-grow:0 !important}.flex-grow-1{flex-grow:1 !important}.flex-shrink-0{flex-shrink:0 !important}.flex-shrink-1{flex-shrink:1 !important}.flex-wrap{flex-wrap:wrap !important}.flex-nowrap{flex-wrap:nowrap !important}.flex-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-start{justify-content:flex-start !important}.justify-content-end{justify-content:flex-end !important}.justify-content-center{justify-content:center !important}.justify-content-between{justify-content:space-between !important}.justify-content-around{justify-content:space-around !important}.justify-content-evenly{justify-content:space-evenly !important}.align-items-start{align-items:flex-start !important}.align-items-end{align-items:flex-end !important}.align-items-center{align-items:center !important}.align-items-baseline{align-items:baseline !important}.align-items-stretch{align-items:stretch !important}.align-content-start{align-content:flex-start !important}.align-content-end{align-content:flex-end !important}.align-content-center{align-content:center !important}.align-content-between{align-content:space-between !important}.align-content-around{align-content:space-around !important}.align-content-stretch{align-content:stretch !important}.align-self-auto{align-self:auto !important}.align-self-start{align-self:flex-start !important}.align-self-end{align-self:flex-end !important}.align-self-center{align-self:center !important}.align-self-baseline{align-self:baseline !important}.align-self-stretch{align-self:stretch !important}.order-first{order:-1 !important}.order-0{order:0 !important}.order-1{order:1 !important}.order-2{order:2 !important}.order-3{order:3 !important}.order-4{order:4 !important}.order-5{order:5 !important}.order-last{order:6 !important}.m-0{margin:0 !important}.m-1{margin:.25rem !important}.m-2{margin:.5rem !important}.m-3{margin:1rem !important}.m-4{margin:1.5rem !important}.m-5{margin:3rem !important}.m-auto{margin:auto !important}.mx-0{margin-right:0 !important;margin-left:0 !important}.mx-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-3{margin-right:1rem !important;margin-left:1rem !important}.mx-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-5{margin-right:3rem !important;margin-left:3rem !important}.mx-auto{margin-right:auto !important;margin-left:auto !important}.my-0{margin-top:0 !important;margin-bottom:0 !important}.my-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-0{margin-top:0 !important}.mt-1{margin-top:.25rem !important}.mt-2{margin-top:.5rem !important}.mt-3{margin-top:1rem !important}.mt-4{margin-top:1.5rem !important}.mt-5{margin-top:3rem !important}.mt-auto{margin-top:auto !important}.me-0{margin-right:0 !important}.me-1{margin-right:.25rem !important}.me-2{margin-right:.5rem !important}.me-3{margin-right:1rem !important}.me-4{margin-right:1.5rem !important}.me-5{margin-right:3rem !important}.me-auto{margin-right:auto !important}.mb-0{margin-bottom:0 !important}.mb-1{margin-bottom:.25rem !important}.mb-2{margin-bottom:.5rem !important}.mb-3{margin-bottom:1rem !important}.mb-4{margin-bottom:1.5rem !important}.mb-5{margin-bottom:3rem !important}.mb-auto{margin-bottom:auto !important}.ms-0{margin-left:0 !important}.ms-1{margin-left:.25rem !important}.ms-2{margin-left:.5rem !important}.ms-3{margin-left:1rem !important}.ms-4{margin-left:1.5rem !important}.ms-5{margin-left:3rem !important}.ms-auto{margin-left:auto !important}.p-0{padding:0 !important}.p-1{padding:.25rem !important}.p-2{padding:.5rem !important}.p-3{padding:1rem !important}.p-4{padding:1.5rem !important}.p-5{padding:3rem !important}.px-0{padding-right:0 !important;padding-left:0 !important}.px-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-3{padding-right:1rem !important;padding-left:1rem !important}.px-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-5{padding-right:3rem !important;padding-left:3rem !important}.py-0{padding-top:0 !important;padding-bottom:0 !important}.py-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-0{padding-top:0 !important}.pt-1{padding-top:.25rem !important}.pt-2{padding-top:.5rem !important}.pt-3{padding-top:1rem !important}.pt-4{padding-top:1.5rem !important}.pt-5{padding-top:3rem !important}.pe-0{padding-right:0 !important}.pe-1{padding-right:.25rem !important}.pe-2{padding-right:.5rem !important}.pe-3{padding-right:1rem !important}.pe-4{padding-right:1.5rem !important}.pe-5{padding-right:3rem !important}.pb-0{padding-bottom:0 !important}.pb-1{padding-bottom:.25rem !important}.pb-2{padding-bottom:.5rem !important}.pb-3{padding-bottom:1rem !important}.pb-4{padding-bottom:1.5rem !important}.pb-5{padding-bottom:3rem !important}.ps-0{padding-left:0 !important}.ps-1{padding-left:.25rem !important}.ps-2{padding-left:.5rem !important}.ps-3{padding-left:1rem !important}.ps-4{padding-left:1.5rem !important}.ps-5{padding-left:3rem !important}.gap-0{gap:0 !important}.gap-1{gap:.25rem !important}.gap-2{gap:.5rem !important}.gap-3{gap:1rem !important}.gap-4{gap:1.5rem !important}.gap-5{gap:3rem !important}.row-gap-0{row-gap:0 !important}.row-gap-1{row-gap:.25rem !important}.row-gap-2{row-gap:.5rem !important}.row-gap-3{row-gap:1rem !important}.row-gap-4{row-gap:1.5rem !important}.row-gap-5{row-gap:3rem !important}.column-gap-0{column-gap:0 !important}.column-gap-1{column-gap:.25rem !important}.column-gap-2{column-gap:.5rem !important}.column-gap-3{column-gap:1rem !important}.column-gap-4{column-gap:1.5rem !important}.column-gap-5{column-gap:3rem !important}.font-monospace{font-family:var(--bs-font-monospace) !important}.fs-1{font-size:calc(1.325rem + 0.9vw) !important}.fs-2{font-size:calc(1.29rem + 0.48vw) !important}.fs-3{font-size:calc(1.27rem + 0.24vw) !important}.fs-4{font-size:1.25rem !important}.fs-5{font-size:1.1rem !important}.fs-6{font-size:1rem !important}.fst-italic{font-style:italic !important}.fst-normal{font-style:normal !important}.fw-lighter{font-weight:lighter !important}.fw-light{font-weight:300 !important}.fw-normal{font-weight:400 !important}.fw-medium{font-weight:500 !important}.fw-semibold{font-weight:600 !important}.fw-bold{font-weight:700 !important}.fw-bolder{font-weight:bolder !important}.lh-1{line-height:1 !important}.lh-sm{line-height:1.25 !important}.lh-base{line-height:1.5 !important}.lh-lg{line-height:2 !important}.text-start{text-align:left !important}.text-end{text-align:right !important}.text-center{text-align:center !important}.text-decoration-none{text-decoration:none !important}.text-decoration-underline{text-decoration:underline !important}.text-decoration-line-through{text-decoration:line-through !important}.text-lowercase{text-transform:lowercase !important}.text-uppercase{text-transform:uppercase !important}.text-capitalize{text-transform:capitalize !important}.text-wrap{white-space:normal !important}.text-nowrap{white-space:nowrap !important}.text-break{word-wrap:break-word !important;word-break:break-word !important}.text-default{--bs-text-opacity: 1;color:rgba(var(--bs-default-rgb), var(--bs-text-opacity)) !important}.text-primary{--bs-text-opacity: 1;color:rgba(var(--bs-primary-rgb), var(--bs-text-opacity)) !important}.text-secondary{--bs-text-opacity: 1;color:rgba(var(--bs-secondary-rgb), var(--bs-text-opacity)) !important}.text-success{--bs-text-opacity: 1;color:rgba(var(--bs-success-rgb), var(--bs-text-opacity)) !important}.text-info{--bs-text-opacity: 1;color:rgba(var(--bs-info-rgb), var(--bs-text-opacity)) !important}.text-warning{--bs-text-opacity: 1;color:rgba(var(--bs-warning-rgb), var(--bs-text-opacity)) !important}.text-danger{--bs-text-opacity: 1;color:rgba(var(--bs-danger-rgb), var(--bs-text-opacity)) !important}.text-light{--bs-text-opacity: 1;color:rgba(var(--bs-light-rgb), var(--bs-text-opacity)) !important}.text-dark{--bs-text-opacity: 1;color:rgba(var(--bs-dark-rgb), var(--bs-text-opacity)) !important}.text-black{--bs-text-opacity: 1;color:rgba(var(--bs-black-rgb), var(--bs-text-opacity)) !important}.text-white{--bs-text-opacity: 1;color:rgba(var(--bs-white-rgb), var(--bs-text-opacity)) !important}.text-body{--bs-text-opacity: 1;color:rgba(var(--bs-body-color-rgb), var(--bs-text-opacity)) !important}.text-muted{--bs-text-opacity: 1;color:var(--bs-secondary-color) !important}.text-black-50{--bs-text-opacity: 1;color:rgba(0,0,0,.5) !important}.text-white-50{--bs-text-opacity: 1;color:rgba(255,255,255,.5) !important}.text-body-secondary{--bs-text-opacity: 1;color:var(--bs-secondary-color) !important}.text-body-tertiary{--bs-text-opacity: 1;color:var(--bs-tertiary-color) !important}.text-body-emphasis{--bs-text-opacity: 1;color:var(--bs-emphasis-color) !important}.text-reset{--bs-text-opacity: 1;color:inherit !important}.text-opacity-25{--bs-text-opacity: 0.25}.text-opacity-50{--bs-text-opacity: 0.5}.text-opacity-75{--bs-text-opacity: 0.75}.text-opacity-100{--bs-text-opacity: 1}.text-primary-emphasis{color:var(--bs-primary-text-emphasis) !important}.text-secondary-emphasis{color:var(--bs-secondary-text-emphasis) !important}.text-success-emphasis{color:var(--bs-success-text-emphasis) !important}.text-info-emphasis{color:var(--bs-info-text-emphasis) !important}.text-warning-emphasis{color:var(--bs-warning-text-emphasis) !important}.text-danger-emphasis{color:var(--bs-danger-text-emphasis) !important}.text-light-emphasis{color:var(--bs-light-text-emphasis) !important}.text-dark-emphasis{color:var(--bs-dark-text-emphasis) !important}.link-opacity-10{--bs-link-opacity: 0.1}.link-opacity-10-hover:hover{--bs-link-opacity: 0.1}.link-opacity-25{--bs-link-opacity: 0.25}.link-opacity-25-hover:hover{--bs-link-opacity: 0.25}.link-opacity-50{--bs-link-opacity: 0.5}.link-opacity-50-hover:hover{--bs-link-opacity: 0.5}.link-opacity-75{--bs-link-opacity: 0.75}.link-opacity-75-hover:hover{--bs-link-opacity: 0.75}.link-opacity-100{--bs-link-opacity: 1}.link-opacity-100-hover:hover{--bs-link-opacity: 1}.link-offset-1{text-underline-offset:.125em !important}.link-offset-1-hover:hover{text-underline-offset:.125em !important}.link-offset-2{text-underline-offset:.25em !important}.link-offset-2-hover:hover{text-underline-offset:.25em !important}.link-offset-3{text-underline-offset:.375em !important}.link-offset-3-hover:hover{text-underline-offset:.375em !important}.link-underline-default{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-default-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-primary{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-primary-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-secondary{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-secondary-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-success{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-success-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-info{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-info-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-warning{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-warning-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-danger{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-danger-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-light{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-light-rgb), var(--bs-link-underline-opacity)) !important}.link-underline-dark{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-dark-rgb), var(--bs-link-underline-opacity)) !important}.link-underline{--bs-link-underline-opacity: 1;text-decoration-color:rgba(var(--bs-link-color-rgb), var(--bs-link-underline-opacity, 1)) !important}.link-underline-opacity-0{--bs-link-underline-opacity: 0}.link-underline-opacity-0-hover:hover{--bs-link-underline-opacity: 0}.link-underline-opacity-10{--bs-link-underline-opacity: 0.1}.link-underline-opacity-10-hover:hover{--bs-link-underline-opacity: 0.1}.link-underline-opacity-25{--bs-link-underline-opacity: 0.25}.link-underline-opacity-25-hover:hover{--bs-link-underline-opacity: 0.25}.link-underline-opacity-50{--bs-link-underline-opacity: 0.5}.link-underline-opacity-50-hover:hover{--bs-link-underline-opacity: 0.5}.link-underline-opacity-75{--bs-link-underline-opacity: 0.75}.link-underline-opacity-75-hover:hover{--bs-link-underline-opacity: 0.75}.link-underline-opacity-100{--bs-link-underline-opacity: 1}.link-underline-opacity-100-hover:hover{--bs-link-underline-opacity: 1}.bg-default{--bs-bg-opacity: 1;background-color:rgba(var(--bs-default-rgb), var(--bs-bg-opacity)) !important}.bg-primary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-primary-rgb), var(--bs-bg-opacity)) !important}.bg-secondary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-secondary-rgb), var(--bs-bg-opacity)) !important}.bg-success{--bs-bg-opacity: 1;background-color:rgba(var(--bs-success-rgb), var(--bs-bg-opacity)) !important}.bg-info{--bs-bg-opacity: 1;background-color:rgba(var(--bs-info-rgb), var(--bs-bg-opacity)) !important}.bg-warning{--bs-bg-opacity: 1;background-color:rgba(var(--bs-warning-rgb), var(--bs-bg-opacity)) !important}.bg-danger{--bs-bg-opacity: 1;background-color:rgba(var(--bs-danger-rgb), var(--bs-bg-opacity)) !important}.bg-light{--bs-bg-opacity: 1;background-color:rgba(var(--bs-light-rgb), var(--bs-bg-opacity)) !important}.bg-dark{--bs-bg-opacity: 1;background-color:rgba(var(--bs-dark-rgb), var(--bs-bg-opacity)) !important}.bg-black{--bs-bg-opacity: 1;background-color:rgba(var(--bs-black-rgb), var(--bs-bg-opacity)) !important}.bg-white{--bs-bg-opacity: 1;background-color:rgba(var(--bs-white-rgb), var(--bs-bg-opacity)) !important}.bg-body{--bs-bg-opacity: 1;background-color:rgba(var(--bs-body-bg-rgb), var(--bs-bg-opacity)) !important}.bg-transparent{--bs-bg-opacity: 1;background-color:rgba(0,0,0,0) !important}.bg-body-secondary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-secondary-bg-rgb), var(--bs-bg-opacity)) !important}.bg-body-tertiary{--bs-bg-opacity: 1;background-color:rgba(var(--bs-tertiary-bg-rgb), var(--bs-bg-opacity)) !important}.bg-opacity-10{--bs-bg-opacity: 0.1}.bg-opacity-25{--bs-bg-opacity: 0.25}.bg-opacity-50{--bs-bg-opacity: 0.5}.bg-opacity-75{--bs-bg-opacity: 0.75}.bg-opacity-100{--bs-bg-opacity: 1}.bg-primary-subtle{background-color:var(--bs-primary-bg-subtle) !important}.bg-secondary-subtle{background-color:var(--bs-secondary-bg-subtle) !important}.bg-success-subtle{background-color:var(--bs-success-bg-subtle) !important}.bg-info-subtle{background-color:var(--bs-info-bg-subtle) !important}.bg-warning-subtle{background-color:var(--bs-warning-bg-subtle) !important}.bg-danger-subtle{background-color:var(--bs-danger-bg-subtle) !important}.bg-light-subtle{background-color:var(--bs-light-bg-subtle) !important}.bg-dark-subtle{background-color:var(--bs-dark-bg-subtle) !important}.bg-gradient{background-image:var(--bs-gradient) !important}.user-select-all{user-select:all !important}.user-select-auto{user-select:auto !important}.user-select-none{user-select:none !important}.pe-none{pointer-events:none !important}.pe-auto{pointer-events:auto !important}.rounded{border-radius:var(--bs-border-radius) !important}.rounded-0{border-radius:0 !important}.rounded-1{border-radius:var(--bs-border-radius-sm) !important}.rounded-2{border-radius:var(--bs-border-radius) !important}.rounded-3{border-radius:var(--bs-border-radius-lg) !important}.rounded-4{border-radius:var(--bs-border-radius-xl) !important}.rounded-5{border-radius:var(--bs-border-radius-xxl) !important}.rounded-circle{border-radius:50% !important}.rounded-pill{border-radius:var(--bs-border-radius-pill) !important}.rounded-top{border-top-left-radius:var(--bs-border-radius) !important;border-top-right-radius:var(--bs-border-radius) !important}.rounded-top-0{border-top-left-radius:0 !important;border-top-right-radius:0 !important}.rounded-top-1{border-top-left-radius:var(--bs-border-radius-sm) !important;border-top-right-radius:var(--bs-border-radius-sm) !important}.rounded-top-2{border-top-left-radius:var(--bs-border-radius) !important;border-top-right-radius:var(--bs-border-radius) !important}.rounded-top-3{border-top-left-radius:var(--bs-border-radius-lg) !important;border-top-right-radius:var(--bs-border-radius-lg) !important}.rounded-top-4{border-top-left-radius:var(--bs-border-radius-xl) !important;border-top-right-radius:var(--bs-border-radius-xl) !important}.rounded-top-5{border-top-left-radius:var(--bs-border-radius-xxl) !important;border-top-right-radius:var(--bs-border-radius-xxl) !important}.rounded-top-circle{border-top-left-radius:50% !important;border-top-right-radius:50% !important}.rounded-top-pill{border-top-left-radius:var(--bs-border-radius-pill) !important;border-top-right-radius:var(--bs-border-radius-pill) !important}.rounded-end{border-top-right-radius:var(--bs-border-radius) !important;border-bottom-right-radius:var(--bs-border-radius) !important}.rounded-end-0{border-top-right-radius:0 !important;border-bottom-right-radius:0 !important}.rounded-end-1{border-top-right-radius:var(--bs-border-radius-sm) !important;border-bottom-right-radius:var(--bs-border-radius-sm) !important}.rounded-end-2{border-top-right-radius:var(--bs-border-radius) !important;border-bottom-right-radius:var(--bs-border-radius) !important}.rounded-end-3{border-top-right-radius:var(--bs-border-radius-lg) !important;border-bottom-right-radius:var(--bs-border-radius-lg) !important}.rounded-end-4{border-top-right-radius:var(--bs-border-radius-xl) !important;border-bottom-right-radius:var(--bs-border-radius-xl) !important}.rounded-end-5{border-top-right-radius:var(--bs-border-radius-xxl) !important;border-bottom-right-radius:var(--bs-border-radius-xxl) !important}.rounded-end-circle{border-top-right-radius:50% !important;border-bottom-right-radius:50% !important}.rounded-end-pill{border-top-right-radius:var(--bs-border-radius-pill) !important;border-bottom-right-radius:var(--bs-border-radius-pill) !important}.rounded-bottom{border-bottom-right-radius:var(--bs-border-radius) !important;border-bottom-left-radius:var(--bs-border-radius) !important}.rounded-bottom-0{border-bottom-right-radius:0 !important;border-bottom-left-radius:0 !important}.rounded-bottom-1{border-bottom-right-radius:var(--bs-border-radius-sm) !important;border-bottom-left-radius:var(--bs-border-radius-sm) !important}.rounded-bottom-2{border-bottom-right-radius:var(--bs-border-radius) !important;border-bottom-left-radius:var(--bs-border-radius) !important}.rounded-bottom-3{border-bottom-right-radius:var(--bs-border-radius-lg) !important;border-bottom-left-radius:var(--bs-border-radius-lg) !important}.rounded-bottom-4{border-bottom-right-radius:var(--bs-border-radius-xl) !important;border-bottom-left-radius:var(--bs-border-radius-xl) !important}.rounded-bottom-5{border-bottom-right-radius:var(--bs-border-radius-xxl) !important;border-bottom-left-radius:var(--bs-border-radius-xxl) !important}.rounded-bottom-circle{border-bottom-right-radius:50% !important;border-bottom-left-radius:50% !important}.rounded-bottom-pill{border-bottom-right-radius:var(--bs-border-radius-pill) !important;border-bottom-left-radius:var(--bs-border-radius-pill) !important}.rounded-start{border-bottom-left-radius:var(--bs-border-radius) !important;border-top-left-radius:var(--bs-border-radius) !important}.rounded-start-0{border-bottom-left-radius:0 !important;border-top-left-radius:0 !important}.rounded-start-1{border-bottom-left-radius:var(--bs-border-radius-sm) !important;border-top-left-radius:var(--bs-border-radius-sm) !important}.rounded-start-2{border-bottom-left-radius:var(--bs-border-radius) !important;border-top-left-radius:var(--bs-border-radius) !important}.rounded-start-3{border-bottom-left-radius:var(--bs-border-radius-lg) !important;border-top-left-radius:var(--bs-border-radius-lg) !important}.rounded-start-4{border-bottom-left-radius:var(--bs-border-radius-xl) !important;border-top-left-radius:var(--bs-border-radius-xl) !important}.rounded-start-5{border-bottom-left-radius:var(--bs-border-radius-xxl) !important;border-top-left-radius:var(--bs-border-radius-xxl) !important}.rounded-start-circle{border-bottom-left-radius:50% !important;border-top-left-radius:50% !important}.rounded-start-pill{border-bottom-left-radius:var(--bs-border-radius-pill) !important;border-top-left-radius:var(--bs-border-radius-pill) !important}.visible{visibility:visible !important}.invisible{visibility:hidden !important}.z-n1{z-index:-1 !important}.z-0{z-index:0 !important}.z-1{z-index:1 !important}.z-2{z-index:2 !important}.z-3{z-index:3 !important}@media(min-width: 576px){.float-sm-start{float:left !important}.float-sm-end{float:right !important}.float-sm-none{float:none !important}.object-fit-sm-contain{object-fit:contain !important}.object-fit-sm-cover{object-fit:cover !important}.object-fit-sm-fill{object-fit:fill !important}.object-fit-sm-scale{object-fit:scale-down !important}.object-fit-sm-none{object-fit:none !important}.d-sm-inline{display:inline !important}.d-sm-inline-block{display:inline-block !important}.d-sm-block{display:block !important}.d-sm-grid{display:grid !important}.d-sm-inline-grid{display:inline-grid !important}.d-sm-table{display:table !important}.d-sm-table-row{display:table-row !important}.d-sm-table-cell{display:table-cell !important}.d-sm-flex{display:flex !important}.d-sm-inline-flex{display:inline-flex !important}.d-sm-none{display:none !important}.flex-sm-fill{flex:1 1 auto !important}.flex-sm-row{flex-direction:row !important}.flex-sm-column{flex-direction:column !important}.flex-sm-row-reverse{flex-direction:row-reverse !important}.flex-sm-column-reverse{flex-direction:column-reverse !important}.flex-sm-grow-0{flex-grow:0 !important}.flex-sm-grow-1{flex-grow:1 !important}.flex-sm-shrink-0{flex-shrink:0 !important}.flex-sm-shrink-1{flex-shrink:1 !important}.flex-sm-wrap{flex-wrap:wrap !important}.flex-sm-nowrap{flex-wrap:nowrap !important}.flex-sm-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-sm-start{justify-content:flex-start !important}.justify-content-sm-end{justify-content:flex-end !important}.justify-content-sm-center{justify-content:center !important}.justify-content-sm-between{justify-content:space-between !important}.justify-content-sm-around{justify-content:space-around !important}.justify-content-sm-evenly{justify-content:space-evenly !important}.align-items-sm-start{align-items:flex-start !important}.align-items-sm-end{align-items:flex-end !important}.align-items-sm-center{align-items:center !important}.align-items-sm-baseline{align-items:baseline !important}.align-items-sm-stretch{align-items:stretch !important}.align-content-sm-start{align-content:flex-start !important}.align-content-sm-end{align-content:flex-end !important}.align-content-sm-center{align-content:center !important}.align-content-sm-between{align-content:space-between !important}.align-content-sm-around{align-content:space-around !important}.align-content-sm-stretch{align-content:stretch !important}.align-self-sm-auto{align-self:auto !important}.align-self-sm-start{align-self:flex-start !important}.align-self-sm-end{align-self:flex-end !important}.align-self-sm-center{align-self:center !important}.align-self-sm-baseline{align-self:baseline !important}.align-self-sm-stretch{align-self:stretch !important}.order-sm-first{order:-1 !important}.order-sm-0{order:0 !important}.order-sm-1{order:1 !important}.order-sm-2{order:2 !important}.order-sm-3{order:3 !important}.order-sm-4{order:4 !important}.order-sm-5{order:5 !important}.order-sm-last{order:6 !important}.m-sm-0{margin:0 !important}.m-sm-1{margin:.25rem !important}.m-sm-2{margin:.5rem !important}.m-sm-3{margin:1rem !important}.m-sm-4{margin:1.5rem !important}.m-sm-5{margin:3rem !important}.m-sm-auto{margin:auto !important}.mx-sm-0{margin-right:0 !important;margin-left:0 !important}.mx-sm-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-sm-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-sm-3{margin-right:1rem !important;margin-left:1rem !important}.mx-sm-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-sm-5{margin-right:3rem !important;margin-left:3rem !important}.mx-sm-auto{margin-right:auto !important;margin-left:auto !important}.my-sm-0{margin-top:0 !important;margin-bottom:0 !important}.my-sm-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-sm-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-sm-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-sm-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-sm-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-sm-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-sm-0{margin-top:0 !important}.mt-sm-1{margin-top:.25rem !important}.mt-sm-2{margin-top:.5rem !important}.mt-sm-3{margin-top:1rem !important}.mt-sm-4{margin-top:1.5rem !important}.mt-sm-5{margin-top:3rem !important}.mt-sm-auto{margin-top:auto !important}.me-sm-0{margin-right:0 !important}.me-sm-1{margin-right:.25rem !important}.me-sm-2{margin-right:.5rem !important}.me-sm-3{margin-right:1rem !important}.me-sm-4{margin-right:1.5rem !important}.me-sm-5{margin-right:3rem !important}.me-sm-auto{margin-right:auto !important}.mb-sm-0{margin-bottom:0 !important}.mb-sm-1{margin-bottom:.25rem !important}.mb-sm-2{margin-bottom:.5rem !important}.mb-sm-3{margin-bottom:1rem !important}.mb-sm-4{margin-bottom:1.5rem !important}.mb-sm-5{margin-bottom:3rem !important}.mb-sm-auto{margin-bottom:auto !important}.ms-sm-0{margin-left:0 !important}.ms-sm-1{margin-left:.25rem !important}.ms-sm-2{margin-left:.5rem !important}.ms-sm-3{margin-left:1rem !important}.ms-sm-4{margin-left:1.5rem !important}.ms-sm-5{margin-left:3rem !important}.ms-sm-auto{margin-left:auto !important}.p-sm-0{padding:0 !important}.p-sm-1{padding:.25rem !important}.p-sm-2{padding:.5rem !important}.p-sm-3{padding:1rem !important}.p-sm-4{padding:1.5rem !important}.p-sm-5{padding:3rem !important}.px-sm-0{padding-right:0 !important;padding-left:0 !important}.px-sm-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-sm-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-sm-3{padding-right:1rem !important;padding-left:1rem !important}.px-sm-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-sm-5{padding-right:3rem !important;padding-left:3rem !important}.py-sm-0{padding-top:0 !important;padding-bottom:0 !important}.py-sm-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-sm-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-sm-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-sm-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-sm-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-sm-0{padding-top:0 !important}.pt-sm-1{padding-top:.25rem !important}.pt-sm-2{padding-top:.5rem !important}.pt-sm-3{padding-top:1rem !important}.pt-sm-4{padding-top:1.5rem !important}.pt-sm-5{padding-top:3rem !important}.pe-sm-0{padding-right:0 !important}.pe-sm-1{padding-right:.25rem !important}.pe-sm-2{padding-right:.5rem !important}.pe-sm-3{padding-right:1rem !important}.pe-sm-4{padding-right:1.5rem !important}.pe-sm-5{padding-right:3rem !important}.pb-sm-0{padding-bottom:0 !important}.pb-sm-1{padding-bottom:.25rem !important}.pb-sm-2{padding-bottom:.5rem !important}.pb-sm-3{padding-bottom:1rem !important}.pb-sm-4{padding-bottom:1.5rem !important}.pb-sm-5{padding-bottom:3rem !important}.ps-sm-0{padding-left:0 !important}.ps-sm-1{padding-left:.25rem !important}.ps-sm-2{padding-left:.5rem !important}.ps-sm-3{padding-left:1rem !important}.ps-sm-4{padding-left:1.5rem !important}.ps-sm-5{padding-left:3rem !important}.gap-sm-0{gap:0 !important}.gap-sm-1{gap:.25rem !important}.gap-sm-2{gap:.5rem !important}.gap-sm-3{gap:1rem !important}.gap-sm-4{gap:1.5rem !important}.gap-sm-5{gap:3rem !important}.row-gap-sm-0{row-gap:0 !important}.row-gap-sm-1{row-gap:.25rem !important}.row-gap-sm-2{row-gap:.5rem !important}.row-gap-sm-3{row-gap:1rem !important}.row-gap-sm-4{row-gap:1.5rem !important}.row-gap-sm-5{row-gap:3rem !important}.column-gap-sm-0{column-gap:0 !important}.column-gap-sm-1{column-gap:.25rem !important}.column-gap-sm-2{column-gap:.5rem !important}.column-gap-sm-3{column-gap:1rem !important}.column-gap-sm-4{column-gap:1.5rem !important}.column-gap-sm-5{column-gap:3rem !important}.text-sm-start{text-align:left !important}.text-sm-end{text-align:right !important}.text-sm-center{text-align:center !important}}@media(min-width: 768px){.float-md-start{float:left !important}.float-md-end{float:right !important}.float-md-none{float:none !important}.object-fit-md-contain{object-fit:contain !important}.object-fit-md-cover{object-fit:cover !important}.object-fit-md-fill{object-fit:fill !important}.object-fit-md-scale{object-fit:scale-down !important}.object-fit-md-none{object-fit:none !important}.d-md-inline{display:inline !important}.d-md-inline-block{display:inline-block !important}.d-md-block{display:block !important}.d-md-grid{display:grid !important}.d-md-inline-grid{display:inline-grid !important}.d-md-table{display:table !important}.d-md-table-row{display:table-row !important}.d-md-table-cell{display:table-cell !important}.d-md-flex{display:flex !important}.d-md-inline-flex{display:inline-flex !important}.d-md-none{display:none !important}.flex-md-fill{flex:1 1 auto !important}.flex-md-row{flex-direction:row !important}.flex-md-column{flex-direction:column !important}.flex-md-row-reverse{flex-direction:row-reverse !important}.flex-md-column-reverse{flex-direction:column-reverse !important}.flex-md-grow-0{flex-grow:0 !important}.flex-md-grow-1{flex-grow:1 !important}.flex-md-shrink-0{flex-shrink:0 !important}.flex-md-shrink-1{flex-shrink:1 !important}.flex-md-wrap{flex-wrap:wrap !important}.flex-md-nowrap{flex-wrap:nowrap !important}.flex-md-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-md-start{justify-content:flex-start !important}.justify-content-md-end{justify-content:flex-end !important}.justify-content-md-center{justify-content:center !important}.justify-content-md-between{justify-content:space-between !important}.justify-content-md-around{justify-content:space-around !important}.justify-content-md-evenly{justify-content:space-evenly !important}.align-items-md-start{align-items:flex-start !important}.align-items-md-end{align-items:flex-end !important}.align-items-md-center{align-items:center !important}.align-items-md-baseline{align-items:baseline !important}.align-items-md-stretch{align-items:stretch !important}.align-content-md-start{align-content:flex-start !important}.align-content-md-end{align-content:flex-end !important}.align-content-md-center{align-content:center !important}.align-content-md-between{align-content:space-between !important}.align-content-md-around{align-content:space-around !important}.align-content-md-stretch{align-content:stretch !important}.align-self-md-auto{align-self:auto !important}.align-self-md-start{align-self:flex-start !important}.align-self-md-end{align-self:flex-end !important}.align-self-md-center{align-self:center !important}.align-self-md-baseline{align-self:baseline !important}.align-self-md-stretch{align-self:stretch !important}.order-md-first{order:-1 !important}.order-md-0{order:0 !important}.order-md-1{order:1 !important}.order-md-2{order:2 !important}.order-md-3{order:3 !important}.order-md-4{order:4 !important}.order-md-5{order:5 !important}.order-md-last{order:6 !important}.m-md-0{margin:0 !important}.m-md-1{margin:.25rem !important}.m-md-2{margin:.5rem !important}.m-md-3{margin:1rem !important}.m-md-4{margin:1.5rem !important}.m-md-5{margin:3rem !important}.m-md-auto{margin:auto !important}.mx-md-0{margin-right:0 !important;margin-left:0 !important}.mx-md-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-md-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-md-3{margin-right:1rem !important;margin-left:1rem !important}.mx-md-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-md-5{margin-right:3rem !important;margin-left:3rem !important}.mx-md-auto{margin-right:auto !important;margin-left:auto !important}.my-md-0{margin-top:0 !important;margin-bottom:0 !important}.my-md-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-md-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-md-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-md-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-md-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-md-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-md-0{margin-top:0 !important}.mt-md-1{margin-top:.25rem !important}.mt-md-2{margin-top:.5rem !important}.mt-md-3{margin-top:1rem !important}.mt-md-4{margin-top:1.5rem !important}.mt-md-5{margin-top:3rem !important}.mt-md-auto{margin-top:auto !important}.me-md-0{margin-right:0 !important}.me-md-1{margin-right:.25rem !important}.me-md-2{margin-right:.5rem !important}.me-md-3{margin-right:1rem !important}.me-md-4{margin-right:1.5rem !important}.me-md-5{margin-right:3rem !important}.me-md-auto{margin-right:auto !important}.mb-md-0{margin-bottom:0 !important}.mb-md-1{margin-bottom:.25rem !important}.mb-md-2{margin-bottom:.5rem !important}.mb-md-3{margin-bottom:1rem !important}.mb-md-4{margin-bottom:1.5rem !important}.mb-md-5{margin-bottom:3rem !important}.mb-md-auto{margin-bottom:auto !important}.ms-md-0{margin-left:0 !important}.ms-md-1{margin-left:.25rem !important}.ms-md-2{margin-left:.5rem !important}.ms-md-3{margin-left:1rem !important}.ms-md-4{margin-left:1.5rem !important}.ms-md-5{margin-left:3rem !important}.ms-md-auto{margin-left:auto !important}.p-md-0{padding:0 !important}.p-md-1{padding:.25rem !important}.p-md-2{padding:.5rem !important}.p-md-3{padding:1rem !important}.p-md-4{padding:1.5rem !important}.p-md-5{padding:3rem !important}.px-md-0{padding-right:0 !important;padding-left:0 !important}.px-md-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-md-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-md-3{padding-right:1rem !important;padding-left:1rem !important}.px-md-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-md-5{padding-right:3rem !important;padding-left:3rem !important}.py-md-0{padding-top:0 !important;padding-bottom:0 !important}.py-md-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-md-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-md-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-md-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-md-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-md-0{padding-top:0 !important}.pt-md-1{padding-top:.25rem !important}.pt-md-2{padding-top:.5rem !important}.pt-md-3{padding-top:1rem !important}.pt-md-4{padding-top:1.5rem !important}.pt-md-5{padding-top:3rem !important}.pe-md-0{padding-right:0 !important}.pe-md-1{padding-right:.25rem !important}.pe-md-2{padding-right:.5rem !important}.pe-md-3{padding-right:1rem !important}.pe-md-4{padding-right:1.5rem !important}.pe-md-5{padding-right:3rem !important}.pb-md-0{padding-bottom:0 !important}.pb-md-1{padding-bottom:.25rem !important}.pb-md-2{padding-bottom:.5rem !important}.pb-md-3{padding-bottom:1rem !important}.pb-md-4{padding-bottom:1.5rem !important}.pb-md-5{padding-bottom:3rem !important}.ps-md-0{padding-left:0 !important}.ps-md-1{padding-left:.25rem !important}.ps-md-2{padding-left:.5rem !important}.ps-md-3{padding-left:1rem !important}.ps-md-4{padding-left:1.5rem !important}.ps-md-5{padding-left:3rem !important}.gap-md-0{gap:0 !important}.gap-md-1{gap:.25rem !important}.gap-md-2{gap:.5rem !important}.gap-md-3{gap:1rem !important}.gap-md-4{gap:1.5rem !important}.gap-md-5{gap:3rem !important}.row-gap-md-0{row-gap:0 !important}.row-gap-md-1{row-gap:.25rem !important}.row-gap-md-2{row-gap:.5rem !important}.row-gap-md-3{row-gap:1rem !important}.row-gap-md-4{row-gap:1.5rem !important}.row-gap-md-5{row-gap:3rem !important}.column-gap-md-0{column-gap:0 !important}.column-gap-md-1{column-gap:.25rem !important}.column-gap-md-2{column-gap:.5rem !important}.column-gap-md-3{column-gap:1rem !important}.column-gap-md-4{column-gap:1.5rem !important}.column-gap-md-5{column-gap:3rem !important}.text-md-start{text-align:left !important}.text-md-end{text-align:right !important}.text-md-center{text-align:center !important}}@media(min-width: 992px){.float-lg-start{float:left !important}.float-lg-end{float:right !important}.float-lg-none{float:none !important}.object-fit-lg-contain{object-fit:contain !important}.object-fit-lg-cover{object-fit:cover !important}.object-fit-lg-fill{object-fit:fill !important}.object-fit-lg-scale{object-fit:scale-down !important}.object-fit-lg-none{object-fit:none !important}.d-lg-inline{display:inline !important}.d-lg-inline-block{display:inline-block !important}.d-lg-block{display:block !important}.d-lg-grid{display:grid !important}.d-lg-inline-grid{display:inline-grid !important}.d-lg-table{display:table !important}.d-lg-table-row{display:table-row !important}.d-lg-table-cell{display:table-cell !important}.d-lg-flex{display:flex !important}.d-lg-inline-flex{display:inline-flex !important}.d-lg-none{display:none !important}.flex-lg-fill{flex:1 1 auto !important}.flex-lg-row{flex-direction:row !important}.flex-lg-column{flex-direction:column !important}.flex-lg-row-reverse{flex-direction:row-reverse !important}.flex-lg-column-reverse{flex-direction:column-reverse !important}.flex-lg-grow-0{flex-grow:0 !important}.flex-lg-grow-1{flex-grow:1 !important}.flex-lg-shrink-0{flex-shrink:0 !important}.flex-lg-shrink-1{flex-shrink:1 !important}.flex-lg-wrap{flex-wrap:wrap !important}.flex-lg-nowrap{flex-wrap:nowrap !important}.flex-lg-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-lg-start{justify-content:flex-start !important}.justify-content-lg-end{justify-content:flex-end !important}.justify-content-lg-center{justify-content:center !important}.justify-content-lg-between{justify-content:space-between !important}.justify-content-lg-around{justify-content:space-around !important}.justify-content-lg-evenly{justify-content:space-evenly !important}.align-items-lg-start{align-items:flex-start !important}.align-items-lg-end{align-items:flex-end !important}.align-items-lg-center{align-items:center !important}.align-items-lg-baseline{align-items:baseline !important}.align-items-lg-stretch{align-items:stretch !important}.align-content-lg-start{align-content:flex-start !important}.align-content-lg-end{align-content:flex-end !important}.align-content-lg-center{align-content:center !important}.align-content-lg-between{align-content:space-between !important}.align-content-lg-around{align-content:space-around !important}.align-content-lg-stretch{align-content:stretch !important}.align-self-lg-auto{align-self:auto !important}.align-self-lg-start{align-self:flex-start !important}.align-self-lg-end{align-self:flex-end !important}.align-self-lg-center{align-self:center !important}.align-self-lg-baseline{align-self:baseline !important}.align-self-lg-stretch{align-self:stretch !important}.order-lg-first{order:-1 !important}.order-lg-0{order:0 !important}.order-lg-1{order:1 !important}.order-lg-2{order:2 !important}.order-lg-3{order:3 !important}.order-lg-4{order:4 !important}.order-lg-5{order:5 !important}.order-lg-last{order:6 !important}.m-lg-0{margin:0 !important}.m-lg-1{margin:.25rem !important}.m-lg-2{margin:.5rem !important}.m-lg-3{margin:1rem !important}.m-lg-4{margin:1.5rem !important}.m-lg-5{margin:3rem !important}.m-lg-auto{margin:auto !important}.mx-lg-0{margin-right:0 !important;margin-left:0 !important}.mx-lg-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-lg-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-lg-3{margin-right:1rem !important;margin-left:1rem !important}.mx-lg-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-lg-5{margin-right:3rem !important;margin-left:3rem !important}.mx-lg-auto{margin-right:auto !important;margin-left:auto !important}.my-lg-0{margin-top:0 !important;margin-bottom:0 !important}.my-lg-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-lg-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-lg-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-lg-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-lg-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-lg-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-lg-0{margin-top:0 !important}.mt-lg-1{margin-top:.25rem !important}.mt-lg-2{margin-top:.5rem !important}.mt-lg-3{margin-top:1rem !important}.mt-lg-4{margin-top:1.5rem !important}.mt-lg-5{margin-top:3rem !important}.mt-lg-auto{margin-top:auto !important}.me-lg-0{margin-right:0 !important}.me-lg-1{margin-right:.25rem !important}.me-lg-2{margin-right:.5rem !important}.me-lg-3{margin-right:1rem !important}.me-lg-4{margin-right:1.5rem !important}.me-lg-5{margin-right:3rem !important}.me-lg-auto{margin-right:auto !important}.mb-lg-0{margin-bottom:0 !important}.mb-lg-1{margin-bottom:.25rem !important}.mb-lg-2{margin-bottom:.5rem !important}.mb-lg-3{margin-bottom:1rem !important}.mb-lg-4{margin-bottom:1.5rem !important}.mb-lg-5{margin-bottom:3rem !important}.mb-lg-auto{margin-bottom:auto !important}.ms-lg-0{margin-left:0 !important}.ms-lg-1{margin-left:.25rem !important}.ms-lg-2{margin-left:.5rem !important}.ms-lg-3{margin-left:1rem !important}.ms-lg-4{margin-left:1.5rem !important}.ms-lg-5{margin-left:3rem !important}.ms-lg-auto{margin-left:auto !important}.p-lg-0{padding:0 !important}.p-lg-1{padding:.25rem !important}.p-lg-2{padding:.5rem !important}.p-lg-3{padding:1rem !important}.p-lg-4{padding:1.5rem !important}.p-lg-5{padding:3rem !important}.px-lg-0{padding-right:0 !important;padding-left:0 !important}.px-lg-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-lg-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-lg-3{padding-right:1rem !important;padding-left:1rem !important}.px-lg-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-lg-5{padding-right:3rem !important;padding-left:3rem !important}.py-lg-0{padding-top:0 !important;padding-bottom:0 !important}.py-lg-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-lg-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-lg-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-lg-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-lg-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-lg-0{padding-top:0 !important}.pt-lg-1{padding-top:.25rem !important}.pt-lg-2{padding-top:.5rem !important}.pt-lg-3{padding-top:1rem !important}.pt-lg-4{padding-top:1.5rem !important}.pt-lg-5{padding-top:3rem !important}.pe-lg-0{padding-right:0 !important}.pe-lg-1{padding-right:.25rem !important}.pe-lg-2{padding-right:.5rem !important}.pe-lg-3{padding-right:1rem !important}.pe-lg-4{padding-right:1.5rem !important}.pe-lg-5{padding-right:3rem !important}.pb-lg-0{padding-bottom:0 !important}.pb-lg-1{padding-bottom:.25rem !important}.pb-lg-2{padding-bottom:.5rem !important}.pb-lg-3{padding-bottom:1rem !important}.pb-lg-4{padding-bottom:1.5rem !important}.pb-lg-5{padding-bottom:3rem !important}.ps-lg-0{padding-left:0 !important}.ps-lg-1{padding-left:.25rem !important}.ps-lg-2{padding-left:.5rem !important}.ps-lg-3{padding-left:1rem !important}.ps-lg-4{padding-left:1.5rem !important}.ps-lg-5{padding-left:3rem !important}.gap-lg-0{gap:0 !important}.gap-lg-1{gap:.25rem !important}.gap-lg-2{gap:.5rem !important}.gap-lg-3{gap:1rem !important}.gap-lg-4{gap:1.5rem !important}.gap-lg-5{gap:3rem !important}.row-gap-lg-0{row-gap:0 !important}.row-gap-lg-1{row-gap:.25rem !important}.row-gap-lg-2{row-gap:.5rem !important}.row-gap-lg-3{row-gap:1rem !important}.row-gap-lg-4{row-gap:1.5rem !important}.row-gap-lg-5{row-gap:3rem !important}.column-gap-lg-0{column-gap:0 !important}.column-gap-lg-1{column-gap:.25rem !important}.column-gap-lg-2{column-gap:.5rem !important}.column-gap-lg-3{column-gap:1rem !important}.column-gap-lg-4{column-gap:1.5rem !important}.column-gap-lg-5{column-gap:3rem !important}.text-lg-start{text-align:left !important}.text-lg-end{text-align:right !important}.text-lg-center{text-align:center !important}}@media(min-width: 1200px){.float-xl-start{float:left !important}.float-xl-end{float:right !important}.float-xl-none{float:none !important}.object-fit-xl-contain{object-fit:contain !important}.object-fit-xl-cover{object-fit:cover !important}.object-fit-xl-fill{object-fit:fill !important}.object-fit-xl-scale{object-fit:scale-down !important}.object-fit-xl-none{object-fit:none !important}.d-xl-inline{display:inline !important}.d-xl-inline-block{display:inline-block !important}.d-xl-block{display:block !important}.d-xl-grid{display:grid !important}.d-xl-inline-grid{display:inline-grid !important}.d-xl-table{display:table !important}.d-xl-table-row{display:table-row !important}.d-xl-table-cell{display:table-cell !important}.d-xl-flex{display:flex !important}.d-xl-inline-flex{display:inline-flex !important}.d-xl-none{display:none !important}.flex-xl-fill{flex:1 1 auto !important}.flex-xl-row{flex-direction:row !important}.flex-xl-column{flex-direction:column !important}.flex-xl-row-reverse{flex-direction:row-reverse !important}.flex-xl-column-reverse{flex-direction:column-reverse !important}.flex-xl-grow-0{flex-grow:0 !important}.flex-xl-grow-1{flex-grow:1 !important}.flex-xl-shrink-0{flex-shrink:0 !important}.flex-xl-shrink-1{flex-shrink:1 !important}.flex-xl-wrap{flex-wrap:wrap !important}.flex-xl-nowrap{flex-wrap:nowrap !important}.flex-xl-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-xl-start{justify-content:flex-start !important}.justify-content-xl-end{justify-content:flex-end !important}.justify-content-xl-center{justify-content:center !important}.justify-content-xl-between{justify-content:space-between !important}.justify-content-xl-around{justify-content:space-around !important}.justify-content-xl-evenly{justify-content:space-evenly !important}.align-items-xl-start{align-items:flex-start !important}.align-items-xl-end{align-items:flex-end !important}.align-items-xl-center{align-items:center !important}.align-items-xl-baseline{align-items:baseline !important}.align-items-xl-stretch{align-items:stretch !important}.align-content-xl-start{align-content:flex-start !important}.align-content-xl-end{align-content:flex-end !important}.align-content-xl-center{align-content:center !important}.align-content-xl-between{align-content:space-between !important}.align-content-xl-around{align-content:space-around !important}.align-content-xl-stretch{align-content:stretch !important}.align-self-xl-auto{align-self:auto !important}.align-self-xl-start{align-self:flex-start !important}.align-self-xl-end{align-self:flex-end !important}.align-self-xl-center{align-self:center !important}.align-self-xl-baseline{align-self:baseline !important}.align-self-xl-stretch{align-self:stretch !important}.order-xl-first{order:-1 !important}.order-xl-0{order:0 !important}.order-xl-1{order:1 !important}.order-xl-2{order:2 !important}.order-xl-3{order:3 !important}.order-xl-4{order:4 !important}.order-xl-5{order:5 !important}.order-xl-last{order:6 !important}.m-xl-0{margin:0 !important}.m-xl-1{margin:.25rem !important}.m-xl-2{margin:.5rem !important}.m-xl-3{margin:1rem !important}.m-xl-4{margin:1.5rem !important}.m-xl-5{margin:3rem !important}.m-xl-auto{margin:auto !important}.mx-xl-0{margin-right:0 !important;margin-left:0 !important}.mx-xl-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-xl-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-xl-3{margin-right:1rem !important;margin-left:1rem !important}.mx-xl-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-xl-5{margin-right:3rem !important;margin-left:3rem !important}.mx-xl-auto{margin-right:auto !important;margin-left:auto !important}.my-xl-0{margin-top:0 !important;margin-bottom:0 !important}.my-xl-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-xl-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-xl-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-xl-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-xl-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-xl-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-xl-0{margin-top:0 !important}.mt-xl-1{margin-top:.25rem !important}.mt-xl-2{margin-top:.5rem !important}.mt-xl-3{margin-top:1rem !important}.mt-xl-4{margin-top:1.5rem !important}.mt-xl-5{margin-top:3rem !important}.mt-xl-auto{margin-top:auto !important}.me-xl-0{margin-right:0 !important}.me-xl-1{margin-right:.25rem !important}.me-xl-2{margin-right:.5rem !important}.me-xl-3{margin-right:1rem !important}.me-xl-4{margin-right:1.5rem !important}.me-xl-5{margin-right:3rem !important}.me-xl-auto{margin-right:auto !important}.mb-xl-0{margin-bottom:0 !important}.mb-xl-1{margin-bottom:.25rem !important}.mb-xl-2{margin-bottom:.5rem !important}.mb-xl-3{margin-bottom:1rem !important}.mb-xl-4{margin-bottom:1.5rem !important}.mb-xl-5{margin-bottom:3rem !important}.mb-xl-auto{margin-bottom:auto !important}.ms-xl-0{margin-left:0 !important}.ms-xl-1{margin-left:.25rem !important}.ms-xl-2{margin-left:.5rem !important}.ms-xl-3{margin-left:1rem !important}.ms-xl-4{margin-left:1.5rem !important}.ms-xl-5{margin-left:3rem !important}.ms-xl-auto{margin-left:auto !important}.p-xl-0{padding:0 !important}.p-xl-1{padding:.25rem !important}.p-xl-2{padding:.5rem !important}.p-xl-3{padding:1rem !important}.p-xl-4{padding:1.5rem !important}.p-xl-5{padding:3rem !important}.px-xl-0{padding-right:0 !important;padding-left:0 !important}.px-xl-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-xl-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-xl-3{padding-right:1rem !important;padding-left:1rem !important}.px-xl-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-xl-5{padding-right:3rem !important;padding-left:3rem !important}.py-xl-0{padding-top:0 !important;padding-bottom:0 !important}.py-xl-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-xl-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-xl-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-xl-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-xl-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-xl-0{padding-top:0 !important}.pt-xl-1{padding-top:.25rem !important}.pt-xl-2{padding-top:.5rem !important}.pt-xl-3{padding-top:1rem !important}.pt-xl-4{padding-top:1.5rem !important}.pt-xl-5{padding-top:3rem !important}.pe-xl-0{padding-right:0 !important}.pe-xl-1{padding-right:.25rem !important}.pe-xl-2{padding-right:.5rem !important}.pe-xl-3{padding-right:1rem !important}.pe-xl-4{padding-right:1.5rem !important}.pe-xl-5{padding-right:3rem !important}.pb-xl-0{padding-bottom:0 !important}.pb-xl-1{padding-bottom:.25rem !important}.pb-xl-2{padding-bottom:.5rem !important}.pb-xl-3{padding-bottom:1rem !important}.pb-xl-4{padding-bottom:1.5rem !important}.pb-xl-5{padding-bottom:3rem !important}.ps-xl-0{padding-left:0 !important}.ps-xl-1{padding-left:.25rem !important}.ps-xl-2{padding-left:.5rem !important}.ps-xl-3{padding-left:1rem !important}.ps-xl-4{padding-left:1.5rem !important}.ps-xl-5{padding-left:3rem !important}.gap-xl-0{gap:0 !important}.gap-xl-1{gap:.25rem !important}.gap-xl-2{gap:.5rem !important}.gap-xl-3{gap:1rem !important}.gap-xl-4{gap:1.5rem !important}.gap-xl-5{gap:3rem !important}.row-gap-xl-0{row-gap:0 !important}.row-gap-xl-1{row-gap:.25rem !important}.row-gap-xl-2{row-gap:.5rem !important}.row-gap-xl-3{row-gap:1rem !important}.row-gap-xl-4{row-gap:1.5rem !important}.row-gap-xl-5{row-gap:3rem !important}.column-gap-xl-0{column-gap:0 !important}.column-gap-xl-1{column-gap:.25rem !important}.column-gap-xl-2{column-gap:.5rem !important}.column-gap-xl-3{column-gap:1rem !important}.column-gap-xl-4{column-gap:1.5rem !important}.column-gap-xl-5{column-gap:3rem !important}.text-xl-start{text-align:left !important}.text-xl-end{text-align:right !important}.text-xl-center{text-align:center !important}}@media(min-width: 1400px){.float-xxl-start{float:left !important}.float-xxl-end{float:right !important}.float-xxl-none{float:none !important}.object-fit-xxl-contain{object-fit:contain !important}.object-fit-xxl-cover{object-fit:cover !important}.object-fit-xxl-fill{object-fit:fill !important}.object-fit-xxl-scale{object-fit:scale-down !important}.object-fit-xxl-none{object-fit:none !important}.d-xxl-inline{display:inline !important}.d-xxl-inline-block{display:inline-block !important}.d-xxl-block{display:block !important}.d-xxl-grid{display:grid !important}.d-xxl-inline-grid{display:inline-grid !important}.d-xxl-table{display:table !important}.d-xxl-table-row{display:table-row !important}.d-xxl-table-cell{display:table-cell !important}.d-xxl-flex{display:flex !important}.d-xxl-inline-flex{display:inline-flex !important}.d-xxl-none{display:none !important}.flex-xxl-fill{flex:1 1 auto !important}.flex-xxl-row{flex-direction:row !important}.flex-xxl-column{flex-direction:column !important}.flex-xxl-row-reverse{flex-direction:row-reverse !important}.flex-xxl-column-reverse{flex-direction:column-reverse !important}.flex-xxl-grow-0{flex-grow:0 !important}.flex-xxl-grow-1{flex-grow:1 !important}.flex-xxl-shrink-0{flex-shrink:0 !important}.flex-xxl-shrink-1{flex-shrink:1 !important}.flex-xxl-wrap{flex-wrap:wrap !important}.flex-xxl-nowrap{flex-wrap:nowrap !important}.flex-xxl-wrap-reverse{flex-wrap:wrap-reverse !important}.justify-content-xxl-start{justify-content:flex-start !important}.justify-content-xxl-end{justify-content:flex-end !important}.justify-content-xxl-center{justify-content:center !important}.justify-content-xxl-between{justify-content:space-between !important}.justify-content-xxl-around{justify-content:space-around !important}.justify-content-xxl-evenly{justify-content:space-evenly !important}.align-items-xxl-start{align-items:flex-start !important}.align-items-xxl-end{align-items:flex-end !important}.align-items-xxl-center{align-items:center !important}.align-items-xxl-baseline{align-items:baseline !important}.align-items-xxl-stretch{align-items:stretch !important}.align-content-xxl-start{align-content:flex-start !important}.align-content-xxl-end{align-content:flex-end !important}.align-content-xxl-center{align-content:center !important}.align-content-xxl-between{align-content:space-between !important}.align-content-xxl-around{align-content:space-around !important}.align-content-xxl-stretch{align-content:stretch !important}.align-self-xxl-auto{align-self:auto !important}.align-self-xxl-start{align-self:flex-start !important}.align-self-xxl-end{align-self:flex-end !important}.align-self-xxl-center{align-self:center !important}.align-self-xxl-baseline{align-self:baseline !important}.align-self-xxl-stretch{align-self:stretch !important}.order-xxl-first{order:-1 !important}.order-xxl-0{order:0 !important}.order-xxl-1{order:1 !important}.order-xxl-2{order:2 !important}.order-xxl-3{order:3 !important}.order-xxl-4{order:4 !important}.order-xxl-5{order:5 !important}.order-xxl-last{order:6 !important}.m-xxl-0{margin:0 !important}.m-xxl-1{margin:.25rem !important}.m-xxl-2{margin:.5rem !important}.m-xxl-3{margin:1rem !important}.m-xxl-4{margin:1.5rem !important}.m-xxl-5{margin:3rem !important}.m-xxl-auto{margin:auto !important}.mx-xxl-0{margin-right:0 !important;margin-left:0 !important}.mx-xxl-1{margin-right:.25rem !important;margin-left:.25rem !important}.mx-xxl-2{margin-right:.5rem !important;margin-left:.5rem !important}.mx-xxl-3{margin-right:1rem !important;margin-left:1rem !important}.mx-xxl-4{margin-right:1.5rem !important;margin-left:1.5rem !important}.mx-xxl-5{margin-right:3rem !important;margin-left:3rem !important}.mx-xxl-auto{margin-right:auto !important;margin-left:auto !important}.my-xxl-0{margin-top:0 !important;margin-bottom:0 !important}.my-xxl-1{margin-top:.25rem !important;margin-bottom:.25rem !important}.my-xxl-2{margin-top:.5rem !important;margin-bottom:.5rem !important}.my-xxl-3{margin-top:1rem !important;margin-bottom:1rem !important}.my-xxl-4{margin-top:1.5rem !important;margin-bottom:1.5rem !important}.my-xxl-5{margin-top:3rem !important;margin-bottom:3rem !important}.my-xxl-auto{margin-top:auto !important;margin-bottom:auto !important}.mt-xxl-0{margin-top:0 !important}.mt-xxl-1{margin-top:.25rem !important}.mt-xxl-2{margin-top:.5rem !important}.mt-xxl-3{margin-top:1rem !important}.mt-xxl-4{margin-top:1.5rem !important}.mt-xxl-5{margin-top:3rem !important}.mt-xxl-auto{margin-top:auto !important}.me-xxl-0{margin-right:0 !important}.me-xxl-1{margin-right:.25rem !important}.me-xxl-2{margin-right:.5rem !important}.me-xxl-3{margin-right:1rem !important}.me-xxl-4{margin-right:1.5rem !important}.me-xxl-5{margin-right:3rem !important}.me-xxl-auto{margin-right:auto !important}.mb-xxl-0{margin-bottom:0 !important}.mb-xxl-1{margin-bottom:.25rem !important}.mb-xxl-2{margin-bottom:.5rem !important}.mb-xxl-3{margin-bottom:1rem !important}.mb-xxl-4{margin-bottom:1.5rem !important}.mb-xxl-5{margin-bottom:3rem !important}.mb-xxl-auto{margin-bottom:auto !important}.ms-xxl-0{margin-left:0 !important}.ms-xxl-1{margin-left:.25rem !important}.ms-xxl-2{margin-left:.5rem !important}.ms-xxl-3{margin-left:1rem !important}.ms-xxl-4{margin-left:1.5rem !important}.ms-xxl-5{margin-left:3rem !important}.ms-xxl-auto{margin-left:auto !important}.p-xxl-0{padding:0 !important}.p-xxl-1{padding:.25rem !important}.p-xxl-2{padding:.5rem !important}.p-xxl-3{padding:1rem !important}.p-xxl-4{padding:1.5rem !important}.p-xxl-5{padding:3rem !important}.px-xxl-0{padding-right:0 !important;padding-left:0 !important}.px-xxl-1{padding-right:.25rem !important;padding-left:.25rem !important}.px-xxl-2{padding-right:.5rem !important;padding-left:.5rem !important}.px-xxl-3{padding-right:1rem !important;padding-left:1rem !important}.px-xxl-4{padding-right:1.5rem !important;padding-left:1.5rem !important}.px-xxl-5{padding-right:3rem !important;padding-left:3rem !important}.py-xxl-0{padding-top:0 !important;padding-bottom:0 !important}.py-xxl-1{padding-top:.25rem !important;padding-bottom:.25rem !important}.py-xxl-2{padding-top:.5rem !important;padding-bottom:.5rem !important}.py-xxl-3{padding-top:1rem !important;padding-bottom:1rem !important}.py-xxl-4{padding-top:1.5rem !important;padding-bottom:1.5rem !important}.py-xxl-5{padding-top:3rem !important;padding-bottom:3rem !important}.pt-xxl-0{padding-top:0 !important}.pt-xxl-1{padding-top:.25rem !important}.pt-xxl-2{padding-top:.5rem !important}.pt-xxl-3{padding-top:1rem !important}.pt-xxl-4{padding-top:1.5rem !important}.pt-xxl-5{padding-top:3rem !important}.pe-xxl-0{padding-right:0 !important}.pe-xxl-1{padding-right:.25rem !important}.pe-xxl-2{padding-right:.5rem !important}.pe-xxl-3{padding-right:1rem !important}.pe-xxl-4{padding-right:1.5rem !important}.pe-xxl-5{padding-right:3rem !important}.pb-xxl-0{padding-bottom:0 !important}.pb-xxl-1{padding-bottom:.25rem !important}.pb-xxl-2{padding-bottom:.5rem !important}.pb-xxl-3{padding-bottom:1rem !important}.pb-xxl-4{padding-bottom:1.5rem !important}.pb-xxl-5{padding-bottom:3rem !important}.ps-xxl-0{padding-left:0 !important}.ps-xxl-1{padding-left:.25rem !important}.ps-xxl-2{padding-left:.5rem !important}.ps-xxl-3{padding-left:1rem !important}.ps-xxl-4{padding-left:1.5rem !important}.ps-xxl-5{padding-left:3rem !important}.gap-xxl-0{gap:0 !important}.gap-xxl-1{gap:.25rem !important}.gap-xxl-2{gap:.5rem !important}.gap-xxl-3{gap:1rem !important}.gap-xxl-4{gap:1.5rem !important}.gap-xxl-5{gap:3rem !important}.row-gap-xxl-0{row-gap:0 !important}.row-gap-xxl-1{row-gap:.25rem !important}.row-gap-xxl-2{row-gap:.5rem !important}.row-gap-xxl-3{row-gap:1rem !important}.row-gap-xxl-4{row-gap:1.5rem !important}.row-gap-xxl-5{row-gap:3rem !important}.column-gap-xxl-0{column-gap:0 !important}.column-gap-xxl-1{column-gap:.25rem !important}.column-gap-xxl-2{column-gap:.5rem !important}.column-gap-xxl-3{column-gap:1rem !important}.column-gap-xxl-4{column-gap:1.5rem !important}.column-gap-xxl-5{column-gap:3rem !important}.text-xxl-start{text-align:left !important}.text-xxl-end{text-align:right !important}.text-xxl-center{text-align:center !important}}.bg-default{color:#fff}.bg-primary{color:#fff}.bg-secondary{color:#fff}.bg-success{color:#fff}.bg-info{color:#fff}.bg-warning{color:#fff}.bg-danger{color:#fff}.bg-light{color:#fff}.bg-dark{color:#fff}@media(min-width: 1200px){.fs-1{font-size:2rem !important}.fs-2{font-size:1.65rem !important}.fs-3{font-size:1.45rem !important}}@media print{.d-print-inline{display:inline !important}.d-print-inline-block{display:inline-block !important}.d-print-block{display:block !important}.d-print-grid{display:grid !important}.d-print-inline-grid{display:inline-grid !important}.d-print-table{display:table !important}.d-print-table-row{display:table-row !important}.d-print-table-cell{display:table-cell !important}.d-print-flex{display:flex !important}.d-print-inline-flex{display:inline-flex !important}.d-print-none{display:none !important}}:root{--bslib-spacer: 1rem;--bslib-mb-spacer: var(--bslib-spacer, 1rem)}.bslib-mb-spacing{margin-bottom:var(--bslib-mb-spacer)}.bslib-gap-spacing{gap:var(--bslib-mb-spacer)}.bslib-gap-spacing>.bslib-mb-spacing,.bslib-gap-spacing>.form-group,.bslib-gap-spacing>p,.bslib-gap-spacing>pre{margin-bottom:0}.html-fill-container>.html-fill-item.bslib-mb-spacing{margin-bottom:0}.tab-content>.tab-pane.html-fill-container{display:none}.tab-content>.active.html-fill-container{display:flex}.tab-content.html-fill-container{padding:0}.bg-blue{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-blue{--bslib-color-fg: #2780e3;color:var(--bslib-color-fg)}.bg-indigo{--bslib-color-bg: #6610f2;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-indigo{--bslib-color-fg: #6610f2;color:var(--bslib-color-fg)}.bg-purple{--bslib-color-bg: #613d7c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-purple{--bslib-color-fg: #613d7c;color:var(--bslib-color-fg)}.bg-pink{--bslib-color-bg: #e83e8c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-pink{--bslib-color-fg: #e83e8c;color:var(--bslib-color-fg)}.bg-red{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-red{--bslib-color-fg: #ff0039;color:var(--bslib-color-fg)}.bg-orange{--bslib-color-bg: #f0ad4e;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-orange{--bslib-color-fg: #f0ad4e;color:var(--bslib-color-fg)}.bg-yellow{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-yellow{--bslib-color-fg: #ff7518;color:var(--bslib-color-fg)}.bg-green{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-green{--bslib-color-fg: #3fb618;color:var(--bslib-color-fg)}.bg-teal{--bslib-color-bg: #20c997;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-teal{--bslib-color-fg: #20c997;color:var(--bslib-color-fg)}.bg-cyan{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-cyan{--bslib-color-fg: #9954bb;color:var(--bslib-color-fg)}.text-default{--bslib-color-fg: #343a40}.bg-default{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-primary{--bslib-color-fg: #2780e3}.bg-primary{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff}.text-secondary{--bslib-color-fg: #343a40}.bg-secondary{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-success{--bslib-color-fg: #3fb618}.bg-success{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff}.text-info{--bslib-color-fg: #9954bb}.bg-info{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff}.text-warning{--bslib-color-fg: #ff7518}.bg-warning{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff}.text-danger{--bslib-color-fg: #ff0039}.bg-danger{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff}.text-light{--bslib-color-fg: #525252}.bg-light{--bslib-color-bg: #525252;--bslib-color-fg: #fff}.text-dark{--bslib-color-fg: #343a40}.bg-dark{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.bg-gradient-blue-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4053e9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4053e9;color:#fff}.bg-gradient-blue-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3e65ba;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3e65ba;color:#fff}.bg-gradient-blue-pink{--bslib-color-fg: #fff;--bslib-color-bg: #7466c0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #7466c0;color:#fff}.bg-gradient-blue-red{--bslib-color-fg: #fff;--bslib-color-bg: #7d4d9f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #7d4d9f;color:#fff}.bg-gradient-blue-orange{--bslib-color-fg: #fff;--bslib-color-bg: #7792a7;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #7792a7;color:#fff}.bg-gradient-blue-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #7d7c92;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #7d7c92;color:#fff}.bg-gradient-blue-green{--bslib-color-fg: #fff;--bslib-color-bg: #319692;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #319692;color:#fff}.bg-gradient-blue-teal{--bslib-color-fg: #fff;--bslib-color-bg: #249dc5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #249dc5;color:#fff}.bg-gradient-blue-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #556ed3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #556ed3;color:#fff}.bg-gradient-indigo-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4d3dec;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4d3dec;color:#fff}.bg-gradient-indigo-purple{--bslib-color-fg: #fff;--bslib-color-bg: #6422c3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #6422c3;color:#fff}.bg-gradient-indigo-pink{--bslib-color-fg: #fff;--bslib-color-bg: #9a22c9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #9a22c9;color:#fff}.bg-gradient-indigo-red{--bslib-color-fg: #fff;--bslib-color-bg: #a30aa8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a30aa8;color:#fff}.bg-gradient-indigo-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9d4fb0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9d4fb0;color:#fff}.bg-gradient-indigo-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a3389b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a3389b;color:#fff}.bg-gradient-indigo-green{--bslib-color-fg: #fff;--bslib-color-bg: #56529b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #56529b;color:#fff}.bg-gradient-indigo-teal{--bslib-color-fg: #fff;--bslib-color-bg: #4a5ace;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #4a5ace;color:#fff}.bg-gradient-indigo-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #7a2bdc;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #7a2bdc;color:#fff}.bg-gradient-purple-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4a58a5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4a58a5;color:#fff}.bg-gradient-purple-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #632bab;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #632bab;color:#fff}.bg-gradient-purple-pink{--bslib-color-fg: #fff;--bslib-color-bg: #973d82;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #973d82;color:#fff}.bg-gradient-purple-red{--bslib-color-fg: #fff;--bslib-color-bg: #a02561;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a02561;color:#fff}.bg-gradient-purple-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9a6a6a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9a6a6a;color:#fff}.bg-gradient-purple-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a05354;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a05354;color:#fff}.bg-gradient-purple-green{--bslib-color-fg: #fff;--bslib-color-bg: #536d54;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #536d54;color:#fff}.bg-gradient-purple-teal{--bslib-color-fg: #fff;--bslib-color-bg: #477587;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #477587;color:#fff}.bg-gradient-purple-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #774695;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #774695;color:#fff}.bg-gradient-pink-blue{--bslib-color-fg: #fff;--bslib-color-bg: #9b58af;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #9b58af;color:#fff}.bg-gradient-pink-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b42cb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b42cb5;color:#fff}.bg-gradient-pink-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b23e86;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b23e86;color:#fff}.bg-gradient-pink-red{--bslib-color-fg: #fff;--bslib-color-bg: #f1256b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f1256b;color:#fff}.bg-gradient-pink-orange{--bslib-color-fg: #fff;--bslib-color-bg: #eb6a73;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #eb6a73;color:#fff}.bg-gradient-pink-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #f1545e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f1545e;color:#fff}.bg-gradient-pink-green{--bslib-color-fg: #fff;--bslib-color-bg: #a46e5e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a46e5e;color:#fff}.bg-gradient-pink-teal{--bslib-color-fg: #fff;--bslib-color-bg: #987690;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #987690;color:#fff}.bg-gradient-pink-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #c8479f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #c8479f;color:#fff}.bg-gradient-red-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a9337d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a9337d;color:#fff}.bg-gradient-red-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c20683;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c20683;color:#fff}.bg-gradient-red-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c01854;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c01854;color:#fff}.bg-gradient-red-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f6195a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f6195a;color:#fff}.bg-gradient-red-orange{--bslib-color-fg: #fff;--bslib-color-bg: #f94541;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f94541;color:#fff}.bg-gradient-red-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #ff2f2c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #ff2f2c;color:#fff}.bg-gradient-red-green{--bslib-color-fg: #fff;--bslib-color-bg: #b2492c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b2492c;color:#fff}.bg-gradient-red-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6505f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6505f;color:#fff}.bg-gradient-red-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d6226d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d6226d;color:#fff}.bg-gradient-orange-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a09b8a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a09b8a;color:#fff}.bg-gradient-orange-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b96e90;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b96e90;color:#fff}.bg-gradient-orange-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b78060;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b78060;color:#fff}.bg-gradient-orange-pink{--bslib-color-fg: #fff;--bslib-color-bg: #ed8167;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #ed8167;color:#fff}.bg-gradient-orange-red{--bslib-color-fg: #fff;--bslib-color-bg: #f66846;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f66846;color:#fff}.bg-gradient-orange-yellow{--bslib-color-fg: #000;--bslib-color-bg: #f69738;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f69738;color:#000}.bg-gradient-orange-green{--bslib-color-fg: #000;--bslib-color-bg: #a9b138;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a9b138;color:#000}.bg-gradient-orange-teal{--bslib-color-fg: #000;--bslib-color-bg: #9db86b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #9db86b;color:#000}.bg-gradient-orange-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #cd897a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #cd897a;color:#fff}.bg-gradient-yellow-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a97969;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a97969;color:#fff}.bg-gradient-yellow-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c24d6f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c24d6f;color:#fff}.bg-gradient-yellow-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c05f40;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c05f40;color:#fff}.bg-gradient-yellow-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f65f46;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f65f46;color:#fff}.bg-gradient-yellow-red{--bslib-color-fg: #fff;--bslib-color-bg: #ff4625;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #ff4625;color:#fff}.bg-gradient-yellow-orange{--bslib-color-fg: #000;--bslib-color-bg: #f98b2e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f98b2e;color:#000}.bg-gradient-yellow-green{--bslib-color-fg: #fff;--bslib-color-bg: #b28f18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b28f18;color:#fff}.bg-gradient-yellow-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6974b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6974b;color:#fff}.bg-gradient-yellow-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d66859;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d66859;color:#fff}.bg-gradient-green-blue{--bslib-color-fg: #fff;--bslib-color-bg: #35a069;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #35a069;color:#fff}.bg-gradient-green-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4f746f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4f746f;color:#fff}.bg-gradient-green-purple{--bslib-color-fg: #fff;--bslib-color-bg: #4d8640;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #4d8640;color:#fff}.bg-gradient-green-pink{--bslib-color-fg: #fff;--bslib-color-bg: #838646;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #838646;color:#fff}.bg-gradient-green-red{--bslib-color-fg: #fff;--bslib-color-bg: #8c6d25;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #8c6d25;color:#fff}.bg-gradient-green-orange{--bslib-color-fg: #000;--bslib-color-bg: #86b22e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #86b22e;color:#000}.bg-gradient-green-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #8c9c18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #8c9c18;color:#fff}.bg-gradient-green-teal{--bslib-color-fg: #000;--bslib-color-bg: #33be4b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #33be4b;color:#000}.bg-gradient-green-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #638f59;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #638f59;color:#fff}.bg-gradient-teal-blue{--bslib-color-fg: #fff;--bslib-color-bg: #23acb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #23acb5;color:#fff}.bg-gradient-teal-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #3c7fbb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #3c7fbb;color:#fff}.bg-gradient-teal-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3a918c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3a918c;color:#fff}.bg-gradient-teal-pink{--bslib-color-fg: #fff;--bslib-color-bg: #709193;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #709193;color:#fff}.bg-gradient-teal-red{--bslib-color-fg: #fff;--bslib-color-bg: #797971;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #797971;color:#fff}.bg-gradient-teal-orange{--bslib-color-fg: #000;--bslib-color-bg: #73be7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #73be7a;color:#000}.bg-gradient-teal-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #79a764;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #79a764;color:#fff}.bg-gradient-teal-green{--bslib-color-fg: #000;--bslib-color-bg: #2cc164;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #2cc164;color:#000}.bg-gradient-teal-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #509aa5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #509aa5;color:#fff}.bg-gradient-cyan-blue{--bslib-color-fg: #fff;--bslib-color-bg: #6b66cb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #6b66cb;color:#fff}.bg-gradient-cyan-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #8539d1;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #8539d1;color:#fff}.bg-gradient-cyan-purple{--bslib-color-fg: #fff;--bslib-color-bg: #834ba2;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #834ba2;color:#fff}.bg-gradient-cyan-pink{--bslib-color-fg: #fff;--bslib-color-bg: #b94ba8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #b94ba8;color:#fff}.bg-gradient-cyan-red{--bslib-color-fg: #fff;--bslib-color-bg: #c23287;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #c23287;color:#fff}.bg-gradient-cyan-orange{--bslib-color-fg: #fff;--bslib-color-bg: #bc788f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #bc788f;color:#fff}.bg-gradient-cyan-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #c2617a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #c2617a;color:#fff}.bg-gradient-cyan-green{--bslib-color-fg: #fff;--bslib-color-bg: #757b7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #757b7a;color:#fff}.bg-gradient-cyan-teal{--bslib-color-fg: #fff;--bslib-color-bg: #6983ad;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #6983ad;color:#fff}.bg-blue{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-blue{--bslib-color-fg: #2780e3;color:var(--bslib-color-fg)}.bg-indigo{--bslib-color-bg: #6610f2;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-indigo{--bslib-color-fg: #6610f2;color:var(--bslib-color-fg)}.bg-purple{--bslib-color-bg: #613d7c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-purple{--bslib-color-fg: #613d7c;color:var(--bslib-color-fg)}.bg-pink{--bslib-color-bg: #e83e8c;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-pink{--bslib-color-fg: #e83e8c;color:var(--bslib-color-fg)}.bg-red{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-red{--bslib-color-fg: #ff0039;color:var(--bslib-color-fg)}.bg-orange{--bslib-color-bg: #f0ad4e;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-orange{--bslib-color-fg: #f0ad4e;color:var(--bslib-color-fg)}.bg-yellow{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-yellow{--bslib-color-fg: #ff7518;color:var(--bslib-color-fg)}.bg-green{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-green{--bslib-color-fg: #3fb618;color:var(--bslib-color-fg)}.bg-teal{--bslib-color-bg: #20c997;--bslib-color-fg: #000;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-teal{--bslib-color-fg: #20c997;color:var(--bslib-color-fg)}.bg-cyan{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff;background-color:var(--bslib-color-bg);color:var(--bslib-color-fg)}.text-cyan{--bslib-color-fg: #9954bb;color:var(--bslib-color-fg)}.text-default{--bslib-color-fg: #343a40}.bg-default{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-primary{--bslib-color-fg: #2780e3}.bg-primary{--bslib-color-bg: #2780e3;--bslib-color-fg: #fff}.text-secondary{--bslib-color-fg: #343a40}.bg-secondary{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.text-success{--bslib-color-fg: #3fb618}.bg-success{--bslib-color-bg: #3fb618;--bslib-color-fg: #fff}.text-info{--bslib-color-fg: #9954bb}.bg-info{--bslib-color-bg: #9954bb;--bslib-color-fg: #fff}.text-warning{--bslib-color-fg: #ff7518}.bg-warning{--bslib-color-bg: #ff7518;--bslib-color-fg: #fff}.text-danger{--bslib-color-fg: #ff0039}.bg-danger{--bslib-color-bg: #ff0039;--bslib-color-fg: #fff}.text-light{--bslib-color-fg: #525252}.bg-light{--bslib-color-bg: #525252;--bslib-color-fg: #fff}.text-dark{--bslib-color-fg: #343a40}.bg-dark{--bslib-color-bg: #343a40;--bslib-color-fg: #fff}.bg-gradient-blue-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4053e9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4053e9;color:#fff}.bg-gradient-blue-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3e65ba;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3e65ba;color:#fff}.bg-gradient-blue-pink{--bslib-color-fg: #fff;--bslib-color-bg: #7466c0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #7466c0;color:#fff}.bg-gradient-blue-red{--bslib-color-fg: #fff;--bslib-color-bg: #7d4d9f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #7d4d9f;color:#fff}.bg-gradient-blue-orange{--bslib-color-fg: #fff;--bslib-color-bg: #7792a7;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #7792a7;color:#fff}.bg-gradient-blue-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #7d7c92;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #7d7c92;color:#fff}.bg-gradient-blue-green{--bslib-color-fg: #fff;--bslib-color-bg: #319692;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #319692;color:#fff}.bg-gradient-blue-teal{--bslib-color-fg: #fff;--bslib-color-bg: #249dc5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #249dc5;color:#fff}.bg-gradient-blue-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #556ed3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #2780e3 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #556ed3;color:#fff}.bg-gradient-indigo-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4d3dec;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4d3dec;color:#fff}.bg-gradient-indigo-purple{--bslib-color-fg: #fff;--bslib-color-bg: #6422c3;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #6422c3;color:#fff}.bg-gradient-indigo-pink{--bslib-color-fg: #fff;--bslib-color-bg: #9a22c9;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #9a22c9;color:#fff}.bg-gradient-indigo-red{--bslib-color-fg: #fff;--bslib-color-bg: #a30aa8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a30aa8;color:#fff}.bg-gradient-indigo-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9d4fb0;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9d4fb0;color:#fff}.bg-gradient-indigo-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a3389b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a3389b;color:#fff}.bg-gradient-indigo-green{--bslib-color-fg: #fff;--bslib-color-bg: #56529b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #56529b;color:#fff}.bg-gradient-indigo-teal{--bslib-color-fg: #fff;--bslib-color-bg: #4a5ace;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #4a5ace;color:#fff}.bg-gradient-indigo-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #7a2bdc;background:linear-gradient(var(--bg-gradient-deg, 140deg), #6610f2 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #7a2bdc;color:#fff}.bg-gradient-purple-blue{--bslib-color-fg: #fff;--bslib-color-bg: #4a58a5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #4a58a5;color:#fff}.bg-gradient-purple-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #632bab;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #632bab;color:#fff}.bg-gradient-purple-pink{--bslib-color-fg: #fff;--bslib-color-bg: #973d82;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #973d82;color:#fff}.bg-gradient-purple-red{--bslib-color-fg: #fff;--bslib-color-bg: #a02561;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #a02561;color:#fff}.bg-gradient-purple-orange{--bslib-color-fg: #fff;--bslib-color-bg: #9a6a6a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #9a6a6a;color:#fff}.bg-gradient-purple-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #a05354;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #a05354;color:#fff}.bg-gradient-purple-green{--bslib-color-fg: #fff;--bslib-color-bg: #536d54;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #536d54;color:#fff}.bg-gradient-purple-teal{--bslib-color-fg: #fff;--bslib-color-bg: #477587;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #477587;color:#fff}.bg-gradient-purple-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #774695;background:linear-gradient(var(--bg-gradient-deg, 140deg), #613d7c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #774695;color:#fff}.bg-gradient-pink-blue{--bslib-color-fg: #fff;--bslib-color-bg: #9b58af;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #9b58af;color:#fff}.bg-gradient-pink-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b42cb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b42cb5;color:#fff}.bg-gradient-pink-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b23e86;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b23e86;color:#fff}.bg-gradient-pink-red{--bslib-color-fg: #fff;--bslib-color-bg: #f1256b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f1256b;color:#fff}.bg-gradient-pink-orange{--bslib-color-fg: #fff;--bslib-color-bg: #eb6a73;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #eb6a73;color:#fff}.bg-gradient-pink-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #f1545e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f1545e;color:#fff}.bg-gradient-pink-green{--bslib-color-fg: #fff;--bslib-color-bg: #a46e5e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a46e5e;color:#fff}.bg-gradient-pink-teal{--bslib-color-fg: #fff;--bslib-color-bg: #987690;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #987690;color:#fff}.bg-gradient-pink-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #c8479f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #e83e8c var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #c8479f;color:#fff}.bg-gradient-red-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a9337d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a9337d;color:#fff}.bg-gradient-red-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c20683;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c20683;color:#fff}.bg-gradient-red-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c01854;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c01854;color:#fff}.bg-gradient-red-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f6195a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f6195a;color:#fff}.bg-gradient-red-orange{--bslib-color-fg: #fff;--bslib-color-bg: #f94541;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f94541;color:#fff}.bg-gradient-red-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #ff2f2c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #ff2f2c;color:#fff}.bg-gradient-red-green{--bslib-color-fg: #fff;--bslib-color-bg: #b2492c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b2492c;color:#fff}.bg-gradient-red-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6505f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6505f;color:#fff}.bg-gradient-red-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d6226d;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff0039 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d6226d;color:#fff}.bg-gradient-orange-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a09b8a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a09b8a;color:#fff}.bg-gradient-orange-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #b96e90;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #b96e90;color:#fff}.bg-gradient-orange-purple{--bslib-color-fg: #fff;--bslib-color-bg: #b78060;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #b78060;color:#fff}.bg-gradient-orange-pink{--bslib-color-fg: #fff;--bslib-color-bg: #ed8167;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #ed8167;color:#fff}.bg-gradient-orange-red{--bslib-color-fg: #fff;--bslib-color-bg: #f66846;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #f66846;color:#fff}.bg-gradient-orange-yellow{--bslib-color-fg: #000;--bslib-color-bg: #f69738;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #f69738;color:#000}.bg-gradient-orange-green{--bslib-color-fg: #000;--bslib-color-bg: #a9b138;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #a9b138;color:#000}.bg-gradient-orange-teal{--bslib-color-fg: #000;--bslib-color-bg: #9db86b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #9db86b;color:#000}.bg-gradient-orange-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #cd897a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #f0ad4e var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #cd897a;color:#fff}.bg-gradient-yellow-blue{--bslib-color-fg: #fff;--bslib-color-bg: #a97969;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #a97969;color:#fff}.bg-gradient-yellow-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #c24d6f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #c24d6f;color:#fff}.bg-gradient-yellow-purple{--bslib-color-fg: #fff;--bslib-color-bg: #c05f40;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #c05f40;color:#fff}.bg-gradient-yellow-pink{--bslib-color-fg: #fff;--bslib-color-bg: #f65f46;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #f65f46;color:#fff}.bg-gradient-yellow-red{--bslib-color-fg: #fff;--bslib-color-bg: #ff4625;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #ff4625;color:#fff}.bg-gradient-yellow-orange{--bslib-color-fg: #000;--bslib-color-bg: #f98b2e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #f98b2e;color:#000}.bg-gradient-yellow-green{--bslib-color-fg: #fff;--bslib-color-bg: #b28f18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #b28f18;color:#fff}.bg-gradient-yellow-teal{--bslib-color-fg: #fff;--bslib-color-bg: #a6974b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #a6974b;color:#fff}.bg-gradient-yellow-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #d66859;background:linear-gradient(var(--bg-gradient-deg, 140deg), #ff7518 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #d66859;color:#fff}.bg-gradient-green-blue{--bslib-color-fg: #fff;--bslib-color-bg: #35a069;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #35a069;color:#fff}.bg-gradient-green-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #4f746f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #4f746f;color:#fff}.bg-gradient-green-purple{--bslib-color-fg: #fff;--bslib-color-bg: #4d8640;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #4d8640;color:#fff}.bg-gradient-green-pink{--bslib-color-fg: #fff;--bslib-color-bg: #838646;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #838646;color:#fff}.bg-gradient-green-red{--bslib-color-fg: #fff;--bslib-color-bg: #8c6d25;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #8c6d25;color:#fff}.bg-gradient-green-orange{--bslib-color-fg: #000;--bslib-color-bg: #86b22e;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #86b22e;color:#000}.bg-gradient-green-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #8c9c18;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #8c9c18;color:#fff}.bg-gradient-green-teal{--bslib-color-fg: #000;--bslib-color-bg: #33be4b;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #33be4b;color:#000}.bg-gradient-green-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #638f59;background:linear-gradient(var(--bg-gradient-deg, 140deg), #3fb618 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #638f59;color:#fff}.bg-gradient-teal-blue{--bslib-color-fg: #fff;--bslib-color-bg: #23acb5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #23acb5;color:#fff}.bg-gradient-teal-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #3c7fbb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #3c7fbb;color:#fff}.bg-gradient-teal-purple{--bslib-color-fg: #fff;--bslib-color-bg: #3a918c;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #3a918c;color:#fff}.bg-gradient-teal-pink{--bslib-color-fg: #fff;--bslib-color-bg: #709193;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #709193;color:#fff}.bg-gradient-teal-red{--bslib-color-fg: #fff;--bslib-color-bg: #797971;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #797971;color:#fff}.bg-gradient-teal-orange{--bslib-color-fg: #000;--bslib-color-bg: #73be7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #73be7a;color:#000}.bg-gradient-teal-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #79a764;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #79a764;color:#fff}.bg-gradient-teal-green{--bslib-color-fg: #000;--bslib-color-bg: #2cc164;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #2cc164;color:#000}.bg-gradient-teal-cyan{--bslib-color-fg: #fff;--bslib-color-bg: #509aa5;background:linear-gradient(var(--bg-gradient-deg, 140deg), #20c997 var(--bg-gradient-start, 36%), #9954bb var(--bg-gradient-end, 180%)) #509aa5;color:#fff}.bg-gradient-cyan-blue{--bslib-color-fg: #fff;--bslib-color-bg: #6b66cb;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #2780e3 var(--bg-gradient-end, 180%)) #6b66cb;color:#fff}.bg-gradient-cyan-indigo{--bslib-color-fg: #fff;--bslib-color-bg: #8539d1;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #6610f2 var(--bg-gradient-end, 180%)) #8539d1;color:#fff}.bg-gradient-cyan-purple{--bslib-color-fg: #fff;--bslib-color-bg: #834ba2;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #613d7c var(--bg-gradient-end, 180%)) #834ba2;color:#fff}.bg-gradient-cyan-pink{--bslib-color-fg: #fff;--bslib-color-bg: #b94ba8;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #e83e8c var(--bg-gradient-end, 180%)) #b94ba8;color:#fff}.bg-gradient-cyan-red{--bslib-color-fg: #fff;--bslib-color-bg: #c23287;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff0039 var(--bg-gradient-end, 180%)) #c23287;color:#fff}.bg-gradient-cyan-orange{--bslib-color-fg: #fff;--bslib-color-bg: #bc788f;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #f0ad4e var(--bg-gradient-end, 180%)) #bc788f;color:#fff}.bg-gradient-cyan-yellow{--bslib-color-fg: #fff;--bslib-color-bg: #c2617a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #ff7518 var(--bg-gradient-end, 180%)) #c2617a;color:#fff}.bg-gradient-cyan-green{--bslib-color-fg: #fff;--bslib-color-bg: #757b7a;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #3fb618 var(--bg-gradient-end, 180%)) #757b7a;color:#fff}.bg-gradient-cyan-teal{--bslib-color-fg: #fff;--bslib-color-bg: #6983ad;background:linear-gradient(var(--bg-gradient-deg, 140deg), #9954bb var(--bg-gradient-start, 36%), #20c997 var(--bg-gradient-end, 180%)) #6983ad;color:#fff}.tab-content>.tab-pane.html-fill-container{display:none}.tab-content>.active.html-fill-container{display:flex}.tab-content.html-fill-container{padding:0}:root{--bslib-spacer: 1rem;--bslib-mb-spacer: var(--bslib-spacer, 1rem)}.bslib-mb-spacing{margin-bottom:var(--bslib-mb-spacer)}.bslib-gap-spacing{gap:var(--bslib-mb-spacer)}.bslib-gap-spacing>.bslib-mb-spacing,.bslib-gap-spacing>.form-group,.bslib-gap-spacing>p,.bslib-gap-spacing>pre{margin-bottom:0}.html-fill-container>.html-fill-item.bslib-mb-spacing{margin-bottom:0}.bslib-grid{display:grid !important;gap:var(--bslib-spacer, 1rem);height:var(--bslib-grid-height)}.bslib-grid.grid{grid-template-columns:repeat(var(--bs-columns, 12), minmax(0, 1fr));grid-template-rows:unset;grid-auto-rows:var(--bslib-grid--row-heights);--bslib-grid--row-heights--xs: unset;--bslib-grid--row-heights--sm: unset;--bslib-grid--row-heights--md: unset;--bslib-grid--row-heights--lg: unset;--bslib-grid--row-heights--xl: unset;--bslib-grid--row-heights--xxl: unset}.bslib-grid.grid.bslib-grid--row-heights--xs{--bslib-grid--row-heights: var(--bslib-grid--row-heights--xs)}@media(min-width: 576px){.bslib-grid.grid.bslib-grid--row-heights--sm{--bslib-grid--row-heights: var(--bslib-grid--row-heights--sm)}}@media(min-width: 768px){.bslib-grid.grid.bslib-grid--row-heights--md{--bslib-grid--row-heights: var(--bslib-grid--row-heights--md)}}@media(min-width: 992px){.bslib-grid.grid.bslib-grid--row-heights--lg{--bslib-grid--row-heights: var(--bslib-grid--row-heights--lg)}}@media(min-width: 1200px){.bslib-grid.grid.bslib-grid--row-heights--xl{--bslib-grid--row-heights: var(--bslib-grid--row-heights--xl)}}@media(min-width: 1400px){.bslib-grid.grid.bslib-grid--row-heights--xxl{--bslib-grid--row-heights: var(--bslib-grid--row-heights--xxl)}}.bslib-grid>*>.shiny-input-container{width:100%}.bslib-grid-item{grid-column:auto/span 1}@media(max-width: 767.98px){.bslib-grid-item{grid-column:1/-1}}@media(max-width: 575.98px){.bslib-grid{grid-template-columns:1fr !important;height:var(--bslib-grid-height-mobile)}.bslib-grid.grid{height:unset !important;grid-auto-rows:var(--bslib-grid--row-heights--xs, auto)}}.bslib-card{overflow:auto}.bslib-card .card-body+.card-body{padding-top:0}.bslib-card .card-body{overflow:auto}.bslib-card .card-body p{margin-top:0}.bslib-card .card-body p:last-child{margin-bottom:0}.bslib-card .card-body{max-height:var(--bslib-card-body-max-height, none)}.bslib-card[data-full-screen=true]>.card-body{max-height:var(--bslib-card-body-max-height-full-screen, none)}.bslib-card .card-header .form-group{margin-bottom:0}.bslib-card .card-header .selectize-control{margin-bottom:0}.bslib-card .card-header .selectize-control .item{margin-right:1.15rem}.bslib-card .card-footer{margin-top:auto}.bslib-card .bslib-navs-card-title{display:flex;flex-wrap:wrap;justify-content:space-between;align-items:center}.bslib-card .bslib-navs-card-title .nav{margin-left:auto}.bslib-card .bslib-sidebar-layout:not([data-bslib-sidebar-border=true]){border:none}.bslib-card .bslib-sidebar-layout:not([data-bslib-sidebar-border-radius=true]){border-top-left-radius:0;border-top-right-radius:0}[data-full-screen=true]{position:fixed;inset:3.5rem 1rem 1rem;height:auto !important;max-height:none !important;width:auto !important;z-index:1070}.bslib-full-screen-enter{display:none;position:absolute;bottom:var(--bslib-full-screen-enter-bottom, 0.2rem);right:var(--bslib-full-screen-enter-right, 0);top:var(--bslib-full-screen-enter-top);left:var(--bslib-full-screen-enter-left);color:var(--bslib-color-fg, var(--bs-card-color));background-color:var(--bslib-color-bg, var(--bs-card-bg, var(--bs-body-bg)));border:var(--bs-card-border-width) solid var(--bslib-color-fg, var(--bs-card-border-color));box-shadow:0 2px 4px rgba(0,0,0,.15);margin:.2rem .4rem;padding:.55rem !important;font-size:.8rem;cursor:pointer;opacity:.7;z-index:1070}.bslib-full-screen-enter:hover{opacity:1}.card[data-full-screen=false]:hover>*>.bslib-full-screen-enter{display:block}.bslib-has-full-screen .card:hover>*>.bslib-full-screen-enter{display:none}@media(max-width: 575.98px){.bslib-full-screen-enter{display:none !important}}.bslib-full-screen-exit{position:relative;top:1.35rem;font-size:.9rem;cursor:pointer;text-decoration:none;display:flex;float:right;margin-right:2.15rem;align-items:center;color:rgba(var(--bs-body-bg-rgb), 0.8)}.bslib-full-screen-exit:hover{color:rgba(var(--bs-body-bg-rgb), 1)}.bslib-full-screen-exit svg{margin-left:.5rem;font-size:1.5rem}#bslib-full-screen-overlay{position:fixed;inset:0;background-color:rgba(var(--bs-body-color-rgb), 0.6);backdrop-filter:blur(2px);-webkit-backdrop-filter:blur(2px);z-index:1069;animation:bslib-full-screen-overlay-enter 400ms cubic-bezier(0.6, 0.02, 0.65, 1) forwards}@keyframes bslib-full-screen-overlay-enter{0%{opacity:0}100%{opacity:1}}:root{--bslib-value-box-shadow: none;--bslib-value-box-border-width-auto-yes: var(--bslib-value-box-border-width-baseline);--bslib-value-box-border-width-auto-no: 0;--bslib-value-box-border-width-baseline: 1px}.bslib-value-box{border-width:var(--bslib-value-box-border-width-auto-no, var(--bslib-value-box-border-width-baseline));container-name:bslib-value-box;container-type:inline-size}.bslib-value-box.card{box-shadow:var(--bslib-value-box-shadow)}.bslib-value-box.border-auto{border-width:var(--bslib-value-box-border-width-auto-yes, var(--bslib-value-box-border-width-baseline))}.bslib-value-box.default{--bslib-value-box-bg-default: var(--bs-card-bg, #181818);--bslib-value-box-border-color-default: var(--bs-card-border-color, rgba(0, 0, 0, 0.175));color:var(--bslib-value-box-color);background-color:var(--bslib-value-box-bg, var(--bslib-value-box-bg-default));border-color:var(--bslib-value-box-border-color, var(--bslib-value-box-border-color-default))}.bslib-value-box .value-box-grid{display:grid;grid-template-areas:"left right";align-items:center;overflow:hidden}.bslib-value-box .value-box-showcase{height:100%;max-height:var(---bslib-value-box-showcase-max-h, 100%)}.bslib-value-box .value-box-showcase,.bslib-value-box .value-box-showcase>.html-fill-item{width:100%}.bslib-value-box[data-full-screen=true] .value-box-showcase{max-height:var(---bslib-value-box-showcase-max-h-fs, 100%)}@media screen and (min-width: 575.98px){@container bslib-value-box (max-width: 300px){.bslib-value-box:not(.showcase-bottom) .value-box-grid{grid-template-columns:1fr !important;grid-template-rows:auto auto;grid-template-areas:"top" "bottom"}.bslib-value-box:not(.showcase-bottom) .value-box-grid .value-box-showcase{grid-area:top !important}.bslib-value-box:not(.showcase-bottom) .value-box-grid .value-box-area{grid-area:bottom !important;justify-content:end}}}.bslib-value-box .value-box-area{justify-content:center;padding:1.5rem 1rem;font-size:.9rem;font-weight:500}.bslib-value-box .value-box-area *{margin-bottom:0;margin-top:0}.bslib-value-box .value-box-title{font-size:1rem;margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2}.bslib-value-box .value-box-title:empty::after{content:" "}.bslib-value-box .value-box-value{font-size:calc(1.29rem + 0.48vw);margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2}@media(min-width: 1200px){.bslib-value-box .value-box-value{font-size:1.65rem}}.bslib-value-box .value-box-value:empty::after{content:" "}.bslib-value-box .value-box-showcase{align-items:center;justify-content:center;margin-top:auto;margin-bottom:auto;padding:1rem}.bslib-value-box .value-box-showcase .bi,.bslib-value-box .value-box-showcase .fa,.bslib-value-box .value-box-showcase .fab,.bslib-value-box .value-box-showcase .fas,.bslib-value-box .value-box-showcase .far{opacity:.85;min-width:50px;max-width:125%}.bslib-value-box .value-box-showcase .bi,.bslib-value-box .value-box-showcase .fa,.bslib-value-box .value-box-showcase .fab,.bslib-value-box .value-box-showcase .fas,.bslib-value-box .value-box-showcase .far{font-size:4rem}.bslib-value-box.showcase-top-right .value-box-grid{grid-template-columns:1fr var(---bslib-value-box-showcase-w, 50%)}.bslib-value-box.showcase-top-right .value-box-grid .value-box-showcase{grid-area:right;margin-left:auto;align-self:start;align-items:end;padding-left:0;padding-bottom:0}.bslib-value-box.showcase-top-right .value-box-grid .value-box-area{grid-area:left;align-self:end}.bslib-value-box.showcase-top-right[data-full-screen=true] .value-box-grid{grid-template-columns:auto var(---bslib-value-box-showcase-w-fs, 1fr)}.bslib-value-box.showcase-top-right[data-full-screen=true] .value-box-grid>div{align-self:center}.bslib-value-box.showcase-top-right:not([data-full-screen=true]) .value-box-showcase{margin-top:0}@container bslib-value-box (max-width: 300px){.bslib-value-box.showcase-top-right:not([data-full-screen=true]) .value-box-grid .value-box-showcase{padding-left:1rem}}.bslib-value-box.showcase-left-center .value-box-grid{grid-template-columns:var(---bslib-value-box-showcase-w, 30%) auto}.bslib-value-box.showcase-left-center[data-full-screen=true] .value-box-grid{grid-template-columns:var(---bslib-value-box-showcase-w-fs, 1fr) auto}.bslib-value-box.showcase-left-center:not([data-fill-screen=true]) .value-box-grid .value-box-showcase{grid-area:left}.bslib-value-box.showcase-left-center:not([data-fill-screen=true]) .value-box-grid .value-box-area{grid-area:right}.bslib-value-box.showcase-bottom .value-box-grid{grid-template-columns:1fr;grid-template-rows:1fr var(---bslib-value-box-showcase-h, auto);grid-template-areas:"top" "bottom";overflow:hidden}.bslib-value-box.showcase-bottom .value-box-grid .value-box-showcase{grid-area:bottom;padding:0;margin:0}.bslib-value-box.showcase-bottom .value-box-grid .value-box-area{grid-area:top}.bslib-value-box.showcase-bottom[data-full-screen=true] .value-box-grid{grid-template-rows:1fr var(---bslib-value-box-showcase-h-fs, 2fr)}.bslib-value-box.showcase-bottom[data-full-screen=true] .value-box-grid .value-box-showcase{padding:1rem}[data-bs-theme=dark] .bslib-value-box{--bslib-value-box-shadow: 0 0.5rem 1rem rgb(0 0 0 / 50%)}:root{--bslib-page-sidebar-title-bg: #303030;--bslib-page-sidebar-title-color: #fff}.bslib-page-title{background-color:var(--bslib-page-sidebar-title-bg);color:var(--bslib-page-sidebar-title-color);font-size:1.25rem;font-weight:300;padding:var(--bslib-spacer, 1rem);padding-left:1.5rem;margin-bottom:0;border-bottom:1px solid #dee2e6}html{height:100%}.bslib-page-fill{width:100%;height:100%;margin:0;padding:var(--bslib-spacer, 1rem);gap:var(--bslib-spacer, 1rem)}@media(max-width: 575.98px){.bslib-page-fill{height:var(--bslib-page-fill-mobile-height, auto)}}@media(min-width: 576px){.nav:not(.nav-hidden){display:flex !important;display:-webkit-flex !important}.nav:not(.nav-hidden):not(.nav-stacked):not(.flex-column){float:none !important}.nav:not(.nav-hidden):not(.nav-stacked):not(.flex-column)>.bslib-nav-spacer{margin-left:auto !important}.nav:not(.nav-hidden):not(.nav-stacked):not(.flex-column)>.form-inline{margin-top:auto;margin-bottom:auto}.nav:not(.nav-hidden).nav-stacked{flex-direction:column;-webkit-flex-direction:column;height:100%}.nav:not(.nav-hidden).nav-stacked>.bslib-nav-spacer{margin-top:auto !important}}.accordion .accordion-header{font-size:calc(1.29rem + 0.48vw);margin-top:0;margin-bottom:.5rem;font-weight:400;line-height:1.2;color:var(--bs-heading-color);margin-bottom:0}@media(min-width: 1200px){.accordion .accordion-header{font-size:1.65rem}}.accordion .accordion-icon:not(:empty){margin-right:.75rem;display:flex}.accordion .accordion-button:not(.collapsed){box-shadow:none}.accordion .accordion-button:not(.collapsed):focus{box-shadow:var(--bs-accordion-btn-focus-box-shadow)}.bslib-sidebar-layout{--bslib-sidebar-transition-duration: 500ms;--bslib-sidebar-transition-easing-x: cubic-bezier(0.8, 0.78, 0.22, 1.07);--bslib-sidebar-border: var(--bs-card-border-width, 1px) solid var(--bs-card-border-color, rgba(0, 0, 0, 0.175));--bslib-sidebar-border-radius: var(--bs-border-radius);--bslib-sidebar-vert-border: var(--bs-card-border-width, 1px) solid var(--bs-card-border-color, rgba(0, 0, 0, 0.175));--bslib-sidebar-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.05);--bslib-sidebar-fg: var(--bs-emphasis-color, black);--bslib-sidebar-main-fg: var(--bs-card-color, var(--bs-body-color));--bslib-sidebar-main-bg: var(--bs-card-bg, var(--bs-body-bg));--bslib-sidebar-toggle-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.1);--bslib-sidebar-padding: calc(var(--bslib-spacer) * 1.5);--bslib-sidebar-icon-size: var(--bslib-spacer, 1rem);--bslib-sidebar-icon-button-size: calc(var(--bslib-sidebar-icon-size, 1rem) * 2);--bslib-sidebar-padding-icon: calc(var(--bslib-sidebar-icon-button-size, 2rem) * 1.5);--bslib-collapse-toggle-border-radius: var(--bs-border-radius, 0.25rem);--bslib-collapse-toggle-transform: 0deg;--bslib-sidebar-toggle-transition-easing: cubic-bezier(1, 0, 0, 1);--bslib-collapse-toggle-right-transform: 180deg;--bslib-sidebar-column-main: minmax(0, 1fr);display:grid !important;grid-template-columns:min(100% - var(--bslib-sidebar-icon-size),var(--bslib-sidebar-width, 250px)) var(--bslib-sidebar-column-main);position:relative;transition:grid-template-columns ease-in-out var(--bslib-sidebar-transition-duration);border:var(--bslib-sidebar-border);border-radius:var(--bslib-sidebar-border-radius)}@media(prefers-reduced-motion: reduce){.bslib-sidebar-layout{transition:none}}.bslib-sidebar-layout[data-bslib-sidebar-border=false]{border:none}.bslib-sidebar-layout[data-bslib-sidebar-border-radius=false]{border-radius:initial}.bslib-sidebar-layout>.main,.bslib-sidebar-layout>.sidebar{grid-row:1/2;border-radius:inherit;overflow:auto}.bslib-sidebar-layout>.main{grid-column:2/3;border-top-left-radius:0;border-bottom-left-radius:0;padding:var(--bslib-sidebar-padding);transition:padding var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration);color:var(--bslib-sidebar-main-fg);background-color:var(--bslib-sidebar-main-bg)}.bslib-sidebar-layout>.sidebar{grid-column:1/2;width:100%;height:100%;border-right:var(--bslib-sidebar-vert-border);border-top-right-radius:0;border-bottom-right-radius:0;color:var(--bslib-sidebar-fg);background-color:var(--bslib-sidebar-bg);backdrop-filter:blur(5px)}.bslib-sidebar-layout>.sidebar>.sidebar-content{display:flex;flex-direction:column;gap:var(--bslib-spacer, 1rem);padding:var(--bslib-sidebar-padding);padding-top:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout>.sidebar>.sidebar-content>:last-child:not(.sidebar-title){margin-bottom:0}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion{margin-left:calc(-1*var(--bslib-sidebar-padding));margin-right:calc(-1*var(--bslib-sidebar-padding))}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:last-child{margin-bottom:calc(-1*var(--bslib-sidebar-padding))}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:not(:last-child){margin-bottom:1rem}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion .accordion-body{display:flex;flex-direction:column}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:not(:first-child) .accordion-item:first-child{border-top:var(--bs-accordion-border-width) solid var(--bs-accordion-border-color)}.bslib-sidebar-layout>.sidebar>.sidebar-content>.accordion:not(:last-child) .accordion-item:last-child{border-bottom:var(--bs-accordion-border-width) solid var(--bs-accordion-border-color)}.bslib-sidebar-layout>.sidebar>.sidebar-content.has-accordion>.sidebar-title{border-bottom:none;padding-bottom:0}.bslib-sidebar-layout>.sidebar .shiny-input-container{width:100%}.bslib-sidebar-layout[data-bslib-sidebar-open=always]>.sidebar>.sidebar-content{padding-top:var(--bslib-sidebar-padding)}.bslib-sidebar-layout>.collapse-toggle{grid-row:1/2;grid-column:1/2;display:inline-flex;align-items:center;position:absolute;right:calc(var(--bslib-sidebar-icon-size));top:calc(var(--bslib-sidebar-icon-size, 1rem)/2);border:none;border-radius:var(--bslib-collapse-toggle-border-radius);height:var(--bslib-sidebar-icon-button-size, 2rem);width:var(--bslib-sidebar-icon-button-size, 2rem);display:flex;align-items:center;justify-content:center;padding:0;color:var(--bslib-sidebar-fg);background-color:unset;transition:color var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration),top var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration),right var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration),left var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration)}.bslib-sidebar-layout>.collapse-toggle:hover{background-color:var(--bslib-sidebar-toggle-bg)}.bslib-sidebar-layout>.collapse-toggle>.collapse-icon{opacity:.8;width:var(--bslib-sidebar-icon-size);height:var(--bslib-sidebar-icon-size);transform:rotateY(var(--bslib-collapse-toggle-transform));transition:transform var(--bslib-sidebar-toggle-transition-easing) var(--bslib-sidebar-transition-duration)}.bslib-sidebar-layout>.collapse-toggle:hover>.collapse-icon{opacity:1}.bslib-sidebar-layout .sidebar-title{font-size:1.25rem;line-height:1.25;margin-top:0;margin-bottom:1rem;padding-bottom:1rem;border-bottom:var(--bslib-sidebar-border)}.bslib-sidebar-layout.sidebar-right{grid-template-columns:var(--bslib-sidebar-column-main) min(100% - var(--bslib-sidebar-icon-size),var(--bslib-sidebar-width, 250px))}.bslib-sidebar-layout.sidebar-right>.main{grid-column:1/2;border-top-right-radius:0;border-bottom-right-radius:0;border-top-left-radius:inherit;border-bottom-left-radius:inherit}.bslib-sidebar-layout.sidebar-right>.sidebar{grid-column:2/3;border-right:none;border-left:var(--bslib-sidebar-vert-border);border-top-left-radius:0;border-bottom-left-radius:0}.bslib-sidebar-layout.sidebar-right>.collapse-toggle{grid-column:2/3;left:var(--bslib-sidebar-icon-size);right:unset;border:var(--bslib-collapse-toggle-border)}.bslib-sidebar-layout.sidebar-right>.collapse-toggle>.collapse-icon{transform:rotateY(var(--bslib-collapse-toggle-right-transform))}.bslib-sidebar-layout.sidebar-collapsed{--bslib-collapse-toggle-transform: 180deg;--bslib-collapse-toggle-right-transform: 0deg;--bslib-sidebar-vert-border: none;grid-template-columns:0 minmax(0, 1fr)}.bslib-sidebar-layout.sidebar-collapsed.sidebar-right{grid-template-columns:minmax(0, 1fr) 0}.bslib-sidebar-layout.sidebar-collapsed:not(.transitioning)>.sidebar>*{display:none}.bslib-sidebar-layout.sidebar-collapsed>.main{border-radius:inherit}.bslib-sidebar-layout.sidebar-collapsed:not(.sidebar-right)>.main{padding-left:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout.sidebar-collapsed.sidebar-right>.main{padding-right:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout.sidebar-collapsed>.collapse-toggle{color:var(--bslib-sidebar-main-fg);top:calc(var(--bslib-sidebar-overlap-counter, 0)*(var(--bslib-sidebar-icon-size) + var(--bslib-sidebar-padding)) + var(--bslib-sidebar-icon-size, 1rem)/2);right:calc(-2.5*var(--bslib-sidebar-icon-size) - var(--bs-card-border-width, 1px))}.bslib-sidebar-layout.sidebar-collapsed.sidebar-right>.collapse-toggle{left:calc(-2.5*var(--bslib-sidebar-icon-size) - var(--bs-card-border-width, 1px));right:unset}@media(min-width: 576px){.bslib-sidebar-layout.transitioning>.sidebar>.sidebar-content{display:none}}@media(max-width: 575.98px){.bslib-sidebar-layout[data-bslib-sidebar-open=desktop]{--bslib-sidebar-js-init-collapsed: true}.bslib-sidebar-layout>.sidebar,.bslib-sidebar-layout.sidebar-right>.sidebar{border:none}.bslib-sidebar-layout>.main,.bslib-sidebar-layout.sidebar-right>.main{grid-column:1/3}.bslib-sidebar-layout[data-bslib-sidebar-open=always]{display:block !important}.bslib-sidebar-layout[data-bslib-sidebar-open=always]>.sidebar{max-height:var(--bslib-sidebar-max-height-mobile);overflow-y:auto;border-top:var(--bslib-sidebar-vert-border)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]){grid-template-columns:100% 0}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]):not(.sidebar-collapsed)>.sidebar{z-index:1}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]):not(.sidebar-collapsed)>.collapse-toggle{z-index:1}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-right{grid-template-columns:0 100%}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-collapsed{grid-template-columns:0 100%}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-collapsed.sidebar-right{grid-template-columns:100% 0}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]):not(.sidebar-right)>.main{padding-left:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-right>.main{padding-right:var(--bslib-sidebar-padding-icon)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always])>.main{opacity:0;transition:opacity var(--bslib-sidebar-transition-easing-x) var(--bslib-sidebar-transition-duration)}.bslib-sidebar-layout:not([data-bslib-sidebar-open=always]).sidebar-collapsed>.main{opacity:1}}.navbar+.container-fluid:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-sm:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-md:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-lg:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-xl:has(>.tab-content>.tab-pane.active.html-fill-container),.navbar+.container-xxl:has(>.tab-content>.tab-pane.active.html-fill-container){padding-left:0;padding-right:0}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container,.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container{padding:var(--bslib-spacer, 1rem);gap:var(--bslib-spacer, 1rem)}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child),.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container:has(>.bslib-sidebar-layout:only-child){padding:0}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]),.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border=true]){border-left:none;border-right:none;border-bottom:none}.navbar+.container-fluid>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-sm>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-md>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-lg>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-xl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]),.navbar+.container-xxl>.tab-content>.tab-pane.active.html-fill-container>.bslib-sidebar-layout:only-child:not([data-bslib-sidebar-border-radius=true]){border-radius:0}.navbar+div>.bslib-sidebar-layout{border-top:var(--bslib-sidebar-border)}.html-fill-container{display:flex;flex-direction:column;min-height:0;min-width:0}.html-fill-container>.html-fill-item{flex:1 1 auto;min-height:0;min-width:0}.html-fill-container>:not(.html-fill-item){flex:0 0 auto}.quarto-container{min-height:calc(100vh - 132px)}body.hypothesis-enabled #quarto-header{margin-right:16px}footer.footer .nav-footer,#quarto-header>nav{padding-left:1em;padding-right:1em}footer.footer div.nav-footer p:first-child{margin-top:0}footer.footer div.nav-footer p:last-child{margin-bottom:0}#quarto-content>*{padding-top:14px}#quarto-content>#quarto-sidebar-glass{padding-top:0px}@media(max-width: 991.98px){#quarto-content>*{padding-top:0}#quarto-content .subtitle{padding-top:14px}#quarto-content section:first-of-type h2:first-of-type,#quarto-content section:first-of-type .h2:first-of-type{margin-top:1rem}}.headroom-target,header.headroom{will-change:transform;transition:position 200ms linear;transition:all 200ms linear}header.headroom--pinned{transform:translateY(0%)}header.headroom--unpinned{transform:translateY(-100%)}.navbar-container{width:100%}.navbar-brand{overflow:hidden;text-overflow:ellipsis}.navbar-brand-container{max-width:calc(100% - 115px);min-width:0;display:flex;align-items:center}@media(min-width: 992px){.navbar-brand-container{margin-right:1em}}.navbar-brand.navbar-brand-logo{margin-right:4px;display:inline-flex}.navbar-toggler{flex-basis:content;flex-shrink:0}.navbar .navbar-brand-container{order:2}.navbar .navbar-toggler{order:1}.navbar .navbar-container>.navbar-nav{order:20}.navbar .navbar-container>.navbar-brand-container{margin-left:0 !important;margin-right:0 !important}.navbar .navbar-collapse{order:20}.navbar #quarto-search{order:4;margin-left:auto}.navbar .navbar-toggler{margin-right:.5em}.navbar-collapse .quarto-navbar-tools{margin-left:.5em}.navbar-logo{max-height:24px;width:auto;padding-right:4px}nav .nav-item:not(.compact){padding-top:1px}nav .nav-link i,nav .dropdown-item i{padding-right:1px}.navbar-expand-lg .navbar-nav .nav-link{padding-left:.6rem;padding-right:.6rem}nav .nav-item.compact .nav-link{padding-left:.5rem;padding-right:.5rem;font-size:1.1rem}.navbar .quarto-navbar-tools{order:3}.navbar .quarto-navbar-tools div.dropdown{display:inline-block}.navbar .quarto-navbar-tools .quarto-navigation-tool{color:#bdbdbd}.navbar .quarto-navbar-tools .quarto-navigation-tool:hover{color:#aed1e4}.navbar-nav .dropdown-menu{min-width:220px;font-size:.9rem}.navbar .navbar-nav .nav-link.dropdown-toggle::after{opacity:.75;vertical-align:.175em}.navbar ul.dropdown-menu{padding-top:0;padding-bottom:0}.navbar .dropdown-header{text-transform:uppercase;font-size:.8rem;padding:0 .5rem}.navbar .dropdown-item{padding:.4rem .5rem}.navbar .dropdown-item>i.bi{margin-left:.1rem;margin-right:.25em}.sidebar #quarto-search{margin-top:-1px}.sidebar #quarto-search svg.aa-SubmitIcon{width:16px;height:16px}.sidebar-navigation a{color:inherit}.sidebar-title{margin-top:.25rem;padding-bottom:.5rem;font-size:1.3rem;line-height:1.6rem;visibility:visible}.sidebar-title>a{font-size:inherit;text-decoration:none}.sidebar-title .sidebar-tools-main{margin-top:-6px}@media(max-width: 991.98px){#quarto-sidebar div.sidebar-header{padding-top:.2em}}.sidebar-header-stacked .sidebar-title{margin-top:.6rem}.sidebar-logo{max-width:90%;padding-bottom:.5rem}.sidebar-logo-link{text-decoration:none}.sidebar-navigation li a{text-decoration:none}.sidebar-navigation .quarto-navigation-tool{opacity:.7;font-size:.875rem}#quarto-sidebar>nav>.sidebar-tools-main{margin-left:14px}.sidebar-tools-main{display:inline-flex;margin-left:0px;order:2}.sidebar-tools-main:not(.tools-wide){vertical-align:middle}.sidebar-navigation .quarto-navigation-tool.dropdown-toggle::after{display:none}.sidebar.sidebar-navigation>*{padding-top:1em}.sidebar-item{margin-bottom:.2em;line-height:1rem;margin-top:.4rem}.sidebar-section{padding-left:.5em;padding-bottom:.2em}.sidebar-item .sidebar-item-container{display:flex;justify-content:space-between;cursor:pointer}.sidebar-item-toggle:hover{cursor:pointer}.sidebar-item .sidebar-item-toggle .bi{font-size:.7rem;text-align:center}.sidebar-item .sidebar-item-toggle .bi-chevron-right::before{transition:transform 200ms ease}.sidebar-item .sidebar-item-toggle[aria-expanded=false] .bi-chevron-right::before{transform:none}.sidebar-item .sidebar-item-toggle[aria-expanded=true] .bi-chevron-right::before{transform:rotate(90deg)}.sidebar-item-text{width:100%}.sidebar-navigation .sidebar-divider{margin-left:0;margin-right:0;margin-top:.5rem;margin-bottom:.5rem}@media(max-width: 991.98px){.quarto-secondary-nav{display:block}.quarto-secondary-nav button.quarto-search-button{padding-right:0em;padding-left:2em}.quarto-secondary-nav button.quarto-btn-toggle{margin-left:-0.75rem;margin-right:.15rem}.quarto-secondary-nav nav.quarto-title-breadcrumbs{display:none}.quarto-secondary-nav nav.quarto-page-breadcrumbs{display:flex;align-items:center;padding-right:1em;margin-left:-0.25em}.quarto-secondary-nav nav.quarto-page-breadcrumbs a{text-decoration:none}.quarto-secondary-nav nav.quarto-page-breadcrumbs ol.breadcrumb{margin-bottom:0}}@media(min-width: 992px){.quarto-secondary-nav{display:none}}.quarto-title-breadcrumbs .breadcrumb{margin-bottom:.5em;font-size:.9rem}.quarto-title-breadcrumbs .breadcrumb li:last-of-type a{color:#6c757d}.quarto-secondary-nav .quarto-btn-toggle{color:#faf1e4}.quarto-secondary-nav[aria-expanded=false] .quarto-btn-toggle .bi-chevron-right::before{transform:none}.quarto-secondary-nav[aria-expanded=true] .quarto-btn-toggle .bi-chevron-right::before{transform:rotate(90deg)}.quarto-secondary-nav .quarto-btn-toggle .bi-chevron-right::before{transition:transform 200ms ease}.quarto-secondary-nav{cursor:pointer}.no-decor{text-decoration:none}.quarto-secondary-nav-title{margin-top:.3em;color:#faf1e4;padding-top:4px}.quarto-secondary-nav nav.quarto-page-breadcrumbs{color:#faf1e4}.quarto-secondary-nav nav.quarto-page-breadcrumbs a{color:#faf1e4}.quarto-secondary-nav nav.quarto-page-breadcrumbs a:hover{color:rgba(174,209,228,.8)}.quarto-secondary-nav nav.quarto-page-breadcrumbs .breadcrumb-item::before{color:#eac48e}.breadcrumb-item{line-height:1.2rem}div.sidebar-item-container{color:#faf1e4}div.sidebar-item-container:hover,div.sidebar-item-container:focus{color:rgba(174,209,228,.8)}div.sidebar-item-container.disabled{color:rgba(250,241,228,.75)}div.sidebar-item-container .active,div.sidebar-item-container .show>.nav-link,div.sidebar-item-container .sidebar-link>code{color:#aed1e4}div.sidebar.sidebar-navigation.rollup.quarto-sidebar-toggle-contents,nav.sidebar.sidebar-navigation:not(.rollup){background-color:#353535}@media(max-width: 991.98px){.sidebar-navigation .sidebar-item a,.nav-page .nav-page-text,.sidebar-navigation{font-size:1rem}.sidebar-navigation ul.sidebar-section.depth1 .sidebar-section-item{font-size:1.1rem}.sidebar-logo{display:none}.sidebar.sidebar-navigation{position:static;border-bottom:1px solid #dee2e6}.sidebar.sidebar-navigation.collapsing{position:fixed;z-index:1000}.sidebar.sidebar-navigation.show{position:fixed;z-index:1000}.sidebar.sidebar-navigation{min-height:100%}nav.quarto-secondary-nav{background-color:#353535;border-bottom:1px solid #dee2e6}.quarto-banner nav.quarto-secondary-nav{background-color:#303030;color:#bdbdbd;border-top:1px solid #dee2e6}.sidebar .sidebar-footer{visibility:visible;padding-top:1rem;position:inherit}.sidebar-tools-collapse{display:block}}#quarto-sidebar{transition:width .15s ease-in}#quarto-sidebar>*{padding-right:1em}@media(max-width: 991.98px){#quarto-sidebar .sidebar-menu-container{white-space:nowrap;min-width:225px}#quarto-sidebar.show{transition:width .15s ease-out}}@media(min-width: 992px){#quarto-sidebar{display:flex;flex-direction:column}.nav-page .nav-page-text,.sidebar-navigation .sidebar-section .sidebar-item{font-size:.875rem}.sidebar-navigation .sidebar-item{font-size:.925rem}.sidebar.sidebar-navigation{display:block;position:sticky}.sidebar-search{width:100%}.sidebar .sidebar-footer{visibility:visible}}@media(min-width: 992px){#quarto-sidebar-glass{display:none}}@media(max-width: 991.98px){#quarto-sidebar-glass{position:fixed;top:0;bottom:0;left:0;right:0;background-color:rgba(255,255,255,0);transition:background-color .15s ease-in;z-index:-1}#quarto-sidebar-glass.collapsing{z-index:1000}#quarto-sidebar-glass.show{transition:background-color .15s ease-out;background-color:rgba(102,102,102,.4);z-index:1000}}.sidebar .sidebar-footer{padding:.5rem 1rem;align-self:flex-end;color:#6c757d;width:100%}.quarto-page-breadcrumbs .breadcrumb-item+.breadcrumb-item,.quarto-page-breadcrumbs .breadcrumb-item{padding-right:.33em;padding-left:0}.quarto-page-breadcrumbs .breadcrumb-item::before{padding-right:.33em}.quarto-sidebar-footer{font-size:.875em}.sidebar-section .bi-chevron-right{vertical-align:middle}.sidebar-section .bi-chevron-right::before{font-size:.9em}.notransition{-webkit-transition:none !important;-moz-transition:none !important;-o-transition:none !important;transition:none !important}.btn:focus:not(:focus-visible){box-shadow:none}.page-navigation{display:flex;justify-content:space-between}.nav-page{padding-bottom:.75em}.nav-page .bi{font-size:1.8rem;vertical-align:middle}.nav-page .nav-page-text{padding-left:.25em;padding-right:.25em}.nav-page a{color:#6c757d;text-decoration:none;display:flex;align-items:center}.nav-page a:hover{color:#8ba7b6}.nav-footer .toc-actions{padding-bottom:.5em;padding-top:.5em}.nav-footer .toc-actions a,.nav-footer .toc-actions a:hover{text-decoration:none}.nav-footer .toc-actions ul{display:flex;list-style:none}.nav-footer .toc-actions ul :first-child{margin-left:auto}.nav-footer .toc-actions ul :last-child{margin-right:auto}.nav-footer .toc-actions ul li{padding-right:1.5em}.nav-footer .toc-actions ul li i.bi{padding-right:.4em}.nav-footer .toc-actions ul li:last-of-type{padding-right:0}.nav-footer{display:flex;flex-direction:row;flex-wrap:wrap;justify-content:space-between;align-items:baseline;text-align:center;padding-top:.5rem;padding-bottom:.5rem;background-color:#181818}body.nav-fixed{padding-top:64px}.nav-footer-contents{color:#6c757d;margin-top:.25rem}.nav-footer{min-height:3.5em;color:#828282}.nav-footer a{color:#828282}.nav-footer .nav-footer-left{font-size:.825em}.nav-footer .nav-footer-center{font-size:.825em}.nav-footer .nav-footer-right{font-size:.825em}.nav-footer-left .footer-items,.nav-footer-center .footer-items,.nav-footer-right .footer-items{display:inline-flex;padding-top:.3em;padding-bottom:.3em;margin-bottom:0em}.nav-footer-left .footer-items .nav-link,.nav-footer-center .footer-items .nav-link,.nav-footer-right .footer-items .nav-link{padding-left:.6em;padding-right:.6em}@media(min-width: 768px){.nav-footer-left{flex:1 1 0px;text-align:left}}@media(max-width: 575.98px){.nav-footer-left{margin-bottom:1em;flex:100%}}@media(min-width: 768px){.nav-footer-right{flex:1 1 0px;text-align:right}}@media(max-width: 575.98px){.nav-footer-right{margin-bottom:1em;flex:100%}}.nav-footer-center{text-align:center;min-height:3em}@media(min-width: 768px){.nav-footer-center{flex:1 1 0px}}.nav-footer-center .footer-items{justify-content:center}@media(max-width: 767.98px){.nav-footer-center{margin-bottom:1em;flex:100%}}@media(max-width: 767.98px){.nav-footer-center{margin-top:3em;order:10}}.navbar .quarto-reader-toggle.reader .quarto-reader-toggle-btn{background-color:#bdbdbd;border-radius:3px}@media(max-width: 991.98px){.quarto-reader-toggle{display:none}}.quarto-reader-toggle.reader.quarto-navigation-tool .quarto-reader-toggle-btn{background-color:#faf1e4;border-radius:3px}.quarto-reader-toggle .quarto-reader-toggle-btn{display:inline-flex;padding-left:.2em;padding-right:.2em;margin-left:-0.2em;margin-right:-0.2em;text-align:center}.navbar .quarto-reader-toggle:not(.reader) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(189, 189, 189, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}.navbar .quarto-reader-toggle.reader .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(48, 48, 48, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}.sidebar-navigation .quarto-reader-toggle:not(.reader) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(250, 241, 228, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}.sidebar-navigation .quarto-reader-toggle.reader .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(53, 53, 53, 1)" class="bi bi-body-text" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M0 .5A.5.5 0 0 1 .5 0h4a.5.5 0 0 1 0 1h-4A.5.5 0 0 1 0 .5Zm0 2A.5.5 0 0 1 .5 2h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm9 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-9 2A.5.5 0 0 1 .5 4h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm5 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5Zm-12 2A.5.5 0 0 1 .5 6h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5Zm8 0a.5.5 0 0 1 .5-.5h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm-8 2A.5.5 0 0 1 .5 8h5a.5.5 0 0 1 0 1h-5a.5.5 0 0 1-.5-.5Zm7 0a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5Zm-7 2a.5.5 0 0 1 .5-.5h8a.5.5 0 0 1 0 1h-8a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h4a.5.5 0 0 1 0 1h-4a.5.5 0 0 1-.5-.5Zm0 2a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 0 1h-2a.5.5 0 0 1-.5-.5Z"/></svg>')}#quarto-back-to-top{display:none;position:fixed;bottom:50px;background-color:#181818;border-radius:.25rem;box-shadow:0 .2rem .5rem #6c757d,0 0 .05rem #6c757d;color:#6c757d;text-decoration:none;font-size:.9em;text-align:center;left:50%;padding:.4rem .8rem;transform:translate(-50%, 0)}#quarto-announcement{padding:.5em;display:flex;justify-content:space-between;margin-bottom:0;font-size:.9em}#quarto-announcement .quarto-announcement-content{margin-right:auto}#quarto-announcement .quarto-announcement-content p{margin-bottom:0}#quarto-announcement .quarto-announcement-icon{margin-right:.5em;font-size:1.2em;margin-top:-0.15em}#quarto-announcement .quarto-announcement-action{cursor:pointer}.aa-DetachedSearchButtonQuery{display:none}.aa-DetachedOverlay ul.aa-List,#quarto-search-results ul.aa-List{list-style:none;padding-left:0}.aa-DetachedOverlay .aa-Panel,#quarto-search-results .aa-Panel{background-color:#181818;position:absolute;z-index:2000}#quarto-search-results .aa-Panel{max-width:400px}#quarto-search input{font-size:.925rem}@media(min-width: 992px){.navbar #quarto-search{margin-left:.25rem;order:999}}.navbar.navbar-expand-sm #quarto-search,.navbar.navbar-expand-md #quarto-search{order:999}@media(min-width: 992px){.navbar .quarto-navbar-tools{order:900}}@media(min-width: 992px){.navbar .quarto-navbar-tools.tools-end{margin-left:auto !important}}@media(max-width: 991.98px){#quarto-sidebar .sidebar-search{display:none}}#quarto-sidebar .sidebar-search .aa-Autocomplete{width:100%}.navbar .aa-Autocomplete .aa-Form{width:180px}.navbar #quarto-search.type-overlay .aa-Autocomplete{width:40px}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form{background-color:inherit;border:none}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form:focus-within{box-shadow:none;outline:none}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-InputWrapper{display:none}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-InputWrapper:focus-within{display:inherit}.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-Label svg,.navbar #quarto-search.type-overlay .aa-Autocomplete .aa-Form .aa-LoadingIndicator svg{width:26px;height:26px;color:#bdbdbd;opacity:1}.navbar #quarto-search.type-overlay .aa-Autocomplete svg.aa-SubmitIcon{width:26px;height:26px;color:#bdbdbd;opacity:1}.aa-Autocomplete .aa-Form,.aa-DetachedFormContainer .aa-Form{align-items:center;background-color:#242424;border:1px solid #dee2e6;border-radius:.25rem;color:#faf1e4;display:flex;line-height:1em;margin:0;position:relative;width:100%}.aa-Autocomplete .aa-Form:focus-within,.aa-DetachedFormContainer .aa-Form:focus-within{box-shadow:rgba(39,128,227,.6) 0 0 0 1px;outline:currentColor none medium}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix{align-items:center;display:flex;flex-shrink:0;order:1}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-Label,.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-Label,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator{cursor:initial;flex-shrink:0;padding:0;text-align:left}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-Label svg,.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator svg,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-Label svg,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator svg{color:#faf1e4;opacity:.5}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-SubmitButton,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-SubmitButton{appearance:none;background:none;border:0;margin:0}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator{align-items:center;display:flex;justify-content:center}.aa-Autocomplete .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator[hidden],.aa-DetachedFormContainer .aa-Form .aa-InputWrapperPrefix .aa-LoadingIndicator[hidden]{display:none}.aa-Autocomplete .aa-Form .aa-InputWrapper,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper{order:3;position:relative;width:100%}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input{appearance:none;background:none;border:0;color:#faf1e4;font:inherit;height:calc(1.5em + .1rem + 2px);padding:0;width:100%}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::placeholder,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::placeholder{color:#faf1e4;opacity:.8}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input:focus,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input:focus{border-color:none;box-shadow:none;outline:none}.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-decoration,.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-cancel-button,.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-button,.aa-Autocomplete .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-decoration,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-decoration,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-cancel-button,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-button,.aa-DetachedFormContainer .aa-Form .aa-InputWrapper .aa-Input::-webkit-search-results-decoration{display:none}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix{align-items:center;display:flex;order:4}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton{align-items:center;background:none;border:0;color:#faf1e4;opacity:.8;cursor:pointer;display:flex;margin:0;width:calc(1.5em + .1rem + 2px)}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:hover,.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:focus,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:hover,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton:focus{color:#faf1e4;opacity:.8}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton[hidden],.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton[hidden]{display:none}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-ClearButton svg,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-ClearButton svg{width:calc(1.5em + 0.75rem + calc(1px * 2))}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton{border:none;align-items:center;background:none;color:#faf1e4;opacity:.4;font-size:.7rem;cursor:pointer;display:none;margin:0;width:calc(1em + .1rem + 2px)}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:hover,.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:focus,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:hover,.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton:focus{color:#faf1e4;opacity:.8}.aa-Autocomplete .aa-Form .aa-InputWrapperSuffix .aa-CopyButton[hidden],.aa-DetachedFormContainer .aa-Form .aa-InputWrapperSuffix .aa-CopyButton[hidden]{display:none}.aa-PanelLayout:empty{display:none}.quarto-search-no-results.no-query{display:none}.aa-Source:has(.no-query){display:none}#quarto-search-results .aa-Panel{border:solid #dee2e6 1px}#quarto-search-results .aa-SourceNoResults{width:398px}.aa-DetachedOverlay .aa-Panel,#quarto-search-results .aa-Panel{max-height:65vh;overflow-y:auto;font-size:.925rem}.aa-DetachedOverlay .aa-SourceNoResults,#quarto-search-results .aa-SourceNoResults{height:60px;display:flex;justify-content:center;align-items:center}.aa-DetachedOverlay .search-error,#quarto-search-results .search-error{padding-top:10px;padding-left:20px;padding-right:20px;cursor:default}.aa-DetachedOverlay .search-error .search-error-title,#quarto-search-results .search-error .search-error-title{font-size:1.1rem;margin-bottom:.5rem}.aa-DetachedOverlay .search-error .search-error-title .search-error-icon,#quarto-search-results .search-error .search-error-title .search-error-icon{margin-right:8px}.aa-DetachedOverlay .search-error .search-error-text,#quarto-search-results .search-error .search-error-text{font-weight:300}.aa-DetachedOverlay .search-result-text,#quarto-search-results .search-result-text{font-weight:300;overflow:hidden;text-overflow:ellipsis;display:-webkit-box;-webkit-line-clamp:2;-webkit-box-orient:vertical;line-height:1.2rem;max-height:2.4rem}.aa-DetachedOverlay .aa-SourceHeader .search-result-header,#quarto-search-results .aa-SourceHeader .search-result-header{font-size:.875rem;background-color:#252525;padding-left:14px;padding-bottom:4px;padding-top:4px}.aa-DetachedOverlay .aa-SourceHeader .search-result-header-no-results,#quarto-search-results .aa-SourceHeader .search-result-header-no-results{display:none}.aa-DetachedOverlay .aa-SourceFooter .algolia-search-logo,#quarto-search-results .aa-SourceFooter .algolia-search-logo{width:110px;opacity:.85;margin:8px;float:right}.aa-DetachedOverlay .search-result-section,#quarto-search-results .search-result-section{font-size:.925em}.aa-DetachedOverlay a.search-result-link,#quarto-search-results a.search-result-link{color:inherit;text-decoration:none}.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item,#quarto-search-results li.aa-Item[aria-selected=true] .search-item{background-color:#2780e3}.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item.search-result-more,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-section,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-text,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-title-container,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-result-text-container,#quarto-search-results li.aa-Item[aria-selected=true] .search-item.search-result-more,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-section,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-text,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-title-container,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-result-text-container{color:#fff;background-color:#2780e3}.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item mark.search-match,.aa-DetachedOverlay li.aa-Item[aria-selected=true] .search-item .search-match.mark,#quarto-search-results li.aa-Item[aria-selected=true] .search-item mark.search-match,#quarto-search-results li.aa-Item[aria-selected=true] .search-item .search-match.mark{color:#fff;background-color:#1a6cc7}.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item,#quarto-search-results li.aa-Item[aria-selected=false] .search-item{background-color:#242424}.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item.search-result-more,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-section,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-text,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-title-container,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-result-text-container,#quarto-search-results li.aa-Item[aria-selected=false] .search-item.search-result-more,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-section,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-text,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-title-container,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-result-text-container{color:#faf1e4}.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item mark.search-match,.aa-DetachedOverlay li.aa-Item[aria-selected=false] .search-item .search-match.mark,#quarto-search-results li.aa-Item[aria-selected=false] .search-item mark.search-match,#quarto-search-results li.aa-Item[aria-selected=false] .search-item .search-match.mark{color:inherit;background-color:#06192e}.aa-DetachedOverlay .aa-Item .search-result-doc:not(.document-selectable) .search-result-title-container,#quarto-search-results .aa-Item .search-result-doc:not(.document-selectable) .search-result-title-container{background-color:#242424;color:#faf1e4}.aa-DetachedOverlay .aa-Item .search-result-doc:not(.document-selectable) .search-result-text-container,#quarto-search-results .aa-Item .search-result-doc:not(.document-selectable) .search-result-text-container{padding-top:0px}.aa-DetachedOverlay li.aa-Item .search-result-doc.document-selectable .search-result-text-container,#quarto-search-results li.aa-Item .search-result-doc.document-selectable .search-result-text-container{margin-top:-4px}.aa-DetachedOverlay .aa-Item,#quarto-search-results .aa-Item{cursor:pointer}.aa-DetachedOverlay .aa-Item .search-item,#quarto-search-results .aa-Item .search-item{border-left:none;border-right:none;border-top:none;background-color:#242424;border-color:#dee2e6;color:#faf1e4}.aa-DetachedOverlay .aa-Item .search-item p,#quarto-search-results .aa-Item .search-item p{margin-top:0;margin-bottom:0}.aa-DetachedOverlay .aa-Item .search-item i.bi,#quarto-search-results .aa-Item .search-item i.bi{padding-left:8px;padding-right:8px;font-size:1.3em}.aa-DetachedOverlay .aa-Item .search-item .search-result-title,#quarto-search-results .aa-Item .search-item .search-result-title{margin-top:.3em;margin-bottom:0em}.aa-DetachedOverlay .aa-Item .search-item .search-result-crumbs,#quarto-search-results .aa-Item .search-item .search-result-crumbs{white-space:nowrap;text-overflow:ellipsis;font-size:.8em;font-weight:300;margin-right:1em}.aa-DetachedOverlay .aa-Item .search-item .search-result-crumbs:not(.search-result-crumbs-wrap),#quarto-search-results .aa-Item .search-item .search-result-crumbs:not(.search-result-crumbs-wrap){max-width:30%;margin-left:auto;margin-top:.5em;margin-bottom:.1rem}.aa-DetachedOverlay .aa-Item .search-item .search-result-crumbs.search-result-crumbs-wrap,#quarto-search-results .aa-Item .search-item .search-result-crumbs.search-result-crumbs-wrap{flex-basis:100%;margin-top:0em;margin-bottom:.2em;margin-left:37px}.aa-DetachedOverlay .aa-Item .search-result-title-container,#quarto-search-results .aa-Item .search-result-title-container{font-size:1em;display:flex;flex-wrap:wrap;padding:6px 4px 6px 4px}.aa-DetachedOverlay .aa-Item .search-result-text-container,#quarto-search-results .aa-Item .search-result-text-container{padding-bottom:8px;padding-right:8px;margin-left:42px}.aa-DetachedOverlay .aa-Item .search-result-doc-section,.aa-DetachedOverlay .aa-Item .search-result-more,#quarto-search-results .aa-Item .search-result-doc-section,#quarto-search-results .aa-Item .search-result-more{padding-top:8px;padding-bottom:8px;padding-left:44px}.aa-DetachedOverlay .aa-Item .search-result-more,#quarto-search-results .aa-Item .search-result-more{font-size:.8em;font-weight:400}.aa-DetachedOverlay .aa-Item .search-result-doc,#quarto-search-results .aa-Item .search-result-doc{border-top:1px solid #dee2e6}.aa-DetachedSearchButton{background:none;border:none}.aa-DetachedSearchButton .aa-DetachedSearchButtonPlaceholder{display:none}.navbar .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon{color:#bdbdbd}.sidebar-tools-collapse #quarto-search,.sidebar-tools-main #quarto-search{display:inline}.sidebar-tools-collapse #quarto-search .aa-Autocomplete,.sidebar-tools-main #quarto-search .aa-Autocomplete{display:inline}.sidebar-tools-collapse #quarto-search .aa-DetachedSearchButton,.sidebar-tools-main #quarto-search .aa-DetachedSearchButton{padding-left:4px;padding-right:4px}.sidebar-tools-collapse #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon,.sidebar-tools-main #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon{color:#faf1e4}.sidebar-tools-collapse #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon .aa-SubmitIcon,.sidebar-tools-main #quarto-search .aa-DetachedSearchButton .aa-DetachedSearchButtonIcon .aa-SubmitIcon{margin-top:-3px}.aa-DetachedContainer{background:rgba(24,24,24,.65);width:90%;bottom:0;box-shadow:rgba(222,226,230,.6) 0 0 0 1px;outline:currentColor none medium;display:flex;flex-direction:column;left:0;margin:0;overflow:hidden;padding:0;position:fixed;right:0;top:0;z-index:1101}.aa-DetachedContainer::after{height:32px}.aa-DetachedContainer .aa-SourceHeader{margin:var(--aa-spacing-half) 0 var(--aa-spacing-half) 2px}.aa-DetachedContainer .aa-Panel{background-color:#181818;border-radius:0;box-shadow:none;flex-grow:1;margin:0;padding:0;position:relative}.aa-DetachedContainer .aa-PanelLayout{bottom:0;box-shadow:none;left:0;margin:0;max-height:none;overflow-y:auto;position:absolute;right:0;top:0;width:100%}.aa-DetachedFormContainer{background-color:#181818;border-bottom:1px solid #dee2e6;display:flex;flex-direction:row;justify-content:space-between;margin:0;padding:.5em}.aa-DetachedCancelButton{background:none;font-size:.8em;border:0;border-radius:3px;color:#faf1e4;cursor:pointer;margin:0 0 0 .5em;padding:0 .5em}.aa-DetachedCancelButton:hover,.aa-DetachedCancelButton:focus{box-shadow:rgba(39,128,227,.6) 0 0 0 1px;outline:currentColor none medium}.aa-DetachedContainer--modal{bottom:inherit;height:auto;margin:0 auto;position:absolute;top:100px;border-radius:6px;max-width:850px}@media(max-width: 575.98px){.aa-DetachedContainer--modal{width:100%;top:0px;border-radius:0px;border:none}}.aa-DetachedContainer--modal .aa-PanelLayout{max-height:var(--aa-detached-modal-max-height);padding-bottom:var(--aa-spacing-half);position:static}.aa-Detached{height:100vh;overflow:hidden}.aa-DetachedOverlay{background-color:rgba(250,241,228,.4);position:fixed;left:0;right:0;top:0;margin:0;padding:0;height:100vh;z-index:1100}.quarto-dashboard.nav-fixed.dashboard-sidebar #quarto-content.quarto-dashboard-content{padding:0em}.quarto-dashboard #quarto-content.quarto-dashboard-content{padding:1em}.quarto-dashboard #quarto-content.quarto-dashboard-content>*{padding-top:0}@media(min-width: 576px){.quarto-dashboard{height:100%}}.quarto-dashboard .card.valuebox.bslib-card.bg-primary{background-color:#5397e9 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-secondary{background-color:#343a40 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-success{background-color:#3aa716 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-info{background-color:rgba(153,84,187,.7019607843) !important}.quarto-dashboard .card.valuebox.bslib-card.bg-warning{background-color:#fa6400 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-danger{background-color:rgba(255,0,57,.7019607843) !important}.quarto-dashboard .card.valuebox.bslib-card.bg-light{background-color:#525252 !important}.quarto-dashboard .card.valuebox.bslib-card.bg-dark{background-color:#343a40 !important}.quarto-dashboard.dashboard-fill{display:flex;flex-direction:column}.quarto-dashboard #quarto-appendix{display:none}.quarto-dashboard #quarto-header #quarto-dashboard-header{border-top:solid 1px #4a4a4a;border-bottom:solid 1px #4a4a4a}.quarto-dashboard #quarto-header #quarto-dashboard-header>nav{padding-left:1em;padding-right:1em}.quarto-dashboard #quarto-header #quarto-dashboard-header>nav .navbar-brand-container{padding-left:0}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-toggler{margin-right:0}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-toggler-icon{height:1em;width:1em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(189, 189, 189)" class="bi bi-menu-button-wide" viewBox="0 0 16 16"><path d="M0 1.5A1.5 1.5 0 0 1 1.5 0h13A1.5 1.5 0 0 1 16 1.5v2A1.5 1.5 0 0 1 14.5 5h-13A1.5 1.5 0 0 1 0 3.5v-2zM1.5 1a.5.5 0 0 0-.5.5v2a.5.5 0 0 0 .5.5h13a.5.5 0 0 0 .5-.5v-2a.5.5 0 0 0-.5-.5h-13z"/><path d="M2 2.5a.5.5 0 0 1 .5-.5h3a.5.5 0 0 1 0 1h-3a.5.5 0 0 1-.5-.5zm10.823.323-.396-.396A.25.25 0 0 1 12.604 2h.792a.25.25 0 0 1 .177.427l-.396.396a.25.25 0 0 1-.354 0zM0 8a2 2 0 0 1 2-2h12a2 2 0 0 1 2 2v5a2 2 0 0 1-2 2H2a2 2 0 0 1-2-2V8zm1 3v2a1 1 0 0 0 1 1h12a1 1 0 0 0 1-1v-2H1zm14-1V8a1 1 0 0 0-1-1H2a1 1 0 0 0-1 1v2h14zM2 8.5a.5.5 0 0 1 .5-.5h9a.5.5 0 0 1 0 1h-9a.5.5 0 0 1-.5-.5zm0 4a.5.5 0 0 1 .5-.5h6a.5.5 0 0 1 0 1h-6a.5.5 0 0 1-.5-.5z"/></svg>')}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-brand-container{padding-right:1em}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-title{font-size:1.1em}.quarto-dashboard #quarto-header #quarto-dashboard-header .navbar-nav{font-size:.9em}.quarto-dashboard #quarto-dashboard-header .navbar{padding:0}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-container{padding-left:1em}.quarto-dashboard #quarto-dashboard-header .navbar.slim .navbar-brand-container .nav-link,.quarto-dashboard #quarto-dashboard-header .navbar.slim .navbar-nav .nav-link{padding:.7em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-color-scheme-toggle{order:9}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-toggler{margin-left:.5em;order:10}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-nav .nav-link{padding:.5em;height:100%;display:flex;align-items:center}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-nav .active{background-color:#444}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-brand-container{padding:.5em .5em .5em 0;display:flex;flex-direction:row;margin-right:2em;align-items:center}@media(max-width: 767.98px){.quarto-dashboard #quarto-dashboard-header .navbar .navbar-brand-container{margin-right:auto}}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse{align-self:stretch}@media(min-width: 768px){.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse{order:8}}@media(max-width: 767.98px){.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse{order:1000;padding-bottom:.5em}}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-collapse .navbar-nav{align-self:stretch}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-title{font-size:1.25em;line-height:1.1em;display:flex;flex-direction:row;flex-wrap:wrap;align-items:baseline}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-title .navbar-title-text{margin-right:.4em}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-title a{text-decoration:none;color:inherit}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-subtitle,.quarto-dashboard #quarto-dashboard-header .navbar .navbar-author{font-size:.9rem;margin-right:.5em}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-author{margin-left:auto}.quarto-dashboard #quarto-dashboard-header .navbar .navbar-logo{max-height:48px;min-height:30px;object-fit:cover;margin-right:1em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-dashboard-links{order:9;padding-right:1em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-dashboard-link-text{margin-left:.25em}.quarto-dashboard #quarto-dashboard-header .navbar .quarto-dashboard-link{padding-right:0em;padding-left:.7em;text-decoration:none;color:#bdbdbd}.quarto-dashboard .page-layout-custom .tab-content{padding:0;border:none}.quarto-dashboard-img-contain{height:100%;width:100%;object-fit:contain}@media(max-width: 575.98px){.quarto-dashboard .bslib-grid{grid-template-rows:minmax(1em, max-content) !important}.quarto-dashboard .sidebar-content{height:inherit}.quarto-dashboard .page-layout-custom{min-height:100vh}}.quarto-dashboard.dashboard-toolbar>.page-layout-custom,.quarto-dashboard.dashboard-sidebar>.page-layout-custom{padding:0}.quarto-dashboard .quarto-dashboard-content.quarto-dashboard-pages{padding:0}.quarto-dashboard .callout{margin-bottom:0;margin-top:0}.quarto-dashboard .html-fill-container figure{overflow:hidden}.quarto-dashboard bslib-tooltip .rounded-pill{border:solid #6c757d 1px}.quarto-dashboard bslib-tooltip .rounded-pill .svg{fill:#faf1e4}.quarto-dashboard .tabset .dashboard-card-no-title .nav-tabs{margin-left:0;margin-right:auto}.quarto-dashboard .tabset .tab-content{border:none}.quarto-dashboard .tabset .card-header .nav-link[role=tab]{margin-top:-6px;padding-top:6px;padding-bottom:6px}.quarto-dashboard .card.valuebox,.quarto-dashboard .card.bslib-value-box{min-height:3rem}.quarto-dashboard .card.valuebox .card-body,.quarto-dashboard .card.bslib-value-box .card-body{padding:0}.quarto-dashboard .bslib-value-box .value-box-value{font-size:clamp(.1em,15cqw,5em)}.quarto-dashboard .bslib-value-box .value-box-showcase .bi{font-size:clamp(.1em,max(18cqw,5.2cqh),5em);text-align:center;height:1em}.quarto-dashboard .bslib-value-box .value-box-showcase .bi::before{vertical-align:1em}.quarto-dashboard .bslib-value-box .value-box-area{margin-top:auto;margin-bottom:auto}.quarto-dashboard .card figure.quarto-float{display:flex;flex-direction:column;align-items:center}.quarto-dashboard .dashboard-scrolling{padding:1em}.quarto-dashboard .full-height{height:100%}.quarto-dashboard .showcase-bottom .value-box-grid{display:grid;grid-template-columns:1fr;grid-template-rows:1fr auto;grid-template-areas:"top" "bottom"}.quarto-dashboard .showcase-bottom .value-box-grid .value-box-showcase{grid-area:bottom;padding:0;margin:0}.quarto-dashboard .showcase-bottom .value-box-grid .value-box-showcase i.bi{font-size:4rem}.quarto-dashboard .showcase-bottom .value-box-grid .value-box-area{grid-area:top}.quarto-dashboard .tab-content{margin-bottom:0}.quarto-dashboard .bslib-card .bslib-navs-card-title{justify-content:stretch;align-items:end}.quarto-dashboard .card-header{display:flex;flex-wrap:wrap;justify-content:space-between}.quarto-dashboard .card-header .card-title{display:flex;flex-direction:column;justify-content:center;margin-bottom:0}.quarto-dashboard .tabset .card-toolbar{margin-bottom:1em}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout{border:none;gap:var(--bslib-spacer, 1rem)}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.main{padding:0}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.sidebar{border-radius:.25rem;border:1px solid rgba(0,0,0,.175)}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.collapse-toggle{display:none}@media(max-width: 767.98px){.quarto-dashboard .bslib-grid>.bslib-sidebar-layout{grid-template-columns:1fr;grid-template-rows:max-content 1fr}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout>.main{grid-column:1;grid-row:2}.quarto-dashboard .bslib-grid>.bslib-sidebar-layout .sidebar{grid-column:1;grid-row:1}}.quarto-dashboard .sidebar-right .sidebar{padding-left:2.5em}.quarto-dashboard .sidebar-right .collapse-toggle{left:2px}.quarto-dashboard .quarto-dashboard .sidebar-right button.collapse-toggle:not(.transitioning){left:unset}.quarto-dashboard aside.sidebar{padding-left:1em;padding-right:1em;background-color:rgba(52,58,64,.25);color:#faf1e4}.quarto-dashboard .bslib-sidebar-layout>div.main{padding:.7em}.quarto-dashboard .bslib-sidebar-layout button.collapse-toggle{margin-top:.3em}.quarto-dashboard .bslib-sidebar-layout .collapse-toggle{top:0}.quarto-dashboard .bslib-sidebar-layout.sidebar-collapsed:not(.transitioning):not(.sidebar-right) .collapse-toggle{left:2px}.quarto-dashboard .sidebar>section>.h3:first-of-type{margin-top:0em}.quarto-dashboard .sidebar .h3,.quarto-dashboard .sidebar .h4,.quarto-dashboard .sidebar .h5,.quarto-dashboard .sidebar .h6{margin-top:.5em}.quarto-dashboard .sidebar form{flex-direction:column;align-items:start;margin-bottom:1em}.quarto-dashboard .sidebar form div[class*=oi-][class$=-input]{flex-direction:column}.quarto-dashboard .sidebar form[class*=oi-][class$=-toggle]{flex-direction:row-reverse;align-items:center;justify-content:start}.quarto-dashboard .sidebar form input[type=range]{margin-top:.5em;margin-right:.8em;margin-left:1em}.quarto-dashboard .sidebar label{width:fit-content}.quarto-dashboard .sidebar .card-body{margin-bottom:2em}.quarto-dashboard .sidebar .shiny-input-container{margin-bottom:1em}.quarto-dashboard .sidebar .shiny-options-group{margin-top:0}.quarto-dashboard .sidebar .control-label{margin-bottom:.3em}.quarto-dashboard .card .card-body .quarto-layout-row{align-items:stretch}.quarto-dashboard .toolbar{font-size:.9em;display:flex;flex-direction:row;border-top:solid 1px #bcbfc0;padding:1em;flex-wrap:wrap;background-color:rgba(52,58,64,.25)}.quarto-dashboard .toolbar .cell-output-display{display:flex}.quarto-dashboard .toolbar .shiny-input-container{padding-bottom:.5em;margin-bottom:.5em;width:inherit}.quarto-dashboard .toolbar .shiny-input-container>.checkbox:first-child{margin-top:6px}.quarto-dashboard .toolbar>*:last-child{margin-right:0}.quarto-dashboard .toolbar>*>*{margin-right:1em;align-items:baseline}.quarto-dashboard .toolbar>*>*>a{text-decoration:none;margin-top:auto;margin-bottom:auto}.quarto-dashboard .toolbar .shiny-input-container{padding-bottom:0;margin-bottom:0}.quarto-dashboard .toolbar .shiny-input-container>*{flex-shrink:0;flex-grow:0}.quarto-dashboard .toolbar .form-group.shiny-input-container:not([role=group])>label{margin-bottom:0}.quarto-dashboard .toolbar .shiny-input-container.no-baseline{align-items:start;padding-top:6px}.quarto-dashboard .toolbar .shiny-input-container{display:flex;align-items:baseline}.quarto-dashboard .toolbar .shiny-input-container label{padding-right:.4em}.quarto-dashboard .toolbar .shiny-input-container .bslib-input-switch{margin-top:6px}.quarto-dashboard .toolbar input[type=text]{line-height:1;width:inherit}.quarto-dashboard .toolbar .input-daterange{width:inherit}.quarto-dashboard .toolbar .input-daterange input[type=text]{height:2.4em;width:10em}.quarto-dashboard .toolbar .input-daterange .input-group-addon{height:auto;padding:0;margin-left:-5px !important;margin-right:-5px}.quarto-dashboard .toolbar .input-daterange .input-group-addon .input-group-text{padding-top:0;padding-bottom:0;height:100%}.quarto-dashboard .toolbar span.irs.irs--shiny{width:10em}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-line{top:9px}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-min,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-max,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-from,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-to,.quarto-dashboard .toolbar span.irs.irs--shiny .irs-single{top:20px}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-bar{top:8px}.quarto-dashboard .toolbar span.irs.irs--shiny .irs-handle{top:0px}.quarto-dashboard .toolbar .shiny-input-checkboxgroup>label{margin-top:6px}.quarto-dashboard .toolbar .shiny-input-checkboxgroup>.shiny-options-group{margin-top:0;align-items:baseline}.quarto-dashboard .toolbar .shiny-input-radiogroup>label{margin-top:6px}.quarto-dashboard .toolbar .shiny-input-radiogroup>.shiny-options-group{align-items:baseline;margin-top:0}.quarto-dashboard .toolbar .shiny-input-radiogroup>.shiny-options-group>.radio{margin-right:.3em}.quarto-dashboard .toolbar .form-select{padding-top:.2em;padding-bottom:.2em}.quarto-dashboard .toolbar .shiny-input-select{min-width:6em}.quarto-dashboard .toolbar div.checkbox{margin-bottom:0px}.quarto-dashboard .toolbar>.checkbox:first-child{margin-top:6px}.quarto-dashboard .toolbar form{width:fit-content}.quarto-dashboard .toolbar form label{padding-top:.2em;padding-bottom:.2em;width:fit-content}.quarto-dashboard .toolbar form input[type=date]{width:fit-content}.quarto-dashboard .toolbar form input[type=color]{width:3em}.quarto-dashboard .toolbar form button{padding:.4em}.quarto-dashboard .toolbar form select{width:fit-content}.quarto-dashboard .toolbar>*{font-size:.9em;flex-grow:0}.quarto-dashboard .toolbar .shiny-input-container label{margin-bottom:1px}.quarto-dashboard .toolbar-bottom{margin-top:1em;margin-bottom:0 !important;order:2}.quarto-dashboard .quarto-dashboard-content>.dashboard-toolbar-container>.toolbar-content>.tab-content>.tab-pane>*:not(.bslib-sidebar-layout){padding:1em}.quarto-dashboard .quarto-dashboard-content>.dashboard-toolbar-container>.toolbar-content>*:not(.tab-content){padding:1em}.quarto-dashboard .quarto-dashboard-content>.tab-content>.dashboard-page>.dashboard-toolbar-container>.toolbar-content,.quarto-dashboard .quarto-dashboard-content>.tab-content>.dashboard-page:not(.dashboard-sidebar-container)>*:not(.dashboard-toolbar-container){padding:1em}.quarto-dashboard .toolbar-content{padding:0}.quarto-dashboard .quarto-dashboard-content.quarto-dashboard-pages .tab-pane>.dashboard-toolbar-container .toolbar{border-radius:0;margin-bottom:0}.quarto-dashboard .dashboard-toolbar-container.toolbar-toplevel .toolbar{border-bottom:1px solid rgba(0,0,0,.175)}.quarto-dashboard .dashboard-toolbar-container.toolbar-toplevel .toolbar-bottom{margin-top:0}.quarto-dashboard .dashboard-toolbar-container:not(.toolbar-toplevel) .toolbar{margin-bottom:1em;border-top:none;border-radius:.25rem;border:1px solid rgba(0,0,0,.175)}.quarto-dashboard .vega-embed.has-actions details{width:1.7em;height:2em;position:absolute !important;top:0;right:0}.quarto-dashboard .dashboard-toolbar-container{padding:0}.quarto-dashboard .card .card-header p:last-child,.quarto-dashboard .card .card-footer p:last-child{margin-bottom:0}.quarto-dashboard .card .card-body>.h4:first-child{margin-top:0}.quarto-dashboard .card .card-body{z-index:4}@media(max-width: 767.98px){.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_length,.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_info,.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_paginate{text-align:initial}.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_filter{text-align:right}.quarto-dashboard .card .card-body .itables div.dataTables_wrapper div.dataTables_paginate ul.pagination{justify-content:initial}}.quarto-dashboard .card .card-body .itables .dataTables_wrapper{display:flex;flex-wrap:wrap;justify-content:space-between;align-items:center;padding-top:0}.quarto-dashboard .card .card-body .itables .dataTables_wrapper table{flex-shrink:0}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons{margin-bottom:.5em;margin-left:auto;width:fit-content;float:right}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons.btn-group{background:#181818;border:none}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons .btn-secondary{background-color:#181818;background-image:none;border:solid #dee2e6 1px;padding:.2em .7em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dt-buttons .btn span{font-size:.8em;color:#faf1e4}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_info{margin-left:.5em;margin-bottom:.5em;padding-top:0}@media(min-width: 768px){.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_info{font-size:.875em}}@media(max-width: 767.98px){.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_info{font-size:.8em}}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_filter{margin-bottom:.5em;font-size:.875em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_filter input[type=search]{padding:1px 5px 1px 5px;font-size:.875em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_length{flex-basis:1 1 50%;margin-bottom:.5em;font-size:.875em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_length select{padding:.4em 3em .4em .5em;font-size:.875em;margin-left:.2em;margin-right:.2em}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_paginate{flex-shrink:0}@media(min-width: 768px){.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_paginate{margin-left:auto}}.quarto-dashboard .card .card-body .itables .dataTables_wrapper .dataTables_paginate ul.pagination .paginate_button .page-link{font-size:.8em}.quarto-dashboard .card .card-footer{font-size:.9em}.quarto-dashboard .card .card-toolbar{display:flex;flex-grow:1;flex-direction:row;width:100%;flex-wrap:wrap}.quarto-dashboard .card .card-toolbar>*{font-size:.8em;flex-grow:0}.quarto-dashboard .card .card-toolbar>.card-title{font-size:1em;flex-grow:1;align-self:flex-start;margin-top:.1em}.quarto-dashboard .card .card-toolbar .cell-output-display{display:flex}.quarto-dashboard .card .card-toolbar .shiny-input-container{padding-bottom:.5em;margin-bottom:.5em;width:inherit}.quarto-dashboard .card .card-toolbar .shiny-input-container>.checkbox:first-child{margin-top:6px}.quarto-dashboard .card .card-toolbar>*:last-child{margin-right:0}.quarto-dashboard .card .card-toolbar>*>*{margin-right:1em;align-items:baseline}.quarto-dashboard .card .card-toolbar>*>*>a{text-decoration:none;margin-top:auto;margin-bottom:auto}.quarto-dashboard .card .card-toolbar form{width:fit-content}.quarto-dashboard .card .card-toolbar form label{padding-top:.2em;padding-bottom:.2em;width:fit-content}.quarto-dashboard .card .card-toolbar form input[type=date]{width:fit-content}.quarto-dashboard .card .card-toolbar form input[type=color]{width:3em}.quarto-dashboard .card .card-toolbar form button{padding:.4em}.quarto-dashboard .card .card-toolbar form select{width:fit-content}.quarto-dashboard .card .card-toolbar .cell-output-display{display:flex}.quarto-dashboard .card .card-toolbar .shiny-input-container{padding-bottom:.5em;margin-bottom:.5em;width:inherit}.quarto-dashboard .card .card-toolbar .shiny-input-container>.checkbox:first-child{margin-top:6px}.quarto-dashboard .card .card-toolbar>*:last-child{margin-right:0}.quarto-dashboard .card .card-toolbar>*>*{margin-right:1em;align-items:baseline}.quarto-dashboard .card .card-toolbar>*>*>a{text-decoration:none;margin-top:auto;margin-bottom:auto}.quarto-dashboard .card .card-toolbar .shiny-input-container{padding-bottom:0;margin-bottom:0}.quarto-dashboard .card .card-toolbar .shiny-input-container>*{flex-shrink:0;flex-grow:0}.quarto-dashboard .card .card-toolbar .form-group.shiny-input-container:not([role=group])>label{margin-bottom:0}.quarto-dashboard .card .card-toolbar .shiny-input-container.no-baseline{align-items:start;padding-top:6px}.quarto-dashboard .card .card-toolbar .shiny-input-container{display:flex;align-items:baseline}.quarto-dashboard .card .card-toolbar .shiny-input-container label{padding-right:.4em}.quarto-dashboard .card .card-toolbar .shiny-input-container .bslib-input-switch{margin-top:6px}.quarto-dashboard .card .card-toolbar input[type=text]{line-height:1;width:inherit}.quarto-dashboard .card .card-toolbar .input-daterange{width:inherit}.quarto-dashboard .card .card-toolbar .input-daterange input[type=text]{height:2.4em;width:10em}.quarto-dashboard .card .card-toolbar .input-daterange .input-group-addon{height:auto;padding:0;margin-left:-5px !important;margin-right:-5px}.quarto-dashboard .card .card-toolbar .input-daterange .input-group-addon .input-group-text{padding-top:0;padding-bottom:0;height:100%}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny{width:10em}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-line{top:9px}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-min,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-max,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-from,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-to,.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-single{top:20px}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-bar{top:8px}.quarto-dashboard .card .card-toolbar span.irs.irs--shiny .irs-handle{top:0px}.quarto-dashboard .card .card-toolbar .shiny-input-checkboxgroup>label{margin-top:6px}.quarto-dashboard .card .card-toolbar .shiny-input-checkboxgroup>.shiny-options-group{margin-top:0;align-items:baseline}.quarto-dashboard .card .card-toolbar .shiny-input-radiogroup>label{margin-top:6px}.quarto-dashboard .card .card-toolbar .shiny-input-radiogroup>.shiny-options-group{align-items:baseline;margin-top:0}.quarto-dashboard .card .card-toolbar .shiny-input-radiogroup>.shiny-options-group>.radio{margin-right:.3em}.quarto-dashboard .card .card-toolbar .form-select{padding-top:.2em;padding-bottom:.2em}.quarto-dashboard .card .card-toolbar .shiny-input-select{min-width:6em}.quarto-dashboard .card .card-toolbar div.checkbox{margin-bottom:0px}.quarto-dashboard .card .card-toolbar>.checkbox:first-child{margin-top:6px}.quarto-dashboard .card-body>table>thead{border-top:none}.quarto-dashboard .card-body>.table>:not(caption)>*>*{background-color:#181818}.tableFloatingHeaderOriginal{background-color:#181818;position:sticky !important;top:0 !important}.dashboard-data-table{margin-top:-1px}div.value-box-area span.observablehq--number{font-size:calc(clamp(.1em,15cqw,5em)*1.25);line-height:1.2;color:inherit;font-family:var(--bs-body-font-family)}.quarto-listing{padding-bottom:1em}.listing-pagination{padding-top:.5em}ul.pagination{float:right;padding-left:8px;padding-top:.5em}ul.pagination li{padding-right:.75em}ul.pagination li.disabled a,ul.pagination li.active a{color:#fff;text-decoration:none}ul.pagination li:last-of-type{padding-right:0}.listing-actions-group{display:flex}.quarto-listing-filter{margin-bottom:1em;width:200px;margin-left:auto}.quarto-listing-sort{margin-bottom:1em;margin-right:auto;width:auto}.quarto-listing-sort .input-group-text{font-size:.8em}.input-group-text{border-right:none}.quarto-listing-sort select.form-select{font-size:.8em}.listing-no-matching{text-align:center;padding-top:2em;padding-bottom:3em;font-size:1em}#quarto-margin-sidebar .quarto-listing-category{padding-top:0;font-size:1rem}#quarto-margin-sidebar .quarto-listing-category-title{cursor:pointer;font-weight:600;font-size:1rem}.quarto-listing-category .category{cursor:pointer}.quarto-listing-category .category.active{font-weight:600}.quarto-listing-category.category-cloud{display:flex;flex-wrap:wrap;align-items:baseline}.quarto-listing-category.category-cloud .category{padding-right:5px}.quarto-listing-category.category-cloud .category-cloud-1{font-size:.75em}.quarto-listing-category.category-cloud .category-cloud-2{font-size:.95em}.quarto-listing-category.category-cloud .category-cloud-3{font-size:1.15em}.quarto-listing-category.category-cloud .category-cloud-4{font-size:1.35em}.quarto-listing-category.category-cloud .category-cloud-5{font-size:1.55em}.quarto-listing-category.category-cloud .category-cloud-6{font-size:1.75em}.quarto-listing-category.category-cloud .category-cloud-7{font-size:1.95em}.quarto-listing-category.category-cloud .category-cloud-8{font-size:2.15em}.quarto-listing-category.category-cloud .category-cloud-9{font-size:2.35em}.quarto-listing-category.category-cloud .category-cloud-10{font-size:2.55em}.quarto-listing-cols-1{grid-template-columns:repeat(1, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-1{grid-template-columns:repeat(1, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-1{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-2{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-2{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-2{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-3{grid-template-columns:repeat(3, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-3{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-3{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-4{grid-template-columns:repeat(4, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-4{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-4{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-5{grid-template-columns:repeat(5, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-5{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-5{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-6{grid-template-columns:repeat(6, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-6{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-6{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-7{grid-template-columns:repeat(7, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-7{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-7{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-8{grid-template-columns:repeat(8, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-8{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-8{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-9{grid-template-columns:repeat(9, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-9{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-9{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-10{grid-template-columns:repeat(10, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-10{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-10{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-11{grid-template-columns:repeat(11, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-11{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-11{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-cols-12{grid-template-columns:repeat(12, minmax(0, 1fr));gap:1.5em}@media(max-width: 767.98px){.quarto-listing-cols-12{grid-template-columns:repeat(2, minmax(0, 1fr));gap:1.5em}}@media(max-width: 575.98px){.quarto-listing-cols-12{grid-template-columns:minmax(0, 1fr);gap:1.5em}}.quarto-listing-grid{gap:1.5em}.quarto-grid-item.borderless{border:none}.quarto-grid-item.borderless .listing-categories .listing-category:last-of-type,.quarto-grid-item.borderless .listing-categories .listing-category:first-of-type{padding-left:0}.quarto-grid-item.borderless .listing-categories .listing-category{border:0}.quarto-grid-link{text-decoration:none;color:inherit}.quarto-grid-link:hover{text-decoration:none;color:inherit}.quarto-grid-item h5.title,.quarto-grid-item .title.h5{margin-top:0;margin-bottom:0}.quarto-grid-item .card-footer{display:flex;justify-content:space-between;font-size:.8em}.quarto-grid-item .card-footer p{margin-bottom:0}.quarto-grid-item p.card-img-top{margin-bottom:0}.quarto-grid-item p.card-img-top>img{object-fit:cover}.quarto-grid-item .card-other-values{margin-top:.5em;font-size:.8em}.quarto-grid-item .card-other-values tr{margin-bottom:.5em}.quarto-grid-item .card-other-values tr>td:first-of-type{font-weight:600;padding-right:1em;padding-left:1em;vertical-align:top}.quarto-grid-item div.post-contents{display:flex;flex-direction:column;text-decoration:none;height:100%}.quarto-grid-item .listing-item-img-placeholder{background-color:rgba(52,58,64,.25);flex-shrink:0}.quarto-grid-item .card-attribution{padding-top:1em;display:flex;gap:1em;text-transform:uppercase;color:#6c757d;font-weight:500;flex-grow:10;align-items:flex-end}.quarto-grid-item .description{padding-bottom:1em}.quarto-grid-item .card-attribution .date{align-self:flex-end}.quarto-grid-item .card-attribution.justify{justify-content:space-between}.quarto-grid-item .card-attribution.start{justify-content:flex-start}.quarto-grid-item .card-attribution.end{justify-content:flex-end}.quarto-grid-item .card-title{margin-bottom:.1em}.quarto-grid-item .card-subtitle{padding-top:.25em}.quarto-grid-item .card-text{font-size:.9em}.quarto-grid-item .listing-reading-time{padding-bottom:.25em}.quarto-grid-item .card-text-small{font-size:.8em}.quarto-grid-item .card-subtitle.subtitle{font-size:.9em;font-weight:600;padding-bottom:.5em}.quarto-grid-item .listing-categories{display:flex;flex-wrap:wrap;padding-bottom:5px}.quarto-grid-item .listing-categories .listing-category{color:#6c757d;border:solid 1px #dee2e6;border-radius:.25rem;text-transform:uppercase;font-size:.65em;padding-left:.5em;padding-right:.5em;padding-top:.15em;padding-bottom:.15em;cursor:pointer;margin-right:4px;margin-bottom:4px}.quarto-grid-item.card-right{text-align:right}.quarto-grid-item.card-right .listing-categories{justify-content:flex-end}.quarto-grid-item.card-left{text-align:left}.quarto-grid-item.card-center{text-align:center}.quarto-grid-item.card-center .listing-description{text-align:justify}.quarto-grid-item.card-center .listing-categories{justify-content:center}table.quarto-listing-table td.image{padding:0px}table.quarto-listing-table td.image img{width:100%;max-width:50px;object-fit:contain}table.quarto-listing-table a{text-decoration:none;word-break:keep-all}table.quarto-listing-table th a{color:inherit}table.quarto-listing-table th a.asc:after{margin-bottom:-2px;margin-left:5px;display:inline-block;height:1rem;width:1rem;background-repeat:no-repeat;background-size:1rem 1rem;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-sort-up" viewBox="0 0 16 16"><path d="M3.5 12.5a.5.5 0 0 1-1 0V3.707L1.354 4.854a.5.5 0 1 1-.708-.708l2-1.999.007-.007a.498.498 0 0 1 .7.006l2 2a.5.5 0 1 1-.707.708L3.5 3.707V12.5zm3.5-9a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5zM7.5 6a.5.5 0 0 0 0 1h5a.5.5 0 0 0 0-1h-5zm0 3a.5.5 0 0 0 0 1h3a.5.5 0 0 0 0-1h-3zm0 3a.5.5 0 0 0 0 1h1a.5.5 0 0 0 0-1h-1z"/></svg>');content:""}table.quarto-listing-table th a.desc:after{margin-bottom:-2px;margin-left:5px;display:inline-block;height:1rem;width:1rem;background-repeat:no-repeat;background-size:1rem 1rem;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-sort-down" viewBox="0 0 16 16"><path d="M3.5 2.5a.5.5 0 0 0-1 0v8.793l-1.146-1.147a.5.5 0 0 0-.708.708l2 1.999.007.007a.497.497 0 0 0 .7-.006l2-2a.5.5 0 0 0-.707-.708L3.5 11.293V2.5zm3.5 1a.5.5 0 0 1 .5-.5h7a.5.5 0 0 1 0 1h-7a.5.5 0 0 1-.5-.5zM7.5 6a.5.5 0 0 0 0 1h5a.5.5 0 0 0 0-1h-5zm0 3a.5.5 0 0 0 0 1h3a.5.5 0 0 0 0-1h-3zm0 3a.5.5 0 0 0 0 1h1a.5.5 0 0 0 0-1h-1z"/></svg>');content:""}table.quarto-listing-table.table-hover td{cursor:pointer}.quarto-post.image-left{flex-direction:row}.quarto-post.image-right{flex-direction:row-reverse}@media(max-width: 767.98px){.quarto-post.image-right,.quarto-post.image-left{gap:0em;flex-direction:column}.quarto-post .metadata{padding-bottom:1em;order:2}.quarto-post .body{order:1}.quarto-post .thumbnail{order:3}}.list.quarto-listing-default div:last-of-type{border-bottom:none}@media(min-width: 992px){.quarto-listing-container-default{margin-right:2em}}div.quarto-post{display:flex;gap:2em;margin-bottom:1.5em;border-bottom:1px solid #dee2e6}@media(max-width: 767.98px){div.quarto-post{padding-bottom:1em}}div.quarto-post .metadata{flex-basis:20%;flex-grow:0;margin-top:.2em;flex-shrink:10}div.quarto-post .thumbnail{flex-basis:30%;flex-grow:0;flex-shrink:0}div.quarto-post .thumbnail img{margin-top:.4em;width:100%;object-fit:cover}div.quarto-post .body{flex-basis:45%;flex-grow:1;flex-shrink:0}div.quarto-post .body h3.listing-title,div.quarto-post .body .listing-title.h3{margin-top:0px;margin-bottom:0px;border-bottom:none}div.quarto-post .body .listing-subtitle{font-size:.875em;margin-bottom:.5em;margin-top:.2em}div.quarto-post .body .description{font-size:.9em}div.quarto-post .body pre code{white-space:pre-wrap}div.quarto-post a{color:#faf1e4;text-decoration:none}div.quarto-post .metadata{display:flex;flex-direction:column;font-size:.8em;font-family:"Source Sans Pro",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,sans-serif,"Apple Color Emoji","Segoe UI Emoji","Segoe UI Symbol";flex-basis:33%}div.quarto-post .listing-categories{display:flex;flex-wrap:wrap;padding-bottom:5px}div.quarto-post .listing-categories .listing-category{color:#6c757d;border:solid 1px #dee2e6;border-radius:.25rem;text-transform:uppercase;font-size:.65em;padding-left:.5em;padding-right:.5em;padding-top:.15em;padding-bottom:.15em;cursor:pointer;margin-right:4px;margin-bottom:4px}div.quarto-post .listing-description{margin-bottom:.5em}div.quarto-about-jolla{display:flex !important;flex-direction:column;align-items:center;margin-top:10%;padding-bottom:1em}div.quarto-about-jolla .about-image{object-fit:cover;margin-left:auto;margin-right:auto;margin-bottom:1.5em}div.quarto-about-jolla img.round{border-radius:50%}div.quarto-about-jolla img.rounded{border-radius:10px}div.quarto-about-jolla .quarto-title h1.title,div.quarto-about-jolla .quarto-title .title.h1{text-align:center}div.quarto-about-jolla .quarto-title .description{text-align:center}div.quarto-about-jolla h2,div.quarto-about-jolla .h2{border-bottom:none}div.quarto-about-jolla .about-sep{width:60%}div.quarto-about-jolla main{text-align:center}div.quarto-about-jolla .about-links{display:flex}@media(min-width: 992px){div.quarto-about-jolla .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-jolla .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-jolla .about-link{color:#fff;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-jolla .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-jolla .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-jolla .about-link:hover{color:#aed1e4}div.quarto-about-jolla .about-link i.bi{margin-right:.15em}div.quarto-about-solana{display:flex !important;flex-direction:column;padding-top:3em !important;padding-bottom:1em}div.quarto-about-solana .about-entity{display:flex !important;align-items:start;justify-content:space-between}@media(min-width: 992px){div.quarto-about-solana .about-entity{flex-direction:row}}@media(max-width: 991.98px){div.quarto-about-solana .about-entity{flex-direction:column-reverse;align-items:center;text-align:center}}div.quarto-about-solana .about-entity .entity-contents{display:flex;flex-direction:column}@media(max-width: 767.98px){div.quarto-about-solana .about-entity .entity-contents{width:100%}}div.quarto-about-solana .about-entity .about-image{object-fit:cover}@media(max-width: 991.98px){div.quarto-about-solana .about-entity .about-image{margin-bottom:1.5em}}div.quarto-about-solana .about-entity img.round{border-radius:50%}div.quarto-about-solana .about-entity img.rounded{border-radius:10px}div.quarto-about-solana .about-entity .about-links{display:flex;justify-content:left;padding-bottom:1.2em}@media(min-width: 992px){div.quarto-about-solana .about-entity .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-solana .about-entity .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-solana .about-entity .about-link{color:#fff;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-solana .about-entity .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-solana .about-entity .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-solana .about-entity .about-link:hover{color:#aed1e4}div.quarto-about-solana .about-entity .about-link i.bi{margin-right:.15em}div.quarto-about-solana .about-contents{padding-right:1.5em;flex-basis:0;flex-grow:1}div.quarto-about-solana .about-contents main.content{margin-top:0}div.quarto-about-solana .about-contents h2,div.quarto-about-solana .about-contents .h2{border-bottom:none}div.quarto-about-trestles{display:flex !important;flex-direction:row;padding-top:3em !important;padding-bottom:1em}@media(max-width: 991.98px){div.quarto-about-trestles{flex-direction:column;padding-top:0em !important}}div.quarto-about-trestles .about-entity{display:flex !important;flex-direction:column;align-items:center;text-align:center;padding-right:1em}@media(min-width: 992px){div.quarto-about-trestles .about-entity{flex:0 0 42%}}div.quarto-about-trestles .about-entity .about-image{object-fit:cover;margin-bottom:1.5em}div.quarto-about-trestles .about-entity img.round{border-radius:50%}div.quarto-about-trestles .about-entity img.rounded{border-radius:10px}div.quarto-about-trestles .about-entity .about-links{display:flex;justify-content:center}@media(min-width: 992px){div.quarto-about-trestles .about-entity .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-trestles .about-entity .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-trestles .about-entity .about-link{color:#fff;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-trestles .about-entity .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-trestles .about-entity .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-trestles .about-entity .about-link:hover{color:#aed1e4}div.quarto-about-trestles .about-entity .about-link i.bi{margin-right:.15em}div.quarto-about-trestles .about-contents{flex-basis:0;flex-grow:1}div.quarto-about-trestles .about-contents h2,div.quarto-about-trestles .about-contents .h2{border-bottom:none}@media(min-width: 992px){div.quarto-about-trestles .about-contents{border-left:solid 1px #dee2e6;padding-left:1.5em}}div.quarto-about-trestles .about-contents main.content{margin-top:0}div.quarto-about-marquee{padding-bottom:1em}div.quarto-about-marquee .about-contents{display:flex;flex-direction:column}div.quarto-about-marquee .about-image{max-height:550px;margin-bottom:1.5em;object-fit:cover}div.quarto-about-marquee img.round{border-radius:50%}div.quarto-about-marquee img.rounded{border-radius:10px}div.quarto-about-marquee h2,div.quarto-about-marquee .h2{border-bottom:none}div.quarto-about-marquee .about-links{display:flex;justify-content:center;padding-top:1.5em}@media(min-width: 992px){div.quarto-about-marquee .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-marquee .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-marquee .about-link{color:#fff;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-marquee .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-marquee .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-marquee .about-link:hover{color:#aed1e4}div.quarto-about-marquee .about-link i.bi{margin-right:.15em}@media(min-width: 992px){div.quarto-about-marquee .about-link{border:none}}div.quarto-about-broadside{display:flex;flex-direction:column;padding-bottom:1em}div.quarto-about-broadside .about-main{display:flex !important;padding-top:0 !important}@media(min-width: 992px){div.quarto-about-broadside .about-main{flex-direction:row;align-items:flex-start}}@media(max-width: 991.98px){div.quarto-about-broadside .about-main{flex-direction:column}}@media(max-width: 991.98px){div.quarto-about-broadside .about-main .about-entity{flex-shrink:0;width:100%;height:450px;margin-bottom:1.5em;background-size:cover;background-repeat:no-repeat}}@media(min-width: 992px){div.quarto-about-broadside .about-main .about-entity{flex:0 10 50%;margin-right:1.5em;width:100%;height:100%;background-size:100%;background-repeat:no-repeat}}div.quarto-about-broadside .about-main .about-contents{padding-top:14px;flex:0 0 50%}div.quarto-about-broadside h2,div.quarto-about-broadside .h2{border-bottom:none}div.quarto-about-broadside .about-sep{margin-top:1.5em;width:60%;align-self:center}div.quarto-about-broadside .about-links{display:flex;justify-content:center;column-gap:20px;padding-top:1.5em}@media(min-width: 992px){div.quarto-about-broadside .about-links{flex-direction:row;column-gap:.8em;row-gap:15px;flex-wrap:wrap}}@media(max-width: 991.98px){div.quarto-about-broadside .about-links{flex-direction:column;row-gap:1em;width:100%;padding-bottom:1.5em}}div.quarto-about-broadside .about-link{color:#fff;text-decoration:none;border:solid 1px}@media(min-width: 992px){div.quarto-about-broadside .about-link{font-size:.8em;padding:.25em .5em;border-radius:4px}}@media(max-width: 991.98px){div.quarto-about-broadside .about-link{font-size:1.1em;padding:.5em .5em;text-align:center;border-radius:6px}}div.quarto-about-broadside .about-link:hover{color:#aed1e4}div.quarto-about-broadside .about-link i.bi{margin-right:.15em}@media(min-width: 992px){div.quarto-about-broadside .about-link{border:none}}.tippy-box[data-theme~=quarto]{background-color:#181818;border:solid 1px #dee2e6;border-radius:.25rem;color:#faf1e4;font-size:.875rem}.tippy-box[data-theme~=quarto]>.tippy-backdrop{background-color:#181818}.tippy-box[data-theme~=quarto]>.tippy-arrow:after,.tippy-box[data-theme~=quarto]>.tippy-svg-arrow:after{content:"";position:absolute;z-index:-1}.tippy-box[data-theme~=quarto]>.tippy-arrow:after{border-color:rgba(0,0,0,0);border-style:solid}.tippy-box[data-placement^=top]>.tippy-arrow:before{bottom:-6px}.tippy-box[data-placement^=bottom]>.tippy-arrow:before{top:-6px}.tippy-box[data-placement^=right]>.tippy-arrow:before{left:-6px}.tippy-box[data-placement^=left]>.tippy-arrow:before{right:-6px}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-arrow:before{border-top-color:#181818}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-arrow:after{border-top-color:#dee2e6;border-width:7px 7px 0;top:17px;left:1px}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-svg-arrow>svg{top:16px}.tippy-box[data-theme~=quarto][data-placement^=top]>.tippy-svg-arrow:after{top:17px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-arrow:before{border-bottom-color:#181818;bottom:16px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-arrow:after{border-bottom-color:#dee2e6;border-width:0 7px 7px;bottom:17px;left:1px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-svg-arrow>svg{bottom:15px}.tippy-box[data-theme~=quarto][data-placement^=bottom]>.tippy-svg-arrow:after{bottom:17px}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-arrow:before{border-left-color:#181818}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-arrow:after{border-left-color:#dee2e6;border-width:7px 0 7px 7px;left:17px;top:1px}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-svg-arrow>svg{left:11px}.tippy-box[data-theme~=quarto][data-placement^=left]>.tippy-svg-arrow:after{left:12px}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-arrow:before{border-right-color:#181818;right:16px}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-arrow:after{border-width:7px 7px 7px 0;right:17px;top:1px;border-right-color:#dee2e6}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-svg-arrow>svg{right:11px}.tippy-box[data-theme~=quarto][data-placement^=right]>.tippy-svg-arrow:after{right:12px}.tippy-box[data-theme~=quarto]>.tippy-svg-arrow{fill:#faf1e4}.tippy-box[data-theme~=quarto]>.tippy-svg-arrow:after{background-image:url(data:image/svg+xml;base64,PHN2ZyB3aWR0aD0iMTYiIGhlaWdodD0iNiIgeG1sbnM9Imh0dHA6Ly93d3cudzMub3JnLzIwMDAvc3ZnIj48cGF0aCBkPSJNMCA2czEuNzk2LS4wMTMgNC42Ny0zLjYxNUM1Ljg1MS45IDYuOTMuMDA2IDggMGMxLjA3LS4wMDYgMi4xNDguODg3IDMuMzQzIDIuMzg1QzE0LjIzMyA2LjAwNSAxNiA2IDE2IDZIMHoiIGZpbGw9InJnYmEoMCwgOCwgMTYsIDAuMikiLz48L3N2Zz4=);background-size:16px 6px;width:16px;height:6px}.top-right{position:absolute;top:1em;right:1em}.visually-hidden{border:0;clip:rect(0 0 0 0);height:auto;margin:0;overflow:hidden;padding:0;position:absolute;width:1px;white-space:nowrap}.hidden{display:none !important}.zindex-bottom{z-index:-1 !important}figure.figure{display:block}.quarto-layout-panel{margin-bottom:1em}.quarto-layout-panel>figure{width:100%}.quarto-layout-panel>figure>figcaption,.quarto-layout-panel>.panel-caption{margin-top:10pt}.quarto-layout-panel>.table-caption{margin-top:0px}.table-caption p{margin-bottom:.5em}.quarto-layout-row{display:flex;flex-direction:row;align-items:flex-start}.quarto-layout-valign-top{align-items:flex-start}.quarto-layout-valign-bottom{align-items:flex-end}.quarto-layout-valign-center{align-items:center}.quarto-layout-cell{position:relative;margin-right:20px}.quarto-layout-cell:last-child{margin-right:0}.quarto-layout-cell figure,.quarto-layout-cell>p{margin:.2em}.quarto-layout-cell img{max-width:100%}.quarto-layout-cell .html-widget{width:100% !important}.quarto-layout-cell div figure p{margin:0}.quarto-layout-cell figure{display:block;margin-inline-start:0;margin-inline-end:0}.quarto-layout-cell table{display:inline-table}.quarto-layout-cell-subref figcaption,figure .quarto-layout-row figure figcaption{text-align:center;font-style:italic}.quarto-figure{position:relative;margin-bottom:1em}.quarto-figure>figure{width:100%;margin-bottom:0}.quarto-figure-left>figure>p,.quarto-figure-left>figure>div{text-align:left}.quarto-figure-center>figure>p,.quarto-figure-center>figure>div{text-align:center}.quarto-figure-right>figure>p,.quarto-figure-right>figure>div{text-align:right}.quarto-figure>figure>div.cell-annotation,.quarto-figure>figure>div code{text-align:left}figure>p:empty{display:none}figure>p:first-child{margin-top:0;margin-bottom:0}figure>figcaption.quarto-float-caption-bottom{margin-bottom:.5em}figure>figcaption.quarto-float-caption-top{margin-top:.5em}div[id^=tbl-]{position:relative}.quarto-figure>.anchorjs-link{position:absolute;top:.6em;right:.5em}div[id^=tbl-]>.anchorjs-link{position:absolute;top:.7em;right:.3em}.quarto-figure:hover>.anchorjs-link,div[id^=tbl-]:hover>.anchorjs-link,h2:hover>.anchorjs-link,.h2:hover>.anchorjs-link,h3:hover>.anchorjs-link,.h3:hover>.anchorjs-link,h4:hover>.anchorjs-link,.h4:hover>.anchorjs-link,h5:hover>.anchorjs-link,.h5:hover>.anchorjs-link,h6:hover>.anchorjs-link,.h6:hover>.anchorjs-link,.reveal-anchorjs-link>.anchorjs-link{opacity:1}#title-block-header{margin-block-end:1rem;position:relative;margin-top:-1px}#title-block-header .abstract{margin-block-start:1rem}#title-block-header .abstract .abstract-title{font-weight:600}#title-block-header a{text-decoration:none}#title-block-header .author,#title-block-header .date,#title-block-header .doi{margin-block-end:.2rem}#title-block-header .quarto-title-block>div{display:flex}#title-block-header .quarto-title-block>div>h1,#title-block-header .quarto-title-block>div>.h1{flex-grow:1}#title-block-header .quarto-title-block>div>button{flex-shrink:0;height:2.25rem;margin-top:0}@media(min-width: 992px){#title-block-header .quarto-title-block>div>button{margin-top:5px}}tr.header>th>p:last-of-type{margin-bottom:0px}table,table.table{margin-top:.5rem;margin-bottom:.5rem}caption,.table-caption{padding-top:.5rem;padding-bottom:.5rem;text-align:center}figure.quarto-float-tbl figcaption.quarto-float-caption-top{margin-top:.5rem;margin-bottom:.25rem;text-align:center}figure.quarto-float-tbl figcaption.quarto-float-caption-bottom{padding-top:.25rem;margin-bottom:.5rem;text-align:center}.utterances{max-width:none;margin-left:-8px}iframe{margin-bottom:1em}details{margin-bottom:1em}details[show]{margin-bottom:0}details>summary{color:#6c757d}details>summary>p:only-child{display:inline}pre.sourceCode,code.sourceCode{position:relative}dd code:not(.sourceCode),p code:not(.sourceCode){white-space:pre-wrap}code{white-space:pre}@media print{code{white-space:pre-wrap}}pre>code{display:block}pre>code.sourceCode{white-space:pre-wrap}pre>code.sourceCode>span>a:first-child::before{text-decoration:none}pre.code-overflow-wrap>code.sourceCode{white-space:pre-wrap}pre.code-overflow-scroll>code.sourceCode{white-space:pre}code a:any-link{color:inherit;text-decoration:none}code a:hover{color:inherit;text-decoration:underline}ul.task-list{padding-left:1em}[data-tippy-root]{display:inline-block}.tippy-content .footnote-back{display:none}.footnote-back{margin-left:.2em}.tippy-content{overflow-x:auto}.quarto-embedded-source-code{display:none}.quarto-unresolved-ref{font-weight:600}.quarto-cover-image{max-width:35%;float:right;margin-left:30px}.cell-output-display .widget-subarea{margin-bottom:1em}.cell-output-display:not(.no-overflow-x),.knitsql-table:not(.no-overflow-x){overflow-x:auto}.panel-input{margin-bottom:1em}.panel-input>div,.panel-input>div>div{display:inline-block;vertical-align:top;padding-right:12px}.panel-input>p:last-child{margin-bottom:0}.layout-sidebar{margin-bottom:1em}.layout-sidebar .tab-content{border:none}.tab-content>.page-columns.active{display:grid}div.sourceCode>iframe{width:100%;height:300px;margin-bottom:-0.5em}a{text-underline-offset:3px}.callout pre.sourceCode{padding-left:0}div.ansi-escaped-output{font-family:monospace;display:block}/*!
+*
+* ansi colors from IPython notebook's
+*
+* we also add `bright-[color]-` synonyms for the `-[color]-intense` classes since
+* that seems to be what ansi_up emits
+*
+*/.ansi-black-fg{color:#3e424d}.ansi-black-bg{background-color:#3e424d}.ansi-black-intense-black,.ansi-bright-black-fg{color:#282c36}.ansi-black-intense-black,.ansi-bright-black-bg{background-color:#282c36}.ansi-red-fg{color:#e75c58}.ansi-red-bg{background-color:#e75c58}.ansi-red-intense-red,.ansi-bright-red-fg{color:#b22b31}.ansi-red-intense-red,.ansi-bright-red-bg{background-color:#b22b31}.ansi-green-fg{color:#00a250}.ansi-green-bg{background-color:#00a250}.ansi-green-intense-green,.ansi-bright-green-fg{color:#007427}.ansi-green-intense-green,.ansi-bright-green-bg{background-color:#007427}.ansi-yellow-fg{color:#ddb62b}.ansi-yellow-bg{background-color:#ddb62b}.ansi-yellow-intense-yellow,.ansi-bright-yellow-fg{color:#b27d12}.ansi-yellow-intense-yellow,.ansi-bright-yellow-bg{background-color:#b27d12}.ansi-blue-fg{color:#208ffb}.ansi-blue-bg{background-color:#208ffb}.ansi-blue-intense-blue,.ansi-bright-blue-fg{color:#0065ca}.ansi-blue-intense-blue,.ansi-bright-blue-bg{background-color:#0065ca}.ansi-magenta-fg{color:#d160c4}.ansi-magenta-bg{background-color:#d160c4}.ansi-magenta-intense-magenta,.ansi-bright-magenta-fg{color:#a03196}.ansi-magenta-intense-magenta,.ansi-bright-magenta-bg{background-color:#a03196}.ansi-cyan-fg{color:#60c6c8}.ansi-cyan-bg{background-color:#60c6c8}.ansi-cyan-intense-cyan,.ansi-bright-cyan-fg{color:#258f8f}.ansi-cyan-intense-cyan,.ansi-bright-cyan-bg{background-color:#258f8f}.ansi-white-fg{color:#c5c1b4}.ansi-white-bg{background-color:#c5c1b4}.ansi-white-intense-white,.ansi-bright-white-fg{color:#a1a6b2}.ansi-white-intense-white,.ansi-bright-white-bg{background-color:#a1a6b2}.ansi-default-inverse-fg{color:#fff}.ansi-default-inverse-bg{background-color:#000}.ansi-bold{font-weight:bold}.ansi-underline{text-decoration:underline}:root{--quarto-body-bg: #181818;--quarto-body-color: #FAF1E4;--quarto-text-muted: #6c757d;--quarto-border-color: #dee2e6;--quarto-border-width: 1px}table.gt_table{color:var(--quarto-body-color);font-size:1em;width:100%;background-color:rgba(0,0,0,0);border-top-width:inherit;border-bottom-width:inherit;border-color:var(--quarto-border-color)}table.gt_table th.gt_column_spanner_outer{color:var(--quarto-body-color);background-color:rgba(0,0,0,0);border-top-width:inherit;border-bottom-width:inherit;border-color:var(--quarto-border-color)}table.gt_table th.gt_col_heading{color:var(--quarto-body-color);font-weight:bold;background-color:rgba(0,0,0,0)}table.gt_table thead.gt_col_headings{border-bottom:1px solid currentColor;border-top-width:inherit;border-top-color:var(--quarto-border-color)}table.gt_table thead.gt_col_headings:not(:first-child){border-top-width:1px;border-top-color:var(--quarto-border-color)}table.gt_table td.gt_row{border-bottom-width:1px;border-bottom-color:var(--quarto-border-color);border-top-width:0px}table.gt_table tbody.gt_table_body{border-top-width:1px;border-bottom-width:1px;border-bottom-color:var(--quarto-border-color);border-top-color:currentColor}div.columns{display:initial;gap:initial}div.column{display:inline-block;overflow-x:initial;vertical-align:top;width:50%}.code-annotation-tip-content{word-wrap:break-word}.code-annotation-container-hidden{display:none !important}dl.code-annotation-container-grid{display:grid;grid-template-columns:min-content auto}dl.code-annotation-container-grid dt{grid-column:1}dl.code-annotation-container-grid dd{grid-column:2}pre.sourceCode.code-annotation-code{padding-right:0}code.sourceCode .code-annotation-anchor{z-index:100;position:relative;float:right;background-color:rgba(0,0,0,0)}input[type=checkbox]{margin-right:.5ch}:root{--mermaid-bg-color: #181818;--mermaid-edge-color: #343a40;--mermaid-node-fg-color: #FAF1E4;--mermaid-fg-color: #FAF1E4;--mermaid-fg-color--lighter: white;--mermaid-fg-color--lightest: white;--mermaid-font-family: Source Sans Pro, -apple-system, BlinkMacSystemFont, Segoe UI, Roboto, Helvetica Neue, Arial, sans-serif, Apple Color Emoji, Segoe UI Emoji, Segoe UI Symbol;--mermaid-label-bg-color: #181818;--mermaid-label-fg-color: #2780e3;--mermaid-node-bg-color: rgba(39, 128, 227, 0.1);--mermaid-node-fg-color: #FAF1E4}@media print{:root{font-size:11pt}#quarto-sidebar,#TOC,.nav-page{display:none}.page-columns .content{grid-column-start:page-start}.fixed-top{position:relative}.panel-caption,.figure-caption,figcaption{color:#666}}.code-copy-button{position:absolute;top:0;right:0;border:0;margin-top:5px;margin-right:5px;background-color:rgba(0,0,0,0);z-index:3}.code-copy-button:focus{outline:none}.code-copy-button-tooltip{font-size:.75em}.code-copy-button>.bi::before{display:inline-block;height:1rem;width:1rem;content:"";vertical-align:-0.125em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(97, 110, 136)" viewBox="0 0 16 16"><path d="M4 1.5H3a2 2 0 0 0-2 2V14a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V3.5a2 2 0 0 0-2-2h-1v1h1a1 1 0 0 1 1 1V14a1 1 0 0 1-1 1H3a1 1 0 0 1-1-1V3.5a1 1 0 0 1 1-1h1v-1z"/><path d="M9.5 1a.5.5 0 0 1 .5.5v1a.5.5 0 0 1-.5.5h-3a.5.5 0 0 1-.5-.5v-1a.5.5 0 0 1 .5-.5h3zm-3-1A1.5 1.5 0 0 0 5 1.5v1A1.5 1.5 0 0 0 6.5 4h3A1.5 1.5 0 0 0 11 2.5v-1A1.5 1.5 0 0 0 9.5 0h-3z"/></svg>');background-repeat:no-repeat;background-size:1rem 1rem}.code-copy-button-checked>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(97, 110, 136)" viewBox="0 0 16 16"><path d="M13.854 3.646a.5.5 0 0 1 0 .708l-7 7a.5.5 0 0 1-.708 0l-3.5-3.5a.5.5 0 1 1 .708-.708L6.5 10.293l6.646-6.647a.5.5 0 0 1 .708 0z"/></svg>')}.code-copy-button:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(136, 192, 208)" viewBox="0 0 16 16"><path d="M4 1.5H3a2 2 0 0 0-2 2V14a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V3.5a2 2 0 0 0-2-2h-1v1h1a1 1 0 0 1 1 1V14a1 1 0 0 1-1 1H3a1 1 0 0 1-1-1V3.5a1 1 0 0 1 1-1h1v-1z"/><path d="M9.5 1a.5.5 0 0 1 .5.5v1a.5.5 0 0 1-.5.5h-3a.5.5 0 0 1-.5-.5v-1a.5.5 0 0 1 .5-.5h3zm-3-1A1.5 1.5 0 0 0 5 1.5v1A1.5 1.5 0 0 0 6.5 4h3A1.5 1.5 0 0 0 11 2.5v-1A1.5 1.5 0 0 0 9.5 0h-3z"/></svg>')}.code-copy-button-checked:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(136, 192, 208)"  viewBox="0 0 16 16"><path d="M13.854 3.646a.5.5 0 0 1 0 .708l-7 7a.5.5 0 0 1-.708 0l-3.5-3.5a.5.5 0 1 1 .708-.708L6.5 10.293l6.646-6.647a.5.5 0 0 1 .708 0z"/></svg>')}main ol ol,main ul ul,main ol ul,main ul ol{margin-bottom:1em}ul>li:not(:has(>p))>ul,ol>li:not(:has(>p))>ul,ul>li:not(:has(>p))>ol,ol>li:not(:has(>p))>ol{margin-bottom:0}ul>li:not(:has(>p))>ul>li:has(>p),ol>li:not(:has(>p))>ul>li:has(>p),ul>li:not(:has(>p))>ol>li:has(>p),ol>li:not(:has(>p))>ol>li:has(>p){margin-top:1rem}body{margin:0}main.page-columns>header>h1.title,main.page-columns>header>.title.h1{margin-bottom:0}@media(min-width: 992px){body .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset] 35px [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.fullcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset] 35px [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] 42px [page-end-inset page-end] 5fr [screen-end-inset] 1.5em}body.slimcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset] 35px [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 35px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.listing:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(950px - 3em)) [body-content-end] 3em [body-end] 50px [body-end-outset] minmax(0px, 300px) [page-end-inset] minmax(50px, 100px) [page-end] 1fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 175px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 175px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] minmax(25px, 50px) [page-start-inset] minmax(50px, 150px) [body-start-outset] minmax(25px, 50px) [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] minmax(30px, 60px) [body-end-outset] minmax(60px, 180px) [page-end-inset] minmax(30px, 60px) [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(60px, 120px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 50px [page-start-inset] minmax(50px, 150px) [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(450px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start] minmax(50px, 100px) [page-start-inset] 50px [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(0px, 240px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 50px [page-start-inset] minmax(50px, 150px) [body-start-outset] 50px [body-start] 1.5em [body-content-start] minmax(450px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(60px, 180px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] minmax(25px, 50px) [page-start-inset] minmax(50px, 150px) [body-start-outset] minmax(25px, 50px) [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] minmax(30px, 60px) [body-end-outset] minmax(60px, 180px) [page-end-inset] minmax(30px, 60px) [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}}@media(max-width: 991.98px){body .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.fullcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.slimcontent:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.listing:not(.floating):not(.docked) .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset] 5fr [body-start] 1.5em [body-content-start] minmax(500px, calc(1350px - 3em)) [body-content-end body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 145px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start] 35px [page-start-inset] minmax(0px, 145px) [body-start-outset] 35px [body-start] 1.5em [body-content-start] minmax(450px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1.5em [body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(90px, 180px) [page-end-inset] 30px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(30px, 60px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(1100px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1em [body-content-start] minmax(500px, calc(900px - 3em)) [body-content-end] 1.5em [body-end body-end-outset page-end-inset page-end] 4fr [screen-end-inset] 1.5em [screen-end]}body.docked.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(30px, 60px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.docked.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(30px, 60px) [page-end-inset] 60px [page-end] 5fr [screen-end-inset] 1.5em [screen-end]}body.floating.slimcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1em [body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 42px [body-end-outset] minmax(90px, 174px) [page-end-inset] 42px [page-end] 4fr [screen-end-inset] 1.5em [screen-end]}body.floating.listing .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset] 5fr [page-start page-start-inset body-start-outset body-start] 1em [body-content-start] minmax(500px, calc(850px - 3em)) [body-content-end] 1.5em [body-end] 60px [body-end-outset] minmax(90px, 180px) [page-end-inset] 30px [page-end] 4fr [screen-end-inset] 1.5em [screen-end]}}@media(max-width: 767.98px){body .page-columns,body.fullcontent:not(.floating):not(.docked) .page-columns,body.slimcontent:not(.floating):not(.docked) .page-columns,body.docked .page-columns,body.docked.slimcontent .page-columns,body.docked.fullcontent .page-columns,body.floating .page-columns,body.floating.slimcontent .page-columns,body.floating.fullcontent .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(0px, 1fr) [body-content-end body-end body-end-outset page-end-inset page-end screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(0px, 1fr) [body-content-end body-end body-end-outset page-end-inset page-end screen-end-inset] 1.5em [screen-end]}body:not(.floating):not(.docked) .page-columns.toc-left .page-columns{display:grid;gap:0;grid-template-columns:[screen-start] 1.5em [screen-start-inset page-start page-start-inset body-start-outset body-start body-content-start] minmax(0px, 1fr) [body-content-end body-end body-end-outset page-end-inset page-end screen-end-inset] 1.5em [screen-end]}nav[role=doc-toc]{display:none}}body,.page-row-navigation{grid-template-rows:[page-top] max-content [contents-top] max-content [contents-bottom] max-content [page-bottom]}.page-rows-contents{grid-template-rows:[content-top] minmax(max-content, 1fr) [content-bottom] minmax(60px, max-content) [page-bottom]}.page-full{grid-column:screen-start/screen-end !important}.page-columns>*{grid-column:body-content-start/body-content-end}.page-columns.column-page>*{grid-column:page-start/page-end}.page-columns.column-page-left .page-columns.page-full>*,.page-columns.column-page-left>*{grid-column:page-start/body-content-end}.page-columns.column-page-right .page-columns.page-full>*,.page-columns.column-page-right>*{grid-column:body-content-start/page-end}.page-rows{grid-auto-rows:auto}.header{grid-column:screen-start/screen-end;grid-row:page-top/contents-top}#quarto-content{padding:0;grid-column:screen-start/screen-end;grid-row:contents-top/contents-bottom}body.floating .sidebar.sidebar-navigation{grid-column:page-start/body-start;grid-row:content-top/page-bottom}body.docked .sidebar.sidebar-navigation{grid-column:screen-start/body-start;grid-row:content-top/page-bottom}.sidebar.toc-left{grid-column:page-start/body-start;grid-row:content-top/page-bottom}.sidebar.margin-sidebar{grid-column:body-end/page-end;grid-row:content-top/page-bottom}.page-columns .content{grid-column:body-content-start/body-content-end;grid-row:content-top/content-bottom;align-content:flex-start}.page-columns .page-navigation{grid-column:body-content-start/body-content-end;grid-row:content-bottom/page-bottom}.page-columns .footer{grid-column:screen-start/screen-end;grid-row:contents-bottom/page-bottom}.page-columns .column-body{grid-column:body-content-start/body-content-end}.page-columns .column-body-fullbleed{grid-column:body-start/body-end}.page-columns .column-body-outset{grid-column:body-start-outset/body-end-outset;z-index:998;opacity:.999}.page-columns .column-body-outset table{background:#181818}.page-columns .column-body-outset-left{grid-column:body-start-outset/body-content-end;z-index:998;opacity:.999}.page-columns .column-body-outset-left table{background:#181818}.page-columns .column-body-outset-right{grid-column:body-content-start/body-end-outset;z-index:998;opacity:.999}.page-columns .column-body-outset-right table{background:#181818}.page-columns .column-page{grid-column:page-start/page-end;z-index:998;opacity:.999}.page-columns .column-page table{background:#181818}.page-columns .column-page-inset{grid-column:page-start-inset/page-end-inset;z-index:998;opacity:.999}.page-columns .column-page-inset table{background:#181818}.page-columns .column-page-inset-left{grid-column:page-start-inset/body-content-end;z-index:998;opacity:.999}.page-columns .column-page-inset-left table{background:#181818}.page-columns .column-page-inset-right{grid-column:body-content-start/page-end-inset;z-index:998;opacity:.999}.page-columns .column-page-inset-right figcaption table{background:#181818}.page-columns .column-page-left{grid-column:page-start/body-content-end;z-index:998;opacity:.999}.page-columns .column-page-left table{background:#181818}.page-columns .column-page-right{grid-column:body-content-start/page-end;z-index:998;opacity:.999}.page-columns .column-page-right figcaption table{background:#181818}#quarto-content.page-columns #quarto-margin-sidebar,#quarto-content.page-columns #quarto-sidebar{z-index:1}@media(max-width: 991.98px){#quarto-content.page-columns #quarto-margin-sidebar.collapse,#quarto-content.page-columns #quarto-sidebar.collapse,#quarto-content.page-columns #quarto-margin-sidebar.collapsing,#quarto-content.page-columns #quarto-sidebar.collapsing{z-index:1055}}#quarto-content.page-columns main.column-page,#quarto-content.page-columns main.column-page-right,#quarto-content.page-columns main.column-page-left{z-index:0}.page-columns .column-screen-inset{grid-column:screen-start-inset/screen-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset table{background:#181818}.page-columns .column-screen-inset-left{grid-column:screen-start-inset/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-inset-left table{background:#181818}.page-columns .column-screen-inset-right{grid-column:body-content-start/screen-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset-right table{background:#181818}.page-columns .column-screen{grid-column:screen-start/screen-end;z-index:998;opacity:.999}.page-columns .column-screen table{background:#181818}.page-columns .column-screen-left{grid-column:screen-start/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-left table{background:#181818}.page-columns .column-screen-right{grid-column:body-content-start/screen-end;z-index:998;opacity:.999}.page-columns .column-screen-right table{background:#181818}.page-columns .column-screen-inset-shaded{grid-column:screen-start/screen-end;padding:1em;background:#525252;z-index:998;opacity:.999;margin-bottom:1em}.zindex-content{z-index:998;opacity:.999}.zindex-modal{z-index:1055;opacity:.999}.zindex-over-content{z-index:999;opacity:.999}img.img-fluid.column-screen,img.img-fluid.column-screen-inset-shaded,img.img-fluid.column-screen-inset,img.img-fluid.column-screen-inset-left,img.img-fluid.column-screen-inset-right,img.img-fluid.column-screen-left,img.img-fluid.column-screen-right{width:100%}@media(min-width: 992px){.margin-caption,div.aside,aside:not(.footnotes):not(.sidebar),.column-margin{grid-column:body-end/page-end !important;z-index:998}.column-sidebar{grid-column:page-start/body-start !important;z-index:998}.column-leftmargin{grid-column:screen-start-inset/body-start !important;z-index:998}.no-row-height{height:1em;overflow:visible}}@media(max-width: 991.98px){.margin-caption,div.aside,aside:not(.footnotes):not(.sidebar),.column-margin{grid-column:body-end/page-end !important;z-index:998}.no-row-height{height:1em;overflow:visible}.page-columns.page-full{overflow:visible}.page-columns.toc-left .margin-caption,.page-columns.toc-left div.aside,.page-columns.toc-left aside:not(.footnotes):not(.sidebar),.page-columns.toc-left .column-margin{grid-column:body-content-start/body-content-end !important;z-index:998;opacity:.999}.page-columns.toc-left .no-row-height{height:initial;overflow:initial}}@media(max-width: 767.98px){.margin-caption,div.aside,aside:not(.footnotes):not(.sidebar),.column-margin{grid-column:body-content-start/body-content-end !important;z-index:998;opacity:.999}.no-row-height{height:initial;overflow:initial}#quarto-margin-sidebar{display:none}#quarto-sidebar-toc-left{display:none}.hidden-sm{display:none}}.panel-grid{display:grid;grid-template-rows:repeat(1, 1fr);grid-template-columns:repeat(24, 1fr);gap:1em}.panel-grid .g-col-1{grid-column:auto/span 1}.panel-grid .g-col-2{grid-column:auto/span 2}.panel-grid .g-col-3{grid-column:auto/span 3}.panel-grid .g-col-4{grid-column:auto/span 4}.panel-grid .g-col-5{grid-column:auto/span 5}.panel-grid .g-col-6{grid-column:auto/span 6}.panel-grid .g-col-7{grid-column:auto/span 7}.panel-grid .g-col-8{grid-column:auto/span 8}.panel-grid .g-col-9{grid-column:auto/span 9}.panel-grid .g-col-10{grid-column:auto/span 10}.panel-grid .g-col-11{grid-column:auto/span 11}.panel-grid .g-col-12{grid-column:auto/span 12}.panel-grid .g-col-13{grid-column:auto/span 13}.panel-grid .g-col-14{grid-column:auto/span 14}.panel-grid .g-col-15{grid-column:auto/span 15}.panel-grid .g-col-16{grid-column:auto/span 16}.panel-grid .g-col-17{grid-column:auto/span 17}.panel-grid .g-col-18{grid-column:auto/span 18}.panel-grid .g-col-19{grid-column:auto/span 19}.panel-grid .g-col-20{grid-column:auto/span 20}.panel-grid .g-col-21{grid-column:auto/span 21}.panel-grid .g-col-22{grid-column:auto/span 22}.panel-grid .g-col-23{grid-column:auto/span 23}.panel-grid .g-col-24{grid-column:auto/span 24}.panel-grid .g-start-1{grid-column-start:1}.panel-grid .g-start-2{grid-column-start:2}.panel-grid .g-start-3{grid-column-start:3}.panel-grid .g-start-4{grid-column-start:4}.panel-grid .g-start-5{grid-column-start:5}.panel-grid .g-start-6{grid-column-start:6}.panel-grid .g-start-7{grid-column-start:7}.panel-grid .g-start-8{grid-column-start:8}.panel-grid .g-start-9{grid-column-start:9}.panel-grid .g-start-10{grid-column-start:10}.panel-grid .g-start-11{grid-column-start:11}.panel-grid .g-start-12{grid-column-start:12}.panel-grid .g-start-13{grid-column-start:13}.panel-grid .g-start-14{grid-column-start:14}.panel-grid .g-start-15{grid-column-start:15}.panel-grid .g-start-16{grid-column-start:16}.panel-grid .g-start-17{grid-column-start:17}.panel-grid .g-start-18{grid-column-start:18}.panel-grid .g-start-19{grid-column-start:19}.panel-grid .g-start-20{grid-column-start:20}.panel-grid .g-start-21{grid-column-start:21}.panel-grid .g-start-22{grid-column-start:22}.panel-grid .g-start-23{grid-column-start:23}@media(min-width: 576px){.panel-grid .g-col-sm-1{grid-column:auto/span 1}.panel-grid .g-col-sm-2{grid-column:auto/span 2}.panel-grid .g-col-sm-3{grid-column:auto/span 3}.panel-grid .g-col-sm-4{grid-column:auto/span 4}.panel-grid .g-col-sm-5{grid-column:auto/span 5}.panel-grid .g-col-sm-6{grid-column:auto/span 6}.panel-grid .g-col-sm-7{grid-column:auto/span 7}.panel-grid .g-col-sm-8{grid-column:auto/span 8}.panel-grid .g-col-sm-9{grid-column:auto/span 9}.panel-grid .g-col-sm-10{grid-column:auto/span 10}.panel-grid .g-col-sm-11{grid-column:auto/span 11}.panel-grid .g-col-sm-12{grid-column:auto/span 12}.panel-grid .g-col-sm-13{grid-column:auto/span 13}.panel-grid .g-col-sm-14{grid-column:auto/span 14}.panel-grid .g-col-sm-15{grid-column:auto/span 15}.panel-grid .g-col-sm-16{grid-column:auto/span 16}.panel-grid .g-col-sm-17{grid-column:auto/span 17}.panel-grid .g-col-sm-18{grid-column:auto/span 18}.panel-grid .g-col-sm-19{grid-column:auto/span 19}.panel-grid .g-col-sm-20{grid-column:auto/span 20}.panel-grid .g-col-sm-21{grid-column:auto/span 21}.panel-grid .g-col-sm-22{grid-column:auto/span 22}.panel-grid .g-col-sm-23{grid-column:auto/span 23}.panel-grid .g-col-sm-24{grid-column:auto/span 24}.panel-grid .g-start-sm-1{grid-column-start:1}.panel-grid .g-start-sm-2{grid-column-start:2}.panel-grid .g-start-sm-3{grid-column-start:3}.panel-grid .g-start-sm-4{grid-column-start:4}.panel-grid .g-start-sm-5{grid-column-start:5}.panel-grid .g-start-sm-6{grid-column-start:6}.panel-grid .g-start-sm-7{grid-column-start:7}.panel-grid .g-start-sm-8{grid-column-start:8}.panel-grid .g-start-sm-9{grid-column-start:9}.panel-grid .g-start-sm-10{grid-column-start:10}.panel-grid .g-start-sm-11{grid-column-start:11}.panel-grid .g-start-sm-12{grid-column-start:12}.panel-grid .g-start-sm-13{grid-column-start:13}.panel-grid .g-start-sm-14{grid-column-start:14}.panel-grid .g-start-sm-15{grid-column-start:15}.panel-grid .g-start-sm-16{grid-column-start:16}.panel-grid .g-start-sm-17{grid-column-start:17}.panel-grid .g-start-sm-18{grid-column-start:18}.panel-grid .g-start-sm-19{grid-column-start:19}.panel-grid .g-start-sm-20{grid-column-start:20}.panel-grid .g-start-sm-21{grid-column-start:21}.panel-grid .g-start-sm-22{grid-column-start:22}.panel-grid .g-start-sm-23{grid-column-start:23}}@media(min-width: 768px){.panel-grid .g-col-md-1{grid-column:auto/span 1}.panel-grid .g-col-md-2{grid-column:auto/span 2}.panel-grid .g-col-md-3{grid-column:auto/span 3}.panel-grid .g-col-md-4{grid-column:auto/span 4}.panel-grid .g-col-md-5{grid-column:auto/span 5}.panel-grid .g-col-md-6{grid-column:auto/span 6}.panel-grid .g-col-md-7{grid-column:auto/span 7}.panel-grid .g-col-md-8{grid-column:auto/span 8}.panel-grid .g-col-md-9{grid-column:auto/span 9}.panel-grid .g-col-md-10{grid-column:auto/span 10}.panel-grid .g-col-md-11{grid-column:auto/span 11}.panel-grid .g-col-md-12{grid-column:auto/span 12}.panel-grid .g-col-md-13{grid-column:auto/span 13}.panel-grid .g-col-md-14{grid-column:auto/span 14}.panel-grid .g-col-md-15{grid-column:auto/span 15}.panel-grid .g-col-md-16{grid-column:auto/span 16}.panel-grid .g-col-md-17{grid-column:auto/span 17}.panel-grid .g-col-md-18{grid-column:auto/span 18}.panel-grid .g-col-md-19{grid-column:auto/span 19}.panel-grid .g-col-md-20{grid-column:auto/span 20}.panel-grid .g-col-md-21{grid-column:auto/span 21}.panel-grid .g-col-md-22{grid-column:auto/span 22}.panel-grid .g-col-md-23{grid-column:auto/span 23}.panel-grid .g-col-md-24{grid-column:auto/span 24}.panel-grid .g-start-md-1{grid-column-start:1}.panel-grid .g-start-md-2{grid-column-start:2}.panel-grid .g-start-md-3{grid-column-start:3}.panel-grid .g-start-md-4{grid-column-start:4}.panel-grid .g-start-md-5{grid-column-start:5}.panel-grid .g-start-md-6{grid-column-start:6}.panel-grid .g-start-md-7{grid-column-start:7}.panel-grid .g-start-md-8{grid-column-start:8}.panel-grid .g-start-md-9{grid-column-start:9}.panel-grid .g-start-md-10{grid-column-start:10}.panel-grid .g-start-md-11{grid-column-start:11}.panel-grid .g-start-md-12{grid-column-start:12}.panel-grid .g-start-md-13{grid-column-start:13}.panel-grid .g-start-md-14{grid-column-start:14}.panel-grid .g-start-md-15{grid-column-start:15}.panel-grid .g-start-md-16{grid-column-start:16}.panel-grid .g-start-md-17{grid-column-start:17}.panel-grid .g-start-md-18{grid-column-start:18}.panel-grid .g-start-md-19{grid-column-start:19}.panel-grid .g-start-md-20{grid-column-start:20}.panel-grid .g-start-md-21{grid-column-start:21}.panel-grid .g-start-md-22{grid-column-start:22}.panel-grid .g-start-md-23{grid-column-start:23}}@media(min-width: 992px){.panel-grid .g-col-lg-1{grid-column:auto/span 1}.panel-grid .g-col-lg-2{grid-column:auto/span 2}.panel-grid .g-col-lg-3{grid-column:auto/span 3}.panel-grid .g-col-lg-4{grid-column:auto/span 4}.panel-grid .g-col-lg-5{grid-column:auto/span 5}.panel-grid .g-col-lg-6{grid-column:auto/span 6}.panel-grid .g-col-lg-7{grid-column:auto/span 7}.panel-grid .g-col-lg-8{grid-column:auto/span 8}.panel-grid .g-col-lg-9{grid-column:auto/span 9}.panel-grid .g-col-lg-10{grid-column:auto/span 10}.panel-grid .g-col-lg-11{grid-column:auto/span 11}.panel-grid .g-col-lg-12{grid-column:auto/span 12}.panel-grid .g-col-lg-13{grid-column:auto/span 13}.panel-grid .g-col-lg-14{grid-column:auto/span 14}.panel-grid .g-col-lg-15{grid-column:auto/span 15}.panel-grid .g-col-lg-16{grid-column:auto/span 16}.panel-grid .g-col-lg-17{grid-column:auto/span 17}.panel-grid .g-col-lg-18{grid-column:auto/span 18}.panel-grid .g-col-lg-19{grid-column:auto/span 19}.panel-grid .g-col-lg-20{grid-column:auto/span 20}.panel-grid .g-col-lg-21{grid-column:auto/span 21}.panel-grid .g-col-lg-22{grid-column:auto/span 22}.panel-grid .g-col-lg-23{grid-column:auto/span 23}.panel-grid .g-col-lg-24{grid-column:auto/span 24}.panel-grid .g-start-lg-1{grid-column-start:1}.panel-grid .g-start-lg-2{grid-column-start:2}.panel-grid .g-start-lg-3{grid-column-start:3}.panel-grid .g-start-lg-4{grid-column-start:4}.panel-grid .g-start-lg-5{grid-column-start:5}.panel-grid .g-start-lg-6{grid-column-start:6}.panel-grid .g-start-lg-7{grid-column-start:7}.panel-grid .g-start-lg-8{grid-column-start:8}.panel-grid .g-start-lg-9{grid-column-start:9}.panel-grid .g-start-lg-10{grid-column-start:10}.panel-grid .g-start-lg-11{grid-column-start:11}.panel-grid .g-start-lg-12{grid-column-start:12}.panel-grid .g-start-lg-13{grid-column-start:13}.panel-grid .g-start-lg-14{grid-column-start:14}.panel-grid .g-start-lg-15{grid-column-start:15}.panel-grid .g-start-lg-16{grid-column-start:16}.panel-grid .g-start-lg-17{grid-column-start:17}.panel-grid .g-start-lg-18{grid-column-start:18}.panel-grid .g-start-lg-19{grid-column-start:19}.panel-grid .g-start-lg-20{grid-column-start:20}.panel-grid .g-start-lg-21{grid-column-start:21}.panel-grid .g-start-lg-22{grid-column-start:22}.panel-grid .g-start-lg-23{grid-column-start:23}}@media(min-width: 1200px){.panel-grid .g-col-xl-1{grid-column:auto/span 1}.panel-grid .g-col-xl-2{grid-column:auto/span 2}.panel-grid .g-col-xl-3{grid-column:auto/span 3}.panel-grid .g-col-xl-4{grid-column:auto/span 4}.panel-grid .g-col-xl-5{grid-column:auto/span 5}.panel-grid .g-col-xl-6{grid-column:auto/span 6}.panel-grid .g-col-xl-7{grid-column:auto/span 7}.panel-grid .g-col-xl-8{grid-column:auto/span 8}.panel-grid .g-col-xl-9{grid-column:auto/span 9}.panel-grid .g-col-xl-10{grid-column:auto/span 10}.panel-grid .g-col-xl-11{grid-column:auto/span 11}.panel-grid .g-col-xl-12{grid-column:auto/span 12}.panel-grid .g-col-xl-13{grid-column:auto/span 13}.panel-grid .g-col-xl-14{grid-column:auto/span 14}.panel-grid .g-col-xl-15{grid-column:auto/span 15}.panel-grid .g-col-xl-16{grid-column:auto/span 16}.panel-grid .g-col-xl-17{grid-column:auto/span 17}.panel-grid .g-col-xl-18{grid-column:auto/span 18}.panel-grid .g-col-xl-19{grid-column:auto/span 19}.panel-grid .g-col-xl-20{grid-column:auto/span 20}.panel-grid .g-col-xl-21{grid-column:auto/span 21}.panel-grid .g-col-xl-22{grid-column:auto/span 22}.panel-grid .g-col-xl-23{grid-column:auto/span 23}.panel-grid .g-col-xl-24{grid-column:auto/span 24}.panel-grid .g-start-xl-1{grid-column-start:1}.panel-grid .g-start-xl-2{grid-column-start:2}.panel-grid .g-start-xl-3{grid-column-start:3}.panel-grid .g-start-xl-4{grid-column-start:4}.panel-grid .g-start-xl-5{grid-column-start:5}.panel-grid .g-start-xl-6{grid-column-start:6}.panel-grid .g-start-xl-7{grid-column-start:7}.panel-grid .g-start-xl-8{grid-column-start:8}.panel-grid .g-start-xl-9{grid-column-start:9}.panel-grid .g-start-xl-10{grid-column-start:10}.panel-grid .g-start-xl-11{grid-column-start:11}.panel-grid .g-start-xl-12{grid-column-start:12}.panel-grid .g-start-xl-13{grid-column-start:13}.panel-grid .g-start-xl-14{grid-column-start:14}.panel-grid .g-start-xl-15{grid-column-start:15}.panel-grid .g-start-xl-16{grid-column-start:16}.panel-grid .g-start-xl-17{grid-column-start:17}.panel-grid .g-start-xl-18{grid-column-start:18}.panel-grid .g-start-xl-19{grid-column-start:19}.panel-grid .g-start-xl-20{grid-column-start:20}.panel-grid .g-start-xl-21{grid-column-start:21}.panel-grid .g-start-xl-22{grid-column-start:22}.panel-grid .g-start-xl-23{grid-column-start:23}}@media(min-width: 1400px){.panel-grid .g-col-xxl-1{grid-column:auto/span 1}.panel-grid .g-col-xxl-2{grid-column:auto/span 2}.panel-grid .g-col-xxl-3{grid-column:auto/span 3}.panel-grid .g-col-xxl-4{grid-column:auto/span 4}.panel-grid .g-col-xxl-5{grid-column:auto/span 5}.panel-grid .g-col-xxl-6{grid-column:auto/span 6}.panel-grid .g-col-xxl-7{grid-column:auto/span 7}.panel-grid .g-col-xxl-8{grid-column:auto/span 8}.panel-grid .g-col-xxl-9{grid-column:auto/span 9}.panel-grid .g-col-xxl-10{grid-column:auto/span 10}.panel-grid .g-col-xxl-11{grid-column:auto/span 11}.panel-grid .g-col-xxl-12{grid-column:auto/span 12}.panel-grid .g-col-xxl-13{grid-column:auto/span 13}.panel-grid .g-col-xxl-14{grid-column:auto/span 14}.panel-grid .g-col-xxl-15{grid-column:auto/span 15}.panel-grid .g-col-xxl-16{grid-column:auto/span 16}.panel-grid .g-col-xxl-17{grid-column:auto/span 17}.panel-grid .g-col-xxl-18{grid-column:auto/span 18}.panel-grid .g-col-xxl-19{grid-column:auto/span 19}.panel-grid .g-col-xxl-20{grid-column:auto/span 20}.panel-grid .g-col-xxl-21{grid-column:auto/span 21}.panel-grid .g-col-xxl-22{grid-column:auto/span 22}.panel-grid .g-col-xxl-23{grid-column:auto/span 23}.panel-grid .g-col-xxl-24{grid-column:auto/span 24}.panel-grid .g-start-xxl-1{grid-column-start:1}.panel-grid .g-start-xxl-2{grid-column-start:2}.panel-grid .g-start-xxl-3{grid-column-start:3}.panel-grid .g-start-xxl-4{grid-column-start:4}.panel-grid .g-start-xxl-5{grid-column-start:5}.panel-grid .g-start-xxl-6{grid-column-start:6}.panel-grid .g-start-xxl-7{grid-column-start:7}.panel-grid .g-start-xxl-8{grid-column-start:8}.panel-grid .g-start-xxl-9{grid-column-start:9}.panel-grid .g-start-xxl-10{grid-column-start:10}.panel-grid .g-start-xxl-11{grid-column-start:11}.panel-grid .g-start-xxl-12{grid-column-start:12}.panel-grid .g-start-xxl-13{grid-column-start:13}.panel-grid .g-start-xxl-14{grid-column-start:14}.panel-grid .g-start-xxl-15{grid-column-start:15}.panel-grid .g-start-xxl-16{grid-column-start:16}.panel-grid .g-start-xxl-17{grid-column-start:17}.panel-grid .g-start-xxl-18{grid-column-start:18}.panel-grid .g-start-xxl-19{grid-column-start:19}.panel-grid .g-start-xxl-20{grid-column-start:20}.panel-grid .g-start-xxl-21{grid-column-start:21}.panel-grid .g-start-xxl-22{grid-column-start:22}.panel-grid .g-start-xxl-23{grid-column-start:23}}main{margin-top:1em;margin-bottom:1em}h1,.h1,h2,.h2{color:inherit;margin-top:2rem;margin-bottom:1rem;font-weight:600}h1.title,.title.h1{margin-top:0}main.content>section:first-of-type>h2:first-child,main.content>section:first-of-type>.h2:first-child{margin-top:0}h2,.h2{border-bottom:1px solid #dee2e6;padding-bottom:.5rem}h3,.h3{font-weight:600}h3,.h3,h4,.h4{opacity:.9;margin-top:1.5rem}h5,.h5,h6,.h6{opacity:.9}.header-section-number{color:#e6b978}.nav-link.active .header-section-number{color:inherit}mark,.mark{padding:0em}.panel-caption,.figure-caption,.subfigure-caption,.table-caption,figcaption,caption{font-size:.9rem;color:#e6b978}.quarto-layout-cell[data-ref-parent] caption{color:#e6b978}.column-margin figcaption,.margin-caption,div.aside,aside,.column-margin{color:#e6b978;font-size:.825rem}.panel-caption.margin-caption{text-align:inherit}.column-margin.column-container p{margin-bottom:0}.column-margin.column-container>*:not(.collapse):first-child{padding-bottom:.5em;display:block}.column-margin.column-container>*:not(.collapse):not(:first-child){padding-top:.5em;padding-bottom:.5em;display:block}.column-margin.column-container>*.collapse:not(.show){display:none}@media(min-width: 768px){.column-margin.column-container .callout-margin-content:first-child{margin-top:4.5em}.column-margin.column-container .callout-margin-content-simple:first-child{margin-top:3.5em}}.margin-caption>*{padding-top:.5em;padding-bottom:.5em}@media(max-width: 767.98px){.quarto-layout-row{flex-direction:column}}.nav-tabs .nav-item{margin-top:1px;cursor:pointer}.tab-content{margin-top:0px;border-left:#dee2e6 1px solid;border-right:#dee2e6 1px solid;border-bottom:#dee2e6 1px solid;margin-left:0;padding:1em;margin-bottom:1em}@media(max-width: 767.98px){.layout-sidebar{margin-left:0;margin-right:0}}.panel-sidebar,.panel-sidebar .form-control,.panel-input,.panel-input .form-control,.selectize-dropdown{font-size:.9rem}.panel-sidebar .form-control,.panel-input .form-control{padding-top:.1rem}.tab-pane div.sourceCode{margin-top:0px}.tab-pane>p{padding-top:0}.tab-pane>p:nth-child(1){padding-top:0}.tab-pane>p:last-child{margin-bottom:0}.tab-pane>pre:last-child{margin-bottom:0}.tab-content>.tab-pane:not(.active){display:none !important}div.sourceCode{background-color:rgba(233,236,239,.2);border:1px solid rgba(233,236,239,.2)}pre.sourceCode{background-color:rgba(0,0,0,0)}pre.sourceCode{border:none;font-size:.875em;overflow:visible !important;padding:.4em}div.sourceCode{overflow-y:hidden}.callout div.sourceCode{margin-left:initial}.blockquote{font-size:inherit;padding-left:1rem;padding-right:1.5rem;color:#e6b978}.blockquote h1:first-child,.blockquote .h1:first-child,.blockquote h2:first-child,.blockquote .h2:first-child,.blockquote h3:first-child,.blockquote .h3:first-child,.blockquote h4:first-child,.blockquote .h4:first-child,.blockquote h5:first-child,.blockquote .h5:first-child{margin-top:0}pre{background-color:initial;padding:initial;border:initial}p pre code:not(.sourceCode),li pre code:not(.sourceCode),pre code:not(.sourceCode){background-color:initial}p code:not(.sourceCode),li code:not(.sourceCode),td code:not(.sourceCode){background-color:#f8f9fa;padding:.2em}nav p code:not(.sourceCode),nav li code:not(.sourceCode),nav td code:not(.sourceCode){background-color:rgba(0,0,0,0);padding:0}td code:not(.sourceCode){white-space:pre-wrap}#quarto-embedded-source-code-modal>.modal-dialog{max-width:1000px;padding-left:1.75rem;padding-right:1.75rem}#quarto-embedded-source-code-modal>.modal-dialog>.modal-content>.modal-body{padding:0}#quarto-embedded-source-code-modal>.modal-dialog>.modal-content>.modal-body div.sourceCode{margin:0;padding:.2rem .2rem;border-radius:0px;border:none}#quarto-embedded-source-code-modal>.modal-dialog>.modal-content>.modal-header{padding:.7rem}.code-tools-button{font-size:1rem;padding:.15rem .15rem;margin-left:5px;color:#6c757d;background-color:rgba(0,0,0,0);transition:initial;cursor:pointer}.code-tools-button>.bi::before{display:inline-block;height:1rem;width:1rem;content:"";vertical-align:-0.125em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>');background-repeat:no-repeat;background-size:1rem 1rem}.code-tools-button:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(250, 241, 228)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>')}#quarto-embedded-source-code-modal .code-copy-button>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" viewBox="0 0 16 16"><path d="M4 1.5H3a2 2 0 0 0-2 2V14a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V3.5a2 2 0 0 0-2-2h-1v1h1a1 1 0 0 1 1 1V14a1 1 0 0 1-1 1H3a1 1 0 0 1-1-1V3.5a1 1 0 0 1 1-1h1v-1z"/><path d="M9.5 1a.5.5 0 0 1 .5.5v1a.5.5 0 0 1-.5.5h-3a.5.5 0 0 1-.5-.5v-1a.5.5 0 0 1 .5-.5h3zm-3-1A1.5 1.5 0 0 0 5 1.5v1A1.5 1.5 0 0 0 6.5 4h3A1.5 1.5 0 0 0 11 2.5v-1A1.5 1.5 0 0 0 9.5 0h-3z"/></svg>')}#quarto-embedded-source-code-modal .code-copy-button-checked>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" viewBox="0 0 16 16"><path d="M13.854 3.646a.5.5 0 0 1 0 .708l-7 7a.5.5 0 0 1-.708 0l-3.5-3.5a.5.5 0 1 1 .708-.708L6.5 10.293l6.646-6.647a.5.5 0 0 1 .708 0z"/></svg>')}.sidebar{will-change:top;transition:top 200ms linear;position:sticky;overflow-y:auto;padding-top:1.2em;max-height:100vh}.sidebar.toc-left,.sidebar.margin-sidebar{top:0px;padding-top:1em}.sidebar.quarto-banner-title-block-sidebar>*{padding-top:1.65em}figure .quarto-notebook-link{margin-top:.5em}.quarto-notebook-link{font-size:.75em;color:#6c757d;margin-bottom:1em;text-decoration:none;display:block}.quarto-notebook-link:hover{text-decoration:underline;color:#aed1e4}.quarto-notebook-link::before{display:inline-block;height:.75rem;width:.75rem;margin-bottom:0em;margin-right:.25em;content:"";vertical-align:-0.125em;background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(108, 117, 125)" class="bi bi-journal-code" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M8.646 5.646a.5.5 0 0 1 .708 0l2 2a.5.5 0 0 1 0 .708l-2 2a.5.5 0 0 1-.708-.708L10.293 8 8.646 6.354a.5.5 0 0 1 0-.708zm-1.292 0a.5.5 0 0 0-.708 0l-2 2a.5.5 0 0 0 0 .708l2 2a.5.5 0 0 0 .708-.708L5.707 8l1.647-1.646a.5.5 0 0 0 0-.708z"/><path d="M3 0h10a2 2 0 0 1 2 2v12a2 2 0 0 1-2 2H3a2 2 0 0 1-2-2v-1h1v1a1 1 0 0 0 1 1h10a1 1 0 0 0 1-1V2a1 1 0 0 0-1-1H3a1 1 0 0 0-1 1v1H1V2a2 2 0 0 1 2-2z"/><path d="M1 5v-.5a.5.5 0 0 1 1 0V5h.5a.5.5 0 0 1 0 1h-2a.5.5 0 0 1 0-1H1zm0 3v-.5a.5.5 0 0 1 1 0V8h.5a.5.5 0 0 1 0 1h-2a.5.5 0 0 1 0-1H1zm0 3v-.5a.5.5 0 0 1 1 0v.5h.5a.5.5 0 0 1 0 1h-2a.5.5 0 0 1 0-1H1z"/></svg>');background-repeat:no-repeat;background-size:.75rem .75rem}.toc-actions i.bi,.quarto-code-links i.bi,.quarto-other-links i.bi,.quarto-alternate-notebooks i.bi,.quarto-alternate-formats i.bi{margin-right:.4em;font-size:.8rem}.quarto-other-links-text-target .quarto-code-links i.bi,.quarto-other-links-text-target .quarto-other-links i.bi{margin-right:.2em}.quarto-other-formats-text-target .quarto-alternate-formats i.bi{margin-right:.1em}.toc-actions i.bi.empty,.quarto-code-links i.bi.empty,.quarto-other-links i.bi.empty,.quarto-alternate-notebooks i.bi.empty,.quarto-alternate-formats i.bi.empty{padding-left:1em}.quarto-notebook h2,.quarto-notebook .h2{border-bottom:none}.quarto-notebook .cell-container{display:flex}.quarto-notebook .cell-container .cell{flex-grow:4}.quarto-notebook .cell-container .cell-decorator{padding-top:1.5em;padding-right:1em;text-align:right}.quarto-notebook .cell-container.code-fold .cell-decorator{padding-top:3em}.quarto-notebook .cell-code code{white-space:pre-wrap}.quarto-notebook .cell .cell-output-stderr pre code,.quarto-notebook .cell .cell-output-stdout pre code{white-space:pre-wrap;overflow-wrap:anywhere}.toc-actions,.quarto-alternate-formats,.quarto-other-links,.quarto-code-links,.quarto-alternate-notebooks{padding-left:0em}.sidebar .toc-actions a,.sidebar .quarto-alternate-formats a,.sidebar .quarto-other-links a,.sidebar .quarto-code-links a,.sidebar .quarto-alternate-notebooks a,.sidebar nav[role=doc-toc] a{text-decoration:none}.sidebar .toc-actions a:hover,.sidebar .quarto-other-links a:hover,.sidebar .quarto-code-links a:hover,.sidebar .quarto-alternate-formats a:hover,.sidebar .quarto-alternate-notebooks a:hover{color:#aed1e4}.sidebar .toc-actions h2,.sidebar .toc-actions .h2,.sidebar .quarto-code-links h2,.sidebar .quarto-code-links .h2,.sidebar .quarto-other-links h2,.sidebar .quarto-other-links .h2,.sidebar .quarto-alternate-notebooks h2,.sidebar .quarto-alternate-notebooks .h2,.sidebar .quarto-alternate-formats h2,.sidebar .quarto-alternate-formats .h2,.sidebar nav[role=doc-toc]>h2,.sidebar nav[role=doc-toc]>.h2{font-weight:500;margin-bottom:.2rem;margin-top:.3rem;font-family:inherit;border-bottom:0;padding-bottom:0;padding-top:0px}.sidebar .toc-actions>h2,.sidebar .toc-actions>.h2,.sidebar .quarto-code-links>h2,.sidebar .quarto-code-links>.h2,.sidebar .quarto-other-links>h2,.sidebar .quarto-other-links>.h2,.sidebar .quarto-alternate-notebooks>h2,.sidebar .quarto-alternate-notebooks>.h2,.sidebar .quarto-alternate-formats>h2,.sidebar .quarto-alternate-formats>.h2{font-size:.8rem}.sidebar nav[role=doc-toc]>h2,.sidebar nav[role=doc-toc]>.h2{font-size:.875rem}.sidebar nav[role=doc-toc]>ul a{border-left:1px solid #e9ecef;padding-left:.6rem}.sidebar .toc-actions h2>ul a,.sidebar .toc-actions .h2>ul a,.sidebar .quarto-code-links h2>ul a,.sidebar .quarto-code-links .h2>ul a,.sidebar .quarto-other-links h2>ul a,.sidebar .quarto-other-links .h2>ul a,.sidebar .quarto-alternate-notebooks h2>ul a,.sidebar .quarto-alternate-notebooks .h2>ul a,.sidebar .quarto-alternate-formats h2>ul a,.sidebar .quarto-alternate-formats .h2>ul a{border-left:none;padding-left:.6rem}.sidebar .toc-actions ul a:empty,.sidebar .quarto-code-links ul a:empty,.sidebar .quarto-other-links ul a:empty,.sidebar .quarto-alternate-notebooks ul a:empty,.sidebar .quarto-alternate-formats ul a:empty,.sidebar nav[role=doc-toc]>ul a:empty{display:none}.sidebar .toc-actions ul,.sidebar .quarto-code-links ul,.sidebar .quarto-other-links ul,.sidebar .quarto-alternate-notebooks ul,.sidebar .quarto-alternate-formats ul{padding-left:0;list-style:none}.sidebar nav[role=doc-toc] ul{list-style:none;padding-left:0;list-style:none}.sidebar nav[role=doc-toc]>ul{margin-left:.45em}.quarto-margin-sidebar nav[role=doc-toc]{padding-left:.5em}.sidebar .toc-actions>ul,.sidebar .quarto-code-links>ul,.sidebar .quarto-other-links>ul,.sidebar .quarto-alternate-notebooks>ul,.sidebar .quarto-alternate-formats>ul{font-size:.8rem}.sidebar nav[role=doc-toc]>ul{font-size:.875rem}.sidebar .toc-actions ul li a,.sidebar .quarto-code-links ul li a,.sidebar .quarto-other-links ul li a,.sidebar .quarto-alternate-notebooks ul li a,.sidebar .quarto-alternate-formats ul li a,.sidebar nav[role=doc-toc]>ul li a{line-height:1.1rem;padding-bottom:.2rem;padding-top:.2rem;color:inherit}.sidebar nav[role=doc-toc] ul>li>ul>li>a{padding-left:1.2em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>a{padding-left:2.4em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>ul>li>a{padding-left:3.6em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>ul>li>ul>li>a{padding-left:4.8em}.sidebar nav[role=doc-toc] ul>li>ul>li>ul>li>ul>li>ul>li>ul>li>a{padding-left:6em}.sidebar nav[role=doc-toc] ul>li>a.active,.sidebar nav[role=doc-toc] ul>li>ul>li>a.active{border-left:1px solid #aed1e4;color:#aed1e4 !important}.sidebar nav[role=doc-toc] ul>li>a:hover,.sidebar nav[role=doc-toc] ul>li>ul>li>a:hover{color:#aed1e4 !important}kbd,.kbd{color:#faf1e4;background-color:#4a4b4b;border:1px solid;border-radius:5px;border-color:#dee2e6}.quarto-appendix-contents div.hanging-indent{margin-left:0em}.quarto-appendix-contents div.hanging-indent div.csl-entry{margin-left:1em;text-indent:-1em}.citation a,.footnote-ref{text-decoration:none}.footnotes ol{padding-left:1em}.tippy-content>*{margin-bottom:.7em}.tippy-content>*:last-child{margin-bottom:0}.callout{margin-top:1.25rem;margin-bottom:1.25rem;border-radius:.25rem;overflow-wrap:break-word}.callout .callout-title-container{overflow-wrap:anywhere}.callout.callout-style-simple{padding:.4em .7em;border-left:5px solid;border-right:1px solid #dee2e6;border-top:1px solid #dee2e6;border-bottom:1px solid #dee2e6}.callout.callout-style-default{border-left:5px solid;border-right:1px solid #dee2e6;border-top:1px solid #dee2e6;border-bottom:1px solid #dee2e6}.callout .callout-body-container{flex-grow:1}.callout.callout-style-simple .callout-body{font-size:.9rem;font-weight:400}.callout.callout-style-default .callout-body{font-size:.9rem;font-weight:400}.callout:not(.no-icon).callout-titled.callout-style-simple .callout-body{padding-left:1.6em}.callout.callout-titled>.callout-header{padding-top:.2em;margin-bottom:-0.2em}.callout.callout-style-simple>div.callout-header{border-bottom:none;font-size:.9rem;font-weight:600;opacity:75%}.callout.callout-style-default>div.callout-header{border-bottom:none;font-weight:600;opacity:85%;font-size:.9rem;padding-left:.5em;padding-right:.5em}.callout.callout-style-default .callout-body{padding-left:.5em;padding-right:.5em}.callout.callout-style-default .callout-body>:first-child{padding-top:.5rem;margin-top:0}.callout>div.callout-header[data-bs-toggle=collapse]{cursor:pointer}.callout.callout-style-default .callout-header[aria-expanded=false],.callout.callout-style-default .callout-header[aria-expanded=true]{padding-top:0px;margin-bottom:0px;align-items:center}.callout.callout-titled .callout-body>:last-child:not(.sourceCode),.callout.callout-titled .callout-body>div>:last-child:not(.sourceCode){padding-bottom:.5rem;margin-bottom:0}.callout:not(.callout-titled) .callout-body>:first-child,.callout:not(.callout-titled) .callout-body>div>:first-child{margin-top:.25rem}.callout:not(.callout-titled) .callout-body>:last-child,.callout:not(.callout-titled) .callout-body>div>:last-child{margin-bottom:.2rem}.callout.callout-style-simple .callout-icon::before,.callout.callout-style-simple .callout-toggle::before{height:1rem;width:1rem;display:inline-block;content:"";background-repeat:no-repeat;background-size:1rem 1rem}.callout.callout-style-default .callout-icon::before,.callout.callout-style-default .callout-toggle::before{height:.9rem;width:.9rem;display:inline-block;content:"";background-repeat:no-repeat;background-size:.9rem .9rem}.callout.callout-style-default .callout-toggle::before{margin-top:5px}.callout .callout-btn-toggle .callout-toggle::before{transition:transform .2s linear}.callout .callout-header[aria-expanded=false] .callout-toggle::before{transform:rotate(-90deg)}.callout .callout-header[aria-expanded=true] .callout-toggle::before{transform:none}.callout.callout-style-simple:not(.no-icon) div.callout-icon-container{padding-top:.2em;padding-right:.55em}.callout.callout-style-default:not(.no-icon) div.callout-icon-container{padding-top:.1em;padding-right:.35em}.callout.callout-style-default:not(.no-icon) div.callout-title-container{margin-top:-1px}.callout.callout-style-default.callout-caution:not(.no-icon) div.callout-icon-container{padding-top:.3em;padding-right:.35em}.callout>.callout-body>.callout-icon-container>.no-icon,.callout>.callout-header>.callout-icon-container>.no-icon{display:none}div.callout.callout{border-left-color:#6c757d}div.callout.callout-style-default>.callout-header{background-color:#6c757d}div.callout-note.callout{border-left-color:#2780e3}div.callout-note.callout-style-default>.callout-header{background-color:#0c2644}div.callout-note:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %232373cc" class="bi bi-info-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="m8.93 6.588-2.29.287-.082.38.45.083c.294.07.352.176.288.469l-.738 3.468c-.194.897.105 1.319.808 1.319.545 0 1.178-.252 1.465-.598l.088-.416c-.2.176-.492.246-.686.246-.275 0-.375-.193-.304-.533L8.93 6.588zM9 4.5a1 1 0 1 1-2 0 1 1 0 0 1 2 0z"/></svg>');}div.callout-note.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %232373cc" class="bi bi-info-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="m8.93 6.588-2.29.287-.082.38.45.083c.294.07.352.176.288.469l-.738 3.468c-.194.897.105 1.319.808 1.319.545 0 1.178-.252 1.465-.598l.088-.416c-.2.176-.492.246-.686.246-.275 0-.375-.193-.304-.533L8.93 6.588zM9 4.5a1 1 0 1 1-2 0 1 1 0 0 1 2 0z"/></svg>');}div.callout-note .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(250, 241, 228)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-tip.callout{border-left-color:#3fb618}div.callout-tip.callout-style-default>.callout-header{background-color:#133707}div.callout-tip:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %2339a416" class="bi bi-lightbulb" viewBox="0 0 16 16"><path d="M2 6a6 6 0 1 1 10.174 4.31c-.203.196-.359.4-.453.619l-.762 1.769A.5.5 0 0 1 10.5 13a.5.5 0 0 1 0 1 .5.5 0 0 1 0 1l-.224.447a1 1 0 0 1-.894.553H6.618a1 1 0 0 1-.894-.553L5.5 15a.5.5 0 0 1 0-1 .5.5 0 0 1 0-1 .5.5 0 0 1-.46-.302l-.761-1.77a1.964 1.964 0 0 0-.453-.618A5.984 5.984 0 0 1 2 6zm6-5a5 5 0 0 0-3.479 8.592c.263.254.514.564.676.941L5.83 12h4.342l.632-1.467c.162-.377.413-.687.676-.941A5 5 0 0 0 8 1z"/></svg>');}div.callout-tip.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %2339a416" class="bi bi-lightbulb" viewBox="0 0 16 16"><path d="M2 6a6 6 0 1 1 10.174 4.31c-.203.196-.359.4-.453.619l-.762 1.769A.5.5 0 0 1 10.5 13a.5.5 0 0 1 0 1 .5.5 0 0 1 0 1l-.224.447a1 1 0 0 1-.894.553H6.618a1 1 0 0 1-.894-.553L5.5 15a.5.5 0 0 1 0-1 .5.5 0 0 1 0-1 .5.5 0 0 1-.46-.302l-.761-1.77a1.964 1.964 0 0 0-.453-.618A5.984 5.984 0 0 1 2 6zm6-5a5 5 0 0 0-3.479 8.592c.263.254.514.564.676.941L5.83 12h4.342l.632-1.467c.162-.377.413-.687.676-.941A5 5 0 0 0 8 1z"/></svg>');}div.callout-tip .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(250, 241, 228)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-warning.callout{border-left-color:#ff7518}div.callout-warning.callout-style-default>.callout-header{background-color:#4d2307}div.callout-warning:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e66916" class="bi bi-exclamation-triangle" viewBox="0 0 16 16"><path d="M7.938 2.016A.13.13 0 0 1 8.002 2a.13.13 0 0 1 .063.016.146.146 0 0 1 .054.057l6.857 11.667c.036.06.035.124.002.183a.163.163 0 0 1-.054.06.116.116 0 0 1-.066.017H1.146a.115.115 0 0 1-.066-.017.163.163 0 0 1-.054-.06.176.176 0 0 1 .002-.183L7.884 2.073a.147.147 0 0 1 .054-.057zm1.044-.45a1.13 1.13 0 0 0-1.96 0L.165 13.233c-.457.778.091 1.767.98 1.767h13.713c.889 0 1.438-.99.98-1.767L8.982 1.566z"/><path d="M7.002 12a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 5.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 5.995z"/></svg>');}div.callout-warning.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e66916" class="bi bi-exclamation-triangle" viewBox="0 0 16 16"><path d="M7.938 2.016A.13.13 0 0 1 8.002 2a.13.13 0 0 1 .063.016.146.146 0 0 1 .054.057l6.857 11.667c.036.06.035.124.002.183a.163.163 0 0 1-.054.06.116.116 0 0 1-.066.017H1.146a.115.115 0 0 1-.066-.017.163.163 0 0 1-.054-.06.176.176 0 0 1 .002-.183L7.884 2.073a.147.147 0 0 1 .054-.057zm1.044-.45a1.13 1.13 0 0 0-1.96 0L.165 13.233c-.457.778.091 1.767.98 1.767h13.713c.889 0 1.438-.99.98-1.767L8.982 1.566z"/><path d="M7.002 12a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 5.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 5.995z"/></svg>');}div.callout-warning .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(250, 241, 228)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-caution.callout{border-left-color:#f0ad4e}div.callout-caution.callout-style-default>.callout-header{background-color:#483417}div.callout-caution:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23d89c46" class="bi bi-cone-striped" viewBox="0 0 16 16"><path d="M9.97 4.88l.953 3.811C10.158 8.878 9.14 9 8 9c-1.14 0-2.159-.122-2.923-.309L6.03 4.88C6.635 4.957 7.3 5 8 5s1.365-.043 1.97-.12zm-.245-.978L8.97.88C8.718-.13 7.282-.13 7.03.88L6.274 3.9C6.8 3.965 7.382 4 8 4c.618 0 1.2-.036 1.725-.098zm4.396 8.613a.5.5 0 0 1 .037.96l-6 2a.5.5 0 0 1-.316 0l-6-2a.5.5 0 0 1 .037-.96l2.391-.598.565-2.257c.862.212 1.964.339 3.165.339s2.303-.127 3.165-.339l.565 2.257 2.391.598z"/></svg>');}div.callout-caution.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23d89c46" class="bi bi-cone-striped" viewBox="0 0 16 16"><path d="M9.97 4.88l.953 3.811C10.158 8.878 9.14 9 8 9c-1.14 0-2.159-.122-2.923-.309L6.03 4.88C6.635 4.957 7.3 5 8 5s1.365-.043 1.97-.12zm-.245-.978L8.97.88C8.718-.13 7.282-.13 7.03.88L6.274 3.9C6.8 3.965 7.382 4 8 4c.618 0 1.2-.036 1.725-.098zm4.396 8.613a.5.5 0 0 1 .037.96l-6 2a.5.5 0 0 1-.316 0l-6-2a.5.5 0 0 1 .037-.96l2.391-.598.565-2.257c.862.212 1.964.339 3.165.339s2.303-.127 3.165-.339l.565 2.257 2.391.598z"/></svg>');}div.callout-caution .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(250, 241, 228)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}div.callout-important.callout{border-left-color:#ff0039}div.callout-important.callout-style-default>.callout-header{background-color:#4d0011}div.callout-important:not(.callout-titled) .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e60033" class="bi bi-exclamation-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="M7.002 11a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 4.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 4.995z"/></svg>');}div.callout-important.callout-titled .callout-icon::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" style="fill: %23e60033" class="bi bi-exclamation-circle" viewBox="0 0 16 16"><path d="M8 15A7 7 0 1 1 8 1a7 7 0 0 1 0 14zm0 1A8 8 0 1 0 8 0a8 8 0 0 0 0 16z"/><path d="M7.002 11a1 1 0 1 1 2 0 1 1 0 0 1-2 0zM7.1 4.995a.905.905 0 1 1 1.8 0l-.35 3.507a.552.552 0 0 1-1.1 0L7.1 4.995z"/></svg>');}div.callout-important .callout-toggle::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(250, 241, 228)" class="bi bi-chevron-down" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M1.646 4.646a.5.5 0 0 1 .708 0L8 10.293l5.646-5.647a.5.5 0 0 1 .708.708l-6 6a.5.5 0 0 1-.708 0l-6-6a.5.5 0 0 1 0-.708z"/></svg>')}.quarto-toggle-container{display:flex;align-items:center}.quarto-reader-toggle .bi::before,.quarto-color-scheme-toggle .bi::before{display:inline-block;height:1rem;width:1rem;content:"";background-repeat:no-repeat;background-size:1rem 1rem}.sidebar-navigation{padding-left:20px}.navbar{background-color:#303030;color:#bdbdbd}.navbar .quarto-color-scheme-toggle:not(.alternate) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(189, 189, 189, 1)" class="bi bi-toggle-off" viewBox="0 0 16 16"><path d="M11 4a4 4 0 0 1 0 8H8a4.992 4.992 0 0 0 2-4 4.992 4.992 0 0 0-2-4h3zm-6 8a4 4 0 1 1 0-8 4 4 0 0 1 0 8zM0 8a5 5 0 0 0 5 5h6a5 5 0 0 0 0-10H5a5 5 0 0 0-5 5z"/></svg>')}.navbar .quarto-color-scheme-toggle.alternate .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(189, 189, 189, 1)" class="bi bi-toggle-on" viewBox="0 0 16 16"><path d="M5 3a5 5 0 0 0 0 10h6a5 5 0 0 0 0-10H5zm6 9a4 4 0 1 1 0-8 4 4 0 0 1 0 8z"/></svg>')}.sidebar-navigation .quarto-color-scheme-toggle:not(.alternate) .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(242, 219, 185, 1)" class="bi bi-toggle-off" viewBox="0 0 16 16"><path d="M11 4a4 4 0 0 1 0 8H8a4.992 4.992 0 0 0 2-4 4.992 4.992 0 0 0-2-4h3zm-6 8a4 4 0 1 1 0-8 4 4 0 0 1 0 8zM0 8a5 5 0 0 0 5 5h6a5 5 0 0 0 0-10H5a5 5 0 0 0-5 5z"/></svg>')}.sidebar-navigation .quarto-color-scheme-toggle.alternate .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(242, 219, 185, 1)" class="bi bi-toggle-on" viewBox="0 0 16 16"><path d="M5 3a5 5 0 0 0 0 10h6a5 5 0 0 0 0-10H5zm6 9a4 4 0 1 1 0-8 4 4 0 0 1 0 8z"/></svg>')}.quarto-sidebar-toggle{border-color:#dee2e6;border-bottom-left-radius:.25rem;border-bottom-right-radius:.25rem;border-style:solid;border-width:1px;overflow:hidden;border-top-width:0px;padding-top:0px !important}.quarto-sidebar-toggle-title{cursor:pointer;padding-bottom:2px;margin-left:.25em;text-align:center;font-weight:400;font-size:.775em}#quarto-content .quarto-sidebar-toggle{background:#1d1d1d}#quarto-content .quarto-sidebar-toggle-title{color:#faf1e4}.quarto-sidebar-toggle-icon{color:#dee2e6;margin-right:.5em;float:right;transition:transform .2s ease}.quarto-sidebar-toggle-icon::before{padding-top:5px}.quarto-sidebar-toggle.expanded .quarto-sidebar-toggle-icon{transform:rotate(-180deg)}.quarto-sidebar-toggle.expanded .quarto-sidebar-toggle-title{border-bottom:solid #dee2e6 1px}.quarto-sidebar-toggle-contents{background-color:#181818;padding-right:10px;padding-left:10px;margin-top:0px !important;transition:max-height .5s ease}.quarto-sidebar-toggle.expanded .quarto-sidebar-toggle-contents{padding-top:1em;padding-bottom:10px}@media(max-width: 767.98px){.sidebar-menu-container{padding-bottom:5em}}.quarto-sidebar-toggle:not(.expanded) .quarto-sidebar-toggle-contents{padding-top:0px !important;padding-bottom:0px}nav[role=doc-toc]{z-index:1020}#quarto-sidebar>*,nav[role=doc-toc]>*{transition:opacity .1s ease,border .1s ease}#quarto-sidebar.slow>*,nav[role=doc-toc].slow>*{transition:opacity .4s ease,border .4s ease}.quarto-color-scheme-toggle:not(.alternate).top-right .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(222, 163, 77, 1)" class="bi bi-toggle-off" viewBox="0 0 16 16"><path d="M11 4a4 4 0 0 1 0 8H8a4.992 4.992 0 0 0 2-4 4.992 4.992 0 0 0-2-4h3zm-6 8a4 4 0 1 1 0-8 4 4 0 0 1 0 8zM0 8a5 5 0 0 0 5 5h6a5 5 0 0 0 0-10H5a5 5 0 0 0-5 5z"/></svg>')}.quarto-color-scheme-toggle.alternate.top-right .bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgba(234, 196, 142, 1)" class="bi bi-toggle-on" viewBox="0 0 16 16"><path d="M5 3a5 5 0 0 0 0 10h6a5 5 0 0 0 0-10H5zm6 9a4 4 0 1 1 0-8 4 4 0 0 1 0 8z"/></svg>')}#quarto-appendix.default{border-top:1px solid #dee2e6}#quarto-appendix.default{background-color:#181818;padding-top:1.5em;margin-top:2em;z-index:998}#quarto-appendix.default .quarto-appendix-heading{margin-top:0;line-height:1.4em;font-weight:600;opacity:.9;border-bottom:none;margin-bottom:0}#quarto-appendix.default .footnotes ol,#quarto-appendix.default .footnotes ol li>p:last-of-type,#quarto-appendix.default .quarto-appendix-contents>p:last-of-type{margin-bottom:0}#quarto-appendix.default .footnotes ol{margin-left:.5em}#quarto-appendix.default .quarto-appendix-secondary-label{margin-bottom:.4em}#quarto-appendix.default .quarto-appendix-bibtex{font-size:.7em;padding:1em;border:solid 1px #dee2e6;margin-bottom:1em}#quarto-appendix.default .quarto-appendix-bibtex code.sourceCode{white-space:pre-wrap}#quarto-appendix.default .quarto-appendix-citeas{font-size:.9em;padding:1em;border:solid 1px #dee2e6;margin-bottom:1em}#quarto-appendix.default .quarto-appendix-heading{font-size:1em !important}#quarto-appendix.default *[role=doc-endnotes]>ol,#quarto-appendix.default .quarto-appendix-contents>*:not(h2):not(.h2){font-size:.9em}#quarto-appendix.default section{padding-bottom:1.5em}#quarto-appendix.default section *[role=doc-endnotes],#quarto-appendix.default section>*:not(a){opacity:.9;word-wrap:break-word}.btn.btn-quarto,div.cell-output-display .btn-quarto{--bs-btn-color: #cacccd;--bs-btn-bg: #343a40;--bs-btn-border-color: #343a40;--bs-btn-hover-color: #cacccd;--bs-btn-hover-bg: #52585d;--bs-btn-hover-border-color: #484e53;--bs-btn-focus-shadow-rgb: 75, 80, 85;--bs-btn-active-color: #fff;--bs-btn-active-bg: #5d6166;--bs-btn-active-border-color: #484e53;--bs-btn-active-shadow: inset 0 3px 5px rgba(0, 0, 0, 0.125);--bs-btn-disabled-color: #fff;--bs-btn-disabled-bg: #343a40;--bs-btn-disabled-border-color: #343a40}nav.quarto-secondary-nav.color-navbar{background-color:#303030;color:#bdbdbd}nav.quarto-secondary-nav.color-navbar h1,nav.quarto-secondary-nav.color-navbar .h1,nav.quarto-secondary-nav.color-navbar .quarto-btn-toggle{color:#bdbdbd}@media(max-width: 991.98px){body.nav-sidebar .quarto-title-banner{margin-bottom:0;padding-bottom:1em}body.nav-sidebar #title-block-header{margin-block-end:0}}p.subtitle{margin-top:.25em;margin-bottom:.5em}code a:any-link{color:inherit;text-decoration-color:#6c757d}/*! dark */div.observablehq table thead tr th{background-color:var(--bs-body-bg)}input,button,select,optgroup,textarea{background-color:var(--bs-body-bg)}.code-annotated .code-copy-button{margin-right:1.25em;margin-top:0;padding-bottom:0;padding-top:3px}.code-annotation-gutter-bg{background-color:#181818}.code-annotation-gutter{background-color:rgba(233,236,239,.2)}.code-annotation-gutter,.code-annotation-gutter-bg{height:100%;width:calc(20px + .5em);position:absolute;top:0;right:0}dl.code-annotation-container-grid dt{margin-right:1em;margin-top:.25rem}dl.code-annotation-container-grid dt{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;color:#f2dbb9;border:solid #f2dbb9 1px;border-radius:50%;height:22px;width:22px;line-height:22px;font-size:11px;text-align:center;vertical-align:middle;text-decoration:none}dl.code-annotation-container-grid dt[data-target-cell]{cursor:pointer}dl.code-annotation-container-grid dt[data-target-cell].code-annotation-active{color:#181818;border:solid #aaa 1px;background-color:#aaa}pre.code-annotation-code{padding-top:0;padding-bottom:0}pre.code-annotation-code code{z-index:3}#code-annotation-line-highlight-gutter{width:100%;border-top:solid rgba(170,170,170,.2666666667) 1px;border-bottom:solid rgba(170,170,170,.2666666667) 1px;z-index:2;background-color:rgba(170,170,170,.1333333333)}#code-annotation-line-highlight{margin-left:-4em;width:calc(100% + 4em);border-top:solid rgba(170,170,170,.2666666667) 1px;border-bottom:solid rgba(170,170,170,.2666666667) 1px;z-index:2;background-color:rgba(170,170,170,.1333333333)}code.sourceCode .code-annotation-anchor.code-annotation-active{background-color:var(--quarto-hl-normal-color, #aaaaaa);border:solid var(--quarto-hl-normal-color, #aaaaaa) 1px;color:#e9ecef;font-weight:bolder}code.sourceCode .code-annotation-anchor{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;color:var(--quarto-hl-co-color);border:solid var(--quarto-hl-co-color) 1px;border-radius:50%;height:18px;width:18px;font-size:9px;margin-top:2px}code.sourceCode button.code-annotation-anchor{padding:2px;user-select:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;-o-user-select:none}code.sourceCode a.code-annotation-anchor{line-height:18px;text-align:center;vertical-align:middle;cursor:default;text-decoration:none}@media print{.page-columns .column-screen-inset{grid-column:page-start-inset/page-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset table{background:#181818}.page-columns .column-screen-inset-left{grid-column:page-start-inset/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-inset-left table{background:#181818}.page-columns .column-screen-inset-right{grid-column:body-content-start/page-end-inset;z-index:998;opacity:.999}.page-columns .column-screen-inset-right table{background:#181818}.page-columns .column-screen{grid-column:page-start/page-end;z-index:998;opacity:.999}.page-columns .column-screen table{background:#181818}.page-columns .column-screen-left{grid-column:page-start/body-content-end;z-index:998;opacity:.999}.page-columns .column-screen-left table{background:#181818}.page-columns .column-screen-right{grid-column:body-content-start/page-end;z-index:998;opacity:.999}.page-columns .column-screen-right table{background:#181818}.page-columns .column-screen-inset-shaded{grid-column:page-start-inset/page-end-inset;padding:1em;background:#525252;z-index:998;opacity:.999;margin-bottom:1em}}.quarto-video{margin-bottom:1em}.table{border-top:1px solid #454341;border-bottom:1px solid #454341}.table>thead{border-top-width:0;border-bottom:1px solid #89857e}.table a{word-break:break-word}.table>:not(caption)>*>*{background-color:unset;color:unset}#quarto-document-content .crosstalk-input .checkbox input[type=checkbox],#quarto-document-content .crosstalk-input .checkbox-inline input[type=checkbox]{position:unset;margin-top:unset;margin-left:unset}#quarto-document-content .row{margin-left:unset;margin-right:unset}.quarto-xref{white-space:nowrap}#quarto-draft-alert{margin-top:0px;margin-bottom:0px;padding:.3em;text-align:center;font-size:.9em}#quarto-draft-alert i{margin-right:.3em}#quarto-back-to-top{z-index:1000}pre{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;font-size:0.875em;font-weight:400}pre code{font-family:inherit;font-size:inherit;font-weight:inherit}code{font-family:SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;font-size:0.875em;font-weight:400}a{background-color:rgba(0,0,0,0);font-weight:400;text-decoration:underline}a.external:after{content:"";background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(174, 209, 228)" class="bi bi-box-arrow-up-right" viewBox="0 0 16 16"><path fill-rule="evenodd" d="M8.636 3.5a.5.5 0 0 0-.5-.5H1.5A1.5 1.5 0 0 0 0 4.5v10A1.5 1.5 0 0 0 1.5 16h10a1.5 1.5 0 0 0 1.5-1.5V7.864a.5.5 0 0 0-1 0V14.5a.5.5 0 0 1-.5.5h-10a.5.5 0 0 1-.5-.5v-10a.5.5 0 0 1 .5-.5h6.636a.5.5 0 0 0 .5-.5z"/><path fill-rule="evenodd" d="M16 .5a.5.5 0 0 0-.5-.5h-5a.5.5 0 0 0 0 1h3.793L6.146 9.146a.5.5 0 1 0 .708.708L15 1.707V5.5a.5.5 0 0 0 1 0v-5z"/></svg>');background-size:contain;background-repeat:no-repeat;background-position:center center;margin-left:.2em;padding-right:.75em}div.sourceCode code a.external:after{content:none}a.external:after:hover{cursor:pointer}.quarto-ext-icon{display:inline-block;font-size:.75em;padding-left:.3em}.code-with-filename .code-with-filename-file{margin-bottom:0;padding-bottom:2px;padding-top:2px;padding-left:.7em;border:var(--quarto-border-width) solid var(--quarto-border-color);border-radius:var(--quarto-border-radius);border-bottom:0;border-bottom-left-radius:0%;border-bottom-right-radius:0%}.code-with-filename div.sourceCode,.reveal .code-with-filename div.sourceCode{margin-top:0;border-top-left-radius:0%;border-top-right-radius:0%}.code-with-filename .code-with-filename-file pre{margin-bottom:0}.code-with-filename .code-with-filename-file{background-color:rgba(219,219,219,.8)}.quarto-dark .code-with-filename .code-with-filename-file{background-color:#555}.code-with-filename .code-with-filename-file strong{font-weight:400}.quarto-title-banner{margin-bottom:1em;color:#bdbdbd;background:#303030}.quarto-title-banner a{color:#bdbdbd}.quarto-title-banner h1,.quarto-title-banner .h1,.quarto-title-banner h2,.quarto-title-banner .h2{color:#bdbdbd}.quarto-title-banner .code-tools-button{color:#8a8a8a}.quarto-title-banner .code-tools-button:hover{color:#bdbdbd}.quarto-title-banner .code-tools-button>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(138, 138, 138)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>')}.quarto-title-banner .code-tools-button:hover>.bi::before{background-image:url('data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="rgb(189, 189, 189)" viewBox="0 0 16 16"><path d="M10.478 1.647a.5.5 0 1 0-.956-.294l-4 13a.5.5 0 0 0 .956.294l4-13zM4.854 4.146a.5.5 0 0 1 0 .708L1.707 8l3.147 3.146a.5.5 0 0 1-.708.708l-3.5-3.5a.5.5 0 0 1 0-.708l3.5-3.5a.5.5 0 0 1 .708 0zm6.292 0a.5.5 0 0 0 0 .708L14.293 8l-3.147 3.146a.5.5 0 0 0 .708.708l3.5-3.5a.5.5 0 0 0 0-.708l-3.5-3.5a.5.5 0 0 0-.708 0z"/></svg>')}.quarto-title-banner .quarto-title .title{font-weight:600}.quarto-title-banner .quarto-categories{margin-top:.75em}@media(min-width: 992px){.quarto-title-banner{padding-top:2.5em;padding-bottom:2.5em}}@media(max-width: 991.98px){.quarto-title-banner{padding-top:1em;padding-bottom:1em}}@media(max-width: 767.98px){body.hypothesis-enabled #title-block-header>*{padding-right:20px}}main.quarto-banner-title-block>section:first-child>h2,main.quarto-banner-title-block>section:first-child>.h2,main.quarto-banner-title-block>section:first-child>h3,main.quarto-banner-title-block>section:first-child>.h3,main.quarto-banner-title-block>section:first-child>h4,main.quarto-banner-title-block>section:first-child>.h4{margin-top:0}.quarto-title .quarto-categories{display:flex;flex-wrap:wrap;row-gap:.5em;column-gap:.4em;padding-bottom:.5em;margin-top:.75em}.quarto-title .quarto-categories .quarto-category{padding:.25em .75em;font-size:.65em;text-transform:uppercase;border:solid 1px;border-radius:.25rem;opacity:.6}.quarto-title .quarto-categories .quarto-category a{color:inherit}.quarto-title-meta-container{display:grid;grid-template-columns:1fr auto}.quarto-title-meta-column-end{display:flex;flex-direction:column;padding-left:1em}.quarto-title-meta-column-end a .bi{margin-right:.3em}#title-block-header.quarto-title-block.default .quarto-title-meta{display:grid;grid-template-columns:repeat(2, 1fr);grid-column-gap:1em}#title-block-header.quarto-title-block.default .quarto-title .title{margin-bottom:0}#title-block-header.quarto-title-block.default .quarto-title-author-orcid img{margin-top:-0.2em;height:.8em;width:.8em}#title-block-header.quarto-title-block.default .quarto-title-author-email{opacity:.7}#title-block-header.quarto-title-block.default .quarto-description p:last-of-type{margin-bottom:0}#title-block-header.quarto-title-block.default .quarto-title-meta-contents p,#title-block-header.quarto-title-block.default .quarto-title-authors p,#title-block-header.quarto-title-block.default .quarto-title-affiliations p{margin-bottom:.1em}#title-block-header.quarto-title-block.default .quarto-title-meta-heading{text-transform:uppercase;margin-top:1em;font-size:.8em;opacity:.8;font-weight:400}#title-block-header.quarto-title-block.default .quarto-title-meta-contents{font-size:.9em}#title-block-header.quarto-title-block.default .quarto-title-meta-contents p.affiliation:last-of-type{margin-bottom:.1em}#title-block-header.quarto-title-block.default p.affiliation{margin-bottom:.1em}#title-block-header.quarto-title-block.default .keywords,#title-block-header.quarto-title-block.default .description,#title-block-header.quarto-title-block.default .abstract{margin-top:0}#title-block-header.quarto-title-block.default .keywords>p,#title-block-header.quarto-title-block.default .description>p,#title-block-header.quarto-title-block.default .abstract>p{font-size:.9em}#title-block-header.quarto-title-block.default .keywords>p:last-of-type,#title-block-header.quarto-title-block.default .description>p:last-of-type,#title-block-header.quarto-title-block.default .abstract>p:last-of-type{margin-bottom:0}#title-block-header.quarto-title-block.default .keywords .block-title,#title-block-header.quarto-title-block.default .description .block-title,#title-block-header.quarto-title-block.default .abstract .block-title{margin-top:1em;text-transform:uppercase;font-size:.8em;opacity:.8;font-weight:400}#title-block-header.quarto-title-block.default .quarto-title-meta-author{display:grid;grid-template-columns:minmax(max-content, 1fr) 1fr;grid-column-gap:1em}.quarto-title-tools-only{display:flex;justify-content:right}body{-webkit-font-smoothing:antialiased}.badge.bg-light{color:#343a40}.progress .progress-bar{font-size:8px;line-height:8px}:root{--quarto-scss-export-gray-300: #dee2e6;--quarto-scss-export-gray-500: #adb5bd;--quarto-scss-export-gray-600: #6c757d;--quarto-scss-export-gray-800: #343a40;--quarto-scss-export-card-cap-bg: rgba(52, 58, 64, 0.25);--quarto-scss-export-border-color: #dee2e6;--quarto-scss-export-text-muted: #6c757d;--quarto-scss-export-body-bg: #181818;--quarto-scss-export-link-color: #AED1E4;--quarto-scss-export-light: #525252;--quarto-scss-export-footer-bg: #181818;--quarto-scss-export-navbar-bg: #303030;--quarto-scss-export-popover-bg: #242424;--quarto-scss-export-input-bg: #242424;--quarto-scss-export-dropdown-bg: #181818;--quarto-scss-export-quarto-body-bg: #181818;--quarto-scss-export-bs-body-bg: #181818;--quarto-scss-export-sidebar-bg: #353535;--quarto-scss-export-white: #fff;--quarto-scss-export-gray-100: #f8f9fa;--quarto-scss-export-gray-200: #e9ecef;--quarto-scss-export-gray-400: #ced4da;--quarto-scss-export-gray-700: #495057;--quarto-scss-export-gray-900: #212529;--quarto-scss-export-black: #000;--quarto-scss-export-blue: #2780e3;--quarto-scss-export-indigo: #6610f2;--quarto-scss-export-purple: #613d7c;--quarto-scss-export-pink: #e83e8c;--quarto-scss-export-red: #ff0039;--quarto-scss-export-orange: #f0ad4e;--quarto-scss-export-yellow: #ff7518;--quarto-scss-export-green: #3fb618;--quarto-scss-export-teal: #20c997;--quarto-scss-export-cyan: #9954bb;--quarto-scss-export-primary: #2780e3;--quarto-scss-export-secondary: #343a40;--quarto-scss-export-success: #3fb618;--quarto-scss-export-info: #9954bb;--quarto-scss-export-warning: #ff7518;--quarto-scss-export-danger: #ff0039;--quarto-scss-export-dark: #343a40;--quarto-scss-export-title-banner-color: ;--quarto-scss-export-title-banner-bg: ;--quarto-scss-export-btn-code-copy-color: #616e88;--quarto-scss-export-btn-code-copy-color-active: #88c0d0;--quarto-scss-export-link-color-bg: transparent;--quarto-scss-export-code-color: #7d12ba;--quarto-scss-export-code-bg: #f8f9fa;--quarto-scss-export-toc-color: #AED1E4;--quarto-scss-export-toc-active-border: #AED1E4;--quarto-scss-export-toc-inactive-border: #e9ecef;--quarto-scss-export-navbar-default: #2780e3;--quarto-scss-export-navbar-hl-override: #aed1e4;--quarto-scss-export-btn-bg: #343a40;--quarto-scss-export-btn-fg: #cacccd;--quarto-scss-export-body-contrast-bg: #181818;--quarto-scss-export-body-contrast-color: #FAF1E4;--quarto-scss-export-navbar-fg: #bdbdbd;--quarto-scss-export-navbar-hl: #aed1e4;--quarto-scss-export-navbar-brand: #bdbdbd;--quarto-scss-export-navbar-brand-hl: #aed1e4;--quarto-scss-export-navbar-toggler-border-color: rgba(189, 189, 189, 0);--quarto-scss-export-navbar-hover-color: rgba(174, 209, 228, 0.8);--quarto-scss-export-navbar-disabled-color: rgba(189, 189, 189, 0.75);--quarto-scss-export-title-block-color: #FAF1E4;--quarto-scss-export-title-block-contast-color: #181818;--quarto-scss-export-footer-fg: #828282;--quarto-scss-export-input-border-color: #dee2e6;--quarto-scss-export-code-annotation-higlight-color: rgba(170, 170, 170, 0.2666666667);--quarto-scss-export-code-annotation-higlight-bg: rgba(170, 170, 170, 0.1333333333);--quarto-scss-export-table-group-separator-color: #89857e;--quarto-scss-export-table-group-separator-color-lighter: #454341;--quarto-scss-export-link-decoration: underline;--quarto-scss-export-table-border-color: #dee2e6;--quarto-scss-export-sidebar-glass-bg: rgba(102, 102, 102, 0.4);--quarto-scss-export-color-contrast-dark: #000;--quarto-scss-export-color-contrast-light: #fff;--quarto-scss-export-blue-100: #d4e6f9;--quarto-scss-export-blue-200: #a9ccf4;--quarto-scss-export-blue-300: #7db3ee;--quarto-scss-export-blue-400: #5299e9;--quarto-scss-export-blue-500: #2780e3;--quarto-scss-export-blue-600: #1f66b6;--quarto-scss-export-blue-700: #174d88;--quarto-scss-export-blue-800: #10335b;--quarto-scss-export-blue-900: #081a2d;--quarto-scss-export-indigo-100: #e0cffc;--quarto-scss-export-indigo-200: #c29ffa;--quarto-scss-export-indigo-300: #a370f7;--quarto-scss-export-indigo-400: #8540f5;--quarto-scss-export-indigo-500: #6610f2;--quarto-scss-export-indigo-600: #520dc2;--quarto-scss-export-indigo-700: #3d0a91;--quarto-scss-export-indigo-800: #290661;--quarto-scss-export-indigo-900: #140330;--quarto-scss-export-purple-100: #dfd8e5;--quarto-scss-export-purple-200: #c0b1cb;--quarto-scss-export-purple-300: #a08bb0;--quarto-scss-export-purple-400: #816496;--quarto-scss-export-purple-500: #613d7c;--quarto-scss-export-purple-600: #4e3163;--quarto-scss-export-purple-700: #3a254a;--quarto-scss-export-purple-800: #271832;--quarto-scss-export-purple-900: #130c19;--quarto-scss-export-pink-100: #fad8e8;--quarto-scss-export-pink-200: #f6b2d1;--quarto-scss-export-pink-300: #f18bba;--quarto-scss-export-pink-400: #ed65a3;--quarto-scss-export-pink-500: #e83e8c;--quarto-scss-export-pink-600: #ba3270;--quarto-scss-export-pink-700: #8b2554;--quarto-scss-export-pink-800: #5d1938;--quarto-scss-export-pink-900: #2e0c1c;--quarto-scss-export-red-100: #ffccd7;--quarto-scss-export-red-200: #ff99b0;--quarto-scss-export-red-300: #ff6688;--quarto-scss-export-red-400: #ff3361;--quarto-scss-export-red-500: #ff0039;--quarto-scss-export-red-600: #cc002e;--quarto-scss-export-red-700: #990022;--quarto-scss-export-red-800: #660017;--quarto-scss-export-red-900: #33000b;--quarto-scss-export-orange-100: #fcefdc;--quarto-scss-export-orange-200: #f9deb8;--quarto-scss-export-orange-300: #f6ce95;--quarto-scss-export-orange-400: #f3bd71;--quarto-scss-export-orange-500: #f0ad4e;--quarto-scss-export-orange-600: #c08a3e;--quarto-scss-export-orange-700: #90682f;--quarto-scss-export-orange-800: #60451f;--quarto-scss-export-orange-900: #302310;--quarto-scss-export-yellow-100: #ffe3d1;--quarto-scss-export-yellow-200: #ffc8a3;--quarto-scss-export-yellow-300: #ffac74;--quarto-scss-export-yellow-400: #ff9146;--quarto-scss-export-yellow-500: #ff7518;--quarto-scss-export-yellow-600: #cc5e13;--quarto-scss-export-yellow-700: #99460e;--quarto-scss-export-yellow-800: #662f0a;--quarto-scss-export-yellow-900: #331705;--quarto-scss-export-green-100: #d9f0d1;--quarto-scss-export-green-200: #b2e2a3;--quarto-scss-export-green-300: #8cd374;--quarto-scss-export-green-400: #65c546;--quarto-scss-export-green-500: #3fb618;--quarto-scss-export-green-600: #329213;--quarto-scss-export-green-700: #266d0e;--quarto-scss-export-green-800: #19490a;--quarto-scss-export-green-900: #0d2405;--quarto-scss-export-teal-100: #d2f4ea;--quarto-scss-export-teal-200: #a6e9d5;--quarto-scss-export-teal-300: #79dfc1;--quarto-scss-export-teal-400: #4dd4ac;--quarto-scss-export-teal-500: #20c997;--quarto-scss-export-teal-600: #1aa179;--quarto-scss-export-teal-700: #13795b;--quarto-scss-export-teal-800: #0d503c;--quarto-scss-export-teal-900: #06281e;--quarto-scss-export-cyan-100: #ebddf1;--quarto-scss-export-cyan-200: #d6bbe4;--quarto-scss-export-cyan-300: #c298d6;--quarto-scss-export-cyan-400: #ad76c9;--quarto-scss-export-cyan-500: #9954bb;--quarto-scss-export-cyan-600: #7a4396;--quarto-scss-export-cyan-700: #5c3270;--quarto-scss-export-cyan-800: #3d224b;--quarto-scss-export-cyan-900: #1f1125;--quarto-scss-export-default: #343a40;--quarto-scss-export-primary-text-emphasis: #10335b;--quarto-scss-export-secondary-text-emphasis: #15171a;--quarto-scss-export-success-text-emphasis: #19490a;--quarto-scss-export-info-text-emphasis: #3d224b;--quarto-scss-export-warning-text-emphasis: #662f0a;--quarto-scss-export-danger-text-emphasis: #660017;--quarto-scss-export-light-text-emphasis: #495057;--quarto-scss-export-dark-text-emphasis: #495057;--quarto-scss-export-primary-bg-subtle: #d4e6f9;--quarto-scss-export-secondary-bg-subtle: #d6d8d9;--quarto-scss-export-success-bg-subtle: #d9f0d1;--quarto-scss-export-info-bg-subtle: #ebddf1;--quarto-scss-export-warning-bg-subtle: #ffe3d1;--quarto-scss-export-danger-bg-subtle: #ffccd7;--quarto-scss-export-light-bg-subtle: #fcfcfd;--quarto-scss-export-dark-bg-subtle: #ced4da;--quarto-scss-export-primary-border-subtle: #a9ccf4;--quarto-scss-export-secondary-border-subtle: #aeb0b3;--quarto-scss-export-success-border-subtle: #b2e2a3;--quarto-scss-export-info-border-subtle: #d6bbe4;--quarto-scss-export-warning-border-subtle: #ffc8a3;--quarto-scss-export-danger-border-subtle: #ff99b0;--quarto-scss-export-light-border-subtle: #e9ecef;--quarto-scss-export-dark-border-subtle: #adb5bd;--quarto-scss-export-body-text-align: ;--quarto-scss-export-body-secondary-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-body-secondary-bg: #e9ecef;--quarto-scss-export-body-tertiary-color: rgba(250, 241, 228, 0.5);--quarto-scss-export-body-tertiary-bg: #f8f9fa;--quarto-scss-export-body-emphasis-color: #000;--quarto-scss-export-link-hover-color: #8ba7b6;--quarto-scss-export-link-hover-decoration: ;--quarto-scss-export-border-color-translucent: rgba(0, 0, 0, 0.175);--quarto-scss-export-component-active-bg: #2780e3;--quarto-scss-export-component-active-color: #fff;--quarto-scss-export-focus-ring-color: rgba(39, 128, 227, 0.25);--quarto-scss-export-headings-font-family: ;--quarto-scss-export-headings-font-style: ;--quarto-scss-export-display-font-family: ;--quarto-scss-export-display-font-style: ;--quarto-scss-export-blockquote-footer-color: #6c757d;--quarto-scss-export-blockquote-border-color: #e9ecef;--quarto-scss-export-hr-bg-color: ;--quarto-scss-export-hr-height: ;--quarto-scss-export-hr-border-color: ;--quarto-scss-export-legend-font-weight: ;--quarto-scss-export-mark-bg: #ffe3d1;--quarto-scss-export-table-bg: #181818;--quarto-scss-export-table-accent-bg: transparent;--quarto-scss-export-table-th-font-weight: ;--quarto-scss-export-table-striped-bg: rgba(0, 0, 0, 0.05);--quarto-scss-export-table-active-bg: rgba(0, 0, 0, 0.1);--quarto-scss-export-table-hover-bg: rgba(0, 0, 0, 0.075);--quarto-scss-export-table-caption-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-input-btn-font-family: ;--quarto-scss-export-input-btn-focus-color: rgba(39, 128, 227, 0.25);--quarto-scss-export-btn-font-family: ;--quarto-scss-export-btn-white-space: ;--quarto-scss-export-btn-link-color: #AED1E4;--quarto-scss-export-btn-link-hover-color: #8ba7b6;--quarto-scss-export-btn-link-disabled-color: #6c757d;--quarto-scss-export-form-text-font-style: ;--quarto-scss-export-form-text-font-weight: ;--quarto-scss-export-form-text-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-form-label-font-size: ;--quarto-scss-export-form-label-font-style: ;--quarto-scss-export-form-label-font-weight: ;--quarto-scss-export-form-label-color: ;--quarto-scss-export-input-font-family: ;--quarto-scss-export-input-disabled-color: ;--quarto-scss-export-input-disabled-bg: #e9ecef;--quarto-scss-export-input-disabled-border-color: ;--quarto-scss-export-input-focus-bg: #242424;--quarto-scss-export-input-focus-border-color: #93c0f1;--quarto-scss-export-input-placeholder-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-form-check-label-color: ;--quarto-scss-export-form-check-transition: ;--quarto-scss-export-form-check-input-bg: #242424;--quarto-scss-export-form-check-input-focus-border: #93c0f1;--quarto-scss-export-form-check-input-checked-color: #fff;--quarto-scss-export-form-check-input-checked-bg-color: #2780e3;--quarto-scss-export-form-check-input-checked-border-color: #2780e3;--quarto-scss-export-form-check-input-indeterminate-color: #fff;--quarto-scss-export-form-check-input-indeterminate-bg-color: #2780e3;--quarto-scss-export-form-check-input-indeterminate-border-color: #2780e3;--quarto-scss-export-form-switch-color: rgba(0, 0, 0, 0.25);--quarto-scss-export-form-switch-focus-color: #93c0f1;--quarto-scss-export-form-switch-checked-color: #fff;--quarto-scss-export-input-group-addon-bg: #f8f9fa;--quarto-scss-export-input-group-addon-border-color: #dee2e6;--quarto-scss-export-form-select-font-family: ;--quarto-scss-export-form-select-bg: #242424;--quarto-scss-export-form-select-disabled-color: ;--quarto-scss-export-form-select-disabled-bg: #e9ecef;--quarto-scss-export-form-select-disabled-border-color: ;--quarto-scss-export-form-select-indicator-color: #343a40;--quarto-scss-export-form-select-border-color: #dee2e6;--quarto-scss-export-form-select-focus-border-color: #93c0f1;--quarto-scss-export-form-range-track-bg: #f8f9fa;--quarto-scss-export-form-range-thumb-bg: #2780e3;--quarto-scss-export-form-range-thumb-active-bg: #bed9f7;--quarto-scss-export-form-range-thumb-disabled-bg: rgba(250, 241, 228, 0.75);--quarto-scss-export-form-file-button-bg: #f8f9fa;--quarto-scss-export-form-file-button-hover-bg: #e9ecef;--quarto-scss-export-form-floating-label-disabled-color: #6c757d;--quarto-scss-export-form-feedback-font-style: ;--quarto-scss-export-form-feedback-valid-color: #3fb618;--quarto-scss-export-form-feedback-invalid-color: #ff0039;--quarto-scss-export-form-feedback-icon-valid-color: #3fb618;--quarto-scss-export-form-feedback-icon-invalid-color: #ff0039;--quarto-scss-export-form-valid-color: #3fb618;--quarto-scss-export-form-valid-border-color: #3fb618;--quarto-scss-export-form-invalid-color: #ff0039;--quarto-scss-export-form-invalid-border-color: #ff0039;--quarto-scss-export-nav-link-font-size: ;--quarto-scss-export-nav-link-font-weight: ;--quarto-scss-export-nav-link-color: #AED1E4;--quarto-scss-export-nav-link-hover-color: #8ba7b6;--quarto-scss-export-nav-link-disabled-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-nav-tabs-border-color: #dee2e6;--quarto-scss-export-nav-tabs-link-hover-border-color: #e9ecef #e9ecef #dee2e6;--quarto-scss-export-nav-tabs-link-active-color: #000;--quarto-scss-export-nav-tabs-link-active-bg: #181818;--quarto-scss-export-nav-pills-link-active-bg: #2780e3;--quarto-scss-export-nav-pills-link-active-color: #fff;--quarto-scss-export-nav-underline-link-active-color: #000;--quarto-scss-export-navbar-padding-x: ;--quarto-scss-export-navbar-light-contrast: #fff;--quarto-scss-export-navbar-dark-contrast: #fff;--quarto-scss-export-navbar-light-icon-color: rgba(255, 255, 255, 0.75);--quarto-scss-export-navbar-dark-icon-color: rgba(255, 255, 255, 0.75);--quarto-scss-export-dropdown-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-divider-bg: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-link-active-bg: #2780e3;--quarto-scss-export-dropdown-link-active-color: #fff;--quarto-scss-export-dropdown-link-disabled-color: rgba(250, 241, 228, 0.5);--quarto-scss-export-dropdown-header-color: #6c757d;--quarto-scss-export-dropdown-dark-color: #dee2e6;--quarto-scss-export-dropdown-dark-bg: #343a40;--quarto-scss-export-dropdown-dark-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-dark-divider-bg: rgba(0, 0, 0, 0.175);--quarto-scss-export-dropdown-dark-box-shadow: ;--quarto-scss-export-dropdown-dark-link-color: #dee2e6;--quarto-scss-export-dropdown-dark-link-hover-color: #fff;--quarto-scss-export-dropdown-dark-link-hover-bg: rgba(255, 255, 255, 0.15);--quarto-scss-export-dropdown-dark-link-active-color: #fff;--quarto-scss-export-dropdown-dark-link-active-bg: #2780e3;--quarto-scss-export-dropdown-dark-link-disabled-color: #adb5bd;--quarto-scss-export-dropdown-dark-header-color: #adb5bd;--quarto-scss-export-pagination-color: #AED1E4;--quarto-scss-export-pagination-bg: #181818;--quarto-scss-export-pagination-border-color: #dee2e6;--quarto-scss-export-pagination-focus-color: #8ba7b6;--quarto-scss-export-pagination-focus-bg: #e9ecef;--quarto-scss-export-pagination-hover-color: #8ba7b6;--quarto-scss-export-pagination-hover-bg: #f8f9fa;--quarto-scss-export-pagination-hover-border-color: #dee2e6;--quarto-scss-export-pagination-active-color: #fff;--quarto-scss-export-pagination-active-bg: #2780e3;--quarto-scss-export-pagination-active-border-color: #2780e3;--quarto-scss-export-pagination-disabled-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-pagination-disabled-bg: #e9ecef;--quarto-scss-export-pagination-disabled-border-color: #dee2e6;--quarto-scss-export-card-title-color: ;--quarto-scss-export-card-subtitle-color: ;--quarto-scss-export-card-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-card-box-shadow: ;--quarto-scss-export-card-cap-color: ;--quarto-scss-export-card-height: ;--quarto-scss-export-card-color: ;--quarto-scss-export-card-bg: #181818;--quarto-scss-export-accordion-bg: #181818;--quarto-scss-export-accordion-border-color: #dee2e6;--quarto-scss-export-accordion-button-bg: #181818;--quarto-scss-export-accordion-button-active-bg: #d4e6f9;--quarto-scss-export-accordion-button-active-color: #10335b;--quarto-scss-export-accordion-button-focus-border-color: #93c0f1;--quarto-scss-export-accordion-icon-active-color: #10335b;--quarto-scss-export-tooltip-color: #181818;--quarto-scss-export-tooltip-bg: #000;--quarto-scss-export-tooltip-margin: ;--quarto-scss-export-tooltip-arrow-color: ;--quarto-scss-export-form-feedback-tooltip-line-height: ;--quarto-scss-export-popover-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-popover-header-bg: #e9ecef;--quarto-scss-export-popover-arrow-color: #242424;--quarto-scss-export-popover-arrow-outer-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-toast-color: ;--quarto-scss-export-toast-background-color: rgba(24, 24, 24, 0.85);--quarto-scss-export-toast-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-toast-header-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-toast-header-background-color: rgba(24, 24, 24, 0.85);--quarto-scss-export-toast-header-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-badge-color: #fff;--quarto-scss-export-modal-content-color: ;--quarto-scss-export-modal-content-bg: #181818;--quarto-scss-export-modal-content-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-modal-backdrop-bg: #000;--quarto-scss-export-modal-header-border-color: #dee2e6;--quarto-scss-export-modal-footer-bg: ;--quarto-scss-export-modal-footer-border-color: #dee2e6;--quarto-scss-export-progress-bg: #e9ecef;--quarto-scss-export-progress-bar-color: #fff;--quarto-scss-export-progress-bar-bg: #2780e3;--quarto-scss-export-list-group-bg: #181818;--quarto-scss-export-list-group-border-color: #dee2e6;--quarto-scss-export-list-group-hover-bg: #f8f9fa;--quarto-scss-export-list-group-active-bg: #2780e3;--quarto-scss-export-list-group-active-color: #fff;--quarto-scss-export-list-group-active-border-color: #2780e3;--quarto-scss-export-list-group-disabled-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-list-group-disabled-bg: #181818;--quarto-scss-export-list-group-action-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-list-group-action-hover-color: #000;--quarto-scss-export-list-group-action-active-bg: #e9ecef;--quarto-scss-export-thumbnail-bg: #181818;--quarto-scss-export-thumbnail-border-color: #dee2e6;--quarto-scss-export-figure-caption-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-breadcrumb-font-size: ;--quarto-scss-export-breadcrumb-bg: ;--quarto-scss-export-breadcrumb-divider-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-breadcrumb-active-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-breadcrumb-border-radius: ;--quarto-scss-export-carousel-control-color: #fff;--quarto-scss-export-carousel-indicator-active-bg: #fff;--quarto-scss-export-carousel-caption-color: #fff;--quarto-scss-export-carousel-dark-indicator-active-bg: #000;--quarto-scss-export-carousel-dark-caption-color: #000;--quarto-scss-export-btn-close-color: #000;--quarto-scss-export-offcanvas-border-color: rgba(0, 0, 0, 0.175);--quarto-scss-export-offcanvas-bg-color: #181818;--quarto-scss-export-offcanvas-backdrop-bg: #000;--quarto-scss-export-code-color-dark: white;--quarto-scss-export-kbd-color: #181818;--quarto-scss-export-nested-kbd-font-weight: ;--quarto-scss-export-pre-bg: #f8f9fa;--quarto-scss-export-pre-color: #000;--quarto-scss-export-bslib-page-sidebar-title-bg: #303030;--quarto-scss-export-bslib-page-sidebar-title-color: #fff;--quarto-scss-export-bslib-sidebar-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.05);--quarto-scss-export-bslib-sidebar-toggle-bg: rgba(var(--bs-emphasis-color-rgb, 0, 0, 0), 0.1);--quarto-scss-export-sidebar-color: #faf1e4;--quarto-scss-export-sidebar-hover-color: rgba(174, 209, 228, 0.8);--quarto-scss-export-sidebar-disabled-color: rgba(250, 241, 228, 0.75);--quarto-scss-export-valuebox-bg-primary: #5397e9;--quarto-scss-export-valuebox-bg-secondary: #343a40;--quarto-scss-export-valuebox-bg-success: #3aa716;--quarto-scss-export-valuebox-bg-info: rgba(153, 84, 187, 0.7019607843);--quarto-scss-export-valuebox-bg-warning: #fa6400;--quarto-scss-export-valuebox-bg-danger: rgba(255, 0, 57, 0.7019607843);--quarto-scss-export-valuebox-bg-light: #525252;--quarto-scss-export-valuebox-bg-dark: #343a40;--quarto-scss-export-mermaid-bg-color: #181818;--quarto-scss-export-mermaid-edge-color: #343a40;--quarto-scss-export-mermaid-fg-color--lighter: white;--quarto-scss-export-mermaid-fg-color--lightest: white;--quarto-scss-export-mermaid-label-bg-color: #181818;--quarto-scss-export-mermaid-label-fg-color: #2780e3;--quarto-scss-export-mermaid-node-bg-color: rgba(39, 128, 227, 0.1);--quarto-scss-export-code-block-border-left-color: #dee2e6;--quarto-scss-export-callout-color-note: #2780e3;--quarto-scss-export-callout-color-tip: #3fb618;--quarto-scss-export-callout-color-important: #ff0039;--quarto-scss-export-callout-color-caution: #f0ad4e;--quarto-scss-export-callout-color-warning: #ff7518}
\ No newline at end of file
diff --git a/docs/2_39/site_libs/bootstrap/bootstrap-icons.css b/docs/2_39/site_libs/bootstrap/bootstrap-icons.css
new file mode 100644
index 000000000..285e4448f
--- /dev/null
+++ b/docs/2_39/site_libs/bootstrap/bootstrap-icons.css
@@ -0,0 +1,2078 @@
+/*!
+ * Bootstrap Icons v1.11.1 (https://icons.getbootstrap.com/)
+ * Copyright 2019-2023 The Bootstrap Authors
+ * Licensed under MIT (https://github.com/twbs/icons/blob/main/LICENSE)
+ */
+
+@font-face {
+  font-display: block;
+  font-family: "bootstrap-icons";
+  src: 
+url("./bootstrap-icons.woff?2820a3852bdb9a5832199cc61cec4e65") format("woff");
+}
+
+.bi::before,
+[class^="bi-"]::before,
+[class*=" bi-"]::before {
+  display: inline-block;
+  font-family: bootstrap-icons !important;
+  font-style: normal;
+  font-weight: normal !important;
+  font-variant: normal;
+  text-transform: none;
+  line-height: 1;
+  vertical-align: -.125em;
+  -webkit-font-smoothing: antialiased;
+  -moz-osx-font-smoothing: grayscale;
+}
+
+.bi-123::before { content: "\f67f"; }
+.bi-alarm-fill::before { content: "\f101"; }
+.bi-alarm::before { content: "\f102"; }
+.bi-align-bottom::before { content: "\f103"; }
+.bi-align-center::before { content: "\f104"; }
+.bi-align-end::before { content: "\f105"; }
+.bi-align-middle::before { content: "\f106"; }
+.bi-align-start::before { content: "\f107"; }
+.bi-align-top::before { content: "\f108"; }
+.bi-alt::before { content: "\f109"; }
+.bi-app-indicator::before { content: "\f10a"; }
+.bi-app::before { content: "\f10b"; }
+.bi-archive-fill::before { content: "\f10c"; }
+.bi-archive::before { content: "\f10d"; }
+.bi-arrow-90deg-down::before { content: "\f10e"; }
+.bi-arrow-90deg-left::before { content: "\f10f"; }
+.bi-arrow-90deg-right::before { content: "\f110"; }
+.bi-arrow-90deg-up::before { content: "\f111"; }
+.bi-arrow-bar-down::before { content: "\f112"; }
+.bi-arrow-bar-left::before { content: "\f113"; }
+.bi-arrow-bar-right::before { content: "\f114"; }
+.bi-arrow-bar-up::before { content: "\f115"; }
+.bi-arrow-clockwise::before { content: "\f116"; }
+.bi-arrow-counterclockwise::before { content: "\f117"; }
+.bi-arrow-down-circle-fill::before { content: "\f118"; }
+.bi-arrow-down-circle::before { content: "\f119"; }
+.bi-arrow-down-left-circle-fill::before { content: "\f11a"; }
+.bi-arrow-down-left-circle::before { content: "\f11b"; }
+.bi-arrow-down-left-square-fill::before { content: "\f11c"; }
+.bi-arrow-down-left-square::before { content: "\f11d"; }
+.bi-arrow-down-left::before { content: "\f11e"; }
+.bi-arrow-down-right-circle-fill::before { content: "\f11f"; }
+.bi-arrow-down-right-circle::before { content: "\f120"; }
+.bi-arrow-down-right-square-fill::before { content: "\f121"; }
+.bi-arrow-down-right-square::before { content: "\f122"; }
+.bi-arrow-down-right::before { content: "\f123"; }
+.bi-arrow-down-short::before { content: "\f124"; }
+.bi-arrow-down-square-fill::before { content: "\f125"; }
+.bi-arrow-down-square::before { content: "\f126"; }
+.bi-arrow-down-up::before { content: "\f127"; }
+.bi-arrow-down::before { content: "\f128"; }
+.bi-arrow-left-circle-fill::before { content: "\f129"; }
+.bi-arrow-left-circle::before { content: "\f12a"; }
+.bi-arrow-left-right::before { content: "\f12b"; }
+.bi-arrow-left-short::before { content: "\f12c"; }
+.bi-arrow-left-square-fill::before { content: "\f12d"; }
+.bi-arrow-left-square::before { content: "\f12e"; }
+.bi-arrow-left::before { content: "\f12f"; }
+.bi-arrow-repeat::before { content: "\f130"; }
+.bi-arrow-return-left::before { content: "\f131"; }
+.bi-arrow-return-right::before { content: "\f132"; }
+.bi-arrow-right-circle-fill::before { content: "\f133"; }
+.bi-arrow-right-circle::before { content: "\f134"; }
+.bi-arrow-right-short::before { content: "\f135"; }
+.bi-arrow-right-square-fill::before { content: "\f136"; }
+.bi-arrow-right-square::before { content: "\f137"; }
+.bi-arrow-right::before { content: "\f138"; }
+.bi-arrow-up-circle-fill::before { content: "\f139"; }
+.bi-arrow-up-circle::before { content: "\f13a"; }
+.bi-arrow-up-left-circle-fill::before { content: "\f13b"; }
+.bi-arrow-up-left-circle::before { content: "\f13c"; }
+.bi-arrow-up-left-square-fill::before { content: "\f13d"; }
+.bi-arrow-up-left-square::before { content: "\f13e"; }
+.bi-arrow-up-left::before { content: "\f13f"; }
+.bi-arrow-up-right-circle-fill::before { content: "\f140"; }
+.bi-arrow-up-right-circle::before { content: "\f141"; }
+.bi-arrow-up-right-square-fill::before { content: "\f142"; }
+.bi-arrow-up-right-square::before { content: "\f143"; }
+.bi-arrow-up-right::before { content: "\f144"; }
+.bi-arrow-up-short::before { content: "\f145"; }
+.bi-arrow-up-square-fill::before { content: "\f146"; }
+.bi-arrow-up-square::before { content: "\f147"; }
+.bi-arrow-up::before { content: "\f148"; }
+.bi-arrows-angle-contract::before { content: "\f149"; }
+.bi-arrows-angle-expand::before { content: "\f14a"; }
+.bi-arrows-collapse::before { content: "\f14b"; }
+.bi-arrows-expand::before { content: "\f14c"; }
+.bi-arrows-fullscreen::before { content: "\f14d"; }
+.bi-arrows-move::before { content: "\f14e"; }
+.bi-aspect-ratio-fill::before { content: "\f14f"; }
+.bi-aspect-ratio::before { content: "\f150"; }
+.bi-asterisk::before { content: "\f151"; }
+.bi-at::before { content: "\f152"; }
+.bi-award-fill::before { content: "\f153"; }
+.bi-award::before { content: "\f154"; }
+.bi-back::before { content: "\f155"; }
+.bi-backspace-fill::before { content: "\f156"; }
+.bi-backspace-reverse-fill::before { content: "\f157"; }
+.bi-backspace-reverse::before { content: "\f158"; }
+.bi-backspace::before { content: "\f159"; }
+.bi-badge-3d-fill::before { content: "\f15a"; }
+.bi-badge-3d::before { content: "\f15b"; }
+.bi-badge-4k-fill::before { content: "\f15c"; }
+.bi-badge-4k::before { content: "\f15d"; }
+.bi-badge-8k-fill::before { content: "\f15e"; }
+.bi-badge-8k::before { content: "\f15f"; }
+.bi-badge-ad-fill::before { content: "\f160"; }
+.bi-badge-ad::before { content: "\f161"; }
+.bi-badge-ar-fill::before { content: "\f162"; }
+.bi-badge-ar::before { content: "\f163"; }
+.bi-badge-cc-fill::before { content: "\f164"; }
+.bi-badge-cc::before { content: "\f165"; }
+.bi-badge-hd-fill::before { content: "\f166"; }
+.bi-badge-hd::before { content: "\f167"; }
+.bi-badge-tm-fill::before { content: "\f168"; }
+.bi-badge-tm::before { content: "\f169"; }
+.bi-badge-vo-fill::before { content: "\f16a"; }
+.bi-badge-vo::before { content: "\f16b"; }
+.bi-badge-vr-fill::before { content: "\f16c"; }
+.bi-badge-vr::before { content: "\f16d"; }
+.bi-badge-wc-fill::before { content: "\f16e"; }
+.bi-badge-wc::before { content: "\f16f"; }
+.bi-bag-check-fill::before { content: "\f170"; }
+.bi-bag-check::before { content: "\f171"; }
+.bi-bag-dash-fill::before { content: "\f172"; }
+.bi-bag-dash::before { content: "\f173"; }
+.bi-bag-fill::before { content: "\f174"; }
+.bi-bag-plus-fill::before { content: "\f175"; }
+.bi-bag-plus::before { content: "\f176"; }
+.bi-bag-x-fill::before { content: "\f177"; }
+.bi-bag-x::before { content: "\f178"; }
+.bi-bag::before { content: "\f179"; }
+.bi-bar-chart-fill::before { content: "\f17a"; }
+.bi-bar-chart-line-fill::before { content: "\f17b"; }
+.bi-bar-chart-line::before { content: "\f17c"; }
+.bi-bar-chart-steps::before { content: "\f17d"; }
+.bi-bar-chart::before { content: "\f17e"; }
+.bi-basket-fill::before { content: "\f17f"; }
+.bi-basket::before { content: "\f180"; }
+.bi-basket2-fill::before { content: "\f181"; }
+.bi-basket2::before { content: "\f182"; }
+.bi-basket3-fill::before { content: "\f183"; }
+.bi-basket3::before { content: "\f184"; }
+.bi-battery-charging::before { content: "\f185"; }
+.bi-battery-full::before { content: "\f186"; }
+.bi-battery-half::before { content: "\f187"; }
+.bi-battery::before { content: "\f188"; }
+.bi-bell-fill::before { content: "\f189"; }
+.bi-bell::before { content: "\f18a"; }
+.bi-bezier::before { content: "\f18b"; }
+.bi-bezier2::before { content: "\f18c"; }
+.bi-bicycle::before { content: "\f18d"; }
+.bi-binoculars-fill::before { content: "\f18e"; }
+.bi-binoculars::before { content: "\f18f"; }
+.bi-blockquote-left::before { content: "\f190"; }
+.bi-blockquote-right::before { content: "\f191"; }
+.bi-book-fill::before { content: "\f192"; }
+.bi-book-half::before { content: "\f193"; }
+.bi-book::before { content: "\f194"; }
+.bi-bookmark-check-fill::before { content: "\f195"; }
+.bi-bookmark-check::before { content: "\f196"; }
+.bi-bookmark-dash-fill::before { content: "\f197"; }
+.bi-bookmark-dash::before { content: "\f198"; }
+.bi-bookmark-fill::before { content: "\f199"; }
+.bi-bookmark-heart-fill::before { content: "\f19a"; }
+.bi-bookmark-heart::before { content: "\f19b"; }
+.bi-bookmark-plus-fill::before { content: "\f19c"; }
+.bi-bookmark-plus::before { content: "\f19d"; }
+.bi-bookmark-star-fill::before { content: "\f19e"; }
+.bi-bookmark-star::before { content: "\f19f"; }
+.bi-bookmark-x-fill::before { content: "\f1a0"; }
+.bi-bookmark-x::before { content: "\f1a1"; }
+.bi-bookmark::before { content: "\f1a2"; }
+.bi-bookmarks-fill::before { content: "\f1a3"; }
+.bi-bookmarks::before { content: "\f1a4"; }
+.bi-bookshelf::before { content: "\f1a5"; }
+.bi-bootstrap-fill::before { content: "\f1a6"; }
+.bi-bootstrap-reboot::before { content: "\f1a7"; }
+.bi-bootstrap::before { content: "\f1a8"; }
+.bi-border-all::before { content: "\f1a9"; }
+.bi-border-bottom::before { content: "\f1aa"; }
+.bi-border-center::before { content: "\f1ab"; }
+.bi-border-inner::before { content: "\f1ac"; }
+.bi-border-left::before { content: "\f1ad"; }
+.bi-border-middle::before { content: "\f1ae"; }
+.bi-border-outer::before { content: "\f1af"; }
+.bi-border-right::before { content: "\f1b0"; }
+.bi-border-style::before { content: "\f1b1"; }
+.bi-border-top::before { content: "\f1b2"; }
+.bi-border-width::before { content: "\f1b3"; }
+.bi-border::before { content: "\f1b4"; }
+.bi-bounding-box-circles::before { content: "\f1b5"; }
+.bi-bounding-box::before { content: "\f1b6"; }
+.bi-box-arrow-down-left::before { content: "\f1b7"; }
+.bi-box-arrow-down-right::before { content: "\f1b8"; }
+.bi-box-arrow-down::before { content: "\f1b9"; }
+.bi-box-arrow-in-down-left::before { content: "\f1ba"; }
+.bi-box-arrow-in-down-right::before { content: "\f1bb"; }
+.bi-box-arrow-in-down::before { content: "\f1bc"; }
+.bi-box-arrow-in-left::before { content: "\f1bd"; }
+.bi-box-arrow-in-right::before { content: "\f1be"; }
+.bi-box-arrow-in-up-left::before { content: "\f1bf"; }
+.bi-box-arrow-in-up-right::before { content: "\f1c0"; }
+.bi-box-arrow-in-up::before { content: "\f1c1"; }
+.bi-box-arrow-left::before { content: "\f1c2"; }
+.bi-box-arrow-right::before { content: "\f1c3"; }
+.bi-box-arrow-up-left::before { content: "\f1c4"; }
+.bi-box-arrow-up-right::before { content: "\f1c5"; }
+.bi-box-arrow-up::before { content: "\f1c6"; }
+.bi-box-seam::before { content: "\f1c7"; }
+.bi-box::before { content: "\f1c8"; }
+.bi-braces::before { content: "\f1c9"; }
+.bi-bricks::before { content: "\f1ca"; }
+.bi-briefcase-fill::before { content: "\f1cb"; }
+.bi-briefcase::before { content: "\f1cc"; }
+.bi-brightness-alt-high-fill::before { content: "\f1cd"; }
+.bi-brightness-alt-high::before { content: "\f1ce"; }
+.bi-brightness-alt-low-fill::before { content: "\f1cf"; }
+.bi-brightness-alt-low::before { content: "\f1d0"; }
+.bi-brightness-high-fill::before { content: "\f1d1"; }
+.bi-brightness-high::before { content: "\f1d2"; }
+.bi-brightness-low-fill::before { content: "\f1d3"; }
+.bi-brightness-low::before { content: "\f1d4"; }
+.bi-broadcast-pin::before { content: "\f1d5"; }
+.bi-broadcast::before { content: "\f1d6"; }
+.bi-brush-fill::before { content: "\f1d7"; }
+.bi-brush::before { content: "\f1d8"; }
+.bi-bucket-fill::before { content: "\f1d9"; }
+.bi-bucket::before { content: "\f1da"; }
+.bi-bug-fill::before { content: "\f1db"; }
+.bi-bug::before { content: "\f1dc"; }
+.bi-building::before { content: "\f1dd"; }
+.bi-bullseye::before { content: "\f1de"; }
+.bi-calculator-fill::before { content: "\f1df"; }
+.bi-calculator::before { content: "\f1e0"; }
+.bi-calendar-check-fill::before { content: "\f1e1"; }
+.bi-calendar-check::before { content: "\f1e2"; }
+.bi-calendar-date-fill::before { content: "\f1e3"; }
+.bi-calendar-date::before { content: "\f1e4"; }
+.bi-calendar-day-fill::before { content: "\f1e5"; }
+.bi-calendar-day::before { content: "\f1e6"; }
+.bi-calendar-event-fill::before { content: "\f1e7"; }
+.bi-calendar-event::before { content: "\f1e8"; }
+.bi-calendar-fill::before { content: "\f1e9"; }
+.bi-calendar-minus-fill::before { content: "\f1ea"; }
+.bi-calendar-minus::before { content: "\f1eb"; }
+.bi-calendar-month-fill::before { content: "\f1ec"; }
+.bi-calendar-month::before { content: "\f1ed"; }
+.bi-calendar-plus-fill::before { content: "\f1ee"; }
+.bi-calendar-plus::before { content: "\f1ef"; }
+.bi-calendar-range-fill::before { content: "\f1f0"; }
+.bi-calendar-range::before { content: "\f1f1"; }
+.bi-calendar-week-fill::before { content: "\f1f2"; }
+.bi-calendar-week::before { content: "\f1f3"; }
+.bi-calendar-x-fill::before { content: "\f1f4"; }
+.bi-calendar-x::before { content: "\f1f5"; }
+.bi-calendar::before { content: "\f1f6"; }
+.bi-calendar2-check-fill::before { content: "\f1f7"; }
+.bi-calendar2-check::before { content: "\f1f8"; }
+.bi-calendar2-date-fill::before { content: "\f1f9"; }
+.bi-calendar2-date::before { content: "\f1fa"; }
+.bi-calendar2-day-fill::before { content: "\f1fb"; }
+.bi-calendar2-day::before { content: "\f1fc"; }
+.bi-calendar2-event-fill::before { content: "\f1fd"; }
+.bi-calendar2-event::before { content: "\f1fe"; }
+.bi-calendar2-fill::before { content: "\f1ff"; }
+.bi-calendar2-minus-fill::before { content: "\f200"; }
+.bi-calendar2-minus::before { content: "\f201"; }
+.bi-calendar2-month-fill::before { content: "\f202"; }
+.bi-calendar2-month::before { content: "\f203"; }
+.bi-calendar2-plus-fill::before { content: "\f204"; }
+.bi-calendar2-plus::before { content: "\f205"; }
+.bi-calendar2-range-fill::before { content: "\f206"; }
+.bi-calendar2-range::before { content: "\f207"; }
+.bi-calendar2-week-fill::before { content: "\f208"; }
+.bi-calendar2-week::before { content: "\f209"; }
+.bi-calendar2-x-fill::before { content: "\f20a"; }
+.bi-calendar2-x::before { content: "\f20b"; }
+.bi-calendar2::before { content: "\f20c"; }
+.bi-calendar3-event-fill::before { content: "\f20d"; }
+.bi-calendar3-event::before { content: "\f20e"; }
+.bi-calendar3-fill::before { content: "\f20f"; }
+.bi-calendar3-range-fill::before { content: "\f210"; }
+.bi-calendar3-range::before { content: "\f211"; }
+.bi-calendar3-week-fill::before { content: "\f212"; }
+.bi-calendar3-week::before { content: "\f213"; }
+.bi-calendar3::before { content: "\f214"; }
+.bi-calendar4-event::before { content: "\f215"; }
+.bi-calendar4-range::before { content: "\f216"; }
+.bi-calendar4-week::before { content: "\f217"; }
+.bi-calendar4::before { content: "\f218"; }
+.bi-camera-fill::before { content: "\f219"; }
+.bi-camera-reels-fill::before { content: "\f21a"; }
+.bi-camera-reels::before { content: "\f21b"; }
+.bi-camera-video-fill::before { content: "\f21c"; }
+.bi-camera-video-off-fill::before { content: "\f21d"; }
+.bi-camera-video-off::before { content: "\f21e"; }
+.bi-camera-video::before { content: "\f21f"; }
+.bi-camera::before { content: "\f220"; }
+.bi-camera2::before { content: "\f221"; }
+.bi-capslock-fill::before { content: "\f222"; }
+.bi-capslock::before { content: "\f223"; }
+.bi-card-checklist::before { content: "\f224"; }
+.bi-card-heading::before { content: "\f225"; }
+.bi-card-image::before { content: "\f226"; }
+.bi-card-list::before { content: "\f227"; }
+.bi-card-text::before { content: "\f228"; }
+.bi-caret-down-fill::before { content: "\f229"; }
+.bi-caret-down-square-fill::before { content: "\f22a"; }
+.bi-caret-down-square::before { content: "\f22b"; }
+.bi-caret-down::before { content: "\f22c"; }
+.bi-caret-left-fill::before { content: "\f22d"; }
+.bi-caret-left-square-fill::before { content: "\f22e"; }
+.bi-caret-left-square::before { content: "\f22f"; }
+.bi-caret-left::before { content: "\f230"; }
+.bi-caret-right-fill::before { content: "\f231"; }
+.bi-caret-right-square-fill::before { content: "\f232"; }
+.bi-caret-right-square::before { content: "\f233"; }
+.bi-caret-right::before { content: "\f234"; }
+.bi-caret-up-fill::before { content: "\f235"; }
+.bi-caret-up-square-fill::before { content: "\f236"; }
+.bi-caret-up-square::before { content: "\f237"; }
+.bi-caret-up::before { content: "\f238"; }
+.bi-cart-check-fill::before { content: "\f239"; }
+.bi-cart-check::before { content: "\f23a"; }
+.bi-cart-dash-fill::before { content: "\f23b"; }
+.bi-cart-dash::before { content: "\f23c"; }
+.bi-cart-fill::before { content: "\f23d"; }
+.bi-cart-plus-fill::before { content: "\f23e"; }
+.bi-cart-plus::before { content: "\f23f"; }
+.bi-cart-x-fill::before { content: "\f240"; }
+.bi-cart-x::before { content: "\f241"; }
+.bi-cart::before { content: "\f242"; }
+.bi-cart2::before { content: "\f243"; }
+.bi-cart3::before { content: "\f244"; }
+.bi-cart4::before { content: "\f245"; }
+.bi-cash-stack::before { content: "\f246"; }
+.bi-cash::before { content: "\f247"; }
+.bi-cast::before { content: "\f248"; }
+.bi-chat-dots-fill::before { content: "\f249"; }
+.bi-chat-dots::before { content: "\f24a"; }
+.bi-chat-fill::before { content: "\f24b"; }
+.bi-chat-left-dots-fill::before { content: "\f24c"; }
+.bi-chat-left-dots::before { content: "\f24d"; }
+.bi-chat-left-fill::before { content: "\f24e"; }
+.bi-chat-left-quote-fill::before { content: "\f24f"; }
+.bi-chat-left-quote::before { content: "\f250"; }
+.bi-chat-left-text-fill::before { content: "\f251"; }
+.bi-chat-left-text::before { content: "\f252"; }
+.bi-chat-left::before { content: "\f253"; }
+.bi-chat-quote-fill::before { content: "\f254"; }
+.bi-chat-quote::before { content: "\f255"; }
+.bi-chat-right-dots-fill::before { content: "\f256"; }
+.bi-chat-right-dots::before { content: "\f257"; }
+.bi-chat-right-fill::before { content: "\f258"; }
+.bi-chat-right-quote-fill::before { content: "\f259"; }
+.bi-chat-right-quote::before { content: "\f25a"; }
+.bi-chat-right-text-fill::before { content: "\f25b"; }
+.bi-chat-right-text::before { content: "\f25c"; }
+.bi-chat-right::before { content: "\f25d"; }
+.bi-chat-square-dots-fill::before { content: "\f25e"; }
+.bi-chat-square-dots::before { content: "\f25f"; }
+.bi-chat-square-fill::before { content: "\f260"; }
+.bi-chat-square-quote-fill::before { content: "\f261"; }
+.bi-chat-square-quote::before { content: "\f262"; }
+.bi-chat-square-text-fill::before { content: "\f263"; }
+.bi-chat-square-text::before { content: "\f264"; }
+.bi-chat-square::before { content: "\f265"; }
+.bi-chat-text-fill::before { content: "\f266"; }
+.bi-chat-text::before { content: "\f267"; }
+.bi-chat::before { content: "\f268"; }
+.bi-check-all::before { content: "\f269"; }
+.bi-check-circle-fill::before { content: "\f26a"; }
+.bi-check-circle::before { content: "\f26b"; }
+.bi-check-square-fill::before { content: "\f26c"; }
+.bi-check-square::before { content: "\f26d"; }
+.bi-check::before { content: "\f26e"; }
+.bi-check2-all::before { content: "\f26f"; }
+.bi-check2-circle::before { content: "\f270"; }
+.bi-check2-square::before { content: "\f271"; }
+.bi-check2::before { content: "\f272"; }
+.bi-chevron-bar-contract::before { content: "\f273"; }
+.bi-chevron-bar-down::before { content: "\f274"; }
+.bi-chevron-bar-expand::before { content: "\f275"; }
+.bi-chevron-bar-left::before { content: "\f276"; }
+.bi-chevron-bar-right::before { content: "\f277"; }
+.bi-chevron-bar-up::before { content: "\f278"; }
+.bi-chevron-compact-down::before { content: "\f279"; }
+.bi-chevron-compact-left::before { content: "\f27a"; }
+.bi-chevron-compact-right::before { content: "\f27b"; }
+.bi-chevron-compact-up::before { content: "\f27c"; }
+.bi-chevron-contract::before { content: "\f27d"; }
+.bi-chevron-double-down::before { content: "\f27e"; }
+.bi-chevron-double-left::before { content: "\f27f"; }
+.bi-chevron-double-right::before { content: "\f280"; }
+.bi-chevron-double-up::before { content: "\f281"; }
+.bi-chevron-down::before { content: "\f282"; }
+.bi-chevron-expand::before { content: "\f283"; }
+.bi-chevron-left::before { content: "\f284"; }
+.bi-chevron-right::before { content: "\f285"; }
+.bi-chevron-up::before { content: "\f286"; }
+.bi-circle-fill::before { content: "\f287"; }
+.bi-circle-half::before { content: "\f288"; }
+.bi-circle-square::before { content: "\f289"; }
+.bi-circle::before { content: "\f28a"; }
+.bi-clipboard-check::before { content: "\f28b"; }
+.bi-clipboard-data::before { content: "\f28c"; }
+.bi-clipboard-minus::before { content: "\f28d"; }
+.bi-clipboard-plus::before { content: "\f28e"; }
+.bi-clipboard-x::before { content: "\f28f"; }
+.bi-clipboard::before { content: "\f290"; }
+.bi-clock-fill::before { content: "\f291"; }
+.bi-clock-history::before { content: "\f292"; }
+.bi-clock::before { content: "\f293"; }
+.bi-cloud-arrow-down-fill::before { content: "\f294"; }
+.bi-cloud-arrow-down::before { content: "\f295"; }
+.bi-cloud-arrow-up-fill::before { content: "\f296"; }
+.bi-cloud-arrow-up::before { content: "\f297"; }
+.bi-cloud-check-fill::before { content: "\f298"; }
+.bi-cloud-check::before { content: "\f299"; }
+.bi-cloud-download-fill::before { content: "\f29a"; }
+.bi-cloud-download::before { content: "\f29b"; }
+.bi-cloud-drizzle-fill::before { content: "\f29c"; }
+.bi-cloud-drizzle::before { content: "\f29d"; }
+.bi-cloud-fill::before { content: "\f29e"; }
+.bi-cloud-fog-fill::before { content: "\f29f"; }
+.bi-cloud-fog::before { content: "\f2a0"; }
+.bi-cloud-fog2-fill::before { content: "\f2a1"; }
+.bi-cloud-fog2::before { content: "\f2a2"; }
+.bi-cloud-hail-fill::before { content: "\f2a3"; }
+.bi-cloud-hail::before { content: "\f2a4"; }
+.bi-cloud-haze-fill::before { content: "\f2a6"; }
+.bi-cloud-haze::before { content: "\f2a7"; }
+.bi-cloud-haze2-fill::before { content: "\f2a8"; }
+.bi-cloud-lightning-fill::before { content: "\f2a9"; }
+.bi-cloud-lightning-rain-fill::before { content: "\f2aa"; }
+.bi-cloud-lightning-rain::before { content: "\f2ab"; }
+.bi-cloud-lightning::before { content: "\f2ac"; }
+.bi-cloud-minus-fill::before { content: "\f2ad"; }
+.bi-cloud-minus::before { content: "\f2ae"; }
+.bi-cloud-moon-fill::before { content: "\f2af"; }
+.bi-cloud-moon::before { content: "\f2b0"; }
+.bi-cloud-plus-fill::before { content: "\f2b1"; }
+.bi-cloud-plus::before { content: "\f2b2"; }
+.bi-cloud-rain-fill::before { content: "\f2b3"; }
+.bi-cloud-rain-heavy-fill::before { content: "\f2b4"; }
+.bi-cloud-rain-heavy::before { content: "\f2b5"; }
+.bi-cloud-rain::before { content: "\f2b6"; }
+.bi-cloud-slash-fill::before { content: "\f2b7"; }
+.bi-cloud-slash::before { content: "\f2b8"; }
+.bi-cloud-sleet-fill::before { content: "\f2b9"; }
+.bi-cloud-sleet::before { content: "\f2ba"; }
+.bi-cloud-snow-fill::before { content: "\f2bb"; }
+.bi-cloud-snow::before { content: "\f2bc"; }
+.bi-cloud-sun-fill::before { content: "\f2bd"; }
+.bi-cloud-sun::before { content: "\f2be"; }
+.bi-cloud-upload-fill::before { content: "\f2bf"; }
+.bi-cloud-upload::before { content: "\f2c0"; }
+.bi-cloud::before { content: "\f2c1"; }
+.bi-clouds-fill::before { content: "\f2c2"; }
+.bi-clouds::before { content: "\f2c3"; }
+.bi-cloudy-fill::before { content: "\f2c4"; }
+.bi-cloudy::before { content: "\f2c5"; }
+.bi-code-slash::before { content: "\f2c6"; }
+.bi-code-square::before { content: "\f2c7"; }
+.bi-code::before { content: "\f2c8"; }
+.bi-collection-fill::before { content: "\f2c9"; }
+.bi-collection-play-fill::before { content: "\f2ca"; }
+.bi-collection-play::before { content: "\f2cb"; }
+.bi-collection::before { content: "\f2cc"; }
+.bi-columns-gap::before { content: "\f2cd"; }
+.bi-columns::before { content: "\f2ce"; }
+.bi-command::before { content: "\f2cf"; }
+.bi-compass-fill::before { content: "\f2d0"; }
+.bi-compass::before { content: "\f2d1"; }
+.bi-cone-striped::before { content: "\f2d2"; }
+.bi-cone::before { content: "\f2d3"; }
+.bi-controller::before { content: "\f2d4"; }
+.bi-cpu-fill::before { content: "\f2d5"; }
+.bi-cpu::before { content: "\f2d6"; }
+.bi-credit-card-2-back-fill::before { content: "\f2d7"; }
+.bi-credit-card-2-back::before { content: "\f2d8"; }
+.bi-credit-card-2-front-fill::before { content: "\f2d9"; }
+.bi-credit-card-2-front::before { content: "\f2da"; }
+.bi-credit-card-fill::before { content: "\f2db"; }
+.bi-credit-card::before { content: "\f2dc"; }
+.bi-crop::before { content: "\f2dd"; }
+.bi-cup-fill::before { content: "\f2de"; }
+.bi-cup-straw::before { content: "\f2df"; }
+.bi-cup::before { content: "\f2e0"; }
+.bi-cursor-fill::before { content: "\f2e1"; }
+.bi-cursor-text::before { content: "\f2e2"; }
+.bi-cursor::before { content: "\f2e3"; }
+.bi-dash-circle-dotted::before { content: "\f2e4"; }
+.bi-dash-circle-fill::before { content: "\f2e5"; }
+.bi-dash-circle::before { content: "\f2e6"; }
+.bi-dash-square-dotted::before { content: "\f2e7"; }
+.bi-dash-square-fill::before { content: "\f2e8"; }
+.bi-dash-square::before { content: "\f2e9"; }
+.bi-dash::before { content: "\f2ea"; }
+.bi-diagram-2-fill::before { content: "\f2eb"; }
+.bi-diagram-2::before { content: "\f2ec"; }
+.bi-diagram-3-fill::before { content: "\f2ed"; }
+.bi-diagram-3::before { content: "\f2ee"; }
+.bi-diamond-fill::before { content: "\f2ef"; }
+.bi-diamond-half::before { content: "\f2f0"; }
+.bi-diamond::before { content: "\f2f1"; }
+.bi-dice-1-fill::before { content: "\f2f2"; }
+.bi-dice-1::before { content: "\f2f3"; }
+.bi-dice-2-fill::before { content: "\f2f4"; }
+.bi-dice-2::before { content: "\f2f5"; }
+.bi-dice-3-fill::before { content: "\f2f6"; }
+.bi-dice-3::before { content: "\f2f7"; }
+.bi-dice-4-fill::before { content: "\f2f8"; }
+.bi-dice-4::before { content: "\f2f9"; }
+.bi-dice-5-fill::before { content: "\f2fa"; }
+.bi-dice-5::before { content: "\f2fb"; }
+.bi-dice-6-fill::before { content: "\f2fc"; }
+.bi-dice-6::before { content: "\f2fd"; }
+.bi-disc-fill::before { content: "\f2fe"; }
+.bi-disc::before { content: "\f2ff"; }
+.bi-discord::before { content: "\f300"; }
+.bi-display-fill::before { content: "\f301"; }
+.bi-display::before { content: "\f302"; }
+.bi-distribute-horizontal::before { content: "\f303"; }
+.bi-distribute-vertical::before { content: "\f304"; }
+.bi-door-closed-fill::before { content: "\f305"; }
+.bi-door-closed::before { content: "\f306"; }
+.bi-door-open-fill::before { content: "\f307"; }
+.bi-door-open::before { content: "\f308"; }
+.bi-dot::before { content: "\f309"; }
+.bi-download::before { content: "\f30a"; }
+.bi-droplet-fill::before { content: "\f30b"; }
+.bi-droplet-half::before { content: "\f30c"; }
+.bi-droplet::before { content: "\f30d"; }
+.bi-earbuds::before { content: "\f30e"; }
+.bi-easel-fill::before { content: "\f30f"; }
+.bi-easel::before { content: "\f310"; }
+.bi-egg-fill::before { content: "\f311"; }
+.bi-egg-fried::before { content: "\f312"; }
+.bi-egg::before { content: "\f313"; }
+.bi-eject-fill::before { content: "\f314"; }
+.bi-eject::before { content: "\f315"; }
+.bi-emoji-angry-fill::before { content: "\f316"; }
+.bi-emoji-angry::before { content: "\f317"; }
+.bi-emoji-dizzy-fill::before { content: "\f318"; }
+.bi-emoji-dizzy::before { content: "\f319"; }
+.bi-emoji-expressionless-fill::before { content: "\f31a"; }
+.bi-emoji-expressionless::before { content: "\f31b"; }
+.bi-emoji-frown-fill::before { content: "\f31c"; }
+.bi-emoji-frown::before { content: "\f31d"; }
+.bi-emoji-heart-eyes-fill::before { content: "\f31e"; }
+.bi-emoji-heart-eyes::before { content: "\f31f"; }
+.bi-emoji-laughing-fill::before { content: "\f320"; }
+.bi-emoji-laughing::before { content: "\f321"; }
+.bi-emoji-neutral-fill::before { content: "\f322"; }
+.bi-emoji-neutral::before { content: "\f323"; }
+.bi-emoji-smile-fill::before { content: "\f324"; }
+.bi-emoji-smile-upside-down-fill::before { content: "\f325"; }
+.bi-emoji-smile-upside-down::before { content: "\f326"; }
+.bi-emoji-smile::before { content: "\f327"; }
+.bi-emoji-sunglasses-fill::before { content: "\f328"; }
+.bi-emoji-sunglasses::before { content: "\f329"; }
+.bi-emoji-wink-fill::before { content: "\f32a"; }
+.bi-emoji-wink::before { content: "\f32b"; }
+.bi-envelope-fill::before { content: "\f32c"; }
+.bi-envelope-open-fill::before { content: "\f32d"; }
+.bi-envelope-open::before { content: "\f32e"; }
+.bi-envelope::before { content: "\f32f"; }
+.bi-eraser-fill::before { content: "\f330"; }
+.bi-eraser::before { content: "\f331"; }
+.bi-exclamation-circle-fill::before { content: "\f332"; }
+.bi-exclamation-circle::before { content: "\f333"; }
+.bi-exclamation-diamond-fill::before { content: "\f334"; }
+.bi-exclamation-diamond::before { content: "\f335"; }
+.bi-exclamation-octagon-fill::before { content: "\f336"; }
+.bi-exclamation-octagon::before { content: "\f337"; }
+.bi-exclamation-square-fill::before { content: "\f338"; }
+.bi-exclamation-square::before { content: "\f339"; }
+.bi-exclamation-triangle-fill::before { content: "\f33a"; }
+.bi-exclamation-triangle::before { content: "\f33b"; }
+.bi-exclamation::before { content: "\f33c"; }
+.bi-exclude::before { content: "\f33d"; }
+.bi-eye-fill::before { content: "\f33e"; }
+.bi-eye-slash-fill::before { content: "\f33f"; }
+.bi-eye-slash::before { content: "\f340"; }
+.bi-eye::before { content: "\f341"; }
+.bi-eyedropper::before { content: "\f342"; }
+.bi-eyeglasses::before { content: "\f343"; }
+.bi-facebook::before { content: "\f344"; }
+.bi-file-arrow-down-fill::before { content: "\f345"; }
+.bi-file-arrow-down::before { content: "\f346"; }
+.bi-file-arrow-up-fill::before { content: "\f347"; }
+.bi-file-arrow-up::before { content: "\f348"; }
+.bi-file-bar-graph-fill::before { content: "\f349"; }
+.bi-file-bar-graph::before { content: "\f34a"; }
+.bi-file-binary-fill::before { content: "\f34b"; }
+.bi-file-binary::before { content: "\f34c"; }
+.bi-file-break-fill::before { content: "\f34d"; }
+.bi-file-break::before { content: "\f34e"; }
+.bi-file-check-fill::before { content: "\f34f"; }
+.bi-file-check::before { content: "\f350"; }
+.bi-file-code-fill::before { content: "\f351"; }
+.bi-file-code::before { content: "\f352"; }
+.bi-file-diff-fill::before { content: "\f353"; }
+.bi-file-diff::before { content: "\f354"; }
+.bi-file-earmark-arrow-down-fill::before { content: "\f355"; }
+.bi-file-earmark-arrow-down::before { content: "\f356"; }
+.bi-file-earmark-arrow-up-fill::before { content: "\f357"; }
+.bi-file-earmark-arrow-up::before { content: "\f358"; }
+.bi-file-earmark-bar-graph-fill::before { content: "\f359"; }
+.bi-file-earmark-bar-graph::before { content: "\f35a"; }
+.bi-file-earmark-binary-fill::before { content: "\f35b"; }
+.bi-file-earmark-binary::before { content: "\f35c"; }
+.bi-file-earmark-break-fill::before { content: "\f35d"; }
+.bi-file-earmark-break::before { content: "\f35e"; }
+.bi-file-earmark-check-fill::before { content: "\f35f"; }
+.bi-file-earmark-check::before { content: "\f360"; }
+.bi-file-earmark-code-fill::before { content: "\f361"; }
+.bi-file-earmark-code::before { content: "\f362"; }
+.bi-file-earmark-diff-fill::before { content: "\f363"; }
+.bi-file-earmark-diff::before { content: "\f364"; }
+.bi-file-earmark-easel-fill::before { content: "\f365"; }
+.bi-file-earmark-easel::before { content: "\f366"; }
+.bi-file-earmark-excel-fill::before { content: "\f367"; }
+.bi-file-earmark-excel::before { content: "\f368"; }
+.bi-file-earmark-fill::before { content: "\f369"; }
+.bi-file-earmark-font-fill::before { content: "\f36a"; }
+.bi-file-earmark-font::before { content: "\f36b"; }
+.bi-file-earmark-image-fill::before { content: "\f36c"; }
+.bi-file-earmark-image::before { content: "\f36d"; }
+.bi-file-earmark-lock-fill::before { content: "\f36e"; }
+.bi-file-earmark-lock::before { content: "\f36f"; }
+.bi-file-earmark-lock2-fill::before { content: "\f370"; }
+.bi-file-earmark-lock2::before { content: "\f371"; }
+.bi-file-earmark-medical-fill::before { content: "\f372"; }
+.bi-file-earmark-medical::before { content: "\f373"; }
+.bi-file-earmark-minus-fill::before { content: "\f374"; }
+.bi-file-earmark-minus::before { content: "\f375"; }
+.bi-file-earmark-music-fill::before { content: "\f376"; }
+.bi-file-earmark-music::before { content: "\f377"; }
+.bi-file-earmark-person-fill::before { content: "\f378"; }
+.bi-file-earmark-person::before { content: "\f379"; }
+.bi-file-earmark-play-fill::before { content: "\f37a"; }
+.bi-file-earmark-play::before { content: "\f37b"; }
+.bi-file-earmark-plus-fill::before { content: "\f37c"; }
+.bi-file-earmark-plus::before { content: "\f37d"; }
+.bi-file-earmark-post-fill::before { content: "\f37e"; }
+.bi-file-earmark-post::before { content: "\f37f"; }
+.bi-file-earmark-ppt-fill::before { content: "\f380"; }
+.bi-file-earmark-ppt::before { content: "\f381"; }
+.bi-file-earmark-richtext-fill::before { content: "\f382"; }
+.bi-file-earmark-richtext::before { content: "\f383"; }
+.bi-file-earmark-ruled-fill::before { content: "\f384"; }
+.bi-file-earmark-ruled::before { content: "\f385"; }
+.bi-file-earmark-slides-fill::before { content: "\f386"; }
+.bi-file-earmark-slides::before { content: "\f387"; }
+.bi-file-earmark-spreadsheet-fill::before { content: "\f388"; }
+.bi-file-earmark-spreadsheet::before { content: "\f389"; }
+.bi-file-earmark-text-fill::before { content: "\f38a"; }
+.bi-file-earmark-text::before { content: "\f38b"; }
+.bi-file-earmark-word-fill::before { content: "\f38c"; }
+.bi-file-earmark-word::before { content: "\f38d"; }
+.bi-file-earmark-x-fill::before { content: "\f38e"; }
+.bi-file-earmark-x::before { content: "\f38f"; }
+.bi-file-earmark-zip-fill::before { content: "\f390"; }
+.bi-file-earmark-zip::before { content: "\f391"; }
+.bi-file-earmark::before { content: "\f392"; }
+.bi-file-easel-fill::before { content: "\f393"; }
+.bi-file-easel::before { content: "\f394"; }
+.bi-file-excel-fill::before { content: "\f395"; }
+.bi-file-excel::before { content: "\f396"; }
+.bi-file-fill::before { content: "\f397"; }
+.bi-file-font-fill::before { content: "\f398"; }
+.bi-file-font::before { content: "\f399"; }
+.bi-file-image-fill::before { content: "\f39a"; }
+.bi-file-image::before { content: "\f39b"; }
+.bi-file-lock-fill::before { content: "\f39c"; }
+.bi-file-lock::before { content: "\f39d"; }
+.bi-file-lock2-fill::before { content: "\f39e"; }
+.bi-file-lock2::before { content: "\f39f"; }
+.bi-file-medical-fill::before { content: "\f3a0"; }
+.bi-file-medical::before { content: "\f3a1"; }
+.bi-file-minus-fill::before { content: "\f3a2"; }
+.bi-file-minus::before { content: "\f3a3"; }
+.bi-file-music-fill::before { content: "\f3a4"; }
+.bi-file-music::before { content: "\f3a5"; }
+.bi-file-person-fill::before { content: "\f3a6"; }
+.bi-file-person::before { content: "\f3a7"; }
+.bi-file-play-fill::before { content: "\f3a8"; }
+.bi-file-play::before { content: "\f3a9"; }
+.bi-file-plus-fill::before { content: "\f3aa"; }
+.bi-file-plus::before { content: "\f3ab"; }
+.bi-file-post-fill::before { content: "\f3ac"; }
+.bi-file-post::before { content: "\f3ad"; }
+.bi-file-ppt-fill::before { content: "\f3ae"; }
+.bi-file-ppt::before { content: "\f3af"; }
+.bi-file-richtext-fill::before { content: "\f3b0"; }
+.bi-file-richtext::before { content: "\f3b1"; }
+.bi-file-ruled-fill::before { content: "\f3b2"; }
+.bi-file-ruled::before { content: "\f3b3"; }
+.bi-file-slides-fill::before { content: "\f3b4"; }
+.bi-file-slides::before { content: "\f3b5"; }
+.bi-file-spreadsheet-fill::before { content: "\f3b6"; }
+.bi-file-spreadsheet::before { content: "\f3b7"; }
+.bi-file-text-fill::before { content: "\f3b8"; }
+.bi-file-text::before { content: "\f3b9"; }
+.bi-file-word-fill::before { content: "\f3ba"; }
+.bi-file-word::before { content: "\f3bb"; }
+.bi-file-x-fill::before { content: "\f3bc"; }
+.bi-file-x::before { content: "\f3bd"; }
+.bi-file-zip-fill::before { content: "\f3be"; }
+.bi-file-zip::before { content: "\f3bf"; }
+.bi-file::before { content: "\f3c0"; }
+.bi-files-alt::before { content: "\f3c1"; }
+.bi-files::before { content: "\f3c2"; }
+.bi-film::before { content: "\f3c3"; }
+.bi-filter-circle-fill::before { content: "\f3c4"; }
+.bi-filter-circle::before { content: "\f3c5"; }
+.bi-filter-left::before { content: "\f3c6"; }
+.bi-filter-right::before { content: "\f3c7"; }
+.bi-filter-square-fill::before { content: "\f3c8"; }
+.bi-filter-square::before { content: "\f3c9"; }
+.bi-filter::before { content: "\f3ca"; }
+.bi-flag-fill::before { content: "\f3cb"; }
+.bi-flag::before { content: "\f3cc"; }
+.bi-flower1::before { content: "\f3cd"; }
+.bi-flower2::before { content: "\f3ce"; }
+.bi-flower3::before { content: "\f3cf"; }
+.bi-folder-check::before { content: "\f3d0"; }
+.bi-folder-fill::before { content: "\f3d1"; }
+.bi-folder-minus::before { content: "\f3d2"; }
+.bi-folder-plus::before { content: "\f3d3"; }
+.bi-folder-symlink-fill::before { content: "\f3d4"; }
+.bi-folder-symlink::before { content: "\f3d5"; }
+.bi-folder-x::before { content: "\f3d6"; }
+.bi-folder::before { content: "\f3d7"; }
+.bi-folder2-open::before { content: "\f3d8"; }
+.bi-folder2::before { content: "\f3d9"; }
+.bi-fonts::before { content: "\f3da"; }
+.bi-forward-fill::before { content: "\f3db"; }
+.bi-forward::before { content: "\f3dc"; }
+.bi-front::before { content: "\f3dd"; }
+.bi-fullscreen-exit::before { content: "\f3de"; }
+.bi-fullscreen::before { content: "\f3df"; }
+.bi-funnel-fill::before { content: "\f3e0"; }
+.bi-funnel::before { content: "\f3e1"; }
+.bi-gear-fill::before { content: "\f3e2"; }
+.bi-gear-wide-connected::before { content: "\f3e3"; }
+.bi-gear-wide::before { content: "\f3e4"; }
+.bi-gear::before { content: "\f3e5"; }
+.bi-gem::before { content: "\f3e6"; }
+.bi-geo-alt-fill::before { content: "\f3e7"; }
+.bi-geo-alt::before { content: "\f3e8"; }
+.bi-geo-fill::before { content: "\f3e9"; }
+.bi-geo::before { content: "\f3ea"; }
+.bi-gift-fill::before { content: "\f3eb"; }
+.bi-gift::before { content: "\f3ec"; }
+.bi-github::before { content: "\f3ed"; }
+.bi-globe::before { content: "\f3ee"; }
+.bi-globe2::before { content: "\f3ef"; }
+.bi-google::before { content: "\f3f0"; }
+.bi-graph-down::before { content: "\f3f1"; }
+.bi-graph-up::before { content: "\f3f2"; }
+.bi-grid-1x2-fill::before { content: "\f3f3"; }
+.bi-grid-1x2::before { content: "\f3f4"; }
+.bi-grid-3x2-gap-fill::before { content: "\f3f5"; }
+.bi-grid-3x2-gap::before { content: "\f3f6"; }
+.bi-grid-3x2::before { content: "\f3f7"; }
+.bi-grid-3x3-gap-fill::before { content: "\f3f8"; }
+.bi-grid-3x3-gap::before { content: "\f3f9"; }
+.bi-grid-3x3::before { content: "\f3fa"; }
+.bi-grid-fill::before { content: "\f3fb"; }
+.bi-grid::before { content: "\f3fc"; }
+.bi-grip-horizontal::before { content: "\f3fd"; }
+.bi-grip-vertical::before { content: "\f3fe"; }
+.bi-hammer::before { content: "\f3ff"; }
+.bi-hand-index-fill::before { content: "\f400"; }
+.bi-hand-index-thumb-fill::before { content: "\f401"; }
+.bi-hand-index-thumb::before { content: "\f402"; }
+.bi-hand-index::before { content: "\f403"; }
+.bi-hand-thumbs-down-fill::before { content: "\f404"; }
+.bi-hand-thumbs-down::before { content: "\f405"; }
+.bi-hand-thumbs-up-fill::before { content: "\f406"; }
+.bi-hand-thumbs-up::before { content: "\f407"; }
+.bi-handbag-fill::before { content: "\f408"; }
+.bi-handbag::before { content: "\f409"; }
+.bi-hash::before { content: "\f40a"; }
+.bi-hdd-fill::before { content: "\f40b"; }
+.bi-hdd-network-fill::before { content: "\f40c"; }
+.bi-hdd-network::before { content: "\f40d"; }
+.bi-hdd-rack-fill::before { content: "\f40e"; }
+.bi-hdd-rack::before { content: "\f40f"; }
+.bi-hdd-stack-fill::before { content: "\f410"; }
+.bi-hdd-stack::before { content: "\f411"; }
+.bi-hdd::before { content: "\f412"; }
+.bi-headphones::before { content: "\f413"; }
+.bi-headset::before { content: "\f414"; }
+.bi-heart-fill::before { content: "\f415"; }
+.bi-heart-half::before { content: "\f416"; }
+.bi-heart::before { content: "\f417"; }
+.bi-heptagon-fill::before { content: "\f418"; }
+.bi-heptagon-half::before { content: "\f419"; }
+.bi-heptagon::before { content: "\f41a"; }
+.bi-hexagon-fill::before { content: "\f41b"; }
+.bi-hexagon-half::before { content: "\f41c"; }
+.bi-hexagon::before { content: "\f41d"; }
+.bi-hourglass-bottom::before { content: "\f41e"; }
+.bi-hourglass-split::before { content: "\f41f"; }
+.bi-hourglass-top::before { content: "\f420"; }
+.bi-hourglass::before { content: "\f421"; }
+.bi-house-door-fill::before { content: "\f422"; }
+.bi-house-door::before { content: "\f423"; }
+.bi-house-fill::before { content: "\f424"; }
+.bi-house::before { content: "\f425"; }
+.bi-hr::before { content: "\f426"; }
+.bi-hurricane::before { content: "\f427"; }
+.bi-image-alt::before { content: "\f428"; }
+.bi-image-fill::before { content: "\f429"; }
+.bi-image::before { content: "\f42a"; }
+.bi-images::before { content: "\f42b"; }
+.bi-inbox-fill::before { content: "\f42c"; }
+.bi-inbox::before { content: "\f42d"; }
+.bi-inboxes-fill::before { content: "\f42e"; }
+.bi-inboxes::before { content: "\f42f"; }
+.bi-info-circle-fill::before { content: "\f430"; }
+.bi-info-circle::before { content: "\f431"; }
+.bi-info-square-fill::before { content: "\f432"; }
+.bi-info-square::before { content: "\f433"; }
+.bi-info::before { content: "\f434"; }
+.bi-input-cursor-text::before { content: "\f435"; }
+.bi-input-cursor::before { content: "\f436"; }
+.bi-instagram::before { content: "\f437"; }
+.bi-intersect::before { content: "\f438"; }
+.bi-journal-album::before { content: "\f439"; }
+.bi-journal-arrow-down::before { content: "\f43a"; }
+.bi-journal-arrow-up::before { content: "\f43b"; }
+.bi-journal-bookmark-fill::before { content: "\f43c"; }
+.bi-journal-bookmark::before { content: "\f43d"; }
+.bi-journal-check::before { content: "\f43e"; }
+.bi-journal-code::before { content: "\f43f"; }
+.bi-journal-medical::before { content: "\f440"; }
+.bi-journal-minus::before { content: "\f441"; }
+.bi-journal-plus::before { content: "\f442"; }
+.bi-journal-richtext::before { content: "\f443"; }
+.bi-journal-text::before { content: "\f444"; }
+.bi-journal-x::before { content: "\f445"; }
+.bi-journal::before { content: "\f446"; }
+.bi-journals::before { content: "\f447"; }
+.bi-joystick::before { content: "\f448"; }
+.bi-justify-left::before { content: "\f449"; }
+.bi-justify-right::before { content: "\f44a"; }
+.bi-justify::before { content: "\f44b"; }
+.bi-kanban-fill::before { content: "\f44c"; }
+.bi-kanban::before { content: "\f44d"; }
+.bi-key-fill::before { content: "\f44e"; }
+.bi-key::before { content: "\f44f"; }
+.bi-keyboard-fill::before { content: "\f450"; }
+.bi-keyboard::before { content: "\f451"; }
+.bi-ladder::before { content: "\f452"; }
+.bi-lamp-fill::before { content: "\f453"; }
+.bi-lamp::before { content: "\f454"; }
+.bi-laptop-fill::before { content: "\f455"; }
+.bi-laptop::before { content: "\f456"; }
+.bi-layer-backward::before { content: "\f457"; }
+.bi-layer-forward::before { content: "\f458"; }
+.bi-layers-fill::before { content: "\f459"; }
+.bi-layers-half::before { content: "\f45a"; }
+.bi-layers::before { content: "\f45b"; }
+.bi-layout-sidebar-inset-reverse::before { content: "\f45c"; }
+.bi-layout-sidebar-inset::before { content: "\f45d"; }
+.bi-layout-sidebar-reverse::before { content: "\f45e"; }
+.bi-layout-sidebar::before { content: "\f45f"; }
+.bi-layout-split::before { content: "\f460"; }
+.bi-layout-text-sidebar-reverse::before { content: "\f461"; }
+.bi-layout-text-sidebar::before { content: "\f462"; }
+.bi-layout-text-window-reverse::before { content: "\f463"; }
+.bi-layout-text-window::before { content: "\f464"; }
+.bi-layout-three-columns::before { content: "\f465"; }
+.bi-layout-wtf::before { content: "\f466"; }
+.bi-life-preserver::before { content: "\f467"; }
+.bi-lightbulb-fill::before { content: "\f468"; }
+.bi-lightbulb-off-fill::before { content: "\f469"; }
+.bi-lightbulb-off::before { content: "\f46a"; }
+.bi-lightbulb::before { content: "\f46b"; }
+.bi-lightning-charge-fill::before { content: "\f46c"; }
+.bi-lightning-charge::before { content: "\f46d"; }
+.bi-lightning-fill::before { content: "\f46e"; }
+.bi-lightning::before { content: "\f46f"; }
+.bi-link-45deg::before { content: "\f470"; }
+.bi-link::before { content: "\f471"; }
+.bi-linkedin::before { content: "\f472"; }
+.bi-list-check::before { content: "\f473"; }
+.bi-list-nested::before { content: "\f474"; }
+.bi-list-ol::before { content: "\f475"; }
+.bi-list-stars::before { content: "\f476"; }
+.bi-list-task::before { content: "\f477"; }
+.bi-list-ul::before { content: "\f478"; }
+.bi-list::before { content: "\f479"; }
+.bi-lock-fill::before { content: "\f47a"; }
+.bi-lock::before { content: "\f47b"; }
+.bi-mailbox::before { content: "\f47c"; }
+.bi-mailbox2::before { content: "\f47d"; }
+.bi-map-fill::before { content: "\f47e"; }
+.bi-map::before { content: "\f47f"; }
+.bi-markdown-fill::before { content: "\f480"; }
+.bi-markdown::before { content: "\f481"; }
+.bi-mask::before { content: "\f482"; }
+.bi-megaphone-fill::before { content: "\f483"; }
+.bi-megaphone::before { content: "\f484"; }
+.bi-menu-app-fill::before { content: "\f485"; }
+.bi-menu-app::before { content: "\f486"; }
+.bi-menu-button-fill::before { content: "\f487"; }
+.bi-menu-button-wide-fill::before { content: "\f488"; }
+.bi-menu-button-wide::before { content: "\f489"; }
+.bi-menu-button::before { content: "\f48a"; }
+.bi-menu-down::before { content: "\f48b"; }
+.bi-menu-up::before { content: "\f48c"; }
+.bi-mic-fill::before { content: "\f48d"; }
+.bi-mic-mute-fill::before { content: "\f48e"; }
+.bi-mic-mute::before { content: "\f48f"; }
+.bi-mic::before { content: "\f490"; }
+.bi-minecart-loaded::before { content: "\f491"; }
+.bi-minecart::before { content: "\f492"; }
+.bi-moisture::before { content: "\f493"; }
+.bi-moon-fill::before { content: "\f494"; }
+.bi-moon-stars-fill::before { content: "\f495"; }
+.bi-moon-stars::before { content: "\f496"; }
+.bi-moon::before { content: "\f497"; }
+.bi-mouse-fill::before { content: "\f498"; }
+.bi-mouse::before { content: "\f499"; }
+.bi-mouse2-fill::before { content: "\f49a"; }
+.bi-mouse2::before { content: "\f49b"; }
+.bi-mouse3-fill::before { content: "\f49c"; }
+.bi-mouse3::before { content: "\f49d"; }
+.bi-music-note-beamed::before { content: "\f49e"; }
+.bi-music-note-list::before { content: "\f49f"; }
+.bi-music-note::before { content: "\f4a0"; }
+.bi-music-player-fill::before { content: "\f4a1"; }
+.bi-music-player::before { content: "\f4a2"; }
+.bi-newspaper::before { content: "\f4a3"; }
+.bi-node-minus-fill::before { content: "\f4a4"; }
+.bi-node-minus::before { content: "\f4a5"; }
+.bi-node-plus-fill::before { content: "\f4a6"; }
+.bi-node-plus::before { content: "\f4a7"; }
+.bi-nut-fill::before { content: "\f4a8"; }
+.bi-nut::before { content: "\f4a9"; }
+.bi-octagon-fill::before { content: "\f4aa"; }
+.bi-octagon-half::before { content: "\f4ab"; }
+.bi-octagon::before { content: "\f4ac"; }
+.bi-option::before { content: "\f4ad"; }
+.bi-outlet::before { content: "\f4ae"; }
+.bi-paint-bucket::before { content: "\f4af"; }
+.bi-palette-fill::before { content: "\f4b0"; }
+.bi-palette::before { content: "\f4b1"; }
+.bi-palette2::before { content: "\f4b2"; }
+.bi-paperclip::before { content: "\f4b3"; }
+.bi-paragraph::before { content: "\f4b4"; }
+.bi-patch-check-fill::before { content: "\f4b5"; }
+.bi-patch-check::before { content: "\f4b6"; }
+.bi-patch-exclamation-fill::before { content: "\f4b7"; }
+.bi-patch-exclamation::before { content: "\f4b8"; }
+.bi-patch-minus-fill::before { content: "\f4b9"; }
+.bi-patch-minus::before { content: "\f4ba"; }
+.bi-patch-plus-fill::before { content: "\f4bb"; }
+.bi-patch-plus::before { content: "\f4bc"; }
+.bi-patch-question-fill::before { content: "\f4bd"; }
+.bi-patch-question::before { content: "\f4be"; }
+.bi-pause-btn-fill::before { content: "\f4bf"; }
+.bi-pause-btn::before { content: "\f4c0"; }
+.bi-pause-circle-fill::before { content: "\f4c1"; }
+.bi-pause-circle::before { content: "\f4c2"; }
+.bi-pause-fill::before { content: "\f4c3"; }
+.bi-pause::before { content: "\f4c4"; }
+.bi-peace-fill::before { content: "\f4c5"; }
+.bi-peace::before { content: "\f4c6"; }
+.bi-pen-fill::before { content: "\f4c7"; }
+.bi-pen::before { content: "\f4c8"; }
+.bi-pencil-fill::before { content: "\f4c9"; }
+.bi-pencil-square::before { content: "\f4ca"; }
+.bi-pencil::before { content: "\f4cb"; }
+.bi-pentagon-fill::before { content: "\f4cc"; }
+.bi-pentagon-half::before { content: "\f4cd"; }
+.bi-pentagon::before { content: "\f4ce"; }
+.bi-people-fill::before { content: "\f4cf"; }
+.bi-people::before { content: "\f4d0"; }
+.bi-percent::before { content: "\f4d1"; }
+.bi-person-badge-fill::before { content: "\f4d2"; }
+.bi-person-badge::before { content: "\f4d3"; }
+.bi-person-bounding-box::before { content: "\f4d4"; }
+.bi-person-check-fill::before { content: "\f4d5"; }
+.bi-person-check::before { content: "\f4d6"; }
+.bi-person-circle::before { content: "\f4d7"; }
+.bi-person-dash-fill::before { content: "\f4d8"; }
+.bi-person-dash::before { content: "\f4d9"; }
+.bi-person-fill::before { content: "\f4da"; }
+.bi-person-lines-fill::before { content: "\f4db"; }
+.bi-person-plus-fill::before { content: "\f4dc"; }
+.bi-person-plus::before { content: "\f4dd"; }
+.bi-person-square::before { content: "\f4de"; }
+.bi-person-x-fill::before { content: "\f4df"; }
+.bi-person-x::before { content: "\f4e0"; }
+.bi-person::before { content: "\f4e1"; }
+.bi-phone-fill::before { content: "\f4e2"; }
+.bi-phone-landscape-fill::before { content: "\f4e3"; }
+.bi-phone-landscape::before { content: "\f4e4"; }
+.bi-phone-vibrate-fill::before { content: "\f4e5"; }
+.bi-phone-vibrate::before { content: "\f4e6"; }
+.bi-phone::before { content: "\f4e7"; }
+.bi-pie-chart-fill::before { content: "\f4e8"; }
+.bi-pie-chart::before { content: "\f4e9"; }
+.bi-pin-angle-fill::before { content: "\f4ea"; }
+.bi-pin-angle::before { content: "\f4eb"; }
+.bi-pin-fill::before { content: "\f4ec"; }
+.bi-pin::before { content: "\f4ed"; }
+.bi-pip-fill::before { content: "\f4ee"; }
+.bi-pip::before { content: "\f4ef"; }
+.bi-play-btn-fill::before { content: "\f4f0"; }
+.bi-play-btn::before { content: "\f4f1"; }
+.bi-play-circle-fill::before { content: "\f4f2"; }
+.bi-play-circle::before { content: "\f4f3"; }
+.bi-play-fill::before { content: "\f4f4"; }
+.bi-play::before { content: "\f4f5"; }
+.bi-plug-fill::before { content: "\f4f6"; }
+.bi-plug::before { content: "\f4f7"; }
+.bi-plus-circle-dotted::before { content: "\f4f8"; }
+.bi-plus-circle-fill::before { content: "\f4f9"; }
+.bi-plus-circle::before { content: "\f4fa"; }
+.bi-plus-square-dotted::before { content: "\f4fb"; }
+.bi-plus-square-fill::before { content: "\f4fc"; }
+.bi-plus-square::before { content: "\f4fd"; }
+.bi-plus::before { content: "\f4fe"; }
+.bi-power::before { content: "\f4ff"; }
+.bi-printer-fill::before { content: "\f500"; }
+.bi-printer::before { content: "\f501"; }
+.bi-puzzle-fill::before { content: "\f502"; }
+.bi-puzzle::before { content: "\f503"; }
+.bi-question-circle-fill::before { content: "\f504"; }
+.bi-question-circle::before { content: "\f505"; }
+.bi-question-diamond-fill::before { content: "\f506"; }
+.bi-question-diamond::before { content: "\f507"; }
+.bi-question-octagon-fill::before { content: "\f508"; }
+.bi-question-octagon::before { content: "\f509"; }
+.bi-question-square-fill::before { content: "\f50a"; }
+.bi-question-square::before { content: "\f50b"; }
+.bi-question::before { content: "\f50c"; }
+.bi-rainbow::before { content: "\f50d"; }
+.bi-receipt-cutoff::before { content: "\f50e"; }
+.bi-receipt::before { content: "\f50f"; }
+.bi-reception-0::before { content: "\f510"; }
+.bi-reception-1::before { content: "\f511"; }
+.bi-reception-2::before { content: "\f512"; }
+.bi-reception-3::before { content: "\f513"; }
+.bi-reception-4::before { content: "\f514"; }
+.bi-record-btn-fill::before { content: "\f515"; }
+.bi-record-btn::before { content: "\f516"; }
+.bi-record-circle-fill::before { content: "\f517"; }
+.bi-record-circle::before { content: "\f518"; }
+.bi-record-fill::before { content: "\f519"; }
+.bi-record::before { content: "\f51a"; }
+.bi-record2-fill::before { content: "\f51b"; }
+.bi-record2::before { content: "\f51c"; }
+.bi-reply-all-fill::before { content: "\f51d"; }
+.bi-reply-all::before { content: "\f51e"; }
+.bi-reply-fill::before { content: "\f51f"; }
+.bi-reply::before { content: "\f520"; }
+.bi-rss-fill::before { content: "\f521"; }
+.bi-rss::before { content: "\f522"; }
+.bi-rulers::before { content: "\f523"; }
+.bi-save-fill::before { content: "\f524"; }
+.bi-save::before { content: "\f525"; }
+.bi-save2-fill::before { content: "\f526"; }
+.bi-save2::before { content: "\f527"; }
+.bi-scissors::before { content: "\f528"; }
+.bi-screwdriver::before { content: "\f529"; }
+.bi-search::before { content: "\f52a"; }
+.bi-segmented-nav::before { content: "\f52b"; }
+.bi-server::before { content: "\f52c"; }
+.bi-share-fill::before { content: "\f52d"; }
+.bi-share::before { content: "\f52e"; }
+.bi-shield-check::before { content: "\f52f"; }
+.bi-shield-exclamation::before { content: "\f530"; }
+.bi-shield-fill-check::before { content: "\f531"; }
+.bi-shield-fill-exclamation::before { content: "\f532"; }
+.bi-shield-fill-minus::before { content: "\f533"; }
+.bi-shield-fill-plus::before { content: "\f534"; }
+.bi-shield-fill-x::before { content: "\f535"; }
+.bi-shield-fill::before { content: "\f536"; }
+.bi-shield-lock-fill::before { content: "\f537"; }
+.bi-shield-lock::before { content: "\f538"; }
+.bi-shield-minus::before { content: "\f539"; }
+.bi-shield-plus::before { content: "\f53a"; }
+.bi-shield-shaded::before { content: "\f53b"; }
+.bi-shield-slash-fill::before { content: "\f53c"; }
+.bi-shield-slash::before { content: "\f53d"; }
+.bi-shield-x::before { content: "\f53e"; }
+.bi-shield::before { content: "\f53f"; }
+.bi-shift-fill::before { content: "\f540"; }
+.bi-shift::before { content: "\f541"; }
+.bi-shop-window::before { content: "\f542"; }
+.bi-shop::before { content: "\f543"; }
+.bi-shuffle::before { content: "\f544"; }
+.bi-signpost-2-fill::before { content: "\f545"; }
+.bi-signpost-2::before { content: "\f546"; }
+.bi-signpost-fill::before { content: "\f547"; }
+.bi-signpost-split-fill::before { content: "\f548"; }
+.bi-signpost-split::before { content: "\f549"; }
+.bi-signpost::before { content: "\f54a"; }
+.bi-sim-fill::before { content: "\f54b"; }
+.bi-sim::before { content: "\f54c"; }
+.bi-skip-backward-btn-fill::before { content: "\f54d"; }
+.bi-skip-backward-btn::before { content: "\f54e"; }
+.bi-skip-backward-circle-fill::before { content: "\f54f"; }
+.bi-skip-backward-circle::before { content: "\f550"; }
+.bi-skip-backward-fill::before { content: "\f551"; }
+.bi-skip-backward::before { content: "\f552"; }
+.bi-skip-end-btn-fill::before { content: "\f553"; }
+.bi-skip-end-btn::before { content: "\f554"; }
+.bi-skip-end-circle-fill::before { content: "\f555"; }
+.bi-skip-end-circle::before { content: "\f556"; }
+.bi-skip-end-fill::before { content: "\f557"; }
+.bi-skip-end::before { content: "\f558"; }
+.bi-skip-forward-btn-fill::before { content: "\f559"; }
+.bi-skip-forward-btn::before { content: "\f55a"; }
+.bi-skip-forward-circle-fill::before { content: "\f55b"; }
+.bi-skip-forward-circle::before { content: "\f55c"; }
+.bi-skip-forward-fill::before { content: "\f55d"; }
+.bi-skip-forward::before { content: "\f55e"; }
+.bi-skip-start-btn-fill::before { content: "\f55f"; }
+.bi-skip-start-btn::before { content: "\f560"; }
+.bi-skip-start-circle-fill::before { content: "\f561"; }
+.bi-skip-start-circle::before { content: "\f562"; }
+.bi-skip-start-fill::before { content: "\f563"; }
+.bi-skip-start::before { content: "\f564"; }
+.bi-slack::before { content: "\f565"; }
+.bi-slash-circle-fill::before { content: "\f566"; }
+.bi-slash-circle::before { content: "\f567"; }
+.bi-slash-square-fill::before { content: "\f568"; }
+.bi-slash-square::before { content: "\f569"; }
+.bi-slash::before { content: "\f56a"; }
+.bi-sliders::before { content: "\f56b"; }
+.bi-smartwatch::before { content: "\f56c"; }
+.bi-snow::before { content: "\f56d"; }
+.bi-snow2::before { content: "\f56e"; }
+.bi-snow3::before { content: "\f56f"; }
+.bi-sort-alpha-down-alt::before { content: "\f570"; }
+.bi-sort-alpha-down::before { content: "\f571"; }
+.bi-sort-alpha-up-alt::before { content: "\f572"; }
+.bi-sort-alpha-up::before { content: "\f573"; }
+.bi-sort-down-alt::before { content: "\f574"; }
+.bi-sort-down::before { content: "\f575"; }
+.bi-sort-numeric-down-alt::before { content: "\f576"; }
+.bi-sort-numeric-down::before { content: "\f577"; }
+.bi-sort-numeric-up-alt::before { content: "\f578"; }
+.bi-sort-numeric-up::before { content: "\f579"; }
+.bi-sort-up-alt::before { content: "\f57a"; }
+.bi-sort-up::before { content: "\f57b"; }
+.bi-soundwave::before { content: "\f57c"; }
+.bi-speaker-fill::before { content: "\f57d"; }
+.bi-speaker::before { content: "\f57e"; }
+.bi-speedometer::before { content: "\f57f"; }
+.bi-speedometer2::before { content: "\f580"; }
+.bi-spellcheck::before { content: "\f581"; }
+.bi-square-fill::before { content: "\f582"; }
+.bi-square-half::before { content: "\f583"; }
+.bi-square::before { content: "\f584"; }
+.bi-stack::before { content: "\f585"; }
+.bi-star-fill::before { content: "\f586"; }
+.bi-star-half::before { content: "\f587"; }
+.bi-star::before { content: "\f588"; }
+.bi-stars::before { content: "\f589"; }
+.bi-stickies-fill::before { content: "\f58a"; }
+.bi-stickies::before { content: "\f58b"; }
+.bi-sticky-fill::before { content: "\f58c"; }
+.bi-sticky::before { content: "\f58d"; }
+.bi-stop-btn-fill::before { content: "\f58e"; }
+.bi-stop-btn::before { content: "\f58f"; }
+.bi-stop-circle-fill::before { content: "\f590"; }
+.bi-stop-circle::before { content: "\f591"; }
+.bi-stop-fill::before { content: "\f592"; }
+.bi-stop::before { content: "\f593"; }
+.bi-stoplights-fill::before { content: "\f594"; }
+.bi-stoplights::before { content: "\f595"; }
+.bi-stopwatch-fill::before { content: "\f596"; }
+.bi-stopwatch::before { content: "\f597"; }
+.bi-subtract::before { content: "\f598"; }
+.bi-suit-club-fill::before { content: "\f599"; }
+.bi-suit-club::before { content: "\f59a"; }
+.bi-suit-diamond-fill::before { content: "\f59b"; }
+.bi-suit-diamond::before { content: "\f59c"; }
+.bi-suit-heart-fill::before { content: "\f59d"; }
+.bi-suit-heart::before { content: "\f59e"; }
+.bi-suit-spade-fill::before { content: "\f59f"; }
+.bi-suit-spade::before { content: "\f5a0"; }
+.bi-sun-fill::before { content: "\f5a1"; }
+.bi-sun::before { content: "\f5a2"; }
+.bi-sunglasses::before { content: "\f5a3"; }
+.bi-sunrise-fill::before { content: "\f5a4"; }
+.bi-sunrise::before { content: "\f5a5"; }
+.bi-sunset-fill::before { content: "\f5a6"; }
+.bi-sunset::before { content: "\f5a7"; }
+.bi-symmetry-horizontal::before { content: "\f5a8"; }
+.bi-symmetry-vertical::before { content: "\f5a9"; }
+.bi-table::before { content: "\f5aa"; }
+.bi-tablet-fill::before { content: "\f5ab"; }
+.bi-tablet-landscape-fill::before { content: "\f5ac"; }
+.bi-tablet-landscape::before { content: "\f5ad"; }
+.bi-tablet::before { content: "\f5ae"; }
+.bi-tag-fill::before { content: "\f5af"; }
+.bi-tag::before { content: "\f5b0"; }
+.bi-tags-fill::before { content: "\f5b1"; }
+.bi-tags::before { content: "\f5b2"; }
+.bi-telegram::before { content: "\f5b3"; }
+.bi-telephone-fill::before { content: "\f5b4"; }
+.bi-telephone-forward-fill::before { content: "\f5b5"; }
+.bi-telephone-forward::before { content: "\f5b6"; }
+.bi-telephone-inbound-fill::before { content: "\f5b7"; }
+.bi-telephone-inbound::before { content: "\f5b8"; }
+.bi-telephone-minus-fill::before { content: "\f5b9"; }
+.bi-telephone-minus::before { content: "\f5ba"; }
+.bi-telephone-outbound-fill::before { content: "\f5bb"; }
+.bi-telephone-outbound::before { content: "\f5bc"; }
+.bi-telephone-plus-fill::before { content: "\f5bd"; }
+.bi-telephone-plus::before { content: "\f5be"; }
+.bi-telephone-x-fill::before { content: "\f5bf"; }
+.bi-telephone-x::before { content: "\f5c0"; }
+.bi-telephone::before { content: "\f5c1"; }
+.bi-terminal-fill::before { content: "\f5c2"; }
+.bi-terminal::before { content: "\f5c3"; }
+.bi-text-center::before { content: "\f5c4"; }
+.bi-text-indent-left::before { content: "\f5c5"; }
+.bi-text-indent-right::before { content: "\f5c6"; }
+.bi-text-left::before { content: "\f5c7"; }
+.bi-text-paragraph::before { content: "\f5c8"; }
+.bi-text-right::before { content: "\f5c9"; }
+.bi-textarea-resize::before { content: "\f5ca"; }
+.bi-textarea-t::before { content: "\f5cb"; }
+.bi-textarea::before { content: "\f5cc"; }
+.bi-thermometer-half::before { content: "\f5cd"; }
+.bi-thermometer-high::before { content: "\f5ce"; }
+.bi-thermometer-low::before { content: "\f5cf"; }
+.bi-thermometer-snow::before { content: "\f5d0"; }
+.bi-thermometer-sun::before { content: "\f5d1"; }
+.bi-thermometer::before { content: "\f5d2"; }
+.bi-three-dots-vertical::before { content: "\f5d3"; }
+.bi-three-dots::before { content: "\f5d4"; }
+.bi-toggle-off::before { content: "\f5d5"; }
+.bi-toggle-on::before { content: "\f5d6"; }
+.bi-toggle2-off::before { content: "\f5d7"; }
+.bi-toggle2-on::before { content: "\f5d8"; }
+.bi-toggles::before { content: "\f5d9"; }
+.bi-toggles2::before { content: "\f5da"; }
+.bi-tools::before { content: "\f5db"; }
+.bi-tornado::before { content: "\f5dc"; }
+.bi-trash-fill::before { content: "\f5dd"; }
+.bi-trash::before { content: "\f5de"; }
+.bi-trash2-fill::before { content: "\f5df"; }
+.bi-trash2::before { content: "\f5e0"; }
+.bi-tree-fill::before { content: "\f5e1"; }
+.bi-tree::before { content: "\f5e2"; }
+.bi-triangle-fill::before { content: "\f5e3"; }
+.bi-triangle-half::before { content: "\f5e4"; }
+.bi-triangle::before { content: "\f5e5"; }
+.bi-trophy-fill::before { content: "\f5e6"; }
+.bi-trophy::before { content: "\f5e7"; }
+.bi-tropical-storm::before { content: "\f5e8"; }
+.bi-truck-flatbed::before { content: "\f5e9"; }
+.bi-truck::before { content: "\f5ea"; }
+.bi-tsunami::before { content: "\f5eb"; }
+.bi-tv-fill::before { content: "\f5ec"; }
+.bi-tv::before { content: "\f5ed"; }
+.bi-twitch::before { content: "\f5ee"; }
+.bi-twitter::before { content: "\f5ef"; }
+.bi-type-bold::before { content: "\f5f0"; }
+.bi-type-h1::before { content: "\f5f1"; }
+.bi-type-h2::before { content: "\f5f2"; }
+.bi-type-h3::before { content: "\f5f3"; }
+.bi-type-italic::before { content: "\f5f4"; }
+.bi-type-strikethrough::before { content: "\f5f5"; }
+.bi-type-underline::before { content: "\f5f6"; }
+.bi-type::before { content: "\f5f7"; }
+.bi-ui-checks-grid::before { content: "\f5f8"; }
+.bi-ui-checks::before { content: "\f5f9"; }
+.bi-ui-radios-grid::before { content: "\f5fa"; }
+.bi-ui-radios::before { content: "\f5fb"; }
+.bi-umbrella-fill::before { content: "\f5fc"; }
+.bi-umbrella::before { content: "\f5fd"; }
+.bi-union::before { content: "\f5fe"; }
+.bi-unlock-fill::before { content: "\f5ff"; }
+.bi-unlock::before { content: "\f600"; }
+.bi-upc-scan::before { content: "\f601"; }
+.bi-upc::before { content: "\f602"; }
+.bi-upload::before { content: "\f603"; }
+.bi-vector-pen::before { content: "\f604"; }
+.bi-view-list::before { content: "\f605"; }
+.bi-view-stacked::before { content: "\f606"; }
+.bi-vinyl-fill::before { content: "\f607"; }
+.bi-vinyl::before { content: "\f608"; }
+.bi-voicemail::before { content: "\f609"; }
+.bi-volume-down-fill::before { content: "\f60a"; }
+.bi-volume-down::before { content: "\f60b"; }
+.bi-volume-mute-fill::before { content: "\f60c"; }
+.bi-volume-mute::before { content: "\f60d"; }
+.bi-volume-off-fill::before { content: "\f60e"; }
+.bi-volume-off::before { content: "\f60f"; }
+.bi-volume-up-fill::before { content: "\f610"; }
+.bi-volume-up::before { content: "\f611"; }
+.bi-vr::before { content: "\f612"; }
+.bi-wallet-fill::before { content: "\f613"; }
+.bi-wallet::before { content: "\f614"; }
+.bi-wallet2::before { content: "\f615"; }
+.bi-watch::before { content: "\f616"; }
+.bi-water::before { content: "\f617"; }
+.bi-whatsapp::before { content: "\f618"; }
+.bi-wifi-1::before { content: "\f619"; }
+.bi-wifi-2::before { content: "\f61a"; }
+.bi-wifi-off::before { content: "\f61b"; }
+.bi-wifi::before { content: "\f61c"; }
+.bi-wind::before { content: "\f61d"; }
+.bi-window-dock::before { content: "\f61e"; }
+.bi-window-sidebar::before { content: "\f61f"; }
+.bi-window::before { content: "\f620"; }
+.bi-wrench::before { content: "\f621"; }
+.bi-x-circle-fill::before { content: "\f622"; }
+.bi-x-circle::before { content: "\f623"; }
+.bi-x-diamond-fill::before { content: "\f624"; }
+.bi-x-diamond::before { content: "\f625"; }
+.bi-x-octagon-fill::before { content: "\f626"; }
+.bi-x-octagon::before { content: "\f627"; }
+.bi-x-square-fill::before { content: "\f628"; }
+.bi-x-square::before { content: "\f629"; }
+.bi-x::before { content: "\f62a"; }
+.bi-youtube::before { content: "\f62b"; }
+.bi-zoom-in::before { content: "\f62c"; }
+.bi-zoom-out::before { content: "\f62d"; }
+.bi-bank::before { content: "\f62e"; }
+.bi-bank2::before { content: "\f62f"; }
+.bi-bell-slash-fill::before { content: "\f630"; }
+.bi-bell-slash::before { content: "\f631"; }
+.bi-cash-coin::before { content: "\f632"; }
+.bi-check-lg::before { content: "\f633"; }
+.bi-coin::before { content: "\f634"; }
+.bi-currency-bitcoin::before { content: "\f635"; }
+.bi-currency-dollar::before { content: "\f636"; }
+.bi-currency-euro::before { content: "\f637"; }
+.bi-currency-exchange::before { content: "\f638"; }
+.bi-currency-pound::before { content: "\f639"; }
+.bi-currency-yen::before { content: "\f63a"; }
+.bi-dash-lg::before { content: "\f63b"; }
+.bi-exclamation-lg::before { content: "\f63c"; }
+.bi-file-earmark-pdf-fill::before { content: "\f63d"; }
+.bi-file-earmark-pdf::before { content: "\f63e"; }
+.bi-file-pdf-fill::before { content: "\f63f"; }
+.bi-file-pdf::before { content: "\f640"; }
+.bi-gender-ambiguous::before { content: "\f641"; }
+.bi-gender-female::before { content: "\f642"; }
+.bi-gender-male::before { content: "\f643"; }
+.bi-gender-trans::before { content: "\f644"; }
+.bi-headset-vr::before { content: "\f645"; }
+.bi-info-lg::before { content: "\f646"; }
+.bi-mastodon::before { content: "\f647"; }
+.bi-messenger::before { content: "\f648"; }
+.bi-piggy-bank-fill::before { content: "\f649"; }
+.bi-piggy-bank::before { content: "\f64a"; }
+.bi-pin-map-fill::before { content: "\f64b"; }
+.bi-pin-map::before { content: "\f64c"; }
+.bi-plus-lg::before { content: "\f64d"; }
+.bi-question-lg::before { content: "\f64e"; }
+.bi-recycle::before { content: "\f64f"; }
+.bi-reddit::before { content: "\f650"; }
+.bi-safe-fill::before { content: "\f651"; }
+.bi-safe2-fill::before { content: "\f652"; }
+.bi-safe2::before { content: "\f653"; }
+.bi-sd-card-fill::before { content: "\f654"; }
+.bi-sd-card::before { content: "\f655"; }
+.bi-skype::before { content: "\f656"; }
+.bi-slash-lg::before { content: "\f657"; }
+.bi-translate::before { content: "\f658"; }
+.bi-x-lg::before { content: "\f659"; }
+.bi-safe::before { content: "\f65a"; }
+.bi-apple::before { content: "\f65b"; }
+.bi-microsoft::before { content: "\f65d"; }
+.bi-windows::before { content: "\f65e"; }
+.bi-behance::before { content: "\f65c"; }
+.bi-dribbble::before { content: "\f65f"; }
+.bi-line::before { content: "\f660"; }
+.bi-medium::before { content: "\f661"; }
+.bi-paypal::before { content: "\f662"; }
+.bi-pinterest::before { content: "\f663"; }
+.bi-signal::before { content: "\f664"; }
+.bi-snapchat::before { content: "\f665"; }
+.bi-spotify::before { content: "\f666"; }
+.bi-stack-overflow::before { content: "\f667"; }
+.bi-strava::before { content: "\f668"; }
+.bi-wordpress::before { content: "\f669"; }
+.bi-vimeo::before { content: "\f66a"; }
+.bi-activity::before { content: "\f66b"; }
+.bi-easel2-fill::before { content: "\f66c"; }
+.bi-easel2::before { content: "\f66d"; }
+.bi-easel3-fill::before { content: "\f66e"; }
+.bi-easel3::before { content: "\f66f"; }
+.bi-fan::before { content: "\f670"; }
+.bi-fingerprint::before { content: "\f671"; }
+.bi-graph-down-arrow::before { content: "\f672"; }
+.bi-graph-up-arrow::before { content: "\f673"; }
+.bi-hypnotize::before { content: "\f674"; }
+.bi-magic::before { content: "\f675"; }
+.bi-person-rolodex::before { content: "\f676"; }
+.bi-person-video::before { content: "\f677"; }
+.bi-person-video2::before { content: "\f678"; }
+.bi-person-video3::before { content: "\f679"; }
+.bi-person-workspace::before { content: "\f67a"; }
+.bi-radioactive::before { content: "\f67b"; }
+.bi-webcam-fill::before { content: "\f67c"; }
+.bi-webcam::before { content: "\f67d"; }
+.bi-yin-yang::before { content: "\f67e"; }
+.bi-bandaid-fill::before { content: "\f680"; }
+.bi-bandaid::before { content: "\f681"; }
+.bi-bluetooth::before { content: "\f682"; }
+.bi-body-text::before { content: "\f683"; }
+.bi-boombox::before { content: "\f684"; }
+.bi-boxes::before { content: "\f685"; }
+.bi-dpad-fill::before { content: "\f686"; }
+.bi-dpad::before { content: "\f687"; }
+.bi-ear-fill::before { content: "\f688"; }
+.bi-ear::before { content: "\f689"; }
+.bi-envelope-check-fill::before { content: "\f68b"; }
+.bi-envelope-check::before { content: "\f68c"; }
+.bi-envelope-dash-fill::before { content: "\f68e"; }
+.bi-envelope-dash::before { content: "\f68f"; }
+.bi-envelope-exclamation-fill::before { content: "\f691"; }
+.bi-envelope-exclamation::before { content: "\f692"; }
+.bi-envelope-plus-fill::before { content: "\f693"; }
+.bi-envelope-plus::before { content: "\f694"; }
+.bi-envelope-slash-fill::before { content: "\f696"; }
+.bi-envelope-slash::before { content: "\f697"; }
+.bi-envelope-x-fill::before { content: "\f699"; }
+.bi-envelope-x::before { content: "\f69a"; }
+.bi-explicit-fill::before { content: "\f69b"; }
+.bi-explicit::before { content: "\f69c"; }
+.bi-git::before { content: "\f69d"; }
+.bi-infinity::before { content: "\f69e"; }
+.bi-list-columns-reverse::before { content: "\f69f"; }
+.bi-list-columns::before { content: "\f6a0"; }
+.bi-meta::before { content: "\f6a1"; }
+.bi-nintendo-switch::before { content: "\f6a4"; }
+.bi-pc-display-horizontal::before { content: "\f6a5"; }
+.bi-pc-display::before { content: "\f6a6"; }
+.bi-pc-horizontal::before { content: "\f6a7"; }
+.bi-pc::before { content: "\f6a8"; }
+.bi-playstation::before { content: "\f6a9"; }
+.bi-plus-slash-minus::before { content: "\f6aa"; }
+.bi-projector-fill::before { content: "\f6ab"; }
+.bi-projector::before { content: "\f6ac"; }
+.bi-qr-code-scan::before { content: "\f6ad"; }
+.bi-qr-code::before { content: "\f6ae"; }
+.bi-quora::before { content: "\f6af"; }
+.bi-quote::before { content: "\f6b0"; }
+.bi-robot::before { content: "\f6b1"; }
+.bi-send-check-fill::before { content: "\f6b2"; }
+.bi-send-check::before { content: "\f6b3"; }
+.bi-send-dash-fill::before { content: "\f6b4"; }
+.bi-send-dash::before { content: "\f6b5"; }
+.bi-send-exclamation-fill::before { content: "\f6b7"; }
+.bi-send-exclamation::before { content: "\f6b8"; }
+.bi-send-fill::before { content: "\f6b9"; }
+.bi-send-plus-fill::before { content: "\f6ba"; }
+.bi-send-plus::before { content: "\f6bb"; }
+.bi-send-slash-fill::before { content: "\f6bc"; }
+.bi-send-slash::before { content: "\f6bd"; }
+.bi-send-x-fill::before { content: "\f6be"; }
+.bi-send-x::before { content: "\f6bf"; }
+.bi-send::before { content: "\f6c0"; }
+.bi-steam::before { content: "\f6c1"; }
+.bi-terminal-dash::before { content: "\f6c3"; }
+.bi-terminal-plus::before { content: "\f6c4"; }
+.bi-terminal-split::before { content: "\f6c5"; }
+.bi-ticket-detailed-fill::before { content: "\f6c6"; }
+.bi-ticket-detailed::before { content: "\f6c7"; }
+.bi-ticket-fill::before { content: "\f6c8"; }
+.bi-ticket-perforated-fill::before { content: "\f6c9"; }
+.bi-ticket-perforated::before { content: "\f6ca"; }
+.bi-ticket::before { content: "\f6cb"; }
+.bi-tiktok::before { content: "\f6cc"; }
+.bi-window-dash::before { content: "\f6cd"; }
+.bi-window-desktop::before { content: "\f6ce"; }
+.bi-window-fullscreen::before { content: "\f6cf"; }
+.bi-window-plus::before { content: "\f6d0"; }
+.bi-window-split::before { content: "\f6d1"; }
+.bi-window-stack::before { content: "\f6d2"; }
+.bi-window-x::before { content: "\f6d3"; }
+.bi-xbox::before { content: "\f6d4"; }
+.bi-ethernet::before { content: "\f6d5"; }
+.bi-hdmi-fill::before { content: "\f6d6"; }
+.bi-hdmi::before { content: "\f6d7"; }
+.bi-usb-c-fill::before { content: "\f6d8"; }
+.bi-usb-c::before { content: "\f6d9"; }
+.bi-usb-fill::before { content: "\f6da"; }
+.bi-usb-plug-fill::before { content: "\f6db"; }
+.bi-usb-plug::before { content: "\f6dc"; }
+.bi-usb-symbol::before { content: "\f6dd"; }
+.bi-usb::before { content: "\f6de"; }
+.bi-boombox-fill::before { content: "\f6df"; }
+.bi-displayport::before { content: "\f6e1"; }
+.bi-gpu-card::before { content: "\f6e2"; }
+.bi-memory::before { content: "\f6e3"; }
+.bi-modem-fill::before { content: "\f6e4"; }
+.bi-modem::before { content: "\f6e5"; }
+.bi-motherboard-fill::before { content: "\f6e6"; }
+.bi-motherboard::before { content: "\f6e7"; }
+.bi-optical-audio-fill::before { content: "\f6e8"; }
+.bi-optical-audio::before { content: "\f6e9"; }
+.bi-pci-card::before { content: "\f6ea"; }
+.bi-router-fill::before { content: "\f6eb"; }
+.bi-router::before { content: "\f6ec"; }
+.bi-thunderbolt-fill::before { content: "\f6ef"; }
+.bi-thunderbolt::before { content: "\f6f0"; }
+.bi-usb-drive-fill::before { content: "\f6f1"; }
+.bi-usb-drive::before { content: "\f6f2"; }
+.bi-usb-micro-fill::before { content: "\f6f3"; }
+.bi-usb-micro::before { content: "\f6f4"; }
+.bi-usb-mini-fill::before { content: "\f6f5"; }
+.bi-usb-mini::before { content: "\f6f6"; }
+.bi-cloud-haze2::before { content: "\f6f7"; }
+.bi-device-hdd-fill::before { content: "\f6f8"; }
+.bi-device-hdd::before { content: "\f6f9"; }
+.bi-device-ssd-fill::before { content: "\f6fa"; }
+.bi-device-ssd::before { content: "\f6fb"; }
+.bi-displayport-fill::before { content: "\f6fc"; }
+.bi-mortarboard-fill::before { content: "\f6fd"; }
+.bi-mortarboard::before { content: "\f6fe"; }
+.bi-terminal-x::before { content: "\f6ff"; }
+.bi-arrow-through-heart-fill::before { content: "\f700"; }
+.bi-arrow-through-heart::before { content: "\f701"; }
+.bi-badge-sd-fill::before { content: "\f702"; }
+.bi-badge-sd::before { content: "\f703"; }
+.bi-bag-heart-fill::before { content: "\f704"; }
+.bi-bag-heart::before { content: "\f705"; }
+.bi-balloon-fill::before { content: "\f706"; }
+.bi-balloon-heart-fill::before { content: "\f707"; }
+.bi-balloon-heart::before { content: "\f708"; }
+.bi-balloon::before { content: "\f709"; }
+.bi-box2-fill::before { content: "\f70a"; }
+.bi-box2-heart-fill::before { content: "\f70b"; }
+.bi-box2-heart::before { content: "\f70c"; }
+.bi-box2::before { content: "\f70d"; }
+.bi-braces-asterisk::before { content: "\f70e"; }
+.bi-calendar-heart-fill::before { content: "\f70f"; }
+.bi-calendar-heart::before { content: "\f710"; }
+.bi-calendar2-heart-fill::before { content: "\f711"; }
+.bi-calendar2-heart::before { content: "\f712"; }
+.bi-chat-heart-fill::before { content: "\f713"; }
+.bi-chat-heart::before { content: "\f714"; }
+.bi-chat-left-heart-fill::before { content: "\f715"; }
+.bi-chat-left-heart::before { content: "\f716"; }
+.bi-chat-right-heart-fill::before { content: "\f717"; }
+.bi-chat-right-heart::before { content: "\f718"; }
+.bi-chat-square-heart-fill::before { content: "\f719"; }
+.bi-chat-square-heart::before { content: "\f71a"; }
+.bi-clipboard-check-fill::before { content: "\f71b"; }
+.bi-clipboard-data-fill::before { content: "\f71c"; }
+.bi-clipboard-fill::before { content: "\f71d"; }
+.bi-clipboard-heart-fill::before { content: "\f71e"; }
+.bi-clipboard-heart::before { content: "\f71f"; }
+.bi-clipboard-minus-fill::before { content: "\f720"; }
+.bi-clipboard-plus-fill::before { content: "\f721"; }
+.bi-clipboard-pulse::before { content: "\f722"; }
+.bi-clipboard-x-fill::before { content: "\f723"; }
+.bi-clipboard2-check-fill::before { content: "\f724"; }
+.bi-clipboard2-check::before { content: "\f725"; }
+.bi-clipboard2-data-fill::before { content: "\f726"; }
+.bi-clipboard2-data::before { content: "\f727"; }
+.bi-clipboard2-fill::before { content: "\f728"; }
+.bi-clipboard2-heart-fill::before { content: "\f729"; }
+.bi-clipboard2-heart::before { content: "\f72a"; }
+.bi-clipboard2-minus-fill::before { content: "\f72b"; }
+.bi-clipboard2-minus::before { content: "\f72c"; }
+.bi-clipboard2-plus-fill::before { content: "\f72d"; }
+.bi-clipboard2-plus::before { content: "\f72e"; }
+.bi-clipboard2-pulse-fill::before { content: "\f72f"; }
+.bi-clipboard2-pulse::before { content: "\f730"; }
+.bi-clipboard2-x-fill::before { content: "\f731"; }
+.bi-clipboard2-x::before { content: "\f732"; }
+.bi-clipboard2::before { content: "\f733"; }
+.bi-emoji-kiss-fill::before { content: "\f734"; }
+.bi-emoji-kiss::before { content: "\f735"; }
+.bi-envelope-heart-fill::before { content: "\f736"; }
+.bi-envelope-heart::before { content: "\f737"; }
+.bi-envelope-open-heart-fill::before { content: "\f738"; }
+.bi-envelope-open-heart::before { content: "\f739"; }
+.bi-envelope-paper-fill::before { content: "\f73a"; }
+.bi-envelope-paper-heart-fill::before { content: "\f73b"; }
+.bi-envelope-paper-heart::before { content: "\f73c"; }
+.bi-envelope-paper::before { content: "\f73d"; }
+.bi-filetype-aac::before { content: "\f73e"; }
+.bi-filetype-ai::before { content: "\f73f"; }
+.bi-filetype-bmp::before { content: "\f740"; }
+.bi-filetype-cs::before { content: "\f741"; }
+.bi-filetype-css::before { content: "\f742"; }
+.bi-filetype-csv::before { content: "\f743"; }
+.bi-filetype-doc::before { content: "\f744"; }
+.bi-filetype-docx::before { content: "\f745"; }
+.bi-filetype-exe::before { content: "\f746"; }
+.bi-filetype-gif::before { content: "\f747"; }
+.bi-filetype-heic::before { content: "\f748"; }
+.bi-filetype-html::before { content: "\f749"; }
+.bi-filetype-java::before { content: "\f74a"; }
+.bi-filetype-jpg::before { content: "\f74b"; }
+.bi-filetype-js::before { content: "\f74c"; }
+.bi-filetype-jsx::before { content: "\f74d"; }
+.bi-filetype-key::before { content: "\f74e"; }
+.bi-filetype-m4p::before { content: "\f74f"; }
+.bi-filetype-md::before { content: "\f750"; }
+.bi-filetype-mdx::before { content: "\f751"; }
+.bi-filetype-mov::before { content: "\f752"; }
+.bi-filetype-mp3::before { content: "\f753"; }
+.bi-filetype-mp4::before { content: "\f754"; }
+.bi-filetype-otf::before { content: "\f755"; }
+.bi-filetype-pdf::before { content: "\f756"; }
+.bi-filetype-php::before { content: "\f757"; }
+.bi-filetype-png::before { content: "\f758"; }
+.bi-filetype-ppt::before { content: "\f75a"; }
+.bi-filetype-psd::before { content: "\f75b"; }
+.bi-filetype-py::before { content: "\f75c"; }
+.bi-filetype-raw::before { content: "\f75d"; }
+.bi-filetype-rb::before { content: "\f75e"; }
+.bi-filetype-sass::before { content: "\f75f"; }
+.bi-filetype-scss::before { content: "\f760"; }
+.bi-filetype-sh::before { content: "\f761"; }
+.bi-filetype-svg::before { content: "\f762"; }
+.bi-filetype-tiff::before { content: "\f763"; }
+.bi-filetype-tsx::before { content: "\f764"; }
+.bi-filetype-ttf::before { content: "\f765"; }
+.bi-filetype-txt::before { content: "\f766"; }
+.bi-filetype-wav::before { content: "\f767"; }
+.bi-filetype-woff::before { content: "\f768"; }
+.bi-filetype-xls::before { content: "\f76a"; }
+.bi-filetype-xml::before { content: "\f76b"; }
+.bi-filetype-yml::before { content: "\f76c"; }
+.bi-heart-arrow::before { content: "\f76d"; }
+.bi-heart-pulse-fill::before { content: "\f76e"; }
+.bi-heart-pulse::before { content: "\f76f"; }
+.bi-heartbreak-fill::before { content: "\f770"; }
+.bi-heartbreak::before { content: "\f771"; }
+.bi-hearts::before { content: "\f772"; }
+.bi-hospital-fill::before { content: "\f773"; }
+.bi-hospital::before { content: "\f774"; }
+.bi-house-heart-fill::before { content: "\f775"; }
+.bi-house-heart::before { content: "\f776"; }
+.bi-incognito::before { content: "\f777"; }
+.bi-magnet-fill::before { content: "\f778"; }
+.bi-magnet::before { content: "\f779"; }
+.bi-person-heart::before { content: "\f77a"; }
+.bi-person-hearts::before { content: "\f77b"; }
+.bi-phone-flip::before { content: "\f77c"; }
+.bi-plugin::before { content: "\f77d"; }
+.bi-postage-fill::before { content: "\f77e"; }
+.bi-postage-heart-fill::before { content: "\f77f"; }
+.bi-postage-heart::before { content: "\f780"; }
+.bi-postage::before { content: "\f781"; }
+.bi-postcard-fill::before { content: "\f782"; }
+.bi-postcard-heart-fill::before { content: "\f783"; }
+.bi-postcard-heart::before { content: "\f784"; }
+.bi-postcard::before { content: "\f785"; }
+.bi-search-heart-fill::before { content: "\f786"; }
+.bi-search-heart::before { content: "\f787"; }
+.bi-sliders2-vertical::before { content: "\f788"; }
+.bi-sliders2::before { content: "\f789"; }
+.bi-trash3-fill::before { content: "\f78a"; }
+.bi-trash3::before { content: "\f78b"; }
+.bi-valentine::before { content: "\f78c"; }
+.bi-valentine2::before { content: "\f78d"; }
+.bi-wrench-adjustable-circle-fill::before { content: "\f78e"; }
+.bi-wrench-adjustable-circle::before { content: "\f78f"; }
+.bi-wrench-adjustable::before { content: "\f790"; }
+.bi-filetype-json::before { content: "\f791"; }
+.bi-filetype-pptx::before { content: "\f792"; }
+.bi-filetype-xlsx::before { content: "\f793"; }
+.bi-1-circle-fill::before { content: "\f796"; }
+.bi-1-circle::before { content: "\f797"; }
+.bi-1-square-fill::before { content: "\f798"; }
+.bi-1-square::before { content: "\f799"; }
+.bi-2-circle-fill::before { content: "\f79c"; }
+.bi-2-circle::before { content: "\f79d"; }
+.bi-2-square-fill::before { content: "\f79e"; }
+.bi-2-square::before { content: "\f79f"; }
+.bi-3-circle-fill::before { content: "\f7a2"; }
+.bi-3-circle::before { content: "\f7a3"; }
+.bi-3-square-fill::before { content: "\f7a4"; }
+.bi-3-square::before { content: "\f7a5"; }
+.bi-4-circle-fill::before { content: "\f7a8"; }
+.bi-4-circle::before { content: "\f7a9"; }
+.bi-4-square-fill::before { content: "\f7aa"; }
+.bi-4-square::before { content: "\f7ab"; }
+.bi-5-circle-fill::before { content: "\f7ae"; }
+.bi-5-circle::before { content: "\f7af"; }
+.bi-5-square-fill::before { content: "\f7b0"; }
+.bi-5-square::before { content: "\f7b1"; }
+.bi-6-circle-fill::before { content: "\f7b4"; }
+.bi-6-circle::before { content: "\f7b5"; }
+.bi-6-square-fill::before { content: "\f7b6"; }
+.bi-6-square::before { content: "\f7b7"; }
+.bi-7-circle-fill::before { content: "\f7ba"; }
+.bi-7-circle::before { content: "\f7bb"; }
+.bi-7-square-fill::before { content: "\f7bc"; }
+.bi-7-square::before { content: "\f7bd"; }
+.bi-8-circle-fill::before { content: "\f7c0"; }
+.bi-8-circle::before { content: "\f7c1"; }
+.bi-8-square-fill::before { content: "\f7c2"; }
+.bi-8-square::before { content: "\f7c3"; }
+.bi-9-circle-fill::before { content: "\f7c6"; }
+.bi-9-circle::before { content: "\f7c7"; }
+.bi-9-square-fill::before { content: "\f7c8"; }
+.bi-9-square::before { content: "\f7c9"; }
+.bi-airplane-engines-fill::before { content: "\f7ca"; }
+.bi-airplane-engines::before { content: "\f7cb"; }
+.bi-airplane-fill::before { content: "\f7cc"; }
+.bi-airplane::before { content: "\f7cd"; }
+.bi-alexa::before { content: "\f7ce"; }
+.bi-alipay::before { content: "\f7cf"; }
+.bi-android::before { content: "\f7d0"; }
+.bi-android2::before { content: "\f7d1"; }
+.bi-box-fill::before { content: "\f7d2"; }
+.bi-box-seam-fill::before { content: "\f7d3"; }
+.bi-browser-chrome::before { content: "\f7d4"; }
+.bi-browser-edge::before { content: "\f7d5"; }
+.bi-browser-firefox::before { content: "\f7d6"; }
+.bi-browser-safari::before { content: "\f7d7"; }
+.bi-c-circle-fill::before { content: "\f7da"; }
+.bi-c-circle::before { content: "\f7db"; }
+.bi-c-square-fill::before { content: "\f7dc"; }
+.bi-c-square::before { content: "\f7dd"; }
+.bi-capsule-pill::before { content: "\f7de"; }
+.bi-capsule::before { content: "\f7df"; }
+.bi-car-front-fill::before { content: "\f7e0"; }
+.bi-car-front::before { content: "\f7e1"; }
+.bi-cassette-fill::before { content: "\f7e2"; }
+.bi-cassette::before { content: "\f7e3"; }
+.bi-cc-circle-fill::before { content: "\f7e6"; }
+.bi-cc-circle::before { content: "\f7e7"; }
+.bi-cc-square-fill::before { content: "\f7e8"; }
+.bi-cc-square::before { content: "\f7e9"; }
+.bi-cup-hot-fill::before { content: "\f7ea"; }
+.bi-cup-hot::before { content: "\f7eb"; }
+.bi-currency-rupee::before { content: "\f7ec"; }
+.bi-dropbox::before { content: "\f7ed"; }
+.bi-escape::before { content: "\f7ee"; }
+.bi-fast-forward-btn-fill::before { content: "\f7ef"; }
+.bi-fast-forward-btn::before { content: "\f7f0"; }
+.bi-fast-forward-circle-fill::before { content: "\f7f1"; }
+.bi-fast-forward-circle::before { content: "\f7f2"; }
+.bi-fast-forward-fill::before { content: "\f7f3"; }
+.bi-fast-forward::before { content: "\f7f4"; }
+.bi-filetype-sql::before { content: "\f7f5"; }
+.bi-fire::before { content: "\f7f6"; }
+.bi-google-play::before { content: "\f7f7"; }
+.bi-h-circle-fill::before { content: "\f7fa"; }
+.bi-h-circle::before { content: "\f7fb"; }
+.bi-h-square-fill::before { content: "\f7fc"; }
+.bi-h-square::before { content: "\f7fd"; }
+.bi-indent::before { content: "\f7fe"; }
+.bi-lungs-fill::before { content: "\f7ff"; }
+.bi-lungs::before { content: "\f800"; }
+.bi-microsoft-teams::before { content: "\f801"; }
+.bi-p-circle-fill::before { content: "\f804"; }
+.bi-p-circle::before { content: "\f805"; }
+.bi-p-square-fill::before { content: "\f806"; }
+.bi-p-square::before { content: "\f807"; }
+.bi-pass-fill::before { content: "\f808"; }
+.bi-pass::before { content: "\f809"; }
+.bi-prescription::before { content: "\f80a"; }
+.bi-prescription2::before { content: "\f80b"; }
+.bi-r-circle-fill::before { content: "\f80e"; }
+.bi-r-circle::before { content: "\f80f"; }
+.bi-r-square-fill::before { content: "\f810"; }
+.bi-r-square::before { content: "\f811"; }
+.bi-repeat-1::before { content: "\f812"; }
+.bi-repeat::before { content: "\f813"; }
+.bi-rewind-btn-fill::before { content: "\f814"; }
+.bi-rewind-btn::before { content: "\f815"; }
+.bi-rewind-circle-fill::before { content: "\f816"; }
+.bi-rewind-circle::before { content: "\f817"; }
+.bi-rewind-fill::before { content: "\f818"; }
+.bi-rewind::before { content: "\f819"; }
+.bi-train-freight-front-fill::before { content: "\f81a"; }
+.bi-train-freight-front::before { content: "\f81b"; }
+.bi-train-front-fill::before { content: "\f81c"; }
+.bi-train-front::before { content: "\f81d"; }
+.bi-train-lightrail-front-fill::before { content: "\f81e"; }
+.bi-train-lightrail-front::before { content: "\f81f"; }
+.bi-truck-front-fill::before { content: "\f820"; }
+.bi-truck-front::before { content: "\f821"; }
+.bi-ubuntu::before { content: "\f822"; }
+.bi-unindent::before { content: "\f823"; }
+.bi-unity::before { content: "\f824"; }
+.bi-universal-access-circle::before { content: "\f825"; }
+.bi-universal-access::before { content: "\f826"; }
+.bi-virus::before { content: "\f827"; }
+.bi-virus2::before { content: "\f828"; }
+.bi-wechat::before { content: "\f829"; }
+.bi-yelp::before { content: "\f82a"; }
+.bi-sign-stop-fill::before { content: "\f82b"; }
+.bi-sign-stop-lights-fill::before { content: "\f82c"; }
+.bi-sign-stop-lights::before { content: "\f82d"; }
+.bi-sign-stop::before { content: "\f82e"; }
+.bi-sign-turn-left-fill::before { content: "\f82f"; }
+.bi-sign-turn-left::before { content: "\f830"; }
+.bi-sign-turn-right-fill::before { content: "\f831"; }
+.bi-sign-turn-right::before { content: "\f832"; }
+.bi-sign-turn-slight-left-fill::before { content: "\f833"; }
+.bi-sign-turn-slight-left::before { content: "\f834"; }
+.bi-sign-turn-slight-right-fill::before { content: "\f835"; }
+.bi-sign-turn-slight-right::before { content: "\f836"; }
+.bi-sign-yield-fill::before { content: "\f837"; }
+.bi-sign-yield::before { content: "\f838"; }
+.bi-ev-station-fill::before { content: "\f839"; }
+.bi-ev-station::before { content: "\f83a"; }
+.bi-fuel-pump-diesel-fill::before { content: "\f83b"; }
+.bi-fuel-pump-diesel::before { content: "\f83c"; }
+.bi-fuel-pump-fill::before { content: "\f83d"; }
+.bi-fuel-pump::before { content: "\f83e"; }
+.bi-0-circle-fill::before { content: "\f83f"; }
+.bi-0-circle::before { content: "\f840"; }
+.bi-0-square-fill::before { content: "\f841"; }
+.bi-0-square::before { content: "\f842"; }
+.bi-rocket-fill::before { content: "\f843"; }
+.bi-rocket-takeoff-fill::before { content: "\f844"; }
+.bi-rocket-takeoff::before { content: "\f845"; }
+.bi-rocket::before { content: "\f846"; }
+.bi-stripe::before { content: "\f847"; }
+.bi-subscript::before { content: "\f848"; }
+.bi-superscript::before { content: "\f849"; }
+.bi-trello::before { content: "\f84a"; }
+.bi-envelope-at-fill::before { content: "\f84b"; }
+.bi-envelope-at::before { content: "\f84c"; }
+.bi-regex::before { content: "\f84d"; }
+.bi-text-wrap::before { content: "\f84e"; }
+.bi-sign-dead-end-fill::before { content: "\f84f"; }
+.bi-sign-dead-end::before { content: "\f850"; }
+.bi-sign-do-not-enter-fill::before { content: "\f851"; }
+.bi-sign-do-not-enter::before { content: "\f852"; }
+.bi-sign-intersection-fill::before { content: "\f853"; }
+.bi-sign-intersection-side-fill::before { content: "\f854"; }
+.bi-sign-intersection-side::before { content: "\f855"; }
+.bi-sign-intersection-t-fill::before { content: "\f856"; }
+.bi-sign-intersection-t::before { content: "\f857"; }
+.bi-sign-intersection-y-fill::before { content: "\f858"; }
+.bi-sign-intersection-y::before { content: "\f859"; }
+.bi-sign-intersection::before { content: "\f85a"; }
+.bi-sign-merge-left-fill::before { content: "\f85b"; }
+.bi-sign-merge-left::before { content: "\f85c"; }
+.bi-sign-merge-right-fill::before { content: "\f85d"; }
+.bi-sign-merge-right::before { content: "\f85e"; }
+.bi-sign-no-left-turn-fill::before { content: "\f85f"; }
+.bi-sign-no-left-turn::before { content: "\f860"; }
+.bi-sign-no-parking-fill::before { content: "\f861"; }
+.bi-sign-no-parking::before { content: "\f862"; }
+.bi-sign-no-right-turn-fill::before { content: "\f863"; }
+.bi-sign-no-right-turn::before { content: "\f864"; }
+.bi-sign-railroad-fill::before { content: "\f865"; }
+.bi-sign-railroad::before { content: "\f866"; }
+.bi-building-add::before { content: "\f867"; }
+.bi-building-check::before { content: "\f868"; }
+.bi-building-dash::before { content: "\f869"; }
+.bi-building-down::before { content: "\f86a"; }
+.bi-building-exclamation::before { content: "\f86b"; }
+.bi-building-fill-add::before { content: "\f86c"; }
+.bi-building-fill-check::before { content: "\f86d"; }
+.bi-building-fill-dash::before { content: "\f86e"; }
+.bi-building-fill-down::before { content: "\f86f"; }
+.bi-building-fill-exclamation::before { content: "\f870"; }
+.bi-building-fill-gear::before { content: "\f871"; }
+.bi-building-fill-lock::before { content: "\f872"; }
+.bi-building-fill-slash::before { content: "\f873"; }
+.bi-building-fill-up::before { content: "\f874"; }
+.bi-building-fill-x::before { content: "\f875"; }
+.bi-building-fill::before { content: "\f876"; }
+.bi-building-gear::before { content: "\f877"; }
+.bi-building-lock::before { content: "\f878"; }
+.bi-building-slash::before { content: "\f879"; }
+.bi-building-up::before { content: "\f87a"; }
+.bi-building-x::before { content: "\f87b"; }
+.bi-buildings-fill::before { content: "\f87c"; }
+.bi-buildings::before { content: "\f87d"; }
+.bi-bus-front-fill::before { content: "\f87e"; }
+.bi-bus-front::before { content: "\f87f"; }
+.bi-ev-front-fill::before { content: "\f880"; }
+.bi-ev-front::before { content: "\f881"; }
+.bi-globe-americas::before { content: "\f882"; }
+.bi-globe-asia-australia::before { content: "\f883"; }
+.bi-globe-central-south-asia::before { content: "\f884"; }
+.bi-globe-europe-africa::before { content: "\f885"; }
+.bi-house-add-fill::before { content: "\f886"; }
+.bi-house-add::before { content: "\f887"; }
+.bi-house-check-fill::before { content: "\f888"; }
+.bi-house-check::before { content: "\f889"; }
+.bi-house-dash-fill::before { content: "\f88a"; }
+.bi-house-dash::before { content: "\f88b"; }
+.bi-house-down-fill::before { content: "\f88c"; }
+.bi-house-down::before { content: "\f88d"; }
+.bi-house-exclamation-fill::before { content: "\f88e"; }
+.bi-house-exclamation::before { content: "\f88f"; }
+.bi-house-gear-fill::before { content: "\f890"; }
+.bi-house-gear::before { content: "\f891"; }
+.bi-house-lock-fill::before { content: "\f892"; }
+.bi-house-lock::before { content: "\f893"; }
+.bi-house-slash-fill::before { content: "\f894"; }
+.bi-house-slash::before { content: "\f895"; }
+.bi-house-up-fill::before { content: "\f896"; }
+.bi-house-up::before { content: "\f897"; }
+.bi-house-x-fill::before { content: "\f898"; }
+.bi-house-x::before { content: "\f899"; }
+.bi-person-add::before { content: "\f89a"; }
+.bi-person-down::before { content: "\f89b"; }
+.bi-person-exclamation::before { content: "\f89c"; }
+.bi-person-fill-add::before { content: "\f89d"; }
+.bi-person-fill-check::before { content: "\f89e"; }
+.bi-person-fill-dash::before { content: "\f89f"; }
+.bi-person-fill-down::before { content: "\f8a0"; }
+.bi-person-fill-exclamation::before { content: "\f8a1"; }
+.bi-person-fill-gear::before { content: "\f8a2"; }
+.bi-person-fill-lock::before { content: "\f8a3"; }
+.bi-person-fill-slash::before { content: "\f8a4"; }
+.bi-person-fill-up::before { content: "\f8a5"; }
+.bi-person-fill-x::before { content: "\f8a6"; }
+.bi-person-gear::before { content: "\f8a7"; }
+.bi-person-lock::before { content: "\f8a8"; }
+.bi-person-slash::before { content: "\f8a9"; }
+.bi-person-up::before { content: "\f8aa"; }
+.bi-scooter::before { content: "\f8ab"; }
+.bi-taxi-front-fill::before { content: "\f8ac"; }
+.bi-taxi-front::before { content: "\f8ad"; }
+.bi-amd::before { content: "\f8ae"; }
+.bi-database-add::before { content: "\f8af"; }
+.bi-database-check::before { content: "\f8b0"; }
+.bi-database-dash::before { content: "\f8b1"; }
+.bi-database-down::before { content: "\f8b2"; }
+.bi-database-exclamation::before { content: "\f8b3"; }
+.bi-database-fill-add::before { content: "\f8b4"; }
+.bi-database-fill-check::before { content: "\f8b5"; }
+.bi-database-fill-dash::before { content: "\f8b6"; }
+.bi-database-fill-down::before { content: "\f8b7"; }
+.bi-database-fill-exclamation::before { content: "\f8b8"; }
+.bi-database-fill-gear::before { content: "\f8b9"; }
+.bi-database-fill-lock::before { content: "\f8ba"; }
+.bi-database-fill-slash::before { content: "\f8bb"; }
+.bi-database-fill-up::before { content: "\f8bc"; }
+.bi-database-fill-x::before { content: "\f8bd"; }
+.bi-database-fill::before { content: "\f8be"; }
+.bi-database-gear::before { content: "\f8bf"; }
+.bi-database-lock::before { content: "\f8c0"; }
+.bi-database-slash::before { content: "\f8c1"; }
+.bi-database-up::before { content: "\f8c2"; }
+.bi-database-x::before { content: "\f8c3"; }
+.bi-database::before { content: "\f8c4"; }
+.bi-houses-fill::before { content: "\f8c5"; }
+.bi-houses::before { content: "\f8c6"; }
+.bi-nvidia::before { content: "\f8c7"; }
+.bi-person-vcard-fill::before { content: "\f8c8"; }
+.bi-person-vcard::before { content: "\f8c9"; }
+.bi-sina-weibo::before { content: "\f8ca"; }
+.bi-tencent-qq::before { content: "\f8cb"; }
+.bi-wikipedia::before { content: "\f8cc"; }
+.bi-alphabet-uppercase::before { content: "\f2a5"; }
+.bi-alphabet::before { content: "\f68a"; }
+.bi-amazon::before { content: "\f68d"; }
+.bi-arrows-collapse-vertical::before { content: "\f690"; }
+.bi-arrows-expand-vertical::before { content: "\f695"; }
+.bi-arrows-vertical::before { content: "\f698"; }
+.bi-arrows::before { content: "\f6a2"; }
+.bi-ban-fill::before { content: "\f6a3"; }
+.bi-ban::before { content: "\f6b6"; }
+.bi-bing::before { content: "\f6c2"; }
+.bi-cake::before { content: "\f6e0"; }
+.bi-cake2::before { content: "\f6ed"; }
+.bi-cookie::before { content: "\f6ee"; }
+.bi-copy::before { content: "\f759"; }
+.bi-crosshair::before { content: "\f769"; }
+.bi-crosshair2::before { content: "\f794"; }
+.bi-emoji-astonished-fill::before { content: "\f795"; }
+.bi-emoji-astonished::before { content: "\f79a"; }
+.bi-emoji-grimace-fill::before { content: "\f79b"; }
+.bi-emoji-grimace::before { content: "\f7a0"; }
+.bi-emoji-grin-fill::before { content: "\f7a1"; }
+.bi-emoji-grin::before { content: "\f7a6"; }
+.bi-emoji-surprise-fill::before { content: "\f7a7"; }
+.bi-emoji-surprise::before { content: "\f7ac"; }
+.bi-emoji-tear-fill::before { content: "\f7ad"; }
+.bi-emoji-tear::before { content: "\f7b2"; }
+.bi-envelope-arrow-down-fill::before { content: "\f7b3"; }
+.bi-envelope-arrow-down::before { content: "\f7b8"; }
+.bi-envelope-arrow-up-fill::before { content: "\f7b9"; }
+.bi-envelope-arrow-up::before { content: "\f7be"; }
+.bi-feather::before { content: "\f7bf"; }
+.bi-feather2::before { content: "\f7c4"; }
+.bi-floppy-fill::before { content: "\f7c5"; }
+.bi-floppy::before { content: "\f7d8"; }
+.bi-floppy2-fill::before { content: "\f7d9"; }
+.bi-floppy2::before { content: "\f7e4"; }
+.bi-gitlab::before { content: "\f7e5"; }
+.bi-highlighter::before { content: "\f7f8"; }
+.bi-marker-tip::before { content: "\f802"; }
+.bi-nvme-fill::before { content: "\f803"; }
+.bi-nvme::before { content: "\f80c"; }
+.bi-opencollective::before { content: "\f80d"; }
+.bi-pci-card-network::before { content: "\f8cd"; }
+.bi-pci-card-sound::before { content: "\f8ce"; }
+.bi-radar::before { content: "\f8cf"; }
+.bi-send-arrow-down-fill::before { content: "\f8d0"; }
+.bi-send-arrow-down::before { content: "\f8d1"; }
+.bi-send-arrow-up-fill::before { content: "\f8d2"; }
+.bi-send-arrow-up::before { content: "\f8d3"; }
+.bi-sim-slash-fill::before { content: "\f8d4"; }
+.bi-sim-slash::before { content: "\f8d5"; }
+.bi-sourceforge::before { content: "\f8d6"; }
+.bi-substack::before { content: "\f8d7"; }
+.bi-threads-fill::before { content: "\f8d8"; }
+.bi-threads::before { content: "\f8d9"; }
+.bi-transparency::before { content: "\f8da"; }
+.bi-twitter-x::before { content: "\f8db"; }
+.bi-type-h4::before { content: "\f8dc"; }
+.bi-type-h5::before { content: "\f8dd"; }
+.bi-type-h6::before { content: "\f8de"; }
+.bi-backpack-fill::before { content: "\f8df"; }
+.bi-backpack::before { content: "\f8e0"; }
+.bi-backpack2-fill::before { content: "\f8e1"; }
+.bi-backpack2::before { content: "\f8e2"; }
+.bi-backpack3-fill::before { content: "\f8e3"; }
+.bi-backpack3::before { content: "\f8e4"; }
+.bi-backpack4-fill::before { content: "\f8e5"; }
+.bi-backpack4::before { content: "\f8e6"; }
+.bi-brilliance::before { content: "\f8e7"; }
+.bi-cake-fill::before { content: "\f8e8"; }
+.bi-cake2-fill::before { content: "\f8e9"; }
+.bi-duffle-fill::before { content: "\f8ea"; }
+.bi-duffle::before { content: "\f8eb"; }
+.bi-exposure::before { content: "\f8ec"; }
+.bi-gender-neuter::before { content: "\f8ed"; }
+.bi-highlights::before { content: "\f8ee"; }
+.bi-luggage-fill::before { content: "\f8ef"; }
+.bi-luggage::before { content: "\f8f0"; }
+.bi-mailbox-flag::before { content: "\f8f1"; }
+.bi-mailbox2-flag::before { content: "\f8f2"; }
+.bi-noise-reduction::before { content: "\f8f3"; }
+.bi-passport-fill::before { content: "\f8f4"; }
+.bi-passport::before { content: "\f8f5"; }
+.bi-person-arms-up::before { content: "\f8f6"; }
+.bi-person-raised-hand::before { content: "\f8f7"; }
+.bi-person-standing-dress::before { content: "\f8f8"; }
+.bi-person-standing::before { content: "\f8f9"; }
+.bi-person-walking::before { content: "\f8fa"; }
+.bi-person-wheelchair::before { content: "\f8fb"; }
+.bi-shadows::before { content: "\f8fc"; }
+.bi-suitcase-fill::before { content: "\f8fd"; }
+.bi-suitcase-lg-fill::before { content: "\f8fe"; }
+.bi-suitcase-lg::before { content: "\f8ff"; }
+.bi-suitcase::before { content: "\f900"; }
+.bi-suitcase2-fill::before { content: "\f901"; }
+.bi-suitcase2::before { content: "\f902"; }
+.bi-vignette::before { content: "\f903"; }
diff --git a/docs/2_39/site_libs/bootstrap/bootstrap-icons.woff b/docs/2_39/site_libs/bootstrap/bootstrap-icons.woff
new file mode 100644
index 0000000000000000000000000000000000000000..dbeeb055674125ad78fda0f3d166b36e5cc92336
GIT binary patch
literal 176200
zcmZ6SbyyUC7sW9!5J7YWX;@miUAjA$5+r2-2|<=_6$w#bgHDkJBm@EJQV`gsB}7_e
z>5^`EXMTUaKF=J!_jAs@GaIZkv+Ad>rbcp!goNbs7Y&kIz|ZSC4FA=@^8f#+8<{AP
zkX*U}aA{yOW_iaEsBa`F0x%VzRs=R%IWi+5`{#Bq02WO`BDzUJ;u&f8kFVLuEx?h4
zMBJa`vT!BIHQG-iKWulOIoKgcE<5o7eZUM7iN_@$6rKSPV75Tb1Z?b=U)-d6_S_rj
zb9xEP3?(69xoUUw+|JFz9>_TZ5y%X{ZajFd$oJgN{{_kAkUs!q1~!(Pk1n~o+dX$6
zxeT<CT#^MIq>HZ@w(f<8mp94fFa;74Vc@X@NAiYJYWru{+ahdj|2!44{bFy6^xU~=
z_orKvk6@2_YHRnB1SKPqF3cq=i+**b<4RZgOJ@oe$MEROB%IQu8YEz^-LPH8w{KnF
zzI}<Gy^9)tJ$UcI;e%k>2PqF8r_z3T{Zecc5_yH0HcUixg`{rq{RVl3LK>AS)jbl<
zh?_rvqw~*LpNhCh7^x@yH$@M*zeatJKB0n?M{^louWX<|&ZoeR`;ml6fJ;GCzf+*@
zsPHM=Bqd$Q^m8PMIN|$sB)V}lxjA(}<`gQrv*Gl)(@TaaFTqU9+_UM0R^qeIUr%j{
z{JoBHkAE=Ntl;j2P2TU^yt&=*RphAEF6gut9_4+0L+>ccbT*+RBhQ4^r}ANOSK)Ti
z>!MHYW{JiQCaNYTBgQ@^%2UNIMHWTXMY$_Qfh%$*HsS`iP1r^riyP{ih>loR8Ssys
zty~(>sxp0U{A5J0%8b!ieMHm8)XLawMAyem)>wb<!%ALYv0PEvN&%DM*+}k+@Lp}8
z(rp9%qJrU}723TLx=+^%zq>@!6-5@#y5Q*Y)QW{&N&*dIjpjzK0=t1@N1nLEq!r~C
zF1tjg6;7L04!en~_nPbs2UjWZ8^0TVTBX8o(mjl<QfexgHsc+6xQ9M|{MpoX*rG}=
zq~KTaZI|3}T(epF2=!k0cu}bF#IXJfO_kNUf%M3xOY->V{ZCCU+2dvBrWc>CtbCBd
zi99qkPb|vlDt;|h689;0#bz&CD!)o%+@+w2LTUwC|4B|WyX4)n(Qe_fn3ZMnK*6f$
zZt5{#NVS}Lc5(mE;_9v4h+}9-d9zCLaPkW8ZsKuZNO-eh@-K&7-D5{9)8wIfA5tsB
znIexNzg4aJie`1QpC&%qQ(Ar_Q{H}4$_K-gE7tWjp&IffCrj$yVP~I0b>vI42d?a5
zk9p3%hN{UIUtduS{1U21`LlmDCoqMnRDH=X@GDbp=L*f<qccpXdCNzek+S8{J^Eu^
zS!?})dwcA%<}G;zOND`HS*1=(smIS{wfld0?OMv}x8%-@?Cf1`G8P=_-XlB93biR4
zW?rJ)dw7OEe_Zy<b(ibd=q&ZzGAyTM*k-wGPxAOllWEz|%<^7f`1B{q<;^|ECRL9S
z`IYQFiYB9rQomK%z_RIrgVmNj`Qw*zpD35#>v@|l`Y1C0Qr|T^D?8U`79D?JA1gY2
z^`0)3(QpPrPof~jsMk5amd8#{(kVr>*L=avD-JfA;nXKdlX9z9b>XSkTOMZt@#NI*
z-unw$UWq&or4pkluDw1B*Nny!MDO=}UXU=F7#8-?mG#Ol^q@<L%tt%BrKRNOr#ib^
zq$Dkb?!0t%uqb|_OF#7Mt0<$6`Vhfe!b-<%sMlLkR!4v5!?HtQdQ?wae*CiQajLpb
z*v_Yb9vSy6p({Pu1)iO+0Ua`~*U}UhV4^={JR;NV9BXOqtcBQmxkbm{3(xgvFF5SP
z1dKE~T1$GQNcO7iB&9xeOous^raDavZ>Ett=9nX>(|s1CE2rIr=zBSLn#SC!Q<!^9
zcXG!ZLq+7%ZF<TUBrB)Zgrs{n7bq*c*TlopT6*LvXHK7=3t!jyh2wOJOAG4_`}Np0
zb5G*;vv<F-&xM0xWJj_b-%V7mjqS?>H8*{;ekNE!GokIK8C2NRlT=|gvAs_n)bQEe
z^>@&ENOkjbTl(>i>bK8b(#IC6Bc3~N);xE6GSOFE!|0|yLD;XR9E*C+JTbao8UOoy
z-|!?QWKz!V`fsjvqkZR-_aVP1zJ{;ao@6jS&8|^i7m}Wg`y%)o?VG^(yz_VYzN&Oz
zGs332?6=vv>%PxPWXMol&Al}hX@Xw0#~6=qeWsn$c+EPW^h95|*SgF}T*zo&&8;=1
z2E0JE_8PpQN1%pxEoeWaVKCHI{%i4?`o4X`cxid|Z~b+reXo;<v%ye@mg<?`9_tnj
zeW6aH6Swz+8phnq<$r(PS8wP&M}$?k{I*&5@NJl~^BbR8XZB5=dVcV`X<z>&dCKWv
zqGerv|E27bfLC$@?_}b}L$fZc^-|B#2Kvd~(h}aqt_HHwj}7fpEAC!34bqdD8v=ec
z#l(jVL6*1u%8Hj=>c&gsidR?aPAu<@4vTyBTHP8Ql>IZ_Kv9ZaU8!$iDlG^a*h4l=
zDR0<~cJBF{O|q4?(ErKu)~_p=65TMD9Jq}PpYn2#4w}C0(>D1+vbE`tTD_tB*Px$G
zL~GBoddW!@NrJAgM;(uQQP4y$vT}-{W`G~rJyo!A>mcuBJY=rf$8}2TAoIzlL~XD8
zyNQ)h?}<tM4X1;5CnIMI)eVg91t(Kyv%Ymx?(q|==iXs;A?__F6K8W2wc#76?%E#?
zRx3z>O|p$I(tqRX!=}PEQlvK$N2mQ)GY{krm);$IJZBH95M0pTDmWer_Oxlu-su15
zbX<7~1Ag(d<Y#P%g8Rvbm~(hoYRja}tjE8+5ZW+o%UJT>{2BkbX;?!`+syLjw%>_X
zb45$1+0IDF?Xa@4_0_|Z;E<VblDo0NUNfyVy9X!NgL!AhYqp+;NY70B{mTq?nrZjj
zn+l<vnUE(k@5djWhrSFOl%LPu?>}@pyK~XVyb^UZ8~P^fd;D(h=`;C`_&vd6&vTB8
zitHt>Bf>eqe7pYM(5bh4TmP=diFs&s_TtRe=J8SJE1M;nqxN(Ai^7Y^u-TR^`NPlW
z>Mgw&Yhhb0$1|tCEp3~-4X5rcofq>5CoO04=P%`#D39Lj2d{WF|Dil#JC_gZVWxZt
zx!vB%ljF}#)kp3WQP~EYZF~`0%VPOJfXplcKD+Wlw^qWErj%0h4ZZTR0p}#dox(x6
z&OmOGY2$`pWP?(sf#mS5Sf#lEcCp*NO78}wzTON`YWb(J#LRR%KBBYjo}Gffh|K*g
zivBlFZQq2r$tn6HSZ9xf#K>>8wMG9^dd!gYCeP0NF_Y<=gVyVICWqX?45m@yv)F&m
zhkU_I%{Oc!%UVZg)BinxO#drlv-S83s~dTG>w%ruA*<VM`N-_n{@LVzuD_{5DZ}1a
z+4=pxw;w6rW@BgboB5pL;?0#=+eZ))R{BBIts{dQ>a9Qjc|4+yQ@`&c_EVKv`F*(t
zADw;-SLf5M1b-J9e(HFR;aY!R8Llk){&$O=xBf<Kee{nIU@m)a)(+oV@z-q>ux9p%
zmh2cT*J<RLX6WufmIpU-Ow)P1tLc>fo4Hl$?^goh?F@RF_*mTZ-H3hfW659d4%&~)
z72O`tw{w;|yHTfiQkOe4%FEq((q3I|wMG@xaoxV`x3nCDIWFYy%R@x)LpjFl9g16Z
zkJ#myqdM$7{TZm#+kblMFwon)7i>?S<J}rLv|wQ7`tjBi9?Yxv)@bhQBiX)akIp(z
z>tL>C`o+%pznz{wr(&VhE$?mG%jP7vCTb;0-_5k|c`8pnkZj+aTd3u5e<$CbJtw#|
zS}S|b<k#8q_q#v0<I}s9!ZbVOcPKmAwjRq^s}y<+*Q}WBrJsL?h=%tXi}vbBiuUM;
zcK6i8t{F@Bx_?c#3>p0<?j1Yz=pel8Px0Z*8FOga#SruD#YnQ8r?rA&H*PC;zX<z%
zx4OO4o^GW$Xt=-Gbka%W+|A}^%in_tRXe9g4Cn6(I?uN5`Fw^s#yxSUT3;*mv9gpr
zNNni{o0mL{4aLsfEqyQl<MWpM*sB(}O|krljm5HSuc8;8K|+1_dG%NR+>I}iW9cJa
z)g}B+yklJ}0YUMfKdSvMs!j{}R*gJp*gPXWSF$l_`q2E3@vQh<{GvXr&FQRVcKC(G
zBiRfp0gB`|E;;r~5UD7EmF@v??^{#K@dKhV4+0~mXLJ6&__`AB?@@B!wKJ~VXpN!a
zM``(!H736wnOpI-yc=(W=CZdweV*^AE%#Kke31O(;O~j2!>Iz}Xl4)7=-AA{>TzIm
zp~u3>acHR0r~59e0*-EO%+fzpJv}YylH2D!Bb+^&C1z4QdMzp^B=>cnGVY-QA2;Pr
zn=pT(9N}6q+DkpQw8_(6F5VMAmYOm<7!q7UA5%7I1Hbo!g?-C&YN@NevH9=o2$ODI
zY1{c9>)I#XH-!As8hWPkF@<g$!X`Q<M!m+8;yq3I`GpSs<%{EF)gPNirX4rG_>DKL
zP3@z4fB$fN?&2lkaclpJ?9=%1u=TM06xofhqJ2_}jkg5qp{1Xs37Km#sWekO8)9aY
zi7yHoL?=@>`26CeM>7}u{Ag-#O{qFIHvCTXPOeX$a^3Jb$fw`rtfh6&51RSxO@CH(
zE(N@tf5WzqK7`+tsQsgSLl|<C@{Q?i%@<j@d)cwf9VT0whx#&q5~N40o^t;FxV;EB
zyl<$wu37!&6!wcylp@uh<=$^lP&8!hY0{eIENE6xEIORJ!WR_L1ke23XSi6jvu3hS
z=hIxTVD(9$Vsw0UxP9fW3Eax1tcJ3vdw70@%qMSs-rd-&5WeE#6I-u$&bM*RC%0a6
z>f;97Z?$`O{@6Dps@Z5}UaLW*{<C|*#41H#jx~;El_N0yoG?YSShm?MJC<yk(h<S-
zPH>isKc|@(@vWSCPB}4@xnAnUI3;%QDX2$wBkM(aFi%)j*>d;M^|Rb_;fva^R?6M*
zR?S(&O!vV}j<&qniWdR3;*-=H6p2dnFZ4g%E$V14w+Uw7kB{%@{Cmq2k-^~9VeaXh
zaZf(p<_Gg!i(Oy}m1AU0TZxc#&rPqk#(#SLl0B5ST9uxR{_--hG%@QnF;hFY9N}Ru
zilUpHHW1CC>VH4l@qPbVkbNzO1O;2$Cn2f#H|^Wr*;)GYG%{G<EQ=g%U7d~n=lFn`
z^UYg3cnZTbnNjPokgnO#h|leiveH7=<?;Hs(&E+S6Yr2I1<B!9Z~G|~$&p@fuPLRO
zGUkELKJ-r}etCYjq96TL<>fUca}XCa+Us{~@@dTvexL41vV*LXZy`&jb@7v(?p06b
z<hHy{@3ssJ?sT7#U1WvLl<hJfQl34$K-;X#@7SF9<OG!5;VqKxy<0iVC)(Xhwv!bw
zQzFKkLfQLp2YvLrM8dU)Yr$wI^=LBKu35~cu77f|x!|I&jFP#7ZcFkapP9Mz#&hL{
z0Xgp*n^B3ZZ=KqImFosW<KC5ZO8!k>;n=GPRBbA<HYzKY98N!`o-*{42r&Bc*@{ha
z3OMWE9zCHCk(B=}zqoiTcvcho`C{6GB4n<83-egSnojUta;Lb~@`v*a=JtOr<JP?w
z1<~JIER>4AW<(m(!uSi*=e==VUCWw@SW(nNK__+-#XczRVV8Nr@H#R}r3jP3g)QQ9
z5{8=)Wg?7CVEP;;x_v_$CdrkL3h9tZEIwr!1=u2<MOvH-A%L@R9GRtB6qn(IE#h5-
z;Z_I%0);0;nRw$6SU=Z@Lf<V7@q`^Ecr39o+Snc@$T_0Yr>!BLSjk@Kh_u!!s>?`5
zyRa_K<1D%YNDEKq8!^LIkk+b2i5YnsRY^N8@aM$FNaH84GL8|wzEzE?<efk`BeC(7
zu{z9yeMGZw5GU1b9%)Pi^MH<+^rdV1U@Q3wH;_qeBg%c-xWF#+SHs~*Q}z+<z9$-c
zNjs+S5MpDju?@_FYec_KT0<7Pn29^?4xlMipKnYbwnv&m_i8f|HDy^BWpUT=<`KhK
zq$$T<LdJ-OEWeWvydQ5KK3s{M)^_=jA+C8FQ%PO;9r=q>T%}J67ujW=JS+rTMbil^
zhTzn?%(I8NVe}|EekWzPJ<(0Yr6eO(vx(d39(<1IrsdL@(W{}0s)QB3MOL$jYxX7K
zIJ*Pn3u}nMFNYzpC+M_?POk7FqMNcyea3UmUQ{JxVJfnkYp*(kQKJ`A$yPXq^o5G6
z_x0fxy2c`gWnc}MG(jgx_$}g^o=Z-KtOh@(lB=*CDW~D`Hls;{Ke1A>&;co@;!>AE
ziM3#LVuo)L#*&9mko#;^@IG~o&zMU2!gykE!f+>2PR*q%BOZ&nCcS&LunI}RQl;0&
zr5VDtXoUOKeI!<R8!4yklAF<w8E0QqfX{TpqK6$}*RkW!MK4?^%IikqXQZ6|iaIO`
zGtRjv1eYhkGKK|_aV%9<eKZ>DC@=QHOk^B%uOTB>a~<PFUO0t}<ET>aqtRSX^kOIs
zK{l(nv}6ckkDv6JX`Hbw7UL-JM|6eZ$Y#A2)M-CGP6XMk`4H_TQ&^I5Pa_Yh$DWAw
zx?9+ofz`ZE41PCk2P;5HK^KkT>hl?DD>kqK?6H0yEiR4#!-`3rJ|A5AXO8<hTdA4b
zwun-T>gRA%jaopfMYSl?F`f%Jdmjb^2~r?&3rNrah9GAwg^dy&V{?L-R4^?NKmvjL
zKwuN>(gzF-F!u@oDS-|%0EVdmqlAH^3joD|WHzv)Ff9PmE@P0PdccCz*?TV;_jAMs
zt=1W;OUHO}+u3`q2KTevRWsLq6ol$@j15_0QodIJLv3*Bw=Q7LVAVR^Ib*G-l<1m{
zuQ=}#O$V0<%$m7eHE1>ca}_$-BT)bf;(p$5!KiVas?m)#W{On=Tz5w7=ndi*W;EH-
zFIZyTrd0tW9WW>X!x}K;K?52~KCMni+n6mTa_BLL{}ZOc7EXy$yT;5OOD?BEN1MSK
zORfj7N*ww-k2B&$oS4WXeL7l87Qoh_qYZuo^l>{Q{<Q%91sdpM0}Q>uA8)y(6}9^u
z#heLa?^*d_>E$>MC(*dCM7IuXQbzC9K}=<;h6Pf>=na7Kxq(!VCYay?T?iY{0E+;e
z1!FKcqybEd0i6UE(8&ZHa?lag1e`u72-88x079?-;D0l+L3kO2w?HTWChJl_co&2i
zaF@v#V6deca4=pl@Hp<{I3z{QFiDd=mZ}y=QKOizM8^e}K}>q8tA@6_V<`uJU1}Zh
zNE{aeK}ZimcXj~s=z{S`(BTA~bWOnN0tY3qfwn$qzXI%hs57CrhacQe4QNjSI~Vnm
z1|cH|{r-dC&b=f7sKWtH>jIqv6c9IN1*R2hfzx8aX;RLFE}h$hn8ef|O>Is`7fjOo
z?qMiDZE~Tmg@}Mr)K`RgzJN2KLPvHG<B)oyq3!6ff=%k_5@X>{O?1|<5aAt){)#Zo
z7j`C;=-eB`n5X9BILJkM!C)E~{K~>Vmf);uQNiOS?@Y+=xq{<J7x}EAE#Y8+O>*n{
z$_m=rfISpPj{GD`OEkDHg3pOVpp-N5EKyQeMG7C*aE2AFYp~&1ARr9{D1ks00wqg{
zQQY5!hOaH_UK`uFLyPEd17HZACFmG5*uvKW-jG)m$OA?$V8o*p_hs~e<W&HWh~Trh
z2rNNaKzl$MdBq|v0Gt4D4QM3?AgvCNWB`(P0Amp_B0&R~Nehe+eGPPwir~}H2$XOx
zo<&p*tEl&$HC*O__O*b2P}-iMM<OW&$#~l=YtHb1zzFh%WfK65f`AJE;Q-JDz!pMC
z{~iE2G2dl$q2U2(I@fk~seyI~U={-A5MbtGzsvNpL~Pn|()RqCKnOjl#M@Rof{+#H
z(7v`K77vVL1Umc(wpj~+!Os#NV4-ns=W8_ig0nC%?k-^F0pq%XaY;bnGY|*_yFvgX
z5ipoh-(^C;ju-&40WcmIWeO&NeH_05pb-H3fp$6I_>W%$KpOyMc-zQk&T!h}NOH%e
zCn701RR|&FRS>d;(^}|X6aD&%-0>M3ZO;HFU~Up@BPFokOWat)&5r=XftR+YD;^=l
zJAt<~4TS<QmE3WjMkFQChG5GX$`;<xl1aXCP9aiTAWis_b*^{qV3!go8t{LzHaNqB
z>Z8av7OX{T)59>|r%vAig`CJ?+yVBx->D>RaOVZ;yI=52^5(g4#6L!6X!zzM<!dvZ
zO2C!`zyx4p08#-!%Pla#8NPXV0|4rTKho#PfO~=eBQ2+PZDz|;L|Xr}n}+MbQRqEg
zwdR>0DD<XeIf59Z!$5n)bWdNcxsx40!f!zo1Fm?q07hgN*_P$x!lE}hc+}LIlZg<K
z?I!jD47xuxLrqQ237+vjxDR*LyzyX&@V6;jk$Uq?r^0{J3EdtfpUyXZ0z+&~_j9~E
z@Mkh_?A#&zm9<u&-rQ2E@Q<efB8@40p$i;{Em3VuyQ_^Bby*0ldE@>(Vr$$C1prL|
z+&6FZ<*D#rFDCr0Dr0>&+ML7}y6J=13M%8`4GKVBF&}He(i6I}G7~s?Pu$^=C2I`?
zU4+Aot~)31R9XTDC~Tl`0b<O;8<J+s4#=dE>9JT{V#%&ElHPoIi0E4}SU_Mz<mqFj
z30m2le(1zdoefUjvlL?}D4e`OPDo1g<t`i<C=WrQGhZ&lNr7?=5TbR(ppB+3DEy#2
z2Za$7D^O$sA%r0ES}@0;TmvN%6g5yfK;Z->9~4JW7C@m!IMC==U=jtiH@JAMl4KN2
z>-n5jLD2<885C_$)Ire)WEqSsYk;BxijJx8cib)WF;Z+PB5w}k4$1~7OrT_ea-E>n
z$<c%d1zb_06XUd*IC&dv2ct^XUTfM^42EdRS5(~Mj7KF}2`%fJN(l!flz2bgeqcU!
zgy|%7^(D(Fn))^)NWhxwzgiV|dIQpcr+1qie0t5nr&j?K7Eo*}Dt0*f(TVQ5ktjg2
zz}JQxe5uIUc%^MbGD_Bh2{*uuASy9akX92M4grV%L^v#Jcl-yn0Z3E@LvwiG2pUzu
z+5xP~KtCW69O4S*t+{}q+t<K;N$}v_gSE&j&{qNTSg>D*6AV#<As(<PACs8Lo~mZ*
zi$O@4k7;6P5e{6Mj+n$Lwt)1KH6{XIMMV&MC828r-NVY;pe~XOLKzTlqQKC9bud`p
z!UG{bb|6#{4TheAP#gdB;IS5UL#If^Z`mYl`$qZek~LpK5W=BGdz$NACW6C!jhz|Q
zuD8t!EQ;|@xsIawd@=Ear(EW`)Y1gHii!!2Y?QJ`s1`-$DJ{%<dyzOmut6zvg=tdA
zp4^A+u8jg<2Bmxus!cJ~ki<~$%?R^e_9G7DtrBvyB~i-B?(9(lQz{fz$FhLoh+G=v
z2TXa;;D{19LjxW3fT0gUta9?X5|HGDdCz$e2Vy4(IrM%gWyuZpDEHG$hWaf^<Wh$f
zU^D>60ZO@Log*sr1j}%|E{I&J2_X)6oDgzm&N-v>PNEnBmq}o|gNn$dkIKXW7%g%s
z^$kNHr#6Kw7Ngux#OF9|69+^|0o(@sR0rxffS&^X4l``GM;I{Xh}SX>YxwkE4APqG
z>PfM=;x(NR{IKQsC2U-o=shA%wBl8Ux0(b7+lQxS1rWa$kP5mBB-RL^+YUD9gN|$>
z5Zo6-4$_YO1s#t694^oa&+t~>*Fg?mAFIS`UPttEaxtQ0qcRX7`<6(|+}I9YGtQ}>
ziwl<3^fH6!zpn(scOVqx<f_YPdXqs!R<pwh^J{rT_T<WBG*+h9T2F6VEQl0F=v+%A
zvd34Lp|O*MBq&!Mwka(Gp}S9q#E6f%5;T7y%GkKfu}xVZyaqyg5NczwC4?kaE*Pfi
zZ3a<e-w}yn9IGI*OIOU}HGL7?SnL|!l!a>y{aHh=f-UG4j1af>8MJHAfHSQJ!s{T+
z1fk!5P#1tt-ew@wt3^OZ7IaL&X~h_D8XGtbY<v<9WcIDWJ*I(~Dlp>;?(r8Zn9&9^
z@fqZ<`*L9B7|h%TGxXpb2`G?xt^;Hy-hlh!0rur43I-RzAU_y<a`$|@W$lcJW~M}`
zhkk{|pFY!NI{FzYvz^NyKvb9B#>ejiCL^9rUJ9cg>J0>zbbvqv5a0y@l0aYs2*?6~
zKp-Ha0hsRqQ!;?qsZ2!EQexE|cUj|mmb95tf5yvH%u;RRBhQKG+wmB62^lq}v44*O
z5N-DWa0SmspT!<O77`<~u;;v;GU|l4x6bB>4`9?_+L4Nuar71n==tkK6n>|Sw?EI~
zia(;)V%m{>FSFqBD4=KN#&${z4PdBYI!|Mv@i2N_CNGIdnFTk#fS$2;L}C3oynU86
zG`=n%Rc2w~{&q^b8NuG&nhgM%G7EohZ>NMy66`5Du$>G#Eb*`u4JI$4w=xU1A^|<$
zpAdzw8{zFK@-cwP2AFzGeqq-FCeKodo(D6W@eT6tWHwIRwre-N@N)wF9Pte@@iH6R
z(nL@F8IJfMsce~zsmt57ezyp7)BMo*pqdl_+y#I(VUCHPEk5XLhRnuKvh7;+O?0Ph
zAQ1nl1r*GvPT6A=P&@<+z&Qr`e!2jKD}IhCM2YEO$p|R2(VbrB88TTrG{mip7WVkX
z)B6E3i)Dm4SeP<X|Iw3e|2|isC}RU0M!?wvVv?v%uW1dYey@80GX0x`xd3<0_{F*i
znXE-h;+9MergP^M4HOdFz&A6{_{{~U?L)ax*7(giv+cu@7IbI-m>!e7)AfMUj7;K|
zS14Ef=y|w|br4NJ<G%a0|GF%3YdnZBS9QS-81H1P?U@4_J3_bpV_OJw2j|L6Euk-E
zLeE$NPLEl)5BJCeP6L&gTCzCeZS)Cqqh7?V&TU{LBYHA%#_!w+$WSNbmR)efH=oDL
z93MZJ75v%HRA8|C!zSNf2{o-FPP*JXY=8>Y;U``095zHT>By2Ue-|@AF-pZkaQB9w
z5Zv{lkDy?=@zWVuI*R)XUmpP3T?kplXnp}4)g&Ps`+BX)*%PcexbfEMS$c~5<GbWw
z?F0u#_kUywdLAf2y4<zPKsK1TTnEAN4()JjnB^}Tp-OcGKY%tXa3H%-vcdS}b_kAn
zNSj+jKA(dJkYqTjPfIYD<-{fsMmlye$jhxEnP&z^g$Ojoyk7Hi<{_)-&oeUu>&Vx;
zW`V#1$=#JA8&qH3gCP7gJwC9UXa%y7F2DXN1`0XpnAu=DH@+D&4Lp{_uY6#Qgy5tH
zw?QETB?goy+!}tk8aQf0!vom4R-iN(l>V<#6KLEOAR824o`T?92em-y0wsuBV-#od
zpYQ;y5pE5p{1G0FnmloCKn~z2cWu}I#1LE=0kUd<qR?K2qo%c14FTEAWf@EVRlw4e
zDDeRl94H@Ijxr?j!Bq~6d8R}jD7jz>=BmM5HI5}9Yg%71kT>Mz>s{0F7*Ntc0iF`m
z@gz{<K*<8-6)0#>-oD<|7*7Qy0+htpyGG-&;3^Z8a8R(XcU6yBNSCv|(tsjKx*WI5
zN;b&2+y*{Lau8h5U^6J85S-DVI=99F?u<vH1s4yz@S4oIEN2)wJ;h^=kk_wwwWWY#
z(>`V=T~6NRAsduj9)hs14LNZG>3%q>S@Sv^RjPU25a_#Zgo@M5&Shc5Qsl5SVdQ`Z
z#=)p{82>V_jr-%1NF$Y+_aCC=0$xFn5$vkF1n!t6>`%x~E_?2e`W<kU>_!c$5Ro|O
zF_8l>l6gMrTjv1jL;#2bVD#n%ZR+mrn57s=o{zj8Mk;1HAEHZBG^nhE-$Lu3il}<G
zKY{$bZ^R9qE4X`pyzWh>N<8z9!Jp7V&<CPC(9N0r=*DbSDq8t;s7NS<D&o{K27ed;
zL2qpusC2Yw14cGrTmlRwz)*#3ozobpBnjq3^Vw)onK!?ORs_RT5l3V(`0WF3X!lG3
zU}#dAp9397Q4sWeF#zR1M)O^y07-EG(#7D{^tqw&@_ErR=kKB0hh8!GIY9ul-~bQ<
z0M{7&9}jM5cuQV1xixSEjd4(sLuml+LD1jn0Yeh5a?Rsobjev3w8B?Zl{iBkgKuHx
zhF(;|RnE^esE%r4q1&5W0L<orUWDdE2OlRwx0fYj@CEA-w7xVqG|CsQLUE=?HMVnI
z711vcgU_0Rpk;%&p+Wm_mHksqz@Sk@^vK2FQ_di0;V>hWj#FhSTCbN-<UETiv5PeZ
zpGdAs)Ka3V!{kNxXz)Nqkq|U(Kn(u<WmV!2yBPdiHVFEjIuDfVJTE%qFdmBh`#=@Z
z<{9Iqf2xkvix$~|=^8r9IdOOOEb^(Q9%bZm?_hTz4xz|bAA9r#;0Z9o{TyyD46l~q
z0;}I$)zqj-0>ps{+0NZ1L)6RR-a$zxe(X`+5Q`C^tosW(9RE25pc4){I-pYt!oGYE
zMuE^W207}rXqeEDC7u0oa&M9pGGDqVfaCU)^`la)o2h%p(<R$yrDV9_TFS0k_pVNx
zrz@n(O3iS?XagC8%to3SB%CBPU_MS3OqxLE?H05Q6QmOw(mqb+Oo~J1DdtP_xc#{O
z7**?->sEQX&hS$Thw&bZ?(7kZ@H9x4HZA<HQ4)<A>zmTCK(d=9k!L-JiB#wlyRc~K
zjA8|~jTfa*+Pb#7CwM$#-;|bGpnxAe?Q-?xI^u==CJQfZdIOfv`a+<>|Ez)VSI!vv
z?!+K91L42Hgv89&JtVTXd6^Ih6q&_pdcNV7KFGsHar~U<lr~eDAy#HXfqxt0P&ZEG
zC_wt)we{v#Fv(+mBe^e6{FzA`Lqo_8`{)e5l}X}W%KwVKeE!%T_*kgp1U?p!x;%gS
zBy(R%&t!aD%dDe@a8x4sdHyWU=#$NiUBL;_e8$gJR_AlW(p{CW9n|p$^w>ymAM&je
zw38O3P@VEMY@}oS$V_exeWH}nx2X*!#R|bu;Qjc4UX^fQ=@&D&TE~PFx+hDprDkFe
zH(yevt{h0`+umlaI6R`nwyo~6MjZ?$GlYi9Bk@h@czb~pY$tPAf=tD#@OEu+Jhsy+
z<h4xJC<z%poQvl#lbovIvGjqMYA{#-opXw`Iw#H9MGE!^1|}7<gBL<Ph+!|>mMl4I
zZ2yT2En?I_1Yc^0_-7f3Ra|(_5&;W+#fNlYHz#&+!&8=jBGAJ2c&L2`ru8Hc&A08y
zU{37SMhLG8V%tkvl*l&EOe$*I%FyjS&3a^;2e&KmFC_`kD;?POscZ#mzc47Qr;{DI
zltv)_r1wCpd+4ynk7jF;&Gd@FD~uNMf%B^#miPlXtjzSu1aWKH3Edf#t;-Z59M!l+
zR#yiZDBt1!U_X=dax5VEa=o`4srUG0vZb#PkbjwcA738SrCeU{xk=j74JS)MJK(<1
z^A)@tvr@cNxx+--vvC3uYT)Iu^_Bnda_kIs+0pMl0M!A=Z1iodG(S4T={65>hYR?G
z%7&}thp15<oxxOtrv<tH?u2SmWgqay95&h*{ct*sgv!m;K(=T{;R%_KYT}?38`3fU
z9>BYsDPuyx(0681EoLb}7b4s}W292x#`&(lB7(tj^*S=;^JmCbMi?%7u`w2!wWtr-
z3J%SWUfj8*DwA!)^Y`dfjjXOdQ>?j|5%KTb57TzAFCBnrXD0rPZNTT!`(f4N*IDD4
zCbXGoPq_jR|7?iDWhdN!f`02?0{)@PpuaVEZwmPmDz(C*>OIUFQ+q-SY&TUW5BPvB
z0lE<s2W(m7)-4Lx4(5euCzII_FKps4-$RaoJ5%u&_}ra%wE7}*!Q;grGjV^ea#W&5
zaR}<|$b_D-X$<D5UmCv6u#=V0^H>grff3Z<B3wsM971nfRg;iG)Q9p*yPhKkD=HV>
zp_4Mj!^oVMJ5LL74*I>>Y8F|}&5xV|@{jJ~I7D{}ut@@hY(<L=-E*Bmq{p@sI!PKW
zpZ<07?I{d5KSu25;P2-QV{$&)*j~KX{PC0&f$-?F(NOc9KUGEGy0>Yt=<_ZcCADK-
z8_aue({s2;#l1yAHns+XbEHVc^~Ew4wiEYrEs??aqhdV1IbBdyZGY-?1c8|8wNX|J
z6bj>~UH*RRgTS3^k7Cgq-7^Ym$J}9Tw1oX&XOW7{g>Do&L^A9iErD>_3pOQlu<axY
zs#DK+22>oz@uJ$z(R_VR@Lki{7tFjc)CKdq{!nT2;C*TQ-^v+H>g+Rt3X$xi20~Zx
z0xvr8sK<<xcPF(aif=>VenssS6GGPjvG_mE1@JOO(*@BmLG#r9U|q1y0^uOHQw8>}
zqS_gYwJE&J;~5sV<&Y`e$3&sz+ju(xdQ6+81T?D7O^3p3><z<01rMypK@<oq00DkH
zJ}%KilR`959@aB|1t6b+mM7RY%W77j1(-J$2ein-Cb1O67y43|Ha=fyl=+$tMdfec
zZCcI?>v<|EQc*nL0JQA00FEX_<sl1fjAo|^PY6xi)YaQBK2;_nXnN#zD9UD{GIq6_
z1z}?&+h=$}X;Rpm3~z^KdDwE-X)Y1h0=%O-8|Czvc8;$Gj2UhNmvKSa?1CW~VKFir
zD?^YM$yqrs8+Q3fC-Ytb#WllkR&+X-{FY&4wc5(ue*|QKw{{pesmz5P1h=`<#-lN_
zsBZ8h2sksE?$!W)$qXS59^LWz+#wmv%7({}<dZ(`dY-E)$<BA8l|33&DC$l`Xl0Vk
z@PE%+v<@ZlJ7rYox8I+xz6qmZzqrP8dM-(LY7Klg$!04YdT`+N*&|?SDu)dl^IX{0
zwMJEHFX0Hjg{KdB8ZVr2LN-I1cFm5c*8WUQYOh@h^W6?Uxc}K0?%%PO#_T~qM?_t^
zI<@Q$h52qz18xAjqiH8WgKBjXV2>EHRH1JAn!<PNbd#rN4GvEkbHE8zc5B)>0(Vu<
z!s7WhE>3VlExekuN1+O2m8YycJ=+f}mTKbhPn+dABbu#r$z~?#;D=0dtPz{DMiuz*
zetZtSJXb{j2`SI+zhvA%n+>}4;GZ~8aFWN33x1j-56zsQQB3P<8Cyi$SsbL^Q<i6U
z;u?IZCKk6<CxPx*JO;ke^AdITUZ5Ua>S5NH6R*K2FJ5R+WVXbLZJ%%r;y1H3*;>L_
zV^7Z$#WwIBI8XIzYzO0*BAp+C%lR~8MssfQRFPt)O#q2cox*JaUjudYPioW2@8}O6
zriP)vTW+w0*G&R9>vtt-*REZlRHK+#-etiwsAavP`2snWsb#S!)qVuwqZ1sNQpf<C
zjFN$QTfJfs`Evud+h&MdiV7p-FrTwUg+5ZF=)ZV$1m`-hw&(J%!;<P+%b9&gC7+ep
z61v|*7y^#!NcF8l-C0q|*-xU0^NZ9msAK^i2#25*{L$OXeOR79d}`K?MK8z#mit-@
zA&8_~1%@bd0Z0LhUxUHx*OTk5%MgS*UxuQh?~m402tt6&$q<!U_CkVm?1*g909*?E
zxBQRFsV;A~l0x#jZ_&UL1OzQr4C@<T2&)ErBpduVFvWcIww{H@3<gAWb9c$|S?U>z
zG`%2IC2X}OLO42anHeT92qt{wrZuij`-m`@rHc`%iE!oVvf{B+SFFdq0Ip3jt+yfn
zygYC$l?L3pmo{_ANgJcmx&O#c>Hq<Qq}Se>ISfEbDS&K{BLcXZ(<Zymqmc}82QHlq
zYk_;L-y<gQT}+m>nG9J!8HxYiZ?JO(1<VQ{E0Sq8I8Sdr$+<gg)5CAf$#**<=th`f
zP!1XCsaG)^mDy|nT;}dZtU!dM#kv1ZPia#flya&DA5?L737CA+rJnI)tT=35g@5xi
zi2t~@8Ug&(lQmqx(qNc#iF7QM>^2YH-T0Y`qHnH}Jy`|){WJsA)Te=j*K2AKju3?8
zL$Uv&q+paEjMip@)^%>MOBL*L1-r)o>q-JGUkH2Dt#zJ1=YAi+odBmyv1FNGd`U;K
zqI@7iEKA>P&|hv!WA4bCD|T@x902+Npu}|SEUVJ>7f3qGWJdw6j1Evx0!1@!EBF}Q
zu@mqHh=u{tcpw_^UM#DB4sfzqVi!eU0tFVgrIQ7Xb=nqlmWguGn1jh^Q)hd!m<M+#
z691iC$F^;T#W7?}+(G)g_hZ`{DLJA7nAIqD<T4l*LaxGkPHqu}HraraPJ;)kdcH4P
zpCK#Qs5?+t#&V&gjJ1de8Un%k@dI_e{gzg3BI1I#Q3qV|{<qWum#+UU^}r?le@jg!
zbIB~3>BXzt{@M2kb0K<A!3H-wk$}d<8vLBBJ}X!~3>b5`H3Xb?>Tt#Pi-gO_b?X3U
zoF3TDlWbLM-=S8w?Fv`w1yr(Zg;4V4jX@dU3d;|;!kXcT(8<)lmhE?mHh4M$@h^Y|
z{e96&2LLw#kOzQd5a<Db6aWYKKReA}oP$=TffE0xGEC!Izx#clh{xkMtPfMjo3lp{
z1L>~#50dh%Yz;xPMj{mrG;(ZFJ6^~~EiCbTN0`R7rHC?ocbxTM+U4mvNeEhd2A;rJ
z^(9GWV_a&x)^*14o4}W>%L|@YNPFhg$nZaPA*kFLqi+W_sh68u_<{El|EU7i$xqW5
z{3~W2==Ewt;JQtPO7uWfwWn7QA}rYg|KW5L3t2!)^YqM9z*D+2aYD&0*jCGPMY6J%
zcM$6^NuI`YropA&CfrZ@FpQensj8aqYO9<`#SNN$Z2RI_I>Yu6Gcu*+3b8zlkv;xw
z^-jQ=0qyqE)*G2)F5q5e8b&>T0dG&eL-h0mZbS)EU^|;0DKYi$a055Y!gxM-o##eR
z?L1Ij%j)DwlG&=ElVk0g4tQ*o(6sX<hdXG|PTRBh;?k@B@T<W8_i1dpnZiD5&hkL8
zl;$9n3hiVymTGQdWi?`AZf^cr&ZpSK#Y&?Zk-N`2*4j~BkES&LZTn<f_;>4riTNuJ
z?DPU;!u`nK3*VLKj(SO}u=Zuz{K{&?{+BPVwodz%*RJ)}HeFm;t00IbBU8T&)Df0P
z(_u{)XPaRcC)q4F|0z@4oVoMq3(F+SjWcVk+L`<C=5zPwKRAl&*;n?{nt*@CU0gF}
zZ9vby+<L_)(K{{A|BAcsc;@uY3N7m4PeHH$RE8sXv(h>IEI6K^zwQN<SqM2;SZjs$
zh+9c|Y7|yjN{SgBXaorJQ2F;tWCTqot8CSo-JPlq*8O>`ry)fxt}FO3h)B|?OunL~
z`Dcla^@qnBbTO@??M;TL``=pcK2)NAp}<X5J%^~wOkk+^q9t*uf>!BB_B?oW>#Tk;
z#CGdgy37U<HRY09KKAU?F!!S=-P#KTPsmsQvxy|?;Rp`jJ68j6eFDG6BP(7#GpHU<
zC3_7|W2NW9a}IQ~I!BBMYY3eC?EKd4Oj~;~^x^M;{u{Stg=5E^2d@)<+LPKA-_mB=
z)nJv@I&{ENf7jF8R<>qnn0YbxTUt^Lee!fu@K3ql_t=XH4fK1?sK-tBKONw$#g^UN
zFWp!>SF9M=sFIlYmm2<yKEJ2G{P=M%?&;vIG-nr@Pq0*xiRJW{m!m1eXch0>lHt9n
zRE$rgNIn)Yr~UUQ>R~S_e2j4*AjhJ#(dYrXCg58I9`5kz_otidg`<?$C)pF9-bUoq
zN&lICrDWBxS&~o;>*0OP%l`UKoQNQQOQz@=6Cb98JmqWKt*-gYN6I-R6yGvKgXFDG
z?<O5>5%_Aq#dzpL1JKi%RDnZ<;||fJ*){g+=&JK8quy?*zbH()NqwJ1+DFtEF&{uH
z{u*?XbydB5zwP8Dc+PTm2g6Ou@%IA@yV2wQBjlbzY?tq1+V$hKl1JsTsbL>-Ut7Sw
z@U4`f@X{17B9laa^v@GcGcNbPY`<_Le*<lq5vlvjdTy4pQT*tlV7NM4qtS~z(oZib
zUQqAqV&oao!QR;VkB!x{9f5+jdIm`9tRC0Hm_DxO@rU>0+4rhoPgjz1XmQnW?dW^b
zam)9K&!+Skw0E#t1W|7#m0s`DM_c0E0%IIG-1_`4SJ?+XkFB~3iTvao6ufl&lUwgE
z_q7K>R;cRFCWF~Ud-4kb`B!XFS4p5GDS7D#_s>~(%KqNl497OSVkUj&_C|D{(dgdI
zpSR156(42(_?5qVO*LRu7geL(ieL$p{~}3Lg`F-2y?TObr~c-1mN)1vUp^UCk)6ty
z8wB59zZZnHV-%GhPbXO#NZmE4QcRDetm017?`tUNRveJ}qUT74T-tRp%%zfjAzybk
z@Ik&^%8eDWaJBYkZ{@pn$bCN#UONu`8iA}2TD&*93al6(9v>0ldr?XIB)=?*l|FZH
z{D#Ebxv4wM`1l}2SorG9lMmx&^A$V$<Y~;W)8xeNB$y%YkWbu+_QH;t+<P&tPZ22;
zc7Z6Rv^)R9@teHcUg{>Xs*VIXzIMd`vU{iUy`gR|3fkt^UAc$JD;7bQHAHn_>>oF0
z`#)7$Aw6&TTyBx*;J^`BSQO+lBlNm<K+mA>SmCy{WK?eZQBMFxq-B)&y{j?bA(wPM
zaL^hU)mKi{>fQaR9Xun#z>|Mqd0nWe-lV8sZ)4QL)AoTaW_d+B_r7XUad9j()1aRr
z?Ss?)o97>F`gE@se0p+@gx<MgAx;uG*K<6X`zwThVEN1GAjHd?p`=U)UHtL<*)7-H
z<12wx@VkCq_=y|HK`T<u;@?-?tKa?Io2xP(O4W5-$+_3{=1r;dTL-wV8>N&&3ya<7
z`Mj|YmNvz|1D~szW%_rP9a*>0GxmE&*auluk!X7*k{~oWcX}iA=-uA3U-5{kJ@Yr_
zaQG=Qg}Oug;d4KGWgP5@CTk|tGp?wA*t?;^RPcJGb~o+7l}y}Chp!Kg&DZT+oF9J6
zCW=#DlkrF)pDpmu1imEuqnm4c-`k9|W01a8oaE<u!ni)Ud=nQTpZxtflB{?7Yxb15
zgusg0*NIK>cYpUAB(py;wY0F9N(78H{OzWv+50f**dnQ_6MAqyH*yb~_dV{fU(>ra
zX#uTn=4VO$wrEwxZ7u78AD)KC>t~O5==gSau&{sEOAd3fOIB{K?^>lS{<7KU_B5(`
z-MFuKw-BN?usg4GMT%9L2f0vEXnt*Eh1VyRF3GXay=Qv4L*SH0vG>4L@s+c5R-vZK
z$H;ZAw;uEm0kI+8MBan6YR0ks=S#(&R+j=#p*BISH)lI!JB@!|*_X(f*r-bVv~%g2
z=t9T$Z0IGYOS@DEHK9~)Mrpe|%e3g<zl7jOMZZIq;@?l+f0xKt+yA!4s5~%*?w;Cz
z1Z|PK+O>EMdgN-9qaW~6#Nr;sm+5tKrC?aXw0>IlL_E<Wop0HrStYH-1O9}2h5U!>
zaI4ZL)J1EF?8M4AtEYO!>%Eqz;h}s;;wD2@VRDAS-7|$6%~a#NUn(OTzST^XL+bZN
z(mtClh>h^9*WTV0x;-($y;x$k!8$)#O;Q`EdmR!?|A{g@5zckxd5mqCR1t}7HPhio
zh*aKjk6q`CUQP!0pa(CkNW$#r`nb!~?c|LIBr=m1j2+XQpMze<dqBT1&)sj4O8zCO
z|8gU@u~q+mPZ<>|a&7;r+QX;_qq;ruOr?{X#CUzKk?Z*nY_ZOJ3k0<Xn#<~T<PYD7
zB9eCXQMrYJuP&EOxE{Xx!ixL%CCu6wU*+>rV-z0)WtLTdsIrcV#Yn0sy=6a3pJ3Pg
znP8>~-^#GfoH?SvmOpu1rh3V0y!%en_?;6hyJGPkF2x`b{WNyh>1Kl}CZ*gvmT0r0
zKyS{`5XtNMT$RFs<CoaO{!c1~_doLgj4Ao#WTIL+z!5uT<b)`4M#pScdyjGOCe4zL
zMqx^dkZxLWrirr?ms+z1=40H<kq)Vx{*2#7=Zh9J9N&9r&+ff)dfao~za@3>_o<rE
z#TMTy`-0E)Dw!O*ON_Irw*<4<ZHsMS`?tHx6Gk7FT}tqQnn&>yNFX*>YMO)U-J~`D
zu6=@=8Czv@Z&yRjlW=a`WLs7yYg$F$=7sVYe>1U4Ro?vuxe>vCMMdbX`N<<qDPxC?
zT`X=iv|%mD(magg>51*7?(0+yW>k0Ssl!8MNhkXM>=`MHmQlWe&PeG%1<Uv4s^PXO
z;kz|YZQpCxGy2b8{hBkqK<8YknG~W>@~I6GrLX7LUB|v8?&>kP@yPZ;*G%1w!_Tj+
zrMMaHm(sXjVW=CoqiCZwB)ytLZ^gE9ndJum8GGYx{-*0>#mO&{#Y~*=)G@RglQ)I+
z7=}p?M@*1RE^3jhnYno@B{$bCk&dP5p6t5lo-vo@XX?o#;?K^+4UNUi_2k^1xjg>-
z>}RXlS1oa4@it2qT?3{x3wWTDZx?6i$X3YpZjo+jr$8;u#Qu+gumFuggrRlfkJVkR
zh_Hh@NoIvhKVN?cz8;FF`!{$$?uO*e8MX}7u<tj>J_W>M@Rww`DHQcE{<+y7V!x=p
zpe}1Wd!bvO*b^OB`{iL4306SwC1>$fp{OKT<-5T<gv%EEZ-6s|Pe|=i`c-k~{GbwP
zwvk``%fpt@yK21`)l@}4FF$k4JzW#Z%AQK+OGCc7<JlbK*7WnS{IARcHHD$q3u3pm
zBv%HE?y_{(n9J9_Nh46PEsfk85c$@3ezUwmnC&&U(WzCuezfe;x}Vc=H=*?>b)MI|
zH^ZZ=hE5$EDw*$Sf`c}G1U}yitibRcI9Zqp@>UkHrm3gxRi(){JTPC6Kq6iSn#)OC
zZ}Oj(G}XL+c=y$r#4Q8w>u1xRgVP@~cr*S@S?`of>>EDsWm(`wLHjG)cKYp|4#?#K
zBhzLs@4k|;d-R~q;8XZSrBd|$4?*%j=<0t)w<te9@O9uYK*%(X4fy96*r=j6>$Ob<
znm^$EX83s<U<_lAsJ6ZZsJ<#(N*(_3p}kppdWDM8goEUflZ(DR?Tc>}+4|)$Gj21j
z?mUHT5qim@y5-jqYLHtI*9srrkit6!XZ@)OpmKuYROV40u4*xTV+@LR5Z@1acXRgM
zlkwBC>M-7#`yd~_-zqw!nEhiS)Q?2U_;SZ%>7hru5A+rr#or45n0TR3xOl&BT;Wd3
zPUdjwxSAj=IX!}67xQFESp8!Awf09&FO;vzxSFt|npw6To|OEBG1@5P0jGj~@FAtP
zkKqAbakKAkemdP<)&hOzph}mFtXSPA7N5*Uwb!LrIsA(^F0XVmmaVk2?h&+_cCna}
zAkkas5l9{_Z^d7DYEgB|@TcVP0IFug<8b&{@_UOyhB31HHwUu(kWp{Sz8{WXr4v`A
z$ySRGYe^TA?v>LBeyv0L!dXliiZdD}9b#T=s})&MU%tcgG>QG`8;Wx7z0d5KE(ITJ
zw0}64FzsJ9lAL<<skXc{F?d2GHvM_`mm4WDcJM-WJ5s&b$g{4RMu&x!HdhZ%#y0Ak
zp3f!yZGYvC&*pEe#;uYj2^<?X{WYcYWT7)f#YVWcN-Dr<-tf;_%+B|0sL_bimkw=E
z&l%o(;+<W|B%}q`X2=X^;?%qSjXVDltWPbwDSy1jUCm13*ha*SRwvg~<L=L2yA{k=
z%UzyXJZtU}Obl*I)Tq;8Hyh>`73)nz2*;@EOX}Lh=lUK6iI3EeA6P!X7)})jT&nt{
zxc9-bLi?@WD6^M%6Cyon`BAmwMB*m~sW|)8q}cFWr1PJN_I>le){Jg{xo*yp<CD7|
zqY4ERjt$Tr#(8F<*hjJf77?LvyUckyF~RwnSN?5^9dY`G6`{u@_#=}6uO_9#OP{W<
zeABy$pys|@U*DFw{_p{V!b9(ZS9F4QGuK)uG*#KZe;0iJR&m}?{VEb(`$}&nrO^MG
z^kGBdZjEC!Ta{;PrS5Fi=NU7ixnY7aRejc@7fnaF{JZ8Z+7%BTD@Z(5b$k{<2<9&e
z!axark$Q=D+D3Fve6C~K^8|jg)ZhQHXXKU0F7zjeW)H(xnD$T+R$(D}<EA^BSaieZ
zS&X}Na*Eu8kJXY~1eK)kFAd3ce~D9lGx!>TaO~T@|B$EiZg^Up%W<S1`!8Aez57Nr
zVr>#3osll=(1)*_9)85pmI`QEbX2yvHFsQXLVM@_FgrF(mKc$q@mp*!o8J4?Fs)_!
zCxP#R{*mC}_cs@<9WNe8zOH5@A3tV^6ZmxeEYzzw{_DFTD$C^T9+a*oTVh9{nyQ!y
zPwJ}Wsf&{URlCVRdzQ1@WtZM7J_r0zEnb$~m{JDvIEi%i@Nmq&z~z3O{y)qlyeqd*
z5f2sazAkmY$@N{NiRJ}~S{<%Q!H!($R?-cLJC5ac?24GoFU_wTx&o)7)zgI{CK+O0
z=Qvl|e_rR6AYWbk!1!AzINW#37-?$kV4mowa{rotSCGz>;?<&j*UL58$NvK_K+wN!
z=oMVk{Cm~KPvVtDNi0*!KJ)`obf6;2_<fOod`7btQL!)bqoq9}oAeeuzlEBLlAjOH
zqb%dxSLleK62+rnb{;1ucqhe1q^smOIwCSd7Uuw0fmQK?!HiFcKNFnA6KGp;CpnvR
zV)$6Y_P6%z{MsCuUz-P8fxi{tCx6D@!`Gp824(i%e6jEJgafH%5f6HnZfAiF?q(30
z69^Y5<>&C*<#XkEIGl?XN~MJ;{U8+Y&&}aO5)SU;2kTG4R`Y@PKJ<4l6+Q^{wXt<m
zbYyMe7<_9pIgCxX5B4nBaG-HOETCPl5~c<c9FM^81*}&-m|@@PycmW2F#+xeGW3_e
zZh5kJLUJ`vg!b%ejz<j34OKHpVnee%LO{vMs%vOw<VIGYJH8fM#>wxx1dt6$QA(Ds
zgLo-wV(RvviG~p-2RspsE=`1CmP}<`*38yS;y_p6#ipi-8VWL%s!9BRezye_=dY@Q
z4t7tA^?}F9JnG<nm&+tJ$kY+lolm>JzY8lDU#NtOY&e65yHtRKICu<er4N0mRNLid
z?KxeI@h%*S^TJioP4g@cL-4@{nCGpynOizIXp4AX)amu-qj@?Q?JcxT4GcSIYqio-
zrJ828J?kmUC*j2xi25{#2M>gz)dvO|Km#zDTKFN$_pJ{dX<q}w7}j0km?CWxXz}6r
z5!DZle)H<uY;=q&>E)6p?%=rPXsxu1mF!yHQ4zX@NQC?FdGw2=8sJQP>x)OBzmPKD
z6zV`MA4jEFl1sV+wY3F8%f_yqX~q2eY4whj-(uY?DD+wE%5x<Nn6MFIA54#-3cO1w
zKc~+OwqP(4_*gR<*a_%ONa7QuGm$(p^q-^kGQ?!blyjXVJP-CfN?(!73DT&~={3Ze
zaA3qgCM!z_>9(Z7KMY})ly7q8F01kz77@E`37@Lc;u~a@*C#yB#t*I0xJIUdxffxG
zQ{QC6dUaz`iF?D6;)mlo9?^;;qI9@E#H?s2eDge+RMjd+Y<IjG`LO{MMLUnziT=3M
zF)+lyWT;K;Q|v^L6tc=mlh$dGMQGO;RRUef40Lv#_6g9xu_H2HOa#0GeT?jE@6K)n
zaqq;MZ4)Szc;5ET$!c`0rARBHxFRXfw~t-5pe!A)%T$u2h12Wm(hWDQkhPm7k+eh^
z{>4E*Yv=WXDG5EO*xy=3PXKCtus5Mz>=n@Sxb>peo6UEO%(Ze?O@}j=vl<CsoaA@{
zy7Uop7bZ6<Nc1-@Vq4<WR;0zzdD-<n=VhmWFl91h5;PMhF(v{g7Ghg~2SdRu9k9-!
z4b70j5EezeNQ#GKQ<BeYV#zy9nhJzCi(29q=73-i<xS!8=#aPJ;rgf0;UPds90FzX
z+_eb!pW#4Bsw!;?1sb^6=DNJ9%pK9}P!{5tO^Kk|?56>Fd;;Y35RzvA?Q|yRFTD8o
zixAxc)Eb)Wc0u#^;e2G$r8P1s)1N|#;<U@fzaC7-{KhCLcyG#I&F5i$ahwuapF|1H
z!t+sPhB1>tJ{#Uv<a6juy!XqfQ@#vEtg$iVmr=ER`Cfc9Y)3QM>J_7=`fZ1R@^lI_
zWJrK3maNN>t6Xsp*F8n9zRZb<6k>oVmnl~~KB6NC^8=R@v&Z^LFY7b1>8%cSlZ56h
zy7^2|u%LzkkB0>dV7wB!nnH<ba`mQBQe2^2wMH;zhQUS*!xD8wu+&L{#mW-C#{54R
z-80$Hn9wI)7}I}m9Q3(4pT^_CCl$XD58_%L9IXf_Q;2f8YrIlLxFS6;t#g{fuP4X5
zhKU^e<%wNZ?)OA{|8P3GDB2=3gqb4JG!A@CR*qs>JE8{iA{p{g^<A>cjMJUm+*H5_
z`#Q5^cfioZMt}6{+>t!E%goQO%Sz7szX6!a=_q&#@3Ch5CKSM`LGST|5=Z*KFz@_8
zaU|)uzF<{ihd8~jM|*j3x}^YGOIjN10}t;R;V>D5DXQwO3E)iDR&$d86LX(WnQPD~
z_HJvMtsPDx@nlxsRg?{s%!#s*@%tOXpYZ-@0xh843u9PA6B}y(3`0d2>+4&C4i#G(
zMx1Toj5cpyh;^3-dJeT_l;xq;TvP>6lRTsfM%ww-CA9O&T%Xp=zcxt<C$;lOCyHY>
z4i)|e+f=L2+YeD;as!&s(o#RcBC!OM#qw>j`ItCuqg%9#AqTAd7-uroRW_ANFi4Zm
zh+F6srszuRe63)(|2~|HEh59e_~EE+gQk$8lc!eHkZ!(HZS}f-e&@5Qh~oiKZD%Lv
z15XhRrBd?O=jINcuXb!N%5UW3a8H<n&x{Wf1bhqESQ1ZGYNaopL}YKv$HtbAz9LVR
zURJ7|M7g*%Cx~yoERQi5)RXrz6IvK^EYzn|qwZNtacqP7EH}6|PmpwivsfV-VgIE1
zY%&N_^w}iAz6^aftUI{1fzc7XyFkVQsduo3(FTpWW27Am!*&Abj^4mVU^ApicS7--
zTZ6H{G|j;q@+8eMY9m~V?wCyBX}V)lmvP0qqhGM-hNv*HW0<5mz~;F4or{3)At7S1
z+iJVtYpQL)<S6MgEf|$(p9RIl3C<?DP+=$4S3H=2V4OsPb0kcqZz&dOsESgTq=l<w
zQ<RR1e`?85s45yUD>o`i=&xyBSzEI-lW4|)W#3;3N|B<dM`T5Uj|Q=nx_DF)P5G*M
zS^9k_D+;OREk=jeWZ81ow6doicD!iO3nI~Wh!sV$rN{+bwuC!Iaak2dAx&d_3TRQp
zHBQi34w&IR?inF$;*1tCGopA5_#rH7Wkal%;**gMMa;~bz~lKDs**5hB2fm28u7sq
zM2-w&>_-NW;Z)!*F9$Q0>&h0Tmh8ILOe<_6l?G!!ZdV-`@hed7J53{fxUitA{U`LX
zOatM&^|5^abRSEulZT^g;}c{ppT^DozL(`=IWz2Hxh#D=x%z1?mN7^s5@8ZhBf4{J
zjMa&pf*r>DU#GC>aoopJw8_T3ESIl0r!Zogi)EA)6P4z%F-i>kSBls&`D5`gy>b7_
zx0(BRqJQO3CRe>8mlLq6(hev?6UlqUQgt~pHM#0(?iJKN`@2`pqGFjSQ-`u~dx4uQ
zHYMpt*-SHXH18D${uS@^sDC9BDipd29+oTVk0(=Os*7cm9Fyg0j2grKl@W|j^2zw#
z1pmq;!5Z<eOqqe0;9uF<%fFKIY0SUcn^gS9R?##X^REn9RTI1`m+xDd)+v3LzLmAf
zzLg<A#$GfYHIXKoQbqmB$L={5N5r4xUff}S_ND}*?lWqkHArM>>=yhK8^sw>Bh9f}
zW3WuCaw?E-6qy4Nr154HNvQa?u{&>M^`<v8<&31=HFifR&XMDs;qlQ~!)XR<#{D7p
z2FU6zkCpngBHtJx&)bsJW|PP_%F6#Sc1MwiXBcCNs!|Jh*^*Q$#md-?qK>ID+lj+m
zoa>wF@XWv;$S&_qE*pl+MUugs`wG$CJ26V<Av$9444*WgI$-FfFnmHn&=fJg5kpz!
z8)Iw)AInZ!g{K89-%&({p|>)Qx6J6A`nwS3F**;?5o3LrZs@b9{C#G&FA0LZQ2Z#F
zgrgu7*34nsx>>k?ulAL@sz>G+rZzm9OU<w)c1bN#x}sJqUP3z_e@Bp6oITn;=$pPt
zcs1CRm&>rrm&y-c3SU2b$ubKX_L6x&b7?}&`;}**9X5w!V#Yc)KC3~0D*yIKVeB#z
zp{+xg75z?xJy?7AvM~OCmep4v=s5lIIGH_4{P3R86zngIQ=h}$g@?aw<sx2)FzjNs
zl4dfP*J@&zWY@Evw>);>lS^xi_Pb29`1v&$kwkp!DR}R5F#ctMdGK_%a4rnup(wL4
z4hvV~9On=)z5eJphqo$}HLjc!{vt*Z@;R^pboD$i{hKUi7XZUWEEm+lh5F3_pw<^u
z`6+B9aHzAscx})vuVs3g^Q#8!=I~(t1ZVhNTyBJBe69dMVpiEwBV2Jq_`Hf{-mMte
zpzppL>18N)n_hP7B`=|}=F+=iWM*pjZ-4+By0pG7=>~}K#{Fm(4erXWBg=R*v*U%o
zCz7zqwJ;k~uu$TDkHwm2Q^!0qyP1ZZr{U-<(!Rq2PhrIP_tmxIhigaID}kCgOY8CC
zMkjVHN=u^T8@NgqL;gh9imUH;tFBjZf4+9GTw9-Aze@E)d3~w2R4z5w>Xh!dnlW>D
z#xxA875HH|ACgjLXTkVf2!$F@a8{y;E3HZW&PkC*{iNrT&hBi}tEg(lYtH6pD?2;w
zR*S57%3NikS(#HjJZmn%*&p5(hPUAo5~)yj2lG*c9al=|taMW9^w$WTC3#(NJFV<j
zXUA)^?2@9C>_(;1$j=_&0Mxy42!cwf-Y8WR+g2*2MxC8KodGp8&ccjx81u(1=b`m8
z%?Z*Td%JGT(vp4Li(6jI7G3Ouk*x7CSc^S~-FECfWzyaBX&T>8p*~Ys5LSefxMHk7
zh$N2CS&&5-vOIRI_e+>%)TY=5F<ew_v$zdfQrrC#gm#CAhi7*%zV^O_Wl{?t59`bG
z@mx$=ni+<sR$rEhW=eC*a}t0mFSj4Z=7WdrBL<!tNB)RfR~>i|V-p`daFxZd2~7$e
zl}OF)R!yaf64h#vqENNgI-6S1J8TLwU5i0keC@n&NVrZo!&Zs$DAxkm(dZZj^X{ar
zvy*o0e2rkXh6%d$t%Os92Lxv{S|zv0%iBe~I6`;`&jp~+wxhXtez^|BsFCIQ5a{5U
zVP&P_n~$4*W#u!q)(~3rnR1b@Ig%3P!;B2-5Mek)%qkT0AS$T`;RMmo@);nHH^E-K
zLwFU=66NSM`;5mlLxKf1Z)MAR*!t8f;yOchCj_>~n&w%dS_1S+YG`?y7G0(g?4k_B
zrfh46EKfHK-Lnp9wrs|iDG^$}{*%kYON3Vl4+)P5@BVINBFO}UFP`qCYg%yOXhBM7
zK|oOFvgM?BuOD$zcP>qAq5&~O%7_`~LbQ`g(8fw7aFA{nbSUAn<Y^j_Q|bEiOtRqb
zAeEFuvU2_PihM|s1QyUwd?shx&_mxK>@eyILv)K&+F2F(s^+2!>-4wQ2(GxqxrJ2R
zIEmXdX?OYwg)jCK&Lrr3GA^x>Q8sbG+jc;dG*g!yRdO|KYjw?)R7cj?eH+Cuz;+j&
zqnhFTibi$E;S2z6#W=vm;~5LiAIU{gp@~98SuSb%p;E*fU{pG!Yb9A0sgh_iqb5NY
z1(0n`*JeP-^?LXKG6D<=Sw>FCGEtj3E0}CD`em~DG8l1upYTTEhptpM<D`E=<ja+d
zEwj&;E@J|Kq7v(OXpHNhxB6G2&g@ZY*JzR2Ts+U|j-maW;pp@wSEB7Mf$pi4Zr8&Z
z4X9q%qhW2xb)g3`Is@a-!_Xhx#XRImvz-cKaSaU)*f!g2RdhrP1n#r8&RTjW?HhUW
zoIwxK+x23yhHf?J{FEI$S&b#2L5q>>tm7V$+`yHNxOU{hyUz@WijGkN8qJM4_OTm!
zu^YEgoIcxb^P8tM?83E2u;8nijk=xLoobGw3wG00&=OxNJeZHTCreCDfdrQ%a?W>h
z3Q){C2_L;8efm+sNrI<rjm}ngXH+~wP6>k$hAA<S6rN_b&DSalqmH70HKLhcK$M=s
z!=vLe@H8iiV@e-FoIfO)4kJs5(`(;)obb2E<M8$WU}_;QH^pOaQ*~hE<b#EidTi5(
z3(h(|e&TbU>Fhu{h9m9ReXno5Oi^BD`R{e(FX32magoj4GDjmE!Q@_g-i__oD~|Gd
zJ9gj4?ku6-IDNXrz9o#na)^y#0D^Srmd2m5>D4suEOjZT{>s>UJTPA_%P%*B$G!MV
z=$T{{NCQw*X>kH5;sDST6e)+JF08VV0D>@#drp>(L4K8Vn!6coAaJyq^88B@mOlZW
zA48k-y&2TH^75A}I<koVfiwbAcD@t4&*T^c^!2Y-31JJIC%`;Tq5=7GvCLeH-Dm+Z
z#KdIGVt%7othxa6N8Gr;bPn@->6O8p`H(2fwRIJnXK!ME-`gBb2h-=d6njlvxy)>?
z6NIm@W#cVO-;ktpW?yz)&;9zqLH;V;Gy^jtQLF6gnjIY|k;rfjgId=vRjQTh(lfV&
zVY`LxX4i`%?>gOuVWb@duI0cW$SHfiqiUL?`|FLZ#=vI8@%DnS%yPTk$s>#Q0kNMh
zU<PB5sfYf^L{UC$NttlBwGZflI}5nplX?3o$)_nlJPnD}bAUw94#UTpR$+%QW~zrN
zAYsU#!1H_JEZh1}hR+m0T1>`yl5}a(>|oYnxO?pa@ek$T{E9Z`IMJ3_{z!Roxi)LX
zF?sKH?KOpZZ?I1XQ52Lq&f!z*_JMO7Lv-djPkAOGT)CSkRHf^<+PdFN7gG0=Zf8HL
zzD!ce=2ql5ea|Pm<%1-St=Zc0<^(D}CmWp-f_3_Iqqco|W8>Tbd;Qc)rcrJHFVDMh
zRJdu+Okx=o2bsH8Q|C*G=k4kjDSF!Q4EU3*z=FTI9LRT-J7uuXG&5?(U`VOjeL0Q)
zC#vg?t{>qmZ{J-2_D5V44NVn^XdAZY*`@`js&;)weKp4gJ$Ng^5#cnhyX_Bh{HF=&
z@_cmtbkVI!vy;nW%ge*ErUDjmGXgBARxTmbhN0<*uJwsM8TGxx$lwZoK*n-|>kxlO
z-!#~=;#cp-!6FY$=1uDY7qh%6Z0>T6H0c-zc?JRyNo)$-Q{)n!(%^rCdJW%rtxcRk
zdw4_O>b3+35z*1z;1)e@S6hkxV}Prvo0etJ)zx<tAbVR-qIBi04%JG5);M>rQQ!|k
zItv^+hB<qn5dy8DVJ6LflE(L^2??0DiWXKnZ|%pD>-Dytw5si{U3XrF0;4-3!YtXM
zW&%#enF*{o+W`1pzPc)v0y`*a)OqU)rM{(G2FLBT{b-N<ft2Qkz*#%QCRLncy#hzJ
zi`E*RGNEjqcZmJU5~xo7-wt|#+uK?hXdK@Ymr9jHx1#WatE(F=JunXBn*T6Be(v1z
z@*&B$yigdSF=nw`N8f6J#wSU$ST*hO*}44)d!x>w*>LLi>knlREi;%;>_O8g2X3on
z1p4<*A!X4weF(;xgD96wUUSLljV008Y<G)tGnQgw5DN2@X@jI%F)4>}r4ol_5?ik`
zZQC>~5)E!f#3Hl+-YvfCc)qENUQ{nT<yz#4x;g1yp&x2ME8Jn(H5ed3^VMv1DuUT-
zGLLMppS8o0r@Te($CVop@*!+MWFMY@j$zujPT<l{xEzBTU{L2+dCr7;4^C?VazWk_
zmKkMW4~=pV6%FqfdPypE&!XY;<s6(JOxEIf?ch?iQmHZ%XQxovk#mO+pIco$ht|N5
z&O>kVL8kLq`Aoc{%Qaj+m{vWoQSO)|)d&E9v9CpPS#~0tUSQO+eiV}=vpx#b%4NB@
z`>CDyTb}2-e=*PyuZYT?6SziT0*_;`xEx>C&615*cPv%lXVg;kL(g_)Su&^wwpJLr
zcqOW~uB%QUa$|9z)37(WMz|Sm#nI%3<FT`Ou4;7qa9o-k7eob^&`+7(NeMA6JmeK6
zAS1C$rmp*rTW;Z%85nNhc;!Wgm4tM-kP;pa1M~u1if^l;ECHTjXoCSsVQeS6wik-X
zwX;Z%;kmG6XysB-T0xEyl-Q)(XWb-ib_TlSXXCk-%$1TkG8dOf&8ZD(jGeP*gCN{F
zclz|Xo#V$x+ewFILPu+mTZOFrE-Swo?I)@CL$xnq6_#SCJb7f2oI>qqp<)KW?i3-F
z3vH;zXHELOf!Q$LezQ(^BL+Yj(0}ce9r*j7^NRJ#Y6bp&wA!v#NTu>&P?4Zf;P8P$
z&94V_iQ1)Bd+E7*?kTio3T=57;J`g9x_w5DqzF*~f_(=f)pi9Ss6NL5iaDTj6WjDX
z_ngcjYUdE&cxi2WmhEdWrMHL9mLW0R+yCllPyY~ywS9Bm)BnbBHy;9wL;bu`kl$J0
zT@T04t$k=hQ<`=sS^$F(tO9ZVbxOvc8tL+%pG=(3BAi1Vej$#C_wC0sFUinIc}fR}
zXi$_i1~(&RcR;p3(^*oi0Fz<`EGd?5+4lF5Fs#KM34(yQaV@-%Q}JQUhgD*HE@gdP
z5Zrq14){4I4E5bvhT=VYXWAbIZ9kd(E!&y|@teY7h<|4SAAZUW#(-bHH3fZI0~d<%
zP!!tuN<J<?`!i2E(oN@x>5#7~-snGDZ`aR;S2J(O)xpexnZQCn$vTTDs7spoP4wC7
zy8bi*`ivgT1i{Q((fhI<Vdz0!l7vByxfK{;R)tkzJM~TtA}r7)mF4?OdVD?wuJUP0
zn!(!p15rF8YEUo+HmplpH>{tn-_1bdV1DZY%LDjPk;M$wSs=!`^cX@}s%>)!0|u}6
zbof*uhjT`w&OS6MWI7xt&x065z*g=~qRe|>)CqsW5KSy05|-FLA!Cth`;+6rw6+~t
zU7JFQ^Agsn{>!~6Fvy*OxtQyP?2D7C-yN-qR3;WaEPt2_Ynk;hV+9U)zr|vpX&YAq
zZG5dz#ba1!s8>s(<;>1HmRPD@7_M!b!|<5y&-hWP6v4+3osqXKPUq>|O?nwrogq-h
zIlXp)IRwuSfi#Kf|KTa5@gu`vjmTVoADPQTaE2!|&?Fm&?1-W%b(F(8oHS568k699
zE&A8%AR6`TWLPdSbJ-E$+H{q8nm-|%Vdmj*y>vXjznt#MDI^2f<SFv{<Oo!7^61{O
zJ50;1eQ3#+bWL<JkCXDH80a6z=m?J@bjGLWlj!+6T%Q)jEFg=;W5p3WFVYlq-*2<A
zz_j@Zj4-@3l0GaUP!j-bPg98!zCQuT5A03e0)0_8wD-{t1*T5{WQw{N)<wee<Ph{%
zN7-CyFAr<NyxWX%Nfh%`p<DfROo6yYq8Qi4o*>Nc-gFp6pKPzO$@8_gLL`;I4^?DQ
zBSeykCaLIWRwZ($Hd~TZMRp=pvXocq#}}&yE0u%Q#pAjm%AyEkBVyPZF7+a!rF(Tn
zC2;=}K_cPQvS+D<y_O7sdyPf$0$3e@o1!CCP4)-9jjxrqaOVsd6FzxWHI}5sZV+@A
zB`L06wHF3$zoS!GS8O+t#aApYF3i^EgJ8ZkyRf*Z?}5UXbt`a7vM7hPL)4ttI$J|{
zlW;fsve>#gbnPYx*d||1hpFdIh+KvfL??;Wg-$PFI&&RYAT#vYz7EtO?S2Q^9UzB!
z=uVJb+nlLWh3L^qTvVsf`ivPLsV0)x?uMcmcH5$qRF9+>JF27+%sGd--6-K0Cq~JT
zH6q!%B!0&>WydjX&p!x1zGs_`Bb)!K17xT!h`tDa3soRR2T4IxrS9pLNF+%#HQRvV
zfuJH$#Lr7w$(4v?2GW2QOb#s=!QVV0iT%>PNS|Z_VXk%<-e5DJTmrXu7nVxR#b#;g
zUAbsZL{mux_&uU)$cicj6$!%`&a0bEo_4Ug`O;KOrz2)$67A_OeqE8OJ}BXV%<{EK
z!Pxq`q~Goom(%^DO24Gi!fK}<rs|jIKyI)cP4K{eMt?zv&Omddn08x?he{?D&-+s2
zmK&G;>PywDPaO^%;ubd>TM52YG3QRLeJOT=!>6u3HmFaq*t*bFvI@}Fn3sQ3I3`>t
z+yb(CpYST-HR$VP$<18}6Jl+hWGll_&r{5e1!pu({<)E)H!z<LzR=HJ$gk5EvlluT
zQ_-Qx#-mQB-7pNNeqY0Fv$OY$8k$)RSu#~cE2CaewLbb;RIMSKqN*unesw0j;;&@S
zcI=%N{Qu;mj$t&~9j9?$Jw3ZqWXY6UvLR8bmL<GUx}f)|rYy4}smHI(^oqZdK05-X
zJ`MlC9y_O<#&$}=5r1DeC7c%SWU(o45Z(+t#GWt#)DcnmXZUye!z9Ic9WdPc`DaN@
zx5r3N`hyex3Z43d=>Do7-5z<}+wQpCzCCv55BXOY2%MhXnbDFFxWTC>rbJ|sJ@8C4
zk-+IyMqu^@qI+I^d+e{i`u00+b8e6PL-X$2$BEtGlq?Ss`wje~EHUf7%wK7wSLrkU
z1wqi$*!mUd={v$fpl}yxd{j7zmQ<MJA$8G49D(*|&g0GmFjVZNgWaQ2hcf@=@PEHI
zA0<sXzxxiKg+)jzZt`&PuVj@Ns>DJi{6qizwsS$a7UF*xTzug>|5YI(S=m<QmaIM|
zOD%;`r6tLaD|n-I-{bjD5Qa~Mf<V)CD67JZumJu1GK|&w4DPl|92Q0xH3!2VmdX4C
zX?HTmZuVc8Ra1HHOwfBjp(szFGLjDXd2F_SKmriOQ%RZq-pSJJnjgnUq{@!VvV5Ni
zcH`r$4JeO;4JrB=*e#xM_c@+$b6}@bnNMUb9Z?lRc>3)Tzr%<hB4k-JO|7MwXA241
z?}KBSQS&SdL)pd$kL8C8g-uzqJz1(OSAQYex5Ri226MNULDlVL@N}PJRWLpRQ<1~0
zx{yN=q^=o;s6w_m5F7wu;E%#O4+-mRpVUjO`KX9HD09y314P(vZ@0Hwd%$wx!eftZ
zZx`ZIF<1Qx_oRT$82N10*dNJQ&-SSpxgKWJaWeXX&Qn6p4?;m(gD9moNtbHEp<N8a
z%J4$rV&jVt58D&Y!kzpyn^Qx%lF|(De72z(b=iN^JYMjOb5Gldq>ToX?X+5F+wHSl
z!jPW3#SH-pVz~VnQ1wDEaFn0R#cq2biy4eu271EPK=FIAFAOm(kgX^=LE_m#)OkKE
z%G3@}xXq&kH@13gqm1mlc%PrMV3FeeS3u_{iidycFxyO{H=jniJ(C8!&6jx#T_b#3
zfK}d@aSaAZKj8%uNusPtx7~(&XGr%lt#u!cug<d5y9Y6G+m~^QWYJQ;UR78WHdxO4
z+rd`&e&G?eYtk5Xiph96Z7>)*Ps-bg=6jU0GIjG^+C|2He)R^aK(M5c)7R9Jo~T{R
zGy8svsL%10Zp++@vov%iwfQ9}ivz;3Sh>4!fO;1@y;l-HaTf+m-qjAn?JJ=noDS(2
zl&@QH%@`XAG&9jpc%0$ML8xU1<!F{P6E+#dvJ~|C*_yh+>?Ts=1bL_+JXRA%IX?qN
zaMNM})Jp}-!aVE5@XT$l`ghXA?8MB32Ab^KG12qevGuC=a*^7hyfyK*#?Q6~cZ&1)
zRhD<@fN-1eJ*@wj4ENytIO$AmVClYFYl8-cLX>p-J0mC@VPPKTZPI81nm~h7bDy3&
zKLMA**)NL4CNxHk$IqP`?3q**=GY$YliI+10c@!=pQ7`IF(|o0Mc|Isi3WeluYj>t
z9)%*S|Kk7m$RmoX4#Ti|NiZ~X`D)U=;8>~$85npr9h84OhoC5roI}?0SocH1MIi>7
ztP9t}c<)v={!R0wp}RWGMt}nh+NHVR(`J@Q9)@;Fvp-lkLDQxH{VR+NLEFX&;MLoR
ze?<~W)PnKZ10q!irysl{IEidrVOt7&hw6r6l|Q4-;k|BfJ>HwIOQNOS=2@2a-$hlr
z-c(*MN$DqPgr;^gn*`W#<JPQESI-sioSTv4iml9B-nN(94Tppa(68ZM)##nYY0i<p
zEE<{BP&f+hr@8kFfl9Z1F)5)4*fTrlV((@aD~<8I2!N?UBi!Z{f0y;%ci$D(E9Lc~
zeb}Bb4>bZo%B<Mq4ytq!7Te@zht-EOGpxS*Mk~wqjCmf$sSVJXH-Id$ShN`tBjs>D
z+!4WoPH-Z8Rm51(4NTF`_Ku6XJdy=xnO4P3ywCOuiD|PG_xUa&>ne@ZsN2RJd0y(2
ze9g9e-weyvy?2_9qEW4VP_bZu<k>5q(>&7`=d}6At%jN&TDI#~U0EWpQdX(0Q5h^E
za!kDD=9`~ajKFpRRjGP*WUIfnV^}cMAqQ_2RhcS|-PJ6$92=#|T%{zdPV9J&=3E19
zOOX{(5uG!^z^8y~!&S`I#x_ta#bN3>LFWnE@noKDWC94|ba~WNbVFC>4oV6&ETUQl
zRiuM44BAMd>MH(iE;yChq@nALWVYhYZ?e4>{*G*rSwR<2kKpW9H!T#mT^X)0VX8Y#
z2#+Is`l?@JwUBzLnpUn*>nG#6=r!n1B_%wzwMH^maVXsasu&9V(arhN>~h>hwp-|O
zC6TDB={#2ok1resJL8%HJROSL;G%Zmn=&FuuGnXr<Zh*VqDOH;CjhCbCRIovO0q)V
zbGIA0ZgoKv6={~h{eyE1^lGU#0=HtvOS2({vH7K-68s|YTT+&4W)O$YtccGRx|Cz2
z27fw+j8x!Z4M!9i7irGF47GL#@S5Km!38Bei5Y7>4zNOhlPZcRE>vHuY8PK%Xr<sI
z<r%LvNm;G}jHQb2S4weU)RvWBw^fxPGry_kq7ck32^t@-LK}0_sDayrR|szvo|w?G
zhgQI}o1^22(dX$rK8;2-LpGD%MAN2(E?j7(pyqLVxpqInc^NdhaWyRl?j+DwSBYnZ
zXn=+3x@6V~ePPngCp{|wvk#9Rn<feHuzIUxS(;{9(ne6Fwj|TJ5+OH}Gh?^e!kEwm
zMbIEh-q{%w?>>k(7zlNC%^&HCA{jQi8m;+=M6((cE6L%=-QrmLTCkMv&u1^A0{SuT
zmI|^lLhB|vN;ffqTepM$QIH~TU5xABk?WA50chKl+Li=EKF`t1DHg>ibCRw(Rzy5=
zh`djwsH^g~@f*<R0#Xv#M{0F#)3?Q(BKZ@^?lhL8K(ag_-oSi#CMP4>jp}zU0xb>;
z-w-y1Bf>G^6j%=T73Onsj9A#1HQ8dh`ayI$6xSW$9sy#)Hf&5N5CsjKc87M_j)?x#
zKC?L3wgT`a?sDEyWSmZuZ>2<$7$lbJMoT5Db+9UXdPh>)Qnfi3$mOQ*0o&@jBS-$s
zv6@5;#f)9ijN$<3r%InSNKh|pR@DKuVMt$NE8g{3l;OiKYi{RYqBU1s_k<EuRB!e`
zRZ}aTy6ifkc6eK-UR9<+QZgjhG~~#r$BJ94TFR_!C|QmWnXEz&c%5($+uxpIFqoKb
zPQ#`_*86-8B*7(wbfL+#${IP0L1Q318{I9mw|af_G_~1&o(c4hCrM`$FWb$wN`DH5
zK%lsgUA|`O=sK5(sDVzYbWM>QQ>h~Bnk>m8A);LI4U^K6*D(zd>_|zrm7j*U4ad+u
zVu)%3x-(t;Lsb^VzN|>1q(E0^s0vjHNJy>cR39OvC8K*@2K!UigF1zB%rXVTUIhsR
z1-dAiKxyMEwhoO4%2Nhoj4Io6W<B4{lg(6OR8YcveJogx&90psY$Thc^KB;yPO#-u
zZQa2PdNb8{(Osqwwn@Fcqxe)iC;BYofiIrZsIRm<3#^kiJGORMlOYw)(seLd!q8X1
zO18C!>aygyC{wN{$@Pac8-`Gd|1{Gg20uQh;|HQM@Qs`lPQ!@$G0?uBD6CEE4m9!X
z(0c1p^ah3=?(*3mPz8tMC>cPVPBHnF3uaP}#TsH(gKWJTI=NV>G)l5L$zCTv+hz^C
z%}_@IF;e72Vpm8gP#JAiHrkrzDdd*)f#~fJ#nZGFd;69aYyRYx9X3GT<sVG1tgYek
z5GUsc@qD{T(rvu+K~C1bD(kwewtas2P#K<cZ3f>cKg5gh>r6Y>L$(X4{v2N!$Bx;0
zc<2L77Js`2E$v>`(gyo+j-KO+sge5~R7Q@NsBs!rZ~|=;yv28=W6K6l5S9w#xzx2b
zc6cs-`W0w1nxa!ebX}zy#Tl*@31C-rRWsNfS$&>+g|_(zMlBF@2W@kA&}&2t-GP>B
zTAGP^LK?b(4&N)meZo2BKuwrgo`yASu9D)tRl@HLkY|Xdcn_Vir@kx?Bf0_xc6vi4
zlTk;ECnApX%VUVAw&r(0%dLR5t$@9W``ut(i#4&I^b(rT9_=I>s9LdqZL@s`nFadO
z7(ZLx@|JJycF!F2u4^V$+i~n_azj$FUDvK8->8%ytdwh8?(%DI?QWiV?Xvqy%bjih
zKy%i$@)Lx?F8FzI$DJ<pd}oB02@eWy7TzuVsPGfQQ+dV<-n^rs5U1h=o%-YwUn^vX
zne-&vh$Wf|4tkL>cq_|PfQQcxHr4uUn!g4PX9ss58{EC1$mj7C4!ihFWt$%JQ^H?X
z<;U=i$7J;}o-{|^<=*S8-gbIOH&j*^xSLx}z1{q#JoK^GD+}o!w(~=;rh8kh5H<!5
z*^aA}8Crd2#3wNJ*kXefwxK!SZ*sT18peUA=%B(%!HO?w5@yTw9gu7f#uTf>EGZ&%
zl9KwIqKZ_3nj=YyFoivZ`_HKo+!I+BDCYI+Y@Hrf7U9mWolAq|$zW-AZm!Wz^!<Q(
zA&@2K*DdJcfA12ZlM1kf!c{zq1E2%2E*s8pZgwe1+KssnT{6J`;7D@tMO&7v-?(hh
znNRH^%l+c9eNGF5AdD|+_eaA!+bm4fkng`=Hhyt9uv5Q?zKw6gcx`J-RDK=T#cr^G
zAs*7nY&%|;FD$URw(*G0ta^vA>U+%8>2J-l80gVJ&Y$IL$#vz`uU7PyX5OnP_nO)t
zNNE@+1}treM>tTbytyf>3YhowZ&zh`^>4Wkw}^jz68;6HUqt<uJZN(m>t9PJ76-Um
zV973zL~8DhW+6cH>WLVBfj7!~_rQ!4Xf1@18e<UAAa{D3a5Q-2FTI#TuP|JO>EiR<
z{)P)k(^%!Pjzi_0*CJmu&1%&&ML*Jq%KrBMqB#}Uhab1>4#|Wq%&?U}L*?#GsNJE8
zzHcI}{-jV}dpg02a<DQ<4)zGAEXu{B<AIdJ0`gTJqd0i0C@UBRb5&s`b|Thvuh`1=
zL?Y~rhf?y4P!ox<Jpy2z@7t2Y*quZlmPZwBB$jFf#Ra#o!w5Av1A}RyQzX-yVTflD
zQiH)&D8jicCL;X72@yjWaMi#H93tDFiDFrlWfH<OQN-_BM1)hgA<JS>jux0r!J{SP
zZo<6qa0X!FzIK>g0XN0y_BZ-_3)e>{gD4FkeAPr+|M{Mfp4y|$7HPaRk;Xg>754#3
zSo-WN4}XEO-^-&rF{AWQq~|a>e-9H=L@}nY;PIU-@KlTobgV*a+@2hDigOyB_U7L7
z8;><j&u^ZrwM7x0F{oIk(;TW3m#n`V`?Z|Ojbo3NFy23ou%s7qCk?MKt>e5K8_I3B
zDf+VFo99@CvZ=8pC0`rVqJy&h-&IADzK-<_>wwh>HT8>_bl7weQ^;FPAs4F!%x+MW
z8%*u{KcbnkqLbJ=XZpkS|Bb2r4kGzGn%Oex*Ck0&zXsn==UFI=<(?A`2#aatZkI3E
z_fvfnWl<R+W+zDhVX*TIACEkP7Nq?Q`@s;3dc6=JVr1bI7hl*8;Wr%<VEj|SqMZ*h
zKR8C`orQ71dErsvu{7H_!+O4#>bgABK$4$qq~UjYHiAxb!69h}PSYr|IHGuod*Sgf
zz#D!3Y=(5^BR-AT>lceZfgyne3@TkSFMie3zNvnlM<ld3#!ql_d5&LQvyJ;O;ox4q
z-cNi#VV+rd{{tpo5I^GYUBJ|h7hllKcCAx`zsKO8)Zrg#W@_!WW&-7|88Xmq{#Bh{
zsodUp0>=Mk&$IM2J|e`cvd8mM66FrI)aUB34rSL${6i3&obDQ1WrL$(%-MCb@IAu!
z3a=G@80h|fmJ1=>`Fud#l#n^SI|VZ-$w*1__ZQec-E7xb{wT>xplP_|R<EBkxHyfA
zY>wu8(R?(|vxh26oRS~mWJu}y!`N3Lx#cu6L{D+GfY`u*_i{3|IGF>^lTR>iat0tr
z|1(i>SL8G{j2{hNzQeCVe*e*wtX-_4Qy(F=oL9|Q@+@QJb6CZ5jGf!t+dGd9)=gke
zU0mhX!Wk2`+%+oU3goTc=0P&F&A5n(xWp#q@2Hf`m#EE0<{fvw(e(Z1!l6>L1b@43
zJu=Ox?!M<#T=7gVY*c<>%{G%8Y`gL)d=CF+TyuBbT5Mi;G7hYgD2kCAm0>LN-$4%@
z2AGyX7ETrS9biUAcVk9$q*ZYXcTs_!J$9MqQ<K5kK(I=(RlD)R(Ho8~EF5jhl*$Xo
z4|>kx@oP^U3e3<_By~;IiApTRiXUv$E3=kciMHZ~<vP}G6TbytrMlYKy!Ps?IkCEV
z^)+=<wna<Slywy}>iipey(4nugvpQGuwj?&LJXP9)>wAgN|bJ%rG~+lWEAePMc&O0
z-%*~q8Pi?n$L17Xado8;0v#*ysR|?Z0#N%WQbML5JIVZfvWthEGEfreS+auoI!5+x
z#kSu)coqJhOW%b<F*hTR6?0(}DfsjLe3(XZM$M>;!FFWj;#b2*gGV2I^h1y0IjKC#
z&L4dg_h(Ma&_SR2Ld13q$Jo9slJrJlhefEoRCqaP)$bP`5*|)l_y>hg2tOe_Dg3PP
zi^AuG&kMgSd{KB>_zGzLW|n{^DgMK)b@**Y>rpcNjAh@5x(a;sQ`o1TcQMt@I{Zc$
zPnZ{Sg!GP(<`EJd!4$oP!t>X=N?HUiyqbCr3L^+~osa+;2K)s9|2x1hbv+>D;y;E@
z1doOn|9a@->pHq1^;-75-q6>u$cujkTzCS%F!aG#vI6DmMu<NFD@P~nA|wD!)8{{%
z>1QwCKiOyD$InmrPxk4Dm&xl_2>0jwew*-vjOR}X9}zw-d`kFv;j_ZO68<%C`+qF2
zd-Ky7RXpd(j-cF2f+0#@j;@f=UrpQ7I42qB4oobMRduCIp2pMz41QLE!6Z!A(+eyf
z+1<DTo}U-qBD`IAr|@p!y}~~eK01}hS!ye`zIz&vTPaV6KGXBD&&hne6yEiocp`v5
ze~f>mg6tU_zdCkjgljiUWf`mCiExx-n+0y&P+(Iq%A#<DH?1-KU4!1IDRfv|^%3DX
z*i_dGr-d`Z9bnVqdXv`)Z!B;oYmGmZw#HvY-CH4D|MFU5wgzDIQz^cE3F?)mt#R!0
zvf5<mPj8Oc`kqYxFaFo*dKS6#q4fX1?9m#rC5Ka5BjX-paC4%j^~X8AS34Zny%lJi
z*cOYcHf|izy_w#{zV^e6)1TR2_l8aXa@*sSKD9fU(f=cwZ*U1sceI1OVZ3|2;<h~y
zg()rjQUqX1Q+FrWX$foRqic53*2n%Yq4m+i`yDX+6tMlCU{L+wgvRlD_T)(r`@d!b
z7Og=1!cYGGKQNkHd|?l}VOURL)FG~md*etCf_f^m6R*%l?1D_UQp<pjeqNqMI8EKr
z%Q7PNFdcd!w-i2<{_jq=3;zM^h9|TC>BhrUyW!$j|6yN2W$NoduFZN=OoluzxjGW#
z_Rx6t-_iWhWBH^<uE(-a?en~_L(gFPa}Wl#J2>5$b~pRhH}lB0BNNW{KHQg|P3o($
z4QKsz)`l}nYTR;u|D?X!kLLHVegEmkJXdHwqb7M#2SWRr&tcg6?ngrV8qMkY;<O#C
zun5oYeCA<-aeg$oriE+nf6CSq4wiq7pEId!4OYEia8G)!@wL*2Ui5MKv%<d>{!sY$
z!q<fh!v7`wKf<>_{_^y+2__!P{u$f5!1i@?A9M@Pn5`c*75GY$t{0tp4&v7XL0pIT
zhe}y*GO_J~*bbLIcwb4&=tFr^&p9mc_9emI%U)+P)?-3-0A&QFj9t}GD)fv0d6Go`
z6&KrP_O(<x^Y6jai(j84(*GC?eXmd*gM+N#<IL(TTA*M9jmqtiLZ6`Hu@#K9j$FFK
zf~2$a0Aoq>HQLL<RUlL7JI5$N?!l%!aF#8{MyH`8wmVI>Dw}2EP2d(j#S6UO&%c+Q
zbh8s&%ix;kp|GCFpOoWTN%U;n6HB!?zqGtH!;wBIIR^iDj(_F<<{y8`KS%|St{FIy
z>^UPPWS3H89T=1YADjG37x)MN8^jZ?uzW$YxjiO?EK^=HRgi3kq9G2(y10A<6ZKKJ
z=)fyyadG9jvuu&&xpw=pZTQ*61EDRr&mV^P=v=$SpTJ?Tc7d<oKE^U3XP(32&bC8+
zW3~~<Dmj=<;_*DgKcY(S#8|#(OrM45JcDZGos#+^SxnH4pTfzsPvRtltor6N`0_h(
zZugyX`ucIEeLjhi`JcLRbnjY(5!ay(LMJn**i|uKV_euPi|@3~=YW1@e-Lr@@e)A9
z3BeRZzP1bZ=sC!|w(1aUl<g$O*HUf`<&?w?$N0wUg<bhrv`TVKx;a;-p<CJBUA}Me
zJB!5~PTA(7=Sdo01qYl*@vL2h9u5HSo3i~xe;(iX`NecgwtVeMHRb<s)=Pw#T@@+5
zMTd3{WU#Zmm%pmY@4y<LMqq)0s{Axo_6}KnX=)7J_$iFK`6S`H6#MWdy#5YE^Bws3
zX<1c%Jd*%L{8YMQWCmj`<^QmyhHEL=RZ()=NpTay0sr4t08!x(vdRk(;ymuB_6HF6
z-c11%(|N}#^wZa|c+62~_(Us>Vje-$<JWu_!OIi?j}R{vf!3KZ8F(fa0d-UY-Z)VL
zGTKsr{n|19Bgcn=nDZU$*p`WGTY_}GV7@VRXCD~Y))NI7zBz`e#}@;WOw?O1M?9fj
zN;#gnPN2}{aNW`90{U#VcD90rru(CufIfrm0KpdO7WO^v7n;u%!i@`Ah5+B12KS)7
zjq0*^f-dWfZ~e`d=nmSOw{LS>lNE1BnpJgLa~p?oq)(V3<9$MZ$~MxM(BKfpPhBR6
zd7HZeo!cMT^fuf3^F`OWlUrOC56Wei!9GM^nr=v1+#Ql*H$$S%$R@*Co4ah?zlVOA
zj%}eYrm3zQ>x<*z_LgDhuzgk8p4AwPIn?s@P#Bj5dd{Z_igA*yGun@&tK5e)_k^~`
z!bkSDb<~2X^UX^#bq4(i&Z$r8i?fYMhx_96B^36dc6SMe&gBC*)b1|7ue<Dqcg20(
zWNSxcxr$a;uD5wXXz`xSCF)wb^%INxqa7M0QPS`Adn_zn{(z!<ApLyP!w-9p>iVP4
zr>P41qSzmtUcI`i()Ewa^2gU{+RpR(T9;B^hj#j7buK=9h}G#meCXlH^&VIY@_N<P
zV3n=C%Ij9r(R+*MHA?ykP1w>2+UrCZlNAp`)&G<vojde!1*-FKg8uD>@jg{m-!Dn;
zhYym7;-O&8glg>dkFUeu$1lk8mPmg_)x|9l{&e+csF?1#Jg9$uQ2X9BKRmV8)xB#h
zw(pR|(=DVs6k<juC>|HjCDA+#o^ViggRb^OQ-hAv6nm=Pz4(HDJ~&TS=uM*ZEC#$h
zD~UJJdsNkC10`vw?1Pg_r`@c4Iur>!QrC^=byk}`luLEA>K$ALygicMHP3^+!f499
zF{5$E6CsP50M;x4_;!b?y>S?}pT6<@V>d1Xe7m~e@JsLmA5RQJ7Q*l`eER7;252Ss
zLkb}(rIfL0AQUd|#LT3fWImejLk+w_3|taFc;hk<eYuOFOQc1M)f~e?>JH1PYq0pj
z6}GN&-0Kf@vI-NvNRCAu0?O%%yIk74Nw3pS`fH?z>AOJwl71(X#g8b;4a(JckgvH$
zh7Y{h-0T{go5AL$(cRqC;l${6yN`9d|7({V6vahJy}2zZx2w{kD7M?|#_fvKzFCzX
zXfzt$%vFuXRWlx(`d2lM9&KE8bE7fy3;ga;p_n6l9&7;IHKUi>R6U+&LrwER#Ow~+
z_ApAdf4be~R=1bgiV=@J!$nYibP4p)0|scLn}BwrsBYN`jbl`haZDB4`m3=!Z<@7d
z4j!DbXM^nI<C0WcaP2To>YiD#+(sM+j=NA(*?lL7<A!ThOs}3qferU>9QrmpDUL7Z
znXU68V7ZvWj;psg?7um7=W<~$#1rlnhk~oSGOue64_KSgc<f)Zc-BisCVD#2AlIET
z?L2kX@&MFg7;2-D9k=fe?tTXTFI*j-4X?iVLh*z5Bu8Y0E#Op0Yp`~!%@%LM4ZQ_F
z=i2b|RI8Prx7+9W)j;o6g;<zFT@$8D39!MXOImw;+mUq5F3kpAbfMXK)bg&eybn|Q
zVb8k8v$mnu9p$ZT%kqzj;>Xx(T;HtX&hAyy*DWvL3q+q~gQ?dqE*4`At3rkCbauQ5
z#bAgx3P{q=6I&%Q4?0H808cnn>F(({Se<vU*`9aMN8$KxQGB<Og15_n*)78Np7{sR
zUU^_%JgdmF3xq7p7WZpLzdf!IY=NFShP(TunV(X-e&Cw`Ly0?5Sd1!PQ-l%fm_TwP
zAE)vJtZYH=<|^6cv810KET8*!zLxGEB&mv?shUK;hy$h?j#^(|YL?jYzEbn~+{TjC
z-J$T%H%l74EgA6DgJ@>eaNHWeHxW<x$mixpal7-y#g5`ODkr`xBumk(^d@Qc%5e>A
zrBW^5dt3OUG{zWr5<J4jJgaQ1kYsruJ6W`KU**BV9ZYr%H<Dr8p5$TIXZaM%pYf~T
z<oPv!hUe`(GeG0nbT#8!hN5{-@HD3ePrylw45Aa^J@wr{oO0vx08i<EnecO>>$yLC
zbdBx9h({r(Zl}0SS~9d}+K>bmFVaPOd=O2G7s+5L9})vE&}$f%F0i!4?6AXSQXUh{
z=Le_12eQdzQlg&~@u=eU=OrrD(9cnoJ`dxVDw92t$J4UX-!rkWvqKfWcBBwoNmvt?
zhbzRU0M}?UrF7I_^noiDj|r!Rmq0&uPIw27+p?6UJU)7XC3orn(~uOShgaw4lL7jr
z7n!nWvHaEfaKO6@FE)YUM^DGXl<jFTBXpo|!npIBgxg1DHDA4vo#YH?<~uQ_Es>_5
z2_}a_-%k2j5X5VE0~~6Uf6Q_CW!@-1#y{S}+vdmlM?v1cXXr~WE0(u2^c`uaJRy}U
z%J$F9a6ST7_-Ww|o{M0jT)hbBj|)xX%BV0d8(+9WVhsE>7LISbIlF=N9YDLA(tzFW
z0x1fK#Q$aU*a5a1zyY=;z=31ULPBu3@@Jd)pgHR|kEP>zTt`GOgIpUZenvP8)Mm?o
z7?n`J_Zi(BGI|RR3FZSp((<%2oBWo_{V$ju1McBeE8a_eGppoCP$~u32%;p3puM#m
z({!-EL_1s5)CVPgicNw&ItUG@Q7U1oXo-FIhr>o$c3mK(?R_geym>fe`_uG~^>MqL
zgHEU8pqs{CXfN23q8SoD#YW7ZLE~$jInzKO(yu@0MpDqINUy^t{5q*Lkv1=R(P@+Q
zpx-@BHsiS{nu}j7a^U7ib1~l&IQ1*9K`Sk@wP-BAJ?(F`JKb18iNu|GF^!O#bdcFe
zvrQe6u7sK)WM$!a>wv5p4=NYGx_I4ERi(aXYOl7=o{o23a=rH>mgxq4FOKJ+(%sh8
z%gTG5h7p8|*DpOF6Pe2Ts~fe`twp-ANEBM#M!@Ex94=hndP=ySWzXWtIlAi`Cs;--
z^Z<c5PkAo2b>K(0qhiV=OnC&{!WsUpZqn|o12=G4Tyl85&o&muWPvO_0VXc#ZT8^N
zdW`v&;x9;w5gJA~A1b0k!kbstZuOi)n+Ge3LVlUJ{?&^b6@AOm%|>JyR5NT(r^#~d
zD~c+KVtLUK6$$6MYlrKx66&_->;5~TU(iHSnh!l!H^k;rf5nfI#hPL(jRW%s4#|>C
zOg}hu=zu{KqA64&!OSm+A|d)*Bq>CaXtG$ArTApU)<iIgm<kbu?|hqdfIhEc`8N|>
zm?W->#|e4}K?F|{q!wVS&WeB=YE8u0Wf`MzrEm-{G17F_w-TI}U!ZFu5C?NL93h+>
zSVH^1QD1Rnu)?ps`FN8MQE^p=DuhTbbiuMied>VNYN`Stdln{kF=~OQ8H%o`C076|
zK-9l)hKfe1B*Ji8G3-zjWxeF6CYAqIj;v-|X&srNi>F$|FpP3ZcT|xYj^Z1EFWIUl
zOCZS#RAZN+2qF{LJ{THQmPFGp0j)9VpBtE%eJb&E*GrH#<$^tkGQAF?KaBExweXPe
zgTniSj|xu;|3dgx;kUr*{S)Co3jay?Z^R^JasV^<6}q6Xu$A7xtl5Y=TSy&;pqy<b
z>_TPdon(fs4nx_)OitN(VM1Uu?+UIo=0hB`f6~#;7R3<{PfP8PJ|F(Dm1muVSH*I`
z=BJ&3lf1o|6fY1W<|^Gnc=#D*PUIM!sO^4xaE_IVTQj07s_jlP1Od<I${;B8N2`)7
z<$pJdGCak-IpFvMB7BIFBEtD6%OdWg!j8`t&i_Od8=}Y#7Y3)#m7yGXK@y)WOZcWN
zl8@rXAGntb9OZ|gU0^GNqNOt+5tzSUl%-RH(;tQ}tGI5Rl3yZtl<@j@b&1HLE{XF0
z3ulsqm;Xr=b*N6#Md@z{oJ;UPC-T1oA;0-`hzk1<kAh%}1T&+-@i@CsWIP{qza+{K
z5dUL-_Q!;91n~G<c!8<>;r!z{HWE3{jvT)gkr7kmA4hU>O7i)PnzHl@Bqbmoe;Y3(
zMS|0V87f5ly9^T|{yqT$$c!ML6Y(hF^;=U66!}zs#=e;n@#@0)BT($?Pb2>9gDemU
zsD^D3j(-bBMom%7^7^A~(}vF(OyS9Mz~FCZRRYa|x@im7*W(^HTN`8v3XE=D2rGb(
zs@si*Vo*t@It=p^t3+kPp1FTnR0;e`hu?f4)OF2-K8^yWD%EA#v~@Kg#45Y3d#Yl=
z*Nrf23D*fX;9l*Q1Pg6<7AVW27PBO?ENKm#;TK(Ty}y2`z&-~WkYa8?-K<p=(UKL<
zuAWL-x0DTCz4O{j#O2RH{Zq|0RFSAYrH&|4yZ-Wn^&1*?Y+rX|Ft6qVt_jOo0-a%$
zCxAv5%BqOH-U$)&;;!k6QWahE1Ewoh6-AfotDCEJPb{~VTV?S%Rr#FZIL7A`m2c}5
zHN&tOWxl+uj-`ESnI-8sZ;k9-Ur<rx4}g4?r&Jt#2Ko{CGlLwOEf_gmC;WgQgsYg|
z!XZ%?j4bHJai66Hnnca$gxKRQ)i|42>~-@!IP$5`Sf#j`L+Wd7XYRmk(~hV)9Ki<N
zz<tD|M6;b|99tvQ1UqOu`N-YnPOPVsh&p$+>TDX3sIvax-MXx(V~?PX#T`;tz+S7`
z3qi18S7Cgh1g?8)_*tpCREDqO>+p7{;+l4gC$j@OJ^k4b?z1a+2xSGn#ov|H@=|rM
zf7$`z`-Stu+k|)H90&9fV3+op<^~g~%Y2?&MOSpuC5;5Zzz04E&7AE;mvqrd%_*I9
zH`&T)%(sa12T+5!$#SUyhwhXpBbJ&Ha4Nmn?oHE3hE$iORwHP%Y%97dvTRgAGEgl@
zDH)QfwBa%}ovtD9K%$TAG?wMvU3s~&6M7A!R5BWv6v#~N2pp>|g7n=bJRrPTcwG3H
z@N>ei2){jIE%c*lIcoA~oQ$4LpKmS_H76u=?T%k#5Nm!-i_gIVp74Hy?Eij}rCtAK
zkPaIC*;<H``HV&^x*KTu9NWPWQ(PPYPSww9#wri|1_HM8bp~x72hP6rc>0_uLocX%
zK2HIF@#|T}L3S^N<BK@Ehbai;M=!pxCM!Fp_WU5t?4+|JgC!M{4%@@8W9EwgpSpK}
zlI$uD1mAuCUuM3{m+!sIs;sQ4%&N@puIj4p%F4`YTHUQ~B(<cH#78Ya5?umG&>)1S
z#n%#G0WF4)B;(Ie4EQ5?%||`P#ugac2hFUpk?q;_5#wF6Xs~yVh4&a6ua9RJ9q%qP
zv^L`2_s^GAnbp;8A$7ffz85zlZrq5taU*Dw+Bm(Zz$UzoyOnz@_W<{C?latZ?)TI5
zR#3h3GkKw=^b<nlgH<sn70_j+yS$?Ce|xzpd}^=EOK7z6#yqHF$~XQc=v(1d2YPgp
zzU@w+q_6gVLP%>I!v2dBcAvZ4L|tc@LZ1DXpyeEQCHG414cuAogWS(@PjJ7*{Q<2a
zKtgw_7sZ@oP+6GWPx#58YlUV2Gy%UR`g&@-`lpwNzULyB;(b#XKV`1cCss{#Urq5C
z0djfhZHDw_m8I6X+d<QLANL5&p!z%9e~9J6dI=KKOL+EUDrPH88q~TxBz-LEvtgbk
zI>|<=mxq?8BEBwzo=21J!N>fv-+DsldNp?^==>k%exCauxUX=3v=fc1g)YLx;uIiC
zUuKnQC~G(oUGWhwb>2_2h7-}*zn@@@^zWTCZ;YaF<qP(l$p=za67u|%D6gsde1y`q
zXvyb^7*jkJD^zeIr215>ra{CN+iG1OlS-B#g!B_jo+O?y)E{IpMeO)Q$OSQG&?44Y
zj((e<_Y`-Mdo6bcte1~+pN3xjdn0RHFKHrYD_obG!kJpv<)v?hI}z*AzXm;e1dZz@
zP1>}=b-9Te*San*E$6tKxDD<;?x(q*;eLhtGOh|APvd$?({-4_b$RGJn$~sc=^g3V
zdt=t{C%DgYUj%FE-^VnrmmV=kR=6$?NuSwT>$>E$+*;`h<E57}X#G9z|9ojvn$Cq`
zj$pd1P^KqxfmYRYslq@{=yY)qOGr;>&72^>sMq&`%$)7Z$rwLHbe$)}kOWB=1)djW
z9$ACO$~uCm!)1dIUe|HMo*{xL3mASR$n=C>=J(PRpG9(+_-S$g0J5Wo^e{hcv1t0T
z25YHRK<{7UuH|0Gy~X#veHk^ukOQ%(nD;Nra86{{(GOz0Idh1otEFL~9mY*L=zF{-
z&0Yc)sztA88LBhmVy)zL)mT%FmcjVp=M2fJ7bR_%xj+kzI_Xx`unVqRu>B&d8$?%a
zTcs+4L1Pt`>AD^xOADND<$15KxJP-6FyS$d;iaqq5-~qp5wx4G%r!jm4zt;)YI?OX
zJE5u{zl@UOt(s7o&3CTUMX%AwX<D@baKLu<YnMkm;I`8TleU{TxQGv<8eYO=e#z*A
zk?wFG7Wv>o9h6WT<Y}pqCodPdWAq->2mk1$ts^8^vCmdRhxz>}FSgOKa5;zma}j?@
zCM_&#qJj@wJ~+NiqxojUVYk!o@&oWh^v89))ffjnNIBr&(e*V>k*>-L5-VUT>LSuF
zs#1`dN3Gw9PB1mc!1IawtG!gU%yyS8;9*Z^JTUM9prx)JVj1h#5XI+Xbc>VL4$1YN
zIAz0JYn=$S<V<*9>SVqmNPdqN01^=GxaADbYOILniI7~i7!kvZc6=}nUs6ljaK2tY
z=r{ix?jK*`Uh_+&+Fx=f`<0hOtH1QV`CV7*V|sm@|K86%%KZ}e6wL)Y2LBCo>ootR
z<;K>(2f2|RCsH36Nwv@BrrOR12oNJ<Y4$^TLf<uUa$=ZjZ>IG6j2ZPUHT##K#Mw@@
zzvPl*Ypwor%(RX$w?3X`{}LqgOJQz(1g-uukUOGv*1Y;RU*h_~cxwG6C+YgA8vUgw
z>?kU|5$f|%-sGsK|7I-P(J;OJQjfp=6hrtj1<P`=gv$XtQfp&<h!VdG!|#cR@=sA-
zGGL=W`*dJK{P@zU@#mtlO_&0mR#{%=O1^HDo#P~`z`(MbbAUivBU}lx+#ynf^61@m
z$7y1UP|S3K-5Yb&Dt^jPC*vMoqrah<QP_1-)x=GS+{25KcwUrCG>60wOQm_t{|%e-
z_BzYs+A5XkW(|(#=?-s`rX=y}f^>L}h$5u}OImRY%^zMWJ&V6#zou!B*YM37HhTvk
zqa5O+&Na9LppUF^SHSpn6?ZLn1B_y)xYu#72M)iRdkc3j@cFyo>5!L#0_j10b*wGl
zD-cXv9oA_t7D#{zf8WnI4>9Ba#g8!yF>yqiN(0by9*+38Nt@#18ylq-U0&RJ_%ub>
zJl(F-*0$&tvFKlzj~xKs76d7t<l`-ZNahYp!aD!C>DRJoYQi0VmygBMA@*#BJj7!O
ziNHnq8p5^otH4WGAC2qBSE?pg>L%`hs<%Y)e4WP}EL*MX#TBc~E3U=OT(qWWZ*{Rs
z!@*%c-Kmr5&e0B7eVyrnrMw4N6*Aj@2W;$UJ<gi{4%n2!cfj`0z(U@2R<<n4!+~o@
zR=Jl0ue_1FhkGxshm@PN{MaC)VD<{eFlZ+9dmUT_infB&?yX?S%2u-_gOGE*hTrS3
z2r^4O(r$&d;QFC`%F_C+(tO3}F3?q)km4+EB~3>G;9AQ|2Nx|@HU56@Eqkb3+V{FW
zvZUO)e-F}n&uw(K?=HhK;NK?Oog;>d*^F^>UNue_Ww{k`OiQuh5~}wT)&vi|5O#*z
z5JiG9_(asTJRFKBNyYHsoT}^aZZ+7!XTS{910F&=Vor%EZUv;#d$^C&oD!*Wc<sC|
ziB`dG&%zO|$!YaEfXx+GZnr#tAixS|ttR^c8$(c#MSj3ecyyF>+l(r~po6P>HWJ9W
z-$#t0+DRNPEbNgLNoM$!_uiVsKafY0Lh{I}e(u0NJ?AH(Gxhx&h!O*=C5jpyjx36!
zvxB&_MWX4Fq-#Xn7@))aAid<W@r*?B#uvcQHomCmi4?8}VtV^NwvIQR`NHdTmxBB^
zph5M!z+?Vm4zd_^onOe6NQKOJQ?oO5Zw3xaxuR{ymi^(PWoL43raUu;Nx=$L*Y@a=
zOCk3CD%X$K(u2M)Wz5JU7#$}BqXR%X%_S1+)1(@M4B#mG(A8&Vrx<kEa->l4Y`0p#
zY-JSENr%rBVmQK@c|m5Pn1-Tk30KPkGx&R0J@xIGppZq^`fDsZ`h3CN$Oa(F2{#4b
zKN4m`9P-6<g^BrOe#h=-GU50Fg@D3Bs)q7Bvg-4S@oW}Ihx`YjVd!0;&vJ1%(E_{=
zFa>rV$iU99s+ET^p|jV(r9U#;Hk}n*7Volc$CKkX{VkY{ZZG!K3R_6u?>=G}0uh%j
z*DknB^>M8dbUl&3O_7W#L(0>wQqZM>q}S=Tuo4}|wz6K;{Ktc>R@KQ=p&%OKUe{W4
z3+veG^@0n?*ee=ul635gx@7CJtmEIUl4KaspHfu>EjrZ%rOI*fJbQE8%V5;Jhx;(#
zO_7n5vD{OBianNl3N}YcJ5-#vz@Nj^Ym{V4HYyQu&TMx8p__)tBPvUl%bd<WDv9tK
zLp`e+!x6f<W+=cGayDMS3uwIXqXF!a#bJmQ-v}(GDg)@PgM<SySx{1FE(2zp%{Z=c
zjDGP&ierP;B3c68{9MSLAl-op9a-rv(VEei8~`mrpa6(FryX9VQ&1gRPZ)=aW3>O{
z@X?{`LXY6$cc2w676tUSX_C1f{AL;*(knf*diuS<H;wD9smB@K*YCoZ)MkEZOk9f#
ztI;D3qgcCVB2Vj@NWOs^!7xN}g-Aro>Y#u5haFoWQ@l_T_$eaT0x!eELfI@7OlRRe
z3l1KX1yR#wUO28+49O4`ebOY7DG_s0S46l{QB5%?86My|FY!Pj9`=gr8B$L08UJ>|
zzfLp?uj9$>a7Hf$`!|v|z(4=&O{@GNULZu^j~rq9L;NZ(59SFGTau#Z&gFDPHVoN6
zlv*OeyTZ)0E=mF~$~v#&P^a>`Eb@XRYSTqY5F|lE)q*GrY$RC|@EWdT^yzyQ_crd6
z-0uWE2uU$Ta~dE|_pt|I3W#n<U^=q<$Bkxm48+DB)^S{~3nYXMG6UQIT6HkBM@VEK
zvF8jWAAE~_8`vbq7sD9ggb74dodnq&HK4X_OW~2v38Gzsql$p_l0Y>tl}oxNl(2i0
z_Pk>cJ^1J0RLvPB_)5tLpB}~;taq;P@*w48ekEXmWr5!p9Piy59PQ(UW!T+X;z?B`
zO)^j5Uy~QAgfB@lC?>Lq{S*`wdA>Z9#wA-3O;cQ46GR!sfGi4!hHy$W=ZJN}XTYY5
zypcc0{c6HH<!9|nO$Eo@3v{YE<#E_<YFSB7Rj2N!r?MR{<+5Nk*j?PTe78}|cN8=D
ziSHa3@J__s5PH^>vL5*+SZQ}Qn(OoU9By6_IwoS%mB&LT(tEPzjAKupiToPNl86b-
za1;886{<_c>ux;+{q_m&xBW`$kx>m6VamTZtR9!|Kicm6BI|nrx1=3XRQ;jF!!bvW
zPq|F8Wgo`ePFb5nSwEFXTuHMd6>>QsAagO&$LB+*QFL@}#Jl#IPdnHo^>xgVxr)81
z73wLoL7Gl_#p}-cjNVqF6m8VuiZSS*S)lHVYezPpzwj4SNq)m29v#`TBDe<yRbrV%
zb+=%HMyJ!`A3pgcwDjzn0FPl#ql1E^atYnj;tta49V|thKcNd>rFr~<nYmPOfL}vB
z=$3!R^Ar$Fh=hQA6f3T&vK@330D6HU{tJQ%Px4cOA&BmZa1H>}eUP8U4)rYx_WIY6
zPG1jeSR?KlG_U!MTjDPWI*uU{_^nf?F%k#!L9ubCETc0G#;jgHjo3G7IkS{AKjP!}
z1NkD!5nVGt`0F{loS!dWn=^7|E(6oQVLGPi8rM*Sw=5VXTw75~b$g{c_2#=@D{DDb
ziR-T_$lAT2!JfkGyG>B6VBqXCSXXJH1T<F;(oAga(4jL1Q64Y>PNPYR`BHg4U$&tE
zFoJ11*_SJs@bBSaM0(ZTikeg9*Hm<?HnCz)7gfb}HF;4t4c9au-I`{Re+T&;YbG7y
zPHsuHpaHXr+hgRTmC60;N)EOPOLR44rP*x`bOdIAE#4;4#Bh2p_o03JmJY9*C0SH8
zo|su(0--+NC7LS1aDC%~*)Z-tU=vdowa7n(wbeXY`ROX|CZtd{Jbm;i0n}LhZVcMI
zE(zo+OKQjGa-SGRlv4}ltjWNX^b&MpO}gvGN4BO#QucI4lszfPhl!pOp%Z?n)cJ)-
zD97YY0W^%44c3hu1AEU#?|pO_zMcvt@?%>giHmaTpiRlf(@Z#KyR%&%mJ`X(VzprW
zG+9i4>%5PX6fF*pNQ*@N_+gYt=8YdpjSnU=)<^JQ#+iN+p18UdK&2p5EV)(|RKCxK
z0=7nEI@X@c1`H8nJsSe|btJ@xwbE3n>^NoErEs-8D&N*gu&`|yroO(8OUc%OHHKp8
zcA6TO#o|RgYtq_^Tq3R57z}$x7K1O(4`W!Iu2g0DYuj+E62r|DP_6@G_ba%!Z-t|2
z(qz$DY<*5QhO=hB<2BoKe(9j^7XwqBPW^hUn$W?7y9^Vc<51L2W0)`03;)irb-k>2
zePsXlTr)S9*XJL~35I4CawSclNAIj)D*0kDuYm1l+BJ)0km8~J`xlIS&Xml2-n@#^
zW%=&A>&rKSA(P9k9<X)oz|CM5Yo;uhii%|`l5s=6V8;C7i?kB!G<TF+=kDd+!*Rz*
z$LqA4O5X!);Os)PL1!7*Fza+YXjPN7xrtuEm=pIPY7Fu8bYqbIfIhG1=W@+`fS`MG
zx>m{+OwAB-`xG5C3#(?EBtnRxX$D|W|MV~>d0oAJ_uZ!!7u993V3#|&yaIy({N=3t
zx-KbpQ7$4bH2s#mDI)U3T<+(#m4C_pc5KA{=J*{hV`2EP{`c4v_5#cg%T`B8Td1t>
zt&!MsGET82`(%wff|^C&r$HPPIRIr0LT!pt8oE~wBg6R!CUFW&e8CU4(PjA<j<+3&
zL(KZHF~v0ls~Z?&yFqb#mQGv`V)N|$Y3z0L2>)rrLVGf*52A+J|EeEvqWGxnkB+(X
zhI;z6YHY3}Fzd@hk%j?vb)#TByB$Ny34ZKwFXwK?+@w3vUXrHhYA<x}hC$28GJ3-}
z_rcI;zarFo{vgZHA4M6;40)J}*|}U?4`p`$mE0S#&c$}svD4Q6K>fX)sadi3myMXE
zO(L(x()Nm&onb=9HcyQyr;d!s5ni7LHm4(&j*?-t{&mN}Dh95LQ9O==5k0Oe3dT^<
zegJ*|mapSta2xzUQU%u$bs;IQCb=uPYiLa%G_SKjS{;Kp?-UTWK{$n>g!qCWFgRTY
zL*ZN(gWw#OS3kZT;-mUaGdSltTtgm!^29J;1~ui>M}^oo5725t+kMqbsjdoJ93QTV
z?`Ht>AN~wIsedNPau>02&_y3f4KoQ3fiLEJx(}&+5EDehFDST?TrF}dbOm0_s}eYK
zwx@C0JDTd!fwLv>`eZm;D!!k~P@eNE%)#atcr4Twx`8&c8#r&MG}8fWT4CShl70(Z
zm+~s^HXM6>kIS}=8X!)Vmjl$Vw(kh({1$V>ylE?%y*lOC$dTe6>h#Fn%X~3^uq_dP
zZ>qXt*GuT(&}GAVGkQLh*Cym|;HSBbyJvSjHQUg62mYH(x*xrpHL7Y@@y0GNch2ME
zu|W(kGqkD#%Cu8E>764ud$#Pb%R@ar+jrgDvwc62?GX8XFxGwx?@yhK?)}+@-sAX$
zG6{V=-WppJv5|M(_$%WPI4O6p+zDk<?c9=zdqZt#1hTEi{7I3lE%-B5JMbuNLy@ih
zOqn!zJZ#UH9V@aArqdpeOm9rKmoDywnMC|bi0L7om-!ev6aJv(v~HE{4@zRo+s-RA
z&nS)%ogKw7bUf6QkEUuZiV0YcrDp{)q0I_<j+1^{pB4?$pl+>spVpGNF-kk;eR3P>
zHzR%bRJ=*aK6k}V`dk#^w{?H}SsFr*cJ2uM?Oe<k7WFri##G(t_qV|5cF~<}Yadx=
zXFKB;vigr>j$x6U7kue)E%$ovL1>Ye^puUS*7SWRQ<j5i3-3~+bz|f}`<<-__~*Vq
zX*1?asth`kEd8@_AKD*oOY1MAXX(ZRnr<Ih&E;fCmvys{zrm0c+fs@pSvEBt@up__
z*Kn=jW%jUb!%n;9cgtkFs}WhZ?BB1(@P%%;0gT;}Hz@ukGxxcR8HaBe%jybJrZCaR
zM1#XpGx^EdGh{?163-9B9dyhoHcVPXtF+&UH<LcQq}*Wo#OYC5vJGu%1+-z_$E>Dh
z3y%SR->^nz(r7K++8T}5NVa!vXO=5VliyXAz#hVKt6Pfns}Z!*PZC{SUss13)^Rn;
zu#DEas*{!xx9b>vuwK|MP$+UIGBS-yl?M~P#PJA%{>3Tubq?AoK}6HVYqRO)bje<q
zC;j-spY2BDJ43fvw7%`Bgm=()n2o0#XC_9Flm0~|6F(o?E!uDAqKvy>TZ!{br%|@9
zJ&u2JELK|1h%9Pl2PJU>vU+_dTt*A7D!4ucV`pg%RzJDpmJIa43<n?{kH9Z)9CTUI
z<Ae-DpXCjp8(pN=avP2r$CsJy+6r8o>Gu5MScC5Pw(oW=8<myE;?5YH3RmK4EGx_`
zuB;E|=)3)(kgU^UF769I@>fng&(`DMndM&i(X;e(pN6j#a8*KJ2eMeuy>Q&zrj4N!
zkSNcGHq#FybLm;SLdS@&+qf1((!Zf-n)0vls|6#zW<<p`mqmXO>TL9B`b*zM&tfo3
z%+QMYr?HxOhz$v_5mcNB=+<%3M2ew=PMe*jpxuvw^9(JU8!dq995&|$LMP3{1YY(4
ze~f?`mnvIMzte4QfglFL=2_flW9cS@VSa6%Vk$niG5XJg6}+|$7bsz2;jqG|Qf8%v
z<slq$j?rerXN^z!E2}d3W{nZK&VJGY<pNJ*b9n(jt#F6AtH$9f#q|#cuWZe5uVK5T
z#MzcyxJFpR4<%0ZFfV?x4Oq?c4(=`7{b)Ik-k#3DLD-&7*wldoW%$pBVBrjF@RiXH
z8uhCR-Ca5Iy)E$-{}&+3M<Ge=*Vr!pFR3f<qnS-cl(;z8*YtuJz5xv*Tk2v^Qt85v
zsPK&`OIwIu+g6Q3G)cq*p&Gz(Nu3$tK8p3vb`CjB^T1ZVc0pa<9{RJAcm~Xu7Tb;5
zfEMU5Fn+Y?GRB1dx6jWi^Cs@yyF`HjoiDB9HB(C9SR{*Sd-E|BZ*9j!-A>C(>3I8S
z9QRJ}w0$#2f;^_9VZG-$Zi&Wlgi}v}EMg0M0V*uk+QhnhO(hiniR{hK)LJ$8_jo8t
z91A+LwFrNPWs0mC_j$i6GHf0zPfoULwd1aJmIm?PUvSyVWEiKI({L%u)8XsL{+c6P
zue>h?ttST%VT4(~M=`k^OElNHe|C8m{;gGJX5hfn@(zDkD;BlGypw+vvG@YJ^9n*A
zoU!v0qM<*k8{$OXb_@4gF6H;c_m`m8o@DjFeK^7q(i;Yc2fehNPNNt|=r(Iaqvb=p
z;ZD2oZ*vgZA0B_kP#;A)!UoG{FVD>6+0%YQJPS|UlY<TU+@n*kej`&DU{m)SyjGyk
zQ<h+S!M(tF9Ot{}(~ar6)}tw?fYXv6!^Qdntq;6Nrnv%FK|QnI>(k|YnB)SN@`PC~
zJfUwttCH}IcV4NguJyLw(}kz6(#+U<6{)BJ$G}gG3;$o-mp={g?%@_uuS$Q#W4%jh
z`&{k$0f~L7-R&#sFXwJi4dIKbq1=&so8@W>(T*Q~^#B|;AW)J%A?tufXzW?tl74yW
z)l=UJ;Syqa#H>9-aoGp1Xr~7MLHs^<{P<Ho#X*}NKz3p3(WwTbDQ;q8i~?WpgDpa9
zXU``Bn~o9OP7{}E%3F1MFEdL-JK+3aX^i4lEK5=sRmrj{u44Q!QyScEqok<WQI+%?
z{F>|tJt)z|f-Dz`hBBWa9L}NCXiwTv=A1Ju?lsN}DAV?E2cd^@eXP*l1$d+El5(Tn
z3~=CE37wuB=6UeK_CZ@WDox92lt13el}fo*?W)=hc%bMih|*l`s?W<*R6Rej(7_sp
zorQ_b!bHI?H?OyI@6Tb{4&2e41!RfAc{IwM;oBXvly}=$3vz{~Ok9Y}4Xl0LPdh|D
zCR_4*C8DccLj~o!3(B(ea(YNNq$0}?Nd<#_*C<H$(1V2BXWyKo*;j^tx6Nh=aggc^
z=%Y=ilLg3sfA4#{s<C0H4@VH)zp*g@l=;v@Y(&}!p|kCfZx_iP&fshqD_hW~Ip39m
z^y#;Q3{r_o7PKgQhZc4JmNmOnyvLeZE*5&mv}^QU6Xhc@oQVmigquGpF3no^;O4zO
z!<{yIT9j_2EKKj47U;&O(LT}y8CJzP3QUnlm!2;1y-xoq?BJ~Nk$k&dGmJd!TT{CJ
z`+Z$i+^H#}_mPj}b-mVZ!yUtbeb1Pha#YpmCHiQWjwUyQ2F1ZE5z$!M8*8;vj7Mor
zfO&DMqZuSNSw(=Dz)RTxZL0q>d$ldQfEy4#D?RAc3s^;5_VPcK_v8XEDH<;mOp?(O
z<i(?lG{KFg6Pec(iLbD<7~f>t{QKxiaWr#3!pm}Qt+AGqWxgcHpOA$g<SZ-Klz7op
zrzKG}$_ky%tnz%7Dnei`8mcHwtER|HuSshAtw7W7N*XRp)}%R%)|Es2upA}(xx2V?
zIO66B8~KBSL|YYTET$syo7&z9K99l3jCq(~L#LQnvJ)Ox6Vr<J>xdM~c-qfU5~Ae|
zCBRF2t<A7<w>&DEU#8}Tf@CN}DHz9Jb)`{&BSXrIdG(xc3akD;G>Wd7lQcm<Oljj_
ze!s3X4OviTB-JPi0<jELno$JVXexT#H=qE&&j7}caW`-qFmfanRJKYq8ZDF{D;Peq
zMA`#UYvFALgPaC#6AK}J0XL>)nJ>`I8Cg7yIyG!+H115$G02X01!a2ptrukRNxTIc
z8`HcLiAA@^sr)5US-|ovypCaPf-7uL-4sMi@^Y+iGCW|eh_SHHXgTru?NqcwH?zgH
z2zFUK8*YMY!<zO08(JFvboQND^ME+pTBF*LUphEL(FgbuLb|k`ZXA^S_p7-=PAz)H
zTOvRK?vuRCU$Y{}!a+%VM$IWoPK8^&I9w%we~u^oHM6i;)X(F@fjWR!)S_?Y?i?4p
z-8jP~Pz`#?5-oo01PY21<TT2}`U!vi?K!+Wcx1EsP4Vq8LY!rWPRD1p+d_h{eJ6Q4
zgU3#e+tZw(47WXl4YJyHeP@9+V8L%sWG{%^_qaUp7Dm4uMtcyYXX9hHp>pt5Nf(KD
zn^d~}j9k!VP+8B&@tEKOS_Z|z_!^A4#az)!Gs={+E=%INpbG1vByYwR(tp|%Pl@o)
zB+2;{gX!M=R?h<+j|rV^vh`erul7Il$?P0GUxM!t`o%A2Cg$NoobWJias7_c_GnvZ
z`hq-hulVY1Zvliz5q_RM1K5#$1ci9zz6EbVykeTNBdB>JUdz`;h)kh4iPy;tymo1V
zK@4c_MU8vLkWLB0DanYTw6z)Gn&V=AeOylfI$3IAL}xG}idkUvTSN)aqma-jI4S#|
z9kR6k2Z9{IfS>0>obc%5?{^ii-J&Bl^#p-3@bsD65RG6O$$*~_&43(TqDb=b`VT%{
z6`2nDG=;fa{y#1Pub7_(XWd$|6XEqt7G7g4yd%8Q%Lp#uHWRO(*%@B{f#MbUwd*N;
z+7@b_*GcdGH{TX<=OFXO<-l`3UTFr2qnP%+m6ij4K1>c|;k85cI8^@Km>7uhW(>85
z4Dl90xJ5K}gjag#e=8HO-;CpJ2yXwQ<iil+mEy#g18t=r+FCFM_Xbh9k(L9@UYj5j
z&<3;;Zld!v2=_FFo*Atyupl^E!SqNV)!LY?HBWpuS?JwU;#nHpGR<w`s5M!zncJvv
z&T^a?Xpb#Oq()@%y9$wnb(R^=ILCC}={12SY%oM;g@>`B3Ijy_Q=-WHQ0$*5Zi-4>
z5P!%f2o$#a7%n0ZbwP9v3bGRU!?BG8nhW$gy7D1denATffZaD%tJ@tk(NZn{Hm2BJ
zp%cY5fd1c%*6{t+|GE0UWaEDawZwyT#u(JkU)rMSUq5$lEz$ZcnqGhLG!3e90#ogb
zo(~2&W5_tPe7_t7ct$idXjK2zH0uFt6>Y&T(C<MSd>Tg2?uc~f8N_GDrCHQI%q6lw
zbFK!`Y8w6bg}|Y=jKO4H(5|q7%8JVx)M0Mk)t)3y0kFz<L7S=pu;&H2raGdu!CLe#
z)}UWdSxU-B{P{G{>O`Tg0I2Zar>3QE#9Ls;XVeDy?6!;Nvw>>POQh#7+T9u7t+U*>
zbPX(~#l}duF&OaQvR@__`9`#wq*;Y;K?}AYMtHLc{W^)l8Fzs<&!^!KYftQ$NuL?S
z$+!%grv0rKPy1oH+mDi+k^UZsE|+uY5;#A42xaOR<o{Ii#{ylENLNlqJTL)P)jD?y
zBS-tg!=!eB1wKIsH6Msr%Qkn+2Y(nYw-gfZy>~ojkYloIifhqmkK<cOFSyiKb}qqQ
zR{c!%i(^T#g>&aNhYK<W|0LMSl`vZ6gg>K#KD`+HY4De@P89>U+YcKOUK(hCMCPCY
zhrQ2MzThVYUSbfPXOQp5*339Rh93xGU6IZTq9}Y)S~z`rlL1>|Q)vY|c^abuW`SR#
zb28VZX@EgBURYo|pv5sVCM|49_-*-Dk?TT=SifHQ!blX^5F`yH4<pMdZG8*6<Pvn5
z;pY4qxK+wu*ylF3rc$@#gnxad>2%uRpVx6Nih|mNJrDm+XnDt|&(E*HKSwjiqUpT<
z-<rQav^9m}Ikhpj@*(aRcMW$p=H=}tldH$(&6PUwp#3;;L33AjFmHC#$rDUt>a^N@
z^mBpvkGajezPqm9>GhlV+)A(8!KB)*hfxAbe~Hf%*Xup&G|J`1UYyK$M>Uw40@0E)
z6*F(>lFplXT`_XDWb!#(mQ+)b|3@@sZs3JQw@`4Ob_<4zHH3&Y>A_Le_FuQRQC^?$
zXSya97BqvXDltns&$~p^3{4}ZR**=A*Q$a7=xp+;Bops1Xu3Xl0xUOt{|VjvfNI=9
z@?|+!nNTZ{PK>@V#m^!ctjBZ0*rhhG`z$l#Fs(5d-I#yZbvo2d*6P|cdI_WMW*p~V
zvoyLaFY%h+tb+RjO&-YTf0iW@)OB^U0FYS}JT5+WtI|rh!8+wS*#d$-LV&plXIwJu
z<M%U8I~yi2+_W70u0z%xD?T6O(KxXl=D7;9U-x_c4iywZ7Hoe`wi6t*TQOgt*KR5`
z^!bTR9?Vw0Nxgv2FpmEjw)?4AzWI;VK2akIb^2X{r8uspp1+NFf45nH*Uzl2af5R+
zOTg>b$5wR5gGu5xK+>0)m{n}E>1JBA#%uQ18IZr7PXGQ`<j&N52WM`Aws9gSN`Gqq
zmhESZx7iE)aIy8cL$l_4IXl?Y4ro}d78m&<&{w`py%+$hOML?ITEeUeia&{rjztFl
zCr^r!{4=s7o)pxpUKV;^UuL&Mh}$1HDH+6+PV%!c2eW0K1DHq=c2?xS406O@o5~We
zY?&pVWm#HnMR$*o>>TocqMO7a72B;=UAqE@rf%eN_iJ#qTJow@uT+I=nwiVR^2);n
zzF3~D<SUb9i`R^zl`oXCu0?%1$~Aw>R@vsa&g$NY-=!<%{kx#i56wmYC(s^app~zO
z7MZD5X6L6Tr9$2+8X9l;tt;}HnRPAYZ`w~|_{Yjxzjgbfoc6yua+Bhbm-mg{kZ64#
z`pu5`m8L$!{VvC)vh{Z7v)9D#sD=GY`0lu??!xyYFXEd<#^u!)`+~@ys6HRMD?c+T
zRj#|3AIJLP1m^-xF*1fqlxCwXE0~V2kJEvy6An~636r9t=-BJJ^#g)POrgZ;xIF92
zRzFCW30&+<QbCpYTCKRhLVMO6yQ3x~av19|LC25@u0uzVFs?5=2m_rD_WEvA*A9z`
z;CzVN>94lKCSb#0C{$!6C?JxA?zi?-T{r0Cb_p~TA__IRU^T9|{)$H9iutk)24Y>_
zOn^Me-tmxXN`aiH>@Rwb$xBBxxzH-tSEr{}uUM@UP$G53_Wj}5HYcwCQJ86jLf_qt
zpb$&|;y~TCV=u4Ocu6h9Ylh&vn#10f%&M62Za1;mJmX8}vvMdR&(QV!LvTEtCJA`f
z1`(XgBE*9UAdhCDww*zPug5`;t+gm|lVFwXPtPl0#`tc3IIsI%{41)|6U|I6VzUmP
zvRrsVR6fr%BbDt!|C<z(rI>%Xhiii3P;{et2o{Xz4;A6ObwA^X$&#;H#yp*zFvXsv
zeifm4G6AT+L*a+4-1;t^r}!sDgy&srlO=pZph;>U&u3Z+$FVqkt@u}QoQb_Pn)hJ8
zpUHefGF?LAeW~0I$+xd(<RA_3dtfx&CcZormnkTQB5{2%4+G=7Megcsu3IAQgDpSn
zPhZA~GFmF3-NK$+W_1(VOMLWDf*C)`%aWqxZ9A{2B9<ml_>w3n{MDktOR`XeV<Lr*
zB)RlyUg3q(L`gJtL6(GKxmXbQ#ExZln@i+IX#m$=KN@D22(@o}{iq@RjB7)2{$Z(H
z2K%w7u(sGdIX-Y@I3;e?6uD>@R3e%NAW5(*c46>RLN?SvyY6LEDQ2`NLyi-4Igt@n
z@uVN2B#TKp{O@cEVi`~Z|CU)uNi@e0;C-1^bsGuu13@663n_6n6!Xt+0XuAlBORL!
zjoBw)OJrdwipAv<Z5gE%L!le~`Iz2~<J1B4s14I>#_o5S3eV@q>VFxUP)?9}(Vi$t
zz><o{7aiIOqhp!f=0rbso{hgR=U$7s;2aom`#mo`-a=CWk^RK*3>XMH-%3V@j9*)k
zdAVMe6}vo-<1-A>7TgrDt{h(q>h%F8s+|!!=#8>w+lnp_8OLlGxa;NC>v$sZrso7W
zfU#RLe-%2X1)bAJMA<9n2d;2&S%fPU(RZD)Lokx1+s+s#!=UxR5-NO^cGXOsH8q~6
zhQv}ZqDS$`i80-dLDQw4IX}j~6|Mc)a!jX=jjvGFFEGyk3Yu<fyEUS7p*~8-bTSAZ
zNql&4P*08NSu{wFtVyJ&a*QXi{Xb=oq5haY3T~6yL^dvWG@{3Kj1uUPC$n4bdM>Rt
zw1iGN*)J2}9fZqX{H#v==dg-V3PGRec|{OQ!1zQkL{&rip(vunUl$xpA}<kmrU(Nz
zpJLdjMX>;5xBz`nH$@o41zrSc>>tR{&Di)Cj_sphc*L=N2<|s7$H<<P75><sMXhjx
z1)X=<$0_3cV5`wl9%*z>$_;;P9|iLxj_pG*U)t@Folmr5lokwuY>QDn;?W@1Vo*nG
z_@5ZTj9b#BIk_ayN&1rIZf(t}%ZhS9ajo@CgD%p~D%=XqT=~klW`j}FOVMh-ew^)A
z#<lO3HA|YClPqm13_@ED+p)Zzxu+@z4k)nOq_~YWY(%@Snxt$J{8k1q*kO{NNsJj)
zy1r>RLel2o!21WS!sOR7?681NSMH2P8Fu3KG|3!fwj#z5`w?@z->@au@6?P;bcP*T
zlL7p9j%ZMd33^<EaY#F%$XNE}_Ou$q+}w(SW80loZk{_Biyq6O?&#)cf}SH<p#k*g
zh8UtrIfwZ%cSy0Z?9%NiHkz5b6|qLKSFz1IY3-=37y-<b&%VJe;Y{!l%hsP@JxbEd
z63gkCY0Ex`vy%^JY_PzMm|?q?=^_wST4w8YmxA=3Qo>ff0<7@YjBl;BM_bl1vau>}
z(YAF_8re?${o!k0_(Z$MZt=)X85!1)kMrEOSv<WTvKK^r3;r?`;OQqtQw2nr;z32S
zq&)|UaJVjuMd`TqEm1Ya(~`I%6-9YZt|&^UWlL0U*12T;@_M>{c@VH&_WQ<ae7^a5
z$rO*b@vAHNDKR#yY>Cp%dqhw~;Ffe+OwOm`+%c{J4nG5*OsqriHykDL)m9^WKKG3z
z{(a4eO&i-0oZlh|SVFx>;r^DhC`K`hS+sodpG451#D4|vybAGl=zH*H@th=Hjh}iM
z$0c><nT(m?bmhcox_3^bBwCS0OKu+*6O+0CTSTOr2iNE<kFrK5!5t)X*~6M>XfY^;
zEPObf;F)0k(%*9bE5MS#8Gh$kin8dPNrnsKZ~lR<4VxQW3(#rzy^yop9#9`B@prfa
z^!<ln>=sT4D&H;U^bcU<<T*j@7js}WtKY@C6(2Z^L6s*20q^5a&IzxE+dRlIMyDnz
zfyC{w+yN`N@~Q;`K%TD)M|qkK+F?8&0u|JUSe`tvRuL4D7b@LO%a$e=yWPF0j@6de
zmTUhPE<X4Gu9XmQ>BMI3z+@h5ewEKjcB|7pP}lR#gOfDycez$uekX$deyp~MMHjdb
zHj7mO?MLNl*eDgFYtIi*YNsJwGm1rHlL~h~h#r6|8m~Q<0IgOuo;HebCrDCFH%9TM
zb8(O&pOOM}DuN^!T+}NHh<xEV{^Ica1@-R)#I|q(KXmv5yiaG0i7Y~^Fybo~m&+FX
z_rnsV!OIg$#Z~aetFA(_(p5P;pE_UmV{l;O@ch+RBfZxJ_@};@JpE}V-2*&NaNh%%
z>S5l(QNJJi-hUDBPWXY3G0h{R%>!Q;#KKP7e4ij(eKlr8gs0%<&B@b+M4P$qQJCs}
z%@IGy8za1XEA1eoHA;<vhEW~r?}kG$@P6b*KM0$(u^&&}eA5o!|02kmBFMcAE@}sg
z5cU)&jX*~f|40x;@ef5Y_3ih;RR#V+UN3TArSgmkYDzN92QbAe;=T7unsmP?-Y*Eh
zAqaf0C&-4d#`9~u6807DtMCT+z0LAqQUv4j#q<wIA^F1a_kZmF9;=zt<*Y3X!0$Z-
zh8zA;Fl3?E;|2J=AmC5o4fozFDlz!~7^8Okbp4==G5o-8KOZ~@n{S41L1?&WZJ<K=
zb)+7;9fR&kvf+PDo`&P4#a-~fIoLx6u1pJxaL+v$9Df$RaaIsg=U*0tx4sn#3%B6P
zjmgtkJ1zE2d!P?^16O{T=kK|P4_|Uq@|5JZ0V!-}_)WStkT%KBjM^W#a+MIgU&625
zm^_7V`Y)H+w}jMr9Jf>#@xQ><b1V^TR&<Vj>Q6>L$K?%)x5>hf*tY?hIH=BtXNcN>
z=Pd3yy83ZjntfZqQy7YX<L~FyLEaldQf|k4A4-b-{mIj?pFKxulEDQj7zc=)`uza?
zUMmDYmGO521dHL!N>L|84gBV}qc;Iaq5lqbbFLeYw2ZXdnARQIy!$zYD~EAK&0<{B
zW}0+NiDpXkh3`kNOxOhbFycS>F=|PP)OM|8`ZKq_dStauH~)8?u2&ExU9-&d7%STl
zp04{h>#GOpJxQz+p@BEy2`#2qqm8hIg^+CyWUK#Nw03Gg)uRt3J@rg;cA{3byGKb!
z8K@i*q)_$Jwb&m-_}6G?HfUmNSXy2ZmocSZ;c491ljXJY<cCs~*%vL-#^1okj>>>&
zuJuh+z+q$CwVM6jfjaF`TP#0IV@9R+LEr}x682LK?xqluF5&*uu?ErXPETW;y?rLu
z<`565s_tiEjWSeBJ%pQD)M`7zMYygepw%_ptGPQaie7>Kj4h|@OgtygGO)&!l+lQI
zKU>XpHppJK9wbE_iI`_t`Yf!_xz3VgVNQF@l?(eriV<tO!BxOcx|Vw#_ZGTt0#}tV
zpa)vRPP)56^Yx%zemni?>a{UQkNL`Umi}ua+R!N@oSRXf8HX2y6fa;^pF~vgK$_7`
zD2`H%e;Prh@X8xLsIX}<vWD$rIaU}bJ`k{>#IqUTg=Z{xK%ShuDE>@LOpL~d>#5n3
zk=XCFR-7t2w(Y<!EfZf(t30G*wA0w`ID_?+Qg}|bCWh*bhyV}Ul=GivIpTt*y*=3`
zE5d(7lNigy=y3)cR)+tuVQJc2eNnWDO(Bm)nO0O$d^D;&=zHlSxXo1kG9S`i=-;uD
zO0_Ziv|ow*v*UE#0Xp1rQFKTzQx6L|uehVp<y1zc>Cp(ZF;LlAPL9JhzgosNm8W-s
zei<dHPmV#ytSZ%Gjm>G9@wSm9^7b-gDVUWh1l5Vq48Y1z-M&W?&rnl;m<-R7CO?n!
zoTOahO`(~i*_<U|(j}52IYlA~6noAPai1w<t`!7Pc*9|tmsWAUK-{w~h>~!}VL@Q|
zGSd8h^F{IduoA`Ih~q<c!D4YlmVYR_e6g71`$_H?_X^a_wyXuR3v55M9LW4wI_Z#V
z$8R4?ti^N^R*9UmcSG?Sqanl^3t$GXLThCAsiwkuEYK+d13<?3k@dHB7jq-To8hI>
z4AI^wp$}B_b1vRzgzGU$(KL9_22JZj2`hq?o>XN?)Ua(Dyg<|~^LYdpHo%Hzv1n@2
z`(x&VOzoba9gCbt>%U{Z^|G5pG>C~Hv28DqOY!Eg$<$s<!%59UJSECf*dPG%1O*X*
zjW>*4@n@_54J#9ky~8gPooJjYEz?&<O~koJS$rPf@c}gTSFmFWQ#{(fQi4`j!KiZj
zBIuQ}6e1HMZug8x#bJieXw<wnvIN#6`(WKS3D#oytpobLl6yV()?K3UP&~w);Nw&N
z;bFGm7##4}_K4wQc<F9|kxraD+EfC$p}<MEZp|kvZ<rRrb<`Fi)*$3|NI@|Ip9^yL
zYrv-!u1BkM4CB$kAuhq9y*lsP3jDvPjEH)sc$zZeztZv3PvTcizG$2_t?PXYp`Opt
z>Z&y8BL=XX!FqS;q*yDVaZsuiNhn7c>{nAcG8FbS=&Yn*TDCKNZ_B1U5Qet+JY`Xq
z3K;6<ZQ51wx#jsmMmm*lUP|7-{37H>%=Q^kO2mwx(FDUo(OQ|Le1F9*_5E1*%=kV0
z5DhKyYvYdIsUHj*m88X1ytW-J2GVpz_Rom4$ufXOBhp<_2CSI<lcd)jyCtiWIlbZ~
zO2?<)nar}KqZ4v7nf@z_rg!X|rqe;iQcRzt<k?ORq)Z=Y1pAodeMI`6oe&U8U64r8
zA)+D?r^L&&aEw?Lay(o_i)b{G-}qXzaRlM;9fBkZC!6@2jzr|cZvZ1GP5Za)675H{
z0>|frbAc_G<0nLlB$+Qcp)E*pG+r0~l5Y$WsY8RunkN&+V3J2(brJo3s2w;WR}3`-
zN8^KsGb|?G5KQvG#xC(ddssp@Wqh)4WSNX`JQk(jooO@5La3MR=N7qZ25kMfvJk0Z
zfwIsa$_^(6G=)$-^Becz0O0{$L-m8H0Wx!3GUl(Aj`{P;or66@v;D>+{;*V)bb>}i
z9f{35F5t`0NWwhND+=G_IOE0t{^F16`$bOYiohXtZjM{v4uZCL1GQ-y&2GnQwfi9C
zaO)`^+xaJ}uyd4N*OQgD((7Xe0@y0;21aecQJyRbNBNF|=mpV`Ct#Q&!#yEM#+;^!
zhHi_ZrmMz;q~rl6o-ay5QRZ#lAvO{0f+QA2xgiJz^`5Ejd_kY>ysYQsDo0PetYwxK
z4mSW*M+9C}gcFiXs&-A}OT0KO@I_fEOe(6WYIBmPYKGj>;cG@+l6b?AnyRMiT22&9
z^&czy*A5++l5BXZD>Zt@k9TToviQa(qKKatuvUE{zORP0HTx;#J45q~#YquS;!DvC
z=ns-a`FM<q(c~!?&yiHcJR)`;$08+96KWpD3Fq!=tG&c4@o7)L$QgA5Y}$>yQQ#}n
z_N@KIVy!ss@{<eI#d$#$=ILxFup}lqgsze^HP4jitM;^Q)r#J%SZRCYgDswKQAg8B
z<l&Q!KhGK926JZ}z?@0_AQluC98)yWI5@9;6f}f7vH`~NE%>z%`m~136o~~*FTi!o
zvh>L`Xo8n-*wuwe-kpX9d=VNlUEvF!ZmQ*py8FdawOZ2LIcNF}gOCbm%$&Q&6KB0*
z&4PAS=VjBAw6dlVeyUxsHmX{=>2TxVnaO%z(ep)qZ^ave=R`XY>BI2+hBV|Y>T<}y
za}=Wx2cm!Z@cd^Pcs{ukJntKkpSNP<HCjgm<^IYRTzos7r^hiO>91O(u`c^CyJdeM
zo^ouA{-Gcwz`1uceEz;bV@?D34vvIMp4#|}w7%gg9pB=349gq__!MHjv+1y&8OP`~
zzyq%cusTBll2v|hX)g|@WHD#zo+-5|_6)86C7!Wrme&vfwHLla8!ZWYjvn2^!jNLH
zU4iirb{dbZNabLNQ(_49mF@u7_7Jgha~!uTAWVf$h|r2*P!!{`6LGJP_mg3xpsB1`
zwwd$V6`|olYd~IC0JToDT-F>-1zhi$Lfx@6V^>;|>0S6y(9X{z0zMzKReJHo7cY<{
zQll|3Ep7$<Eq&AqVGtx3Z0N;Ll#DwxB(cB5%SsEzPd|_UO`!5>Ff_oHDDM(Q9(IaI
zbfO%EJFpAx;A4iu!?Q(s|B;?qnsxZ%wEdJjBh=P;1%11)of1S6KdBSk3G|Z4<B7zh
zH75Q_eV;sC;|tSsGj@Hk$;lb*+t`&BwlU=t)h}s5s3mfg_gp0Tt~C7}Qwsu+7+v;H
z<vmj@PUJ+HsO^hK$rWwP5{+pm^VLZQB)TYgwQ#*Z=4Qos)ojUebCyK-$BYi|*K%AU
zI>q}!YPLDCUMG#%wX9`Ze>8xhWfqRyV1d$K^BY;8heqyi`1vrR?_WI*1OaKoB>4ep
zM+9vc@wNo{iq@1Mxzlb$l_?|%YX|oN@Gi~(Q+0H~mp-kw@4RUB{R3dxqvY|%s_fQ;
z8J9X1zNtxHLP&p`=O4xMk81OdvHZfqtk77T1~^m$WQV4qKh_Z@ro*viiTh_7aejp6
z<p=53>SN&)AAq<J1j~wM-5s(7>+wokC5FoD-760;xc&j*_y<G}0X!`{9ba&D9$`H%
zO3&9k8Tfa)DeRl^AqC_}!c^;&m?&Sxaf(mt7N>G$Zi-gSKANbt+K=^PZ{&+<wzviq
zbNm-}T%w#D>C)r?hva4Y%#}nDYm%TrHx*8fbm_w>K3BuG7wO7(%o2_H>+gZqkIL1;
z#i8lHjm-bYcZ$I84DTwMNW02~3p>Rq7s`rde~eg5$%+JPd&2|=npEo%|E~EFsIUM<
zK)Sz~-%3`TV!~iHAsYU2dap7)1?`=iEs<#$#{4ytaTs5{V<?9*!VbV6M(UB8b+`o*
z3g}p<038J)h%gXE@`vi!^SHX!j)@%<dAPn7U_KgGVw0<fLlO8?bzXQgEo=B@-kB*H
z%cuNSN#b)+YeDbc2l8-#eWn6%7AN!k#Pp`WT=hh-?8T2wSR$G)7q1^nGO~N#P?8Ob
zNm>x%iMW{Dpe@;;wb%plw4!FbFy-NxU!N2AO=D{SdS7PV5+jE!pA4IXYf?eiMZR)r
z=4uL1AxOTCT2K=gjifl}VL>iQGA|WmNu1{uNg%QX=bsp0k6Yn81w&dA2rr8hs`MLa
z#+JkHvzXL_U?biZ>SwKC>e=9p_Gpl=P!)_xm9NDWwU5WtvEPX+Z66-Bt*5C_p*oj9
z@K_K9s-I28q)l)`7U9I(4m)&g3-RLt-z{^;x!bvSWMZf_1VQw;J*p5;G7;GyL>xOF
zz#Fdv4->^0SyTww2p&MEe>{Lq|M*w77cPI0!Z$~2j{Eq<@$*D-)Z7W4Mjs7_wEM4j
z)Q-4cVt%+^qCjHPuGub$`Dm7Ph&SR4ThAZ!K~z8<Ua^JUFwEl+&p(i9TMT*|<Z#wB
z%kCI(G~D9D%H-UBGQnANtu9hIPl9QmX1fRQzy;6sTD`tp=RLWNiG=&~y1c}HN7#3G
zMf2+XtosDUu!$&=>kU!YMYABOl}6bH+3U<1yeJ9Io(ZxswNII;@v}?QlkM7X@Up}c
zy*o_=d)~C$(<PE4n&uSgbvf5q;6h&9PixRD(0W)ab>1nvxN?y39#$t`p$Hup{&Tr%
zNmrztTQr`~i@H(L1sF=^?isgPo4Q@e1N#COTY9Nn(nP_jt&QK-IKOtc@}q4rHJ#1B
zTE<ND5JtW}@f7|p#?v)`>&EP;+YpAaU2GX4w#P=}`)5*Zg4gUB(P<g^c%m_6==d@0
zINQO)#(*o}*v7G;#4CqmW55-HNNjw@CW@Nv-9a1A_hcMG0e{H%YGhphaLTnq+x22h
zC+<)yf2e0?_(@vkW7}#}F@WtA0`72PkDiEp4K)MYVzZ~HJExY_J{-W-(IEp{bjXZy
z@}VsK=t6A#KLBIY>&K#Ab`ysVYpm@+v#{yGF|-+uh3y+YY`~)kk$6oCT0QJ|7&eC3
z3uF8EvQ93-$H&+oPXhiAbjPhbz{oznL)5KzDCO|mqHkpT_yXCM=XBsD%=RLO61U&(
z^#e&JEA77bGM-Su`q2|#nV4qssWA0??)g8HWF|)SuM~+##g8?)05`bU`)zIs?Y7wa
z+f-;C6Ox~yVxGyyh8O>6>D_L9qO6jcT=?-^Ue8fkxcH$s7<Roq#gp8=d<2b}m&#_5
zO22O-U`l8^x4Bgs`T29)FLPh^Wjn?|$SepVHYMf9m-pywWRjCE`H$!_{lPA~p9ngE
z4#I1iAo02&U=e%4KbaGG)W8)Od5OBXn@<`th;p?$=zkS>T_V6)M3#um6G`U<dWi?c
zLJqFL5H8vRFVO9!{fVSW@PiKz+E8jpf7J0`!`;a}#(jqSCikDl(e~p6WFY||1(!(g
z7zP$$?0SlZ-evvS=(fjg5#3<}vI;4myb!1U;w6221dHbfaW+7^FXRJ)39zsd8%cA*
z@@4uBjz5&Y7Px)fE!+pWU+2EfeLp2#kciE8K*UChh-1*ZGy0?L(E%8EYxF1+P|-_M
zL*zkT(D;b7yar=9;D;wf*9n0Go$vAKD`NULZ^gI79)Y*uBZPqE02_Q{CF?^8rsCl(
zgM3HjS#Ub&e&u<?CoUzpZ1C`(AxBq;WMD+ih<q{XbMvEpcj!|RAA9O!Y0WsA6It+O
z-iNp^ao-{Q7_$Nc67MIIebbMQE+b#_Qs?aCg})k90mbBW;x~Rmc|Gw|=o4L}zMLB9
zbg1v&h!u+mygGioo4eBMdNhq8P5K%nI6F$(9*uUB6&DULKv-GvT#u%Qk^5M{(SK3e
z`(v{dsMlcdW)34MK7#Q_Inm(7vS_;{O;98;r{<=bQ?kBMG(;gMnN8}vr>p1^&Y`Em
zRiY&fe$C;lCNQumhp%7J4YTa3s%AE3ZKrsXoQH8UFG|OvwGC>B5A+-L!9u)|yMucR
z_pY%#NV|or5j{;8i^A<4Q5TKZC|}HCR*X^@JQv2Z#p0E^9V&nlF-m)bWPU7;CyZBW
zl<0EtSdh|Pd;COxEM(`dC|v2kp1}F2IBXXmqvQ<<$-CS!N(<Q`N?}`>pLu*Q^N611
zk^IU2oEgBgOf)|yR@9R)sjjz#b1e#;5yTNGAv-1~TZ)@g=2j+*y-Q8GIH?xS)j|8M
z@s0g6WU@V(H!WeJWl4@B*F936tuwzc^_6O1voEolHMkTEdm(6NUHp8*|DM}M%usiw
zg8mAM7C-_5*lf`_UpnjqfbdJQSTH5UFyyi!s=PBZW0)p|t2}kynXm8!JL(heEMNDu
zh10VK_kzJC=p_TX^%H6ybazXUl*e<PtSW-zngTD0#j38EJcGVAX*V5c6PelH;H%T8
z0Hd`ENOQedE*V<E%1JygOHqu=|3mYBrSy}*@%qVFFu?R;BcR6;_m@=H8C}B}T@`qA
z+xUZM-nt2L*x;rxpQ;UV6sy}|!twTChPZ*ehTr5xMIq~STNi}Ghp{i6pcyb9NS>0M
zsDQ5V0^L5tt9TQ7&T*PPQ%Ie29G9r$G0h#sm3!M}dmRDd%nYy};rW#nJ``a4lcz%x
z!eXYgm6b?B3aN80%0>4*824wxEUzqADP76ILSLf<YWLxK48X^6T%B?Fk;K`bNBEJF
z=)zOKHaFsH7Bq|{?l5;P*mieOuVSnMH+DAY@3aJW*_|d1ocJ}kQOGJ?Ws5Y#@jLEO
zo+gduJ7<VnGRkhz(lniviBmF)j#W@JgE&P^&FQ+SOPTq}_R>VKYq+URcj{!ibF?!}
z>YeEa^ES!lczenc`8lG=xe`5{v;@9IG-Z!yDjMnYT3#n}4`e1eTlU`z8!db<Y$Kmv
zGgxELO{`Lpoqjx*X5u8M%4}LKNqiRPHJ%lv`6w2Qrk{NnH*1OLg^-t<)F<Bd^V<fo
z6wP+*SV=+S^?d$l^}DMWj>kTHJ`6v5E`sXalC&<0>yl1>z!KlLm}>A`2$vxU%YqJn
zlDr{BdGMRm4?WLy>3qb{_Is_MrrBy+iI)4)T)f?6`RGnIhE^qAM;L!IEEp|HVV=`C
z%I+0pX+xGMv~Tu-hm8$y!PzKyRa`~{c<qmY*A)EQptcsmISjRr#Xt*c4DSbSvngZ?
z&>xS{RlH8~2uaB;FXLJ}<61xC+Wl;`JP0-Q{AoO-n<q1`lIq%;9j&E*K5@2o?i`$*
z!!y{NO`KtK0$!Pbeq&Q_{4}lbkSrGuf<}SAcGKx7oo+K;CI0aTAAIo92l2mnvEa?k
z^J{D84?Vd_=SAo@Fn6#HHc}5Tb}d~kNb4Loe3x^}UkppvMAG%JbX_b<q~rFORRAdM
z_PeY+T2n#wB2QXfW@7`)CyZhZSM)8{+LM#zTCF?@>i7C&?1ZeJE_(1p4WILhXXy#n
zkFUWISz`}fPvSpWC+uyd_4QKtD_pNu!#ed|k;Uo%7{=TETp6R5=gWD1i9ZU%0Odoa
z&bJGs4=p`>^7vxT>oj;nYiR~wU!J_`bocb5b4T{bwf@PMAJnu$K~wjv?dzWI`r|m*
zQ*HX*S&XIty&j}<o2@DIP)a8jODX*V*=%01N2{#N`rnvcsc3ty_~qHP6Kk_{^E2?p
znf}#RF4K=`%U53ApLulFzd5^R;-{G`Ey9?C^^Gy!*OMFVKBfvMRFQC3Jfwi1MHmo&
z46}-OUoTaTR!XH-Ny_t<#per+I|Wf>iC$s9-<UWtVf^{EvV5(;+lMq+)(-JcNQy+}
z3H*ZtbDu2l3-hm%Dy`#sl)!ivixxRd9>%x#_h7et9=mp%XquvE-({8@=Z~2!A_M%a
zlI{WSt=yxrI9w$twbU8B)b2PPrwNSK>~`9%9*9M-E>}F{Qb4f_3bf~f7Ta#MVc>;L
zLqSAfeKa`fALaYFa8LBGxH0~?k12RT*n^F_((f&ajpvx8srj`${Gt>!CMVxx!+)jH
zBoW0qQ6Z(hwj$00?nJ?`O^h-ssD-?!sitA=vkX(!#5`PqCy8krf;3;TO6X{mG)+{r
ze7?L|&gV5n)HDfaQcBi=qhDt+cQZ`TX-qE9Fx0J;@bgDN)zkq1o)>ZQP!$EiSXgx@
zELA0-?-`(RYnq%_bty-Ps#+87>VhGH25s%xzi3KC6II<pDFB*-s1eCr+`q@t{rUg@
zMh*Z!YqTojewtN@5de&+u>O^YtHgn&U8N1kRglkt?|HigAT}FlZ=hn$<{YSEjdkk4
z@Zo!X*D2F_JD}Fc_haYwtXrU$RxC>(7M>Q#{NAP{)*JlHp_A9Fdd9vhD@H}qjrOdn
z3As{Hbjr4nFBTM0b}P|EQF3few)N8E27QZVYWUWQbpp>(96aanf^+QJ6AL+~bJcY(
zo4xSQvT71XES7SDrp~q}57?TnSw&fmt`!TKtl4D)L}P3%70a!4I3rVGS~HHHcbs^-
z4riTKWT6#WXj;n6P&kK`TU@IY*4DwgT(qtk-d;D<PvK#ygu3*n3$C4j?bF;BX~b!A
zWMdK<#^4~W!H_+4IwvmS2lp$zZYPlZ*?X+m=I2mI@4Zx3st5g4YBFe0s%%eA(X?=b
z+esT&9>60de-Ab%&4-Y&O+0D`8QQE^;<hsx4_^G|4@K$VlD5>xxPQw%$^D)`rgnm5
zYpMN;8wN2A*@LAJ#1;+N0~ZEiM?>~79KiRKG^=jI${XU2kiQ*H<xr7(Fv49eWC^l3
z>NiMjEW)it%I%3TrP+yyKf+pX3dq7LW(n^G2$~(}<qoTpbc9d?^}=5~hcB7O&5x<2
z`#h~xxgKRkJ8N4majbZ4NrlRky{tY+k`(RWd-w^ex6b;&C`}~!ERg<L|AVl3GPJMl
z3qLZH?<qzJ#0rs(XF@;cEg*X+1h8#wpwjx=n6wZz&cL2!@GQ{WJ+I(ykF|fetjjDB
z)fwT~uQ>)LKD7t@mPkR3kPz<q^btJQqFEJjL(#lxIgWKMLiHLON9X+@{mdQpjVI_3
zRGYesI}7^aF#m^7k95}=USz62+DWFf<gd5D3vFle5T3+CA2qOf0Y(lwe9&}~A^A@v
zP#Rk<B@WqhT%vE-$-WUGMoGzezi-giykG};rzy?s`3(xrMhk{CdPgzh@)K!NrFjms
zxo{RMgnF~iklddUceQ9gP)g7*(5v)AKhYTg67kM?))UsLDc3?%j)LU-mtXuFavj*c
zv((R@8Sxnp`}A5NB>s&q;G5dBXvlt3lo?6o4q>%(RQXXrb5j<72t3<j&xy~#<#|61
z6fLj=Hg9aaU*KMgSzm27zUFzwAICut_Qz|`dHr#DiEk@#=3j>={Ab};{`d?}&}W;z
zwpS;Q1J!4G4W8zw(fLMiX5hjDd~InGu1+r1c$OX{ec=q?cLr!o6TS?2i+|z4;cp2p
zEIBjqIw!JS+1yK)JIbBpUWqe&ls3>lpFGe$pF+?+pFV&G90%c62W-I(_0aKc&{Gu$
zZed;bCcL1}kg(DN%x{AQi2`a1%Z*ZFS+Eh-Q*eS89|$fiQ!K#W;x<@-3oZNs{4o8F
z;H75~r;Zc&wGVJFa4zOi3D)M|{B~Pmvpir4v5Hf?AijXJq^_s6TtS$y-d?PV)8wBD
z6~)T`S5c8la(l5V8rT&ck>1G{r>e9YvUO!>8#vq)cNKRJ_p|UN%#y<<^p3HxsD7{2
zRvOJd{dTiJQ;2w=^cQ#<;l{6mS#}WTVUF=Q5utPr7KoeiOgDPQJDB~N*drQrnrX3G
ze7iLv2yRQSxHuK834)a`h|ZUZC}2#vh_UI4Lcmx9(@9W+(?eiJk?_6@7!rsepvPR|
zVT|a}iEDZnPx<8Cr@`iX1d(Nk)y1}40#on7>qM_s`b$|6cuf|u*tUWb>nctu@{%YS
zYT`=9GXfd+AwRY#pii5-iF+6K+3hH#v^ze3^j{*h`cG4TRpTw~?RsJQUxaGa4}MTL
z%?p>Ac2tI84yPHgxsP(kLFc*-uDEi6M^w_<?&!@}#X7WLOH;>%tF)SEe!Ex~vX2Gf
zmvLDK7OU;{6}f%jVCSF$wC?nX1lZfB7>ZsZns=h2l9H~N-b}d&*8h^I++Y>!jx0-x
zQ@8S9?#_5>fe^cA6H8U^e;Dh+19UAwQgIG&sC~&$EK4!Iq$#2x@%u#HCc@3UOn^WV
zDGD;bDUe)_2%9`V!3#v?!@0>oMzyw<Qp>~(cy>#9_4iYJL4Uhu@wFk6tB%yvKN#pN
z9M)mFk-G(RKlPoMVICZT_OMD*WclI7zGJ-^9fewNSjUz6-LV{vQ;rO^GXig8%nxh@
zGS&1-g<`!*=tV=|ix}%72t2Otmh*UY<xDQ$tTe2m?M$wDc*SGeTD4r^dBG_<)%a!l
zCQW(^bR53K3s2<C>O^5OAGGuPWCHZ1eKfW@n1|{POhh@!nJCAPw(hR;b5rG+`N^rA
zRTd`sxmtO$F;M};3iI_+VFHX7`_4)oL7AQKCKd4{Z<%f#SXG$y%2Tqi&KGmnDqha!
zk2fUCv~2#QU%%*kpvz&!B^YgiXS=|&t#$_;dEF*X)_Yy7Dy=lp!M9$PItx}ISE|oR
z>o?qRy*yVd<qFo`j_la>`}XV?D#FZE$tz7x2^DdnRr0v7UhER+0*An6c_UUW>6Tp&
zYoxa6SGpbg9fy7g-H7mqaVq^KKF>=DXYF|NcMG#b%N2MH{u3u0RZ*2(QJ9;bpA)RY
z!6~V}u-t}0zqY(~U~=-n+H&pZh+Wi+<w9OnbW$Saf^L=aIYot&f~p#trfbj6t}0GR
zv~#MV7R%60OaqVP>NH=OD@hZ3A7T@E_{Oej5yK!j56$D$t63nX$y;85=Vys3%?XC2
z@&|dv)X0oai|2tBSOc@;BGDa04l)VSqt(WyQF63or|dP?=Y_KUsWXNy9DO+m(#d_c
z?Kbx)GmqqR<r?%&Ua8tfzLZxCz2s<Bl{D&w$=G*LD}kQj0lu-PJ6@-a?$<v37Hrl#
zB~Q5H4#6$0-n{3OHEmnMM3v9&+5BPAvaMp}4R?@Yp=vv3nTkrkFJrDKuHW?;sNYnY
z?WRXb3V)wlT3Tu!n_lHRGsoI++&WmQ*zdgi?swXi(!s!oog*sNtz$LsmU%jR4d@c?
z5lp^)t1SN$N>2HoWck)MZ^G4}e|-z&$O(|rH0Ll#WXz*Pdp?!Oq1T3rW_lH~CQ`k#
zgSEs%mkb~p4n1W<63e!#mK;<A3wcllsXi!5Febu*Q_8xRvqb3_4B~=pNRk0#1dNTZ
z1oiQif}p}#>Y@nap8K2r+&F8uoocy)j_`i6{r~wokxaiXiG_F15b?TaIIil)lP$ss
zW^yI2Li6kG;|_2=u%AzG*K)4_S&RMf4EJ{Ko!kT5L)=HWN9d{w%)RIJJQ%1H55zo-
zQ?A#i+csWd*ZUp3GED}qOZ19VcKWTpwAxi%#gpjorCuMNW5*sIgUS*+j$esiU+J{v
zWfJpXY{HnMX{4=dAfm6=bU{`3s+y`Qk7%l{sCyM9FUx<bD5fmWNs=XrKa`}RD9uu7
zjAar_0)OG}Z9+&DHui<@&?}@s?_c<F>?i+)zT}VT3my9M$LNQu^rI&!0CTzy&>RY9
zNXa6;RG{z7u}{)>P0;sB^o_9>R%0*B(HC0ug&J$5O)t6hb|v<M3HpkOSR<8KBR%?c
z2LO50Upkw}=U7SQbBw!(;<+57V&Sn&jx)g<)AgELTUBFg$NetZrP&(MoOpJXw#^{p
zK~RJC39NaS$e$6LVh^9@FwTTyb>3x8=km}STyBEKgzV>5=`8fi!8Too9&t0+>$h`v
zaCd<XbwBrB?gQMz&?i30{Q~!C?iaboxnIY=p?Dx2Iv~gTgMWufzva;mFwZ)}YFn8i
zoq5niI!6dfatX#MDZU5sC=FaMd;KE>4yTCMi}fDpAMou3;;r=CAj6vQTQBw3juCTN
z5(Qap7K@Eyu~{lL=)nZ&HG<fXR9zHb<W#8u%6g8J><ix}wrxY7xbXd8E8DRX+`?#B
zwM*o}U(sI^_UrG961?&+=(oN^KhY?a*hhal)-4E~APfF05r_WRqi?Fv9~bG1W{R=@
zN+tTL=v<{ws<aL9H@E<N{0jl&$LnxUX4K)1O^!w-xStAB8O4ilKNfI)vgY>U2vJyZ
z9?kS1em!FTA+c34e)jdX4E|q_UK4vh@YPMg^Lw~^fvdSrB8q8?_1SIEJ<CeZ>-Ok$
zEk<X`%&Nh1Q|%@6tb5Q>uu{V_uz~t=bh-kaA7^r@GA3hT?H`otrBb~)T`W#d+Bg+$
zvq}Pzh?4+CP0bg<R<&kso3U)%?=R#vIls4-GgWP7KjC>292ZoxSn^M9d&JDuUJb|o
z&i_KQUfQ5@4Aj}`f9MubuIL((Uzu~%d|q{O=W~uy;1xY5<>aa?7IZBq=Oj4F6Jlxt
zoX8bP%CEsb2meg?Bc_~7;C2c|(|4qCtI*7|ET1FV*q0ii2diREDqyId?&1o;y}ORh
zQ+s7z%44QTV;&RW-f<~#S>av}dPx$O?O12+Ut%;GhmbFESg-Cn0@vBR$Gw*VZ*yQ*
zJwVqCkZ_3i`eE`)#8X%s{!+7Ih1N1Pp{XWDX4ZJHopuM8=O`ZOXYQNA_)>F~t}0kH
zF}!w)|J-h&){pT*+`gPU1^xRz&<eOq!hBvL#df^#Oc-ACH@k$s7b0Ff!Dg037DDs8
zyM(vTL`+bYM3(PDl-B(qBRMf19pY{To$H=)#-s+F#6))*x&YcOrOSM4eaBT=veJ|h
zoq+3B+o=;u2>-0?Q)%k~Xk$NM*QQ7=1CD>$u;%WZvkan6tmF%L@7>bDIm;yQ$bKRy
z^n}r(xYd~RyMWLMhF9F3E$FIcsd~ZGWZNYL#W{j!c|dr%WhsV5QJ;^^qp&e%39PqQ
zV)V~8<IHX%DLtsXXE!jE9!I{h8~CYkd$}W^%ip%e*s-gqi3W<*-9U;zHaxo<xR~R*
z+;w2lym`d&K-f*Z3=03(?HoVb$$KL26kke+FAMOs!X4*s0%>$Nwev0#8U5`A`sU72
z@`+(GoK$y&i<K=!Hu4qC&l2J3<6sr<29{`yYMmS-w2p~PPh`xx0W3*n8a-)Yv@G1l
zR2)5e(!*>Cezifj*Y_AkS9<M>K<W{q>pUTbBuF})(~@~aD{OdP5Ouh05W>?{z&d*d
zy>EgfijpAH6MC87TV(N)JEX<i23|;Z1Xz#8TNX_&9!GA$nA_v70sZ<-S@a9^6Q6`g
zL?ojZ674Y7(I`)mn4TkhEO3nl%0&SnC|%%sD@Bn9Ejy%1E@wDeUMuY1ACN(k@M(D9
zpd{Ib>dd%kFR!b{nrgI+G)6zGIQa;vm`qUB^5psemzusT2x7s1C|^+xT1FGzJ5QGb
zU_Cus)u|kK@yEbU6QJ=K@lj;HmFK-auI*{Vu*ze2`YsG0M9j}t1ns6Pa}7_t!)!LH
znqF<pPN-^i=iK{NO{mP~J6GzW*J$K4t=fRs!gls+JClw>#*(DoF{Bv9u8y(0I+jFo
z^FCVc0EQ9?M-tR-YQmE{97yDhapa!ekdIyx+q4cvMiJfK%0-C1Ya>)krin}IOdMEY
z%Lm6hlw?+f?c3>l_<{Ea{wam7qiF!2U5l$O!8GFO+&V6jz%WiQmHHUG#wOg`o)e%o
zc1ez30&KUh3oRww+W~$|iW`cH_^JO~cy8q5jb6vr<w+ruIBT@M@e8!sa3N<IN|9J$
zteih~q7C^=+!rpD1-V;8DXt8n>pFJJ;QZn)kE^?7r@mpg23jn_1)v#W?du7U0I%ZC
zV2o<tXV0KZ(_SAArWp4h-Yk1QeFM^ZPD|-^+esPH*k-tW0Vyz}set^wrJ|`q?`J23
z1V?O;wSBr;#r&^@w8GsL1LN*D0F`;bIwDZeOj7WV6;*`8Dj++jYsmQL1CqQ_CYt^#
zifhwkDF4>b;TQcZiSd_u|FXv3OVV?O567#!)B}c&&8K$FRh8uj9Br)riS+|+J(gdy
zd#H`!miYlcaH(YYsKfnkiRP!aANzMp+WzGT%77Hp1!h4PI7xE?B~Y~5^Drt#j<5(w
zT}{vcB_&<j8a)Nr%=q1_;;1lx%RgZ9b25fo)C%mxT{Us`qj)>bT&LnUo#G)cwi1{_
zG5M<C35K?UxI$%UfYE{&u6w}xnK{noTHG?X2BXULG<Pgb-;>91iJq&pgN2ywsC*_{
zj#8EpUl{)uEY)PYvfK}Dc{EQ9hG8A00e?;T^JPz(**7D*<#|Ek6@wNr-w0MExR%XU
zVY2O0%<JG^=*dM>=5y6@d8I$A?42sTLvHS?P41nOE4(Dmv-;=ni)J-z{>p{_m$@)<
z>@7d`ul{ecSyXr}*X>T^mJYQrQLGl?1lQMMB;6u+0!G?9X+Hg+mCnG*)bN%UUBR|0
zvDRZo8f6uiKvJ|8Fynr@oOgO^_xTVJuzif-BF`?YvDV&PZj?(R!;9ybdnd}xvOTrX
zR2h1WlJ}&K*UezLA#Q%mF!H~!Y1|x}d;Si)_%=oo8{Py6q&PB{S7zYUnH4AYwJ5Sn
z()9iQ+6uSuy;3x(9OEloi(ljBxh1X-J?)J&V#`T0krHxBa6qw&I!U+ywVhf~!d4PC
zyL2sZ>~FQVarRuqNt+CB=L1%vt@|1~`^5(_0uwjJS<oBucD@9Ewz0XszB$N1!4jCS
z`Adh`;#?wqT$OLZ_rT^nXdU5yxSq-k@lc52;actu+}$vG-b;12^h`eDCuF4rmmxkn
z!NCCEc->egh;XIMN>2f_bo@VzA-OeZwEXU~XBi^SC7A1D3`xHk@yxe;jxkiglWKe{
zznou!zx?y6d;(ttoAtG|Gl7$k?tU$~(CU1|D9=CdhbK@CZQd{fj0N#^|37W-0_8|{
z9fsAb->V<03ZFs&-Dse@(Ez4rdb+0@jYjwUXLe_Q*blio`{V8oX_4ZPT<&s)ACVl!
z<dU<bBuYy~mN=FuJ44Bl4%haPd{%N2nUlnd&E#02R?az5$XE)ap>up>>Nq)$R&vZE
zEpzXC^?3yp&^;@<XSz@*RHN>_@4owa_r81IAe7aWxR>P~gnf`dFx}cgU)W3&Sr{y0
zqjv|C-^Zz;V-N40w5Kn;zv??B)}wcW;dqGwy5abHMZ1if|H^jpDm$|G-{XyZOAZJk
zzJ<y(qm5K$xc+kT@z|03=mPvtf_?t+gJThn)$^FZO5~Mw(hkS49FEPaX;`l(yXGEG
z@oz+%>WHSEPf5{8YpLx+6)W9sc1<KuBfV}3v=T0h`ye-1#g2FGPqQ2j80+HB-qRrE
zhzu<g+Psf#^#|H$mvAXsz3lkN3XCf3l-~wWUr8@dqs17XMCkfsydqFX+<O?~{vqiz
z($7dY@0<oX{F9~0cny?s8clvao&QAs%AsM(W3QxSYb5&(&kJ;IB)6W=yY)b{&TwBs
z_WgO!n!>7ay)mAHg{wtz$taA04nJqqjB5t`XD?(WImMT>Z^athyC{{@3RJu#R{Uk{
zzaRaT8RE;AROOm1UsF`^3*n;=!8HRuiuQWNd12#Fogvxh^s-QXnSDp}Rq`0jFC-t}
z3xc2Kb9$K1Ig!^En|Zt8(o%Q}50`i?2eKBuhr?^U41<`<c@#Ll#=1!H3Np+3a4flP
zF!rcx{7bH(V}A<|$rEqzCw@G-4u^93Y$=;ZSHmpLi5;YIOeZng(!akPJ(fS?jpU)P
z@(aHfp{5FSY5`uG(f-332+;}s{>CJ3uki1`!Cy!{7snBY<yZJi1{TEkaL%|*Ji9kJ
zlG$fpo*dWgbKCJKYZJ6M{#K~bJ1|*BcR-yp5s{=UoSaVm864h^Pr?ZwXLp<{zQ|vE
z4%cgwl#943!&`@9`sq3JyX&n`Fm+E3Ov*c8Hzz|^C#9>NL)ViB;lYHU$<Pn?<H`nq
z>=*dEjj22uf%o5K{wm?vG(MlD`GY?kst?~9`47a_xucLWIn5o53wOlliD;lRghejr
zXGf=TuzAqVjHNI}#*{I7{I4}igNfl9+~IKLT)LAniw5h-3Hg-DaYqWkm)INMPZ94+
z@9|2_fafyX(YLk3G#Zw?Wt;nPbynPPaV6S;eib}<w5{rE^~gI;F9=&PR{17#A3g-%
z0Y1Z9@Hf67$Wp#XV_Di}e%thdma&W9V-N$q2Mh;fI~*IAL_4K$JK9eEMvg}isKshm
z1nX^T3?E*dvcJP;$X8Efus<K4u&a{tgTDfNwuu?7g-OG1uN&x?sL<P{I^6^dH|$r#
zV|}3YX1m*4K12G|t>M;Y7{F(nK{edk+!^+FdXp3D&4opxs>{o&N){IywxyEgRbo)$
z*o0_g<>3{g*#&Bl)n#Jte8u+t(DR>$m#XjoxvLREP4MFkxYs?rcIfDw;}nQSiCQeG
zgleSVICIZE{F9}6Dfu7g^0PV`N73er3q(XMp|NHYZYai`uiXt~8Z4N`Vnr=-RddlC
zzJdo=d(iQ*yt-}ZRJ&bB&h=5*<T*TMnZ3Uvd={OH_9T_oQKt=XP3$!Isv)kNE)Xmw
zMvjn5D5TMWb24c6a7C7fp5nU1r;wd`?PaV3?-y=sR5xsP?u=9sV$3Zv>^VhMP3^n5
z3)|5D*to4I$^C^Z?2Z%xe)T2U)UFtKzjhUSG{yG3^!rkYS*~2hv`BWF$D~_dHf&vO
zsp_BLO2_wJXI}U%ToifTHcsfK?8&w<qIPH8pHA&%d}aX8JeoXJti^?uy(VGNmV$QF
zMf?_X`M)QE@#go5{wK{>#~<0rKvWueDGYNg-c!fAt%R>IL=O@(&O$u_fYtgWes<yY
z?Z6Oh?XhcEq@L6t6%Mxz%Q9}Uaw`+hE%U-M(RXKxrHU3k_^9Q)uP#Sld<&jE2G>u<
zFxiRUhR+S8X12ylk{#R+tC6d4+pyCTr48G-N^RS-ZQNe-247fvgW6vsd?{7HDPVFk
ztsf^oK^e5e*e}{;%WlFW$~SIY!Y6n(-{KQL<z>DOoQ~H~w)^Y|;~BBeIX`%86o-5P
zHBsno;Xy?k{OOk!?S=)k+lbcnqDA@dIlcuXEbGc&y#cIs$>QiGacQg{*pb#)4ff=_
zhaAluY7TdB(=LjipkKThJ!(y{q6H}qkEXn=`c_%{*{fIi<Nk@FTD}vNh5ai<FO%*Y
z-vg-roCy1ueiQx~n3)>qLUILrEww9RnKUOgSbbo|M=>Aoj4e2Gr#eb&MGCUC)(|ET
zTlB`(^SHvPeQ~0`{f9Mm1KEt#x7tAC0M1sX)Ul6iz8;k}q!XY^AH&r!ZnGs72O^G7
zAQfg_my08|GQf*Vg}rW6Z6T@A%@7+>ogs!x2w;HeDzCt%>Z~A|_;!)##3QoO#7(Tp
z3DF;^$#PBBw10vJI3sKMe;>bH<G0JVqWhor(l7Rf?~VKjq(R(^@|{?5rcAkz!J6Xq
z09f_zlxp#e)P))kT|MQ3o%mk3uL^(rw*;Wc48r`xO!A)7u;?5q!aLvu{$b`lk)rSU
z-}WKgFV3caW6Jj@@3a08uTQ-podzoiEp2YVabpgvMQmcWSr5zyteuZZoZa5$4rQn9
z&C`Sqs-WVlAr>&9@E6P79^3T~H=s$?gBcaNM6foGyPj8U&DqVW^K5OcsN2CpFz+3j
zt9DkaCB3s=oZmR4>DuAtqU{%73Ra7T-&!XnvvyQg4XfS&xwIe}yBCV9RYg|RdZA?P
z6+P1|*}WCjS?OA;+}yaVg06SW0&&}=QfcjdZow-q`WstTwNPp;sH&{YuZinc6ewfk
zuK6JiX>ZsY2E&jJ;5CHzH%8+>-#W&B{^hY_8y||!BYA_hUP4@rLL+y`3hf|07@hQh
zMdk4nsdWQOw7W)a&Z(HCpjdZ{&AwjHP1`Ekj@8_5RjP%#h2lc1R1KFJD~;xM^A8HT
zQ!E=nF|G%~;!joZnXqOl4oLJbs4|aYfP=yx9rEM?xX>coQ2||wA2WD<+@K(JOIEdJ
z6r%F(o!VN-uNAPKtml>dpjIlnwoxS&yLbWMqYP5AU{K4fhhA;2P_4AKn*ikMUZ-3M
zT62q`rYfp#C^GPG(W#TF8$Jb~Q-(wa{v)gd@GST_a}MqZ^7`7=TK&#I-aJyTADPR6
zHtoT&0;78htN09$ox&o+tjsX{3mD*0y_;SaKL}aI980O=cWv?-IB4~P(MyM0*eayE
zd`VVy%U|(9G0TT*b22exOaH!Z$p(?<YBkSQ6ji3GTyWd>bX<?Md@yaNbHA$_1a2UP
zX4n2tqWQK@9f$gkr|SiGw|$fv20hx|bqgW(n?iG!fQ1YMBiLX<R%A*}tJQ)m7pf}y
z;c9|oP1<3(Dxfz3?mn;j#cM9pw%$d-pmU~<3i`l>Zu}2!VkF_iw6jIG&<_COv@u~H
z{@++!DHH%HMOR6ouy}T{0M3r7XvY+VOcLNQQFI5$<6Hn5kWPolDz$4)`&7{|2{HaZ
zaGFe`c<P@5UaU*6ihF8+V7Bb;+1?l>^g<yp9mmu)&#)E6GCj>WYny7Lx^(2oQnjI)
zHm?Wxm&Am*Tn0~(Nk`_PV09+Uc3dZI8ZaNHNf;F(ui&&$6A<l)n~jY|bK{U{YDTA|
z*EbhhI>;yNys5i}iQ&`v-aNecVG&EnG+wT)7-4nJv5tj%+s(^;oIQ$4L6m9gDhl+g
z*pCIxdc^__!0MDEoNEAQ4|`STLb{Ev*cMCX)OkZv9_`r|ftz7B43-vjOS8JJ7W8T#
z02q~p&AMLd7@BDw+Gt+3i&Ib{^=dDCxj3i!e%&h^D{UOET|{zh?}f)KXRaC=E48os
z&OSIIPyQ9_wPS;vXt5gh(y$63_m<HL!bT1(_Y^Z(5M5q@8n|z`#gka8ND7>0dKf3E
zykyiC%4M=(g2QgB205l%^QDC@)fK~a;P?DKpMPiJkfEuM&8gs@!!18k28?O(Zc?2u
zsX?Q7WcJ62-#*9eK&gzaw7j$c71i1BCAFDh>R^6tKQ?G?5>S-Heip6H<vN?m^P6Cf
zZqBnAUT>rO@EqwX@Y7=L2JQoip{NgY6ZK*29<1J%^dXQFSdK-tFIyzZ#|Wz|<2MOP
zQr^zC9UDs6B@it98wvpQ96q<51JE*7D|pT(+6%YI+89i{EUySt3vB1>*W()eJic>;
z(Fx7-c3c+pNL+X%CRa)(wDyMm@V}59nLY&;7FbD7{T;-8$0sT<sPPr-qM#Cg2QU7C
z`V%VYX`1UhKmF5=|8=nS&!||HuGLZ%LRD-B)-9J4c#eYqv%_N_t&HCj(<-G=<Oczn
zc4&VPH_B+!{tcZ$W=#bO9H6JD{DIr77|Cz_g!-S8H3aZx9&k4u-_>5fiG+NuEc!hG
zdZNPsIi59w8YD%9+stojFR^9BrJFiD&;eldOPpG)KIxnPhE?8}8wH9}ptRs9f$zNX
zisRo(YwM=Z;vVUs9~E1Rzx)JY4t;>Lk`;_kDze?Y6yeXG0|;M=!T1cT!?FX9`zn7F
zU>F1OzkU_({%SIZ(BATs71TkT5E&54Y=Pza{LXxR`D&!*pK7#Pji**|9T4Ou_5(Hm
z^7{loa)=F^Xcv>XSkoq+Jq+SQud*!#E)KH{!i0@7mRv-k0@4O0%Z+epX?<zw$4TQD
zkbS2~D|U77Tv$51=;)54RhEERt%5>Lus}64Y!)<EEpHz!nV^b&y|xybdU35+y@X$~
zzVyQUv6@!Y>VuQj)|s|hv*o#7#cH%_x^Bb$m5Q!7y4xrdP<dLJHn!@jt(1-(4m1Ma
zTj?a@gEVq6rp#Av<*lY1t=uMwk7$|RsIS#a&84}eW(m&FIOWt^YWB<ZBZE`hjmGw=
z!I63yOw&UVC4s0ZAGAqJDo6`)>^kbqHXN^_{1dvbe_SJ$@qZf@(G1&o10@+82w?UZ
z4c9=l3r13&R~5e7mlm<IS@!$Go=3(W(P;<$>|_4&;~%dIPoi(UAw13b%xCtJ2bM}=
zeRl`w*2l<7c0cfG!h2Hic*77=Z`V;6f}vAoDWa@X;1CzUhrE+T#lr@Bf=9F@V}l8>
z!EDdg%8H_coox5kd$yumirHLgnlgQ#0V6G|9c(kK*{S#QM+%k+G!>oVvWEe8ei^$F
zhhbSWRlivD75Q=B0exk{ZJMo^MlBN?<?H$DpiAFAv+fdWlE#d;(}W|Xvrdu>mk7CG
z4`)k@HH+K<10AMB{>uIWNc;#d8lfu7U*M>D49~~G3{aHhZT>?4T{19#vE<>VZ-%l;
ze<{W>!5NHP*)+rWLN<QM_rVXPvuH78VlO>h{@mss1|S7un^jir(zu7)PO?!n$u2<s
z1J|)XW_MY1QJ>YRflYe#N}wsG=02!`RjLFPLzltsH1xt8U+)|7mmg7OQ(sunO+b8I
z;FD!V*U&;z3`%Uu8li%MrG^S8woDamI}?6-NHujZI*4wnq0qyO$8U}R&KFq^vsp2m
z@reCSC{O`gf^LH42=CAfU>#QA&fX?|F4Cf%&jr;C(jP9kDOE7<URFO*aM2YIuAGK3
zPH7Y7v&Cx}Dzrs1VQxS06I`OUc;!VBio}w#^S;L;#o2k!3M8=oQ6bwCmFCnB`A6A=
zd$a@Kvm{s~3~Niqq;(iU1w0|yRvosx0gmrd@Pp`$mfj?neGH6MQ<T36ep~dn*xoe0
zT|ti<?$^@q!FH#3{<GCq75<OOfdQ|qJ%cA^;5LIK2?ynK!s%&<J~NFpS9xqOp8>u(
z4QXW>nV$EN7}c?Hud%L0)9XK^^|H3Xs*5Z8Z|?|WTjvGd;qW8L**lV|U@6GA#8mi^
z(6b;rK3&j2XZNj-7eg`-LCn<m+P88ZhZGN&($jouUydRzkMO+WsCvNo913{$O+Qz_
z&_rXpY(*qinG;hb5*d9Z(Hb<-B=0j=TxLQJd`ga$gNo{cYs4frFs=uO`_MO64^9Zm
zJ5YN&87X*RMfbv3R*s*^B|uAM4|Mv4v1){urRVc<3^J*4eK;F#?f`P=q|i-V*6H_H
zmt$7gbuh#tuLFup2?47X*dFxb-(Ejxi@%~4e3hfP$Kz*o!SV_G_}0tuNlCVJ?Pu|z
z>6RR_GqYUYD3uze(gG{T*ND#rrZB8Rq2j_%1z#@~EKP$e>=^2X3;%0|?y^^x|HQ9A
zvSKp+*r#k+Is`nh4L;pz>AZB0^nmoR^r-ZN^bVlhM}>VM0<ogn>YHq3rRotNNRzeI
zZU;joA#Q^JmoFS(QOC5rBIfA#gx~CbEs_7OXUVpU6e|>!=;fvs%GMW=Tp^O{GkDf0
zFEBDS7|D|JB_ZP@<<I}xVuV}f)H`JrDh2#?dEM%oj{A*tw>w6;52nK^u<Y1;ngeQ{
z#5|GgtY@=^6(0tpIgI;=;TyWcHd9r3H@9VGePKX>1?4h9go0|jwyJm)?{6e$zpkjq
z?>-LFNY;zpZhN(;%jCV`VNX7M-(lH1EHEM(_oj*46l{aVgiQ;>b@nPa(qLDq1xHqi
zCOZZ7$w~>Wf>K3{necqyC22!C1sw7utH6)TN(uBmu!8u$Fx(}^GI(|dJYy$3ErYzE
zDb~~b!2Vg~%=u%w?3l;z+A{$nZ}09sMPGe%Rj;m`Tq(3y+HXg+6(qp(N<xNF?l3>m
zsJq9Ou{Vki_=0Pq7qEEnqN2c1zuU0Agc!C+lmB_zILEpCTu&b2J<)~85yjSV7%S>x
zPjOrf$EEO{C~283HRAiKFsAR-YQCaY767oy=XFE1dq;b?udiBf&IMLQvkMCqRrW(s
zraWQECBGb6o)cOhgb9Gc5vBkrtPki`=y+CG_Dk}FoL>b?=iF>NCj;`ZmqAAWKUdIS
z9)tcz1<YO+Fv*wpSF(5r80uM^=nvlf!C#VN<QTK)@0IQcEA=_77SY+pQs+$tZ1Wyd
zts(YLO!ph<++s&tP4`$s0j_tM4c}!Z?6AKX?2#e-MdHiiyiIY3FC}ZuoeSp&3-aRb
z-RI}#Esbcj>6#UTM52DdbkKk=m>@$ip-dyP;nr>RfeHp#--@Dv&9A@(wOA>Fhh%Gp
zWn82o)+e4bs1?#1?bC<7;@X)Dr&bH)uvD?Drt5%%tQjj@^}S7I>-*?FTdoF}Y2XCI
z4K{^qvaTrx9NJH5mTFp-samF{Z5vv}E&6`Zt!M_L;}0S_E`Zd(!1~Luu249y<<m-2
zra(R<X$AWmc6$e+-yR7zd&y!_l`XcUHbaJx-wRKiJT^xfn<q~k1u<@@_5R%#_LiEv
z7w(=j9<nSQa447sj}Z~DGQ66t%iuRsz(z*L@;uiLC%`@c3aSR!7aEWh?tQOF&CRk`
z=nYQ}SDfPNZi{Fu=Qb<O;WPJ~J>r&X+Fgor08&1{jVH9dG>yt&RU=^)V?9Gv(e|5{
z)-WnLZu{YS)27a)Aovo|eB&XCs`z1$wE-1PHvt&H_dSZ@tZl7<8E<Nz)67<UT@j2x
zonYTnq64?wt=ej}UazV)7^uhMtBy`4<ilZU7qtH@`7i-5$0tT7^DbXVT+0IgBPm*=
z^fX^WToV!57BF;#^Kttdeg0|occx9*bh~>ubr7yjv9_O#yn%jXVvqYTxc7LLeVyhp
zxnwx8!m_X5vU>n)`f{|T^WO2q_AI?tv9rb$dhMVCfRWCvL`}?cS7N(gv2-Oe`#&^=
zeivj=^reyqOi`&;D~i!deU)s!FyGCPsmGU;F3a;$LjKuycLso>V-i6qYTGdwRRWWL
z3`$LaG4ZD|mO)PhndU-zz;Qo-KSk=fEbz%m3{GUZA<MT%qiExFzRlx^J9P^M_Be=F
z_{nir8ABmcSAT1A<vMO$hfP4Xqb#d9XJI+cUKqy;>>z=E&davBO>MJc+D6~BL0JR^
z!>tlCFi6!k3W2gVjv1vX2ES{%wjslhVY_C{@hh&Xx)t9!l|f4(8Qw1fPu<N{tARNp
z@EE)xx}wNh+zyLJfOwOkooK^kcR9@ZzcD?4HqRw+#w+g&wy|a9;ib#AwPgV2U(L0U
zKNdj`FM!K?#}TY?#CLN|^|5*0lzQW9(~~RDv7M3QZ$~_|%ai<t!P7JB^4VxXa6%WN
z-q^mgbz(Z3;j;_BrCCdP-^jrDkz|=JlfQ;L?vKIzykB~1oKm<q0%s;JXJFR5NHC_M
z!8Bt@J3BmECEQ8kfBha4Mkt?H7*izeHC-uGfCaz^YnBv!gJZJc_%))DdhsFb;-jdt
zWqEa<+MAK<ey-Rw6dkxJa6knKBXLC499tunS}f`(IjS#~=MT&B;ra5ireKnfXBE7n
zdXkITOU0sU5zTh^BB7BPuO7_x6EH(AV)fCPE4eA&IPqEi3b>ZD?2j8Y&;{r+Dt3D;
zDLkc3{wSC2P@sZ=t-tF?ol<7>8@f_zf?WwzsW_^>p{XK~@|ofZHKmDHD050ZnBkN)
zRZnQ73g=^UYnAO=%hjP6-^~aY^rSGUB)STx@^YRo%?aK;#}pLzab#+siJsvm4)al&
z>mn{QJXJo4>wVn1rmD;rRVVe*eQ#Ya?KI`B({l2`9jm7$?rwnF7JD0arm4`C(KM5A
z-%O!GY>om~WBo4di*XmLQ-caBq`hEBhv6{9Ky?Vb*a-kt+RHAVv0Pyc%tpn{Sipi3
zrBe8Ap`v{G#tZNNR2Wj%*FI*K$%@nN7U>b8%oM1cKxHkM44q;G;olaD#lkc9eE$3s
zBiB`(<&x(*fA2ZZH<X`EOA)mcrKY0GzKFKyvLUlfZ!N*6DJ`*{WpB&uPi6(QdX#CY
zYP7lG4CxJmFji&<L5-C#!4e<!56Bk19Cc$YKlRl0>#PRHw`O6-`r=;1q><x~c;|Ej
z%y&}ti#vlk^^r$Zumb<>K1lvh#%#Q%7^^C{b>J}zF_c4D0K!t$Krs$9X734+0CWkF
zU({9ER>S`UW0sHFos8K8c6Tynxkyfq*|S)awG47S1Tojv1}(xcWDW3oG#r3#WI6#A
zK--NIzfdvs2kB0wC<`C!m2C|JJx!zsH=YKSExx0u>%x$J8OHX_|M)`bQ)=1zG0k^r
zc;f|H>@Ayc>R|0eYCE7vO+t#QLF)mj&bx(xxCZq}qrI{~8p{t>scI7n1N}^)_}_f~
zE_-A-u2))iRF^z=mtqvp_*JUwy6aeuM>T6zQ40mRRG+UYHP=>LWvQtyw3ljz>bVrl
zQz!c};<10f^pRvQZQNlQ2mtJyZqD+^DLLj2I#!(n$uK}N^b$Ix8_3%0ajqW*4Ei9h
z2irX1ZW)^h`J5@JC4ZoPoozh%rKec4_v;^X{pAK(SZOziPYsw?1$Vwmf;#AyBBsD1
z;V2fEbb7W*tKep_Zs5QrhJMa_<T$dzMZfy7o#(5_EQb2Xi&bB(EwvZa<`R&_c7Z_i
z^^#mDsLHCPQ4a{{0wHNr^|0+&?wmiJo%5l{Kf`FwXRI@<BN|tYXC}HZ-9@%wv;BpX
zqdTCIsuqUsJ*}YDWjUGb0s}nDCi^4lEOKR8>DVZ-Tus(jRVw9-_ntJm!||St(d==&
zF$1Rq4BCUUk;@ySy*nHuD(=EN;P715)VP_!fd-}+g^VZ1;-0_8SY#kS@GuCl-`M~Z
z)3`Uyq5H0M{s5HEb}otcmkOKNgIPl$=)-w2pJ4hR2&(W>T&iC?sp_n}PqIHN9ghpr
zBX=qd_r+-T&|I8?(*XwOOEZxZ))e?SRk4Q8<OprC@d*dxg#WHpRzDT4zK|iVV~yPU
z`OizD4dF|n)CKAt7r%X-P!ttm1r=)*TQ~!G`c>YWuh$*+0yGiQ5J6%HJwnXYa(UGx
zKS?f*$P)|xGiuqs4X`}pFIab(WOVVoIaiS#x7BhTxiZJVJ0_Q1({#)7Hj(?EU-us<
zS9UFj5XaiBl<C=`zyA6DakI2gssup=UL<iplqYTC+Sf!)9pr*g4}ehBlu|oDDZ1T5
zI|z3t3J5TBhi#qg8um$b_pGLm!0x%NTk=TP&hDxwZF)5W`fXm9E(rWrVK`STYx(&@
zhOv%u>2<?6^!&1At-w(4W&l@NztW>*73hw&_*zl6IBtdj6BGgK;a0B5eB0>^dtWn4
z<`vWabglMj+pIkQJORrjaK7sVZ+;uztf=Zqtz_vBxb6eGRnktKJ{dgr82B$sCr?YP
zA93_N?0J2J(@J<Ei@mi7d#)B%!c8!*@cqs(${gtQHnEp?hE@?okH~zi<wtDu%|96#
zkCA8nmSwg4XO*Dv$6yK_ddwKz$j~eScS9}VkeAdA2K1U&To1rXqqGwA5%~3@w-%Vi
zW-+K#0(imUj`uaKN-I46QSL@)%oyekf@xp4w?ExPI}9`E+%(NAIP-STslu46=U~h=
z?WZ`dj5BQWb6IT4@Mp5A`75Ti{K1|<k5c$KN{!wJmo0O}*G%~#j_)sHF#lHo(?`c}
z(h{6JsU*#VB-;}3N*N*$7-E=%5y#<!VfaB!vz~tX`gM-Sn}7NA(^oSXBru<K-LnYc
zU}q<I_+fy%v9t5JF=#<YNBo295oKvkS^!zG3X-Ci%s09Lf)n3rbiivUguO8Cr9SLk
zJ72V4wu|Q*pDT53veYS>_Vdr*_%i1Z08c=$zlA*i{I$+ij0|1vxbHIWsRB1FSMM<a
zh{oxYr;(4H02!I)!Z8F$A|kOJ3lZ7+eBeQCCG)692gA&u@1AcoIB2DkX?JgTY_oVz
zI=BAY=d+A^v*WrP&=vxxR}oUpeEIY>KHHu<dF@%VxKOOX54=c5G_GqPwHI0cRgBy|
zjO)V5j{7h>hG$5?3j4NX0V8&{A!k4zV}~v)ky%VexMXTtxpZ$B-ph9G|9%PdQC4#O
zbe2H~MadVK;>eqUhD>7Hhh6vZdvU+aGd6OIRuiy%MqnKtryj!@qbqks8kymimc-e>
z{p)Sc$DZ7yF+&CH><j@EBQEL<fw4vrh9IP2psIcsv}~!IAo)6QwfSdHKMhFzF=*-Q
z*Pq6YrU^>^FC$7-ullwrSRMc1&Isr3!+Rgx$+b)Gzf<1<QEn$2OOnkdtn|SF?bP#X
z+N&+<-gY2+wZi{IY|55nyFT&P@grtocVJsK2fsVBY4Z@3(jGVB`1KFuS>U2+Q$8(s
z+n^U|cL#K#E6*COcNAFl{JABpa{sKe<eoE?3M+m;eWaEO@gLNemmI}B=PsRO1^FfJ
z6GgDc<8=0py^LHeW{=ACsdyY<M4B@Y`k~AvbXPGHvISzH%C<d3o!xLB!@=g#Gm4U5
ziv2({%20A@xa0W`dtS}L^uY6KDz0}nyQU+Z@gF><jSpe=M!pTsjBT#71g({^w{c0a
z^wd)ysI#?3rk0+1Dr;Y}Wy7O`o$5Ic7Zi=g&HzE+`?;2gD;3lITv&?~0AMvz!yaX;
z9k|f-3wF!C&$chy-nMVA+x9E(UeNUo&Db_HqS*HR@bP8a-hh{j?_SV+Z9^0Kzw>9z
zW3+SPDZJ(>_8UYTr#|m%@SRV-`$=7Y&(sHhMAz{SPyb8`0mpHkAe{${yDHUwQ+Abg
zo{|v2WQTLg875}Bq$<)^Rb5jTj?I(i)`|Y2jH9n~PF^^^EVuSfDO6Y9$+}65c-l!F
z4)S9E1aRa0u?1Dr)#bgGynOt^$<E)DHA8+Eb59>&{U<W9y*9|$yQIgYcS|3Uu4B)<
z&Oj{p!y#%X2W|<B*Hh0H6W`tm$2n@e{`%PGz5I^k$>%V+)COn`T^Xr5s0&t$<aJg5
z9Sg(!csOLriX!8o%`f2LaNfn&lCI`|#-<=!72mk7>hPt!G_a;NboDwsqOZse{D#ay
zOWrKyUHpI2;_scfle&@A^rgIiR3t}RtnmXJ5m8=98R?@hV{eEaTMB30c5t6P)>-R%
zthfx-t32^wxU->$?kC((_hgNlvQ8`jzN_eKr`Ool3ezIGY*J1Bl0x=~EQb!!<Q+V5
z$@dDrUubKx>KH8zCTWGXSy?+-uCub;ZCZi(tF;VAm9;q{pOn>AMV3$COI%+J@F@a5
z@53G#<G}TnS-Q20rDbq%!(?Y}b!(WMvwCbBEYBH_v`tgez=074=mDgVF*9=b8-T!W
z29tZo(N{G<Uspu*JtR|Iag9$HuIg$w1>4y^_x7{42WPg1yB$;XE>0zjHb#|p!6$8M
zS+s5I3SHC;+b4=rv<+>MD!MJDb*<K<lhT9IQ_{NxKZ4g2;vCP4xUi9CGHL-MuDq;J
zkl7iO5-y4ewFmtH`wdRW!#3UQwaZoSs#o<f2BfIEtjd-`6-&n4N`vi({(ZI`x}&Sl
zA-BU}t1&m%SY54of$O>+cPqYJ_1R=Nr>d&1Qgc^TOa=a{H(9Z3)38TTMXyWvb(i4Y
z|GsGX)|oS#n~9#8V>#{fAY<OeyUmTw@=4%D3pKJ#1u|{$8r3Wi6mFsmQjH2p^q8_@
z9UPFj=OfH?b&$br;8g_@z;M-kMjjIjd#EyDs=4ZwZ&=1v+{CM^F?QdGk^9FWH=_Mi
zwqc@Yc{3)>u1SGbD$oq2h#!y}AOqfb1EhdZznbU1jOCMHZV}tv3e$*n<QvZnxnWoU
zjiFOw8N6PbxPj9)*dxyw_7xIWRLkZa;xmuU%-i5uhPa`NeIBr6o58e98}<;HB?Ti`
z`SJNnhK104LIcm%)EJ1VUgq`9F~=sS!g<wW!K?vperOWDraM;ce1afjWzUBbM8Q5@
zYNiG{b{gHc4m!=J%&hY0KR-^Nnf21o<dBK!3+eBJe8t{6QufgaMAt#WnVLc2AAjQi
ztKt46?c><+=2*6_AL1NJD_9G%jG}|jBX02ko@{pI(*0YC`cYMRRDu7|$<IyFepUFG
zpIG_y7hp-~e}Y6ZFR{#+B%}eZ2u-8^;}c*#_s~NheCVP3Culzj+j+e4gAaXu7Jse@
z{5dYSa_Q7f?20Rs^jS--f>9n}RN<KsUB*nR{zR{re7f-Bj%amIl0bBQf++R0;Z8Y)
z?e#p0C_i`;pjl@<uQ)w;JsG`8`b%3_5+y?q4%GI}*JR7F<j<;#Z4enVwsP3ULq`2W
z>cdm%S8U3}_IYYwv11{o3eS%DWjpKxyLhC%)b4f9$m77|ljX~Vkk8_Nz&GE-<7X@5
z(EZv3iFCN{4~b0R{~uZcpdknTlPVHPJul4HSEcWQWzFVmE)WyLkSih6{ttD~oKjJ!
z^qDe58?9<)k%Qwf0Ef$~cA4Jxs~x|3`#LyhsKD@2hh43DUEoaHxQ?YtGd(8eb2z*Y
zsuX?$_}!??&sSc@!{5Ac>C%OB=g3<09uD6OZUjdvqD;5p`>;}r@D2M}_b>79eS7kt
z@A=N_s9E9gF7|+%=@U1i4}2%parj*H@V43RD_&nScbSW(mPNal*;kfmMbMBp!PVlM
zlarHOZU$g^qf`z4!&K|t#*aFB`gi&zZX~=e=x`wyH<B^Zo4_we*xrS76TmR~$E&_o
zTfBZ&)TI;uWul+oXmlI!{~HR}s&26o%x%vF6)dDj>DYyzVQT~{c(_r<ziE^;z2KAz
zzG>U0Unn^R+s?ipM)$)ZT|024fJLN+xcg?<)7Rx;j}?B7YwK<G#$fO5tjjn><ujzW
z)0D&BF#O$SsWWfrU9Vum4TkOc0O{b7`XGj0r)I(H>c)JhRFkP6*u8-r`1JVg!tLWt
zcDYyZZO4ou{$~jerephs9{c(s{)yZh;1_%f{!+jJ`K2$t9cxoEa;K%`!ks<`A8|s^
zDP}YRdWdNVSg^u$9Krq;T#3d77kw}Q2iSeTQZ&_={)4((sx;=RLD5wm&#E)vJr0@2
z!1NKm$25)I5LD+H<+3|QiPr)8r%Rz@BqQYNt{oHut>l^Y^jqwnpML*4RgK*cu$wdV
z`7NYIqIZA~m!&=FJjUTWVwF9b9oUF9I|6?StT~Lgf@EeBS<A4d?{U^-10J*HoB!S|
z0^YkZk>c|Yb3T1MWy2f$u`JTWc4N{+r+4#7vXRaJl!43^?2#T=MXQk&ilt_dX&2(<
zynbd$4@35}{e8VmPq0!C!53s7bLjkKRsKSit+Sn=^!wLq*K1^v;!TRBR=*&t%RVN{
z`SKS6w!n9Smi1e3S0V@g7mx#6Y(8Nby=C0Jhe^S4y>;Bi675;TA~BIPW4nEqleTvw
z@Hkk7{&U7sdrr1d64^?7Z2?ry!dhG57v#D2l3_Pr;_30X_>1lPv|a*fSDjZEt7sQg
z7u9)iA+=JS`o7(bYTsjDqbQ!&X8S?OGB52EGCvOU_F5$1YKgXk({kj)EFE9z^_p{5
z!7dk^!?e9}eCf-p7gpha=j>tq{cB5&7LhvzyRr6u{qR{S?xPs*_wE2~&PWePkAWZg
z0;@ycr5GWGc^ZTjx^o8C4}cA;0UlboS^_7_JpiLRtPXK%ZVzB|o4D3D#Nx4V+(7y!
zJMP1&p3yC@;F)t~I9Dza(pt3CdPy;>hOE1O%PCff(>_<Q+){Zigu4#67HzFoh8xV&
z?Ph$-6>c#Yxz5xZoq1Dk&KVAMF3B<}(V*)%?W&D?<>yPfcc==!B%SX#ln}?5Yj$M0
zM5}PmO4X{HG`?w`+ZI=HJuzYZ$&<R0GzfS}^kf-fhh_doeh|whMWZva9>F?%o2H$#
zz(7&YY1T>2x_u*QlDMKMN7}lkY$z7|YI^$!94_gZT#o=oaDY2&Slw%844gsD%)Pzw
z(ps4;$DP5ivL{;6^Z!k{04>_Ezn4B)OX0aIS&F`qG_Fi*${}wpkMCXzbI2f8UMY)Y
zuu=;DULFy&%wDPbY!AGbKL%5>=d<IH?|=W^)S-!cfl2?4)yS5H@iVt~LQOdTl3{!t
z=07Ec=;ZIICG~=4nqM?c?SgPFrM@+Mrz}P5`a{g`NWQK`GH=6|Uieh>wLeJCZk+k_
zgTnVdntdDP;cSp~X&tPlv#77w+97KI!OG0F2rCkW8mii7-6@dA!YW@w?QMg7#dKDC
z8SY>jp4I>54JA;mkwdCyl*9SCX1`o4Dx@0V<_ftS7?xK&w7MY6ZcS62T3OZA|7&|t
zt@c)2TmH|who~8v%`>|@>qn{Km5$BVY93iuH9|Da@zxgVTP2IittHIGR9j7z80D=6
zaQ|JED1~}xf;;m2_iTjWMv?wRvt(JJrJkAJI&#vjgyG!bf~_gBzLtI;<gW&Dk=LpV
zdIQ`i%H8?M-W@Van>N9T#&n9oD-Cc4!zKABi1efKl3VaeV_|{ReslSSLM2=ZuPVym
z4C;q714X$SE>)CDL2t+pEkBml_(Nae5InD)Z4^n9{Q~h43NEMsCksO28C9!d$f5)I
zcCg!aRTEv}OPYFbfO@nm%Ux8I!M&;`)NjU05W7)@{k$hlw1gcCH{6(f{Jlb#*F-qH
zo(IRJ6wb2Hr8Y~K-XvWR#%{czn`Y_{1hTwm94aG2mQPQ>qN$RUM;CZtWTf!LE3{mI
z6Mk59$sM29r_n_f=&}eh?`#gwjE5X)ejJIvcS&0y+92xe2=gy(6FExzkBSuh_ElEz
z7Y`<e%HliY(o;!4Z@%urduU?kiOj?MS>X749;RUL_apPg6u52zaX0Oa`br+nM=6vd
z<S|0G^QoWWqcKl>h3g!cHeCwN3J8Yrkw|uJ1#Bgr4#0DNpTD5g6x;hiO7$1Kc@O^I
zw+9zO=hIECTG%Kwwf`vv<C~QrI1~h%wmozU-i0`afcL_7rAo{T$xg`Lb{})1mV3jX
zg7za;yv7xS{mA*g0z%@+zNIYqwzBYe)d|7!y)Tc7Lj>tiu-YtC>xHV;wCoFB!5!Lf
zOkv(sUz-eKsT_5R#!_rafUd{O+OlPY5j^pB5Zq=V`UbCP%Z|W?*I)Lkw>xa@rK=~k
z*F>9O5%<8N-Mqv)oh&VPqeY!$f?!$Su$?@*9R+{}@pG`cEzXWIL%3XxPD+dvmwER1
zzd%rMS@2~B5bK<bVEGvtwwN!&VmZqMuP_fgd$OT4nAF%*vV!Zgy9#^)dak+!?v9j^
zSw(3HyIc;lLL9z)zhVETEN7nh>Adm&ZPq>|dH9uSGD;bn_mdJTDmYk)qXHoZm{%S=
zO$0)GRI#yKGCFo$MR^K-yXXmG0p-lX?+4Hsg!^KFn@C-_(^9f>vlY#iaVR49v=T>3
zCi=g(vukaCGYebh-EX8lx^E=}{2#`E%)b3Ve#iDKJ&$01=L9|DIA&bvJ1JhhRTYaI
z3UqyR0h7xXKq>MxdpKDi;U$!YB4IB!^z<GK<7>~yr4Cb?Ho(U}&N{LQ6}2aT$&@Ua
z=#oB8UC^K7FdhLp9l>}mkG{7txgZPewWzkqe>sJ?;@k-*c@?YE(*bUP#Et=z#?e;w
z*WVq}lxP=pb+?0ZV!Gw=>xOoZcD}o7-+gZt6Nowr-l!|pdjV)E+ZSX7SLBCrM<Zm;
zw9!#G=m~)K_-d@6+Hsk+imNCo81H-hat!L@MY|ihY?og{Y5_2^t#V*sa$k7uB8T@o
zj{n+QpjEDMc8cWC>cjy^b!CE3>b+UH7lIEwjMjNik%`A2h*()brVxu+8}|m$Bxj;1
z_*fpXv<$7#lDuy^y}~hzt+$`WmP1G@vD`E=+jL9l`de3cN50sAixQsH4qoo?ywMa7
zpV=6vabN8)n<ueA_Jg;-&j8%o$L#!S{%Ek42poxA1ai4pw4Kn7Taempuh$GqPT=fW
z7Ts`iRMs<=-Se?!_pio>;s&?xj@kAO*mrm5*>~$qXQka<Q(#3{_G9&zIY5mF<i|N+
zZ_lUiLmVnA274(2&(0@o8fUeE>l@6U?epj}+IT!iX_}??ZY0GY2OkIP>7VD=d+?$0
zb}+CSU24BFX7!~uUzx=nqz(|I&2^8?pM8Ra>2v$BnI{53Xt`?_FEiur^7%4@fLnWi
zn!@qz#~=5dPHgHiD44-FN(|7fjGXXXg*i`?vH(o6<V@)~W|Fa0?u-$baLogIvb=z1
z2R8*pJ7EFkbCM^^b;oc&=J+P=Mcrg-1B^WRqxoF%izxNBs$hUA3$lEC7B8f9n6O%e
zIkk^9m_wPtL9D`rO!uH}Yqn6cbrq^~*#S{)HC0;{VJ(1dxQV+nItg7^Bh-YKV!=n#
zyq2h3b}!9riQ*Y?W-;(ngK-uk;G;^aIsk(p`}W&$0_8=<XUtA{ZI!7$xDC*SdklH(
zy);2@7OR~~?|p5wI4LfN@{O5R*(B|IsofacX4iNtNomY}9I_nv1H-l-FV8j~rzn)y
za}7wG+=r*=649?%0zSXU@)Kf3NYhM$oN?&Kd!2In^mLOpJz*c^{Ut5VmebyrV{vvd
zW^Woqp_JGrplsU2rZUf=!$|whNF@?oeQjcl2r!%D*I7@W7FWbAtW{aZ>qUZ=X^JOi
zTRtE?8(F47JKn;CwS}^Mk9Lx_M^sEpU_?2HNfsd`Q#%Kzb16Yy%^%ELIMxwrSdB9W
zvqnmli}?ebmFHVCW>v`c(g`K@mwAo-Tmg3Xp+f)#Q4SxL0kq_es^*BV?PJFjjVfDx
zg=z{v?{BG=EWn0Un{ltQ=)s0nma#KDn7{C%Vyo&HH0;gQeB%CwDqCt6BzFGvlzAd&
z=$P0A)kMm2>wud9Bodrx3~<%|>ycPQ^*=Tyn={6wAGX)!BdlpQm;@M-!*1*}&whUF
zMPkaD?P{96ZEKoqYqFUm&bqd#ypyWLvWQCGshC+Fsq>ET!#3@omKlYyzZRyM_klOU
z{%{|)@_RWz?B!{ZTN7}c&(j>De5wGKv~k5UrhQcSLdUMTv2hkx(rTsLF%0>j^!1{)
zvZNKX0UMW#ysX)=a}2z@kMVJOe_tnbtD+EX@H>38%DimUp>u~3KK-r*kzNMt9+|FM
zh8gm(MXMpRkXZF3CJ>{-Tfdl4<?&mFeORXQ>LJBcV0?<Na>%a7^f+hJTZd&L*LGE+
z^ezK=GJXvO9$E!!=|YrQwm@QG6-G2TL#Ekq!Pt&twjzLuXx$QajzzhG{aa7vxNC-0
zw0G^IWd@5?R@}T?!>Z*+WmT2`aGdRqR}{xMis=?d4M(YXRiWk|;t+tncT_ro9dU-a
zFj3aXc^UUXFjh;79_{CtiD2$^0W%S?znw1|UY&r}E0^JuK11#(pqVcvRAEf&V2KUV
z5t*h#XN<s+0}=hk0Z6T8=zK&vwI7x5oT@M186|%zmwwqb2kZ^}#O(E~xFd{E=Zaag
z*c7cbxqKK3yCrfY=&q%is%@*LX6aPtZ&>#WeA~rM3`xR2KbsV<O-E;S1a+)0t~+=m
zT+wXM<g6T@jzt)siq;>yys(IfcL%h=DQP4<C#;*r+Xv6Y;alty&-J1wyCg|j6w?>?
z8^BlKxfcAhZT4aywwOi%hi;P0m>$mOkM1|{vl$rM@X>u)WIK5SpVf!$S^9A10G^v%
z08U2ciO9oEpM|+YYl*cR)uG-;zg-qA@I1A4og3Rv=j*a8o=0Z*)&Y%b)%01NT<qYM
z=gV}$g)nq2)6mPNJBf3D$-1lHd39fXyKc#voYnPZu=o3n&v=D}Zkwx4)R_CQ&+D#P
z)(z8gLja(&m(Lyl$h6^^mi~6tuX|qMF6;eyV^0%!EXfDy)MGtl-pleXG4-O2=?gj?
zvLg4GR2VP=yr*NA_A}UteTpSUn=Gee2xbdw{fO0Xu<teI^Rj}^EGa~Pp9(*QN@RVA
zWdyXgs_L47dELq}q7dxdMczJP)cqaBa?4h!Y?g4>Em8EkV;k^SnoUhb)^yW$>l(dF
z23|2`qL?d$R1Bk{Vh3dSX7U_`DQ6Kv=%>f6!+)r&dbJ3X$0fRDgY+SuUva5rDT->F
zwL%c6o?!`eL!bXX=`rbP=|`mx;SSB&V~6!7crwd6;q3$}m=_naqMa^jW3x#{nIUU8
z8T5HpXSc_iD)YQBxMznYJi-lx<`4g~8d%j@-38A_mlc!R*oeWfojUk5AZLuSQ7)Mk
zLj$6ix=dgU+NPyp6JAwUkp6`=h2SvAayN%@{#*tE8lD;iPTjOs1uRJFR(ubvN!7M=
z;#sop)>XL+!(jr=O}(iQ!>~%40#+1^iF}?K02u)bbaMS+k?kdzHapWb)WpUVa$fT$
zDsQRDX`||<Q><)PijJ8hryk->Jg@Vu+nTOZ6tZ+Iwt{x&y4V{iXV8T-U61#HeVv0b
zHeW?}R2aCpji^BMUl-ue{*N5La+b{QCRuC4Jef@_?Yyd<*I+hG5Mx)<l|!<v^b{?N
zzYX>k+TDR3r3rTTwOOtv51L~2{ewqz4W4AnhZOyb2Z$f_XW1sx<C%H!w#iuLBrT3l
z#4!bU8q#5`Dm-)bX+PbTv*7fxJRSGi<Qh|VKi`e>$P9{!E$+gx*nSShE^Xk`H<HG2
zFXvigMXvLA`F?FU=;Ldp8%VW%*GlsPL;PL-BI&DQSGyS2+_h1b^TWXVdQ!}9+>gL4
zo%VoHXsWgYqa7wg+W+_~VOS`apPkTL03pGhtl)SwgRAGt^X0FPYoM}iTw}>a%_B_#
z7GJ7KT^?r|dsYwuw;zXH7oW}GdsiOuz9BsSB0_jMpS^?_6PMX&JKsNJRwlb>!eM5C
zV26%z-wYdb!*C4v@7LS1vVy@!-Sa9@91?qpBjCHa7U-Iee`vIeZJCr<VO-O3PU*3A
z3@=&+r$SUa6EoWn%A-OL#-z2%II_uNMLz3(RX{?G5o?L)DjbnqcD}*a##f`Pg$EGX
z@*+pp_6*xJZR6lECK~M#!*Ccz2y@2c+O6x?ua~>s?j*tARqB>IopLur(>mCgE4-t7
zo~6WNZFJBN@Y!sgx6eA3!}G^2om-$ZuECSRq20R<uK4xqqM`Em%hY}yKJ36Jof3GL
zU12Wqs4aH)!1OL~VcXzg&RS=0vC#~0P|>+%&C~~A#$8v)Ap-xEoYjJhS5%Fdnz~o;
zap#9u*i$u=XdgR9NR22pLVkO4Azj66!Y<J$j33oNt-^buZY7QO(hab~-Vm!dME%t(
z)zfmfhYi>VJ=|^~ptYZ3z7#|01&aakBPhYoiZGk9v~VjQR&sBG-a~mxJjFwi0>EX1
z^^Fgv(>~?Em<zA&YI%IPySrQ7MNS-bySMXsk#S>OUq$4{LjXq~V~hvp$$rY0_{l8Z
z1h9YycXzMvPUX)I=TP{`uFG+~eT;Q1m`0(rD{VWIAQcK9<f-sPKfaPGOEbx5&6&<z
zR5l{fawN@})P#9elRCU(`cPO7Nu(o&Zi0-%pTVjZjB#iGa<T?;SzHi_lW#BPvF58B
zGPF*>kkgmBUjL&ApV!WHrrVUWqRb&{TEy0pF-+sKT8xyxbaU4SM&&ZmbdA6Z1wXP8
zzrmM=Y~+M$b|1cgy}T|wtg13qui*JT_u%o0P89*xmUNU!uXV!u-e9krkiYD#V|%(D
z-*>&=)W1w@L(YQ5`ew&)zcR&A*Hl9nfm5lhsuAeJgT%pR#azNnUL~Cw$KM2R9MczA
zDqIk-{H|(l4F43|iHQ&a%TKFf^ggJv%uZ?lIG9i~Y(J8Ib`l<TJ|ztHYm-2+oUrF-
z=Y$z-f61=&Fq57OB&5BmL53bm%^;Se{aA_)xS1bTb)J%l!A1~tm&-=6WZav_%nK#_
z;SFXDo64G2)NJ_GWNLq^RMHk4XW_q^@BzxD65L;6{~xz$sj1tFqriV#Z<fYm;Mn<S
zO?p&_fs63T?EnjcZIC=ecLG276)m(&wgEEMAgWeU9e`RTMyEuKQjwCEMLVp+S`Oo(
z8(CyG@wF(zT`X!%+iq$Ph$a|AlN|&TS%_PrJ-M$PnQs&3ntJ1T(hp`b2Y6+#@jTze
zvzwVQaDt3w3|trTe)r83c#X%z;c!#fM8v}Aa|)FeWwDGOmKQbI(B{~6p1LaV!4c^M
z;FIPE#3Y)PBQR09VBZYk{c;2Zm9v|dF8n&;!4cC8S)0VGk&xorwA^el$Nz9nW_B2<
z8x)6J-^5t~LNfOC{j&jJ44&H9P(XDnTU(e6rX=rJ4wJZ+vBuUic2Zq8{_uVGbo~uw
zJPjazqmXjm)_^}y1c0*ODaO-NHVP1dcF|stbasv6@ghM?1e`(A7|mC6XwIVky|x(j
zcV#MWbNXAeg8f)Vrem23`7l_;CfL&S!$dKyY@B~Jjq?xtVw}GzhyDG?n-}K4o=4W(
zIdDwfb|JurRkIJpG|k5O^CSH5-0d{J|I9>ezv}-|CbEBdGOC|lU#+qA#dyVFLXgE%
zJB95Af95D1q~B&V=qaXhXjP{2;i#|NGh;OXZ}b$R?$%@?A0Au#LErfFL8JP(iy|+k
z8%&1tuL%3)#7OQf0LLiR;~+Qr7b&Dy0@*Q-+J%CB)O6$krHPcW2b#(LxOhuc$V16N
zwnQ||-~b@*yA9C*Yh@iBSTh3kQB}aDo4ZvL&?r;8$Kb38kQe0$wWIP^^4U<y(0_ee
z1b;TQFU=(I^?k@H@Bq?vo$<hIs@^#fKY2%V`|+uBR?HvVk<Oz<IJvv$fRXyxL(4A0
ziJN|G)xt2frP`n>`5V@+4&Uxxnv7g$HUc?1$Sg&}NHq_{tNe-%3Nwz)Kh0b=L+ixe
z<#=rVVd>IAWAmZ|DwY;Hh*^+>D8AX9L{Yt&XBvDvMK2FwcIEykB+pTOHQz!Ib=7>H
zuF4&#mvXmay%Mc*iZS@cL=1jEUKV2TjXWl1OOTp2WqynJOhm1aS0^Ly0$%lvDB^DF
z4I$zV&v<Crd>6>~y03na=Ed4s^aJL>0?9sLxF5Q|W4h@>-;sy#zn5(@^YpV*W^1|*
zYJnHrWV?FanT@nB(;XgaujYr^GbT=$gxLEB4zX+Mlo)$zE|m`yTz@}yhsWT$u|h1%
zq!mTOa4~!txkOEBVsu@9DmA?RP*=3Ds40%C*_NiO#4A=@-Lhp_b<D6(t!S=cNkUv5
zeVY5Fi_$x!%hJ`T3fhcu3ipYt17k;(x@HE}+i7HdlsDMrgM`&BYgSxc@5X_3k%0wH
z`Iici$Z)kv)vlY4D$ACwyOp9xVAw26_bo;93XURoU<eG1t!ph=28x#eHlko9<?{y2
zK#Z-MS-Z*uyDHD(O(HW<h6g7qP(qk{Z63YeU9JZ%qlFvPMFg(b78f^`mVU%CJS@Uz
zgqS8FJR9rGtYha@X0~Nx<Nd+-=*B^k+u4A<Y8&aOjCiCROU?YQro*tn4~3K8&6G|2
zIzqLvIA-_f$x<OwF3QrJ2swQ1ix?YMEcn$mDH|K-HTDMS*X<``(%5(>kLO-|Tl%bj
zHbu6GxJ_g>&EDZVdsnZB^Oejec&~u>eB?LIsPK0n)tmW#W#N-~>h;zHnD5(nf3C;J
zWtIjVk9-T1`MYWU;bg=cNV6LU5>?cHco1zf)U!uf-lwQ(gF9WY7zxRb2^gI`0)3A!
zdM`z=XzI83r4MEMOvOldR@qF6aHo3XJ~6!f3I>jErs97j2DLMGEA9*3!d2Jue(2L~
zLcEBL!-%oEc{KP00U^Xf-Pj8Sfh~l*=;j0+MurV=v8ci|1YFA09zmHonA{Py{T3qH
z0!O`gr;%*nvbRMsY`-nbB{sO0W1q^{ky3W=XPw#2=h>R-bZFnIseC;T2QhFR;c+Ez
zeEw*EWTqI!l4vKGQUR<f?1IuxW7$z*j%3w)YiB?QxT4eVMQg)265p*>Z{W`rzxwKK
z;CFW62g&`w2-I1()GhB;{qAnbua4l27&r8#Ik4ZBfiI3p+nAFv$-6NW?b%=moZB_*
zTP%zXKUWO<Mi<^-NN|j2M<Z%qLEzHJHUO8^KSLhbPSjbB%!?x0yrLm$;PHsI;6(pi
z63O;Rbuow|*<U$EWK!e5KPN1HY2X{sD5auH!CTPA<JY^qgSj5XujA5`?KptlrULZ)
za$zQd?M(!*pUW@xFahMf0y-~FV?1ku?;4Enj<5*wZ0>7khdzuV$`L1y@CWq&rQLGO
zi=%vo+>KchevmKi9KJlt`0}K5R=WF6xpNFmjCJFulbnjsvCz<NIp)<p1b7y|c4Ch0
zc6Yl{R?gy@<92REDnWu~H*sdfY;K{FpG!oMAeO9s<m`w`p$)D@YmrQf>z3%kT&uAv
zPUo-(U2epiN8*JKgdGD1VC)#mYXa;Cm9MvFA#r00gNt@#Uge#AqTmAIzj)B-H_-c;
zhwsMV@!b!zRu5Ox(DGI6HJ{2htWI<;jZI7kWIv1h88wY^=fOwRLY^i5Eutz@K~?o9
z_FwbJ^S-7=yaj4r*)P%!&-@6xqGsaG_*YpUaGpiiq#>{Dl=t9$*MvW!MBh9f>CSau
zep^xU08@+~o8rre>#sHN8Sg=D7RP?a&C-DCRbJQzcdJI*(uT2A>!-nB_=f(?5~0N+
zeLOW<&qNVnwS*C279%bv5nwou1-8+Yy*VDkw%-*8xXY!|L_B)~^9N_fv*rEd0O;w7
z==L=+p3R}}aO{ZSGQID7JPO8cg4Xna_9t_tLK4{?^PyY&ugc`Ep-fr!h<J1^*^d~_
zl&IK7^aRkXab>?T_vv+%X&yWaU#hH7^6QonJ=IDc%D@4KjW_a4yFf{q=d+r)`23)o
zU^qzXBf^NJ4<Co)jq84up?y69g;n;4XZ&;if(L|#RQ<1WfNpu!H(vFt2+J+QW*ZB?
z9>IjY3x5#B49`gSN+aN=CwU%0o53?)rv%Gm^UC>{p@s`ztV`8toH!u{F|eksPKW)l
z-)qxh*dB($Ip!Z=tRgOCQj-<Kq^fHa49lg|wRBrGT$P%}PwREvub1k}X4%{l&M{Y^
zY#Xn!Oodm$fd6w{H>s(sx++s!H$2<4JyTVvje!I0M5QtsooY5uMXy=ka6YoFqP&Sg
zKM!MtwgH3oh72gc{6G}0ToIA@ySfL!`MP2f^Q=1o;G5qok49y?**rQL34R6caM%W)
z{UPaT={>V3u0%zO+k?|`4_Sj*B`)cqNyTjvF!jRzkclJOhU>#VuQ7}hd$_~+3C*u@
z>c7@999=b3MOTSqwY-9+D^#^y6<$pH94T5Q0!A+hyeFD)r|uY@*OMuA995r0^IMwf
z7#hJHI;LgoR8xp8>xOP`t*&m3HaF2wb8T8bqNqBZQ`mfncyC_1pY3g$H45E9AD3E@
z7T7IJj+$#Azrk&oLfe!he(Z@n>dILb4Fk@iNQ}`)B*%-gOH>(*7qDn--BhVlbSNVR
zCAw-;Q!@&t?$Cj(8o+QKm;({lQcE?6r|PCLNfzu5A<IP8s4B}c(8vOY)n!e_#V@#E
zxVW=&eI^&jJ;*t&?vXwq{lY%fs)T&i9#BPHu#L4Jf__qO8ngO%h5n7GA~o<1uJB|1
zftYxp6!|@aFrWBgMEpusAU@I+6Sz{kei(SlFbh;Obz9S^rK>J2_`oZs227=b)u@v*
z5lIcz))`I91<EO^UKt;jf?v?_GexJk8<h<UwQ@-RJxf;<_&NgzK%+=>AiZuC@B$E4
zlR>=&)Y5?KWt%Fh$><I&OJx`d-O_`c!LzjS!ILJ(=fih4KG6`o<?(6v`r+oVKS}#5
z6dBj0fVT}cDd5V94|5cV84>^7bmIR~sa`A={r?oST^gM@k<mYxvvQa8i1e|U(lFpS
zvkBdl!vUvROp{&^OG}C}5N?)x?cqV?fk?U2nqi_V0|(D=1J_`O>+_CBspmL2=M~@r
z>!<ES8vH<lPc;q543nK`X3?@d)9@|J@(sfqjLc%qHyzzJfIcc&oh=k7qg^A3KjQ9~
z7o<<+*B4X92sISxh@Hv@k<Mr{y>eQ{`@@9yw#Rr6zB}T*VV?r~f!-Mc5B4)<%GOw~
zBX;npSN4E}6_0*y-T`8Iwrp7%mP0o`p%^F=67DM%xCx-VvskMYed7CJSn|lXuU$*o
z?=LJJr=rfu>07{AF&X{A$SBr|=9w+vPu;NfnOu0d)19C1cIKtYJrL*inZs!Z!~T%z
z-H`T~trH6Uk0z(&=ayQnrDMnL@qOZz!Vq5kVr>ns!1c*__OB~;#Tkv}=R4iGxlY{~
zITd?M-duoje{43abbDQnhfijmTrj1zpYp&CjJ}LDOoCwoJA%HuD}w$p7079MR5cu5
zvsJn2tJE?T7%EMb$szBEcWkLSfx};b;8#uGv}nPE@wPPGvo%hA%d&~0$%E0T*&I26
zgX=XBUS*I57kH1JJbtEn80>F{_M(j)Xc~#562jFCtI7(QW{G~(Yk6Ml$nxgq^5WtN
zS7gK4wx>T@S^~UI7e-FmUIODUS&!m#w#DksotlE52swAlf=M$kVEdK?Jbr`$yXOUa
z6;d)|^X*3$H#`yU?~E1~N3B-j^yxz3bZnkE&M02@E;0DJ;2UG)<UD%FxiSUJf=XCL
zX~1@?zz_^mV>wJ_MRF+v>X&6#R&3349nI437i{P=pmTguEIuY&%S`@%4cw+^MAx?a
zg@UTvk$v3+g|Zu+N3<5^$dl5$r59(<5>N(hHc@mz%rW+om)GTpyFgUJ9O8eEVWJ-9
zLF0ZMu6Ho=nSFmXn4J2GOgu%^fYU%IQq@~fsqGaEU^I|p#&0U1K^@oAY(;*!>3PiL
z1~m#sjzj6hwEA}{m?2<eTPO<QCe8LxV{2e(p;l4p8rn<+r{C9gO;!e@Io)<0y{=GY
zp5izwf(*GIv!VYQ_IJhv>~X$evAl>cCavus6U;gXQyU^{8=M0j6r`zvv;ZQV6uIQB
zTGtxTT-J2G-**ZMV{vYGR2ZM1lP-b>_EG7j>0JKTz<gZMj$15D-iDhfnR=P|@t&;R
zn@b>SX9@IbM&p9A1&!+(HMg+jI!l>3^i~MBoEk4$WV@K2#dNRN*cxrY|H!`<Y~iP{
z+Q$#eSDBc3%=p+$jF0gN_lX0z9`^~;?iY)-u;`<~0TO7M(Q6mD(Q9bAMlXCom2}$W
z+ZPy#IK7X~#vaKiS%zp*h{gtROpj@TNrZpQZxE0iL9y{DXe-cJhrK#l7?T$Kcb1n&
zHFN1miy0UT3yqD%MKQLE_V|u;lGPrWWsi^RBo|Z{+rJ3l(oNh=alqAxUlkPAhp4Bo
zxTe{$ZPH?E4+T1N?IBrXDAXLsb)C)4a=E*iU2qWTe+2Biv-{AWP0T)1vWQzaW+5wF
zs%0&tU*+%@5SRPs+02LyT`>120RImRXa>Al48dmYE7U9$bRlkq<Io3QdGL-#A?pT#
z0y{#+4fW~Nf7^tU^A0~_iO--v#{*m#&!ozkZf9<eF~2tE=iQGHt+(vu(J^%U?Q}kf
zNF5c#A@e;zCc4It3#k56^lwRygFDqzV`Ey~I&^4jbnMt@bo_V*w+EzmfV{gh6Sp?E
z6}ptwdfYxOZ}DWp01Vd(cTYi!G-ckLbbMq*TcZ-&8smr?hN0*R#e7Txg4+bWk8Y|4
zg9$DXQzKMYRsEL*Qhc=OdZrD^SOE(GcyQ9<c=InD)JKG1!l$lirmbRy9je;+4rC)Z
z0gA5c@=`(`Fgl!q1G5i}4ryGz45TyCLw6`&a&4_}XV_0F3G9TMc=s$0zcOxK=?jKl
zHBEO#ry;RtGI*U!6;)T}sP5XEC~h35Tb4Azp28T|U0CmMrskW?<sldMV;)&n-hVc$
z@Lld=cZ0oPFV`2F$<}^&#&OO>o|rgR=nroKM)$_EcDl~W<IdsaqRbI83#*^U;Q>~(
zWd%nE#$-Yw-II&ncQT_>PK1%ESyU-Sh>hcQisb>n1-!y2n*lfL&rovv;C7L@Y>g(H
zdwhjmFBFtXEtf@Z7aTZr<FYzVlts9fhE>C6%&r4^JiRIC!_-6)^xDDrWOK8gj^E69
zI$s^9L5D0kF8toPlVfrT6S*~e{XS08yfMC)^|!rq-krkH+#^50@D9c=Y=d*YFXLu}
z47%w_PMROB9UKPh>>P0JlT+3Y^7(NJn<k=W`eNP<FvB$6K{Jd{9nG-0i7UhE@!*N6
zi+111uxdN=?hvj-j1qG}SQkbg;!*HU--}uLGo@)Y$Bzi}{fKG&oLvIl<`$l5W?Tx;
zcfiBUrvQ|}r%(tw6>!1O9UYt|bSjRFkNT6bF=*R-)$=WS7dyZjF~860dn41YMP?)D
z`0+K&t+f8t6rP{L4&1|tA_|u-WrBuH#7i9YJ+F>F(<D!4M%a!#3@KyVt^*gd&P+2R
z&TO#4eiTOT#byz!Ji-FZG&2A2m!cP)l|?Wm3x!q`Lw3RNDO3yKqv>bNviYFvmVKKP
z3I$^O6}ONxF&m9?dGx@70?6nqH5lTc*)+CCL--kOOJjO+mGx(Mkgr<qM|Yh~HlTdK
zQ=UVX8;wQ*4fwc3jM$|teQrB%9pyE2kErey3kA;utIg2vSzNRnU^|U!)UgYe1{_LN
z*9N(nqf~UjOKCP5XAreBBg@2^l)!$DshzFJ8r54>-O%M^PL`ScISYQ_gVM*PU!AsQ
z((5<niEts)G?@hxalP=s^S8H8JTemu&IqwzAjBiOhUO|;bfSf#qZ1dbRy7VL(gJFz
zhlI7qwFAxgeqQvLhbavqd~6#N%P{JGt1wPwG^*Imrge~*n|YvxopgOPh}OU^OFt+5
z&h*jDC3ntv9%xzqAVZB72=M>Zp`LtpW~6@+M_R%BgZ+m3k?BkS8?#4V$0`*t^zON6
z?5TodZ)iM34c8`O)OWrFH#;++dk~#*K$gg^2k#@nrmTwQPQjrN%a{~k|L8yy2Z;A6
zW+=baHQ>9@SZ09%<oo{^>7Tp7LVi*D=u8hZe|>UYo)k+x)>S=q;`&>|us2}N9qTLt
z?9A4L!zgewmX2d6ELN{s8e=vMgUSW3kcriQ1jM<)VyPaCv$m`nwgG6Xt|^0Gq>fta
z-dnUR8ZJx$WG>UUOv^W9;}9C%1RQ%y16FegsEWc!T;E0dJt+Oy4~cphFWLkWRfG1r
zvY$)xmWX#B%VsFP1RN);x@?$=34WospcsIep;8Yfd(N`Qfyft(PMj$0=dB2~56?YA
zuX^8ANJJm!P?)R%&vKr>1pY<YTWvO%j$O!glK&L_iNa_!YJxjFO5FXib^C|+jR+lB
zh|`RB5v|s5PD1|igNa*Q-S~OFP|uG04cN5g6kO~WrOWe~9A+Gu+1L-FMed0OnocJ(
z8b`d~Bf&><2jb0don{f(Xb7@-iJMr#WgiIm(=)jTqGvh=36Sk_(G!vmPI|%dJfP4O
zixUna-!r=&y(tZN+<NM)XT~@Ki~-jd^Ff7VB@w=#bxFn;e@w{wyt_Ch4e49kktGf`
z`sie{c@mR~5?ggS$&JCaxGA3Uewi`tG4~>W77C_EX?)e01>rtia4QLNW!L>8PSAT9
zUt|+oK-q)Fj}u#1MoW^|vI9S(!5}+v)l44clh|EiKh0e^m_^mgrnRuOuE}Y;CwnS0
zv~+Ezp|zJ`zR{w){+4w)awp8&`sq<($MT#7jFUVny=%%y-@yt&W+8;>>l3nYX)`n8
ztweA|+9%gTxR?o|Y=fr$86If#xEK@J;9zS3k??945;3L<^EAmWneJI<hIdMKNp_ZA
z$qeG*Op?j^Wb*TOB018@WSK}AIB|aN&5i*?GA>>S*ayZh@{ni!xRAljQeY;<$HMyJ
z;zlD2T!F!MlpK{j;HVI0RZG}|Hw2A+Mv{uWZ+lSWo!bZHXd9any;9+IIQAZ2ZLn=I
z2mj6Q8}{hNjT^6b;-A~eS%=i)Tk!TPS8iM(dZ)A7DdX<8GTsUYw%cv5Blb6SSe?`@
zNs5@Avv*vC9T`8&Y#<Zg+*rME=+K2n@o?m8xC0Cii)lrgPSpMu4$*CJcpbw+z^3PT
zeds|_>s%9Hs!Ls<-&x$rwjRk=T<Q=(%vOq-@@AdfSUK5lpS%wbD=E0u6tYeVZ<k|$
zhjR2ca44s+U*y1^;-H^oFGQ%hAbmH2o$s>2Y9C1z?o3JU_)V8>_#+zMJ9j;)0MF|A
zy{>APqY}3#GY`&rUI*9X+<12g-+C17RuSJ2swWR*&x#MPM^Bydz?5J3-pHM3zfHO^
zs2vw=nq---h<!KOx>WVQ6(gxFrX>WryqFh<QrMhx35EJ3+o&17@mMw5<5-b;tV(}>
zI1X(FsC@Q8TnLS~<xrA9{$&Bkj!^y}9XBil5^p9F3V?h&f_hJ2AVf<T&w@0Ed&h^}
zowR=P6!`GX)qO#UN0i+r@{kwl!nRbB)>vfgIq+{kBmIo@E7Cuf{zvIwNq;K+2jWB(
z$8_A-h&30Hez^4&S08m`XI5mRLa`kAU}q4*_4Hnt29)oK?BcGt9`2Bg{%oD-tcoN2
z0nPjYenTa>0MiBj3a+_WWQ+X;wpZa5*Vy28d);liRn_?)b=ltx*`J7jz{8n489#iP
z0=x(ud}VYI`?iSCvTqP!Lty?kkkhKOeVTPkTdLxcCm3HgmYa~z;fFOs4TaQ90^<ue
z5#SuSzh)fXXu&<z<pmvYpui-u7P5Oo(AXbUbq8!is*sAUea_!$sJN*!JY0|sU4BWS
zieJ=Ou<s@KQmu;RlB+5q`ae)K*>V{?;MWuVOO5lpOEAFm$96*%ETe)<u+WjA7hT0@
zt6*%%mH|#J{SyR0F$jY|Mib&SaD<l^A5s;WDi1HTJa9&F(-`s*va;^jtk#}o7j$*8
zd9SJhRQEL&RP7L<$|IU$XpgAM+p0BVrG|e~oYRX%1#W8SYPtL%t0Vnjv7~ARP3<h7
zQ1CG)7ME1*ff|sRYD6nm6rvh|BD<XmFQ?)+ZP{4YEv-_yADnRQ0lx$!!a~3cb3GNa
z4|=sZRr`hF(jIFqFn_G!$}*+ci`Z+kDyZRMU%3oM*ERyUt*YQ{*GgeUVSlX_gJP)a
z!d}7!lx?UB2iv`FuZo53cfh)>QcSx*Rsqd}%z~w|usqgjT*pLX5t!&kYrAPVbpmFr
zMvI5%RbpDzVr5~wRa;c4Lv#(OQE?Sn^$lyWzP>m&zdUDoh9ZYMhqTZI3!&()*B968
z<@s8{Efz3>?JR_H$?_c<7HZ7*B+1^u0lcD^qrH&o?PY*>6DC6&iy3j2F7k?nL#!}X
z4L7S{wT~Iyn<P+Sz~>jsAl4jPH;v)~OFV-A-eTF}7tP{=Qy_<B?G;@yP190j%%wFI
z^`mJQab3g98c|G@iD6lQ?TNPv9oAxE_qW;o9trp@Ey8X&Lk<qUX3$o}cCe(bW*YuN
z;I%3id#kgzcYt11kxh1wY|>9?aH<kKk9FfwRu^ystkYA>;$sDexVniNwJ<k}E8_*i
ziud;iTx(;dN`=+^8uY+Q^_@o1e%US($Cx*K$1FNubBd<p@7`jj$*VV*h~EX9&v*RG
ztUNH>==|^T8~A%B^wn?hmy4rFM;?uO9h{IZV1)bs=kDFZB)P8pu)1|`y`SCHRn<My
z^XTrG?wx1vOwUZuU|-k;hy~Dib3uV5b_jx?78FTomnK0H6d&T!79|UkNJ=A%lmtq2
zEUxt_k*H6JEF#h;#q=MFKOmO!Ih3QsSz4d$k4oai*!Z(31IRh|)~kDZc6PCV>;#yt
zu6Lbt?(3X;&-tAIbjwbGxCFh7lagpD2-buhewVbDa+ns-x8q8JQ`Vvx7oeAZ<4*<6
z*MJbJFsQabIx(3dPf|uWS9KtrGq4v3eu%cSo6yf6tY2fwIov;|us>yVQceyY6wJK(
zy>omg(@9o1PqPBEI24Gr*+>tBw=|@=$Eg=V!DuRg>in(bgZS}($Z;fqB;C4dB1M7+
zoAYzg)Nvkq+tSbP7V&9UIEH*5^^M?W<){id_A!xZA!@mPLjwE;L#=`FEYM{6y23kc
z`=n%_R?<8!+q|K=eEuHomSfUDdZ+XO=~L2k(l1EACjGATN7A23|GV_p1mqhautAK2
zHfFY0HNFsuS}k^r?G8d5bohvjV`(2Xzdi=2#n3B8aZ?RJBpBI$O5SLrf9Dp4^mjQL
zX1vx#)NdGFWdvN=?1;;F%(Z)H?pxUzHG<)LmB5aNZxblHVqe~B$9zGQpTIt%1KqNW
zzwo8D+F#=|;DIn4;w?Pi#arcQi>C%mk4YzmdiyX7qJSAu#Uxe$L%V6)rpkBCN@QH`
z;*xHGwGE@H4kcrKOe;kprD0UkFx4pmdq272Tklt)m}Clyp$^4pm=l`EpbH~%$+sC}
zzojzG)3SRS!&etjE7eP-2}&nQrT>{Jng%K+m7(BO&1*B;FUbVToKoF1l>e47O)p%(
zi+K*bB)w@~pb?Yk#@QUXNG)AUiP1G{PcRC$1~PhH2mvmk5VVC6;yOTFHK%QFn6~?e
zul@*XTB&@`flR293&kIi=r4_2Zw-s|tbl1w*5sEt-Z=W=-6OqSdNR=g5??f7bcVw4
z?4b%!0hOiQ)Y=^ZSi7}>V2#4h07fn?TJLl8LV;bVR$@M+38>E~7T8yyFH~BU^zTT%
zd!GwExb3@*)rRMJn&KF7qY)d9dQW;9v=UEMv-4e$t8ZDO?-9FKWvaR={3&qv=lx|L
zv^jW%u4w7WSW&+p&cpDD=h-QImv0G-Dd{z^N_MzRo|Zl;eO~&q^fitXiVYG5>Ma36
zTtUoB7sqFg^ob1^6(r~axgr(u4DP6@chpehIN1m_0fIY8cw=ofv3<tajX<kE#z8|h
zo?UU1_X5~D`w?~pOB_-Y!C196??RyCmBn0KCzb0zcU1>QcURR_1Ud%g*PZfpM|I(h
zWtJ-j=;7fk`YITevI&~)F{o3h0gIOT6!UN~7W7QVF+owEQazn%j^ij)(`0rxuY*hj
z{gJU#xMNzLxI|Si1RWd~l0D1Bx@VZjMd9p$u01`e>lCyRv<5*(p}Iaf4a%QP6;MM>
zMO-2wzowXQPNDaMiUE2zvup#E@Hq{&Zo@%DP3G+~=6axnKn-EE%F{#|CbS;MG@gOF
zPl&ie8T$Z{p~rxhK&Wo%%6wA9**G>K&4KQMcs4Ol<8q9rDDNiF5~0@XZ=vGChc6hk
zc!{N%odk#oiXQv=b%W@#jXuI31~tpJ^namSN?DO(f8fC<y|?($`@xQRucli1_bA;4
zCA>~C+sHcAuazw4vKN$1?Q^<q>-|l?W_cEvIf?^)h4u4Phx_4nOi{8<bA59W=&vG%
zZy#EqmrS|<Nqu1|!w7&_;~hljQ^VgG_%SE&5517svQ8rZElAmOs`g$-)!w38*J?gc
z5qOs!tK>yMfh;=;Dd2J}C-OS}I^<O}-fu!w&tpm}krC+-Dg9PDp*BqGeXoPo>*bKV
z3EE#|JV|9SSL7i)d@j8ipAv~BQ-0F2#@XstN_3cV#`6xf1^vfh;FmR_XYu}oqwMw)
zI5$EwLK99zeTct{B)<h{fAAR7VSsWi*HC=BY`Qu-#;zfmo}eCTh+ji;f$I0n%S>@8
zGpzDLTv90*07yW$zt8eKHS+muq`M}qOJ{gy`vb}940Do~OPm&$sNJ1HIsGr=2>tdj
zS9!5GysT_uX@3YM_efeuUl`B8y9=x$>swe{pVLfJ`@N*B3gv?JYyjM4%?>H{NUOh*
zloB!)wnW?Ht#JxF`xteu1Ckt>Y4h~3*+LD=sQ>VXIcJ~dO%~ejLp)niNL5`P%3}1d
zL>sR8oBf3LGB)i6Y|*m(clM=668q9{eLKQ2yGQuS`<awCEHp$@ON>0Ydzk*bKstF5
zK;!wWq6;Q6Vv`QYNQ4coWL8CbGrGW*k)~3d+pV~_#gU`2#fllMgfz@iBaBW#RH%*6
z8u<2x9T~0zR)cN;Ws{|ElqP{v-3YF#>|7MM-Uj*yDz@yK-!OgIR+R68O+p;g^KF~?
z#uMc@E<a)TjGM`zU!2z16Q<{xPcTi&v5YjZ(&myp3PpO;3_&RnAg2}40F(LaB89&F
zBIdGDQp*1RRX)*~({mZ``G(3?IVpn+*xjUaCepu=t|8j8$aR7SIuAhGL|2g>U*7wR
zuM$HxgFuHN{XQ7L^&l{1<Ju}H4qrrV)-X`B{URt5**uIP{WF|NO>+-ViCFpQzZt0N
zK(#;pY5Sf&b8(o{f_nlUODo2#vh;0`#~R1#6EX+|U{#XFc;h76EogAz_=-!+SxTgD
z5U>oNd4?&J+(*chpqVQ?QQklS4gDG1d*)k22IFu`WdRKNeL~45y64b7@MX%|l>6zh
zdzhM!p(_c_keItjdj8y=<?|Kc5Dplqx{<%<MQuJM?nek12_`gJChf=<(2Mi@O60+b
z_B@Bb(ya4-RG2fdq(?=z@_rugk-=CAfn@7^JRcNCm*AQ-YMTj(Sd@G&xD^&~7?0rs
zeSu`e4Nk*L&N~J!51XoUZ{%I_B0qw0#^eJIlPwE({EC9^tzd*#eM{%pUgFrW^C~-!
zVunr7bjyZOc0a7SmTqh8kGU`+-@xl&gCApx4}H9%D>h!Yk8|@JEOJA)1_D|3yUFsC
zs($Lc6;6d#wx+Sd*EA3}_lKm9k6CdUhr`h+Kwb+(;>}!&ViKYw&|%;f-U{=aR$(@y
z&ha>tbd&4RnS@#3C}Ed6j%I+CYaR2Vf?#wDL}lXS1Y@C(rz}fmd>A)z?%R*NV9(Sq
z!H}t`>9U)8Rj~cQDW=)F<!ZkZ`7aC63@nlU3IuuF#)~#CxPd%<F4=QDo2(xasD<;_
zu0hQkuFq(-<cOoUUYSPZAKY}s>1Kiy&Ew`O%C%f%!*fs`@cEwoIoIC6W|U4__O|n?
zKY!EgU$LO*&AnGL;P9sp69QHD2?6_;Yq?&9c6&uykWON3V32dqdEXLnB~emrQ9e7D
zrEQ9mDxa-XZi&0pbbSeikdmG1-pZ3;C?=Y??JIx4e{kuCpL~H7?u3fEEbh{kTzSWZ
z7`*3OLvQs8FM%^f-hzy`oTMz7YN5or6B6g5qfL|j3^#JooJ^-1x1eRhXMLLwACD%~
zsz{SUzQNQa0V|#*(@fk)_*IqElw1tF0u-;ZZzMa}%kwv@`5J}~x>a1n1okk2u*c^{
zJMGQ7%C1tCK1{$&RJ2DZ;XO*5(s#atDkLC^^1*jfnzVQEuGYbC>xWp&`<ol5$i6oR
zkH~HLk@4@Js=?+d=lkGvO_E=G<+WExmvp7G(nIjARk1D#BR{6;jQ9Z`w&AnWxMs)?
zwodWfMiIh-w)sI@WVzg0<@9y5bmrnWXdF|SRc*zNp?}p(M|BJXHl{{ZBeqn#%H;GO
z)jIH=dOSP3Iy+0M2S0sNU-&J2+v=>OzV_;CuL9|-(q-w#r7uW7EB%7>OMIS~`+}j(
zB6ydRJ#<Mx5ujvnq9+o<l9Rnv;)?ZFbS^_pM@U6Z_HdF3Z$g>jByC%WQBL-F(wVr$
zCvp4mycT!VR>*OU<GvREZQ1i=nP`Sdp|zQm$dsc|II_x^EQ3mGXm~=CWyVz5!D||o
z2{lV9qo$z|`Ig^s@v+Uy(WSa9*O!h~Hjfh!gztO36`4%G`5!;@4L@<#k)tDv)gzJY
zoV^43Y-jGsob#4@d6SY0W&P7xUFXHXE4wQvd(%lF@+zRFHGf)8_D&{MfEQwvHiMIL
zvWE)>X_Jt*(l)UwCwsWO0PUmT1@x=`85C<!Tb`I$USFD+So+VRAg|Vz*OzM(OF)st
zUi*)){YP??9F=UaFwRTAjW&jxc&uVPoakzr?fx6ndv(c3>mS?Z$F}sb?VGpez0zc<
zG}*+>v0AeRfA=X<QwpK3rg&4f4sy^{Fh70xVN|!LJVkRUxq*9npu19#t><fSs8Gs?
zYU`?|!k=!dgi)%%3pC#YIp4&+f0B?3n-e7@xb~B;i@yE$kXxp0{Jqe|wbx$bdPYks
zffiK->v&pffR!!<$xKH!yW5}dXsv#0M(%gPXpTE^zjYf`zOK-1S^f!X%FjXlbemo!
z_~}6Hc0cAdpIA8mM4!wsHjp*?O~zi3RmxUYwzq?y`I%s4CHM^dmyF%GhP#OvH$j>x
zKGk%zNZ9COCm~m5TxRTRtGvdYB;V$W26u0|T`x@sBN>Ehb9%h8RyujsS#6sW@Q>8E
zGZxoRHctEVvmwPS3{$8D;A3stT3ZRt^2vt-xW8jetO$Cge-HH9JahPPyE-XP%hw}!
zFJG4VP<3&AX_AIVY9fI=eva={I}xc>nhSW-LbEccE!E_BVQqOzZYP=KSr^D|%M*(Q
ze~Sre2D9hwmmkJefShEM6A&u`EmuLgOL7M4>ixd%&39K@j@ei}8r=W+uo)8gelrZ4
zMB|QQTvb1Ne{giQVLGkV?!2e_4-7)Mr^0Vc=}6?#I8ZKvrUT>vpE{WLr@6II<QePC
zJEaTRd<Dlg?qkp9k(e)A_ggnvn>PBL6nI`ZsKh9+tv)8Q7e0He*|c$*(e|vV*(vMf
z#`SAlFkRA1HqIy74Jfk<y6o1bCw0f{tdJIes$t`tqwU#_W1l>%HionFei<K-?Pilx
z<4jMMh~osS6J`+)9kJe}B&pn160#6h(9Dbb+1jIFgr%BWp6Q+J&6M3*DJ+W|W`+^P
zjrv@DdSe>?V5bpBml~M3tTCJ`sV#Le-we+f$djlgBeZt?4`=z*{@kesQ4~}ipATNU
z`w8;Ha25<iLD%OdW@>R<o0*thTdLKT*8cuDLO&PHMDQ1+^vdYtC|!fP5}=nxf3zx+
zxT1>GZy`5>Io@S|jVFbJVF^!#==g;QVjbc2wG|D5#*t=3{h11u&wQ7$ip40dwwcv4
zRTJ0dD6=Xm)0NmQo4PZ3_oSnnWjAK()k?D>muEY3aZ;!0(%)v7*-*w8K&P!km(TMc
zs+1t6J}BQ73=amF$hrkeI!pbH6TWn*&MlENam~Kg_PSWy6Ec7D7C0rfBun2L!Erht
zw!-~<Hl$<HnIs=?$Tw@16z;#(5Fn+;NaA4CRKv;)>1~RY>QLkq+)R9ew(FhA$xfZL
zbLaUssr3D-PtNxV{9N?ybI;oT#aAkmlNJ1@FiQ_~i<r1&ioxfn6kvrae5UF-+rM(f
zx1YO`U~nMKM>x6y=D!*6>{nyy&b+Hq3x7QQ8i$bSZE8a~oQC+|ljQqZ*}K@jSl&Iy
z)X$w)*bqPSy7e3q{~X`l&r}@)47T?aX6!vD{QO8Z_jFW}<~z7j6NARD!3GB$((i<5
z6*SQ;XcG@LT;ajxovjVhzO4A_WK*+EN}HNz$X3Zw=!~W@$9H5+2j#c<c2&i@KC6E1
zvI_Ug)U>q*RUO|ZhOf;$%sk()e3!nJmW}2+R9}JHKCAk}wB090BJ|meqpcG}#VSvE
zm5(Eya1-l0X6xi`8C&Zy)cs=wy?#ZTh(fIr^~J<020Ki2rxm5<>P&{(>FbOcu4&5p
zj30W?6flU%a>|{uV?zB^i0noFv<YL!^y<Wsw~*x{Z&_&UESiMzG=YZVDhL)24HN7x
zjk@JdTvByV)X>pKXP~m_YANoNU1|gij1hBbXBKFX$ZLNM{H;N~ngH2jjw;Mv)k=Qz
z6eOgxfS3$q^5P^>HAP-9+{#6!vx^n?cgYVR;?B8qj8XRf>Y8_s(sSNgaqiT{sP5r$
zJmOxO@Ci)lDjYUrJPa$nlN37->}1pyLR+Hc5J4{^h64U0G?7Bc$j`HyJLwTcgXXU>
z%VIjgd1Kve)>!P$s0xrMQzcZCvM&7OnPRJ+qZky1;jIS-U5YYIV>2H11Ji@<L#IN+
z#V|14b3x9t3p1{6Vk{IGP8HKKWEf&K!^Vj!Rh1PM-u>XQpnm&9#d&v&BXn#@pTf1B
zlV~AvqX-vVBkpS^rYbb3Sjj4_ii2EHUVUyhVV!StIe=Jr!Qg(ov$je$p~ZCC@Tz2u
zSG+fwyF+hRG=9HT1!djREtwHx&D7vWrc{T2a>r0KT$!T^)d;;Um+UXQN?8pRPa&#o
z>QhuUWM;^^Qqe#Pz_e0pTB@#_xF|&tCIUNO52M3X#g%Dwv)B2Dwrptd2pA3WCQ+bD
z>eyGy=&rJ=$eIDqF^&9rF8_a8)u|dN@J-Jq%rzZmQA3p(6oZBk&hF}VT|pOHSp~h)
zRH><&x(@fq#HQu>;(RvdsC;;|_EjPwA4EE7x3*Ba&`*k7JHq)S!ko&L8DYXEE9mgI
zE|TiHaxq@%JqkfUI)j<8!^%^>rh}a&s{t5Xw4#_gg=Xo342WvLXi7Y4P>jna?uq;F
zC|0{!hQ7c=Hoy)i)vq(zRPmiq0D-J3#HWg?7>q?ST~IVlS)g*nm~VS?E9(4kdwb6a
z$GpCAdmEPo(a8x|7cSJ;rzK4gy0(h8rowNcbcyCB26rJ8o=BIQ0i+;-XQ6B?n{uxs
zpe$e}LdC)u9@>$k>k6W~n9!?%7-l?6$0${q0eeslm>dFIL^CG3$8KAlM1SvGAYEIv
zR0=hdsti4%2+?h5$3%7_yHr*%*)T{_a~##QND0&KP}Cm`8}wS0fy|-K1sa{+ns?p#
z{_#z)Ow+k=TPjO6uyp32Wv=o_z}uy}I4a=VrFTi+FMU9gHnV9WjxYj!c(1j^(^^B{
z-cq%`D#p)^j#i2J9*m(Vp-ryCwxT|6W(dDo?S$OdtcOSZLfnbsUb_?Vo5P&x=PJI!
zR&EDA)VZiilm%11Tiq^it3Rx+m)F&gGR3bfsL-poB4|@x*}O6rwcze9K39diyX2^B
zE+{0{lvXr{&Bm`33hUR@DX4v}p}VtZYCQRPwcD-6r>k+d8`aKO#+2G$)eYC1?Zwqv
zblRPTo7;VL%CLf{R$atr-sSkSuHH6HxEDoQlxqNENki&LD~VLkN)JH)|2Wi1%un(0
zadM3i!UFapjc2YFl5KL)@?_U2TpUn^bX5&=t!NRKal94dGf<sFp7**R#fgOZYa@O&
zk@41$kQuGSC!nTQ14|~F=f?CNZF^#QTZ^9}YPD9Lh*k0t%KS=PR!wGGfvk}(jh&x_
z2W)0`@eSdM@5B@Y4(EN1#0uV5RT+P5Smw|Sh-C(#(JBv4Pb%7UXJcYwb#-Dwy}e#l
zS69_){VqeBQg=vK4@|0-SL#!pX>HvsL-!UO+2L={@wurkoL45N-(})$rry+bd@NbV
zdhJy*NmfB)ba@^}X`auqX|dMPrsaNT6Q^6yzmixuQ`kFxytMNP*$woi4}S0yJ3poe
z?hmV#iaNdBH#aVsj;yTw(R)92Y2#r<b~1ga1Ns8S31ZZiwkBe>6RaIg<ZFo6&}(xI
z3bnOKS~QNzy=g9uh5ANi(Vq6FRs+*mJr>%&{P1sjUUR2X60wn$V9Byu+f$P*PmkT(
zq!fP+eb02@Xa&a6QzG*~Kbw4nFLHw=S*47_%-<k4%;VImi7V(p3i_|sOZ4PTtY2bx
zARKw0!<E9gRQgxkF`m4F#>OjPQ=3e=5Y$mYUY+br64<$PMrEr^9Spcn{GM=sJ|f=2
zCp)@4K!H!T{H1cX-mb%6H^6QE0(ZI(a9fy*SO(2u0h&~=)Ce2>e2o~_?6`K^!AUBi
z4aAtrD<HL3|Hl`UWb)t+s?)VQDJX05KsM;w={`n3zWT3&b1a<_IQj^qM|UZzOBWbj
zdx*QW41K?zet(`1i6bA4*7+w%L<v1N^0^0Ooqz3q%3ogi%oFq=9!F+Yc-FooP7eJ*
z(O={SdHilViNe<2O8Lh@hvPgd*Wr;*QjdJ}BsrYNe+TC;41MNZlw`Wc3XdV}X0u8d
zv(0PUNZ!)T(S+HSzN>-7bzZlw@a@QEI66DKGSJdJDaD3c;qfe5lDqvF#<Y}r7}tVf
zE1g5Aibp&YF757q{rvg!FN~!w$jAVpeUGT;2XOaTGIx?+5&R<PoRF?1vN|O_2&3;l
z?}L7O==c;FxvpLR{>{<%1(wFGTd+`f3D~yWzy-;khac{9q+8z{dy84P<7#qw{ATC6
zSCMTy<DF(`A(#p8L7ofmg7i_414rLU9VUj|kJ97xI;Jrnc25$=n2V&x5db)92>&;l
z+i$e22lLzG=-w4lDoRZ32kp<b+b_&thMyo85U&*c;OhMR<?K!0pY6SE%(<2M`S~3I
zL$I5DyPeY@Wpg1K)R{<{0$pQXT1KDBP2meT>IE1I@4tT7J0Z8iSVT{Cgm+*PnRSgj
z0%9M2CKwFPpGo$)D`3gqvq~lFt13G^*yUFSnCye!dm23ujBMV#SciXzM?a6|D88AR
z+-VKp-)X5M_p5({HUB0Xhlq7=XEk7CZ##XdIa-zYi%4?Eb=yxllY8EN-U?dg1?l}D
z{l5k6=Mzu{m!&UBUz2_dO5)JA6kVrnjk^A+QNLHbztGm;U)I)!+uaE=>UB3#nm5=E
z_qE6arI@tMJ>E2v7MhjB&bX#Zw08Sy_Ko-E@B03%^nPpKW&CV@eHBw?@8YJtEn4!6
zc?Gzb6E|OM)oBiLfxQs%jNkJ~Ci)clWSoyLwC-av!m^}|9PO~Ag(F-N77$Ksj<oc=
z4G+qqQR6{*fMKx7t*^j|!TCYj=!b63ea4FIe*d_2*f=%XI(>T^KMk9`!**Xuev6yk
zG3P|=)!6=NKfR*<#XUYB(r(53uZsxN<oZ5-T$0ycumtu!kJk$kss(=QX~J@LX6PhI
z@kPG7|6aH>baRG9lFQ}{k5WxuVh*29ZvSe2HU&=^zPE5Tucs5zOWYorfSzZWuQ;O@
z*V{S51?sUITiH&0<7dENN_S$OZMAq0HZTv`i}<-+H%%Xo#Lw*_GE5gg;)s}f@_Tyd
z$vuwmWo_fW2R}T1>&A{uk{@QWkX}e^vwgA%1mR$jr8^wjBRL~>v~l+wdgijAVMh?S
z^E!_RMv;0L{+fssWF@3X4iKs-WNMnX4?S75Emk-Vk~A;oH^f{=Cr9k>6(_iMYbR^)
zSC}bio;<mC=8PoeF;^=bmuDVoRyLgn_O3GY`U28v0d0uXM!z&3!Keve+&g)4$UZ`B
zp8a0ph419xJKOx#MTDI}`oer3R@fj?5tDUP3!NVxd^ca@@BMB5M*i+163$ieiFTga
zyO{G@%u}>Uoy%>DI~i%Jh5!)R;?W5bbH}8OF4=++<D0l{SYE2sq6(!|u<F2ot<qlv
z6K_hEe{EuxDz8wLP+QT=np~50xB;#+`&Yq0ti!#(;#2<J*Ipw_z%SzNBzQI&H=Ktj
zl7UmOsI}IrtUVu02C?Z7RaPgNt$j>$>4YX@_?yKH&!yCLop1NtAaI{zv}E088%I=4
zS66Mf=hqZdCAv;Y#&h!RAna{n6bde$@IXd23*?d8+{cOs8$VAR$Ad3;2D30GO4cT*
zt39K;H0~<QkMt=Zj#xcHbui*grRMisdsWqSedPfobY*SoSr-$i(~R$v()Xh$MUeZ6
z_XPxVb})kz&LV;!S^dB@52T7uF`vMo|3q?P(m_5zaPg-xf14s(IwqJ=WK%=zUPZ$o
zI-*Qa<rx*&K=eL&MX`w9241H0$Y)A7rmp_XI6-&7x>i5J=WPhC@^D9*DMD4H#xcHZ
zy=NM$Oj&)9YK*QhMg4-RpXR0W?35<Eq4B^9NN1#U+LBb1(P`;@(hs~284(_7c`CiO
zu@&D~4Lej&A5-Wgg0-_i(N!*|!z$zr%85Z$y$!}+R>JF_PepZ|irnUH=oL~T9h|Y~
z3$IVU<Q3!cX&Zy%G3*4FCAeS665=oMl|)Ub9~?p_jy_8w2O6u)WTG0A70p&<Py{F?
zGGkPg&A>k6Tc$w_)pTsnGov!xtWG*U++k>bd7?(BCOf5)MwwOy;TSdt)h$^Jv5WA%
zzBRCYs#f%pT1xuP3iQ`IILqBvg;xaMoJe@!$WO<39={4sX5&A;I1Hp~YM1cepp`L{
z=wpJRn8$PRwZFKBtIqh&E>2Y;peFC-`s*N?f8az!1_rfqWctLj9aA+x-M0KQc3`6C
z9Og@+gqSL|waTPnT5#{m8QZ|yG}~8HO__{zo8qFK!iep`uWXoAV3jaKRdK;kg({?K
z8o?dkodE&pp#r5ku)t<!m?o1lt7Irl2eC1xm2^hE5_*}+WN}F|WKyOw-lfPmW2X_l
zYF6~1si+uOzOt<71o1eDs({*}%J32yYOM#K)R>v{vlX()XSIZ<Coaut{Vgu4T&{fi
zo%ehlbP7#M1M#`I`1Vy2s+xya5@NZcmXu0ROJwx3EqRpbB~77*ZE9I9Q79vX3xz5n
zuO;aiTOwUiPlC3nxk>dEd~kT8V?;nlVUZ`59-Ulqm3Ip(R?r&HQ=f<bf)G?uU*dJX
zwjPTBz!=v)<8_eEjn)$=p?&ql)Ni_;kR|Fnvsh6oP-U$9EFXjnr$;IZv@{BGb!3f_
zrA4SHul}EDMS;zI@G-4sBsB$U8m}o>Re%`NP&Hmt;K#?BLM<ur+XA_wpV<c2X+$(y
zg5J{xt$;h4%P4OdHHBo^paYr;!qRN+f$9-7RvYE$?itgU16g$f^Eb@EQDNUVXV9VG
zUIfM?wa9C?5eB0TipjvSK4VreS!KoijOB<3biPf-f@$t^H7DJfj0H$l+NJL!3Y4#8
zNPx}*3Nf8N>i3fKK;<C38^|0-;~G#m13eY-o!@YLs0efhql!4aK&ha4nBKB*7lg+A
zBTf)FKLS-*oXLDN{}1U^81K=0s={eFg+YOv8<_KN9k+>>QA_N2z!Nvpwe&ZAsMFJ0
zzlh|Rv7%amS+%T+fssIqP@r(OVykM&j4f3OOn%#9ebHPz1;Y-I(^C|)TfuBEj2&H~
zaE2W<nc6iKrOWsYcN|S)MEN{(Rchd@{N{-E$#bD#xJS0r!5iWZA7P*y$<)v;hgwt?
z1USMe)4FP!E#~N7)E#ySQSJg%gA-`d<mW77w`;nt74&*7S<A|`>=2}tgj=E!zP{Dz
ziLlF6grZM%@YMOSm+o*){At@4E^)qHwo8QXUm(8C!{YuYL2*OCHF79rFWN5I7#xQo
zar}3O6TTxt<HR<@!_>Ham(I?oUYDKJ8XMB&L$KEpw^;0V)%IrM7Q3<4+vIbaVG8&)
zes(QEMy(1Iw(Ugl7m8e!b8atiaNVnk_J1)4V|`KZV2ZaXSSAN>o%!HdKbt<R=4&Wg
z!Kxh;ODl_YO<B248SOuVmJlzwpqyD!<zRnlEesv~*j`e4ZC+X=%=d>%?sB#V+ub%Q
zpnAqnYO9~+>kRK7SE9M;on5*CzJ)_%>NGhFn~D&mXr~UBZd(wUG%01QJc0x?HM8(=
zZ+maJd_OPb;|PvlI6y}2yr};!pY!C$cX0XS<t_S%HR(3#jx?q|%ICcHm1JksN*>=r
zzq7w=*AF;<xj^`Ku_OgwV;bsE`Qkj^5upFH^sw{}^aYw15U2YQY*U{0O|<mW!V`tL
zAM<wK6JybBzb?pYi^6{C&^s<Ag@c=NK?q$T7bbpwq2M+f<Pzy7h`0%8P11KkpO5vu
zBlU2c?Ps04pc`GDQn{d^fwdNtr?8`m&L2A@ZKQt+hpX^4J}9WG$q2Bj{_75--=VA{
zt6x+%@~&w&848R9>e;~_scxH(;<)grX+L?Ekh97~!M-NuF_QTh=_L`5nv_BU8-hVq
zMZ9DYstnKTV5ceiALFNBU;l0fj{KsfpHM-6h9Q#?T|KE;^LhPA=4#+u7JRGAzm`V+
zw~{*E@5SQF$>K*#X@oeb!eL6f8QS!Bkq5<9^napbi{$ITJpYyRPaOE*74gw|zRE*f
zMR7sx+|jJ^j^a24pI{B@@}h`18-b8O$8<8>yKukS+Hx(aKISy~{#!|Hsx4o`OZY+4
z)OTk<hnnfl?BW4@7cSQ@M)9wcxi1%e69OL0j*hSzs?~8*V2&cr&YH_34UZ*WOnVb6
z{$!<e_VK&QLDlO%?M<3gHUeFy<@M#jS*VE^6%v+xTADv{WZq+?@*i2;`;EvYwbosm
z;UO$J-yDl27{@n$Qyh!woW$b=6|dHHMNtoqP(vx>sj_0wBUULD4XM?ZgluqKUW7ng
z3hJ0f#!{GX^D6<|1Eh%<FX{KWZQfI-iEw>(NeoqhL+_0KCQv-Rq2K<e9%0U8sS*U0
z(xk&4X;EFFq1EJgITk;fnL#M8mc@~DnpTKMm#s)Mt*N?YsxjP!5IcqK^bA7vw8DIw
z(7OV*R?O|hX_><$B|#>g_DMO~^NuVwD}=cIY-?kqHS4>CRGN!NNZ?yqty7(tnwgzz
zHs@w%rY5SLn!xOoUwakay9)1h_)K4z#Hn<67-ar(?nsh;bZ>QL(zOZE&4p#(*=WM2
zWUwoa9Pugjj~r<PL<S<7|KX+C*?PSi2Y#iZ_%AnGGqZL4vSGN4xrTu>)n{i~f{yPY
zJ~_17GDpMV&Tqn#rO6>Np%nUjVhzKYz$8pvyF2I;TAzRP(fO`QC#Q*`Dr9DoDz|~v
z*t@+-Y4!FV=)4;(er2jITeNj~8~#)8d>Y}7?_XrW;#{K>G8T3kbBm8rIQzakp<Cm*
z3F&Kp1$6a*u7<S4;}lRbni^K(77`=`!wCX|K}fB~(<wvI^a~`W=p(1ld|l`0eJK{n
zp5resn3_>>O|y4AqBJ_*GflT*Xei%jYqJ|0v$f7$cYV0P0s@w^GYXxY;U}jjsd@%a
z9sdumuT&~tsS>qkry6pFv5MT7nr%mwl843cvvYHEt?ugTujhPX>afN8@%ZUGaXc8V
z7d3(M#JLth0`U#i5zZ7?w6;4rvwDLShSTl`akXBboxS18BWMNF67+>>XdPOtr6lsG
zS~wI_3qa$q&eiAZYt88reA%tfcUD{Vdb2w-<vG-!neULs%3ynUwlO_juT?vfuIwxh
zw%^uhOixW7ZP#o?Vo5t(gZ5GfyXO?geH2B{ztr#WsjRI;4qP^TIs@dU#R&3<P_0n@
z?L?mc;B0d$4o^-ZVOG#kXmzr5q~U4W#PkEMdt1E<t+>2BVdy}qp;KC`L3bNe*G%Y^
zOTqNo`ckJ=jV%uepFX~(8>gFwQaw_ene806%$1eS+-w|rFjfO6^-$B-?tn2fR-<Yu
ztWD15<D~+1+~L`YqyRzT1@X}2FnlFF&_v_oN7x)6HzoqvcBiKun^Y!#Op8p_sMd7N
zjLvwA4I)?K_4T+Slg1)x8KtRG1^$DdC7o4VAO3SI>;v9ZG%?kH|E4CQDNag)!@DX~
zkQP269Az`dI1}_m#x5~ZC-eczKE+@Q`;08f^0v%AsnDkwdx|QbWHQ!`{9UyJ-WK=z
zT{8BSAGz(GiT90p@BeoGBfa5wPagDc?clfD6Zd?0{QKRHoabLC;+kW>kN%5dYB-H&
z;lM=CGZS79t^ZQGLl!8#LpGGpD29ATzFWj1r%3aBsFnH$yp?E|jD0Wu=byQQsQ&nu
z=kHdeDgN%x<|%UU+b>=m`~EvVbH@Q6Kwkyk)|bc<v_6b_TjjNt$7CZ6YO|vy^GQMv
z2Js^ra+d=ocf$C8ux89j%hZ34vTcOgXKhtcCJ4Qve&y_sI|O>eNyS&l$2Vd!Yg&^g
z=v`MR(Pc}Skm)v~KUewjvtMxu+%u`b-QU@migSAfea1^t1ah5~<^^r@S!R}XXgnb_
zn|`a^=|QK^3p*Qq*lgvSURYdF-MhoJ9{+%<eDqV4eD*T=?58OE=m#|8@jGYd3ahM}
zWdbUVVFrd#o`;_{{+dS6E|)X9R$$~hA>A*Xmp&+cQhHAM#l-ecX6J-Cl+7abQu_em
zt~U8vDb`{QO<t`7-CMk=^@feAcv!XD+UHF%a{uV{(dRFr=FgQlF$=<7ZMM+@sF%$$
z{)wOWVEmFivdQ=V*u;6;?4;Od*-5et3Z$hEEd8=>i5POeI{=FfjGWN)P0P2!5H_$I
z4){9=fAc&>4^h5Z%)a>c!VQbbRVr6=dayY1+(;#EnWJU<Zt0TrCF$p+|3bPV{krs<
zsD+J6JRWxK+UKtlkJ!QcE&1#CYZd)>3K(~omMK-kw|J&{o~^#ODLnH++{cVk=#=#Q
zP`SK66S!Lte0K8NV@l1<54O1ra8dz;+bueewc9)8i-B+j8w}>RU)+9iY!S(Xjwykp
zTn71P#+Ko~<yJ*nDbnh2ZF~FbRZh-}<x7|GzbvkNLfV!d$h;#HNBwm9Zkl9Ju^B{c
zKk5Gx@Ist2<oCtcH>TIyiTN2!L=66mCW0&-QQ?%b8kTfRraF}^y%ch!CS5Jp7I4kO
z#YLZmT3J}A!zTkp34&Rjd4|hFnGBEVv$*=5$FfgIeb8<0m)?n1>jAW1pXXKwbmSgX
zH;rEmg+g{J3AmmS@9B=RG?wuz+B4b!S!7JnP1`NIwy7(iumk(S81BZfjB>aZc#h)*
z9~0~{g{f2_$Bq#NbA2-|kkNAO7Z3wCvehXRLo%T)wEDCV6ER>4gH&9+T+{2Xt6b}2
ziuC{<)$;ygEf4YZ*c(w-lc#zkBDopvkg?CDJFlzeCft>W4hwa7nM5!iACd0BtR6#o
z-1-NAUGN2fHutP4VD4s~ww;)Rx%0`1@0O>_<>}T;sWdYgAihNa)QTV&p!s3Y4?c;I
z6rcPg0%I)mL2Uc6>Oxg^p0rEJgRSZEF0Z|OrapQ8zf=_d%qRKt%O~+#)p(4=i`|NE
zKbWn#M4(^vrE`f+C<qSo%8E4hM;VBRJgo|ln3;=4TG79jGT`_b9^Y<(m))ENusSoo
z6G;$1VZwe-0&w@tZjqg@<-)6o2a6&omfr-=cz5Rzm)9$nc&L~)H;aNr!<2mO`o;3C
zoUFAPa}*DelFdlgG&$qwfMWSnE{}5ys;sy;XJf$#!;0%<qRj42G}cVviqR{wOOnai
z44sgpuJo_Z(T@Q~OPq~m&ev`R!&u^wEdT5ESz6*LsZK6)&UT+kpb6i@(Nw;4U6%Hh
z(W%#0Mx8>JbK}Kh?O;Vi?3tS=n1NVh7%C6+Z@lAxERRT1=f(@_yW_|^eMtH2qwZVR
zFO(ui<P!JGOVzkkwJQppmq&W-S(Ubttox`+Yn-Gt7+@~K&^MrJ{S02>6G)dCtqCcv
z%km}4GcRWI_xRm;?t3uwZk|YJ3<cU|pL-}IhrgJ7@ElWMsCYCZ<oOF1P762YIQt|U
z42G%s{f|8T^hbbHmEAkq$In6SLxlA#6c>89IDtyv#6=P=3*jz#^Zk3ZPF0QCPzpBA
z=GJ9>aQ%vUnHMO3`Uj_YH@Lw)v^RJWPRXwv(#4PT7Z2WKe{lmsxQT9KKlg`q9R4Bf
z=ea4<zT<Z3o*V8xlzn4Z;zHQY)^`u-M~F5wc!||Gz6^F)XL5XK+!AP3Oc@AtxuH6V
z<#s3LG`NY9y8+G)D>pKGJ=U&QXb&I4TGIYoyv*hgDXk2&XM9Oq;;GvX5cb6fF;d~d
zB>sDqe{WrSR=RwUEJdwv^R=v)gPf|Orr^oA^Twb#YHG6nJz4^sgMVz4p^YepWf_Y6
zUb}SZ?N(`y$fikPqfo_UlpulzSS)C2-~m4^6Jp}V68C_KtRi$JZe&^9W4Trd!?*X9
z9~W9lp_*DYWG`^3X@b};K1Qs{>-rAaNAG<+Zansdzpp5ZMd?G*58qsWlpyw}7`Rc2
zfk{%>TQvp1Bq@AMJ4L@FDY)%cKNc-ueXDa=Blwfo!|a&CS>WuK$YD*V`o@S`-4Z2j
zr~Q8`=^r<)U-LWZqJC$yGc;{KwyOU1rHdkY-bG`gBHQlFjBCr4iSgAdyqL5`uP}8l
z73p)NHJ)o@=)?Ve7D;DwpDB{{5&gX(N}_`0fpj`~#~~6??_Mbovs*AU((*`plG~rb
zO7YBY)-SJcc^&~>;qF8FFW!+%30U}kI#F5(_Ci-nG=tI0aeNo|_Ue>um>2BQOMb-t
zI|gXzMjj?vKWLPZ=VGO7&V#s>$P4m_!oZ8l=php5X71a-+u~{I-kWQS8$6><27b>k
zc0?>2+VY3Btlu#8XeQzz`npMWQf98#INr(WFYvP?SmA5W3ipriCiQ$7`uUr%BHFGT
z_qFJxjOmkQ1oAn|fcTegqe_hjZg-k7dzam6)(o@OY)t6-VdH*i&}zf&-EOkD*>nt}
z)`Vwa+Yogt;X#osVNJ}BA?+p%3BnGdvwcYUyzd}7fBi74Vyr$rJZF;79A?lK2no9N
zH=$4GI?WIAEN7rkDyUAPEnAnrdYCqStNQc}2;oiDr@wT4{>Rwj55CU%k_;)fFu>qq
zHFOy_ZksE%z;+cFt^ybq_iw^df2S?*T=N&n4!&rAS-$mF%JN`tZ=yV}$L<Z5>VfR$
zI(Lh&bKcyz`n5wx(xG{}gNILC=jLP14D@fu_tU}Ai_xKFTuFq(Mew6Q)DNHo6!mSB
zXa;4<QvsZ7B0Q1WOs?$NKIQgIO;vvYcj$ZUlA_#4E!S0^(iEjBB-_O1Kn3v>n*;jm
zq_oz-t{kMR!>DO;FMXbiwAgkJBl#4`y(T{hl-i7%%s$64pn*(m#J_SN10Ox7usuY5
zRaKjr6rOtqvz=fqpeXFe4rb-?bD{*KqMv;L`pX99>f)2P8R1+b$){j?83!Gn&#1v2
zAFuA|Ryp+d(98V4D_i;=cR7Va9>^h5dtbBkZWp~dyItLS7Bi^gaTxmDdEp|OVe{eq
z6H>A@E%Vvq4A*#C?yIYE0?|4x0)JNJJlwguwK@Uqps{^syHRc|gd@ju(JbN{|FtdG
zyOT&#W4i%=VSCb49DWX@m7MELm``|!Y#ilSD$5dq{!o`p9&3C7C|v9BkZ?0x3)fn`
z%^~o_Es>xeMrzVQSV;|`1Q9Y2Y+fGlwgx6*tim1#5JcZV3|vBAo;7rb+Ukc}JfLk2
z%*3B1gyhTePqyYQJM_z~kj!{2bSulgy0}xLvD@$}5wW$oiDuifD&h`bsd$y8(#Rcq
zxQIi3u7IjoQmpAx+%{=_b2@6d6S_@|Qpxbz(+sT7`M_OXb?}~Y+%{N!YbtDcHN%q4
zvakE?8Is(zvLY2R7ImKeEpwOW5LsxHg`(!1AeSc*MtAZ-S(Wcvrz*Wuz#zRM;KQCE
z^sx<@$d7RS?p)qSuW+1(CiYyT5h{5CGX5plX#KSQKwT1X65iD`9^!2J<rUrW938Br
zYk2(yu2=@>`g&##D7dQun+G&U*BuSoQIQUe*+#EIF~uSldPY{nL!>Vy$Ky-hm(tG-
zl4$VAD-g40>ED7c!S~=}cPHe)2AbBh89B>WK-Fa#9lA6Pbet}DZ|6?iTfHH~992NS
zB{M)sY(JUypW&Vv=WK2+{$AvN6J5|FUwmJ;uK1B=S^q}2)_}<MyMWxLrC&x%)(1}T
zCTR1?&r4vG_?LNK(1z_YCC$6Qa-F}x74)Rvm##s-gTA3qGk~U~xo=_a*cOj0LN{|Q
zO|H1nKhNmSIjZ_qU!~`EDEqwOe8@50B5Sg5{)Oqo?^SjdJ5S$c%2`a{`+bz%23@;#
z8{<Vr3*T+QC_nbgaf>HfZ^cQB{f$s^&pA5LoS^4amHq`)|NHCHwTu1UQK~3(Z-hdc
z<%vhR;Of*3N+uMJ6WML+#jA&ljk2ec>rXQ!J56fz?v(1gDS}Z^b_^3u9%FpA9WkGn
zs-^U!-BqB!#a9uZ8qdRx$oRTE&7=qjeetH7Es*gXSH$hAczBrq^QAB>ttF*)eM#6y
zaY_0pw(WFIdV<7K1ZBD@y&JPH23nYn7YXu06FVzttI;|~Gr<#Zx+hk}Q}3rd$wIsp
zi%<+~<{2;`2EK}93jLuc-(ITKIjBXwR+1gPIu+}VOy!B&>Ie`!d3;iq<9Vh+TUM)t
z;7EAd@<|X7-{K26vJ5>k^E?LK_4E8jUm@BF!femhb=a_zx-Kh<tUn~9>kC3Z%Y@!X
zKzTvofCghK3`S!Y11ntg3s!<)ax$PqK>%9Vn=0I!%?m?I{f-o4hLpajIegNNXGuW?
z3Z&7`Q4u$`wpI4&BdA7gW2$RVppJZ+soxM2)q6}?pcZ=PVOncYkrtE(IqoFiVB32I
zeSM`rhrYDIV=7`z3|%pTN4ql5xC4XoNtyfp_64$L=lwG?{^IoX;<0IGes0n*Cg<jz
z>B4?lT=su|Zr+<-nx1dD{5)KBTYR72>8{P;RWH}c(C2RoF+_*0Wldej#oAU=OlO`V
zS%!&RN4TvldkSH`UQt|v*bX%T&yo-LzwwxfdGczfv7oex6s?98zFh+upnrBT7nC+c
z3XhkDqUfR1xe5fOL;ZLNmD?6E_L;a!MfVv|Jh>teW(sI5L4iVOm4vd>HPmX&&x~Nm
zNWTL*Ynkg8DHSp1wGx6!>G-dAWD9itK$|o}Rb~b=<w>vN!Qby{h78s9ZL@^vDpU&h
z>(FvT&y$ApJkD|i&QTp>AK&RJN*@3EtE0RUQ34_lB?n{qyF(0qFar$A%K&54j!HWi
zy)fj&xG&PolMg6}u>oiAI-nei&@n*{RP)0J6oe>;E7D&|KKef&TpRYuBKh6{C7jCN
zm(ECU;}{^AQxc;yg!30h7iDfXQ5+>_c_fF(u-V`0r}4C#9KXAryMcz86RNhEd#EO1
zLA_mtRRUcvhuAB4uEKaWP1v{|?u6xW*BqZIy9h0+4NwX4?;W64=iN8pV%Ml@DT;+z
zqGa|EaQYBfc5pOul7UgTY3cWJ(+OS^z|Xy0O@%iX^i9+YHhA{l8)9gvcYNU>A}@)^
z$`=p%{s8PZZp2|fndb(nNyk7(d00Bn<FHs|nysH?8><Xs?c!;Q@DeWN;~4FLg?^It
zGJs)i$XrU&>SB@YEiO#Qwk+H6^uoD(chPYc13&aIX93Z@(62mFDI7MZ&m(S^X+9Rt
zh2h*H#*&3j)q%foYKkn6zE12;idO~ahTr^Fekj*WY>{SInOmHzgssI^IO$X;s?OBn
zRH2WkaS!O5%HrG~LRds|;5vDop#XlCPj4TR;(3ol3xc-2>qQx6Y=TjBN_s2UtDh*u
z>E^z>dGz8C<-H-Q(>Vbxnd47dAA*zh;^A|=1>(68ufN`fGzFHxSLPUW&!(15VCpX*
zFE8k35MhQ}gD2<W4eUBcS4AFYeC=i7)AwT1xd!E=BeS$1@S<HaeHf!lw|V|sL-@tu
z0cajZ7UZd9vcoy4hd_h<7)Pv|BFVI#NYIwd@4VS!1Fa>3Hbmz^YnjIh-bi`JElTZ0
zg`MT0f^AuD^Ps}B(1G2u;0}=6)`@hUcxZC(|ATVd9VfS2uW-5COt9e7F%M_ja7b(S
z-)u=;x@fY+Ma3jLw<sS{yR4|RR->x2ck9itAeRzYP4~bka|!*@dSQPHT3#lfH(fmC
zhN8>{LPp!SD5AZ~9||Z9K9maTD{q>dhC3?gx!!C!{Zm=qk#))!p`taaf8iEI^^drj
zM)|r_v`#PHvZU04oDkRJonVJw<{kz`ixL!-WZ`j!h;9o}rQT%O{R-8gM}dAtS23nL
z96o&X7A5(IH17rBbs8Nx{@|@wLM@XmYNw~A_evj^o|Aq#8#mL{u)+c-7xpR<la6WJ
zRt51V+hHC?(0tR|Hw+XOaB0FHvw%@?fX>L&QZ^EmWoXmBb*%+)_ACmWfQ7(ecv75E
z6<OD0e(D0($)Q$?*^57W1IHN~tjas3$1s0S|8G*%7q~~TN<jha-fBlGES4|7q5UVW
z>z$LHMnANEfuDV`9Df5<awN0qE$c4d8hV?K@4xT{$WG5>r-LPg9s=Jdy<hsY^rgbO
zj~3&Aov$AMCv5A_83<Htlxi(XecqN`Z>re3+sWouoA_U-x^C-wtJIQ-vB*Ve+Yi0b
zExIUwd!PIToK!MrRh06<2XD6gcXJ7k>sa5oF7^Twz;cT$!3X0KPmUgYBW=W@buBb4
zoCMqF`SE5;_HV&}9)`@KEaG4=VaQ=(B6_PD>p!CK7Dg?Wy)k+FGq);5mdOxTiUx?9
z7re<*x>aGM_}#bEW^=x`q<f`Z=|ef+trTZ*1Wc=y#ZV8O(+*&JpLkNl#4HN0sqXH1
zd~J7+3uBK@B%v^H?N%IEL3HeTk}<PgWDZ61O+WPcO17nc6W@Uo$X7tfC?+quXr%YT
zdv=B-)=h+kf=kwhm&vLWf`kwsAM+_Bxd1j%W1J>2dd?I;AG!$lSrM5%p}}ox>I=T(
z_|a_WIiAO=QvrEFWuMT?`X`u5Kl-FB|B&bXkSss>QA#u?r0*g+jtn|^7u;KxBk!kX
zRYN|>*h$$?XDR6gM+hZdP3ywXk)TJia@3@2S>HRXP2FIh+s05Ns4qM;DMKL}vzLTj
z98Z6EmzMlUL-aV(L^4a<E`GIhV+36yYD$rM?+uZ}Q$HL+)2+(rdSu+Jj9SUsN_PlZ
zG@TQ@VZ;e>ojWzXh?|Q7I1Jax#Sd<-WG9Q@|MLlw*mjauh9d&(@&1VSOu1fKr1ilR
z4?SNA+B^h(VKjfhXf~5qIP^f;SUDS)bK(d2?q0roz&cUNP_nt4Z8%ft^MUKDE$Pps
z0Q1V>x+20`^)Txq{@-#*CtXN8Q=8W&y35v&*XuzLl!KtA&$h)1+PT@Ii*6<27vR5A
z*cGcunQzBmz<9zVq%emjynib-@4HFbJhADt5TOon^QHyhcrtuvn@)=_z^7{L%v#s^
z1kOkBOa|yui5tGhOsBgxGYDtZ_(<CT>&>Ua(URXLJ8t=ts;t$0kyB)^w%`OYCPcv5
z61*k-N8-R)czAMQ&MpOtU%2E2idWhe{HJ+DrK@9dX+1CfSJLlGFG>GW`nS?+#Jiq<
zlo&=v(K|+QX0vQI2VcJ~8puAhki(uIxqjI5H}QU4cIo6RwAXU{kUz>mI^rSZ4np>y
zezJ)8lI)$qO`U_1<TolAN=Pq%pfT>`2EK?tKAU5RaFCG;9C8Ge7XLi9vbSg7Y!yC~
z?ofKj$@!e<B1B2{(g*VEN_LVSa`K@uCuJNTMeebcp?dm+bW!>d=?l^?OTQ`of%MPd
z+W$=mDUpUG%^$MHtqFT(7|`;@d%6=O9)7`QXE+U9XR~#aedp-)8#pcw9{aA3=D18O
zxYY4_bL?V%d0%aWz1_TY+ey`eO^}~GPYPvKs0;IX1$;4oU3=t)>z&=szE=R@nNO?o
z1)08>C$|ieYQ5`4A4y?!WVtIBs_g#F{Egn(-6xiV_GGf}<muv#>GzT`S^5r{;22?Z
z(h+Gn8Iwi$6J`qKb49`jtAqenJjGh}?;KMv3Kf|DxEGWENDXvjt|7nrSGjskG^V^f
zAIt~say#*TdF?f#^R;&#11!AecqWWwnu>dq_2UYMO7s#;oOKS7v(^z{g<E(k#M}?)
zbk+e~3ol3SlGR<6Zp%Q((lTh}%S%9ye2%e-#VGWJidj8VH5KR<DYGh=L_BuOrtVDM
zJ?ZFX*^R-{X{+jGrW2wOVame}(b;?83RPu%31L*%DZ@lUvPuc4+)OhaOIMM5Rks{d
zV?;@OeKTHXL_Ed;+@1nMppU+UZxqkb$~#up9&z3P&Y7lt=udB+`!vUtCYjieavilH
z^7J)CZ_u@UqWboWU9!w`@e!?LKmWY#|KoyxGCpre6|QZixf6uOIQ*_xn2K}K!!Nw1
zvY+?~rUu2=CS`)YA06og&o3?3n38aZ)a#^?Ij1D6D_SS-vC)MS{dZi_>R}Q$;jJQk
z8MzVsi(=&`K;??#vx(HW7g5^^DE&k34bH9VbK+;WcyJCcQ||sJ!`=25^LWQEf-V(+
zR(3bfvWV!)!t6v~ZINOz&r6)}7qkoMuu@ca;k?dGX^z3Pwb3c)J(vpt(cV{)KNWSS
zu;mtmJl_ER5QOpi48l>);5^AV%<EPe?8NB=TQW7B(d7}D6)C^4seF;K=M?2RzHPO#
zy?3jwN<3K}P>T-4&rUhtFZY<pC7syvWhv%3mTe)-;zw|KP3!dIFY_1o`{^fe?+YLQ
z_{Y<)Oiksl=ll1ZzWw&xr(1ku1`)sy{7Q`9iTZFf_G^RQ$3NaX!x+2XS9c~(pPp#R
za=^cywzD*E<UVV1b406&NAkC|vw5V{FVKH6FHfT7CSf^`1K3U`Xhe?JVoU_5;V^H(
zQ5$irUh$&Hb$WwT6ltho8I0zs>x3!tww!)Bn@of6@kH3#3nD#g9)j`G)qI|ANY?xt
z`)!-l6;94D4KsCr_X@uI1tHl3z98M=H=chQzwzt^eBXrwzoBLRofh|OJ}TWV-HUiz
z{o*uUaX&?aLj-fJaG^gc<kCJN^JMSmhc_4U12a4Md|ntlwXL?p1$O0{P*%3Z&zWQi
zV@dQYFJzVW<yXI#qxctNrbc-zc$^iz&$QxnPh50M8!dp+z5+ON6FjCxbyZ~d<@wjY
zK88niAB*0#VHOKKirkZ%VBy?_T3J6w?_1vr4=lWqRvM<`a$>`#IeM?{<86qiXR{)J
zhN@cHHK^VJ6kX!<zA;>2SMyR1aOzsKw!f)cFG6GgW}#m0<EKzI;eiD;I+K9>H8{Q}
zH()TOC9=AGeSSWk^+3?=fW(0ztsOu)6&X@T_`zb`$4}9Zx+YCwoYC0)$VhW3D78xY
zyunZ+WQCpC<@wHGXE^ooHOz5`SsPycUg8^@<Kkkz?HInf#4|HL$a8C_xveoRHqK<X
z;&J(|N6*21eLPpp8w~gD4pO)TV(M3R6O#&BY#Y03k@5PQT)hxFMs05|;QckGzlNXr
z-QBXKrO#cRpDzMKwA%qsm@VFv`p$^D5l9nKgXc~eiA->b*4gT7tEAJ9;ywZrbLb9?
zYmCPPDg25#7l#wp$KpA2)0+5u@Gp$#esaRvg!3D5qp{j(<k0|6io4#CJQMQUH2#SB
zn|SMLL&|g;i}a+BM{u-pv4GnISm2lk4sD~f3b(RwqGr2Ix`VsgR$Y}nGznynDy1^n
zPL?DYHIxm-po}Rmv`+-ncicCjstx$l6n`;#o~4Ty%uteg=aZw7D@9QLTiAm7aghLy
z-k@sZ=;#PEp><4U$G$4ljFdlkKD=#b$#Z{cKS}Et0DKj_9bfRgugVEk4?G{b-qOy_
zg(7wT4+h6vMbDpPtV22xD;N<;)etl5!Pk2{xR-C@G+!of<YEK17PqNRkytPSKdbN~
zbbVnFEGnezb768~hY1!Zs9d_wW_rnGu2VkRopbE6%SpcNBxu5%$gA{Ucz(R-p|CMt
zCF4WKL%CEpHGeri?UZbL?&#676YZdgDwJ|*z-FjUtnoZ-g*wsV(jP4X=(5B{6xS)N
z-yB!A$I<pz$xvN6kiE3vE>3kLJJgx|2rwJkQMXuw_A`}~-T+-dqQCC!=5|Fl5SYKR
zy*VpMBA=rK))&+&DZo``p7_+|25<|3$xo6$p)zTnQWdsGRW;uwOJHsOprt%7gCUyi
zh`EI1ea!>ksPR*MessXwKf9@@a_#+mJMK%*>&ja{1phJ~{yls%&d_2<zuJFVq4$CG
z8fR67UPeC{(mV<8Ci9f(>?uq6I+Yo`TT_zxkmP-H2fp#t5Elo${?y_rgb94?w4#v4
z{%?KhCv@fROK>j3m|KQ^uqz!0tp+nxigcwSO(iB`*n+Af;|he<!fb!MLyy<ifjsQJ
z#lRyD&`O*lQ1Q|;pecwERKy8rYs1fS>oVKH4t3U>h^kS*Cm5*R+JS!4t5XY(2w9}E
zOJr)v@(Ds_z&<o7Y-MsnCW|}@9Pc1t+VBloZfrL?bADLzT%F0PW0<ni+?ay1KfM9B
zoPkfkd;72_%V*$3XMs@I?JUS=i0q0_CO28sl!=wb3B5`%8XWt=B*_g=)9GRNk-A>C
z`)DS|eGd5`T<xQf;|yw#)k=-7-!gPovf;h1qH0ZRCbB##&UvP*W~cbQzga7-HvFbJ
zF&U{Aqb}$MEi0O_JiBY)IhVSX0G@zTuI)_V%~@YDA$?1#^XTAo-Tb87X*FAsNRrCu
zDz^GfSB`t_jXVfb)i}&;>!Zt<DRSGb`b_btte9nUn<wd-QjhVImE=?eXAFMxw;J7s
z6HPisrDR6Hgg$ft7o{9)Y6bL}0Q8wY*Jp691!o-*Qhl=zN>(3d9z@r@xdGi(FPWWe
z^T4UK%~qs^jzJ{Dag5hsdwwOdUE;*9YdtQ@wqaW<Q?YO9>IxKPtUHeW>Z`b||LBi?
z*{Qg0#rg7m8fa|3clQI+GrD31`t7HXwA8;=OV)Z3kM<omyx!Z%9G1}|w7<0;AABNp
zY2H=X73^sb*hFHf&ZII_0Q9Ln;21G6@Hk0X_TbRZG_KOgvWY`1jpzg84>^)ZL?wSn
zhZYEw_k$EJNFSDdMEXhTE7I4c-<JMRdMVRI1@}Wj(*7o|q#k0o{I=Y+YoM?faQ6ki
zTR3}#Bd(90eF=A88a=y*_wL=;+3{aXvVR=#HPT|mba{zSUV>@axS4C*(d-5u#H#WC
zulgZ+!yB$2Y4Kmn3*=f}q{C;^A{{=P25aBQ+3{c7O(s+i|J;9)3X<R-X8tlG{@Di}
zoKx91kzd%%=iIdOvt#?Vyw4`d_wJFNke<HbK3fwBfzcW5q;TjB5wAfN9&CIDywh3}
zhMz~)dSPp`FYfQRHrKSxe(#VT{3P_MOks#cW1^I-l4&tbA&jaPoMOxh=E|61!MEYR
zu{{1gohg)3P1j775KYl^LzUrG>@m&IO^w;6UdHmEM>@KgPaPe{?vXwq{nU+*V_Qk)
zNley+i&ZdQA<ipkGzb6-7y9uVa?2s2+aNljgqcjH2K*8gMk@GI3Fc%6rPgGy_eR_@
zR_arrXEF_8HY^5eB&8^Gi^2(woqnCyuvm;lKYNwyK(o@av<+IuebVDlZ}&yFhOMwS
zRBG`7laKr67@?+yV_w$A#v~F7l-s$ejWH!~EzhxRmdl0bUO_Z3g>Q_&5Yjp!CLRaL
zPjVSHLAQ8!UN?b%Q8&B5|LGfVuc>=gyl>}_+O)0l;Cy}PXKcy40>@x5^dtH;p!zSz
z_G<@I+|4_)ahn03-GomUf3kTX=^D=+KRqUM{BYrBS>d@CA8at^JBG<$9H4SONZb@)
zldZR8?_NFh?V}Eb`@Z&>Y?ki8?*@s}1Z*TNS?`*F+J96!Aq}LrLGAy6%=%4Y&|5=M
z=3y&YFi#Rs)6JGvga$y;HG%yW_BK1cY`K^Z=phee-%P!fq3&-p)v)~_nTzVybz3(~
z-!{wbwjqk)_j!_aV|SaQ0@CxoZNL*(44gaVSy|W_Je9b<28yb~fJi;#vhDBe<}<^A
zfq>q&v9}k9Me?rg@A-Be{_WaaozjQoaa{U7>D|y{UzDCrR^yUI+c-@{lIMU;EhCYT
z3=DPPww=PGZ}zA6WtQFI;XiRt3ww!lwNQ=t0=62)n6)a4^Y1dnw0+Y6!PI#wA`nyn
zd;k;ec9O|?&-U9Wjy>GWm+W%A&HIVdM6mj~-DkPWH2Q{sur3=ORZ&bJE#aYy5<LUW
zVCU(4JphHpWy3cpy*iij0uXRfVHYl5<g{E&Y~2Nr+Zm2#$Z-nMizVjBb0Qg|+T}H-
zqir@5So09aay836iCz>u#t#J}P|jN+sB>m8;aav)qN<`0;s%X|4K_HHy9zOt1=nsg
z0+$d)QE3TGPIn?OmD?4QC|#K%yH9E>PGA|XJz+Stsc6RW{gN8Fea8vq1_S$Ty6;A6
z$@iU^XsYHI6SiwufwQ7L$&_EEa6>~5r8K7n@`!UZqHr~ArF@=7OY&r<nFIL@J8~zO
z+u{taL(S1#jx#R!<BmQ_y3&ZJO<6_m+gT<;p1_o&{9!I+#6^7`0ZzsU1bTa2YldY4
z1+IXOVwi!gktU3^Iz4_I#nI6004kA270@F|Ic#bMK@_Di@)M2f0KvIj?{#&TmMpJY
znT%}BaxBNym19*|Q)PK+38fYvQ*;;3TbdnBR;r#=qVSA(cD%gPF}Y3Jk|jk_Lt2q`
zcxEK@NyRKzqDf!_5h)`uA#zL96rMDqnC&Xx?`h%|m&97(9sL|g0B6bj+Q{)YL|%bM
z(WbBBT=#p?Qm<kVv4i44jhp=eZkpV`idizP;VoV9V2e@*6r8H~nZ@jJI(<eQeJ^_f
zPrh8m$M6$1{McFXy!boKWQ+07uV$Y*0dH_9yqW*ZQXLc1JtK|^m`_yF5KU(u`jQU*
zOS4(G_}CJMEm9i{zBw4|@@0{W(QHI64Ys$pFY|?!g}pr-G4>>>h$B>%5Wjah%a)yn
zh#<0NeBg%=#AyJ-)Sz|zUHd*sYOUjQ+uMAiYa4ESaX$&^_c8bD6i14Y`k?>k(L9|>
zv$NL5`OT)lO$4=TIspuCZ6-OqlT?A2TwE2GzU5PqLV_{S>Onp9tzb~Ioy(25(yEBO
zey|faE;gQfyHAsN(K`3X#XReG1lE*)=~CPX;I2!!#B~XFf<}zzq%4OB*kC$HZBAMT
z8(rv@JS$mmA7L5IID%y3MVA13*wF~)21rl5SD^4NPBPor<*BV0{i&RvtAdze0u!pw
zmAN`dmotV6ehVn;C61dvEa~e@_UKx)p1`WE_t&=8Qwi_x7wZhC4=F9k_QF_+KER1X
z?AdJYlE<dZNDCq}dR*|-?zftv|LhCjs#T(GB;kmy&e38m2|=t^HsISlCD`t^5GlG)
zRa)<I?5E<!<=ay^GWMKLxwszXl;N5vk~Z+CED-SzCIZAP=qdK%GG9xW!PR0Q%p;v9
zE)9jH3XBr*dJhJK%?{=s>a?|Z6NipaBg-0l{Bc=64`NAoUNlTymcLJvW$3V5(18X)
zKot7wUuC=c%d-43A;iZGd0EE%&xvncHSsQ$2%p;ZNVC<^RPIiCimdnI{+8AW<6eK$
zBb!=C7N8qCFUyZV4xdkU-ZvPO2ZM_PvPjvhe?=AkO_dQX*`4=Z_TcZ&U-n)u_zHT!
zb$^V{i%9t-)+N!MDoOmK^?-&Rs6s7m4ZD!syW-nS4Y6l!b07l@7*u|qOuCb#63@=Y
zm8-miKoHE!2&sboWD!zj{=|vF#e43#DCJmF8ux8llTP9~<$<2!NrqA_nI6vPi-%CK
zw9R3<$GShg6FTOZX^ePm%Lw?=6%#gJ$ho*?rBQPU*fHS<$RDPNsgQxhoJin~z*4z5
z$ob0KE9MF70#C&hJz58S2r<7qjA*z{!WSU>h;NtVJc*0xe9u*(I-P-%t<K!lZN&9E
z&)r#%8xeJkh}brX42Kq$-M{h}j#`hErY0VG@aWM8ADVdBu*ls|bF{mOg=4w=+N(S#
z%PNfLHNGw=fQ&h=u}C@qYTrlZmY3&_+~4iq|3K@ZJ3I2?S-x$Nuzcqucb3gMZ0kyp
zo`-Ek*k$l;jA`28-atZs++5q>Ub<`TPK!JD!vCSruS8<c4xVXom1<*i4b{^=3{y=$
zSTz$vB}Iq}|G2)y7&<vPsxo~CxpNwHEJM`@u}cec4bL@om1u_PI3>d^%c>PE-8r)~
zyEIb^s|aOT4QprFlpoHP?0V#zs_YQrpl_~Mjy7(aS+q=<V5mb)mI?e%Jy5&Qows*i
zwZ6Dmuip1StzLGVa=n)O=7HT?0$Z^oaDrP+u`&oFi9D1zF@v(c61U^hOx1}yu@67z
ztx6o&b`V#rwt;4pu@`qDyE;=!ej)@tTgFW^wYYfO;$m_9M~vb!^e9KT)t$D4%?{Wu
zAmPqt;=0<4xo@>t1i$82efSHu2bEJt_Yj%PyIt26s34|l$hB9{x_A*ce=wnbSXSfN
z50S{j+sLc|V(&`%ILh<7twWE07tf@YQ*eT`J8}BK)N?q^<XTYq_@XNGE(KD7(zRYE
zA+kTv6S`Uq39ZH}M;04t8q?V$^9$xf$FFeLma;Z&>N)XjR_h_TXE^PcGsh|y;t6Xd
z!=MkTvz-SpH83dRJAGuy%EmIxXul)fBi%1OEIlf{T{;hz=X;CuP{J*`;M=p$oGiwl
zfB`kq^J*bz6bJm{af5Hu6%G0)P0>Grhp%u{HH^l5GTkHUW8w^&>|YPIgPmXyY?rs;
zpXk9<oMlr*KbCh^`JRy-)Z0V<<#VMeR^>vnp0Ur*YF}ha8l^>34dq}TGv96_C@dJt
z`S=p#j(9<Frs^V+c@dcAZd&}gb4h>+6tQq;3WSrBNPVlG6+g&^HpEMpTQLtIw&8(Z
zV+UoteqG$vSsG4#=5;aJ&6h4N-!#uF94Byz`~2MpHF<wjY&-F+P@A1w;kTa6cXVW#
zxlERI!pN6yg5%{}WUHtFLdX8@8|F9l&l9+asO_AFUg<Gxwd3M|#R3<!%QupH?mc*Q
zn;X^GFn<q54-c0p2$TP>*C&X1F>`F(v^<QVIxl9^ACVrH-X}eC!}VSWDemMly0r!|
z<a)_rN59KEFCnf!(sFVAyIXB2bunixYeH~O++;mU<%T|<Z;>A4aywjVZ<wI=lD5PJ
z#g|cM^t@61%&iKid6QDf`;w4jM{jVi!gnTlQud8}TG%)tb2c&1L~yG#!qk=hK)#0#
zehSp|=G~bK^x1cy3-sNZTCc0xS#*d_e70W}h*4sb!v<smSfI2S`B#w1D}J=Qy^TC7
z^*RR4>~gNZI>c?tw3fs}O^ou~M@uQ24Nj{-XR|RSbILIngpkF2BLlZD&VX$mcreVf
zclS7+weK9tGc=)An6tdIQ*Ng$x6kvil<_PlI{XT^lh;I)eTv`P;fs_|c<AM-i7c3?
z0Bxn|F3}oOkAQ)4<j9?iQK-;q)gmmaJ-sYbl^&H<xusW>;=xxI&a~TS-g>6364{}e
zHleUe|M*=Lx`8=ao=}-y*e~RFnC4!s5~`CpV}l6MtC*?Bl@Z5_yG4Qa3pNMW!#OT5
zI^T9VxCp<AtLR_$93Qmoc(Ue#plYC9Mow)iJ`x)x!(Y)<JGA^s589Tk|G&h&3zQ_s
zbtRUOk^jufZ)IjxcUAwYt7>|>r+cQWtE=bN0}Q@I10+CVAOaAYBuMS>&l!mn$>9f-
zh7v`IkVH^y?j9Wz)XI`ES}yIwvs`^JG$mOOy;=*{Bd?Z_kL;t7btEsi(y>f<y-Smz
z-;0R+RCRSv4+MCD>8#AGtjHHHB3`_B@!oy9_a3Y2PFH+1t$KC$kE)h4fAW@&GdZ!)
zoow3en@_h@)w;>HqOf_i8H9#o*6)>rXeF*1VNgHPEcUH1Zen9ZUuS0wG8qRDyK+g6
zge}0ymSy~Mn|nmV2(7jbwW`o=Nw;=fxJjtW@me?0UPRm7HXa(Pf`%KkF|obI+QNj~
z;WJ7{o>-`{%;}jv?~)auZ5cF$Q_L<_)w}DQo>*TUq}q6b1nq+uJ)dEH%kuYh(?Rz|
z)g$np?KE`b`Ur&vP=EJ(wE07mjr+Zcsh_@9xwhlGo^CKQw=1qUN+z$!l^=d@l;|_R
zdVkBC@V*AW8m0Y|bQkk?zERrIUaDnP(D2(Es7RPhr9P{SYlglV-dK&I*T4cL_Iy?`
zm&z7&zo@8Ru_uXh_-!KI^MJhtmYvJusdy+m@Y((P&&PQ>r4IMh|LXPoP=lFB*hI1+
zdfqk~Ts5JS9bvh2P$#<s<0Y8k!)1}8vCbqZn?e%I_EFBZqtCUF*bHE61jib8Hd`?o
zd%hgT%H(`Ksf70)tAuNsPU6;5`*#g5aLs5UJWPmA=TxIv{ku=7x~_gg)AjSWO!$97
zoAaq$52MPHq5G(z-WRu%`u||MCj6^};WxtgUR<Ew9yh+CYmdQ?@Nl$Jx&8KG8$8<e
zZO|5^2)z>EPZyt``OIg&l>H-1pZe6Nq9>ksg6ZZW-x%&Q2hxPpg4tyQbPQNuieL({
zpAUU=<W4o|GRIK#f#&P$4tlS5*=`o54#6aS2#n7R)U7#OVYgWNsooYx89_Lh8ECIx
zFI~p*a}=|AX`}(V*k8c3*&s>nPinso@2l!|q?@tPF*U=3zlNb%x}(Unp~LdDYDe4=
zAi|Vd)5-AF|EcIONx^?RigWq$hG|)*WfDi%9aLI$RiV13SQcAv=JLJFlb6qtmr>h;
z*Cdmf(sMv@v+&jsD~+C}i?HBLg4VAbK_#O+JIT#L`JKyl;Mq=y=4-mmkVZns_Aqmb
z%6-k)hst-E#eO`l7oK%|+mtnyx?47Fc4I@tK=Z*Wl)9dPE7MB!ACcBshTHo}o(3@9
z&>F{hx{NS7gK-*fMt{8b$$41EjIC6$EVg)ccK_q*>#n+etI&Ext%hL$#Z@Vv!4X*T
z-v(>`hglnOw}~u*d;B$MssYxJCnna5ls?d&py{W_dZ=Zz))Dk_&_n<6W%>J9nTCLc
zYwWsUb^VeKN}lXE&FT7LZ?QhpbSgyGbz+!_`z|*&&pID>Ks3+AWC^N*t+l4IKi+aI
z=UziYKpMAmw!b_vvD`n~@jOkd+IFK}Z`eN(O->wcLziz$gzY;_ur@|(uMQn<ieMHE
zP@7n$8J?vMgvLdg$N`x@({WvR9ZhbUKYZum{-R>Zq^lUp^j!N)n-Z#9v@vz~j>AV^
z8$v5eLwhSt=`^!1?Dw8&rZrB9Oh2wr98rD4FoDfXs}ynTLah`q1cSY41VtHSysamc
zCb|tP8@QCSOZjF(iArr(mR*~w<%@mlw$6WR5S183QQ76|c9$hC*xXR!$uj68@4O!O
zdG~^;aG>$K>8rw~c(w~wZS*43TD<e{M_+lSRfC(iHrsJkhBwZrFPnpMsoyT&Fz*gp
z|AW?}$YqL1xJL>U0^B0ZHjkPAYgflCfS9g6;fd&RWyJ6Zw8trZuVHVnt&2ZXbYc=p
z&#`}Ssw;N_^T}E2Ajr<*du_eb(d{2sWryBO6gq=3O^7h1DKju}O=ADb>1o1h==i=g
zbqDDfy5s?Fai;oi@;>{44}9S3V@Q0Ba+cDw%4_x_b+9rR93Mlj;noSMyHjP`3T6IW
zcS`q5Z<d~tehc@E`=id9r%CS4)!%i7q4R7qnj-7{LCLoZyTe#d(2|_TPne}Mw=wYv
zKkf+_P4UUowhMSt;r}-ZleWFFv6XnwF<=ER;n<g@<MTsv@H2dcK6E(GLj?KJHgIPH
z9yaVfZeP}vHr|87m5q*gZQ^y<UU<Y3%7*M)OZeXvn2*fijBlQFI9WT_d)v_M&*$E*
zL-s`;Q+`bPu=FwMap`xYKgzKTMN<uJ(eBke=xY8-v`{Z@rK3o7H`K3209v0PQYn6O
zOatD3Fz#iyH#RPb_S2LX_USoI%Xkj8<mU_v#9xlvSsZqT+w#iw`%dY0V`HoEW4^La
zH+l|DWQ>rq@2$u3vt^D&V$GHh!%foPs4uN9)!E@Z*X5p_Sgh9<Cob@h&j=oXa#UX9
zI#L5Atk~7(Yge(6dRlCy!a|n$lVmI4Wg-ycU&mIo?66fA@2)Ssn7sy$JMmn#Iu~Ey
zAKL{r_#p`lBxa-0qza2;b!F&LgE;f261j;VuXWOM;kmSP`f7b~ZLv<4=;UHeJAYoQ
zElz&$zKJ6f@Q<$%S$W8>7&z7*!!zc##oo{<Ua-N6WruT(&e`y6r?bN<_OE%_@8lo9
zr7TTp_usEgExq+VmK`h3_N8^UwrukZ;CvOBc46$L_=c05b6wnpM~BERLg@ARMt%IG
z>-wo~lKUUj9hcM>>mA_F6(Vag8XzgT>N!ae!Kxr|lTJjamR62&YI^PFY@nJ{jplt{
zPUfeQ&@t$H^R+V4gzLH`ko%~G>*%qGmWwHwQPl6EZ)HdCGxvZd>Be0i<){PIVjY70
z0lhV0Wx^irUj||=$973ZDj`7NjN;p&WuLnlvr>9%*zO%@aZ`txUL&>vwIO6FaMMLg
zEh_L%q}cn>hpA?IetqUW54oMlvfey9;JiI%R1^}-M=CYdVAgXBtnfNpv&I}(CP&9F
zVDkIDxQjRsNc%d}OuvXpkh_hic#`BWd@3@H9m9EggJ-5reD>j8xw3Qhs;pa}Mc76V
z<@S^m-seb|;T^rii#PVmFY(C19PzIKAwW)KYR3(C`&m`tco=wieAht`_~oltL1U6_
zVw<zkrLh4>*_h5~-GfT~3TFkpe#3OP$CY37AIwEi<M<SVeDRMWrv$+@x)ERCQ*aBw
zEtxw{;$P0a^Kas&t}PDZdTj{JmnSqg>}xYCv$HER#onsU_J|Pi7RMPw0vW}T#BrUQ
zDT@A((>^*ejjpe^1jGj%q6-2g_9kut-r$yuH;cN!p_8LFy;$J&ZfsoPm5ORJSb@{r
zobg<~$?%t90r`v7a&w>i2GnX*T9-~po6=e7c4k?9z4U<ekhER)B-48(LnN<ux;hx$
z9>KVq(%(S*>3e*#MZe-I2%*5X46ni1726!oJpJ_3TkO9_I-TR~_VHau<_GeO4NDEN
zC5G3gLG@?2)j)f$6CMwvDEcHv8b%mm_McSFb6l0?q4uwn-omZ*%yYhpt1XwI|Mv&W
zED5=WK1=JUY2$qaD4dqLZ&#OPhi?+@bYIdcB2agj6FJJa0fw3Xw`5(b&8dVCS#Krk
zs{BS+m^m?~+l?JhBfn1#*&?!Hs!##TR;ocVi!zL;(>1j-qr=Her&mu@>ct2a5L7g$
zTHE<NA<A4$yYDrwrp66;KK(EZD{2EX5Zu^PoVO4~Hat}|75G*O68EGso#<1iZ?Tnh
zMk8|iv0sv3`Va6`;wfKLD^(S-GDRME>Hn6dq($c8CPQsnr9p?L-QK1=CF^AmaI#c8
zw7S@kJH8WPlfa=PHx^e9)!JH?AVS-zAFk7+TgL}N79mpaCKPTKF+jrh3h#HM!_YtP
zKs#YM>m<H;JW;!_jLM?$S!U;6K<(g1IbFNxUCBeZ&KDU9uN9idSx)TgXdSFV+lumv
zWPJROz{e9P(=h)qbea}M>~sqBuilhA$n&C(rmR^_o`tDTP8W_XHHSGnad?(56>tH<
z3{VJ#RXe=8I7#C}b)bs`Kps3^>n4w5<VyAwI~nEk9nudSoX`7a*%^hBJ^<fA-yN;X
zSMsJf0RLBtdSxf=qDi2djog1Evp+9MeUN;2NcTz)N{>qKl0G1PT>6ysi_$-l{;BlO
zq<<lOUHU`mIq7-nn)HJ7|B!7q!!0vGn43k}37Y1I@03oa3K(+TCHy3}3g?sD@Ch!V
z28l3-_x5-QPj*i|gN&d-_IliK8Btdb@cQmhH&}yyK++BaT;C+!By|Z8nBpeS6vA^S
zIzJtt9TklcRXEJ?Q_xJ;@!KTTx~w!j?R3}SHF~EDwZuDBC`&-OMSQgriTo0wxV!61
z_z1-UDdd+HU&$q0K}oX@m6gOg(ynhG5~1B%UJe$Sy=Cat$N4H}>`UHS_~?E)U0mNd
z>dU%D6dlxQg{p`!A<G)V!|E^tPz>5pbV^l9iKc?CN;SA`g5qk~@SLF#>UvaBp$JMV
zj*KZfpc@+sCUc*heny0)^O|YtezhL`l5Z}IWpqUL;q+x*dEs>9@=`tV5)W+<aNFbG
zKDysqo-KTzcbdI+U`(YlOnK-v07qMwoeFXVcaA(4gQ;K=v@BScm>8*}A)*g*1z4vm
zz*0o!MW!kk2Sru1Xi){<1ANrv@I)?y4}-pwGK5j|%|=6729AYMNXw)JGd_-Ov_EPO
zsHS`~0!f?dCyLX#%G9LiSE@%+2gBShDprP5!A}0A?oCWh7`Aihh>)xJPE~433oI8(
zs<*m>{$6kAqm5}5Iwn`(ZW(i(e|OVOpSbC!vGhD(J5Ex!11svRnk1ZVO$Whr3*^nk
z3l}biy9xYZEt!Zby4jhv-r%>UL-@s1806SAI<qvsQTibFykP(L*}5`|nCr8pWH*iR
z$i9y|;`@Q#V<^qjEHDrECA)FLvU%eVHJt+o4`87*`dFV`4zT3bv`lrwsRU7Hp<03d
zZ#rI3O($c|R^|JefvFg#?fH%aV@J`A3Q=om7&tcEb%LN8>bj=R&7{-R31qK1c`9+P
zlCP^lPyua7)pU5ut>Dfi47mx-Fe~n_Q_B6zJvv#bn6~9%Xq@cR$gSG?lB-Ts91Y!c
zJz^-TVd>VVJH*r-(>5*Ja$()(!^|@&>n03SyVALKM+gJe)D0a03N%9=$eQBOSl1Oj
zk4^eZwx2u@v#r=k7S=H_X>x5vTouzO6WV+DtrpjqnXP{@g5|`<(_9x(A`Ca;XPWYT
zK6njgyDY=2WM0EpI-Sn8s<3&J!XQ&_Rv1#+io!6?;8R`U9?Hc5jgW;;G1PH|)$d`L
zRo^$XYGO&m3%ZMVUhW~4ndsnKp;|CSv(JWRf49$FZ&%InUSp`1-`~ZBj;5HF>sn?+
zwHH_xm4Xpb)Gk;NZ9G3xbL0!~LslpTercFqK5{|#Xi_iqXtR;?WCf$CbKb0QpKFra
zRMGZ?J3K&l4sZx{2Z7N0Hwt|{g@4NXvlFs3bjeO~ENCpNSs<*B6ga?Gqi*Zm=+Eir
z*6+Dz9i8V!O8$d;-hIzd+0%QLeTL{Pl&QPB^hfT2V(uAJ{8_vYZ{T}C-y;iTL9)43
z1)&JnI$b@9bskpHTTNjJ=7I{kKIm$#7rnOYdv4czq`o*Yu~=^=?WsF&Uk$?5+wYue
z*CwBssI{XbM~_rH^~pyjaPQUcq*iTk#~p*3opwCFZzp6O>VO!NBn`6sR4LbhRzZ1X
zvG&;E5VRw!I*X-=6RkC*8trhy!eO4s=8~cYaU2lZk(Ey>a)<hOr77pDLVK;Yt`cTu
zm_5!H`)^m3<I2SB1nnw3Z>-X#a0xfnr>k=dLT#4#x;3XghV{fI^CwAiOG7?qU}hXv
z5m;no!h><Pj{qdhfl*6imjygCY%v5(#D@k5_nI)Bp2;F%P&pA!%}<5wa5ic*I*rC<
zwmt6;)s$exLxiID9Yvk_=uCEck#OEbAul8drf%jIDdyccw)QJ9YJ$|)l*O5uMR-#h
zsLDn@mm|JQRhrG@%5=62(O4|XklSsfDtBn~QD*ECCr=kK6(AYZKSD1AIOae;Tu%Ga
zOVIaMn7<if<ztyRb1|GiC>iG~OhbSZAZWV@TmEw^K?hwa_uL;500bcK&*+t(uILp6
zbE_1mvx;n^NU~++r;F3KQx=~5I{ra7I>68uP1BE>r<#$Uoq%TK`V(stK5n=26H3||
zZL;kKmX=hU-H&*?Wejp2&z2zK#!x(47pUdXcsdu+aXEbLnc8$RQBRbL>sA`2K()D@
z{^a81-R`72D8&*x%znCvV?7%_FqG7x*JUfp)<C>2YMr|CREr%*z4Yddq&|^M*Vg!}
z=fxwfQ}ZP%fQJ5EjEGV_mtgiyS?>{V*p-JUr{R}b;Sc7Kg-?o?ejdyGLSaPQ!WR{G
z3tVHGAYW#g9dlQWMj>lP@z&1|Rqe6D7rdOj>sns7A1y3>Sumtg(4vzz^If_Rb2JVx
z3{Q)(yE60y4i#~RfW&1sR<uNP*H~T-W~hgSo{lznrv7c>y+MjT3wldHB=n&~t{Zmn
z!S{o@6nIo~H$iLHtV5Bd+uA$Oie-5=A(5xpha%IkwZl%uwN#l}zDJ>VyIxg+r}c+H
zm-pMILf!|OzZQg!X<4v>_)igNVWwj^)%s}-w7uw%trA&#2UTpxl6A|{Y1Foi=#Zts
z?TTHkR0Gid@2OW5YST(R*5NDVVWOCAA65;Is4AgmSaAr|^r(ahicx+?rQ-}~mc=jj
zbAV*@K3MA>>-1rSW8@w4Lq`9u%=p~R*4_D?f1g;sX^F155O_qELF=s;(Hk5Lee>_M
zgCGrpe+&l#0?y=j-=|?H+ho>>J<q7drf=wkXm0|ENZ{!W9y{0IKlo`ZBn`^GnB2ZG
zFWn-&tz<{f@q)V9@*nR&e*lpWldr~uI8mNe(_=}@T`)vYD|8H;J6PhC)ieXjDA=>Z
zY55`k@m@}##VcrEwM^eiBBP=j$OA9PDz~Gnn4hO&dS*qdL>T-ai&6Y6ua&4O+KAx#
zqk;k3lNiXUX}}ZHH{+_|dD-GyCMwJ{=A@5Ff6!4LR)%gS9GDLxv;8iC#xyG(mX@Kt
zS4(jpc|!SeuBS7Ib($#$X}_h@_&Cq(aSt`<SxR5~R+OsWPG>KEx_^HYDNX%n-gWn*
z42Dj5>pc`6GUR)hC2=-U;Gy^M3QcTn5&zqnUMM|JT1;Y;<1EE_9#WZ~XMVxB$9O+X
zrobi+LJS4N9F5itoUvbo89QR;@xn*o3Py`u!7undHZg~}ek#2P>((jenc3~Pa@!v3
z4B5;Icb@^p7L7B)clbQpbjF?KZ-1LvPnqHZWW1oLr9I1-p^RW=P?)J(#nuieCxzr)
z7DmR*-T5kdc&;!r-&5SExsw40mcky%5{9K0K}g%e<RBuf$};f?%WbgCysf$IdXtR$
zD==S#1T{V$;)#{7;_&Xah9`_g;*#^+ww)c=3c&P}Wlrr=j!)1;8T(;Vo5t^+Vj(|T
z^Vwk<x6AbVaa0XN7!@3Xs@QJ^`WF~}70(ljHbSJh+Yvi%FkU&)Y}=PBf`|`MibG)`
z?=wmA3F2Bil_~<uC;m=gTgUm{Q&E?z+*Q11`SPS3R7^Fnp7xW9QTgKsx#s(>VZDjd
z7QpygWpXsvM7dC0CPO^tn<s*kFr)MlwqX==i@)a-(mNAdIgfwFa*1gNSQX2tR(>Xn
z=(nFO?Vmw4Px5u(@5;XQ$3S=DKWzo7SqbD-k?Qtop0hJs^BiI5w^FpsqWD{L=pmqU
zZ8@&uPIfxQEH}N?P5Cr#?%;$8qEKGUqGZ-`5QDE9)daX3*j3+fbko!w!>_{S_}dz$
z1!|(5D_*6#MjD2JsmS;FMfvuRY#=7Cp~}fB)eNb$h6|)0!x?Q9f_DJ3ImWR!u|^10
z5m#NP1hZ1Fut@jQq`S;Ea{FtWP;ri^q@%wq8v-mz59?QH9x}y1esERs`#j50UL84f
z56TCA+qI^wu;#^~WreX<3x_U$o9AK5>U%TH<_F~yyImPqXntAJi3dC%>C5R2I=Za$
z_@B;tnk_SLtV^si-ylQNp^x*`*bVlI72CKOWzEfoeSuiG5G&`(!M%_~`VgCQIDx`_
zC1!gB{a3e(T;>ZySurd_!CLlT(LS<`m@)8%52>aK{}>xCyuux2L~h*jGJ8sCuDh}R
z@k3j<s>3Pp2K&WDZqS#?a>9|`JA&@`>bC!Rn70M_551x-FZ{*c^@=xkxvk&uGTVBe
zecl^u`vtzD7uzzRdl~8tc7pqdevHKp^@oP^ap9@A7s;<~|A1y3t>Vfn+E1PtVIRD0
z;}!2S<+<c0#JY`;2h-$+_$JgimoV5JX0_8WNEf$QuyL>hqUr8^H`Ifsxd5JmbyT$$
zxv@fts_InrOj{3OF=5_dfBGt$qKGx6Tbf%7(~7M*x)s2ej%UfBloa#MwDc3w&tmV|
zw?eOO|I)Kb?+Yf`g9Wc_e|v!Q`$Nti{Q0R@w(~tHCO~NjC#)eYNYW^R&7d4R+ACia
z=BE!>6-*+rVPPVP%GOSX1opX#br!xFJ{0kj9E%&#cBH-Eay^^a@VkF+yiM%cafC!U
zU|GIEFotZN4AXen=D8=CW!vX(4v|zcetP@!qwuE(eq^u20N-*#u5KxMyc|hB-q7js
z<Q_L8Ayv0zF^z`F9i>b7dtVdS&eC8{tMz$q*?A}bc)@or;39SCkn2LyGpxLc{mScI
z95!gbPzF564q49R!@#pPt|r^-9R|I?VqH$c;7<4^I_o@q%evd?>f-8{fWhTI5FLvc
zFw&#m!>Ud?M5SvAlyX!viKWPTw`Lm^YJ{5k^OQ~=o@kny;%Ry<X&L;9qlEmmms*LY
z{$Gk(Lpa|}N&~|V(@v}_+a#(UQIue=9e9Rz23->*sj$l}NiC@-osw>sUXQle9J6Lj
zX4>9fuC9C$&GtC>NxxfC9fuM)Fua;`+eMDrNA72Kh0L6w{wp<4po><T+<2kN^4l?E
zHXJbS^s?henXjo~NBoAGN1S|1DJMl>iLjv1!!o4`r6rrBC)Jqc=Y;e?9)pSiiF_Y7
zGg9Q%if-8|Q?^`m%WPx07K;dvosu&xr=wj&#&V;<`NFdCIAajACp4J(<~*t6uMsTM
zHq4=V{bymO5m18l0J(sLZ!u0pY=4f|Rk|wby39QOhTFEK^JZhCfm@7+vuBoCGb);H
zr5yZ9)c;&mLdlv+zFKM|UTc1|_-|`6w1*-~MIQ7cWjdiZB2H!=6-!qW?yT0!2ri+z
zER!;w+j*YC^TIJn1nDjoWSV*JB1;*Ev1%`Uw@m5_qJFj9U{~PKRbU3W1y<tdGr`A4
zk4@{@Zb};GW2Mki(-MYwVXb3Re=pAZUMs_`xde@U6k7Zl<|{^6tnc@^TiM4Iwm-;W
zy5aoOXElTe-q~V%iI>_u=%&pC*9x;sGFO3bt9NMBQgbH>i)xoeb^tsUxH}V)Db2%-
zaE2i`{Ub>_0~O<vK#YVeP^w%Bwjb${E>mdqE;FeO_#R~!s<%>vK@HZESkEik>u0t9
zdBdR8b?HxvAZ<-lTKOiGJx{Gr+oqL@tn1WODID9j5?8S@)mX7tfC(#9Gq+d;e?_0y
z<?xbd#)1p4p)@&_XcZ+m8Yr)It<Z9*YBy}<mMu-C0rR;CsH%OqU`#oeFUHiY<QDqz
ztz5Kaezv;_H+Yk6acVi>_E1y((NN2vufTUPA2W*;vLzoqHrmYTJ5IYXjN86Gg}NuG
zuQAaZ?(FT04?R?0dFY{V<+V+`>dl;c-70<J{SQ7SNAY712GKEoLn`%a9)}c)40x`b
zq|oq~k&%?%QQp~qymPAQ)E^2Ts@siI-FuSm>+bQpA1r2hQm2cz-PL&LvBw^2+;!XH
z|9<zoWj}oP-LC%=g|>jNX8RzA$RUhaiUylJm0P0mSYz~hI73)O=`zDDP1jeyPZh^8
zP^ufm_&r(P(sU9L@-?c_(~3=%O(QXlm^4xB8;%2yrN-|OctqE>Wch0{kv9=5b`yTB
zXYs%M{YMW-0V?$}OW1xeS`@0&Z+IuZLFvJ$ie>p5C%kXS2E~~7mjq6dDa{~O4$CE<
z@@QK=EBEpg1MHx$iKxT!p~W&)VjA<o-vc$%w@p*JbQ{VUkQVHp8ba5!__a(FLfX{S
zwuy|MGw?VNbNseTEUHt%^aCJ;LL&1skSi9I$+l+FHfkDt%ydW#XpT|aCPFx@ra&_z
zqzB6_z2r{@f-xz<dS!ELOj@#BZAa9Qjm~2-@rXi?Q1U(&vw;#?#A&b$2bUb57}TIq
zyGrTp?N1PSSyp^`k!ghjuZp>g;MSDG##P2!jHsHI`T}Jvi*)C0pANcdSC30v?t0J@
zFhDHh@L$Qy2K;o+gWp)JEozIGIB(U{a`R3ZbJAeM8uAm((N}X0{{lRG@8Ts@aJDYr
z*^~)JegRi=Kdr++qoA&YJ~H@4=xINv%4_iaSD@@QStVUrzNl!whWNa{#$4fYJIY~(
zMJ`3DC}OZKQ13G*(810}RCx)?dd)A%>dy@qO{#BGFn0T&Xv)Rn4Ov_PqpvlNLUq?m
zkme8Y_dEdwF9FTo?$>#;{3iB7N*_l!j-j`f*K}2>$@~gH!)akmKFcM66H<(Rc->W$
zVd$Wap=bl1zYw>e(!!hlvP@M&x1#h`Sg>MdjIG)XnO&KkJ*PxL7|w~iH&-!&Zn7gA
zM#ZZ-p667(iebnXW>?tb7YwV~uBiEK?ph=jxkZrHIfjpk_=peVo_Kq;w@G+BWQnl3
zV{0sH1gk}N*t32UeIz>FxvU~5`DQqrwpi7B-CQ-#cDz6o!;;>BmH4e`Wa$XwLa1U?
z@2lzx-Qod&ch0hqvrpQNuCWS7uT!deGuWVtUhUq*zMT~zbGtYYYPxNvq-NT>c9BPY
zwbNO~mfF1vVULA|HB~pXiD?WDPS1(0H2$zu?Aal{QB~mFh@GNkhC!igxLt%fcXKy_
zIh^y#=se$pF{oA)^=ZZWHQe&|N#O8C`+~#v>jOK&Hkte$492HbrJ_DfZ(;XSKjO`G
zQS9u8JYF)NAJKMz(SRqUbE6rn_c5b(vnI<Sh0;!!w-dBfI@}*$V7uSUcY2b+(xdjw
zRRIjW9^OlgDR~q0TH8;Wu4cO4Vk5B)XjxG}YbxBvU-76DQO>(Xn~1=?%ciO6mI)n#
zSh}3Gr6jQVr2oYHH`Ypi8y<r|z)5hG;fG@!ZHIyKalGxhE_2vn5bwedW75|h6L%v)
zR(r&BbYHesD6t%Rh$>dYQsA><%F2o*aF}oZqK5lS4<p?V<7Sig#fo)GbqrZKN{ONn
zdQ>Hbqn<L$v6?=MjO-6#3h~*bSI*Xz_Z%_YGBvbQL^htw-DVrSi+;e3>$lcpmuEfK
z3l}H8G`PZ>WzECbbx&b7oq{f#MfR=bNI;B?JmzMXyH`fMAn#H^3xCU{Xp&Ldh`toq
zpR@zrn0j=|&~aZ5X}QQLer&S(Ev(JAs*@`?_x+Koc%Gt`ah4<K%cnsGotG|T>wm2P
zldfv5vY&KEgc5dmuBXB+oc&<7;5MJR8T7(fQJf8ihKRLfRl5XLWOF&whM!TiY>@&t
z$&`{k25T1WZ%|GfcG7qNI=gAAsu&X(1Cjp**O}!X{T~0w-&cCs#;6>7N(_+PR*pWd
z0mvfI4BNYQt<~}KP3FmzP4o@Q_nnQG{sR-kPSQBnNF1C|9>Mp`&ZhrUv4;##%1`Yz
z!<6)Olt)}j-Zd(ZST-R+C87+zTp@I<4g%7{U2`jeg_pc}qY~);1x;36+!(jpR95sK
zKcM_6?+M2&&s!cZHtbuB@mpe^#93arTa*KPIycRlDRBJt!12_az3FO=ar|^CO6t(9
zTT_wk63>pNwr)Knm2K3HG|Rk1hw|U;-|Wv8$mdV*)!t8XOUMTf*v>WPA$HqHJD2Tk
zgO}OF_HmqN7$)P5Htv=P$2wq}mg_UmP~h&vnI&sWOdvDJbAq5VPM7C(N>*QP_pLRk
zrc;cXad2EUWMk%Fb-c{#$4Yr`6C>uFHHPZ3=1>jYNS;1mm2-K^ef%+7#qBI||3)e_
z*mFA|pUs{<ShZd(+U{&-Qe@rm*@M)lQ82cuDtw<(QMY)eDrtC+1@)v`UY3>wHD>y3
z&VvK#L{DMdqw2Y$t_M+pz5-RMc;T$WDmX$v%GHg7OV#~|-b04&N%3AHb6<K<KJ*SG
z9rh!2mn_(m$dRGHFKsy1&)o1x-;d5c$Mp!k2OY+{$S=t%TrZ<<<LUCK-9?;w;4wNv
zBGIllCWncm`wv2eVQMc6x*wI^y)V@Rkp&18_$k66+om|=#5CgLt}vsQOFA4{j<3Sl
z#X1Le>ulnK>w`R6sXM-szfCMlCd5{$3hKTrE1Hb8go7-@8USwFm^8)01Ahh<ry524
zQL3>Ec3WQHj@V>8Yu0x?PhQr0*SV{>Bjx>gg7LF1Yz)Kv?00f@rfch2O1yGhph%V1
zDe-2TcwW{=Z6lsZFj!*Cus005W0z}}ROc%~0NW8lW<|EaQ1iD5tj{CE;(hf8-$zJt
zII}ZJ(C)Z$I?$6%8DXYbvZeim08!l88Z}>gQ)Mny-sH6A>ZVnno1HL>>9H;P4wKeq
z-QG=>lFrRlOtU@*zkuWHyT%WxG4fzowx(#WSjnaaL_ERp!&g`y6hrO3%vlq}fr`=q
zg>!>GvWGNKMr6S)`;Mt}xLP88n5dqpR~ILd@k8Ygidw0TgL%Uo6jpGQv-9!2|F)uX
z53|SQT|dw5fV!}m?PV_$itNVdVr!ePF84Hl@xGYJAJli=&%Cy7NZWa}WYYDNqyOer
z5>or>ROem=nN_S?4XFXEhLPSX4_O-rXi$R_xQ4mw3ODjyfrZM`YH8u}eIxN@%bBP<
zwj3no4p{CAKWBJ#H-P`#x|h$P%Vd(-Kn{Vvk9!yE#q0C~)5U-g7!<+tYG9~Z6zQrF
zEF6g1mSSs0#W8?yv8H%d#c?W@r`&UZN-Qd}<(Z~ujp*_wLn0cm?6k}_#@BsgmxV*<
zXS))|r@dXV9v8(5@sZJ`W3j5VvM7^0P`rHD$&AsX!}Jft&V2zPLU4;*0C(s3a(wJA
zN;jijVa(W#anqewv$vXRdUuZFC&ax2&41ZA^tV+mT^2pp)ZAv`+aS9y?l~Qbada`v
zs)bN>1{+e^o|9v-j;}(SoxGmJ>+S1c8tt{1e_Y{S9Stq_-`rD1ZVWxf+(Cx)O;ZYC
z?Xk|{d`ls5S(F@*yC@IUK<7}3ovao1twPqhI-HqXqeCa6ieU%w$vChLg}$)6URmc7
z`vAW_GI}aP#))>>8CO~1=S7Y%#3N~t<>*^yK$R%5`<tX3XvSjv85*OAIm3{@Os&3X
z*d~J0+a^dvg&K}Ao35!;R}?t4%+P9cGxTr$2D(T+$n^gQ@$^YVU|7bpUOFpN*-&Qj
zqKg4J<_b4O=dyGH#>7!J9)^$rtH}W4k#Uaa!#6+Uba&CYBV`Uk%Q6#Y*h*+0*IPS0
zeEkA<@Zp#aPvNJgDkk3i<%(w9d#|e+zf7oAaaB7^&!(ZRx)qC(pX5M>UqHlGo0L)P
zE#s<TSJco|uj-~k)yVcdJ5s4)>P&`>=!(p$w7<4^fLZ&x`s4Lx-|H%r`<q(I_QD$5
zvDnMPo{Dq>rMae=L49N@Ssr6eP!O3y1HzaWU04qvEaI&ro)61`A>jK!j}r-WL*_3i
z{XN@5IaQ8*?Q?|TZok)s)kC(Cf?Zs=6XshiO``K>#a;>xv7bjFO!*YFAInzdsyYQK
z<CI#JRU4r@;TsTrwkbAK{`&N3?A52Y73kvfnyOM$yE#ab;AYLFs=9_5N0n`SrZg@M
zjt`gNdL3kd4#k2BYU^W$W+$D6PasmzcgK_a4a?cIhiA(S{eP_e@CH{A|3BFe54O;_
zj`gt6H@Kg)i2mA@RXNiln72gPJuxjVToUV?O*(GDuXi);xNu&+GA?Onl>3WmVYStr
zjb&o%KS{*&QRH8~q;OqA?eS=rUIuj54yFU#0yNdf_>58bpmOYxIgqP2p&w=e`<7`;
zI!x6=S=xr+6Z`qd&eAoP8I#3>>)ZUb4_~KtpU=nqcUaOhTYnSb72e@>W!{!p$wgkn
zMXpRKUn=En&M;9`mDhS$7q8(S(&+qAP~Tzjo_Ya7#O0KjCLYP@ON^*~cRex^6i19w
zF&F$)iIV41w@M;6Epkg|{(VQ7e_vgik&a0lBKKBCQ*SIw<5nut(Ibs5BoYBunY&`Y
z*N4BW5!+#phM3sd*t(+Ini<wx-Fj%eZUa=a4g4>gx>&wA!cxb<SDh=**#@ZcNhLtS
zJ008Tw4?T}^~@>skI%6KF?3}xv~b9)us#@KWs-9FPjntcd=WH+4=EVw=SAz0XJ{9Y
z__r$By9HliEGGUX?iz7P*D$#{3{%Iq9LMsVzhlg{Ou^tMbeO*3hOim^;3erLHWw|x
zTy&F&FlQ96i<FHXnU)i+vmyf*aXpLR+^}fvPm|2gj76L9zW!A%D~~JMr0o&|2tkAz
zGobBh5UVCJ9nB*0T&fazSq(JxBc>8hqTrm>SWX-G^*XG4?onu<(tC9%;<fXNcFbJU
zblKE46#A}-x<%#Fn(XO6W6sAQp*i1w@O`KU&IiZ11y}Tw7+2hU9gYii2CSH_Pss?H
z)J3N*mLy>j1n`tRNHJH<`rsO;LsP94J=7Of*{G=PpcBr!)o^7__No<IsVaV;*v8t2
z9A$bjtOkn<(-plUW{N{rsx?%i&MFmKRwgRTvIC-0o?VF|4+Q*%Wf|5zhSLZar&Om3
zzneOII4G}&Cz#FSG|U+YhdaCm@JM&pz+4}+a#`DAGM1wxBeWm$q(L7UQ&qR}5$ZdR
z<}R?_v*2nDW*R2MFdd)ciXtj7As4t?$(6+WsuRS)73@A&f*2a=t6nm^em~%vrIe<A
zmz`<PHOu>DgC;TCZjRaZbRy38T)C{va+0tJ>I7|3@Tg|USj33EobPj;yFt6pjoQtw
zHwAz7;D)q)=i$zQ>}r|62d_W#L&;ZfIST3l7H<My;54h+OtJ;t>#ABZ!bQBQC-`Fd
z#uj6eRL%>D{`RB3GCZdOss$T~kMa-?7kQjm$?pY&0eqfEO%;A^OVT9hCvTQMB7F)o
z(~Y{56iy^*?l#gHR?O#ELd0dlI|RqD(vcPpOW8sDUyCDMLC0|!%r(QmV#&{VzISA1
z$&&IctgSHRAl3rrYMX}!l<xc<D8;&B!4%-cizl@NTNjbP6pJdcdi+pmY3lbB9W-Rn
z;S@8Pr3jzbN^Ib%_Sq~wiGpEY9NQ9E#9T}->h^b7a$CG;BPZF}-Nb;>5dN_O4Pvu#
z)TRb)qvLOEsp!uiBkm+oRCp^j6`~o6YQhpd(zUl0*dfB7Q+kQF<6uN-=w^O@9kEDB
z{E$cNpG`cZ5)T)6%DGJ5u(GD5L%VV1TO&zI)U208LIJ_`kC^x|3Qp%e=6U?;B62~z
zF!2y)&chf$Eys`Z7{>mFbUV`l_v(*b^e{tgGY0wFS!?4Ongth3VYaw6eZ3>TUPmaj
zk#DyWe+!Ab%TKmBZJ&~{t%HD};d;d2pB}&aw=!mq!n9Gc{!NTGAi7SduFFQD@J0s0
z+}j^hbcNE{yJistRB4=Ti_eC*4^xG#Ru5NI*;M`clv1iG{=Bbfa8^BB<Yv?b?~hA&
zNe@Zy%A?hC6w3anW|mJ^XeNd6;*t(Wc*mrS31+QYCTv8k>*5J*uLrIo`r2Iv&sWGf
z)#knB0xh@#y!s9;cm{X>L_y97X6aw2I{Ostda+PFqmp8JmFEZeyj<oIa1Yj@B{sqr
z6GTT&3I3raoGNgVb5PH7J}hPZHrk74V(Cxwt=K$DNXWRv7wHbl5u8!6#((L%0<D7n
z;JZKgZXPQUfiBw8u}i8v_F@qCmSwW^#CeRac<K8r^dfwmb~YbBkLN2q=;DdTJDZ(y
z45kJv+6XlAJn8~UNqNETaRC>X<hk{Scu>VlFN}$|xHPjq_p>Z2W2F>nk-g6c9q2ac
ze(4eP`$J3AoN&xWAT|-in5+&s63z<iaO(q5>8e_1y}u5s6NA+T3PB5jc`fO}{Q=8Z
z;NnJp4(R>)$zv^;d2FJ}o~@V_7i4D#B&Zp7LtEB5u;%vwaX^m0{7?stP*r4j2oaTa
zRc-a&0jf47RK9|6)Y0KcQJ~GkRtMBl_(!~GKB<_Kmg_ppib3gws!o8ETyfh~G-*(V
zwnZ77m*T3l-c9a=JD?j|m{dm9P^8L;vW?}HbQhnk(H@X~Li%($uL@4oo7^9hp^E=Q
zSf+2vB1yCk1+?%BbeToi1{N*a;V$zmy&PxmHm`IX%m<2wLtufiqw{ik`$?s_RBs-g
z^gUH};8C4c7pjU*tk5B<XUgb0NtA%9e<vE@H#wgY6=pz;woqYoE2nR`3EmjZM73@j
z%-D*y(4bsKD3gaY#R=dClS-AhktA@xV73=y8iO$qoVlg55L(m*LDsD^ZII#L4c1Vh
z`YnY%BFlE5Dko(kdoZc?PbyW_R!;RP^_!-3!j{RD{0N9J`-Ejqd6f1~DYjZwPWItp
z4}Nq~Q3E@(CGz(lX6Q=?r6D~o<Ju0il~xB9L|p{w1N9~CLEq?V)q{|z6XmR;sk%`y
z-;V58&I049PVXiMBEGA^&*6&Dw}Xmd!4<Rtxtr<e8J^k^jJc0^&?{<CalWR=whJ^4
zu7?oY{k7L3rmE^t1w?zIDO;MRZ7EuUwq=JbP*w3AO5Z?KHA<lC+^;2xb`E-D5^3ri
zD0O@ue*!CF+zULyaJ@N#DO=m>y{_1uKqhOg**V}CBK%`!3PH~@IYHGG3Yym8!!I5_
zjL7pVD%}7!z4X_hGQB36Jv5h4f4YjO8SRG@mi7!3C1p#sEcH*Sb5%K<>CUI}+ffI5
zJ6&WkRXIH!X6@XMqCr$3husiqzr@w~NB5_YSYk;tj?!|zvIS`ovyU9S?XW|pF?@@9
za<9Ws72zwGs#;TIzlNPbOBe30DNiVtZfb;Bs{C$QJ5YnysP>3PRn3I|G?i-G8i+2L
zoBe+9|Nh`VlVebqTiNWxGCANd>htJc78th5Avj!&GsQq9jV;#^f|*T(&^leHPO=G0
zsh#=F#Y(g?OO-&DeH)fJDs}DZWUEn~_B9ow*c3NNk4#S-u2<(1y9#=?d-&YRO8YLd
zG`(^At!tHrw*?~3kDV}y?4Ua}ITua^W&{)m3eo4n+Cn`!w7OWgjCrqCsh$0y`;O+h
zR6uX@*}Ofho$0M~Bv`)T<XC!_vpg;!NdpEU$bkeeAp9`HAXdvZ0Mu<(nAJD!t(P?8
zy#fcs_XToaBa5+DE)_XsUdXWkedntJ0T2%71pzGRG>aM2dH%g!C>=ay3=4uRtjLdH
zKEbn!QXq0X0X<<Gy1@#yXlXwdbF`yhby3R7$P0GiZe(Tefw=L<r5D{Y4roIvV<w@`
z<OtG6sX}FJjejm)TSC=v$JJ>m!|&J|Nn@}sLJM&ej)%`MF&i5ELx|e3*cpFJ*=tuA
z2~sZC|5Ex};HfmwHP+$qJ%9E=!}E*}8lS#RHO-$jP4zY<%xwVI;CWec;TPSQ2U&4c
zhkyN^j~K=Wb>G)_&V2lCSm0(q#@x&BvK3~R!dz8STl3^TEGU>GC$7Jz<Sv@sfsL3+
zfnk($Je(M&OI*pvI!X}Jc-QMt<Iwu9<NEV6#wEp4k2Sh8wdt=pwqxJh(_og=>T35K
zEQ7nvh78N4u!(B`)p8eJqg${T`<nM>TE(6|ksMbn+f<s32%j=1NSP<lCL3a1ka;Hb
z1XO9O8l&leNyb)(`7$u@ACdNt@0L?-1HO+GHJl1%MXN`(ib+C{i_whd>V&{;Ok+Tn
zRAq%lZm(#{28a*O2&{@jDwb<jjmU5-h8s+UFfG7Cg8!Vhv;P3<ve>25<e+qVlhv>~
zc|+80tspe(is0!Fz6UbtDD!~BiIBQ7VlRX4E0Tu{*3l`eD?YGapx&!tUjS(hkF%}B
zsJ?vS#7cwIjvPN(165@%Iu=e)T4|ycq&eACWLs&vptDYd)SEMj)i`^0ZBm|IKYO;{
zgjsWEs^4)d^Zk?SUC#=MPUk!&n1tsmQ`Dx)6qI07wOIkZG&okBHETxI370l+J-y)i
zi>K!>sJ||9X7jaI2(n-b`t)JxDD>-7&?<LIuanM64@+;6&P(r+eoXqX^qBNh($7hs
zmHs(O`&PQ14lvsY28MS*DIBn47w_=<gC5?QCcEyd?2cNE2c3bs*4Gm~RoU-V%qyVt
z^AyiLQa&%=Dc@i1b+xX@g_@<b&EXkPfbPIVW16IEnJ~@WWm?oQ77Wuc^q(^gOE(sD
z-PN@JQP&+!n^#p|RlcLDfuhb*Ig}}ZWqwHIm;QYO%Yix^qL;qK?hucC|I+tMKl$HS
zA%DmUj9KB2vlrFbOYdL=%v%;v6M`PZC%gQFEMe)C6<*=*W`$0$7pSc0QpxpF$rV<z
z&q_9VZ6kIoVPDruKbT-eOq6P?muj1`x5q%pGsDC&aKiH3acFez*rx%?bIIeoD_)vb
z>SNw2jm$6W`_3=>`TmTbZx)nMg4FTz&b1NJn4Us%yLTg#WP80`QMvK~Kji={Dkd5(
zUE+zB*LQ+Uz<hQHE$Jv*So}D!b>DXk?GLwJ#JvhUEGomP=HTSf0uCmA5qIeE0E^pQ
zY1^Jv@;rdELMKC+VZqE9N{KW9ce~ONX-zsQodLOchqMKf;|(m^%$uaQO7D~|N*|Oy
zCVdiQ$1g~qlfEGRvh<|%G|CVioyU73<<7|8Niyi9Oi;AB&{*s0Q0+EmoIZ`HjeWJ>
ziq-OGuisL4-LIyD(*#bNBwa5_8IU9J?nRQqvJ{g6z;9Fd>%s3*alT)NUQrdQ8Lc3g
z3xZY{&asmT!_cTQry7=`{6I2PPf<vPgwacXN}?zN+40ifWJfJ3UEmR>s(KWXm;Rjn
zn($x$jiN%4|CYV$%j^|%VZc6n<D>`4vTE`an<Qq1q^!gmD<ohA&IG0ZLin#$zBbsC
z2CGZCgaj&4e#3t90qv2Cr8PmOq3;v^)X6u_&yNm9=$1=`A@cswA*hz+1FjWLAx_M&
z9dy@PVZSj%^el0MPI9U)>=qJxSW36XNr^&7{Y#V+`%9_auNo_P3ORxG{)F_4Ilfhq
zi5=mEu;&gRW0uHp=*YH^#CePDMU6T3*{0H-#~$~TD({rspX0*s0F!Lr4l(tN?b!_k
zLo~)68u0DNlg#mK=&xbRg?0VKV)@n{ek>}L6(aZ|yD?f=`KhRQsC^Z~*{>S5ElEWj
z4Dz}uJudxMySXh~wNvi6&0MZP2C-1`p(8E}vhyBx{&B}XgZs*lu=XE1ws`5KBOTcR
zD_DPa+I8gWglSGxWygKX-aH*;JXD|k?cG3)A_sEO2s_nk&<soy4%OfJ2RB3Uz)Rod
z%>f6i)b}qy-~V}>d;bgf^>nED&%S#7eS4@qK@7bbeg2E3MjY+`5zOYlG^YOpTmFkn
z9OgoK2@mRp3!ixvg^r<p3Eu*TD?1ynj@ZfY=y0bmf|(81%eS#zVy2&r*(v5esQqr)
zBXTI@qU;?x7-3!fY^U>VVOe^TdsBZH1PGdUZ$?-wRb1Rv+}tT#2zNT|-0JjrE`Ba%
zo{=+r_g=s!-jShGaD2DIClYOKn4|L+>E2y-#a3o^W}yJ+aK!^;WMTb7$few6lH)H8
zT@-4@1s?Np)<}F}IrRz0B0=&S9F0!RQ<CjEC3<NqcO_4I5*xBV;iCctAF?HUKj%a`
zTOQRamD<x5n6sy)dvF)B+;)R<t+EIi%t@Lx<H-50elRjBcg9upSN5u3c6z?(2OXi~
z*!uH%oBr5PeMuHbBLF@-{FU2zXSh=K!jvRq$hN!4diCq2M-gdpS9|Z*%gY#;LvFG|
z7lR_LO)jQ7yPRD?ll#oB4nOYfNx>?FSM79$`a|aZY8ykne@tI^V}U}s(Ghxc1=f2M
z?aUF({yL($gNT%I<5lMY36F>8CsJ#6Dm;Dn>3C|^jB1)cb9iaKO6r#{KT%(*_~9LQ
zt&rur?udN5y>xim(I>l-RNzV-ft4C&0=<v<rWGFRy;S%arMSg|Ml814WZp>KEE__p
zpS9vlw}uu;b1BjVsdS#zPi^O}acbr=SGQdP3Ih+P*%C83L+~-IHe=S6t5>fI|Kvt{
z#KR7XsmVd&5yU>Y%p8jh`x!LeKLe7KEk>SUzOv<<LM>RAvYc-*OffK-;QVx;T3qh5
z+wkv^y{PQ8JDv94RAhbz+|L+MhIzln7_e46f`B?Se*2lhpE(z2-Wo8Fu7bqMd^Y}F
z=Fr4EH$q={X2XIc@uJ8}b`W+J=8)sM-9ewD4d|Kp*-8Aih;+;Gaaml?5=h&7rMHwl
z(1)N}bKVT$o#sQcC4gN!!y6<F6E!YYVQkAzklJH(1G%gUk*LjOR==GHzRX>_;fbgK
z5*EVjGFArb#e?1wf5XEnve38A1`C7B(RbnYalsAT0TDR?rE4q`bdPy|3shh(BT%Ve
zpCkDPSyIw0i@Qh45{5|=jJPxG3pO6HKHn0K_Nw~lA|CcH)Xe!_y~aq+G=aoB4peg4
z*<^97#CO7M*2w)cfM%0r>^g%{m+d^Eec9u0D4$cQuP2!!8~-fN6G`S4F39{=MtA9O
zNpv}4=asa|GsBY^)y<(90K46AMYmYrgkqpM5cgEf13!ZMU>UU@!~9zvU>bq0=sN2O
zW}?C!Qtr70y=@U5)4IIdogu=9)FW}4&lYE}3*%-f^ZJ6{o+Kwo;2cr3KqfLZCrk>T
zfu<aB0-~y_XX!RdWm~sAxIFly|A!N4itbfp8RkNj$b`rhPX}!wa@HeLugq6;Gg@~d
zm^gMnrIjB-ELOpjw@TmKgDW@Alzn*uJpq`eu^>Z{KNa1nxt~7v5Vn&pJe=63J8n>S
zJKRA)xSirpdHE<G#GbsC`<cHwd{1!?SyMdE!OVq<0*s}O=iv;kD)X{#3AS3gJfA;`
z5Z(JeI@+4@K^jxu-IixH^^BU=Lm^lXyrXKmW!MPdY#Wvim*vNJbxV(Fns`i-!rcDq
zOLNkibh~ty^jgrI9*`cAwweC)PU({Laaf)GjP#itr8ze;^+y2hS^~{37jR>s?4aE-
z)P*HQhtD)ikdF7@kA0%L!e&@RvhJ~f7}T-gS-1tWZj8$nv8%oUQsEju$CZHkJlkXA
zHxTJAf9&5z(Z(Z>Y#l$|iJ~?;*@2Vo?Py~=iaKG`IUYvQ7JO`Vwm=PQWd#;K1gsSP
z#?npp&`xxGdkYHMfO4;}^NsB`6pQa_hZ|8ChL1#&6z6tD%n@(1m2V0f+BD0Xy8_xf
zTD$L)eyFrdCPj=Ttv<3!M|R;kW=2PLF;+1}gefE#2&wl4+qjA%{(5l=2mF1XF&RUU
zke_%11Cn@IEoi)0ysy^>jb-cN_SP2Ep>r|Xx(qtXg=lMonQJ$ITiaX5!!0;{5zWKx
zXbWnK72b$8;37I6ZEauKXcN7&bpd{M8Onk}wyy9?D73Tj2o%hI!~Vn5P8)c|?agqu
zf%i5T8;5dwFzleA-6E7AEICY!a!W5OL?QtViw7vOqz<Jy_6a4V61ysmRrJJ_sh+BO
zbHViqq02DQ1eR6WR8pvHIxzcaimzLag5j02X$87fiDT1FWcp#tv*6!IP7ogbzjsa$
z?)G@yTI5L-#_v8^gvxu@r(p$HoALWmzl$~5r)HORtd&*x(xB~TkmF}_yeYnI4XY%S
z@2nzn$#kBj1DexxP$KeVy1!x6j7x@b$xQL%>zG@^cNdzvn(B;vO^vDQ#ub^$P1C|U
zka>8#_*FpV#5Uu&V(EcpqU*D)IF{}!m|(#%izX2{;xpU+9n2<><5M#v$}?oNHaFTb
zKPd>~AD^cNy(G(*_{YCENc)s)2Osare~lllgEo`!i}+t)?#Go2#xx8zWnP7%O*-o=
zf!Q4Mj>yMH?|zucl=fzK*0)u~qZ;u7*QriL6ehhz!*K$~G5=6i>QiKHvK5=YLr@|r
zLCrG#$<VD(^-FeW+o3bG-<pE1iT<A*o~ItDn8uwm<TJ3ctKVS1`4&~)uh7?1_$N+&
z{~&#5q4WcO(l~hkDb*$KKispCDOFtU;p&ner@cYHtKVq<Ik&4?kKA+5yYIQ@9S7<=
zyK8qAfBf!y9=x&fy;K_Cqn6nlY^TCJ^5j9<r?fqH@t_T(lxP<1a@}3K8*P`dJ9)o)
zphg+LxA*P?4`nfy|CRJ@mc?8e+$imDJqVcxo&gGf%_XVEa(`hU^PJwrB>=1|cpt!&
zb!o9USQT14CQlsnSg6@txAR|HM#a}P!>GB5@7PXjtJP{dNf6AP4y)Dh^sVqQnd*Mk
zCRM6B-ci4DwrW&lUpFixs2}e4`|n(f)w$;M)^u|YJ~}gNp###%astbB3OV>9{GT!e
z^vvq8wkh}eb0WE3YwTUtXcN?^SV7I#ZDUJk*QR6nb#`qV%erk<1FPxQr-GaC@sS(S
zknS=8n)ZmN3o2`Aue-^uATjd^XbmYS^_j`$wBt8wvrX6c%h|~0c21nwR1!a~SNoL-
zw~$Xf_gR<K7qp99(-_88?5(mENn70(Q@q)$OWFacm#8>VYxvG|b8@C$=VXjkK<4VP
zpD3FrPW&VzsE{Q(aKDY{H2f{QfVzih`2>5zhYGn-y!?-`+&7gr$g>w_witL<1$1p5
zGdi}0*ULwDezdAz9Pc1|ZK$p%3LNuHP8h<UXM!}P2N_nE%$H7U%QAD-?k~&e|2}$?
zAQ~#axsIWC3BSjn`ZGs)x=B9TJic5<BhTuowHZ2bY^r`Jrjv`ShcU-e?joN|s&h-z
zXf>`jYp`O~RK;>EMMZ?1+4WgvVwI$-+g@+t3_jJHCXHjK*5>3$ZZ57ar1HdJ=0#Pw
z(noZDsVUDUc-M<**sRkkbug;Np&!GWmOQt9^NB+xY;^34m>-Pc6GdF%<y~?ZeHuho
z7c-h98u#tnBwc*~_dz)xV0v+eg;I38-zj?M*W5XGx^ZhmBc`Dnb-xmX@;|M60ojme
zKga<JNWJLTmx~VjM$=r0qKBQt)Rdl436^W`?&(v$GI#O;kw?AQuUrE8Ta%7~4uiWa
z%%Pq&9M7sO8j<LGf1!=J-^p)tY0htAHK7$5k<fN9#_CJZ*6LJmvo{T2z*Ra6pVii(
zRuvwYO<5!^5{#j_&(EJi6NR!MT+&npZglF46L>nkHZ`?2jb{^!_4aHH&*HgibuPxU
z+HA)*@B{}KexeL>#Mqh_KG|sLyjyys^fu{&^ik=n(w~kXdhfk|!Y-!B_??xphR30E
zmy9<~vUEbizX}twakDZ9HDr#nV-9jEb}9Baz%qCPQ`~4bbof#p$9&Mke@`tLQxF2L
z32PvWn6iZt@fZ+K^<8JaC9`Wot~k5TEPCyEZtWf3C7Kf%+{ERt4rS33Ce}8GzRE8D
zloB<R<rn!gFUoSL5`4+-KieF06aGx~@BbdUjwR&aJZNT5NKZ;%mVR6ML+O8({zCd&
z>AO54EQ4Jbb4W7v8u2Q10bnrZ(8*Nq^%RuveG0?_YUX>EG}ldEMt6QWLttLrr=cCz
zGp^Gc@`pe{zD~!W`S(s`_t}BF_|8H5HffFTEe`Kj*XZ6Iia$cHVV|q3f3EJ?zxMp<
z(0+s91N(iM`=@|zo5{NjCu&Sf?|o6gNN!#4RBUA_<30*{wgq<8`fl8+-YZVh{n&~O
zXJIT+8BJecYc9kn*^Q-jU>aW*S$X2gQFi6DJtyM5PF<h2eQEs%l2>HD9p+sgk9Ow9
z=^MR%#YlV>zkQZ_(v)M?YGY#7GJ{c}APdBmB3`Y<gE7n4p$5aKC}DS9M<EwwJ~mzr
z-{vp4JT6479AglVT8+n9gT<>#Ypl17($DqTxqagXvS|15izHFxcss11AdfGG{~hF2
zsrVhx{_0Dsc-Ro1a475q{5SReDBP*Gs!Vr;&!V&~aSIO7P{A}W7cu6hSUf-1$A{~k
znS8Etox4*MT^EcMO7(k-;zi1d(mM{qU$Gh(Ta7dP?c-~+ovHP_Jr9u<jKG_uH-i?k
zT!M>49RajTR6Xzt^h>H?ggR437+y9vAaMb)Oqk(Vn`Su|dAt_4A)^JEfhu;tLMwiS
z&d9Q6$TQTUM{BA<)n%#^n4v9M`}N|QD7sFxqf}EjaU5>y6+`(RwN*L;x8C7kTvtOR
zH*(&AUy>@q5)Bkxk?Dr2zClB11}H{3f@Rja8qwJ`h2mv}*6wEsVU3EXZ>W}X99D#i
zEz_oqaEb`Ygx{2W1sG=86wBN2Cg$P5?Mf-j%guls(8@EXF`S`1;qy?U6@5<cbO&iQ
z5$0y#ZH$g-EaOnj7leIo1$n=2nibP*m_|iMU`{m*1C=N;(JSU8ep4|o7vtqlQ>(SC
zS<5j^TbJ8}+O4`4fO@2-hmIbq!IHVM!+&%u0i8)9D1?cu5S@}W%{1Y6RxvaqEg&np
zW)gT_RdgcDYb4Mt&!xKG2!e*MQ%g51Y)vTk)>SEFR=J0OGq^lq7+YQJp9r(f_qo~b
zH8X>)p*c+F`Iq5a?ts=s0H$f8bNn|E9l~HTRxmXR_NL=#4nz3NI89^4qW_ZeQ>tr}
zef1dOm@3=QursV9%*|oG6J-_``qap77<BeK(b2V|5&kkBoEpT?oj`?Rr}zT#<&zA-
zg;+m|J@8OfmUTm}Rx#s>Ts@TE)A62x_i_w?SZYcqr7ey`EYPOd(#(|w+|Z`nOF5*F
zD>Kg?F<6*S(5BpFH*!w&pVj0g8UAS(>W3#}o51Tp+pW%3|DZU%F!anP9c2~=NdUaK
zHc2UT3zf8GN+)Xz#}WpL6Le>z+5BwunMqBB5@1LKcd&_2F4YaOXZx%+93*H{W*ggi
zBr?JQVZB(EW(JtbcEzDyo%LrgpdrO1z)WJJZ6&AsPC9xgd-M+{quFl2pnZ~K3$+rd
zMBa6aiiagh_<k%)e<e*azqM{g3qwjvI~*yE(b;9Wd8;BL0FN<VPj=4f`Z?XL2Qx7Y
zyU4400j%@flQKd0jJs9YQnpn1Z;KlDoK}40v|{Kr&1!aTNdwo=DXr_q=M<d|2qhaJ
zy_5m9V4XOiM9AJ1=`W>O=}ypUr2&h@PkQS)ZF8+-X_s$R!Sa5sm#!i_4F1p3By#c@
zDk-iDt9klF94!1gVfhQz5YX};6U14tBY)1T&j%53A7mpXP>n}rnfP83Bo!I@z83kl
z$$8CkRE}w?I+k`fNJU;x#gbL5IVI16V2EM0AS+hVt0Cls1#%^p$p)cEh@v@Oor0jJ
z$7Uz}fH)efylGoldCk%uqlAiUMYSpXXoGThUhZ$m+HBtzd*1_^_JCfghtqMxbVEP#
za7(VTXFKFJ`$+H{Ij+Zu^ft<I^s*Bi?;Se;$3Keu3aaASRe#Q{W!!~z9b+${f#9yD
zjN|S<kdsmwJHkV>fPP#!{4$4i;2-Z@+Dds$=J;x#RZk&m$SF&`Sy66QujLN7$-Jz!
z76mlHVoTr?tWc?}%Vw75^)as*Tu-t6lxmgUef!=x47<q~kH}EO%sH4lco@xjJR?e~
z?))$(N*D=yF{0SpNZ@_AGgXr1Dfb1$tW;@P6;R7N5}#H1%LF5KikH6hH-&y&0%<4{
z(l5o`7Mga^Cm4o9iu+}Pq_vK&md?K@pc!8JjyU<%5<tUkWv~XR@b%<44*}+Tff=e>
z6z7pa`BMLK4*l?hy+9ydKY|I7`OzMcD$<N}i*!-?WbU<>!8EVIh!yQ9T;h=F9H|k;
zDq1dCx)-Kwv^1SrJDC>2buC1KX^LLh%bjt>J`4`;MN)FK-9$H5P~R>REsxCDXhzG+
zQPa@d_qTQ9oGb@nfN<zxRVJ=aRJaZi2tAD0b+P5pNxKR6+`P)&IFoMLc{>Zi3Z}yC
za4IBO(yljzHS~n5g$ZV_#jLe$A_oMMSEH|^J0lZ0L|`VoLA)9HN3u<bIPIn;^QA5&
zu`uE0x%23`d!7NAh8&Og2h6v7N$gB;JEJS39dU&KZ*`4h33H<&2U^9wegqm7NQ(OE
zfY}U_Z#2|~O{k`kShc)pE%GciCg@y<U%09!t5jAMSLX1$4cDzwOO9b>vvS{DCD_Ca
z-L#=<nr&;E+CZ5~Hv^}2?{Z`-=<Wv+@(~V9`ipc@Q6`xjVz>*-Fw0{d5Jv(i?5;8T
zL0YrS@ElgShgd07grySaIfvmN=r5acqUlMB{vT{te~|puhNHry2;{127!OV#37jRk
z)M(4Fe65{;-V9%9@)1YY>_(uZ4PW^MFEne>0{jB8Q9pcUJq}^Qq%xvBtCe+Cc9vk}
z0#&41+ExQgpQjp>dBhHW%u#|y3V(0+&@mC4FRGc5#4s&I-3+MPO`uV}S$Yp<u-=qA
zt~|)9#*m&-X@MGhC}eom))7Yxn<`G?MDHOd@UitSE<N!4jNIWMu`pV2Krzns^<~m!
zcN4h$O{0-0K2Ueb*-&KX2+UzJR6$SLnr|5`R)r-eIC6S`)uk$Q`8PqAG-}Aa<l#Uz
zJzKBoR`tL0!ina5*Qa#0n)rv3P*dd3hcgHx3%^^4YQJRy8<CaE>t~Jx_7TSKdEE-s
zcB*lH(<Mh%*DETr6keq1;jdIJDBbo<`C~x>ZzpwFuiQmcEl4{-5`hNnA8I%2p{xBH
zd%6nk1FHo`XQp^9@B4<dF5N8MEj=o|vz$|bP4*(qJhB5m*eZ=Pmtphq=^us$&L%ou
z_t&~Q%M*YT%^9JsF<`B^E{D-tAEanohu@2w1`puaP=IsrPf=WZ&p|Zl8m_KrpLZ*o
zqPw@rvTapBYWT~3U>Oz5o^cy)EL$18ym#M9XrWs!>d1Dua&ODobBHM<1&!HMXDwqQ
zNQq}yv#P28uLPPenOPX54*Zv{ADT%$>pxN)dXi)cIXwYuzd7k%=|SM%yQKF^ACrC>
zxX7IyV5Nfw1r`jGBI@BPdN$9oYxo8M3o}(fr~_H7Kpq$|SZ7#<I$tbSwE_RY#7Cod
z23d~!Ze|;KyIGs4nTkynXL@cLb0sv5nt`i6x8uS>?+yfHon6Fu6H8Uvigp2wf|`jp
z7H1KLwDsdl++KNa7Hh|3b8TqL=p2u8`i=$vDb!Z%O2wvzS~CoU!FFNctQ}UVslLs3
zI(8szph{qO9RF>qN!535u!tr@HHGzE9k~|y8z0S#^VSgW7nwgOW`X^f^tiNhAnq6Z
z<faPHd3WE*PHwEJ(owIYgSN)K*XQ)r12ovzkhu(<?DF;7<wH5AInOI2*NXW3{J!lM
z6^7-rSsRu8v1Us#OGBs$rLA#>SCH<$v)hI24|)%*+MW%HAL<WNYg2F$&sEV)mrIng
z|Jcgv13i2U)I)e|3!cWS>Rb$0mqkjV+{TsR-ky|RdjRjTJRc4M?G4i%@4hwGckD?{
z!G>+@Ri`~1#{~BO=M1}4zgrEvRKI*JiMsh3<9-|<{e6*RV(70I@BTwqAS`WrYpc`Y
zf9RXd|7?RSVF*3wSW$L&YpdPne~+-h%%Tt2&;uUAc+3IMU4Tv6EyJOVewG=HgIA!}
zCpgV80DGpJF${ZLWpy^0#+pU){-z#PUC>1A>Ev^c&-T@=^3ivYfzDG^Z{B70v#MoP
znQ6_P_o`@A#T$xj|0*ATAAw2Qtwx5GO#PbgU;?hQD*Mwi7{dEm2{5dn$fAV&hGNmf
zneMEW`dgRfpm`7T^B-O*76=|ApJTB{vZ{AcIXwWo)>4cGU1x!rk>JL)ahzh|mDlhP
zaUuLgnN?{Yw$a0!Te=MjfuF-Pmbumz5(*BF6pM-mrTcS{f*#Ks)|o=599!^lg5}-{
zV4c^5HRuA%D$6y+7OESA?)o6s2K^+}JLw=vTb;BAI^envI&Xhnk2~E#0%}XQmlRXz
zo_%xBi$_FQzOemBlLE1RwGyZ`-;I^9YRL^5{#jMok6mA_1(m7_cWILZ94dZ$pWb;9
z7A$FZxs#@yo#Kb&dN1*Gh6w*lzGDK%F|9QI5lb}jM%q~x<qDr3<RA9K)3ADYO$nt9
zb(vB8_=ES#9>O9G()B@K8^j2dl)@UL*GXIG*QJAo9k1(s1Z3(CI^7ls(Xty@w!Lt`
zuKCp_fsT?^RW+DbmDruE%A6-M+>I60Z>egvGEt?#yriiA{hU+Z7jM8h6&NO%7HPcf
zYl?QU_%*%Wn<i3hFZPB7kqsb|!|0cIXv{drYps;AI|Ju>GH|vsk3#=~xr)Q#W@^A6
zjE7YmS(&$?mpJ|t$ME)9Sw4%;9n0UzG6RpyyxamMWCRa-K)KN$M;U{aDFQRk!HhFZ
zNaCH#sa(#IH(bkq@abW?K3(`odI_dpD0nl!H*Xj^IO7b^KEjudrYb8y%?er<M!9m)
zR##+sMYT)(IJpNudMu3)a)ZaI9fTJ{FTV~a&H*iy0xjo8dBE)uCXB;rkY20I-ZMxC
z7#5taBMAJmJb;nn%IE`{4sy#wx0jurHEs9STV30{t?9R_T4TO0+s&pe*XJ8rwdFSx
z5sUlDAocw;c)^)8-v54M(s|RWVNJDCd1BsOSa9bj<g_(q8LL0c)eDDz2pb>fAjZ5B
zcMf^_r3l`<j9`iwuZgFapC;`MFldw|X2dlkW{%n<y#d-GRuO`L@0?r_um$==I7#Ey
zwCx4T5vrb3n)9{Vd{eo%A5JOin`v0BRH`AJU{DFaYZ$6u^<~#q2?D5ppG{1c3H(5h
z=4SQatP&*3<otvZhxh!j`W4Hfe(WhWibGEgL+Bv^nV>pR1SYS*t(?ZpzQKTnwq`zO
zVtladQP@qoDNCEemJ-o%Lw#4ETFMz!ozgt@3@kR)cd`Uf{|xh;Y{&+#$fE_#ly&W$
zXy87hdfJo<Pg-j5F4g!5?oe&OLO?EKDg+`I?yW5E3y+dvUA5a^?+nfMz4Ap_{~46O
zc;O?Cam3`n$Rzj$jy}Jnnx<_S$2k~&jbD)^c+mc)n6lSko<>BPp@W-Wm2L`qLXMM=
z!YsTZ-OMtsmXSyfaE>17dd#ECp}y$Iw^jrT@fFtxIME#>**4olS0DUHj>Aldam4tn
zrT~|ZPzEIR!mjF!Kb#jT%QtOf*??(AUZ(V$+#Lq@)I@)Y(Y`)d9UK!V7g>TlhDuz{
zwS4K&%w4K)t6R0%jk6oGvm3J<X=$3{E^XAdRNGhYnmII!U&0e#Vkk*fhL|LA8y}Nh
zN2Ll#Nrv`bE<eq+ykjUTHtCc+<STJ$=yjLH>yknLa4Odq*Vh+oml`XL#>&Rg$;qS7
zm)^aS5gu6X{OVwJ)~U3PT^0{+tTdRv$q&BA_;FOKBAz2!nbC<Ibjyq{WyAlUM@Dk=
z)X5`3W#Po>IxGD4qP%z3PoG$*gi9w+9Ziggk@-YHS-#gGU1O26%^{B|6gM1ekzrn-
zqkplPa&c?xDO*2&T(_^XTr6yVC)(QR==PSK?c<}JF_I=h#=(q<M$!(V>+0Hy4AQv0
zuAf113_@XOM}^H7m>^xi4YNsACkrYagmaUgOwFl(q!5+VwJVRC$>Ta9KU&I$2{T8m
zC>-8L$N7;gDB90byEYp-RC8BuqK<rle&hdR?p>fHxy~}dxDj#V#`~F(k*}((%B;++
z>Z<PQs;sQ6>Q+A`S#p=&w4S!w!Y|1WLAMQ-E${f%8ZfMz@i0t#mt~O60Co?|;t{*}
z>>%I_UK^HS+Q2SrA7D5upPk*ooEc_zb@rTLkhK4QZ$xBdR#sI@Z5X8Lc;~(U|KI<<
z|Nr~1l#MGYAs-qA&T1-3MKxuW&Y@bibVNkS=3B@|ELEfqM=mV$*zTtYs0h%SvNB}J
z4RW!8D-H`+Y6`WK#sx-n5@^42<xhY2ASNzje4a<IEK-vmK28JWY2}40kAC?81}4kl
z^o^GTU#6TzN5b580^8`<QxFP4OPSYO67N)$1i#njup^si3NMM>WpUz0%VHl+Zm=+R
zGjdRA?7_sSJVFWQ2-|Q6T@ZCK)FOJAox~W&xJrDABon~zUOu4Z##3ZjF|=3zZz|Jw
z&l-yC)mveVl4MuPwGka1mC0oj(Rn=Ts8v}eP0dg)%C>TP)zuU`oUd5QlUR7_7323t
zY`e0obS%^%m&)1_8b3_Re^{Vh1P31C5<e71kKPFBf1wk0p#hU#S*~4Kz@;{N<$mHG
zMElc9K#w0*iOST)E01FArKhiKeBZv5@0R<<jmmutj_t-o{v=hm5ffrtr8g+-Kr94I
z(ed>gl<^cY4;1etod*dvqmoX@vd?ScUeZ*?s$yY86=$n`@jhZ*9A#e@3rFI(pI~48
z_ah!Jh_LR^BHjWQ46d_r;Yh}7IDgqo%E^tAKULhnyC3;e<^ARTX_xc;I~zs^1*vml
zxxfq3#3uHhRFWf!y&GQrf2pjd)s5S3xJ}v<eHjJ}H&$29sx;zDVYohk${6iiw1qlT
z5<e1N5$KzzPOCp>Z?ug%n3V(T(_XwBRIjFS0(I=QSI}%MEt4brw^%SKgaP77x!XR8
zk#0AWAxv%eDFNp2xc9Yr3}F9Od)_@`C?y`(en08|jFbJO+M2f&<BU6h%W0>Th&q+h
zE9AbiU(2;0xl^V5pZ6zss(?@LN$eZfckIEY<M4G`H>U4zQIzQcT5nM6b!AFySG>L%
z-9MPE19g9Yz7DDZdvm4aJUNn&u?$I>DK5!IddaZI5*)|z9MbXIwT(xeKfNCZX@YJs
zJo}CN|I~SOL(1*UO6lIzDvuL_a+pcyRfCz-m-Yk^{`84B^)id<YHRzp?bhhvxJw^c
za{rw;IsDOIx{2c?ZdIICw#0eM(0pHe3qKl7@oYS1!@ZOK6dYYT;j29yPi)u4MivRI
zyZZuNpT`_Auz9-Gwcd*7_iOYSJbz3fpW=I);`A{^<&WRb4F1_|3*U1(Y1~$@X!}L%
zz;p{*DgzDP4P)%eG~g=M2KU`_6xo7y?5{xcR=MxN2Kr#TJl+U~;pXdb1VJ>@C9Yxj
zdbu@YTeI`l$5i!W*21hFw&p$hF=M741oau?WAZCOO$+Ac0?j$@y2o7_x92Q(J~(DE
zvoc#TnRP51@1^<b>29`EjE{B@87#NP&n{jO3wOs4qVIX}-EXG*rq$Fi>0qjF6JI`=
z?Vx6Q+5|mqNXMkRfCFJ#!wo<LsvEbj5y6>zJXkZD-J}7BTd8R)2M|fd*QH@L`p$Zd
z66RRduxjXx=>}X_4kNTtU1%)z=6aP`9Rn)c^-|p^*(C$cw#SI2*FP=G+rq1XDdjaq
zCoY33y}8<(s|E~h$5>FE>#c6W6p7(ZIWuKPKi*$%*GigcYNc9xxqn=DWaA!~TWzI2
z7xB8Bw_zF2w@UZl9M>a54&fcncbEq^!TKo?npakJQ&*V=72%EY{vx7ULt-DM28F*f
zsu<s;{2!KPv1N){oT-$3j`AV6Kf-|nlga&ttA3GoGa944-mBnTYE*P&rAQt#isb3$
z6F7y+OQ)cZNE=$l9Yh`L6{Vh=9UMSojped6(|EW0lKK3ZkpNGMz`fGA91k5#X&Orr
zHzN)pZ@`VLadfvoSyv|GU<+D~9ZX9eWNUZmi34auEQ1*&Z3ZsH<Y9jzd9a%X?zv-2
z-w?2~WAqLWrZ*XKwAFOIlxbpR|Kvf$cH$Tfcgh&hT*<Cg<$%(D0=}0mHBPaIakak;
zzZE3sZO05pAls*5c-y{}oV50-x8w3zJ6Jv(VI;{MOmV;I0Ll2KuUaOkj|&=PTB?6Q
zA>uwxS7l7?;;znQzj61?wV<U5+Goj4w4UE*8aZ1ujNLcfcs^yg8Ou>@%l-t<cie2R
zF)|Im32xhIL$Ra2@m@k;lWP<FJ$!Tg?HAhSCis0ooos#R8|M4$M7sp(<@Ihw3Iyg<
zf1K_1{%l`G{QDw1?r>jrN;ghzR_A6oPErG-tqzOL0W^pln>Qy8oi;Vksin0)^yYpe
z3^QspL$_acUjBc!)|OhLx;y^zZ%}Fp0*<b99)s7TJP$*TTik<gtO@FI3*%lYg54yH
z)P*Ix=y6>ii{H?zVcBw(1%HgZ(DND7gib8{)FiF0XJJ{ZT2O7YmX7p%s+l?I^X}|?
zZ&^t6FE|?$y8YnI>9!c%PU(;~ZbGxs9${K{rM&TxuPjgN$rQcN{|b5Bj5Kb7<u}Y(
z)(_7U8@+@C5-5iDh)4ELFx&dH=4;PTV(J4+zo1(K-6WJ~n)Y$cvji}_<!PBLL|O}b
zr<U*-2qHvje_GOi^~^H^-MV1uyYg^8{&Bc~+uZlK2KDzY={?dX#&9TczR1O!9=HR`
zkXtkl#Cq`96*u8;vuqPyc=eA5Q7!Sf-C0<`;>(eR%oa61Z_qUs)a!;8l_HHFjCws_
z4|rt{T3UY8yx5c1KEEcz8CuH|s`e7R;znJs{$Jd*iyp!_FuKOQ+B5<QZR?imC5CS4
zhN0s>xJW$J(i6+`1gLf4tg7;{W3sBI&$c!7qCl<@<6O8lsd^S2jnuQ~5p-6>Khwr!
z`a$U#>4&BN`mal_V&i&G31jrYNxYAC+<(I&4tex~OOt}`Z*5)R?os@()#-d!gna=~
z5+9!s`Lq0~h$OL{wxfqg41K|!()nyJpSK*m&FII7D!{O{fJ@#^@U|sG5?{wg<A^ZJ
z=_UpD>$q@@W`ea+U>l|tmZ~$AIG8of*&wdWR7;^{8g@{sHH0}~5{LLMj1#zjnaYO=
zIV{uc{2djlVaf44r(}7NZMLGQW!ll`8L*i>rE_l{xnuu);Tl=aaMV^LD(RBTBR?OK
zHl>r&ZPI<xV~9dA?5~%Wp_yol3&xMVNl;?(Q7oXp@gIJVFo)@{#@-D22o2uuiFZ$+
z`^#<AoW-+;2Xzd9`F9-OceX2)FbJxe!)#4)YxroGB!TZYG#YAJFoREirC#%c`b_;_
zjG-}yT}r!9#cbN`nvHIscBlpack1;{-Rww`Q;>nhadcbKqI3lMkK3ekAcZHSivlNd
zqb->4=cej$f5<qWt1>Ho!TKY96K)3*_c(z3dI^yq;lo`597lZ~J_h_Sltmxd4l5N?
z5v+p_RI*q11VRFVriHW(PoaUf;xXBcqdmXa?P_(Knp)R1K^D~a6MVA8CE}wz>YYw!
zu3rE7F#(PPK1+VO=8EzPxQ#PLRX)q{_nwg6FMUY*nDnF4=cF&9#WV7n3`QnZ!@_mU
z<uMLB&L0)`3K8{As7vwbFol_)u2)HlL7x=f4ms$B0N&(wUB#~AUJzVM_@~?>PSEAw
zGFrAh<OD(Aa1Hp2P~Yg1xCxjw^XXI#wE*5MYgWVRa9!;{gQVifLHkq|I#BzQdflf~
z)9{}-@~fvD=fY^E`Nd8rrr6HmLFR=rx>x4rDE&;P&-EI-ozt|scrXtGOe(_ww7WIJ
zJd=V#c?oN{!_u~NJIrn$mEIw}5BinopuBz(%IkkXB#DuZ7H#-gu?(jc$>bNjcnbBP
z_$%=?DZcF&%12bCU68Ef$UX%uo?D@KQ1{bPVWJ3Y#ga;&RiMEB+n`WlRVXk$<ONpt
z@)FbF_}Q$`)bmrlx?kiq#mY;@$%`|XD6`R<@~HI8I+ottTwZ!{R(Nxcb8=LC*G7s_
zz-Jn*H3{7N`V`drLR_eEMEjp=oHzl06QK1&eAbvqH`zPg@ic!4e>^2csff>yn7cjR
zmhuL4ct9KYgOFx*pXThPtB}F1`!!WJzzdhqI4@tnc&Kq`f|rHHlTYr!3+Lvb_WS}Q
zxv&o}*CGAJLpRJzWeP6|Hij@89cSik(MDq<TAg5LK0bQ<_|bS-8sjO$L1R6}V|pCK
z)Yn#5G4&(+@bxd3mw8s-w=Zi28Z6M|Wt3|>?%$xEe44AXwx6bO-F_XqvN%P)Z4Cdz
zv(^}D4XzO2*l?n9C5kLB$SVAkm84--iPJEdIwAj%3Q!$NF5co)iCcF~eY`BH(h^59
zk2&@*WW$FKo2d&lnzSDz6_uF{hg8kR7)7o9#s`sZ0x6{G+;S-=@ICtptthjlDKd+3
z0L^g!w>=n>-?-2|F`d>)p-m%is?7Gv^&px3>~uQzQ?1u8Gv&4a{MvseYh(>Z+Z|rt
zj*ycwx`XxFIrbFHG)~GG8<EVAHUbB;LAMsf!*e?`cWbO6CIWAs**Q0igW8?U2<^Di
zPjBxcq^`&?#8*l89qEtWtVQPPs;g_2ij?ma9_IVzGtxTlNDYc8D9KRF2D?2NH@QGa
zMq&XTYsdzpzUlNuD>B&evKt;feWoW)wbMDJ%}QYt{4*DIcXic_v|DOtP9F{3wm6kf
z$vN3`K&G0@32B@|{Yw2c6OK_Yz}c2b3>UMq)`P(~4j<9F>^})fRL@rCo_fG`s^zVX
znVmOp#laV>TBqT-W+uxsPY>>XfF@qK?rp%MtvB!7XOV@_bu72w7m4{FlKO~pI*wp7
zK>{m~m(v2H(IW4+adLX&oNrX-D5vYNWff$4J5Bt!?)Ao{y7Ba&EJzgbMtA8`2d&=G
zv8{jumZYR1)C!dADUu8btL~vt)i>*UWEB)jzA=4TDeBX>Ek;k{dYp#++#F6Ju?Hlr
z6gA2Yzsnkx#b(7a)nK-ZJ`;z4+HL7gY1@?y6H#pKAvU^D?}~u^gBUu-^7;@|ZJkF)
zKsTdOGUUr8={MN0_3yaDG*oJkJMQS~dS5qTLT?#{iMJK~2Mt|S75=JS(oId_@35_d
z-#y|}MZZma$Z*t3N@2MEA^!Cjd|jcBiFd1p&0hgKMjk^5;|#Qbqhrz;=}zfh5la~u
zC>ICTNO-SEarSUHE$|QsgbSj_Qq~2_|655n2&JZ_8>XonRQ;zTPD}&tH+Kac>K{8+
zI(Dot%Xh#L?jPhg|19J25HhEyT=AQaPDUL{Q^<dtCWJ-G@^ZDL2*wY0N#|iM`aYb`
z6lNvDDlzU&3HAdVBdkTQk2?U^-gfy?Skgw&h8RBuHOMC}#hGZf8`K{BrT{!#L#G|@
z)8ip?;Nl<J9<d3`MQl6rDlS~>wr<)bukVz%P6ji+KNA=;2#XNdf8Vrh+ropVJFM}!
z$rw5hD8FVKR0G*5it3xTWx#ZS>as$8(>Z(kv}Kh|KMZ|51a9os&W&QCu1JOy@=@0D
z0>Vl^&aqZ9>)c(vWp{i9@T#ye(8Z}P%sa?fU-`?#(z%ObS2the;<2{Gceu#so93^N
z&ln2OVR_7K99vX5vW7r$oS<SF2bm+{LH+%u0(qT-xtSPL=EPo_;0z5Ztn2%Y0Y>f6
z9JJUQ(rM|O^nmnU=>t42*6Y{8vB`I?kq0XLu%E=N-8+5*n&}ATE~g4w-$TiNO$2xP
z{u|cLTCV8>2lRAd4|RWnhnyN4Z}xVLH+yUx;u53x4iG<Ve6;z6Y#Fsts)xk4C+Dp*
zF!t*Z75>Rey%DXhS_Qv15o-%Q)X|%87g{Lh+uJ{aTkm?k)W}J&{phZJ0<>L;8d(?C
zfuU$4uvWjmPc*j4b%ALz8gT5G_F!Xk3@3D4Mju7O3Up422nksoqivde#&b9!P-~`l
zGzAJTOvq>9Z&p4?6(X#l19p;*zq_jwI`ofajg(}0TTzH(n|7#clR*IMva%_wrmB>g
zR<>7GbX#RpWVBay2gvBN-Yx>F`-tgUZ8tTaI!I21*o3`ZaS4Zl6cIX!Lj{ZNqv3eC
z!JpVo=R`O`GKI7$a!RJ>c2HgWsonW|-(@+m6pI&eKM;$&#XSy=Wvkyjz!=bSgtX^d
z=&sGzpOYa%cnpe5pyE2=3$S0)G;;^zddZ(u6-$5hRozn5cQVIdhOV<xP^kpaDEk#X
zbb{cwtJPmXn5sW!Xu`+NxNhS6olaBhXohBL9rzFF)N?v@7^PQ$(gi+N<jBZ3K+|A4
zJKA6sy$JjB{iN(EPfnxJ^@C~r`~yNhNjQ8Q9^mpTQz&naGxrDZ<sb9l(r_YD@Y<`d
zy~=Ie1?EasYJ$EThPm3dbV|CN$FF^iW1)Q%G&td=8;n~i^MFolW7yLW4W!3!OK|AH
zYktQxxI9Jpiyr=@$GBlg!hhe}#6%Q61Awo3{5?WuBcuylV>G5x!iHiR-~E763iQ$i
ze*EXQ=2EL9yQ=ES|GOTPl=3~|#)Rj#{oztcE|(v+eGuHDZ`*G8b$<PyY>T?u=lR?F
zW|+%lOP^VAs7YOFoGd9`5GhgkrunV;$X;)p-fk%~^;xpW)71F!WT^}kyva0`q%4~s
z;^}?HG+|<|m`6BmcL5oydTSw9L4n1|6mMVs4{S!6(Jh&6sHG6{3=UIAF@Os5FXp#y
zn_ag9GCMrI9Y!O2298M^2|ppRs2{hpiVD~L5pEe=rGriETaOU!N3_RrjP>|=Lknd&
z)Qt08+`45K&YKvsC8XwgW?H1F=_Vh{g?dHN<SM%$c9J6ow@L+OC#EW=Yeh&;U)ti9
z?eXiRh=b}g40_z|4KE)^8jJ^@b=_~i^exvV!Ttn7QW&NALI(sV<TeL%HYerQ-%~=}
zvj&AuutSZrsYH%i`i^d1x}>ehOj%|}xeQ3U*Ui&DzMDPFEEmM(?gfKJ$ej5&@bb8c
zHxV@-f3ZNe{#43RzK@P+FYpzc+@c_Druj^%oi^5+(6qFn${;?+Wk`OMS2xe;;ziH9
zc*%7yU3&_%GIgFWd+`FR%1FRgCkh{rqO4y>na!ZbM6qQ1)64MkF1Y~>Pc@rgBf+HP
z;KQ9xcV{%$L~IESX50a^q)llJX53=D6Spxl;FyW#6G!5`n1`(A8(OnOfM<vS_{2Ws
z1*&#e&vaFK@#4iZt8f#-bpz^?&+#(!1>Jtz1;m2D!?#r0=}@&1b;B8zcDq!a`PCFE
z;`eO$bSto)vjTod@gZwGwt6I$p`C0<D^LfH@s$t0u7|`iVfMg=Fg0i=tnZONYeBmv
zNM;;}TQHe&mO4A(WNY|G!JMK1VIALo&hFQ0XWH$H?e>{kt*a>Ox`hnDt0y^&$lHg}
zNd9F!kcD851lGH{U$y;r)!x!p+OxF?$nfp2cHd<Mp9e;FfW==&6a1I)fMxL7Yn-0|
zq<<J{@@>%8Kg@aQO)#^buy%}pXyAmKB={$=J9aV0k#`0HZg?=xoD(SObE+01caEZj
zwb^#$+s{ToYsmS2UQryuW~ix`Rl0g+l1V3!sruyzxZEQA_rTn7o&a!vhB(M@^ktP^
zhf?ia#6taVMl7duMg}~GP`8$)b!n6HeLh1f;^?xa6s%+%0!u7!aOP4hy+mw__fZPa
zv%~ax5*)mXoPr*u_$4#xeyFmbsBZ)ALMTJy;>A_aP%O@dP)a-?>=CN!?{VlBBSlyH
zJyltdw~(_=L#11w+VJ}>FnJ|a(!j#86ps5=Nkb|@E$&LI(0^Z$E{Rx`Y>hh`Z61-)
zFmaRR?nh+=lXgj&Wf#M>G`OpVCT=)RJSn4@U%8K15^!edLW$5I!y_Wf1tgDgR)(A1
z-;Eb%3`Pt?B~*(W9?V<mC-wG%?{yd4V=`3;r7~$~4pS^gSFU)b-JYH6%^h1v%HiUi
zZs{;}RFVZxC-QH|_2r|BWu?Bnwiy2jL*QhY83vPapBXXfE2=ZM6vI<P^=1+S(l+&<
z@_ci)<=ah+3PKUez9y^CSF4tNPBkq*TD0BLLM2*>qlD^KMJ2?HBUL%>S1q+PJ6E-7
zY4*L0$%KL4D9kjNqN-+TH1`qllUAq1PhtaskDf0kq^RjBbovOx<PZ9$+84%Debww!
z(;P)u`ti<;MwQM?9?NN_qfm_>E@-rXRi8$h8V7+dtZJnA@CTvR{{bSkkJ0n|-pKVA
zIBC=9+n$PGKkj3=G8rxR0D%=3v>)3gvcd|sdt)*av7V0Y8;eNLvo}%GV?cdBrIYkc
zj{s3XuD=KMuarIHCuq&_<)nYdZMmn*WkjODcKt!!`;hK=`iDII2et2{fxk}NMh#<l
zTZ%=0+Yk>Ygl(i!$fJ!>7KXXD?4eD#jLi~5<yj%}J<kfb{q)mMf9C0@zq4m{oS693
zsgIpHHC}%@Q|eEbbDM*uh!Pkn<Hr##u#b2IOe@rvTrEcum8QjXNOBw37KgPKAe#r&
zj*qALL7AE#-J@PimL*UV>>JMk>%y+c(Nmc()_{LnV85vQ`|~e!{^-6sznOyt6=c($
zuIbPrjYLBBQSwQdNU3g1e-3&V=V)T%d=I`ubt-fjE?&y@_(yYf%G;DQoyuM*^%BMt
z<Jt-rpr+H(`%zHLgWB~%!V~4g^Bk$gW4qeYX?i-X`cG-s56JZ(&D~AQmA|oe{Q%O7
zTD`mbnX23S%IpT(^`6l2X^Jjdg6Y~m^WAIL#hRRG*E>^5D3D0nZ=_*Y3VHB${lI*n
zX(sWokXzoizxwji&8H8@GX5Ag><72&Vt<9jg`TW98v2z!R8Imj-Zc=~)voRpt7q5H
z=b+m0D_rdaEY9VF>cxnR>9t|sc6}=I#TxLTBKzZmntk}ck#;>B#*Ak^UDLmZ(gGbf
z)vk}mo_Vnhc7%S3QNs<l>(f$#amZL=SN3VwIiBGmD3>|uQH)ESY@*9jTw*Z2apiWa
zr=e}mU;{B)PZIQJQey7AB{o9K!!ky#-xR?!dKl9laXL9v33@BEd2IMcs&Vsp62-)F
zs&hdlZk{NYV&dwtJy)ZKRdWrmWSG>}87;-WhhUbLVcQBkGYIib<}t-!x}%wfR}aaG
zqr~N?-!RNl63kT{TaA;bU$@Na@si_QHZ*22Lo-a3D$LR`R<x#;T`SZWg*lOIP<W*h
zcm<OmBX-T)ZsRbI4}vp8#HBqc-N7+E#dh^^l8oBLK^pjT2Gl8aM#Z_maH3x(y#f>!
z6A;@cM6A%%k^NLyiz27#<Ea)loe(bO9C&mO9{v?Rw&!-gz1N24-#7Mk=v;@p(An$j
z>ti{M)x*kUJuECo6l>qZh#hkY4~42{pljSEJWSNL?*8>H%>Wn6rq{Bklr*zT%JeGs
zhl<D-4%u9QL0e}d{tVxz;R}E3L{*gQ2wMXOxYpsU;h`8aOxQI~n1PRyD3?ZI7}l~p
zwPL)5cV5b5Gd7+S@u3fnsW3+Gw1^37BVP9ijqvbDTvFc7d<*`^7`nyu3u61j@!x1E
zCU8wMxE;j0u!D#&dKPZ*dxS?A>BnvE2G$-3;FXi4%cj0S7?+6`--0TQn9SA@r8Zwv
zyImDd%0D@81wjIDj)~iUB3{DBK76}9_9<a0rgMm0ON87<I|ANW-0;Uc0-+@e;y4KT
zLCWkrkUzw1-IQ*X9+BRK_J}zwU%o7@C8O!&Mt8Uyajc~ci(e$W)4<p&g)gYOlL9%=
znLOzEM*1N2Kp#EbDfm>UAyJ;>r|0p-^V-}f^vJWSc=&l9MP>B-g~IoQ7JiV*p*AUp
zNN6eg_KjT4$NcPB(J9?cwV4&a;QPuvlE7e(G`=<>>+^I98_$zzd{DCc7o^H%N@(VV
zzM96gFL<}_D49^_37sa@(Hvi_Ss#<m<+G#|Y7QqzF-C{faMd%dwFo>l+PK-k@Q@<t
z;>1GxHOQ@llpIkt48ROCpwM~53H>szN2o@6Yh(-P0S`jH#UsIef~htCIlrbdXv-bE
zIm;P0(?{LHy1V?^D||iku=F13e}?v4ghA_d2RtBH4`Z<~^hd_Nm^&DvyE633+-tc>
z*2^N+RSuD{KC2|<{u(+a;u0suJ6=QYM07tq1>E;H`IENBT%|3bHCh5pqjf~a?w>m|
z4*DBBJQ@6u=h7SS(SryL41X-v`sjHN*+AZMoJQR&lVxbIV_T7l?CPpU;Ygv6mNhl7
z<Bix3R81xX3ec6Qrs@ulc4c7@Us*L&)i7lY0IVs@(^cKlojC`Nst%(EXzXO7$fiLR
zqYA?V4So<<w4|9;ODj<;FjN_a4ZaS;0Zk!9bu~+onX2kKhN7d2YXr-o;bPP=qB5PT
z406_7c&TWn8|&wps@OiI3XyUAf$-wC8Q5mTY@0=<9U3-4(=VnOibAPx$#AqFSE|UG
zhH<)pa(JX*AY&Y`s0zji1GeCWrNC=jfoC?nR4n*JS542dbrl{OK#ztAD00#;nUCXk
zPd6JCd<049427epx@~!;n)R8e_n(razL@!-AO1i@mpjF85f=fDO@bTH;<k%3qF#5i
z+>{5sJcdFJpzM(tHKgouAAgM-o4}A%V9<zL47Qt;weG4MCp_{l3?*UqWTVG^?X*D{
z2%E+jx>=0_s_3+J`cxYvt*Svx4H{FTo3e4bhT9!B)NThmH29%x>XW}a-AaF_Vg%^?
zmlznzh6-49(80fYuG3aju3kJ!xvfG<veD@nM8WlQ8$*!W0pbg{JLjsnP2yBl{&OS&
ze~!8B^=qI`mPbFw@XGk}yr1S}@TcQtkcO=pq}a)O6iHq(7@>WlfDk6FDI>~Hv7n|G
z*hIliFTaUmoK~DXz5y>)J~x<Bx;6^0Rxt<+Z)iBJs6$_aQ#(o8ATZ4j#tI$<Y6L2e
z6K_<60!*lZIEoz9V2Vya1~TIrU`%+`52vZfC=YBi98?&^87WN3<^^AOTacFKo72?C
zq)39{I0goxr{=gaK@+AbZPVv@x7<3{D!a5>xprl>Th$6VA<?6(iki1OtAZkq`aoU!
zERQCa58(<**_*D0gG>Ptd?;%ghI^=5GF7=^P^a>wHnEZZU><RDRrKX*rcAx+6a{M6
zYN$DMygaMD2{nP=?pBR@y*2fxqV!wcPUm^y-}!v535|+3t?<1S52i+@JYEJ~rLnt!
zF@@b_x_1Iuv3>kVxqPIY`b0>7MI3IEo^afq_F$O$E%!Z=dXBeQFI!n0;e7L{-v4`U
zZS{_|^-aChO6Fr-wTb;SFj>Z6R~?u#bvP8=HRpB`IBz1)R^RcCs<*iKOE^^dN{scR
z3*B9pgQQ4BdKmWuMtWFILsB4ZFM58hK*+kF?fKpG{ATr?@2q-DOTS>hiTLpEw+^I7
z$mhsj^6~NoK-6&&2?d+6y~KmzPWBd06N~sO95iN(oL%xcvZs95bb9s@&p17k;$eHy
zBRZm*^l}=(<QD18g`I+lqA9Lk6e0Pb1<0;hdkpFPEDV7;fDYscq=6d8JE_lcde}nB
zcto7X`N8C74HORAcksc_PL%0?Cl({O(i7op&XgR<u+P)BjBz}gqa0~tlAk=>)NpM2
zl8yK`xD`VQrmmyw^!XHa&Wz|aM_i0^#u>$!J?hVKc%qJinye}19~V>~5)0?s>B3^l
zTdFIs67(vb6R{yuyL~jMfJr8wX02m?+snNY(C~>3hI`enKlB0;5uI8VG%JQ1mRhET
zQT|NNG;0w_Dxcb2*S_eQDlK`M9yTM(8Pxk`V8Y-1@3_wogCCe)@iN*`&zpkwC?$;@
z#KT^+{~6Lp_F>?oyJ;uRJMRZ7|7^rUXpUI;`KiQD%Kg;qkoz9B_Qu3MUWfM;`(t}l
z=t;SsdL44#gZ3K}`#A0Ui+yoA@zdL)jHq-8E2HXfxGlmZgUQ<K?B5ja(H4#KlS}z;
zv@JS2tz}jBZH%PMCb7oXS#gcG54HMk>9KtEC|wiUBV5->vUeb_y?^t3x~?NuZ0=P>
zY~H8E)1MTluZWT1ze!($Y}M<??{yW?ke<WKu|tC|xQ)Oc`BfpHZ(+GC_xr{;bwiAi
zJKVsf6)hHcZd$y?chT1nYbVAD5X-Diu{2W)S6qQjxQqBk`Z6<AZU2I)nJ__D6=o`|
z)~Oj7x|?YOnS6kdj}X#UWciXTKT1eI;=;8m-pfBWm%va_mUW#PHgo(^qfzn$LY%Ut
zJ`Ufg<imu*uT=RlqR0U$;SD8^!HYna#rx4Xfbls%Ak9mMxIaYZq{qgBgbM4@QAA)|
z=N*d*dQJGKt%`-%RUUeZ?^kxpb8tOZ?s91P1w6bYT;V<?@VI|}w5Yw~$M9wOPW&zi
z@MB3D#azqc7O$mfq7$p^gGo9waAPJKyIGPH(Z+M)NA7kbqIX32I1R3mdS6g_Pf9-=
z_E?pFR77b@kc+u^I|Uj~IR}m;R`5L+_35m1uk_Z*`ZQ|ux*KB)u47FG-&Nsm&!}eX
zh<s9>m0cO}2St3cayo&!OJPcY%|)QhP<0mTz^qDNplTM)#FaJq9#xT<X(*@V_br}m
zxc!~`PRzjZ6N3*Ae6@1)#EGL773-P6|5UYk1(MS#9N69YJPl9hW%*XcFk$kk+%Ie2
zW{^^r-owpi<K*@s*FCg-GPTL<6T2Xb^U{;4&DG7x0^#Ik=JM4~4WSoXuAK0QbxC;>
zEj{0uWd7~i1CN}IU2WbU+&1v%HT~+`c1Jgr&>Ef_TA^a<o$Z&zkChpf&dG9!?JUUY
zZbgKP5vT9rBI5R%_bK%5m7N9e#KVUIn+5^3gUUae-8eO?GbIcarq7<*m`&?IUHUf6
zXy&AC=~n4M9%nk$^%it|0}cd+2uAr<KiB=CFwBKd`l6i`0e(gQ2~+JP+p%n?c6}lw
zCr-q9u(aVx@{>WS6queuDixx5vSC}1WqrVORG7GTV5lHFx={(ju&kNr;QxG3>BeOv
zu-FnM(9k;Ss;2r{`>bNWwNi3QiNV$`d%96_EZcHjM|Yf(uGwb--*ZgH48QJ&%w{Gt
zEW2ixDkWFb9rzN2YO5D0p{q1dwT1gB$?+y|53(*D=3`E=<tVlv6I21ddNJI@^5<dh
z(|g?TW<zY*aXBO1<p#bByPXZ_X9xG<(vAAiEsE^RraF5bR~uBi^mt2E6gg?lo~|%k
zW%WB-ojO%CtGI>wEtSY~A&=i#T9EZFm6g!62sL2ZZ^ep1mw_pw%a#urG2K@zQz^uX
z^&}~nu+}ITWUxHIYH@E9_wuA+koKtUxETAjRPJBBgxihzFC(JyhAbMTJGM{SJa{*A
z>IUW>?PIFP0mnJAcznYI_hN?w;TtyW4>&+Vj|UME|NgB0AM3Rr`k@cU_r*W{WbkDF
zS?eP+m+9sDeeZkUef8R(`0w*q9*Q5lZ>zHPaJ|+T+4_(pk}nYZJ0jA8=!ZGjK;}Bi
z9S-T_YhuhP%8ya{V~R5OPUf=`zZ4IMHvaBupD_6j8UBF#HC6o!ae0TL+yO88{^;6Y
zS5=9mzH)7D#i1btaCh^R3~ih&@Nt2R+cblCXJL5omL<hjmfor<>RFE%+5@J{*dwQ?
zO`Q8R?S6;Y^wcAa$q2S2r<k5?9#K%@!wpEWlqSjABCCzfVZs;l@n)68?-V&GR1rBm
zpil;RgL+?)<=Y*i4)3CJ`*%flR98iQ%TR+JIR)ck@`FTfpVQP0lhE6~!tGQF{H-8|
zWfbWqMz86v%Fss382N?``IA#Tz#-H45+ct&<y?C!r|T^YIKMk5qb<|BhAMGxSIeA=
z-O}jf#N0-SxuI(3+KMFN4offd*te&oxA3^4O&BQ%9dJn&{8lKyNg{kd_kMqhY+#p^
zst~>%a4Q%b<PASdOya1ay+G-L9jeaw?JYzj40?%JQDW#e*!b;dR6-TglV#6T&|*b3
zU0H_d)K}ogPN>?%l<iSkwrmo(buLd$E4y8S6f8(hfut?Bc!O={Yp-$Jmky7{@+OeX
ztULmuVaaAPOi^B4xh*0y_Y&?4!P14>@=%PZTG~*I7B@CE^U(2m-h|$F>ymQ6<#bxB
z^M+rx2E+{G$dMa%#WX5*WHLQ8LVAB02TSFy9Rx63dT;Y+oj0AG)4h2{Het}#O6=G-
zI#y^ax^Kp+S*e(+u7)8~bn7F~vNV&#5mo~E*!?iHE$A;R1u^t>++io(utl3MK1vjD
z1zpxo_p8y9d|>HR{c}tImhGQKYb4+PfMvW@_wDN!E&Ty#?!Q;oi2p6!BHvM|ZuxOw
zU$=ezt%mghJ0Q>NmIhxaN)ZPM>DUl4CV0D<5rY%ENlIAO*b9Gl4E4&tZrFZG*bC!S
znS9!F-u9nVwdbgMA6fD3oVv~!aeP4Yq;H?6D!otDGMh4P7ZC9140*er(vy+H1+$7i
zHy^Nd%=Ut%e-F${P!?5R<e|emKjmLmM*9nxcAF!3?Unj2W{Wv}lh5cLME>Yys@^}I
zUx6j<ezg!ML)cq3r`TIg^j)J~?^(=s3lAeR%YTqQfy2mdAWM5S_d&>0JIJzfl*bFx
z?JBR2GK{7h7Z>fg8k2ThlO1zmAo4<<bzyYlP3KS8b1m&_Gd#iRbd#<^`qNFiMpTV#
znEo98_XNJsm<QuU{0tA@5hpy1>>~eMl=pBFOL{^iXGg!~%EPgnuV*RS&!$f<^7^6X
z^UgIM_vkLJ#c>-=7RHWQ251A(9u$t*Y#Fy^^VpP3+?TXVez^lJm^V7W#M3K#macS0
zhq7;oX=FN?OoLdhz;n}~9Wy&b9uDWIW15}(;FZs&mR1k*v;RSU1MTD}4m-~MYrxxE
z3z$6#UO+*okUc<cTHrA{IIyu}Rzzri(jSbDlcK*W2)D&)F^#OnIqs^!`{+>i&-o|?
zwVNOALiOK4^NQ=Dq2EE1Fh2=aQ~th)wdr7igckb-B6EL(Sy^401sN=*?K8Lg!Z3)~
z2?7h)*FndQXauZ_xNCetF3t=;kBhy|XSrq1{GdS>j<tHH=d3DYilwM^JTNxXMU)yf
zqAEvX=ZkDBRM$#$@s``R<_^8-t{KMAKB0z-qm0#PNz>_jPV=R$O_gSTNw2+v7#=XD
zJzm(~DA+9J%b{@@M$KX`mT#vr=&BW592gF8nXxcC!d;|o4iB%=PPPOF)Zijv9rr*P
zD|tq%e{5^9#AerEM%!riXJ`8-`^{$mtN8UMF2<=>`J$XfYw%O1DK(WdUf$kbo1yWc
z;qmo3)uQvw{)vs**^LwZW>#-sO|gTei1mj_e~tCp$?#*3Y%DBn)M~vmXUWPh+;-c>
z#wR~HHYeHMZM(@@aVa%j;{KWKR&RjI-h+)a;t#s$t|##kGP3M^f552Y>n<}E=V$6p
z(=O?BR(CX`J6Eq*jiR~dyDXTv1E0*a!+O(RV}{Kr7M5Xqjiz~2W25#nl3wO>n+}e`
z`$@moW{7bvT&zkkT#PyNC|XNl9h)I1!dt`2>EzT-ZL|IZn#m}gzjdJmL*K9&o*fKc
zc;R#R-uuIZGJMr?LlWyM(`}CgHYx7oD;#}8>deRUG#(}=2wafrtA9zfSO4<zWjz1e
z_T0~%JNI)x+|Jrs-aog*P6x<M^pKl~g$SeEq$>dSHfLeT{w=R&I)Os9a~f3w$2+_f
z*_}jlTZdY%7B(XiHqY8tqed-If<|bR53MvvX|YmSERmquZiZrR@K=0pfGE*Vh<W-9
z4x@Wr?kGD%w_6x%uomJRZnmMqp)E^-<%r8!AV*wuJ%ZD89&Vz=&CWE?Y#6eqVa-5I
zWi(&E8Z>+&D^A?D!tNo5N6Ccdfsp$M30)bcIf`C&@S|YHcNE2MRCK1wmVS?pf#g{E
z$h_q(B&58svQQ@Rd?=R}Vh@HNjew~dq*Gh!mGnSH<E<zzA<jsnNzjN(Htk?$zTpwa
z&>{Vo9GGoGOUx|CFy;=g%*s}|6UUvhM&(Lpsa?v-LMC(A@12?e=sEJJ7HW^GsyKb&
z8OENWHWjCQJRsIa+f#5-ah*1s$F%q)W%!R|Q+|k$hs5cFs%_iqgPQhWc0poQ{fMF|
z9}%a&ix|_->Y@H@c9Hkh+_n~RBvA6{KtW554O_zakPjesO6^jqz1S{!o^jMkN{X(O
z66dJld8Ia~mgY<RuQ%&kD{hjgYLd7smOtB@FXdQ4HJI0)1x@AAySf^5x#*O>iPJH3
zI&U#P598qz*196sLXZVPn8w!7KQO>>2+6!3#jB0RY8?6V2@jnhsDinL3uG;Zu8q*z
zwi_bdF3dTc<SVTZhu`z%pdQ65b8{<kR1eDYz8(G=C%T-NoNC+f5v|ZSY|fPl>Y5bl
z+zQ8L#af0A<igN$SQgC}sxU_17*)p~b{D%9v?Z*$6__EpHMA_ObQj%GeSFk6-B!u1
zIJ#f)YBMu6uj1=Y#Vxg5(@!g9t}AV*e}|+;xX%ZVUc*~VSMKGN>J))GHQ2!UIhP(5
zXDO?oSiG_5;nUg)JxMP$KSdc|1%eW{Mk|IFG#5e_Diu4&pi;R(XEyYV3az<GpzI{N
zRhn;GGDhu_L-U-inKDr{O}5(eB}-3e#VPwi>{Ot$Y*f93mYq^Lnu{qbYp!X!TA9&!
zE-IItGEKZ{gMP4Nd#xGUa1=;JG3Pv}=~igMHBlVHo@se@2@*-F)x;^M`yc7LKu=nh
zhA_grH$yzgP~Jdm()hD&SH=^MgaTh61QR<#GK=rQaO4x6U2~|tOMDBiVFfUVF-$`{
zM<|?NloMDpTixy!w|>^OBf`pPm2YOj_yE1m)YZVcW_SuUG!acvGpOPj*Q}t^SxvY6
zMw@@dD@iJ>i?)HcRjHqjUAt+&$qDU?<9a&##C}v@AjXWNIrQVpFoMK^u?a-Ih|y&6
z=4{#968MJCMOP#2CpnTs0ELJ9iK@U9fx#`V3x4B8_-}*}d01MIR-{eoG)8!ivO(o$
zI%4N8Hf?cCy{r4fzL-U^tjot~B7zSB-3Tyoq3m^2DlydUix~QaQyJ}W_{9hLb_FiV
zBRa%Yv1%%yC)6>536NXkNE9!=fCo(j!nIUK4)ZxtPr6&WU;5sO9g6<mH$9x=r*JWe
zCS;DtO+B<HPff|&43V%#GaW9HOJbj3N$gS_5_ifOm6FgeTZUo>)()bis<L5~{ZLa4
zP$o;ih~s29T;E4K;OYV>7tAirkm@kH)(n*}lhV+HnJ;M8R_7%gBj^6CN-l#Q*Co`X
z5m#ya<-R7)lZ>Gz!2p@$Re;__w5$p=INcYnLKWXz+c`xSqe!e*aXLCMV(+K%gI}k7
z`@Iytabxq@8<6JJ%XqLjfd@+L68t6%Ie4-$HkTgln?EUi5Vfc18dX%7q<AwN%`?`~
zP-Ap6=HN5tO_Yq{&X!9(eshv9WM0ksU^tKM*8SvY>-*)=<!>uW4GkCT3cDubQN#OR
zQkYUl8^)UQOVC~N2}@nxNz;YHOH&iOFqYxFClW#9k-B_nJVP`sf%|OSFMUHAV2d#<
z%z4sjcNT_9%mw<nQx0!4x-Dp=IU15!sTJ|Nyt0Ue1k=N0-tw*aBpN;H+7X;gHL|;*
zN)*GQ6w3N~*~2yvX4LZu){A65f8bh-S#GEkwCh)h$`xSbdsOCjEw@VdNss4Phul_r
zWO*|j+v604%k6R2*``2E=PdW<`TVo9=N{<aQ+O0%VfD%_X73ou`kRWFmw|xi4QFY6
z$uaIgmPrkdum0<)B}Wojp5bxd=dtOSw{cR~94Uk{(t;nUtj*4@RYXTNGq&3)W%xTI
zc^s4Ok=`Ob^#&w@V>#BwMTz9EML17-HIKcWd9??}rG#6<H!r0Puhw*&X3cwWUvcfF
zKf{qB;~=w`0}1@C)t}!_Lsi=?e9*E{2;m!0pS|_xbxY)Lq5j;slx|3U(sc-s)$VnO
zQ)KEo1j_Bc@N8H)m7Kn-wpst)J=P&wgkk_vwdDo`#@|f0JiZLU#|W<yogCq@R-2-a
z!&w0z2=|jRZ+&|(Jm8&KJAOr>vLb&&H3?O}N$|=*M5^lwJ^cA=P>aYVnbJ!tJ*?!H
zs{4qfr2C6{o(}>QV&IY?8aWQRaa}%?O5*<IJ<{?Xo<c#IKk#D~_TX`k8qrsmq{Gr0
z^h4XbcjY@N=ve{ytet|Nr4ik5JdJOQ3;#R#-!`)z-@&nJgm)L<Utwea#jObbhj&|B
zTOHln*}{vRt>V1j$Nm0joSqXTM=I~87EN$?E`=T+CI9Vj2(a|=eUl1b5h`w-Aby?L
z=j)~`VlYY)MqH!x<I5ubd=HML_FB7E<=fa>XCpM2mC%gX$YS44jR~T3m5v)ah=iva
zY6&XRJV!QS2U>?I*;@BLsqSIK@akU*$nw#CQ^@zZa-Z@7uzefnk9^rbhWQe`toigI
z3Mbv<>)6~*;vrqTOyRK)r!$$ubUud*xqo(;qF&~ywKJLZ`o;AkfnhT2wJ(YELt)vu
z^<W_nOTYZka}VaO*}}~7_5y7WG3%-49(qp7Z7DiH10wq$Dj))8IraugIWs1P><a?u
zn)*XJeGanKJjj-u6-)oOQV_&*)Tab3M~d0*5J7Z{>Q7kdZ=rFaISnee&R&wX(84(Z
zE~gM9Hd$Q4a;h-uO)i}N)^iVDerl4k)QnUW6y8ZcLx8^LKvIM~1d<phb6dzXOsS1^
z{2ac9G_7=9A+;Yp1+&GsNbf@Q{U{x|;sPf(Dd1C<eB4@IACCDNZ=k^eE~_BQJfFny
zV%9idKjU_#`G!)Q+ib?YVvlu6w^TRLJee{>)(A|!u(Me?xuPj>HcX<)2BWg4C9Z1e
z@BL9t31v-lwpv?`rpcj#ErV~pC@SBzbpPd?&Z|Y_J1vU1luZgVxstZP{hJ8keohNo
z3FcC?sXK~7TaF(wYl|*WgUB*5-~wl<9;8LGGvnvcxCdp2u^BJ|<VEST9L>DkhuIB}
zQw6-SgLog_ytu{ESd%-r3vcHW@{k7(KS|z#lIm^sxSX_iYSCV@ie6V4eD8;KS9Phy
zppCZ9+<)5g;L4(&;{GonK2=5GJw9D{Uw#8+V(4v4Q`j6@G7v=}Xk;+Q6wPYu#>8Ip
z>SWv<6e~&y9mLDrlN?QEhU?rB94EJ0ar^o2m(N%P#!kLxoKR&lZ!y{EX}XGrbec?5
zJuYlZ%SMATActy8L{^)oX3!dppuFmA$5((gPS$I$y!HxTPr~(Q?DhHH0T3h%?1cp*
z-~8DewyhZhBqN(LJwn+M^R#7KwjyiB-MLhw%pxsIwwE;zUcm?O8A!;Ofe#q1UD|PH
zaIdhC-`0US78mhxUY=qj^tu}dk@Bw<U0zj`Rh&&Uxx>d}P31uP0{8|0Axx#4&@?q=
zj!ROY`4nIA6*jUwH2G%QLE*4$mHT)?)q0~`-L6_RVyaSo?;U!`B9$&f+3A6HL_=!T
zTi+1V5mh4=>UX@is;DfYmeNu&1xPf|mo$3p4awy7mE*93+=`e+ZCv_2NmAF=168W3
zD0golD;%;yr}rrOfyG#pbp^x3k_@|I<8EcF94kQcBPrg^G^*Im74VQO?WX$MH&?b~
zX$n)-n=9Ihay)!+IgUY+3=S&1UrZ@QAskd@9pR6nj+K)GG~Jz#d(&KdH(59@j;Q+3
z&6UY_Qks`;x)@$}yEvK&6g;A!Avvf>eqCVX=-S=rhIP-z4U|kN6_%!M+=KJ>yS2ZT
zHgPYGZn90B!thNth*RZJEWcZBp!|x6Ksk00PU;66n+KKQAE{cSfXKsvd=4x{bNqV_
z{IjbZ-}+rB$Nf0F$+Dcv_)V7P9&&ko<vC2SM^BL&QVfq&gKuE0@-Lu!EyuWMUoPV^
zSj=4$<M`n5imc#g=roEy$7}v`g?-n{b$<?TvRwBTz)cqJ-tv*h<a5)~xk&-sET8GR
zuJj7`L5!sbq<2f-FMU|Lg6q1%45{4j;!I<>nZ)JIZgTVl;m(*XG)kBu!XdNV2i|Z4
z5WY;x{q-UKLug72qneb5>xhH}GQm{=t-Z#9s|F`Ye{H?o<K~UCMC)T^=Vl@>O@!YJ
zFc~y<XrrVq5mi$O)Zt~ODKIr*hC;|KhE*z`E|*Nx(Ns-_|5V%bT~CL1a(6C{=g`qs
z(LL8UZ57|cXX```E!S~eD<txKxjcU_GY!Wi#u77#=@{m_VG8a!;am@X+iS|ro=pBd
zAB@32S&^wsm<nSl`Bq{yt93sJ{Cc%%wp=%IZHM^2H*Zs_Q1}r3;DXxop6?UKh7a5=
zA{z)NWyrXY>pC!{b0g1-;8b$X6PWDD@**ph+lz!Ow#!l;tCqoBe31Dnj!Zkt<J;C*
zFSrR4=<&G%RJO4(STZ)5EV^v-u!#tMr#^}6ZP&drvTo`B6I8aUs#en~wyLQe{0PQh
zUkhj3%^+yDXTv9{x{V9%+bVsTuj;|TGh5T+-JfnP%8-n$R?)^0tJj;vVCV#Q(&Xs`
zsWW$G4_)JUkt>okn6xf#k9@Ox8M>uFNNpCNuV2^_?%104QQ>{9aOdmh%V?nbA|4(T
zD@qzZ77q1L-<zYgq@w)a|J|2=Z+usEd>{3u%Hym`BYYacZz3$1NPFYuQWEJ1-+g!o
zDd+K7Ydl`dDo3z<P<mVH<1>waZsgR(*zY_Xwl+(6Jh!6z{iJn!E^oAoN!O1Tp_n{X
z^>B1%+e#N`<6SWj^$hOY*nAhqhc7t)czAiF5<;lqQ@Ggu9~H%;mg@7pC0o|lQ)7=!
zl!VQ9ntW)cX9TqTYOXZju=E#o^QD(eI9Y#lx+0{;m7Sygu4b`%pi#alU3rUwg5*4>
zLQMto#x+J)cBjU#bNi?o%++S4Bhtw{{_AMAR^SxboG>{~CN>x@5mCqSu*mCoGp+4+
zj=UmsPsz*a6zna$rZ=;S|3q50a0q$vW`&#jB1kXi_9Wx8%if;*mSb;rMu-R<?pnkB
zhdNWAibZklmh8j{zGVz^N6<4XEyLXJ_6drT{go;n0~#&fbVnrEat6IYmDyS4{K!3H
zba;I9c{;CMP)1-dPzcNsk2eG(8#xeYk%t+MesJ4v9tz{e?#8gUCkyc&Ec{kVJLR1G
z49)1K#j9Nmq?8M#i1(oXAe|W8*iE-+A(l}x?b*HxF@>Iu8>VNx>g_>LcC|ya?7g-h
zN$NOFj#QHtIV%2mJ>Qd#RWi1jQpok(RAMMOw@X$Wgk?k-Zy>m|l!C0*c<Zx=GUd7_
zfT_J4ka$mxqw2M&(TD`tTtk=^$W9fJW3S=c^Z4EQT-ok`$!V{SFpUa&yEjnxY`@~;
z3qh+|*c(QC>+P|T5Dt)c`B{zIq!n-KrR$|SKX5-k?t29z>Md0EEoiP4#XwTv@Ga!x
zxu{M^J^K6z`u1U{5nI@@Zp&+MY2c;F#Ys~iXtWg{N~fhaOYbP`J@c`B?yYnL?d$sK
zvy^WBEVUeqS+?Wd&n|jx);}o^F+LqyK;sHQe^yTVjlg=*3XqTdBEkbzUxXGcyD-Fd
z%^!)=)+c0LX1>RmbJAhV^O>x_g5Tb%lxijA*3qT7$8t+w@zN9fOyX~=V_PbDMyEvA
zVVdv^Q7v0icgf<rD2X@1@f+nXwJmu?`Xg@ZFwa|+63z`bvk;Kn9Z_7k@(~08;Zgd~
zsYDo|04FcFaY!w+*fY$XMz>)bMzz(d8ioyOs94Ob%vMZhDgIf@SDZLf9zseUQrs6%
zTdz2+dHA+E-*T?`_Ep(zG<?6&Fy*W0gZN*BzAqqsU-;Wg-z*mD><!VU@-+=5r|-Xd
zLlh=A0!X|Iax_P-&5Uj@cyT~Y6Iwpq+Nt*%wvJ?14c%F_1H}YJXDSvj8a$wCHa1$-
zCgO_9>KG(xvpEJysv@u_OxJnaijC^)mQtvx6+SqwErc9A)VF1s<=~-VXvn(70brRd
z+djt*4Hs&BxLZkkPi!L%E^9xR^%?0NXeBR7e@FV!H!k(udU?z&ncKpnD;|c(;PWaN
zP>qfH#*lrA3B8^|KyjW|6bE}bSJoe*YGr<&vy5{EmBOG=%sf@C4E=?JLUiT6_~GlW
zO;8SELW}rEd1`Hy&L5eFjy|3%(`8}iyf&gKqS=o|zMQ;18XU7m8fK;{{QxD5C`K6E
zd4NDwya2e$wm1ZD=0+P8s$c_Aw+fZ8OVXGLY>b&JX`fr*sNnN$!uSa1%^UsogvYn+
zvi>-B%<v@PAzp-&l8kF=8$%8f56@T^gTA0+78<~z+htp%f{sM+#RgxpYky+SvSgDv
zx_j^M#EnLbhX?f7i>orzEGJ&vj1$*&ol0DviCa!&R)R)%@la=`r>X`~b*C280ypw?
zx58k=_3@Hi(lz}Vkq{n!%`$5hMb#6hG#3%ePU`p7Dz;?>%XH2)R*x<pX_O77>5&yc
zuBJhBUs1Fi3t<I%%7<`_*jpRpmbCU}7hTYH^UK#Lp|ycyi7!fS4!h52zBv;$TC$=0
zetmFfV=gE~!CYgZS6>aPUNh<RH@DYnOBD@jGPS~1(n`Ww$<obvgEDfs3QBL7q2b8x
zud0^kwC7Z6HpAL-&GnlL?_Fs6u3J9FmP6;%>FtxjY{>?7sJY2ZsqWgcWm>ADlO;nd
zYg9Mus+5jHrC%m~>IXi;e;e$&c%kTrF8;QJTloV*DIX94JftjLNcETVbTc%z{;rFu
zi9EYIvyr{s3XGd0^miO)=4M4)i3#L@NVh1|&9Cd)9JTaseklz%iwjXdqSHAFe{f-K
zjcRsIKjz~J{EA&z4^wic^D%4p&Sn{?-yF*sU2x~VK6h<|KDjeaLG&YYbE7%SVaU6e
zw$R1#wy;*7SbSLDJi-w)hQhKk9l4E`Rb4VDuCelH&fMC~R%v#F!JuJ6GYvg)<XH4A
z9ECSj4T?ik8+bFL@EgSx76bI{sV&`dYMGha!=>}Wj&!&5p0WAP5yGc3vm=a-X)AI2
z^DV49r;TCTQ#p!Kek}d^KEi0ggE;nipkZ<GwUoQZQf}Xr!XEA^7MPfolv=B%@a@<9
zl8dBJ&Y6Doae=Zup=)Um%C8#*@+K+YgM=9?kgP&p!XI!b9hXi?w@P<NJLva#!y1Nm
z%p7X29iv?{YsbgQfXBfWXLysq^ZisTimpacbTNu9ivvIHty!yKxCP%w-Eb!iL%fRM
z9A-A!z>_kDCe{R=aOM|_nAB{KbMFQvp5xTG!h|rD3D&aS4p4rz@b7DTm(D8%ElMwb
zKY;&Qc$m+xKD~D#HpVn4yUf}|QMR|fj*{G`&b+QVv3JSst^6ZjXha_BLqFS(v5SrK
zwdN#7;-PQ7)}di}HYo*Ob-BM>tuEJKfVVnNR+37iW9t4VYWT3fTq7NPJXaa@KZnzG
zpJnt%;3ziBqTE;!*``<GcH!J7Zy7y8uE(uXsTFU9rBb-Xub0rj?E{eYMV=Wyy~I7<
zoW+$z=jzA<Zl}6Zt*)%&Vbot8=KV+mYdQD%%W`YSMLI5w(uNcF*IUBOUry75!?jUH
z#Xid5UNO>ebEDT!(UHfId*sa9@LCk@B=hsh)e&2z>T;FTglI2RPgY?h8;!DwMwyTE
z6G*d&t1lYsZZ9<k+#K61xXkm~(dcPh)ILprDJ?-bT!(CQ3&zXl(W1m?t{3QsFb|{S
z6Vs%YBIqbWMK-!({X)z#Km=(~3p&mRd{gf^ae@JUuhIBi1gCnvX2O3D@pY!>ieFvk
z8xnjI;UauT8t_kp-x0VmD)ov952De|#b_-VBxCX@TYZk)p4yW@v9QeDpq~I$C*@T0
zMsh<*&XWF6h>w5p&yL~@=BQ<zPEg_J6B4&#0;Sf9<Cdw(i`acF%Gy!xxbR|(FU0Fh
z<+(Um3?2s=kC*0)x(U6NX~6r=$RSk9{C5tMRX*N$HzKkmWo}!~x~?1xr`ln{wDQQg
z;Rx|aZ&*Gl!|<1_k|BJUupTxty<}`^m-L$?PFU<ItMru8kD?{bBsyJRb84|2&cghJ
zu7?_t)f1Yh%~E<&HE9s8JGGK6FV3??cc!eXYwH^OsH`hPT-MxFu3(Hv^b<A>!_ai1
zszlcevJxF*+WI;)O^NF_Tv~4WYaz@f)svLYLh5LJtc7a|^(zizi)?;T)#Dj&eVyvW
z)Sg!$Wtg)KVJ<(Jw1F84Uue+J!_uA7=cHf2nZkyEm`G;|BC1v*79P?lTN<~P<`4;k
zu7HW^im+OF^vqs&n8w25YZh&-$8D^|#9<awXmex>jm^s1Dt1JvCHio)mrk)ahihGY
z#*Nj7-PB6L<8UeK_#&1Km{Tk~wGdjGuF4pXhA5g0a@TAa;yHHRDVegaDjuUU8ZJ<R
z=HarURb1OrRavdfD5@D&E3*lQL~SlQ7Lj2tp)wugVyPBP%{9$bRg)N0VVYfbYL@N6
zyT%+s;x=n4M!nK4AmX=WS@+v-Z~MBcW*4tQ3JMIcaH-tT91D_SzCjeKDaveY8BAxM
z6;=^<)d_7|LA*rQQ*4!}hC?mifEPf2MBrpoB2$<#GNxKF%$4DpjKOkL^g{3)S+x)n
zCeSOP>1i+_wL=fBRQ=F3EK5;zeU?D#x{h9A=~>GB!m#`?Vj`kOoRn^d8uPI91deJW
zZjLiUhIuE9U<K$0l=K8-!GH~x3HG@Cs6PrJn#4WaZEweI^kZbRM9jJR*+uT}kYtml
z_cQGT(>r`?w4<{V8hi9f#osB_KjSM;)=OJk83g?M4uP_VzN#KF#9qWNilvWFi;?T)
z)YWN4nN=PADQg=Wm2K-O-BD+i|FL<sURu4{yt7mXZ4?;seTQ^9#OXhtLRX}xZ5<sT
ztXk?CW##oGhJK<4Bh5F#xc^b<?b7?CACUeITBh;VMZ^!x#wJ7Ip#c&?@cZ5XWR?uN
z!sUbeaE#>I!!hW3f~O(JSr~!myChw&&MrpM%Ua>LY3w`~V7>IEQe9`NZ}0d@X9vbO
z=N11cLT{0k^NM_n7I6VbXHkA<L6S;l%E5Q<fb*XfLjCNxG?x@E&adKf(pN#0fA4PF
zzkJ!Z@2+WYrwVA+tU}*Tg{-xHhma14^@5P_)qMpkW!M5WatLHQpJUUZ)x&$ZMHp{w
zio@9+F6HF9HaR~|au1pe@jYJy=w+Cmd>$4HjK&OE__rE|8gbL}xClM3`Sx7pT6uKC
zI9<p0@b7B6Um+Y;-b2;xZIzzq_+zBS7!CDC@I{o^7lVc#mW4FBzTL&ty0(8!NJ1Ar
zi@Lb8gy+q(D&2xMV=IO5ACu8CM>oAsdQ5sJ=-wx|&pB_cXL46jY)6OFCAUjDsS?GR
z10Kw*ds7TV*k~A@8Ij~F`kq?17S5Qaz)c~hIdgGBm>(+&F4Zc2ZT(%8*LKLoR@fFh
zn=2;qL6$x(*<Zxr`xosJ^o6^m`)|f%IU(0CaVc&IVRFb9UIb5x5=mnwW8Izyy}Ka2
zSNb5-=pRBqtqenNw;XfyJPZX1bs6#6nxu^bpdR<=0x4;%E&P$8#=|*o!@#1?O_S13
zhHVTexhaxxNm?-7v|tAPU1ind*nCmD@0Yk_OMbr{ab)-EvgHthzDIJTDu+y6h7M8I
z%|nVB$<Rk=ggDmnlKL5S2_I|l!P`CyLxH8EOE4aII;*2XBzK9E@Wv(_-7g}RQuBWr
z1^&~1O|eJ>-|T2|WSNT<+b=i#MuW*RYrsX>w=0XL70KG=u-tCPk<6HUfvAu?{HZ@S
z2A?r;lB8Yb@uYM~dJaY>|M-T>11ka(FnKVv>xz(}O~kCk0mKpB><%}f&a^j&I3PKW
z;9AEC+DA^7LBcCUT17U~*glix*zV=!$l^VVsCCS+h;5L(G}iMnBZf^Z<Jfx%`3Y5C
z&I@muym!1{pjH&_A1juoyqcHh-UYU0%bxkuhNpOqvg=euA?1Zwt-=fAr%g|`|5sfT
z<rx{{<ztK&*p{hJDPfZPjIN|U`vYjNic!Myg|M;4yInL+E2nlzXa|eKL;Tn6Z6c&P
zLUqE}@j*RqSK2U^c8K$6*}v?Sx6vvI&ReNT8n!e#a2Pk4I#({wQ`Rg;R<i9G9*old
z`^tq26}Q{n!C!5ay~}9Wv|aYDTI@O9B$oahvwqcJzHTrxEa@iG!g#h+N&-r8O;vyG
zH6HuosN}<3cpgRthhd(yjn)r?jgvBu5bMe~97a?zxG1}FublLd5g7PRIiem{mWP|m
z_CnX*zW0t`uJc5+b^MT1s<3e7_=;)O$);u9smf32`a?|pK8-%0v-c?SmTBz}YFaNe
z*4wt-ss(r4yKQ$DyyNF9CFjubt*Ci?C2Uesw?bLD(=wN-_7KyL>H7UD+pw%HSuI;8
z`X-CDTA$awBhqmmmn18YA}Ed6N6B{7#cI+Nz^Ek7g5a!|SFtRNvVYMp<5s$UhQnE9
z%dCB&Rkvh_Vd#^qTi0-&ea+G%W^u&SY2Z&r{z%IFZT%GAsv}|K`-f9}tLYRLz|DrK
zvqY|p!nQEQ5QCq^-be_m1QhYf>C~ScWAv;57dxX{UzPbHt^DQd*T1GImyint#>%gZ
z{N~y&-YwQ8UqS=AFOew9@e*oM1SN$LefShR=D>*DOipzFY3$=}7lvpEb4upq*nP;2
z<)55=WY*R+Tb%xbEQ_#Kvi$29toe5=m>*FUN(WlxiM%>_9;+nH7SYp_K28Bq#_(yD
z%Uqi6zibeTJ;W^1@K-liUruxR^)%P{FEO1Eoxw?$QXD(p{x_vV*^WUJr$auL<vj|}
z%^^5=P!OSxNeL19H@X<(JZ4&oVb?2$Gr!KUYfYb&y&v$(#OG*9XE}KD<2!ud(J)Pe
z7^d-26{g`Z2Q}#lQOLwctKHx7N+tKV(Q22Qi}0<Mv!92Cf${^4WfYAnpI0d~haxaI
z--pq1-VBcUbsBoOM{&h41Mx}**#>306qN{&hf~5(ipO{qhF%HiBd$(yo3A01Jq+!j
z`P|rS7xnvDU}c9s!a1A9^NerlKcHLw5d}EZH3RoMfNN7*Nx4PeM&Pa?_)IzXehWGO
zIx|i7^^o&Q;HYw^NFfB3j`1zlM|F<hm!f(CZ!ACH8a6t^XNTTTE2gpq6J@!Jv*jkt
z%FlPt!)+7A+J#$%EUM~*s(C^+PpZc1s-d1V)e{&OS}x>u{Kj%?gUp8roYN%YFx<o^
zn+H?zILxDaJpxx9ILOT}H)X1k&R*2fW5-N-i$PDEpvEnf<5BRK1uGc_`{dF+n*h?A
z1lA+^|LMiChlihFbOSL>Hz@l6UNU|lE8;C4{t%!2I3A$gXX^cUCFL<Z&@VuWQiN5P
zwC53!_Bq+KP5G0mV(G8Gs#}UmJPQB)F~eyJ%j09zM4XmrgT==|ZmaS{GM^)#L`>OF
zmUblvL!tkJ$mEaBVzPyFISzWXfopDM4^nIQx~FjY7uTU?Nq>I*B<cM_Od8XF_0`E-
zh@9m@{+M8#^~s#19D`>CMk%-8jzfmyD8P3)3TT{gSEF<l7G0B05e^zE;Cj0WT9Lr1
zj5yocDmNG-eqz#i#xp&~tHUf(k!7m92>K!`ElTU~)bwUzY9>CB{eady^PhV)8u*HU
z%k*3&nwd>Zr2<z=nyl17-&DjGs423xq_|#1F_YPuNO9+tYBN;k=9I8mRSxkz*9o59
z_g&zsR}w3FY5(cr8b-g@ZTAF-XkTC@Yz}Uc`(Gi7gNwSVrn8zX+t1muTw}VX;!>}p
zkeigweXcWT&$t#<*ECaR#4rexO>IrpOn0U|aNI(?Je~U;WH|B|iVESQ(><LOsBcyt
za(LMkuNmhI8%GE$!jfN2!>+AV%kk2DIKOpv3yo`PM>Z?P`})zX(Hi!%<z^ZFuGp$s
z>&5e1b8}l9jIF-fkl|Uq3(w}^kNaUBZOo+0=u53qzTf_BzM{8*8a)(4ZKy)KfJr%s
zB=EA^EHZ6^_;OV|OTMMj$kiK~Z{u?N1t3nhu4_LfCf;8VQ~HXhQn?}1$P8Y>;X_+D
z7op#XWBw1uG8p%wXbwr$MOAJq=ZIO{E5ZGSTcq2i`}5If5@%RP{c)g^v4sf{9Zg{7
z;8J65lF{p@yYZmqXG?XAn~mNY7twJ;)6IJ61->rk>PI$?=x!vK`7v>NH=jv#aY8^=
z>ZNNIbhEtYoZ)b1TFJUrs&C;pZ-)6DV=$3>0%57d0v8;+WVF_m#h(;$3TB|iA432C
zp4?6a#!>lVl}oY(mdsg0gnLV^LSkTcyBN7*WCPnSSi=@=nR|oHZu^3ee7bsdNiYY8
z3pxKFi>pV6t6O~jYU)}pDi|58+JRW)A%T5?3y7HMN=wEvxmpm#n{_LCGh+|nJakfY
zEV91l+t;`yx9#Wl&gk#6nXW?P?r*~Jba!j7OOD7xU@$>PNBC%%51mTGS-#?$Y42{`
zhHP!WsLoap2PTa7Dl{ww^eGtnsMl0_-3(}^($~f`{X0Sgje?4sp-IfpQd~m^WdW(s
zkMo&{=EF^@=#Nj@0_Xkz63kS72K0c#o1VhB^rIt3n-i|E*%2m{H~i(0dzH4jn*#1<
zGqw7{FbX|afQX6JR9rHFnJDAG;<55D*&*MR;h^O3KOsz=H?1fLBh!q+AhOJPYRGgR
zAHZXL0FUQYMZV!L`#Li$g;63Kj4>VBiAH<aD;c_M`jKHUtrT#C=+%nvQiWOQsTL`+
z!c-TpYy*Z|vhB{T%~nEY-VRwiD(;l%+0l_TOLib_3PYas=n%c7D8wl|f}<ld{-s_>
z0G@p$v^)c&+v~RJmuB2fsZuh^Ny%1JquxO8J=cl-c?bFu*^n7ia6tsR6{gyHV7am1
za_R+pFCKq(QJV01>Zhx?eA#xHaK!l&xRQI4h(1-=p}$0#rdxz4s&Buf{r|;%3y>vO
zd0wA$y8HC&cK7YR-F@HlxO4B_-I<-)o!Pnf&aQT}TCIe%gQNv52us>UDia<Cf)*?g
za)HZU1+amlz+otp5J89>DWM>ximPmEsR~KqGImP1pb}KBkg8QBPMHwOHXBz=#UlTI
zPQUIvRx5)l(rkC%ex3K}|3Clte}8_wA_eyUB7RgJDK4bPrFcIM%&;eSq&{Bj(oSw?
zK8cp*H8kSliWy9~C@H{*!v3=^*4s)s`X1`MAW*(2cE;7K6<>w#v~9q{SDpMg&EU5P
z+<HkqaS7E3ao;1QXXJ7E9jggnYG0`#k6#FCUNl}CuY)AWk%*A@qO~FXyJP*-=+MLQ
zp~b<Ge?1#TC!&ZfjsD$e^w9Rl7YBnc4NjdpmDhb&U~53<;z@7Fp`(Fl2`~BK@>|}2
z;f{Nk-+AfHx7^fxszti*`@);w{41Zm?+bsHW1HYJh4@w>>7e0L=lE|KEu_n2sK*H+
zKH>!9b%1P&IGka8adc=9)q+YjsPVAo9qd=N{;i2)*StzQG7MeQXKMa|n!y*tR^#A-
zLq;Vus1w_&$7_Qco$Ev^E~wP;Jk&{0uehX@;~f7?`D+Cud<%X`<h?{UaLFc~+3KTp
ziJ<s=cfJvNwQlou!`X^g_1d%5pjw@D>Oq4Nns`y;_y4JrRPGtMO*ikS=+9$=)~}3H
z0!WE8>T!5|^r^;$M&rV#8;2Lk<jT2oSKj!>D|5}+FJ;dj5YG;$pUuLr@jduDKiPPo
z(Rko|&<(k2-S+-_SAET#U7wl!-raZq-ud(Yh`3?Zb!=|m<HuwpG+i(5fKZj^bso|i
zD61Jtp$FrshpRe6^smS8MMD`gbTjSoyp*j!4XaTUco9j+?0A8=w(D39nE|08Aweey
z-QZS{xlaY%z_n}585sB$q11MQTC)|}uHOvaDf5|Qbub3Fp|}DoioqZ;gc64$L&-}i
zU+dQ&2o@Jkllhrdk;Jtt!+7r2x4dI*eKUKZEZ+d$*G0xcbZ(LGa|9scyZM@!&}FI1
z<$5MzQKz2x728lF<L&pN{n&OJ^9!BkTt_7Lwn`=GN4jqOQhXe=A}D>cnWZbsaS+l)
z2FhcK((k1x;k6Vt9Lai4F~Tv=7Rt+cAPaTMO*gC))vg4wAM{#&%dCXe*?RZQ@mvIw
z@mOT#PYjI@7Hn>MeycNAapS>itKN8P5Z8va`0g}QGN#tIq2EUUd_aT0$ODT@PCd|P
zWLU$s@iN`+5YwZtp_WNVhZ=YYO5p20YHAd1<{JN~p%cr3qYNXh{BIaTeU!kG-az!v
zM6ePjGS<A5zk2I)Q4h=e0j)**lLeB{cp*919ngd}hZrmN38LTP^uJ;1oF|}|ctQ2^
zow`>I8osIqfm3UC^Z8%-pJLqNxay_;={>1ABrv7DB^jO>Ce0(wk39YKUp>|x47!iK
zKW??+w}(o$KHF9PT&%12a5kEX%2UL~-OOUqA!2TrN5?%B+!7?4$>!PSdh;tJYArWt
zM7*ZgOB#)^71@5JG9QkQ9sAnX{_)*+U#?!L)sw(-+=?45#DN3XWo$KF$Z9L89E%q5
zRpe*MBW~pf9OuO$@_DQchPl0ZEEGx}!{2a}NNc6OSZy?VjmCcn8jXc|{pO@^L%ogW
zJl`Rk^3q*_XZeO-p`QO@&3BiY#5s7!2OA6UK;t6b&})22gGp7LJ<OS3J0!21NH&f)
z0;@i$J0Bxfl#tM4gWS$^Son_2Do2!4$~omB<q@=~;uuRJxzZ?u#O8JiNl=!c{i#I1
zx-(WC-|1bw!+N^QB|q~y7j~-E@n8ZXB#4cuSMX;x(9e3qkfJth<ZoOb-pILc#m}wu
z%*2bI5!=|C`0-nWQ?`Q~^dU9arS!eHc{+P9yz=SN`RCH7H}FOH4)MF6zA$zF7Iz##
zck~&-4iGXiWn}idX1b<zv#8g1E06DH&Mj4&z-jKRwu13L^@J+I0f-ZP#|uxAT`=rS
zzPVhO4K82_27W|hzb_A=FcC8t$dtC~<lgHcLbK%017)Mbg<YG!6A{sbh$_N&h|P9!
zcpGDZYE$M2$LZ(b7s9gSFF$|wQVPcYj1-iLx?B=6P#|LA;)=u;GJTD+pmwRd4ffdi
z!6t?Bj;Gm3N<WgRsE0*F`O?|bVY@wSwXO(cA!&u7fC;vi3w+=+1)T701v$azoT}j8
zt)_RuhdhNCK7)Wm#2r4Jf)Bq}-17&;3+^paNTcHQi-bzLb5md`vzj=+yI1iX_;JI%
zBPWFBS)G*n%Vk6>uFPG&ejnJ9sK&bV{wR9HlL>+r=hpUX<{wE^<N4evgye00f6Due
zk~s>K0&<tehLRDAqTX7%9Cebc^)g2{3hg;w*Go6bhwx*KD^)0lZL8Yw)PSfeWlYuG
zuBT~?!VgV1ZPT=(naDCtw8ZIlWb2^G>*;mPopE)f?om}TI5ouJf2u_nj`t{22~`nV
zrs6`Y%5?e&wQH_hbBG8TXmL=iD#jyJXBw{4N*Ed;ps)3gFHlR>a^LJFfo^$^^1)ZJ
z29OPDw2HvPWF#D%a7|Re7UAbI`MFh4GMsRK5TKG290+tXNlg6~A0Dg@s(kYEtiA($
zb>1}NxP7N()oSxWFkibPY)6S{&aY;&$2H|e&=?j(#6>i5AnH{{SVIgf8biu<vTM5f
zs|RpypP)~>p|JzMs_M^5s2EhxK@15w$voH(YWEZ#DM~5sd!xiKzsuks``ccO6Q%!I
zKkjF7;QQhyOrEijHwt|HQIJE|hfz7k`8to*g=&$m5FM4V)<tv_TodUK(s*oz(lKQd
zLRKC-%*ceYmF34CTZXf2OzHifxP8<jjLcFx3%l0n_D>Mf8e<SU%1%+zZoTIARvRS#
zTNs&x-B?k|F{zAnHY(&Fo<tmDjIEE4twxE!{SVj2=&B$P`SG~CKFo7yJM`jh!*P7a
zId@WLy5{S?2K$p#B`W<YZtpL6g}+k7p#85-)q%g3zVb%sMT~7xy2Tjuj5n_6Dh7p8
z-%*ekmx9}3pAhy)Sq3G{gSukOIczb3`;~S^WP(CL8`l(5-?>}f2*W}BYs7u9HyEnH
zNkk0z#|_mazI1@Q&q%>5#5nz!s#8sUhYEih)z!z+cIByE5PROv={g<Jv9xCnp7&z>
zre4C$#krMPe0NM5DEj!~yMED+B}yNM%$66IV+-`sxM44w%1$|C&s2X$#7IJc>3Zp3
zwhg4X#*I>rqZC4YpHyBi@MiB<KB9akTkF=#AF5vI6DG4JJdMSP9VAF|MVZ7inDbHU
z@hp^!u{?H3ah#o>mPJd~+N!3bbGAU@J+7-0^u^W%@y9n+xtAwNSIuslz%&0NMLVW|
zh<^4ncO0G|%G40%<@_CnZcV-98*-PYUrPM(T^?{Rw@K{_V*ayX2E@K_vydHrrV5(e
zU_oDa<r&V_tC~t%upH!#E?RcF2aop%I|Th;3?B{=?g9HN>IvF!7L&E19YoRgy&$w=
z2bC+V*u&xIDM~kl-1`)sed+){QChzQ4IJxGDT6@EnA{1DfGfCpHuq0`UdlNu5>==a
z{yzAwy7(fM-?oGOKEF`7qy8XG0ydMJMeIrWdQ%ErCu5{7jMKSVc~SVNS3qN(E5#V;
z<(_prM*SthP<#Ph^xl9@qcDUuRhl2SZb6s5ElqtEtL|BW(Utw{YAKDS8%c$)<<^=~
zf?Ect+FOnfsa}xG&E<4$<4J^Le39G17X@&3x@~0a@YEl0E>CB=%a{VFETSF#?F%5|
z&=)AB&uc11H^WVv(Ev}wMoHtAqD?XNQ%*w=q;D6IL3TRk8QL61oY0}xW_9DKyk}p<
zpXadN&Y>sqq;Me4;&Wg+`}!&Gm2wizt)I_F&QsGV|KT6bzVwrloAP8j{c!f&xx9SW
zpjT(!t9&}0^->-36dVRxdXW<JgR(Pry_`zjurG?~fCwcZFDLrZ4F%KHUjEp7<{$!L
z4T?J+trtV#+39>-xuQXlXSHqN{w-3xxI05(9rowW&1uH&@v*pn&lNaYkx_xJ-D9Na
zz9(#kp0Dp3FAv&QQWu_|og}gB2i25&Z7S`JSC%@~2|}=YD58v~2ayq47Eic|kt<**
zskKT-oSH^<45!z-PR9#%%hE%y<GA<2z`EDq-7t>BE;oeyDPlDqR9=f#2mvAzthBfo
zBs6CcjRVmXNk%9FiiUJ)+QB<<LM&EF1kzysGaW)%P19<IY1*n6`OLDsN~D{l;;T_R
zX1;6imhBK12jdEEkoI-1#wh6!oeUPK_8W|v%r!L@)GC!)z%;k6nat*Dy&Z(@M6L3Q
z>gzR!%I87o<+d9BvPKu_)I5yyMsIgt?89JArHkGie*fmy=MEwNtC%)kdMOW@_UEb=
zVtnpr7!C1yNLf3bO+d-n%5`{@nUSY(MLDIsPI)UxK#B-z%n{a$ak^6-1TFj$_O_O@
zdFEFCWs>JoXfitk$oEC{qx4_SI*gocvst5#J2T@twATDZ<hf-DmgI}LNf^3geU~A@
zZCjjpfu+~s=SzQU@@qx~N6||^fgwcnD`wCMYEe{!U0d7(`avCD32&&2A8_hXN{P58
zPcnG}{TGDWWS*jlFeOLC*s?sASBpCO2LdB7>e94|Xh}=UYY5jPqfB#y8cIX>A~JKj
zhUCKK`tYa-f)OWW*#<`1bOoI(a(%ui9Q~NsiW;havRu~Zh{c~V9E|6Jg$k}bYmsWV
z8r5nM?8qd2NYrytv0s&Ysre@94bTMNH{fZ4@Yiej!fV&ArTw)>dSXRvkjGr~D2}T!
z+5{3*_R&_!Q;jV0$RAOIfl{>KQOo$oacc8TdXq+uen}(VzkOI3_3t5M7IY_O(1lHI
z)8qf1Tht3ZqJ8No(QbP0J!rKz9;Suv*Cc+yAvfOXhqQA>-<Mu-awQjEQie>B*@Q#-
zj?D|PFPCEdIJEP%+S8I?_4`?L!7Oh}`TA>$FWn<=*=X8%ecHXEw^t6Y>~Ri*Cui%%
zVQFc3J6$%0(?MyKSy9|vA97L;#5LuFa;sR^y8ylZ-O8iFyKY}~xrmpP9gKBJea+MT
zSQoe&s2jP@nOl0<K_V`o`?B080}*ix!4P6!QkegsP@_{xQ~u2|8xaKb47GLKzd^H1
zuF^*}(}3OkQWr((TlZcK`L+1_RTFpjGR@W~=VzF1tFR}HLao1CDEsex1+VDdc+QHo
zl@%i55B3G?4@y<@d^na~GO;lH_jC?F$Pztqa<LiIT~DpKOC7Igz?a4`2ZSY#6>qM5
zf9bXN9QlPKZZK>Tb!}2vKH<Z`9hBs2dzsB>QG98ng)Lup-LZ6?wK=r^HoAq+Lhnz}
zSw%}+Ejffsfqz2+Y{H`c4o<m;=iI_szgxj*ALAV?1ErwGc|P~-d!-JtP)ah*@B>Bm
zQtVY&G;vrIBhCAu6xnzu=M2={CzQ`B|4jLp%D>5C0>(nf5Kd@<`&q2mxTIJe(1lon
zS%zj6_9Z+P(o3=4Fwo`F4!l6D2x3);7Y~XB1@Q=00K8(82*Zqc^(c3wz|XS?BadI9
z!rCXz6RV?b!`4k&Q9V~xD@--in0toG6P0LM6X%4crV%y4u5M{DTxUr7X_>Gyb*PCQ
z55ph;;r4n@s|XMh4=>A_-%2rRDa7xsDNv6rWlwX}@c7P$Wy;L7ka_evNtYhpmGAP?
zQ>D&ObRC(cTcuy*Byajs+h^X!AIa8=#H!pfh=v7@SBgKP6G*P-P3e>@#ez*4<V?~X
zh~TA?LP?^3k7Wyd*dY>(8Y0#LcR6c|j*kXrXi(GhHI!L7h*6%#xMM_EA(7#5rg@s}
zX>i>$p)lf3&9k-Z-PV=uv>|ID%<80Qrt}Gr4Y*~oSM=_^Zo6nm*y7>5Lm<j83&WYp
z8b_4xrjZ&Er^(lrMWq&$9Fp^2cK1B^=Vi-$t*<o>FU96U{pq0K-)N`w8)y>yq+2)r
zm7yCPT<&Ic{gTonlxDabGw`DF!<3`s*uX2i1d>?zE%Bk?ff;wnGq!qo5tY8h!>T=T
zoQi2#mg_iIK^N=kY)jQa^U=4eGmdN4n|0H5W-=WhO`|1|X**LHj-^xo+G?_IdTX<@
zRP7JgyZ1}?9nJSN=%hWBdWJRnA;@#f$R3E3(n{ffHI_t-JENVXR^bn8mbQe>D(cdb
z_Hia@%2{^?4XkiXZFQ#0jK84zw#bOO#ULYxSVl;Vx8pGy$KhB}3NB;JPg{H!cIOBN
zm*78uEZk_`43I7Rea2>g>jpE~ZPE#C{Nlb@&9E?-&%N^at#ax+-kq-(rO&rtrj*~>
zvm|?FIQ`BWgs1DQBY|o~mLkDwGu-cZcls|R^?FjNMR8cE20^vr`o8Nrm5S(GX1$!J
zescWO@x%>}ZLGn~hmRbtR=ob9zE^Q(=CjWBYrE?}o673~<n`44xt$tk!|M!?z4C3R
zxrW_FDc{!WtgOxb3UFt6tQ_BCFqL8jhk{{Gk*pHyy(~oVIx>|;`6@(|F&==>wO$pn
za}`+&jg=L_ytA@m*pIKEp0OephPR}hdJBmoMBwGnwu49>AZmV6*8lT$NW^j1MTA|<
z20BmHgfF7Z8%R<;ttY5m2E$$?Rgv|!<D)+$?_#`&OTKfB_;06rWwoNyw+q%T!mCbS
zXO?g!RrNmdTi8Yr81^HA&a33R5esc1V-FKh$SE~9(|&sML&0<awH5n+Cl)Gj>Gf)@
z6~b24ehj*&arLTUUlY+E`eNy|Rr~a(Awf-ja$wj4J72dXhm@Mq6r~2WI@+OCjYYw|
zosk#*TO}fbl>iFp+jo6}YyUA<`pAG)D~Z6zBU0WcAl{HQflU{iV!o0bxacjZ>{<%p
z(iLE{#5aOa5ZMb{q;Ek(Qzq<!;ETFV8z=R?s;yUophm(VsJy8XxFJK&F&4T(WwoHv
z5?t+KWnbM6Dm{Pc=%U|47tda0F4VL@)51A84z|nOSShEs@;3te{vnWQUlsKGt8wS6
z0fK)u?)(u`1r#1c;%r+1xrw@^6R?FYaO{Sm@arlg?lipK-S^HL1mc%CHDj@6F%kV(
z5HmneJv!y1HqO~FQm>etq9mht$pzI9QFWkF&E<?yja@|UBj)nd{o+CnWP1c1w4&%O
zQn`MRG0jwo&Q+D`#MJJW!r)o4?gH~8Y}?X1Yem}j3Vl+!3f~Z9XSsa`2jwEhUY;%o
zjK;nA4su_W0>zZc-)X#^H>tW|mI|clj{2sQ@Zc&U124%8A4cVKHo7cLl2=m1o~G(r
zRyI&ZwAkCiwHTy3*x+_MmAic`UhJt^Y!1CNr_U=(%2^n(50*7%j3S`VkQ%xzED}U~
zigj_lmgGl6^dj&?S;r&vbr>KL=LA9eMi-Uo6n&c&_dAV4_4=Wshw4>VrO=ho<HH?O
z6S=-Dels-;m2i{)_gbK07%^IBjCsB~v)ZB7#zJ?sbD!rLj(Ml)7_jR^DLr7d3-VKy
zIYfI<rOYsNZnEs)NNu6Qw1`j_NjV&+wODU2G!CkUX(DABG&a1+%;(%i=&|_CMVyu_
zMJ?guc<rd*P#$m;3CBYu_gh=-QUv0Smx*wrk@2~u$K^<d>7M6Db-}HiM3Sa?yo#g|
zKaIwn1GgQhMfc1NV4r+=zN0NWES#UrHxBngGPlu-42HDq8ub&CBDn~zqGHfJLsvDX
zaaA+zC^nOd?rVgqM2V6EXU_HRih=`Y4s^$VT4j}Gt$W+XT-ZC@nA@0p(56&3ahi!d
z=w~`*&{Nt_$+~7FF7%s<9h)krwqr4x$&uexo=CX}GQySQD1Pq+0|76Qu&j>1&4}7O
z-c$*roJB_>#%UjtFl#1kE6iYTXP_{H?hJDhXlI1=;SNoMn=6PgmAcCRRj$DIZb2aa
z@$I@Gau*(QtNdlN@P8nsrC(ZNMB{9Dame`a<?{4t$Ww=4@DV~A4LZ1P7SlES1>&S}
z-&d3k!9}5r2g0WY*OahVBz!*2_c$5HiCl%`;t09}2uC;L#385^xme<vms~=tIUaU!
zZ2YZ8b8veX8k)6F%?H)qx-3_B;Wjt$#bI!!3Zq1p?F>kxc(4CJTC$2L=L3l2{bStx
zNKayFLNRL*o9~DMhZ^^|B@XRFRVcbB_XDnBiyNIsWx*?03BQY~%^%|?UngIHdA+9e
zMI9*p5weRHm4Kw=qr7!)ya{owzWbW3tl+nuYB4@|#m)NrP?$K$3mnxbh+GEibZ)Gn
ziI8ECvlO}JSJ#rL2JP$78G)?p4MFgt;-o7Y4}@K@u8j2RAd-#o5{!IF9<?^ciJ+G~
z!4Z`=77U;$zWq+4L71*Gqtl~fIM6f#6DzGa>{iWEt9~=`n{7kWZI#?g7&ocWYS6CU
zM*P<ZMm^VRG0bu>KU^ak%%qj^YsjyI`iVe7jK4@M<MXD~TNo`2#wQ{N&glp-g#T;D
z*Kewf?eq*hpxXx)(XtD_3i%UpDjoQ4-zy}39t%h4Y_YUhLloGm)J=L+07lYFnXsN-
zjpUaQZ}|ZR&VeskiU<y_q5up+dh2r{lp21me2H9ukKcPV_3HwStII3i*Hw-1wy=d~
z&|`GC9=1V)QDta_*$7@^3+6RMjwIRXSd=k~%g0mx_emZ*GABaKS35Hjso^wqOKLfJ
zpr&!oSEB73J2Lf{o0YpmDHqhGEE3(1i{NzOHQRXJut{Uj*pbht@fgL{l9mK54?P!N
zOrz0V>Z1tp<nG9kN-6#h+LR9Ni&!Y*S_BU)6n`$7OA#(5#KPCZW;6T_Zhq+5-?2Sp
zlNSdU3!uhx&ETs+Q*4%O7(liYe#-VdJC{e6N<Id6irMvL+%9%rAMb#yNF^Dv8;d*V
z)-%A7$kIxYwKwD^JDCTD53<)wKSA0#7$Ce<3@g94T`F4<8SlrIvAu0rS8XE<(I!@u
zP4=}z%Ch*%<;$=o`bW9eKHLHI(bnf84XS-2O+anA3sbV9`wQUDOMi*xdi?hTDe9?m
zSCPGCQkWjMHa3Fo{l>w&e*WX&x-mR|rg5=H49~ys_}D9gXMu4j+V7ZhVhSBR-me9V
zj*ImL-EXwU*VmA!GFeBMST8O4!RZz~S&qmEeUJxR6T`mvw9T)|Vwuy;Iu(@>Bcgu$
ziEYc;e!?hm6#W0q|Ko2n>`y%Yv*dqCaMFZI8^+FQ<+UOTO5R6HN12rarh@VUQ8pn6
z9j+8JCS#vq*}3y+D7=m|zpH(j$1NWb!~Sz}7x~ZF#&ghnnznHTRkSOHePxOkHRC}$
z2lpc&KHWWk*^jZ)Z5#I1MKrTt3`HmG@B2-9HOAugVx;~|V{zXYsej)QxqpPzLWeE)
z>&;iztE0J$ZVYR0?nZiai2eCRZvR+*7hbMI_eFr>|7;mgh~8ZI!(=fbgbzGV)_XJb
zBq(TWzaswo82R#-zf1!Et?#QoMjlH(^@Zevd3l$VBg#my>Yh_B2(Gw+<nG3zpxTN|
z^W&qdvEJhe$e{BB(9Vk!qaNt(Lp>QM`Zz|Q-6qwht>7gi{r>_jk$}SJm^3(CpFJL!
zl?YTw=B>1A74)Q;ck#FCF4x!#2j({I3xk8hcN1&anI|v(MPUo)gVT4P9*E5sEK4`3
zrGf(P`oT<#fb{+G^J9tvFdY8o8LNM@|8!&fk$>oZlk0KgQuX+m*^(80n7(IlS}E6O
zB35cn2<_1$$rv8JnTXAh%TrUXa>#HVCwvB7h=zP1OabHChpDQznP<ZeH%uNguUDkM
z)wgTMZmPW|X!wEW%`E(`@7E{4#dK<#S`~D44U{pG9bYbL;1^~(8?$EISvk(Ne)w?M
zb2e&P2}7a7+;&(*sg+rojzE44jl#)vrsGXI+^4KRz)YPk5)YT}x@TmC(^;Q6b7KCV
z>{H-vnLoW$?+0(~gqyEhC>O)0y^}Nh6ut1LIRa&C2?ijgFUl74Rp^#uL9>7=B8F~@
zG3OW(FX8$C3xk^>oXU|TWU1=a$K-tAH^Mt=C)aBB`)ot6HJj!`u4#L!N^Os|<xU<m
z^*1EP%L}!6XA<f68^>4Li7|7${gByg)^x*upZ@x)23nD?nHF?&aa`3*lW5Nt96uR%
z*@`HI66#$FW2Z|oveq)?!x3x}Xoj^J=4mTzeK=mf8I(UNnEO9aruu)B@wnq2;B()#
zG|AhK9R7Kr07BhP%-N$w*ntmtoybrR%<*5_i|t=4_(u*){*nIFG(Qz0dU~QKOGW4S
zBb-uKS8>{2T`e7N7mXk9x+srJ@!1e}GrAt;W|({V?zXQU+Z%BcvprK?WqSi|{uA4)
z7;r55V#>>-D;1>%Eh%{fVHU(G@z{_LIW9>I(a)@w{41kwjLx2|oIN`>j5BA?j^Ot@
z#SgMz7-MmzIG5g#+8@(&e3+5CN*{d&*75+kX~b!U{%nJJu^g~MusWcDusWc*wjJYf
z1ODoA7miMz95ExfX~fv*EFQAamiSTJgF6zBZi3U7%Xg%=izZ5KiX2NmE_qz<Yzy=Z
zVGlioU)+ZuE*<~Pt~2sjDf%GJ!8ks0HvFmZW9Y-TPvehPzsJ;HR@v(r{G|uq+_T-G
z{~s*fBoFA`jjos1GL{6Z%1t@1KT1r(8jejP+zNS=tc}-){Cd6g4ZE^=;jX*hb=O^w
z?$=NE$n9x-;9Yl}*}tzA_2~)R{!TsoI`ufn#gu=(U;Qm*+N~Ggw|~X4QXWLT?eMT$
z+Z9pydi6$l9$22_IsDqbmDZ5=qVXdv<#zuXOZ&I-0{q_=*z^7Ka^HT*UT5|Y>m_1b
zW4wsN%I#<&7~`rR5|0bIvMv(Y!rxHX0gz&Ali(ttJR1*j`EW&YKZ&S~aSd6a==yM(
z>M}q}mmuFf*Lxpbm~o7D)1A}TDz&PrF8J>3hb#w;0IcR}8uuHvUK^MyW8^PsWf&P|
zwW3vYM-LFR#hIl(VKqFLTGhbmB~1gSo1Q}mL(1#Qa?9(GRlhzkd911GZT>ubgT30p
z3P^ccV)4eQ*ojQ4+6Qj|<;FLd<`}J6YrzTvt*Vnh(M(3DW`JC1YLrxU5!d>sKUMw;
zzK@Br0y62ll#dA(lTEA)>0}C;R8&j4^k$S(;S_=sDVoo22I+Hs#L=%7=Lop!QMegv
zF8v#?jmBI+7zGM)Mi|~!$7`GLGI$^nvaeW!7nnJ`9a*jDvB5nR%5{x$^|%J2DkbQw
zW1HX7(blG0+M@8fw)9QYV&uKZb<50_Wi_FA_`lxQkp!2hOl6h^GErYeKs=>k!r(tB
zJ0s+ss<t6aGT1g<W^lM-a#M#J-@~ZKnMxvxHa*9MwowT^?_l+Tth=b{oavl+RMS<%
z)GAvJ%+3sS8?;?PD}z|2Tc@$rA!El8I?vcK&UFyJse?grIN#KDny74sB3-Sak6?5|
z({Lz{&h)ZBGy3Bl7MS)D)#Y#}20M*QB~P!xZHfGr3qQcG=$2#Ym7o$&IV>*-Rc%2m
z;ld~UvX1d}bhA=<AwSxh_Eat?AbK|)(`{S4w8W!!6cH~X;u`c9IZ^Y-ML!9_XxCe2
z!*w}hZ&LIKSRM0d{5jfbE>|oUSF@pZUCTYvYSC8P9>lIRpTRk2V7HpByWuNHh)TV%
zYJ>p|Lp7|q3>Uwt?$+Q}NCV%lh91;2At&!YoXQSMDF^w2uw+kxiKis>9FM6FQ}_6k
zg*^3-7jT)u=<)VSyS*}jY#!MbgrQ=yO2{3WZ);ojeX6#Sf-v-DaX|E~^nuAr`)zaS
z6LXWxgxsmwzIN98ZCzak?X&!xqA0uafAmK4U-oHux%~dNj5vdvm&@jTQvLsbQoMRf
z0C=2ZU}Rum0OE?8jfdm;ZN4&aGwJ|E7;02EG{ETpKmR{w^kg&!ayb~7K&k;!1`J04
z0C=2ZU}Rum)L~!%k^g`Gf6VB~z{r3CI2ZwDk_3tX0C=43S=$bSAPjZ?v;Y6MiNc(V
zQIOIW4vGm6jfsO^PHS%)hGBTUpGwXy<o-mSapyZMUq|1E=?>z%Vj!@oM88@XJcTxl
zxmYX3n)Bl(zlsi1J~p}bQnsP(tI505HProfJvRM&iC`kklSk~r+(YFf?!EL}D&L`V
zVGfTN9#WpI#v^5mipPxC$%_w$KU}`O-(S=>fzE9dFHL{W#Zd2II!TDi`>}IUep>l=
z*j!!4e3%8Ne3{PNA0u#V%<hMkei)4lef;^^JCm`Sxhj@8-;zBTxQ3?WN8iu7xNNbW
zXmmoy5!ZWi{d5VQh`Y6NI@G^optGg@)>>>9*-gxJ8y?X+hyGDgH#D;p%BEDm+5+Zb
z{Xy7Pir2PB2z&n2lltu{ogutT{F#au3JcG-iky$ydn9Xxa-R;Ly^Wxj+5L%>O<|Bb
zM|gQt_#a7#Z5Ea6auRyfz*>qWtFt|m#I{;Gm0*8IZ>!k@hW$X6JZ0WH%lQH#J$Z!y
z0C=1|*L%2EWAg^^`L4qjLJ>kQAtWIxIv0vi*$7cO5Q<7~Qqe(_3hAtNN{S>2QAk3O
zN-9MNQFM^R8;THqAOHOJb<MTb^E@;6+;h)-*Ip~)|NoXm6l$?`5lJ()Ga{|QrZV?g
z8|EzU&a8`uN9<RLZHg$>Ct`oG`%jKIpfVd3abQIzwscdrGU6aU2bW?CBMyOICS(6z
z=SP%vU$$q&q3{mf8*$joh;joX4lm949|7ZteGx~>UEcjsgCmYc`DnS1fn8xs#D6-n
z<q^kLWzLV?9&ubFhC@YX$5)Fu0saZYB2HW%Q3-aXaS<osdb0hKT`Q}hvfQV-o?60i
zsnUw!c$&ISvsYDIRotpNI~||X<vybyqsf`}&)gMJjW*TfIIA&pJsan<`BlfEI-lxd
zYrv|BYfZ5=#hioJxfNI!rmkB2YORT=?OIzcb!b&*QAAyQ>f%_BZ#~-7$EUs=4fLj=
zJPpM*DrWMX*OK9NzIx7|9&v%|1+ya><JDNbjrlg=(`0wVh4wC#?;?F}T85d2i}mah
z94|Eo&E#(;Uvul`^k{+SWqNQKykdPVrt{_EE?3JH@W94Zu&y=}C`T)NuC;%idRp^u
z4XX|PuBU5T>$do`)35gG>0ll@z`cR*jWBQ2*N)C_vc8FSH`DGG2|B5#6D>N|WBA@`
zhHiC!n_9cz+tmzqb>B^G-Eh90KDXo9-F|oL|I(?4Ts`>QVMgwtVNbog(|#}9d*jnv
zUwW(QE_L6HLtnW4aO~&4zu5j}Xn@#z)G*K--P<YRK6wVwY>*s--QSPj{qrJ*z!-x2
zP%%Tz^Dwy{AkG8sAENbebNev8MyP$HT1V4uw6ig48f#7-f%yoW@%T-^VS<<v*An?l
z=sQvVi7+26$MBoPf0CLe%lVjjdCYHR3Jz1`f86?UKGX1iLXA&aKWUDp+nX+ChFO~_
z*DSea<NrLZpT}d4nRyYXxn^>n8F<CayvpYDo3FpG!Cv5gf&16pFLeKg9B;sT6VJEY
zzol1;)VCP!;tdgRyMM?1JL+Cy9+v3oyZF4TzNIu+s-~szmYIX)b0XfO*$T6>!ruG(
zxso=ka9J&8HGXSgtQGSi+>cy8!uw;IeB%5QHGQhS^?JHNuQvF7e5vlQ<p0V{ZlvSa
zYTXQDE3Vu4ZpU{AZaeknN4a+C>Cb$2)B9Jmvsa!!aN1}8Z}!i=C?x%&khO|JQMD+P
zst|<(%17bA^-(CjJqia`jlv<7qfn-M6v|p3+9?W$m1e`EP_9N44!1sHWfaQKj>6Fk
zqfi0PvEq-N6NTeiMxmnE<4dvSQ8-~-6i%$j_*HVP#OI`DY+V#ihI7iWC{%WKs{1O=
ztH3<XdDY@5oZc}CXTUvULln-mu9mS)Q8;T&6wb!0`jRNrkfSD^=QLuAqTu;ds8x^g
zs|~9Tzj|Wp!)bt1gZWWtI4TN_@HmeS=d0raIE}?MhSda46WTNpf1$mL#9bu+#p=Fz
zXA~|Gcd4^xayG}e`P3-1P(us#Tvm(WRxID;joI!fT%oQj<-M|mSzlEy3Rm;L28UL3
zYlWZZPT@MduG<xb)@pB!Q=2lZDT9A~B?jmEfA9B3p)G!G)!Np5JGt7)-5$^O>gj;v
z4bE=l+fgrWf_F2YTUe(yQRuAKo$<eQUKDOqcUS*XyXtE<+ICah?SrGxU9Y>t{4bmy
zxb_s+6URH%)=PXZ+<U|A&8LspKKR~6pS$4QZQU2we(?J7?@yBfYPko$duB#qV0(t!
zy)?O(Cin3jM5{qK4R(IN-1kq9!Vo-$z#l5lFj&L%-~m1l!hdLB6o$hbZf+h{#|S!&
z5IfS|$jOZFD43(<9}RCbtTB2uhVK}4j8)TEx{ssdIKGdVmq&1V1pat_<8d61n`c;I
zf?N~Cl;Bx{&qV#3D1M?@c~lLL%JXPXhKuJ}VG@3m_4P4zPqBX7`Qvg-g+HxN6rS*V
zdy?i)%H=s$cuLP__}x9tX2N|&pJs`Bmfvi3%@+F{F3*d7!QLENc*Yf8<o}Yhxq3d&
z`epWt+F!ArZ(iop<2AFqfF=vvzb@Y6qp(m<-_XxDVZY^fx=6l7xGh%qVw$||>>YJ4
zQQNz;e;2={@+?#1axu%*{T{#f-LHhblD4bxTBVlNus=}y8qbflc&_F55v+CUS?4+M
zvHefkXEfR%-&eS9a{jg7`8T+IV|F*|$!6CrW@@Xmt>U)9-=+uO>dST-Z5Q{Q{T=3e
z2jB1I-Kpju<op5uA8`56*^hj7$@h~we!}G!*n9M4k9vQF`J4Lp^4qI-zw`N>rhm}!
z&n;2#tStN`=AY@2#IQ&TrP!`W68GLcldK%;$JRxXmuJP16qR9ZA}Q5{EsDfDXR?2%
zNDl1C=0{T6y0rB{OCmXhZ<(f%l!fn|GAUO%lEbZ!Xc@_ogCp^5O^$|h%<xDmOpW9}
zD<e4$hvT+Ka=e@;mX4&7oG0-+1((VdBRRE3BvtxEa@xR1s?LeT`(JX#%t+3Z<E(nj
znP+Emw)?Z;R`(iR9k=T0s=>FWI%<kL$NjmL*uF?=Esvx&?sfRpg<mgYV(a5te_<pI
z8ZqY$t1<fxVKu_Lk($oK>AZQ7obSA`*d~qHs7Nl9_d@zyBu`WIUQDw~;9jD(OJO#H
z-E31N&7HTvsRcc}%O#iLU5r<;JjHx37k35RE9Gu!|0-Bl^SP!O+Ym{sl1Q$#z7BS4
z*EZrjN0YWaBWcIK9gZFFzhPV?H{#Y&u8xx<xk>(;;NQ&OdtB1Vey1&wbf(>{G{1Fs
zB>tTzU8^#)&`ob{7uTKc-r15KINgCmPkHW?x0m={bnLD6KCL3T%N*P#=iRXT>SJGX
z)KAX-6`1-5;4(mtdvF-2ANQ*NK3WWtZ;+l0R@?n%V2JvM&~d05hT3~T&Ie&UB!0NO
z4_iO%e1x1M_>YulBp##W9i{fs`ZUH&je$QF_E_^V7S1^L<IKz>@W-3e@nR>azeG<a
zngj1|$)m6)se7{Alh;S`n4FJ^ouU`s-;&4KRM=B#GR+)J(~Bq6{REsR&BSzZ)7AJC
zZco{pf$I$MPpfmL{4@PVp7C3oMY~!2pN0Rd{IkvTb8<fq^Le~p5ch)obJXOyn7pWG
zFX_QtJ)f)AdGvc(%q!wwF`KW_<yE}rn}hkZd=0+^&R$p3>wZfM;k*I!&410@zNMFM
zxh~S5#eCm31MldoSFU8qzgIQ9OTVS&X(_*DG+U-`%k6m&OjgLZ0`B{8-j{zRpH=o&
zsdY83AHe+p#u_}=%DLA4hxmPj^Ex%GQ{%_#_(V;gT7N3`Q+lk&bG@_A<oMkF=QQ|2
z?hQ2GApe)lJ7e;dd>e7wC}tC%Z)p1s?9KRWHji8McMF}jdj4$H^KJNSGc%sG$#%2!
zojSi$=MH{5aQxnk>@>4Gar!|Yew6=5zk&a$ahKljQrAy@3qSMw8NXlf`~{cY_V@7p
z)%^S>@9%v7(1Sna+^6QhJmdZr^ADeY_D1S^KP_yDG}#nsRxi@LC9_|&F4Fx5Mp~*k
z(*3JOdO)j453CVs>5OfN^q~2X9=s^hL&il~hF@9hL)%1p7~gW`B0U`Dk-H-;&-dsG
zksc%d80!k`KRqKowhR-0-0Vmzvg5^{(4Os!^u&3QR%#mQNphXU_as;+%W+C$#;*#_
zr*)6iH+Xt_u<4PW(TBl4Q|#GrtFMo==E6wN>Bn|PdTvRiwK_#wyJDnu#ME(K7e+nn
zdgAM;xqdY!wt+eu(xK6)NY8_NKD)r$vo&puM`L(R*hMfd=HpqKHltlLd(Gu*fp-ho
zmknd~iYG^UxjL@s80nSe8J;c0UB&0Bm62Y}TB+AFG`)@nt<`vaBQ`bCwzO<ZlXh_1
z=}~*<?bY9be}`R>-T=R2C8qA1X@3jfPW0@I@2%>(ZBC?J@asyeuDEwy6KOZG-PC!z
ze7DQhU5|YKrak!etjy@$OMI{Gk@lV$sb^{0$KG8yd+$v9>T^GQ`imdXEYf@Q@*eAf
zrI_oz^t~7N`^?UL;s?=n(DFzJ%XxnnMz<k24i!I?Mnho_!)cfv4Wo^3;q(D@Jvcbh
z;q-b~-6QZE$wtXPT8=S%#)==Ou5r#Eq49Wo<JC8TMkUTBn!Abc9<?`#o|EZ3#r_oa
zJ`QIp|7o<Frp_nm^rXG%_<2^QGtAly{!i=uOf}BL`5FD1rLI{teipaca?jR_=U_eu
z^LZRzfHMc)i!ffq>m^#vrN>;H=338FulLCGWxQU|k5}P%k4#^4zCiu2<GN7%n>2l^
zR-}vkzTS3yhn7q5TPo)=G2Rc;_vpC-*84D5(rG14J@e94;#bq^1A9Kn(+~CFW14-8
z_b1LirT2O{*W3FHzt3s&x!5n{{6dZmt{ddt;5YH5^DpK3$}DY^XOp_VHX~oF^&2z2
znZ}!GvPJ*4>ho4Ho^9ziv-B;l->PvtO}>NkowFTk{9f!2&i_ZVU9|m4?LXu83)^ju
z_u#fi+^@L*hRa@fznkqp%-^4$OZ(*Cr>4Ke{q6Vhw;0c}^q=*SMKRkMSz$?JNqe?1
zvUGN2S!FgevV34<MZ+W8uT^BFnnkvMy~qx5J#cVjrAs0^$l1Ys4ynL4L{_FxWMyka
zc4#@aCbGj|9kxBPa{D4X-2D+<B0Eyf^8ebr$c|bb*)gRUtO{~g*c92Z{3=dmdm}p@
z_VI8|Xw2|EaZY5u*Rx71BRi=)!|`NXPGObZp9<sDU6EB871?S1B0IAulfT-y$j-vy
zZ1^>Tsi6k{nx&Z>=cw!4Mr?XywfNTB9a(MhwfWb<xvsr>Z6d4Bw}HG3@N0lyL+eK3
z8maxf$&sDU_X4<$heg(8U1S%EyJ%5lP2pY)^HRB+(a5teYtE-RPAzb4A<t#i7@uPN
zFNb%9n!H11Ez3l9mAbB?%hj~<ot|CedW}4-aJn{Q?yrS;T~j7sYdqS(y<S{fy=kkz
z?fABX-+_J|=0|n|zBfAaO`dh^9@$MDBlAt3-6G#DFnpJ1zRR=D>bR{E)2l9UyTa@W
zr(4U&Zr79U=H|b6_7K-&OJsMLiJoe_Q_P+Gdg)bfK7E|`(Z9RoyW4qR`TDB0AMX9l
z*8sc*=+!-DVW9rp`>%<=Pwxkr%fV_MtS9%=#rtJ8)SL~a|4<kYID1gsLwfU&^>93g
z)AnK5Bh0`^{q+r=jgn)O^(eTbX*F6+qxEqN{$ptFT{9c!ew^3|t`o$U$X^0`qV**D
zOcFC0{$w1cRE_L$+@`rcp)SwFY&tEUf-^&(W~k?B`ps0|OqkEeI}7Hsc+WN)&(mWL
zPA}5>CHlO?f38`Xr*|)_<z;%lBHyd(p3i5#9>0dm0%r^8;QcUr9o9nEH^jf8hBw(_
zeS3%QOT;Xp$Gdnf#c?Sv%j8=o_cAlLTs__iv*lvmGso}Yvx1Lz!fcgU^*)%b#^nRC
zYv`~>jce7m7LO0beW;#~V1EQ_o%4^ye?s3+;jTBYpYi*mQ)FNIoqcKMzJj$8&rN3N
zYdU?ye~VhSz}TvvTlHvLEoQF1^?cZ_Z@#Ou@940DhCAf@-Yk6&Yp1$@!1+hmKYB*|
z55_J%-G%#4>igMz{vy{e=4`j#?bhc#wExxqZ|d2rKHu5dUf93u<sb6?A=jVc_p!g=
z{cZ1Wzny=2Mjln!xX24NB2Owuo|cL{E6rv`o>yR7A}`t<`F`^vFBPmy<ohp*{6KM~
zD@A^gn1fqIUPfHmO_3j3iw%qXu#B~2Ya;i(otK*(`Qg<VoFh82?U5f@5_x$!%ELRV
z5wqtzJU^x!b6;U+<lYf;?}+(vlOwOVF7o4x+1|)c=pOlr<r(gk#8z4u`AMB3Kbil@
zI9DDV`KfSDwO@r_6}48`75QnndS}e5ZixJJ+|IzK8oz3|oYj=AjQs3A%z1VEtIJ)(
zwT5d=*P3|L#LIVker`oJJ@Q&O)q+!-Cbi+zajoN8*R`&^b(ch54{m)v4d~PWH{bDj
zLtGo;-*9>4jp%*eoXF3Y=K|+VoL^Kg@{7e@IxzBPe49;;yg3b<^YMJmFKZKdvD%91
za=G;t%_6^&o>$`6QZKH8do`Zd!0~RGUyI9i!y|9qp2^#qPa9`#;9l=~y*<y>ye%H>
z#I{#!2R-iK{svdi*8B$A-`J0>kGvx-JHqRzS2y8#GrwEp=+v0GcE+u<^Ult0H5a$h
z={D;wX2UZ#?`p5BI=hLxosQkjV0YaAt4}@j;0`%^^7G8idpYZEZu-FWtj+J%v%B%Q
zTd(`lt1k`u=|?{`^w;11W}!c<0qVcU{y<vYi|2iO?=#PX=snnM-0ysdy`i`c6F1D>
z1A67XGk-|E!^J%;&j{L%gf|L?Z~1(TI>zepIJG^3(|9$FSH}diUjnPdyicUVM72E1
zZ<6{Z%k`K(K1SauwD640r<vVp@;!n7ldjX%I$gf$>Uv6_X6X0RdiS*1nF;$DeVzq(
z7QAQ8*R%GX-52?6JfEY<b8w!wp2J>(HP`IS<NJ!=?yF*7Gs6q${<>OUmv155H|X<*
zvp3cC7LDKXyI6$7Vlj)&#bTPi?fz}pOYmRf{9SpM@?Yk<obL+vD`@k+THkkFDfdd)
zt8iMS)>US9wd-oV^G=$7K%WoH(i(AV+^^N&wbmc;Sx5Je_2(0uKBe<|c)q>!&(!fb
z?w_me3pu|qzZ=B+{?5OWcOyMEnwyQZ+eEWX`tda$U&HuDf4r;an_V~4WQ+Z+YT7D(
zn>^p@$#xvJtL;13-#h!>41I4_cFOxB-n;xZcDeop<0lwD)8c3I`!n9Z;O*Tt->uF)
z?)T94SAF```Zt_@H&1_9|0(Z2dH&M(zvTQ|KmKVGMNuz`3XP&DsT4)&zcw$5vQbf#
zSB;{g;waj$3|kXLrRGP`{@bJIfXQrq6dkxTib~IpqJxG-(ZT!<wmzgw6qPB*`bAM$
z*RtKC=+M$^Q4}54nC*_DaxJ6iaQjD=Vsd$REjmh$qh|geH|O(W0C=2ZU}Rw6U}TJE
z&}9GtCLrblLI#HaU_Ju?7O?@X0C=2*kv&esKoo?>`9s(eAR$pk+klWzoF7r8p-Vc7
z6zOee?KrZ%)_M~u4JY6voPZN>1Wv*Mm@HelAp7L?_h#PgS~7qee8IzMdAPRwX?1YH
z?vJ~qJI6ipz2iOtJUbpxe{t;N39pU=+~UX+yxt|1A>JK#aD@-YUFx5Xd*pA&ect-x
zcz~hjJNB{m9vugG@ZMsjOk;FZkMcxS%}QqbBGN6j)vl#(a#e|GIB7XcSxFrkxe@VE
zG>2?vOe#{XO0iItkwu|It<_E@CfpiR&&T7`>0zQu#851QhL1*s8YARLs8!TfkjSt{
zK}VmN{oh^lB+Ykjdx0rJOwMGM%v3fP(U;gT7xVuJdIx^jjH*G(KIM!;Nm|(KX}Vx3
zDz)`?R1)eTwl-B`jxj53&4>2(@)y9?b&vo60C=2rT?KUGMgr~d*p4BzP-afsO}5O;
z+$)o8D~TK1axFWsWoBk(zA`g2Gcz+Y-H@b_o!j?f{r?9wjM~}YZ2BLXZPI@n00<Dk
zfB*^@;2;5#AO$)=8e~8g<Ul9r0^Oho^nxj1DwqbQgW16xU`{X>m>bLk<^}VC`N0BU
zL9h^57%T!71&e{j!4hCe&<B<ROM_*=4A2h-z)Y|#SPm=?Rsbu4mB7ki6|gE;4Xh5<
z0BeG^z}jFPur62+tPeH-8-k6%#$Xe$DcB5b4z>VWf~~;TU>oosur1gQY!7w-JA$3S
z&R`d?E7%R}4jhmN1yBSo7z9IL7?i*sU<8yw1yq3tYG6-L2R>+kCKv@{U>r<<Szs@)
zH`oX43-$y1g9E^U;2>}?I0PID4g-gSBfyd1C~!151{@2H1IL3Cz=_}_a56XroC;0@
zr-L)VncysNHaG{I3(f=QgA2fg;39A_xCC4ZE(4c?E5McDD)3)$HMj;`3$6p#gB!q&
z;3jZ0xCPt_ZUeW2JHVabE^s%v2iyzp1NVamz=Pl+@Gy7;JPIBIkAo+`li(@vG<XI)
z3!VedgBQSy;3e=fcm=!)UIVX#H^7_VE$}vY2fPd31Mh<mz=z-?@G<xVd<s4TpMx*J
zm*6Y#HTVX63%&#2gCD?;;3x1i_yzn5egnUQKfs^hFYq_`2mA|Xg8)K^Ach1|$RLLa
zn1m_V0n;!8voHrcVHfO%J+K!}fm7i$I33Oo=YVs<x!~Mz9yl+Y56%x4fD6Kf;KFbb
zxF}o<E)JJ~OTs?56kHlE182Z~H~?qDW#MvgdAI^x5v~MRhO59;;c9SoxCUGkt_9bI
z>%ev4dT@QX0o)L71UH78z)j(1aC5i?+!AgDw}#um|G;hGc5r*R1Kbhr1b2qJz+K^P
zaChjyJS@N>bm1Tzg2S)`_kbg?3@fk-Jy?T#!aDR}12*9(9E0O<0?vYa!M))=a9_9|
z+#enQ4}=H7gW)0YP<R+T93BCWgh#=n;W6-7cpN+)o&ZmTC&829DezQy8ay4I0ndbI
z!L#8x@LYHvJRe>FFN7Dti{T~kQg|7>99{vhgjd1;!mHsm@LG5sydK^FZ-h6&o8c|+
zR(Kn{9o_-&gm=Na;XUwPcptnUJ^&wt55b4wBk)o97<?Q)0iT3V!KdLf@LBjAd>+04
zUxY8gm*Fe$Rrnfw9linIgm1yO;XCkM_#S*8egHp&AHk2|C-77F8T=f60l$P_!LQ*r
z@LTvD{2u-Qe}q55pW!d?SNI$J9sU9Tgnz-m;Xm+SG#dg4B7`s^h$4nKN}wc4p$?Qr
z8I(mi)QP%KH|jyXXbPH&rlIL*b~Fc?6U~L@M)RO~(R^rrv;bNVErb?Ei=ai(VrX%+
z1X>dHp{3B$Xc;sE^`ika6D^CDL(8KT(28g!v@%)+t%_DdtD`m0nrJPwHd+U*i`GNy
zqYco8Xd|>S+5~NiHba}EEzp)|E3`G*2K@(Zi?&1CqaDzWXeYEY+6C>3c0;=(2jx)#
z6_JYu(GVI&CA0?`L1k1yRpg->+7s20j~b|nM$s4=M-ylk+6(QC_Cfoi{m}mC0CXTa
z2px<LL5HHl(BbF^bR;?o9gU7b$D-rV@#q9}B0343j7~wPqSMgn=nQlwIt!hR&Ozs*
z^U(R|0(2p|2wjXWL6@S-(B<d~bS1h9{TE%0u0hwL>(KS+26Q933EhltLARpY(Cz3B
zbSJtC-Hq-+_oDmI{pbPoAbJQrj2=OcqQ}tV=n3>BdI~*_o<Yx|=g{-$1@t0%3B8P7
zL9e3M(Cg?8^d@=>y^Y>M@1pn6`{)DoA^He?j6Ol1qR-Ih=nM2E`U-uGzCquj@6h+?
z2lONQ3H^+ILBFEk(C_FE^e6fY{f+)X|Kiy&zz`#hF~Jlw%y9xIaSC_fG|u2G&f!kn
zg}ZSN?!{B^R6Gq&$Ft)(@SJ!qJU5;P&x_~7^Wz2Zf_NdkFkS>NiWkF+<0bHtxDPLd
zm&VKB8Mq%0;F)+?yc}L0uYgy?E8&&#DtJ}A8eSc*f!D-q;kEHPcwM|6ULS9OH^dv^
zjqxUUQ@k189B+ZQ#9QI5@izEBcw4+3-X8COcf>p4o$)SsSG*hE9XmLW3%H0~Jcx(z
zFfQRe@CYvB3a(-g*YKXWj(yy~O+1Rn@Hn2pv+!PcZ@drQ7w?Dn#|Pj8@j>`td<Z@i
zABGRdN8lszQTS+l3_cbghmXf6;1lsl_+)$vJ{6yaPseBAGx1sYY<vzr7oUgE#~0uW
z@kRJzd<ni3UxqKoSKur0RrtU7YJ3g87GH<2$2Z^`@lE(<d<(u6--d6;ci=nmUHEQ%
z555=QhwsM^;0N(T_+k7AeiT23AIDGNC-GDGY5WX+7C(od$1mU)@k{t+{0e>*zlLAO
zZ{RoaTlj7K4t^KEhu_B^;1BUf_+$JD{uF<PKgVC-FY#CSYy1uV7JrAo$3Ng7@lW_?
z{0sgS|Av3Zf8amyU-)nQ5B`_TMgV~X5ljf7gb_{>BuP@FgQQ7@WJ!*6k}lFsdPpys
zLZ*^wWICCh%t7WPbCJ2pJY-%nADN#lKo%qmk%h@3WKpshS)43EmLz>-DY7(KhRh)S
zWPr>h%aY~D@?-_FB3X&7OjaSQlGVuSWDT+=S&OVq)*<VX^~m~U1F|96h-^$YA)AuT
z$mV1VvL)GyY)!Ty{~_Cw?a20I2eKpCiR?^vA-j^@$nL};c~T%n;*vo!M21O;>_J9I
znN&!Xc%(-5Bz5AG25FK}GDgP91erzlB72j4$i8GhvOhV197ql#2a`j{q2w@fI5~nG
zNsb~%lVixS<T!FXIf0x=P9i6hQ^=|0G;%sQgPcjuB4?9x$hqV^az43$Tu3e=7n4iK
zrQ|YlIk|#dNv<OQC0CPc$hG7;ay_|$+(>RBH<Me)t>iXxJGq10N$w(dlY7X$<UVpg
zd4N1f9wHBuN64e(G4eQhf;>r{B2SZN$g|`*@;rHgyhvUmFOyfutK>EEI(dV<N!}uF
zlXu9w<UR5}`G9;#J|Z8J0r4je*iXo(<TLU)`GR~&z9L_fZ^*agJMumGf&55*B0rN~
z$gku#@;mv1{7L>If0KX6zjQVVD5QvDN+_j_a+;t?nxY*vO*1r0bF`Co(Qev9d+8K9
zl}@A6>Fjh4Iwzfr&Q0f`^V0d~{B!}jAYF(qOc$Yx(#7cFbP2j7?W0T4rRg$s2JNQ<
zbS7PvE=QNAE6^3`N_1tq3SE`1Mpvh6&^75=bZxp0U6-y$*QXoM4e3U7W4a05lx{{h
zr(4i1=~i@Wx()pg-Ii`gx2HSM9qCSVXSxgBmF`A&rw+~20xeRP4$>hyOiOeRIzr2|
zLaWrHHM%FQQ=c|ylaA6cI!-6(EV>uno9;vRrTfwS=>haWdJsLB9zqYLhtb375%frU
z6g`?ALyx7$(c|d}^hA0RJ(-?DPo<~P)9D%XOnMeQo1R0@rRUM}=>_ycdJ(;tUP3RW
zm(k1V74%Aa75y*0nqEV%rPtBx=?(NodK0~w-a>Dsx6#|_9rR9m7rmR_L+_>c(fjEG
z^g;R%eV9H%AEl4c$LSOFN%|Chnm$9HrO(ml=?nBl`VxJazCvH6uhG}(8}v>37JZw(
zL*J$E(f8>G^h5d){g{42Kc%11&*>NROZpZ4ntnsSrQgx-=@0Zr`V;+`{z8AHztP|6
zAM{W97yX<5L;q#7F~A^03^T$gV~n!|OR^N}U}=_NS(am+tc!KC9@fjIu&Hbso6cru
zbFewtTx@PO51W_G$L41Xum#ydY+<$tTa+!v7H3PaC0QR^iY?8SVKZ1i8(=fpvTQlF
zJX?XS$W~%2vsKutY&EtzTZ661)?#b3b=bOWJ+?mEfNjV&VjHtf*rseVwmI8^ZOOJ`
zTeEH0f7rHcJGMRBf$hk4Vmq^4*sg3hwmWlJo)uV;xonUPv0+wXd$18!W))Ut9;>lE
zS)KW;!J2H8jj?ey!Dg|&*xqa(wlCX{?avNi2eO0M!R!!rC_9WD&W>P5vZL71>=<?|
zJB}UCPGBdpli11Z6m}{*jh)WUU}v(k*xBqHb}l=QozE^{7qW}k#q1JxDZ7ka&aPlr
zva8sC+12bCb}hS(UC(Y{H?o`9&FmI-E4z)|&hB7$vb)&b>>hS6yN}(^9$*i$huFjH
z5%ws1j6KetU{A8A*wgG8_AGmjJ<ncXFS3`|%j^~QDtnE+&fZ{evbWgV>>c(ldyl=(
zK42fRkJ!iT6ZR?ljD60&U|+JY*w^eE_AUF4eb0ViKeC_L&+HfWEBlT8&i-J3vcK5h
z>>u_o7xO<3IpUZTPC4V8CwP*lcn44O4A1f$@8n&)oA>ZuK7~)^)A)2gJD-Ek$>-v8
z^LhBZd_F!uUw|*j7vc-^Mfjq8F}^rof-lMY_)>gnz6_ti`}qK$$(QBJ@#Xmnd_}$z
zUzxAMSLLhm)%hBHO}-Xio3F#y<?HeF`38JLz7gM;Z^Ad_oAJ&07JN&-72ld~!~es#
z<=gS?`3`(Xz7yY>@4|QGyYbz*!}Gkri`?ade25S865oT5@G`IPD))Ge@5$@j=MCQE
zqkN2y^9eqS@5T4#`|y4Fetds^06&l)#1H0&@I(1w{BV8*KawBCkLJhlWBGCXczyyu
zk)Om*=BMye`Dy%geg;32pT*DS=kRm+dHj5S0l$!6#4qNT@Jsn+{BnK;zmi|Y|I4rD
z*YIokb^LmM1HX~q#Bb)e@LTz9{C0i^zmwm^@8<XLd-;9*e*OS|kUzv9=8y14`D6TX
z{se!LKgFNs&+upYbNqS!0)LUe#9!vG@K^b3{B`~Yf0Mt(-{$Y|clmq#ef|OekblHK
z=AZCS`Dgrd{ssS%f5pG%-|%nwcl>+)1OJi##DC_$@L&0F{CEBb|C9g4|K|Vje-pDM
zKmyK&X7mrFm+32%>V>k~H&`l{dBBA1@7Z+fp{!YYM$C4=glyXmSh_!EJ77Y#Z3iqp
z5VIXHA=|bCmYx~29WWu=wgZ-4HfB3uLbh!OEWKRJcEE&e+YVTI`Izm13E8$Cu=ENs
z+W`}@Z98D;6=SvoCS==oz_?RrltxR9iC(8vua%vu+viq?N>$fa_HwOiIuw*Q0ZTe%
zr(RJSQBeH4<4%WDE)7-t@?N9iRSYS()rMP7XyR6jMy`~K#j=~y#BVtDhOyG{YE+<_
zGtuRgYr{_7ZS*y3HMd@Hd=Y&kA*bA+PQ{t!RgqIEGN)Rsd!-^b&;GPitM!$t#Ztj(
zcy%Ng5r1X3!>JdBOQZUAm?1f*UiZfOR$Qj&4)qniv1&{xyMv8RTd0?Yh8r1MY1RzQ
zJ9XuOMWyp>M3v)?h&OA-uu%32BV#4sonpAxlnK`=OW*Ab?`)IjuoM}%ZF|b(W^GQa
zqSNL?n`K+%IW4Z<(GGU%|1oTLWCh&rNE_x_bzAU<y?VorvgKiCo1C_kW-S>p<3nrm
zb+*YlOR*!PQ_6}=YqEB>$;n7D<)iM_Tqh`db+^&1>$L8QDJoc#SZyia)vkBil8R!?
zu@%Rzc0FZD(==`j*S+S@aNn>iDzS3cJ&8e&)|xdtcG(tjddOQ-zGpI%7VB2bdnPkU
z$Hdt~)|P0!lNz-;u!3uKpp7zdHKHofqbOP)Wm`lZa2<l*=TwJ8Bnw`(Q5P+FcT9E1
zYohs^X0(D=E;}{f?HhIL4bco5zaxVwhHGb`wRBAGG;FY0F8hVL>sC`nsd%Gq;AP;J
zYToJiHMbxtgwrT_>b*K_g*(1z*h>BgbQ(!#%&8YmM<OB4JB5+7{HcjQ-RlG_6M%lD
zCqfz;48#ovLW7xcgPF>}7tl0v5v{x8ZG2Nn+vG&3h&UF9+`fTg5J%07JafdBXO0+o
zg_yiTAUiQnoWK*&J=k*H$c2I}7Yarmj(IX1c;d%oKad+0TW(a0JnGr<s29kwn2ZK7
z8H>eatRPZ#sIM^Wnv6??G%Zol@rMKZnkgU^<uGbM5+PL%YvraNIg~6Z!MH9q3L^2_
zk$1{v&#UUTD<F}t>6pX}6MkwJ!i%c#qFQuHI?0$JqDWRpi2RWStuEduZ0I6dE}1b>
zCaz^8DoTCLPlP;`cl;4odqg$v(2xEgctwmjV2cB}ywebsXhL}<Mr!iqwIFM1NQRyA
zATxyJ-C?I%5JlK6mvw8SkY5h+B^hxnA<et9N^YI$M?cG#3KOCs#CZ{3R{VH{rWpJA
z#%;*6c{yg;v*|Tl)zaoHu3ExGzT9*hp4S*o<h|lVU&9?2n&$<jMj{z=EK#@osW|l!
zt6*i~T31NJB9@Lx5!1OAu&8Wg;#%ZmSoYE}DY78Pt}Pa%lnvRqNuQAO0EU73(Tgjp
zkU#8}2Rj6+<h^l|`I<;~4PW#CH9csLXx)`K5eg#IdeN=-IVySz>cM%y^ys|9uu`?^
z)>DSatP8B^(RyIbYg%sffYuPdF;RAdK*dNt(8o%}#xT{SCoe{}MNx$M<a_D9)$g#y
zZfVdmzU*nybe9?zr$n?=jhs$xmzC?zHa)JWp;k;$&U#{!8BO%IippxXMYZCqrUemg
zjgaxSE2acV9Z5VW8fgi9*9ls6!;k=?G>rF~`Yusa=#vr05)$#_is~zd7Ggda^wLyw
z@hFL|FC!lApqz`DG8@ooc@;e|XB1A$jlN;QOm%BFnA)P1#oOpsyG`%0q|nc7i)e=t
z_?3xkNkPlyl57Ff`MT#6MWh>jwNf<^GT}muUSzEhBiD*3?uNRecgqH3uvB*kWgRr!
zcLtq$N%-D0O%G8pm2VcJ)?HzqZw{HBrYYL%W~r<Syi+V@LSlN$fa=C8WTJ@ngv=c;
zl*J5K45&o84Mgd^^szgxm}WAjNxxeq5y14gWQEdVc}SF-7D=}AEg`GkBp27joHs6v
zr(--~GK`2x@1vwrKBqz|=5HYuO{{^T^`MCjsIVbMwC;o}YzVWu7*Pi{3{4ke3o8K=
zvc0eputdQb(Gr#@DSvdWBMN54@K|<_4NvzL$%uCg=!%g>-R|MyT31MmPD2Nij!B&s
zo6tjUTZw`Y!&vjCnYb4Drz&m8tUfZXMOG@Ms_7&%am}(K5_GuLiqxVvi+b9a6!}pX
z(;T<O52ZzZjJfVeiyu<ThHN}ce8YJFdjimJWka?_?2dSt5t+EwWz&gxWRzqD6lF%X
zMYKFs^Vx_-`H^iAElW|3WLrerQkW;%7SVdmFjq2h&GIM8mu!n@`4i?$Ca&cos`Dmg
z35gJg!3|87Xag;YHV{v=fytsc&=SRgNusE=Fct%$<C!`@i!0AE5sx{XDPF4L)*U^m
zG+0cg%YIN*BPB2%Efrlo8S9NH-r%4f^|h)Uwm>yu`q3{+V9l4!r)jhd>yV%+TDY1V
zT^b@@qHZ^cA(aM2T@T77ztN$nD0#9yO)65VI76}}6j0jGNRIABLe)iQsK#DuzHM=P
zQLIf)MvC!6E$CQ&v@NW)$;n8`X{c9er0uD;U@v{O>nTf0Yuu~_1<f#fDW^r`0hb01
ztBpxUJfmPZiAjW1BU3g73TtH|q$!1T9ikdz`5^MAU;PfK-<iUV2%zAVs!4^!bfqwY
zB^0ma`);EV^lgS=H8?87`oly>rGfR{iI8T*+*D>=BZz81HPVSBku@TXc(;No8$&NL
zam}JS8$xO~l5x?pq-UfpmXv6PElY1}*lNBSQtdeMEE#bfm@Y)&OJrL_o9pTx@#sBr
zt*UJ;3Ov`U+EEDKCEFq<Bd)zQ`5DGJ*Q&+@T^lVSigst49J3fjyt7S?#UzY+u2oH0
zg3hDRZX%>uqJ*w<jtWd7t2lSq3fH7OWb9e5*<mY0k)-<VS95FwgOQ?U+A`Hl<cB*d
zT2r<86RhLrY*Ds#UACZCVq<pHo|G-<>9DOdYhh3YuTm4C==npdsAjLNqVle*Rc+RC
zkz`h&1EJ_O^JP~B(W<HGdyA&$yDeg~))6UV)$nz60)%F2VIm;Pqh!FEwUm{q0hVh?
zX@i5zP^_m=F4gj0*sYn$5f{57F2;9`X$)eTMRh|IQoA9Fi<TvOv=$ewu&seuOtYNS
znqo$ii+C`giiAXndqZT8cXO?3k0pd}pKTFsu3NB(+bvj&YFmqvAlIteQ3w+u+alV2
zsbpyLOR1ffFm5Wer6ePgEQG_Qw1&fCMB~*bh+-l<uUWJgq)d2mwc8fVXdx5VjGHJA
zQkG!cORvhF*(J!twJt;JrCGDeZW!T6N@`{~9rT7w`xww{i26+*7II@W?3Bv3k>sJ#
zEH`d6%gsfw-+n74^k`gG%QL~Ge|oD}cS_ZuI<=c*TSOCJRE|=XU@TXH&4FaZjZs*z
zk`XsXVLW;*E(-AIgq`P+nv4Wv7Ok+SEFm;>%`#ES5=_{B)huQuBW^O$Z&vM06tq*L
zW-Tl#9kxOg(Si78n5eLpCM-$3gI9FT3X6uS*~AiKIdaU(T|~DamxW9oMZ8uv^WJQW
zn2fmawcM;!{k|cm#tatEN<}sFvcK_l9GM|Ptcqwf>ZO`n#F8XcA0&OO(}L%Xlw{0m
z6TDDsDwjxrsfD^*EQ!&zZ2kKC^1+s3SGztfE=3cd?nw-Cwx;tg5^$mJ)e_>z_eCwK
zCqvZF3#JX|kYLzrm{-&!A)j*Dehd|4yU?uH-D+W?FJEftBoBn5+`+A<L+&=~UT4I{
z#b8HGMgyfL*VZFiZ`NG5Bcv0q$c<**Hw#k%mcu5hahIBrm=08o{u*vEXG_YNB`}4{
zL)u+5Y7E_31FDoIV2L|qkCK<`%W`2fBzj>qq-tR_Vsc;;GJ9b(6lGXyVKlDj)wQ^$
z7DihnxiA`+?1j;|iCP$qOKM>>F6lNPu8GNETo_Nsc*NAgXvoyUXvlQ64QaEM4DmP*
zV7BOvmI`v8SQp@A!~-MWj~fY|DVCg}x>M;hJMbY54F=){104cYysBxB0;2XM4M`QH
z=QDKkqp_CyEva8i1C}(PrJ0sAQ%lQQ(z04w&XSfBvGeuLHI|6UAFo~%vGc>Wiy4wL
z&zfh3F)2&v6<D-L`KFk}411!x6-|Rv?zYt6WJIp$i8>`3SKl=ah9zVa_G+$a(L;(v
zrQtzGpD5N$vyLU=qI=Kh^Rl*y<|glrcgSbi^d%wDDXmGW*c==*^_6POU9;ee1YqJX
zFFJ&zD+-A2?TLaZ^=tA&V=WC>(g1gd%(y~<ddV#!fs*bX^5}>5O~Ra8@%AXmLo0Qi
z)+tNqHCT+bswIEeq*ks~H9}F0aAJNaVY<d*+E90WUsTYtYkHHO$xMu2qG30~xRfPy
zX*^UfRf5AJ>6nxanHgI|+`Nk(0=pBz8Ov7H_%Kp3kxWLsTf?%`92yP=N}0H3B3N~s
zqUR{v5j2ts&##nB*3W4R&6-~-y3r7J>i;oJS-N>IG2|F3%O#@Ndqrw@Ak=I1l4-#*
zam~DXBPfN*h#RA^Qgy^Ol6;z59d*m1g0zmmyC*T2(xRCjxU)^pMT)8EmJs=D?a{=w
zu8>Bj5Mt8wkXe0)DVCF%M2_RHW^KsCwI~8%Q!*_sSx59HF-XU>$VSbnxjK8Mw`h@n
zJ(HPa;$jrPXsahCML|X*=1g46hScojM4SgO<<=eF#F%PKUB4irz}?2MTd%s}RY$E6
z9uHVn0KXCCOETh9?L4y&ShnlaY{~Bax+gKn*jjlg=GH4ToFT8;?$#K@;$$ygx9ihw
zNpw#7#GuZ(Nla3f$RutS-Lz;m%cVjoNHfDE@I-wUi8~S0@d-Nz6Cp(cCB>iYjzoEo
z&@>f%P_(4-&<h%Ls=b49XMWJDtD}MWhM8Nps?%bEl_lMi7<9%o)kE3}8+096G`M;8
z_YBJRucdj(wuoI8H_b{qCUq+&SL4KBsAdXuT+3-%syabuXbC#NeBE_S8yInIWbGnE
zT!)~hZ7Tgkstt-IvxAff>6{9QDyt4CwY>Gz|6@4&B)Un3-bsz-h^g^ZnKHTAw749l
zQuibV@rXvL*43`ZtwyQX)vm{57N%-vn;f?orgCSS91lDiYjw5jEmp@1lUtQ~Je>A4
z9SmB#&New7irU1RBow8`{S24LI@{!_ZA$+neky%>Oscr@(uRJ`p2<w&K1T~aUF~{{
z)uvx6OlDQ;@giF*noZVLHC`Y^BjyL~^CSlG#0sYtoo#ZfMe|0j7Mp&HMbDosOQqcM
zI<>}RuTyKp>u|kP7!Eg2dM7oCr)a%dHUspoc0Ha{ep!qp{YjEa8_X5g#PHlHigCV~
z%}o1$rt$O$<KEdO$1RR|rW5FFljD&ax0>uwi>J)2Qf-p76>5hqWDN=GdNSh1D6HGb
zi0c+Qib7b26Cu^EqdJ?6xONlP(L_kIA?ml>D6SL4u7e%6qFqLHE*WuS6xOk1#C4|_
z)F~b?onlrUN?C%Ad{}1^Aq@<I4e*c}gP=|%LTZ9Hb$|y<7ui^uPsb$vMo}40MqEeT
zD$^-T(B-4G4Ch)^9p7lPGNjrdETf5#>OjmEzs5SoK^f!$!xAWm>Kr9eNabW>Le6KL
zLMkGq$RJ;-MFs`SMZu4e(TEg1oxcGYkBr=LHzGr&!N}gQTe4gYH!9_b?ct0%k+VH&
zLs5+O@GWbikXY7yE8G!xA|jQU)+<$tmO;^SQt_-;s?-K-GBYAxV=yA32wP;hW|8WU
zMQV+-O{`FBeldz$&5Cs08H)_+`N+L5hRYTc*%I0Fha(UCJ<Apu9g5sE#0VrZYS`)4
zu-~pR-iXMUGaC6Y=Gi`sm#qj1V)-{-u_7^1Da%q>k=3pt&)7x-_H;~28Ky468LaoC
z4Y6iA>6k<{6CPBDT)DtdZD4s^H!)Z?_e`)vsX;MIteO1|jXU(iT)Z$uKF8ep4D$@Q
zC>vf`ot>A%!;~Sqsnl4thk=2b*c$AML<rV1GI5O?I!-k(e)X%lI5ULE7GG3EYZ}80
zA!-8KJJ=MxNUd3^iIJuobT+wGH4_QCg$Zajy5}WoZ>f>A?tKk5t5xG6)i^N}tlM-e
zG$t2|3Z5`3G~8k$)UZTo$gSjt+^V{cP|25unqJdS8)49`I3ni_lQ=Mui&!Ex3~SV#
zsxOGAPKT@aH-rzF({dslFCiKmZHy&CL!|~4f5XkZe3YGhW~l7tMblStMPu*yJ%;)v
zr_X81Etqo2nWjZ~LqaaB`ChXztgLiv1G(!Wo6kY%1yLGRzx}Bp&l@t`71fvz)tYK^
zD5vK<YIdb2M#)IXXecI8%7h0rJyb6h`(~I^{5)Wt3JeH&$lL_!uoWJVz}`Tcy#d?a
zKtchXY64}&Pt97^PAeW>6O8Tt@PMUbn>##Ubl9ngw9XF8n}#KkmAVt_wYbQgN?sS&
zRy7$3+J#bmYu~+9?Y4zr-#xB%NE+NfV}{^ic`Gg06Uj+XbsM7ZHCywRke0X}-cbx#
zMgn!BPFJmvB7p>}lynM9l$3`jYr|gE^%eBRP+_#r3{2_OHC;%oBXSzbb^V-%(RqiI
zB@l(P>epI9h&WZyPY=7bXqhQuG5{X1j$wB^v=b7ww_r$0Uc*qk@oqgCV&S{z*GdgD
zmgE7;4SUVHTKh5gk+PBC*UF_vI^qqlmIzr!AiXtck~g3^jjUr7mXOsrT$fEoJTUTt
zbNLYujZ7&^Mtq8ft?Rg*ZL)DA4UKnUo0*KbiF}mFDN8W!!f|{u;>MkIzQ7XF8k^G3
z1J<E|xlv{aQBUpKoDI+z4%ZXI&AM#js%~O<qE_`9rCDx5PdG%on~bI&L_w;4{bZ>c
z+)W}1cBo(DJ2NCvsusMVYN_E-S-q<bUaE@TyVRJ-3E;&xm@R=RTprSv?A>Hbz!G<s
zs<mdL&)U=OuoV%$sGM@AQlN&s;9F><_Yj4xDoU(V&NnMvAqyA2a&f6y>kWk1HN#x1
z8hWhCWJk!nqMHqfnP-Jo)led|D#sF`x4|?eG*!n^12WDG;^yfKFyV`~TZAmSxYFQ+
zK1df3HN;|urAC};a4(i>%*Ci$cd!Da#b0#ZjR%aDMNM?;2~qlWslyVb%1%vGV(pp6
z8PPM!yNIG)l!KSFre-LUQq0V&43<p|)p$<v5Pfl<+?tXLg(3^(v21-myJs@PDJ`P0
zn~BRE0hR5~9EParS&O^wNz7)qwJ}ka)%O{c^{#fKAeN0`F|z72$6u009BT|_%B4Xu
ze5i}DT3z^&QEh3yS=RkyS44x$0O^>N2w87*Xi8laHpk*x)lB>-RwATWUSUqEG3(4_
zi>~4$T>gl1T20DB{^p#ZhUQR@dGiiiOerF|Ix#e43VUVkfxIs3ClqWr{)jegin1fG
z5QNPhRu>k^&7q;7y<vtV%gx|^7g3Oi7N}H6iQj5)15e|Qic>19W=Tu~m?3HN>{lue
zm;k!SA_3B}{)Qa^4_HzHIV=@3?uv-IFz_2*(W_EfkDHZD#T5fA*}>~{4XH4%c=d^N
z#jQ5`<fm%1w}#~u$TvmDXilElf*NJ3Y;09imXJ_bMv^JitO>>UT{$zt6}RE=idm>h
zOHgSBhyDYG1jvfNy61c9%)3+Z6CoESVwA5gxqD90E%JaTE223wUj|Z;S3HrAO)(1#
zkKbh#QrWJ1w1{H~b0Td7i5i%cB?v>gLlZL$1>sZB-4X>;>erlH&{#_YmNd|*m#wPu
zFnV6Rs9VfhiY$l8h#Si_(`Ly5k*y@E?wGIkJF`{Ur-=IA=Z}>dh2d;fZXQ^RXA#dR
z7%ggrRJf5_=XJN(ROjY+wW&)q5`b1mOU06tsE$_5)kBFo>}|u4sIi(M89Ap6&f<-_
zU0zKM>BZEiXuj8xj!9H9QK&`hBqsCe*e6_gz|yhXFFar!HAkdEpS}SrNIAS-!VKZ&
z*`H9UCXe6>RCm<uClMY5#1f_DPk0oE2P{`}MEcr;KTDXRaobp}O*Xz*M@TIpt8sKr
zG~(SF?%5P=Ti|;pt_dp3Ww4i-2x+HA_Duv^Nou0Xvx2ndOo;UAtrhx$(@N0{Eg7k4
z0V4xgE97d0l0L3z<}aC0@a6G99?)dX4Gy{x1$l6o011$iCC8iuP8)2dc$!JknofjF
zO(y~-e31u|bQRQ?$*dQ~NwXd-5B~+QUXc^FPDAy_Ms!kHN_Rl44+#WLqJq(3E9rpf
zPdOPD&Jt3V5MsIQYp_*xveqiLjC!ZdRaiM&kv_%+;q6pgh!smqSF;6MkiB9oG&T0m
z*h;gOw&{4vUUD~B`WvK;J`7?LKM0{5ekm?+sHPzXhm|8`2|;SrosvBJCMViK?CM@w
zv~%M@xO&zS5LKvp1J-bKrWWP7t!#^$Fgz+Fqv65a4qFL{yds&lNq>wD1Th?Q$|E9x
zz}vB5*DV+1QQqK{Wlhqb12V$6X_*Y88YH7ZKmbKE*P~%syoEZRqC3YHu}kyOQobHu
z{f<dQu}rPxT3@C~gfv?#4Gm57$<NoA@Ix-#RQsZnAc>IjnmNqYAwe(@Vum33(S(|s
zFNxMo*frxM0n(~P3Ys|{u$%(+WX=*2;U*83NEw+!H^Rt#G673O<xfe}KUvP=$B}Kx
z1P}^LVKG}&Pd!;yZMya-em>cjB9F|BQsNG|r4;Uv2gLfyP)uTmDK#vV-669MbHz0t
zP*RhN&K+_xP^gzwN1V1ve}`J=HbW6+NTycz<XKCzs*wn3s@4qeiBOF@_N?o34WhhN
znIS!4cV>UDtwrn&xqZwj*Q%yMp*$2hv4<(SRy7LVzEayF@_>^)n_k@^>e^sBwza?m
zmX7Tk@PLuJQ*`Pn@muoLr!ie7FK*<$u}s}9xFU)B3eAR?Q!ztR@~INOZy5{EuyDVH
z2P`}@E3juBX-Y=C%RqZqBqpT{Q~9QGhklq%3q#~nexym*DHcl&Qg?^kajw54*f)b}
zAl|K3o{%>db?FFcQX|#*D=4BYr>u6A1Kwp|Jj#}oGR#C-Vp8|=o=8J4{5DFr!3}4`
z4Q8{MxJES<u9`tHryBFR85|-3sN9(q6G`70G^1gOlJduVh(kePJ{p>jO7qdkTtMy@
zO1|&aedZTDPi`m4{g1I?U6hS_hhL2Dd@w^geOG=+)T+xJwn8;Y`|eOher-vNz?{(@
zF%xUMK9+3}yDT1@FN;aqfprKh;u!^_tK=c=Rj73|oNHCJ1C<*JMRSvo>d>vvhebT4
zpxLR`7|RFM3mVKHcEk(%N%1r_k|@~W50~7sd4(=R?E4%>ipEt_ZRP*jRIPo^R%HCs
zwlJQuSYS0=d`MYB5J$bsXNc{ed^xumL?B9^4qM@dj8hFnQUrTeL_y)RS2OEHSYnZA
z>d1bVFF%u8be&>fkVWa3G(|H{&Q|-}uxsdSleNWg;WH=;bF|BTu;ws{6KOWBT{hKw
z5`#%BPy9d26P;~Zk5VeGT+Isa{%2dnUW-Rp>vOHDex0gk?)A&hsYXd<%RPZ2W~L*-
zI$5?woEmU*vp(>yt6kTU=xWOK;IdviCS`p6VPvyz+8a^5v0}ZCr6xi_r)BubSW}Z@
zT2DlyB+9mkvq#)lVMN~bPHHqqn;LmOWpXne=0sTik`bp8IyzSBXitA+L-TY9!y=@Z
zTE>lDJJNVS1*ilTD)xcpkRpLZP*taI%q``;DgKBU>jsO~VV~I9CQpwm_IGS2GwF(p
zx-i(14#Bnh;SO8r35h*cYZ22SR|BAc^(xSI*t%7n78C5l){_`?#<VE9M$t~xH^0LG
zGjVN-MeXk|x5-&c(FG&q#2=ZcxkO~7m?7)jwU=3Jg=<nCGWP5bKG+ITgsEd*b-&X*
z`@>Cf9w<q_B4RKm*MPWRb!zhS6Ifo}CGw|V{SI{dav@87(cGK|D2u8P*E%gO&jpkl
z>6oHf;5reqUWKYnb-*Ds?`%`MV~P$T+oGiu*JjM%gjzHj69Xn;+N>8`k+(yl`xc|i
zkzgufNJ>22qE~U{ISvbpM5?*vvg*Z!T~F;h*)y-;t08jh$9&&IZ>)PFz2#XPqG%%g
z0;fjm%b^kxwW2J3U8j;(J6|QUc)<gfQotMq(m1V}({KtKu$%(+4AT-4VLOc_%E}+B
znNCJLBcUo6L5vNh^3=(wL-h^kgx}2T86KBNzGlX6%KOFTX5P3AXs1H4YnHZ>Rm3_!
z5OHnf*BsNiCnKJgF#O76E@b5)^YW3v1Dpv!e-uMSd2m-f8C?z&ajb^7!k8h#TeX>|
zvqNDa$t4t(7Tx|tMHHR7z2?~&YC$C=4JR)qAL`e1R0hWvT5_!_)dFvrLm^0jg!nV-
z&k|)D&UneXIh2ezZ@6Vw?j&a<h}L_iSk!6}JKJQ-qMU#>t8EsI+!HC*mqS{_sSy`b
zrO3OkcD*~2qK}NkwOmBaM@1|l5#nUSt;$8*zCHI$iyAH>;^@WBh^fvkWa`{*$W-^7
z4Cy!zU`p+s>TlKN6xOE&^mlM<L4RLZDbg&XP`I0rRMfIFCe*A%qH&SmU-L_|++@gQ
zo@$trJ4_YbZt0^pF+q5kwFS{O#Vp9?K4w9b?Ucnk<VLV#zRJK1>DIf$>KjZ!*-l29
z2%ss?#N|D65-w#4J+dj2-%WEHVmxWiFK1iCl;OIRYAouSD2`6U4GLeYOtm5&BMomP
z+3<$s@}b;x4xo55h5DtM3~`ld=yUVzCk@Xl`$Dc)#cXg|Lw{(a?&{%U^!<fN403@M
zSV~#K>>8^Ex2`^05?bq>)F`2j6_&;8btojti~KS|Qjp-At_r5quh~r0SW5$zG$4F)
z%>si&LKVH7myTV!i<sIM39sEX>RxSlBJfS(jC`oAIQjLNig4JJb5ar6yj#o}W+y`^
z5QSMolr^VPVvSK{12#r!Lv2SU<U<z5ab{3Lp8@yEMW)YX&oIw^^BgeGndZ5yc`lby
zPkFMdROr%Yj>S8Pa-#nWh>Vg{>6AMNSW?=|H>-_irdiSxd0$?cE+#@M4b-w+WT0uf
zS;@;J)!!oG0prc8ZUcw_URC}!CZ>vCc@#nJWEKkY7P#5#Y1LEBYB<@a8uMnYpq50D
z091cNCE4`zeR|uRD43DIYHdjOc_HNzsg_g%wQnBTP&BPkwPfDIFeQiU-aKF`WYO1&
z+(7~+M&+j^8}f8Ui29SGdd0WqRuiKocTDYEcPONW5N?r=mgG5zhD%0E71v9SmZ}rx
zqQ8RTx8C%r`t47QdZmIZ&qL%!<rxm^C|t@C0>khKLCO-c1_w)f$%tnRG%H()kiyY=
zYRoAIy*JgE=|?|N!zKAEhL&S)zF@w2O*Iw|IoRf7;>UMtH9A%rEcMOMkAAKmB5*80
zVyrYGKPOXiuv}+JnR7d`!^#BM^+h{Z3ytJ%o59x6Xrrj!%;4ZqQ9xza7802m+>mjq
z0n-mZA#Zx9&lAiwCYzz*KBp*8Wy!ILJ^q~b|4cjE45(Jp<FS?TfMv#GgU3Wj@wgb>
z2~&HaJ`qyko4ueOFffkC^WHd~aLYA5A==sr(Xuglu&J4M*(}eih_0Her_g4b?SHsI
F?~0aZ)an2L

literal 0
HcmV?d00001

diff --git a/docs/2_39/site_libs/bootstrap/bootstrap.min.js b/docs/2_39/site_libs/bootstrap/bootstrap.min.js
new file mode 100644
index 000000000..e8f21f703
--- /dev/null
+++ b/docs/2_39/site_libs/bootstrap/bootstrap.min.js
@@ -0,0 +1,7 @@
+/*!
+  * Bootstrap v5.3.1 (https://getbootstrap.com/)
+  * Copyright 2011-2023 The Bootstrap Authors (https://github.com/twbs/bootstrap/graphs/contributors)
+  * Licensed under MIT (https://github.com/twbs/bootstrap/blob/main/LICENSE)
+  */
+!function(t,e){"object"==typeof exports&&"undefined"!=typeof module?module.exports=e():"function"==typeof define&&define.amd?define(e):(t="undefined"!=typeof globalThis?globalThis:t||self).bootstrap=e()}(this,(function(){"use strict";const t=new Map,e={set(e,i,n){t.has(e)||t.set(e,new Map);const s=t.get(e);s.has(i)||0===s.size?s.set(i,n):console.error(`Bootstrap doesn't allow more than one instance per element. Bound instance: ${Array.from(s.keys())[0]}.`)},get:(e,i)=>t.has(e)&&t.get(e).get(i)||null,remove(e,i){if(!t.has(e))return;const n=t.get(e);n.delete(i),0===n.size&&t.delete(e)}},i="transitionend",n=t=>(t&&window.CSS&&window.CSS.escape&&(t=t.replace(/#([^\s"#']+)/g,((t,e)=>`#${CSS.escape(e)}`))),t),s=t=>{t.dispatchEvent(new Event(i))},o=t=>!(!t||"object"!=typeof t)&&(void 0!==t.jquery&&(t=t[0]),void 0!==t.nodeType),r=t=>o(t)?t.jquery?t[0]:t:"string"==typeof t&&t.length>0?document.querySelector(n(t)):null,a=t=>{if(!o(t)||0===t.getClientRects().length)return!1;const e="visible"===getComputedStyle(t).getPropertyValue("visibility"),i=t.closest("details:not([open])");if(!i)return e;if(i!==t){const e=t.closest("summary");if(e&&e.parentNode!==i)return!1;if(null===e)return!1}return e},l=t=>!t||t.nodeType!==Node.ELEMENT_NODE||!!t.classList.contains("disabled")||(void 0!==t.disabled?t.disabled:t.hasAttribute("disabled")&&"false"!==t.getAttribute("disabled")),c=t=>{if(!document.documentElement.attachShadow)return null;if("function"==typeof t.getRootNode){const e=t.getRootNode();return e instanceof ShadowRoot?e:null}return t instanceof ShadowRoot?t:t.parentNode?c(t.parentNode):null},h=()=>{},d=t=>{t.offsetHeight},u=()=>window.jQuery&&!document.body.hasAttribute("data-bs-no-jquery")?window.jQuery:null,f=[],p=()=>"rtl"===document.documentElement.dir,m=t=>{var e;e=()=>{const e=u();if(e){const i=t.NAME,n=e.fn[i];e.fn[i]=t.jQueryInterface,e.fn[i].Constructor=t,e.fn[i].noConflict=()=>(e.fn[i]=n,t.jQueryInterface)}},"loading"===document.readyState?(f.length||document.addEventListener("DOMContentLoaded",(()=>{for(const t of f)t()})),f.push(e)):e()},g=(t,e=[],i=t)=>"function"==typeof t?t(...e):i,_=(t,e,n=!0)=>{if(!n)return void g(t);const o=(t=>{if(!t)return 0;let{transitionDuration:e,transitionDelay:i}=window.getComputedStyle(t);const n=Number.parseFloat(e),s=Number.parseFloat(i);return n||s?(e=e.split(",")[0],i=i.split(",")[0],1e3*(Number.parseFloat(e)+Number.parseFloat(i))):0})(e)+5;let r=!1;const a=({target:n})=>{n===e&&(r=!0,e.removeEventListener(i,a),g(t))};e.addEventListener(i,a),setTimeout((()=>{r||s(e)}),o)},b=(t,e,i,n)=>{const s=t.length;let o=t.indexOf(e);return-1===o?!i&&n?t[s-1]:t[0]:(o+=i?1:-1,n&&(o=(o+s)%s),t[Math.max(0,Math.min(o,s-1))])},v=/[^.]*(?=\..*)\.|.*/,y=/\..*/,w=/::\d+$/,A={};let E=1;const T={mouseenter:"mouseover",mouseleave:"mouseout"},C=new Set(["click","dblclick","mouseup","mousedown","contextmenu","mousewheel","DOMMouseScroll","mouseover","mouseout","mousemove","selectstart","selectend","keydown","keypress","keyup","orientationchange","touchstart","touchmove","touchend","touchcancel","pointerdown","pointermove","pointerup","pointerleave","pointercancel","gesturestart","gesturechange","gestureend","focus","blur","change","reset","select","submit","focusin","focusout","load","unload","beforeunload","resize","move","DOMContentLoaded","readystatechange","error","abort","scroll"]);function O(t,e){return e&&`${e}::${E++}`||t.uidEvent||E++}function x(t){const e=O(t);return t.uidEvent=e,A[e]=A[e]||{},A[e]}function k(t,e,i=null){return Object.values(t).find((t=>t.callable===e&&t.delegationSelector===i))}function L(t,e,i){const n="string"==typeof e,s=n?i:e||i;let o=I(t);return C.has(o)||(o=t),[n,s,o]}function S(t,e,i,n,s){if("string"!=typeof e||!t)return;let[o,r,a]=L(e,i,n);if(e in T){const t=t=>function(e){if(!e.relatedTarget||e.relatedTarget!==e.delegateTarget&&!e.delegateTarget.contains(e.relatedTarget))return t.call(this,e)};r=t(r)}const l=x(t),c=l[a]||(l[a]={}),h=k(c,r,o?i:null);if(h)return void(h.oneOff=h.oneOff&&s);const d=O(r,e.replace(v,"")),u=o?function(t,e,i){return function n(s){const o=t.querySelectorAll(e);for(let{target:r}=s;r&&r!==this;r=r.parentNode)for(const a of o)if(a===r)return P(s,{delegateTarget:r}),n.oneOff&&N.off(t,s.type,e,i),i.apply(r,[s])}}(t,i,r):function(t,e){return function i(n){return P(n,{delegateTarget:t}),i.oneOff&&N.off(t,n.type,e),e.apply(t,[n])}}(t,r);u.delegationSelector=o?i:null,u.callable=r,u.oneOff=s,u.uidEvent=d,c[d]=u,t.addEventListener(a,u,o)}function D(t,e,i,n,s){const o=k(e[i],n,s);o&&(t.removeEventListener(i,o,Boolean(s)),delete e[i][o.uidEvent])}function $(t,e,i,n){const s=e[i]||{};for(const[o,r]of Object.entries(s))o.includes(n)&&D(t,e,i,r.callable,r.delegationSelector)}function I(t){return t=t.replace(y,""),T[t]||t}const N={on(t,e,i,n){S(t,e,i,n,!1)},one(t,e,i,n){S(t,e,i,n,!0)},off(t,e,i,n){if("string"!=typeof e||!t)return;const[s,o,r]=L(e,i,n),a=r!==e,l=x(t),c=l[r]||{},h=e.startsWith(".");if(void 0===o){if(h)for(const i of Object.keys(l))$(t,l,i,e.slice(1));for(const[i,n]of Object.entries(c)){const s=i.replace(w,"");a&&!e.includes(s)||D(t,l,r,n.callable,n.delegationSelector)}}else{if(!Object.keys(c).length)return;D(t,l,r,o,s?i:null)}},trigger(t,e,i){if("string"!=typeof e||!t)return null;const n=u();let s=null,o=!0,r=!0,a=!1;e!==I(e)&&n&&(s=n.Event(e,i),n(t).trigger(s),o=!s.isPropagationStopped(),r=!s.isImmediatePropagationStopped(),a=s.isDefaultPrevented());const l=P(new Event(e,{bubbles:o,cancelable:!0}),i);return a&&l.preventDefault(),r&&t.dispatchEvent(l),l.defaultPrevented&&s&&s.preventDefault(),l}};function P(t,e={}){for(const[i,n]of Object.entries(e))try{t[i]=n}catch(e){Object.defineProperty(t,i,{configurable:!0,get:()=>n})}return t}function M(t){if("true"===t)return!0;if("false"===t)return!1;if(t===Number(t).toString())return Number(t);if(""===t||"null"===t)return null;if("string"!=typeof t)return t;try{return JSON.parse(decodeURIComponent(t))}catch(e){return t}}function j(t){return t.replace(/[A-Z]/g,(t=>`-${t.toLowerCase()}`))}const F={setDataAttribute(t,e,i){t.setAttribute(`data-bs-${j(e)}`,i)},removeDataAttribute(t,e){t.removeAttribute(`data-bs-${j(e)}`)},getDataAttributes(t){if(!t)return{};const e={},i=Object.keys(t.dataset).filter((t=>t.startsWith("bs")&&!t.startsWith("bsConfig")));for(const n of i){let i=n.replace(/^bs/,"");i=i.charAt(0).toLowerCase()+i.slice(1,i.length),e[i]=M(t.dataset[n])}return e},getDataAttribute:(t,e)=>M(t.getAttribute(`data-bs-${j(e)}`))};class H{static get Default(){return{}}static get DefaultType(){return{}}static get NAME(){throw new Error('You have to implement the static method "NAME", for each component!')}_getConfig(t){return t=this._mergeConfigObj(t),t=this._configAfterMerge(t),this._typeCheckConfig(t),t}_configAfterMerge(t){return t}_mergeConfigObj(t,e){const i=o(e)?F.getDataAttribute(e,"config"):{};return{...this.constructor.Default,..."object"==typeof i?i:{},...o(e)?F.getDataAttributes(e):{},..."object"==typeof t?t:{}}}_typeCheckConfig(t,e=this.constructor.DefaultType){for(const[n,s]of Object.entries(e)){const e=t[n],r=o(e)?"element":null==(i=e)?`${i}`:Object.prototype.toString.call(i).match(/\s([a-z]+)/i)[1].toLowerCase();if(!new RegExp(s).test(r))throw new TypeError(`${this.constructor.NAME.toUpperCase()}: Option "${n}" provided type "${r}" but expected type "${s}".`)}var i}}class W extends H{constructor(t,i){super(),(t=r(t))&&(this._element=t,this._config=this._getConfig(i),e.set(this._element,this.constructor.DATA_KEY,this))}dispose(){e.remove(this._element,this.constructor.DATA_KEY),N.off(this._element,this.constructor.EVENT_KEY);for(const t of Object.getOwnPropertyNames(this))this[t]=null}_queueCallback(t,e,i=!0){_(t,e,i)}_getConfig(t){return t=this._mergeConfigObj(t,this._element),t=this._configAfterMerge(t),this._typeCheckConfig(t),t}static getInstance(t){return e.get(r(t),this.DATA_KEY)}static getOrCreateInstance(t,e={}){return this.getInstance(t)||new this(t,"object"==typeof e?e:null)}static get VERSION(){return"5.3.1"}static get DATA_KEY(){return`bs.${this.NAME}`}static get EVENT_KEY(){return`.${this.DATA_KEY}`}static eventName(t){return`${t}${this.EVENT_KEY}`}}const B=t=>{let e=t.getAttribute("data-bs-target");if(!e||"#"===e){let i=t.getAttribute("href");if(!i||!i.includes("#")&&!i.startsWith("."))return null;i.includes("#")&&!i.startsWith("#")&&(i=`#${i.split("#")[1]}`),e=i&&"#"!==i?i.trim():null}return n(e)},z={find:(t,e=document.documentElement)=>[].concat(...Element.prototype.querySelectorAll.call(e,t)),findOne:(t,e=document.documentElement)=>Element.prototype.querySelector.call(e,t),children:(t,e)=>[].concat(...t.children).filter((t=>t.matches(e))),parents(t,e){const i=[];let n=t.parentNode.closest(e);for(;n;)i.push(n),n=n.parentNode.closest(e);return i},prev(t,e){let i=t.previousElementSibling;for(;i;){if(i.matches(e))return[i];i=i.previousElementSibling}return[]},next(t,e){let i=t.nextElementSibling;for(;i;){if(i.matches(e))return[i];i=i.nextElementSibling}return[]},focusableChildren(t){const e=["a","button","input","textarea","select","details","[tabindex]",'[contenteditable="true"]'].map((t=>`${t}:not([tabindex^="-"])`)).join(",");return this.find(e,t).filter((t=>!l(t)&&a(t)))},getSelectorFromElement(t){const e=B(t);return e&&z.findOne(e)?e:null},getElementFromSelector(t){const e=B(t);return e?z.findOne(e):null},getMultipleElementsFromSelector(t){const e=B(t);return e?z.find(e):[]}},R=(t,e="hide")=>{const i=`click.dismiss${t.EVENT_KEY}`,n=t.NAME;N.on(document,i,`[data-bs-dismiss="${n}"]`,(function(i){if(["A","AREA"].includes(this.tagName)&&i.preventDefault(),l(this))return;const s=z.getElementFromSelector(this)||this.closest(`.${n}`);t.getOrCreateInstance(s)[e]()}))},q=".bs.alert",V=`close${q}`,K=`closed${q}`;class Q extends W{static get NAME(){return"alert"}close(){if(N.trigger(this._element,V).defaultPrevented)return;this._element.classList.remove("show");const t=this._element.classList.contains("fade");this._queueCallback((()=>this._destroyElement()),this._element,t)}_destroyElement(){this._element.remove(),N.trigger(this._element,K),this.dispose()}static jQueryInterface(t){return this.each((function(){const e=Q.getOrCreateInstance(this);if("string"==typeof t){if(void 0===e[t]||t.startsWith("_")||"constructor"===t)throw new TypeError(`No method named "${t}"`);e[t](this)}}))}}R(Q,"close"),m(Q);const X='[data-bs-toggle="button"]';class Y extends W{static get NAME(){return"button"}toggle(){this._element.setAttribute("aria-pressed",this._element.classList.toggle("active"))}static jQueryInterface(t){return this.each((function(){const e=Y.getOrCreateInstance(this);"toggle"===t&&e[t]()}))}}N.on(document,"click.bs.button.data-api",X,(t=>{t.preventDefault();const e=t.target.closest(X);Y.getOrCreateInstance(e).toggle()})),m(Y);const U=".bs.swipe",G=`touchstart${U}`,J=`touchmove${U}`,Z=`touchend${U}`,tt=`pointerdown${U}`,et=`pointerup${U}`,it={endCallback:null,leftCallback:null,rightCallback:null},nt={endCallback:"(function|null)",leftCallback:"(function|null)",rightCallback:"(function|null)"};class st extends H{constructor(t,e){super(),this._element=t,t&&st.isSupported()&&(this._config=this._getConfig(e),this._deltaX=0,this._supportPointerEvents=Boolean(window.PointerEvent),this._initEvents())}static get Default(){return it}static get DefaultType(){return nt}static get NAME(){return"swipe"}dispose(){N.off(this._element,U)}_start(t){this._supportPointerEvents?this._eventIsPointerPenTouch(t)&&(this._deltaX=t.clientX):this._deltaX=t.touches[0].clientX}_end(t){this._eventIsPointerPenTouch(t)&&(this._deltaX=t.clientX-this._deltaX),this._handleSwipe(),g(this._config.endCallback)}_move(t){this._deltaX=t.touches&&t.touches.length>1?0:t.touches[0].clientX-this._deltaX}_handleSwipe(){const t=Math.abs(this._deltaX);if(t<=40)return;const e=t/this._deltaX;this._deltaX=0,e&&g(e>0?this._config.rightCallback:this._config.leftCallback)}_initEvents(){this._supportPointerEvents?(N.on(this._element,tt,(t=>this._start(t))),N.on(this._element,et,(t=>this._end(t))),this._element.classList.add("pointer-event")):(N.on(this._element,G,(t=>this._start(t))),N.on(this._element,J,(t=>this._move(t))),N.on(this._element,Z,(t=>this._end(t))))}_eventIsPointerPenTouch(t){return this._supportPointerEvents&&("pen"===t.pointerType||"touch"===t.pointerType)}static isSupported(){return"ontouchstart"in document.documentElement||navigator.maxTouchPoints>0}}const ot=".bs.carousel",rt=".data-api",at="next",lt="prev",ct="left",ht="right",dt=`slide${ot}`,ut=`slid${ot}`,ft=`keydown${ot}`,pt=`mouseenter${ot}`,mt=`mouseleave${ot}`,gt=`dragstart${ot}`,_t=`load${ot}${rt}`,bt=`click${ot}${rt}`,vt="carousel",yt="active",wt=".active",At=".carousel-item",Et=wt+At,Tt={ArrowLeft:ht,ArrowRight:ct},Ct={interval:5e3,keyboard:!0,pause:"hover",ride:!1,touch:!0,wrap:!0},Ot={interval:"(number|boolean)",keyboard:"boolean",pause:"(string|boolean)",ride:"(boolean|string)",touch:"boolean",wrap:"boolean"};class xt extends W{constructor(t,e){super(t,e),this._interval=null,this._activeElement=null,this._isSliding=!1,this.touchTimeout=null,this._swipeHelper=null,this._indicatorsElement=z.findOne(".carousel-indicators",this._element),this._addEventListeners(),this._config.ride===vt&&this.cycle()}static get Default(){return Ct}static get DefaultType(){return Ot}static get NAME(){return"carousel"}next(){this._slide(at)}nextWhenVisible(){!document.hidden&&a(this._element)&&this.next()}prev(){this._slide(lt)}pause(){this._isSliding&&s(this._element),this._clearInterval()}cycle(){this._clearInterval(),this._updateInterval(),this._interval=setInterval((()=>this.nextWhenVisible()),this._config.interval)}_maybeEnableCycle(){this._config.ride&&(this._isSliding?N.one(this._element,ut,(()=>this.cycle())):this.cycle())}to(t){const e=this._getItems();if(t>e.length-1||t<0)return;if(this._isSliding)return void N.one(this._element,ut,(()=>this.to(t)));const i=this._getItemIndex(this._getActive());if(i===t)return;const n=t>i?at:lt;this._slide(n,e[t])}dispose(){this._swipeHelper&&this._swipeHelper.dispose(),super.dispose()}_configAfterMerge(t){return t.defaultInterval=t.interval,t}_addEventListeners(){this._config.keyboard&&N.on(this._element,ft,(t=>this._keydown(t))),"hover"===this._config.pause&&(N.on(this._element,pt,(()=>this.pause())),N.on(this._element,mt,(()=>this._maybeEnableCycle()))),this._config.touch&&st.isSupported()&&this._addTouchEventListeners()}_addTouchEventListeners(){for(const t of z.find(".carousel-item img",this._element))N.on(t,gt,(t=>t.preventDefault()));const t={leftCallback:()=>this._slide(this._directionToOrder(ct)),rightCallback:()=>this._slide(this._directionToOrder(ht)),endCallback:()=>{"hover"===this._config.pause&&(this.pause(),this.touchTimeout&&clearTimeout(this.touchTimeout),this.touchTimeout=setTimeout((()=>this._maybeEnableCycle()),500+this._config.interval))}};this._swipeHelper=new st(this._element,t)}_keydown(t){if(/input|textarea/i.test(t.target.tagName))return;const e=Tt[t.key];e&&(t.preventDefault(),this._slide(this._directionToOrder(e)))}_getItemIndex(t){return this._getItems().indexOf(t)}_setActiveIndicatorElement(t){if(!this._indicatorsElement)return;const e=z.findOne(wt,this._indicatorsElement);e.classList.remove(yt),e.removeAttribute("aria-current");const i=z.findOne(`[data-bs-slide-to="${t}"]`,this._indicatorsElement);i&&(i.classList.add(yt),i.setAttribute("aria-current","true"))}_updateInterval(){const t=this._activeElement||this._getActive();if(!t)return;const e=Number.parseInt(t.getAttribute("data-bs-interval"),10);this._config.interval=e||this._config.defaultInterval}_slide(t,e=null){if(this._isSliding)return;const i=this._getActive(),n=t===at,s=e||b(this._getItems(),i,n,this._config.wrap);if(s===i)return;const o=this._getItemIndex(s),r=e=>N.trigger(this._element,e,{relatedTarget:s,direction:this._orderToDirection(t),from:this._getItemIndex(i),to:o});if(r(dt).defaultPrevented)return;if(!i||!s)return;const a=Boolean(this._interval);this.pause(),this._isSliding=!0,this._setActiveIndicatorElement(o),this._activeElement=s;const l=n?"carousel-item-start":"carousel-item-end",c=n?"carousel-item-next":"carousel-item-prev";s.classList.add(c),d(s),i.classList.add(l),s.classList.add(l),this._queueCallback((()=>{s.classList.remove(l,c),s.classList.add(yt),i.classList.remove(yt,c,l),this._isSliding=!1,r(ut)}),i,this._isAnimated()),a&&this.cycle()}_isAnimated(){return this._element.classList.contains("slide")}_getActive(){return z.findOne(Et,this._element)}_getItems(){return z.find(At,this._element)}_clearInterval(){this._interval&&(clearInterval(this._interval),this._interval=null)}_directionToOrder(t){return p()?t===ct?lt:at:t===ct?at:lt}_orderToDirection(t){return p()?t===lt?ct:ht:t===lt?ht:ct}static jQueryInterface(t){return this.each((function(){const e=xt.getOrCreateInstance(this,t);if("number"!=typeof t){if("string"==typeof t){if(void 0===e[t]||t.startsWith("_")||"constructor"===t)throw new TypeError(`No method named "${t}"`);e[t]()}}else e.to(t)}))}}N.on(document,bt,"[data-bs-slide], [data-bs-slide-to]",(function(t){const e=z.getElementFromSelector(this);if(!e||!e.classList.contains(vt))return;t.preventDefault();const i=xt.getOrCreateInstance(e),n=this.getAttribute("data-bs-slide-to");return n?(i.to(n),void i._maybeEnableCycle()):"next"===F.getDataAttribute(this,"slide")?(i.next(),void i._maybeEnableCycle()):(i.prev(),void i._maybeEnableCycle())})),N.on(window,_t,(()=>{const t=z.find('[data-bs-ride="carousel"]');for(const e of t)xt.getOrCreateInstance(e)})),m(xt);const kt=".bs.collapse",Lt=`show${kt}`,St=`shown${kt}`,Dt=`hide${kt}`,$t=`hidden${kt}`,It=`click${kt}.data-api`,Nt="show",Pt="collapse",Mt="collapsing",jt=`:scope .${Pt} .${Pt}`,Ft='[data-bs-toggle="collapse"]',Ht={parent:null,toggle:!0},Wt={parent:"(null|element)",toggle:"boolean"};class Bt extends W{constructor(t,e){super(t,e),this._isTransitioning=!1,this._triggerArray=[];const i=z.find(Ft);for(const t of i){const e=z.getSelectorFromElement(t),i=z.find(e).filter((t=>t===this._element));null!==e&&i.length&&this._triggerArray.push(t)}this._initializeChildren(),this._config.parent||this._addAriaAndCollapsedClass(this._triggerArray,this._isShown()),this._config.toggle&&this.toggle()}static get Default(){return Ht}static get DefaultType(){return Wt}static get NAME(){return"collapse"}toggle(){this._isShown()?this.hide():this.show()}show(){if(this._isTransitioning||this._isShown())return;let t=[];if(this._config.parent&&(t=this._getFirstLevelChildren(".collapse.show, .collapse.collapsing").filter((t=>t!==this._element)).map((t=>Bt.getOrCreateInstance(t,{toggle:!1})))),t.length&&t[0]._isTransitioning)return;if(N.trigger(this._element,Lt).defaultPrevented)return;for(const e of t)e.hide();const e=this._getDimension();this._element.classList.remove(Pt),this._element.classList.add(Mt),this._element.style[e]=0,this._addAriaAndCollapsedClass(this._triggerArray,!0),this._isTransitioning=!0;const i=`scroll${e[0].toUpperCase()+e.slice(1)}`;this._queueCallback((()=>{this._isTransitioning=!1,this._element.classList.remove(Mt),this._element.classList.add(Pt,Nt),this._element.style[e]="",N.trigger(this._element,St)}),this._element,!0),this._element.style[e]=`${this._element[i]}px`}hide(){if(this._isTransitioning||!this._isShown())return;if(N.trigger(this._element,Dt).defaultPrevented)return;const t=this._getDimension();this._element.style[t]=`${this._element.getBoundingClientRect()[t]}px`,d(this._element),this._element.classList.add(Mt),this._element.classList.remove(Pt,Nt);for(const t of this._triggerArray){const e=z.getElementFromSelector(t);e&&!this._isShown(e)&&this._addAriaAndCollapsedClass([t],!1)}this._isTransitioning=!0,this._element.style[t]="",this._queueCallback((()=>{this._isTransitioning=!1,this._element.classList.remove(Mt),this._element.classList.add(Pt),N.trigger(this._element,$t)}),this._element,!0)}_isShown(t=this._element){return t.classList.contains(Nt)}_configAfterMerge(t){return t.toggle=Boolean(t.toggle),t.parent=r(t.parent),t}_getDimension(){return this._element.classList.contains("collapse-horizontal")?"width":"height"}_initializeChildren(){if(!this._config.parent)return;const t=this._getFirstLevelChildren(Ft);for(const e of t){const t=z.getElementFromSelector(e);t&&this._addAriaAndCollapsedClass([e],this._isShown(t))}}_getFirstLevelChildren(t){const e=z.find(jt,this._config.parent);return z.find(t,this._config.parent).filter((t=>!e.includes(t)))}_addAriaAndCollapsedClass(t,e){if(t.length)for(const i of t)i.classList.toggle("collapsed",!e),i.setAttribute("aria-expanded",e)}static jQueryInterface(t){const e={};return"string"==typeof t&&/show|hide/.test(t)&&(e.toggle=!1),this.each((function(){const i=Bt.getOrCreateInstance(this,e);if("string"==typeof t){if(void 0===i[t])throw new TypeError(`No method named "${t}"`);i[t]()}}))}}N.on(document,It,Ft,(function(t){("A"===t.target.tagName||t.delegateTarget&&"A"===t.delegateTarget.tagName)&&t.preventDefault();for(const t of z.getMultipleElementsFromSelector(this))Bt.getOrCreateInstance(t,{toggle:!1}).toggle()})),m(Bt);var zt="top",Rt="bottom",qt="right",Vt="left",Kt="auto",Qt=[zt,Rt,qt,Vt],Xt="start",Yt="end",Ut="clippingParents",Gt="viewport",Jt="popper",Zt="reference",te=Qt.reduce((function(t,e){return t.concat([e+"-"+Xt,e+"-"+Yt])}),[]),ee=[].concat(Qt,[Kt]).reduce((function(t,e){return t.concat([e,e+"-"+Xt,e+"-"+Yt])}),[]),ie="beforeRead",ne="read",se="afterRead",oe="beforeMain",re="main",ae="afterMain",le="beforeWrite",ce="write",he="afterWrite",de=[ie,ne,se,oe,re,ae,le,ce,he];function ue(t){return t?(t.nodeName||"").toLowerCase():null}function fe(t){if(null==t)return window;if("[object Window]"!==t.toString()){var e=t.ownerDocument;return e&&e.defaultView||window}return t}function pe(t){return t instanceof fe(t).Element||t instanceof Element}function me(t){return t instanceof fe(t).HTMLElement||t instanceof HTMLElement}function ge(t){return"undefined"!=typeof ShadowRoot&&(t instanceof fe(t).ShadowRoot||t instanceof ShadowRoot)}const _e={name:"applyStyles",enabled:!0,phase:"write",fn:function(t){var e=t.state;Object.keys(e.elements).forEach((function(t){var i=e.styles[t]||{},n=e.attributes[t]||{},s=e.elements[t];me(s)&&ue(s)&&(Object.assign(s.style,i),Object.keys(n).forEach((function(t){var e=n[t];!1===e?s.removeAttribute(t):s.setAttribute(t,!0===e?"":e)})))}))},effect:function(t){var e=t.state,i={popper:{position:e.options.strategy,left:"0",top:"0",margin:"0"},arrow:{position:"absolute"},reference:{}};return Object.assign(e.elements.popper.style,i.popper),e.styles=i,e.elements.arrow&&Object.assign(e.elements.arrow.style,i.arrow),function(){Object.keys(e.elements).forEach((function(t){var n=e.elements[t],s=e.attributes[t]||{},o=Object.keys(e.styles.hasOwnProperty(t)?e.styles[t]:i[t]).reduce((function(t,e){return t[e]="",t}),{});me(n)&&ue(n)&&(Object.assign(n.style,o),Object.keys(s).forEach((function(t){n.removeAttribute(t)})))}))}},requires:["computeStyles"]};function be(t){return t.split("-")[0]}var ve=Math.max,ye=Math.min,we=Math.round;function Ae(){var t=navigator.userAgentData;return null!=t&&t.brands&&Array.isArray(t.brands)?t.brands.map((function(t){return t.brand+"/"+t.version})).join(" "):navigator.userAgent}function Ee(){return!/^((?!chrome|android).)*safari/i.test(Ae())}function Te(t,e,i){void 0===e&&(e=!1),void 0===i&&(i=!1);var n=t.getBoundingClientRect(),s=1,o=1;e&&me(t)&&(s=t.offsetWidth>0&&we(n.width)/t.offsetWidth||1,o=t.offsetHeight>0&&we(n.height)/t.offsetHeight||1);var r=(pe(t)?fe(t):window).visualViewport,a=!Ee()&&i,l=(n.left+(a&&r?r.offsetLeft:0))/s,c=(n.top+(a&&r?r.offsetTop:0))/o,h=n.width/s,d=n.height/o;return{width:h,height:d,top:c,right:l+h,bottom:c+d,left:l,x:l,y:c}}function Ce(t){var e=Te(t),i=t.offsetWidth,n=t.offsetHeight;return Math.abs(e.width-i)<=1&&(i=e.width),Math.abs(e.height-n)<=1&&(n=e.height),{x:t.offsetLeft,y:t.offsetTop,width:i,height:n}}function Oe(t,e){var i=e.getRootNode&&e.getRootNode();if(t.contains(e))return!0;if(i&&ge(i)){var n=e;do{if(n&&t.isSameNode(n))return!0;n=n.parentNode||n.host}while(n)}return!1}function xe(t){return fe(t).getComputedStyle(t)}function ke(t){return["table","td","th"].indexOf(ue(t))>=0}function Le(t){return((pe(t)?t.ownerDocument:t.document)||window.document).documentElement}function Se(t){return"html"===ue(t)?t:t.assignedSlot||t.parentNode||(ge(t)?t.host:null)||Le(t)}function De(t){return me(t)&&"fixed"!==xe(t).position?t.offsetParent:null}function $e(t){for(var e=fe(t),i=De(t);i&&ke(i)&&"static"===xe(i).position;)i=De(i);return i&&("html"===ue(i)||"body"===ue(i)&&"static"===xe(i).position)?e:i||function(t){var e=/firefox/i.test(Ae());if(/Trident/i.test(Ae())&&me(t)&&"fixed"===xe(t).position)return null;var i=Se(t);for(ge(i)&&(i=i.host);me(i)&&["html","body"].indexOf(ue(i))<0;){var n=xe(i);if("none"!==n.transform||"none"!==n.perspective||"paint"===n.contain||-1!==["transform","perspective"].indexOf(n.willChange)||e&&"filter"===n.willChange||e&&n.filter&&"none"!==n.filter)return i;i=i.parentNode}return null}(t)||e}function Ie(t){return["top","bottom"].indexOf(t)>=0?"x":"y"}function Ne(t,e,i){return ve(t,ye(e,i))}function Pe(t){return Object.assign({},{top:0,right:0,bottom:0,left:0},t)}function Me(t,e){return e.reduce((function(e,i){return e[i]=t,e}),{})}const je={name:"arrow",enabled:!0,phase:"main",fn:function(t){var e,i=t.state,n=t.name,s=t.options,o=i.elements.arrow,r=i.modifiersData.popperOffsets,a=be(i.placement),l=Ie(a),c=[Vt,qt].indexOf(a)>=0?"height":"width";if(o&&r){var h=function(t,e){return Pe("number"!=typeof(t="function"==typeof t?t(Object.assign({},e.rects,{placement:e.placement})):t)?t:Me(t,Qt))}(s.padding,i),d=Ce(o),u="y"===l?zt:Vt,f="y"===l?Rt:qt,p=i.rects.reference[c]+i.rects.reference[l]-r[l]-i.rects.popper[c],m=r[l]-i.rects.reference[l],g=$e(o),_=g?"y"===l?g.clientHeight||0:g.clientWidth||0:0,b=p/2-m/2,v=h[u],y=_-d[c]-h[f],w=_/2-d[c]/2+b,A=Ne(v,w,y),E=l;i.modifiersData[n]=((e={})[E]=A,e.centerOffset=A-w,e)}},effect:function(t){var e=t.state,i=t.options.element,n=void 0===i?"[data-popper-arrow]":i;null!=n&&("string"!=typeof n||(n=e.elements.popper.querySelector(n)))&&Oe(e.elements.popper,n)&&(e.elements.arrow=n)},requires:["popperOffsets"],requiresIfExists:["preventOverflow"]};function Fe(t){return t.split("-")[1]}var He={top:"auto",right:"auto",bottom:"auto",left:"auto"};function We(t){var e,i=t.popper,n=t.popperRect,s=t.placement,o=t.variation,r=t.offsets,a=t.position,l=t.gpuAcceleration,c=t.adaptive,h=t.roundOffsets,d=t.isFixed,u=r.x,f=void 0===u?0:u,p=r.y,m=void 0===p?0:p,g="function"==typeof h?h({x:f,y:m}):{x:f,y:m};f=g.x,m=g.y;var _=r.hasOwnProperty("x"),b=r.hasOwnProperty("y"),v=Vt,y=zt,w=window;if(c){var A=$e(i),E="clientHeight",T="clientWidth";A===fe(i)&&"static"!==xe(A=Le(i)).position&&"absolute"===a&&(E="scrollHeight",T="scrollWidth"),(s===zt||(s===Vt||s===qt)&&o===Yt)&&(y=Rt,m-=(d&&A===w&&w.visualViewport?w.visualViewport.height:A[E])-n.height,m*=l?1:-1),s!==Vt&&(s!==zt&&s!==Rt||o!==Yt)||(v=qt,f-=(d&&A===w&&w.visualViewport?w.visualViewport.width:A[T])-n.width,f*=l?1:-1)}var C,O=Object.assign({position:a},c&&He),x=!0===h?function(t,e){var i=t.x,n=t.y,s=e.devicePixelRatio||1;return{x:we(i*s)/s||0,y:we(n*s)/s||0}}({x:f,y:m},fe(i)):{x:f,y:m};return f=x.x,m=x.y,l?Object.assign({},O,((C={})[y]=b?"0":"",C[v]=_?"0":"",C.transform=(w.devicePixelRatio||1)<=1?"translate("+f+"px, "+m+"px)":"translate3d("+f+"px, "+m+"px, 0)",C)):Object.assign({},O,((e={})[y]=b?m+"px":"",e[v]=_?f+"px":"",e.transform="",e))}const Be={name:"computeStyles",enabled:!0,phase:"beforeWrite",fn:function(t){var e=t.state,i=t.options,n=i.gpuAcceleration,s=void 0===n||n,o=i.adaptive,r=void 0===o||o,a=i.roundOffsets,l=void 0===a||a,c={placement:be(e.placement),variation:Fe(e.placement),popper:e.elements.popper,popperRect:e.rects.popper,gpuAcceleration:s,isFixed:"fixed"===e.options.strategy};null!=e.modifiersData.popperOffsets&&(e.styles.popper=Object.assign({},e.styles.popper,We(Object.assign({},c,{offsets:e.modifiersData.popperOffsets,position:e.options.strategy,adaptive:r,roundOffsets:l})))),null!=e.modifiersData.arrow&&(e.styles.arrow=Object.assign({},e.styles.arrow,We(Object.assign({},c,{offsets:e.modifiersData.arrow,position:"absolute",adaptive:!1,roundOffsets:l})))),e.attributes.popper=Object.assign({},e.attributes.popper,{"data-popper-placement":e.placement})},data:{}};var ze={passive:!0};const Re={name:"eventListeners",enabled:!0,phase:"write",fn:function(){},effect:function(t){var e=t.state,i=t.instance,n=t.options,s=n.scroll,o=void 0===s||s,r=n.resize,a=void 0===r||r,l=fe(e.elements.popper),c=[].concat(e.scrollParents.reference,e.scrollParents.popper);return o&&c.forEach((function(t){t.addEventListener("scroll",i.update,ze)})),a&&l.addEventListener("resize",i.update,ze),function(){o&&c.forEach((function(t){t.removeEventListener("scroll",i.update,ze)})),a&&l.removeEventListener("resize",i.update,ze)}},data:{}};var qe={left:"right",right:"left",bottom:"top",top:"bottom"};function Ve(t){return t.replace(/left|right|bottom|top/g,(function(t){return qe[t]}))}var Ke={start:"end",end:"start"};function Qe(t){return t.replace(/start|end/g,(function(t){return Ke[t]}))}function Xe(t){var e=fe(t);return{scrollLeft:e.pageXOffset,scrollTop:e.pageYOffset}}function Ye(t){return Te(Le(t)).left+Xe(t).scrollLeft}function Ue(t){var e=xe(t),i=e.overflow,n=e.overflowX,s=e.overflowY;return/auto|scroll|overlay|hidden/.test(i+s+n)}function Ge(t){return["html","body","#document"].indexOf(ue(t))>=0?t.ownerDocument.body:me(t)&&Ue(t)?t:Ge(Se(t))}function Je(t,e){var i;void 0===e&&(e=[]);var n=Ge(t),s=n===(null==(i=t.ownerDocument)?void 0:i.body),o=fe(n),r=s?[o].concat(o.visualViewport||[],Ue(n)?n:[]):n,a=e.concat(r);return s?a:a.concat(Je(Se(r)))}function Ze(t){return Object.assign({},t,{left:t.x,top:t.y,right:t.x+t.width,bottom:t.y+t.height})}function ti(t,e,i){return e===Gt?Ze(function(t,e){var i=fe(t),n=Le(t),s=i.visualViewport,o=n.clientWidth,r=n.clientHeight,a=0,l=0;if(s){o=s.width,r=s.height;var c=Ee();(c||!c&&"fixed"===e)&&(a=s.offsetLeft,l=s.offsetTop)}return{width:o,height:r,x:a+Ye(t),y:l}}(t,i)):pe(e)?function(t,e){var i=Te(t,!1,"fixed"===e);return i.top=i.top+t.clientTop,i.left=i.left+t.clientLeft,i.bottom=i.top+t.clientHeight,i.right=i.left+t.clientWidth,i.width=t.clientWidth,i.height=t.clientHeight,i.x=i.left,i.y=i.top,i}(e,i):Ze(function(t){var e,i=Le(t),n=Xe(t),s=null==(e=t.ownerDocument)?void 0:e.body,o=ve(i.scrollWidth,i.clientWidth,s?s.scrollWidth:0,s?s.clientWidth:0),r=ve(i.scrollHeight,i.clientHeight,s?s.scrollHeight:0,s?s.clientHeight:0),a=-n.scrollLeft+Ye(t),l=-n.scrollTop;return"rtl"===xe(s||i).direction&&(a+=ve(i.clientWidth,s?s.clientWidth:0)-o),{width:o,height:r,x:a,y:l}}(Le(t)))}function ei(t){var e,i=t.reference,n=t.element,s=t.placement,o=s?be(s):null,r=s?Fe(s):null,a=i.x+i.width/2-n.width/2,l=i.y+i.height/2-n.height/2;switch(o){case zt:e={x:a,y:i.y-n.height};break;case Rt:e={x:a,y:i.y+i.height};break;case qt:e={x:i.x+i.width,y:l};break;case Vt:e={x:i.x-n.width,y:l};break;default:e={x:i.x,y:i.y}}var c=o?Ie(o):null;if(null!=c){var h="y"===c?"height":"width";switch(r){case Xt:e[c]=e[c]-(i[h]/2-n[h]/2);break;case Yt:e[c]=e[c]+(i[h]/2-n[h]/2)}}return e}function ii(t,e){void 0===e&&(e={});var i=e,n=i.placement,s=void 0===n?t.placement:n,o=i.strategy,r=void 0===o?t.strategy:o,a=i.boundary,l=void 0===a?Ut:a,c=i.rootBoundary,h=void 0===c?Gt:c,d=i.elementContext,u=void 0===d?Jt:d,f=i.altBoundary,p=void 0!==f&&f,m=i.padding,g=void 0===m?0:m,_=Pe("number"!=typeof g?g:Me(g,Qt)),b=u===Jt?Zt:Jt,v=t.rects.popper,y=t.elements[p?b:u],w=function(t,e,i,n){var s="clippingParents"===e?function(t){var e=Je(Se(t)),i=["absolute","fixed"].indexOf(xe(t).position)>=0&&me(t)?$e(t):t;return pe(i)?e.filter((function(t){return pe(t)&&Oe(t,i)&&"body"!==ue(t)})):[]}(t):[].concat(e),o=[].concat(s,[i]),r=o[0],a=o.reduce((function(e,i){var s=ti(t,i,n);return e.top=ve(s.top,e.top),e.right=ye(s.right,e.right),e.bottom=ye(s.bottom,e.bottom),e.left=ve(s.left,e.left),e}),ti(t,r,n));return a.width=a.right-a.left,a.height=a.bottom-a.top,a.x=a.left,a.y=a.top,a}(pe(y)?y:y.contextElement||Le(t.elements.popper),l,h,r),A=Te(t.elements.reference),E=ei({reference:A,element:v,strategy:"absolute",placement:s}),T=Ze(Object.assign({},v,E)),C=u===Jt?T:A,O={top:w.top-C.top+_.top,bottom:C.bottom-w.bottom+_.bottom,left:w.left-C.left+_.left,right:C.right-w.right+_.right},x=t.modifiersData.offset;if(u===Jt&&x){var k=x[s];Object.keys(O).forEach((function(t){var e=[qt,Rt].indexOf(t)>=0?1:-1,i=[zt,Rt].indexOf(t)>=0?"y":"x";O[t]+=k[i]*e}))}return O}function ni(t,e){void 0===e&&(e={});var i=e,n=i.placement,s=i.boundary,o=i.rootBoundary,r=i.padding,a=i.flipVariations,l=i.allowedAutoPlacements,c=void 0===l?ee:l,h=Fe(n),d=h?a?te:te.filter((function(t){return Fe(t)===h})):Qt,u=d.filter((function(t){return c.indexOf(t)>=0}));0===u.length&&(u=d);var f=u.reduce((function(e,i){return e[i]=ii(t,{placement:i,boundary:s,rootBoundary:o,padding:r})[be(i)],e}),{});return Object.keys(f).sort((function(t,e){return f[t]-f[e]}))}const si={name:"flip",enabled:!0,phase:"main",fn:function(t){var e=t.state,i=t.options,n=t.name;if(!e.modifiersData[n]._skip){for(var s=i.mainAxis,o=void 0===s||s,r=i.altAxis,a=void 0===r||r,l=i.fallbackPlacements,c=i.padding,h=i.boundary,d=i.rootBoundary,u=i.altBoundary,f=i.flipVariations,p=void 0===f||f,m=i.allowedAutoPlacements,g=e.options.placement,_=be(g),b=l||(_!==g&&p?function(t){if(be(t)===Kt)return[];var e=Ve(t);return[Qe(t),e,Qe(e)]}(g):[Ve(g)]),v=[g].concat(b).reduce((function(t,i){return t.concat(be(i)===Kt?ni(e,{placement:i,boundary:h,rootBoundary:d,padding:c,flipVariations:p,allowedAutoPlacements:m}):i)}),[]),y=e.rects.reference,w=e.rects.popper,A=new Map,E=!0,T=v[0],C=0;C<v.length;C++){var O=v[C],x=be(O),k=Fe(O)===Xt,L=[zt,Rt].indexOf(x)>=0,S=L?"width":"height",D=ii(e,{placement:O,boundary:h,rootBoundary:d,altBoundary:u,padding:c}),$=L?k?qt:Vt:k?Rt:zt;y[S]>w[S]&&($=Ve($));var I=Ve($),N=[];if(o&&N.push(D[x]<=0),a&&N.push(D[$]<=0,D[I]<=0),N.every((function(t){return t}))){T=O,E=!1;break}A.set(O,N)}if(E)for(var P=function(t){var e=v.find((function(e){var i=A.get(e);if(i)return i.slice(0,t).every((function(t){return t}))}));if(e)return T=e,"break"},M=p?3:1;M>0&&"break"!==P(M);M--);e.placement!==T&&(e.modifiersData[n]._skip=!0,e.placement=T,e.reset=!0)}},requiresIfExists:["offset"],data:{_skip:!1}};function oi(t,e,i){return void 0===i&&(i={x:0,y:0}),{top:t.top-e.height-i.y,right:t.right-e.width+i.x,bottom:t.bottom-e.height+i.y,left:t.left-e.width-i.x}}function ri(t){return[zt,qt,Rt,Vt].some((function(e){return t[e]>=0}))}const ai={name:"hide",enabled:!0,phase:"main",requiresIfExists:["preventOverflow"],fn:function(t){var e=t.state,i=t.name,n=e.rects.reference,s=e.rects.popper,o=e.modifiersData.preventOverflow,r=ii(e,{elementContext:"reference"}),a=ii(e,{altBoundary:!0}),l=oi(r,n),c=oi(a,s,o),h=ri(l),d=ri(c);e.modifiersData[i]={referenceClippingOffsets:l,popperEscapeOffsets:c,isReferenceHidden:h,hasPopperEscaped:d},e.attributes.popper=Object.assign({},e.attributes.popper,{"data-popper-reference-hidden":h,"data-popper-escaped":d})}},li={name:"offset",enabled:!0,phase:"main",requires:["popperOffsets"],fn:function(t){var e=t.state,i=t.options,n=t.name,s=i.offset,o=void 0===s?[0,0]:s,r=ee.reduce((function(t,i){return t[i]=function(t,e,i){var n=be(t),s=[Vt,zt].indexOf(n)>=0?-1:1,o="function"==typeof i?i(Object.assign({},e,{placement:t})):i,r=o[0],a=o[1];return r=r||0,a=(a||0)*s,[Vt,qt].indexOf(n)>=0?{x:a,y:r}:{x:r,y:a}}(i,e.rects,o),t}),{}),a=r[e.placement],l=a.x,c=a.y;null!=e.modifiersData.popperOffsets&&(e.modifiersData.popperOffsets.x+=l,e.modifiersData.popperOffsets.y+=c),e.modifiersData[n]=r}},ci={name:"popperOffsets",enabled:!0,phase:"read",fn:function(t){var e=t.state,i=t.name;e.modifiersData[i]=ei({reference:e.rects.reference,element:e.rects.popper,strategy:"absolute",placement:e.placement})},data:{}},hi={name:"preventOverflow",enabled:!0,phase:"main",fn:function(t){var e=t.state,i=t.options,n=t.name,s=i.mainAxis,o=void 0===s||s,r=i.altAxis,a=void 0!==r&&r,l=i.boundary,c=i.rootBoundary,h=i.altBoundary,d=i.padding,u=i.tether,f=void 0===u||u,p=i.tetherOffset,m=void 0===p?0:p,g=ii(e,{boundary:l,rootBoundary:c,padding:d,altBoundary:h}),_=be(e.placement),b=Fe(e.placement),v=!b,y=Ie(_),w="x"===y?"y":"x",A=e.modifiersData.popperOffsets,E=e.rects.reference,T=e.rects.popper,C="function"==typeof m?m(Object.assign({},e.rects,{placement:e.placement})):m,O="number"==typeof C?{mainAxis:C,altAxis:C}:Object.assign({mainAxis:0,altAxis:0},C),x=e.modifiersData.offset?e.modifiersData.offset[e.placement]:null,k={x:0,y:0};if(A){if(o){var L,S="y"===y?zt:Vt,D="y"===y?Rt:qt,$="y"===y?"height":"width",I=A[y],N=I+g[S],P=I-g[D],M=f?-T[$]/2:0,j=b===Xt?E[$]:T[$],F=b===Xt?-T[$]:-E[$],H=e.elements.arrow,W=f&&H?Ce(H):{width:0,height:0},B=e.modifiersData["arrow#persistent"]?e.modifiersData["arrow#persistent"].padding:{top:0,right:0,bottom:0,left:0},z=B[S],R=B[D],q=Ne(0,E[$],W[$]),V=v?E[$]/2-M-q-z-O.mainAxis:j-q-z-O.mainAxis,K=v?-E[$]/2+M+q+R+O.mainAxis:F+q+R+O.mainAxis,Q=e.elements.arrow&&$e(e.elements.arrow),X=Q?"y"===y?Q.clientTop||0:Q.clientLeft||0:0,Y=null!=(L=null==x?void 0:x[y])?L:0,U=I+K-Y,G=Ne(f?ye(N,I+V-Y-X):N,I,f?ve(P,U):P);A[y]=G,k[y]=G-I}if(a){var J,Z="x"===y?zt:Vt,tt="x"===y?Rt:qt,et=A[w],it="y"===w?"height":"width",nt=et+g[Z],st=et-g[tt],ot=-1!==[zt,Vt].indexOf(_),rt=null!=(J=null==x?void 0:x[w])?J:0,at=ot?nt:et-E[it]-T[it]-rt+O.altAxis,lt=ot?et+E[it]+T[it]-rt-O.altAxis:st,ct=f&&ot?function(t,e,i){var n=Ne(t,e,i);return n>i?i:n}(at,et,lt):Ne(f?at:nt,et,f?lt:st);A[w]=ct,k[w]=ct-et}e.modifiersData[n]=k}},requiresIfExists:["offset"]};function di(t,e,i){void 0===i&&(i=!1);var n,s,o=me(e),r=me(e)&&function(t){var e=t.getBoundingClientRect(),i=we(e.width)/t.offsetWidth||1,n=we(e.height)/t.offsetHeight||1;return 1!==i||1!==n}(e),a=Le(e),l=Te(t,r,i),c={scrollLeft:0,scrollTop:0},h={x:0,y:0};return(o||!o&&!i)&&(("body"!==ue(e)||Ue(a))&&(c=(n=e)!==fe(n)&&me(n)?{scrollLeft:(s=n).scrollLeft,scrollTop:s.scrollTop}:Xe(n)),me(e)?((h=Te(e,!0)).x+=e.clientLeft,h.y+=e.clientTop):a&&(h.x=Ye(a))),{x:l.left+c.scrollLeft-h.x,y:l.top+c.scrollTop-h.y,width:l.width,height:l.height}}function ui(t){var e=new Map,i=new Set,n=[];function s(t){i.add(t.name),[].concat(t.requires||[],t.requiresIfExists||[]).forEach((function(t){if(!i.has(t)){var n=e.get(t);n&&s(n)}})),n.push(t)}return t.forEach((function(t){e.set(t.name,t)})),t.forEach((function(t){i.has(t.name)||s(t)})),n}var fi={placement:"bottom",modifiers:[],strategy:"absolute"};function pi(){for(var t=arguments.length,e=new Array(t),i=0;i<t;i++)e[i]=arguments[i];return!e.some((function(t){return!(t&&"function"==typeof t.getBoundingClientRect)}))}function mi(t){void 0===t&&(t={});var e=t,i=e.defaultModifiers,n=void 0===i?[]:i,s=e.defaultOptions,o=void 0===s?fi:s;return function(t,e,i){void 0===i&&(i=o);var s,r,a={placement:"bottom",orderedModifiers:[],options:Object.assign({},fi,o),modifiersData:{},elements:{reference:t,popper:e},attributes:{},styles:{}},l=[],c=!1,h={state:a,setOptions:function(i){var s="function"==typeof i?i(a.options):i;d(),a.options=Object.assign({},o,a.options,s),a.scrollParents={reference:pe(t)?Je(t):t.contextElement?Je(t.contextElement):[],popper:Je(e)};var r,c,u=function(t){var e=ui(t);return de.reduce((function(t,i){return t.concat(e.filter((function(t){return t.phase===i})))}),[])}((r=[].concat(n,a.options.modifiers),c=r.reduce((function(t,e){var i=t[e.name];return t[e.name]=i?Object.assign({},i,e,{options:Object.assign({},i.options,e.options),data:Object.assign({},i.data,e.data)}):e,t}),{}),Object.keys(c).map((function(t){return c[t]}))));return a.orderedModifiers=u.filter((function(t){return t.enabled})),a.orderedModifiers.forEach((function(t){var e=t.name,i=t.options,n=void 0===i?{}:i,s=t.effect;if("function"==typeof s){var o=s({state:a,name:e,instance:h,options:n});l.push(o||function(){})}})),h.update()},forceUpdate:function(){if(!c){var t=a.elements,e=t.reference,i=t.popper;if(pi(e,i)){a.rects={reference:di(e,$e(i),"fixed"===a.options.strategy),popper:Ce(i)},a.reset=!1,a.placement=a.options.placement,a.orderedModifiers.forEach((function(t){return a.modifiersData[t.name]=Object.assign({},t.data)}));for(var n=0;n<a.orderedModifiers.length;n++)if(!0!==a.reset){var s=a.orderedModifiers[n],o=s.fn,r=s.options,l=void 0===r?{}:r,d=s.name;"function"==typeof o&&(a=o({state:a,options:l,name:d,instance:h})||a)}else a.reset=!1,n=-1}}},update:(s=function(){return new Promise((function(t){h.forceUpdate(),t(a)}))},function(){return r||(r=new Promise((function(t){Promise.resolve().then((function(){r=void 0,t(s())}))}))),r}),destroy:function(){d(),c=!0}};if(!pi(t,e))return h;function d(){l.forEach((function(t){return t()})),l=[]}return h.setOptions(i).then((function(t){!c&&i.onFirstUpdate&&i.onFirstUpdate(t)})),h}}var gi=mi(),_i=mi({defaultModifiers:[Re,ci,Be,_e]}),bi=mi({defaultModifiers:[Re,ci,Be,_e,li,si,hi,je,ai]});const vi=Object.freeze(Object.defineProperty({__proto__:null,afterMain:ae,afterRead:se,afterWrite:he,applyStyles:_e,arrow:je,auto:Kt,basePlacements:Qt,beforeMain:oe,beforeRead:ie,beforeWrite:le,bottom:Rt,clippingParents:Ut,computeStyles:Be,createPopper:bi,createPopperBase:gi,createPopperLite:_i,detectOverflow:ii,end:Yt,eventListeners:Re,flip:si,hide:ai,left:Vt,main:re,modifierPhases:de,offset:li,placements:ee,popper:Jt,popperGenerator:mi,popperOffsets:ci,preventOverflow:hi,read:ne,reference:Zt,right:qt,start:Xt,top:zt,variationPlacements:te,viewport:Gt,write:ce},Symbol.toStringTag,{value:"Module"})),yi="dropdown",wi=".bs.dropdown",Ai=".data-api",Ei="ArrowUp",Ti="ArrowDown",Ci=`hide${wi}`,Oi=`hidden${wi}`,xi=`show${wi}`,ki=`shown${wi}`,Li=`click${wi}${Ai}`,Si=`keydown${wi}${Ai}`,Di=`keyup${wi}${Ai}`,$i="show",Ii='[data-bs-toggle="dropdown"]:not(.disabled):not(:disabled)',Ni=`${Ii}.${$i}`,Pi=".dropdown-menu",Mi=p()?"top-end":"top-start",ji=p()?"top-start":"top-end",Fi=p()?"bottom-end":"bottom-start",Hi=p()?"bottom-start":"bottom-end",Wi=p()?"left-start":"right-start",Bi=p()?"right-start":"left-start",zi={autoClose:!0,boundary:"clippingParents",display:"dynamic",offset:[0,2],popperConfig:null,reference:"toggle"},Ri={autoClose:"(boolean|string)",boundary:"(string|element)",display:"string",offset:"(array|string|function)",popperConfig:"(null|object|function)",reference:"(string|element|object)"};class qi extends W{constructor(t,e){super(t,e),this._popper=null,this._parent=this._element.parentNode,this._menu=z.next(this._element,Pi)[0]||z.prev(this._element,Pi)[0]||z.findOne(Pi,this._parent),this._inNavbar=this._detectNavbar()}static get Default(){return zi}static get DefaultType(){return Ri}static get NAME(){return yi}toggle(){return this._isShown()?this.hide():this.show()}show(){if(l(this._element)||this._isShown())return;const t={relatedTarget:this._element};if(!N.trigger(this._element,xi,t).defaultPrevented){if(this._createPopper(),"ontouchstart"in document.documentElement&&!this._parent.closest(".navbar-nav"))for(const t of[].concat(...document.body.children))N.on(t,"mouseover",h);this._element.focus(),this._element.setAttribute("aria-expanded",!0),this._menu.classList.add($i),this._element.classList.add($i),N.trigger(this._element,ki,t)}}hide(){if(l(this._element)||!this._isShown())return;const t={relatedTarget:this._element};this._completeHide(t)}dispose(){this._popper&&this._popper.destroy(),super.dispose()}update(){this._inNavbar=this._detectNavbar(),this._popper&&this._popper.update()}_completeHide(t){if(!N.trigger(this._element,Ci,t).defaultPrevented){if("ontouchstart"in document.documentElement)for(const t of[].concat(...document.body.children))N.off(t,"mouseover",h);this._popper&&this._popper.destroy(),this._menu.classList.remove($i),this._element.classList.remove($i),this._element.setAttribute("aria-expanded","false"),F.removeDataAttribute(this._menu,"popper"),N.trigger(this._element,Oi,t)}}_getConfig(t){if("object"==typeof(t=super._getConfig(t)).reference&&!o(t.reference)&&"function"!=typeof t.reference.getBoundingClientRect)throw new TypeError(`${yi.toUpperCase()}: Option "reference" provided type "object" without a required "getBoundingClientRect" method.`);return t}_createPopper(){if(void 0===vi)throw new TypeError("Bootstrap's dropdowns require Popper (https://popper.js.org)");let t=this._element;"parent"===this._config.reference?t=this._parent:o(this._config.reference)?t=r(this._config.reference):"object"==typeof this._config.reference&&(t=this._config.reference);const e=this._getPopperConfig();this._popper=bi(t,this._menu,e)}_isShown(){return this._menu.classList.contains($i)}_getPlacement(){const t=this._parent;if(t.classList.contains("dropend"))return Wi;if(t.classList.contains("dropstart"))return Bi;if(t.classList.contains("dropup-center"))return"top";if(t.classList.contains("dropdown-center"))return"bottom";const e="end"===getComputedStyle(this._menu).getPropertyValue("--bs-position").trim();return t.classList.contains("dropup")?e?ji:Mi:e?Hi:Fi}_detectNavbar(){return null!==this._element.closest(".navbar")}_getOffset(){const{offset:t}=this._config;return"string"==typeof t?t.split(",").map((t=>Number.parseInt(t,10))):"function"==typeof t?e=>t(e,this._element):t}_getPopperConfig(){const t={placement:this._getPlacement(),modifiers:[{name:"preventOverflow",options:{boundary:this._config.boundary}},{name:"offset",options:{offset:this._getOffset()}}]};return(this._inNavbar||"static"===this._config.display)&&(F.setDataAttribute(this._menu,"popper","static"),t.modifiers=[{name:"applyStyles",enabled:!1}]),{...t,...g(this._config.popperConfig,[t])}}_selectMenuItem({key:t,target:e}){const i=z.find(".dropdown-menu .dropdown-item:not(.disabled):not(:disabled)",this._menu).filter((t=>a(t)));i.length&&b(i,e,t===Ti,!i.includes(e)).focus()}static jQueryInterface(t){return this.each((function(){const e=qi.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}static clearMenus(t){if(2===t.button||"keyup"===t.type&&"Tab"!==t.key)return;const e=z.find(Ni);for(const i of e){const e=qi.getInstance(i);if(!e||!1===e._config.autoClose)continue;const n=t.composedPath(),s=n.includes(e._menu);if(n.includes(e._element)||"inside"===e._config.autoClose&&!s||"outside"===e._config.autoClose&&s)continue;if(e._menu.contains(t.target)&&("keyup"===t.type&&"Tab"===t.key||/input|select|option|textarea|form/i.test(t.target.tagName)))continue;const o={relatedTarget:e._element};"click"===t.type&&(o.clickEvent=t),e._completeHide(o)}}static dataApiKeydownHandler(t){const e=/input|textarea/i.test(t.target.tagName),i="Escape"===t.key,n=[Ei,Ti].includes(t.key);if(!n&&!i)return;if(e&&!i)return;t.preventDefault();const s=this.matches(Ii)?this:z.prev(this,Ii)[0]||z.next(this,Ii)[0]||z.findOne(Ii,t.delegateTarget.parentNode),o=qi.getOrCreateInstance(s);if(n)return t.stopPropagation(),o.show(),void o._selectMenuItem(t);o._isShown()&&(t.stopPropagation(),o.hide(),s.focus())}}N.on(document,Si,Ii,qi.dataApiKeydownHandler),N.on(document,Si,Pi,qi.dataApiKeydownHandler),N.on(document,Li,qi.clearMenus),N.on(document,Di,qi.clearMenus),N.on(document,Li,Ii,(function(t){t.preventDefault(),qi.getOrCreateInstance(this).toggle()})),m(qi);const Vi="backdrop",Ki="show",Qi=`mousedown.bs.${Vi}`,Xi={className:"modal-backdrop",clickCallback:null,isAnimated:!1,isVisible:!0,rootElement:"body"},Yi={className:"string",clickCallback:"(function|null)",isAnimated:"boolean",isVisible:"boolean",rootElement:"(element|string)"};class Ui extends H{constructor(t){super(),this._config=this._getConfig(t),this._isAppended=!1,this._element=null}static get Default(){return Xi}static get DefaultType(){return Yi}static get NAME(){return Vi}show(t){if(!this._config.isVisible)return void g(t);this._append();const e=this._getElement();this._config.isAnimated&&d(e),e.classList.add(Ki),this._emulateAnimation((()=>{g(t)}))}hide(t){this._config.isVisible?(this._getElement().classList.remove(Ki),this._emulateAnimation((()=>{this.dispose(),g(t)}))):g(t)}dispose(){this._isAppended&&(N.off(this._element,Qi),this._element.remove(),this._isAppended=!1)}_getElement(){if(!this._element){const t=document.createElement("div");t.className=this._config.className,this._config.isAnimated&&t.classList.add("fade"),this._element=t}return this._element}_configAfterMerge(t){return t.rootElement=r(t.rootElement),t}_append(){if(this._isAppended)return;const t=this._getElement();this._config.rootElement.append(t),N.on(t,Qi,(()=>{g(this._config.clickCallback)})),this._isAppended=!0}_emulateAnimation(t){_(t,this._getElement(),this._config.isAnimated)}}const Gi=".bs.focustrap",Ji=`focusin${Gi}`,Zi=`keydown.tab${Gi}`,tn="backward",en={autofocus:!0,trapElement:null},nn={autofocus:"boolean",trapElement:"element"};class sn extends H{constructor(t){super(),this._config=this._getConfig(t),this._isActive=!1,this._lastTabNavDirection=null}static get Default(){return en}static get DefaultType(){return nn}static get NAME(){return"focustrap"}activate(){this._isActive||(this._config.autofocus&&this._config.trapElement.focus(),N.off(document,Gi),N.on(document,Ji,(t=>this._handleFocusin(t))),N.on(document,Zi,(t=>this._handleKeydown(t))),this._isActive=!0)}deactivate(){this._isActive&&(this._isActive=!1,N.off(document,Gi))}_handleFocusin(t){const{trapElement:e}=this._config;if(t.target===document||t.target===e||e.contains(t.target))return;const i=z.focusableChildren(e);0===i.length?e.focus():this._lastTabNavDirection===tn?i[i.length-1].focus():i[0].focus()}_handleKeydown(t){"Tab"===t.key&&(this._lastTabNavDirection=t.shiftKey?tn:"forward")}}const on=".fixed-top, .fixed-bottom, .is-fixed, .sticky-top",rn=".sticky-top",an="padding-right",ln="margin-right";class cn{constructor(){this._element=document.body}getWidth(){const t=document.documentElement.clientWidth;return Math.abs(window.innerWidth-t)}hide(){const t=this.getWidth();this._disableOverFlow(),this._setElementAttributes(this._element,an,(e=>e+t)),this._setElementAttributes(on,an,(e=>e+t)),this._setElementAttributes(rn,ln,(e=>e-t))}reset(){this._resetElementAttributes(this._element,"overflow"),this._resetElementAttributes(this._element,an),this._resetElementAttributes(on,an),this._resetElementAttributes(rn,ln)}isOverflowing(){return this.getWidth()>0}_disableOverFlow(){this._saveInitialAttribute(this._element,"overflow"),this._element.style.overflow="hidden"}_setElementAttributes(t,e,i){const n=this.getWidth();this._applyManipulationCallback(t,(t=>{if(t!==this._element&&window.innerWidth>t.clientWidth+n)return;this._saveInitialAttribute(t,e);const s=window.getComputedStyle(t).getPropertyValue(e);t.style.setProperty(e,`${i(Number.parseFloat(s))}px`)}))}_saveInitialAttribute(t,e){const i=t.style.getPropertyValue(e);i&&F.setDataAttribute(t,e,i)}_resetElementAttributes(t,e){this._applyManipulationCallback(t,(t=>{const i=F.getDataAttribute(t,e);null!==i?(F.removeDataAttribute(t,e),t.style.setProperty(e,i)):t.style.removeProperty(e)}))}_applyManipulationCallback(t,e){if(o(t))e(t);else for(const i of z.find(t,this._element))e(i)}}const hn=".bs.modal",dn=`hide${hn}`,un=`hidePrevented${hn}`,fn=`hidden${hn}`,pn=`show${hn}`,mn=`shown${hn}`,gn=`resize${hn}`,_n=`click.dismiss${hn}`,bn=`mousedown.dismiss${hn}`,vn=`keydown.dismiss${hn}`,yn=`click${hn}.data-api`,wn="modal-open",An="show",En="modal-static",Tn={backdrop:!0,focus:!0,keyboard:!0},Cn={backdrop:"(boolean|string)",focus:"boolean",keyboard:"boolean"};class On extends W{constructor(t,e){super(t,e),this._dialog=z.findOne(".modal-dialog",this._element),this._backdrop=this._initializeBackDrop(),this._focustrap=this._initializeFocusTrap(),this._isShown=!1,this._isTransitioning=!1,this._scrollBar=new cn,this._addEventListeners()}static get Default(){return Tn}static get DefaultType(){return Cn}static get NAME(){return"modal"}toggle(t){return this._isShown?this.hide():this.show(t)}show(t){this._isShown||this._isTransitioning||N.trigger(this._element,pn,{relatedTarget:t}).defaultPrevented||(this._isShown=!0,this._isTransitioning=!0,this._scrollBar.hide(),document.body.classList.add(wn),this._adjustDialog(),this._backdrop.show((()=>this._showElement(t))))}hide(){this._isShown&&!this._isTransitioning&&(N.trigger(this._element,dn).defaultPrevented||(this._isShown=!1,this._isTransitioning=!0,this._focustrap.deactivate(),this._element.classList.remove(An),this._queueCallback((()=>this._hideModal()),this._element,this._isAnimated())))}dispose(){N.off(window,hn),N.off(this._dialog,hn),this._backdrop.dispose(),this._focustrap.deactivate(),super.dispose()}handleUpdate(){this._adjustDialog()}_initializeBackDrop(){return new Ui({isVisible:Boolean(this._config.backdrop),isAnimated:this._isAnimated()})}_initializeFocusTrap(){return new sn({trapElement:this._element})}_showElement(t){document.body.contains(this._element)||document.body.append(this._element),this._element.style.display="block",this._element.removeAttribute("aria-hidden"),this._element.setAttribute("aria-modal",!0),this._element.setAttribute("role","dialog"),this._element.scrollTop=0;const e=z.findOne(".modal-body",this._dialog);e&&(e.scrollTop=0),d(this._element),this._element.classList.add(An),this._queueCallback((()=>{this._config.focus&&this._focustrap.activate(),this._isTransitioning=!1,N.trigger(this._element,mn,{relatedTarget:t})}),this._dialog,this._isAnimated())}_addEventListeners(){N.on(this._element,vn,(t=>{"Escape"===t.key&&(this._config.keyboard?this.hide():this._triggerBackdropTransition())})),N.on(window,gn,(()=>{this._isShown&&!this._isTransitioning&&this._adjustDialog()})),N.on(this._element,bn,(t=>{N.one(this._element,_n,(e=>{this._element===t.target&&this._element===e.target&&("static"!==this._config.backdrop?this._config.backdrop&&this.hide():this._triggerBackdropTransition())}))}))}_hideModal(){this._element.style.display="none",this._element.setAttribute("aria-hidden",!0),this._element.removeAttribute("aria-modal"),this._element.removeAttribute("role"),this._isTransitioning=!1,this._backdrop.hide((()=>{document.body.classList.remove(wn),this._resetAdjustments(),this._scrollBar.reset(),N.trigger(this._element,fn)}))}_isAnimated(){return this._element.classList.contains("fade")}_triggerBackdropTransition(){if(N.trigger(this._element,un).defaultPrevented)return;const t=this._element.scrollHeight>document.documentElement.clientHeight,e=this._element.style.overflowY;"hidden"===e||this._element.classList.contains(En)||(t||(this._element.style.overflowY="hidden"),this._element.classList.add(En),this._queueCallback((()=>{this._element.classList.remove(En),this._queueCallback((()=>{this._element.style.overflowY=e}),this._dialog)}),this._dialog),this._element.focus())}_adjustDialog(){const t=this._element.scrollHeight>document.documentElement.clientHeight,e=this._scrollBar.getWidth(),i=e>0;if(i&&!t){const t=p()?"paddingLeft":"paddingRight";this._element.style[t]=`${e}px`}if(!i&&t){const t=p()?"paddingRight":"paddingLeft";this._element.style[t]=`${e}px`}}_resetAdjustments(){this._element.style.paddingLeft="",this._element.style.paddingRight=""}static jQueryInterface(t,e){return this.each((function(){const i=On.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===i[t])throw new TypeError(`No method named "${t}"`);i[t](e)}}))}}N.on(document,yn,'[data-bs-toggle="modal"]',(function(t){const e=z.getElementFromSelector(this);["A","AREA"].includes(this.tagName)&&t.preventDefault(),N.one(e,pn,(t=>{t.defaultPrevented||N.one(e,fn,(()=>{a(this)&&this.focus()}))}));const i=z.findOne(".modal.show");i&&On.getInstance(i).hide(),On.getOrCreateInstance(e).toggle(this)})),R(On),m(On);const xn=".bs.offcanvas",kn=".data-api",Ln=`load${xn}${kn}`,Sn="show",Dn="showing",$n="hiding",In=".offcanvas.show",Nn=`show${xn}`,Pn=`shown${xn}`,Mn=`hide${xn}`,jn=`hidePrevented${xn}`,Fn=`hidden${xn}`,Hn=`resize${xn}`,Wn=`click${xn}${kn}`,Bn=`keydown.dismiss${xn}`,zn={backdrop:!0,keyboard:!0,scroll:!1},Rn={backdrop:"(boolean|string)",keyboard:"boolean",scroll:"boolean"};class qn extends W{constructor(t,e){super(t,e),this._isShown=!1,this._backdrop=this._initializeBackDrop(),this._focustrap=this._initializeFocusTrap(),this._addEventListeners()}static get Default(){return zn}static get DefaultType(){return Rn}static get NAME(){return"offcanvas"}toggle(t){return this._isShown?this.hide():this.show(t)}show(t){this._isShown||N.trigger(this._element,Nn,{relatedTarget:t}).defaultPrevented||(this._isShown=!0,this._backdrop.show(),this._config.scroll||(new cn).hide(),this._element.setAttribute("aria-modal",!0),this._element.setAttribute("role","dialog"),this._element.classList.add(Dn),this._queueCallback((()=>{this._config.scroll&&!this._config.backdrop||this._focustrap.activate(),this._element.classList.add(Sn),this._element.classList.remove(Dn),N.trigger(this._element,Pn,{relatedTarget:t})}),this._element,!0))}hide(){this._isShown&&(N.trigger(this._element,Mn).defaultPrevented||(this._focustrap.deactivate(),this._element.blur(),this._isShown=!1,this._element.classList.add($n),this._backdrop.hide(),this._queueCallback((()=>{this._element.classList.remove(Sn,$n),this._element.removeAttribute("aria-modal"),this._element.removeAttribute("role"),this._config.scroll||(new cn).reset(),N.trigger(this._element,Fn)}),this._element,!0)))}dispose(){this._backdrop.dispose(),this._focustrap.deactivate(),super.dispose()}_initializeBackDrop(){const t=Boolean(this._config.backdrop);return new Ui({className:"offcanvas-backdrop",isVisible:t,isAnimated:!0,rootElement:this._element.parentNode,clickCallback:t?()=>{"static"!==this._config.backdrop?this.hide():N.trigger(this._element,jn)}:null})}_initializeFocusTrap(){return new sn({trapElement:this._element})}_addEventListeners(){N.on(this._element,Bn,(t=>{"Escape"===t.key&&(this._config.keyboard?this.hide():N.trigger(this._element,jn))}))}static jQueryInterface(t){return this.each((function(){const e=qn.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t]||t.startsWith("_")||"constructor"===t)throw new TypeError(`No method named "${t}"`);e[t](this)}}))}}N.on(document,Wn,'[data-bs-toggle="offcanvas"]',(function(t){const e=z.getElementFromSelector(this);if(["A","AREA"].includes(this.tagName)&&t.preventDefault(),l(this))return;N.one(e,Fn,(()=>{a(this)&&this.focus()}));const i=z.findOne(In);i&&i!==e&&qn.getInstance(i).hide(),qn.getOrCreateInstance(e).toggle(this)})),N.on(window,Ln,(()=>{for(const t of z.find(In))qn.getOrCreateInstance(t).show()})),N.on(window,Hn,(()=>{for(const t of z.find("[aria-modal][class*=show][class*=offcanvas-]"))"fixed"!==getComputedStyle(t).position&&qn.getOrCreateInstance(t).hide()})),R(qn),m(qn);const Vn={"*":["class","dir","id","lang","role",/^aria-[\w-]*$/i],a:["target","href","title","rel"],area:[],b:[],br:[],col:[],code:[],div:[],em:[],hr:[],h1:[],h2:[],h3:[],h4:[],h5:[],h6:[],i:[],img:["src","srcset","alt","title","width","height"],li:[],ol:[],p:[],pre:[],s:[],small:[],span:[],sub:[],sup:[],strong:[],u:[],ul:[]},Kn=new Set(["background","cite","href","itemtype","longdesc","poster","src","xlink:href"]),Qn=/^(?!javascript:)(?:[a-z0-9+.-]+:|[^&:/?#]*(?:[/?#]|$))/i,Xn=(t,e)=>{const i=t.nodeName.toLowerCase();return e.includes(i)?!Kn.has(i)||Boolean(Qn.test(t.nodeValue)):e.filter((t=>t instanceof RegExp)).some((t=>t.test(i)))},Yn={allowList:Vn,content:{},extraClass:"",html:!1,sanitize:!0,sanitizeFn:null,template:"<div></div>"},Un={allowList:"object",content:"object",extraClass:"(string|function)",html:"boolean",sanitize:"boolean",sanitizeFn:"(null|function)",template:"string"},Gn={entry:"(string|element|function|null)",selector:"(string|element)"};class Jn extends H{constructor(t){super(),this._config=this._getConfig(t)}static get Default(){return Yn}static get DefaultType(){return Un}static get NAME(){return"TemplateFactory"}getContent(){return Object.values(this._config.content).map((t=>this._resolvePossibleFunction(t))).filter(Boolean)}hasContent(){return this.getContent().length>0}changeContent(t){return this._checkContent(t),this._config.content={...this._config.content,...t},this}toHtml(){const t=document.createElement("div");t.innerHTML=this._maybeSanitize(this._config.template);for(const[e,i]of Object.entries(this._config.content))this._setContent(t,i,e);const e=t.children[0],i=this._resolvePossibleFunction(this._config.extraClass);return i&&e.classList.add(...i.split(" ")),e}_typeCheckConfig(t){super._typeCheckConfig(t),this._checkContent(t.content)}_checkContent(t){for(const[e,i]of Object.entries(t))super._typeCheckConfig({selector:e,entry:i},Gn)}_setContent(t,e,i){const n=z.findOne(i,t);n&&((e=this._resolvePossibleFunction(e))?o(e)?this._putElementInTemplate(r(e),n):this._config.html?n.innerHTML=this._maybeSanitize(e):n.textContent=e:n.remove())}_maybeSanitize(t){return this._config.sanitize?function(t,e,i){if(!t.length)return t;if(i&&"function"==typeof i)return i(t);const n=(new window.DOMParser).parseFromString(t,"text/html"),s=[].concat(...n.body.querySelectorAll("*"));for(const t of s){const i=t.nodeName.toLowerCase();if(!Object.keys(e).includes(i)){t.remove();continue}const n=[].concat(...t.attributes),s=[].concat(e["*"]||[],e[i]||[]);for(const e of n)Xn(e,s)||t.removeAttribute(e.nodeName)}return n.body.innerHTML}(t,this._config.allowList,this._config.sanitizeFn):t}_resolvePossibleFunction(t){return g(t,[this])}_putElementInTemplate(t,e){if(this._config.html)return e.innerHTML="",void e.append(t);e.textContent=t.textContent}}const Zn=new Set(["sanitize","allowList","sanitizeFn"]),ts="fade",es="show",is=".modal",ns="hide.bs.modal",ss="hover",os="focus",rs={AUTO:"auto",TOP:"top",RIGHT:p()?"left":"right",BOTTOM:"bottom",LEFT:p()?"right":"left"},as={allowList:Vn,animation:!0,boundary:"clippingParents",container:!1,customClass:"",delay:0,fallbackPlacements:["top","right","bottom","left"],html:!1,offset:[0,6],placement:"top",popperConfig:null,sanitize:!0,sanitizeFn:null,selector:!1,template:'<div class="tooltip" role="tooltip"><div class="tooltip-arrow"></div><div class="tooltip-inner"></div></div>',title:"",trigger:"hover focus"},ls={allowList:"object",animation:"boolean",boundary:"(string|element)",container:"(string|element|boolean)",customClass:"(string|function)",delay:"(number|object)",fallbackPlacements:"array",html:"boolean",offset:"(array|string|function)",placement:"(string|function)",popperConfig:"(null|object|function)",sanitize:"boolean",sanitizeFn:"(null|function)",selector:"(string|boolean)",template:"string",title:"(string|element|function)",trigger:"string"};class cs extends W{constructor(t,e){if(void 0===vi)throw new TypeError("Bootstrap's tooltips require Popper (https://popper.js.org)");super(t,e),this._isEnabled=!0,this._timeout=0,this._isHovered=null,this._activeTrigger={},this._popper=null,this._templateFactory=null,this._newContent=null,this.tip=null,this._setListeners(),this._config.selector||this._fixTitle()}static get Default(){return as}static get DefaultType(){return ls}static get NAME(){return"tooltip"}enable(){this._isEnabled=!0}disable(){this._isEnabled=!1}toggleEnabled(){this._isEnabled=!this._isEnabled}toggle(){this._isEnabled&&(this._activeTrigger.click=!this._activeTrigger.click,this._isShown()?this._leave():this._enter())}dispose(){clearTimeout(this._timeout),N.off(this._element.closest(is),ns,this._hideModalHandler),this._element.getAttribute("data-bs-original-title")&&this._element.setAttribute("title",this._element.getAttribute("data-bs-original-title")),this._disposePopper(),super.dispose()}show(){if("none"===this._element.style.display)throw new Error("Please use show on visible elements");if(!this._isWithContent()||!this._isEnabled)return;const t=N.trigger(this._element,this.constructor.eventName("show")),e=(c(this._element)||this._element.ownerDocument.documentElement).contains(this._element);if(t.defaultPrevented||!e)return;this._disposePopper();const i=this._getTipElement();this._element.setAttribute("aria-describedby",i.getAttribute("id"));const{container:n}=this._config;if(this._element.ownerDocument.documentElement.contains(this.tip)||(n.append(i),N.trigger(this._element,this.constructor.eventName("inserted"))),this._popper=this._createPopper(i),i.classList.add(es),"ontouchstart"in document.documentElement)for(const t of[].concat(...document.body.children))N.on(t,"mouseover",h);this._queueCallback((()=>{N.trigger(this._element,this.constructor.eventName("shown")),!1===this._isHovered&&this._leave(),this._isHovered=!1}),this.tip,this._isAnimated())}hide(){if(this._isShown()&&!N.trigger(this._element,this.constructor.eventName("hide")).defaultPrevented){if(this._getTipElement().classList.remove(es),"ontouchstart"in document.documentElement)for(const t of[].concat(...document.body.children))N.off(t,"mouseover",h);this._activeTrigger.click=!1,this._activeTrigger[os]=!1,this._activeTrigger[ss]=!1,this._isHovered=null,this._queueCallback((()=>{this._isWithActiveTrigger()||(this._isHovered||this._disposePopper(),this._element.removeAttribute("aria-describedby"),N.trigger(this._element,this.constructor.eventName("hidden")))}),this.tip,this._isAnimated())}}update(){this._popper&&this._popper.update()}_isWithContent(){return Boolean(this._getTitle())}_getTipElement(){return this.tip||(this.tip=this._createTipElement(this._newContent||this._getContentForTemplate())),this.tip}_createTipElement(t){const e=this._getTemplateFactory(t).toHtml();if(!e)return null;e.classList.remove(ts,es),e.classList.add(`bs-${this.constructor.NAME}-auto`);const i=(t=>{do{t+=Math.floor(1e6*Math.random())}while(document.getElementById(t));return t})(this.constructor.NAME).toString();return e.setAttribute("id",i),this._isAnimated()&&e.classList.add(ts),e}setContent(t){this._newContent=t,this._isShown()&&(this._disposePopper(),this.show())}_getTemplateFactory(t){return this._templateFactory?this._templateFactory.changeContent(t):this._templateFactory=new Jn({...this._config,content:t,extraClass:this._resolvePossibleFunction(this._config.customClass)}),this._templateFactory}_getContentForTemplate(){return{".tooltip-inner":this._getTitle()}}_getTitle(){return this._resolvePossibleFunction(this._config.title)||this._element.getAttribute("data-bs-original-title")}_initializeOnDelegatedTarget(t){return this.constructor.getOrCreateInstance(t.delegateTarget,this._getDelegateConfig())}_isAnimated(){return this._config.animation||this.tip&&this.tip.classList.contains(ts)}_isShown(){return this.tip&&this.tip.classList.contains(es)}_createPopper(t){const e=g(this._config.placement,[this,t,this._element]),i=rs[e.toUpperCase()];return bi(this._element,t,this._getPopperConfig(i))}_getOffset(){const{offset:t}=this._config;return"string"==typeof t?t.split(",").map((t=>Number.parseInt(t,10))):"function"==typeof t?e=>t(e,this._element):t}_resolvePossibleFunction(t){return g(t,[this._element])}_getPopperConfig(t){const e={placement:t,modifiers:[{name:"flip",options:{fallbackPlacements:this._config.fallbackPlacements}},{name:"offset",options:{offset:this._getOffset()}},{name:"preventOverflow",options:{boundary:this._config.boundary}},{name:"arrow",options:{element:`.${this.constructor.NAME}-arrow`}},{name:"preSetPlacement",enabled:!0,phase:"beforeMain",fn:t=>{this._getTipElement().setAttribute("data-popper-placement",t.state.placement)}}]};return{...e,...g(this._config.popperConfig,[e])}}_setListeners(){const t=this._config.trigger.split(" ");for(const e of t)if("click"===e)N.on(this._element,this.constructor.eventName("click"),this._config.selector,(t=>{this._initializeOnDelegatedTarget(t).toggle()}));else if("manual"!==e){const t=e===ss?this.constructor.eventName("mouseenter"):this.constructor.eventName("focusin"),i=e===ss?this.constructor.eventName("mouseleave"):this.constructor.eventName("focusout");N.on(this._element,t,this._config.selector,(t=>{const e=this._initializeOnDelegatedTarget(t);e._activeTrigger["focusin"===t.type?os:ss]=!0,e._enter()})),N.on(this._element,i,this._config.selector,(t=>{const e=this._initializeOnDelegatedTarget(t);e._activeTrigger["focusout"===t.type?os:ss]=e._element.contains(t.relatedTarget),e._leave()}))}this._hideModalHandler=()=>{this._element&&this.hide()},N.on(this._element.closest(is),ns,this._hideModalHandler)}_fixTitle(){const t=this._element.getAttribute("title");t&&(this._element.getAttribute("aria-label")||this._element.textContent.trim()||this._element.setAttribute("aria-label",t),this._element.setAttribute("data-bs-original-title",t),this._element.removeAttribute("title"))}_enter(){this._isShown()||this._isHovered?this._isHovered=!0:(this._isHovered=!0,this._setTimeout((()=>{this._isHovered&&this.show()}),this._config.delay.show))}_leave(){this._isWithActiveTrigger()||(this._isHovered=!1,this._setTimeout((()=>{this._isHovered||this.hide()}),this._config.delay.hide))}_setTimeout(t,e){clearTimeout(this._timeout),this._timeout=setTimeout(t,e)}_isWithActiveTrigger(){return Object.values(this._activeTrigger).includes(!0)}_getConfig(t){const e=F.getDataAttributes(this._element);for(const t of Object.keys(e))Zn.has(t)&&delete e[t];return t={...e,..."object"==typeof t&&t?t:{}},t=this._mergeConfigObj(t),t=this._configAfterMerge(t),this._typeCheckConfig(t),t}_configAfterMerge(t){return t.container=!1===t.container?document.body:r(t.container),"number"==typeof t.delay&&(t.delay={show:t.delay,hide:t.delay}),"number"==typeof t.title&&(t.title=t.title.toString()),"number"==typeof t.content&&(t.content=t.content.toString()),t}_getDelegateConfig(){const t={};for(const[e,i]of Object.entries(this._config))this.constructor.Default[e]!==i&&(t[e]=i);return t.selector=!1,t.trigger="manual",t}_disposePopper(){this._popper&&(this._popper.destroy(),this._popper=null),this.tip&&(this.tip.remove(),this.tip=null)}static jQueryInterface(t){return this.each((function(){const e=cs.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}}m(cs);const hs={...cs.Default,content:"",offset:[0,8],placement:"right",template:'<div class="popover" role="tooltip"><div class="popover-arrow"></div><h3 class="popover-header"></h3><div class="popover-body"></div></div>',trigger:"click"},ds={...cs.DefaultType,content:"(null|string|element|function)"};class us extends cs{static get Default(){return hs}static get DefaultType(){return ds}static get NAME(){return"popover"}_isWithContent(){return this._getTitle()||this._getContent()}_getContentForTemplate(){return{".popover-header":this._getTitle(),".popover-body":this._getContent()}}_getContent(){return this._resolvePossibleFunction(this._config.content)}static jQueryInterface(t){return this.each((function(){const e=us.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}}m(us);const fs=".bs.scrollspy",ps=`activate${fs}`,ms=`click${fs}`,gs=`load${fs}.data-api`,_s="active",bs="[href]",vs=".nav-link",ys=`${vs}, .nav-item > ${vs}, .list-group-item`,ws={offset:null,rootMargin:"0px 0px -25%",smoothScroll:!1,target:null,threshold:[.1,.5,1]},As={offset:"(number|null)",rootMargin:"string",smoothScroll:"boolean",target:"element",threshold:"array"};class Es extends W{constructor(t,e){super(t,e),this._targetLinks=new Map,this._observableSections=new Map,this._rootElement="visible"===getComputedStyle(this._element).overflowY?null:this._element,this._activeTarget=null,this._observer=null,this._previousScrollData={visibleEntryTop:0,parentScrollTop:0},this.refresh()}static get Default(){return ws}static get DefaultType(){return As}static get NAME(){return"scrollspy"}refresh(){this._initializeTargetsAndObservables(),this._maybeEnableSmoothScroll(),this._observer?this._observer.disconnect():this._observer=this._getNewObserver();for(const t of this._observableSections.values())this._observer.observe(t)}dispose(){this._observer.disconnect(),super.dispose()}_configAfterMerge(t){return t.target=r(t.target)||document.body,t.rootMargin=t.offset?`${t.offset}px 0px -30%`:t.rootMargin,"string"==typeof t.threshold&&(t.threshold=t.threshold.split(",").map((t=>Number.parseFloat(t)))),t}_maybeEnableSmoothScroll(){this._config.smoothScroll&&(N.off(this._config.target,ms),N.on(this._config.target,ms,bs,(t=>{const e=this._observableSections.get(t.target.hash);if(e){t.preventDefault();const i=this._rootElement||window,n=e.offsetTop-this._element.offsetTop;if(i.scrollTo)return void i.scrollTo({top:n,behavior:"smooth"});i.scrollTop=n}})))}_getNewObserver(){const t={root:this._rootElement,threshold:this._config.threshold,rootMargin:this._config.rootMargin};return new IntersectionObserver((t=>this._observerCallback(t)),t)}_observerCallback(t){const e=t=>this._targetLinks.get(`#${t.target.id}`),i=t=>{this._previousScrollData.visibleEntryTop=t.target.offsetTop,this._process(e(t))},n=(this._rootElement||document.documentElement).scrollTop,s=n>=this._previousScrollData.parentScrollTop;this._previousScrollData.parentScrollTop=n;for(const o of t){if(!o.isIntersecting){this._activeTarget=null,this._clearActiveClass(e(o));continue}const t=o.target.offsetTop>=this._previousScrollData.visibleEntryTop;if(s&&t){if(i(o),!n)return}else s||t||i(o)}}_initializeTargetsAndObservables(){this._targetLinks=new Map,this._observableSections=new Map;const t=z.find(bs,this._config.target);for(const e of t){if(!e.hash||l(e))continue;const t=z.findOne(decodeURI(e.hash),this._element);a(t)&&(this._targetLinks.set(decodeURI(e.hash),e),this._observableSections.set(e.hash,t))}}_process(t){this._activeTarget!==t&&(this._clearActiveClass(this._config.target),this._activeTarget=t,t.classList.add(_s),this._activateParents(t),N.trigger(this._element,ps,{relatedTarget:t}))}_activateParents(t){if(t.classList.contains("dropdown-item"))z.findOne(".dropdown-toggle",t.closest(".dropdown")).classList.add(_s);else for(const e of z.parents(t,".nav, .list-group"))for(const t of z.prev(e,ys))t.classList.add(_s)}_clearActiveClass(t){t.classList.remove(_s);const e=z.find(`${bs}.${_s}`,t);for(const t of e)t.classList.remove(_s)}static jQueryInterface(t){return this.each((function(){const e=Es.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t]||t.startsWith("_")||"constructor"===t)throw new TypeError(`No method named "${t}"`);e[t]()}}))}}N.on(window,gs,(()=>{for(const t of z.find('[data-bs-spy="scroll"]'))Es.getOrCreateInstance(t)})),m(Es);const Ts=".bs.tab",Cs=`hide${Ts}`,Os=`hidden${Ts}`,xs=`show${Ts}`,ks=`shown${Ts}`,Ls=`click${Ts}`,Ss=`keydown${Ts}`,Ds=`load${Ts}`,$s="ArrowLeft",Is="ArrowRight",Ns="ArrowUp",Ps="ArrowDown",Ms="Home",js="End",Fs="active",Hs="fade",Ws="show",Bs=":not(.dropdown-toggle)",zs='[data-bs-toggle="tab"], [data-bs-toggle="pill"], [data-bs-toggle="list"]',Rs=`.nav-link${Bs}, .list-group-item${Bs}, [role="tab"]${Bs}, ${zs}`,qs=`.${Fs}[data-bs-toggle="tab"], .${Fs}[data-bs-toggle="pill"], .${Fs}[data-bs-toggle="list"]`;class Vs extends W{constructor(t){super(t),this._parent=this._element.closest('.list-group, .nav, [role="tablist"]'),this._parent&&(this._setInitialAttributes(this._parent,this._getChildren()),N.on(this._element,Ss,(t=>this._keydown(t))))}static get NAME(){return"tab"}show(){const t=this._element;if(this._elemIsActive(t))return;const e=this._getActiveElem(),i=e?N.trigger(e,Cs,{relatedTarget:t}):null;N.trigger(t,xs,{relatedTarget:e}).defaultPrevented||i&&i.defaultPrevented||(this._deactivate(e,t),this._activate(t,e))}_activate(t,e){t&&(t.classList.add(Fs),this._activate(z.getElementFromSelector(t)),this._queueCallback((()=>{"tab"===t.getAttribute("role")?(t.removeAttribute("tabindex"),t.setAttribute("aria-selected",!0),this._toggleDropDown(t,!0),N.trigger(t,ks,{relatedTarget:e})):t.classList.add(Ws)}),t,t.classList.contains(Hs)))}_deactivate(t,e){t&&(t.classList.remove(Fs),t.blur(),this._deactivate(z.getElementFromSelector(t)),this._queueCallback((()=>{"tab"===t.getAttribute("role")?(t.setAttribute("aria-selected",!1),t.setAttribute("tabindex","-1"),this._toggleDropDown(t,!1),N.trigger(t,Os,{relatedTarget:e})):t.classList.remove(Ws)}),t,t.classList.contains(Hs)))}_keydown(t){if(![$s,Is,Ns,Ps,Ms,js].includes(t.key))return;t.stopPropagation(),t.preventDefault();const e=this._getChildren().filter((t=>!l(t)));let i;if([Ms,js].includes(t.key))i=e[t.key===Ms?0:e.length-1];else{const n=[Is,Ps].includes(t.key);i=b(e,t.target,n,!0)}i&&(i.focus({preventScroll:!0}),Vs.getOrCreateInstance(i).show())}_getChildren(){return z.find(Rs,this._parent)}_getActiveElem(){return this._getChildren().find((t=>this._elemIsActive(t)))||null}_setInitialAttributes(t,e){this._setAttributeIfNotExists(t,"role","tablist");for(const t of e)this._setInitialAttributesOnChild(t)}_setInitialAttributesOnChild(t){t=this._getInnerElement(t);const e=this._elemIsActive(t),i=this._getOuterElement(t);t.setAttribute("aria-selected",e),i!==t&&this._setAttributeIfNotExists(i,"role","presentation"),e||t.setAttribute("tabindex","-1"),this._setAttributeIfNotExists(t,"role","tab"),this._setInitialAttributesOnTargetPanel(t)}_setInitialAttributesOnTargetPanel(t){const e=z.getElementFromSelector(t);e&&(this._setAttributeIfNotExists(e,"role","tabpanel"),t.id&&this._setAttributeIfNotExists(e,"aria-labelledby",`${t.id}`))}_toggleDropDown(t,e){const i=this._getOuterElement(t);if(!i.classList.contains("dropdown"))return;const n=(t,n)=>{const s=z.findOne(t,i);s&&s.classList.toggle(n,e)};n(".dropdown-toggle",Fs),n(".dropdown-menu",Ws),i.setAttribute("aria-expanded",e)}_setAttributeIfNotExists(t,e,i){t.hasAttribute(e)||t.setAttribute(e,i)}_elemIsActive(t){return t.classList.contains(Fs)}_getInnerElement(t){return t.matches(Rs)?t:z.findOne(Rs,t)}_getOuterElement(t){return t.closest(".nav-item, .list-group-item")||t}static jQueryInterface(t){return this.each((function(){const e=Vs.getOrCreateInstance(this);if("string"==typeof t){if(void 0===e[t]||t.startsWith("_")||"constructor"===t)throw new TypeError(`No method named "${t}"`);e[t]()}}))}}N.on(document,Ls,zs,(function(t){["A","AREA"].includes(this.tagName)&&t.preventDefault(),l(this)||Vs.getOrCreateInstance(this).show()})),N.on(window,Ds,(()=>{for(const t of z.find(qs))Vs.getOrCreateInstance(t)})),m(Vs);const Ks=".bs.toast",Qs=`mouseover${Ks}`,Xs=`mouseout${Ks}`,Ys=`focusin${Ks}`,Us=`focusout${Ks}`,Gs=`hide${Ks}`,Js=`hidden${Ks}`,Zs=`show${Ks}`,to=`shown${Ks}`,eo="hide",io="show",no="showing",so={animation:"boolean",autohide:"boolean",delay:"number"},oo={animation:!0,autohide:!0,delay:5e3};class ro extends W{constructor(t,e){super(t,e),this._timeout=null,this._hasMouseInteraction=!1,this._hasKeyboardInteraction=!1,this._setListeners()}static get Default(){return oo}static get DefaultType(){return so}static get NAME(){return"toast"}show(){N.trigger(this._element,Zs).defaultPrevented||(this._clearTimeout(),this._config.animation&&this._element.classList.add("fade"),this._element.classList.remove(eo),d(this._element),this._element.classList.add(io,no),this._queueCallback((()=>{this._element.classList.remove(no),N.trigger(this._element,to),this._maybeScheduleHide()}),this._element,this._config.animation))}hide(){this.isShown()&&(N.trigger(this._element,Gs).defaultPrevented||(this._element.classList.add(no),this._queueCallback((()=>{this._element.classList.add(eo),this._element.classList.remove(no,io),N.trigger(this._element,Js)}),this._element,this._config.animation)))}dispose(){this._clearTimeout(),this.isShown()&&this._element.classList.remove(io),super.dispose()}isShown(){return this._element.classList.contains(io)}_maybeScheduleHide(){this._config.autohide&&(this._hasMouseInteraction||this._hasKeyboardInteraction||(this._timeout=setTimeout((()=>{this.hide()}),this._config.delay)))}_onInteraction(t,e){switch(t.type){case"mouseover":case"mouseout":this._hasMouseInteraction=e;break;case"focusin":case"focusout":this._hasKeyboardInteraction=e}if(e)return void this._clearTimeout();const i=t.relatedTarget;this._element===i||this._element.contains(i)||this._maybeScheduleHide()}_setListeners(){N.on(this._element,Qs,(t=>this._onInteraction(t,!0))),N.on(this._element,Xs,(t=>this._onInteraction(t,!1))),N.on(this._element,Ys,(t=>this._onInteraction(t,!0))),N.on(this._element,Us,(t=>this._onInteraction(t,!1)))}_clearTimeout(){clearTimeout(this._timeout),this._timeout=null}static jQueryInterface(t){return this.each((function(){const e=ro.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t](this)}}))}}return R(ro),m(ro),{Alert:Q,Button:Y,Carousel:xt,Collapse:Bt,Dropdown:qi,Modal:On,Offcanvas:qn,Popover:us,ScrollSpy:Es,Tab:Vs,Toast:ro,Tooltip:cs}}));
+//# sourceMappingURL=bootstrap.bundle.min.js.map
\ No newline at end of file
diff --git a/docs/2_39/site_libs/clipboard/clipboard.min.js b/docs/2_39/site_libs/clipboard/clipboard.min.js
new file mode 100644
index 000000000..1103f811e
--- /dev/null
+++ b/docs/2_39/site_libs/clipboard/clipboard.min.js
@@ -0,0 +1,7 @@
+/*!
+ * clipboard.js v2.0.11
+ * https://clipboardjs.com/
+ *
+ * Licensed MIT © Zeno Rocha
+ */
+!function(t,e){"object"==typeof exports&&"object"==typeof module?module.exports=e():"function"==typeof define&&define.amd?define([],e):"object"==typeof exports?exports.ClipboardJS=e():t.ClipboardJS=e()}(this,function(){return n={686:function(t,e,n){"use strict";n.d(e,{default:function(){return b}});var e=n(279),i=n.n(e),e=n(370),u=n.n(e),e=n(817),r=n.n(e);function c(t){try{return document.execCommand(t)}catch(t){return}}var a=function(t){t=r()(t);return c("cut"),t};function o(t,e){var n,o,t=(n=t,o="rtl"===document.documentElement.getAttribute("dir"),(t=document.createElement("textarea")).style.fontSize="12pt",t.style.border="0",t.style.padding="0",t.style.margin="0",t.style.position="absolute",t.style[o?"right":"left"]="-9999px",o=window.pageYOffset||document.documentElement.scrollTop,t.style.top="".concat(o,"px"),t.setAttribute("readonly",""),t.value=n,t);return e.container.appendChild(t),e=r()(t),c("copy"),t.remove(),e}var f=function(t){var e=1<arguments.length&&void 0!==arguments[1]?arguments[1]:{container:document.body},n="";return"string"==typeof t?n=o(t,e):t instanceof HTMLInputElement&&!["text","search","url","tel","password"].includes(null==t?void 0:t.type)?n=o(t.value,e):(n=r()(t),c("copy")),n};function l(t){return(l="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(t){return typeof t}:function(t){return t&&"function"==typeof Symbol&&t.constructor===Symbol&&t!==Symbol.prototype?"symbol":typeof t})(t)}var s=function(){var t=0<arguments.length&&void 0!==arguments[0]?arguments[0]:{},e=t.action,n=void 0===e?"copy":e,o=t.container,e=t.target,t=t.text;if("copy"!==n&&"cut"!==n)throw new Error('Invalid "action" value, use either "copy" or "cut"');if(void 0!==e){if(!e||"object"!==l(e)||1!==e.nodeType)throw new Error('Invalid "target" value, use a valid Element');if("copy"===n&&e.hasAttribute("disabled"))throw new Error('Invalid "target" attribute. Please use "readonly" instead of "disabled" attribute');if("cut"===n&&(e.hasAttribute("readonly")||e.hasAttribute("disabled")))throw new Error('Invalid "target" attribute. You can\'t cut text from elements with "readonly" or "disabled" attributes')}return t?f(t,{container:o}):e?"cut"===n?a(e):f(e,{container:o}):void 0};function p(t){return(p="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(t){return typeof t}:function(t){return t&&"function"==typeof Symbol&&t.constructor===Symbol&&t!==Symbol.prototype?"symbol":typeof t})(t)}function d(t,e){for(var n=0;n<e.length;n++){var o=e[n];o.enumerable=o.enumerable||!1,o.configurable=!0,"value"in o&&(o.writable=!0),Object.defineProperty(t,o.key,o)}}function y(t,e){return(y=Object.setPrototypeOf||function(t,e){return t.__proto__=e,t})(t,e)}function h(n){var o=function(){if("undefined"==typeof Reflect||!Reflect.construct)return!1;if(Reflect.construct.sham)return!1;if("function"==typeof Proxy)return!0;try{return Date.prototype.toString.call(Reflect.construct(Date,[],function(){})),!0}catch(t){return!1}}();return function(){var t,e=v(n);return t=o?(t=v(this).constructor,Reflect.construct(e,arguments,t)):e.apply(this,arguments),e=this,!(t=t)||"object"!==p(t)&&"function"!=typeof t?function(t){if(void 0!==t)return t;throw new ReferenceError("this hasn't been initialised - super() hasn't been called")}(e):t}}function v(t){return(v=Object.setPrototypeOf?Object.getPrototypeOf:function(t){return t.__proto__||Object.getPrototypeOf(t)})(t)}function m(t,e){t="data-clipboard-".concat(t);if(e.hasAttribute(t))return e.getAttribute(t)}var b=function(){!function(t,e){if("function"!=typeof e&&null!==e)throw new TypeError("Super expression must either be null or a function");t.prototype=Object.create(e&&e.prototype,{constructor:{value:t,writable:!0,configurable:!0}}),e&&y(t,e)}(r,i());var t,e,n,o=h(r);function r(t,e){var n;return function(t){if(!(t instanceof r))throw new TypeError("Cannot call a class as a function")}(this),(n=o.call(this)).resolveOptions(e),n.listenClick(t),n}return t=r,n=[{key:"copy",value:function(t){var e=1<arguments.length&&void 0!==arguments[1]?arguments[1]:{container:document.body};return f(t,e)}},{key:"cut",value:function(t){return a(t)}},{key:"isSupported",value:function(){var t=0<arguments.length&&void 0!==arguments[0]?arguments[0]:["copy","cut"],t="string"==typeof t?[t]:t,e=!!document.queryCommandSupported;return t.forEach(function(t){e=e&&!!document.queryCommandSupported(t)}),e}}],(e=[{key:"resolveOptions",value:function(){var t=0<arguments.length&&void 0!==arguments[0]?arguments[0]:{};this.action="function"==typeof t.action?t.action:this.defaultAction,this.target="function"==typeof t.target?t.target:this.defaultTarget,this.text="function"==typeof t.text?t.text:this.defaultText,this.container="object"===p(t.container)?t.container:document.body}},{key:"listenClick",value:function(t){var e=this;this.listener=u()(t,"click",function(t){return e.onClick(t)})}},{key:"onClick",value:function(t){var e=t.delegateTarget||t.currentTarget,n=this.action(e)||"copy",t=s({action:n,container:this.container,target:this.target(e),text:this.text(e)});this.emit(t?"success":"error",{action:n,text:t,trigger:e,clearSelection:function(){e&&e.focus(),window.getSelection().removeAllRanges()}})}},{key:"defaultAction",value:function(t){return m("action",t)}},{key:"defaultTarget",value:function(t){t=m("target",t);if(t)return document.querySelector(t)}},{key:"defaultText",value:function(t){return m("text",t)}},{key:"destroy",value:function(){this.listener.destroy()}}])&&d(t.prototype,e),n&&d(t,n),r}()},828:function(t){var e;"undefined"==typeof Element||Element.prototype.matches||((e=Element.prototype).matches=e.matchesSelector||e.mozMatchesSelector||e.msMatchesSelector||e.oMatchesSelector||e.webkitMatchesSelector),t.exports=function(t,e){for(;t&&9!==t.nodeType;){if("function"==typeof t.matches&&t.matches(e))return t;t=t.parentNode}}},438:function(t,e,n){var u=n(828);function i(t,e,n,o,r){var i=function(e,n,t,o){return function(t){t.delegateTarget=u(t.target,n),t.delegateTarget&&o.call(e,t)}}.apply(this,arguments);return t.addEventListener(n,i,r),{destroy:function(){t.removeEventListener(n,i,r)}}}t.exports=function(t,e,n,o,r){return"function"==typeof t.addEventListener?i.apply(null,arguments):"function"==typeof n?i.bind(null,document).apply(null,arguments):("string"==typeof t&&(t=document.querySelectorAll(t)),Array.prototype.map.call(t,function(t){return i(t,e,n,o,r)}))}},879:function(t,n){n.node=function(t){return void 0!==t&&t instanceof HTMLElement&&1===t.nodeType},n.nodeList=function(t){var e=Object.prototype.toString.call(t);return void 0!==t&&("[object NodeList]"===e||"[object HTMLCollection]"===e)&&"length"in t&&(0===t.length||n.node(t[0]))},n.string=function(t){return"string"==typeof t||t instanceof String},n.fn=function(t){return"[object Function]"===Object.prototype.toString.call(t)}},370:function(t,e,n){var f=n(879),l=n(438);t.exports=function(t,e,n){if(!t&&!e&&!n)throw new Error("Missing required arguments");if(!f.string(e))throw new TypeError("Second argument must be a String");if(!f.fn(n))throw new TypeError("Third argument must be a Function");if(f.node(t))return c=e,a=n,(u=t).addEventListener(c,a),{destroy:function(){u.removeEventListener(c,a)}};if(f.nodeList(t))return o=t,r=e,i=n,Array.prototype.forEach.call(o,function(t){t.addEventListener(r,i)}),{destroy:function(){Array.prototype.forEach.call(o,function(t){t.removeEventListener(r,i)})}};if(f.string(t))return t=t,e=e,n=n,l(document.body,t,e,n);throw new TypeError("First argument must be a String, HTMLElement, HTMLCollection, or NodeList");var o,r,i,u,c,a}},817:function(t){t.exports=function(t){var e,n="SELECT"===t.nodeName?(t.focus(),t.value):"INPUT"===t.nodeName||"TEXTAREA"===t.nodeName?((e=t.hasAttribute("readonly"))||t.setAttribute("readonly",""),t.select(),t.setSelectionRange(0,t.value.length),e||t.removeAttribute("readonly"),t.value):(t.hasAttribute("contenteditable")&&t.focus(),n=window.getSelection(),(e=document.createRange()).selectNodeContents(t),n.removeAllRanges(),n.addRange(e),n.toString());return n}},279:function(t){function e(){}e.prototype={on:function(t,e,n){var o=this.e||(this.e={});return(o[t]||(o[t]=[])).push({fn:e,ctx:n}),this},once:function(t,e,n){var o=this;function r(){o.off(t,r),e.apply(n,arguments)}return r._=e,this.on(t,r,n)},emit:function(t){for(var e=[].slice.call(arguments,1),n=((this.e||(this.e={}))[t]||[]).slice(),o=0,r=n.length;o<r;o++)n[o].fn.apply(n[o].ctx,e);return this},off:function(t,e){var n=this.e||(this.e={}),o=n[t],r=[];if(o&&e)for(var i=0,u=o.length;i<u;i++)o[i].fn!==e&&o[i].fn._!==e&&r.push(o[i]);return r.length?n[t]=r:delete n[t],this}},t.exports=e,t.exports.TinyEmitter=e}},r={},o.n=function(t){var e=t&&t.__esModule?function(){return t.default}:function(){return t};return o.d(e,{a:e}),e},o.d=function(t,e){for(var n in e)o.o(e,n)&&!o.o(t,n)&&Object.defineProperty(t,n,{enumerable:!0,get:e[n]})},o.o=function(t,e){return Object.prototype.hasOwnProperty.call(t,e)},o(686).default;function o(t){if(r[t])return r[t].exports;var e=r[t]={exports:{}};return n[t](e,e.exports,o),e.exports}var n,r});
\ No newline at end of file
diff --git a/docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.css b/docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.css
new file mode 100644
index 000000000..3c9ff8775
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.css
@@ -0,0 +1 @@
+.glightbox-container{width:100%;height:100%;position:fixed;top:0;left:0;z-index:999999!important;overflow:hidden;-ms-touch-action:none;touch-action:none;-webkit-text-size-adjust:100%;-moz-text-size-adjust:100%;-ms-text-size-adjust:100%;text-size-adjust:100%;-webkit-backface-visibility:hidden;backface-visibility:hidden;outline:0}.glightbox-container.inactive{display:none}.glightbox-container .gcontainer{position:relative;width:100%;height:100%;z-index:9999;overflow:hidden}.glightbox-container .gslider{-webkit-transition:-webkit-transform .4s ease;transition:-webkit-transform .4s ease;transition:transform .4s ease;transition:transform .4s ease,-webkit-transform .4s ease;height:100%;left:0;top:0;width:100%;position:relative;overflow:hidden;display:-webkit-box!important;display:-ms-flexbox!important;display:flex!important;-webkit-box-pack:center;-ms-flex-pack:center;justify-content:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0)}.glightbox-container .gslide{width:100%;position:absolute;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;display:-webkit-box;display:-ms-flexbox;display:flex;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;justify-content:center;opacity:0}.glightbox-container .gslide.current{opacity:1;z-index:99999;position:relative}.glightbox-container .gslide.prev{opacity:1;z-index:9999}.glightbox-container .gslide-inner-content{width:100%}.glightbox-container .ginner-container{position:relative;width:100%;display:-webkit-box;display:-ms-flexbox;display:flex;-webkit-box-pack:center;-ms-flex-pack:center;justify-content:center;-webkit-box-orient:vertical;-webkit-box-direction:normal;-ms-flex-direction:column;flex-direction:column;max-width:100%;margin:auto;height:100vh}.glightbox-container .ginner-container.gvideo-container{width:100%}.glightbox-container .ginner-container.desc-bottom,.glightbox-container .ginner-container.desc-top{-webkit-box-orient:vertical;-webkit-box-direction:normal;-ms-flex-direction:column;flex-direction:column}.glightbox-container .ginner-container.desc-left,.glightbox-container .ginner-container.desc-right{max-width:100%!important}.gslide iframe,.gslide video{outline:0!important;border:none;min-height:165px;-webkit-overflow-scrolling:touch;-ms-touch-action:auto;touch-action:auto}.gslide:not(.current){pointer-events:none}.gslide-image{-webkit-box-align:center;-ms-flex-align:center;align-items:center}.gslide-image img{max-height:100vh;display:block;padding:0;float:none;outline:0;border:none;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;max-width:100vw;width:auto;height:auto;-o-object-fit:cover;object-fit:cover;-ms-touch-action:none;touch-action:none;margin:auto;min-width:200px}.desc-bottom .gslide-image img,.desc-top .gslide-image img{width:auto}.desc-left .gslide-image img,.desc-right .gslide-image img{width:auto;max-width:100%}.gslide-image img.zoomable{position:relative}.gslide-image img.dragging{cursor:-webkit-grabbing!important;cursor:grabbing!important;-webkit-transition:none;transition:none}.gslide-video{position:relative;max-width:100vh;width:100%!important}.gslide-video .plyr__poster-enabled.plyr--loading .plyr__poster{display:none}.gslide-video .gvideo-wrapper{width:100%;margin:auto}.gslide-video::before{content:'';position:absolute;width:100%;height:100%;background:rgba(255,0,0,.34);display:none}.gslide-video.playing::before{display:none}.gslide-video.fullscreen{max-width:100%!important;min-width:100%;height:75vh}.gslide-video.fullscreen video{max-width:100%!important;width:100%!important}.gslide-inline{background:#fff;text-align:left;max-height:calc(100vh - 40px);overflow:auto;max-width:100%;margin:auto}.gslide-inline .ginlined-content{padding:20px;width:100%}.gslide-inline .dragging{cursor:-webkit-grabbing!important;cursor:grabbing!important;-webkit-transition:none;transition:none}.ginlined-content{overflow:auto;display:block!important;opacity:1}.gslide-external{display:-webkit-box;display:-ms-flexbox;display:flex;width:100%;min-width:100%;background:#fff;padding:0;overflow:auto;max-height:75vh;height:100%}.gslide-media{display:-webkit-box;display:-ms-flexbox;display:flex;width:auto}.zoomed .gslide-media{-webkit-box-shadow:none!important;box-shadow:none!important}.desc-bottom .gslide-media,.desc-top .gslide-media{margin:0 auto;-webkit-box-orient:vertical;-webkit-box-direction:normal;-ms-flex-direction:column;flex-direction:column}.gslide-description{position:relative;-webkit-box-flex:1;-ms-flex:1 0 100%;flex:1 0 100%}.gslide-description.description-left,.gslide-description.description-right{max-width:100%}.gslide-description.description-bottom,.gslide-description.description-top{margin:0 auto;width:100%}.gslide-description p{margin-bottom:12px}.gslide-description p:last-child{margin-bottom:0}.zoomed .gslide-description{display:none}.glightbox-button-hidden{display:none}.glightbox-mobile .glightbox-container .gslide-description{height:auto!important;width:100%;position:absolute;bottom:0;padding:19px 11px;max-width:100vw!important;-webkit-box-ordinal-group:3!important;-ms-flex-order:2!important;order:2!important;max-height:78vh;overflow:auto!important;background:-webkit-gradient(linear,left top,left bottom,from(rgba(0,0,0,0)),to(rgba(0,0,0,.75)));background:linear-gradient(to bottom,rgba(0,0,0,0) 0,rgba(0,0,0,.75) 100%);-webkit-transition:opacity .3s linear;transition:opacity .3s linear;padding-bottom:50px}.glightbox-mobile .glightbox-container .gslide-title{color:#fff;font-size:1em}.glightbox-mobile .glightbox-container .gslide-desc{color:#a1a1a1}.glightbox-mobile .glightbox-container .gslide-desc a{color:#fff;font-weight:700}.glightbox-mobile .glightbox-container .gslide-desc *{color:inherit}.glightbox-mobile .glightbox-container .gslide-desc .desc-more{color:#fff;opacity:.4}.gdesc-open .gslide-media{-webkit-transition:opacity .5s ease;transition:opacity .5s ease;opacity:.4}.gdesc-open .gdesc-inner{padding-bottom:30px}.gdesc-closed .gslide-media{-webkit-transition:opacity .5s ease;transition:opacity .5s ease;opacity:1}.greset{-webkit-transition:all .3s ease;transition:all .3s ease}.gabsolute{position:absolute}.grelative{position:relative}.glightbox-desc{display:none!important}.glightbox-open{overflow:hidden}.gloader{height:25px;width:25px;-webkit-animation:lightboxLoader .8s infinite linear;animation:lightboxLoader .8s infinite linear;border:2px solid #fff;border-right-color:transparent;border-radius:50%;position:absolute;display:block;z-index:9999;left:0;right:0;margin:0 auto;top:47%}.goverlay{width:100%;height:calc(100vh + 1px);position:fixed;top:-1px;left:0;background:#000;will-change:opacity}.glightbox-mobile .goverlay{background:#000}.gclose,.gnext,.gprev{z-index:99999;cursor:pointer;width:26px;height:44px;border:none;display:-webkit-box;display:-ms-flexbox;display:flex;-webkit-box-pack:center;-ms-flex-pack:center;justify-content:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-orient:vertical;-webkit-box-direction:normal;-ms-flex-direction:column;flex-direction:column}.gclose svg,.gnext svg,.gprev svg{display:block;width:25px;height:auto;margin:0;padding:0}.gclose.disabled,.gnext.disabled,.gprev.disabled{opacity:.1}.gclose .garrow,.gnext .garrow,.gprev .garrow{stroke:#fff}.gbtn.focused{outline:2px solid #0f3d81}iframe.wait-autoplay{opacity:0}.glightbox-closing .gclose,.glightbox-closing .gnext,.glightbox-closing .gprev{opacity:0!important}.glightbox-clean .gslide-description{background:#fff}.glightbox-clean .gdesc-inner{padding:22px 20px}.glightbox-clean .gslide-title{font-size:1em;font-weight:400;font-family:arial;color:#000;margin-bottom:19px;line-height:1.4em}.glightbox-clean .gslide-desc{font-size:.86em;margin-bottom:0;font-family:arial;line-height:1.4em}.glightbox-clean .gslide-video{background:#000}.glightbox-clean .gclose,.glightbox-clean .gnext,.glightbox-clean .gprev{background-color:rgba(0,0,0,.75);border-radius:4px}.glightbox-clean .gclose path,.glightbox-clean .gnext path,.glightbox-clean .gprev path{fill:#fff}.glightbox-clean .gprev{position:absolute;top:-100%;left:30px;width:40px;height:50px}.glightbox-clean .gnext{position:absolute;top:-100%;right:30px;width:40px;height:50px}.glightbox-clean .gclose{width:35px;height:35px;top:15px;right:10px;position:absolute}.glightbox-clean .gclose svg{width:18px;height:auto}.glightbox-clean .gclose:hover{opacity:1}.gfadeIn{-webkit-animation:gfadeIn .5s ease;animation:gfadeIn .5s ease}.gfadeOut{-webkit-animation:gfadeOut .5s ease;animation:gfadeOut .5s ease}.gslideOutLeft{-webkit-animation:gslideOutLeft .3s ease;animation:gslideOutLeft .3s ease}.gslideInLeft{-webkit-animation:gslideInLeft .3s ease;animation:gslideInLeft .3s ease}.gslideOutRight{-webkit-animation:gslideOutRight .3s ease;animation:gslideOutRight .3s ease}.gslideInRight{-webkit-animation:gslideInRight .3s ease;animation:gslideInRight .3s ease}.gzoomIn{-webkit-animation:gzoomIn .5s ease;animation:gzoomIn .5s ease}.gzoomOut{-webkit-animation:gzoomOut .5s ease;animation:gzoomOut .5s ease}@-webkit-keyframes lightboxLoader{0%{-webkit-transform:rotate(0);transform:rotate(0)}100%{-webkit-transform:rotate(360deg);transform:rotate(360deg)}}@keyframes lightboxLoader{0%{-webkit-transform:rotate(0);transform:rotate(0)}100%{-webkit-transform:rotate(360deg);transform:rotate(360deg)}}@-webkit-keyframes gfadeIn{from{opacity:0}to{opacity:1}}@keyframes gfadeIn{from{opacity:0}to{opacity:1}}@-webkit-keyframes gfadeOut{from{opacity:1}to{opacity:0}}@keyframes gfadeOut{from{opacity:1}to{opacity:0}}@-webkit-keyframes gslideInLeft{from{opacity:0;-webkit-transform:translate3d(-60%,0,0);transform:translate3d(-60%,0,0)}to{visibility:visible;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0);opacity:1}}@keyframes gslideInLeft{from{opacity:0;-webkit-transform:translate3d(-60%,0,0);transform:translate3d(-60%,0,0)}to{visibility:visible;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0);opacity:1}}@-webkit-keyframes gslideOutLeft{from{opacity:1;visibility:visible;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0)}to{-webkit-transform:translate3d(-60%,0,0);transform:translate3d(-60%,0,0);opacity:0;visibility:hidden}}@keyframes gslideOutLeft{from{opacity:1;visibility:visible;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0)}to{-webkit-transform:translate3d(-60%,0,0);transform:translate3d(-60%,0,0);opacity:0;visibility:hidden}}@-webkit-keyframes gslideInRight{from{opacity:0;visibility:visible;-webkit-transform:translate3d(60%,0,0);transform:translate3d(60%,0,0)}to{-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0);opacity:1}}@keyframes gslideInRight{from{opacity:0;visibility:visible;-webkit-transform:translate3d(60%,0,0);transform:translate3d(60%,0,0)}to{-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0);opacity:1}}@-webkit-keyframes gslideOutRight{from{opacity:1;visibility:visible;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0)}to{-webkit-transform:translate3d(60%,0,0);transform:translate3d(60%,0,0);opacity:0}}@keyframes gslideOutRight{from{opacity:1;visibility:visible;-webkit-transform:translate3d(0,0,0);transform:translate3d(0,0,0)}to{-webkit-transform:translate3d(60%,0,0);transform:translate3d(60%,0,0);opacity:0}}@-webkit-keyframes gzoomIn{from{opacity:0;-webkit-transform:scale3d(.3,.3,.3);transform:scale3d(.3,.3,.3)}to{opacity:1}}@keyframes gzoomIn{from{opacity:0;-webkit-transform:scale3d(.3,.3,.3);transform:scale3d(.3,.3,.3)}to{opacity:1}}@-webkit-keyframes gzoomOut{from{opacity:1}50%{opacity:0;-webkit-transform:scale3d(.3,.3,.3);transform:scale3d(.3,.3,.3)}to{opacity:0}}@keyframes gzoomOut{from{opacity:1}50%{opacity:0;-webkit-transform:scale3d(.3,.3,.3);transform:scale3d(.3,.3,.3)}to{opacity:0}}@media (min-width:769px){.glightbox-container .ginner-container{width:auto;height:auto;-webkit-box-orient:horizontal;-webkit-box-direction:normal;-ms-flex-direction:row;flex-direction:row}.glightbox-container .ginner-container.desc-top .gslide-description{-webkit-box-ordinal-group:1;-ms-flex-order:0;order:0}.glightbox-container .ginner-container.desc-top .gslide-image,.glightbox-container .ginner-container.desc-top .gslide-image img{-webkit-box-ordinal-group:2;-ms-flex-order:1;order:1}.glightbox-container .ginner-container.desc-left .gslide-description{-webkit-box-ordinal-group:1;-ms-flex-order:0;order:0}.glightbox-container .ginner-container.desc-left .gslide-image{-webkit-box-ordinal-group:2;-ms-flex-order:1;order:1}.gslide-image img{max-height:97vh;max-width:100%}.gslide-image img.zoomable{cursor:-webkit-zoom-in;cursor:zoom-in}.zoomed .gslide-image img.zoomable{cursor:-webkit-grab;cursor:grab}.gslide-inline{max-height:95vh}.gslide-external{max-height:100vh}.gslide-description.description-left,.gslide-description.description-right{max-width:275px}.glightbox-open{height:auto}.goverlay{background:rgba(0,0,0,.92)}.glightbox-clean .gslide-media{-webkit-box-shadow:1px 2px 9px 0 rgba(0,0,0,.65);box-shadow:1px 2px 9px 0 rgba(0,0,0,.65)}.glightbox-clean .description-left .gdesc-inner,.glightbox-clean .description-right .gdesc-inner{position:absolute;height:100%;overflow-y:auto}.glightbox-clean .gclose,.glightbox-clean .gnext,.glightbox-clean .gprev{background-color:rgba(0,0,0,.32)}.glightbox-clean .gclose:hover,.glightbox-clean .gnext:hover,.glightbox-clean .gprev:hover{background-color:rgba(0,0,0,.7)}.glightbox-clean .gprev{top:45%}.glightbox-clean .gnext{top:45%}}@media (min-width:992px){.glightbox-clean .gclose{opacity:.7;right:20px}}@media screen and (max-height:420px){.goverlay{background:#000}}
\ No newline at end of file
diff --git a/docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.js b/docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.js
new file mode 100644
index 000000000..997908b59
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-contrib/glightbox/glightbox.min.js
@@ -0,0 +1 @@
+!function(e,t){"object"==typeof exports&&"undefined"!=typeof module?module.exports=t():"function"==typeof define&&define.amd?define(t):(e=e||self).GLightbox=t()}(this,(function(){"use strict";function e(t){return(e="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e})(t)}function t(e,t){if(!(e instanceof t))throw new TypeError("Cannot call a class as a function")}function i(e,t){for(var i=0;i<t.length;i++){var n=t[i];n.enumerable=n.enumerable||!1,n.configurable=!0,"value"in n&&(n.writable=!0),Object.defineProperty(e,n.key,n)}}function n(e,t,n){return t&&i(e.prototype,t),n&&i(e,n),e}var s=Date.now();function l(){var e={},t=!0,i=0,n=arguments.length;"[object Boolean]"===Object.prototype.toString.call(arguments[0])&&(t=arguments[0],i++);for(var s=function(i){for(var n in i)Object.prototype.hasOwnProperty.call(i,n)&&(t&&"[object Object]"===Object.prototype.toString.call(i[n])?e[n]=l(!0,e[n],i[n]):e[n]=i[n])};i<n;i++){var o=arguments[i];s(o)}return e}function o(e,t){if((k(e)||e===window||e===document)&&(e=[e]),A(e)||L(e)||(e=[e]),0!=P(e))if(A(e)&&!L(e))for(var i=e.length,n=0;n<i&&!1!==t.call(e[n],e[n],n,e);n++);else if(L(e))for(var s in e)if(O(e,s)&&!1===t.call(e[s],e[s],s,e))break}function r(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:null,i=arguments.length>2&&void 0!==arguments[2]?arguments[2]:null,n=e[s]=e[s]||[],l={all:n,evt:null,found:null};return t&&i&&P(n)>0&&o(n,(function(e,n){if(e.eventName==t&&e.fn.toString()==i.toString())return l.found=!0,l.evt=n,!1})),l}function a(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},i=t.onElement,n=t.withCallback,s=t.avoidDuplicate,l=void 0===s||s,a=t.once,h=void 0!==a&&a,d=t.useCapture,c=void 0!==d&&d,u=arguments.length>2?arguments[2]:void 0,g=i||[];function v(e){T(n)&&n.call(u,e,this),h&&v.destroy()}return C(g)&&(g=document.querySelectorAll(g)),v.destroy=function(){o(g,(function(t){var i=r(t,e,v);i.found&&i.all.splice(i.evt,1),t.removeEventListener&&t.removeEventListener(e,v,c)}))},o(g,(function(t){var i=r(t,e,v);(t.addEventListener&&l&&!i.found||!l)&&(t.addEventListener(e,v,c),i.all.push({eventName:e,fn:v}))})),v}function h(e,t){o(t.split(" "),(function(t){return e.classList.add(t)}))}function d(e,t){o(t.split(" "),(function(t){return e.classList.remove(t)}))}function c(e,t){return e.classList.contains(t)}function u(e,t){for(;e!==document.body;){if(!(e=e.parentElement))return!1;if("function"==typeof e.matches?e.matches(t):e.msMatchesSelector(t))return e}}function g(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:"",i=arguments.length>2&&void 0!==arguments[2]&&arguments[2];if(!e||""===t)return!1;if("none"===t)return T(i)&&i(),!1;var n=x(),s=t.split(" ");o(s,(function(t){h(e,"g"+t)})),a(n,{onElement:e,avoidDuplicate:!1,once:!0,withCallback:function(e,t){o(s,(function(e){d(t,"g"+e)})),T(i)&&i()}})}function v(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:"";if(""===t)return e.style.webkitTransform="",e.style.MozTransform="",e.style.msTransform="",e.style.OTransform="",e.style.transform="",!1;e.style.webkitTransform=t,e.style.MozTransform=t,e.style.msTransform=t,e.style.OTransform=t,e.style.transform=t}function f(e){e.style.display="block"}function p(e){e.style.display="none"}function m(e){var t=document.createDocumentFragment(),i=document.createElement("div");for(i.innerHTML=e;i.firstChild;)t.appendChild(i.firstChild);return t}function y(){return{width:window.innerWidth||document.documentElement.clientWidth||document.body.clientWidth,height:window.innerHeight||document.documentElement.clientHeight||document.body.clientHeight}}function x(){var e,t=document.createElement("fakeelement"),i={animation:"animationend",OAnimation:"oAnimationEnd",MozAnimation:"animationend",WebkitAnimation:"webkitAnimationEnd"};for(e in i)if(void 0!==t.style[e])return i[e]}function b(e,t,i,n){if(e())t();else{var s;i||(i=100);var l=setInterval((function(){e()&&(clearInterval(l),s&&clearTimeout(s),t())}),i);n&&(s=setTimeout((function(){clearInterval(l)}),n))}}function S(e,t,i){if(I(e))console.error("Inject assets error");else if(T(t)&&(i=t,t=!1),C(t)&&t in window)T(i)&&i();else{var n;if(-1!==e.indexOf(".css")){if((n=document.querySelectorAll('link[href="'+e+'"]'))&&n.length>0)return void(T(i)&&i());var s=document.getElementsByTagName("head")[0],l=s.querySelectorAll('link[rel="stylesheet"]'),o=document.createElement("link");return o.rel="stylesheet",o.type="text/css",o.href=e,o.media="all",l?s.insertBefore(o,l[0]):s.appendChild(o),void(T(i)&&i())}if((n=document.querySelectorAll('script[src="'+e+'"]'))&&n.length>0){if(T(i)){if(C(t))return b((function(){return void 0!==window[t]}),(function(){i()})),!1;i()}}else{var r=document.createElement("script");r.type="text/javascript",r.src=e,r.onload=function(){if(T(i)){if(C(t))return b((function(){return void 0!==window[t]}),(function(){i()})),!1;i()}},document.body.appendChild(r)}}}function w(){return"navigator"in window&&window.navigator.userAgent.match(/(iPad)|(iPhone)|(iPod)|(Android)|(PlayBook)|(BB10)|(BlackBerry)|(Opera Mini)|(IEMobile)|(webOS)|(MeeGo)/i)}function T(e){return"function"==typeof e}function C(e){return"string"==typeof e}function k(e){return!(!e||!e.nodeType||1!=e.nodeType)}function E(e){return Array.isArray(e)}function A(e){return e&&e.length&&isFinite(e.length)}function L(t){return"object"===e(t)&&null!=t&&!T(t)&&!E(t)}function I(e){return null==e}function O(e,t){return null!==e&&hasOwnProperty.call(e,t)}function P(e){if(L(e)){if(e.keys)return e.keys().length;var t=0;for(var i in e)O(e,i)&&t++;return t}return e.length}function M(e){return!isNaN(parseFloat(e))&&isFinite(e)}function z(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:-1,t=document.querySelectorAll(".gbtn[data-taborder]:not(.disabled)");if(!t.length)return!1;if(1==t.length)return t[0];"string"==typeof e&&(e=parseInt(e));var i=[];o(t,(function(e){i.push(e.getAttribute("data-taborder"))}));var n=Math.max.apply(Math,i.map((function(e){return parseInt(e)}))),s=e<0?1:e+1;s>n&&(s="1");var l=i.filter((function(e){return e>=parseInt(s)})),r=l.sort()[0];return document.querySelector('.gbtn[data-taborder="'.concat(r,'"]'))}function X(e){if(e.events.hasOwnProperty("keyboard"))return!1;e.events.keyboard=a("keydown",{onElement:window,withCallback:function(t,i){var n=(t=t||window.event).keyCode;if(9==n){var s=document.querySelector(".gbtn.focused");if(!s){var l=!(!document.activeElement||!document.activeElement.nodeName)&&document.activeElement.nodeName.toLocaleLowerCase();if("input"==l||"textarea"==l||"button"==l)return}t.preventDefault();var o=document.querySelectorAll(".gbtn[data-taborder]");if(!o||o.length<=0)return;if(!s){var r=z();return void(r&&(r.focus(),h(r,"focused")))}var a=z(s.getAttribute("data-taborder"));d(s,"focused"),a&&(a.focus(),h(a,"focused"))}39==n&&e.nextSlide(),37==n&&e.prevSlide(),27==n&&e.close()}})}function Y(e){return Math.sqrt(e.x*e.x+e.y*e.y)}function q(e,t){var i=function(e,t){var i=Y(e)*Y(t);if(0===i)return 0;var n=function(e,t){return e.x*t.x+e.y*t.y}(e,t)/i;return n>1&&(n=1),Math.acos(n)}(e,t);return function(e,t){return e.x*t.y-t.x*e.y}(e,t)>0&&(i*=-1),180*i/Math.PI}var N=function(){function e(i){t(this,e),this.handlers=[],this.el=i}return n(e,[{key:"add",value:function(e){this.handlers.push(e)}},{key:"del",value:function(e){e||(this.handlers=[]);for(var t=this.handlers.length;t>=0;t--)this.handlers[t]===e&&this.handlers.splice(t,1)}},{key:"dispatch",value:function(){for(var e=0,t=this.handlers.length;e<t;e++){var i=this.handlers[e];"function"==typeof i&&i.apply(this.el,arguments)}}}]),e}();function D(e,t){var i=new N(e);return i.add(t),i}var _=function(){function e(i,n){t(this,e),this.element="string"==typeof i?document.querySelector(i):i,this.start=this.start.bind(this),this.move=this.move.bind(this),this.end=this.end.bind(this),this.cancel=this.cancel.bind(this),this.element.addEventListener("touchstart",this.start,!1),this.element.addEventListener("touchmove",this.move,!1),this.element.addEventListener("touchend",this.end,!1),this.element.addEventListener("touchcancel",this.cancel,!1),this.preV={x:null,y:null},this.pinchStartLen=null,this.zoom=1,this.isDoubleTap=!1;var s=function(){};this.rotate=D(this.element,n.rotate||s),this.touchStart=D(this.element,n.touchStart||s),this.multipointStart=D(this.element,n.multipointStart||s),this.multipointEnd=D(this.element,n.multipointEnd||s),this.pinch=D(this.element,n.pinch||s),this.swipe=D(this.element,n.swipe||s),this.tap=D(this.element,n.tap||s),this.doubleTap=D(this.element,n.doubleTap||s),this.longTap=D(this.element,n.longTap||s),this.singleTap=D(this.element,n.singleTap||s),this.pressMove=D(this.element,n.pressMove||s),this.twoFingerPressMove=D(this.element,n.twoFingerPressMove||s),this.touchMove=D(this.element,n.touchMove||s),this.touchEnd=D(this.element,n.touchEnd||s),this.touchCancel=D(this.element,n.touchCancel||s),this.translateContainer=this.element,this._cancelAllHandler=this.cancelAll.bind(this),window.addEventListener("scroll",this._cancelAllHandler),this.delta=null,this.last=null,this.now=null,this.tapTimeout=null,this.singleTapTimeout=null,this.longTapTimeout=null,this.swipeTimeout=null,this.x1=this.x2=this.y1=this.y2=null,this.preTapPosition={x:null,y:null}}return n(e,[{key:"start",value:function(e){if(e.touches){if(e.target&&e.target.nodeName&&["a","button","input"].indexOf(e.target.nodeName.toLowerCase())>=0)console.log("ignore drag for this touched element",e.target.nodeName.toLowerCase());else{this.now=Date.now(),this.x1=e.touches[0].pageX,this.y1=e.touches[0].pageY,this.delta=this.now-(this.last||this.now),this.touchStart.dispatch(e,this.element),null!==this.preTapPosition.x&&(this.isDoubleTap=this.delta>0&&this.delta<=250&&Math.abs(this.preTapPosition.x-this.x1)<30&&Math.abs(this.preTapPosition.y-this.y1)<30,this.isDoubleTap&&clearTimeout(this.singleTapTimeout)),this.preTapPosition.x=this.x1,this.preTapPosition.y=this.y1,this.last=this.now;var t=this.preV;if(e.touches.length>1){this._cancelLongTap(),this._cancelSingleTap();var i={x:e.touches[1].pageX-this.x1,y:e.touches[1].pageY-this.y1};t.x=i.x,t.y=i.y,this.pinchStartLen=Y(t),this.multipointStart.dispatch(e,this.element)}this._preventTap=!1,this.longTapTimeout=setTimeout(function(){this.longTap.dispatch(e,this.element),this._preventTap=!0}.bind(this),750)}}}},{key:"move",value:function(e){if(e.touches){var t=this.preV,i=e.touches.length,n=e.touches[0].pageX,s=e.touches[0].pageY;if(this.isDoubleTap=!1,i>1){var l=e.touches[1].pageX,o=e.touches[1].pageY,r={x:e.touches[1].pageX-n,y:e.touches[1].pageY-s};null!==t.x&&(this.pinchStartLen>0&&(e.zoom=Y(r)/this.pinchStartLen,this.pinch.dispatch(e,this.element)),e.angle=q(r,t),this.rotate.dispatch(e,this.element)),t.x=r.x,t.y=r.y,null!==this.x2&&null!==this.sx2?(e.deltaX=(n-this.x2+l-this.sx2)/2,e.deltaY=(s-this.y2+o-this.sy2)/2):(e.deltaX=0,e.deltaY=0),this.twoFingerPressMove.dispatch(e,this.element),this.sx2=l,this.sy2=o}else{if(null!==this.x2){e.deltaX=n-this.x2,e.deltaY=s-this.y2;var a=Math.abs(this.x1-this.x2),h=Math.abs(this.y1-this.y2);(a>10||h>10)&&(this._preventTap=!0)}else e.deltaX=0,e.deltaY=0;this.pressMove.dispatch(e,this.element)}this.touchMove.dispatch(e,this.element),this._cancelLongTap(),this.x2=n,this.y2=s,i>1&&e.preventDefault()}}},{key:"end",value:function(e){if(e.changedTouches){this._cancelLongTap();var t=this;e.touches.length<2&&(this.multipointEnd.dispatch(e,this.element),this.sx2=this.sy2=null),this.x2&&Math.abs(this.x1-this.x2)>30||this.y2&&Math.abs(this.y1-this.y2)>30?(e.direction=this._swipeDirection(this.x1,this.x2,this.y1,this.y2),this.swipeTimeout=setTimeout((function(){t.swipe.dispatch(e,t.element)}),0)):(this.tapTimeout=setTimeout((function(){t._preventTap||t.tap.dispatch(e,t.element),t.isDoubleTap&&(t.doubleTap.dispatch(e,t.element),t.isDoubleTap=!1)}),0),t.isDoubleTap||(t.singleTapTimeout=setTimeout((function(){t.singleTap.dispatch(e,t.element)}),250))),this.touchEnd.dispatch(e,this.element),this.preV.x=0,this.preV.y=0,this.zoom=1,this.pinchStartLen=null,this.x1=this.x2=this.y1=this.y2=null}}},{key:"cancelAll",value:function(){this._preventTap=!0,clearTimeout(this.singleTapTimeout),clearTimeout(this.tapTimeout),clearTimeout(this.longTapTimeout),clearTimeout(this.swipeTimeout)}},{key:"cancel",value:function(e){this.cancelAll(),this.touchCancel.dispatch(e,this.element)}},{key:"_cancelLongTap",value:function(){clearTimeout(this.longTapTimeout)}},{key:"_cancelSingleTap",value:function(){clearTimeout(this.singleTapTimeout)}},{key:"_swipeDirection",value:function(e,t,i,n){return Math.abs(e-t)>=Math.abs(i-n)?e-t>0?"Left":"Right":i-n>0?"Up":"Down"}},{key:"on",value:function(e,t){this[e]&&this[e].add(t)}},{key:"off",value:function(e,t){this[e]&&this[e].del(t)}},{key:"destroy",value:function(){return this.singleTapTimeout&&clearTimeout(this.singleTapTimeout),this.tapTimeout&&clearTimeout(this.tapTimeout),this.longTapTimeout&&clearTimeout(this.longTapTimeout),this.swipeTimeout&&clearTimeout(this.swipeTimeout),this.element.removeEventListener("touchstart",this.start),this.element.removeEventListener("touchmove",this.move),this.element.removeEventListener("touchend",this.end),this.element.removeEventListener("touchcancel",this.cancel),this.rotate.del(),this.touchStart.del(),this.multipointStart.del(),this.multipointEnd.del(),this.pinch.del(),this.swipe.del(),this.tap.del(),this.doubleTap.del(),this.longTap.del(),this.singleTap.del(),this.pressMove.del(),this.twoFingerPressMove.del(),this.touchMove.del(),this.touchEnd.del(),this.touchCancel.del(),this.preV=this.pinchStartLen=this.zoom=this.isDoubleTap=this.delta=this.last=this.now=this.tapTimeout=this.singleTapTimeout=this.longTapTimeout=this.swipeTimeout=this.x1=this.x2=this.y1=this.y2=this.preTapPosition=this.rotate=this.touchStart=this.multipointStart=this.multipointEnd=this.pinch=this.swipe=this.tap=this.doubleTap=this.longTap=this.singleTap=this.pressMove=this.touchMove=this.touchEnd=this.touchCancel=this.twoFingerPressMove=null,window.removeEventListener("scroll",this._cancelAllHandler),null}}]),e}();function W(e){var t=function(){var e,t=document.createElement("fakeelement"),i={transition:"transitionend",OTransition:"oTransitionEnd",MozTransition:"transitionend",WebkitTransition:"webkitTransitionEnd"};for(e in i)if(void 0!==t.style[e])return i[e]}(),i=window.innerWidth||document.documentElement.clientWidth||document.body.clientWidth,n=c(e,"gslide-media")?e:e.querySelector(".gslide-media"),s=u(n,".ginner-container"),l=e.querySelector(".gslide-description");i>769&&(n=s),h(n,"greset"),v(n,"translate3d(0, 0, 0)"),a(t,{onElement:n,once:!0,withCallback:function(e,t){d(n,"greset")}}),n.style.opacity="",l&&(l.style.opacity="")}function B(e){if(e.events.hasOwnProperty("touch"))return!1;var t,i,n,s=y(),l=s.width,o=s.height,r=!1,a=null,g=null,f=null,p=!1,m=1,x=1,b=!1,S=!1,w=null,T=null,C=null,k=null,E=0,A=0,L=!1,I=!1,O={},P={},M=0,z=0,X=document.getElementById("glightbox-slider"),Y=document.querySelector(".goverlay"),q=new _(X,{touchStart:function(t){if(r=!0,(c(t.targetTouches[0].target,"ginner-container")||u(t.targetTouches[0].target,".gslide-desc")||"a"==t.targetTouches[0].target.nodeName.toLowerCase())&&(r=!1),u(t.targetTouches[0].target,".gslide-inline")&&!c(t.targetTouches[0].target.parentNode,"gslide-inline")&&(r=!1),r){if(P=t.targetTouches[0],O.pageX=t.targetTouches[0].pageX,O.pageY=t.targetTouches[0].pageY,M=t.targetTouches[0].clientX,z=t.targetTouches[0].clientY,a=e.activeSlide,g=a.querySelector(".gslide-media"),n=a.querySelector(".gslide-inline"),f=null,c(g,"gslide-image")&&(f=g.querySelector("img")),(window.innerWidth||document.documentElement.clientWidth||document.body.clientWidth)>769&&(g=a.querySelector(".ginner-container")),d(Y,"greset"),t.pageX>20&&t.pageX<window.innerWidth-20)return;t.preventDefault()}},touchMove:function(s){if(r&&(P=s.targetTouches[0],!b&&!S)){if(n&&n.offsetHeight>o){var a=O.pageX-P.pageX;if(Math.abs(a)<=13)return!1}p=!0;var h,d=s.targetTouches[0].clientX,c=s.targetTouches[0].clientY,u=M-d,m=z-c;if(Math.abs(u)>Math.abs(m)?(L=!1,I=!0):(I=!1,L=!0),t=P.pageX-O.pageX,E=100*t/l,i=P.pageY-O.pageY,A=100*i/o,L&&f&&(h=1-Math.abs(i)/o,Y.style.opacity=h,e.settings.touchFollowAxis&&(E=0)),I&&(h=1-Math.abs(t)/l,g.style.opacity=h,e.settings.touchFollowAxis&&(A=0)),!f)return v(g,"translate3d(".concat(E,"%, 0, 0)"));v(g,"translate3d(".concat(E,"%, ").concat(A,"%, 0)"))}},touchEnd:function(){if(r){if(p=!1,S||b)return C=w,void(k=T);var t=Math.abs(parseInt(A)),i=Math.abs(parseInt(E));if(!(t>29&&f))return t<29&&i<25?(h(Y,"greset"),Y.style.opacity=1,W(g)):void 0;e.close()}},multipointEnd:function(){setTimeout((function(){b=!1}),50)},multipointStart:function(){b=!0,m=x||1},pinch:function(e){if(!f||p)return!1;b=!0,f.scaleX=f.scaleY=m*e.zoom;var t=m*e.zoom;if(S=!0,t<=1)return S=!1,t=1,k=null,C=null,w=null,T=null,void f.setAttribute("style","");t>4.5&&(t=4.5),f.style.transform="scale3d(".concat(t,", ").concat(t,", 1)"),x=t},pressMove:function(e){if(S&&!b){var t=P.pageX-O.pageX,i=P.pageY-O.pageY;C&&(t+=C),k&&(i+=k),w=t,T=i;var n="translate3d(".concat(t,"px, ").concat(i,"px, 0)");x&&(n+=" scale3d(".concat(x,", ").concat(x,", 1)")),v(f,n)}},swipe:function(t){if(!S)if(b)b=!1;else{if("Left"==t.direction){if(e.index==e.elements.length-1)return W(g);e.nextSlide()}if("Right"==t.direction){if(0==e.index)return W(g);e.prevSlide()}}}});e.events.touch=q}var H=function(){function e(i,n){var s=this,l=arguments.length>2&&void 0!==arguments[2]?arguments[2]:null;if(t(this,e),this.img=i,this.slide=n,this.onclose=l,this.img.setZoomEvents)return!1;this.active=!1,this.zoomedIn=!1,this.dragging=!1,this.currentX=null,this.currentY=null,this.initialX=null,this.initialY=null,this.xOffset=0,this.yOffset=0,this.img.addEventListener("mousedown",(function(e){return s.dragStart(e)}),!1),this.img.addEventListener("mouseup",(function(e){return s.dragEnd(e)}),!1),this.img.addEventListener("mousemove",(function(e){return s.drag(e)}),!1),this.img.addEventListener("click",(function(e){return s.slide.classList.contains("dragging-nav")?(s.zoomOut(),!1):s.zoomedIn?void(s.zoomedIn&&!s.dragging&&s.zoomOut()):s.zoomIn()}),!1),this.img.setZoomEvents=!0}return n(e,[{key:"zoomIn",value:function(){var e=this.widowWidth();if(!(this.zoomedIn||e<=768)){var t=this.img;if(t.setAttribute("data-style",t.getAttribute("style")),t.style.maxWidth=t.naturalWidth+"px",t.style.maxHeight=t.naturalHeight+"px",t.naturalWidth>e){var i=e/2-t.naturalWidth/2;this.setTranslate(this.img.parentNode,i,0)}this.slide.classList.add("zoomed"),this.zoomedIn=!0}}},{key:"zoomOut",value:function(){this.img.parentNode.setAttribute("style",""),this.img.setAttribute("style",this.img.getAttribute("data-style")),this.slide.classList.remove("zoomed"),this.zoomedIn=!1,this.currentX=null,this.currentY=null,this.initialX=null,this.initialY=null,this.xOffset=0,this.yOffset=0,this.onclose&&"function"==typeof this.onclose&&this.onclose()}},{key:"dragStart",value:function(e){e.preventDefault(),this.zoomedIn?("touchstart"===e.type?(this.initialX=e.touches[0].clientX-this.xOffset,this.initialY=e.touches[0].clientY-this.yOffset):(this.initialX=e.clientX-this.xOffset,this.initialY=e.clientY-this.yOffset),e.target===this.img&&(this.active=!0,this.img.classList.add("dragging"))):this.active=!1}},{key:"dragEnd",value:function(e){var t=this;e.preventDefault(),this.initialX=this.currentX,this.initialY=this.currentY,this.active=!1,setTimeout((function(){t.dragging=!1,t.img.isDragging=!1,t.img.classList.remove("dragging")}),100)}},{key:"drag",value:function(e){this.active&&(e.preventDefault(),"touchmove"===e.type?(this.currentX=e.touches[0].clientX-this.initialX,this.currentY=e.touches[0].clientY-this.initialY):(this.currentX=e.clientX-this.initialX,this.currentY=e.clientY-this.initialY),this.xOffset=this.currentX,this.yOffset=this.currentY,this.img.isDragging=!0,this.dragging=!0,this.setTranslate(this.img,this.currentX,this.currentY))}},{key:"onMove",value:function(e){if(this.zoomedIn){var t=e.clientX-this.img.naturalWidth/2,i=e.clientY-this.img.naturalHeight/2;this.setTranslate(this.img,t,i)}}},{key:"setTranslate",value:function(e,t,i){e.style.transform="translate3d("+t+"px, "+i+"px, 0)"}},{key:"widowWidth",value:function(){return window.innerWidth||document.documentElement.clientWidth||document.body.clientWidth}}]),e}(),V=function(){function e(){var i=this,n=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{};t(this,e);var s=n.dragEl,l=n.toleranceX,o=void 0===l?40:l,r=n.toleranceY,a=void 0===r?65:r,h=n.slide,d=void 0===h?null:h,c=n.instance,u=void 0===c?null:c;this.el=s,this.active=!1,this.dragging=!1,this.currentX=null,this.currentY=null,this.initialX=null,this.initialY=null,this.xOffset=0,this.yOffset=0,this.direction=null,this.lastDirection=null,this.toleranceX=o,this.toleranceY=a,this.toleranceReached=!1,this.dragContainer=this.el,this.slide=d,this.instance=u,this.el.addEventListener("mousedown",(function(e){return i.dragStart(e)}),!1),this.el.addEventListener("mouseup",(function(e){return i.dragEnd(e)}),!1),this.el.addEventListener("mousemove",(function(e){return i.drag(e)}),!1)}return n(e,[{key:"dragStart",value:function(e){if(this.slide.classList.contains("zoomed"))this.active=!1;else{"touchstart"===e.type?(this.initialX=e.touches[0].clientX-this.xOffset,this.initialY=e.touches[0].clientY-this.yOffset):(this.initialX=e.clientX-this.xOffset,this.initialY=e.clientY-this.yOffset);var t=e.target.nodeName.toLowerCase();e.target.classList.contains("nodrag")||u(e.target,".nodrag")||-1!==["input","select","textarea","button","a"].indexOf(t)?this.active=!1:(e.preventDefault(),(e.target===this.el||"img"!==t&&u(e.target,".gslide-inline"))&&(this.active=!0,this.el.classList.add("dragging"),this.dragContainer=u(e.target,".ginner-container")))}}},{key:"dragEnd",value:function(e){var t=this;e&&e.preventDefault(),this.initialX=0,this.initialY=0,this.currentX=null,this.currentY=null,this.initialX=null,this.initialY=null,this.xOffset=0,this.yOffset=0,this.active=!1,this.doSlideChange&&(this.instance.preventOutsideClick=!0,"right"==this.doSlideChange&&this.instance.prevSlide(),"left"==this.doSlideChange&&this.instance.nextSlide()),this.doSlideClose&&this.instance.close(),this.toleranceReached||this.setTranslate(this.dragContainer,0,0,!0),setTimeout((function(){t.instance.preventOutsideClick=!1,t.toleranceReached=!1,t.lastDirection=null,t.dragging=!1,t.el.isDragging=!1,t.el.classList.remove("dragging"),t.slide.classList.remove("dragging-nav"),t.dragContainer.style.transform="",t.dragContainer.style.transition=""}),100)}},{key:"drag",value:function(e){if(this.active){e.preventDefault(),this.slide.classList.add("dragging-nav"),"touchmove"===e.type?(this.currentX=e.touches[0].clientX-this.initialX,this.currentY=e.touches[0].clientY-this.initialY):(this.currentX=e.clientX-this.initialX,this.currentY=e.clientY-this.initialY),this.xOffset=this.currentX,this.yOffset=this.currentY,this.el.isDragging=!0,this.dragging=!0,this.doSlideChange=!1,this.doSlideClose=!1;var t=Math.abs(this.currentX),i=Math.abs(this.currentY);if(t>0&&t>=Math.abs(this.currentY)&&(!this.lastDirection||"x"==this.lastDirection)){this.yOffset=0,this.lastDirection="x",this.setTranslate(this.dragContainer,this.currentX,0);var n=this.shouldChange();if(!this.instance.settings.dragAutoSnap&&n&&(this.doSlideChange=n),this.instance.settings.dragAutoSnap&&n)return this.instance.preventOutsideClick=!0,this.toleranceReached=!0,this.active=!1,this.instance.preventOutsideClick=!0,this.dragEnd(null),"right"==n&&this.instance.prevSlide(),void("left"==n&&this.instance.nextSlide())}if(this.toleranceY>0&&i>0&&i>=t&&(!this.lastDirection||"y"==this.lastDirection)){this.xOffset=0,this.lastDirection="y",this.setTranslate(this.dragContainer,0,this.currentY);var s=this.shouldClose();return!this.instance.settings.dragAutoSnap&&s&&(this.doSlideClose=!0),void(this.instance.settings.dragAutoSnap&&s&&this.instance.close())}}}},{key:"shouldChange",value:function(){var e=!1;if(Math.abs(this.currentX)>=this.toleranceX){var t=this.currentX>0?"right":"left";("left"==t&&this.slide!==this.slide.parentNode.lastChild||"right"==t&&this.slide!==this.slide.parentNode.firstChild)&&(e=t)}return e}},{key:"shouldClose",value:function(){var e=!1;return Math.abs(this.currentY)>=this.toleranceY&&(e=!0),e}},{key:"setTranslate",value:function(e,t,i){var n=arguments.length>3&&void 0!==arguments[3]&&arguments[3];e.style.transition=n?"all .2s ease":"",e.style.transform="translate3d(".concat(t,"px, ").concat(i,"px, 0)")}}]),e}();function j(e,t,i,n){var s=e.querySelector(".gslide-media"),l=new Image,o="gSlideTitle_"+i,r="gSlideDesc_"+i;l.addEventListener("load",(function(){T(n)&&n()}),!1),l.src=t.href,""!=t.sizes&&""!=t.srcset&&(l.sizes=t.sizes,l.srcset=t.srcset),l.alt="",I(t.alt)||""===t.alt||(l.alt=t.alt),""!==t.title&&l.setAttribute("aria-labelledby",o),""!==t.description&&l.setAttribute("aria-describedby",r),t.hasOwnProperty("_hasCustomWidth")&&t._hasCustomWidth&&(l.style.width=t.width),t.hasOwnProperty("_hasCustomHeight")&&t._hasCustomHeight&&(l.style.height=t.height),s.insertBefore(l,s.firstChild)}function F(e,t,i,n){var s=this,l=e.querySelector(".ginner-container"),o="gvideo"+i,r=e.querySelector(".gslide-media"),a=this.getAllPlayers();h(l,"gvideo-container"),r.insertBefore(m('<div class="gvideo-wrapper"></div>'),r.firstChild);var d=e.querySelector(".gvideo-wrapper");S(this.settings.plyr.css,"Plyr");var c=t.href,u=null==t?void 0:t.videoProvider,g=!1;r.style.maxWidth=t.width,S(this.settings.plyr.js,"Plyr",(function(){if(!u&&c.match(/vimeo\.com\/([0-9]*)/)&&(u="vimeo"),!u&&(c.match(/(youtube\.com|youtube-nocookie\.com)\/watch\?v=([a-zA-Z0-9\-_]+)/)||c.match(/youtu\.be\/([a-zA-Z0-9\-_]+)/)||c.match(/(youtube\.com|youtube-nocookie\.com)\/embed\/([a-zA-Z0-9\-_]+)/))&&(u="youtube"),"local"===u||!u){u="local";var l='<video id="'+o+'" ';l+='style="background:#000; max-width: '.concat(t.width,';" '),l+='preload="metadata" ',l+='x-webkit-airplay="allow" ',l+="playsinline ",l+="controls ",l+='class="gvideo-local">',l+='<source src="'.concat(c,'">'),g=m(l+="</video>")}var r=g||m('<div id="'.concat(o,'" data-plyr-provider="').concat(u,'" data-plyr-embed-id="').concat(c,'"></div>'));h(d,"".concat(u,"-video gvideo")),d.appendChild(r),d.setAttribute("data-id",o),d.setAttribute("data-index",i);var v=O(s.settings.plyr,"config")?s.settings.plyr.config:{},f=new Plyr("#"+o,v);f.on("ready",(function(e){a[o]=e.detail.plyr,T(n)&&n()})),b((function(){return e.querySelector("iframe")&&"true"==e.querySelector("iframe").dataset.ready}),(function(){s.resize(e)})),f.on("enterfullscreen",R),f.on("exitfullscreen",R)}))}function R(e){var t=u(e.target,".gslide-media");"enterfullscreen"===e.type&&h(t,"fullscreen"),"exitfullscreen"===e.type&&d(t,"fullscreen")}function G(e,t,i,n){var s,l=this,o=e.querySelector(".gslide-media"),r=!(!O(t,"href")||!t.href)&&t.href.split("#").pop().trim(),d=!(!O(t,"content")||!t.content)&&t.content;if(d&&(C(d)&&(s=m('<div class="ginlined-content">'.concat(d,"</div>"))),k(d))){"none"==d.style.display&&(d.style.display="block");var c=document.createElement("div");c.className="ginlined-content",c.appendChild(d),s=c}if(r){var u=document.getElementById(r);if(!u)return!1;var g=u.cloneNode(!0);g.style.height=t.height,g.style.maxWidth=t.width,h(g,"ginlined-content"),s=g}if(!s)return console.error("Unable to append inline slide content",t),!1;o.style.height=t.height,o.style.width=t.width,o.appendChild(s),this.events["inlineclose"+r]=a("click",{onElement:o.querySelectorAll(".gtrigger-close"),withCallback:function(e){e.preventDefault(),l.close()}}),T(n)&&n()}function Z(e,t,i,n){var s=e.querySelector(".gslide-media"),l=function(e){var t=e.url,i=e.allow,n=e.callback,s=e.appendTo,l=document.createElement("iframe");return l.className="vimeo-video gvideo",l.src=t,l.style.width="100%",l.style.height="100%",i&&l.setAttribute("allow",i),l.onload=function(){l.onload=null,h(l,"node-ready"),T(n)&&n()},s&&s.appendChild(l),l}({url:t.href,callback:n});s.parentNode.style.maxWidth=t.width,s.parentNode.style.height=t.height,s.appendChild(l)}var U=function(){function e(){var i=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{};t(this,e),this.defaults={href:"",sizes:"",srcset:"",title:"",type:"",videoProvider:"",description:"",alt:"",descPosition:"bottom",effect:"",width:"",height:"",content:!1,zoomable:!0,draggable:!0},L(i)&&(this.defaults=l(this.defaults,i))}return n(e,[{key:"sourceType",value:function(e){var t=e;if(null!==(e=e.toLowerCase()).match(/\.(jpeg|jpg|jpe|gif|png|apn|webp|avif|svg)/))return"image";if(e.match(/(youtube\.com|youtube-nocookie\.com)\/watch\?v=([a-zA-Z0-9\-_]+)/)||e.match(/youtu\.be\/([a-zA-Z0-9\-_]+)/)||e.match(/(youtube\.com|youtube-nocookie\.com)\/embed\/([a-zA-Z0-9\-_]+)/))return"video";if(e.match(/vimeo\.com\/([0-9]*)/))return"video";if(null!==e.match(/\.(mp4|ogg|webm|mov)/))return"video";if(null!==e.match(/\.(mp3|wav|wma|aac|ogg)/))return"audio";if(e.indexOf("#")>-1&&""!==t.split("#").pop().trim())return"inline";return e.indexOf("goajax=true")>-1?"ajax":"external"}},{key:"parseConfig",value:function(e,t){var i=this,n=l({descPosition:t.descPosition},this.defaults);if(L(e)&&!k(e)){O(e,"type")||(O(e,"content")&&e.content?e.type="inline":O(e,"href")&&(e.type=this.sourceType(e.href)));var s=l(n,e);return this.setSize(s,t),s}var r="",a=e.getAttribute("data-glightbox"),h=e.nodeName.toLowerCase();if("a"===h&&(r=e.href),"img"===h&&(r=e.src,n.alt=e.alt),n.href=r,o(n,(function(s,l){O(t,l)&&"width"!==l&&(n[l]=t[l]);var o=e.dataset[l];I(o)||(n[l]=i.sanitizeValue(o))})),n.content&&(n.type="inline"),!n.type&&r&&(n.type=this.sourceType(r)),I(a)){if(!n.title&&"a"==h){var d=e.title;I(d)||""===d||(n.title=d)}if(!n.title&&"img"==h){var c=e.alt;I(c)||""===c||(n.title=c)}}else{var u=[];o(n,(function(e,t){u.push(";\\s?"+t)})),u=u.join("\\s?:|"),""!==a.trim()&&o(n,(function(e,t){var s=a,l=new RegExp("s?"+t+"s?:s?(.*?)("+u+"s?:|$)"),o=s.match(l);if(o&&o.length&&o[1]){var r=o[1].trim().replace(/;\s*$/,"");n[t]=i.sanitizeValue(r)}}))}if(n.description&&"."===n.description.substring(0,1)){var g;try{g=document.querySelector(n.description).innerHTML}catch(e){if(!(e instanceof DOMException))throw e}g&&(n.description=g)}if(!n.description){var v=e.querySelector(".glightbox-desc");v&&(n.description=v.innerHTML)}return this.setSize(n,t,e),this.slideConfig=n,n}},{key:"setSize",value:function(e,t){var i=arguments.length>2&&void 0!==arguments[2]?arguments[2]:null,n="video"==e.type?this.checkSize(t.videosWidth):this.checkSize(t.width),s=this.checkSize(t.height);return e.width=O(e,"width")&&""!==e.width?this.checkSize(e.width):n,e.height=O(e,"height")&&""!==e.height?this.checkSize(e.height):s,i&&"image"==e.type&&(e._hasCustomWidth=!!i.dataset.width,e._hasCustomHeight=!!i.dataset.height),e}},{key:"checkSize",value:function(e){return M(e)?"".concat(e,"px"):e}},{key:"sanitizeValue",value:function(e){return"true"!==e&&"false"!==e?e:"true"===e}}]),e}(),$=function(){function e(i,n,s){t(this,e),this.element=i,this.instance=n,this.index=s}return n(e,[{key:"setContent",value:function(){var e=this,t=arguments.length>0&&void 0!==arguments[0]?arguments[0]:null,i=arguments.length>1&&void 0!==arguments[1]&&arguments[1];if(c(t,"loaded"))return!1;var n=this.instance.settings,s=this.slideConfig,l=w();T(n.beforeSlideLoad)&&n.beforeSlideLoad({index:this.index,slide:t,player:!1});var o=s.type,r=s.descPosition,a=t.querySelector(".gslide-media"),d=t.querySelector(".gslide-title"),u=t.querySelector(".gslide-desc"),g=t.querySelector(".gdesc-inner"),v=i,f="gSlideTitle_"+this.index,p="gSlideDesc_"+this.index;if(T(n.afterSlideLoad)&&(v=function(){T(i)&&i(),n.afterSlideLoad({index:e.index,slide:t,player:e.instance.getSlidePlayerInstance(e.index)})}),""==s.title&&""==s.description?g&&g.parentNode.parentNode.removeChild(g.parentNode):(d&&""!==s.title?(d.id=f,d.innerHTML=s.title):d.parentNode.removeChild(d),u&&""!==s.description?(u.id=p,l&&n.moreLength>0?(s.smallDescription=this.slideShortDesc(s.description,n.moreLength,n.moreText),u.innerHTML=s.smallDescription,this.descriptionEvents(u,s)):u.innerHTML=s.description):u.parentNode.removeChild(u),h(a.parentNode,"desc-".concat(r)),h(g.parentNode,"description-".concat(r))),h(a,"gslide-".concat(o)),h(t,"loaded"),"video"!==o){if("external"!==o)return"inline"===o?(G.apply(this.instance,[t,s,this.index,v]),void(s.draggable&&new V({dragEl:t.querySelector(".gslide-inline"),toleranceX:n.dragToleranceX,toleranceY:n.dragToleranceY,slide:t,instance:this.instance}))):void("image"!==o?T(v)&&v():j(t,s,this.index,(function(){var i=t.querySelector("img");s.draggable&&new V({dragEl:i,toleranceX:n.dragToleranceX,toleranceY:n.dragToleranceY,slide:t,instance:e.instance}),s.zoomable&&i.naturalWidth>i.offsetWidth&&(h(i,"zoomable"),new H(i,t,(function(){e.instance.resize()}))),T(v)&&v()})));Z.apply(this,[t,s,this.index,v])}else F.apply(this.instance,[t,s,this.index,v])}},{key:"slideShortDesc",value:function(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:50,i=arguments.length>2&&void 0!==arguments[2]&&arguments[2],n=document.createElement("div");n.innerHTML=e;var s=n.innerText,l=i;if((e=s.trim()).length<=t)return e;var o=e.substr(0,t-1);return l?(n=null,o+'... <a href="#" class="desc-more">'+i+"</a>"):o}},{key:"descriptionEvents",value:function(e,t){var i=this,n=e.querySelector(".desc-more");if(!n)return!1;a("click",{onElement:n,withCallback:function(e,n){e.preventDefault();var s=document.body,l=u(n,".gslide-desc");if(!l)return!1;l.innerHTML=t.description,h(s,"gdesc-open");var o=a("click",{onElement:[s,u(l,".gslide-description")],withCallback:function(e,n){"a"!==e.target.nodeName.toLowerCase()&&(d(s,"gdesc-open"),h(s,"gdesc-closed"),l.innerHTML=t.smallDescription,i.descriptionEvents(l,t),setTimeout((function(){d(s,"gdesc-closed")}),400),o.destroy())}})}})}},{key:"create",value:function(){return m(this.instance.settings.slideHTML)}},{key:"getConfig",value:function(){k(this.element)||this.element.hasOwnProperty("draggable")||(this.element.draggable=this.instance.settings.draggable);var e=new U(this.instance.settings.slideExtraAttributes);return this.slideConfig=e.parseConfig(this.element,this.instance.settings),this.slideConfig}}]),e}(),J=w(),K=null!==w()||void 0!==document.createTouch||"ontouchstart"in window||"onmsgesturechange"in window||navigator.msMaxTouchPoints,Q=document.getElementsByTagName("html")[0],ee={selector:".glightbox",elements:null,skin:"clean",theme:"clean",closeButton:!0,startAt:null,autoplayVideos:!0,autofocusVideos:!0,descPosition:"bottom",width:"900px",height:"506px",videosWidth:"960px",beforeSlideChange:null,afterSlideChange:null,beforeSlideLoad:null,afterSlideLoad:null,slideInserted:null,slideRemoved:null,slideExtraAttributes:null,onOpen:null,onClose:null,loop:!1,zoomable:!0,draggable:!0,dragAutoSnap:!1,dragToleranceX:40,dragToleranceY:65,preload:!0,oneSlidePerOpen:!1,touchNavigation:!0,touchFollowAxis:!0,keyboardNavigation:!0,closeOnOutsideClick:!0,plugins:!1,plyr:{css:"https://cdn.plyr.io/3.6.12/plyr.css",js:"https://cdn.plyr.io/3.6.12/plyr.js",config:{ratio:"16:9",fullscreen:{enabled:!0,iosNative:!0},youtube:{noCookie:!0,rel:0,showinfo:0,iv_load_policy:3},vimeo:{byline:!1,portrait:!1,title:!1,transparent:!1}}},openEffect:"zoom",closeEffect:"zoom",slideEffect:"slide",moreText:"See more",moreLength:60,cssEfects:{fade:{in:"fadeIn",out:"fadeOut"},zoom:{in:"zoomIn",out:"zoomOut"},slide:{in:"slideInRight",out:"slideOutLeft"},slideBack:{in:"slideInLeft",out:"slideOutRight"},none:{in:"none",out:"none"}},svg:{close:'<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" x="0px" y="0px" viewBox="0 0 512 512" xml:space="preserve"><g><g><path d="M505.943,6.058c-8.077-8.077-21.172-8.077-29.249,0L6.058,476.693c-8.077,8.077-8.077,21.172,0,29.249C10.096,509.982,15.39,512,20.683,512c5.293,0,10.586-2.019,14.625-6.059L505.943,35.306C514.019,27.23,514.019,14.135,505.943,6.058z"/></g></g><g><g><path d="M505.942,476.694L35.306,6.059c-8.076-8.077-21.172-8.077-29.248,0c-8.077,8.076-8.077,21.171,0,29.248l470.636,470.636c4.038,4.039,9.332,6.058,14.625,6.058c5.293,0,10.587-2.019,14.624-6.057C514.018,497.866,514.018,484.771,505.942,476.694z"/></g></g></svg>',next:'<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" x="0px" y="0px" viewBox="0 0 477.175 477.175" xml:space="preserve"> <g><path d="M360.731,229.075l-225.1-225.1c-5.3-5.3-13.8-5.3-19.1,0s-5.3,13.8,0,19.1l215.5,215.5l-215.5,215.5c-5.3,5.3-5.3,13.8,0,19.1c2.6,2.6,6.1,4,9.5,4c3.4,0,6.9-1.3,9.5-4l225.1-225.1C365.931,242.875,365.931,234.275,360.731,229.075z"/></g></svg>',prev:'<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" x="0px" y="0px" viewBox="0 0 477.175 477.175" xml:space="preserve"><g><path d="M145.188,238.575l215.5-215.5c5.3-5.3,5.3-13.8,0-19.1s-13.8-5.3-19.1,0l-225.1,225.1c-5.3,5.3-5.3,13.8,0,19.1l225.1,225c2.6,2.6,6.1,4,9.5,4s6.9-1.3,9.5-4c5.3-5.3,5.3-13.8,0-19.1L145.188,238.575z"/></g></svg>'},slideHTML:'<div class="gslide">\n    <div class="gslide-inner-content">\n        <div class="ginner-container">\n            <div class="gslide-media">\n            </div>\n            <div class="gslide-description">\n                <div class="gdesc-inner">\n                    <h4 class="gslide-title"></h4>\n                    <div class="gslide-desc"></div>\n                </div>\n            </div>\n        </div>\n    </div>\n</div>',lightboxHTML:'<div id="glightbox-body" class="glightbox-container" tabindex="-1" role="dialog" aria-hidden="false">\n    <div class="gloader visible"></div>\n    <div class="goverlay"></div>\n    <div class="gcontainer">\n    <div id="glightbox-slider" class="gslider"></div>\n    <button class="gclose gbtn" aria-label="Close" data-taborder="3">{closeSVG}</button>\n    <button class="gprev gbtn" aria-label="Previous" data-taborder="2">{prevSVG}</button>\n    <button class="gnext gbtn" aria-label="Next" data-taborder="1">{nextSVG}</button>\n</div>\n</div>'},te=function(){function e(){var i=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{};t(this,e),this.customOptions=i,this.settings=l(ee,i),this.effectsClasses=this.getAnimationClasses(),this.videoPlayers={},this.apiEvents=[],this.fullElementsList=!1}return n(e,[{key:"init",value:function(){var e=this,t=this.getSelector();t&&(this.baseEvents=a("click",{onElement:t,withCallback:function(t,i){t.preventDefault(),e.open(i)}})),this.elements=this.getElements()}},{key:"open",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:null,t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:null;if(0===this.elements.length)return!1;this.activeSlide=null,this.prevActiveSlideIndex=null,this.prevActiveSlide=null;var i=M(t)?t:this.settings.startAt;if(k(e)){var n=e.getAttribute("data-gallery");n&&(this.fullElementsList=this.elements,this.elements=this.getGalleryElements(this.elements,n)),I(i)&&(i=this.getElementIndex(e))<0&&(i=0)}M(i)||(i=0),this.build(),g(this.overlay,"none"===this.settings.openEffect?"none":this.settings.cssEfects.fade.in);var s=document.body,l=window.innerWidth-document.documentElement.clientWidth;if(l>0){var o=document.createElement("style");o.type="text/css",o.className="gcss-styles",o.innerText=".gscrollbar-fixer {margin-right: ".concat(l,"px}"),document.head.appendChild(o),h(s,"gscrollbar-fixer")}h(s,"glightbox-open"),h(Q,"glightbox-open"),J&&(h(document.body,"glightbox-mobile"),this.settings.slideEffect="slide"),this.showSlide(i,!0),1===this.elements.length?(h(this.prevButton,"glightbox-button-hidden"),h(this.nextButton,"glightbox-button-hidden")):(d(this.prevButton,"glightbox-button-hidden"),d(this.nextButton,"glightbox-button-hidden")),this.lightboxOpen=!0,this.trigger("open"),T(this.settings.onOpen)&&this.settings.onOpen(),K&&this.settings.touchNavigation&&B(this),this.settings.keyboardNavigation&&X(this)}},{key:"openAt",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:0;this.open(null,e)}},{key:"showSlide",value:function(){var e=this,t=arguments.length>0&&void 0!==arguments[0]?arguments[0]:0,i=arguments.length>1&&void 0!==arguments[1]&&arguments[1];f(this.loader),this.index=parseInt(t);var n=this.slidesContainer.querySelector(".current");n&&d(n,"current"),this.slideAnimateOut();var s=this.slidesContainer.querySelectorAll(".gslide")[t];if(c(s,"loaded"))this.slideAnimateIn(s,i),p(this.loader);else{f(this.loader);var l=this.elements[t],o={index:this.index,slide:s,slideNode:s,slideConfig:l.slideConfig,slideIndex:this.index,trigger:l.node,player:null};this.trigger("slide_before_load",o),l.instance.setContent(s,(function(){p(e.loader),e.resize(),e.slideAnimateIn(s,i),e.trigger("slide_after_load",o)}))}this.slideDescription=s.querySelector(".gslide-description"),this.slideDescriptionContained=this.slideDescription&&c(this.slideDescription.parentNode,"gslide-media"),this.settings.preload&&(this.preloadSlide(t+1),this.preloadSlide(t-1)),this.updateNavigationClasses(),this.activeSlide=s}},{key:"preloadSlide",value:function(e){var t=this;if(e<0||e>this.elements.length-1)return!1;if(I(this.elements[e]))return!1;var i=this.slidesContainer.querySelectorAll(".gslide")[e];if(c(i,"loaded"))return!1;var n=this.elements[e],s=n.type,l={index:e,slide:i,slideNode:i,slideConfig:n.slideConfig,slideIndex:e,trigger:n.node,player:null};this.trigger("slide_before_load",l),"video"===s||"external"===s?setTimeout((function(){n.instance.setContent(i,(function(){t.trigger("slide_after_load",l)}))}),200):n.instance.setContent(i,(function(){t.trigger("slide_after_load",l)}))}},{key:"prevSlide",value:function(){this.goToSlide(this.index-1)}},{key:"nextSlide",value:function(){this.goToSlide(this.index+1)}},{key:"goToSlide",value:function(){var e=arguments.length>0&&void 0!==arguments[0]&&arguments[0];if(this.prevActiveSlide=this.activeSlide,this.prevActiveSlideIndex=this.index,!this.loop()&&(e<0||e>this.elements.length-1))return!1;e<0?e=this.elements.length-1:e>=this.elements.length&&(e=0),this.showSlide(e)}},{key:"insertSlide",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{},t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:-1;t<0&&(t=this.elements.length);var i=new $(e,this,t),n=i.getConfig(),s=l({},n),o=i.create(),r=this.elements.length-1;s.index=t,s.node=!1,s.instance=i,s.slideConfig=n,this.elements.splice(t,0,s);var a=null,h=null;if(this.slidesContainer){if(t>r)this.slidesContainer.appendChild(o);else{var d=this.slidesContainer.querySelectorAll(".gslide")[t];this.slidesContainer.insertBefore(o,d)}(this.settings.preload&&0==this.index&&0==t||this.index-1==t||this.index+1==t)&&this.preloadSlide(t),0===this.index&&0===t&&(this.index=1),this.updateNavigationClasses(),a=this.slidesContainer.querySelectorAll(".gslide")[t],h=this.getSlidePlayerInstance(t),s.slideNode=a}this.trigger("slide_inserted",{index:t,slide:a,slideNode:a,slideConfig:n,slideIndex:t,trigger:null,player:h}),T(this.settings.slideInserted)&&this.settings.slideInserted({index:t,slide:a,player:h})}},{key:"removeSlide",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:-1;if(e<0||e>this.elements.length-1)return!1;var t=this.slidesContainer&&this.slidesContainer.querySelectorAll(".gslide")[e];t&&(this.getActiveSlideIndex()==e&&(e==this.elements.length-1?this.prevSlide():this.nextSlide()),t.parentNode.removeChild(t)),this.elements.splice(e,1),this.trigger("slide_removed",e),T(this.settings.slideRemoved)&&this.settings.slideRemoved(e)}},{key:"slideAnimateIn",value:function(e,t){var i=this,n=e.querySelector(".gslide-media"),s=e.querySelector(".gslide-description"),l={index:this.prevActiveSlideIndex,slide:this.prevActiveSlide,slideNode:this.prevActiveSlide,slideIndex:this.prevActiveSlide,slideConfig:I(this.prevActiveSlideIndex)?null:this.elements[this.prevActiveSlideIndex].slideConfig,trigger:I(this.prevActiveSlideIndex)?null:this.elements[this.prevActiveSlideIndex].node,player:this.getSlidePlayerInstance(this.prevActiveSlideIndex)},o={index:this.index,slide:this.activeSlide,slideNode:this.activeSlide,slideConfig:this.elements[this.index].slideConfig,slideIndex:this.index,trigger:this.elements[this.index].node,player:this.getSlidePlayerInstance(this.index)};if(n.offsetWidth>0&&s&&(p(s),s.style.display=""),d(e,this.effectsClasses),t)g(e,this.settings.cssEfects[this.settings.openEffect].in,(function(){i.settings.autoplayVideos&&i.slidePlayerPlay(e),i.trigger("slide_changed",{prev:l,current:o}),T(i.settings.afterSlideChange)&&i.settings.afterSlideChange.apply(i,[l,o])}));else{var r=this.settings.slideEffect,a="none"!==r?this.settings.cssEfects[r].in:r;this.prevActiveSlideIndex>this.index&&"slide"==this.settings.slideEffect&&(a=this.settings.cssEfects.slideBack.in),g(e,a,(function(){i.settings.autoplayVideos&&i.slidePlayerPlay(e),i.trigger("slide_changed",{prev:l,current:o}),T(i.settings.afterSlideChange)&&i.settings.afterSlideChange.apply(i,[l,o])}))}setTimeout((function(){i.resize(e)}),100),h(e,"current")}},{key:"slideAnimateOut",value:function(){if(!this.prevActiveSlide)return!1;var e=this.prevActiveSlide;d(e,this.effectsClasses),h(e,"prev");var t=this.settings.slideEffect,i="none"!==t?this.settings.cssEfects[t].out:t;this.slidePlayerPause(e),this.trigger("slide_before_change",{prev:{index:this.prevActiveSlideIndex,slide:this.prevActiveSlide,slideNode:this.prevActiveSlide,slideIndex:this.prevActiveSlideIndex,slideConfig:I(this.prevActiveSlideIndex)?null:this.elements[this.prevActiveSlideIndex].slideConfig,trigger:I(this.prevActiveSlideIndex)?null:this.elements[this.prevActiveSlideIndex].node,player:this.getSlidePlayerInstance(this.prevActiveSlideIndex)},current:{index:this.index,slide:this.activeSlide,slideNode:this.activeSlide,slideIndex:this.index,slideConfig:this.elements[this.index].slideConfig,trigger:this.elements[this.index].node,player:this.getSlidePlayerInstance(this.index)}}),T(this.settings.beforeSlideChange)&&this.settings.beforeSlideChange.apply(this,[{index:this.prevActiveSlideIndex,slide:this.prevActiveSlide,player:this.getSlidePlayerInstance(this.prevActiveSlideIndex)},{index:this.index,slide:this.activeSlide,player:this.getSlidePlayerInstance(this.index)}]),this.prevActiveSlideIndex>this.index&&"slide"==this.settings.slideEffect&&(i=this.settings.cssEfects.slideBack.out),g(e,i,(function(){var t=e.querySelector(".ginner-container"),i=e.querySelector(".gslide-media"),n=e.querySelector(".gslide-description");t.style.transform="",i.style.transform="",d(i,"greset"),i.style.opacity="",n&&(n.style.opacity=""),d(e,"prev")}))}},{key:"getAllPlayers",value:function(){return this.videoPlayers}},{key:"getSlidePlayerInstance",value:function(e){var t="gvideo"+e,i=this.getAllPlayers();return!(!O(i,t)||!i[t])&&i[t]}},{key:"stopSlideVideo",value:function(e){if(k(e)){var t=e.querySelector(".gvideo-wrapper");t&&(e=t.getAttribute("data-index"))}console.log("stopSlideVideo is deprecated, use slidePlayerPause");var i=this.getSlidePlayerInstance(e);i&&i.playing&&i.pause()}},{key:"slidePlayerPause",value:function(e){if(k(e)){var t=e.querySelector(".gvideo-wrapper");t&&(e=t.getAttribute("data-index"))}var i=this.getSlidePlayerInstance(e);i&&i.playing&&i.pause()}},{key:"playSlideVideo",value:function(e){if(k(e)){var t=e.querySelector(".gvideo-wrapper");t&&(e=t.getAttribute("data-index"))}console.log("playSlideVideo is deprecated, use slidePlayerPlay");var i=this.getSlidePlayerInstance(e);i&&!i.playing&&i.play()}},{key:"slidePlayerPlay",value:function(e){var t;if(!J||null!==(t=this.settings.plyr.config)&&void 0!==t&&t.muted){if(k(e)){var i=e.querySelector(".gvideo-wrapper");i&&(e=i.getAttribute("data-index"))}var n=this.getSlidePlayerInstance(e);n&&!n.playing&&(n.play(),this.settings.autofocusVideos&&n.elements.container.focus())}}},{key:"setElements",value:function(e){var t=this;this.settings.elements=!1;var i=[];e&&e.length&&o(e,(function(e,n){var s=new $(e,t,n),o=s.getConfig(),r=l({},o);r.slideConfig=o,r.instance=s,r.index=n,i.push(r)})),this.elements=i,this.lightboxOpen&&(this.slidesContainer.innerHTML="",this.elements.length&&(o(this.elements,(function(){var e=m(t.settings.slideHTML);t.slidesContainer.appendChild(e)})),this.showSlide(0,!0)))}},{key:"getElementIndex",value:function(e){var t=!1;return o(this.elements,(function(i,n){if(O(i,"node")&&i.node==e)return t=n,!0})),t}},{key:"getElements",value:function(){var e=this,t=[];this.elements=this.elements?this.elements:[],!I(this.settings.elements)&&E(this.settings.elements)&&this.settings.elements.length&&o(this.settings.elements,(function(i,n){var s=new $(i,e,n),o=s.getConfig(),r=l({},o);r.node=!1,r.index=n,r.instance=s,r.slideConfig=o,t.push(r)}));var i=!1;return this.getSelector()&&(i=document.querySelectorAll(this.getSelector())),i?(o(i,(function(i,n){var s=new $(i,e,n),o=s.getConfig(),r=l({},o);r.node=i,r.index=n,r.instance=s,r.slideConfig=o,r.gallery=i.getAttribute("data-gallery"),t.push(r)})),t):t}},{key:"getGalleryElements",value:function(e,t){return e.filter((function(e){return e.gallery==t}))}},{key:"getSelector",value:function(){return!this.settings.elements&&(this.settings.selector&&"data-"==this.settings.selector.substring(0,5)?"*[".concat(this.settings.selector,"]"):this.settings.selector)}},{key:"getActiveSlide",value:function(){return this.slidesContainer.querySelectorAll(".gslide")[this.index]}},{key:"getActiveSlideIndex",value:function(){return this.index}},{key:"getAnimationClasses",value:function(){var e=[];for(var t in this.settings.cssEfects)if(this.settings.cssEfects.hasOwnProperty(t)){var i=this.settings.cssEfects[t];e.push("g".concat(i.in)),e.push("g".concat(i.out))}return e.join(" ")}},{key:"build",value:function(){var e=this;if(this.built)return!1;var t=document.body.childNodes,i=[];o(t,(function(e){e.parentNode==document.body&&"#"!==e.nodeName.charAt(0)&&e.hasAttribute&&!e.hasAttribute("aria-hidden")&&(i.push(e),e.setAttribute("aria-hidden","true"))}));var n=O(this.settings.svg,"next")?this.settings.svg.next:"",s=O(this.settings.svg,"prev")?this.settings.svg.prev:"",l=O(this.settings.svg,"close")?this.settings.svg.close:"",r=this.settings.lightboxHTML;r=m(r=(r=(r=r.replace(/{nextSVG}/g,n)).replace(/{prevSVG}/g,s)).replace(/{closeSVG}/g,l)),document.body.appendChild(r);var d=document.getElementById("glightbox-body");this.modal=d;var g=d.querySelector(".gclose");this.prevButton=d.querySelector(".gprev"),this.nextButton=d.querySelector(".gnext"),this.overlay=d.querySelector(".goverlay"),this.loader=d.querySelector(".gloader"),this.slidesContainer=document.getElementById("glightbox-slider"),this.bodyHiddenChildElms=i,this.events={},h(this.modal,"glightbox-"+this.settings.skin),this.settings.closeButton&&g&&(this.events.close=a("click",{onElement:g,withCallback:function(t,i){t.preventDefault(),e.close()}})),g&&!this.settings.closeButton&&g.parentNode.removeChild(g),this.nextButton&&(this.events.next=a("click",{onElement:this.nextButton,withCallback:function(t,i){t.preventDefault(),e.nextSlide()}})),this.prevButton&&(this.events.prev=a("click",{onElement:this.prevButton,withCallback:function(t,i){t.preventDefault(),e.prevSlide()}})),this.settings.closeOnOutsideClick&&(this.events.outClose=a("click",{onElement:d,withCallback:function(t,i){e.preventOutsideClick||c(document.body,"glightbox-mobile")||u(t.target,".ginner-container")||u(t.target,".gbtn")||c(t.target,"gnext")||c(t.target,"gprev")||e.close()}})),o(this.elements,(function(t,i){e.slidesContainer.appendChild(t.instance.create()),t.slideNode=e.slidesContainer.querySelectorAll(".gslide")[i]})),K&&h(document.body,"glightbox-touch"),this.events.resize=a("resize",{onElement:window,withCallback:function(){e.resize()}}),this.built=!0}},{key:"resize",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:null;if((e=e||this.activeSlide)&&!c(e,"zoomed")){var t=y(),i=e.querySelector(".gvideo-wrapper"),n=e.querySelector(".gslide-image"),s=this.slideDescription,l=t.width,o=t.height;if(l<=768?h(document.body,"glightbox-mobile"):d(document.body,"glightbox-mobile"),i||n){var r=!1;if(s&&(c(s,"description-bottom")||c(s,"description-top"))&&!c(s,"gabsolute")&&(r=!0),n)if(l<=768)n.querySelector("img");else if(r){var a=s.offsetHeight,u=n.querySelector("img");u.setAttribute("style","max-height: calc(100vh - ".concat(a,"px)")),s.setAttribute("style","max-width: ".concat(u.offsetWidth,"px;"))}if(i){var g=O(this.settings.plyr.config,"ratio")?this.settings.plyr.config.ratio:"";if(!g){var v=i.clientWidth,f=i.clientHeight,p=v/f;g="".concat(v/p,":").concat(f/p)}var m=g.split(":"),x=this.settings.videosWidth,b=this.settings.videosWidth,S=(b=M(x)||-1!==x.indexOf("px")?parseInt(x):-1!==x.indexOf("vw")?l*parseInt(x)/100:-1!==x.indexOf("vh")?o*parseInt(x)/100:-1!==x.indexOf("%")?l*parseInt(x)/100:parseInt(i.clientWidth))/(parseInt(m[0])/parseInt(m[1]));if(S=Math.floor(S),r&&(o-=s.offsetHeight),b>l||S>o||o<S&&l>b){var w=i.offsetWidth,T=i.offsetHeight,C=o/T,k={width:w*C,height:T*C};i.parentNode.setAttribute("style","max-width: ".concat(k.width,"px")),r&&s.setAttribute("style","max-width: ".concat(k.width,"px;"))}else i.parentNode.style.maxWidth="".concat(x),r&&s.setAttribute("style","max-width: ".concat(x,";"))}}}}},{key:"reload",value:function(){this.init()}},{key:"updateNavigationClasses",value:function(){var e=this.loop();d(this.nextButton,"disabled"),d(this.prevButton,"disabled"),0==this.index&&this.elements.length-1==0?(h(this.prevButton,"disabled"),h(this.nextButton,"disabled")):0!==this.index||e?this.index!==this.elements.length-1||e||h(this.nextButton,"disabled"):h(this.prevButton,"disabled")}},{key:"loop",value:function(){var e=O(this.settings,"loopAtEnd")?this.settings.loopAtEnd:null;return e=O(this.settings,"loop")?this.settings.loop:e,e}},{key:"close",value:function(){var e=this;if(!this.lightboxOpen){if(this.events){for(var t in this.events)this.events.hasOwnProperty(t)&&this.events[t].destroy();this.events=null}return!1}if(this.closing)return!1;this.closing=!0,this.slidePlayerPause(this.activeSlide),this.fullElementsList&&(this.elements=this.fullElementsList),this.bodyHiddenChildElms.length&&o(this.bodyHiddenChildElms,(function(e){e.removeAttribute("aria-hidden")})),h(this.modal,"glightbox-closing"),g(this.overlay,"none"==this.settings.openEffect?"none":this.settings.cssEfects.fade.out),g(this.activeSlide,this.settings.cssEfects[this.settings.closeEffect].out,(function(){if(e.activeSlide=null,e.prevActiveSlideIndex=null,e.prevActiveSlide=null,e.built=!1,e.events){for(var t in e.events)e.events.hasOwnProperty(t)&&e.events[t].destroy();e.events=null}var i=document.body;d(Q,"glightbox-open"),d(i,"glightbox-open touching gdesc-open glightbox-touch glightbox-mobile gscrollbar-fixer"),e.modal.parentNode.removeChild(e.modal),e.trigger("close"),T(e.settings.onClose)&&e.settings.onClose();var n=document.querySelector(".gcss-styles");n&&n.parentNode.removeChild(n),e.lightboxOpen=!1,e.closing=null}))}},{key:"destroy",value:function(){this.close(),this.clearAllEvents(),this.baseEvents&&this.baseEvents.destroy()}},{key:"on",value:function(e,t){var i=arguments.length>2&&void 0!==arguments[2]&&arguments[2];if(!e||!T(t))throw new TypeError("Event name and callback must be defined");this.apiEvents.push({evt:e,once:i,callback:t})}},{key:"once",value:function(e,t){this.on(e,t,!0)}},{key:"trigger",value:function(e){var t=this,i=arguments.length>1&&void 0!==arguments[1]?arguments[1]:null,n=[];o(this.apiEvents,(function(t,s){var l=t.evt,o=t.once,r=t.callback;l==e&&(r(i),o&&n.push(s))})),n.length&&o(n,(function(e){return t.apiEvents.splice(e,1)}))}},{key:"clearAllEvents",value:function(){this.apiEvents.splice(0,this.apiEvents.length)}},{key:"version",value:function(){return"3.1.0"}}]),e}();return function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{},t=new te(e);return t.init(),t}}));
\ No newline at end of file
diff --git a/docs/2_39/site_libs/quarto-contrib/glightbox/lightbox.css b/docs/2_39/site_libs/quarto-contrib/glightbox/lightbox.css
new file mode 100644
index 000000000..46432d990
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-contrib/glightbox/lightbox.css
@@ -0,0 +1,26 @@
+body:not(.glightbox-mobile) div.gslide div.gslide-description,
+body:not(.glightbox-mobile) div.gslide-description .gslide-title,
+body:not(.glightbox-mobile) div.gslide-description .gslide-desc {
+  color: var(--quarto-body-color);
+  background-color: var(--quarto-body-bg);
+}
+
+body:not(.glightbox-mobile) div.gslide-media {
+  background-color: var(--quarto-body-bg);
+}
+
+.goverlay {
+  background: rgba(0, 0, 0, 0.7);
+}
+
+div.gslide-description .gslide-title {
+  margin-top: 0.25em;
+  margin-bottom: 0.25em;
+  font-weight: 500;
+  font-family: inherit;
+}
+
+div.gslide-description .gslide-desc {
+  padding-bottom: 0.5em;
+  font-family: inherit;
+}
diff --git a/docs/2_39/site_libs/quarto-html/anchor.min.js b/docs/2_39/site_libs/quarto-html/anchor.min.js
new file mode 100644
index 000000000..5ac814d1e
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/anchor.min.js
@@ -0,0 +1,9 @@
+// @license magnet:?xt=urn:btih:d3d9a9a6595521f9666a5e94cc830dab83b65699&dn=expat.txt Expat
+//
+// AnchorJS - v5.0.0 - 2023-01-18
+// https://www.bryanbraun.com/anchorjs/
+// Copyright (c) 2023 Bryan Braun; Licensed MIT
+//
+// @license magnet:?xt=urn:btih:d3d9a9a6595521f9666a5e94cc830dab83b65699&dn=expat.txt Expat
+!function(A,e){"use strict";"function"==typeof define&&define.amd?define([],e):"object"==typeof module&&module.exports?module.exports=e():(A.AnchorJS=e(),A.anchors=new A.AnchorJS)}(globalThis,function(){"use strict";return function(A){function u(A){A.icon=Object.prototype.hasOwnProperty.call(A,"icon")?A.icon:"",A.visible=Object.prototype.hasOwnProperty.call(A,"visible")?A.visible:"hover",A.placement=Object.prototype.hasOwnProperty.call(A,"placement")?A.placement:"right",A.ariaLabel=Object.prototype.hasOwnProperty.call(A,"ariaLabel")?A.ariaLabel:"Anchor",A.class=Object.prototype.hasOwnProperty.call(A,"class")?A.class:"",A.base=Object.prototype.hasOwnProperty.call(A,"base")?A.base:"",A.truncate=Object.prototype.hasOwnProperty.call(A,"truncate")?Math.floor(A.truncate):64,A.titleText=Object.prototype.hasOwnProperty.call(A,"titleText")?A.titleText:""}function d(A){var e;if("string"==typeof A||A instanceof String)e=[].slice.call(document.querySelectorAll(A));else{if(!(Array.isArray(A)||A instanceof NodeList))throw new TypeError("The selector provided to AnchorJS was invalid.");e=[].slice.call(A)}return e}this.options=A||{},this.elements=[],u(this.options),this.add=function(A){var e,t,o,i,n,s,a,r,l,c,h,p=[];if(u(this.options),0!==(e=d(A=A||"h2, h3, h4, h5, h6")).length){for(null===document.head.querySelector("style.anchorjs")&&((A=document.createElement("style")).className="anchorjs",A.appendChild(document.createTextNode("")),void 0===(h=document.head.querySelector('[rel="stylesheet"],style'))?document.head.appendChild(A):document.head.insertBefore(A,h),A.sheet.insertRule(".anchorjs-link{opacity:0;text-decoration:none;-webkit-font-smoothing:antialiased;-moz-osx-font-smoothing:grayscale}",A.sheet.cssRules.length),A.sheet.insertRule(":hover>.anchorjs-link,.anchorjs-link:focus{opacity:1}",A.sheet.cssRules.length),A.sheet.insertRule("[data-anchorjs-icon]::after{content:attr(data-anchorjs-icon)}",A.sheet.cssRules.length),A.sheet.insertRule('@font-face{font-family:anchorjs-icons;src:url(data:n/a;base64,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) format("truetype")}',A.sheet.cssRules.length)),h=document.querySelectorAll("[id]"),t=[].map.call(h,function(A){return A.id}),i=0;i<e.length;i++)if(this.hasAnchorJSLink(e[i]))p.push(i);else{if(e[i].hasAttribute("id"))o=e[i].getAttribute("id");else if(e[i].hasAttribute("data-anchor-id"))o=e[i].getAttribute("data-anchor-id");else{for(r=a=this.urlify(e[i].textContent),s=0;n=t.indexOf(r=void 0!==n?a+"-"+s:r),s+=1,-1!==n;);n=void 0,t.push(r),e[i].setAttribute("id",r),o=r}(l=document.createElement("a")).className="anchorjs-link "+this.options.class,l.setAttribute("aria-label",this.options.ariaLabel),l.setAttribute("data-anchorjs-icon",this.options.icon),this.options.titleText&&(l.title=this.options.titleText),c=document.querySelector("base")?window.location.pathname+window.location.search:"",c=this.options.base||c,l.href=c+"#"+o,"always"===this.options.visible&&(l.style.opacity="1"),""===this.options.icon&&(l.style.font="1em/1 anchorjs-icons","left"===this.options.placement)&&(l.style.lineHeight="inherit"),"left"===this.options.placement?(l.style.position="absolute",l.style.marginLeft="-1.25em",l.style.paddingRight=".25em",l.style.paddingLeft=".25em",e[i].insertBefore(l,e[i].firstChild)):(l.style.marginLeft=".1875em",l.style.paddingRight=".1875em",l.style.paddingLeft=".1875em",e[i].appendChild(l))}for(i=0;i<p.length;i++)e.splice(p[i]-i,1);this.elements=this.elements.concat(e)}return this},this.remove=function(A){for(var e,t,o=d(A),i=0;i<o.length;i++)(t=o[i].querySelector(".anchorjs-link"))&&(-1!==(e=this.elements.indexOf(o[i]))&&this.elements.splice(e,1),o[i].removeChild(t));return this},this.removeAll=function(){this.remove(this.elements)},this.urlify=function(A){var e=document.createElement("textarea");return e.innerHTML=A,A=e.value,this.options.truncate||u(this.options),A.trim().replace(/'/gi,"").replace(/[& +$,:;=?@"#{}|^~[`%!'<>\]./()*\\\n\t\b\v\u00A0]/g,"-").replace(/-{2,}/g,"-").substring(0,this.options.truncate).replace(/^-+|-+$/gm,"").toLowerCase()},this.hasAnchorJSLink=function(A){var e=A.firstChild&&-1<(" "+A.firstChild.className+" ").indexOf(" anchorjs-link "),A=A.lastChild&&-1<(" "+A.lastChild.className+" ").indexOf(" anchorjs-link ");return e||A||!1}}});
+// @license-end
\ No newline at end of file
diff --git a/docs/2_39/site_libs/quarto-html/popper.min.js b/docs/2_39/site_libs/quarto-html/popper.min.js
new file mode 100644
index 000000000..e3726d728
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/popper.min.js
@@ -0,0 +1,6 @@
+/**
+ * @popperjs/core v2.11.7 - MIT License
+ */
+
+!function(e,t){"object"==typeof exports&&"undefined"!=typeof module?t(exports):"function"==typeof define&&define.amd?define(["exports"],t):t((e="undefined"!=typeof globalThis?globalThis:e||self).Popper={})}(this,(function(e){"use strict";function t(e){if(null==e)return window;if("[object Window]"!==e.toString()){var t=e.ownerDocument;return t&&t.defaultView||window}return e}function n(e){return e instanceof t(e).Element||e instanceof Element}function r(e){return e instanceof t(e).HTMLElement||e instanceof HTMLElement}function o(e){return"undefined"!=typeof ShadowRoot&&(e instanceof t(e).ShadowRoot||e instanceof ShadowRoot)}var i=Math.max,a=Math.min,s=Math.round;function f(){var e=navigator.userAgentData;return null!=e&&e.brands&&Array.isArray(e.brands)?e.brands.map((function(e){return e.brand+"/"+e.version})).join(" "):navigator.userAgent}function c(){return!/^((?!chrome|android).)*safari/i.test(f())}function p(e,o,i){void 0===o&&(o=!1),void 0===i&&(i=!1);var a=e.getBoundingClientRect(),f=1,p=1;o&&r(e)&&(f=e.offsetWidth>0&&s(a.width)/e.offsetWidth||1,p=e.offsetHeight>0&&s(a.height)/e.offsetHeight||1);var u=(n(e)?t(e):window).visualViewport,l=!c()&&i,d=(a.left+(l&&u?u.offsetLeft:0))/f,h=(a.top+(l&&u?u.offsetTop:0))/p,m=a.width/f,v=a.height/p;return{width:m,height:v,top:h,right:d+m,bottom:h+v,left:d,x:d,y:h}}function u(e){var n=t(e);return{scrollLeft:n.pageXOffset,scrollTop:n.pageYOffset}}function l(e){return e?(e.nodeName||"").toLowerCase():null}function d(e){return((n(e)?e.ownerDocument:e.document)||window.document).documentElement}function h(e){return p(d(e)).left+u(e).scrollLeft}function m(e){return t(e).getComputedStyle(e)}function v(e){var t=m(e),n=t.overflow,r=t.overflowX,o=t.overflowY;return/auto|scroll|overlay|hidden/.test(n+o+r)}function y(e,n,o){void 0===o&&(o=!1);var i,a,f=r(n),c=r(n)&&function(e){var t=e.getBoundingClientRect(),n=s(t.width)/e.offsetWidth||1,r=s(t.height)/e.offsetHeight||1;return 1!==n||1!==r}(n),m=d(n),y=p(e,c,o),g={scrollLeft:0,scrollTop:0},b={x:0,y:0};return(f||!f&&!o)&&(("body"!==l(n)||v(m))&&(g=(i=n)!==t(i)&&r(i)?{scrollLeft:(a=i).scrollLeft,scrollTop:a.scrollTop}:u(i)),r(n)?((b=p(n,!0)).x+=n.clientLeft,b.y+=n.clientTop):m&&(b.x=h(m))),{x:y.left+g.scrollLeft-b.x,y:y.top+g.scrollTop-b.y,width:y.width,height:y.height}}function g(e){var t=p(e),n=e.offsetWidth,r=e.offsetHeight;return Math.abs(t.width-n)<=1&&(n=t.width),Math.abs(t.height-r)<=1&&(r=t.height),{x:e.offsetLeft,y:e.offsetTop,width:n,height:r}}function b(e){return"html"===l(e)?e:e.assignedSlot||e.parentNode||(o(e)?e.host:null)||d(e)}function x(e){return["html","body","#document"].indexOf(l(e))>=0?e.ownerDocument.body:r(e)&&v(e)?e:x(b(e))}function w(e,n){var r;void 0===n&&(n=[]);var o=x(e),i=o===(null==(r=e.ownerDocument)?void 0:r.body),a=t(o),s=i?[a].concat(a.visualViewport||[],v(o)?o:[]):o,f=n.concat(s);return i?f:f.concat(w(b(s)))}function O(e){return["table","td","th"].indexOf(l(e))>=0}function j(e){return r(e)&&"fixed"!==m(e).position?e.offsetParent:null}function E(e){for(var n=t(e),i=j(e);i&&O(i)&&"static"===m(i).position;)i=j(i);return i&&("html"===l(i)||"body"===l(i)&&"static"===m(i).position)?n:i||function(e){var t=/firefox/i.test(f());if(/Trident/i.test(f())&&r(e)&&"fixed"===m(e).position)return null;var n=b(e);for(o(n)&&(n=n.host);r(n)&&["html","body"].indexOf(l(n))<0;){var i=m(n);if("none"!==i.transform||"none"!==i.perspective||"paint"===i.contain||-1!==["transform","perspective"].indexOf(i.willChange)||t&&"filter"===i.willChange||t&&i.filter&&"none"!==i.filter)return n;n=n.parentNode}return null}(e)||n}var D="top",A="bottom",L="right",P="left",M="auto",k=[D,A,L,P],W="start",B="end",H="viewport",T="popper",R=k.reduce((function(e,t){return e.concat([t+"-"+W,t+"-"+B])}),[]),S=[].concat(k,[M]).reduce((function(e,t){return e.concat([t,t+"-"+W,t+"-"+B])}),[]),V=["beforeRead","read","afterRead","beforeMain","main","afterMain","beforeWrite","write","afterWrite"];function q(e){var t=new Map,n=new Set,r=[];function o(e){n.add(e.name),[].concat(e.requires||[],e.requiresIfExists||[]).forEach((function(e){if(!n.has(e)){var r=t.get(e);r&&o(r)}})),r.push(e)}return e.forEach((function(e){t.set(e.name,e)})),e.forEach((function(e){n.has(e.name)||o(e)})),r}function C(e){return e.split("-")[0]}function N(e,t){var n=t.getRootNode&&t.getRootNode();if(e.contains(t))return!0;if(n&&o(n)){var r=t;do{if(r&&e.isSameNode(r))return!0;r=r.parentNode||r.host}while(r)}return!1}function I(e){return Object.assign({},e,{left:e.x,top:e.y,right:e.x+e.width,bottom:e.y+e.height})}function _(e,r,o){return r===H?I(function(e,n){var r=t(e),o=d(e),i=r.visualViewport,a=o.clientWidth,s=o.clientHeight,f=0,p=0;if(i){a=i.width,s=i.height;var u=c();(u||!u&&"fixed"===n)&&(f=i.offsetLeft,p=i.offsetTop)}return{width:a,height:s,x:f+h(e),y:p}}(e,o)):n(r)?function(e,t){var n=p(e,!1,"fixed"===t);return n.top=n.top+e.clientTop,n.left=n.left+e.clientLeft,n.bottom=n.top+e.clientHeight,n.right=n.left+e.clientWidth,n.width=e.clientWidth,n.height=e.clientHeight,n.x=n.left,n.y=n.top,n}(r,o):I(function(e){var t,n=d(e),r=u(e),o=null==(t=e.ownerDocument)?void 0:t.body,a=i(n.scrollWidth,n.clientWidth,o?o.scrollWidth:0,o?o.clientWidth:0),s=i(n.scrollHeight,n.clientHeight,o?o.scrollHeight:0,o?o.clientHeight:0),f=-r.scrollLeft+h(e),c=-r.scrollTop;return"rtl"===m(o||n).direction&&(f+=i(n.clientWidth,o?o.clientWidth:0)-a),{width:a,height:s,x:f,y:c}}(d(e)))}function F(e,t,o,s){var f="clippingParents"===t?function(e){var t=w(b(e)),o=["absolute","fixed"].indexOf(m(e).position)>=0&&r(e)?E(e):e;return n(o)?t.filter((function(e){return n(e)&&N(e,o)&&"body"!==l(e)})):[]}(e):[].concat(t),c=[].concat(f,[o]),p=c[0],u=c.reduce((function(t,n){var r=_(e,n,s);return t.top=i(r.top,t.top),t.right=a(r.right,t.right),t.bottom=a(r.bottom,t.bottom),t.left=i(r.left,t.left),t}),_(e,p,s));return u.width=u.right-u.left,u.height=u.bottom-u.top,u.x=u.left,u.y=u.top,u}function U(e){return e.split("-")[1]}function z(e){return["top","bottom"].indexOf(e)>=0?"x":"y"}function X(e){var t,n=e.reference,r=e.element,o=e.placement,i=o?C(o):null,a=o?U(o):null,s=n.x+n.width/2-r.width/2,f=n.y+n.height/2-r.height/2;switch(i){case D:t={x:s,y:n.y-r.height};break;case A:t={x:s,y:n.y+n.height};break;case L:t={x:n.x+n.width,y:f};break;case P:t={x:n.x-r.width,y:f};break;default:t={x:n.x,y:n.y}}var c=i?z(i):null;if(null!=c){var p="y"===c?"height":"width";switch(a){case W:t[c]=t[c]-(n[p]/2-r[p]/2);break;case B:t[c]=t[c]+(n[p]/2-r[p]/2)}}return t}function Y(e){return Object.assign({},{top:0,right:0,bottom:0,left:0},e)}function G(e,t){return t.reduce((function(t,n){return t[n]=e,t}),{})}function J(e,t){void 0===t&&(t={});var r=t,o=r.placement,i=void 0===o?e.placement:o,a=r.strategy,s=void 0===a?e.strategy:a,f=r.boundary,c=void 0===f?"clippingParents":f,u=r.rootBoundary,l=void 0===u?H:u,h=r.elementContext,m=void 0===h?T:h,v=r.altBoundary,y=void 0!==v&&v,g=r.padding,b=void 0===g?0:g,x=Y("number"!=typeof b?b:G(b,k)),w=m===T?"reference":T,O=e.rects.popper,j=e.elements[y?w:m],E=F(n(j)?j:j.contextElement||d(e.elements.popper),c,l,s),P=p(e.elements.reference),M=X({reference:P,element:O,strategy:"absolute",placement:i}),W=I(Object.assign({},O,M)),B=m===T?W:P,R={top:E.top-B.top+x.top,bottom:B.bottom-E.bottom+x.bottom,left:E.left-B.left+x.left,right:B.right-E.right+x.right},S=e.modifiersData.offset;if(m===T&&S){var V=S[i];Object.keys(R).forEach((function(e){var t=[L,A].indexOf(e)>=0?1:-1,n=[D,A].indexOf(e)>=0?"y":"x";R[e]+=V[n]*t}))}return R}var K={placement:"bottom",modifiers:[],strategy:"absolute"};function Q(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];return!t.some((function(e){return!(e&&"function"==typeof e.getBoundingClientRect)}))}function Z(e){void 0===e&&(e={});var t=e,r=t.defaultModifiers,o=void 0===r?[]:r,i=t.defaultOptions,a=void 0===i?K:i;return function(e,t,r){void 0===r&&(r=a);var i,s,f={placement:"bottom",orderedModifiers:[],options:Object.assign({},K,a),modifiersData:{},elements:{reference:e,popper:t},attributes:{},styles:{}},c=[],p=!1,u={state:f,setOptions:function(r){var i="function"==typeof r?r(f.options):r;l(),f.options=Object.assign({},a,f.options,i),f.scrollParents={reference:n(e)?w(e):e.contextElement?w(e.contextElement):[],popper:w(t)};var s,p,d=function(e){var t=q(e);return V.reduce((function(e,n){return e.concat(t.filter((function(e){return e.phase===n})))}),[])}((s=[].concat(o,f.options.modifiers),p=s.reduce((function(e,t){var n=e[t.name];return e[t.name]=n?Object.assign({},n,t,{options:Object.assign({},n.options,t.options),data:Object.assign({},n.data,t.data)}):t,e}),{}),Object.keys(p).map((function(e){return p[e]}))));return f.orderedModifiers=d.filter((function(e){return e.enabled})),f.orderedModifiers.forEach((function(e){var t=e.name,n=e.options,r=void 0===n?{}:n,o=e.effect;if("function"==typeof o){var i=o({state:f,name:t,instance:u,options:r}),a=function(){};c.push(i||a)}})),u.update()},forceUpdate:function(){if(!p){var e=f.elements,t=e.reference,n=e.popper;if(Q(t,n)){f.rects={reference:y(t,E(n),"fixed"===f.options.strategy),popper:g(n)},f.reset=!1,f.placement=f.options.placement,f.orderedModifiers.forEach((function(e){return f.modifiersData[e.name]=Object.assign({},e.data)}));for(var r=0;r<f.orderedModifiers.length;r++)if(!0!==f.reset){var o=f.orderedModifiers[r],i=o.fn,a=o.options,s=void 0===a?{}:a,c=o.name;"function"==typeof i&&(f=i({state:f,options:s,name:c,instance:u})||f)}else f.reset=!1,r=-1}}},update:(i=function(){return new Promise((function(e){u.forceUpdate(),e(f)}))},function(){return s||(s=new Promise((function(e){Promise.resolve().then((function(){s=void 0,e(i())}))}))),s}),destroy:function(){l(),p=!0}};if(!Q(e,t))return u;function l(){c.forEach((function(e){return e()})),c=[]}return u.setOptions(r).then((function(e){!p&&r.onFirstUpdate&&r.onFirstUpdate(e)})),u}}var $={passive:!0};var ee={name:"eventListeners",enabled:!0,phase:"write",fn:function(){},effect:function(e){var n=e.state,r=e.instance,o=e.options,i=o.scroll,a=void 0===i||i,s=o.resize,f=void 0===s||s,c=t(n.elements.popper),p=[].concat(n.scrollParents.reference,n.scrollParents.popper);return a&&p.forEach((function(e){e.addEventListener("scroll",r.update,$)})),f&&c.addEventListener("resize",r.update,$),function(){a&&p.forEach((function(e){e.removeEventListener("scroll",r.update,$)})),f&&c.removeEventListener("resize",r.update,$)}},data:{}};var te={name:"popperOffsets",enabled:!0,phase:"read",fn:function(e){var t=e.state,n=e.name;t.modifiersData[n]=X({reference:t.rects.reference,element:t.rects.popper,strategy:"absolute",placement:t.placement})},data:{}},ne={top:"auto",right:"auto",bottom:"auto",left:"auto"};function re(e){var n,r=e.popper,o=e.popperRect,i=e.placement,a=e.variation,f=e.offsets,c=e.position,p=e.gpuAcceleration,u=e.adaptive,l=e.roundOffsets,h=e.isFixed,v=f.x,y=void 0===v?0:v,g=f.y,b=void 0===g?0:g,x="function"==typeof l?l({x:y,y:b}):{x:y,y:b};y=x.x,b=x.y;var w=f.hasOwnProperty("x"),O=f.hasOwnProperty("y"),j=P,M=D,k=window;if(u){var W=E(r),H="clientHeight",T="clientWidth";if(W===t(r)&&"static"!==m(W=d(r)).position&&"absolute"===c&&(H="scrollHeight",T="scrollWidth"),W=W,i===D||(i===P||i===L)&&a===B)M=A,b-=(h&&W===k&&k.visualViewport?k.visualViewport.height:W[H])-o.height,b*=p?1:-1;if(i===P||(i===D||i===A)&&a===B)j=L,y-=(h&&W===k&&k.visualViewport?k.visualViewport.width:W[T])-o.width,y*=p?1:-1}var R,S=Object.assign({position:c},u&&ne),V=!0===l?function(e,t){var n=e.x,r=e.y,o=t.devicePixelRatio||1;return{x:s(n*o)/o||0,y:s(r*o)/o||0}}({x:y,y:b},t(r)):{x:y,y:b};return y=V.x,b=V.y,p?Object.assign({},S,((R={})[M]=O?"0":"",R[j]=w?"0":"",R.transform=(k.devicePixelRatio||1)<=1?"translate("+y+"px, "+b+"px)":"translate3d("+y+"px, "+b+"px, 0)",R)):Object.assign({},S,((n={})[M]=O?b+"px":"",n[j]=w?y+"px":"",n.transform="",n))}var oe={name:"computeStyles",enabled:!0,phase:"beforeWrite",fn:function(e){var t=e.state,n=e.options,r=n.gpuAcceleration,o=void 0===r||r,i=n.adaptive,a=void 0===i||i,s=n.roundOffsets,f=void 0===s||s,c={placement:C(t.placement),variation:U(t.placement),popper:t.elements.popper,popperRect:t.rects.popper,gpuAcceleration:o,isFixed:"fixed"===t.options.strategy};null!=t.modifiersData.popperOffsets&&(t.styles.popper=Object.assign({},t.styles.popper,re(Object.assign({},c,{offsets:t.modifiersData.popperOffsets,position:t.options.strategy,adaptive:a,roundOffsets:f})))),null!=t.modifiersData.arrow&&(t.styles.arrow=Object.assign({},t.styles.arrow,re(Object.assign({},c,{offsets:t.modifiersData.arrow,position:"absolute",adaptive:!1,roundOffsets:f})))),t.attributes.popper=Object.assign({},t.attributes.popper,{"data-popper-placement":t.placement})},data:{}};var ie={name:"applyStyles",enabled:!0,phase:"write",fn:function(e){var t=e.state;Object.keys(t.elements).forEach((function(e){var n=t.styles[e]||{},o=t.attributes[e]||{},i=t.elements[e];r(i)&&l(i)&&(Object.assign(i.style,n),Object.keys(o).forEach((function(e){var t=o[e];!1===t?i.removeAttribute(e):i.setAttribute(e,!0===t?"":t)})))}))},effect:function(e){var t=e.state,n={popper:{position:t.options.strategy,left:"0",top:"0",margin:"0"},arrow:{position:"absolute"},reference:{}};return Object.assign(t.elements.popper.style,n.popper),t.styles=n,t.elements.arrow&&Object.assign(t.elements.arrow.style,n.arrow),function(){Object.keys(t.elements).forEach((function(e){var o=t.elements[e],i=t.attributes[e]||{},a=Object.keys(t.styles.hasOwnProperty(e)?t.styles[e]:n[e]).reduce((function(e,t){return e[t]="",e}),{});r(o)&&l(o)&&(Object.assign(o.style,a),Object.keys(i).forEach((function(e){o.removeAttribute(e)})))}))}},requires:["computeStyles"]};var ae={name:"offset",enabled:!0,phase:"main",requires:["popperOffsets"],fn:function(e){var t=e.state,n=e.options,r=e.name,o=n.offset,i=void 0===o?[0,0]:o,a=S.reduce((function(e,n){return e[n]=function(e,t,n){var r=C(e),o=[P,D].indexOf(r)>=0?-1:1,i="function"==typeof n?n(Object.assign({},t,{placement:e})):n,a=i[0],s=i[1];return a=a||0,s=(s||0)*o,[P,L].indexOf(r)>=0?{x:s,y:a}:{x:a,y:s}}(n,t.rects,i),e}),{}),s=a[t.placement],f=s.x,c=s.y;null!=t.modifiersData.popperOffsets&&(t.modifiersData.popperOffsets.x+=f,t.modifiersData.popperOffsets.y+=c),t.modifiersData[r]=a}},se={left:"right",right:"left",bottom:"top",top:"bottom"};function fe(e){return e.replace(/left|right|bottom|top/g,(function(e){return se[e]}))}var ce={start:"end",end:"start"};function pe(e){return e.replace(/start|end/g,(function(e){return ce[e]}))}function ue(e,t){void 0===t&&(t={});var n=t,r=n.placement,o=n.boundary,i=n.rootBoundary,a=n.padding,s=n.flipVariations,f=n.allowedAutoPlacements,c=void 0===f?S:f,p=U(r),u=p?s?R:R.filter((function(e){return U(e)===p})):k,l=u.filter((function(e){return c.indexOf(e)>=0}));0===l.length&&(l=u);var d=l.reduce((function(t,n){return t[n]=J(e,{placement:n,boundary:o,rootBoundary:i,padding:a})[C(n)],t}),{});return Object.keys(d).sort((function(e,t){return d[e]-d[t]}))}var le={name:"flip",enabled:!0,phase:"main",fn:function(e){var t=e.state,n=e.options,r=e.name;if(!t.modifiersData[r]._skip){for(var o=n.mainAxis,i=void 0===o||o,a=n.altAxis,s=void 0===a||a,f=n.fallbackPlacements,c=n.padding,p=n.boundary,u=n.rootBoundary,l=n.altBoundary,d=n.flipVariations,h=void 0===d||d,m=n.allowedAutoPlacements,v=t.options.placement,y=C(v),g=f||(y===v||!h?[fe(v)]:function(e){if(C(e)===M)return[];var t=fe(e);return[pe(e),t,pe(t)]}(v)),b=[v].concat(g).reduce((function(e,n){return e.concat(C(n)===M?ue(t,{placement:n,boundary:p,rootBoundary:u,padding:c,flipVariations:h,allowedAutoPlacements:m}):n)}),[]),x=t.rects.reference,w=t.rects.popper,O=new Map,j=!0,E=b[0],k=0;k<b.length;k++){var B=b[k],H=C(B),T=U(B)===W,R=[D,A].indexOf(H)>=0,S=R?"width":"height",V=J(t,{placement:B,boundary:p,rootBoundary:u,altBoundary:l,padding:c}),q=R?T?L:P:T?A:D;x[S]>w[S]&&(q=fe(q));var N=fe(q),I=[];if(i&&I.push(V[H]<=0),s&&I.push(V[q]<=0,V[N]<=0),I.every((function(e){return e}))){E=B,j=!1;break}O.set(B,I)}if(j)for(var _=function(e){var t=b.find((function(t){var n=O.get(t);if(n)return n.slice(0,e).every((function(e){return e}))}));if(t)return E=t,"break"},F=h?3:1;F>0;F--){if("break"===_(F))break}t.placement!==E&&(t.modifiersData[r]._skip=!0,t.placement=E,t.reset=!0)}},requiresIfExists:["offset"],data:{_skip:!1}};function de(e,t,n){return i(e,a(t,n))}var he={name:"preventOverflow",enabled:!0,phase:"main",fn:function(e){var t=e.state,n=e.options,r=e.name,o=n.mainAxis,s=void 0===o||o,f=n.altAxis,c=void 0!==f&&f,p=n.boundary,u=n.rootBoundary,l=n.altBoundary,d=n.padding,h=n.tether,m=void 0===h||h,v=n.tetherOffset,y=void 0===v?0:v,b=J(t,{boundary:p,rootBoundary:u,padding:d,altBoundary:l}),x=C(t.placement),w=U(t.placement),O=!w,j=z(x),M="x"===j?"y":"x",k=t.modifiersData.popperOffsets,B=t.rects.reference,H=t.rects.popper,T="function"==typeof y?y(Object.assign({},t.rects,{placement:t.placement})):y,R="number"==typeof T?{mainAxis:T,altAxis:T}:Object.assign({mainAxis:0,altAxis:0},T),S=t.modifiersData.offset?t.modifiersData.offset[t.placement]:null,V={x:0,y:0};if(k){if(s){var q,N="y"===j?D:P,I="y"===j?A:L,_="y"===j?"height":"width",F=k[j],X=F+b[N],Y=F-b[I],G=m?-H[_]/2:0,K=w===W?B[_]:H[_],Q=w===W?-H[_]:-B[_],Z=t.elements.arrow,$=m&&Z?g(Z):{width:0,height:0},ee=t.modifiersData["arrow#persistent"]?t.modifiersData["arrow#persistent"].padding:{top:0,right:0,bottom:0,left:0},te=ee[N],ne=ee[I],re=de(0,B[_],$[_]),oe=O?B[_]/2-G-re-te-R.mainAxis:K-re-te-R.mainAxis,ie=O?-B[_]/2+G+re+ne+R.mainAxis:Q+re+ne+R.mainAxis,ae=t.elements.arrow&&E(t.elements.arrow),se=ae?"y"===j?ae.clientTop||0:ae.clientLeft||0:0,fe=null!=(q=null==S?void 0:S[j])?q:0,ce=F+ie-fe,pe=de(m?a(X,F+oe-fe-se):X,F,m?i(Y,ce):Y);k[j]=pe,V[j]=pe-F}if(c){var ue,le="x"===j?D:P,he="x"===j?A:L,me=k[M],ve="y"===M?"height":"width",ye=me+b[le],ge=me-b[he],be=-1!==[D,P].indexOf(x),xe=null!=(ue=null==S?void 0:S[M])?ue:0,we=be?ye:me-B[ve]-H[ve]-xe+R.altAxis,Oe=be?me+B[ve]+H[ve]-xe-R.altAxis:ge,je=m&&be?function(e,t,n){var r=de(e,t,n);return r>n?n:r}(we,me,Oe):de(m?we:ye,me,m?Oe:ge);k[M]=je,V[M]=je-me}t.modifiersData[r]=V}},requiresIfExists:["offset"]};var me={name:"arrow",enabled:!0,phase:"main",fn:function(e){var t,n=e.state,r=e.name,o=e.options,i=n.elements.arrow,a=n.modifiersData.popperOffsets,s=C(n.placement),f=z(s),c=[P,L].indexOf(s)>=0?"height":"width";if(i&&a){var p=function(e,t){return Y("number"!=typeof(e="function"==typeof e?e(Object.assign({},t.rects,{placement:t.placement})):e)?e:G(e,k))}(o.padding,n),u=g(i),l="y"===f?D:P,d="y"===f?A:L,h=n.rects.reference[c]+n.rects.reference[f]-a[f]-n.rects.popper[c],m=a[f]-n.rects.reference[f],v=E(i),y=v?"y"===f?v.clientHeight||0:v.clientWidth||0:0,b=h/2-m/2,x=p[l],w=y-u[c]-p[d],O=y/2-u[c]/2+b,j=de(x,O,w),M=f;n.modifiersData[r]=((t={})[M]=j,t.centerOffset=j-O,t)}},effect:function(e){var t=e.state,n=e.options.element,r=void 0===n?"[data-popper-arrow]":n;null!=r&&("string"!=typeof r||(r=t.elements.popper.querySelector(r)))&&N(t.elements.popper,r)&&(t.elements.arrow=r)},requires:["popperOffsets"],requiresIfExists:["preventOverflow"]};function ve(e,t,n){return void 0===n&&(n={x:0,y:0}),{top:e.top-t.height-n.y,right:e.right-t.width+n.x,bottom:e.bottom-t.height+n.y,left:e.left-t.width-n.x}}function ye(e){return[D,L,A,P].some((function(t){return e[t]>=0}))}var ge={name:"hide",enabled:!0,phase:"main",requiresIfExists:["preventOverflow"],fn:function(e){var t=e.state,n=e.name,r=t.rects.reference,o=t.rects.popper,i=t.modifiersData.preventOverflow,a=J(t,{elementContext:"reference"}),s=J(t,{altBoundary:!0}),f=ve(a,r),c=ve(s,o,i),p=ye(f),u=ye(c);t.modifiersData[n]={referenceClippingOffsets:f,popperEscapeOffsets:c,isReferenceHidden:p,hasPopperEscaped:u},t.attributes.popper=Object.assign({},t.attributes.popper,{"data-popper-reference-hidden":p,"data-popper-escaped":u})}},be=Z({defaultModifiers:[ee,te,oe,ie]}),xe=[ee,te,oe,ie,ae,le,he,me,ge],we=Z({defaultModifiers:xe});e.applyStyles=ie,e.arrow=me,e.computeStyles=oe,e.createPopper=we,e.createPopperLite=be,e.defaultModifiers=xe,e.detectOverflow=J,e.eventListeners=ee,e.flip=le,e.hide=ge,e.offset=ae,e.popperGenerator=Z,e.popperOffsets=te,e.preventOverflow=he,Object.defineProperty(e,"__esModule",{value:!0})}));
+
diff --git a/docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css b/docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css
new file mode 100644
index 000000000..05b0f7099
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css
@@ -0,0 +1,191 @@
+/* quarto syntax highlight colors */
+:root {
+  --quarto-hl-al-color: #ef2929;
+  --quarto-hl-an-color: #8f5902;
+  --quarto-hl-at-color: #204a87;
+  --quarto-hl-bn-color: #0000cf;
+  --quarto-hl-ch-color: #4e9a06;
+  --quarto-hl-co-color: #8f5902;
+  --quarto-hl-cv-color: #8f5902;
+  --quarto-hl-cn-color: #8f5902;
+  --quarto-hl-cf-color: #204a87;
+  --quarto-hl-dt-color: #204a87;
+  --quarto-hl-dv-color: #0000cf;
+  --quarto-hl-do-color: #8f5902;
+  --quarto-hl-er-color: #a40000;
+  --quarto-hl-ex-color: inherit;
+  --quarto-hl-fl-color: #0000cf;
+  --quarto-hl-fu-color: #204a87;
+  --quarto-hl-im-color: inherit;
+  --quarto-hl-in-color: #8f5902;
+  --quarto-hl-kw-color: #204a87;
+  --quarto-hl-op-color: #ce5c00;
+  --quarto-hl-ot-color: #8f5902;
+  --quarto-hl-pp-color: #8f5902;
+  --quarto-hl-sc-color: #ce5c00;
+  --quarto-hl-ss-color: #4e9a06;
+  --quarto-hl-st-color: #4e9a06;
+  --quarto-hl-va-color: #000000;
+  --quarto-hl-vs-color: #4e9a06;
+  --quarto-hl-wa-color: #8f5902;
+}
+
+/* other quarto variables */
+:root {
+  --quarto-font-monospace: SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", "Courier New", monospace;
+}
+
+code span.al {
+  color: #ef2929;
+  font-style: inherit;
+}
+
+code span.an {
+  color: #8f5902;
+  font-weight: bold;
+  font-style: italic;
+}
+
+code span.at {
+  color: #204a87;
+  font-style: inherit;
+}
+
+code span.bn {
+  color: #0000cf;
+  font-style: inherit;
+}
+
+code span.ch {
+  color: #4e9a06;
+  font-style: inherit;
+}
+
+code span.co {
+  color: #8f5902;
+  font-style: italic;
+}
+
+code span.cv {
+  color: #8f5902;
+  font-weight: bold;
+  font-style: italic;
+}
+
+code span.cn {
+  color: #8f5902;
+  font-style: inherit;
+}
+
+code span.cf {
+  color: #204a87;
+  font-weight: bold;
+  font-style: inherit;
+}
+
+code span.dt {
+  color: #204a87;
+  font-style: inherit;
+}
+
+code span.dv {
+  color: #0000cf;
+  font-style: inherit;
+}
+
+code span.do {
+  color: #8f5902;
+  font-weight: bold;
+  font-style: italic;
+}
+
+code span.er {
+  color: #a40000;
+  font-weight: bold;
+  font-style: inherit;
+}
+
+code span.ex {
+  font-style: inherit;
+}
+
+code span.fl {
+  color: #0000cf;
+  font-style: inherit;
+}
+
+code span.fu {
+  color: #204a87;
+  font-weight: bold;
+  font-style: inherit;
+}
+
+code span.im {
+  font-style: inherit;
+}
+
+code span.in {
+  color: #8f5902;
+  font-weight: bold;
+  font-style: italic;
+}
+
+code span.kw {
+  color: #204a87;
+  font-weight: bold;
+  font-style: inherit;
+}
+
+code span.op {
+  color: #ce5c00;
+  font-weight: bold;
+  font-style: inherit;
+}
+
+code span.ot {
+  color: #8f5902;
+  font-style: inherit;
+}
+
+code span.pp {
+  color: #8f5902;
+  font-style: italic;
+}
+
+code span.sc {
+  color: #ce5c00;
+  font-weight: bold;
+  font-style: inherit;
+}
+
+code span.ss {
+  color: #4e9a06;
+  font-style: inherit;
+}
+
+code span.st {
+  color: #4e9a06;
+  font-style: inherit;
+}
+
+code span.va {
+  color: #000000;
+  font-style: inherit;
+}
+
+code span.vs {
+  color: #4e9a06;
+  font-style: inherit;
+}
+
+code span.wa {
+  color: #8f5902;
+  font-weight: bold;
+  font-style: italic;
+}
+
+.prevent-inlining {
+  content: "</";
+}
+
+/*# sourceMappingURL=df2e132ae23d8dbf2594fe4620073e17.css.map */
diff --git a/docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css b/docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css
new file mode 100644
index 000000000..c594ad1f0
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css
@@ -0,0 +1,185 @@
+/* quarto syntax highlight colors */
+:root {
+  --quarto-hl-al-color: #bf616a;
+  --quarto-hl-an-color: #d08770;
+  --quarto-hl-at-color: #8fbcbb;
+  --quarto-hl-bn-color: #b48ead;
+  --quarto-hl-bu-color: #88c0d0;
+  --quarto-hl-ch-color: #ebcb8b;
+  --quarto-hl-co-color: #616e88;
+  --quarto-hl-cv-color: #e5e9f0;
+  --quarto-hl-cn-color: #eceff4;
+  --quarto-hl-cf-color: #81a1c1;
+  --quarto-hl-dt-color: #81a1c1;
+  --quarto-hl-dv-color: #b48ead;
+  --quarto-hl-do-color: #5e81ac;
+  --quarto-hl-er-color: #bf616a;
+  --quarto-hl-ex-color: #8fbcbb;
+  --quarto-hl-fl-color: #b48ead;
+  --quarto-hl-fu-color: #88c0d0;
+  --quarto-hl-im-color: #a3be8c;
+  --quarto-hl-in-color: #ebcb8b;
+  --quarto-hl-kw-color: #81a1c1;
+  --quarto-hl-op-color: #81a1c1;
+  --quarto-hl-pp-color: #5e81ac;
+  --quarto-hl-re-color: #88c0d0;
+  --quarto-hl-sc-color: #ebcb8b;
+  --quarto-hl-ss-color: #d08770;
+  --quarto-hl-st-color: #a3be8c;
+  --quarto-hl-va-color: #5e81ac;
+  --quarto-hl-vs-color: #a3be8c;
+  --quarto-hl-wa-color: #bf616a;
+}
+
+/* other quarto variables */
+:root {
+  --quarto-font-monospace: SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", "Courier New", monospace;
+}
+
+code span.al {
+  background-color: #3b4252;
+  font-weight: bold;
+  color: #bf616a;
+}
+
+code span.an {
+  color: #d08770;
+}
+
+code span.at {
+  color: #8fbcbb;
+}
+
+code span.bn {
+  color: #b48ead;
+}
+
+code span.bu {
+  font-style: italic;
+  color: #88c0d0;
+}
+
+code span.ch {
+  color: #ebcb8b;
+}
+
+code span.co {
+  color: #616e88;
+}
+
+code span.cv {
+  color: #e5e9f0;
+}
+
+code span.cn {
+  font-weight: bold;
+  color: #eceff4;
+}
+
+code span.cf {
+  font-weight: bold;
+  color: #81a1c1;
+}
+
+code span.dt {
+  color: #81a1c1;
+}
+
+code span.dv {
+  color: #b48ead;
+}
+
+code span.do {
+  color: #5e81ac;
+}
+
+code span.er {
+  color: #bf616a;
+  text-decoration: underline;
+}
+
+code span.ex {
+  font-weight: bold;
+  color: #8fbcbb;
+}
+
+code span.fl {
+  color: #b48ead;
+}
+
+code span.fu {
+  color: #88c0d0;
+}
+
+code span.im {
+  color: #a3be8c;
+}
+
+code span.in {
+  color: #ebcb8b;
+}
+
+code span.kw {
+  font-weight: bold;
+  color: #81a1c1;
+}
+
+pre > code.sourceCode > span {
+  color: #d8dee9;
+}
+
+code span {
+  color: #d8dee9;
+}
+
+code.sourceCode > span {
+  color: #d8dee9;
+}
+
+div.sourceCode,
+div.sourceCode pre.sourceCode {
+  color: #d8dee9;
+}
+
+code span.op {
+  color: #81a1c1;
+}
+
+code span.pp {
+  color: #5e81ac;
+}
+
+code span.re {
+  background-color: #3b4252;
+  color: #88c0d0;
+}
+
+code span.sc {
+  color: #ebcb8b;
+}
+
+code span.ss {
+  color: #d08770;
+}
+
+code span.st {
+  color: #a3be8c;
+}
+
+code span.va {
+  color: #5e81ac;
+}
+
+code span.vs {
+  color: #a3be8c;
+}
+
+code span.wa {
+  color: #bf616a;
+}
+
+.prevent-inlining {
+  content: "</";
+}
+
+/*# sourceMappingURL=0b8ca29711c79dc68f7309230d5ff6a0.css.map */
diff --git a/docs/2_39/site_libs/quarto-html/quarto.js b/docs/2_39/site_libs/quarto-html/quarto.js
new file mode 100644
index 000000000..3c24c295f
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/quarto.js
@@ -0,0 +1,911 @@
+const sectionChanged = new CustomEvent("quarto-sectionChanged", {
+  detail: {},
+  bubbles: true,
+  cancelable: false,
+  composed: false,
+});
+
+const layoutMarginEls = () => {
+  // Find any conflicting margin elements and add margins to the
+  // top to prevent overlap
+  const marginChildren = window.document.querySelectorAll(
+    ".column-margin.column-container > *, .margin-caption, .aside"
+  );
+
+  let lastBottom = 0;
+  for (const marginChild of marginChildren) {
+    if (marginChild.offsetParent !== null) {
+      // clear the top margin so we recompute it
+      marginChild.style.marginTop = null;
+      const top = marginChild.getBoundingClientRect().top + window.scrollY;
+      if (top < lastBottom) {
+        const marginChildStyle = window.getComputedStyle(marginChild);
+        const marginBottom = parseFloat(marginChildStyle["marginBottom"]);
+        const margin = lastBottom - top + marginBottom;
+        marginChild.style.marginTop = `${margin}px`;
+      }
+      const styles = window.getComputedStyle(marginChild);
+      const marginTop = parseFloat(styles["marginTop"]);
+      lastBottom = top + marginChild.getBoundingClientRect().height + marginTop;
+    }
+  }
+};
+
+window.document.addEventListener("DOMContentLoaded", function (_event) {
+  // Recompute the position of margin elements anytime the body size changes
+  if (window.ResizeObserver) {
+    const resizeObserver = new window.ResizeObserver(
+      throttle(() => {
+        layoutMarginEls();
+        if (
+          window.document.body.getBoundingClientRect().width < 990 &&
+          isReaderMode()
+        ) {
+          quartoToggleReader();
+        }
+      }, 50)
+    );
+    resizeObserver.observe(window.document.body);
+  }
+
+  const tocEl = window.document.querySelector('nav.toc-active[role="doc-toc"]');
+  const sidebarEl = window.document.getElementById("quarto-sidebar");
+  const leftTocEl = window.document.getElementById("quarto-sidebar-toc-left");
+  const marginSidebarEl = window.document.getElementById(
+    "quarto-margin-sidebar"
+  );
+  // function to determine whether the element has a previous sibling that is active
+  const prevSiblingIsActiveLink = (el) => {
+    const sibling = el.previousElementSibling;
+    if (sibling && sibling.tagName === "A") {
+      return sibling.classList.contains("active");
+    } else {
+      return false;
+    }
+  };
+
+  // fire slideEnter for bootstrap tab activations (for htmlwidget resize behavior)
+  function fireSlideEnter(e) {
+    const event = window.document.createEvent("Event");
+    event.initEvent("slideenter", true, true);
+    window.document.dispatchEvent(event);
+  }
+  const tabs = window.document.querySelectorAll('a[data-bs-toggle="tab"]');
+  tabs.forEach((tab) => {
+    tab.addEventListener("shown.bs.tab", fireSlideEnter);
+  });
+
+  // fire slideEnter for tabby tab activations (for htmlwidget resize behavior)
+  document.addEventListener("tabby", fireSlideEnter, false);
+
+  // Track scrolling and mark TOC links as active
+  // get table of contents and sidebar (bail if we don't have at least one)
+  const tocLinks = tocEl
+    ? [...tocEl.querySelectorAll("a[data-scroll-target]")]
+    : [];
+  const makeActive = (link) => tocLinks[link].classList.add("active");
+  const removeActive = (link) => tocLinks[link].classList.remove("active");
+  const removeAllActive = () =>
+    [...Array(tocLinks.length).keys()].forEach((link) => removeActive(link));
+
+  // activate the anchor for a section associated with this TOC entry
+  tocLinks.forEach((link) => {
+    link.addEventListener("click", () => {
+      if (link.href.indexOf("#") !== -1) {
+        const anchor = link.href.split("#")[1];
+        const heading = window.document.querySelector(
+          `[data-anchor-id="${anchor}"]`
+        );
+        if (heading) {
+          // Add the class
+          heading.classList.add("reveal-anchorjs-link");
+
+          // function to show the anchor
+          const handleMouseout = () => {
+            heading.classList.remove("reveal-anchorjs-link");
+            heading.removeEventListener("mouseout", handleMouseout);
+          };
+
+          // add a function to clear the anchor when the user mouses out of it
+          heading.addEventListener("mouseout", handleMouseout);
+        }
+      }
+    });
+  });
+
+  const sections = tocLinks.map((link) => {
+    const target = link.getAttribute("data-scroll-target");
+    if (target.startsWith("#")) {
+      return window.document.getElementById(decodeURI(`${target.slice(1)}`));
+    } else {
+      return window.document.querySelector(decodeURI(`${target}`));
+    }
+  });
+
+  const sectionMargin = 200;
+  let currentActive = 0;
+  // track whether we've initialized state the first time
+  let init = false;
+
+  const updateActiveLink = () => {
+    // The index from bottom to top (e.g. reversed list)
+    let sectionIndex = -1;
+    if (
+      window.innerHeight + window.pageYOffset >=
+      window.document.body.offsetHeight
+    ) {
+      // This is the no-scroll case where last section should be the active one
+      sectionIndex = 0;
+    } else {
+      // This finds the last section visible on screen that should be made active
+      sectionIndex = [...sections].reverse().findIndex((section) => {
+        if (section) {
+          return window.pageYOffset >= section.offsetTop - sectionMargin;
+        } else {
+          return false;
+        }
+      });
+    }
+    if (sectionIndex > -1) {
+      const current = sections.length - sectionIndex - 1;
+      if (current !== currentActive) {
+        removeAllActive();
+        currentActive = current;
+        makeActive(current);
+        if (init) {
+          window.dispatchEvent(sectionChanged);
+        }
+        init = true;
+      }
+    }
+  };
+
+  const inHiddenRegion = (top, bottom, hiddenRegions) => {
+    for (const region of hiddenRegions) {
+      if (top <= region.bottom && bottom >= region.top) {
+        return true;
+      }
+    }
+    return false;
+  };
+
+  const categorySelector = "header.quarto-title-block .quarto-category";
+  const activateCategories = (href) => {
+    // Find any categories
+    // Surround them with a link pointing back to:
+    // #category=Authoring
+    try {
+      const categoryEls = window.document.querySelectorAll(categorySelector);
+      for (const categoryEl of categoryEls) {
+        const categoryText = categoryEl.textContent;
+        if (categoryText) {
+          const link = `${href}#category=${encodeURIComponent(categoryText)}`;
+          const linkEl = window.document.createElement("a");
+          linkEl.setAttribute("href", link);
+          for (const child of categoryEl.childNodes) {
+            linkEl.append(child);
+          }
+          categoryEl.appendChild(linkEl);
+        }
+      }
+    } catch {
+      // Ignore errors
+    }
+  };
+  function hasTitleCategories() {
+    return window.document.querySelector(categorySelector) !== null;
+  }
+
+  function offsetRelativeUrl(url) {
+    const offset = getMeta("quarto:offset");
+    return offset ? offset + url : url;
+  }
+
+  function offsetAbsoluteUrl(url) {
+    const offset = getMeta("quarto:offset");
+    const baseUrl = new URL(offset, window.location);
+
+    const projRelativeUrl = url.replace(baseUrl, "");
+    if (projRelativeUrl.startsWith("/")) {
+      return projRelativeUrl;
+    } else {
+      return "/" + projRelativeUrl;
+    }
+  }
+
+  // read a meta tag value
+  function getMeta(metaName) {
+    const metas = window.document.getElementsByTagName("meta");
+    for (let i = 0; i < metas.length; i++) {
+      if (metas[i].getAttribute("name") === metaName) {
+        return metas[i].getAttribute("content");
+      }
+    }
+    return "";
+  }
+
+  async function findAndActivateCategories() {
+    // Categories search with listing only use path without query
+    const currentPagePath = offsetAbsoluteUrl(
+      window.location.origin + window.location.pathname
+    );
+    const response = await fetch(offsetRelativeUrl("listings.json"));
+    if (response.status == 200) {
+      return response.json().then(function (listingPaths) {
+        const listingHrefs = [];
+        for (const listingPath of listingPaths) {
+          const pathWithoutLeadingSlash = listingPath.listing.substring(1);
+          for (const item of listingPath.items) {
+            if (
+              item === currentPagePath ||
+              item === currentPagePath + "index.html"
+            ) {
+              // Resolve this path against the offset to be sure
+              // we already are using the correct path to the listing
+              // (this adjusts the listing urls to be rooted against
+              // whatever root the page is actually running against)
+              const relative = offsetRelativeUrl(pathWithoutLeadingSlash);
+              const baseUrl = window.location;
+              const resolvedPath = new URL(relative, baseUrl);
+              listingHrefs.push(resolvedPath.pathname);
+              break;
+            }
+          }
+        }
+
+        // Look up the tree for a nearby linting and use that if we find one
+        const nearestListing = findNearestParentListing(
+          offsetAbsoluteUrl(window.location.pathname),
+          listingHrefs
+        );
+        if (nearestListing) {
+          activateCategories(nearestListing);
+        } else {
+          // See if the referrer is a listing page for this item
+          const referredRelativePath = offsetAbsoluteUrl(document.referrer);
+          const referrerListing = listingHrefs.find((listingHref) => {
+            const isListingReferrer =
+              listingHref === referredRelativePath ||
+              listingHref === referredRelativePath + "index.html";
+            return isListingReferrer;
+          });
+
+          if (referrerListing) {
+            // Try to use the referrer if possible
+            activateCategories(referrerListing);
+          } else if (listingHrefs.length > 0) {
+            // Otherwise, just fall back to the first listing
+            activateCategories(listingHrefs[0]);
+          }
+        }
+      });
+    }
+  }
+  if (hasTitleCategories()) {
+    findAndActivateCategories();
+  }
+
+  const findNearestParentListing = (href, listingHrefs) => {
+    if (!href || !listingHrefs) {
+      return undefined;
+    }
+    // Look up the tree for a nearby linting and use that if we find one
+    const relativeParts = href.substring(1).split("/");
+    while (relativeParts.length > 0) {
+      const path = relativeParts.join("/");
+      for (const listingHref of listingHrefs) {
+        if (listingHref.startsWith(path)) {
+          return listingHref;
+        }
+      }
+      relativeParts.pop();
+    }
+
+    return undefined;
+  };
+
+  const manageSidebarVisiblity = (el, placeholderDescriptor) => {
+    let isVisible = true;
+    let elRect;
+
+    return (hiddenRegions) => {
+      if (el === null) {
+        return;
+      }
+
+      // Find the last element of the TOC
+      const lastChildEl = el.lastElementChild;
+
+      if (lastChildEl) {
+        // Converts the sidebar to a menu
+        const convertToMenu = () => {
+          for (const child of el.children) {
+            child.style.opacity = 0;
+            child.style.overflow = "hidden";
+            child.style.pointerEvents = "none";
+          }
+
+          nexttick(() => {
+            const toggleContainer = window.document.createElement("div");
+            toggleContainer.style.width = "100%";
+            toggleContainer.classList.add("zindex-over-content");
+            toggleContainer.classList.add("quarto-sidebar-toggle");
+            toggleContainer.classList.add("headroom-target"); // Marks this to be managed by headeroom
+            toggleContainer.id = placeholderDescriptor.id;
+            toggleContainer.style.position = "fixed";
+
+            const toggleIcon = window.document.createElement("i");
+            toggleIcon.classList.add("quarto-sidebar-toggle-icon");
+            toggleIcon.classList.add("bi");
+            toggleIcon.classList.add("bi-caret-down-fill");
+
+            const toggleTitle = window.document.createElement("div");
+            const titleEl = window.document.body.querySelector(
+              placeholderDescriptor.titleSelector
+            );
+            if (titleEl) {
+              toggleTitle.append(
+                titleEl.textContent || titleEl.innerText,
+                toggleIcon
+              );
+            }
+            toggleTitle.classList.add("zindex-over-content");
+            toggleTitle.classList.add("quarto-sidebar-toggle-title");
+            toggleContainer.append(toggleTitle);
+
+            const toggleContents = window.document.createElement("div");
+            toggleContents.classList = el.classList;
+            toggleContents.classList.add("zindex-over-content");
+            toggleContents.classList.add("quarto-sidebar-toggle-contents");
+            for (const child of el.children) {
+              if (child.id === "toc-title") {
+                continue;
+              }
+
+              const clone = child.cloneNode(true);
+              clone.style.opacity = 1;
+              clone.style.pointerEvents = null;
+              clone.style.display = null;
+              toggleContents.append(clone);
+            }
+            toggleContents.style.height = "0px";
+            const positionToggle = () => {
+              // position the element (top left of parent, same width as parent)
+              if (!elRect) {
+                elRect = el.getBoundingClientRect();
+              }
+              toggleContainer.style.left = `${elRect.left}px`;
+              toggleContainer.style.top = `${elRect.top}px`;
+              toggleContainer.style.width = `${elRect.width}px`;
+            };
+            positionToggle();
+
+            toggleContainer.append(toggleContents);
+            el.parentElement.prepend(toggleContainer);
+
+            // Process clicks
+            let tocShowing = false;
+            // Allow the caller to control whether this is dismissed
+            // when it is clicked (e.g. sidebar navigation supports
+            // opening and closing the nav tree, so don't dismiss on click)
+            const clickEl = placeholderDescriptor.dismissOnClick
+              ? toggleContainer
+              : toggleTitle;
+
+            const closeToggle = () => {
+              if (tocShowing) {
+                toggleContainer.classList.remove("expanded");
+                toggleContents.style.height = "0px";
+                tocShowing = false;
+              }
+            };
+
+            // Get rid of any expanded toggle if the user scrolls
+            window.document.addEventListener(
+              "scroll",
+              throttle(() => {
+                closeToggle();
+              }, 50)
+            );
+
+            // Handle positioning of the toggle
+            window.addEventListener(
+              "resize",
+              throttle(() => {
+                elRect = undefined;
+                positionToggle();
+              }, 50)
+            );
+
+            window.addEventListener("quarto-hrChanged", () => {
+              elRect = undefined;
+            });
+
+            // Process the click
+            clickEl.onclick = () => {
+              if (!tocShowing) {
+                toggleContainer.classList.add("expanded");
+                toggleContents.style.height = null;
+                tocShowing = true;
+              } else {
+                closeToggle();
+              }
+            };
+          });
+        };
+
+        // Converts a sidebar from a menu back to a sidebar
+        const convertToSidebar = () => {
+          for (const child of el.children) {
+            child.style.opacity = 1;
+            child.style.overflow = null;
+            child.style.pointerEvents = null;
+          }
+
+          const placeholderEl = window.document.getElementById(
+            placeholderDescriptor.id
+          );
+          if (placeholderEl) {
+            placeholderEl.remove();
+          }
+
+          el.classList.remove("rollup");
+        };
+
+        if (isReaderMode()) {
+          convertToMenu();
+          isVisible = false;
+        } else {
+          // Find the top and bottom o the element that is being managed
+          const elTop = el.offsetTop;
+          const elBottom =
+            elTop + lastChildEl.offsetTop + lastChildEl.offsetHeight;
+
+          if (!isVisible) {
+            // If the element is current not visible reveal if there are
+            // no conflicts with overlay regions
+            if (!inHiddenRegion(elTop, elBottom, hiddenRegions)) {
+              convertToSidebar();
+              isVisible = true;
+            }
+          } else {
+            // If the element is visible, hide it if it conflicts with overlay regions
+            // and insert a placeholder toggle (or if we're in reader mode)
+            if (inHiddenRegion(elTop, elBottom, hiddenRegions)) {
+              convertToMenu();
+              isVisible = false;
+            }
+          }
+        }
+      }
+    };
+  };
+
+  const tabEls = document.querySelectorAll('a[data-bs-toggle="tab"]');
+  for (const tabEl of tabEls) {
+    const id = tabEl.getAttribute("data-bs-target");
+    if (id) {
+      const columnEl = document.querySelector(
+        `${id} .column-margin, .tabset-margin-content`
+      );
+      if (columnEl)
+        tabEl.addEventListener("shown.bs.tab", function (event) {
+          const el = event.srcElement;
+          if (el) {
+            const visibleCls = `${el.id}-margin-content`;
+            // walk up until we find a parent tabset
+            let panelTabsetEl = el.parentElement;
+            while (panelTabsetEl) {
+              if (panelTabsetEl.classList.contains("panel-tabset")) {
+                break;
+              }
+              panelTabsetEl = panelTabsetEl.parentElement;
+            }
+
+            if (panelTabsetEl) {
+              const prevSib = panelTabsetEl.previousElementSibling;
+              if (
+                prevSib &&
+                prevSib.classList.contains("tabset-margin-container")
+              ) {
+                const childNodes = prevSib.querySelectorAll(
+                  ".tabset-margin-content"
+                );
+                for (const childEl of childNodes) {
+                  if (childEl.classList.contains(visibleCls)) {
+                    childEl.classList.remove("collapse");
+                  } else {
+                    childEl.classList.add("collapse");
+                  }
+                }
+              }
+            }
+          }
+
+          layoutMarginEls();
+        });
+    }
+  }
+
+  // Manage the visibility of the toc and the sidebar
+  const marginScrollVisibility = manageSidebarVisiblity(marginSidebarEl, {
+    id: "quarto-toc-toggle",
+    titleSelector: "#toc-title",
+    dismissOnClick: true,
+  });
+  const sidebarScrollVisiblity = manageSidebarVisiblity(sidebarEl, {
+    id: "quarto-sidebarnav-toggle",
+    titleSelector: ".title",
+    dismissOnClick: false,
+  });
+  let tocLeftScrollVisibility;
+  if (leftTocEl) {
+    tocLeftScrollVisibility = manageSidebarVisiblity(leftTocEl, {
+      id: "quarto-lefttoc-toggle",
+      titleSelector: "#toc-title",
+      dismissOnClick: true,
+    });
+  }
+
+  // Find the first element that uses formatting in special columns
+  const conflictingEls = window.document.body.querySelectorAll(
+    '[class^="column-"], [class*=" column-"], aside, [class*="margin-caption"], [class*=" margin-caption"], [class*="margin-ref"], [class*=" margin-ref"]'
+  );
+
+  // Filter all the possibly conflicting elements into ones
+  // the do conflict on the left or ride side
+  const arrConflictingEls = Array.from(conflictingEls);
+  const leftSideConflictEls = arrConflictingEls.filter((el) => {
+    if (el.tagName === "ASIDE") {
+      return false;
+    }
+    return Array.from(el.classList).find((className) => {
+      return (
+        className !== "column-body" &&
+        className.startsWith("column-") &&
+        !className.endsWith("right") &&
+        !className.endsWith("container") &&
+        className !== "column-margin"
+      );
+    });
+  });
+  const rightSideConflictEls = arrConflictingEls.filter((el) => {
+    if (el.tagName === "ASIDE") {
+      return true;
+    }
+
+    const hasMarginCaption = Array.from(el.classList).find((className) => {
+      return className == "margin-caption";
+    });
+    if (hasMarginCaption) {
+      return true;
+    }
+
+    return Array.from(el.classList).find((className) => {
+      return (
+        className !== "column-body" &&
+        !className.endsWith("container") &&
+        className.startsWith("column-") &&
+        !className.endsWith("left")
+      );
+    });
+  });
+
+  const kOverlapPaddingSize = 10;
+  function toRegions(els) {
+    return els.map((el) => {
+      const boundRect = el.getBoundingClientRect();
+      const top =
+        boundRect.top +
+        document.documentElement.scrollTop -
+        kOverlapPaddingSize;
+      return {
+        top,
+        bottom: top + el.scrollHeight + 2 * kOverlapPaddingSize,
+      };
+    });
+  }
+
+  let hasObserved = false;
+  const visibleItemObserver = (els) => {
+    let visibleElements = [...els];
+    const intersectionObserver = new IntersectionObserver(
+      (entries, _observer) => {
+        entries.forEach((entry) => {
+          if (entry.isIntersecting) {
+            if (visibleElements.indexOf(entry.target) === -1) {
+              visibleElements.push(entry.target);
+            }
+          } else {
+            visibleElements = visibleElements.filter((visibleEntry) => {
+              return visibleEntry !== entry;
+            });
+          }
+        });
+
+        if (!hasObserved) {
+          hideOverlappedSidebars();
+        }
+        hasObserved = true;
+      },
+      {}
+    );
+    els.forEach((el) => {
+      intersectionObserver.observe(el);
+    });
+
+    return {
+      getVisibleEntries: () => {
+        return visibleElements;
+      },
+    };
+  };
+
+  const rightElementObserver = visibleItemObserver(rightSideConflictEls);
+  const leftElementObserver = visibleItemObserver(leftSideConflictEls);
+
+  const hideOverlappedSidebars = () => {
+    marginScrollVisibility(toRegions(rightElementObserver.getVisibleEntries()));
+    sidebarScrollVisiblity(toRegions(leftElementObserver.getVisibleEntries()));
+    if (tocLeftScrollVisibility) {
+      tocLeftScrollVisibility(
+        toRegions(leftElementObserver.getVisibleEntries())
+      );
+    }
+  };
+
+  window.quartoToggleReader = () => {
+    // Applies a slow class (or removes it)
+    // to update the transition speed
+    const slowTransition = (slow) => {
+      const manageTransition = (id, slow) => {
+        const el = document.getElementById(id);
+        if (el) {
+          if (slow) {
+            el.classList.add("slow");
+          } else {
+            el.classList.remove("slow");
+          }
+        }
+      };
+
+      manageTransition("TOC", slow);
+      manageTransition("quarto-sidebar", slow);
+    };
+    const readerMode = !isReaderMode();
+    setReaderModeValue(readerMode);
+
+    // If we're entering reader mode, slow the transition
+    if (readerMode) {
+      slowTransition(readerMode);
+    }
+    highlightReaderToggle(readerMode);
+    hideOverlappedSidebars();
+
+    // If we're exiting reader mode, restore the non-slow transition
+    if (!readerMode) {
+      slowTransition(!readerMode);
+    }
+  };
+
+  const highlightReaderToggle = (readerMode) => {
+    const els = document.querySelectorAll(".quarto-reader-toggle");
+    if (els) {
+      els.forEach((el) => {
+        if (readerMode) {
+          el.classList.add("reader");
+        } else {
+          el.classList.remove("reader");
+        }
+      });
+    }
+  };
+
+  const setReaderModeValue = (val) => {
+    if (window.location.protocol !== "file:") {
+      window.localStorage.setItem("quarto-reader-mode", val);
+    } else {
+      localReaderMode = val;
+    }
+  };
+
+  const isReaderMode = () => {
+    if (window.location.protocol !== "file:") {
+      return window.localStorage.getItem("quarto-reader-mode") === "true";
+    } else {
+      return localReaderMode;
+    }
+  };
+  let localReaderMode = null;
+
+  const tocOpenDepthStr = tocEl?.getAttribute("data-toc-expanded");
+  const tocOpenDepth = tocOpenDepthStr ? Number(tocOpenDepthStr) : 1;
+
+  // Walk the TOC and collapse/expand nodes
+  // Nodes are expanded if:
+  // - they are top level
+  // - they have children that are 'active' links
+  // - they are directly below an link that is 'active'
+  const walk = (el, depth) => {
+    // Tick depth when we enter a UL
+    if (el.tagName === "UL") {
+      depth = depth + 1;
+    }
+
+    // It this is active link
+    let isActiveNode = false;
+    if (el.tagName === "A" && el.classList.contains("active")) {
+      isActiveNode = true;
+    }
+
+    // See if there is an active child to this element
+    let hasActiveChild = false;
+    for (child of el.children) {
+      hasActiveChild = walk(child, depth) || hasActiveChild;
+    }
+
+    // Process the collapse state if this is an UL
+    if (el.tagName === "UL") {
+      if (tocOpenDepth === -1 && depth > 1) {
+        // toc-expand: false
+        el.classList.add("collapse");
+      } else if (
+        depth <= tocOpenDepth ||
+        hasActiveChild ||
+        prevSiblingIsActiveLink(el)
+      ) {
+        el.classList.remove("collapse");
+      } else {
+        el.classList.add("collapse");
+      }
+
+      // untick depth when we leave a UL
+      depth = depth - 1;
+    }
+    return hasActiveChild || isActiveNode;
+  };
+
+  // walk the TOC and expand / collapse any items that should be shown
+  if (tocEl) {
+    updateActiveLink();
+    walk(tocEl, 0);
+  }
+
+  // Throttle the scroll event and walk peridiocally
+  window.document.addEventListener(
+    "scroll",
+    throttle(() => {
+      if (tocEl) {
+        updateActiveLink();
+        walk(tocEl, 0);
+      }
+      if (!isReaderMode()) {
+        hideOverlappedSidebars();
+      }
+    }, 5)
+  );
+  window.addEventListener(
+    "resize",
+    throttle(() => {
+      if (tocEl) {
+        updateActiveLink();
+        walk(tocEl, 0);
+      }
+      if (!isReaderMode()) {
+        hideOverlappedSidebars();
+      }
+    }, 10)
+  );
+  hideOverlappedSidebars();
+  highlightReaderToggle(isReaderMode());
+});
+
+// grouped tabsets
+window.addEventListener("pageshow", (_event) => {
+  function getTabSettings() {
+    const data = localStorage.getItem("quarto-persistent-tabsets-data");
+    if (!data) {
+      localStorage.setItem("quarto-persistent-tabsets-data", "{}");
+      return {};
+    }
+    if (data) {
+      return JSON.parse(data);
+    }
+  }
+
+  function setTabSettings(data) {
+    localStorage.setItem(
+      "quarto-persistent-tabsets-data",
+      JSON.stringify(data)
+    );
+  }
+
+  function setTabState(groupName, groupValue) {
+    const data = getTabSettings();
+    data[groupName] = groupValue;
+    setTabSettings(data);
+  }
+
+  function toggleTab(tab, active) {
+    const tabPanelId = tab.getAttribute("aria-controls");
+    const tabPanel = document.getElementById(tabPanelId);
+    if (active) {
+      tab.classList.add("active");
+      tabPanel.classList.add("active");
+    } else {
+      tab.classList.remove("active");
+      tabPanel.classList.remove("active");
+    }
+  }
+
+  function toggleAll(selectedGroup, selectorsToSync) {
+    for (const [thisGroup, tabs] of Object.entries(selectorsToSync)) {
+      const active = selectedGroup === thisGroup;
+      for (const tab of tabs) {
+        toggleTab(tab, active);
+      }
+    }
+  }
+
+  function findSelectorsToSyncByLanguage() {
+    const result = {};
+    const tabs = Array.from(
+      document.querySelectorAll(`div[data-group] a[id^='tabset-']`)
+    );
+    for (const item of tabs) {
+      const div = item.parentElement.parentElement.parentElement;
+      const group = div.getAttribute("data-group");
+      if (!result[group]) {
+        result[group] = {};
+      }
+      const selectorsToSync = result[group];
+      const value = item.innerHTML;
+      if (!selectorsToSync[value]) {
+        selectorsToSync[value] = [];
+      }
+      selectorsToSync[value].push(item);
+    }
+    return result;
+  }
+
+  function setupSelectorSync() {
+    const selectorsToSync = findSelectorsToSyncByLanguage();
+    Object.entries(selectorsToSync).forEach(([group, tabSetsByValue]) => {
+      Object.entries(tabSetsByValue).forEach(([value, items]) => {
+        items.forEach((item) => {
+          item.addEventListener("click", (_event) => {
+            setTabState(group, value);
+            toggleAll(value, selectorsToSync[group]);
+          });
+        });
+      });
+    });
+    return selectorsToSync;
+  }
+
+  const selectorsToSync = setupSelectorSync();
+  for (const [group, selectedName] of Object.entries(getTabSettings())) {
+    const selectors = selectorsToSync[group];
+    // it's possible that stale state gives us empty selections, so we explicitly check here.
+    if (selectors) {
+      toggleAll(selectedName, selectors);
+    }
+  }
+});
+
+function throttle(func, wait) {
+  let waiting = false;
+  return function () {
+    if (!waiting) {
+      func.apply(this, arguments);
+      waiting = true;
+      setTimeout(function () {
+        waiting = false;
+      }, wait);
+    }
+  };
+}
+
+function nexttick(func) {
+  return setTimeout(func, 0);
+}
diff --git a/docs/2_39/site_libs/quarto-html/tippy.css b/docs/2_39/site_libs/quarto-html/tippy.css
new file mode 100644
index 000000000..e6ae635cb
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/tippy.css
@@ -0,0 +1 @@
+.tippy-box[data-animation=fade][data-state=hidden]{opacity:0}[data-tippy-root]{max-width:calc(100vw - 10px)}.tippy-box{position:relative;background-color:#333;color:#fff;border-radius:4px;font-size:14px;line-height:1.4;white-space:normal;outline:0;transition-property:transform,visibility,opacity}.tippy-box[data-placement^=top]>.tippy-arrow{bottom:0}.tippy-box[data-placement^=top]>.tippy-arrow:before{bottom:-7px;left:0;border-width:8px 8px 0;border-top-color:initial;transform-origin:center top}.tippy-box[data-placement^=bottom]>.tippy-arrow{top:0}.tippy-box[data-placement^=bottom]>.tippy-arrow:before{top:-7px;left:0;border-width:0 8px 8px;border-bottom-color:initial;transform-origin:center bottom}.tippy-box[data-placement^=left]>.tippy-arrow{right:0}.tippy-box[data-placement^=left]>.tippy-arrow:before{border-width:8px 0 8px 8px;border-left-color:initial;right:-7px;transform-origin:center left}.tippy-box[data-placement^=right]>.tippy-arrow{left:0}.tippy-box[data-placement^=right]>.tippy-arrow:before{left:-7px;border-width:8px 8px 8px 0;border-right-color:initial;transform-origin:center right}.tippy-box[data-inertia][data-state=visible]{transition-timing-function:cubic-bezier(.54,1.5,.38,1.11)}.tippy-arrow{width:16px;height:16px;color:#333}.tippy-arrow:before{content:"";position:absolute;border-color:transparent;border-style:solid}.tippy-content{position:relative;padding:5px 9px;z-index:1}
\ No newline at end of file
diff --git a/docs/2_39/site_libs/quarto-html/tippy.umd.min.js b/docs/2_39/site_libs/quarto-html/tippy.umd.min.js
new file mode 100644
index 000000000..ca292be32
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-html/tippy.umd.min.js
@@ -0,0 +1,2 @@
+!function(e,t){"object"==typeof exports&&"undefined"!=typeof module?module.exports=t(require("@popperjs/core")):"function"==typeof define&&define.amd?define(["@popperjs/core"],t):(e=e||self).tippy=t(e.Popper)}(this,(function(e){"use strict";var t={passive:!0,capture:!0},n=function(){return document.body};function r(e,t,n){if(Array.isArray(e)){var r=e[t];return null==r?Array.isArray(n)?n[t]:n:r}return e}function o(e,t){var n={}.toString.call(e);return 0===n.indexOf("[object")&&n.indexOf(t+"]")>-1}function i(e,t){return"function"==typeof e?e.apply(void 0,t):e}function a(e,t){return 0===t?e:function(r){clearTimeout(n),n=setTimeout((function(){e(r)}),t)};var n}function s(e,t){var n=Object.assign({},e);return t.forEach((function(e){delete n[e]})),n}function u(e){return[].concat(e)}function c(e,t){-1===e.indexOf(t)&&e.push(t)}function p(e){return e.split("-")[0]}function f(e){return[].slice.call(e)}function l(e){return Object.keys(e).reduce((function(t,n){return void 0!==e[n]&&(t[n]=e[n]),t}),{})}function d(){return document.createElement("div")}function v(e){return["Element","Fragment"].some((function(t){return o(e,t)}))}function m(e){return o(e,"MouseEvent")}function g(e){return!(!e||!e._tippy||e._tippy.reference!==e)}function h(e){return v(e)?[e]:function(e){return o(e,"NodeList")}(e)?f(e):Array.isArray(e)?e:f(document.querySelectorAll(e))}function b(e,t){e.forEach((function(e){e&&(e.style.transitionDuration=t+"ms")}))}function y(e,t){e.forEach((function(e){e&&e.setAttribute("data-state",t)}))}function w(e){var t,n=u(e)[0];return null!=n&&null!=(t=n.ownerDocument)&&t.body?n.ownerDocument:document}function E(e,t,n){var r=t+"EventListener";["transitionend","webkitTransitionEnd"].forEach((function(t){e[r](t,n)}))}function O(e,t){for(var n=t;n;){var r;if(e.contains(n))return!0;n=null==n.getRootNode||null==(r=n.getRootNode())?void 0:r.host}return!1}var x={isTouch:!1},C=0;function T(){x.isTouch||(x.isTouch=!0,window.performance&&document.addEventListener("mousemove",A))}function A(){var e=performance.now();e-C<20&&(x.isTouch=!1,document.removeEventListener("mousemove",A)),C=e}function L(){var e=document.activeElement;if(g(e)){var t=e._tippy;e.blur&&!t.state.isVisible&&e.blur()}}var D=!!("undefined"!=typeof window&&"undefined"!=typeof document)&&!!window.msCrypto,R=Object.assign({appendTo:n,aria:{content:"auto",expanded:"auto"},delay:0,duration:[300,250],getReferenceClientRect:null,hideOnClick:!0,ignoreAttributes:!1,interactive:!1,interactiveBorder:2,interactiveDebounce:0,moveTransition:"",offset:[0,10],onAfterUpdate:function(){},onBeforeUpdate:function(){},onCreate:function(){},onDestroy:function(){},onHidden:function(){},onHide:function(){},onMount:function(){},onShow:function(){},onShown:function(){},onTrigger:function(){},onUntrigger:function(){},onClickOutside:function(){},placement:"top",plugins:[],popperOptions:{},render:null,showOnCreate:!1,touch:!0,trigger:"mouseenter focus",triggerTarget:null},{animateFill:!1,followCursor:!1,inlinePositioning:!1,sticky:!1},{allowHTML:!1,animation:"fade",arrow:!0,content:"",inertia:!1,maxWidth:350,role:"tooltip",theme:"",zIndex:9999}),k=Object.keys(R);function P(e){var t=(e.plugins||[]).reduce((function(t,n){var r,o=n.name,i=n.defaultValue;o&&(t[o]=void 0!==e[o]?e[o]:null!=(r=R[o])?r:i);return t}),{});return Object.assign({},e,t)}function j(e,t){var n=Object.assign({},t,{content:i(t.content,[e])},t.ignoreAttributes?{}:function(e,t){return(t?Object.keys(P(Object.assign({},R,{plugins:t}))):k).reduce((function(t,n){var r=(e.getAttribute("data-tippy-"+n)||"").trim();if(!r)return t;if("content"===n)t[n]=r;else try{t[n]=JSON.parse(r)}catch(e){t[n]=r}return t}),{})}(e,t.plugins));return n.aria=Object.assign({},R.aria,n.aria),n.aria={expanded:"auto"===n.aria.expanded?t.interactive:n.aria.expanded,content:"auto"===n.aria.content?t.interactive?null:"describedby":n.aria.content},n}function M(e,t){e.innerHTML=t}function V(e){var t=d();return!0===e?t.className="tippy-arrow":(t.className="tippy-svg-arrow",v(e)?t.appendChild(e):M(t,e)),t}function I(e,t){v(t.content)?(M(e,""),e.appendChild(t.content)):"function"!=typeof t.content&&(t.allowHTML?M(e,t.content):e.textContent=t.content)}function S(e){var t=e.firstElementChild,n=f(t.children);return{box:t,content:n.find((function(e){return e.classList.contains("tippy-content")})),arrow:n.find((function(e){return e.classList.contains("tippy-arrow")||e.classList.contains("tippy-svg-arrow")})),backdrop:n.find((function(e){return e.classList.contains("tippy-backdrop")}))}}function N(e){var t=d(),n=d();n.className="tippy-box",n.setAttribute("data-state","hidden"),n.setAttribute("tabindex","-1");var r=d();function o(n,r){var o=S(t),i=o.box,a=o.content,s=o.arrow;r.theme?i.setAttribute("data-theme",r.theme):i.removeAttribute("data-theme"),"string"==typeof r.animation?i.setAttribute("data-animation",r.animation):i.removeAttribute("data-animation"),r.inertia?i.setAttribute("data-inertia",""):i.removeAttribute("data-inertia"),i.style.maxWidth="number"==typeof r.maxWidth?r.maxWidth+"px":r.maxWidth,r.role?i.setAttribute("role",r.role):i.removeAttribute("role"),n.content===r.content&&n.allowHTML===r.allowHTML||I(a,e.props),r.arrow?s?n.arrow!==r.arrow&&(i.removeChild(s),i.appendChild(V(r.arrow))):i.appendChild(V(r.arrow)):s&&i.removeChild(s)}return r.className="tippy-content",r.setAttribute("data-state","hidden"),I(r,e.props),t.appendChild(n),n.appendChild(r),o(e.props,e.props),{popper:t,onUpdate:o}}N.$$tippy=!0;var B=1,H=[],U=[];function _(o,s){var v,g,h,C,T,A,L,k,M=j(o,Object.assign({},R,P(l(s)))),V=!1,I=!1,N=!1,_=!1,F=[],W=a(we,M.interactiveDebounce),X=B++,Y=(k=M.plugins).filter((function(e,t){return k.indexOf(e)===t})),$={id:X,reference:o,popper:d(),popperInstance:null,props:M,state:{isEnabled:!0,isVisible:!1,isDestroyed:!1,isMounted:!1,isShown:!1},plugins:Y,clearDelayTimeouts:function(){clearTimeout(v),clearTimeout(g),cancelAnimationFrame(h)},setProps:function(e){if($.state.isDestroyed)return;ae("onBeforeUpdate",[$,e]),be();var t=$.props,n=j(o,Object.assign({},t,l(e),{ignoreAttributes:!0}));$.props=n,he(),t.interactiveDebounce!==n.interactiveDebounce&&(ce(),W=a(we,n.interactiveDebounce));t.triggerTarget&&!n.triggerTarget?u(t.triggerTarget).forEach((function(e){e.removeAttribute("aria-expanded")})):n.triggerTarget&&o.removeAttribute("aria-expanded");ue(),ie(),J&&J(t,n);$.popperInstance&&(Ce(),Ae().forEach((function(e){requestAnimationFrame(e._tippy.popperInstance.forceUpdate)})));ae("onAfterUpdate",[$,e])},setContent:function(e){$.setProps({content:e})},show:function(){var e=$.state.isVisible,t=$.state.isDestroyed,o=!$.state.isEnabled,a=x.isTouch&&!$.props.touch,s=r($.props.duration,0,R.duration);if(e||t||o||a)return;if(te().hasAttribute("disabled"))return;if(ae("onShow",[$],!1),!1===$.props.onShow($))return;$.state.isVisible=!0,ee()&&(z.style.visibility="visible");ie(),de(),$.state.isMounted||(z.style.transition="none");if(ee()){var u=re(),p=u.box,f=u.content;b([p,f],0)}A=function(){var e;if($.state.isVisible&&!_){if(_=!0,z.offsetHeight,z.style.transition=$.props.moveTransition,ee()&&$.props.animation){var t=re(),n=t.box,r=t.content;b([n,r],s),y([n,r],"visible")}se(),ue(),c(U,$),null==(e=$.popperInstance)||e.forceUpdate(),ae("onMount",[$]),$.props.animation&&ee()&&function(e,t){me(e,t)}(s,(function(){$.state.isShown=!0,ae("onShown",[$])}))}},function(){var e,t=$.props.appendTo,r=te();e=$.props.interactive&&t===n||"parent"===t?r.parentNode:i(t,[r]);e.contains(z)||e.appendChild(z);$.state.isMounted=!0,Ce()}()},hide:function(){var e=!$.state.isVisible,t=$.state.isDestroyed,n=!$.state.isEnabled,o=r($.props.duration,1,R.duration);if(e||t||n)return;if(ae("onHide",[$],!1),!1===$.props.onHide($))return;$.state.isVisible=!1,$.state.isShown=!1,_=!1,V=!1,ee()&&(z.style.visibility="hidden");if(ce(),ve(),ie(!0),ee()){var i=re(),a=i.box,s=i.content;$.props.animation&&(b([a,s],o),y([a,s],"hidden"))}se(),ue(),$.props.animation?ee()&&function(e,t){me(e,(function(){!$.state.isVisible&&z.parentNode&&z.parentNode.contains(z)&&t()}))}(o,$.unmount):$.unmount()},hideWithInteractivity:function(e){ne().addEventListener("mousemove",W),c(H,W),W(e)},enable:function(){$.state.isEnabled=!0},disable:function(){$.hide(),$.state.isEnabled=!1},unmount:function(){$.state.isVisible&&$.hide();if(!$.state.isMounted)return;Te(),Ae().forEach((function(e){e._tippy.unmount()})),z.parentNode&&z.parentNode.removeChild(z);U=U.filter((function(e){return e!==$})),$.state.isMounted=!1,ae("onHidden",[$])},destroy:function(){if($.state.isDestroyed)return;$.clearDelayTimeouts(),$.unmount(),be(),delete o._tippy,$.state.isDestroyed=!0,ae("onDestroy",[$])}};if(!M.render)return $;var q=M.render($),z=q.popper,J=q.onUpdate;z.setAttribute("data-tippy-root",""),z.id="tippy-"+$.id,$.popper=z,o._tippy=$,z._tippy=$;var G=Y.map((function(e){return e.fn($)})),K=o.hasAttribute("aria-expanded");return he(),ue(),ie(),ae("onCreate",[$]),M.showOnCreate&&Le(),z.addEventListener("mouseenter",(function(){$.props.interactive&&$.state.isVisible&&$.clearDelayTimeouts()})),z.addEventListener("mouseleave",(function(){$.props.interactive&&$.props.trigger.indexOf("mouseenter")>=0&&ne().addEventListener("mousemove",W)})),$;function Q(){var e=$.props.touch;return Array.isArray(e)?e:[e,0]}function Z(){return"hold"===Q()[0]}function ee(){var e;return!(null==(e=$.props.render)||!e.$$tippy)}function te(){return L||o}function ne(){var e=te().parentNode;return e?w(e):document}function re(){return S(z)}function oe(e){return $.state.isMounted&&!$.state.isVisible||x.isTouch||C&&"focus"===C.type?0:r($.props.delay,e?0:1,R.delay)}function ie(e){void 0===e&&(e=!1),z.style.pointerEvents=$.props.interactive&&!e?"":"none",z.style.zIndex=""+$.props.zIndex}function ae(e,t,n){var r;(void 0===n&&(n=!0),G.forEach((function(n){n[e]&&n[e].apply(n,t)})),n)&&(r=$.props)[e].apply(r,t)}function se(){var e=$.props.aria;if(e.content){var t="aria-"+e.content,n=z.id;u($.props.triggerTarget||o).forEach((function(e){var r=e.getAttribute(t);if($.state.isVisible)e.setAttribute(t,r?r+" "+n:n);else{var o=r&&r.replace(n,"").trim();o?e.setAttribute(t,o):e.removeAttribute(t)}}))}}function ue(){!K&&$.props.aria.expanded&&u($.props.triggerTarget||o).forEach((function(e){$.props.interactive?e.setAttribute("aria-expanded",$.state.isVisible&&e===te()?"true":"false"):e.removeAttribute("aria-expanded")}))}function ce(){ne().removeEventListener("mousemove",W),H=H.filter((function(e){return e!==W}))}function pe(e){if(!x.isTouch||!N&&"mousedown"!==e.type){var t=e.composedPath&&e.composedPath()[0]||e.target;if(!$.props.interactive||!O(z,t)){if(u($.props.triggerTarget||o).some((function(e){return O(e,t)}))){if(x.isTouch)return;if($.state.isVisible&&$.props.trigger.indexOf("click")>=0)return}else ae("onClickOutside",[$,e]);!0===$.props.hideOnClick&&($.clearDelayTimeouts(),$.hide(),I=!0,setTimeout((function(){I=!1})),$.state.isMounted||ve())}}}function fe(){N=!0}function le(){N=!1}function de(){var e=ne();e.addEventListener("mousedown",pe,!0),e.addEventListener("touchend",pe,t),e.addEventListener("touchstart",le,t),e.addEventListener("touchmove",fe,t)}function ve(){var e=ne();e.removeEventListener("mousedown",pe,!0),e.removeEventListener("touchend",pe,t),e.removeEventListener("touchstart",le,t),e.removeEventListener("touchmove",fe,t)}function me(e,t){var n=re().box;function r(e){e.target===n&&(E(n,"remove",r),t())}if(0===e)return t();E(n,"remove",T),E(n,"add",r),T=r}function ge(e,t,n){void 0===n&&(n=!1),u($.props.triggerTarget||o).forEach((function(r){r.addEventListener(e,t,n),F.push({node:r,eventType:e,handler:t,options:n})}))}function he(){var e;Z()&&(ge("touchstart",ye,{passive:!0}),ge("touchend",Ee,{passive:!0})),(e=$.props.trigger,e.split(/\s+/).filter(Boolean)).forEach((function(e){if("manual"!==e)switch(ge(e,ye),e){case"mouseenter":ge("mouseleave",Ee);break;case"focus":ge(D?"focusout":"blur",Oe);break;case"focusin":ge("focusout",Oe)}}))}function be(){F.forEach((function(e){var t=e.node,n=e.eventType,r=e.handler,o=e.options;t.removeEventListener(n,r,o)})),F=[]}function ye(e){var t,n=!1;if($.state.isEnabled&&!xe(e)&&!I){var r="focus"===(null==(t=C)?void 0:t.type);C=e,L=e.currentTarget,ue(),!$.state.isVisible&&m(e)&&H.forEach((function(t){return t(e)})),"click"===e.type&&($.props.trigger.indexOf("mouseenter")<0||V)&&!1!==$.props.hideOnClick&&$.state.isVisible?n=!0:Le(e),"click"===e.type&&(V=!n),n&&!r&&De(e)}}function we(e){var t=e.target,n=te().contains(t)||z.contains(t);"mousemove"===e.type&&n||function(e,t){var n=t.clientX,r=t.clientY;return e.every((function(e){var t=e.popperRect,o=e.popperState,i=e.props.interactiveBorder,a=p(o.placement),s=o.modifiersData.offset;if(!s)return!0;var u="bottom"===a?s.top.y:0,c="top"===a?s.bottom.y:0,f="right"===a?s.left.x:0,l="left"===a?s.right.x:0,d=t.top-r+u>i,v=r-t.bottom-c>i,m=t.left-n+f>i,g=n-t.right-l>i;return d||v||m||g}))}(Ae().concat(z).map((function(e){var t,n=null==(t=e._tippy.popperInstance)?void 0:t.state;return n?{popperRect:e.getBoundingClientRect(),popperState:n,props:M}:null})).filter(Boolean),e)&&(ce(),De(e))}function Ee(e){xe(e)||$.props.trigger.indexOf("click")>=0&&V||($.props.interactive?$.hideWithInteractivity(e):De(e))}function Oe(e){$.props.trigger.indexOf("focusin")<0&&e.target!==te()||$.props.interactive&&e.relatedTarget&&z.contains(e.relatedTarget)||De(e)}function xe(e){return!!x.isTouch&&Z()!==e.type.indexOf("touch")>=0}function Ce(){Te();var t=$.props,n=t.popperOptions,r=t.placement,i=t.offset,a=t.getReferenceClientRect,s=t.moveTransition,u=ee()?S(z).arrow:null,c=a?{getBoundingClientRect:a,contextElement:a.contextElement||te()}:o,p=[{name:"offset",options:{offset:i}},{name:"preventOverflow",options:{padding:{top:2,bottom:2,left:5,right:5}}},{name:"flip",options:{padding:5}},{name:"computeStyles",options:{adaptive:!s}},{name:"$$tippy",enabled:!0,phase:"beforeWrite",requires:["computeStyles"],fn:function(e){var t=e.state;if(ee()){var n=re().box;["placement","reference-hidden","escaped"].forEach((function(e){"placement"===e?n.setAttribute("data-placement",t.placement):t.attributes.popper["data-popper-"+e]?n.setAttribute("data-"+e,""):n.removeAttribute("data-"+e)})),t.attributes.popper={}}}}];ee()&&u&&p.push({name:"arrow",options:{element:u,padding:3}}),p.push.apply(p,(null==n?void 0:n.modifiers)||[]),$.popperInstance=e.createPopper(c,z,Object.assign({},n,{placement:r,onFirstUpdate:A,modifiers:p}))}function Te(){$.popperInstance&&($.popperInstance.destroy(),$.popperInstance=null)}function Ae(){return f(z.querySelectorAll("[data-tippy-root]"))}function Le(e){$.clearDelayTimeouts(),e&&ae("onTrigger",[$,e]),de();var t=oe(!0),n=Q(),r=n[0],o=n[1];x.isTouch&&"hold"===r&&o&&(t=o),t?v=setTimeout((function(){$.show()}),t):$.show()}function De(e){if($.clearDelayTimeouts(),ae("onUntrigger",[$,e]),$.state.isVisible){if(!($.props.trigger.indexOf("mouseenter")>=0&&$.props.trigger.indexOf("click")>=0&&["mouseleave","mousemove"].indexOf(e.type)>=0&&V)){var t=oe(!1);t?g=setTimeout((function(){$.state.isVisible&&$.hide()}),t):h=requestAnimationFrame((function(){$.hide()}))}}else ve()}}function F(e,n){void 0===n&&(n={});var r=R.plugins.concat(n.plugins||[]);document.addEventListener("touchstart",T,t),window.addEventListener("blur",L);var o=Object.assign({},n,{plugins:r}),i=h(e).reduce((function(e,t){var n=t&&_(t,o);return n&&e.push(n),e}),[]);return v(e)?i[0]:i}F.defaultProps=R,F.setDefaultProps=function(e){Object.keys(e).forEach((function(t){R[t]=e[t]}))},F.currentInput=x;var W=Object.assign({},e.applyStyles,{effect:function(e){var t=e.state,n={popper:{position:t.options.strategy,left:"0",top:"0",margin:"0"},arrow:{position:"absolute"},reference:{}};Object.assign(t.elements.popper.style,n.popper),t.styles=n,t.elements.arrow&&Object.assign(t.elements.arrow.style,n.arrow)}}),X={mouseover:"mouseenter",focusin:"focus",click:"click"};var Y={name:"animateFill",defaultValue:!1,fn:function(e){var t;if(null==(t=e.props.render)||!t.$$tippy)return{};var n=S(e.popper),r=n.box,o=n.content,i=e.props.animateFill?function(){var e=d();return e.className="tippy-backdrop",y([e],"hidden"),e}():null;return{onCreate:function(){i&&(r.insertBefore(i,r.firstElementChild),r.setAttribute("data-animatefill",""),r.style.overflow="hidden",e.setProps({arrow:!1,animation:"shift-away"}))},onMount:function(){if(i){var e=r.style.transitionDuration,t=Number(e.replace("ms",""));o.style.transitionDelay=Math.round(t/10)+"ms",i.style.transitionDuration=e,y([i],"visible")}},onShow:function(){i&&(i.style.transitionDuration="0ms")},onHide:function(){i&&y([i],"hidden")}}}};var $={clientX:0,clientY:0},q=[];function z(e){var t=e.clientX,n=e.clientY;$={clientX:t,clientY:n}}var J={name:"followCursor",defaultValue:!1,fn:function(e){var t=e.reference,n=w(e.props.triggerTarget||t),r=!1,o=!1,i=!0,a=e.props;function s(){return"initial"===e.props.followCursor&&e.state.isVisible}function u(){n.addEventListener("mousemove",f)}function c(){n.removeEventListener("mousemove",f)}function p(){r=!0,e.setProps({getReferenceClientRect:null}),r=!1}function f(n){var r=!n.target||t.contains(n.target),o=e.props.followCursor,i=n.clientX,a=n.clientY,s=t.getBoundingClientRect(),u=i-s.left,c=a-s.top;!r&&e.props.interactive||e.setProps({getReferenceClientRect:function(){var e=t.getBoundingClientRect(),n=i,r=a;"initial"===o&&(n=e.left+u,r=e.top+c);var s="horizontal"===o?e.top:r,p="vertical"===o?e.right:n,f="horizontal"===o?e.bottom:r,l="vertical"===o?e.left:n;return{width:p-l,height:f-s,top:s,right:p,bottom:f,left:l}}})}function l(){e.props.followCursor&&(q.push({instance:e,doc:n}),function(e){e.addEventListener("mousemove",z)}(n))}function d(){0===(q=q.filter((function(t){return t.instance!==e}))).filter((function(e){return e.doc===n})).length&&function(e){e.removeEventListener("mousemove",z)}(n)}return{onCreate:l,onDestroy:d,onBeforeUpdate:function(){a=e.props},onAfterUpdate:function(t,n){var i=n.followCursor;r||void 0!==i&&a.followCursor!==i&&(d(),i?(l(),!e.state.isMounted||o||s()||u()):(c(),p()))},onMount:function(){e.props.followCursor&&!o&&(i&&(f($),i=!1),s()||u())},onTrigger:function(e,t){m(t)&&($={clientX:t.clientX,clientY:t.clientY}),o="focus"===t.type},onHidden:function(){e.props.followCursor&&(p(),c(),i=!0)}}}};var G={name:"inlinePositioning",defaultValue:!1,fn:function(e){var t,n=e.reference;var r=-1,o=!1,i=[],a={name:"tippyInlinePositioning",enabled:!0,phase:"afterWrite",fn:function(o){var a=o.state;e.props.inlinePositioning&&(-1!==i.indexOf(a.placement)&&(i=[]),t!==a.placement&&-1===i.indexOf(a.placement)&&(i.push(a.placement),e.setProps({getReferenceClientRect:function(){return function(e){return function(e,t,n,r){if(n.length<2||null===e)return t;if(2===n.length&&r>=0&&n[0].left>n[1].right)return n[r]||t;switch(e){case"top":case"bottom":var o=n[0],i=n[n.length-1],a="top"===e,s=o.top,u=i.bottom,c=a?o.left:i.left,p=a?o.right:i.right;return{top:s,bottom:u,left:c,right:p,width:p-c,height:u-s};case"left":case"right":var f=Math.min.apply(Math,n.map((function(e){return e.left}))),l=Math.max.apply(Math,n.map((function(e){return e.right}))),d=n.filter((function(t){return"left"===e?t.left===f:t.right===l})),v=d[0].top,m=d[d.length-1].bottom;return{top:v,bottom:m,left:f,right:l,width:l-f,height:m-v};default:return t}}(p(e),n.getBoundingClientRect(),f(n.getClientRects()),r)}(a.placement)}})),t=a.placement)}};function s(){var t;o||(t=function(e,t){var n;return{popperOptions:Object.assign({},e.popperOptions,{modifiers:[].concat(((null==(n=e.popperOptions)?void 0:n.modifiers)||[]).filter((function(e){return e.name!==t.name})),[t])})}}(e.props,a),o=!0,e.setProps(t),o=!1)}return{onCreate:s,onAfterUpdate:s,onTrigger:function(t,n){if(m(n)){var o=f(e.reference.getClientRects()),i=o.find((function(e){return e.left-2<=n.clientX&&e.right+2>=n.clientX&&e.top-2<=n.clientY&&e.bottom+2>=n.clientY})),a=o.indexOf(i);r=a>-1?a:r}},onHidden:function(){r=-1}}}};var K={name:"sticky",defaultValue:!1,fn:function(e){var t=e.reference,n=e.popper;function r(t){return!0===e.props.sticky||e.props.sticky===t}var o=null,i=null;function a(){var s=r("reference")?(e.popperInstance?e.popperInstance.state.elements.reference:t).getBoundingClientRect():null,u=r("popper")?n.getBoundingClientRect():null;(s&&Q(o,s)||u&&Q(i,u))&&e.popperInstance&&e.popperInstance.update(),o=s,i=u,e.state.isMounted&&requestAnimationFrame(a)}return{onMount:function(){e.props.sticky&&a()}}}};function Q(e,t){return!e||!t||(e.top!==t.top||e.right!==t.right||e.bottom!==t.bottom||e.left!==t.left)}return F.setDefaultProps({plugins:[Y,J,G,K],render:N}),F.createSingleton=function(e,t){var n;void 0===t&&(t={});var r,o=e,i=[],a=[],c=t.overrides,p=[],f=!1;function l(){a=o.map((function(e){return u(e.props.triggerTarget||e.reference)})).reduce((function(e,t){return e.concat(t)}),[])}function v(){i=o.map((function(e){return e.reference}))}function m(e){o.forEach((function(t){e?t.enable():t.disable()}))}function g(e){return o.map((function(t){var n=t.setProps;return t.setProps=function(o){n(o),t.reference===r&&e.setProps(o)},function(){t.setProps=n}}))}function h(e,t){var n=a.indexOf(t);if(t!==r){r=t;var s=(c||[]).concat("content").reduce((function(e,t){return e[t]=o[n].props[t],e}),{});e.setProps(Object.assign({},s,{getReferenceClientRect:"function"==typeof s.getReferenceClientRect?s.getReferenceClientRect:function(){var e;return null==(e=i[n])?void 0:e.getBoundingClientRect()}}))}}m(!1),v(),l();var b={fn:function(){return{onDestroy:function(){m(!0)},onHidden:function(){r=null},onClickOutside:function(e){e.props.showOnCreate&&!f&&(f=!0,r=null)},onShow:function(e){e.props.showOnCreate&&!f&&(f=!0,h(e,i[0]))},onTrigger:function(e,t){h(e,t.currentTarget)}}}},y=F(d(),Object.assign({},s(t,["overrides"]),{plugins:[b].concat(t.plugins||[]),triggerTarget:a,popperOptions:Object.assign({},t.popperOptions,{modifiers:[].concat((null==(n=t.popperOptions)?void 0:n.modifiers)||[],[W])})})),w=y.show;y.show=function(e){if(w(),!r&&null==e)return h(y,i[0]);if(!r||null!=e){if("number"==typeof e)return i[e]&&h(y,i[e]);if(o.indexOf(e)>=0){var t=e.reference;return h(y,t)}return i.indexOf(e)>=0?h(y,e):void 0}},y.showNext=function(){var e=i[0];if(!r)return y.show(0);var t=i.indexOf(r);y.show(i[t+1]||e)},y.showPrevious=function(){var e=i[i.length-1];if(!r)return y.show(e);var t=i.indexOf(r),n=i[t-1]||e;y.show(n)};var E=y.setProps;return y.setProps=function(e){c=e.overrides||c,E(e)},y.setInstances=function(e){m(!0),p.forEach((function(e){return e()})),o=e,m(!1),v(),l(),p=g(y),y.setProps({triggerTarget:a})},p=g(y),y},F.delegate=function(e,n){var r=[],o=[],i=!1,a=n.target,c=s(n,["target"]),p=Object.assign({},c,{trigger:"manual",touch:!1}),f=Object.assign({touch:R.touch},c,{showOnCreate:!0}),l=F(e,p);function d(e){if(e.target&&!i){var t=e.target.closest(a);if(t){var r=t.getAttribute("data-tippy-trigger")||n.trigger||R.trigger;if(!t._tippy&&!("touchstart"===e.type&&"boolean"==typeof f.touch||"touchstart"!==e.type&&r.indexOf(X[e.type])<0)){var s=F(t,f);s&&(o=o.concat(s))}}}}function v(e,t,n,o){void 0===o&&(o=!1),e.addEventListener(t,n,o),r.push({node:e,eventType:t,handler:n,options:o})}return u(l).forEach((function(e){var n=e.destroy,a=e.enable,s=e.disable;e.destroy=function(e){void 0===e&&(e=!0),e&&o.forEach((function(e){e.destroy()})),o=[],r.forEach((function(e){var t=e.node,n=e.eventType,r=e.handler,o=e.options;t.removeEventListener(n,r,o)})),r=[],n()},e.enable=function(){a(),o.forEach((function(e){return e.enable()})),i=!1},e.disable=function(){s(),o.forEach((function(e){return e.disable()})),i=!0},function(e){var n=e.reference;v(n,"touchstart",d,t),v(n,"mouseover",d),v(n,"focusin",d),v(n,"click",d)}(e)})),l},F.hideAll=function(e){var t=void 0===e?{}:e,n=t.exclude,r=t.duration;U.forEach((function(e){var t=!1;if(n&&(t=g(n)?e.reference===n:e.popper===n.popper),!t){var o=e.props.duration;e.setProps({duration:r}),e.hide(),e.state.isDestroyed||e.setProps({duration:o})}}))},F.roundArrow='<svg width="16" height="6" xmlns="http://www.w3.org/2000/svg"><path d="M0 6s1.796-.013 4.67-3.615C5.851.9 6.93.006 8 0c1.07-.006 2.148.887 3.343 2.385C14.233 6.005 16 6 16 6H0z"></svg>',F}));
+
diff --git a/docs/2_39/site_libs/quarto-nav/headroom.min.js b/docs/2_39/site_libs/quarto-nav/headroom.min.js
new file mode 100644
index 000000000..b08f1dffb
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-nav/headroom.min.js
@@ -0,0 +1,7 @@
+/*!
+ * headroom.js v0.12.0 - Give your page some headroom. Hide your header until you need it
+ * Copyright (c) 2020 Nick Williams - http://wicky.nillia.ms/headroom.js
+ * License: MIT
+ */
+
+!function(t,n){"object"==typeof exports&&"undefined"!=typeof module?module.exports=n():"function"==typeof define&&define.amd?define(n):(t=t||self).Headroom=n()}(this,function(){"use strict";function t(){return"undefined"!=typeof window}function d(t){return function(t){return t&&t.document&&function(t){return 9===t.nodeType}(t.document)}(t)?function(t){var n=t.document,o=n.body,s=n.documentElement;return{scrollHeight:function(){return Math.max(o.scrollHeight,s.scrollHeight,o.offsetHeight,s.offsetHeight,o.clientHeight,s.clientHeight)},height:function(){return t.innerHeight||s.clientHeight||o.clientHeight},scrollY:function(){return void 0!==t.pageYOffset?t.pageYOffset:(s||o.parentNode||o).scrollTop}}}(t):function(t){return{scrollHeight:function(){return Math.max(t.scrollHeight,t.offsetHeight,t.clientHeight)},height:function(){return Math.max(t.offsetHeight,t.clientHeight)},scrollY:function(){return t.scrollTop}}}(t)}function n(t,s,e){var n,o=function(){var n=!1;try{var t={get passive(){n=!0}};window.addEventListener("test",t,t),window.removeEventListener("test",t,t)}catch(t){n=!1}return n}(),i=!1,r=d(t),l=r.scrollY(),a={};function c(){var t=Math.round(r.scrollY()),n=r.height(),o=r.scrollHeight();a.scrollY=t,a.lastScrollY=l,a.direction=l<t?"down":"up",a.distance=Math.abs(t-l),a.isOutOfBounds=t<0||o<t+n,a.top=t<=s.offset[a.direction],a.bottom=o<=t+n,a.toleranceExceeded=a.distance>s.tolerance[a.direction],e(a),l=t,i=!1}function h(){i||(i=!0,n=requestAnimationFrame(c))}var u=!!o&&{passive:!0,capture:!1};return t.addEventListener("scroll",h,u),c(),{destroy:function(){cancelAnimationFrame(n),t.removeEventListener("scroll",h,u)}}}function o(t){return t===Object(t)?t:{down:t,up:t}}function s(t,n){n=n||{},Object.assign(this,s.options,n),this.classes=Object.assign({},s.options.classes,n.classes),this.elem=t,this.tolerance=o(this.tolerance),this.offset=o(this.offset),this.initialised=!1,this.frozen=!1}return s.prototype={constructor:s,init:function(){return s.cutsTheMustard&&!this.initialised&&(this.addClass("initial"),this.initialised=!0,setTimeout(function(t){t.scrollTracker=n(t.scroller,{offset:t.offset,tolerance:t.tolerance},t.update.bind(t))},100,this)),this},destroy:function(){this.initialised=!1,Object.keys(this.classes).forEach(this.removeClass,this),this.scrollTracker.destroy()},unpin:function(){!this.hasClass("pinned")&&this.hasClass("unpinned")||(this.addClass("unpinned"),this.removeClass("pinned"),this.onUnpin&&this.onUnpin.call(this))},pin:function(){this.hasClass("unpinned")&&(this.addClass("pinned"),this.removeClass("unpinned"),this.onPin&&this.onPin.call(this))},freeze:function(){this.frozen=!0,this.addClass("frozen")},unfreeze:function(){this.frozen=!1,this.removeClass("frozen")},top:function(){this.hasClass("top")||(this.addClass("top"),this.removeClass("notTop"),this.onTop&&this.onTop.call(this))},notTop:function(){this.hasClass("notTop")||(this.addClass("notTop"),this.removeClass("top"),this.onNotTop&&this.onNotTop.call(this))},bottom:function(){this.hasClass("bottom")||(this.addClass("bottom"),this.removeClass("notBottom"),this.onBottom&&this.onBottom.call(this))},notBottom:function(){this.hasClass("notBottom")||(this.addClass("notBottom"),this.removeClass("bottom"),this.onNotBottom&&this.onNotBottom.call(this))},shouldUnpin:function(t){return"down"===t.direction&&!t.top&&t.toleranceExceeded},shouldPin:function(t){return"up"===t.direction&&t.toleranceExceeded||t.top},addClass:function(t){this.elem.classList.add.apply(this.elem.classList,this.classes[t].split(" "))},removeClass:function(t){this.elem.classList.remove.apply(this.elem.classList,this.classes[t].split(" "))},hasClass:function(t){return this.classes[t].split(" ").every(function(t){return this.classList.contains(t)},this.elem)},update:function(t){t.isOutOfBounds||!0!==this.frozen&&(t.top?this.top():this.notTop(),t.bottom?this.bottom():this.notBottom(),this.shouldUnpin(t)?this.unpin():this.shouldPin(t)&&this.pin())}},s.options={tolerance:{up:0,down:0},offset:0,scroller:t()?window:null,classes:{frozen:"headroom--frozen",pinned:"headroom--pinned",unpinned:"headroom--unpinned",top:"headroom--top",notTop:"headroom--not-top",bottom:"headroom--bottom",notBottom:"headroom--not-bottom",initial:"headroom"}},s.cutsTheMustard=!!(t()&&function(){}.bind&&"classList"in document.documentElement&&Object.assign&&Object.keys&&requestAnimationFrame),s});
diff --git a/docs/2_39/site_libs/quarto-nav/quarto-nav.js b/docs/2_39/site_libs/quarto-nav/quarto-nav.js
new file mode 100644
index 000000000..38cc43057
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-nav/quarto-nav.js
@@ -0,0 +1,325 @@
+const headroomChanged = new CustomEvent("quarto-hrChanged", {
+  detail: {},
+  bubbles: true,
+  cancelable: false,
+  composed: false,
+});
+
+const announceDismiss = () => {
+  const annEl = window.document.getElementById("quarto-announcement");
+  if (annEl) {
+    annEl.remove();
+
+    const annId = annEl.getAttribute("data-announcement-id");
+    window.localStorage.setItem(`quarto-announce-${annId}`, "true");
+  }
+};
+
+const announceRegister = () => {
+  const annEl = window.document.getElementById("quarto-announcement");
+  if (annEl) {
+    const annId = annEl.getAttribute("data-announcement-id");
+    const isDismissed =
+      window.localStorage.getItem(`quarto-announce-${annId}`) || false;
+    if (isDismissed) {
+      announceDismiss();
+      return;
+    } else {
+      annEl.classList.remove("hidden");
+    }
+
+    const actionEl = annEl.querySelector(".quarto-announcement-action");
+    if (actionEl) {
+      actionEl.addEventListener("click", function (e) {
+        e.preventDefault();
+        // Hide the bar immediately
+        announceDismiss();
+      });
+    }
+  }
+};
+
+window.document.addEventListener("DOMContentLoaded", function () {
+  let init = false;
+
+  announceRegister();
+
+  // Manage the back to top button, if one is present.
+  let lastScrollTop = window.pageYOffset || document.documentElement.scrollTop;
+  const scrollDownBuffer = 5;
+  const scrollUpBuffer = 35;
+  const btn = document.getElementById("quarto-back-to-top");
+  const hideBackToTop = () => {
+    btn.style.display = "none";
+  };
+  const showBackToTop = () => {
+    btn.style.display = "inline-block";
+  };
+  if (btn) {
+    window.document.addEventListener(
+      "scroll",
+      function () {
+        const currentScrollTop =
+          window.pageYOffset || document.documentElement.scrollTop;
+
+        // Shows and hides the button 'intelligently' as the user scrolls
+        if (currentScrollTop - scrollDownBuffer > lastScrollTop) {
+          hideBackToTop();
+          lastScrollTop = currentScrollTop <= 0 ? 0 : currentScrollTop;
+        } else if (currentScrollTop < lastScrollTop - scrollUpBuffer) {
+          showBackToTop();
+          lastScrollTop = currentScrollTop <= 0 ? 0 : currentScrollTop;
+        }
+
+        // Show the button at the bottom, hides it at the top
+        if (currentScrollTop <= 0) {
+          hideBackToTop();
+        } else if (
+          window.innerHeight + currentScrollTop >=
+          document.body.offsetHeight
+        ) {
+          showBackToTop();
+        }
+      },
+      false
+    );
+  }
+
+  function throttle(func, wait) {
+    var timeout;
+    return function () {
+      const context = this;
+      const args = arguments;
+      const later = function () {
+        clearTimeout(timeout);
+        timeout = null;
+        func.apply(context, args);
+      };
+
+      if (!timeout) {
+        timeout = setTimeout(later, wait);
+      }
+    };
+  }
+
+  function headerOffset() {
+    // Set an offset if there is are fixed top navbar
+    const headerEl = window.document.querySelector("header.fixed-top");
+    if (headerEl) {
+      return headerEl.clientHeight;
+    } else {
+      return 0;
+    }
+  }
+
+  function footerOffset() {
+    const footerEl = window.document.querySelector("footer.footer");
+    if (footerEl) {
+      return footerEl.clientHeight;
+    } else {
+      return 0;
+    }
+  }
+
+  function dashboardOffset() {
+    const dashboardNavEl = window.document.getElementById(
+      "quarto-dashboard-header"
+    );
+    if (dashboardNavEl !== null) {
+      return dashboardNavEl.clientHeight;
+    } else {
+      return 0;
+    }
+  }
+
+  function updateDocumentOffsetWithoutAnimation() {
+    updateDocumentOffset(false);
+  }
+
+  function updateDocumentOffset(animated) {
+    // set body offset
+    const topOffset = headerOffset();
+    const bodyOffset = topOffset + footerOffset() + dashboardOffset();
+    const bodyEl = window.document.body;
+    bodyEl.setAttribute("data-bs-offset", topOffset);
+    bodyEl.style.paddingTop = topOffset + "px";
+
+    // deal with sidebar offsets
+    const sidebars = window.document.querySelectorAll(
+      ".sidebar, .headroom-target"
+    );
+    sidebars.forEach((sidebar) => {
+      if (!animated) {
+        sidebar.classList.add("notransition");
+        // Remove the no transition class after the animation has time to complete
+        setTimeout(function () {
+          sidebar.classList.remove("notransition");
+        }, 201);
+      }
+
+      if (window.Headroom && sidebar.classList.contains("sidebar-unpinned")) {
+        sidebar.style.top = "0";
+        sidebar.style.maxHeight = "100vh";
+      } else {
+        sidebar.style.top = topOffset + "px";
+        sidebar.style.maxHeight = "calc(100vh - " + topOffset + "px)";
+      }
+    });
+
+    // allow space for footer
+    const mainContainer = window.document.querySelector(".quarto-container");
+    if (mainContainer) {
+      mainContainer.style.minHeight = "calc(100vh - " + bodyOffset + "px)";
+    }
+
+    // link offset
+    let linkStyle = window.document.querySelector("#quarto-target-style");
+    if (!linkStyle) {
+      linkStyle = window.document.createElement("style");
+      linkStyle.setAttribute("id", "quarto-target-style");
+      window.document.head.appendChild(linkStyle);
+    }
+    while (linkStyle.firstChild) {
+      linkStyle.removeChild(linkStyle.firstChild);
+    }
+    if (topOffset > 0) {
+      linkStyle.appendChild(
+        window.document.createTextNode(`
+      section:target::before {
+        content: "";
+        display: block;
+        height: ${topOffset}px;
+        margin: -${topOffset}px 0 0;
+      }`)
+      );
+    }
+    if (init) {
+      window.dispatchEvent(headroomChanged);
+    }
+    init = true;
+  }
+
+  // initialize headroom
+  var header = window.document.querySelector("#quarto-header");
+  if (header && window.Headroom) {
+    const headroom = new window.Headroom(header, {
+      tolerance: 5,
+      onPin: function () {
+        const sidebars = window.document.querySelectorAll(
+          ".sidebar, .headroom-target"
+        );
+        sidebars.forEach((sidebar) => {
+          sidebar.classList.remove("sidebar-unpinned");
+        });
+        updateDocumentOffset();
+      },
+      onUnpin: function () {
+        const sidebars = window.document.querySelectorAll(
+          ".sidebar, .headroom-target"
+        );
+        sidebars.forEach((sidebar) => {
+          sidebar.classList.add("sidebar-unpinned");
+        });
+        updateDocumentOffset();
+      },
+    });
+    headroom.init();
+
+    let frozen = false;
+    window.quartoToggleHeadroom = function () {
+      if (frozen) {
+        headroom.unfreeze();
+        frozen = false;
+      } else {
+        headroom.freeze();
+        frozen = true;
+      }
+    };
+  }
+
+  window.addEventListener(
+    "hashchange",
+    function (e) {
+      if (
+        getComputedStyle(document.documentElement).scrollBehavior !== "smooth"
+      ) {
+        window.scrollTo(0, window.pageYOffset - headerOffset());
+      }
+    },
+    false
+  );
+
+  // Observe size changed for the header
+  const headerEl = window.document.querySelector("header.fixed-top");
+  if (headerEl && window.ResizeObserver) {
+    const observer = new window.ResizeObserver(() => {
+      setTimeout(updateDocumentOffsetWithoutAnimation, 0);
+    });
+    observer.observe(headerEl, {
+      attributes: true,
+      childList: true,
+      characterData: true,
+    });
+  } else {
+    window.addEventListener(
+      "resize",
+      throttle(updateDocumentOffsetWithoutAnimation, 50)
+    );
+  }
+  setTimeout(updateDocumentOffsetWithoutAnimation, 250);
+
+  // fixup index.html links if we aren't on the filesystem
+  if (window.location.protocol !== "file:") {
+    const links = window.document.querySelectorAll("a");
+    for (let i = 0; i < links.length; i++) {
+      if (links[i].href) {
+        links[i].dataset.originalHref = links[i].href;
+        links[i].href = links[i].href.replace(/\/index\.html/, "/");
+      }
+    }
+
+    // Fixup any sharing links that require urls
+    // Append url to any sharing urls
+    const sharingLinks = window.document.querySelectorAll(
+      "a.sidebar-tools-main-item, a.quarto-navigation-tool, a.quarto-navbar-tools, a.quarto-navbar-tools-item"
+    );
+    for (let i = 0; i < sharingLinks.length; i++) {
+      const sharingLink = sharingLinks[i];
+      const href = sharingLink.getAttribute("href");
+      if (href) {
+        sharingLink.setAttribute(
+          "href",
+          href.replace("|url|", window.location.href)
+        );
+      }
+    }
+
+    // Scroll the active navigation item into view, if necessary
+    const navSidebar = window.document.querySelector("nav#quarto-sidebar");
+    if (navSidebar) {
+      // Find the active item
+      const activeItem = navSidebar.querySelector("li.sidebar-item a.active");
+      if (activeItem) {
+        // Wait for the scroll height and height to resolve by observing size changes on the
+        // nav element that is scrollable
+        const resizeObserver = new ResizeObserver((_entries) => {
+          // The bottom of the element
+          const elBottom = activeItem.offsetTop;
+          const viewBottom = navSidebar.scrollTop + navSidebar.clientHeight;
+
+          // The element height and scroll height are the same, then we are still loading
+          if (viewBottom !== navSidebar.scrollHeight) {
+            // Determine if the item isn't visible and scroll to it
+            if (elBottom >= viewBottom) {
+              navSidebar.scrollTop = elBottom;
+            }
+
+            // stop observing now since we've completed the scroll
+            resizeObserver.unobserve(navSidebar);
+          }
+        });
+        resizeObserver.observe(navSidebar);
+      }
+    }
+  }
+});
diff --git a/docs/2_39/site_libs/quarto-search/autocomplete.umd.js b/docs/2_39/site_libs/quarto-search/autocomplete.umd.js
new file mode 100644
index 000000000..ae0063aa9
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-search/autocomplete.umd.js
@@ -0,0 +1,3 @@
+/*! @algolia/autocomplete-js 1.11.1 | MIT License | © Algolia, Inc. and contributors | https://github.com/algolia/autocomplete */
+!function(e,t){"object"==typeof exports&&"undefined"!=typeof module?t(exports):"function"==typeof define&&define.amd?define(["exports"],t):t((e="undefined"!=typeof globalThis?globalThis:e||self)["@algolia/autocomplete-js"]={})}(this,(function(e){"use strict";function t(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function n(e){for(var n=1;n<arguments.length;n++){var r=null!=arguments[n]?arguments[n]:{};n%2?t(Object(r),!0).forEach((function(t){o(e,t,r[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(r)):t(Object(r)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(r,t))}))}return e}function r(e){return r="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},r(e)}function o(e,t,n){return(t=function(e){var t=function(e,t){if("object"!=typeof e||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!=typeof r)return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"==typeof t?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function i(){return i=Object.assign?Object.assign.bind():function(e){for(var t=1;t<arguments.length;t++){var n=arguments[t];for(var r in n)Object.prototype.hasOwnProperty.call(n,r)&&(e[r]=n[r])}return e},i.apply(this,arguments)}function u(e,t){if(null==e)return{};var n,r,o=function(e,t){if(null==e)return{};var n,r,o={},i=Object.keys(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||(o[n]=e[n]);return o}(e,t);if(Object.getOwnPropertySymbols){var i=Object.getOwnPropertySymbols(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||Object.prototype.propertyIsEnumerable.call(e,n)&&(o[n]=e[n])}return o}function a(e,t){return function(e){if(Array.isArray(e))return e}(e)||function(e,t){var n=null==e?null:"undefined"!=typeof Symbol&&e[Symbol.iterator]||e["@@iterator"];if(null!=n){var r,o,i,u,a=[],l=!0,c=!1;try{if(i=(n=n.call(e)).next,0===t){if(Object(n)!==n)return;l=!1}else for(;!(l=(r=i.call(n)).done)&&(a.push(r.value),a.length!==t);l=!0);}catch(e){c=!0,o=e}finally{try{if(!l&&null!=n.return&&(u=n.return(),Object(u)!==u))return}finally{if(c)throw o}}return a}}(e,t)||c(e,t)||function(){throw new TypeError("Invalid attempt to destructure non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function l(e){return function(e){if(Array.isArray(e))return s(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||c(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function c(e,t){if(e){if("string"==typeof e)return s(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);return"Object"===n&&e.constructor&&(n=e.constructor.name),"Map"===n||"Set"===n?Array.from(e):"Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n)?s(e,t):void 0}}function s(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function f(e){return{current:e}}function p(e,t){var n=void 0;return function(){for(var r=arguments.length,o=new Array(r),i=0;i<r;i++)o[i]=arguments[i];n&&clearTimeout(n),n=setTimeout((function(){return e.apply(void 0,o)}),t)}}function m(e){return e.reduce((function(e,t){return e.concat(t)}),[])}var v=0;function d(){return"autocomplete-".concat(v++)}function y(e,t){return t.reduce((function(e,t){return e&&e[t]}),e)}function b(e){return 0===e.collections.length?0:e.collections.reduce((function(e,t){return e+t.items.length}),0)}function g(e){return e!==Object(e)}function h(e,t){if(e===t)return!0;if(g(e)||g(t)||"function"==typeof e||"function"==typeof t)return e===t;if(Object.keys(e).length!==Object.keys(t).length)return!1;for(var n=0,r=Object.keys(e);n<r.length;n++){var o=r[n];if(!(o in t))return!1;if(!h(e[o],t[o]))return!1}return!0}var O=function(){};var _="1.11.1",S=[{segment:"autocomplete-core",version:_}];function j(e){var t=e.item,n=e.items,r=void 0===n?[]:n;return{index:t.__autocomplete_indexName,items:[t],positions:[1+r.findIndex((function(e){return e.objectID===t.objectID}))],queryID:t.__autocomplete_queryID,algoliaSource:["autocomplete"]}}function P(e,t){return function(e){if(Array.isArray(e))return e}(e)||function(e,t){var n=null==e?null:"undefined"!=typeof Symbol&&e[Symbol.iterator]||e["@@iterator"];if(null!=n){var r,o,i,u,a=[],l=!0,c=!1;try{if(i=(n=n.call(e)).next,0===t){if(Object(n)!==n)return;l=!1}else for(;!(l=(r=i.call(n)).done)&&(a.push(r.value),a.length!==t);l=!0);}catch(e){c=!0,o=e}finally{try{if(!l&&null!=n.return&&(u=n.return(),Object(u)!==u))return}finally{if(c)throw o}}return a}}(e,t)||function(e,t){if(!e)return;if("string"==typeof e)return w(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return w(e,t)}(e,t)||function(){throw new TypeError("Invalid attempt to destructure non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function w(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}var I=["items"],A=["items"];function E(e){return E="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},E(e)}function D(e){return function(e){if(Array.isArray(e))return C(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return C(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return C(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function C(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function k(e,t){if(null==e)return{};var n,r,o=function(e,t){if(null==e)return{};var n,r,o={},i=Object.keys(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||(o[n]=e[n]);return o}(e,t);if(Object.getOwnPropertySymbols){var i=Object.getOwnPropertySymbols(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||Object.prototype.propertyIsEnumerable.call(e,n)&&(o[n]=e[n])}return o}function x(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function N(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?x(Object(n),!0).forEach((function(t){T(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):x(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function T(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==E(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==E(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===E(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function q(e){for(var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:20,n=[],r=0;r<e.objectIDs.length;r+=t)n.push(N(N({},e),{},{objectIDs:e.objectIDs.slice(r,r+t)}));return n}function B(e){return e.map((function(e){var t=e.items,n=k(e,I);return N(N({},n),{},{objectIDs:(null==t?void 0:t.map((function(e){return e.objectID})))||n.objectIDs})}))}function R(e){var t,n,r,o=(t=P((e.version||"").split(".").map(Number),2),n=t[0],r=t[1],n>=3||2===n&&r>=4||1===n&&r>=10);function i(t,n,r){if(o&&void 0!==r){var i=r[0].__autocomplete_algoliaCredentials,u={"X-Algolia-Application-Id":i.appId,"X-Algolia-API-Key":i.apiKey};e.apply(void 0,[t].concat(D(n),[{headers:u}]))}else e.apply(void 0,[t].concat(D(n)))}return{init:function(t,n){e("init",{appId:t,apiKey:n})},setUserToken:function(t){e("setUserToken",t)},clickedObjectIDsAfterSearch:function(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];t.length>0&&i("clickedObjectIDsAfterSearch",B(t),t[0].items)},clickedObjectIDs:function(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];t.length>0&&i("clickedObjectIDs",B(t),t[0].items)},clickedFilters:function(){for(var t=arguments.length,n=new Array(t),r=0;r<t;r++)n[r]=arguments[r];n.length>0&&e.apply(void 0,["clickedFilters"].concat(n))},convertedObjectIDsAfterSearch:function(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];t.length>0&&i("convertedObjectIDsAfterSearch",B(t),t[0].items)},convertedObjectIDs:function(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];t.length>0&&i("convertedObjectIDs",B(t),t[0].items)},convertedFilters:function(){for(var t=arguments.length,n=new Array(t),r=0;r<t;r++)n[r]=arguments[r];n.length>0&&e.apply(void 0,["convertedFilters"].concat(n))},viewedObjectIDs:function(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];t.length>0&&t.reduce((function(e,t){var n=t.items,r=k(t,A);return[].concat(D(e),D(q(N(N({},r),{},{objectIDs:(null==n?void 0:n.map((function(e){return e.objectID})))||r.objectIDs})).map((function(e){return{items:n,payload:e}}))))}),[]).forEach((function(e){var t=e.items;return i("viewedObjectIDs",[e.payload],t)}))},viewedFilters:function(){for(var t=arguments.length,n=new Array(t),r=0;r<t;r++)n[r]=arguments[r];n.length>0&&e.apply(void 0,["viewedFilters"].concat(n))}}}function F(e){var t=e.items.reduce((function(e,t){var n;return e[t.__autocomplete_indexName]=(null!==(n=e[t.__autocomplete_indexName])&&void 0!==n?n:[]).concat(t),e}),{});return Object.keys(t).map((function(e){return{index:e,items:t[e],algoliaSource:["autocomplete"]}}))}function L(e){return e.objectID&&e.__autocomplete_indexName&&e.__autocomplete_queryID}function U(e){return U="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},U(e)}function M(e){return function(e){if(Array.isArray(e))return H(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return H(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return H(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function H(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function V(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function W(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?V(Object(n),!0).forEach((function(t){K(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):V(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function K(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==U(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==U(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===U(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}var Q="2.6.0",$="https://cdn.jsdelivr.net/npm/search-insights@".concat(Q,"/dist/search-insights.min.js"),z=p((function(e){var t=e.onItemsChange,n=e.items,r=e.insights,o=e.state;t({insights:r,insightsEvents:F({items:n}).map((function(e){return W({eventName:"Items Viewed"},e)})),state:o})}),400);function G(e){var t=function(e){return W({onItemsChange:function(e){var t=e.insights,n=e.insightsEvents;t.viewedObjectIDs.apply(t,M(n.map((function(e){return W(W({},e),{},{algoliaSource:[].concat(M(e.algoliaSource||[]),["autocomplete-internal"])})}))))},onSelect:function(e){var t=e.insights,n=e.insightsEvents;t.clickedObjectIDsAfterSearch.apply(t,M(n.map((function(e){return W(W({},e),{},{algoliaSource:[].concat(M(e.algoliaSource||[]),["autocomplete-internal"])})}))))},onActive:O},e)}(e),n=t.insightsClient,r=t.onItemsChange,o=t.onSelect,i=t.onActive,u=n;n||function(e){if("undefined"!=typeof window)e({window:window})}((function(e){var t=e.window,n=t.AlgoliaAnalyticsObject||"aa";"string"==typeof n&&(u=t[n]),u||(t.AlgoliaAnalyticsObject=n,t[n]||(t[n]=function(){t[n].queue||(t[n].queue=[]);for(var e=arguments.length,r=new Array(e),o=0;o<e;o++)r[o]=arguments[o];t[n].queue.push(r)}),t[n].version=Q,u=t[n],function(e){var t="[Autocomplete]: Could not load search-insights.js. Please load it manually following https://alg.li/insights-autocomplete";try{var n=e.document.createElement("script");n.async=!0,n.src=$,n.onerror=function(){console.error(t)},document.body.appendChild(n)}catch(e){console.error(t)}}(t))}));var a=R(u),l=f([]),c=p((function(e){var t=e.state;if(t.isOpen){var n=t.collections.reduce((function(e,t){return[].concat(M(e),M(t.items))}),[]).filter(L);h(l.current.map((function(e){return e.objectID})),n.map((function(e){return e.objectID})))||(l.current=n,n.length>0&&z({onItemsChange:r,items:n,insights:a,state:t}))}}),0);return{name:"aa.algoliaInsightsPlugin",subscribe:function(e){var t=e.setContext,n=e.onSelect,r=e.onActive;function l(e){t({algoliaInsightsPlugin:{__algoliaSearchParameters:W({clickAnalytics:!0},e?{userToken:e}:{}),insights:a}})}u("addAlgoliaAgent","insights-plugin"),l(),u("onUserTokenChange",l),u("getUserToken",null,(function(e,t){l(t)})),n((function(e){var t=e.item,n=e.state,r=e.event,i=e.source;L(t)&&o({state:n,event:r,insights:a,item:t,insightsEvents:[W({eventName:"Item Selected"},j({item:t,items:i.getItems().filter(L)}))]})})),r((function(e){var t=e.item,n=e.source,r=e.state,o=e.event;L(t)&&i({state:r,event:o,insights:a,item:t,insightsEvents:[W({eventName:"Item Active"},j({item:t,items:n.getItems().filter(L)}))]})}))},onStateChange:function(e){var t=e.state;c({state:t})},__autocomplete_pluginOptions:e}}function J(e,t){var n=t;return{then:function(t,r){return J(e.then(Y(t,n,e),Y(r,n,e)),n)},catch:function(t){return J(e.catch(Y(t,n,e)),n)},finally:function(t){return t&&n.onCancelList.push(t),J(e.finally(Y(t&&function(){return n.onCancelList=[],t()},n,e)),n)},cancel:function(){n.isCanceled=!0;var e=n.onCancelList;n.onCancelList=[],e.forEach((function(e){e()}))},isCanceled:function(){return!0===n.isCanceled}}}function X(e){return J(e,{isCanceled:!1,onCancelList:[]})}function Y(e,t,n){return e?function(n){return t.isCanceled?n:e(n)}:n}function Z(e,t,n,r){if(!n)return null;if(e<0&&(null===t||null!==r&&0===t))return n+e;var o=(null===t?-1:t)+e;return o<=-1||o>=n?null===r?null:0:o}function ee(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function te(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?ee(Object(n),!0).forEach((function(t){ne(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):ee(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function ne(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==re(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==re(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===re(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function re(e){return re="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},re(e)}function oe(e){var t=function(e){var t=e.collections.map((function(e){return e.items.length})).reduce((function(e,t,n){var r=(e[n-1]||0)+t;return e.push(r),e}),[]).reduce((function(t,n){return n<=e.activeItemId?t+1:t}),0);return e.collections[t]}(e);if(!t)return null;var n=t.items[function(e){for(var t=e.state,n=e.collection,r=!1,o=0,i=0;!1===r;){var u=t.collections[o];if(u===n){r=!0;break}i+=u.items.length,o++}return t.activeItemId-i}({state:e,collection:t})],r=t.source;return{item:n,itemInputValue:r.getItemInputValue({item:n,state:e}),itemUrl:r.getItemUrl({item:n,state:e}),source:r}}function ie(e,t,n){return[e,null==n?void 0:n.sourceId,t].filter(Boolean).join("-").replace(/\s/g,"")}var ue=/((gt|sm)-|galaxy nexus)|samsung[- ]|samsungbrowser/i;function ae(e){return ae="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},ae(e)}function le(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function ce(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==ae(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==ae(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===ae(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function se(e,t,n){var r,o=t.initialState;return{getState:function(){return o},dispatch:function(r,i){var u=function(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?le(Object(n),!0).forEach((function(t){ce(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):le(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}({},o);o=e(o,{type:r,props:t,payload:i}),n({state:o,prevState:u})},pendingRequests:(r=[],{add:function(e){return r.push(e),e.finally((function(){r=r.filter((function(t){return t!==e}))}))},cancelAll:function(){r.forEach((function(e){return e.cancel()}))},isEmpty:function(){return 0===r.length}})}}function fe(e){return fe="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},fe(e)}function pe(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function me(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?pe(Object(n),!0).forEach((function(t){ve(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):pe(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function ve(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==fe(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==fe(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===fe(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function de(e){return de="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},de(e)}function ye(e){return function(e){if(Array.isArray(e))return be(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return be(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return be(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function be(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function ge(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function he(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?ge(Object(n),!0).forEach((function(t){Oe(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):ge(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function Oe(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==de(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==de(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===de(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function _e(e,t){var n,r="undefined"!=typeof window?window:{},o=e.plugins||[];return he(he({debug:!1,openOnFocus:!1,enterKeyHint:void 0,placeholder:"",autoFocus:!1,defaultActiveItemId:null,stallThreshold:300,insights:!1,environment:r,shouldPanelOpen:function(e){return b(e.state)>0},reshape:function(e){return e.sources}},e),{},{id:null!==(n=e.id)&&void 0!==n?n:d(),plugins:o,initialState:he({activeItemId:null,query:"",completion:null,collections:[],isOpen:!1,status:"idle",context:{}},e.initialState),onStateChange:function(t){var n;null===(n=e.onStateChange)||void 0===n||n.call(e,t),o.forEach((function(e){var n;return null===(n=e.onStateChange)||void 0===n?void 0:n.call(e,t)}))},onSubmit:function(t){var n;null===(n=e.onSubmit)||void 0===n||n.call(e,t),o.forEach((function(e){var n;return null===(n=e.onSubmit)||void 0===n?void 0:n.call(e,t)}))},onReset:function(t){var n;null===(n=e.onReset)||void 0===n||n.call(e,t),o.forEach((function(e){var n;return null===(n=e.onReset)||void 0===n?void 0:n.call(e,t)}))},getSources:function(n){return Promise.all([].concat(ye(o.map((function(e){return e.getSources}))),[e.getSources]).filter(Boolean).map((function(e){return function(e,t){var n=[];return Promise.resolve(e(t)).then((function(e){return Promise.all(e.filter((function(e){return Boolean(e)})).map((function(e){if(e.sourceId,n.includes(e.sourceId))throw new Error("[Autocomplete] The `sourceId` ".concat(JSON.stringify(e.sourceId)," is not unique."));n.push(e.sourceId);var t={getItemInputValue:function(e){return e.state.query},getItemUrl:function(){},onSelect:function(e){(0,e.setIsOpen)(!1)},onActive:O,onResolve:O};Object.keys(t).forEach((function(e){t[e].__default=!0}));var r=te(te({},t),e);return Promise.resolve(r)})))}))}(e,n)}))).then((function(e){return m(e)})).then((function(e){return e.map((function(e){return he(he({},e),{},{onSelect:function(n){e.onSelect(n),t.forEach((function(e){var t;return null===(t=e.onSelect)||void 0===t?void 0:t.call(e,n)}))},onActive:function(n){e.onActive(n),t.forEach((function(e){var t;return null===(t=e.onActive)||void 0===t?void 0:t.call(e,n)}))},onResolve:function(n){e.onResolve(n),t.forEach((function(e){var t;return null===(t=e.onResolve)||void 0===t?void 0:t.call(e,n)}))}})}))}))},navigator:he({navigate:function(e){var t=e.itemUrl;r.location.assign(t)},navigateNewTab:function(e){var t=e.itemUrl,n=r.open(t,"_blank","noopener");null==n||n.focus()},navigateNewWindow:function(e){var t=e.itemUrl;r.open(t,"_blank","noopener")}},e.navigator)})}function Se(e){return Se="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},Se(e)}function je(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Pe(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?je(Object(n),!0).forEach((function(t){we(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):je(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function we(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==Se(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==Se(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===Se(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function Ie(e){return Ie="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},Ie(e)}function Ae(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Ee(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?Ae(Object(n),!0).forEach((function(t){De(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):Ae(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function De(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==Ie(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==Ie(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===Ie(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function Ce(e){return function(e){if(Array.isArray(e))return ke(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return ke(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return ke(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function ke(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function xe(e){return Boolean(e.execute)}function Ne(e,t,n){if(o=e,Boolean(null==o?void 0:o.execute)){var r="algolia"===e.requesterId?Object.assign.apply(Object,[{}].concat(Ce(Object.keys(n.context).map((function(e){var t;return null===(t=n.context[e])||void 0===t?void 0:t.__algoliaSearchParameters}))))):{};return Ee(Ee({},e),{},{requests:e.queries.map((function(n){return{query:"algolia"===e.requesterId?Ee(Ee({},n),{},{params:Ee(Ee({},r),n.params)}):n,sourceId:t,transformResponse:e.transformResponse}}))})}var o;return{items:e,sourceId:t}}function Te(e){var t=e.reduce((function(e,t){if(!xe(t))return e.push(t),e;var n=t.searchClient,r=t.execute,o=t.requesterId,i=t.requests,u=e.find((function(e){return xe(t)&&xe(e)&&e.searchClient===n&&Boolean(o)&&e.requesterId===o}));if(u){var a;(a=u.items).push.apply(a,Ce(i))}else{var l={execute:r,requesterId:o,items:i,searchClient:n};e.push(l)}return e}),[]).map((function(e){if(!xe(e))return Promise.resolve(e);var t=e,n=t.execute,r=t.items;return n({searchClient:t.searchClient,requests:r})}));return Promise.all(t).then((function(e){return m(e)}))}function qe(e,t,n){return t.map((function(t){var r,o=e.filter((function(e){return e.sourceId===t.sourceId})),i=o.map((function(e){return e.items})),u=o[0].transformResponse,a=u?u({results:r=i,hits:r.map((function(e){return e.hits})).filter(Boolean),facetHits:r.map((function(e){var t;return null===(t=e.facetHits)||void 0===t?void 0:t.map((function(e){return{label:e.value,count:e.count,_highlightResult:{label:{value:e.highlighted}}}}))})).filter(Boolean)}):i;return t.onResolve({source:t,results:i,items:a,state:n.getState()}),a.every(Boolean),'The `getItems` function from source "'.concat(t.sourceId,'" must return an array of items but returned ').concat(JSON.stringify(void 0),".\n\nDid you forget to return items?\n\nSee: https://www.algolia.com/doc/ui-libraries/autocomplete/core-concepts/sources/#param-getitems"),{source:t,items:a}}))}function Be(e){return Be="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},Be(e)}var Re=["event","nextState","props","query","refresh","store"];function Fe(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Le(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?Fe(Object(n),!0).forEach((function(t){Ue(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):Fe(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function Ue(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==Be(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==Be(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===Be(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function Me(e,t){if(null==e)return{};var n,r,o=function(e,t){if(null==e)return{};var n,r,o={},i=Object.keys(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||(o[n]=e[n]);return o}(e,t);if(Object.getOwnPropertySymbols){var i=Object.getOwnPropertySymbols(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||Object.prototype.propertyIsEnumerable.call(e,n)&&(o[n]=e[n])}return o}var He,Ve,We,Ke=null,Qe=(He=-1,Ve=-1,We=void 0,function(e){var t=++He;return Promise.resolve(e).then((function(e){return We&&t<Ve?We:(Ve=t,We=e,e)}))});function $e(e){var t=e.event,n=e.nextState,r=void 0===n?{}:n,o=e.props,i=e.query,u=e.refresh,a=e.store,l=Me(e,Re);Ke&&o.environment.clearTimeout(Ke);var c=l.setCollections,s=l.setIsOpen,f=l.setQuery,p=l.setActiveItemId,v=l.setStatus;if(f(i),p(o.defaultActiveItemId),!i&&!1===o.openOnFocus){var d,y=a.getState().collections.map((function(e){return Le(Le({},e),{},{items:[]})}));v("idle"),c(y),s(null!==(d=r.isOpen)&&void 0!==d?d:o.shouldPanelOpen({state:a.getState()}));var b=X(Qe(y).then((function(){return Promise.resolve()})));return a.pendingRequests.add(b)}v("loading"),Ke=o.environment.setTimeout((function(){v("stalled")}),o.stallThreshold);var g=X(Qe(o.getSources(Le({query:i,refresh:u,state:a.getState()},l)).then((function(e){return Promise.all(e.map((function(e){return Promise.resolve(e.getItems(Le({query:i,refresh:u,state:a.getState()},l))).then((function(t){return Ne(t,e.sourceId,a.getState())}))}))).then(Te).then((function(t){return qe(t,e,a)})).then((function(e){return function(e){var t=e.collections,n=e.props,r=e.state,o=t.reduce((function(e,t){return Pe(Pe({},e),{},we({},t.source.sourceId,Pe(Pe({},t.source),{},{getItems:function(){return m(t.items)}})))}),{}),i=n.plugins.reduce((function(e,t){return t.reshape?t.reshape(e):e}),{sourcesBySourceId:o,state:r}).sourcesBySourceId;return m(n.reshape({sourcesBySourceId:i,sources:Object.values(i),state:r})).filter(Boolean).map((function(e){return{source:e,items:e.getItems()}}))}({collections:e,props:o,state:a.getState()})}))})))).then((function(e){var n;v("idle"),c(e);var f=o.shouldPanelOpen({state:a.getState()});s(null!==(n=r.isOpen)&&void 0!==n?n:o.openOnFocus&&!i&&f||f);var p=oe(a.getState());if(null!==a.getState().activeItemId&&p){var m=p.item,d=p.itemInputValue,y=p.itemUrl,b=p.source;b.onActive(Le({event:t,item:m,itemInputValue:d,itemUrl:y,refresh:u,source:b,state:a.getState()},l))}})).finally((function(){v("idle"),Ke&&o.environment.clearTimeout(Ke)}));return a.pendingRequests.add(g)}function ze(e){return ze="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},ze(e)}var Ge=["event","props","refresh","store"];function Je(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Xe(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?Je(Object(n),!0).forEach((function(t){Ye(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):Je(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function Ye(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==ze(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==ze(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===ze(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function Ze(e,t){if(null==e)return{};var n,r,o=function(e,t){if(null==e)return{};var n,r,o={},i=Object.keys(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||(o[n]=e[n]);return o}(e,t);if(Object.getOwnPropertySymbols){var i=Object.getOwnPropertySymbols(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||Object.prototype.propertyIsEnumerable.call(e,n)&&(o[n]=e[n])}return o}function et(e){return et="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},et(e)}var tt=["props","refresh","store"],nt=["inputElement","formElement","panelElement"],rt=["inputElement"],ot=["inputElement","maxLength"],it=["source"],ut=["item","source"];function at(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function lt(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?at(Object(n),!0).forEach((function(t){ct(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):at(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function ct(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==et(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==et(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===et(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function st(e,t){if(null==e)return{};var n,r,o=function(e,t){if(null==e)return{};var n,r,o={},i=Object.keys(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||(o[n]=e[n]);return o}(e,t);if(Object.getOwnPropertySymbols){var i=Object.getOwnPropertySymbols(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||Object.prototype.propertyIsEnumerable.call(e,n)&&(o[n]=e[n])}return o}function ft(e){var t=e.props,n=e.refresh,r=e.store,o=st(e,tt);return{getEnvironmentProps:function(e){var n=e.inputElement,o=e.formElement,i=e.panelElement;function u(e){!r.getState().isOpen&&r.pendingRequests.isEmpty()||e.target===n||!1===[o,i].some((function(t){return n=t,r=e.target,n===r||n.contains(r);var n,r}))&&(r.dispatch("blur",null),t.debug||r.pendingRequests.cancelAll())}return lt({onTouchStart:u,onMouseDown:u,onTouchMove:function(e){!1!==r.getState().isOpen&&n===t.environment.document.activeElement&&e.target!==n&&n.blur()}},st(e,nt))},getRootProps:function(e){return lt({role:"combobox","aria-expanded":r.getState().isOpen,"aria-haspopup":"listbox","aria-owns":r.getState().isOpen?r.getState().collections.map((function(e){var n=e.source;return ie(t.id,"list",n)})).join(" "):void 0,"aria-labelledby":ie(t.id,"label")},e)},getFormProps:function(e){return e.inputElement,lt({action:"",noValidate:!0,role:"search",onSubmit:function(i){var u;i.preventDefault(),t.onSubmit(lt({event:i,refresh:n,state:r.getState()},o)),r.dispatch("submit",null),null===(u=e.inputElement)||void 0===u||u.blur()},onReset:function(i){var u;i.preventDefault(),t.onReset(lt({event:i,refresh:n,state:r.getState()},o)),r.dispatch("reset",null),null===(u=e.inputElement)||void 0===u||u.focus()}},st(e,rt))},getLabelProps:function(e){return lt({htmlFor:ie(t.id,"input"),id:ie(t.id,"label")},e)},getInputProps:function(e){var i;function u(e){(t.openOnFocus||Boolean(r.getState().query))&&$e(lt({event:e,props:t,query:r.getState().completion||r.getState().query,refresh:n,store:r},o)),r.dispatch("focus",null)}var a=e||{};a.inputElement;var l=a.maxLength,c=void 0===l?512:l,s=st(a,ot),f=oe(r.getState()),p=function(e){return Boolean(e&&e.match(ue))}((null===(i=t.environment.navigator)||void 0===i?void 0:i.userAgent)||""),m=t.enterKeyHint||(null!=f&&f.itemUrl&&!p?"go":"search");return lt({"aria-autocomplete":"both","aria-activedescendant":r.getState().isOpen&&null!==r.getState().activeItemId?ie(t.id,"item-".concat(r.getState().activeItemId),null==f?void 0:f.source):void 0,"aria-controls":r.getState().isOpen?r.getState().collections.map((function(e){var n=e.source;return ie(t.id,"list",n)})).join(" "):void 0,"aria-labelledby":ie(t.id,"label"),value:r.getState().completion||r.getState().query,id:ie(t.id,"input"),autoComplete:"off",autoCorrect:"off",autoCapitalize:"off",enterKeyHint:m,spellCheck:"false",autoFocus:t.autoFocus,placeholder:t.placeholder,maxLength:c,type:"search",onChange:function(e){$e(lt({event:e,props:t,query:e.currentTarget.value.slice(0,c),refresh:n,store:r},o))},onKeyDown:function(e){!function(e){var t=e.event,n=e.props,r=e.refresh,o=e.store,i=Ze(e,Ge);if("ArrowUp"===t.key||"ArrowDown"===t.key){var u=function(){var e=oe(o.getState()),t=n.environment.document.getElementById(ie(n.id,"item-".concat(o.getState().activeItemId),null==e?void 0:e.source));t&&(t.scrollIntoViewIfNeeded?t.scrollIntoViewIfNeeded(!1):t.scrollIntoView(!1))},a=function(){var e=oe(o.getState());if(null!==o.getState().activeItemId&&e){var n=e.item,u=e.itemInputValue,a=e.itemUrl,l=e.source;l.onActive(Xe({event:t,item:n,itemInputValue:u,itemUrl:a,refresh:r,source:l,state:o.getState()},i))}};t.preventDefault(),!1===o.getState().isOpen&&(n.openOnFocus||Boolean(o.getState().query))?$e(Xe({event:t,props:n,query:o.getState().query,refresh:r,store:o},i)).then((function(){o.dispatch(t.key,{nextActiveItemId:n.defaultActiveItemId}),a(),setTimeout(u,0)})):(o.dispatch(t.key,{}),a(),u())}else if("Escape"===t.key)t.preventDefault(),o.dispatch(t.key,null),o.pendingRequests.cancelAll();else if("Tab"===t.key)o.dispatch("blur",null),o.pendingRequests.cancelAll();else if("Enter"===t.key){if(null===o.getState().activeItemId||o.getState().collections.every((function(e){return 0===e.items.length})))return void(n.debug||o.pendingRequests.cancelAll());t.preventDefault();var l=oe(o.getState()),c=l.item,s=l.itemInputValue,f=l.itemUrl,p=l.source;if(t.metaKey||t.ctrlKey)void 0!==f&&(p.onSelect(Xe({event:t,item:c,itemInputValue:s,itemUrl:f,refresh:r,source:p,state:o.getState()},i)),n.navigator.navigateNewTab({itemUrl:f,item:c,state:o.getState()}));else if(t.shiftKey)void 0!==f&&(p.onSelect(Xe({event:t,item:c,itemInputValue:s,itemUrl:f,refresh:r,source:p,state:o.getState()},i)),n.navigator.navigateNewWindow({itemUrl:f,item:c,state:o.getState()}));else if(t.altKey);else{if(void 0!==f)return p.onSelect(Xe({event:t,item:c,itemInputValue:s,itemUrl:f,refresh:r,source:p,state:o.getState()},i)),void n.navigator.navigate({itemUrl:f,item:c,state:o.getState()});$e(Xe({event:t,nextState:{isOpen:!1},props:n,query:s,refresh:r,store:o},i)).then((function(){p.onSelect(Xe({event:t,item:c,itemInputValue:s,itemUrl:f,refresh:r,source:p,state:o.getState()},i))}))}}}(lt({event:e,props:t,refresh:n,store:r},o))},onFocus:u,onBlur:O,onClick:function(n){e.inputElement!==t.environment.document.activeElement||r.getState().isOpen||u(n)}},s)},getPanelProps:function(e){return lt({onMouseDown:function(e){e.preventDefault()},onMouseLeave:function(){r.dispatch("mouseleave",null)}},e)},getListProps:function(e){var n=e||{},r=n.source,o=st(n,it);return lt({role:"listbox","aria-labelledby":ie(t.id,"label"),id:ie(t.id,"list",r)},o)},getItemProps:function(e){var i=e.item,u=e.source,a=st(e,ut);return lt({id:ie(t.id,"item-".concat(i.__autocomplete_id),u),role:"option","aria-selected":r.getState().activeItemId===i.__autocomplete_id,onMouseMove:function(e){if(i.__autocomplete_id!==r.getState().activeItemId){r.dispatch("mousemove",i.__autocomplete_id);var t=oe(r.getState());if(null!==r.getState().activeItemId&&t){var u=t.item,a=t.itemInputValue,l=t.itemUrl,c=t.source;c.onActive(lt({event:e,item:u,itemInputValue:a,itemUrl:l,refresh:n,source:c,state:r.getState()},o))}}},onMouseDown:function(e){e.preventDefault()},onClick:function(e){var a=u.getItemInputValue({item:i,state:r.getState()}),l=u.getItemUrl({item:i,state:r.getState()});(l?Promise.resolve():$e(lt({event:e,nextState:{isOpen:!1},props:t,query:a,refresh:n,store:r},o))).then((function(){u.onSelect(lt({event:e,item:i,itemInputValue:a,itemUrl:l,refresh:n,source:u,state:r.getState()},o))}))}},a)}}}function pt(e){return pt="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},pt(e)}function mt(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function vt(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?mt(Object(n),!0).forEach((function(t){dt(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):mt(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function dt(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==pt(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==pt(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===pt(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function yt(e){var t,n,r,o,i=e.plugins,u=e.options,a=null===(t=((null===(n=u.__autocomplete_metadata)||void 0===n?void 0:n.userAgents)||[])[0])||void 0===t?void 0:t.segment,l=a?dt({},a,Object.keys((null===(r=u.__autocomplete_metadata)||void 0===r?void 0:r.options)||{})):{};return{plugins:i.map((function(e){return{name:e.name,options:Object.keys(e.__autocomplete_pluginOptions||[])}})),options:vt({"autocomplete-core":Object.keys(u)},l),ua:S.concat((null===(o=u.__autocomplete_metadata)||void 0===o?void 0:o.userAgents)||[])}}function bt(e){var t,n=e.state;return!1===n.isOpen||null===n.activeItemId?null:(null===(t=oe(n))||void 0===t?void 0:t.itemInputValue)||null}function gt(e){return gt="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},gt(e)}function ht(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Ot(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?ht(Object(n),!0).forEach((function(t){_t(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):ht(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function _t(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==gt(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==gt(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===gt(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}var St=function(e,t){switch(t.type){case"setActiveItemId":case"mousemove":return Ot(Ot({},e),{},{activeItemId:t.payload});case"setQuery":return Ot(Ot({},e),{},{query:t.payload,completion:null});case"setCollections":return Ot(Ot({},e),{},{collections:t.payload});case"setIsOpen":return Ot(Ot({},e),{},{isOpen:t.payload});case"setStatus":return Ot(Ot({},e),{},{status:t.payload});case"setContext":return Ot(Ot({},e),{},{context:Ot(Ot({},e.context),t.payload)});case"ArrowDown":var n=Ot(Ot({},e),{},{activeItemId:t.payload.hasOwnProperty("nextActiveItemId")?t.payload.nextActiveItemId:Z(1,e.activeItemId,b(e),t.props.defaultActiveItemId)});return Ot(Ot({},n),{},{completion:bt({state:n})});case"ArrowUp":var r=Ot(Ot({},e),{},{activeItemId:Z(-1,e.activeItemId,b(e),t.props.defaultActiveItemId)});return Ot(Ot({},r),{},{completion:bt({state:r})});case"Escape":return e.isOpen?Ot(Ot({},e),{},{activeItemId:null,isOpen:!1,completion:null}):Ot(Ot({},e),{},{activeItemId:null,query:"",status:"idle",collections:[]});case"submit":return Ot(Ot({},e),{},{activeItemId:null,isOpen:!1,status:"idle"});case"reset":return Ot(Ot({},e),{},{activeItemId:!0===t.props.openOnFocus?t.props.defaultActiveItemId:null,status:"idle",query:""});case"focus":return Ot(Ot({},e),{},{activeItemId:t.props.defaultActiveItemId,isOpen:(t.props.openOnFocus||Boolean(e.query))&&t.props.shouldPanelOpen({state:e})});case"blur":return t.props.debug?e:Ot(Ot({},e),{},{isOpen:!1,activeItemId:null});case"mouseleave":return Ot(Ot({},e),{},{activeItemId:t.props.defaultActiveItemId});default:return"The reducer action ".concat(JSON.stringify(t.type)," is not supported."),e}};function jt(e){return jt="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},jt(e)}function Pt(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function wt(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?Pt(Object(n),!0).forEach((function(t){It(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):Pt(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function It(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==jt(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==jt(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===jt(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function At(e){var t=[],n=_e(e,t),r=se(St,n,(function(e){var t=e.prevState,r=e.state;n.onStateChange(wt({prevState:t,state:r,refresh:u,navigator:n.navigator},o))})),o=function(e){var t=e.store;return{setActiveItemId:function(e){t.dispatch("setActiveItemId",e)},setQuery:function(e){t.dispatch("setQuery",e)},setCollections:function(e){var n=0,r=e.map((function(e){return me(me({},e),{},{items:m(e.items).map((function(e){return me(me({},e),{},{__autocomplete_id:n++})}))})}));t.dispatch("setCollections",r)},setIsOpen:function(e){t.dispatch("setIsOpen",e)},setStatus:function(e){t.dispatch("setStatus",e)},setContext:function(e){t.dispatch("setContext",e)}}}({store:r}),i=ft(wt({props:n,refresh:u,store:r,navigator:n.navigator},o));function u(){return $e(wt({event:new Event("input"),nextState:{isOpen:r.getState().isOpen},props:n,navigator:n.navigator,query:r.getState().query,refresh:u,store:r},o))}if(e.insights&&!n.plugins.some((function(e){return"aa.algoliaInsightsPlugin"===e.name}))){var a="boolean"==typeof e.insights?{}:e.insights;n.plugins.push(G(a))}return n.plugins.forEach((function(e){var r;return null===(r=e.subscribe)||void 0===r?void 0:r.call(e,wt(wt({},o),{},{navigator:n.navigator,refresh:u,onSelect:function(e){t.push({onSelect:e})},onActive:function(e){t.push({onActive:e})},onResolve:function(e){t.push({onResolve:e})}}))})),function(e){var t,n,r=e.metadata,o=e.environment;if(null===(t=o.navigator)||void 0===t||null===(n=t.userAgent)||void 0===n?void 0:n.includes("Algolia Crawler")){var i=o.document.createElement("meta"),u=o.document.querySelector("head");i.name="algolia:metadata",setTimeout((function(){i.content=JSON.stringify(r),u.appendChild(i)}),0)}}({metadata:yt({plugins:n.plugins,options:e}),environment:n.environment}),wt(wt({refresh:u,navigator:n.navigator},i),o)}var Et=function(e,t,n,r){var o;t[0]=0;for(var i=1;i<t.length;i++){var u=t[i++],a=t[i]?(t[0]|=u?1:2,n[t[i++]]):t[++i];3===u?r[0]=a:4===u?r[1]=Object.assign(r[1]||{},a):5===u?(r[1]=r[1]||{})[t[++i]]=a:6===u?r[1][t[++i]]+=a+"":u?(o=e.apply(a,Et(e,a,n,["",null])),r.push(o),a[0]?t[0]|=2:(t[i-2]=0,t[i]=o)):r.push(a)}return r},Dt=new Map;function Ct(e){var t=Dt.get(this);return t||(t=new Map,Dt.set(this,t)),(t=Et(this,t.get(e)||(t.set(e,t=function(e){for(var t,n,r=1,o="",i="",u=[0],a=function(e){1===r&&(e||(o=o.replace(/^\s*\n\s*|\s*\n\s*$/g,"")))?u.push(0,e,o):3===r&&(e||o)?(u.push(3,e,o),r=2):2===r&&"..."===o&&e?u.push(4,e,0):2===r&&o&&!e?u.push(5,0,!0,o):r>=5&&((o||!e&&5===r)&&(u.push(r,0,o,n),r=6),e&&(u.push(r,e,0,n),r=6)),o=""},l=0;l<e.length;l++){l&&(1===r&&a(),a(l));for(var c=0;c<e[l].length;c++)t=e[l][c],1===r?"<"===t?(a(),u=[u],r=3):o+=t:4===r?"--"===o&&">"===t?(r=1,o=""):o=t+o[0]:i?t===i?i="":o+=t:'"'===t||"'"===t?i=t:">"===t?(a(),r=1):r&&("="===t?(r=5,n=o,o=""):"/"===t&&(r<5||">"===e[l][c+1])?(a(),3===r&&(u=u[0]),r=u,(u=u[0]).push(2,0,r),r=0):" "===t||"\t"===t||"\n"===t||"\r"===t?(a(),r=2):o+=t),3===r&&"!--"===o&&(r=4,u=u[0])}return a(),u}(e)),t),arguments,[])).length>1?t:t[0]}var kt=function(e){var t=e.environment,n=t.document.createElementNS("http://www.w3.org/2000/svg","svg");n.setAttribute("class","aa-ClearIcon"),n.setAttribute("viewBox","0 0 24 24"),n.setAttribute("width","18"),n.setAttribute("height","18"),n.setAttribute("fill","currentColor");var r=t.document.createElementNS("http://www.w3.org/2000/svg","path");return r.setAttribute("d","M5.293 6.707l5.293 5.293-5.293 5.293c-0.391 0.391-0.391 1.024 0 1.414s1.024 0.391 1.414 0l5.293-5.293 5.293 5.293c0.391 0.391 1.024 0.391 1.414 0s0.391-1.024 0-1.414l-5.293-5.293 5.293-5.293c0.391-0.391 0.391-1.024 0-1.414s-1.024-0.391-1.414 0l-5.293 5.293-5.293-5.293c-0.391-0.391-1.024-0.391-1.414 0s-0.391 1.024 0 1.414z"),n.appendChild(r),n};function xt(e,t){if("string"==typeof t){var n=e.document.querySelector(t);return"The element ".concat(JSON.stringify(t)," is not in the document."),n}return t}function Nt(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];return t.reduce((function(e,t){return Object.keys(t).forEach((function(n){var r=e[n],o=t[n];r!==o&&(e[n]=[r,o].filter(Boolean).join(" "))})),e}),{})}var Tt=function(e){return e&&"object"===r(e)&&"[object Object]"===Object.prototype.toString.call(e)};function qt(){for(var e=arguments.length,t=new Array(e),n=0;n<e;n++)t[n]=arguments[n];return t.reduce((function(e,t){return Object.keys(t).forEach((function(n){var r=e[n],o=t[n];Array.isArray(r)&&Array.isArray(o)?e[n]=r.concat.apply(r,l(o)):Tt(r)&&Tt(o)?e[n]=qt(r,o):e[n]=o})),e}),{})}function Bt(e,t){return Object.entries(e).reduce((function(e,r){var i=a(r,2),u=i[0],l=i[1];return t({key:u,value:l})?n(n({},e),{},o({},u,l)):e}),{})}var Rt=["ontouchstart","ontouchend","ontouchmove","ontouchcancel"];function Ft(e,t,n){e[t]=null===n?"":"number"!=typeof n?n:n+"px"}function Lt(e){this._listeners[e.type](e)}function Ut(e,t,n){var r,o,i=e[t];if("style"===t)if("string"==typeof n)e.style=n;else if(null===n)e.style="";else for(t in n)i&&n[t]===i[t]||Ft(e.style,t,n[t]);else"o"===t[0]&&"n"===t[1]?(r=t!==(t=t.replace(/Capture$/,"")),((o=t.toLowerCase())in e||Rt.includes(o))&&(t=o),t=t.slice(2),e._listeners||(e._listeners={}),e._listeners[t]=n,n?i||e.addEventListener(t,Lt,r):e.removeEventListener(t,Lt,r)):"list"!==t&&"tagName"!==t&&"form"!==t&&"type"!==t&&"size"!==t&&"download"!==t&&"href"!==t&&t in e?e[t]=null==n?"":n:"function"!=typeof n&&"dangerouslySetInnerHTML"!==t&&(null==n||!1===n&&!/^ar/.test(t)?e.removeAttribute(t):e.setAttribute(t,n))}function Mt(e){return"onChange"===e?"onInput":e}function Ht(e,t){for(var n in t)Ut(e,Mt(n),t[n])}function Vt(e,t){for(var n in t)"o"===n[0]&&"n"===n[1]||Ut(e,Mt(n),t[n])}var Wt=["children"];function Kt(e){return function(t,n){var r=n.children,o=void 0===r?[]:r,i=u(n,Wt),a=e.document.createElement(t);return Ht(a,i),a.append.apply(a,l(o)),a}}var Qt=["autocompleteScopeApi","environment","classNames","getInputProps","getInputPropsCore","isDetached","state"],$t=function(e){var t=e.environment.document.createElementNS("http://www.w3.org/2000/svg","svg");return t.setAttribute("class","aa-LoadingIcon"),t.setAttribute("viewBox","0 0 100 100"),t.setAttribute("width","20"),t.setAttribute("height","20"),t.innerHTML='<circle\n  cx="50"\n  cy="50"\n  fill="none"\n  r="35"\n  stroke="currentColor"\n  stroke-dasharray="164.93361431346415 56.97787143782138"\n  stroke-width="6"\n>\n  <animateTransform\n    attributeName="transform"\n    type="rotate"\n    repeatCount="indefinite"\n    dur="1s"\n    values="0 50 50;90 50 50;180 50 50;360 50 50"\n    keyTimes="0;0.40;0.65;1"\n  />\n</circle>',t},zt=function(e){var t=e.environment,n=t.document.createElementNS("http://www.w3.org/2000/svg","svg");n.setAttribute("class","aa-SubmitIcon"),n.setAttribute("viewBox","0 0 24 24"),n.setAttribute("width","20"),n.setAttribute("height","20"),n.setAttribute("fill","currentColor");var r=t.document.createElementNS("http://www.w3.org/2000/svg","path");return r.setAttribute("d","M16.041 15.856c-0.034 0.026-0.067 0.055-0.099 0.087s-0.060 0.064-0.087 0.099c-1.258 1.213-2.969 1.958-4.855 1.958-1.933 0-3.682-0.782-4.95-2.050s-2.050-3.017-2.050-4.95 0.782-3.682 2.050-4.95 3.017-2.050 4.95-2.050 3.682 0.782 4.95 2.050 2.050 3.017 2.050 4.95c0 1.886-0.745 3.597-1.959 4.856zM21.707 20.293l-3.675-3.675c1.231-1.54 1.968-3.493 1.968-5.618 0-2.485-1.008-4.736-2.636-6.364s-3.879-2.636-6.364-2.636-4.736 1.008-6.364 2.636-2.636 3.879-2.636 6.364 1.008 4.736 2.636 6.364 3.879 2.636 6.364 2.636c2.125 0 4.078-0.737 5.618-1.968l3.675 3.675c0.391 0.391 1.024 0.391 1.414 0s0.391-1.024 0-1.414z"),n.appendChild(r),n};function Gt(e){var t=e.autocomplete,r=e.autocompleteScopeApi,o=e.classNames,i=e.environment,a=e.isDetached,l=e.placeholder,c=void 0===l?"Search":l,s=e.propGetters,f=e.setIsModalOpen,p=e.state,m=e.translations,v=Kt(i),d=s.getRootProps(n({state:p,props:t.getRootProps({})},r)),y=v("div",n({class:o.root},d)),b=v("div",{class:o.detachedContainer,onMouseDown:function(e){e.stopPropagation()}}),g=v("div",{class:o.detachedOverlay,children:[b],onMouseDown:function(){f(!1),t.setIsOpen(!1)}}),h=s.getLabelProps(n({state:p,props:t.getLabelProps({})},r)),O=v("button",{class:o.submitButton,type:"submit",title:m.submitButtonTitle,children:[zt({environment:i})]}),_=v("label",n({class:o.label,children:[O]},h)),S=v("button",{class:o.clearButton,type:"reset",title:m.clearButtonTitle,children:[kt({environment:i})]}),j=v("div",{class:o.loadingIndicator,children:[$t({environment:i})]}),P=function(e){var t=e.autocompleteScopeApi,r=e.environment;e.classNames;var o=e.getInputProps,i=e.getInputPropsCore,a=e.isDetached,l=e.state,c=u(e,Qt),s=Kt(r)("input",c),f=o(n({state:l,props:i({inputElement:s}),inputElement:s},t));return Ht(s,n(n({},f),{},{onKeyDown:function(e){a&&"Tab"===e.key||f.onKeyDown(e)}})),s}({class:o.input,environment:i,state:p,getInputProps:s.getInputProps,getInputPropsCore:t.getInputProps,autocompleteScopeApi:r,isDetached:a}),w=v("div",{class:o.inputWrapperPrefix,children:[_,j]}),I=v("div",{class:o.inputWrapperSuffix,children:[S]}),A=v("div",{class:o.inputWrapper,children:[P]}),E=s.getFormProps(n({state:p,props:t.getFormProps({inputElement:P})},r)),D=v("form",n({class:o.form,children:[w,A,I]},E)),C=s.getPanelProps(n({state:p,props:t.getPanelProps({})},r)),k=v("div",n({class:o.panel},C)),x=v("div",{class:o.detachedSearchButtonQuery,textContent:p.query}),N=v("div",{class:o.detachedSearchButtonPlaceholder,hidden:Boolean(p.query),textContent:c});if(a){var T=v("div",{class:o.detachedSearchButtonIcon,children:[zt({environment:i})]}),q=v("button",{type:"button",class:o.detachedSearchButton,onClick:function(){f(!0)},children:[T,N,x]}),B=v("button",{type:"button",class:o.detachedCancelButton,textContent:m.detachedCancelButtonText,onTouchStart:function(e){e.stopPropagation()},onClick:function(){t.setIsOpen(!1),f(!1)}}),R=v("div",{class:o.detachedFormContainer,children:[D,B]});b.appendChild(R),y.appendChild(q)}else y.appendChild(D);return{detachedContainer:b,detachedOverlay:g,detachedSearchButtonQuery:x,detachedSearchButtonPlaceholder:N,inputWrapper:A,input:P,root:y,form:D,label:_,submitButton:O,clearButton:S,loadingIndicator:j,panel:k}}var Jt,Xt,Yt,Zt,en,tn,nn,rn={},on=[],un=/acit|ex(?:s|g|n|p|$)|rph|grid|ows|mnc|ntw|ine[ch]|zoo|^ord|itera/i;function an(e,t){for(var n in t)e[n]=t[n];return e}function ln(e){var t=e.parentNode;t&&t.removeChild(e)}function cn(e,t,n){var r,o,i,u={};for(i in t)"key"==i?r=t[i]:"ref"==i?o=t[i]:u[i]=t[i];if(arguments.length>2&&(u.children=arguments.length>3?Jt.call(arguments,2):n),"function"==typeof e&&null!=e.defaultProps)for(i in e.defaultProps)void 0===u[i]&&(u[i]=e.defaultProps[i]);return sn(e,u,r,o,null)}function sn(e,t,n,r,o){var i={type:e,props:t,key:n,ref:r,__k:null,__:null,__b:0,__e:null,__d:void 0,__c:null,__h:null,constructor:void 0,__v:null==o?++Yt:o};return null==o&&null!=Xt.vnode&&Xt.vnode(i),i}function fn(e){return e.children}function pn(e,t){this.props=e,this.context=t}function mn(e,t){if(null==t)return e.__?mn(e.__,e.__.__k.indexOf(e)+1):null;for(var n;t<e.__k.length;t++)if(null!=(n=e.__k[t])&&null!=n.__e)return n.__e;return"function"==typeof e.type?mn(e):null}function vn(e){var t,n;if(null!=(e=e.__)&&null!=e.__c){for(e.__e=e.__c.base=null,t=0;t<e.__k.length;t++)if(null!=(n=e.__k[t])&&null!=n.__e){e.__e=e.__c.base=n.__e;break}return vn(e)}}function dn(e){(!e.__d&&(e.__d=!0)&&Zt.push(e)&&!yn.__r++||en!==Xt.debounceRendering)&&((en=Xt.debounceRendering)||tn)(yn)}function yn(){var e,t,n,r,o,i,u,a;for(Zt.sort(nn);e=Zt.shift();)e.__d&&(t=Zt.length,r=void 0,o=void 0,u=(i=(n=e).__v).__e,(a=n.__P)&&(r=[],(o=an({},i)).__v=i.__v+1,wn(a,i,o,n.__n,void 0!==a.ownerSVGElement,null!=i.__h?[u]:null,r,null==u?mn(i):u,i.__h),In(r,i),i.__e!=u&&vn(i)),Zt.length>t&&Zt.sort(nn));yn.__r=0}function bn(e,t,n,r,o,i,u,a,l,c){var s,f,p,m,v,d,y,b=r&&r.__k||on,g=b.length;for(n.__k=[],s=0;s<t.length;s++)if(null!=(m=n.__k[s]=null==(m=t[s])||"boolean"==typeof m||"function"==typeof m?null:"string"==typeof m||"number"==typeof m||"bigint"==typeof m?sn(null,m,null,null,m):Array.isArray(m)?sn(fn,{children:m},null,null,null):m.__b>0?sn(m.type,m.props,m.key,m.ref?m.ref:null,m.__v):m)){if(m.__=n,m.__b=n.__b+1,null===(p=b[s])||p&&m.key==p.key&&m.type===p.type)b[s]=void 0;else for(f=0;f<g;f++){if((p=b[f])&&m.key==p.key&&m.type===p.type){b[f]=void 0;break}p=null}wn(e,m,p=p||rn,o,i,u,a,l,c),v=m.__e,(f=m.ref)&&p.ref!=f&&(y||(y=[]),p.ref&&y.push(p.ref,null,m),y.push(f,m.__c||v,m)),null!=v?(null==d&&(d=v),"function"==typeof m.type&&m.__k===p.__k?m.__d=l=gn(m,l,e):l=hn(e,m,p,b,v,l),"function"==typeof n.type&&(n.__d=l)):l&&p.__e==l&&l.parentNode!=e&&(l=mn(p))}for(n.__e=d,s=g;s--;)null!=b[s]&&("function"==typeof n.type&&null!=b[s].__e&&b[s].__e==n.__d&&(n.__d=On(r).nextSibling),Dn(b[s],b[s]));if(y)for(s=0;s<y.length;s++)En(y[s],y[++s],y[++s])}function gn(e,t,n){for(var r,o=e.__k,i=0;o&&i<o.length;i++)(r=o[i])&&(r.__=e,t="function"==typeof r.type?gn(r,t,n):hn(n,r,r,o,r.__e,t));return t}function hn(e,t,n,r,o,i){var u,a,l;if(void 0!==t.__d)u=t.__d,t.__d=void 0;else if(null==n||o!=i||null==o.parentNode)e:if(null==i||i.parentNode!==e)e.appendChild(o),u=null;else{for(a=i,l=0;(a=a.nextSibling)&&l<r.length;l+=1)if(a==o)break e;e.insertBefore(o,i),u=i}return void 0!==u?u:o.nextSibling}function On(e){var t,n,r;if(null==e.type||"string"==typeof e.type)return e.__e;if(e.__k)for(t=e.__k.length-1;t>=0;t--)if((n=e.__k[t])&&(r=On(n)))return r;return null}function _n(e,t,n){"-"===t[0]?e.setProperty(t,null==n?"":n):e[t]=null==n?"":"number"!=typeof n||un.test(t)?n:n+"px"}function Sn(e,t,n,r,o){var i;e:if("style"===t)if("string"==typeof n)e.style.cssText=n;else{if("string"==typeof r&&(e.style.cssText=r=""),r)for(t in r)n&&t in n||_n(e.style,t,"");if(n)for(t in n)r&&n[t]===r[t]||_n(e.style,t,n[t])}else if("o"===t[0]&&"n"===t[1])i=t!==(t=t.replace(/Capture$/,"")),t=t.toLowerCase()in e?t.toLowerCase().slice(2):t.slice(2),e.l||(e.l={}),e.l[t+i]=n,n?r||e.addEventListener(t,i?Pn:jn,i):e.removeEventListener(t,i?Pn:jn,i);else if("dangerouslySetInnerHTML"!==t){if(o)t=t.replace(/xlink(H|:h)/,"h").replace(/sName$/,"s");else if("width"!==t&&"height"!==t&&"href"!==t&&"list"!==t&&"form"!==t&&"tabIndex"!==t&&"download"!==t&&t in e)try{e[t]=null==n?"":n;break e}catch(e){}"function"==typeof n||(null==n||!1===n&&"-"!==t[4]?e.removeAttribute(t):e.setAttribute(t,n))}}function jn(e){return this.l[e.type+!1](Xt.event?Xt.event(e):e)}function Pn(e){return this.l[e.type+!0](Xt.event?Xt.event(e):e)}function wn(e,t,n,r,o,i,u,a,l){var c,s,f,p,m,v,d,y,b,g,h,O,_,S,j,P=t.type;if(void 0!==t.constructor)return null;null!=n.__h&&(l=n.__h,a=t.__e=n.__e,t.__h=null,i=[a]),(c=Xt.__b)&&c(t);try{e:if("function"==typeof P){if(y=t.props,b=(c=P.contextType)&&r[c.__c],g=c?b?b.props.value:c.__:r,n.__c?d=(s=t.__c=n.__c).__=s.__E:("prototype"in P&&P.prototype.render?t.__c=s=new P(y,g):(t.__c=s=new pn(y,g),s.constructor=P,s.render=Cn),b&&b.sub(s),s.props=y,s.state||(s.state={}),s.context=g,s.__n=r,f=s.__d=!0,s.__h=[],s._sb=[]),null==s.__s&&(s.__s=s.state),null!=P.getDerivedStateFromProps&&(s.__s==s.state&&(s.__s=an({},s.__s)),an(s.__s,P.getDerivedStateFromProps(y,s.__s))),p=s.props,m=s.state,s.__v=t,f)null==P.getDerivedStateFromProps&&null!=s.componentWillMount&&s.componentWillMount(),null!=s.componentDidMount&&s.__h.push(s.componentDidMount);else{if(null==P.getDerivedStateFromProps&&y!==p&&null!=s.componentWillReceiveProps&&s.componentWillReceiveProps(y,g),!s.__e&&null!=s.shouldComponentUpdate&&!1===s.shouldComponentUpdate(y,s.__s,g)||t.__v===n.__v){for(t.__v!==n.__v&&(s.props=y,s.state=s.__s,s.__d=!1),s.__e=!1,t.__e=n.__e,t.__k=n.__k,t.__k.forEach((function(e){e&&(e.__=t)})),h=0;h<s._sb.length;h++)s.__h.push(s._sb[h]);s._sb=[],s.__h.length&&u.push(s);break e}null!=s.componentWillUpdate&&s.componentWillUpdate(y,s.__s,g),null!=s.componentDidUpdate&&s.__h.push((function(){s.componentDidUpdate(p,m,v)}))}if(s.context=g,s.props=y,s.__P=e,O=Xt.__r,_=0,"prototype"in P&&P.prototype.render){for(s.state=s.__s,s.__d=!1,O&&O(t),c=s.render(s.props,s.state,s.context),S=0;S<s._sb.length;S++)s.__h.push(s._sb[S]);s._sb=[]}else do{s.__d=!1,O&&O(t),c=s.render(s.props,s.state,s.context),s.state=s.__s}while(s.__d&&++_<25);s.state=s.__s,null!=s.getChildContext&&(r=an(an({},r),s.getChildContext())),f||null==s.getSnapshotBeforeUpdate||(v=s.getSnapshotBeforeUpdate(p,m)),j=null!=c&&c.type===fn&&null==c.key?c.props.children:c,bn(e,Array.isArray(j)?j:[j],t,n,r,o,i,u,a,l),s.base=t.__e,t.__h=null,s.__h.length&&u.push(s),d&&(s.__E=s.__=null),s.__e=!1}else null==i&&t.__v===n.__v?(t.__k=n.__k,t.__e=n.__e):t.__e=An(n.__e,t,n,r,o,i,u,l);(c=Xt.diffed)&&c(t)}catch(e){t.__v=null,(l||null!=i)&&(t.__e=a,t.__h=!!l,i[i.indexOf(a)]=null),Xt.__e(e,t,n)}}function In(e,t){Xt.__c&&Xt.__c(t,e),e.some((function(t){try{e=t.__h,t.__h=[],e.some((function(e){e.call(t)}))}catch(e){Xt.__e(e,t.__v)}}))}function An(e,t,n,r,o,i,u,a){var l,c,s,f=n.props,p=t.props,m=t.type,v=0;if("svg"===m&&(o=!0),null!=i)for(;v<i.length;v++)if((l=i[v])&&"setAttribute"in l==!!m&&(m?l.localName===m:3===l.nodeType)){e=l,i[v]=null;break}if(null==e){if(null===m)return document.createTextNode(p);e=o?document.createElementNS("http://www.w3.org/2000/svg",m):document.createElement(m,p.is&&p),i=null,a=!1}if(null===m)f===p||a&&e.data===p||(e.data=p);else{if(i=i&&Jt.call(e.childNodes),c=(f=n.props||rn).dangerouslySetInnerHTML,s=p.dangerouslySetInnerHTML,!a){if(null!=i)for(f={},v=0;v<e.attributes.length;v++)f[e.attributes[v].name]=e.attributes[v].value;(s||c)&&(s&&(c&&s.__html==c.__html||s.__html===e.innerHTML)||(e.innerHTML=s&&s.__html||""))}if(function(e,t,n,r,o){var i;for(i in n)"children"===i||"key"===i||i in t||Sn(e,i,null,n[i],r);for(i in t)o&&"function"!=typeof t[i]||"children"===i||"key"===i||"value"===i||"checked"===i||n[i]===t[i]||Sn(e,i,t[i],n[i],r)}(e,p,f,o,a),s)t.__k=[];else if(v=t.props.children,bn(e,Array.isArray(v)?v:[v],t,n,r,o&&"foreignObject"!==m,i,u,i?i[0]:n.__k&&mn(n,0),a),null!=i)for(v=i.length;v--;)null!=i[v]&&ln(i[v]);a||("value"in p&&void 0!==(v=p.value)&&(v!==e.value||"progress"===m&&!v||"option"===m&&v!==f.value)&&Sn(e,"value",v,f.value,!1),"checked"in p&&void 0!==(v=p.checked)&&v!==e.checked&&Sn(e,"checked",v,f.checked,!1))}return e}function En(e,t,n){try{"function"==typeof e?e(t):e.current=t}catch(e){Xt.__e(e,n)}}function Dn(e,t,n){var r,o;if(Xt.unmount&&Xt.unmount(e),(r=e.ref)&&(r.current&&r.current!==e.__e||En(r,null,t)),null!=(r=e.__c)){if(r.componentWillUnmount)try{r.componentWillUnmount()}catch(e){Xt.__e(e,t)}r.base=r.__P=null,e.__c=void 0}if(r=e.__k)for(o=0;o<r.length;o++)r[o]&&Dn(r[o],t,n||"function"!=typeof e.type);n||null==e.__e||ln(e.__e),e.__=e.__e=e.__d=void 0}function Cn(e,t,n){return this.constructor(e,n)}Jt=on.slice,Xt={__e:function(e,t,n,r){for(var o,i,u;t=t.__;)if((o=t.__c)&&!o.__)try{if((i=o.constructor)&&null!=i.getDerivedStateFromError&&(o.setState(i.getDerivedStateFromError(e)),u=o.__d),null!=o.componentDidCatch&&(o.componentDidCatch(e,r||{}),u=o.__d),u)return o.__E=o}catch(t){e=t}throw e}},Yt=0,pn.prototype.setState=function(e,t){var n;n=null!=this.__s&&this.__s!==this.state?this.__s:this.__s=an({},this.state),"function"==typeof e&&(e=e(an({},n),this.props)),e&&an(n,e),null!=e&&this.__v&&(t&&this._sb.push(t),dn(this))},pn.prototype.forceUpdate=function(e){this.__v&&(this.__e=!0,e&&this.__h.push(e),dn(this))},pn.prototype.render=fn,Zt=[],tn="function"==typeof Promise?Promise.prototype.then.bind(Promise.resolve()):setTimeout,nn=function(e,t){return e.__v.__b-t.__v.__b},yn.__r=0;var kn="__aa-highlight__",xn="__/aa-highlight__";function Nn(e){var t=e.highlightedValue.split(kn),n=t.shift(),r=function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:[];return{get:function(){return e},add:function(t){var n=e[e.length-1];(null==n?void 0:n.isHighlighted)===t.isHighlighted?e[e.length-1]={value:n.value+t.value,isHighlighted:n.isHighlighted}:e.push(t)}}}(n?[{value:n,isHighlighted:!1}]:[]);return t.forEach((function(e){var t=e.split(xn);r.add({value:t[0],isHighlighted:!0}),""!==t[1]&&r.add({value:t[1],isHighlighted:!1})})),r.get()}function Tn(e){return function(e){if(Array.isArray(e))return qn(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return qn(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return qn(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function qn(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function Bn(e){var t=e.hit,n=e.attribute,r=Array.isArray(n)?n:[n],o=y(t,["_highlightResult"].concat(Tn(r),["value"]));return"string"!=typeof o&&(o=y(t,r)||""),Nn({highlightedValue:o})}var Rn={"&amp;":"&","&lt;":"<","&gt;":">","&quot;":'"',"&#39;":"'"},Fn=new RegExp(/\w/i),Ln=/&(amp|quot|lt|gt|#39);/g,Un=RegExp(Ln.source);function Mn(e,t){var n,r,o,i=e[t],u=(null===(n=e[t+1])||void 0===n?void 0:n.isHighlighted)||!0,a=(null===(r=e[t-1])||void 0===r?void 0:r.isHighlighted)||!0;return Fn.test((o=i.value)&&Un.test(o)?o.replace(Ln,(function(e){return Rn[e]})):o)||a!==u?i.isHighlighted:a}function Hn(e){return Hn="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},Hn(e)}function Vn(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Wn(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?Vn(Object(n),!0).forEach((function(t){Kn(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):Vn(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function Kn(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==Hn(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==Hn(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===Hn(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function Qn(e){return e.some((function(e){return e.isHighlighted}))?e.map((function(t,n){return Wn(Wn({},t),{},{isHighlighted:!Mn(e,n)})})):e.map((function(e){return Wn(Wn({},e),{},{isHighlighted:!1})}))}function $n(e){return function(e){if(Array.isArray(e))return zn(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return zn(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return zn(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function zn(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function Gn(e){var t=e.hit,n=e.attribute,r=Array.isArray(n)?n:[n],o=y(t,["_snippetResult"].concat($n(r),["value"]));return"string"!=typeof o&&(o=y(t,r)||""),Nn({highlightedValue:o})}function Jn(e){return Jn="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},Jn(e)}function Xn(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function Yn(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?Xn(Object(n),!0).forEach((function(t){Zn(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):Xn(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function Zn(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==Jn(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==Jn(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===Jn(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function er(e){return er="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},er(e)}var tr=["params"];function nr(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function rr(e){for(var t=1;t<arguments.length;t++){var n=null!=arguments[t]?arguments[t]:{};t%2?nr(Object(n),!0).forEach((function(t){or(e,t,n[t])})):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(n)):nr(Object(n)).forEach((function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(n,t))}))}return e}function or(e,t,n){return(t=function(e){var t=function(e,t){if("object"!==er(e)||null===e)return e;var n=e[Symbol.toPrimitive];if(void 0!==n){var r=n.call(e,t||"default");if("object"!==er(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(e,"string");return"symbol"===er(t)?t:String(t)}(t))in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function ir(e,t){if(null==e)return{};var n,r,o=function(e,t){if(null==e)return{};var n,r,o={},i=Object.keys(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||(o[n]=e[n]);return o}(e,t);if(Object.getOwnPropertySymbols){var i=Object.getOwnPropertySymbols(e);for(r=0;r<i.length;r++)n=i[r],t.indexOf(n)>=0||Object.prototype.propertyIsEnumerable.call(e,n)&&(o[n]=e[n])}return o}function ur(e){return function(e){if(Array.isArray(e))return ar(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return ar(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return ar(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function ar(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function lr(e){var t=e.createElement,n=e.Fragment;function r(e){var r=e.hit,o=e.attribute,i=e.tagName,u=void 0===i?"mark":i;return t(n,{},Bn({hit:r,attribute:o}).map((function(e,n){return e.isHighlighted?t(u,{key:n},e.value):e.value})))}return r.__autocomplete_componentName="Highlight",r}function cr(e){var t=e.createElement,n=e.Fragment;function r(e){var r,o=e.hit,i=e.attribute,u=e.tagName,a=void 0===u?"mark":u;return t(n,{},(r={hit:o,attribute:i},Qn(Bn(r))).map((function(e,n){return e.isHighlighted?t(a,{key:n},e.value):e.value})))}return r.__autocomplete_componentName="ReverseHighlight",r}function sr(e){var t=e.createElement,n=e.Fragment;function r(e){var r,o=e.hit,i=e.attribute,u=e.tagName,a=void 0===u?"mark":u;return t(n,{},(r={hit:o,attribute:i},Qn(Gn(r))).map((function(e,n){return e.isHighlighted?t(a,{key:n},e.value):e.value})))}return r.__autocomplete_componentName="ReverseSnippet",r}function fr(e){var t=e.createElement,n=e.Fragment;function r(e){var r=e.hit,o=e.attribute,i=e.tagName,u=void 0===i?"mark":i;return t(n,{},Gn({hit:r,attribute:o}).map((function(e,n){return e.isHighlighted?t(u,{key:n},e.value):e.value})))}return r.__autocomplete_componentName="Snippet",r}var pr=["classNames","container","getEnvironmentProps","getFormProps","getInputProps","getItemProps","getLabelProps","getListProps","getPanelProps","getRootProps","panelContainer","panelPlacement","render","renderNoResults","renderer","detachedMediaQuery","components","translations"],mr={clearButton:"aa-ClearButton",detachedCancelButton:"aa-DetachedCancelButton",detachedContainer:"aa-DetachedContainer",detachedFormContainer:"aa-DetachedFormContainer",detachedOverlay:"aa-DetachedOverlay",detachedSearchButton:"aa-DetachedSearchButton",detachedSearchButtonIcon:"aa-DetachedSearchButtonIcon",detachedSearchButtonPlaceholder:"aa-DetachedSearchButtonPlaceholder",detachedSearchButtonQuery:"aa-DetachedSearchButtonQuery",form:"aa-Form",input:"aa-Input",inputWrapper:"aa-InputWrapper",inputWrapperPrefix:"aa-InputWrapperPrefix",inputWrapperSuffix:"aa-InputWrapperSuffix",item:"aa-Item",label:"aa-Label",list:"aa-List",loadingIndicator:"aa-LoadingIndicator",panel:"aa-Panel",panelLayout:"aa-PanelLayout aa-Panel--scrollable",root:"aa-Autocomplete",source:"aa-Source",sourceFooter:"aa-SourceFooter",sourceHeader:"aa-SourceHeader",sourceNoResults:"aa-SourceNoResults",submitButton:"aa-SubmitButton"},vr=function(e,t){var n=e.children;(0,e.render)(n,t)},dr={createElement:cn,Fragment:fn,render:function(e,t,n){var r,o,i;Xt.__&&Xt.__(e,t),o=(r="function"==typeof n)?null:n&&n.__k||t.__k,i=[],wn(t,e=(!r&&n||t).__k=cn(fn,null,[e]),o||rn,rn,void 0!==t.ownerSVGElement,!r&&n?[n]:o?null:t.firstChild?Jt.call(t.childNodes):null,i,!r&&n?n:o?o.__e:t.firstChild,r),In(i,e)}};function yr(e){var t=e.panelPlacement,n=e.container,r=e.form,o=e.environment,i=n.getBoundingClientRect(),u=(o.pageYOffset||o.document.documentElement.scrollTop||o.document.body.scrollTop||0)+i.top+i.height;switch(t){case"start":return{top:u,left:i.left};case"end":return{top:u,right:o.document.documentElement.clientWidth-(i.left+i.width)};case"full-width":return{top:u,left:0,right:0,width:"unset",maxWidth:"unset"};case"input-wrapper-width":var a=r.getBoundingClientRect();return{top:u,left:a.left,right:o.document.documentElement.clientWidth-(a.left+a.width),width:"unset",maxWidth:"unset"};default:throw new Error("[Autocomplete] The `panelPlacement` value ".concat(JSON.stringify(t)," is not valid."))}}var br=[{segment:"autocomplete-js",version:_}],gr=["components"];var hr=function(e,t){function n(t){return e({searchClient:t.searchClient,queries:t.requests.map((function(e){return e.query}))}).then((function(e){return e.map((function(e,n){var r=t.requests[n];return{items:e,sourceId:r.sourceId,transformResponse:r.transformResponse}}))}))}return function(e){return function(r){return Yn(Yn({requesterId:t,execute:n},e),r)}}}((function(e){return function(e){var t=e.searchClient,n=e.queries,r=e.userAgents,o=void 0===r?[]:r;"function"==typeof t.addAlgoliaAgent&&[].concat(ur(S),ur(o)).forEach((function(e){var n=e.segment,r=e.version;t.addAlgoliaAgent(n,r)}));var i=function(e){var t=e.transporter||{},n=t.headers,r=void 0===n?{}:n,o=t.queryParameters,i=void 0===o?{}:o,u="x-algolia-application-id",a="x-algolia-api-key";return{appId:r[u]||i[u],apiKey:r[a]||i[a]}}(t),u=i.appId,a=i.apiKey;return t.search(n.map((function(e){var t=e.params;return rr(rr({},ir(e,tr)),{},{params:rr({hitsPerPage:5,highlightPreTag:kn,highlightPostTag:xn},t)})}))).then((function(e){return e.results.map((function(e,t){var r;return rr(rr({},e),{},{hits:null===(r=e.hits)||void 0===r?void 0:r.map((function(r){return rr(rr({},r),{},{__autocomplete_indexName:e.index||n[t].indexName,__autocomplete_queryID:e.queryID,__autocomplete_algoliaCredentials:{appId:u,apiKey:a}})}))})}))}))}(n(n({},e),{},{userAgents:br}))}),"algolia");var Or=hr({transformResponse:function(e){return e.hits}});e.autocomplete=function(e){var t,r=function(){var e=[],t=[];function n(n){e.push(n);var r=n();t.push(r)}return{runEffect:n,cleanupEffects:function(){var e=t;t=[],e.forEach((function(e){e()}))},runEffects:function(){var t=e;e=[],t.forEach((function(e){n(e)}))}}}(),a=r.runEffect,l=r.cleanupEffects,c=r.runEffects,s=(t=[],{reactive:function(e){var n=e(),r={_fn:e,_ref:{current:n},get value(){return this._ref.current},set value(e){this._ref.current=e}};return t.push(r),r},runReactives:function(){t.forEach((function(e){e._ref.current=e._fn()}))}}),m=s.reactive,v=s.runReactives,y=f(!1),g=f(e),h=f(void 0),O=m((function(){return function(e){var t,r=e.classNames,o=e.container,i=e.getEnvironmentProps,a=e.getFormProps,l=e.getInputProps,c=e.getItemProps,s=e.getLabelProps,f=e.getListProps,p=e.getPanelProps,m=e.getRootProps,v=e.panelContainer,y=e.panelPlacement,b=e.render,g=e.renderNoResults,h=e.renderer,O=e.detachedMediaQuery,_=e.components,S=e.translations,j=u(e,pr),P="undefined"!=typeof window?window:{},w=xt(P,o);w.tagName;var I=n(n({},dr),h),A={Highlight:lr(I),ReverseHighlight:cr(I),ReverseSnippet:sr(I),Snippet:fr(I)};return{renderer:{classNames:Nt(mr,null!=r?r:{}),container:w,getEnvironmentProps:null!=i?i:function(e){return e.props},getFormProps:null!=a?a:function(e){return e.props},getInputProps:null!=l?l:function(e){return e.props},getItemProps:null!=c?c:function(e){return e.props},getLabelProps:null!=s?s:function(e){return e.props},getListProps:null!=f?f:function(e){return e.props},getPanelProps:null!=p?p:function(e){return e.props},getRootProps:null!=m?m:function(e){return e.props},panelContainer:v?xt(P,v):P.document.body,panelPlacement:null!=y?y:"input-wrapper-width",render:null!=b?b:vr,renderNoResults:g,renderer:I,detachedMediaQuery:null!=O?O:getComputedStyle(P.document.documentElement).getPropertyValue("--aa-detached-media-query"),components:n(n({},A),_),translations:n(n({},{clearButtonTitle:"Clear",detachedCancelButtonText:"Cancel",submitButtonTitle:"Submit"}),S)},core:n(n({},j),{},{id:null!==(t=j.id)&&void 0!==t?t:d(),environment:P})}}(g.current)})),_=m((function(){return O.value.core.environment.matchMedia(O.value.renderer.detachedMediaQuery).matches})),S=m((function(){return At(n(n({},O.value.core),{},{onStateChange:function(e){var t,n,r;y.current=e.state.collections.some((function(e){return e.source.templates.noResults})),null===(t=h.current)||void 0===t||t.call(h,e),null===(n=(r=O.value.core).onStateChange)||void 0===n||n.call(r,e)},shouldPanelOpen:g.current.shouldPanelOpen||function(e){var t=e.state;if(_.value)return!0;var n=b(t)>0;if(!O.value.core.openOnFocus&&!t.query)return n;var r=Boolean(y.current||O.value.renderer.renderNoResults);return!n&&r||n},__autocomplete_metadata:{userAgents:br,options:e}}))})),j=f(n({collections:[],completion:null,context:{},isOpen:!1,query:"",activeItemId:null,status:"idle"},O.value.core.initialState)),P={getEnvironmentProps:O.value.renderer.getEnvironmentProps,getFormProps:O.value.renderer.getFormProps,getInputProps:O.value.renderer.getInputProps,getItemProps:O.value.renderer.getItemProps,getLabelProps:O.value.renderer.getLabelProps,getListProps:O.value.renderer.getListProps,getPanelProps:O.value.renderer.getPanelProps,getRootProps:O.value.renderer.getRootProps},w={setActiveItemId:S.value.setActiveItemId,setQuery:S.value.setQuery,setCollections:S.value.setCollections,setIsOpen:S.value.setIsOpen,setStatus:S.value.setStatus,setContext:S.value.setContext,refresh:S.value.refresh,navigator:S.value.navigator},I=m((function(){return Ct.bind(O.value.renderer.renderer.createElement)})),A=m((function(){return Gt({autocomplete:S.value,autocompleteScopeApi:w,classNames:O.value.renderer.classNames,environment:O.value.core.environment,isDetached:_.value,placeholder:O.value.core.placeholder,propGetters:P,setIsModalOpen:k,state:j.current,translations:O.value.renderer.translations})}));function E(){Ht(A.value.panel,{style:_.value?{}:yr({panelPlacement:O.value.renderer.panelPlacement,container:A.value.root,form:A.value.form,environment:O.value.core.environment})})}function D(e){j.current=e;var t={autocomplete:S.value,autocompleteScopeApi:w,classNames:O.value.renderer.classNames,components:O.value.renderer.components,container:O.value.renderer.container,html:I.value,dom:A.value,panelContainer:_.value?A.value.detachedContainer:O.value.renderer.panelContainer,propGetters:P,state:j.current,renderer:O.value.renderer.renderer},r=!b(e)&&!y.current&&O.value.renderer.renderNoResults||O.value.renderer.render;!function(e){var t=e.autocomplete,r=e.autocompleteScopeApi,o=e.dom,i=e.propGetters,u=e.state;Vt(o.root,i.getRootProps(n({state:u,props:t.getRootProps({})},r))),Vt(o.input,i.getInputProps(n({state:u,props:t.getInputProps({inputElement:o.input}),inputElement:o.input},r))),Ht(o.label,{hidden:"stalled"===u.status}),Ht(o.loadingIndicator,{hidden:"stalled"!==u.status}),Ht(o.clearButton,{hidden:!u.query}),Ht(o.detachedSearchButtonQuery,{textContent:u.query}),Ht(o.detachedSearchButtonPlaceholder,{hidden:Boolean(u.query)})}(t),function(e,t){var r=t.autocomplete,o=t.autocompleteScopeApi,u=t.classNames,a=t.html,l=t.dom,c=t.panelContainer,s=t.propGetters,f=t.state,p=t.components,m=t.renderer;if(f.isOpen){c.contains(l.panel)||"loading"===f.status||c.appendChild(l.panel),l.panel.classList.toggle("aa-Panel--stalled","stalled"===f.status);var v=f.collections.filter((function(e){var t=e.source,n=e.items;return t.templates.noResults||n.length>0})).map((function(e,t){var l=e.source,c=e.items;return m.createElement("section",{key:t,className:u.source,"data-autocomplete-source-id":l.sourceId},l.templates.header&&m.createElement("div",{className:u.sourceHeader},l.templates.header({components:p,createElement:m.createElement,Fragment:m.Fragment,items:c,source:l,state:f,html:a})),l.templates.noResults&&0===c.length?m.createElement("div",{className:u.sourceNoResults},l.templates.noResults({components:p,createElement:m.createElement,Fragment:m.Fragment,source:l,state:f,html:a})):m.createElement("ul",i({className:u.list},s.getListProps(n({state:f,props:r.getListProps({source:l})},o))),c.map((function(e){var t=r.getItemProps({item:e,source:l});return m.createElement("li",i({key:t.id,className:u.item},s.getItemProps(n({state:f,props:t},o))),l.templates.item({components:p,createElement:m.createElement,Fragment:m.Fragment,item:e,state:f,html:a}))}))),l.templates.footer&&m.createElement("div",{className:u.sourceFooter},l.templates.footer({components:p,createElement:m.createElement,Fragment:m.Fragment,items:c,source:l,state:f,html:a})))})),d=m.createElement(m.Fragment,null,m.createElement("div",{className:u.panelLayout},v),m.createElement("div",{className:"aa-GradientBottom"})),y=v.reduce((function(e,t){return e[t.props["data-autocomplete-source-id"]]=t,e}),{});e(n(n({children:d,state:f,sections:v,elements:y},m),{},{components:p,html:a},o),l.panel)}else c.contains(l.panel)&&c.removeChild(l.panel)}(r,t)}function C(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{};l();var t=O.value.renderer,n=t.components,r=u(t,gr);g.current=qt(r,O.value.core,{components:Bt(n,(function(e){return!e.value.hasOwnProperty("__autocomplete_componentName")})),initialState:j.current},e),v(),c(),S.value.refresh().then((function(){D(j.current)}))}function k(e){requestAnimationFrame((function(){var t=O.value.core.environment.document.body.contains(A.value.detachedOverlay);e!==t&&(e?(O.value.core.environment.document.body.appendChild(A.value.detachedOverlay),O.value.core.environment.document.body.classList.add("aa-Detached"),A.value.input.focus()):(O.value.core.environment.document.body.removeChild(A.value.detachedOverlay),O.value.core.environment.document.body.classList.remove("aa-Detached")))}))}return a((function(){var e=S.value.getEnvironmentProps({formElement:A.value.form,panelElement:A.value.panel,inputElement:A.value.input});return Ht(O.value.core.environment,e),function(){Ht(O.value.core.environment,Object.keys(e).reduce((function(e,t){return n(n({},e),{},o({},t,void 0))}),{}))}})),a((function(){var e=_.value?O.value.core.environment.document.body:O.value.renderer.panelContainer,t=_.value?A.value.detachedOverlay:A.value.panel;return _.value&&j.current.isOpen&&k(!0),D(j.current),function(){e.contains(t)&&e.removeChild(t)}})),a((function(){var e=O.value.renderer.container;return e.appendChild(A.value.root),function(){e.removeChild(A.value.root)}})),a((function(){var e=p((function(e){D(e.state)}),0);return h.current=function(t){var n=t.state,r=t.prevState;(_.value&&r.isOpen!==n.isOpen&&k(n.isOpen),_.value||!n.isOpen||r.isOpen||E(),n.query!==r.query)&&O.value.core.environment.document.querySelectorAll(".aa-Panel--scrollable").forEach((function(e){0!==e.scrollTop&&(e.scrollTop=0)}));e({state:n})},function(){h.current=void 0}})),a((function(){var e=p((function(){var e=_.value;_.value=O.value.core.environment.matchMedia(O.value.renderer.detachedMediaQuery).matches,e!==_.value?C({}):requestAnimationFrame(E)}),20);return O.value.core.environment.addEventListener("resize",e),function(){O.value.core.environment.removeEventListener("resize",e)}})),a((function(){if(!_.value)return function(){};function e(e){A.value.detachedContainer.classList.toggle("aa-DetachedContainer--modal",e)}function t(t){e(t.matches)}var n=O.value.core.environment.matchMedia(getComputedStyle(O.value.core.environment.document.documentElement).getPropertyValue("--aa-detached-modal-media-query"));e(n.matches);var r=Boolean(n.addEventListener);return r?n.addEventListener("change",t):n.addListener(t),function(){r?n.removeEventListener("change",t):n.removeListener(t)}})),a((function(){return requestAnimationFrame(E),function(){}})),n(n({},w),{},{update:C,destroy:function(){l()}})},e.getAlgoliaFacets=function(e){var t=hr({transformResponse:function(e){return e.facetHits}}),r=e.queries.map((function(e){return n(n({},e),{},{type:"facet"})}));return t(n(n({},e),{},{queries:r}))},e.getAlgoliaResults=Or,Object.defineProperty(e,"__esModule",{value:!0})}));
+
diff --git a/docs/2_39/site_libs/quarto-search/fuse.min.js b/docs/2_39/site_libs/quarto-search/fuse.min.js
new file mode 100644
index 000000000..adc28356e
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-search/fuse.min.js
@@ -0,0 +1,9 @@
+/**
+ * Fuse.js v6.6.2 - Lightweight fuzzy-search (http://fusejs.io)
+ *
+ * Copyright (c) 2022 Kiro Risk (http://kiro.me)
+ * All Rights Reserved. Apache Software License 2.0
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ */
+var e,t;e=this,t=function(){"use strict";function e(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter((function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable}))),n.push.apply(n,r)}return n}function t(t){for(var n=1;n<arguments.length;n++){var r=null!=arguments[n]?arguments[n]:{};n%2?e(Object(r),!0).forEach((function(e){c(t,e,r[e])})):Object.getOwnPropertyDescriptors?Object.defineProperties(t,Object.getOwnPropertyDescriptors(r)):e(Object(r)).forEach((function(e){Object.defineProperty(t,e,Object.getOwnPropertyDescriptor(r,e))}))}return t}function n(e){return n="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},n(e)}function r(e,t){if(!(e instanceof t))throw new TypeError("Cannot call a class as a function")}function i(e,t){for(var n=0;n<t.length;n++){var r=t[n];r.enumerable=r.enumerable||!1,r.configurable=!0,"value"in r&&(r.writable=!0),Object.defineProperty(e,r.key,r)}}function o(e,t,n){return t&&i(e.prototype,t),n&&i(e,n),Object.defineProperty(e,"prototype",{writable:!1}),e}function c(e,t,n){return t in e?Object.defineProperty(e,t,{value:n,enumerable:!0,configurable:!0,writable:!0}):e[t]=n,e}function a(e,t){if("function"!=typeof t&&null!==t)throw new TypeError("Super expression must either be null or a function");Object.defineProperty(e,"prototype",{value:Object.create(t&&t.prototype,{constructor:{value:e,writable:!0,configurable:!0}}),writable:!1}),t&&u(e,t)}function s(e){return s=Object.setPrototypeOf?Object.getPrototypeOf:function(e){return e.__proto__||Object.getPrototypeOf(e)},s(e)}function u(e,t){return u=Object.setPrototypeOf||function(e,t){return e.__proto__=t,e},u(e,t)}function h(e,t){if(t&&("object"==typeof t||"function"==typeof t))return t;if(void 0!==t)throw new TypeError("Derived constructors may only return object or undefined");return function(e){if(void 0===e)throw new ReferenceError("this hasn't been initialised - super() hasn't been called");return e}(e)}function l(e){var t=function(){if("undefined"==typeof Reflect||!Reflect.construct)return!1;if(Reflect.construct.sham)return!1;if("function"==typeof Proxy)return!0;try{return Boolean.prototype.valueOf.call(Reflect.construct(Boolean,[],(function(){}))),!0}catch(e){return!1}}();return function(){var n,r=s(e);if(t){var i=s(this).constructor;n=Reflect.construct(r,arguments,i)}else n=r.apply(this,arguments);return h(this,n)}}function f(e){return function(e){if(Array.isArray(e))return d(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(e){if("string"==typeof e)return d(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);return"Object"===n&&e.constructor&&(n=e.constructor.name),"Map"===n||"Set"===n?Array.from(e):"Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n)?d(e,t):void 0}}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function d(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function v(e){return Array.isArray?Array.isArray(e):"[object Array]"===b(e)}function g(e){return"string"==typeof e}function y(e){return"number"==typeof e}function p(e){return!0===e||!1===e||function(e){return m(e)&&null!==e}(e)&&"[object Boolean]"==b(e)}function m(e){return"object"===n(e)}function k(e){return null!=e}function M(e){return!e.trim().length}function b(e){return null==e?void 0===e?"[object Undefined]":"[object Null]":Object.prototype.toString.call(e)}var x=function(e){return"Invalid value for key ".concat(e)},w=function(e){return"Pattern length exceeds max of ".concat(e,".")},L=Object.prototype.hasOwnProperty,S=function(){function e(t){var n=this;r(this,e),this._keys=[],this._keyMap={};var i=0;t.forEach((function(e){var t=_(e);i+=t.weight,n._keys.push(t),n._keyMap[t.id]=t,i+=t.weight})),this._keys.forEach((function(e){e.weight/=i}))}return o(e,[{key:"get",value:function(e){return this._keyMap[e]}},{key:"keys",value:function(){return this._keys}},{key:"toJSON",value:function(){return JSON.stringify(this._keys)}}]),e}();function _(e){var t=null,n=null,r=null,i=1,o=null;if(g(e)||v(e))r=e,t=O(e),n=j(e);else{if(!L.call(e,"name"))throw new Error(function(e){return"Missing ".concat(e," property in key")}("name"));var c=e.name;if(r=c,L.call(e,"weight")&&(i=e.weight)<=0)throw new Error(function(e){return"Property 'weight' in key '".concat(e,"' must be a positive integer")}(c));t=O(c),n=j(c),o=e.getFn}return{path:t,id:n,weight:i,src:r,getFn:o}}function O(e){return v(e)?e:e.split(".")}function j(e){return v(e)?e.join("."):e}var A={useExtendedSearch:!1,getFn:function(e,t){var n=[],r=!1;return function e(t,i,o){if(k(t))if(i[o]){var c=t[i[o]];if(!k(c))return;if(o===i.length-1&&(g(c)||y(c)||p(c)))n.push(function(e){return null==e?"":function(e){if("string"==typeof e)return e;var t=e+"";return"0"==t&&1/e==-1/0?"-0":t}(e)}(c));else if(v(c)){r=!0;for(var a=0,s=c.length;a<s;a+=1)e(c[a],i,o+1)}else i.length&&e(c,i,o+1)}else n.push(t)}(e,g(t)?t.split("."):t,0),r?n:n[0]},ignoreLocation:!1,ignoreFieldNorm:!1,fieldNormWeight:1},I=t(t(t(t({},{isCaseSensitive:!1,includeScore:!1,keys:[],shouldSort:!0,sortFn:function(e,t){return e.score===t.score?e.idx<t.idx?-1:1:e.score<t.score?-1:1}}),{includeMatches:!1,findAllMatches:!1,minMatchCharLength:1}),{location:0,threshold:.6,distance:100}),A),C=/[^ ]+/g;function E(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:1,t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:3,n=new Map,r=Math.pow(10,t);return{get:function(t){var i=t.match(C).length;if(n.has(i))return n.get(i);var o=1/Math.pow(i,.5*e),c=parseFloat(Math.round(o*r)/r);return n.set(i,c),c},clear:function(){n.clear()}}}var $=function(){function e(){var t=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{},n=t.getFn,i=void 0===n?I.getFn:n,o=t.fieldNormWeight,c=void 0===o?I.fieldNormWeight:o;r(this,e),this.norm=E(c,3),this.getFn=i,this.isCreated=!1,this.setIndexRecords()}return o(e,[{key:"setSources",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:[];this.docs=e}},{key:"setIndexRecords",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:[];this.records=e}},{key:"setKeys",value:function(){var e=this,t=arguments.length>0&&void 0!==arguments[0]?arguments[0]:[];this.keys=t,this._keysMap={},t.forEach((function(t,n){e._keysMap[t.id]=n}))}},{key:"create",value:function(){var e=this;!this.isCreated&&this.docs.length&&(this.isCreated=!0,g(this.docs[0])?this.docs.forEach((function(t,n){e._addString(t,n)})):this.docs.forEach((function(t,n){e._addObject(t,n)})),this.norm.clear())}},{key:"add",value:function(e){var t=this.size();g(e)?this._addString(e,t):this._addObject(e,t)}},{key:"removeAt",value:function(e){this.records.splice(e,1);for(var t=e,n=this.size();t<n;t+=1)this.records[t].i-=1}},{key:"getValueForItemAtKeyId",value:function(e,t){return e[this._keysMap[t]]}},{key:"size",value:function(){return this.records.length}},{key:"_addString",value:function(e,t){if(k(e)&&!M(e)){var n={v:e,i:t,n:this.norm.get(e)};this.records.push(n)}}},{key:"_addObject",value:function(e,t){var n=this,r={i:t,$:{}};this.keys.forEach((function(t,i){var o=t.getFn?t.getFn(e):n.getFn(e,t.path);if(k(o))if(v(o))!function(){for(var e=[],t=[{nestedArrIndex:-1,value:o}];t.length;){var c=t.pop(),a=c.nestedArrIndex,s=c.value;if(k(s))if(g(s)&&!M(s)){var u={v:s,i:a,n:n.norm.get(s)};e.push(u)}else v(s)&&s.forEach((function(e,n){t.push({nestedArrIndex:n,value:e})}))}r.$[i]=e}();else if(g(o)&&!M(o)){var c={v:o,n:n.norm.get(o)};r.$[i]=c}})),this.records.push(r)}},{key:"toJSON",value:function(){return{keys:this.keys,records:this.records}}}]),e}();function F(e,t){var n=arguments.length>2&&void 0!==arguments[2]?arguments[2]:{},r=n.getFn,i=void 0===r?I.getFn:r,o=n.fieldNormWeight,c=void 0===o?I.fieldNormWeight:o,a=new $({getFn:i,fieldNormWeight:c});return a.setKeys(e.map(_)),a.setSources(t),a.create(),a}function R(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},n=t.errors,r=void 0===n?0:n,i=t.currentLocation,o=void 0===i?0:i,c=t.expectedLocation,a=void 0===c?0:c,s=t.distance,u=void 0===s?I.distance:s,h=t.ignoreLocation,l=void 0===h?I.ignoreLocation:h,f=r/e.length;if(l)return f;var d=Math.abs(a-o);return u?f+d/u:d?1:f}function N(){for(var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:[],t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:I.minMatchCharLength,n=[],r=-1,i=-1,o=0,c=e.length;o<c;o+=1){var a=e[o];a&&-1===r?r=o:a||-1===r||((i=o-1)-r+1>=t&&n.push([r,i]),r=-1)}return e[o-1]&&o-r>=t&&n.push([r,o-1]),n}var P=32;function W(e){for(var t={},n=0,r=e.length;n<r;n+=1){var i=e.charAt(n);t[i]=(t[i]||0)|1<<r-n-1}return t}var T=function(){function e(t){var n=this,i=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},o=i.location,c=void 0===o?I.location:o,a=i.threshold,s=void 0===a?I.threshold:a,u=i.distance,h=void 0===u?I.distance:u,l=i.includeMatches,f=void 0===l?I.includeMatches:l,d=i.findAllMatches,v=void 0===d?I.findAllMatches:d,g=i.minMatchCharLength,y=void 0===g?I.minMatchCharLength:g,p=i.isCaseSensitive,m=void 0===p?I.isCaseSensitive:p,k=i.ignoreLocation,M=void 0===k?I.ignoreLocation:k;if(r(this,e),this.options={location:c,threshold:s,distance:h,includeMatches:f,findAllMatches:v,minMatchCharLength:y,isCaseSensitive:m,ignoreLocation:M},this.pattern=m?t:t.toLowerCase(),this.chunks=[],this.pattern.length){var b=function(e,t){n.chunks.push({pattern:e,alphabet:W(e),startIndex:t})},x=this.pattern.length;if(x>P){for(var w=0,L=x%P,S=x-L;w<S;)b(this.pattern.substr(w,P),w),w+=P;if(L){var _=x-P;b(this.pattern.substr(_),_)}}else b(this.pattern,0)}}return o(e,[{key:"searchIn",value:function(e){var t=this.options,n=t.isCaseSensitive,r=t.includeMatches;if(n||(e=e.toLowerCase()),this.pattern===e){var i={isMatch:!0,score:0};return r&&(i.indices=[[0,e.length-1]]),i}var o=this.options,c=o.location,a=o.distance,s=o.threshold,u=o.findAllMatches,h=o.minMatchCharLength,l=o.ignoreLocation,d=[],v=0,g=!1;this.chunks.forEach((function(t){var n=t.pattern,i=t.alphabet,o=t.startIndex,y=function(e,t,n){var r=arguments.length>3&&void 0!==arguments[3]?arguments[3]:{},i=r.location,o=void 0===i?I.location:i,c=r.distance,a=void 0===c?I.distance:c,s=r.threshold,u=void 0===s?I.threshold:s,h=r.findAllMatches,l=void 0===h?I.findAllMatches:h,f=r.minMatchCharLength,d=void 0===f?I.minMatchCharLength:f,v=r.includeMatches,g=void 0===v?I.includeMatches:v,y=r.ignoreLocation,p=void 0===y?I.ignoreLocation:y;if(t.length>P)throw new Error(w(P));for(var m,k=t.length,M=e.length,b=Math.max(0,Math.min(o,M)),x=u,L=b,S=d>1||g,_=S?Array(M):[];(m=e.indexOf(t,L))>-1;){var O=R(t,{currentLocation:m,expectedLocation:b,distance:a,ignoreLocation:p});if(x=Math.min(O,x),L=m+k,S)for(var j=0;j<k;)_[m+j]=1,j+=1}L=-1;for(var A=[],C=1,E=k+M,$=1<<k-1,F=0;F<k;F+=1){for(var W=0,T=E;W<T;)R(t,{errors:F,currentLocation:b+T,expectedLocation:b,distance:a,ignoreLocation:p})<=x?W=T:E=T,T=Math.floor((E-W)/2+W);E=T;var z=Math.max(1,b-T+1),D=l?M:Math.min(b+T,M)+k,K=Array(D+2);K[D+1]=(1<<F)-1;for(var q=D;q>=z;q-=1){var B=q-1,J=n[e.charAt(B)];if(S&&(_[B]=+!!J),K[q]=(K[q+1]<<1|1)&J,F&&(K[q]|=(A[q+1]|A[q])<<1|1|A[q+1]),K[q]&$&&(C=R(t,{errors:F,currentLocation:B,expectedLocation:b,distance:a,ignoreLocation:p}))<=x){if(x=C,(L=B)<=b)break;z=Math.max(1,2*b-L)}}if(R(t,{errors:F+1,currentLocation:b,expectedLocation:b,distance:a,ignoreLocation:p})>x)break;A=K}var U={isMatch:L>=0,score:Math.max(.001,C)};if(S){var V=N(_,d);V.length?g&&(U.indices=V):U.isMatch=!1}return U}(e,n,i,{location:c+o,distance:a,threshold:s,findAllMatches:u,minMatchCharLength:h,includeMatches:r,ignoreLocation:l}),p=y.isMatch,m=y.score,k=y.indices;p&&(g=!0),v+=m,p&&k&&(d=[].concat(f(d),f(k)))}));var y={isMatch:g,score:g?v/this.chunks.length:1};return g&&r&&(y.indices=d),y}}]),e}(),z=function(){function e(t){r(this,e),this.pattern=t}return o(e,[{key:"search",value:function(){}}],[{key:"isMultiMatch",value:function(e){return D(e,this.multiRegex)}},{key:"isSingleMatch",value:function(e){return D(e,this.singleRegex)}}]),e}();function D(e,t){var n=e.match(t);return n?n[1]:null}var K=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){var t=e===this.pattern;return{isMatch:t,score:t?0:1,indices:[0,this.pattern.length-1]}}}],[{key:"type",get:function(){return"exact"}},{key:"multiRegex",get:function(){return/^="(.*)"$/}},{key:"singleRegex",get:function(){return/^=(.*)$/}}]),n}(z),q=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){var t=-1===e.indexOf(this.pattern);return{isMatch:t,score:t?0:1,indices:[0,e.length-1]}}}],[{key:"type",get:function(){return"inverse-exact"}},{key:"multiRegex",get:function(){return/^!"(.*)"$/}},{key:"singleRegex",get:function(){return/^!(.*)$/}}]),n}(z),B=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){var t=e.startsWith(this.pattern);return{isMatch:t,score:t?0:1,indices:[0,this.pattern.length-1]}}}],[{key:"type",get:function(){return"prefix-exact"}},{key:"multiRegex",get:function(){return/^\^"(.*)"$/}},{key:"singleRegex",get:function(){return/^\^(.*)$/}}]),n}(z),J=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){var t=!e.startsWith(this.pattern);return{isMatch:t,score:t?0:1,indices:[0,e.length-1]}}}],[{key:"type",get:function(){return"inverse-prefix-exact"}},{key:"multiRegex",get:function(){return/^!\^"(.*)"$/}},{key:"singleRegex",get:function(){return/^!\^(.*)$/}}]),n}(z),U=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){var t=e.endsWith(this.pattern);return{isMatch:t,score:t?0:1,indices:[e.length-this.pattern.length,e.length-1]}}}],[{key:"type",get:function(){return"suffix-exact"}},{key:"multiRegex",get:function(){return/^"(.*)"\$$/}},{key:"singleRegex",get:function(){return/^(.*)\$$/}}]),n}(z),V=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){var t=!e.endsWith(this.pattern);return{isMatch:t,score:t?0:1,indices:[0,e.length-1]}}}],[{key:"type",get:function(){return"inverse-suffix-exact"}},{key:"multiRegex",get:function(){return/^!"(.*)"\$$/}},{key:"singleRegex",get:function(){return/^!(.*)\$$/}}]),n}(z),G=function(e){a(n,e);var t=l(n);function n(e){var i,o=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},c=o.location,a=void 0===c?I.location:c,s=o.threshold,u=void 0===s?I.threshold:s,h=o.distance,l=void 0===h?I.distance:h,f=o.includeMatches,d=void 0===f?I.includeMatches:f,v=o.findAllMatches,g=void 0===v?I.findAllMatches:v,y=o.minMatchCharLength,p=void 0===y?I.minMatchCharLength:y,m=o.isCaseSensitive,k=void 0===m?I.isCaseSensitive:m,M=o.ignoreLocation,b=void 0===M?I.ignoreLocation:M;return r(this,n),(i=t.call(this,e))._bitapSearch=new T(e,{location:a,threshold:u,distance:l,includeMatches:d,findAllMatches:g,minMatchCharLength:p,isCaseSensitive:k,ignoreLocation:b}),i}return o(n,[{key:"search",value:function(e){return this._bitapSearch.searchIn(e)}}],[{key:"type",get:function(){return"fuzzy"}},{key:"multiRegex",get:function(){return/^"(.*)"$/}},{key:"singleRegex",get:function(){return/^(.*)$/}}]),n}(z),H=function(e){a(n,e);var t=l(n);function n(e){return r(this,n),t.call(this,e)}return o(n,[{key:"search",value:function(e){for(var t,n=0,r=[],i=this.pattern.length;(t=e.indexOf(this.pattern,n))>-1;)n=t+i,r.push([t,n-1]);var o=!!r.length;return{isMatch:o,score:o?0:1,indices:r}}}],[{key:"type",get:function(){return"include"}},{key:"multiRegex",get:function(){return/^'"(.*)"$/}},{key:"singleRegex",get:function(){return/^'(.*)$/}}]),n}(z),Q=[K,H,B,J,V,U,q,G],X=Q.length,Y=/ +(?=(?:[^\"]*\"[^\"]*\")*[^\"]*$)/;function Z(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{};return e.split("|").map((function(e){for(var n=e.trim().split(Y).filter((function(e){return e&&!!e.trim()})),r=[],i=0,o=n.length;i<o;i+=1){for(var c=n[i],a=!1,s=-1;!a&&++s<X;){var u=Q[s],h=u.isMultiMatch(c);h&&(r.push(new u(h,t)),a=!0)}if(!a)for(s=-1;++s<X;){var l=Q[s],f=l.isSingleMatch(c);if(f){r.push(new l(f,t));break}}}return r}))}var ee=new Set([G.type,H.type]),te=function(){function e(t){var n=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},i=n.isCaseSensitive,o=void 0===i?I.isCaseSensitive:i,c=n.includeMatches,a=void 0===c?I.includeMatches:c,s=n.minMatchCharLength,u=void 0===s?I.minMatchCharLength:s,h=n.ignoreLocation,l=void 0===h?I.ignoreLocation:h,f=n.findAllMatches,d=void 0===f?I.findAllMatches:f,v=n.location,g=void 0===v?I.location:v,y=n.threshold,p=void 0===y?I.threshold:y,m=n.distance,k=void 0===m?I.distance:m;r(this,e),this.query=null,this.options={isCaseSensitive:o,includeMatches:a,minMatchCharLength:u,findAllMatches:d,ignoreLocation:l,location:g,threshold:p,distance:k},this.pattern=o?t:t.toLowerCase(),this.query=Z(this.pattern,this.options)}return o(e,[{key:"searchIn",value:function(e){var t=this.query;if(!t)return{isMatch:!1,score:1};var n=this.options,r=n.includeMatches;e=n.isCaseSensitive?e:e.toLowerCase();for(var i=0,o=[],c=0,a=0,s=t.length;a<s;a+=1){var u=t[a];o.length=0,i=0;for(var h=0,l=u.length;h<l;h+=1){var d=u[h],v=d.search(e),g=v.isMatch,y=v.indices,p=v.score;if(!g){c=0,i=0,o.length=0;break}if(i+=1,c+=p,r){var m=d.constructor.type;ee.has(m)?o=[].concat(f(o),f(y)):o.push(y)}}if(i){var k={isMatch:!0,score:c/i};return r&&(k.indices=o),k}}return{isMatch:!1,score:1}}}],[{key:"condition",value:function(e,t){return t.useExtendedSearch}}]),e}(),ne=[];function re(e,t){for(var n=0,r=ne.length;n<r;n+=1){var i=ne[n];if(i.condition(e,t))return new i(e,t)}return new T(e,t)}var ie="$and",oe="$or",ce="$path",ae="$val",se=function(e){return!(!e[ie]&&!e[oe])},ue=function(e){return!!e[ce]},he=function(e){return!v(e)&&m(e)&&!se(e)},le=function(e){return c({},ie,Object.keys(e).map((function(t){return c({},t,e[t])})))};function fe(e,t){var n=t.ignoreFieldNorm,r=void 0===n?I.ignoreFieldNorm:n;e.forEach((function(e){var t=1;e.matches.forEach((function(e){var n=e.key,i=e.norm,o=e.score,c=n?n.weight:null;t*=Math.pow(0===o&&c?Number.EPSILON:o,(c||1)*(r?1:i))})),e.score=t}))}function de(e,t){var n=e.matches;t.matches=[],k(n)&&n.forEach((function(e){if(k(e.indices)&&e.indices.length){var n={indices:e.indices,value:e.value};e.key&&(n.key=e.key.src),e.idx>-1&&(n.refIndex=e.idx),t.matches.push(n)}}))}function ve(e,t){t.score=e.score}function ge(e,t){var n=arguments.length>2&&void 0!==arguments[2]?arguments[2]:{},r=n.includeMatches,i=void 0===r?I.includeMatches:r,o=n.includeScore,c=void 0===o?I.includeScore:o,a=[];return i&&a.push(de),c&&a.push(ve),e.map((function(e){var n=e.idx,r={item:t[n],refIndex:n};return a.length&&a.forEach((function(t){t(e,r)})),r}))}var ye=function(){function e(n){var i=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},o=arguments.length>2?arguments[2]:void 0;r(this,e),this.options=t(t({},I),i),this.options.useExtendedSearch,this._keyStore=new S(this.options.keys),this.setCollection(n,o)}return o(e,[{key:"setCollection",value:function(e,t){if(this._docs=e,t&&!(t instanceof $))throw new Error("Incorrect 'index' type");this._myIndex=t||F(this.options.keys,this._docs,{getFn:this.options.getFn,fieldNormWeight:this.options.fieldNormWeight})}},{key:"add",value:function(e){k(e)&&(this._docs.push(e),this._myIndex.add(e))}},{key:"remove",value:function(){for(var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:function(){return!1},t=[],n=0,r=this._docs.length;n<r;n+=1){var i=this._docs[n];e(i,n)&&(this.removeAt(n),n-=1,r-=1,t.push(i))}return t}},{key:"removeAt",value:function(e){this._docs.splice(e,1),this._myIndex.removeAt(e)}},{key:"getIndex",value:function(){return this._myIndex}},{key:"search",value:function(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},n=t.limit,r=void 0===n?-1:n,i=this.options,o=i.includeMatches,c=i.includeScore,a=i.shouldSort,s=i.sortFn,u=i.ignoreFieldNorm,h=g(e)?g(this._docs[0])?this._searchStringList(e):this._searchObjectList(e):this._searchLogical(e);return fe(h,{ignoreFieldNorm:u}),a&&h.sort(s),y(r)&&r>-1&&(h=h.slice(0,r)),ge(h,this._docs,{includeMatches:o,includeScore:c})}},{key:"_searchStringList",value:function(e){var t=re(e,this.options),n=this._myIndex.records,r=[];return n.forEach((function(e){var n=e.v,i=e.i,o=e.n;if(k(n)){var c=t.searchIn(n),a=c.isMatch,s=c.score,u=c.indices;a&&r.push({item:n,idx:i,matches:[{score:s,value:n,norm:o,indices:u}]})}})),r}},{key:"_searchLogical",value:function(e){var t=this,n=function(e,t){var n=(arguments.length>2&&void 0!==arguments[2]?arguments[2]:{}).auto,r=void 0===n||n,i=function e(n){var i=Object.keys(n),o=ue(n);if(!o&&i.length>1&&!se(n))return e(le(n));if(he(n)){var c=o?n[ce]:i[0],a=o?n[ae]:n[c];if(!g(a))throw new Error(x(c));var s={keyId:j(c),pattern:a};return r&&(s.searcher=re(a,t)),s}var u={children:[],operator:i[0]};return i.forEach((function(t){var r=n[t];v(r)&&r.forEach((function(t){u.children.push(e(t))}))})),u};return se(e)||(e=le(e)),i(e)}(e,this.options),r=function e(n,r,i){if(!n.children){var o=n.keyId,c=n.searcher,a=t._findMatches({key:t._keyStore.get(o),value:t._myIndex.getValueForItemAtKeyId(r,o),searcher:c});return a&&a.length?[{idx:i,item:r,matches:a}]:[]}for(var s=[],u=0,h=n.children.length;u<h;u+=1){var l=e(n.children[u],r,i);if(l.length)s.push.apply(s,f(l));else if(n.operator===ie)return[]}return s},i=this._myIndex.records,o={},c=[];return i.forEach((function(e){var t=e.$,i=e.i;if(k(t)){var a=r(n,t,i);a.length&&(o[i]||(o[i]={idx:i,item:t,matches:[]},c.push(o[i])),a.forEach((function(e){var t,n=e.matches;(t=o[i].matches).push.apply(t,f(n))})))}})),c}},{key:"_searchObjectList",value:function(e){var t=this,n=re(e,this.options),r=this._myIndex,i=r.keys,o=r.records,c=[];return o.forEach((function(e){var r=e.$,o=e.i;if(k(r)){var a=[];i.forEach((function(e,i){a.push.apply(a,f(t._findMatches({key:e,value:r[i],searcher:n})))})),a.length&&c.push({idx:o,item:r,matches:a})}})),c}},{key:"_findMatches",value:function(e){var t=e.key,n=e.value,r=e.searcher;if(!k(n))return[];var i=[];if(v(n))n.forEach((function(e){var n=e.v,o=e.i,c=e.n;if(k(n)){var a=r.searchIn(n),s=a.isMatch,u=a.score,h=a.indices;s&&i.push({score:u,key:t,value:n,idx:o,norm:c,indices:h})}}));else{var o=n.v,c=n.n,a=r.searchIn(o),s=a.isMatch,u=a.score,h=a.indices;s&&i.push({score:u,key:t,value:o,norm:c,indices:h})}return i}}]),e}();return ye.version="6.6.2",ye.createIndex=F,ye.parseIndex=function(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:{},n=t.getFn,r=void 0===n?I.getFn:n,i=t.fieldNormWeight,o=void 0===i?I.fieldNormWeight:i,c=e.keys,a=e.records,s=new $({getFn:r,fieldNormWeight:o});return s.setKeys(c),s.setIndexRecords(a),s},ye.config=I,function(){ne.push.apply(ne,arguments)}(te),ye},"object"==typeof exports&&"undefined"!=typeof module?module.exports=t():"function"==typeof define&&define.amd?define(t):(e="undefined"!=typeof globalThis?globalThis:e||self).Fuse=t();
\ No newline at end of file
diff --git a/docs/2_39/site_libs/quarto-search/quarto-search.js b/docs/2_39/site_libs/quarto-search/quarto-search.js
new file mode 100644
index 000000000..d788a9581
--- /dev/null
+++ b/docs/2_39/site_libs/quarto-search/quarto-search.js
@@ -0,0 +1,1290 @@
+const kQueryArg = "q";
+const kResultsArg = "show-results";
+
+// If items don't provide a URL, then both the navigator and the onSelect
+// function aren't called (and therefore, the default implementation is used)
+//
+// We're using this sentinel URL to signal to those handlers that this
+// item is a more item (along with the type) and can be handled appropriately
+const kItemTypeMoreHref = "0767FDFD-0422-4E5A-BC8A-3BE11E5BBA05";
+
+window.document.addEventListener("DOMContentLoaded", function (_event) {
+  // Ensure that search is available on this page. If it isn't,
+  // should return early and not do anything
+  var searchEl = window.document.getElementById("quarto-search");
+  if (!searchEl) return;
+
+  const { autocomplete } = window["@algolia/autocomplete-js"];
+
+  let quartoSearchOptions = {};
+  let language = {};
+  const searchOptionEl = window.document.getElementById(
+    "quarto-search-options"
+  );
+  if (searchOptionEl) {
+    const jsonStr = searchOptionEl.textContent;
+    quartoSearchOptions = JSON.parse(jsonStr);
+    language = quartoSearchOptions.language;
+  }
+
+  // note the search mode
+  if (quartoSearchOptions.type === "overlay") {
+    searchEl.classList.add("type-overlay");
+  } else {
+    searchEl.classList.add("type-textbox");
+  }
+
+  // Used to determine highlighting behavior for this page
+  // A `q` query param is expected when the user follows a search
+  // to this page
+  const currentUrl = new URL(window.location);
+  const query = currentUrl.searchParams.get(kQueryArg);
+  const showSearchResults = currentUrl.searchParams.get(kResultsArg);
+  const mainEl = window.document.querySelector("main");
+
+  // highlight matches on the page
+  if (query && mainEl) {
+    // perform any highlighting
+    highlight(escapeRegExp(query), mainEl);
+
+    // fix up the URL to remove the q query param
+    const replacementUrl = new URL(window.location);
+    replacementUrl.searchParams.delete(kQueryArg);
+    window.history.replaceState({}, "", replacementUrl);
+  }
+
+  // function to clear highlighting on the page when the search query changes
+  // (e.g. if the user edits the query or clears it)
+  let highlighting = true;
+  const resetHighlighting = (searchTerm) => {
+    if (mainEl && highlighting && query && searchTerm !== query) {
+      clearHighlight(query, mainEl);
+      highlighting = false;
+    }
+  };
+
+  // Clear search highlighting when the user scrolls sufficiently
+  const resetFn = () => {
+    resetHighlighting("");
+    window.removeEventListener("quarto-hrChanged", resetFn);
+    window.removeEventListener("quarto-sectionChanged", resetFn);
+  };
+
+  // Register this event after the initial scrolling and settling of events
+  // on the page
+  window.addEventListener("quarto-hrChanged", resetFn);
+  window.addEventListener("quarto-sectionChanged", resetFn);
+
+  // Responsively switch to overlay mode if the search is present on the navbar
+  // Note that switching the sidebar to overlay mode requires more coordinate (not just
+  // the media query since we generate different HTML for sidebar overlays than we do
+  // for sidebar input UI)
+  const detachedMediaQuery =
+    quartoSearchOptions.type === "overlay" ? "all" : "(max-width: 991px)";
+
+  // If configured, include the analytics client to send insights
+  const plugins = configurePlugins(quartoSearchOptions);
+
+  let lastState = null;
+  const { setIsOpen, setQuery, setCollections } = autocomplete({
+    container: searchEl,
+    detachedMediaQuery: detachedMediaQuery,
+    defaultActiveItemId: 0,
+    panelContainer: "#quarto-search-results",
+    panelPlacement: quartoSearchOptions["panel-placement"],
+    debug: false,
+    openOnFocus: true,
+    plugins,
+    classNames: {
+      form: "d-flex",
+    },
+    placeholder: language["search-text-placeholder"],
+    translations: {
+      clearButtonTitle: language["search-clear-button-title"],
+      detachedCancelButtonText: language["search-detached-cancel-button-title"],
+      submitButtonTitle: language["search-submit-button-title"],
+    },
+    initialState: {
+      query,
+    },
+    getItemUrl({ item }) {
+      return item.href;
+    },
+    onStateChange({ state }) {
+      // If this is a file URL, note that
+
+      // Perhaps reset highlighting
+      resetHighlighting(state.query);
+
+      // If the panel just opened, ensure the panel is positioned properly
+      if (state.isOpen) {
+        if (lastState && !lastState.isOpen) {
+          setTimeout(() => {
+            positionPanel(quartoSearchOptions["panel-placement"]);
+          }, 150);
+        }
+      }
+
+      // Perhaps show the copy link
+      showCopyLink(state.query, quartoSearchOptions);
+
+      lastState = state;
+    },
+    reshape({ sources, state }) {
+      return sources.map((source) => {
+        try {
+          const items = source.getItems();
+
+          // Validate the items
+          validateItems(items);
+
+          // group the items by document
+          const groupedItems = new Map();
+          items.forEach((item) => {
+            const hrefParts = item.href.split("#");
+            const baseHref = hrefParts[0];
+            const isDocumentItem = hrefParts.length === 1;
+
+            const items = groupedItems.get(baseHref);
+            if (!items) {
+              groupedItems.set(baseHref, [item]);
+            } else {
+              // If the href for this item matches the document
+              // exactly, place this item first as it is the item that represents
+              // the document itself
+              if (isDocumentItem) {
+                items.unshift(item);
+              } else {
+                items.push(item);
+              }
+              groupedItems.set(baseHref, items);
+            }
+          });
+
+          const reshapedItems = [];
+          let count = 1;
+          for (const [_key, value] of groupedItems) {
+            const firstItem = value[0];
+            reshapedItems.push({
+              ...firstItem,
+              type: kItemTypeDoc,
+            });
+
+            const collapseMatches = quartoSearchOptions["collapse-after"];
+            const collapseCount =
+              typeof collapseMatches === "number" ? collapseMatches : 1;
+
+            if (value.length > 1) {
+              const target = `search-more-${count}`;
+              const isExpanded =
+                state.context.expanded &&
+                state.context.expanded.includes(target);
+
+              const remainingCount = value.length - collapseCount;
+
+              for (let i = 1; i < value.length; i++) {
+                if (collapseMatches && i === collapseCount) {
+                  reshapedItems.push({
+                    target,
+                    title: isExpanded
+                      ? language["search-hide-matches-text"]
+                      : remainingCount === 1
+                      ? `${remainingCount} ${language["search-more-match-text"]}`
+                      : `${remainingCount} ${language["search-more-matches-text"]}`,
+                    type: kItemTypeMore,
+                    href: kItemTypeMoreHref,
+                  });
+                }
+
+                if (isExpanded || !collapseMatches || i < collapseCount) {
+                  reshapedItems.push({
+                    ...value[i],
+                    type: kItemTypeItem,
+                    target,
+                  });
+                }
+              }
+            }
+            count += 1;
+          }
+
+          return {
+            ...source,
+            getItems() {
+              return reshapedItems;
+            },
+          };
+        } catch (error) {
+          // Some form of error occurred
+          return {
+            ...source,
+            getItems() {
+              return [
+                {
+                  title: error.name || "An Error Occurred While Searching",
+                  text:
+                    error.message ||
+                    "An unknown error occurred while attempting to perform the requested search.",
+                  type: kItemTypeError,
+                },
+              ];
+            },
+          };
+        }
+      });
+    },
+    navigator: {
+      navigate({ itemUrl }) {
+        if (itemUrl !== offsetURL(kItemTypeMoreHref)) {
+          window.location.assign(itemUrl);
+        }
+      },
+      navigateNewTab({ itemUrl }) {
+        if (itemUrl !== offsetURL(kItemTypeMoreHref)) {
+          const windowReference = window.open(itemUrl, "_blank", "noopener");
+          if (windowReference) {
+            windowReference.focus();
+          }
+        }
+      },
+      navigateNewWindow({ itemUrl }) {
+        if (itemUrl !== offsetURL(kItemTypeMoreHref)) {
+          window.open(itemUrl, "_blank", "noopener");
+        }
+      },
+    },
+    getSources({ state, setContext, setActiveItemId, refresh }) {
+      return [
+        {
+          sourceId: "documents",
+          getItemUrl({ item }) {
+            if (item.href) {
+              return offsetURL(item.href);
+            } else {
+              return undefined;
+            }
+          },
+          onSelect({
+            item,
+            state,
+            setContext,
+            setIsOpen,
+            setActiveItemId,
+            refresh,
+          }) {
+            if (item.type === kItemTypeMore) {
+              toggleExpanded(item, state, setContext, setActiveItemId, refresh);
+
+              // Toggle more
+              setIsOpen(true);
+            }
+          },
+          getItems({ query }) {
+            if (query === null || query === "") {
+              return [];
+            }
+
+            const limit = quartoSearchOptions.limit;
+            if (quartoSearchOptions.algolia) {
+              return algoliaSearch(query, limit, quartoSearchOptions.algolia);
+            } else {
+              // Fuse search options
+              const fuseSearchOptions = {
+                isCaseSensitive: false,
+                shouldSort: true,
+                minMatchCharLength: 2,
+                limit: limit,
+              };
+
+              return readSearchData().then(function (fuse) {
+                return fuseSearch(query, fuse, fuseSearchOptions);
+              });
+            }
+          },
+          templates: {
+            noResults({ createElement }) {
+              const hasQuery = lastState.query;
+
+              return createElement(
+                "div",
+                {
+                  class: `quarto-search-no-results${
+                    hasQuery ? "" : " no-query"
+                  }`,
+                },
+                language["search-no-results-text"]
+              );
+            },
+            header({ items, createElement }) {
+              // count the documents
+              const count = items.filter((item) => {
+                return item.type === kItemTypeDoc;
+              }).length;
+
+              if (count > 0) {
+                return createElement(
+                  "div",
+                  { class: "search-result-header" },
+                  `${count} ${language["search-matching-documents-text"]}`
+                );
+              } else {
+                return createElement(
+                  "div",
+                  { class: "search-result-header-no-results" },
+                  ``
+                );
+              }
+            },
+            footer({ _items, createElement }) {
+              if (
+                quartoSearchOptions.algolia &&
+                quartoSearchOptions.algolia["show-logo"]
+              ) {
+                const libDir = quartoSearchOptions.algolia["libDir"];
+                const logo = createElement("img", {
+                  src: offsetURL(
+                    `${libDir}/quarto-search/search-by-algolia.svg`
+                  ),
+                  class: "algolia-search-logo",
+                });
+                return createElement(
+                  "a",
+                  { href: "http://www.algolia.com/" },
+                  logo
+                );
+              }
+            },
+
+            item({ item, createElement }) {
+              return renderItem(
+                item,
+                createElement,
+                state,
+                setActiveItemId,
+                setContext,
+                refresh,
+                quartoSearchOptions
+              );
+            },
+          },
+        },
+      ];
+    },
+  });
+
+  window.quartoOpenSearch = () => {
+    setIsOpen(false);
+    setIsOpen(true);
+    focusSearchInput();
+  };
+
+  document.addEventListener("keyup", (event) => {
+    const { key } = event;
+    const kbds = quartoSearchOptions["keyboard-shortcut"];
+    const focusedEl = document.activeElement;
+
+    const isFormElFocused = [
+      "input",
+      "select",
+      "textarea",
+      "button",
+      "option",
+    ].find((tag) => {
+      return focusedEl.tagName.toLowerCase() === tag;
+    });
+
+    if (
+      kbds &&
+      kbds.includes(key) &&
+      !isFormElFocused &&
+      !document.activeElement.isContentEditable
+    ) {
+      event.preventDefault();
+      window.quartoOpenSearch();
+    }
+  });
+
+  // Remove the labeleledby attribute since it is pointing
+  // to a non-existent label
+  if (quartoSearchOptions.type === "overlay") {
+    const inputEl = window.document.querySelector(
+      "#quarto-search .aa-Autocomplete"
+    );
+    if (inputEl) {
+      inputEl.removeAttribute("aria-labelledby");
+    }
+  }
+
+  function throttle(func, wait) {
+    let waiting = false;
+    return function () {
+      if (!waiting) {
+        func.apply(this, arguments);
+        waiting = true;
+        setTimeout(function () {
+          waiting = false;
+        }, wait);
+      }
+    };
+  }
+
+  // If the main document scrolls dismiss the search results
+  // (otherwise, since they're floating in the document they can scroll with the document)
+  window.document.body.onscroll = throttle(() => {
+    // Only do this if we're not detached
+    // Bug #7117
+    // This will happen when the keyboard is shown on ios (resulting in a scroll)
+    // which then closed the search UI
+    if (!window.matchMedia(detachedMediaQuery).matches) {
+      setIsOpen(false);
+    }
+  }, 50);
+
+  if (showSearchResults) {
+    setIsOpen(true);
+    focusSearchInput();
+  }
+});
+
+function configurePlugins(quartoSearchOptions) {
+  const autocompletePlugins = [];
+  const algoliaOptions = quartoSearchOptions.algolia;
+  if (
+    algoliaOptions &&
+    algoliaOptions["analytics-events"] &&
+    algoliaOptions["search-only-api-key"] &&
+    algoliaOptions["application-id"]
+  ) {
+    const apiKey = algoliaOptions["search-only-api-key"];
+    const appId = algoliaOptions["application-id"];
+
+    // Aloglia insights may not be loaded because they require cookie consent
+    // Use deferred loading so events will start being recorded when/if consent
+    // is granted.
+    const algoliaInsightsDeferredPlugin = deferredLoadPlugin(() => {
+      if (
+        window.aa &&
+        window["@algolia/autocomplete-plugin-algolia-insights"]
+      ) {
+        window.aa("init", {
+          appId,
+          apiKey,
+          useCookie: true,
+        });
+
+        const { createAlgoliaInsightsPlugin } =
+          window["@algolia/autocomplete-plugin-algolia-insights"];
+        // Register the insights client
+        const algoliaInsightsPlugin = createAlgoliaInsightsPlugin({
+          insightsClient: window.aa,
+          onItemsChange({ insights, insightsEvents }) {
+            const events = insightsEvents.flatMap((event) => {
+              // This API limits the number of items per event to 20
+              const chunkSize = 20;
+              const itemChunks = [];
+              const eventItems = event.items;
+              for (let i = 0; i < eventItems.length; i += chunkSize) {
+                itemChunks.push(eventItems.slice(i, i + chunkSize));
+              }
+              // Split the items into multiple events that can be sent
+              const events = itemChunks.map((items) => {
+                return {
+                  ...event,
+                  items,
+                };
+              });
+              return events;
+            });
+
+            for (const event of events) {
+              insights.viewedObjectIDs(event);
+            }
+          },
+        });
+        return algoliaInsightsPlugin;
+      }
+    });
+
+    // Add the plugin
+    autocompletePlugins.push(algoliaInsightsDeferredPlugin);
+    return autocompletePlugins;
+  }
+}
+
+// For plugins that may not load immediately, create a wrapper
+// plugin and forward events and plugin data once the plugin
+// is initialized. This is useful for cases like cookie consent
+// which may prevent the analytics insights event plugin from initializing
+// immediately.
+function deferredLoadPlugin(createPlugin) {
+  let plugin = undefined;
+  let subscribeObj = undefined;
+  const wrappedPlugin = () => {
+    if (!plugin && subscribeObj) {
+      plugin = createPlugin();
+      if (plugin && plugin.subscribe) {
+        plugin.subscribe(subscribeObj);
+      }
+    }
+    return plugin;
+  };
+
+  return {
+    subscribe: (obj) => {
+      subscribeObj = obj;
+    },
+    onStateChange: (obj) => {
+      const plugin = wrappedPlugin();
+      if (plugin && plugin.onStateChange) {
+        plugin.onStateChange(obj);
+      }
+    },
+    onSubmit: (obj) => {
+      const plugin = wrappedPlugin();
+      if (plugin && plugin.onSubmit) {
+        plugin.onSubmit(obj);
+      }
+    },
+    onReset: (obj) => {
+      const plugin = wrappedPlugin();
+      if (plugin && plugin.onReset) {
+        plugin.onReset(obj);
+      }
+    },
+    getSources: (obj) => {
+      const plugin = wrappedPlugin();
+      if (plugin && plugin.getSources) {
+        return plugin.getSources(obj);
+      } else {
+        return Promise.resolve([]);
+      }
+    },
+    data: (obj) => {
+      const plugin = wrappedPlugin();
+      if (plugin && plugin.data) {
+        plugin.data(obj);
+      }
+    },
+  };
+}
+
+function validateItems(items) {
+  // Validate the first item
+  if (items.length > 0) {
+    const item = items[0];
+    const missingFields = [];
+    if (item.href == undefined) {
+      missingFields.push("href");
+    }
+    if (!item.title == undefined) {
+      missingFields.push("title");
+    }
+    if (!item.text == undefined) {
+      missingFields.push("text");
+    }
+
+    if (missingFields.length === 1) {
+      throw {
+        name: `Error: Search index is missing the <code>${missingFields[0]}</code> field.`,
+        message: `The items being returned for this search do not include all the required fields. Please ensure that your index items include the <code>${missingFields[0]}</code> field or use <code>index-fields</code> in your <code>_quarto.yml</code> file to specify the field names.`,
+      };
+    } else if (missingFields.length > 1) {
+      const missingFieldList = missingFields
+        .map((field) => {
+          return `<code>${field}</code>`;
+        })
+        .join(", ");
+
+      throw {
+        name: `Error: Search index is missing the following fields: ${missingFieldList}.`,
+        message: `The items being returned for this search do not include all the required fields. Please ensure that your index items includes the following fields: ${missingFieldList}, or use <code>index-fields</code> in your <code>_quarto.yml</code> file to specify the field names.`,
+      };
+    }
+  }
+}
+
+let lastQuery = null;
+function showCopyLink(query, options) {
+  const language = options.language;
+  lastQuery = query;
+  // Insert share icon
+  const inputSuffixEl = window.document.body.querySelector(
+    ".aa-Form .aa-InputWrapperSuffix"
+  );
+
+  if (inputSuffixEl) {
+    let copyButtonEl = window.document.body.querySelector(
+      ".aa-Form .aa-InputWrapperSuffix .aa-CopyButton"
+    );
+
+    if (copyButtonEl === null) {
+      copyButtonEl = window.document.createElement("button");
+      copyButtonEl.setAttribute("class", "aa-CopyButton");
+      copyButtonEl.setAttribute("type", "button");
+      copyButtonEl.setAttribute("title", language["search-copy-link-title"]);
+      copyButtonEl.onmousedown = (e) => {
+        e.preventDefault();
+        e.stopPropagation();
+      };
+
+      const linkIcon = "bi-clipboard";
+      const checkIcon = "bi-check2";
+
+      const shareIconEl = window.document.createElement("i");
+      shareIconEl.setAttribute("class", `bi ${linkIcon}`);
+      copyButtonEl.appendChild(shareIconEl);
+      inputSuffixEl.prepend(copyButtonEl);
+
+      const clipboard = new window.ClipboardJS(".aa-CopyButton", {
+        text: function (_trigger) {
+          const copyUrl = new URL(window.location);
+          copyUrl.searchParams.set(kQueryArg, lastQuery);
+          copyUrl.searchParams.set(kResultsArg, "1");
+          return copyUrl.toString();
+        },
+      });
+      clipboard.on("success", function (e) {
+        // Focus the input
+
+        // button target
+        const button = e.trigger;
+        const icon = button.querySelector("i.bi");
+
+        // flash "checked"
+        icon.classList.add(checkIcon);
+        icon.classList.remove(linkIcon);
+        setTimeout(function () {
+          icon.classList.remove(checkIcon);
+          icon.classList.add(linkIcon);
+        }, 1000);
+      });
+    }
+
+    // If there is a query, show the link icon
+    if (copyButtonEl) {
+      if (lastQuery && options["copy-button"]) {
+        copyButtonEl.style.display = "flex";
+      } else {
+        copyButtonEl.style.display = "none";
+      }
+    }
+  }
+}
+
+/* Search Index Handling */
+// create the index
+var fuseIndex = undefined;
+var shownWarning = false;
+
+// fuse index options
+const kFuseIndexOptions = {
+  keys: [
+    { name: "title", weight: 20 },
+    { name: "section", weight: 20 },
+    { name: "text", weight: 10 },
+  ],
+  ignoreLocation: true,
+  threshold: 0.1,
+};
+
+async function readSearchData() {
+  // Initialize the search index on demand
+  if (fuseIndex === undefined) {
+    if (window.location.protocol === "file:" && !shownWarning) {
+      window.alert(
+        "Search requires JavaScript features disabled when running in file://... URLs. In order to use search, please run this document in a web server."
+      );
+      shownWarning = true;
+      return;
+    }
+    const fuse = new window.Fuse([], kFuseIndexOptions);
+
+    // fetch the main search.json
+    const response = await fetch(offsetURL("search.json"));
+    if (response.status == 200) {
+      return response.json().then(function (searchDocs) {
+        searchDocs.forEach(function (searchDoc) {
+          fuse.add(searchDoc);
+        });
+        fuseIndex = fuse;
+        return fuseIndex;
+      });
+    } else {
+      return Promise.reject(
+        new Error(
+          "Unexpected status from search index request: " + response.status
+        )
+      );
+    }
+  }
+
+  return fuseIndex;
+}
+
+function inputElement() {
+  return window.document.body.querySelector(".aa-Form .aa-Input");
+}
+
+function focusSearchInput() {
+  setTimeout(() => {
+    const inputEl = inputElement();
+    if (inputEl) {
+      inputEl.focus();
+    }
+  }, 50);
+}
+
+/* Panels */
+const kItemTypeDoc = "document";
+const kItemTypeMore = "document-more";
+const kItemTypeItem = "document-item";
+const kItemTypeError = "error";
+
+function renderItem(
+  item,
+  createElement,
+  state,
+  setActiveItemId,
+  setContext,
+  refresh,
+  quartoSearchOptions
+) {
+  switch (item.type) {
+    case kItemTypeDoc:
+      return createDocumentCard(
+        createElement,
+        "file-richtext",
+        item.title,
+        item.section,
+        item.text,
+        item.href,
+        item.crumbs,
+        quartoSearchOptions
+      );
+    case kItemTypeMore:
+      return createMoreCard(
+        createElement,
+        item,
+        state,
+        setActiveItemId,
+        setContext,
+        refresh
+      );
+    case kItemTypeItem:
+      return createSectionCard(
+        createElement,
+        item.section,
+        item.text,
+        item.href
+      );
+    case kItemTypeError:
+      return createErrorCard(createElement, item.title, item.text);
+    default:
+      return undefined;
+  }
+}
+
+function createDocumentCard(
+  createElement,
+  icon,
+  title,
+  section,
+  text,
+  href,
+  crumbs,
+  quartoSearchOptions
+) {
+  const iconEl = createElement("i", {
+    class: `bi bi-${icon} search-result-icon`,
+  });
+  const titleEl = createElement("p", { class: "search-result-title" }, title);
+  const titleContents = [iconEl, titleEl];
+  const showParent = quartoSearchOptions["show-item-context"];
+  if (crumbs && showParent) {
+    let crumbsOut = undefined;
+    const crumbClz = ["search-result-crumbs"];
+    if (showParent === "root") {
+      crumbsOut = crumbs.length > 1 ? crumbs[0] : undefined;
+    } else if (showParent === "parent") {
+      crumbsOut = crumbs.length > 1 ? crumbs[crumbs.length - 2] : undefined;
+    } else {
+      crumbsOut = crumbs.length > 1 ? crumbs.join(" > ") : undefined;
+      crumbClz.push("search-result-crumbs-wrap");
+    }
+
+    const crumbEl = createElement(
+      "p",
+      { class: crumbClz.join(" ") },
+      crumbsOut
+    );
+    titleContents.push(crumbEl);
+  }
+
+  const titleContainerEl = createElement(
+    "div",
+    { class: "search-result-title-container" },
+    titleContents
+  );
+
+  const textEls = [];
+  if (section) {
+    const sectionEl = createElement(
+      "p",
+      { class: "search-result-section" },
+      section
+    );
+    textEls.push(sectionEl);
+  }
+  const descEl = createElement("p", {
+    class: "search-result-text",
+    dangerouslySetInnerHTML: {
+      __html: text,
+    },
+  });
+  textEls.push(descEl);
+
+  const textContainerEl = createElement(
+    "div",
+    { class: "search-result-text-container" },
+    textEls
+  );
+
+  const containerEl = createElement(
+    "div",
+    {
+      class: "search-result-container",
+    },
+    [titleContainerEl, textContainerEl]
+  );
+
+  const linkEl = createElement(
+    "a",
+    {
+      href: offsetURL(href),
+      class: "search-result-link",
+    },
+    containerEl
+  );
+
+  const classes = ["search-result-doc", "search-item"];
+  if (!section) {
+    classes.push("document-selectable");
+  }
+
+  return createElement(
+    "div",
+    {
+      class: classes.join(" "),
+    },
+    linkEl
+  );
+}
+
+function createMoreCard(
+  createElement,
+  item,
+  state,
+  setActiveItemId,
+  setContext,
+  refresh
+) {
+  const moreCardEl = createElement(
+    "div",
+    {
+      class: "search-result-more search-item",
+      onClick: (e) => {
+        // Handle expanding the sections by adding the expanded
+        // section to the list of expanded sections
+        toggleExpanded(item, state, setContext, setActiveItemId, refresh);
+        e.stopPropagation();
+      },
+    },
+    item.title
+  );
+
+  return moreCardEl;
+}
+
+function toggleExpanded(item, state, setContext, setActiveItemId, refresh) {
+  const expanded = state.context.expanded || [];
+  if (expanded.includes(item.target)) {
+    setContext({
+      expanded: expanded.filter((target) => target !== item.target),
+    });
+  } else {
+    setContext({ expanded: [...expanded, item.target] });
+  }
+
+  refresh();
+  setActiveItemId(item.__autocomplete_id);
+}
+
+function createSectionCard(createElement, section, text, href) {
+  const sectionEl = createSection(createElement, section, text, href);
+  return createElement(
+    "div",
+    {
+      class: "search-result-doc-section search-item",
+    },
+    sectionEl
+  );
+}
+
+function createSection(createElement, title, text, href) {
+  const descEl = createElement("p", {
+    class: "search-result-text",
+    dangerouslySetInnerHTML: {
+      __html: text,
+    },
+  });
+
+  const titleEl = createElement("p", { class: "search-result-section" }, title);
+  const linkEl = createElement(
+    "a",
+    {
+      href: offsetURL(href),
+      class: "search-result-link",
+    },
+    [titleEl, descEl]
+  );
+  return linkEl;
+}
+
+function createErrorCard(createElement, title, text) {
+  const descEl = createElement("p", {
+    class: "search-error-text",
+    dangerouslySetInnerHTML: {
+      __html: text,
+    },
+  });
+
+  const titleEl = createElement("p", {
+    class: "search-error-title",
+    dangerouslySetInnerHTML: {
+      __html: `<i class="bi bi-exclamation-circle search-error-icon"></i> ${title}`,
+    },
+  });
+  const errorEl = createElement("div", { class: "search-error" }, [
+    titleEl,
+    descEl,
+  ]);
+  return errorEl;
+}
+
+function positionPanel(pos) {
+  const panelEl = window.document.querySelector(
+    "#quarto-search-results .aa-Panel"
+  );
+  const inputEl = window.document.querySelector(
+    "#quarto-search .aa-Autocomplete"
+  );
+
+  if (panelEl && inputEl) {
+    panelEl.style.top = `${Math.round(panelEl.offsetTop)}px`;
+    if (pos === "start") {
+      panelEl.style.left = `${Math.round(inputEl.left)}px`;
+    } else {
+      panelEl.style.right = `${Math.round(inputEl.offsetRight)}px`;
+    }
+  }
+}
+
+/* Highlighting */
+// highlighting functions
+function highlightMatch(query, text) {
+  if (text) {
+    const start = text.toLowerCase().indexOf(query.toLowerCase());
+    if (start !== -1) {
+      const startMark = "<mark class='search-match'>";
+      const endMark = "</mark>";
+
+      const end = start + query.length;
+      text =
+        text.slice(0, start) +
+        startMark +
+        text.slice(start, end) +
+        endMark +
+        text.slice(end);
+      const startInfo = clipStart(text, start);
+      const endInfo = clipEnd(
+        text,
+        startInfo.position + startMark.length + endMark.length
+      );
+      text =
+        startInfo.prefix +
+        text.slice(startInfo.position, endInfo.position) +
+        endInfo.suffix;
+
+      return text;
+    } else {
+      return text;
+    }
+  } else {
+    return text;
+  }
+}
+
+function clipStart(text, pos) {
+  const clipStart = pos - 50;
+  if (clipStart < 0) {
+    // This will just return the start of the string
+    return {
+      position: 0,
+      prefix: "",
+    };
+  } else {
+    // We're clipping before the start of the string, walk backwards to the first space.
+    const spacePos = findSpace(text, pos, -1);
+    return {
+      position: spacePos.position,
+      prefix: "",
+    };
+  }
+}
+
+function clipEnd(text, pos) {
+  const clipEnd = pos + 200;
+  if (clipEnd > text.length) {
+    return {
+      position: text.length,
+      suffix: "",
+    };
+  } else {
+    const spacePos = findSpace(text, clipEnd, 1);
+    return {
+      position: spacePos.position,
+      suffix: spacePos.clipped ? "…" : "",
+    };
+  }
+}
+
+function findSpace(text, start, step) {
+  let stepPos = start;
+  while (stepPos > -1 && stepPos < text.length) {
+    const char = text[stepPos];
+    if (char === " " || char === "," || char === ":") {
+      return {
+        position: step === 1 ? stepPos : stepPos - step,
+        clipped: stepPos > 1 && stepPos < text.length,
+      };
+    }
+    stepPos = stepPos + step;
+  }
+
+  return {
+    position: stepPos - step,
+    clipped: false,
+  };
+}
+
+// removes highlighting as implemented by the mark tag
+function clearHighlight(searchterm, el) {
+  const childNodes = el.childNodes;
+  for (let i = childNodes.length - 1; i >= 0; i--) {
+    const node = childNodes[i];
+    if (node.nodeType === Node.ELEMENT_NODE) {
+      if (
+        node.tagName === "MARK" &&
+        node.innerText.toLowerCase() === searchterm.toLowerCase()
+      ) {
+        el.replaceChild(document.createTextNode(node.innerText), node);
+      } else {
+        clearHighlight(searchterm, node);
+      }
+    }
+  }
+}
+
+function escapeRegExp(string) {
+  return string.replace(/[.*+?^${}()|[\]\\]/g, "\\$&"); // $& means the whole matched string
+}
+
+// highlight matches
+function highlight(term, el) {
+  const termRegex = new RegExp(term, "ig");
+  const childNodes = el.childNodes;
+
+  // walk back to front avoid mutating elements in front of us
+  for (let i = childNodes.length - 1; i >= 0; i--) {
+    const node = childNodes[i];
+
+    if (node.nodeType === Node.TEXT_NODE) {
+      // Search text nodes for text to highlight
+      const text = node.nodeValue;
+
+      let startIndex = 0;
+      let matchIndex = text.search(termRegex);
+      if (matchIndex > -1) {
+        const markFragment = document.createDocumentFragment();
+        while (matchIndex > -1) {
+          const prefix = text.slice(startIndex, matchIndex);
+          markFragment.appendChild(document.createTextNode(prefix));
+
+          const mark = document.createElement("mark");
+          mark.appendChild(
+            document.createTextNode(
+              text.slice(matchIndex, matchIndex + term.length)
+            )
+          );
+          markFragment.appendChild(mark);
+
+          startIndex = matchIndex + term.length;
+          matchIndex = text.slice(startIndex).search(new RegExp(term, "ig"));
+          if (matchIndex > -1) {
+            matchIndex = startIndex + matchIndex;
+          }
+        }
+        if (startIndex < text.length) {
+          markFragment.appendChild(
+            document.createTextNode(text.slice(startIndex, text.length))
+          );
+        }
+
+        el.replaceChild(markFragment, node);
+      }
+    } else if (node.nodeType === Node.ELEMENT_NODE) {
+      // recurse through elements
+      highlight(term, node);
+    }
+  }
+}
+
+/* Link Handling */
+// get the offset from this page for a given site root relative url
+function offsetURL(url) {
+  var offset = getMeta("quarto:offset");
+  return offset ? offset + url : url;
+}
+
+// read a meta tag value
+function getMeta(metaName) {
+  var metas = window.document.getElementsByTagName("meta");
+  for (let i = 0; i < metas.length; i++) {
+    if (metas[i].getAttribute("name") === metaName) {
+      return metas[i].getAttribute("content");
+    }
+  }
+  return "";
+}
+
+function algoliaSearch(query, limit, algoliaOptions) {
+  const { getAlgoliaResults } = window["@algolia/autocomplete-preset-algolia"];
+
+  const applicationId = algoliaOptions["application-id"];
+  const searchOnlyApiKey = algoliaOptions["search-only-api-key"];
+  const indexName = algoliaOptions["index-name"];
+  const indexFields = algoliaOptions["index-fields"];
+  const searchClient = window.algoliasearch(applicationId, searchOnlyApiKey);
+  const searchParams = algoliaOptions["params"];
+  const searchAnalytics = !!algoliaOptions["analytics-events"];
+
+  return getAlgoliaResults({
+    searchClient,
+    queries: [
+      {
+        indexName: indexName,
+        query,
+        params: {
+          hitsPerPage: limit,
+          clickAnalytics: searchAnalytics,
+          ...searchParams,
+        },
+      },
+    ],
+    transformResponse: (response) => {
+      if (!indexFields) {
+        return response.hits.map((hit) => {
+          return hit.map((item) => {
+            return {
+              ...item,
+              text: highlightMatch(query, item.text),
+            };
+          });
+        });
+      } else {
+        const remappedHits = response.hits.map((hit) => {
+          return hit.map((item) => {
+            const newItem = { ...item };
+            ["href", "section", "title", "text", "crumbs"].forEach(
+              (keyName) => {
+                const mappedName = indexFields[keyName];
+                if (
+                  mappedName &&
+                  item[mappedName] !== undefined &&
+                  mappedName !== keyName
+                ) {
+                  newItem[keyName] = item[mappedName];
+                  delete newItem[mappedName];
+                }
+              }
+            );
+            newItem.text = highlightMatch(query, newItem.text);
+            return newItem;
+          });
+        });
+        return remappedHits;
+      }
+    },
+  });
+}
+
+let subSearchTerm = undefined;
+let subSearchFuse = undefined;
+const kFuseMaxWait = 125;
+
+async function fuseSearch(query, fuse, fuseOptions) {
+  let index = fuse;
+  // Fuse.js using the Bitap algorithm for text matching which runs in
+  // O(nm) time (no matter the structure of the text). In our case this
+  // means that long search terms mixed with large index gets very slow
+  //
+  // This injects a subIndex that will be used once the terms get long enough
+  // Usually making this subindex is cheap since there will typically be
+  // a subset of results matching the existing query
+  if (subSearchFuse !== undefined && query.startsWith(subSearchTerm)) {
+    // Use the existing subSearchFuse
+    index = subSearchFuse;
+  } else if (subSearchFuse !== undefined) {
+    // The term changed, discard the existing fuse
+    subSearchFuse = undefined;
+    subSearchTerm = undefined;
+  }
+
+  // Search using the active fuse
+  const then = performance.now();
+  const resultsRaw = await index.search(query, fuseOptions);
+  const now = performance.now();
+
+  const results = resultsRaw.map((result) => {
+    const addParam = (url, name, value) => {
+      const anchorParts = url.split("#");
+      const baseUrl = anchorParts[0];
+      const sep = baseUrl.search("\\?") > 0 ? "&" : "?";
+      anchorParts[0] = baseUrl + sep + name + "=" + value;
+      return anchorParts.join("#");
+    };
+
+    return {
+      title: result.item.title,
+      section: result.item.section,
+      href: addParam(result.item.href, kQueryArg, query),
+      text: highlightMatch(query, result.item.text),
+      crumbs: result.item.crumbs,
+    };
+  });
+
+  // If we don't have a subfuse and the query is long enough, go ahead
+  // and create a subfuse to use for subsequent queries
+  if (
+    now - then > kFuseMaxWait &&
+    subSearchFuse === undefined &&
+    resultsRaw.length < fuseOptions.limit
+  ) {
+    subSearchTerm = query;
+    subSearchFuse = new window.Fuse([], kFuseIndexOptions);
+    resultsRaw.forEach((rr) => {
+      subSearchFuse.add(rr.item);
+    });
+  }
+  return results;
+}
diff --git a/docs/2_39/stan-users-guide-2_39.pdf b/docs/2_39/stan-users-guide-2_39.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..011fadac0dbab9555aed7a23608f94276ccaf8b0
GIT binary patch
literal 3281908
zcmcG#V~{REuqHfw$F^<Tw(XfSwr$(CZJ)7u#yDg9jBW4i-rbEGaW~?-f4=@x6&2Z4
z-Idu-XJtM~rYI^-$3)KtO*XeUvI@;X$Vg~!WChL33(X*FYG>|ZLCDO+PWT@eG=sRM
zjf<%hA%nP$p^K@gsj<C@DKtMnw6lwosi7^j$9kwDp8_Zo?AAR^cOvZZXk=~zTB>>j
z5eSS?g@s^h`>ze_*H2Qu1Vq@eq2~O{V+0s`1H;M-0gu!rn*n^vA`-)~dO_S9W=2RT
zKxD$oP04k9b?s@6<|kuX@G0?mv?Fm^`E+N-u@4Uv5;@XJh>s`^p36*;d7cYC9MY*1
zCdW{Eo=Cnnhs(BMfoh`C$?T@*E9Lc29mB?fs7Z1TeRVHe^?6MNzDZz!upSZF&9EeU
zzfAoTG~X-faTLK7*OHs{0EfDh+XBiIgsaBo^B;a-y*SxzXj40r|NV+Tz5lsAtp917
zR6QL`2^lmMjI2zJU7#6MT#a1**NBv@p}8qEgNCKa&nB|4vO+URnp&D$xco<7*wV#W
z(bP%A-qyk1&eYC@kQ16g#NNi<NyWj?_-8&*Q#VUvQ)LO^pBd#0ovjI(n19w({=Zsb
zVr1sz{9kJM|JKT-uceEsTBlpBD_YL``t!u`iiWLisu1?bD6k21rU;}Iq!ghbCa56*
zsu;FZmI$h{L^-u;LKc8b%_0@0YIRKwEQBhW0uTZuj#uJsynjx+%Tp636qc#DmK|%&
zI!U?LU%9RL?cBr9zT1U@0R5*cfjMdl;-0ScLOp9+b^(Wu2ghRFud^kuTSK93MTdQ9
zZo}u#*jF?x#}bk$oRmE=7I}5mNSZNw`O?CGGzmNk5Lh&_^b)y_7m3nx(vo;nd(IRU
zxyqenNg9|*B2#7#lce1K-AN8{Q%z7^9WhR<s+Np9ySDoh-f%q-5Wh&z^k~LMoOqKP
zftIQ`GPGnxn2gMGzt~*mtBWC(UUPi6js%M~=86ks9|n*E2p>_5oNX163=OTy{dhaA
zkY*1HZnl)sgNYxOR!{tuMZqgNjGr8k1M&#_#*BDpV;I!6rAeI_t@<@AiK`84b7R_t
z1!3l>`o}Vm0|hL*jyTGeqTIjKxm_D#vW@G=FKXuM2%C!5D?1`=VFw7HHR3H3v)=cZ
zO9(TaLxXodCc@|``qmnbzZdA$MLl>Jpb5Mg_WI%|TKaMsCi*k2J17(u<4I1f9hK~v
z4mv=mnSD)pr0K|+{%?AcxJ{BySX=4}Zkaai0s$3D^FEv{iS*TkL|lxv9H-^Pk)bAA
z{O3F^s2tFUNU|8RmU1z#Dj5Q8NP|VnNz1Y^<^?ujV6mHK=VjR3QV`9L#1vNJ8)~Xr
z=A*u#V4=#*T($zaec^CMhUNSR4<Jn~fz0C=c7**5zZbiEv*f459^*N}tj9X7s+hGL
zae#M7nnyZIa+Zerk-?SaW$8(D&I^A5e1MI+lzao|X^kK68~A~ptoV6Y;a-7v$rd+k
zbZ|~lPyC1bG5M61wVai>fjDVFFk>gQu|w$tRxt}r-I+~Vjmn^IAjy-dYUjDv1PxV)
zm#MAC03>QlBI%IrutQKoU5_qJz;2|emB+eG9dXpzSq^Q$=5yQIdFzt<_Xa$FYjA9=
zNi(d#yuzzOxB4Kw6T5*d%~Ae>74RU;`8arjg9SS6t;#l4#Mo7basb(-f@I0t)R7Tk
zqk<lBbd?ZK(pUnFK-WlG5=V<LDW8jg<+c#Cnq&)EveVW2q5I{@)93YC638ZDer&uQ
zJ)NLCI?$?V*+OkSfn1WnN@e4Nul?S@9tC>iiG(qMHpV=r@;y=(JvbPD89_reDOH@7
zKM9*Ti<TwYaY*Pm&vKHS=1KYf&qI2(W)x+mm0$*71;?eoowuntF)MEVx^8aA&gE{0
zV_rHzkgY$Y3~2SeJimiTEadC{f}qu(!Tzs)?<i!mX5)q^#d#q-i1)`s^1*|?hBx~+
z&x<`fqhK6iZ_=(diRj?29;|xu(!1mmk=LzExlJF9e~;Aqh7vC-#NF7Q=Lu|(9{;Dn
zFXx4i>lvh)?=N(kh0wSubecQ_px{NDWqrHibb1HF_h(UXj+(w-B5{PeHr0PXgPgt}
z$~hK9D7fApcpzbSp~1qPk}aqCQ@~8$P8BMNM@5El^8FyX<u~f_LLZ^ZxGbb-L&V0c
zz56eWiY&?g$j|53zixh96Tg0XPm$cj$AGHaAg4`^D$%WJieSejRHwhRe0smWyS9V;
z$`wR`3RG15`=FA@MJDPS@cHoVq^J4MU*q-$Ck5tZU>N|DhnCzu{G3fQ_;XW>_zG0b
z-*m;+wt9{t212dK>+yx6#!vbtk-u@KWBfvf#s`0$G-FE!{p4u4?x<c*J2j`jx5)=1
zQ5saP(ZOuwq~@Tb@{zPVOXFDa$X4i4s8;<jEP>x5w}WSF@^nB*ghv!`l?D(5e1Ni#
zaI?^}G{H&}UNGy3aH_O&XH#WF_gi;H&_2B26l$ERQHh~E9CSnSzeFfJS6*+{WxIvG
zKc|US87Rj11k3)BAR{i6e^*D}wS3ln_+v`nVXjT9Umzp(4s*1n`h9f3Hu<@~_GuuZ
zwa4Q&)4^ZE!?4{c&_6C4)RVrxyDMUKph2-}m0|SngHb$6RT`mv5huMug=;#UG#NFF
zuookq1}JherQ1}J<kUVSGK@cC%DZ5L*GiaU3_;z@o`&XLe(4(yZZ>|>w1fPa!;XcK
zHVo#a_!0}r<Acb4{0j?W+h-6|`W&QL?8syW@nT@WuGJOKieJdl8&vsb0suac<TIY;
z&nQvAN?6i*en)o|$d`#RxZc{|;>Iky4*E7lWAxuH25~{Y9(e+A)5D;j=`mAuHzh3+
z6rQ|6nb6)o<2Zo(3oe<AdNJ8+d__ZkVms*rpGDVc?E<d?grec1#e0S^`MPGm7qK$3
z8}tG%#K5b(HdrG79cG&FU7Ob;51(SB)SM*e<|yfxp=(a#Tx1vRI|%wiO~cFl3Ag8x
z+2@4Vhhd4_>M!nJys*-XzbEFPO4v0&<_UL;KFzts{RjK&SpkW56V-|d^CsCD(IW@=
z;6Pr@AMC#4f=(aVHoAf9CRw3O#Iw{u-<{Cai8V6~J$mSDLrZFm=I&tq*;c2w3k|4}
zaixVgzvDvnx?bAL;i&>PkwffDbXioFHi}%vAaz&23OS6y1oMaj)dnY{8C5PwDR5So
z{3$`keAu;;<PBu#U^kh}QJahd1dKm1#-gH2((o==GlLh$Aia!*f%zO_3?4Z=js^E;
zcO=VWHEo*7j3NqFqzRM32?0P|Xjsx^vF+QC#@B{$?O98p#mJCPX?VKn!fp*cm2`wS
z<U7;S33#!nTY_^~iF*vh{-9*b7CBmA)qUyXrH2n%n+|iS113z@Y+IVa>py5QZ72gw
z2?)Bwl2vc{ZMd<TD_cqI0f*=!AoQ|KD=xs;2iv+cUZHEn9hMF2rAdjXKi~l47PGk@
zX4MT1?UaJ&LsE7jsHIPk3WLz!3$JT8bP3*}iXM3)_j^ek43v~T-Y2XKXzX({kkc!I
zM=q_vrsF*@)b#L-Ojld6*LwydZDI?mTO(eGNO75H;K_^no{i*m+Jg<~VL)63_!&NB
zBE(0FCy*D$Ww2lqEo?%&Iwv=v3@shyp)f}<fhe%4l1h7&=Z_mZF*1-Z0Wzc=si$vC
z%yHPnSUe?c5HBnqWVJyBtpK|cjS2;$lti8WXfD_*G!)pP$R^>69EmZ@_&bFIj!^zx
z`T4_x*1A*7iS}t3ccCg&%IGqb6C9wGt$-HDt3%8qH%~xenqa{PW=m598D>qw_G+=N
zp3)jEh(NPTq3lRV0PPi70XA6ozB>A0PD<OZm<QX^iK?r+?%DgrjqF=3=rY!%TQ?on
zrb3(%$ZMiH{bWzUGBkEerL7=kOZDK44>w;V#oHDV@Unk9y9Xh-@^Xphb;oMUIe1kr
zS80Ty;^2g~g8d2$WfW^pZe{xB!*$T<SUs@)I;1vwCbb2@S{+)mDr5Yx-5HCd&Al@>
zoKCQ+x<$TWQ1D3d0^E?jn1@>~7GhUPvH-INois~-SmoQz246PennlQtfPfJA(4`~f
z=f$F@F7$GlC3Y5(pA=3#P=+0C6C4PQ69nvY<@?s*9ya-Wr%0BeL%aZOXaWC=FiA%4
zdX>tGci8VL_jKoWhBLUn4Z|+=HZRZyB?uVm?%4Wa^8VuZm%SO!;MDQ27!Dub9tDx0
zQg2gi*q7=XK%fiubGZCpKJ_;{yi7HpbV-G7UEO~?0tKwCfa5V%%vIfJ1Hs`ep9Q8T
z=X_<c{ksxR0nlctrmKi}>JVnXzL!JG7vTBD2OEmjtnD&w?&5^E>O&_njq7_Q$pl40
z6^e#-+$XY4!<>yZzWa{&#qnnqHZ29TZGEj)&pw~f;k0-|N8yXD=8+GZ?rdiX(8h`L
z*kDsM=0{MK0zBkV)I3J-=b-&=zU@a)8M!jW4s}Qxgd*nM1^Vq%mXce@>ZxPBDvI7Z
zb@o^~yW2bhk8<%c2%I`10Ux;ghfWLg>vtD#&`g!V<sQSCJcm$@meA%XQ?IIZbuiF~
z`;p;OaC8Q)z@yPJ$EJ_Q54FQPG0nO^j$e3}?~xwx8}HdH80cQ6AdWq{rl1qtm`yy6
zU*Y8zBkvc`667=KVcH71x<|Q-24;z}JmBRsgj5df0}?q5xboA-{A`-$zsK*>_0)kW
z=J@(LHg5h5K7hRp%2WNpfm&9dblC_%n?-q0TnkQx%BP{>L`Nn?AHT&!fj0S!cx{=i
z!CA~+n_hKNKZ{Dgebs#nY^yfOy1L|{?CQyqt<bk3$+jZpNu=PIp*pCsCY<Gv)w-Nh
zJB>f~@k?9AHOrsMP!k<3tYYAzuidS&ZYWTV%LS3EHi3DU&9c8fn0$i6esvHh;;0PF
z|BIeQ(nU;kZ<_LVyO3elJ?`H$m5_vjTQk5<8@>~Ob9T5&wW>*S<+kTOCGm~&RNQC7
zka%f<X>p89uN<tF>crc5P(EV*u0fCFlv9O}aHxFp+%M1`5q)Lo(|ZvG>MSjaesgV6
zVsR48XHR*@im$0{np7xX^C}{w@uHApwsO^@CpCz86FlSS*SOZaMqIrqPpd8A*GbR0
zA>bO8vc3u}kmNg;n^{a9FC0X6)FcB94TU5>WzRd@TZt5Rg@0mMb)HBN!8Uf@9@I1*
zTmSNYrC;1ahelQl!M<!DOmLVqU{$0@z2yuN9cAl81Ak$2NQBAx?76V4dwHTo6%QZX
z=`Xk&%%JeH!Fn{NiekqwMkM0A1q;PKCM!S1*|T8o+me4wUAd``)aUlT5XkZl3u`j<
zjnDruF?@{D7AtK%0uJEX+jruCfKwln<!bP$UkMJZpKa+*FUCi2dnru+<No7p_;^?>
zMmMWSln&tp;9nC{Iy=*i!)CUW^t4&+Zko%Tvo!yTlD*=gT*bQY<6y^JN5bj<5p<Bw
z<FP6lkS?xqlkh&sOp1?Np7Ju^SOAXP691~ZOEhExO?Hc)VLXryzKWfKWnt|&gIHc%
zHF1^p?$BX#xN@0H!o%{1%kup0C!C$`tLDH=NMZ7F$DwCGq1I6C$ic2zqbR;DYQVkg
z`7@iM+vQ-;L*3;+9K;VA%*FBT{c?ii$UvtYzZTmC*QpHb?7M4GnX}C_#qN0sJYn?=
zo8P-1IAFNV{v%nwrxHdxjn}*pgagk#)KOO*d8k7A<3P&jca}c$coSM$i}%e#@&AbS
z3`A50r&~CT3x&r_xi+e<ZsS*_O=_6=woKv|R5f12TG9|;;!e~<v?u%H^W3MV5bL@&
zxW}z~u6lAbG$@K6h)KwES$_%Wm7b1-GRRA!$Gy5g0z@=04qAix9E+Ai7Y+1cb%O#=
zLx(e`4>{<wGEsML_ilq8f!v$)74I70j(&~}8uXN|L~N3kDKd$3hFv%R$nJCKD&xDc
zsh|;VARqvbFxod~S;(6{H}nDxe?pfHn|RSt!w(be>mM$F>D$Pz$Gk2cF%vxl=racs
zj=VPgZ`^<`x(PC}7^1rgUHzZ&1IAI8O)Uw$MK<ww<@x!!Z?9!qidp6iQ+P;pD-p&_
z_Lh#f<S^JXVrN6iq((tEOT~QX{=E*{+F#m)agC+<!KTXBGWn~}MUOv=nLf*7<ndTz
zvzhlGECzhs4Wv`>;`BMbc)Kz}XDT15MjE{GYoS7D$88-!sG5oyF@U!Z^9Kum*}pUu
zQj$)@J9^5$2DUJ-%3)WM2>j~WLb1W&^3LTj3lbd=h|O5{|MfPeN5_bm>EITr;%DHR
z`fvceRoe!$Bd~c>vLoy16~Ob_zcZ8{CGB~c+rx-5?))oK1v$vnyHp^;_m%h@>>+go
zp${3D!hz!3`l3c?Tx&I)+;E24t;#e(mH5_K)7&JtAHBA&ewG4E3<Vz+>pMLHjiFC3
zz)<|u(Td+uo|o{pdv{K%Nvryjg%a}VDj~iH>)B_&b#DX5kpC|hlY`wEqc(%7K!T#6
z8riEaAhdr}&x-y5MvK#=1KvA<RPdX&$im+J1R{ux&#}Am_rJn)ivHhQ2SXyod9gb$
zqyKyx$SoBJ!lkrDQA@~YWA%ZYpK`vroZbRTViLP<R-+`<Lg7h6zcQ8Op}!ib5y+=(
zxo%H0S|AZb{{W>1EAI_5Q+*!@;57qA&ttKUDL#_JXcr9Qq-bO2xe}0Tt5TeZPR4I#
z3|oKY8haTo!%L!s&Sr|iYXhuF>BxP&{%UJhuN-HDg)}?M>YN+e1(_z1P|Z_I3Ft6-
zL;~E-s@08e^oiBGTN{zZDFo>$oe>eb^{9@Bij^!O@bR(1NghwVekX)#rctuaFp$N4
z1PsgYf&MEU;(gW3?c7~*EoBr8bg*j-=H8R8M8Jw&v-KoqgbEvn$-hzXa;0O(hcsBM
z7)vCmNcbY^ZfSmP+*2_~<*ifu6Ts3ZOY4w;L7V)iIc%tn<Lh|rlu-)*9H0<J=V>g%
zCYwYcuxnv`o8;7iJMv>83SfjnyPiI$l@&{vW~c=On!5DG=5Z3{89$a0$amKZefo;E
zAzo^HA9>9=lo)5BlzA~@GB(j2F7~XO_&|;oE`}gIGq{dI@ERx_l^)aSH2n1WDU3dd
z8G7@EJPhF9b&+r#6f;$A4QKIa{PzQGjN-k~03n4xHF%j(k!Jh{z+CTvaZy4WKQNjZ
zgZ4D`K8V6D7clK*Ju@c6+wePQKPAk*_FHCHnwIfU-yJpKAq=@kjNv2#cN&Xs14dwO
zoMWffmzrAaZZ-!h#}^`;FTdS(%1&33p6ky%3P6U&LHY(TSWOU@2|kZheh&`6A9&>L
zD7}>sDW4ywbq}=7v#Vmp*!$h>!iwIpCZM?Ak5E(bdq2esS60&2&u_Ff{2K|b`moUm
z=(7;vqkn=rC&?IXUk0X`fFSf1EKeJ%Uc5I+Il}t{>X>U=Ai$$v3w}fKOH~ybrGAjz
zSD@^{=0m(j%r(0iZwX36b=5(do*ahm^SI>Ga#%H^JpOg(%Q=z6qtc$U9I^b6TwhP;
zPC%&|A2aBtKO841e8d2}2q$K|ncL%!DBNd&ISt~14g|ye1B`SS?6|pa2+^c#y9d3a
zZW^cnVn$R>G|Aq8cgw7t_w)RUv0)WIrj&=H8+BdI{eyz@!=a;hmp(dtl+2g0-Wp5e
z`Mm7|dwym&{uv0xxX%?{uRCs9@M1sArd(JZl_$B`mWT+&pjppwd69B~pj!@VMUK}q
z?`S~tUDZeb-Q5Pbf23w$R1wF^&*Sbwdzoj1QRehd5rb*)S5lVYVRmaSFt{k~qzfDL
z>2?N<we3NB?UDBU0Ffcq%tRY>NbWQ!h6z7X!lF4ME(nJYyC2GxnnI5HZVG{B+`RGE
zKMVkwIUiPQ$@Q+{YEw9F!fzaos_;VN;}1cy51?x*k@FvXZVdyPZ-S7SX9}C^saTzF
zGUEA{W-u%$2)}nN!>SO*FjU3L@iIq)65gi~uH$0w+i&1FiV!m)o~szZEXIOD#EX_1
z)LLB40$G8zz0o+~M9mPvhD}6<e7AT%Npqhcr41gw>rzOFw=$}Rl!%w;g_{oP0{Qbt
z`et(*BB?iA@w#3U8jiq|YiD!|tBM@7oT49yjEt<eukzt{+{phk)9p_fmfdMfVmD0I
z=-iaYeGQS8b}%eD&XKajFtxe8hiLHS&R6)dJY%Md5XKU%7Xgu`UZhHC7q{Bzujr)S
zS6~u>c6lM-j$sGh6<9I{F2Fbvuw3Y}OxE*T^3#7XZOmuwE?<ITpP4ac%#n!UMrD0;
ze~z?`P;4gm4}WkFW_pA*V#1LyZ$F_nf?DBko_Hz*gwz%5>jaF>McQ^YSIFeV5$O=~
z0}lSTFxoV7IBE6q4~+aI1-(bmXc;By6piCs`-PNvxP0PMWy7q^=#UZmKcF%bjs3cD
zKDNe3&QLn|-YoK7eg?Zf9SVf<)9z~s-IeUGl?E6Qi2Gy*5Mg!nX|MKY>4xg<apQ^@
z!LlJYGuy{yHhvFqjuMbRpXzwGK4OxsqZyr{2I-~HB+vyiQ3Q*r2f0`3W^xcn^dKEA
zt^LpuM3>WQ*j?YK9pXIs(~k=y;)Z!tM@dh7^%@8Xw+}ogw9#mc;vLfe*>BCVy!#BT
z($;YdVH7(`$krKNAY!Pzkzm?M@rU7(M)LHIM1;bwBH?|r(I5R*WgYd3xDk;E_=HyC
z$_gtb=TX!ksbJepmd#>_&FR~69~M0nLbDg+C!XlL?x-{L15n=&>LAzh{r1Zfqa-hB
ziCS0a5Zn+5FoonGTgTPb&%9_YZ$H2Bn|-jnr);l-NzieRc#>pMt4az#!TD`|p=e_;
zK0H?+%fzg444)3XquUkY*_;}Id+;D&<P$B&{y45xC4f^p18;u&<(}a$drtnBXGR&F
zHG7VeD|B3*%q#^od}BW<x1<2vXAA(t^2c*~$FKFWmz#uJxOW&Vikl&|C}^S~OHcLL
z&j5YwAGvKLN;`Ib7EivmrQknd@`QMaKtowWlF;0v(9E{b^o(l(_h3CMwybcH6T2pU
zdyy-}{JgrB?jW${Xa^qSqI7BwA)!Y%patp4p<A*|3HL7*r&sZ45ELKxqFc5U4i~8O
zbdy-%Ij@1kLFzdaCphqWdYcRufz|SaZu1k@^x5`2z+u;_LkJBqV`R`*h_yqNLHwxr
zH8htPDHpmzguFWWj^z*rFH2%)@t0Vcq>^%ji{$X1f=*ZfI6?g%C$Z=oo2MFsf|1>D
z?{yaY$`J&%rIxcNPF{#8eE<jt;m{C(br>#^c23~?xh+@qxgAx@irwi26ejSi7{%Gq
zj<(o|>j|``0nr@hu3YkC<Kg)Y9^MosP=iFp?dok$TBl7uz>p6B&5c>q3agHdrfV#Z
z6o+w`x?yc@FV*E(*g}C|aGZmztLj}IK-nEI{>rJ7KtUE0^^=yN*DvYu1_~qaho_hu
zw}{pF01C$Bs|qcs&r}wZ&nJTC3&_p3+Jn#x`=>Z-9_{6~2t)z<7AHk^ctUkn0wLHn
zSDhQw#h!{q>&Iy_he6^r>=N_kH3Ux4&v*<J6sR*1h?_l_o$)I`1kf{v$ydoO>fie>
z3!>j;PcIdiPssx-V+*KfxSdzo&baHl&I^dA8h#(vUQ7jWW-Gu*d~cC}Edl2=-a$6}
zf%JLQnT;Pnlq^JwN=RjZl_3VALI-^fiW4PRm!T^?=J-7Q&3+1nYerFyp&Y7!SZq0B
zhCC<2*UwGzdB_TZ-#=rjS>qdR*Qc=$Fhkw%7izP!gqN}}!<cdY2aK7=FLGnp<ZMLm
zdd;Nr%!qIhUJ#w5B7DwY@ZC943en)K0~+_NUfU8t?GyR*WPull=Q0&Q+f<l^%RS!n
zy5V`r3^g^%<qnH(n!M}MJlK{I-AK~lP8JY|iafYlna&%K8^*plw6RZ9b;Bio$qOq0
z>$cD~L7r_jHlV<J2Fg87P(Lp@T$<~c9tzci>SN%BR0X!1i$|j@dh$5=i{pEv<{fB2
zkOBqz<;Qjg7ummQKfs*KL(+WNPnw&1S^RnGe-)`P!R!n)%3_u^QN9dy^Er$$AGOoE
z5w|~*?6)GXqU3!BEfF1Pw3ChhTYzd14)Td6Am=lEI@Qt?r$+GV0a}I07APU&dZD|5
z9CuM2;9&G5MPU|jVIoVXSL~hI$%C<}9I|)BFoqK)0uud4NSfgBuHYNM=*SuWZ}ukr
z{P!N@fQh4qc$E&?U_Wb$b&17S6}r<M>+C=K<M9O-`m&AAT3E~qVMfT6Oro1Kz`^}$
zeV^de&M$;bRT(6QC)H{wC73Y;K^elYAQql4=L=M{h*y@6JKn8%$v=fmTh$Bdkg$ya
z8KA?^BAQRMd0rT<`_O=?9emKAtW5+k?v!2gfMB(-M}~fsz65Ubcl4xWW$NcH^u&jM
zkI>F{NVgso!Y*n#Bm-u{kb)`9bYP8S0qlgCslcenPpr`}Q((wn&(s@O(`xT8E3mO0
zbfJ-1B3fT!0cI^pW$fW6N(#nOzDORAf*Re06Y__09XRN#dT792`ER+oee?ak7~Zao
zpUB|&!r=;zFzfeqpP57mdYWN;0*}dN+4VDQdsL;^OQ}cPN3JJCf&^KAoOneLUy7Sz
zC}u)JJar%fg#5jf84Q<85+uuA+x?)Yw!wf0Aw*6IN$;lM3DkZI`V@)x`$At)l4`d?
zi2D8Y+pa-oD!CB!nBFKxnz330#QoE*#=#f-6fLRNRQWebz0Lb!0!MGrJN(}f(-R;k
zhG*pn=pk4+N=&E0$c~zdvOUNSr%1@Pr^}&$rSMfj@OXIdU?QtYixw#&cG+?3A^T%H
z_Y7J(b9W!#H-)$A2HPEapl##Q@s~GmJuw@#2oxv}+_jMkdkGRYyyBf-NS&9>_2D_C
zG3C<qV%M`2M?yUo6Tw?!P=9TR6fvJB9N;|5v(yR7KCYe}e!V<@L1ne(%d+pb-Q$5a
zM<y%<%hS=*gb|#OU+Alf-_afpTk6w3Kjow|C?621=OBhCG5=!zu~^}cjIfIW<4$vR
zr0T+!+yVg~I*^U;SM1LdmMUA*Z~KGegG426WNTLGegsjg>87RlHI=V4V;{Rm7uSF8
zKhd^ner)LRe4vwjvbjAYj*(F(Z6>_J(w|g~?{(<CDr)QO3*`cn`V~%ajlMw)nr(c+
zANHb>k<kGwH;3NEQU2rGYD}B_ak&gwvgS;3^rBscMJ|xs2yZYxB0-Tswxo4EcKYb8
z!+Lw;70*x7FsB`cmDAhVwYS54Cv2d-wg#*$5s7MTG-Xj?zicpmKB!mH+DJ|QvK>7n
ziB<n6+%4~fx-E%>yJ&025aG@Ys+oP~_OMlzuW0?Kp{G9HS+}>=#Y9X#8|e_!K%c4v
z)*tVHGZ8kQIq`m<S=qp_%@eY3a+Z{wz+XO^O0IIyi(q~wGaM@mRaqir0J$ivJv~}`
zN3E4rNmO$~Vu}Inw1AB9xIh>`!>Az+Edyd(Q{;@_;P?pK?wzBr=#X=oeg4&(ahXC$
ze#Ta13`zyY!@b)?3%q6a0gCsc&n3`zkE1qnOCTVVpQ?JM<QMu0;PO)C?iKbuRQIwC
z;^T<@$yQ(XVXREZ0LzTCFoEC)ua!D_IrkcH-OA#j-b1@54x*iZ-$HJ+<j>HPdrZmL
z!7enoT88Q()w4i&*Y>mZB=|Pn&q)Q=U44I@&%}R)S+xoR>m<<qsT>1K6(h1IMR&-&
z5Brw@C$9V+h}h=w>6H6}<5dU?ws034d}9v-@}CVtSMmkTW*OI&+aBsj<6MIAKGJ@@
z<g`OS-U`m8IcxGEGjhpjE<_i44JB_sLGCa}I;inz&GMD8K>eWQ6I0laR&2SYnX^xe
zn1fch1tIi=d%!3nd6%&Mabbxzg8Mt^Qg(G=F${m?&u6c>^6g=7+RnX#ld#H$yodL|
z3DnvqZU0?h&OcOEmK-}afA!CP-}G$F&bKwvfEF2m>7>$|ECt2}slG-Z;C3##svFz+
zG9ZW-6KHLBPg}y-jYe>4uQif&!@20e5H-(6fwuU;hz}|Ok)IT;F<RJCxaFB&*6(m5
z7K$;tOdPVhws8*?>^!C~x1qC1DZYQvI}RZw9L*aqpoT967~z_9iCibh+|k(_+1=(Z
zJ9@7oEj;BR1}q3CEIo><_&EXAFQ-XIZ8#y-dOmG%{@h@^gao+NMNHf!pjW$X+3fPd
zzCO1?dAWk*k$;PLcKN@aCx|Erwglr4pqHVBPl>P}Q2YEP{JPe9OwyS?ev`YAKvcaI
zFpiTJI(mtMZ0@GlBe2u~P4FrDumOYS0HsIFwt<;xIkX7QyFfHaqMAi?6ERI#!wih!
zQv%-16D3PPe~HA*AlHqsfriqdPmm{KkA?N%qFIcI;?d(!ZmtD3;Bo@%8+0$D*A5Xg
zC_2EdDE7g!&@P(asH~9_Q3HjxDD>~A8wT6Ht)o%eLcO;86%cYtDTtfkhpOF$a(P$i
zqdS2mD2Sg&3Ri|O-9=}TwHmOMfD;lp4bU=UWf=|_1a+c)iq8FkRZ3Q*MM1>emgbNs
zc7l9spyqVKcncO5t(oVfLpTQ)tWO>X2hhYea1xR5&*CNpPY|7u4K$ERhX)04A|_Av
zA_mI9I8EMSBNthQ`bNs?DF%>(jF~#mDS&4tP$JcmU@C#}nIvS1TF+r1`VtV)BOHOu
z$;nTV6{}_tra_vaY6casf#@wlj-0F;#h$=O%2=1!M>Fku1<NxD%LK7y;HK2{6SC(6
z3@WKsWy}>v(!huie&{4F3;;c`hK;g-xR@D1HbL=-g@s$L3e14EERZ!;WeFyBN{QIO
z3HF`{;c!^_zbfCd6=0^Zw^}a5W@x552#X03xE??Spwgy4N=O6kN?2KR_#O8~MT7))
zCG612RS2X|GA@kylQfg4mTSKkj28oKvDF5?4epU_BViN=Ph{Lh4RTs<SQ2%T0hJ8E
z<`e2S1vAqzV#+3r#4IQWH}Zu5f5bv`+37F%#>BkYh4K5DL;+GG`Gr?AnX^3Pw~3G2
zGyZOGVBj-?!J4_?@lFHo*as7kJN^k2Qpd7p3`GPDVC<@q$YbLM?I)rb0SAa_eh+zn
z8$tu>G1N_Sb;8$i2rGn2HfE8k34AkuCWmV8^GCG4z{~&Q$Gh$K?;iV@*S{6k43CC9
zaR3);zZAU2thLHb3x}f^%trW&6g~^{gtxc{mLVD73JLtYlZgs@asy~90w37{aa;oa
zgC|~VE6pD{SClZuBI0b35(s9Ia|XV-(l<1UHnHKio96`l9!|7}=L#AE7O<d5{l~OL
z?ysSK=d214TI_UjpjEX@R{y&l&ij^MfE4rv+t3Dh2tOzi0#IEzgk|e0FLEXf%p>;@
zE32}-qrJ7o9|?gId4c|A08dazTpwWTD}IXu$PdVYAaJeK5`4-X8Zh~}0i1j<DcC9z
zAH|&J*9CZy@HHs-v=6eAfn0q6E*0)K1}o1V`n^2v_4G6t!bb}<daN^gtSNr=)_=bf
z`UnaN+W8Os>1{99bKr9f2I%U4y5J!*+-&h+MnU5v4+7jP$Vnm}%<Z^Pzd7LE;W6ER
zq7ioQpKAhc)X{*~gC~$2<BUSfp!HGFRv}3&FSCX$*>Q?J?8yjOh92CK?#J5rp7>N~
zu+2SU?)|_oKbyh=fgs@iuAg{?gsJAioumyuR=M!9fj@cwU`Ahl6f8^5&C~0%@4(J4
z6pw=F#vP8izGiRERRj#Q8>}BUnwS4imSTTzE(OPi`B$PbT-qEha>CI$lOQ?6$psQD
zq!F$#f%kxefnqtz2?7Ps4WN%z$Doz0mqh+%HqC5e*5~=)&Z1$0IG_#0_!=Dru(=~p
zf#3xeP~a2vmtA`aJz2r{o)$?vT2f5fSj{9T?NcIRrX@y6Nq{ab065?sQYCxzq7|%?
zDp~Xowq(0vas*-vnYhBi!_>~45|+BnHz<(3cbF1&YXf?!OgU;o<*SV_QX!c!i?5Og
zZn`PdLW48kyw_w1F1H(p^`pJf0Ra#sC<$}?TcvHWn@%RDk1Tn%iY$8rJ#pNtHGd?>
zoCO|AMY0qCyObo^g;ZB&hz<RpeTD$D+w_Odls_SV{AhC^#QVxP0K1N)m}UHA%XZ42
z?p}Z9Lt!@u_&>4q{-aA8_Koms0=o6TvK{{y_3;0&LjV7yJ4!eidj4OABMUPl<A0ZT
zdb#E4kn)K4=l0r+2#QEJ0J2crgfsR?(Xb#0Q?wkl6x^@@GYa>yNDZ}t6W&7+6R&V!
zkPvYaNs3dnKL|G%xr4|+)QDJ^P>>igP7p+@bJFbj6%A}_>*5i&&-|H-|GMwR`Iyb{
z`SmkzM<@hQcr0dX%Y6!m#e8;pLc94@9H+SneKwoR)1|L<z93O)1Br#>Fu(S9mu5eh
z_mRSD_P;o}z_bWtaoP!*W+++Vpjuk{@i87ljDqkoe>E%fCg+^3+pp__6Sa#5wep-8
zk(nmCgY3b3vq%+p4aPdPE5GK<ai#a#&N0=~vt`z3DY<O98D~Z`XY^@#c7ihqGFsT&
zSw<Aw<kJxl7>)GiYbJ23sfUF<ksh^E)}i>BPTM+Dq_I?vx&8_`b3Fs3>bnE-QPs=T
zb0ehb)Qt^~Doct?0QYxPp<=BH15>OjAHyi4v#QogW|)%+&VDqzUq`Yf?TYx;?DX)I
zGb7x;^@IZRr8j4`%D4NAfhf+ND(%1ZEO5t?E5|V_EUB^wYJDXFODV!-!Sd%9({h{K
zfv3frqIiw+?u=W98D$tt@*}buxhsFt2W$dQX$@`prmM@+Sn$-@BsZ$&tuy2&Eei~n
z!|R#1o{V*TnyZ=*cfoC|$n#CO?OV_(5c7X#(YTBF=hyJWOmg^9uBK9hH<2l{zO5_6
z0B)UWNPSs|wG3-kaxrTu(xL>+qB;4SQQV^-3#;x^3dtOA-vn?m^@(JbOUo$YW-BPF
zmCjhK{%&-rTI|LM%0Z!AZ+*=<4ZGF!hGZz$Jq){DCdq5h);9>d6UQ?+<2B1>76c#W
z9k}WHZCrMu=<&?r-pG{vaG+;sa{14otg^y%Rr9)q1;zEZScay~5~GSMCvSaEQXNxr
zivT}tjT}wJ=|rSjF54Ffi=ocac8&?1VVO<lL^9)hrj)ma+OlkO6SHATRE<63{XY*~
zw&j7KNWHbC8S2LEYm>686mEM<6swH!jq(Xc@;@1;+4Ugj&<)J&X*}o6f<1|~JlB;%
z6pHrhgYMo&Q)mg>R@rDn8q)|C6an`ux|N%nIx9g7dQ@#I5M=!iw9zQ%%8Fj;qoFPN
zKvLJ#I?e{MwIC*2Zq6R7Z^$vda*9)Ln24XM3*x8VgV2O;G!lE>!$*V<JCdgLkT<l5
z1x{k`7PBCXifhoC=79_jckUi!2EPa;DNO3=CV+o6kdJRE#SqjM=q0~yzoVW6r4vQF
zvhE5Wo{^j1)BMvBZj~6Vu;yPC<`8AJ&H=S5<RSMQWW=#B!X`>=L0UG9IBOqHHWmQQ
zZ0ApG&4l!?BSs;$z25|4Z1g1xYzc+=RRjTntUl-qPk-x6p2dcJAB<du<{LPRYanHr
zr@3OKDK$=hhxKHtD0Ob@SZA7$w2Y=y(+82V2#9u^4o_a}#b5~;(V;sH)<TMAW$q{K
z87b7O7nbr;Bhnv(1xm^Z3w4?fPRxN8?xE<HP3zXLBb9}#_&c~;he1eMA3YF{50<dW
z4RF>2Bj_*%IQKvjbm+kkuEX=*oujTkqMVG`dm5}bos88J6Iw&;QdU|Mu6Q;m{rxLg
zH}cac0BioF)&$pZg?QueRZ1}$^kh=qUfg>CSf*<4rnWGcrCaB-Y<Tq2?`;ArLk>#o
zNd9$f4ocd#Df!m={YBmMy>-b2MctMjUTUR$86>J?J80SSnpdO7QT^ui0|o^*W9Grn
zAQ?9nN_s1xjC)FcdMh$NjT)Xh%)j5Ss>e5%afo{h#hg<5sb$_>b-|GNtzKRDp^*9V
z-d*a#Aqs5Ub)4=Y8oxA1ID#PRm3j3D0wL=2o?LPv5q?b%qA^ze9$0Y2;PrUSEkBp?
zxH-!yr?FewO<OS{#^Loiy4zfj)e`peoI)KH4A#HgtaBF%)<@{jdlwW*1zwS~xH)%i
zJ4dy+q1n3R#=!p2-gnc3h4ZT&we_z0J$~5<!2Vdja{C1NX|CP2pyB*{58Sq(;Qa91
zx~_r&eb9~`dDM8WU4Ae>L(X0QgoE|LJ!~ta!#C?T3JNFg&&e%rtSPS6lu8-4>SziF
z>SOIzWPgTYHy%lT*4(&!LuY+q!_jc8{wvycUOczEqKJE2BF(o|7yBdJir;1ISOMCH
z?Oj7Ld~~~sOv{<27;BSv9woCoP%V>J_r5n!Ev&GDcC(wh!gr#wH$YSG>!>gotYO*y
zM>4fL&hNL8LBVJPkVdZadFOKnJO$xkIkWvr?9YDF+iUPzI8twIv`~n?u=a1Tp%eR9
z1x~Urj{SP(ohD)2o4@U>TeTW4ojvdT0w?FqI&|LSlJz+BzMoMmaKIeV_R(!b&l3Uf
zeHQM9Ks%bgb@fH0o{)KTZP~9-9_cXp-0XA3rI)68bgy3%Zl|<u5m>qUu^ocX4(*Eo
zr8Ji1Yq9bjL2<3nuj`{+SS(Yy`|GG|60~Ld{?NPj1xY0|U`-~7>KeE}UGIKzsbJ6!
zWi*Gw;|QpePJHgWcWy$OQelbMdf=wrE9W)6c=!wH>^+h;hTw*ML5o*7*nd%N;ZDs<
zFlSN8EGDf?lr-$vW3}E68q|QqoS9IAQVm1Wa;idIoJ9~RubI(8-CIbS@|=X42OUZ&
z2DaZcS8>U|I>bWGF*&7dt02{z45|YcsnWxYbS>FSN*)Q6Y8Jt0c1e${h-T30;YwPx
zNNLE7SGrS_hdcMb;HZ2lvQm``$a?+u^-h7>ho6GM)X!iU8Xpt)qq60O-87MDT!S*&
zy(FDW@d!~Cgd%Gep=TS6&Zw8r+#Mn;(@oT+`?b0X%>mjn?jyg|Pnb;_lxhDNil}ZW
z#?izE>E4VccKd<pHOT9jbe8RE(P;Ka*;NN%wAY(7E91s7z-4^Ozt**%P&)XNkQp*K
z1whk9=_WIt-DNkkQJu44zF7mPE+Git)Q!|Gyg1N@-)7oe$d2SfjJL{=;_rY?^?F02
ztywOv9wwHPULL?``EcaBwLxUvtvec%T2OQPT$tkWX<7yw&eH?c&G)siMeQGPZ{f;?
zzjV>f)thj^sEx%QJ-d_-pc~X%2)|sJKi%L987k)~o4e!*UEk(5YY39HGrd5S<HXoB
zsw2x=<3<y_M)Fx`&P&Zo%%<f!F=w{(7HgBK2#l@0lh?)UnFy+@1uFG7=@()l!jw6!
zY?{Ov2Q{G;B16lZd{YP2CW%oRdxiF@N1n?JeUk^ZP?9fR&_`7~a1NQl&n2PVE6F`l
zPu^HU`ygz@&0YD05-WlSj-fAlsF=x8u`(YgdQdBoVSQkHm59aayuR6}&u0Lr^V>}S
z@`h_DqZ~+JU*2>)a*6d<q>CHZ-;&X->>Bjz5Bi8~P1qKT+Fj>MGQ)7p6wkdff2(G*
zj2iTO5L#AyO~&-Uxq?{MA$8`5JI0XFE$bf3#7D0uh-ViZFa<IGk<O~^v12&rTuJ_&
z^XRMhHQxERSn8nVRo=gb;)(?CA!bIN$w@YnSTChPN}<O4ig@c`Osl6Nk3i6i+A4>>
z7xqHsEOEvvqwY1>Fl6?M6fdn?Z=+_XEJ6OfZtuuZ;yoeGTeFxaIm*7zs_{1gyQM)B
zzUjM#LyiOc<x}{%{=~!`^=cu#SBl)y>fKdQxt%}bq>tgibaN=ToulmHQJc{0<+_$>
zL)%I{!TI&mmmUDlt4V_dn}N4%_IfQ1<}xcof9<02q{??%z=DdV&M4hm>8C;{_bX_m
z&z0<NmD=~+dd#c5&?1~hbScAoN|q|Z!ez72V*Zo(&AH$hj<UH;%NI%KsWek7ov8Qr
z%Du71-wmYob{|_I>wE;~yVC4B7M8fYYx&;>e5YOEZDGE(fzB%gAJbs?9`9x6Rk=&v
z_PqUd)1*BsH*3CzK9#%NtpDJ@^Nsg9b#_26^;qmaZhM6R6$5kiLw6c9sB-Y#ydEXB
z2eWmm-9dmxh;hu3*wwnbAkg<-N4emyxT~2K`I(N}_!zCoW0o;atu9{&(OO#r3}ylZ
z&oIr6s@%3T*avEEB!$r`sdcj}3Ncli_v&`gVeYH?UoZQ+E9#x<6iL7`*bnna@_i<Q
zp^wHoGA{%yR|~}NdsF3O4I@u^jmJerhQSVwvo(IVAwJQb1}<h_YNsV^p=-15(V_Z~
zTkP?DtuB9<ka}nCK3cBf&zB+1lJK39yPmwg1}{?eFQ@4^ow^^UO7B=oDdJEMueMsx
zhNOlF&a*aw#36}75dS)SLqrNo_5Ifq6YKvh*7`purvF_>$Hl?&->##p)tHRiU`O4W
z)v%xfWtuD}rQaa7RbiE6xve`DPUB)n(qbiDvweFc8=>*4m^*-<O9gws$s2k{pLxQX
z@$>7y`7pe{*-}U{DG^P9%(Km7RyXNRk{n~+xM&i%n~p+ETG>#qT~^)Q<b6GkFT+ve
zfUg$d=+nmGO_oIe?Lh9(Y6eam;|W7RU)5*vm@eUzmMWQ+!C7KZK}q9n0D{~gmba!2
zYBJ49J+Rxglw}Ze-{!yaXQAf2XRXrB;aXUg%e5>AQCg9Rx*`5Th<@Tj>C&4GcHL!u
z&HfRI7>URReGK$4Q;<rLO45u@rz|r<b0}ougJY|*^O^QDO0~l@4?mA@S~G25Y!nMc
zMR_vFG}Rt{H!(hb)>FY%pJ1uN>r;VmdzU_=8Dh~iX<fv~43PfWvr<ODdzu`m!y(Sc
z1)R@R7MsUx_SSRd_GM~urcRct?<h)^`=~0#7Z3^Vi1=X9dTjW>nmZ55`Okx8Q<iXQ
zGE<2h3mAnZifh}fK3s_-nkTOeGouz?KN-4I&huAK4S=yR3uLidpsn~<oBBevN2Z(?
z_J_RWIvhSKO1d+-ZnB)aw<H*j_fh5m4aQOoy`gd;Mg#^CVo0g+wNPUSG^tNUg&ULc
zyFgetvd+H{uvp?!N5BX+nZ<`#b#I<#R&cB7O%|bJU`BiEj7tlM#e8;yZSaWPTXOwA
z7#ZD#W<nryZUT_DB65KbN7Vtgog4F(ZsLJZ=GS@887<Wyt8V7Li;%}$P(Fx%lh|pl
zoZKb0YR^(FejoX#-nt3Q=4tN*6O0vLQHAN(*{_x2C|<m1F4AD91*()aHwh2Kv>9O_
zx|G{pRDF_8$KbtbwS@PUO4|XKcaG?!wdHAy7aA=bv?AgPBU(Ki{NIXXc5jn8e@1Z^
znuAz|O)~UkOAZc%V1zg5A`>{jzg(<p|KuWsd8}~3oprPH^mKC#vT_l$K@+!X<if4R
zW=cmx8zOOIDO_=?cD_iis*{E|ZL71>5IOm(o-4HGSM6<22va4gRUqfN^rVdS@VkTM
zx}jfuIN(n=LFiq@$zK7GJTu(@Bi$}}>N9Sb7_CY<HIqc&2+(HXSlBJjo+~BiDabJv
zXw2~#r&IR@nh-GD^g?2`?kisY&vzgsPO*+&Dz_OKN#7Lded=*Ev@?Ut)Hk(f&IOut
z-G6+r<(w<(i|>rIv-ec&yAPs*NaRI|QBQe+6sPn;D%+dku>N*m^<_+MZ8pC%Lcv65
zL<#sD(2jNBAsjsf3NtBM<KcO9dA?vk@d(TOS8!lrWc#n^z{JGH{NIgsq-u=2?Q^1T
z9n)wIf`}!v(b9nWcJfr7sq8%zg@+o!&Qmow+a%N9J(B4J(@6_@q0rpi^w@V^L&>wF
zaK0X$nR&YQ2r9~xCcrk2JMj_OBq;$1r^zRm4zEw0Mo%9WriVUl;#fF8n;K>}81mDf
z^V1cY{nPvO7$~v&h(rPejey=TENXd*X-*jHF_2?xj<<c{xDb=p2#^@FNwRKFi|w%W
z>}XGNty1|aolFt@D>|2~y7*T8{TZsiCvIZb(QK7C0!bi}L>3E;q=8H}ou^C&<GI9<
z9Wvx#QTmw`U^7=pXpPOP<JM*wA%z7SIS~$GjW90^;&4CMIbXbHgoZK48Xm@A$rE#x
zVq|ge7SxYJVeYR;W{iL_hk=pFh;o%+rSJ|#O4WhQ%Gb61+m2n;x`HQk{MAd>@v(DV
z^s_z&2Yho5_hXFcwW#~LK3kXUtCp+iW*~?TfJm?dB3mM5@Sz}?XF8!-`UkU=_3tLJ
zge|nMCpQB+UcA%oqnj$NzE8KP;0Fyt0VcGkyyLROiif<H(KynEoRXZi=0@0!nx~^B
zhn_oURkGq+kPtXr`2aU@CPsb)pQ;|xf;;3qmZ)2w#EpQNgp6Oq`vD1sonp+5U?9te
zkCCZ$=e|ZiRHj}Z{KscEeT;ERVK7p-_2J}}N$@dIS~;zRvicEqSFR>SThshkw;jxu
z6Dw|DY0lk4kP40{r5<Iz44+inij+<9B_7;G={_AhDEPs#L_!Qs&f*M_xtdQqjI7pm
z9`?&<5<@Ds`ZAx#=Iyg52T^)9>du@yB4$!9>Kvu1s3$SGGXEz$J(r2n9SFHonSN$z
zv*B(lMF+p4=vt+ibFzz-pEA1uW)4g~t7d=S+Os*W+3@(p#Igd<6IS}aURJS2%u|W4
zlggF<IzJ$UnOWU4ju_JVt<IZTs?|r>$Hg%d(Z-sQPWS%B7_o20FKKJII*S%)y1!T_
zQ+~3idE#(>Uhyv+DKK>jVw!JtB2nv@P~G<DWbzS4j|7K`RVc<H#pqcyzIWmS3Y~Ww
zEtWnOP8%Z%fpQxAN%T|(Rq10E#)4WmD@5ikjzs2~&qNp*X{@qOy--t0u^<u2j>kQU
zGh)(nws)MZfl37T6)BO)#Wre4+f`B=H#_3rHJ7fq)nKCB+y+F+4{YX`r53hefCn5S
z?+gLnOJ4L7gTG<z^m^2-`v@N$w=GZtoU5#sWoGCZ9mSbY2>xwgsM&SLX{?7IxcDOj
zE4o4b#TrTYHAbNJX=?{Vqft*_!y*c2j{VmbVCe<isLs@&FBHxr{tK?l;-n!}J-f+D
zJGYDB&PTuc!$AtDV(n$DAC@fBvpPe|O&7t8$&YG^AeR6wBwFLqtlXzavio0N50614
zFqSGvVTmr1$>Q&>)|hn7y^Vk4gsdY<H@#KU<m-Cx3o1I-_sNq9nN&fOfiiE<BU9WV
z`?oYfCb1nc58an5Qy&+rEKo=ZdcyANOr7?5jSJrxTeKQ3H+aP-#FzQclhp(Q_eWO$
z4?yE!`hNf#GaJ)?7tnGu<dXN<5yP(UX*h+|Zcc`hN_CZ4OfpoPGS{>vj}i_G0y30R
z1tOSCUY{*>kXu`fq@3;s+0P`lKHb_N<hIEz?tGrxd%N_@T!>M|q4I0oQ6?y8kD#R2
zOnP!Z*xAb02+j$v(=o>b(F|#cf(Eg2=e^QO7sCG_C^T~!TD*1so*g-#d^<Jmu~KLk
znK}AY-XD;1_Fv2~iA>nUr%YbMEvL<<zdI<LZ=u?`NEFb}$t9pgb})~eLbRJ5$QYV!
zq!<kO6$6+dmBUO#;pEwy>Kb%lC3kw0I<8_;v*cq-_5F%Nq>?MXEFbi-i+hBPuKeDo
zPC$e*{s1JjVWS{P{H}6~g%L0d&S%O}%zP{MkHUo=5S+r9f~lRrG>$?U7^4#vMILqt
zkOW4OV$$pD6117FgOkCjn`W|~dD^ft&yPQ=K@jg)&2lQYMVyR%f32&!6kzuq?E@o+
z?KfFQxyNk$S*M)7f?*=(Az-3QEj;-21VXG};<B%hm=wuHFjcHFXDK6!s|ulN-qbkP
z*WRkxbGGHZ;;!3CF_GDrxKQ<%kBUv~DDe;o#U(;N#tZrkdf)_U#Dx|fg0B<*@{%|K
zJh%2D7@20&w$4Kta#ehsHMURS6k~XHT+d|}_X+PhgN<JI8U*J5VC)@$LkYHZ;n=oq
z+sTe??AW$#+qP}nPIl~M$F{9M=fC&9cker=-aA!YGpl-PYNmUpzFMolwZ0y)onmft
zrl+GSNItIuLfGT~UBb~4@!V}}ssM%V{N5&uT@DPQK<14rk68_DiYl)l*32Rk`|yik
zJzxxsQc|LeKs=zm68u!UzA{v#IkayG?pVG~UDt{y!v_N&1pIkW`)#*%L3ca3@;p=@
z-<hvB5)C&Dee>?D8mN_TZKh+fd7`>eMQLW}p$)UuVY`xM!fdOp=Fqp=fK`G*3IvUv
zp<BgIqHqr*fyevzFyD#-Bmbk3H}<jWkX|*=sCgzD<oIfi@>&`DHvDLR7#h|=`lAwr
zDaF~cI;c&0=XC)U)V9^H3={C~D}<&k5f!{YXbCGWeQg&S=-xZbIH}pqQ2cHnMi@`U
zRMIwkQwSq}>L878Z%?a$dec+i>ut82c`v*zpx=6bVSc)L{ncAdn<d!Zvdz1ZCZUu8
z4o4=p$6)kW0DZtde^Xj2+dY<g@#VO7_Fe@xWPt#As&Qh7Z)Cv&QR`wb-Y1Bg6n_Cl
zqdC6>QC7iqBxxi;-3!4tMQOX`)zj|)uywe!UE7?L&A8dRPW9aQjwJ;G!^L8%f7z^7
z0TAKd?FV87eq5@{?FUoV!s^}Ti#-H$m|vVoTKB(HsfL1`6lcLr;zu81kLyzLOVW`U
z<zR97Vo?QMyCmtdUB!%r9@GFupb<x-urG?~o&>j&g40{9Z8^i)%V~}ng(3qh1;#aI
z#6=X<ONVh5&W`!5;?qHLN$9)aD_Fr+WBBddg+~O(ZXAd(?8z3~ny42Kx}w8oL*m7q
zsRYyP6$0CgAy!y5fkQPaCLay>s0siH;E=Ql@bDXON5pQaMb~(OBCs$<B#e}@u$Abb
z3JBnrEINanhs>*LG6J9j1XV1dyqXcXM1nntB`QHtDn?Q?s`OW|&!{cmcY1BI{j#NY
zp4;c0jYgE}_u|sEB`l`m<65tq)3x1=1K1gpkLH^N?qE`Um#zQa1dNe9SwD<4z4OPO
zhs1%GniA=CB!Au9Hqytvx%V@?i$4kEKV#;94-o%@Qu`4QQF5||qL(FLWMKG*sE32&
zk17bukNIChEUe7`q4+`Y|Can{)|83GX+`Xwsoh~5_aCe<K$YlY(36NWAP|6&04ImW
zJ$NKg*t>wLRMGo>?;_GnAf+A*p9MiXJGF7Ce^IN76^jsrLJA51OTe+n(-4CGgF{nS
z97icWVh9=XoFv5f*SxG6cMOPhm!A{?)22&X(jN;e0&GN%mXaY~Vj?brOc5A50|hQh
zENK=#!$B5&M2@nJ(oOgUPKI_L4=Q;+U`L^_rlu&$b7UqXa0k(cU*UuDOW|Y51p!)d
zHav+y<rdGk+I}oRKff`{o)rpn&}apl%Y<%=%E%fG3Jc9l;igN2kDpz}OpS&@z=Ki%
z6S$u|19`nUFN1RtW$~r~XhpK^IPd3K6l6R>(<91oD*hFcWoR@jXcmwr{J@TSpga5V
zDkpQZ#lV{sDBxIz6bW|p^00*#X=t=wW>Cp_p_Mv}9l!IClFU&n^jfew^oD8lVA3F@
zg#wr&)E~|JAl`%a#2zI?G7Y>4uE%=GuES9|0CUOYGyJraa(|~uF!M`ecvDZ&ip=7g
zL^VjsfdLOH3##26brbtT0;2ZmX#IAk9qaY`EZVc}YE%Y9r?i5=-`8Eq)l+>521M{Q
zO>+{A{X+AObHNUlh5-V;X<^_IcZSvXa4T#x3}8gGBfA<5phSQs^M~=a!%xcx&1_|R
z=m(umhz-uf?!oRRp$@I>ff_a4eK~k~wdUZ1QVno^WWmAj?eInH`NFD&A3g5v;tSq)
zrT_Nf!olC#`kQNWwH>r~2{&3=%HC>wT_a~(clH+eE7!;C@vi@JW|wMLKvY-bVi+)Y
zdzTx=fW_0n_5F!;xn}#p#^Xm{XTNd(zB85GR_C34r8j7Q&~#L+^t-ErkJtN-0H+uB
z+0|Aq(*a-k>#NrH9<&JKZfYaf$HV@`3%UxR2Jy!C?-?T)8cx*cT6LwbpX%uK_ub$x
zxt0B;*6)!7&+nto&W&@^(sw&hD?DDf(c{wG>uy*(4~(tJPa3znX1UqP)9--m%*;&9
zv37^&qqkZ4NGB~Tt%<fpd-Kl$Lo@09fqkZ;o45fXD0oVq|KEs!p=2r=iz5>H??!Ik
zf|w%*pXd9acklN?Wu`A~47|;?56`AE&V1xPe(W|LzK%k#{SE`EY5`;Wu~N*{nPmj%
zLPm6fgJ`Ag0QvG~=eci}m$z@nyU@Y*JI+-FsQCj%kO(@q-o*Z!I;-C5bG(bHK!)A{
z-XKj!>4XGvsw4kJdC)S&k$<J=xemtwKyfjJRqoO*ET5;dC$f1k8X2NM5alSyb@UF7
zX0}=XyJbEKHX5l^4A=%jTFIC&?3yktW&Gx|YBc7Rj2=Fl5b<>_SZNqWL>ww_ON5;?
zGlP>aZ5_o<bXiN8O`dYWpSrruX1ywCIg)w@Y9dyL0yPu0q|aswH00mmp3=R`SPL+b
z9ldBLc!=>CBLQAvkwZL#1(&s;XrN5GKqRoI6#Mq+`U2*@taVgl=#aq3e({b)m>5mH
z-@*y_ys_4#zYhxk&>m4@y>rVS{6(bXRb9n(w$&-sySV;^xXhbUPuSpYA2!A|Y~cHR
zni38L=txwbm*-$xqr~#$npY-b;5@2sxQuvmjsmyAyCy)3UWu-CrZ6nkh^By^J>~Mm
zE{~$8(ESl`e>aeUaN0bp&fzA|6WD;94#s(pSiKS<&Sz?I0S_*XywydIc&XtI#HD2T
z4SG&#1`eDO;e}DZT%vc;tH%r4JZ0I>&kLGuQ2*`lh_YBt^{i12Hp5@4=GCO3@XItK
zWfh+_eR<VRSDn2iKi{R{M+&O>*_8U?KADx~YGHmoJfFUz+IX?UIH~y`txP*)<UdU>
z`B8{7{-T(@HaJRW@p2(aT*b3E#n2?Kq9(#(SW0Vgb0JA=#r3%PkR-PM=#ou~rlu|H
zLM^fOOYL4A%)&mqX$*c2XwK1$TrOuBB1Ah0lIA>YM5BY|u@mKxT?^N*eb-+Q&rX<r
zSltFS{S(ZAXBPm=7EkT9ff5UPUm$*r8yn?>SIMM#``}YB`lZ8R8Ug9XKvYC}Jj(d4
z#N*`a1CcHmK!4j*tv_xeute<P(oDlo?B2d{|8}6)rzW+~sCR(+Absz#GH!JcBM)rO
zcbJaV6;vrX-8RUw_N*cQAUzlN>CW9;)+6MXw}!wrgi<nLDq{%97!Z^(@%!=x4wVVK
z;<%h!TOT_LrQ^D+KmP3qvxbV97Y#!!z#a&mp3i%i$U2~Y`XLTf8Bug}8CUF1rXNHr
zh}`4inyZbd+?9vPbsj=Wt<>%WsHN;_cmT*Z<|7iCD!1I7M(+ke%;FYSTbPz)D9S0f
z)GcW<J5-5z$-d89V0kpmRrFZ>8`O*d*@}V(Q$6iXw*A^C$pH;kbBCwv&n>RA;1l72
zM}E&iI^O#=7-AL}duh725>zBQ>j};J^~`m`A>NUC95(tDieiBaT69A|_84u1tq5^b
znQfRSs>RO`>R1yLu$ywe(rke^j=6^o0rO9|39p#-P+`NtUw@DNe1aqc!8`ywpj4f#
zVTWbch|#Zh`^`T|ao=%4J&1NL<rlq>`r<x%=xWVA4MUudjTDQZmBNk>fn*rnxwuS`
zYh+wE<UG;2z+6&`UV3^IkBwBlyV^ceQ(lzdN+;|%QO)!`+i_sG0vqFi>*AOb*tPYy
zvPrEjubcj1m%>AADC-ntzgn1UGP}CQ7i4m5?wk@!GkLsS2Mqpk-aI%ZlVf*p8E=%y
zvA(lPG|lAkbTu+eG&Lra19sZ6eIP0^`Nm}vNgiD)L`>nd>j^41mrBEW{_1h;V;sbz
z)3;a`ED;^6bwE^F_|3!$idi)#A}ymiuqe*b8mnyR|J?C&19f^v6pnBkGizgZF(`LG
zL$F%KQ}ybe1p<*EH-3tq1oaq}_@h$&ndX8Qq9(A&?^#xHRUAa+qae_IFW4zWb%9j^
zeSe}Uc?<hv%d>BegNH4CGhBToc|e!yqafHh$&r$RqyjHQL!qXGhW@H|`)er*?1JPh
znfnEj*-74U<yh1#SS^Zq#@{Q%y29%SwEEU4>7kFs=hGKR%ONQ^T?0h=?1=Q<Ra(Lu
z2-OGH2#;|_I)WQj9D0E}1#AY{_^Pg3BNMM_QeAOvw}xu%6Cy*f)f@vmES2s01lZKp
zIQ|VxDn@`3eolYnVV|AKa<`Jwj>2g4Aa{6*i1)~?KYn5Eyy75g+Z~w3MSWhFBnoU7
zx%f;~8QKOh5bCIC_ie?|a6T}fV3XmB<<sTanhobP;+r^gK^v33`?lOIR5|jaWxdy*
zbcx1#jDNEuX+0}PcXei;>0xqW+LSYM9-mghmx6vnCsK#t{tL(Q$4UGr?tzh!k>%eF
z#*<ZL95>hyy5CiAC~$*z0s;KDveaxA%rB<Zm^qys{3z87Cg78cwAg!l2mo(@feLTK
z3rvEaU7j!bAVfnUM0>Zq{%(4;iOCBCL4>F`d4@won*9NYGH==t^R0F{C(+sT`7rF<
z(t}cseLi|Uz145^UgfRsz4ZCE!%sYgWC0@)z{>lz(+u72kH)x;5CE_khn)pLmj)@w
z*mn+VpjOpl+rV*nmiib%&}PwaHLQnctgURElF?cr(a4i(tpbmOM1T~=1&1;6cC%{F
zp9L0XqqQhH$jrq+m@-~!k(&%2xQs^|KFX~be$-o{68!q%K)1kFIRSt?1geLitG`#k
zLsoBmCo+DdRg0)wE|^>msN(z`J>Ba|ThTp{i2y7LFk)sD2ZFgu5ewK1le2zNEaZt<
zfe<jSZZ`%_;x8!9HU?}UG&i|?kc?fHSY@2Ap~Bv>Ot)K)tn#(aN<{(;gn^_A(M}Po
zP=KibsM`8|t#TB>2dqMPN(cn#?=(cx@327upV;ZJk>o+)PT4l`L^vZ!dM+XNCU-pY
zZB4kr<#`?^oCZYJac9UcZxy$+QBKud$CM7vWKi}!92cij$KM!)00OymY=B#xOO_Jw
zC^{?b+RTx?;lE^PUV2Km;9>J{Aii-qV_XT>3<HH}8RoV=RTt+DMDQb5R}8@EJ<ijc
zb@rz(Bo&g#YfAy|ilX;|0(W`+7%^W=y0T#*7VS+Sxg4s01Z8NUBT{b@6Fh6J0UCRP
z%y?!gQ^1P@eP$B{G!9{fn$up~gEj;Sg{yXTGWOBAI9$zJ8@kGMZd=IuduqK<%6Xtb
z%cP$&?*R)M{cx^t@GVVmso}4xi-C@D_V}?Dt+n`sm|jN{0p+v31Vn3Yi}ITJA_)y>
z^4SwOB#B02VFRO)02tduAwnpHG<qOoOrSf<RDhfUp!@(Gg`34bhWr3;JLzv5*JWa5
zYnSmOwqIq(8eEfGUydo+G9jI@&0yp&L~1}2%@Dk!vLU-ZWg6`}gKuvAW6As0ajqUO
zKWv~%e6fwK`DP`XRO({z>Q1jmB<=W*9#`FX9p6(;v(5ANCKxY<0NhO*Tgr}A5;*$z
z#aSWXw)bN`hOVJOArBeLK~N;668{YcA4I4Un*!eKJ3XEuu8+rU-YsbUD}(+YzY{ud
z=E(z>w4CWj%crx4N8?);aOagF4t9z*R(BYL+mR4>!^Z5@UJ0v^&25^%fs%EPSyVH0
z<GMM3n9`B_hGWMsKD$ry;GsmMP<mjt38}v_OI;&Mqi2@|u$`d6Z#8_Fb-^;m1w2OV
zdcEWL4XPVk%;D}&HXQ>c`LTh9pbSu%A>ns!ZMSa=;$O;3mj`GrIIFBHBzY^uxJ#iD
zv#dcipm*0H15lsm86=g`$rCXvTODKVt`*coetHdO;%@y;LNQ&kVFDk|HvZ|;QshF$
zEuTPC0{bGtJ8QaH&wsCMzQA`D`}_YHs+bv={{>VrF|z)<P!;^Ypz4HTN*RI=!f$L!
z_9BK`BCC8!XC6;4B4G|u;s>g@dwZk+2`Dj=WP`>~w+0=yH%N1=P2HZ(m!{6n2DJxF
zpeO}~t(wyr8TZ5aId&VDJe~}{4A6XQectWF>aau<rV(W@FYIA7XpNY9dtMzn)f8in
z;a#8JPIouXo;J?vH+G@8#{g_4Fa;uz)Wnmh^w~+E=7tmOk_Wb+X3j;pekG;b+&6D(
zq}s$E8Y?Y9Crp19Hrq5G!tf(ehD(48lZXb8<S-fk1*2p#5kd*{4O>ef6<0JiyuS8_
z9&!8Xu4!qegHp1gUA7Q_NIzb4H2>l~fc;C7HpwZY*2<~Z4@=Uv5o*w+M&;(yTIsZ5
zd)LimY?->!Y@Qq0#j1V*A`uaGk|>M31s*NOkDN4spF#q7fS1;X2%^IK;P5MN{o(Sm
zczhC#sG=d9Od;=q2pAg$Aupy6ec`BiixnSTN_5BiFrD&&&aG0$#4qF(+Aw2{+t^rL
zL|lOgk<3Dm)Ue7+nLiSD@`B`Ej3+hv4~xD007gQZ4#1-VpxK_ILOYTfT(Gt>w&#@Y
zv&vdA028XvF+qj`qa%)ps{$aG>q%~fLG5`D4B^^^w{=qJ9%XGIj&g)i0vwLGw0x4H
zuy`Kup3{v=YxDTs5qjifsgjcB($})H_Bs1^Aw7H??#-_YSNE(mE{j(3k&=oeNcRk!
z`SSx&_8ydo3RZfhVX?d_h~3maEoA_gJ^M|28ol?HSUhP`3PfN<B%uaIK)Cxo_S!8{
z?ECu@)%~UnhwR>`M|s@!g$%C>xWkTcDEYjgc>nY;=vh(Fzg=j2t}UDR-!zS03+DG`
zaC*pVnoVFG9#VAjcM5={^l`QB3QM+Vj;h8Urt1_xL9IQfUb!3OY8imkg8_~1orf7z
zN7+ZiiSpeU%Z)>l*ov)?!ig-^F?1NziH#QnEn+h)J}N`<7Nsj#)9!fsybiwR<!^OR
z*r6ok5<-3h#co3cr(G5S9!Tv#j}?v!jfKd=uOgN=tB;yVYYfCnt&=FPGA)6^S~^tm
z5`pu)Zk4P%#s#w=CJ<6r$SDy}vmRf*E9qtY=U-+THKFjwyi8%-#NehVfmFgsw%BY`
z1bx%^D;mL-{!eSXTZfr_B2Eiw*6RGWk-sSdF~y?Kh`ID%%KY@1re-}i;~1!mW`Uqr
z;tLk}(N1?Bi1<v}%KS8JDSf{EU|9@c7sP&+z01l0&4&{~4IA`?8jO^s8KQ`yy3-VY
zeQ_~<&K-<On7(pe!!$p9?!t{+j#I|<KM^}7J5AiQL0JV2V_h8Kp#8p4=cm_iF9Vpg
zG~tM%7FYq+FZxqTRiD|yiML)I<%)og3eX6-1Ece~lN~7efn!!WeJ1fM-LUcYn+qj0
z`^xK=A_!8KzL7qYP6{2>PWq7h4~xG*T2h+9>-yDD#WsnKf%%w-o~rZhXtplFN*!@v
zCY`-=Q~+e-MHN~|w)Sw;S`%^GXR)$@vmVDc!6O)>6eMP?S)$-TM@4YF;X<GLSngFo
ze`l3V9aDr=a>~Lt*IDKl?PP_g7vy){>71L_7?w?6H2A9IaF=~m+ge%q4cO5O#`Q1o
zknLZ&ZcJ?dw(C}`CKLM)*X_D?c8)I%Q9=PlPNV@Jbe#pVLDthOQZU7gUcp-c-=?Lv
zhd4gTimuxxF0uR0?|N$gPv%@mKOsaQzU`K^?H13DJP1&D+EBj>{w#-{LexMQ=myV=
zceD1r+2hmZ`}<=^o!#TDfbW*iM_u=KowwF^`{(JDkg%Lw7}5|2sNbnyp`ziej2S~9
zEGZ0PR3Pjw1!6rMF$(E9Y)rV?=F*z{M1|^-K;UI4)wzncGc%T0K&YGWAfgx!6N4EE
zA)X{QHmI)r9}{7H1pLqj=?&B9-`6t=kGh|Z#N9892(hv^<>=eOoG%<8Ko}5SDQmMX
zXB6LJ11!SQ!q9pLpezb>1YmvgBzDzATFvK1pyKBAj?o(>Dmtleo~%jSYiHNs>f>zL
zLv9jbdO$1V=z`MVkY18u@?a`}PzW$VNXXdvkD?4c9k?A^K)x2|+8s*4XH&sET_vJf
zc^DQhU~(TT@TjA>tVBUmkK4X7lO;Dgwn)UHI$1tBSP&5jL1~LY;l^aF;-%#q@xC2J
zPS9%)=1cF?8U$#h%w;#KEa8Z<d-qAJS@YVRBjXh|8SUYG>Ezo(!9ghTmdes`gFzT%
zzocHXN$PeWu*~##Yu%6!XY)cvQe|aVGqt`DLy8(meM>ufR62JZ{PoKKd|>|ybMC|g
z`qoM=m0}5cM%gFRZ&FM-5sVOt?+A<Q2ay8p4ljREg^+D~PNChV<2cnm)A&>Dmx?8;
z#(GpAV~8Zt#gKn|((Saxamygm%hXdFK_(cLFr3w-R##~D9{N}88n_cZZ2^vi2_JmG
z?X?rzoEOL^OK-FvRn4(0>5pj{e}?_yRXPfIkOTjmb4Eu2d<Slgu0?24fD_oDxL(Mn
zVX=6QJ!>p&x1>o|^u7d_8RjW-RDaO~^MalUGIPr}c1jWon~q49(e5W$*29EIW7~AC
zBeLVr#7YnGNtaF<oOqg-D3&%OSW5h2Q||OAbJ$pGui{#wK7D2l0Z|XL>#da$fP;Uk
z=E-Q1k%FxMnh2jOCkpSiD<KFKx5}<oeD7c#Y;<pq6?d>GLZ7ut0v^4rVVRz2hj3`d
zxAkdSw4xwu#$|PK<-FX|Di9}%;xL95*=0>l%a?CE8i;Q9N@+Ye+jdgl74=*kJff_I
zyZZ~`qaEm5)rUKu?Td_oo`ymH$L^q)N-ZRf7%x{kEZTA<vC6s955;`<#$ug?+r3As
zvA6j?>hPf204}M(GD$i~XncKIs+mr`F7E>>DcT{=L@3?qdOG-5PXQ;`6s4|JRFID{
z?qQy2;?d&YDjvq?Vh+F$_nO0U%Wf79ngw#M#Ia5=sMT9BEK_V7gNQqyQ)JT0KT&y?
z9)G_=Ew+HB?ez=lG3eX2d}Q*pgYPZkPa^gW9ygRMJ~Y!WiU$wCYnA%$kmjhz*e3j2
zk`LuDyUzY%lxX*6R`*NC-cO513c&`xmp{0Wb!my{I|WxoJB+Qbd*M7cUKPE^;Qqyl
z9Bo%+3*6eeu&oKwF%11QUF)Dg^y!>cfTQF-_rYwg`_VI48q=p+QO9v}<|(;{c%lU^
zbPT6fwn`ETUgOe$Zb%TRKgCKfXLRP?rlTix;_>M(Ufu5In&^)%y@FM>75i&geaCrC
z+{Jc~sw2u6{3?Jm_Ua4&bnoMNk1t@x<Mqz~%Ea)`!tDQ&R^ep)kz4;ST7`}6KLzgp
z-?U1%ruIK*6|ZgG{?UWr$*Ut(-~hmlfCK<Y0yPhDqL*DFdL+V;YCBh;EZ^J;^)rLX
zxw||VU~zWtM&{uhgTxEECrN8UP;;yy$+)u@8i8uiYM$%-2yhwJCPh1|{La1AK$1gr
zH+~CPD*k#Sa8$?yN2rKzMJzvI$cWORfQLVsjtJLgVt$nKx8SPHH+VdHxMN)!N$mSY
zzX_=MJqPfI6Kyc@S=21L4EoKoEZ|Jib(8hdWdH`K`kO#EbOb%?w%mt>WHsO{D2>Yd
z7$MbhqccLXUIR=BQ=#~qQY#m@c$RLvHHN_%j3C)P!3Zd5&e||>Qmjt%2wG5x*rAKL
zG;BGOE^k9Jaf2)?2{2)($0{;l@%cNQ`402Z5`q#V<X#B!TejfWS(R0NjPr{pY8}k-
zQ1K<rOg^ACQ}hF#&@@PYxm_-g!Bc?2WZa;55S`QIO99TD>`YJYa6J4WyOHcx-q`{5
za9)Lp@i0Y}jnzZX(@!0u#>>jKr344xe<A2^9P8h)%9)GFN6*gTr{25;17jkE>{#9L
zyr~A#iPkZh-q<lI&&!O`Nv86n8JtrDe@@xe8=ry)yAwu&D25N4-KQCh-Tmu@W-Wg8
z5GC4L{GPWVtYPm}K<(NR;;9emvDL@Hf8znRjq)PFttNn-;cp9SZ6<(^`B!rGl%LKF
z7#H9mfUOUzR|K2PNw|r)#H!k%3nMJp(cxlFms(7p_VeaqZTIC(_gZfE<?-_Jb=S`8
zan}}p;}w5<^XuO3<H+lI@bpS%U{+=u-`DqJ@OFc}AsZ)VJ9~>Sh`m^Bu<j2QtanHX
z(FProR7KnA`OSqo#}ApaovoWYc<cT$#f#_Z+7q2n_`&A&xCHoZHgbOxa-BQiy8Ck6
z`@uRp?C<oxJBr`??KTs6e>n*D67Uh5sS9x&vN(e2fs4!71E>%H*w4(5qq~*8-NV&?
zX6N<#A;Qta_xXT;H%Bkr)%E&*YWIOgn{D&6CxY0Geb3eB>)__b!Y3PEtCTO-=grko
z=DnOgF@<4O@<^(WyJM*7+<58+o5l4Ug(+vofclk#)&9$Ehi=UN(nCPqlwGPLt{C<x
zPmD1RcnypeW*d{8^Ug`%l7A_%9C!{K53U>6jnmF)U!%V%@C;ZEEKlc-_rT5fi3Mm*
z<5X9%Fh)Q*HeIS|Psy+6W^b)6-rv*xM~F($7YVELiwGOraa#tgHAZOE9ekdV=F{R^
z+i@lF_i+a(jMe=N(eBgYYugBIBP}mUILKgPHMA|9#Lm7#4MQOK@i{x#S?M$rS@A8s
zgczh7ihHQj5sv4GBY$8{-!TnH$!HeS@A=4Vg<|zSOSKcezfICChX(v!5?Khi2Hni=
zbT8fBV>6%OOFJpv-^-iPxXBDpd5LCE@H)qxnI`B}r@a4^-AD>2?2)<*87%X=TQUnd
zee3kwnM)%wO=CW*(G;wB8h;B|-i%KZp!{~%A6pd&)hR8kIYFN*OZ6h-GL#wSDp3-8
z(mg})8l|OCg+T|)9XCv8ZunLVH6r1{Y{%#eB7Dn=@9HeI5DAKOj|#bIQeT{*VTjG?
z_^zs1D{T7}vN>^{2w6mHrwrnULFw+e7~llbZv2p88bV{I+>atDXxB4raz<ceF$N?i
zgJp@?zgxmh(3QsA;43{+91KSd6P{*J8Wd+{G4^GNd&e+xdZX|HMw+K%=WQXr>3sqf
zAXxGmGKTCAvD|ZZ_%?R=-ss;)Y#gDOR-4{uNK&d;(>R9?qurk|02tU-cNm?NRbq1S
z%T7eGXg1|wUJPT@T6Xkn9}N*9nQ1seju4{%Z87X~)C&0I2;1(vpdZ;_u^iL**<X>z
zWJ;Nq!w>{Vqjm*6f?ixJj;`+)zu=%pWIeSz5zT9D5b4ApIg}J1O0rKalegHqhhW77
z(syOa0)tl(MI_W683S`6kDrEVfxfgodAsi|G-3W<?Nn&u6u1_?i|^*G@BDuoGvGP!
zJos*WH(oohefR#yz&{=DNRW`~45-M%PQ-GFab5yFiP!_E4;Tsb2fT!xLvQKLub)$!
zgHX_YV=Yqi3@iTcKhhOR=acy*pj?mvz!q=f3}t+ZeT?mV$L1P!I3~gk6PK|Q(%>Hl
zL2cGQpTHRr&`=U4(QZgDPd?!&>Y|q@gE(lb<p4$dfI<3a%E!Jc3;e@H@%85e!_yUi
zXa4*N9pddnZP6ig=E)9_OA#*Ke`w^vB?S*k)N&DPpxHeTs<l)f7&d9^fX7Yu`BbOt
z(XZSI7bu)f!sz0p5QH4}o0eDGAtUQ)%q_+O2!&V~nPGo}Jl0hb`DL!hqnD9}6$pg+
z#fcI^&pGEJH~3zaYH9O_S!=TY<*-rsc|rP1N-&9|7H+co`Lc}@r$VHVrq4%W?4Af7
zgmi|6J(SMJbH_2DZjiNH0w+TU9j>Wz2;etoKBwRSVS$;lL_M@hO}*cDi!q??Wrj$G
z<Dz1sB7l_obm_<j&`#LGs3ahjc$3IDw%p`K{yxgkJisnZD9FZ{AxkE*KT|<B-=o&*
zgi_C~jYe2dsf*7D{RSE-S*jS29yA_nL}R>Jd#_I&gbnqReabE(M}sARWN2(M_9X;%
zo8QO++O5+%fWAw1^C#xhGsL5yp({kQ3a`sQrW_M`2%7|^2zZyt6^&ahy|Ml~K{Jl7
zu8PJ{aC<0|#niALjAO4-)E!aT6&e)xmuaCl(RSUH^=L;PeX{LYEtUUEh+MpFm3PH<
z9Yb`>>a6{RqEZuTv386%4*gqbG!g+?p>_;c3|ovl4!91UoBPh8e=4{Rrkn9jvcD?W
z|1_5TRl#*|-JExpl8gDXfYp&+Pzb@lmR6R4>gOS}Z;~_!F!m-b{4x0IP>v~^`L%M;
z#JSQS)LqI@ne%w6F~U;DzOYnBpvH-CBHokQ6s_3e+t5(N)MLsUg^_NdEfW}+$#bh#
zIO+0)+DI4^D#I?+0s^g6YN6jd@}59{Qv5fcfT?3JG5e06(D7c?6hG&8ovXg5!>rU9
zq1gq<r(o3cbXmf~MPQ@y)XZWSvyP)^X6a{1vsH$qy_t)P0}h;iepwZ?5XJ@MM1=mT
z<@O0e?kuBmXsCcT!fBb0b+%zxeG?jwZa&FW6cEC)19(7;rFxhUkBKUIsh=Ka8*B#Z
zAi}CVxv`L=D4McHKfnMCA;fE}U|=Ia<LNP=YFIIRz66d;z1MqAV(2EeD8%X{r#H6j
zc+70(27b#yiZvhj{l5dK7$E{YrbhnDDItwaNaH=IknZUe;MV?4MjK#IyrljyL$oy#
zbWH_nkueId?+9dnOg5I^oP<n8>Qj_$?6qOmD)_Kl+e&^0=hUD(ag1V^0;XbPF9W;g
z+l`#}_OyUo5gUbyh^+m*09q_z@V9v<#?F+?KiRt^u9iYtP%;z%pyi>OgJSYe&?hd6
zzdrScw*gAgFa-2QQs`%Pqeqf;|1msf2Z2Dw+RaJoH`s0uZt6d~wk*tS|6;9_k@4Ry
zk#AO=bofU}$vbshML8%U52PNZ`#+XStIC&b5~h<1C_riPE5tqayu1Z~KuBRwjNjB0
znr8kw+TkAuFszF)yxp3#_qgBlOOW_W+hbL`B1#a8DgjVNE^nL|eiS5We6)O>P5}qp
zfrMfLt0CYT2UVkIjo7AMU!g;qF!*>d%ygE*pTEPO9}^D1HWlF+20#KK5q|~(6DlDi
zK=+|TO4djD(!s4URz}otRMe`MW~@WZD{#c&6c9oYA=n)}AOzr^1X&1@h6N;XlI5qT
zk=)io2gz$4Xy4P4g#ipNGAiI1HBiJ#tTS)=awbKeDR;AnTfA=qg})#Naj1%HuJyWT
z2yc{k!Jh75*o|euq4Cv=xN{*a3g!jfZhU@mFrdF%{Z3?8DW0)OwvM0&4>Z@zPKZN;
zA?=g?qoF1~8Fbf7Pz?&2kih<k|H(J08~7*-SZHP<J^08*k&F1Fnn73Dy5`na{dN|1
zY2Rvb5~O#92^cnMd4YD=h&=fvb&gSa5vMlKM=uzqHiMnsNf!)8GKWcw>UWAL$uJ)^
zpeR`<PO^V18LOa<c6hjz0u<L?sEERaTFMVq8-_%{UNs1q&{8|Ft0Yl7)Jj0l*K)hv
zDSW8yClJ?IBOcgjE`SGI4<i>^(Bp|75$(jPZ6+HC>W9J*P#7fN7w*r;#|JkErr%|;
z-anL4Ncm>iO&udrea_qSrH6N&X|15~{-Jf?8^aqwVJe&$b3BJ7?O(hER0@<ks*5{{
z_M1LshqDsy5P@!1r~Ib~0{D+{U4A)V@|Sl8rwrfHR^hL7B69u8%ZoC$Pf}Y{n7nf)
znu$cl<}qz0+_yE6j+vDAo7zKtHf7dsB3nWCyOJ*l6%0(*!ibW$>*f+92u95$KY?nr
zL-XQo-MQZ?-sT@hUaC)=YL?<?^EqMTtUog{Zm?o3!{oB?>(ZsOp)bw_H$t`nRj1_V
zf@EoNjjFSAgJ>jr>v;5#NusrKIlC63&It?y-$r5HZm^AfPRMBki%k;=b>{lAsqW%1
z<9S52`)B3M*)#>lHtYI_*_k)8h97!|lp?_m4`e{_-#7w&u*W~uan6;dE1cs@mgN4f
z>g8;si6IVT7#3tu1Uw3a11bxo1v(>fj@8CbT%<y4rZ%k78Emyzf(52FgnM6#I&nt)
zg7zx$0`3uHz}ymlOk}06$jQEc8_c#x)hVbakq=@n-CcExYP1FvF{i@FHi5b;J9W81
zf4&mb=S;ghy=k06)+j%F?9vI(uL+SrlC2f+NFLej1j5TD+q(2*e(o=p;xO=Vhr`H%
zE%Q>6aP^iU;%x2Ph^nT#zH^Z7R&oLv^dKof4AvB6PjyyUaE_giefwJ4e_oEyQ(KTH
zmAYi)s;oIR{;uv)GAtI#4szDqONR^Q9?yfpb?_qjsbucuC81YeL%+NP*txwdzD|8D
z5hCP$7tP6eKF{n0)XF>m7lm)E4F4(>oPmS=-=%blQ)L_r*)T@0Ur}%_1DicW<{4$P
zTezpl8gDO(F9EUl%p&@MIWj+=k_CXOyu)}eA!F}?zU5}PIo3ru-X4E2>eWfEE{0{E
zF}m<kYG_aV21B2&z!~e4Q*(wica!d0Z3<m^XKmm6xy$pTLC0JD*UYz$@0Zie+6!#J
z>h{y;<#*<D;yA2J#9V+vaM16$0Lh{%ZOdVby?fB`@*G~3k}9rvAy#sWlwy;OwAL!_
zjUJN?-!q-xM6`^Y<4DtpD?{fFIKWoWw#L4BY7ct@;k8?dpM<W8hmoT%vBF%UR7b$^
zAmLdm{3zT7;P*Mj3wxV$gzOtdjK*`gJ#K$h6ICQE8@!v*++9297idEL%A54Vz;$46
zxe5VPbL=3qz3F7;qKu{~I>oY4Wx%1^XhzsLLkN)OG1m&~CsBkn7$ya20v!-wrqDIU
z)@;>t=<s4=3?5SsM&oT02Hfv!sLnlZtz{w7N!uk#<0w$Y_YskRy92C|7`?C1Lc(UK
z#S|{IC?-3``D80us4W)KU|>+18F$W*5*wk?%1kS;8}Z%ADm9=cL56}MNJKq={v1qG
z#2Lz7)p&M3zHW$Jk^OL4V*--?n0+jYik&dCvQeEdZVg-2nG=#9!Hk=-lKR0ICfz8Y
z-DMo}JZ2bO@E4BluD25%%YbY^2HZsZ=s{g7c?W~d(mqDb5)p*{HSGnnPuj%Ba8%rt
z{A*%xPNq`&fA8BR^qiv>DP$<K*ruuEpYL}pTO7R$zuLoUodX*hM`GIXKyqNGsQ_*p
zcX!j+$!4~uGuOXE3tPOY_yT}OgIb|>KGNE#7kb)xyu3T;cv`rcvAD27-@AV#oaJD3
z&3HOxz;~z74{z%*+RYa?CZMX|$`RrVM)goC+#<rQ>zK;icB#Hntt>+z{-Q~I6|&h;
zu4@hjHe#wFWUP)5%+%7nWwbVuvWwJ_6q$mIgM-xbdo4Qa0okGw$eYRih}$l6M(#_^
zPgBKRy}PNea=OsOHtYIIHU@9t4MHr$<NTcn&PJ339prKG*Y53*qMK}>EfJD8nlx(V
zCM9sv)EDBNs0@E7xkNazk@9e#KWfDSVB-Z~k+90_dq_xX<Z`t7JL;r0Ff~)5s?szR
zwzoTh6NocSwQ~cg^XSP>t&*F<+DY}3j<4I%dqpmgFS5FPg%LRrX=jS6ckA;d1FEsc
zLOd9?e}i!y6_5&T*JeZJV9lIuDM`R%L7hW*p$}y*8lap?x3uv(kKVk7AG2WDuqi=`
z4MrUbUb!ie4g;nBWPOp}G)dNknOf5;u&w4YxPO;W(+K6{o%8L}p`LqMY+Gb2VVof2
z`zaM}Lmr3c-v_*@4oUz6V&0P)Q+Bsazj?ql3XGk-i5QMIkxT8?eNqL4wUzqqV;E6R
zAJin-BPWAQ32rVh&DZ+r*xd7qZaaJ#_|I+=D;vYVaGMyI|80u-CmY6gg9D-aSoMSg
z_s^XYNvupfh&G7K9L46K0elt7hTx<A*j>q{Ztp)BmEho3a)mLYt;}?GIhGg~#27pq
zo5oDn=PVJd)1(rGmS{CB$Q+_SB1xz}LSwN9*B7y_&miZG1XLge5dsmEWKtxt+uM9U
zL?o5nhm&nCeh#0pYi`3a;~=a{7<C)0F^IY34nh2Uxy%>|(s)Rb@+iJ+;MB5{h&nzk
zwm4yYN?H^pi&*(=4mA4Eh?xVNP?(@wkUow?h4sZ~EkFQc%r-upv8J-SJ4ltX6rlM@
z8d!r3Mf#nl0Td}Vt<i@=)p|B5+l7>;8=mwz)pdrDN~*UX?JiEZGM&>aj*~1pKP8$S
zMr3AvxUJ+0O9?HT^Agv^6@#+2XA}KKO=~DI!-RTRFj-h%S#6!20yWxVy_<Pt13qCy
zJ35>lSE_d|0;+4bA*lV+jl~<<7cq;}1O-wAP1igEJ%01vFi!1^vHp7g1o~zWmmlk1
z7QHytDj<5Z;GQ1|%gGUoQ*?U1Y$%Y97yXfkJ`O_DNB&05qeETx`eO;Ci-^AP;e0a7
z*pgUv+nb07P%z6<ugRiB9X?92l{#^Fprvp>7K1u%u!~eZ%0qi@#I5sJRkk#~tcn{J
zP$7$UNxqpIHHkr*&hN6+>Q&dxIY7+!F@5riefgHam1A+zpqaS1JFK2ajoo)|x2)EB
zx{b{`cz0BxggI&kI<w%tC2!qVlayb1b^A?E`R)TJuS*eHi%^qinZvrsGKLGPN13Bm
zGR$WjeqtaRJrG*;1X-{e`R>7dF74&Ce}19ojdpQ^l!Ec)HMN1A75pwMcZFuA#TPfy
z#hYwa^*)#R>c$DcMZ>XoFHuKaj*8aEpIjU7=N(OEadwNqu=ipcX5kURZXS@=VJv0C
z_aQ_k)z4i@&<(lk%v15;jSxP2nek?Q5x|wk!%C7>s6hW#q`IKR(C??~4mLcdRTqQ^
z_XG~Ww@jaM+HjpVvC+lL3B=c0m2bw#B-6J<w1aK8BGmJ=bicw<HBhh?Yqydr&G;)-
z;;`I$tuynYRbkFJwo>zr=imXyj!#t+_ozVK@bEl7?V5V(!ph#vc6nvwV*G+k8_)BO
zVUG4CyitZKVnVMXE@#n)h>45Hj1-T8Bil$}Nid7f@2dh3T^Te~&D()>;`WNscR_tj
zw5W@zorHNYe);|)XrQ#RX{jFMv=715b{=r6%XGliMxhK%@|aoi`i=kRyvM2V6d5&l
z-7#Oz6@>hWK7SlRjoE_@6Zhbx;5W{&(JR&T)|QE5n|02*#S3PW-bqY?ErV($?Z`}&
zOJ%jyMJm!nYjN)sbsy2fs}B`J&kN#iuIt*AFCa0R!N0&3&i~Ff{3n~h$jtI@hh)X7
z+Kw9(DBd%*vl|o4jG+E}Cz^a2e_T2)W%sA2Wjo-N$(WF(93JCYzJ2KJjHN0PtxR+K
zPJoVa=$`ekPyk_T<S1H>D-+pO6+vi75hF$FI-uQ<!lV<_O8PBu|Ayq^Io<0$?C<3+
zy|b$ZfG3QRNLo>lDw@>6!K(EcCWxbjaS%#?DM_Vutf#ruM+L!XV=vTgnEjI9rd31p
zDQ207UVn@AuFM=}Lfy6Hs63g%+P^*<0ph67R6rUPPy%`XIzCU_S*EpLD2LkT5#Sd<
zlbSYr@jP~2qg7j7ojM0s!?i!fulQ?iU)!K_PU|DTv<`pukyW5@IDZ9@6wh%&3bIT2
zTqLE0D*pR8bHeL7RE7q8W&He+uk2FUbbb0D5Ndx@rg?RNNJ%Fuk!CW0bW$9oqqFc0
zOXQXXO4f`c59EynRyYYELj><Ux1zoR0*h&*u8sj7u9}}x8-B;)8*?ddcfHAdyQQRt
zVcjt>?zd162OWdOg!X11SDQqSiuL)5DI4(Lykl<D(7a$L8c-PtD&C`92`0fp%y9)C
z2M|=^jXJ=0J1Y1JM(5aHP}+oVZnXf?n7Ch#qmu-!s5V7fZBK(Ln=*HxgbtX<@p~et
zQ9Antk8!SYX*B@b2cN%L3m?+Tn~ewn6VqVk9LMcCqbPAt`+515oO8B1X22B|TgtE(
ztS8LR8W83ey(|fg&n>(J!1OSh=3YMqO1tTFQm(rNyt8P4AHwX&cpe(IT}E`p*r}o4
zq5RvbIDLpqsCYa{OV*GEKX-@-?b(uOIxUF=n>S7^th<OHT`A+<GIqhWZ1{K7_>w?H
zH#yf`0(b3ZDzo1EyXag^EF-f&L5*!M4W*sa4w};WWr+MHK*JM;40(-g&h$=>WZk<c
zfyCR9&rmur|LS}fU*=>r%{qg+yqv)yfi1lOn!We@dS3Lm(&#PV2EyR_cJK~l);psa
zR)V9%V<a-9VZENIu^r1oO%sLr#c@a?kp}-^Nm$;^%`?rQ&tIdc>BN(=Yq<Hc*s$Fh
zo!UdsHt6S!g&n%35B~1L0yxU`S@`~Y<+-|*b!0ETHX0E~7;lCkr;>aVD-*z_TxzM@
zALEdi8->P?heT_K_E+WNQ#tD;A~~*{j6OL-^SH9|3&iXv%Qz%xUZOcWvLmAUi+Z47
zGU^SjUnvvM{XV$p(~8I@d-c<&SQ${y^R(x|&I!i*8v6;%+LpgA!B}HZkp!3FiIKe=
zl)K$BMRJgX)9+kZ*Y#EvK7LHexi9<pl^gq&=2B+(QDF=NT@SfG$$0w_d3{3(a{aI_
zlmh#R!o3^#6=%B;BhDvdN)|Jj!$Bp?cd^*f=c-s~trhO;+#?s9K%fCdZkK%y*pBRO
z<BZtM*3f984Bcfoff-yABHf-A()Fh%$-S0~8mSfSzTrDx-2?g4y%+M0-b51^L1c*0
zdkf{v3pYqP*c%?;kAQ`$(3rs*$&OLj*0Tb;um%y5=dYQ8Jx4gLQt{$WVUS^x$sR4+
zEp%tOc))?cXm;d`B1cHmR#%3)I^G>G4vr6QDT(X^c=O<8o5_>+-|iT{mawjC7DH>%
zq8Q!STn8i3E=td~dK^m02=-R<M?mT6QjUFzPo=h@1Qn!5CFFK+t?(PNnV%+hyZ;;k
z!L(mL?7G+ZhD#tNMqQ?7Z4>zP1Z}LezV*h5BPZBDO+a56BjFPq`arsngrSTxe8x?O
z{=L*xA@Djb)l@ot;s8b~*3%51q7=RGCF~^29O}``{(zPU$cOr8FlPB5f#?6FqV9iE
zzyE$(^wbaatJMnYovWKa+5(TIKtu|~07Sz07tIWicw9(x5Wc3-EE=qH!~?eIyXV;K
zz>8ztOkWES)FqGG@wzyMB7!KktvDr4upFxhrHS{z!pxFFA^=Ux)u(lq1zZNJi&FzS
zp=hT73{k4hOrYUiA`yH}>sAWh0HG=&fJG7+au2kX1pG9*Hz^L(w#;ly9TpwU!Q87P
z#A?74i-D3PGW2iO41n>BUm_r{`e<+v5w7j-iXxP)K8uFC6yPsFixG|tAgVr8TUhiL
zlDR|^B8G?L*MNP^e#fK=P<&?blG*iF=q(4hIz1~L$GV_U?j>MUjjJ|YUW}{Uu)eBY
zK0Y&h=s`BsuJs!e+n*wJdC<Y7l6;fEER%e+0Y9P&eF~_yW{09#AL~uo2zE<y3b}mJ
zm_^d2C`iDL*@!(_`ZP9|T`UdMy<uUtm_<voX}T^@x!LHwNGTceM?9~ng3=wsB#aj9
z&J)c0*zpye0PfJg`iR}Pn9$zRq$T;LXP87D{W~;=v$TMY*KP6?0K})sXP0;$p>MF8
z;*xdD)PpSq!+KN+P)9>~juINL*ng<)`KtSRkjNKr1ZJwUDKu{y_Sn<X!d#|-YnVnX
zXSsX=xL{Mj8mqlh!L$IEP`vwiUg_TayzhPv{(E#*T6*~&OlEQp_Ax--eWHYGv*2iO
zCX2J+RY?mme0n8ak$jQ^l7W;vRD(KIP^vg17DrtL>ip%3iw6$}<EI}}zS~<t_s6ZR
zuJEPzE7#wzKYKA9PF?!S9@Af4^>(-W#@9J5Oy1{C`||m|3XY^MyjEte!0!gT9}uxc
z<5<YWOLZ-mp~iaQ<?uir*ABkMx^DjszF)ZMbagwO`Vn%)yL{eloRNLuh7mqR6nenP
zztcm~{*7@a$H)UqFKLbNp~CTfdHoF`WXk)u=liNN6+g$_#r2&5gfEfh&)(e&^R#!y
zidH9EEn~MGSSmMyy^oLY^V>N#Kiu)LTyF3^JNx#ppwwS01NW-m9h}*{!gB1b@Yi*#
z)VjxN+ne3JzTWpXZb7#Ty0$&u_h*rv*MqZg+M%peU$2#(UDzCJ`|bIwt|!TGDQYKr
z^p~O6&Mnq0d>cOI$0sKm^wv4n>6c05N;%T*Oo!t?hEC4iT~cak7ti4#CB2-x`^4mb
zbg7nfF&!pHcBO`5rG<|S3|}H~ey3PUw=gxhwB--Z2{5$sVsYAq32XP?4C?odj~4xv
zhh7vQ-^hz0jq~A&1M3ZhCDM#tYI{F`fYt?!cj7^6Tn!0cTy6JXLEF$EB^o7G2oL9y
zAjNueXO*4DlHkLG{2i`HK1K9IMjO*zK;xgZ?}xHViG!D=Srpi>gB8iFSA}T5RZHal
zd97-TZ*ygMh22t)gIG;q(gy;F@<gr85sw$k9v5(`mll$ymqsPHM4|ngKc*F4cxnza
z3N@U~?=TF&67<0Boq|C*gub031*Ql6Elf8zqqX)5--TZs3)?Z<MlE8DG03u1hjq@O
z3w!Vg3EN-^MopY3tx!a6MW`cz4Z}Pn9tz0BNcRQ;vrn(61;n8%-UNe^84hIWMs^I>
zyvQ;-Y2Uw0L=`AN*EGO__DzG>g9m{|)kIDmQIny18R=ewcr^4A(=zB(B}B7yim6qN
zx(F{eFfxcN0CPgEQ+-Em1<E{>8_|B9ilrX`K(~IIAhzpDC`PwK@st~fVn+qCu-R<0
zwo-`g1))9&7LMNn48f|)tdq_k=UNElJ)GOsr?Uug-Qc<d$e{~9o+fO_OV|jOl1|94
zV|nbBPZ(*yXjL51j3|*+z@x4Ok=turPxJI~j~l0-2`N3{{Sno%%bOswygHUam=>Ib
zap~Pz`vCTQIBt=sI^{PJQJy?+NkdRQp|#(k5H@k5AVk1JaaFSdACZ$d&98*nSIXdf
zM~7oE!DD1f9H~2KsM2S-B0KRfxQIeos3KV?qj~9dGSGRl;`yU{YelO6-6kxRa^&3k
z4quaw{}1DlvR2OBd*WP4C-?3xX{EGl%SDk>dpB_-&tZZ{A=Df98b}|ao=I-hgda{7
z1hNk))SB2)ayp6}U*rftd{gWQqQ%-amdFv{;Dtdl8n29CK(q^BJ6{}7qg@s~X`kIv
zge}8lWS7TEfG7FQc>v)36-*Q|g>is9WNj!|&pN&Ycr+xMP~TPbMzNmp#w?CzS_2>3
z!MQ4p4UPqII-Vbs^PNa_1@<9a?9CyhnMcF;<~p)U$n{2IHQ!xp+&ObZ!RDc1sNP>z
zaA&senGsh=7D}sbp;<6Sw~f@2g$hL9x1<~8pxiRrsxEF3aVcS`-$9r}3E}HqgBY9X
zarzHN$=R$capljhsMYbX<y>gt3d9FcSn#j$6?FDEy~8(sK<<o80#|RV7In|g>P2VR
zM<VqIh_WXBV&?}*7JsU>S3)oI-kAajQ*`?bG?V-F7P~+i{I7%rEt<8;P#icR;D4Sp
zYRfba?WYyfOkGNrpV}lY6RXQG&(lC7472O-Y(KAiPJ~;@J(d+=2C=3-4dB-#iB&+T
z4z$L9WHuW0tY0uIB~sP4JPJk?>Jz?3#i>&U+@Bv&ZD*zDz~uVLCkxYb#{k~Hr$IQ%
z%=sCZPef`~D0$2tGvcd8JDNMgZQ_wMNT4VRGLK2Fvm(4vr%d&fA`k%`S%v^I4mEgN
zd7y~S0=M`VUjZEi!B|ckK@PI0MQjliFwjSsxc=^{uOm;cINstdP1HbPg?wK8Q*dUr
zIK?b!P<x$ggKYzTM%xSrtzTOs{XejeeRLu*U5fkgG^tc6N5-A)a4pI7e;J-;gv-_s
z0(0aKFBBSO)^>*L7>%(Jf~(Y)mI#pHUr*{yHaoxKS|Q+VW?9WgR;~MetjZV_i>}1U
z+o!N9<za7UECTOoFTmN+00aowPT-9}nm`%?(Q8x3z}o{;`6evqHHf+YQV|mrCGyAw
zo2ux;{9GsObsphgwXGgvp&G~h8S>a|WQ#JECGO8M^H(w{o5aQ=Yh!+&_}twGJ8GH*
zCMa76SJ;-KfwH6(jjN~+<=W`DTO(;4Y!(7Mk7r1*n-QQlp=<l{SO{w(0)hfDL@?90
zg5M*rW%Nl46h0w&(Aa?jDxWtvc~J9!@9WrXRO=|}%L8r_!t(#e*gpm5(nsCiaBLeZ
zw(Vrawv!d(if!ArZQHhO+s2CZuKj)c?76G<(<gn<2VMWFu9~yw9OE}^aL5z#!XDZ{
za(nXIQlD5SeyH_-NfCy;4m`kw+0xt@qzfG^@s{+J5@_yx_>)O5s>NTX3XU*hw9O=(
z-;3HWFcHN-pVO)VRY*}ERd73r%l_gjpYp(XiJy;p$JhDyx%jI9-QEAln{5_Sd;NR7
z6^s76PNh7y;%**~&0eTte0-(;8QK!dlB<^O-y(!~VDU(dJRqosP^@Yd_RqIp^Y=eG
z#Me*0Tfrqm!;bWMABAx@dnBqA2E`h!T%Bf?ej}%Wz&in%&Gps)Y*X2|xc*}v$@+g4
z-v0}*gX%OB>;NND$Ua~1!HCNko2w-KICWD2A*V5FsE}4_L;=FL7j$*NA_w2D<sJw`
z%S2V@c*QviFfN(_qC1EbL^hHJ0pl8!yPvZ26+!KvMvmp0-=|-Oze2Feu;`$}v70cH
z+A^NE^==>A>ISW`PEG53T^6}up%d$pVuvJvS;a&DPCpQsHNFEu+0#J%&jYgmr>BSO
z|G)kr{9hc9ML|NHQNiw%xla%j$cPer|G(bxUpjfVf0+9J-}V2NaV~2}{~w)v&p(~K
zqK(x*y@pT*J`5{xgp0q8H)Y+pprvCsi1B2+v2Ba!$0pB#Fk)y6Duwx6B-#96`aZ`l
zT<Fj|^}zfkjo00$z0cF9suVFXGFgE5<3|`7vN{z?dDO||8T&sdk;}Iu2T4aZ@|3w(
zq540rSJ#$|iOSyVszZ5X2fFunL?flBxlp1SfmnF~Bf9v|83wcz+L><P&2HwDjQoK|
z>voE&a;qM54Y|m$zP`{x8wc@WGn=8w7H+=7^~vd_b$+!<yseYk`n_(VL&nzMlBU_G
z8nXwXTW14`+P`AB=bD`uckM4(jzgr4Ii{lDrMer%FVA(4c5vyUX?ZG87D(YHNkk)M
zkepylBlirT-KvPLScb=V=#;>YzZ^-qnrCJR;FzcKRXCn2Uoye%8V*hPp4n=z8}B!6
zCJQJD9;YoVdI=s+`g;I7TPCWU`_=F#%zCWFjp^1Yx?2L#j3<0sw8de{0Cfe-;}+E+
zWsb;7&Rte^CD?B3E8j?^-BuD?#*#>F+l+E_XU7FMWwpyH%ABh^=Gsi^-c7cS51v}K
zj43QNpbSmMC=a+lpdq!4Pja#6L~%eVEWyY`TkoG5$FhYw9#?;8uPot<^h`z^#q$_n
zv9`<`N<JpEU66?(>QkW$pn=7ZMMN#IpO#>yNJ*IjK<3TFkmQ5W9YhL(#K1^OpqMWF
z_=s`Bx}6L;v9Z9JjgtE&q9t5JP*1bR8uKusz%X===$p9x(G+zLzK$oQ;-thTxlfRi
zz-;A0rv%l=1Op3$NUPVzb`ArND-hNJeyS4;jndFOQ3M1yK7rT>?qQqiwhJYqNL&!k
z(k_A?fua();z6al5c0y7GQfY%MKAKG{FSO8QpUKXORPYi&_%5w`tV1#N{$mS8%Fq_
zI(2$xValMNH&=YGDzR@W(<c|;qyR6;w$Fh2+uxm%^nk}>dYTd<5L5|t5F+Nc^ej1f
z87sfHa-de37T^<n%BVoQnDs4jn_rkQWk)}f<daRXTLeU4q=rEu$XM-zic-f|!1Bq6
z3BcUR_<sK^aNnoljLzYG{&=i-wh%U{Pue;+7=hW83^1eLI?GLDQV(t$WD;SsNyNiL
zsfUhwb+4;*8oM#i7qr)!P-TI3G%}%#;O*c;2IH9fDV$aHbNfhXZbcIqnv3j4HF0W|
zA8I=UNs~31SC<bUAt^m$8>Rf{X(^@(nciCP_r*dYp|cj~cpsN1Pr;2o8G7BF7a6=E
zjDD0<2_j;VPpUjqC;}=KVWeQtH3d@QFY3KI7e)LzM0sFIwi0%vY1yne?e^jvFw-Rg
zZDdn;+mzrXuS41<vMzQgM5K5!c?t~g6`!erFJ0DomLP!&=1}18+)_fqK6WwTy+3Bj
zp_cyd#1f}CA_Qko_tNg?SB6mWUy4H;c+p!}6!-b7{4yrcS-mxSULb_pL<rRvP;uSt
zq1m9<t*f@x-!Hj&1ilF;rI_rNM7IFuoVS-@<a>I?O0!JW^*_QZs}zUCN+LxaL54&k
zG}e{hcU4>ToHQ+n>&JHZ%ATc}43$rgD25_;wh2hiJ%3DN{y>WhmDLW>b;@V`!7Hru
zkEKaHm)yu;dN?2N_>T2n>&dr<fit?=Ym~&a`gqQsoITfHzgBbqUHQB8jDVdN35~gD
z3HjIkv%G8RAwrT%>8~96Q3ZE0X3<|RH~9&d3WaQSl~#_-SxP(09UGaJ(5+mZ5Q$B=
zy*!qHNXS*-zYF_^@hl>!dxS~h`Nej<WpNT67@-i(e@ZmN9chZ{M1)2{bU;7Lt{d<h
zx<TD9|D0`M?iwkuKwm6z>fhRO*bdIX*^2LzN^ezX4<?Tu*JvhPCf3Uqw}a!z`Epw#
zu0#Vp?iW4ZL?ZhdQ2uT+0=3Q3D(FlMtnb&=2h23a8*GG_H|ZxPloAEp=t?~U%ay`G
zm%%`;rK#{+O9(Y$_Jxc*LMsTx(io4D5A2D=39*DJ@khss<mZ24HZ8k{|3~3q;`pD{
zmznv0zYW~R-gG!@dD_<NZ(Y#AmHub4er8{EV&~axvbl8rIJg3(6GWltDkBA~PdDxE
zfM$V)=NVs*xMaxafkA=U-KdN7GlvDdnCfk3b$?}uyT>*1MlbG}AcwKWNlp}qKyW<n
zu^&Y>K1!#H?dSg1?*90CdIQiF-o0;sT)GnQ`@Bvrzm;b4_k3|9AJP;VBhLZ{bPHzC
znIikc5vS0ki2}FQwHfk`2WH~9tXylWL*<=)$X0v42MRB`I4i2gzS`<O3#;!~L-4ug
zPU;>y&FMrpT60y|t;|M9k+l4ZQ}(l=hU>_x3;*X+sgG6EqrW_W3O6D?2%>76<AJ0_
z)ehmPOpw5eZ=v>ZJVdD37a~m$n9rfRbatIlU8`{prC8BfcX668C>}OA4%=?wtS|kG
z657MC39)Kgt((?tTS%Fc+=1;8LVD~aqhtq-Hq-}9)69?Aw+%VEo{#0SiVn~8sXH6e
zLh}>6i;(Vw%>>-xJdTQ`-1R(S(2cr#vxDz<hBG(V3t-7dZ7;Z;ba5P088r<}%+ZTN
z>!T)|Y}A<0=XROgE0<rr&2_)no?hplb$LM7hv$_ro<e!$t(=|ekRZEK2WF&@T*zDO
zKe)1;6<uI4m2Vd;M@g?IC94m(FH%JNhJ^fw`yjzaU-nSlJHY*Q^H%`2cW8efMBEb9
zF68H(%!@1uQI&i~5MT&gD3mmgb2I6_PD{Z(kJ<ctF1R4~YJ=6Y+0J4C`=jku(=0@R
zE7BnA-;MTo{nAWqs0|9vc3&KZ_o!Ms@p*_h-7MDLGNNgeGq;{4*FtM5eT&DZP#xa>
z^>MGX&K>4vNk^3(e=AJ1kLpwL*X^TP4Wd2EOE{j*7NS(^er-*ZRAGp(?bjm!<Nz7w
zQ+*07FW}j6)62sjy8paT4NQPIrTBN1Z*Pyce`2gr3QWLVfBWCJt}UJe|AX=D9c=CG
zf=6fI;&j2P?zFV^r7;05i(Q`8Ol;9Z(Rtyx-?IsE@uX2lVYD2*<ex0$EHUMx#)jc=
zU}<6=G6~`bP93q0AYOJjG*OKbUMTd=g`EWrOJjG33ap%gDuvt6ZJ}l|0|O5R?M(I+
zquygRLy=}qRTLh8Fs7gKB~S@NrMI(uL%w*=D>Es=5jlD8pc~c!v3C8r*3#iHk`oa6
za1vPJhI+0PsLf#U-?V*Io&l%<DPhV$Pg$X$Wb~42H#5$|M!x<OdmZB1y67)XNb{IM
z-%xFi&wc6&%u@A0r;^NMhTuMaB$6UhB1KV6&f-!|5)-;ck=_p)8qQ?J43NGAN<HV|
z9kHl7oXO%$a~3R4v8P#%DnEXWlrCrzDJNoN4k@BZa!M;nxAAHah7heUX3a2A!rBjZ
zn=ZTaOLXv+hJDszy}p8#qLAZ?)u9ezt->RkoZ(cHDR9H)2emg9EomhF?ePml2bw|V
zO=DE4!LZ}7l>agpidE7Pae6eV^|x=*%@x$_dYLvTX;JO#?A7uP$I&Urs(@m0S!wMm
zQFwzxmj;mwaTbA*<?WAi?f0s2u!HbdOMy!R0QW(-MZl%`pban=6sQi0{->pT0_DlV
zQ4cX|b&e=u0e7(6C)o{2#c!rmhjH~%VVgVqas<w*Byrkl1r7n_$j0R3<iIUH8nKnB
z%EN!aEaN$a+AwY955a1Jiqrox)F(h__aqU-)e2#7^dCi%(s3T~qf%kjrWEh17j0mG
z;wF@gKv*OU$1%sv-f~<zZ))BqANv4Y1TjK$pG6H@7*Gthk7pT+Gq{wop69`hSHkB`
zTISmfqIXZp+Sy;(9W!s5ZqsGMA_F<2^WdbAF6rF-1vzSnMYc==Y3@0`egktU#WduK
zKF<)5DIzH$UT{lX1b>z<zHl;8%Ox*$35?!3edjPI-#Q%K_Hw$0cA+;~GWgQvC*>Ni
z_qw}q?WJPGrJvtG#yw~Se7rRPvx_TN`G)QsO`vz7EHFrPP6>7lYYAQ@GChYIDoAiy
z?er9H7%1<i+u59qY^TjnjLl34u6`9=y`l8Hd9_U`Z~l4jO_}2OfgP{nd2r-UP#7DQ
zrR?f12)M_s;33G7rxDDa>nMc%!52I{CWQfOM~wU(F0yfGnYdyTqG=fc@puY#2U_F~
z)@<@Eg7-ZrOY*?*2NN%eQQ_s8qCSr6Fx?{tlN|8u*Rf<ZWIu_pS>yJ5kjr>{K!hgu
za8J{TWASqT80Lu3ZHYHuaE4O`;`M?x{rjz8XojnXHmkm?ME>p^U0&*ztY*Wpn?Jm=
z#k=-2k}~)}*hCrnyv7k_o$oP=@c|1iYLZ}lj4A$iuwNMV(Wvr=kD`x6VQy}TU`-zB
ztc-=n2NldoofVD%v3mcCO$CuGI&G`1rs^H?gG#2T(6C{`+LRhC^qJpOU{)h-Vqprd
z^NEA<;|GHwiU6s3US5>iQplU9a`NuFoRAD65AWyYVco!q=+#b=o13>k?51kCH|p;j
zYWu$oDx?U1$53xGT$tE+-8h{@7uCV}5Jr|%4)zIly6y=_^tYehf6>>Ru_He>wOVXC
z_wMqUvjFl$gcRZtf|nZdTPsO-`2p#N;W+U?1hSWl?6n1V1TEmKHkkt_*px^&O+$0g
zo6r8WDV~xy5^@hZxI;cI(Pmf1nYJ356p+cy+dO+P*Lgdt4VKq3+`;BEK?P#@`|Boo
zCR^IB$9!nUFBv!m8o)O=aV0wHwUrCJ{3x-Bbqm3sGwZGjm>!oMkuLxas|c8026n<c
zlN0VG2qD!c%o<Iu(Qu-#7zqwf+L+-oT+2*-74glW<>pN#b3>a8eErtzl7^}{3E(J=
zlNvTEw5q&c+P5I&Vp?$pA~(CKSNs;B6h63|fs_nwt5SP@QSMUsM9R(5!fQ9}dtv9%
z<4niS8I-pXpem7_&p9P4TAHlNWgnFj6q!d}c^?X!)S}s;E$QdugLggH6Q6Z%XEooj
z5;QIUu`pw2{tx&DGYi-M4Jr74|3pt?PdXg6+}G6ZO{)+K0tq4&wR+j1R9EvJiBwV(
zta#>0M1^P~aFKI)s8GK>v(5p9?m3uDTskKsq6i{wpS@U`-C@n`bvE*E=jwDvyT`@T
zh9~U#px==q*$bJ`27EBP5s;q}-28YvrOaD7gcVasYQRxWrStp#T=U=7-p(+P2nvhM
zFo+5bc3$4z)a-V5zU^bY{M--f)%<+R8SxZ8yGSu(LnBY+ZX4O1EpDF*>Fnlyd5!G-
z!Kw3e6vBo*pe0L!96ph;eH7`tXASfFkb%gCQ8WE{*TyM*0MNoJ#pw(*UkS9K7*81{
z6rmiyc=)6%;Cv$z$ZlFcyi-u0!P<NR6ChfQ**}SU9X+{Z>+N;a;&X8I)p1Z75ToRU
zS{6_W8+PO6-CtOvjiL?^<0Phu%ayCEhqX^_H)pn}G?(^q*`^|+1_2c9#)${|1>j14
z{v`NB6Pl{QP$-p33nsuhQLzxaV{F;(jeInN6dLi;Zj=yobXh=Hj^MY?{A!4@;}%wI
zovru?AzsVx#GdW*Lm_CexVr%k4;xTIciuS`Ufy81zQAHb#?t2<G+*+}!a%|*lYg}%
z`T2?}`8UPZ#D=G{;ahtdy~!%hpMYFG$9=*}Ph^WrmOv2)nMb_TZY1?9M##N5)zLj%
z*bAlz_Uro>#>I<CTouB0!~T5PQu)gYU{4fZClJUo8QMYjTX(9bGP%U5IHI5U?EXBg
zf?aD+x>I%gFZ(n)NfJ3~i+~55P=LYY8C8mak>h%q_}o_F03J=ohR?1wzRIfQw;7I)
z7UHGLOWTMo*Kc;ZfIY3{swT(Vd#;<O@lSM2#Ahwtl!`6^LyDjg<oG-%)@JNLH00;D
znbFE8X~nYZIR2%ZZTyPOllfiX`<@M3YKCjw%emPaOHW2rZ!&wOX!0g8CS*c@@FhG{
zvSTIRj8UmhcPf9A{KY|<GUE&sYI^Jgpr3>s&Fprn02_NlY$`drQ%)uy&#WMu>RHeh
z8N`cJA4yW#nx3W`joDf8kwU6abjXexSN`i7Y+v(%YG0pS=P6)F7>J2j#JQ}28luqI
zxQdrtVg^gnU~<q}X;LwPGqj1!yX-R!%$*i0H6GDEz?PmEZb&(8lA^dB7rWI2LTYUj
z+ZZ?7*=lUCE{{!7r2_jM_HStqpM`!Bs#tNXB@r+KGb75NA!U>yL3p^-MsBQDR+I#o
z!6s-opR<k>5tbu?8CqJ8iWUq+h9mjB5+OWSd{cBZ8;z^(@vsL>I*u=2iRm3$`O~NL
z*$IkLRvK`+&>&fs<j8(pj`DfJWga{E9&<&bm+jeXNle~)<9?;Z)#)}y?kxM4LZ+~O
z<u<%%ry|Bs5#btNzf+Q8f?vH`(Kh|*X?!wNg)DS(1msbY^OTXhu|qQCi^l=;44Ft7
z_8d6#daEb-G)PuaL;JxZ;fykLaiQ$1kyr`49(2eX40}$e1(r%+W@nFs`!G0tclKcV
zlNh`z(=-?D)#|ll;WEzrok`v6hK|m=x<JP{P3AzFoK)k{STW(<@UnquH;qomp!obs
zr);35(cb&I+Xg|jaNCCKz)(U4vyIhy$a8TPsUY|io?ST_c#nDR()R-RU+AD>w;>y;
z?_gZwGJO_TGuUDBF)W9sC}HLfl_Sh4ZGi;Ik|*Y{rm8&cDtwm~wdD-#JftnZ;5|z~
zD<mY-AFx4Mn^?{6Z_ib&zAe06OfuRm(F@qEW_n;fdV&Yt<d{IGf-|g*(zWCKYC35~
zm-(rbra#+c8AOg!0H>m~n0vo`Z$=&Xf2>~c$vM_^Hkgd-eRXU2>HaWvSg0a3ba$QJ
z0}Ql&bhI^?PJ~&3{?vWj1DT^$R&l*5)R=@YW8&=A0e!EeQoR!N5|L0}6hH_1G1SrZ
z*18Lq;bM&nY&4mOz>7uHhZ8Ccam>*f+)w=jpZD9Ri2@n2-})s$9jg*i*T!JPLSgQt
z5gMpveW$Y0gq<g!weCl0bKjH4H5cWS#2Ekg+UvU-B2wK7lOZg<5i5{9SnX{lVkiX0
zVX5SJ6K6h9Qj$bGoea#KvSIJtu0u#%xVH>i++SimH56R6k}qo5U3mA$<Y<e<C^fcX
zkNS3m0ziF;zI#{VyfB6SWalK54Od~LJuMJGV92Y@TGgvbM35*DHY#{lTpZ3^=D<lz
z*fGKn?k|yv&;Tj-G-N31F{P#1abLAJa8{ePDNVnLfUUN(!cnI#UN<VC^BK4}&&r*^
zuHdMM&J(0GWM)2Lx9$`*L}vJePCc*v0hN<zAuRp-B!~_)XvrdI&{LfI!%VGt$m(^$
zKN*qj2tI^jDFrsH(w!leCJvBr{)lA8g`ge`5$!-BM>IXRpD&-q*;l#HvXyIHYaXDl
zzkePtB3*z2y)-BL$|9@q3qOH8F(c9{hZXx3BZ<6s`4dYf-T%Wx>2qQKK40-3GpgL@
zvae}d>f9vQg7!-(LX4Ir3$%nYlv$Z<jpdQY+vFdC{@C^9lvZU_yp!9a#6fH#pV~u+
zuSs-n1%?0tCn-&;Zr(@1y<CT1cF=Tkj+<k+mb~}Y6<77(xo##jxF!PFC+AL)l{05a
z+qpa2<OWpE>dS-qo%P1Jw1|SFCY=KMX9l&A<;zGw%!t4ws}`R;ZhtT;RW=tNFsH-W
zXL^QB6c>a!`UnbEp7y40`QbhL`&Swb=?5!)IsT`l&_APN^`knAA9BquI=k@7Gdv{T
z_;;hJi%A>JK;|;s=oY)dr6T-u$u9x`0CLIfHJXnD!5fuBAeZ>&(xr1Lltn@=VRH)7
z)piO=94B**69sxW!l%UOG;7U(nK$y!F(=4{-C&c%JJ8JuK`CWX4LR#+f&mYLM04X)
z+tlTb(2Mg^8@|jg=ttrSLXWw4GA9xs|NNG=z$9yb9NJUidN>|8J)ba9q1)=Rf>ESD
zQx2%DA$n8Sa=a8qvJa+nLau9HR-V-2Fy$^_8HHB+elLy)Qg8j^+7GTqZPwyvu)hcu
zfCl4kae5z@6C$N-m7wwiL1FgC0hIouSvC62x!=a-A=U-XGywb?r*4o-n_KSg3K3?c
z#JFuCA5nD`4WXksaHd*`oICwI`&)7#(@!q4Km{dzSQjk-tJ9`(phi3Z9IupdEopN-
z9K8I=7&0rfKJL{!!dnRBiywYeHt#>CR?Po3wPIubUt@u`aWri=$KJMk17a6eaAv*r
zGeM~0iq`cPn^)wgA%N+W*@45AB&b`ouD+jr92E}clFTM|(9R+y;ZHknu@8^9`^P&I
zd-bxiy2;i^A+RL@4ovcCGGM7tj`7;sWVvEsimW+M;o_J-ZY((p{^F;~KUqC*0IRdJ
zr@v3Lw)8tPyFZst8u#<VAHF`ne0yHcujovzyMzd3;NqOklD8cKKnbQFud4qg_l`jc
z(#;k5XD9?mvD2k&?2VT~_hWH(jT<6NT4kr4Ggme;|G54+tt<#-0YRhO>xZ^qFbYUg
z7#~KF=w<*yPA%Lp2uXG3BKJu0o>b4As3}u)Nu^(nylQH+kH^&fBPO8!R;3$Utog_7
zZnEyDW$Ka}_Gn?^QL^5OWrI<Y^B`k5YX=MpDOi7ql?X$G1eh{lM(>R-D2!cct^{d1
zrSo?O2y`>s)^Qv}yFgTBT6<pBW+ogfFf|Wl_J{<|I?c9E#UB#z9#r6Cxkwki$u4c&
zA>i1}_-6}JtCL@fa}Hp4^|`SSy?J5Mc!4rKfc`6`ST)BFR_j=x(keN`dqFTrPnWtK
zJd+@a2&`+yT=kq+!SZlb-`csPx)#fI^%m!>UB;4bGlSH<=^Dh)_L90yzgv-|RCApZ
z*OfrR_M+isqTPHYVqzXcJp?3GBv}!>d}<4U@q)rCBIIb`S`<&6Au!DO!gcCQEsFY#
z>|;$u^|IOQWT7CO*i+0~)n$`(%+*lNP_y_Io{c`wDw+6zY%Ln)`AAF``)3InR0|UI
z(WU;LxJ&WYIC$*B-+as?gq<~*m0i!3^z9qV?%b7MTHc8vvF7%-sWlDMz4V_2t~tUL
z`dNhV_&D|~NIEk5jJj|Q$FIy`cFD5uS7clZA`ofI4&4<EG+I?dX3CnC{g%q2`qe$-
zu)Rmp89+_=)UL}@bjPCI519*f4b+#J%TE;&nBOlXOF)uhf}G-?*@$I=Fe3JlBVuKR
z=@`17DOYj$-~yfAnNQq~H{8H03kM#JcFQOc&%h@2A%4oyxw4*Xi4xYmxgK6z6AA?E
zFk)aFS1LU=UqfQ>%kl-SWY1o_`vs{?W$R)#<$Q1eU{OR1gunvft`K@0iZojoJ#w6P
zh*xHm2kl9}IQ6C@<bc{^%u9Yfh$)G_%w|L6GxphMR1%noAlB}s5>RNQ0=)_>)>il>
z6eaD$zj(KpUpu}Vg(CUp1O7qm`Ot0M<E9)yRpIg%cTV#l@~a;GENF-BY9STBY#*=s
zCbvp(J`URaO*t^w5Z=aox^%2D)S2>^0og)YYgIch$B<S6AqtpYVx0Pr<49nta9}Q2
zjBRRj#bsC)V|!Ri*-40$vY|?RXgFHe8~UZ2_?S5b%S@$nK&d|Uiukk)>O0im!YOKR
z2Rpu0qaIDrc#js!AMiC^f9l|XZof8cHTI|{UGO~Sc+ANtOwa-zE;T=!pRe2TySGPj
zKDe5757wyu0cJ8jQglHJjhJD+s<p;oUt5}tAibbo)%LO5)=F(Hq+z|n+i=28vsv?;
zr!?Sr_4Uoxb*}!RWFagApY^umZSbzK>-q0M!^LJX@=LU91TJLHZnYT`l3u!)DTO6f
z#1ziXh_3h3M+uxL<6%4l9E-;l*jE>>8s1vSlq4(<TlxqqP(nzXqGV4e-xUO9lhDZ5
zwr9)<xX%|{7<#&?RH)Z%48y;yip4{n0fhMel~51iO26l*fcsj~<dv;>74?U;P=gDh
zs-On^5mXJn^*Z-|riaH@6Lmg>{0-&`{iDX5n6+}e-lGy}USE|#obfM@R;+b|fI&*E
zYC9x+__z?o36X1M7%4wIO39>~&H$Q%u3I_&cDcul#8K>-V}V7rS7#gdyJ}6(E|w@I
zJaKA@r0$dj@=rA}c=_Z4G@C}^Huun?899=;cf7{5!dmWzMtxW1Rdh*q0j^g4GLsl!
z3DGgu^T>YOy#_E61EJo}sS0{lI02dngU|Gm4aHE_X?+*@Pp-ru)OLs_MpMrec$>GJ
zf%#i$x{IxLL{iSiZ#AOehI4|jmk6Goe^3x+v@;<|D9%_Lc#AYbxc)|FUa%(3(sBa=
z8s3}}eWfwy4C#bqt&W%Z8d*;nw}+B?9*Gt~xwFQMGVf#Jp5r2VSldn2U>~Y1kYS<P
zgN9z|A$cd|{DR7LS3US+zbxXHA!F}I-D2#p*o<{|rLX4eNk8jis5UeWYUPF^FN!3m
z<7C68U97pT1i@QQ5pJqh6lzh;05%J@kcZ)RCoz;HR$st?tIasII#O{DU3Hp<#ZoUk
zvkSJViVCBh{6H?aa-=J)e(f7%{3h-z6yrIH=M}z_M~Ao^KhTXsn(f4K@G}K#Z*do9
zclFqn_B=-EdrCGXPaGkm=rxn{P?PmJlQh<6-l_?goYPfqb>zJO8>?^x9WbRe_7^n!
zK|XlImJa&GVt#`5CS@zNH(Ye2HNeG8W0%~scSa8n3boB%8m_*+?%qT*oXa{sBX%(=
zOm+McHC3ZAg6M|?d1fd1m_r@HG(t#DKd30pqN^;)fL36_HGP>#1Q>t2)Fo?2(1n~@
zo#ri0&=1%X!lE2G*aAhMF5G7pFe1WlhKrfx)w9}IvGru|4f^`d2(2m`11GmG%32T|
zIPwQyLQSrY`H3Z_@J{U?Yu<RQ*x%n+bx`KGtd>;2>O^RAJa<lVE}z@C*VDx{tWr{+
z|MvB4vH*QJu{+q#UD8&*PTVVcL_RSx<w4nj*<wHcn5QG|4l-XdZ-w+UDGIe+&BSe@
z_`JK}7=K>nx{glbl(`shPaOL`*WDWuhwv0a5<|@2=>2J0E@^Qd_&Fo`_A~c1p_B=|
z+>J$m|3y+U*;ID*mcb{t4e>z~0=Ja1ecZ2%($E>NX1~V2FgVj&r22JuXudE0b{~q6
zeEw4SK*rDm03_qx>f#+l`@RkOe4Nz8T}iVA#y7v==w7XNvbZ`u`;6V-dCU;wkJ9|C
z<y3As9`JaYov->rX?xLGejFG6aJdU5I0F?`YDo8&*G_IYF<34lSeouev#l6b*Lx2c
zYBXB4$?~AZ0mr>J9{7I5@+}J<!Ul#djDmSnnjp_JOJi>T@H*gkYU~z`ih~h7>nc03
z&^uvha!sy0J;F)|?=ztnLAye=zaIxT{H!p}e(|#g4Jb7!9ZN?}JmcJO>HJx8{*v!#
zqVCoGJ;<3rQePPyZQamd#f_Qkfw<}lQ!hG{i?%zttlGjk22^+>m4~-6PRcrmT;a*E
zK#-5u?CJoUS@S1(FEnNHW3@NRHQsw;;f=i;uh?}0$}q|ay%$#{WH?75?WCUQ%folY
zc*}jj?^}<gFl+ihn#KR(L@=|n|6c{p|9h-gt*PmF*oN%$Qae|e)xZhX$5LKIo$luP
zoOTs6YMWS1Clx@uqEsYSMC+;acKhaMqX0%H?Rhj@B$&tW_uiQ%i&th~d{zCYt=HpG
zwTRq3j)FOTtCpNPeMk`{NRhW)PwuyD<7MtmulHjz)MGz+JV20`IIhlfr|07%_NCQz
zl?6Lg8CiSwfHZ4o+sE7W<NlC?|1)xB_vdbQ^OhgSLY>GFF_1Ueo)??n6)|!3_x9J0
za_ftFZY|ha2r82zHB@k#NI8s1jvPiNO=C?0)<>gcl#xZlV&d|0=q`;>ZN7p=D%bZ=
z1cEz_s~057%9Q@DNV_zWA`&0~?laG_n})l)D~HU0^Q(*DqUK6!+v8-2F5F8%6bd>V
z#Z~$zf4WzwU|c4MIgYqL5)MtI5e)ySz|sX@y2|JMh9beDrt^T|gr>*sbP96gLEU5P
zOS@@%7VuT(T3u4iKVPWDLyW{4a^}zM0uK&-QN4v74Pb4`ln}|YbFI41<PdS5KU=QQ
zPU^V%^T*u6<99aOo=g!lQeAlIjIU-F{{d!U-e3%`OHaB?0Vm7@5^x)EXDpz54ozsg
z!)iJ+&F3m2W2L3jGF~<_RHr^4{FJ6k*>@*3Rup_W^y8^P+X<lD+SFEH$@V@v$mmMk
zk9)xSeC$zzT6;;fr%_kvT2_`Shc5fjJRH)z$!=Y0y5bLZ!-2C#0`jAc6s`kpV9G32
zGTCtLa_nV`#zb?jh=@&2@fV31ftkw7V;Gt*<TDt>pXd~?gKJWZYz_O3;v{uVzwy2Y
z8Mm^meuK+L62-XQx1oE8J{pwhcnpp1zHri?Z(S3A(quMw?f%_%<z8b-R*`?PRalvU
zZFj-oYirk>y-hhdDbUywn`O|=f*EAjgq1I8J0oTNzW#YXMops58u+GTvGYNCzl$)(
z<_RID&iuT$B|kCqE1;Z+z?oEsm2Nv-$KOxm3|C5D@Hd)jQJsNE0ehufCJXgwa)h~&
z=ny{{LqNlEm3}5V9bQC5L2^ygObi<g7;QFJsro8$?g4=aQ;_G<Uu@*Y>NFWGpZT)Z
zu;%lxjMXt#(aj~mmp}=@8On&@K)*%{J;r&MF5B|MtdEcQL8KMPlmcNJPX54$XMh58
zC>R$`iLzz{Yc?t_3hiWENafc^Nfs@KRL(dj2s-y9YVI5)0!UIN>)4cOR^M_FDmZIG
zoZSzCu=B<XfWX{6FX_P?(-<VhdbK(;w1l1-<+E;OrmQKSCtt}$kc`aHFM7=M!%0`A
zxDGRToktkDry?i3GM+d1=$Da!OIHMMkU7#B##DNt^CGq_w6gX@PqP6m;uWPC)sRvl
zKyp8RJ;(wR)}xETrI$7K9}c>Mk_*!bo{uBid4bZ%lil<3B*e{|J`j6z$jE>{w?(!#
zPYF2+<t^0Yx<$k8`C=uhvH^Ar7>Cr0JzlV4#`uym41nk!K1+!tlT@bhI!~-87C9!i
zj{<xtI%wwqSA10nr5lZI!El%;n;(m|JB(N=kn%;MHga@lAmiweI;Pk~QDp<*Bir*<
zQ1c0~jg}EV)ixsFR|?BKNE3h+fjIlCFa%UI<^YTxEhK5Zi}gDz_!mR`bPXDVyO}WX
z?hblp!$q8Nbt1s<y{DW<c6SG(GeQz=S-4S{XmhRv{2D>0On)2}{nNIJb!r2N(SMO9
z9x^?SRA29vGLWZqlm>G;?TB@keQv|L>Ls(?vlRd_8v*8a7Vhk^8H2aEx3}hM<=g_N
zkr@KQM8f;4609>C1^RQZey!i{X0QY(&$GZa(?hIt-E!)}y*Az)g*5$v?3I5RF`Z;%
z4#BfY8e>ws);&d<@CF4a-xI&W1%snPd(CE$<ns+ZjZf;aOiG}fYOws&x_uAEA65sm
z`V?8*PhpD21sAi$jZPh31hMp^<7#8abVd_z04U6(fmbh1j~v@j<ux>Bc}+q`6Y53i
z4*!Momk~z{@#qG`Ok6%q(NS?F-L3D-H(j?K?ZLD_&D5)3({@2)9(0AtEZ$PVI^*+p
zQnkMZeFKm%CI&N>CPpM5;qhZLk|ECDVRyeF<1C^YuAJQhbm3Al+Ua=Qix{Ly!;6d_
zL|0*!tiGO;2RDXm|1w;7${yYpceDpUbtK;Pl@{|%T%&yeYXdPfsAXNHZ~fl|;Pj<6
zyK<UAUi2XI`jKgn9R(N*cst(}Pa*Un31<ApTkXCg)W3<;r~~*+hY>v(Ki5KC1(%7M
zM~mo2bYw^raJRR_oBtGQCh`^Q6LA|80u0)4_hKDFp4yt$IB^4p&-HAFgz(`yIS$qv
zO7fK=R*<MC<B|+HYSRI=t3JBf)?g%rM2<Sb2am93oGnJ^w_)jf&}?5NBqfmn?aHJX
ztz6Qsj0$gc^fceShqu>0JLxeNMN?utN;ENfJ=-`QnP`Fo$r_j~mXSs?-+z685rP%t
z$Oj4^ow(tjtfs4IrIK7tYGnJ`M8Iku5yDbQe!ILeowP-jN~79tme5UmZ_nXa9inNl
zvnEQex!0+<;qAm<wLt=Mq37E~XtT0Y?r4XQUQCJZ0qRVL5XXyN#Vf{vBUBN^z9$1i
z;A`9nBtK)HAems(ePNScyKiEooB4uFE<7ZDLDwRj1!`HC0%9gM8L=}88LWZ|Y>+Ab
z3qmqJwl*zbU|Z=Iem7rL+SH1IX-B=1^vMSVS5LD|J)3aUQDe$Qpf#TF=CI1})Whdk
zQhr&==ggPo2iXOHbJGdHHBqFI&VB$=)JbJ~coSi)eXmK6#Ym+jd%7fm$03J4%Kx}D
z9cs|-z3~3oMP+=-Y?Wpm0k*Be#DAJ*vCe=CgXukM&Omi~aoptgWIG?5f6;BVdslNE
zqB!v%`VjLYWmF?%WI|TA`jcdN3SkA}xM$1Z%f{k1uD6=utIA3fG4dOfPa>dRIt?bC
zX`&!uI}nvgqW>%?fYlW9qi4?^yMt#tHhNyIu>_MYD6lUM<HD7fKqHkg4c)2UWXaex
zf<o5#1ez73d&s1*wu7j!$ux;uEJu{G9_B+U@)z1+JzUV}K)LV&q>k#>M6F%ib)V}?
z^`H^n;U5bOwb{TvcOn5hkd$Ld!mX=k9mhH#8dWhhGScyk!xX1Z0rNVU$1GDKQ}B^|
zMAt@jgMa)3L>b#be0Js}1Di~!xqP^AR18i7e+vR#EY4#S3=fO5+;T$cb>&Im9Byuy
z?v8bSiayM|zRY@xF_s0A?ZZmg$A>9ax&a;w&<{><N%>)dSRR=KDeHuQbxQf|zoG7!
z=@V1`QXu?&-881d^j~QdBNj5sbkUUI^v)5n8lVOZ)3V<%(?*^=MNc~r_=;GS&X%ia
z#Jq8<@q9a-xZw&Bt!<)47!KCSe;Uj^2)UZ!@%769)ShpoK<1UUXU#{&-kq-SwO4`x
zP7>1T&zYrU3tuRHt-|sPTuDUh+zDu<35oTd6-L#;fo1(xe9#WVT#v(e8h;eOBVvzA
zWpUaGU}PYw*!{e0?$dP<I$nzpTv18yY6P>6s#0DudH8K_Yri42{Pu|cqkCojuV`!z
zj{mi?ds@TV_Mfu*J4Y|y*0CjRKe7zd3AfJBx<22se&{0!1KgU1ok{cIZ^8_2%We-z
zY!jKP+yblU%W#3*(e~is{SDWmKKk6=@c82T_?x5HA|aXiP@%pHUKnMZfruGp;5u_#
z1|v$sVtLfr(ZSQ=F1=}~GX43{@?|c4?*3isD*ta!clURf=f}541t&Jie}SU?ajHof
zR1`~^2qoW-Nu#8aBw0&Us(}`l!4r)*qqokDBZ`ePb_y-i1zX17o;inRnh81SkEc|y
zSS001Tlxk1a!5pOrjC3X>dInNSIt?i&B97a9WEj4$IO$L@!iZq<?B@NSXY@M`fO3C
z#F{8m?D}RV_MziNHY)Wh%T&_j(j?&&7M;PXZb}vZ!scbe9UXOy8J8!nBCG(WCeGE9
zAjcge?QP^@L2^gd@9<d5n$3c*Sz<b$_?>+3Wr)8!9S8T6CO#sQKwH!}IyA81P2gXo
z`$W)?<%BstoT!N*Br2mP?7NnnmCxNb`*evwNXE%$t1&0w<=Vj1eN+$KGF6}0wOHTD
zc|avCNWi^AP@S&nSS<rbd(y}@HkBR2s3H>+SkL!@w#ldlM7-V*T1L2`8p&X)3V|#0
z{F<v2%3vK&U;`r^BOuQUBj;!O$wLE!0kFpqk8l(f;9$JJYBw@7Z=dZt-Gs%Af067Z
zx9o)Op?}{5h5BJ&A-WXZ*{?TER+*K!9Yu}D6B^W$+K+vtWW*Dm6B|L3Aj{m);bhh8
z?;E>hL`a4UU?NT7cfvbioezMj4r9wX;-d07W7*@i-X+~hc$s0^wz!m=)T<3oaIYC>
zI>b5GO!6}928!#f5SmXF!-f2lE&uv(x<Bx*%$b(^=3KuO7q|$5JSBxc)UW4aazK_$
z%LSv#`j<$gilLqR`K}O0Poy*eHpaE(ObkwcAxW=|zE2B>!s&i!s3j^yE|E5rW=3SO
zOvuDC_7w`w8z9|Mi|)(wG659@1scWNw4JDp#m9SvlhXioC#Y(WCBT{A@#r{45Vs6{
zZu;?IWdoJ;tg4QE6zc<??sYl0EAGB)?ad%iCWz3vXc?aSliet7zUPB-+MIJV5Ktt%
zv|AV(@f_aGy+}!_H1hQiRJ`YyAL~x+UFyCJqG|-*HEkJyivMB+Zw6dM#vYc57x0Nm
zz*aKiI*m(hKA)4Xa=h6*u+Q<DW*Uv^6&=gJz4WJ+9R`Et)Idz~S2WM7a?KSXwEVty
zS-xhL$Uu7fTTsj2<Wc47hYbKXg>DLNDwed?D2^Uh2LgbI6n~^!JklkK5o(|~cfYZk
zuFsb8`1c|8%Rd#`X8Kmrcn~QPvcy>sg_A?pTsN(H+~d<gI#pJ;wdzdsp*#BPfa>18
zk_#pIFJ`*qm`?rT(sSb)SXFEr0n}M}OXY$nlICL)mkehn-uj9@{rFVpl+NI@mn~KI
z*9?L<+-|w?m3t^y{GENaFF!HA&pr+KFGIZVVKNFPG~~|E=nW#lES{OQmGuSY^fj=^
z(%8zXG3w;YUS@IKPPFCphDqBwWJ0>gt>~u0Do<UQNMG@JXD!+czS3amPUL|x;jp76
zrqC4u6y^rD4PFilO9%(Vaj1FebtSE9Hkf7}J%k-U8cLMIAcRgy!%f1!Q|UCf>ftvP
zNJ^4tgE++N(}u)1VDccCUb)4)cM3767QO16ro{PGFGyL@RL^M!G^F97X)hrVJhFb>
z!;n)6+rW@%i4p@n4!14@p%yic+gA#1g%9kVxvk)o;1(`Zo1={3NEN6G**nZIsnMc_
z12(Oc%0eV>{=+<7gv1j1R&`T<$%vUbMASX$b6kW{D1yjioH{zsH_#Ft(|SP?ZfpIf
zvF>g-DK2+@W>`N@9yCm!^Lm5idrF!0Ft1C3mkg&6&zP*>BNRQN@I4z%>3!c<Cc37i
zce67=)Gq+O;ge@4GHl~x>p4odG3Pkud~PtVl<Sb!YrF6$XZ<~8@|VpUCen^o#=ea9
zrGlYp`sOlufML2HKRl7u^?f&`V9lQK5UTH5=rc{2)nKQDnCX?0_5nv`Q0<aPh|R@e
zs!Zd}5lNj`qj&t?yb<{|_obiM9+IP={$rdbJldd0vL2^RgZHVl2YCygn-;zuQ?nxR
z*d<s(qOUexLhmKrz2(GMJBhS7qzdg>B98QKP|$1j+8u7gPvmz<_@D~vW?W@2+hBE!
zfkJ#uuyu(cy_2ySe5$#P>*vRO6LAphQyyY$!2sw{&jR@yxYi1#OgI!<@5^sLf;vuj
z2|27z5E5UcU+*KKqd!c|C2w7w-Ck`pgR@)4vyBy$o4l*HGE^<q&)(cuBs$G*T{RxB
zK^?CISGsqq|4}^u1+M?k<TtMWJ&)#EL;C+DzxDi|<hNFg01kgA53@f!))caT#7oK3
z;mGDOa3fSiCR#SkeLXP|mB2(bbBfFz&qRy*hCwv0UampKSCM0Ka(L6TcsurJ7bK*q
z3&&N5j7i%>(jv{6V|AmUetWaL-Oq#fcaVrC2_@Lg83gxyeP`xUoTio<QQw8-E!M}m
zIk|uTefT_{B+%t`J>>6xz4=_1F1!Z&IA&U4LMOi9UtOKgmro_&zVvc<_zvo8;3P&i
zf<i@sE<oDn&Y+W@;!FfNQa$onHFt7;m<+wSKa_L3^AM0=GewfHK(VimN1@K3-6tlG
zM2nIgE2C_j=vkPj)})whl3TrzR%_osC!e(BS6M3dti<$LNOaOk%H*BhdR1-LwRtF-
znHYel6~I#@>D}Ni$CTp-M-NXeS39TmjAoKOJy|JSd1XGJ77K7}J$FBGn!fPzl+q_|
zI5;V*U=g(t0e4i+NnX`dHDx)}^g1fh>l9aQmmO&1qvl0@L@IREWO4;PPWIF-7449F
zxO#EA5qYSRNNoOSqdhgh(Y0;#*xxD9jIU~yYuj3CbVz59MEfX3LoqMePRTY0hIDto
z)}46ka<Hwf?_<nbz7y(FQR{NFRnIguf%852o>4Yh7`VoYEzxSFP7bq2k1iKN%d;>r
zSLJ>km=^hGf$F>LkEU$#x!t!+d7-2zIJEjT-!}X@ok#3Xd0EDksqXQ&IT4;0GcUxR
z#xiP8Evl1wZy`Z6h=kUxFliybJ06Q^ypW04Zfo5j%yEL#P((ZnS%iv7!1oD4-0{`y
zR}^rXP4yXb0h&D=Wz@qNS)UJqM>V}v+`_`5krDbzvlQItIuk5@6aAPnk>?5g6cKVL
zW*behu_ZjazE(7p%rDpL6Gqk&J3EA2zdO>Co5iC0p{h2?)#RT;Mt>fkj#K{|y@o%)
zVijVdwke|1MxAXSk<{>>VaTcXe$vd0Nev}V_~ucOSg78e_9T>s2hl?2TN=@D++?Ad
zLC&B$h<A6>z=*MQ@bH~&>y04Jd(ED|5l-+1{6&JA=RA3eLO>tq5tI=8v9@Rhc1yVo
z>pW-niLtbWw#cDC%TTur!-wt44}Q-<Yh+w}2&|SR_<>O0bUpz&=^<$1y9`Q@#P3u?
zEcbosQ$JU41iI<32qUQrU|qCbJL`XZMZ@N~0JEz{x%OLw6u~6|LvX4yos)wG+z?}5
zgF9eJlpIH6ZmQ6y%tK5saqYHE6J`iViidD^A)J&KvRJ3`{Us&_y+I?kIq*@fZ9FeQ
z-Lk=4x7trH?x5ItLI(9Ag#%Rz6VCxE-RAq-twGe&nHGpeir7bZ)DJU&)P~`MwR#2C
z3TYU+a=qg7NUQ;Iljn4qH~w!`90EM3qq*9^n7?fe$AZn7YBjst2tX<V3u4;i6+&WH
z_D;fS0%FW%H>e<ax5NF5ggiRf`q<ATiWf_DAs_=?OQ9d~QDchu*Iy?)BS$lIjYy2i
zC|}j}6DjmlpqNN|<1iGTiZPatFtm$$dko(A6ae{?o@zy;R~$S^I;Ek6_^yJBFGF5C
zI8FLy5f7JD;|XO^A-BU5Jlly#2eF?2l&n9)S{++GOlTiLv=D6BEE>yoPdo07KNNEV
zvcTD21*<9QD!dQv0C|1fo`1jd%aclb-6F)&m!y=5AiEvdQ}ymE?Jb;45}Dfdga9>|
zX=tWCD3mS1`(uDF#yUeu0wE8@d#z^vJk8)byYB0N502a^y8%;_G8DjeOAF*r<lQ@N
zlfX-7ksF`#<%J<7%-{j2nGZ_VT$$iNo_?nIKI{RT7f8<f`++jWs&n95!33-Vy#${c
znpe@17!WVL>ysDcnl+|D@EM9k0N76nh>_bpm+ATkV_esGeSP?9mOYtr9e1ux4C+_9
zfjnqg#*EUwxuu&LlFUPdz~^@@ue1N+*JcB+8-gnU%8y%n)3*Aka61h8wy^u>LfkAV
z8pYRn_-*Hp)3~E<Lk5N6_oJWB&p*Ppd3lbSQ^KPik7ZbDSI6C}1zZXUzE(wAytHG$
zmm@tx>eusJY0QzybbD_><rPjtL0r^Fh^xYX^NlM$U7xaKm#yrOhXCs%+B0|c20N5w
zx^)z6n7IHaZN;v0?{PL#&6aum`=}2xW)NhUOM!}1<A3+|BV!i@J!Rf_a1sc$uSWrn
zy3qMLfL1d=A;933i+Rv&i<mBPy(yghQ76-4<8?v8vy}Y1WU~rc?AMw>UO>kJ{5HAl
z?P>`)X0aZ5w{%J+FQ4NBO{QP+VwH;?BzG^Lhr$NP$W0HG9gk$c)nP&u?oY2@70I~w
z@p`#Dq*guXvPHL!bNUg+TCrT_i01JkGsa(DSuKzn$0^E#`r9pCxD<Gp{a0XFtr;z=
zEj&v8wfz#hxd?EG7kD}$BYmQw;95GLZMqJcyr!oyCeaS6-rEouX3E?vMcGHS`R*O)
zwPLNrdO62PhkbMlY*r0J&V=wkL(mw<RH`JNp~4iV`j&I=;ge5;#nce9M9jYxj8HTd
z<0b^wd(<E83!Nx92x?lX%u{y|Zld>EK%<3lp5&pVd_kS_($4;?$p4r3nd=|@=zr^d
z*4+HZ0Y~zFQNOEpvMg*%(C6_sqp&BJ&1i{vKk{H^CDY8x7f~%j@#Ma4ec2O=_}6Wz
z@^tf(0ukxE)@x+(!7iS$7tBW{+LBY`>M>@BVvnQbh$Om`CP<72DU=tCvvx#aOOBnl
zOPPIc%h$>)W6SI2^nQO2PmZ5+sk8QYeIGo$ZMHJ>`Z{{>y*)p|QiU|2k*O3Z9E&iq
zoCbLuBa@}e%@rdLsw5cGD#@UdH<~pJeU2NNB;B>G(YLDZq}+bFE}m4H_izEYb2{Qx
zyERXe&m-ObuwR#St+xQW9eQV4D+hcuQL$+^*2lVldC?gu;OW}d8q4#SwWW#O+e)mm
z7984ZOeWVWvfXhflN)d8s$5CjjgG{l6iT@H`@U3=qgSLXs}YNEZv6rAA(G)(H$;ab
zCA?^Ng9TfS;rUWxY?4^8srysls^3s~KJd1s_TI`ZH=EqPE|%h^vWMEy_W-_Z)t=1V
zOgWTYH*9d9%#NBTaG!^szOiX<1}v1-Qren_0_|;X#Op2+7D2RA;aftu%6E)XU5uMa
zoO{W`-y{11vae0g##^VM4LWVQ3^uyzs$akx;gK2%#oFa&TQ)X@qdpy8_D2<`iHUl0
z?_V%ksF8MMB9FAxDHJB!q%e=pL!`*m(d><8J`acWVh|EYBFXkH83`)nH}5-o)iv_7
zs8WZ!I*u%xRTtK($^6Qq$fhOZl~2U3{JtyG7GHg(HSK}FWB@0-slLO#JAXs-ry=tU
zw@OJBg*xNjaQv&CcxN+L{x)#x)mk&Ix*jvT3L26(HGiy+1Yu(Z#J$}JQnC>mi*X|{
zPY#qd;^YDe5fepKN=V6Zrp6l0>Bsh6$V}LU2NUvz5MC2ORrk7V2qFDB5!QR8*oQmq
zlVWN>1-_ekZRp{#V&q^V6KPx9Wnl?^4ZM9%{SHztP{I&uDG4eyxdozU2!o_L@k9TU
z$U2{oo;U+-AoD87Z%k+xCHadxEwLU{tQ&L$A<=6Uq_6-1Yu=^r^#_O~19@h_7dR>}
z0QyaY@uB{e!C-+N2pKafUWUqMTsTO9c*`z0M;kor>>^8X?Qy%sV$v~j*(+g#u*i!O
zuS@4HQG&UkY4%VJNdZVn>FCy4Yf1Wt2~wNQ$IIpqpG+avcIF|I?op59A(``i<V};)
zVq{B)D@k##6(eMokfAH)Qwhm$R&1;qr7hFkjmc!-HA5E@GeHcF&8H3<+W5H>3(n0g
z+xDNku;OPRn!qz!l{<@G63;-51eE?kHpYA^Tk2ZO9WYrY1nVsz-V5%=E)7m)Cev=~
z?XwRr<r9-9VBvTknW<2VTUQzv$}*_f8SMX$vVZE%gzLI+(b%?m$F}VXD^A6>ZQHhO
z+qNsVZCAAN?bi0udJf+8CuSdG%-*japWcgH8GAVzB$g7DrDvWK(_53Vo)bpg8VDMX
z5`TeuvI_<zxB5zA`#HMHDSF(181fJZNdj<agj3=546<>x^jrF6x|i`_9{yu}T)g)^
z^%kJf)4njcWct)%%)>%7-s0Ogo!~z|^H0|b$gh}RzJw37Cl4p$FEIuN?JhhHl|`4)
z-{4Mib&`aWPZMFhr9!eI38;^GCDnkBGfR37+6Bb(yCU3YPV1|(%qI9-kU?Q;GqqL7
zx5v=wJ1E&>MWN)!8!%#0QYHE{2;K2{#<!f`J5t7hxU#Ey9ud3R<wM@D61YRREXr!w
z{&E#uRY)CtPty9^*dG3xOLZg$9dJaEG94I{!!by^{lVejRuLZHDnXScJ@~w|lQ>A_
zz&7%`*>4!Sl@hD{(LDzgJb;%$U6{2joiD_E`<V0>CJJcGxCS9uug8mz76(1dSj*U)
z14&vh(@_QfiJ=Ihj`y2W`=b=f4j%DuyrkDlH{6G5@?OI6sM>8U0qs1NJ><#j*jBhp
zPo-@>)~l)c7SkNiZ4G$!Uqa_1(7k*bFPvx+B~=fyJz33-vW|=d@x2-qvEdTXV}{(k
z7ZT6kQ#apBPcgG!p!9ZFzGg3DP`rHS><;2+lV5+&TswO`TMr&|-}3#oZZ-{HIl>_2
zvjbXuAY1%otWcuTjZp{ds_fXjQ+?|&EhZG-b4_d*`op^6nD95eo+(r{yDez-)zk5u
z<6L$KRQL|u@RYao;Tmqv&6WsE%prCB_e<Rl`)R<yp3O6&KOcuPx3Is?YrtNy@IgNV
z|J4g-`)b-wkkoE#v>3Wvkxa@o8wmyBF5;>9&i>|Kdod`Gl;rM^&M*!tAb|_GYmL>}
z?JXpMKNCbLo(nWUR^C0&5LbfBq!#o8_|zq)<jDmXrLd~bYu<unHuq=3nM>j$iXf<J
zfZ-4)bELmK!UU`RS}T)b^TiTemJ>WhZSnLv<4Xe(1;=gJ1^ACT)fvo2F6tP=fwv`@
z*R}W_3-et|V>Go}dJRV^*ddxyZUMc2T1_4~FL$TwzfF8x3v?9VZ^@0Pmp7P2;Y=d<
zswXX30S2fqqOp;!hwagacD>8Dx9Kb1k*TcXa$I5~dE@~BQ1iNJk$gSB!d8)Y8sk^l
z87i--x?Qd+Y&uh`TGWFWusrv#5aVZQ>HkB$f4V&XUn@E@2mAkwqdvm<LB_V-{aATz
z^LbE&fxw{HyJt4pdYY!$R=7Q+nK_ccHmoQWk*PL1LcgA`4TQprsHmok6VD#|$=g@K
zT`(g*0}6K8!vfSW1*Yh8zOg2WjLR{&)3(4AA*q*XBUBhQA#Q}}=>?{~{(VRg6ywv{
zz%BBIaN#ZFetmrIn5wO$D&-f%HBOc<{C4wxI{$ZmwbA?W5wh$1{&ut^Z)@KYImRHM
zp4lh~#0%p_gsvnG7J)l*z?xRAmmJE|*?vggltY>W0b$nc)Rd!ONuGR;%KydOD*IgX
zAsB>$C4e*%GQ8Q8z)PBBXk||fl=WCTYl*}tHE=xhFI84``J&=EcWIg`L@cR}RZ8B$
zD+)hfaNVg7p-dpAz;4;0ar<xyLf2V6xs)@WC=2Ht;f-@T2+Wa*9DilJ;Lkc)_p!|#
zk%2iiVrLc<L75~IXvhiW%~e4;`&(@c+|+T&dX*qxT<q0bmPmn;*Zs9+{8s?f$aIij
z6D*LXQFM23I)(-lUByNh@&6%8Wchd4-H*cYYXvn_35o|n95$0xAnmEsQ3BGsg{FU%
z4?4GutG3i%vp}<$mbGpq(`DjdB~rs*M2z4CN-8NYWeAP|3y*aPSUb8&33m=9CZ*ZA
zjEj)`8c>1Q6hQfBp~?Ez0R%1egW!xaRd5*UpW#A{C+=C<C=djf&Iws+>BmH6M+Z)M
z_C^a|SWpuW6a)7FjCr?=H@feQN`d>uqx;~sLgoS}qpG?&xhk5;(<r1xJOcyUNF>|G
ztlL+$0~&=`RE2{DjtH%3kW;g*=R(q1$?eT%4N<@JIE=v?j+zco1ODw9{RQZG>7n?g
zRSQ+Ol<akMn5(XldLNQmIY2jDL!l-ew_Vslgeto8F;v`d+jMC(k7QUMbG3y{*H~IV
z#%S&;oc=zkbk--NFf3(HiL%6(+578{Gw$hph|ASTZOkk6NB(8<w1}3iVE8AdMB6Xc
zM@kNV5=P$lp2?Xe9wos<+#9R>cczRxazqAzM1c_2=_0<CjqIh@VS1HIK(?TLt>5V!
z86VvM&IDTU*N^7<nZiGeyu-?i;~cu5%Y2*uUNkfCh}B^7y9MJP-hhVOGh9mu*GQI(
zx`9!o`)*~e^6UzB@%;>Ikz4`p;^W+L+MafMHWj4ZtznAUeCu}N9bd2rxaTDJ?j!@I
zuahQ}xu@s{n1QC+S2?4(CI4rdnL65)e5|)ttlSuJJIXnYy-5fUi(Wa$^ur{TJQ0Gm
zngk+S8$)Zk#n_gfYSaUMmo2H~3e3`pM!tlhwP^UhNZem4UbrfNw{m*45lRvj?=&&N
z{c+4lpQT*o(LyOTW#dEPw$ohsAZya*t;qWBe5*50!gQr)=d3&gu{e3cFh6*|(;k5d
zZKe;D`|6>r&bDF?Ja&oSl+EkrCd6n+jjlKG($eMr;&6@{^l#i|n!Xm9I(2O#Eer5h
zav%z%QSQ~%-Q#^i-3g=CgtJaK<)mL}F`BuNZ;#3%HXjTgPS5Bs<L{_m-kXJAF$8H4
z2N?T1kvj#8-bafJWd(ZF0Mzn3a21t+VB<KeCaT)jM&sqzO+9l=aCO8CJ#TWMmlisp
zWs^0Q35xbyq@Clqeea<aC?#h|i$Aos(XSl=Q9ZY2&AEdVQPTg08U>}uyl6WTh<+cP
z)DZX9$fuA(zbeI*kG@>66yWe;fo6=4tH?|MUdsrEGg&H&kh)3O8Yc2o<Kq26MhX8`
zY2_G*_NT{n$=~0z5^{$e$>t6jC#=Wyo8<=a^tEUx(|pHUv#(vFaq#-LG5k}ynfW^c
zaoHvB@2rV}Smi#OWnQgO>TT&_XqgJD{PMPd60wD3#8O}ed?tkVw(e+eib**CBwHLd
zSd~w)#@tm^+2zEzgd<?^HZ|z2(4Ihd2K(txqiTmzby-i+>|^Hy;mnDA%24+eNmj)1
zwjUEB5y2YQ-{(?NLDyL?$StF8ssMdyoUyUdRaGw>A*S2Oc`m0e-_hfO-x^FK0c@G<
z-NXk^<vS$7DZ@FWe?9FETZ;ELVeA%LuA-csdP8j2(-{f)<px<=MJ;KZb*mEe7DF9-
zZ!(*1`BD51ZQLxvDt-Ea#P+#4Kg4-F92lX!R45Kj;bw@N*d|+Y_8fUd3Yj26e8)EJ
zy=>#K)s;wYBP5UeCR0KY+1{iGPh4?W`|_1K9_=0=nWfMFS_(jZB3c<OsP@Z1BaMdD
zcN^>I<_G%hue|;ENCR@oblNpTmq!df3*3Uk3HLS$3bC+{{m$0q07;$~U^Hm8L3FrE
zLldd_mc;J6z|5WYW_*$rqgLDhwhk7RK)XNwRw4cf6V~}s4ZbXNaJ_f6<@fq(y1^sW
zRjJ1b=$7e;HlWZQ$7{KG@*R1y>@DuBO}xk*pW-sAuBbjCDr=7frF(C0oCARa*p42%
z#9ME~j&OwCb#K-jyNKM>{6=uH&NsYBY_F3<SN`kmGDEic8w^}^y)=F9ywZAW?w{kV
z=JH%)od(ZeUgh`a8ZIzfgz^xip2&>OnkodUY<2Fl-Noj(JF-HFUKvO~29?_-mEl{b
ztUgp_*G4)<f#@cNN8pYXe0moV`P{+*&F$4Cb8v(9aboLhd*7n8Q%Rb>Fi6ov)gp)4
zmqo8A8a@Aao8j8Oyp`lV8HF`HMVjz-O#0@w#7FI|42<})1J)Xc^bdjg7tnkA!4cBO
zK*tZWRNIPwS+md)*JcoF%ZR`ziTI7^jdkS8h8)YuMuV6wM^<Cl9>h>B*oSkVu^4qi
zc0;|b%F^S-O>1U|(2Xj1@V`GZ9ba5SsCexmGeUVh?^UB{rco3eLJv1d;*!Md<>g$5
z!R}Y(HBGaKN`BP4-CJ7t+j!DL`Y5zC@89%U*&l(2KKJHL<VRu#n1J~!biWZ<u@M=p
z*=X;kNb7@|R0H#_x6DVqukv2QK#FAjtF3AC7v$mVue#RdbQD!lEFB0|)IaO*h<a1m
zfBcNLY5#l!i>RE9W+e)a@W_!kuY^*_SEXFTG_>|~K0Iy_TBt=1=|8Q{#vZIRA_1)@
z!h=hK<R^+S{z5&gt*|Q;TS#f0yc$61%XZqU$uTqQ9`Db0*E`{lCo(-mQD6w`U&#Ji
zj^{TIb=Buw9C<Xg`(nO>AWiTe1^2UM&bq!ZbG2)}Kl*pwM(`4iFogzTw9T!2NyF~N
zj`%#K9}<>n&~`pEG+Q;3U&Z8A_mOUc09k7D=y&|o9-IILC;l9sr5)&`-KkoS=tFID
z(`jh+i{Z46oxs5ihf{O~OV!hvPyK{DuHj0x?i1eIW3?)YWi4esum|9~ka|cu{M5s;
z54Fshfd2K1kOFK+fca$cm~&d@+HAIrpE09|R?%L~a=`rLZ>?C-<UnUqwu*$SWiQDT
zl)un(m^Pxbuym`zD?EsxqeLg<I8Iuu#N<C$o%1mZ5;^s%f5HWm0@I`b%_s`<*orOx
z7dt)w!Y%)3*2gRt-_YGA*^Vl4kjRkJQ;O&M$d>inx{#($ht!>0p=TI|Yg3Tdg|~4q
zpKa6OWpKVCu!nM67o(1ma>f_4>@2K7%$t|qocmG(4A`3$a|xN=NDAfiD%bso`LDFI
z)xgod#2axNYxDj&O;M;T<3x_fM&Pt2!)L0ejSvAeq$tPv9kk_2#r2aPE!yex#uoUm
z>G(h3h|H`^|MP8YTEp7*fF0%k=l&?&HK5ZK4Hvd2Y}vXrs5Prig{|qcjtsV%ghtl5
zmvYc~p8MV=Hl>V=F>KxF93&AGRXVie^g2%83lrl&j@-`n_V#{0r!y5(28fY2oO+;-
z(<+*$)G3RtOrDBNCGv9RmPZW>WMPu{ct8E?oV=K<eEIhH{_K26SD^hmzV|$B^9t<!
za(|nC;##@esrj=1dU(>+?Up@K?l6E(<f4rkD8_djk<3=Jh2_WQeJC^~k#~yU%X*ot
zz@Czi3U{!vVBy7A4<QSN2&dh*5s)Fr+19Pq|F^PKu=$KmIeFh^t2Wr-)g<6#<#EY6
z?Y)-XfPH(?qM?c(LzybQxT)tF9xJx-2pYX~H)mk0LT~f={#0U3%}}L1|7Wmm)r3cD
zmCmoGaKyYHl?>+ZPjB5^(}j!1(yCL(fTP=bnwZ9l{A;7N*JPAxMatjC0;<+Je9k=T
zB<20Wos9gfl}Crg^p=N$(X&<crlpCAdP`cks%46{=pg7`)MTS933GZ>Vz%pJwy}7a
zAN!WdN>X5ZZQUR>iOQapneT>(0|iZ67^id*#BaNnZ5iHa)Zb+a|Kw$uvWD9;fJkJ6
z<oM2eo*a&pZQ_e~A4K$&E`e&uJrB6>;!*bQKiWpzQx1O#ET-kEOMzx-5Xfb2sYT{r
zc{DA3K<7Tw65xwglL5P~TK_i9wR5qwXy!DMBjqP1|6}(=D|>LU5rOwJ;AZ}IMUB3y
zbT#!4EC-m{jbT5Aqcv|NZRxqxby==M1|W7(t)S!a_1A=2WA!a3K@=cfFD*PqTJNaY
z0Ag)7Acgg$uk{G{Q05`|5;Bh(o1xPBoOlSO+44}I*i<{wvT+E^HXeAR`w39G9{IW@
zZbb~ct07q`H*gK{d8YTp*3>XP14o|jPP%L`U01!#TAxrTX3QD?9X2#bta9cy7yR+c
zxd%@t)AFO~Q&P|2J%YJ!qd6_2e5F#;gHAoj8|+hqHbqV4hI(WLzA_twvIq?+8UydM
z{x;8l9iS0BKAG<5;(vcxGm}YYoYVb-pgM1G)g@>G6LKMGkr!T8S+KOWBcLW0(9TBm
z2s$08DV2rxsP)?qgSS~C!8ifLiF&tNFf*7aH%M_Na-3ELz`L?N_e~6&t4fqn>@v!o
zVIME4Ff^^3DgwEU$!-0y2k$00f|WPHXVL`IIJEa=9#aX%r#5qN1=hoGXNtyJ8**=`
z#_x6X;RXRU$)YG(Al&<bJQhXn8SdYIxR6jwaxWGF>;a_KIu$3q;UER>lw#=COxn?K
zre@vhLT?S(@=8K9AoaZ*$;30hx;M~E2GAa|m0dwhv1N5FG9Fz^Hpw}A1!TshWLWc?
zIukkrUu&d@w2#VSlY^zercPQMQl>f2IJQ!~b{n=is{1s=DkPXP$Xl%9I#gXaXdpkn
zQKFTIAnAhf(rQcZU%WSLt9}lr_M@_Q1|9B817N7c3eb)3fzhfp+Ybc9ar*qqC-_5I
ziHyf1XaVrtccw)`atNV2&#ND>{2dk#$P{nt&$#0NllMv_rtW&7Q=rJAZx)PnCL?)b
zpzfxr!D{;z(9Djc&Zdk%juW+$(<K%u{GNn=2q#M+hm??dBL{LQ5Q8<2=+Ri`N)>P)
zTaR3RXHQ&Ho#2Fjw^LICCuCJP2*Lw1CC6HLVI$D?g>g^!NJ3(00GG|^+U@ROy|>3n
z(J!66GC{sjIy}7>d;=^XzoNpv+RQ93W(le=QD44$J5RQa(HnGtb1Kllr8^+`SSV$G
zKl@Y;QW2ZiQ{82?9T%5((m^g<Zf%Rd@#^TQ{zkawFS@HSc;E2ow5j78QI<`HaC<&p
zoad8FA*bHae4e&fA<K%Gxwo4M#I})Qsb!fHH|^^2W{X;>^GeNv;xvm_)EbS-Pt%bg
zTfD~7`ffO%;xi%phlS~$D7;)LMo@DH!Xmr0k0abcJK&y%iJns~<iFdX8h4*z;}krb
zz^Z-a(Z;+FE91)g2CAB5*2OWWb3Yh~N58^FCSZu?ZN8Hp+yG^&@5a|<D^RNdy~v%V
zCU+siYf}M9NCf5jyyeF*dir}0J!_+yU|uPVL|+4$-VG2Y>x|y+w_!{$^JyU;yI7LY
z7kzfqnxstf;|!O3EN;T5zqAi&maL&OEy}kZlm@Z#Lvhq3H2+Nb1=AlA)ooST^M>lo
zo}9}q_)E_5c&UhgA^XCu6YHZ6-47+MZAAC8fCe%*QQ>Z3g89OtmvwxPM6_6$CFtjx
z-9ejZ>M8!`s4qFnT)vW1Y)%=Q<H`@;$%*8exIy2N*7Fy0nqP9&3Ld3RR1RXfWyyd_
zFP{$OaWDi4?EqINbEft(LvLg!X4#4tEp=x%@8uQvWEo^%oUp1o5|-DMs)kNkJi7`z
zxevn<8ny!CE)~)b3qfFC^vu+P6?i;>tnBQbq9QMWD|Kz<=M{za8@lJqd^@PO4^T=+
zfNMLO7}mq6Au^SPq8AFAB$8=A1uGmORr)Go4x@@S>DD&1wZ5)c-)y;>i7ySM*xFWz
zCs>_ZW`ZV4(}lekL}r+t=)k?@kEJ<rPx=k3TOB06)%xCM?g^BReb>s+Qae3;<yTi9
z*~hQF&{F`@t9#&2k4co@MLG+8z0zlyWw~o$&@kObSqB+L?Cylw4<ga|<g7|q&2Z}M
zdy5PHrT%sUO?ek$U}*dC5{$`D6SNb?DG#IR<%)zX2K+*!M@IEiYWxf{ftQE(z5BWB
zHBiQ5Giz(D=*J5*>}ARfKCXbk0VcJyOgKJd{_1V~2R?xC#rQ>(rqo`{w7GA9O~vaH
zEr@X6$Ra<|KKKkOQ$0>0QUj<TN^`|nlGonq*G`BDX&J?3&YEz;E&rJSf$?2Sxe$0#
z>$;0g>@_%$lSqzA0r#Qfk@ga2EA4)(+c9+w!Q@lPv(4MC_e=m*X6q53a{|20xl7F4
zcpq=jpjb0@n1Q;B--mHO(MW|J%Mq*uHAOBo@tL7euzWN+zb`cd!5)*hTZWL)d_Two
zzQ76l_e`BmjdnFq17pS7X0}2mP4+48-)t>7)*j;3t++v(m|z)VtA<LD-=!&-B#^BK
z{dj<Kw6ez<7+ST=5^AYtc;tw+BU6SfW0-B)fkrMYoqK@nD1(<TKV<?}?Gt#!_YAOP
zTVAO<0p+P)?LOfwo!OouYca`p*N>>6tF_0o{m1Oa@gM!f%<TUkoVyyDKac|SpWNB5
zCY4~uRP-TIBrl+vL$>kC1|QQ~5(Ahn+KL3N9G%KDx#07PkL4uFE#As>5rYyg?J#{K
z-SKTkjJG4&?eTd3>1jCfra8hPrIb0kW+nbtp=fE;1)xJmjL;z8R6zIpZ1^SMBv4}j
zD+LDL^Hr+%`=BznhN0F3HvLw;ZjD6B-OKsy`27C!y;a}~{LJ_34lz}r%e7P6j@!q9
z-L&1_^Ip=8KS*ErWc~>v=Ja}E`Ks{X%7*-YwPwP2<<dk0Wj6jxZ%b)-vOrw8{tt0+
z-_1|#O8d9#>k^wA)A-`v@TuE4d}2A)gTiivgnJ5!vGDkKfk!Z8%X_Ac8ZFPEIeaUl
zePcg^Chbj*fkc0ddC4fxPMnD-iZShYkZwF9sZ3K`6z;<AiD{cEOb_!N)Jkg#R!c9)
z%~HShawGV!eLE6fDJcx+Eeu?b%8A*QpDI6-;SWVu2lebaR?K+zzV79=!R8&IaxBqi
z>duDp$O&Y@%niDTsc@K5R!+s5OPL{OwHicY@5m+>LKAe!RaP2J;$hr`g(jWhjRtaI
z&JWTPhZUC}<bsR7Zrs57U0q9_I0Mv(F#~KinG7{b@CI-49cH8!52@^8n^DSi?t;@y
zfBlqMWo98rsKrcC3`RgYfK;MbJ2(o=bOLJc*cgtLJMHN{iV2P$xMc{8kZ`t?13iHS
z$ZsBS>khtr|APHN)5+82sif(l(OF6a44R)x<@LyH9ltyDU8x;EhA>Sqe!5yGE89?=
z@egC@Vkf&XcV3cOgT7(62R1#D819hyy3Yy6i>?C3)Xl|l_Bte_i|23;;&%V)%$3gP
zQ9Nd6dq59cojk3$GY0%Y*NpC<l}!%Vr`sJBQ`RnKo5$}REXuz!+_w4`14HYDgRV2A
zL@=r4eOI+M0Y32QArzzHE-B}L-n1vuXqF^?0+N;f{NPFGfaWa>g<EHczjNcpWo1F5
z<o5N1$%<UR5RKq()ZGdCNN?#>&g~NaR{S1%+>M<ouyJsLk2hAr<U`IXo=y?Tx@>J?
zWC8!`dwx78jALViG%2w#-B|ZDVMG>EMBKNSkoPk)#xt}dK{%3^h$n+v2%>1U(5d~E
z;r<q5p`m3L{4$$u5lk84%?%H(egK0do`qSqn1WLhn_`w<ZiH>PLEmy(bEX7|n0g`%
z`GQf`2yVndp3QUXP+%)^UjADxqUaMfWJ9Z<%Z#S`h@wdRt}R=ddTVX}5Yf5dzrBMl
zS0UdLXg?r^WIEN?<Jb^W4Ia&jE;|2E9>+ffc4o~x`L0+;`+r@IKRA$J)^XB<Q+MfY
z<Z&!JQKD{>ypt(pke6AVk76brB&8!}zqW3X33km~TOF+uE_wU+wf;6ZLL%k3qBwwv
zY(po)ess0+a97cBY($U?^_3uE-J0mTF-GPmQWs*$QwE^6Rh_kYm<P7Xez>T#OnPsS
z1aJxHKvL8QX0)s(hshiir+xGjN++1P;^Oc8e94It60GF~kL}K2o5V=SgQ5$XtW%&3
zY)Li3UH;WQP3F-y^}_|ibOwSTdhAfg02>mtCcZhnyaMj+=p5L~XGd_!2u*1zC=Z<;
zap`mj&+Vt#T4tTcCFQMVF{Y4bkZs)vM?e{OB8fY4>D6$WCNV+~s%bu`u7b|XEEoP4
z2SXSb$4x=^7#<2liZ4j)<_K}xyi3G7fx=Inoh41r0wf#u&OiK!y7Ly<r#RDS7BBS0
zIl-mP3}Pu0Qw%OGLqPO$ggqvUo#v57G}Es(L4+C--3CitxswGdpHPUVE?v;y)VPLB
zq(?~gN@Wv2Pc;q3HBXZYC#0isNd$k5XVtlRy8SO;7!kku%&rePtDI67U^No74TI;b
zgcql!Oc!rMt4P!o^Q;K^`a}bAAPj#gCw#gv#6I*;qoq7uEf3;^;%U~<3I*Q3;?Lif
z_V?`0z~+=x-p!H@!7F-nFk`&^j1=MKq-IcvW&_=LX~5kercw_Ov~6W4lG6xg!Q7>9
z-&b)wvX2^ke8JRPpZN1>i~M-XiNr^Hj3tYKBqzMhrogOHDhI^jS;2G>qGLleZC|yY
z!Th6PUY=}98eY+XqaQMa!Cb(w>83I;i>1N3{uRci9I%Rq7*Vrw2|{0Gb4)9wb2q^=
zb1Hd0xQ8%t^g;uWtT!~ZFUCuBG2v&(OHlJ;Rx_n|sFJBbXH-;oDv<#LVTE4L5w;2C
zPBFEEe(4J1nB2g-EHRCR^iV?gV?eOwI3-d8K!1SB;AcZ4z0Nb_uSs%y&JmD&H7Z6k
zokFY0R0*$t$;`)A!-_qZLp*C_0aj29LiUqvt8fYPuqqJG-J#EO`LoS=nZZ)-tt;tk
z8GPC~9l*!n)=}c1{3XOEqdBCL$r*P<6{gz?oJ|X#hIM<Br5a?)1c68Tmb3C{v{(zY
zemhDBJbQF=*O&i|_JqafylJCjYY73ElCThieFPeP@Pf_@<Bl~Dqd_Yq=ZVa2%rZ8<
z-Cm!pksRSo<<ayM9uSFuYB~+<g2dwfuEE<-L}Zve(~1~p(Ah1Cs;98+M50H<s{K6i
zHBdXr^iD*A$s=w^I+8-c5hC6yxi@Xr{ZoMU$~l5Bf)4?DWb-!6e>Ruya1s;a@ggHX
z1O-vUqa$M6T#FWQU5;8TlQtulV(XO$^l6}O7vUa`_{GYDk|iA(-lXLyWL8BNx-{8C
zV1iRh7HJn(M0F$Dw`uU#me>`_-M+@`GSyqEQ1VG-MrlD#`&a4cXxE@6{Vj10vQG~4
z9vb`RUQgMxe{fG_T;*K$yuHfnwYJ<C;%u_x13bBw&sM%JGqCEOp9)G{i0W>;Cdeo=
zblvSoGvF}ujR|-})FI#hl*c`=?@dXCHnyC%*;?4$zx*59hDh<K&+F98&p6vqNU8mh
z;wf3f>;(_F{$oE`AU7HnsDkR`n^0ezvSimIk&%?u&)|6nnuR|o>Ie=7)Co;a_e^kE
z=i+-nV(XG!><BOGa7df~_#1oR8hN<XoGCZN|IJo?3wO&7GL&xTQ$Ejgm)=i;H@|6w
z%|LwE*5fA|#i1=`{M0l)_ZxhxY34pdyNm#adV;VFZng~bQgXWI&mQkb@W6cCU6n}X
zxC+WE&rm&Yu3JYutrEGD)!5D{SBRX9K5-lE3<<D0$t<LT-MPC01fjGhj8^)ywPQb3
zvGa9vHN?r5`q-ICd(=s3vyaAOce=~!hm(E7cQqO20TwJ{g-Ca3Z?m9{d&yoN1$=1#
ztOYU0Q*{Zv8ADN-^Y85qYGrk0!ru~!K8~;yDEy$~`vj&2sCjh?gStu{exZ$*)~82@
zFqI~u@2l-UeXPFY{)pU@&7x@XIp1{R77CR!5?Q><c5&r5=nq~-Ac-L&*A6ZHRjK+Z
zI5Q{#k0ISyGjHuZ@Lk4SF4j|OF4j)m`f~6cEYKH)628Y$z9TaKSPlv?`aD72>6lZ3
zbF+H7F^zIA!LjYy;n;sQ+Ma{+WkkV1ks4;_K6?SQ5@RX4w_W?6VLV{X(rEr+7cl6a
zLhE1csO&cm{l@^y`5#TuEX@D^YL@>2obX56b)ya0`v;t$d8QQEk12F(^$>Su;<#{S
z^JmmnmXI>M!Cg2!LUPUCbA4wU7EThGmF&vo9Vrp4zhBSJ^mvBS%0<Ai(N%3{N4xu9
ze4|y$a5{#EOKXl4pL9}8`Y#jL$~6HJg3MQs95W2>w$Ht*s|~`M6B7M-U4bso@4M^I
z+_ST32A@~kR>B@%Ux}Ai+nueR&)2)p&>tkK`%9{t05z8uxjs`ORa2e{3eQP|d$-BM
z^`}-*{W$Bi2jvZ+<a5LJO>KL41hwtY<v;MPWP5wj`dwkJAmFUOLMu#tqDa$O$dOat
z#5}8Lb6Uz^5Pms2{f_*Vx7at!*nm9^Z9|F$!*oQDH`CsVX&F1WoHkO{z{C}Isv}TG
zK|%ED=g0iQqc?w4M5*c$KX!sH+RMt(S55h-Wb@d1{#BkZVnmvl#V8y5BWT$1#bu-E
z-4Yc)#ab!_8CmM+cegF!`Fs0f#%NTyoF$(U#IiyJYfY<rFhf!}hP%=_x-tYY4bG%%
z=Xtu(LAlkLv*x1Xg-c|nFZy-Uxx3@)lvH)sX3Ng}W9;xlL5(Wi^yd%?d746-gGX;+
zEa7A_i@&zZh6WFASvGlnT#@sgv)YlSDO#SI!ximL?(-cUBnjuJJq-HTnVgVV>u?8=
zrfn)X{AZReBIAs-9b6Dh%1HUK|J=L}${9R(d}YD(WJNw?z`*X+J@HQS+(7iQ4pV56
zyDcIcV;F}0h|-_G2G#a03iFvTroKeF4PkE2DENR6e@(nM&{l=>zBWp;8t_ZmFl9KM
zhcfIdDLvfS;T{pIwZEAKTd9)>ZoJr(aHT6#n&lPv6$^bTw{4UuZ~6Vd!1+*5?W|8>
z-1smy8dm?=gA|6CXCC!vz1H#uEFu&Z8aw?>%G6fz2{1&HWf(i-om0zDPn#04v6f%*
zse6nhMPQrkns{;2y@~jom(>Ug4K>&quH1yImzDL8&ha}i*hq@~vQHdz(jDD5K)fW1
zX<W+^;Q=GU%<-AYUVUH4rH@o4>#Ebji@9sm@hsjevgW(VYKd&i)b1W0$$;)Vzze+#
zJX6OdyyI8jqk3=|i4KLm*0aLRVvhz_tSIDA9Zm9!Ub#8<Xg)he@G@aTX{)KM@F^fl
z`Xi?H70l!QO*d47ApK2(h{znS6e3+>kX|t(Hn~nXt+e*9Xi@M1ea&8sRI{(R@x(+=
z-*+Hezncb3{F3(CP?4Rbt|VnSL%6iwAx1_6hDm?mIA9cdwnK^|N#KAMU5Xi3_vNKg
zsn%x4`#{b1rlv>hi9PAo$WzL2{K%@kz2QLdhy7(@cY)^btUI-==KHNm#>b||bjSjP
z>x&{q2u|9XvkX-O!46JqY_SiJ9@oLmD}v!1eMKt@6I<VS*+F1-w0-1K4Bk4q^E%F`
zKkaoQR$k>jMhx!9dK|S|;KcGDK^hc>FpB*vCGy%-kiqc&+!}F68ZWoai}ao-+c@aE
z^9GIrFO@>Qr)Tc3cesorQ<fi|&Jjt{*ySMxJ8bW@JmZMqbb7n-{XxL`a?v$5&TV9p
zU>c@LqGz%hd|BVQkZC-cjp1{Ctk6u>Ai6gQ9JUg~<H}DG1g0p~W~H=kdyk4%poHnZ
zm{0#Y2hMT23tEV16OL$8>J$ej?H3AG5aVV{;3!8U(Whe(0##w)ZySh;3pS-sRFzZ9
zbt2)Ab$1ID#8iJyIkDwTPPG+w+OxX;c&stVsOB!d5NwQX-3LBWuO{l8s%4z3HOv%X
ziq>De^DaAEbrzy%XBdlVew4-c1eZ>8jBn0`UYY!aP56ubtua*E9smsrHIHS6vQ_<H
z47(^t^#ig?=*i#$CyO6$h&3GYw32IZf;1LswbP8n+{Tc1Qd*L(_=$VW;o&{BwGN{3
z*dfe7I4@KZ+CPIzjsv2vyahG*mn~-tQtbK-b_j>1&Vglw(DRM5>Vgmsr!B;$`tmID
zr6E-e`<zk`zqDC27+i{hMXBuq1)FLK!0&-vDB)8^D6q|X&e-3>rH)OF{j~^6#a$X(
zo|+ls4N-IouzsQ0A=AZ@pt@|s@k#@au_~u*Wo;0OBkLQzN->p33v<y!Ox!+9KxLJI
zh#>%K$G^Tr?ci2DoK*#~{e@&ITpjk%0KcN<Zk1J4AOmvbA$DK2xFP)&W8e*o>KQLB
z-t%_`-`j>O(7i+{fim65tFsk&ukXEMDnj>vqwrx!#Pa)d;2Iz^1+ZZ(nis5u<1%6}
zeVwjV3KGM!RW03zKscJhpK2&8=aPj<eUe8oKoVj%!{=AK`lieM6cAK}Cu3V(Q$13Y
zCzOBv!ojtsjY5*Z3-ckdFdmOgpZ3=t4(y2pC*pU|Cfab5k|CVdc_HVTSYZ^t@$;IP
zz%ic`L}FQMee@Sc$lpV9ScUfRt`}^O(^u3bs;@!0Ca2uQC6E)|i`mBl46C%KH$-C!
z+g+1jiJwMa!79(PI`{%z`$=Q&-(3px);toQf7_7cr}8tbnBp@f)(cHQCuwS}GJ?v_
z71~lFY7}eD3=fE@Smli>t)Ur4swb{N74tPYfIS%sXDp)`<&sf<*f6>KnP-uClitTF
z$&$z{7vwf{A^BjROD662Ptfm66<dkM{d)EW<~jVQ&sjR2{0<ZiGFp^B2OF`vtdx>=
zo9RctzLfoWC7WmqqgX>NfT{B4cb~bvn$%cm(kIemX%^RZ9S3EPHb-(6)5_1zAveln
zy(nk9N_7u1DnDG0LjMk7yHahxTat4y1fqc=oi70!#ms&mgj1w8A}0tmbc+Y3Jhdpq
zC8xFdB0XtOy8C(6EK|tiabYn(a;&(Ld$EQ05CgB)aI(sTA}Zpy<`=XFMmfsmVcsUM
zw>vT!+Tiv`8}_?EB>kc5E}HX2vWPNS?qSN}95=}Agxji35$4T-l8>)sp3HB-?QGN4
zM4qcyxy(qW>?W9(UIphs_ca<ml#*sUH(Et;Lw(qH!-5KO9#^C3k;nSNgM``M4s|P8
zF_}R$0{5JxL*#(!i8}GLeETcEnB#94AVO~=nvRRSc_BW&b^EK0FaZ{-K6-@p&j!Nj
z0_?+<jCCQ0wq}O(>%5%vpHao?%sqf?mD%N#-XVL0^Igg==PU4Wo$u7fV4d2U3KHMg
z#tc@@NUar3;}f`?ZrG1X)t^)6Z?bE*z8u5|nRGrtncNCmu;wMDC?fde<FIjaZgT#s
z(?x%;=_x<=M2knb9$a#L_xSDc`8@pDW0t{RpEbS<g-{>gbNVhwC*-@pZuSdfez4mW
zodWG((-h>4cUZOx@oHgN%d|jX7N~x{V0JNUIOV8*f){-)hw?W7&xBF7FB>T3DcFnT
zDiCgImjl=#2m};Q*1z#yX0F&09B!CAB`?;+)sY<=BvbtUHgYymQ-u=uUzxXWxxsE;
zj0R=OI^1&?Z=g;xdpg_6qlm{)c$vc73hD`X%C{`VafuXH1@#hB=C&N2nV>gYS?%!`
z#S$UC;8q%Ea$$9ntS%xvZ-CTFkpqir%j#N^=80FCH71nmtW)OVunF9X3F<RuW@>M?
z-oz0zv?xOIwb|8K!)8mm2LYpbH!Zu&x6c856-g4E&<_f?s{kt&o<Er}^1=ZAGT`If
zS3k&D3ug56e(w6b;rr?c?9yqf{tpD2>(li;+Fty56&)kC`BeZD=lx+HmH_?=x2k2H
z?mx2T4)xmXfI=ZBsRq4B+3r$;6o}wvY)Ohi3uo?}0U`Y;Dy85|uzXo#1<)F9-jaMV
zHd^kq=hE?`y<BQs`>|Bmig9oWF}oD<!i1yD=!s}?JtluPTI8hxj$!>Zl{H$bul5vL
ztEi@Co~FkZ-biUp7MN^^djVZ)Yz?c#&s$UpYbP~EVNj&tfzOtRej-j?I~3*Y7H(GU
z$is^@nTmlnyrFzBenEW=@aQ0=jbF-E3K9sIx`8oVHMwiscS^slW<gRdCFlHubi)g0
zm9DW_mA(+a#*NPxh7<&(n`f+{$9bm@Z&nwij1{yS=2TN&L6+SY@z?`rz5o{O)jWTb
zbo|}RE<u$QkA(W<oUoHbj~ug}J($@^#aAs(jM;FSwyemxb23kHtSQoE0}eyD?>O)x
zCGWyfde7CP9MK1-D3QY$flW8Bdvl*X8gj90zV~MUtpdEE7VE3UPA`Ev7}ggJcdaSO
zbJ-`1#FedAVSm4|jK{~+&a<%33rsXts+i-yeMzt!sBlOb-oB#gj^7C@7>i#mCUMm`
z5}_FMwfI+C10Bd%v|BHaHp9P1{fJ1#w{m(0^5sO->6}+uUDdupf#&mo{%iU9kLx`X
zJJbJ6joZSS{7DtvUf1r%+t7g_f&fZm)0^xUSj`vKJGgS-#QIVGgez12Uv#*)8<v(5
zx`B37ZoQ~GY|fzm2HYi8yf93B7NO^Ykk=z4jFO^#pZZ)^R9KApcMv)AIbvaBTKV9^
zS8Qw!bB3*6JuPVjncyI?_xIpI?)xZ3C?CD~)R?|JHBK&`01p>GpZC{4^Z{Rw&%@WF
z@?sZ3zIO?E#}=XHmH4Z>B%^lW{o2fYn@6N@ne+Jc){-ILXaY75eh%kgVTcSyVJ^FT
z<j5cZ45t*eBn0OtRoqo<Ga1Q22bYls9PcIM+n8T$ROyp~yqyL{vNqtZkB~usmXO)!
zqDc&Z&2U_Tv)2IbV<wd%fh}1cl(-N%_7G*45zM@-^Yl3p9tBVgp$HYzbnBT+7q6!m
zRLW~IfUc2q5C_cO$wXr0+qavUHkYq+ebG_dexMKN>62T}D)XSDBpI<@2@h7@x){%8
zcK9t8PjrBa2wG0*%?GGKi^Gm9%>mZG*b7fvsuN8Sd7U=}++6FJ<A{>_LzzgxA}7xL
zgJH>l1tQg9o>h4*oSf`$Cq;@M(*>bXfgsfsNWckTh$#3qxQa&CSC@m!0aQeJcU2pM
zaIoEj=m6F2wSS1VJDu(#Hq=1Jw@0FWaM!#;l6~-ky1g9Z*4mswB%0yu<+V<Zh$=mI
zUQ}&RzRV$Bpkv{v3AWOoSNih5ae80R2_U41uGF8u8KJs|iE0U~knVQap*n|&<oQ{J
zBTA@JHil+(vU>&ASSZFM*s201)r)dR>LduM@Gu(JK8t+zv^%<Dr8P*PyNdJWOToL;
zS2(gg4h@g-dRJnQ@rnT|#K8iDR$w?N8R)W1OVu{Vp0dP_0}GeLvwbyy-oM7092}1~
zj|8VJZK2Vvtm|d|Am<EmZZ`z|CkUHRf(*uQ3~rDYW8N{yS+}fSx%oIFf{2oIzt4Mb
z8?}wQ^AmIqmN6|UUfr5HFKw&^in7`p(N4Jjs*>!`4lj1W0@O|Iz5VN)nL0{6OrYaH
zF~N)`Kb=_;qwJD{OsDYL5uCZGmhpkdlrGVMHs$UCh~e(FRU+u1w_d$kyrFT5)By0B
z0}UMAXjv|CcX3$Br;4R@>CKo(m)sx)>U+2@>o=~8tj`u{q#VKNjp5uo{64UZsv2-g
z7u8P`X1yf@5Od|HNj@CS*zhHE(K;63Z37XVVJQ-nq?h*SP{4yM6fJva5+JtrkvOv7
z(ERcR)(lP}*^%&rWl3R{^`o!z45)GB@AL;0Vr}>`dJ|#rS>0oBv_EW^v|~AiG=s2X
z+HhpW%J>wAD1->(HJw(|GMBJJ5kpa!Z#BL-tRo0fQ#CScs6>^`AF$=rj=MA+*-(qE
zs4+3iW6Vi}D&#*rsDhjABa6bz)cNtm+7IylV(05@6nB$iZ<)Y!*a!6ces(L95pL?8
z3W4Gc6`EiwIGVbD`+O!nxHX&BK}UNYnpCF@?T8}#6WY%1J);4_i<)3Nvb9JR9qAaT
z51zK2kw;nyfS2rA+csilXO^d{3i`-lEnBvFmelR1Jgl^E^Hiuyl5LCSX}oibeOF8t
z`NHRHd%^LJ8p}Vf+U~5u2)Wie?h&K^wg$8?P~oIsccO9h!V4jOK*iT8hU==HL@^l8
z+EO6bt#mCLH$xlqMjoS1#QFgLYhI$pwL-$+?Kn{rUITaG23nEJ>xG6yaJZ7>ar)^w
zPs0d9tZ&4{Yxv?WRx=#W)nQfrbPrP%{H2+8EM-=L%=uN5SZ%~<Wi=OZWf_-rX+e-O
zXAQHRaqSfY1xp_SN9Yvtpx^F>^zm_k--M>Ldp`qpH;OS0M2avtG;@~(oq^%h5~*Dw
zqmL%leLuJaXWk_YE`1)abf?gMw9w8!48|*jTF^o(((vkOR5`$t{DL8_FNw3O4FM9w
z?i4V-{!6d4mAU~2RB*U8uulk-^nv;LMS;~;H>jJNDX<g5Fkej#OfQEzNggiSz(Qf=
zr(?&{r@4d+ZI#CMOopaa!urcTOU|<KErws8rj)*E(WO%Cd7A9BZyqL;_jCHv7vC?i
zB9mChf*{jH!i(t*pQqF1uM*w!7F@m?m@OG3v8CrA?B?Dhu}Mb~w+<SM!P21=9>DFd
zmtggG+$gKFy8qB9uKzgDF*E(|*_o?9Ted%BQ{P$Ld<pQdb^uv|K`mXJPJ^c>RErtA
zZM#_hY6P8fik4Ii2WOM7XDHq|i^HYSVt2unv{AI=%*&K*k%H~n@0-E#TpPLVZc}B?
z472fKSH~Oh1=M^1E86z-K<5+o{Gk&$;K}!&o4dEq=Yc$-FK;h*w!P5j#U$cNYOa9q
z$3y8N1BXDbCn3aVAQ%~|LLcPpZ<A?g=d8?&&X){#xHPE#=^Bkj>k;(%_-p9-o$IX`
ztYRXij3j$5-QovMk@N{84RV!ajJF^ew3wjJzlKBv@qK?3#FG@Zr*T#{t57(Dq<ep@
zWYU&fIF^1qu~#nAx=S*9b<ktHD!rKp7hy5b(9BRpeyA4B|HG5waqALR++x%oa*hHx
z20f#8X{=`&bV^}}lNJW5Mx!jilFAAzEKtWCJNsL(kb8KrwCYF>h_G!svRRj=gu+Ql
zuG!6!RG+iv;=V&2&mM4VU>Ij8yM4o=KPJ72g|{v{pd(>}DfXrL<{ShbT3TBtSD^xi
zGNZ}R>e&$79M!BE!TGx!BLzhc|6ma=Nj;lQ4{-NVlh!B(*+=i&L=3D7{PoFdXB4Ug
z9oE2?wC@o_;pEp2l#7WH)oZ190Pq1G)-50(_qT_$myE(OqXb$R+BuqZDUTjl$1h;6
zrpqrWU`+uF1yU)U(78|~ZmYC)DMerI(K1o&vQ|Oet3W$qG6l#m>dO2~AL;pabKdKq
zn>)o%Fu6zIicxHfUT5rn6_Z^96yY03USC>(fJf1X_^`Dc1uGSU1j#bhKw0yvHq6eh
z@TmB4CPAp9Zmc~d(g0LoQO~G{vmjfXcrr%Fd5F$yZhUm)2rhXs+5ix7v%`qfBm+Ko
zB30+SP>Pkx)nc{tg$n3N@Cr>bT*-#EC~Fw?1-^eaS186u%q35}uI*+8p^o2pgz{)~
z>fvIO*fGz6)I~vJ@hm<|hX^|PKx9YU8b#Lsl*L-Am6Q-R6S9#<z;p$W_eKk(2wZmn
zstlWEleY0#{N_1rN!^}!xh4x93YdZpW>oPoza6GrQg=dR(+&Cys*A4l?>bhGpNiCr
zn2>++78zk<OB1sj(<NnA&EXkv(8<23$M$bwe+@K6+@*3RF-!6UMfYw`N6$_V6uwQO
zg4V@yFM_hb35>Ln{LAX(gt8BHY_TErSs!jn4V#x@mCWWyg(#Dd$%ALfxte*fn^5AM
z>zv)g@K1ss=1Wwx@{g=G{n_&2&>#MiLC#rc<??Cz$uOSA6Ck|{zpf=|<U7K=*|J&^
z<>}AI=UiH$4WfA5N}5QTlyK>!2_VP9U_9*QoY0{L&v8re)Q~3!bK{7~%5xu%3r3da
zTpprpe}Vl$rq_Pxu~I6LLiKEvwx;3enXLfvt=@-xapI;gvhB#c&*x`@*Mf)nCI|!)
zM<c<BczP1N4h{_v_2@RnO9dC|y))a8$a1bNeH8=yQ-FzNZZlmG$;WRNjcYbAQjz8&
zs4_bj+Sz&xD6nvoaRIWE#?=6N)ZV9Pu@9YcwtMx*=gm6bP4Q}z5J4tECbaj^Nc2J3
zM^@6YUE^2t=N)bTb>*rutCi(2Ak6=FXn_CvIH}HZtHW|juo`Gf?8`;Rc0=5<DoX%u
zO=MrDoAMz0tisR;jD#A2Gn9=Iz@(mk7uG=FZ_-0J-jb82Fe}W*d;P;D(Hw?ux)~WT
zRXXX5TSYG%E@B}a>l~a3vQGlN$>(^{=fDsboiHmig&cyLiI@*U(8q!a;1OZ<8>y)m
zn^N##oZTe(8j+pBgxE++=Y=9dS+JN0MR7VtD2%?#7K%iQ1QnNQ+V4-q&{lDa&C-_z
z4T75oBJQjO7G?qd8yj=p!m3pLmV8W?1?N!^@JZ>ISLH9UZaZQf*waC|Y{D$hl2<k-
z;0T0jnm!;bsy@Fb)@H8_lCsqfddmoP02CnwN~je2+F%eGL?r{~sH;_C&o4igG(N51
z{|5_KP{rx>cMD62Q`Em$q_~3{x!;Wx)qk|xmsFsAQ-^#T6?C4TlB#oVFJ`WMD@<cQ
zskL#?_y<OT5s8!zo&L&b_;o~Gbjw4R5h+K0y`!<Dp<$;j=fCE86B~$+cp-4u1VrB5
ziK>8FsM1brhH}wgjaY9*ZwwY@jyvyfuBkgCz{`r!d%nX6aA3&9d<gn--*={}iP9hx
zdwJ2NZudy65bCsF+#es_P*ex_2kQ?I`S)ek%ENK_sg|o8+TUtnG1kg-y(ji|sg4~t
z697Xb*}?mY`ei~SS-vbFn*f)M{F&(Mb~<DPQ#zk;VCjC+L1ixS#BBq+t;t0BtYo>-
z-}ajZAhBt2u^NT;D0+#_d3K(Df1Q$!cGv+$R-wZW5TIU3s(kIfD%v4``Ij|$K_*$!
z0C1&0M;FtJ{`S7^fkXKJ!_D}5-_b+r7$wU6QiE?`5U%N2BC^*Qi9Hy!74YWzTGgZ(
z%fwLy5x(f*^xBc+@hr4kwPD{8z$4H_L7)v{QVXId2>2cyEHd_ObZc|qpKe&;d5TW#
zqIL1TGI{CO$JA9=uT@polMhc2q58VBkY3mm%4~~TF=;B#X)c`$DnDgg-G3E!zjMGP
zw`czXvpBuDg58vfm0IVBzrt-$g^pABEOE>Km3r|ib|%8LCqr=XuM#u!Meux+%=rVq
zq4bleX$2X0Rj?Z*DHeNH3mmVmmjOYvKhGjVelAiCcGm=KU9We~mmUxKkKx{w|5fV-
zMfJ}lcznRg{h;P#1ckcc_DjRo7IWt4dq`R*>4LK{cVh>n1Fv!8-uQ7GTRf1khwE>^
zf%o)#&CFM$))xNj-xamR9s+0h#KqP%?3#@WXIx~{?ga4GYSNr)h#l60*qT%hif@p!
zFRT9|Typ(anGZ7?$N$Rj82*8D9kw9%T&Uej7PG4R5eYKZ;9aU;0FExn7nG0cRNFZL
zCPTELc!*Mb?8zVhSG=g35ud#^#3i}Q>US;Pd&_242Wxh_HOz4l$zf|qPy|Shhg;t8
zn-m;Mm{SM4AbCY%c6M+5TFh*XDoIq4p(6Re)wf{hd+(yRs@9fXEk2K!G?j+f+|BWG
zYx?Qxig2e_<X`TNujl7y^`47Rw%e}-01AnE%&4}V&H1sTtM-r4p+k=-1%h@rOJO8H
zomp4~=kOg?<V3O016?xhh??_U&+~3&R>i?We`G<>{|>-00{TF%XN-Ybj>5VtFi#s~
zzl&UaD{qI!SO~PkVM|)~J{AQ*gMo~|I%&3nXNsXS5rc&+Xy^0p5A(;j#f)5kysxqf
zv!+;2@uL)rIqSEL4mF4An8}CV?S^X}L#LM1uR-6maXs|HQWJ8wFNK}G0^hy5`O%KE
z$_my|x{FZ4N)t7g<I%;sM-CtJRPQosq8)6F;~Sc!AxbIHbHH8rirudn=i9H3gfR^O
zxhOR%zZZBsRgf)SJV2%21CeCq1?QetZlGoIoP<Y*866`8B+!SlvBt_m%Wf=YO~>`$
zDYNG_C`CZ=E}7kkjJW3YGxjxir6!utoS(c2u4{+oZ@1$l%9qfJsh>#tlQVRDr-U=&
zC`!TVLMBeJ7%fcSl$OwfT+$&knQ(YJ?ZIm9d^|-KfH&u+P;G{QH{Iqz6OAJ3KDdk$
zg92R{5WWPMTs03a*uxW=q6$}u^5Wp0{fF@}jt-wm7bS(0G7=P`aM}feSs((xSAThZ
zM>ANg@&L7MEY-Ho>a9)0WP9GiV`97|`#SI{wV`gFXGNhL+^T)NYJz@iZXG4^gHouG
zK~Bl5*p=AynHw?X6?$oY&0hQMFZ^U9cV^X=aU=_yA4UM_;XW?h`u{NY4l$xMY}8;I
zw{6?DZQHkP+qP}nwr$(C?c1K7nPf7_|1Q2&Z7Zooo#)VKfIXg|Y!m}#_Z-|iXgb+u
zlM0ELx=FyOg)`rKGw@<oJSaKTS{J4iKG9HKASuo*riP&v8Sb_byf7^+KQBEFgo(&M
zR8<7M5f)9E(~^sr=i;g`8Dclm``!#bZ3|YqVq*H;%BpuDU6spyL|7e2X57hn&{rs*
z{#4@F4L8Su@Bm`g(7M!>y@`gDZ)>2C5t;ayHSENBMV5o7{4kO2qdTO+Af<D<6rI+k
zJ1K|F!_tgT9`m@kh6|dGl_SW2I6!I?myoI6l4Y97j-3NkWD2xwFC&6}eFv|@LDz@U
zx~pF{is|1PTX4<HQgD6MpD_vCQejcro*$RkbN$Fl=J}13OiA7SS@1eN$EO27`J_0*
z<C529)eMKN^KPDsFBk=-HaRRudK;!hAe#>RELc|mOqmu23WeltrVI;J7IwKi&J1;`
zU?6*Hybu6sp|3a<p7$9e1p3iQppuQi6`R!(bNVc$DPr`?5rmKcBOhfMM?0v7epXRw
z(3IN!rs3t^TB7|3P22PdipBYa?;N0nO1_uQ?%9}`(!0W4etz6{n<z2!yHK3$!ETb^
zC+8vchb)Z!n-odeUzTn3CJlvD^d3!EZ%0k1ARiiPAAgTft$yG2yhR=EqF}apJ3i0G
zq=bgAolY}29kcXlQ9p&%1(D+%ZCr4FLc#>&M@jcUtX_Qlo|zSatds!4&<{mMn~_1f
zq*)F(v37Lj+M0eCOm4p4y+fNgJ05nTN_wnq8m`n{3}j51bYVArH)m`q2tdo_QR9OI
z;<9&zUpSuLzv&DQ1Ym&tJNFrG@1MBjT?%mKhYy&mrQ?S(F6b8_zRlb$qP6Z9cxSbi
zsO}y5-Q!Pbb9ql$bGYyD%*n?p7uKq+o~wiF$mYlQ9|_cAy1%KqrF{F3-^2xT=kJ|c
z&twj;x-?_EJsV+l$ck3|Q|bp(+BeR6BLVz5*OOb*mz{A}Zc{n&Mt|<ziaTwLYa2Io
zS6)$HhrdJoHDFYJAaBjps&zTQ(IW2t4RJ(?>EasuU#tJF!M0;H&Dk*+RvUvoWCAYt
zNP_?!lYe#qU3Ywx>~8m3=)}X1oP5rzyH+3qr<tJ7yd2LNqm{*k&O@h}yGrF9j!uT>
zbB!<onDM5_sl}*v&bN!pqmeguYWmScvm!8?LN}x1khjwSzkmhZLWkbeYoL(eGalAY
z<=J86BTEZ5mN;#`!g!#eVO(Xbb;3($c1|jL$lDShwL^kLD(#US#E3y%C5{1lzCrnW
zGurbuwTC~zsA->(fd{Lzhc3P`Db50Up4<Zr78d6}F3dGC&s_O9sl9PX%*ms6)GKM;
zPIa5Vpo$L%X=Js%tYB`s*aExU^@0uA8WOms`P#r!)5^<1{@X@NZFnAKq{rhW>SJ$)
z6Ck$oT;%9@1b(22+rsl2;s9cGOKfbN%il-F$GZ+}wrlccD$ql!or@L{QafpZeD*fw
ztQR2;PxzK7p3KYcQ?)aUro>L*Qr5Wn#P(DXPQ#h%9N#NLzf7@K1Jr2tFve@OvxX$l
zIRg!}{j@}1Fol_%GM|I?cgKh%pv*qJ%hwONt7==G8v+7zg2g`4>c+fm86g%XhNf~u
zj291ZB24$hI)2<-jdo-P6Ad=e@lybd@NO#<0CX~R%c<C#ui4ndB51xvSFwZYo8t4q
znoALXWvm2YPK^-P<B)WfEP-sU<;nu+&=e6a_VP~*6^gaVo6LbH&+MuC6Pi5vUM#XL
zrs3;E-%)EV*hzfe5KB^muIgO22icl|p+1%bdangj*3b1hB$4C|?%4J_-%Z`_$y`r$
zr#u*$%81V_@=tbwok0df<vy&BTZo}i$4YOb?QWgv2q|QtV<!BoQ&ZqLNKP7&^#!;z
zDWEdEXKp-z?ymjFt77O@HXEmkNnAY*L0C>1P?#dxh$IaSIh)%{S`m;^Mm2}WyMz5}
z%D#R&R!L{Sjw2$;sV=dkzMkFA(PRGJ-U;y^v9SY~Tx!daJPiLm2G#9uyKa`R?+6{4
zj+iN_2(4pe15yZp9*k)=W>QuaT4Kk&En9iM)Bnlyu+cOAM@I@h<NuxlRhLTKWJT<m
z{S_k`F&IGb@G(|#tx}QXCdNq*$H!gT@evc`g%JPpimEDR?&g4X2*fvJ6=hwm=z!#{
z1H8XE3+1;y3*Kf{n!c*(`b-ZDPz3~tZFTOtF}yAWf$JxDye7Ynm;yotUaCw8{TyFM
zc9y!%>hAFRzF&Bt6zBg8ZeRD_=;-!#{=7`qCc61ZVuI;s^^>@X>$}wmcCbQ6P)jP^
zdR?{bl=5zS-_)SJs{QmB{MvRH=wbrn80f|q@r`xXwJ;o}l=No1+2PnHyJS5wEAjZ$
zMkIt#fq5wR$fE5R102LgV$~0h^;mJa4jYNZ9YS0-&usMB?NF&qJ`sR<{+J%O>SMNy
z{gBo0^=94!tT-iE%wnSk6yJdxHJ;LtyW&ln%tVr{uN`pXdp+}nP1dM!Lx1T`iP<Lm
zxgKn3){}<`Aa~yWEAbAM?mehW(XJfrP`$P6RtcPc8ytFs8)XV(bA~x`JxL&Q;~ugf
zNiF>et$J3|@<x4wwn}KTJJcCy#0To!Vzb+pv#!P^E`e2dLm>(F4VAKM!pZ%5n!d#P
zRJ82AwT+|hki|Cto6>L_FVcj9@S$V9DcT}*%c|A5l{(gf@I~K=GO`+&_p#c5Y)Drf
z&fn=p>X88=l2&jwC_C^~H6}}zIXoBMQag-KrKR?#H$_P1Dsbff4NVsyN|?UWI)1^7
zC^<d3AXRm-jvfuc{=JIL$x-`K5xQb~p`*F=q^LVzcaAVTph>SNm^IbU*^b<q3u1X<
z23zT0g_uJj*<ruMnuZcaM;Wn0jAxNm)u!|i4i3|`qBUeFau3KdCGmJP$FnuMXmNtI
z1XsoUwon64z`awckIMx=fBI7PQ#e5+V0D&S6dK3S1S*+l8wzy`s-;ThewZ*cQkoT)
zLjN{!jif><5xd-Z3;6C3!RxK7VJzPH8MG5fBi<6r<0g$FuJP(1?vc`vt%0-)`Mvbt
z6af=r*}nq&p`n{~YMRRtm}+`6As`O=GDNVXL}t5%SE#v^%W&C$V@tcg0QAin3I9<^
ze77!Vu}8p&W}#(sL$ROw{{CStdI0nf1L*&#=luE(9!e(zT>)tsm}lcZr(c75@OH5F
zymeRQcsn72C!u~;pKq0Gzit2Q87-BMeI-_XajDp<-Y*;oj&rp>-or^%B_1F>Gk@1E
zEA5dprb?Rk0Ztskk3`{r?Ffu_ARPxfJ<Bx*g4%0G7i-~*P{Dz~7w=3$y3P2FD`N^4
zi_KHi*O0WdoAQ#56OJ9ykEayKh<485(=7pYBj(Vu@e$-(@-L&}-zx};7)QOaouXr*
z*u$5ucM(n%w^WWe!&wq4H&m4;bfOxtGxgK^U!Pn_6;erzXawft29)2-3{NcJA1(22
z{gprQsl^I!EI+}zP`{a?Ch2os1}8Ouk@s~J&B{uoT{>!H4MQ(3^Dt^UMiEziIY*V>
z_zgspa>nA@6C!)BNTdVTKTmiSMNDYfM4C$GLuj>tO)9eCOOQ~hrHgQsC8#4O%PH~@
z1Ws8}zQ~$bdxX@KR%lB=fL>x&Tp4{7q)`CO))JSH<|kavE8iq>KsC5W8jh7jZYCC(
zaj{lDm>*e^|6pa!%;b^l{ibhvY%YmrU^_hRm@!`iEtY22Y^ej$a}3>&MDBnL5m=x?
zv$<_iJ(jO5)n557P+V#>)TLS%Eqanw>{NvTgwUZPmfF^e3<dz4@_9$?xqY8#0kQ62
z>+Xvavd<>6Gxqd%DYl={6t}uS8WQF56krkvMX!DOWcCKJuYrkhodTg0UNKGKs(Il<
z^Kfhn$#-yG8wzRaWbdSUf%j_48o_htf%WptaOlw;SGW7Y=I5^T2!L$!vmhc90~_F@
zQiLQX&lnnWmB+kRp#=-ct?z0rkcyH26$DBFjD%qb@eN)Ll9$lIkO+l@c1knUxWt6m
z#%vmgyCGM&09X&jn263E;Je~8V}v5XW*~eR(=WGeW7x9~HZ^j`A!=M@K7kSO7N#bA
zu`=-?KAF_%gDchVZc^{<J0Tx~j#7fS@>6M9k8E6L%ZJSxoz86+DZ`F6ypd6R9?~1Z
zs!pHL8_FaJaH|s}It`$%zZ`lkJ{7H5F3#m{@vk+l00XYzMun_wc8=`2bdZ525QZ60
zN-ts+EfXi5?^mnzh4O^ugFTM96`6)A<3ny+|8+mRm6=(7DzY~K9J?9q7act>T)fVs
z<b=s_ckUI|DLlh~SVxa_muH6;1P4#s^22<5$3ZD+oY0SgXU;jVuNHRI3Ky%U{E*Dv
zA8eTYce5OM9O(6mP$Qq;5k@uNKwzMcICOGYK(K*sFHOTufYJkFxsJb|S4C&-dndp`
z6hEDd+`M5flkE?($i`H}0z*LVcYP@V=^}WM{a0IjgfQS=UWG*yivpYZRAG4@zsHk9
zpKFiC+T3A0Fm#?e#NLfyqU;~IDHCu;E7*+$H_AV%8<BwSoWwvc9!XzIlT&qERDCnb
zvXdT7blURF`6qqsxT{9~Mu=u51oBuE8(0@&*rgJ{5ZYSGtZ*@HYX|$GxTzzW9@uvE
ztge`PK*>2TXi?LR_%leUo+r7fLu$?O-HkO}{ouZk&FBq9LSX*DsDHC`>!{@|moe>{
z9EIVKQ8$R7k@7O>^C225>EHDx;i%zv(BP;s_s>7EDk8hw{tGqy7q7y^%>KXfD%Zce
zcU!E8KC88RyfiAs{&sNy8tYcchvX60L+j#+WL$7*7?MJsA=r;`1(MgkyI`2yp@um2
zUFAuW;Qa7r%pF*|zSH?T&0^h;g|;VRrMsX5krLGHPbX?QaXRSYf`N0HwOwqyrK7zw
zx;Z|hW?PxV1!3Z-e$ZU!Z?Ci8OnjZ}IrnkzQP{JB5?_&xkCBa&lW2V1Blmc|?=M%O
zQbgBb9oJ%>vzYZvHR8xp$UQPnFDW6=#Ybwgc&IO`U{OKbHcRc7CFF>dCZ$0^gJ8sA
zz*$NA#4|yo<Om3cCMkrBu-!Y%F^eE;#hbLG2+7knXzYe|h;isGNDo$$#;EWUB21^P
zd?gNd#L1zZB0&^4kam6p8nvXbsE8!}eBBykXQ`E4Yb#jVBvf82$4B`X`uK3zFyGO9
zH#8G>cl(2|w<X=h10vM1$8$t;*3fasqudXX)jQ0J?z)t*;`o}o{CD%?c|-zt{NOP&
z{KtJo(&gU>sEICc?w;GSB|{)Vm38R~vba>l4ACB?21*Vl2xN>Pc#Uxolwg$6a%lAs
zH@_<K;^$XUEOB^2-=)llk~cHGCcys608n<5rnFmG2E{U9CxuJzb`ha6v&$ioJ8+=i
ziNe~rXBb1yrAK>D=8`PH$`BmA@LJ<wl*F+kqO4|0zx!(}SCGXFu`p7QD?cRq=!|##
z{&ww(d3bj*ZDJw;K6EuHL7jP*={JvK^W1YRMH34AM|FAGCY`2?HAH&w;>us}Lrxz_
zExaX}Cz)7O$-3X~MoF>I)i<h=kZkVE7R<_I6Tnfob{^;GivRabm^rGS@W_ye8_@P>
z%-#^;UI||cJE+Q4$ps<k-Yn@&%%hI>F&i7`!4BleQxwJEN0L6@!(i_le7k#}F>A^%
zsVV*){U@5`gc2RcW7Z)%4LIb`o|*G`%q}bBm7V@*Uza%wH{ne*4orT8<8rs|@*qvb
z$LDhIinV&hU3-6!u(lD!PP@O%H=THUSTqF6R~+3VSihsj*WTx-yM9rJ%;Qmx?5Sd=
z!Kl&oM>Ao|zG~CCL(9;$di#QLy9Dcc&oI8~IT{cN<=+ql;9r4u6+d@EkQ$Ka17ZHV
zsaAZq9@#5MB*&;ZwmRz#FF?f$2glIZxNz%_Lpakli3!SWpw&ULJRc<+eH+~wihO{w
z!06I|;!>*9Jr_3)b;pN$ZNL2qv$lc`Q#5)B0FD=g&akvHvnDC`Sy}<%Buot6={Xl;
zYi<h<OsrIf;62vB_4<nYgnDMBv+BI?X8%}{Pf$mxE#u5uX(PztPit+}HO;i+h>X;8
zP04osD--VSftJ*SE5SGg1GUzAAUKsBw9(HFVSyH8JH$q+QnwcAy^^u8%&<mD;LiHh
zj(JXbVNhA??#=z~Mf963F^WmAPMHRj6!Zbzt^LW<OU}-b{GtJZZUQq-;5>PWuo6FH
zUPUCEXSUhAR?WcKRgcbP_6Etx-^FU@<P(gZC1a*l%$`n-KNTqY1CR?rJcBiMuXEel
zPEY!TL&E%oN*Drhb$f;s2$Dw6;xBGCv&<5>9LjiC4AdE1FAyh2TYT9Q)LG9`o2H3a
zRAcR*xWYGys-84}g&aqo7FWcOsW(vlDnhx;tmrfyQ%v1!Tl;HZfv+3;PtiL|<D)&~
z^aPMdheFUxN_*vHi?@E6V&Pbd(sr@t&?08q56@4l!Rjqcd*)9Hpz^qXuGyhEPFVir
zlL(zb=rPnV`U_^oE99gzNS6N;NVYT*Qn<RSx@geQWf9PnYyrF5<#QwnWjS_1lP`D_
zqhM9aT2Wh{b+@sjF64H0{t2|9+Z3CCuEZBu+0d3im($)P7E8;wC$4ObEb8NB$3RnK
zIa;g80+7W8OCeE@MuOrY3<C_F)Fc{ydZf0;;ez)Mpux|ar@~(T=VrZo{0u90Z+)c}
zE@E)zT_!^a>WOXzQpdPvrpw$0qn(+MuWdG@Rw)vgHxKXU)<>fu2(S>mL+W=l;-l3j
zHPFsP9Caqr!C*AVs}B6U%bLRIKYeK91?mojZO%B_u!F6&lQUi2IzH7^_Z_g3LHn|a
zJfc2}BER?$)@~)QeBSWYWE5>vZ~%g5H^`4sr-RBL!Wb5vNh~#iNTxpB5w1{*Q6n_`
zj->UV;q)TFU|1NKBQ5ZcYUvU4D5ZQ_*#9ZjeLq39x&31zYS_nIgRsjhrTXrehBuci
zvwEiP_9_1R_1q+oN7_KSo#@tuL;wunXUBmSwDe}C&*-TUAxI|qQ<di7(>v4oFy|YY
zU7s9NQgOdu?YUvx4SA9HtVcExh`<RFO@^z-!lvrs6xgV4@gAA}Ul-UbN6kfoeWleF
z0UJF?9?`n<x`6}K8GZHKx-+A6`+{M;e<OObEvhsX$k$T#X{PQB9b^GopHZmJ(t;1H
zAu4X~a<UEi8=^z#4Yff802yC1Y->GhcxVz#I~%Y~3DAy%Em6o@r3)RJJ6ri7dldt`
zT`lzH7x2M9)m?qldGThCP4Y|LanAJ2Xe#rxLIHf=_8^ZE3%-+%)=Te3;B1(-z;UYX
zstn|OZMsQqvUZ7#qW!+1_JXU@b2w+tIYg8KnwuD6me$o{T0|<?<mEONl~sQ=`5+$a
z<@L|AK?O?)eFP|gRDlN_8|!{D;-z%T2LlKo6i<-@ZJe)9WgKnQl0HGbt<N@ry^;-)
zz<8)vDp>Fp2<OVAfv>2+ptrcogY}WMI1>7@@eWX}4Z4RIrO%rx&!3%?aE7*!F|F-a
zgT4fNn2%cUNbFqDGo!J;%lJ--2MA(s8`n}Y2sS@~Lk6!nQEMApriRr&FzFcV^OLn%
z$d8q<2qQ{M5*Sf=F9N!rs6ilQB84;O6p~hB;pbR)jMlkX+>`(#sA(8hNfB_9;k}tk
zI^L+IPEPxgYPbi<tGUBeZw1m3SYM``qG-W8ABe9f&$V!&BW$Y6au<4#<sxN2v4L{D
z3hFDJ+lNg)=Yp5pou#8G`ZU_jJ3*?GQd5D(JcWA>2>(cizE22VCW6;KUUNBzGCuFu
z4W+%p>xn)b?{rB<ctu>gKbH2*dz6SL|MaGElM22T%zBwX4?*mIe~3*qozO=8B_H|*
zH+TLC{XfAM3&;Q8KC0|)XN*rLXJDb^Wc`Ou7N3Eh{{K$TIXF7uv$L`Op_4YYF?BM-
zXJlaipTSx8Uv1mN7R29d9lo9J2DlQo5A=9sUZ0PiEz}cG<TR>T=UCvXDv}|Rr9Rj7
zP46uRN#<WgtMP_gP{gnVatAk)lPT!)xB^DXq9XD!7OK;#KN!{%Z3zcg_ArzpwO-2U
zca)mR#hxRkBc8(}<b<hYxNziPUEvYbpk4K5#L$LaglHHribvxG{=7TmiBan~6m)7a
ze}oddj0er!a_I9Bm^u(e$Ak9CDJ=2fGy_E)+z^vUA}Oj>R#s}e8fFYhrObwcz^Jh3
z(KWqhgcJSPv|fi-q3PH1<JCg|QNaounI<&eAP70Z2`~&fbLr{p90>n#m1;pRkWr>G
z%B+zbHzWfx$c|L3VI9MypitK^mI4=uo`3}HOl0MtzymhGs#US=iNM5<0^|9E)ZI^|
zDTszgISb+AF%|tmfDmNC&I3UF1X4C*XydMfl+quVYak87#4e;0h3L<OG<?U8H02IP
z1Yj_MguK)@sRwY1`M1!gYLpM+q>%y&vo8ZrkDuc8%TI|R5QjpFjiLJ!h&jnVfKrDy
z#*b=<0B;(gixV?+lobRgpie}w5K@$#89Vn;fUOad)z;aFLmNgT&@)vrlbYf06%H++
zMMGB_@TWyV0H`Bgfx?s~1d<L_C8EXa4M|)@FhC*4`VVg$dk2-51kDl!$!Oe-QDPC<
zLSyP+LZ<?-BFHrkUld4CdTQWw@Q1s>&xN@=*fUV<_<m_s91L5iSd$G;T6B}I;z4uI
zX`%3;@}bn)0l&6oAm2{Bse0!PZ3sU1Uw}kag>QJW|M7$&5^A0Nh}DGF`0Ud1rqt2W
zT0%mmJJ~dDrlgJ3-T5N^+1<8k&1L(cjp%%HYRs;7YVP?}a`r<ZzeUB3DJ<^oumW4;
zJBUI(+hl_3&Y8<mm%Ws#Ciy0u+cQ!ZJJ&CKTr{cGA|E1q)_1mJQZpbH{@{?*%E+L~
zVy)2_mv&E&M)K&UHeNr>I_PX~t5nL+d|x~0)UY0x{QKwJ@fSI*yp5*TwDr+KAqT$$
zhI0TmESnU)mE5myHt(;DA$%>b(8Prg4_&5WZ<iO3^P*a7<Ht1Q0ni?D(3U-@Z{KZG
zA^BF#zC-Ey=#`;!FU>bgEw8UOU9D});<^RHc8zJ|gH6kK6VapPGeO__FX)cY_5INH
z-vEkoa`yG~Y;J{vk^}m?dG%WI*Lrz4m~zJtrEmTj+_rDr2zQg7)@;zbGcnA>z&n&6
zqV>|kM*|2$rOW$E6y<i3PSbf=`1Sg#BMZ5fUa5V<rjJ>J7fl=wxw$;h;_IyU*nq=R
z!&{*lu?ysC%<uUcTDDOHzo29HhHw!8UlABWnBkiJ78HmYnUg<CiyCWHr>Y222z8RQ
z36+Xsw12n-<O+$#S>$lOU|5HT`QwUV%=Ddr){K2!;lg6GI=SiUJevja<orDQNd4s;
z7BDgA3P<bcz?_u7K9DK6(9Q?H48|ncj<!LIM@ldDs?g%yJ!iJ8=WWDvrXOW*Atu;`
z$Dub57!fHoDrTUE&LWOZ<v#bqyErfV*5|XZq9v_*>8aZyF5#%5o3(C}4E?%~gTwsD
z!gHxpirpIYr!b=ggIR^#oKT5ZBDm#p*rf?=6PNDW^gjBS)z6I9$2DAga@Tabu>WkM
z5vH&oU5_c6x7**UcbM6#NCdF+$qu3q_0F`Fc@)Pk8&3-LMF6$@5w`~JGMVyr|9f*8
z2S53{iX1a~L4{JAcv7o08&V*L-3-FnOVz1+YiG-MX^a<YJf<Y_VYWlzv-z_Lq+-&7
z$J<$npgFUr(s9Y5Ic;1$(>r~8en@=g9O4n*k|iY|p!!6@VP8XWE9v<!Gf?un@$h0$
z9b}^<?hii5go&1082pWwU|U`>Eiq$9jbnVNQiX`%_=j(j5QunI4oE$O%J`hT#1#ff
zd<KgCq(!K_x&AdQt)wE<5%2Vx%al`o?hmGS99(*y@nunME|i9;<;O?<3FFA~OJHo_
zpO;ba$}b8PxiikBl5yI&QrO1WV{JUnuFGxuo#%3#rLJ2wi$b^nO!sV_*O$hz6h-k<
zXrF64Hf4?KtPTuPW7p|iWip4n8p#LsWDrM>1!r<8{PiU4B*vzw4*wTe3pcXOf&SbO
z{nimEFXRxeb76?nDjut;jexOksg21JwgT-qYyG)OiJsDbXYl6e@_r$z`sQsr$6RZ-
z_2S`ro|p<n%*-9VM&@sOF(;iqEtU+jH~*U6le|Yo*J7TTTuy0iTvFYnOL}(aMS+t$
z`G@OghP+k;cP!VVUkP}NOerqaZA(pyc+d3|4?22=zA`PQdBtJJEINIato5`oYqo?G
zEX3{&Rd70u)t`PJc#}2%)NMpuRdJsOazOz4N(Epauor}sp~hBVONUc3J=seq2|lmM
z&5@yJ{7162qFKFv>jmVZ`0*_?OOG8;rTB-9Dzk{ES6-4Kvw5n~XX4nfrCTvccOlTG
z0T`@y^Uui*r}@%r-b(dPa<#aKhqh)(qj~j^%ow5|T0`S-WAqXeFq<6l>Dio5kF2<O
z)T*OaBvKO+@G29Xz#4H)atu>5J4w?-Gxad_0w{8QQ|;Y&HXBkK@dS@<N~~eU`rDD_
z9t`n4x%HEDk+Yam-Pi0B3_H#)bQjL3`LGy*vPfU(|NZElMh4EzMOoP@CgolVXJ6)$
zV~150ePIw)mdB=ER@a6hs?WSiz)n|-MO{`VX>k=D18Uy1drAu-3N-S(RTW5dfs;GA
zIrJ)d#XiNFS#9zirP3dskz!tmw5^9Gi*y`O$3nP+Gwvt04i|)LtB)*%e8!zcF$+A`
z2WF`>SdIP_WuQARwTi<rfu#4e;t*lx<~N>pBnKB{B#5#P`wH2P01jmZ%Al=dbHwh*
zcU=?y2785^C;C6Z4;#aO=~ayXdu!>khLq!0+yCR}L9ICOzatKpIgYCy+qv?v7F!@G
zF4J0q788mTnjjwiOY&>zaXJ)0Trw-VD$LRf!G~~Sdw*jI5oJS&-of_vw0C;QP{1)S
zK*|wDyA&smPk_u<6hW&zV>Ah3aOjeUjQ(-3e?L69diosva6K&@g{m?0y5BzkXf%u;
z@ol{2>-l*+-+Vl*jJ%c8kzydJB+n-VtqGwtZ956$<RFxEO6iLeg-%HjZWOPNqAxFu
zq^as%OTux0zpqeNb@!ZVXtQH#_kNPeRdZRZss@QVCkIAA^e3bj%LN3EkBIB<<YANx
zHc*78ZLm{Xy;T+2rsKVhsG-Z;aOj<!TnSh0DECofKY!c_-$ePM;_BE%e&K#;9!t#>
zI*0^0LZnP8@)7llH!>pi(!e`Put*?bqN=maQC#dJRaB0L&WRbV_&2^LL2gLqrj<HM
zUGw<T-^049J?}Ff{SJ#kqc41k@fJGyeo@Cy{o4o+)9UYuzOq6|ug*%N&ObWbP#ZZD
z&ZS9m*mZN$uh=hk3K)7_3|UBxyuVobRX`U6Z}J>qPngy;Z@?2d5}13fUrMY$vSl)Q
zeeLtkO_09ggVK-U|7CSHuI%6VF&22pls)Wfz23wweK<X+J->5o-9z&Z<c-t2o2p{W
zr`w5xg;h$$)oJxRChOBz*Jo;LURob{x+7cYVBf+f5X*$4J~D(RTU(7zuW55z;h^yX
z_uG5$FPUSy>_wb3%PeKX@hQ{9wXxm<qzBHW$;xGU0=5>4&-iFae`mck_U{Ey#>lDn
zji@LlwpAuWg!!kdq{zLJySAz-w@I@11LhLfy>gwSXxj<xs*l^HE0srz>m&?x8d)!8
zg*UEHxF1N59%4>-v7yySF)n0^cfB?{%`S`cX%ALG?G<}%#|Ujtb>_#0jLXY8B6@VR
za^p08_dQhRMm+A;pgL$#5x0{CH_-<@uAO|*t3&B$m&?kUvk2STENsO|Nl>%YtLjQ?
z%u0>UdyrN08!CpoFIz$0&bqR0V8XQB#g1rHxiguF*$L%kNkzs-%-Vyf(?oK_$?OZM
z$Hn4$W^ViC)4|!}{N87Pd<dK!Y<}VK3&g2Nfu1)i$sLNMhNkdkVehvVf%b7p5M<hE
zoKhH~ixR!#i98s1XYBf)0wp7$3v~+``uP`RG;T?UILXDS(pq0m5<WVCrZ|jvYDZl1
ztM8VvO85{hTEvKz>i5rIQad0^5+nzwp@9H*Dl0!uOLlgo4WGiNRZCVj24>mXp8{B{
z$1E*%t2nG2n~q~aiv?<y6{4Zb`Z07J<8bn}R;PxeGNDvCS<N+5S64-c>;$y3pQzfq
zA$C}+Syw2N@6%$Vi`ECYS2Ow2xqVB;(K{v&W!DC7WIF&-q9~{QP`8(O940HkVuYRz
zTwNGIjIt3D9f1<6axyV3o>U%5M0~)lD$=hsoe6Y~;Ba46m}so4Rq<?I&faAm*|G{k
z{_U-SIVZwWT4V5*&j0tSp&}z#l?x?70XSDqnTCaAd4$JHN2D-n1JAyFc){onqXx!H
zjD0eQTQ$=7oy?q;1n+^Sls}6KjH!}3xWZlDZxR9dr4#~F_RY^OV}~i2E44zZY~k?2
z0%OTJH-W!u@)ickQC4NT&j9YL!T2K^1L0DB#&FjBH25i~;3}V$@i5M^G~e|HPyJIt
zoZ74UB)oJrKS&!|FH?uLZP5mXUh+;YCCdJ!XH>NGYGNK8VejDcmwxDI{O*S@0Pw+&
z{qa5BBTlP?>d-n%>Or>lIgB~|{C3^O&T%|rM0dXP3XwrLVqvOd%jYzq18<6>w3ir-
zoz-&jPm@e<?i4n>6pstyD5ExYryE*nBM@5a%l{}^Uo@+~>MZ!%Y5&Rx4VezKJq1Yc
zuSfQI6)Z?gPz&fUizMzwHkoC@UQj7#U%-fL#JR@uNR(~mRxeMIx_4EkOQzZyrk?2?
zRE5pJZ>&+rdD0^q`u#ylpX}Y`RJ0|YSIY)xeh)h>?<Y5QU%qu(Xwfms)b2njO7Ag|
z<311rxX3_Kk!94N_8ST)ge6ny-}_%L3@w;K)84_V-1=B3RL+;f=g9agEufUX1b<bH
z*_fv;0o>#Uzda%@@5h-P>^xZQE2+&DFiLv1ASK;S+KrBN6_q!HTUZb7CXt{M(HlxL
zDr&+GoKs(4?w8QOLI5Erk#2CTO}TN~n2awR_)}tp1vLuYP*UTIaH=F9hJMCx(L>Ez
z3fE%=7jI3S+q$z3soeVmk6rH~LJV<mD9BS@fA@%3g|N%n>~_AJLUO=(rrv=tYc=~3
zzzd{!0e7L=8Sw;(CCEYxCu=zcnRG$W_<5uKz*6!OWF>y_Kr`4FV4E!XfX0jb`-#_x
zV(5o73F9ryGbI4cL^|KC$m>41g#Aw@?FJ`G7-7mL9_R^kLT<VqHDGunPSBq$YVUSf
zAYjUbl8@>@2<pkEZr>24fpTI5gE^pL;C6%xl06o4_^(*zKR6V>x$)y`?2OU%H}}W6
zIYy@Lzl==H4Llv<`N{D4s^PZrYge{4%nZJw(bqJo=3bUJrN#yV^wL)sD)SkrMOC&}
zr46mQ(rJf^z3a^2FG}0Vl?I<)Wv+_wet*FZrW}sHMJ>4D`dW#lfd1$&q?O#kXOb!;
z1W+h~l%%RL|LtdXkq7d+*`eOZ@ZGC|Q=RANQwAm!zznJ@Z-(=89vg%tQ7nh-a;+&=
znlRbUeN$2t`Lk(bBJ!=&-5_y|o54S~pINYBKo(-?h!P!mJLQYRE|_|lRMLXEFKyd;
z+U@!5A@*Cb%kY%Dz<*Z`Q8_5+ct=GtNV9S>R{O7us`m)RNozSUu>*^QA+SQjU@*dW
za6SB<b+{|V8Q<eB^-mDvrlFQ5JmxAOjGA)^3}Y&>@je!QEToB?UDywR6r<4Qe_@{g
zYHDX<Wc=UNw*L?FB>Yji=LZnKQ<y24kTAOS#BsOLSw&?T5`S_S(?}rFkRa*x=F96}
zHDj)<y4uEezzC=oEzH$rBfrsh-|}>RSUnzJpQOK2Oc7;#NU@{~<3<sU3Z2kA+0oRU
zSahAFo7MS~BikmNP$wP~r=R})@)o%|`53tvth|MG<&;l&qnp*+!S4Hb5_#K0^3JyV
z_4RqlCb|LXI(nPkeKt&MG?l<ixPN(~OzhK5K+(E7e36kNA~x%+o$_ogMYrv|lFrmc
zY%GC{l}?=k8Sb$i$Dk-KXNO8GsBX!AXupN^xPn2wv_vC4)^ys2#bHolpwEGv$Q~73
z?x^K_SC%HCmsaO3YQ@Co#}_T^o54;l&d^_&bBtpp_x4*-EVZPksiBLaO-JX)8t(kK
zk(dP;??C_sRJ!(>^(e4oW->(F#i1rMln5Npr4X^(f4=K@4kBU3ZH&zBNdUqMGL(Vi
zVijeiCH{|^xaGAJ>S*<2o~{swpb7?7*a4YL10D`cIn)@$oU_B_Ze{3$oiu4jxOiUE
zk*HqGei^Eh+e()(A>ErVhZZPi7r0|sI%>mL`Qu38<NL|aUJ*tj4Nn|wj3I!GK$JBM
z*c`Kw9pYrX&k7?D-yY~S2#p*$_HF>3lyGbkCM{(Zb4Iebx7<;Cp<%mPPUPxvbX7)*
z8JMsh6yY)D*nJ*_&BikRvVsa;AYB{Pcz&D1Ve_VG(?`>3^|_`nHD(6-(^_ds2C7Xc
z;GNDAiWb*`4bmf;h%O3%RzPy&0&d+z^THIYB5Apx!@<(Ssbd}csR@Z(kY5abrr3vl
zZ?+tjP+C1AE6w`Cf>XiNZ_$v9G+|tQh7GS$so!JI=SRw;7?8=G4Nd!abm=!N9Y@f}
zGddYTRzX*J_%4nyAm{nMc2c;hz21COSKyG(sG)~y-C|!!lcuD5I$S|@F+nWqH$9nS
zzj{F+L<)&W_*z`Bc)_ARxMZxlyOUvvE1}R_ww#`J1t|4F1BrA-_eeAPMJo(VFMjGt
z8;eI&WRSnHskBmAwaAg4-VQr+?o2JL<Qj62HT7k^(l<>W`_fJT(Yu_XuA|H`e3r7K
zd+v~^(lm*|0*v5>&5uZUTg97;W^A5D5+{+bbdN44fPCoQId&G0c5SiR5=PU#X;vNx
zU@#8LAJrj>gs3tQMoIKkA9S5k2N^8BwSZiz;#trLjUH-RJ#*-ON7AGp8g<mC6SQFC
zW)34ExF@xVaI$CIY1!gjsR=B`&-B1@rmK|&7T;ThcK>eK2;t$!!u&7aGGY>E7y$%0
zw#j2rW7ja)%GO*(+3qMrx(Du@VU&oadPKk@_d|ehTA#%0ip>!Zfn~lyrFBx0@I)gO
z;aO6xquwXKm;JTI=`u*LF-si=0#j1c%7tWMK|!J?(S5gBAV8nO<)7X@zQQ-?x<$em
zhvKsVh3;KX093Zr-_iO_o5_H!v%Fzpb9nx_02LkigxZddb|S@<8~aU&W>Rl|ZgF&c
zxEYQ--yjWqhi5Zp$bqq4l@%FIDugWq5n)XLllt<_7}ZTM9Xi#OYkjMRR%&U~f_0#W
zI7tEOugUDgnJpt5bs3A#`RtISIV1Vzg&31Sz@Etu@^I?-ll(NrfTte+f_4GK$c8On
zKs=Codk)Mh&C3cF>zqG`6xk{o?yG-y4`*r3>;IbKL?pO@9kBq_od5MFAt9`8It-F)
z+9&OBWLc6WtAU+A2uM@bF%?nbv^sNM1c6(MHjZ@-Si;ZfnMXm$9?1*N;^Nqb)dM(n
zh2;dxLl`3b%ylU_L&(AigjB1IQuamH!j(ya<Bubt=mpFG<e=0PQ4EIdOy1?!>#kP_
z!<j>`xDL`dFUjPUA{QaOX3A>fG{)|saxBp8s4>^8AzE<uA80Xe>|mu(-?<(ney5?0
zt0_!okx}!q!^Xpsj8<vf)y!#c+2^#a*+j9MNhxi&RrO?~{OV_^Q>_KV=k*LHa9)7F
zMo}(F2#K?$mp~>l_`4u$evbk*uv!SS4AaNhk5hT8%d8E-V8XisvSE-G0b^|ety9e~
z;GeBwRgP;L;jzjYZj^LrD?ylVsqy*PlV@PN>s_S=NsD3+K&3_I=r8f%C7zr)eO>R?
zNW&U!1umO>-6vB|QDJ3-F>ng9{B$KVifAkySwXwencmmK4%=501#%;_50S_ysr4+(
zU(O|V7V0*~y0~^=D`a6P(<r=j2u?mYpOS_4vS84Fv)mO>)VUr$-gDhJRmvK1A9}x5
z{FAl}OOWcl@sg8X+%xGV8NmCcR!nsCIAEu%h3-AYAk?V1^WE6~fcw5UT&dfB>E-^g
zIIhPcN~mAMmN5M9(ZbQ-OrYq>noI;TUxIBzGxzgwmGZntDS!7cp>GNeKP0nso4HV{
zP0m~~2$xp#Bt&p`fjBAYi^SvL`b~}g@!WW;WZ|6TM|Wntx5~e@PlS#+6;~?(XuoDY
zT44X&erHo;C?2#EkJdSr$Ms;E2Y1!t6F<J{AI9-O8O30^vmgdKCT~ktc6Nf6H7S21
zCel_HLH{b~@w0N~D0<6no`^3q#zmK>VQ3+b(pRf)PvK*ugXh<MH4d21qk-iX_0FOp
z!~K}8#iQEHi5%jd1pcY&&%aE}w^Omz;qHbgjiQA6#ItIfn3cjPbw(fhkE!g*_ER6k
z)7jq_pd5c~{+@)-th8!$7qnCK?{s$JV}1NKGb;)9nYhCSNFxI=y}1$tBYP9?ClG5M
zUMBUVWOo=ZYD3FKRjSMhH6||=Z2gIy^F_)hQ=K~$QeII5pS(<4hC&dbjtO*U`hHa+
zEdl>F2P1O7peZs^*aPb6`dOdaRtkYabw#7Tsg=Ag;daHFSkdmif}M(LXBDzgfkw;O
z^#&ROAh*NYzydH-g1mUe1ppn5W)7#I9mKiqZ-Sg!eQ{#_AZQApu$yKak6}tm{;~fJ
z0`n=K2x*%03Z~EU&z`IDfddCxHhLfCkR_iqdf$1;18cT2p~&my3PO3#1zNNLx|NJk
zZdTXa@&=&B?pn{Pd?4a3c|6_zv>nJ*w{??Ue4Ea{ZMFUf#A5ufsYu5E9b!$ZYdRjb
z{J(|j=PQDC1mg7RyiH1_+tLX#u3ro5Zrn*A5?V?{YDFuBE(yCkaO`=)79^W8tc%>@
zLpb2458mK=^Na`cyL8_kj`zc>0jCNx56UI;(}arVMTtrG!eVTXXWG<h@K+86O1Qlp
z?cX*JC-2vAwY|Qc-W>56qijEIK3^w_#`nEt-`khH+P)tz4=*Q~bUm_5b{f+dY8Woh
zm6FFJT6g&1rtYN8F)_WF#!fYt({$O+-j4VCf_j>`;o=k{!C{Dn8l(+oEj@PEl*{P0
zgvRQG1<}!{Xg{CDj4{kK4`ci!<qnkoDNN*=#MCSp<;66qmj^Z)Rjb3t#K+v<>AZ~w
zT$<NDM;VD5=%Hy(rXUjZ7X2gY(XmcZA(cP+ExT%XwEzf1YBdNjNCfGKkP<yu<6O;|
z9qXzp?rJQP4UPdHLfxIcy~PMuu}{w0BF}Y{JKer`9i*W-<=0@2-@>`iFEh<t3h#K~
z%m`cNPQ<Hn`(wQ$E@Cy@`_cqPq>2s7!PYjaDOo;N2JuYs89V2jHa==>na&E?5b)WI
z<_|t+2AZ=S4doOS6Dd8LD>LOy4VhB<2A(3z?O*YZ8n4}p+Z6i55v2DHQf1|6=RwrJ
zjq%sU?Oe_30nJ8y25#gOMj%||fpre{bQ%LHe08sStr@shY5WC6cuidGaV8FJQGF6J
zxIY|G(~J54z<wvq2DBxSa?#?L{FC*S`=29=xS_YCDnh7CsApeUeSIgEs3(Q_{04&v
zrw!~CIsF2aPt-Z>9TbFb^w0~*pGo1=OEofpFZHw&4VO6o(P)W|J~mvl9y@JvY5!Rz
z?~p>+Y_8_|EskBXS7?Sw=H1X(wVLovJS?`GYtT_lt^S-EZGhpbYE@C=RX(n=^YlVV
zxhg}F;<A6O%Myi@OB;#rN_!b>`BYaHK*V*Ou7$ju6~&?)=U%{`2P>mh%ME4KlV@j4
zgb*!3k-=>4f?Ogb7$M-y=QD<0yb}sYG+`2akjoRGr+Ls%S`p}<we{_41&;NmPJKq0
z<&)#d)jNZT8e$;we&fmaHw^iKhas+ufx?^xrVP61e19oM4$f~qiczeWK4&P*!l&ii
zhjztzmt9no)27sysPT1taPXra0VZ)A@%IDfjx(qB{g?9R(!_>JZHoMRektX$fs$4E
zJmc=#G-6&_OkQ#%*0{Am<mE8pt;<PVURl|foOO7^#l<n<Hx?r0np8j6Z_`R9Z||lx
z1mZ;j?EImj!yQm5JY0D)kAesu)CdEs4qU?f_u{jzxf*<gj5txN5#)(}<M7)&o*!G9
zFiF&L0`<HFk<q8w&4G_ynsUnRW|Em8dUE*qzQt*UgakF1DaiIrRYax&VJ0m+#$5Pg
zZmAlk4aN|93HnTxXAqeglZgp3yD;Usx!}^Odsx=70-Q{s++k?3MLAu-Pa|!C`mD1_
zTdc+La?aG2M0Yc3Qf2iPDI>yTrz+be`i7|`{tVv)_W3Bmo|sMN6V1K2mnZ6!M?bv(
zci&2E+^vKuNK%D6(4$jE;k`<>SmJ7}E&pI?AuY6oCOdP0o>J}pup+BkFjmd8k+Rfy
zHRR2EL7b>$g`!Rl5sWg9T{49zk(F4!R?QUl259>-(R8P&osE>T#Z2m!5&ht7gso-U
z1$F_Jsr&w}TJJBGDDq+hWoNb&cu~}5b>eO_p0&pvq4EVOL|wJoj6#`X58ovIfpJE?
zFqu=F$^N1-w)f>DL~Q~NOFz<S{*wIek|GZN{(N7^H>kObOyP>m<}+1OKZ%0BMvDj?
zUkH!Kc4CA4GeWR_?dSGe2xU-HA+tCUXaWNKwn7EpAQ5}z>;sK=B)E}^%Z%*W!hWan
z*lU7U*lkQOP}QSdXP@is1wrm%s3q94pjQF%l_Uxy0ki@lW|l}5vneUg<RQirM$-gL
zfmF<C{Nwv|>fH~g_<~d@KOQG2A#C_(KLxR(6v`dNmNtMDMnFl>sR^YEdOu8)NEvz=
zobD}l%}tCht&4Ids}4gr<W)$CONdi)MDS}BF_)7n3`xK8qdvpc4pu=or+-ZwfgiJ=
zfixG@@E4`tFb`NB#lHy4XFvZEYng=3%K{khXkf6^QX&Wxy@uu#PR;=F+)cvG!l**J
zMIwJqCjMz;NIqGX!e~F~nZ%6g9Q+}CY7tna$EX<qFJA}%F7YewMfkY30hh!5Ni}Q{
z+f7z-d2mqD`+YM&6<W~Q@f1$C9<qPIhxlyJyj+9TMq-m7T;~dZ_=|i4NP@=WbSsHc
z+!z6(CN&!FrBna#4kFM>h(EbyBtC~{gV2%#)iY)Y%}UDMCQGJ}h!G0j%g&o)*8a&w
zC~o_Ql8GaNpV3|0s(e$3V5qtmIPmxd&8wlj9J4W5_qX*=D=#kXRJWs1+hwLe!pS4S
zG4)wLCWZi9fPpU>3LxyU4iGcX#=*{Qgo6g1R#9jX?BPX8&J$q<J|WCA{hZD`y_|%Y
zt5PXx8FyO2m-dzjD5JsSI8@1#ZB*|vREZL3O`qB0>aY4e$c3tZdHc5huvn_P1@@zB
zFxhC}yTgSKlw*2GECA8(WD+6)w04zG!82Yc4r2CxMJK<l1P`)0ndsE@z|nyxN$fO{
zBsqd~3ajNm)`D|i#c@}4e!m%Yw#*HlEGY==(6#9<&wEBQ&M=LQbDLfr2LS0)bmdqE
zjjH1FG^PvZUMKZgQQ>~I>qTMhEC;8>BH}s*cD%MQ@;18H@)OjBX|BkcV=u}y=*@Uo
zZpeOV7mnFQRB}&G#!C+#0qTt{ILm*oukL1uT~xj^p|*sbp!Bs68pJFKBjx;q&L7K2
z7L#H(Tl9t0v+V^f!RtCESD_0USY>Vh5RLCzYx04%%Y;Qhas0v14dr=pc&~f0bHKjx
zZ7djyd<*AoZDGyx5>nIFsvDELH-Luyp!*6r?e!n<hl$}oQeZ6qJN)U^*#3n-2)?hn
z{jG{PP~h+&;<7ok<#gS$8B(|0i=k4aLJD!g*dmmK$6TlRzMii20H6~M+Y%cwc>G>8
zeqaBtZH;rbNVwl#E-p7K)odNom8%xgWo_8Pm%=Al4%;PdR>wz~e=XYE-`C4n{gr0P
z<WeOfu#8&;w%_mXp_`enk;Wx%lxqg%%yPTCULW5lgQ2arJ*D@(y&sd)u6XhrD6gZA
zswnA-<L;hqQaKqOKc24JTRLAJBzx*PVZxFl!Q#lZX|y5>S2cFOW4)J?!bjAz&JOhB
zCl`3VuL<=b!DR`ENu|c}4A}XT0ZpxphTZBGGmE|oi_f9j+uvg-yEwih>z0^!SKBo%
z!1QE2geu?po7ot-WRubjRg~uj&aK}>rk|@*NoZN-I-ft&Noaqzv=Ufxd0qIXr<xc;
zYxr8KzuK>FufkqBDqTC@4@>3hmTJG4SlBgMKdis*Q*~t%C-KdoGw14lhz|dZH0G(f
zWlW?Y)7J-<dhfbUjyM%^Pz$GLuz*<U6n(tQ-$Fxg>|V^jZ1WEmo^JtV$M8f~K{t2i
zyeQ3D)jblOY8VQzTG!*q*i>^Y**G^g%WQ)IS}wxgT>r~c6n&lM09J6??cc+olpC$S
z_XF~svLy8iq)dpdlP-3ueteA!>u0V(I!xGY@qcY&nIFtE?Sug{gU?^UukfrAMURHl
z?I;aD)6f9qmtSZkRAKsyo(|1jA0Q7fl#tA}Nz>&+(FkO7G;1~GQ$&Ei=6>{>-C)<`
z&#Sbeoh^j1_MC7Hnn^z}Wi!9Hk^9!;f-Tved!5*wT}OHL&vsn=b;^j-Qs`xYprOCx
zkF)|eyT0r=#U>RXX`n;}7PeUeL`mwD<jQ<O77e&MGp=;!yBgb_p6Tza<n2n9wYS+`
z82;W4o5Cp&Ge(eZL3;qWsxohIg&&~R1VQBvw7G+s!71e(ZbMGZiVX3&-N+0F!AO-+
z{aaS6-+KLtQ0amZGnL$5daf<ia5F%~jr^hv$4l&f_R!`C-R5mRGr7b=rl9CNT(?Pe
zIlYxt4yt!d*Sfk5E|5^p$JH|D=jB)wdJG~gyMkyA<X-_VK?doRC!GQU8p*|HUBSSC
ziXOc|v((!w*CyxQBhwttaAlTuFRQ>XHvstf12~+0q`()Ue(J~?@V8H~R1!aEN%DTi
zfV-q5)WuVaEw<dm`f&M|?Pq8hNW<yG(sFcWB%#lLv#+7n$cr89eun>1>HGApI1MuB
z7_F88YjwbzcQ;hyj04AWpbi{j{$Zz{cv>5VYs4A98}qHfoF-5O(YCUIQ}K9p&4r-S
z+g-YOskXlQhvL*MIgNI)K#Z*`JlgpTe0}>q@0&IPE#!SCRwb>32QGxuK<ov;2a$^M
z0CEO8+D4(CjzYkzAoWFgHbgQ!6BM-NdOgT}7!b-l+JZY6ap;qiZkanSanr;1FLh7(
z^mUE~!GmJw{;`L5&0`re_aQKRHUVwTA29Km0-#~Vt&GN<_0GF(kVGB$fwb}h^k}L-
zVi6P9PYiWFVPe=6KJhqJpvg!?M>td)NM`ye=%^Ch_pQ1f{9P*(jGj?)5~06>K1sb8
zKQSPWqz}EXhG<^VV}l%hXoI5@9@}ge4i@(eUXH~L8LJ<1D|kN-Lk11Bh)RIWYc-cU
z$+L4njGpYzb(TQqNSEIr=lGwjdsgpN4nX)K&#1_=Bp)Fpv);#`o24t0*Lf%k2atBs
zrl+}SiCt#wZA_hqh$TWor$O8-+yMzOwbj1jr6G_$k_9dYuRl;CtsyQ30T43@r{`F%
zvM~Viz=Ua*7YLdv1j5SiFJFQePpDk&rj^X~EkcgiU=f2*SQJ3h-Pf^$hU9*oR1i?*
zz}?PCPaYAoX{wrdiIqNEq5?IBD8u_WA}kM!qu(STa@m+p@3Wu_oA|G58b`8#f^$i!
z4h|_o$tiC|GDiY_=O2LQRZP7>NW7aZ#Te|GFVEt?mnr=<1Q<+yd|fg4fe}6-z4*Co
z7B-L)7}`0wVOL%%70}kslYFaloR@dda9s*uJCz$t<M81Fe#H6)I5sKLZRbwxf!_bc
z*gG|8(r{gvW!qI<wr#u1W|wW_F5Bp`ZQHhO+qO0RO~gdZ!E^Bbh0KWDEB9LK+G&}Y
z@|-*ATS&!~KOjSCP6~_z=78_EJofYza3D`wOfeNq<20ZCUB;s=sN`5HcyG@eFHCV<
zFyRXZce|D5j)J`OX9SW|V3KaB6;|OMjoE)B^$EBRpTBm}`Y7l~r)ymuK5<E93C_?H
z4Jf0bUPMlV<$dRPlxLT4BW<wR5oTO1v^2Xf9^FsjkI4THnT~Rt-#0f{A|&_@`&Zxc
zlAlNe=~GkIf%_a1Vs+o(UXW7KNZnIm*}gY$dkU^n_DS@5noII2NMA#GhTgRnN8p-e
zscRg#QL8gS?9#F&3{5zHje@czQi9Wz>ksb2ygib>I?KZ>f%|0A181F~xVewXS0LF-
zL$;<tYQK}dOcgjwUQ2_;A(W3akWk2^XCtkN2>tdOISl@7AY3V<Nb1M5JTg2hC@Y((
zlPhgdEz%fpMH%Y?WsJv%QzUa2W+p=w{{H=b+f;P}|24fEg9r5opM+^KUdH_A*VyzT
z9=aWZg1-)jjX;n`b;&b_6kYf4-Eo@of(rZC4PN9h3F9XX($Vv%3i_myJI1DF3QWG{
zgMqU~SUZKxq_!U+u!eJB%SC!Lpf_#BRY3a?q7+|jxes4H_~wl&C%^TFQAhT&8`E_O
zw7K^eIaSPlSkD!NIj`EFLX2A|S4SyrT70&x->-H8=3GA#79poSifXAWDf`5!$VZk*
z8;fxs?IJ^46)u!LY8c8@GhGm6njSC~(md8|0<>Q`><hFLUOJ*>3soVT!qB-zdJ-TG
zIVBaNR_>QgUgCxH4~y|avKK$n_Q_5;dwH)7G!pSLkcM|S)bymn(wc^A2c6ZlVFHQV
zrO2TmXNAFd-qx}mhzmqj;K(tkg}?A6>Yq?XyG%npakxkNl^C%{2ck@XqEWK3VD_zh
z4L51M6Nvt0@9@Jm2-V#;z9=H`v=Yl)&^V+=X85W|Z<(z&q6UZV9#Qk6N@vL|6~jbR
zV#}5iCm34}G||puWoYKno7vbs+9?ria(#+-hp-3Xab3m-5zSO4V|7WvOZ5y98!sIG
z7LcJawgxadwu4}m3CwnEaL=}|@K8~HNW~-}wE6!Su7NOx*Eom4i%9QHU|;)o9^+fs
z?@{HUrE{`EE;_YAn`%u0J*~fJt2li9tDy+>t{^tr?<~WMm2Dg~Q0L%oip*dB{>qfR
zs!M?Jk9~S5N$s}~OS?X2a0P76nOg~@E!%$h$a?37jVS1<bjMQ%W&jTplnJx?v@y5^
zY)pM$oNQ!lRQQ5ow@DKGmsI@!r_cYtbU7Qx|BO^h($sR=Xh!jwshK5VGs<g?C1lKc
z*e#YjV6~a%ymps-7nE!q3?XJCHW~SPGV|BDC7yHq5C!@}i{bg^in-q~6*c4I{;=?s
zL9h*NYnH4@_jLJfQ2a<6&UWCy+@UwqEgAqlbQ~%AaM~ikKfS=&-t)Di4TY?;v!(m-
zHswWxivRU;{k*VchTHSu92)r2!U0$T*zN3ecYTM^^mxC&9Uq*$<-8F1xM5c8jX|4f
z7ZIikXpsLshNn<8%hW6|7&QZtNRu0)MGgsHoxTPL{v=U!@}b3usU^=`Q=(U$*VC5s
z^7Pt)q-00b56&@<+_2LSY*PMRr=#ODJ&FstVy`ea^QK)nsCufO^7Ga37H4-spJpgm
zb=SsR8r}jMB;OMn;5C5g*1c^X$1Wd;@aKND23|=}%jf1l`M8Xwi>x=#p-4PwTzH64
z9x7xW{aMx{3}^mKd@43yHw<nV&|an7=^YVuUqEwl2^K&@EYR^J)yAx7=0_3}5m=I)
z!-(OknzyK^91G7SR|_Z}aaFzPkI`%z7D}>Xo*6znKjg;wmlnOy=Jbm}%+2uA{KCL%
zazf6Fc4?gG#_Txw>#XE8F=42+LMn{_vu;ACb4Ik8iVHSf{Qy}~SCcuR8mDd}e@ZvS
z83VaL5@O*$A1Kg-1?joyUNW~TZq?eYx3xRvwz&LCM-5Opvc3^U1X1cc@zFxs(l=`w
zkK2IZOG)T(jBU5%Lk<cC{2>%BfuCQD5xc35g5_%sn%|8&te>XZM`K5)^pN6?U^|0S
zn|5{%zL4KAY$_sxAc^XhDoel0EA8&19A?`u3|;m+GHHGn2YXUJ9@0Jdv{Mm^W;xNI
zFu;@s_bCb>`&-y)2tSXGSP0F`9OL@L%@Fu5?*p1+=j7`NB#7}y2Kiek$VyuLNBc5@
z*$GCfXvG<3R`Iw4LcLyO?vPWIt(e-feyiKL_;x~GtYb*AuHfq<^VQ!c>4j?Mm#AeX
z01JLwNQSQ_6U#FzXQHCgkXlxf4YDM1HTR<r;60kHP;rf8|4BYVz^n8Q8XV=5uVbWl
zFOB^L0vz@yVGbRyj!dmf>WlR?MCc`$q4fp^&IL%Gv!RmN9#0m^Urvz#S<_84Po!8R
zuEt*J7#o@ks=1_-7IVTYgYS5dNFy*>DIU_AC{-3kG7KJ}I1$HmS`p!+9w}1<hBw*I
zS^t?9DmY|-sSiG;S<Gh?fhH`a61UZu1^9<zI0`&63AeXLZE(t>_@LC}u!^hSx3D;Y
zH+rHH>fhcE9)f3=yZ0NJu%j9$MK4R-1&57`5ZMs`jpSFRLD1l+*zbW|yG8CzEKlZD
zXd<l8eIMa}dwY%qw8+m$)|VQ_jN~MusV=7jYyK_>)Z-Njlce~)^}8Sau5XBbQVfcv
z<&+kz&`5y9AlNj7BCB>L3lv>75~*?gg-(+NR(CZKsiE-{0VcJchFwbv+N}Y>(zQ{_
zc|nc^UJ!_1c(U*x=ROFqV&A;)$BypfY0?|a2)E^E$DB{`wrPK6q0%@*^w+8cU*JjI
zxd$La2@3hwR_4SFJx4mjF?QLUoL1GT;PRAo&S4Q(XWzacO6~s1;fN2Z<kU!F&%`k>
zT3{U^SwRNN8c8a~(-5;7W`Hb?m98|y=eps8a2ZHBjVUj2U{^s2M0hzvq<N=ZSX%Ok
z*?F&jg3BZyDhrRubS%_sX@3))#dhG<wifrAJh0n<ZUYIi2Jv->XvGb@67ReWSX+ut
zBYs>#0lpIhHf@@MO$=|*!bDTRVAeud>Z*&tcbHqapxbLR?R4vI_pz>GE?K>CrEGL5
zUL`cR0EIik4Gu^o^HtJ1O=dDN2dTWEKworj#eY6zfA|3rInswaV$mo`#CUOrhx(F}
z+P7XJG<{Slj~?lHrtb^rb=`nwlImX9-vPz|=$Guai%!*DGd<=n{IB>F9=(G!LSDFS
zqixpBf#ySA7bzwL{02TO@?&z8@tU<$y2>UE>{W@vMRIKX2sEGoT;EB1ZynYOqgI$n
z@1v`~IT3L9Ea=LTOiAAR2%ZL5e!%aFad4`_aNw*(Qt3=9I}8^J|Dfm*I5*h9nS-^i
z!T)~QeLa6d$e_aC;~DKuAZ$5JKlkbx5w-~r#kJT~$}YXj128_wAF?mRpbX_o)ij{1
zn1Mf*N7@yp#jl;_HCoAdkD)1I+J5%pt5GMEc=(Szdx*d<k$gn@Nx=?5%HgGFz?Wz+
zB*_}ea2lS&AOqO+{UZMW*qUMwrh2;bU5GSMBJZ*Z_IP&$cH2hAF@LI6dKZDLRJS~C
zUD`z6tUqO>Y|{_^hIz37g!FfqhwLpBN@`P{P8-c}%cGlC7yLGENqPTjLl?%%;bjWF
z(cnf06R%D}M14EGl?hTzO7LJPWU}wZ?q)WU5bpqlt;C~qD{6a;q|3Y+R!-1L(Fm&^
zb+%rKh-7P1ROh~Ml~|i>X9t$8EWg{4`p)Qg=pIBG=$^fQZe{9oO~~`bN@x+lH#n<x
z>v0DyiRE0o_t)7Cn4I4V=q5TVE8vR^Z)|U`?3}5<lm%fiDMeg;r!f7HePz?}XXgZB
z+#7t;4B<@^$rV%*WC;6keo1`8ky=W`GcVzh5M@y!^L1hbD69-oj0d?Iq^=jqXvAk~
zWAcLbV+ibbjwX#WUUdRmPtJ+`1|!;ZZ5j^-4%yHf@*|t4XfS0+Pi3>Ab4n9|w$i?G
zxU%n4<GyKTl$myad*$;=*d5w!B>b^gfQ%qD;O$geoFEr)rn95mqcCQRaG}5bl=>5<
zCkIQVjTUqrzJ95-A{G1x+DY@&cfbiWo6Dm`R|8%>e57f0Ml%DRz)82NRa*p+&&y5u
zI$J?FZ(mpcvjV<3hH(he6t|#uRn-|>j~ZfzqZ_(Voj|XSD}>^SgEp%xYfgHOB-dia
zzTfL@(`8Yl0;Ug0Ku!wo+#>^rXOkZwSxtdE_XnfoE_KCmd^Oj#AhT}kV0XP>(Rlid
z6b;F=OlqVO_Wb!Q*hUx}b8fzkl*1^wr<*+Vh?XnKNN6bz^O%Si_mgBad7Tk~pKJdm
zo!J<MrFu9+SZZ@VawL6I<1q$(uMBA!N&};$4}Pj=Zt<Wih&%Rp*Z$iv7Kc2Z6Derr
z_%<J*?Q~TJ2Lfr1e9W4NETOC$4qf@_Ky!Xhkdg3X9Gyan=U#WV7-K8=yp46v0F&M~
z{;JSFYHMq|Nm^MwToV@{`cn;q|2z5F_YJO;J~-jh;Ewj9WD-ldK5d&@uXt7Rr`$99
zF_brvZ|aQ%(|!ydw+Zvq^4t}=!P{WAQ$a_=7UCR3*$r2ujtX>cz&Qs*H>vpfQjH{q
zDfTQhIHJ+3UBRKPVdC)`9TjFv@ywq<BeRmO5JLWj_45O^p3;^99Ti&HyNSt%1z{)S
zgX0HMV&Br{KNKeGf1$_OI2ixexLebo9lL{naETecz2A_lwV!m*q4@S!SGF4$IydaV
z&v{z7##(=unu-?d6OMa&VsMp`(T2xBJQ<K>l@9GWAEr18Zr84&X14Xb08V#PJXNOP
zG?k4F8Ba26X>-XAHl`dEYc)<5489$2XBmJ1%s@@znKX19t5?aLpW7DSoz5?NcQW`K
zSXXxw+8kfEkMDtviwgqZH_WG>hm@Y4uebFo*S>-0hKknNq)83k)mm+>Dh4tBpTRcm
zWmTtK@at!n4hjo#VkfmjjlVNNDcM%Z6^5=215!5?SJl?H?XeH_nu-UZ9vwx?)jFzq
z?yD+`8szN1U%Z=m8$bRwr)(l5A+x7Qv6MyDfko!>8<=9Vq}|#H_O$iyY4YCPoWzhd
z85nljuEl9bmfPBvwXLTh?p0oN@d$UoQXrP_*&;r9x7AFD96euvhb}u4r8#`GH>tY-
zR#zeq@@#)8eTjIlDp;u}<;AeZt3aXU86#I2rEqFbH(pMK+CoUvYcc3HRQ8*a9lo-d
zE0_$>ahjzyI|N@0VkDb1r|ii}$!X*mwvXz2$B)q#l<7GeDN&C0tI!8!n5I&hmPqEb
z?i=VbDIr}iN9YPTGo&!*ENp306&=fZQZ;a0BmY?)o;tK|z^$-HReu|Vm&Ze4Pdgh^
zm6;)jeT!$4GBx;H!eyt_VyKZ<3}8lH@1J@SG>u!8HYdMTOKtoseq4wTRhSy<p82@x
zV(iWk6C{i7)h%aZzT5gGdUt#;Wjut-Ne&~4xJSTsGsBlmW+ZpE_mvZ0PpKC6U2lsj
zj8!a$ip52F;J90WbIt(wb_!T(_HL!k6jU#-eSkD$U*_D=!S9&<ss*#G26#Kip0tA;
zwwZPm_d&&1&vfKpI1^&0Y%fm0!xMCr7u;!Sn6lf&0Ms|myVw8Tfq9%57Mt{0(@Sjg
z#7HcalVd$ydz%Mlc&4E=XJB3$?{*YU!xZfY^~_pM>v|P9z{JZ5;rD|SC+Yw;Rm-+u
zqD{Y1ET1-(_Rz@q+C2a*RpPS}iWQqF9p?&H3@oxM-0;PWHJNvloV)=vFn_*7M>;HT
zM08W?SK~}5TLk4Id`>u;ExXGqywB_`!eMD!7B5_)WA_>j8<3tWpI_1tuM@~oR|{Mq
zE{&qb_rQiE61Uv@Obmb^)=tr3z*p_jU@|FZLIZRWfxuXlxhiKuBwW39L4s7R?(b+i
zj824oj;Z>pE1F-X54Uc_#AY_$yDdv4UqH<j+Ol`7_k;56fO@;Q=Li<l#uEH<yP~qf
zpC&9a)v$?^<4CF7RMwhg+ro-p6q0gUV$@wgIsB$;7h`#&@d|C#mUMyz;tcrq_;u{F
zuUj`oUnFsWbEzO%6$zUdz%o|A=RcR?8WGXyO^L(!F&W7|Us&JJ7=Iu8uP%rn0OEjj
z4dW+>WF(V%kS0PVkV{4O4c!W2sTn{jJ{+zZFT<#|sy>Q9=11Sm2>*!a9Od)jm!|wn
zAT8wp7%MErwy5wUuRk0;E2kG$;Y!O#9J3BYQ4m?k5K1%|!=Rym)dcWNH*g*}r9-~F
z3j_TwR5jY$p1ni7ZG<m_p^ki(hhz7^7RQiptwF(9#@M&qqTTq?>0k&X&Po~NL}&dP
zOp8aunkAlG+0$9;MOGRV#Gto51^%X?7ry;{fDl3Czr|1zJ9mp@fgV%1G33Lv3y~_Z
z2?Q^dySaJo=BdTi>AW>ET|(?b&Bq;tSIR`hgw!%xFnm+teG)Fk^y6m1dg-lvjVP#^
zft_pVo?CZRT$N(5EwZL)w8DLT$uRLXICW{Cens-79duGZL=N9c*iE2UGqCQ0>FpFN
zCFNHY?EozO>C^f2sC^&R3Of@hn`@M)^OkM%g`MZR&`guw&d)xfAB$BaaiY%Yx_Cq-
zdWG8z!*_ErlwV+>a|V&IQ7UTy8GsWzLaT8bV+)!!RfA}9pe^Ewzj^>I-kGqw*2pSj
zM1K-_e`YTT7vjV~P?iw@`A%^%gakfNONNKoh3K#l0<nB_A3NWq7Nx}mXFcKCQt`<w
z5h~^3yz%bpWWRFZQ)P+haAD6K<6wJ@S2xwHZL4j4JoIU{2@d~E23!;vdkSZ*UQWm`
ztY-Vu(F|MDH1dU77Je*sq%=o6zv%~8JOVEZHy6vHz+&9_rEuePJrWR$7g$2*r!E|}
z;?Ev<9%_c!e23%8GLTLmf_C>*%!~Vy4Yofb@`lw+?ZD5ZH^-p;1%^Y0Ri*~4>G3)M
zk8eSxpkoHi_i_6>s*}_7$%GWslY6F_7aikt_c|UdM~~#+dSisT)sSc~v9pGnV^f6+
zKb1&x*D`aW#7Uu$tmN`z18NHV)Z3uSLyAVtB~5CzP+$BJJ>;i&Sx7uC=l&BoI7_0C
zbV1u!k=j>+9xMS3T8h~~J&ad4)6^l2zK+r2z3|z{k%+a4Ej+1s9Zkv$&t`nZQu*tr
z@r}06Gr<oEG!7g@5})NCSt@FVltqmNwC7jX>z!8|W1z?2D?H+3CXAmpIWJO{Pz-C0
zW|mP%wMIoF-i-q%Ly};;I%k387<#}7A%XBl5}Cw$2^hu}7}F%W6HIJi>mow#&hHUa
zKWNy7df-e*)wwWLh9u?&IPCm%jj=-V-Uk)32hCdP4;<olCRDRuB~@M;Tw24_m-!L2
z=(I4bltAPTbt+RVCEN?>SdA=Gdu_vyXc(ubXMWqhKGpf5(18}nioYQT-yARSrOeZx
zLx0e2pO5!rA`|?%(!)xYWk_VTujzfhqG!~woNbF@QTSiWc(iF_M|37#UA^xJ-eZ~O
zs4!61{UW4O3$Ve1sE}u>B(t0oM_ZNZe3+8PD9R>DbyO=ENlc#TIWTl_$mNzUV#>CV
z(weY~|6wyexYO*E%sE8(sS3tfRIlt3Fe<%<v+Mf3^EgW^iS$Ssy%f#aP&}+9=BcsL
z?=4cNFHuE6c;Vrw9<n(Y3+pCYhgcD43c;5nALt8ELo<f)3MK`!W-1diK8rSy!eA)I
zG}TTe9Y-O7M~SJ?S}`dn3kZ|SocUiqn;lf~->>}DM-m~T!(!Xr#J=RwkbWN*R3V(*
zYcXVjUAFBqdWFWqcQ6E?)!~MiuJOim%2VrMMJ@iCa<NvZJsr4b+yg>b|0dC?EJB(=
zlSe^<bGfSKwzjM_`RSUHcP(PKls~uWr|@M}Q5q@Utgh*1h?$UeGkHw1QTk^|NBNZ*
z|2x&Yvt!ugu5Q~B$oV|Zodu#bMsCt>)~8q0rzhK|hj@4C65O5}O6q>#bHTtcFPTig
z;ICo2U*Y8tPdgDo)gAK`M<mht9GJlM(ie_@D&uQkKbbvxFOZEKoWuC9Cq-1xuS+x&
z%9reddZ8ORLQ2#xEhPsvXJr#=W@d9zW3kTisxrQI`UAO-y80vF685R|U(%B8KYVTu
zuKyX0QS{H}{zp6ge|_!(Gg`D9FeWDei$(ejcLL9d4M95pEE;WHoJ5Yq+Su1qOD_nT
zq@trOcXpj1k_PSbLyL5JB}#f{+w0x)e&|Y6`XExm9c^2eyl66I8Zx!2V~327JhnlA
zHoc1Y0%DK0o@Ls0_s3H_Q*w8!`%Aw-PrJ6g*n;m?=ZD(^^hk}s_U`UghdKv?t!}sX
z%l#1BwAbtFLdAuAitpw>Y(f*-Kmt0Min;zI!DN$`lMrYHG)dw>7&)9;d{muiT_&SN
z{01GZ?;MDG!Tg8m2?X=iGKLwxt7_)PdcJK9J!ZgpWBbTTtu0{ep0%<XB&!@kY9qhH
z*kWmsQ+<=|qk1Qp^xxUzpsFc8IX;Yz^3!)6S<X^xjBD|NNWa=LVy@43=~d5>F*7N5
z&9!!?%0p&4sTuu+ceAAGTZf<B>V1StZtd0Tj4#vh1D?npRUP(@P*ZhhS8TI(h~I9|
zSfa>%vtu(CAq&>IxENf?0SXB*YuBcYT%ARelBTb2h^vW*wBC3#TWH^EG#4hK(3x`+
z5;#I=C#`kcb~l&P;Q`W#bn0baf2t!rOd@$|;M|nil?}KJpfTvL)U659xx>;t`LfzL
zXSfWoe?2e*?tm|NLoK-Wdd~lJwEf25R~93f{ahXr*&PTN^QPnYFj1|U)rW{5#ol8@
zEcwXrllB*nldvG*&K<M&(DE7prrqq%i)kJ@>8Q0+<y*%I1^=$3=@1c1ynvzoWj4p#
zyEGOBJQ0ODz*_(4aI;j_wQAKvfIr(KxUHqdP+->E`NH-p=DC=i!`OfT%BZ{9g?$Oy
zvZ<=4_zXup9F?Xd$;@|r)nZneNHv>@eM8kBz9P^wB6LH=?Oopf6z4-2VxqEpjNxo~
zVjXcJdddsY?xN8z+E0nhCCaXmfC}!}A3u{=!F65sYgp`(8cgPuM2f6Kbw0*!Tl=*u
zopS0^SGMpt+)prl&y@|PMY-=d-P|TVI5x;BgvTetB-1|_$7NT0wsz8(LKE_?+nU}z
z``Yl<z)J!Zg#YMLP<f844#esYzNCw&1Tx({FdjRJs-(sYKh9?Wi=?By@vS@v*jL5I
zskiwrM$Tr{RBbgntc$I0MrCjY?LRl9{IonJpX@Xg#~Nz(%w*c5H)!hO^_di~Qdp|W
zY0*zO#;zK#7HVZ7#*j8&TohhVSCCaX7~N@(L&vT2egdnp6*IgUPC%drO^|TX8&mib
zxK_TZI3y9+NQo5m`-P6T%!HI=pEE2@hnMO9VA~a<?6tO19~JnNCcA*Hqv<p{Q3g&+
zqVVa%SvQK-u~vdGdu}i)R-=LWj{K@|#>R_%CO>kFTWOh0M!~>PV4O~TEM6HoIDAti
zrMkwd<|OF{iK=jwNLw#C<h0X>G-nc0Q4Ms#om*N+C9)U4eHk3UxAAJwWM^By>4Nkp
z!ASxnD$;y|mrSzkni)_wDbVs9HRNxTUPF+1fsk0&p0mw{$9bm*aQ^FyabRVy)V3e_
zX75t&$eWwW0VJP}0)n_Dpg+(lk+{n*v&PWorRThsK-N#@S=EU!9_$0X#f;D?%LCNt
zb%i-ER_f@*zNu*;y`X1488Ul^I>A6g{x`49r>3{L`U<&FAu^DRK&x_YJ=I}k_~&A1
zM6ayOEx;sw2zD8@`L^7dr$-m~y%0H-p%S&m$jxX^D3DRDQpEChFnS{9Ng+-Zf)&>1
z^INdB03I=z0GNEwRAeM?JjXs?<Y_kGj3&OQNG@<c6<ACvb|EbUP{SnsNcR5DBmT;p
zbZB5mZ;odxuTmiaaKF*!7%Ull&fxe!34Hwzmg}NjCGDsd(VP5t((bpDPQa74xlpQs
zQT=6@dX`Az(SfQQ33IX)qXu`Md)k;aCChEX6o)qRe0t6rh7DFUnoDkm|5~eX-2f!(
zm_6&d(V+{{SZfD5tZJY4-Rfm$gJW9BuSiJzJv}>z<2Aw*p^$uQuYO6kTK>_muOOWF
z*CY{)02S{1iNDtn<;22acl9)#*bx$ta~E-^PjROYx{o+UAd_Hmff$`DvYQcRiRK9-
zk4V&r6zvte6hyqe_JmIAbZp+gcCjY1q>A+I^hZ2mrSb%a!OkG^{@9D?MYRdFY`1W>
z!%n<Z4fuWQ1ojYyPDhrk4ayo+eZ2#%gkMC`r5c7BrSLo4U<HTVS_}$GfaFF(0|A-m
zLb*jD(t1<64lTP@Px70Y3WH+vlLIjA(e+P>-16EaFrree!H7zNon}ZA#zhj5EzCZn
zLX&}OwRUIfK~}6Lyf;EV`;nzu)t52fY!sLywZ?~E%zIaFxzt>0`vJfI8bF!>tvE}p
z{gt9eXd=YbI9QZv8~>0AmuB?tbw++xd7A%s?$!DNLGRI73`g>n9<|wXx9^YFhf!Sx
z6ER)QUtE9c{QB5cfkcLv3|Zysz#~kiIj`EePo48aAU+*SQlZd-VM6dPAwRafS5MRo
zg~({{M(=`NJs-^^od#G<s9%oWn84ms7*B7@lD7;HM1!2X7hf0i5;>?;P&m%l38{hH
zw)@dr)?K1hh%13(ElZ}Lus_TD2hE!6(Wt<v79GAo1GTMv7)Ln2z3r~_hn>0zLEE&V
z+CH3`?E}&&ZWeF>@1WtlPA;p}z~eEc4dDT4L~Z*AY`X-ze_sbc-U;T>fj;s=u&&y;
z0rJ2I)k~tVH)c1`un4LT=V`FxnOH?5#(YSi;GA97)KK<Jw(!K&0r|-$QKAjccvy^-
zCn_h`dD!Mc(apL!P=a@?ReVYWwwiGInAs7{_l<6i)3`j1w#fN0oU8MFVy}y0)0$~a
z-Amku2EN4Yj&FQ(O_*fC`oqI54}Gdb)x(6OFpp{iFjITbxMuT$-7^TXV@^PEJPFu#
zZ_-pG_i@ni4t(U=;@$pbvDoj|6)1V|<4m;rzr?|6@!})ZnHKpf6cU?|f43Iu3^lrM
zzq(aJFIi5mJ+VX^4R{w=HlD10*N)hDHK1qm)6uN(RChR(`XLQ6X)5D0Q(;+iAQBPn
z%!`zmO~(O)FuR_{_%F?hFjylUe6;SU<5jdJI&7BQ0G8^s;3YSwiIM;9*UC4@iYJ4r
zM}ACXf6x6n+Qs-Pzq!!0%~Pwvhuw*-fYI|5xl=a^wS^r|zo0)93@ATHK3)wIr<+Qc
z7VvD4v{LE+LX4E7YVKKBDtp$Lxo=xueS@fg6?4oyfp@FH={JWQQrmtLf9^lNJi--K
z4Us@vjQu{xPS!|96@Qq^zNHF&%2Dc_shu<((NkUpoA2wO(DGXIie$K{Thx$Y(n4@a
zK=alD?FkXBu>^Isdcm*I*VDf~j|XN2iu99!UFK7HS#(Twq_eS)o*mlEw2#O<Gp@a?
zRDaCd!~Qtoic8~c{z<5EgLyGalUwpiz_fnw<o<|V2(9}1_8h6$i2TupQ`uuC0cvsd
zy!#+v)$DJa@BkrxOwen5n&P!avW=I={28`8v(Kc_X2pk`&Xvh0^`qo^9Yx{p5Lado
z|G0*2^~cNYi-Pa+Cn$%<Pglz<VMDW!&TNmlslGz(-8nA%QsN5SBmZ(Iu$NNxq3Xva
zHrDv=u5&1dALRM;rtYVb?f{jBySV8NR(&eia!Cbg3nNf<Z+8l{i{X#F6Cu%mrjgtk
zbn-E_>iE@Z<t?C({;myGefpTq=;S<OkO%sLyTnyo<K%&w`W+zoi3s<b2=YJVGuwZ}
zM44Ea|7ZEU^$+{IFOKAMqJFAq-RCBrJ{M=K+$6cW#rZc&e_I2SO&mnjxmNgIv24mt
zKE@PG0&7G2IayEN+~WPUy(b&AGMcT@+wybXD$}d5_?UQwnYM~!UPjtzQ8<;k>cql(
z%RXJbYat;9tb@i5c|-^!FSo$$Y-NVv``o2{s=2RLPN89nwi3XT)1EakvBNN3^f?o`
z!)=x#tOZK*z_x+ahA>(u&YIxwlh3qID6CV3>7Rt?q##<BJ>!JmJh-|X=$2TBJQF}Z
zm!L74I$e{%n^0qls)sOak-+H@KeL|adl0IG^~S5leUd<m$j8;K<`m$TTG&8(ly;L8
zPfLXk>4Xf^+HSu(yvt}h3JeDuYGx5zD4TJ9N(9HmtgUV(5)@)ZiHph*HiANb#toN0
zNP`<<fHZM45jCQ8gyZuJ<3~SNDjR2@bS5uf&U)m9rRWVcUIb^K2x^jgUcd`iWRCJm
z#ejq>7eJ&wA8b_Y^UD~Oi)ws;wsp?}sC3K>9L9-DE^)2$(Ks!IIz3sOE&z<a|Bl%V
z>7DWJx8XQuRJ6hF$eQ-J`{h`lN3jqk3(=m6Wau6x^uVFo7nGE2HiPrOr%K8ce-xp;
zp)XV6TViChP;gy>XXfG&T@wusN8aBH0!w0@vCe1-RcFEEwlyb{LIfeNtoMuUdwWT8
zNHi);Q-#!vQCY5BqP<)btJedPdYW0aC3lt%s;J(NFZ&K>0NO#F91FuS_gr=7RlV<{
z(PkGql?hxA&KSi6;Z>s!B0nd)svbX52xvOkN&(BKU5xT3Ne7N9o&X`vl_uVHkF?tJ
z%c;S}pfcyDw2={3mqV<$c3vq?;^wUU1#Ba%t<w4AP}DN&K>{IQ71kC6k^7|6@Y7vF
zmBbDef(ESfi9BA2o%A5aJHdc28ICKps2dZGWsU$b5v@GluzRX76Q6<V01YEL@MJGK
z{@`-dos*qU<&RtX{f9@`*F(igjkLGeDQNLZ0A7|wkDP}VB=xu#`^NNzvZ}_y*>#c_
zo@lr-LuHZ&^>gde(k)Q%QTk|ow535`94Y+uGEGI%ax%@EcMD{+3g1e;PWUAom2*2=
z@cf~x{Q<1}N@5mO<9%Sk6LFU~-O#T99`(L00$xr|aKwbs904hjRdub=W-H0BW~X1h
zs$wkJ_oFkPj}8o}kzrStj<gDyHCVw$p4_Xwz47jtaRSGXq1st2LkgUGq1u_mF@v3<
zu$C__BaLxGnd;m?bSMdTeCF3EqG?|eYR4k@@4ty(Ln3%8o<5mkh<zsI$g3s9cRV|G
zX5pn{tG}WM%xo(4No}@jgxzpgm3o?&R&Rh547Mw@RVM@QbXB9^jghRf%7%z(JxuR5
z<QJwUz4wY_bOHRwe$Y&7OA~ZeRnJ4{s<Q{i3O{9*hi~?dKj18#>Xv`BCh0G6!j@e8
zPzxn4h`pA!BETE=U0!FL3Dw$czSy&yHTM%s>GBvAA7$?>hr%_qe?l1$@jP9Tv@T7d
z#MzznynJ=ox4aZIGoDQ+Dc$qwfZ3wr>4}3Evgj%{479hK7uR1GEUG@#R9ju;zW1Cp
z!y`@?MOU563l+njOJMe1U`~T~b-d+)9`c|nAw4wkbm~OaN)<kzm^Dy@vDG(XYPen?
z_Gjw-F>mYKbQ#;5R$apVlEpTCN-=u6j7?}6P=gxoUQ}_MolG(Xw6SfQn6E$XmO{i`
zTihOP+py2&O9;Cyt17!^B$tmJ*Iy6|)&V6euLP}ai)H`y0sgQ31tTNN|7wl>t0@Bh
z?Js;@YIaGQM*h_l-7Kd0-8KqcIH~~XH{GpV|3X5NGL%Zll01^~FD*B45Q!3+#T{qO
zQ?Qb7{%L*s^gDi@G%=n#*<V-hmuu2}s0;>@lp;4CXJxYsLZfcET^GTC16Xpj3>lZv
zFM{tUOBC?$r=uPUVW=NNS9gOE26E-Hwx5`_qoqqW@1D-|xe-7LgDGZ1O;5+?`#!p&
zPwV^fP{m3D$I6a?Cw6JK@xSUKGNmc_1lUQYLrM9FW<1#K7$2~*W-63?WNJ9!T{Q>e
zk4IJ>&{>UuZNrvpiaHNY$x_zZQL)U{Pqc;W8|3@6l?Ow84rhu<v;E;z^XNh+75{vz
z<_$UVbA@ZuHPqj=4_9KZhSfB3`SAKm*I$d9s8JJgdW_U7v>`1^Yz!32|IX+3#Wem|
zWjLE@SZs88-u;=jj729bb*$JDuOthW=){u2re-a&cPs*XICrD`v`na)g#LJWOd`S$
z#;;~M3OHOYqsr6;Xs-KZ{Z>(+HWosG*JLZkQx~*IrTdFch}t@(ZW~lAzH_?9ErGV6
zG+#c?dNqx9y-5w|HR%Fha>%{3fJ11q%&^(uy<t%N%V}#<`^TB|+ogLwv))zD1R6L+
z)T8y*>-3*C2`L?Zt8(DJmeDLYa6#x~<7r6CwjUtGkhxD4HI{A85FKy1qoWK+-2bkp
zQ@$`I$v{`~v%+KY+DRXxJ9p?9ojQjmILHi^g2UG5r6zm73xiogkIQJn7Vy{b_tTdt
z%V|A+mYJ?{DP|74j4z9o3HNhEe5IH~J^1B^g}j?Dgw=q>yrAn_CL-Xua|}quj;OQ_
z1M^`ul_N_<$p2STd`Zs7UAk^e56S%S&BJU{&$%DXzq5xWi!5-X+~U^1+^K00DdfXB
z+J)ja=Ehg`Qh_95TkseS1rzy9KE)JDK?)Q>8M*oB)Qs@}FC;`p<Pu2pCp9IkhK#h{
zhc_+>uK%c%lA)I<27@NZnKg{lGhvIaAVWy`P(gpG!~l~Cnd`G2oi@dvJ0bzrbXDJ|
zj0tR>o>(fSISU)&#;NyVX8!PytgJC$K2)fwXx4~Da4K1yUt|vkOBJs5=qk*Q1l~;T
zEVY;|L-<wRpe!kAK(83c40>!e`*3;k-6#g73iQ`AVa>y;M5u{FA;=QvIt3tE$v?80
z_r*OC=1)@|ip@%_r!_xk5{6Nk&qpT)0zHuz9ZjXB%m_dL1sftt8I%FxuWLKB+&Ikm
zOsLoh^ZPps22gFxaGIxkLw;0rfJ5iL7Re!WU<A=zPlrG|5s9V5Igrqb?+$<w>Y40p
zzrb=%IJU)390ONer&9#FutvH{?awF4{D}LG1#bzKoD+H3kC`kbo+3VToYd_S(;Yyb
z*$7NJA7)(}E|T9=$O2?+)clF>aOf#5zfupv%!?Pk^ccC>8uz?hx!jo1e%gP({8|G(
zWu(6y(nN!1*vcf*3C1}dvDc+Vi))$aJX#YFP?v-Z&by|}odCl}VsoAIV|E(V73SkP
zglz1R(?H`K(MGY~XyBWsP}pTd!o?lo)#Zj2p!?Qn_JhN9uDo|7#1ur;h`_HS^Z^~e
z%(ESpzj8qxklP#nIUf<p`%F9c%C+u<8~+^1xQCa0N9>Z4h%N4|@>%_HG;u+IzB<0r
zBJAO!kcuwCJmRH*20`F=20?><J!)w^{MLYI(;wF))F)69W_ILDr}G;aJyn7M?w@;o
zXWkbVo`xe-_`6U3Hy0)NC#N;f7O?vp36&XIb8yT5>$8Gy^Mat|p+tmZ5ip!VkVWC$
zDn`-n{qg)}HiAs>zCwW49~%2%Xy|6Q=-vChr0Gc-;E7z3K-8lLRfI<tF;Xu|@u^-I
zv<r}i212~OeZK|Dpb6^qc0)3NjSPbkIN?d?o=8s@r%JMwfMR8Tq%L1%pB94S6L4k$
zYBCo+B@WmolwWzY*f4f`c~>oT&A&pQ?EF{H&V^S4a)ubck@dpC#AD9}1EM*!zjKFd
zI~W)O$#p-29;I<O<i4`SoGn!RDOS9Rr)<s0&t=7FuK<zt2ACT_dTXwR`>wZVNMl0o
z<p%E5;2xjc3i&LDeoo>OnsOT<8w{{^eo5g%v9Q$%fr)!YY;B|T*jltepP_s_c!?c+
zY|@=4uli(AUc;Bxv;%y#Mq=4v>>lVZf=pS9N^z~s{Utb`Dqd@-fdySp3J`!4gCZn5
zeF~FlT+|7s`~8*0-YTXo&XVbSt0;V$%@-=k1C&Ml7v-_zBoJL*#bnN#&7m6mGcN~{
zHI{T%(ES)2bQU7oP+Nkb9IepSkFZW*;D0IiUm3Zp<la<lbs((819N3Tn#MV*@<v2<
zR0m&ejQL!`c1i)6@7_L%%~c|{BxfD9Ysuf%*fi`2131JnG>t6`vTZ-XgR+t>3DH2G
z;}9B@(JQ(TASyQB_P4bEG)7)8esFIc(YM~d+r_o<ba54y0j}#O|5&~6l9HN9wP)I+
zfM)q3V?p7GPZ&X=DJ^{@XhHVv<>GBOwnVzQYsBOunzM_d0l>d~{nQo+$zwMdGDA16
zfm`3VleT+$xK!Q{^v;mns9C`Psm6jtv8az69_5Atp5mF<Mg<#3C9q#7ft|Jo%S$7q
zB4}fjT<ni__X!0YZWq45MQRH3|Dh4s|0_Pj%*gsbYlQzJWDX;DKdA3wFc#rWgU%E>
z*5XOAll?{*8y7RJGsWDcr{q<0i_KRavjfaHBSHyEkHsA(BBGf7;tR~quIcE(ceJ~3
zxEY#jN%>7N1A0_M8*O~3v+YKG-%n`B3)M9->3Klo<K^P+dar%!oKF@XBu4DVRyhyY
z@qJi0I(xg53i*eNazhv26cF%s@bPfC8@TYz{Tx~G74UX3P~+Kubf2Er2+3>lYSy|b
zQ$NY!xp=%k-HP*@ag=wHVxkYiBqfRzyR<1MAyHR^dJiu_!GNln>dGki=MqLML&C;+
zK)=!}4Fh(!urg^uFvytRZ}!~K9j<>c30xM`+CS4mJ8iBzE-KGzUThNJGqy*!1w^V`
zwl6qZTGR<qNzWr}il%OQ@>4Xi{AI_nhQV=o6a2C$W=xP|q7JsDB#L*<s|+Y8TA7Wk
zUF|3Ms=zoh(x<)<vG2&Pwfs;k6XL}px@T`95b9hCBI83Yyb40;M+n5^`=2#PeX4ij
zm3l~mJG>S+Z0;hjBa4fh(vt}VHu4Xl6Oi!0{+VYg8Jd*fum(6(d?Y0f2kx5<zj79i
z4Kap6O6m#0jfJ(5sq+QYDzP@H+oI6@>R=9b=rnrwgeW!83Q=3+1B-CRyIXzWum#Uq
zLAuFbHPR(c5@LgINn&HE2vxP!^w+<tS1;j$gb8Mo-`_g&1Xo=`Zx1(VIB|J9)2g1p
z+Kb7du$kPrGL|`Bzna7VyB9dV7tKqEoYuCt;k`G6>NLo2JM3QwC6gfeP77D;3VEf^
zvKtrez99@9!&v9gK;e$t{vxQzsLbRI){Wpyt4nEqaKx#q2xyLkEe3E(Rsx^hc$xU(
z%wguyqYN3QF~nWA*Y{?WE{Z?Mzb80NsZpF3lChut7EX9T7bk|E+1sl#@VglAfYfRI
z!_~;F7+X|8g@Zn(x*5tfx2aCcE}~RywNWna?~estL34Ps;d1a3sEA)LT}g|@IAog$
zJa~1E<!nJNMh^)oBl8l?--{eNVMr$@FC{djqYHQ9cJo(A*EeF=Hq_K9YeR$nw1ZV@
z?0Ha*R}S3Omb5g27%YWFfkuVZMhGWlkJK<+#qgrjK$PLw*oELfXJn-DijJfM(*uKU
z0>N2hv@O`=dB*FD13OcV>6rg&ThqVf0En^EOG}?>wOF^zJWO80`cx7mk18OWN#Y;@
z-Pwo8D%xQ-mEo7O;oic*nkb&68*hoMoGfX~u`ac91I)HK5Lj9BsBShpkWDlljAUAg
zD3`kKgoe9GeSMnJNeZBNi;~$W!Zp#vdd#L2g+_xUYL7oNybS7r63WVUVAOLY^%ZY`
z;LDwNFmB9>a0~5pQOrEugJ)m|wyI^fTZ*hm4Pn&Xy`s;32Sz%C%vOL-&betGmlu0h
zRJ)fNU28%u=sp|2dLf2>uM#p9P-yOEeqveK3VbY(_mw-)jdD8C55h;+|Bww@o5epM
zMrSZkCPUt*upCf7X7JhDp#wpwG6R6?uPb7dkRI%?lD?y2s;Xl%jDi&(ri!T!@|=gi
z&J?g=u~(Vz$H7g9jUWPOf!bpS1rGOrIFhI5o#W$7fVzk&Lhb|U932a<=Vx{<EcMp8
z@ugBeWt!^FTDNC!Mi@QnyB~mz$38t1TDc@j#6?U<Q{ga|xqmf$noXNTx1ma*g|qRG
zn-)mp{>fM&Ta#HFt!;MOM?_~zLjv>`f$Shb@$_cmz4Cackn}mCOwN30^Gx9}F(-ii
zDLwu5X)zD{m%M44i#`+OSId)VfDr7!barme9VeClv<VY5rR?8CW{7PCXGqGUdMv3j
zAcels!eew;*IIf@^@i(6pO(_NovU=_n{EC|+s3!ilGY_7o2eb;w##1bS)<TuE}8GZ
zYD7UpqoXIti)Wrgo&i^mmXW5S$SeZ;(sU<=IAQx?<!cjd&JzkKtOVOh4Pf4j>Qe&T
zlNtNl^CN78XB+=)MgDc0*g4|NaR(OQkX7;<1sA?%uu?8p7D$N{3Bsq4vC(H(={Y6L
zMn<{ME(p2Q7q0u(#r}2P6|U|QP_RR+=eh{d-Jc5$`&Ytuq}MnDBL{t>;Ym>(7#ciS
zf;4lfN7^1593+H7iQk`>u}6xNhE6Lcw2&MK)ka!loDP~W;@4}7N1%V}i>I@!MM$J+
zrwCV)PkyoedjqJeYQ052va(VDu4qDmCoWl+tT@Rq$;-cesR{)30&sb-&R2Tdf4H%f
zGcbSq*p6k5>1Ghua3H>7FoA&PrgCMrfuooMx^ywg(lg$dLBzC5c)_GRZ!38w_ET_!
zu{3g3g;E47Ou$lQ_4qj#-bf$8(KLuCUG#$@91uxP&>@Y=2h{&!5cD>r?xlfZg&Dx6
zfDdA+)skRpg3POKM2A~mTZ114C`*Ut$ZZ-ziyC*W1Nlx<e5t)dd5(j?Wu%;7#ho2E
zpR_U(>B%GySC8j<oGLI2dyo|jR4I%a+Ox-MV31O!^V&_5CPt{)(J*5Bu7xFzk07n5
ztoN=%)yguzi)ikfM_F?nV-|#CuS41JK*}d@Tsun=D8_`gm4XI<3Pe&N>1;cP`%da_
z0yuouw?5fSp!q%t3iN;x@)rF!4w1^o)z8#?U=y10rZQy$eU+zGq*Bb&#JL|?<2Y<q
zMw5O*zs<yGKU$mxYr=-Ikrnazf0q9QA?wUf+E%sL7`5B8Y__<oDeTu)DOOzA?l}9^
zx*Oe{#S9#uIyE3%l~vp`@Y;J!&tbU7K8u;>Js(GSVu!)cs95Z8X%&r|uCTJviJa*G
z-EQW1z)>EAeIvyMl|crnOEI-M7JHs89wLI`lY^SmSd-W$(F5A?8RpL-_D1&Um@3rY
zvu#n>-wuMgLmZSBzYenph8$TQ<!2i)#eTkOUE_A(C(wTnNv#CZV`l%%hW3CY^Q1TY
zm)QA_g$NTT<Nw@7Yf_i~7wACh{!qVFqi)jYh2k$>2+ZKFLN(i@wq&pn7l}(86emTG
zueSoM?`(r(6hdnz<56@$&4R3g|GRzmC6QGml9lt<Clx=d(~#AX80BpfPz?L~_3xvX
zFyEPs@=4bO6sj?6GQZ2tw<{A2FTmsbBaG=dTW(tZ>8D{XRr>s!nX4zWbKtFF0$R?8
zUmzn>PWRjE`~DGx%#OjAx5JI&ozdI-ejHU~A|8brq&1bDHg(-ebV`*f9E858foECz
zB(G@9Nj^(9h98fKQ5x?UN{BMhUwVJwDt=#LESSN-_q3LhRQGT(t%39XsV${3zmJ##
zEax>p65=6@0OB;!ca{PPRiT&UH4dan3q`@>K7z$AH=NsOI2<TVtug$fb#%SbmX_gW
zJq6j|Rthz~C&zjV+jAEaV_iJ5wd$dVHjvQ!=h%()naV}axA*S3qh#YOn`~CNzI<8p
z;m))<ETOsFOj32_<qDezvtfc*Yj~#ix$#9Ci)W&h(t~Cm(nWYEhq=*ElJTE57|qJe
zRTCcVn~laM^S8_Svb?ppXtl;r)iI{pJ`Q&?)Q%PGdU^mJ>h(bzH6yajZo=}@=`01`
zy#FFp<Mj3)zzDT-ka%$8+E`qxT&>>lYUq9olv*F&FoIPgG)_hddi?p7kT2EnBy<BS
z=Z2&?4H(WuhVHn$Rcok_A&d3cy4te(9?G*=(1h-bS+Ufi+n8qi6U9<la1rBsUYO~U
zR?dorsBoT|Gg}x!mUg;&zoq|j+(X$mxFTK6U;)fS7uR>-z7m@IBg#1I2Zg8x%gCi;
zvZ}J6v6t!@=ZU`J&Z}p(st>52DK(r`>SwHXv3Y4c%<~>@^Qd<@=C+;nQ9YUHj9R()
z)hVW(&s%-58deS-fuu3TBvqsCUf&<wDWuqpRR`2B^%mA9nVm#eEz-s$l|X8R1ImRW
zCUioyvlj=3(0AurQHftS%SJ!G;g5~wKsOcdVqJWEZpt^@r-H@%$F)Wka(;5Nq{@Di
zy`y;igFBIFwKl*ojY6;zjIpexl(C{(kc_eXKsVamkkico%D`{9TrZ41LE|rnYC_R^
zmH)4r3deA(75z&>EWayo{04L*>`uh6J{=7pTE)xE$ce|?s!@`-O|Al~TOsAlw?WuT
z<>S2dm)mh-h6m!FsVD(wUAJs4+sHNqR46te$m#KoPU@ym;A5}dJwKEf5A))6W0y*H
zb|jh{X&Ek}HZ9W?&NHo@zLe1TAphDV0Jx*wz6-~{Q_6xZR|(X&<E|T9$3;{nuZU|g
z>9AkcR#qt$uaA8lCK8k?ST=&F6v1OnK3%~Fd^2!);k_3!X6~n|n-m6%F9d=2!^ZL*
z5n35{H;cOGBE?2m;I`QIsbE1L-?PWHNm^GY!E_4%)b<QlOKeu*rzIoXm#MwdRLjXc
zN~t~g@;apKjxwh!Z=y}|@D=fh;2S~-Q3<c>wLkvaxZN;;txWn*%T(}4jO}y@s@n7}
z21rWhU`w!(Ht=Dku*x2f^@toRKf`z(W2)$i8|WU-g)ogmw&hoNAZ;fvP%2#99Lxzx
zJ|}2RArKYLuw+~7=(q}DX}k)h>A}K{y0X+;pcAAUVw{m6D2fXmNRuP-0k~`$EIWfY
z_a?|Q?d{--cOvr0D*XdU04yk<_2Dy_xx=*m=!kRkSL}MIxTVGN0=aW9{VZSdjn60}
zPZS}0L>PFUL?{>(Q6=CM1phq8O?MlP&1MP;`z)9N&>vT#koiJt_9ZocWscSA^L09&
zDd$|423Y&2!re5)IU%4I?o!x^&*_{2Mnc`kVNZk{f!?fC9XhNhs2iNrJzKb6;ils9
zg|_&Z&E*@vCuz>f3GJ#xPF0St841M{CnJXZ{wglJwm|$*JJJ08gF$(@hU;`dgqHlK
zs+P~ouwb2#bilVrZWx<Y>{BH1At<&WJf*Zw%pnHt<FTuq@@F@|h~8Yi5dTpJDHdpa
zX;$v1=m83JFVB{3pg<(HUv}L_hZ+C`5To#~r0`*!k)Bf3$2h3w>EKai3k!>y4YB>}
zA3CKt<H$muZU~7|JJOh800YK4q>3lZw(nQ$hwn1M%{gJ?uNzWeV@EkK);!6Eul3La
z;GQJn?9H|nj}oUpvuLgDyB6sN{{l#bvWtp=TC6y`F}i_Vdj>8P{;`84fiT94D)o!u
zu~=*0A{hJjfNO<;v>9M|km2A4>#Cy&9a}_Ibo<qC)#W~zZ98#0#l_3g?Nmh#lx3CV
z3v{XK?*xi8NjtRv0oR$sGp{@>Yg%!rc!%wDO@ijvim;xaOv1i|3H7AP)IUml{R)IR
ziFfY@>fqN8GZ?JZ(&|lGK<Jz)buqfBjV%NPz*hj`&js3yb#HLdL*>lC|AL&$PfGr;
z`TAezX-4+{dA=U~(fq&p`hSLl*3n{s0}|mXYuedfp`=bn6W>(e#oR0k84I#xQofhT
z|IJrm{9H;+wN2w<Z)76Jwvm=dK0*Vtef^WC=@&Dej~eac*HKIrPBWuW=+F)jgA^V!
zl9SGn)9azNjW2bu4|r-~GL%7ZeDFC$0q>WKwZ1Xi9LG%(^OhO#>nz+(elAWvZzr$9
zn;*x|gTB48{pP^v*JSG`FX)##5g4{G(ZQ#z1}svzg$u&S0Anl=mT?b<ShKk*UC6Of
zTJZwPv{M%!Ri*!jv3KeYg=?ZUW81cE+fH_D+qP}nwr$(CZD&V2NoSmKdh|uV7vK63
zRkf<-d}gSip$tJUv0%D>2&5R&2$I7h(ku9>d&3qw#%u1y6iN6IfGzS&RKNo({Vnm~
z1X4^L%LLU@`55BKFdD~{Ys7U)(Fj8}w$Uo(OnFqgIi(Ng>44|LbP5D8`Blpw7I2#1
zOg|1qyEP_%-cyOemVD^b&A2~>byx=K#TqOTiTag&BNr6ew-+xnD(dT~t-I!Lij3LV
zzjETU$$(M&U4}rW`^0CdaClZ&(q%#nM#*p@AS61eM08vQfEs!UNyf-PNpk!{;sJBX
zQp16Gg9ZIlqDvKSf;;2k1aae_I7qy~ZZP~<Z`B76HS9aWTf~B!NB?H|_+Z)f<~^p_
zkTHTDvQ6fK$xO|vbs;>m`gjS?T(U18&|F1ThHN*3euO8gGVR)a0fA!7L}>fwIO)Ch
zOvC_^x^!F&f=mey30i_+UR|L`hcbnJAlql!#l<LB{SJSos-a4G@6E2eSg1${(p2`(
zdih_r#=W+k)@W#Sz=Yz5NqYyKHx(u5!IYdSl!(1;P_6O#GcUrh3V}zspf+2Z<P_>E
z-+N=NKXgTa6%ml+bcs$p4d)=(aeWk~wF(`3kQ-<xP`aSXxzL(wID-CJt766tfWDvy
z^iL|;Ch!}f1UBboFKRGa>b}GM9C@&sR<l5ILy30}=&^$Dz#KxW^1&z`P%1l7xZ);l
zCc`jc0xtzN6kxE=(l$C*(2D;SD3}yCauzDaBPTSBUwyy5j{b2c&=pahr(BSXnIO@I
zlmNq^OZAUDzy_b!YefF|3!xqm5#2*>X^IDx{+EqrMu|PbE&3`}Bq9kp%qLbW^-xW;
zAtneW=#(XYo~t@ZJUJDl1p7S1Qe)U@yjl4W4`9K`CEth64hplBC;JHMYi*!aljd{*
zGQFzkP&L{T1!RHQ?A`O4bM+pqo!J2c{3C_z=|{w%6ylZLa=qkB3MEMXn(1_%Mw}4E
z4od_I@(gObo8X@q(D@qsc7T5f{+q?N^hhE;yO(2Y2cPq*%~^U`$N)4ZO`kq%EwBYf
z{s!TKs_4?V!n8=lP+gtyEwMGWZ6OAt1*ciAT^NSXESuS`qXP`|cAy?+9cnf0pP-jt
zOh`K~y@F;~dKY!&14z%++25te9H**YUU(IZ^;Y{%32Jltu-G&b+HGRN20EE>T(Ey;
zg6>lvtYQ?36>;xeDo%vw7F%mcBT&cbOeR|i++<e~7%GJ6P)&vSV~Ke&*pDKgozWtK
zC>ZA?>`;yaO0hzQEaESg6-3BFEO@YVEjsh@rX+F8p9t96CN+MTaLNQOQ;`43ChM)x
zx5z<b1Tg?QL*)!@8E`CdmT(YbxZ#@Fj}N(Y=sE*{LUwxguJO46d^<Sf$6|rQpF{+1
zM4fWUl`TD+#y%e)zlBkex7Kb9YrLEvzogpXX5aklP!b$Q2t^3s1hmE;<-Vjt-Zzuy
zLdn#g_92JPd6>1^x-^enVF>gAl7u0GaVXVmyv<He;<wB(q(mRQC*n;KG(+HbNDtfS
zb}?UbEOdE$S<wGi4>qqV%mw!PY5-trg2oE&B^z*ge{ja@1M~V<NbfGsjt+baDOBx(
zCEKqZySzKwG{_*+Wa=?GduYJnR~15AoU!PDOGRk&y=OaZHzBD2?jFJ>;*wtT0v7^v
z`>Bcu)`4Lt@7HgwJ>GJgjUsi}y6UB61-p^97PD^1=9RtLBtdGe<B_cP7vDha0mvrn
zvGLI!j|MD?1S^eBl{{P`2E1N_RsJfLZPo@$JRQ16Am#D(mGwviE#N5XSFR!>EFCf<
z&@6bNgie+FTK?tezQEDDyDKu8@4braVR1+zjn3SIQ+nXe;|GKtLozi^15+yB7IMA+
z(XDk}0vT@pNO?<u$5QojA%*L3K!J5xSk)V*(aHtHT^L1l-4_32WtFU@!9Evg3^yZE
zJ?T<#ZMqUtest-n=_lhsNTAnZ__N_{P0HBg{%}4W?jB|Hq*lixp?pf~R6N-qM-k$a
zj>!4iN_9UHr!iX06{NvGftl2|itdum;E&4?j=sc`DSH=v=%q?nXEokrN(@M4h;jT=
zvc8sc(OIsK0><u_MZli2lDLL@bS4hA>HhF9iI%y`BD*%umB`=d2>^^-E2`B)RI+N7
z2j@X&Qss%*V(4Y&4=y#nOEijCPl!u5hdf=iKo76H%(RYrLMhS5Hjl>--fpLPux^$O
zqL(x$3$FPZ6~Ud=%QqV*A3iM7c>O^0b!x7g10(l<r_0BI&jL97<Y_xjJo}8SH0-L>
zLn`7+-KDIJC2?-C&JG9qV^|D>gI}A_Ub_stXh5BwUYahS$)eVi62CZ5WC!(?QE*e0
zA>HUoxr}!oy3AQcZ$%tPWg|?`i+0j^K-aCgMN^*odm9v)0S!tjP^DDOk5w=i0>ZZM
zL^-;4*@6T$i}Vx2MA0EWc!=hRA30r_fi(+VEM@hxel0`X53usp2j_px80^gde^Uk{
zCnMYcda&40-;4W&5Bl!)3E3^6>C*XyCkHo!(xr6`vG$*KGR}u}5yabUQfHIekb1bq
z6n<@S7ZuR_G4pcDLIHi$k18HNU}xTg2|FAh+aKtPx;sT*vjde6N-h!0)bE8$hIBT@
zh;wGp1H~5}5E-VcYd$y#+hX7A^W*jVIhwhCl(NL@{d)L2bEU>Vv$xmnL|kXvTQ=%e
zr~9+7tMkqCv-Pqi5yQTa?SKj#Y71XtKMU$s4m7(6qRE4Ri<1Bl%94)G(9+XXFIYUt
zoyB|IKsm%V)dm}*YQ?@R@0D|t>Snu3Ww*(tcfHw7B}htY%yD<dYO}gIm`YoKN(77K
zZ-p|53KuPda>SHLFzMQNJ&D7{mL2v}01l_@@(^J)YevE3uXAYokP<HgDVn)=#sTUZ
zxT>=x>`JV+Da$}-2u{XGbK)OjmFl0uMKe_`;^N$XzRYBxa%^Lz#?oF(FA605JeeU8
zCfd^YyQbEn<$py=-Kcdi^tJRF@aVWMc5byKOKe>!U217dTckPt?T{nQ$siBCztWvE
z6<HpQ*O<@hSt#f_tV{mW8{C#`+8q{cu1v)n&C6rjCkf{-Hq-zFnib8$3yvfUC|v~^
zO4icWvF<pvz@RVwW!Dt?1o9?enR@Z1)05E{<C93NDC_r)9<D>ie8);~I&sa0VZ<>^
z73fC+;TseF^$vvG{ooWcYDC{e#hoN<uJjWyF(Oa#S^rVt)*TJ<W@u=4(-$JUtLojj
zUzc@q-qdA-H3la_I@g<mFnJ`%l7r0nkAoSaarF>mWTmw^Ly_Q`MtoGh<oqFp_T47#
zbcAr!s7$B*jvupb%XI73uXVVnwVbP0RG58_%SI``A5hYZl7Rdg9E6}}$s}6_2jAp3
ztip9{jGjf|0?>Kb%aK?x%T;WG$jY38{-R_16to!r(!U$UKd`#BS7uO$JWbc-d+1$7
zNg09WQ<|`F9?}Mqw;S&oJg1cETBS~c%P`UZjsG~mn?P-9z;Rw|>gCxQEK+}xkZK{>
zddHL5n8VjAEv>4kE+R~0BT>zC=rg6K?F0%qur$qA72ux7Q0&4%YAk+JkSEAof1I6e
zcx}~IF!t)jcEMX+c}G}mVym~U^emk@9Qui}=svia0ZLwC4AK}(5c5D-$A$nF6^^Sk
zD*$tBsR6Uy1cb*Bl3YtDv&RvVsYpBt{Z%WC>P9IvCnHS|Hu%j|i+AzwJm`*%>Jn7p
zbEvZ=mN?ZBgf{On^_EOAYy2Mz2cxYWpFYtq(P(V{yhaSUGPt%7Yp~#3eTt*#m>3h2
zwe~IY*%PKaIe+G{l6lCbP4u368`tebUJSr6YZ@azOfejw{_r6)(Dx!M9(5Vhbs-s-
zB}FraI;ll_@Qwn`<Fg}uRGc|QIhY21cohra2$IJ&XN}Ib%??<l+34rBT^^^?q9On@
z5_8WMbL+48mk8(#u(CzYIEWdL;-QB4GPA!Z_QsFu9#-chj>_Cq0I;i%g8XGg3`kJe
z7d4oq7Mq-4V@6#AT(McbvCP7Z$e*YoO*7`RB&3uO`nj16zUlJb&{qzQtGF|%H?I7W
zl62Shjatq1K1^J$y;u_bP>UXeI%3&+gA1w$iU3(Yx4{b<*zn{kYGEV?MI|03CDrhJ
z4%FzjJ7j+6aFRP_VA(y?XsxAeCo;^ukY;DUpNQjOjF%bu5+83eeM8V}`7akaMDZQT
zhr0dZa770WO2@Nr;%NV9l{0xW187vnT#?@8GURq~f#8AI$2AwV@bWS|L8%@oATU73
zh4q^tm#>8lJY>WQt~i}Ph)V_wO|O>{m8liRh4dr-yJsk&UaCOtJ*1y!UnpZVa>acx
zE398J6EJIF=5T4`=VEl8GJ5v|<H4OQ>Fg|45QPs&=pqcstm4CRg=(Xe(jSSoIL%EL
zHirl$<{*&|I$hNt;U&$E%f<RIweW7MN~*z|iX~sgL!<+7v{=Nwcw1Dm=@D>3at}&E
zFePCcw?spW7>0c5d{?semn%12=(V2AeQ#$Y1K{=&0`I^osxVwMQP4Q5Vnfd)@kn7s
z`K<#Vhj;X#7q{lw11FevPWNS~w2xx8J8rO)g$c3=Bw$7SWL`B)R3c*NVtbIi9zQ43
zpZ|6h{$OFq?1=IgzHs_zVC=9A@&8yRDCzWPYvSY!4oFDut5rUtx6y;6FX$1O>)9Q*
z0#lZj|48Xk_`dz6hCw2=PbBw)3A#rH>+on0++GVO#EmZT8@98_8SpDTTT;6o@=1d4
znAsbU-1W;bhE(Io_$QbXLHXMD3-1#C^LM!->`MS7O#4FVXpg@2Bp9$L@hw~QK#J(_
zn6%<99anC?b`94qnEuS!rs?KRf;Ud;2mO-uz4Q&sOWUHIPPpK5<fuYIJo9CNue6<w
zK;gh)3fIssm<H7&bx=xuLife$^}zG7=eYQw*4Pa><%c*I-`1TRkYXa=^qC?yzjONN
zH>Y{;WvY^{k=9qht1y5cmLxX@Cwwo<*X>AAC+ZdS+Na{kuDXg$p5pEsnoLg0?!THj
z$A2_)4$l7_-l49QwDlXLu%|a5UTI|DK<HkPHO(@y>9SS5)MEOvZ_E@AA{kDqNJwdu
zBKNuF1qVlw@VvQOy@L)20%vcFKP$?kDbC~R@iKFL`5{_Z7>HUSQ?x<#|E4H3nLM+p
zP8^qQ=+)^#=suT=<Tfdj*oTpS{H6E%HgI)f&yboi8X!~3d3?J)J=vMbVz1NPTl{$a
z)hE7xVt3y}g8$7%%~<BbihJSum|xvdlHZxG_3c6GPZcLpjC3F@EFQ;HTQGDBnW9<|
z$Q4Igrf)27UAVl&mrrT1(<h9i0N*NsvH@S6=^rODu*gmLlH*VN>&qm|8TTVCtg*t~
z^p#-sT{pFhZ=Sxj%2Q-<w8{;ve8{<xa)DB;KmmnLa;OhV(&bB-R7@Uw3@M#_t?l6>
z%YJ9o<s7xDLh5r@S+WC{&Aw_y6=GT#zo0Bp3V`$7`vWq-Hx|A|548Y{53ZV6LS_GU
zKw}{+z-*wswYXZ^BK`62MSpU&WtJ`Y9Pm28YS%!Rd4{xIGDt_D3HSc(Y;+YlVGOPI
zkZD&<_UKd#S~p&3<Yu&vnn7`Cak`eQFx>B0TdtHymqSZV1n2MVOaz&Y=$cF5;$<$!
zi!%_?8cZBcrhw0>&jMOA_C=Cpf2S%8?&-_L29P!7;d`rQcaU3Ed#o8Ycw;CmQYlww
z$xRK4LkoD&+udE&qBPu+ootWjJ4qa3hsg>Ul2Tt)EQ_9XG<AfjH<H732!G!qixkqV
zW38m6T5TZqWJnvY`-|@p8g?d~jAT*_=OzFsmTOq7>Cq}SJ?Hd-xJP%R{#k4<W7k(}
zR@dI3^<IuWZ6-$dARqXO9*y_JucU|U=#BI-88UoNkHnvstoV(9J(}vAs{&I#>`w?|
zD@#oQymy&zwwgFJchNEd3PU{TqtdWjy-<P7bB3Y5e>s(~@=n&~HBEc<yZQDaD6+=x
zNS`10oNuNV3+NL$5&mcwaeqZ6IE=x|&7N=P@a<Wg*+5v@+N$-K+Rw#8Tg(io2OWK+
z+=Gpa-BlcxrNocQ{*(ub?;}9mz+?`3VqD30XIfu3Z=ppdL(NlI(^yjdeBb%fg|sAD
zlOX`GjD-v^s!oYe<F`VguXbYS6bBl>4?eFKC%gu{(=NCRy1ghM6Yl{m1e`1KMG@Wr
zZ7S0N;alR_dzzw`iVsld@mPMIyqvr#EkS4HlTIy8NR517*lTs%L1&kIbj@FJJJvkx
zo&z8zWC=jaXqjSfEih=m74Y*pmkvI)B(uaZ0~K_5|Cb_$?`}gog>=Z#lM<YGD4Gpa
zlQz`l5h+q-V9FAsN!oVfe7SK$Y5YE%dKzIM8P0a5ws4GMfSwe&qjy4u;I2gJzYt$@
z^Cp5M8Ia)-o)qXRPj_qSJ}Ubw;X<%;9R%8vo&j904Xg!NIFokea#)*U#xz$@-MTmm
z71xLT7C<)I(NyR}Y$od@DF%(?N!eGv+C}{rWx9MAl7+Y=H}(UQc~!1OSDxYtYvqke
zIG0zojYZV2T21GbFYgu#12@=S>Z(gvbV7XX&DAr!GpfqPw_xNDuK~-fHaGoaIu~rW
zW{la#_vfGozrXTTg{~?p2K$q{JBrX*#Ed{LHbNUckLr?vgNu%we0q`6Je--o`oNH@
zUmU8u6aU)EW)6iraBTC(IsOI!<e?`tNY62q|Gn{24-5r_;!c1{1>p!`127gLF5`S5
zW@yD;3QbXIGqe!Ww}jhrbw6I+=HpIEWnn&vAcH=4-^9`kW>d&x1bnK*Y2e7*#6+m{
z5CF}Yk|0svDk&c7*&@Pe1k&Y?=8TK0-7K`|M}eH{*0SG%hT1}J+poOq*4MVaKEy@D
z(0uTPqP92G@$^WTqpVJS_Hp}c)Pg_>{?sGTSq!zpcV8D0JV=>DAeo=;6p&&7yaOzJ
zP*p*hj4$4apWw6rp?<2|2~J7u#J^T(;D!~KU`V&F(Fh&Vb>O*{5zf6Y4H!SK%xtw@
z>cmxVG=PtQXYIK4e}>wIR{Jc^(g9*`_qh=2ih8gyfQzACYu*_09yqdZexS{X3y}@q
ze!1`t&Xr!gs`hy<E@-wo(&_~_4zVUvfsk@iQ{_H0k^ri`48e=z-s>71?RgRl$tW-|
z>k@0$jPVfIn{p=#jXvh|HkL6X!ZpS?FAN5ufSz%rJ@-e#hvz`T68T4N+JB<42L1U%
z!7vO4y5I0etEb*8(q;k^5`nwm%$z>6@%Ao+sH3|D<tGnN(m38uY<%g!54z^?Z?hY8
z(@w6rhek@O=~F!tlR%pDj@@pWd$Pdl9}eqzH-Pne{Kpt_4ljw369e|LW!^{o+=gVc
zAd(9siv0GV>{~<hPQ1R^0Ci#*%^&c<M|c?A7!>nCR+dbNS)Qx!WC5(DFS}#T>sRB8
zi|)bjo??^w0SZo2ti*&51}-F30on?6Dxtb9LLnhqh>(oQko)w0M^&*UJkoBGZv@YH
zMeWLY!CkMVOvmNium@Au7k=6@yV(C|*&P2Bsm=IH;rc%hieAjZ+S$aBfL_enz}ZB^
z#K_Lr1d5Lj%E{T$#J~p1eRD^B$8Kx<|G@vnp<38JXGj;R<IyA$J`;s)O5kTeK}Bmw
zTahk^DY<PIetMd+NktcvaXuq~H_`uW@4SD{T*Fw@!<pZIU&lo?soi131f>sWUg<j^
zL*P<WK&DWsR|nD)63ZlPXj@?U{`+@ClX!4Aq-Pr4a6hvW#y<18wUJGY_5JMeP?pEn
z+mq{~quuREE}B<S7#s;<%*GSqXf|8ER_N2~iQeSTh(tZ|8#qRq^%Oo#c1yin2H!i4
zS(3l!t7YaUnL*kjniA=ggT}2TTP_kntgz(IN~R=6gbEskEAghT{9-+8h@8rd7Ul{I
zkGxqq8QJZK{U?1NDYw!(rs`8myBym4HZo=Bqvl_J#uim-3hiCuNB^T)<@w^>D)#5T
zN7)2guqYP~c#=5|U6i3!(!cEt6b7O90SF62qJi<za1@c6C?;Ufd=!MT!Ud0%6JODe
zl#X~ZnJ6FKAaAgMRhdw`JL_0(?z6@2b(0A+bbs;ZabjEI0bK9*&s~pu`|s62wkpb$
zqHqR0enm{{m?KJq{DS2_Zvz+Lgz{0{bV-%2V2psiQ&c21)|cK<WvUt}w^fs>xLQHh
z#B*qCckHXk>m)lYn?O>AJ4$NOhUsE1jKTV70x=<O6Rvg-c11yl=REi>hbYGfhG<ds
z{Wi)*+(rdsRb}lvo_x5T30ARl#qv9^u(rS$goajeak-a~$e^`$%HZB6q^04s*k~bs
zH8vep_jX|puZ~e=w~eS+GDDx$yZe33=F!tieyx>L8IZ3QJ>qc1usm1V72%=UX+lV|
z3MHVSa+Sj8?a6fgsN*zWx;}buFCc>+yGZM}>D@m4Yd*KUTl%#cp4T*Uccl&3-y8%c
z@4f44I@_NNC(-HaX<<7>OixpMfR|iNHD`|Skz?XL_-Pt3cS{Dvx{Pl}Pd<Tr1pU7v
zCrL@8!st2x@EV9!{(JYn?*|IF(`f{_RJN3t&NK0N7~gB}a~4oqt-Tz&G7N1*3w0$M
z=bSG^;gq6o=aB)<oi*{c-O&!pc4}=az&~!o*-?V^tYF-K2SR-!rjkOo9voHvemtd6
zySg{F50rtzMtWfX+g%|bICDAv)7iwjTt+rMxB{IVRF~4zWTMo$Z<2MwMmbi2^_uK}
zlBFB#5C_{ex!6hZg$PfIcS}uAHZaC;;Z7eX9zjp=AnMf;qqG#pR*^%g4!{->rk`4u
zZ$Bij8z&VK*70<rcjKDX)(kU$kM+|l{>}DE#9;zshf%hhjU=e`%qW5{1QOd?ZcpVd
z$yUKbaW$xx$7H2ssj}@Alf?EDd$P(&l*YIcMAw}lVRg>LaU3jWg9k@WTu8K<tsERQ
zD>B;4KKX!P#-p#FU@)B%3GmHyj{9k#Np&FAhP%*q7GBgjCXNh@t9ZBk7_K<!yy(k5
z0s;4m(x?O_dB5aV&Y}i}K)#<(5Sn>sL?KRxx<vOf=35&0k@PSQ@wSC^Fm)0wj2}p<
z4uN<o+EgE;ob#9rh^E{^gELA#nL{EodV{$+2nFQ8ffU5u75nO`h%!ufD{Y9J`0<q(
zNbu-yHI8)<#j{GV<YpJQifk#HjI)>nw#|Ss_8ieKazE&Am(-vs43#@i9=qR)g{YgI
zI+8D_M!a>V9(#M{wyGjS&N1i-2gO5Gs`(noS%ZF-Oo1rxd6{mQjZK8geHd~9wJ;I1
zI9-Cdh^K^POSnt;xn!K#-lvTZAo^b10$`7DDz77BNg*+v8z5r~gV|w|)-U@$laQO|
z*x|nrn}vAOa45auv`+~_Gj<C47JP-o<b<*`MPA&opbw(}KRxCIuLi|)eTE0FF%Zgp
z@NOJon*>M~2LZ&k-{?&g0I;PPN5XB#F)<;T&SYQg|BIvc#3|h#N8?OOVhCw0I7y-i
z;mphWp_0-l4xcCHMkwP0;6qT7?a-2k_Bsk36*EIC#cUSa4VuLbGslO{_RmUqyy%Zv
zu+{gQg=GD%{2D`ki`o~BBD|7q&{0vkPYN`@^m1yRWKne?a|k64-|x;5mN$V(Lfp=L
zh<`xFYBD8?31v}dI2CFveaUH8<4y=g{zVzq<so1ONz{h2$Rhb?e8p~V>=6S0T}8MB
z+Wf>3gV(kQMRKdYh$LQf7bkMoo6H|^SBmMfr^A5LlsoL*WZ%=T-)pt$(m7Je-PGE%
zFSDWUZSIL}UANuJfH9Aps5dvafW>x61FmaNt?Z%_i*eGle&wd`?2zAKeswZt#*R9u
zK11qeKO1mwl&Pc{nK<qgW8R)_d|Z-Rl@oBuFusz7Rqq<pFb~0|%e$0q(6fsN80DK=
zw~g}BeqwI)k2qkM39kouZD80k`Dw1o-F6V5Al-bE=iF(&DXacoi@~R--(29m63cUy
zk&kNpTn8{I%(9>DmqaW!WbjSB+FxH?`9xRj(aHXDT+JYZh}H~H9u_3p$>}fwbPp#6
zwt#9;1K`Kr&i;BYk!sDeffT_F1v?Kauo5a9m`wTUpU!~E=bu-|xd&!uPHZ;XqpC<6
zSkn$WX?M%*qU*S7`8nzNY&JJXWufV$L<ZA8^#U7DUuZkJf8CpLw1x`HO?IgSq_cU4
zQS!Zaq!fYhBQV152ui6ElVn@{fN$z3W$E_*f(er_SIKz)aUCCt!X^4{!xf6-I9{p~
zl~7Eb7)pOnlh^CZyGavE5cXoxcYZ*9`S%xw_rk97pk(d-fYrA<5RPS0820<iFgRMK
zLw3^G%<6UE6Gcv0M%`Ff5v7dajyp%(HSaU|@h`niPB)LlyzbXcIsyGYm)2e4UjM{f
z$V0Lps2yKi+`rE%+o=U)x|<&Wc|O?p{~A`<|6{+v#QZ-C5^mJA+>Tffe)047MMD<K
zQ;psR8nXe5bk5Sy7Q0gNv6mbX74!qHo3LJ~d$SV6+^t<bq|jmDx<n=<GbdRK=S?Ow
zFjMaAZ@x?%JPzrI@%=42MojGKR`z40sY5Xn)9p+d+}*j^c)1GH<~N1C^4gdFa4i?-
zV|Tt8IB{XbSz2VILx^s)PTv1)y}do&mqs`z{d(UH-mVobvDG)bMS<o{x6k872HmYK
zy00pZCide1L7P$qVMoJ7D-JmL9(ZhnF9aqtrVRS*+_M>Mu|<ui!TWA6jgg5=2-<+(
z4q)J2NQ`norGSUP{dLFyLs8u0AukcuUJ;D{pcpO%zz3_Htx<ReU>;c>Yo>ajhp6cC
zC_I0AmZvVgWT9K}@H7-awCZY&hHnN3$awJFWn6$;C=CkcN1d}rX>H7wRzPcE={T%b
zF+)E_@$-p>;7+_1IO&8m$E~fM?bC~n^G%8x9Si(n?1L8U?44w(w{vKZpbBWJvws8N
z-pvQIM2rL11Qv$PTULbMfW`Z&lG9KeS)!r;wP&L&y#Uf_k47m=@N0eS)RxL)J@{C5
zuL)7~PhLfxV!<_q9|d)wOy>wlMq7`u7rn=S#5+yF9Z^8D3ro>H0jyP;1S6xa=TI2_
zv48}NcxoR;gy#BY&OG?Zf{m>b$_yXm(W$B8(0Wcj@=;v!uDvgyEvt|(592FK7_Sd0
zC_(xOw+>Bmgv6tIxd&g7u8?@4nO1?&DE}!dq&kELkrl#kY%|qBK-hz}tWG59TZ{Fg
z9W!%lSOgv9*b)mtrJ~j}tIa*yN!$AL_dI2ag#r}tFl~<(4qgR5H;hI;B?ZAf)A#{8
z*WR(>eRU{YYLpVeU>)GB+4qjKPcyO?!jBrZDQw&qqH)U5u~)d-Ce!i?Ug8=0!a|rE
z=Al<c`J*xV63URKa#}%!t*C^ise-qUnWBg>x0`1PTwi1)EIvOA#{<JwhT$*Q+^m=n
z91+ivLb0-PP$^)b9%;<Sg>+@#At;E8Ad}L^kcyg(B$dI(6Lo923>8OZl*_rHX%i{+
z0*>@XE3T6TB9msPfhxSG9kznS0&ZG7K<A(x0R+_&nfW>7A`962Cn1$05V}xAe2OPx
zatl`KtgsiZo<;opZ78cI_1BR}d1wiVc2w_ohS3&o)wF5lfK2W{iVe&<!KZT>wE&8T
zkhbB|s{eT}A?>3VN@tILEJLUHQF19^*{RQcvnF3ShGUJ}ScQ*w2rK|`!5i$we<?eV
zEZ=+M|66uG-v7NhCNcrZf$g-tMpvf3$2Xn3-9CGMw)STCJQG8^ILJ5DmR%b1izUsL
zqFXet4{e@Vtg~|?6Pya7viT1UA@U0MY9E5b_IXs@@XaOZH?6#WY@yMV9OGo#Nf5!-
zotr_9-G-D${K=dvJf1+vkhP{?F9Bm^zTx=BKi>Pw^gPY9PIBHLu~?H?pMSRg@ynh2
zd+=_9fStRCb}8e6GH2M3UK?p%IlfJ`AtCV6!+p+DM*CE^aw$JSm1l@LS+uxd6aA57
z_{rFyX2jt>Vquc@%Ia)rzSbGraOC3n*>$<O)S1%925^RYI%Q+p{I0Oj<%iFb#RIFw
z{SxQL=_+8wN9Y3j{e++Jg$S{}Cm*3R^dD7{^FKfxOw3IGbCoQ?oO0Z5f4HqnCss+Q
z2Y^H{E@4xd9ql@?)~<~_(LTiCz>ye23R`|8^!R(VB)~2aSSq%A%rsmC`25N0?bdhH
zrMLU<<?jqUKXmS$V``a0XIIY_>4n@xvO|SyTi8w?@j$Sl8avg=LfG6-`yx}P&rcmS
zp~&;!yt}>5_pYie^@sDLv#mTnU-#FGOa3;$p98ZD8hVxN2a%zy+gTB<boT7}b%#_l
zhC^<^c$%3Jxrk`Uj5}L-0ADYb`@r$SyjUBXkFchklLehWa5iDwH@LpwATvdcK2)8R
z=2BdqRH3yWkzOkNfb1NXwfkeqHcRk3Drhk$`f%+^&+euD7CWh;F~V*?y=<w;Ir=<*
zQ2>BD3^BcN|B4#bNgbJA@03^|sle6}wAR~O!k3P=d|{?}x|dtts`*31^*;^+t~=nQ
z3mWYUfE!<oFOMj+JQo@XTX>$<w|b`$9+~7appHq8Qj7lU2$RRl=pDP#@3EfA+4&gs
z&{i$Q^wjQS2Zn?C$K#qZq%=`pZna4ik7fcwEZ*7G;=#1zzdYo9Zyt}3;|rN^$D~lt
z(f~z9MkU~ajG5TLF5FO%4ETacQ64$uBzzTNPn3&UVx+H3rasLa?9C*Hpbs-$6(nKm
z6K^r?Sb+Gq2Q~Sb>vc>0VBbDwN`eOFR&0#2LNakfD6#(L4RTLl0dQR!!%cn!6WnVP
zjzvd46OP->wu(1o80b;Ii&8$ER9wl${1#dz7_-ta4!Laao^-%rK5Wz(ln-=iZ?q7z
zvVaMs54_1!pMq$9ASB$CeIMeGN38F`pgL?afqQ-=PN%C5496`$zc_nqP}hQMd%6Qk
z8wNw-4WycfNbepKTCf#LSY9=539r}!>$ry?D)__)I+_YA+r3UuJbQAe&E30tPW51I
z`YfEr6REfff0~%l$69I;V(P|q<i!;SPuwt87_9dj&PeWFTXqpUOLCw)+B9iy{Mc+I
zR#=WK!_;v?JOilDbBUMJ=cZ?%GK7aGMo5Cpz`|k7BXmv1;$R;&N)SDLnzV8-H9D<4
zg2RLpq;;eRbkPwJJA;9vf#;=d4NozU)S{cmG3Yj*TyFcoO%*tu2ld4qVjo(q8>ED5
zKl$hk4i*PYbSKLSCu#*xj2f67;;$?Y(SiAhdn(;<5{aiLzvMYv7sviBK7wv)z53Vz
zHhn}~)E(mf@G1}Jvx<95TK<r$Uc0+x*eVdsuBuNKVg5bP--z*y*+F#QJt5sjl=*+a
z<K&Vs;*0l}!fHJ1;akU+xT_<Z=Yl!ut0=+W=+YxDcjfx#Dll}K@8HkYrxliCyIr7X
z^3{Agc8ze3g0PAm5k?5!k<;!mW>ou_?p-<i`X)l^9I%S1U7+i6TX&7>C44^rJQoD3
zbB`_#b5RzVdc19ZfE1~+bU&9?MFC(k0>Bx!aON-E@|}yAQYoO*>!(kV_nZo}8(20>
zDFb{CAj!wQML}SA5mDZ24JBk5K1d~zSZlO61XsI8C?&ES`5p!;H$qJ$(k78%UL!#R
z(*+&%&y?}JH@%^Z)22qG4?)MW!@l7qOg)&?kb0mMEdP?a3Z(!<Gw64cvMfG)a}4bJ
zAJ!#Ly;%Wjb^^iUjV|)LOh9Y=tA>ZY|M$YW>3)b~X&*i~E1H&OPI2~lzx{LeWSVU}
z`*vv=?PD3~gYDs*J)p!`VxXiBqW7!~D~^+~neA;eP!!a#B6ifX_uVaNpV%Z5HUFn4
zTYY%HbR~0|lsQpjqOQ#rF<w{h<MTf8Td1SK{bGzZ6pZZM_lXGf6okWxtX*?UXB-r=
zIcQ?qKTYHdmnXjNkwELo|0;^-ZK{fApZ)>BT=PMHidcY<E)}PJS0YIn?3BC!;F<CD
z-TYMBL2-TARa7;83rQ+{`mmpJcX$Fc@5BMnj!NZCKveQ@ANi`1Q`scIq_1WZjSeC|
z61+MyjXe7g?V~uZN^+tq@J{UPy)RG4^F+Y2V5p@2Omb1JVk*a~47J_Mz4NQ~OQ_AW
zm|-q+h9d31j*6<Vh<X4!RAavS-dS&b$WQF4>9BgbC`yxD5lkjhnk==N5CRriWmAOW
zM1&vtI^|5t<A&8E#TReq9g_Xqb>-nRUw~4y+6Ez)H$WsNsh*wfOZgF)qHh6I2#Kyh
zx^qNr$S$%%K(JzrWz@GwmqA$~<LA`aq_mJfQ4-3s3poxZf8IYP<>@GZ$8si2;-J?x
z+g?@pl~1A`2`iROF?v^&x1KVSSB?Il#pgthBMc(f%iC21dh)5BB(Ov@>&K!{+n{op
zQk8ah+j!gd@Po5DLY_Eju__CoKcB3<*=!P4m($+P4hI=}+j;Nm;`X;wMJ2%N2&#TV
zJ5I2e?gk^!U#AHAkUs|^fYR|15>|H-JCOy}Kv%R<gR0X*V_jeDE5}xr1MM!|mJJNW
z@Iu-ppl!4+0iTNYvygkH0pr@k9zA!$6I;)~1`wB`jcs|_ajlsz6f4)>Xx%tWFo_;}
z1cE|=xQX_1#mUSAw34|p;Qm$K?zRs2oCz<9oljD0&e5h?qbHR^dW5*20c@;O>AXA{
zrql-xR>j@rAuoe_0<O59Y#HX1=JY5G#PU6_%W6#j@Qf2x%bwQn%$50Bb;H1i$-pqJ
zZB270=#cY~;N8*?qvOg#A<XKYA4q03-@UXyrGB(Cs+Ii;o~M~j$`A1r1fvuX4u_Qc
z^5x7I`obMDv@b3+9k|~%WQBvls@9WI=3_@yw4^ex(r3$EAlxI$=)C$e+9FuHX0k}(
z_H)xHAc*!q0V~hZv!JG!aZIThFQ9`fQ-vV>2<+QWJjon<7RN8G1C~y08b+N!@16BH
z@WTf<kj}YLr)5;V60J_fZrI-%%WNr{HA*eDwB+quY+aXPZ)T-)jOSO=;YVngGrg!(
zJ;~g<7(<>-qfCA7VYhfWJ5>-IHMXF%)KhUC=@19d``3?riZ^{X=X}e-*rK<TC1fdM
zaM4@x!+)yo$im9%%X>f&?^s=C`_|x7x+=er{k2h$Plg~`Y7)ar(@2axnVM9&vACzM
zMV+0v^<+9p`?kz$Fu?Eg{dxO$#t*Ps50Cx7=B)p+pRut0&p@;rjV;GtdA!e>+I@|(
z6=6Gab=GWgOj4^w&gyE`)x|uLu@g}OmE?v4*Vii@U9qTGuY@a6+!8vVHc%IMSJVh#
zNYD}hc%2tNjMtp51KKb}sxsJ+WWO;%yGVMJJ_lT^Q0d3mkX~*N5r;08XtGeE{k#GA
z*V|ImUYHm77a%GJNjX0PQW&^b9sFN7ct78+yW_qdpN`k7f`!$gT^=z9kr0SZ8wV`y
z*Gv5+4X>fyJzgMF2LYf_1TvESU;;@MhRm1fY)QZ)6#)oQp4cEE@T6V*tU~Km35;<I
z82-QxN!p13cwkUJs{49l7aRz2ZN#&~2_kG2i(m5+xRAC)H;Xg_%19;L6u<2?E&vEH
z9G(FcM*kuvK*Q5gwWS7hEEv*Jz4r9^n>@|^&I*2r<R6kWT$_W1*#|7H>IW+ZQva}L
z8@Cu0<Y*$qJ(~r&cc(fW;of$9h<KIVMD2r-9oD;DOkH+Td+c)(dfRDGsd#{ZIZ)&}
zXQL&N!m6EI(;QjvF6PelXuef_VjX-Rty27ELF=JSNE>psjHQF4r<o_;oFdonKc3d5
zcbySryNx#^e)v>4A+>0tu0|s2!NumPI-UyJgmTMJ%It?d@{*Uy-WUa!ofPrW1reBr
z3IU<a^&C7)3jLTiQt@6=(rc=szk>NC+$WNuB+(N+HNAI0ja$(OD><}_W~^e}m<huz
z&T1rjsYVAaa7Y)$(oOhk$2Gdf2D-^gOdUYEBUW{}>8qD<f673H5?fl)ymEy!uy_Ik
zJ_m9o6E%oH4Zc;SHCggr7YzMnZ^oR<hBQ;O5Fb5H6CWg9<J!JsN_VQS`C)2DI6%P~
zqho#2l(2+$hok1|$YCm>m&wk6L=@3z8t}*|`)uWiyNd-DN~AbY5b`DO=UEpUfPx%a
zF|%5B**y<fI;!~!%bb$cCKm1--Oe%5QwDj!=k<u}eUwZLZSBmaNfJ;sHcA-ZMikTl
z!Yg9%?N9u$cgc#Vr(({tPHc)p2_0O!G+PPkG~!|+MR*CE;>|1cTHuDIvUtv03y{+O
zDT`B~C?5rT_ro&-_!{%|8D6&iAVzd5k~ewhjHo~;;*-`0Xd@iS+lVOHQS3QNNWmEP
z1DbD?j*~aIzODA*wzzo<--0OC)J8b9w>Ox7yST=7u4{x2)Uk<Wr}1femBRO|a2tE;
zgw=FORpsyBv9GVd3eR2~>H|6`9P_F;-A@7*4e4|IKYShAI#r2CAjNe(e`+V9TAmKH
z<0@x$MkqX;uKj!WzOs$c$MXX=;whX`nL;2=Vxzj#l~c_f4<?Y~bJ0+d{)%i_7{iYF
zv$K{+Z4)EZG2cWeXXXG4PM4jr>HsY!r1UqL$`PchYY8ukxI3h##jawZd5z>zppt>o
z95dww`iL3jmyc{m<1>WPVi2<gk2ag=D(gC)hH)f`Ef0SbWo5QWGAhTm?x_5;f+QtH
zB%-pEb?Dv@nwe^rn&U-~K8)d5DvLlLTB69Cd6F?x>1dh+4<^L>)SBe3Iipi<VCZdu
z3$8tRn=iH3a<7wq)yM3>bW2rcr_H2j2X|O(eLgrfNlMR$kg~R#9pAO-FEmM+&xDRS
zS)6~p?xT6R9^PS~*~KcfU0ptwj*@{!F+j!#+BKcs#^LypQG=amfqdZzr$+yKyZd!G
zYE)dm$>u;RMG#v;%~u0!3AIuqgGMwLvK>4@1+&KaKI}26SAWJ2t7g$Fp_>`UyG2R^
zA}%p($c^J+6>YKy&Z)&dF7gKj*%STrj~_}i8=hc3r7+^>BGA_&xM#O({t<?OV4i0T
zFp$$8^yTF>dq&?~wfBvksT%$$_;=MS@fIOsve0Pnx?;O2oq}Kj(v^x3!-F5`wkv7%
zQMd9jg?OTvZ80g|%U)R3Yem-XyZ8iJlGW*ZdtLUc#W#Sg8~#4iuFXN{J-%YO)a8FQ
zga7bKF|smq{;zk`5%vGC87wag0RZinSa2tuvROD~bV}(ZM~4*>WHE|zk&uqvKJRty
z2PTJ)N78WLMnU-Z-yUAyC@gcjls`G%Z@K>p6N(a$6UJY?&3N3}?<5Y8sfw9%Ig!&4
z$tLVcd@g_WZk`wrZvQUt2#uzF9RFQPGW+r2cqDe-?Z)NxdOiKCI~aSoi_?YZrBqfF
z@&Mq%&fr<}0$WxWltgh?<oesh<6`US;(cCmb~JDX2;8!40CyJ5<gNfFPp$#;%=AiX
z4Uao_6x`5HnZk_!&!aqH2Ng5Sg+35};z7agq}>H9DjUK6sbf?YOk^<7IPlHDf4{_I
zM|dtH+FNQzKJJC8IyzmP1Nrx;?(q(tOMpTb&QB;z>`6VA!(v)t-zN3rT(|fNbg=*p
z-&}`iNqcC%&`sghXd}S83SfD751hinug>}eIc;%;N?qIs8)}o4I>x=#J9u8!rcWn4
z{KL6^p5UK;X0R96CT`<OzP#GjO};)nQ_kDPzf45kcFHYywNb*@m3b^~K7L&!ncZa;
z>%whD5^=Nh@3^p>%%Tbz|9v$$*Jg`rOGU%CyKa%mFd4*`KkJ8=XA;W*%sWNJ(#`vm
zB)T)5MIciYdQv+_j_(HiB(<hV$K?4_fRZhLn4M`e@NqY6SU2o;osRYcfP1jjMEX#C
z`aviTgvNOkdY*$%sb8SoS<NA1bV)`<La^GK1&JsgKyWGPUUlpXZMU%?-6JqqAky>f
z(7RO{HK_66v$qBs=m@(LC@z?(+4+}a-8my0qE&%ar;UEPVHi)3uB39<o_=)Oi4%Ie
zXgyyNZm==0<OAwtKd$}xi^Oc0z_Zauv2G^>U<XNtpqdw;O~)EY>=<cQJfW&xBC4|o
zjrxkwZwR-SxX7KZx)E@3HLn9<xyA?ToD9e_RQKPlrl7M2p*Z3A5U5BqL1kKFxzjz$
zSIkSJlAo>)C~WX6?WJ%Md!7C!geYj1Q-e?yXB_xsB0({1jY>b|0TJB3wl)1-&sOHL
zY7R9H|Ito;P!VM~p*T<`H-dPX8%kzoFS!-f9)(t^YP5uJWdu&jkYZ}mw+`>>GS+a2
zfckc0n+2eI>#o0W2Z)jNgr6S$lokC%I*jowwYSahsp=XJuP3;6N$a5jSVcq{sJ875
zzhe3_ME{;(7N-4XfHkYUboW!WaVtKq_NVgnbzzE2>ZFiQETrjEwNUQi^Jw$e=hNPj
zvLJ2!mj`eA&EpTX2MCu?GpW`5v*m1%wyFBVW#JhSaH<(+2RJ>B{_AMj%j=`(nTm=6
z4+W>v1BZ`>HCuR}2@K0eO1O4{e8V9uSo;$p)ndXik;=?`5y`VWH?C5qE>;DViCs$`
zl6j@pSqCG|5N4WL*}W?%$}>8!Uv3s4w~d{tYt;sh*Bs~KQ)25ipp-OZj*oFRT4Ug{
z)m7hQgJX%Lqi3_+;@x{|JPjRQ%x5{Bxf<(OPl0<=tOMx$T2DWy9E)AMMP;fDz}nc%
z&btfjR}!kdd?Dll>#_5<AKFw{H+?sESR*s<(6}<uU4ybJiClDugd3>7wj;|Yhfz}9
z_@3X&>w-z%3dceeOjR~Uv2Z*AlD3K1W~&+W8%xvDnV*d)1rb9&DrHeSs@WIHgTInE
zPbAp}$(VAx7$g+P^de*Vt7_CG8hC+-V66srErH8f71F4qG{<3wo61>RVEf2wI#pFq
z*l`mh0ekmW0Ck>fsP|@#UE$=##plqYhuA#Zz6>^Vq+@cyMS#Ar<x_~F<%7Qk^t-II
zNO(MUI0h-d#s#y?LpW7LIyq#Ik$oHJqIvfu2p%0>v=;dXB2Fh4Rn~i;UaD^9oT-+p
zuZ9+A^$rw=7HESy<q9hC0;aS8+EzLhYa4SGI}<`4L>GRNqs~8-ZWFhuJ=TF`Ch}Oj
z493Rd0q@hV{5l+*40#|---tkrsx!4SP1M6y%!Y_1{8C|Mdy$ISXC-SGwJhel7h^e0
zcR`O?vWGS?-vVj2i~h%E0N55?=au-YFK-&z{K7BxOy@tc@Hl?4T?_<vhL%t~JW%xi
zH*J}bk>h^`e%}0FzQvW=K7|Vn2!TYrH5oX2!;-0DSBi@hFOF28@v1ikNwVf`abFKK
z9YsX*;kaB@G4n6s!my!hd#z31SP;Ww=hI-@%k>fpX;FbIG5SnI;;<u;X^r>@v*WtU
z+Md<-={mW|J}e3xFl8GQlXYndKh6&`FB(jE%*aO$Tv)#G`_~kAetdXX_QYKL<NbYk
zd=rweYNX0rRB-2xcTsy1=zVSeysSljwA+OT-V$w31Y(knoaTfDktWlWTL66l=qJG8
zR*np%NX2yb^`T77zBdj!pM&7b6Og8voB$gH{0-ATq{!m1PnQx9N4IK{t@czny=={p
z9Q~5!fN3W6TQfRPh(?BRIkyp)(RGn4_!Hg%vc~7z37>h_Ih6e!TKwB-j$EXP5$NY?
z;nKV`vVL@BUH<GMKVL%i-QK<r0CnXW*#w>8(&~V7W_t?Mf4|3g(D*f4(gm0hQ%`VD
z%o0r8lkZd1i(=kH=t>jZbzXfonQJKPGmF=)ERYJLH)?^b!UEmx>YRqYGWBr@?Ph#g
z;eHQ(>?r0Hk<6Cgh1i7ozVS1Gi;}5Dg%Kaw5(9U4B@FcKj1#;lB@7UE@SEl>jdKYA
z;ux$qE5XkLq$()Pxr}oFr;K53_G15iaZw}$pot<`6NgnHWTZBw64jgmw@NXdQpVko
zKJf?lDS5$yZ(4cAa&u=GAzhX1?Z>9r9REWFY-wp%>u4-%1k1bZZW#oY81AnG{#eF_
z`;OMGtfF}Q?DX$7ZXku&d&nNXHn_N8sE<t>(@A@7Ab6;rPk9H3VHIAhT_m*iSS#ns
z#yPh_G+N~zo$g}hZlRfC%G^_@xUy3t>sH_B`9hP;NVZL1w!+ZDr8Ak1Xqc+Vu+X`i
z*3wD_t{kT<uEpPlmcw)SMzhZCo}CBi<@F_nKKOmKW8uaR?QMrXf)1PznhS-Y2$Uo=
zvg^*i!=@<GI&Ck^2hwGY%HKNWl4Ov{T!jQn-(Y6-flaax5(e6pfF>AiN;tg(5o|PE
zJ5WPpuqR`{;4JvdFg6)%P=8B%{4m2L%%M_>?8T&=NrM=b5sGc>LL>$2)q!tlX6nkg
z|1-&;WG!aZB0=pQ1OT9>bN+-O4~lZ(l&Ed4#g%ISyB{*sxl=ebm7Ir)?F@xg;nX_i
zo}d_J7pNt1i6l<RU&S3+Xbr1#dI%n=cZpr6BR~T{P8ek$PKT^*JpNp&#Hm>R2qywL
zB=|(?$P`K5?p^{N8gx;*nEY?x3i$_gO4TD@q1XLDw!NMZmP7N$4%MtnA+pI>f}|z}
zu4&e^Lrw*i;FZ}P*r6~+1-7To@az7jnqRIXBr6nz?%6aOj7!bk+sD9|#n)i6yk|C3
zI=*|E=rVDDY-?PB%uev%!sFTW2g9-k+9j7{G{15tppFE|-E!Y(VH4`5<G-&hetjM{
zXcTPpv-;sCg%`S|E)pAcht<39`?)Gy!iU3nvyt0)Snbu4o=y&CzM*|D|E}4eoMboQ
zCVO#{sd*SoI*F(3M&Di~8#>qe1O34QaC@jR5L=&IlZsqJpx&{K+f=9j|1&fP!++QZ
zn3&lAcQ2skZ@T;_s_$A|Kk;(-RgaO|rt_#Ii7Zo0_2i^<jSM&uAtWOt0|=$ol7-*z
z&II5#kbrWs;p!p^kajMY_w@<4+QBhz|M=f|*Hw|48OW6gF>QrR?H$r@nI|38CRs~5
z){#O;ICm!@k<-)D{k|Ey)5AC8>G^(`8QD5QMCI^tI?YY9llOZ+m|3}s!r%ROtxD(n
z{@%4j{p$NmGW1-r;r)L3C{r`b4mpF0W|$J7@uDFQJ%#WR+&ZgbmPHbZM+?l<DRSKk
zH<h5qoy577-;IG7%^&pUqaAy1tgae`RNd=1>!wsq@=Q=m@;p_oT)A#lZ-4zPq(XDu
z@;}FrGqs12gD4Lm38!EX%otC+!VGu%Qox`i$~iGP*?nxmF%9ry91g`GLJ)2?T>hTJ
zrnx^^8Swz7AAX?3eqGX$X0FRJX`nceN8llZKwgQYLat>w%#T8Cs=nRFDO@bXLMN}!
zl2*r-S4(Q_WcLGNe*G-Q@tOz#5qaTvC3k7o(8NhvPUJ_6_5P<2y~mF52gb=LgS;<W
z%~p%g#`4uwtC?s?;E+Z**pmFAl5%=OsO8f1!uuqy#ZUIp<Q5{+u{BfE3wwq{lAkB}
z2>dmxNh4Km>NP~`AxkI*#%|2B2CiDe1?MHq@d8uT#q3(ac_M*mF@mqQO44E&a7)u@
z{MqI}w%Plh-mR*VxmjZ8A|MWDfu>INeK~=H8Kca-YPIRTS>1C|av>W1#QTN|+xc67
z+Mb{~U&8;OrhZ1LWtzWJ#{M=mO(kg$oqoK)bAcdDB%WLliIlhyVP(1G{-iLfaVD5K
zbOuNR@?UUA(u>#lwWHzOF^^*gv|(;-<wq+eS{$3}{+(wF2sdgB^FieWM45@oan$Ua
z{<GJsIad6Xr)=>V0@TU}LmF<;>d&9^Ll{W>j!`H%Mj?^%403b@F(yQt_jOF5RVKp;
zP$G5zgLtILv|a-z7e|SJ`eVeSr14ip;?bidL8Zb9uXq%j^HPe#Wb4Q6((FVD6JrTZ
zVSLN6k|j8=8@j*d;`!H5h|+tY*847ktGauH6eCcphOY8j+XV2Az#-~pyn;sj3Q`Hi
zVVr?NuM^{dx%n`w1Seo9|CH4fC<sWdtjvG+JtTJj4`b)loe8w8>)0LJwr$(CZQHhO
zb)0l;+fIjHY}-7&&&@ew>~*pJ!n~MOqpF_g&0d}^&Czm?nfaHiYe(&+@y5%#k*(e}
zOIZe9oK;tG0k_OAqUZErw$9#UW)y&r@lc9V5WK@aBit@1yqFo#Tl&!E0~BeQKjLGH
z5q8Y-=VW9S6iq1i&hz0aACxGLbr>)oq^CfSzCvrKV>$ZfT245>m}t&zzHx~jDjm)|
z84jMcLgx<k$>@@X9J^nAQAJ(kHnUF-))2b~yeuYzpmbb@#X&)@sGGxTM`62^c#>c@
zrloQi;|t;ExF@vJL$?7MWyG9syjVP+0Ge$9bZaVO2*;Od091`?7nKju0ksvh<y;RF
z7a3|@#$~9Z8G2gZ>~g~|hz>?lMz|G(PEGB)gcL4(cm~Igd(R3<a7Rjr8*Kv|D@L<5
z!H6To6AchT55DK4wOnbi)NawNNDro8TYqmOA{i(l9-;RTlHKp6W`ec<&Hk)5*WDlb
z^yfl`H)xN<mFAeV`TAFyaYZ3A#LOTottm4(Wkoo_CsZk^3n)BuV=R<dw3n^x&?(_u
z(vTQQefDF?vUVOERcX(l?1lQC1LS*_Xlhk@|AkabY|Ugx^J1G5bY@1CD-$Rsx2X^W
z`+76c_HhEGaGNWs3ED8*jTJU{qLg|Pm}%cHv0_1L$uprhmg;(nH}m>XW1#7;)JV1v
zm&kMQi>BsiV(c=gY2mz9^Nh{$T76ZUTs!CsefUbsiiLi|Q7BjVOPsH)KO@qD;H^Vw
z-81J>Ol)a7um%}wyv$Xe-M){c{?s!YQ4%dG@ZE>kN@xU|F=MVyID6RLusyhl80ftT
zn~LEPGW4FpIw`jn-~~UD%s3DtHV=Z&*64uKZ3KArgrd=iqR?zm-Gd9-0}3yABSYNo
zbk_`JK@{qznl&6{X<>r{r_1&5jkqQ;*w$)&Soj4F_mDTmgV2G9JBImyt;0|fvess!
zim=EWY+0e!l>;9@{V7I=zqk;sv1rHkO?L?Kz1UH`gJI;t9HwRjYKy<k?x&2q4^qZR
zVNXnjkMJs63NhRY`77ZA8mmNRv|JC9=?|v$&>-;)QdmfY5MU!12&Da-t)jxjC6o<M
zP{#&_CC)4-kj_+SrNS@~9C*6mS92<&)@}SqXU1a@sEZHKH!lv?Z5?mrD?u1j;}>Sg
zP4E_$ak3xX#0PG7Uy$Tq!*e@LJVxp$<vR!y?lo9O$<_7C+s+<4U*xVaif^(}ijDwZ
zWP{~bQulpX$w1osGt-#1nu~&?iaNtlKXjNeo?oCX9Vn(KzFjmFlVppsZ#3m5zY?c5
zT`_IhsW$T>KuvCu_w5VU+SdYG_^(@K>|X$8*Y-;7V!QSKQUEC?-T_?Q4*onI7w`^Y
zZSbj1q*M-dlICq@kdy!b+NM9nru&TIb5ePqhb7im?_@QZ;qqVrY-ire!OM3m;#n+A
z(qlo)hbcfo<Ah<;&D-zsL#doQWFn6gH=7k~%+uX8Ju7s6JoRf{f-o9g%-Q4dB?D$%
zzl7LWrv%t@p7yHD@@R4gMr9DM%;;Nv&9_l~GvLea&c~5SCK8@qQAL@D_Ljk6`^zGy
zPQSk53IjphSAeS8?1r1;YpO}>*{x{)O_E}x`CE||UhzmLfF1&)BfyZo^+cRTLtSz+
z<GpT-B+AMA?gSKWaOSnhID{AdiV)K|z~ip^;Kvmx;F7D!!s+`JWcH^{{_~_US3Y{~
zNP4V?&Bqblw$0!d3H^_M1DTIL41d~9WBI&d%3qm#p2i@APO-iIfY%?k*o`;W@5Dad
z{nmyay7U0rRR!onO2!|Gpv}~mrp<_Suz8Zd7SVJczZ}cmV?hj~0T8m|^sB!ky_}1G
z#`vl>ViO=>HX@y3c5ppr-5mn%J6w?~?0{CkYu-!Z0Y1V9KMVn;?c@KUX=3^h00$En
z$A7<t!klnEV1K@@+f|xfgK`H+nJh|i^5#x+Wwzh)R;J1hCX-hw(JpB#xxaY`gj6OK
z(YcmqmQkZ8egx|F8~Q!LT{JP3!jMi;;7=xI9JQ=VF?At&Dk>#cp*rE5)~zS$|2}ko
zq0ndi7haypB+?w&F;v9h`)lN7DrRcQrS>#9Y00JZZ2x}$=xF30Z`tb;ck9si0T|(M
z;B7_0o=`DXsa0kEx?Y_y(a9k2xiI#+{|*J#%B~lM62_7#VHLH?IRu~}U}z6`rIbWd
znkDzh)GB=6HOM{_GXp!??@2&+vj*#m(>N-Y1B2TvWCckP{}TCCr+4(Mb)jB<lD#<c
zkyB$G)9N@QSzt(R4ylQG(4TZY{m6pv4<RwZ^*h(G%xEAAXayu#C!_7jUlh5}!Qf{N
z{b=$lu$ynUZ!TZXmYN3-r6*r87iI!W2`G~JJB3=?kJYMdPcospf_z(gK;{<m5*aa`
zFoOjhlgLzM`HQUzzTLc{$ME-|Zg2<;i@K=lC4idx-Q`pD$v-De;JUgcgOA1|2ae-A
zxPWF=yr1H=<K`}xsE}#}Fx_;Eg=~>pGAX_f67}P}md!cBpake6lO~sflGg-;RmqnC
z{T9-tZxg&WbGo3OrhHO6cy8k-Z#RzF`cBOX(?Q#o$jNNU#5VsEnM@@{iio+_Glr9J
zKhR7#oqbnYqb;UxztO$jNipG^k6cNy&x=bm1x`VIbRc7amz7lg#Gx!{;4BPb=O3$?
zot#>Ax7<}st29m%k*$B#)MXh=J9km|qNeo6sqIgJa`3#UkCP^h4l>o{Er<m#@DDNs
zGt1`uux5I$$h@*!<P{b)Diif^OY<ErhHHF;%Q$M~FG@2E6+9~@(qG*{r6%*=HsPky
ztWR1cV-KiOoGheX0yH3y|B7l8mnbsaBE@}#7HaG{31!J;)aKl|F4Dl2>WOVQcE~%`
zZz(#mMU15)sUk?kzqIK&#lUrmXT}$e)bs$kmHf9-=nlUAO~2NywCft4J;#RXoRzDX
z>r@Hn_E%9js7P!|f0KOtgrFi;;;_N;nOExtT+9cr#9zrMYN9l}$JwoG!fEkbd)R%H
zGelCZ^ED&OtrgLBcPTcvbzM6`S--U8HpD<n@lkX(w;}?BNN!gnMD;ajPQlM7-6kry
z5uZ2ZVa}-^z==d1GJ0E|Ceyb6g0+AKaf~B6ZV4{O{uzG(&FNp>;xEx1pL`YAY5&w?
zsWZ)X_Xf>ls^h1mAU@8*puBxgr90=A4#?mu>>;WI6I9yrCaY7L@CVf!2_~m>Kte}F
zD1`fg*h37qeAgjA-M4k+-U*?9pw$Ob@_=HX4)smqYcCt8egyqTAfbUZrzA@l2-3j<
zN7D3<q9e#H08tfrIa2f;tRoFRZs7^%82hLg*PU=cMvhLnM-o9D0(aRK*x_MT4E(B#
zajbvXWExA1&qDB<bDB?a*iQSGb<r-Q-=TN7?r#8=4+NDT8`D+1wF^Vc=`aP!8+!@V
zJkjbDcGNMubq=9^MjYHEc|582+m;^^YTiZfSJF)VEVz_ibPWPGFda8!#4@~;7vz(@
zTMb<xJ$&OM)WcUd#lv?z5mmt^7Z`25;}W3`GT%6hnz`2|Tt3Xs<Q5(#J_l2T?obe)
zJqQ)e?-_rSEvzckPLxxp(PTsaXJyr%Xbaf^%x+atR<X-aUvNX`JL=Km)c}ktKSAu+
z4RU-e2(#D#jdY9OdOWH~hu!9?2=J4|i4Cl<HCFcVG}RZde-F^yDOtN`d~JA>KdD7j
zKHyaO)<G*-|9&i7nH(=9Ych2#*`DV(v;C3W)UGdy6Uu9OGuMuQ@M96Pn{%4_{5{Is
zbwb#84<~uA$l4CftH;v8+akDwq8@|0*<bfqF+DU&Xl<gUq^AWvCPD~&DTlKG99SsR
z-ACn!G-m{JN5KNdYSysb+zvrO+x>Es<uOwnpGinue*V{A5)mgzKf;6|wv+)zv|+X%
zBRD|6CEGDY-n+pn!pdj)otrOW6q!WR<Mb&PROz&Jd6zGydkid;FB(Gwggl=-1jtj~
zSczbipEocv*iJxGGC^_A(cn!Xz*U}CBs1xE{!1Ufvhk>Z<2MI|v~CbE+)qFz<rkOC
zd!*K4fP3`g^jngzt3T;?{o6tb)LVThDgB}*ih(N4^!B`~)C@X<UdL0~aGhSv&a}s%
zdc!VQE_2Z2la41wcJ7~t8w!L;4!5n)vdtfxO}bF^r~rtTnV+Yo6e)uf3wma_(ltm{
z>Lp-^h+Lc4t5N7EppK{MO;T`FIEk-NDlX@8DtzE-A;$W09n^7pBG6z$)rx_VDhxF%
z5Qg|WXhoJLT!G6~xzJY8cdU^ENuoBGvH*uKCH6X@1jh<tv-ikPcX%}zdy0ZOkqe%o
zsPI7=J<3zxsZP7Qv+}b+v)+b>+uPBlq5%LgDb0oT7HThE6b@8C^Xdh_e1}`!*0UKy
zJ}OTWDgq~eU(a*bN`T7Np|R07vsFXZ*;DB5j)$HXpOPOxJkRFE87d{=)j81ZMHUmE
zue(^}LWs!ATZGral;egp1zuvkH-u+ea;noGLt})8V-MQRvE=IchOtzj=8FGs+H2U#
z;f@4nhRJ_w1^(0ggz>+}Bet}395(;&{A8D;jcLrKkJSno!%`QlB4`8Tz3|njS>_6P
zn~Z~0qYAsZug8MEJ9;B64FVVhl-PsQrFN$y&Z$w(@%|fTE<d27*ZZS~2wEu;c_8JY
zD|8~%CL~m8;)%RNib~tB_q#|SY10(UC0oBPR5C^+-*57Q*g+i&jq}T=W#T0c@bm05
ziyRVLvWPS?Zpj&slFS%H>UbzjTp2V5><Pql7RkjdTOuccyv@<tvS+7<=lnvhqO;n%
zd7M+PpSrhx9ATmc@OxETrYT;M)2~aFU7$7rD_`v*R@S?@YzTAs9$Rpj$?3K|IC(CD
zVK9hxK6sLX8WeS-4vgZIko}iE;|AH)AqyCkfpX}9&@F*<NaffQF>DEo0Ew4VN?Y%J
zMOFyQ+f#MXz8S2t+kg10{%}B(ygfG5y6LKV9pj7>iVr<aIXDOwbmgZHu*1(~ABPkG
zLGxTk%aPL;Z`*j6aTC~%<Ib7S(ts9`K0jU7!!H8;Nko+Dk$Znhz?V+Fc<kW|B44@N
zpX3oKUB+&Ph4|akWX_1|{!$x~#>9rFhg0c=k@;w(0k(AhXh<+b)#XrC@0ugffJ+|P
zwdTm3m(7cPT3a;~^WeP_Tf8el>*z3cYzY{=Hi`=chdi~C41*#ahZ6Fu|K~Uu#=HVa
zqzxq;J>!9d0tx{g{2I>6AW7t0hFWA5e$C^V6?B!oD4W5Zk4?bu;|<G~f{dqiyp)Ii
zy#D<mliQm&%j!qDp|_^2{xopi^-?@B6u#mQy<%czzv^_;Kt0J`AsmT$w#Xv^9omp<
zeb1SzC`q}m042&K^^aIX8mKw~$FT})S5{dR;~IlFNThPYP+K^7Qz-A&J&{R<jpamf
z+?cv;Ck(mI;;^@@7QAQo_Q=UVY`23H>e3%$JpthlY|u8Q!P-{_si7>Eme1^)mrtz^
zH5_II8k3>=r63OQABG#t6OY#m+p;NF(E3v$&j&)R#$6`O7CF&5125r(gP(F(DAG4U
zj|{SNw;j1dixe~i@)F;!_lTCfIj#FNtUar)eM-@)ENO!>Vm1oLZM4bbM@WiP+6fRr
zR|Mv^vAD@!&N-o7y7)`0ElwEnG(I>rEvZ08>aF26Py(XQT1<~&@9W=YC#biOmf((&
zl>Kv&S?5!@4fvw+57h2JU4RQ2Sd*#oV-GHhW3r8%D%N676^hQ9LQL*+f%1xQ?Z85K
z!*pKTxc;DRvK5x>(VgpOH=<0$=8YbNpmZGykHrqdM4W&soE&!Mpe8g>TLgO~R+T>K
zPmKRL%aN|bee|zY;~sY3J-xtoO2cMJFtROFqpa$LzEH!;2cErkqV(r(wKMY5JMyrf
zS)Os*>q;am5v8gG86U4qeeO0^r81qCWm(nDH8CM&_H{pg4hCkD01<o%oijvG<X8hm
zk5^M^o(H)kX!4Q@eGIIRjr;1P&S`BGTMx2|ZzwX+hG&`=t;hI19b}###oxl-+OY5e
z*uX!g({1y@^3_YeI3UX3zNwCY$h&f~P?E+XT@4lbA(Kerjx`&ngn>L!ka)~eATaPb
zuUpj+(0lSXVcbp>WhyC^T7j%XNKvFq4d0umTPm4N>8cLsRv*_F6T}L4^dL9hn>Rf&
z$(5$si-O<~5kV|dp43ez9I*GN&1GK*&0GMcPQp(<o-?|E^%P-FuD9HkLLzDSJ0kA-
zH@JMKC-YDjLz?if_X3be)B15YbwIRl18CL{=-&C8y!xka+J4910DUuL@d$(+|1crW
zW;0QRU(N{2J=qDSXEPCVLEi^$MGk@3&|k-ghXML_9K2+%&4}Y%aN#pr7tkS*i*>+*
zq7LY@t6ryiV?&uLYu5=pA86uvXZN`4fbOp!<u9`Oy#8B`AUhRX1a@H=;Q{1#jX`zE
zkCSE-`&4f+|16eSWA@C#%i2!E4JimGvTENJAf~N7`h^*oMYDjJFc&Z{iG*3cw=smG
zc5aVvT;7SI*J+gN134^}iGU#zUEJ4dGz@E_yBswFvb`$=rmxLZ1JJhE>C(?nb1+Vd
zTgn3HCZNuy%4>|}sn^_en=DC=U=@uj2mS+7&?$j_Cx~BLAv>;;P;1Mt5Do>t!JSvt
z)UG7871fN$f}z3ei!%H;xP@1$nZxu8k>i?yXM|<o)8u+F9(djq`_`t{4iPJF4Oi)d
zO<Qg=Dmw~E6PE0b7pkQ5;=&Tl$EFye2ayO5mjI*WB`9uf6l_u1nI)bu3pLVtU>hom
z+zH0CU!guyXkxl@2yh}Bsyux8hxQr6M@{3she*Y2%7m+#U47T&XI++lU65m`tyc`t
zACp?2IO^5Tm$a7Q5AXA#WC0CVl*0-OBi!w;u`iiKh)XIjb{MP^nTY--5F{NefesG}
z0K4tS49M$euGaetJ%1RG!;iHz4W;H6v5&STyPe7e@hg3Hc*6x>BU}P8Fk<>}B(3&u
zY+tY@c1-^@R$YH4wx`x)bJv%V+skVkLaDGm*>S>NR)fwT><Z>nX<hCc4*Lu!=}e}_
z!LJcb?)WzpTKT78XWtvabl@0pKWqaPK*;oC@}8f*SS}w^ci%>0Qmk<d29b$$`1rcm
zxGEhmxB?ky^+`w?Jj4cniq2uFp;_cpRiD>rf$QF)y{(bhGA{O9eKxVl`wxT&2UAud
zOQhNikWVZjMV7$C`XQxi!4EKtYCQOV=pvZ^!`7LZ`5#pDUm781G=)44T9N)IRY9_x
z28>V;@v_y=0pPuKWz$;u2QUgLSxnliSR$?L{<rtJ%_9&?nUoB0XTNbwj09@Ph&es|
z0?fMx#`}7F{_;>=w?l(fLc5#<e|#Z$PN<&kga-Mbc1xvt{_g*MQnv$}4=)~-%3{ZQ
zdK>b6Tkf$l<6vKBp)@@dsW)dlGxQeW;9%~-(MDi@U4H=h2lXE_c*t*|l|NC$xTIr4
z<MAQ}#b(KiA;J!YN{76`uVgu}lvBg^o$yxWpyMnTPn<{C9%*yyuCEmP8XTE@gim8w
z85-iz^}IEQ&ojv^dUlWBrL>PpDQu9-4wfh$csV#OcMI`MMo$COs%r!<_8oJUR|i9z
z<HTB<Bf_z-+7X_H<>%NLI9SV^)4_x9>|w9FJ*VM5eQ46vI{wFlZxsB`z1TaF&~3Xh
zU7MRk4gG*#>b}nY3ZCsZXZ6kEA0s;HSv5*SARw0U$zIQ<jL0S&UoO-vX}@dv6+QrR
z!@yi;N&N}JkW|!KLU;MRt+$N7d7Qfse~QWjT~nnzu}?YZ_Uxv=%$O#z*I88n&?C-L
zTJUsFad3Z*syGb}c1f?8sG8jo4v!IukwHMf%J!eTHR!Z+J`?U}nF{v)%Ifda`@;yb
z=@CWpfb8DTWcMu?0{Jnun!@CnO@eSHh&4cLKQAB4j=uq!gW<kM7b$i+1T}n>4mPuB
zOn^|$7G04LG4+Rz<2zF@yAM$?`pnaI@XqT+TxnR9IZK7lluax1hPfa}Ni%7@m@Fad
ztBAAM@{;XTwC*hz+fy9JoUr=~`5V#joSo+RwT0`jx@J}4M2nth0sDuE6CsC=J%F-W
zq5>o*>EU(4%VbR75wAo5RdgwBQ)XK_Q>yd33Dhj({FQbQv1S8|rNR^`RY%~Zc05DX
zxm^MMC+W{tyzY_SasAc-N(X<<4oIUAd7)-8!w%Y2x^A&+yTq#9RX+1hv2W0_lnB|W
z3%GXmr)hstA(3Wl^MgCG`;(kUwtK|Uv#uKRI!vW^^H{^|7`H={OcV-LI=Sl89S|C)
zw-VpHg~0$PehBj3e&kh|$DjT34uti_Kj))33`1kxTk;T(*dva4Vk_0bTct*=43~=}
zzhWP>%Yx{L%ZP3u(z`rqnzWKKG{x%->!63V8Ncc{22}Lgc6^U(a^9AA>byTlMH5(9
zjHpDvojr~YkqNowWC?$TY-n2yyq`N1*pMW_&*~-$s}3sAIxR{%%M3$R{Urn*2w+~G
z&O~hjn=_p@AElxpK^MwZ6#rf-fanK~G5QM^Kd1yHd#rHi4X-mwKr&yS1O^wfuwbc#
z`1yYD^(5<p|Bdq_t#J^2<27y9u%nIo4fBJpB@<kCd*fyk)6}xXOrW%UiI)8t%3@b2
zyV)mXhwq%XOUEq_)P3Ht^Zi7I<dYRF3>`BHr&^VIMv29!m~~n|zH$Wz<PbJ9DiZv2
z&b-Nn%O&*rH1k#TPy6c>+wa$0!2}Nr8=4Liz}s8_D)?IqgowNGgk%;d9{4jv3KaUV
zEU?;?z*F($^~iYZCKYGbWkIv$w8C6K+A>|3iXaZex+?@(y6gOF&z+5mUzvMRR#NdK
z$L>S`A%L$0q?p5NjAz`aSK14PEp+-C*>S=`ypc6M4>JD|IrS~0`_MY%6ardU2`Yvt
z(s`r=*W`32!A`w1F7XX7?(|Q9#|5)QJCKBsfkHkP#Vv778r1&mP9Ekh91K|_mItb~
zSNNVOdQQuM`2jx2-~54QN_~KyKiwel95(7DE@Nj@Oss0HI=or#coQha5TiS+ml+P&
zVHuX=++%YLB{=ig@9RnC0h8iHRNQ#HlbOVTP8!g}b1w<$hHbYUR`zotloA<*oIMjE
zzi~B<VM7{i#E<KZA()y>zKmF8cEmCB?xg4YvpZmqNVva3BA0Gy9Cmp_Tf44$)ltxC
z63DNxAsKlI*O9`=_>*F8lp2v=<hJl1erj#s)@g|af}2fH9y2_s)mGu+Ks{2z-5J;V
zOcVt|fgAJ^ZCt5r0rOLKy`rlHnK$rjTD>=Gw3{;zHCRph-3=duYYTsumOUo5_^M;a
z*tmj`OW>B1-`rq0(Aw%Si`T?RL-WfIq}@<qys_Jf*Jiz#Hj0f)qMe{#=vOT+NLt6v
zDs}DObajMf`-G(GI;jZ3?;dE=3%bB3L<i`}yY7Je@FiHzFcIPKfeRI8_OJ$?@QQ~)
za()j=Gwp}BL!E7IxgPdXR*i*!5oa2Jy%q!QQ%ob&Ml;KQkNtHJJ|yG_Y?JaxaI&vs
z(UGm3au|<bU|=LmM@pP*maySWh>sEwq8U2~;Q1i_em=Hf#-X0QnEocN4)7pua`@@O
zRfHsFlf~?0^>bV%kt)JoGs|K4N(qrF+<lKRH}>TU`sPQ_vHkENbUmmhgsR`BtH!IR
zr~oKdp(P9<*j$b*f*hQ&p@^2lK*#CNn5d(2E~?=o)5lx<{nct@Y>q56f(LPKsr&c{
z&9t|)Gr-oPIPgPkBe}nT{Jj+eA=T=2P){BY%rl1l(%RZp^MnZ@M?Hu|C8qhjQLagf
z3=ITGnnu2@YR!2lHjkoMJjkH2dM>K^SdD?y!G$<=sl+S~uFR+6SMu8%o%2D~G*4f7
zhg2><XQn}AQn8+fK~gJ*KC{s`^%EBQJNkIITS?XfHCkqpn$HV#`{!0{8gUK&-G^Uv
z5Ay|~=Jx0}8Yfr(2}yW%j1)A~doOX=dV)J-{k5<@tZyWk+7O+aB_AO#F|#bqel;KP
zf~eqa55<M3s9|#c`YF8Wo_<<K`59L4!7lk1ZbVyDf<NIO24Q+qngedUoR)^Jn1rt1
ze>fc%t$C5K&gZ4UR6W?2Jj36V#Z3e_y}_Ul92EQD;N*dhl4TMYv`K$%Rxw`s>Ct+6
z=W)K5EJNmKzmDjSk1R;=aPgYEB9@(-uN8#qn+(DH$}c*217-)_cM;?9<4|+sS(hY!
z-@DE68w|v*7G7V%x!*`|`)zb<pRHdkD|p=`3GDdHJrXqT3a~}}?ER-6>OV|knAo}g
zYXo}>bLt-&bX)sxDPv6w1~j0t=t^#v*6fxdi&u&<2ALgf8vmv!Iayo2<nM#AR!X|q
zXiAfZ?@g56Z@kmhAIt+A!Q%kAqX0R#1BK6p6ytlwv($0$9$)S|nsy><G5e;-JRoi|
zcQUkEP)33TFOVy@oiE4l|M~VZVM)u!@B6YhHc>|)fw5$e-=}oVPL;rN4beZLteii9
z<+%U#DM#9xI7auZ2|*j{yZP{b$@0Rvh&~7y#2|73j<GBtmX0W%BgSfh5)?0N_u^X8
z>JibMEuKd_j0alCVhBBO$_NW&or$A0@8i)u71nSH7f8npZqw*weqMM5=prd@tYgt`
z>@=4(RJyTKB%j66BgR^Jp>r?^OW9`py-kK9?{^8`J;IBhQd=q+F&@Hd2Bnk4*ilL{
z#K#;uPfGQ+7k!`ZkX6=WWy2<?bBqNvRPA@UTZaoOANR*--l<FRPcI6Op$*Eg4$;bG
zH=ZYvUpDG8!$F{fFFxdcYtrRLq|a{Sci1rF`C`???K8u}r&AjTG31dFWO}W>013E~
zz8Z84RyO~;Zg;OqMLo@h#yFOvWu6<6%gbVs7Ntr>5$1fSCo6{GMF$FwGsHpK0z)R=
z*dX*9&vEPjk$UEv24KMwkdt7Nxd=gY3D9N3b7a0hFOYt8N<s-VpwrPf5aHn0P15X7
z4T!x$gWq3G=V}po8JypUn-o0kolQ|-iiXKX-Tv4mzDVGNf*@arB)q<T#EhlzwFRqU
zS6OAm=Jhp!fGEIIZ9I(_NZ{mqK+_%kyC3^q?OpfQ6A4WTD^apiE9ot{$El+~qHrs%
zkg@g6`L+7#RGDK^;_^bHS02Orjldg6@x)Kjl{~Q<`Mz$!TXuL|UyuQb+^vE*U0Tkm
z4Mcs?{YS?K3&*r&=H+o+r7GqWGPlFB$1`D-?qG+43>J6t%jLAFeL_R>XZbX9lM9hn
z;brAX!%K#5w`PpfC_{u}yYHj`qnvLiaouO6hwX-?u}<py;R5#a4I-lJ!PwW&iO}Pt
zEKE-H$N@LHaq?h#jnN=4mYr=NvO^3I<BqhW!Q6X1yJABo2uYQ7Q`JiIwnsry?<uN5
zgqe{O(~b&^(mNRQN=g-jEMGt^7Z@3Rn*Hb2T7Ot9d4nHRJ)3p)KwG1-7AfR)eMGG%
zy6F-gW4}}X2pgkzdcvIkz%1`nOP9kcncah-p67IO1PzuC6_)-fJRM<15`qH-?5+$!
zFmD(u=%~9*J=IE%ka5-q%wnS|L@uNb$dA-E1MF6ttf_SPj=VN9z{{kIyb0o61n1m)
z&vB!b!!Zbf9XnD|IQXqXIPZ8$b~TO(lu?UJ1FGdJyO>{=3K>7pbJ?)E41*C&L~nII
zeGi+{I-T80Yq&Xw7RQw?#)h1kQg|3%B?fUas3gA;Z5+cLP@-Ar^rk3U#EM$8;`>@T
zzGxGdY3i^YjaxXrCvGaJt#bTV6ky25#Df1=Rmm2m5M=WqRV<7A;3#7%V2D)y%(U7n
zqze2_AkJ*men&1l%8Kn!d@%#>N?C{<vA=m#B}BONP$p6Ti;@KrI4m)$OyKddBmy8O
zt>aZ#<k8lA%X=)Qzj`~b!Ex78%OlsNEx{G|?`c+G%vzOGaq2}`qxGAR(%5V&YOL1&
zG%}{0y^6Q#0Lnp_>5%N*B3pmsTDOFDI=B&vWI`bszs_2dhPweO>zJ)K7%FiUiLJA@
z{p$mlry4spuFs9C(#rjS0xMgLJgcc>TeF?3&!&E*ZL@?r3gHO|o;{VlFJF;>RN!$`
z&#+9sGZk&~s?6gI+HsSlQq!Y@71w#z>9!tuec2}Irdbu+xbF<&KNQ`<a>{=OT2G+%
zkX&ng$t8OwZ@UqOU8h?KA7T2p9ZGb;dhaI<LF9%tlj(BGR+0e^M7&Q)V?Z;B|0y*0
z*BU8<TBdlJgb{V8#2esp@r!bH*zCG^ie(ozcB$*C+t`Jih7BFjh?Rhe*D53C2F>}h
z&_k>TTH#KOi_#U#G7Y>(_H{M6H2zl9b_@vmaS}OY_knAst%MGc#H-{bFH&?{!%sw=
zM!+EN>lz#G+9K=di%}rdD)IKc2lWk34D+MNyF5Y;W;ar(XG@85;XNK)|1}GZ$`wI?
zNT&QU65)ORK0q|1*A98-v*ImCiNjQcVXr6lvKdbpDYk|}ZEq9g^i$RTFEOszKy5J5
z39WSifWPaBoi3*H%f}&KRd3}=;9~n<mW3?j575eo*MGWIXZg=lga5wbX>I+>ib4)}
z`KNw9Hzs-|_v&>gUv<gZWOdQdai}H0jEd(ZN&ZK}$~jzrKVO3oiYcbnaGtpylOF(q
zWpu^t_!~A94~zvu2<U$LIo-YcPz8sQ{xeD5A^ty;WS8Rk`qzFzU7rOjmSRIFGM>$@
zQDm>r$MH_gMb7_-B%PUaXyk@Qx?WZv02etPt8LFC9|2zn?_)Zq9zuD(yHkGyZ*-{#
zJA#?W8&!(jv4PA)Rfpk9Nm-Ky{v`>95ZTaS{WIaEMrJaVg#1y&gB$O4GOg!cLm+YO
z%QH?^5k=nJlhnU*!*(nZUCfu^{?nT7D}BWTpDdaLAuME!T`;DY7ZJjZ6Jai02m-@S
zHGe<~g)yP1&;G}+<n3YlSU6cUcGok0Z;ftPs_iuk=A07c3V>RMM{LG+;G~IhW{zwM
zuir)Sokcyi7{e$fXd)rZVBu2sJ5jFD1X(Z^@^X!o6A9cvD8*7!CI3wDcHUWuR*~1m
z=U4@5HV6kCimfL1u_AI<3D<Fri3jj|QL06>X5KW()q8;ipO%mWZd_k(@^@(U!{NxJ
z7jh2FKOn4|KvC`vgcwTzmMToQ6mcPxfSarzlOjA8rQX=VR+{3_YCOJ2q#pVNq<P`T
zvY#<UW;QC2Q5RM4yEb<Q+{8$2P7}4Zaf81VEuMFJ(j3ykIFhETclI)-aWcoAt~$mX
z*^MHFEHNYO5uD&fOo){~{~#vVYa9>2T8^~m-TdI%$AE8NP44h821l$QNZV!CSg%!v
z!=b0X4r&Z0zRTh{-PCAzxr$$HrZ%gD$LNTRpa@A&;uyI;)8>|)u?`b2Xf<%098SRe
z(XDZeS^+XGFX5Zu)tVbkits0>5iTCUnaqE()JZA5=9Miwin%s6M?||gMELQH6-MHS
zHR}g{x&}cT9aa3>4#}ozgf^+DX>k5%0b4Go)l03+ZTlYqyByL4*hp&jT0pYc#dgbQ
z%Xp=IA)P7P$Zo2H{N*c5iX_U>RCO3f7ug>GaSaSR%9rYbfXShCAPI2O-q2DjzSyUw
zz{+AzkMDmgkPE&igLnt1etKDulGF}|1SE=tyqnC=Irz6@QXwN@-GzuUQX=w6xug$^
zis;U9kw8~i^J~@M={wl>`oxsevsjdBQCF$8IxRh>Alx8ta~*1QxP_W^If-@*eY9RB
zD&*1OSIv}pTW_*UAv>+Ddk*Unc!3jkUshj?j4)vIs6Y!3h|{h>M4A~PVrT>V*`_Zz
zyP@;yem*MG<slIsg4aBN#Zg{LY){5@NjY*tm@uZJUsDsYYNFFxPB%*f9t+<UAP5}g
z_DRn^PX*w2PRw!^BZ@Q_$KR);rlwX>R=94oXvchjJydEmeoYrF7c`_f<myHn8GeT#
zZ;U^w+s7um);zWS)7Ci_bUN$ScN2loiQ!>5j94>7Y_OmYUG$#&er-6@dW=BQf;z@$
z!As=qI=23C=6w#I%#FxR09z9m@uSXl+8jt-VG}*D{Ou9*@nGj~ky^u|6OFL8_QU`*
zb;biyJ-eZ&x8b!BKVYrkJ(Me!csC3c^3%J9Q#ogu^A8CSa$3p#D@>M%yq^(@u|2AF
z3CkDm+wgHdU!I_q=ReKM<7?K2D#7tYCN(4j5+J=O$pLVa?jv<^<V1jHiReMuMNJk}
zXK1qffHaH#AxP2nFI}=|+xm9WZc>RK*1wj-zFAs;7C4_e7NCjQkiJU=2}Z&1Ddfk3
z&FAs)-Y2NCusweRQt!BmL32DpzJe)LI_1BiG&xE%+QB(8!FazITsg2b55dl*RP2dH
z0(sK&aftd}^@4QYaI7A=Ui6w^aEpOuuUl`?5J70csEb4?D=7f{ET*w3j01(PpS)R!
zM!fA33SaYxUw?Le=6dn~!4{o=*V1azhLcX0ed4E6GWqRCMZP8{2kgVpcg92za|gvl
z#9x*ES=;GHHT`RCWAXf5ve|=4{v{{d0b38Ffz_(H`J@phx9k@Tj;`VT!?Zb@Jg;W>
zY73qEr@u`5PEUX`d=f)<d8w6e!yvXh*9!)$O7>vuVCr35RAF>}u0>(xs<&nXn0fu&
zy6<GD3|Xa+;3!=LvJ?^B9FtC&k*di+hsS1@JxAl(uPL0SI<MA(wkChmA`K4`*GK+Y
z{E{)^q?s?ubr70VLSgQ=AnY6NJYOIpW5fM=BdBj?XaY$A3=Ti5Q|+09Q&uOPA10eM
zY7}Sd2-s~rU5<ocL)F7h{F7yLl>^ZL3Ol^X5+|$PP<5?O3Z5%T=eO|BEEc@E&(Ofo
zN_`#vgWB;sf}Q~J2OC{$PHb%auUI*8<kI#L<OJF^+Q3;ofAZW|8V_+Iw;qYB`}^N2
zv^1@1dE9QRo_!VBX06gW>;}dWwu_Esm#2$h&lDA*jf9Pwy#w%CARjhY!_BLKZzQd1
zL%d)0XU~?#V3aFr4H(uL_mDgwj$U@<w|@EQQ`3H_*E!y|_3Lm!;&+Z(o#$R~oS1i`
z1l{!9oa`g`;XBeRi^j$^EQ(5&MykU|YXOZV)swZEnPX8(uoO1!TbkMEpi47K5+A()
zoc1Plr`MIq*l=cvo>59*2KPS>(+LmcL;Oempga<T{3~m)c4Fpuh4j7!kb1fvHQvLc
zL0*u^z8?1g)j|XkWDtKX-Oa5fL+~lgZjj=a2}CsZU0GI#U_ea4dqHVK2HHA-c44tv
zR3$Ohi$4?N)-Y+YZ7LUPWFq%`tR?p=c&Lc<+6ooe+Tx?qfH#*nNt~W~;0t-P(I(sa
zkiE$X1PEc{Geebxgh|j)U6;8%&nM7mjnijAzrhLwyQch=qWV#{15ZGsD?4umECS}$
z4N^$+A}*<*q%QXJv}BE5idQO!r>&^-rXB#BuKj3j!|b=iO9+&Fy`F`$C2p_HRkB#5
zi6UG4rr`*rCQ;0!(jWEw!xeA@*r_y~2730_G7JY?G&)jJD6O)tQK?4m;k{bcjkd&0
z31Gsacatq7Q?qn>AAzHsUL4(HgcbIX?OF>2S?Ab>d5@brsOxW@dxpA;|2;e4VEX^H
zBC1}FW`qn1#?~sXb}$U`giMT#|KsF#c5x--<YfAPR3Inoe@&XtXanqVIpF+u>gG)z
z5aUGkB!R_&D1vf;=n&Va=2vb*4s1bxN3A~920h;Ra9&L5cDJoN&4Zj-n^Q*d@|I9e
zT=qDH>oILp4Dn-1hwfNPg7wXY2Hxn)5<6t(PVgUO$*FU}h#U9Uagd!rtBi8C7&7N$
zCX0?jA}GNfn8bSyjg|*bNXDK3?IFfy=*|<0%z{q`?F3?|&hIksCBh|HW3DL%YGV}>
z83zNYg;;_%QWYXC@T?xdU}xQl&rH;l6l)m4h2#(TaO{m+o^mphcn;)B*~upc1rd>7
zsm;$*DYFux*%kWGXA&1>Xb(d(BR2;D%M?QoMV_LHR$%iD+>=^X@COn>L{K@iGjYo%
zfnrzrhRYg5HL)<!P@W=cg<&0_f=s^V;wp-lA2Jd&N&L19Synh59wA)|6BieWLmdAr
z4&_b8D-@NwCy5>*z7Kt1gN!*9tRUtG{?z1gn@`FbNP!W?Zymze_l}==6!_LIng~+G
zh&cI*@T)x;ubvW9;)kN14OX9}c?SF%t^MDYXy5u<j5Rfo(Ev9!1av{s9%!xxOi;n8
zWyq7bkYS#b#E_nb3v3ezZnrdBJ+)AOEHnoE6RA!ch$tu_E*`Iti7-8~2<$UGlXr9^
z11dxma$Hcq-%`l{3X`s?rT>Z%Tx1}2U)5#4k8>F}mj2My6-+#^8mtM8)7{yTv6BDW
zhmUPJ;(=jF>dD-_B~zd7m@x}6YAog4le=JRTUYkP?RB}%KhM9HH-PWMUj9i;G?R4J
z0tL<~q&IO&remnO8Y`Up4=A?7^!nj)^=RtTi_|}zp_P)rkF(d)ONlpM*Bft7WrVxM
za@V=v)A8R`gsSP0!Cdb<b}M^BhmeOz_11HbWJ{!%L>^X1$TDuG<+-K=7L^`0fu0W=
zb{)gc$~XqN=603CKU_!LC%8H1F9P49yg_jzx?BEV-mTK)T-8PS>mak~he-3)7X0_w
zgMyizOGus_!}5LofS{WK#Ac?g7U*f9>YIXM{<H{(lD`3%Bv!TFuVqxx*9E&LzIE41
zc~}|2^9}EgEM1&gAi}R7S4oxp@_h`Q`e#T*|8&tjNJUY<YsTJ@2GGsvSlh;mjr=e&
zm;2lB{<;n+f+g8@K_UP8MW(G?XahJ;{HGYK-^Kbt6j2>Nw<c!?=~rx|8&46P6_>U^
z5Yiemb@OyoOfbOHr=_>hSeAMM-1&UHEd5KZSz0}7dl)$ynYvi_W$Nk3-I~4<*x~Q<
zgloi8gPq{Uh>MVSYxHQ<=Ps&x@i$#gO+jrvI&4+yNiXWeEq!&pWRK)bwH<hwP=`o~
zyBTkV)1iwKpFS8b#21PZ9vWPiJXo-HLLMPjFPPqcu(~pnG49A3eOV$Am5m+A6>d0u
zW5P~u6YbVVIpr@VmpMlfr_=UqM5CjyyECGoOI;bT(_hhMrX;S^gwXMjpTK+fpxlOm
zKo#S1W`IQ~^SNH&8!oC=k<<}P-$--<dAE9rUC<5ew9!PVcdPb6IV}PbUxieJ0M!)N
z+y;Ql!e+DkaL?D6=*1M|H3~*Nei&tXV~*D2g)4P(aV%zXwH9iMO(2@Ic1{SH%d9yM
z$GY)H{7Q2^FH6Pb7?V&r?p1zf?*fjoj(`UB+i`WwQ4rb6dzZ1L<>}}O(r`9W9iG$a
zFXQi1!L)V6?c3CDal*ECyrvi-6#?bG%LJCAgsFc^;36(mYI3ZH_6ZvIq_YwcJ+T)p
zZ{#Y@VfH}EQ!Hn^(W7u}ZI@#>w^Ew5AFi13Lp~WGofSwNo&ymNP$}?Y#3Zpj2N>3{
zw-;<_)FA;sX{DDVfT~~rD+h$Q-mi^2@CNdFzFF*$Rx%K0f(gTv-<b!M1B6lpCA^rV
z_&K3hn7W{1Au=LGiQiV$?L~^?=#bfTSfs9tOK9GscCtPPjb+{!D_+&oIRlC=1n12t
zf8$#8_A4L(QZ?kXe1VQY%%O%8;63dP@+$<uo);n>ZW+AZva$JK(tVlf%&Bi9aXqvG
zJbe;t{^|Z^Duael2QJqV5IJjg8k6o%(VROhO8IS`I7Naet6ePfG2=!M7f-uCbLIOb
z4#xPl#16U-8_aV}$8L@N=ONwl=$00d?xo!9*Izk$>-Xj_(3>23`PV+a4POhV+ZGf1
zAEA$3NcRoQ^EUp=Vh(RF9py3$$X4Krwi95KFz!E|<PNY)&uvBqvkoT@QxDY{t3c-5
z1q!SBi&y?mMBCC|6b!S<&*YJ#>MMP|4Q3-~lLrl89xxbpI-*TG#MuRjuw?7G_ud_2
zG~eCt+$C8M8!|)LfVURNw5!7gZc30=DPbS9UyFUI=-A!;ik{oTE_fv$V|-5s`K_@Z
zN8sA9Fx!3Yni{UPTxWK-7L=$(bJ2o4LJN{XQ`M&n@ca!}gW*uXxPNMmymfSblyCsU
z{vzQ?_TkVrm>&doB3}OgyHa!hOUd}(d_EUD>wkMGafPk-FYfB@%O9Me^cd<O1q{H=
zQtRTrDRur_oH09s&@4HTTO_GfLf>J&_%9zDy#y+nVkcr_+V$R%82Zw<52OEj!4PkL
z_iDrd9fM(;F&hk8iWna}Mvbsoa3WH%GFS|KbDDF{6u&BRwm0BsX4&KB``I1eS<E7)
z`sU~R`+eEt_33JP|NQ+6p{}mi>uG3z>%t=@U{-+k`+YwSb*L1LOm$dry_VJl)I|u7
zfnl4}76{jUo+u$oo_N^n<NW=pO+Z;}oh$@W<UEIaYRWvAm?U**pVd$>S5#0?Nr<(d
zmc4#sUxgo`bg?PHN`>~$El=Z(o-QWf2n)XCqjh^GEC0yGbGWJ7Dq1I^t){xhGNw~n
z2KP0!tpmI^$1J~8PKP$;8U6$gfcG=aXtVaa_<#Y%bTD!s`rh2EQ}k_A8DPXd6=45B
zZG%|G)|i=r?_NSM{N@=UBGy9tqNVglVr}NcFM82DvOcc~Ytm0A6mQkH-!&I;7L{4=
zhK|EI9gQXgRG<mBq1r57MJ!CAWbxYL+Q)3~{Fgsh_@Qk|IXt1=mTJSq!V26eeC$=`
z{-|HeJ(mqq2n!OUbg21sA!Tvs_2-?0^qCdG{=XBlmG4R08dQ8TQR-`vQh6pS{?>}r
z8cpnzMv*6&f;P9TRxzus(I`7;I8}TxQebDwHNK7AT&Mx=^^HnLQssO2elP9x5Ohzo
zPmm|E-@xJ^Ov7e~%$hk;qP(GcvZ~X>sx~!ZgwPj(uR%kgs*+N0%ArWHlnHXPuk+G~
zKMnAlNk}5~87k9Ej6w0-UW-z#VC`a1?5Xh#+)@n0AdNdfm(^0<sg6&K-r827#wzK_
zq642m*4A`D$?q}LVX;|~zZXKMlEi0>H;$EVe<8E5pFKyE5)XP2U|)vjb1qyqXQQfZ
z2@{m}jBqa}Wq*c|CjC`e;^MvG^^TVCz8{oDu00ZzIi!r*pCtviOz*ht%cO8spA#`s
zv1l4|4ccL6%g!**O8-YeUe>Yzb|n^eKU?h$T{_+7t<t#G;CJjSJ{rMwyU^CGQWPD(
zd^Ha3|CGsH@wav5uvE!)A_}jUj!r~btu5=W=&Za=mUyM8ri<yf*f(ul=AiE^HoD><
z)a=w;lj}u>^pJiiS4>pS=QEA~ojPwUcz!mw=aa60Zi&bS!Sd5_8w*G=?|K70KmFAg
z-)qqzbtA86ydx^CY4c(kv8RpnOvH<C=aa|7Z#$X8lh{xa@eb%!;Q5Llhr(BoLUsZF
z_s4d*Iqv*@0#F$1(b4l8JT{O*fgfW<(3X1DZ|Y_`AyEZJY$?hhpK-1o4V>>2C>$-~
z^-u*hLjtrw1f&U*Ih?O*VQ?3#Ajrf9_0UMeK7zO!vE$6Fa^mA3;gC0gfkhXj7O7q_
z%6Ax+AV4J;GhZnZimmyq^VA5M4>i%@B^*|o)K>(Ag#Fqn8SwO^L0>@Ew-=wpYw1#4
zr=M0I7>gZ$zb}!X60-%#UeJ(kT`hLq?Q_H+AQ(K4GB-}`aUU>*93&VfzSqdnzTuj9
zP5#KmE~8$^P1XqS;s!}@rfL$V85GcNNxA`CGwi2Z8o8)`OB0Tql@G6^(rlf8<b2)m
z&wmJN+Y6LFGAUJqMWKBCRHaX`nC*PIf9xD&=P{G}#4=^Af9zuJhkf0w__ngi-JNTp
z|5`{d`#y^OU~>2{o$1$=!hI9+Y_;9GUaZnu?ux4{z*<C*-H8>kMm(3YAfZWoFUSMW
z-<xCFM9};MIBe$8cp`10HN*AKG~wQYCVP9is&Ym|1i9p`WMTM`XKqlWxR2Ry;I;D&
zA}q`%<qQRXxn(3gJ&%E})Wog$3!zB0gFitAFf}XkEZ(2!*@TYBtrJtkN}_#id{}2*
z-F+d@D_#`9eP5T2+}tA38zNo1gg9Qu%V#9$-AM}hodn8Dn=NgoD$1K!pj`4MXZEQi
zz1na{%ofz<(HL1As&Yab(mvnR2%Q>F3yc-H|HLK0v<5b_c5=8290DsG`y6@CntiwX
z4G{M&j#sqK)h>?9mL^SEw$AGK4oD+*8>X0d>=?~W!MV|asK5^+0Xh=T1CLr5-_=%L
zEwhP;V{61&k!rU!){iX9Xc#dYTKNCsrbI?^ihDKsY7M4ew5trdb>2htDf_Crp{c)d
zkK9*ab;wQm?MU7!tFNx??P{ue4l}kWq$Y>MIg`bu@eZT8`-y6}py)BJb{kVzGHNV~
z9q2DTt6s<qC(`ecbCWJ_YXqVU0yGsQ36TX=drvRd5Tk+>lkimyF7Y+GOzgVYBG_xr
z*yt~+f<RUG&WCM}wUf4N)4Qb2)H%m&r+9KjZJOjQE*-B2=Yn)thf|xu6X~*lPdIV$
zD3iwPRsV5Zwy}k@SUdfdW7kV-8`O1C<G%d28E;t);l3lRN-81-h4{C~`V#q*3FsYI
z>^emm?GC>8fEyce>};G%wN!Q@ayx&jyOU4uDVFzoeOoQxlG_S6WC~N#23e*CePjfr
z2@4}DqVIl?TcX0gQ)2e8@u8)>cqMwM%$bwnZ`aNyA4A+lM^a;ixec6(>zw$L6|%|E
zZ<)ZWaaZU`9;}X&<^6qu<B{J4|Ec)@r-cU#*MBYN8PU>mIcP`nf2rFg$!lTclw!=0
zN~5^5C2vug@KjBU$nq;B8#fmtwo7h2u)PXUI3R>pOg*HLN}B=`D(qSqTirs8uO-3#
za=kb5zrQB2PzrAhC13DGYam7D70QgU?2YXm(ary6{<-K%0JGSpPBuy%2eZr+3wWP;
z+d09=C}*K|lwCNJCFRTyc=u)c@OFAT+pqq*eRKF1N%2-q?h%algX9v-80LkGL;{%<
zc}tt^5Gd9(z6x`C%i24)s<V9RgnaTRPvvcYG=9VQ_hlQPic$wLe2+GNr+r4|3C~0u
z-A*aZU8}_z$3c~4posEg&R+4vtf}fePW!wGOc*xR7#W&2F%%SYz~O;=r$T*MFkykV
zS6d?{;JhdnmgC9s_qhtOiJ8GKimxqs{UK2JHCA!s00S+)HE0jYmcLaP!UZc7ma9{o
z3|l_y8#l0S0Pcrm+GN9sP%Iue@m(yDU}c|lUq7D|lP;5jJ>)3M?68m7X3)Tf$B?m^
z&Q7(6HRN1k`e85!=S+@0R3_#PsIo@iKj1Au9)Qhr3oDBB5$Oxe?~zICF5UK)(a+oF
z3xM6#wuX7}My#ZR>NPs%oNGaDjv@LxbOx9bZC-&U{EJwNoo%<J0)Nv`w{B;UAFf8x
zF698PsZ;5OANHjfN%9atlD%AbGY>^UQb%3vfZx5LDk7xHjT9L<fOgWi$DfxBY|H{w
z?GJG*GZiEnv=j;aH7Sezh-Z5xtzOwo()xU&pXKS>prpuvtz3hv4c=Q%l{B&;nTYy~
z-j=-OWQL+i;kahILkk^-9m7nT_MuKS8g_u)Ln8)$NiP8X88t5Nszv<ZMst^pl4G@G
zH%t=&+bdz-`6f7arGi{h{f}SA+y$`*d|XdyczqV2#R2*?z3$3Wi`7m~*RF-Jf+N*}
zB2r+M@UnF1JRn!N9^(&$>B>Ur0FcGHA&a<e<QuGCC>zSa-R_oQg*!fZxL%HR<w|<7
zS;HoMDj*T?Rz_L3Y7Nsub(#W`S~`for%$(0MgMJwL=5d(XtCIC^`LH-I>RVCN|5^p
z!4ta_OPa_ABKLeKQCur}F*VGG6&H7(MEN#$A-}AB`%uDj0@aAoc)WUw^nIROBba-o
zsvF`1Um8zQRFS=yO<=kJIvAOGW9caVi5s<OOpQSUDLgrJ&2OsP&G0O*-z)0>F!oN}
zkv3c#ZEPDA+qP|^W20l+wr$%T+eyc^)v@iJ?_iIy$Jo!o`wOa0?z*p4Yh80{x#rpk
z&}o56l8P6dkp?kRP@)EAp2YP<AsH{K^Tk^YJkGUZ_hZ52oZ$nen0kM?_+GNu#Zmz~
z1I3})r>_-dpKpP<x|xJ1s6XT)irn;##EwAG5jL-7t93r}>a^1~O>}aFK{2ihPqR?X
zl{Q*#&{xJ-x&{yi2m~tcpXJ7df*6uN?as;6m3f{{LDKNS#n?P9NH(du92*X-sKxbC
zC5I6ozL_HacEd+V!7e4;bi#N+9>$iYL`?JMZ*`_BP|P7D?-$leMr){rSdQl*qn9uk
zx1CFqVxBJ&iUs(y40&+#_eCogk`z8W?<*ImyPhUjru4XGy1KyZ-O{1)84!bJHi&6b
zkjj%cO{0iXQs$V48~-x2DK(uA-WcRUA%=f$iwyPD?4LPg)l;=BId(Sfqy#e05-~~$
z&P?Rlb@=Ij0L_)lY>)s8nS(jxG^P#y>E1`hi#(O|#Uym5m_9)=e?mv=u#mka37vG8
z9Mk~WzmLz!<d@Ez-KD%tC}K<`B9@eVm*&za(ETbCVH<TO;u+8!vCd`el<Q^Ndq&$$
zNSJOc16#<I?F2V30x2G4t>boBWr2p7WiH068Kp9XK2#-pk^!5I5<8@$*5O;?(0F=0
zbC3?j9x{Vz;3~p$sL4GZXAT5E8R!@4GQFnQf^Zz2jj$u9D%l0$gZN~W0w<BI-1_+%
z-sY$}mK7!R+>g}EV&OshV}G?>R-N8WyJ1)rs94TL{2feyB%&z}1DS)KfTW)&6CL&@
z9tQg+CGG_w`w%LPi8TfGU0E}rXe)WO2YIxJ#D}rv2fdLMrTb)>Bj44+)HWG5s`ho^
z!-@s{GI$dQ`gnpF7Eqn-;TDF7P6&hF3k;oJCRKalcY=_->uV%ew;!e|LyxI6OGMon
zPiaiHfK$qLGyGxFBU>Sx@;|0<d)Iif=0VUfC{1ZqPwtbMBLCp2zQ<<Vb7#h6lgW7Q
z&{nDVkd$W4Q+*eq+?o_(z&7<0J-f)<96GSv%<#vsCS?cT-VTCi%|LO@kQ%^3B?u<u
zjqj?DrlKs>IDXv{`Kih!r0IzR%B1R8awA*esuLTO9I-qkN4iLPg+vJhDVT%&w8j>c
z@UM9m<1vnUGD^9Le7si6DyLc;+^}jtgBH&be04Xg{-|gJ`@sGg$~V66{*^Xi04r!G
z%Cgex!@u$+6n2jISm*83sg627*|f#7ilD{|7_t^{vwnU6gX0$cf|Drr>WVgS7@8nN
zyy55X^0b)qkS(hM_jV%EZ&TPZT?7KDvh1$K$61*Z{)XSuXZ2lPnd?zV6X>>I8U8rD
zDz2l)9uVCb2I^b*-M}@sKs)ygJdd%jc)m!jmE-ME5nro03S|#{rv&j0)|eHTA@h6>
z#VWP^36b2pJ5PBj4ed0Aq=yrXNz5w5Ji*QKdA8`4&B2M+q$SIf*RyC=OGvjdmm=~J
zMv(-_3qc{3WWdqbY4KIy9rqf=42rlAkfPl&_%{zVHlqA}>zrx;nDKK~b=LGxZ$Wca
z4aBD5doD?fmzu{<-|lV%o4EWZeNcvfehzluWvS|8N3W747+Xd~r#cYLXp2K{g}LIs
z_C9b+)H~(ZfWq7Jk&f3n$JNcU2L0(3<h#Dp8*Ph-gUSd`4aLMgYRBM84nToHUC^Tf
zdnl^#mU6cL`%<d)l)gNk3i%b2y*y<0{_l+glhW=&#`>B{5A{%J_Ul&=;2PS|>G2~{
zaFlY)6-0N*<!FEN%kfSC{^GbiB-V{b(Vw_>YG;#a<+T~_1hz?yNhjyCH=gi?O#=2w
z^fzI$&^eJ_GFL$F7w{ezQjKkvk#35eE#i#Al-w?|iY#y->ZPW<#1rS1S6cjb@an~4
z*h<5mJ~6wGOv5QtL(nhOSi8d19dP_d=|q7ZH9&18Ngsh~P=}L+{^WW}Z5xf2#ICmv
zI*$Pm*!%1{6}g{4>s0pgTV0)n0XMXS!f&N7Q09)}o&QQ||8<PX!Sp{9GPZt_+7GAw
z|B_nKq6Q;Epu~pBhz*_sw+o9ro3F#2piKNwXqdFJ{F40J#?SR7O8?(j#CDrEBv37o
zC_9A<;>9elMIOEvTW|Y^#JSZdDZou1?hG%o`#d7lV+lv@k2*je3_I3D5YoJkLGRO>
z8{i-XIVv);y1Rckb$be}JgxU_f49Ie=ypI5r}tJkp{Oh@26c?>>s!47P>k|V<UCaN
z{A|ed;_&gjpVg%}2--ow0?6?FgBGF;3Xnko>X7oyKP8T@=fBecAPYQEA59@l6Tr+X
zF-LVnIio<;pkkvygN=Ekh!RnY)19-@L{0qLvPV<oc8?v@rly)+&Qc%&+@+|OU<Nb~
z%;GXJqTeTE=+4oqCg~IoK!XF{DhwfYhyV}OCvYtT2lCy~0N4q&3Ej+X*e?v~#PYp_
zXYw#U<u3ZgFK|izQX)u#5e4r<p&gU>8YKq@(40s_0jVnw7{8E!6Wdez@1Br_$!Q~=
zS<o_+Q9I2(>dM;>47SA#2q2@Fk09!*)1aL(0LaHI|7hScbcGT9)>jF8US>=1!PjPc
zF6M~68!o6_@c7!1_@umU0%oMC()L;CD0q~k^Sj?I(gc!W@0u@k+wj(-@Uf>CnftZ4
zVq8PrlovxyP>X@wKC(@2XLVj!eMh?RXPp&aT4PJ*V5aAU_3I)^>L83`#GIlz%j;>o
z;*%*wq%P4Y_^!n>ppq;!AXN0_hA9kn6qXGDjRUXBW&-+xi9mSekAi@scEq-9mJOks
zno6+fPZ~Oi7YgIXNX}ROOW4fyg^&tTsetznx=~}%$+@5@fHWy!g8Qf7jCL!(q%G}H
zPzuFfk+7*Mr6h9pcLQ;vSlSW%&nh_5Cf+GrMYg7LRn%Rj`ajwOGCuUsY!BX4q#A&s
zE`^*HMsw|NU<sz-u_vAK;E-^og|{fN+%sLtIiQDC>(pYGqNxPEnwF|FatqbkNC>|(
z+8e_Jd5ovsYI^Oad}mmW8}}ipzr?}o)N(<OpkDa0e;&HW=fyPTnHVhAtAEsqL7LdM
zzg+tXh%3N>7o?PTi)sI`U3LAzO*&$hX;zx-KU$XIf22Vcz7X=;6V$)Hqmioqwe~{g
z>b0a!kU$vtMsMylw;2wh%$0@@smO25rX9&nD+~ofOnd@u09ouOVCnP~mBYI?y}qn+
z+&U__i#kT~Y;Y+f#OYqL1o#bF$~L`tE!8ub%r5n}y4`>C(f88}RrsGo>Yhoi9?it9
zmx|o3j@K&jm{}3MLs88Zmyjq?#BQmJ;z}zlwUH3W$QIuSa*TkRaGdbT&%dO9jm+l$
zNKaJq51BEKWs0?DGTi)v<}SC7tg?^Ahe16^W3072NM?L@QR7HW(yc~mV_7b>BI98W
zkT2C4B8@&^&%kjC?xCwOq=rm#Op!~loCZ_QWQf5Vw{mJJMQa-{Ba>s60+xJKZmz0b
zmh4bZDy^Oum^X?gnBBe+qllGTJ|PJ;=@Z$PnAa*vQXnQE5*@gUY+(kvq^6;`?tk`{
zhr|^5`wj{z<hDXOMqWkRU;*2M#kCux=a6+EsDKw3#)%~0mhs}&Aqy5<zgW9qx|InY
zpRpL>NWF3-Xe^6@u`Y{PTHo_WON&9PRWCSmb~<RV$y)20r8KveJT)G}r3hsLO6A+^
zmwKv9;Jm++P05^^I2byuL;ypyf);L5`gE*IU+xARLa?0QbA@I3NgI<10|c~EDEmXK
zOJXdflUh0T7$8P8ni0EmVkWjmmnfKLExvzht7h8mNF)zxcfWKHKdXE&$VVIt!u-vE
z7E<qggm5mbNpqLBf3KrT_~kM<u7ER*zMNH)1h1~SO(paJ5GKTYV6$g*AJMHmW?b3T
z&1WPv3uvPJ<e;ULAPG<C<~7<XL1#hKwdE*nrtpDM!`)s0!+F%!?)MjMc$al3Ya}YK
zS8$5FY5M)v{MCwhMY~0hUR_t|6@K@jME416?J~WQ3OKU}r?H{3Z0AH1y>^>}L^9v_
zwOT~75mM`zO-tjFlUmhbBfp@vc2#48cXb5gqZ^HN$O}yW*vlB#*3WxD!4hX-ync0P
z5gh>C#ISa>`m}u>HnEJqf?7+0ih{B+%VDw2y;Fk$_QEvRI-9$rYGb!U)n+F`1T2EZ
zM{4|ve5-k#Q?4wv+!)|<zSH^Tl!B&kD8p0iE^ES*@K#5vcaQN9eiHhn;ES?VS@{4O
zvxnH+@@5YD{-Jy9VS7EjlwhM=xpskctjXLO=Q_pzg<HS1g<By0hKtXU{9klUw*LY}
zv9SNorteD)ZKnfPv>$oqY!PToWz-R@X~vJZy%P>^mECTOzbzsO$6Ly#jEr%@?pRVn
z-nvygvlx_iQ05STmvdfLM8d>O5#5E|g^!4!perSe<2mqy5Kh2Xs+Gv7`s*%Dk2f6p
z>);wufeTh!hu35gmN(uoy~oGvW8vk{n-e9HD-c0q+<5wBYi;ag|L|jL3|v0@^JQ!7
z2mQ3^ulX4r+M^ixFzP7J&*su|UYT^e>)IzoN0Kcmg?%jgH@<^D!vs_=H5cT`KfDk8
z+FN<Vgx$TjD&Z2z2|p32@OX@T1fZ6F<1>`5NH63MwI?B!yj|cwJE+pXl}aiuuA>Oe
zN>WZ=)Rl>DilGoj631kYoT7xW;-Kj=u+0$;E9Fy>>HPI&?Ip-}g6Ol<d_bSOD7K~q
zy*P1;u7%D}NlJG|QY8)?TfU4yCB%jY({b6d^^e#8w7EMhOx;#9Q2id7-uwl9DK*ig
z;qy$${=Gg92Xxjd!feu${<(;jQ%$g5+|M9#iH9QBtM5;0${}AG7ReJuJ<+A&G(mu%
z1mX4ZRTb0f1d}fWzRBSLMxqeMEqh~H^V9<kK;3EpD$LsBK{<Vg|E$$ZZ0i#@D3^?D
zU}yy4^2>_vaGutk@CFFHd!>DoBG2c~l%)^q%G^;6neW`eW&MvOsX+Y)QS6f9yE$2Y
zlcb{OcHgZ5KNV?_%^yKYRem$58(A{EuipG8mQmS<C5`hN5+Ug~SG%S~y)`gO7$4Z*
zR3uc@N7LU1e3pL#Mvsk~!PQeH8~u8hzGcZiuXFSjdp(=4zKhOSxGM2WYvtbVOk*iu
zCrK*PUe(I{4`&M`DiAQSjPck^%~XQ<J?O7Ia8I|M{7D|NXc8pA2;)16u+8(YY;+Jj
zc8A<OqUO;ABC(@w%Sj@FEsCo;_AD3}kAL&>_4z$Z`52!{sgb>I!UV8)3%|`wRJMS`
zq#HHNwqnVp6A=yim1AKPwQwP%1OWDuBT@xu^MN->q6L01$XaQLq3g&`A=paAGX+g$
zNv1eUeUU698}MQ6##qMEItBoL*t!RiV;4j3hnj2k(c-_?jEPP(`>q+XzEZy^3CGKz
z(rwAzPn5_wwQ|XNEmzu|2#hg;+n0VzYkTrXAlkP?u-pA1wa!ZT2xU5InPO5|VfqJJ
z-dGM27hiqb-Y2jr<!?P;#C9kab8DLc*J=jEnW=-BUftqD5PZ|0tHzQ<m+G=PP35hs
zv^)~GjnwpuMFinl0Ee2)IwEAED%sT-=4@V%Dxe1J{oEb;3RJ^LaA!+vIOh>S2y7;P
z-*N>*z|Z|&0?#1Ehfvx^uU4-Tvm!o*lhV_ujzL*v-86^lj+@=rh46g)y4K+O3RFfU
zx*}!k!6e$(COEesO0h`K&e$vqpJ_tF)?vEjqJxX=5)H>L?y6J}V8OqPv>q(eG-&$k
zRIabYDVt=fW8fu#Zif;WYXj?2BwVCC>&-YSk{-h%D3;q2$(T3vmb{`bXDQz4YE_FR
z!ttZ6rK_Gruz`*|BB!eyx8<*TzGgz{u#ea?RU^X_bIGprGeMWE2hLgl1g|V#Srud4
zudNZ`vWWKNcM!L1s&55XHhlPjoLZTd<CE+zDZOXykUBTp#142&QUUr_`=cfH%kLnF
zV_e?n(azq$eq?0Hx^JjdhReZU4!XlfIaKRQ2n>3<GZLF>s+6r838JRhHTaz!{5+$p
zpD7N>PgN}*9P0-Wv(G|d7ZQ*`^Sp5w6FAfPWjo>$yik26ukH*j&pEpp3{BL*c$g0y
zb9x)S9Ky(NsOa1N6aO)%u(AKgI+>k|`F~v}Fa3Y#6s`Z8Qx=@bbJla6*Uc=ePJ@yl
z(*sH6spiP2JT~RuH#&hZnj~LavWppr!;RzYdUgGg;LU*HW1o8nuJHg&nGjS`B>3P5
z7C*b>P~?IFxRIn?>pm&TP!jC9{;Zu5kyN<t8~}bEAUC(i$B*;>MizlyukWWc;}u^P
zlW6#Dg4`}HTM0kTk5lIthmV7PB}#U5De3a?euOFsW)}0fLeu18rZnagfhtY=joC4k
zUwE<4Uz^7HsSvbCfr=v__!6R|ynivgO7%ri{!suU8l!-+rXZAh_3|%`FH4Z;=at=%
zglI3Ga@`^uS;Pa`CouL2IU7Ntz)B?#A#_Ps6@u#9vqw_m(W=+!a3mmyqm96gXP3$&
z{^sV~H*6~i4N0`eetQ7`;Bq8uQfpIEJh?ByFMY$_wxhdpkq<=cZ*0Tp|GiH0DOZ(N
zKUB}V<fP@zwVKu9Hov>4oO^SLD6BMMIxB3RZSQJ7oDp$zZahRvkV>gst19W8Rg!es
zsknVvEy`H<3qTjHopLpZ&-X0taMhoAx44VzD2ofTl;NhDW~|^GnODxbakCkrP&1$3
zp5##kNp}Wf9EoDO;Sh4}IYgc9AgTkRIZ04NT^?#&Ah1^?-mO273)Xz*I?5kNXENEl
z%ksC6bb9(ISs9WqpE;|lERVMXWv_R-WBDf|j?ONaGKSu+3LEPhNNYCl&kO2KYj2ls
zO4#YtR%osT@}0!Hg?=OZz>#YlM`z;&IyR#klH>``BtKdHYBtr{R;knV>2vqVPh?k6
zpCV45A0JQaj_x6d>~Rn`m>}DRJk4TCp8)K(kU|t8A;b}HjpzJR`sn4h@@6E!7wOz(
zSI;N*>N_cC25jdr%VhEcS4Z+MWzlu8w|NdxMwogI%CEqTeB4uhopUT&7Ql$)n_s@Z
z$Auco1RD;+UKn_2E-6rM(RK2($bYD(zeTU*pMH;+d5K@FW>=`O_r5)_Pe5+KWxO&|
z>gQ$ZlcKZ7ay;khcuAPg@o@_rpLe_EFoFD{T%?<pBE7jX;SFPfL|o_>#TL&>t(?S*
zbaf*^%|Tyfau}?)STjHPwQL!X#IS@OS?=W|{aej8k?UWKrS6dk`)u>M6O7rFVXuRd
zzm|M0nPy<RptFC6B0ARW35==B6ax#V8Xc|`Z<qD~nb=_Q>tyzCBQh4@_FjVskZEHS
za-ew-)h@Ow<=CqHy+F&XW3h}+;9wNihXXi%kiDR>7=41;))w2vk@;X!!+EuKoh^HU
zW<z2~C=-Ay=8||)NWJ+rxbrCJcR)Z9#0Jk#k5>#m)Z$l=)L))U6cl-xzJck=WrRk+
zMUYLP0qa7su4Lij3mJ4&yEf!u^&`B})cEi)jPW>O3?`R-Q`^97Y7P=h87rIg);k8|
z$7c|>s+DeqCJ&Hdtq<{{W=h_91jz)13MEB^Z1%ABABOC$^Ddo&e60;a>!fMmzC@%-
z`9%-|42Ry<^(P24Z~KZRf@!CN)y2h8a{Yt+&nD)W4khuj%GxeY+@BqzoCu|OVAx7W
zJ&M1eg3}3YT7Jnf*AR4M-c4@ZgwZ8x$762{=hL}mL)@@YuDR2sv`$Hx*%hR;N2<vD
zTX+ZB`1hTZwRD6~en|Wnd{Hrw9<%vkr$v!Xdt|+}*a}mKYxf@vK^a2@W31LV8XWI^
zEJnh*MUMgp%Tw?bOap_?Z8pO1MzsQf)ZW%U5)WXKw{LG5U_eBogyjXp`AJmot0t?X
zQpMeRBKWS;SvD@Kh33Jx!a30nK#<1`sd3UdsO%v6!~^#q4ZPtz%XD%AUcCVxCy3aq
zX_PL29p}rGrn4GUv#Y4tBSDntz?RQNA&NIH9Y;K{ufP5oWf>CwtjC*0Xin;$Y{JO`
zZ5DA}DUmNQ<v1qKEl0sGSKT|-$oVqv-3RgI7s4haD%RKDa;QJVwxT?lQfB0+@t?3q
z+!wBn3t?oT7#-)nYn|fzqB3KH^|Mb7FE)-@Yfs#vsMPmYvvONc$ZSwsY@KA_9J?9l
z{B|A6YNP8{Ts=369lm=ugh_7G!f6M(QPNx;0G;w=%Q{^rL+m;uvtFQYyZHwf5Q>P#
z%5ZO{SBvuu^0sFUTxdYT&UZ;}Ey0J;?4|ZY{}EHt>&X7N@VgAoW^8$!i7f`pe{cFe
zcdldbw}+;vILtd%e`@I^`+z@MuDQkh{@m(idbtw_$&=~YMqT|TfNI+n{EwF%JInw7
zWyj3K^*^)zFMs-CKa`pO>x*UZ1GcnSffETcthk*lE)SX8eOJ#O`r9GHgGpOR3+@JM
zGcCTK=<8)7ijO#ZI!m8tF@ZH`F`|Zzo?DZztpPi)Zx>ZZYez0voE2#>^w~%2(7~u7
z1NVb}HUCZL;NIRpPuMZkW#q-SX`FcxBNGlXe?Db#`040vWiz-Rc5d?Miw?M@+3o1k
z+1A-^>uv21bqai3opoxHTuqjJ`Y^<JpmE)|tZ^#HZF!$m{(RET43_md&m=U+jQJ!q
z_I)Cu^PjWHSx{tPzkS9LzFY|TqU^`DJbu-Qh8S#_9I#0W5MzEudpK7!A}l6xR9Ee#
zvHOr4pXZ&`#f}P-*KUH>-^??R=(iMxZO_3V+JOK+BYi@hE`#Q+v>7sgn2Tm25JTWp
zXpM00*5T63v!5I7jVr$&v0+|yqa}{&qm1>F8^c`k)U!vNGRA6p&@a9xzmO?gZG+>q
zRn<@9rN9i!ymjDeRxi!4ZXq1&^Psdr1+@GoXz*fPJtxMEIUDm*opZB4E1MHs_r!#X
z(Tv9@{O(rd{zrsW<>pF<9ut!(Suoy*_xiLk6|;^4s}owmNX&tUTy|%A7s&Ny8h65V
ze&e_nMM{zdHG*5m1S2!MKJ|^)!%EysyZWP44o3e-25)F*fF*F?#|ZC4VdQY!mHy_K
zf@{;aidoHLxE<N9WH$~iF>er;EUfmldptvUrA(_?yn?n;PV`((XKehHN4BXVh;LZT
zpl$GC$P%uVUeeobyM%oGQ<BZ|xEj=Qqw}Jw?d0J)wqTAH(fne9-9^G1&sELDfA8(9
zu?q%$&&ex2ku|0sP#O?(XQrFjK@(Eb9CxOja?qClx*@;p4>2Ivw$gENScP%Qk_1Aw
z;FF;cQt9VhISrYx{BPDaYD%4LbwypW7U#j7K~B75&_e~I;3}nu;cQBHyGq<P`XxN3
zHoCovk2twy0iF>7&Yc$EqV4tDV4f`@dTxpcXJ1h+1duo+ZH?eIjo>DK?-Hv{P=N=n
z@B+6q@@WbiS4Q4^QB`Ua`6k#eU-3B2V;V&m&>GC#JD>B&yjRJ+z!h%eKxx>Dra-gV
zY|Ycw21GeQH`m9SRXCCa_;v7`MoOB)e&R!K8BY@zqpqR*S4P9F03yv4$Q>`aIJJ|Y
zQH5X*6B@7|<!0CK&Qgs>-GXyi(taQw@aElcx+hWxIF=`QV=rkIWybA1&?c8D!_U7y
zL8#iAwFHOwkUYAJO&<YWi{Yf$yJ<-@(Ou~xgCPE{596UM3v{1LqG71l+JabeYw-#4
zy(bpSck!SYcw87v1{2{zAF_mb%qHGhRUI@v&p!0O#5A1`uSUNAUY9gba8k(TM3YE)
zcHtrz_A3bUZXy9oEuZH99o-(@<L9|RBk7jTq+J$~B@WcBaFIxS@aEqb7dN|k>tl(8
zv*6Lavg+pS3Tb#EM;bxHj3mOUyG4hTXmF40+ZqdPqKgt5?f{=M|M1a8(pO&%%St%|
z@z5H_6nl=m6RKeb$1b!nZKU8DBDwb_cdwaH7D$I=i9)~e?qVz~GU*5nBZ{dqf?&7E
zNIaPBEuh8nQkuM&0x6YRm<2XxGd`g1H&@GZxX&Frc%yv);e4T>n(>9f;ZqvGeh({!
z5^rRK;kAJ9@%`-oodyd99%hcvT9+%dj{3!bNuV~b<4xTBzIA14>6g6Ri*2~Q@{QgG
zmGVY3A_0`d8P&~LiuIjVQ)%un7AoNRa)~sK5{|m+N7+$x)#<^=!)qLg=Oa^r0<_=5
zK8kCkL(db@9aEVCoP`#!0wq%^q6aU*G#JN=fm}SOI|!v#Vl6ES9N&E=lZ6=%twi9g
zr5k{;6mKl|($b#?hKx-@JsXO$PsKZ-aup)=Aj!tL6#>S{k08?xbT@0VHykHyjt~Hv
zxNj0tF`BzZ$%-$a9r{?nMYu_P6m;WnKDn@qkj(5~uqF!rcbC@@3&0Hxox2TB6L=d5
zgq39cc=4|Ao9nOzX_G0sED>9W8kLAYjlu<GQidz9*86+X7R1%Kk!`oRvOVoYD&f7J
zj<$=VM??_=mY7iTF0JPxQkTn7<U+{<C_i6VSg6O%el5pVZk!-n76G4fLOqYtR3q@0
zWC<M6kh^5#?CWP<TC1pVwB&=%C9P2`m4ryA5}vEH#p5PY)va^)l{Kk<fzGKH59rf#
zOea=+#?Vz(wm*w3pe66$HMYOGzb4^k{u7Pp4X<-l<b|`H!XTu>Y=Y)8C^hqLphWnO
z#LxKrw%=2523NB`Da^BElLVZB{<={@+nH00#Hut(qC=z{3~Q43?}o4c8=@_6)CQhE
zVaXjv*e8PaCdicmN+4YEJcb9GKgRQ4!(Qr3H^`U(3o#Wn0c&F=fkFiu-9%3715G?G
zVGX#Oxtncdw>Mj|0<7~loxPacHd;zuPNa*<^PW@LZXDQqvS3K2AwSC3sQ^&%5)%X1
zn>RaAu+NSF=>?vlK5V+;V34kd>YEG3T9<B4e~)r&4gZZuV6TC7$Sn;>7}{0d0Gg?l
zYv;b_u>CtIkxTo-I*njUYqhPX(GCxFEyj`h->-R~6N}!v$9WXBvc_@m-=isc=(S++
zH&lv8v+}yioY46o7}#!IUe`Qa@p7urgps@&rQ@N6i=i3w?cm3NuXdJLT-Y}issID@
zGJKgpB_;_neVlje7fMdjTc}vXAPx7yUp;$`6pN*wm1@|-diTv1PfAs*);zeblWFx`
zgbx-VB5G98FwueJPQ4*e_8);yE=mKAn!UscZQ!7u&vLz<a_=;cMJd%z=4&mwtyaB2
z$p?CRwzA*|#lA<T_xa?(m?6pEE-ui973_H@z~WV2JRP*_Ex#_wP-(wHE*GTWw10X^
z_l%}{MSd_{7X2SDh2NmsL7yT26>|PV3d+p(6Mg<il-?1JuH8Y~Enly&m_8b-*$Y+T
z@47x0FR!g_j40b^kl}59$`%w@b8UX9?M2F~-klqfKa?^p2POgdZ$?UJTrAnw*f<h#
z$gw@WZxat^Pjvm<^O6*-5v7R8GQ|C;!R_It_a`c{34$xA3A-=0o^q-mds90--B~?e
zPJ9I*qu*~gUmI3q2dACS*QMXCPTwy#E7#jQx_-G|pHFn=y0ai8lcWYM@!%|0we@8s
z9@{-jq{LtrMCEI27A<n+qFSd2np>B1WzhXt<<(JZ#5HH$|H@yLX&1Ql|2<B3*F}1?
zR6~>x0s#p3!Zq*>0*mA4oWcJJr*}anaWG!C(bq-9Ha0!gMOnMg#A)mTOixC-B41>*
z{+)WJ_Om*`MnXQzZVd+1v>vVdJ<T+EsH>^(q&q#z52T0#<SiAHZOmac;=rpp6nY7^
zqvUcheRVLeHa=aCF*~2BvutFXK!i1{p(x;gR-$MVrnkv;1biACu;$at%{yC1#NHdp
z;(3vp>K@9DalW~~Sl;QXSiA@W87~2n(xT<Akd^o7jOk29*l!{|Ab-;HZ?}jyXq912
zZOgE-ov+k;G!LK6f{hECnE|ah@&~5_gPCP6XX=xYE-Am7v)GB5X4K5u%)}cDos*}}
ztb57^p<5C{E@pG>izWAe=K)(WxB6`V%^CSRB>Y`h5*!|ziGmo76p3a}<lnY&9WN<q
z<o1KmjUWeexU}iOSNaZB6vmb=hcV;{G1lAr-K9Iz=-BRr!f`JSyj9l3FV1BO65T2B
z`T!=CJorOJlv|tGaBkn^`dA{*!N$e{$yQHalS}yNA|6ga#rd_7JM=gb{PFy&4)%gR
z%LdD*JHReYR_@xqjw9z<SK+nTDaS99c11R8<BnL3w_OuvKWKBLA&z<tpbhQ5?W<A=
zka#YquAdlV{;p^S-#yDUibYnkoIx}mPHB*%-!JRZYg0PIui<jUx0`3yH5CVrv&ewp
zz#`U+K1p9-<gYc<9j7X-=LCbZc2(tW!E%g}kgy8NL8>wG>vGoAhy|yxi94zZ+c(+O
zp#{byCPp9<le)U83$?K2MWoC>WZvADl*dU#z0wWJa2y)UC&EvjnXjY$jA%hPg^Cd+
zTXqzQ9*rQynG|VIQML8w+g)OY;=GE2XWv2(0LensY-d(K>@Hpr-LxzeOa!SACESZy
z5N~=nb-f5rB!MF9Z?Tr__J17Hgjq%@=G;$&h|l4&=~4^8ORpbPE3&2b)jD)x*TnQV
zqNo&OdLM)D+{4l9Cs=$ngGR`o?tNLxfCU#j_2H7$%v%%N%Pv0B8x3U=NuHojk;@)!
zkRh#ivH(rZWDE|mbgU$5VTAf_(zHLw=cDu9m4ciCbIuV~KY5-E6C1$h!Bzv479`CP
z3le4=b@D*5^*|Uj;>#Y$F@jdwT@reoIfBLkVJ1SkBZmsyO+ZeWr{hC65V<T47XL9C
zK{XVDe99J1|FMIx0bSG}gSTN|gwz0+J)s@{YQ!#%0y0!(#fm9KWC|t`*Ur@gfC(;S
zT%ovl@e#3>4hAEOqOK7|!%0gEHRQuv24I<uLOTa43*$q<07~awvBeTuCBJ8L4G|$W
znQiD|;gk`uD_IL|k#mU7mY_&$JmNC((W1+L7w|@x;}hLgNu>uOUK65z-CX`5C}8A~
zd$pYWEm{tuNOFQKPa1yxc$B<xmP6$4oL{<BALq|1PNkALkkJ6ciCG@~n$0<-!q;s$
z=vkDB+8FB6G0L6Oq}4%RTM;>dUR)F_yT%85s6qIPxS*T@UX`B`$rQFq%I(y0w#v%D
z2v-RLNXY;Xcui{iXlKc7p(?NL$py&WkeqmgC|HP@Q(M>PExL%yo7ICWXy{{%URO+(
z*g_`2)Y<J(VW;IJcdida&M01?a^#mCgd#{K7%|;}fk!<4gA(;tVX?S7u_!iULC7#6
ze|qL#x+;^o7asp87CuDE-!`sU&NWsch+-RJ`Zz)}=>1=xwon_vSZ^p$cG<+6;^C-s
z;sBO+A9qxSxvkt!6>#<`1l?Z@i^$uhG+2&+#J|WhvTbaaLFx0x3Wb6=-RaC{iZ9sm
zw}da>rEN}%X^lQ_(qu#WSMN*&jZL%*k41K{;Xa^n54ovomu6|N+?`xnANKGKdBJ^o
zX>fQDC|Ce9{q9X<K#(oV;8ux6>{$2ET05oynt_6sAZ;RnXMDUXI7Cgac<3^#G){ra
zI9;V6Ckd5Kr&y2#qFZ)>-g4dZ^J}U_$92X9ZcU4qM%wG=aw|(n{o{24KBNriAr|Y<
zZI`IuCU5?L8Xs!ImKxzXKP;?JxhlFM5vw#I^?Hedsh&v6)kN-|FJZ{ZnCxt6II^#1
z#+|@nh3Jk_jApD<`;aCV9giuhIn<Iy$k__rg~nE)rxXqDf}6qQHfVlOC<mDH<ptzB
z`PX7ahYdQYpPBYfkn0v8-R{l*>N5B(DVSK!!hcWou~@I&6UWA!DwOqbNq1fSbUNpJ
zJI__U`idy~Wx4Je3})FK4Chy)^XVrTTTZR9j}Y)^PzC+K@e;DJgdubr5Qea8uK!3U
zulDr2E)h(D5PQy53m=MmD^Z0}4ClWB9u03Yx4tHC8@JhW5tqSjyE1-$YliyVfcQZT
zUjx#+{NdRo*aM-cJ(sErRExu@^foQgB|>7u%UA61Y!TKe6{tAHS|rKI<~q5usLfWy
zESk)&Q|xJFfi^q%vi5TAH+JHe38UO8yY-Xx2ZFqhW$&%?pI>B);C4&f_|<bfKZ2O1
z;1m4Eydz`Qr_+OH%qtj6?Z3&gfv`u}D4VJ+W|?Gt_Rs-QWV?q_s%CSbs*RW^YnzKC
zv!aUB@yMj+wknMGbKi~rfuPES>Wt3MGa$r5sOb!dmt<DYFgqGaSJwH^${lLK-=0so
zBk<B7Pn$ClcP?A+gMuGrjY3Z|!Vxufd=~j=S2&lGMcxOuMo0<oN<(|_X8c;j4tZw+
z`U)t~2d4ylXvQ1i*iwdwWa$oz_w<hjGTBY)?q26WU)y|}TNT*tohlsc{0-f<Y^YZq
z%?-Qer{Z~BsPLPHQk1m}1{Gp!0`Fy~=3%~Wucv%1+oxN@IoKIIK7M(2dNO|4#v}j!
z5gjV|s9vB3_LrS+&-u1%T-rlXcAW!CK&WAem&jZX(T9m1b>o?xh`Qfra3}gMkBEr|
zRFA!N$e);(;JDp{7}Pr<tojxL4!VaFz+Atr1(-ZyiwJ%G*y%-FT$(u{tep`*i;Hh=
zm?iYi_Mi4}gST#*4QzUTe^V`4f#p&A>69?kyU8Lia`*tR!MbxiwBBH(bt1Qjlpp@7
zBP~J&HoraFw)tD8i)ETPwEfuQn0$+{^XYbq1^f>}6ra?RL7K@+;jgnFc(KYB`F~_e
zj{oW%u>L@s|Hn*usi_^e(T47ur@xm5lIc1bk47b}`%sWJX1i{-LB3sPt!QuUDXWS(
znn;nM>+I)+=uuw9mP?uh^sd-fPZ^o!B(Otty6-B4xuf@C;c|P?9*&7UhLTIGz9E)P
zTPz8M?%1$SyDh4$)1&+OxaK6@*iV_NlvoU1y$AXE+9~km<nnpuI7Sx3*{(AlEMA4l
zP?uI$r{CS?8$+P?b-=LV=XaUfnQ!PP%{*M9BbyhtF)Wg&=W|tkzBJqk3$ZZ|u06z5
z0Eb^eoJ}6(;;{F2AUg2k+`k`el1+zT*3kX3a;(gi%w)Vq41P=2fJy+R5s;k$L`03Y
z=WEW823<18@WPqus*M*(P*E&{hFu^?{8a+?jQ`k|T91G`hRI#aZ`iW)Y66}g;B&xT
zU_8bdks+9A<)wXsykzC0bA5$3mnY2P!#s6YWI-R>3loU#lf92cwcYKJ__Y)JLf2pF
zeDy*X`)ieLcE54<{%Y2{QnK{@EhM!~&wQizavQG|Z?qngQ3iTuS7D@Wv>fuu{V6*L
z!HXdC1U=hSw14sFXV=3m+~>H-me<WDV<6d@N=HF->EIz3fy?l>jQZnqJOak_xsfBG
z#Qz1o%ozD%n-!4;*UOiq4v7QIOdq0zkIkUV1{36NR}+BuDq?kR<1<F7#Ei>wxAToV
zJQjGt60`qhSmbUrbYGQ$_-?>!w}EsQu{?MTvL^z8f(QpDKz3hxpB&65B+5u^qa*?%
zFpC6gugbZpg6`$hW@2H&9`2Ay&FeP8rt!d`Z4hU1Q{fb-uj483FO7CSre>UGWo*NP
z)pP3vMO&<yiA#PiOD-noJ*Wnu;Iewt?}{8$se^Gl0W(^WpHQEiv;c0*4vuzjE@w~>
z)ERR`MrvOp_So8CKvFn}4n^Fkc0{Fi`Z56sEzyX#fWJ3ve>afAi?)3~fu7@-eF9Bt
zNfc`MbqXs+N_ycIFiW^9)O2?zpIE-FcK+Q~>c~tt4U<yB0>}dT(_fF45c!G?Qk@m#
zftmn*Zi`=l{t|TnT@x1`%#vP2erWw~4B6p>>d?Mc5(L?)%Q=LYrV0YQ^x3NjaTR)w
zwAE_k@cCsbb#<pH@Ttc>#FdZHk_gI8aQ!}uqw~zY^ZermU;F$l9g?CO+V;=#>s$JH
zsf%8tOYH|qiMNehgsNaGd~ErxgO#+?2}L*0*`J0OwtieY$9izbea}QvY+g<@EX}~=
z!q;M{raDVT-199gy{Z~7>Ry7bWx3|<(~br!@7^kZldO$@_&XafKrEI(jjI(_-t0WM
zT51;dicY<)eIEFw*Xsp5u)_B*sFJc?9q8o<#0<&QfK&WLnm;G<1l9B_BV6D`bN4Mf
z-m}B}9KT-FHwGfsQ(Ir?sfI_yIdHV!$5Z8EE^Fc{)dK9w{eGWhozb%<T&gh15VDja
zg(jkn`PgM`tx={S{{t9sQh<R^bAkb(lqpgVL`Os2gvZB!&mu&GiC~dLDDxHOKXk<v
zQz4JJ+daFCV-`X0)sOt%U27ZnM*+74Os(<{VcZCqo%bW5IwjskH1!!ALJubfE|jT6
z_RkL)6Gc!7@=M2-EhEMwDD4?b&E%KQ>{Fe71UqA*m~zia+cwQ~Q}+YvRFlVc3`S>(
zlB2Y4T+8kmdHrX@+G5mTs8}+Zd&4vwMlpUa@>?@kYzZi!4;=d!q7@|te5Ip66eh9u
zD<TOR_LJ+Y?;*t9z1cd^46KLwjVh)=fDNKN<OS&*!<W}4l>;B>J(muOu)*zTeKK2&
zTN_Oxnvw=Y_QB!d#K=y0m59$@Ww;ba2z$AHb9Gw6F-iR$)x7)#-^IPcJ<3X(9X@S*
zrdjGi0N+EnKzo`V+l#iKn`vs9ZQkx!BE&n{C@i+~UGEuFGbPkp*8shG$=AZl?~Na<
zj$$)OS4@PS!MH<V#&H8h)y_?LGsE+Opl@4g7DfklGz*1|rDap-N@D=_P>eXsm1f}B
zhhOMK>IDKM3Ja8P4}~T8fCZEZK|I6MhwA>g3!W=4wWEf=YYnTf1%+J=-s5X^H=7}v
z&Ab<Uea9tNNN^V42&~$4Lzn`Z8B(PpBBZK*<Y@e8qh_i9XGp2w97iTgm;>@1*z1|F
zleY5e0>V%QVAC8`u&Vc~PshW&%OBXb2L1{1;Gdq@r2&|YcB5@|7{pEcQ=f4piU*uz
zE+m}FMxJaG(j(Rm{An59C+jT1$HhOa@%CqSL=xmV;UWlU(~%>CT2O|UGu6fY;pJh#
zD*g)<dOSami2jtjuyC*mMJ{6#Ir)}m$nmq-K~skKG{#nA>lV*JeExY=RuDu#yFtfw
zGsg$Kuf5I-u$Qs$iUR>Z7l}(u1kQ)ODrif0m#)<{+HvP(2aGKveFImrPBwTvukZ$K
zN4s6~zit@Mj=S{GI9fqniVYDSh}TQcJyT<ZZV3D29y%MRO~S;KIVZ33tSTjosXlH&
zS&T(Qd4Xm|H=4vb!rKnuO4y+Amb!ThBp)Z<WT(g(Qp0M{_-7W+Fxqq%P-~5?ll33X
z8y;l=w}0O|O9Tbr$0pK=0l}2fm{f%!WK_|!E*G^B36j7)S0LE?rMdcx3+nyOTq|il
zS!ifRW4%<Cl{SL^dSig6`y6(fka&Z02TY;7?nVYG3S`_H6$8ppNi=>?8qPlmUB0mO
zVj$EmjFB$iO#j^2#%uuKlpZ>6I4?^mn6kV3%J+|65YL_TpTV0#t`Q&v8yv@DDfT{~
zu49hFCIUWZ!?^v5;_^XvwU}o}aGSljDlhE8LChT^RzPnZQ4*z?3Kym52m5=uoe{so
z7=MdI4N?5;G-0oI5|`zTI*2vOI->hQ#gXpHZbQ2}I0!nzSvayPRdWGy%a|YFW=0LD
zCS;;-2Sz6nZE+<YSaFlU4K=JAk&uvCULf{LX>zc4e%rz@P812D9WB22(&fm?nB^qR
zD~m1`j-fE($Jhgdnp~{D-vg#kcZTK^4NK$am`T78p#qP@S9yX`V;XvP=MDLwE^JiY
z{j|i%DG2|ic<GmL^1+lw%^w(dt=0#}LV3USTF^^j7gNgb)p&sD(y@PaOlSn1f>?};
z=_^8u|F~Ndi*?n5=Kpu?Q2t1>je;y#*4Bgn-}-Oalfw}hr)O5wF3pe)bi0=wZ*~q%
z;rX_zFD>?LhSyNUz*;6=Bj@WFI@#YjDQ}5NVe&l#Ig=02<R1bky99*)73BV-mcY#Z
zztbb8aMm0*{@&{92S`x=$hr&=s;vmvO}#k6+-P8E82(tn5(^<)CFTPwwq#59@NGHy
zV_}h!)#bEFlfnHvxjOB{EgrKGP9k9VV(oN)OPr$}lPqnMHk%MF7cGsPRV8l}bL86-
zO_-=qb-r83%M<Xsy+2Bp;r@X{Eb05bmJxj4eGKn>IlP)ZrIj(L%t4Y-6$pcp!%9hS
zItlSznM9rwtBoaF88IKZPN|20YA+qxv`?*EUDE4c4-U-l8PUqX5vqVBm{JvJf$LNu
zI|4<R8H%iN8eCSccFYyJ%TeuMz&>leoo-%N-P<%)pUG+L9J%yRtrxo1ahzLzrMAH)
z$~sXc?KSqu#nS11zGRf?P10}c?w0%>Kp}!MrIo@a5icI|b_%!`mRhw-gySq|ET&I9
z8r~5|lVsqLPWSi7?H8chEcpFz{2asFLP5M-I~lbg1XXxQRm@A)hFA@IJkvp_DPAb`
z`_5WGL50~gdeDUw&L4qr8l;WppN-n+x(8o-ZnuMK&E~<+-ILXVr{ni7zMMzrSe^$l
z=?S6Gx*$eu`0sx(c7qH2ztW9FO^7%DovJ{(Xm`iWepAS<WO;vPN=xTG%npL_j5S&a
z9@t)y)ptCZsARtQjU?es>1f8*6Gia`<$5l&UY}u5z4QG%))l|P)UcUwwr_LwvzJ`v
zChuML)_kV6c|(&dZaAMF2>58&-^~%1#|tuMvF;r6Getaq@+A33PW<<!d65}Ri*tP+
z<P)-hXklKuC0INn`blSon$85X)yNntI0N~lfBVX1R(h|<QlHgetxO$1;x9q3-E<>`
zAV!{M&<@Hu=BU&6XG5u&>JNE4o^a!6MSR{KQtBWH@^D!X=8a$=Y7O>&dv%T)uQ6PE
z)D5|Y$LmKj7Syt*DGg3|rjUy0Vo(2p_C5s?>o{hJJ#qNh=9Izi@oV0QcF+N$m*;$U
zB!dp5201umZCyteA<UCF=3JnSkbu6h76Zvj(H#u=q_Jjl3ydP5-ZQ}krON0)6R}Z=
zOeG`2E9I@CkepI6#rhfA7$S!8(wd1ZA8mE*na0Zj_;n*=bUd{Y0Ge>nIVcPiBuguy
zt6I*=(HlN0`W1S=nSix&i!BPNlQd)-$ZzIQ?o`7^D8oQxF-UVL6nc@!Q3^Sxx|S$@
zU3sB|rZ8&CmH6s2V4Lj_0t!SAxrP3Q(z{KKBb!}?gHl%)%_ZA5$~4LKc?v2@p;Ax{
za6dtkB!vC>?ZWlGF;$4R3hu;#P#V*}+!!KSy7wFhXA4B;a9zN12Tqsc3H80YIcN>C
z&>Y_3POcbbsY7rUWLF*XlZ>kv1X>)C!Eo18zOCA6Vh3%Bd4-CC(-is25sWrZ<|nAa
z$W)NemDxa9g+BXZ<(?qiVxZx_sA;lUgrV7MW^vPB7D2^85LBgEL^*kakW>!fEL%@B
zuA-soErKU#gq9&!lxgKV4B_@H?eAxt2ODI-Qx+^%o45|jg@j_THqV=iQn6>J_EAUi
znmyWz{vn}q;fDXieqkT?XR_H_J^vA>ST{p=*{L7RE7(585Nx6KNBulebIMJUmqmij
zblDO={`x}`jSd7|)Anu`l`=fIB!Df65nwdOOafzpo>}a;TIIOEf+frK<xh^>vRd<f
zFarkBi8bVdU1eviy(IkE$1($F*d)G58j>AzHI2e5nL-2YjS;_b$bg{gY2Y(*r-*-(
zj4h&is7=t;bbr|2gw~L#o*&hDlo6g5l<haS-2{){nIu3x-tjm7jlmgDVO}VJ7Qrc)
zNe~-NBBwBp=ZAs*Mn@{j=`|7W5I83YER}rAHIBIKXSG2sC8q#vP`l~QzlnbGl-EWl
z|4<Jo<KSFq>Z9LWZ>#Bf<HGrZr$AX~;1`lm<?4VG+d#i<U$BTb$iymVge*5RS1O6d
z8a{A>G!5YCxV6oh$)VxTPZl*&EmgFqK9KKW*ic){wCr<V)(g;~DEh!sg==fq<<#*V
zc|`P{p!wIcaozT9)AXkv3~4tLvW6B~dVJyAjV1~=UVfyTslYIgb^$!hJ`LVWJ?`32
z)~A~wC8Hoo2#QN;D8G63-1PMcrR&@CHxvB$cAfr)xX^4n4#E<o>3C<ROXnYYo>;<9
zJ*N9V^@ASxfpTBV*SeRju;%tegh#ay{qBua>y9d{z~0Sv1`Vf1m2^PT;eD@VDVUUC
zSZ$^J5Cw93Li9FUQ3{l<gdMuVr>@%hBxrjO^=4&Jwcn?7SQrDQoA}X~roP65&<Z@^
z5QLs4>>b4O0|g3rTWx{P+JF9H-JTdn*9GgP5a@_vjJu5QWz^YlV(c~rNCJVqhL(Un
z1G+v2I^4NEdZ~|TTF<VQFIJYzK!IdPXsgFJQl`udbY3bguZoCd*<Q`e@o)ha{v^8H
z!v~gW8E}Ek3RiAWNGuIk9r5|anR0rCfk6GyX7QZs!cBh@hJH7C)i6w%#C<ecZ57X8
z-Yjz_bdGlV=(PgVFyltbP-tWe`kXOvWk|^D2^p32b$F9w`@LXE)|hi*r+SkP-i>Io
zt4jtx^KZa$S4D#u3Zz7dy&T?B*X^jgbx#&^Efm;ej0KBhp2DHXn|}bY8AFyc5;S3k
zs>d-;uz5eiI3?{QTS18&eI$XRSGL5B-bqqGR1Sn^=gpX0WwdNne<hriQl0UP?yH;x
zFM=5<e4qG)$)Q$cI5vY7olp>|G0LM|IF1h;c7k3Enq5>mzpi4hhp>x^ge+6!rya*6
zR`0g-5&l(?yU91NFNOeBbT&FNrI9_m3rey!Isqe#67mKZs=|~6zv4s>ETgL4GlB1d
zv6?HE<m2qib@gmGq@VkM46S5&Jv+BpS9j_rms9@EeiCE|u$-0@7J`V=&m(=L+P#vb
zgK^kf80Uh(Rhxb`h682%yTN+j;E5X_!OvVC4`2RZxM63!+5``WzcI>WO{G+UHRvWQ
zj$>e^IT@F<)MXfdH}=x!m>dwv_!M6I)ASd{e*n&7q8aI{T?{IDr8MM^?bcVy=4GD3
z=cf$381>u?WxlYLI1Nvc2=x+2(>S)g#i}vDiLH8|DF+7)K@A*^zYRm80_!A3vjds0
ze13vP7&c4iJoo^Y@cTp4vdeS)Y}@|>)z4$#=0HJd(1Xm_a|1f4>ZVVVgCrP@`vX>~
z`oWGbq=|nTdR{)rurD$kw$+iHR%3HhUZmSKm_{sDjua<aT@bo&ZGZf+z4cORub_Pr
zd6Tg%-)_=1gSOv8N?g%{N!Dj*46)|P7}YfumXX%4AM&#3pxYC2;i;y!M?v_x6CRC)
zl8sJPtgEqVv6J=6thehMY(w90gPRKgC4vm)yFrToa+dVPO)72b9lGVgx0F(`{WJEQ
zOIkn#l0++(y4r$vfRhe}-Ju|-A4C=Hh4;}gtGc0(=~KsYGoTngk4D~va-9Ap09!2C
z#Cz_M(0&ICru^5ckL;11k84@auZdE^!{nvl^#wkd%f$7K(5%Xie9~|bU~#30ho12P
zmw?xy`Soh#daJJuDZ}v}!IkqribTvT|2u%R`~U9L|04UY)b(lSQ#<fdFruaru;adl
zlTajRrc}DODkDKhXhivg_;)D3UY~N=!?6+`S?NO|on-WM_TulJP!G&y6$#MA>2>|1
zRhA@UjYv)Ve;3-R=O-$(i3j>=ejc}(25g1IFh!!|^<hMBua^pb`nkCT4CQ@HVxzDL
zyY8=77tXi04<B2n67TyLc7FVy=Mvpp$gUq0lh6a{1F(REaA?xqI6@TU+QK7@mk+0;
zG<g8=$oXbwFv(t>d(&YbnHVuyC{B1=GzOSdg4XNUZD9RN+~6{(Lm*{JPA<XQ)&R3i
zY};WEm>B4~6M9|cAG&efW-Iw>u5~%vor}z_mN<1y3xcC(j+Noc-*p6=-;QcE_rAIZ
z>a|g&HBnKz+N!3;I~cZMaW-bMOB>=m7_g*!mWSix&AJU%n<M6$pY0tkiPp@GNxpD_
zJdt^gZyk=xs(8KZy_GEaP%j}Ufhdf@EIY=1-c$qC{|{yF5S>{Uwe7}EDz<IoiEZ1q
zZQHhORBW?iCl%YalT+XM$2gtyHom{HdwcBG8hfre*S)TbqQZ@?QK;Aw*8!Su&j%F}
zHY))ztA^vf+RP&40zs&SDf*d<KZIZUg&kRPpWeK7(%rzr;@vS_=e{n4__$nqcXW%O
z(_9rF8RKj$@->KwZ|4`-PZfRbnB2z2h>Xv%7MvP|jE)qOL26P1RSXl9^robHtscS@
z;YI$%28NjwGP0!O8H32RW-gcp6*4l6dlTFkqz!(t>E$ZW$|{;u_T4&)PTV_LE;ywe
zDBvk<@?eTP#C7?zt5mp#rn{?_{`~%YIkl~GJ+Vw_-D$CHHma^JnyHHszqPN617}Bs
zDE+p$`>mJ?(TyPL(X5$2SZ(Fq1S`%DKqo<%0TxQJ7&=Sk$Xz`BYdKmk7ocC?U-$(m
z3ZD4|oC{72@GN4<UFFO@<%uK>MP^oe=MV>%fv7B~tL(+mg-De~Bs!DePKc61>vJ+H
zO$^yWwlhMR;D`{V+z;9S|G_qqPF$L*>0tyO0ztW4d}9Y4gRX5s<^S_;ENpSX9hBTb
zmn1pIC#kp1lF8Qgs^rX%mPWF0(}*w*vnTkc_u3F;3Tdhp0Vc)-k|+)>&ef<|r7fm&
zxsAEkqdtD6qjusb)z>bYwL9Rshu&r6Pu*4aImr66Z5csFX|WmHax~IbGPU-{s+;Vd
zMAMflIx#rDr%_-2I6U!@fr(7;hY@M4q#5jF5VQniQBxYRm=2{qi^2)tp&vwS$|I8t
zM%A@P&R=)YF6b+gfj+CjCwdj~UAc1+;dbf5Spx&y^ntlxK6&8K*b05{D-f-EzL`FI
zx%d4c`c6WpBNtlpM*6@K?8Gnh{o~SY;|LtcKO=*mrzr#k57|xsG$!9C4hYnoRGhu}
zJxLF;M@Icg>}wdP^@xzcv=ULM#q^^OA1i^EUESf8<S5loc5J(+4xNm{A`Etby8?B}
z`<P)+D3&#gg?Z2X4BOmP*}aL#Kw_@1PUM*dCNniCik_Bju#Esa(;oR6s()cRcWuhZ
zB$YrqA|j<TTdvrHrB=7CZ&p>uZi$6n(O3M`GERi%K7E95KFxv5jAsdcv3@P#{i$Oo
zCj~33ffOGlZV19bBxNhqI6%N8kB6k03q?XXiCGhxVa2y7;XCE#T`<}nHz<@gA|rFl
zAJ-q!<>X%?M&}R^c>I&Gc;lX%x=$6OqB<Dr*>1qRIAJ?wbZXC~)50c5B%p*JTE8F}
zGf;r;xmrA|s^R9i9(TVxfC*mvQ`9jxf&39>Dr3m!^eP!VD!gJ`P2m#eMAk-Zdl!I6
z4Ss;F12TzRx?P`u^{9ZmX+GIT5yx%HZ_IB~NWk;VNdxPMc>DtM&e%R35+n#c)XL4n
ztKr_Vm{IUwm98fs^QKw1F1#C0Ubf3Mbn%AI`ODtojEN&FKvt10b>HQNtOjhtpsOI4
zYHe+3dn;Kmdu%bv?C#CpUybc|Bfuu8o;q-$RFVXALx5+r7w@`GU$aS<L7E{Z&7XpO
znZh2@2Jwkw(z<!MxzF%TgGWE^Wq~o(KkKo9#ybih`yJ<F;t}@5(z81)ukyRwC#2IK
zwkpL7c2I7+jlB0jr}~N*YQCt1m34QHWi6x;f+d*O9I!*r7%P`pI13?g->5A_83@jW
z59}@QWdbmk9f7-J@BFP?RKpW%mR3n;Zkn#9UF#qLqfbk#sMfwE<`$QwyySNfb0*j)
zZ-Z&hO)lEC57yGmgIWXFzuFPc>0d-W`8lQ&fJJ-P{Zquq3^)s(pG}py6-El~N%>Ow
zS|9_uHxJc9=Y2LbzL={Y9tGkCZ!1!=l_#sVv#u?O%VrC^`75yH=ea~CB}HB+Tt@4s
zQBKdZlLc`C+wTi5HlfPmts=0{5#S8bza#5#nQ`$Q_F5mJmbp&(GE$f4Si(l{QGY#&
zY}`s>l`$_N{nIz@LUUb*ho3Z<mElo;lAT_Z+&~$*ycWT-r$KmTpSEutKJ%F9$Ff!m
zeui#jA_?awKV~)*z{Jyjme9E*NqXa-Fx89iQKMg14R-wXM-M21!|~Rh(y*M;^p{g$
z?C`wM@x8h&Vnl4v*+@QZxuU-KY^@iU^@bA%PN=LL&juw1_5rt+5oz=_qg8Bmb<h0H
zD%`JfBu%({Hq&89&?gCbyB1Ob5p#AmT082Z$ubio=+w6CLl!!>keB`aq{{ko{yD8)
z$EAz^byAU@F%Th4EC-A^Z_&cL5dXCL3tTkJf!8CkVy1Yq*x#o%;Tn`uOu?TKq{)CZ
zL-&xE)1APQ$^gh&XnIp!i+Hcc;pFY2W}2Kp4O+1noKg9Yp~&ZNmK($E_YYtm6%<^h
zR{XZpMdAU#_wB+}b-a8+Ui?6eoBqvsN7^s5!bBfj+L`z3pZ$@BEr+(AxoPzs>qpbN
zgI-qjea>^4=|%@1-mAvWR{dSE_mO<Q&v&0<P3NVKD`;gF-tJ~|bp?%u@w1bkylG=>
zBup<*acEb4NxMdDrPk>SGkXmvAiK((u#xmG`vJ0j6Qi-2=#ypLgO9l-G-npB$AzD8
zmJ@p?8oxN<+(u0CYjG~q9*WUibfH*ve@8~v(6H2(9c`GV7uD7+vmXe$KE9vM+r-|B
zZC45mvC!+o{jS$h)j9&pbT8$e&or?GOU@3V>Q{anAbk7(+I{`kEkRDU|255ISyS72
zL*)Nl<7WB0(PN>mC6!5K8wf(Bz}y`oBi3pvE|L~R%FzSVm)nFf-W6rTA!l$Z-K=+&
z$D4lRb5au&Y#I=3AJ@yp>-ESsz?>ifLmjjz+Aq>C9O&>Tc+t~=y-1?q02;&N@oMs9
zR==AoPa#g6LSEnh=is;7=fML8|EH6W-}_eaZ{g#peN0320ZA?7_@yioo7qtbv6P2`
zpPICRR`uV%wx0GMPvc2I#`XiiVu)lU1OURU`vW$vCC8b&^?V2#A&OI$3Z(=4rzgTK
zmo|qW$C+M{k!y$3-K|UohM2><fPs{Q{vV`}zz|B}2FLAU@iCOZyVLR#DK*PvwU#$s
z`YGpB@~2KIb<6p`kxqHXyx}Mi<w}()24YzD5GlZ{h-Vf-!O(>f1I)D}*)A(7sT62t
zvr}M$fO>x;>uQb%Ee5l%trNAKCu173zO-DeLd9yS<Zg!N==d~<ZHl^=N`ZzHI;)3!
z@+Z3Mj|jzn#mzc@CrERPA;Y@gAL*lHuB|#`>ofx*0$PI9o@PI;^^v>$*u;W#?OD`R
z=c1am3mu$qUcrsJX($s4B?UIaKK1XHV?G48C?IZ<5@Nq>>5luf%sW(Nth3lOH8qw<
zWm9{688fuU5F~myrpDYyMp{KQ$!VgagoSUiIw{?Qvyq1-)8YCI<9;LXWkLO0?-Nma
zf1kaN(Z6{DOK$3CZ_pU)Qvvw=UD(<B>~Qk5gp<)#-5rd>6`REv`lPC^ZmxAAJbXM<
z1p=bX_sT8Pk>*F|URxz<k<#3MCQI>5)GhOL(~dS*BB#h@vZm;5<I-JoWu3d4?2r?l
zB@Yu<UY0%0??gMK(tjl&UCzZ&^5w>Ov<N<h2e;&f%n3HU_KGBJP1^32kC70_P!Zhg
zK^`<sfpp(`mIa*W9AdmZ`4sxRT-W&im~Zgqnw}V!+n=^WA`DjLc5iWIaOl|il6_8j
z)cYN{*>n9>9*4A`rM={V)kf&4Yve7sdRSYx`P_(&gc;KX3cA_CN+62J)Ocx^A=&e<
zdk}OS8B%Sav`zK)&NXkUaXhf8lMYp0P))^mUo9u5vAOGRU~|l)PjcOE4&CNS#^1PH
z?TIf^TXRjU%=>7KvtOuEU%jw#%`LZI#Kv>!bW=Ud|8C6aSREm+Osm9`-+Ahcw3g-M
z?R6X<jq6j?YdJV?T$*x!Ni#C<J3oJ0OU@m)-k_KyO0QlXfnhff4fG#tH-<#&y}nwe
z5y`Rd@Y*^#5V*014y-}CqJW7DAq6%hN@Bw82qKY3!?gI<%_(u1HEUq7_+l*V=y|bC
zM=5d8AV7LI>X^g%v^%O2C5nJZ`}-&Bl(x(ZvZbA89(G`{*5)P1FEd4opYecZtR0<u
zhrU}Z)xJL;4#Q9$n=#U;qg;75-id)3kf#tMVS%gMXm=$U7$8h8azui(WX%Jqf!3?`
zus)w>da4QAbLrB&#Bx;El5`xvNo8n9*G*$fSpbnGjf7a#5KU}un@K1@8Fkhj>C%LC
z7!i}2zL4#?cYz5tAPp+gcEP?FDPUNbNgPrNLlEN<FGvcGu0buX?0^i1Mg-2UuQ3pr
z(V}NzXif!UT)=%|#jfwDIw@z9!Csv>nfjg0RiHVnS||-USZ%xZW>R4Y<68p@7$c;t
zES`TcI#{+rmN}l@1WJUbxP;YQEGetRq**4aIiO_i+(R2ABnc>mfO8CfYk<x*lCuaL
z7F+7g)J@G9Wa6yiO7korM!P20oJ_OH#GZ>NekfFMUtpaA>t=aB8^zzz2GR9)s)Y`7
ztHpN}TtNX3A&V*=O8OM2c=VB>fu;7S+=fDh7;jIO+aeg!ih;f0H4dH8mQa?=CoDoP
z$X-fv{F3N5O7aXs3WyI?NTLIvQ+MpnQs%N5FRlVJ{gL$op%6#Z>v=m|8)s>W7x-0p
z0MMV+vTM$))bo=`h)lp{uY2tRTXQePI2<uZFg9RmW%;8+;h4|re#U(0zv_!R1QAhV
zT{5k15LVOiI0576TVajI!$Dh^ZWmDitJYlzkH07&LTS?KLUkXQf}g4k<&bQVY2ysX
zQke81G*Unfy)XIQCfqXKN>}J4wfL-%xY2N)EG#qr?IevKVj@KTf=%X@7=vO%OtgNx
z5IZiHM1;&pS<<#=Z2b-flk;XiL>DnUM}7x+t-gK3o(s|k;{>@#b!$D<n&AiQv6pM?
zf++5?0#yh-q$=mg8Yj+xlU$rRquP#e>}qMozN5VCV{C5z%x#;GA=3X3(Vucw4!_3O
zrdX8I8pD@MzDPTp<!-j}I9Fp${P&bOdv<ZKds*LU%&fOV2AgHV>UU{b|05iWd}Mx=
z;OYRP_&kw;A`S5{iII9l!XO<?=z?BR)WAc*EodxYGE5MGjO{weuZ6D5xT7r_op*Z!
z0!9oR#`Cnprn|}12s84pwm*g1W|agK(%oHgzvS@gSp%H7eE}<7`nhqKdpL671YlvW
zwzaJc-3Y=^bAvM@9?{B#-tGM7vvt1SHnfLrOM~-hX@wcnUNC9Gb98M;AQFr6&Q}2j
z)O^jifb<6xA;@~|IA^;J|HH@8>+zS=aI?sNl>L7gfd4;h{{JHOHKOr<AZDxoaI?I|
z1H-Z<vgTPUmW%ARRu^>B9+c2Y8TSdEtd3k?H(O)^S}AB_ZkxvggCH8T&x38!;saa}
zK?LNpX35DXL6BjXBPxvA;5y(+l*rInQ)vSi3ZvroX8gH4<%w&1MdZZPYJ|nc@7H7Y
z`qOf=ab+@ML?q(Ml|>HE`$uPwpO2R#XB`_JZyPV=eL2PLl7~0(5%V$=2L3fG17oc-
z_+IBO>>W7@?d&iX{hX1M@FB+`*m$(}&~?saf&2)hxCer!!OMp);pYGi3C<c8Q2wMY
z;o?*l0&p==#Nc4Gn;`*kBqmpUy^TRNN#k^JWi-(}(`;1P^^#Wn4wONW={Qs=EDP3<
zGy5?g$k-sNSWH6|-~lMeWiH@<gb?*&9K*Y(-}cw^FRtZ&TCCnmokwtRv{})efemh9
ze6B;}vkaehsx?0mCMl^G(adcZhkqSwl6h{<tdv1uzs|3%M2)#0N-7s#IN+Y{i6zip
zK<Cd6W7W7=gnRwhFCb=|&`T#v>F8TkIr}WDb}n(}8#c#`V({j@6F<f-zBwyn^o(LC
zuTAlnmI_CplqlQ7-rU8>$b_PRz9S-mjGYiw@a6%{vf607Fy#C7YZ|0*U_~3+<)TR)
zwYF^vGs&J=9guT$=`<YcCJm0_+^KQg;c)SKib9tEn1QNCDrPC*y8wJp+JXf{h1!qw
z;O`fJ#8I4O5F?#e*{@aHED2fOaAY_m!H_#U8ieigNM`20hTdMMy;P7tR1%Gf4nLP<
z_YERQ*?Mhld1Ol53#5_LKnT_GB640Ql9Hq$icn%Dg){DB;Vzdwbnx?LQ~s`VE|Qq#
z+<9gBb0SzmPa2FVxcDP`Hp%Zk8Z9G5l9Pke=#q*lvBf7Vv9_@W>!#X~#RwRty|y^q
zQUd@VZDK9FKRwlscF?i_59gAVf3Hm=K8V9HMSzbC^BPSVOST*Rh^BNPi9`!ShyOKs
zye$JA0G=wea>sS<k|1oZJE+(|fF})w`-4uWf|!n6f5dpwD2u1a-D=W6sa;VsvhG+r
zF=5;85|C=E$<ZqH@g%Al2!yh3y@u*yv8}1y{z(Y>PnE3_+%Q<VK7ObZI_csn!WI%9
z)+#h%G`fdSW}?~OJlY|a(M5)>LN$0deYJhRSDz0lu(B&#HfCNQy!e_^^MKF^yo~K+
zxMboX88bo}#u!5yJF!_C#aJv6lWO?l(L=Ri<2iSX*XJY!*y+OD`KWzyL_&!RX!W+b
zCfV-?=0(_Lq$diDAXV)3NO>Taj0Vj_3Z10(R41XIr}R<mS8L>P<0kB`@(tqpaARif
za%yS>99oZ(Mb>>>J8HZ1>%q?^BS;AIl|y?p=VjBpy0x!4KV(oUJi&5o@vw_jzQVF?
zy@IuW_jz64UT{zTeL+nz0A-lT&Pctnajlle#54<&dV2iF4DDHy;8(F0i!2F7zjE{D
zul4^PJsJp?_sK#OQnO5xC8(j;Z(r%|1JK#ca)_SJQP?C?kAFde7qq#ex&Z)#?(pa$
zq*ouLJ7(xETW=1#LO|r_V?QqK;9<Zr@I_1{AYADPs^+>4s*b-)vRwex^(+@zJOx%8
zFCOhz80i56{;sunGHP|kla6LSNa07_E(RO<K!~&NB+(_$7p{x+zi~NV4w`~hbA>^X
zv5r&x(@a|Z_DRArFLRd`uSG$q!~bT5Xq?`KhiZQga1rnco`Y4Ql-z@V%N3Hle|*aX
zp{{SU^$|VSYY3UN<IA-Li(I&USW>ug;$9rK0Yt?)kB#PNuqsgfrE<`R@L*>fz2`Ml
z8NANOJo3B}<~oQ7?8b&rJw0Ci+W81@Q{a*R_!&av4>sNT?v{-=`5qD;{P4aZSuOQ#
zX4zK%*AM<b;vh@_#{ad}YfD4le)E3>s^xRVaKms^Q7Iu^GzkT=^XR;r9e&QiptVPJ
zkrt(~B`wydFI)CuEJ>v;%7Ae^1WLG!O{DOL8;&6&(?)5N`>#wMtR21F9zXR;5j0|D
za{0Q}$W}=`wLuk8rxwXNWxAOi{V#wssgjB*uySU_u$JQS+rc|;%?~G>u!Z{xM&|rk
zo1F__QfbvL;_=sN`~cfD=+|fE)%C_(^oD)`>t8Rresm$TD6rT07tlJdFZxwUGU{wi
z6`cWOGhrMT`jm-?qgwA<VFyBS3{UL!+M8L`#P`-`5$w2d;njH1FDF2T%Zc_*SD0nE
z@Zp(`uXIX7u_HfL3{SQNqh#UTH#=fDFab}3t5B6wp;eiPVrt2Gw{{xFzh{Fr#!tSP
zaSm!drILS+@zw1S=Y1E#$no(ZKu<J(e)eu9wwuiUFr4XfYbws`;N))Q0b&biv+%-v
ztNhf0#HKJwM~#@ldErOh@-W*?*lKH0ZRf-6#zT8{L6@YTs!@Wal9jRYb+L@!Mt%Qk
zfyiwSQ5^YHeB-X}-)0KsoKeh$$hqw<cBfm*8$!m%C`#5v6>5agk2P=DMu^vqx5=LS
z`3Rq<32rnlm?@+#S1sUkZ`nqP8+_njz{uDc`5$0%C=c}%0hJ#XCREaAway7#meN@f
zyQ_KN4;sks3Ccg&<%D>gcs|L;0{BUQ1(T<Jq8&E#vIa|Pzz!u5>!I&X7vvO8?Sh?j
z^R?OS7aj<GorR~6(yZ!4O^!5Uwek7j;pn1VcKwLSr|JaUKbEflEAKIBOkr*IK^O+4
z*|(|7RW(H?j-mvvQzBL!<8FNfp)H(5&6Z<-F5KaxgvbIUlp+CPjBf=$qxceV@5o;i
zOO&vzZUFO8!&gA4{jRTeQBfExNqIlT$`MxliIhs#9hY@xD*ChCU^swbp>TtxCHVAN
z26qdS@WPtb?zPtO>iX2puiz$Q2<#v1V&dvotT>#ysF$Qj)p7mfl_Q72^Zkhh&;@mM
zXrm^X%_vzA&)pp+W>CyA7HSF*kD>T{e7rcM1K!YZaNX!#*rMeD64F&?N?;GZvO$hg
zr{hzzpav;vRc!c%kTn3E&6rpW#X3dlZ21c5%@qqc&=?F%u85qm72gD|Zj7shMhSIQ
zzCA`}ixqb2%7SeG2s*=Cb^VGka1{q$<0QlT(v1%yovz|?5G+$%sM&DO=#MpDqI}cV
zh%sV$Mx7%OypFdad(4+-CjP~p1d>UckGO=P9fs=5B=*8q=rn1MqqAVtit`&Mm4#+1
zwDvuf-cBbdvn-j#wPZzTpbI1UkemSdE(S0azboT;Nq9%Eg3^R(_24n@6*VH#n-EcP
zXwEs<7}c>q>q_E<jMvHguH9dZgr8o!zXUi%A(4qTeqVBx)Kr31%MIy<kuaCD6Hau`
z;8|u9Fm%&dzd*ks8<{CXJGm0sp6l=m&@UAoY4nfD9sE>HE@JJn>c0E0BTg8X*e5O*
z6Hb&+m36b3xCmrvX88tYP&P^eS44)kk)R(WUv+*gXxlrp(!_kR#%ZIak?{O6Cd`C1
z!IPCt0{Z?ug)Q$Dah`dsvbnR&shzC^SsSB``b3{Wmv&fd8g1d>xb?^8OE|;Gdf9fB
zon41RAKX&JNWSYb&oVHH_x}~j5E3<ZMaX4afiz4E=&V+alYk?t&x>onqTti@qgr)W
zHz$}igF5}2hL);;80$qqx3JQG&#br}1w#+gBj0NN^QlB;j)F!A8Wb+KFx*_Ea}Q3X
z{05>Np=cS}g7#)R5h2Q{70d{#uFg{LSKh*q5Q_<F1a1s*l0g1F7e_Hd`*jiB#?u0B
zhc7l=1srvX>RCwTZ`^@wi2jorby!E4t|;kQ^~9l~@cqru|E142MCD24)XtZve9vF4
zTp;CBt88@POTEkW{_!>0)*A=&gkQBZjurJ7&gn&n1JkBkeS5=z5Ts+@T(_AI7aCZl
zn4+I#HOqOcEOry0kA&SJw`2bNiCs}w@ZukMj_Bt7!l@8i2c*xf2d4Tb=!aOPJm-B!
zDD~Rc9~B&4R`>2u;I*naX6#AQDjDS0BAopE#S8Nq5SBgE-=nQD(GQWMxtbAa`Q%R!
zaVwtN0@eP9)`(gVH<+@)a)2g17q}fIbvN&C{T(;~@$6xt9bG4amoUB8XS|-SJBAxA
zi;&i_uysMdh@vr+*ItXF`^VnkXg+sg&Nh|qc#`!ddq#nB>zF*PKR&MT3(C=NuW{}R
z@gO?Ry7UYRlL%L@k?u)`R?-L7I8%%)Zv!xXvhM4Dd+yl&jHQn(Pul=qxpq_Ctlsi$
z-S5_<Xr%0cG&xuMRDbtEsgf~k)?sfweVrn;RYSZ9Q9)TMl48t-X2xuYDc5=c2mSy*
zx_kZ;&)b<eyEvH|{=c$vdm}3tPR9R0&M<Ls{$D7{ZSDVYpuh9<3*+w%ntYJRV00l(
zth-FDo9LMs;Vc#BT9crbrKzRmx(z?BxbG=9l&~q}7F<u$CAjXl^Sem!&T~DNJ{)g8
z?zWTD18<Fx!+9P@%k!Xku|h&17I0>}QTR)O-4F8fyG>ZP|BjFhqDj<i`@LHJOuvrI
z8w`^@fn&v(NPMkAj*UTv^Aq2_eG|fbHoP^(r+f{ayxyB3I{W`v5EA?Q)c>5PzI}qG
z`*D6gy$sFNeXTMG%r`Vx#B*U#Mv~B%uKKH)Z_7%HzZ`inZGW_n3D7-dP=FQ2cm^hs
z8YOZ}zrg8+AV1T{QzA3|mg-*Y*}1kpFiXOW<^{j|6&lHJLK!5|5D-E=NK3nup8RHj
zs&xJn9A8x9l-pC9bUckH`S3`)Dqtut3n%y<;Yh487xiaRS2PBbt<2@{GZxw_VUbsZ
zS;!N|U2+g}A76}gH(&vmHEuR}KmzD8iaeh*)zjS1Gm14rT8S<R%kRu6Y;yO4hi)c;
zYq4dtEt;9`oYQW8Ie=;>GB2%ZxZ&J0|AM)8*0Gi{8VMqO_zs4q5JN%|b9ZE2o7B8h
z4kJ>K<3-TW$*kr$=?zx5)GLY1S=~LiyX`&$>@3DT6kiKWqfYLU9o{es8DCsi3?<2S
zVm|}SH%~Y^^=bm}D>A~nPS_t&#$Yt2@8U8;!9ZksgZkNQG4zWP<E&=OLdhUs6vw6o
z=^`K*3hGo>^^DBsxF{9D-ZrJ}ecEU+xSht<(>XyK6jOx)RwaG5T~_#n9kX6Yf3>b<
z^>z=@{A^SgAJcuSH&q#<bHP&y)Q|Y=R~sj(kcAZMDYuI;u{u&L8IsB)UMGZ;dPY!^
z060#l<-Lm5n3-i^HJOr{TbGO+atvfQ`2*LqYwlFot8Ovsp7{~#c^Pf1=Xh}d<vbgS
zy<KpXQizyi`eZ9iu|MDPMj>&4Y-Qd}tMYb{`k#pY=F?mF-)KNSCJze`xY$G*^<rGg
zU}5n`)_cejA<~F;!+-Y~9Nq#_aoNq=IbE|c#+&J@A7|AhG<TcqJlUm|PwVHIiqsS!
zssis9SCl#}sZvG8>)F`yOG1s6#DwGQE0?iA!E2SiRc29qx@fa)!hR=%wFzdsY$Pe8
z)V8JETV|oUcG*^Cr)_m0wJk%Cy8I!wievv>NQ(x!3rh?d$KH0n2QXS5T5Kupu0d(r
zClvm2?s3SDBF*IF-m^|qIU9WT69=-8&7KO#b(vw4*BdPZPt=f#s|eKbCZUf&#677Y
z?4CTx?FWQmeTk{NRyEl~QuCo6neNf*VQje?Fnbqhn#xE33y-#3fo5m?$xrVP9^$FR
zf!ms+V_S6gvEFm{%mc8;#&?hRyDoZLQ;dW#28o<XW|cl6N0!SnwV2vv&M4}b#JJmW
z`rN@AgNm74QX7I+z=UwH?M{zsQ!Pf&y5d0_`@zjsI}Vv<nYNFVVNr@QhkhC!fH>(A
zdq#mr)%CZHn$nVx9-XBEzdb>d0jAJTl~y*{`4c#umzZB!A!$npizpwm_lNp%860Su
zwAJNmUKm_fNXs^3<&9}ctPGA>HOhB6X<%Q{xB|zXtY#OOTJT%tY>I+IOJdu^9BO&O
zU!K<xsLvkz7g6`iy{L{zQ3buLD`<45<o7-;7c3C+HVP*Tfgueic>)FR3e^*_JC`yq
zI7g!vG*y$W-v$IBfJqHY*dei86i!+EGEI-jq(G#cN*|)9=wTh46wNd?>q`;=YyCV5
z<)vS=WpR*Vx~Y15*yO?jXC<v!0U@!`ox-pV(;pqXFbG<P+$}54L#f|$uC--)6hPjy
zWklrYbrfi(viMG6{@s&kE`&ty?lknfLD-XH9lDBxA?64px8g7+=L&I~@G(PAX0d80
zPf#nVLc?o?3s*b39z&WMBaX0heL5cYtMKYH0J*4k?~x!H;olD*JKLHjI~WMH$+a?|
z7DJl%*L6J3UQ{JeKmCG#&aV{WY7(36pWHPz!{yzfBnlkR*8`xMK|W;UU&r^lx}td-
z!rJjyFq)&|yJx=;e08pOjc?5K$hf?R|B0?&MDO8!!>^K@08ts)W9kH`*M6$dpv0x`
zkF(TWA1lI(E%RR14^ddqCLa06rIb?(w>0uPb-%ZVXh-7BuiYEtRc+%eAkHM-%yN&G
zZ_Hl;`(8g4ZRqKH1yC3FwNS<Bn}7aMkA7uvc+nfTJ63-z37Ok*mgYNdA5N<?JK!>g
zy{z0!m~%K(xA+Yg1fOBY{u3%q7HFsC|CZiC%8v<d6WYEgfqp-baJzD7OBNSBb%3<H
z_>jfqXYXt9z0b#TGo_$e3$8Wc(0HqHKYio>JJGh$Da3g-=<Tljc7kASeJ;!p)I+f;
z{sZ1NKU?}=Rh8pE29q5B`}4>v4b6Xj>;Kpl=Cn<hCT|g`pe<oarWG-416DrLzyfI<
z@rt-|Qdi=%oBMi%5n8%ZW?Pb5q23!yBxVLO?actv#mNz33@^ReH@sd07WhY`>N=ti
zqNFOMD54g`#hz$Aao|WM{^V$}@qS#N-)~J{Sm~_aZ?~tXx69I>9?vfPviluho}E8#
z-v`AQk$;aVsS1g~1LF#fo~Q7is1=gRbtWvh=;ezpTK#$$mLJ}>!+9Z$EQg8z=azvK
zDhcakapPLkg^(m5Ulhho6rIks`R~V3uM~uunP*C%Y)U;lFl@PjMH*sisYN=8^8)Jc
zdG?SX`qyNp6yzEp=YripBEvlc$n(siN9ORh54$gD1oM$3{#NlKtYWo2tbZg@FtqPQ
zX=9MT_qcu~0+s_X65fwNp!3JitR_n?sdb>TWyi=lx2yz6kC0WvScwiWn$Vk_0wX_z
zg__I3o~v$?WfeOLlG)3^b^z@mn659kUyh%oh(H!A%n^K&DhQ|20Lcg1EDl1aZ#`iF
zvdWe&s&%E8SIvL=Oze*yslLLLw$6L+Zy8lM@EMt|k;8Tfa(YtX;5fm46e(nSA>M!_
zU8J#%bLZ3qFeOm|_2%W+#zrteNxI3Mwh1&fusDh)q{sNf3wBKM2=-ON$dJWQV-rLM
z=TDGzV<OfqlEVotmMc9TaI9>+v4O!kJRAu^`ADP1Bal*J#fT&8bL@+Gf|fv{0AGdY
z$Wt5Yaeyy5xk^GJxG@4_@`qz36ecatHQm%eNq|eD;QGc2tP`Fw4`~B?yJ!RI9}BP0
zl3f)D5b!=1%Xo)oJ!LrQhdc5O=*(o+TqtKnpQa?@NnlA&3G?@Ag#l)eT^Gn!qJA);
z{x|^KWYHByf{FwG+Mq}kE+c3B?DXw6fqO*AxSI-hT@R<`BxdLM*<QEsbt;uyXR8CW
zlkqPF%t7L07`BxbI~VRpjRaf5PH3lKNdw2Od0B*le=`eM9Qo)qtr~5-H<Tq1sYn=<
z%s)!RsX3&Ljn>|b>6Vb6*FJVxB4-g`*;^CXHJllDC3YWZVjlcS=tCiqJ8k2=ydA^t
zDjt)o#Hk_DVH4X3e5pyOgt!V%s|NPkN_of-s*D9gw8>-Ty7myxWp|UHGjQvo?4e_*
z(K{d02X@P9ka;zFo)BBeW24On3yeyS1Zo0(s_cz_r@QR>8657jjNn4)!poMjB9nDq
z0|2h%*Hx;PSbxB<`cLnYZTfu!cCPc3APT`I=~HNR5ryk_FEFOw?M!q%49%cC-E4gW
zX-pN<t2;4W*kTbrPE{2C=Dg!kwu;f`17X}(%PQdPMv07=cj9)IzHN~xYFyGgSogC~
z_0rGOq#A=#?-KK|-<B+|)}=~2xN_Ff*PL96Fx1HF6q!0&x@ha|zBQB~eQ1o=P24{R
z;o`eYsSibRzGWl@Fx_?nuPbO0MwSq@aUZIdF%K4+fw`ErRrP!tL&BnW=6k=eq)YBK
zT}#O@+9(Tbg-z$xH4nyxWA5;vfbuIt$(^HoFfo3@qoirRI`%i5H_OM_o}_ElRx{$@
zf0$UTH|nV#&{H?2(tH$g)?UE^VdwqkS+!%!se@ql90BRYB<V=m<oE6nsITR&zWsv#
zrTD>BE6WLE50?U+zI;h$lc1C7{5ndEt0Sx&|0xWsb>!NX%*Ou_C}?gDPK>}0!~;6R
zCBN++yHkC47}?|Z2L$=c4_!Fy_6-{uSVf<vV+3@Rg)Z9dkT1r2>|M!YP?ofWIP(K0
z+!cBsxv?zPUz)x5%WDhBjao;?mfm7r${*LhotEB4U;_7|coB;hKHRoIQ0-IDc(j?%
zm6Q&Dl^}P>aX#z@ESLw0RMO`PN2b`kM=uM>enkippxdX$YvzNGb9Dh@%WWgUKw=+l
zk5tbUR?DKB816!sUsK-Np&nr=J<TkNY|+FgDes%!y7;tZcw*TzqFkgB#{>{tYRRQg
zC!HLcd5~)?%v7n*2$wDOg_7mEw!hvI{rFz3R&zheIKS~orRe%x6e!p7kM@16b_ePn
z<HjT#hvI1-%KMsRfgxS<#u*_QDXp#ySe@CnznQ`QIeIIBn8A`Pc^8M9s{8Q2D6H4W
z&ATpZY2x;JxuB!)4##P-#$+rl$VGE^=7f6RIr{k*R>3^v?)c<W>|GM&^_5$Hie>}w
z-}JQ${xSmJ(b5*yxl+<*@h~_%hh-oc(7#*lpZ`!j4k=wTce;GPsr?p1@xZ&@o7O$w
z@lD!$N^H=ElfQHV-v}q97nth{0=(p>iKj?Kq$G#0pGc#*hhPRjbD}v#H9$TgB_pk8
z_5qK8OlJN^xn%l}CX|_p{eM5^)%@3l{z3Y_)*rAh1&ZFI-(lzh>Fhep(wpg72SEs(
zg<4tCA{I*IN@}b{zp}rXno(F-Xw3O53Hun`&YiQB`{EmRpWAW3&7bGP!Rz_N4wo}E
zX@teSh(i&POBut<ACq%7)!|~y*_riyw5+{n?ybXoX9B~GnRfHT^Je;f?8BOtcmuI*
zfjiRs@z50=?QYJGd>1}W3B5)2Efc}a`4I@w7sGreHzO{%FSjGR-8Xi&bG9sR?*H{P
zwV(C*34zcl5Gm)QHt^CYB9_SF!|r_2pnfu|`();I{}CBr>?;ls+<g-Q6#kETORkqk
z&<jz#Kc7oTLf$NJ(c<Lz(0(@6sk#^{fS%<w5=6^Hab_)5$V3iCFADS#`IQOSJ!&p`
zF~p}dOU#z`1NiR#GD;%fA;y_A&UeSQnHq8f^Zl4jDUccB3lwRgGJCS<RYAW^on+{d
zj+b4+E5~1xXBQIrE<QBAYQ87{{p%vx)#FKRP?DP)UQU07#g>MH)dr|nwvq4qN0L}_
zPqg3Eb{YMll=$pNHoilPe%apdu&J334F0PmsHBB*biWSM>++FBE!&vj5{NiL+~$XP
zypGnkx>K|2-g)^+zpn1apDgsqVaI+3U->OG?Gxe!P%Jy9KfI$}-kcpdk;_|S0&`{J
zCAo~fNjvyps?SBjFteKou7@gR-Wh^Y4jbFT=N4^Mc4KK#r8<d`kQ(TVgid^`)Lvav
zTk&_UaZCi!@I0{pSx9v{4j;W-L2x*DsDKuaU*<g!@+G8)@prWx8k*(0=K4izzRT0@
zmwhh-O+^p)2W~Pzf_XN4-5#%-lintQJddFKIqQ8iWP(o{E@v=VK^-1ano@_c$Ik94
z#*Af|3fmBHqM-E_YW2o>*tP7b&N0@U#QbPNdHKL0|AFiFLfBtxngKy8F_L7<2He*w
z%I2??2wE1XFfKA|$c7E+w_B%DoYuJrFPp!j9FWHZWo6sy$+Zb~iUrKqv{G0UFg-}>
z$k0rVq@Y}dq9Fii_+Uaq8B(M51omj{-yL@BRiIHZ3jn!YmYwF>uE7BWQ#2J#uNusE
z7IBWDs>IQ5H0hvvH!uJyb`Kr>DM17~v!@fjz>jQ7A3z?6)lEByus;ynS1e2tJH6z7
zq4(2*{j#ADGh0v3d-qeVlDY!8-IsaqhapBs)IB>1G}=e|+o&hY#=T=ZkQmrl_X%fq
zx_1LXWr-~4344-laqFhJILQ-@94atHMq!PL`ig^9aUn}D)!K1+ZO}!v?g-GSm$rH?
z7t%t3>A(U*yEU<cj6fYSg@Tpq;nu5%!rnn&hl>$`TBH1~TW90E_Ao`@B|`q_pDu0p
zB34{jf;BLs)5i23Sx_NIv@?zSt&TY=p|?yI;;Z<uq~c&Osal=6aF_hD$Lti<1okNu
zw%(NVryFqntM8~k7o#l;bk+tCF-6SKsZ5@}+G*ZFZnzw?QyWt-=kEDy6V#niPqO#6
zKyL0Vx$4*jo_VfmQB{3c7X_@cTFrw}9KN@&di5w~Ti2S{5IL=sDpP{}Ea0)5es!{{
z6k05MEul<BJ+WRzS9#Mula@wQ4}?vy%pq5l2wK{2z6nGbUKY<R!30qhKK2G(w70-e
z+ZZ_6(90ol|CJf;vS#(M3Ilpfn=FBjqZjC$?YFWgLaFF%yT~8frY@8b@EB@gZc0NK
zgj{tmc1pckt@~4iYfqw(jjkMyfL(83xep`WdbmPF-pfETmHWd<mww1j?LW9!pw?3f
zPv7@&ME#I&Yg<7+C6Rl*4|%P;e|dHb5t;Q56q>Fseg_SG*`%v0QUfc9>R~#4bq&GH
zQHTr<K!Fmcy{oC>DcY$jS6nP?PBV`1NWGg*pn7A{Qudg!UO1w_-9*_5k_df{0qLWA
z56atc9__97F2W1Ar$Kqox6CYenqo0<P3#ZnE_c>`&*8NXlCokJK}KJBwNy>}W|-L(
zA3GOIXE_!J-BL$1Br{QzXt2n{veJ$cQmMmv)Tw+o=+6oetepn3RV;aDFDBSR6q`=3
z@DVn~7q*7__U23@qJsy~$v$ZsC>L@SwJEH_*9ANFT%SI*st`Se(WjA3PjqHwdxMOk
z<LF)12Bs5#lX^f~dkOFV-GtzsSYL#;Kd%rEwOx%-a?f>-fMZUtH*5}d1{H|<jky`R
zWGiy5quyCaPu=W5)k_LMHL78Lk9V61bKHRrX(gnBMMfqC5{qF7`p5q)f%{Nnhz9Ds
zda)`XLzP2vt{0p<g?B~bO9^{%=xH+ez7AUtujiO(M)w)3a9YH-!LlKZ#&i5dJRAK<
zLi-E(g|5`Trw*puTL{RkLf5~G1hyP0hOlXB8DfZ&LHn+%=-pTM_DC8xic|J*D0%KA
z4XDCMH-(phw^x5j!Isa?VpNMy1_)^w(J-ds4<&R5A8?LZcBAT1)*o%q!Hh-Nc2p#R
zi#V|+p=oPIvZ~~J@EEvSMGC|6p^2L%v9{0#D228>0pMpI!!`T_4WX^P%MaVB;&EKK
z0tZJzEb9K<RYi26l0uC@2clKq>q`*+;00p=ih3l0z(6XFerTXlh$X#d*lBKE`%}(E
zPF!ImaOLOffK*doj%zf6BDFvwS}ed|OFdZq-}O_Ijoa38ZG$01w_j9R5#iY(f(9=*
z)W>!z%R&0-yawy8E4&L|J3n-Ug+z(MBB;Vj<v&M1V3+n=yPVxJ4UdF=XBSt+`JKQ+
zrI}xqAB*1<n6LnHOrKxj-gZ#l->;Xm&z}ST$bcGnJ~4p0i8hA!^A8FkOa42ICCUe|
z_Lp6XrIR0mf!biSYmf6}xycX=p)c>SMWvnSrq`b#%YCMD-#-|xTjvbHb>Z@q5I?|g
zJm3ERwUhY|xG59he-EOX){y=mtzz$m`Yn1@i$%aLXiYg}n=F@H&J3xew<0gTL<sGQ
z4}jEkZ2Ga!JrJBU$d62J@!0@H10;IO-krZjqT4y%gJ0ecka^ixK>E-~C_OmLt?5a7
zw@(Qzf$B}s!SSMr2)U@4gZT3_-{mge_sNqR$|U2*n^%_T@ASBlqI6Es%geR&aPs<2
zkyGg(zPOm2{=q=Z&O=HLY)JLDU{)eSi-l;E8rj4Wb8P!{EzkR<$LnKBcUGSKh5|K?
z4F3re7OGS)Lj+i|y{(2^Y9<MD#ThYp-to9)L0fj4XITo;Gs_cKKr#aaTtOb#2OAt(
z9w|fJeA0E%<(@WY23hFy=I)L=a-0U&&smyTm=+YI^?+Y)o>Cm7)x)_bkZX6SvLL34
zf2r-K<9)E97nJPQEQRvNj1>+ucCPQA@^j+frw95n3hl%}I$!id>MbGB9no~7)KnVN
zZBKCd!T>WMFo~>|FL%lEFJ?%;h$p5C;a>E|O?R4&?N7Q2a*2b+%uSu?HvECZdeL0l
ztk$KDJsv6cg@n;dX#HRqQD<fs{<eSuQJT6tI=3Rj6g8PTyW)^X&kXP%eOaN#G^7*3
z8zEv+sE+IKyjf)RC)L&!;UoP$7;&NjiyPEtd)unk74;hyi3+I2bSDDj>0j;@MJr2V
z=}SQdcY{AWQwvaB1K{!#dE;jO7_G(A&Y7Y87-p&Hr;yL9PbS^Y(|RV&{|ow^6s!I8
zMbUKjwK9(CA3fu)oL6|noupPSzb|!cPx~@4$o;ym`hDPO`>ZNo>*(Yj8;?M(@X@9p
zE)?dz^VPO5-2E`8`cBrfR-su63~d{n;bDjhX$kYC30{u(G<RLL#K83g_f>TkW3slS
z-bin<b&%aF_-0C#hAcWLu?`|7Eqz~<t8^`_t+>5ASvGp@nKCnqTaR<k<!sBm<}AL-
z>p(g+7=Utp9m1FIcucV3(MnlpUm>cmgBAQcTQ<}BP&`3k2bXFxB`H*EisedItaUrk
z#1!&NYkzXNj8^fn*b;L}$`Gg8o0o#8puzk8j)Y#h>~wUkG5mz?P|1eN7*(r$s(Ctv
z3C1Y-UG^zK6VB@ufSvh7b!Wh{oN{s*lH++w3xqGBKJr8KVrjDi3hsDqH*bkk`b~h7
z+ls_76I}6!$Tj*K!n|n4aqc$<UfmRqrQ51(JgCg}OSTY+DJry2H|nIBV#GR>v#Uwy
zGVlZ{@JVAAt(WC&;vWgY(KgHxK{ymNFNB_|=UuI4SH=8lg4~8UM0KPfO(Pv(rpJD~
z-TYCmO~Q6Pj7c{71*iZhUm!vC$X4;pn(XNg&GJbM6HgYUUK^)KV#VnDvL+bsruqtX
zOz*1+dKcDU8R$yrN3Hf(XNoriEzG6KqO>B_)o5xpY$DJB*cw?(Gj@DjX{0>0j=5zB
z(D0lkLwG(q&dL@sqX~>K+1T^&Laf)JVj$w+?~Km4fAn24Ehz4)*iWfP+q=EO<}zA1
zrKC=n_;QvB_dYk-Dwc9f1IvWrjUBhy>x!_z5mxZ+Xrq@D-e%7+2!ncf_zfc5N|9Nc
zB4-(wun1APq9(zj12h1oUe-c@dEzusB!S!^<SrjBmnev{4Bh@oBZg?D-=3vq>})MW
zv!rRR7x6t|ok$OoDdYJbb3hP;CT@6{Vi-YCOv6BVz9Nkr^zKjRDVMTqfrs)ST4B*<
zcCpiZwD>9gydwhLBHq(%KPUQ7Xxq7IZ5WYPiHY>QB#qMi+=Ug1(pe>#s<EcHy=9fO
zhSH06OS@LaNZhp6bXhpBSsP>3C{pom%+U-CQ$0{6{vTFMx6c(9WJ|>IB_C~A516VT
zk!gGP+csd5wYX1Ym;qwc#?8&u@GMl+2SQ7sYnAlUP2sZ^pD@*oDzBS>Ic*<90Qnt8
zaQy%)0~h7YNEt(x*DnsOvY4bRT_RES(`zu3)Hz8vJ)oy)8)HVG7<1rgnJ^6h2^~lD
zz3=gq_q3A$ex_uD?zAqib;WHEHBj#l9B+;aK9DrJc||+qZ)D?@N^4lOvhSj^)vy_>
z2ODj}Yt~d`DU;rP!ZEttvUX$^3hHYLq1z}hIHCF)@(1PpBSJNkKLOfxPel{O4y$Ha
z)>1cHhwy%v!E;qN5tc^1PmUH=%I#M}9tkkyhK;BOtXKv&_D|!i!F1>1$`KiPHAp2>
zZSq53tz$ug6x1xpX0k1{dKqxAqSHudHakmNZS8_ikq#%JqIAMA>sJHdC$9m;mm`7P
z*`id{!B?Xj9Ficiyzx2FNB~jCetZP`EO>-E*xX&ls#4Z;g6%Go1XL{Ykq9z8L{>wh
z6YMy)Ivf6cEK6I-w$4orfhr;_y+|jLE?^cKi59vJOHhbNuyjB}Y)p$c8C^qgWvZ&l
zPMm~aM8o9L0mb3%j#Wd|@P8CsX14#Bw6nAPul-<G8voz<wR~>O8tuQ_<x&l3W6N1%
zi)K0|CJYCqNw!?``N*<#kj-Dek2tQ01``U`9259@zlISKPah8+K6f6gT&*RqmWLsS
zle1ojngFmQX=^|eu4RW1GPRC1phc_|wI_LRR{jMGK26Lq-hqI1N!ahx^5*LHNYTDu
z*NVG@!-1@22N#!s0Jm3YFS;j>;<E62O?Z5J5BAez1Gc@#eaV@}+oS&Fy~VC~s7@cj
z|Ld!s&FW@aTH{oMWWCS~7Vk4g0l~?t!HZ^!LI!Mw5je**r8IO69+sBiSfV`7m|Q}N
zhr&X>a~OKHmdJO%X@<*?3w%jwz3|)l$}w$(S|*nEhKc(ZgZ<zDDT&TI9w9-aW8u9s
zTMrhJIsRSAsSWSYa<L2cz<YAN3S~PWn%o@ffhUGW6|(^(qt_OKj-eaqtX;%w%BfP?
zr{1M!%KLnwhDX^maE!gu9iG5^uVe7HWh!a<YljA6F9{1GIY*GEY7n2;ZL{RA<7!Li
z&s(Ld?&^Fsop}%Lwa>by%_30>le9<9D^ET7cw8FyLg-!Z2pQ(UL@L|Ho2_=tXgrl6
z!XHC;DIG|P7E<x!&Q@Ot83quk3Y=Y2IOhW$*bzcpygW=AKR<YD)&x`paoGu4eMOTc
z1xAqXe}0f=(iZG&ck}#5tL5~r?BK$c0LT4+&$6v2d*JG@dWdzpE0j`G^)Yvx3y;p(
zObsp(0;tsoY`$0)->-GtRX!I7Zhok&yIFf@%I4_KJ~peFh)HzHs<+QQg(mv*n^+n;
z3C9<D)&@yRq&gt(R|yx$rtN;<G;Gj~fpn-#$+W2^l~U6{UPIx6&iQ~#;l;(vsD0Ax
zD$H`w8#2n8C`#Uu*(|uHu^NztIAfOTs7Z?iGf(8vFLp9esR8H{5XBW0A0)IeC5EX!
z%~sJi=zZSUg*g}5;-aZy$I@bQsYXrn1*6$m9#;KC!zZ(y_~0SDYs9gaF5H;_1F+X4
z|H=KU>JoW`I!FvShB4ZYDOD?7UOczO+TTpc2^6eVMa?bO<g3nP(ZIelAR80YzHSx8
zTzWfwtq{D34rLBFOk7Z)es0C{>;291b*IKD+gxGw)`J_#6uNfnLT^?9=Ni-4?IsQW
zD0mFwl>@iGzL!<*fGdS=I4a-&I^yP%7p}BNcVe3KCI~jyyuCc*hpxvIs=|RSxm8UU
zUrtk4<?Qasxs04%Y+m2v=Z@iS-Y3Rhqf*LGMPlST<;pF-AKsMR=Uu)pf7$t;+)RGG
zf_4z}`M(^vyYc(cyf~+A{^~mP8tdMXN%mBi%mKBx!uB9vrtO<wOL4ex3POAaP&ee7
zM2mApFf@J|6L1q+vXE50JO1jKG}hIgb~)fJ^+7g#hY&u;26ow>WO!dZ4Ym2rxX49C
z<brOhQ3Xt1SZk9Bnd8wLtS#Z9CFAmZ{uj~}FJ!|MzYM&DGcv2l06Wca2UVtul6}F-
zE-atOFI4t{%hWCKii<7OUw0>?V(@fcGbOWjr7ANdl)96O^@R$lvGORU$@%G0vE+A7
zU?gZ;cCS=P2t!Vf6IBG<)Rtv#ZExVajlIc20Mgz>m?VchY%MTu8XOmgi*IUy;qkAS
zwn{CQ0p9MPL{8YzHcERNs82yhe#}cqHjW)#GOH88$WCP-(|#8|Him2Wo`5yY(((B=
z2I4Q1lffos)h66O2`0QB#PR3Iv=L^iTmuU(TgH0xb4<S-dVgV0mMzL;nW$Qsu)UlK
zw)hT_sk$nnd8tGg_OBYlt1JIR-DQ|{mnm)5)XSt6R7Xs}xLOmJ_ww>YPcciXVh$U#
zxM!7`t&6JU)vlG{b3s=NxweK%qu<&Lw6g0Mr`yP87%osuu`Hh`+iI(4Xf?vJ90zOU
z3a9itL4u_LvVDdHPYm%x4>Sy``yC*K%dC)xVi<pORKJ`%{H>O9EU5H_id_ONA#1H?
zCFW9;t!9~O!!^jbK-PBx`+~%tEp-LN9bBC6O^!Gz6Ayk@D_7Al@1j#P;7iWqLTx*%
zoztOSxAtD^xBl}}A4c`sOyFOfg*w6+%-(z+uI1_e9(ndC^aZYVbxuN9_zm8ojkc8Y
z<0&}p5j%4%CICVa%U|qmJTZS)53cUxEpPYJLfdGldR<3U^6)G1Qr|K&SnR(F)_-+~
zEdP7`MUT3E@&+rS|7`g!&dgjzn#m-)48g?Q1up2+9Bg+;cAPOyWLaxY)6qpKoR6ED
z7N>f7VQEh=hIPxjZmr*kxD%>PFr^3QhaIQ4mtZl^A4&QQsI=^WXu?SF$bC``wE^lw
zQlbY@;{S)Se+sVc`?_%9*v5`++qP}nws&@H+jg>I+qP}&*v`($TXm|wuj>3R{#R>W
zty#0?tkL`EZM3Hq-j+>PRZZp@%wW(4jMoaxEOP4V|6?yp4xKg7<M(*LX#3=wxjcRQ
z^nM#$j7snAl|o4fFWU>+6d_eN|K(LnoLELH6}VWx3Gv<ZcrMedPiw)!0<$R&49`h{
zTp~0D9sv$NVQ&7WOG=mrt|c9eBt#K3eVNAK<_y9+9R$aSyk(qNObRS6Y(m_C7BJ7J
zXXVXj5Kak`x1ZnJY#5LNZWg?_CZnwT^^PShMeK+tfF)K&j(ba{$pz2dk1t#B1s)@r
zd9<ka?{S0*l7==+VPO6>7<430pHZ<xAgA>Nl>S0LH~4dryt~($Wu7;LDHQYGcI)hg
zG6JD?(~Y51N2}&BX;v&Or_cqSoNvXSQ1Q0}SY%v)3C;9BwY|0PU!1xqry}q1w)_?L
z@na*A(&1wTg_h!!N3Nha^d2Y!b%wIR(wj^$)AArf4|P^FS|sg&&HI=n84)j<&Y*Rw
z?qRm&Lr)r+MfY>PdH|U=_IpwMdUH2Nl`Gjzu?Dryc9wl^h=`Txu_M<?FVq1gkrwgX
zvU(cIN5%u1_#%48Zj&BpW~>r=>FWs*62JE90Wq-?t$gRNS*hfaZ>?33F;9f`3?p_t
z^|^+Tu5Z{d^YR$)8sMhFmZvp11;>49yR3oK#q|Nl($dtO*82Bo=eIglKsW4{oJBR?
z>BOKczn9}z$I87D?;6Q|KRyBCvmx)==$+>lzF2|zy;afy6A4=K2t7DBepFqSilyL}
z;;fY=Ee-r`Wyi>HGlq4j^J&E|bVqUym&|@Lw0dPJyLc=25OM2Eb-|SoE|{%fISq6*
zqf|{eTdUFhY?U6F0FQ+5?N){v=fq*OSwo+st9)YU)u3khIL>2n>$sGVdvnEqzHN~_
zeGOGwr(t*5(n71Mh7!%#T%}?qPvt6{MLH!>+TAYjC~e?V2;wPXdITr8OK{_7?9)3B
zSpyegZ#V}FzG^vpA6BY9F42)La3g%wDRzwn(YXBrC{O`;zw|V4y@#~^s;o<`LWYVx
z<Gwq3F@Av6m0>b4&a`z$h0>Wq#W6QoNBm`)2s$1(f$!MfnHa&FVP-n4Uf7^1dWdH`
zS6_iVVw2uftuN+d2_5XXY_TiT&=Xb`2Tf&3Eb2Wpi=@hk9tRrQ!LnW*)?S{#jc><t
zwQ?z-AOH$D9#Zxk+wcLN9N%6ihL7sz&}d5pmj62e45jrrkvqJQu1dTmP1pJ1+n|V9
zyV@EfI<b8tsm)ncIU)j;2Wwo7N7&WRGH(*t@4_M;RtDa;2)1=e_6Z%q+kax{rHFit
zLwfb*8Huy;_S=yNw6eL`B8@ORC$T$ibq8z%1&O&e$H#dsq!j2cPq7^==b>ZvW?Je4
z#<ZjRCA_`yV*YZS!_B{aV@kSCL1Y&ir6s?pr41ZR?=)sXE)1bI4RcD}>V(}+`_m>q
zi@+5uiS{qI`(-_<v=^L+zZEgmw2E`o`FQAS`g?Ss`!!)sDw@8X4=|i|&6D2oK@UV1
zo|;EAwVeF_?5NDSI^b+QI5+e7DWjmA)5wA;dg5Pr!aoalF|L^~PYb#>C*|3Co{oBc
zO|HbGIQ(t*m{%5O({G+bNBQ`rsOI8>#L#_*@>}oRx;zZR;#r!%@_c3Q1E}HYD8qjS
zwS2j>c?Kb-TW0*RcQWoa{}bmb`dZ*W)@YXhz{xYQ|7<n>N1XhyrcB)bP4RAO=aeAa
zv3kge<q+jMgnTIHhI}?o1UK)s%kcrEH!GAIyE_G;$z-*Meqdq@zxaqPdb|+DiaNu^
z2^h}iAHv>-vup;5Q;}wpqfag*&m>DjDU(DW)Le07m_L}my*QD#^_i%>c=MP>4$L*^
z?9yt?HFW3O>~G)rt2R2Imk?VTV}JgEidZE2WMjNwsJJgdyc(l|b)G7mR>;&JWTuj{
z_V%rAJPk)THN(zhIy!@srm@l{p`-o<3-C+p1kr7pHbw#{1+x6WjNPPukkiBpSD)Sv
zozqje8|DTib{5YTL=Wf@xFcDV6r`Z`qOhFc{v5&WhF$AACa__{uI18sNczxDp&G+I
z0mG|R!Wh~(%H7({6@^5cY6pj*;~)^KKPYYunnB5XL8hwfD$}?A;=aTBN0}tX29u$d
zBIdH@%)`P*Oa;e}t$O|?Jtr(_l@QqD%$$ZVg5tq;y{`7wZRZRW_FcC2USs2$$v-4`
zFcsNkwhh*~vV30Y5i>Z8c#Ry2I<5kCgI9D|_#AZ%*g|J-GmQkIvQd1zz2T@l`#@a1
zHI%GdK2JB(vybBEv}??h4yXea@&iKVhR9GJNEeMCMdF0PlQIJ>|HYx%6!`7p@Gw9h
z89fC6!?-)0>b*5LnIV(Pu3_O-qvLCDTAumU);gUE`R2x^sU9tO5$xH?FW)SYy9B9_
zY8d+dXAK;&izduc*{SXs)tA8uLzA2~ha>i0g;LCvl5Qq)c8U(!IjWbiH5cHPt{g~}
zNbN>l;;kP@#GqN=A?HBziaIv6iTaC(#Sp8~#JfCJkS`eWv!cp4j;7fD!aP2sRtZWp
za<{QXY(Inc9I2D!P+2J1v^-3VL=bq&cV1_y)7tItDppq?St=CvP0OYVj|hR5>2^@<
z$s?Fco>$y@Jdt*p_-O_4=LNe4F(!s^Pm>R+5eJfHsuRJ>vOL_f{7~o0>aTy4mcy);
zdg{#R21TjFGyj?Z`ntK4u)O<2eT@=ZfthX>b7xCs4LYoirlfF``ad~3tuMgi>DXeX
zjzlsj`SlgWU;CoMssrz+hTIc{G~&ex=~&AVk_7t}p@GX~*#!6YhxJ^YRxGawY8kHR
zgmxEQ=Z;#nvww+5_hgP#H{f&kq9vkH5|~#Aj6{_*M!d;(ax{d`lIu)*0v9A^q}h!6
zOwa4zlNH#~Lq5zm>});c>Cz+r96nA>b`4rFtMd~@{F=Ai^f-MA>ltd!1mP^`Q8HX~
z?U){YQ<zCv5UdnjZe2FRjDEiZI^)+h3IR=*Wjg-jD9QC|qT|bzTZfB9Ai)KhJmHpn
z(54RzNnFmgZE0nx#M&fXE*8aZbA2^8LJ9Ksg`2P-11U++Co(o~TJN!0Bagr5ektxD
zN`M|B2drRE8`BvjL45LlAR$D(jxQ#iH5hg;P3$7dzyOwc3kxc!nrN#p4#m?_`0eNh
z6&E4{Ef5+6{swhUX|Fp^O(Y8U$zWva3}}^<jr{1;V5duAGW0AA7Y!me8|q}UwTfWz
zI)><*Hg&YakrL1LPiW2ATV>aSGgTXuzjCNEMX~~)q<8fD$9hU=#bH`210xhzr)dR8
zxWX&>U^V6}OYGe=EaoDW%qFY5fAq_98pEtbIFh{J2ZRL~3=(<&>zalBUNw%xt)Byi
zA|@+)M6UI}m#TTZ3`ZLr9do^vCJQEN&}~XTIr}bBw25sRV<F#^(^k^0A&PIAfgj3<
zl^>j?RI>DU5z<-bw$@dC^u4Jz);=JR&1%5<2L#*?=`QRMM(dGAQu4leWYf<A5wpYw
zlgueV(8A~+oYNZK2ipFIvLGXgDZOO1`CDd_n`Lvqh7h=aLBZB#^*O|J2kXl@g$`Cg
zs#Vg-CDP}8eVOjVM*?$dq%w?ux`G@;yr{k@>!A@NjEDFKQ9R@g#mPpLj!vJ-Sl}8n
zP19Qf+kyMsad!eIo>08!%!S<MZ_F>B)N>je=lbOMjV-62IVk>}V*5xpHLc7Xt#h9*
z`oi8d{zcIf*3}yUVu#G`&~SX?IhyDQJ(6`0zHa>*>)9p*+FpKjHi-W_5<3U`-tj(R
zR<PvO<d9e=u18;xx&34FcHALh`*?Yt0;0{F3UiTAhA0I7$w0k1*EX>J+Q3Ma<NUuW
z)qfZynOQg)|JO=2t+nlN_+yZqH7ICRab!<3Kp|`5%U^Y`rCR{m=2rO2Ki85Cl|rVG
znB<1I`hQ1FsiGnuygDtB35#EuaA)8wezTO$yNMQkXMg!PJ)h7Oc_k<D9$Pi7i4w=4
zWS~$dS=Pq_jHNia@+hccFmCAmKhGy!$PR}&4Dx2Zgb!=)7C%4kKfew>I2=ka7o{l+
zg!`3JRV$}5S<VsSFbi<KS&g4nQ<2FZEbBu`1Tco*bjwMv9c(FL{IA_&L~bJG=5txl
zF`g!B-i20u^uCH*9{1x&rIz$1M85H|J&vKDhFi)%5MBWGj{wL(V4F7A#iq8C(smL0
zcL>PmoYPu9jk_Lw;OS-}U=OaHaB|Y1l^shxcRwBa7r*r-Jn0+2r9+*CgB7=$OH2;l
z7GqRaiGHfuI)P(dwTAmzL!8L=8T-ixMyBEtn90w<R%&G<)&9lc@MJHw$Bu6LKcdoB
zpR){gWWm{UMljn&Faf-cRf3gX{99~9RK8=f38K<nHwUFnwDq2w)ln0!Ub^-kYVFh?
z=YLcP#aqWHS8M*zoH#XtrmQ6xP@;E?s&%&5YG1P32chCcb*2Dd6l`OX)IcMtoTx5W
zYIc4-RAd%QLt%`uVG6&j=O@irnJr&dvAL(&pLKN6S=Uk<Rz6^7Qo6SBhD@e5FJ7-<
zmU_EN19{=62o`a5At0Ej!x{uYKh@Af)^x%ywynQBn6ntLe|TS=Y!vRcx_JmU=MUJD
zLX1mMXXEy4_L(b!?tpUVKrEx%e6}syL0=w(bu&ZaZ^b_I;@}74W$6AAuUx3yF8A8?
zINngQE)B!Cg4fZ}Gr(X`B`T1OdBsSEI{Ac5jQ<<q0T#5vT=3Q1w~qDO9lmAI+&2bM
zshx(VNGoG#xCZb{iDcy$oWf>7Y}}_ZlAiEv=%kvU8WC^6==^QFUyrAj+@EOc>Fc-x
z%?9DTmsl(tI@IqEn>plgwaByzc`?7ov)w!?_ezXX$jNKY$SI9lyodgKpdJe0AJG#n
zzWi;OB>QXJ0@m?-w){u`J{QvCB&h~&*J~3S{=Wb!$Jg4;r|E;s)<z*rkYHFj02Cxa
z{l>-_abLy>byPa3a;cJzXjWsM{-h~Vq{w`rk_rOkseapv!gqk&c9V`CmVfW|tYbM%
zZR<vxB?T*^AdLQ?s9=CC7>Q#?y)qYfV!%p>5t4!~n3B^3foO6+do4XKV%m+shrcR2
zuG{R81VEb7P`GZtV)Yen627H_-*A-o2M@5kGMmk4SZhrV9<CiagoxP1F$)*uDA~xn
z6?{z=QB~;D{ZD+gM`P=-W}SBgLxHd|uqBX$K6h_2*k_#FcmG4-zy=JtgJA=5O(t(P
zcg&fXd`pkNKWB_giFV87@o&)_?E>0X28^#aJ3J*La~teKhL1=0POWWr062EmE;1+k
z11bU@>B7_TFDNYk@$#3n3?VA_H($L!c_&YP#S|z~U~Xd+;o2-WYh9-gOcs40qezzg
z?Dk)z_gF_vk$*S+R>M5OSljS%t?!KfXhqkj!xh5cWg`7~!1D0C;)PT+qpvph+f26U
z>U{c~ggGuZ2IWB*<CewO3Zt6jhLivKG4(r9t#=vN^+E!sQ6~{S%t#@5a|Xt>7^@I}
z7>-%{wlgrcurtv+cZ1{_fW)yfe^ukRo`A!LS7^L#3gi8A8cPMN&xXj6BCosA4++b=
z?eZgt#BzeB(q-waHhVg;XBc2q!ANQb$kW9!-44*pdT>bOb4xA(t5`6Q2QXcUYoxU1
z{{v8#wk`L3+cQq6V1XNEMZ+h29Yl;d5S~-52fav!ik0Dxa!!2m%>i-L;t>yn!3#E2
z89o-}2b>PsM_~@!!b9}Ccy6(Zc)^FR^5KuU3@i~Iee>h+iMnJ6rH>Qx>;rL>EV{z-
z8z;(-Y6$fB^RLrHk!}cUs4(aCVy_D(QpB!kvw0*XfBznkZQy;m=xs~yem9+w9DtDi
zzc^$HgMY9A%l(IyU~}xe*sqWX+s;VQgpkft{n-I_rNZ2N(8Zr8a`7($W3#yU9nvCr
z-#0({loDHb)*6~=QM>3UG@UoqndfK*#f}sI$&Q#fn&;V{$Dd;`i1~`hsq$-98xnB=
ztHo<a>5~+(Qbg+@23VK%LPNSJSx!hyMToD9z-)mkfou4Ryu35u3S|3OJk!J4FW%;X
zRl-AGn50LzxB~tO`5}g2*%o!j5i~G;J18qXqH`e0N%PP)2y*>oX(18>CMv;W0AG4O
zA%>Rt6pJFuN#m#i!>LqMXA*jyo<RH&-7`$1gF=$@aXiL9LOMr|I<{t`%1$Ks8G{`Z
z@m|tO@FxHAltnU8`L~0}^BMEFhzR}=E@;3orXZ=d*<|e;+MZ-9Rti%q@L@<t$eOi{
znhubI4mC!BK0~}^BKAdW{GovY1JuC<6?vHP=UmEpEUTvMtfQ3V<J$CwlCqa(YItD-
zoqhnf#WJg<o(7*>TFxHQ%8cxu+i5+?>Aip2_>~)3F7LCF;@gkj{~-~14+lp}>!s5Z
zexae)=I~(qWo@@@v9GY%=a<P7v_nCdXhx+pVYG_8=K5M&-uY!%Y7);WFZJj-tmh<}
z#_|5+-0cN*G(ouppF06-6Ve@P?8R+k0&fvnH-OsWr{9Aynu;=-ZDD1XMI4F81L^SO
z1Ib)!=6!7MJ8=oC1DwAAN6mIVqtQ{xV@`o<UZ_Mq4{N$NE)zuC`i;-{X)DDI`MpI9
z!#~CuBa)L#zlIaK%rGOgK{bH_$9}LvSAn0D;p(;MX+}5!mfIV>d-Vp2vgL61%?)aO
z-G4L8^mop*JvJ~72q15_1&`(_Odin#VY0L0uDqK(jVCcnDG&3-ES}aasnd+SJs!1^
zLp@PRmADFUIi+Q+_O2kQ`?^un%B|n7@+naQW2E6sj*F-7BuDus9O`dVKDgG4J6FG_
zo26#03m)uMQf#E$r^c2u{B5^Rbw3t7)rZ%#;lUwy`Rrf#YT%08g9(mro7kBMwdy}8
zS&!IuQ`y|PWr2vc!ab|Je=i*+vJkHc!2l>zWvBGBN(;>u?3{#SI4UB=8aW05xklk^
z$$zt4kTzsNou>D`ERuq(AGFv$|I^4Yc2hmS)5s8fcsWljFJ=sRqyyU*05*%6viOhY
zjP<|pL|h#Iv$u9yQ|HG8^y~jzK>tUtU?3(dyaHKY+BP=g)Uk;6lYAhYwKNYtV@_dp
zf*8Jb*cHK7{*`h!<UTKEKs&R4m3_0tpDCUdF8<B=@dI;yQiup=osX>g>Bxd3O$(vq
zOq(>t2#iy5{B=Jco)n_b$EinHFfANI^&X#3Ili@Zcd`u}r4EC$4J5@5ho`TeHaSkM
zP#Ea_hblbm$2TW0UT*BW9wl^5|8+5?<tUZmk#Qd@i$6HDM&M_4c>M)T4zD{{zyV5-
zqk;*OGZdtoOy?W;S)2<<t7nN*sRoqM-Nc&sKaNYZW7(O3odNrCXl+#STbv9|siF|B
zWL;UB!m6o-km_&okZS>>>AALhk*3-{VPCN}d5#{g{e9X6`;!qsc$fbgumU#)AHR~~
zvNs2?5^Pm#tZ$;Ou}VMpu1ofaKa-fcIO4Z7$<}o*dUFgYr4KJ|a)Bii6N1f%2)$O1
z!`}7@00R79Yl!LzfU&A|CiUNSFatTsf=BzyAhrKyqXNnb;5)k}FN`TW{*itvi#%E|
zL=>C!;W53&1uG+A|E#lK1*&P&wW<rCpGNe|fLv58<#^R^0(f6{z|vb<dqAsyV8HX9
z_jh#CUjxjhQ#8zX$$-C;aygy2>d*<yA4!RA_8MYt3*7zC3WM@UdTrRaxNaep->Vhm
z=x2nos6{jX1OqF{;xL|9Ywvi4$E3bu&z(Oo7LNDa>hnsk&byek)NTyqvh9>Fhvm&^
zK~0J^Je)cRQJGQ?5jEt)Hsg9s!h=b*@ex}E26WAb5`b^F9w^YEOCtg!u<bvv?qkfH
zfL$4+4vl!;Dh?VXoMHu-c$|zpa)&?6GZ4)-_bvf$Qzo$i7WUcFg$sMnG`ipA4cXAQ
z!YG&RyQ!0Q<-3)d>Ri-4&hO1_2Qyge=zejDMp--|Iu<f_oA@-D?cOk`auJeSHP3?M
z@WPybeygs0Y<wEtopx>Tcy;oJ04K#m2ZmV?#1{39@+B7Awm^n=?e$G{bFV2V4QDtA
zKHBX9Imjm?+_dr4V5}l_v%<b;ruD_Jq+vGW271VwaCG+vI&1?7`cAb`U<8R?XDlLu
z2q5O@@IG$`2N4Gy`f53qbVJJoK)%c2Rj^zW*U5tf*j`QG$Y%_r-e{7Kmh0?ZvTE{n
z(WfhcIP!B{f~IKpQ^Y{-wW6R=upnP7Jnc4BA%klzS8Bi=wH_PibX(*ET)nJr8c^Sm
zXpvvZ>kHlV)i^sC!tB<kX1GygJ+6ARo*;n85B&3Fum(p|%3<YzpXTB(Nb-aB*N3i?
zg+0mIe1FNS4td4x?$^=Cci<K#33Wopf?D>pE?+|CYg3>bQISwJ)fjd<F|zdVSlo&9
zgPEQC3GAKCNIAcBQacMQv%AyZ$pa-!D+ngrTq$D*qjGG3WeAcG5wqbW9eP^Zc*auh
z`o^klB=?~%Pn+y08&DdVCS8xUEB2cnv~Kg?tw}Wj6qVVlPzrPqkQSqvU}>hdj=K$l
zi6s0uGh=sj(Vpx%_Y{H%?~a3Fk3rwz@&~~Qfpq5!gUd6m4b+YStz1c_bFwF2h{v4(
zTb@<Dl^&(&#&lqciwYoidav?9l5P8?EQ)AKPWrWEj;bUKDsfDO0lwDV4k+r(T#_Y`
z`byo8ttQJeRrk~A?-XSOCj+%L8<yK6jcFB>jDR~{GAVMsO?I(^z_N)3=m(}AIhC~@
zc2+w|0Sr!~`q2V9i3+>rsU|N9{f{@`Nrvcnouazope%PmVA2H?7zVV{q!eLE6JhL1
zKerA@7*8iqDd-o$mCSVOEIGGVYPXDFiVhoM&3RcyyhYft0^A3W_fNh;lxn%rVjo&l
z*4vfGiVPe~qok{*mI@$ho|TZlCeP^|{X7;k4@AsZB-+CNG;<XK@L*qIUO5x9k%UeV
zb;JX7dg(HIu6)04zrA_(1;NSmGb7FU0yihkK|j#rT8}5xVKB90jYf&}Nl@5k)()=J
z+tp!h?&|@VPMZgP>`Y^|154Z`pTXfip5?0YmPI-l;yku;St-;#c9r)kOeiIcv*{?o
zF6c*CN5>);BeKh}CVR!TX45@ABI>p9UJcBv>9Tl_E$ehE?3x||RGFp4I8*p*9w(md
zu>x#YekK_xEzPxDE&GETUB0#yG(B)0b|27xz3#|rJyQH51c$MpyOM<L&ZF$2o}`Aq
zm|v<!DFUN4?Dn6Hxc&fFcG&Q|_&m}^KXg7f@`x&8lh-zh{)vc|G+^<Fg;_@ms>@M^
zIy=6to7|gZGH3WGV0D{;zyLdChEgjVz8k0^4_5*QS^d~$n}K|JGkU)rAMu7FzR7d-
z#p8Qj?alvWmwgAp;}akM%i`=IuJ_D`jS(DwO2IN~6Yxz$Kmaf3Cq81r034tO+Y?)S
z4#yLiQJ<N?XY?3QJed`f)dUQ00ER)#q{RH&K$u7H?SshUw4E~mc?246Z|3>8=}{7P
zE)%gbM&ri_1Q+umQ=(zCtTqsbJIIHFC5QdC3j9C;O+a7oKr<p7KdZHSlS#&*j&yqb
z3erMbQJB6)^+}9`m9Sm}WzS`$v(}8I#*U9M`=Rc|3J_DMorZnO;IOU;ltsRsk=*m{
zPrdsFI)E`n|F2HqKlsGVKk*R%>zHO+TgTyW1k3NG)^L}!FLNqKgcZKd$~_8Vt!aIY
zQqc!_4#ljZKvJS`g)Q>A^<^pqEv(dc+disCa_Khf<~J7$yBq*D^K<`bvoWc5hbfZ`
zYn*j;e4QDYTS*y}TFtpdlAb&fiaCc$j-uE9V^G2W{r0)@Z!`_}+x0YSN)pS*o%#FS
z<wxy~)VxIzWdU&r+NhXC>Pav}d$K8&1~t}ihuS2QTK+EA)t?&}C}0jP1d|CPO<ZdH
z1lA_LoeTkBYXluo%vp^p<XglVc;D_ywB6a61Ke`=MWB1BwBL|EXES8saLdFl{K+HO
zD_*vU9qz|GpfD%BaoWhqC$%NMfAYeF+?h@4)J<PcL<J#R{q@)qSZV#fOjAX4I7uSi
zDpQ{9>it5gdSRWZgpUN^9-NC(5ap$~m)^7f%Bpa`?Vr9|zie7?ieVFHy!m@fbd~F0
zqwlMYw{CvgQ|=FNeI71q8F0q5i0LC}wm)@RgvN~o;23>mj)^6Bx*xH>?J`sp&XDrf
zAw|1b=rb=SQnE>Q`S)n929Vg_+x1Xo8h83AnxqtbXQVwQ);Dma60*PXON4e2Cr3LE
zWc+rbofWyg&efBO9vqeB%PfoiGXD<T!uSqq|K{pgvG?m9?-i}LJr`5rx6BcIF20+R
zGcTx@sMf9bdHO!?t@Q33m)V43|5(g5Y9aAFJc|#JtGBpMXOadoEey8&dcam>5%h{y
z+##mPfl5V=d4do2)0uks0syH`y%uY=L$cSWE^SiDPz(>$a(W(yci1zzx<_rNCy18v
zqM&BG5cof{9_A}3IeqzE<Y07Ao|qwo$SOj0tmxNQ<@?M5Z1NXXW%pc!hE2C$Jz6gs
z9u?M1yAk)pclsGv$%G-sJAGYLxQ&z*?i&DrD-#&<=?MmJhy8-;Q?Et<jY7PJ(V~5B
zS^%;A#VGEmiU`=PY4NDJz;&~{6K8;B!GWH71SBc+2-K-al$|v{u`HAbCIJ-5%5+DX
zbU#=BuqvMJ6idh<#LLG9Ev!(AR+iZUFVSS1FpV<PXxKuvN|4nf#^7NN{XC&Zyye_D
z8TL-~y+Vk=TXNb5@!aa9pqqyZ@1>I~yOD80%ai|=cN3l}q75C~y9syQ%M$GD<bnrj
zVn%K)yaZo{;qHm~E@zUcH>2R$I0~|3lpSk>l+3R}qJ<V`!RrZln$=w9)|8W9nQHoc
z>08(tMGcx&56DG6;<u`XDCwSvclR^8PeMgw4J6`Vr>0i3JXB<f6N>aQICo}D4%HfH
zu+1@H`gVtDwRXHs%`v0;cAcOejn02nZfZ3|^xpt?3Q<phbwKSi<V+_7U<*x+P*bDF
z+xwkb#B!voCxGmUB;J$Zz-%@uoeH8c`6J_gw2}(YS~t|pWLy0b<A00qvNKQ{k9DXt
zU&9`0i95qZt!O@A^H^%DUNc^BNd318rq`icrps|(W+G8fqu}4*J2fvHH>i3?a<eW?
zuli-?Q)#1=Xg!Kub#dIXl?LG#t|16pN(hV-_d`C*=q&=Q8WV$ir?1bczxT1%7|-`*
zbE99p)T{R@+S>tZd@V8dH}5RXw<EefK*YDJeJhm0(I;zG<Hb1uB-Dy+LkxLHvDLGq
zVpT2tfg15`z2}~&eWG^dxT76btKPM+i5$Cx=`dRoYi~NQm7j+5Q1CciWef;%4qC4B
zqA02bwEQ1BD!F)A|8>WfbcA#UiDIx4Y*T{VKJ(;;tM{H8p;;znv94hYH1&K_t+F^6
zp<3J~hl#e4$Vg|<@LRc4%OCD(0qrwEK2eCqr58@DrU^zxpo3x<!t2@q*<!vcr$R=?
zWD#e=&LicRb*`k_h~bb5HS!gvML!nK()@n;A`mS#h4u{1Jud5{KFokeEI0WVX`uYr
zc1aN4$zlmlTF-a3e$Y!iFoma}msa)>zbdA87Cu9?W*RZ(c`0r@6QdXvx?%ezHACU)
zITc%47%Cu#U8N`{4{FJyxaIx}U91WL6Bw@84IpvH_TQWb@J<kCtTyfS^L@p^6Y_gx
zC6DoY$RvP{zOD=DV?p3>22^rrJJ{nyXZ{nqAS}>Rl5gaoHu2$nU|{eq4cb`u>huQ<
z-~A~jIyU)A+5OwZcAV3wI99E_5<DD`^rMvo`^yBaHGm?yK+Y#7Ae6)?+k$ghP$nYW
z8m~n22u^4FtQL2>t6MyjXGk*mQZJZK`~}y;8}XhB(gv%8U9f-vp|y$_Zmr%jB!;uN
z%(w1tJ0VmOanx#+{ZGBOniJd@7Vt0X)V&j0hkfMJUm$8xTLG;h{>QVk|E_#{*V8a>
z5P~0kkYH#g{n%Q2x12<Q30qJ?i|DguyeW*u3tANUhX3G18ob!RUHut1#r$l@IdfJ9
z_)|?hRRTcg(pS}j&x-}{_CC@od*B&|uV=I7BbLrZ0&dl?X`Uo|<MYhysq1c>aPqQl
zbxR>!9?r~ygypSKvy2Z(){YodV2?AGg8+<~k~0TiHxi~Zm%H_}X0Bu|1lY?OoyH$O
zO8K@`)GBCd>>%`pwWK2C?N))mb<%KY(!!ardy`F~^^SgzUR(&s@-=P43xDkOR{?u~
z1Pe5$8Qnla7>`Rd6dVIM0VFKkB7q$~UUkZL7StaIcg;X~RO`YY69s=iL@q*UpJAnR
zMb7!bWd6Wx<~Chj{5O{>=4ybuQNf^XxHe;~GA9KKkDk`1IvL)+`IcT^wZfYL=F-?(
z5$fi@43yY6XA+)fL2YuX&?xWiz7gL8^2VTe=ci9Eye?vK7gVvNjYc#j;bHJRaVX{@
z5U$#dq<)83L+)6UFkrNJTt7|mVc%LM8X}$VZh|;4672?k($_Qe*zSy*@&QHsH;urv
zhObj5`M!cu6V;j1sN?dKIwhRH39Y`D(Z>ESP_uDZ(r#(83<4^*60>OU;}$Snyee?T
zS_*h$@`wIYP`i}j{-9SW7kK~QR)~}3{~wK_>g8xg$e>_st>S72!yr${#K;K4Aa3RC
z;!4QL`Ex02W^dtYNyx;-@;`gQhqX6je-tgh7Y(E_nfp1ksMX4RWW&UK4q*2HLvxVJ
zQYJ+FR_4XE7QTx~<NFt`lAfW4k%k&|62VS6sS@IgB_c0L?nb9j5q6>5WB5Zvu^($_
zA~de!G&u_Xs9_?SFGbSup-H3$i-YtD^r$iEpca}4JsBD^jBqF_);;4X8e9mjoKt9F
zklljtk}z{JGqu97;)sAeGB`Bqh@NqnJ1J)>5`59Gn<aypk&o!gy~WXDj~^S6$JlFd
zs*CxdOTSL}5{4aXZk>#yuA(5LZXh&$l@xQplH46-d6u#))a|gCY&aE;9||%bhG8li
z6NIdo4TyRlES-5+U`t$5gnZEf2oxK2Y$RDAFg=r{8ZEhah~6#bAl9MN(z|AMgs|n0
z?}|PRx=1Pl+k}Y<3Di(rsNUHdk{_(Cw46~&92hwKft)Oujy^LP7#rCZ#y*M!tuQY<
zsJ>KWsQ?I_m=|TGHQwP=iI6m>4@5|B9aJSCuX<}v{RGT2xNv*|LD}Tig!nc|2#MD?
z37x$F?j8u}4q^Rvj1wG<nYSqOAi=gE5CoGH-xd131M;=LKEMer18PkJg)jj~0Lv2E
zHw8oly3QyLNQ7A|2N_#k6k5_-qA4EU6Fe+aOrz-!oEPM5KobXH-&~zKp7}93t~gnc
zHw4mHxA#idKqF-Wn-2^@{II|kDol$gIL*oU@p19{ylxQrnMbYM)6Ms%`CcZ7PG8&F
z&USZ}@5?sEbYdg$AG8KwS?dgaIt&~aFRMx#R+weB8(k{0v6NAorm%^iUYgyb`+1aZ
zXpP55+Pw(@{Qz++FrOyc%sUDh)2sm%_%o8@af>C3d&d%21vxA^z;7)&3y1TVcJ^fk
z0sV|~yDu-3+(NI9(K2krQi5}-Cs5)lj4U?sq>;N~Z@t>I&E;x+4p0@B3xlWWtH+OH
z3zr&p<iE0Hy&`Oj=sO|XUsauL69}JYDpCl*8xT$yJH5i^ftw(Xh+FKpW&K>3O>4k(
z?*;k8bBNlb<R*_asq+?)b#7@geu{`kPO#^C0EE!YPQ}xH#kA80ms*|{P<!BJNUNx%
z&bgZaJ`l@s4?!7wO3R4Yheop?r`Ea{`P|9bL0fE-BbjzsWnTE=A-wp7vnlJ4PT1kO
zKQ3or*?+75RB1#FqW}On%jI@T^Pb<^`R%*GBfpiq*L!wZSHHXTK0TEEdUdy#s!FL}
z0Phv=?eF`UnXq<Sa7qCJ#ER)uT52@W4B?V_jqt@+-pht>BdXt;J$l%tdlz}X4d^+_
zt6T14h#yPLeJk`S%|sNQgp)U-cT@u1l$L$Oc@+Jd`m2^j2<32>t0pH${GY5_+V$uw
zIhJc{{a0pJ2{^J8E*z@P2d_>JpSyGDPM5t|^mMnDz8zqeue`f!R?*F<4cN@mrve_q
zg;0dE#lNR-F%#Vsi`R7|O$k7W6>BVlqWm?$Hbfc(_@gg3c+Z$;R)pfCxf37ZdU#wm
z_w32H?bSnLX)+W9V|0QX?vc8~jbx9P%bLB9fJlAi<KzSBohRF|JuH7oYZRgju)>mM
zb9W$SZ@jyT5NwzLEu(9aAwpE*mf1{-HwQL4Yfd7d8DEXNC=}tZw3N~FQ<kUG%!1^0
z631*S?q!S^TFzQmK{lx~?24Tn`*3-%;pUa^w?*wkVFTp4_#<=YCq3ryVUk_P$**bi
z-J1vmECnL?GH==x$LCQnP<>l+zfp9E>crys463FqFsS7&ou3Le02j+09a=Lj)Y7HX
zIE{ZhLWiGEehnNS`u<k(&z%w2UVc6_|NUay4gLe9RKBgY*MZ2ZuKWHdR1}|tDiU_U
zu&;k!+Rp69y`E7`40G08NWVzwGi7q}G|mzNhk*2D;`c1Sb~Wu5w)f~?D835TQk`&`
ze*x!lWvvo1t=&FMEpP3Z_4&ipsd8+!c;D9*by2P3=I^&gSex}cT4%R;^x-<fd1Ajw
zS(i*tYsE<W2BztZ^Gibg$6c|dT{P0j;T7DmMTE_1e)bmKB8E>qznF14<sS^q;A5DO
zq}IeM0ec>zJeP9(m|^0=Ga9W}%Sg2LJ?(%nZF-7mJ-?1SXH1p8e`FQ-EMIMdy@t24
z;<m^gMfJJIKJ^CQo@WI8YC2EC>JZfcdqo8qx(a%wh`k6+0@N%4WL(44wM0K;=q!`A
zRd&T1{_=5D>cnw7c3#gQg!Xsdy#Er+PE?<kho7BDnZnFhcw(cIFuBn!@#6~rB^?)x
zGKMEU<YZ_iCe<d7yZw2;TKC3q?@;F20uG^G#=-z1Y>dHciG_7&Zarl|<vxHY^bj(v
z!GrL(B>DBGSarX$l*E<r{TcV2Z_Jv<+H0XjD@t2dAg}8j+#@e|pC-eZgFK}%VdRL(
zxq^2LnUmz5SCj-T&gydIzi#nEi>G1Vjy3rk*N5LK7uL?(w3xjc)h$t2_S?E!TL6x#
z%@-ZFZhHpOCk$MYm(mt&{-)Mv5Q+&D9vvc-?KytTID*&~yX#{WN~5Ec*S4@C6c>!*
zMfMd^_g@%J(J{@rdx&|bk9hqrb~@K~n24L>uc(<RUJC!;9)OFD=|4<ctSl`5>p0{}
z<Nuf2uq+NGkDo-&QejedWtzC6l)E4ckw_VVicc-I{`B~IrDrIVSsXVczu~#i6~70v
z_Vo<ZM)1Ur|CB4=lc#V^AhkbCir~37y_yTfO9m=ja(_Z|(@e1NH|tH|B@Q#jrwTL9
z5H|p!^K36{RA3hI$1@2iXBc-a1v#IMGvE{O<b^L05bzCDVc1Fx8a!n%z-?WiXdC0j
z$1y=YpNq+766PAK809Spk8CT`6itur_Cg?(Z<S&*OUUEwqkV=O6I|=IHyMLU+jlfk
zQg5HxF`qoPj`68wo-8JejJYg*a{W~~=Fk#>HS}BEUQrpKR7d-umwF^Me|8R6hLDT9
zzo$`oHYtwr+6;;)+{-JaA?25YLoTD50JZaL(4(r!zEVQt9m$#4D-GbLWpsqvy>&2l
zt$^`CP6wM@Om|{>3}!!&555$Q$%9YkG&R$R(iavjTv#k>>Vk?60K=SaxJOH7^bbm;
z-nPwJXQnzwLA8K)5@67f3YBryyK#XEN#W4h&TkVc*&3a;+Go9<XtH!9uTgpgP%1?E
zZ~|mt+c8A3s0TtOO%6Y+^&}I`N`3k$^PN~a=D}!EhKMvKaC)m^JeHU47Im(<vlV`4
zxvqi#FjsligJG`e9wHK=1pD8|ObQ5<`nsv_R+0?tqZV?6rPG19;`R%`?6D<dxA_J#
zE>kKa0Fe=@C{Z(DcCH;^&C{SL!%jLy{%x;4yDc0hz)&WhF2w097BB@4)3}L<>Ritu
zES%4Htdu5gR{UEjVKek0X!|6_mL7y$ErfqDpNjs{<}dmGYtdaov`0e2LLMi^K)Wk}
z7}f=9jg6#1*He49wcT%qf6YSF_}U<@M9GO!)p}cpUz7+#Vo>^E%D^!_7|ED?x;d#e
z(_<C@NsHE4;dA&US-3<w<iCx5L5akp_m|fyv9jkElP)i9w7pB3t~#Lx8QbnIanS*S
z`J+nqF!%P+doLxhEh6AS-f>`XKW`|j{lYqbB9(FA8^>S`HuHYw<;6QTgWLbMg7*)*
zV-jom*Jju%MgyVI7CQ5eN84Q%zoG3X+@%KPKbLdrM`a3eER;bwKiNjca~>lx*B!e%
zM#2KqHG8~u@?@3OseT~m1XYTb^}t%xoGZK2QM(H-IhHB3^nL4cd=;O1)N3MRBS~WV
zun*a)u|XF_%%2ZEhRQ>Zxa-9&>V?&=<+#U*Uu|)07emJG-D*!HP<j)3^Lf$T{!}PH
zRT97OiPkU8^luAoj4~F83+1~Vm*mPAlJlo<I^!P!pN0xn_KVvXNz<da4V(0H5SvGv
zvKGqicd-R+{r=EjH{l*aL2{TbevM9Bm(U}mAWci~g3P1`_<*@#7)U%vnbc_0p9EFd
zZpp?bE*Jad8go~pb<-=^dQl|_CByX|c2&crLl@*hEN9bcCP<iEi%Gvy9`_*5sUd!j
z`^O==OBkRL45q-jI^G9<4L)O5JF`vWa-NMP0n8KULZxp*$@LM*SFMJgz9hpP_a*lL
z9jXNN%z$<n03Qo77<l+#pR-4r?gWufe1ReqQg^?(OMCuR&9M}OdPHD|$SDGB^g^LM
zO80;n`yWElOx}wF5^vzbTah@tlFRU>eb<pg%~!DA-@1rUg_tO!Uh+ubxIxs*QByTO
z23>9Oq|r-AV~sS0q#ca?wM3Y9&Q2g%lB`6uVtK`-F(!a90k_Gq5`)37L|{Mt7@F+i
zTOJQ9%{Jnx$)uL|70u?ia#_7MOeL<&b2qJLpM;c2509WbX_p6G(ymfF)V!SAhHzx@
zUX-0#1<zxR0Ur%0hP6^sIQkp5UQFTF@v+U{t(`=83Q?11uZ>x%5k*+v_!lYP94j=}
z^cQ~aCL<q8{)_WCzt6vc$Y6;1|7)fHkHI$!)Bn4fBmBR;o~^5yGa-Yxt&yvlsF{g_
zsTmAEKg|CNsrK9`QIb;tVM6MD(U@}5*k*3LYTCp^q%<^!6Wz+f4)nE^)eMKlmtESm
zPeU0@bnKt(I&$N_nZ^kQL9!oqqu1sWhLq<{!X7)#hpS^pVH4pGDU*~TI}C;w=h*&t
zXt;JYBsGEQwo$*Vq@U*gnYi2Rt$h_|lw4k|u`XY)`U17}x6qZ{EL5Z#hVz+~_ep~T
z<0i0+79>C!B|^}GmQE`@x$q>#Xg4olqL{qgoDl=D%-fb!27#F%LdlJkU+rR$j84v6
zgf_2e`GQ*>&iE14u)T8P?f?sGGA>H6W;fFL5~~9aOIlf<$*rdj#vMD_WB$CP_@x*b
zMY`mEOWmCX5V=k+E6{7LdossLW3|1D^-(l#+s@+K;^5hoE_>R1@zM7ioU;LxR9wb<
z)Agxa=SHBOpitl+Lp(=Oz<<0C_W!87%q(30bLFkkl8W2pM(H`#I4g(PfYXo~+S4;1
z2TIcVBfGnBLpXWfV%~(Ax+Q<Qs*oTiT}ZnL&3xOBl4e|Epr*v?9b@^pis*XxJz9oH
zAYk>YKH0I<^zjiT;%5R&E?qXiNdJyabAS7>xI2P3zY1TJIGR`4f;|xiufh6H;>nH^
zZ*qtB^HnNR5;f;`NYNyC0gZhjTx)`wND3kuC!aSl-fBw|1(7!9OR$yZVu#nl=_Sy_
z>taViWS-65Xy=8(cjmw-50R7max$ekg`|z=<n&jV{+^=d^3VB2tKd4@MyyR`lmr;l
zO=nQ$mt!5TDVv9Z1HEO$QEMOUO28F?6YxDy(lWp{%-VC*XlwT1p@;gJ)ATLyyU+&m
zs0=WPN59!&$JO)eO|i#)w6ZU$VqSN-M;kyjzplF;h+4_&%x4kLYYT=tV|&6mNs;G@
za0+(==_39WTl?`j&A`1Z0!||_KG{+mTIu>Rvs{-~-rnCb6{lZIo8dIZDrFXk%biqm
zDDL@~ns)+{YXlVS160*4;!s@pms0Ia3X#^GH0u+OpvC@QJi$XL?u5zH_W9PURMgfC
zX9*RsA8}d0@$4H9c6I=bcgiF988)3HJez98OBe>Dg9p2MW38W-W^29qzD#3XAMF?)
z(sswV3gW0|S`L<hr5NU9*Ij}>lctX}JS}d{2j;;T4BX8<(I|z31UB|owN?jrw3jtu
zAx~e07LPa;v$e!M{}O<h3M^}kbG%$M6mhhYrz0Oz>*`O353b!x*jcdQNVs|=&TB!i
z4W3;a;Q%J$U33j0cFA2@!%U2<eBw-qso!~wgJC$`96X5ptSUXY5kA>J)~l@H6_TF(
zBFCbMgQX(lJ#;)VO$i}{eTn4A*xLEm&tYfVRE#K+NWb5IUP2t}1zhm|eh1{(V}1)3
zh^S?sf1X(`|EV$=n*kAml4eGv#>~YMB^z3W2g-9?--n8}n_LjO{h_aZF-JWN765!P
zwQ>3k2`NL%NG@imH&-j^Bxtk<v-_hBv6PgFX1#2hYx+c@_%1h=faGgb3L|YKrAD_P
z^WmlN1dNU<+pc2aA$qV2BuDL00wC~F8LNS_$6F#AYd~_5FH_i7%}vv0*Gi!G1G!X1
z(=eEr!QO%0eQ@vp8IwHH38-_Kn_3x{_P2NC`A*0`fv$l#D+|2h5+B;_XxQgOwah~c
z{5F=IM(sF7EtGY{V*c>r6DQyALlWfKYO<Q>aw|J+9=99pc^k|$Zu=$bBZ~tbSPuR=
z`sEKGWVya@?xPwr&|`vV7KG(H*s2hCdnC88w;;W7WQ=A2&No$gXra&AL8fw+@Y2yY
z#I``5#gyVL?N&QpFAMI<BXi3doWwgTSO+P*sv3mz+yqpZSNzV!!u8x0>sunpzl9oG
zMZUc$BCiN3HTg@^r4=jc22l5QQIRqva(lZsP*k%6LyNIULYoRwV(K%C_i-X0?G;pQ
zTnLA;*E!6(Q<s@DuUO~5fji-%Oqrs`AvWxpZ#DmPmP0|bQR}RV-qaj?s8BwXLG9bp
z^xykj*Zwu)B!}qYXBOMkh^f1k26DSMp`z0{_L`K((0e>xKu@qb7cEjx&zp%J`VT?q
zTf4k8l62g7w^{P()HHHVHPnav_Rt9V4-uo<Z~ik~kv^-20rWIS{Kb9bX;KMk)rrek
zFkTvP402YLv;J?l;EZrgGU6aGvsE*!I$pZNHrepEH|$C{k`leU<5+A@YG_br;3WDW
zN1)d(dIuOP&S0^IwcT21==Yg52B>kYJO)s8um@#+6|oyAYEzhF<1q?9*XT%BIabSK
zPvQZ(S=~yt-^}F?i&z4e_yQ5+vUFX8<|bI*jXZHq1?EmdLSg(B(N0E28&8tYSi-ga
z6tZ~>2i(bZVb`2On;l#ksqww4o~nLgXtUU5$$}q9_RepIw$$3N<gLVFl9(n{W%MNA
zl?4Q@g{Ie{v&mIH#+2~eN5XL4Mo_t=U>%>184>9?fFWQzltCo$Cvti{e%Kr7f@rGo
zMA}4BmXN*~5v`_zxo^{Xh%HTe(NlAU(kpu*)zWT)f9#RC<?$B3fpu75`;Zw41--B}
zzN)ES<y1c1M-k#jA`W1A>L&j%-cDe+GYTpuB=YVIRR;^*C4OcnMWKmGKN=@>qQ{$v
zz_TZ+3^7kjO)TVpt=?7Cg#M_vnlg}C0tW5g_hBUvh>MQ1rM0p@`sv)*iNV!<Evs9L
z+-qs0+lvGT<^e02c0>I82ltrEnJY)M<W+W4_aLV#{PT`cMde=@(M}yYyC|tWzv?g}
zOZ_|Dnj<gsE+*cC5A2Ub(!_7Hj}chl29XW5P8eww<A6dfI=D~g8^SShZ0{RGR3Ba4
z!}Mr$df>r9o`3b}y53Ztv6>y{hjMP_fuu4~e)=h#;DYG*HY?W{?7~z7<v=+uRT7=f
z9hpv7gTpy{UEH0Ndwz%weg-1IGksDa_qsL1ejpiM)pY7^^}u+4T7ds?HW5Gry3UnB
z^rXS*-3ltfeC=B<Ghj3p9r5xp+2V3}mZm^wjhAMyL1dtjTa;-+yaT3^CqmTe^S8#l
zJ4XVXm1Ez+bXkexA>}?GrX^1p5^t{qq6i|*;ITnGprLd{3fXE5_Bt7S%cZhmXK1g1
zOp02yW)J3~(8Vl(txvTnS5*d=_DVABZ*RBt$(;NZ`rjHBQ5xkh)E@&}oWbE4!_PO{
z4f>cVS)#drg!+P6X5n>WX@@Gk9T#^tfgkNK4zO8R#G&ztYp73tQ1<!h&cojD@+Dou
zRxq^6CY)dGO541_hMfq-Cq8*Y^9&pgUVDV`+@5R&H&wO08W%Bi6Pf(4Nm51xSuUwv
zu{V$m3$7$03<aklFjgaFz|SY}yQNVB)BCL+sPuiYpH+gr3w)SCx$bs1sQZhV<Yzzq
zzJQFo<^Qp=>`+4DaFcXt_IK|m7;o>lAO)5b%D?Y&xlghb)jE#+kFy@woGrcPq*-6n
zd&KVU)b^j}5b{g~56;@m7Gw_z)Pd9)iTyi4MMXjly?%3(4U!Vw-j?UE-l1UhRbe%<
z`E<?t5$aua86uorR#n<Q0t)HCN??4VBhGP~=W^%&58m!63Kz{+_<WBZGrR4R{k66c
z{P#H=J+~H2!VD{#I(A~u$3V`#;k-dE|0kjZXNmTIb!z_swqa&x`=51?X>FaP%@*|E
zy9WKO&J;A58X%VDJV{e)e_NaqHzq_z#y#6mRwqM=)grVvtv+^r9E$|$)oE&SS`gXW
z2^}4;PkH`0KGP4bb66P6&gOO#YFc3b4g9tKNp2!W?is*{vF@AEJq~Vs=1UHf-TUwE
zc(UrIhbvDh;NiwUyEE-AFRwFAW7gpN_RFKTwx{R)+4^4+&d0v);%AkHLANiz|G)Fk
zv^Rs??mgNaS;OEAF#SvY#0)NrH5ZKBS$)ey<iIARrRvNOGZyLub#7x;{oL*h<miB#
zZELmG8(fKPfCs~u-LXFVvio-P=6i?z`M)9@owO`xwWnqBeios0v^wA!sM|Q0Jvqa`
z;`nu6aNns}wsD^P!Kg0%nk;LzW^)Dl&)nMu_bj#bHm<67k>1U0)Rd37DtoOrhdw)%
z*5w7i@*gTX8*P%8n}$bc4bHlHRL}K>=kJ8x>&fQ<Nn_8lfG6TZX%d76trk3`k_WSH
zJx+ITmO>Ic+wEwv^^P=~UQ<}?7TJMLl{&BXpO^xd#9eP`byi1;naI!c5vnvKWE+k;
z3C6_qXs=HyyI3L8$OJ{}uF_}Ov}@!hx(92+2P7Opm`P|Hf4Rm0vtn&3<bsXqT^${)
ztLLnDZ;NVEJMxDYH~$NM_U%Gff=dWA4CvJG6G<8N9ph#UVPc{PJzynSbTH_F3CM@q
zI7j(<QQb<Ws|G#p=a#633WR&{5J6YsFZSU~pxiO5YIJaA(KxFcHAr_iCIlEgotLYX
z?Gq(4<_g7t;dJYVbiWkQ<i$H;D0Up91Vl7kPWpq}z@-M83;ts#xFHp>v0y3;n)uM^
z&*1fylT<_u5q29D)pj5Ra8udv(t#RX0YgMUBi4~-{DS;Ly*TtVAzB9U&{PmQIHVaR
z6gSV61+AmAi744DBk|hii%I_I$YYbbP-$-&Em{0RkalLVb~vHYfLLVkz8PV?bUczI
z%QGc3vbk+iB``5&Asq^qt4Otq<Q#oOic>}daKKuCCtHNVSpp7Hxmz$1Jc7jq0l3M4
zdNA>DARZ<LRiUABtQWh=o}M?FRj8kpc+Q8=32yd<^PEz)ES$@45Gfs)6YOWB|A(=2
zY7C`YfNX5rc5-6dwr$(CZQHi(6Wg|(6MHiAFi+pZ{Sm#pdUw@Y`o^^us)3owP(GET
zOHc!i=Bk@HN-3MtnTmYw>;=}9KfAkZe}@}#69WgDGe;VEZ$`>YmLqD}i9ooa>KE4(
zZbL$_JpMR30o+}kPMH_jFlICC3BBPw*qrX85dg<^*={wx1L#8oDCn?m^5xFvPo0A8
zCu6^0)^DuPnUk~EV47z#ChMc)$jSL*Dxo9{rd_}JNXBv{9?isTU2XW6@6~O@-L#_J
zU_5h*8EAY(GuV8=A(nE<;utFn{HyZS6vil5bHFvgR7YZc;NEi9_4@jld04M?>&f%}
zVrY+?L7|u|KO|FvK{P>NZqYp+L{*nm<`u8E7KgP{8U~1`;B-H+eR8G}A_A8p(Lq0x
zFXw-4fH-wdS0-EA7q{mae|<?BMzC3h4^E&mABJ)W|Hnw$3-n2!oGvi0wBK(PV8|IJ
z^9pWSWC6Yua-Ov0BDW$aYx0-#hsefqWF0Y-JCZ4@Qm!cwlCX_lB_>Y7F-bH6F=j6h
zdT_`Gx_Jhwao=3TqumB@3Z{638o&{MO;`|to<-Nu#~t}^g7Kn`;OaQZbC0o`&IWHg
ztH&fvR<}tmt6c_qR@+q6v=#s#2O03^78_8A1qc}po-?RJbW-DDj&gW&5>T+(C`ZMs
zw5(*QtY?1GF#yn;y4RCyy>QPxnSF)pMa|r>!QqT)N)1<J$aC&Z)gKrzWJMXDH1yeD
z_Vzzyorb^$k<}d)Y%R@aC5;=Y?b`G0-PPj*wYL71QTGe}3EOW{h`?(aI`9Zf{1Dky
zWrtPK8`UaDr61F`5Hk3)?vRGI5ocHZb(P)%JmKIM$Bycui14YQ#k_p+Sy;Mf6$~kO
zl)WnCsx7!l&X?eSc5k4=1@6Oi`LN~`xP2#c$MCHn51*cDY+u#3N-A_?Re%LT2H5lA
zCejBv25yS_Bp8@<aShNfdjUaLB|Jbz9@|}rZn?m8wGOp6TLGvrR~sfRfYk$?FSZd6
zNUsaqu=r%*A(GMb;z*qb3|H&KG(+ou0Tix~t?SoV?ntv&uHnxO!GrpwqWqWy5Ei$J
zFg%1Eo?wCes@s7MB2T{5kkGjbIv0Ly4kD|l<sLhkRd-Ap69vj|<+#wsYh;QrOi5S(
zv*<mcN)JX2DtG5EtZDSCHM2>87Yl-MRZ+xX9)8fRS$~f{pXBhrheVCYTX%%<qr}5@
z$mKh1(LkOUlKT?g;~sum9>LB4QGjy>^44O|=<wM&1};CrCdJ@9k0e$9gf^J54x)S<
ze~_bO2{1Bnr0gBUjSUSW%kLx04s#LdP!tK5jTTxt(B4!oE#>_#B{72}Ram=x2m>mQ
zu<TE;-vPL$q*^p=;3Y*=0+`!x%`d}%U)sl1klTFa?VWZ9{tiQJpX_aeDD#ZqGY$nC
zgv1%C<4x1EiG><ou)?Xkm)~ba;M>p*?a}QnZ3%e<6Be3yOV6n6Jhi2S*{5Q`v@CPg
zKji@XLzUie?^I=as<Z*JuF2(2bZ&~Gdw~J$xc_rQtYz@~=o#?`=Y2KuacV_rh-Of<
zbant!WIF}vKUfGtKXp8b5hSX>=Bs$0vzrj%XLs)NT(<FI1EFZ^(^2v1vDf3K(pu7r
zJ1$S7o^!r~6m^p^f{+8iRDg;Ql)E9slcbM_$jFzguzi73UpnQ}m-rog#RY{<ZrkUg
z=e162mm%~Uw^<D#?v&~l6%Yt}0ux99r}`T31Y-cFR<WpWu*fp2$Ip$6(82^){k<P}
zju#OL2=9=NX!AIV&!QTR5`NAL{wQa>9s1IxfoF@9&uqqU66EK`7_;bv2`RP+*b)v0
z_V86FE3mw{XY|T&<2OjQ5C**SfeYaoOD7S(kDy9d1LdfO0;LGbsf7X^ikA|dthP$9
z?}5~1)3yuLim79R-?2q|evX5H%zHg^bRrstP|@z3d}pJK9vn~_<W{$3RuXqlvUsu*
zekLF4yUJndbsvn^#8g(^?fas{JX+W>yfM=Wj-6*qU*(+()6)(fx}OWk^z}Z&{c7-7
zbiQr-79;lWp_$9&X1i?+<L!AriKxo`?JPO4u7$5Vs?&Sfw5sQ17=V>41~E}UPD>@&
zR9>B2*~uL<d}e}ml47u_jfvB@nKS>J_`wTD!po6M7apTbpWj6l;RB2Cvu5I1LD2W$
zK=JqU#O0e2U_C@=9NxBNJh*=^b-c&@X)Wt#|9LH$-efYYYZH_D_IXTIifAAo=0<Ac
z8|5IOW-5b!ch)Gw7IFqoUNGW#cYLS~Y&0GII`~&VgW0UbXc<1VpD^xqXD05~j{`>M
z^dn$ooPiyKA++|IOR?>imWuvj-)G~oS)Ilf`YFctPWO_Y(}<mM<Ki%A*@Y|PaH-h?
zbQhk^zMRBNlvkaA9Ac1H4VP%R{0S{!)X2o&XuRgFSen62xcdcOHDy)vUzznk*j9`T
zOzi*je?2UVUys96mmg@Kis;mON7$kW@1uC@x(NDaH57_?kbf#gR*kr&2vzL-bKh-N
z5K0)<dSsl9Cjq}W>~6Q!4$sKAN??-Rj4v18#|4X7gs7%s#*`XlQp6sT0d3+NqX(PP
z=jYewW}Pl_wU~UeelR3B4%(TU-|IB|gPPr}64jk_{)4hOx7XLn2it?1?RA{pH{;L4
zrWzlXw%HnTp1drqq9|rJMY6@Y{A?X`nBlf|N80prwsF%5V;{|`iQ6Swm!VciTcgXY
zNy8sLG^v-vM)1iW-4(UaC=l}iamtc#D!Ivc`#~fI!)S3~$a9lzb_7Xt!`JCO=zj4%
zb_oWewae=|^*szmxFdc$ul`LueC$Id@ylxRH#YddV~B)$&VLA2;D~2RL_}b<bcPh|
zsM|uyh$ui7BR3+1p5d&OSp0!a6}I6#LCIDCMG}_E)h!?IjDryQi;f7UezKM@#PP{A
z;|O4&UYr<W@eU+RDaFM6`mCN%!NdY!_W-z0`kJ9rh@xKQog8KB)>_Ax<*XP0*w}uo
zN)i<#qa@%S^lKGir0A#l?kq(LqJXK4xO{B42p@EG2WFyc_NEvWqm}92lvJ@Ck&3`x
zQ<o_?t5#e1WMDSka{fHXCE(Y?4yDNgW30MhjnwvmW%eXSG=a~_lrj2?;r%ntFdE()
z!}3&j90>@4_OrP8(HZ4}P=d0Yp`S?GUrp8tdDc-^<J(_O-Cqu>kU$v5DCgF%f``vg
zU6}CjZ{{GFK$lADXg;o!ihAWHIT*)FN}?P-l|1)F&%#E}#*)~3%%6h?2?>zwGUY*v
z;qj441dK?90rjk5KHSWqXpqTal`ldt(BP$DB-gc`+=;}l@Eq!nR>cRh#3|tpcEyH7
z8~YgsQ_rGQ2dr}KJLFI8E07|BCdASp2Pi8J<ZLhtz%6pzLIeYkeJnp!daNh1ICRhx
zf8w6c1OnM0zdgV!d<TN2zSV@ZS!)hsOOWH>IVfaCY#5bb0!hn51Rsv}eEV!zu@~1i
z$lTP@s!Hv`s{SdpZNB4qX1SRxii#UyOEP>Cn*#MSi{dn1bORmPshTq79G?uBZp$Rm
zW?C*TiEMY!#fF|wK!ZEwv<=_^+-)h025H%Hn??_Zi23R)+17N?#$MTMzZ1R0r-e*I
z8V~m}o(Ho(r`c-PucJ&t^TAM{z19BNq5-lH2@x-yy;gC&THLh5YrkCnyeex7kO{4?
zNNom@X);#G$#5YXQ0?DO=7*uWut}(%`F@lq8)eF+@58{Ah)h2TYVvKLS+&POG8&70
z28`-y0|dud`#q5;8iRvS^Ua3BA;`FQNPC~z!zt}3*D!|x!id#9tV9Kq?+}YHE?D-G
z+_}J;a;4)1J17DerQ(3~zVsIaQfV*!ZzwO=Dd+%pc|#H^a(%ygqH0ASe%$F7DVYvw
zu3Jvz+9uo7L{8=V+&v<Jei^mw`SEW6F5&a$=G1gsMYhriKlM8_YG5bNI$t|JuVr4!
zU2b_sV&EGde*1UL$0{j&NGbCVA=JbLROssk&9i^-d^qpTmAePd`B|&u{ZVir-s(n&
zdtV>XI;n_W2-7nWs(m1y6>_Aomq!OQ1xly|_4U|6AZ$1%1ljSefY`-{Az(O2ZpeoC
zyR9$vTy9+Em$qBGw_IpjFlI7&!k+)eWw1@7Im*JKM(XIGxmbnHfJEj)w<WJmj<0++
zzPZOIQv^T52zHrFkkchad7HMbJ2lOqe~rDDUe$ihb>au8P4kW)Xf!bLNz-*u_)v{B
z#r<ssb?uAZ3S}bM=NPvEYHcW!uqg``&-xyuu#gS1O18NFX?E1q+_x{J@v_%MgsJe^
z=8>TI;psT!<B*YOnnX3|c;f!;#V7z7vVd3s4dev0thJlleWVjLvDSTNL3>==Zg=di
zPJ6C*t~ws)A`k#dPJhO3iH)zR=;l!{GkeZyf(`dYv9x&kQidw}bef$7b4#&ey60|5
z?e?ixOjtJ2^8Te=2wj8hn2hcLqPpzjFwzJ`$+A~%U4#(^U;!(9cJk6v0X^yuv5S;c
z&Od|2-(Qp6R7@nCP_f@-KM!Xa^bWV7GZLcK=S(WwrV$rMst|;!axjPf2lH2+m%l|S
zKdoT!D8ud6p)uI**;CDIJSL<%|CWtOFS|MkpYZ%O*4!Ecg$f+VR~kel^Sv8^DSRrH
zgk%C^ch)dDRF#J`vlF}_4yKEGF<9_12*Ktg?x3A!N*eMgY0Qu|DE}*F90@0kLmsxU
zh$d-~<HF~Kd+aZ)BuY5xHN7AbPe&8}N9|jW<;vyBh~8YCc7TdZWHypy^-QvQLPxo>
zyx=`?+A}!QijO8Yr*X-PoTXg9BYR~|Y~)rwS|0O|BYzaJpDF(v|FsD7pg=`*SXL-E
zrzsM5mi7fcBT=3|A{7`-ngR`u$wx$29<8Y-UoQL|=IrFn>4ksAc(k8$(Ids8M=9YC
z@SiZH=U_A6&)+(RrkfdQhA~itbVu;vCP~lBqe}1Dd&ySc@nnhJcr!Mh;bK*b4jk-E
z&Kr(k64cqBo5>Rl%DbO?v{m3(c4t2)5Ne;z(^pLz*Jaxrw>j1|P`(>xC!n@H%_fzG
zv(G%Nb!@FVl)u$^T9IEANUlkg8tUswxZm{=Q<<JVmpdCW4HPoKYcJ7VF{4ZEyS0-f
zs#Q8~^NW}$7jr9sE{0qaHkW-Hu_#Y5&&{q$Cll7?TO-pJv)0?rt3&#pdv5zizR1JA
z2hip}FIpy-T8|?y6FebxLJZwERa;i1db>w$*K@jZkpyr&lxg5a`&>$jr&N+>uJUVj
zZ{#HZ@&KDie^mY>s<8bB;*F8*e_v!>YyEDs+fjeF+39Vu^DsjnC7UQ~0J0Y{P8y>h
z^6a97UdCe6(MPeyTswt7Jz;6{@r@^jURTEGQ2>~dL>_RnPh2E1;HSXc?mZ9x>`&jQ
z1E>25h%0b#a$(bPAZQU!4DH$Ra9}}yJ1Hb2c)p*<*PFjLX>MZ5d;1@Ha6X53w_d#t
z^)L5BA8z<BdN)5tKX-ZG+hICFVSIR>udcWDxB4%ycV(w_>`@Vr(ez6FQAH)e*G)md
zZnJurNG^EOUUB7u8{&K?1bj8M|1=bGy%Nz?w)>4(9^w`cteiL|huSn;6s9u^sV?zM
zAqG*-4gdjr<3Tonn2HN#_fa5leM2?U2@pZqzzK<RNv8Pe1%%6(v|j;}<zJ)I$%bPQ
z!WkLU`pIeRcbjn({+{Ic>O9JeQy_VY=Q9z7W8ZY8I62agnZ+;N8O~qt`aJUYcb2#<
ziovgq6vyo_W+2JttRN+DLvtIhlMB@%lrZBm=m?3>3kSDDqZT0Cnpf%;QU<vcQ#OX8
zf|B`jsT$iXXDJ+BTi6zM3Bv{jpdPogPiHQJiN^2?W;#EYqzc>`VLHbo$paB{COR=%
zWEjJA-oYjLowKnA{MBx?TR}MKr&2qw7P$^l%VF_22i3~)FfB}Q?1vQy%%m>|k=r5l
zOkaf!hA0Wu=#f0rB%tAPqItQH$<C~Exgf2SDz^|W5L&txxuvS^J<8ro(<PpRuwuGh
zz{VnxN6eY4Ujj`J<rh>6sJZt53lJ$IUU}nQ1KxKd1QG~zduqqcOG9E1SExu?-f7Jq
z-4x9(x+-ECxFF%8BolFBy3J<UR&`N4ZK|3gt5;E-=$DQ;r*EJMuY=vk!0Zu^8njfx
z6gI$PyLB;DBR{+2rhCv>IWqcjcRchvo&Wta_H)pjsjF<F%Sv5@T`WiGA$c)rdBeXN
zE}0k$cr_S)H@>MTw+Bd{l+rp7N{;B3c1=_pImM=H%I5J^#tCby&!2A#JmFSs+U4oa
zF`jb_7z!i?R18Gv4?^-}uwrE}Hw?uO8)}T`iT4x%_75whgx0er2+&YvSRTOMdl;dL
z7)%$PfxxR44r8$7i$%RK?pSyp$70Gb)Qe{;bcvPyDFpToW(Dk|M5*~|rALXP$2T%H
zN}}#_K$%OD4=yp_EMY=+Sn@}><ZKy5)LHi45K;s-p0%_c%$k<77TDhqwYuHJ;xgt)
zOT%?^FL*wy%EN19B`H8z&92x^2CGrOu+0VoXqVz>4i=RdkNT+C5Z1P72GFEV-GEZw
z9LvBO7?D0OUCYw=!gVfNiK<GcGOWpBMJ^GLfj=acY*+c;CojvIBKnb4Fd5fL_CgYX
zv-KM+7Sbec7~pGuQ2F9W>aGs%N^zW~ST&Afb>K4Q+4g;0EDrzaocM5@L?1={)rOoJ
zO7gA{%9*Ojfg2WkGUL>odl{fqJ-L<C-#Wbe4mG%XdKw)42D%ad67R_X)Xj3bG9|Vn
zOl~D09l54SPCXqNMG}E(NPIWgp&UU$CRDGFPRm4!U=J~*9Lq$-_)h7rN7W>hikK*g
zIe<`#9BIvzBy74#vX8{o-g5yrm>&!P>*{8F;8x+u!f|pogHuXsp-ZYP?UNRvLicvP
z_q4<A@yPKe-C}T4;w@VAfc|$IB92{}#D_33#~zl)F*boX@5S_Hnu5hL{<wcT<C9(r
zk0oks%EW&`r;1)RtL^!KKWym_iG`AOaZI%{kww_ZT}%Ez`ZXgR4#fLejVacJrM^7R
z<n#>X{9H9}%1aUgLmMj+!S!D_cRD#ogB!j6G7UomD0RAO7Xjz{yBk~?Fuk|lIB>Q|
zN{AG&<fFiZ2NDzjn|%$nA0<=?VRZ<pkZ?^TCYf3{(V$fMZO0*3L|pn=TrV?-3Ww~w
z5rTO*=?S5kAY6*dH=|bekV!<e@(91UVvzvwf_(jgE5kP4e}}?z21+&3g+nA*fboe7
zmG#$8$q`L%bgmx?F*Zz<K5S3_JZ0RB!3q&L=bxT|_x?2bzaGib8EfMUECUsS1Yt7V
zuwHEcG7ULCY@E%dEA3CT@=RTKS-v>wazAxxZ`Mu1(rm5pd=G8(aNq*Kh6$)dIwLO4
z8AnpDPZW=lytNPMwx?MfDBr;eesTaD@oY}vL^lW^>)u@x7b2HzNm^C`s#BgQ>4RIu
zB30eivu?)&4U1?fe1o#OcNtr@Wlj8U*zgCD9xxFrQo$&*o`&1l9p7se-W})~5a;zt
z0m=mm$L2tD&2uLB!I%^qRS_Rk@yjL>=gN+M^n4cW+p;*=GA7RxHC|7Wj`Zl>lg)&n
ziZOb}?AxdTI~7RvS*t*%aH#=P*pa|<`vqxO&&mlCVnDP3r-THGhu_0DAj=^o@MVi4
z4F8tRD-R_aCAy8rNDWc$sw&*4zZO8V5L;;~C=2)GI#vZ2pH{gRk*LsVAkm_*Mfj$T
z7<5Wt`H>WT^iJpZ9fZHz;7oD-YqY$vbP$G#c+9wfB<PU_S<Ri&PsN*-@5&2&`ZO=C
z!PuHcv>lvA)Dnz7eHuCzSv#b29iQ91;<DN50C=Pm#C&KNpTOs2nLIzO=AIQpS5aW?
zeL*qJ3DzX7^!sGdcnu52n3@5%`Ph4!*&B;_beg97i}u|s<LwFPfIkNh-<Ib;WM5O>
z<7Y*#i6X1^mmS_}7Kr?9)!LI`JOSO_Q3HLm8`4E*O=P1gYcN-$rOmlXgLOK~8_l$O
zW^n4fW_y{OtscaBONP8Hc=c~e*PGa&47DhUa%kRz#f@I>x)fjQz8$enaFn0N`Ff*X
z;iO~>54Vw_I<s%7h3nx*cJO)Wa>{bs*NXDD3fkDILCXVjQLvd|a+f*g@Q(ULtLw76
z`xl3uog^h|AKeH+Y<A~5jhr&3-tCHkylSjZ-D4?gx4E6>@;65mpKhw&#6@RwnW84F
z-djKsM<0o)x=;q~5X<k#3F4V_$cK)izDKbG9r$>H5)p55@_f9zX_4-;lv{8&SnNz*
z@2D(<N7AVmgs$<<Ae+TtCyOXA+KP|+RnU4xm?2oC{o6%VhHTlz%5^3yp3Tu11@+lh
z=G`jCVx0W0<|9VZ)d#a4d>=kL&OceMY$h%P#da}O9wH|`!Srp<`t<aJ5%Y8PgvkB&
z8586oD3J|(pALWmU>wxCnGpr;1+@HxByVR@&Vj7ut@3Zv5A1=C+$N!ARInHb01+ey
zxo{-~6zosuee!mTd>{xzl(%=z>`i?#dH`+f!15Qk`^LP?PJ8+#pw-mVuB#Z*;eHnW
zAdTRLj87FMj;X|s`;Es`PilPL#?;GaN1pwSTC2_1#CNOTi9SA2UZlo`v~LBs=<rQH
z=J$u{d{>!#0=$0fapV1p%Mds8x^?PrAoX~>ad)cYc9foR^oybt1%CdT{xV@XZ6mHe
zQ+4W5d>|ey@J7Nl`rPgL17C(Ll5UKOK1fjb=u?Qh8jhTUR0AEA-@Bu@mSseTBW^GA
z43GdG;UWyTKX@!`7ITg-!ZSzm!K^$z@&@zot&v5RuC2Epje$ocU3vvq`yPtkg=JH>
z*mz=DN)D0ozD9iAvWj-548C4Grl3+$$Hzn1%gtn4HQ3)@IyE>PJfLERPQUX9I1eBG
z?|;PKUqU23ft{fx6gT((U)eFTF#PX?V-2mi&ENQ&r&k~b&#CH;C{pp4*E*_LCJ1&z
zqm3e72c+>&-11oBY9dd@<G;<1d&}RE5E<m059R{OxD($g7Y@?42+8~5@-4a=oh~W&
z#38rI|D{H9#|S5mkRZ+*;`U^XKfa&-&C}~HB*DKjOdbv<YZtP6AIj_fnLIeYe#(Yq
zM81F0E;fF?c-mZzN;AC<Tm4c0yna58yQrl$j7VZnlMXlM#=9vT<=ytZu1Jfj(yOHy
zOAHsy4)TCujA+UhsoGA_Jgw;erYV%bV~T~IG<SMhtvg<r6z5pch~*2+5*nDOVF3za
zWjbT~wFogsEJ|a6-fp;PfKLR8nDcTE7?L@e#rtLS%o5;DW5$5iW-6gM(HG3-O{nbR
z0R^i2yqug~dPt5{+urWHY&OzO{b~g+<b)Bv0FmFhCWFNVVyK@GEYl%WZByuJFeBxf
zV7M8{r=!qA`zzILq~s=RmaB|(9#uBAmu9!NF~`<b+_<QR{l5Onc&(wfApM2n7dQSf
zR`u6ER;3P$ZgF`4En~4@?H!;-gbuZ_-QLHz`?O+^9xuY-n|vA3l@mO)7R6&{t#Hx4
zB$50)sghVk(fZU;VUkpqt1_O_b|=K8zG$;csf>Q>B~eE%rp@)ysK4oQNYC8+LWO<W
z&w@$$#J_$tH7>J2<{!P~RrmQh=`v96t$VqQL3-QQF=G4Ngy|xV-AeD65qkBAedVC4
zmKLJ=c{2@K|C~n;>uP+t(rwJyYnWmx!$0QQc6jOiY*%9xxB?utK!PSDSMQ~mYExcC
zbRCGDN^wUl>%dxI7m3lh;&&*B1fNFMl3_r-%U9CMo0tjAo@2SvLHVYRfT8m;lI0f7
zW86w2-Otc@YT(1GwP0<EuHM-0wV5crow_a5&@&8nfW$r8FR`Oz8bm^5TH4&3>Xa|h
zQrK!kEeCFB;C6vRikz?PXyRfXy6)v^$KqOO8m{4%Ff8ql?DWyYiPTw73zKRz?x*JG
z7!Uxaq`gd8NT0Y<2fb!UoL)Juq>c7mH7&7QPM=LJdDBVXUsLWmlz};O=SiYLOAJ%a
zrDrc9LJJNtJJD&`Ts;;O<^a}u(K-u0X=*%!(fe9qJi#6Ktlmr!W6q^6vy|Nj`NXW0
z-TOmpAArGYmgm?Wx>?{YXY>RYU&*`}wR(!86F3V4CgCF+bj&czjUDpf>+;cf-DcWK
z>qivKQgY0nC1XQDes1bOv>l?PikN@aJ1gT59A>2gWH+G@>W{P#4a*2ZbYczgb*TX-
zgB;2^Gz49?%)1u|6--`lXD`Xcd=`s?MXkw%k(N3;=(u*FV|A>%U@MD5JBM94AX>q8
z@#m)35_Y$W6`r@-$D3FfGR7nG4MC-KpDc{X&c5<r=m|#Ei;7N`#v+v-+L#6Q-wm+t
zba<5}${6YC%B!hdU#+#eOIK%aog1mc4DZtGf;KI(?ln%W3c2eMZw0ODzv~((smBb6
z5r`(~-WnsOoD{##h|9@OsHVp|3%~Ma!G=}`<R1t7-4dMG(xg*HV-cYM3%UEPB$84h
z_*s?Ifg1zxA8uvp6m!P|%sNjlSEY$~3mRlC#MWaDMkm;LNoK8+ReFvH>(=EqBONaM
zkxB+pMXqL;a#AG?)Czx~ToC81=S&kO(BvzH2_k>c+yXweb>IfLFWC4)lRph8C7h^)
zus}6)$(ydnOjx;nFupTlpB9g986@wLk=Q{-8wyJrsD$Blt33+KK3>(UqA|x6&F@;N
zN*=j@vZPZfORblcB~p{Le5%uxytUVc2B9`8c+xZKHq^$BSz|SahMf+`*doSz?F~^s
z+QAx$q)=p02B|34u2dIAkOr1W(f0fP5p=}iMMDMd?de=fXhq2&%6QZiLN^qa@?=SO
zd-)^uV=K;I1?AQ@1*&dS#`4iJn^Rf;;MTvDchp_6yWF(0vq5wjg8mW4*I}MIj9BmV
z(nY~P1#tv0m)-SsuF$w6&#mj2*v#^(W&JJ9e6iW=@QJ-XSW->xk|Q(TZ1~d+aIs>9
z{*E@N>taOW#8F=x<zu#ZT1;(BEIth(4=e+$wrAFwWU$<*Qb3QdtkAMN;2x~YgBn&T
zU+Y6@Q7ENU4KGpE>WY=%AwGh9O+C=CgN9-B17iDp&S*h-Xze|oRYqYJzRwH~uTe$P
zWORz6v|eeJE?A>o6U+6K$)eJ24u=#Be?gDF#2xHl(+P&E@1a~@!nDiUxz@Jo+{y&~
ztvnY|9#XJ>N?A10=xJ5pt*k(y3Wij%Jds$Q`7n)4BL!fQ&u3KMx=qb6FfVZkpRge+
zT$3SsWSHpB0eV2!-rSL$J--1=6ZgBD$T)3Mt1%f=;pDiCwJlxmV0c>}DP`_l0kXCx
zB}@vTl~Q(qjQDo6Whcpmu9PEj$gDSq+e=?ReTJ_UzBh=FGzt_os+So!9@mtb!}SpV
z5q3CUn{sC!<sWVz49!9G#427N7zv|v05F+99F?$TV@VnmFGfU!ZT)hR%n%gIxzsH!
z7oeocQq+C@fhh)HRo<RPVIkGK(8{n_hmhv9sN}EUw*YuiW(k!ok^_K%10eK3>WmVx
z9+Jh@%`V+rOyzjFtAWOzwf0m<dfKQyCacpQcdvhcHYOF-qYdB$Qj9Cp)C+53mlA`y
zIN>KxG#w6!AeO$`9Bfoqby?<17lE~0UaOW4-5~tKL3vC47+n+zCJ$L^K9`SxUVUa3
z$Hm0Z1DKJ5FH>0+*$LhESb30HQBih(W<#S!hu}a9(o|;|4a~%7WQj<AfA(cB3rX12
zr>8_A=w`G`u4g3!M~km;{Yo08NP>D(byFW#Ta7~Ht~MHEQ~Ultz^y}7J*PhGZdZAO
zDVR&6b{tRc$HwBay<_Lu3}|nirsKihA*ykXli&L6J||FG0FDe2PaV$pEK-yc{0fJ(
zLZx>4v1NF>&bLNaUI0Ftif(DNWpTQ8J-OZc^6o2qR0@CvY^g-n*+70heUf6Xyg274
zVF;Kf!ZE7(5>EjGHdDf+wX~9A<C`Nh)p(n;Kew9&?4?xsQcKS%ByJSGM5V`hTxN7U
z+4N7TijBiv?M)kQ56c)2T4w26yO6E?^0~K*L9_$E9g}xwv*8Ru!L=6qMd};`fJe+0
z8>uR4+fj#mUYWtzI5}p&BP)s?cDe*eYzX-M7r=$dYz{g1@h}b%@rjox@F2U*#d~s9
zA&U3Pl|r1TxhvQG{ZIHJ)yRF}bC*A1b@pnk%~R@b&F)#cF(p63U@^`~L?<r~7#iPd
zk;arHHKQ&s4hsW7_C1XYeRHz0k4_4E<w#mo&OBXy`_<9miLVQCGxonS;(uI`GO)7#
z&usB)O)aNIQ54_VUwkFc@Z^r$G8~<We*sg7#f2*;XeeT9cDa$nh$_&{ao@|8NULh1
z5qYQtFgL<zRgu$X(=NvcC{B(TyQk~Fq1U5*rNBc9(S~A1tXU#5;sMe@ZQ(<UM@s?(
zv7v-`_q)xc(L=)dpft1DANTKrb$ek@ywAtam-{+1IbY0P%%9ggN;4T5AmS+$jK;`h
zMn>|rtPu)T<?!;a4(>Qt9?q_7Ex8f$TRDjIzlaeSLh140(09ye^CWx>m(o!2s|iQp
z^jLou&0YUJ*yL;li(^zTV)#Ow=t8N^e8R~O62p2VRqp@}BCZaT+N#n9O%22*v=hof
zBGvCLequpuv%djPwb>v?>~*Q7pw{3H#Z$fjdpErYrZm;siQ;d*iURr6-)%$Nr@{Pw
zv<&RFxq1#8NM*!|1`~nFRo75y<d&r<eeKP+K%&4@xGR-C{QJDVVD!uK)lSzQRQpKy
z6{1Bvp(%?<`j@2s1)ot@pnWj@cI{1{nTonp`-}auhyRkG*ZVKwM$SxcW+$7Ir|ODZ
zE>_v(fTIO420~Oa6j0!d<QB4*a)Q_F>T&U}td|knPw$ImEXVnuBW!_b;u>M*LKsL%
zo!C7Ltug6cJyW%D?mv}p?--kTFi;slAo~vBP?*JXyDTV<RPlwo`FFMX^s~ISg#88~
zEr?+N_IwKPVG*5icgJ&Fsp3s&-aj9h@vtP$uy{b@Yu@z2A+aMI={S1E)WD9$^fG!z
z(oxkD6NA-r0*!&SneU^^B8nZy%o*D4E&sNubaY9dVYy_Gh_Fhn<_h)G`w<~v>e}om
zNKhK=AcP5t!$+T<pJ6Iu{3*iJT3}F5$2UCO60}Qr<4sy6q1!mMG9%ijw%eGiXctP7
zsLWi7E37hIHu?d;tFpa-*GDfFEG2Zj(v#l`Y4W+M^>0i;gcQG$EHtYdEw6j|3qKNt
z<)sNMO|_b1Jok?%XoL<;u>kNv3Uu`!MExn!?dBWhNX3-z%L&_SQ5#DIK^Gsv*uNi%
z)LBs==_%Rf7Mjoj;saNJ%6zDz=?YWFyp;J*4Xx`UW6-O1w)~N*H#+W}@U^4OnK0wt
zq2l<DE%r}IsGkR;23*4F*St$;G=&N>wgZ)_GV+jgnW)~K1qqFSbS+3oVK3Ip<;Am(
zHeS`c&l%%bDL1GR!xeFx3F%7$aiH!LiD6#^@p(qRDwamPV(O6wYFj0Rh&!)~!O*&u
zx0FStorqhM72qoHFqm()2`TIZ*A<R81E8ee`wNwhkwCBvKd5<be<>fOXj#m~ogIh*
z3_+J07@H8d+H(O6a0uid(JhHkYO-q%fyw$cm0^j138HWYTV?*zo7v#LS(DUgHAt5v
zNKPx|5j#M@9#k9g^MH+_uPvjvPP~D@lJPqRka+6;2l+rKA5ig_zp@6N$4sCth->g;
z9nkXk=qHqXFsH;ex26y|24ny#VVnDkpI7l3Aa}u9WZ?LbO<NeRAd0=l>yfz0LQOJi
zrYw6~FSbu2FzD6%y2gKw$EyDvGiP#^z|Y=o=O>-o%cf3`!>ZZcE=%bbO0ZEQQJmWq
zcE!n5Qcj{K`t&;gRyixCOkUN^9PyJrx|MTYTB*!}uAo2mwM#0(Y|FT)|2EkqZ?aX2
zwzSWeu4zTyPd?*0!|^-qju5!Aao)WHejkqTk+HpRU(;}egS+P$c$R<<sGDpHw4sPD
zC(m$kaa!k?!`VmLLLt3b#4W$nLo4nudf>VEQGzAuUcxxSas3q82}4WYG8<XoT+=th
zTn$-ZJTF-xq}_~tboOz~W*-)DE1O%oMZ-khNjvjU_A8aWxlXrGV)dIIVP>KdhAk!s
zsRs*sqdE1AAawEk{qW%>pZIH-mhrrd1wgz`O*avU6bQ4-M^Setu(ER~(Az`#x?=<5
ze%hK1Bh&$ggA@4NlMsn>4JzxildUf>Bdw^M`%AnIwkCB8OoG3Jz`0EhEJk7iueq<s
z)~2Nc8W&)bnmW^W*7bckf;ORt*bQ6<b6pXy?`yaXN!y5kehZx`f`89%YMO;S>Ti6S
zM34tU#Mm>%hhGB8p>8LE;9{C@m>dqd9{yZ9lG_l&71<0oG{ZWDSJjo-s4pRKX130c
zF0dX<7TWaU7SSDIRJ`+H>d5g%7>$#!n^})CTqMfXpcMxZhYa>Xf5^3qe1XpdSq=Wz
z%k00ZNi3}YvnO^LOULQ3^?tiAFz)d&dM^lIEaRE5XQyO!++`z*^IAF){CDk0f>IHm
z>~m9h3#2YjAd_>QWyOfe6Apsb74-#{%G@Di{%?9YPwlt=J}s;mFX)g`?HVB;F@;i)
zA+7ogyih(}<*WUEUe*$gf{0|2q?iN<1NQt{t@~;4eAs^vmCBTK^dUx-+3o%K=vv0+
zd;Z?`Q~i0rJK>7Dj3m0h`&r%$B#llGl(bFQ1-W9=P9mb?hLr$t=;nU=4pop`bepcd
zs5Di2%2oB_#M`tjZ~b=p+WUb&Ikdk37_%{JG!vyj^-m(3y3S)PERj_OC-|rhIMztO
zn>Yc?!zr{%EkiZ5ZByCq``0wos|geoAmhc}DzgkTW`-%3n`yJW3}4z&lOX-%&_tKD
zS*g96a*XMY%`x>a(6}CSw4eb;GBg>O&$rrI9%|@&>7$?&_cMMSi0%YjvV#;*5;?l8
zst0qO$tT2WidMhkv%leLLWo5f7pSd^1AhJn_aR>%(}v$_(0!kc9#Oh<!N=pYxUh`T
z-f48E8F%n)8W3Ybd)lOSu@+~FMM~?(RSOiU1fQM7bk#Ks!bLw-th+M$NZLc!i=UJo
zJn`6H&VP5IZMH%EstYwL%oV`eSiY>CXr~L5Vm9ec5!_KpU=6YA$=}l!HxU_<%N}O+
z<g}9Yqjj6}R)WGk8LEC^z=Vi6+vCnp4}Ea6rQXxOXFsYXsr16@Ru$;XQwZ_A^J;p8
z4=><@$R=PMc!uZoXN3|}8V82E+AZ8Dg9ADFd;*G@ppVVvI&Y%W_V(+r0^Tw(-+71C
zjD*hl2t@<#oim@(TGDQYp1t||$DEb#WBi%<wIBoZnCURs065#tSEZwFq6Xg;PG2F5
zzwY`|yaSucyO|LOj5;u4ghb&N@A)TuYQD@bKT-84wOi1R%IdL-i<<PKzUqo#hD-DA
zC+%X+w$m|vz_&cM%*BS)1lWCj9d4!(8I-tNUtz3CawunF3i>w1uU9N;t@RrhI1(V=
zNg2earbm9Zb7sMjDO+->44yCOo`Gx#0SORX1%(n$xH4U%dya$B`DT5zUcUInBq;F?
zs*SMyJb+ema&35@edK6=uq_FXJH;WK4Iww~SKR&Fq`MoE^{Vc%7?O;RqToXVFII;)
z8SbW#xkWDvLc*Fh2Y2Ec_t_#7qhZN|gcIN9J7vr}xA?d=m=3M_y*y+yZAxFy#~eq$
zsr({tw5_(SbFLpQ7npzNgw(ustPir~^uk*#)%J>4y~8IkmVE@v$Eh-X_&7nVWp)RS
z7)AwCp|?3_)-x{396n^W4mnMn5`l33M;_Q<Q!^1a`?VT(tt0K3UMncDlj#~#;yEP=
zGqH0wYufiFGjkBX4HOVeIj*D!xKyyzI&gt}eWJp2NpVwJkzy*vG?T|7n>QPU;}9}g
zkO(|sy-*4*_*~v;9U(wpS!5wpy>@i)ioQ|^xgrBNz8C>0S;jHLi=s83nLo))0_5=m
zi2*bG33`|jg2^*t&Dsug+#D~dDZw(}kH}Ix$K{z+I|hmv$B(1q0;VU;PjI$q{%l&3
zCQ&Ov84;PAfGnng1|kG)qw^3)e#&51&o}JVv54TQ^suA<KZU9a9ASAUM5SPs5W@iD
zMcoH(mja{tA;(0@`!z?}s8qo#eaW(DT}BZ+Eh4QY;anly5&hLfh>1J{5<PL`iwrc%
zyfVl)B$=y8&zE3UMghl}yp>Sz07mtDb!=pBHf{Pf#I<~XPnfpm4L;(0K2G$FA+e~6
zF8ro49(x+gU=w?g6YGPxr@}F4_FS`5A8)~}qDLG9x@(d3{!*tBUa^(QLa+y<H{?QJ
zoT96FJ%JW_RKZ;^iLi-h?bzrAzC5S9ULf_D8|rEuIiYQb1P4&D@S{*9(PM2CrRa3t
z8Gm@Qa`tS8(=`%mgzAnx;mUemuE4KmV?v~KrMMz982uu09|NMy2!MEgs@(DZbNIDl
z%qX_6G7+KBKsx+4ZB~Oh;3(<zf6!!8!%jwR0Q$7PcAZ(Kyx&#7EbwHL+AV)h14}3<
z6n2IXKJ87fE~$bb4e6r&2I#<Ijb!h0!*#GbZC=*e3}ms*xW=1goSQvNvN1m)WbabI
z2{X+6PnJuar;-aDFIxG+S7X|MZJFx(<v|g=#RU^*ugHfmL8`?(O{3Yzz)EMU_X4KN
zBpxK%bDC_Z3<Q!x$G}^6!eqw*Jo04`rS$ZHk_V<qN5Q|zkB$t8XyURU$~yOfL&(g<
zJzxlQ`33CR7g6JZ<Trrnfs)uexpG2Ig&w>bcSNUsV-j%C#kmBn%b6`P>owHr;ax5v
zn9TT3O9?*R;{KYd;pe_Tv*54eHBq6G@1R(y`3m$c&{$ue%>S0oc~3M>h0&GDdffN$
z*F#^SX9_@Gb38RC4&1Qe@;JXYtr8jT+ATc-sAmdJX3Z*`1IfUd9YoPi+=sEfyOl$J
zj<2W)i7pSC8sDtR5e9+y%>f2yu|(XmqG$1YJ2J<Ug#@KRgvn4dPMXg#a6hijqq=g5
zvj-`15l2=G*?VH#MqN6dc~~=_33jLP?4I}x=JY4!2`Cc8+OFQW()>q1d{}bvUCga4
zc&@QJSR9mqb|&33`r^XE&P8f^qGn7iL;2Cp>m1fsu4s^;dpVXNKsn*IxlpfxV-*66
zlx$@@Hk)}|_gHP5PT3ry9D@Y>G-?-ibgw)=#z|Pk^`F_(gWl@8GDdE+eKuY7o56pj
zOI0!+|4@!P)=!n^_$D3@XdN$^>|>4B=GfU1x@)zGg$!0nhPd1k@TzX?u`2(_Vp|8X
z51cuhc^T+J!alR?^!ff^X(4|sFn_IGIvP>THH^~h=MnGNnj4AqL<32)IyZD);zZB5
zpqnT7dTo4KUxZnWs7L06w4?5IB0j{5#<p$PdPe!s*%8w38Mqbjp>B|THEghVtxI#R
z68O)3w-b1!9KjF)ur^FQ<Z!R)IH|dgxEKqkf~qyfok^mdT}j%GxHw)76?oqPTtcOF
z@fQ1RVAUl}>yrX20vS+F)F)iW=dQTV^4Vcm1j@6p7Wm{5Zk^}RotL>iuH~GS?Gkls
z=zKY%8ihup0==KuKYhh9uMgH=w?`1k-QdmvQgH}Ctp-*l-PV3{_(lL)h*{+!qZ~~V
zC17UvHhSc)0*J2D42u&aR0wZ$WzS6lPe1J(qufli4TRP@ht%QT$ju1qmtM^(a1t@7
zhJuds>oWxC4FbWg3(7-u(3+b$M}PT!GA~M-`OC=03kv6LqD``74?lvyd||oz*EF`e
z>cNfOVTNvR4gp>?&1{A~VUaF8L9+D%GX?*v!JOD&+lyRM^X|#hv#b>|8LmfV_B9un
z#n()~^BAp`Vsnvcd?l1NmiYnZ8?9LTk4VY!ANM1S%xwR2q@2dwblhxz+WsAA&M_eZ
z`GY{ZUh*1WTK6zbm&Ur5+9ck~Bat|yp^Fq(v)laifU^ULC^n;bZCc=4LlM~DxGCBl
zE8K=ftu98X*<r|Z2a_Pib)PRl6UhZ578Kl-Lf;0_g%hM9BP#ryTxVTwUbgM(_4Q@j
z3w*us^?AK4UYj6%EMD>*tX%f(Kg_cE<$QZXgY(Jwc|Vc;wxd8vnZp5l0P)G+r$Adk
zU@J1(50UYZGlbpGbyV0avsAQ(Ti10N&#Od!U3Xp~UQyMyt?SmS#wr|wprjEX!AnM=
z23Sa#Bs<L*K<HrZ3&6DR{RZTZX*ceXaf;XXA}m$Q&%7?i0KCH!Mgl~U<;a*s7jfMA
z;xtBu)S#llk7UX8Whr5c+z#X7!^U|HsB?Nm9@CnH9uJjkN0klUHVZbI(w@$N&V#|o
z1q$W__A8QKy3^Qu8*>8zNx*~#g6`Suuzq-iEMx)%+W|hQrx1>%Bkkyb5rqiB{&G%c
zM)!<243a?O=^PU7{WxoJ%5#WIf#Qhnn#pKOlm)=(Ni*}{VJv5cF%pT*&!3YN;3LlG
z0Aq;fxD%KV!erkmA!uGm{uU=`cuBzy=M1uf`or3<9|F(}2-7r*aJs-4=4(pAdClPv
z>OpppuCk3k$O97r`X8yKpwwWYtob-w)F(XSv66(z*9B$b`ueZ>FZ+lbz_|wm99A?3
z-3|3dx$cbdGG3+(wA{Y&Q|W1fR}flFq!?%rCN}2&5UKQZ!4T8^fYR>inf*v3z8W@B
z%%i&^&n4>u7%@H{iaru9YXg55gl{YRRSp5H?*@HTuPwJK2>19gd?KoRR;owaEh~zU
z#^B$NVzR{L$khON3*su%A(5IHDMf=<lH>tmLDA-m_Kzhhg=E!nUhI}m=Ysbr=auRq
zWZjwLWq7HwAficlSzAUeBG@C08A-rU+iRa&n22{I=ryGo`<tdYZT8%0YAA>G`IZHb
z+rvZwSCm&?XYdkzbI_hw1yfvBxc3azXr%`Oi##};^eyXi-#Dq&+NCE$ggnN*jqJcb
z7naeH#W-a+o1G})X|lqw2$OwPX;F8mPi?D4%ya-%<dp*=-q}cRdh8UNt<OmiCJqNp
zac;O<_)T4n6jGkkfan-LFWbn{(-{SvG5nh+LZ)vzar`U6ju?^Bvm;55U@%|kzWI)g
zET-?3!)4Zx7w_FWj*rIF1p`UWCXYzbR6bl-^^96fU;tVpp;*GZEGy9QDUTndQ(*Bb
zu*)$o$+F5kzV+?84rFYiHdNE3wgeHRmRHIPp22Yq#kymFjtSceNIxoJasuK|X@Y7D
z)5>5u!K=|Xo%Igo_*ZLSHAo;pk3&%C$IiTAxoUe2X!ifOS4{5U^`x$-mQGD1zfsjN
znP`|(dE5TWF(IPf)-}N$w8m14jG<oMAmn(_Q%hHC_Dzb-15r$bxS$mh?G#w8UU<`7
zR_`uEH$;uXMrcJ-4aIatd;m)suV$>M8m{P5EZHc9-O2wx)zM_tNa8qa>dyR$<Wo4m
zvR__k3_QqziFzT374Nt9yC=uy&{fZmJrLrx9dM&y8kUmbu5BSsQ{iAh12Z+8RDrD{
zMho=GsqGEE3!S~5@VZ*QhM(<r;putxb`b9&gQ_PXg1OlDa42{Iu3sYwJ>Y-~k8W>M
z+4*-sH(RpsT&+LP>F+QG1=2mBeQ~|DA(|D0Cyk50Cdw85=w)Hm+$<5+LodaN?hcY-
zpjGxcH1Wc1YtMrKq|A<dh8g|m>|+m+$GZOM!w!t&C?v?KDYdzKqv{%2y`#~cvK-<^
z;Mnr++0_;{!CCwXpY~TZ|Gn8oHi>@GMIKI?OcG4iL7pn4w)wkX6Z{Eo$9r!l*b_RL
zYZdsj-GiH~ww8k4B5w2Ze1SUnD{VyC1YuOhd4l|2ZAuaS@aF^+UtECXFB3NeAc`n5
z(;y9@N*0-o_tUMpVReI-TYkj^q^V*O-DGk^2F;7v@Fa2RT@(8JlWSiL4zSjmYfaI&
ztMO(~zCy$t354!Xzb+cu-(e@#i9aEit{D60_97qV?5>&nUvv5&#YNAv3Ylz}Z0M}P
zF7GDwi$h!#um!$*g)3feQUV-?k8{$&iYAqWZ&CT$XY7p$zYyf_yu;u)$HJdpp}(VZ
zqmAw)$=NWBeWWtHx@Q7p!MSPTwv)-&I_6^#Qe5sA8I>sCG)*6+f({=Ed`_e}JDP$n
z-EuOhjpZ&a>HUB#a&9ey$m%Nq?Dd<@(ANq9ACo#9Z4cT(o!$lw0ue=u;Jw<-M)9&X
z<<JHN*x)i>?{n#Ys7&_HLa?!@S5FLTnzd>ih<n0byA)d`mQ0_Bof|8?6}o2R1mtHL
zT9$2!Ey||l)PC=T`{cwnxxv2Io)24_k$vLln5+HuZAd!NLOb0rW^9_4I|qnkY_eR4
z757&P8|SRm7S>pjgWiHJK|Kd1Vk!y-SGS$x*m319bR5cht!JEd0OiC%s&-sfky?mv
zWVmMeZbH<>;gi5O&p-k<MNb(TNr&C;ns<(M=eQSF&*mYPaC<%o9M~g}R#DWA`auC!
z`2vdq{C(GBxgG<i7y-7@UMFS0o}{<{&YZ!cg`!Z`Z-Jj|SF9I3+P-cb=?L)V_T&%G
zJM(&bsF%MJz}8LuZ$9tN9^i`N+ww&-=^@V-V)OfE-9$~y&6Z-#f@6b3Qmx8-ik*u0
z16U|Nlg3)D_ym+G=6Mvo+Y?Aw_dSPZ(gC8r@Bp)zJ`}MXO*xs=F)n%eGu<LWSZ4lI
zY4}3&Sj?O7@9I?HDPF3EPk3AhfMX_#ujQ{@vFOqxE|T}t-L;9LK2=D-ilYg3oDF!e
zYoGa2>les^4f_~O%KGUUMhpC)yZ~&jy;A%EEJpp7X%6Oh3`^CUnL`RZ8F23Ozjwlc
zm;@E2^=v$4?XZ_7x_f?*U1t4s$}*U8F>!(ha<6xzcg#u#qT@uij)fdK4nNDE#Ml~D
zG!siPU}hc~gW0Q8Oo_pShx!pkMuPIQ+hseiyGTyvFy{^sUmSRLksc#N9Ov+H!`#ci
zV?PVzDYC_*M-nXa$&C-5aN4+0K)Mh>x&#u9i>~hKoZC$yqg9w=v0h%MnY6)RXk8kI
zoJ&%AX+d^;#<A;=D}F)nUC^k#NVCa2_ftZxb@J>_nG~C|qs9SRmdT=(gGA_D<R9U2
z%q^TK3NR(<SKClhD{5~cZ_ybpdXMb;%jNM=z(odl`@oPL4aLTd?CMTNfQbqk{$8c|
zs73tIovZta&7HOl@n6C3zofft?EkaP<yiB#JT>zF&>Uj4mNF;2I4G3yAg%}bf40;?
zBiDTMNovztge^rZS1-f9pL07Ba3T|3QOPOh$P<Q!m`?xYY|690rSBb$R^6Ve?RFo~
z%!^A=CB4ZE8WXh(r$-t0O74_k_I}g<yk)6@Tmqt!Vi*htgpt1Qm)-76>3uu$6ake1
z>C6TXpI5E=`SJLjuY5nh>pl2=|NB?7?-KHLQj}vBa2`$dJ9AV&jX8{0yJ>5ut6b5s
zbz3Eox)kZ9ST*@DcPED-u7QP)VJD6Ko0X#!fRx#fMngP&W#+0xRem>9B{4jQ5k5yF
z?|n{*f)uF^6&oyT3IggdIbH*jh=ayiX;;Qd46-ksyDoER-M-y7<Vv$6Ys|@!AK8Yy
zGUD4dN^GcvM#1Yp&2mGcCIOaB(iSLH%;cUxMB85Uz5FobNWfDMJPP1-RdK`3R7zw{
z2s;Hbn9SMu5)CZ|>3+UWSzHW|WK81ZY&>DSXZ$BILr~bA-TT4?G+QGdQw}?6e~f3|
z7HJ+2%6%dqNrC`$j6L&OAt|x{ga<7to4qhHFG%NEc&p)b|3<0oLmsc7i{HaR`ar5c
zN>1Q~O*}g%MLl{CUy24{e?2&+D277Ozbg0Spr<_<9lRB?i8#QA&BDY!6svGfbY^+-
zZFxIIV+69Cgd$M0pG6?~%($b&^97Kc+Wnbn7zOx}SA7CDm`^^FyFqqG_iHF6nY|eH
z|1kE>zqS8g^k;3}+O}<;+Pt;hTie#HZQHhOx3{)!b2^huGRgOy%;)?8ub)nyduQ#n
z)`KDYHi#xlWptb(U-K(Zv1(WmZw#SG;cU=+v^#s_*C_5kQz@$z7SHBI$TkTLyEDht
zlM2#;?GgxG{MBQ}I1eRy4^!$j87MfL%fRo>UUB8PHvuV8qcXI3>cHs0yW;|`WJ=-t
z%|cV8u}W!dy#nX6rhE_MEV=uCjjyHoBnIMrvA1$a#yI*5-yyiughzt7^qBL6TV9|l
z+p9k5g@`7ef2z$jTUv&u>0GARnx>kr($DM7bz2{wnjdhJtI0}E^PynR4}G#)hQ_N>
zlC&VduJ()+(K^0fH~DKcCwL^X!TX3396)wUvSuTkyFS2vxPRzH7#P(UF-IxIvfZoI
z6KDIzejy&iz!quYui27j2X3|SL1Y0W(5}eGotjyRVoaq$7n1W%-Y9<w-BmIp{W;$h
zaR1bM8(kPPWNEeDE|hB@SzlIpG}I(pYp7k(e}FijLc`eM>hGlkJwY4L&k0iZI_>ad
z_L`G#T!pY0r9Gy+>MUhkb~pV50%*dzC|z(|Re#I2QH#Y)l+*EF-8FTv1m=A@3xHas
zX~obXA)PP66=F5aa?H>^04g8y@)?7|(@Y2nH0oNML4PGJ(FeTRHvnlh9d?Pb%Bz_y
zT#XNP)9G^Nhk87ZX<%lz0|5nkZ=p}HKZ^x}V+iwZE=HGVS;I*#1TZX}i+>+Qxqsms
zwbRO)YXT4}BY{1Hx$M{18A(Fj1PknOet|C1(1A6j`Wr+F>Ft-$j-oMn-`R*U4xiQx
zRc9xmLG=k=7$Q=(wr*bkKJVADJyWf-Dc&v9U|>%$ZBLN1(zhLXnL{VF5}HB}>9+;+
zv(rB>5=Ff(_<8yj1xIVFC+289JM5mG<NP=eyUQku%PQtVwlO-&w`!J6LVuwT1PK8u
zgvIa(0GwF&!zE)voN3zX?_EphZ$zaFfqvL}4IoZ0+D{j6Dwe9xFW96@lE@yaf#@DP
zAXn*bDQ=yxAD-{%AdP;x20;X_G57tjJ)s=e<&5f8AXEJjcgtQ4lrB{rOhswtva%Wv
zqGWpDIW@nFn9#vqIZNCo7T~byeJ1aa4!7>GoSWHumSpU9+3q<wqWLKi-yDeNbKrCn
zA#q|KEiL!r4Td3by_o`}!@85mK%F}Q;&bCrCrI%VvR&&BL%&>)wb#94AiRL|FS^1>
za^U;x$}2QJMrk&kF8!_5x+GL+z?3y`|MH8w`?^X_T=$9HjzZjy3M=<|Dffzbta-Wu
zmpPuYvLumCm(<pB;JmYH;J_H21VTWvj^=VD6U+Jc9oI(1*hQG4mq{S1YX2jWS82Kq
zlo5_0Em7#sZB%*p*mg>9C05YI31jW^iGH(&WyV97IJr<D_kz(to$%UDWfs@oU=GNW
zVD%)4)^OZRZ_^YKZJ&_^xBP;m*|0Xwwz|h#M5ukd(oD<eHBu=;&+$7ZzHw?)y?i~2
zxyrA?^N$i~i1jRX#u`WH7lAuS!Gu0S35(&P*O|08xu@9>D-Jl^QgMFlk7Ip5tnwYX
ze_@v6>NYuT{q0A!00E~rzF7rW6enPgy0!A-BY>JQNY3Stpgp%wYRX+-8?O<u%-cr^
zL?}H5UA*`6xKf-!RLFts_h_K3K(OBkWqEdSiS$0*vG7s2p(+%TMQC^;1RXpXj2qhw
zhh%1lm&5YF%G^3W?~+GpR>e%TJrU^BU*pIvCbFx613jrd12w!sV!~{XQegEhwGM4@
z)V;4N1(}o4x%klWV$YI!o>cZtqm{hLtvMnoWg8CVls;||P0{f$v;wIVHpEiJ$=6xU
ztYk+4)w)q3&wDJsJ$Ds<%dag8A&MHR;j|M>(qJ9B+Dh@p_DInWQ2+p*<HbJ}9-NI^
z3DB@5EBElm4n%Ml^XD^ZhHU*H$sw)XvS8dT-J*iV^chD!xXu{Py;D^SLVRWqkdlb7
zLSnrv!4T8!Axsno89LjF^#wvp%aC)B2RYrB{0BVZ<8ZUTA40()hhfq!ECmuo`=bjs
znl#WJQYrMMav_l8hcd6GFXR%%wCi7r?8s{=+mBGs(rnMXx`~Q2PVqh6Y|eGmP9)VJ
z<wt+^Qgj@P9)1#*3FbtT55Xe^YzXHxpkWR3c`pp2d6yK|s4^YDuiS|}VslDd_Cvc$
zq|@TjPc)U2pFgdPMW)q6xRh_u$2oy(2Or4-qaD!_FZz2JtS4I(Y+AV99R>P;rI8%F
zoq#Sn(P;}f>9s{$P(bc1NvC*#gb3K>41jJT`q`;2ol}A-hwN>kfdn!QRp}*?u==yk
zfX$s(T6XR7Z?ZwSu3H+ui^)2=>T<EX1PjUFA;Z19a!5?>OoC-1Dr)?272j%$G8%fu
z#Fun$q=J}*YjbE$04WkARazF0DX!OdAeST_&rZ5x2vL&KtO8-ks!6gw1>G$^3vb#a
z$sg_XZh5ay&f;N|BqRV?baqO_{g7}L6*>%@f5-muzIkoL0vSPl<x&3j@jNks5TtXU
zNRAO<Y}*6)cC=J;#&aO%gQ*QtgJb^^Dr^gk>=hwF2FBhXFzFF%)rug`H6+;n?kpuH
zf#owI1iFbblpgRMLh##+li7L?2p(6IUY*V9VNh8uhH8-yDb@<mJQH#^HQSY6zZ;E8
z#GfHR{zdx-q`V5+E>K=o$CrT9(!gYUrK=C+8g`mC*6?Zjhf3wddV6-H=`=Fv>tv5b
zP`uu=FTAf6BOg%}p2gOIV@3<s)WeQ1Gw7mm7?4t!aB9MW4vCD`yjM`~Y)+YS&J3xD
zhnmf|ll2#XO&>cjnIxJgFqR~FC{qXZQU1M>V+|ov96}`(U-G<gWU0u)`@o%8Y<mU~
zE{=-*4a#kA{pc`?FBrYzn^<d0W(c0O=3=ZX+D%i9D0@RS680<Xa8d?25vFxVg6h$D
z;QKdN_&rqNw=`5IA%^Y}!DqIe$**DfC$wE|r72vAOodB^<6UguL6{zfW3_K^gf_jP
z|8rsfq5Aw637O-6FSFa!kak>WN9sOTyOEHk1tXM*wMw5RMWL{!aMl#>S{!pDgs$hx
zlSnJP_)$tdF*eA^l;ZX5*V6E<0tjnkZ1vQ`c+3!7x8QW6A?5X8@p+gcl1egfH{!;j
zy@px|IQP>W_aJ2Ib}1D$KKE~4Hcv_lmTmRm0NZ#3-PqBcejN<a<~RG-BQ;z2doL-j
zH3-|?1U|Um9}i6o9Al7~_E;d#=^msP>A)23nQ@==-iMK$v7&YTs{4+C#tR>j5MvYx
z<j3y(1J7mCc8qyR3tpxt>$FrM#12!9QZ@GEc|+F~rt6USHx-54;(vp~vY<qM&xc=(
z#e(Nqic~w<o0jiN*kAFxcLPKP47<b94+S*{5q0()vr8;|RGAkuhw?DM<i0gBx!(El
zdfa!a2X1!^2?ZIipv0wygY9?^S*nT6+A&w$%Rtl#xUvcAk+gFer@e000Y#)04&<bh
zo=hLGq&8=hxCs;<{oB2p9>mY7aAd?>3rSaQ6Ulj~tuh-x-xRx^Jho<}s4~MhU@nA!
zE`hP~YTNAA?d>#UJq-C;!Nz?5S6O>R)oc_!1oD30IKl3$w0%v9t&}&&hkpfT&p<(5
zn7~9I&Wb1J{I>}OF|HJzO0ajOz0^MkDDGEzzpiS3@Jh}DXbBBGhPU^=^jrG8BKn7Z
zm10WUMTW36sDjWz!I6W(Mn-P$LBbRL@9VX*oj76FCT$;{vLSrLifP%RYm{z~yJCvP
z?|-Ra(&ZQlX~idj+~{C)Mvw$W-Qbzz^Em$!fZSgE))vfoiCxc6m}YGIn)macXcPF<
zfvAH!sRHnZU76JNIHhN8PpsWvT*ci4>z_Vg7QHy9tk7;=qjH@O>Qi-cj0qHDkDu*l
zJVufu_``n#C-B8=@q1m>fW-&fHzLe7EI2fAvgM@wR&#5L;jU9xJbyS@WI}b}>X)iJ
zsjNzYnHK&BA5THK#4JC~JR`SoXt)*pSTj}Ilz5@1%xf~4&*)b9B6=&~nvi^yfs5M`
zhyl%)KK4XJ@Zh#V;1R$FhJmt+uxJgxx<k^!f5~<1C7UyQUI4bQyHF{JfemQbgQSH6
zn04kB+ljX<J#YY<7pu5V*oZtgpk<E)P@ertJBm0gXFimj1jm);ifL}u+A?$!zrt@c
z7t#v`Jt0bRpr{x}D9l(XR&rO!s~sESNK{ml%sy?r@-V2cd?00WUbmQw$;WT)oo=z*
z0N2i;x`~=j{)vN!2RfZzDmo_g_%#}NP9ZeCQfFx1T2IjG@MR=%mfuhdva12l>nw#I
zqwq=27fQj=Sv6O4kfzOnM^BW=O*KIL<$m}>3S?DU)E}c5HW+X*0#)B7$w<KYT>oYW
zTwLL`KMlM!H3TE4K2?&jL4uU)8-vw0g5B*5op)m+;4XW&5oJ{og8}SK-|mGVaq_R&
z;b!9|_wB3jF&DWw3qI0qkQj@Q>yBY1;7imM4@;+e(ixgBqKnZmNaDD8EW3eNq70;A
zw8JEcHl9IJY?)Wp@%DZwH%VL)f;_g}lHS#4C^3}blF1s9+-G@R6q8y0?T@XkZPbNC
ztJzo)QMpIF-n^BO{WvhIdqiwL4QW2FmI+H%6Nl1H%z|n;Rn_8d-S9eu#6^>xmZQ53
z{|JRcXrD9-s!iNs1*_~y0u?$L!W_O!Y%)iODQQ|zh)051Pvs3Cn~8iTM>rJ0v`jr<
zRXcqzJxCV4xFimqy+T>z5`^p~b+3GEY?U_8iKwsTKCOrvJSZ}4&3Gz=WQ>HFwNgC^
z0%MVJe0V`0WD=$2sILq{Ts?WV!9+b{Wq!;?Saq!#x9}Ru+Z$H*iS?&fYh@?N-_to;
zW`>Ct3u6m{x|=T-sU355{4?=T4IN7$D5^<qfHV(xuSt`yXdMh_fpdw)OS&Fg2po!C
z#o0SzmE<=XH%*?7hJLgP-Q#Rl;p+nm-HgRJE4N)!ULZ#Qs$V@CW_gPBavfxXNjGE1
zf*QnaMrvfdHV%yari?-RS7gyZq6x!puQ)9OV`^HZ3^j~$0nQ4?AX==V%B%{U@}gvd
zwH73bJnml95%)C^;i@GQ6L>UJt!jqsjdl4%d%cDh(g_$Cuq<-*h9hRtl$FIrm!b*b
z3{>CdQWzft5cFN8iA20@t14<O`=SJrR?>2qbS$~-RH9P08N=&cGY|#1Tt3bc&!ws%
zpNM1ci1bPp#u!F`EooP+<m47Mq1Nj0i`4;{MSQ(&+H9aB_}{2}-O7^|pmO{u`Kd`(
z>&8{&;oR;Xo2wdm<jRkw4qxgfGW)U^lxCT#@$&CKqJPGtawu4n6u4pPY-r?Wex7Lp
zkQ&UE{~VbAUrOHpOTlAe``;xdbsgJ#R@Cn6nj4ex4i6W@<G@>D;5B^-6~=%dm162Y
z8+1wuiKHeE;9ooxbqm8+BWqUsAsK`A7moM>$1FrA$jA?uJLh(G?gO}}JH1UhsMB%#
z#M9a`nBoSJ<jJ;(*8|rKwv^D^6;@J=d2q(85#6s>6$At49lN7PgL`hAd-QcSHX|*&
zO_}ezPs=wc9p^cJG(htcQE=VZk4HN(Voqi&I=VZm2l}o-VJxBZN9}=lMHio#3e1xc
z&MZa_Yd6H@dh5jbRG?T1J{B<!j=~oG-Ff_AT@T2ukK|T+A(*{$mz_096hhME$$IWp
z*(>6CY1qDcDOpc0_7I*xdYZ=c94?`OX2XDqqGQAshY^4*o+vbmEC2~{(qTze^QRPw
z2Zy#n2@9$+m8F#VO+<6)8#zSzbG|@J*gLZZEWG<WJ3n(pj@z|Os^}#~iWj&f4x?6y
zL^orVnEWWOMh?RK4SX(d`hKqzu!8ECs(_$~LwjD8-tLBTEolmNH({K*V(C5&fy2i2
z)+xjP7u()QeV{dDfLm#JSeSr++_S+uev_u7Ys-!;k_I~W30JLUSisg~Nm!+Vn8#9M
zWPpCV7Lr1(Ry}L(gReE3_=sN6@eE81m1HeaWHOqtyE75jC=!0;L_Vb6QdL9)IyyUC
z{Vc42ntNPUkg4J}CkMhiV6)eo3Qt|{fxQtwHsV906RDT}DRAn|5!Y?my(jS@D&u0a
zxPCLSdyDh6tjQ>5;&SHEsLW;&{0`#v8|^7~5t8Bf`>)UmaA0MLM)p6IOaQkD?s{gH
zO51Q%z_<73+ENAwwf=?OC-$?2gv@OHG`NSD%#pa{f?bz1^o2KWiotXj`a11QIjPp)
zf)d4GU2BVo8u%=F6^f}eIF4_-Cyi5G0S<2$l<I`2eowlLB`4Q(u!b(L8Ow`KJ+|MN
z3%uwN`&oaZtQi<9u3yct%XS3HW>#K*^+%goRpsv~vc4i;?si*asXAAxm8sOwy~|W<
z9>gf0v=#5onBeK(sBf23D2h6JX|ulb0RiX4rJLxJ$u>1X0!zwH@KR0iX3!^}cHZ~>
zs{l0z`>K*dDnhhhY+m|AX@?PYWmwk4&3>4&9K1=k32agW`VF%UQDc$a6zv;&QSr6g
z3o64`qsM8!Iy~1f_2<h5vnN_Vc>Bv>cUDsQq^6e`h)hPv=!G<dBx7X>&^^9asy;zM
z32IO5Yr{J8s=ZY(B#huUL-bgs-4#=>j!@ru&tJQ!++A+!PQAD%vYDg5EL<8j=Rie9
zA3<6^Xh^rf7W`sB=Ys)`{{)>fo2P@hN?Tig1U1?6h15XJu!F*u%wVcWMSq}>e4JHY
z38p;EIO!B@n^nA8qPR9B0*W9v<*Z+%LOpNNgs&u36=Aey{bF;l!>5D1kMeKbXY&Md
zZR@y@=@*cj!eHaDlCgx5bDEY&(RjaBxpy>{n>0zi$Mhis0BT%Q2_^N>F@7o?IxbvB
zRyAKjnoC7OA9bh>*Xm=qG~mn!58(^k6SIZ=zuq*a|6-3Yv;WWRk$;*$)UQ!wpXHif
z>GIrgrtEi#d==}Zcvp^SS0k3B^3iI#*dP&7LZJkGQm!#y`5FI&TvDm6iz88>P(D7s
zO}!k{=xP$&F8<engM&lxK62(!3F?9|)j?y@|64XUXO5va@+ak%4H~@Np5FG(@p^yS
zn^G$YA4#}ZD}hGi{5rFCdp%;f*6reYzV?0l`uCdgTtYyx-PO5MIidU0VI8o5=Q>3@
z1&wgm*~wT4$Xr~6R+YdxefDZtpNyEyX`nsUsKeueBq<=f8ASGHi398YY18pHuLmPf
z2;LXmAF^?ja>C_go|*Gavi+QA*E<qzzl~7E%8sbrbv(jMl<a=<W0J10x}w4@!N@jm
zuQ(GG$3tkrsbBZfB*?C@JhC`cyI`{G2qTm^J0A_h2>@gJqHY4J^&c=&n;JJU6+I@;
zE~4q-o%(Rhb`A=fu^{|u{>bmh-+$uP_YIN#;Qls4Z+)E@n81$yZO<iJ;vT2wx`l)~
z5lx0&7c3}(VEoN$*uJGo^(*M%xv;{WCU`+soN1g;Du`lIO|^LqA1PvhsWZfYi8w?=
zWHx;#I8=*~JNo#wdF?>|U#IqK5)EUek?_t-GAxvw^T1IxnW6;co&;FFqA(K221=D!
zkYX4%x^q~k!S04d?1^cDnkqVa7W}R|7r9b?p<{HArfkKC1i44NqA*DS@`64VMFv;^
z!s=?OVjmG%*`uF4N&pj7mKrR>>6+0CKcPjG3<~G4-#}lN>WCq3u0<bnF{rSh)zFUU
zkD73Mlou35Llo=X3kNAhf=Im=stnUI-Lme$=E4pdf(r+(0PoUnEgu?tVwm5>i4HHl
zid|OJ8)mMo-1uZ6{}Zm*OgEdH#<j03h^8f|?KK)d45R)+9Tg~C=!xkQ7Bocy6~~uA
zNKuAo6trZ9P>6=Iz<p$7+$6?IQVX<bGoR^mW5qWV3FZVDVIL$@VIc)S=eZPTn9{@0
zkf0`UGlF4q=iPV!wey6sMhuXJ?`03`lKNXpB90HDiM@Aoq%zX+uTrj=^NVQ1l7YL2
zvvI&o9oOi9t&+@eM-+Mv=}(z}i_*xuM%yj*R`tX`UOk=nL*j4o-v<%(aJ)4;K|WF^
zB~d<g{_BRI92`#2>D880C9VbdN8U^ZaQ4WVMtm^dH>_gZv%ZZ3SFiaNf!9I;i6ea;
z3>1O5;vLMbA!UjX(R22!QBb{+hJzMRH*0Cwfpz}Wi3vRw%Y-C9E)`!E@@yeY8z^J-
zvVmpNjiWO1<RdjB>}3g-_7Zy!g?&h6%K<P%lr29zZNyF~Y)CuWXW^G#l^_m$83Z7Z
zf#i|Fb|T8B`EW&J0nX<FOz>P~1E%3se+71Z*t*Eu9N^bb1DenjUap)>FYdAmwR|fe
z6NMiEh*pJFD_h%Va1OT{%#GO8$!T9ACC>kAQvDEP?ma=^oxU0|Dm~?f-8TR)R20_Y
zc1`ieTu(ndfDocT^V}pLlD0y_1<`LYgunQSG5UqNSLiUgIA;^_PpI|di^&FOjc3iJ
z0)!g^;_tdmZPetHH+1)K44AU=);fo2sX>J*fi$5Uv!9M@&y-~33puK;8}B-utEi?B
z$w|*ii;zpvl;2gtl%ETUiZUsJOuEXZgTzxJA&D3f^9`Y-BUo|wPiG{=I~03ZZ^_n?
z{SJqZI4P;(!nQ*UNiF+dZz}N<T|q#}#9RN)`ys~@)aEwx8BINoMUEm>%e<1XHqR{2
zdLBe5^j~j#mVnAW_{z%hDQ3F68H+#vsMl#fGa%tjjXwEEP$kh#IT2O-f3H3#dOx|8
z4kNOEY9O<16va$ehMoJoLgbv^Md9dm_bDw+N^mFOzkpH7-=B+ur*ug2r7#XsdL|^E
z%eGe@Gkm>vOX(o5<4vg;q$)2tW~*wztBSHM&LAyrpmkE_xVZCpgQ$XunzWn~=Nt+#
zO?))pJx|6GOaohKk1)I0R2KmBx;#Gzb-E&<{?S3GD_J!lHbBo+uYZ`I(}3G6otP(x
zs>OolP#UN5tbyuhfxuKgx5;)Txf=YIf{xL?9PO`69G|XSq==ekO%Nnlj0fZrO*$F%
zZ63mSaQ?OTazS^B+Kv8-A=i!Ff`R&_(~j5b`db-(z2*zejnAuWH|m#vYCRjVC{lS>
zu4;SRH1*8uvf1;N0%PIYg<9hBG$e*blIE_@IS$=epIy1{ux~ak6Ie}i{dl;wOO2n3
zaJa3^4ICf1*w?R%*Kp#A>~av4Z}s=mC!;!D-ABcH+i%MGL#OPe-L*$%yiIaY`i9ky
zaygM0unqi`VR_%exSaj6K<z}oGAnt2wI<YvR}0M@cjITB#lPLR3UcVgj`FL?b1%jz
z)6?k!iuhjHP{jxv8P*?5!2v8|SrX{|$b$;qCPN**ZAbS2IM=r%%-w7_JnCsrJvWig
zKR#ZAFM+yc#j11V$RVHj=|_GeGBBFL%xeJ2pN9szVIuT8qelD0ypUgUF4_>zbFCt=
zcV#On=63qF&35L(`9!?wMs%?XE~T9ERzmrt=i)V{K(9^kNIqMw*puWsR}TZz{i4F(
z{2y6kY9Z&uS^xSPvywtjFWB{}BdiusQw$v1gEhZX6cw@cZV_Rm%8~r}Z_MZ^p&=v`
z?i<5}zvXFQTt{Q{^(Tn2s6e~zsxfvBSH4()K-tdvg82Qc)Kp~0bA&vX2&zjK$nQ~Z
zD;SI9TU3|e&KB~t8f7#ypEt*5F{9w@H+pdG2G0$~jdu$b`lSmBuX}bJT{1L7Zwb~k
zVhYldMB}FVvv$rk+O$;t<=aQ++kEM2c-@m2Kyx&v_~@F8d^?_^+%(Z#eGEFL>t9Dt
zf|Z}OQc*lxn(>M)eeY&$8Z<sTI4V8yuW%{|YxLA}{HbfYRyH^9Zz6d6$eX{_1SPz7
zeaM0H-8GByw5A2jjP6^OVt`+E<aLC$qem<Q*7P2yuX0Qop$T$@&K-*Qt)52{A=J+Q
zP(|izek00<@qdG>GsqhL=c4o<=np2A|6MJ;RNuATXGQ(q>go5AHn?4lz8s7MfHd-i
zh=3PJNOV|?Yqms4B+yH8v4i>Y%+xK=TP#=U*t!aPm2G0YcjauquYQq%$s!BcdTt|l
z)@<DXQB{&Am~#sGiv&o^v<eO|aX%Ap^<aM8oWYZwZBdFrlP3noc+9$P``$0y98_Mj
z2=V&AcgGHTI&yR3Wvj2pl=Q&+^6|V5y__LqbfTmg_V>ymiTd{v<w?v!DOb<hK`<TJ
zBhNS$Ma+X3ifrXvn2N#NKXu}8WHN*028JoI@b?qP!+$Khi~`K{PdQccFVVHrG^*@n
z^ruZcwKJnyc8hS6#{-@+ok)y>NknsCG(u8cH^u(=0Ybn}56(-kA%i^rHHFMmoh3PL
zmSp2I96k9er6;@9S(D$Ej>E3cP&5}EU}nztN>l%F%PB0o)UjAEe)YUGmDTrvucYEP
zj)lS%(gLM6vrj+kpNkEDHx3R8tsEQESC)fH&NCNFHC*({MEAya=dQLy>JKR8vCtMs
zw?QZ+jn22eC5)S6-l%idFcYF}0P5nTb6l^uuS3a2h;j4#tI7Oo3Q8JHg3k_Wmt`5C
z+PG%<%Y3K+33l3gIMjQH50nwA%grh4L;!AF5QOUN41lO^KnZMYkhz8y@TggGKR;dJ
zU>2S+R_D#AVO&5x#&SFtjCB}MX&GStX9B&Uf*Vi2vdqNF{b<`ijv8pi03Hz>VtQG6
zRzsDv-3feDdo81xa!+f$>T?M|4NpZMa{(Q!IvOOn*_@gmil5cB>9{mj>$;#-4N@*C
z3$UjIu%<so^2LG-i-$oF71N3YizRH$yyrWWF!Zqk@hIPrfbf)=a5*l620HSCu#fd_
zB5G4kx*0?NOKiBny-Y0kY!CsWIC0qjIE??`1UkU+fkeKe8UO|=I|kHg&ds(lKr?x@
zf!Q2fhY`xh*j=5>x}0R`UF_r|d;^th`2pUYT|=D*+R$Jg8%?$~u4}SfK@#f(kejx&
zlF{_%b?$EyZk;uR)+9FLz6kL!hd|(>5sGzJxAMvcX^7&Dt6JqH#V|5%F=*@NY@)kq
z6L_qVmYp|Cnm554sqkBaWtz!r;p;;jvvoM}d2^t$Oo;S$uJEnYpNFRBPQ=#ypWzd~
zJs%}(+#H~)(c3jqSC^~?>S%_T*H^PZCrKtPN^D|4eWuek+g81iLJ23lhil=qQ+Jze
zY`E4{L^G3WiOkk4a_qI+#MLbL-!@&ur#4rh;zd^@Ycf@+qNZ7hhtg_vO0;DnU-5{n
z5>F~E#YR+>evh4?v?)<F(rWi84FHM9dtcBbQRSv8H3wCNr{O89S>Xtw4QY>hS1&_7
z)Wv5tm|bSycCUH1G{tz(ytnAW=hh}!b{+BF+Gz+3M<WDzFy&l_c-%D6ZTPXO-KPtd
z)x_-MQJd)UfxM;?zw{`1{<hR?`omTmwT*h$R<uUTyso(PVh;rW1}B_qiz0J_VKA6Z
zY3JP5rLEg?^amWgu&U7qQ<86Hf{%@jud^88>pk7W&JAxWr>a1}6B*AM$_K%#!#s;0
zBDly~yox^Ub+6r1XXlE3NucwoY0oh^#<{)NKEa_(J8PxM!=)3`xyi=an&31k^&iV@
zGuapEAT91>e0_CtQt~&!59To#tLd@UWaZ6tNHDP?*)BMKEunHKS4{HLD5HXO!b~zV
z9gl~zo}Ka!8iT5;>wc>Dk^|FBLj5GmK2emJ+P)iy(9@`v2^xm7PE*TSv#rqtV&j+8
zUoV>s)|~m(qAy3A+&8?+(G*-*WPG0<XRV#x9{9)w(Eog(|0_(&{&VE|e}4b}RMRMR
zX~zvtgdd9LgY_Y`7IIF!rM$cQl2(Qc^{tKd<s|;%W@83g(;`XPK4*^)5QvJjHCcAV
zAf!Zqn-@@O)s@=nSqesttzKs*x<(uI!n{;RdzThf2lHBS8hge1<yp&Mp}U8@iB+=K
zhtvIe$qa_tyzy8CqHWpj)!W-<#(KtgXTjCF8b+I(&Fy`-82|Q1%&`ss?DyxCukH8e
zU&!t6x2Me>e!Y+HryPcwrt~|5dRAqwTC5dHxcSb})Y~Kp7U(J~GC$^~B6TdqLgAE(
zb-vi`tBs+gU(Bz36UNJki>INMcAm?2Eo)Y5Ml27D9e;SN!i>ZI=yg<E#5iA;by#HN
z-JkF8tcQq&B!8W|tlUayzi&EY<V&RHS~ib<MgEC(G@S?e^4~AEu9k~1PNNvah0@E2
zrw=OV0_AUBto9x#=q~w28sl&7zK)aESYw&Ce?veRPjHdH-!G4YPSw&^PoZX0`ZX}P
zMMg`ja#wqggXBL6Ul*^na_$e&-99zz^3}24ucKlS|7Tmp`O^(JB<03qkFfIWN%gWv
zZIY%EF}#n9oOx_k2X(A&HBtm!UT3ETscaHvuO(ohNy=k;W4V;HD)7S(*d(IR*1}^&
z6=21b^zH&>Y40tc9xBO<o@z{;w0OzKw$<tGDeGuo66w=z`^7#Bcj;=UA5Dp>oIF<1
zu7}!2H=n;aCkBxysK}gAPpMAHxY+tmiFg7|*X!iWxwvKRu8N~}kIe{o1AP;^<13t4
z0fIYA?IhSIBVnXOcgICRa20CRXF<^!s5F1~0pd%6=(*B1?Cp$K=VmTcd(Lf6JQ2U9
zs<#<W8Yv5ZUN2C6tSLk={gdU&YjMdqTPg6V&n*oJx$dw$$9v^-!U^4ju5!_q@a4|q
zQ+Y0>T!W0<(hDmr))kGEYhC0{V*jB<{v`|Q=yUe81PAZiX6uU^y`RG_W_-TG_LB{3
zWEIQzYWSC=Ql?}XFvapy$0lV=I1rO%%fp`wWA0C}xr^+&6ZKV5kY~-NvUt`fkX9G+
z<@s&R>B8G;+fseyTW#OzU94OIxA}E{3ses8mP&^Bv7P4-h^)fO$lFL*@R?_gnX*47
zgwN|&(s1e)uqFszEpv(R0GJ$`lxxEQh$JT$W!|gZ!{q%YmR2PaQIGfk8pm`oE{tQr
zj)O5GnYvDMAI3m^38*vM8|&Q)R{g8aPfSl|Ku~Zjy(@A%`L_u8K>6Op;(^+4P|>73
zu`E7-Qdc|%x*1zRI0Yy{Py5o<dLl%`uaYD>l_hGow{$L>vY<&k#*!zf90|PtBk=yY
zW5-CM$w&LfSJl}6zL%0@<#j9=ipSsgijy^oLaEU=%!`q$9cVTXRvXM{5Wy1|DG;@J
zk$5)f@N?(+9}-7}aA)9e&k?6%+QYG5kyIIg_Ix8azQ54qz3u1edy1-|VPHnUk#;ty
zimAtFXrEW@mILn4GQ`rHQhm_abbi)oE#xI52A%Xl1R1cS4AH1UbBg2}ATJe|g69Uy
z_}8~<<U3Rwb)><dNc9Muv8mt`sIj?+qWtUk0|8j_oJjRD&edkpYtGm0HAIWxXgGM6
z3NiTvtLdvm50bqrngw_YES4@f7C?sTsLJAV)Kj!(e~#753)JUQ6)XAR1IT?qq=kW3
zMjNWW)#LNXeuHd%2%5&J=G`~?4K>Je0e=!rLj56<s(8&O09t>j#EsO-p+lKy<)$B7
zP)ff~sJBv7(4dw=VfY%^LvzwRG7K0p)bNQl`4R;lhYN{lkY+kUqfiOz6(w=3q>;N>
z0r=3xn}MKUHnT^d+mZM3PW@xChkjDqORHsNvs4JK_3&|nl6HT>=#OWVzZB(lBn~2z
zs63h&t|gu9%vavQZ;ML^m$Q6de53^!jY1T%BM&td)q;Lk4GhB8f6_fJ$2^37fRXc3
zO*&|91+o(nJvI^Kiuqsc#hU^=!lJgW5g+&6k4k{H8u#8<a2bac9LImxeim32#m?eZ
zHR>W2f#*#~#Qp+?Iuo%|O{CKNLSrN$+Jz3X?G>z;R^~G2%ui#gf{`SK2=hF+Ieu%P
z(GZ*~=hAEK`zg2uHw`vnf9n6b$r&-lh0-7mS3a#X9y}Bg>2t0}((f#ZqqgZXikC#q
z^~Z(@=eDwbF_br|nXL0UJos=jDUYKv1V!Y7muWrTYcGmFRAH2dqm19UH5^ZAGlV=H
z=q6hB=kkijER|ffqTR?D#<j@v1INrw1x*@}Oy`EcHGj0}o#$?-L@JGS?R&7FMjRn5
zCd5G|D<wiRYR>zdb<s077$Q;7^$!L2?R!<=-SyJX){z|^rI{)eOnn3EB?>}&g5I{u
zI|!C%(Z59;`yj(2kTy7@XK6DOI}y9qQ1hNd5f9vwou6KV-yp&HJ1C?Ln<r$Ek^Spy
z1RRpFZ1iICeEJf`Z+v+10YbJnd2ESAA#Pe29BOsh97|9s>MUqlWKR&?3rg}NDx4ct
zgE9ENMBAm^`<0Yu+?le=E*wfoL`^yw9Vl>Nz6+MEAmHHs<ecuW?kY>@Zxj~Dc{-fq
zkj9FHMCxWjL^N**DEWwsQZ?MCZQLBqD5}hUQs?VGm}7LnbwCt2y|_knEMvb-OmCEK
z+aLq&y{`J$B{1&@!-0pwe%xP`NJj3?sT((_SBf$JP!T9wV0}G42hfw&T3sJU9Xrz}
zptb+aZX-%F{LWV?i<*JF1ORG`3SG@#j-qvskzmWyArMCp&M&e(FrT>cOiCEY2798e
zU#+3)56)mEf=dt@x0h_mA8b|%F{f^Ck!`%-j|JGU6vI@~pyalxow}rzXnHs4^-X{>
z%k!2{d2(?=y}dqHS_ka@u>(^WWJd1zJUxk>H*e(X>9s&|=&5B^ML>t7+u2UAM(n1e
zrOX5bAh4hFr0*`1CoA93VFu?eAXxy(WhHEa|A9EhJ?|C#MH`I)Q8MW3S=q910s6tx
zx{VbNW3-^{KqPQ01u%qS;Iuk5RRm)xM4K~bdRP#QCAP)Xiwp>vObb!+nD_CO!iRG1
zQEU$?#kU@kds*iGCp>uEf%F<hT!!^`M%ar0+>pDAbQVWj(&5-VH(yeShF{+wuO&OD
zz`zaVc)4>z;gSkL!fxoc@C<32gj*h%`**MMbdpR%7&Yw*2~hm^^t%K?kq~404MI?e
z5*Gp!0Jxm(Z0R;Q?T^BMLdURmxAr_xD05mZAAn|UsP0@t$ct~of%TFqEbJG|W!ND_
zDo)gUyg>{LBO_qV9z;E(e5;p^3O`R{8$OyQw+x4*d|rAKRq?(--8b4Xb_*~@UXL68
zu3Wat<5lHXxJOV`a9Otx@utScpZEt^LR2F;Z!MRqQ4L&jw8B)n-y`?0o_5zr#HSqr
zbgbQ4DE=eh>0ku{vPUR2sw9)^NBVt1;N(B7ZKrvHE#;=xdj>09Fk5@^b$#EHPo*TK
z{jc4VndSd%Ze`+R`=75UL+YEhKUON=nV;rX@Gz{nMaW<q(6u}o_FTMV@D;!^<Z!-G
zUj2cT13%-1?{=JPBH5~Bd!v>f?U}^Z=hgcGWh+Ad&XgppXLma<*EiKM@J#BCB(8Qf
zNE$@lxJJBLp9yj&qO!}|?p?~R*o{wLfqIZ_)RjxsYHGWHP8cbSUW@g@p3*Wn{Z_gz
zgYHs$`}O#A{m+Mzg{JRR3V;7br7hGa+?MoMcN#ojkX!a%O(|};=G|}(;Y?;zYS(=>
zSx%AG*&QyXk`Wzo)^dRut9VMRs9oI|9`<iWh3qscfgO7#Obwlasd@YZiVI}dt3e86
zhH7V)#%ao&(02%NGu_trPbR=lye~!9mlfTfynnr3kEHGoJA_4X@PG!C3urxf!Te1O
z+g2<(82)`g{p`A@!*W+LY4`d`xx*+)as{#jqf_ER7%tIo{eB!8VRD1*6*DPje?j_I
z@w+6d^FJGy7CM<4O8CiguAHX7OL!6cLU>sx@Mk#Kt}NPZK7&vpu_QWdAX6pd2d+PU
z<xxeK_B$pIohnkOlfzgi6j<+|ftS?^Z)@XtIh%?7Rxk!XtjZ<-t>8&h3Q9BR73&g#
z5cCfnk#=Ig4%Gy%llfUfSs9hVIjYZ1w9blfi=%x+Wd{@+=Yxc`wQVMsg)|X$cN*Oe
zT*<PjBKBy}W$d3(J8DN!o{BK!4WCsSMWkbJ?bK#7b&83=xRCan2)9#=YN03-(v!VJ
z_g51+oj`K8hZi|wkenrvy&8r-&QTP<SWc15U%0uByFzbw5Q^g6=J{tE=(MKuIQDWj
zdMW?LRl4QY^6?mm8G`<xOC}l5VbwC0jX=RXO(CZ2n0!DnlQgF)cBJ@umxrk9SrJ3D
zXoam#VVHZ<x~g|u)rn=EXH*II0%Gv|GBu9v)KPh*H;{o$Sgm=qM6sc}ktVNG=fS85
zW47{`{WNNus)>;fzRB*4qlv!fAuTv;K~}JVXm4<RZ(QSuHJ2g2vf+x(UhwJiMj|*l
zjNNQ2fC+*I@y}JK`|Ly}Rg%K2JRk<91Cf8hy6$O}UHC6Jl4@%K(DvDHTN&Gvx2S5D
z;tY0;i?Pf@N#as@n6W}S^p*J6xMEz<T^euO_W9+gsZP-h?}%X)zKhXVx_2i@_3>R~
zQJIz1-m|ay5F37UFk27lVzC_eq5Thaxr2qw(X#2j{$;K(U?`o<oV(eU&-_BAH6w_A
z0}sO?kZUc1cV4OIY!qrEcV0;nrl-(McpO7LwILy$Bup)rl8g4;RCzkAFD$aT7zN?C
z5awTddY!#p5e{v`y>#3?))ZvaN{YN7S^lTs&aE6Kk97bO{|qqX4YF7mhGjBq!eMzH
zW)wv>F8^_dO2!TUMjM+kuW8acZuZH+eUZSF{>E7<SY82Ywn5WmX5!$m!^JIfXvT0i
zjFSZSSQ_#VK{tbHEl?MmGxuUXq&4w<cNLZ3NcCZ=C_OAKD+#_N!Ox10`LmM^7_iLN
z>ajvF&=W8KnRq`=SVG=(@*A&zYnaITqq5LT@vWlJ>Ay=tOiudfNw^<t5+HyZ+;mil
z%Z(eY;Aqcfc1F;)ntLH1!(S#2_92Ob0P;QZLcHF9HIVg}GSD^tPt=Mx`Ov|`tc(XE
zo3Oqz<zIC#2iC7>I?XQ1l+XPzV;B>KNg0Qhp13;F{VTD4unPvxh<h;53Lj!6X7FCK
zO7HTuH)3)8ze?y-*p&Ll>TA1S$SaJ#gBr|TF5h5hF9wT7tmE~+m6vSVf6ZFfZGCK=
zzeP$R@`tnrhC~Ciz*Zcbt{_1hw{w>-1nLLB-$az8RI_@YK)<ivq>j>t$os8cc9e)v
z_Z_)<Nn%~hsZP11dPBob4yDRSH}GM?yU}YL2|#~hp8cBRVAy=^{;fJyW&21wBN7jG
zZ{~@|+vt3Il-8oFm+i%kMJSCK<{m4;s9W6U;Al9;EW6VR(<<xO7q|pgk+7W&JO4e_
ztM3!MN)(zSvR4Fs81UtXgaiqroh<bR8elft_(_0NEhbw+GJQV>4ot!qnTCx29WnXd
zXeB{UNyHpl;s;brOJo0rGxdAdHD}79I$}rv5E$ESyYX8%=~~hTjJ}^>NbxGjVOLPh
zaZw<XUMUrNyNi$>T1HexQ(kn?L|EwI7SkKXos=c!=mQsPsbeb1TGz$;82Wv%oFC%G
zeJS0t$M0IIH$eu^!N`jS^Uw3wCM2r&v)5;5Yg)kiQVYSerCGa)W5N0_ex{3uN9%{q
z?=z1#8#eRT*yzv~(P=9Kv5#M?U!NDWe~Uvf`D5F~6-oVwlxHO0G1s^sD*nzcomhU!
znq1~rTx$9{FLXrjb&UBUNh4Y~n!MH2z%@^P%f>NKjU?Q?$&TBr>DqSpf<Niyk$SQw
zoQHhbe*Avr`DnQz1WNrBaha{_F<-(Yl?{DU@6E)dE&N3BJAzJ<tL?LUK-~R>&-|a`
z_&<zV%#3XRb4lhujm;m95VG&{&p2+IV*Wr9{k4ofnkN^Rp91i{0>r_BxT`e>UrBk!
zv%4rfd_CbfP*#=xv-EC8$z`CS7j@_kWf7P=Vb7o2p6>49_WVp%M8Pst$`(`Oh?5{m
zD-0rMZr2ng*h^$|=#Hg|>hX1)#*Fy(zFp|xxSqk#o_`6A;1KvczuOCqcvj=Iw|9Tt
zFY<lf*JJpy|GGW9uGuCxln_)B4hDI<@~XN>XF21_AEcx=Vf9&RnX6Mvk>7N-cYd0{
z0<C8?j6(`yMkk{pQOPiG<7&7kPO8-vr|LxGF26rHy2z0?VOHQ~iG#>0>^}oTtrond
zB`W(}n!<1#kvRXN=9Q@0HMS(6EZOk*!yF$I*xTo}{Tw8?xxInOk+ts_(?@)tF2^)y
zz@!KulhC)D1x99t&dUY)n1c(3H5DZ1425_K{HxYu)KV~k5Y0LMpu6-9nhkz;F^R<3
zHR&?Z!LXfe;huR6Ua$lXYVtjk^Nnjp04`7(;usGjzw7NBGnT(<-J+VXpYufj=Q6CX
zG^2dRx%1wa-|SpxYp1mxt7>EESZ3t4T)kO!Q~%3A2ElEImkf74Q(h49oh@8qXaif4
z(<B!@9y14W;;%|E6o&<?dLohiL1CpP?ng=G)r8vU@#D1omI@ofos~Ps&`|)atWr53
zL#7TGE8?9tnK8dH`j?PozQM$B+nBI9ejqFi)xK;t8;`#7&?`B^I@P$2lyN39N&f)8
z$6H?gyhQBcSCfn;wz8&<dP(9ONHs<gJ{h7kD{v@`I#9w46Fju`iA6P{mDhYf;J1qK
zq&l8xpym|6*${V!?y;R(vGCxvjonsT5#B-5LK|EAr&i)q8m{_4m$oSxLbW4C1}J&2
zHA*qGvV@Ov)tz&9eRmQi0jMN5uX8X9y|L-PV$3?_frXx?swiSOT?7rg^4fYG*ZjOe
zPK`v))uYGaKB^SN3)C}Vc;*qKMMBn%XEE8b-LU(~L9#-?W){ugfS`B<$wrymjZ2T?
zRQ9eenPYI5`pWR(c9b9=ui*LAJl3N|)=c2gzc6Fj<E}9%VU9wL%)4`<3yu~hW16f#
ztQNW1e2efq^zTY$&y<ekA}z?Y28)QOb(fRpekfv;=)`(wlNe;^nCpnA>1Yvc|G7Vz
zA0{bD)^q4j>9yZFoqfL|#Z&EkjA6A%_Bmdvso2+>e6y@?@yo0nTx?-|m9$>bWE*}e
z`IPwvfJf4ir=t*6ZH@RE>n32huc`y&eAO3saKp{aHrWCaDn1<JnnVwyi~$b)Z)E|D
z@NMQLfZF498sMK)b4DeK5~!T|i?nN&S;h3Cs;il2NoU#8`Pmlss)ccG6zZ9d9k(kf
z^&UFG4oVWnWEvR~uGmDm)mr5P#&I6;CD|H{5?Svx%ULqUIcaK8hx(X3MPJ_ReC{rW
z1sSD3ae^=!h=9^gvHsC~3l8pCsCS1!6%lwPE9N=<uh631RssKTNt_>t<M-oRaBYc*
z`fzCtPhgs~CxU!P5C*NYN?`oa<DbfKXrhAOH;r~iVk3NVI%ayqqSv9%1YJNSGta?}
zSpd$E5&e?KF^_#$`6bBxkQEUbfNXAr=W{d4({IOz?5bQ*apu{DAg|jT{Hnq_&)_YP
zRiOI2{jT$wnX|uN0pXSnn+sL(!ZsVAsNzw3^d}m9=rW|$-I-PuK0SuwTU}FURtbRH
zg-HjZz8$HUz-Q#{9=|;tn7rDI&oiRONb`J!+0;PuwWO~jr)b6Xv$5#6v(vQOgTL9~
z6c2fuB`GPkS{TiKwFAyg7ZU-tl*s_j?rP&HX04bi=k#f0uINop6irT;a9%zSZ$g7I
z7I9s>epR<~G1!waGMvp^L~%l<a~lBd%>n@Ng?CCOp8LXdUgkdhfd|mMdFdYX9Bn^(
zpci}^=;X78zo68~sqjupbqed&`eIqTU7L_V%BFw^NZKf%T-mbZ+eGjEs2{hz*qE6u
zqlIPEK;(i8KCOc*%UzcPV4Fv~0|^7FcPtxgw5R6wV)c~92Kc6D8wxoG+1d7D7b~%A
z_x$qKjt7h)p`ppyNXQ6L`6kk4W1Uyjq4Ad=Wa*XPo2LVU@Ra636=a)lZ(v=%s7ePx
zvY(I;e-k0W*zVY5vlV|nfIObXP7gFf@oC8j{WTAu3qCeVVp!GyigLy^eL0DsHg1?k
zQW2D)q6zg({FABen#oOY#R$ZufcVeT%h=EDL9p)6vl*Mdl3fxh?^wz7sj2SMOBK2B
zt=%zAp}VgdEnoF)9)bpx7`PyKzoqH+k*naMwmGI)E9R|tfQe16dJ#!-of4qgleI7c
z95XvQc#*meGG<3A4YGyM$ts0=W8aVSh$V@UPk|^?P70T8-AiKy_pn>8RqaYzY<`!-
z5at|(2!S;6ot=>a0endKiQq02;WqEX4JOnWVH?+`i;mwF1g_H3wXL`^kbx`0NtkmP
zP%vqi<`7?MH#4C4<sro#tYcAI0m$-DvATu~Ral85HM}Un#>4eog!Rzn3`B*B+ObQy
z2gUB${Kw8<1;JGX*%vUQiYLS4Y48_}KHIpk0%a1KN>Ugg3!alRTS<Z>=l`mavS`1l
z6Sv%I56zVqi06HJiGJWld%Exqw4WGIh;(q(>xy5YXDWMza^T&<^&*^C93QzR?!TQ0
z-?F#N%28{xV;hHEubxtVJOk$m`Q6#d{SNPOS6cI5JH-!$@c)FfaI*Z*=cTD1qvX#{
z@ne+SAyHk@w?`IN%AhtO;mOeC`CF($UH9yto=3U{P{MK`o#^4;1R5oxv=rNNj&?};
z3yjmd?8-rzzCp_B?)dONQE;s*9j9)HrZ#ryOso<wDOx%iKcRZZ&vx{3JR#uyoUX`|
zKOf1Mo11+UDv<qkcz5yP*2j`g?a(=jD!ZMXrTcODadElYA^p8~@$oU^fAp|P@Ue5b
zdRUiydb;<pc`84WT=L!eatngUn-9DThEI|>z$=>y7>C=b9Ww!^EH_oH)atQK5_s$T
zv2i3fyAIWV2vRZp_LFDK(A`4MKCn?{#}Tlyl;3>3Zl9`?Y}p<0?%qPb1*V-^wr1fE
zFBa{hN@*G3sxffXr_38nM|Ma@29*!(tX!wBNg0B+A~b#*;P$=1MYWknWG$G~M36r+
z8^=~)N?@<JobU5*+fZC@eDPXt)N+Fd2<#W`{jJ8!V7$5Lo3Q>Yv%wdm(6Q$EF0dJI
z))ts1O!*xfc~5vx>^hRxs16<^Q<72t8AQU#tmiUS3{O5-Uq?QWECZEa8TR~np(avJ
z(5L_Fk}zU9u}PqThF$eMbgBt*(Q%A71<}NHom0-ha8OuvF-uRusxyp(Px<IP^;bCK
zpc&@wqMO$QRyL`@+LO%z97N_ohN9MQ*K>WPu3rW(vsTMA0uQ;i9|c(>B;av|c8fDS
z*s{!1xUvvP8uj`jnx|uDScU0!3iDy3?=YLM<MnMb941?64HAvylW>B`j)J7Snw?N#
zxS&GT9$@B;UHbvZFGHE-oOY*Kyv;1{W+JuzJI2vT;0^jdeu=Qe%w2ko**Ua(p)9ea
zWc`dsTuOqB#P5RO@QE&P_)F?T9#+ndaMA5$cg{UD1lQy6aZ(S1=1-1yf@u@kC#I0{
zC`Oe93L?EQj4J2{pB-Fjlpy0ek)_#nE-Y~u%B;$VGArX@i6q4<A50DhDwzGl79^nd
z4Pc!hi=<tDbhna`n1U2ef00ljFs8%C`K#81C*@ENR76x$ytP{M2xz9{^aphMJSfAv
z4G$x_W|z+?PZST6mZ^GZ))1aVOomcJ=C0bT^~<i+1{>+4`t@CKruqZ;!MqZpjYLP4
zqBHDm+V2f6yR{8`9P&XmA`NL%D}W*Qb={7qnst*Nr@qW1e}NxDN&f|jaBO7^vxnbM
zNt!W+S&%<8(@_sGBk;IH@_40ucvD4E(#$z@^0+k5sEwe21;V=|O>DtbD0|YT4N*zH
z6rpM)%C8~#U{AYn^h7TtUr<%X@EJbD6jy>w5B;d)yB%=p=zOfXUz&021Qj7=7|&R*
zel*tErvN}DS?C-PrMvECI&{(7vb#Iq4MqZY1#rU$qyb{?giY<`g#ghb&^;wjo_m9M
zAa8V<8IqeBz=Yk-llu@u-iq&K6}1T%pxqrF4=i;sH&f$eYgDm+!6RglrUJp`-HO$>
zQ91OV6fiuFFQek!3wYnPTfrxYRshI5;VH3|b6bE{$ge{V-<zIo)bsM^zr>j-HGsV#
zK0>tcO?DDjXnZ!>uS_Eka>c4Of&YuKcjyu<Oq+Jowr$(CZJU+0ZQHhO+qN?+ZQJ^)
z2WOqZ+k^fQ``NMXh`3_RQ^~+lg*;I32YFx-fF7p;R+_%mZLhCCv>3^{$S*MFjSw$v
z?&QL_;pz*cYW-7SGX<6Cv6z}4!(+g_uhz#%h?j*bVt*hSET%KRtX;hbij)z~$SV-i
zOdg}2=2;O&jS-8?75D~>#*}7SBR*hK#B;NCt1>;iu?Br<les@+PlRdo&mLD;gs{}z
zkav0bzWBDmtPVfeF5Q-`ve5R{?R*+C!9s2?-R0#;x?%10<mY-eU`ilV9JW;Y3f@Y2
z5k@Qhr)NdQDHA?QV>Wz5O(tY}{Oo)Dtl|c97lU;1KCdG86;rK5Ga{tQUz%lFZ?*tO
z;-ryqs#d2vaX3)A8Kbek?*<~#R4Rg6p+OZBAdBc{OwRmv3dMYIfzftD?i+7Bd((I=
zfWAPYx(L=UBK3<c>`th{b(FPCLQOP$U!pFG&oV!k9f5_QOiXcZxwQ6RG@FSqQsd?$
zzCINa`p9^s2?oJYh}+C9iXhn0vl5&k*Q)!xU{^hJI>L!S=-MK5^VF(9@@y=|ig<7Z
zO}H(zJ7zxK`Vz-%;nC-1g_og|#I>xT6k(Nmocbr)kfb9z=if5&$l)JM7H1Z%&m@EM
z8NP^Q;9FVE`Wd7u7kC$k>Q4~JVJs{_o=gLnNGmIy<wk~Wx;y`D4|H?0uBMRGd5Iu?
zC%Di$c!|_GR*N|{)(5|>WtAr0T-1Z(9EndEa-)iO(HaP42l_z*5A*FdqH?7SSt;gh
zNGUy|I(TF%(8noh^-Jlex$Z5mIU^S5Doy8g!d=et{c^-Scv*ju`ZBC2K{2eYTQV<3
zLu^Z<sv>7GdSg<1K-1=kBOyuGM)_16C=IGZR{gNdThZ?e1QQvKKv~9HdB0Hvu6?KE
z>V15`%k{q`LvK8Pdt+sR+p$wXuliN5;p$cA+aAyC*K$m`gKmvAd{t|wZWglKQ$C|=
zBrk_hKmuust1oSOo=gGmuIxBN_6t&jF)2_}drhtfWt>*fvOfRPZkGM2j0h>D<GOax
zW9Q|VMQPX^wfcJ<=x}UvJ@Bt`(-SDUCNrRN{F$HCIJ<U?zN<LU+*J+}RThIXl@F6t
zaLkuoC0A(Dut_KEh);^tJ*JV?Cfk(RbG#2y*W=HOyZE%tT#yiDcku9to)iYOFxC;C
zi`YHS6_4^ZP@Lb(M(N*f{o%7GFN|rIk~nJ)V~7(S6_f+A_nVCzL<%F-3QCR^%$0(c
zUDi?;89q}_5XPp=3(3}X$wM#{-Bjess_a5#E*4s83PHn2Q@x52Tm$Od+W$>2O!h(;
zW%NpDpHxWXq?<9SJ)y7suXM0sm&N^CC*vxX4`3>B&{P1Jt@;?$=po4#e6om8G|-qo
z!NP?hd@Cpvw*nKe26NSemu>x*)p#;#{w*R`aQ5^Ldt4gp!hq<`Yvw|o+?f=f%o(b5
zHV@0un$n;5AERTvW4O4k7{s%ATqSmvF7C&LMa2)x1At46<%wtN;_Q2lMRS0=+Zb_`
z1U|M=s=nEQ9r5z$;H>Tk+QAz`#FPV8BVfEl#zZUO3UmkD7!4GCE$C^4CTpIH2|6)`
zW#y4ek?X-z;ygW|uVPg{IU@J$yW;^_b*{JI;@`LLR4u~sDb@*I!u%n${b=TlKLNCe
zT^-GWzsn{%Env%_;7`>O2m5tlvE8#D4$tA7mUuTx*c72A0?k=zDqPu+*(j--rYJ;P
z-A|3o)@=FBT4nH)$oLg+q5T&fTY~Lorm}lfeJuco9yb1u<NM7YnNHoq7&^KkmSwaX
z?RWzVLEG_(9V1>3_+XyzPY1H42W1cS>d7zmG3fHHB6!ZOEKdBJ?H?YWcG~+#(82LK
zM9i;KyIq7Rzkj_-OSeshe0FZ{fZO`&QTi36cb$&a<r|V9{!DHf?}YwsjYAcV&6#(h
z?O%=Btk(W2Q<4Y2OoDCj5Vorp%P*B}XWzofE;#<q{o<wBdUd?|H~zr>nXFFgOPeN}
zF+6^A-#vtEvXh4?@0g+t=`U-?IAl9$-T3~f>;Ov@eryP#H`+-*@|lyUgl)Azi9cpY
zH_!|tWrn;XW9hwl+_nMuHo?JMvq-cvxsLEzBA0PKS~u6RGynD&m<ltKy}k!Vby@gf
zQ;g9Fz?4Sv=(~(pq>RF-Zu&_4TJ(RcX~Hz;6%Z-8muPz=uDtWs^}o$b@WvFo^YBu*
zn)GMH?tgYYuE<gUo5E)KZ}W#t|3(r2#{k#0mbBAhJ7Vvr`W;%Owc#C+fC+O&W`%nx
zTgo$8tCNIBX(>fSV}?SpTJe=CTh?ywXb_R0Vq>)fQ<{zpM4Z^`8oV>`=zx!ZP@LJV
z-Q12hfd(iPrJ$%iH+(t-5go#bqCGeCjZgptRt#~#(Ce*OaQvNJ8vAZv7gmf#@DM)T
z?)KO7$3xgumG+i{htKz63-;$de3~DYPw)3(7JRk~dpN_O%-~O`Z}n3)^A!ga#Gr-%
zGWhVFkda6`lr1{Lt8pfM;sHCVb^jp)Wo4PO>DJ>Si;4!niLHAFRsshNxkqO`u8Zc+
z(7(bMcNaxewSP-g@hslbA6>QDOw<={2jNSirKIc4>iKAJjEeRlx>yAm{h``zGNgUu
z#>qsZOf43Q@wGw0zeb`6#?#~qHv35wiNn)z@zEar@$>ww{>}i0aLOReDAdQubkI}l
zsYBh}Z84UaN9Q3n@~bVjBKt{MrU<fW9!xH{nl^p%n%@T6(EYLH^V$ruQCzzkO0{J=
zG>b;5voOKMBXQKBL{hYA@REG^=vv$ztQBIc^E6R}42G<V-Hf!r`T(Oi!n^4T!w|R8
z{YHZT_`^m*PZt=Fl$j%W3`4f54^N=$P>a)4lZY|GLj^?=g6a;{(>hqha9qdX2&b~A
z*Uht|{e~#f+)I4>5b))we?(?xF_J}sVI~Mf2N6vjsHH_hCj!hmIL)nQnia)^gNgj{
zMI$Au61A-ucyv-#bJ#8n44LX;s0SrDD;8Ib)&6ebw#am?%7IwVdJeM(6-pE@vuSmQ
z@S^O5$)|ZCNe_d(x)Fe}M(7Hc$Iq9BwTEJXEhm)VyaAf<HhSG%Hv*6QLw`<{N*WR}
z--9ZE>}J(I7K8QGe`A08<XJCH2J@yjV+j&WF5%7v)Kf7uB_~-Tct;k42p6F!Qqty6
z30e@>=f>Xe5$Xd&Cuyr|!-V4lDri%oe@cktw!wWIxlZne6$jRrw(50pis6(|1BEF_
ziw*vxY$uh6gCPzh=(o~6&NLu%uRn7rYQT+OM}X)Fty3bsoA5-i1ywG`;=HPyPaRb_
z;?q2_8bstVhdIb3>;Ywm1ZhX~YQC8T1~N<czDewj;X;0amO6LCKO3YZdOr^=o=^l6
z5V>3g6bsW?GLpei)(*|-Oo3-4b7mUPSCLMbk3DN;8cV<s+Q$lF4B#vZ12_$9!TWIc
zXA3l|sug*i7XT^~&I=ckVlz_W#$qCZyfGRC5|90a0R}P-DuHfm;s)lTkK{wdBYGoA
z&!u?N^b;~JgxjND3+!VfX8Y$@oB~g-F~+oPiZA5D0=_C{d3z7K&tn(@7W?Fa=jra`
zeml4F@<F~qrL&CXgXQz+W^en94#0r9JPZ&B@MKg(5g0)z+iZB)JH4Q5SH5*A)-LfP
zGJpn-4mttK1H#0L2?9rMS>yw<QIwz1zqo8lT~_U;xyGFws;hm~Z*B;-v;kdN=KjR7
z%&|z`OJ$nDFdRQf%p~c~2$QWa0^*|rTGTzz0$eA1va$aW3Oq<p!=Ab;#Y>n!ab?O*
zWmbn#-7vD4B7l6h%uSO|Q4g6wn4Aw&Su0f5z7Mxg1=?#Kzo@36yCP9v<NL#$W1OWq
zqjx?BX}#Y{efo`(Ac}a9@DQntGs~o&v~uHSCarEYbIop1$)K%4o<uWu5p$hg7iH09
z)YGi5EO1+`lS<8tx-p1P2^nHI|66+#<RiR>FRycqO`?IhkOHxCn$Rk~JqJ<@j+D)Y
zW*-`1kOJmZKY-lfoQAbIm!CJ-5lr+4-Gg`ShALFZJbD1>8PcrQ2t!$7_@+CH)a5`o
z_3SR63%Z=4q}ORsAE$Tj7IqYFk?9zHLR0Zmo!M>wHkNE7Ed(i?NPD&ar9g*AFS*TD
zABL*Y&&cTY*|QclseSJGR@^wi!Tos~8el)ZEW;@uc%1l6qT@ojuj|A5+z!S^!)OuD
z?zoQpFcg>23~0(Eg+{5or7Gx56e^mXZv-LqNUww~?OMVHM6X)kpTalwWKAQ@jCOap
zv+~L}fdd-B-bC!)D$Ad^*&mpcp8|L3xd$ICKehqxOfzYJqw&}7v!TsKw#I>p44q!X
z=7rP;s<ZL!riCCYZ7w|sJ-hj7WJ0@3A~Fl(!p0kh$mfJb@MeWtH`~|C)RF=CfS#P0
zH&h1e_0TV;`etb}k<ZVIV~2rcYm<-uwr{%e{Y(1eanau*`a+s7XXbfsMt7g+<Xw7Z
z__+LA;HfSlN*o<mf4Lol8wP~LExz4r@cP)|Yxl&$YNH>axS`vR{rj@1Y<#Vl=-XuN
zNMI{lEv+foiNqjb0x1tqDGQfvCEP5-wqV{4X;7cT|F%Qyf`Lu=+&xTqlDI$jotyQR
z_gvm8iPfoMX%_fOX=EBZQXJG;?eYCIOvdly&!(z#zOs#Qe^kk#Sm`SME{wUa#^7~H
z3nkwZE~LWA4W`KHcZ1x6Sy(YqHW87LpT#aRGWgq(hZ+SrMW|lfPO}cpzh*yfFncHp
z&?Bzc(+?~*tr#2|I|C;Ga=#qxzFwlGbd*t#gnFm#q^_V5Ug5z!DyM3WjN_WG6x#`*
zllF4OO;i{iB!gh12bEWvaI0$i>wr#Okr+Or1r)^roXM69TPvf3acz1d`iIG<oS2Z$
zlzw|;Rnl?><7fKDwvq1oq{Vh`MoT7;oN)+;i|HAWY-LM|51_URc%C7HILgK0&4(YH
z?!O4CKNDCj-itO92j&Z>a1kT&;L5~&dxh*i6s<am@YI>I$P9Nao-`GEYpv2++WH(#
z)rBgXX4fEcYCx_Pacj;V+>g%zGsOc@IDmMBd>*P_HMDhMG2{*-M_A2f>$8h-raZi+
zag=Mh)?j(niGy~A{1niUv<_}7aqx26&`icBQ>&Zp38>7OOi|EU#0LiNG49^k)YOt)
zL+5K6dXPCJNrh)@1M#g&QPQo-cb7AiSKn8{OO}2{e}`oMf)_iPf->ZL&8>9sH+|%<
zep>f|%Y?c`<147s2mJ4qjQxLAPXDEw#Kg$R@xLmk7EMhj;<o?XCez0ZLxvC1o3h)b
zu&r>r3T25HQ=~S0{X+>N9&rA0k|<XkwUq^o!kGObm8zocGxrCZKYP3c6Ya?n@42Pl
zsiWWVfe21ZWI0G^+wmlN7$hKKf#`W2{hW|LKS+CyyX9lZjCCD}W>^<=_-FVzzP~=N
z%ovRbUx=GDKoMW)>+A0PJf2*9TpVJd*V@|toV@7xe!+pSsb7lXyror5#ijY_7XSTy
zy@qc)8HBzdDJF^WfKZ?5eV+BQJwK@_t6fFUqr=dV<~@NFqzIBnLY5=`XUM|niV$AO
z^N<lGI-aadZs~m;)+oVoF#$T~4(2KB6+j-3dPFuXOZP)2FvtP0kL=N+#ZyEqAiq*s
zRe2f-<o`U~r;Sh*F7x{ur*{9#LL5ap5NY_6|3-!B5l(c$<G|$%=PQeWYfcHZRN9gm
zO^3iM5BA5>1^zr<Z1t$d$NQ7sC<UBakv?Xus_hqko;+upMe+<R{T^o33==07Th!U<
z@s@v7`;>nE9^I561ig_Py59yn8iy1+RrHD8vU{qJ)7Ottp^(TWRjB61*0*hm4wFD7
zv2Pu}k%nppAUI8=*`@LT_#*@zZSUW4OZ|kAll;0$T#oV81jXMmyyYNum9tUtR~769
zkDZg0^`M%O1nVNJu^}nG{O*@kUkIJ>FU9++7Ss3bHTR^=aQZ{Q%t)3YJibMDi?-pE
ztewAc#(!rTF;=bB6Q5J7Nfe^+A-V?(zSHfEKI^e7Sc+a$?ZS1;Izu=4bD3?1&=xn9
zw0wd-qQ-VL>eUa|bvhQYwbv;$p3|tJ{*&c)hg!hxS0oDh{2nMt1Q}9~;o!N-{$qdv
zK-|1e0)lM1rP*ZHdi(hSLO}<3m0(n{Vg?Tw1>J3uAb%nxL@*t<4$vT6%FpwlahVeD
z&La*WGsMwF;(1kM;>I1FZW2LaNvu7KIsK%8R*(YCJ);=Kr?WYYE@(<_Tvj?tcBWj@
ztR1uo4fU^XogwLxJA746zcHMUz|Nm;#vTuVpyElw%gickLsk_?mi_bv%QY*G20GOL
zc&$L!_<c@C5o*AEA%O&Esg2cpDR}g<=$c&Zx`?$fIG6eh(rF_XGQi2fLegyun-x|e
zk+X3c#C2DJ;lXn`=~Rd8SG9_b=20GJ;H_JZ0b!uq>bJOj4%BNKAJ#~gTd!0NB`T^w
zfpyR-q^a$03Bbup;5%}|0Vv?%J{yFD1%z&yKp)UMM5Q2{jbd;JNf`tYzEpYjEF90;
zxK4-aG$^P21Tf<%ly=ye^1l;;xk05SQaGM-HhY<<VFN^BoMD9GEM*Jd2Q`xlafrk~
zkUy5jC<-iEtxTrSnEF~fKf;&{lPfz)P)s5xn|Y%BCG@UX5d3RLU5rynHI@*Cv2SNW
z?XZN_LLa|F0fCWmsI=sYU6Wa&ebLy{RghtSNLCAIT~lTcDl`NN2rHflEH+xovF^>b
z3hsQF-{;6D?e(-3kdR<71lLR6F-@`+BnB%tpVtquFVG4)N)zAt$Kc9#YdLE8*4512
ztb<N4#TyHPzzZWWf9|?<m1zv3eh0FQG(tGd3@^Txq~pnxKCW_M@pxf%ctKe8muq%<
zeptP#i7w5>S42uJ<+bt(xdGk)1I2iaKEbt&X@m7f2$X_3oCDTk7WmACoYV|JDKeJ-
zfT0OU8L&FR;3^x7drL@Cm!&`OaXUUxAn}codK<^VNncf2U6pnLD59g4>Mto7`T$U<
zVx0ijB7j~Y4pt0+z@p_{MUrXN)**uTOO<p`pjgYO5F@mgS$Khp0ZLl9!xmLo>|fAV
zvBQn(8Qfiq(q?$NuEt6Z6;g^N@<?X^9=dt!Y%r9}g>lmQnRJ4ieG1GfTl8$(%~j}h
z?Q4Tw4Q?GqmvuVH>^ZLVluASrD;?WSVj(*o4d7W@uMGKzMiMf~v&B`gly64h85u8>
zd|sWsBe?=FhGxc(ayF0#SR|8Rn|Veu{}dQT#~y>|j571Q5qq4e1vH#}4P(keGUiPz
zoj<@1&c7R4HcyZ<T?2q0wB06DNsL6!tbzg;1UyjcuhXx};I6H2e^yOvKAFqP>!D*6
zSFi*ag@XrAw?>oG6B!fEKSErI_es#2n1k*`-!xg9NzM;E9!2KHf#eqZNtbfys)UST
zA;N?h0pQWpmLR7A*39jQ5d@*rsK@@CT3!v16e9eD;f*+NG&fEsU)qGGo;ceTxP7*U
zwE@EYVa<e;BDx`1Oa4lM{>}k)TEnzRJA)9IH-1H>?r+9ENB6GP-xI<MJgDW?z*}h@
zgL`EASR)EqYPnYPZ{7iBVp;FzIOH+(gz#Hn|7`=Wai!Xw!uf0qe+NMO+udld#%Zaz
zf?yEtj<|Q_x~`8?C8IULekPoM?ieL&Uq?Uh>0?N7&%B2I_G0QyLKl-?+QsBvHZLB3
z?UC|r=?mP%Y#cS^81#h~|I#xFeU7BwF`e2GVhSb2)Zi>x%_JVIWr;#A|4qOh6mu(6
zpY3DSF$qIoYoL^RUa>b64Qpt&0D3`X1m{H<g~{>**?lB;<;O!07X{)BVeC;~qH$ou
zJI?eg*!J1<MhNsdbP%EFv5A+&@W+*#L=qxS87ZTpLF039iN)WRJcjIHrV)^a1bYc5
zH$jf!kyzmg>wK}nc>{X(2r}Z(VAwU3y)Kbfl7QqbT=F~F+F*ypJ>*!)K`JpEIrFLg
zujyT)U26_BgS)%nP`*wWT@+MRu}6d2dV35LA~iO5PkkOV(0E9={Ebfz(+2dUfH*+b
zVcmlH!!GP=kCy%y{s@?$(0|h<{}NaK*E3~i`JX-0Ypu)w<gk8k`h>ohS#MHLv`v;!
zB;eh`z-$6Wp;1EUAxvsk)f%baNo>5u9DcpR>U0&WeAXQjeIbCvbs}+Q57~z>l{h*i
z?H?XD^v%rB`4Ov;W15SZGH8*=kpDASOw#1x$eIvd_*fC|{t$<6c1oGoo!R?wTGf|l
z#q~aSyh}H?JFBPvHLRHz*bmW!f>sa0!e~5Oug{bb`l_5sR8oSis&=YObywU)t3zB@
zRm(|ksYI_k7W3KC3+#$9khEd5(aHW;J@r*Qoz(H7aw6J0eKl-!<2fjlri+o~!4&1$
zQGKg@v=QPb5njw_{rQrWH&<OviA34Zs6?8uN@6IHxGX6|?7+B48B#G+M%Z<_l2tbQ
z;o$mmiP>-kBo1RJge1pFq(k~X897|+e@~@InW#i2n?N)B^Q=O~PJ{&Zt%6O;$WD3T
zaTp*5DUZ@G(g?sJgD9P`l+|LS@8U00=!IMk(tZX|_4{vhyE;cEB3KSqUIWhk&i~R#
zM^E<0n`xqeCYeBxtr{>g@&vK4`{~jMnE5l0$acH@G{zJOzT;==UB!sU<NE*+Y9Gp>
zuv@76r>S%TFLbz0ua7t9D#i3j6P22<+FE!*UD5iv7P=A@iHzY8WPK^Z6X&;kuFrsC
zh3d}9VPCRVy*9zjX2JJ%Q&-t!`(L)~mZfAwDjg}jKJgyoZug4YcK3vNvrYHZW5KuS
ze(iPtE9$LYR*S&!{m(Z2<-j}o;~pdQt(H8}@Ke8MpBEZG=?WIvB}SsTWq?uITR|-!
znyk7SJ6}fawv*;(tS8^=gss}UcY6qr`;C&xt}+YVlC7Ad*Pdhuh_b*eG`uV#3e1Ex
zC``AOYuoe32^xgrzLSUf(NrVVqxE@L788rl2^?)+aTorYscIctkoBl-uI?q4tm`dS
zRHvpXu0|*mP@V26l=rYA<)EkLN@jrqkiW<TnQ$tkvcHWO?xUb~0iA~W+H+@S*kA!D
zptq{ScYTH8rlTH<c1YNTsmP)1YD$YTC1GKia!rB?(gjhpl0tdQ(U9w2>pDWX@GKuG
z1&ZwLHieYA5WvyoG}}<{;)Lt+6e9Pay}51A%V=HN%b0(FA&g`JMzQT4efl9IqCXB>
zwV<ltcjZhvh(eUVcDm6-Tw}nv2qrr=G@2yZS_*`4O+SKJ?_wtgDn69jJQ*BJlwv4n
zAtiiJrGsgiP)vhV35IcsQ2v#}KmyxrgXs%;^lL?IIBcI+3)yH(Ygmc|A-5-RNqZ&_
z6-ES*I2bd7AfbwqGNWG4$OMjFW(x+#PY-UVR50b%bhoJ^<Z%(VXm|<Ju#E0BAy35<
zh=lz*%2oMSy*;o*`ubQt+o?VR5_fej7^MAf@>2#cM8p#X30_ePt^~{z8UZ5&AlboL
zFjKLTAI?*lqNj360RR{}8i;{dF8xhBv4c3VXhanEfrW#ah_2UHV7e;Y2dJuqg6lNB
zKoLxt1gP8JDAsNEyThTm2?cK2k1P8|aLWYRfeILZ0j&~nfXFvkYzgb9{zn9%U|#%S
ziufV<00m#J;Btw9S9u|XBcN=2q4)w)7;+6ktse>qpILj-JXRf8hT`%OgfhlZA`O99
zQNovclSyTQOL5!9zum5yHVZo|E8a=<xZ}Rsb*MC-9ho3bfN}$Jp)Orw+_m1LdAU_1
z24oOJFQyFJ+V%*=B;r6$^GE^nv#tbSQ6S*~kNL^43}{~IDtj?dQmo<LliGi1jN<_q
z0g8oSXq*66u#mJ}7-+Dh_I#6p%QyC&11v4$NqY3>coOSnXcNOZitS9Bkme~6egC=<
zYZUtX#A_F_{0)qjtdsSTl26EE-21re1sH~qH3q%9w*;MKD+Fki2TV{e>yPket7B7&
z*%ac-q7AQ@i%SP$0mF>sgN?+(4=yYyi6IXlwTIyN49sknRJj=WCDahozK7%`@~w=1
zl4%T*NT%c^>vevAvyfnUf}4OJZbxhQ`g}pJ96c)gJi;ePpuYG1oZ?dsXH+A?4RVZ#
zr3v`U2{^vXeF;#lWE4&I*I{6K`9sE0{W2@G>REQmmb(=3mcF#Ct#5tzWpa_Fmbb$L
zFfw0te=P0C<pl402!{PHUV#SC6xZ+F_gy&f3;p`|5TZCoPJ>%X7BTu;i&mSix67s}
zK9nDl(u%HkezR?qeby!_yXYak_Eemy_Ug)^twM%~L&3~X9i0MdKKaVfr?L>mqC%0+
zF|<mPjl^{WuHaC%DqfG`ib9PI?<FI9<zzKA{~V~1vdI)6=GO0BIJVrEhuo~V#%FNR
zBTAykgG=1(*2ABa9`{BN<v_|pk<m@PW$~kM+o>6k=}&T}y}FnC6Y4|`R~+@(curw^
zCm4}X%g-e%FC`-vvv1Hz*l^^bIB2}uoj#W!y-x5eM6uZuU2wZrlG25($=;ZqC$jXB
z(bWfH*nk<T_W{+1ZPh`L$JfD$V|p<n^xy9Z*TG!^B&&YU9R#xLI$88ta(NdUCNygL
zk2{C!TRg#KIS!FGl6K#V*0a3(S{meIx-~eHj(?%xmQ6~BnVp=@U__ycdCAEF_ky0=
zR&U2bWtlIRwUb_H1MNlO0NUyEK*_Rcb`w9AHEx};jQ9?Wz-x?;uPur-$8(gYos9TC
z4f^f6!c8(Wwgo1izU_D8s8+ur5W80C_T6op)sSTjkf7p_j-(b@E?`XPD>k2>EW`v(
zqjXUFKI|fh&X9j|aP<i5#jgr}2DRT4)%m>OrQXpf#Y^XV<KG_X8&;yFpaQz0mg^H4
zqzz+}FUQYZN4;6Mu*54{yO+m2a=DLNM86DQNA^zK7^ar@=+SHu;@E_(;UQGx1gps_
zOxL87U5%SnaKw;dSYcHn{w#~qkGI-sOUzl-VNMKe=QlQg=76JTT-vVPXqqC#_dPe>
z(c#^GVwRvD*Gju@A!xGu-rrBST$+x2I&?bK6k9F_<@Inh*{lC}n;WpoAr>FY%^tMT
z+<t`>rin7n|8`qHx>!k}`bl=#&vljRr0Bjcz)xj1cIl$(W_-RAICYH`$IglF=c?jO
zR#t17P2Kra#KBc{;=@i(>Kea7m2vz$AME`CXwn>6{8!z@_J4I3BPS>0|9THv(X>h0
zVnO)v?Hk0;**~~gNu@MB@egVhhSnKkkl8X!9(HrMm$u~QRRZ(nrmjie?EPEPlZwXI
ztfcl}>DFQw7Yl+82-(a1Vd?68syz_IB4vyQ-t_1{7J%aL&!YJ=wP%suIPthK#J#JO
z_~J^+Qdl3fms@w6_r<}@n<bM$Vu&MK%0#wbw3I~7pO?eO<-;=+veeXs|2H>jY<KTy
z;oLeg{LaMl;qB;Ca&k~VkH_5zj)zE22&n`z$$r_~nlA7%b=LNC<^1aG7%{|HQUG|C
zSjd7PIT|SwB<f*&-h;-l52`&91U8d~ga6~amFeZyf00GOU?&~>Q3>!7#5ORJLJd;u
zow-?BL;&tfUXjO00C09-$wEd#0zA}tm>Zd!rH?kZ*FE+Te)0`7q4<I^?`1AX)beMW
z6(*39l*Wi96f%kfwfscimsaYfrn{^`d`E6SR@rZ#DN?!pjw1pD_?CWue1pnvKsf{^
zZG%}zBgM@_7gGvH??g1f>Jetlg}FddV=k(CS+`ff$TzmAw)IW-8e)6WXVn%hs@lfE
zMbWNuHs^fKH;2&pOO>7D4?TNle_ZD#7VMKM_iGF=p{d9~NDY~LX0l-m<!Oc?VGf1B
zWu^`~EeKthF`oN@lv}$jyGK^bn8YdvjZ?g2$VTDN?qhf5)W4>Xe)2ODwf#VbBp34l
zk#{=QIc^wftyar4L?WA7PA^rUsAOd}RZQ*XV<)Ay4AF7vwS6dgy!{f9x>0iZ*EGGF
z`^`uxz=+;wP7U2!$d=MFI!8sCG{aObh(tfD>abF1?R$G|X%&xEvuTgUF?2JCLux(F
zXzth!M{Y2FppcTR@$!9I1d8EKjemrZoGbHvrFI1>Y+C11d$Kx5;GndVDkW^!6qZ{H
zE<%e-A?eM-Ns#CzhyYcSoV-g(l#LZGyae=hdZ6KT%u{eIaBigpWhl_)bLV&y>6^kd
zayT|Z`jBo`zQy!&I(}#kCM%o`FDNsr7;0AvqiyT&KIBFR;|XWG2DBAW#A<FnzB*Ai
z2zMRMRO@!~M!27zF3QGy3PA=kh6>IEhv>vw=R!XBC<tRJCg-%M;@r0jR#?9#FrnU7
zJRzU#Rqd7>uLU5<sd2~S2zkz{)b@vRRAH}({^t>E95Nuf^cX9B0%F?L5DG7nBAheW
z1bZS5KbFF*j-g5X_b!Uqp$nu67`|IqmlZgvg^s)_mZCy`3$-5?F@7pRI+XJ1N<AEF
z2HlIQfPK2$uFJG3j86MPPP-9Tel{jy#@QvBu;X3?`a^70+0OeFpuMB|AJs7aCMRU<
zvA3a88s-K(&m_juNPz<BY&mZYZTKw{Z&DNUY~9|j7jHOGeo62&L>bH&k}ItG%|ao+
zLj}Rn#Sjbn>YSWS8(+l>LFs$OO)zEiFkN6~Owp2FtNIjOfduP<=2`Y$i#;-AH%z^P
z4TD5jwKT)u({$xCQTOW5zinQXlJW#lhEkeLUtrvjh`m5z6*dc9RuGWM7&;mjZ9^`k
zf0u>JMzVOK`|7MmjzBkF`VxB6^A%w-4lJkego%GCm(T&I@8rNRX~=4;-0<ven%n#B
z7mWV?j?@UJjMkfuK}RkinB(63*1etk?_;0^=kc>7L<C$M==7uBEAnmM^bhVUxDgbh
zUo(A|@58N)xetc`L%_?(#G?SijEd@}vaK5Ktk~9Qp=6arpcYzx!Zf=iZSet@hB^8l
z%;Uwzwzzh1!ZR>L`U5e)(Y+Sdbock>86h0RmuC+X*Hm^2pUZM-JCCd;Fb8QnH*FWh
z<)-9kk^T*Uzp4nfGci2)BoVH1$nLUkY<<huGm9ysXJL6_-=c^>F=!nq2zxN0vVrG#
z1A8QacJnEJ2kc>PG@8$O<#M+(Pw+>tBrK}TWkAJZcg<5m^KWxBy~7Y;ZAQ%(qOa`n
zY;+6s7wXt*evWp}Q$#8`uWj(yZ;kIwXHp_)-BR4KMwatu#<q0qjytD26e^<_IcweO
zp%O+St1M$=9Vddly=JWmtZrKRN^QlFvbLra`-8E8G!KJJtPMwRVlV~#1F8e7<q57$
zg$Y>OmqTlZ2C9=2#0#GNI=G2<R7xpe8{l=@huqw=iMv1&V*t%2o}{u~2J=A%-QNrj
zHX{!V26#z#x%hR`3Lc=snW6M}G>=D2U=j@gF^w>H7?y504N}o-$W}~<Odhy%aFjA(
z?8@eLSU65yRTI=(Y=?U`O%d}QBvNYw$Lc?O6wT6o&Iq7bAnu?3qV)NvAt(vX<7F>q
zD6TiLAKYr$%TnGg-X*^YIg5<9fs}yzfWQl0-<a7PDJQ3$CFe>RxV+ar<6Lggn1n;d
zw(KT=2J#dci|gOQ(lO~%{|$-m@=D0=PR-~opNasY^3685e&}E9*;`egbDB5DqAI%E
zhbw-dybww;@4voOpcSm&%V!x(=7Q&CB6FP<Pp*F{`F!5Kt~ZD_f6^2<LN<p=aE9z5
z*znEQi0&1+zuCUuAu>LPb}u>#{(9TfPEDJX#FY~CsSVskX$Rrh+o#+xK<+JG0rX>C
z^}WLRK|@YxT+rv-KY8iM!r-QKKfp9k>m6g|Azl&By~D0K@4bgHg_}{m7M1?Ke$IWC
zt}H-IeB)<>EE4`VVZioZ4l<0)EUf>lFz8X=vfUCx@L5y8O8^?2(UYY6ASsfIFQlE<
z5Kv_C(NLzgfYw7vY`g@xZgX=rvxtbiW}}=(cdBDHo6XKhv1pVwe{gkZ(d+gdDwktf
z;G9c(W<&HSG?6P;nea^d$~*~9GA(t?{IT!RLu1l4t-@E|rDv7*>+o=E`m*`nzGPR*
z)>cSXQ1lNn5P<gkBUmIY<vfrgMc4OD=V5#Ky|&Y-N6$xx(Ov>69ZZniAc7VsL6umk
z`R0;9&7n``-?7JZZqumyZz6=itePn&xnGiwhrHh$aF|K*p;16@v<G*FX^K+zcTj%o
zIi`YNK{V@-S_f+e@#G3yo^XBkU(rpHWgaK5BgPvU>V7{jk?>&y3IL4QITc64{TZ|^
zNdF}x!VFQEg?T)MQ|&YC7r)UH&tbHD4rLGx;&aaYx6k~bP;5_x^IR$6@1+vu;9i$)
z|6wKG>$XFi{-12dIuxA08<to9rpRpVdmV<>yS5I`I?>ca7ouAmuFuL?wq1GA&U9jw
zCGANe&EEgM4k45JfTf*AA0cdA#d6|QpMwwETUYg1xVZvjon+_k;mg=%!h{Gm&yoWH
z_#YJmpbw<uUPqPp7>}H|)u-oz{G62~i%dj$%^Iu0M4sY-gYl0y+qunuff(qA5iFZ!
zS8<*)e3Q>g%SzGp;`AC|C6eZj+FcZIi{G$5SQ>rESX3TGpd^ng6$W^*d!{T9l|q7j
zh6P;<*LstMrUe=*A(zt22<dw&3c{ciL$mgoY|cc*N=c#6|7+lKdu$}p73^L_X?tzK
zwDvXuhk9A*b`9;%Zf~H4z1kw5Q7j4HcJk%cbb^IzGUaRnydvhUF;GKN7ye5tNq#V3
zP$`cYw)L!Ur7s{fi&5}nqF^N}nK!@LtYu<N>LoMvJY>?Se*Cy!YUw=iq7<6;ChSiF
z8lYLYfjbOaf$@8pdEpCZsCj|c&#$I<)CVyy;F#=~fbsCb=o$J>484H!UWGr?xGSQG
z98-zmS8Fqm7RHNOw&B5kNebrrR*2vkT$XOSwP1>=?|ZvoY`eo;4OKiV4doDK-W%>z
zr-n$ZVN~c>a{(s=;|xwSAJA*HZH;LIyPsVElDGYN<8dog0Eq<OH`dJ~pJNH5^h#?O
zNQ|qIX~V#n?n%Dife<d{s$J)s&KC}@h{~k_VK^TaCT&3Z!$19{t%X#J$tShu7u|}R
z4+)E|Ckv~(mAT*%!mo>WK;uX`p4^rNmKk8TDOJi2Zsyjy18y}%B!B4XOk6<d8iQ#c
zeX&G!A`-FGpxPxkdgE|dQ-i&9kc(S1!^NVBdSyQ;^2EAMWh-!*H1KAcLZPJayA+EE
zI1o$$CvmS!VK6Y#k?u3|Ns96hB#d)VHTnUl?)?LLN{OPr)~-PCqf@>p6G8OtQ4w&`
z3z{zU&_8T=R%&r&5^#19Z5|%rh?kMufE_-tPRZ|}A!th&-j2TlvG0tnUG&%RL?BaP
z<d6ge@iR&yZs@QegJfRIt<h5x8pXG2@vV1xo9>b=tGr?7Geg?#N|w7nX-&k;%+%??
zZs&MbzZhp{C~>^PuJ>_0z|-AOHU}15y92&I;ua=*WO_5^k^ZkgxiNce68E8{llIo#
zl&}{?&`6gmLe88e&*f$%U+B(R+bLE5w>dKW-U}RmP+gbwj(Y+!*ssPO_E}PJfL)oT
z9|>5lx*Zv`;!jj7(p37=S9L6Ab2XESqd7Q!*_txxd;~PwtVAJU0wzxzXYR*$L}<Tc
zZPbx-==85{l>VFgh24gIP|bgRlK<OoDFfsGoO(>rkWAWYMe3ca+e@fSL;%|JlsQ;e
zomKM4n3$LaEM!1Jj*N$N98J>y?OXsrk1--I0q*~2+CN{0CZ1uCE_8JrS$^H@-(<C+
z#<>rG%q^bnGbVZ@r=m_(5_xin;2#9q{J6XG=Y;cJxhF=xk+YuJu3w3y4%=aAzk1F;
zJy%_<VD@@Of%kc2O_m#}(m5XG)>kI?Q$)0PQBG;<4a%CepZhA{Hpbr%m`gX!wsSA8
z$EHP{gBNPK^ft@7J$2oGz|tGj2S(E;L0C5>@kHIi(z4_VI?enTRolGO@bjnlXS}Tn
ze+Ev9WRvpfhv${bpd}9Dd#-G#abhmg{M@{{y>d($FCia)w*@$|fp2RS^05y^l&gF8
zZ&rg}zonZ<4zX-4Hu5V>gI^h3BsB^vgFB1}X;Eq<s*>1%6KU*?4G38zjP#w6Z^M&s
zTUxc5HxFV;lMP}*k8Oy(g|H<?L|Yk%BsX2?eNA7UzT~LvE%9c&NG>(>V0H!Z#iv7K
zdW=M}kYm>2W>v!i?S7(T%}ri&v*lQ3_4$~PG+_D!2O6=wBTXTk-5G15J&&1Z{=VGI
zf;tW~26hRcnNCPSm`*s77?j?r7V3chETUpg5uPpvH=#JesrqnmQm0wK{B-%VyaE33
zWSGf^VJ>9vI}l^y{Xz37)6(*_?SG-~S_pL5zBq!>AVKH&+*6X4jOObfrZMA}@CV4<
zaTWH=(dP`mWiDIF^AQ8(C1`CG5hVAX7zCibG2qtXILqN+)+$mku<V?fQ7(Rd;p#O>
zXi{?TuaGqx&>~F76PNMd5m%8=OsB67&ASHr_>tuaaTmxwuAXGtk7w9~oG4BP{PPZ*
zFPmtf74EWq9`}!%MW5o4ZL~YqgN4N~W;Ndv$qN@0{8<@2eP_d+J?##Hl2|eK(Ht#V
zC2uSYVQjII*;MMXg7@63E+R%xuyBlt^ur>1im8i3-lYCnY7>gC<%iRSRU%qWX;S^Q
zszzOP5?+m_K`jaeIVO4Z9d+u7VRZ_*KSL-~8I=OWz=_bJX2aWxjmK@QNyPSYK_h?*
zmLzQZwwy<=06gB^Mz0Mte*^X+MW>OPM&k3>mPA%}K)%8bMSvRAA#r%Btej^)5&=@=
z5o}OIsSB1QPrMmEHk;Wzfq%iEsG!Y_aJ*0ffGipWZ6%{6Aw7tbP(-gC#(WceK`x-7
zVwpiA1Jca3NNSmHonW57`jix)WJiDr12R#Uxr(l<B|Vf$y(heTHTK^Apq=NkRZD3m
z?1rq$Z|ja~tY+=XQ;VIK76rH8LM@_MKS_OXGi2aO4Aw_R>-w3t{Uqr1GRft&ud)UO
zkUATWC7>$~>4^Jvmh4{>MF%`CUy0jty1qVf1}zCauC^bwuZIi}K}BXghB+=CPRhWO
zjmBUFR<$LbNgpeyz+!f{L3mY$;+PQ^EUHdQVPMb>#E~xp&_?-?E^D(ObX5(YGj&pW
zBQO#0T8dDOjzH|6(l)p7G!K?@o!ZKR+y5RQ(qX#Z^wDjk*fDz+me-KC;hD=PW4p{|
zo=iZ%lNkIp$po%Ku%qlmP~Y!(9&`&73MmqmXb@#8%-H>gl=wc%7*o(zNm$6LnA3ik
z!MAxrzs*K&Oc1~;?l=vC?TuMI--H@br!KhdZOAx%OcHlH#^s#guPu+<4mdzeS}6+h
zt=SQ}>hpFs12GWkt}696R?Z4#r2;-LoI+hxH9UAiDzwluo$7emm2(|(Bv}k5+E1NO
zv^@U6HRKBq=oh1sekDf~^w*MeSb8cLwF9CCW3@h-TjLy2g47<kNYM2Sb**Zk-284+
zQa?FC1$!-S-)Kp(57bG5%(TKQWd}t9Z!o5>2_JI?$9Bp<sJPo)l)iG~0dq<KIX8_)
z*2iMsbd(<|!AaqT|GA0sAC0Uw=NX>=ev`8a){XK-k<uOFD<#?a;B;f^N7boH+jE<8
z`vp<t8^sXEnU=S(n^m`rR^Uz_50@<2r9**>1_3e+Po}zdTL@8l6p=8zTLL<6ATm^6
z*4U;$1)*KQ%@lfcPb4w((VEf~0h6FqYSpnX1Ah7seSaj+=72PSG!mA>TPADc<neH#
z%b^3J@llkK>U@aF@x>MYUv_>c!bz!iI3>m;yzs#+)X3eY0?Awa-jPDU9V9{&b%a<2
zvkvT>XVL`FIgH`Q58rRp4A5>OeO#zo@c=j$3v$MvAOTZ2Fv_*=tvX1MxeB1nHFY2x
zVR*qFp@2GEAm-$yB^21P6G^wK;;;Svo26}_a87x~uFJ>mF#g0d023NtB+p2*wsV|&
z6(Q}Ik@}dM+aGa1fPi&`T2Vm(fQTGbM0qIxK*#&WXMaUhh08UHOj4SPJBdlC1X@Lu
z&gso;`a=XLBqbpWS9~-mk1Ed(dwFdf`Oy!T-!i~n(*i-)aL8p<asOZ!asT8F;=49E
z(nl;2XnTp2LO`-U61%7OgG3y<Z6uxV;xkf*yImAfhIgC4VTGdsklz(ZFl^XpRM_Cb
z+o&`XX$W4X>JvECjg1fVL+60zGiM;>&mVvsza0f&|M1R!#D!G1s<+tfhDeQ24Pgq`
z__*#gEV0mNxyEKd)@!Bkzmc*bdLG}5BUyLBxg!;_Tb8do+Uj?HA9W;Qv9L6tAu;5b
zngvpI#J|7OUA960S1WM-<B9$McDjm(y(s~`ypg4{vkerz9KpXm$Nw;H9i5yBIGI_Y
z=w(c8&792%7&+MgXS;nyYgZnJ9l^J+Z!m5qYVGho&LM;-L4+kii!xBfGRj?IUrKSY
z3k8n#`}Q*?z~0M?Keb2n0+?YccI<_nIdAMG^(>m1)bs0KDVC5de8D~Mo*nz;g9;<o
zNXn5+fG*yc%Ke2*nL)(J(HXhW$k8FRn|aW(6=UKpEB_D!RynXyoIaO;#5qG_MHS&m
zcb*bUf#QtSRCr;k*yPM)oer~xWKT6TCy}u*IswUWd_x#wvm0aXtPBDZA4$sqj8A3C
z$lC;BaMVnnOmtp{2aIeUQKl{&!!?1B<li<js!aGj{kkCppbu&wz@w<Y{QOA>DuHTj
zXwqG=0g%pEbZiI0s7@s0tf8Daf{p+>sbIgxI49FU5ibc_zydpoKd!eAbPt4f{IG$P
z2N0!DWF)m9y?XG1G{PH*-9j3Q4dAPY<FgIZ0=JJeOvI%&M<Vm+(Ox4)<1AAq#zYbH
zWPBkU0L183C5Y1jii!sEXgDIdVk1_Nym0`o!~_X?Gc`S#V)<`Wyrudy8KdZvKZYb>
zxvq$ApkYdWn(#_c3Jt4sqvQpd1YTlj<j`9aG)8@pG(yM<)Ioe@3WChZ^cgV;uoKq-
zR`DiEC#PjK_t&IkC1P-6JeJ^YhX{-P8)>a4=sRH~jD<Nw$);Mg`yyRD-Job$RB%E*
z=H4DwgkG6L8VD63hNw<UV5<_KP<&J=K8Z*$atn+Z+AS*>i!g#%Qc#fO<Nhwa@>2J&
zJ7*3{-D%T^b89;zmx5U=`ON*(A73vYzMcwAo!{rBr-O^bs&{p#N=`hDob*{T28Wo|
z`J~F%2UFJH(d<+yu(8tj<F#FcxjSFRrmeZWIXvgs`5I0Rh8e|>Pr3K-Y&b4{9zq39
z{%&Ep!E2p7gMXX@UVi=zx@pQk4=3X+_@(;jCt2UH$}%~WK-&e%(LQDCxoTE}Me>X;
zJs5c>BPaH3SwqVfre?TPQfbfM#U8AEz3F=D+Bel6TisTAhky&dr9FAZ>pP%TRRB&6
z(RWWs)?#fq6M@wEZjPUyKx|p~b${|Qhk}AOg`M1c+cW1flTJ<#-98?--%q!)MxOq;
z8y3Gmmj|F?O-EtP#LHY<lX^xCo>$951T~H)bi8>*DXKl<zd^^sI+s2s_9x)CLjpt}
zXwZ9XF!whRDFgrbjo>$)&4t-$c!g8OloD=&ygZ>s<e^CLvPmeO?c`Uk@_Ei$V4mFV
zQvj_6J20nB#a8D+IpD&y#__h$t+WCygd<3KM6Uui8{)<U>B8UxSD%FGk5v1|UFYVi
z?HO-3l2a<HT5~}v_;~K|oZ8C?c+FvxK%zR0S`F{RSSJeH)q0$aGw_e_PQd!UZp<|}
z=1tG^)5N>ioUssW&=i&<f0Dr|wP}rWW=eQW5wI-b5gX?Y+DNrO!*p-0ZNSfz_Vy{y
zlsBkhMqGvza#2F3&QiCtHx>m$(EBqC&dJHWu_IAKwSh_xi_li7@wOV1|G_>OmbatP
z`B5w}3FF%YIphrNOx6Z4T-R_Xgf*4LOr<9$Jgvzo(W)q-+6q(<YN#z}6??vo`SkPK
zDoWs#Bd%skQfE^i>^0%LCj@!_Bw^Rjwx$jAPBcl`Un7YVhrU2XEy~s~akN0iH3(v9
zjM&hLe<1A`z?4K8VnrM7;P%ksT5CmWU@CpoSW~NoTQPU`pdY;i?hg9=u9KdAQQbYC
zRlz?LyC`CRcqYI6ZtW#GdCagJ;w{O1-eTo)mW?CUWH@8;SvVf9)=^8&`z_cZdve#|
zIT>LO4l%~)%0#^OvqU3EXk+-IuG|AsS^-RW0W@|!X(&pOfi!wrcabky$Ihg@ZH3<%
zpzyjWKC$44-g@P)D{blwxS(VNxs-dqa(8&yzC2n{?7o(G>vzVNHmJ8=yKIhUU3&tr
z0|tLD0eJlFU%n*9mG9xja-cqhHdK}qYh&>oKbIa&1=G$tus+b$#@um#SUNn*iINqi
zZ?#_b5{~N&CnBWI_{8u2?a=V<c<;3+W&WI7DXXG~s>#vUXZu*jnX@xPS8tkn3GnVo
z-O6d8CFK<&j~cP<+WF3vUKiW=wQhRltCCHup>p*Y(4+GK7k9O{BGL{m<4QxEWd|c#
z8THzV7=5$y&`|yhf){HH!E40Te9M@=E>*~A$JYjazhh=~bddQ5&6KSf)MvKpH5cbn
zd^2%jAzBO|1wp{kkRRW9XKY*_r8VRpb|Iys$PetWL&keNhbhONn`m093mic`rur-#
zh!>#NSuBxC`MR|2YtDL=?{ul@kcGBc(RU8x><&?83-Q{0qM!BgP5%*Q>Wtm-6Z`=)
zU&6Y9|M|!7LHBI(Oq>FneRGJMc3ZOSsacX<d1qVqeS3}lRR|~Tv%cD34v9ibfuO01
zoqwR{$jJfyu-K<kifVO8kp}SyO>wSpsyMI2^iB$AKPdJ)e&J!Ys9SEsa}~ab=3HD!
zlZJZ(RK^$Jg#X6ZKIZ@az!@3X|BErp#K8GKb3%``G?TVO5PWz4fywL|a;JS1iYX+7
zAVv_wcOl@lwDqFK7mX&c+4^$vX3j8bOt>9{LJGeKEdDI7JvsYrGsW5x;qdmoxjNjR
zaR<tp#3!hTMY;TcOKT00zIl`)3&%PVJYEmq?EL(GuAF5+b9z4y2d5!j-{)y%bH3g#
z?_&pH;=2y3WhDQ)jDSU%@rbT7pfA-$9XV$GwsNS8Vx6jOqM7Qk_L*H)DYnSsxd;nI
z^e{%5EJss_knYToSPX$?QL+aTwXd2dH?1R7_iYoqZC%oZW|Ep@)a=dlOJAegqc@2o
zs*4cqQd1>fMGww)hq~P%W!vf9Do`YXyTCxw!?IP1TCs{5$}nBVHDZXWJC-|q$@Sp|
zR;UN%7Q)z;b&Ny4sOz0Y;FtTYnxJd|vQO3Hf>SKBSe?!CY1G3J=Md05WEA2eE!gwS
z;X*=)eL&cVGmsQB;GQ5~7>P(Im#Ki#OqSTBh|E|LP{RyPQB9Cr2DE&BTk~4e3UL88
z4D5<_8-!aB_H<_0SZ{mpzD;vWE)I~R8#ZGjBofl^^ECSoqQ%=tj;y7@y!W&zBNXKG
zO3!Srn;1|>kT}QZ>wE&{g(P<imXmpV3SfMmk=uG&S*RA8Xz$6}802Hy5mc)*mPbQW
z$#D&<hrL+DLJ%Aa31`cXQFUYH{*P@kiHJ~G)f;%R^ko#yB&9mJh~M^Mz+(APDJbF~
ziwKE(uh~e&Gyyd(Ib4VsK>OXz1Apf=Q%`)Y^~p-Gt1?Zei1bTTwT>K4{zPt>psSC}
z%5M2YG&R~g**zTC+>jj*5=y{&ba%l=Xy979#r3l{pC6%Ris1w;VILp)c1fmTXIYld
zz}b^FT^gyM<{#8ZYdN=unlV$i-ke8t4Zly;xIs>pGWPLAJ!biMCndWin#d*=1DEWt
z<SsJwMD&2DaZ>YGfTSTpz|0Qk%##+2q~X;i$$F=80j*<>oxD%((nR^dfd!Iq2>{6V
zz5#jM0q~QEG<QX8e(vV#FL&+YY8A2=4XzIhdWV|)Kla`!NVe{4_bl60yKLLG?W$e2
zZCkr++qP}*vTfUD*ZXy!?sM@*obN<;MBn^ZM6S%pm^mZY%^V|sV~pnkd~rLWQF|;5
zS^nkbLGuYs1<e+v+N-@G-q976;=@W;zLGx1Q}n7Gx5;q=5Myj$FR&;GSwb(ZjMah_
zMyAELdC?N9`w@Qr4(HvwulrtDxe9Yh1IR&sNxU`O!FE;U(@Agl5;<9b8ivk~U}P5$
z9Ko*!6QdTD(}}!01)?5@XF_5fU=T>;8@q^50lR?!ZWH>+0YIPJIh=X~>i|3uHRKHz
z-v?pG58@{{>BK>Z(pp0(CYUv7P5$0NGRu+2KiJPK?j`t|$=W}PZ-6Y(Vn#?Q*)MBQ
z+&azY#Oy)|fXgb=G>+@(a;8s;uLm;91I`pPX@W3&Y)2xTWdy@x4FeSh6S=0DR6Wuv
zh^c>hMJ>jZq|AeUITpHElzo{nrLB~hOZA+J_WGf{w^tf5JT=+3J<=KKSQz#@glLif
zhNy!uKpSJZU7Mwfg(ts>S*U3_Hg!1ZiYgMDV>E$>C5$l_Dr6q0gNobegJNs5&<=Ip
z)tZ~?iC0H;GOdcLpsf5vJij<SpLCms7B;p;nSIdz_uf@`lv1?oBGz-8;DpX^dmSor
zEPFSl(z0Ut{SfFT(pp6c#1z8VQtnFNBm|8%IGxoi%Sb@28lGosEvfy>6G2ql35d5}
zwB6vv3h1R&k%h6*l~gJbi$3ax&}fNG=GF#lsVA(G8DPQuzcgRs3@*VoU|=?u;(smQ
zI;R&cc428<-y0{m6h{t0h-ftUM-~W(<B!SuH-hIGZF8c`RH+c0R|NJ4h*2>juyXbP
zVKCZZ?}0~)dhz>PM<@gn9tn6?G$Gt&T#~B`AaJ9I4^pT$r5qUG&1D2YQ3?Pv!~^lx
za6abNa?T-+W0V{lJC3>>g$`;3c*1C<=_UAVMpjd-lM+ZtbfFE%ttTcCnvgDsTyuZ*
zSW@6J6B>jtfjcx(09Qh=-L$}4{X66nR`_K04zstk7*9R(xg3<Qi~Z144`(av-l>rS
zu%)YdJf}crq5aV#80#5sE+k9n7LW_34rzv>rX!>aZfgH;FtqzMDeh<O0OO{327S-j
z1m7ye!7+=3kSBoTK~7;2hl^T?XSHeS{zSK#8QhP>&j#%gh5cTsAxyZz;$qHxm;7U2
z*UVJDV>nCx8ZHR{dk_c+q*5Htk!0H&28AAHi(1TUzz$1+K_V>*EEE#&0jUIDE~X#%
zBO4yM?^!c8oUEO^h=j0t#mOM9gN`fLuYJ6hnz4@Vv(-!eS-z~1z}xTqn2P8J`}Z7>
zAH!ReCs_@$pnQmrdVAG%xyPIbFT6=rRco5-8uiV<s=1EB80~pinV<1hE)|hLmkt->
z`H!(VHi=}LaD!hixlhLRm^I)4Idz;C-n1*KnGxRKL_3cY@jJ{R-l113BLc1$AWkK+
zVy@q~+@5ZsZII|qQ7@(u5k0gXzsd42F-AGYFNdyucg#TdpDIE>OW&JynxY^=yo{))
zsXd%;E8gLOZvxHa+4lYLK-dAD8L`>0oIAdH<fw<A6t&W%F}I$2pun~WnhB)db;DY-
zKaXg;u8~!>Yu1~UF;161pE@m}hu;S0oDHz-LyGV-g{%^GGgPBJ&-fq{)jb-@>OI-o
zC6`%Os<i2*KUw-{r#73`2EU+1G#>x<PJCLoM>YgcZG9f+eYx$58o0JeX}LMgd#DsI
zcEcr+U!VgRATIF$CW~*iY@gD6be6V@9(pUj_5uR?BXq$FHLr#Mg<$IR^?d;*a50tr
zTl&HNFJ`X{>_5t8|8@O?53HPw9q{Qyt@NFYg^dkujf|mqd7&Jg9E|m?q1@IE)wdo0
zNk67_3AQfs+PALf6`+*X8hiOm3Vo!wFosmD{Ar4sr8UsMT%VlC)JrlAM<5&jq^xk{
zJi9!<Kdtj%h|NQY(M5M-;qZ7(sTam2D~lFHg#Hz`3!_Jy{i)BxHDY(;e7@Xp^tCrc
z(1#Vy=@o1}JiL?Dh>nmAq(X7ao1sI8f%EZTQ1YVV^kA<(n(}q}csOu#Hh*_vT5!5K
z+#Q&zakyJgvG-4vWs5QBL+MxLr&1=Gna6bg1U>=@urwCc$&C#6FHP2IQ)7<0?_;WW
z&c9oF(h(ChIli?u5sl4^lMznY-*IS4Xf%j8@Z@pg4TWJY#?NlDl7lSftc4Y^NvWkE
z3PS{NQ6RG}*F~F>kWmn#)C7YUze4MW*RZJVMsdZkBim%AF_|hY!L~>hSv-AU)e&PA
zXD>#9;+n*`A-DvRtXw7n>*xnV&p?Fm@56Apw^EhF$qkrf`m$f^9q3~=Xu&FA%&V2N
zB*U2z2uM*#B00w+BVzxi;sY)1fk;U307T{VmZmEX%k;0Q;kAUVpvQ;KB?A*D!h}Ke
zT6BU-xr(OP%xL5IL96aH5QATY8-VswA`L|n7Bz;1#n2+MP?8=C`fep6euEf$n6MKj
zKF@n^l4UoBY|7-Ubjiza6~UFhk;5$YwghQ_k-a0n+{bc5ZC~~ev^@~pY9F8-#*zje
z5@yoP4Sl$1J&h9iV>FK7z-s`m%|00+4XOq<p>8Y8=t8a{?8cwwipS;}z@uQ)3Op(l
z>Z35^j&h?^AcXVN3%cZ$y5yH~qv&FQjC(avMp%<G%EH9~@QFhNA_=KgyVDKhuAmF~
zu&uj9h!{dpZ6mDPeVxZ5_9Btp9qHE$blM8Bq7XljE=u5ExOjY7WPtkk-8~$_#wB--
zQX&KMI-p-04#)LJSJzZD)*{=xr_sz)IvP5vKOCPsSZr4gpKeqY4Hd37xokHyymmO-
zY*#j`uST9bdy5JZ3?+C7FA27tjPEGJ32~4-&li6|8#aO|np3D>+5sKTiEbTDJ3`Ef
zsg3tF#A4{Q)D|NwMGRCmvKwF11-_Y&#bY!hZOzIR_^X|h2fw57J2;Xdw`30{s?Mtk
zVl9#H%7k;ksZ=8pwvPAxN_)6FkDzpNsgs!0u~zUAv{t~GK_#FA$_3ak9LC25UriAh
z&7(5tx;~(N%JiZi#J?&teR`c6H{f~+-DEbC@eb;ybb0)Z;oN+w((-Mo`*_fgYM=_*
z4bfWQA3TQ^0bUJhn0jTEV<>syR=C)#6|1+NM0Y_|YOZc1a=3Or)Q+y|K40I$_X-u-
z2-Y8{<{ner{0!UOv$I`hJ+o;YDCBw47?mQko+}ELYYma|Oh^oj%}`~_!)dsBj>XA1
zcTk+fd_Z0*viIAt_3c+Y{kBUp58UCD8FtswhG&|;C}~(pHC;;?Rf(rA6|($Ab*s%^
z${uT3S8+>dUpieszT&-hjW+~$8AmK`$!mXjySa@;w)JMYO|5rBw$PkD7%^f4AU#q<
zV*p~aSu=f&ZJY?6tm*NJuf#5Lvek{Mr0Ta#F~gTp$o|2}lYNUT-0<TmGj9sJDZ@oK
zIwGk1A_~R*7Y}3H^pV8#$+qKjFP)*LUurfuu(0J*<cQdiac;>3$fmJD4X9o{P@J>^
za=_IuK0LM6T9bFT6L0qFA?jb*a_Q%z?&@vRbH+N8sOm{Em){kXJk!VSkA1=0kT~P*
zTnna+&)FCKsjQ4#Qw;_eb8ZT>IB>7;7Mn>=aP)Q&m^H^`5;IV%?EY98><lxE0?Hi$
zo*J@N-7PyCAy*zt9@NB84yGu+1A;^8=ehaLOAEP~wDW(oYZdyk%U8Lxi-wpT*T*|c
zFC6xNbyjI2dbOwjs7Yv2dZ<W06`vDeEP%HU-NTsiEP#ZQ2gfzK)P_LB$;Z&gF9Rx%
zH>&xEL<Dc7P_SqJo>3FJ7%0xw{7lu&uHFKRH}P=NoUAB^(KX+$n$Ks9G#_VY1{PR9
zj^7@sdi;LB@2wWY7R#*&hf=%fqAi(ij*->DeyQ4@o$V+{j)u{<%Tz)aH)GbpPBSJ2
zwF+ql<XPgLS2LL%nxzZjJoO>^bBf3wFi{J=4E;`GdpP*nDy7f3M!&TH9nxQ7Lz-In
zd+q&lkq*(q<SzNcxBcOY7fX3O+Rdb1Z_=rGVy?|q%TaY@+=LH(n+*X<^N?*M-mL&K
zGXjw49XxpHP0RMM&;{wa>&paPDiVCW;bNA?0Vu_=Xpt-h*?XEJZJU$qh;-0D*(R8k
z=ip0G@QiA*Cl0Begrs1dIEXAXqe4H0_|zfI=LCxnr)zWj!u)cyi&P~5jOkIK?dc-U
z^CCz;)s-)%N#gx{XSCLKG=p-Yj>}kSK$>uYpgv6xecVM#_%b*xcG7W3%=OemM(DIl
zQ1=~+WkYcJUt{mT%;+<)FmU{LWACQgkI_0Sg3t4heke;F0yantu??BEV`D;r4D&oV
zTY<m0_+BdEuXx4y;!E4>0B8+DoWJ!|tXbU%J3tYepHIC4S!qI9rexbuvR8b@%npP|
zuN$tSN@6e34_<;ir;$j<8?X>$fq4&Q@1OT0HQ78EFl~ib(>1<cEd5vG%Y5AKw=Op}
z5A&ZT*(*8{ih?2l0qKA8dnY*ZRG3K=dW)RitT`NAV7uCAmJ??Gz9WMflnr1LKp0Q#
z`pXHAX1UoLob;#A^EtJm#R6E-H8JtL9yV*Z>=?%(1Ce*r=K?eW;!kkzg#ZF2{A#$R
zlq(6$lPmKKo<u0^QIHV3(T9h1n2J>x;4;dDI3l>t38l~nI=F}6Rm^q3fjV*erT{OF
zB5drKQB1PPv(#3>p?h3RLDct)cRoE1EaDQPBgTcUlK1~8Qc)Mfp26K*BfLtW57*eN
z-wULDUYpDEH91ra1^WnsIdmbn#UwG;J-3~)eA|UZhzkHwSaBZshKrjRuD(X381N-`
zJ6AkhY$QkH6et51W!m2XBCtiCf;kZLzc{N~ve!=ZCY}<zrbd+lDJb!7zA*ABOEJ_M
zJ@D1UmbJH2g1ZXJnCvmPm*J)-Ts7uKcP`hAFyWPM)1sR6%;<X*a2ay1&0Ve@dCTQ9
zixr&V-PSnG1J3|8+q_+#RH(QPN_RDyVby8HWr^TliCMWn(<U=(A%C`lLN)*W>y=%T
zwiqlBZ_1Ku7v9Rk3RhoA|GuiKV1>Ij0<RW)FEr8x?sap9PvcS@C2Z<xIB+KjT%lx?
zT?mUftEh6(R~?3zHbU#zxjgc-5&Ik#<`b$uhk;9W)-0lm(sYgUdTaG-9_Y%>$#JSo
zC4%8(S-JTzS(9sBXj;UT!~Ze5Fj9UC7ntOBl0iN=NOS^&hJSCrPrkM{>X<Ue{6Yx*
zn8jswUIvY^&YW*z-Yt5QxJnt63+^kaeCbzqm?xy!kRv;UsIt=`8;8Rqepe9-Jo+t9
z%{Z&7Wd+TPH_^9Kju!!8e;$$(>fHMuXn5JG$`N3!1cbRPGbL2@6AjR_wqbD+1!4^!
zyoqFD2p}VC`=j_KcE`WkNo3+mFFC!l=&0EiT6iUO=Q2iltDAw<6Bp2=Hs_;)?7Q{2
z^JR-a7C*};MiZ81SXk_9{E?p<IG34z2aVNEtee+_e$cd3*?E{W<#eUAyKwVI%M@E9
zb5QGlEN;?^+Q#O#odXsOejAFdnuk^IF=jRHlh(^v!UjcuL9nW;26r9kBUJ-$kjkk{
zM=Zk3`FqmvXWQ)qI$}WtS3UAKmKl+<o(W<BQ}o%y=a=krzBB7dGbg*(Ya^4QR6xS8
zd};|TsSj6@CzI>_Zes4xyarBlKoY!Ng}hUN>a{`tn>%i>=NtpFExOAp)ukIQaW{Hi
zpR399{3%I-@Mdo1^M?odLK+Ye39@B%O0wXyKdmnA*$o29x2}DL{COGhqZlo1Dgko@
z^VtABjCF$hCapZKFgCWRO{Iv$I98qmy<v$Rz2c+ZVl*m54a5^!xl1%DY>|KHUb#jA
z-<2#U)?z<71ERKE+8X0Eu|foTy*{kg0Bku(Bz@XWG>LYua<Ph*GmTan-@7+aG8p-n
zZX=(n1MmZ^AbB+_XT7>Y@uL@D1$_^s*a^thJ+p*U^k7PPC;8Hftzv-(1Ru8p)+&9Q
zh778C)JZ?~ADN`02KOkYrE=-t2|DyMsW8@?cXAOhHon911s5t8O!pUkUYaoijWTaQ
zniB)Y#Ri*n(CezSLM!bKEZ+&0p>$T0X-}z(g=*_p>l&-!4Tne$DOulUbi%bl(5<k~
zxS_eYh$hkNBX0Pd2kcASgwAqC!}js<HSV8(_5cuFBG0d32Yq(2d!$S}(KtRJ3-ntZ
zh}su^C|3)P2Ri+TD&zWXKmF1%oBQM{$?ipQxsE&PT9NLP?%Xx{a-r{g(vD?!)x(y=
zpvuCcSoCd!46hC^IriILZp(F_E9H6@s!L1{psMI(?!N|*fAcqGV`Tg929Pd|&4l$<
zSl{g)LEA;yWf(pYbDJxDIy`I6NtTq>`H-%xNUOg(EcFL!_0P@oU)P|Jiph=j-V#jS
z>Ix_je)u=Q{B*EOfUsXqkMkdALs%36LBxnL0k4z@3b{tYrOJKW&{zG$-Q79gheKGR
z3An$I5yPB-xT$By`L?@rdcN#<_QQkIwWgJY2YR-2EHS2Zzqv<VRc-mYKW;p}Ft@jK
z$+$Z|?cU)`c)gezt4o!_<AoiI0&X2!;AFkuZmKGDrfa&j7!ag@Q6U7#4gC`t5ILce
zGpLy82tgn0bAQc-gLZJMX@A~TNj_apsUZ#O&55HYoCV1vf`?1=@1tL8*zhH}UOrOi
zGk^TF^~ldaeh<PF!J6=k+nJ)(`YZUa^Dw?4;*Gv<F}1{4YctDedsDV#EY>K`s^awZ
zY!ap#^Pvy-MJm*a0m;dNB$VFxUXXESQnvv2b$Gp>#?q~Om<lI8zRh=2ssQs;+T33^
zUt=9GvyEwjx(n|OR?x~zk-ZVfF;y%-De<bAhBTN88|RFwi<%+HXdX4reu;%(iZBo*
zdX}gVGs@ERsxW{&f)Q`Al0vwza4)<w4%`xsn0B$FJZ>RYx-3(}okzJN>#Hv>Q-5eF
zTYAZ97E-QZWs+|U3Y{&bh$NM8Vbh&YHc$9ct&z3-7-TU*1R7Eg2U##Ict%sn+AIQ~
zsX5o0q%^~~Ddl;KSIxwEGlRd=%QTv_9*y%(>S@=3kVm9K9K-8j6YS5V6WKl)TW%%T
z{l2~tI;UV20ewm{7D-^WzUnqH&)ECvlS!jJ^NtXb#fdOT=%}IUcnUdB1DR&5teVo<
zC684`EMunBv}$&f4Dm`(rRhI6R7hs;cP!e)be7X(1{2ww%$2?@Vev}kM3@XMo)h|9
zKqLq1J?u_^d>4h}Hjq(6X>)8`n%j|jO~B|Swai>mRqwW3YaFXgNt63-Rbiy&O>s|Z
z@->U4_AH;1sS^fj@Wxp>*KS{Rs(|Y)E~O%FQMr~dzuTxVcRiY*Y$UGRD<yi7^qdk%
zBCHb+wtU;!?K^7ga3RFV1wi&1ETCmV=GoC+D%()t*&{H1IgCwy?`*R-wf<Powsw@&
z#|1nSP=+aljYGY?|AJXkN7zTu4W0)WIx1ni&=yc2EURNCgW@jk(azg&`4c`M(*EJK
zQS%z~&H~uC5ND%)1de6EUuU~v)YTtm(qra&S}#K82XC&v6k|$5Tu55CAOt$uP1hS8
z+zvQa(8qXgVL~?@pz4Rgo+6%1oAN%{9aK7TLtoLjdCk8_<^>wtIE`ElfUNExV(Tg?
zuN2x+P)v$75;Ty>Q^F>)&Ixi|x4EFM<8fk!XZLV**pY(Pd=-Gz%e1!!-U7=8p)s-m
z^NI3uf*d`=&-js~xV2}ut@qMD6W&auJQC^i_5M+?ZxW-N<^|jrxlgTI6gtjdTS+zG
z+QmkGHDNGkvcUz`E1&TZHAQ!R7R!ZFnLJF2<jg7;ILr>D)-RACNGn9upE0m))ibe)
zlq%lu(c@El41w%?n0iDd)WJC7JnA`dq4sAfPyHpx_R?_EJRAG>F(et}0v*IRz==x&
zEHM?}?yXX{v^6s}Q+;n&YN9o+Y!CJZTq3_JvVMaEgXw4iTj)$s_dU{%;_p-*;nc-<
zku+E5K2)?N@b68d2H6L4AUA1=g+o(IE?deQPFY*>BKBbBR%@7A_EIV>kAX{hPv#Ov
zxDY#(unS}C>R*JAaDWt3egcGHvV*bE%JrrL{t7490Q_JP^wR;I+}gZ5rCdy2S-EP0
zMfENog1>F%+tpWgn0N|B()Iv*&-1w)#ybnVuO9W7$)n`!sCC;HuKU2pS%Y=@1$(E<
z0S{SUvi_{2I9S<GLg4(F+T5zfX5fT8r-m80M?H!wqh8=Ik$`?~?j>+1L1_A9iJ|=|
zB~-1;wt5HLVhaR1Q##rV0+33-o>wifAt&CfsZv27k4lAxPyA8f%ajJ7o=1|J5NfBI
zqh&<s7FiLIr>jS;J!>#jGS0TS=D<56#G*dhT}SgBNL%wy`a5Z3@<=_}T85~dZaXIw
z@|r4TZ?!XQig}&TLln~aGY_uu0)!O4>7q*gG*zGvX1zU5$6vqUf4rl4ycy1Uoz44T
zggy&nN2gMbH(UESwS7<&8k`%H@7`OqvBmO?R>%bBQ$A>y$EOZnY6&nmCUnKn%5{Jv
zPn``GGW1)x^@wQbtZ6&qED$!9@hHL?jXJZ>07``2tL~5;ExW(>83hj#sUy9+O%DK<
z$_-2X%0!W;htOY9u5>qCRY!An88x`>YAwW9)*kd!*)#kjzJ?y`a8dOH3SZO+C1!%p
zTpjRN;8j|(_T125E;_~jTQzO9F&$wCKNxKoI~Y@7WEt^&R#>2_SrvmhV+=kua}jZL
zFQiO_oyoDZAPE!$<Bm^%cNQy5x)sg-E+NrZCcYUDI{v4gjJt$<BEViGh&w=ys6<JK
zPHYkYC6z5<hQcL@2px5w9@R2L1~onTmreE}vfeku%rsgnISONL3m?qvmxky2z``w5
zZ|7(9eCRE!1wK`eyxkxq1MQlw!!9*DXhZpIGOZIk0+{faQR^_a2n>sTikJTLDy?^=
z-9T3bk((pyJis%rb0)L|XjUR__`9r{YhNx9cM{N}S;?`}B7k$k5;kB~8v`G-BR_)c
z%$J2j2ZKJ^j_E;*<f;SYIr_Z&ZFS3D`SNh8`}u$gCYBY`Jo96H?c`}2XS9-?AS;SX
zneve!QgOmAVQGHy3QWc}0<kNaNGH{Ll;UwhXDmY^&)s=qT~<e(lkY64^OMGbQYb#C
zLN|CtuBhzzo^D%LwO(~g(a2|}m_DrRL<n||tqGpH0SpEk$9Z`ApqOfBkbnkq+DYCQ
zT5NP&YK)7Pjv%K9x!pS^1*8HYM)eDv{?O!*Kc;yM8xrq!s1J{~;T5)Xt2ov*uJfX=
zqHi8(npuw!^HBJT4r=%=4-2yV)?hW$DHJ%ybHBkDI%YG(kFmyMnByh!aJ(>S{F@~$
ziHAkLN1?I109-+Os2TC-fqi;V%aN(z=+x=jX7k`zw7}WcBDl}(^Nz?-RJ@9KTg@B6
zT7h<9L_Ra0L}WV|*rjZ~_A^M7VV#8c{as~?`_^P@gloZFVqGFW=97EzTQq+D8?D4-
zRMWo3QcMB#hb(f>-|FSSxnZs17Pmi)dE+o|W`KJ1``C6ux)xL`uKalS>sURp5axtE
z1c;1z;OL@<aiL-slFC4|TNL-kiiyPb58vx8Z)e`%OmP6YQtU$Jk}&VLBW0-x@~siQ
ziPu+CxW1G)A7E2RCrx?F^GJ;Oj6MF_oAdaAw#5{P@-ciGV^Oo=>g!2;BNe(Ia%;e1
za=DsO0pf__iz+)E`^_xMZZLnDnZeh%l?g+f>~f`Jg)yf=OpS?m>poWrf`h_T5ENw?
zvh0xrhBzZ6P&no$XY)yWNVbl&@7<Z#;^d6RIoBGN8}NDx7lzR_*bncwi-93LT*nR*
z$sVCG0A{AZ$B-i@rs;4BIw(|;<l~&**Q85fQD?F;Gn8m?cpOg^%;>im+SqzMD19v=
zXg{|hV!p3-vv-2OLDeoL-Tp1{<@h&p!o>XFPkdFS{vjvb=Rdm89;5-cK2D_IBQeZb
z?B>Jm9K16~Gz~&=^%AT5A8)h(0fZEi>9X_L5CQM~EfJeB>>8r%Z_Kak8JRkWR)m3(
zg^cMu!qTDvQbBE@CJ`sBDxvuaWwGqM-=7!fCF2Pczu3R-*N<zMdOE*c>|YO$4s4!A
zPJfwdO92s0ps+gxN7gYSZ?r~3^rNV-6XJ&j702OYBHHAyLJjKT%^xG@?=NVQ4hmyR
zOPR<dh$=`94)QbTdcsPY3DX;-Ba{ThEpdZ5VpWFQ^oGjg5UUW8ods30<ab`be5Ot!
zl;w$w&#IP+0v{s$fl%1Mi4M@Cj{rB}v9fe4Wv`mcuf|u99DX#kK%?}LxRYt1#)IpV
zsuPq9v-f%)5~dJ{9Pd_Caq~T|FP}fI^m-zo$xc`nDrJ~JFs`&c7dJuIABC~l>*G?s
z{a~O-E_3asZOG?6*+dX#z>-<hxfZ=R!>=$bf`B&ggGUt~n3wSK3QTQ{Oh|d)P&XNx
z>q)*jQ$;O1)BA%cu)6`=aq8A+jV+8mSa(K6GPIXVWQ@RCT$6z3SjdrxgvD-Q)5=AV
zcw8=dpVD}DP4)e&SN!h6wA~D0b>SMTG+d<O1Z_*B>F>E8+h+eHtWbdB^$UZ&tdmya
zQ`w$KyKIDr#)MLjE~*dTZO!-v3C#y%;G4{k(C*f69pVkr7nAV|06S%eLE`I>c-%Ib
zv8Sr;vA7TUz_<X^x%s~_t$^x6U_QCmOoDq%zO8sa0=Oc*m+r>%d$D$AhDMDiy;YZ+
zvs(y8qxodQoh}@dXc=u%4LdSVJM;6T6law__qH>s8_3mWOyHC$ZYvfn?UNwXEX~&m
z@6_9M^OQ|Ey%s@mwOw}zZ42U9S42H>m)h8X8I;24(AFN66yhMuJ+HiU5h5me4#qG-
z^NVQQ8MStu;%7$8WitVwM13<up+tWoEx?D(BK%yy6V&>Zh=SWvu<;URy4I9Fdx2uo
z_CqOo**X7YY%vf43StU>VKM=s421U8fz|PYUj_*RkX!G;7bIYns{~5?QX8Moo-HT&
z+?E~rUOlDeXwQ_%Cfz)(@JF?uLRcbBuOV%5W!LBx^W03GovlFD7C|BoT^ujS`1>5Z
z1o(KiK(z!RRB#-L|JQ6k#l9{I7fL2hUIBnC7qqe$a%zJl4JZe=L@75Y{Az~?v551A
zoL&skfB3NQ1{J`Qc%KM}7(MYo<w8hvV_m*-F<CI4M_vk;m*R6;u?IH2qf@ak2?IOo
z(KZeQpaiE)*BRWUrU#&zq%xC02kugu4+6Z>9%jGfbaZCkpMB)Kv+^>$mDk)=OQ;^2
zZK?dputfsN1|#_lV`6-kRRZ7l67b0t`wp76=yL6BrwAmH{<X!<HH5U*7A~?aVxUv2
ziml1|9(7RWurzb%rdSrL2U7ByvPY{BKCFmJcJJ-u&5Jz}{ew+!b*9LwX=?r#f(M`T
z=Ef-HcO~xlaxF_HO&3R)_MCUrju{klQgHvPA}v<frbKv#L^u~P3LEeO&1c~S=uI-v
zA0=*|ZF16AuLLsZRDa0Yuolk1%a6+-fLZ=7JmHf+)_s5v!UDiV8!mk%knszSQ7*g9
zfnE*vZ|))m5lEPO&r22vX&M>x(F)5<?`fn^)hYvVLgSrd$r;H_{RDnsA13BV(bH@D
zPWI_8_-jXbo>NaDZcvGkK8Pjs6EL=ljL;ljH?ldby$UfTg%n}N8fxX3JKeSm<sf6F
zlZ(lIMvEKorSsk{vt~Q7r9FUVw`-LDdMN+TrR4v!7hUbA8Qs{>35rh1*}&<)y3<8%
z9e#S#|GEFy3I-@Td1Vnb+MhDBv8=weF+Mphow>CsouQe&jj1t>ovpcz)6d=Urv`0q
z>p*K~WJ2+O{X$;f)L7Qp+Tf=+{XaI5uraX}HaB$oX?*AShm0sf(FqFLy5Vd7Gvdt1
z&i>C!8;VZR*wNP6!O++dpPTz<GY8wBeAo$J6Q53ASQMX5+1Skq|7Vq`t&P*qf}hGa
z13tq~C;dPB(TOtt^ZB3WRQxG|(}}n_i77e#6vqGAiI<m7$kxi%LCH?v@V~yGb5s-)
z{AWGmfBhHUpX>kW!DnG)`(GB<Gpp2PVo=x+x~6Mx?4N6ge}8oc2oe_xOQKVN@AyF=
z5F`^LP>l6VQ@KtznxY}Sk%XM;)O@bds<!DQNo^pLU1hz0-V8i=-kgru;<c;h!0ib^
z*!cK(JnCQiJUlRM%Het2e$!k{4>X|Bw&uX^W$!#+;R6D1xQ5(f!dqK=cza#q>QDpm
zrIdS9snkrCq2?sat9dD7eI>ltRx!UUWn|<u&-o7atb#yxms}l{Kq7fXXW;Kf<BzvE
zK2o%=2QHmRsbyTl%Prf)of)pYEk7~ms?a%O);6{PhA)OR!m6#a?pQYazKRc28ov+E
zWup40$SPJ|nj->Mqz|gC&-*EQ<@~&}<{Fx5I@@(LFDVFxE_9Bz`C3l_d$9RW11BFb
z^DDKq$pSAOG2=&erRA7l!N-Jv$AB*RdIe^2`Z!^j1orgbiH&uess{db?{rqZuGzfZ
zM0a~Rda2nt^K_LrrhD(hu#BvB#=`<sQ4+&@TZfk=wH|rZH=)&uZy3;mv}0;GS#dO2
zC2<q6?WICAdrbEq?&tEC)!~}iqhfl#5g{B>?H`kDFr*GgrhA{IzYHp<UKInY;w2$t
zNN-t-k>&k8q3vaO6ig&mqmmQ0|2t!P##QnDs$&+m5Wt(PGIPdNNjLYJ+C}-PAc1a%
z4JJ9R1C|I`sRQZc$kJ(Ukj-FuN0IU&iUcB1g$dbwRtzl5E3ljEDU*G>E0qbmFagr(
z9_MXX41L-Nw~FpS6XK7Qydgo7Y?~3}s+%~AqA;-<OYtIk;f41gpr5OtHQ|1!dHB%*
zgd4WLBoG0T1=?R7=Uy%rt`jT|G5i9QPAoEW+_+UA_7{-jiJeGC)CRN@9@Fro$e3J)
z$*g=5BG8+7LgF2xYu+Zn5~GkNlM|d-7!s}(lm#Qr5)^JMHVQ%hu_HwO6!yw$`5PM@
zNyh^#-muvOLWilyj$VNpm#v{G8o=pp!RtEvZrZ4Ze8fJszN!Eka1n|~ISlrG$WH1W
znaI`DVZBC7@^ECeiZ~Q*6MbxF9a=!6&ZmLPT%l{bR+tV=YGdL*J2{Y+AhSJjiFKfc
zjS$}gbYQj%dl`^b0kyGdIB~ggC`PKoM{$5ZQxlTnDBP<ep@gCo(slvZ0K~@j>IFNK
z;*uVX(gu9k>HyK{YHH)?G;@AIu_(7*MkwIe%^L>xW4{TZz$GLmGb@BiW5?;chuJH-
ziRIbd!U)U`fbc0O;tIrxf$7=qF%v6<@pWgjkdG<j;QR_BY0y>6#KHL$hZCi7jl^Iv
zk?pV(UM$q^$U!iYW;%i$plS*8T<;ZeJYxE=uY6iTI*=1nOAnU08#n6=&dFE%WC^0G
z&QY-ay)>^h$XDYePW{+Sj^p6Oar0=96*S+-i=R!b1bQ1bjiM|4*zL=S5|I8Z=k9EZ
zk11l3<~JX(6xTT?Uf3IL87PnxxTa>rp)^Ow<DB}%EyPPB*%>K%Ygsey`h(tyDvl`>
z9{$W`cRN`;b)r}fjt2XwMvPtniG=MK8R<D|ZPTOt*K_1j7CY#_YEG(n>s>X_I9F}5
zT#cD&ag-*>05zPqWdsT-kAoy?4Bjm%a*&0&fjs=EnwU&WvAXncF{lS6+pqgo=5SIt
z7kaB)zGV&{1yp*QQ?(Q;m~Cbk(0{0<gc()bMUTt{V~Kdc?3YFa*htp?ovQeNsy%$Q
ztanUsDPYOm|HRt#*XajSy0cg^y)Yi~QjWj2nfOe<=#so^-R501I}X1jucd9-E<w5!
z8FD%2Q#GT5YF7F#<M&jkY5L+~dHxnFY&yZZB;M9Z^gB+!Uh`G}%9iRb`Dz2=R`pu`
ziU7Kyp7U*1E0in%@ois<Q@LNxh#F+1-C=Ws=c<HDh>A#qtmZOl8K>^}oxby$z3KDn
z<y&pkJz45%f^9utKtkTvvJ0A{{pk!V<)m7p>(lA88!dN~<~qSDW6v}hISNvO#IfiA
zd~<R3S!26tJ6)smd3QIkzI=6vXOr>u`q%}U=^N4m*gmbg@-Ly4f&PDXniTk_Bg^KW
zs6!`g>|$<c{1ayW8TVrPzoQEy3kU1}I=W;fyDBUFXe&&2v31GEBZMc;SehvUiNkh@
z$I((UKq5%?#sUzs3M<<QE5sE;Fp}^16$r)E5r+Q(RH!=#!4O`sw?n~7mN2Tq*zR<{
zp3K<Bd-k!+>6+d&z4l&e2Z;R$1nrkrgY<*RQ$yK39tw?$Fx3MGgt-9#wgT9(YiJNe
zeGvsp1M*slv`RyMLF;OV!+(9Joac+|1NhEKY*Qy>0FY0OJidEb)xtpg%g&>kNzx+s
zH?^5rW{O^Nicmd(Q6<6gj#|^}hdsk{+%pDoV~jga?HS)3Y>@3oaUgX^#_3_89Yegd
zF?N$020BRu^AFn|)M0qZvI3CD|Fk?hDj8YP7{-$`2GH!g1MG<Z)xDXcQ=6DZ9)};A
zn$Vy{A}$HVAk|U|j!Hm|*qxr1IgS{bpoe3NBHCkENyAXb-gN*lnf_2maks{(C*)dT
z{~1-Xrm^lOU+~rb$cTS-N!AJS{jpKNdl`6A^+|&Ts4X^bT73(#Hva{neGVw!@fk9y
zM+XpI-|Y<NZI9dTrK32wM{P<~v9X~g1(;(=L3eWfmILQ2r0X@U*u8musPA^4z#i|!
zAf&@)W8zUPJUBYFXZH$oloRrT2ma-Zi;DO`+#O1nC?Y&wER8Rn+@v{*915By7|j2Y
zll}rf@Kw0lmK?VBf;f|`22@0V$?t;;Ku(E>59w({HFNC^JuC+NIvFvWmYVfG9q{F?
z+A94uic0JoZ73$p*oveub&>v|0mO_CAi@uP#*cLZjNA*-+>6)>0KWqaBtRYtVz0-r
z6Mzy6D2$JW;@6amo)}Q2M^*~voQudD$ZCh~<Zrmc%nnZLKduL|0jNk1m+eP!12w^q
z5cFph;g@2F6TZ+WcKsi<SgZqJP$5Kl>_i9>AwhAVU;O8}ilx})!Do3UCz>bZPMGZg
zutF`l3T8-ezdXQ+=pl>yG-?s8e%vnQ*dd~W(E9$~FtX#(2DJ1h*#UbYTm{Jur0gK7
zLGk&QAo?L7a^^?SBck?;3dXq+Ln0DX#3B^}ne(s3;uhkyg^S0PjB28QiHI=BOD95)
z;r;rvm!F)clE+$5t^h8dRm!{Ydx6{n>Jd>!)G~i^cIpJciJB85D}q*3H8*1h=cM5T
z@qzvA7g%7PVI)H|y00`g>8}Q$28;%*`Dl(m7{ihJ#dT_xoJ-i};jlvvwFXuU%}|~Q
zo=BdcT+!ZvM0-wl&Fs2b+_IrmgTHq)?8xj??T}kBI+3;zt|Q@x?r+9#jbB8);C&$Y
zqH_qyks~3igTw_X^=b0rXGGbhTBY0sh>;#cWe4r@{aS<7B)EmS<*-S*Nfg9Qh*>H8
zi3zL;F3GY8HwnA&6aI=C!8>rPOK%C<kme9<ll3U%QmRpckl&!FP%V%<QEDl;6^E)a
zD?cb}Ri@Uc<+%yli{F+Z>HHNBs~S|$C{iwPQa%lyYh?VrD$1eEA>`iR5OD{3D275k
z6uKw0s}N59PHswWs2o%-XROq=;JhGurhdi_E5uX`I-_V_)M#rK*huj!JUmOXQ8HUH
zYRvu@1x_B;Ebz(c1LK?ioAVpjZ<^n7zq`*!7g-mn%4-&nO5ICu7bO?g7DX4f%9s|s
z7v>jQ%3RB~=lRW3&4VltEG}l*D{h_OtfR-VuQKB~LXJ0<f><V*iI`Q*uI904GMDC-
z?&inl%I7!>9+fkyFY~g>yM*3@o(kc9kI0DSHV7RTP8kM64J5=(l^Q)T6g#mve>#)2
zKeh+9GkFYqD0uwwxb%2_fPAC=B>u#C>v~fH;{+oB0|l!BYk|EDrXD~U&<_3#u8E3_
zGKgeFcgEmhRAGXnpJKCN>9!PPwP#Z|nbUKaL@^v=lwsPm_E?2sZD5mPs9?Zi44K54
z06prM4567pl|>bd0E@tqMw(AZ8?Z8t)|jISq=BaCpw3apP#e;;Za8asGqbkMve`8A
z9tk!;Hd8X2H@vT?N~2kB9)4QWy3^Wjt2lO<_O>y#8EIKJbe+$(z%%IS>t6l}3@iv7
z&70Jpo9$(oty{VUix`G7v^L3d736Z~sB+;vdRSWRQ@`+9c|}7jmP(h37%y-izJIyL
zJdSERYje;#*DkVwX%TMevQc!^c8&H7d<Ta2hR+sD7zrvVGc7+Bm^Rf}(jlyYt68=(
zeUf~lupL~V?ilo(3E4wFRJwHBR-I;BZCP|3o8GehvrXaK=o|aZ`i%vs0l49J<X6T&
z(wpS(El`*XDzGR3A21mp(#Jz!C$2a86x(qRdklT-`X^w<Z{{-Zxj%9s`HweVG$cH9
z4Wuu+KC%MBY>!{N-rs;^&>F%n!g0cD;WS~1e2;=(1<3gs1-c_{BZ(v8hE)ct!#d&f
zp~!>IBZm70`x$#BXm8XTZFby!ym{=qZ-o0I?h&-qOKO?jFE2<#B$*^)NV8!KVKiaa
z9C?{wQx9`~MNLvAtyWjGSIF7mw&1q$*JE_Ya>Q+<Nn|{dE;3u$+-ZB+ig9<!y{O92
z*(Gl!OGVwX&@$^ePd}QsT|B@kajEvo;)unN?~F}M_lzM86Aj@Fh2xnims+eOP)I-X
z&8rdh9Yf5k(+hrs!u2|g>HEn_3HPLIoX_If4%zg!dTpzBGUH@--uFE^{u+JSvzob@
zwat1?MaN@rzh}yeDm+xrYiKoix`%Bi{hNZC%q@M%8)>QOyxM3@wB_PdX4CWQ9(~R=
zBYuU846(7cT9vMcu92^>x0+uSm&@QuxsjkrRY_4vZrdK&biX7`swU}ruf;bS*G}YF
zO{+|(>@J<JZ?xc$tI|%>;jQD=cR7L^iv5jF;Yzval-E~PXo;wms+ZEh&|1|tmTzi0
z^LCbe6pvDs@vJhrEM8mgaPO$OxLui7Dfz9c@oZD0cRlIR^EL%m5FGBE>do<v%PPwD
z<U9G+-X>8{)_6MLv+Hx}9}4U}89g~klTKq?Wmi$?e5cpnH@@xWJHI(>RbgAwl7b5q
zLL7qYoAGk_I3N(>8B-hG=xzF<Z4KU3P>6h{wZ*ZRG)Hn7eU8hE>DJgXQdF|<Iez_&
zH_xNusCDSJ3fUi)kDMcWl#|Ju<a~B|dN6O;CFOCQ8OGySMb~M2V>PpSf9y4#b^3h4
z=f3s$Q3pMR7Oi8XebIhkSLWF3le-W8j>idx1I87C9nwz6rG15w&E~Zi{+=|i+EtU?
zfogX-H~JB`d#<o-yCvRA(Dk?D^+e^Z>bS+xW_m;25zV{hAbYMW{5|~{%ZK^1PWud>
zhxd$!^G2^PkjrqRj*QRv`@yI4gy2<wC}>Lfo@iZcgw&gij!);s?L%TIA+2PHkLp|N
zed1u%wa4v)k^UDK%a_)}$9zqOZfVcT_uQ57wbRPOrz%E`iB3wVg0JHa&AIh=N2)6_
zQ`>djH5Ff~&&ezL!|hc`q+E^<@N>}p`@g25Y(F%^*1!VlA9(UV6a|=>|J%kwRjTK<
zKj{4*0e`$-xhAhOj@sZx)^o-y1yY~ZI8dv_*7@}(;ffo2E8pEjF^ME>r|mWndT}HB
zUGC4&QO$@^I+NYH6WKasgp?9#x6~R|gmEI~@rdb#vj#YyxJ;6CUuqvO59zZGgTun2
z^_?ID(dR?(DK%)SEohF}bL?bj_sj*iaw%x;NOaXc>iD0BlNo3&su5HhUESpFUhbZ+
zO!79{YB)@1RZ!a%Ly>&}qSOYtD#W1r(!v#=7Xy(-+|H&0{c*rOWZYe=5hq32pJ2WM
z7J6x#Y4>q6a)Ev9pmYi1H}Qo)+FdH^-CcSj)PXG@84+!I%sxt&@J(7o?oAcxRhf8=
z7Mz9Hji*KxPVc!jh(w8C^Z1aR(h?e?5j>@x=JEk99?SJjny9W^Wg6V(Rg8sau<?bb
z3*Bdh0(Wy6o+XhQOLu?qf&SW399EsidXzS`sK2hSY3{tLYn{_5PL!APR9P&M(cUoR
z7g8t0N&Ug+o}M8iQpJ3I&5iNo;Xc{5ufkMsYIP~k-ff@F;x@D{Um6m_E9#*a{_|d1
z`+{4DL0smjos(+hjb1`r(#$yHLk&2;oMquf&RXho4QyBeI0b<^LMZ&E3}(}j@_Y1Z
zn2$L*S~8YNG&~;yeiDDWP?9M=k(yp_n0&3|{j$1S42Vin^oG?g$7Y&?D%7upJA^KT
zi0r6{u|mbg;cOguEYHdvLP3(TbsBVZ90?}DQL~nKYfQ9RKH74_=cJ-oDXad_{CQlX
zW^27D^m(zX<h7y;vV}8h+gTImb8PKYan}7xXi>#;rKz>6;!G!9K6;2e1G`--T;&NY
zVN{T%*I>1xJ)?sVumDc<$&x1|T%!X7SZMw_rFu^9vTMGo?zEV@d+AHq$4R@dn{8^Z
z;LL?!+*p|(g=sgRJt$TXZZ6FJao4|A&I>W<*2}?^<}%O*-{zQQ0_3kqv{q70l115d
zEzyjU*vd)!(nww_yqdfw1TB&kT>obtWR=ikM=iL(r4a3XByu)M4%I~_iaUFamsq@O
z{o}7HPAtbFtN@&(L=L(6laF<qf{|z&UGGDhROu3RB9J^7QDrE2hL*s=c#&ivPecJh
zDa0nmwH3q%XVPE=0X^B(H$S{o(W|*2hjVaF)4$|1vCb+-#B#`y2PY&ZiP5@yZnWNt
zd69u4>f$3gya(S_h!4M@(oE8DT0yc4ua#ujX2A<>d(@8FA~NF=Vf6}KkarxXdZj4F
z{J-0lr3h?al;{_@@-{e$R@)ZkzQ`)x2X?pByI$as&9d9Xc<FZ4GDTML#~nW^@y|6k
zDN_rs_53QXwAZb+=MtB|UyOC&3W(tixQ|}Xd$?=+GQ_aZCt%-w^lWN3gtDdhf=31v
z2f2*f^FaUP(aKhZ#$-&og~QR;lM9R}C%t-k8UqY6#z|k3*GyeJ)Cmwx<wm6TLJ9}=
zk-<X_p&&)CGw)MSFm6v+Q{@=QF4scp1Z*+&t3@6WPlD=r<vr8GZt_@)#kT>DB{1`z
zJ@GWV;J0I#!W(z6xm^)L6TjZ#t7ci&A3@^2hGpyzHYEb$<TBG4lyE=UVIB;p>+__6
zH@yC}fM9rTwJS23cXy(MHB<eKgFiv)S6r>wgdvX{O++50Li66I+-mETY7Y48H#px&
z`QnsNM%I;OjHWX(GiYPv%m8eLOl?Lo*#NXzcvuU!vVMF_9R9_Hn{;^CpD(>p3M>4D
zCXF&U!(>mt*F05Te>k;++ZuV@O9c$J8i6x*A-7pil*(UZmViAb{rjrN`ijF;rE%Q@
zZ{*$-iQS?4449}cT*1~Z*ZLWnDx`L&juiW*FzXV;Q<Qw5H=|IEw}=P5iv3WxMZNkJ
zSA}9L<L@DZBnU}XdTbu^mlZK3)h~W83#@T~@d+beuAf*zpL0|Ha`Mx2{KK*T$I1Wy
zlJ7gx7@0d7Iv6|s!}$Mq>HZH=|1YBc{|Vjyhv+l>ZxVfme<Av8%>Nyr|69fqO8}<p
zx@KmUS2=6eOBZ=fF!v8@Fs>jRfqiZs!(egH$Gaq#VU5XVQ2JBteo{_h+?LsvMk@d`
z{0+0$@9Fo=7CBS94lHU+L;jkMb(z3vx^Dp8Y~MIvB5ruqWxXAnTD!jK0X<l*-edg_
z474%zPpf6SrC)vUFQF8HQsNk)geE`^h&W;It}D_>m=h#{Iz@AA``0C+FsQoxHFKeT
zJH?_Cm2v4@_KNT^z$78F%=gE3t9s|$Z_K-sDsX>hP!8oU|N49N2`30O3=s&!EVO%C
zsoW#2Xv8ex-k4Z~7%@<USWtrJ2iQ266;Iuu0n4x7c#W1vSWse0_oAb~E~>!IHW8)^
zHT{ws1ULO9w}={oeK7c}tt`d1E5QxVtgJMz=Z5sNZUK(KvN+?FRE{yr#B08y7A+oz
zpCr_Dk|a&M{F%rkvUvbPsa%cKY!ikFmP%!1L5=&MaX3!~ClxArPCh$*C1v25<-+h)
zRzVjbp|?vlaa&0@-Wf2^mR^qW@%L&f{dGY>+*W&VYw87NCPW!sWQWlgReBcj85Fdl
zv<UQcZ)L69sI*DvAP#|<5U{l<+WTNW8|)!g{5=)8@K!!sLqgR1D8B<CkNA}u>a>c+
z!w0$RJm=WG*c3p8XKlh#v{aM2Keg)!=YifM1aw2t;=qkQi-`|cqE=P(L4XwXpZOo0
zR!qW~no#hUw5<y%tGMg^da3l-(IDFSRD4?Ez^6t@qe(OlV>VFQxxDD6OVT-?Wfwf|
zj9KbtHRbxdaUm!s%@KUJg0RTBrY{S|V%5^EBtYZwsTEaVbB<5W>Q4L}FgBruBculp
z`~0OrrTxaCf1{4AUHt{J13Aa(v_FC)eSP`*`2%>iL8JjkZ2fLVly2RWh)ah?C%N~7
zX6eUHkfoCLVT$acxY@15r5Th*K)){Bx#j)x?R440@g0bdb%$hf)8+4B?_TlJZ6_Dy
z{tZkI;6*b2)4!Y_3=IE)nE!#8|ACnQftde+nE!#8|ACnQftde+nE!#8|NlTthJS&W
zjP&gPZH4tkbnPF7_1jM$P(1f}<VbiV;!GD#8w1n$rTK?SUXE<6@Brdy3PsY=4z0DX
zXWCk+@CEPIAEU^3Lit_;*ShTv?ln>Fx34|9@d(vx5VkmdQ7If6wzxTkbhF_EW{v8C
z82e$A4z0rUk?U=(Z+H98r*$?yns1-aCkFzhmG^U3+wZ&GyZO5j)oUack%Yp?63kK%
z>HuhXr@pY83Rh9-Tr}VpK@(~=OuEm7_AQs+>jYKxT{P|GW|Hp^JbmnFe(7R_iWKvH
zyoce30F4{K)rV6+$%BLEV@aDuqlW96gXiI&dByWp)>f(yE>XU0{9H3ggjq;Lf`rp*
zW7FhqHFji?2BgzRDP#Ep;Zz1^Ct$d5sLS=G<A&M~gQvonyZzdW*EfZiuB^2<_9oH}
zNRnXO#G<vzn1vqX?FR4;C)lNGkQhB>@7kUl9>QgzfO(-bKHvnuCsDtDg$Tv!YbJtd
zas-@r<gZ<M#BOLK>Mf~)%uW+L)0F!s&Y*c8zF8omgS$tKEhxlekO}}2m$i34!%rG?
zw4tbfFdZYkP*bhDK}o=IQPf}tohn?rUs&x^Uo~NE2)^k?{|<o<tdf}}m>Aw#g?ldF
z$`D>;jgHpf#NS9mgHupnr!nA(gSt-R7PT~K9UMV1cE1R}!(~S$56QK*y+F2pzsz!e
z+JMEj2`cHg@a0&khtGs#J=hj`vdqVxP@uAnRy3vlyReh_xXX}mNR73bLeRTmYMjhe
zrrd~7iy0Ps9pkK<$K|f$D$-DNEu_ivX*dO_l3wDtETi0yA}6}cqb@1<lrSt4Vc7w+
zPI125*tZFM(uj>NAzo<?Gd#Z)he3Ah({IJ8&q`tvsA`151<-9BaN6g7xG@V?Oi8wu
z1mp$x`V>id<*e#9z`SMp9I~HF;0waNHxFeB=@A&+G~&DCz?xdv$G`L3jUCa%Hj30j
z)!!uMx~?LlpxP76_ICNSf{f*QiGXrf#^W%+Y=)V2?hg@gf7Z%f4aD;zd=KvD<I4D$
z_VVyRJKCwfUya4Q1k!ipI0>}#Fpcq~RNC$_kxvG%2&R}^-1;V5j?SD6QGE9rWjZTl
zu)vw2yTT_&m=r73;E1`y%?)vcVW19nl{8_y^wW1iGB3y@VMWjc=K_FduGMncw@Y@W
zT!JzOV6jk4D%C63j|6}fWQ7G}X?y8cw5CZYWe&M>=HiJ9{Toi5kHA%3$gld%BGiIP
z%I3@wxG4OQd*@f{MMQUnkG`5NeS&|R@Udk}R5oY(*VmtxEn2#6t;iaM*m_|Qg#gHG
z(ZjF)YpN_DFI>np`pJgUhS{S%Te`z|3$iKxA#6d~k3uNANtf!7ej5`Noh?ixbI~}=
zA=Zj9z2|)-loA3hYQ9|!yzahQKzkiV(4gQ<#OQB7DO32LaWpbr`Nzq+GFD*9oULwx
zPp_<J)$2lW(n$D3<U(nd)xtz>DI59!gRysN5+%@<b=$UWyJvH@ZQHhO+qP}nwr$%s
z&)oOk9dRC3#9j4Pf1q+yj*(yP^?@1F&Bh?pmvD=Mg|HP*<yS<WfeTphv2V8fjAhSB
zkr*m_(S5H<R)KUa{oEUrpo`BS#s8YrkN?uvQ$GF@w*CymOhp~`60pPp1Ti|3`G4Ha
zhf|aDPILV<VFz$1MV6<BhV2p(7{Q-zoRsifvFUK_dS5{<^4g!<zjM>YV0;kMp{1_2
z<++*eIe_)#`mg?Y8{YXxpgOa7Zuj1qz4(QSe#9s4VvOX@94LWEsyW6>eXc6O>T`M-
zu0r4^Mekb*>lttIag2zp*rFes0$JPB+O-F#077#-JM}V<{7uJ>>(8q1=_bUPRUxC=
zmzuV%xiVP*qMCSXL<00wp+uDojahEbhdKKfwg!b}73XMSjSJth_XX~rC;lo9fgGDz
zBs1m7N06H?Er;pnw;Ojz_d}&DNy?HEJ+T4x)AP8Eym>5WG`2sB#dYTUbg{O4Mg8l4
z6P!y!kz!d9weu#R%TEcq>iw%<WZQR11px%-1#Wf8lEW9@WqD43B2l-|0LnkNJjaXq
z2jnn%tdaeUFi+B*0TRHaC$;njB_^U|1HGC4DxdQacl5Nrm|fYK2$)S$TwgDJFHvFK
zZ>53uF*twXV42(Ibr?+}rMi@#+-S-UOe=kDrH7M9ukXYY+@_CkKu|AH1aS_2am0>6
z5M-89E#s;WPNn*%aL^e?l*{S@WG!3@gd4(Qb8g*i-&<^Gx;QLQ7|TN1KUF&RCvDV{
z1Vnd<4HF^^e0P13uRjpkY98jdl@_&^Ob75|Ksg^>I)M-lW)S+6a}Cf*`)SxcoHoyx
z@lT;jaT|Wb<$99&AJ;lKJoXq|nJNWRoo|cvG6(KhRe=*JYHzgKE_;`nQQn(22Lu`j
zT_BcT>Q0MbmS2oSEn(nxgqtq>;MN(~<y?1PEFe98l9g+}YV%KhGP}){)<To^^e}LR
zqJ%-_lKlRJ+|E4)@!VPHeJ384!$B{C`|5l1uDwx#LrdKK46aw^%!T__>()?!ZgzN|
zzaX0Z$a;T8;e0`WUL)aMj621}+6y4}Y6i@4@O7lBm)>yReAs`uxWRvIL)@u3IJJRS
z*ST3i0}4d2u>c{X!tq6(?G<*3_3tbl32^yQ#6kXU!~feZQ3)esPE`Z}*aw3f3G6vT
zgXbUf))B4ZufDpyBLK;C@+?@<VnS4;TXZMKS7smluoE%R%sx^mPn{O`{firvz*~&1
zb72CNAPVAn{Hle-!EA^Exm}l%bJ@gv&ch9eIWDDhLjHD{=Cl##hWu2z*Nam(wf%yL
zB)r%EC-dvi|9u?)ziZRKfBtLE+O@j)zr+$s&!_60in#@$8f1Yv=V8YUt38>rlcYzU
z(1yvNFlu?5dFl7HcONjNKnX7SbopYWn1TD-PC#q8$SPuV5AVB)^TQD*C6S=i)S;O(
zYB;;XA2Ov1vt;2~lc<Mx`pVoKuhBA0T0<hTdRD%nYL4IitKkcSzq)q9rWsS{R3(k3
zrjE3xm64~^IKMBWsi%{X%b-5pY1llRnmO77V-LmLhd~}n-;WR*0roOY`tt#n327Ly
zKGW?`F2G79BXXGfN-$#anSOPIea*SQ20aK-GZ)^?Gp=FG)sa-RnqL$aUM=q><xF0?
zMVmIgnT<v%xi6Xx0LXkLLL&nYd?3iNp>!m-`LyKuk7?X~-R%RW$u~Pk>!FW=SLyQt
zVhI2*L3V@pRm;*bk&i9F4&N(3T?ID|MRP7bMX;m;B>Ezr1y}b#@g@jkkC6ThP%h<o
zgz@5(PTZ=SCY|()HD(q`0K`jplEfyqVM-`)Aakf$_trZ$k#H}N`;cp?9p_sv3=sKi
z`dt<cT|=vz1y8oxA7rnsG)AlV<NbT6TP=HA;y3UN?F>PTxHsomTz1#<MQ;GhuEO26
zFSI_ki`p8;dHtn#)$doU+-CLyFhqp)QgKflyG5TX5wj5VqHcf35PK`T8^PT)KT&Hb
zkDC+m*&I?>+%<hqQkC2_@E0%R_sKKX(D1NmXj`?kWC4#E5e10m1{7qS;7P>+pxkIE
z)7UK1u<?O&%P6%Xi2KY#h*gSgA_o_YKBZ%jDPxF-nzXv4C%t<i>7`G@|0XsS**Jjg
z5~~U+qKUUr^+f~Yk1gtK9>jXp$(<g5jbPj)G4T0eZ-sY_8@*JOu}Q1b>!|(knYb+Z
zVWB)TwMAj_=N@RaCHP+kFT<qzC)s4cnAT+A!g;(@{n;9FnK6NTBRZmw5h4l$$rQbW
zS;Oj#4ja}Ro?`zkPrlg(EmT}J>G$_G#?LzMQ??ta4n=@8+{u>XZu_Yvt0l-HFGxG%
zA<zM;!5ADHAHFzhP25lTBTd4GjFkt4XI+t~+B-)PIG#(yzDP1(t*Iegmmu$HgW1jX
zj#+5uM$?*42WY@Z%((mt4$x=H4DJ`+Y}<KJGx)GV(C;1)?q*4Gr5&5tS8(;(&wRaV
z-AK>mCG9GFDOx+XLe0AmxxdGujcyDm#LB+K$>m1-h1kkAd;+F>o5qy}!WKlfV^PY*
zwo@h_ml0+8lMa}qUBBb{UEwUxrW~Jg-1*ng%7vra$qjr_-4Cq|u6UNvBUWlG?hV0D
zS8*RULC^yUgw*X?^%;Up8YyYGm>oeW5seNK+?h9S!H;`L(8YyATHw)t=-9|2|GaK*
z90gJgFqi{sMD&V8NRH_sL>dQVKpgJkG@Jw~N9KEVSo74y4L{==T1e$GI{mL2`0Gml
zs7fDvdK5qLKF>nzIBYmPjR5q{s?W9e0)A97WI`Fj{ktbd5SLsaKJ<&5?S!?&+_oF5
zNZ3frjm=~l!3ac6zk)kH$p~A$CwLM_>w~)=Y9ZtqW!}TN)d7z_1KbR3jiO!EbYAew
z2O^M{bjQ6vU{6+xf2Mv%D;VMt7W}cohCtGO{w*LBKh}^V$uVJZNMW{ewElXP$>{_1
zegn$kRD_;`C3lO${Jo<q?lxQV*AKBPD53;}Ksu@?kQ5bm-c?%|YL%>(xG8(2i@Osj
z|2eVz3n>m@x5C)fz9{4tN)(Ov{djs54Nh5f-Nz9XnZm~bon+^e48NWTHbK4Q)SlkY
z(}VZ#dYsr0gP9lqi6FTWt$zH2#mjM_INl@BNc9>(wUtf3B6m)5?N>XBQV=iXpkrWs
zF|c^7yf3%zP|^%tu8$v|5m36?!!Fv``!CRanJh2j#kb3+epzKfi~aM%lI0K>Z*BIm
z;uR|c)G-~29Dn0cZ7i6SASzt5_9uZA=y9VA&q$-x`ZV3B*DW%p48a0CD-QLqBg*@L
zBp!aQXDj2vxMOW1#sjMJ&=R>K<O&45J%JSw-iQ0x%sdjXLQZ1u{e0<0jG?W4>YLla
zdt(=zi|7-#%B!ux{x)88Cy%$e330=v#{17udUg6SOePlpU*pjHks3H$01P`!L@vhX
zIE#j!RB!cA)Q&4jXG~TUz1GUETNgPgfe$O^?2esKfPH<{QN1)5;?Rk?n9}bCuX&eu
zG;r&kroNbI_F8A^O`BYPZIS6I7bAmcZD`IwksF9AM!{DAyop(sH$*HyU)e<|63C=a
zhps|fJ{+U?wU@Ts>PP9Fmh+qJTYT0YN)O}o2JyX;vCEpLMR#?-&%vJ9YWYg*#g^fY
zr`;OCr39W~Hc?Qos&EyHt!eqG^zgrbkTHCqmpwFG4+qkpx20K)i3G{fr9sfF6?l7I
zIvWO5iGg{s+gdSY(4GM_afOOejvRVk-x@9<OK5M*Z8J7Kw)~7k!Kwn?qUK(at^}wz
zFZ0E_5u$K478I}Q1q~0yvK0_Up7XBi;I)c-(U*G$VjJ9mvAWDNXb{qLFA6pqp>?PF
z0d^Qgt4l922oil=_)31@JH2tv{-dh@haB=B7N-BYReM_FAHi)5@gKoWZzoT~-sPs>
zh&ow=b7RTldOD^0h;(8w4nR1G-w)JJVZ4`z8=cNSsViCOA!r_l+uLWG8%1Zk_U@wA
zc4w#O{f9i@S%R3hTx<JAMXClQ6~eK;V_M}`3dP{ef|f*v=jZJ_^|>_DmG{Tx^Hr8j
z)sjl@#@4s}{qFrjxAnI7H@uM%lMVq5$;Ms@!o+0}8;4E@F^DqiE}|-pPa!dxfRae7
z)&{=r_eJ8cD1@^~;-P|@MHK|>=hg-v44+qx&M5=s+Ui(P-ukjZ<2S|CeeooDH>FkE
zo<ZBkyV(SUoJ<NmlsHZ{@gt4x=dGN|#o6(0a<rrjd-4m4xO#DnxeO9IX<(JH>(P@h
zQz&YQZ@`5=?r0IYm(I-EE3hbRejSs%j9|Qcx)J{>E?~+hfo$_b$;9Uy4%(_>YPQw0
z6F_l@-o}9!Uf`01H^n3~G&@$f_0=;fmwicwsXdp5suFb(72ZEUHClJ`FKT>$2~;BL
zDh#9q_v2`Xb)>ZUvaPnMp@uI#M#sZ0z4Eu($+|aQiR;hhG*qCWH&?Qq@0M@HAJab=
zE!&iy9YjGSV@wZIy=_Qg4SPB%M29doaDg?HExI{hZ)O>EG`9m5TfSwDT}^@)R*dS@
zm!gx=xHcAg*9p9HD8v~HkcziofD&sFdP52VSCG+VD~wS$3?q?7j1vA*YtNKO1xy0=
zgRYUxe8v^K5BB|>qa4OClsx*p90{tId6}m1$XdW(n#icyhE{3`O~b2S2_FzCVg`Rn
z9~Afy2jUkRGvkMWQoar)7Kf5bxrP?Vz%4y=LI56Bp6?mMBL$tV-P|k*1O^RG$qmm|
z@)A$#%^po{pbj8l3_!hAnz_?2jD&E0E^w$0NGzugg;5zJ!j=>>a_Kaa?E(~vOH4H!
z*;8cb?L57qq}tb(*frj&au-qR+l_AN9Qo6wLA*RhlmNDk_XbP@tN*%?+lInSjipQZ
zv{d0eR8UuZmhFZI_%kEikre8SM`?vh1V$fPB-=i=uuLo7g>LvhQmbJZlR*opD8E?T
z635GV2X!fW;R3F%C>hd_G0Zay=g_``po&rNtG`^FV&$#*O26e0>?&KYrWHiwc`zs+
z7X=&xT-g*R&RGZWZZ6)dvH73Xz-<BoO!pVd{JI)(hCD_I4#XvqP&6|tTP_9#0*z}?
zW!Rk}$7@@}`eEq(rCjhEbKsL7FQmPyN?!HpY@?lQCAB+OwjkFL>!Vo?WB6nI5=AT%
zGzh|&6Y^v^oe5ryQJV<9h%zEdF7<Mt3nM*_h-@X{+1eNBKm#uHb4XMHO;U%o5U3^v
zkHeQSDm`gVCY<r_I@zwD-ZUL0B6-pr4EXC2ePPNEREZcAbVx{f)t^pLGHkGollQN^
zXgm90&4J-_yznsvMd;_$Td1(Wo{AJbhJgqqP5xUz4tW89Q3B!dYgMkQ!UW=|mAtqk
z>nAZ~0%b1V`we!Qdp6#zD4PETnxC74Z%Uwu8EU=xqZ4Busi4-yFhK?*_&nMBBqLEM
z)Ws^3z@I8|fex#uQc(5$4}Sx}h~*6^i8fxvmDfp?eSVhSz`=0O{y2w+t=^N@epdb;
zrZWEHO1(qjrMwH$wKzfCuY0NBN`}n!(CkN%GMLYSIPtuF1rV_CZ1VOSiJ_H1@}8Ns
z;~!1yL1vQmM9OcSF?(F2ww2l|8RN8BhcvpurdxB6rHtZSjMM92xClx}0E~*Gv=l+s
znD!E39iB4l(TDXOeOPIhV0T-=&2H=7s@+xN_okmLTcYB%SAJ#nrcJOKh;J3&{bQ3p
zHno*wajO?E9&(3wg-*Tgo9)lpd>p^<HO*lb-#J+X+P`T*{N+aw&y54q3r0d~GjK}z
z5zwTQemHYIeEE@!Y8o0{@-4r$D~U27WWb6^rTHfc<A*(KH{Wxe;4nG3OtL?7F18h)
zb>%*H&lp+MaQbwd!GKB0$dAB@Lc}$~Ln4dUzG<(6JJ^~OZlc4-Jop*NyW~yw{ah%=
zi%q9LUE8Z@E{}hfqE%i|42_k~JkMEN`cA#WJ#;&wx%jYktp=4#poQhFVoPLs_VD6}
z$m!fg%hRqj&Locg%y{X$ne&yRB1m4F6tm&F9Y<Az3ML{#i307cYfGc5!vTaBDOGwy
zm?}Kl+O_wjV(e~#3=4V4(pYRGI4lP_!|&-K4P8+AfIeMZAaB=ki&cVyh!$28K#<Ne
z`Y?Md&w{a7<;8HlD>`(DfKL~hZ_dzY!K&XZ+Q`mMthH0y%fT$IrO`7et#(sQRJ5j$
zw4&+h(df_*v0mnzW~~qg`kV;Ytjg%LivJRX_0W*vSBRkO@++_iu^ZQEhQhs$uq*i|
zF(l0YUZ)v_n{O`mKaL7wPb7A`L$sSEc81DiDh4ofgm-o`a{+?3>Z`qqTRrLYb78?r
zRe3%w5%-h}C<L=^G@9>MJFEc_XF_HgI9_=D>q|==42!KJroPH8i3**nx!kwU4IL-I
zje}TNZ{+7VE)Hz&p~tO<V4OArkd3sJ=KOO8R1=T4I+AZT;_DRKjNdh#SrPK3GqqrZ
z?d=q@NEfc9%O~i;8b+C78tB%FGeqggKgGg(n;BPMuNKW3i~&QxkKIT)QQ#TjFr;@^
z%b>3n;<}e=7%UbZLg$>0KmDO*E;|i8XLCqwt4agL$qb-0T~60D>ZPJ4Gl{`<Q$PN9
z475nJKFxGDuER<8qs@xuLYh>^km}youwn-^ayL4+#gy~{`~J~=9ugB8vG}84u|F+`
zTmKwILQB<daH7Ozm~P5?V8?S#<~VfL_uYKER+xFct(2|WR1IwN>z3O8jfsG$&x{Iq
zB=;kitAnHB)Mxh?Sg*<3-1Ii*V7#7jvu5qY)S17^R2s@IgG|nFlV5jY<8PV21#dcj
zjPdtUi21L-02-A<?+q4^=_=h^CJU(PjyFr23snh?w>A`Dj8KKHqEj7uWK84^vkE#$
zH-g2)R@+{kLC2_EDPd;nsmL%vj#y~osU*^!EDPV}B<eU%D=z}bU?@g}{0ptU)h<et
zo3oep>$mRM7Gu;V>4l}ZJk;=QMu4#r!`9!&i(eV>zn&2QYXM4swVXoL-J<Jm9gzE0
z8$^!@6X}iNOY&|IYu(5(cb<O0CFCfio%8b1@O;Tu#e~c`iaLkcOj%YAbH8u7-YljS
z=XnKWE0CZb#J@W#NSfsx<^5L0@=TDpJns(WR-@@SAHjyUd2zbp^g%*xsB!(NlF!?G
znvpqL4&uZ*R&37?cHV*rwq@Og&&>7qExp%J+$}plP)0p)PP8F!dI*CD$BE>o9TL~o
zl(z~7+rGkfGdJ0O!5pKd9{=Oe_@AhaENuUE<IlgLF>!MQ@n7Rlzi1sz%n6T{eX)_S
z%#nmb%uof_nh${Z6&)2xJz3xK{pBJPH#-;zWlgvJ*HC}n{GzMxVyBY}TVZ4Gj#p=w
z`+GJDlW5k7M@v^WB)6ghGS#wGd-``8Bh@J{eXs93ZlMj|ND$cuyB<=R&hPay{3P4$
z33QE_$(UC@^YqV;g^f{**W)bd&F}H`=j!ba&F<bJyyIx2>aET<u***OtLfGp#j;qD
zu;*F;wn6p?@g>=>5^<$;yzQP;zgbHc{ejNTk-hTDbiNjDX2Zsz3}&z8O=O#5YG<hR
zS3~(->)V@M8dD~!Gl{P1r}TOIk>(bFj~P};2-;DC7^UW-mq>;Xv0&MmRqCpu7Fw3Q
z0Sd)54kM5_?{iWJ#UM~Ik(34jakcTp%}`Dla*dnkK_n&XU3W9Ba3D%nIA)=c+6Wh9
z5a@E_PhM_Wo)|?ICKz}ysinKOCeF*UeJ8DEj@Ou?A1xo;hJ%vR1Xh9_Q;Qi@56MJ_
zWfeh>87NbOz(Gen><>)-o|(ugw7K%jbJ86S*AT}dkJM+J=X!A>@n>lZ5zc@KAJ~Zj
z@*jz?FuJZVaz8ty7`r$ipi~Ttd+pSJUY7QiZ6(6@Wr1u;3>p#@{j2mRk?jZ9ka7>b
z>wC*|rdMw*(RhV%0xz?uR4pt&Dh_rTfP~_$jAYwk{fN@0b-m5R=-7d?^V;#{FHKMA
zw`}C=a0UBbq;@l93$lrR&<9E$6Dx9r8M~|mRE_`92B=|B$DWZ2zG^cs<r1`;&E-G%
zAdxWMZcsyg;CEMPjm>v$7+4@|68Jf>v?7knbDUOydMvJ|t><lX{*VV*{ukz{+=jIK
zWpD?r9P2F%pgjZ!^YT72Pu|wDY|=xuYvJx=75V@Ng0Uh`N~WP*qOT%1+P$(PJbp$M
z#a;a~>ebNeaZOqN9+rM%>-}C=O>7<ZP{>m&aD6&a95#i^y}ftt?QMjYvdtve(UDR<
z<Q_ULSR{|v9^o2I7py&4C6Vs#XL1OMsd}^|LSEM<dM08e$L_FlWyoa2+b|E(>gb^1
zdz$V1iFX(V6rlj9en{cdd>(aD&yxtdb7EZrA1N$;*K>nqygr?^K#)Z^Em%ur7oA=3
z!+=35_(CU=LjBiPNV2Itc5gR$Hnf)QBbnPyZzyZ)h?*HuW#|{cgkYI8YO##;cz#cr
z(1j!D%C~LD)H_MCkX>rn%Rr@xr(?J6gEzLL%^ApG_yQI$6eLDO{Mek3Ccb-wX&8i1
z${{>E#Ka`n?zh9#YjmtU&9L0FZ1rdrn2^E3;fi8}Wm%6$#R&NZe6!vR(xCDLXHhLW
zONt>+#JKFWX4N<jHR{#VIbN=%S1&zYWX&3hewF!IO4&l5)}ZTx@w+x2lHx{H*Kl=1
z_^dO@x<PeFH9_>yggqrzdb1Ei3Z4&{l$kAsAGkmXzB`G%48v}z&K_C@;)`6nf6m(v
z;ZnB+=-*ZBwZ^C_9(3zy<*1@(U?~b%dm^-}lb+ygZ9P&wC&f38XiyluLatFkSdlG}
zjb$SSEWtFnxpmrkD%QzHMl%MmWzwv9gRGdV%rRO(_dOtJCBIz~?84R|RkXdZAVYIM
zez(dr;4;-Y);?jyeFDHLKZvJ>Bmt+iyG(Ib7?hj3V~2#v0P{92h_~b+3~%{(5Y>9|
zu&Zi94`#`8qvt#O4RcpcScvL6)wT&i0Zu1nzo5Bgt(WDa`3y1ZoNF^Sjt}>0)(0}u
zlnG0{Cgm3)ySBd}m4&tnA<n6ZwPYO_MUbuJK;#{Ru(2MEuH?^RT+OZy_U&0ox+x#!
z0^~M+nv_Drawq3oje9;kGh7xkCOf$`GhK;5H44l#v~RL4dR%CGr~)Kv)t$;d6qiK*
zUY4#5KY^fohD*?WxMPN%L%VF=i@K+X2A;Odxt%gs>y3j#g0lhY_*ty54T~b(;ssk>
zv}E^(=Qzx+MC$Jb>0S2faQK!8ur4>PWM}V8^W569oX1vtmw5qxp?qDHch?f2F>LaD
zki%Fxz%wdMyv}d|^2AdIs?W&OhGSH2Zs`OlrUmrKz`wvCSP-%?%jCI^ajMEcvD4t#
z>YyL&#c*IrA0Z#qjeRV?8B-P9^N3P2Fw6#taq%m!4H1_ud6e4R%Jw3bcC>mzEU|?m
zYS(FAU<mGCpAf_R?Ls--f_B{$AMm6_jJK`ybiJ%ffzsF`;E=%ac#@{QxA2Zfl+vGD
z(6w@8LVp0P-`~ku=7p4T4Bf2VoucvObJ)JmagzjBI$maQvkkN^QPlDk^{hE_-?qhb
zv)bR;4ZJj9WIeEQ*fU#x4hFR*Hn2xGV8F#kR=t^n#ZQp2w;;uBHUi)x&446jrR_9E
zl7ln)LFeb}2iNEM1awfFvV$p(xYUUBoymg0ff!D1*9vD#3OhK%-2>O}#ZC2-OnkjC
zoL9Ia1dF1ap8E-{aLynw3;+2^Nf0bI)prfo)jsZ<L3mtWLN4fPh~Tb$#qb%8z^Kv~
zTN{7d<llJeI5kNK7RuP0^$w`W)NBVTsmv0pS4^`%iGG%5k8UEhq3=qQ1x*v_UcuCW
zE$~nP_{}))-(I!OX)dblnau~y7I;f2MtWZh$K}MNfg<)8@p({=x4jb|?)>kOh)fxQ
z!ubB+6@75yRGqohHDJ6-n@}=Rxf~gCs_q`&Q~BodqJ8gHT^KxdWGCJ_B9QL30Q)&N
zn}dV6XJJ=Oc>9FazK(8JurE>eTQ_g3;3KSUcd74uIV`GD^i>EDf&?Zs?reX(F=bot
z6~9LM6@4OzxyIS>RC$wFo8H>~1pBh%koo~q7zmBYQ95HgK=-A*djS@|_D?2)R+^5!
z2T=OOLT6D8%5Wq8frf2GJT6Q`?l+*8&O%fYlB&upar}iAg{-SOF`*KkNds6|X}v1V
zAedG?2@B2{Ya{#vXK-e8jFw_ngTx1P3@V}4>U0XtZfdu9R%7-F+1amLl-tmgWl#`~
zg1a<cDmDAHCSSkwa;0j<^tTgG>Uydys|bk-**2`VH|0y?pHSj6B)Al9q$QWy>kbqG
zoU7OFc{H5-!P97tFFzwDZ6?;ZV+c^(AS`y1A)Vh&L?-7YT(t<QK&I`BRmel5nm4$T
znkH!Vs0bC;iMj|%O>_Czb2hO=<e;%UM1=ljPEXEffC4U0KBZi9y+Qh!U|Ad?tjb+V
zWV(GjQiTNjH)(*!jlRRXYX6*BqZ3i=Yx^|nG{8!Wk!;^u)Wt@kOaQ)vPWVWfk|>wx
z+bM$G4FO1zWQOTU**_UAaK1N=B=}d$Lk#*m@w;L#omEWqnc21E7~_E9{-nJLRaMCd
zehPUpgFGD#ao{Nkza|^cN86OkG5LO~5n~oM`qek$!TgCF#$U`A=*99^vHJU)zC29K
z<3Ej;O#gQ!^#8YM{KLTTU$;tJt8F`Owjuhy{u?hpun6A>-Wbg}9L2LcOIFP-D`nd~
z{U|gBf{2t8Bu1{ces@4!C=;W;5H<01Z6Amdp!Mq4zQ0}25BjP0`r*lKPsr(aA#*@6
zCk-&cA1C>S{bLM9h;gS<`$HM@BeG|lrUrYxeScQh-`6*9e;I#Q-`CHjEaJYuXH9vv
zvwr^g{yu0Ig;)?xx5enA*9zNRlV#g&u<y}G?ni?U+)^lz70{&F>fOY)Twk0(u_B0%
z>Gmm#>>F>h^W$Vp*hsA(n6<Lv_N1RE8R`4uE8myrowu*MaAHUf8i-fW(C#Qd7tm(N
zn#iNEyUQP$tR^IMP>qN9NV4-0V59<fwwq9Lb#EI<p}89>*ur={p^@nI`|CI5s|6$y
z(}IPDG6JGe=gjSIPkit1ABKF7-2`%OR`3R4xbaS)EGoQ|+IbX^&`59YXRhkpQr+^s
z?G)@KYNOFgRAA+{04cJSG&%<ZHE~7MF86r*`qO86s3#TOZ`|K8Laa|#)R=&O;S~pn
z=M)SFzUL9B=x?UmoYF*+f?*pSp^~3le<|hdJ$97YDd?z5UPitC1mtIcK`kb&2tow~
zertqfEZD!z;3JcD7<xE@w~SCqHbPhr8)F+%YgWezw3|o1fc6xoX4TbnjL^9piL)um
zmL$Cx%5OMp)8B9AqT~dgJrOS(ozG@XP9%M+9NQ??3+k+74Qu5e+h%<>!KDW4N}qi8
zef0Iutr%A8qy^LVN8{lTj9lxR%v*^m=^9MLap4_B?U*Z!z%cyX;pL4p0y~i?PDe55
zGBKaubqDVcWk&a4Go3l%*2(gx#X6IR*;V$>iG=JS`17WnQB&3-SffChNJ&~3Tpd+d
zC$VT}iPcES5sI5^vKLuA4d^-JX@gOB8hsyFB0$)Xh1Ntu&8_p@1Sg!mjdF%U_yovC
zRh))2zt(c&!d4O17$m#01Z$(VjJwAz&yn@YfX~TqkwTi`CCg3-GJ228KTAZGPZH&{
zh3MFiitNxpkAyep(hMq_I_4qvLhHaGLe!rKLt}^24tRiAz4vKa({2pvI45q@*00qW
zfROsa+tjw>T=WU>9c*AvGsiv<d%C^Q^0%OlSmUN1$P)?$5deH^_zRMr$M?l+<%>vn
z^@dMayec(|TxLFJ-7EKjC6dX_D5D$%Hq+2UW-xL_>22jQbahAh+8Pig-bjF)xS0*a
z&NdvNh3t^DWSAS4r(;B^T5^rYYd%cw#!-L6r>KUE$%aLhG|PN%+U|Ceb4po!;t#<2
z*p7ueHKZXL`5(FRhwC`qhHl)Ic!1oZL(&>fzlAq+*C8oHti57*1+qb5^;3NB!Mxhh
zNsnN9B~%JF-TXPnVQy-H1F*U9?tqI-ILKjo-9j_(?JSB1uh{B0toxZpaiMDvFwh6J
z)dNYm=padAsD*nkYw-=VnwJGnRV#COUJJGsW>_D$9}i(;$aGw2M6X`TjEbG);$hi~
zYpoAL+Lnt{EZY`K{zm_G4&(SPnf>PxgYYE61cVRZyxR&JEuDP-doqoe%s!*Qfa#hm
zv2|#TplS%IdR$ryabIZ`%@|2^^Bqo3Of}f<4#Ax|SL}u~i-&Ru)jlY$pBiTw;yDvQ
zJ&u^=q=|Yt<l_=`R6V;X_qWhhHiqI>skan`C>CZg%k0zVj#fMQD?x~<>#)hxLYlg@
z^@hy#{`3&F93w5&R|Of&R|nN&IxgK8wj_|?5M~Ck00|R-?}jQ~!}t`+6}|L)&u!?A
zRoHmX`Bs@db*7B373a+N(REKb50*{}w4SX&B??k})3qFU%{BVV3I}JpLMgr|)ic3W
zC^tGK+9yxxNZ=epTDZTm9O;)_dE{H~mx0)CR-G(xuyRp?V~)Fhlu#kahDBsMDVV0_
z8;#j?Oltlg8=&5UO^CydFVi<={;t7wW>(!+_ahL+xEytQA66>;(t@YFTz1RrUq`#-
ztJla|2?W@JDgb^*naJcS^_DCOW^DTrS^hmlZCvlCg3(2j45oc7aM$o(X2Fpn3634*
zl4mD1>lazMW@H4fADk}>P9{#{ISbeu7tE6%)bGCI-LAK|Qy_I}#&=0Qg-@5#i)oej
z#Xo{?{z(pYXg@_j$uT6SJ`R7hzbC(?KlUEuhE2T64r5<e7@tjH%B4uVJqZ8-<Zo|h
zR&^a6c$z<C(^Sj*dLBqJA#)gDyYSn;SHW{*V^O0wi?C(OnaC2J2=nZx5YPX`fAL{6
zO8g$YnU?^7likRcy4>oP75J7nHS7OvDrYS0fmd2BKQAm5;yTwV!}f||K*^tBWqOg>
zpT(Kbq>Sg0NvG$Xmux+IX{d?}!WZBxy6bvBT@N+!zQMi(`(`O)WfVyWK|}B(p?jw-
zz1n&+@C=*(6?mxniDrcOCcR^y*DDp|+~({x#a2n)RAHF$&;{sX>)&ri*RlHq4coE|
zwC2BeayR3bJNPE*jVo>w-@%7(CgdbYk-@jQ{wuAPZ^2bPqW?+hc&7qpVc*m>KC5w_
za*|)64lVtnaKBz2{1yWD=HV^x6c_FB2*8@N<PqrMHo;k98Y1lYM=C8ey7+b7x*#8{
zZbP>jKDmtd1SZznku@ACb@vrW*k$?-no!FSnhh&`OTTxLV&;C4y%~SB@_FDT#=flj
zkE6nWa-x`6|LdcqWvq3Ft+sn@-Tv0%1}qHn*bR2&2t(VWiTZexTq|C?kfe}i9EX&n
zsDh%0Tk||XIDW;3nk5xH8o4tO-7_1fR;j>DKQP{$FMIcglRy-HiG?uH{s)f#Z4dba
z&l4OaQmUJRzw4#PjGNDh7*8SNAJ*^jdit1~a!W6Zt~Jj<Rtk+ee@A~>S`YIt6xlNL
z&8w}|-Rafq72Esml<{!)_3%|&=jdMSo+U?(N@_lp3{_(3iM$2g;dN8~I+>^|i&0k$
z2pvR_RL|!io+cIW{33K4DF*0+RFub4wuRRJJkFa#eYAELRm=}W79GTB4n&6fM@}k%
zB7DAXH<yk`Vz8q9T?$p(c0$&SQh@)@S3JYf`85DF`!sjszzBNYz*8?i4JH{t?jSc3
zX<Sq*u+fg!KTm#0iy=ONB;~F@93U>fadsoKKvcii_p(0wq8ucAam{_E<m#iO+cQDq
zUe$u|aCg6e))=<Mv!H*is=rEONBK%;pXkX1pjGB@t@R{mxpM1cG8qBh6S*v}>|Sq4
zIFIhvy`@xsmx4`LyGdR5sBED2v+8EDho%F(I~p*YIVP*~`*VH)z;Oc7G@f2q)97E=
zw%4~Mf{RF;!)U&uopuAfnei=r(DyTGvrw^oHiw>tID(V#rE}XbNG6x`N%j5>yz5JO
z-gLlydy54VWXL;ma<Xtt@((?1HoqH^tdJ|<W5TKNhn66Py^X<QUE3IRSoUnvgO#$1
zv2UnsmnsB<K)WXzYPf=$f=nJOM2aRs{T50obpL@}YV&%_S!9bKQI?1yRSHNLunfW{
za}9V4qdX%MI7;~bD`?;_)FeJyj6GoNbC83|X^Vxvh0H%(d??f?XJckE*af~U);+BG
zo^#|-&;zELL|Gla)Jo34tO@7{uW^1o?-jYeWAt6u*axhS33rWmdvj<zs9y+hJxKUp
ziTaYFZD=lHTslT5As~HPcBtkVm+ze{S|@%o80u~!>KgX52tP)ZZM7zcBSB^RkjzSh
z>}$@*4vQwV56jGL!r^34KpC5MmXvXi?JG;k&kpnZ!ikk$cIkw8B6%r=F-8$YYnC~S
z0t73JV7+E2NJOYIDNyM_(ll?yB-$Gb=o0<J+tu&&K%OiTA>sO2Z<gCZt%6mGSxfdq
z1V|wgIU^t)DYdEq#Enh$mpz^FJHKsoCLUg%({DMQ2Dhmjo>*1K22VWR4mdgz#J%4f
z<tENBb{ftu|H4|7_RfIl?&3<G%6rSrBjI4x1az!+I*!uN*P>V)etrue_j@X-nOs;|
zsw+&2<6>W`%R2n=r-2dWkez1bk!?ytnxz6JydSODXZ|jEG-e|gSP}nX$xw#s2wDR*
z{L&GCT}<KbEa(Ywz*+`D#!jowO202x3c-}XR#WfwrH=K2KU}pR(v7|A1ZyQ&H;pXV
z(U9EV<|KFF`?nt(AZMMl)R7&OQWI1s2J&aG<Swdv2fBYy0w%-l0qmqG0JQkmPDi4!
z6i#ohc8F{n0*M1V&Kw5$UYsGRtZ?hBHe^1tll0e#N30ubANWmJkTOVz+>oKF=>;r)
z&ta|KT+GFp)>QZ4+pM)vSIU9M3S*9Yx0)eJ26|Un54R1VoRW;wHk)_TDE#S@R!-!f
zLgR;+i3B$81hAG;ZS0z*A@QuNK)(+-D(zN<(eVc1y#*a-jRyAIQRt4ZObnoqM(RVc
zbvDHhB*LZER>LpD;sFA1)?jo55Ky&#!#ezsVoAtYNHI`L;>sT%<W4+Gk2o11xG*DQ
zrIoj7Ya*2AW14fPVW(;x++e=T4{9bDQ;84js8=uuw8&PFBjK|80+>K>_&g#}4+lXq
zS>X;mi-7%ZhIhQX23V!>?NdSPoL8RP%y*d9>RVaE1@Rb(gaaH>4LkY1REGQK0`o$S
zS{S{gGZ^4tJ1r5!Eg{TYA(-3|35>7ppttHK@FzxsDEvCUCo)D!T+wu%Qbd+n-ju%@
z=`XzKChgOaz-CD~TJ#GR`T-D|XL5iT&tk7kqw}h?leynMaBWsUX){pxM&9a9E7F;R
zy5w2iONBv*>2OQAE}6I-G*wV<sxv#Nqu1jXqMC(^DnIMz7VdrX?t-9*#DiQt-tQhW
zM*Pkn1(%PS0ut@WHwb67bQn7?5Ebm<JhR-T=z5+nL_yl7CY+skl7!crY933co1_&{
z=~4yIP}ipgAHszYAge~FZeqZ@4ur`%+{B5e;E$z(fwvFpJ-IW9flm|j$8Stw>Jgww
z#EBMrA&tp#HEvl*em}q*W!$Y-4H;%A#YyFz5X!Rj!C4})7LKa_-5)@7yzD_CRP9iH
ze8OZZ@jH{@D?nM%fHO`#3KRC?Std26J{8xpW*W<NGR~AXu0!O-ed!ppex}8TrSE<y
zB4(qS0TuIV%pE0<#tP>!r$y*7J(q$RZ}%RkMIa96aHM$b1cmN0?W_OXfdq<Ckf<HZ
z;{NKw3fwY|4L8CXaFbYWUfD=|Ee@wQzow2R{u;@2RrpV%aD7<Ac)dv2+-FJ&+-?yt
zo&-_e_hOLx7}AveqNG8vVnIJB@yWXT&HU}8e6md5XaRrm47Kw@MX%uiWmbJP51Pql
zu+_BK>{yaY&#>0-pPO?beQM>&q?y&OHm4g)v6eF%h#qDVZipI~&5A^mNFHfQqN1sB
zo`vQWps?ZelNf3Q>Q{BCuB^2LI1IPh#hVB$H0_{K`{saUEg|L<QDkDAN8!}6g;tCN
znzpJj`)6M6DMfYkqkIGkDqdI+NM|SK16#FJHq3mK(5~VROnl5Ksj&eBP~PfqudYfb
zUv3x=M2q4;0iQP>H5xD45s@((;rk^7?dl%Kqz~N{@hhJ&_5+R1kTylVwGK4YMVzs|
z^j9#FLUEYHo;dVmAcHM>@{%pHsHoJid|PzbWGH+-{BCWESDiP%HlIgzynAqDgD8SH
zB^cpIMtsIps6kW{0fpq?nX+QJY^HEGY24(sU%>fZ5=?1`4vq&CaN1Hw7r9Ka)UD>V
zT$*+e6i3ppc5Jh<hu@cQy3A2q<5QMjhNweCYi^%If*HuIhM%X)Em!&0tEEt}KiKTb
zl1s`k8(oEnDxb7iy@GqLeZT)s_tnUm{GjtYuiE{mX^i>*<|F*?i8w3Ee@)i9R@eO>
zvX<{_>0BX<5)5T{(oz-Z1}*J^_1cLCd>}eX1Sz?Y(RiW}_D<bx53zd!?YcPfXkSD%
zh1|U8!_V%yD^#-$jYT(p7cTtF4kMO2Vt5dB)&0~%(6Bdu*=)7Ff%KK1o6(sa|26(@
zRFA6&`fks=``6`}!C4!Q8LzkN{ov^S^4s7#=kBKQp`$=M0A&Vrp+5(69>tnN25<~^
z90Srx+7#^Ob9L5EhH(~~?B7_Xtr+UYw2?>;bvSjiQb>NlyI4D$WC?^C(OYnhZofM}
z^XD}D1OlTJQ4r`TtKLEaQu)1s<sZfRZc|IAtLn)6y;#7-ivSV4{3vxg)1c3MmTb#J
zuKs;5v-SWaRe-^z3|ODPDw8WBCxF}0NvT|p<${>{A#GEw0ua@A88G>D`armBnkS1I
zXzt7Q1V^ZmE;jbD|N1?T26a!ah*9cr@VO3xQwFsT{E5;ioY`{8V)B-9#v3Yl)2qaM
zOS5=%riNlB|Li=qZ6sEb*UfQk6>gj)f&{P0{knC%d!I5%0O2`jGg420Yu^e`0Rn2*
zZaKJidGYVDSx+`=GAW^E{lHsDK3>X+e0EU!=u~1~cK#ZZ{x10oy=D&thay|N<Uw~@
z6bK}CRwkYeidROPf+SIr*!|>*=<tEqvUG5&hW0nby;YfR?3zd#$FSEqCe1WWp855R
zk_fJnZODequD`WW6>F>RD%PlfHOF(6#CuUKF3)o~9eO|5G^Fy;Cr@<87<`@O!RDfe
z%dcH3jk=IeBJ$`Z?aAM=(LCxrZ)2v{P95TXPI;r&JSr-Gujahxo~1(-Bg+yBMf~AJ
zF%0(|LwfpJPlqHYJmP=lJ(@Y)TiZG7hj7Mbm<|sEkOD4q$sZbVjAQSSu5>zc(^`A(
zV2Gp*TIMxmQb@y}q>{d4&ymUNDa&M&vUv{JjDWa9RzGqR?-rY~Fjec%bcmr|c6%m@
z*1wTQv>XN0Z>L*>L9@6v&KXOVN|~lul9JMZ?xr~G0`hSzZYCnJ$IV=JR*|~RVI>QW
zYSS+pCBd10LuVmlRQ0n}^Ob_irwDp$EhR~PxK3IMkE}q$O%0tOH5l5JnUi%d3_@Ax
zQ8f$YA3uha3;&*;r<Lbh_TV=(P^ns+1wh~*D6@8{&AD6Hxs9TaS?PcdOo<#`Y{Kis
zvk6^(Bo9h{$Jt2xWvS71(NZ4ubx~72Z@S69Hx(8d47&40i=SLgyibOd*nx2zzrS16
zN64qxo5jsIYV7H8E5%Fu`t0q|da2{^;SBG1lVxiN@4AR0^vpeAwG*i4+!eU|`ErM1
zSmWw6;|EZX+gnr>({~WA206n=nTQs!td^ao1sf6mz&1AV2HDEnz3>Y~DS=?ddm!xP
zGaunX^0=9CH5`(b2k=|xlN*s|4Fv#5G10RF$yubhi59Tt$s4MYndGWDmdIzz-MlnN
zCpEd{V37^xSYByLqfc=b>`G21M0XeLuQpIu{lo8r6+(4EBW+J=*$4{Td!}{#GRh%h
z^GGDlw1fys3(x`P(lH0oVh1A;ytVPnXdDE%v%)H~Sg|pj2ZBHseYG1BZ&~X6DBlaP
zUyOwCoCVlv&O?iozMFPE!U8pR;J<dI6xmi4r*u?Kh?U1p`-I$n_HES|nYM2Po@&`%
z$y{5>1EXp;<r-~mufhzo+R$T-If0alwT=xfZweu<Gr3-}TqJzuE!O)g8m)sQKvYiR
z1SEiMHx7S9P^8%=%|DBd=I72U1igc%Va>ah?LB83Z!MzuotdG?aCvEZ*&MS{r;nnO
zc8R)X@j%;9%a)zlM-|F75@&f=P<~>|Q8(8AZOry3{v`<5d!UEeVDrgx5{3wc0m-eW
z9{y>dYZ#!P%9hIfQjZHcbIT@1KRAt8y(7Z0=<KgF?E;hlOu0ouqOD{ZEe@PIM|7&P
zVwidhEpb`y<(q~{!w!al@~AV%s`~0h;ex#Hl|TDUQNG??reQeh9xN(nBF7rc6Nb8p
zf~URdjgtMm4$SpEgTadlVQJpVh}OTv+=-W_84k3xp2AdQGNqXr`ACGj=W3jCc5h!L
z$H=r-6@}TmH0JCy90dJ}r}K#Rfd_l#;Pzh?$$47ublFDZg(r07dTC|U$xLg0gX-Bs
z&Mwn{%8IJ7OND@xhOthfje#jBn&s~_!20sc8@5>B+I(LCzs})_|D<Is|3m2dUsl-v
zY8lHvZr1<v{Qt7aH?GyU|6eWB%@@Ut#8gP4u>XgN)b4Oz3z=EgQ~_y_9v!k0Pd+#5
zulw~Lg{MrZVvkAmfjN^j!*w4$^uBk6d8y&L`euG3#Len~D<4HP1$<m+7fnYGhS_I$
zSV(u>N6v!TE$z+fAxpERB~Btlb3`yu>Hc`Xvi9J?iwS`EWW<Ug3i;NZ4Y5BfJ?$+`
z`8mvVtsOZ%Z~yk-Jo5Kt<@toAICS-h1$5ESA37w6=(<tUc7K(Qjy>j(1;+;iX&J`5
zrw~r-LkJBcoyx2M%aBnAxkQc)!U-n4r`n$e_r<9+b`%lt$Mb`f7ot^+{Nu6%DRxGx
z=PJ^?-Os}478XPy71sRSRi?25tcut(jD&y^KceV+03q4=0|48xi3}!*rT*u(4K}d-
z3SVvp!b*OmbD|kEA%%;;>l5sWbP#cH?N?ORqY%MM2-#K$NQj?puytzXLZGMhQfGsI
zk*-v|oHcnav?mx;V-1=*L=Q!@i}pm9DR^_J3+<0(c&(7TJ8eltS(BaVT<9!3ko{{M
zxr)JtnQM}zMe}*ReEtuwe7sA+0Y$YZ@wp<o;v1&p`Fj=Fsb}<x$`ol~q{p)l`5^I9
zvTtf2dPVjzH`6W`5hm<lOfen}qDjOy=OV*=U|Nn_q7$8+TKv+gdJx7cuwc@lcTWKB
zV({Fn?Or9-<SN1qwf0&-*hiz3v5Q3mO^NS%?z)3^n|WCQ``GYCzkf^Uc8oud;n=2T
zi?>nR?IOEaUAw#H7rBamW^fi=m5PJ4c*S3-{D$K1bXyf<{e(4Q2=Hb~J`)V0Dtc+!
z#<W2Tyib)zLUFHrV}}#di5ZG^q|X=$pa?aLrs0a5Lu+$>AUJ$nJU&VNB=*t1kyUll
z#&L(=6D|RHoK{flZ6(iU0BZh_d)7uq$}62t_misqvGngIwj9!NuWF%_71T$bNJSt-
z{7u*f92FqEq;@3Lf(CqbW#iVD(r_k><pDQ>7_V{~V~s|cAY3-f!W_htqqct+AreOy
zA~o9!BKmLINd*&0G!EoEwQMLVue8IE1l(o5BKOqsmOPncYrl2l4y4J<_dvVS4SclJ
zS8Y@#2Ipd|0`<CR$0q8xaJvSufM5#QD2^u5GgeWC*uF8SRv|7|mkHR&LTYu%TS=ro
z>B{R86!O*2Y~Y_$S3=a6&r+VoKzNMfduurML{Hsf2D*H73Nzp1BJ1g0ZS|7-d7D`c
zXsgI*{0Vs^QdJX=(md7G0~cqBu1W3pcwcE~l@5hQKE}M+`swgDDbF(>o9<q^@jzVo
z_jT2yPK5Yl86W}a3#_S>(qA&^3uX#Z4dGf6&YG1L-IGt;<|8z>B3cMcZFYjGk?&If
zzU5?&{3MqI>JygpLz1(Q?f}b)OX$|Rj1lrkyR>Rc`sCYg#Hlv3T3VST%ZN8GTNTN4
z=na)pCufwdYmLuVBw%YmBWF)mI5BY67pGS1O|I!8pyACQDkImx(ukrYG8U@1aBBeh
zq!c8$AkL|%n|!C8h57{1h>=*=DZ|lj0BlT2cVIw0UIx8RRV&s{J(QENrO$9>Uq3|6
zZ%$7}UKf5g?Q!&Yh-e@O{-g7&ZEpt$1uv%Ct^f^D=1Yr$T<7|I?p?tEQ!Sy(g&LLr
zHYBTKp993U7gXciH4AHIo#%9w1`d#I)6Uo(`Lf!ax;f0v<&SGG_nf<kA}E5FxByTE
zHknN$yV^$K+LH?`b~u0n76o073wV6R*eWv@&*UmkTLXQ{rQl`tH!i1k3w&MlfWrKu
zX>)p~``DYJ(X&wn5KUZNKfOcQw66gaW$IqwLc)9WMUt3V>r`_EYs|3tUe6rd&Gj(|
z7hT)93g8<N2fz`B-tDe#q$}0I7l|1;TmoVHYlZuc6Ai{k^|2{ByNlYt!Ht^0obcy-
zwkus;%ua%BulDzU=^kXi2YxAYE2fhlwCvPH=j7#=epj}0`7azg$}2?o?7i~Vw7#l!
zxCV{aar3`^!Ks$V)BdA!{HMPlHunF%+E2~;f0W}VM>pSAaV3NFSs)&lHSco6U^~RR
z{FRaxOt=I&zv1YD)SS_&*IN{iG_v8UM4Xc#jN(EkmlN++s#`MFgI>l*D~HQtrtFC%
zo@;N%;W~ILl32zJ#d9**^@f~;q%<;3@wb)Sw3lWkhuiyM^r5!Vv^0_+q<h!b<K?n)
zk!><HHb-<{O$r4W%W=-2$PZ51Nub(5!U$=op3QqDbthle)b-e-h0EP_fRFY$4Lv02
z&xvqSf}f~P%isA=!dQPDHyViWy?j+_TknTc?P{k!;{{|tC<sucHUfU4?zCYNC`^IH
zpKNDm?~c8skKIGV@K{iujaz0X0*8Jeq=qt6>gCTBeQZe&KRaZon|<E?IU7^(splD3
zV8;VsD6MKji|LvNN2SLR`R|O8Cnn|_bL`K~?Q?+okOkx<yYw~)t~Fw|qNslsHQjlu
zZvTX1RNssUM<p^MQ&QMjz`P$wROb?YwKf2=QAv=meVEH%PQBH_Q$U7m2aAFh6>HVh
znEgc>7~}y!LXMy%EHp?Y#(l6aLt7?sI5n?a;y;vFe1ro(c_Rq+ZV+HrdQ)IZbhSxz
zobkOk6IIidNV_T*EXDlku`UlENszk6*7yAmO2YEn%x?=X4y@*$6JOMzvs{vqf(I2W
zDs6rb@y&wb;TVv`cGA;vf%R*WVp$A>{e(9h;o>CL&eo<9rIA1Y@cphvyh4>73+#|F
z$H$~4KC{F7vKa)}V3+jIv{c&LYP)5*sn)@jO}pmG*)72Q8Z~zFhd>ikp9IRN7KIZn
z{jGR3Y^Ojodv@XS3)5~hAp82V4I;O!jSG2FiYtnArb47bB;T*f&W7IS+Qvf?&NV&4
zV+09Gt8itD3oEC_w<jeU<p+FSB7LDibUgljy<@e>D$>A_Xf!Q*sX+uvnbxZns5U*9
z3k|_hiqUX%(5^>i%QW;6RGyWoRvCMD^2wNTtjNZ)x;<Fo?<M`rgv=P2I71}iM9ghv
zQg+3<N4CoGy2{CrJs2`LPe69fpRy4^=Pl;hHYi3?flicb$!}1)|NigeugGj`?LbTX
zuTLH&suyeOP_)x)D?JR0SKO81M;h#mXrcEngbW#(3Wqwd!vSg5EFY#<^nfpMS@{gZ
zkuRxfTGGtX&xLi4uE9wHneirN_B;zOEFBp#+OxKjf&UL<@7SCP*9L3Hwr$&**tVT3
zwmGqFo0Cjz+qP}n&O2Xr)mydis{Q<lRo&<6?n9mnZ>g^2<~Y&kQWGzMD-mI)WD4Ip
zsLe<`141YW6-ik<oq-|K7%htQ8CqVH?T9ybnS}&;)Jc==RicsUmMmgWbn5aiAvJ(l
zmnl$zEn{*2JEp(*U?8?iu(oWWe(ub88r23)5wC$cel}Dtew5#oWYcdJPXVu?q;4fQ
zZJAEZzVh#bnxkqhC~^KD@UR2sYKySmG_vS0oL_Mq2eV3WX##CC<ljT>jt~RFaG)cN
zxQN^dz#lAIueRE>3w0e^DY3F3>Q=XGh6CgdtsOA3arVI9C(}Oox}aocN#YnzpJh%H
z#wi!RoIMS|c7G0!O#DAwAiL6XV|%*-4(?|Dq;Tjx@EJxCX%{#cO{_>LErg00&n|mz
zHJ-HPIQqVBYDLQHeUQWsk!7=;<mJI%t&D~0fWGhYAEX7`IVRvpeZ+4IQ&5Z?%1+69
z{uExqs->wx;y1hX%VvIA`0m`Tb20r|^W&N!To2f$*sWwfxsn_N*qwiQ{1ajL#B&Q(
z@S}@3z7Fx(O%Fer%p+5bvFUV@P?IXt;nZK{3tOPgb!tsRQ=N5$HFvm|$rYbG2CkMy
zJnjUqTHreEGJ~GVuTvY|9Zf1p0}h>0i=+tdF%u^nqd5X_``mkNZR4yPUZq^DkGAai
za`bTRIF{VfA=^Tz*(Dpz4fuR;`0a0{^8hp{8E6(vKH_i#tz}x%P;?@lzg09E&0ieG
zN7e9YIE)r9SDoB!+fFNs7t1i^BLFPO!%;gNB+%fqEjMus`E5j~%@ykQ)sM{?Hu&y+
z%C8G(2-xN_^2*qkggUZ_<Z9+W`}h_gzac@0p+Q&ihrPFO<7V{wSKW>6Z2Iz|zY_cR
zpr3a&6T12koduu`qheGH!1C+Zpn+tv-fQQvR%?RzM&5tGVoC50>H-yt8D)UruN_ur
zUm{r_|NPVmJqA@XKcZmjBvl_6A#gCx4<C}#$z^Iye-qu}`|~8T30=u)7@Vwxf9UJ)
zONFY`MNKgsgFG!p8=rxKwn_pdnw*z*!SBroy81jUc#N*$@R*N#Bew!V3RBf^k9Y=e
z(I2F7CU+{9CK<t%goQ0j7hqFL^gt8)d#6W@h_ZE;2sv^Pa=!yLL3C5^g_}5~XIc<`
zknFpx<!kg$7ov7@NTcCq_fW(L&hyBF4b<(Dtn!$gKOuvV&i;#&%lz+n_<z|j7ADUB
zdFN{CpA9?smj#>oXTwUuv3(;FXYB$uIo!7fnpTfLkM+>%D+e#v6BQDT4<o;C@SP;Z
z7T2dKzIwYfNrv8b9gaC!*<uJtaK0Swwm$dYe^VD`A4XPK_!2^+lPe?BBw2JvVlW#&
z-amYfT-;z!Nlc_Gik5IQ)m3c&yd58WK3_WxqvDhvI;B$K{J1!0jC=TgtZ+~NoDIy}
z+}NG8aJ_qEJiH!XA2TvJzceH`3gxJyreGDQ_Cn$?R}f4<C3&*#1nwGST`6m9QD!p5
z(^wG9te`{1d;V+t;&;3H%st~r(qDhbU}(_CdiA`bmULcu;Uy1M@3f@RMODydYpI*o
z-E6J{8icI$b~vw^PhCOjtOdfNAf|8p8nZyR^4tYc6Z;mf4L2%jW@gSbIi93JSK1v2
zGrcMqI(aLsQw1ty6wwMkwK6A_6D1?&`LX1sdE{w;eXOnF(pe!`Kh)kHPR!Q&%rOMi
zX|u7Q2cVg%>rT6!?fUPjf{Il73360|WaOP_MD%hoa0aPiKHj4T7mQyzb~A_!C*tvJ
z;=HVDmy7hT`n1hekyt?Eri2OJJ9QJN62~Lf#cpKFT9&i|xXB~K8o!V|Sra519EQ1&
zWdVX!us1cYuhmu;W<6^bc-1Z|EnaP$@`0=qg$)uelREvZK@mClY!iULob7%*&9K$V
ziCQ9>K5kJ(=7Zz~lQq=Y0=X>FK!iXnE$`{nC?qUD_<9+MytoKKRKq0D=83SF-<&Ue
zp)4;Qra=+|*)OLjGoKY*I+@1=0vJtx=kySSU~W&RZ}hZ!0=qUEYQ9$@1z!~!VQnQ0
zIDc_Mir7kX^1nGcnVR8hH9yDnAL$_e!QUBxE9B;$V0}p9A;2|3<S8hNG28{at0$&>
z)vBr$;Ss&e547XSW_AiABF@;LsS!gKOw_Q0E9RIC<A(=oaS^c)3;&$bghY@YgG{HR
z_;g#jHsA!X@giF5RqJK%MG$V#h5^ZU=@5v(g(tg&2L{FI`ZXVFerGc~m{VhSngpM~
zSqfB^x38F8@Ta<q>DbWTmS2bKLd9k)6b30Lws-Aliivv9e(^wo0Argq{FM&cc%PGl
z_Y4nu>Oh$VcSWjuxemY)Df3#K>K;(oBH+-$VQcRv|I+SI05w3a$t?esm@k@&oY3K-
z$%zC8vTAJ|^-=14o#sr!ip;WF^#q8uGz~?37Iq5&1d`DmDlb+vAeZ?k9iom!y~!jx
zi5F+a>+SghPfp21s>8?`N*MZX?V~sYS#DN7oBFx(p8)eydQ#bIJq^KHDv<BqX-S$z
z@=O<W^5mUOdHzupF#D`*ByxOhH^+OTAR9n?RRpfa?D~11=mt6<@>das^03R&ev$jN
zs07j8=O(D5&OeI}>awr5@C;U=kXXtqRe09m6E5%5{;?OM+VT)(zyOg1T^j-{3c*EE
z<n-I<c#8PY0V+cXlb}h>#qX}Ah2{*#S{e&&-k<X}35pEH>{?f*w*Uc~+>T2Hoc69!
z{KT+uA}45gmSE`?^~XOurBvkkr<C<=;vGm>S)NP9C}2@Nd^xDs!f1*4`}y|>FR-I1
zmW+Rm)^Xot&RpWeoxCO0;XBTYBoucCH_-t_Oo^nK&o)Z;fP}wP8E#3!$5^&q-{e*;
zF~^$eCpPxibRER(Th9$<Y;bWB=-Y}PifAdW0LQp}JVjezASlreU}7`PTO^e?YHe3M
zLsA5zW9aWClcbq$0^B3qDN;SwV_>{U$FN$YTpT0E2|QD%{{Ru<$M7BDNA-&-$@HbQ
zt(TO*8lkSIl<_&3=PIBS##mD<-`%;55W^Wd63$d08cB~+fh#aqGwmZrM70BpwaIe{
z@q)l$WvGahaWgpdw#7z<Di21&h&jzzgDt3kWH%s(D(kBHFL8q2+ae+X>mdxoZa#`Z
z*!s2>yeW2P>d+eq+l-AfdM&rlpCFTA?%5;aEnu8V^vEG$6(a$l8K18bI|RFY;vtI;
z`=b1tCBA@@S;0A14tCHop<#ITpXPN_in97aA)s6uEhixV$F{hvp4E|*X@ox&F4uo0
zLeB2IFmQki=4-RRRxhLd;Q6dd{aioG5kA3qZ$#CyBACv#a*=3)^3#m?h(<g|W24|h
z$vq6~galp;th~L*9fIled$0r2^z!)zMUSBC=D;Du+W3`c>Z6mAE0jQ;h7l9t)-PRz
zox6Ee{3Q$v{0Bp0r9X#aB0B&V3>FAOsPb!G<N9y1E9X~s2kQ1w&l7M$0|vOo-{m8q
zV4Wn4Pu@zZH!U)qlG4%C_@YOo6~)Pt!(Hv)<HN4dMYNGA7bU+}Tss%xXu|{Opx<67
z+sPF`t$`Uta4Hhgy1l*L*8LGzAd+BH?KSj!wVbHj^ou$rv6WnejTr<o{C8(X!iL1t
z5a2eWkKbCiG-fRp5sr1ksJ=YM$wEJyLed3Jr>mHc&mgj`-6)<ziMNOPCP-ZZ<SxLx
z3!(Wse*4#DHz@vc@&V}xyQ)ntvk^9B@aN-!{G|jcFIcD_;tmcy7-#vjo52l+pacfw
zu#jxvi~9v!{6Z$@mvsev(AGu&dBX`+Q)3zT0$VD*VZZi};IvbZF^c2ktNrI^FY8x!
z%_a1A4%|zA*0ExfIlcM$g7+)!@fIykbh>=gchO(O+Byq2`x{y_=|nQAolUuzF}Gv;
zY$aB;AN;-L4@^fJr7hCMs1?ma)sCm@r>l*XPQ`&AwHG!Yknm63xT;d^|1#|y|9>>Z
z|H+)Qu>8-J#u!Z*=R*#J?wOj?qIN57((|gKm1)^*X|wX>F14)8%Yf8VyXvP%s<pGv
zcW@|0xt3plyTZv{L3-nQCB(7@DAv8-ZsWef6!OGWZOd!N54;=v*2&%|WFDBL8>YI~
zyb&Xr;9ENA;sQIeY!e`Ue{R~iJ7R#h>|9^>_!hiRt^6wA+}iPed%rUB^P9kLY593S
znYyZ3d_UUQLKG0N`Dyq$TR366dy#8k8h28hu%iv`6(S^AepH^y!R?8Ps;zX&d=h)I
zO04=;d4XKFCn#I=DKc<r7)@P2x1=54m0D_OE0jf3HwFF>$E}m9XTk&|iCJn=ESc@}
z@pD$`QQW1nlf|P}SJWayZq1gdvu_TN#@>fC$ppTYl>tmfnzUmdob<_{#Vu6(Zm9@e
zD^3>F!?|81_2MQjm*+a$w@~QU!<iy;px-E}_L)Qu6u-PQwHIk-f+WOB62)66D>3`}
zJeBT<E<~v?2a3qiuN-F0mf7V<OAuADio$Zm@zKL0{Y&|!4^mx0r{y;jj&s*d(S}JV
znslBDwcWE1Sh89qY~%u*3gGi8JO?Ft*xPf%#)2_>%Lm27Ysc$X3HXR;pp(e%c5mW#
zymum~K+-3)=FgvID;<*f?6)jDPa)k9+!mi$#@<Ege@n0ULxxcd>g<+SqM4NpY<c6Q
z4j`*bRrFFV^-Vvo=C4_>j!R8PW;U?R9FumOjtZZKV&_R7Db(MDVUs-*<-2<nD$kS~
zuXLTidBDp%V)tjTW*g27Fj;~-Csq<s8V<RGG@F~6UC`L@!lSB7FXGOYN=ge;nFq&I
zErafYrifZ6vcOI{cA0YR0MP+e-{h0Ba4iO_1yk4OE}7h&+x$9zA01g}9=JHC<x^V9
zcXkMj!T~EbQeCjA>VgosnIC{@U#QB~?rbK3IvWIb&@WrN1beNIgtmA03(2Atf-(DJ
zgyMMlTXWcvsdCU#vzCNuz3d`<aDa(zlpthlia3)`ehUMG$TwjjB7vdzLI?Q~BDBj;
zwZ0zx%S^Rtn|o9A?M*ci1nh6V${T&;ekMR0Y<i7c37eP7Z>|x}#zcg7z!zcI8^Kz(
z5tKX*v_n#mD8OR<0cDp^S6k`jLg`6Wr)C7|q{=JTH`&ho)YlIj$jb=1gLc=m@(>p(
zjhq|6zVNK|Te^iBY8%)1_tB(vo&O|F6l_C&g$Cr9nGk2z&@bKdoDeZEaFbWn;^Gf0
z7nF#6>`VGkoM4E`K{#SGt(3a@OxUng45`~SeVtQ9zfn6IbbZwX@SMx`J*IygxZ++Q
znOGTv8m;R{>m)XlaNctxHVpkE>^iYqsJlHnGVG#NPYfEomzZ<wRHS4h{t~jY<7Jdk
z*$t9rCRwrDs8Swzmlr6jcL(}^OLS0CiKvyHp!>Y0f`TFavblOhprbKz=bq>;($6jv
zmk!p<i<QTU{J=|!jln?;dl@wx3#k?uF?w6NwOIixsu`uusv(xhzjO7AGDnPc-~Gp#
z*7=qs7`a#1arkz$5z@?KpqK4WPItF_kZ--r7p@W+`hw0d>-ef86<do>c|l0G<nr<C
z8~%oU;(^%fCL5ihtZZ|`sZH@^E!a%aG(#HWi%ug%Gm{mlW&k5+c)~~WjGB1_3{hZK
zPRSH0QTF;0Y(91#m*I}?fW}j4<~YNRIzFDk^!ihWWK5Xn@a|cKOz~iG`0%NU_>@6E
z-!+v;f+~6>&~HH{7Z7Vo<9t+p4BzyoB#>LrI-cI?k3eUI=FHFT$_uA1Bz|lS&j(a{
z<NEH?!5fd1j2P)^kM)rOn}x(hQ8{DcTEX9U8(mVdNMLu2s{ay1m$3zjt#_{`!e;b3
z6tb@;Zv5i?efEmBT2M57nDXy@1H7yMH)b8cMCZInNG?IX@sG=Hbw<V1tRHr^mWYc5
zt_rfRE>Ri2!ps&eCqkX$<a?Ct?B?bh`<#312#%I`^CI1V*Pu1MGNJI@OWa56K$c{^
z@p${=kR~JT{bh7i2Mqke<>Yxf0)CygMEx{?7}~w|hpmhNDU7jt$dS59cO|CeUUpn1
zA6u)~*PwCe%jS$lG6LqIGNQ#)c$3gDmPiX!Jw(Hhc%1E!3$S3ITjKj3g!*;^#J#S6
zK?A#qG)j#ztVoV)(CDR>WQ08DUY2oGxA2T^YRR~zc!>pR2#v?!fVTr)22gA*G)E4T
zYU}>Q2~HdZPTZn;DmYo5bq%i6-e{qNAx|51s0;q=uhi9S&GA@`w6&RRsLX5Iz6AlD
z0X`|Dm$IrsgaHbnb%iX{^sqWn!a~Qp)?OpT@HE^PH?k?aOR!WxGMR=~bS3ViApD)n
zYu`gpEJb7bAT8ZDao=G?9&AFG#6LniW7$^}&yeGVQ!1uug0}>vR5vq~UxHuM+*U$P
zTr4X|fBPnB%c-V3XUKVn1}x?1VYb-spD~d;?sr4sS!lH-?SOEsIfR#|;cPNfBC&IT
zl%w)_74$AM?R>ElnGBRp;-u?VdV(|FJ4IaZn7e?Rk%o~?O>Y?OM2kg2KRdPgtK}tH
z6ly-=g;2+{9ZS-sP3<PtFU_`>!eb&zv=!=f==1dxBywv4oL2oRy$CRwjS2o41tB_K
z1X8DvvL7ywHC($*^hO^Gn^+f8M+EEn{5Aq~;Wz|?98yAsr;rr-^E0{7443*eGWUia
zYa1Cn_Qsqih@<txTL1yiDB}O{EzWGO?-|&G#|!dQo8GheKmx92nl6U7+qDDZg3N2y
zCcf6Az-$0%`$`zO@!KaG^f%i?aXLk7+Urvuoxb;ELv17%?{-bJ6mB{8A-}9Qk;3Cu
z#G$>L#HI?*o|b+)<6WJJ%YbLcUdk(VV`2k}pHk!Q&4Bdj^cN=oR3cLaSA`C)#&jqD
zA)(nA!7^qKHBaD$(91I-p^Cq0H!q^Cm9Is43T7`n(6{3q=v0cprNAd*axl#Ke+r=g
zO44(2{m-SlE^V2lf2F&B0%(^$yBrxqA<%^e()>!UbJ~u(*{&H8D}r<`RaC3k?DFf$
z5=bsSpK5ISk1`Ju*qsF%j)C76lAJoU+_&Sy%iEMfk7NQ`@u)}YBq2%yxwKiZbjoDf
z*hez6?}y{}g+dRLHe4oiLV_$NQ@HihgFs$SMvwRJUk8zYGB>AH#TWeedX`vndOm%w
zFAO@pULH2m9@4(ArfmBbPVDo3AlU79*0}Acq}fp-5=p|mGHXSt5}8w}Qc)?QOu9qb
z@&;UV51EKi>4wrUkN;}F7p}DlCv<=4UP~Li1kjAb==CPOWm>He*lFc1PaJ^YV(-{$
zXiT-NHM312Yol4V(YsgK*Hxt*4YzoeA$Hp-F;;Aq=iy+aG6{ROfXp(gk%g5eZC}1*
z+0XKnS}yH7pX6P&KZZCN&`;Q^*`KR5G2t4lCEH$52By9D2@D)6`fB6iIcPDizt65>
zrgA`(vc5ba<6g??tch`J{T4DM4&UE|ZLh|`DyGTmdH;Ht7b;-Zp>sX=-|xTl-XrM2
z&vH7G(N|H_R5xp?$ZFPped*fq7i;5fS6+*!UML!aYtF;m%-V)sz}c~}VGF_2l<Pyu
zU{>C>I@b5fWw3SE5N`8mVcD(;3j(;qQ{~^;=i4!2sSL0NEY$Ima@)K*bQ{!!j|wia
zF&6SeE~G+c&*9iqGR8N61yq9X=p~9P!jMXU_O$<D%tRh@T`aYlxDv!@#LC`QE2_&&
zAtjX^6&(*WQT-4yP=HM|UzMOKM2q{N6Qrdz4|mhvD-PJwQnyw6m6c?<kcn|w)wcj2
zv%W4cWv4MPazD-KUB3QAp3$Z(dg}3Q?*l29Y*rlv!ZmN0#uV5xPE%QFKL{N@OMJYw
z$kYseqwRf9&xHzIBzoxo+m1qw?28u<L%BDjy-+e2^ssDBx^wF>iXN@NVZM+9s1(j&
zosX_{9bq5<e5$Wqb@QyAk3~KhJPld|t4a-Zoz}+HG$PO%q?C%_rTfvrLasN@$}%pc
zYA^*(FY1q>lQ3vA)DAxK8S*Y>-^F%dDd5{i9}E~mUiHDFQ^6cTpAz&STC!wc9+>e8
zxs$78ysu_3r^1@GA9=;$wn)BqBe3ofUAQ$P)EZVmAG?bdr@!Nl@Z}gtFBUgwX|Hly
zu2p=D9Ju9j$l(_S6|A7e#Bc}14-48@cb*s0r3tk^{UV}mTZhMf2?e-X9Z?_8+bmB(
zx_JQbm#y$zP?Muv#Iz|d1EICQHP@dti|1b*xZ}ntI)g=PcwghaZ|W5}Yg|Vg3hJf_
zs1ik?tZ99%phXN>PA!5vcy5=znr}?^4@L)2iR`byyp&;TDC71Colo&*@RdLz&_s6Y
z6*L{<&nEUrh4KEqC|{`!oGI!E?+uHLCZH#8b)BYv50n(Q8X;AxmgzP3lQ`2#Zce)V
zr6+0(!DV(@Nh1|)+j%6=e&%1-1wq=swc6?6mtW=TTq!dl*PP|4<~#pb#r=6uOwCHk
z!QaquknUFGPYtv0K1HTL!?iKQxB$j^ZLcw#v#ZsC#|>|UrkbI)2KXZSUi`(uQ#5qA
z4$jHEg}cP}?vL){P~7Z!dr!VmquS|$=B?4O?s%4=A6eLnfx84P`J8R6d*iFBsNMeq
zT4=TxW;0=)Ku=CYG5hZEmyk<?GzvGz2V(ozYlgrdyoN@7_xY}u0jhp-+}30B=S_%5
zmoSW#x~%8R1T4fPRv;u7c)BY<JJ;Q8tsBo(0C%tr-Oc+?8iNLC(^NYII#GZB<0@e@
z3g9pA)pAtyl}=}!c~{0f*+BrfoMom6HHC5)_+%?XI9MD<18>k+14pkB_KA(505f$h
z48|<~<<%kxm{A+fUv3;4P<8y!O!g(W9_pUhPIE=_e)iQFHVy?TAha^7RdQrOi7zXF
zbPTWsnMjv_U9|i9wu(o!fa<ScY`oFcJf)Acvmy<_SC)fVGlYn{aGi$2*3qgx!{feJ
zfxH4SoH?BVCE(SDfcV7G$Xri#6Nb64P9hvhgT<Stl(xvpx6`tjS*q0tC8kI3DS|@f
z$~%g_cBs;BNqJ8QXAsEf<4+ec`D&rDUdAH%ZSr+X<C;ZSHh++1k-ie3f_$>1JGPZ)
z8NthbV&y0vxwgK5-w@UzQ-JQP8;-Fc{V7wgb~*(IXUsEvGpx!;laYqbtjRY)&|4Ou
zfo$XL(BDr-l`4dbT;}FIY*UUGy5;zPpJ#sK*?{v7B?gt;_{0_A{M>S&$z9A#kY@I?
zN~a6+`P<tfE`8ON(x4J_l@c&>$&L7sP||_s!PUh?F)?tj;dRkP_5P57O&bzIOz|rU
z)tMp}z`EL!BzGKn;w(D!pOug)2NQ5K`A?B7CI(w`-=h}=JTfi2dxoogBrt?U*a}<l
zO=~;$N~V}Atw+KKt~Mx=>0;rz{`j3ROq-=gW=p<eL31KBgI?ENB-BhBWhYdZTT{2_
zqVOc{bqxHrtMc6ZSTE;hVoHhQ@l>DS9e1JS)U+g79X!0@{v^z?Ch8|!0L05=lZAk!
z`*#mEqb1O2c8vV?3Vh+CYy-Qiz4Bl5%d~VZT=KWr`I6*?0n&T@_xFDhRo4fH$D=xF
zBZv^(`^5D-@}ByZhg<yfoEb$!#ozO8>vhBr>&F-Bo58U;Abe+H?NZj9Yoj|)Dna;y
zfh<tPV$DchRNWLa71@sZZx&6dk={)y1b__F$^~wpIbW=Ym5_1?G5{JgA}EEnw}xbs
z(7<uBHGf*!?A8Rk{){thd}p5s*geD*9Im&o53scds+_t@G+mDY1P#pME(J3g0!7Sh
z3;~`6+h)qJ2Vsv<<l1XNEZWX)P;upWAz)uV!*tbNUFW>kxU@{kXtNO(n-Ux7AK$0O
z@P<0}cr=azrb#LqrIhD@>S#s){!A_5^Sq)8P61)Ud~VF@Of!Hsgyq8xr0DJm;K+La
z3ynvB+%Pa7J0#X(_;TF6u1fDTA3#z>HCGb!M^R6i<n0H?LMpAMnC_S=T6k6>L3sLV
zh#?ew!)l`>N~Dk;ixHP%-NIh#d15v(P-w?Uw2q;aRl~gb7pf>U0jwuR^P_uk=BHKu
zUq<lE1Q1sttjDO}4znI?Bi%L41$mccuYYt}OiS(BGT7}|02A<x+LI<=$x$O9!uD%|
zzZb#g;vhvQF0c*vU{AEBW^|YiTeZ$X@6)=*pZGjA{e7qwD%U3#vzJ=>0gznWerloe
zr6kVqee`)eF|}KK3?~^Oi?NaQLTFhZY&9_e9z3E@J;!0Byqhvc;pzTxb6XR!{d*KH
zTO`54ArG6-q;1Xx26uPl$3&4S1k3>PeVl3T2jJ1y|8{5_T7{@V^gOg<A*i&<#!pKW
zxPk4i?6GG_=IscLZ`Wj|PQJ|DJ#1?Nm9PQ45x~;X=N8rX1D}t--J~j)KB@LG0EHei
zrKi(-UJd11vk&J@DRJ=3I|upq8L-7mGG37WEKeY8AB1^hq|Fu#MA{jI%paU^I`3=k
zusc22AwcXvw%Wz#DTsbH-6&CuohL__Ia~je)C+Zn?vCl>>$r_@g%P~S*Sh!T15tD|
z8KCX~n!a6vjq!xAPH<t&ddcJT5@p;`Q=zvp=dUz;ZQ|Pz@|BXqnT@Wkvaod&9lfS+
zqqUPCpn3)@MS#r;hR`N5(HV7a5c3c^b=R$hSMGxgY|821Rc)d&C&&nM>*CP9TY{El
zxU?lG2+SfA{vlqLO35VY$J<#=5gbqOcn0kmtQfN1gQZ2>I7&HcEC2<DuB;G0;tOX{
zuoli6yUY|G&0_{c6fVR}<Dybca|j;2=i?33e240PE#PL&;Qn9Iobx|7dYS&`TE!90
zgwqDseQwQcZfbQClLsj8WM7hr@|ay>Rm@8*J1e`GQuL503Vx!rjo%l?ED`y9ya}b-
zGczVGSXPX04-rpW`s`Q~r9h6r+n%}jf|RuK)YIYTm*{CkVy5C>ij#*i#t7oXY}u4Q
zV+{?cq)E;%CzmGU=yJK;KF{}4u)29&o}p3Zj5GS3o*&nrM+>+6XA6Gc2bX6(w`Wfq
z<}*Lf*M}_+8UmkmL)UG}$mMMDM{%aIT=|r2iaZ%^Kq_|Oa_oUkYFv$#6W(=n+TNX&
z6XkaDcUbaI)7-MsfS7byY|21rQD<zTy$mH1I%9%ll<{AIz9oSWWiI}4{fVxMIy@l4
z-BF5EBu(o8RW|VgvdvKhswfg=(SZ484^7d0m1!)a-ri(OKr-Cx@<W-bsB|)Fb(*yQ
zS8#mv-w5LIT0uQ&dLqh22-vq-!tRj$;1a%oQfA)h;$=4^!lL_?L2d3e?z7+Lf{t^*
zQ-DWWz?vcgVbO_He>ye`vKpp{P7&&{q-v@{F~mm@>x{<jQd+&9zwHECcj;$&dIw|{
zT{UCWJ4;1_+8l-}s+w)@cq0)FK41y~wI2>t1XpQaG?kb-FRHW3Y9_waEYA_#Mb|=*
zGW66eOlo(xS~{YOtg^nDq2~6re7LK!y5XYs$v16&I00r3j*gex0PXq^I$fToC6F^&
zu9)q+M_?ih1yjxAo&W=KmDNz|FTQWrO2{O>ZaASU5vx;v0Gd0uj8L)L;F{G!<LGK-
z@6Vc(me1WEa6~L^&wK8Q8b`J278lXRi!Mck87^^5bHj^FL;;?mgG@oej#2F%@=gd4
zYPqbchyBNR;b2&Q9(<CXJXS<D62YGPZ6RC6n?Rk!@`S!*c$U`5b5ygWsQe6oo-G4L
zPTx90P_AUH7O>l{z%z|8%v1P;o6>G!gK-OH@3whfN0ru`@TUW%ywK)_Ww*mT{Oa;W
z`Be7DuJcnHl)=*O?GnvpXqOE_O;#mCLkVTt)$BNULHs&z)l$0!l<=h}KTu&JR=6+0
zX(`DJM@-QG%Lt63MdEEZ2n|98nW#R%lxOhmTypXNWy|9wt1}a`Iye13nWE$YXCjGl
z$&6b<+brOZ>+-)5WW2K~r0Q0ys3&Nd<JWMUfEQXp9uWGKHxwey2avpmaDUNp;vg+D
z8zl^O7!rfzHUX_D{S|GzC@onBCVhTUFETP#sfamJVcpXgNH29fc&)8KcsjkyS>CqF
zazVU%#Hd0`8K@nxpg4Y2(Bd?9J5x`D{mGgERI@a_D3nx`Kkc&0W)?~ZiK)U`#?T7f
z??)HA+DAM(shmgeAlAcHsF4z6v-IbvSvLJ52+8`YNH9h6T?Yxm+Ie;25Olu`T%ri-
zl63~Z)$%3~*Ii-cXCx%H@cu-rNKz@j!gb%xLCEK%$sq_!p>nNhwXs<o-O`vq|D7G(
znillcj2I>2_lTj!Y`fZDs`ki-+9er!5Kuj8ADZ>mV()bwZJIyA>Ct;@m?bg9O-ckR
zSV~V3E#=H$TRV5cUZhS$n4{^S(a1u#W0;5=s@GLr!ZUzCrK@njF>4J75t@|FVZv{_
zAv$@z4|u=t+Yp-p!LJu2?mQ;8=oeWQMVF>8SZ61tX4YHtX)8TE3EX1P344X85}8y*
zcLB3X?zeL%^uc_l4g^`Jw2Uo0MT~GZh6?{Z+bh^9lSEh~TrWyE@Uk?I-Gm2m?>w^^
zhUGlYnYUxWkL=5qkV=>~(e*wxLK(Idm8<YOvwX3Z^ox;A-)S{aKARGST=uqQ_R~)b
z?L!7yNYC+wUHF@3qd~$}!KdTu>|p#@BDf;rx{X>SJcvloclbJhqQUyJr#;2Sm;iwt
zvy|_qPSy+xy^G^R6~2uk(C++Q9l@1j5DicO>>;_eB|21H9K#0jTlZUM(U0x%G)(bl
zZ?<!xNLYyRYegKKxd+Z+*Az~<lx?^Ij2ENZ<$#=482`CqmnPKZoX6)@30BQaGU6yz
zx$N}oafW>qV0|ed{?1Ot=7BgHLdS|!7K(Or95L+nIer=pJ<sNP7=$eyny=ZC<B+x3
zeIGkaeM)|vk!H7#3W{`t)-J=B7>2c8?>2r~5Q{t41=3r#6w;1R3_;&~iS}+)a~_jL
z2UA$aQ|VQ9Tu`vO^cmKA{ML-TQ|kF^{!_6aQqlq4upgu;?(fXSYiEEIs?w<kP^0VM
zX=}g@&G`U9mLf-2njBv4dpq=~h@6i-pdFl6sVmqo!kPY+;|(tcu1Ac0EP&YIi(cTD
zo1NX6)5^^c8gwW1oEI8_Lo+ivo+bpG3g==8@pv|4dv=ETMLeU(FXozMBemust8|&I
zX=;WqE-v+IO(w;Vh?&ni*ke+y31_3nx#E%^P$_G?_D~6-GnfF(OKI1C*a@yTJc~Nx
zd(&)hr~_4=`+!-U6>QL(#xu@iT@3~2G{rYRndz_Ak;bFc75lM&ZoO^MkOBQ?p-uHC
z?_~a0iLz^p>N^!?nS0l7?@=<&Yz5Ym@B|}@kNJ|8M%?#)fEow<^r>>jWCI86h8+e%
zo19{bs<i;ywl8^-)iN6Ub82nZq}fn?pDSl@*byC&eI`OJUY>sS3YXSa?Cb4#oYY<w
zv+i44YE!2D{WC|9q6w?J43OE19P>=L%sAcK2lN3C@g64hTAEYr2N!PGP)8iNnfvio
zP6-A4Ul_MqpJ7+Wl%uM&z<)9;vkln%<$K{zo_0MG?UTRhzepT@@9QwsDCac*%t@GY
z$)#AZ+#$L3T(}HOqA!nwd|C|ei5EW2z7RKlVE?U=JHR+f#{1D__qPJ1w~DSFhUdl{
z4OOhM%<H+C`GV5zrz;XTulWQl=SxVSd7aaPy4BaID{q!*GRjatI(*{#!e+GDTS7t8
zPfqyCBVAU?L_NIe^2Zd<O|VSTwO^d2RtRhm1PQLbU^{MTPUM8p;2%6BsWn2MWS4xJ
zgLJSP#S~_SYt3ar`9f+6p{LoR;5kD*Y1UK97WG<OD{QX(;~>qBDnD^x0_DWLB<gJ2
zHBvM$ME#Q?a1dN5@H!$2BoVga(pEtTEYc@K#)G+=03wyB?e81iXkOW=S0eI*QOicO
zF@r*RW4+#1mA%nQJj~eNIpWUT4w*GIcya6DvA^U*3(gK`T<je!Z<KZKi-t;G?Se8#
z-fZVUms3?3iqI(pGh7b$_m0JjaS(znMpX*mqq0~GF0+b)dGcwC%6{~OBtNqL^gOw{
z4&{l|F>NGIxV3+!sK5|YE@^v22t8wL?7#=+d~e&R_q;IMdi?4q8htrlUqUsx4O+hb
zd3d1wf$&x0?D|i2_#cN`?*9XVB>I0Bm5wxZR2Br$d~a%2M_h_0wv*6Z{|NUfyVf37
zxn)a&f_bqH!t%Ut@QvA7Cn;M3LXZtIr^c?^yiYb7%w=Y=Nlx>hH*0KfU}M>qBdHJb
zNUGz}U{Hw@1O)lQl$CBi!ZtpVPeUw|X)AT%rR%ji7&9e&D|GCcr0_6TaFvlyOdGCY
zM(n$uW4vp8IyY?J=pWovS0*jpvT6k9`Q&XE!=J2v8!Uk9#AMV#IM<b?FneZ$85dFG
z$c61u;ZX{q$PEu~=*ogw(+jc+p*cpNTciaWBT~GPDjwy9{iWbo8OyfREzM=+jtWXZ
z&X(VI>CG|W@ML_&gP&x=+UUD8<!a88ND|^=8#o=OYmyca3!0twBSfIgo4Tb>D2|wV
zEtBVVye-5Jhv{r#rXph-O0+!aVOQ|YphQ@zhDbrfHK}vUNG^Mgo|!FI*b0iQ(Eb+(
zv%RJn2?p7OlQ}O;Qldw9CO2vi12d1;2-!!<S>9H4olguk92Y`JN1|)fN=|a2-@-;8
z1+}BHdt=az&`Xb>vqB;h3W!cV4*{12hCCVkG283f;+WM8K+=@OKHe2+=EtGrwkGnb
zZypEG!Ep(lR0B8kw~A6JROx|Gvj`BdyE!M?9h1>|eN}EN26N@)p!h=}!k+WN37*0T
zuBQO+A)>|)(D*Y2pg%-o3gJPnh#hr3r>gB382B*i)G7J%aq$!sZo&5I;1&*zoHb^z
z;AgF0<R~x21FAjwTk`soOi=6@@HdB~z1$NK>IYx+OYI$?Ll++M;W(rXskglGJ<?}8
z5tD%QSsWBW5diXAp8>aBPQhd~7C3^em=s3dPa9k%!;4Sl**Bc{<{Dbg@1}osr}p#2
z$vu>yPLG8>EMSdFU5GoUG)}#fdd(tH;7siXBI;!o_6@zkzn-bC5?SMyC-6-1I&{^o
zrh2>Gj(A?1We&U^rgH{(va`EBZm%z<|9JTFw3(8gMEZDLh0|m;)6-`((Qpr3sD_o`
z1qw4Sw7B&1*_CSGfNpS#hyr6q15|90uf#C+N}cl;EJ~Hi4+-|VUrAesLg&6yx=mXV
zVI4G_){Dz#mr^CJ?(>K(OYx*^ySjBrtnZx5VXD7`G2}-jxV_BP+TEFT%<@*8#W26V
zvwtk{&C}<yXx`3o?7Jm%W=>yvJeV}aju}&S=0Zi-?7p24;Q#C?cVmac;Rs9`BMr@s
zy)nrugEhEa<y9lQk3GXLL;3Wu?=#R|MB*r&62Sg$>&a};BJ|1FS+l;{mb;uA!rlB?
z^!Yp!NDVQcsQCgz+M$d5uUCTWzcv-QIR57xZIhOa^X4#e_k+eRg-r;a@wQ-zOTMkN
z63Q5d*pZ=D35dH8${s_lwxN`g+S|=pZf>8<uCfyiCIeHtgh0=?SGwqm5=s~E``!L&
z7$)ynVYI1u`sqGRBE|f8t?1b)0KF@Ye!<4EaGtFDbz^Nm*j(!K-NgvK`#nrUpsNJ2
zr!#-&p&`%r?cdO11t;S`;6qx?cJ}wf(v7*m!e~vD>|p6hY0V77^z;q`zwYN(7^XDB
z&}^V3v5+(+_z4arKDqvFu&KHQzMP-?%A9fD^v`8^LsXH11YWEh34zoY$~0+;m04R%
zY8EM8n)tvcTOzdz?>hS2(wx1Y`mnqR!a8e0A0fTvoW<d0hk^roi6JNHoE^^RP=loN
zkgV01hi!KU!}?H!RN^b+jD#E5&A?@!)0j;vL`BWTMckC{L)~lV7wH^yx9|Jwk<T0O
z_N#$o(di)7pQaf3Rznmr$BS=~Jdgg$jJJCMZ8H?=Y;qDV3D%?l=V()D!B4y0!Kf^N
zBMA%YzaOHyVF`Tu`>nvzJ6Elvi+h>5*<i1o`4bL0t(2+xDLQR|`YHD}h;28tdZbhl
zsZOkZ3R!99)Po7xctlZ5^*T0uMsEG=QL>n<=IxXvcIaKxg}d(Z^R3~r9;f4Ll0{Kz
z>tSN)EryA9?G#(Bpz2y>%Dd&iss_htN9{I*7D(Q|Do|$9PZw@cr;v3;cXc!Nu{Dq-
z?L5*016W#;0QrFQ$C5nYJFB+50@iR0O`)XmwSnFt7>IVQnUf(iDXV_V79&#b)b;1c
zuMGvMNu(ew%rnuagdagi(}CcFHhM*Cd@~z!^m<tf?plI8?ot1FBWW%0n6L;~oClZW
zRtA+g>I-{;Nn4n>del56B;a<rL@AuqNF#>7Y`z=E==^I86^2M4SJnB0PSoHcdCy1p
zZeM~b4_0DDS@+KA4hEGgkFGn4(F40#?wIyxe}PY}zPM~==TFN!`&Fu+<@SOhqDK#j
z-oZstJo+MZ)sa5X-gvi{O_w-E9_wqCk8FizlGXlYcqP5lYNE<Z5QX?W=r}iA>jD^u
zA|b9Po<r5U=olGsacrE@#Zd-yd?X~9L8#-Ah!V`V%dTzyHbAfMO2%=h2L=IIyL#&%
z!XMrrtxpNT9#8jk^;A|Pp|0OE*woTq=!u{rKiKF_idmjGk*Bk)X?A-vXt1kTx{we}
zJd%xw*X0DPli<ky_Qe7xCo%_&oJ`cfS*TDcqtXH#@aHF)pl?YVqDwf6$kQGoOT5)t
zk=pXeGsYLm?nXX<1fR~h?D0k`RQXKtqHPaSfD!|me2T$pi$zMHWK#rP3FzNOP)CXt
z1L+8BJo`I&>kw#r2D$^Y2A-D`)>o#1Is=?i-uc*q>m6!O&B)`HP@|%0kk48?13<@E
zod7D^z)GC(QZ6(j3qY+1Tv66X!hlATZjXSddAc%tGjz~DW$n~E8szlh32*PFK-%Q2
zeL7<*%TfhM84<2P?PMo+;ECU#7;|@mbkO;D23PR{j+IEM<o=aAsEr`Jhm9Y9@}~%Z
zLnIF86f+m3s+-wUq9#aG^Xp3zIsj`t`NNpyNw<7Eb}NY6+ED%jC?v!(O?RSo+s#LR
zJ`+hN22e7DYzx~n!JAvw8GRy(v>^@hRo_ZGt*#43XPrfQe55`*^=`4i;5G(UDSmjl
zkt}!CE(8riO;okv;YO0?ynWzJn!DC!`Jd3ga#`NLT>Hc@KGO0YKAhNDq)>Xg=xG=+
zb4OXT_A(}a+b)h#2vVq6q;a)fs=10?7MBH5=W8R_1|oKd;oN*pu-L;Qz3fF~it$R(
zj3!#n+*#&n&^ujn4<E0XLv$4H-FMIMc(bg8J6bh1KTb`Szk=?UIO&4xI}2ogAecik
zNK_+rBc_K6e$d+PYtZ|E{sxA%df%o(8X{I13?$}m<jZw0a1W)AjDj24uU&9Jd@^DR
zPOO?AtKN_x(jEOV{vbwdB5a5O-4!!KU`G(Sa5Q#kNyr_6jH)`46@>E%GOcYvQh3B5
z)l%+*3!b;I7QGo9Hh;Sh#e*yZ_piOE-{FY|4%<#aqa0m2gKTY_pzyeSSHyHa{V9(k
z`{fTSUp(h*y#EO?{Rqi#ZP9ymna3Q5h%HMg8?+*>BXEqdLEle9H4>FuD6}k0a(<V2
zr2<(p8v$4%WN_l(yt6C=l9!9yq_8`$z#v+s?WsxaqRb!0>%}2^CCSj-*V%)Z-xPL-
za_X}`PboUz$7ihbomDtkFPV9)uXGRYoWckKdg?Q(3}Z>gSz9UpHUoj=^gL*}W*=%9
zxCO>XWR1X)`Ow@hb&mKc!_K!Czh_@>tMpR)a|M$jXA&u8fir633VH|Em0^F$wxY>~
z(+E29LA{uRJA&pw6vofeGd$>MFC)!qj(L4kaa$mhcw6wt7D6AViV0LmTx}yNo8(}R
zR3XkjCg0T7{phIM9HRPweB0lH`Gj~>d|`)~{|NQFRRye}mlh&v_0SG=oW)^N78IiY
z5mnZbHepmskdwQ0hkzk!JMWfE${X3;1adM&@2MWUAyuzV)YYdY4%K~6dTt@5x{`y8
zABydmGS#&6fbt5#GPB{x28_Z<mOGe9R)&B};%vw|()0##bb{QY>~?2daMwzDl<Du(
z#M2?op;8mBNg#Z580^r!wV7B8H}G}?!8Ul9G~DWnfe%#4?YZlPoxvS;oRWdweD$ey
ztLMB6`vHTqVe6WTqjGMKaJWcJ``@g<?PdbM{wH`@x-b45q>Iiwl15Ipt`2=^NN;sA
zB;{S0(tsA%0jeB4)A@1yw5KN|ai{mQ80ldZ$(ogai#66Y0-xJEysssqzUXDn3WyT`
zNQ8?1RJFFPnunxnaC}vjTr-B0(am5uLFiB7{)~r=p7!%>po64pA;UIu7KYws+*>)V
z=;m_}2{@*qE_r5Qe={+S_aq@3!#A|$hS;&KUy0HD8M=ENV>h2^6NP0KHVhV(J!@ez
zXgaw$=~8528?rO*+f@aB0}%|rYIk{wrH?0no%XjXrRS3|x0VV%Ht;-aL~DFBb$q&X
zZNq>;TGzBs^o_8@e9#SHvvN-*9L(&VOR^zkYp&IKt8MYmH@QuifK0OI(9x!fEgb%;
zk+cBRby<{l5v=qQrqo0v=d*S6(B{0l!X<)U=DQ)n4Y!lw6%nQ>DmKBNN&t8CbOW0g
zy_quAp8h<6#!VwWLIpPlfL(dmU|DISiCRQrV0BgI+&*^ESNL?<pLSvL=_*V8^7|MK
ziEZ`wD(7|jbW1m-0xxmz1o^=?Q4R!!raGdx{PL+wWZj=^QJ;0hlI)zj?h%-?)*egF
zksLhjferVfQe2z0@3yl-fhybBeoCJ&u-Yd0Y{51}*GKb?(<GTl-?Qp8%g}HE=(S}z
z^8V_1;&8HcJe%hkTw*Fniq_h@Ly2Dk?|PmJov_-G88AHa^C1v%Ijr97jqlL}DnG6^
za@oDsMywZhAMzA)rn+)Xd*=CvP1#6=N`b8T=Ed)^zD4bwYWTB2e&sqwr?P1P+qo%H
z+(A-u`;VCvq$62)cC-ZA1uWcZ!bOq&d_K|b`Vc`42qeL-x#uV?t%qJR!i}_5d5u(h
zUlU@+zP)^{*yK<u$5!C|OU=(1Wh>)~QyyAT-d)JK0njRBjY?UN87}R<|6Qrn_rCj0
z_z+e|4eIRg)ZuF}ng{fbx*6KOyK=_daJPY2Cdb$a&X7qk*p)HP7Uc65pK5!r)*6xd
zjmSCMQ60LNh6Lwsq~kJ?z7)FiPfh-S)a%-^{-^rp{ErqJH~0UnzOS?|k~W%=e|-Hy
z`0vf+neHHX4$ipFxTQ+%P<3)9U+)4amE2v(lra7#&G`9XGNQ~XA^xiVm2n&;DcpMv
z__t_&1@d<4mK^8X`fBHNcgS5t8q-JwV9SUr!*ZlhoPM=n!R;Dj!brQJ@bkOJJK;GG
z{VOPJMFjZm^K;OUN4dhtI1+r4A)02)aDPcS!$3YkvGSAM`rPyNc>a;Ll4Skz^iV>v
z){r)%FE}xw5jrAvUE+1Gdl0_D#e|C*k&F?0<hr0fQCOwk6085_s3)&uQKk`-#Fc?5
z&JDc(1{FsofyeJ5pAg8OM6USpRIRHQYbjsv+Lp1BRZ=43oav*%n=GgH>Q}MUxv2|C
zNvW{(1h>@`*16coTz)D)sUnqVu(bTP_3kKD7O&5em+PX7nJmJy9P(lg86^F~yf6BT
zdFPlnuODm!u+;yHL*-<xa686)sa}_z=?$=llXc~!WkWCtIe}AzWyoW@nfJ8Gx6v`*
zRe1q1Sa}}vku7v0=i0!&E^3#`2y@EYtQ^R|jG!EeoP>isnG<ZLy#T25XdUF55flj?
zvoUXFN<f#FyB?^J!?8_nm_5!-2)tBI)E%@w^F&P~t$YT@()Zp(M1eBSL1=ff$pRUi
zc75I4dUNHY2_8o-7!XF173Y#r1$M_Oj3?o>$tbOcErC{I0tygGyAYy5fhG}2<Med~
zIj}$^9d9RS2@>TR#q+`{x;H6{9%T|GgU#{fE>M?717zP+xC*x0fssp|(BR*BDQX8a
z#tyN&DKxm3iUM#Vx<aVc8;V$<V_v@jzOfL{e;i9iL+l^$5EBcbP;+E&1_zUcm2WJo
zt@^jg=9ba&c`R<7InAbwFt=5&hTkAwv+Xq&YG<5N`5wB)<A$8NsJ6nMFekZE&T=UC
z2?(2aBg&VP$}@$EuH9fn@b;xi7R^^Z0NLW^Rx@0Lwk@3(ch{nlqu_8we=RiZl&2-?
z7bW^n>cK(~%CZOOQk=-5x2Z5<Lp^~{;Pq{F>~UYZ7&wON*^v9eP^G(O5;AUjU}wBK
z(ml0@?2njnDU^vpP{;$lQ{gFn8JPDFN0ubngt8ZspOh#C8wI7yT?R{qjIe`65_y0j
zrr*xuXOXaqkIkj!fYlAVN|}O;1lN*ea?G1lIuWHMih!56$3tW-@u8#y^ILGlD|2oA
z>>YV@mLZg|Tv)5pCcKWcqn_47OWJF*kiV;NUKDgX%u$;eJ~dt`YB%TIo#zB*uA_#U
zZK`?IW-OI3QY|M>i*npC7a*f^(=)iwPV6R8Ec~xGb(7`Y5WR<?2AVYg?w*PF&idLH
z;{$eqCI+eKRCepv83M)@1BqUPv17vqhjtCq)7-6>NXm*36_E0&w=Qf{E<bxzF8H^0
zzORO#0c*{5uS1_2Vbf7(ZbmlbaE^@Idg>o^#!!}@Od4xFaG{qj_ybkhq_=2&s+;Q*
zlZ;D^Hr_!GYxPM()a(9qJHU^A2I#PHbB5HL6ZO|hnb`P#AqCY}1I<p1tcVZUW|KNy
zZ&Sq1phe}+(^ExxU`uEeC7@=FCU&$yHA|E-@Njjk>BG47e%GBwGvj737>7M{PKTa~
zP@U{0h}Hy}*64L<a8lt+sb31dzcv@;r@s_Y2Qg5j8~|gh#F}UiAO_KG(j(cuL^!Js
zi3pxPf$-CPAwq$vaO){s^IC`~Br_@29!nS%w`uSU!3T6@3t5dy7FSDW-s8--!Qg=3
zfjwNrtsya1_Co#Dpw$Vid(#=0xQRsJMmJnLE42yFmFAKF;D9*SaO*9`c2Qlxo(QQ;
zJ2M0q(F?M}&IH*w$!r9O-cC?&yNmz#Qvj9C{A~CFW*7h>`Ou$>I)Z_LAhzKV3@609
z6Lm1%Srgx0afGDgzIrtlKQh`7iI3*Ydt|j58#|ywG?9i8zb_~<&@?GJW;mXppY+&8
zTs$?@l1XM4hPNXyoL)^=;op4%jCIg|M6R|JjfirTU+>?ycv_s<ZRs5^TBVP92LrtQ
z<kTZ(kp^Z@>-)sEoidKmo?GawZy{r-ENqatY&$T0X31A?GP78w)Y(#jk$nuaP$WtX
z&-g2n(rm`RRu-7ZE3`(}F`07nH(-)SnmG^Q8gBl;nb#x|k0^64tnmJUb!h>YG$k#7
zh3OLq`8jFZ<U3yD)J`HV_dVcCD8XNiK;o0%nse9*mdN|h=BLb|Z*r70HOd&GEV7r&
zXFC>d*eYKAO}zA%;#OLPah-_&MgQ+01h6iKMi#+!F8uvlX!cX5^FRZgcfhn>7wru(
zi8S)~FMc4Fs4MlR5P?ZCL5m0Ecuz&hAD1$m^}^OcknW%~B>#MO6n~Vh)^>sbE(PIG
zL-i#RV~~iKF4v-9?%`RMiY@mph!$<1dd2C=$f;Ga=LHxpbXHc{NPO8qp{Hg&RoY*W
zSXO(;&|`F+SgmCN8m1~r3;TD<2M``EN3;&=zH5xtQb$dXDPd-M@)a$f@_(vVTPw|v
z`U~JNXrk5ga90lF**epiantr{eUMHe`p9C<;<8B~BIO3TO*n+SLBfTgGZ7*k#mQVC
zsWi2TpR7S=bZRdViXS9J5{ZD1u_;{uP-nsyN|p2A*bKI9bj03J?Qp<p671~6k7<Rm
z67fPZ-s1lsWA6|pNW%u)wQbwBrfu7{ZQJgiwr$(CZQHh)`E!z!EZ)Vptxc_}9^VTX
z<pzkng-Y@`u3h!VN&I+=>}4V$EcqX9h>;O-4!mnS7^rc*=8*7Bu1ypr>}pH-%uE3d
zFx@Wpuh2@pO$)Ee0?)yD=XJ&Kz_$hL04W992wXC#*JpBg;19CLiBMD@A{Tj|z*hf(
z6$sQH6pAzcym0m%xcVT9S)+Pgb`~>vL;#jC%`vqwOz=e5Lr0UDm!)rtLHpv=@9c~t
zTn1|X=eY9%?$>rPyVL|&Fku9g2QmDzpY69N44h}n`?+-cL#w|lu2t=n$nDxjW|v&M
z)WhrG6*UCbvi1`{_RA_++#2$`d&YzdIEZ_P98$-8x=O*qI*8dZ?A~>KGo0$-q>6O^
z*!;iCOAr<>m&K<0{#IB}H=ATxe(x?-YK;{xW-f%zb7S%Zu_YU;qF+eZjRi9zl7k>&
zk<#xD-4s-Mu8^}$!OYZ~1)6VEa_s<C3)|z`YR{Adi##_Kn_RCg0Ne-QpVNTerY7-d
z!5f_g&m^aK55f|h<A9Zv_?DA3)|j$*U`ozFuEDR;v>J$sFU~7ik{J$!*HqQ$GMC4m
zn{cWO!ZGZB)_VlgJpG(sRjNP7=7ZU|is5k}1$VP@c#%-_4#g3EZ?d5YLLQMxAy!Ve
zPIF#iMF{MBswzpk@Doyqmq$D3xk#`w9}PjHvmNS~t?<sVDz<jC-uUzeg=$oYDW3`h
z;la$I3^P38nL`c|pYbj(c)pao!?LSh=aioDf`|cJ^+j!Nchzp^p%@4diu07c8+Ha~
zs$a_<FL~NsLGfR9>E0j?pm-0GqJ~eEz0V~0rGE&L6B+v#R<ipN`xZQ`a-GF6c_7YW
zKlk2!p|8mz;Lazfg`y3o4Rq_B=ykk<@6?^oWu46~EbWfE(pvJ%&BL-Xj#hrj1Rf`K
zuy5!SeJ;nWx)ONoaiV)c2A}RuKW~ztnj;2s0SQ3<ZN*@06>(!u438II|2<nv0P0cI
z{!y^i@M#{_C%^K&g#r}~jxqxOvsXbOfve#;MfYcdLkd}!aNwVPMCh1-5Xp{QyI9fm
znTG_|SAGBEIuiW#T6Vh-8Y@TU{3oU>>$5Zm9;Xr-`Ed~NI!MkWhJ<3<fwpyApXDoa
z3th=9X%k2rP#%~JU4q?hnR)C5jkh`I=nPDi5XC?Jq6U7b*x9dMA9>`-E&U2M_^k&_
z9%9|x%@Ued{uSzU#QGZ-MvLiX{W>;ZUQQ@Mv-uu1tbtO^CegU^vcsYxbJ!#mwRR#q
z^&YN7y~cL3m;Io){SnC)c9inEFpIlq3y5&L^u3-I)*Ve2@ts-a6TjF}B3fkp@53Dk
z#xMTN5R=4z<^LRv|B?T*FtPs6Fv0(0Sr<j}U9H_CRyT<`<yAZeC#CjkZgxdVZ!V8K
z1P>3Su@(;`<EMP7-OT|80Rw^LOfdAImcxKKIXgc;ytLh$4@1Q3=6v_{eE1^mH#Lrn
zSB@!JB_<OKlnQRgkf{~DMG67Il4~0of66sGgk)s*@ZnV5>IofrhqwC<!`J&b9bSj$
z`#HTj_?oWt92l;AWGnsU{e8Z|@A)S#$R9r+KEG@{UmxDQ|2*%_={>5F3C$Ci2UF&x
zf>3asgM<s_^Rp1QdMd)>&4;oe{w`|VbncO;^nN$s&Icl2mN~(0H1tST^Qf0T?IHd&
zN}jG*a($i?L@24F#zP#%or(*+T?DnhZ%|juG|@D9lr!2hda&j;NDIr}3s>{t(29+a
zZJ>JU+k<^7Bk4k4^%WBo_L{82XGB31J(Tg|ktQAn>5$<zi~q?<_8qM<h|&t{&v$hz
z$sn##@0v)@6@bJ=j7}pPl2sU+fn|HHxn+l8thq$Q_e4i5Yro#D`?qvmqB&_Vj9Lr+
z&26s~Occbm`h0~@i>Z*3p{wf_o;-%JQ;@P$s6K@*)LQtn=|&Kd#FP!}(y{=-CbR>e
zrNJO<etOc_^weOnoEBva)c{$^GMYkc3t=96e^|>lsZj`#?ZqOVcCE6@eI{~~qjl@Z
z&e<;P*f~a{<EhdE!|B^PQ%MD#ZSk(N>p*gl<UgIqAM=G}mLYGGfZlBhVYJ=tDb=EO
zyOg(6-phZ>zYtOe{zQd|f!u-Xz*jtwg`-Th&)qkVe!rtBqP-ia6UkuQrg($pPhleh
z(i8_afq^~^+0R3T<W4(mUSQzFck{~2iXqIO8Gv-=j?`)hiZ2G@MzJrhz?m1`cYmGy
z)pVl+mZ1&{No^zg6csKYY3t@WZ#C7AbFoeAA{?Fx9+x+0njY};5dL0YEGoE&2p$v}
z@)Y`dL-r^S$Pm0RuH&AP8DfXhdWWkyF$lfcOf|vOn-#`(xmum*7F5e6bPAxDl596x
zHbk3Lvk2N%Ep>+k1<dO7;s+8~q2(JGIC#MFmfR2w>I9w-A)#uNPy}#;q|RRMol82v
zTba3K2p^buY1q9srL?Q#=GkPt>fcUANw+xZouih!5!EE>!hq^+S@uf2@m9kt@`*8|
zS-5utbxNcT3s;^Evu3VjE%}?T@uhLZ7BN6TMx}f0yHJ}O{S-xCpW8XGYsy{>fK?Rp
z4x%^$A*YLeGr~N3URfpi*lxGF>Il4`ytS;B6vp-5Cb6H872KEk{H^mHs5JtR9l&MF
z7Y0QVmjAduQZbP1gRTpb&!;zYl1aTgY4C*e0#d3!o<!oI#4ZtZzbvB}f(h=xWwKZ5
zABBi0)P_4ix1O#Dx^^S}1&h$n<W?U_Y#L9;_iopGvdIeZkb*1B`lY`?@HaxwO_fG9
z69Ll!F*tW}h3p3yBX_~Cee%?4H<jKl>fPpLs<%3FA`?&y73N8*RQKDv)E;o9Vd}aH
z?gfN6q)j!a%DLGwL+2}=ehbQ6&9C1PXfh2z%`))#ixkIW*G4=hExh!~t<!k*R*~lQ
zeLEI-p_WB4ZO98hT9BoAk2vSfBi1Y>64w2Ml@bgBolLh~wQPy6Wo`Fzui4Mk!$e03
zp1P$4!5I2+W<#hdAiy3CQ;uo{d>P70nDqL1#ur!^#4uYYg;ke)Y2-5gUXH$h5=_@x
z2em8d-SE$BwXLzqbVc)FBogFCyan86(5stQHa*umc;`e%l^_)nZ8q6@xXI=3Cp&-u
zPq53N`p?h7B%y24=-|lrfW+$vXLjQ`A9e{$u5}nq;h6u7y|8B4x}Hp1N9Ggj<zEu=
zEikB)HT48;X@+%zIUP8+1B|~A%kbEvXJXMtJ=~agZZL*_=Yjz|8Tw@JI>aqJ(<DQO
z4@tmdI(lNlp^PzlV}nHx0QCBgTS1>T{ZBsTiv^kPlZp$hBC-Dn@CY=q;IUz&0X!~^
zLL`tz(ddMl25n>$rqUEa=f@`}hh(hhE3?;^+s_>S&C#d&f&lLiRUAXXNkhzT+yMlr
z;=sgES3Q5AatVp1z9cD@ouJX;7v3*KP60@2GI%S3j-dd?EFy%vR}j2w*r$tpcJqz7
z3(G=?Y-&cVK{%S=H(N}MC7*r<SQW6Na1eEAVfT2&i>=wq{EL$|U&zJgXpk-x%yaJj
z^m246|9!!a2+hDVN5Vugamh;zl9zzQ8^{x~f8VxNBLl&t{H|07ILX?-5*1*f#8j2e
zrdAo}T3mgJXmePMz6ou!5+A-$9T(5bh%$Cc_$M(Iv=hq{(8a+pIXxV={6^h%A+2@4
zur+ZAaUF&dz#!3mlo%EU&f<)qMwJMr{8F!VGjhPiq|u?JU0et0PzWv|%g@!I64@L{
zY`sLsgyvfr-(cTnYxyP=@;cBYJ2A3JqHQmM0#XoziCTef3KS!IV2ny@b8h6AQ`=&Q
z?~#9^!8}`1th9pBKFz`jIp6M&oEhfDI0h2n8s3yQHB!QJJ&8Ju6_K|WHbglBL^@A^
zeDyPqHaCP>)1Z;q^Yac8h)RU}ZIA*_yng1v{V3QovgO!Mo5Uj@D$>n}lQnw|7uot8
z)Kapk4qS7sp3CGN6V+g6jJJlW1{TTC=pNTku>C&Eip{+6*<WKl^hXbeRXsqGnP_aY
zY|MSVPOvrf=JHzPW?WL1FUsYZ9+Pn2GWVq@F9X6r(zaoMdD&FL{BCtrt3Fe}+mt-{
z9%10!s)*<9nf^9s*KB~-^ioIW+Jm9DZGOvSC<Iv|Lqj|?sjH9*MIthju4e&iX;fE4
zA(lgPuBkHffa)<A>K4)uZ+L?J4BvQBpB-eqGcwLi@*(XwgKiwA6C01hxoV5lc>;X0
zH_K$_UrNBtXcg{5Z?PCuexrGfRp=agiK_$1!Ty5~AY-^4Bhx`p_!di;V@at3I1pN>
zPRa+3lg)A`(4KC{*@`>4SPs*ksSDn%V2yl=`3fn)QzF_)ceCFUzcWiO3K>MyesEht
zozh519;@)O$Pl~Tl1uL9knJAq68)3Tk(x)N=#-Mg0h-6-Dz=H{<MxpvT>O<AHuYCU
zz#Z`u;LUqL)VqF%PrbOa;3R^99QRO%U&vPu%oTRbIjk)z)gzNm{`u60mS0RwL7y3|
zZ+!hyV&ur9+WMy~zHpc?*gi18HQi!zeTfjxP_7SbD(bDr(-<B5j&Isd-y9jaAMj+S
zDUSch<rvuh|D+IQ4|@{=dO1T&C1)EbdRYQS28RE$apmabOu)hR2a5jx1xIr*G5)W)
z-tqqdM~fo)t?JdsK0~)R?s`E$k-W4++Pl#v0IaTz0Ta*H)HIwWmJz+~?%Jug2uUUh
zM@j2J0YAiXapCu7udm}KI4g(?P(fXTT2OlT@gWH-#64j+CO{%dY9h=)nk>YjYLW^<
z8v!BfuI-T#1lmhJViRWtsbdyFq<91(&2!>K5{aHbf^A4J9;-83L1WC3$qbPgrO_Mq
zVFkYV15TPwgtzOGLZyI!0+RHjSDQxKs9r@X3nFI9DFzwxV(y2xRIz!=2j~O{Oqg>3
z;)~LPK@`c;J!l7Z<)J9Wp_5t>PlZRm8zW@2cm|Nxw^u1vf{=0KyNEmWn*lkSlQtC>
z0a>T)6Ox9w29yfMzFd*e2g414Q4l*DbFTvv>v)1ct@FtXU>vVeR29t6^uel|MjHd?
zO_R(4R58qu^CFEPXI3$?*#pG&Z<T2<;mJJFV7_Td##oV8HCX2o8n{5r8C2D=3fM7N
z;!X~3Xpn1UD&$!wAcZ`~P)gxYNJs)C>w`JUP6%u4w?u_lzyakAdI~E)OpEUCVJpeD
zRole`HzrYtJM`tUAcvCk$2#FK9U-Lf_Jbd3i(5gZ>srF1Lb%A~gzSx08gqBOQenng
z=!&F-rM-|zfQ9a0v4T?W*@!y@-SM`gu_rGS5Yclt34j9;h>rshUI2qr)QalA-AD2h
z((-d4Hkk+^{`9M}vZ%|)%Lo{Qj&caQ!_@hU5toT4s@0OcC$|Z_c^W@^#s7Xy)`rv4
zNTEzl8{S(ye7}OHxtj=$<ohHjcukNDT$2<MC2H(>XU$JYlk>}%`rFc<m^0nFg6OGD
zYIb>&K3s<PwQO>%T0FWJaV)ymzd<HG<f<+|328)O$NxIJ&_2W-r)qLfe!lgFApsBT
zaE#Lvc51I!?X=^}<{sTnx-@IZOg_67O5Ofw%$y8oUPQGSB!fwW#r=Ya3_~?~d@5^=
ztWKJMTCi<?Ctj~BWMYuLhfSYy^86b0eEB5wma8^8CI?T-y>EW$_<9JP2VAEQAd7G=
ziUZ@ilpy?#G(2%Wx+aPe4Z3w1I13u{E8|rKg7ABNX{oR*qh^Lehml71w(}BJOgy*X
zK$L+v0;V3yBq_;OFZ{x~(TA6plN;kVn|{Q&g_nCoDdG^CRcy=SnH-AqL0X4_+d0a*
z9|R(j7vmpF1r<h~B8VmaK^;c3r4@EluqIp{XoBp=8}MONp$LsB12(V*#K25*?7tz#
zc>)NQ(F=|pwm|;tuoaDK`vGVh$_P|`Zc6Po16AMHAA!q=@HxT{BV$jeW}+}#d90$4
zdeB+fOBrDVKjV}z8~p(EP_f=$AN%MT5dV4*rgDF`>D__GXL}JQ;}idtVN-G(DdrN2
z<#AtDd7y@F1CIGM@-stkk?>pDcFHJk7wM9YM)BM1-qxK#f6I&h`~k|OtTeLecrT_g
zTRKeg`{}4TO<?PV<r{oV=I7=ye#PvYy>O@`G{2ZtJ!S^^-g`Q^I{Ex{dAI!a(gJeU
z#mg3+{N>xTRVCv5?#D*K5cSY<0_?+E>3`tKGkdE#jBlbV&E}FWcUpOV^5d%DJAV0Y
zsCx1`?5sljI9rLtou(*Vsy^$@8W#ENooz;;+4#V!C*Z~_f<S{d{5ZMGI@xL1+^&a`
z$ys|-Ve;<9ORK)~k)M`w4TL5WOSNh^z&Ms34IdTM8e@8bL+#h;@$?v*`#UI-yWHI=
zr-q4@WE^|ak`NtxP)L@(-wQ19!tzo|q7e&{e*===s~vb3@@5>^ly09tM>e?aaduv;
z;~b;1?J8JQ6M&V1uoRErtSd+oTm~O~<Z7$H`>5FMoy(zJJ<8V=R{&RZ@Gf4?_l(lz
z)%XS_j$QS*zs}cVe|zHXF3K^0*KXNdsd$+%$GO`o`@`vp_|->LWC*V83!rT9C2lFW
z?Cgo&HYk6+mD{7+G&3S|>9fW1JovY1dr+owpRncdv2`cdS?tW54<K;fPX4<37=t?u
z=-k>qdL4YUEHz-!T-$*;?OLLH50Oh9#recRsMVNh`g)my`JAGeyg0g)Q@wt@A8SL(
zDLs<vd5)VvLDFOeG{0#okb7B5$+0zc<nPnsa_{)lmQ-RobUwaQQFJdy7q<#L8c#GP
zUew^h>_p5P^eK61#9-p%Cf9bnvugHj)ADYEi;J}wCw|}@U;s8i!0$-r1IFdg(wQCS
zwXs}j_PW!CLTq*6@KNwR8MASyHqOjzZy^%bFVk=4@W}e8;o(-<Fj>bGqMr<`20)2q
z1&#hBTh}tRmt|DxeI%?$axs0Nf>}VX;zyjttwH>36|aW%l5V|2zT8l7_Xi@yB74@$
zK{JS!#nz$oF0k|CgsW@XM~qea>S<{?n^A81ryt-ZhfZsab@k>51-j;L{N4^czin{g
zOFq0DaXSttoqE8G8%G%)=7yo8phuQO5Z0^vYas7<KfI5T_fh`bye+gVz<}*h-iE|9
znLJkIVc<o4w2b&{D3$Berr@kEwn#U)O46WP%^>V3EBk1MRM_D)yYesiI1)?ne}sHS
zj{g|nGIFr~&v3A7O)aO*5!8R}n?9|4Oozr-DYFm#&W6Pz3*-nJ^5$=V1ZpHyh&2$3
zk?&_7FF=U6yrPj7Z8_!~MzqPG#?zndZB*G5NLk%%p4=Rt$Arcq0^)M%5;n-hh&?0&
zx`cBgk8GrZ*<IP*&lA%A5;4&<VMR>+gK}N(&#TdYv-9_0s4OaHUPPo)KVR1xch2sw
zNtUVKd%0Em8oJWd{p(RDp+z&bSzrJ3N~_ip%os)7a2AIs%<4o)6>)6Ll&l(S{U_hd
zWSJfBaBDUJZAo)M2Ez=5Q6a40bn{^!Tos~%T8;wAR+L|yW5c+FBGUIk3}iusl?Kvw
z+~a}xf+kWAY;iCs=ph;MOK$2Q)9DG@nF!fy8PpsDPqf??+Xt7GH4*pi<LQgn7&k#k
zraoUmj)9`*iq_ZTr>pPuMR0NwLP26#k%49b{mb`{t7ok&n!VOrq`2)f<5S92SyZAd
zAQ4BO7p3rgU}^w*U_JZ4@dysn_%APt;uJ`fh@{+v3`TEpF_hs<2s-5mAY<|nfP(R+
zEb=`D+R=>J`bC7A<XBs%=_g)t{4nnX9TF6V+PH1VG|grth{ZbMaSn10=P(5tFq6pl
z;6hbP=uvztrv(UaZKUt3I~zXDSs_Hja-}E>Vk3p>IZ4OgY2fT{seQjiuv2X6dyJ_}
z!Y`B#3+dVK{;-&+nS{m0Oy||?DgH19Q(C%XW{78jD$=NAaeXZNf`p>ZOgg~Je5xzQ
zlxlHx!`@oPGP-nw!i^J`Z%Qh34`h7g7AmG9X<ny^?HMMGT@WfTWRGOLhhgs@I?Wb&
zGa@l6?xCD@Eti-{p>3*Y7OpMHnJtEs!A|=sk3?oRm<H-a8VYI#C?Qpnwf>pq6L;3C
z@ogE)u1=?tf4%;_%asu;+Q9${Olbs0?bl=IhR1-iR6IZkd>5e39yabE*;(rJ85D-X
zbp!0r2WN<bDhmF97frk@y<J9*4u;4=ut?7e44yq=o-Fy*Y+(V#To?~lC{B)hC1emt
zDO^%MK%LiKFcGk#(?EVxW9%?87*<dwLtn@e{63@?O-|Y9Ata<dQ4W;s(kci!d|`w;
z6-$0DoXiqq22NgZaNV`hxH)QYi`00m^sobeffsiI7@lS*oZa=NQ5ddt#*<Dm*OvXD
zYXvy@%^$1ML$7Wfx_s4<^@=73d*mEIF#Woj2c~G(^)OFy7XJR1%Akr#ls-l>fj#-E
zUAsqya2soIQ9L?>0|YuqLO3*x{`1`nfOh_8?oa*8-DRCa9Ig|WH1!iV)}NN@AbBKD
z$1@DfqvSV~^wJ+QrdzA4!yMMfH{M6Gr9IvAoyG!&;AzBQ$ltR!0R5s9aT?ZW`04R(
z0Q{D-T;ee>Pp-Qg`6EM`5q9BEBSUH5j0?CL@(1SOwKv`T)`Lw2Xo>ae7p?iX+Ncp^
zE@fP<r#y$`x=!G6%+zrHzU{>}MVOC&ME-WT=<_y8^=!s)<iVR;W&wP2TMm}Y`r+@L
zAmpZ8j7#GJ{+vRm;%-=sa2wu+Z<ef1UEsGb)k8STEo9faJ;0irwh<TVoC=38#y_%t
zB7bgvH}hHM>xej-3;jR9H!P+EoqlY%J$o>k?iP|G3HMKHFucgfzbrAl)@n0&j?@I#
z(MM{Oz>R_}g9Xk5)^B-yiKB@CrQmwl(|tb`-KoCEeqIjLbnTy(0|*ipU_z{7dA?;A
zVI#j*(B3O5(1L=Nz9}tH;q@Bb%Q9>-wyyaO>(6vlt%<4E@;Wd^#;CD^vat?hm%c6>
zRN@C-mP3iv0C8A7SgEbA*)nN->N!@=U2{GcMSzge75obz@Ye=v_N{6oo;D5HHt$<{
zgR`ruFS#M;4^-e`Ok1)cGR=Fj!RR(SxH^NgeeK9){;+^c!W^xh8zx>~Fmyp~p18PQ
zUVYS5V0VsJ%A~<7Uhsg|F<<SqF0(cOcA8p2=MK3cq!mRm^w3u77@$jeO<n5m%Ar27
zGu?cyE)R@!`)q?0vTo{QEqPk*Q4tch548F{yBXfmI)Lw*a5WgzRUh_6F>#|3d@BVQ
zs1#y2A_H~w&Y6vJKN6f^zA&8NqKz781RaErR&4l3c!K<U7Q$v3WVjFuU-9Qa9md=3
zcR4seywX7VdxE(`Aj|MyI|$tR)W$!w5{)fm7@)bACDZz5;R_>-61)zL=`QU(69&$s
zJ-n5<syliA+$AyD0L5wIS2Lu%D+_&7r09wlb4fw$CNkk-SF~NCOhtruHj}94=IeYG
z26y*^q7(BTa1T`XZBZ!Io2_Ro_^{1=6aF_2Vcay``LI&WZ2a<W6=<DKuPa|!cXuO2
zY_9UR6o`>#1Pfyfo<yXIiGKSc+8!Jsn%|-}&puhI6SCH0Js7Cn|KjknzVNw)__`Zi
zx{4X21G~rrm+Z@;^tXT0COwGo(a+2a&BL2>x2=*&ba9(1w|0a90(nRsCdDUus<Knw
zw($JGSVNQEVp~+Pv61zsxKBluV~~fw;r)6B%`3wDDf!;q1@~s+gNX|UnWg$#VVrEk
z-(i!?4w@U7QC81}E%T<>=*R7K6?b#9y?gT1U<4F)ILZ0u!h4O^0Q(2OGZ>5HKY~9K
z!++~M{$~-?_W!5zh<!j{_C_YH-t@mA9Yo6OQw08uLjxwJZoyiAAmOlJ_`h`?lu`|e
zrV{h;Lh+F&liT<0onhPy`iOITtHhiiKkQtb95Nbxh65TTLF+CA$^?l?!WKAFC;cXA
zapRxs8m&ZaSxi(OZ?1)xWVzBF!b06SMu&p*Xzf<NRa$?mTWz|hH*B)mL>I=(|H-Nq
zhm#e=Br=K!(!r)GVo($y%X9-STB~)j5xKL}>egCTR4zLrJqHpPW;Y)2yfUs$m6mPU
z={HPtI|F7^@l<PSZ=^vLW|Y&`R}x8@Ry6mg8DCl_Ej!f>KPxO#O^H{Wmzi}Wi3|fu
z%*@*%Y8}5WXew1Jp_KDgF?Rw&*jJRP*{qC~RiiX(RfjzTvfrD}yF#nu$CVi{XHwc9
zA-{+fP?n<dtn}3JHR`upMA{Z?bc`ViU>b{envl_e_<?sbaGzO8*z*-4v8h&QH&}NI
zFwnxV_@>Ebp>$-~Ha(RzLRuCn!&+2+5I1e=2aYA!XIFhkXxrvJN2YdCZ0aeaq+3Or
zS>g5++*@fH;GUaeIi)0MP_fqiGbPde)ack-|AK0l5D<&ic9d<ald75gKyhuD%<E4c
z1kD5URcv;-0#{Zf0sx`EbJn^<2tCs6yat-538dLa7by(MfYoiA3Ri?1(*+|9z!l*Y
z2Q6Wn`*2TsS|UkeB$l0&TE@hadCDSBNU?U;<u>)=U$yA>3t#)5cC_QaR8?q^Ylf_1
z@KEE-uSEnf(ne+Ce0D1~gBzwgbla4rk+em$48$$aHSkX*+d9&xyCfLc%P~RA=qh05
z7(xi{igj-!wMTVVX|fG-yBYCQuZ?lro9e9|mdSZy5xR8FCG&h1hKtX-P@tV1w?;t%
zFb?ALV|c8)4_GzW1kJ->>QwRk;R%3oS8sZ^y!dJ@U&(>11{(E=3yQNZjDvQ|YLvpm
zyU13PmBozmtX~;+8j9FEx6d>S3vvRNN<tVu*e@vA22yL1K?n9hC?=R{tl<d9tA{Xl
z9INS1cm!`4u9Q!ppbNA5**N(d03`>$Kg)MF{^!98F52O?4cF!sIg@LYw196M+WDcc
z_~wa16Gffn$9rzC2!=x}t&T132~$`;!F?v@RHOVe=`^rP4ts+!lNUsL0#fP3xrGhz
zlna=!P?#h-odfag(077M@?8L4ejxdw5zZh7=-ws6S?AKLY0@;M+erNkhd6=d1Yzf}
z7V;ZBlmI!W^&)kBUQY~nAW(jdRSAb|j6iF>37IM|7t)y-X`H0&aqr{o^;Gk`)*Qqh
zY)vt790(oOlWyzzFK{Bce<9M%2bsFCIT3>7<@x^oAi&u+P>pzk>)7x0-FMfG{sfA1
z$z2FHTj@AXEu<09`^`AfsMkKgDILYcg{`xmhd1>uSND>A{x`irncJKZC6lpHrC*vf
zYF4vZ?GG>kQviYN&#r$~FABocb^P6AqY?Sf_u{bL2C5#)o-iU-Wi1c`u9nYN$N*49
zUM)V_D=!TZe=NkH`H}?YQ{Hof-S^U5iTE&sw5+T=1L2_5ozq=_6*FLMf%j;X`@IJY
zX_>G47mVT7;>{NGT23bt?^Qv7)b4z^uX_t%OldiJ5n@j6>*pFfvAm~;7vlXjX4>SO
z7d<~LowjlpMsp^5K3Ok=IT!s^p)0T8G$DiG?B^tW45i)9;Vb@^_VF`T<Zi=TV-=c+
z)e9Wp&V>};E%sL$ppI?I$3qZynkS%ZL;b*4w;=XM`l0k&6L!RcfufHmE96|M$uGtd
z__f#mv6VB2N$^7rKOy3mKGH?t&@!;!o)3?~jSE%&_A%avD=NTO9l8gG9!jgPjGbjK
zoVcQ!PMGhnB9M&VAK9}SfRLNc0-A2f!wi4GM95>yzy?kFugMV|m0g_A)dXp~@4`So
zx5DjC?B`qCrr!zC!E4W)Os?OJ-jD4I$4fkJYfMttI?B&0!hL@ikx!UPL`Vyr0foz*
zPUAMJjFQiNYSi;6$46o&dLFkMMra${=+4pR(MY8Uh*WQLd9}Se-vA>#lJ~tjZh1P_
zPh&_4q(4d2FZ=$)X`{s4yh`0rkfe+#xIuqT!DVD62G$d0RQ07;m^5xOahM1EslXRw
z!pz|d;1FVxy^Jr~lS04MAkcM2vM&t;K2DzR>k93W>Ba<-Fv@_T(j@g=(m7kW!)2ux
zSn%=(4+3b)a+Ui6s@#JY*$)7vfh*e!xjX<g9usPkXx(w~6&@jkmGZtQqX4SdYHf%I
znDwW;oA@TXUjCNI1^D%q^=H^l<PU#tN+}IYF^9+Z$p^AGb(;%!2l>dNiU*`Cm5WQ`
zO7hPH{gLPTg-*rk4*AbG`X43_Ms}wEeFqwI-F8#-Zd<RvRWYP#C&;VL0;(+Hkc2i`
zS#=J4lma8F#X_lwNYNp>;QM9fhK4l0;g}dKBaC}(%gFU6JzHabk0!7GXQuWrds{6Y
z!VrnNCh*?kUkD`@F((PSCt-)eEbTtL-`rM<9<kDABqclW9|Y`n=kaFS!F~qQUYLv9
z;itmF>$x1}{-<MmwfAwqLjo_{|M|(m@5Om>Mv1xZC9J`T5xdok-i>{Mqus{6;JB_E
zhiG21=mR2*h}0lRC!~oCx<|pl$}N4wV>W<dcfm1D{?4eQxYwdJZFJN<Jff<cZ3F^@
zzbxuoUUZBJnukZd!G&?ycS@jNsDMW<9H<zQ;orBAPT;&(zVMh0j!N_d>@q(Z4knjS
zNNP^5CIu<WmNeYE1bH*hPliKR;jAa_%gh%hck$f@HVV!&&DVG<DZ5}vSNCm#XvCV8
zK~KgfB+Ju6%U$&Nd*-a3`X*!l3VcdLAB6`!YY%voJ79Q}PxZyg({CH7M1470B>DN5
zeW@y%2o}CT9buef!Wb3yv;f0>9DT81KtffzA+cyCk~s?)!D+(sMG-y?APCPiq=NS~
zZ-z#dJUZ0xr3i+tEFYV>{YF&E6?WfY>EyM54S=nu&q?Q2OQM1$zrl%xl1ytHQzXd(
zza3<S@@uz9RKmI-IafFo5$<T4@)n<fj<YJdOSB3oQsQHn5}pjdtf<r^CsnP?8~_Im
z*}0q=WL@sD+{tgP5vXl=RU@aC!Cvx@cl!6G+t_`-6b{Lwhhw*nw`ee<jYQ3D!WIUS
zP*y9PiOQjewT!?70`Q140<ZZ*#G;b%TyizrJ_-)In=LTnA5pzP?)o%e=~H$(Im8`8
zs#+>-mjL}(HZdH$U}-+ugszJ#yMo|2Sw8uRDCY*4Tz46$a)RJ4E{?w)%oaJaiv_Ol
z;3bpXts>je*ahm)`;zM!;?p>7B+(W6ec|yNr$4+x>@)6t6l>9pVQI*Pfq(i=TZhDn
z;S%}5Tvs*numzci1Jcpt{VKH)Jg}I~j6e?M{*5dHXpEykTL>{qqbY{D5;#6-O?Tub
zY!V#%BNASblr;rhPzZkwGxPd>pdRQX+70=&wn)o?eEO_$C;ovf!?wfKoi1V#@)nnX
zk0LthX8Rg(KbqshiGWsTsD?WN^D{^vq5SiM&kiHyBv_u%!J@BRLJLu=UB)chtI+kq
z2t+tSv?yO6fsi1MWDX<~y6<O7vx|{O5D`<@@Pb|n_l!H7oTY;xt$RYgij8o!Ip!=7
zAc@*S4e)Soz5<%nMt5MDO7=OK%+sOU!t$VkI`<b~Nj84aSTWi*z|ax*Hbz7?FYuDX
zjZ%FX4QVXOx70=k#OOiI4u$LB5MKGMvE`d!!*|1Qb|WR;AkOZD?N+nv6hxneWY)Ut
z4eKA&0psDd&gy$hn=2h#KsrJZjvYSN`lrzk6Qb*p7kLjOdB?GSO)D?O9KF7+Co5V|
z6meH-Z}$Qy%XCb_B0jjw<2@Uo=vq3SKgMD@2M(^)Fii1%3Jp8TVISgEWy~JF;6~aJ
zh40J$s^=3o_eD=P*wpz-jdqoPRd=OjViYAyWE>(~{Mhmyl7;I3xg;BuZXQFXSu)5`
zp6;OCp^_Hp%M4YdKJz^_>$lM={1(fx9F9bgx2z{v2@U{2pmf#ne>SoK0yCin-kx3D
zLn%0nm;MNi1p-wMGtU!-;$G6ke1Az6;`LvUr`-QkI|~Y*$HDExEZ!DpLaW8ygNht(
zyHD-y_9S&jKulpQqrtsinHW3RpXb~M1l||fl2z#JYzfOFBrF;p@(I3V+8FoMP-vAO
zSeVUu|J<{+`YYsKUyRf@d-u>?x{%}5kp%XWdoO8K9gUM%HTg_mRuE+Ndcun#94b83
zgxX-uwGxkyi?fS=9dC3bfm%s$4e|X;(ef*NQ<pZ4H}b4yw-6N-ww5H4z%TT^UA9}I
z;P)un-R0e-Fzi-%!%}_AX0TK$zO_2rb1>6Id1BYe(%Ar$cd=OE^O7rO)!@`^c--=r
z+ZkHh4h0ULocoT|I=;2mL+g62q0=XW8gJU4UfWzr&6z5>GrMpyO9bQ38|&6dDe$Lf
zIGL+v&?QdAgzdSrk1`JNT<=@JPca3$N6vS0s+oq32^jM7`1~=cKc11pH0Zow%i?sb
z`wpJ?!@7r~v%@}I+p<8AcSM=Tu6Jk`Bci}B&i0ilyqm|zQ_s&Im6w!U73Dlw#&|z5
zA0vEjqvK9o3}uaO@r$abOSPK6gL)hY*6WY`EE>B`y#sx287io}qJzfom61N-I!NWU
zSn9jt!=CE|faew43K9hZ=aL$HZql=N^R5l-7I>~bBsZ^#FNbo;(y8#GKi%f4sxF+f
zcQV<6vzq>0t2`s>Yvo7Z=p9AB0GK@V)c+N28UO3vpOxi*X2xw}O8xU@z31o==7ml0
zSKU!2Y{~{$xS7_`)`>9hEMDQDXw@&fkgO&oK0nRA67wY_;0${>dz}xLP&|0`iX^xQ
zCdi@t>1u2H6b+G(F>_2N4%-Yo3P*Ai3z9{tJdiG9PMSKUCq8|Am&)Py<W8rPoWI_h
zNftqWKH|CY(%1NXOdgb8!snXlq5E}zP`n|O#bL&C>^U<~7QW1Z-Wv!1R>p>sC7Fs!
zn$}Jac5^7D&zthk-+4Zm$U*`FMLl<D6LM4*P}?uNehxNpT}`?MGGrh^F^CJ|9tN9A
zRv0^gn-j6*FB<mJ$JnUPmPt;I(s%q5G>JL0E3(<H7PYlsMSA+I4IsvCllHW8yhUXm
z?Cff)w7$IAM6_D^<yH3(%8cAxS2nm5*n5xlg-<!g70!_%_f;90)|cZXGkq6VsI^(z
z-s6L?ES4uKkZ6@iz0O^IH4no5;<K9kt+Z8wn?)ZClejR^#J{nf@Zl2O-D*1i={c(?
zv5im7fd}0{lr%}gJTd}D$X2C&9mOnW`3poK1SsD|;Zo8*5_PL!e%V0xI*ew3(uvEY
zLBsaeVb%4VG~80L+V(E*^7HJgudk%YT%O3(<>J5hUeiuLng{LO#be#t){x?<)O5Uc
zFTiB<utNmeR$vU|v8LLquAd-!X==n2O$_WBwUS%8ahz`xZ<?X{rw43#AM8PIM{T-)
z*Sw!!0D12ri|NS*Oj{Ulf2^T>6Lk^}Uz7>Q0zUkd$Oy#(K5+eyjn=lcl=nkH^MUov
z7v7rLJ2liU4=gCSUtVg%o#<YdojwQwHxIv@iD?;IR_xzyp@*ig=C(~6sY{Q&5h;ca
zOO+2dG*AVBjylGC<F#%Pb*Q6oLHc`#>zs_i8kadhbS$(k#}PO+d{&jU4R%u4I^KtL
zB?qLOWBsETd}mUWAYj{H@En$yID{iY?k9s-f3b06**_4`2Fg4mXJt&~Wn5&;l2`T8
zQrE<6IfhB3Nygeinf)C0^2y{uMz?XU_L8vund}TV!>Q#*05#MJ{e&3~W9$fX8f5l>
zk=epDD8cN12JTSdW8Sl=b1W(XP*WHUb^Gd(faK~MaYl`19i}MocTV#Fc#hu9#PyLI
z<}v>;l}t;^0Tm*g$w=48@drL8E!soik-Y=Gix=w=&2cY^_K4(twIZG|lMVG5W}W;0
zrPhvUBiFY|fJn4R;hXJlCL2H|^OuP^?U9K>E?P?Exixi?{F`wE@MgTToIb^zuQJd)
zCVYe3_6SJ@BT{N#8K4$@O*jP8Pndbfm{&SgI64c+M;Lo|`PkM3lzv1%@TWAgV{pZj
z9kpAcU}UF=tB?tel*gvgKG^ZR-m|+ov9HQuQP(Chy3pi<xfR)3vJc6JO_*hNh(sK0
z*!H&IG7*9|(1tntObIwlTB=CS#m`McH<X5c(Z*wQAFA<gZ@B^lx4xr+1K5orU0k}=
zQT|~}-9<Vn8u<V&rQb2pfuRQR?<k}PtcM|j9?Veke6c*OxO5j5axLk?@phQ&m;fB4
zi|Ar`7AGAtd!xc=sfw0f=SNb|y`*DIEe;Oi#KQb@DUm*JvM@g5Eg&_ZKyECcS`LL2
ze$Cr4$3a=M!{s1-$qmasj|gr^nHY)>4J%1hdkR=2))~{xJUAl1!w=fW8}pL|+`2rU
zNG?G0Kf#`(Gs>Efo&Ru;rX8m?SOEG>GyrlbrE1Ju^ft5{+jJ24!rpJgKXV8%BQP^@
zTycAC=)WJa<-o-9$=xsJ$Doupp5|_l{$B_#SEI|?9id{lkTZ_ipg@I9VJ6{k8_qW|
zMNAM9A-QuKgwvAs8haJ}&jKK74^KU7_%p5v3w-pP8%MRPs2Qag6EAZuFGNP{ymGr<
zg}x0|fa49;zh)TA3+3L<zv;HS^5TE3y+PZT;-zrY7HgVTTbf42ygkpn9#%3?=Fj0>
z+SK36`LEb!C@1P@QVpf2D-FjFEhQhtK;Ma<2;d=DjFtPN?*syfpCEa~bI;xGCdqjm
z$`oY#M&*_~RpD?r-Ky6nkM0JTw#N;$Rb(dwx@6%D3Bcc@aq_fTN;kTn5(^#;5>9Ir
zjO3Nq10PWz++~t^NMn1NIH+SGDBKd>W+aBW=Y7Rlg>g?=tq}kpxD)vsjGi71TCFx5
zSsM1vafeoE>FP5K)GOAu&IXySx6wZfa|!Ly1+zx!CUwxsumFycE<sllJDbA^(QOU9
zT^?H!MRDK+Ltlc5JI=L7ea^|TrVTF1g-0u3CK)=-SDd(p)>Si0G-Sq-nc^_gAe6__
z6TFKgCsYLc6M_}lv8RKTYtojtUhnvv2)#mH^sDy;{c|sAX20h<THz_Sn(r-9$wh9%
zJGhOchfSRu4p50&D^$0rG1*_j0S5c9jLF*J@-9qA2ugNPHzf*9He68ZyR+#gg-4W&
ze%MJtA+xEP8xTaZAOd2uNy31GjDDX+#d!}@UU^=(QFzY^(lQ3v5->eq_U4XJSSNL^
zNK3rXAVC}P#3N*hf<_p#1i-VHXY!p2Y)5MHH0^KE4B9@u_*yFmjhmp1tnF_9HFhIP
zv8OtqAYNGwuEG+<4k|Kz4Xa0iwV=M;WS8{yGe<efBb^ktZezH5Gq8p>`td#paoX;|
z|54x{l9RE!`_a&JJoJW|<V&0IhV&K9^Y@~8W6O2c-yYe6=W(zU4vkz?QfmO~WmXyj
ze~5##3a-8L5YbK-z}d2B1`gy`j~lGDAdKA0D%6%2lGPGEAD~?CHv_lVZ8O6X=S=vN
zrmBbVKpCI_rzmz+swX<CJBVYPGIiQp_hBlwJ8<CwD8uV72ZrXYvPG-Gvlq*pINv~!
zT_ar);4Ji}JL+vO$tMglBA$;(Z97)%1lNF|O1IG8(JBZBH5Z3&{~eTZDHw4vf*fnQ
zoAfz>XW4yq&i#WW8HV`r%{SA-v-y5W!?Z1E+vD~$V9Ywe^OosLb1#W$e;A)`*okzL
zQXzRE9+NKg7WOk$`?5f)CudL9#v!T)HGggc0g75|(u%cf6^v6Z2*rkU@to0219CJI
zRxl&=)W8_yjXv0(!*cSccFTi1S<x%2hjeVzA!PFlvpNAXQ6v~y4=mQB0DMwkgpbzR
zPVBJwV7VT5e%l?shPZ0c9tzd{aIRTkx7FRCnqKU2?hTHg|N3TgKVN(-V%zzsYkyr9
z)W#4RFMvqnb$l9oxEZv1z&YrhgJfuek&gFB+tUBst6~kFuE8J?0lMD-i-dQwY_z}V
z#bBaFgdQ25`rJ%1KJAQ{A)LCG?h;Jxu{R2}q(%Q+n|yMcGJusuwqwe{Q)gYn^gx#S
zc{7eF$zp_1u_S<1(7)E|A7_tB@$<bEQXb%5YjY8eQ!e{uOgjlK;Q#LBuiY~SLwd~%
zVO|qF8;W=Pdq#{%?p;6u{=ogp_O|p$o|k&h!ZOS}{@TfWUHcVBXnpD3UZW50s-yR~
zoS6v=jr4rIp(^0-&SZ?b4s2jve$)ajjPnH#{{lFh{7L-JO!YtdLaZGB&uPs6z^^M-
z(Xq>9K=C<MdqZqGSOep7U&d=-y>60J5<62`0T&ey#3BIhpsH@!?SW>sm|B8wRt&B_
zymYFaZLf#aBn+IXo{j&*`N%_3k0PCC9>E>ZM$d515s7AQt2B~zd3JdgjpTtq1_|d4
zhNCk#o;q&A0*jmKU<khE&`$7jwGcm*6WmxnK)x%xOp=wAEBE~6d=Tx!WyyyeGF5PQ
zLDB(L+Q?cf;#>?e$#{qZJ8Bt{qY1U!VTGm9O9N#ASupm<kk23<!677J+gwM^9t<AN
zgb&RmsuG#9^4f~l&PVhNV1~rIrIZOnvJ|R7r2wD`QQHi(g%xChLIpW85!+UAkWGm1
z{}^Qj+D!zT(HV%qMV$dC1-?3(5e~h!-yBA$jED(=@OLKd79WKg6fSSrc{f^AR;;aJ
zVzWA9W!*;;KFf~-*6~lh6y{5r-GWgx>_~nf9cgv&$PQoa3RH|Ze5^Q2d};WDW4vvS
z%q`>8NW<#U{xey7v>}|b=)RnL3vII1=Yuo*o@9|6Cq1FK9}sOz05>Z*F6-KTC+XrJ
z+Q|!z{-^q8eLe_sHal4v(BP}~2E#c|`btKc8Imu#qF8i!dX)q>RpQk!>=9^b$P4G>
zQjx_R8@Hn;xC;%Q-ge)Ag;Quj)r0$@?U<cGH~h4!$JR)Xv2U)R3@<mCB6eplM5?LP
z>u>z4wr9{OI1$kNZWj(Tq!2$NFnA%3qUdFn?Jnf5?1YGF)dolF;dDvYqWdN{ZW<jf
z7QY)piZMa4=65!kjE2dpLt^r9JXDvK*|;VxwWI$;#k_o@ENV*9rcQ2$t#eF4M59c{
z8OuMuH<0ffLQ-Xh-_6qG3R8LM(v>NsoW&g5RKR`q@HUtb5gb;$PZwvLq`X~&u$ADI
zW2Wjh<#cB){{G)ja0SR|<Nx{+nEqqZ$n=Nlf3DezHU24hSdn`Ebp*uKJP^1c8yqE9
z940QS;9ssNcYwtTNNPwGNfnuMzMnq>VC6x=60VMf5ro;WzTY~qkD{gn>3#i%$!GO$
z^}e6z(4h^|QAFu)R-u#WoKu(+H;2D1@68e#SAW}oPx<TL9%b!m6aB7aZPf58Js&@0
zv2wmDHIh45Xur?1Rim<w%cXJoW*)eKDfPm0S|v0x0$%aDesx+)g*?2gHXM)F#*Fn`
zH%*rpCObG50OaCHGYe{dPT7*iE6$*>R9sf)%ol5|WWi_i&UU3dHCQ`!+OH}`9bcuH
zpImhCEcQeIn6n1PdC;`1N~(zJp+|KejXBQzyL*%b$zC_9yf@Imy$e!R*6CuGh0c3j
zHWR>>f9j}TO}0^WmX~~M{34H>)ZWrfug7z+u((yIIe}8Lvrf9&>JBw@Nio?t#y2})
zH)=G|xZH#pk+!V5|8;;pEizriTu}z1VKCdGq8p<t-J{^qY5d3<rl>HHx34D)Y$b{c
zUM_Y~aajZt?JA9$w{9k8GM!qjQUuiAzCy5nMmP2D5TaGPYQ0P#M%wWxg^s2!may%b
zD{!FQG_(6Lj2o=28(>WRV^oWOmC^68uzobTQ*A_A0H<rrH!q-16OJCFb6XHLYayiY
zxr0V4ydPMD(i}-hTVXCZI@22Dp?MG>y|6g!<W)t<oF9Fq*tXDTS?sksPC3T%N?4qH
zHCukL(|7Db+3O;H=Xmh52K(l9BP$#SgU7^|9yS9GN{cC4pmZf_KsnS<y?f0zlUbZ7
zVnF}lzgn8o3UGiq7p&sn+)6iOt`7=b?vsnhM<43BiM(CcCd)Z5gU&s|)wqxt6Ymsd
zh!PN^$%ywjgz-STOX>y}!?GF2Yd$W}(Vv-Nv8-K-_Io~7vS-c;Bw~G$d%flE8p|&X
z*#OYnvN*BWan4-05;h!5e4O2<O>~|5*2JCM-F|jcehN=tChiZW(uN%xF-CGd_gp8*
z0-cb&{qGS#7>~$N3=btFu>+4s#(3Az-4q#*zfv1Q)5fv0$a5#Q5&F3a_gu;M8)^nP
zBu47J(UB$r=0b~s;6aU{_8y{?5+;z%Os-b3HBF4orTt<%>SzQd<67y0Se}*ne6}XX
zFT5e>*vR~WqvkcX6UOk~ym@a~&V)yzS@H0JA`S>+pX)bzf6*1^%4RVmfq^RUrna<W
z3DPbW<1)?nm02`dRsPB^LiBvzJx3)?6oYU-H#ZU3j2PP*GM+9BoGCLg!&n<-#G!7T
z@K$8C5k|}ilpatfN+S5!?Q4IsWM&r8^}4X2R;~6vwZ)q4s{tB*ex=D3V&e-rA7(;#
zlvZpmY#KRulR0t8+IVf?Z^6;Vqr%gznK30yp-uw((_z~SNNS`IzP6d=SXWN79u=y9
z^4ap3b=^)FihOq9tRMi^pMN2(y@$3hER7^+GprJ{H>L*sB_?V}l^cjW9mgjO5(;+$
z%O%L30+9md9gWCNw78=I<EBtsGyMfP+f)VT^V)Dx*}kH#up~17<gHQ!tUTgY^}(ac
z*N#2^YfwhUuHcnD%w^o|`2}@5$2`&|^(Rr2{s!cWZI(DR)ligrM5&qoe$<h=G$hGe
zVKbgiMKK!MT2Hn1RD7wdiOeSSi}y$>HM5o@H6dL|gyv}T@>G$^<o!uIXmfC6v5sv@
zvaM)J1B>Ao@qQaer#WvwEEoQGVbv543$)~;ARqjfu!C7m{??VnlcH9whhtJ0>-PN~
zY{=n=#Ks6K@3-gqS&I2;i?5sZBhq3y>N2mpmW<h<ACG{VyVH<A)4G7X$(@Ry=KOp^
zVG)JKTKd_WtZqtu9vPr-zlTB-#=}k~sz@HXKYb!Dr5<XSk27?hNn%p}Y5qjElT3KJ
zw0)bYgh77ng|fP1PkHbhboAg21Fv{zH6M@ylBZButeA6egw6DwmK`-R!nvRho-H^r
zfjLKRs8HQkDARMl!GRXXsOJ@(Z`14HpIOX+n`AUhUQ)62OP*V~v*(X7aP(g)8ub-Y
z7lDqSZfk$MJ}1L^+L%f}+*x5XsIY3iyhC%5-IJj`@!?a<u#<lUyx9z>u^{!#wMrd|
zMRdx5q;&<EdASXY4<~8U&{zBCdFkx1*slvRi-WF!1a|OoJ-m&y!l7*tqLMS`n3LR(
zyU{<CPhswa5YrZm7z)NGTT$|ybcgpDRCEIm`qy_JZ~KtAQJ}N0YZRu4WYgg36-#M<
z)_FgF596tj>K>r!sk>H@C@IkOBaGdq9xdp}4~>csyGtTTG!&GhCNkH4Q?-3643&S1
z6GA6~IAQwJ@_46x*iz8hAA4~5Qd2=4y|xv3V1*H!ArawaZZJ6P_A)nKjuMUC4RNdz
zYc%8tws>O5;`DK{=m1xu)F%W;?3+lik&S~j<Pp0XstIv2C1fS<EFPx4jMM&=Of4~K
z*vSTOD#&HRy8%=Dn`<159<=e2lGV+1=?xSBE{>yO=M^CsrsNo;4Uzizy!btnDK2es
zv_c?gq|{M+BadaFq<v|tdn3F8`DD#R-DiXlPi8s4DS<r~ar|?5_V7+m1&V9pNO<%M
zAls)`&hUvpQrAPhelhz5V|mOauqg!>OapG1^p)O=dg3?}1yQ}?^Q*URK;VcUOn1RV
zv)g|pR*BTfPW4Oa(}hgj{xrpHWQEQ|_vp$;;uL`OOcZw8?ZZ3fL+>*8f2`oG81fRO
z3(A=){EXd4O8Newi!PoRAyMm=bj;K8o}1(?wtZhB0OhQmnox)O?bMIWQck$`uEEtL
z{S9=Cz9^ee6HF*|0?dz>7ik*zh4Yqol8WjYbw8Or(ppig^Nl%2JPiX#h+R6|5($OM
zf%?A~d#7em+GtsGT5H<2ZQHhO+qP}nwr$(CZLFzYefsQ(zW6S7-P8}Lh<fuKnIj()
z(*ru`28`O@&`iSOcX`+RcWSkV9Gy0r&hqv2;KC?ev>*j!86*2;3XcUxNS|^C<dQGf
zM^U~!0XL(=zn(^<$;N^s#BGG&E5#y+xJlJ^*FZ~7`0MG$m|Ubrq?+_-oJP5-`Zkye
z+H>pTevG?^bf;Gh$h8=)o`j&)B}l(m0`6lSm@DP~N}CuuQqpEly18$=TCA81XM4^{
z2d0!Bc;c(&4=|9ZKCfc3paZ0{$F%B#$iggnQ3Rkn$9`{^vC7XqjN4h%Z!1wa(cG3!
zsT`ece>UOmtO5(auPp#tMb{;Ce$*uK_N8TWy7h$nr1bdn{Hb8bzu4w!O)yDr%oW<P
ztl!LgCtJM&IK$a3Y@oR;Szr3zW)yd({dXO;mb3gl1{A7gz12>Cr?X>UKsdxO0>}F5
zvpw$i?<`Q;n1AXTv~V^1-Z7n>DCGg%PX2(1)@iJWsL2YTVDJJr?*JZFzxo*iBrH_!
z-Rgbg^)D{v|CeH9`Y*-EO3(Jc-8H+>(26^1Me)tkE9mv`X9;tYFjvao&{~pdyP18q
za8oe1o{3h}a-{a4g?!xmR467^OugBlLCojZqt4*u=H_m@!~RESMu^$X_3?6XeGrW#
zC#g8)?CgvPj!j+;mPFB^DT-qnJ^47R>eKEvV15}_M?{i9qPM8f{rQmd>+}3R@Sa7<
zDLMA8WzOmO`pHW}|GNFw;`&+pXI_1OmmIi??)*iwQ#^IZakD<-CY7k9d78Q`ZD1^3
zBsz_TScSZwMR?!tm%NYF@11RGYz?GV?dqIp6Y;Pz1BKELR*cYKHmqSR$A^TdT@4_P
zrtpg`Qj09+hCS=%Y6v|gRw#1uG%&gRv-M-wdPbdbqOs&FXF8*U3%|*yp@Y_T!bz9a
ziB*4$B^_tB;^L}P9qGhsm3#eWqQJIw=(&cBq8d7mKw^;y8yQbQ6j`R)f7z=e4%yb7
zxbw&oO_!eOlSS94lcQC5X(}C^Qq?$(X3D|koWlm^p{qH#)-_V8E(rNrhd@h(X6vX*
zdvovEW*O;0SA8C=LVQY{=pe~Hy$t1@XA@QJaG1Q}np1-vXPHrF`S~Ny?Dk626epMb
zIJRBWTCPb7e-wgKF6R!ch8{EuBv86NfR<|o<0n2S5%eS&ElaUx%l2_J+9|B#R_I$8
ziAjSoLq8Y1pQC8&L0kV&5))R~JSRntp_K8R)?q$zU8K@lCVE+cpcge}YBpjIELt}#
z$xr?k91poLJ$bLwRe$>>IA`lxmEzMiGv%FSAo5mnvPQ`p@<exgea>jb_}N+!`Pj}P
zEJ*$s)aT2Ala<qq9U))Np03!=ksT4PK!`Xhm^DVMLYeO9WQalo0f%Nl2ekB%j)W6d
zDU{xjyak?hLy<`a0W=asK^I#|RwDdpAM=)^B@>=+k)~=FcPAq+^DG`+*oxXjG5I$?
zmxIYw;yS!o5NWi)fV(3AV(6*NKu}tFdJ{6yd?J8C7>G^NqRa*jDW`Ir=qR3~WXV-2
z7#MTs)}V|dYKg}PVC=lPljC_H5yA3iLMVm1R9~Oy#^gK^HD^Wc$|aV#x8Bdq#32Zx
z#m_p@-U5#!kAam2e_tG5@-%yh0tWzZxz@eZwWMxGPFY?I^2wvJ<g{gL70BYWmGf)@
z8cQ(Mp}J#EQxe~sZQ!^YQb%GC%xQq$+m+!$8$SM#r|Z5!=j&uW;zaj(JQ;et7!g-s
zfH-xr^L9w*_VF}blWNcyQL%}K@)#^+9;!T&vtP@=yP(1Dt$*8cYv?h7-N>H~P!EnB
zFR)G)vlI!@3PV;Cm0Z|TN&R_zV<L^k3+u4Fa1>iT9gc;BbAe!@Si>d~qQ+C+_1)_V
zWK4xac%uQkggnCN!K4#-Io`edr<-;)OXWwLlcyw?$wIi2zhHhPM$N5DReLpJipF-6
zfp~j|^A56cRMFXd(v#T)R%UB1B&&Yg6t2Xlr(wd_268*Os~Sg#^U_;kxwaK}!llKE
z)Dp13XfS>1K<GPCxpQ`Hjhyr@5~i^t8A@yF22IH&rQGG3Z<)yDcDKs$#dwb!@L}$f
zkVT<>yiLZmb<|YObvkl&k!Kraez@EeQIc@$u_0W%qE_%Cl3b0{g|k>SjE5d>%heRv
zn_`r+s{_-Qd28i%y}xr-?af1N>N7-@46OB-_`%gK{N_mfgSGNU2&O%HBw>BAeM(*B
zU#DiHO{lqlNj9SQMC$2KCpj044&O%ElMH61)46)71JZbr=l&O?Fs6yQk!+kfBG}oq
zGBqPiz`cc~hB8R<hRMp4zq}75d@3r)8aRDCneZhf{IN24)~SP@Vs)8oKq{Wdc{ZHq
zhlM&EsMuD_o-mN=dDd;5QnWnZa8J%81ui0rs2;y>E<FO7L!fiF9U(&8>$1_!$#X6u
zz(mvQu-q-h&mm6?O~|g)XBVfuV7l?Zpz(l_K=Z0^BmIl6Y5YKo{pkvb+^Wm6kk;BZ
z3N5$Ql6_l^8=eym;WOm)m4}5Mn|$4;%ePKw@B{m7sy36eG;E9|IONg$oN5*R&s7D0
zG=gg9+D9K5uttf4yj2X3F#pLYAo}&#F{GHnsDgQ6AgCBAhz=o9+;tYkXU6@Fk+Ftx
z+b?Ev%b4Ooqx9{VoqZxoFkRNE(EHj->1I<d{|cc_Dx78W>vjoUfa*$(8-UZkxyWsv
z^0TZZkDwrjJ^b@iRH=w<^fjyP$|P5Kg3V{*Y^6-gju`p)Ogu;kReJ3fF;2mj+B@?Z
zbmW2#B~Q92=rsO5P%=H`SStk25{3~zcQBORlIIV7N)L*Jq6?!JnG!5^?oFZRaqnm5
ztIAlJ53);zYF}(zrw7qUQ-!_gFG)t-yPS4agiWdN124mI9M;&B*+{-Sd4NW$t%?L(
zwi~^Y3P^FX>_4XuxitK<>FKrl^JQft&P?bkuo*>+T84lhlV;Tnk8jC5!j2RZH1abn
z)e)wrI3lAZl6LNg!9jk{X^?pbPh0vOOmmC*)U>YhsZlM+6C+y?Xj^&)*xK?Bxh&$q
zj_LxSEJKJ@)-{(LG*G8BTTIObz@kv=Gf&#naq^b`IyCY6ozrCYfKY52C(;0;d;}}`
zg>K9_#p;Xg!L0}cI%9v-fRm^X5sCIo%;iDwVFM+Zu_%h*ibTc{PsWfE%D-}3$PJL}
z_ox7-N{gJ}S+rI2Q`WEy#LbLbx1<20$SnNfP**lsDFRfb!s$c<ie`bGr5q7Ec|~jK
z5~sV)%1UnESpo=f5b$ic=gS^u>7l86mh|I2hL@NMwfAHdCha43^_R_pLh3Hlab{WD
z`G<GFlEfU3v>MxgC>doSa7T!BChDiHACR0-H%jJqB}mWFAe_vN9pz<*i23NHvlcYw
zgfKv;gT(N0qDB*cviWj>+(s*FN-<x~)Z7I!D7?%Gkn<VPOXwQu7)apaS&tc>$%Dpi
z-B5+QzvS(!#^i(Le-EQ(zG0>lcC~y8(?oRbS>~rIs}Df85_Ih<S1LX0cQ0%M1A~sV
zo<DF-T$|f|a9x%grT7cN<8pd0UVX36VQ3oOM!*J>BS+_MY@E-R!rJsWsQRszLy#>L
zfX1BvteL8tgFX(sXY!x5&lF+tKdVii+YVM#F!}?@gF~rtQJ{b9+g{03O5($0`g-Zi
znM1X^VC)2oxev^$UoJQ`(+d^-ZBn|uC^-<u463jy4bHhEtk;7P@^fdMdQg?GpjwRr
zvuj@CpWB{fv(<qD-oWcu>$0VPEI>(&E%`b6&{pE~`t=$c(Pyx>BITqV2v5=dyNIQy
z-_xY0tsZ(Zi*2;&hV*u7p_K07WaiyVJ*T8y4v{X5zo@)^{!_OZic>#RwKvTAEi>|4
z^GIFQ{Yg`On1}QY#X8Ztb#2kz7k9_U$H(DoJ>slpWh7e1yQRbJ(^JXWAl6w~d*$tw
z$Kv(V*=I+v<#eXM*Y{8DZfY1jgSh{I!-hlE|EB=`kKsK7Gu!`qWV!Nxmm_ZV3hbO(
zBM+*GYp?uatiY+&!&-h)!2%4IV&Qv?6Sqle{*CX6;u>j29E{e^>*+{EaPMO0aPo*`
zf<*X(^u+k&M9b>XR{}^m4;AP$zzLy}qaji#)29QU8c4IVgDDG-+WfjRv$dstxqL`Q
zMu9Toi6i-QkobfmdS}z^2{<F}-_rPce7i&t<PZy0wkH^BtLMW#^(Yn?)F{zgUkKf;
zFJ}s>bY(^TV$jn|0RIHzCpY+8NU9eGS}-&o5h~*6^;hF-KIXFyct4*N=+?fD1CR}}
ze31u4c@o)qPqQZh=A(!dMqwloOBx2()+{w<sJR%Jk{JkJxTm&Sz)6S0FJWOa6DigN
zzC)Vi?MuUxCJS>y364eplx_drEcTegj{d$k&KcTy2q)Hb>!q=?lj(7yljXfQ)^qPd
z4h-dk6zb|pV4IcTeEusoT*n+W)vc+W%hLuVP!1B|vc2~#36|E~Pa0g{ro6!PZs5yO
zM0^X6)c$@Kv{2wfXS>8nBW-nQN+v3agi01Rj1}8uEU{;<xpwgl-}%%ORaV90rr*Q1
z)LMy0{`Ul^fjV`rM>MNac%4Ythu(__B56A~!T};9M=VZGPT_=Ej}_n_GHZYRI+(^-
z5V1nV?^Ep0ZjjUGusJMin;bjF`dWjhB)1-)P}ytTmBUB=oQm8ru#YmYd8LQf$HTpK
zG_k^!<QfT)Jg94`38X~)>Bbl>f_E-!P-HuwJAoAbe)QXgnXV6*peVAIfcZx?H@i1A
zP{O}5bZqO3CkK!CZ=Miq^u{|qtxFMfKAdY8@&i<Su<kZKYc@TQtgp!O92ohnSODDM
zr$wcq8Iij88mU5+eOsuYLX2An%p6I7v;9h}pVd95%GXAQk8MF6>NzeqLF?POsi+vn
zYj@&Z!4vg@lwHNb&ymmdFFj?Zq`PVI0m!(Q`13ci{P{A|Iqxx{HWya$YohBLtGtvV
zrReR?YG_1rB7)xslj0m5!lkWmSle((ySc{iAee;xd(Qod{B2`4(z)6y%h$cJ&ab`U
zzu_y!@pv|xbkb3keO?Wpe{J{hs;;_9)=f3EGrP2+OvWa%w9n|U;~aK~f?rAGHU^Gz
z5hlCdngclj?k>^?dZ4LM^&K2x(%4g@XC%gv7-liI9vw6dvr?FxtoF;S74j{X98)_<
zn2-mYf}6zkZQCC)H?4wlD9-&|9@;0rCSOQCKP1S>fj6dgn!#kp>^sr9vy}sR>#D;a
znl(?Eu5~|XnDG?bGyV?1X?Vr-RJL&Gcx~{X9Akbx2c7m%G^80i<z!&8arVrX?!Y!E
zw%ckMe8wu?E=Wi@_dW4IrAI@?5juB!!cjVPX=^mLbyihvBeT!@BLwO6qjPz%$ZS<p
zMlAhVkOgZvj_?w3ZU|Q42D*Oox~{<=l?#3w7<l5$(C7a*PH$GX{TMSJLLepZO9~)E
z)fyPX({{8FuhmgC^13Z5P?6f;f^n*ZPy-&K^)o240S}quylH|JJxu{G@(T3aTRttg
zkT(T#p6ZRs+VO=<g-3ytCuSWuatYo<V%>Fj;!S+&4_6rWUVPInjtj^zPb~i#M4_^A
z@ERgbf-aAFTa&G7lnK}rA}|_O9*4-2axKn$Gd&#moS3ZL{cz>9$^3_bOqIqhZQ5c;
z0RibUiUlX`nS89Ckv0-AUJN2nit(etouSG1f=#%&*a00kHMEi-SAO5!tRNDt_c`$a
z2;P@(i?u7p)hk)T%C^4-u!)=@rG4$1h}_!V=IJMZVh)2Mncr@lm<-XGH)&R${U+=g
zTKH(p*7WVCPz#Q}pi!S<CJ%Pzdu!H{?b#%xNxS77&MgQdeTs%mI)yb<2Xfgdk`(sw
zkw&&>f-{uCzsosFb+g%3hZPdbKWcx<0^>t11z{B4-+!tz(QKuieD?vJb`$#R2DR6l
z9sL8U5~x4+U*`hzf5i`({`bh*81?_LvhK$#Fc(0?<A)g@Hg~PB-KiE-VP!4<6yig&
zAXgk3T~E}>0q@{L#&FAC06>6X#`t=F16nvoFP?u_;kDK2_Dwa3lQ!;)H1h~+AVhSF
zFKP>K`8;!&cDH{zedev*sZ**gsjHBBXa>NuUCs3Qz2Cjw)VX2NE~=BdVJQ?pWWoJ@
z!hrexJ>iM<`+BZx!}D!g)GcZLy>d&eDYorihoMW7L?!>~v304#d`h#3Beyt*C9))W
zT6v5lVLZ_-%d;5u1dh+0L1Hzl*#tBLW!0CMB%YQeLNy_z;13pWDh`y`Vxr#8Wj;?p
zd+@LneJb}X62-LW`l-gezMG+cHD!w_wYm|<@MAcF?31a%Xi#&L<=67cfI{^d`6=l$
zQrW&K(}W`zI0!4{((EZ>PRQS`w#aBCPoqZAE-WS1FLNzrI7({KMZI(3n)WjvExD@6
zlyyPJ;j-S7buoI?C)I=`Ba}JV!>bUNDNOQ$f+t;NC9Y{l&No7-d@*44%m`2H0M$Pu
z41*A#<QDtA$Et;d%~qAjMJq6`BP1Wghh-u?lsl4mQ$5-C-EkjM(S;T#HyIT0Av1W*
zVDV5LndR>kKpXNi!!qYW{IgSZJBzt`F<>)s=!WCNKvU^t_dQxejd3MaWEpKjbt~+6
zHa}sE&Z(*gA<_G;q0BuQH7L@xlmR3Kg$SPG!*+vtc~zHIyK-Zm?Tubv`<Y*jJ1gHz
z7_Elh5y^+Hu)aLQ9Njr4)c@<$wgyAl*;qL)umr`jT_JGC;PoKB2AJezFL2=WGy+i}
zEVA!RBp6vtEz+FX47=%!(i|>P$y4Dl3;@33X)X4$vTPaVHyBT7SG+QbvmeS$g=mW~
zh5P!(MPO(j;ox!S1AXBIgHBa9v?jI#7mT9+G8APo!UQ0RYWw+QL!G5^qLFE=(GDhX
zkNY+(uQ!O~Ot?~o(J;wkxFb#?FRcoa%#<2|@hN;@2muzFH0Zp3^a}tXf<Uc$W@|YH
z$cN7!FkI`GpQt~w*MROuZWaggjHaFTh|1a-EB`Nr1i=A4;$MjAq`+%BiYx65Aoc;V
z%d-%MG(%tWg7Xatm$)cE@5s6zAh{_BPP>nu&&PaJkrgZJ<^>jG+uvu}-+EKe!|Hzu
z^C}=BY4KhJ(A{;!Q-e9Q;^M~J)!&E}Uo5bMD#Ze0G$ZInd7{G8Bgs1P??tlWY{NQd
zrHo)AU!pW_Wae~*#%#$74j~4|Zy_jT4Pkw+Wxj3AO51OX`YY@L#c0>RyFX~Ch`My^
z5MG=I9S_we;KRr=9q53{);?9j1_A0CRCZNA5lKl=p?N(%u>@@wh9N9V2P$k6HNr#-
z0!R_EU%L$WXA)nJskv6V(cbsjV`!0?g!Rz=Qa{efMiY2eJ{l;t{Z;Ix82l(;wi$fq
zN~1{Q45faLx>rR!ijfv+6ymVWEm24@5T8{Eja=XiP;B;c;C2*5^o(09PF97>s8Vj7
z=}d|Z8te~xQF2%cJR&Qc!#zC<%&V$dQ`F5ywlQ~J4%C@ftO^EIA|;6!7zxZ|1C<3t
zv^B*dKn$Gn58S%0a$Wwr7bSo-<Yo2Cz+i{Iddp}2cA9*WnDV1^p^lBej!Kh!vUiDf
zCRx8TMAdv3hX>nRL+Y45n^$BlUii25ntLJ=;c>&}-OS)SHdAF;z$Dz4n0afpnH-n1
z8dr->mxu3Wl`>$wl)45YTNZb|>pH7t8CIZ{xSzQ2+&vHa)(-voK^g?Y%b5A??`7h?
zi9k>-PlTq4Z1#3Lb}`5lG}3LFUCy0H$VnCOAIRVk$w{m@Z47W_;49p%CRdkf827$)
zJ{vYevkk?&-~I03cs^i~rn3olk-{9cI+Hvck+vQ&b~^UGqmF1w`R4J}ZGtM{!6*5p
zr|0ja6O+L|g)lLyUT!1^=hN5~2DcUYSAVZ!^MzI-d6`cI*iM&SywFN4WP>+mwg2IZ
z?|+_c>HMA=UoXNEL;;;^B^b}MPW(%@(cXG~Dsv8W?0fWa>t!=0>_6V7iArc71`yL8
zT-hGgJI}90Y2;1gjn*<dRH^s}(VT-ZQ^Kj?+?rj!e*JuGf4^d>0L9tA{;VeQ<(T5M
zMmlSFyX9<UlZlA#_q978Y(RM4kEJa6^FGzb+Rsx=P3?Qq@T{MLw9-5N(K%=$!=*O@
z;lzu#?J$nS$vMb8U8{7dFzdVd3cU3$INtUO*;xsOodsB4jApN`Y*>UzhJ_uatF<5K
zi>vfI$8FHdLl$NO?JlCEcb9*0#{+5e|D*UMe9%WpOUqIJjcSyYGi1X(DZgO14VxM?
z(30<erCD!&>g?)fS<}<4qI%H3kEp{aSE@$t8n*_9WjNJRIP@Ax+8<ufhPYn^7H_Sq
z;FfB%wIQapmV+S`SKr*K+N{oJd5M#3Ve1b}buE<C_n>7`iTtfs67Wc`C>drf+W>V3
zJ`bI*A@nEP?1rIgGAI{i=@w>J;QE>F(s5wOpWCTl!MP)if682Bi2zco3>zsm5eag>
zrjMnXww4J)nMlWYS@B1Sno0b*zhNecE|GOUfB844@!g@6VvjB?eH+@IFzsVyE5nZ$
zG6<mWV^ATqdb&53Lp9Wg>7A;wu#Fks!qby8%LAS#dq&kij?9VMAhLt%<5bnZp7~)o
zb=E46r6ioM1$E8|mk6na=Z{<~4|r=p!8C;hp8?AF2b+1)C$m|&*1W;h<q!<0NrrMS
z=I4H+gaQCgZZ!|o!BRedl&dd#P%*i9>y<8vp~AZcAv1}$y0k{Mv<CdMO|22?WkvcG
z`Q}1j!MKU4!2~GiYvk<N3KJ)s5NieaFUpoCjHfzU)MAC`3&HarL!X*bfw>Yx$hb(L
zszENcW<6vY%&xOaV6A;>M0qKnZ%^hRKS~7{o*Va^)HHmE<A3F16EoKL_gwGkxK?Lx
zMAJEldv*t~Qgx}Wja5Cem%>6c2DfQvY@6GvkJcff08oDwSdeayAU?|DdK^v1z~gE?
zp4D{cR@WlC8sk`&0YLqhtD?%@Ox+7!Znkc!`Tzmi!sq@|RsF|po00v0UnAY3DdlwZ
zKLH^PbnXv!*~E-wwycfKXt&B%QVw}U<Azj3l!Qm3K3}YJFp2r*?27ADIG_Nq`SY>0
z9siF*-vKl}T^~2Mm)oNKkwSeD%B7JV*Mh~61&X6h>RZ;RI*9@1F5S#zvb^0r-w*qq
z<WRD`o<3AT;@_WpjK03#&(FC$JHB1t-yhDUiJcohUW4CHIX;j;Kc9aQiK&}1ygI+G
z&#kUF&6u|3P=_QSF<&tPCNbxchzU?Lk=lX@w817aM^fj8_ATItn*kgB#Lmt=m)eRX
z(<q&t*;dr9&M~mh*fOgGvYE=(&)3jVSJbh|8KIU|gP7UH$+s0?Ru@HuwYu!7+A7$$
z_m0{*8#B4*nk1j52A3!!m8;oC_~<8RXAoO;*uEbKogl_b*}3_7kw!YWx%HNc*4Vw8
z7wavdu$`5H%gV;uAM&nBj^=d@Tvqm;mKGf(5}a5vn-0F!W#8&ftAuM(WUlyp8w=eh
zeK`{y@Ld!Ao%-mg5L#p=MZ!;k;*-kCs@o6DX$rt71xeC29Xb%Y^hR4OYn-g34qCZl
z@=NZv;)v6A6)(;mN0nArBvTb|MMitgo2gokgzT*+Pq8d$#5wSYD{Xp6oLrMnXAsDb
zlc3?ezc3AD90mBelS-gF_;si9tH$3aO)sm!&T(CncF;2}9{vBaJ1TPq=6m@-4^x4P
zOm@Qc*PV`*FW7PIGES*znb$^WsIQVBOjyQ1mVAjolPu+NBDqhuMg*vW^@Hda*fDge
ziVXA9<1IvTpPk6rL1k(BPeo0JPC98eS{$uw0`ff@9|yaCbUWYIL-*Yq^%+&%yX!S*
zKW7{RNP%@oe!j{cxBC6xY|l5;<aRpJDzEHy!(`&s!<NgD&NUHMFhIeDwCW7cGda+n
z56Kf{T8I=?K>#9i@`5yB;6{}W;p!!ITS$N{I&tPBupS+~n!{xJ1mmsgY0OtoBH=@j
z_FzFd0T?#R(So!3EN?FUEw0)g#9=hC@6i2yrJH5t7r<c3z=x()ha#_1zvSeGm9wT$
zEFU@+AgmnMupIC5<z4fSndyD}DFjzRIR|hA5T_BX9s)oF+%`A^^Wx!UUqtQmOHsrb
zx6I$tf_b*U*4c205I-!vy?&bJg}4JjGW3q;Hjzv^3b0y1VN5_k+Q@D>=Z>Tje`{|V
ziscxfe!erhFa|g+_w!Omg0!~Z#Y<CaQpzyR&NhwyG}@`N(#o}r#MSE8n(}1ErZWK;
zS3%&)M+j<*n%bo$HO0e*!KBi(gva#`h8GKvA<k-~Y5*ND52Hzh&c(W}LI;&DL0MYF
zA0p;TUvw?I;^wFWMr>~f_L%V6eECqX0j}`5B*G5}FIrb|-oE!DH>ja6qlJ4j>py?Y
zJ#~4l@Y3NPj54caAwfTuRH|?HI5<MNc;e>gg7g{YEW;k2q3!~JaM|G|d=~SyP=|W2
z3pVB$x#Y47*;Pa!+Ezno2K=_7ud2M!EZjU~hiQF%W7432^~^>aj1I_U1xGUx9q#(&
zMnV_;lGIn9be}BM$a{mwk0|izXc%AoU^iuuQ$6vCA}TmTFuE=EAwZLd0!Kei_Op>a
zDRMhv-YCSnd#Dn2^ZxgpCrzm4)_p$q2l=g?_&IkEr%sZhkOdpE8$mtGVR;jRWImVN
zk<@mF1KRNOqAJOWsTeWpdi$mm4kIOs4mr4_Yjn57ByN*7&1cIDM&Yd)O9S#s^gI8N
z#%WxYM}-qcq}~WS1j0r1GEJ*51W)DZkd5VeG$~U&&wU(a^OGfL@nrc&a|1MsMsUyI
z%J5qPhE8blN5innp|NL!BhN&&Wix`Ar3F$@ZG4dkV*HaSRgixARr6!`2o01rAZRp5
zO;vqVAUJ~{jDNb}iOto2H3+6_h<Z>EK{pbS5y>CB?~&o_LST{g&er5LaLHevtnPF7
z+xYYO&!kUaJ0H05e-GFnJu6})Thma;IshQ$kjV^ZB*f<h2KI_M#9i|<KT!WU0>Hz@
z57BgW*(`Tf4{VMe;??R<dP=)8rAc_!UctyHJTql{vs2?{HhKFmzUNAia*a1k04%|b
zzyt!O@~mbb!{0i{$;fC8`T!#!`rQM&@aHGtfO2S^a57;rMjnNzq>z-#<H@ayfj;}V
zspu!Vh^*^^5f_M661)H|a7+{?={~f4l3e+Se0M_UZx|3Ic-vzZalhC~*y(P&b=MTN
zw4Gp!xvwTHKoOBWcCskMQ}SNZ7*U6rkVJuxewhlrf0PZnFWC+rBQ9emYyzbCi$mLU
zD;&TSrJwZI5Kx?JgP=IgcejFg06%SqfPk5vTlE=P5S9*d>A=W!`TZ#kLx0kPpl1|o
z==m=oZG~$OGvy&7A<&d-_Xa^heStZ{IWqPG8M7^tMwspP106?8l)29cY0H;8ym05<
zFHp4iMSmsq<vC8_78;pKnw8_T&?$wjF&q!1EMSf$FW`}MSEN!+14KEI4xTsxU!U!t
zAZBSA4?~jE2+y7jlQ-T^AI~&Qp;8~B9D!l(ePbMZNW(AbK>l7*ydJsq|4F0Btz6ky
zA!yPT#GWmf3~5eBNrZl(>0(Pe*5Vgui9*<@({}i`PtKW0=Qd-{di<q%3hB5PcDS&p
z$yDPB;rK3M?A=Wp+*h4;arg8`V3AALjCO4ChcLr@J}VV*QJpBL`P>_}^|TuTWZ;CP
zTiPJTH_Sw<Lon7h2!Aoz>(9ylHXuQMr&QN_FjAF5<(wuw+XL!gt1^9zotwpZL3KqS
z7HIUqohNdP<cu#uG#?$4b%MYt-JlE5^7z9_=dgNCPxy*~5eGZ>sJ)zw6Y#4Y`6O$Y
z%n5Z*wNN#Vtvu(oViC5X@(U!e;0Y-Z7OX?>)(89~sE(bzM7Gcz`_%q0hx410v2x+y
z+2jtIKIGjhuk<RGM7mNW11|1jDm(J~v)DOuBXy@F#2ab1nZqo<3+(--*MPxjG`<6t
zJdgS?$rmVrr3vXfRuCn-!6xwL<SRdFMPi?#0}a8oNyD2?ZPSlo1mNfKhOi(Vp`K>C
zLvkO4+u<*?+g68C`&%*gA6gVZZ|w_yPx|zs-$u<ySWpI>ZR_Ewt9vWMXBSv52|Q5n
zRA=Xa6WmAIfHV1Kp^uab_lUOPvfsw{Mmv<fD66=F`?Vh&a!!5`Io*>JMVwgPoR5>5
z8)v`^BW^S7rfKq7WOTE?V)H*3E`M>u={%oFvEcRK%0uXa!wga&A59dgp<o|EFMH29
zZIMex*}0)?(f$M>CogF2x2XUnk=Hq~fQ7&7+zjok+^H2+vxP<Cov(dX)%87$fOY1a
zP=%|+<(k7~%FF+7a#f(^)%`@kz0{NR|I2zBQA`my%mQHc6NiCg4fYN8EE&qq4sCTf
ztVni_gDA<Xxf}RCj}&-yXYWcc1&IKrJQ!$ntY6x7iD{nD>GD8Pnim}4#tj{<l~+I6
z6=eO;SS8obGk;;wZ0jX2r3~D6b4l_my1^&}pDA)}KcMVt;VT3h?X&SBcdeQuUEfc0
zZC;%8b}ft-VZ{(`n<F2@>?(C-t}ce6=YMyakwLGW?Ws|!ZF{Z4eh>X@!cup0=OV}#
z&F~R`q*>FD%2=`cL~+0yOYY~INwuNF{45f6NmkyHzMG=`1_c$SfuG<`UY#E2Zcx-g
z-Lb|NSLba-zFOM+{=!oRK&In-Ak$aB6#fNX3l*}M2oC>A>9?iq<`DRFI0cd#!)5sf
zh5MunUMsf$FKx~8UlU6PR*wI5l<J?h{=a~92-Wr?4@3Htag|gWst9S)0(3odoPN-D
zVT?c30Mr(ne!Dz$-4L{DQj`^v&IhnyJUn`QWn}``)j-(YT%Jx2&zIn{Kq1P)Mf-=G
z`U?2&!SrYYBAD8t!J}h%bU#0r;DIat(Zh%i*uY3aFFSgD-)AFRPe-6A?8@iv1ccbW
zZ%2K6{l|N=yBi-%e%`;YV{n;U2(F)UtYi>M>0^M9>HG<V0r(7r+MsAw>nDe#T)Ot7
zgEER_SJg9@qlQK7n2rwA5C!2-ET<0>weHqhf{=*rjieMyIiL2S<azo%zJ_=M^25i*
zd4JP~3lb8-ZoWzNNJSj%M08KuG%UhqzA=OAo=SYac*7245?Wk$N@uG&0g;9kteL^s
zNh4Fx`wasO>03A4jo5l+BCEAcyNp4bm?%f9^5iQH#TO<*J%DKmB-WMjC3<J4iD3Ok
z(|<BcUUjrKBw%DiRXRb9O$k4nWuT;t0@TzdRCJGRZa=iuT72?Irm(m)No`syFTKvD
zg{L9L2!;9CYta#tHH-V$ZQxPiqt&IubpdDbsP4XyUFF?(4}rLe9)pjM-35xF2H1ra
zK82>`qKA?=jF|-yE())|iPY=$bB$Sm1@Xb8aR3`Z;5B&$2nor-(*1!IM0CW}EOcwF
zk5?xwO{Rg4iffZTO--?)lNToNJv~5@lHm%|pzil%TBlno21q?rHYKbW2Zk5{<=UL=
z61|KR&`)ElTcq0?x`EPWs%(tRGOa5%IzZVX&Khl~ol!|s>(fFtETqHD->aR1{h!d4
z?Gn;dwg^ygEAG{W-PJJIq_cEpMV$8=+=t(95m*`pOa}y`SN*%_!^7A%7y&1!{o!m5
z`%6+*(&zJU@;7W*Rc>%SH$s|@BYeX>4OUe$QDWvWg<+L}9t2dFCsg+L4iHz1UE6J{
z*X;N8HY<w#rF8<q>~K7mRyi^6Zj`)|nq32;p0~JJ1fkWBO)|eN!&JZN?HguteoWFg
z%eL*dGE!iT^QX_(#_K8Zosd^4BD)<ZD{F6}4=VdBMUN8R^l(chveXy1J91loDJh=z
zsvSMj+tt=o6qMamr}~-f>_lgWxXv7=terK)C2pxzaFXn|G_#<M2|z>=xVWtS^{<Z5
zsTHl)d@G#`u9!>9GD-DPc#CC9L-x9DuJLP9w_H?PZ9S(k533)sZ&*pouzPyg|2i{2
zyMk5(Q$#|b>v!;WeuW1i>sn^fqky}GI^)2LBc`7dYk=Uz{;((|!NaO1bL%q*Q5&-I
z0g^{ByfO$w;*g5uD=(gPPi4>t^$+T-Qk!}BgESL;unz@M3WK~L=QnGxf>crf{DPqV
zbg1Jlv?+&Ap((ZvlOqo2=s0!A2xi8Ghpm_~MFbP40xgIS4g?w^d<w-+R{$tO0>O<(
zx6whA$VZKO0+gY@f!{_M6o4BH$WP1<ctCa0I_o0H%>ar6CQ!9F;nrhAIRVbYYV)qF
z4<dIdMyGhOR&kJ$Wo8+nOBC<8@V5>@BRh<Ab(SBkE(xsH*N$>t6#=nEq!i7U<AJEo
z0bTyKM`1FK1rCH?M>~WAiX#g(%bg<@YT)EgO7sU=%O}8$|1hiR8VaLlK(K?6LKsH-
z>8j6v3nurbD@}Ye6TGc}_>__ozh#)31)fjzl7^poyc8R+GK`TG{Ob3)TUT(1%JVvj
z0{}OEBO-ps5flFvgpBA_7UU#{)|+-_`iv;JfR*+~z8zD)%>c+Tzm5PopZI|r)Ezkv
z7=nw^YA2ek9eCJmFmf~}!jCA@4@ii{?#>%%c0xcUij?Rth_t^rShhreJt9{3C#k6V
zS$jHIb0gN9L&vPGW8S@>MmV%T0<9pyk>ClXC>So?G-yX8Voy>JHz*GT+}Pn}gM8qY
z<|})W-4w{1k?mZQ$eWfmOw1|eHr})0%sDD3Qj=a>61HqF0&J^be_BI<$`#$GcHi{@
z-Ic?XLlz>3-ICP~E)|PM*|1UU^c}KMJeyEC$b!)^D|nI8_Hs}mvszQK*MTEzgKCa=
ziD|xjk{4%3ULpu?`T(;4&t!}g=GJsH1Y{=$sIZhN13I0&`Ab=Kv;IuE@3I%*c!Rjc
zOA(nnSM0@qhz={ZHuHeyyH(cT%M~ED1G()sg$q%qgrC*n6Vn+F435i2+Id476NVh>
z;q|+G3PeQk9_00|zK4*FY-xs;Oc@J-t=kH7xwyMkI<uzp2Iq!@z3q~WCgKQcdm>Vi
z)P0RvR;tq6#8r!<x|b2`kM1=3i;6ntum%~UvIZHVv6R;&bX8A3jxF9q6R+hn7)7dQ
z2erfz6gHrS^mNTjkq*jZ>3*Q_c1sxTcj;OjUz@W<Rp^BKY5r-({#8<aJDE^v?YGO_
zg{GEqM^vI4(7Sc)mAT%sSZ`Od0VmEW@&`fM?@Hx<z3RygycW}h7HkE1Ju2vvqU0$x
z*A$_csoWC3n4DcB2XrmI0(SVEH|Amz?3Ci2dKt`h!TfaFr?*Jn@pT}d2#YaZ`4;4*
z72DepOAIbXzPhvjeaAvgm=nfI-yT1!M~vw#6YgXv290)5)Gb7<E*1>|S2V=e)87J5
zqpZDQYXSwtV<T35a9J;#`tu%_N@0z!cnOjn=XmSe<$d!e%3=R!ST?LE!wEQ1TEWlA
z<@sk0TuadJP5GPR#s=Ullu5K@B5kG#1^_G!tHsbL>bGv=pp4E8oLy_F&J;Zhkk2r5
ziOvj+y|-7>fR8G>ryWIG=W1vm_OUTw$S=c(qRBbsSblhFDb6d4wB4CqZ@wELIZXP;
z&2eYq566jX(+BtRU5?SWYr7GoX=Dz#Z_C0jUQt-7q-%gJ_*H}+PnC4mmtey?c%pYS
z{C`TB|Eeoz`6qDx&-4HCNvxes9P#Nytqq(_giVa>j7^|;d7+$~9Zd{upxifaG&XFv
zS&@D6>gJ2U#^;MsC9O9BTn)ivvCv}503##tm@#9*>QgIF=_oes?n-DmQjd2WHcQW=
zMhe7thn!|l2$ime3s<)mW}TZ`#>(qK5Vj9ioh>@HA9-ZU&$O08Seif|#@!E{Eb(mb
z{I+rAu=6&#wv0y8zpri)etf@=UKbXAc7LvJ9QHSszVl!kd_V7FEbgZkoZp6i4z?a1
zo_AJr&s>E(c?)>Kg>8gSAy&+VS;QrU_uqS2yi>_mZR&WrTh?x!pSQvV)GZmNTp&%R
z9lb??_L>f@i!XI<g+AvUfx_AYtWqLcu)n{Qle0J&(JpZS{k+k*UMzXh8`A3q^<%pt
zIp8S)deX(*(eb?$G(^*Qe)1h%-J@0DAdF^6vlX>YU+k0b@VG};bBDTN|8*pE1t74~
z%qBImz1!=;k`=(0e#3Fng8qGw8>hFeD;=?IBDh486TP87Y=Vh2?eWFFjH?@^Njjrr
z0Wxtor2MWJ-K!D+(M1cmOXLqW9j>9_(O00YPliK0A#l1j%B3g3|G*~q&oTJ>n{}>d
zQuwwh97X7Bi!~6hzU~(^XQHrcR{v?VcSw%p>;ef=UBoz<$VpK}_e!^7pbhPmI_#pr
zn|HSDZ_oFo#Vr<At~Uc4o6t<IQ&(6vMF)^P752C`qG<_SEyaO<V<81;HjHAc_x_to
zh}S~+RP5_2P}_LhiM!Sjx&;0NuKoL9tZLzxuZ8FwK8O|!ibdn=_kK(NAb8kNkJp^I
z!6`#qYr{qQ=W+!t!FJn24U^bRA@kVbF~#xV)aGS>Ve8UH8pEwt?O)SUr{kHR>Wlu+
zdA$2y^u5lx>#b^a;)@&r?J)L$bB)!^bDmB%<tx_g#XQs9l4vyxMZZE`n>1tc`n0(a
z9}l_7H#kJ7p^GFTCRiC(qlp7rzu!xCkTS4mR<!m4hCVKNX@aOsN4Q(nTYty;o9pLy
z*V&cAt^({%#rD*Qy5r)psstY<%Wl7T%sev<y(sp|q$Hr{a7C6qRsq|jq0h0C@jeU=
z*t=9Au21jOJ&G{N+&L!pC84gPqp<V9y7!q#pNX;PfioP@a--hz_uxOK9JNy1BQjtJ
zW&hYO4)3N&3L+iikD9)P%L`)gA{Bc$N^s-Z@gTeU!*gmpK!@PfHwOr}6<m-H!5h@J
z@5n0yliXqN%-3)@DMqN9nD>McN-7BJ6M{ach_1>PYzdoO$`q4mp01{u6z5=Thpp8t
z-ON2y9*Ji6tvOGoiT29Q{rl5MLh!wvhGP~@`~b*+{YbF(EUA$sybQd^iXNH8jw;BV
zQ;icxny4HCocy^|oXV)7@l?<lTQ!y$B9#oTJZUeL=Fa#`X%d(-Ny`A0Q{tzLd;@wk
zbZZ*ma3N|=CI;f(c?a5XL>{I7UIjf!I;1G5E4+NBu9beH3b7#_wwuZ}-sBUnGWuDl
zGP*4Gz&uG=s8i*d2hsatI{Hrg&4EXCv8%-`(^1l*7a$=4)_37UK(g3u!u={bu>{iE
zY~_krBIshxfd=sNSX9X~bkbej6Q}M)9V?H6WVLH>p7u>CZ1^~r1BCSao1DOcnIXg#
z|9&Nml}wG`m^HS=VQQPMDF;C=AN#Zjv0)1HZu}OYY)~qPuNume=DCIGfS#z8Y%~wZ
zYp4pnvt2BWo}^KtZgq+4{fDlO1^NYh`#HCxY*v4kTVA)RDSKQxnLzUpW<94S;<ZOz
zoZx1s4C=xT6aK!+B8u$~cL|2%mki;e$hHSVmaU0cf*ly+67jh_0#zBu%yCPghMocC
zIT1-GnV9q08nIRlHi~igA%+AQEVfG!X)|wT$f`LWfnYkDJA+o^c1JPRqlwfKwY`C>
zjLW2g?@-583N!)DpBs|WNr$>fFL9=WjSd*R5k~1=xn5`~4Y_fi<G}t&3OdhXX@&u9
zB%AhxLb);N@hkvh^4fhY%gk8oIYIr~Le1}^n{6bbZG2T~3kvZ9+js$2#}w>M_(VQ}
zxqgJzBJunNkp~igEB#I7f8_(@RFW>{lrh=q_rN^2b8$`=umaKgUq=k<{~v<&KaeYy
z|2^dDMnl4Fixs~2O0Aylf<v&LZsw^#28V5CaaBRCRS%a&snyalBDL6N^XC;7BA(Ek
zJaLWFqa&T3IUQrx`w0;E@R)jF!uRS^_v(RZ7BMKgnDGJzkrZiwWKf&9XX&_J2K^TM
z&DZlQT3-o?NRuFzg`A=|_vd|wuBRtRHVtZ<wR{N$u`RFn>jp-e*RSX9RPXz1Xm^TD
zRqKz}C(Mx!A&R0hnN9$L!ZO7inb|HV;U%rOFeC*}Iw!*a!Czqd0%8P^Q<22h_cQQd
zD~w1Lz)?Yl1QUh?3rZ{6!&(&NtdB<oJW){uM5mVIiWMiSYb4&wQdU%WEMFj+Y{JtS
zmV3n?7BGCFKBOQLS8Oh7h_937prkZ0@E|KFjNHCR0;y7-xQNSClMD*$M<A-o1&Sux
z&zrHk8TNTGP+QfJ#<Qk~-4QMvi94;>b0`}yX_)B4gMt+Mmtzt}GLz^S!KYCIrV}cL
z+OQo0n+F<n@B*AtIu7$+ccip9GzQXjqX%6xg#+jJ7v=;4=;z26XN>(tI3g@XDx7AJ
z>clKOCZ<!pD~n$pRiQ)dq@3Il-z_jTC)sC!0<0(l6h^8<0CVD%++KrWc`?^Q0F0%;
z|8pq(c|pjP<y>w%7y8W{XX@`XjImKlPW7xKYyb*kEE&-fo(Ln%k^@MeM2bzT0Tq4<
zVT3uu32GkT7+9Z-RT<h@nIN#jyg-luU(64#PQ4(oRsbx73W%fVx!?eei_KtE3l&Fy
z1>`{c$C}9(O=2%GqT_&#fDH(7#VdBLUn(#rvt$MxYgjq}WT|zv0X2=1&A-s3G;<v)
zRv@4gf2Ha+tGt)?WWQSk^tuG>X$zmFfutfsi>xK4Tm@zs%uYREHTfjrn;~WxwUCnF
z>t(-ooo*hTppVktU;tiUS=HbsHu3}vnu3*y7zVvjs=?w;n4u6M=xOyvA)j9|ofNco
zB$4NS8y3_SBnb-r{&)IIEwh_Y2CReGKUEZ?=uj8}zO{S|QUbkyB|g~59*8q{7b4j6
zOrvZn-1h9k`#|(bg!QH9X(FrxYj7L}wZ6Zv1@NDt-$>ZM3zk4XhA9f5YIioZuoYP)
zxT@<Rjv?SbqW_@JaZ4AP)WEbS_q{*-*^_Nnw%a6e;pfRX9p+~eFq9G|UjgmO3+%0M
zjz=e!0qnT2HIUyotz@<!TYRL%mhQr@)v7)_iFdj5WYTN@n|4*bW2hfhi9Rjh7OK5v
zJ|PZZiZ!>P(Rf=H^leqUdDXJpQrTu^-yg`l@k4bHr|Ft1F=ftKfJg+*%rccIaJ6>z
z_Z-X6-0NUHid-H!K1>UF8vDI8M^LI|FkvP(Fg7On`{I~+3M<DooEvPhl5jANw>v9F
zC<2i<+{i@KJ1&0K^dc{SC!O0H^t)!b{rk|}>m_U?QX};g#%1J7gA=nv4octfE}#sj
ze}>TI^v5M`iD3VUdk!^KY{}a*SF_T+v4v^|oedh6v`G$ZJ*;B%1Z|B~Cf~*+rS^cG
z9{L=Zy_0V~4h<?OBWP+i1WJO*X|LCws|V&tY>mLF9H4XvUV1Jp3~lTx$QOvB!Vp%<
zp(RI+@~6XrbKPZ$;jj1NLuEp&l?RT@flpC7A3Ho+vLpMWvdW;x8qan?&HCBy>C!?!
zX#V?!-5a@k8K5J^z`M@C@1_T~AMd@N&*#0F-hp1&??tg6J8$e7zjh%zmZ8_NdA^UC
z&xtnxjKD)XOtCIc{=~M`w>QNWcwgF15yFoMC)X6J?91#e88A7zqUaiJ#^mbg=%SN7
z4rC$`!Lla)I6JJZ<Fl=ARO1YCH(iN%YK*VlO15&&+@~|EwA`b)SDC=rWeMF4a{`_4
zyxzEFz*RzNugc_Tlt+=6XWr=)^{Sm4jPWw=qQ(s^sb?Mde1kzP%W|8&ihaZBR@7?u
zZl~S><0HXN#Rb><b^U?jKbxRJ6!)<tbN4dzRjOS%35!a^rdJtBOQ$lTt5?~|JCGzr
z0NGaSr(@0P8{C>v+WkL|5Z3=-VzV-^{BLh5F&h70m|?DNe)pUbYaj?<Vsp}f^l6-U
zcVbo4EAyCNqJ9P%Pcxyh=<BadJ{CN=6E_7Oh<}V|5uTpgkhe?UVHI56Znd7zo9?}(
zO2edhljdGcvc|<WeeTtYXWQRw73robEE{*%&%c%9O)p(?S5;UQ9<Q%Qe!M^TM>kxV
zy}do(C*Z^%?^9(}dux0?Z@g+0XWC4)Fxb}^e@L(wsAL<fZGi@2sMR$vh(G7KH*q^8
zY!O`>4JFU}Z+s8<hl=3?_AyJcaj9WnhuAlSn6S+~+YwDztuJMkSeGc^*N|dehhEn=
zd<T_4vb}gklLJvixXZQTh3!kFyFHi)mprkvK6h{JTWiR?_^@TA#FPUeDTaRN{*WO@
zr>Qss$ou2Qm~K)i5ByXP$NlogvxYo7rsNYLC!mS!-+M?vfeH>Y{ctKFOQQ`c9Zw%S
zx)7=Bo-o;HV}q4aJyQ~V`YJR4(VEcO(u7EIp!5^Te>DsXd#MpdmWj+zC}j5-4=g#7
zMvWKh>K6!Lf5V_^o(32;l_SA)MyY+^co!ZD&^YNaRxc~)f1c&q>J&4BGT|e9+1H1W
zeZ`!4OEe}12-+!5ev(XKl6o5liabj3@hMrT=D@i242uDXJTyuGdOP<@%KnXU>cv(T
z9LYBi6X_5ck<8`V3<NkaBUBoZBv(Dzf93~gqmqV<jIUM}FaZ#k;?D5{p36#1fp>b!
z63?gNaqn<nI7|QX70>~YZ;W<tKUep#&Av}TKzr?l)@y!;IS&?VvG`$e1JmP1@?F^<
z*<)<~`L{xD29`IG%LKT41yQtbI8dQr+g?6Ck^HMJuQ3w6;1p(x3)68lcOch}oo2<Z
z7xco3|LzWcKuSUaj?TRrR=V5Fs$CvPUlFbFBRE_PbPniY8hv|Js|6H(Y)t^ORGE3_
z)yVt(z9iDxwADHqf-|Y~#=Ncp0k#qLQaXmCaDO5HcC=%E4m8rsyPOju9kwLHW!<MO
z=7Ss|s2m}&H?btmD{``-?)HhwqztQR5x43mzp-R*ZQ)@Ad}EZqogII<R&iV4DL-g@
z1x|P+(AjD81J&FlTuqDUV8oCTAuL~^Q2+wf31i65YBL5Xh9&t>(dgYdvgft61x&1^
z`Low>OXi0tG^Ik=g59n-VJvd+_OssFek0@4Xr%PUZuOVTtNuGxw)c;Th@Q|#qP2P;
z!>Gt%dSxJU6|!zz?39Zq_OxP72L*>zQ_elR`V-=ZNitOGEf7!VALu~$0|0E-q1nGi
zPUMp3!!r6?JNW|WX#(c;;9+@*O^Suv`%JZ(kCd}g&PwktxBf(T`1f5noia;o-#oGs
z^@t+q4$_8c8d`SNSnh`6{rl@=S0b%dWt&~ukF;Wqq(4Q{6U+dGXv?umOMq1j5S=!0
zTy#PyHlY^=64YOqz*gPyVc}IB36s`q(Mtsz*kz{pB+zlM_tF{o`C!*I;z`OVth=nF
z?x~O=)X<BcL(JQFAnM?W_^i-ahel%v(96p<UV|Z-;#lO8Jsi>Z-HNb(8c!orL?Rjm
zdAzQ_OjD532i@TBrTDmC+A*uNe<!4P57ENhMNAzKN9>jNiTCA`dJr|Fuyk~8&jD^-
zh2G@Sl^L;Wd_te)RRj_h<N>Ubiqh*oall5Awq+i!gKW6&(j<*t2U|K_IFC!@3KRdF
zI3<z^2m#MXc@r+g#l-2W3(9swWCzk5D1iejc)b4$(bfiyqnmNdJ;S_2FN63-kiDFL
znO$*&30-lt-M)@hxE?!Ky)cE*E-?9nAC$XYz@rd>&=!g(z__(@YO>$7w}p+Ff}C~n
zTa3U|<qtmwwzl%h$9uW&u8u9}J{whFS>Xn|`7F#EZL1I7S+rZ9Bx09wj5K%MjHiA9
z)@!GdC~qZSud1o!a^!Z8kM``ax{+lszYCRv7Uk36@<Xoj1I}u@bEWSQ7!Q#LSf?Z4
zs^P}zFhR}6vWa7$nT-nI9DqeQ7T_32A_#)WU|dqc1W;No8q39mF`y?5N^`@8CRa>r
z|B?=dgYb=|ECiAauFb<PdDvLB?PBE-1v7WZ*{iFuLr#4)4hEp)WlNpRxoiR^`A<A#
zbq)MKl)Ynfu5A-;8{0FsZQHhO+csxx+qP}nwv!p#$x5xN{b@f{``!0XxT@~1?z4{`
z!~c&+D)|G5#1>5a(2erUuB$o1>%lfC>oc$HBsZ$tq0}-Cvoj25OZwwa!kQ{@ONU*J
zntWiyp`0T}Gq}vr2n#+!M;sMoXpJ5x9z><&EE~7aAq+52)Jke-n;Jl@7+sdC;3ClU
zsWsW!E7mC_oDVm%Ory<=MH5WBnDJNr=Ia<UQa<p?i3CmFag;hjeX!&r*Mu#MTE2|j
z3V6mQFX8c{{m2gpQ!x-j`J6AiK-Y{Sn{u8Db&n3phrCgmr`jqpmN&X*F6Gm@7jXPX
zOSAIn_lC0!bUKuwqA|F%hM>5&-$Zs4;zfWLlP)9PJAPO;>N_e$emzB*=N;vOL_D7-
zLDQmdY%cl9F~>=W*b*i`4&T!2;k~F-Q=wc?1`HS%zz><;{5-B;bnw>>8fQqqSH)N|
zZG2$r7y+`fC$O&rzaqa_oL>sgnLzAi5NOn)@%i}mO{1<j4_z}5V^^U;KaV`Ly(ooG
zO&j(L550EPCr9&$L_VS9c<_9_JOOmG{zgy9#*T##f1`X&KynjS{H&#4?<0FtY?`2m
zbbMMuA-r(8i7WQuXr!=mc_U3oxjM!~ygBmCG0$7S?`}?5a$paV-yP7K_5)=JV=|>u
zJ{VCf1#?1UQ|&;u;2^t{pdi|uvl2@~x!4pZ*hL8=vvBcjAgybVw25@^oQt5c!8fD)
z0QXX?`ud=)b;2hjEwj!BKs!9y=NxQoQmV8tmD)bZK7pe+cfvpr=i14o$=)(uw?vy7
zHN)v$Gh6thHbMVY(=UY5HBTM|;UDANmWjplvTmlNzzcSN3QP%c<yh^`kmp_2w@Jb<
zLnnOO7Tj0&OPC@_GSL+Y=0W~vW<H*F&$p221HY=r!QufTS{*5|SqGv1w~_gKIgbu%
zonE=#Ea(MxJE1@zDaBIh56_)CNuK$fz9<M)d7ilPxSxaRkIkDm5&P*n4SFfroaZKX
z#+48j0J7(xfmUdP?f$Z0uAN>7nPde3p_Dn2VoFURf65+ty^Dj*{MF{_63Q@d(bVt1
zK7;S-x}>p&9vv@(hlO(#6~W*vFQ7Oy`1!2!>f4vVLU7a$NRq^34(F?+d79;5tJDiq
zUJ`L#PT3Q`CIV)obq>N3@rR3eY;rpjb>GVS88K`sDdh)?pQUT9y}#}ksPSt!4!3Ql
zdX7|28XmyXpX?Yh0$}4Ep6C^#TX;PK>g_b$!p*0__Oyhb>vjgy>>Y`(ZiSzP{*cd}
z3z9#(Gj+%i^1!|wNHs{F!DfE95`@(|L_&Aq9HE3p?Si?zu)Cq;0>y!Mp{$Dr0Sc7=
z+P8Kq+ZGDM+A$b&Wv~3>y2OJp(K%=iO<)@paQk8V_3wv^+8_VF8c{a-|7b)R*_i*=
zMs!+J!tHPbv1hh+57AT3`lSckJI!5GmeGiePUz#{4v>%ll0kwVggpB3LaAHd#EGT&
zG!lXtb6ICo?TIB=?1hN?{cd}g^rokC-`Y4XUO9%;jfhk%P%5|`UB<L<7ir*Z`6kiN
z*G;JIT>AgMWk5JI`o4aA_{s5ppIS<T%EEMDLWd{f@&4R{g8ke*#QJ`g`uV)RzC|WK
z!-YP9N5)#!8&$-yE<o!Y?cN`Er`BugA!Z|28tNPmHwgRM{dEr3rvg(zce0biW&HVf
z3V$=m8==AaLiemXI;I|S8y1zuj|5^*6Nli;tic>uCTY9KnEO5%@&`2amv-gYI3~%L
zQ@jZ7$3j&?-&OUgqx(Movd*Ygs=^9AIH}Pxu{Yb0!N(T)TFOX~;(JCl@`wd9Fe3(H
zpn0Pb(`FdPc=hsj+cmLL>6}g5K&cfJKB9egRH7b)ytp3sLW!tT3l_fui;Q|Ut|KhH
zvJS=oWKjHmqpcT>!76bqZPo>Kv4A-^R#erHjOS_^5MKe|rCloLOZ)_^QtwC_8n2<N
zdhzLM^mQXMUmpa5r+B?e`!H<zYf<eiQFjf}CKz$Y>H4vf$&ZsClGzszEHMgf+b4h_
zdbcu0#&aP>iCy0iOJC+NG%Se^l!lV>ix^`LTYPU{?Cgp;@1sE4>VtzN%)*`v9%0T8
zj)oAG{&TQuBI)Y#7J&iBB;#h{J}RzgcPG>{Qwsl3RS~jKBQH)&izVeZ3?$|vpv;oE
zP=+rqynrY~`1ephY@`7W13)ZHILf{aAq;`XKX@5MV#NcIdqpm5jTf@Mt_Y0l$_9v(
zc~(a~XphrjGzkmukUnjsKxnS!Q{7n{;#PG0&WN>o>si3w;ddE0(KUg`#P^;y!<G&T
zC~AK{KZqplNr@fzm*<!_v9EUTj5U*945sS3p}fLo1(3PzfieL%Z+#yyuYaydea11f
z>3j>GT|G!mYdmObr@<F^29tq%Z8ul^pl2h_H|p7h>mvS$6#T3srjL#Zc^~s$5!!<t
ze+4%xBm^0+>%);Bl0ms3n<URYoRhZ3$Pr@79O{`73eaCBDw>l@L4IWK=o;YBs84BI
zEkygJFP$Hkl1CoxQWEBgkJ#8+jhaPVYQyi;({9_)AJDWeQBFYxWgu{QrHwBSk4-!w
zKppFCWxhp~=88n(+M1H{bTh66jdnUohRF-fvZI@NfNFY*vwGy1GV=tLp=9vi2y(Er
z&bGF9bSK51Rv03PONf0T95_xFJz2^>S_Y)>f73VcOrbKU96XEDfqJ&k)Rciu<rY1f
zr0wC(rDQAEN(n~wdBTaCDtT05YGsrO146n1Jn~t$j{xHQ(?I}EJFbY|^KcxR0tfoA
zX1Rx*1BsJIJzkyhF_!o()mO<~2w>U|CU%`Mn=Cv(Iv8y)I(2Zj#&mSr-2!PszJvBi
zad)_H!5Q)#W~t|+aFqTc785(Wf`59H57;}8MD+5H><l*{h%-rHYd!QdT?nY1oDnk@
z@t<WZ`$j?r8Y|1V>oqL7wgPU4SlsTtlo^mS0>C|@n<EMt8}#j=oF@iJRLxQvji~bH
zZWgWE*;EM$e<Iqbwwo5t6s6!Ux9i|$PRM(uTMs<Ol3(g?uQ9tE1dW_`!K^vLM>|CG
zf0rK9?fduN0}dyWpJO`Z{qfdcTZCaAqlP5t?<DO&zd#CglAJ36?bz<`FxF~u9m2Dx
z{#%#i6Z7;K2MRy*%7&js7*q^CgxRW~lN#G3S4t=_0GZG<-lbP4P>{BDB9{_t{Xo`M
z`L_7XvABktK)=Cf^WLCAoe`*NT_GS6w3JyBpz}95Lny=jM?(FS9SU2}MyN^%_|;dJ
z2RYMg^0ZC1NBMj~ZyL%C0#t3X;2w<?Ag6KSZ5B2uGnq#Knr@AUL#^Fsee~Rr1)Qel
z%LbS9M1U5(Y|8G`prLZ(7W$QUdkXND2qYXS6&;6k99;^PSM0_~4Dw36cZf+8SbxoI
zpT#oVIGK)Zxa1L@Q`PyUb=Fr`qjLa`Q*Fr*Xmz=(QHUA=W#YzOA5o#!0raawYXq)%
zpH1(5x4`puZR9e*q!|#96tcF{NKkG-BzjQ5PwC<&g+fvtA6vW*A-)O@Zza$u6%`;P
za3vP>xjk#xvvkxsyL7Li!QiUi6tWBm&Bt2n%0C2^UG}^_^a?>Oh_Qo45N=`))o8gx
z<G6R82*oXK>n0LBtZxkRBMj#qVhx@TUr^owbp@&9oh7Dq8Y!7j@*P0;f+N*sPgScd
z4MV+`Zb(xv%c)^?b&M4$RrUPU{~vTbt<;R;v57NgQdf<voI<uWciiE&^SCiioXLWz
zrVXkWvw#36J;vT(*Z1|KPg5<}AY*9;(#Y&b`81Nb5RBTy3$Zw&-@p8P;SW#@6o!2M
zye5BD86t{v)KxnzSM&1BnS&1qnz+7=3`j0+S$!qf7wIlh*XyI^gEl9>JIwH|<r~F?
z^bm`s*%aLKoLs))FBoT8+;dYH9^__QuuwS8dq4qejUigT%EM3VfBqfR#yVjZ+UVQA
zYi6!L9F+OzkZ)DGbh%zBt1VQaVf)?FFKB^aLg7JTaQ}`T0`tTB*q^s-tVhobhzu6k
z7WQ3cC&A=@tbeI~|8#G}rSW)?4#oOQxZz7$WfeUuud4bFqA<O0g~xpfFMg~*>d|GK
zU-XEuF*H4qgt34a;eHw%GIvfMx;&rd0#1G_PBC7(2`X{QH*|rSgKL{LBugbE3l1Y8
zTM7_Wx88;w`SEwGaGw*%tDe-UKxrS=FtJ}_+EHqRgs-&pz0~MYn&lrmKtP*vd^q`|
zof!bAze|G={cZsSz=v-Ax3}wexz@1AhiD;#sph!QHAD=}g%*Afv7)u;-Irx&CSoEt
z9#ta{XX^N6=pY4*9}=*`CuydwSPMt5ai(^t3kNKMo>Y?Td}AiZ?NjEXqdE?06OF6+
z_gsy4bZfhpg=9DabayxY5JDNk(cQfq7T@Vj-(Tz=FMe=;p@ROiXZjED90NPU|Btk!
zmRqJM!mklM&+cqU^9E(DWroA{vZ#j6LMOY6+_w)YAsmq$V$;QOV|OP&2La!Bk})PR
zs|%zH)713zH)#hR7#$uE&6d6Uo_)0nA|*s<p)h$C3u}rIs0HC<@F+IYmn5ENE2qB3
zy>Yd%b$OHf*vUg#*Y0NTdHe6B(C6#)dVYRWP4Z~(O5&yJXY|=>W+J1VUtL|FT}c{U
zDuh%0!w?~uV#^@NSQY^skv94p+4u*hhV&0#mOn{m#&}5T4!zUSY~mVUujD6rRwYTV
z)Rg>o$G-SBOyTWDWa98cyymKB8HVtmW=6b@0Nu=C9mCA|X1@Cq%V8D9L-HJ5P}_f8
z?j5&i&_{hN4{YuBe^8S1u_`j1<aEBysc%bk)7*)g>F@g!j64U4+m;F6<9q+GZB1Ot
z|7|nE;*msbwn6D#I;S2B4pU4D4<RS;h%=K6MzTbRN(1iokK#`{M!Ko*(}d6>;lZZ9
z(>^l-&cf!{ngMN5Ek|e{Vt)xv=5dHSL{{Zxz5*|xgr3=Ael;Kxy?!Z_-)AjD*o$ow
z3&{(~J*5h?1k^!OAi;3t*~cHj_KzW&B>lIVJuGTNY+c|F3`z|R$DSFEW5r;K6-+R$
zbTld&i!UKSkl)8vF)d)ra(+2KRh|<>Efpz?HnWM~ojxwm!_|JlCE}0j>95p;fFF%`
zv^C*v)+H)OIT;YF1EpHF@nGcVj_597B#dF+{7NcZEec2oKO?x+cGQfoJVuTP+><iI
z%e-ERu%IIV8c?Zxra(Q7CNrZDMU6n)@4E)W1sdRD5Xz&UGXMubH9twZhJh4g4en+U
ze>&u|h5WY(zcd~E2(_79*}fGYe}Fuk<-k^eYoiK89)<$C!!wF#bKb#{7&xFHqW1t(
zf3>X+*=~*Gk$j(?#C5?sEWTg^4$}z;sh-}Yw$Y4AP&f`2<h<A_X)BW+u4aVXnsE{_
zF;)s4di;bDlBX(xrkrT#uao*U=Sf>UqdhT#{C1zCY_%$s_y*$S-D^{UKg9i<yE62k
z<0*PaI9qtf5<|Xtn}AV?Lkv67BuRm%Ygtm5Ty(o#5UvK7|CYiH-iDre^Nf1mMuNI7
zAtt5bIm`>L(T2Jr17RdOx5AQb1}lshZJx;BU#c}|;#64`mg_*_{up~uUeq;rdAXiS
zv?OF7OE|=V7VD9+gnUZzV~{Ti2?@o7e+>WnGyq_r>(R^M?h*TBz|mY*Vn_aYHnncb
zv*h(ocrW=c0MsyIO=K<|j$B)#ZdKfgBJiw0Sjui&H~9BMzQ%nq-y8mD!r}_YV!W|~
zLI-fkEiGV|AYT2tB@{Jo#$>dcK@Qbc8EXHICubNqQaunvnM}aRTrHWBkP=TZG3O?K
zwr_ZiQ?dfj;+(r21I*y`(FcMGI2W>2oe;aQO>Tve)^kmh91_-uM`wy}C_@zbMw5+H
z<jU?^zb^UqKGZ8+W)i^kIs6SHiQaEYq-BC`7;=UAYvLNsW3=^Vr9}2boV^07UdoTb
z?pz*9Nmc53Um83JLP`^PKAs!k#=0UF3v+TGoO6;y&BMRH*JK`=>*9XX3TpuoN<rRf
zL1!pTmVrmQ`~nF4u-dG$^70#qF0zLmI~ayvn}}(N1SEWdZ(5nMH92N>N?7UkUBA*m
z&kXG~r#YvBA5tu;TT#?d*m-Dp0XZN<D|~ovSwPp7S(ZVX{@m{}bnC`fVS}aT&rOv+
z?CX)k-r>Q4V^iR^#QAH-jWY&?9+M0ASAqF6V8D|+t0_hhiRAp8P!8PXG}P&AT<E4;
z-{EkryV}md-m%3O8q=IlD|j^!b>{Lkz^#%*QFq!D{^32FbY=Qv2+95=PBvQ&nlXeX
zPFyk=^h=<iHc@z#E>duZHJZP;gJVr;)tc~?5v2342WiCSAy|-sI=fylmV3D$4BV`8
z)0V#LzAap0Q^TyWln#h&(?3_#kLJ8)x0bIS1dmcgj+cZLKn>$jM%>Bz77D_GmBOEG
z_KLO&^OFJg8xmda-bH8v#QJQR5`}Ag*({fg!MmW$LGF*8kJ*G0?#zTpv4GS@Zx_zu
z-7%}=+OqIPI!|scYZ`ZsJ2N!ji+nAFUINGBHeFs~67@+6hG<b;`%+-)uUsIkuFP_*
z>+=;|2ELwBeC1f+bFY*05x6ro5%%0s%cSmJ*OhE3l=}&?aK=_&Ue#SIr7*tAS+^_#
zZ9k3V;w$|Mk5@fe^e+{rZHE<7C7oCk@>-_HHz>wU<nKv4u<ya<z0d)swl3h$CL%wx
zHM^|$eYT)43D^&pSa?_vi%Q^;9IT}!<BLCToHQ}8vjG1NkuqAN&I5_9$eNUBEHt<y
zlXetR@iU6NRGM+&WP)z3rPzKNLsrmG_Ny4p%dy<|s<gpypA<@6w{P6O!Rz&9|I?iM
zi_@aVw==Ya;^u~;`+vd~?Em{|!nC@E({I?q_oa4^h$UJdXm4E_*O9}*>LqNsr!5^u
zS^$ZZ8yUAHOZvwPO<kVQLd(J0Yy%BY8!>XYbr(19lp1H&!|B%0)!zNciHtyEOwg>w
z-u(>3SDc&A-qh)Is(356EBDLWUAVjG1v8rHz>9&L!LHl;<MU<x!Yz>knwyOv(E`Eq
zdh(D@PT+O1)=u8{ezS6Qv{Z95!0hEk9H}1Pe#EHR+4kdN+efWC2gjQ;ude!3KGf+z
z;omD_sF6L{G<;K5C}0=fZ!DJZge%z@72I!Yb&Pi?iBuM5FGM>CO_&%!fZqpc1d;F+
z{y>l%oP;_c*Z&X>7?PZq+i>Z|;vtK|Bsge_+)}tXY1$amGv{Nl5pZMma*tTDg?DP=
zIJ9G$Z*2PAj_x*va28bhOvy?EArMkKcfQbkoj)+;&1aH)Em_#0wSS*LGr2sLQK_;a
zi*MU%CrK0l`E>9BBaa`17V3)6`**PHVu=8g<X&MQ#m_gDMnHf%qz=W&T}`BjMN&gy
zXQc{HL^4H@b!xWQuoN^}{Z37FN?(>ERmyGHhT`{wI+L3mM#h#$Z7{F87ttz;ro2L(
zklfUqxhg3ZFrK`eyYn%{wz96ma;RZ^(lA8GfPE@5chV?VHhv|iT&{9lFH6$zB5Mi_
zI#GuqHfG(LtVl>VZ-NxHXji3>oj#Wz<Rup9gr_w&KeSG8w}l+N15zvc^RL82i}|&S
zoGPoRt~>yd6^z|73G<9+7YH)2y&_XGyA_o&E(!W0qNM@XB0H4=08l?jh+Sq2EnRsf
z&M>C$nHnA^N0>N9sOdwUx=93DgGJ4g!_DQ0em2s1&m0mM7>l0Eand6bGc`bZZxRil
zgZ?xa+=wJ<v^K{LTEV_u@f@nuUHtVX8ECbd(PFiR%4DW+8M#w!bOoF^kpI|$KX8n&
zHwA3b0vwpywCbTWCuR>M+yFo7MA<PLk{eZ=Hb9X55BrhL3c1sSXaR^h92D2Fh{iD?
z)?%uVqG>>}ouA)A_84@i@<9N+a3pm)%;47;icKSBB_`e+q$2Zkz1O&)*-mRM6wlC}
zWULL$pg>CpP|->O5&|KB#!;tmBk@Y7&OQ}FysM7^6)EgHU|ltL5`Q>Gns*m0y0|+c
z%gx}pAl^XC5GIsTgdJQZ4mz*7Dbcc0RJuuK02c2gO|b=@{TLbnGFl^lpqJwzoe%1?
zBu@!R0!y--JV4cwL&pG_Nq%Ck=B=A)n6W;snEk0?;XUgrJOL0aExA=6@>X6|!VHiN
z<<VK28AWK|;GZVva$4j3RQjbMg0#&CY1=<#jN`E1qGV$E4wxXLTGQ`C$!!$@K~<OB
zsRc?JQwmL7oZkU)b=T6~WA_ZH1U8;iC6PF9GC-pDo0nG5!M<gyj4ByEI-V>;!49;T
z0=EA~M9<&EZ90mHb}%*g;(dj{ZoGg7V}=Rt^5y=8;yB4=!U--JmrR72gJ-ygGwcf{
z+e9a5ZZMYODH@<<O;*?7kYau<P|mETPtokJ$@1Ycp2K-$OH$uUsEu|mHX2(oNzc~W
z3ro2aC=DZtw82VyhOdoP6JBe~tP9WCS#i4Z#{|R-;Gq*XJV{=T<_IVug9fR9ZS|;R
zked%{mtf3cLFm_mrb{dm<uo8TiWxeMw20Ns=TWFnaos-_0OrUrKYm0Q%&aDCTG$;b
zk}ZNa*~YqJ<v{}THs0Cb2m5(^vbP3wGQXaT&|ls)hdx$MfwkS;^sJ<Z5X&ai+<iOV
z!1+oF(bHi(uj}B+LG(i|$jmyIEF886gGn7|FIHP$MT0>0=<s$y?;Kw<ai5-DI{{LF
z1ld-i6q)sQjp95!1G-^=rILL8AOK_CigpeS{Nw@4bozju2IhSLpwug7YR$S<q_n^0
zj9!3#e@kJ#gy({=X>auGE>k;~$%by747vf;>dgrY>JA{Ic0NdI8v?Nzh9_#aGIad4
zzV%R}dMUQ8>5Us2)DbWtdfxR$H%`LGuGUI>S3GI!?K99)bM8c0ko6|JV8)J{F|u5I
zqcIhDy5sWuJ?F~<D+DU_X}tV?)O0NEc9s=$18szv{9FE^x};oCM!<En@V&gULFHb~
z#tNIJlo|<^AzFoIpGs-N(@>p2@ss%@_BN-(oW@gu`%C6u-2d54frk<mo$|HbSSZ{1
z@gXN*ju=%rx<F-DGPm)@gy*5CkksBz>b<Udo+2i^c(BU&Zb-N9R)il50u*l9aZdxM
zo3ymG!m-CTJNYv}F5&AP45Uxf2JA|<8?~(lMQj`)7FrWJ1G_doS>x$jToFtVPU!mp
z-W%%2bahHRnBNIHR%Yod6};CvzxUDh)`D>UX)U?DI4WqzV7Sz5nv^lSjY|9#Lf!ka
z(t>w2vC_7G&prA*l+p}54tvL{EIu0d<sPVa&upnzC;sM+gN`H)?3tqFp1ZFvmC8&1
zz4?>zokvNJnUDSjoiV`J{a>>%`+sF2nVJ4)2IjVgmeL_BvhQr|ZQ-JAFtJuolC7CO
zG);(3N!&T~<Yk&L;Ou)sGY<JxQ*ZYur=uZPI6<oGM$g_K4bIdXZJII%JNd)Vox3_W
z`-w8U*CDqyBN(#3$LIsD5X~VB*jQ{Oav<mM!9vrKg*{f8gNg8i-~P~+zJ9JQ&)1`o
zf4&~h4i4^dY&kkV=X!Czo$W2PTxMGEmD%yt8B9~i*`CT)LXdWk1<|j}T<xm7T$-=S
z6Af;pVguAI=?5DiOs4AEWzw!N3ft6bv|{|fa`uA&YwU~{(jI+xRL8-#rr5XvhU+Ee
z0W8rF%vc?`(E{EIgNM=SHG44VWvNY9cl_a|fUdVf`33~%qfz4A?I-r<1Yv9#BNpqU
zFwa__8rb*sTb`G1w4ibAMM3OE?@${ui!@5@eMF&u#6hE5RsA{YOlGL0H5tVES6@es
z#_H-Fo2#@F39N2|LhN>edaC&B$wk(p0$>46J!Ig=2;0K#95!=b!^{*AeI*yD;RsD0
zE@6Ax;BF4{jy&7KVD@X_q8WR`R7CNXpP5w=@H>f^;o#)S34|+)4MpMwVrID@ag)UP
zaEipG8xgiD#w0|L;q^}g^-^a$4}4OytXA3RQ=JafHHXfrBkFu{z-Jl79Ft5nay?3-
z0tI6ns>Mt9A;vwPo1n18Un(*V&{Jk1l@*Q9gVGMo*d;?SqDBoIYc0IZx7eEzTODET
zg$Ppw3f#zyh(*th;z9%%6A*B-ARjvVcv{kDOBR`_9Jkm77X1}VLQqP7MVrj<?qy+=
zD6Li*UU3SjV$%v_3g^U^C}(TX&L9QA^+pV%*u23UH3#8zjR+9CNgmBFi2L%NA5%%<
zvB%_44S5iY!Z*@nlYx@fNobmonvFva;qY|HBzIMtW~_Doc+pH@oG(^oZsErA*wmRU
z9%YYLsMC2Y8@5u<!UicLWdH;4L+s>A6-Czt&W{-4l~FS*H57ellY*vi3p?#HO2^Rl
z0TB{75Itr6WFQV@{4V}3<q`TPN)SCj5k{9MZJ4C5OAZ5;0e#a%+8LuVujGYh+KGvd
zcSObB$P-2UBV~mLP^{8-+K5{s8E(nU{Qm%sCG9PQw}0e4+h^zeGoTrwg9xGhbdF_X
z5wF)&!u3x1t`kM@!h+0>JWJb!90R^O+(A0(q!-8~a~&I`bT&8ph|nGS6IQwgc?9s;
zpP6pEO7P0_%hnSlXC2TgPbYGMIr20SN1hzr7P=p7RnK@q1u|9n0?ps+O6aZ(Ix3EJ
z)kgpD8q~DDg@FQxh*_^-lH(qinMbSCc{=%O!J>s~$w8R71a4?qXnF{x8V>8E9>6MC
z%K>Q08mUruPM3;Q)G%K&ro8m;Tv3r!Y9^CzOqkF->zkGS^lW=xH)|h+c{mF!$J0%1
z!QFj&aClW7_i*D5nr^JTceb0|H@Mo~V4w?Q@BSWJ;DN?Z{gCp3Xn+-W;aJ?*>E-6I
zNsgo4BWA4c0JBQo3>s3fjT-1Sjr>wTkFWTk4&FxaPYk%Dj~owv3KX(<gT2J=D5giO
z;w$&x-$G_F_WiZJcJXdIC{Z=yj`CEfn`>$>Iij?NXc#P5v#0y9=;cS<hN1CdA7*<i
zIKbV`3%80NSg$NxbkXm0|DmjQsuVSM8!~8JFHvf?l2J8Y@cH=|*7ovn==pdpbX>{H
zc7;Dhscjc_RX=6D9Z#s?L|WRG)q4zoP0XZ!__^wqU%ing4(f-CoqwxBSf_GS4wyP!
zCT7>QuJZ5t6qk8_KatYc2s`LddgzoVb#W#q)F?|jQ_ib2BrIEOsvgySV0I;!tn_|d
zce%jp)@QKL`lG4!a>?x4v9ebmMX2mI3Pm2>8cYgziX8A`f0K^dRoq!JrXM@v{eV=+
zw&sux?&yt`;(PpOz5lQCDh8JSx!!lHZQ35PBKW+h-70{sz=Wbn2+srkf<oOgT{oj6
zd7UE#>rJ%6X>}!JGeBE*ck;z25)O&ikk2AX6rB#!(+3%dtOmv|#@!yq$J>#Gm2mFk
zrqqOkg~#I<%g4MWz9IO77x_-@<YofS1JH*Mlc`0CvX{=`cfT3=y!f&mhJj>CnNS+`
z9uHcjw!9x6tU7-dJ8x!bX``DVoxU-a5!E8Fc`sNzE*C#iEwiV|@W}jaF#ricOaO#M
zfno{93ewVi5Gj*l!ws1a5|$`5WPE)r;)KEgnZpsg<k4>tb|DPFBnH_V0D7nvTo|)M
zqCkJv#C)B;RtqXF;H4QNF-*`4zry=H@vicenuaK_1fsOefqEQSmBR)GgeS71=ixTk
zIZhSf3P5Wp{O`+oIqL70&plJgT%IYKFIpOAb<&%kr-gG*Bz_@_<OM?PraQnr@_-BP
z77Z>8qQ>!;sF@#h!-s&FwLScu50&3&+-wp(8QHF4O<0rLfDMrq)*GDT6VG{}IkEeQ
zQQdZu<v#6||4fYZGK<zu&KGErTGO>vTF-S)&#M@fj#qN)zInOkzNPqZZc!wMCfJ$_
z@<&Qkp<<zob@L|QNSi;0UD%K4N(+cpG-Fk0eZ9RjWjwH#npJ$3^1-6$Fs<mAq!nG8
zFD7t3<UrgsI_vGGN!Q=)*Xr&0Lz70mtr_kx5e1}0te@4wn8n<65Pcq`^RSUsz)d?x
z(4`z?SrXpGt&JRaL3=lCD?6H@y;ygiUXc&_UOZmv#SsET{yj<$nlcyb>^_kVdx=WQ
z65(jEP5USmiBA8DYAeP!rj{}^Xcpn49Eed{pTj&`&W#(4b&@4)GyYU+06_0FMX%D$
zZ*}V!Pmh&RI5c*3w4y`d!2SWw5LNGC3jQk=Q$=T6GYUEo0Ix*k&$xU#@&}bQ3_T;o
zT;4N(x0`^08HB(6I{n~X{g@o{Vn@Z18C#lXXZyo^Pn{W2_iYaPP9CS6JemdBBi&>J
z_vvpdCfcV-=;<Ml8n-Wc3x}_;3djyzLgi(d4NL%k`I`Qs#C}%;+pxM{TQ`YyS2q{?
z@BQtwbRjyKpb=VzMERL){y-l}$q>8sB6R|{3r(YE!|&KV?rzQK0fJr=68m^zMhF_i
zjE%E)-17u(2z8+yJ*>Olw;9dVuamKL^z!Vj(Oyx4u(WnjBV$rm-B{65P<hl+fdj}+
z?p+x{CVptQk+_hyC=!*Je#YUms{ZbfkAX|svRrfNi4o{*GF7`EV%1CdD9h~>;ZxRC
z%4h~Xw?Z?+BGGQ&Vq-j7(MAW3z|1<1CF=18LFR=MLqy)NCvJ8e^bcP3(h&sxFkR!l
zDC!17dG8fd@F=3#NQ@$=OY;zCiJ@FsK*J!$@uQflyJ<K)H-lTtPO~#)aF~$Bq01hF
zK{2g}+`jB!gQ-dlJui5~a{;fO`x#Q#j1Yv<e(L&QE75Hgn+L1aLdRH$^Ew9(qGc8x
z`jXp&-JXBQA?Ml0V01JTV~+sIboTEd;t1be?m({$9-8NW)in-=|MxQ!BR$LioP&u*
zA5XY!x%%!Yu$xz&p3`o^)k`_8z*ej=*5+8_s9V0Y9Z4u#U_88A1t;cbSza_bB}M`P
zMBo~NPX*@l{t4Oj`T2Z5?zD?$4~vjcI0{{T5J(Q~vQ?i~4qaW`FO{}`e#~9%wNr!R
z>xRggy!7gxX<Qy$jHGsd9z4=ynl*iH+x3}T&Nq;ukdljUEcsjci4SyFU~B?A<b|Ju
z5bwb%D_1a1m=CZ3Xl%dOVNgH#Pn9zZZB_TFYi;F5rA@E&MWrNHSD2MM{4!WA#j0!T
z0$!K{$%TvBR1nFOf$S$PNMF0C;gmxuS#Z6XEU9ow<z(k_91ab{;)kMuTg;~^TZi}d
z()iS*VbrDGmY?~>l$4vRR_J1%$;(rLd$#Ah*I8UOJ*<NGP$Z~;5Ingu8%kvYF*Zn%
zdy8rqc&EyPt<-ES$>pzMRjSp?9kF{NqWT{iJ5fJy@eUXD$9f_|*Vl9>j<sQKBxUBR
z2`S_z7acQA<>(2IlbP4-M|($}n%zg^fR=P%ujKarz5aEXHu{ck(XE*rdy7y99sXJo
z9kW`s4o8NdbY@DH<5q1p1(Eo%2`aD%5sgz9?Cm2S&`X8lj3f`(jL+e@1uhPE{5KAO
zk@1djI5D=G?3_4JFPE0+)d44NW|3Cl0dafed(u2hDAGQoApfro8dh^F2J~XShd_8+
z_w!5rM-Z+6d*^p(UwnKV5f|sKnMbEFAl{Tr(t9d($aWzWl1dU#=ea6s{2iDL?qHD1
zSy4JFMt*;_=AU<Uisy6TqWCBPsomp>)HXuFAD7>ql{20eB-%hqzqEc->J>}){X`Z@
zV8LeMtCD_27h1J5PCHeSrb(t==%B$>EJ4yf2nLlimv#(3d7W-pMYdbfbv5YG)>2UT
zpv8N`G?I&tS-2qMJo7x#7{6`JnFlR|Ch7J&0Eukm;LSYE>gFIss~g!OVmZ@B@#yyV
zQ^$(}U+HY>9#>=fj$l;ZbrBjP+t35qt_p1BMW8pDv8>Nr)+MSAXb7h#bHtuf^zMsi
z-m?CV17iBUzJCa|M{aNuvizOn(bwUe#;i`cK6WwjkS(7p$OhJgP(ePosU>-CUId|6
z??zlK^2U}?wC4hka{Bo9iOXSZ3|_oZlPw{xU0Rfa4Sh88I+n5I6r#(V89)nd7;6*Q
zOE;K<ug^&Ih#Zntom2%a5Vg-u@S}AM9+#j%%Irpx4CHXGqUThk`g~xPZ~R1g+r}Ke
z&}nlX46MEP^^NxSDt#<lO#W>j(&xJp3Y|L1u)qoj74*bWnK<QXzf=C1;GSSzO@F?D
z6x0lbnjmfAziLG6=>#&<@x$ta8ETGn(A6Wn58VN4NXl})h6HH}M-WL1?uO@f<~rcH
znC%p>HyVo$7mPa_K(B%&zK(zn^w>+oYiFkVjfV6*{fIrSl<r7N4aeJP!y9PlE9P^z
z3T6w#{nk<v7k{2G=yYgRil_3_tzG83bOjl)0^2+Vj`YQE*dI;8@V(2P=hw9xSZ3Kh
zg!-I-FxPkBuU>8>4OSRUTuXv+e#cr2@l0hAd64vKCJm7t%60CH6~UW(t*IKX!rgi0
zog*E0z=MyGR;L>|nd^@>`7m-auK*P!TI>N(Fpz$Vf>GOUZSfq!2_<>`sW==5`0M>^
zdBr~{>gwL5(6o4!D*H-au~g2Q14J3YARMZjlP-WqZaC<<xW@!JjTP|<tKaK!nss-z
z936lFE9-NJDajizmT|`{$Mza*nK1UKl0?k&Xq*vlqHG`9?KX)WPoi0u_N)+KL^dws
z#1pijFh<eF9wG|g$r}=_PqKAPX1#ZTO)J4C;QmK#bJ(_Um4|WlE{QA{gRRAUdoSC6
z3*cXvf$FkCBA_~}^M)}2e4Q_HkZd9|m=sRx<8LSyAI<UX)AlNtM(3xT9?aqcdStAR
z6vg5#LOgq#;n5x9E^UY<yNC5^;7Wq@cy(tujSpsH2qq$iWr3eEmO9Fs{#ptYii-je
zJ9^X!k#nOi6toXM$q1X(&Sc*LnAqNj<s-$elgPkuC?HqKFRA)DlIHT<Waz+QJHG%f
z+SBJC5E|u%qB}Bx0{h8gzIy7~1JFF{c=dSQ*uV{Ey+_r70;Aa*AmgzKR<qq=E0mq#
zJYkEv=;^r8a@skPvL<j?CFKnucDv!F%>`=niKcB7K9O^u3_nvKCGUHp6&fIPDplV2
zE}(Z0&j6DlAvsGJbihg6zzLn(6jYcOKtWMee(>TsRDL~%wAEKqEeA<ucdEZX)Ek{j
zba7|sfS$s55@zVM=E5w`j4YRSoG%X0JXr^wGU4PKN?F2Ck<)&YeifCm-jpfckQ*4^
z5C&>(R;J%{{_QYyhzXzJQ4-zH#Clu6GEI~ha(||Q1C?li3E56fn#SR#5EA=Ya)OU#
z)B2Qu`+iBRWzjKn;oBuO8}~RjI}rmGUhe}DMfw@s7cXkDVChAH<raU)jL_MwR#SL&
zg1pY2+7gVrk#beHr4h(U*CUu8c5u4swudV5*n<-pR+bJ(>m@&%r149ro_Q@Mr@hf!
zXS6VxwEnvlWDD$mG2eot5H3QR&m_dM)_9w^QOHV1hh9tm=aWVHprZe%^4}#UG7vz8
zAHc%Jfh_CN7*h@?-Q#YWYt#=3dX~B^nw@z^X81eIJKS$@NS7#R$h%_wt7`3wA{vgy
zV;U-L7bHp~a079ykL#$n+xHF6d+mwyKl_URIP_v-W?=kZ^R>%bT5g9eh`;MA{oSKQ
zlt3zpA~GJcOzTDF(oN?pUJ{Qyk_|(`@CfnhtM7N;wp*Zxf}`mZS4Y-haeewUj45v>
ze~U+i`LoH&-fdl<dp8O~8Rx0Pvu9o;hazL?GUfRz#&_2d;a}n46(8RlS7FC$baB#<
zU|`tQ(;J_zpVP<5_EQ!Wr%{sjtctXmUeEW<^t4*-ZpZ3w*T>7LovrQ87P_a}k`sN}
zuYuz(M|8?@Kyjcb#+EIuaN{_VPz&v6G8LcvxQg=TV$|AiIzb}_0~DpP*78AOI%2{_
z1C8(G!fdrhE&b=+#I>w~aI@ouOa8ZWB&@CIa>RjCsrRwhd-xT0beh(8&u&|9id^e%
z^JEUgw3kE_xog&-9h<t9(QpI+=GIQQ=zhIg`<82X-0ZAWZgs0C-8AyBYaLw=l2*P9
zCtC5LK21~xmiLT(PH`g=%MPsXb^vCPA~q2!j$A^>Hbr4=e=R?86ob{U#5?k*DoDnv
z&VcSiLUWm_w)wiaXmk`SU);}kchJ4S;$X4`BTJA+7NH6|T%TP&%L~<yt%zQGws2j#
z6|1R14Y#$jxwJI@%`<DQ>`Gtv9)IU!U4LGOopRNElLIBzN!pxvDwBE%9zULZGdT5Z
zH8nyn5iSMxRT5>F!SdBAWDH7g1?C#3dJvtu)=7>2_->;_rCM!T(Z_VtP<h>d3B4MK
za_Pn1^;9*DRM~m-C$<Ti4zZ$+l<BFahfvI{!$SO1V?unG^!tIHb49HQ3F>X5Oa73%
zy29~3RJih0UEWf&cDlfUeOgr#aYx5!UvO2JC!f}G!`}0IKj=7=P5SWss6A`fe5uKi
zXkmRZvZX1}-Ws3aVX&i26OM$bMqsU+Le6t$4DyHU?bIqTq9glov6*haz@<5~+9ieI
zeS8|k;)%j1dV!%L1_6(=?AO~9b~kIvoLi?ar(1C(=kNR3!CiFJ!R0;scy$#mVX7=(
zqUf@daI=>ly*AvS)g2X7_|P4ics44(^^A4ODW$}d5}P=*#k3Mh-PnOSn=yPwzHAJm
zn(E0+BYAyX#QwXaBcL>B=_QP+FrN{xP^=xaP(?B)Kq1&yo7rUMp@%b|IRgYX%IE~o
zjNdn1z0EGiB=EHK;0)Ns(aApF)ys+6zVD0I(LL-+KjX5KF<f(9jqKC~3cMj*ecicU
zb~9Qm7S0O@Gw{vO)su!!gp^s5Z||(K|HaohEdC{LI?(Van?28dNSez($dx1K^$kC0
z7Y~y@PCD7ggEU;0VxCPV9*<O>O(spMpGU!^;`=5^_ES{;g%pSdB1pT(f|}4i+>jq@
z#fBfO`{XQxWBGf7pNjsq^_S!qJ{EV4i8?nvDO3Y)pQ;?-!;y5V-eNu=P1!MV8}2B}
z-BvqXni`$@a)w!H|7hm1zDD>7ia}E^^53S%VEpzT*_Mz%msg_)yLQt^GmrL}XXAtr
z>GOk_(e7OsMXzdFV>7r%PGs<Auxs*U<wfH)WO4j)cHXAICZ38Rat55-445r+H9Ig{
znEQ@_`sa1tvz_0|K(adt-e{FG<IVp^aH^%)r``e|(yWUQz5xoBY*)7E41E`gy-|jT
z^)Ps5F)mJ*d<GB}v}ulr>BPB6M$S7|jm%v}tJ<4~Q%<l9DfN~r_7yhCc*~-)2oLjU
zcNM|k!;FwGnvp16R=v)$uS985F8qj4{Tl$3BCSe3$^LkL&?s`U6d}N2SAzYPO~uJh
zL0qy;pTN$vF&w1FoU55~WUqU_Ehyiv)Xa#RARKi+?d`)`0qgASL|axiV?a0=1~Q9P
z=*{3BW(q-OxjjIE3*<FyUFyo7Qz|4c36Pm?6}n?{$V{4UR7xuNx~;%FiT=WFhzawg
zZ!H`h-Af*<4@esWz<@)PpUL2@?M&Y=6_wZD&hd{$h<In0!-QyDT_s=uLkS1*7VX6%
z!%@I3#Vf`dWCoZEsTe)@-LyN|pMU-`$8AYmk__=W2(O0PI^tu69(FB!(G1hbAQUqW
z(@zkaWUVyvC{n`Yiz0=Iom)JOS5DnPWr7=#SM;4+OV{0%VEKq?{UH;=tg{G{pknuf
zhx)zXTR_B)tmV?L!;gyfLuO6owDTotD<!0g9xfwF|Cm+Eprdbz>Z%vUvtx!ZXnSWL
zPv(>*xodQAQF~$%-d;>VA76QM+1fgvcQY;XEIKy=FpMKLV4bm<U=!giN=oYr3A>~)
zG8!Phi6linLd$G=A3!%{fwIQ>K-QF9(9UEpbfZBcP>|p1Z%s;9qXAq)U0U2&+P+Ea
zpkFfk&>}ZF(nX)!pN|W;^v=1IS(&bqbG>b3+>Xmzu0w0B_LTW->Tv|!YN8A6qYQW7
z2_jWmn1YZv9<W*TGN2T{n}kT~s$n6Xw?3W6QoN+Q&}l7%y~a2(wq~-dlxE()_QqW8
zj0>NM6<PX*2_S>^wIVlT6ZEqsa){?;6|?(XGfGCu3Act2>P$)BR(AeI#uI|H&cC6b
zhw{Ye&&J^t7ux2`j;NZ<38gX^7d{~nX&a$3mI1*GmFAy%)S_L^HD*7S3T;4Gy33Da
za!7a%(4&8TabHTEt5Y+FD*(NuHJysD&~>$Uzn@ot>hdEavBv@5P@#xLPA|9a{(2K*
zOH%OTjQKWdu!Q6`rc7|QwZF`c)+&mN`-q{37@4<oG_K@SGub5)>qd|$H~FFa;BS+m
zP=sl~=tI$qOcK@>m|DX;;Jy{-A!Ni^M3G6$>K;x?5N<qaxtRVD{t0t){aGJFG92)*
zda_<Z5(r{}{D}M-u5&d}j&@m4mbt;>ZTG9ATS&P$E&~K287pH`F!KSX7Y>kK!l<zs
zt`@rtRj5f{!Rvd%PgYP4wtiVi?}H~u!M(ZA?-j%YKt2J94b5|<r^r}JLH5dgnl1bu
zIl{Tq2NeQ`tVm@S_gxW)7;g*Y8St1DP%ZpmVi5VVxG}4s&42M)RwZAo<L<vQfW@Pc
zgCSl{B1?+a>F>J~POsyM%p+#KF6TR0ULuBEp#a~ynw3saOqg(Hhs6JTHUV<jn2sC=
zG!7<I$+XDO#(_8)fZq`$z@Wu*bN@xVX_kqNK7K%mp_({9F2KkvSVZ<=mg5WB;t(Gr
zjR^wZHyy8cFl#Kg@l0p5F?bXwIh#5D2uKXP6LE6nacuB=2E(y-5{ZW;-Z8c6f=W07
z_PT-#)H}NZF6Zxo?Zv0_5%+w_d2|foDq+MB#wq3kPw`f)xCiOqo3g(oJH9Vi=A|Ua
z{4rWE@w0&cR)AR^WFM4km@BvRtgI=2E-9dl&s2u@;C-RnA&g}gdBx~AK{gAf4zq|x
zSfrOl0X)Re0u4W!MYBe-_R5I;Fxc)G)E<@i*GO6g5VeV>8^n_eO}1;X)F~yl-&1bn
z=a1IeP)I@&F@!E%e1tyc`K!p(tJ2cnbf$to-1W(!{*3K8lS5ks4yr_EYk`~I+nNcn
z#G1~gqf-USPBl(yX2UBaUMHNPn-ig-$zJ;}w*^~0bgQ8wgA8u6c3|8oH3n0TsfT{o
zNTAR)q;<E*Huw07<oI{aeaBy6rGMs%+m6bbTc6Qq^g@beE&jBCLVN<bZZHSt8i@_&
zlW-H`;C{Wg%_`<e=)0i3WP^vOW>tu_f(bfKL=~43rw9=8^6#ek2?vXn%>lL7Y|fIm
z_ud|z4l=p2jAg|bf}o`uZ+@ZyC5o-PzJtrP4vS5-Oe-L11H_Sgf6Pj?@RYCZ$$WcB
zwY=5lT9ch7UQIUI+{HK8II!Bu%$RLn5|wQJ3v|x(16ah%u&$JHymF>qg>xOm9>>@2
z8M$WT>2Bme0Se(Lb$-I{FDZmrq<sv*&VeAZY1N~P`ottz-6e$hjJL%9SAZGz59e}?
zI4T591!K6yVs634NjQOKn}*cwqrIKdv%()iD;X?FwsD7x{($U$=z0Sr4m%;!-!bv+
z{V>b9g}MRUYkr-idwTeGx@K!NJ!H1qYWuB)(;01G@mju&!^7&{Mysq1459X+y4~5!
zdt62+A2zJ%Rfk*-4wg;KgjR6f5L~xe%-wvGi2~8>3r*u#i-sZ-Do@TJ5y1YSrZ1Dt
zjBds8fen*U|4syQYhE^jgl|VejKyD<<!(@^sV6Bt0E1IKv={U@z7Sr{Sn96w@B%1m
z`{t|E-I2#{YzyGz(6N&Y<Po-C+|eSgWFWZfJg#-K*+NaZ%e)nZBUJGXd%d)~2S-;7
zjsW+ZLQhQNh}jKjtG*v{3+u}vk!2JX?|eqinx6pQ*afmpZlxUOOhQF)NN)NJwzwx+
zTD}Zx>ztt$Ii+@e?af2yS0TfAG3kDN19Fu2k8EQ>bs5n$^vI^au2pJ2e}ZjzuSmPi
zhSl!RQ`--j2yt{bnnqgFddb4`b%+JCUrwLtBx5b6%zNyzf!U`-V%F{lXOUU9$5lhV
zD;K!GU8bYOf)txok3YhLJ^~l6@cuSs)oG0yRoxaWTp1uA`h$u3EK_!M!%Ggv7>Z)`
zi;YDfD>Rgh?Q1aPHy%eq9pQgWX-xG0@%qce@GF@AzrX))XkfLb&M(mk$>*hZj>u<4
z^iBkbefyeYJ&TlVll5xky2to|U#e+Zl@KSP;qdv=Z#NQN5k%&9#RaZ~3;_cc)St}-
zJU=`gNUNLs`EY-fsTP$HLO^=BesJ$e&>%v|BxusnrcJX$7KJ1$JG&S~ie~5g{xphK
zDoL7Zx6{MRb4nqj+x_uI_VYO4*X4V6F+7}!X7|;qxAXOR{}CCv2%q}%ev*uQ=wU2r
zo*@9s7fs)?2nm}(gPt#a7*D$<SVa5_IcJnf`!#36f@$xdrU{ykUmfbqZyn3@P;?aB
zF*#&Jmlz7b{7EXB8gk6Q*dB3W3YgIz37S}$-4=jk%c^(YeU(O{Ae*R-{8!S+W0(5Y
z!OqBM7AVhy;b&T^CZSv1g*(|{A)786@my1*TJ5*F%tfLz*nIQgprt%1_EXaX=kq0a
zIn~9+i5*ZPe7HF6(OsRQqoJnSu~3JlB;kNoLQ_$r{)VO&`@JYvDSgkm+hUWh*0Ua4
zmMXIOzUn5U!S=l^w7w0uux%M(a_ZGSd|C9*SjFzXfl$^;a%xRQ-^Kz&g5xFQ=X7a|
zSkDfci|CK*Jb0Hdg&-5IRGL%5e32$n%mWgvM$JWw`Dd`Khxc~?m}$bTjm}Z!?G8=E
z-O&@>9e~|mF187bb++9Uw${_d?V($P$SObk9PX`m$jGkjb)9mIeP!_2sq8ka(gxIV
zA3Hjl*fr=|fiy`Y2QOI_J}#q(I%PTK+vhohw^!{|c-wUQCZ!Y@su0ORNkADSPcP=a
zxYNRbu{y1!Va}t_C)5bU<Z$;WU|<?J|NgbJbkYrRKF2>%x~gv*2P9G(aAur#s^6Qx
z+ZU(KDWVl=4uina1_N|JH1|aheBBk5Q)-P`ExIi(99-5n85P$}mDhjx2A@6Sm6-!l
z9``l7p*!7m_uGn#QAI&~k1<t&M2$m$V0EYe90HZtL8GWTfi^!$`In5kpJ3rbQdtGs
zKVJ+(O=>967hU0$u3#|pcS&ml;VMb}(s-zl#JH6JA$fvr=!p*vI-)e1Ph0pP7$^c(
z<v?IRD#R<&Vk7xI>e<bNwKm)5+>KuV7q;H-)E(g~=ICy^JllUZQkJJV&9znCHm~Cr
zup6d@&*egw>+d9?5n}^hswJ)5PEXU;uK=3LEAt`-Gv{0pY=h~ErA(-Q*6Nd?Knxoq
z2bbWVT)XeRcsK0(!z6~+yi%XOxC`)4AjZ8&nM$EK&zf%Nho}dQe2EM?sEv$#6A3j{
zF2sI7_%~lm8=z=>N}mr`xH`C$&B=KKAosG8aWz0%_IT|OLQ)Ne>>Y#pK0JNkSJTE^
za2IeL5#tfzrs9+OAvhfox%5xhZvD!ujEsn~a3<dJp!^;Z`URO-;;k`e0ZkgFu5O%?
z;x#x`Pg;p4%A(LmjK<+!10w2F*I!4y5K=g~YbUHaiB<t?^xqnVI+&^<!RNhlNS>ld
z93K6_>{&*v*(8{gBXw$j6eZ12@`O1%&OLo_oE%aRr%5x-gpiDGFL&$%ple#Buy%!v
zaJ!+o5n%(c4<OsF>xw)s<1-?*$5A#AD2?H{iE2|v@UYhx#hLL@n?-g{@)n*uo@rtl
zI}}jKwvE>Igwn!;1q{E%V|ns8&Z5^28=}nPW*Pf@5Cb+{t}$9*V6d**e>hbw;7v;c
z$!iSHhY^#7|H7(p;s%?0K|mvi7N)Hi2j#|OEA8hs?3{kHR&pcF0y4!?S`jNaHvD>v
z#sRy4!8Jf5aI^5U-6c+hNH>6K#D8ZSoZA4#B8l!RY{m{clUI<U9nFR&PG~F@>i7?2
z@b2MzAe@V*LEK1%V4jWKW9R?F*gFJ=)^^dlv2EM7ZQIU{ZQHi(>|n>XZQHi3n^Scg
zr|Nu-zq4AaJ>Q8j#`8?Gp`$Dldk=!fQgEFfJL6#OLE>X7G08toMZAY)T@2HJvP!Zu
zMTZ(uiDCMA9+fp}IF92)omw8DFDnDm9G_WZR~W=u>`x5w`ml;PicEGM5hwx3$U_y3
zk<z#^RZ89>78ZZa08F0~Rru6mv+<T{i*eExB=z3CW0!W}sCp<kThS=uiRr12x+Scr
z5=I$7goSBXd~7CK|Miwe2IWIMCe9&9K+#V`!}TdD$DjmO%p|%Q4Whk5D=9c_T_)WX
z18R~G1MdKki`e%kfq$9=(mqH=OEo1~Pu-V962J;=axk%S{a&&%SCsVwGd(*0(`gNQ
zlxQs1<j~9^4W%OL#39W(j8^}bXmRX%ndD>AyqNEvProcuAkn1jR>huYf^OT=;^yYE
zFm;$9M8I7bu&=LOYL2PcQEnw|JJ=d}pwF2=W4!tN6^_1EPjiF^a9QcCY-bZO`s~I%
zZmqZUh*@*Rgvc(n`*q=(2@W|OzSzN1{jgB8!Z(+n8|{vCi7%u}<sEgwctt}*4bZVX
zGo3em445$o7@wtySOjMlVEH_XII%xVjtwNsO)X<_N0NuV029fx9Xx&}Ee+*uJM^&*
z`Io~N@Sg`oyFJ&iR%0v_tUG(C3H9ISDc$dKon!TGMm&BY;%S?<R{}X8WlW}RiKn1l
zr&=M#=CFDdx)@UJ+q0tm)xFjNk&Eu|^HUY>yV`H}w`H!@tuh{)hL8<M%=r*+3iMwa
zo2Jc$&8b@xtYa;<U^{m-NqfxeaY~)2^q=|Ujv?CVHUf^x1MbAr_x&$UU}MtXWg_%3
z9+H(8$ENM|e72&)CMhr99;dY&2Bdi~za5GCJ~H7TbWY>Zv8`@n!P44Dv#SSkrVv@k
z@Fu&Cvm79c^|@*t9Iy`;iFNsw4dA*#v33vI$_jeg7rS$f_uu6j&c>U{vNe{m@NP7x
ztG5GNWO&til0B6npz64#(RAT>_sMH#%pQEM_J6acK+Q_VV_D;FJywq9?aoW6&W$CQ
z+AipgFUl;>!-EIHZE*6-cy%+tG93MILVm--eUvIYX3+!~_k)1c_$R=4Eq|3W1z;<9
zWjLA7{w&($b~th8=Usck&!MiNJG{^<V*kU6xGdYiGtFtU%GbN;CIQCktE{J9CNvM}
zr_X&y*}>~?|M$GuiO|*dUT7jRzBV3>LH~!~{nseXx#W+@*LDwcvZPiW6ty<*vZtor
zsL0AI%_h3z;jD{s(qVKtxKe+EPqKCoEYnKaf2+EE<dlML(9zNAJg!%DadcB4r&t9B
z0*{))=7xc4iz-Ct6OKVWp|8p_bdXR8aBumCfDTIq3^vr9DSeh?!xi9svjt$Recq`|
zfuY`EVnxNL(RTcy56#jTnl`+!03m<?_+tGYeQE5deD4}c0Z8a?8X7mqF!7(I`H|{s
z2vkrnyZM)Bt8Vjft8N6Wn+9J;ZDDbTG)e*E(K0pB4(oAhdUlklPyiY)4Vvo-!supE
zeVJ6?)x9&SRrihOdRNuXaLhlNQv_7W>+f0%-51=4E~CD~o+k~~LqPnA$nG;7%9)fn
z1kxB!AGUFBuu$g0DrnfTyiNaTWdc~Dy>f8%&;juHvap0y*}7z7kSO0fT>{GSv{zo4
zfRMFE<4_Fbt7(|>`A!_@M=IX<)VAmqtctC<@Ybg36fE94LGNd@2#nbJvQFj37Fd~V
zi)5SK5)LYnwhu1tn;g$=+u+b51@~VE=d*~%oyH_A$%4FjbKnW{)vMsztMTtgW=VI)
zn#;+~fA)~SJWCIa&me`_tnIpadGAooNE3rt17wm=P&l#LGqyy&P^J7hPJf+dL_s$>
zfN0ofW}5{NK4*1#R0>3@oe)unV&wFQj0y2=_%)A!>XfxHPwl?92_RRoF?IkGJ~nW=
zI6Fffn|{%Lghho}agvUM!K@<w3B3(hVxuI*_=6N#7P9<VSwr3GH+p`1WjFNle7Tpl
zCWwXsxHze*QAfvvY*92W>=0BgBW09mE95`lO1Q#*LYyLVh&Olf6IIzMEIrVDmxeQR
z{4dR1XEtlrF@SGZ$1AoTbNi#2c!np<0ajbtgkfK247LuOf<6Fy10GLOxf<dL0u|)u
z2M%v$&^UFW&92q01zB4&ds2LlmAauHlrH0cz-!$e#At7_6Rj(r_~mApi#hCizF>kR
zuCCF$1cCOZvdH|hq5OWH2OfsH-6B~)Tc=6xny>P*1JjfC7l)O9ZlYl2jRX<to0ITd
zsDQ_NKE@7CI8sWl!C~e;vv3X{7L<u|LwMhJ`Tbs>YXpS-F3IvZK7YW8WpQf$a~@&+
zf7K|AObiVF>pXI$p>4PEtE&DrJr#8>!AMFbui66F6IrwX=%u;K>mk&#B~dI4rJN3G
z)_pzWMygm+#_DbeDjN7mNZ{UY!o%}t=2MHzW^K<DZTC6s5(j{q;l$7-%qU{m%LF3C
zl$u}8`9IEx_u%0u6QwrKr5jv>6BKVhy<hEh%*|Q&^$P7P27Uy^Bi~Hz;hUND;K0$o
zZJpWG#7$4fPDgj*IiMlX5TS*Zai^-EmX|E(w#ABd!Sg~{LKcp`FcvV>BH%>Kl3ZZ6
z&j_#L6I1=cfl#4xUi<J-C&t$)=9|s%_baf=&`MTLgOCD^BplpiCk&+#iHE^5z!PVi
zY_Tkisz>F|j}1Hd0wf=tIClM<gDc6@)rjSnl%PQ;+b#J*SGJOha@3M0(M6^IH%JIk
zSd(2mC5MZ{j78%iGV9ND30(DjTz??N4iLSM08u>{4L>$rK^aC1U!T~i9F$cw&st3z
z?+W}ppU&N=h;5Ddp1GI6g6#TWcI3k{VU8dG_~F1^Onxv}Ld9ukX()fhdiP;dA|Tc-
zcd5_*UG3u3puPnnpFH)RSo_qj{iuf4JF%BaYVWjd32w>A%m&+R3BsYH3b&&fu@u%u
zlY-ce!T!WnJMhQbX<Nl6=k<DzA9IJNV$oBhdB+iMS13Sk_0P}A!fzfREw2*0wdd-J
z2gADFyL2^=cxBF*lBrZ>y_R-gK`#m&zj_}&zc#lDkxp8pas;+*<p=DV4_r^fpNjb=
zx(}EJiAEhU;1xlww$>K4pADU+ayh#D!bxLGH~}`dBbWQ~{S}X9_}o(hwpGG?utk7$
zM22{X55Rw>`Hs$2FaSom-uj#E<uTH{Iis60W#kJLwZ$*CE4mrxRrjGKmkX;-(n&;F
zV-a#US0|nFT~uQ=mua>vYRs&X#+jd0rLY@SG)e8C1>gk3h{%L2Or}>UZz7t5?wm8^
zT$I21u$q$Q29m52D&@B7Ow<to+Un7OTnQUiBzUdG#$>LH)tMfZj7pT7q!6>9T*Y9c
zzV^{1!KFcx*Z$cCpGX-gq5#cNn<crD)QcJ&xax3StgsFi?J$xtmlmtfS|@SyqJYL$
z<s-CeYbR-*?q)wn_327p?s6JwkDiJI-{#diFT_#jCEDg!pel_trW$~DlVfA766`1`
z1snm;YXy~K2P!T%3o@4s8a(hiU}!llgF{PKK;EN%mIUZiUDid9I#&B$>TTtZiPdY`
zW!aIRFUkx*UHP%yv$R9xT773g+iRjMw>xLIy$r?+%g-;^Y2Tm*?mo>^=YKJ#gfZE(
zCJfk&9BvJ5%?s~D0hj_B{$&X#uA7t8He>CEj+FmShK>^R$G?Wx_|m3fj+HMP@=Zqi
znRrY%A3)(a?CM6n&smmRQOQsN4w@oCgASIEm^Sm(2TMy+f0b!2Y<W5x8BwYaAmN$h
z0xG~iSQ+#s{zU}+lb-+<!Bdr3>QSQ|(v1n^t!~r{Z51@Af$hWi<9{XtTS*;+3Cd>}
z!-h=ITq^J1_W+@&-9t2zTET0w<`tu4)-VMmRRcp@#S^)iFSs(+uQnG#NN47SxB=&R
z7}s;QV*C$(fG6xIbN;gea{j*+kco-yf7TgqX-YaBw4!!@sNYhk(y##hA*^IpUdqrL
z$86!Q<NbThH%ny0m>>}&;Zd&g?S*#>5SdR|enk_eBZ2_39opgzUpS{LQY26A*Usnh
zkTAz0C0WrLS#4LOA*w_n6cSziaC#A}*gxh>hMJ?3<MVy>xup^{_Ov$i_bghP>&^Y`
zh&gt9?_i;0r@DH2TX%OupRVWge(&7!{p)?kjxVxxMqEPZH|Z`u#=vwA(j`%pxyN9b
zRZ=Nr;$HT7mf-9iEw5$6ay!-KsML;KIaTHywcA9l%zA^xL~U;aD{V5Wl#oJm+%s{`
zd>E<UaJv4pQ@8oZr<t;$x%bPYX}ftUTrOGxZo^_Q8dH=gBAKUM4={y(>E41XswulM
zsvia<v9FECiWx^$G=crKtIEN$d$ghhXv8FBNA1Z+b8Bx?&gkLz)VeZfvXziNRnC9C
z1_V}MF{hV{j7LskcZs=)7SgKLTuc&14|(;>b8mS`O_f<IRdp}6MAw)$wv%fHzVZp>
zz27Z+D`0!^#g?rh>nn=?+q3N-{;iHYg(Pbl6BZx4+g!JNTAp6p#lX|p4S5cukq{=2
zh>*%;;oJG(<cL)NJBd`Eku#P}<mvNaO%3Y&5F{y0s7G#pT(pBSgVR}IHZvmWZaI~!
zBHey6!&U$It3LlH0d;L-OD=VI-y--0f3{0p0~op@gN05=5{5|a$3z8e(AULb9KVcX
z8I+kwy&Mb2=br~4RbNMMN*si55|}qI<}@`4w&eGdGZ{s$Hun#pAb)v?RF0s*pFA*+
zAIl&P_%XJuOACq5i=h{M@{c7bNBzkS&xh#Qj~j}U18o^Ig~LWD($|y9D|um|E3(SW
z4bKKDM`KTxRVA^(YNmGsW@ZvLN&j|WNK*xbIdtPxg~Is&J-h9El;HRvArQ6*;jB+?
z$P16LV%_Crf@;a!eAHqX?a<na@Wm7f=Jg{V_wsY_m#%dBt+dT5tIdY{k7)U|f+4tD
zkr#7sO}ll<EELUcV2rRA_6L}BHoLp2#Tr4$m;xcCqVhRXdUeqy0F|*+laJt6LY8)^
z@a4(8&kw|KKn?SjeCI^7iPB-k9Vm^8n1F>G0i4LVSWqZuQ&wr6y%4}_9TSrj;rUwS
zGD|zy#TyrBeHR|ip^SW;fezXg*jW7WBP4P5>x<?-V5xjHZampGE&7QzMku4jq^`B~
zV)JYYSx+pIOo*_G`tx6}Aa3xfF++IKHr-anHi{3zKAE;z=Muy`cDzk}3O28?yw72u
zr&Tmt2znQNTWYXkOd&Ao?3w|`bjpZYL%QY3J;Rwl%1Y8cVm*Z1^aX0~@V_S8!{sgX
z=UueZc7}xk)!SoNxJBUl@JcYc2~={f7{&4WHKXuO>-B$`UQVVHZ>|<0T)}0W?N{_s
zo|-IvZqFT4GBEEAqf~wrUyT-`3~C_&LR%wq>=%EL-1p+8G^iBO7xE#D&UL>PlJ=K7
zy^E4(r1N8vfYn^#4g}sf6pqt##&hWcRWPQR+1(D7r}wn4%{%1PZCXytp%?XDdNG5!
zP_(GIq&okd^wzX<iw5l$OA4Nf?FQvM-F|9!Y+^5G)Rk+$R**koCF*jH#2hm?$>78a
zftJqToVv;xeJ0(FExC48X93CkD=+Z_(fdJB@?cWbVQ@BR(uC+@_7{*Pdk}D&H;eV1
z#X<5>RR9d{Vj(~TTZIMXgyVo1iUHdbYGfivb~t{<3@=tm0eddOd{tEBI!gQ%Ll94R
zp>*`;=@{B6D<)PXhDh>-#I4KmKb%hix`ix#JhV>E38f>sb}%YhblsHW4wG@cBSFqD
zmx*t4_tZe4OU#|P#s%cU!@Vv%NZ4=)#7h<ofqqV|PXCRJ%5)4?uL#Gw?&Ro?Q4~Z1
zpy(^j5RMW2Dtc|5N!=(!CWlWX|C?DXz(wHtJej(yD+2G~iHaM#`j;5pzJdl*)p57!
zcwcOEt6Z5+3WI>?N)4A8eI=UY&n~zLT*myOEzP%P2PM8PB{%)L@cLkAG!kLk12NMe
z6{)v55Ih|h1lr~xgE=FMi$$5bPco>Cg2F|qbA*m2je~&cz_O=I-f@jrf2`CU&!}*F
zAjI6z@j>n<vT&fI^fz1TWd6vqZs@y@3dSfJ_iigW_~XA}C7S@N{8CCDDH$mx1ny{E
zR*)jYEJza*uo3uGODlu2uRu75S=eAArLcd~aXK}EYZ8wFou=IS1oB=R&a-y_LV+=R
znB?nD*^*~(naB(!)I&DI7l4z%rS>oJ{Bi56jWUqV^<5$SF_D0z;x6RxJ1%$-Xel_(
zc)N|r$NUF8EOo4`we-ax2~0t@)G~1Hh?ZbJWoBg7Yu%c7!5BK0#CoGftW>-Z105eR
z{Fh^aTxY4-ad#=t<`(|Q&u24$-^Dt?^|UfYvkTmeEqtOWUM1QhjBe>nAK}>!f@QHj
z$ZG;Tf_5Np=gu;-AiEtfUf)Z2@Y2d2lQfPe!Cq5|aFa|xd506bNmi;fp^k2=4&E>-
zqymLAqhT@>wE#1hDp>5%fwK!W&|UtqB=q9CGi$dZ-*!x*LchV@lM)}|lhDjV(IX#?
z;3$*aV_W>t&61(d$m5*G)r&tZyV`+t>aW8MP}m>Vu)0llm=8#R=bcYK60d50H@_<<
z=p8#V8!JpbB|r#63peqQQ`s9&ftWp8uxi7_?`r_DZ#wC=kIcy*KhPc9yLLx+=0;$D
z7^qyh*Be2b4qD3O6$N8(-51AzCg=R414KG39qZmB6xIG}=S`Ha%^#drNl?~RK^`-f
z-z*`+S!-3#%975owc$m>Y(elrV8_a<1S~Lc+zI;`#7ag;T_os|?<WX!RNzdQ7L5_f
zuMi}PuxriDw5MgCdS6$|b1@zdij5%y;nWknnf}HrgWa=#X0&`4n_5rbKgi4S{}8sO
zqT+eouAmxn++N%q-N@$$mQD5XwKrk~-am9WHG?*4BAaVlYK|iZe`qN#`IEmYEq*@j
zeeUT6*fm%Fqg!I+{J*;;6%Tt80($vBmdegHQ1o&Hj0_C_=Yr+v<V?WH&h~$WXt4gz
z^JJHnjU6^S(s#9fU))ug{Tov@A-q^XqhK}YwLg($fL=_#eU7Y^4$Tn>=IGZ4pYYmP
zb)%M@niL;L#uhFd=S=6qb9b%7@Q2WQ+_y2Zk`hsdLf~DWsPG)+o{W?xq+M_k(ijTQ
zYxoaww+Nf)I1qmAY)HEV0kyhmI6UTN-+}O>D$>QDuid)9NQ#gmlFMu2T#j=@o?H!;
z@Db0uU|OE;)OOJ67~*n{nf%8`LXn`qQk=*_b4`6&0xFcJsZ7gI8NF1Jx1p+>0T6T*
z!T_Yj*dTza@_7DoaGu5fvQD6YtHJvcJD>{1D$!(IbHsp|WkQ0W$U!yKgCaz#L}<v!
z<varDNfb!#(fi5pStDDv2w2b;0d<Wk>7&?=P@V*ghn0pDm{3YRXu|Gow;UjfN|<8-
z6#@v@Qjxa3;QKn1OeC6s(g>X!taXxdQ-HNH_ubvn`4Tf@+r{y9q5v?IYS;(i3gV2w
z&k>c)G-Q%Z+JJ(da%2i^P<<W!P;u@9u`bNhp<V5&{$rSlwV;;+b^Ww`44hj)kxZ2f
zU=pT^%wPmliRLF3*-`g<#ya)a&`D9y6~?B-=gs-`h^x?E0$z}X)U`5YxuA%@-0@aA
z*^+kqfEV>1|D?z==t@7y>OezmN*BPu;zdx1^2<>}a1l*cOh;mrQ7T9k2F?mdCI~2(
z1t6&$_WmlSu<>EC17|5w1)UH9`(rE_1a&1n9UYoX^&h{T%;S?v{teNG0n0X<w2heI
z27+0AnoRU7Q<s)*TlNl&nw#&f?zXmf-={tuu1*4XOC;Ye|9Dn4YNJg?K3zKXc-OWG
z40jVf`tFbF1{)I%7W$j-)#VevEt;PW^KSjBoyH5Q%`RJCYhO#>*2gEo8K*ARv>|P9
zZo*#cW$*YgCW}t43(4N+MID{=;}VPSHqR!eDdvYMFlak9Y3kH&R*u0h&fE;?o(W5O
znTHEMUttYG8`OqQub1ls!#v<Ls@t|>;1(m_YgSRtHoy7g!Mtt@xWw1n;8Vl*K}dMu
zmoH?Qq3soN>y^(=J>4Gb0^;cV%Qg$WCj91)+AD2p4zAi%vZc1=`OT`6^0trGwa`nD
z7pCu`oZYi!QdzobvSniuwJiwBVKPPGt*maXZVyifMK|+YHEmY<EPcJtV*Z`armgOs
z&ms%kU(rLex^#!u&swLZu3ct}+r)T>N4@P%^XzH64ckr4a@O<9M&;G}@xV>x%hh{D
z{c2Bh%T4!}@yCKUg2g%#lB~=&Q$KowDJv)OjRu|XB4pQ-?+$d6W31;MA+nWJQ>F#J
z2?@SU05sm+MIcS?HE;dN0WbgdgU+Gn)Ea!YDD>O_`@EeA2O9zmW$x*%NAKR-15S5X
zs!X5Wdwl4hB<Q?6s%OEgKxWj%wb3{C@J+g}N#W~0T(N`;#u$$)JR_Plr1LMt@Q+$o
zz?=66j!oUT?TG~n=xlQWc|TZYSdLcyzP_$(z5QX~Ph^6I6gXfL8W|cR!uKF1G`w*Q
zj;d(GUV%vQ3bX2;BD=<OOhqOoURio0JG)f&jpwnesj#c9_N~>Qt!_?zUMOQleMVLY
zOWQ@l5vcb!j3^fz@e9TJup<f?N^wI=&UK^c=-4V50}*PwhJz7I-f{bAWD?Jz0eJ<m
zq7c-{e{sz1205b#(L@v&&RVcy#Iii;T|}<C-pk!ga3IE?lm3_chzDnyWzfp#(Q|4g
zdB5<6{Iu2I&q9TFNH#vWqwgW|Y;N{L9Qyp3tRDK$8p5oAn~<+BD1qB_-GWoCX9tR7
zvm%2XQPf*uRZjzBT+62f*&e~jeDZO`vnP>UI`g44sNBln!xYEv*dTd2W6nafVhyxo
zHFE;EULA+F2K@j=Qz=a5-AGI2652)++n_MBTn3)Goo8?QDHMNy`mP25osogFeiba=
z_Up83I19qc*o?dlW%k~1Y2K^&xH$T7mu9m*-TPTx8rO^xVIoW~$DJBou_FX=o*rQ-
zmCx*`H;4-x4i4O&0Xsh2^9e>GS=id)a1nak$X#w6G5oFs`EEMd;Q$_Trn$VDRPJ0s
zbm-k^?M{e0P3+8W**+M+G&z9I^oICNL9^QuV^h*~AP0U$K!R|;cNs}8@lXuM5`>1)
zizM&M1$sI?kLjp@6PY(q|4EM)7A4WhrYjqcM}GKWxhTm3AkLa!S0<PKS%4VPU*zOt
zdhad{SM&>(alG6|D4M9e*(w(!7q<8zcmT~~V9m|o`86Xh$BXCHu6gem@XQdQyuS`^
z%^SLEc{a_+W|BX0MDAHn^2Do=BeOV3zr+#jD>z?PobhC}(&C6B-LyO3Mr1pq#H5c&
zJ($Odu~FQ;$KTB8sd*jtNLvE?@z34}TcO>@j8%do%`f06=9la*{X|#69JP}Zqf9u4
z7`j}4N}sWDKK&`C{8AZZW>3E|Q@-ls9hUPCtN(bbnSUWK|8JlbJL~`M2+)xJKY>=i
z^srlrCbVew2v<)D=Gd%}*7B;{L>Vwd%BU(4E|R0c{}Y%24dk7fjN_J>p$CGbae3Wh
z2j#Vb_dIRJPiOXbhT{Q6DN)L<?K%AsBn`r8kCyl*MxG<5N4uNDN38ZdJ^_JvjDw-P
zEcg5I@ag{2u{$5=1ElCpNa8Y&|Ly7g{4EK(<LmwY<gb~!x|w?D4h%Oe*gj-G%5$AK
zm8?WKg!;B4IMcW+k3PfXr;cAeJUiD)M{D#$f2UK=Y)6(2&f0(rA9+FE)n%64RVIoi
zBzQpH772BSYBE=hvnZ}6pGE4;<1>h!KLQ+WPST7|7uAcWm4bh&oO%F|3{L<kl;Dlj
zPPHpo4=UnD8ij8Zqijt5F{|X2C+=T|>Iu(}1HHc{N@NZJ|K{SM$`H2gr)s*&izXh#
zx4E>Z`KGbFVGMzw!G7%bt6Z5IW{LscXNjpqbb|Hpye`)0<Cf+@LBxdm`Iy5D=FgSv
zPUb^sk2awR?Ma**CQ!4f{}A6lI(Nrod_WZFbsCCb+!l=2nxy*t{x?WMs=K|ajSz`(
zsrqk1=kE~KpTX9#(QxN-a6#Gda|Q*FOX3UTw}rR+R-*0x;*3O}H*VeIv6EbZRUapZ
zKTU1!W-IDD+?*p<oGixxx5>0&3>Q|DdR``WUj}JQWE;anH*PmY?kQKc^*~GlyJ@$=
zRHA!yVtoMT`tWQ^h<PrFcL+eg7cqa|;3AG<0LTv^y93pXqZiyG9F(+oYJwdT#ZF#`
zOhpmkCEzPs3PGSUPdLz!s$;*65{jaiW;HBzUs8Ejsp7c;+5~V$B!hxL@e+bQ1AL;M
z(I&TL1mL9`PsqBBjVJB2YS*`81?$>w+8@~r+(+Gfg?bxPq;U#ZN~qu~lZluMZFya^
z0c>^OoN<GnNtfC|mkydOE+XN2J1wo$>V4!Dgq8TX=eqMMRB>ggo0bWi(xOP<`1DYn
zr|s=B0pMu&El=G_IKk(tX=GTYsgCpHX)nNi4Ma48(`O%F`m307$q;+<50S^4#actO
z{YY(Stdva<BsS!5$A?hf*hC<e&)OC=R&Fb7Gi4~|a)h)-dY)$3vtE1DPzwLIDmQA%
zVHj!yu%bmJ6!tqR#w6*Z==k?``$}DFEfT-Qkq?$~9EIWMuDIz-4Cf``PBt;dPIbxb
zYJwmDUO;s+IT7j&n`CU=brxZ&OA71G0AVwTu63OHT!>v;hF0Zs_OlMe50^uVNMnZ9
zmh$MnAj?%2t3<IoSr~o`3-<RSvZH#3F3u|~@Ql3$QCXJ;Jew(LZST^c{vw73!ClRz
z>)|566?+b0Li^J#NgOPRQYiRSoSBH7>8jcVf@Wzq?+|6}tUWmoGxlfEM7<YVkqFrb
zRSU}A2hBwUlM6A>7lk;zuQTjfp&p`XctxFmy|FGsbNKZe?i{TGKExoDgk@HY$%MmY
zAIAZE5U`fqi-r^vGl}($>zKo?oGTxN5{;F9>>kaLQT^-K#m<G`ksKn}4#4McR$hC`
zO17S#nehxuj`~8l5pjq0RVyNngN#>b){uFEA6LtKs?&gGr%Y`wVXkbo&Z>C*LCw-7
zxy4PMSAs=&Jls!m=L#sgpjKDASrFjBPdYqGr{=?}1&rfZ`{|5IE#@z4W4&#-rKN)G
z4{IjW>;gmt^Gtx;j3iT8gzX%Ec+iM2Lw%lPf)WM6y%L3J>EZg58CdP(AR>Q(L3a-z
zQvU*jC{PzsJYYyBDZoq^Vw4JdW$baOR@|#<Ho_~FxF`=8y@Bon*>~QI{^aup!}hts
zF<tNP<wFFaFguk#Nby->Nx(0;ztEblDwYevP^<-yJv=_^+_<(Mi+6l-fbTSag5W9U
z5%*q5fMA6C9}H2j_amC<x*Ee=0u7UeP7RV(2o@48BPR$1?N02Lu4RY%rUV#KDM7%L
z9f%`7m~M{1F=d4Qixo;mayTtStEVm;0x`x6{~<l})uPGQp+J~>xc4L~lhj_q1M_Iv
z5lUD_wdLkv9HtTm>89LMZuXx=MiPRrAVGs?l#mV+k7^ZC0Jo~riyenSKqEfBc?eiv
zM@s-t9@-i1hY_P)N*=0vL(`d4D@?kR64{9B^+t&Uc!BTI-#QTPhy@2ri2tF(giv9A
zIbv5D+{?!1YPBj)R2&d+15c)1^2=lm+K1m;EbjM&0S?-y7h`z*{L@V=3Eo$1l{u~;
z(<-)C_s-pzX8Ds2H)2@c94n8>YyDGOG)snB4Lh4hLZIVb#KxFHs=z06BQkdllvM;N
zMha8Vm*2IGA7y_Ka`D}hVauyM{S6p0kZk&2FD27|u_+k;cTfJ(?-lfb4ed8uYIb8>
zBkM>uxpj+<q2m-}4QTCb@D~L#6xfWZ&0T+(a`?;O`tgGIpwPUh$RSzaLje?u<l^CC
z_uVP5XpAv;_PBdF8dA5zTuLC*azLfMAu1>op%OAdqGg}6nWR88Zk9la+0mm-Q(sAI
z0F^iY1PhyB=il7a>1rGP>vr?=6lZ5=Ysydi^R}c-8ER=FWL~&iS*8j_?kpHYh)A3~
z5G_W71Sz>oXozX3BtF4A`s0;4%3<f$5F#OJZBJ3@X$O_9KH{o-b>-b0&jKi^Glh)<
zuexwp*}t!>9nxZA)zV~PQ$y6b6OtnB6zo$9BXy2Bv2SUU1%_)(gXuoX^}tH1R9^Sx
zhD?ar{*P4Bf@z!C%}dS_Infb~l!XxUa8#P-xOw&P2rcMl$~DR7h!$2JT$%l?o9f|w
z)%k|K`4w7gupBErS+PGO5-l4!w=}(i3xWx;O=}dxDUouH>!SXR7?nlhtbi+iCDKJn
z{phP??MC)*9Q>1{<f4GS5IGQ;k9D97P`N2pOp>IF3=-X=0DhLETY>|Tfk__{1}9XO
zq#rYWJo=lq8guSYx5!_aL(Jp|_lOCW*2(Tclalbs`rUegy<^BU)r#+Ux;Xe_p}qve
z-8wQEhGd`5sz^^in)T0Z_ZFvMBtmZ-B@RxSYvI26J<bCXz;FWqKq=?E1LZjY5;@LQ
zzWCw?N_y743i{a_vVjyJAbAnLf8t1#UIlUTC}6xHXFbZQV1O%<<c4IMf%_Al0dL7r
z5?AH#=aPO%qu&&7&~h%HX30y|>CB?Rhr9?mpK_9k_19iR0Rl-dr?XMI@);!(V6ia8
zncE#nbpGcutwN?qA+I8^0{MbrL|W(FK&xivLZU=L50#+%KPkjCl>1G8DaH;`hC+~m
zUssmC?X6|A)JrHM#dR9p<o3m--nd#!qCW6ln+I30t)OOTuJ-^l&_Q769w({2jz#OS
zHZp}E=6g+Zoddi>i?eD4+Ht+w#{8Xey#k0S6R$VtYJu=q2n;)PbN2-kPWr*rVc|rd
z48ER+$aXgB@4o&1kds7lIT8M)x}{PmUE4roqX#i@aAL5ES~&04_Fg@@z?JK=VzB)j
z?Hl5bx}#q~^CZu3G&f+0xyp+K64l+*&aSRtJH6}N+~uKn=(Lc<RH1S6iO;|j(X>n?
z;d#W2$1o#IgRF2;28E=7(s4%J7i)ySd<3k=)^W*gonvW**M`!dATX?!*!<iEn13@J
zkI?DPV(e|_><%h(rVsVnm@Z8qEP4jRL8Q#?gTd7lF7d1si;1~2wR1`n?@>C$w@$dJ
zpbGc2a+c||1vr<<o&N`b1j3@BRr(r3dX9Q`eP?jNch)VvZ*Mt{#@$^A!JB$2<6!55
zG47(^;6}4@?Z2n&5~c)(FHd@v8|p)H9>}iyCu}5)s!)XEJiX-~H4fc^`%}0yn6sTU
z3nE_MAn!Dsi{8={&aW21en(iLiTTCwIFi;{G@*)?HhZoU`IKtG&lWU{fP(d?w87i`
zmb0-T&;@+QC{h}3Ux7g-ZysukGZ@!;Xhe3q(WRR!)IN|T)JXv%j|`@`*}(QysmapX
z$BrF2e=xv{vT=R8ImQAuo33xE9D~6!J%cXKXMIGJXy5cVG77d^8x$1#B2i`Cc8!Gt
zRS<VXes?~s-qT-bM{bW?eI31ueQ8L5L^3Q%a<dl^azTC@umTT3W6Oaywz=;u;&J5P
za*j|br=e7AkF{@3ayr;PFdX)kgY)Gr7cxD2>%aS9z8O8Y%H-Lo=<Y4=P!VJ?j>jPR
zwF}F3_G_3z_e7)wuyCmFR!?Ty*Z`DvD%V?t$OV`hV=t>rVmgpbCVN5XmDbK+iQuet
z<!>iI*s1b?3UY25u_Qyc!}IL9*WvXA!+vUxPrx9NC${sGn>{JtsBo@cx^*LT=jJ`E
zOGP%^Buck59|~fUg>Rm*);fs9qRJmyAc|ZF&i1Z$?)D;<1;I+^TPJIW_Da0|yMKZ2
zLAobyae}wI(V({+!NZ%ha4~Hc)4M8#pyU<+5zMuOy!+*hE%vRkqayHaph!0m-YDGW
zzR#~zJNQu0F_K=cAz|wB47kBQGOlIhD$xFO!UB^iA;)H=nmF~_PjK-xv@?mKdy^Bz
z(#x}A{uQ0T?cRClrFg?xZnYGi<ZlXs3!>=q;QVi1U={fv8tKShdfm{s*kz?rFZpMN
z8Dtp{VQxpzK&NGPuv-JdyG%j2aD3Jkx#Z%ySahmJc$3r0{P@V`7Ik*xLOaQA)h1Uu
zoS<F?j!+z9^SLO0-yz)eL9Rbmbro1Pn;IR_djQ3REvR}o61s255DacT$o5nQN-P@5
z>BTjTeFjb?D$ADfk2%7`3{F1S>7yb{$>G=6lVfhUf%iC-aLozb*CXUO0Vdu~1x_N=
z{RL|_p04iq{m{P*(PmZaBr43iUNp#6fC@DSIRr><1#F5sqHgr93;-CVb|YmVSK#<S
zZLz94;rW|^Bw!rR$AReZ%3=7WUuR+a?m8hjHu+tCqDFp%&G~<9T34ww78xAPr}~>v
zuK$=R;QTzTQO)@5!AMoI!`|r4xE*EP>YhW4{y3GdiJ4L7rroaO7cBW@(V~>xJ+06~
zginSWuW=BRR!<$Y-EUmv<<>keV=HNQn!`y>SNVAd1UOr-^D3CB%C8->@_SCzh9s(l
z<m|l^=IQ6mGb4+UZ}~hn993X46ne_GnWAv0c2R(Q5XbIJ3c<ns^x^8FKW*L@p5&#6
z-8<SWiew+z;hgl59;^grnfK3Mvi9Mo_?9I`Bz_?=VBlWwpu(IY8z|gV)bs><avAeu
z2k7_$<|bwq{%7y<UsFB@2jl;$;M&sIaQvO}eOA<O%S%epU=aZ5yQYfq$St<o%d6z@
zHlx8(2_lU2P!bcbjMwdMfS{#FXea7EmeG+{wt}`Az~dtt?86^SZx`I{KQo!hpZ9y=
zv+p(y1Bl&s5$Uq;Cfgh-PX^!*^kemvky=vc!rcK63<m#kd3>C2y*~coxCeEhybC2p
zM0lC1<fo_i^L2?GxS{jq=j`g_=~{dS8zdso!+96t__AxW$IEJWKdWB2+dc3P)ZSMJ
zV~kTce+7JzK;T@k&L7bvFz^8x7$hcgl{%UP_v=JCdX7%+YsDYH@3U7nJM1?NoL`L~
z7zSKm3}p$9haK-CfMT33NwgsMW_ad>^A3nRSm8UO1^3+*b=W*)ptV(9W84NeUKXEv
z$N#4f<S@XGDB6>g01ce@sQANy0~*}Qq)U6`rtb}8Mv{&`T)<q|J+jO2GgGEAMl?{y
zLA?7Hp$quuYkvUID!icJUwZlD?pgZH+a4QMUS0xao+prui@#YtT~fz)dChd(?dA#o
z?Z?I8C1y4O{k=nI9{z2EhV0^&<-G%~S&~KFTNH$&+hSpvP1*!d>FGq@RRNXh=fNOJ
z7jOj-N6Ea#q|QbIL@(%qr7%qw7je4B_CzPV)V?+;C!ZL8g9u$DjzC7PlxpdH`?=)f
zgokMyFMVzq;bEjn2A{+CpU4G(RBIbjj~g0!bW{9nw$`pQ^E{26+>1_wS3aXL$5t(k
zxHmKPwj+h_f~3PR<lhBMaxE7?izUTAFIVl9b#RaZ4=(tENkYZ>u`iSI0+BXqis9l~
zs;rMGz=*>E|EuKvlrbUm0RKIX3-Xr&F0WCG!&k2mxHC6WsRVazr6=216jM_9@Ym_E
z+7fS99d7Bzbg6aDa>N#!1yS{j*+o%R>6LYcHQ0yfS5GI4XB_7_Vvol)1bz5k+dv3<
zsC_UDPiz`guBqYSv@va+OAgus_7;q8zBJVYA*`<IUbt?9;sVyin_!8C8OtBMhp*}u
zugyCQ0jwaf*TRm!=|vO!?Mg>tQ#AgBA;C4OE)c+*T6i(;@B^qQYl0Zj4E~7DEf#M~
zxk=tl)$r{mKpc}^BduQRHV`rpp5}X)QxSIa#gYqpZ~%LN6HRtq^>ikBfa8<F6wg`F
zLX~3_ED!r=*rK`H8!iN3gEO+B1=`qR0>sKpBdJqta_GyYx=)grz9?7}C#mwoF!!eP
zwzHn&MuT0Q`sI$K(i!DiDOkY>)nvMdRF^COX0FrKXEzyW9-x8}&CswauX@_t9tm-d
zP4O=EiU4Vh-x@lShobgzSl(I;E3DR3&Mbu(kdXQ?C@i~m1QR01${Rafe0Zc}9N`wL
zF3zs$czm$D_`bF-&m^ergN;JOV_v_^VzW$GSZ8I~qjsipAeWk~E@Bl<`dC+;XCZxc
zQvg!IxwBFU_E6fuj*Cw)?~@Ove%*|jT1EO5S97X|8Y>-L(zS7WgiuD&O<lun#^dp8
zTcmB%R#~hqQdp?1pj85qDW9A_NLpg|v!V;CF}z8lugg=;$P{>(WTdLPy!DQ%WTF=2
zifHQ%@NY0d7?2&4!yIU|w}c`32snLNQG*H<Y-~m^gWF_9Ra=M1V&J(j$NhaS%mTV(
zD^{}0#M#X*wrJb?RKZZh*j4%9DKv%0<o4j(e26zV<nxEm9a)K%(pAglqL+|ab+j=O
z!r1M=>*Wlm{JMt6?{frNbkonkaat|#A^nzij87&6ghct=Iy|r{Q1gvSZ!ozNvbA>|
zpxZ>i%W7G*I8a;=|2%<Jt%r@F_&Jii1G<_-{r`w<s*K(HqucztZY=j8XA#eZvxO>g
zZo(MZfhPXAQL;CKW-<(*VVQAXTU;Zw$O~x97EwpJB?r4w_9?BcvN%|4*H~Jp{-y3i
z(c2Uh$g*T>;vP7+7BpAEgtGc;KFb&hCM`>016lpc<3T#MCgr1=BVuA&1-!q-{TfVx
z6T`SzPp^3>53qrkUDu2xkPh~W-njU2&5%h8ce5}?z*><yTBTBRNn8^VSg}O>M`77f
z9>}Ai*nou6Mk(f0P-|srQ9(tVmhj=brqM$?j$K@EO=3`Apjtbjd^vD~^8sX$4cW{Z
z$t$e__s)VpzX~c?Qtit=SUuiSIk{9)6TTHfUDak;qHv9o-Q7I|a0w%P(A*oUvxllQ
zdZwiM0R<=oGt)oY0w@P#T%j+<$~LJR3~UeLkwc}M&gUSg?f?c13yVCCA!PAUWh7kl
zR#!|(whm%JF-#FXWW#ad>H|_7kav$*VhW5SID33@q8?^4x$=Vt%@tN0{7Ae9o-iW0
zsu$8@$Z8Km&7##L|0IkknRj`*;X?{SxY`)l3`i)2#oMSDy4;lVukBxcU>KoW7w^){
z-j>|e)Sa5Cq!zkIVN=jt?^o4m(=3C#Ztf@(t$kzSZ6<^B60%myQ<-ZJap<?F#1t$4
z@oljz+j8{S-`oo*%Gfpax$7P9XD=?iPiURV#<u@@s+s>|&A`a~zvG;?{(r*6O3kiz
z`EnrO_9^RuD^lD+bmsc`W8a-Q0%9V^Q9*oi+}Quw$kR|HHR5PGE`)VL>?U@67}|wv
zK}Tu*4bmqII#`Kzj6%la2w=_7wl#Ru6DT{;PFiiR$0Vi0jQ#x8;LVUFHQb%qAHFW~
z^&LGP#!XB#Br)=Jb-Z4!sHde3T#RKMtmJN>UAzUq1DcDL=hpCHDF6@t3FfLTLR)~)
z5J?}Yk_6kCHsLf*pnb@|iPhn+O~P66%us2ah}0-;^^y1CO6})ioEFY!;1JvmeO~L_
zV<}V<Rh?j>lp<am?MSiU%0$g)tdUlnz!-Zt<{pB-rIm6=X*N*K6REXao%l$hQ7=Tv
z06|S*LJ%H`y;)6b8cj$xu2eD2mspfcIM~=Ak-fX}lFGq~&;}(771r*ZwN3IY->U;R
z$8(Fg;ee3MZ#lk4W1n!rD~d5987ySm`4>r2L%mr<XQ_@PgRBm+wF!%F8Pn?q=QcZI
z*XQmLW6fSDGZ?0l+3V3lDz&b-i;R!g>5ph@@WXkysCKDnKGXPSKs;a3o71UXGDcIr
z;Pn8R4&r9+vRVv_v)Z!Ti`>el{Wk{MUtu!FTL++85BWjs4L)$(shn5Quxx5e6d%wt
zF3CR>0zc<GFo;h=XC(!P$*Y^q7+G((w`9`63@Bx7s67`o!#AHyCL;&krGb(;p@P6I
zopFIZre5dIt5OGDJkkEDwSBoiyUl(Ee>Oq!<ozZ>=h;3afp#KS8zZ3(u&bu2_^Egp
zn2k6VB@dB;Y`fAvZLA%oX}k#GLkAC3`*OtFz$FJ_6cIXQVpGHV87p<8m&=#}J(FVE
z7;d+;%swjDp-~wfMY}-#WF;$^FtKtv2#x=?JoN}M4@*czWpZ<}5;mzR(O4x}9iBBL
zB)Mwgi6kxLvS|GIYxk1=M99-(Wz+(2o1QEfK6n;$TQHNomjng;KTS3TWA@v<hfzmi
zuxpqqC9Q|Po3KC@(mbSNpe?{2<Q>scxTF3@@-PoIkOZXGQhtwQ<gj)PL+5hK7NjzX
z=oXA@e-W<M$CJEh54Vd@AvA!5fpY|O26NxC(5+z?lb=`Xgs%n;1G(maRJ4}*%>xR4
z%q|q2V17j5F5K}Du(9_tztBH|<-Fg9@Imf~cN&YV;S|yiJLtZw-M-zc)Ud@WydBhH
z3pdXI5*@R)hyUotZB`o(y=XBtC?JFAPIkaM8UqXpkxRommAXWi3`(5MS{=nY!#FMS
z>|-b(-ERowHBm!Bm-d!i^BCq~WMbWS%Jk=x#T2aW-?G2F<uH;ffG=BU+gny|NZ+={
zcIY#29wWTz(}wskZ#qa071>LBxU#~Xn^3^DkE~*nsWgr0o5y33Xsqt|Rp2^teW$&v
zg|lz-G5Z>C<pM7`9&zF}f!o5AOsIR<)6z8`5~hh4j$^J{79IvZ&X`@Y*?mOmqn1$U
zOc&y;5akU8)K(rkiqy7oZO#+h|FZ9@fRj{bVwXp3!h!l+KY7D7xbKRDKfSI#ZcOnC
zj$m&$cHkcJn?F4K5%+Wj<24ET+}%}EE?)_ky5ClG{p*dNmEoAhZ~A`y<@@h5@c1LR
z?Ys}S?~3&OP1TksL*MQ5=XJ<!APmu1{Q0E|YHtIPPaCcL-*<c7Fp}+veM#bY^GTM3
z=iy(t7qRu0jS5Bnfo7n<TG4Gkd&&E(Ga=&)1&-(Pe&3yI@sC)OIFiPPGg0)WZK#(a
zKR}U(zcBw<c>jZ=$Hc(-KVL$B*#UN&!>GR;y<K8Zjf1vZD0V=d6^A77l@yi7K1iV7
zqd(~+DQcpXiS}>h-I>I6<Q%jlrG^7~6qI%}oi>?1;x-fG(E#UhcY8j$Xh_=;r5e@D
z9VKb)hE+itRP1+00ben>WkgYQ{9bs1r|&`L$)iQaG>eGr9jx|zSXsfZ-qFj|A59Mn
zY<P&`*xTz^`Dtx`|9<kq`nG>-VYjQP-qyqPHQTi4)yy5|fJ>dR#pPjfdDE$ASuyL$
zh73OnfJpwM6HXAxlv?UgyPlijkTR%|2Wl~?LK7qh-eTtGSDolY44nXuIhjUb#HYa#
zQXo2*NJEN@_Ow;X(IdeD|05meOW|j`ZmYe>f1iFJ-^o-8JOfRbfrN<oyyW>oBgD&-
zAtR)ChO*@)0v@=59vd@vBH$BX->b(v_EUiHDyybDSY@CSn;WB<IM5K;KoK-ZOQM%&
zrs=*zg!Di{Q@P=k5c&$~Vz%wvFwZ-+b-Jf_G3;HuS3CEpgdSuIrghhLTCrd1%6!8&
zYd)8ZCyQhYJ>jLxDk@9t@+7&kRBAI3v8(}6@jywHM1X{eOj&NDVO(H{0bRBOi#=Xa
zid$hC!=fw{vZ6*L=69&$`Xx|VnVUOMiedh~eYb{^x20XDZC6LHJ0Iw}GWV)GhpNqM
z{^FtBQYjxRE=uL^x~My?a>6*VAM9-R$g4)yHUgWM_hf0l<>e(kqs4B;K$eeqw^<Y0
zF$$6-c^Ks`j{*-GI7*6~ac`j}s%tbLx5z^&uBZfsHb9J*A(@)gUo54GL|YaeHV=YV
zzuhY~a#GzQ#!d{&OKE#sZzKK7A3jRjn{t+J8@=b@#?U#697-TDwAwm_3y{jCI<4DU
z5;hTbwU_%FU9vWY?*0PnGzqRGN~4rP|MVQLd8{RsQb!`|Q;05SyrJm-;%1{vRA}LD
zJ~Hd#l`#*0BY{QeWTg4<H1nWC#GhJ#(SQP{LS||#M|9&FDu@*}9VanLtStjFVREd@
zcGWg1WW`d5!URxe7{gG^WTZ$mB4r+b>zsdxc)2URTN@^!LJ<p<uH=8RoA_P~W{y7w
zWGJOoUn{1r*)xD=<3xoi0>h%K#GeU6B-$qc<DmlfEm165xCr1N3Csq`Ab>a0n<cA#
z@xAc<2Pp1bf1S2DK59tS2_#S@kXN8S<rGMwkc<#MvcMAiC8l*GOj1508o)A4JL<rl
zUyH;%GGA40wnCvZOtwSY&ir$`O1r6}9Y<eURyszG_?(=%>mZmkT(i<m^GR5J>oU7v
zF?2l|WmO4j_W^C18#J}fIFUa6I<CW7NobH$`eBpY(H%a)rbzw|FgguuALn1+pKytY
z@ad4-2)c17?5jGLC?$lhUWRBHjL?^$$RQL2@7M&zLE~d6OV$3+rCN3n&dxt=iAHx_
zv<A=mW<odT<6^kSw%K3Nx^dR{qS}Rv?)~o;ACytbhLHX;rkM2uP0IIx*s`yX-imcI
zub}15H;dJozOx|<$SEOu5F5Jd2ErYt^#x7ylR{8sX%&c9a7TuCVQ}>~>`UXUgvnq$
zi#?aUZMg{VrYCyMR|P(O_1X`Hp2@1LRitGnVpd|I0w-+LB&XPh8qu2CUnS@2Ie64M
z9oTADYA?QKfl!K<?|6D8H67KWPAeDR0bHyTf*B_;N12xlCxNk}*T#p&VQL}3N!aQd
z<=@NG`(f{-{dfq+o~?i6MM4e&2W_P+A8=H4S=;S%(yl-^rJ#D-b`apB$*YUirUbZB
zJC;-=b&&y5tNl8*Sds4``Pj%zO>|pcu5mr>d|ozB$7HM3Y2GPC)DDm4XiXTSyTX=o
zak=?pV^a=S;cb>)Y?xJFlZJ*aaa1WCW8b$Q<I5dINdqHn><tGk<4zV8DS^jWur+yQ
z70jxc*&3Eh(u;6yHfC)51D=D{@-kx?Ulz~tEzlkOkITg!PLl}dEXg+OfOg2L<dZE@
z4rKrYd3_nqL38?0<T^|<<qXGhaW;-m+I91mnL$;CRy9Gu=3so09Q*@W{x!Qb(aN(?
zF-3M*UoSgrF789NF`6yb3L6!hDi>=R@`xC9%x$d#0}H{|`8(;AE^6FGC0)s62d<Jk
z_`9Ltl5ujH&8Hj6nLAN1(wd84pvR>9lQmpxV^NLI!?LpY=L1PCKGrP}4^4hB(s)-a
znsRbU>`^Q8)B{E|Cz(azb`p@1i5D_gicLi>rBh?LeuaWgasm*B870tXR?+sYx!;01
zpXuKyL^oc9C{VlFo*DVbK=~_|)0_PO#N*3Fay%I>6FZ8Vyh?@0xpL+(4f|e;)j(0@
zJ&;wK+poLE+rdq2Ep8^bkPvPoWqAam(L8hQoWaDNtYSG@0lSs42WZ@&H@S)E5QluF
zMkUp_nxQRh?d*@sh;SyIs&tR%V44X~9_qyAX&@%BJDy!H6^ENzzzskbTN=mWfbEOT
zOxpf7yz!6|vk&b98$CH54&EoFXFwv*iB~Z?SbR*G(8G7I??>OPD-&tX#KRccn;v}K
zZ6$?K9UX-ikzFGwLHIjAhitbTKf6V$u(|_@6M4ClMixGc`8aFm(Gk9Z03Q*<Kf_~3
zJofw<%-!sNT;Si7F6_n@Kh$U34B=c<?pe_3TK2abseV!)CRF$InPa{2X*J|k|MIu-
z;=qb#(k>iAD!L+?&>&^dH!i5y^~f#0DL#gY=(bqs`vl1T2?=`s>5#qrVa~Ndeacr2
ze?w=`6Y1=@=K;z|ZNQ7%dHF$`AcwmA0oV|ZdHv7s<G;XyoNUbh>xT9hESQuvjOLrC
z-$yJZ*6K!3&BH;ru~K)yVN`Oz+p1zyPWlTF2V{{}8nc@@0~H|_!{TmJeGz4uT6?G2
z;iN_R2aM9$`fBTXcgiVd5ieFwK$KnuL<KZ05{v<FI<e~$t|(G8mJE%;=f|1V<^Ax7
zi4-S9{_<cbDQ~CC>+LYc$II*KV8QHWTb6IK+wU=Zu3KZ6co5XQNlEix#Q(?GJH?0+
zsN1^D*|u%l?w)Piwr$(CZQHhO+qQM*-pNVs%~{D^^;pl9`ZdNkXycebgw3Vc;X-CD
zaJF90FC^P@rfGdT`|?mh+4GnM(7}Z2z{82vtmvM2QLxXD)<pE2F_kFQw7;H;Ra{Jm
zkYB_h`LZ-fZBHjagY-~|_c-`B{q#`7iCbzmQzVl;X9D35nr8V45lgEpf9smy<Zz`R
zV%Wnn3=rk%2kvw$g%wqD1lGKNPX>)B;UQIHpUKu<(s%_ah}EVvvzM~vS(RNps~#6F
zteksudOg1Egn^Op$_|sgy)5xAx3#u!PI|A89eT)XeTA{NIsc^JvMy9tN<-w@S!T_A
zhFui>UBf)R);fnrT!G_UlfXYAu5aP>Fgm7XYNv%t-(*X!tftKHkP|=EW{|4)H&qF=
zgL|Ab4xgmE4m{|WN3R$m3~(i*r1~SuPj1H-?wHl;=xmjlBFeRSN8h#EHE&;k!^D9p
zg@F$s?FHB6q1z^gir4jq=lixHEOH#(R7}5&LGu;#(@2xxqeS4%vmYs)Uiq=O?NyF?
z?4r6N@%f^hDlTgO(8gu1+kSO-wI5jVvxsz771Y$+<-0DbqBXnf(3!jei+1ox0Lbrk
z_BuJc$666fymbkx=$f<c+?a)C+??Fr^64`|rw>%in=~D?;s267+C<gZK^y6=Q^S(_
z#Axmee_*V@XZKr&1oLkoO%V#<M6s=Quj%h?p7qW~vFuE`N!;*Jnk?4XXkB$koeXv$
z?$l=K7gyUc@oQ99kBS08uZwN6Rs#XqKt<ZCrAfQP#s|kn5AA?|iAKoRm$miZV^>1L
zmY{FwS;`c5E4&)5?rMz0QfGEqPI^g4iPlhUkWVSxYTadvVGFgTS!}Lid}mNyv59iu
zy<0hBk3lEtIj(dhleQx^xMI@QC~jzJM=BOyNO%k*%L@QV0t&#qz{&%rTeMS0ov3LT
z;JsGY(|25`*)QU9R8z}52Rz{@d%rscitAZNv%C7T31DDw0f_HNj!e<XV6S1a^L^=r
zDMs<Au1K5_XE>sLq3NCFu&0BO!zS_D3*lV^Gsc!#Tm$tp$Ud)2Yt4*aSfO7TiMwnR
zm}o=eqD()IC%(kLh8P-172YOq?{YHGbjg&=Q&!$9zh1L<u7VvoL>>@sW7(+%WN~=H
zbwteMSVE$|@npZ2Ua`88KsAjTplD)?{|N95hGK*bt_0=Hv->n9#r}BR+rwO}G3wr4
zb;iO|r|w+3F=izMlra<9I|_MBg+S9!IZfZXa6>>v3>q?2Q$W?ZPKJdS-iK?jXeeN|
zT^nO<l4U!l1=b?xTv@7fn5OSHZ(CU>+BRKYE0qKSd>R61^URi3Kut+8InQyfd>jc<
zJhg%Aw}Y03o;l6(6I0Hl9015Ep%&N7W!Jin9PaN9#3+fz8D@3yZf}oX!A=+?9&r<H
zRzb)TVTel00S6A#3ygN&A}L$HqGn!4+q`Pza(@v&UN50eC4<rh=;7CL2O1l$8p$@s
z_Y3&oO+Rz11;X{G`ZPRBQ~TqZZu7Ux@_4QXTa9rG;-j{ZzkrM+oLoX9;-YERfF~9j
z){cX;iN}755Rxf-6o2;2uWrvC=TJvlEolD_dx_zmavG<>#v>)KwK!z~$h^Ovwmfr*
zK20kZObfFMd?g#3t6wOW`gpa4lJcfBlz7GspTj>1k9wNGVy+K)<VZblva(K$GAt5;
zefQj1kUH>093bWtz#*Npa0zOvA_Asg0D`(Jextr4Dynv*?uP63p)5NJCF6kI^lhbX
zHU)W(NEChSFte`dNVo&3I3mMQV~uitR=3AXpr`S?k#xxg0u{3#I^YiX8v`9$?g!CE
z7$`P=%jab=C_LBoRAtKWVsT@D3$=XYHU*zws~DgPJ$G>6w|XS@cYEB^sQ~wa4s>+(
zgVJ@_)d^I)?fwU^40cK-a9u{%1qn)l-LOlxu{`QzRVHD3PJjNfD#lcLTw`hq8i_m*
zv<LbK{pva-p&N2)2|9i-m|~hu8r1NOpA7vVak(9Ssk!3Kjasc}B&tm{Hsj%67d&i%
zMda;g7Z%7+26SIc7>*|(A4r(|@<Q5=!nIQU_QJSALbCxd!jQ2Ua6Xa%;VcqUp|~_W
zme54$Kf8xaLmo=Vu>*ftU5S~)UgA3(X|8o6^WRPxj(*V?kWUTWN(%Wsjmk?$r!oe5
zMfejVHz~kr;acL|+6eHG3<T2{UCrIWf$l=_0q)`V%!*J$*ZumxYi%pryX{Qk<2}a8
zuo|tdApPv?<(pkmj80IETCe_DS|h!kX)?RHde<pc$wD~+KGO7N%YNMf=Ndun9Eu(f
zX7#8K>P>&g(_<LVK+9+GXDRmMCIhLtIF{t|kQv^&Nq{omPtW|KOa@}N&{We+#h{sh
z9ziqtP}#K(_b!UQejT<S*RKw2BLVr!|5Nd2X8ONH*o;j7-v00N{}23I#+-CGWPP})
z-J@};20fHEP~@1BnU>OF5w|)$N_Ut?G<GD4hZK*#CjPwLgz-ZbS4^mOX3UI)0FJ&9
z>BWMFu>r#D;`&$}o<G;501}Kk;cQC-69AL25GaPEHR(4VGxGYnKiFJsjmm&UkEECv
z078U%{<u`-)9K>i%CQp`6dAu4GRN%ke13d?XRF=mJja{$ar-{``kLlA;3~)o1bi0{
zbQ>T-#JZkaZ8g7%Ltb@Gc(-5hQmuSQ8`iharAg$Z!klaW957*1BSpO;olFeQR5$7^
zio9weH^Egr?s+t*C%>uiB`o!q7ZUx09s!{^!tmtJbTK@j6zfe`21%MkPl0nb(Qi)*
z#2(MHz%3V~7&0dU0ruhzJOUh-8-^js4nr(F2l)ASnOtAoZOP(LU$7yNig08IW~`S7
zd>My$LSjoW3AUgVM;P52P-a99|DRbvOhi6R4Mq++mhc^Z_%Agg^8gk+JOps_FzJ^C
zG!7GS47wPjM^QYkj9mCx9(HmAA*F@hiJ?cb5I6#Of%=kOfk7neU+P5IBb8`w&8%1!
zMx~3J!bbb*p)Qp7eMIg#nNKmo0cb8IIG%g)0--I%!O3)Dk#;t;e=HM%N@hL4QfzDv
z*rNTu9N3b{!pT_XwElxuK)nFPUG=%*Xb#BZCh(b9A54hFEOUDNJ`^Z$gpYH1@#`1~
zTGD%O2Jt^QbvQc((>g5Ueu04*-!}sce{BE$njl!2xiIbo792Li&E9R`;?O6kiR9I*
zmBBK^75)M0BO+xX=n}fMQOv~ucLWDngOAL~lYS!UUOXohXm{m=OH&?H(w<sijf}M`
zOl}7I00Q)067ozL1^LV@nVvOpLM(_{pF2;oE~jhcH4eu*x-`}>SuDY(QgLpX@@Ztj
zt`C_2Hy#L;1!O=N=t&B&W6mPU>t$!9rq*kX8gvdOB`Un@l6jz`WRpo<J-0G}81&sG
zs}yFG#`!92xE})Cet$mhsMwyO29-Q%7rl)_22e3pNwTrg@RFFYI&iajTlPmzLZ85?
z73Y0s<<UJ|l+MmLYV?1mi^#nHXzHym58U*gZg?%Z6lVAiZnMZt3!|cs?QiXZXramO
z=CV_zJB0wm3|cBBa2P~Wz@34*(?uGF&qWW1*T<Y9n8q0^z9q)*^e=~phKgI>F4>t%
zzjb9#<cB_~2)1=2P8_1g17#DqYNdb%VC_u8pD&Sk--g0KWN3qQ09)=h&SP^7acjF$
z$n|4Uq~=Fo0+~W9D=i&GlJ3S?;~#~+`(1(UpyoaP8Z?xXU2{b<hOAi|#?a#Yh^!1T
zLK4pvf|c`;D0P;PA;`|<29DjXkM2HNPfKW!!u?Y=Fc~{5oOt|<&IeaR(!YNzE5gwv
zITHq?{R~|wb?Kab%PfV!f4qiyQZZIhF;?4`P=i`|jNSaY$No9!jrjT=%0_+kC8<h8
zE~tk5jPW+x9|D^)k}nc6w990V#;$Gl#YBTe=#{=5O992szSh=?VdSpam1R$-(waNn
z*UI4R`yf#B@H2eGEQ#jq5O%2Vj#Tbn8oeRu264sN!q;)@dh?k4xm8uO@VUHPduikX
zY7)HiPs`~>!@4rxvCg0GkBf@3Q9zwOU1Or|5D4van!XV>%h?+hZgd~^IEWl3>1iAD
zp}+4dzB{-Ani1r)4|YE!BBc7JKTTA8h3qDWZUf1!<uE-(C`r{C=8QBW*v+9$hp0=j
zp4_fkj#b6_crZcw=p)}=T)S8i;K!mdVk^NnkcWO6fCN&9?YpPZy6hmNH00H6^*1A1
zCoERw_fLR%tD(T3QQA)CrI9PT_7Z=B)A<=sOkkSvy!MR}!@}^C4!Zhz++qZVLTIC@
z53Yo5A#_m503X|^i8exvU=E^Bt_dC`^e4~R{QfBG?Gc#=G3SWu>mJSwE}TLrQp8s6
zoe@ef7*>SL92Z`TR;@H|m(Ape0(ki&Ia(zBpi94?mY-1RtT}_dqBi#7Pamxgne8D5
zo^EHR$D)eGh3TF3R-i0l+mSHP{t+t=#u&7LkyoRg^m<*yA5?*$WlD<5Bml&cLEA6F
zsbvfYC0x;o=H@zJEFqH3Z20f|EoW=m1=$Oeqz?Yj9h21S#Vz<khj=n5NiBkVe$jD?
z3VfHXPhi+U9uPqOO+DRIah>T%hxCsV{gfp`AH-3)!RDW>2O(`f=w0*29ff{^#`nRz
z_p7V2hTR1kTY4RU-+xZ{!5Ss*3klGZp4sh0#z8TucU&AB>9;yfN}em(TsdgzyKfk+
z-qiP2WZPDmbXApSRhi2gY=<Lqn6KP5SBUO(Y3b3B;FZ*F1;Z-f_e2R8c<vO81&%e%
zu3ZEze?zj#Wd<lTiV*7Z0jnM^4t1o90*|#m+W>N6NQ`x=*Aqe~T?+;w#brTmNmX;+
zO6*+W9=E=DXl<zGXnq-fYJn&C=}~s%py3M)YymSehN3lE11iRIJo|V_ESR`)R&0%(
zQ=5faO4Xo!sn-u+w1$dy?5hhHB&gT$;_$*Z+OcmKENCtC2z9I~y%{hn^mR9A-9Pha
z3d(PSylcu!Zh|(aHh^TLfP-_$9eQ0zN18>{Q;{2sjqArp_s=CQ7c5b&{W$Vs*L1{%
zbrP0r`?mOnBY@jhj-)|OrUG*YrP-K*WqaNOx1%guj1d&zbb6kd8J&yTvf$PC;x>3q
zxuVMa)qu(hwuvY<s`!mo=_kpZ1i|kHrKym6f`85D1UYsGL*wMiBiec)7Fm|}FP+b2
za##04<hVXgQ7?3vu~-aY*?QvpCoQC+sk2bbk40uwkl{*O%O}gL#`CZigY^?l+h-eb
z%4G8UGq6nRx(xtYYO44L$se1x0V`0FwAY^p3UfQLx@-sx@}m95^k(U4zS>LOZfv6C
zj@ddVK^DNvh;qdOyMki--V9-oNs(4Y((G^g{0_Y!m0+Cw$|YrEFY-nv+6fSD533U&
zNUKGiw)KO}#m#&(yB$oY?foFU$~4;APmJh2Q<qzT`|9z^_dMp&su=NA)(hJ!mp6CM
z>(M;oV_kUtrGpk;Oi}-#RA_AVXW#k<bl;E?=2NRsq0V>3KIoK^e?fJ>s<i%N<Mp5N
zLpkVK{@cduT3!8r+;9IwHX~wb*24ygb$Zxyxr*Y#X>CqE7fhJ{i)d88|L@wK)wKBY
zLJPY<EKS9mRG8g`z6ZIc?WJ~QeDyDo&)xoR@M!NWX>TlC?nrSrof_waYwHo=rHx6Q
zy;ds#L9meFwcgW%H9af47d@I{^rjv|2fMeci<^s=O}AI)>)~+x`1ox4sI*Wqwt3wH
zTmL3dX>d7msQ}mtrJo#<{8Yx5x+bgBr^D^4CULyK6Bx*#UJ!c(o^<HGEHn&_^YQ8L
zeEO`u_nDBLAHD<RIW?S*XH~*90t7H;^d6k4iDIaO{%s)QECe8X?(!=?_fe=nKkYOc
z<GCIfn^?gMYy>&Eb^MQHwev_4K5XSk5aFM%XW3y&ylC&JZQm())6Mf_F~xgKM=K$G
zXuC8%0tQZ3@;oN`0~~KZyEngWjVBt|JfL@-%9u`xNTY5<oaRXjhUo`Fl<<Ym7ZDTI
zlzU~5+UVDg@*fs|dU}G~5D_(!2r6|9hFzXrckQfve8`JSi6Kq|M-FL!{w$2?1fz~E
zqese_M!`V^`QU6{xrohoi4<e%hI8uiQ`=JGuhJ#9eIXEc?lT~c*;YZTe)6kSk&T(%
z&HJqwy})EfBnRNvG&?)2zHXvp+sHFT9huo=_#pFd5AV@I?BG2$QVKDy*-AIsezf4N
zYLx;7rrGY6Bw?Eo>e?$KAuS3tm+o)p98+WEC;?Zd+9)ETpw?l{^&|ioPi@c?;F{WE
zx?;cq@8aW6XT;e<%6EDoRcKpNDI(aDjPyh|8>{7LNu~hnD6un(Pd}=bkJC)Z^GJNQ
zy}B@h8+oEr=oymXgSj>f7|1Bc`1J4-ihfEXeemT8T5%zg#)+xcg0Nxg_~Q4dM$Cd1
zam6Wk2rN@y5+-$k*x>qXSFCttdlcYUz&blWx1r1G*&G92&q&y)_dSJgAE&*#Xz%mA
zJ|KXpUa&b=PJEzS7RWD|zP40>y9I_J%mdflFO`|ft~0JD)l=Ij6-g#;Ho@!LsA%cP
zD+@l{SaLP=&m~$qXh8UvbZuNBzzBiau8Y)S>OBU=U0&l8q*kSTQTZn%6=XY$WS>2E
z18__EFeR1g9WV1Ul(r#{L@85;dbyP%l`kusi7(<6%fTD(EJ)x|k?vwLQNUn8?rc2x
z!b?_E!J7gfYj_9>IZn-wdpnKeaT`2IFO*R8)aWtPPU6HkB3zNdf@<>sJ>#6tV5U2-
z_)}*L{*NoL_|H#?icg5JPQX*2;*yUL3G)RMWPcKtRA`f6^dT!Gm8$ua4fn(i7Alz$
zO(eGMq~!GSO}sEo>z|;|J?5?&_;_f7&2`j1Ra@0UiP0fs0$t1G>Cg^<!t4ha$^~XJ
z@oDOrD~ZoRlWeiSF$fnjocO^saRPm~OUoHg$bC3itMJFgkPW0)$M8ClRlsB}jnb5n
zZnPR?V*SNye&BkWf@fNQBdh3&ZuHOVnAA!MJyfoGfB|Jv-|#zPOXJw&?CSu*`bDdr
zVg7K2V`QwKVE$Hn3HX%&yoW0TUL5~23&U1oe6%HtVkjR3z?~>@gp-!3E;#dM|KLXM
zBBRp`Ygz$?p%&>pglY$T!y-A0brztrBa4EtLBWQqxxpuAD)B=%=n>r3pWzM%MxHgE
z5dhr~=h29b`Qex0V$q&ej;V*q!ViV%ObF0}*!~PBqD~MT@3qiZLdGXcmJMEX852;}
zGa1ewPZc7jz!ULZ98+y9LNDdM{dl<&^Mu0f70Bv}#cwIg?q-T-DgT<{exSHKjCyIq
zwm@n0@)Z^Oi^&>snbI72SloQ(M<aS+K_?ovWWr4}5cE_`rznORWWecs3a2wS-OG!L
zI{OMDC(kf1t#YdV%|OPcR+i1klJVXW>eRNi)q>iDnsP!v0dtA%!x3`8e(B%l%8LUM
z@J?R#W1;7T<al8|TGNR;ME7-AQUDYlAZ5Xe)1Nki*=%T0gb3{gR(Qgwp5KgRxj~rs
z6ENtv;kn{oU%^+wTUo(2ZE(M&RQyJ_OL^AuJNyKMy<Nr(tp~RBN+Th)|FaxQeo{Fb
zUiOQ}AWV+=czKQ(lAhMveO?}CYQo#@=57#(-BPr8m`f*n!TnMG3)s=93xqb86h5SA
zJvZ0Fzh6_1cDC?O|B-X61GT*aQb2p9rzJ*?OJi*wF|kxL$AW96F;=`rtZa@(;V7TT
zy+}ezP5!8)vC+`Tc9vZ6w@gx*oHX*mg$^{P4;<N~<+M=psE@S9!vjZ^N>WQbVR_ce
zO1Oc6XS1(q;Yw@jp_+L;>7!|S55Oy&#4p94N?P_FmY*uUN;<p>ojBYA?$)BEY!Mz`
zaTB+^Gi5QewEzq3na`Ma;RYfvE{L*j@Ij76m2lBXFKe9_z0{^X<xx%f+VN4amF`av
z%hum*);dWiRhSo*5?|n=^eeWFd<9|N78-=IpJwQG_Ag2a{^imev;dtTwY%cUKJYCR
zqTPx<Tjjk~W$_rBf@%v8(5~0mM{V3n#R;Bs^=A>c^<N+!qGhB1RKi*QLs$X>2lIa|
zZJ$<`blhS^{9gn6vIjo5I3WdIR#LVRT*RsZC0MN_s69QC+9Lx2M?!M-*DEu)ozUFU
z)P?-I76?3WUT3e3*S4QsQ;gly;J9llndvQiByj?Sbit(ZP=eSUh(2w>r1292DNwFY
z$JgmPSHQ6u<~Y3sdZ_e=`tI-B$@lRlTqzW#ofJK%b{NPgU)THb!9gk8sm<GWXl}2s
zr~UoQz>RA__aF4}BuE75O%oiRr^V3H8V^tRhfN5;dUiqtLvSQS`*eb-f-dg*L23kg
zo=9?h{Kc~Q`HLf6*=74E1_H(jfbJmmCxAU<J5p>C^$U3Fr53sNo{HA$K`?!}vG4IP
zB$OyW5Va69Ka(f>cs=Z1pCQz5paQgUqrx}M-f%HJ*2WV3UU7P$%&hLhtk@|c;=tTt
zcx+SC((X6~@0FU><zO47<j+};<50H2z80q9cWt(t#yRUt59`&mE<yhaMLV=038An$
zoFQ;miLd=?_ZTuV(O`V3P(4KQczCIDA&SI#V+cgLY8H&y1DjTwCO#qm5<zl_5OkTr
z<wQ3(HD+;eLN~(ACY>h{)tXji>&jH0Cf*_#X>vr2T-8#g=?$fdHXt<ImB?ZY31dW?
z!ir;p`M<o};s~XLNgGO+U?3Wp0f86)q!Xj#bRrZ=62$E34iU~td+U>NkRn0-?8z_e
zjIHG8?|b2)3HReo08jylv^aXk-Alv$i$mlE<=6l}w}~A3`)rV~$9>S?`Qi%fHrq?g
zW1M~EP>@*?Jf$x1RUM9mKldMb=>hz8H-~1l8^6(@v=NodH4|Kls<m4!E=un|I9~9(
zy8y-*%G%m!&Oh%JptPRG$G|KKb|#1eajR^OpRAc1zU2m;7)4(9*X_3b{@D%*)d1Kw
zJOv0`rKe75@g;tQnWkf-g0X^Ys~tctx73?zpuU^s6-?Rl*ZgqMV^W0IZLE5rA!~$4
za<)!8%hSx|-rT=OO%D{O&9n5lB|u9xoRe&HM>iaKzaFlp7yu(b5kX>SJjshu_!B&j
zx=xUc2*(s^t1C6F)-vL#t869^_4Bb!o!2ckoS*SSRCc+V`x!tM0lXDgqUDv@m+Ke(
zGQB(L1)At|#sF>J$34=?Xr0C7D>sWNnB9RQ5KzoA__*n^ytkH%Ei6`}^x>%^DY&H|
zMQhsId)4QxHKkU-`bxFiuKE?THe)8-Y{Sm<YSvZyE>%ka5woEN&_{6*`HW-uC}1AR
zq7SsBDd2%5a<PxdqrET7ekfg*p2St0*R;D>41r-{dwPnk)9Csbu>xx;uip12qpsC9
zTDIz3y%TK}T)Xo#{tKuC-kUbgHY%<+g6$>rxz75m&ex}~gx#yS)OhS!Em0v7x4`@0
zR(e#Ya-qgMDxL{6Hc7*LE3F@Yq9oL<qLT-(GQ<1VqDGHvuD8;t6e77^n=2v<dYm0!
zCN&yslv*zjcj2D(v#wiDh~RqwfBXE(Pq2r;6B4+JJELj7&<X5oG(`E~f>DuW<O)D4
zV?k1g!vrPZQlCUL{Z}`dUx?wk>umXqXd12aD_OA91n*(l3Ye4ZeAFl`Y%Q`vEU7a$
z?k=9~&o*x+C_u0+R7yd}fnu24;5_=?J(c^3Dpg#oGdnP&<|uck`eli?_OsSLFa!2Q
zp1<t+#2|>F?YVT^UHR|`*-5^={K1tXq3WpYqpV}JK<JBxgFvu0i2*(qmlols4>Yg;
zbX}Gej}(s++`sR#m5_nF@uPJ(&CTVQ#L}Z=wbfrOGvovW&_E$v@!u35JGWO;(Q$JT
zW{J1W1U#DJ!xYh_?R^j7BcxPJvhm1=^Du1GaD~ma%SX$v21lo;(~$4czBC_p@AW0G
zt(1Uo?R%@!Jg0K_5x>z#iZI1T^P&DKab#4xMlVYE%9CtS>0T#qCqj5(+Qv9YTV5%x
z?ouwQVR^12Kd>OrPzPE{{#GR7CnWSEd&ZnlpU90rhIh+`X-6o~Klsd0>q0oeU0cW?
z3lbwS6_A$5ClwnISMg!!UJfCw_iDaDLb(ZV;kgllzJ;{oZ@vx9d7h|rvevo|FE~q=
z@BQqJCR=W3slv-O>oUJ!g$SxUjdViBLKJ7euKpZAaOh6RwV%4hk^vU|riYu8w}(of
z_(AI(Vq|$@R=Co0F#v7`Y6hav6;Sb%SOUG#^DaB{Cpfj0@Lq)=+k8a%9ETLX86iV3
z*yldj&!1+=fAw#DJ4QgkUl&d<PZo*=8tP0bppGFBayV3h$URfn93o|nSvi&c;NH>!
z)!@KeFP@@778&L8|Mt$(5sOp*3y3L@SBA2o3hjJQaK6~1Sx40;boW0^``Nc@V{A-y
zi-GvZhLuuYEbFYLJzPc+x-0$tF1f|6gi!)zirsNl^;Y<P;h04?Ojb=?4J;2%!skC?
zSIr=cB#m=j_n}igi{0b<iTPblCSQw_ivKMU-BAuDy5h}{?QaaTpFWuXl}Y>76=(vm
zL5Z6&-6g=q8%#7U4J~jbW&AW=2R5~<#8kepc{}|lZ7cXfIwI1^p8&WKv8~+-xaV4Y
z{~<YTFJdwG=lTU%3y9eIk7fQpIM^8Jng4qNubQUgW*f@?nbft!>&!V5x7AD6Z5EkT
zTH(Y;fF<LG7=$gvCt?;~dS3y+lEl(2O>u4|@vQvd?e6xs?r`SJQ0C8`$6Xr}8y!U&
z$%)hl63tqogb*px;VG4wHHGjDM8feC%O%XbeoR<BA77r0CQzbv^&dZ$`RHhOx;!Fv
zZ+CvbKk;a3Npns3a<V1o(Wp@JDgCwOD-3KFAYlHIk1NzDIc=>}o~`<L{4Oc;P09Fh
z=-1Iis}Y0bLuV)<X5X9jaM&Lb%G2wLN|^*f!J8ExkMre3IOJKS2xB%SgvqQ+B?y(4
zrTi8sfFt4incm4lB=MpH=9`{ojpq<!N^^i8V1im#1k?P%>}h2o<DTa9{Nl2@q817Y
zl*O>&M(>X{_C2I974ko#I1~bqpu4%UVw5J2h9D3%@mUz}G7B%zRIIg%JovqP=qIA&
zU!cB@QA@MdKY4Kg26|XPPTi!m&8X<2u2?p%-=^kM5WbbW^Q~*|oSF>&_KMKWRBXk1
z3nH9p09JauIB{sFI<?kG*6%=4t*yK>j=0_k@*%skHtSca_CKHc?!L>E;+PII@oJlF
z?G(LhOs286*692T(0e7deqwEt3HCe=Aells%KONU$qowYYB+w!&gQbC@~kI`{(bj?
zRG?w-h?qI!1=>fu>3@IWaBS`R5x2a6+!waHZXspK-iqG3(Rvh9Xf*+TtPP|{Y!4wt
zLhKKoZLK$|I&G8ziJ`kra+Z3&24qs5@V0UQ_h_L?UAJj<&8@q$FWR{yORm}5KxwCE
zY0+3;T^(pf=GHc3Q*&hiNPE1z4V?D(AvLHI`?T*nC?rdYMAo2j`+zm~G1}gwZ(lnx
zm8wfPMvETYxSTMdSaAH6u&eq;^nGl52`tlM3_~<by4?*tk}FS7kBK1&f%_R<NWFH1
z-Ykp;jqQwy*P2swu3I|IFEikf)R#%e_9yE-HII|l$t$zf@(RQ1<>SQ=*jrW&d&T&E
z3j{Ej1?s99rA2blZ@)e_RWg!RIY0_E(9nMT(AnA2)@nO{rUBP#B20{VU`A6E868AL
zQJ?M>+qyne@TXsQa&G4pw#LoM61$qL97tN4pW#09ZnK3Ie#Y_iPDY?}CU9zyaElc>
z)kXQJ@~fim&AYSZI1J1&8_+x64g6__0(;_IjadB)_Bzr~*vC53OIsQ0YrZ}9w?8yO
z_joq>W%fA4w(_F`LR%BLZ38R=O|#k##K6(U<_dj+nV=k6d=Fp{hb5l3`P+Flh%PpQ
zjMB$8114$%ag?L@0@O(n<Cl9JA2h{sF#|~II)PBu|LAzdQ|2C_fS9gYGP{ebH1X{b
zgxuI_)lT^|TTVnb3Q&ePo;zs$HM-kHWzHNPG`n~0WWy8mi68uEy*zK@HV=-gFE$48
zEg6(O2zhB*M#}^?7tLA+8P=?%-)F|rl1^oG8Z$9J^#LQ>GlvaueCz7Y<!;Y&423Kz
zIINX7JfomZ5c4W~>~-T~<}>1QP*^dUEe`Q7F8NaIbCi%k3`U*3sLxjNz`s?gQuV{1
zEiSuT*EIv~nmTjLQ&~7N4!_xb8<T#;neIsYHah&wCD)qT19$zuOaHhFU5@f;2w$iJ
z#*2Qe{eS_ggSN}chL>2Q6k8t_t2VCIizG73p@?>O)C2*M1Zi}kD0YUWQ?&pWgg9t~
zj8Lf-?1W%QZIHs9$C7J-=hPR{hch2`nmh&YIRKx0Kmz+-)Cde%+vc~B3GMa!3(G{C
z{L1k-wxRD^1rGrPTIe`7K=YmV{RHgGLB<5NGb=0q_Jf|opRwIaH6W;<GrTvcaYJ9k
zRytT>z#Jz!P0U4-W8?fvZEDD*3EJ!Fo1p7<r(jaF%*ZNz<6wKYBT}%$STHqz{@|TC
zQ6>Q~lpm({r8LSBR3UuWG{f?@wGeFl*;hO|em*BxOL0d8K*L>QRZ@z5FZLLtQ>;si
z@L1$k;2)G)D{Eyz=hVl6dYf4xAuc7n4SML2ix&_t>1dxJC!$OMu~_ql06DdV15hk*
zJ~8zt;<O<Ohxhb*!oRWG3$m+u75*Zp9y?PKUG%Sl-H@C8Td%MV@yS94Aj4I$Tgb!I
zCj95)S%#v{QI`g@CYxuS6NUth{GStlM%x^llH>>q*$lD;g<TQ7L)fO<Y~UCGH04hY
z)85bRg?!gs@p%v9H+=~<Kl(1#O)IG|?~RxK#$^2yvt5_{q(IfY_w(!ccCnKds*z&y
z4<CE>7iVcAPpKy{$T$ybQCI-n$rz8;6feQgBm&(m3dV!OYJF~gmYQDg#`0P}Ch#ZQ
z-;gM>2EzRu?6&#H79@Da4@gK3Y`d?WW++QYEf5aew0TUv8wS6atx~Wv3bP1~fLUM^
z$V|6@y>gB~Y89ncI8;Y?1_jvTGz8(7SCc)SfQo%W;^*!l-{j1`$9%u_ESD?l+|W$G
zNZhk3>v|ZLQS6ArhT;8V-Y+)Q5_-7@>H&dLEHB#gMgdh40AE4+V}!I8R`+((Kq$Rs
z->;NZe^bxTq?)RwQQ2IetD@t_E-#;Htax7>99|870Y(0{J#<bPL%1GUchpyG*^Icp
zG`AJ9X@a>zGHJ!t6PqiHtIFL~UN<v22|UU7#YA0bq%f{|$%EFGAD9U^xJGyw`M1)*
z2QVUBPpMe>TwV-kw<uV0m(A<j&<vcyA12PRRJ7T<9oH>^tsl6fb+XI<*irnyDYa6@
zHl|Ky|KPOjtpDxqe_8eakni@yq-oIc!Ffp&12gTnSTRQHml7}c`-q9{1>zw=HzO8L
zT3z}7wunhKP*W_P3;X@myVu9ty+LFa0Lgat!EuV&^@J8?f6&H*2?=`-^W-JP6%9nz
z=1A(?zU;>vy`>2MmueEHm+SsNsU|nAiIM&1#;vCg2XX<OJU2+ES4Xoxzpsgojb8Te
z=iBDJz8k9#oc|p0?2uYB@}5;9X4|Ec+OpZOoVr}dH6&q#BlP*m?o*>WLvoDgXU->%
zYj_3pWx1;2%vWAJC`X4#?QoY;zQjw|7FAjrf^9jW0&?}UO%eT&farl9?0Na}kN5Vi
zr$`5SQCJcnmvd(4W^hsfrk+c~QH+i+c#Z~8)YkYzkPobZ_~<o+Ytc<&CPHi~QDz<d
zPeoG3J@>{_!#40VTQ{FGwioTlj&WVflN<ONWq4ENcqVP}%JMrW?x?57<chusGC}VT
zkr*L$Lc3>BtQ+Qdv6iA#04~~gW>|}OfNN65;o*~wP)sV?n4AtqA?wssP>nvh<@M<h
zorGj;jMByyKi<fTTEq)B=`xmp$K09V#l}lV^Lgk1*t7w0hPq$`_jc&=%xM4}i5Ou9
z{RcW}<@L;ok&Kzs7Fu^;c8Vg|M4g*9;zIIC6H{bBbwV@bR6;pCV<~l7<mzkLt_2fA
z(zMsh4z!5LzLyFU7Vo&uBgfrciOSZjA3-^xv<C5!9CjknKRc9!>g1Cbja5nSwG-RP
z;jj16f%2rGbN;n7BSuq$K4*9ljJhX?g6S}v{zt<=c#mV8A9pQcO*<x7gXv7zOStWH
zrEYn6s5K|Z_`qD_J!9Nhsm2!PeFj_pl57nMB*KXj^U3Xev1qCmHVw7j@0{@NWJt;?
zl|itvIs~P>^{_c^&BXG`2bz*ej?Uj23>bF(klb(?)zQU0wlodbsnJG*s4w!Z%Jm%U
z{0IYhIE~3_2VR`><w0vtkF6Lhr<l0<m>|+*>y_ko1bf>m#tBI*$*36vJ|(Wecq;2&
z$^8KRGq61~6qWYw>|nE{hrd(ftNs~yIOJM0a3y4AAjxYaZ*jC0Zd4m8ACZorb}(0#
zOg{93noLn8^;xoqwXKBF{nbru)7OBO!@;iHA?pMcd)Ddu_0Xv1n5F6H=6-n8Wn|CG
z{kOquh5HeEwkvsHg~5BJPmq&Qg_QDBA;v86tnnoR;n5P2<S)!WCONwU`5qnX(Hfi@
zNC1Y~oRNe*<C8~{+?A6uhYBa00&(9E)e=}#viAI&Gao0u)-~FiovXTC4b6%LnBvBc
zhnuOB{d)dHY@qJY-9p=<XXu)g`BqKK+Bw;)D1q60uOS(D(NrxC_YMadh?=K4rdf@9
z`HwDbbZn_xV2z%Yo0HuzoC(2IdBF|yV8rTWQpOIB)zW~H$lOo^-5$X!;53uHSm}}v
z|LK49PwV}EkvJJS{(BPVwuXA*<_Ln%uC9NVJ=H%c3_$UivV&KWLj8dl)+fJ1Ile<%
z5DXAAz)il=!{#Cn6Z;>i1PuyNWn342Xp@rLoLa2Bd1s-*TiTwEj<*9%kwtQn@)ORE
z4Iu&1|51WP)U3>Bf))zJO%^XW-CsvmR-)1Mc4|2Yqs?v~cb@uset!%<rrz9ys`|bk
zveoM9{&+o?()A=8bX6+_QNk!~*GTLYL0w?v6Gyiz%s*-9ijLxTe=I5QYH5GD^!ri&
z@01fI%qUN6V&Al(z1KDec|sO17G@-GT`)bY%a=T=BS5N`Rr+tH`M0WbUrrA!mIv}!
z1bY5c64p;Bg_u1g<H}~Ttm2%<i|+mSa5`CB;#6;+YI>t!^hv0h?kc>J(crC`x@ucq
zAbfA*NUr*C_>kS=#-9GVg|i#{x~(?P;HKK`rmI-4z0LRvbPq@(LLKdE3QLJX>MuI9
zQr@1r*0iPaO-1>%7CoNP)&02KK(-D^y1y+aX)qE8`bd|14@LTJtYWyANsA}W79}qr
z2OAz0asUs&iTg6oMEj$yMeCyOJ)SV^tYaelLm()>9k|)GeL&Ydw$+!Leeu2K19!O~
z?3{Z0A#71U@pUm^WcgZIy;Uf^(p2JN=-$qdQbnPv2k^;lTOAYkyp!%soJ<Ifz0r`3
z61_`Jr`g$~fu~Y(Pp!W2fo@siLUl2zMuQIwj!LKVX8OqqOysHZRIq(A?G)d=ZTxy8
zs@`mK;aMBIpoRAC{DZq>u^sL=0rpLn=Q}Q*MJfYds)=B^M5EKg=&X(tLFn2l{8LHs
zV69x8t3#`Q{O_2tm#BUGbaJ!e-pJ`ESQzEDi;m2{;G{psGFG;@=5ZC93O(9lt>Ot;
z^85tF7p+*5of^|jM;;&UB;|X{Q`~@Xgji(0y=+5-9&F7n97;xOT4HsTj1I8(yY>G4
zD5#n5uVNFqfqX(rl^v-6=GS`FM^z1qN2>Rf=r12{QJ{Q<?hQy0=-AyhT-~N&)rfT{
z;u2agZTlA@j%Q0u=LXLD@WoUr8r;c357Q0RT<l<`dN1z&?yvZqAY0r?nejx#=r%Io
z7EERdasGGyy#+qhIGM@DHNl<^{Ir+H*GJQx9^2TP_!#a-{odZF4)C-sm??Loi$P|m
zx3U0Gw^1ZF)fT!D*HTm?z?N&e?FBMnybu6a%0k6s(Cdy(rXY-wrK_X0grcENy_5Y)
zcV|y=QmeI&j_0d=s!1}6Po}quDVGyqtr_PEVH&WaUg@ZZOl*XoNdXBN^S!$ZPobDJ
zVzx-3c;-Z-O~dkvr$G;s@uH{x8A*5x7!{8U(8H?U1DT^IwL9b=P+t51aLMj%-)pU?
zSZ%3<wATPbvsr8M@T7)Jr!D0}n+ZBc2^C^-&z?s0{yrqq0DH{Xa_|9uw!K_juW!c~
z_%Ya^ltS}o(F%qwQAFHI8xf{Tp1=Ie80N9l?lB-(ey8+nYtmDmE+OJsc75z{)tEDT
ziZHV4^5)EQ?yO`efp;<{SM)DV@#IGxz{p-I-V|*^4$M#rex_WwiIIS56kT$EBXndF
zaPJ91<Ve#b&Y|y-Q;Og%>z|sccLet)%q9Gg08hI^h=Kc_r)QZp2APIHfF8E1v)h5i
zaRkT<R{acUMxe`_xd4}tqd+MkXh15Y4MQ3#l>UsTK-d`qQbro0<-yBs6|@MwEnd<n
z0XSv`?2?C784DoBxo4rwy)nde00Ei3?tEB{RAt!hd{SmEQx*fY$CmTfsQ4S4#m-wo
zVvWnyE;dm?fClSd7s5{j)pNdxMJe+eP05zw0pTw7L_sKC`4<F`&znb#9B}jIz77%J
zO3Q+hBDqqQE%r~#{$YCr;4<P0gN2zAA`%d^lCt~L-1f!1s&MIdscX`1l2<zF$it!g
zkN4EqV!$+nA3x?qdJZxUas}ATF7%JUapT%3v`;1sI9h)6+BgnBF`0j+Giz!a{}r0x
ziVjWeALy31I5ol#O?H$%1wE>QXkx;sC}75(0CK9Prn%)DJSSrY@GQNV%0XBXX~9(4
z20P_oL4sw=;<^zN#sqtw6f9}eI3h0^6|)9Q(R-|A1Y9yY>c??7?00ET-uGN5)S(X&
z(l#y$Y}Zr?EP7zL!GFmVYsUcU*LgX$`jS;X{0Xc=q~!9970fxW>*^keLIE&iQ=@Ji
zI^y&gotvVa>h7ubxoN@Vz5}?_Zu$v%xw&;9+P2-65|T`W+s7np*oUtZjcS-vml9wA
z(8ZzV2z~YET;IFM<P1HW;MA`b6st2qHNgL+M9a`x&$iRWFA>Qn50=|5RR-#?v?b>O
z%eeX$+emnsh8?2hlMvX#n3imUlPocg0by^WpHMUTV)(VkqQ)kQ<~Nr+j2=HlFF&Wl
zSp}Eh*wp?@oxi@xp6_$iJAEN5iW8;?1WG)sQbQ)q0Xv%^DZ-B}pxTwx5?1D5<X&Cn
z81Y|vT%YCbb{v}o*$joFoi&ol-$Q6!APEe>Ggh_6CT%#*hXAt0gM>hC@WbqghWsR}
zR_)?&_YsIwq{vCankuEIP;OTqtTJ0~XZ4SZ$5F`<wcS7c2v;6Y*N~So9!XAnPC&WA
z{JVheH(qYB@;RG2@#x&X%1YB-?HiqdlKW2CSML}$^d_nPJ^IdVu4oH2pk%0cT8q^l
zh#=MIEXj80Ud4Sq03MUX9jeOhGYs~dIe<0H+Jkg-+hHu%4Uu#uA|=6PE4<#hlK_;Y
zYx~%A90zthpfutIMs>ut^grb-A&K{K1=aU@eGsJ2gqqD@R7lV+5O&F%GX+DWx}KI(
z;vwz@*ZO><zOEsLITD-NShf7RJSY*+dq8_J_X(d&Mj)ns%4&5cx@$6|VwoHW*W%hJ
zBUD$VlFa6v*8_w&oALk-z{n<S0px<6jWwQ2Ia9=O8K8MbdZC?#pdfi|LL{S5c?`PH
zk-X$&0oASR(R}4;mJIj8^PMcj0WcolBUKuF!^aS?d=<3!UPXEVjEJM-M8ubt`aU9z
za_IAE;<I8Q`j<hO9GhxDV;P=H^N}#}_(6R$ojfEA+I!BiI6JQ|Yldskvm1%((gsM<
z!ORznBGJHwG8rNe#Ns&7CEgs8xE2nmvKoH5c^CKUkfzB}!$O*DZc%E_R-D0V(|el0
zP`0b@71Op|Edhb!FwS9L=qm`s{*)B>9#f7>SsyXD{g4uJ89`jGtUD*Gnv0!~hXhN1
z03;1^VC6oQSFT{C4P}+2Fj*ChQpq9+XI-K_hbWeLJ4t_}{iexPX_rUfWrg)se!8QU
zOd(N+fHNZF3TFdly6eKE>pr<qS~BLLT-F(GV@@03yX2(NlUDna=uV7)nOU5kexw`p
zoJEq|@Sz{ccKsBq7wWC~_;sjJ;N>a37FnO3U6(PpHh{~mZ7f~|^qZ#v3D90q)p4k_
z)S+i411FS$`qr+Yfc5gPD^$@3Jp{!3tO_|1LpnJT^~BNC=rMZ;9XwfYGTJ2<S1{N;
zpoZ6~4_&|iTEQfD`SzOKXy|6CQE1x>a=i)O*J-}YE%W0;R#-{BIRGmq+uktnc`nB2
z`Z)qv+|LD<w=QQHBNcr95$38F3Jq-Rm=W>#-Su_E@Bb~;#ik!yd>i@f{hS62J#dlZ
z(7=cgqzsBBp_UrW#p13*$TAb8y1v!vJT(_|3Bkr+xcm0eX<3ZL`nj9elraXRE>)Y=
zkl)VuKq%mTbfSp<;<NLXWJ7m1PB&7l>3zau=k^W~Aoj6Oa0?Fml@0d#9*4ZgJs@;H
z0ZxnEd5JbY8cSu?{iHL$!R;r5;oyFZp7DP5h8nK*H0iZ*?=mK%{!2-RC3e)ZYsOc{
zfi^-x<Zy<{M;e6$d71*--SUz>9uE|9>(ybK#JiJu1!R(Pu2g=i{H{*n<Pa?MLkCXC
z`1l_?`~T$aWMTNPHx2(^q)!3holV9}QTcVtO;JsC*V4V29wZ{izHvfCwaLcr4k&7p
z_(bZ`lv4>EP}`8sF0NVxx?~_aSs`9Y;h^Cts{qwgnyD08$TddIq56=0^!ZjBbNOXd
zZAmXSt|FpK%)Fmk&b1%B^ToBS?M#X2=^=z;xOg~AP8%+R>`qs2+uQpz7jbU(@5{^C
z!NHlp1QvodV7)+YdkuqDW%2!5<)e3dOi5uT0%-G4UWm13kV2sRC}f$n(7M=y^=$B?
zzkcFeL%E^SfLfH#?&?HzY@VMQbr9aFND1z-mMBmZ!@tk8aZ}FES!a)EjHK3$6jOeg
z2ZVr``KqoxU$`F_Cw<oK1r!w>iL4GA6CG(H={6!)6-D>;3Gn`5v@$iaJIgI7EqU);
zZpxJol`&|<wO^q3nSsVe_?eh=sxg4xYPNU9140Pf7bwER6WO>+JVeEpPbBo<U>SBz
zVSrkso3~9DHK`}@7lNFd#;^y9y2kolKhxpm7de?z-Q=^d!0|XSETdqfV#rmsOpv*Z
z-z4K0?%}God!ERfU!4K%Jn?Ql?!>1}_Emq$fq#ishHmcIKc0&`S*=~4@0)Q)X?(k7
z26$=x;>X#gLr^S~y+>-4g0|LPJ%!0$b;d@;i#IYt7B5Wl2A0#=wa#4XE~4i{fk78}
z78!a(T2xCT`L;U419cl3MVvPcIfrf_-&!H3yFMBg@b@KK2vYwuYWc8FO%G1u_K9(I
z?L@UZ%&Pb84_aA*>7{*O;|!NrS<W3ft%AI-*6ax(WkW+*@X(7B(q=XLa~s=2k53t2
z+dk1BhBP*{yiYqV{M+3&;`h?^1XGMum#YP|ZGsyxE<37aO~e-YIr0jLvvh5?M|B+~
z4B6uC*n!fFC*CN(E6Meavl>sSSLn#VnPU6dZAqfe-Id>ID-Sp9g&8fnXe0Xd%k$qV
z2}pFd<uBS#zm2RrWo<I<(b5J=LitOkZv^K%bAFnEjs49)I&cnn{~YKi?7+{;BsTFL
z5Pm@5at($nd3(J~GE=dz)#5VF3gmjs#Xe-VN)Xl_3g+f;P#d?IMdxiZ3x7+cBWFf5
zjV^AgeB8Bjf3To}Y;jyDrKCXeO&Ed=J?Cv=gemF>xr$d%wKd`eNO3Au2N<cC*ne?u
zxJv$nlt}Sk$-KHXl<K!{xEpzdHZ$dtYK&mKZ7jXMl~K29Hy<fjAZG486GFi%6MspZ
zyAe6fq^Gd9jOi~bVQLY}b+VW2d(1Y%c1HwmLD+Me<rA$q`HOt>&3S@Ka0Gr|=SrA{
z7*C3uhF<%bPzdm{W+oU|A|aDl&po8X@%Ih<D~TXOL)%BfR;5|16YKjP{&-ZDCMK@q
z?u~qKBc7EmwdEW)`s4mY;m+0zoWQx!ch3!L_VblkZVBd`NI&?)lb(ddU!nRAVfXn0
zn4%jB-tmtc1+h0I_L;UHcAO{?pm&&n=2O{GF(F&codM;y)5I-QGP%W{J*yO*7g;7v
ziGeRZSMZA@v)k2Gxq&<rXt2_t#+WyDQc(4}cGuCgw37?-sy2^c|MpPy^^DmI2Of)H
zKxbax01R0o@UJ3b<#>#NDx$arkWl4rvO`-!@*Y~Lx>i@Mfk>i2{u2y_Qtfl1WL`{c
zW{WvNDUWHweb*Dxr2;JY^UV7x#?xiGDG0JentoQ%)p}yV@Ny{j7iLo@!nOTHwzf~r
z+rPsu?phgT;5zdLAXbe*2wkb5c&3bCp-|ujhRLiKDIO8qsq6~=n(rc!`dDVCS6vr5
zc2Vi{KVIYHLaoGJfW%}sY97UQ4<3^lW9DpAg<o##MZLQZ)!ddyPbYCt5Kq3a8z)+x
zI0A0DnvqXy%cpuML0N65SsJU5cG5dA(t&E7BAJy+j3tylD2NNb{43liO@@y{1Ce6d
zGppG3cOT`TIjFNvA`kGV+yX~PtJeNlBLI46tyNjk$!NWzAt*blz~U6{EI2@;Uwb>Z
z;@N{4=Nov2ex~j}RS&lRv=?II_^+$SwT5KkW*b7!Z0+83)&zJYSpVn_Db{9s(Sa4W
z#POc9H9sPWeJWuyG0Vf-YN@}?UH<-BO_laIK{2qu36dFe2i7;sP@r~S;Hs*xmX7zU
z#2xjR8tNpwnhh`t7-s}VtW(_<$gMzsy1^<G&g0$kwuAfY$@8;-1@6wl;qx`$?d`42
zuHNlvtMBjr@%Q)h=do^&CrbH!UA&3bC_Pn7MYD0o0kIf9`YB`vH@wWhKtTlK{s(}m
z%MOc3>76v?^h55^j_Ze#GOSIM{r)DJ9(o*wjX(-X;$-*ce<?dQt0hyKMjW?~CmKZr
zFzeTjMXQypwzyKNODH|=^wRnj&NiSe4ioWdT8+p#GE>o5HM|@5h2&QjUOr8W#01Y$
z&H1Y~yk=Ii%NsHxxy_}A^)?+@Ybvp5SGB%HXzw*;B}~V7Qyo<2fdxP?IF}1f<m{&E
zoZz^|5#|)~cgj22kN-sAupZ<UhvkABBTj#;{zBl;ep@bSds225Y!H|?P%&0nY(#Z9
z2mqGTxG3KvojO=o%phHcS|Xf8lep4A$ioe=SH}{|f+lhck9%UYC*J?b1dW$kUs=;Q
zzWj(wjJ}Fa^cMV-p*HE}WyzyXf|MG@j@j9gp{b4!rIWoxw3?5z{(86Nt0QyCPY;nn
zmP`o;0sUTe9;q%aAJVuaVx1B0Y^6aSP}s7tGW4|Iw9KtpT+t-dls&@)_S_`w-zdY<
zlMO6vrpw9?<Rq*aK!LqT(=cF~A<d;!bwmxsE8saD?b<j(F<<N)?BSK=4k;AZ=x)`~
z=AdZJv%OFh)VS`e->3n&kFzUda}TFH>Idr5F+mg@+I~6E9F34hL)eHpspK&|X0xTv
zh*Lj<2|T{?gq9IFlXpVQY80!~BLJObhNu&kp-|Eb)T0bCmQIr!2(8a;I~YnHPQ1@s
zcf_em<W93p4}llw>;#qY_tcsk>OepjIxkH)ELRwJ!Ulty0a^7Cm^Ll2JPJ4ti`|`~
zVKQWt=~D8|&Svr_uMRnWUoSZnqW;&7Kk2JTQyi-EEz=DRjB*Byr9Yqgr>iH0^na+^
zoP>m>NqrDxtUsHPj!q)XEL^$H7wZgfNuU9BsTD&oGb6$C<GddBWzc7XBdQ9e!1V~%
z7^pC04A)KgUrzo%l)Y1TC|VPC8QV^FY}>YNbH}!A+qP{RJGO1x?0jSN=)UN4@%Epn
zRaL8=Ii<(2pw(v_QwWvNhI6-A!=&tw(_0)Z<)w=}Okfke<EiHin*vN*ML*itO<PD0
z&W^SxPpOo;QHvfX#+CV|9VO*LH-!Z0{mi<uHL$E2qCDzBgJ9$T5@351CLV|-1UJHw
z8bRQ7ul2f*FrnPI8nD?qK{Y|c!6Nt9CwSNAO8ixz%whK)|7h;b-f*VJO#?9u3(JAX
zsax!zup5S%{xRrm%lboUqA-86oNL02%~Wk*Q25UT74lDha!1{goXCD{*JJqP!6M|L
zBnfxhLMIp5%yX53xjI^};WNOMj86AyoyO5MbI|hPV@r_Ax4POR(zK=$8Xm)(U(bxb
z3R(&)A6CrMa<SpSF-qJ+_bN=I5p)9NxcEqUg(9SsBImk+nQMgD+AsxN5RtLnvre?!
z<Wo%2{S#(?g^N22{bl5qPyO&qZ}z>#SD4eJi#3vN?#cE``^}WJLv>L|nJ7k|iGw87
zB$!p;P}7u)UhX(-yiP;m)tx3yg_jyy=RiN0lu*`gNcnBJlswk`b=Sj9%j;z@u=gl}
z_swahjb4-EK2%taCh}i~YMI`oFriFJO7X#na{^oK?_uLMrqIAM8a`mINkfR;<$H5s
z_E`-q9T;e%&r9B~c5WLJ;8pyDARz(4(=H5vlinNYr-ES@^NOv1m-5>n4T_eo&%^CD
zKE<ln<|TR^VPck{pA^v`Q@_!~2!a`#dV@<Me{04mL<-{`l9Fi@lIG_Sv%1j7BR62{
zTt8fx<A&c5c`~es%G{B)J$*5r&mj$)c%#Pcf=r@06nofgkU-()5h~%}`(nZu`mTSc
zx0o&<;f`AnAYEU}oMp9u=L!Yl-%E#c-n$)$%WxP$Q3|Y~F+g*ZR!q6QKSey!>v2az
zK4s7+CKjT9018MQiSTv_937P4!{h`8;5D3I#Dx%1KEylvaQ7{jlx*uZmeGY^<>P{K
z#qp>_ZMvTje5i@{9tw&;tE@f;kM9<*5w!tAI5Neo4E~nGuRb(zuvFf*+`BDT{^8W!
zn%xA5GlQas2l`}uyyi(zhYtG)dbps>-H!<@hj4u@N4Jslm{n)v)z7JjHpivYjU_R|
zToTy@7c-u{4CwiKeoA(H*?Ud+T^eHDx2zIQedAwVP|Dc_1SNSe7@9iD=A-n{M_|B(
z6f*?*p$B2=tO)Hh=gWX+a0&SBe)5X+hSKui5S>nDW0^;#*|}|SNd|hy-p}(A0@Rl+
zADJh9*i#iP*X5fW;;orM&LffI<3Q9<`QlQf_Iiz#C)!qnY<O+Od&a3$^yg?f+EtIK
z1oq)t(^QB+S&%}9H`Ae!D+k4h;;Q0FEAX&|gH}ogvB_1%mQu%1Y`CEflEw`Wo%#M5
z2+)Bw@^d!vtHT$QDd4ru8m+a5Du~7C<?&O=1+P0Z1+4nU_FLKk>eDU5+`Cksn90`w
z!~zK-Lndwh))Qz<Tq5^7VuS)2BF;qID@&Luho2SwyLTSuymT|@Mgs{0VJSgqtJ7Ek
z76c5r5_Sdu&`j6|^FH9#BkzuUhFa5TstGFyTLv8^tzK%PAS&3MsgWPlwyhNo;8!tl
zbF!BX%BkeWnhnnmR`ltkDuY<$L5U&3TuaK%(EQ4|6hcSAI)pwP5-wv<zBf>Gd;yEO
zK$yP1cvH@KN>|M}l=~}Mc@ckOtL7Ak+lHYMTl}QBP({ELOLISis{##;BMblq(a+D_
zkxK)XC4>A!$lG4%aa&(|hH-zvIbuvNeYnc-eNX_hJ5nFuD@Ont(~&>^{DU>$T24?t
zlN?7|6RHSNwreVp@{TkQ4PIxwsTObbVP!R6g+J<mF3f)>B|a4O!j5L#agB1Ok&s_1
za{G*ho_BnehPeUKX@#6<!$sDQpD7-j*Ru;5sTh!xc(_H<k!Eh0yC`Z9d&<05Jjdl}
z8d@|;NXM2$gA8e#TyR!h@SF#r33`Y3r_qH?((R*AAXvploG=qG2U@*!H2+^+63n#u
zctv@WGYE`d8A2^+a4KqCskwfFk~u_OjioEy<`FH05Wl81_oDWJ--wa_T=%&h9?bya
z=CDP0@P){VyP-Y%!1{ZH6!Tb)q}n_w|2Y-8GO0bQ^5unx5d(ef;89Or1XR9-8pjzz
zm5r?Bdrd{<fbMHMm1ZNQWEGe&$hy_qu+GSuWK^)ZvgVyLt{!~Uho@HMEn=NCK((mT
z8MmPRK9YZUzTIji?34jHbe|Yr3`su4>|cBG_wN2|Ky@I$j0yl(75L$+E!%DB*7l&8
z&VMn1FyMX{TOhK%+PYsZEP^I+vZK4$7lTWkFWjpDYc*@Qv8JO=ey^2fuL@u<YCnCx
z+WcmIj|398WswGCjr?M9r_wX9t_jAF=;(+;Xt{uomi_Z-_%&Z_qFK{u7{6m+$E=eJ
zz>J=(btnzBDZkNYD8y+2?tars^wE}X*B?pI!g?_hqSN&2{XcYv&AV<)nZk|eUr2%#
zS#4x~v*t%fU3AMIXGf0PF2jJE4NuQTpS^%1KhGHF;$5{wcM>?}PT|}T)WTz=VD@hy
zz;A4(>aCvt***Wq?uC(snd5)G@@UrB`n?-P^nKCmZ!L5r4Uj|XlEs_P)TVXKW{qAt
zYigGcNx%<>6`>?X<vM%x{Z$|MWV9r1KK6q(sb8X_MdI;eT>@kMlC>7j_s`@MK`11Q
zaKS5e@b~%urOB5EEU~&FGotGC=zQHjl8G)L{#i&G5)42XeL9$#(XGzu?TKnDASgU!
zRAhkF<No+~dYUlx6}Ycl@%_4cdwP1Xq`C?4fHwL4RK-4J)}KZyflM^7;|z$g0C(WG
zc$!-jaT&m~T-e4?+K+wKuY1@>ZP94Yjz)1eZ@~I^B3Z~^6G{v!F~6_qY5->Jx|^KJ
z=S0^XXulhR$yCZ#)18*t#;uZbJSZo`(lq6sL{bhMs^U$@@?CuUo;7Tpx*vF<w2J&p
za}8W}+o-pRc8A07V}B^b;e6P$8lNJ=-pZbkftF>d!(v3;<D*qV<CHLO`zQI=!DdZ1
z<oDSdL@8hwdw3~fc2+&iokBxrqBB6%7T+C>yoA|nB)qD%{y#<<Uac*@*G&V9rV$1*
z?r30nsRU(xpsToeCS3k^h^e%rB;lcgiTa(g0;zrC_y)P54cQosiMgN>iP8HTkxcCo
z&lngE(r}9Kgbt%N&6Yqi`_Xz`l=!eiM8u4j?;JFt0th^@Ac09|tnayMT(zJE5U=o~
zLV^PX#(^@Au><zK?26I}fdkHTcP~>==g9!GqH)YEXsIws^-?%?8eh&kJS(g;5D8?K
zRTKtX2Q}H8T3TLhr?m+~iL!bBC<0CHRcyNLuHt^Q5g))(k&3`SqwYb(WP-)BGyxZp
zy);RKx=H)O#Jn1Biu5Q3ReIlgsoc_+^CD4W%Cg)+B!5Oxl)<8qK{dVm*mWi?^n<Hn
z^x4Ym`!uK(=>g_tM_@=vAN0jJL5N7|Qx?>vb($Kgt|jQLc$Bx!<+LRP7}|hT?8n5W
zQw+gL4O57f<(}`D{UV6-_>rWtZ4fhwG)r9_57{Ko1aSq9?|B{91n;R51bM<lJ0`E@
z9qK^>O04`Pev4Z6k>oCU<h4QeCDy4!aZlc@J;!2|_<jDT%&@(oobAlpR|0kEUHrMR
z>;2#nKOCXWO}Vb2W_QZttuU?Co%=54UVRRz4?{a)38q0@c)TBo9}ZKCZOds^g&0)R
z;ZLZMVG2c@QkNd>O_v{$dsj2_Rd5%sAkQVx)p^*I9*4$rcynCBY`Ca)Bq<YnyOY<6
zcE(y($-6@4$Lg15KbB+`NGKnQLI!w(WCEo>#^LcVt5Du54FX=nh>OuF4HyC#8*rbO
z3zw+~-DHIj5i#Xpi{|~5#c@2Jzp?!MLq|5m>iFIumpyd8wj^MyCx_9TFke)r4PhkU
z?6d~ElWOD?I)uBd33%Dqa4s7{Ewn4WGGG{y0vzS$js;BRuC7Y+(u=XDQ8Wy;`_qk*
z7ElydW{nUHBm=M0rtzMO?jWrh{1Wx8WBOb!6^6Rwet%lmX1m#Exg1DtIk;aB?bwAg
zdw@e2)eaJ|4v#?`c3=kP4?s)vXdxISFdf(7djCG&Ef!E{FjaG7%QpB6Y`4!{bR+5S
z?||RZrH9e5pV8VpD1Hj_T-|{pe=~KNOkH5(>!0dJHEIAod8cq13OmAsBFLlWUI@W&
z{w;B%zD+k|lwTX?CBBDiZraukhLl)i5`g>;dce;sKA#PMGC^3_tb7fiw$o7=s|JGt
zdRMri<mgMhs}EDx#)tQ94ZF4LR{EQ9$Bmkk=M$>pW)${=6Yc3EY5=>H55;S0FyV&<
z^i5*Z$VbqX&s)b-BdMXCZ4@#CTMy65QD8W5sAJXqUYxxP_nnLUD)y@!Q<IS(rWbHy
zineFKeuRZBGSQ%(9G#fW#yn7~d%?nD=Jf(^hV&Pd=jYVlwFIN)DOZlaH|xyN6x%E&
zO353pHxCY$OZ+YDEyqGzG1NYc4Q(HT4ZN@~FJmXf25j%^{0-@MdvVwc`rF&4!ps;9
zEI(*}Y!5veFj|dRdOfcO`KYTZ*=~B-LBf>jV7!#B7tD{X8d*K*kGeG)HO(2AHzlhh
zw|5{xpT^H}0xR><-d+~Pk`V6m5E1dQ0E!|BUU^EVw?cVbd7j1CO6@?EAaS?ot|oG#
z?zUt|{e*e3idNBSUs{r3Y_(E)%6UzkbDB%~ci;m0CeO8r`f=Y{ugAbE*49Srv)7uY
zmXsG59)N7`kNtQ;IrYt4YqUN(jun(R2XI=3isu8Fej?mx`|@UR95`NMJj0X$Hf^#9
z^LB-FoQL$feG6N;MwhaiaT%Of(S>njR6W|9jC!b`V%MZYNBX`wW&=HcaYfx=Yo4Hl
zno1o9$j7(DRxuh1>|H}F`o|7CxCJJOs3rEV43Ixw+g&rz(ECNe_JaBl{#$ft;{WOZ
z51YBr*~4sr=XkaSH*oKfA~tbD8=a5x@L{kkGI<vZxee^0%+Cn6R~g}-Lz0_8mPK+x
zTmvNMz?OKxfFJzKu)8^OWmUFZ`0c;UNs-VJAy?(aea?!v(rQGay&sRppp*H#?`_K=
zER$ujn3>6dHM(ec39PkD*;Y5&4JMI;^HhU#VHJroLBt1OlwKiVdDiZx1B<JE+@kaE
z28PA@qSp!M!mFQn)H1f&QjpZM|7CSyuTghqV^~)jrKC>WAbS@<0JF~&5!Wx1Urhyo
z9=N*9&OA9mb$gr!kg<ecL2tN?Sj<(WmF*FnT@AXB%dkhu6s<=myw^fvY5p(>Q2Eu;
z%rjbz$Q_HXwZ}sDMXZ4*-Fs<w(m;8q{+n%in^WQ6WlE<)P{lZ&O@lRV`+N4R&N(h?
zko!Q4)!qLNlJ}QO!Z+j}xX>$A=ai1U{<Fgw$K(~Hi{WuM0MIo%Rbl7el3nU>kx4$)
zp4~=g2E{oKj>Kv5#|^cUQ=~CA+=!bA?;^%8b#<un#``Uf(U4A?2(xlH6^X^iO?Or7
znZfH(6p5mv&2yxMDMcCu`4i<~W+G`&B4nhKj-_P*BQX0pu6cDLBb1f7G@}pHw_uMU
zl0ERFRF=%K1uW2_VZ@1GxH(|j>7V~Dn$_f<+Qan$a3Q3`U+(udvfW#D>$;&mC_)Ch
z_)}_r#bYgHBEzb{l-B+ovwO(Qkt7;iPQvTn4D(Nmlo&Y3c<G)DNKu0JnJE=pZ^pD-
z%C<sxnAxLNh)&i)6TRQnfsIvM2J6l)04@-rm+j=g8S1aMoYal$S+<<zwB-%HFj`9A
zePO$>eU!b4kbPVTz~T>C$gPS?*Pm0X`2cnPHn+ZrZ|hN8S*qH|(-9@R$n8+$A&AQM
zxQbiTI_#RuR7gX}KheBCiS+ip&ZD-xkj^t<Y0SoZvCpvpCLJ%?<Gmls#o;aT_);In
zY*ZaLDW#4Z(unpeym~sH@EZ}vasSmuvNQgFkMkHfSlRy9HuCB>W&D3@vQ^!&^4$_o
zElID4a~^j~T09&a@<_%FMF~|B?hDhuUuf&_Mdq7|+GM$|9t{Dc^y_17@#4(waAtR%
z9tWKsmn=nP^uy+~!52jEo$`uEl`Hf);b}*skOX8&B^T~4*cQ$*8y6BG)to+&LKjkd
zb+x<P-9su{PxyXb&-eEazaC3#b*1J^expZ{gTEtgu**W2hrqj{^!XPuZo-%u_fw1b
zL%~obn!A6ZvvyXUnOC;S9TbHzs;OhqbUdS+&3v2WgB3rcC^<N$Ssze-7%R2Ba)Nl8
zen$vb>BQk*!Zd8#eL`I_6-9h_S?A2~lfE)SNhgk>4?1IMB+<g_bPZkhw?v0T(+4Dy
zG6$}TJH}p-0S_*%eQLc|n&u`p`#lyHF~u3e_Km82W~t-0u8tiXU;?g|S%2m%a;n;7
zh~f4S0+4da(eyt&x>k)EOI3ZhE=Vz{_Gc9ANoSlo!=SX-iC)RO;}u6E#sQ{m%TwZx
z7Jr#~e^c{nc|eTNqZeK|RTgKM8|{iB+;NB`S@z0%tP-`UDv)Mq%7I4J3T+^85?Sr$
zG=ZD{0Gc$i2(Bp4tSJEz*dxE*tj%RCDa=EXM3v7(X!LeN=Q(P<*+pqI;bzf35hZ)0
z)FL4T!x@$akV=Ik%3`W40IVZ02+m=fO%TwWxt;2%2@C#qbK(9aBUG`)hxY>(0I`Xi
zN01Rp+9HK6$5qQ;*04mPSmM#Tg3pahoOP1RJNR@iy+fPD^0Z~PQ85D7z&Kbk!CYBj
z9QB6K!*|!CvhzvlDoj#q>j%psv29UFHd$#KyRV{D6lVz&lZ;~EstiJ@A`~2m=I8h$
zM=x$@9L_YU^|lO(ROKrdr>`&}9Ie@D-#l)+mE}p{@Ai&1>d?7%C=wibY&iw3=6tyP
zdH#pzz~8Hngdu-Mo>#N*juiYux45#YqZI^LZy$<}#WPF&_KbC==XS?rT2*ah+_s-u
z-^Emakh}T5y$KHFPf6P-rtc^o+);9?JC_FrnI=_ixHV`EVHnBS9T)YS{s5sJRHdaf
zqg+2N8{~+122BECKMKgiMw8|~^)sklD4n`>gcsP_9e?^9f>DzE)t!eyw#xd@l+dKT
z3(|&wkWQxrV^PvSTdky_8tw5D);T}Jqn@&T7_&GKuGH<T)c6O8hKREqfthuw`suzR
ztuk$5_Pwn37fBL91EN;P{x85m5$p!{Rh~e=g+)3GP1TT3OhH;95bb0M`@~6|cX06n
zo%X0KvSLire>63&E-h41k7dRu*#`aMQGZ<J5=(dN1wtQl=*253f0x!p!c{!`fuf+a
zX-o=Q(dgvSGg7pld1QAQuqo<!t_Wc3+%yb}UcifRNckx7&B168hP&DCQ1ubDd4_M|
zw+GaEfh0&fnrmnSEIM}{jRVqVR1?!b0DO=cLSJhXJu{*TqryUA5aEmlN_i##0xRb%
zsX~;Phcp~+;X+d7{aNv~Aw58XZmY6v0pp;a`7oGdU{5-?TotGKBvDB!FuRWG^|LVM
zBk5*4*n(4L7H^>fN6o{d_ANm?p|>Rvr7{e@3`62T{24g0^|RpWw{-^gd}a>%gz7uT
zlYnFVQd5D;sW)H`P)#J*+~~{exDM6ar<(3nmOLg)2WcvHdj@~bz*#*g*%$A2<!4}H
zz;l9+8%^y`R_7|uPFocooFW9v*yG~y&+6mA8=p?^d0>xk=UNnjn0uYIR&M!8=AGXw
zcMTsSeTQB8_4B+0$SBx};$dC@2Y_l-^+SV0d*XAiGij0ClrM~=MLJc1SP89KKjaAn
z4eY70$Shef5NChVO0ImFu~v7I&T<go&)4?%a-~yZM*`11J8T`6r#*^%l!doa(YV9H
znH^r{18{xdK(o^A86V{QP$C=~WE;kRNd2;Dd^*d{ubr=T*)<e<AwD;Tc<Q*s@($EL
zYydJtj^hPJ@$Q)`Ehsdj#kxCJ{BFgO(3*EpTrzK#La3`?m@m4(qrkre0%?A#@(-++
zkFkPo@dK~K3=hPGUG;{EZdJwzpr1zfIo&Q7-p-KaoE^#$u1gWj+%O!|7ZZY2>v#fc
zkd#9eHHiCx8cvXdANe-@6|^<$6u=+{aL(nYziOjqLBS9)q@aJOh*shXE)wWtE8Bho
z@tLZDH%%%?)4d;$dA!)%<bV-3!Pn&JeX~W(K3MU+J((wZXyf8gErkPHnZAYHiYRy9
zJSo~Be)tzVJ-Lz`$%rqczP|nmD{=A~?RW)<5M$s9))>XNk?()-zDzJZf4lXh$|e!^
zr@BSP7VdaQgf&rZ!XDOlVI)o8_5470{W!|+%7p~1LW6#L$@3%bsY>P@U9}y=l5O;y
ztW<<GI>e(}e#kes;z~1w2uzgaCKwVzMXSoik0ym`LOHaZmA-D^=>Hpf#ChPm|8uya
z`b2>^O2_A@&aJ2H^`JegcL;8^bC+!W%JeK9BiA@RHR{$xmtDK^#0T|(_VXUq?z6Ub
zNpu{aAK>DezI4=2v>&6PNcBYiknS*pjNSJm0=N5tgKq);C`f<6<9jQGk8?kIQjIjl
z=RDC37sv(<UFc)}D&_5H8}$4o#~UX1pEo?exuE~gbv^?l>;G<t`G3<w<jssx-~vc8
z3z%$1rPW+E7L{Zxh501n`)TmOMbzeuXSO#1_^AE~PY&xsv_Q>6*FGOthN<{wQ4k(G
z@9is`y$<XN>K&6J-H7|NNHLM&MDn&_*Bi>O*`eFECu&J!&DM)#t4DAc6t3MLNBcrI
z_brYiAm&N|T&4pgS1;(kTU%cq_C`Z^Ha~8z?yj!xrB}`3!G8K@3<UCyQzNIvYOSRf
zI?tMTS}2=H!iXnn@)6{F>hDOf`za}oYJEO<98vn?-bob97GLg^<0?@}fmsr-h@0dZ
z2^*BiNs@=aqqHm7LRj2mV9s?Ho*^gFpz$JjV?N9Qz+PE;4ChrG`n1o*RgM_Nj>#G$
znSSG6KHWdoN2vg|$D^eqwIuB{0tuw&xtyl71XorGJscQ(oXFUo^FpRc25|XanOPun
zPDozBRb#c|lDIJk2O4)&s%--IRbV0+EIgY#+!%Nj^tk-tr;Vlz?{9yf!>r530*LX7
zHL5~Ju9(LH)KMuXToWWQg@;6@&P$2f@^<w+21e>bqvG=m0KXw>GlU4NpV-M8?CQ@N
zExnd|d~fvd;w=ox-uUrWOx!I~?wV@R7QRO{=r7^tPkO`gBd^!pFKgaWy9!v?_kf_X
zk<v(h-(rU5U1ke^0U^_#eh(j;OSnP(EK<9_GB~1?suJ;Rq(_xhwn^6<Ehx|(oGlrF
zhld5Fv?Y}{f%N7<4ZWyV%6zK_R8~#t+jurkx-j0klbA001CBc;yYIZ^jOpC!Z#!Oh
z-pQD#)>lwo)l|%lP)C-Sc+&IYM{*P2^04!68OM^2p4}Bi(cBvARF*r|7m8EJ&+crC
zwF;v4P(_#2#MAT-MN8#=1WG#3X&FVNWxkV0q%&Xk2&4X_r)l*B8K84Kf}Kv?27AMg
zNwEC^+315lKsY>M%-86s6PPYZhRK}bZ}}D%&TU00%jnT_UksZz99EC{ovbOn0Izc3
zC+Z}W&=DS1g0->{br4HA^%~Q>FGckdc`@altE+!icGMf#YT?pyH*43<cHC0xe+rEn
zr&6wrLRmGR6Sl}EZI(;oJ~S9vhrNWVQ(!k-l}`X+8x`7W+do)pR_X?8>GFU&6A1O_
z?7AcS4QvZ`YovMfeL1Nn3n^N=I)XMp)KZ*+KN7S`&=$?Q4zn5kd15s<1sFHaM(pbb
zt&%}E4n*D3kh#Y3FXVf5M4*rv=7H9hM>F~(@oTq1g1^rVCpE$cHmxWua^^?1jb}uW
zzo#O=tUCwxdIE!u8e298u~LBs>YeR0_)uhlTA42FC%t=eG{jm=|0yL*BR`meV&q8m
z;jC|mr+2{A0L(LXZ>9t|v!l3h&8wbcl|D$Yr=^VwWTbZ;E-+&ri$a8rH`j=b@&&2a
z2&~`HJ^LTVfJwCa!%)#Xm%nx>kDG@UqcrzKYS*b95if#3YobQrUla;WOq~L>Oq?l1
z;1)XF{>6RbAa6c4KvA)!;&MljD_kG<DnKHt`J6ppC`$@XHfWX)#B|eLg4CiymW&6s
za_{N)2Bb__V-u#CHSOnKqTj1`0=S^AZB0H4K;>n`8x&R<ka)^VEE8h=fjgzjVxTUv
zB-mV6TMpc(1>=!2mDAfB2wrsDfZaG43Dhw0giPf8?8DVr@>sL{g0A@)(fh9fisQeu
zhAiy=Gih;4{r?%D#AS#;-~&K1T9|A|?~1_LnwQS{|M{norz1xwON_^w-0yVu1u#ln
z7*M%zP7UXwsqYT*_UNQ}IL3O62I`FlrrKgI<&bh7=CkFv6F&Y$K*AEY<$jv>{-rkz
z#>15VJa4&uULU%#?<c&Av18+Xxw|}HpMv;y^Zg#LWHvVLjo#VP?SK_@TjN6}6J$kn
zN?N1AmI3caXMhsnA{Gm2s}{d;OWEC2{#8ULv6)JHM!Cd+YHU>&mfW)PZrpQ{2(ma1
zPk+J4$0lAh%84UhoP=&7m87)o@)J3bVpp*{ql++4NS~SQ9gFx!mL!`fftD&%@>w+S
zk3E-F36K1UW4svefVpf0Q?h2CF`ACC^#`WrFP{ZB!z0=7KH(H{t4@7IYNZG8VHogV
zR76Y=2=`9!v;`<3q_DC|mJR*tWdTH@h$m%TKR&`eJ-w1dsaF_O{^`o53aU?)$w`d8
z=*>ssnndoJ1;U*klZ*^^BT^!DV*41&s_>{^7NYjaGD?!!QX1i7fH)E@0`_|4x-pnz
zDgC+GRs*bw8BPH{Yx-rGdgufHdnLAgf9x*7C=;rmci{D#*UOtUII*G*j<{q$i12yE
zK;vAB-zjatBpted%fD|7#JG>RpX@}F&VEQtg#M1*c_{eUMkkv%_R_OcTEF6bT?B^$
zf}W~rXU$Bvgvl*o2ggK$hs+5-{HZ(II<p@gb(Hm}>nw-HH%809()Mj4a{t!e$tSS<
z>V*vGpo~%`Ex_j_&Vfcl5Ioehj>?S#)@2Sw+Wl5r!?&TVB_zbN6-wibj6Lcd%t5l$
zm-2eW@3DNRTUG-lkN|SIfV#X91YEGAKq%N>)0-g@n9gPrxhVjH?Gd2Ade!W8g$VzA
zGcCk0PnkwyPK|f0^_Op6JCLLz<_o$qTuxx&!y}#LB-oRne(MS0yh0M1*VHl5<x!hI
zi31SkA^7o3=xpvcOX<fM#&6jWGJ|cHt3_aQmIZa*0+tIGRJt{QXNL$@dT^@Qw@7MV
zPh0s1!Z<OnM5w(+-<_v)EC$yS=d{s@?WV*^s!^qe9PJURS|JnnjjzCaV#Uk1dmz)L
zhW%}qML#Fgpt@f0Dl{*{-&AG%v)njkfOjUl`LdCZSU9Ug4u%(DH)Vh3*)_wQy?CZF
z@0UhGK22tc81t(RjyX_KDXve%?yY4-*TrhWK<b0urH7I>rvVh0u$~25&&_jg=4&=l
z{%(kcJbq8Bf)ERZJZ<&|6$Gwqw5S(!ttja*>y1FazToT`ZeKOm*>Nm_01X$sdadl6
zU=xNnIsJ3w<A}Dt%n*i++{;=$!pwl<Z?z6y?)TPE)iyE<^kx+-U7hhg1!av#(3v7g
zs`6c-No6{W9mblyEBwdsuL@Cb8nKG=O@$x=8fV1_@-git6?=$&uawhJ@#DJ@$4iS$
zHZES-lL_~$XB`-MULYXpeh|a83>q*IT7J;c_(2FyUM~f*E$i!nd|XX(X%J+b;}{ho
zD5W!T>Gf8BzG>Tps*wPi0A>XJaEU#RYYxQM<FWka6Q?^L*TXg6A4U$4{E+yKp5gOS
z_bu<ZtPegOw~22ZcA|vpd;xvZBchu7My}L=$Xoc?Se{lQj%IufyeGOr1@Us<0bQN}
zLVrR)C7sA2W6YgrS3W)iJiMNp_FzTQ)s?2_`uogL{vx<6H$ywQb5}Dzf{@<u9do>=
zNV16KC5N}G!lzs@5h^B`dWfC}kmH%8zRxox41x??$Ox~OTRQNK92d5Ve7TxDC%Q@K
z!K{u}dNWZVOdLfFlp$1Z5H9J)tNU410Jx7g3|}qKlqjbi25`H*7p%8v6&-@XyHHcv
z$kK@jH96GxYmVNnXB<86yoBS+aO_yR3RSt*ox+zKy;WnDpW7>~&rxKsuxv*@qF5`3
zl__^Pm2MnFw#hKfMIvW5<KE9uxZNKj@LO=sywm`Wo43H+K(j=?_2ymo${|MDJq=AV
z<K_heMP#&cAxWW8#to^#6?)rHiC*i6)3nwuYTUsZr0wSpbef%PnSCpFOuj%ByNAKb
z@2%e659i72ti_A&-gAbXKRZ$2KR{_Fj-&ti1pNm>h~fW_WkG$z;dh7Uy`pxjU_pa{
z4n%w*VVV_n#d_UVUM!1;5*reapH4vsCDES3b@qP}|0n`;_0=WWXORN_k&U69rs#h4
zjZSrrbDlicIJupZkn;Ftb|cWCkfEJG2B};J?M!FL{@`o}t)nL(;IK^lM#O2*_{mF|
zc(OauGd18s<Mz4w%oq;-ALymMF`|4qx@TvuD}22if1;;Fp=v{6fun$89j=>3z0SCw
z&);WNOb@n8aRFOb3E+Mif%^#42+lgDDUXZX$k><Oq3#%Oyp9fM)E>Rq;|-X`f<U98
zuaGUUO!o1&JUD$boaABj0C|9)d(wqkOaLBUvf&&21JiNxL;5v9ln?YYXuoF4H+^#Y
z>)HWs>Aw^Y0}-B?2>CGXQMSnMCyj4@p0{{zo5%KKrp75+v<5B?&V?E@vWt}&+@AGn
z2<fB18>UDp?oPco58@x2@89Mg?)P3|ulRVW9&W{<M;4SfY724~cD&ift~U&$FD3#Q
zwKhTzSvw}T5mx`~0jZTG`O7l0SPGkWTblR<lowaDV}FsI+)$|+SR?PLQO1M|4Q9!E
zd*e+1w3_C(R53`x27?gOA_P3@PQ&?&Y<Lyc-%+A)9!A;x8%=7s7UA8hjz!nczOE&Y
zL_%xl%b1+<z3F$BM~;YD(xq*yK5u$om;Z{67?(*Il%ddA_04P$nwy{=TKdn}H2xmQ
z*_JqROeCnNilc5!Xi8Xv(p4?7th0&YpyhT9s3*=AD))*4OEE%>qX++@1a<YDDZ}Da
zQ$azWiulgyc!bZkoG?ix%7Fg9@>Zh)>{|<9ck4SJdBE?@Ju1WY7w9BCm_y<4M4{e%
z1(UMTY5naI=*aD?D(?q3tbWITVAdM7XO3<(R<0<{FMb$4B-y(z<8x>{xW5^j5w@pL
zN)J79c?n9%PEo-8N9HnP0gVEFm4k68t(l8_?B2Gv^Y?V>!Jka@-Wa+h!zYhSuTmG@
z?Bh{?hjrC=?)A*)<RxB8GyBJ1R?*_+Q&mR#!D&qeU2P}>ASq>bBQz@pZSoT9l}v2u
znL}@hCpfXE9X4x#!ca++$vH*d)WdmpQEp8z7LSRf{oFnoK5QrT!Aw=uRdGh-Oz`~7
ziS%iBpu;EbT;4|Z6iBuh1TvNL03@pyj;$eV03}YlR`BKBf&ZNgyk^HiUV}1N^@eu9
z*#P+U%ZfA-sG>9Doru!5Q@Rm8c=Bd|aeTf4o;)Ls#cUsdA<kOFQMGzs63~NMwIWkq
z_WdH+y>ae9`*=<#iT0`Xewo^j2YdQLLFA8Wu=DJ}lpg|6=gczi654~f?|oGitOFux
zWt*t(shkY+3(ienpq--(7iGLEnsUB;#zQ_rx$MV=b3;WxRGXpjgjV7qqh<VxbcC}u
zZh9n<+bYRwpbfa3K2)Koo;jCf8Rrrc9tc1P(3^k=*n`f+sXUArZ*51&9}KQ>yVMhc
zM>sj5j*=8st(mH`H6ejrTdjdnxKvyu^`|nd6DivMlC%4s2W0XZTit90$Gst?8`?}+
zh=5^agc>cb_8S-uv8ZMt-Dcre!%L*nyW?a)2;oMz{CtRpr+5?K^Oq%sf!Vu^9}@uT
z^MRi5x>Yd5g=u`Ku$Xfv*o6t5ZSJ-ba8fD^1f3#PHPxLz%*PsLyqn5@R10aT>yXfv
zJ)>OrANa-pPDH;iU(X@BTx!APX7`}k$}OVxI-jHA>h!yE(Okojsn6le|4=0i#T*)k
z*7<SmETEoI8^$VVsfs<@4@CVIp&e&a^W^_5$p1lLVPs<ZpV6(|8WN5hW2nFAR{T<V
z6%)A|*rW+?Q$`o}EC<FEMGkUWRN*G!Nk0H^fY!;}+!bhXW{`vgs}qpCw|DEWlp7ZN
zM})o8ui>2<Iv+PNJ0ueFcxBO*7sSN)U8G-d>jia3IzkjiM;25#*$uC@p6~19w}T{T
zPM34+i|voM^V8lH*~?Sy%+KdtX=UXE-3s5&ZA>Gt7A*oAQW0lO0!9J!C2Bat0Lr$V
zFd=DNm^ra{6og#C1?a5paYqmFoiNvOP&uycKAU7X3XaudhrA$>k%V;EmVZ29HmxnU
z9N{}R#qBQ6zflERC~8BTf4=p7Mp5kh;Jh0#@D~pf{7I+cK2OBz&uQXG*yQtwFVCA|
z<RSr;v#_-zZ3$m+;fBKlI8T@qQF+27Gw}n&hH0T(8NX4iq}{d?bq7)O9DLdRILB>m
zSxnMImzu4}$XpYZrt8iQ*BPgFY8Ur6{hnKYFV*S=PPvKnpB}yBM%G%%nP$$oKpTNq
zK_CNvLpM&5@uLcBQKPEJ4GaY|oAsO)bqQhu`~aHFuIxZXVcCjK(G|<=w^}Jijph?4
zd<G$PSoQ1aR~|6mQR2;qbLEhPpuloqiXTilg^0;xCXR7<;+gTeK4OSgjDacAuEIhq
z=G0OU#z?UR+_U<qW995gv*1(zeQ8l80R=0ljj-HN$7m=}%8|X8ckvp9=SSs2Mggc8
z)~MSNk&Fo(H%P^)pgai{!uRsye|RTwau`VZRY#k0f)$3jK?t*?@KXW*gg&ygpq3_D
zbj`?83iqgsSKba0BkPHHGe9Aoq$UN2Zz|U*TGCG%+m);#Z4{8Rj~daa{S@cf6Bb=t
z`tmxvd0skPK-Q?#BaVn^8kN#vj8o`w@6zz0R2C~9?wTNHZz$sT&|J%O1cFjU)|cOY
zK>)y4=n`_?r5n-9#UJc?Z{%v?f;<q!wll`Shn52aJlka8BjZV=#^Gssa_o-ZSxlvU
ztRbIJPJRal$}VI!n~ABwc+>NuFFdQ3_}wsmJ_ux+3`-lvD(4MT3(F`@zzsfn#TwZy
zZ<F3&96&pvqQ*U}OBJ^t*N<s;Q0LSD{HG?n^gcC7_oaO}E?Cz?5M-s1M2jiHIg~L#
zX;*i%W+Xwph0iFXRCw;%?@>HZkRsf82?)ppd2u9@FXz}{<N-)L@??LF$X92N^EnA+
z3j*`l3PEuL=yC0*1QHx!%VfCyN)K>mK)hn1_9riJ9<9aYcER8$B~Hj=VzH^b<b)Kt
zmu(=!K)s;!z4?q*HFircB6sMXA*C5W05Jw+LgLL@+S=X%3SoRcF*RAF_RWkW&_nQA
z_Y{TI(0kj`NfFvkK+{yhXK4oeh}<w(fNzji0j}8L+JZu1LxoY<JccKwK0W|Z5y7U_
z#?f#A=(7SA{0IuS&FLwk8CXQ1=1nXt0t+YM42t;kho&_A#)kn|K+(o{Z}EvtX<;Z{
zVJfmOr{iOeId|R7N|lfk8LKk$&j$<<>Cr~_Efd1nvsmkRAp{}eqCEm)g7YLA4=?a{
zW<_a5^bcweSaX`?VX0x@LWuz7fd$9hnz*nTJti?oIcLS_aYHP|VuW5I;_-RkzZv(d
zh6+mX8x_De={aDokf2CLcj*MQZ#uT&uaw4-BgrQltDfb|;-i0>m!1X|k{b<^G1J^E
zGNKsp!7-t%M440tw15x6L-49Yy`>;_*$pMlav!pDiYvP-9;OQz7qqOWBCN&OB0FZG
z%~oZ^9FQsjzZq75{j07pK`{aFF?t7OgaC&0h5t_4r-XZP0w9qA&I*KMSfDG{%ZCr*
z$l%S=|DvKPbBigSG*{r_*!c#e5^0;DDPSNGL{XILMvT!FSi^#{w42S(woOz_*2vk5
z3f9|1*s0`BgE3aK(70kViU3ccTPJd~a_ux&5Te>eZU_3UTWzW<m))1|z`9d|;TQWc
z28Chn&iqZmQ+ePncSK&z!=%xU2MODkk(2WbHJ!M!Z<hwW`p!q)RZfpvvjY>;b5x+|
z@!23|wKohb=IfN=ZgWEX-ZAf1S)?`=L8_RsBNqFlBJmCSi>ouz@`pit8o-ZzWd6K3
z0mI*HLePO?D7M6GuRE)~>OC&M%Je$uxmGTBTRrNnbSMEqvC%lQ9&3yczy-&$({sVm
zni@ipcalH)h)U-5kSI3fc_4%%#RqlarnF9InzlNtx&IQo^zBF>U0v(&nts1b_n6&h
zi)*(02RhWy`aVX!yl&u}_m;Qf2O8k{nU0rZu>qc765<pL8wVqF-`^0xCcLiPshu=A
zTN*1St=Lzg^H-;)pWWL?88y=(*!LKjT<puR{13O*E`TsZ0ii%M@u^Ym4bD&}yXI*v
zJ!{s%T~}=W(!tpUq}{8Kgnf7)IjfAoh{01qShQ{{4{Nd!$Qbjanadf5+A(Ri&q|YO
zG~a1=wp;7~k1y@lY@A3+uu;n4u}S#FKsG5HXcRiLDJtzf;8pyaEYUqcV{_E$U^Cht
zSAl+Lr_26WU`ZahpDW+lhhE7^0~rd$k}-mF<Hl@vL3B~a3G_Xdei53wr)ZTrXtk}!
zjE1W|QBGBqm@Rgh!Mp@2-`qwyz|Cs+<1NG=FWhVN35)r9eEWiS1#N^gYRR&LI>2B_
zjL|Sb3SmYSThz@=EL)!Ak!GV)p8uH&n2HKPWSK?JvS9<I%Kh>cMByCH2TOa6V^EwI
zDy!LK?e}FL94Ej~{D)inv*+!l3z#O=Pe9$`h)9InQ}a~id~|zTIya?kxLLcQ6UHMG
z14!-M@#i*c)5{$Ux>BH8M-Y}}V_r?pRD`~{)lRO-LoBENPpEcp>l&Rx^jrQ1#7vX=
z^g^nZSClnRZ4b`^sjs>hxDJy}Q#p<4^8*KDGQv<W)bJs|v}lpn{pC)Zn2t48q6(KG
zaj(x^J8q&Vu(`}PpLLP-8FX>BWahsqviISC?WMI_ZHP7>2OMiqNA#Oloo__>aj@Hs
zy6=28d_JBFoX2>!jb7fJA5bUR&4my9wyjzYoyBzU5oA)D8&E$y2YW8{@6Wh}LR<jq
zTiXjifqB$&csADkZeJ{`C4sPkWO=E4fU1da4gb+>G1LD?jF5qz@qb>xUa9Xm7XBaA
zjb45$X}9JaJaOKcGikSdjWy@ll1uJ9%;4&(!L*+oX*m<y*G?zC0HL7OAc=R|StNs?
z*ZSA<h7M#vN?6`YU~WlZqGKo#-baE2vkpFOyd5!V4F3X02WLS7j-H$^4@c>cFLSg9
zh}1QJILF1-jOlN1{q8Of(m@!K7Z*w*W;^$bBmL9Cr96}`%GZ}ILs~Qu$d*3}6y$>z
z8HlCUB0R^F`TAqojPsQ_uZ~&(#XbhIc+4P8F~sCw9>%d^pA$R>ZeXb&dj9ZkY>((N
zts>mPWGG($lhpFqDoB7Lo;!d2A?Wb@;N(+$9$ZRJrMv8a_OMnYKSp;)(GCtg34vJZ
zqvlULI}K@n5P@-;l31@#K3>mK>JevH%f;@3H9Ne?+LoShoHZy)%`#j5TdfVFo2cAV
z130^*!?gY|xEz)$HKVcntW%Z(wSJ#?B|X(uH*+$5@kqo%$TI|p&X1ciuqM$#=?JP{
z^QAcx(vIGI0Oi!i2_uLQcn}z=8B`nOkc!2aOl@@<xA(269xDPVD(YXWePg6$KtSyP
zoxKf|p9Qq99U!jn#!M?zYu;_}2eTBC3oE4oJ#K!sCFIG2&fOeMe8{Nb2Xb|hNNxao
zu*%CDATTthn9T*BFiPER^|+_Y23&ip=_r37ZB5YQKT3^NvyhYB76J9JRn!Md0b?5J
zKAr%0CTV8$|LTdO?Yv-Gm{%JD8*L|^iwtCn<`x1#|5bzP?W#<)xSa@uyB^8jK<O+o
z^|0JC6Sxuc=^4UjgdE-I)Tz#O70O06&qRQVL!o_eYzk!eV4g?T*ULze2I>5p3rmL6
z26tSW?dOI|jkB)gZrUUdbvCEdmY?Hni!^aBiRg6jP}!Dpp$@@t0zX!ztfT1FIkJy4
zp@dZPmrMoKq(BLLPMc+mrKV7`Lcv&5)YC0MFC*_&Ih%Ir)1el@i3Vj-<-m0*TTy-1
z9o0`Xe?SOmZS)l2eQj>fh1W$-nd=^J;F5hXiG9m&rE)O}28mJzSKH2TetDT}nSM2D
z7T~z*6W59oDjxVq!I^#9Yg6kW7v8JzFO0NBNB-lmY`$TUeHQ#x-^)o+x2?F#Rotg5
z3w6my?%xiz0DBppSYdQc!@xTnWA~Aiw!e7Xxy8J(s=m5a9*H3C82vswf8}=96xt*<
zW6>14q0z%%n7_mP^7zy_!KrM3bQHgSiQ;$5;M~m%WL8;5^XV;NTu(V%sfu<lZR~%g
zn>tU(?($Weo#?>z$wvjNI^Njy&+<&4pi52ik~q+!$urY3o2~P(>Fg+F58RhL;6d%e
zyfP5b8V?`tNB#*Zx2WgzLi>O-1MVBN`eNMSKTFDg+OydH=aRCdp{2ATisEykb~=|c
zGO<lJ^+<3P1XC=bX_spjqF{x&xtc&UfskSD>p5egmZCbwiZRwiXGW9FlX|25oCPYD
ziWswv{mIVF&P})~42CIT%$y}GEgB#d)E*XcaM~qCZ&=2Mfb<d*6FPX{YWogMX?K(C
z+2#K9v9@+s>3Y3VTYBT`;q9{gL;v}98`enr|IBV01;Cf6{bS^m!izq~sdFWAr61QG
zJlwBd{QT50;zJk%LBdIPMUauA55ONe5`d4KhO79ZZb@Wwhr6xb&Dt9fK=cJ*-w?;F
zpucB|p$N$4{IZOZS@Pn2(12Ky)!1I{NHaj61rPp1ZLObP^;x>p2R?5Xp*_Cs5f6v(
z)OK&MXotlcQB^_+jq8T{C{Wvef}vtb4<W4VSn~2g7Nui*`H2iUBdHotlF7XZ8jN!D
zhHDOYHV*<=<`P6G+<jS15-H%+8}bT@f=5P^npsG;XM@I}lO^U87*Rw&VxFe~Y^AQv
zL`eRAlz^Jo7v4WH>ND&Bz3a7eewIVw&lq;CcZb|0bl57AydW{x?~gis5K*Zy48+X|
z1UtPGbTlvD4)#dm7&^v{f8WlY4YQ&e-Uzi)44fO_C`L8XX05ZU9x5mAK1X><&Yp&i
zx#<p&Jxi}fFfhVd6X<a3I#Iksg{<M>1R_i$E7|2eHKOR|sdGn}DqlvTqpLQ%>x2H8
zzSBUuOsO>*p1<fQZ`Z*bQ{A}RX-weg#P}$)dWfb<WE|SZM`i1mF`6^bl>BqBj6u>j
zSbWES^{}<BVn9)tXFTD=M6GTBu$|da$OS-tzyBt@9dz57m(7(ii(^iWj1<X<qPu^b
z_c(`s%IT<1c#>)+ADpNq;QOqI!s*2&Ya}e&jeXw7Wl8>SwV-6uilFRPQ|K-}p@5An
zpJBLZl8~5n8Y{^_Pg=nLQBU)YH79|Kg)wJVWBn#im~6GOwT@SxOb}?HM5`QNv$(Tp
z&l)NJXdY`+XwdG^wr;inG0=1tV<f~nw0xxppEj)wrg`rRyx5NR)g4~(N%ju1ElSlJ
z9cGr|u@w|fY(J4s0{QDS_T91#C=*L~1P54J0tDe74F5+=1W<k>%ff6&lF?jPSxG@`
zxGzYvLqPMCQd73G$eg$?3?pHP>|x^yLV!mw?!_MyE#z~3Sp-^A+iK9CGC<t&Csl_2
z^rOEo7YMAt3^^nb$3Fmcz73Uzo|RtE^F|2|w~+B&pJ661rrp>+IVoKb3z3W^A%X+<
zpe;;iY`P>Dsa4Zt$5p_h&-EAMvr2gI@NN$Xy1cCHs~^IXZyo@t6-A7iaRlg%^HP3&
zq{NO52sDzH_6szsu>=oR4)iaXbrw1W%1TgcAiri*Ah01X+s4mU9$zTA$7cAKDbgAF
zM+}?#Q5YVkgC9#oGaP_YfO9_>xYfDZXAdHP6AZ*K`p`aAP#--EB$#<wz(<i9I18qw
zM@wnRxyC$eesY|ca4flx!w)fP6bW;-%K`Qb$dSKDAfn!^Zx^Df<b>>5HZqn!ie)u!
znq8HCA~C~43SlAQe5w{l905#>CV1R0xh!oDHp6n(aW*f@wW2l`HXz$zB}M3|gNWxW
z378zp1<(5p#~audfLv6C*w<y47DbVLjC=S*){|L{##0bYUvA65H5C6*MtBRf2K7Vv
zE@hByh_{8LZ&|ktBj5(eVP<s)SbPoWUX~C`1UE$>6Hb}>W=@ScVroM(NpRglQ_@g!
z@m+s?Sdwt#bARe8T<}+&WCL*#<w2Myb+R!yq*}o<*ppqdcsZ?9PbG6v_q|MV^k-yf
zgcKUPBp>L{P~IrjMqy?p@eT4;qj$)$3R`pe+F~C7f)%6NzD29VlY>)eo>hcTy)$ng
zJsv;u>gH`Rm2sbzaqhVmd{@4@ys_MIMopukpAf{Bsz)XoLkLP+?O;LN**YN@2^I6H
zK=GR5l8F$lotAAr!Q;_$7uAsdwQ@%?V>Rj6X-!onMPBv7yX-Q9eB!!NqcR3eNtoHz
zy?~OzEJ#PSK;Y#doFA%?W(GhVG0nXe`uV}o@+b=Fs`FuLq=E~M&`Ea8=S}SIB_{D%
zi@Fhz3H^AdHnLzpg0V4I7npZXe;^Xp4A|NOEBh2uKC>dJ^c_2nZ8L5NlpP94KsVp2
zac8Ih1{GR>e}5lP;I>DvsJ`oWD>w>rQEw+|fmNZtU;L?<=Lp=O%GGcfH1aOY0|QG5
zTUUl4g3=_u%Yt$Ii?c0D13LM&F{}N|2q@gAW|DDl#}I|yaHEFH!3xz6HRDSbiT$-}
zO8S0LHB&K*zZ8E2Qt97hzN5pd=YazEM&_L9L?;d1*EZY4Yn#A9`xoV_kwj_Giw-qi
z<=ySOAxyi2Xzd|)Zn+5J-0x;qF4ssoi5${KS*~;e^@U|-#IO+a2AdGhT1I@vz1M?p
zM&L}-bZMY$kODWgm)~F1^oVF1Fxw7QL<7h7oy>q|UW=ugLe*<-izO!9X4i%x-A0zn
z_k5<jJ-70w??nc(amDR!TCgb6{^7nK5|j<czm~EsN{2khf5Y2;4db|wCn26G%gkj<
zypAgDsRLDC)a5*9Ablpr;5kpd5uXoFx4U|uq8Hfz)d(>Bhl7ieo#B6$b6jctYBbrA
zd}eBAH|ABSKwjc2+{a27*Rq*VJuUG}N?OdqTiD~pW5tK>zTd48b^)P~j#k5&R383j
z*Z@WI_VCV*G+tolbh|&tMw-s_9$A^h#Vb$K)|g07gQP+@GIXj>`$&Wxn)d2=y?kl~
zO!-BbMG+s=i3>fQ=hE%ycKNunmI9H16gIZUvAaLq@0^_cc)TTgem!r^)avHu_VTt~
z@C@bz&OJZ4>~W}w|2a#kRnyo*qj1%FxuqIu8oQYMYMEryP&hJ<|J|fsRE@QmXZzlQ
zWICk3@#!S{K&mhC6r~zVrK-_=+J|G*850xsO=c+u!8%Lpa0p7S!pKfC6Qn2-42V;u
z=rIiDK=t5?!w(vXHA`Grf7g@Gl*zXIVfQ!Uhvh{8LX<VZG9XzmW{9yLk0*6v#gn(u
z<UNa%+j<~}qaG4q$~iT1g7wGvLoI}@kS7P{m6>vUbLU-_LNoj_=Z!KDkUc=RbHFwW
z3o%@la_A}-5E_9PJ>?$3?10POa9sEhB-Q7x+&c8^w|NTbfUIO_w8dImup$0eTG;C2
z(gnPOvks2HB6p{ce~U5SI{{;7cNdAFFM$IzTH(6v#zQeUh){Pq$nq7SnLf@n;5?kX
zAV8FONvZFpqw!g0ksijd&$V~X+=uIVHGZ{BR9iNqnqM#@{!{21lTb5@5R8=qj0Nfc
zV(cB8Lv5R=&DgeW+qP}nwr$(C?PLc#wr$%^c9O|EH6NyG?y7mNzi?J{udBP)v4jnr
zkQ^DV{j!5swlU3I`@3LQ5S<~X1tdiR8#w__%T7#0O5;Kr36;Stx8DDDuk8z>Hb8Qq
z-h;g=_n{7DlFeyAHtby51Vqx3CSJ9E7>mRi5W^PiV`efc)R!^OF#c0%j)wJd?w)8f
z7a`Y_p&Rn2pcQMl8ywsl<A<FR-;=mQ;lT1i!3igCM@+uB15k{%l9yq)1FA#txgUSy
zGzmqxgBvpsN`0IF^VuN-<QeDNNfA|B^3*bye!Ym-?P`%6`up|4w~0m<xi@b6!k*3)
zVWQsp^bCb!O(5Q>m)hOso6c^xkc&){4$Pu3j6)o)>ZKekN&)xacrMWxLli2~<sLGa
zFw2jB6n`h=R?T4vba^u>t9i`!nnmBtAn#lm$A2o@CCiSP!~2t;Odv;hyER+2^+wNU
z65X{m5+_J17?eCOC^o;5LB~N*4i?gQ_D30u=@6rLZm9CGs1dx5xB8}&0z1V%FWnQ=
zzhi4T!M2ONy2T0p+4dX{?Ks>eyL{NFV}&UCG%eTAAex`W<TMZxF<UO;$wY`G?%jt#
z9!vw)4n1eDVv$)60o)j-_sygn-gT0ITXlb;byZ%CrU38kD)6;#JO;>Tb%%BG+3pZJ
zUaxe;tCRDn6gS6l<qDhT?MaTChc2<epCKY(SWKj?$SE(obC3<Nx(7EJ=z>Q~;*y!H
za={f2;o`D}rq-QqLG@<#SXJL6mLSn`tcEa-mB*zHye0Qi)8*cAkv?gqm}|!QywRd#
z6TrtvsyKsa{&YfJu|;`%7nqNWXDp5W0Fu7llJS0+cyrMp-=BfO5!59DJHPB7(xd=|
zNp<cjTB1pUaL^z15)f`s&U<@B8vvUSWOxtG#*FP+!1qNm=C(o<nvg~^RW&t!$)4WO
zqTPVYZ_0~Jg`~ao9RAE1AmmaBz&#iaP7V9$bFXZEB@fRKFKwBXO{1<IviJ%K)=~*a
zFH*H3Ps5QI6<#5ll2V+D4@sfPO4@*iVe!}iC1ij^Ly?k)3Hmf}$L-<$&p|H41b`R2
z*Rh5;T9k8O-T}s%%M8<4YL73<SQspnhBs<aS3jnzNAO~<d=XbGIYHoi6|PU{W&D7P
zPg|w$64uAT;wdsC&%dwiH><bVEj|6dmW#K6$WYoiID#$|R2CXa%7ccVZAER{fMT9W
zb=3>Jg0Aua57jU2H;sZn64CY4)g{%)Vc0!!zq~>1Py|nlo*$}K$OQ6@5YRq@-vuqw
zF7BTQTC&74wz_8%^?8AD672htxad@)Fc7%-Z7J4IV{_;dF)vdEbw$7vvBCNa+paOA
zi7+9Z{HW!Z1mKC;Xm+p^haSnoe(fWX<RWGTqhWfWOdNYfgg@pBujdLlV~KBT#V=hu
zFHmlNL7>{;q%I+j;a|x_lH+G@&Lf$(($yr=)y8-b^p#}&<y<$uRw+<;Pkgv4&TQSv
zEH@!{xQ=oaLgXs@p-`wRbs>Bgx^2Q&@g*1#*<qttuC}bUbl*Tpfz+|}eiF{<OqQRo
zak7rufCTI$e)fnNkoynv$wV+q!X!tU2vtD)_<hm|NOxW(b>Q2iEzr~kI5Su@TP_0@
z5Bbc4*%yI}CW&Y_hMBh8FhD%)>VS75NxG@%t2f87)>_*pn_TX`Oczo(1e1I_Ced8B
zT+_vVn5kesBl?6eR^<(38<z#N*C`^t?qU#_PlL~UZBPk75QL{CcD5X-l0MkoTB+U7
z3;gEKOSW)f9!NE=75-b2=+qvwgYg>~Y$e113??lc2v^8VcZ{4|^QLVQ?y3Ar+ii5+
zlMt_b^&2z(*WY>c*V_rMbyLy?>NU+1$=e^QXr_Rs##y(Q(d(NAUm^XB=1JtcIPq{B
zJJRzBnhR2`q)aeO3}{G!&<+)_{$P%F_K#KPiP*<OMWycGwK}+-&Ar@aTq!|rppfht
z5f>?fDSK4^$~t9&ww2r6TP0S1?T@p0f8SF<v-YVbBXeU#pn&AYuJ5}6#RPZ{?aP4J
zLwy_8xc4PEV~odcO;tj1Bcss)#~(eoq(702gy3X??gPiK#|smEJA-}vyH9_UD;L<l
zR^LBua_c7+jMr^@_1XbWu!W*kwRFtk+G!}F5$yndEXKLEjOx+{d)^0v^PwK=m0yja
zb<(h3w{&cD>f-gLk%bWrX|`@I1hx>i#jz|2#qc?=c0sP5=BB~srn%1S{Klh;DX}mD
zYh#0_<EBJmE>dTC(B~~>5G4R)#tNgOh_MO5QYK@;XV`jmQ!iaTFM}%nU})*5Z}#Zf
z>WKt`3e!%B=XR<397IHkXK+A;Ir{O>g(EAt<#9NQ+%h|gycUlRj5EP-$d*V+xN!c>
z9ywNdp-eb0PK+~uFViQFf9Z%hLO!%hdJrg4U=<*X(qtAIYT;b@byG(4Msl%Zww3rr
zV239$S#0~@Usgd-&`a_J!JBu1p<a$=%sNU8kB^1Dc*W-?0wTa89S9;rJVfqjwBO&f
zC=wMNDR#|dVzCjbGZIo!q%5bv0NU3C8@kf?2GU@|aT9#PxT6<;!v#Y*Fc6?!-%b@;
zl?~0%1?h|vPXjJ~&H#?qONz_U5jw$NTH&}-kF}I|b!9m$p{PX-!Fmxeu4~NcAGZ4L
zQmonAVVCryg@~qW!RENi6Bgm0esY~siSK|@l99Rz{B`S@0+~8PHVpG;%LtrWeVQK$
zXo%r8dfY`;C~@=a(@0g2?CpzQrxea`J^9{U)9S%wwn~1oQM9w0pjz2KE^wHU%WhPl
z%4zi2n}+->0zQFa08&_n%Ye4(H4srxM*N}#{?Ir9lJR$V>L>(j@7FIZw8Fx=NY79I
zm&ACLd_wzTe+nR88w_K23MPw{sq;}M+uNyWo?YYcIQJ4t&QY(u(3ZIFrfJd?_Rv*v
zs!O=t<MX)k)f439)^3NDB|9py&V)-oP4ny_h9v~OYHOH5YkZDusvzom@V)<etJk?@
ziJb`4EGongr8I&qXp{lZ`_*!C)>pn2@l<CwI&>?S)rQxuoJI=u4P5{A{&BGW1L&}Q
zJ@TKL`aiDKI9dOn*&4yGVz;z$F?AxK7qc;RF%>a2wl^__;^TvIc5yN_w1x86*w)yz
zTV%lKy;XjTL)DrMUz6pK1a7m0L0y|r%_ORJ?G}O~g*8SEG9QghZgErZGJzqnyI6kq
zzkB^6hYzF_*z0&Un%m(2I|hvR#IbL{0GI-N;t+uY24>8#H3*@!cEc3+5+?^1oc!(B
zEse3kJvVtj5dQd>9Nhc;a&vV(Ir#YaGQ9WvohZDz%mqU<lB!yp+I*z|jT%RA_6%X%
zua``@_Ohs{TsbIsDGg2C3}GNZDS`nUE)WBW5n5}L?Nk=m-R~Irp?aQdu;8`3G=344
zQRRD32`|=~(CDmoC)fb21Bz2lioX`Pb!~YxLWBzD5+!g|3yA=yJ4+q-T{pY)m+j1`
zj*Yh!Ng@@#^B;{AthLqUd9CV(J0S*>PAA&Xu8Gni)htxEgklN5PR%tIw|Qss+RKM$
z_ax%RC4VJg#Gw#)_YXTkJr=M~D>{+n(1IG@mHw9AHG?}zB2mNA(dQ~FjtH50p+xff
zbD4ea!3DaP?g?57MG&u@)!TfA)?_LR;=B<Oc_EH<G56!z;G8Dv6qn}}(HQR~(dHcP
zq`-UPk0yDJ6Ch1@9_)>;kS55M4hd08r3_l%aJ(l%HP#M5U-rT`*uR^kS24U@>z!g+
z&wQN`SEhW$wMQD6x+99%Wd3ldYr&Z<|B#7>8Ze>^{V1+E2XLm9&)Gp!#_X_zIzZ;G
z7~sygo+ommXdHhOa_oKO`@0c;?1@SXg)ht63wq^y^b3ECYN=1{1&Y9uRSP@nvIw$T
z`=WX~PVGUDFw93Qz+HNq{4`r~{W52Lz1c5%{b7ya;brlmIoEPG@gVH`&B}>frN~>`
zv@>-@Z>^27EcTLhZ2(j~Y;G<y9BQ)&tmnvT*_&JR;=eL7hZ3kLxb=ZBrYUqauZy|F
zE=ac&RKH~Ydt_F*xU16rlq$1C*VJv7&0bXc^an@F=BJK)$NK%xI_GW0TRtn<rt-lG
z1LiIUBj?9(@$2QMmSV59nkop!6m?NY8gWOCcyy-0mzJm%*gTkXsm;NvtYabe<&f}d
zhlkV1r@?wbu59qhDZQO%F0tcj#!3B$C6ZZJ)b<vU&p<w7Sp5XG*q;7ygy#ygwUB>e
zAu4rN;(iyVXu9vph+&!H6>`kR*~0~9g6ne)scAR@gV&;5FT||((FBj))doA7){qNK
z2~Jrs^LvlwJ=GtkD$AiUJ0E{7uPy-@m<#{fr5bhDH_o&bns*kOmJV=mjKjIDm?b}7
z_?7!FYFzFYJVOWZON#$jF<|-+_XIQR|KmRYABsV#lC1Qg0K%VB^{Hu*DL1mud76X5
zHcQo^6|y|3M1#dSVdz*yko&&w={8brgc3cbyOZPT?Dk*<puc0?(>*f^4s$Zq5`t|Y
zbplYs3KC;z43CnP26?<LVIf_j#Tc0fe<N6GzIE-BNEZoWNIpVnAWDV&m%JqxKX;Nt
z$rl~@B^ODy&pkWb17zHZDfe5{NdK*5<$bCrd1}W`X@CIDU?K%73^A<SwY|7;s4v_Z
zE0j$dS3qfa&7Nl2gv*o=vfWeqNo0a3JFcF*+A<ps#R`;6i#t`{5Qw!f$4IoKAP7Za
zuxk*jpk)_LTpA&UK+W&^9c6o|)<?OMKPhvYkhrMh>OEf0+D$pcXoyU>HqKmHJ)g(0
zoY^>->cXD1#jT2F>fc(7vUZ?aaHY8zPaaxBce^y6@O2_XI}U{K?VJvOhvn9iGSl~v
zYoHTKdaul|g^#|^!se(sNAYVb47va3v;9X6XJY&Xy8hqa|9=gy#Td8S6no06-=k%*
zk~P-{Y7xy^0R%#!zyhX4I2BM%8CEG$p^N-_^=)>*%t~q&3@<d>>9O63OYR0Cug~9K
z|G0c^#;=Ewj1WqJ$kVY$B15i&MG_-jy*$#Dg6Q(_|1#@H>d{7=vi!JkxN_4pp7rZF
zpkW-7prQo}VgOXgK?s+TAYmog1aA6Xh;mBByM!)U92|AheT$u<XUAwh=<V6=w9WLJ
z!22iYeRuiYK<BpKyxDp@T0_^&MxdE4tDWAitBqGBaP~4lY!zj!W^Svwtyza06O(s0
z&S@0o$+uZMJmSzbi<9hbntJ_D?svc~@W3b_bH0w7WY<MyPcMOm(ZC%rR1*H@w+`#0
zj}~d6tKxkF7eVK`d2VtRfp%#y)+}Dyo-f&;N|VDdM>R0y?fYcoZWe*@Oq7^{Mnp)#
zU&1LQFF6$kJ_^3Wjnvz3)^xR;a&6;0)HO~xf`<br+{AYq>8seHk%atd8*5j6*H;Ib
z(ron`DiDC03P27<NzWTM&ONrx)Rtu`TR7lr?!ZP+wm2Ae0S{(~vGn+A=%%w$_}KVa
z_W*qT-mQNYe%=q0Y}<5c-Sx^Hi1vV~=jQnP46ED{bm|flr(*@CV-zZon>Wdhrnt-!
ze&LBsgYPF1QPL|NcH5Ujt`NAGj&9cpjLI1uYX1YU!5gr=0c{YncX#IiH^nNOsrsGo
zpPrdq+PLaTPdWx6SR4QQ&nVe%g<x#sutcgp%<<)&-I}*F$>o56bDwVdn#)r}va*J<
zpvOu7_-{`{n_Jj$$eWay$@lXXuBBUv#&_opH(?bR-5cI37!=OnB?sg8hhJyI{lGvJ
z9*@S&3L29;b9=mb0(ch<dS)s_pa-67_)k2G=%U7BVT|WxIBJc}?%V9(NF5LDW&>hp
zuRGDGi9q5zDNi1^^`E?JIt+)?oL-BM=D+yEa<gFN`p<YGLn3f92$(?&EHo7yX<mB5
z5H5fgF#dtGZyyd0L4)e*vc+Xjf2#;*2INpq%^X=@2n~`AOFk7?O5xH@g4k-Zh5x+j
zH)qfuU~CkGYK%gN3V~twaIJBacX|V9n7y?~SbLD7+wL#nuIU`m;PzM;od*IiVksIM
zQT@C@r_6;I46(cg{sGFLnai2{0--0A{Fn(w^5}$g3XGTxhI_~#m`6d1;(*9@c5Nam
z;KJ4EC?79lA^p2|iOe98aAYLPg5BHkqh%CKyia2_5eAz1FhJ_=V{ka&r=7BW;+QAX
zzY~8Y6fiepXbce?gaHC~Ug*b%Z(Ok;z`ycU-n1Ro9(#@CM&^cj1@ahKXb@9DYeJKy
zY1Jmiv38PF<>egE5MKB%(m|KjGEO*-S<RVjOpEUKH6=JCD9tM9gNvWm2fQpJRuhq-
ze8h?Xc283_9rl%V+idU03@g>fKpr$7jU=&kh>I<N3c)04dOXyFkwv)@mNes)5)-Fu
zz%Kwjb|5ok!Jy^F5mkHIS3i*}C@u+Yz8k^}`!UmZRES|b2>|-I3|B|)Ao+SK;7G)Z
z)S-8#5>v=~IEzLQN4Z{3lu<a%nMFlVYy%SWB^}ZDK{Odu;vm?01sSA(Qt^(kj4Jda
zv<tSnb&S$C%2SzPr?-a_W?@-uuq<XOf3*dpWMD>tzU0CNs<(|n;tU0-Gw#*p?1#!i
zKk=MVn?Zn>f4oUyfnLJBv2yPjCJ0f&8`Wo)ChW6jA@~Q@q+;>0k7ZO;33~)I68|B(
z#LshCa|#+U?fQeuf(wK!GQ~GRWBAJ1VjBeNR^|{1A-yA#zRxoVqv;=iw{n-#%roXV
zjA)Iw8{~=Ob(G2iI#b3KdMP;acT%hg+^ReS4~ayhG@{`IqC%6~#{(6LSN5ms@)Bbt
z#xcPZqY{v>m4EOHn5YcS{W7vKo;2H{{MZOCS9}6+`k(z~G}N9da0!k5jj}+7b1L9-
z_mScSk6ZwQg+yl0%g6_e4A7up2)Y~YvY!>-{R&Tw3fIrAQZLiTv-`!^filsQ7A`<d
z*Msw)8avx)(RnhkDL@{06~vaB6el<UZH01G$d>3-_7h<l-+tqM1Q*ciRMiyyA}FLN
zF$TuS6_3e*sKqefXAC$6gAM_7@Zfsrs*2vggXf!VZinRsJgdm?i$FhL;kdo5Upd8n
zX{z$_NHqCM!Z|R$o)~Wk+!?^sv!f?|x-6VM&iiq}o10(VdFtKP*ygO6Y45s|P1*1u
zY1cb^4K}Zn6Staa*L#p{b?y6hs$1>C=zPCtnS_Y!h0yj<jqPc3l~-=ji!ulo8b_op
zrYx@jUmcVAT+0uM^3aI!8(GrufG2!YqL$k&hTiHev*C>dYRGn+C{9!?hq|qFdvW<%
z9hw1$a`9CzJ?3*HOR4;rqo9U*7n&<yx~%w7PJdWjwmp2WM!TCa1TacKDTS)pa-$+j
z6)XPy0lIYpTz$D5(tdHhJ?|dWjX)AzPPrDg`MZ8Pg(oxyAv|}l#@G0hOq<rXmt83W
zSs_xhDMfK5q-%hwgNH&P^vTiq3zbS3rJZoM%`sXH?S+JIDCbOf+T-CDT<FIh98Bzu
z<k6kt{ub}IYa~le@=ecLqHpl5B1?4{1KD=+g}CtVBJowh^Sd(u(kisnBXk)<U9SYg
zk_yix%Y70E2AVoL3MUPN433$hl_E7tgH(Q5a?mangesg<UvG*f+9prH9vMbb+8dQ>
zTTnQ^uYhQ}iZ>>V!cafly1JhXf+_x+j)%$<B|Z7a?4n0gD*XqjYTD_5REq}1?lB19
zLM<^yVRs=M3Bf;-fLr`7q_&|kyGOSOjFbhr&8$C*pZ0gsF=PVp7ep{*%|on^3cg=>
z_JN3wCZfndYGFqx21GiTi!$ir>tp@r4C@=~zvK+_f2dF<W|sdMf^_@~K|1`uN%V7y
zbxjN&NJlEy1DBZgNm5kWn&nig+?)!(e1CBw)WjzDhs#9+c8P$Zn#`k&U|J+AmXG^q
z=!57}Q2l>TH>0EOxA-kuCN$C}O&&g46()u*e>>QsN2d=~OC=Vv{rSGnzk>HP1;|>B
zR<sNjl(&EOZ%#+${kt_K2h)WbvQs7B9=_}Of4*kj`#QdEIdaeV$w%u-rZlq5?hG`u
zNXK59Q^+=8jUbS1f{Q`!2r1=AC=)cyYR)8Knt3#9KpthNOOk}>K+LI7fyR+wp@Em-
ziR;feN@MtM3~xAKM@?7blWAJXIBuBPeQrdiTUNgNcTb;@H)ilEcDU}?OPGYhXUwgh
z&b@9}1i&y1N-^Y@UB7^s6ch+dk6uBhZs4cl*@<-1Ad2jy;luq}HBT?C)Lr!*8>_k>
z-ZO3^UNTm%m7TZR-*$LpJ!~H6;vHHvG!qi28?sHLDxe*-|8+6t(7R7R=Gc=!<U`T>
zIs&R;Qgj^1I5<x)E(B4E{arMun_G`GCfBf1ikihN2C7z}2P)dkPHk96`m1Du#4qxb
z==5eM$2x#ySwEyo<|PSm!@E4Hxy6b7+~m;&-H*(K>)<eugeF0j)&o<SSYt(Q;cz3O
zk>biNZ~?<2>dbLn)-aA_rX<@ipWd9(bgBtu(jHmLE0xVzT^19(kyT0rTXWGSHD_+R
zZ|#iczG5o(gstq=a0t%~AeDjh#bwUxKwaXuKG;Uy1>l$1Fa?v7cjEGU^>Fe-gnD5w
z9r}^2ypL(Nx%dqvLvB$wAuZ_;|Nh)r^{C4o=z2lmg1X!oda#i_YXM>~6J`}Tj@#_H
zp61G~dY^#*=)0aex^lUQHfp82Jo^|jTGj8K{@zC(COT5`LEoa9*#iZ^tBy@v3^9U>
z-HQNSof%lMA_+wc5787qWdhD@X)0KXy>C_Q4M%&tBP>m5E3>9O@u#?SE4ev^JU7}d
zNue~yk`wVhN{uRXEl`ebqZ#O3m*_DDd4|ajeWSxCiiD645>OSosa{@pc_{;|)KICz
zj?NTmLf-yJ{PxPqe=XIUL377G0wGYq!uK^Gzt4q}*Y-<&+d3FtU#bfFEWtmQ%E#f~
zCOl&ifCO7&qsknxI7@*&K6K#&zz_}bwZBcjkVC=Ko~m<hJdAak7GLf~5wK1`7j50q
zGuV05A|A?36}l1HdQ<rs`z!Y;&Y-P+hxhu94C!edk@AhaZMeMkMq1y|GM)bIgr$H~
z*>p+8MC2DxKwt#eMDYa7kwwbg46cCyDFc{QDD*blt?3*<z}Nngp%aT9@^y1L<aL$g
zen89v1l|IUV`49>@y{<<-{u@BC>w<IT6-S)1rtEVM@{J!hREQn^nO#Uuxza_;<4PU
z-CA!Pd#4iO%4yl{nDC-Ax+yoXi!F*Isae7Q2cSqSGw?D}(4c=wE%6d6sERpA4!4b`
z&PD7y@L2b`bW3%!Y`x|2+?+#%Y6i*Tx>%WSs|IB@?e)D5V&<<!pz*^rnEG7#{s1w9
zL91t$Yr}#}g8yXCnxAw+!hN=+iEB&c72Q+SE-ojsdk0U7>`=HJ|1!zXVJw2tIUugI
zXBBlHMH>Bkqf;Z#Y1|#cIH~|?q`cz|f{&*=$5k&;H@nm1*JyZJMAmpeyO#Y7EswF-
zxaHQg#UQY4gUu3hAAlo!T;3>p!gUw<QfvROMqp}I*r^~0QHW(qfNON<gn>Sz%Rwh9
zo4HY<yJf_bTfi`sojv{=K$BvU+kPB&cEO3?PXLv#A|4>Bt;f!{%pj+DZnN_RzrRrI
zXklRx6o#HD&`wLi&STfOoo2W*SasglkMf@Btcs>9N!nIL4*J<yBi)sfAgo?B76hF8
zq*OIG=d^B@t$vVEpPl}IW~tXsx*eTBcm8Vtfo9+VQD@6Q*9CN%AXbBH1!=i)#9T8u
z#qG$dcA^>q@Dz!!YBQaOi7s7rvC(V(%~8jWVI-8wfx;qV%s?)dJLG}PiRXmj6g-g!
zwo?iWhP3yuq3kv$^Plb$a&l(J`?{#62U@dUoP?{;B3Wa{rr=~<4qUCWpEJnyk~CU$
zk25^3xRCXv*w@$fD5E{|#pqhAgx|Ihyw@}+Tv<Y@IF?rJ{eYPCjSsKCE~KGii4TgU
z<-A{77((H2BT6d1b?q3gH;#bq1stWw6n)U`f^m3NdbrMJm`1@r*>oI<S0dUD6m6G7
zW`k5VV8)YKpw2WK&=pq+72_^%fD?dB-9OzKV`<i`dALpxPMS9-n#7P#axk%$@3zZO
zt?uxPI9BcrC*$MS%Qz|bKyUU4&!P;o0oBEV2?zk(Jsd<)=CyafTW#fI0m_JEF<ha@
zr_#}{*^{6s&7Kfk#Jxfjt3epX0E7CwfqSFof7z}6nsO)(FP`oR<s&ZAMN8*IwVH8g
zZI3M8IAW7+58#_zvKN)Ema_Vn?{yrurGEN;65~s=0d_>v9dyaQSPb)kzKa0^2%}GU
zGzDcyRrg9Rn;j^VFVP@IAogSDz!(xLeUy-Q>P?6`jcP(EE;H1DuXeLHMzgB34Sw$;
zLV9pYekr^M9V$YHCiN<_b5MCds8cMfh||a_m<}j>2(lV&chuY6<DI=MC`@P)(m7{Q
z0Me|Y^>2>}BfaL?(yK;&ZTj%q_%WQuoj_~ahAH;`=pW%|SE@+o3ZuW^ZG7ig?x0K_
zR6$9f&){B!?#F|aI>ORjWW)(W?gTp!jtslc(Zg*)ob^x0sHni^b(wbX4-#fMxjJl4
zp<bZRl9XL|PMTq|{tgH^kH_-wmA2IWBfba#`F^w874FASVBI*-U-hiL<Bp2uGW{EW
z-W38$q6)X`ml?{pN+ywCxa?x9qP(wo%KuDy_};F}OTqHW=Zgnub6mv?ScyZB&ZUz9
zm3p6<TiJfG9{jUaIe=1XxjhWkoWoBJ*kwkx)wJ#RPf)_nVywk>W1Gekeg%D?1oGRc
z{cPis&I~wbGI@`uR+6yG4B98yI^2Zx_M`Wf^tIgnW?|cU{Rif$DR8#1T~AGDK4D7d
zY(x1qFJmL@@psc>9IMRXOoll>su*9(=iMu;OB6ug|BK<1_`ZD9{Ri+%tj6R&#n*qC
zC0UsM=NO=EEp7YF5hTCY`aR-UvFtz2?6yE1jyNS_j*COTge)vD+Ico@$tA>6j(g5)
zd;Vs0y|f}awXLEv>5dXJbF<TrX)cbgNQbA7z0c<0a&nMM31ZsHn3+0aGU5T!zd9o3
z#b-QZAdD^?2uP##@qXWT`_uF1`vFimegE#aXYJ#+K9>99zn^ces|TZ-@$>Q=R?J30
z{|W`MjLKRGy9E3Jnh%hQVL<$ie<Eoq#zJS0Lk+~o(i0(|)Eu@tlb|R>9EhM4f`Ju>
zd5k(jr)el&C>ahX&RSxb)MG<pB-6Hm$WcCj^Aw%h7!n~AN)E$1#1pm{Vrh`&cui|?
z^Kf;$%LfM1b`&8>vJlCL06~)o{t8bHeuTz}-5bRxLMT^tF_$EF0I8j7b9~z0BR?Gx
z&>D`ijwEqKs!YTnm1qP(QIv6s+2&O>xYqwlG(n(tuv1m^xjHV|BWW^ZDgdc(MgT0f
zZW|U$K@?%|g~3*>QL|ZsL|1@LZX3SIP7Y#CZBqeKEI5cjsQ(wu8dxe*!)_uQ6V?VB
z^kkeb!a$p-*k5J33KALO@nv~Px^PL#L3CRc`l$r|GDnuZTO%nbB$~jjt#sF$J`Hey
zj>!w2GN_qSteYGNVo2&fEzrmc+!<6wzFcf#AXJ#q^o?@lzl(yRt$c?VqtdL!xUhk9
z#C==<;l#kAnH*$PeuV@-Qzuv9Tm$6!b)QE-WG%M)#0fqnn`}FySdhi|mzbm6==qeZ
ze6Z>;+qE??xRWJ3R7~fxfGtF9&`dMa)wpO`rTh<*?78w2;oL#5x>`_+92Y0<9m1H#
z6!Md4Wk60PWL}Lmo>M8+T&rqf<pOnVAW~Cl9T2CX2|02zu8a)c=+MQyY)?_>S~WbF
z69HH2tj?Jm&_OhF7zsEBQ=qPU`})6x3XV(bVZVG71vIYD71fM1%{*b^IS#&80c<R#
zXx5y#B{sfu`gje2$HX!SJ`8V{DU(a%{%R+mh@VdfZ-VKNj7mraktMgp-=zB;LI9zd
zl>R9;1}kv^;;>v?B#hjMvKzrYL$;Q6CuMC*_)XOOCGP&Rx#D14{mUYg?&zl-Iwllb
zd_fd!bVL2t_|-j=N{uC=M`4U;`j43`g0zMgLwy^^a<vHen2ZmV;K;O*XzjkS^86;x
zbN#YmpQO0$kK_AD^4#*p2P{9zZEr{0yrHIs`#^?TC!C_-EVjCwv11l?>_EQ6`yon)
zWcfwMb?udGiD1C(BcJTm(cXovh%k$wVzT|b9+#qROG8GQlNqQ(mqheMsm5WRpMOI!
z{WNmr#jwK>WCTbbP6Wr(X44i)Yfe3y%|0oz=<f&7p2&KnV^n3PK5Cd<rlvKBY|SzI
z&(?|Z=r=`wUC6ru>`D*8=+!WI1lOK2O-Thwa|FhO-AJ_UNs*KegEn_8z^(@ImR9-d
zO`Bhoh*;=J_|2$!J5TvSZ(Q>iroHAV+!mCLIUtn0(N0*AIL&d5U9|!&_qrvQxLbfM
zPITA<3;+(drEKvLLN5<+E1(l=u>{-})X@QBk$cSzb*zi*cwD6!k)Nuq8ublyVQC5h
zCS)4H<sxC;vS-Bh*z-uxrj-?zDTE9|?$8<g80bDU;23~HI28I9<RLtZofqLLjCd>y
zX6*V)f&n=9jOVteF0h*=h-j_UR(+)g6YNH?f?9OuJ59|w_zWh(E(iy9pK123=mUMr
z0;qtGL4qv%0cs`qk8)8C9Uar<?JqVoX?4KI!H$WtJMEc2D>2MU>s#iYea>^IM%IT?
zUZUn3GH6FD%O}fk-VGDGCbqV4gj*a2_7b4(65_FffU@k8t-RtuXaJsZ1DFB8uQ!I~
z9XfIX&P2`Xx^4N}{;c&tu5wPDI(}=j&Dcj9so0%c2VAqPu_HWOITUm@&`vr{^36t7
za(+kQtQJ1hTm?l)xjT@6bzmByizRd5#x^Bt3VqW2dtCWx`}M18ci_ha0tY$rO;;r>
zB#O4soZT7N({UdJRv%L^#`SsL<kcqY#+eseBTwtkl%wCQ{;Vs*Mg$b3vOR0MK6A*C
zs~5LK4W}^=7PNey?yGgp{}`~NZX{4iIG+%Xt`ic{)b-~?a@$>thgtx#dc3F1JU&_F
zq<&#ECkZl&DMc$lyyYW;Ytq#B|A-ZZrxpQIi>4az;FU%%`BMRHDdSQxrU&&zGZ02w
z4d*|5YB&bix$0VW+iQ5=Mn3LTycGFz#OKD-wQ<QxlN2z3%NfJFX0`D4<K3Ea;eDjA
zf;qMh)>4PG&x?}`NC4>BJYIEGxP?eo&HcPAo20~9!%p=yR7h06-3JHH9@$YT!*Q1S
z(xnc%TuGldv8Xx<Ga(5TAWkFz%Fv|W+C6?BEKLCeUGSmqs{!9^F6Rs>AfR{h-u<Os
zo)RcQmrKTygO47!fh_ZAJBN@leq2o#i1k*-!B_^U&5TF$2Ye5Q=B!>VT|9R3&<<et
z`!ouA;>$)ag<ot>=a>5pS^sjW7jb?kYNfO@L%DZ!P>T&%y_Op#U{+de*|IOTQIe5{
zNEAI8?ORg(BbRQPS+4xr2q}ZMdmL`YQIi4sE+0(U&_$~-Z6MqjK$p(}E6yn~Q=ew0
zLOsnOV&S8uA@0chx6JHfpQN{Ydh}VmuYxtGRk6ZlQp%S3`o8K6AliV#wM2GnQ2G*w
zFSqbj@JM|7^dFhl>*D9C_YIU?zhTKJfPIHp&RQRQ)~pN^LHIlnAGN~|Pjnr;--f!R
ze&y63a-VVrbOA)i>>&EvFa5P?Q!V88J<%coSQ#CWO)osPeg;*84fQTOzVFNFZe2E#
zxR7`*nszfZxPT?kf7>%O-R6M<VhGz)neOZV0L49Tj*T1=Kf$dUQn7tJ?Nf(zYxO%w
z_`g=18U52kB1deVbb{#Rymwaegew}by7+xBFyzvce{oH^+1)swJygx_<lf{;U|v^d
zEprgSvH)FU4Rv3OTTcJfJ4{pLWez=k%HMR}Q()1D|K0yHy4lOp{Ke)?zBh{7>d(m2
zzr4y3_47%ta%)Fp2>4n&)c~*g?{227efI%KeznR1EVwC+eR{`NRl(HVme~83ERnTW
z4AFjhu*{mJ+bOf2++J(OyT!rIysylvzRVoYGCtGfmM7aJPEW?ZpkgJ8xL$wWd0=Ch
zNT_&ZP+je?cCER4*Ux40@fy^Wd#ZD2ja4dh_d+4L<ORrl!+(;(PW^|LV`lt+)5KLg
z9ZU)6<&CVAU2LJ~<p>xV82-;@(8<|_fRlp-ieARl&fLX<fRT-Z<$v89E^F=D|B{M+
zy_y4~FNR3|IOB&1C=%K(v<-m5i-4O!JSCOsrqoviZUHxayY4PYT1_U@DGP@1ka8>M
zH*eeS;z}uwTvW9&%q22SL?EWfeum{au@Ar@{X`_k#E}T9OGVbTMoHzhij-iTprW>0
zF#LjxR*LOT#>KaHv!sc4@kJW0Vwm~)tsO9A)qDCb2V0(zL(w3>IKxnO8rRP>*-c?h
zs3Nu+Ph(V<J&2Jhq)7HA1dVS5XQ1?IcNwvILFDj}!`O7T4@iJ2IWJ&*2K|<3_;~>m
zcrox3$ADZEgY6!VHFqf*kXQklk;CE=f<62B(VQOu)HuWRc+QgIZxD)!<7vW5Ce6nP
z1g4W7F*wpH9x*;*NS*}=h4FYwD1Gx3QEdEx-jb)?4e3aO-~=yR#N%R~)CSRteUk9j
z5`sPeBDaz?07&>wZ5af|IE@Yj#F+-6nna)+e3}DPiBKC7G?WzvF&@%zN@6^XP(W>j
zA!rdp+EEF6IDML^LFifmEQCN@5@1L%LP38@EWsiL`33@7!b$EE=c%+v0FqP{7lOMX
zk#Y{%l^Dq712!2UUO=*jD~7=}a5W_)0_e!H_ke$M#UfBkf?TJNI_#1};QKX@tQfr9
zAOQiG$;QDF3=eFkC|}s5B^}2G2B5|9_W(>zGJ=7ZGjm}Oe=yobgE~!yC@rQX!-91j
zq>$*5(px4^2r$#`2VI-~8UTGFXK-?JWi8vMMYnQsDLrEYGsw_Ay>=k&J?F;I5D{Zz
zmX?%!v#EpiL+sz&ByLH+s&kdU?sG~u{^?6vDpt9<C9H+tVIn>l8H`46{y5D(9v=6p
zl~etCykxKXX837RGiTmPdgK4+%sU$UIQd@V7uq<5ewo^mfpc5q$FEOoZ%=npEgL`4
z_r>z#;`;6ME}9!99f(<b3==nFf7e2BQgF-LqgLMeczovfb?vUgFqXe`@bY-t>_>Xe
z+9+kA4?A4%VNbj@)Ejz;H`LDD9IK6OZ|NIo>u%=dtQga~jC=NYI9DH!9!*-4Yj=Va
zNj^(T)bh`KTGQ9*(=oT+zenpIZ?!(1J>Rr_6S}WjU+wO4y_{@sLO&e?$xsML;kCEA
z<aIHjD&_HszGvGU_e>?1;xqgtymTlHzxMuk^>GnXm(wh0p(W9M>IxHbrj(Y-gQ@@-
zDINBkqN7bj#!IuS#C-diiq@3H_Ks^)>pW*0hu%zATvj`=)y{$f%BOUvf5tVF{ekU+
zirpcHh>kd}7GB^ONRJh><(i8U9N@Lko((6+jI3;wTH!Ti8JWvh^8-u2Be;VhfkTOV
zU^KFW`Q1S=FD9FTjaxmfLi7MpJ2YCRQ^T0tK-2egKc612r=-#qz+^lCH4klbRA3Ge
zjXhVt)Ukml?|Vs`d+UaJ*l~c%)Wc3JKuDPeb?Wqk@T%gmK`C6r0YrlbkM~A^OfMV^
z?lcFC?a06_4A)vECewIZ?Xu?)ZsfU^D>ixiK?#I=4XhAGphHf88gTn~^3-c%u+e=Y
z%I41Ajwsi9FWRSi%`VL_Rg>aOR+Jz!wu~qW*K4x<j|SbU1N$hLi#b2$bwE4*rW9Vc
z5>ife!RRe7_$^yD>OxK4O3rty(1hfiBB-MX>JZ0QQN926<{4tGnDSFxJ%)H&H%AlQ
zXyWuz_YzsvbnGf#h&Q%9@{-e@5V!*i5XaB8o(Bp%mut`BUg0fE(l)M(*Tl2_I|FxG
z=XfeS^H(^1^NU>{_H~c@zEYg=C=|yqBF6)WizRDV?_#`fJ=KPW?2N!dw>1u@HMOYR
z%uQZOpwmC3AUw{qHAV}5qdCKen<n#a-H!S*`ho9bRmQ&~E(AFfh-@E``l+b}&Gz@2
zaa%Vj0>ZN_md+t7B5}RP*z1xL&4&hwVZUp2p@eU6-pWIr=quvK-q!V!Kj9U+-O{x&
zOz_>4W2x{^i^8Zj3Tzirz4QRw)q6wF+itA4YL|z_#WsC>Goz``l3$ab?hmh5v(F!1
z&%V0;Jw5-byR?>9Czric_49V?)YMO-Z|Txj+w*5HPp5M!)&WoNWJ?QrjwcFK;Hiwf
z1~JYWik91;mc&!9>9%T83z7e%!`F_DK%xsRIig9piJaqbL2s!h*{3Rz=9A&7EvYz`
z3`Wg-Zrjsvq2^X1oEY-vQoXC~7ZfqpCEdp4IYjq{gKg4Kmk-@+Wa!Y8ha3W?SWwtL
z5a~zoQhY<71F<puM;Bya`HyHeMyCJ0%e87~D_{P0xnI{%F*B*y!*<RBCf3-wi$x)e
zZ&8f+nv}vl>8P7ifS#YXdK!++DH!A_X&b6#>!<H--O7kX6vM~Xr{fkko+jCY>n<Y>
zOCG_ZVB7K8TGKp-P&<~R<B#3H$NLC_KE~+=nQd2H2Bm4NZx=V--i~v47;Nj1v#-Ew
z16vQ*uhHZ6TO(iiYTU+~?l9Zp8`k_HS^w>;?hB)>&jsQJS}~B!vN%Y=kl*nMED9Mf
zqa7F9)N4Yh?lyBjfpGGU{ij!FOkxIu-P@gUwv(dukr<(TL_AOb5+i6CKsWRmva~yo
z`aze0VdgO{fZ>xj=Utd?oH$q`mUc$n&6?h{R*tW!?V)ZujQFchU!H%}%Wy!>zx#--
zc2*~4&d6@P9|v#0ReTN%Yev}{;txs9W?T(3D1yeO6|6)>jcE#=xzp;GxcJiLoHUu8
zX$v`IwWgjk3Hw(y3jv!N?y^uYusZ{fYu+%;yylEF9_eENp~)Qn@lQww)oq`$$D%zI
zS;*#fo-$nd9U<LpD406TF33fy_`l&aX2XZh4242Of)IJGi26W&m^!33REXfr8fP+E
z%_Xh=uWBfEz^NvP(d|+bEG~n(QWv*>L>U+&?1GQga9HJ6IcK@9Zu^z3l<L0a^~zWz
z&+^wsZWI5^lkT74R#)3ft^#!kyvU;&36!G!v?FFghH!OlXA0`VfVbYGK}y-u(j1QV
zQAL8a+QE-JGpO8s{Odi6r#XSwL>oW#99TS0E2+Ebdw3AsX)M=FYic=Sshbq$JEn|{
z=3Ve7oz23KZ^8}LCJQIh4amL8J#a8Wa<+^v&(&~STy>ShXtDC9cH)0mRiM7-F5nRe
z4w{BN-lvITLqIc+x``+SrO!VW&a9-K8=rtbR7%A&ZQa=KUaW1|NU-~U_-E&VKh$hN
zWZ~B<t2+Nv?zzU+Y}!6(QNU(JI{sm1;yKpi!*ztfDYG>%c+keDu~|RfM0N8RkKEp_
zw{@6P1}c!z(#kFWerZr#S6kA5=kqi_Ej44YB`DN|Si!@idru0H?!~8s*(tZ$KEKV*
z3BeYik-`#dQP(54?~m12IjMm2XKO{VBEqu^q7J_!0;OLgtf^IUlX{E{*EOMU>M)7k
zVkZ5xH_Aeq(ueozFYKQ??>&oWh)4(W1iF@w5cD+jVa4jyYXZ|x>w*ul+&VAT?532+
zFiFe($($fN>6tcn84#cQf%3sk{h-1`vY8Jo2Ug^kxi=BModyQ&1|RO<hC9Vf`d&yR
z5WASV;V+@9a8su!<61;fcbQx|FPFzw!W5k!e1XaaKrJ|M0J$ZwG1C+Txn7e#0mh5-
z)Z0h0iH?uqvfCD`#d!DkdL>^e!tQ*VNy1?{o<@Mz1?r%(x4;D16IV^wjF7*-g>Mmp
z09Rl`eOG^J$cy#9C2~)j2MwkIU6K~h!knDr{-^{AL@|Y@hFp!_$w4Bg1sCj$roc^r
zXDz3q<0<ZFKVDrP5@Jnx*8m3qzI*J&g`;N#vIckfDsuNxV<~-^BHJ4B#bwnwD_U@y
ztgYo97&ukP%Yp@rcWjeE`o>c$*D1`@ibYkqfcO?!y}HYZ6G?E3voAk7gyJZH#>ptR
zB5u9M2q4i7<$yZ#P(q%F&6sj&JYm`4#8*B${Fnz_r$Go@9dbo1lPHvT-UO|gt?T*u
zx3J*vYZes$S5;;G56%k{6Vv~EbC|2~8yC+3*L$k|R#6v)nYiR^w?6Zv(nc1)@w9?R
z=3QXgHr;}!Cuvvy^92NAA#LYQo*XRggABsw1As%%7VU9ATJ68p>*Y**=%g6Mq`ZFc
zVprHKnq@*Hw=jI|b+1lRDlxpwG<|UPu^0}Y&!<PLiY1=EySt-vEcx^@%ilLQwi$n0
z;rMlanEra}_p_><)8`BSr}}5-`*B*fsC<>AGGY;?q(Iv#GFoq#M&pexjf*Z+3SBxi
zx@gMcS-o^&{UH>kF-WZ^AzCO>h%{OSO^EFHpe+y<zpJ%Yx4SdwFzU=2r+TF&m6;Xx
zNu4SFs^fMu5Dd<4OUc@LdhpPRrLN8TH$Ygw*Ri(Jv(wVE?jlRqwy)TRU|c7U0~M5F
zb+WmQd$Z0~TT^MB;YD4vrCuO}T2sU=qtEWwsbLDJPue1rrz}J>(Uh^9m7-N5HD9}`
z+w4@U4y@PZ=I*hcTgtte7&bbJyVa*6iRtQYND6VL_BrobL5TB#Ro;k-eV^53yUJc6
zYvXOZ_Y!M~1q@9Z+x?QHmm~y*icV5!=o%2iT3sf%_*KO<)0&71P!!T?=vH925*9Oc
zx5k2&r>Xd3L3G**NKDO*S7TfGXy9e8){3jA?7ybI>44yiTW``rdwwtPkX4c?@%|nn
zkxB;9<joDY#A&lGvL;(>zm1y5r&QmH&D;f!abT%6VK8!i(t-(h197zk+{30TeeG!7
z@mxE&sJPV$rR=axA{jE_l;pk&3wP7&<A_!J%ERpTHod2cN&}=rdkws2TcGG1ZVbTc
zA;j7wxKOdaTD-aJGsNJw%xjN<Gn2VOM-L<=tyq>ta`hto<w05+gHaJ)3q6)4g`Y1C
z?`Oap*SI|I?iz9=SqyO5dKos_b}<06Ob_@BL}5b7^p(=1<U4NdS{1Z<0QIEB4^lGh
zP398J3^8p!yD&fVbSOZ5P6NBwfbNQWDg?}pTq&z@AOp{Z$dz6YE>xkEgqM;*>*WlL
zT>i!|F9}8w^B77eg=p=zgRLB^Yp+)4c6b<i$FxqTn65PQ`U5tbgpk`dP;x|)x0HX(
zgA8y_<a-K;f%=|2I-SWJN`9O<mYKslYZ#-51lM8!%&Zl^J!FioGqmLDPVf`J>K-7p
z+c9*=&~urqkC;C1yNSDZ6<dngeDTsVh9$2(_jhiYHa^JAJjpCl!YEFUc8gFU5R+n#
zaWaISqr6&nHfI6U&Qpkk5&}ZZ1$9=TMzV<8UegSaqmGHS-r)I=ke9cW4bWQL0L)XF
z^dL#~at$s|k;gIq#d7W8152dfocwN43#HNI&2w5PqInlq?BD7VaFJc*GLG*Ga)du0
z4E*OeAW20fdzTrSrQ_M0;$U{MX;Mh<VX$zB4^CVg)grTGDvIO)3hxFP!!9QcIYIIF
zjM_~c_Y1e~Zw&3<XkUdN&T>z%r-ZrjoOe%Cuy{{U0D;DoH~>Y6fQkG{1cbVs&jv$F
z16#2{Wj%_uK?b#PN&7b|TpzTwyEOhOKa8CH?kc<XdQUKBEDm}K)_m#O2^o}_ALS~9
zov#(@qKzQI)s>ck5RW&?pGq=Fa2fmVK@{9M2>Q}Frt!W#L7$;1$;k(q5hBEd?))4S
zc1nExM7*j*X5?OkHQ%8rBJ!X>KjWMD_k_dk6_}Zqp2lLv)gb7buDmY^+)l6+nNBeJ
zR~_d8Vb!{jg{wjv{v!>MeGH@#ffizLM@KyUREY%>JoTqZ$51zib;nEjXUB(_)%>s&
z91I0i!*X~6$EO=7IN37%GPv-7JAjN5IR3Hz`G**QsAG?a^m|Dr%#<vN#7Z6%%sYgJ
zniZ%;yrqV;N^+}zLR4wu-ecIEaDOY)dd=Fa0k?2I97^0?)3p9z-&8JVtk~EHTLRxa
zs3Ndjy(S(Q7|+t?2ZE(i5n>ww0WfWCBIwC}adZ%pSh-bEO-N>DiDg*|z17-4*X-QS
ztZbMgwf|hhGGLJ%x%mQ0G0C4QNPU|{6*25WFR%1zkI{Y(D2tR8XYa7Q%ss#IkEt<H
zLP?;SixaLTad_=6D2DNY1Iq!<<L0tcN9;PjRIptwHCPqAkdW=$=^q$H4@`_^>|@%D
zHz9aZEJX-4xf~SpC7377zz_DWzxQ9MBm66DJOG8C!RrY%nEHh9vz!PgSirEu@fh_p
zXk0uWCdcDRZy5`LGu`1)hptBrSxV>?n1L=6I@$LE7@RZH9<)CKRltkgx|rVB|BU#0
zPAR8j%mq1t@X9^~1-#sS0XE*05n?=OfkI(o2${Db%;yA)OJ(;oUlCclr>rwnf$Z-F
zFpDfzszL}BpwrWe$Pyx^&;S0H$vL4fjXBEro!%d*gX8OVh<j9as`<d$5DuI;#fZ%j
z{6)e(S1AGLfCLpv^3U=Sf1-?K4a=EEo<-x$&f;#UF)Ks*oa6(@v@rf(g0*1mx;L_=
z476F(8}DG-=Bjnp>sm;ocQ^JaIh(=u?#u-%8W$x&nuUT9naM;6^$(fp6kD*8`{z7P
zqiPWTC&Yf?9>5TPmFojgO!|yG=!--ES;a~qv3p)lw`xV60mJAI@>LmpJJ8T3U{iHi
zmx+=^2O8xx-zWlO$r$A4_7Oy^=RXT~U{+p217lN9gAbc2G1n}w!2T_RHgNnQC<)OF
zJPXKj|NVf+zHN;%B!sEL6Y%qk=+4lO@hn!p>TD-iNx08%^WZ!VEx8ihhfwzi^vSiu
z$OPtS4rw!SBLjd%y==>sHH2&u&_b+gC}(`A<itzpO}J#0W{9USYN5s^=m)rTl<0hg
zcl_oQed_Y~&9WBLlb^;M7uC)D#-&Nky~QdLMWRKohj3^#P|yU4Si!)ksnc_5D+=~m
zx%;;HD>ta>;J7H=;Yi*%+{<{RCVz??{Pzt3S%%(&uWVimL)q(P;_2A^JIVfzC-#rq
z_WEylM0e4;{}MoKEdSr6reFQ@KMzl*|BnFry;A&CKO3uTvIqdsubUWfypmPwUax(A
zYncfYHY6cIsYD%bB=6(HpM#BBvM|e$gyXCLIaog%eRUPCX`^oZ*k8u4*Z=cgv0N21
zNi(jt+D@HXtCTw|R9<~KebFv)GQPZ9|MPKDjB;$FRiv78%E@-$yz}$EO}~oo_eYDo
zGNL4=e#Mfk=#~HH;eFISf9pd%NS=q=E_aNG&gtK&qoP^M<_L~VH>0hs#h@RD9_qXX
z{NcS68se*4`xF0Fhp{p&s+^jV2!1#5bEf_B_gqFhG^U!;UV~HX$hsuu`5e7u5}kQu
z2MtP=5vjT;QaO|Mc~eYui-zuP<3SdjTrqxC<E`!Tg8DY@NSV_$&4&8x66XOQi4Xsp
zFYDa|p^f8R@V2b0F*f%dQI{dsS8?;T0CtxBTqq53GrOFI1lzbeiC1Yfk6bI+`&6!F
zc;1tN_QwyA$by=!ZS_e>%W>;!Lw*A2cd~}$SBl$AN^-el5n7jIk^CTpG~8t%pCo7f
z`QhaZ8xO4daIJ$rW@fQ%`x3XxN;IW8-To?-1sc~teR<~Bm!EyuvuLh~q|QXn5!Xi6
zpRjhc`Pg+;cQ%fxhH5aY#_Qkxukx(IZG|RlPZt*smbQBEHNk}R(;3-$@(>nag)*Xf
z5bOJlx7eR%?`FoGQI`=b7xz_9mGB+kc1BoMY9$C=^j#m_DLoK$F}smK^~2uurE&xk
zQOUns4i&6rx8&hUC)s~2JxZ+Dm)2DE98O<K*5Zhr&;ZE$w$^E_-GZarzCc~6aI1Cb
zg4IUN_kseEI|N#Z-YmfUQQxJcVbXE>|1ZYgsY$f839~KRwr$(Cja9a7+qP}nwr$&7
zWt?>)Iy(AdUwnUHUe3%nGoLa1uwU(4I}LOG3WxK-yF+JZ+L5vf5J;wZ^?=@oXMgWf
zerdFP+qi8vS-J1*r5{L%>BU%VpuMbKoTnBPm`Fb(Ox!2Z6QS-b*xG2b)=ls=NX*ia
zK5|fVT#l`F8X=NVDI3F;NO3TVSnSFiyn&f5-xH?J%3^xqyVr8Malu<lMeRGsOVHtj
zlqi@$0Xw9>e}3_8a_X-gBc8e*uQ$MD8r%#>H#jFD?BN75OM3`Gq$VnfJ$P&rMk@6<
zCqHY7FPODWJZT(XzvNM*W58RQPY>3~3;GRc4!R<JQsnm=@}JQle3e!9kh^m`gofKz
z@!pz#2<UTr8HFJuPV!tr;ESRJZa<^LCSGvNDRDh=JJ<-8$~TxkLgex;TQ?3Phg=Jt
z!PxexVb6&`h3|rs>ZzEoi<(lU=AVqP&6`1eye$Wye*Zz{>x4%aiuspb%^BbHQS{Dd
z1Pyhw+hdLI60*Z_CCWm?BT@Dzxz9yEnE&fq@5c8rg$&hds-4pr(E~lUJJ8)1myh&#
zw6+YY_PpBv#jO0EO9(StkqycdWScVHK2MWR?ny&q@1HS<1cBTL9X(w#!x^67jOdN3
z=%S_OZ+)7zyyJT(j6rPQ@-ncR<V&XzjbxHtX$*(!d<p&btobWF$s@KX8Rqk*CcIwb
zf%HbcCXF~LMd#*!bwLDOTV}d^%|{l~0l&4YD(=23ke7_~aXNtWdSpL%IDH~KQ2HVB
zBShTk5^iaP+jp59w;2m3|G6eYl0;oXC@CShvl6O>XUS*<Lf*o#%)x0ts&+m$N*<pM
zQ+5EWe~G=7IYY8Pory)5cpMHF)+N_p=eoWS+q&cOX=|mm6yr0_!ISM_d!>6y$27a`
zg2wcEK^DO5lPHW?Sft6Y?0LSL#BeF2-9Jb&X9mgGVWgLZHhR9^6qG(<B1I>$94G>d
z959Nup`8CZ2pkj9_F(nFvInOdg)8I_lmeJ1{oo8IZ-nAiZ^SvF%15?Vz@;W2>-5|_
z4FTjfcs#xggEf_0S6mGJ3-V&#Z?={cK9Uhm6~jC;EJ||HdJPmPJAb^MeNfrv<u1pX
z(Rr*DT-DUq{0%^FY*A3Ofm2Fo%%iPyNn#d%yWJZhB*Qf3G8Ee=yU(?Tjp{=*2T43@
zBt+|I%pR7*rsGU8E08D<LQrZsN9)fzEh#7Y7y6;}@ZOQU<G$11Awsw=n30Zzvo*)K
z92h-s+e__84mCfW<z2*adOMpp=Sd-T81$>fEWd&*(_>m#Vw$xwnWWlVn(g~~9>!y%
zc)S8dz_sbB^Jm#--|!WH0mh~Oete;dHLq2`XR23ysS5z;S`_pr5l07-T9JB^Bxj|b
zP97T`JZ@P@fo{QvRa_34L8`E8m1lPQxRbdJZXZXP7|-hJYwL;`P;ORcK&ao55yzWQ
z<evhKNw=VRu0${X{eg8|H`%#kuDecVdq)*e;pGWxc+@%*`8N>+i-7ZekXZauhd%b}
z;V^-qjD(+CiEpNg%0XVh(ZkNnux<yJ`bnUo&W}(t4_yeyxR)tOhW7_k)P%FFxAvq!
z%tZ7!VtuW;b}KLBSJQKKdubTV)hP9FLN0_?sv?w^zzM|lY~8{aEwLo^AY8uPA5!V0
zfex6H7XDS0E$JV1_2v5bO*9U+uSS`$kY9H{QN`O1v9Y83!w>`+*P%V@$RBr)4^oO0
zY9rdNpj6vfo|iB+S8&6P5}tdfl)oZY3g8u03ZRjIS8e-Yz3*IWXQ}=<{mTmJbQ@}q
z{&#NH45(A}CK@>vRqs1{5kKl#BwtEC(acbnY{RV|IywX*TlRBE)H%ouZDXq6q<|vk
z;25cy#}XES$(-?<{+jWrOh2ZRiw3{2>Nh}fK2$81d6GE2ZZ@XyRBMuM6V9T6ZYbc_
z&lT8i>WjZ=>sn9?D&iO{u~eU!Zh05j_ncR9>ix6A*{)+p-EvpyZ@}@~L9bU#t2xVZ
zn@my3+;BPSqBCb0yx(-fpR0?U8=6mj1E;bK>y!$PB%M`VU(!FNR8Q|;m+4(HQ>!}p
zpW6#*r!ZIMuU)p*YzK^qr#9kUj0JbOKOIc>tT0MT4BnoixxAT1sW$Wr*b-8ZoT!!X
zyHja02tWKfeIM+13X`e-)eEuzm$j1ZfA58Q{*UtJRsFtZSq9{2QhFDMebVHR47tiD
z;!{8}J~*97_0RSBT^9jBX0e5)92f4K27m+&8o1dUU3gDSSdW{T-Kp6bbD$Bun1cYY
zUGIy=J|rH&bO^hi$DX&4f<KuuqW<sW)5^*jKbbO>^XF7v9-WLA8)S!{!{39Yu(q|N
zzBBwevuqby$AEeWeL*7z<Bt+%Yajc*v!--sQ&|fXVnloMV2Vg!=J(LWe0Hz;R9nWj
zMHv&xRuxIhZ5>ioL#_AD#0(2GWTf}UJgp2d3o*BMsZy$~^xan-lN5{6_Z}J<<)8BD
z?3`v`H>Rq|0Jm>m%-7OjBC&M~ZXyU}j4`|i2(obK14tSLkTaIOfv>vC?i?9R)E~Nu
zL|FGC9}|}Qq;EdesbVD$wN7@pQ$@0qNf^=-gSI5qg49#Vl=5XB&554r(oI&4&O3lO
zOr&Hia5X0F$t1TKKhHHDqVU6U8qXZ|a&hl^Y2L+aC3Y|~D=>9SUcv(SUvCacb5{3-
z@Qs=)H#Enr?TT&&-V_&8u^%lgCkC5vSisaL0-#REqf<W0%0-9SFEwrHv=l197ttEt
zJHZ24P*K9waM5s}WLQH&u24q(DNwgzzdE+93n@TUDTdp(y~GcPa3riEhGN?vD}8{0
zAJGdQ8`_FX{ZW-x76y_M8a?<g@_l!dPT2XEanau%J7v7#pKZXzm97wIx-d0ip`rRx
zcz-ehbxn;mwix7hAO7q8J%&*{T3QBI8X)c|Z@n@#+|>!Wl1a${A-b(LCC$wR?C6ZI
z0<;Poy>$F7=GWg$2tb4vG&df~Z1kzg7B6}XPxjmnC!2Rz-f-}siCzUsR-|--^9F?G
z2pKS;K;U?dPz)p80elWwhJKffJJ*htnJGAZX7F_*Lj#ntZ7~Q6w4nX|L8viS2vdV`
zNpA3dPjG?nV*|JOjvxYAmTOrz8w5k%fl}fG%TO=lLcZZWlE*6(Rqls7uzwT`2$aV4
z_Nqof0Q#rFX=&`g4nbW$xdVwo=>#yMiqA44FhL)JAiw}fVR%M_hdeBBJ<^wrDw3vw
zD?l7bTey@ah)4t_yvzvK*$%(3Adi#jlh{ea*DCPWm(5CU5ritti6<j+_&K(2A=fDr
zL>}TsHHXdZoiz#p;iuUz1xdkRiu@>O!IEDx_V5b9Cxr)_4E&BX)s`9+nQS-*@<Z#a
z!UVPqcNrEW%=Q$P(xHgk527#%%W+^<@czqA&NO4bzGZ6i1L2HF>#IOMr1n@Pos~A6
z)=E3!{c2mr<4@qZF3FGM9rt-A>@h?lhS2At4ckJyniHl64A!Fe97Dhg?4zHh&l?~G
zeg-%G35zUQG)s&-&T2^hMVe00^EItva#u5}FxJLkjV8^QomX7ia$|C#p#-%BHR_p<
z$EsOeOImqnS4H<MOy7-_6&4*Nz8Pd;RChU<GgoV+_A^Z}(rr17uE|()s&3CFtTuCV
zZd^BI<Q<xKufGvl{J>|`5{+Ao^zzAcVgcy`Q$iwEavb`ounGb-)<T$w?M#9Akny#+
zLWae5-n_yye?*W8^~;sFS$ve0{9#C^YA_o-(q13wUfYOW$gK!5<B$dshuB8w@GEPB
zf|?}DO}x;pR4be)uT&S=hAX#Ig-^?Zc}T!h-U_Kl+oAWCK#@;k)=KpYZ3a+c(Pf%C
z0sLr<pfk?%`TLX~cO&1E+VItBKon?(#ib<MTfDms0H-FI=6tDfMaZCelBdziL^JI=
z?Yo&&2t}ON(26TOvLl_P<<t0Tt4Od5H2W1XK;YBDjSeK9Q4<sNAjau$dimN*_TewF
z>(<?EKCOUNo>F;TR|)~s^vfn*V?pR<b}CtH_V`bnM(Qr5&A7{M#(cd4oK;F1OL`Kf
zQBhhDVAKlb=8k{IPUU!LKYx6b;DG!Hr!4JDLUrcJ*2vFx&PL8fpq0E}bFR8Jo*ymj
z!L)v+ai3%z#(S?YWY=64@PtdOn?l9g+YJb3kUrlWM6TSsoHxZ7PuLo7T0k-4(=#7H
z^uqvm%pWD*OV^s`b^i8->fRYJzRDiFJ~+P)PR{F&l@y(#r|g(M-fMG~Dj<eN2~#WF
znRM!{S%{}k|HYhSdXhbFkQ+fc2#0mKlw9U&s~RiCx7qma&B^<QR{PMs{jZwE@&6u2
z{h#9}CYFEx(EkydP=cu&x83^i?HjUNDcr--Cn5=}l`ojo%&L$-i|Szy6V$eN1GOr3
zmEwGMRd)vywyf0E$(+a+fj@LGZQnbRq&t|DXZtqt`q`D$#-B=r$eBd6G_fOGC{q}f
zXkwDPtvV5=BH{Y^aGUp(a6BYtPrKB|TIR{k=EL6Iu`?cBcK5k3>a?ZG^7i@Ezy~|#
z<5LMP^7lo&CNVKZ469p#few&aKDdjUc^bw374SBc+g3DTk+-#Hhhks}unu$7lzYDW
z0`K#sArz8+X)-o7_28rJJ7QGat1-+IgbKoPduu^>0Mz3W-Y($n!8oh7*4*v-!b^{M
zCoS1>Mx{1<k*d^LzYV)6>e)6fyom|dvU~Qv*)JayJQv(v%j;Y|ETF%dKa|PIo<K<z
zMW9W*y~=C<>>0-(<Zh`(;&sMVzG;;B>Vr{%J~G^!6mw3B$dBXsyWI{jZ&}l`eypg+
z&MLD2VS5b{R@@!mm)6($W2o6CtFjfh22Q-|ulU_nO%52p$1ORQHSlqP-?P`MWL;kN
zB*(M_qZz6BL5AT!0y4Hd<3wqHo3{BD8i{a`SJ!tmu^>(V<kBOIVlrn6SB%TZeG4cL
zb|l=|(5uDoHTvj@m=nj#(0qJ#VHDhpQ$DplQdI64%&>g4e|>`4Z3)$Rf`h@@%d79Y
zTeF$P2lZJb9<Xy)h2-|2k$?+$P?5njqm0d#XMFF)D4`C9{RW2X)=Ln6B2;*DYq?Hr
zJ|dzXb_f+SnZOK=l}HODP6o3$S?^?a9n%jLv!2P9D$8{smFIaA0Spupe*BrSTl+IZ
zfT%O-x-a`?>T2w4p{IKzFfGf_ByxpNW1!VxQ7dq7U!n-EH-45uqs^0Ew>YG|pzr6z
zojR-zCx}_Oj4@=r75GQu&p<ymkiHfMP~f6j)DIAjiZ-|R`}?^va6a|(S+<4Uc9J@S
z9ek5o@<B<D$)C*u{?50W3;nzEDjaRClI+wfPw}_Q0cs>nCiwxt5rPnKMN%H!(JOl8
zCv$CgTzg)j^5>b%G9}Fcky~M)*UO`hSiu-WB{XS!NkDN&WbthKe0{CUlYk@7;7s}n
z=uwwe9;ZfCqb@)Wq(5U=i7e*V$1P4Gd)F(D_2B%Drf;Xyi;Fr`5KgUI9~(Zrd@C(&
z01!ZUqZ6!djxInE$ED0_82I_7zZ4fu-f`n43peK|Hul9Yj?_dj-`6r2GQpGdIF0jP
zmr`t>ujzj>nrP5P?&gZ3kTRsgF5Iu>XEjY6G0k&ii6xSe_P>5I{Ns#e<78Kyxi=e|
znO$buJwRyEiT1fj;W)4ebv=&4h8fnoY|04q6+)Saw2Wm;{~(S7V%OJ^xz|B0#R3<j
zaLf?*S((0?Uo_<{6>l(~&PufoKeqmjoCj(9SO({f=z-1FGISuPa|x1Gv%K3I?eR;y
z?*zoPd}jJAvR15Tnj*;rlkZQ@P_mRm8<FZ9_xQ-88<tD_&?0L;fLW|JChl)-9&<-`
zhU*@--CkOs*Z^IFASOg1ieg9wWflV_gj-X;g9(-b(NTp6V9=kZo3ba^)uTZDj6xC$
zfJC^Pd*yIp4HiCTQk>pLDcG&4SNAV8o*)Z^Vr#R)3;OoItVM*>#7`oyl87zIAIe+9
zm`y(zx%x7_*8x0q@{gB}kl>R(o~x1&0Cf*wqG~=H;hfFFzqRe%2H^$HV9-T@B#Ed7
zz_f#B$LD)24wPl_g-zz;oEHYF$eCTJ-aJ0b@>AArJcda)36UuVGUn<#WH1v|^u#O&
zDP+%gV(`;ULOvG-JmeAf@exEyuyvOV=P8Xm2JoDFitg{#ffj-q;vn(X&x*&@g%m<P
zp$QM<4p#>W7)OCw11V4~PR?h{Z-oM>B03SHMxGe0SMm%Tp}GpSh<1`GDj;6RUps~X
zZ)QP<auXGOVM4ua9e@0jFH~VDI8wspSYML?GY&<F5G6E6ELzlp9c(H#2ll{}w9%H{
zjrO-BZ9iJLHd+Q;SJMd@b85f@knR>#egMfDUj&R3mrDmc(re8&1;<F|RE3p+URUv|
z&>es9WpZF&vkcSj+Kmy5wc6B9O~+AvZUG2iU6?I0Rwwnx$FF0CsXXsRdd@~Zrj7--
zU<-}PCHk5VG^6#C6VyR5B;_Adaq7+@t-JZ}PuZS_rNMxd;5vC6*abTlLoIa@pmzXQ
zF#$No$^`KG^&JUQvnfV7TRAg_a4IENvxdj7@;wykTiK$1=-DP)d>qr-u%N>curC+*
zC<R^_{;p5=4o3KdUf-u*FqaufowJijDZvZ_;Y|N+$YQeW0>bkQw$!vtyeDKCp)XCn
zw@*(92|>dI;MC!?a*zxCuG`(6ZmXyMK#Qje8?c6ar92{U%`Lb7&MVYPW3o&J^y!{Y
z(bEVh#w+QHIjuXfCUNZeac_|1AFhBuY5PLZrb2E3U;ROL#P7+tRwuTF$C#0KJ-o53
z$>ijjNe>w%;r0f%g$pPzodfJO+b`MHYYR|p9hn8a)BGVc8BjhzcllU*+}cn5h+&u*
zfbRCu-HT^CecMvc9=OKuexACs9v6*&M=tt^hDs!_>iUsj8O_QTYF#h5qODa6lBwVq
zKI^9B4d!3Cu~gD|;^w3oG_p+daPWteE|<6EXD@X5$w=xu6wX{D7m3V+Vs+T*x_t3l
zUCRdoX!CUh@7Ik<UwLaVV73Bn2NuWp@&Ae<<X;Sih8S91dB%g$m_MEYA%cI(K-<_*
zV*#$^rlh9<J*F<clEBR{xS%d`E1ViQoJtigqAQqC1Q-Y;SgXgekW&ur)0ptE8F=_)
z^B{y)sKYJ`yfhV;@%nx=*K#t0H-g-OCvftKm1tzZEjtGzCcFK;7_ha{<LMZE+^H-$
z#c~{*el=IXxM`wRP6-@eOX{??%1pXD7zlfeI#cT1xgroisqLvLPpE#o=664yo8>U_
zORZ^^L?JmFq`CM`<pzFQjn5hte_2&tce{D2dOg|iZ#2B=jcTrQ{(H!Ib+ngXi6RsA
z?@4=N5UuY)7i1UtNfI&_-~Z0{)+)V%zn{W7w{(~DgL<1_strCqns@E_!B;S_KYET{
z4tkB_$>05f(g?6PX)SmlU5NO|E&(<q-+TO}#nc__SpGJ1sCnfd<Y~P7DV&X&WqVmx
z@Ye?05kF1!O-I#{qO@IO-@SlhEuLBQ<fTt77-505*dEd~`c0kk3H%aK5(44Hz8H;0
z2{ThT>IY`l<~1xK^wk&gw_E`uMg1$Z4~*%#RO=dz*0J^d-D7|W5)$!^9x<144O(A9
z`xl%gdRy^7d%^#hUodg9{LjZMJzD>W6KqKT|1PK1xh4d`&atggb>mtlw<{&*!GwAO
zFx5N?Gh#VBnfK@QLJXLMLfH{d)r(OO!u%O}M~^*+9uMvPu&T;$_m9s@>e)%5aS5WS
zgI8a?X+erX$70mh-@~zJXtMRcN?AOgk6CVdzQ3p6Gk=xNdAxpy*6em+?exCR^jO|L
z{P?>)(0#jp-e0!ncBwKF)iX~v`!7UxEOneyz(dipPk_YU1HI(j2imF7+qgBhTHFI?
zK4e3HlWeYZcidRsm+|=K%X$AiRg1U!pOo^>Ur(|&Pb_$~?Wu%W{V_)E*$X+Fy?I^9
zOno^J5U=~Jv7%B4Nj7;Iaj$$b-ZW||?dTMaP>7Ep=`aQ5=^>F-t+|NjsmiLY+s3+Z
zOz64OeM*b<<z#b_FB-F4Nh0uA#n)NdX$#ZUSXRZY$Z*Oz&j>3DS*>n%_55*T4wZaI
z6=P1Bx+pbjvJ6-Ys>&MwEaG0iwbgdnh2+pGjTm%74q2N%^)8T_XhsmRlIQvaJr*)Z
znOMD17ACH9@%CgMA~**W%R2LM!JrM}nv#7vZ2*cI+sS%)tLH3;zirY?uO6DHa&2>?
zw#P8$F*9Mrl10sq=qAVoZz~$=I$G5h4?@Oi8!);ZOgy*#OD6$!wlw{Zg1YK9)@zGx
z_e9jyL#MjBIH$UVE8fO(RIq?4a|okssrV4FM5$a~s4~nyNnJCJ6uS>gQ|6eWYTHtf
zCqEsnyfFQYcEPh8r*FgW{QM>f@m{0d;?`oNUu~~uYWqFaaOJiu#w)9eA7{%wR>vkI
z?$TDCvt5Oh=h9UBg8F4~@42J{-ez*st(dUVJ^a}Vy!f+*tXkDLzPnv?{hBYR@B8Tu
zTL8~W5E6lP|LLj^woHM$4V$Y-R!ub$+%SCgvETh_$=ax5Y?VovYCEp}1^ckX;8~i0
z4GXt)BfMfauHD?GD)}0Gn;nZjLXS(auz&11O{w9iW|?8WUVe^nbyQghDGu+*i36Gj
zJQaWk8@#e5tC6`2>is&)m?+W1A74BlLXUj*m-)s>C!ERS1$ozFA{wS+Uz9{5Y23_C
zx_6(V0u5vHaYRXitVD)+ljQ2F{n!HzvX-yfkY$Qe6vnO=QqYcvOKevXNB2}Fu!yF^
z0LU<c{~^khO$2*1CN`r!qo5>%k%>+t_7E&J7tUo*y{WlIUkXlpFHV|fDHZ$>?KaaG
zOmEI1UjxuXKcgDTY>iN|rYeuX4hC}|<YGT5f0*jH{v~%Dnhp)eWw=9Ks5V`d$MHkC
z_KDMDzfseruo77+Rg=;Dsm=+L&P*o~&|?~DA^MEWAaw2$gix}15`mWCdb;H1)>JN^
z6g@s28m_h?1K~?mc%VW@FLh6#T#xLE4}RIY5(EMw<N>~j>{t#5LIF!bdd(+tZ1Q<M
z@T4SW6QNmT&;?gAWnVZ*Zf{0X<1LDTzQ9CO5v*sb(c5A_0cM-SK>nJ$K?jn2JR<Uk
zjXKE$VVAf7-sv(RD+9iJnp&*s0mwJKMFVi(aR2qmBaRl9Kf&Bk?s08d>INn!kdm>5
zG)oc=DX^GM$lvEEsq2u9s1h2Z2?Sc+tq;lm`rSXNez?OBi-xZt+5x;{v%^KS=qNWA
zLI2BXZ&emwZRJVb1YnHG>_z2~A2ARL?EKUa0;?n_81@f+soB%i^<|=DKTg^8qq-au
z0#-||E%qSVDLW`7ZcwVarh@Jk0N*3=nhsMh4OBZcIhXfHW21<PTpP_950Z|WhR<Nb
zn43}%G(qON;YBf^Pp_y#^wSD?+m>BbR+Nb%oZEKf1?1g7Jz}YT3RjK0=mwD)Xz&xH
zL0qJN6N#o&g?Hnyg*bd3l>;6xCV3(G5r^j^OLOc!#3L_#X<2U;B6wkoY)L{L#-xxd
zpDHqtX?1#>56EZ<rVUs=IOT#8&<Y+n&e6EE=unPV>3}Zkf=C;X1jYhL@nW3#@I^BP
z4=!(OZQbWlCs%;kR;$PTQ*Y<thPrB36@VEXI7*(rlp=<yN;!v`Z4@bBTCf9>J}m1-
zkG3kHe_6OVp*s3k^c9d{dM=*x#@w_AzjVhL+Kq|~rQQqDABa?8!z}BLbJEl9G#PhF
zSYr!LQg3yeZKIWh)K*%Hv_F^df@!g2-e#l}J*KE#9p_hzK#gd9dXZ}^RH!Hzm&26Z
z=o!~nFBOA_!<pBn=k_ECP)fkXUMa3n$$$UCgv-+JjNW{EGwL5O^979~)DH!p;=g1h
zE7DO4%Tg%g4-4gNE)k@Izd5rU%m@#47MS%p7`Wi(;od;rpkxl~NP)q2jpk}MlHxQu
z<y{g&+h<Uq7i!}XP2RQ_t!VIenxbZMx~cxEs0MIGW~E~Yhn;76eH?V~E;?39C4<AR
zBjl$^lFM&2_&wL#+{vNrpQBrQyqQi6na=-Fem@<9rU!J=lt9=wVPhv=#dtO%S_rEQ
z5zwep2+TsB&9gk5b>BV`b*q9uXjADSyKPuc;&@);sS4EDpOEz%{(O5~r*QgirhIAk
z!1FSvM{e@xrkI)+55IO@4tyBl;pus?k8jm__{QnSG9t@oYWGVL{)zsDzqNBLPmpVW
zeVceXQ6gj1`Iiz87W2)=SCs!b+wA+a6*PB&l$C_=d%1iH5`g3G6{Mx*joXR(cUu*e
zw^gv!16=_N=TR;6n%2G^)(V}zVDL(UwqrA1_)qk}W2?K4mo>gi@|P;L8aMI6li@%|
zm%wqz+F+?uK<Zul5~vz>#rgNH5~jY8OZit2Zgwu&f20J7!3CT!Kasxl3sw_a6sIxO
zZ=l|tkRFmk-P%Lw*1DtW=kBm@4ISf4`G+TU$wFTYKnwgmD>YZ1?mQnJd8gLqQ+9cJ
z_Qq^G6nI=+O|)h7&HBrNJ9?VoIUn~7Y`zyIb?7l=2>Q@Z%0Cy<O$LJ*h&K}+-(P>3
zt+{{Us#f<C8sR`MQ3C<t0U`2EN1T`T0(g08;O5ok2|=s*JL)Ezcs2zYi}1%k)NH9U
z_{Or)zz-}DCQPC{e)1wWNPDdovVF3QSii&J`$1q$^8-!=J!!0XMJ7QoRHshWnnC$)
zO<D?iyq^X1)Wj)040{3!*?#dAkYzo6=8hq5G#dE5or_0l?$_F53f0$ztH7%lOWm-1
zbjPY4+|tG(it;3D&AtneS;lpHrm;>xz`ewT!(ddQ>^3+9veO+xbczZf;j;e9mh3Xe
zyAP*6S@uAkF=*BF;}S<daaA~hH=jx`bwvA$x>~{Sue^=Ygv*kzJ;Udn&G^)N=of9W
z5l@sJ&yWia^3F8*L=QKIg09Q?tWJ%HJ%b_Mf#@CohJ$E)bwKczDhRASldm%<Ld%u;
zB<`(cIv>x{RX%j>pNc^d^RLpeJZWSN&$#>3u8!JrQ%7sWpMqNRLev7n*?ceZZV!{Z
zgnJoj3iLNjyTUzYi@kJAz`0GBzVu{Qp||w}^?QxlB8%9(OPiQd&KW8DLI54ZjF<(M
z_tpx6j_V<G>JJPE=3ew7m6to>{Q=;af~b3?w1Po~j{aZpe!cQGhl;ymkk$}oxLmWY
zG16m8--a5v*NOGVmfnwQ@SG3`P_smI<QE<^fQ8Us?t1~Xg){NGrag+zVR%DJMg4gA
zk!`!*LIzSu;t9RacGfE`<?s-<vpFIf0A&EjMS5d1E+LHH$dN>=y`dam4urBtbn-mV
z(q;m8s&!!Ts`<HB@jpEsJs<9@hl=H{P%(KxE%Q+;x0hpiPa{~q!UV(z;pBaeZ3w9q
zNQ7gq4?mZE2YG>ububc@<5M9aK3v&v4yPj5K7pnN8DKg-o`#BW<c?FpcU{Hga-t~b
zY2E?f%$g)Zz0Cz05bZRiHSc5Dy)e`)z82VNymMMgr+l&%i)4C|p~Wb91W$Eft-!PY
z)zJSN<1!G~8CpW|@ciFoD-#>r|Ezp((bTj%9!2zBtJBXF0?c&H<2khUkE!8>8QztE
ziq`Lepnww>4ilbD#KDF8+SA=xk!|c)zLtvWhMCB{ti0svtt7IrN1osRJF!<M*Xuzl
z8lT8=sA%5;DS|?QiAbet-vYNAMiVQ?uk(FBLoK!hnIR>C;<RrN-s|=BxU=#;auG?5
z#Uv-EPE~mDYDYiY*X8-S6qjc9Ha8-#Ue*AkcbRwbF@@hZmc{<->*nk5emnC%@?Mha
zDO8ioOiC$<4CpuS4=R*EVG>{nX}|<q>zY5e(mD^L+HNPWiNv;6MpGS$Tf39;jdu58
zvvIIPw{<u50fxVA*f-<pLq62_`y`;jzwU?K&F1nzy9?`glht_p(CSGc*m|_RoQg!>
zCc4XB!&lCySh*e152}&?QhUlWGr(YE&@^BbHx?Aq+D&QIEnbts{aDxErJ?;{+FCc|
z`eLh)qIpuP66Tt8G9tr<V@9xU2Tfc95C=Uv<cCvPi2;`^!-LW!P#Q=v`ykV&AgWoZ
z)Xp>`6PyozFEwQ9Gp^_Wh}N<;<fECQ`RO0mrm1bZma-6d6(br|M)y%okKF~ydK#2A
zvOHyYU(0|7+|hQ@c^-Wmt-$voL>6tInZG;AA#{&mE}h&l3TpN0v+zQC&uFCuN~qyp
zJ|(+!JNW}BBsy$q97Mym8SS94>=A%XFb$^}LgJzPs;Q}+>Pr5B*?MfEp1+73Qxr^6
zAriATLZAHDxg2`IFaw8+l^%-0v;3YN`p&i$#Q9Lr@EZuGyB7#&yTxZk7zoDoR(Jah
z$lVx>%Az6Y=qRBCjK-f=$}_!*#q)rU%a$*1lj8ifz|8UmJ5=?K+?Zn|a4GDM!YF}=
z2jX9Rb-KDM_cb>J>**6#4VrFqd}BRP%vL=2Hi(x{K|n@jY6`3`UkoHj^j!|GiNBUE
znh3$EjJfEn-;*z09j^2fJ4LFC5~2EmqMI08&hp4<G<jcd>)HLlGh7|^woeEQK=kFI
z-P!DHF(w{;fHnDOs!n5qnMM*r>ubl1dqV{ofMN`YkfE_=#TXeq1eUiNw%V?u3|xIf
z?H<@jgcLJoHAU(raOmq*A6-Y=^jK;e|Gs%k<yB0<-B#8g;?ghyr^?^ol1>57N*@Ry
zuymm_Rr%Nk7{!B(Q8O<99wiL`|01uAXdfrdWHriD1#YJUziws%8{To9mNt7xlVL7P
zi5OY1steh=q(iDQUixhpD#nLlQ$icLn97S8Qd-S<(B3SVz)do3-|){;Z=z|Q-#c@?
zq~pKz2J^aOGv$31HG9c94+)`5MZ?1GWoC!<2Y<5C4I%6VUSWiy(7}dTxMzy53jA_V
zg*UiS5*P?z2tzXE5N)ZQKnP-s)Pd+5cHk4WC^_`EvA1Nk4K2_HWxn2~-qZzJ^3p6E
zfFW`D1!P=JVdzCA&3#7mpEW5;v}B8&3{c<3Mp|`O9lqVrI<Yi=OjdP0ryQAPlO-d*
zpsu<UL76EM&_G(+1P4S#bT#IhU)G^tLr8fGz&Ru$6GIw-qr1<_-az*@vU4dgRDnY0
z=V}aQLPu&K@_~zHBYsL+I9J_uNQ-;{u2TIhiM7jT)kUtn2?2snTM2I)m)m=hl8OU5
zDlQBOsI$C9x8OVkOPGY!C+cTU108qj!Z1LQtJwP#KTaG-88KpMeXoey)*S9nTH6T0
z2ej%wz$b#1{9=H#+`5jPn`?5{)JJRg*ylE-kpg<HFk^!kf>DRnpQx2YV1m!$p><pU
z2S|-P?Pyd7oIh9VQQ*h((%}>ctz^SSN1n2VMrNJVO*->`PSNxa(|43g*Svf9%-UFp
zAfLr{@$Pk#$quNIRf538NuWt3pkObQNut7$Q>L3&WZf<1YL@ImYvUJ#J3bgpOKdmh
zOaUfqdSyUhBBch?FCv?#mti9|Po$-k20@%ewhIfLw*AlPa$zL25prB&_9sJrB*WOt
zn-1qshS6&2#hzk)klV=;=PL*0ia*ZbM#{n>Awk)QSD>j`a&sYR@<itOkSeD@b91Lc
zmZ*V_1nMINU893NmT0-~!3G`nf7;FM;?tlw8`yQ@(-TD~#g|h47BWa0YTSo-=%_(S
zAlcBtou=qb#FPM72<P*R0Of8dGngXTLzD-@s;NUi_J;vg^ZkX#{OFVflECe6?Wte4
z)Z*Cc-(jsHq?OTTcC{DBeh1lRIr9DOz|%z{%ji41E>9*QMe5|W8%sLeJ;N4E6-%oB
zjULn~`zxcz4*zV+l4Zt;PvG`0gHzpqro&(MPvuAT#f$aEl<bSK<xqv%tKvXU_69&5
z=g)X=W;yD|XcVjzx+tCD;t)hyXWzv9oDmm0OK;Owx?sDMDW@Ixl-Ha-g_hCF2u$6`
zLZ*mQG)in1aH6%>PDwmh`K4gegUv)1^JDCZDA%3AiGs$`qqBY0QO(hW4E0CroGk$u
zfA36VuZ%&yOp%0#TU;UfT*m5NGmkdUmFcV=M1id_h*k?*?;Qo#+4fCHK+1c=xR!mL
zbJD(_xep||4m|%xeGxt1$O<us(Dew^{$+?NU>?KY-tWsC#*!=O@*skrd(&17F&bfd
zufeZ)Iv`EVh7dFMD$#s7C!=s?)x`AQz?s&`Y|qMS+a*CA^TWV~)=rYS+dHzZ-n?+e
zN(CjNLrhh$-0PP+cA-?f!mO@BXNUnNw52QX>;^-8G5v~*v#XVTz{pRx+xN8gsVXUs
ze8^!RV;@ie6|sR}FX!9@k&-|@5F;8Iw2C%xW`aMjst8xXWefBQr!dC9Y=+G4WY8CH
z-(S^;@6uZd0jD-4v%BQ}%!EG%)z~h(-Tfo4<>8DKLe29sgzL1R6fW>LI>nXAIqZ(s
z3d(f|{~Q#?>faX-p5AZa=c9(+Jb&+NA{a3aByrvw{eBI1&Pt2;i0N@!mg!kYmxR+4
z<{C(W-z#p}WAIjkYydl0n}+oAd?es*hX0vYW^@<ucw-9ifsRHj^(+BKFpoO8B*qCD
z&+XUm`b5G4en!d%_vRTjOs8+4yN!L+oMc^o?^6KXPKj<Mz1PXTn>*>uLr9oZrwl$>
z;~M~elbEUi*H=&4uwP?J`m<a%2HsqK?t5o(XlwS7LGAerxX{%4xPyl#3zT>lVc|T5
zkfkw(w}S5;*tC>}e1(4N%iD>(vkl{xrvwp0C^-ytV(WrfX*9TZ&}!B=GTO8&e7juO
zFgkvOoa8$x$8p0;Bb=|pp7^TL_^2QxIMVs^gJiG!ji(u^bSX&@)jMS>B-UzQ@y+W)
z`=0el`!C%>?ZN4c8fM$5+Jm`FE2m{}?4^v61eVY{+^S!vWLsE*t3X19xdpq{fRVYC
znvV>NfVt%Ll)m)3+R_>L{C5niN>9o$qs*F37rkN1mNr~}dP2s>eTlrSK3*xdQ`ogX
zcd)d7DXyb-c3a0kDjL>SKOyi!`tZ1>hFo0cjq$C{I8hoXq<C2Iu1?G7=aU)qHcPj@
z(zdj~m&B>Om0{5DtWMs(`I+9CMikvb@_&Hqy?=23vu*v4OaUVk<Nur~2vOUzJz_)f
znNvSk1Tzd}-@)6lRw|}w6g@1|xPA#pHO~qs!Rj2g@AH*}qtxDYejOnh6Mme|U=}_w
zM?W}!&DvXKcZ&l-R}6?uE?w}%ZeWpek7ST9bV2orffR+*qDev-dzGhu_51Dc)TFTp
zJW31e`-ArfKX)xEdW`zB$@k;y3(C*eH~3-6QV=4d&;9B09a*NAr~1zy^xxXwcCX{>
zN7r0}>|!t+om`XE3ox&4L^08XbjsLpVJSxucf=a(Bb}?F>+1&=G#k*Q2YZOt7`&jC
z<$t46p_U-GJA>^Cw++uFu|8>Jc4D&H9uT9Ih$FFNyLzl<{MX4l^f2Kz0*`9Zj<$B%
z?Jpo&IR#s>u+^J`;cGPEbjkGYD9>cxdSS>Y2r(M+;YZ;wi{W*s%(=L)P{x-odrv{j
zSd0nak}!CUK7UE`Sf<Pw_*Q3M`&*DuGYTW?e?iq(63oVinLCGT4>TBBeiH(-aEA{+
z7=%<^Mg<VCUd7xm9Gnr9k~PRIiFAD2dj%<l58zxqy^(3v<4C)mKa_^n<P(!nGCq`+
z9#&^Ge77lx+01E$ImzNd&q&DzNDqMp!Af~}2$rm=a$7XEqn-*ewIViaJbdTA@i~jQ
zG}I=w>RAp7U=ZoCUKD7z4w=HCwh3cJUk=b3&n<oxIMI!%a4ei_P;-!l9DU`eORS@4
z2t@Z{7*9^&kfn(7lvFTTwrl)RBoQn7;S)v%VKlWX49kmIl~S7abjD45m$$Ip1b`kA
z0;WL9+&h<InUavdHUR^!?K$tK4)j=kQzSRKu02p=9lH*==7EZlm;#}xfibJDEzg4N
zhagEY&wi;Tws0+%uX50%g<5x&Wk5_-Qml<FX|Lw1B6oPuTLKCRB+C_yT1v{&rT+9-
zKh;;hx=6|z4?UNY4QRMHdlSOsUK?<Riy*OoSNCY;d>}@GcqAA8;-dsf9)&0<E1zt$
zwM7GvQZEHP(Uf|3tQnc1tI=<=8CcnkFt`@h%Ga|zm^FYnaNF~)>pEVm)qUo?`fsoE
z(>R`0U}3(rFFm(!&B0ArJ3~KNj<zMi(IZVp;*><F+ZOGqb^!x*>tp3ntF~CS;p?L!
z?OE>pc|gO%G1`oXzhJmrCC8o()oC|_);_)F2fx|l#E~}}vRCzau_-}!IIgg$$2lxt
z?y3~9oz}t2<(nKZR6hDQ16EyvdNFcG9oT0Eg?4gaoJxYw4II(l2?=9uMomNJ2KG@L
z2PkOpE6`Idie|V<`rK^KSxdk0U53FFFPIjH>>sG8<_)3S>zHH5_~P$D`6hXwWCYj+
zS($ZmZ@DQt4hx0d?d|;(bPlSXU+$ylbJj2{)c+Aan?P>C$37@Cw1Q+kPIpsw`K~#W
zWdpzAENj@-g3#84^C57GH}EU-lz%uq_G+^Eb5l(IiizSqX@58BhndkzW2GrO{|@Rr
zi%=#xk65n;_ZQL?uRvHECWZ0=>GTgKw4rNq(CtSCOZC1bswWJ$-z{#5cajb2A^h$c
zHx(m<y^YISxYjp$>KbkaV<5!C4f*(EBV3swFE!SX*EQ*(K_{C=UQ{=FE2cYs)rChE
z?9tH2DYqBIXWf7N4_O)h;~CG$$^1X><!UrF<F>?*eDm}M+U)Dqy5#X7zxHVkbx8!+
zuBNk$KxivU2#FGm60tOQ`SU7kPdlF1k1#P9CN6I~J1?uKVB=m8<9=L5%;xazYX9_3
zh>|Fgg)A)0&?!k4H)Ut=#MrhjbZJnye*c`F{YX69lD4Nl+^rn+QoOEFTeyFH)Vp16
zqG5l0T?|dMimI}?Kb~OM6lR}9$i(;Q>i`hrLDEl>&`by{BZ_?QZX#iv^Ym8LiK6|p
zF0ZnRG(2nMMGALO4Yn3-ZErT3$3AydS$bOBZkMj>O7MR5vea;c$*s`Y<6OYx)h=yh
zVLPOn<z$AE?i1IrDpwGx+S%$J?dKm7pGpm5+W{)9#X`WgPuN79M!u}Bi0}__D+Y)a
z>+N*Z3_cWD*;U;@e@Pf_AvAF8;%+ab&@!SpD8kWjl_VuNr5JA?7eCO8LU!(dkH5{@
z{xo`_SgCx0vw!W0cJOX?c9giREhh5TU!@>%rDkKVlm9{3`NVtyCEkyT_AlA1Y~luC
zY)f;|wT_SjL&PNw^~qVi*tZ%oDHfeo%85TElIMha?KD33G%h;Yso36+IY^7hGSu6A
z0SN$h+dzl`0%|oh_CXIn@(LsM6Le`iJ*qwP<Io9*Gia}fYI)_ZMiDeUdvy?aN)FhQ
zbmEqUMnc`yt>Z3>XVInThpuRFdZu9y@NPbVyc?b}-Y@%89hi*DRjinnAvK7aite32
zC=(I>_;r|%Z+oHaujWH@1U<@B+txyx74-KY?77gosz02~ztciT_2v?O)}%v}l<by~
zCgfP%Vvq^bk<a)83+Pky3}kCjdjVvzx7}?(H5KoB$IBmaFd9+pOa>d2h}ZuHO+bRV
z;?Y(Z!4Xc54?P(yUs6=Q(Nt2!QX}^znhrWqDX!Q+*lBY+Hd;p?0)yC?PgxfV6LMCC
z41}^SahGelL{v<s)U$LwIu4LL91$O|(!}G-60(_j)Pp`ro*GCMVZhOd#5R;B#hxh+
zAS+#|vX0oWAkn@Y4WAZh>otA1J8sG0PqL~9g_N1_?eyX#uYwJNcB5iud2+<I!g&v&
zvjQ<l*fv=P6&}cCg#arAjJ-RYzu-n^MXdh<;7ewKR3%-{-*GY2yi7jXQcPObDJH>6
zJR5Cv`~ks=*tQX$1%GG%A(AwZFbRBer%JE-fmc?lko<6b1;(~;$cl)$wsbWK9&j=<
z?bhd0W$YJsKBwd%Y&J=;$r)QSHHk4bLBf#vBu+3hfY-J16e*)PR;s9|p@U_B`R4e?
z=0gge{xzxs8qrHI5CJ!ZCCWQZlDo>G0xsl_finr#uWLqxSa%o@Ew9?s)d?AWfP_F|
z1rlfTE)+g+ca9Xll$8$aUOlj)XwrC7YRU=o=@5CJVj!MkB9iI_&1NJTX3g>Rk?Ift
zIa!v?7fVH&_MJ8$W__$<OfvUI8u^$LQW-%)JR4vZm~i5N0SOYb#b+XZp3&ioTk3wh
zT#WVd^wm>pom5NNuHzuOt2W*plMCPvt5OfV7+2o$E9lOmPx8*+qln(TsZaiP8Sw}L
zB)wQ~qOUWQeMh~G`F&JbfY!*~tV4G$fvS^AW#O76s5ivT347Z_<;gY+X>=r%gb0KA
zH`x}C<!|qQYz{--c;!{sj@8%8W60nWqvdzcG^ZgZRwLX<nKkb5UO^?{H6Qq$;~_bf
z@dc4Y>eL7Y0DjsRY8?|@f*a~J8SMi46kmJtXEK#H6TP<+$Ae$9mK$;jZMxML|Me-`
ztW?wlfxF}oyf;=jv3CV?;C*5n8o`q!jp2_a1P^5!i^<9S{m{YPkMvN{%&taTebLf`
z^pcPTbD&|{)vL6Ee$-rTig+PdPICSE3ot|eF~ipj)dfim$68VE$C}omgz&8SiT;m&
zFq^1B;Vj7zF}fUe{R*p^0PD3C9}WQ{n)y}Sfi5FHjG$I3vr!P2iWETT6NcqjD5k}+
z*o1+&!dK#$(3pP^g-Gl-aKl1R{J(d0J~#)h%12+zZJS(s6-zy^jaVPoWVM<kiHWQ?
zoN5LD*)AYLr81WcpXj95{F@7JXt`O6JV{C$DisU+i}NB^DFK*3r4!~qB3ArOAftX;
z6?wL`d^&$aDIs{j41jA@M6!z9nx*VZC%(*qE7ZEo-Tv;`tu%agL>L4!<5*+BnEuQM
zKeBzHkc!}$#2@(+J}V~!*e^($XMrVS>*)`2r=Ob^)Fr+W^i4(egGW&*o1?(g=FSB_
zq7f<pb;C<V;dG)Cb3^#NKW;i*XAVa+*DW~RCZ_<{%c8c5BTW{5Bw7(fXnle>1vY1J
zy0x~7uN}gZ$FWO5eR~86Rcvpr@@=Ry8EEdxFSXX*n4A{A?R<)hM6>s}zSJL|$N8=9
z-I|_LGNnxST<81?Of)Oj<1giL20u4-mQie(+@U-p1N(m6bqqONQYZ}E=zh!ePBLkv
zDeKXeYJbtCk2Hw9|2FEc1`F*IpOx(;0p|JUGjXH7T_|>sn7Q{Aj-6<x*VKZ0dk_ub
z=PH4l`7NgZI;IHx=JX#Ik*IVq$V9Saodc*o*b(KSfB$EB|1WPH)BirD-}xtc`B&b1
zuhjR^C`bpNhKVb1aMEtvrL&f~DtkES*5Yvz*8Kruz$v*>etUVBm<a$BBaJI8mAweT
zVlr}b#a`F+f=Dz#N!-0&j#f6J*prO~(M%Ylf~v0k#{y6sM1o;zQwPH0AH&7Zm@$*?
zSEi<>-1z%=^B5%utMdLF?Vp@%t?~Ezz2Bnl{(MZGc#IDE_4)e!JY3?_??UjY&>4#u
z!*~_lk}xODhBJqWXbu*`l&*QAZvQ?1UaR%=ad~-<9&9Da2RuV8U_q#eM#=<<n#o9b
z(ir?=*|#7uSGd~$xYpCNIvPJ<fmXVgoQ{$2-++9}j--%7ex6E$KTi$!$9zjt+BP5o
z9AU_c@rDSD`xjS=zX$Ke!7GX0Ex6vQQ*=!;VeW3W+NmU``r4%}!#Qn$@g*5MA%>*!
zdk9rzyyu2AO%}y5UOj8E?drt2$m^xd6={(1x%X|*24^-XB*vxwT+;lTtdLO*7X$IM
zKvl#MWFC_w!62~&d+pw`n(nuz+hP7xIozf8?JFm|)6UIbLG4Fh<#RFYMZR-3gC#};
zy#+-ot9sqNTRAHD$Ivj&=ZHDNgeEcmfcOZGJYU}(PlzZX{2_BMoCL~k`$ANEg&-eY
zOkjtzfA(~~+0$rmXq_Jt@K*v;aon=Xw#u?)p{8~w<LH@QPH>=5Lq=uYzUP-DS$KTJ
zx_dSc#1%J><lvU%zL>AfqFgn=O?ss=Y<)uJzG1?bJ@4|f#(3W>^^v;<XRGj13paY=
zwpHh_KO@&aqUMi9S6Rf-avvxJ&_E#wh%2xg8R74ti~|-U*{GEr-wAmv%Q_7xLdt}D
z0<Ypln+W!etRGRiRea+CYB7YE$~E!iEOs%{c7ZyK1Pg_|BE?V+6&HJTt)lF2_1Zyq
zQiLE9If4NKBN1v+y4g~?8Q-sayWUw8>m-B2LFFWxczQdHj!KrzTIU$u4JKYe<>n~v
zS6b~T_=lKHYe-g0H3GechMJtsn)x;ki!mEZDJ3|4Rw?4LoEL(X6&n0@t|T@<eM7v%
zThs0Z)-;v#QvW+g!+|k-fT}L!C#^$5Aa0WVijDII8^xxxY!vbt7O6)EMJFj8W7gpc
z&Q>O8eZn{esB@#(OWg;W<?WMe46*%mAuSy8oAsk=1p$(^NWR2ORZgrNL4u%VE?Gi!
zm;~%HQMURNaIDPoy~aM6*uE_DQIW-|ZOwP>sjZ^R2vS(3In^@L9llmty20A<$>hwC
z<Z8_do(YDvve|=(n18GJ-$JUaRzCTmS-BBrZ3>3i3e^qkDl&H0KRhKbyx1v6{va~p
zlqp7x#yd|69bxux(yyb<LmX<R3U|jx?&7B&V_|2-E_yS=w$$DIZ-fBD8|kp$e#{EX
z&xlvZ_&rvWOceEteI0gkDZH2uxk|pjU(E~45?lhrNm$Y1n8Rf=S7~UMyoc-9h_%j9
zi?Sj9ac-sMDDOfoo*n_n(k%`BhPhnMC}_`DF@H`gU7-KYp#5cm1qj}9IGK!wK!u8g
zQi)-Z5Mwd{EMtlE2z|Lt0Y9>=qe0<<aV$2eIQ2+q3v(i1GdRZKv$T&Wx)NXp$gJ7(
zv$^tK$Q^w7^q?-}mL`;N=Mr&atfn}%t?(lbP6;0zOhKYMN4yXW51ENxwLDvX1|6@l
z%=Xqm-x(TO<yhte(eIrZg}ISVFTxUM10Flwu4SwCH?|Mn?#7@SjB#N*vwb<EyBh+`
zwN(u=8JVsZMGw|%&QY3y(yg^d1uY=eft3@%0~Q(s7d$4teSuDG|1&~Df&@kfeoCcd
z;y7ZpT^xPhYoS$JM(VBb#d|=Qd!J5b=4S1aUBb{=!<R|GJ`5aeMMgTLVFc+<NP=@_
z{l;qoL&SW}F7YOX8ZuQHP*ACH;ZgHAl26<fv<Gj=UbJ66d@@~P+t-lI3YldK?Xo$-
zO8lS@pq(S=5Xi;0c)vlLKMlYmFS}%qE5Py4OY^NSf!=`f;(J9Av=~oz1j_*c1xkjL
zw_UC`2|I&d*PJu)jSba*M}tk@)Q}7Ccg|d+zz%Ky7tu>zY?ggj_)K`uH6XAjOWjhM
z&%hXxqkNrc;NON|O+%mDu#;3@GFCwTG3j2yJ+=K3fQDAKjE@kI*0plhrdlY`NKeRm
z^<RP)i#H43^fWTD!}4(y3o5(bN-@1lm0I_5v(;T3wW?C<kk8c(<@RZC#<KM`ndG{)
z=X>J;fTajAWy}_HVTThoq4zmuXp;~&y!ooKTung@DmfjP#TKm@(sp;z7#jt<%o#q!
z>!zi(*DvcDEZZeaqY<d=9gGGKBfD5}GK@E>7!0_2&*SaZ^Q{9AO#EbwF2Zv@zv!Dz
ztOgzSw?bQ_#YXkm_Fz>Si-(&vqz$v#@`K;sX7%iTy1@n`n<|kdmiS@-m&#~^cmnKC
zi6hv6w6*FMKh=&IUW>;0KPE=Vtzap(vvB3Z`DVu}xWjQkiv?Ftchy4yP-V4XuaU*E
z#hutJbGb7t)K<$3#y=ug4s|L6USNl&k>m6@IHQ4W&nGhSHuN`cb7n?eb3GKVR{w+B
zhscW}17}g~|3}$7Hi_CrO`2ufwyRFrwr$(CZQHhO+pbf#ZM&*Z$4tzZej@svzp&$u
zedk^)^UBCJ6~J{s$!UuWF(b|kWI~3r#XD35&+I-0w)fJ_{N^iGCv0B&R(z;EC>3+7
z$Lw_O)ebN2Rer#S&3}tLMI_`f#2;LM9x9Y}gLS|Bf4^~_Gn)bhP@4V^qx2f$R9|b2
zo9&RY4A${IXY{p^nn^B$NemJuf0sm#pJU5|`$H2Il;z$yxy1pv^_<@~M$Rn?)$g#W
z$||4mG<oHgN;x&9DWKj@E$3>WPVREqq)-(_d!>)Q8}&sPye>X+>n3<%E2D#%X5Li1
z?bSX(4TLd1pC9)-^iO@TX-=?yJeQ=gIMks%&s4ITy9w0Zfn9RBOC37Ah7DV{!cU%M
z^#>YB<omPH2SVPmLGKqtqcSb9Uy{6}Byg{EnX}9dA2%E%GvdB?WxfONqgyw6^K`j9
zYQI=`phar^IX}SWR1T^Clu7?p$->C+|8}(h-|Kx_H2#rh{<)Lq>h^3nRIo7UdM&Ns
zW8JP?CL55&)_I00(ZvEuYDpEz2xC;VK6e-Sfe{80%nq!tO@hse?p<F05!Yl^AZ2uQ
zeAxOtUlb#YiBx7Eje8}9bB!6vl}#tSGrlq#6Lo)Ce}C4cM5`7UM?ouu0R^ur(SO~4
z4!>T$pJC_BCqX)TCmb_>zHVk*Mi%{geLtbha{D}7KDR$l>-6+rU&{1y_0Sn}`g~ts
z%$z1sT|5;OQMsoQ&+YnTds&}<=59K3^n7{^?P+Axi9iaXNfV_8jWcLm)l9XIFstd3
zpa3W;i`6DBF3R|-jxan&D~mzoo%S2d5|2y*%74%q2&Mf2bO-XJ^PkK?f_c3rGH@gc
zCt$Y-PD!Eu`fS5h)$UdtKg-Bkbk==(mNj35LA+E5j!uS*cc=V5hCNHzVjS9^!eC`X
zq_+aaS{8Kiaw3WPxfoLN$M{Kx8WttIx-#V5{J`23vq`S5ePRE2C#f@_a(#wwzqEf9
z7_!<58)J-8532J&(b2XNEin=GPi#f@`9hxvB56p3b$s<1`8H0<<Xod@7@W0=JvXSl
zc3SbxN5ho~RBcZ~zP0m~8>GibpKa*zz7#C?^M;CT(@REj_vXhX{*q#lleh~-7?Wy8
zC26-C-LTCV26Zp%H8HAu%(n%#>9KC)NC`rp!@3WWX5WJ(qm$M{S5u0sfWCl~yNf6R
z!_&ql8({AD!`RlOdw3Yq)Ddw*Kk&+XO2|v7qb?nwFHb@gy=zF1(uOlkV{9eu1xCg!
z7{voNP7eD*!cE#A?^+?}f%Ov4&l>c*N((P$1>vXWrZXi5Q7AyPQ@U@!3D2@Y6^~QZ
z)QGHv<yf{L(oCVO!JRVGQspdP{DGs3;Y1IdBL`ANvCof&2dVQGlu<nv=9d@Uyl~}M
zD6weKton(o7?iq0qn{E^IiVajmt~wdwCNhkR|w$>%lT*&GY%~{pD&2PxGCxH;1G_k
z0FMUfsfh8XlIwEyVyuC&bz)YC9g=Ehg3*HR7f<$?4%SO78|$>Y=&HdQKLp@k!af{}
zq=4T<3C!nWVU2DBtb))Ce2`_cLM)<b&hEvoU%f(tYXsrcM~v6#2h^Nlkpf1U=i_2S
z5f(f;-<0j5^vV>#?RsusbbY&=F8nktsIk~`5D2uuc*5RRZwplzD~t2Dpe5>;&<H?x
z!0`r@06a4x{A-3$PmHy-Om3~N7O025JB4t*0z8Wh2tNnR$|L~jo*=Syj~LOdaIM*O
z*oFGTNwAlEt-1Z|($s%*F|WC`RU9)QwV92za8&TU<obGj7wpF&TPRNoxMC)YlvuL6
z8g)}!p>>5haH95UF&+iJBrDF}!qCQesQ&~M-3wl_6$_UZoI^a>{5|?j+qwOdFB6@y
zV_?4V3`!105dp$@&j3=zF}`-)HdDB;#zk!ASzj)s#%!@^8O2M$-bUNE=@x`l-zVQw
zi{0jQ(SM>r;Qp~8^ACa1#BM$C4xOMI$*!v$tm<7%L-z)uvDicw^ljjI1vU?x))8L;
zJb+Oc&&X@{+<xCtvDD-aow<;Wm9YSW!L*@QXL@Cr_Kbu{j#Alm7EdR~PFjz9j68y)
zHPC=ig>FCVq=-T*0+1Dih^=ojq$!^^%{Ds1K1_-A^jP4qI;7x6iis)OjxA*j{)Wqo
zZ@FbwUJTNw;BuM$R0S}PBv=)Ws;xt#LxRZfeEK&;H<23Buac2zc?B7D9)r95w5cL6
z)C#XFxjJ97_JF6Xti06UY0`bJl((`8(wY*^#Dt~M3Jg9_```jl9eCXaTRj;wl9rLg
zEgw)Ul@}AMcWg~h_i6Fop?Xg^HHis$qBU$gCsI5WQ>;?RuxS6T9y||9w#k4K(^d(-
zeb|Uz(plSaQ2W5Id?4e5?ie4-EEnIZ=e%{hQ$xS9Bw!#<5-A@HlstgG)wBB!#=;P!
z?(6{Kkuv%tTthHst+IDzG5kK*ZO~;09%T4kmF=_U5)@X50Gy6ejH00uAkXccTC$LX
zch{4K`6lanMU8?pnL=LRrb{=h(AZvOx-hhuekkV|VVS0bqvSGx#*Llt%ocw#D@1G2
zJbUYHyXwY2z+}&a*js~AVuT6bTYYyWDrCb-N!xHY$e><-sSkHj;T>dzK2^|lYa2~-
zTaudG+2Y}*)?!;%4<=@qi$f9~@FJlQftbSyl|*-}la_nUozJ{V^F}$y!0vYh^0RjD
zpmxvjJSS%6CP`)PX`6cq5bGNR?DDvgA+umtFcK`xKe2nDEU~~Jja(35fY|T~uj74s
zuW|%3lJ1+j3NV8ueO4u6#g)F>Y6bhcYDA7!pU<B5C2Mg{5Hc0E5*GzVO06U2Qk2PO
zbVZ&&tf==Q_yYLG%^TrMda$X&S@?*w)lw&G<VX;WdDql2k6|~1NCc%;I56~3cySnf
zFlL;pr~va`*<@oyvecjmWEDRLp>InYf{MJlLBIcq+Us(mkfr>*^c7bfI%`IpZtm^W
zu^C{9K^0+NNm$wO(RpnZ+N3eFC9l2&V=mZl8S~b^KIdKF*cB9MF90N%9Pm<J%zLJK
z8_bGGT`}z2tPJtIkJa+RftYwcdY&74$Mq?@r!woB{sG`|{)`)Wj%?k);jBT`BIorx
zpLuk$ck@HL@s&8J%o)42NRInA^g-s&#1XcI2|bxgs>5%E3IgF4xhz>YNz$zeTeM<k
z??0xE+a~xfk73&HhSjKm!~2BZFnl66v10fH8xjO7x(%<t^P9KhJq1h;?2(vt=N!mb
zDHqJk2(%Hn)|=R1s-=n$nqycdmsYaRU**({(6IZkO0&TFd!>tnFl(?B&F=l8P$N8<
zQF&x11>vkKa)?W^#{NNmv>(=@uYbX4aMCA$*&E-xDMak!($Q$>@1_Pb13PxQkG8PS
zo*cbzIEZeG`u`F>4FAO#W}*KduYXE3G@Q2D|F?^CPpg~?Ix5J_=2+k|5w~pRGKM)N
zjpX>io}MY%WBMVU`_~uy3=F)uS+(1P2u!T#z96oSftwwK+uiSZ@#G@RE$~!e_Cc6z
zz!blnB)lm2o&t`wYfrC0p2l#M&0%n;Ezfs-%kSg;Lv)n;>zOlc|N8rStZQcb8~CN?
z(>X3?X7{Gv;q&pcEAPgRuX`j9z8)*}`|;-IgD<P^_l{QaXI5d_J=@=~-GOYnm|C_I
zjLlRqS#%&qS|gR5{rX%zMJ#zodMLc6X7rXmN7kjpt=4V#d6fv-!kJs~u;HxAP%Y)H
zs@@4ssc6$dzZ`f<33F`VV-r7+53AGfmt1Yzrq;I$$=(<*oS$+u1fr0}6sVsni+Zsx
z5#z)xKw3Q!tOxEz?f0#Tla@x{NC50Lfx-+@n=yq-_`};L$sX?ql}|{G?_#dT7pv>#
zac%kcgaLncs62~}egEs7{@lP{PcZY{L_zI(pJ|$XA-YFH?7H;LGEAQr01tvtV2s4S
zMLGfk)&CO<I`jA0iLZ@nP3!CJR~sduo>6?D-T5^hL?N-B25q1eP-Q28aC@AD6gzN~
z4h?cq79s07_&&T<jBlzI*G5duD2$<NDFb(MfL4<q;4fTwp|nbS-iBlK){6fbEYUc7
z9QXBM_z0EIwToVJiLcPf4%Tb9qZ6nK6U*AB?92)tAXq1lM^XoVIH)895CE7N`A3V_
zUzhJRz^=Gc8@+^G7Ef;1CA>Nj_w4UELPM0$&vEZ|;)&vD`F&FL2JMgXbH!F_u~k(<
zsM@Fo&tg1)QLYyt*7!C&5&8W6fI=J!SuHsYNL8WI0CNfW8j>%5A5dRYEx-&@tQS?_
zKh}|*t!vmVaUn{G21&VuG^urQ^mBAwxGk*^QuiYs>wo)2CJm8QXDA~L(y_KkBPCRi
zG+3QAefBq9*;U=H2Vz)8&^2MKDutkFCrvb(iQ)3@7~#!#gF+(2Ad)l+EOx{T-?7WW
zHc@B}PD-No7?SiwUtosT{*f~jOUrBC@iZlYxVo$)OJRg^L5J$r0}FVx>Yd?WV^U%=
z2;D%Z{#bxY<Be8QijIrTv8Ztfcaw<CtQH#CYgp}+TLy}fZIAF9^<tP8-Dk>f^IZU{
zsZ-2A*(8xPQza_Q8?V`{p&ze3n`r|vk~<<|@Sy>-wrVP^_GWG1dYMWxND-J-L^pvP
zRPoGT`wsi3r)M!`W~!GVRtC=+4X!3|I)l#mBFZXZZp4$*GNfFcEy1#mkwfR~r}Y3b
z!57N%;KH#(+m^Q2&xbl=cAP>}O{7PT7e#&Kx9LlgzUR$2MzI6VawWb7x-6tWo;F&^
zf@WkpEr&?`8qhZ`5N8&RT`ffMWK&YgePTdlODlZV)OL|GaXA7VjQB24IkFc80>V$H
zf8lz%X|lN|QqW8)=8YO^U5XTIP4W+QH|K@np;D$o9b@|`;62wfMoQ6#;yZ;(9;|zZ
z3#b!Uh_0^YX!F`kNZLm^Tx@!PfNHQwEzC@ZB4kAd@n{*&ok)+kX_I1<RMfVFp*g`3
zpa`qKptAF;yd~)N^Te+Czde|?92DW@IIPYl(w^upv0vtkC5DK7)X|EWnaJ%4vWsr=
zqWV}BHjeCa`Hy4HQQRy>FbG`Mm0=N4kYe9B*>ECkfJLwE>AMSwp=^aBBR{i<K=5q|
z1p7Nzu$x)PuGw4-sDPcc^tDn&*)!U58NM1R$gL14W8GYT^OF~rtXM;d#=~ku3~q^Z
z>3gvQ>LM6jGOZ9x0bS82ddvzV{ds<VzY4^Fj){PT5nneMH;53fLWjbiL+^)gR0d<5
zQcqTRW0>~5W>T0WNtTTA2rROO^b!a#rY@c$G!(y~Cm@`;<No(T&wCEIS6(B9M5tRz
zF{q<}aYzJ~(xRr>@rIsfHsHTajq=pVa@2<{+1uUG8|IKocE@E%NLQr#!5bH@OlNMp
z?@8~ebOwVaT=m1ZBnqpflS-?4p$ka$_VO}H*Dk3u0watUQS^p4Wn^Ko_~0{2<(Sr8
z#3VpokBZmeH5dG3t5+g2jjTF;=s3~dBY&4_1E(Cp$kbLX24haQd(v-*7~C*HG?P53
zCo%#6uU*rIF3&a1ee)(x()nwWos58LC<e<$D^M+kvp*iLUhL)|ahps@p?ab$9U(Ov
zV9on1;_;G{&UUp3nB=L6a5;X(m?t$J@z<g6+9P4qC~j_j7%S{G%}5{8YXMy}>V3v^
zc5CW<Zo~^^{seZz>@a*7W*)k7$mTfq-6L2RF^Bn|g{2fYBAp1RDjUr-RtzhpjX1q(
z(tvX!dWam!1;|&jtC?DX{D@@SQHSaosr_irDbdo)&2wsuk)1Iy%$+QzKRk9@`L_Ob
z-T+_mFZ<COLuid(78O4{5=F5XVY10CJ!+*=cCi)D>aOrU-86jsYH6xEqO4dxcPrHc
z1ka@7^%QUp%tXT;0pl0H@_G5?Vtwg?%4Z0T`=-p{78kqJwZHx|s<E{f5+Y{?mra*9
zQyQT-I|m^(e*8`gFWr7HH8;;MFk6#le;dYqR3nVqd6m%ek2yrdCSd4!ZvOI_VLCz_
zpxV`wkqX8HiDO)LnM2^VY6=Gi>T=tV;U;%5HV*)kGh2c?xH!Qm2U$BO=c_mqoc0a+
zxT_YD4gS*}UMj`bnJ-P{i-+++lqV)w8<r#~dP(F;S~m6se_MRH#+FY{z*c#3|E_=z
zbmkg(*hHgPIX9Yf$_{FN*Me~O03t!YJdd#@Je9|W&a?V8S@C8<lRJ*Ef1y;xA$B)x
z<FYGOOI5>4v)0<FZR7>h_{qz9lp8T}Dp|5TKCPVamj=;ZRrs9Fc72v)<~+^L0K-+>
z%}h2BM9sErQ6KCuh~K#)g7)srLpp;L2R%As14ZcrocB~Fy&=@3YE0=-#}|2^(2DcA
zlKzYL{I&c5N7elm9$nN2w?67uHus&%K!QZ-$ltjlg-#G?-?;_E4XsQectwlw5@!%{
z+pcqa638*b?9-U18<RYrOb=Av=#u?fr%(M7keCPcX7FppggYR~NjkgMG;(&}UHVM(
zP+VFb|2!3n?7#-hOE9Z{P=i02N-}1S@7R_!ZW~OqnQfjJHR!r8>`x*_Lz0H~Ub#Gn
zR}bb~Z)qjdSvd)%h2w*jG!{&4W)~`tR#-I7r5Y?{aD`xG66*TTxneQ~RzuzI3zph;
zI@2%tB0B&B5M{2{E_=>51uuu2aZhc@RmmODcwRAr8}}w*j?1)blgJ~De>9FxHilzL
ztyo~q69WJ>?1^F&XQ~~3eg}MRS=IW02azav|5F&U{h#rk|4$bZ{NE>HK1xzf8}tah
zkJQe|W|8q<gPcfU6glzD5sOk~-~pf%&sN4B;y*sdHCydzxvt;Who7!$rDuT@8#6Z#
z>xS53;mEn6l#_E5k_t#=!K09q&`;1!G%&uWyJ%XX?<6rHi2DnFSsgSk>#A<wPbkyF
zAV+A32Dl|CeSN#(-NXHE@Ldr!wr^S8<f^{0c-i=LRR<VH-+%7+|D}}`>Z}@$tp{20
z&rkLfjk!Vp2DvM#=n2$y*2O9Y?Pd?SinijG|C))4nwwit7*a@zTUH=pB}|_jrCev}
zmDJ1$YJZcCuaO79vI9i{1_a3#f}K(juDCS<3^VI<EjA+{1Dp{I`TVgx*d|wRRm4~Z
zNu9l+H+5xGI~K(d_N$atn8uV4go_N|K2bMchW&B~O$md$9Kwp<REf3Fk1g~cx6*3?
zsDMKt{u%#()~}ft&8cS8=>=AN(p*O0RXMA)tJw&5?B){+Y@o&wRvP5U3)gi-Y6KBZ
zt5h^)M*&u~s+)w#mN>&=KU=cGt%sS~IK?Y=;9q1tR$q-3Czs@>{riTY+&7t&Fp{4r
z@>Y_xA!b+kgqrgdJt!7j8%DT&Lrz>wx&MM-8!YLX(>{LRvcA|+vj$(6C%ZH6hhO1T
zH^?P+PHV#R?wZ($ku2vIw_y9Q4qi<0LH=JK(!aiv|1&4X#KitTg@DDVOT`_r!u=bq
z-r4A^S%^Idr8;GiFm<UDK-O>a!)4$snW&_>X}gZNx|IF;P$V{*xIHiBFHmT}M;4C9
z;gc4Qi3|UFlC$>lZS{TLD3KCtkX6t;oc$n}P))j|pr~MOuP)p+OebyQzy4;V3HR+v
zB$VCP32i%@y034SgoOK9!f{m`2H{$>v6PnV#QEt?FFkFak$AsS&J|KsC|7Rr5>+^Z
zA&HPU%n6c84&{=sJ^Ce`X9pAU!pmC!E-kqh;~n}cZ|TUwyXM#a@b^7e2Y&*Jlz$qR
z8R^9}-^@Q2HrVB5-XwUAr8$rvPUO0$rmYnfALnoK+-waHWk6u!k<KMmPoGgVaXc`#
zHgViNHc<@M@%?jUsF8JbNuRw5O(`h=zU9QaHZN(%0^*mlo>q7?)+kZ0GUU89Ay}B5
zrp)lmuwRg_wVlgbwfjqgYMNOPMMgHCI3)!vx?h~Tfjo~5LHlqIzU1KBUF&!D;O;7o
z4Ve^*F3Px`NpC(7%|~j9EiTuTyzDf8LTmGLWr1`4b~7W#VW)0L8{hI<X^@)KieIX|
z!gonh9uDs+V$q;5PUMVrh48KQbmZrxh@O6qiQ6plz6delF3>5TPe-q=&BJkDI?%fS
zZ{Cyd;C>BiSo~X{D9YJ%HDvHpDvZo3n%t0|V{ivx%`#lm>n?BvwI^3UoHNSeK=OQg
z>?WkBV6ZR?cd?6^sqNk5dU~|$ljfIYh;HKDv9|QdfSzD{#_;>S&?8Cgp?*uHoFyA&
zxBfLLPh7cYQqTHWA2#dCGlysi324HE0SRhCeCc8*^2ugqc6bE&T@h2wop|#tt9t7w
zLk*Ks3ktUt9D5?fp{RT)j$g;tJuWPdoWaC-iH2Lq`bP8t1X`G<$$d1=T;a~o8d6_e
zpc}%L?k7I>K3&R=oe>Z6j;jB9${nS#7B&(a+5_#``$M-{ty3IPUD*a-pxk{ccRCfU
z5JnK9Lkz$2V%?S58X(tAc$^~wicpO~|4*7@;J_AZj`~nW0lb-U4Q-2O)INV5OO~qR
zN&#7_T(SU=M44iHk2g{+bmsrlNC;OIM9!sFoj{Vs{z!q2`*W+!z-u($!QlUx55xZC
z&##x$9%?EQaFW{+ImPiV4il%$e^wT?O^=W%0Da4gXj0<iIUe}5m6+9)B1u%<Jz<S|
z%$V4jN{Ap5(%2ptr4l<NB^9bg5V;dbK36fL`Tl~xi%=bc2gyU}9KqU7r6DF#w{Q#^
z6joTx0}qDxz;|vz<z=H`E9>t|D@#;zRDeY^k9*6`aZ?>r3y1o4UU_h5F^rYyR2<0Z
z0)S&QoA#}xfEu!hlPV@Q?36QKR<3rIrT}?4s&Z;;mI@W+n4W%0>I3vh5(rh7d^sLg
z`j@8cX;!i9izES|B(LfEPx>si{OkE?EX~Ss=ggdFv#*4S9hwk~m%4i>dbgYx`ZDE|
zbjd9z(#fF#rHBU-Eg1$yI1_wt|2Tc#lSXkTm&+J2re;8Zd8Yw;OXw!6;BWq2At0x=
zRlmp0mNq-DS%a&OHp2F&22?*>yQ#%<)=*C6|9j+_Q1QFV7JPObvj;~4xGiVgvF-<F
zOK!n9G~9i<AM$DXw|X*U*S=pUYQIk#2yX*~SDd2FVsZ@njL4A;j8@7ND6oY?!}j^<
z^XveeE0#O(vtv|{XbasGBzJ8FIXC-q(H<T+=yMzxvAWU9OY&aS{S+zikCw)+0xY;e
zzv9f#l(+8HKsS7!Gm-W4Qxy4NEOFfQcnE4h<n1@VxyV(1GdS#7wc#Z@v24!VN9Y43
z3E@P=V}TgWYzZ;S%VmBsy<1J{E(j?AF-5Sdl_^Z14DJ>YNZC!cGaQzM`zTP0@_D=V
zfD!@tQcwmh^ga@yNudOc3?cQMtA8ds9LuR#AYGKP<qV7~|Ds+Rf=HqEzV^AWm#nN?
zi3&nO=T?3_HV<8Mf~P2j|2!G}xAT1uylW;sMoScnH1EbDU5#lN#R6e=vdbB9Sq&N$
zNsDAUL;@HV3qK$A!EUTHEUe}Z)7KY^STMs$ILBkJiEG0G6z5=%`HZ(F29M50MaO0*
z6-FD%jS3LKWFKVqrm0<Oq8ek?O93!0Q;(CbW4-_o$teWeyE&CJQKfayB`mxg;I&F>
z-KxC)U@lvGNgALEuJLC|c%Gd#v~JmwDeMYFZ+>u55i(FmVIZl`F<wE>0GIa9LmXNb
z*Hq;3EF)^0jM)soO?U8>)>$u&&QtPbj4fa|*a751iz?g<w3ly@N5S-b{H`ja?K=9&
zNtO*_nef-`&$_{It+KKGTy!J-1ryZ(wP+#r%4}8fUEH5w7bpUDcbAC{&M|<it)W>X
zf9FZA49$aPKOztw%^{&+OWFh3P`p%C&-+$WJ{|7v$~WU)bv$$S(J{1HZg+7=Dd~(9
z{yGe0jKXwyM;A?x0}7ync3G2<HGUGBLhWd>w0pa*!te@nevjNl*6!_!^NUwPsSu4P
zc_zo2n<$>@8ojhn#+f9lq+DO=qt9oPxj7_xBm55@6ld-9PuU$AJGi5Z_v?T=+$;T;
z)CbQIQgB|V{);4yR;UuYN)3uMWYXu<m!3uftT48xL)Qib2E{jE{OHgzbK76^Jh$y;
z_<}o@1$<ow&5BithRGzF<7v-6r#Q~5i73>yXS6ulA+f5O-w(GqM=Oh#mS#iTKJu4g
zV}bWVWRj&9eS-7p8yxNcOa1r6s>EZ7q+J)zUd=u{KG_qx#Lgo;-(M$L`t`>*-==Jq
zzdLc;>d=6!vuV5!AP`<J!z=JFPBu9D5taRCxpVh_@;*8p(haOaYw+WI_-IXZ+}TfQ
zeqNn#!Y1Cl@-z_r2fJ6C=`PWeIr_hcNPW3qhtjCCE9NbfljU<6%(=C|p)x3RYKqH8
zu6%nd6MA#8`B3$^x1SHl)<U|kF*#to+u8VZ5b=c~20XT3-LL3uWUj$;oA_S#q|}sM
zrAg@We~kRu;b$gbVt}Wm-%g{Y#oY{u*>Cl|35nVr6Mn=EXNzhVF%ifY^0jx)ce|vf
z(g2%w^}w1=ak;Fl4w?l}z(`8PU5?V*+)OH(mXG3IZ-2nH+$XO7Q*-@CA15O#3)}xx
zbCvva{T=>4#VO({8tMQb_~W_RxO3G_85uES70Q<6;NgWdsqgaQlXx?Aa=_d9;+m~)
zP2@l^YQW#mtFZBOH|XB?`v-My=zQI3`+`ikQ|=CJI>GluiVTNTnp-w#eV{}tQ@>l^
z*I}=LmkMk)QIOO7oZnB6%XMz6ZBr@H?gS`WwT0>6d3L+o-kqH~p%Qm?)}C&UKi|*S
z&t+X%ReeQAANarBnq*Y%UGHD#Q95UVNtR4KAcsgt-h|naL?(jgBRZf}c3Eg-QAzo`
z$LeR95A*0U+d6_g2>}&sYxP#S)`Yx^s8yvpk<9t6LTQWbg}F8IELq*bKA{b8JD<-{
zotH8>$IX@vB^=kiC)j(4Qu~*zdoK0cD)B7Ky%mq0ZOasAlP4A2hKofOQQboCoVR%o
z*`vYHMaiy}foB=TlQM4p*Q5fv16R0`S$|xcxV&mh{}M(O3$yU`b+w8r0bVuEHJIBA
zRximc;HCj2_vi8-$##sE`rL5dO!o=_<`(~9Hrd%wDhic3B)_bP=b{$QCuXVnKJcwQ
z@LiP?K|?lcO@boxh}>kDR~mRn`yQA-uVQ!KI0gnu9*ekU9fTYoKoYZNUk9Wz2R9#6
zazkQB;yz(#hI7qln|yJP)>zOGnf=Euq*;bSeGQeC3st7XQuGK91egIuvZIa9*|VXJ
zE2Ws}oc<FV0%7ZUr8gg4dWDtDyou+Qu9>%nC#hK}fGOL_Va(hxhGeEWRwN_L5cRcR
zw;jMKG>HsW{~=rtnG?fL?+RJoS+illg+=d3P|opfVE<#2XT}pKhUD%K{o++-XVi!*
z*;_P%9SY8;s7?t67mZbv25f@L;$QU;4;b)k+EscU`GIFeuB}&?KjmaOaM)3QX^$on
z_2SErY{#cf<2!Vo*(o+6iPOebQ}MuS$Hix`f~*a;0wS``^d(HQVpw22?2NQ;dvmoA
zl8VI^S9&wUqjJ4axOVk<upkk_vn$qCZ9{doc`GfKBcaIZ`e+16G~_H};~;t{3U~W~
zM3!fjV?VDMOcM)gDU<3&!@5#QC^uhK|F2HAa0g1>42AdDq~iik<Tgx$zP*kM>?{Co
z%|8(pYKyi29HV8*!uA4o=)qbIJ%f^GdeJ->i&niH;xwxpoSaxua4S@nWpIK&ct1jy
zN$3pU-FExA=3YMZBMC4Q)nE@tt$@M7I4)l`2su9`yI4$OK?VXXyE^#2lc^fHH1j@Q
zot)Z92(Mm*Ntv9z2p{Kz_O{_mc<KQl;~E68;wk#^ojtOjxhAi$f5zUA*_*!dQ3$Jh
zw#&DENc^3|6jFrN&Xbbmfqxvgg+GPB1lCS2scb+d=8tI;WwoEiD=cv|4Z*ByX|G#k
zSb4i^q)-WCmUk<dOvuM3Evt<^vgc4xJ)m3_gcR`$q|pcpUkSv}C_|z&1=wW@hbLJh
zXD)Fd1Qp?tn;-(>m(_WGLlH#RkWvi&VSt^=61JFr1)DR-!1Mfv;%D$r*FZQQeTWUN
z2kTLsAf2N4>A`%7Q!!F7Vk*$JjI$-aa&VBYB%n#PMtb>q{gfKIBudoa`pIh6HTI?X
z?1Au4f=jc^RVlE)bk)ssh#)0kjc0E>mz&kCnkvjb#G$h*+Pcu^-DsqJaN-<u3oXEW
z_j)1V<p~aTFrgF&gX7{YLh3y?t>9=Am?#F408Kpk*ip|tC1k;jVB_Dv`Q(8nSHve#
z<o)}DiNQ)UJ75YJj$}{P7#x~`M^I{MTy>rr2y@Xs%21_>lL#vw=e6j2{#=Va2{dmw
zVZdyjnOtMG>2?iHAY+7WOT+N9hLZ`C{1TCc0Nye_!-Ifpg(30EuOxS(fraGSsJYCT
zGXfDMj=^3LnAQS9494rNC3jYNj3Oc#g;-^^X0FCPe`F$>wjBknyj<+f&i)Ef*BdPP
z%4iq@vK+oNdu{lxb^%yvflD@~1WC>7KJHQ&BY>86UTvCcCBI{`$FLH#7;8#{cP?#`
zp8JzkD5V)`@j(4G15OiWkEzR^x{`m{*?yzQi<g)zOn-pn7QZ%XrRH0LbCsX$6G+{o
zjgl`wsD2vH%&b12gMp&;HdMMkrM1ZJ!|Fg;dEt>)d*mb(Aj7__IH1GdH}0YY$3j^w
zK+{I<8=6L_y@z>7H_IRdmH7I4l{bdxr8rp!&N7`6WB|PBLZVF4=Xy=QSD7?_L1Ne>
zH-`c8{B+MUQ=CA7uqj`31*^K%ORW+YRh^0lsyyzbFZ3!z)s(jGX&xZBMA9tZ!6WFc
z^W}KrnE`kW)M*+JeZ@oK{HRsigV4&-wozEZ<pMZrGNuEIA_egDQn{O>((@SJ<rr&f
z(23Hg_}W0*J8{<us+_0%MPyp2(#hfa;B`427!$I%xY-XkoCxu+0QEu@M1cDjkqq$#
zFQWDu%Dx^V#6uGu7Amkb6^83t<<<~~ti@L(>7Fwn7kSA}54)<`WC?RTbdsMO-ykAS
zVA0!rX6mnj;PwM^IBi1nT7OYaCW)rz$BEmk0viqi`Ua{!dm)_=)#(@OCBMZpCa)(|
z*wfklcS_4zh*mb|6~#AE`=F#PcUAod!<5T-WZb}PUVA_oOpxD2U|WLUB$+!;G~hXR
zmzBYKY7G$I6$AX15|Dc8&UZ{l)hQNSP*wd+_i?Qhfpi{Stf!rIvw(sC*pvc@uFQnN
znP(ypV)kZ==a;QOheBo-2gE&o#+ZS;I;~$m5Q^k^r<+>Cck=+<svTYBeYWe^ZC;pX
zCwo!`I^U;fH*QCbl*Lc>c8Cmb_0h6thi|#S=Oa!}6-{T*z?R*ORkgf<L(k#y_ApHd
z24DzS@nnB9C{s);@^R|$dtMRe4-hp<AzIyGTHT>q_c`ct{b4KdtmU}_Y*xOx*9L<r
zbEf;rMD>AxQ+RPvfe+*U$}qqow|cG}qVM+hBOj4M>0@KgD9z7TS5%hL#sLGdQMg51
zK>tt^L*yOxzc4tNG+xbTUWHdQc?m(*@ahQI1+V!dLTSRR+)FG$byiwvPeaSeJ*K^6
z;J&@YUz;n7J+7+bf9-EIKuvfD<{p>G!1Popcl3~>Yp@Z&l<J`O+%|Kplwy6TBY0aA
zOjWO5Gs430wqcC=QcOA$K=Q^9O9Ar#=<&Wi9LH8!%oe0XU6P|O1>A!~QI17GQkvTI
zI7gGFY28c_QBweBlMtxV_d5{SrQc@AS@>n}1d+*WjG|O?<~L7>^Sp@W<g;O&5^s2+
zqxdph<8k@pcNPmz?iIcN^w6iZ`*4LvGE+hJGJDuj*hSa`5`hnO)N@KE5)2RiTMQdM
zMO*I1pCT@JrTv3bN!Ah@jvTOOPJaf5h)qIF9p5WwvEf0HwAuYJWIPJVwrq}v0gYBn
zQW(d40{PmATxpj00&gQz8ZkZfo@^=ZgEs`7ZF{w5`#2>wU5iA5Xdeb5R+amHq)2|S
z>oU}tr;A^?ih|WG)~<CPHiu5x=R*)e@ZtT&qO{~QTReWcG-(%*`H6<{z3E$ZL;nS)
zxQm(kFUiXEUqKaS*8d+4Of9F)c2wWje@2A!42S_-eX|$BO<T^*46?=A-9(>!%1OgN
z36bI>mwvmzd;TQGlMQVX?xP2wX}4Q_z<;l3i`MySeS14RKXVs|sKpC9W^GwB<}(sA
z<trMqTwW;666qa#6f;lod3Ut2Gy0r<ZGFGr_CH>;{Ca)pM!Z{<gQ;Mn!X{KZ1k_Mo
zrzj|^L!>erq=#TxG7MAcHH_K3((7yEW>y`|dsn`nBdHU1S5(^ShMsG=b(yku_ifx@
z`kq%iUDO!uE+5*xVB}V1^jxmzo6ByN`XJ{fU2*A|RE^R2t@inuKuFlhMITWVT>QHA
z5S(el0emot|9Zee)z^NW)Yg2D9?B{>SB8Uzd_oGHlFfH(*Y%{k?1q7VS8Rk<1^F#(
z-(2{}eFegWM$kKx)euTf*6)3m=l}h-uS-zcWz&iJ*bLtN$vhTna*#Sqiu3yg)#G^q
zzIrC3nO`k`H_$55*pH(~!M$y~|KQU<kuzx3YjN+i)PaQ)Il#<wyki_N4xV`|-F6=H
z;gI&<%oi4Xmo_efKrbJx0UzfN`@PO|g<$j9et~UQX5t<V3HxZLtm5n)i$aeWd2m}i
z;D!RNa_*{oW~g;n;GccW@=rX?QsRo`N|SU_q(>fhgj0qie}ZCDJI%@`&G}_bzSg@V
z3|xQ!%TqPMrnS9fe+3EHqr>(N$y`Jx^Fwdl^4Fd|a?#q)_gT-K)HzsyYbq30Vir*j
zc-?%9dNUsuLxSz`(W{SUP^1{~2{)mp4M#A{4OdL5#VCnS>nw=&^u{V3!N89+A7z}_
zD=+{&=|aIAsxLuk&7(Y~wf#Qa6oZi+)+@3L!3f;P;B%SBIKrKe^HcKU@UTY^PWN1b
z(No4>%~ZEhvX3(WGvf{#8w7GsKiJ9{0r||J>TIqh*&1gze+xBirwk7{pE!X?Q|bX(
zl@zQIH}lpArvCPaq>U;PHVtnP35M2?^7oO(5T^^TO_@<`a^U+zmC}XjFgF9LE;w~t
zDFQf41-P1smKH<XPPkDiQlz&d%AY``Yd}p1lYDgWTfb~VBmcIhL{A9XCa+5Gnb~>|
zh)Au<Knkm%Znq?F)#bYF&81E=45fCybc;_dF98c(94bE1gi1`7NaPuSC_I3Xk_$$P
zkk^7#lnOz$h)LJ-H*)hpkOF{spA61d8$lKk%wLx@yLc~vJLDp&ExbjBq~+T1FYt$v
ze;_He4xHLO&!9_kAd@&>B3<Sgkam6;e@fcc%>ymMLA84fvHwtDSI-*aTicKe5b=mo
zT{L$Km0OqmwYxEJ!fy#$j>bnvwG#bcvN{SeAVhcAxdyNQ9$Lac?WNFmZGxLPZhhx&
z>qgEBUX3suF>GNdxXlsp&n!cUS&d~lBGvYrnCebPKaQY;R<7+&x0m=ivOVMBoY^Bt
z^j~@P#S}B##lnZ^`xTj!uW<<3yA2>3Gqe;pK->Mljq1(M$2O5b+#T*AF(4t4fR0}P
z*pOh~Ej&pb!}SzHas-#3U__@FxiFiK>}HpP{<MF4CmtUbg*S=OMgMZy!JdtKPoY)Z
z7jUP3diJ*los{twBS*@B<+KK*{@6n?d^Do@$Nd4&vjy~hFWSO)eb&>l1!BN*Lw?yv
z9rIabMJ>FkGHs&{OGlnXmC@LsfdKT?(wF+3z3dqG;REesq0g?jYk#G3xE-2H=Ptq3
z6Uf`|vS<fsTx2sR$~HPR-ws69Bo>M&d9Sm<i0q1Hs*fmkay+I&Ly(U1ZB3U;YbQ>(
zF|Q;DbT9*%H{S_evV4$Xx92C!PqA>F`6KRQE&E1OY@CEItIoUk(OP{ov{~Q<Va95O
zvjXEDU{!<WsV6{KOPn>F!~Z3;%68^!e~Qic_Gc(g>Y+Y{4#n2G-{&mWHq>rpuY%7Y
zU>5+0mVaZYw)(>7deftAmV|zbP&{mavQNL^;ypbX*;p5YNX2?ptn1P{s)h;djB{R#
zq9yLF_sVR=-_XUnve{XrB0Dxu@;I<-sEtYJqML82ftS}h%aW^bB`FB|B9Q`tKf+^g
zZ<eL7WFwu+{`F?>kN;&|7Qq^f&7A|&47-0B$<A}jj!Ozxs6q4HjWjHyXqdZl)`XWh
z+}j&qHrjS&Y$fMEPQ!@3VG)VRqxEu_BDoExNymv+G{qIG8cz0};&b2@jqYemjgc^K
z(Vwyc3QT2obLSI^yc)X@|C*;nWjX!w;D<-(5sAR7CWAC+6#PvLwIn5oeeolNuM#&i
z;_IBI;>@t-s90B?N1a<dOUF1%tyZ_^?E8v{dzs#%euL0!kA}lcq>(2acxy#i^|ZSq
z7xNbv&G#EDZmrTo$a(c))@|EW9uN&(S0LeB9vyBbBsP6AWH^zhX${jBMwdX;pjiPS
z=nf#obGY_z9J$?E-;RzCN0!#i*k%hjT!b~KaG93Ak0-8~mAk(!r~PY9;C#;TJ9gep
zeN(<&A8fvV#ChOX7|~E*Y>7jny#luadOHe2#`6v0==4z0&EOz=k`>mi*Bs}df6Sku
zy>v4``ZGjBUaq`H2(;|)WkMM9(^NBW3+2H)G#q}i1(&TdJ%6eN#|-#n%AA@EDn$8!
z_cayMNCQ#*sJ?S-PRna@vphELqwST%<EU)kO*)FwpU3MImrGDi(+i9$G7<5Va7-%<
zskBE>UiMcM{(?p7m4A)6_ixQDN_Glz1n<sJ;{AkqrIit(<|sJxIToKe@owav3XyHE
z*1eTL$(;&<`)_)g_VG%2EIRaO-}I#pMwvkUw%PKKEjo;yM-5nCK3ShnIE-cF<P$*}
zoVo^*X5YkpBbd~}n-{aPQXr=`_JEkM4)}yfNP%ylTj|^jCXA0Ss#Ad>NdK$OvmraD
zKY$Y74eSH(%XuSDls>$$T4r}vNze4w8zjU<db6<~#H`Bd{WI|S+3E!m*S0_C@Z)6H
zlqk#zFSBpH@AIJWDBmU57Tv(@5=XK)Xef^cOL35a1MQ6FSS-^-;~o*HtS_kO{{$H#
zr{x#xAvz2LQ$+I;#BhO1it@o<6CQKDH^aJ?dHlQv8VlF|@Z8|dpO@}}`j){EeO8%d
z#WPt^H9F^M?U<I=RXY|ZgRlR@;M$14`urBM_)Kwp&#=Yr`eD1dGG+jE$P@(dX#(}<
zp>DC(`qRsp4F_kgp_^@=wHeL~503Rp>ELM7#=4N+b8POab~rp=_7a5?>FBUWSxORo
zQpyG=*}Ea{Pl<nT94O_xu%89#7w_fgS%+WFZ(!dyd`#iD=YL5)W`_UT3^OzSPp6dI
z|GcN0<Nw=x+E1L0H0E@mu7;PazT-}9ayvOKG<hj0KtfR<h{OmmX;AX8+veqI4?}|2
zm?3GYH#~r|t@rx$M2fcs%KdhKAlDS7X4e9xm@3K$Wje(#60m@HxB=26az}|G!Tiqs
zdA_VQusG0)62>rJpY`*VR+VSA%9d2Y%#D&MQ%ZKQAC|>-yQ8-)t&H=Nle*LG_4)E~
z^6`DOy{u<(@{!i3BV&j=yW`{ji-@UX4{j4IMNeyx3r-{-DpCOvEuKDaL9YcIxbDN7
zK}9&E$C0M8JujvzBaE|Sv-&<d!(8%d^SavAIkC#DYs>anO2PJ4GFa%U73Ja!DU=Zk
zuc31aEw=f`OqPcT7KtC5GO9I%N-^~gafND5)!Do*Sd*UgTd6J}Cl;*#(W66y0QZ(e
z8=s9PK+g+`CEoepldKgheiU&~pbAjC87n<}o(hm$9@5KgRc0mMviMvj$cxeKtlqNN
z6j40`{FNb(&BR5RgUql(`9~m(Xp_dXTb70U!lg)QBs%#9PoBl2sYYHkCpd21GD^{M
zK1hdpwV8i#C?8pO4A^J@^{>;-b==T<YoSM#w>F?w;Q~=z<XAN&-ecxV>J$UhQY~s=
z6Ep63JOr7CAR~<<^Bb&%F2_vOP^KJ5)N}Z(?#RDV<%&JDXutRZmdrcL1`2s>{dEw1
zeK)v8?k3ifq0MZv$Qkbzo#y!{*l$j$MPMnVo<=Pfm}f*%8VQyG!poC=U<?Lz5}#!L
zCc%=@eiIGV%V`~)YE*15kW?Hayp`h%hO!R5j(vK)u9UhFOA+Wc+fpVkDgSCyFE_cU
z8>&U}yCgu*decMa^(`}Va>e~^HB~!BE-L!>{TY@SfRnG<MnM4n0JVPDPW~ul*8G>n
zpXDJKO8oGx&Oh)PDK(E|b(Wc863}gB9ds8j>Nfc82isACZ=U{oR%C=o>VT;#3%!u-
z{T>-~3WVkbk3jhZ>Yws5?c<qOiLEK9GA3ELGT!T^Szy%@y)Z>uEqiFxYO?rQ+-{>q
z+bx$j9P_cW$!H;<jnkW}>ha(=IfxL2AW|_z0*?Luf%Pj95c}Zxe^8=`7$A6GCeSX!
zQm}W;mOHmloIa>7Io4!Xl4s%0dG+^KZ1U1bU|`x@8al8!f=oIzQ}nt#cHh^$fD7@n
zt|}q`cQEc8!znh_F?U>Yh)oh&Up)ghIhe@GZvt(-9}h+DCjMP2#og!ghv9e<GD|0~
z>%bT(=o7dhB`LUImyJ%ThF?Smh2Nkt@!hzu&f`ALK2#EzYO~PJ`DY&L^fWO8WP$f|
zl3iwy!N09at4&dK|Jbc2HDY9mBSerVR#g_Lm}_}!+*GnbRPd?jecsgKWoEO%fgTwL
z+mxI~Gh6M!bMVmR-RcYwXUWsg1^Vtmh~Q)HdWGlgK6qxV<1VAcnNngS{LZFO#{tX}
ziub0r^oYrt@I&c=*;K4HH}4nA!4!t(xQ?8l`|9Y6!9$O+K^5^Zj{QbZ>pop5u?yni
zv?qBAP%<<xJHtLZ#&5C^1ATKe3iwEORZ*O~bVc!$$b)Z!yk08%<mAX=b*8^0vSfCl
zG^(+3eq?7++l3x7JIkjL&~yg`FZOT5-&eiJHy-?xLj+}AP2t=*PpslYeAx<+<XEXF
zLQE$=hjr!rchrjsz%d%ei7R(rTp1U^7WVHuB44Log$orW4bjYtjCiS|Lo`diEbDv>
z1B$HRr4`dP>~fuTz6kO5W${w(+L=QlS>Z<Z8x4M7K@$U#BK-1lqYpTAc_j$>Z{*$d
zugVQl&<Z!TFW?1Sb6xOJ(ghfDDRM4u#mdJF$nv(m(w5d(5aE+M6^m3x$y&FlTdHG|
zZynxSS+X#(I#`gl%OwN3T|3}x0U%-2zXUvIW>pih1^s3T7{7ZCN&Q635AfFw5OJ8v
zlwe3T3`|<wkk^XtOjm#g+v{x*mYTfbun|@ivE2XqB`@J}r47lR#c4JX`y(;18LiZq
zKYf7tm#(q4Eb`E##8VHJTH-q+H6_e$>4P<U3U*5YM8$G@i)mJfH?I(DxsvE}V`tY2
z8F_y$f*GX!GrNbnh+&AORS(QC6@wsxZDg>8dCj-wKLVi@i6h|C-OQjFuF`%uv?|JG
z*Y1!uJP`q4FGSQam!fPyIVn0zq6g+-5U;7mD7xo9S(Q;}CtKT^DY4rKy_}=Yj>=qm
zE1;W6Gp(w2K3X<~LfE1sKOXAsf)U2W+Fyi&b>VE&V@v$XM{u6g3ohOyi>`x&nlRP;
z6#X-X8@q5X`q~W}&boN+)N>JuN0DY=fZv$rOY}vHuS=y0!RH^uHnj%nH`?`9T`RFI
z(FfvdkEdGm2Lcg+A#8YpETCwisMadP3HTC*PmO}!n~K!7Y#8rE-Ww?Nod5tq%OYA<
z@z4#Lyy_M3Ls>5lmAgzorP$BB4z?~MpjYMHKCPl)v%a;TaKK>Q87%p<YM;ZcvTUM-
z`_)Md$CnCOQDU@i$GWn?kYWjA4l{s9NpeMZC|5t-{8cvHZ+`~plkM*Gf^4}sG_$=L
zV4wnXB8$OKn}q?a`J+yU^Ktmg4Z^(d=70fhsh1ZP!8-nKVZp~#CsXl6>q3mT8cY^>
zKzSDs##)}^3v)>&-bG^64V>5DP+gYT3(o?l)jal7I=VN{1g>mzHRkSWG}(G}xw~he
z{yA@a><tQSn$OoE)w=ESrBLm7zOun8U>KvJxd-)${ecDMt={r3nGliD(iWR<kJUf=
zC=iEU@|76sqS|DxX`3s@+<xI0vlgXQq^Kud0Hb#VK2urIV0!&n4uwyn?7^r`P*{Su
zk3(VCzmaIm>vHAAcUjIU%LaC~tdZq7@~^mktktaJ*wgXK-Z=5nSRJ{Z^ilBg-Xn)R
zga1Xp?t^Fqz>DkhqiUFW`gM4vlgVH{P_MJ<2_h$--TCDTiR5t*M87fC{FFDrkqTV&
z0zagkLcE^X4SQ*<oOG4*w9eC$EG58yeuA5&&?ft-La+42M~5QJr}}XzonlXNx$-GI
zIIipY`iKK2BMKvP9Bsu(M=G{^!9O;aY}-7sXD$i^H@nyKc|QQ}6T9&xZ2tm#qjPus
zPu=w&qc#Rsj{h+Nb6G>nZj%+ocdl-}h-Vn)wlLL&Ka3POv}6S7mhunOvZ-M~YqF#x
zm(iNvZb2-GROHo0Y?Cz|*~rxX*sZp^xYF%T(GDFsUyR%=EhcQYKS-vPo%bo^HDv{4
z$|Sa2&n94!aN@qW1!sGwyU*j;^T_f@*1i2ODz3J+USH4X#qwqzoo^RrFaEEmE2W51
z?*AUWO%<>ob_tL-3-e3tIhAzVX6$%nb=Jb?xG?a(dk^YsTrkK%2x2}KGY4bt%b=0h
z>X?Hb#+(WfmZWLk-W<IO39M)t18sN!@Ja3!t4_I;K%lvUg)bH=g2dOw#7+A&Byme<
z9eK17%be>sNS3Dd6h#`|!d(V*FNHqxTD+g*DMn%%WQJD7mJeKi`ixiuHFNT$7RLig
zTy>+IMkhj)F;2>BSAh=a=s3~`wn&z<8D>L}DG-~P4GCVw<dEDut2;XMX|J)5)lm#F
z%1(wNMB(Y&H1^a+INxzM;XyGS(wUiN;qLXB5f<7IuCnTtGORo<2l+r=zKMPzd{Vvq
z7RSa8>55vU-X7M0@>-YU@K~{b>qdJ2RZvVdO9OnwAP!LM-XdoUw(9dRb$c^LLC{&J
z*D6JGyPCDhD)p}~r|pM2(U?Po<Ck`^w3Jg6iZns5|C>cJscOrKO@hyq<qP>p5y6C#
zzGWcC+q4G20n}>jk*-xOJBY!>mkK*IEj#5Q5y3E3B*IgW`hrlpKX}4AQxB}{6l(2d
z4XXx{X>WjF%fl8~#O=uGg3po(#i#>6t*T**4x$107rWZu2Jm!V)!rBRPpZ?Jk9QpM
zVI(GjtBfapPqe)5zKSptxE8`6Ohe!6R`MW$0p#SDIs1G1wldP*;~}9$jVr+=`3WL6
zkjo}|9TEp-O15;F!gp^7>*JIiGlLc&G?u!zchYP+DTRJnz&xexVZ9_747GZ2QIJ=}
zuTw{$Lp)$PmU@nKM1GvRs_46ASIJ~K=2M#r1TmBg51v^zVd@n;HTEj~@ti%_mM&3c
zuLgFbN;4s8D|>;yKTgYC#GtIIHsw&oIt_{Tda6MWx(vMcA!`X3V4*eq`dJx^HdqJB
z9BMlXLZ)H_*`LB6rFtT9B&+Isx&+a-bB+tX?1|cgneydeY1nG7lnJ;DFxf#C4EdXg
z7?2XZyFvP&fn)^CXVqu6;#=T)GkeSek5t(Un7E+`*3ppw^1{js;R-;)o!J8sgXRpz
zrv47{fmNM~uoxqZ;IA=gXnc5y5QCHFs$8m}^6jF{spo9`GWzqZJ!}a~?W7U>&Bth4
zpU(Z%Qo*r!4;lpF2c<l(xFz|%rp7wR>YK4ai~i$q(MX_0ac?Ku{bcV&`?1q?OyQpS
zrHNYDvZm@)$v`^G>lGLFYjd6Q+2YIV`wm58qM}<Ys*3B|hbw!y2!H+oNbs@yt-M$K
zKmR=Ua}Vm6+eW&E$^I|)zB;JttqU6v1d9?ul#;S2LAn%CL`qTV5(%Zd(?$dY1f*3&
zL|UX%LO@ysM7q0M;(Y6O5UeY_-+X_4Gu|2Iy?5q4`|Q2;%4a=mpZS-4K?5}Tt4G4V
z;_ebQ)qZ_?FPFl*Hz$h99tWS!%%_+lQY(G*a;!5AM}e48zcA|T@`Nhka6oA(H@QpB
z=b>Bl<%{N)9p|oo@a^G<m3klAZ4k<9$d0u=Fm!S_Cw9ovk!R<A_A~pB*c3g$Y%Vk~
zxlPKjC;LO<?L9JtM{PNuHI|0pUG?B;d2QiM;&;8JsjY<Y4pyNRZD~Mnii<JBYpS^^
z>`7z7&+MAquS%5#d2AgtCHd7U<;llX*~1U34`-*O<7lM#2G#L#Mp^1P8D9@>XqrA*
zaVEAhLMHX&%ZQPsDXjSd8OIQ|j(5|aO7=}e-Ev-x_Tp%A8tMkDDciw3NBP6vY%$Jp
zlh&mi;v#|}`q6~e@fNxrckq&jr%!Q7sCCui&J@rr8s$V$^jAKgK9)!{Qqgxv=jE6=
z>#fexYgvQ>>CBO`Lw>@q9GY*W(8Oi)FA0hS$G=+SW#k*>R4q|Ep19|Dc7f)sXnF}|
zS=w!iIM27ewI+r*3M*p|=tmY_G<xn0Z7e2H-<MX5hnc2(m3oIh|6oc@UeYdnN+^y6
zc9=!#TZ<TM+s*rI$E`EB4%F4`e;44+W^eEbRrp@nHC86+z=4oJuGZ_v6T-B7g<lMM
zR1@qG>1w)N^DvBB7+d(X4l3_Zcf<;%k8Xdpc~6LN-a$>DG)@zf^Wx1;c((LhsZ2vR
zp1!&A*5NWLdv~|MN4hM*9k(9ge`0gInEx!1@>O_=URYhNkXG#r%Pv<g;g{hr&2Aph
z(vR{x<L{RpuN75jrDwkSq#}DdO7yD8*b0GY=Q+0jh<7JX{v7YF*qUh5u*zuaUA8dX
z$tq2Af{hIsW-v3ipyB0$fBpV5)hBp)|L0#asZ?q;KFGb>xmx{s$?S7JH<H;CS_d;p
zRd*Q_)>!2?yg2gdv((#HxlE~zb9wDv4}%w%iNdR$zMP?Gqde;^EK48BPmOg@*8%T6
zwihZyk}*g6p3jj>yAN{->9QS(;@K;zc;@8uBZstJ=rY_AZ=-f4^c88~uYIZV#6{(U
ziAv8ET<y!*RrJLqwY9Z@vO|;t*0EQL9Mv;uEe?&-@SnSQk$}YMi2P3T6N7h%Dkv<^
z)vJXBJqwuZ3)9Qz!{U2ve(5NC&(%}siwSWHDA=AB;pRl~T-hVL9p|KjDa}OpH34n8
z>#`VrPmXu1UD|g>r;tTtyMg9-zMM&k{9=>qM?DMDLW*O~?6F1@3@)V~y>vyc9y1b;
zT^_=hb*iWed?I}7!JF+4Ma6^IPmiU)*hWMD`c*Fd-q)Jgcd@PtTsYwxLogft+8HMZ
z%P?gxl^5Zd-*HS%jP11AeqOpwWZGgKB%Eg{$j{S6z97<kwnI_m=nykkfzVMuk1LIX
z7PKZzgQYec?5?s8u%AlE;YL%x)HraFQ2wN3vDn^FzB3NzChi`SKBe8eO~+f4@{(lF
z8_LHIMTRW5A7T?p(0IM`xt6Bcu{$Adij(^aYGm(ru=83zW5a5>YWkp8^rg)q3}U*V
z(?l=SF!z3VQ0gH1AP>9j-ZRS!_>ZyK7^1iBeJxweF1JHJd;0@s#iLb>OM0B=w205*
zKGylvOaH)^BJs1QmsgNj-@XRJ5A04lLBgvJH`(|#YwSdOZLDk;zYQ#nb8|52axe|C
z2^p1KHg_-xDxo@Nk8%0Fz`1yrYNHzj>5g5C`)n&Q)x+Pvrb&}=;SQ@}_?)<Hzxknj
zntK+<Wp6jP;_v2+t4O$|=4Xe)9NZq?T-P5FZL1yBP9?}y>0a9(ox#~r91uf96%yj%
z8FZgb=VX|)MVsEzrD=ncH+37MBFdylW9FU6bFz3AlU5AWqOmY@pOC0uf8neDXnRD0
zo>k)V8{1-8!V2v@9h8jN_jTQw5<^cURhH@;q^L7JTO#gwNyVa0L-ejZ(Z2S*wFB~m
z`-WU^+~0Awl)?R2QYaB`H7(AZ#>*R=*A2BrTN9typPv~q?4kP5V{n<k)+df7KU97c
zQ^oU1de_wXLRG4a`{n$SCcN&&nr$CT?#7Ddg?dv^4C2P7O20Xi@vv?F{9|`<(?mL1
z6-&{bS-dZ7*{ISU&kenjl6BMzIC*$k{#Gt=ngI7b=0Sx>N>1U_#_@BFmU{D<wx0wH
zsk*TI#!~d`ck<_mHC>C-DrUTL^@;58#gBp?y^D6|_9v%ZE;Vq<-S2#Zih{b4JP0?c
z&fTg-z}U%o?g}cGH(WY_v7M{3IXO9Rr1)iq-r+iNj=-=-<Ke0D$>qLfj_obCea_78
zmg7JC?D`H*ox`NX`|aq|=tpAuo5~E>Gt%oVS<m|Jn`sYgB5i13YkG3UH0eg*+*OOr
zyn7Tr_32^<jp-g9jW~PsEA#2@^Ea*c2s?h$%6peQI5JuqmP#EgbGWW<ntF9{tf#De
zsj<7Zemu`<e>`)StTKl@Qv~5-RxZ6Lyw;{^l%zbfJW*vX)|;hmVhuOL?}o@-Xi{w|
z+qdu`f^}|0(@w}<pY>~>gN5y^lhxumvWDRKWK+tA3pqx*^2Ti~MfU{5ow8LDU-UK6
z3%<3#zoSX;-0<5O4GXGD{g$fNgR1_@cUp(Ext|kAe+`uLwX!OV(VY{V7-bu5lgz^X
zT6Lb`Ic{aJhBCG;(YvhSjxH;g6XF%`QqR$z7477d8Z{YdB&nd7%g>cD=#PAScF8NW
z+3K8Bi+ES=!Q?SH1G-%6Ozrofx06_n_RQH5j7U(ANTeOC(7KtiW4x}%ax#5ehjfSa
ziBa;FR?23R^v08ek;cAiT{~MMWqj5AgJuW~thR9-mB_bD7Vj_8K#e|1=X6z^GCI#}
zRm(71zh9qu-VUQEbGf!;$G3Dlz56KB`$9%s<f`p?ai8bLy~h0|%m<V2To_d3^WV>F
zJAA}gcZMNiS|+REX>L|bz20X<{m|*hXJQ2eP1QUP3trfNK*sJ(CU00@l7=2zECH)J
zt5MpH9fHrPbR(WiO<;??wQH-s@iHX!A;GbcYN_eVOD{MlzHs|q+H=UgVBc7v(eA;)
z@rM0F?9XHHZoJ|iPZ@t>5G<2AbDsHJBi&IH+0jBrm4*eLJyr2`^ljUU$KD$r_16>_
znbaUFT9p{VxcMc4ivHnz;h|Y&eB%2>&CjKz^(FlUB|ROb#>_)Fzc49viw_NoA5s)|
znVxZYjr}w!qOsBA3z3eE?_<Jk{{DyWi=MaE&DW^Au>B>~;Re-Ld#=uya2lr%PaSA1
z?u@6iKHo@WC5k<L#zZa+o1*DM>^O7Ty`HoTmiSDT(L4?Oq!atp>Nwa4N<#Z_)4v&+
zNmIOYr#Z*!l6puh)or$Dhx-nX3rB7!ooVaap*T|%7L=~(O}l&9c-vr>Me12$Nf#Yg
z?Wo@DaK@3)*>4^$qQQhALP{%CBJBxgmS5h`%Cx18k|cXvER(<GyJR6p;cul<c5l=m
z);R8ID#y3{#4_V+Bno(?z6z~RM-{cXDbwS6vOJVW-^Z{maK<E#B<-A5<A48PdqQr(
z1vz0%ny65_nrgmSJW3O~*VL30Ckfh868O%$6E*adCtlAj+Ks~6#$I~wKN1H{&i`sq
zu(AJN6w0mecq+d0y!v}FJ88{1clDeM+_WU+c_W{g82fZ_4)P01O(7RwipZLBhOUkm
zvm}y;OUQ-ra16Rz<7i?sk?vY1)oF_>$)>K(4G+ypm;EGh`y{b9x3Vq!l*m3D$-8X^
zgE^D7J|TIbE6ZIEli!v22i@Gw&QwPrth#J%Wos_9@6dF{&Z?+uMazxctO@F>Eg|J!
zP{K}jwvLvTw&sg^R*sX@FJZhP&wW~9F?qN#!HM%>)&9T_w@#8?8}KOk*1$hE*k>-U
zT2&<!VtTP$U`(8xtx7P`hvr&RnnE~v<|B~~nQWfjFDLXy<Eoq%RSsrR-Xp4sTrA~b
zIdIWvz~zpt8rG`kOH;uR>D~9SjXZ5{tvXM3IUG3Ar(4b--N2`xnxZ!R_Ux_r;Wt!^
z0TLJKWFxM=n8U<*cU!MbzB2nLnJYF|j1?#Gj?)EoR9PLT@vwLLv@4V<^`Lgw3t;(6
z%c*(J6+9JlCZ6hXF`KSQe|}2nG47pl9Sv5o{*`M&c9(Hn5A8K`EwIe8mdbUilfI!I
zRXNe7;7u@-bM`7~fy`6;YC>|Cs%qPTG{MuE)gOGq4GvO#P`Ec)EF4GnWqL$?S#@5x
zm+3*F#lT(R`S2Ip)N|Qn9<iS^p-(EmdW|shaSXl6L40!iJ;!vdzxK!ta%a7&*AFp%
z%cndn*SJS}x>J?L)Z|b;wce40oH|3m2fn_KpnojRr}ETK;!0S6ZLltmV!ckBQw83k
zZW<}Eb0c4N>=G^Ac0%b%eZRnD*+DiL$52bg&ZnF_EF53xWFNlCE*^AjYwO(Y;8IKA
zZ)r(8MIwWR>GoA`c+_XfFxSb#SA0MHO#zY`s;28vL=Rbt>NWN(K3%HMuAME3tvsH^
zagEt#IkArZVz<=0${h!>x6ys7J$7W_B8f_mDQdP{;v}u{<C-hQVmYEg(FX?TO4<x&
zXi**~Iu&9^_x4Y|>^Ir&B;m8#7Rh$2;uQl0d(9h8*`)1X@$&g@Jh+IvC-Tj~<s1T{
z>FXKqNL`+#ihSDM6wr(_c`CD+l<<8~qxn)q)%IrXW4GDlNU9l)sZ3HH(6D5_(OUVq
zJU&a?%yochudT;yJU5+i@Vll@C0y#{<?_BOZ_f1HH6@J3FS#nQ67Q5yu<c$n8{S#n
zr-IdXuMaY*Y@e&J$uD9%E$$*kOT==z)2GOGp44nF@zjeiT!wpA8SuXvj<oCYYaTVt
z&MPOgUhsHIOlB=!oqK1*edVki$$ky{L*)2*sfuorA6|S(c$Mze)_E37#k*_!qa0sW
z@saB7vXR8DE<qev(Jp(m_Y0F<5Zu$ibb<1m|7769Ls#)0J&ld;#1_+@tsEIWyiMd-
z74_^3%J<TA14r>X_s`|?-;BB=(&EUr|7AuDV^PVCPgcGkq$J};@(ZRLSJY1J+>z&Z
zyq2q!eEvjeN^>SXR@)w(%M<=@w(WIFIpFvqW}B}g`L<Lq#UzI9nK9D^eH1iIDdJ6$
zIkFSxxpaao%<re_Q&ki~53|$tpC-A?j44cgUO49o@y=Z>xbBkvm&L2C_JwahF!5p!
zc6Q2kAABs2Fe^gs+A{IkC_)S`Iv2Oeix=%L*c0P2c@pTAnBfKS$Ic)5_;COE{fn#n
zKMyO?$fiq|mEKg8uurWO4R+3^I)*c=)$`$Y$MuVnlKZ?1&t9R^l}-_#?jB*r;0?TU
zgKVIk`)qD$v!+<&+cVUauM7OI@{aC2>SMIJr=BR_nwy)SbL@Z{_KjK$Ke6RH>H8d|
z9h&rAd!_H6Q9a*P%NKkww(nI^+P=PKGgapKr@M}jPKp}Mdza%~8b4Rq(t4xY@|02I
zaV3YS9p^nx9L(STdI_ry*Yx}>N^RUUgZ<fYZek+d`*%u>(zYjap4S^UaKW5DP=ooY
z?|@j#=b{rsl<X{T43j9#bb>uth&sp{-YDNX-l43UJ#triKjsC-qMON#6RBfvI5@Xv
z0$(R{^zS$r@Z9<C4f)e%E_bf5%-o@(Idka^ZvST?<F4+AH*I^T9q;<Cm>b=~Jf!g1
z3s2XXspBnUp_Bv9l+O?sWt1y%AtyOs|7)uA+Aan-Dtauh87m21?-&g|yqnh+yLyf?
zUnR0^{)xSHb_HuGKU;bG@ObVH1;%GQ$IJ4h6d19y>-SmW6BHcXrTq+lUWqmQR`s%y
zZt<Z@s`dlr7pZO~94_b6^AnG%W;B$}ueRJhN{EdqoqO(J>22I20`CWDqKRs<WFC(b
zzBL?bp*bY+l8pFZc8AK;ftHv>ZnFTJ7g&*93AO?F?R&#8ZanCj&|ZEQfKl16K6~l$
zJ;5*6C`iv@63|r?koDp#AM3ReESYAUpniC1a36t0f=kUaD`L4zL*64xf?3xMq!a9w
zcICAi4n&x0Hat3gfB-Y=NC95_j*{1B8r}8j?%x=WEv9mroy=+IePB&JCgxsV=+2RQ
zpl^UmQ-JM@%fSK#Rrho3rw;V-iIb9Jh|_nQaVRn9_yhzSx~tHoDep+#(^=G+$lc6j
zhP8BF;~0+l0op8Ub~}xfvI9Pe6NN(f<Fn%XXT`tdUbT*qP`$jgYk)dcu}Nfi#<oDR
z`mP)g{LV~QgZ3x4X9<GMNOk)?oCxtLG?h!K?(9))e9pmmm1n5p?x_^J?8agqf%Ht$
zh7k-QH)b2F%0$CmVxkWXk1x2?5Sob|$Y5APDN$evvp-;FdDYO?L9K971Z#PhY3PE}
ze`XFpLk`~|1umet{;$-AyI0zc>(~gLs?|G7M3q#+of+B=-Z+YNg7NDFwt!Z27CC2`
zzt;KQgp#DtPyPj0t_p=$Ter_IvS16++-f^(Tz(@7`^Dpe3$kt>nKg-bu$<_`@;f-g
zx!sdjTK15*gL1Z_vm(GgQS9Q;G^R;Pt#R45_STmjR1UY76Sy<zmMo^{i=J1miYl)s
zA@O$6P;+O;(kH~md!2vbs69ss#%rS&{dgSSt|pjbulR}YWL0J>^>;ZI)=H5%rsiej
zhX{t&MLAFLlS^OJCBBUR;L=^3O2L^(TbseXrPoJ~Dr<M}iIYE0mn3nUmw6R$wOUbf
zm47<sfK;QMZrbed^?Ghq{h<y@z2xSr*Bh5szQ_w(7^W>@V`w^DiZ7OZPt@{c?|p9V
z-N&=|vWwY*KKL6`J9J$Wev~;mce!5PgUV8P!Li;b`<fC*P+&2eTtx1~E9{sMT*1Kn
zPnyqpHI~G*uOAQavNNGgvia;O;MH8U+kSj_47U|!QLy`c@}63+uVPCdtK{Y5xW@{c
z%Aau`%ASsylblubF*F`hbnH7CwKsT29$!bmgO4q;<pOi!Pw2ZR{raWc1Zai0y&o%R
zkIZyUYd^=oOWzR^rj`-(`2Fq$QHKY^@vZx#F1cW(4x{d2`Dki3FVIg4x!^u@(eQ6J
zw?C`#v5$B89CpSN|LZ}Myf*|ceoUm~6Rv)04uzd=$kLIiou;Kv^s6|AP~K!__i&z*
z%?frGjh@LTI%>1$(F+Axt4T`yGlW#wUGMQN*!nvKL!|alO4fyi3T7;hk6CBFuOq{H
z<!us&D`u#hb$(~xAp>=*YXc3}_kH4QJ%<}b9*>%BTHeE@uKAx&ft>t#ns5z2Qvd)>
z`?|$W*2|Wf7HC9G-0+&N_D)v$n_3n(Xil;tsM>k$o4PkFzW;~FO$+mj+Gb~r4NZ)V
zw2dricy_X$F*YzZyKHjpIv{eQ+EzEOYs-p>z?xFm%=KwbaB?D#^2%;NavU5dPVk^}
z@2y`S8EKVz&^Vi$KY0m$5bWW@fkjtIC2{u`chir&e0_&BgM^8G=N+2o9_}?(+lcm`
zCb#e=)h5}2BQ7Iy+UMdE?Q1hVpF>`pWgg?|Qpr$HFX|03YSpPgWveH&@>=k98|S)n
zI|xxP_X}fU{QKvh1pZ0jp9KC%;QucPJmnTfEl&%RCa62F%&M<Od!go07<dbv_o-K>
zxjj*0?bt^MU0v`<zW>cgx!iK*#zyr9s5?+R>nNdiUN*#bLaof_?x4Xc6T-kS;l8G{
z^GTpTmJ}@^q3ilz?v9HZ+V~I5xiKG{A{xNNaLICe|BmDV2QyBPC=CtP+3$ajd!OZh
zq#*1!!h(<1Y{S5?-h2DzQ9@m%ohRK+ox*<j{qGy&DOW#UnyOZ8_~<iS42&_x?RBy=
zRqse{K#BVEUa^jwp9;y^fqawz1EWQ1_p8WLAvcfK&|_f|9r^w@5R8E%*XP1xy%+f?
z6$VCn@P0`NSDrfAXX4mcv=_hs)o^3(jIF>l9b?3Rj|w6uJ@M2Z%M@1h#lseTw*F^~
zMWPV6t*b=}13qepfze^4$;OP+1uHTV;lb&@|AAZ+*>Q_9f-&HuGZ+~8twkD2JB$7a
z?w{cPY1}_k#s25x{so7B!Qo$2{g;m5LC*OvkNcMZ|EnC<s)v7-!@rX2U#<ERAn`9a
z{0k2Mg2TVy@DGps2W0+3fd3z%^HU`wBO}8Jd8^c@(Mf>!`vP^<Y<vA8#JpBz_96g0
z7L(@hXgLCl|H#Qma|nbMG)g=v8&$1ar+UkyR{c1tc)I<SjcmUSMmc{QJRH7@&=teo
z?{^3v!Rj}#dDpWeq1J5~(^dQd%*tG|B_UQ{Z_c;bkBd<|_S9k05^78){SKc)gW(5e
zlz7B(wqAa|#k~4cq(sze_P*nC0?QMH=H})?bTn9)q<eqw3?PU40r|UE2OH<Yx#gu-
zIJsL)Zs*y6Tru3;w%BuDMt_)chFR}thxD0G675b)hT$a0=WcKQx#FJ1{_yayu&^NE
z{xJKo_qTB!6%a{dZ6RroZMgUqjqN8VmF3}*yy?Uo)Zk;1d|RBYFzI?pd`4eT^j*`X
z(X^!k7d-5z7F&GP{iX3dXT3v$lSO-l*g=4dL7N}xY{$xcyLzs};;l9*6^q)lpbyj0
zm_&GN8NWACkZ#G3?g$f>b*p9L=YrtCz{&sR|1X(Y4GpZ%mt_Pot@p;r#Z9IB4__jg
zSwdxRFeGZ*?ZNF}f{5(M<|6YQOKmpdzh=FcKw!d!0RO=$mmS)__p_jJ^`lXxVdl8H
z*e@eE`}yI%<I-PRi#)j729mdd-ij#x)*V9HKtI|(PkF|3=G!VND!$DQ2e7IOS!#kA
z*yai&B7d^E$lWGdt=mW2^PHPg4fPvh72j%=eQxjnl62jP61ic~&2Ly*N6d@LajNVv
zw^55t6>q;}C<Q*X(8xR%99CRw)5CVq&x~b_kHI&Meod*XtGj$S@XUSuMI6LTi8ntk
zk9b8@RqjHU`#kLZs=O+vB{9F<%xiU0<h<USpEs{jH{3DRetyj3Aa|H$gW~1GIojp^
zm6er?asR_NFDO$8*qHQJg$_5w1tX?M5hI^4m$gllCjZXcO<jvgfdgVTbED0fmc#Y(
z>VlksW7*Tvd1C*ABJCP2^`^_R9&bxG?GDrp;~l@hyB{w2&D9u0N$9);y2y6^58CDW
z>)z*=?$*c18@Fa!w&yy6i(}1Oo^s<2;#tLAoT-yp(Lqe&`Q~<45z*At<hV3v*qS-}
zCMG^;KKU-V%vj|#BM=d_!^C@IluUnLUwnGk!@VZlro(Y*-}WKuY_|FRE({AT_V~7D
zSsS;coA!S6n|6vbDPssQ`1<Z<b(mng!*pfd5+;@7Of4)rITVxdP4$0S)S2I;BOm17
z7^loFC@7dmwo@X2MKxKcGS#qoZoFNXn|t1Ql|DH+dDn`(bc;oulv6ii0XH_cfYm1u
z_Uy*na+FgIGC(E!SGcYfKimhy0O5gwixY(es1bs>f?cdkU|F4GjGicscE|ZpTnxCc
zvdz6ghu6YFx0f|o$F;(He#hmjmc$FR15{IQ*1Zz6p93LqJNMqtUK-0Hn@I@yl5LwD
z@V_i-O~GTL0czL?t$PS^{qx(eHNDja%WncPtgbBe4@ZTDhN@;+jou2A7MQ6yqlWW8
zco9@@OhWEgP*HzoWtQxT0^2-(p3!%84GHSP930cY3t%ZCaGn{c<<xIrYH2O<VUVyI
z{o3n|v9+!52Y-^S-x$B=AXkXfx4~&ABa_A60OK`PfcT(1AY4kO?qFIc#m(F@STr*;
zlUyw{{2p1k>-U!WuTHZIpMob5lo#?Ar!pUgX?yiVgQmCLsIT3qsg<Sq@DAb6X%LJ%
zZWDx=eLT9$g3t5=xo-B{SA8l1L?Z5+`k6xm<*a!v-vyH1amzOED9_SJGPeP|o!=N=
zEwnT;zH&Wxwm}*GYb9A21y?bi-tDQMkTW-rN}7i71P*=L9|LUg^yW55c^{uLC@@el
z&uInnrv4R`uBmd?_P5&QZ8`S3y1FuNgl8pKb1XnzP~Lg=nqW@B+{RAei@Z<?o)o-q
z_3GcWqxuuTL<{c|*vlg-Px7<M=Q}4^qfWeSoPCbs$V$$u;<a}|-6>R@$QOx}ey$}M
zxx1JR`}Xxj9sMJLdV{Zt4!^f}dFim`e&QE(SGoxkMu}fscN;zLa`YUZ-nCAb-4s>h
z?e<kog1XjLu_<X+jZuTtsg<o-#MfghcqxSXLs}h$sZg!sQj8hnw&AV)0gX$&w{qO@
zW%^w5A>3VCV`?P61AonVz(eHu<MZ^INeb_>h;wps%Cz1pFd<Q4*D{4CCBGazvehHA
z+LJjPubk)PsB5leVq#M6&&&<}BOJ9DF1#{R2mgawj3G*gaIlDK9y`p2x_MY8%%T#+
zo_NHD9T&T*4b2wWg@G~0$R`lE4GV(xoi&0-fzL@!&Z}{9$xS04JPy*K>WIOJ%Cy-1
zE@+#XnN6PSV0D^w=Q!5^cbR9_`*AD^aeplDgp)8qdz|Qy6E&fE92S@u`E8>251XIr
zfQ*K1jkBU~K@Cz-vy2ewD1K*|LrQdXbQvn~axu)VoUX}U`G><DrXJn5h4`0c4>zW;
zfZm%iI*rG{z_8y-8CHSm3W;sO8foQPl75Pk$GM1$N9c*&O05~XCT-XQnMmDPb7o`{
zi55~49OI&*qLxm#I-^$NX<>t;JX>w>rF58p;>PjS0~EkzeEITaVPT=C=OtdqeEv3q
zkjWY<r-f^c#E{MI3PcV#f+o0RVqj<(zbN$Or7?jD;OQEfhl!<9Cs~Chs%|hhH@CEr
z;hdP*^rAzKT6l2qUe~<72<nv^^L4|cqmCm<B`TMYL{|YhWAYk12n)mIPc@lHQHrz6
z39Ihk>n7ThMZ9g>Hu%M{E62d2N?=>L<y&P5a&a}Ru;Jq3hTHa$bCV+5^xbNk3SyoV
z<kZyE2M;Qz=w0m@CPq>LRL^3yu+wrno<KQXNMK;bVqb6;^+*t>zM9=2LmC;f@Z+r(
zJ_+Z)ax79d+D3qCF|%l3;D%6mLgVnLl`?kN8Ek23X*K)NlvV;;NYuc8oBKk1PORBV
zNkwY(Qif*jgu?TN-o((h`|YQokc^a#l)X7W*6R5!POJmGM4D0S*OiI{$nCF}c=GBZ
zqCK*OXibPHg&b@g92^V`3~X$8rYzlNha0P%r<VJ}c}xo~i5+xh1xp;NkI4#0K@w~8
zZRV!f9te4N(U(4lW<#lBU?g(ac9u~PLJ379xyJfp$<|3$Jmx1GeMNxE%S!%5RMcVb
z^F7mxb~$ZEX_T{z?)l?R^I2!koN3MT<R8yAH!`Zai3ESw2`zlQkQ_8=iwpx}M9Km!
znRg_3YdotvyRKVVrK%<LPcM!(r-`jbG2FX%@6(_~hDuwObwWh029ZLXTy^^5l$(r0
z!T+}6^ia+7GkzBse9Lwd5QycnBv}qO7;q`a$;pHZjlO2RfbC!2%i$yib?vilxR4-<
zbRb8yM-#?aFfhIZ51{u9RJHFMul~ZKBGDuTNk(5tPin_5R5)wUdLJR#33l6R#6RRt
z<%;CH<auu=Crx#ca>Bg|F`-SQ`NC}IbF@+%Y=kAF$EOtat_b47+Nslw!z0u#5`8gB
z8wU?LrVjtKFSnYEfHz4`-@==8GV|F?Xhh0_w}Wi&frF|h=wy>Ys#084MFr0g&Oz5W
zx#78H<MGw2_yh!~I^h*xUqk|Dd$y8*j;7`d1tvz{G6g2m(Rfi~W8?1<DCA}J#|sGx
z%Iv&%Fw;3Xn9J~MhIv#WvTAYH7OUzS+u3Ex#q^qGFe|5u!qr2P;x8Rl-a7^P4&|)o
zqkJel4BK;ameK_H`0RR0iTVi;=@x7u-Kv3n*KKG8U{cQ&02}c3_Euq$kEz$u(V>{t
z^vc&s&M@u%FxI{>+o&$g$=Q1x1Gam&e~azr)wZ;#MARN3bqD#9x}VKDUfbNv4@m*_
zD&&#yQ-q|h(~W(PmN?T|t(0Txr<ftv>A;%aLjTn{*Q3=dsbE@&R#3Q9A<?QADlkvG
z9bz4OWdv#)8yi>HE80)AV?hjHVcQ}GJWVt2uR_PDrKKf0^??24<Ym2PDN2S`Rt=YM
ziCxuKhUBXkbT0@^2Uai8LQpk1WV?A#C63L=$T$~q=?Dq2TM>D9dY<BXV247n@X~#9
z*YL#(-Ar|#MdkTkR_7*bpoY5a`tZv|@z|q`#g~u}-@bqvT11Vb2xxZaeWU&uvYil*
z2#*G9N2QnZa<@|49N<b(kPLY)%SgnkX3edPJEPi{E88n7D|z*ZY@npJ%42hMbcBNZ
zNy0Yp;bb2U?s<<|%0o4Gg3M206Mx3RaN+p2Sna$@Eg>IpbI=(B1LWpXJc9NNy@KPm
zGS1G<5F-NAofb=kS0{DDZ2O{wSE3_GUGruj4Oxcobe=EgIC*krdAfS(90mr93yVx2
z1wMug9{p={+5xtRf#-pT?ZPmkE4<VMx!g=swbL9Jt-FJ~risZ!>$sD3^c6_H6JS;2
z9(B~Lx^ed8hcWA$P}&(0VdUe0<AL=^T+v=<fk37Y1LGFt1sa2k0nUp7#?DL42;_xi
zF%ZTfj|2#Aw$j*XT1@zxH>{}(?guqzjG7}y0XXpa$v6&1NA{E!J_f>&1!r6J3HFnW
z%Ce|Tt2xN|35A1~pSRMSw=#tCDqKd@24WmqBLg#$Lq6;TEMZg!!RoWTJYg^yC>y@z
zE{!dXS?4YJhA$Uga?j@;?s$J2O2|>ClK8B#%)mf$^*tcfs&MWDw@{7FD*=U=A`|1N
z)p2yhsDUd;o~a98oe@J{ffokWL^m~RK^}Fc>6b6pi<65%`+s=p1u~%RxVV?2uYlZA
z^aBvN)fN>MK|!6Vo)K)6c~eK{L)tb_t?3?*dmVWT#jF(ggtx+%dk*Iva78Uwqedh#
zh|VXX0}aHPI~5yoM(YHhSx+e>L68c4^gI4tpv#9rMJBsTC{yqr)d1?m$;MW*^3Abh
z8}*H}=3E+xxFG~vL|i7?|4t9o;@tD6`-jz0cHk`MV{D+hT)c&tn$Q-h!Q1@x>?k37
z>f_H6N!IlOiI6d3<xSjW4Vlytf+{eV4mK&eMZ9@@cW|P!u)Dka`t|FDg@xV;5(cfA
zoWYe<&Z};F4^yPzWls<-K!AO*v~P59&~~O)%ux#B%>(}}QlniLlAD<PE<|5BM&%_2
z?y#=T_dNa1yCM;_AU;neY_%f1kUUrC88PS2D<AN8@BTb&*-io~L%T)1DLrew79Xqo
z+vif#&i<Yi=yZy90I<!R*79j(4kDQ_m`N@ZB8cK&JkB3!%wzXn(BYqQ1Jfq$0PW-^
zy%akn6RIqDl;2(AmDf!<KH%ibZQKS0E=xNZvSV~Q^!FU06o>1?iAzkRzGYf6GBQy2
zOjAGoE(((>riXX}HWMBD<W*`{G}WPe0Bmm74^XX+xGkbX57yJLu=skt43#k-2^Kf-
zhfK<;Org>N<-JfK^XieRVsiyoLW*ad*E4R{n6=zX0#eAf+(KtVgmLjHUSByDD5(RG
z2&luOM~}1+De!S{biY>oOtdhPT+O1K8oPAYxqYdPo144e9f8u6w~)g59!5sShsB!g
zfscwcAyJx{n`0z)6M0^9<75bj#8E87S#eQ$S?8gsS10Z}nu3nm9N!|RaRcbUB2KQn
z*K9o^D5V_pExU3I>0<Ed`Rie?)~m17lcSggCi1uC@*_6Ou|<4rdl<@Z4at`zW3t2I
z6jIZiJ7<=m5+@fN8fy9Z4x#IOJ9vL*k|c##+31CDh}9Zxu}$0LzP`Q#9_OB4JWo%Y
zq`>r5)vc|qO^aQTiB#QT@~oJY3z1E=;JgKjo7k%aH8#Fl#SOCqF6LY0o~4RGK|yci
zVww!B>Rw5Cqi(tZQlrHlcp=hk<jV;$sZydZdMRT7P>i9}<YE;jA#Ak=APaMsZ(#{z
zNY(PU9Dxpt|5a2JDD%8CKa^4S&^mGOLT>Bw7GhXz|49tq;}?p$C)Z>E<(8NS-m(Xv
zMh>u=6)MUC;38@slYIFdu-)V>lCM(MPyySwN^v^LO@I%dgQOmCDoI9&6pV)tQ*zQy
zJhB3)4*=saR4fvl7M)Xbr4UtM*g_S8Ds?hcq@(H}oj~ls)3a}SQ7Mj#pMUr@k*o2?
zb&vp6WRx>2he}6O!G8-G6n@rAS=R&*!S3C=U&=<yM$dd3PT<!Q?NFZV@s*jIpSMPl
zSwr2hBm(uK3D*{D-iF}iJvlj<l$3NeM(!+}->5ZnQ6>zv4E_pM)wq&7my$3Grh;mm
zkCt)#XpI2R9FPobR($O~{geOBWjVpe7gXIpJZqQLYW4X+{&c4+E)`FAUeSGOr!S4j
zdShGcG5|F7YC_}m;xd5$`xfJiIyy9t`5p7S0Qm*7;;s%fTHx#l$)lIx+*4zuvRB<A
z9<rRcPe4tzR@i5lX(NR_V7TbQ-n8pZ#5ruW^P^VwtLp9KWI>R{k*l+SzixZ5dFbl+
zRAay;5h;7j?_o*z#>KA{v!G3yf#0zLls*DP18U`v2kO7zv4sH05&(BcBO7VP?NlIu
znwpDzp)3LA_H!eqAOIXqGqb6$`U#_}Ue;@H!8C%qcQ|hm2UINZDfvzc2?=p=aR~@e
z01D9%r<~kmkfJ2$`0X9YfaMfGWuYkRuPfA3oQ*2x;2heK#-5YM7IM%$fsR{XNI%u}
zBs=e(i>QUVo|D$Zb*YoUIUTfZ3>6RQ>$zqbL=c=?M3`Mo(U+t2Qj}N%vY>T}fN%f}
z3|eHWu#7jSRbQgRB9q4kyr5k*)j)CHxf_~d{Vgpa;|D+=^0r70(lICnY!YRo!3#o9
zLljKGusIc7->(?g!orr=RjA`a-Rj{P&|!9LM0UG{6Ue`N{qN;%p^sa>S%$+0=T!tS
zNH_>PL}PfH;9F#k>mdtz4}S)21s@_L8?;3661o>t+4c~%^B^~c99!V;8HUXzXb?O@
zAG^0a0AP(1?Cd_3Ix+P_6%~&n?nP3%j#@2&>l<3-c%G1uAQ!Xv2^`B7;l<{Vl%yny
zd$ATfxqI=8R=<JK(NNTvMFQu}yJR68*!wo{0y<3_n+06q1=n#(Lf2CrTbMz0T6A=@
zUJCeD$e}0}!~GR<>?|35Ay0(Hfca-a1Gu7)3<NPirGI&Mv#u(X-*FZ$N~CY|65J+i
zq$Ohwi(Fk@{XK5eVD(e-S=;~&uGQHBJUJDwSy37Q1~gmYkwihk!TKi;+!Z<GeZELD
z>4r6foxME-za)j0r0e%1wi7MnL9z^S#|kBG2EPO>BPEpqbsKiU7Mi-yASESrHKyKx
z%l-pddA~+w5NDv|E7@pCX=y$^Lf2M6%Hb-lQPF12hG{8E%+|>CAU4((tJ+j;AkhC&
z-QUI&u$K{-6b0C5KAf$|pn(z@aCF7kqQ1{P3RPI`4KRFBTB5~xQ2{C$M&Dx1Ntg_^
z`b>jW-6Ws)tzLntVFkAe%H*4k=$(pmZ*g5*n?cba*ZpqQ4t*T0&Ql<DR_DcXBWO<(
z;bCu)#3#A~6v+}$H)sKegm4fM5wYlytCmoWzypM?-b<xoLS`TGOsAnsm$e(2M<qn}
zDgKV?=d0I!M`NieDLtTi*U+HxyWn>r654IZlKu59EiLwt36BXs0eCWZI&?f8dQGq}
z-sb5=z9*fS=OZsKFAtA6l?-;Ms{xCwMBrFzVW*b3ZA3QjfgFJJG+Uid#R1fT$eZC5
zHs4+v>`guFZUL;|N}cE{7b0aRZ4ihHif0EQ^@PUe(bV$ov=c9$hJ-A~wSS8n6OM)C
z7TT!ea-gvaQ=3}uehqBE3eoZF_u3QC=v~`sO%8oY)~${VQ2V-~+c>^lH68+}axo-D
ziSIU5Ft!jp0f&W&weovn!US&ZAeX^YB#6&X>=*68I(gzm5;VNVgiDMv2VViSV2#up
z#GCqqd<JUQpK5P_C86}9^)Bt7p0&$qHOi8v!6JJb6=tWWq48|VD}}(JW1CZuD+r-r
zQz4|&)0wP8(rG!PhDg%={P}YWK2??g0OyByy5gyb=1zJTLv>??BC-Zoo<Ri;-2d37
zy1){NCtxvf5e?3+)!-^tP1<O1HcZIzYb7R8`gzCBUFtlbwD+m(O7|db1o};-YO+JL
zV-sXLi;V^=z(vNPYp?Ed?fgK5S0+8OO{&ee`3X<Ja_9=W;0jPN6Yp_W(u6+rM(AdM
zeHOiI9wl&^xZUc!8;W`X=+})QIXuI!aNqC6xBGD{0xAK8=qv1j6+ok*RS)=*YN*T=
z4339`gJW#_gM0=9E2}wRmJP1bnL@Y$_Lccn$l;k7HZ=tm5+a`U+8a7&(3?n0<Muv}
z&S5%*T<~@u#`2;TUiGxjj#!c@t%Tn+`kGR1h)7rKkIWPLd>iddl|&*QK|$3RX!f-L
zFpY+0<>EeSvG<zUqE=VpF(2f!(M<rHAg^M2c0Sx+9xD6V2r<?DkReFi{v+)IBntR&
z=$1nMT~gPj`SApF0lqlZyH<erm}NlW0R?z7E_m$V%D#JhAU^guG-kC~4t+jqy&W-#
zKjJw|qi^&+O%926Yz9#;e841H&<}H*YvLG-&KY+c4@R9H{ffASkv}W~`hOc;PXxyK
ztqKK#itc(9s(z9NE6ZkivZQ;`W}f;DQrfE`X_9;Brb5_7?FvnGC3N+*va<5C5TN>6
z!V&FIwCnK}t6J!HqO$40<VIQwFaB@{Kc9c1VEz0gJKaRP9?6)u$Jz*lNiw<@cOfZ!
zlnb5!H@dW^PMsFc$(!;^NEZN;AZ7i-B;cktnqU2#STaAiRFRHyV|?ZJj`Am{U-Q$(
zfau5D9riUXcb0h`*(q)B@3HZhojQni0Ky9C@kSI|ZzNY%RejrzNwojDT#RK2rL_g>
zP_}0ALqNcn;t9dXic5d5>*0k4Y<MgI-A~iFg5UvvDL$7PFo<kcC1qp|`<9If`vTo&
z{^Rk4^s|TrC^waW@d+$1DnULWp)knk489OT{Q+dD4<A1C2|JsaeG0&V(ykj2M&K=K
z&xcndpt|ce+>m{DC_oFmH`M}jcj(5fH5UJf=8K-N;m*;voFxFi23M260k+O>JM}X8
zRt<C(Fa1$7!2=E(t#z9@L=xx@>-U&$V#2l_(-+lX;aWj>D&x3({K#s=QiHNFkZUaE
z5_{+fHOl^IS^l!{2RwldB$=GLChYclhZPCwjo=k<ddm)nkBpCxuU=mUCY{yA#4E7s
z;N<VM_|K~@oFn@3w}RAhmkXXs;zoV%y^7TLr{Bvy1eaT}nKpDJoJ0IUj#_p9G;%f2
zYQKH^_FAjGDY>L<8?8yw0=)O+{2_dTKzKRv_N}v&uB#INmDT)g57VR8-)07(4H1lx
zJJ-PUFPWD5T)~vo)WUE-&{gfwoJsO>j9v2j1}GTNGb>KNf8l-##XyMnXF>YIGv0}9
zGC)9Wq?HWL1YlHXxVnkHxOn8`0S~LJ(bo`sURR>%p~uhPO})<!=*%Jem`IT=Gj5{5
zc$I%GMq9HQFM$}2Zq)mv3TyQ~th;l|b&G)8d))5=qc1rXm8XMh*V}TIa-HOQz1f+;
zI!~x&CF*;6dQuwInpe0B5Hj0nbAPNx=d?POE88D|{YU7@|MVM{3~0%BA?)ej{UHC;
zNNd(SeAV#kuC5^ccwVcKr^&Z)s+@QHM?vzF(~+oNuS$Q9>r3H#aYZ$rBFu0|Pc2c}
zrog2Awo5cIf05Po)YDC3gGL#TNyn+58_rr<^ud*KG4+tb0kI%IUl7PuK<*h{b?>g?
z9}fu*HkV7(*mBvV;HOW2?Xw1Npy@7nJ@Qag*$A|)%$EWgYxtSQ`>Cx&s(Erfl=oEz
zBXGn<mvx{v0=cSeZ=}d?KZ+Fjw{P7o+e(`5rBJj2D9z}r2oT0kokXN-2V6yD%@7Y6
zrzzi7=L{%sE<|=c)4<$vpICskH{Li2Jrr8+2OAi=Z<%cLJ7`6qpPO0GNogJhh%;ts
zJO~&$N^yO$8qX)U5<tP<NKr_e8Qrj(h`W)Z1jIV~`F4ItP=y!X6MzvMp(VuolX~;R
zWSQUz;J*d&<LEbeBG@@OEerQ>Tc9YR)PDuYJKu2y9?{KBHhn-TP7ZqQNFb-3QgvGo
z<IrV=2B~O=t=T}$b0CaZqx?>;Q3G^wcHQ}}lKdwNAS*BPJq)j&a|O4in_h=I`zege
zVq+orBtii0Hhzat5B2muD!vz?*}_(FfLa1-z!y3;JnRmH3k!>r;nj`9vl$t@NK2bF
z@mxd_^ccqS-WX+#rUKa^w&TU#&5f!93oha(PT2PC+ktiyeZ>x-hqE*;c-iS;HeGHU
zEZZS);^L5nZgJ}VY8WN>P*PIz;e%Y9(!RCW-6w3jIt^(t3nI*d$$EGpo7CMB#XL7C
z<NWpX`(lR8kA12Ku$}*|KJl;#U3b+X9r3ZeD$S%X0l6~_uuJbPcIgdgp0yJUuU5CL
zzlxyd`1pX7fZE&78zF)uyX)o9L)Yq@RosmXvb4Y+X*b_vzOw1UMCTf24lG@yXW+E*
z4UL)aJuLm;6b4axIxq@y=5xkJSA&gi+!)|*U-V4A<t~Dab?pz}{70~MHUBGgjNFb&
zHn2>PB6ciqwjo}12}qy($I~aa(03jQ@EPdy8@J~Qg2Mk)ae^6%cKE<gw=B-3<>f4~
zz6UxB@L>6mLyk@6vPpJjVGTyICV<Y)PJqi&jauQ}05Tk`i$d?!XKtj)d@wS=3UwFS
zmC_+Rov}#M#Q2Yb3y5IfDS;CQ*CE5Ey1Koi@kPtH>pi}&qp8U4OSJ7`%}zwB`8;al
zwQ61$2S9Fwn(%8k;wPc1^Q~8%y366kLbyM{#WIL#)(Fggz6?acR%dx9XwNI)9lVQ>
zM(&htk^tfVrQD2?2`6>Mpk>o)U--ZcP+u0REqgq7iGl?3>*<oeFL*~|{pcHLsT4i&
zU6JRtHxP9yWp$i%R~P6`s731c(||Pa$ZQfj@jkA{<KIzYxLkK+J>w_U{aGlX?=5*s
z^!DCwvOU605MtzFEE<i{5VvlHYUN{Nn}A^gv<`W1l1N|M@kb=yU_7DE9L&zoZ(q)w
zW`VlOFCefCbQNAiFKais>uYX-n^cpCvGeMdcu_l-@VC{jG>$>Q21>joqCzwxn<ysJ
z9)9aFrhsyQD%L108>}x~T}0XGiE#<2vb!E0&qVFa{WcT>70nQ$?SFCxepFO<iuH|*
z;@5G}b%<OFfksUM%0{?GW<n=}D}e-Wf$~_6x6T><9DTWJrzp)S`c1B0!Q$1cS8p}~
z5kM?ILBkJ`t6!W$I`Hbb*-OU#;GmE(5M<SS<v;qQKjK?~CNgsYZEXv_*7kO4aP+06
z=UD<E;DK+Ut<eTVjqFpmmoSxANfzqWYR!G2o@a3W|6mLJuy!V`UvS}~{QOhU&;x#Q
z-tyb;PxF!0!k`PWdvPYJYqh0cs2{!N+08EM6mYp?(LL?LdwZbn2acB&@Ku4e3dQ6l
z_ftS$%E#$iU4jbgpa7}6WAAZw(~D50w(Z;`$tU&P1QkX{kZ1N|8Bt5d(>4ODamJ42
zXam*VWS=TP2>6AW^gRxLXC`6Lnsmy)r?hUVT?tqMKr0vS39bA6f8bz0DU^E`j0~+Y
zaiI2m1R2B1x7q1%0n{R>y!B^9qgZHR>T2AfpFa!rLLE531kMX*&<>Ph6THgg5VR;{
zqn|*YIcf6~D4y1AD8`Hg5L!!!hQ|}K-!8GPfMvj}ArE4x#(#2rek6DOKx2H<IK1ev
z`pB53`u%!07v0Q6?{p6|c5{1tBr>`4Enz@(sz)5XQK?O8-A)1CzaYWFNIMsC%qD}j
zFm5LwuhNu*3vKs?rY1%7vLl-mO@u%V_=BIk(Z@_d$E9Xp8KI{3_I5#nvD_siTqGuK
zw&X-ZD4!*Cw0ox){|g6xjX0Iv9}!v~{N}oa48(yBMEC!La`2;iu4;OI<~}sU|GQ*V
zLL!}p<LY#{@O3U^z!e{{`KC?qh|bEO@0HTjBh*s0al<ds7%k1mw$w{sN-_@vX?Og(
zmtY@ykDGxqy4~QzpdA%Zu4-1sfWY6ttB9-{G|R{~ZWlsp6HVT$RwHO?SXR5efB)Wj
z;l8tp92jWkCXwX1JE$D6c7SsOcove37&iNF^e>TXf6Stzt81LQ*f(}FNvAT1lHUrg
zp?~rNew4}+d4G+_Ji6#eV(;tg+o)!5IREJjP^`yVvqJYll;{3~n*76!ij=I!<zEO#
zPs)3dq)QLIDmQf7b4K2lXUdm+@3x~AlzWq?x1$9}&1kSdG1tKsEOZ0#18O#URomh9
z3Y_)vL1?uDh=BI_s7*jm;%7g2xuP5Bcc3Jxdoc`O)q*nxK4cF9`NADq68Iq0d?A|<
zL<Z`Abnkx1fY@`Rfou4c<c|g}I_{}CP6xL?D(P<R?tw{eOT<C{2isu7nUwm>1M$J{
zzoQdy1)=rF@-%O`JWt^IYk`H1+i2BK*hHX;K0i^7))-goRU&}BKx%}HM}-ODhGvPC
zJK$VBwhwLb0h`2Se;?n&G5}r}aAg5M9HHfdujdGa!HmF=5DT<(Hmil|TX?lwc#I06
zAUmVxoyU<ZVr{J9=c%M0TFQh8WLn_x{h%2=dbEc2BQzt#jcA>|nW8jqTWz0NQZX?x
z0nL?dXoV7k@bvx{rooTU?0y9z`}eN3Xzp?N#mC1NA}>UKg4%U#ZP7z837d;_+0!b4
zTW&ShGKlPSlOV#x0?b4J@v3!_p*@Gp4**jcIHTZ?czFlc7ER)@7^ndxE6ciSU|;}B
zOGKGyOg4$5?jqk=MsE@lSbu6&!T>7V67uwq1&@0zz~Pw)DZkQut;3Fl(o?0ITox&$
z5Pv5deShhf1rtbjEzohC$Icv0Z3YK}*#Cbp0)EJKSLKg-IbMYv(kK|-?nmln8K{qd
z6^1ajB`OeBJ`zjrzio=WjRjrnKL~d}qP~{v4X)rw*`t0Jpn64U3*SemjDaP40C6Y*
ziKuDkRT(DI;jnYSn2thneHzJ4WJ*>$GGg#IU>E2$n4%gyLQ756R@E;RnC2o`9m`H1
zLi*J-f6t&E?-~bo{GTLYu!qZ0(!6@2NVuyL6PofKEBoBsyfmBO+`6{tfz1|u)nwp*
zA@a(xeO@IIKrk5`6#rf_y@SjI=<|fGuED{<<y!2myY$v;a1r3I)BHK=^y7~D>;H=A
zuF;DyMQQT%baeV&1se^iT?U-T5!i;I`w*?@+$5axQg76ZYhCI9RVzhwixO%^V6Y*n
z`9bgn4ol7LZ8__m^;qO-Y`i;vuf)&aS?9%$3=X>e+Mx^~LPB5iJKgFqj@71w#gBZD
z1|^I9HwEB_1dMCfqtx%!BbpeDTviVS4E*kbR}_Y*RP$gu0w#XDVXGJtVSnFgX$J2)
zpj#fm>IEL)JD_`Yd>%eZ`#5bZIx6hI{`~_~8c5iOF(r^Hp~lTkBERkB_0w;#f~%ZI
zmhW>P*wg<AQ2^25gHVY(7+H=d!PIm$C>5qZ*Csl}5%irDZ8Y6iiR336-N>1NOKUn6
zBCNjLMOOVqhd^-V@ohkKn67PNO~zU<7T;^e8!An(dp9u35NcI<uU^X6k&##+*CC<l
zk%LgX;3qpfI~Z@jgnJiyrf)X!O-?`<K}Lnp@a6ibP%yd?>CT!v08Ru%xhtX7$a&&6
z^+=lm02?yGEAQaiv6#d++357&>P8BFs|vK)(~vcOXU6BW-@AW59NLn)$ZqZazT5oA
zIWRnpKK}gt{C8&Aug51v@`3i1*7nghgof<|E*gF9Zkr_fXMj`lQz!B_CNNw(lfrsI
z9@PrCt0P<<daD;U+3Kl~O=-YTJ-t2hy&kX5ZePl2=eHbsf)*?J-^JST>F*%!nmRG@
zp%%PQ=?IWpW(_0SX-drKtsdB9tC3OldiBGz4npDQ>5&1?{rk_O2Rs|&Ro9%p;s}-V
zGO%~raxjEvo)hpMfn5L7CJlZT^35h(*YwoiM~upr9-!QW_tlPL=J-2-=77CP#CKQy
z9-D_Aq0YST*I0+Ik!BlONh60?SG!m<b+0jzE`5~G?Ro}YV6(=4`u>gSmCT_xKTod!
zgJ^A!Sd+TWbGog^Pf&R|4=cBuU4&%xY~|l^EfrB9jSHa3uc<Jw4e=9sz5rD=9q!y!
zv}5gp?!2nkYc(yQY;`w<i3VV5&4(avX>ybNku3M8it#7pe(mV!mI#~HdL`c24GnLn
zg+gk+<6AWMStS3C<PN_+jsE?CDG8u7=fdD`W=T(JX(_r_4!b?E*>0b}ETjjTNw(rV
zH@2EF_Evf!{c!(D!>{**e1y&)q<XpC-{Y>~<A+494<0BgG8A-dKP__iM8^fWYuqgn
zMR81>rv=g<FymwI_y0_LErx-q#^2vZ_J%n1&95h=q^p{QHFf9i^K-P5f4DRybyMb(
zuX&D|Fg#QcuVP-k)LcD#sJFM*Ry}~H`w_<hXutCD!Jrl;=K}<+IVA)@cNQtD8U?jK
zL=EVCTg$<nivJF!5}nu}tB~FTB2HRrY6Fa|J$ce&Ew9;wD<?0X_U@g9wROK4uh<@>
z#$boEl`9n9ikb{XUT@N2Cr7OAsmkwGN3;<aJECDOBB`pX3LNDLlzHyqXR)!d@$vD{
zLQP9+ymT80Hx9DMbp6enHyzBK`uW>U5&bLsopbopUZs)8D4_oR_U&6Bhi;#vs^;LQ
zNDmJWN=iy-r+Xei$674`cp2sTD(5?Ye^OjzlcJk!vIt%UG1u4E2YM**)yCsR*#lvd
zMmKK`j*VGWK_MkdV><(L`i}x5pigOq-r(;;uRm?jGu+kHb?wY&-AQ{_WF#cRg>ulR
zc7%6X)^?~B;(%Ier#APPhl##WBqU1i{M{u!*Z?Zn#^=UbgQcT}hlh15o_fz)x_$ik
z5o8MA^8ERs@3+$iGc$nGJ48qabB+4RtC=FVUhq3Y`QQi2KdlV`J1#B`MqZ(<mzPCc
zre)Ss;<-El^S?a0==@@bf0<V9OeAaZ;o%Tob7)u$At?*a?*nc>oj&Xl3<6n!<Kp6C
zQAng22Q(GB8nd&rgBIT)q5Xb8Btyf)Ssk}_bxm8=zBmg{BA6kY{ey7)(`IGzp~KtU
z+&nxqbU{L5I+x{=qGHqM&!3x`0?V{0K=Yxpb_K$;>RT;whmUesuQn~7#l|9G|2=Uj
z+9-i-Lf$qRSX*05NJv=1kDRXGe020pLwKmiG6op2PzxND9M7Jv%C?>EDz}11K|sZ-
zz>8n}qYeK<={2%}4ejaa351-74<BYjLd0v{_qnkV{bYmC+GWWua2d7mIDVU1_=rUN
z<fd38Dw6zB5B%8eLXy?Um)f>AHF<e%?T=5g6C|$4$#Fq1*UAbEk(&@vdo45Dxh7p0
z9&ng!1MiE3n~N@+D83k^Bw1Nm%1TOaVq&JDXa#25*RNk28XBPS+zfnJIDg=0D!YN_
zC$=pB0Bnt@{vR~spDqa)(BO)|Awl{y16!;J<xhM5{2IXNP@J<IL|;<vLamn*iFyI3
zT&OfZMA=iPe@|OP*NgR?Tv+yYcCaNqef=oOKp#+i08Ti$xvecNwUZUCcO$2Mxd7EW
zAT%d(ABtUA0-(hLt@ytW$8Btr<O((nDZ&b{4lB&>jb{-<MFn<XsIPAizqZqg0I{@>
zUvSGdLK<Z0Kpzkgkhl767=7<Fe-)?5gZ$gL1o4W=ubw~ed{-x3zLZ(Cyf`i9G*wzH
z?68nHjE(qE;wRu7U{In6cls92E;I-L-*z|uw|Ki@P|*+Q65+`a*j;KW7fhjpy953w
zyz8Q>s+yrv#SEn`Olmg(m{rZ&AF^O01hbdG>l{GLG%Dn`%7Rw&J2~yL_t_4psO074
z&CREEVeUM8{DNdK7ko!5hyiG|M(}O|@QA$u^}BNAN>SR(T{1PB?-?Oa%3o#hkPeFg
zUJC-a5O~>BQ&XS<$ZJG&!nQsA)WHA03p`I6tj>fSc=+S`s|ESoDF7hRRZ!IL?|y9T
zweF>7K0F}?TLV1+#Ijx0&Bw=QXlTga-+yduOogRC3^<d7u7u<Bc_>FDj_7c6aq%JP
z+tJ^-`x`oNa*%ccm=?e?15v_DXI5>fccO}liXc^c`z2gzhYln3q7!Zc@F*Xss!#^Z
zxvL?egy-sCkIyK9J=UHl94Za(1Ax->!)3vXNgJ^#9{a&5U8r7ouEBLqj)tD6U$0(x
z82-BEP5~^K+GzRuRq68O%d)aQaC~oS1y$A7PoF-uwuYAXw!qnu^k|jgadtX@DqS!X
zC@BOiV8r~{CjY)L8y4>N?R@<F!9hWgEQHJpGoJv5l$n`1H#Y}<CeIPjDcYJmY#(c&
zMd#)U36+<B{K#eSr32Bey}wi4erT)-tqBbOLEkSo*Qu`}5V917OpBX<TZxI4)YLpy
z=Yxp4EJ4$)_c9Y18UJwjTT6y{nqhMvdfneU$Q$bpJ`wAH=A*K*vc8@kv74xqHG@x?
z7I<#(-`3V?2<;Vf=P|6KcHnCI!=0TRlH}#&#(-F-2n8(TqraI#-BGvvjd*=`7g-;~
z>xv-G7&vgKv@=ZMJ0YWHM?&!EPa-e1!MMfXf?Yy<ypUPX1H@-b?*3bZdFA#fA6_7P
ziB46>Idpm)RbRY)`xZ*K<YW%tvV{~ZTEb^0w2vPff<?oVf|Cr+(EG>VvxgT+YO#OE
zMK+w`fpNSFo#fWmR%T}A`uh6y(ES$jfs~ZlxU{yZ7;NyWouV`jJ<zNwg!Vyq;c9UE
zvJOjtIpP{8e@C!?n1wVESb6PQ@WR<e+JguqkRu8$3~D}$Qe#jwjJtbXCz4>cFA(OG
zs272C7=Yw2kGTITh~;w~S>w6Q=-?%<q!jdX{sQ(-hKnRBZm0Iw-6jy?(V0L22@)M@
zOn+zXZfw+Iis)&j&PMD43k7umS`MUe`ufAr%UFI3?&~5C5nlLbcmj}|8pXiC0JxJS
z62T^ahuVKETR`*$P(+yC`0_>J(j{-iCalL=Ak~B>ZewG=HCQmRQ=V%u0uBSDUMRw+
zz?6VD%tjF9?JoM88z1yN*8wp2S9JdJ>X%4{j^B9)z}QHnu|?D#S**g;2^H$nMK!h6
zzF=c?g=6zqW#9|LqoAS!KM3h+bbNf(QhX8IHzc#oU%x^hKrjz~ZQ%kN9=n-=vt+6y
zD33&wP7!rP=99mdw;O^W+3QLTPf)<iOa~4e0C(;OS7tli=ds-59o`OtNJHMx13}y`
zv}uOfN=3kNQ8^Rv;D%L?eGy`CQ@`V{KkPVx&O}3_1DRlgBo#)dr+^9sFUmpt0CHpl
zuHgv?5^!?^>R5s2fp8iX?qs)|R>%M}Nf@I<Dvr`WUIklElU^Ocf_F?-$4ArJb79=D
z8u}~0L^de)@hR3c7~JLM0l+yzPh3B@994vDr?c}KdU>+HDKb1zgb%Majsc7rBxrBd
z*2vIT9o}A3g9JR7EaK5aWKDu1K7YstI;JJdbk^-FW7d=;K$TGcn;+%@5Nj>LDRFpd
zfd!)t4KUqQUk|Mr3QokO`8w}-c}ie)p&JZ+b#FJk0%DFRarWO-qL|sY%mSU<gqv6K
z+P3>0Iet6_o!m4dH9yQPoa&rrbtUmcdNGsX2Z5RohxbEFI7HwtC%^avMwaq^3G496
z-$r0V*p(Q*4`8FuTI`17Cc~+suM9!PV1jJa_isV?VcB<guCEU%E7<Ra3l|^{g={o4
zE6Y411lel7R*yQ?(IY^Bgeh97<$zM~M(XyXoX9=#{DC2TFPJpm@>`84z@vVbFV`Xd
z8cD&y@oq#m^b=cURe9L5H-OB?0xxhaB=~@K!`nu!i0(>1{Y$WT`Vu-V0q8QUyu2K0
zkR*%2Pmtz=zkL7xG<w+=ov`0ifTRgPBG#RE@!0X>$dnS|9S;9aL|8WoNfrR{j12z|
zUsoOvW!Hw0O4}Q$NDHYKB}EC@TeMnQRMtp^7Ew`1mR1z%6<H&ZNU2D&L=usR(xO77
zq9jX&hWV~D#>~(>-}n1{|Gd?i^PF>^d%f=Kw!5B?Fl*MVB1Qm8%gp(cNoo(!^QDvU
zuXT#B-w?Ff-*IlO>l_=5Rz;)-)rhGWK_v<&>@6*$k^ewQgDA@dINJr*C0CP^EdZ*1
zOdn2$GzzSJ2U|RJH|PxFBM+L;3ZIZiS+t$xYia`rXF&RG(9qp13tw<%63IgVSDZud
zX|aEO^8+cu^$ndA6d-TmFsvFvwtB7Yd(p5=T&T`is@HJo&0nGTl&)Jdw6YMT^v#Oo
z9p^PZ9lU<GV`FFbW?qSldqfh{nnF~PE8ss(qY7={FJ7cNe8?lB@Fv9u{u!b%fb3C#
zKxt>Y<k>tz#CN94FU9reF0w@xcZz7dc03rhYz`kj3|A4(CUgL<7%mhXsRf~o{5fKW
zqPjmd^xhDU)h$TRGcs!RRIfnFNGZIzUw!M(9M|<eXr)vjLUQ<s_Srv!iEB$%l)3O^
zp`v2G#~=9VtmwX;D#(#4K<`suC?-35b{iO5;E8Rpu!z!+|I^(~q9!b>doWb@6OikZ
z+tqAAD?#EL`M-esOw}rXgB5=MZNW)_iv(gJQ#9Q=kB*w_O_wrBwJk>$iinZK-~va#
zYyUVCI7pZ-IP!4kp{PO30DcXGpJ*ZXA@K*oZ!6@A)&$S-p)X5JOjLFJa+Ww<`BW0$
zRWZPPJNeu@@L5SwF$qywaxy_Kk~Dz?+S?<$+CLY1%tagMeg}tV<kc^y@(T3Vqmhwu
z;qTAS+Azpk5i}@1eWvzg<-HM_M_g&~&(UEJ{yRVldbXTuo6?rf&gs{#T?1MmBSUp?
z07^>GxSk)toB?#(IWIx&d-_9$ln+J+^`E4QJ2Z}2inZR-r4R9~*rNj2a?VAmx3!NI
z*PPS0$KzVRpDi8h*R|0JK`sfrb1ep!iTdC7wx-6$ox69#Y)oy}a&R4Bm6*aKA12=O
zWkqY$W|*bEI6l{c>T|_+_kV5|5Q$IWqX2r2kRCn?6xzSc?bV-q(f|!n(3yOF)`lxt
zU??RL`QtN%uE>odM8$m`N-;wlD|qzM7~dQ)WsxLvQG;D*T?7u#pWY_c6?T|eDXE=7
z@wzQZ(<(v?J3G7CLB-I-t5&T-hylDGV&%CPgz*#+#OD>Ed|-6TN{4ppHAfUrrGxlo
z2UujN;T=16VAba4W`q;~rji#f*k+41#%=C}k-){_HW^6_I?0S&<e+P0UeFR-*kin9
z<OBN*5#iiL#tnHlAhii(5l?4p1ahyJV`3zGhctr`4=qq656nZMC?obm8GwKfGX&zr
z4}ZHw3U)}@y|IaU&*%$%T2Bx9|4Cx<n|KVmp2!mW-TTJ^iK@1uMp(YT2@06WGKa)(
z)!BcLShB!bBHdbe-#%1fTfv1KS_4KAbRb<Q{mY}SFHVPNe_CW->7S_J5Ii$RoPOBe
zKACxVQq@2OtoRc{Te)r9W#Y<sQHXY4K8QZv0v&fjS1RDZ(o0E6-@w3FgEs1Yy)??i
zGSaH#-bw6!hoJq@l~12~2pIvAW)P&xG3v%5umnF$(E0Nbxp_;FuLQkWyV+djs<Acw
zX6|W-y|pIDGmv@ci|7eD5byu^(fMuviqiFz+s<{OSD5iTb11G(H-wP&CZ&|B7EL%w
ze!`>olr%gWW^6BKa(+N;Ur*m4!u*;LXVjy`4ovNBjE^FD?Y}_|OqVJ|6hx&~E-o%x
zUQP<0W88*LMXqt}0bARZeOgxF#JupQ{%+I#yH<=wlVWWWS26`=z;R;4O;DN`&i<Le
z7pyCV_};|S^zNQF;~vjL4ws+=h}FLZ5I!LbZjEvWpCY6(JY^iI^bc-f>sN{eZ{c6I
z|GXg8wwE&leW_<lACs=2gwV=;ctL7tJS?P*ItD{ix|%E0fq4j`n;{1qH*P$UHX!+Y
zw5PJK?m}uWzGX-rIp8(sdyGQ`g(SU<eYAm$AI#|5@BGy_ozkLh_4>AYW?)gD@w;w%
zw>z7Wegpl&%3D<c%6)s2QlBG;I!Oi6{Wkw{r8;05>N8P*?*;`2U#xQSfsF%J`zPXF
z4J06IyvbuBaGB$^{ds|$i8(#Z_Nee+(Z~^kd=wIPCdzd0y#~YUYa&1mqec5Lmfl&Y
zg?e2MgD4ULd&xNSpdEB2?Ls|R5gVeHMK9p}i^#SAviI!=_7%NHdLJ3VDk0P#j-FyP
zj~(b`3MSuO*Pl%aS~EQW;a8D?m)b#L1?2Tf)po?&kgJFsU;yFbY!hK>=@pn9P}*5m
zgmmGZ-LF*1GVdstAu~HhybVtW<Y8($Q>*kY0bvdhTXi`y$UfiR`(72t#Oq5H9uySi
zofsv($DS~YAF5@3N#}0Rx_t~+=)HV9D~tA2xyEr5(mqCmv9pqW9o}2jNjKpJVcT9>
zSE(8EC5eAUSDc8s<2u@*2K+(|!g|I&{-X#}3YH!^^gCYaC6aBdvnZ@eu`z(zp&i<>
zabqP=uMt7WYH3oKLD8{<t+_VaC;RYk)q_mMe#RPHE~prtrl-PdYN)M^#{1*Zp^9qi
zKY%fStU+4L`q#*RjoA!9Qr73p#mggQ4=nMBDs58#0jZFSSOWQ{Led*W6r(dp-`zkI
zMbD*y4fAU>{452E0z#fMDtqnA=IY^XAy538=~&5MLCcal$YngNtHBrX&uT>NW2hwm
z6v&?ay2O8!Tt?+#(E;XJOo@k;mhN4!;M_{}U$3_@mKaX8L?CF(Jo0`?feuJ@c>ibA
z9|F&LfyYjGcu1{3SnQ+iDM^uzIkAs0(&)XeX3Vh2WqySN4b>aMM%@|k6ot_HC?i3A
z@#!9<x>ISw@evZ|V>sLCN9|(PMhvzi8&h|Bo<IF#1TxKta3diaAulLj3J4g;bPe9a
z!WMt3UPUmPjNrfMYolWv+L5-|X2*_(iyDWCgoR{LE#vOq^~6Fyw>lq4!=|hK-(<xK
zaLf7wTm&Vns9p;VtW&MMfx>X6N{yMXMS5F*Ec^L0V%bG)=e{N}5($GC09e}S4*JlR
zNh;{C5k_?U8g$7^3H+J3d8=IeL)RtAXz_#yk(FX80)-Q;n~aSY=^qv*AgB5XeQPY)
zLMqWnKDO3ef~?&=26+$kr+T=RI1E*N5BoZ>+gxNVF--$hird`%Zfwb}B631m9j|X%
zW6@pVJV%e#Q$s(Z6z-J8R6>uF-`eMB^d3`0WdMy^L|k<xspLe9C}b>>WOApt0!^F@
z@3dg)+zY#HZN2GpBEcz*#)Ue@p;E-3Ct3-roVHgNne?w{5Il5~xe-yVzL4`YT{Cil
zO(|&87{WH5N=o!qRaM*471pg^--afUYXA*A==W~3K&{>vY_2s&u!#{$E!YxrTZ7u#
zm@adk^W%p79_`8=3##7!w|#yW*oa7P<Ay7`lMZaNLUat*H=BMYzTcr-4}3o6Z`9=&
zT}w;^-P=d9IHgzV45v`qYFh9+hlgXz>4eu9zBAH<CK%F$BH-nPzCcKBGBh*;gdZ+f
zHz6ZNHuLi=rc(3yjdtuaZr2Ku^^=$(DK46-9)2Qg^0aBaB{rH1ssjH*tb}PEXMWL@
z0*h4OX)v7d@!sIa(vQ;AaCfQ$&KtZuL^q75@*qFHUpiJ}eEY?ZU|w)q#=NVLIRvN<
zsA!HV8D*tL?B`92i#lsTW{hMxx7yBI1(XN|2kng~oxwz5ubSDLGghz*$>C2#S^w3{
z*&oxmlc@>{Wh=jDe4V?3^^ZS;v%$Xx{pRBu9U0R97a>>wLslo0!ldGQc<K5rTil`J
zSX(Iif=s%<MP$hQ9mNFkNA_`v!e}g+)&u9p-$i>r@$V)3qnh=|L6UMX4U?^0x(KS9
z2n1E>ddYn^l#Y;Q6KigK3>!qIJ)Q55oL$C@9-U-#OVZuR>cP>WXmz|ntl(dq|NS0f
zSKBdhn#8B~D@{$qXW%a~RMYlDwYgr%?@Hr>OBR&915Z6!WGEFb#RcT(rT`37ojzya
z(fu4S27S!uV0f82WS-oF2@_hc3BwX?_3+4HY=^2O?XnLuQVP{YT)+l11%1Q}_2658
zTSpXLzgZfo91zZuB`E99`qd~0sR6}MKFv1*!=<MAGBzT?cj<AGMb7=yR7H2Em|M5X
ziL%>3g*A~*hA;dn=rOv88oilEyVi|FXYltrD}7*~3mFSgRU)K!A{0z8+?D!!I?NWh
z*=8|wPM2ivq@@+KpqbPMVdx9SB7=Ee`Y%Vi5u6V`F5+#Z)v{k$#uHZ1N6GqdJ5aq~
zNbLqHM3eO8%uKa&E6@CzAA99Wfl&Vz(m{iavJ<`onQQbjs>p(tboF59Ak%C7u$_=s
zDU>Y%60e1I79FU;Bguj*`BWO2Y4=u%IE_+N$vFMjsA5fNm-zU^qW5;cm@SdO9!fC-
z4*Q5(a{+~6RqR9l0fiQD$RqO;$}b+p$szA<;NKq$en1DwSj;IrgLNPuq*g-KIY+gQ
z712B)ha@cM-LIzh%h-|>cwk807AEVP!7w8hWB5Cw6mD09r`g;?4~B}I`NcOgRCn*2
z+mHJ1ce*xs?Zr{duG=hqzw4>Jznifd*^UC$cKAT9Qj@Pv??lIUbN~4YCV{11DpVIL
z`yKKL44X<-3w9&ggHZA>Kiqk4bsS1Q5&tn?&XQj^8uwjuka1!xL6x6zTaR?LL&>!O
zL}HKn#l#B*t_AM)F|o1RQ5Ycqp0Wz-CW!;XB(JH>MTN+|fuFazUm<BTqnypss7&@p
zIR!x!PKxF^m|Cq**((klNa>dozH_=#2LS2E)tm=drnO))xtcTL6jw7P2MDR>23i`Q
z;tC1cvi~ePA0ul$FXKxL!~~^yQo>`bBTTst@xnia7J}GAxZh;(_vR5a3`R#`6~efW
zLgpo+j_L<DYxlq5!baF@hOYK11QY=oQ<kpIwfKq_fvzF`=hi^9k%=Pv`8b$Kt3Tu5
z|D4pC+cWA3R;u5OS}LO5sOik41(3EI0O67<BOW4Ufwb`Q-MfU*=W_N}qjZN4xw)Pj
zm`a~d;i5-xSD!(~9Ir(%$Y(7myf~F<1a)gQ&586avfkmD##cGLfVz2GM5dD#`}6;&
zn0jU>Q223iahEkdt9(sLflg=bqp%+QT0RV;On8HdPun%Pg|FW#E0`k#p-{GnJRy~S
z1?fN|jik6Aq<qM*4)P@;6kp=~8wtAj^$F$E_8V^82-H`<Rp<Awc^uz?5BiCPI3gg*
zDrWv^kH@V@s<Vvxk5#b5W^S#IVEPc)zP*1WlL(456$V8Cg-KzbeA^b~;CU7opER%p
zSA`d#6BrS!6BHCg8gdcj*ZxIl$5hwT3rK~(-FX{&jagR{aT9qPxh+{L@Crae2W&ZG
zjpJ0Ik0Mf_3S~-jGr!GeNxpsstT3vy#AACjVKd+vMAG!BurJF~fZq;Lo(z8=ie<@F
zT$pfNT2nqrlj188$EomA+)k$v3CBxz`|ZAQk7d2Q+`5|WJxtfHABT;ARGYYt4tx~a
z^~Wm&lr_E0HU7W!FUQ7yBq@^BOczul0C^!V1HT<Vn<|NRb6Zcfjol=VwQbr?MMXuH
zl$||9_;YC0!1KB_CBzeH@1(*3{}4emO`6lnx7eZ#b@AdfmYB7HFq)L!lJyBh?%K5{
zNa%UiQ{ZY(<otL%BOqTLGm<`qvrLz$G&Agh3(5|ARP?F7=0Lx$-dX@O2WA^K^y})}
zPSUSERK!d$NA;F%c?|gTY8d;Wio~>Dy?i;beRe%z3udY7`s>g!fTlrSA+?x{w;%Fq
z`B#S<Gw<2C(!+y}46$a_;WKz=-_g_6Z%VyJ{6OGR5~ZaDv1P(zaZmYFLD-7#$nu30
ziekn~0PA!@<ok<fP5H{8@&pEi_|{a$bQvMv#v3cYxRu~blatxI@#%mYhS6Y92ncxJ
z6VFYQ;*7nC(N?gRhk=M<Id1Z;1ng?y_sAv_^ERlCR0w&naA10EljlvGV<ayxU(33k
zSw!3m8Lp$FYBR!&+(L>PdY-qbefGjnKXpKO!n&9EXBeV^M*|;W3&s=-$LsA4qp-uk
zjvAYos8(g3UQnhEJekVB%uvN)<e0Hz$y^KEF1iFTcQb=%H4iHDhY`$6OD>-#Ej14P
zMyvMxWM+k{DIXQC>Lb6+i7M%(K$M=(&F=a2miW@2Cl2(by_TRz$0bk^r-;7BrNiS4
zxjV5;z5NX9SWX}Z$Ba<=8yZknrm#5wZb#F)HFD7<vZs-;>AxE7#c$yD?2M0`n4`LG
z^X3bHI6$v;5*MYGlZu#`MxUMuY7(o9Mzr4FIV@8i&2xQlctTa78%Nq4LJB9`xpT+=
zYI@(WLQ)?Hc|jtS5G~0`$c`Q+Dv6?KIxYz;Nd;L+otV<|#oISrfS(jl@c-jQpiB6E
zy3hKmw|T?*QWn^bH)8g6fXk+R7}u>96C=%F0OJLp=_LVVcV@dgNiw0Lc(c<f-t4$&
zv$=QCYE?eX5WOL&&ra4)_`?1m8NNPrC?d9`q&;8|As*5L3ZWktjd>aQSy$2#<4;*9
z&{u~c%+2FZp|}GW6do=r_Iux^J%v)EtVS(apsw!r{oZH)U;dIot1)IlB~FlCf{_tp
zB0yomFfSC9dmWq{RodWRMN}K-ZJV}^&_4Ui2r@d2fO3=+^eyTy<s|DHqLicQW<;t(
ziWCav=zK**LnHvJ+I^7g`3_dwSwm5XWOhC;IR}s@qmHv#dTw7GhY*2BQ$%zlipglG
zM_G}WC89QE9r)SEDKSn86FD16CLgc6hW1GrKq*vov{fjq>+#DuIXN*gR;imA*-cEp
z+A#7ITb5!vkUT(`igQOOv4MQw6pHW5Yn?}Lo2OtP)r)DMZ-j<2D|7h1CAVCEJQ%LE
zFc=VZ@=@AU9|aop8kDkD9G*i^mFMVQ8%FxF_ab=^oF$K1^dM!RpxMNqF<V=J;;Z=U
zNYSihn-4)q`O7GBT$<6~VCMf6kB(bnq%-XCl;%*P<$x4wI==*97JLNbk)-dKABw10
z9kPwu4F)tK*YRj@YaBC*SS>&p<wBIl5z4|DBhRwGE9MX&wWsH1$iPH?$UA|o(nXYB
ztSFAz7pTo_Kx|a2zIr3%CUD?59`<t+c)$*B5Dp(fmg)bLlV-UpS37SjdG1XDX@42!
z@JJ%ue3%Y(0HuBsOd}@={R(V@4u#{pdLLZl!?T0Ln%6CZdPib;HFLx)sU`~gle`#k
zV35eo=1!GAe%$;&r6u3D#Keb)2$2rk86$yj>ep-0>L|o2(087Atf%}Q`9I$P?w0So
zB$@4uBu{zu5;9Tn!jnaDmK-PaG})u)D=PFks^wKxv#}0^|9f8YAiCHJ)Q=n8-5WV$
zWGH(lJnwM_t%rdsuDew@4F^M2p;~>*6-!ucSxnB*U&etghk1^qypyaCUlK_*^chlC
zm>xBavglk=BJ54FJTbW(AF@Fyyr=tB&a@m=g#({TK%1h6WvHr5I*kBadb`kmfIP5L
z2zVx@5KlKkH#3?13~h)+WSV^qDp0<IdFq%E?D@Jt%)msSdih0-eq}#`L34>AMnVG9
zNwpm9P>5s+uJPL&ycH=mis1Z_4N?*c_S5eFyl%fpl98YzH%$fu5d<S*b9BTa%t2pk
zAcc=}UJ{Bv3)tZDEowMOpu!gY%UWnTnR8BH^N>1FeoF<Cg)UHuO#AC?0gu$3Z1?PO
zT-IM(7jXLs?)=$ua()PhVUaL?2^A|UY~uJp3^*eCea0#1_xtqTKl(TtWZQ|#)B63k
zrZ)|cM5JTMIKcnA4*KH@vmDalF-;X>1Uke<cr;_YPv;AxP03pT0g<ADIO}Y)w6ug4
zxtwhm0;mke(ngMBts(x^U+ud$JOimfGK>+MaHLS6<NZ8(UU~SrynR+tQRmG_;S#MN
zs@GTLL71XsX96z?OHj3Ujz0yuQ>5bpekE=#jaW2;J|vz|EFvmF`wa3ynB|5XC;+>k
z(USlNG;r_`AePYmFrarBC>V(K*3kcW_zN)oSotfMv~A(=U?^_*^h7yVpw@H%2&VAL
z;{?9d<anTtb6Las>eZ`2yI<V^^UJvFDFL88{&!6<LO7%lFqH&(Ny>Ih-(1W$wAwPC
zh5?qDGgwRH0Qq2psysFWrm!hfri@3*zKlEGzj*syFcLvY`TaV%6=4>lIi{>EllbAY
z=Uw_f<ae+TAv^?M3AE*3ge}a_ODdBvk1325t!PYT?Ipq!uH#&=mmu?q`S#HNxfEG~
z&=+{`)7m{hq(t?6gI;ExNJI#xNOf-5MyfJW0GSV-I^BOYtC~?g@^iWSBeSNblxP*k
z^V46Bo*Tr~lIKF^c(j14m8zyQ-^H#fH38Yl)D@$m{=EB)7slQ~EHat@!UYXloeiuX
zzTz(yW@f%5Gz&Rnt1WYMeLrLD6$9Jmm8T4PL(Jor>$A-{Ak@&2(B`G^JLj*mdzY?2
zvh15aV4P?c>nJ}y$<56LTO847vuBeq2kwF`#iP+(Z_$}2Vl{0xjf~f1uThR1W+~~u
ztMoxAb^tyFK8DJ(3q8}@u0>z)$tMrpZxdGZ*H<5BGF$DE^<mSpa&j!7>;IaQ!VFFV
z%c1L4s_Ks~lWZkp%=)yepjWOl8Q;Sh;~<RwYvePt-ZsCHaCj=r6Yl*W-ur-cw*_TD
z(&2hhHt1~?r?0~{{+g=Nzl}9Y;TkXp%MS;h)s+G@UekGK5_(j&TEKs|)!Pf~=OtBm
zddsZlxgcsbZqkC~<vy+Q{*$)FNHv9&&kPg&FJ#rb1qHUTrqPRaM31J;IV$j2(o}Hm
zxyt2AZ$!ea?GiKRhlvWU(E0aq*H4$P5mP)8ZVY!#Zu-;SRj*-xM$_C$>vz8Wesc>`
zWYYP(TsImTo{x;2Y985%nXEJT`C^C2Vv>4?6wmav0*<^q+V+wU`dWWOcf7v2b<*#$
z{maNd<UOt(-RJ`U_3~k1GytN7b0QzFSY!|;M!O1o3d-<f3VXEjIZH<8<oVHrS<&fh
z@UF?1(A$kk_WM=~c=0Te@DvnKJfnG}d)a1bOtptEgbgUprP#ov*qb-yttMQ;0BnDO
z>4Msf{SOlkC?p_*FHOrp&<*+-IjkT_u|lX78p(mH(Z6%t0dwB>R;3orn$FM1*vhlB
z=oKe)KpGjU<pBEB)mcp!jJ{<QI)pD?nODGTNi^o=x(biz0~{5Nhreo>KP(R;vzxsc
z$ALMtw$4cWlD0@krx>8e035-eh@GG~lrw8hOkCgXD~zksutkT>s#QK4mh>O(I#SR>
z*`d7B2l&Kmsi}qn{I<@UK6&Og>hUaz7W{&V^wrCb_I0|?C%c$>3&S~&ja8LAu8p1G
zaKy?{hp)P8FK!;04st_uz8}~29A>=(?6n~H;vY%o^?uQt+1J<pZx6cZ{+%~(US;v(
zc>PGgNgB?xzRC=^LA%PtjgE|>xL;xY5((|4w=rs2-=0s9?*tjr+aWcsule}DUXmHo
z`8Ij{#EI;^R6GEF2{#;bO&ASB$=Jxr?Gzh%N^%8o(G!%$+{b`e;JRM1?&z%4urXqe
zCFjPEACEg5Q&5bF=h-8)1;+DX+WWKcm3`E2tHb_080#*0%UXWVYxxz&1qCF?6>FPx
z;c}<G4}V-*ikVumf&$OvlS+8Ko`>@s*Pf0!zdYGK{Cx}zu_*x2fY<9vm4XTyLbu%t
zVO@;Gbuhq?rfCQX5aaM@Mq(qm6dO?`gNG;IvXpo2uw`#V46{W65c9vG|FNM#&gdrY
z>Z(1()mFTW(K&rPC1u0)iEbbe^(m~^AJH%R9UxP`0gRMZhU&hGRnpl=G2mjdsV*k%
zMtYsEclOa$fl`ZDLQ?Stom2X1&8c6S@_|k=UcX+FpYMe@#rtX8S)JvA*Pfs>65j3-
z_9u0l-m-c$LOMbavi2qBG;pJM$S;JcmT4qoV2z~ViJhF$FY*#IvM^&0Ll*Wx%zR~U
z$jqB3!pGZmjN<VqD!R0N^wv3wIP;xQ{~JEL_=<jH748LJL-<DVfHf%|@T#Sx*)?&(
zQ7~%&@2SQ*k@-(B0wm@3Z6d0F-nT$hRc$Cl^}Tv~$pWE2L(pU8$Nmmsg2yP&_QAqX
zBM=X#Clumgk^qMb9Y7o8#$v)k7LaJ{%N^|m?3xbL;(-Iw`VlfX&FK`Uxs&K+q3|Kf
z?;Qi#yEp*7ZU&$Z6bSgzfpl8|DU+p=vK8QL1E?r5(hFq}G<Q_6pIkXEvz>ShkTOii
zAszjYvdEhh&tG%cKT||+$v=xX5NC^rFwt}`jcxEfumWlO>X+kYeo?(XA9VjZet);9
zGEp0w`s%jPGk1g+F)4gdd;*FwnC=ki)>ecsYcEx^M+lDn$Qg6j)2j%oY;FF<TvQkv
z*?M|4I=A1U&jwZlaTOl8pupP6REP5b9jK}pPqAxhG(st0(%82oWetxM@cw`s`etb8
zZFM%oDJxK1RAVhE{~3)%X=eBoA%^jTX40$h7Mg(<e)*a(X$i(gwoz<E>^<Hp*a6nd
z4RdUILCTEmja*5*dGklej7H29#R>~3S0FHV@9UmOhOi|*ER8fRKw={zB2Y*k<;Anm
zlDovm?IX$=AdQ4gd>M|7>o;xs#I_Pub*-Ial;1%G*uJ^lfJca@(?p?+D$+4{OMKb4
zA)}|NzF<9H!J}aL^RsuG0a1fplrztO)w?>8;yC?mQ8)ob#+o}L9IAH4KeAGYXX$zg
z`0wt>bS~d4t$OHt7Vt#V1#7=h$jHuRq~<I6l4rS}SOZH&1d0KJ2Qlz8^GLvm3(@_6
z@3~B2wc>8gF-etCQ&X$t%O&g#&g$P;2P5#bt>FneD@CyoLRe$tP(c9)MGE0p3?>t4
zgr3te(VT5Q^AypRg!j1urEnC*eSJsbbG)gR@I_B15}k2je>|FZTO3%{?`a+1L+fls
z>l7XiLI@pHC#n1-zdDQF1`ElPz0czf4zd0{|93zVwLY!v{?#lC!toUD8ld8bbxDYj
z!uOW<nRi+0$a6K?It?7{C1quILtEj&@>AUEn3ii6J-hOg=!A)f_0wk^kQTya5^ZR3
zaSpsxD+&qq@|&urWc7?5+2U5n60F>spbHl)K)!@e&&tYzzK4)+q>|0E2x&@zgD{-Y
zix+`pqUyi>K2>Md`^-ah1R@9E;eq`Gn8a~>y#G-N)=EU;0U<5)&mO(;?6cTzdUvk|
z2&iCTuv$&(&OkwdA7iO*GsI}lK_ndG!@eo+ez^xC&wBK{iK!?=>qn|Wt-AjDh5~>-
z#G;*Yp~c6B>74mVs@81D^1g{;C}bOFI~VX(qd8Lid^9RA{^TwA*T9V}s?^Sb5=QmW
zadp3N2_j<M*{2CfaJ@t#w}$FeL<u4%Uy+Waiv#P19TY+O)29Y=fGxsXqm&SWyF=8c
zyRgvyZCZ1<{ryJPTdS4Pz~H4m<BO?AxMdG=%Lae@sTDk=r0Tt>*WTlgDa%svR91i@
z6IMZAzZBjHuAM?mf7-z`p|2-FsuQ4i8bN=L>(}UpYt*OZATNgd1tV01#^E$U0sc}7
zAK*wP@>w$qego6MeyMARAP>UGVM1pBWV3JIEiiMIP%M#Mb#v=dNaFf2-!lg(<!Ebk
z5EF(Qd=f$oG!((&JfIN#^~vrXm&J8-n@^3Ngya?p@eoNwD*+u>gI9nhFl+$poj`?&
z!oW0@Aqj-u!r)z7JO=4xo;#`Ai~hqwL5UH#8$;B1xo*Wc^l~pTr$P_AL1N1m)~mwo
zKm>`E?*LTI<LA3{i<=_C8W-mP7D9A#NH3ba60cr$U`gxR%>{mwF9Ze(!(T<`3cg5@
z>leu$co9hYvI9cO4MOA*?K6W#?e}7ew$Bt+tnd1a$YL$mC%5U-)v*K<vT4(%#6}vR
z2es@o)r5JQ3NZ!AC2~^wm@2cCn?6gKzv4(INo6ZUMrj_&V5^XB!GJ<=YZu^70Kl<8
z&3ifogsZDuJjH1*0&AfIT}XPejs8`wTSiHTTB`drB2Z(hvJ(7MXK*js{pYm=sW!O&
z_&E+aO-7(Np;fDVW`twKDnjzMNjs3Mw~Q}&?+TKw83nC1y}#~GWw}3!p5%>y91j~o
z7APe)Hs6?6FP!RAuExaFMI0zywa~k)sK_|OF@Y7Cd7eZF;w!ALG9BJSxsFIPp|IiA
z3!z%+3&z`>3CXK*uJ>bQ6L`fcFq+8`sR8Ii5-zp1?Y9}KxRL5yg5Zr2JYFR&+c?Eq
zc2g2t!k*<03`K_jzhwZIRkZ5t4UD4YvqkK(sHU4Uga_deQrr$}wF9|2A2``aGR_vC
z6S3=CHT-K*5Mfy?+c{)ZWS`}QD*0!mtt*wZ3ZswHPc#!HubhJ{z5FcQ9R`DGdi5i|
zM0-kc54(5xSG~QpZ3T4mhB5Ej+uEw#+Vxlt8IRw0g7>|lpSr7Z7qZ^i#%ii<WaB~3
zv#N3@-PWwAY~?AzPSW2>8g6drPeJb*vONdPv;(Q$jU2j1$*?i2()&mL9As}~=QzIl
zRx*@zGF0o!>u;e4B1NAma*Yy`O4h9*Rqh!lzYQHalr1F{v-#0`f3UIQUx*r-;+tRJ
zdU%Of&yzyb?%dk?@Faq;II-lY{l&B3nz2oIg&aTHQ&j6VdrNhEf3G(2^de=6=UfdZ
zHZ%F|yZvkO_Z-2vR*uaq;KzZcAgzWM^z}yO+-U)7suGu}*4PF^u?yv&#Ds2eEHsI)
z^k(Kqcp9hhtvDlkj*$yYM`Sg1>e+ItJGuoRzou1eFB!hWzN2<Q%+br;<{xX3<E}sU
zu^H8~(cQOi-`>4@6Sum;pupLRV42QX4*}YvdG12U(i9a<(r`^3!ax5!#j!4IM{A3$
zRfwiL&6#b$Ckc`J9T0~^Oh4;G^&wuE^0`P>R<=^K2LudkcjWBi>GBg#n*$z>Q-#r*
zXP)swfp1bFR};L$UknwA`$7s!qm=Bz*jl<|q!(Cz=xrBrra;)<-D`(sjF@@Wm})aO
zfD&7-TQ-ki6Z=CMo={N+kr@(3eFhlL+9!Fh(Omss=mIBm$8w#7PoJQwVJw|k@A&Mc
z=?=h0!f!frL%*Y`@>)_-i3?w79>oXkazXSYdGZqq(?RSHI`fYg^#5+KJVGgf9pDCR
zYOMBhuDJl?(E2ani9b-Lp<J1+E-h_?f(~2m=WFo$i)(3OP9UBT1}~|smzY8E2KRNq
zGnBM$pd+pRfo(!}9?vD7SI7(()bb-|pY5eMRh88MbJ;}fTb?+}W$)haC|a<^O)*r^
z8Q{fbzopPEfI-EdEWH#d%va$$-(H*1y=Zaa(2e;^YEh#TGmm`y^l7Bx&i_(|R)kVW
z>x#>0nLoclf7L>9dxXbO(;VBlC1`}s8<fPMXW>RK@aT0b92G0a)%Ofq<iFhA-96ud
zM(BswcOx~459a&`ko!ocgIo)mcDD5BSZS(7Y>_{Zj&N9#c*}Xqmg)gsbacYWTy#@k
z1UmwxAJazcktm-&WancFVVTiQ$fLYNof~q#a$L}<TtKl7V(2q+r7%#WpVda<J;f~x
zA+=(M%J-UuV?f~laDYRt*d>@gC`|cu)hm_i38Y%|G;J2T{DVewV?QIW<GCm-F0`*!
zhrpo|M{SqHb*imEU5~-{>o4TjYK#qJK>qu^VTlMqBhG?JeX3Unl?ggdkcw7(pr8t1
zD@g%4eli=jaLTH!TQ3b6UP0xwB_wUh&OjiaU|mCozt5R?U^Oe$vK07+z#KLmuc{<O
zp>)d1;S{Pic0Wc|E-Sfz+SiwzaXLPUs7P>DpfW&GHGmdPReSb@;{9A*&(zilhX!kF
z{>gkGX(BvDkXuA^>8P1VkEN*J1T9K*tlYR+=S&u1bLY62Q@NmIBrRDG6ZD%C@5L=z
zh{ArI`vwenX>sxDh2OwYXrn&u5bIbZ_;0XkCet)r&j$yqdQjAFl1k2D?9JTMA&v&&
zN>5oXm#1LJO?Hd|eyiKJ=O^>nj;A<W+A%nNIwk2jt$y|@EGsC;r)B^B-M#B3cYKi)
zJluT8{nV^yG8Bi~IXv6-r?0339>*(=ZFG)hg(scK^zM1~ycD$`*#mCXovrmVmcOm(
zetDOD#R6U%`vR2zqM}-Kc*bxK9e4QWoGz?K0J1mn(m@QZN0U4|iWTrmU<WA(!(2ru
zYCcf(Ef>L7>V4@KWU7Uw0JH_pAWF#ryA>inSU#m4#p9EQ*veC!=1g?J5ZVIWg{7sS
zX}x82g5}!tOqVx7f*Q#ZM<>a7^X$?6S4~yD+4O0psJgR5{+vI-rhC~5Bnkd$erOap
z^7ER7-wpi;U`Y|J{95lhjBBP7b~FwK7s1WK%JG?`VhIXF4ndn1Hy~(%I;rW{;@Qo$
zwno-Edi1Eb(64zOU2E8Kzr^wmni=bk3+sDhLugO|n{(SPPuCsX1HnSjIL}!A8Yu^4
zk2v?T=@m*KZUaIV`{7r$C{FnuY7;_%U&(+94tX<LFc0M%wDKTjL1-$nr$+ZuD951y
znc87wt-->S@5oGmJ(1&@ZWf#zD}FI3Xo7hpNpfdNL|>#@<~!-A$<NKepis6idpy^~
zxA#PR5V6L{K}_ZS`GC90nY*1(zoB70apzAT>5J*4Z<ggTEwOBVfAFmxOcxjdkw<ma
z(~jhttTYZC^@hwg_Fnh%a)4$NhvX&+qp=gd4@#afNOuE9)_TP=gW_sLl91K~1%P_x
z*$m`uZW)!ZswbX;Mw?MTIq>0x2vLz*y`89Rx)^IurO11XdI_|osKO=pwj$}NLVDqX
z0w_F7kYcT;;Oiezmr_$Zkjj8!as2wyuD-ypT4e2J3e}RA3IG@mz%Zo|+7N8SD*uKl
zO&28fT2e6OoqltjW&zv~{Y4GuQ7!|u*WGP}|0IvTDoi0dTf5Q5H&r3T9br87c5%So
zjH0%;);)s#J`L}Z5y-gg)^4Uy;zMnL!~(Zm!tx4~dN6!T-$%+ofFEU8=JxjFk6o~>
z;LJoh6ZkTO{(-JGVeMgxeepIyG32Xv6XL{W$5Bk>Sm#~Y0`RXPi;FKL6`)7FcPN5r
zkuajnbz?}zX2mjz<IF=^i{)~c(^rDQQXQFZj8J;ddxyf{i-c{-cGWp^>2pbFy|)0w
zk$ra||3o#R9V_6w-gX<C0I`L|Ux)0X(8+UUgr#)7Ll%>CwZFH60K4)f;MMpefn8)N
zk9mY5OV;9O>j4sQ)jr6>BVqJgfH^H>>whEq)$IN<!zM=g+(lBMhs>`~pVk7~nXi>j
zG^xJNz~9)}_5kyacU&-^2)Fjh6Vp`Na~A|&42*4mUTV{_^F4i9S~o;~`{T1T8tp#a
zW5}@4q6${|;JXJQB_UE)oOLM6xU3i9&lkPeMWnR8=~JEpn$5k-+ZmQZz<0nIX-3(h
zL&O6A0JSi9i(68x{eBA=UKrp}JH6F3g3a7upi%kz>V%0COT(3c46feYp{iZunzp@Y
zT(*y*4FR1Ln@OOtnlL3_5mQAm|Ax`6yjeoy`$der0Qu{h)hO}+^Y?<8++4jnYjw2A
z!1I&^rh}yXH>Bt<&T;QS<<Jd@l!ERbL3DF%`z7{y1RfJ~PkZ#F`7^ERXqUx%9t#Pf
z-o$=s^6xc2Anrg7kzE72J7S8)0M~q1OG=sMVZ;EtYUVh93bmQN&E>M&3A~RldSCPi
zNF!yw$WdaHzP^;JOuE-^l#Vaw6nNc-V;2oB70ny3vU<SNd2wG+0mbu-*&?XEIf6ZH
z3|$Ua)*;m)#gq{8osz?zq75u_hNO1C0wM)iOIP>O*7`!6C-fP(jVAuRdIflDF8|@{
z>gvL*kEx6He?%f35!}d8qb}2F#JUXsXNqmDJ;qqOW{=1o(I*#SS-5Z#D^z-S0do2t
zMUf4Ie!0s11v}u&-cV~hz_hHLjKs$3jlwcAGS1MpNK=DK|G7u!H9}elrK<rx@`T<h
zfFBnA5HSXCT3ey0yk)MiQ+(SBws?kRk?f>Udjo*XuY`o82Haw#Rx>1RGIA5e#bV%u
zc{Os%xdF1d@E1rIetmtFTwM3?>blFbA0F(maEGkmt{Ogn-rRK@MU#ae4;g1U?`yMh
z##LwSV0{0SO`pOL)~DUUV9H8|dpI_ToW74dN7Y-vYiF+)cr9N-P!Yw(t_sXDmIxd8
z>q<K-VrR>aC8Yec!^P_~x1&T6px(B!ZRs=+(-~-X#Xh+I1o_SLv`N8p&70)D60!Pl
zbN;JTGIK^KumRY{*Fr*b27L5WEeG&6pcpp^!NT(^PeBxsZ6<Ma;4g33vSrJTomSw@
z!;L4tybLg|dBbH=W4E-h0k*VA+!BSWEvePV0a*G+_JH1NJiE8J<6T$WKmfvM`WW=P
zJzDJuJm;m?9>o^}k%atNLl0lVz&l@nZ`GSCHX1hxF<$6g1z$KY>KBOLrl~vM6!JJ!
zfS^9hmUYNC-|o$nvx-)?eI~JD+2a8^{!HiVzsho(M;jJ#`X8QR&PYm*8aeX6)5E~6
zneb+aY@iEb@wOqauA9D+ud&4e6looSgH(No^{JIW3zY*wZg}}400;P{SF?oX4>;lt
znq3&sh_TNkOY|lMak*Od4vB$@goQVe-2NE67@vtr0c-1RFy&mKD-<rk?htzz_w%&p
zXw1*r&HB{D77zL#p#vX3JUpc)$tT^1o#-=2*3V8^q}r~-Hlb&c|8ki#u|3AiV}K{D
zJ?6|b4b^))<Uws;rG5aCR8%%cMuzYah}BLW169`*kIC?+93NVNMjg~=H|U)`jEM*9
z)_piH5HUc%8%KE{xgnHevG4<EMSlJDlTK4r93-@~wC?QyT*M^2kldjMy|*afefqv_
zL);Sb^71~4_n^<i<-MygV0bD?B@0*8bd?^GEGR^$o6|DXwg^Im`IYZB%gml#Tw02J
z@jFlmRWJPEGC&O)Zbq|PP>yIFDw+umfrv`)eUvUE6!tmgIJIkgj$#mb?$?*$K$W39
zbEDBX-C=8fXF6)2t<LXedG-Gg@y$YfYU(gXp$8QqBcm@)Lp%m#k+hw%X>Cu6CLCLL
zd3j}T5nz3K?rm_MC)X<5&ZNJC`9z)gF#Dr42s$i8%pg*=QV2i8`mE5&e1AF`h5M4D
z3#`v5|1)#odoV}){+|6$ED?X+@n{2VrF&D{eUFxha034Gutmb*&~;@p&+$>QnQ(id
zR<DB9b~ltt7NM^N#lAkbQ2z2#OVlp`^MQ0Q>z5b)p^=B6-jB8OY(jLo7C!{|#T}3R
zJEW3tE3ZSTBDv;jTaLg?`m^avPddM1ToP(y3l&I^8zPbexU#<hqo~+t8%sI_(~+}+
zN>W5p2GQmO5s||!16>WL^>(Ok7(}Y)5K$Owk?rb~?VeCPHj1_sNsX+%41t${0{su3
z{&HPkQao1W{GOXg6b?=>IR6w>8DM-9;{XdNuFeN8=Ih5UI0S}m4?mYhStMH|qB)`K
zl>ej!1$~lm)N03#BJaX_7-rptGcZVK%S}tK4SWntUZ3L`t}9@c<9_7ATysQI8m=8O
za&kX^P2WCH_E5br{ZDonjATNAZso$QdB;C~#(m$}lJf#iW}wFx;QFFyHVCWZQ7;Jc
z=mmyEWUg5!DDvhR-Hc98cgE94ViSy~FTx+&5>P4*deY_Pe!ilD9xaD!4v^Q-?>t%x
zkMe@lfvEck61S+K%$1crz$;-iXixZ!K-pmdn%%bM8l)wls7!p}E<+qQ2Cv{!F^j?N
z{v5F2=YgFiBRV?Tepe9vJ1N>B(m<8T!d_B1l;NnnQbaHxg-Hhq%O3)8^61ec{Sh4Z
zzQ=XC!+;bCg01g7#|P@#UF5m#x3C1~q0>)TozK!*9)OPg?2dz7iJ&e0{4{)XmpwAf
zBZQ?se*XOXx<8`q#^>=ZZ|^2=-BOI||AmF;fhTf*PDX~$Hb`#^0G*O20Ppla9}mo?
z$1NyYrlX@Hbl`(HSQ(jeNur0`D`I?PXij)IN<`_XsJ!D9*pTS;2e{r(4xWM-W-+|?
z0^_lm^LSoXwkeHp?KvU{d$7vAH+u20rk`ujr?{p_tD}J)pW)|Ib{K>|TaX7b9Fnfu
z@!^#CFFP3^CptWa@|;b>oK9!d$>YKVK-&qT!*{cId4Oa+?l)yoTgZ$_s*YEW@!>!4
zNTP*a&gDWc>Q>jTeNHyIB?-yt>iSuw8M5erb&yu7zV9#y#2=&@zy6H-qQ}ECf<RA1
zhlBJm+g|`HcuKWA7%qN>5rE`ipyod#$+|0dRK%_-%Ipa8xOw9uA*!GK>L(pxUS-XN
zFm;#x*$2nKbNdGd$0N$Zo{#6MVW={R7Y$Qt!DdG5D)TtFnk=K}f=O__MF#HU;3QOM
zb)hLRVgmVtwQ;zenJpCtq4VSL10XBK4`_ket^3b<LN!uuIBXgSk<gJtc#F}kVgiOl
z)RuZtTp$1pI8Q5RyD`q!+@o@A_6VNnid?TDGlo3~@(@3?jQjw99rP&SxJRH44@$~{
z{73B|0PXI5H1AfWX`}&!eRaP%9RYq~&b*%}SglYT+Sc9<7*Q7Lo#>`htQ1DJ`7Xp7
z|MN!hwjlnZW==If4>3OInU59YMV+CwC5L{01Gx-|Dd5;$m%Jq#&l`m5ZKRio@v%P`
zS-Y$?T7bf2xb`B2cg^e-{l~+xUt8Ab(DJ;=SHfhj=!!rziP>fvF^)HpFX-;1xeD{-
zE(MdA0{sXQIa7qlM_hv6dZ-i#Q-rQy6Yha)sVYnUptijrKs*XRgQ-*a8K?{LGvwx5
z){L2=BHtxFas2o!=x^M~jqlG1<wm08j~5#ZDr>z~#47Xhcpb*d1V^an4yVH`IVS_i
zb{?LJmvBxDFt-Okh=CU$J+z0A|E$@9UTA~rxG2a2)B-~MjW@oE!!-Bae^K}7w8+1+
zFejmPh%CL$F#8y;pt9@pt8gCA-1p!K5ZYD)9_1DMSVsFMrt7^0&jwZ}V=})yn{ETl
zH4o2B-O%~}x?FIt%lE-MN3vz(y&H6V@Q!D*z*Og)cYMS~_jAQsTRv-j2UO?n%ad)G
zO|aAAc~H|`FT>u}BRss$2mt?a({IRel$!JbsW`l(9f@A{z`NAA|FMhb@5kjFpTbA@
zSC#r|005L_On#OcCzehIw0z*1{u&nM^U~-cu!N_C+lKO}h@r(*_dDO2scLCAE?StG
zMbK!l4>s=>lcs&VzM`6jb1SA1M!W^T`3!rHd^FJ8oan0GAWj$7=BdU&2l}6gVvk1M
z`FUu}9Iw#a^H99E)*C8@SxoqWl6^^WF|s){JUJ(g=n<zohfh$QBEb)kW|WpVAAl$H
zSD|Q=c(4VUrWLwh`VcQkwc^_G#QuxYLhvosp1Ei?G`qRIq(-0iz2LJ2+%Y8ebD(<O
zk;XO9u1xjI_-$D>J{e{@efl%Q0co1Ga>?1hOtOE8S(jZjPtHDbG8OZzXCD>@i}cIz
zl?Hv5W=9NoOQfty)vQZISFjFPz{}$hCu;f2e}ga!+IKJpWbQoy!u9c@5|b}`yI}T0
z<fjXlHDQL$)!WJ#rR13H)3LoTf-f+JpNh<M;)oW!29bcm$Z`Ee5DETXa~+*W8#kLH
z*=SptSc$yDJK;g#hO5EI`EGnOlu$u8ee9aC{LD)CPEZvs=txH+!Fifff&bKNti;F@
z72*u%J0B`OKZ|V-1o`L>z{*LwI}wQ>zJcZbl&CGdjWbWHeJ2Yk)s4TT>RO#6)jdGn
z09WO{o6E^*uf%(Qocfr&z$#yW{ax0f%Jy3bF6hB`^3phD;^u17K@#Z%)JF{B5dyI@
zjh;0C;S_BGrPP5FoOy^BvTXL}6dAX)F?c>=UVIoptU)xVi>^(x`hPD~F@=!z2yJ+e
ze%5lPM@i1nayI;WwTAN>pvj;Ts^Ku2iQ3K|nEtpSG-Ek_Y2sWnqd(FuRd=qzO#<*L
z>XOYuFsk6o_J3T-{-tBOVT!Y?E*=Jrj!K^w#%M-4lZM_^q|*sf@%Q(srysqWWGgo4
zcaV~TcuVPqto<WR=F4>zfh5W3)5VPqrwn?R!&)R&FLRyGn7tXcxVJH?cUI+0Wi_=w
z&?XGmpD7v6%RonpYM~B19ngVpL5avNdc=Sdb#w->v=`nD6;*MtOkDN6JY0V3lD0r%
z^Dhl|g2{m+@zGv!Tfu1dcWofykkuJ91h~KAIKkR=VrmEX=<Dm_$3W=f`1dc7w)FcO
z(YJBOWluxR$?o4RQMA`!bJ+f6@gu8Sv&Zg2+SFCJv53)QwR1VT9_SSp?w{Y~86Ft8
z!OZO2c@BN?1?rs%@(v#oIsj4VDJH74$7uglOJwx!xO_r3A#u;7Z*Ya~=ae!4if`?a
zY41UH`6_K9h_!Y_svm@!+~|I}*glc@0z!kWj~9zhRLJf@`wXX|x|y*zr{~v~>Sd@A
zA+IrI{C33w-t#WfSzF}fiFSu+nr`U@WfAjttZrMd!cS~tvc-1%%!v~xinPG+QgglM
zAuuNEcG%b9xuao+p-yzyJhx!>AC%j{2LL*yp5wD`W(>}GMoic~ut#iH^!$~Z-v~IP
z5VJeW&tmP`v$R}TlFF|lZvaUOvWN}NQM+6L^LEW`{}JWI9a6g<_MLaP?3_7gLf-a%
zv(aKY%#5ZqqXoB&7zM(<79F$nXU2<94oKNYTa!UvAJ^w15gD|rdb}NRmklN(qlBi$
zrk|!$U?72lDbmV-6A?rodZ{4by87|iW$_hovVTfd1T#9fDile>E8_k5u-7C<6Ru>p
ze>KBz@7u;UHhe?{m1x?_&5zd(@Y!GJk{CAI6A7-$%4BPjQxdB^4T~CA9Mwx}F9Z2T
zx^<ZuJn5jaJ)TnD)ywSq`n~iHkmeT2f;)s<cSc|MLd;UGNfmMxKn4NS0NDEqG^M2+
zJu^_(J;k_^^s858?uo5kyY?nZW4oF^Xo1an<EGL6y~>(<-~hHJ&N0vE@b<84@R0wd
z_W@7z`hy~(`?a)kiieW{r_XCLtNFEstxdv8p|=81JT^Q=rdmvTakL+hw8S!!42uAw
z43Q{8Nc7Gh*fa<|bM+Q#MYG@X_Wb!hZ`Q0uVbWQ!&2VCt1&wvV(28ptH(qU<-u}5v
zD=fzi9!5f<FsfYL@B#zcwz;8@L7qb5iZRpFJCOw_4_|51*WUO14d~b>D8wCsvp-dc
zZF61*p(K9?gytEltw5N^$He3rFCI2fJu49=v2qo-6CQ~<Kd`Jz^PN}l(l$3(s3G}{
z?A}5n`#O*<?B2R{YvR1Nd|?AL`)Ouqw8HBlC~8DG3DZ!xwYYBa@RD-5OkZQ&mFoz`
z(WWX@>6V7HSCGO(l7e6gi{(@?Eq6Cys-TBR-E-~P#RIjg9~@mr#6bzge1Njm7Y4vO
zbwTjzTb*}QBN_w;{8Mo4NxF`NZwgjai#wXj<qiu6VRUVka)l@F5TF?|;8-D&Moa~|
z+l{?BNFbRa9B&X8M>3$pVvCvCJH_gB=B=ioF-BM+;`*75oTl{ayV?iG0v7#8GL4HD
zUm2*CgV9I)9&y|6>odr8!ma#nbM=UQOa<6z{=5u;(|D(d1pJ;-C8>%U{cVB;w@G&~
zdN;pv>I}_y(7Sy&aG*IB$vhSq_-?96nL+xZbIh-vqj39CL3&j?<i#p`$<_#Hf7Ly@
zhMIOrS?V>Kf7Rnr+<GOhv5`>WiarS|;*ihm?$!VgWbK&npJPWm2IhStJK_CDfs?b!
zaP&pAETI=Z-dZdBhO2N8+A(Kqt%^J^g^KBxb?eIWksa-O4=4M*w;^u`wiJ-SfBVrV
z+5$;h2>MB_A0`{J5~}5KKjr8^%#yY_p*llGk&Gj0`ZWq=nYCtSq-G&T|DyU$Z|3rC
z=)`yHC||MrT?pY{CfV9>`&hQ5UD}<Ttz;#pM!FU`oV%Z7!l{Uv!(vwN!~RWYO9#rC
zDH*7PO7`P(GuqqR176Q_27??z5O=slCcV!xn)PN*QFXM1=s<>>YjWu8UjB>9fGnK`
zViLop%Kg=wRK@rG9W_&a$y#)+CP)vqW_-;9!F0vA2-F8N4z=9Jv#=Sub3mV+r)wJm
zUJ?ty7eIMC4;`q>d$fY+QDA<aV|3G!#LK^1%Dm^b;oT9JK1jS1d2TXEg)~*7ZV1l@
zW<R6ttD$=)5)jGenQ3iWmf;`gg*@v^3YvuN1p!~LoRxsu&Wf3wXEnweff7m@NIrI;
z{)?P$>Bx$}0a<faceLX5l5k~+yKH4AVRdxu%iJ(5!0TLu#4ljY5jExXe$6ScQowg|
z$6=X-Z6dgO?>ArZkf(4cEu}1^MM!-bNj3)m<;tZ49s{e9xT1s9g}f8K0%P-?(<!|*
ze!_$uHa5U<APUtSKD3cL1lS}@laSsTw!iFes$>i9o@E3Hf~U_rWSEM~N{UXz3dL3s
zD=iys3W9zy&hzxUfv)eq7B$ev7A#)eJwSJ``;V_2L&gG?G&QS=Zg4PgzMVaet<E$u
zeG)L_J#Y4-E{{9mj=xG(oH}~4(j2Q2S8`w+j;yXQ3^vK{z$6t;J!FQ?jFh8}15)Sg
zrfNG!VC><=Ck#tMXN*U^pK=68{Bk8=a9UDD*4B~YW|k`hS?k@n3vB?LS9>M==hs(<
z7(iqg>i~q`i)3xN<8Gf<6x>1Tn&E6*Bd-VY&lwgze+D%Re}lA(BNu~eLc|0`iUzTr
z^`QGZC3!NDC^Sy*A$1`Ls!*zdpHBvZpz3Ec)b9~z4TasNUFA(iqutt8Xjo4qJox;1
z6%IwGYlHngdHNJ+9U3agj%%M84bU{~Zf(CWq^&X}P7`eZoVMn94RSuNCFhVFRFk0q
zap6!<wFpM?*KorNbW%0HyZ4;atS%=!Vljt>0agI{GB)vpY7-5gwdgMS<DGInc{w++
zKu3KV9b9+>YWFhf4F;j%ROJVF&n$+8otKS}HztidPYwDTdDz~+{4oNM0vjywyM2A*
zxyFnC>ca2U`t`k|sJKnH_GbJa#iuVN6xDVOKW8qq%tz+I2}w~0pOb<MuU=kU_;1WH
z)vN0s3W_)m8}4oDEi`7_)M0B^Rh{tkwVY9WU+J0Rk&H5$cwpjPOboer>iVnS_imKA
zXOy*n*j(N6e&(JvUXR}dr+FhI%pV*-N@OmW#QF%<`Mh(L;;&pm3!uV>=RQb{0qpuH
zDf&(4_J6!1GDi?9zp*?#@AQ~UDCOW`TsVJzyh%8+jPe(rgpOk}mGCaJG);xP!ks>8
zAU$mCv@8@yQ;vT60P2|i-=%oK9;Cs;<Dh)*rrc0T8`CP<-t-SAb4wk8GG5M@7|x)u
zd3_MPTOdJ|byHQq9M`Zgv6Q3MwG>N~pwY<!wKQ8(!ievtX}_8=6TlM*8!x*mLTQe6
z{~wL{D84p*{rYrdSQsdr9^VV+0Y6g*4^Qs<WqE1iy(8>)E(k}i%yMIfTi5;>j6vfJ
z`WIa?zZ~pP#|@-$JX_a3d*L>J__EUX{1SXR(ZhLozDtcui_`wNXHED9r2M4<XQj>r
zT02Rof6$pW|CIuiO2&86y%_xJ(b22AC{8sVCHmn?u`u(T7o+F`oB!~TFgR5)?OjpE
zR#h`}h~($Je4sFThFsqv;Sq1wOYBdJduk1`2d|%B@_mH;KsLxY03J#CBi5x6lQbhA
zTc_CP^rS6bEZmCGAd3zO6Xv36JUr>)D_;lmolB9kQv!Bp1WoTAO3wM-+DtMb3BKp?
zmo9B>dXWIZKn8<kkCkL?M^fzp+Ok?;j}A#jaUS9|upN@pCKf?FXa6p}JSn=I>Ku-N
z${nRVk*ZS@MMRKn`h0~g8X;B~0u7+=s3?8h`8PcLr09K=3lUg=*N!hp<ZWNZmWl+7
z-%dkL@RWfxF??|CL!mRHN_fv5Q04{m*B4$M&rF}=cC!Q*y{)(s{G<WR4kNEGUbA{N
z5;uD|#7m(QAqm_jG0jL(IP&J^k;aR+Q629t#4}^RX`}G4lVdNfMt62?#AMA*$GVe~
z_dfpIZavg5542t%-|6vyD9aUz<GqqwO-C*(O-o%>`SK;nc|ZG=g2E}1z>5E}-|tnz
zRf&=fP8g7=N#hkqQhboq&yRryCP$8v_raL1&}KmL?$7;qfF)ws(D+h{UvGp5cojMQ
z6TuhFGaTPw1j8)0j2AQ_V=|^j4<3h{@$LNR+VNC(5{ZP-@ZgImS6$_JBg<u@%IF>O
zT4SR0fdD2{=+;Bysm{R%>>_vJt31m{0Z}1B6At+hO#u%+xY=>H@YRtpvf1QNLG@xE
z69SgkPQ9O$a}?OBP@9x@fIm55AdP5<phQC^L`<PL%_}!?6?hlptM(J%J<ew#`%<)f
zt>JqEVI>fy51i3<zyPd=N<W}D;92kXKaumA@EIdJY&j%^W)5Wt40m3<Fl=l4Z=>3o
zRBAZ`aWN{NvCsCmgZR6&VL!*gJl0A?D+4&)A!%sp+G%SmFRL=0;u-sWm)7c9Wtq}N
zh_E<xL01}~GJdm1$xU@1QpS_UM+j1!OTkZ2C{IKVyK3h@la)%Mjtyb*oMCY4S=1kF
z=7&NjBh22p)`3QGKa*dPJh74OT+_4?j!lekM&gvA5Umj-3>#WlL~)=(gfG%7XXqgm
z%O?r3ER;l{BL=)cKyJVJ21ZTkV@K<$%#%0aJK*0GI#4BR`zPGL?~Zgc-WnV;51>A+
zhZcUMuun#CtR@&)Z2>Tl!^Az_N30c@P_7wCvWKV=JRO-wbr4TbR2zAH-hxnF5Cm{M
zoBcFI(1*ds1?zPGcXkV+mnTUQxfJKI4%r>Wo0(bNP1iUK#984x05}0_0E3MzG|H}9
zQ;sg7*olETX0C$yVx*ni>C>m7JP&MOo63(fne<0E7nJos?O&8~w1whWT${FFszq*b
zn~m0=P|h!^*mT3}4Inp=4pJ0a`y-NSE9;IN=|SFZl$<OyGsflWj$j$Rvm075vk}S&
zZx|fc(~ljsZ|83HG;|1~yX6KKgax7;G7|qC*)-K=BjgJw9Xsk#$Nh3$XJh2ZA0&xv
zS0H6s2*V%3_Jkz-NOE~Jl%j|T0cH#vN`K7#W;zUM=-&M2n+&Rz?i^+__SVPtF7&_P
zE5HavM_?Q>ffmW#eQ8{!=J;hT^u$OzX}F=c;dgOdO{@^irXET*ZmQ+zkwLdc$b^|x
z2V3=Utkx7Gz0%MHg&farrz4g?XboLPrr3ZQLteI13(voti`5KFK>`)GIMHNv@P!Mg
zJ3c$l_9%9aM8Kj2Dg~%C7+(#CT2EsyimxiV2U{>?auE`EoKINK1}JjmR3K?&R0VRL
z);b@lz9)|`!QgOgMcyF0m${N<G_}L3NJ0sJ4A}=@IpgHXVJVW<6iQM0L)zQDqvlA8
z%rUa$(9ep;FpqRW_){!_=DTUfsQ*2x<M(G{=6UK-b<hWa%9z7|R;ZI{7cjy=SPaui
zrTUZ?NG0<9DFi#}Q_ic=M;>rwjUML1LA6dEzpsr#*r$`8^ID5+aa0|aE<MKi`w?wj
zg!5Mv>S)$17i9gP`U+6k@QY`Wi{V!a02*cihuI9Xz=Ql)6!<V+7@jZ^%nN0$>M7n0
z4d~{}U>LRi>^w+0%f<Z_hj{x^2uEus(*1m+3em78#_@LLY{3_Ygy?8kU6^vfXRLM9
zDNZ!tIhrbmTq}^hk)FqSqWfTUk-Q!mrB8B6m{pYIeVk$!`!WE`7$rC9FYR0*M$&bW
zs!mu11vVIs>LhO}L2o5xkRg)`5Ug;1%k@G)kOA|>Q$c}DUdD<d406@&a&yo37CLe8
zG@L*rFGD#mH<6HZh3ZNno&x{#G0C4)r_qQU(m!)1f=y#Dhlr;xA?Tvcy^?`N1n@CT
zCxb#VPFB8p6@{j-6KTHSNl#2H=6v;$tU!baj?DTJ?#fW{8(IPclEx!iImoFCD|gx2
zhNwF2>U4~O(f}Euf^0=G#S+nJX=$kAoCx-TyIh0Wxtxa)20iA7E>@^QAoeghE^I;q
zg_RPMM`9#bsuex#oMyM8d*kNKW07fcQw3u378qPoB#%%W>8M4}Jdnpx&4NhIz<syj
z2q02NAf@^hOs;UsQ1{Woz<1+w^C|St3H7$eGrWe@5%$l?9O7>Gn2g;*4<1Y=khEW$
zno4{XDWZ@beZaT{<_LlziHwZnv=Da2gCh~0T@;R{dpv3Ozo{IdcIXC&?XH*z?x)oB
zbQOr%?w50pNiL)~lN>w31*7UHld6E0$TGC4aoTRBCg7#invDQ9wpkl{jN;YZ?Rk<z
zTVrY78Ab1k;&7~1h93m5<YbE!5{j7JY_&GllVSxeB$?MBdK=k1a`<S`vrih-C$lX>
zRjQk&7ErW58kb>IH*~~P-!zKIs3VF-B9}QM(fcvMei+*c8IKu^1e*Yf#jpUhUt8;h
zf8wqXy93a-czTbLsjZHR<CkZ!L!1jas(>VtKq)jMzu>%JZ?<hpHK!00x-Eirubx`8
zu{JzlCIK+s^FJxViY0HU-RWrc`0RA#(3NGYWGKA%oIikG%n#MQ^t2Y0<F>A>;X{}+
zYIuZJ78@ct^X<(7-16Rc=~Y!GDKT;s2B#;J5X#@2NUNduSR@Z+AB`aiMv$1{2u7-?
zLacn$ghY%4wxFy3Vm~7;QZuvCP$k%-G+8q;y*cgeXkVp53S~fcnXB{zeSD|m`PiS4
zLB<aS7Qu5yjc3)d7(1#b6Ha0tFO6Dc0keeJ5@PiORiaZZQu9Mg(1AI|S1F(3_oK=P
zNclvtCsEAq-3f;{HZhZw=pkcD;pX?<?kz6gMDgN4A9l@aOf9Ux`Z)F$>eg9UXkoA*
z7C@;206pQ97`-3dZ2xIDQxp7M+h@_aQF<8h*|w7Hj`<Yh0nWu7GAx6p0Z8o5O{wl&
zrR$vi8v56qUfG+SVE=!cg~1Vw3<8pNlBs|MlXhS?)z<8mBLv1}tt0!Ho3x9xr={Mw
z@f{s;vs2r41KxlrJ)2{xx>Up?m{JR`w?7_<S;*CWKkZ`LEe?VKuOl_6L4I~&OZ&JK
zW`-R?y00$ZY_83wR1>bBRK9l&Mxebzr~8ud`>0C8B)Cpv=Zco?MLiK=74l7lAZwV;
zwFa)=)STuBJ`b;jH8TR#>+G~M2(f+l_J0V?xS*GH-af*d4~+yfPUht<E?4DhK_A_d
zgEyRu(0PmPgICxJ2zp&7pWRCq!S4HJ2ZTD%fD=l1?%x3FPX!s1>D(UiJ(at0ghIwL
zQyE~vu|w7yU^vG!=>sI|=P95ltpAvl%I}BE%$@6wiu)Mfmj&F!cW$0tBNBE5aA<qy
z6*LI)>CU2OT?cJw37kjBCcX~#mX2<N65maasMec$=$+8(ui0>#whYaO7!KH$;oJ)Z
z*8`PFa8ycr<Yi69a{a#7X>q1$dv-$3F8NiX(haih&AbX^;(JHY6y#<xKIG}~z_sS)
z8dmw<u#Cnwk$2E3nGzGk{ejq-c7ry2&dt)p$R25o@CH@&vS@=(BxB7hWGfXBfiEN2
zXha+YFoPb0B-=4u?ORM$Ff@)r_|st%UB-+VbCG`;8H~bt_EmUe##^>bgX#jsF9vb!
z+LWro^((KY#gTCnmZyKCj_o>Da4Z>5fWOI@v+_(=Wda7UA?k#uOPIHMjmF4vHBxgg
z+BKqS6aqN1xrIQ@%06cg_hw8+;rPTD63G0RN0Va2F)=&1i)e~mHi%yNo@62~?mSv>
z!k`^FA)cn~ESwLCITkaN>k>M*<&MBc^X<+A%-DH^;N*cz*p45o{zIGLZK9a`INEj=
z*N1y<&zeHN$s$H?dyta;HBLflpiDlK`8%MQ$TiO+@-WR2u7KBx_W|x5Gs<`*F^xm$
zOW-0>_0Z-I8{rp+Y){+ABHTRm1N^M82~W$o3ir>q6>gdyNF7cvsYL3l2h7DkM@uXj
zv@!2pyLN%vOBlxdedcfA5=(roxw-`xy6qZK)1UbIxgPiqCVQ$3R4PmAG3CAt84v_q
z&=b|j$QqYNa23glY`N>n$pr9mX^gp2_;nb7VG{a-_VoqX?RV&}$74Y}y?fU#?2Bvs
zcHz5SBY>=YU+=-ks2t*HJ7`}U2yhczCP;0CDDGT;<Vd!0IrsQ$eU=<X%&B4E;z+qq
z02C2cbzbdMGXLku{NZE>hqv~EOFrnh$I#E2(B(k)9gA_{Y9!;FO;yoEnY-yb(%G$F
zmzE4#rEy>k6f=qYDo&=Z0NkJ@ctTYnpxQ*ixF7Lmj7woZ6~zQF9?x-8K3B9IGtn&9
zDlX*udQ<nV0We+obs?cQba(Z}zzt5FqId_)K}Z`Q4~cT-XkRI=rsYDhT_ehA`e=+?
z5}poaym|BHKaJD(54yb<Vq#)I`=As7$jCzwH<b8Z;eLBq4;{#SyVJo?Qrz+n870NG
zR7)0Iy>@LfiW$(rsF?%Fswf-g&CP>0>rRV<K|sFZz@?#P;NykF(7wY25BS6Y3^PD2
z2iaB>Q82`9=O)FWT+gb*ZVlm);7_who#gjr_?t*$_F1d(UW)Ok*!*m%(Siu{fw4j-
z_;J^CJKk218E`-lpZK+xKyp!)5uXZ2g=IelVnc!LvBBR!wc(7oRy92@j(aUCZjY%e
z3R=U0;iWa(_tfFz7#S~y_Fc<gF%GHSjV2~Ib5t=W!%a3W_vZfoQsqcDoKm=4>fXr!
z(jfO@F$H^~!$Im(Grrm;k5S1-QWdWm*_`<@V&}M?dRST<X`HZp^{ZgAedj%1%o4rd
zSIRk?wgVG8;-@|V6mYa}0oRn}g)o*OuzLeT&`@)5IHo3aGu`5<7!ugIDX%j$1nuLv
zV=?QvKN4$G;?E3A+e136>+d4w;aK9Yz-a?lH}HbLT_X_5;;)m<0dsb@g(na@uW#f=
zY;*D8$m!86t|7*n!Zb$=81ZVTg2dCF;eu9TlmtojBMt=N8VJ$mA<joW142=iq%wi4
zDVg;;_&$(CUU#aHzUnFFTxH}pRKtM|fbs|I9z$1Zl2mqbUBvE)X;|53f}XX#w8_wr
zjB2e+9__8|d0wZl78|=9IVEHQaKR|#T)TdKp{%|?*E=??37%_wd*)5OOo$aNpL0K9
zhMP8#hHDgHo-TZa`JW*ZxEGXe=-IEq03!nIvJD7+Y7M8qZzW^q`E<_#TAPu!#B|zC
z$LhFOmF_g!H4-Y#;lZ|`u{&KO0;z30-8YtGRUE63BW$3)b!`OKaPnSSrrgtp{ji58
zUfKXA55Ml`AA~>>JD3%rS-%DbQo>v%_tn9>aRd!%dE_9HlYjPc3ce#@W5uz)*>ezN
zX~Q11a*Y;tUw&Hw_Z5Ba@z3g#0AwT7E%lr`N!UCfp6cz~mQts})rGlJZr8XiKx5+T
zu%{O_-p(6^YwloPbAByQbSMgvQs#vVgo7(dWg2&h*=K5EZhi#Kf+AHH5c4$K)4ePv
z;(%u{57?&=!Z&=37z`<f{QPM7=KyyzT2+_U0kTPfIZsF^t0jx@r9z}HG#eL$bU&Kg
z@d^}ny|{|@vhWF8x$kDLX+|H?NW>%%;EQWWdX^#HX36g|Td>`t^g9CCOtdir^)Z2$
zY9&`RCJZ~cov89~a>7TLcfAjtVmol|h!o&#rYsiXn$CQcmKKM^7-GXCmzrOb0SB8m
ze^y5NK)aZ!sH}b=nU3-zVnj5&*sUGUU1qQje&&GXLNqd1X~}S&%RN}Xrqpfl=xs*(
za<5XoI++#+Y<IDB%86jBr6)(DX8Pk8r@YXf4ho{USDAJ%eOp1WnK#G(A-zPe_&L;N
zXE@X)Afbr8!J<*)K|hjG7sXZS-U+g61Rfdfvm>uRB)A1P%@)Q|{b$mdX*O3GEvZ-v
zew5Ol*Or@TGq~E?wA*r#D*LP~fxpdc>7K>$lmvU%IXV4Afe#miK0D;Bt{Jx;;(p6k
zrrntQJv_}4{VBI<gg6w#hqr{t9VZLt=Hx$9U2AiBQzW_i-mj$6;x3r&Uu8Ko4Q%f@
zGdON2?@%|2qEDVY!NbO1DFRaNsV|+pY>mO-mEApWD=_<Gc(4T_s_60NP?!R|10p2E
zW7a3Q>Rj2H;7{#c_z@A4!_ZaN%`;yH!24uLzk@D#95yM1<5^?suEsFRRT!GPJp%>5
zo}-)_DXb@BjF8qWjoiKV9Cr=%^lb$O!U9Y!Qgskr)gG;nMBEn>!#JnUFdnb}ojYXg
zaYoV!6GleQ=3cA1?qZsO#>m{X-H0>XH90Kn@%js8pY1amuRusvtt!HJyKS4wxmwoU
zcivW*&E1Urwrv@FoNg3jJG(HA>g;VMa75n_<sMR?N5u$8^vr|TjUREX#NVjgyM{#0
zx&NUn_j}u0UJiW_aU2>Mz*kfz%Hd1`D<87)I9C_S<a1gF6f07)vr?;(;e3wjBJChK
zV5UJ6khc4)>@#69SNVT=PwohYVhqW=?(yfYNBhG8JMsm#?e_re+5hOe(tsMbE=)2{
zr6NS+R+LInktT$zfl7%+Qie2XAe9oLOmPVn8hp_}AxU$HNGU04&QJ=KXrk$R_BqAr
zJ@5I^KlXX|-fOS%S<gB_TYoZ@th~6%$Y?*Bu?jcEqTGd)YA0i<LBn<hA@^H|AC#1s
zPVj}7b$Jr>hTyiN$!{c0Mr|@CV|IFMSjFZv7$gLk2`HfRo`ij%e@EpHA~7VokvQYl
zK!&IH@O>GwN&_L_vyy`wgHg_c2K2RG<BA7uq-1S-@7lqREo#$QUgOG-?&xuSOkij`
z(5y5{U<B>SZc3D@!thL}vhla?k~Y|s6eY$w`cs+qXyr!I`i$V>;lnPJ8y7_t)4v46
zilWf|p3gqB$RwQ}bJo4+V0i9(sQ*c!>yawo1~P*kf%dd2>h?~P6qL_NTHQk)c{L13
z?&M;zBD<7aU0u;tF&Qg?gcDxbjzCLJEB(f3;-z;|e)qO`x8#sj2A$?z`UH}8d@@!U
zgrw0Irs4u^wMqxTio4|*U`6IigBpC-s6XW(%TLeGp=(ht)`0$XidGbX6lbKuCNdbH
zN1ye4@;?Gd{S^fvB-cTVjR1RbY#VKId(Xs*BA}#59oDEF`>HA6W^rT+ulx;fh&HRp
zX`~G(OJ_ibo4fDcs@D#LLRSz`5(6O#PjK0gpeE`9{dct5H9{5fZPwEXUh=|W$VZ6O
z7l44|F0l3mVMVkl_PD1>k)*#Nyv{)E{1)h;nlUW4C_IwTh`dkrkBErCP((ok`l43i
zw+C%tx0<~0rdQ8L;cg$+O&IH{h62Iu4ynC=-vzGGj=)K@QX=(G%5K*#LI56~Xs~y0
zD`;$b)XCKhswgSxOE1m5ei7ILQc2JbL5Ab#jeYF{7l505t`=kTi*x2A)|a?PcjHE9
z5XPc%LGWq#>JUQ{+7PQ$e?W)xb}1HS{;HqJSm@7F{CXD=K7jpL2)upwy!*WpY~Ii-
zUmKI(uzDjm=@IC7=qT3nHH?ml`2tWMqzx1maR_T-AH1YXkHxKi5Qzcda4V$lA0-+b
z*s0zntjBY^kaQR^;kt;#5WeKcoTiumb9Rl0Ie6cxtAO0bYZ|MGd%GLiJSIjB-QqJ0
zX40lrN_#4bP)BY@s;jU$8Z;Y3eW#+F@h2|1zXQ$FOr43AqsL+8v=Gtc`kITcuBj})
z<K_;N(mS8^CEl(?HDF3c$a5<>g0_yLFHyw>B@x|?hJ4BUuJF5!baYM&k7Nl0!gUCq
z9Uj?0d{gp(R%YaXK4`P(Z?_mBd6P6-$RG^<6O<H+6_=KR&OsD*786GT)cW4tDyPNn
zCD&J8w6DoX$JzK^)BywYGx$WkBCOYB2RpnLJ=iF{0lG#ep=?Q{53KW(rLM|uU$gbi
zeaH7J)kd>Qam76pj4;~-#*t)(n#!#=*&U|!%o6Oh&<10wudlx=X9-h_e;&rF#ZQuU
zw|I+8BrZ{{oFuGjyXkf)r=^ew_?>llHvs2e<5{If!G<-ud2PXM(@Xmlf?-oooPsu4
z85@ansQ-B&b0jwQ=T(I~z<3wz2OG1`__gDRO&zCoej^%KbjQdTl|Nyix&Oq~;kP|7
z!G44{|9n}G4%;I(`Sc5HGcm?gq8C=<>@MIkz8aA;8MI0N9tZD92f`Z!mlI=3=-<#W
zJ8%G-&)6(ZMu)`@V-c6LGf+)1>ar(R8u~>s>Mgxz7o8N@lF@Cy_0p2PA2ig4v)*s;
z5a=2Lg%CqYT9id;aYfO1>4#y3Qx}|nfo?l-FD}-5x$vUS3Dk-}oe}a(AisB%l+%lU
z<AXLbJ#MS73pzK0)mT490(cElgoWM(&=GtlCAW+PXdRdv6&DqjESVsznpXHD3cCj=
zL(!Xj&!7j+k8ue(2uT5{7X+vbnc2LUcG{BCo)o)AP;HC|xgKJl^5Q#dlvUJ(y7)pF
z8Mq9mW-2xl_*#i3{jOd=SaRtlro!=;PseNobm^8cRvPreEp@kSA<#p}KA1a73k$by
zx=jl)GUc<=0ymQIX2JX`%ierq{Th*0bcR|dlwl13qV04!EU?j5YY)(l9Ze>?42$uT
z=#vXpW|e#cb8z?#K-fw?&<ibsM!?TUb!d<r=RVepey0pg4m+(amayx!$Pt?@v?lg+
zhLp+x&QZ!A!^{ZIaXU$24*F1jFxHukH$tE;G4l=cTF-1roI`unYX%heTz8<f@<}>&
zFtIgveS3R5;gN;K2XX`6_l^>M+WpMfZ-Pd~Y8dmCso@}JK^-P|8S9zkoe^5*1ZG1-
z!k|z{w|+S_!G@Q9XQ@(ThQ^4K;Wa(=ildO1|D(NmlEGUw0i8BT;-G{N3mRnjliqL6
zwrd<jJBqrX(^5#qy*P_p_MpIYg|NjpHuO>C>4#G3br(&zy2A(o#eG(Z`4|0hakp*(
z|0E0cLQ-ANr&kbnA4FOlY&S%^d$>*^X_V$;b?&m%jL|91Oa~EkTU0&0hS-c?py!jg
zk9=W^SLgEUtm)?3&@_;={e`yKD9|77Jt^=?ZXn4MsY7tf-r}JYv)VMzo8Vhe{#((>
ze?sfR=%F49x!5|q4ZN@+i&=CGmv7#cWx&MfnVR+>GcRbcm{ut)Z?|iNlOx!rm;(&d
zp69IY0yomRrU%<gprz1JfEvxSTlCFJmC+7b-UON6p-PbHRxHnm`!NbGU>Xue9jH3J
z8#lLYlMXFRu9-h65?yQKY(3-BerWOyIaD_T;M;^%m6`Z9QZ~1t?>f5qP-@oV9ope$
z8^q2NN*x_RmS^Pi@jgVA*=V!ojL-p)2PN2u(B99@dw`C)?B4p?8Ch*X8P;<Z|HK~7
z+@MVx?k}n+LhBIvfBB7D0lFrwL(fQRb}$2<weD}X4LV8-&}~!FK>M!OyPNz*>@>1t
zxni~3^)l8JrlS*RIE_tBg;4eB@9UlzLti_(TKV8U>c(iG_ElI8O$6p^1rruS!oB_}
z=+QPTzeteH1|MlH^zvl8Mi?OB&0jI;F>}#o)_Y$68i1yk->Wpxg^r2A6JmI26V+F<
zAKV`RL+(=oQlRqg_T!IObC`}_P$x%D99cFb=TZJuh@~$b?VnOnWPvF|KEAIYn%V-a
z7ro}(%Q`Z630WnWBcwg@V(2SJR~sI*f%=Q^>)p?7EQ)*ybIe$3Fy-yD#(*ulDT$`l
zb<2^1CrI^oA3^?=^DnD@+GUDl(onEl>_%GMvVUbo5oin0j2l@v>L#<DXbo%DTYX~`
zTYR&THU~qNf{!77Cj&e|*AB4COreHq&-W+jR3)jP4cAO9EF3n8=hJDJ=B}hjK<iP%
zT^0UR2Mt3&^xIrx)l^HO(fZ1?)|g^qqsw`nq-ys)TlXYveZ=*IHRT)@{RjDm`Ct~)
zi(k+N<;!2$H3EBLNc?UKXB137J1t|ZGiYhU4iG>p(_+KDKcY02_G#xJ22ijw-fHm^
zjom^2us-eldW@fdXJ?gAhV<>sVt@Kq0PqU*7!<4Nu8oUr-`xQ^@;uqlpAT6E*fAxz
zSx;NDu57i{^2W5hk#Cm}4DuS*2Nhh#7S+rm?G0>AfyvwLwbXE(ByF<78SwR>xvZO#
zvP9Y9$f$_`eUo$$>Bu>Ao-P`xlue2xo~`Xpw7l#-#%c-fUtp`+l(f>?lkpB(M)Bse
z-n}&4`&3{QNmvioq10a*Gx;px3Gs=q&#QCV;i^r0L+|Yxk(%DTYu@=6R$yzHNU(T=
zu8_~vq`%=zN~72jB{AAF$$HR+4Dp0p+kru_aUZgmDTJ*Bd1L@j(Y!xOfEMdww2M(a
z1<TMbUJxktz*h(&FucLAitBbRA=)m#s|-6nN`O}XV+~R=82w-o=By|xv0kom2)a8Y
zQP{@P-8KEyVdH3JzWbHkZe+DFd2!>ek>?pj(^J6k{VB-#Y?>chfZ-!j2&1970I;jM
zl|g^G`|Y(X;0|Z?)xlFxR2=d#H_@Y5W_=0D0IKrAwkM+!ek;(Ps)GZ=<!=8vt?RlW
za)XN*>n!BTXZCLH$|tRgKx@If#CX?K*RPurev32j_(OkLYU!K>FBiA=cW4(l3aP((
zIA)FVglF?-tlH`?JLRLzqS@n4y@+|;5V7M(T(F^+&6L-ItCVl8*#3`Tfid^Cr+m-&
z9=Qz@*v2Dp>`7Lews>^;?c3$kQXTfiKIq%z+AH<)fdfx{VXtXuK}Cah=i<(j4l1Pw
zrxP#(nxenyTmWIwqreSX;QPEJm`W7%su6gy9U_H`8Ew_vF{WBtFA)}onqS>zZ;#f~
zgq`hs<MM_-Dr0nJyPz|>TEicFOO@g`A_>8PtffYzkJiSGZ$=Ay)`pXm{dV4lTtNL)
zgmAd>?-|W2o^I~0mdGsuu-JA>^#dpju7Le^$VVNtG<HXLEq+5vxG(;iAY|;`G#~u1
zNZ}KFnJ&fwZhoEV-QV^ld-u(&Z_COWw*)@KsLGufMv8`O5Pf>^iGps(Q8@G>IPP!(
z4~wo?Ltg`p*qTk!hIhxfSPuX{)$I;|apA`=Pj}Z37@po5rfC{9;&1}9%NXpQ>Hbqd
zr2a5qgH}AOXZBgrf;)TK-%rv7aF8cz87l8DRZ~CQ_VNb>fnx`wjegd!VRsbuEWx<f
z`T9OW<J(AI#4a(%^fea~laQQ|&;+Mehe-}VXt-^PQ2@AB9Rf)D_^AuOhtYq>jmT+c
zE_#+}PDq@&Z^A2nZqA)D2*e6oBXII59Pre}5`VZ^sQr)j)}^~K)7{3ukqIV+_M)wI
z&Y|6=;kw^%QIAIFYeNH}$0~~D)6vn9$>_5Dbn_G4g$3S-BnX7i%Q2+zHk+S%Jg#<j
zb~!@M+PL}RC_Pf8d$*Q$A%e#<OR_gp4UJ#2VK?~LzkO>9qkPu2u~tujHz##eNC0-x
zgTHJ~Rl0pv5_jm$`sGrx+GLuUiYVmJN`}j95U9J8E9fp*rC?TWV2X&+mneXkMtUwC
zSjuvW&!l0)-i6F3a}PSfAoIy7S#$dZLj5?Km+Ubo@DDcV`DL^?Dagx58i}u8`u^z3
zaGs_Y148Myj$3Kiu&oXeSIAo_&FDzXo?s1chJB4|m<Fq$pnz{d!|0V=0Z+kev($r6
zU3);P>%ds`obZlbzUGMu)Ia^m(X1mcgNUIn_e!xUpsfA-0~-HB3h|h`7|ljB6o8r2
zbUtq?-H8rlcMT#!{}>IRXy<dyjg3q1C~vFQkiJA98y}T<jOq{7e$yEy>p!427K63h
z9f!&1d>%9vZj1#aSQdDKa{pEHt5NasnQ*?3C{LtWz*1uBNyJ!gmX`N0O>M(iDDW+H
zL{kt@@aj2Mn|JQ~yx``Q|IDv8l*ZS9t!dW#-E^n<My0dzyx^rPs9nzHJ!>bzGv*b`
zl|1O^=!o~Mlf-KPAeaF3gS`1^j)c4T_3OPLBr}?E0Vy%0cRe)KANZ@!S<14jpT;I8
zlb3(H1Cc`qkBPDIy<VY#7n=PK7xSsV0{2}5IuNzA*P;6F88GI#GVd7f&wux&aR+%x
z<YkDkE&bpJ4`N5xT)KVfczAS!R!Jg%EioFPMcXizx)%3co01p+p8f|tsT1<Hy82p)
zcrEjB&Jp8nV5lx9Xu%~IET}bMEok&U@(xE&(vV%9AxvE$0Ky3^ezo-a=7l?cd`)Gl
z(jiit3m%y@u>89h$|1V}tm^?ybud1#p7xkJ;9GF%8{*fSqHj2;r<vxYk2(>(zC)zc
zZ2wg}1|YZ(I}d7N5uOOJZuP0AlF@2bEhp)|BL^Kv1TVFFh8(<1GC~wP<e_}XF2MX7
z)BRIYQZOIF`kM*eLTlaU{5WaRgApTst1!4Zc~FD$ju{RWn(#_VGF|C)VIi4Xw(#H{
z8VNq^28<J$l5}@Zk^}2V(agm#6Jv&6cecTNelnV<?TZGCBJzn>g5tl?ToF$Vm~8k~
zokfA|4Mb@5&=1-M4%%M8b=Sv}M#tYxNj0yYb-IEi^;I=D&-pP-Gfbwy97Z$|d-v|$
zqOR>A_a##uK}mm9gv+NEg34@67<2Twf_)_!Aq5ef1k5Y0(z^B~3<$BFL$3T$5)hTy
z-#)?tP?Q=O!Rw>$$f9T*5r)o&y@y6emp#IvsCI`hG+V2OGTHOXGoz?@J6HnI2iCva
zc$~((A>YJJ$SqLhf+Y7fxAetbzkdBj<_)GDJ#<+k3=AqAV!5E;4N?UEN%r%P;m2na
zy3#2_K8Z`%9em(Z`s*V9YdP5BA>TU-TAmnk*5~>B0fC<+o5mI0*$#qkgy$AesfRS?
z@LU<R!~8z#FF@HEba~zd0@zRN9!a~-#{e}|*+w@MlytOwE|Ucatd{wwN~HW|r<hZ6
zB|0heuGc^*vFY&%CZMsk%xf*d2<QSxu>qubJwuBVP0l5}AN50=2tt3L*fyQC$fsK)
z1l^XF&j__wPZ>m66Ou!a-X|D;wI7w1FDJD9xa(1qgu2kaSXya>_r@kUv4kr;+vOwj
zzS6zh@iE#%KPI&B?OO#fm_hm7+4h9c=#~FAr+MC1`gb??Jf4=JyE{(L$fy(PnZUkv
zkW4YW3vod9Nu$PmE+41cw{Pd?=R@g$PRHTho$zBB7wE1VC8@{Ry4yipD%^XX=BvKo
zp32aK3jgH1m`*uZ6K{LYqj{_MZeG9s5$KJ$C9go=0?F-b5LS>H3G=S*NHE&WzbGs8
z5g{tnj9>4G5l=Gw<K$aiX;l34YZVTeDzY!Tsyj0}TfN`m+}8~`_c_~8-8_BzH1-Gd
z_|%h6zw}Gf9hUZhlGl+XMPi^jJbOa%)F_|zZz|>pB4db68kz20-R{jxvqQOhBs8^8
zEh1zXo*!sdc^8R6qz6LHl96|h=`o?PIcG%?z06$ZWEJ`TDVlW%W#on)N%m7u{fHX!
z`W78jD+vdFNsj>C84>?O?ASE5->(C5`vurVph|Dy)SdakfiUnDNi855*^+4Sh(^Nm
z-T^9ocJ3;P(^7z`fuhK`&jFeM&d8jlr1gOL_wB~Ya7T=}fb-RX`iNR@<A#!7zv$lW
zsh85J3uP0!8aH_Vj_xO9vFBlf_|TI?ny&bLvYIOZ^v5uPYvxx4k?;maBli0f8VhoE
zB7papW3n}J_fwQKfu`$=?vG$!Zzs6u?!gbbiBXLq$S_u$+J#oqY}EE4x~=%GpCD`p
zfqkplPpsezHGH}i6VODXaz_!SR)}BVF@C^h_Bc4ix3vVb9~mLdY6D(BU1=`{CAjX}
zx2a&~J{sLHBID=PmVd<fGyB@(`TFD7cX;h@07cLR*Cx}MxZ9Z1P&$S0M16gIoA+Cg
zGE?5n)_^Kd>vVt1MJ>}pH1sXa2^n{ivDC7ciS;V#PhtO%Y6NJKMIL;~ruM);m3-Hu
z5rC-OGMPP2W@cuQ(S&a7#8b=P;)sK(@)Su>=}Zp*{MzRIS7_dE@qI*3;%la$Qr{)*
zeVF}w%)d~>rxvgEG>_M&6!IS=xJC%HqNzVu!5~rWNr$<KNrh~V(0RzDyiGA)a@k4Y
zBAO#Amxphk{m%y!Kg+HPBl3jp^Qk3WHcY>|P_){+zZWfCFf5an)*eD<eaBmI#KKgI
z5Vl4I><|9av)74Vy&s~M!**X`vB)s*00Ih1?)n)e8|aSoAFVW+AZqnf%e^qQfj1@t
z$%P?+MfQF3*`-_hK5|$$Zrp%{87qE8cRs&+Ujm#ms#l^*CRtn%R6!xfxzqas+nKAE
z8ly(BefxGq^H-c-HU64hPxC_Z8*wq8&aRS}I1A<#c#i%3H`$enOLA&UMrvG~G+qeo
zdFS_=PBaHq-$R<*Ek8IS$K!P)bO-yH&l(a?dHk41w=a|Oii(^7VT;|Bh3JF;gy-|(
zUUl|kvX=u41@0s=V7;9+HQPfGfwFlP<v&Q9r0gEjxv2Ve@)9<a(ja#A>eZW9oHX5C
zBf!BxF#83bv*{%BUkDyb-;z4hL3aBcP-L&Sp>W`+Vs#K-4^H`axIJXWQBEwG-L;7B
z$(-%)?C-O%%yvX?O7o)vnorvV-{APHeVgG}qH9D-N~)@N9i3PS;NJo89lgZ#9l&F1
zJ)=8nbJ(?<OA?xl17<B(z#V4uu0@A`qKVou8eb#v1I89>X-~z_($H^Z0rKo;Hk^#X
z>k$vvWJX)n2Se%XRq!1QbpYaEQ}BXQC@Z5B@6mCd-9vPLhUq?U0EoefP2YJ1#xS*U
z1dUPF<?DMd+&5QN_c9`*+5J}R5;6NH;Ks1UurJc&as6(hId_9h0NL8n9$Pej8wRb_
zg!GTt!@ktDCEk5K2lv&1<*wn31)XVW?{40o8FG1EC72F8em<eu*E>8+s9-#Gywop0
zYQEAuzd#SPGynYM89DbcJo)mnvSdZI8g>nHoEnJ5^`_J9Eg4Ml^0F%)0_M{3<8ug`
zGH*+Ra{xb`Sm5Ab!f7Sl6G^vF0x~#nOG*rbX9>X_hT4+x_GQ;F@$ZmUe!W&0DGost
z$>0lC8D_qwd4@I}NM@<XF6{zrGqW?3gqlPB#IU3^Fei<k?Y0%#s&Lh0%x2tpsYi+C
z9mc$6ScvI_e_qS$B>P9n7=sB1JpphMH{Q@DXp`!GY1)Nqz^g1)Ka#xpOFdKBg`m0}
zDZGh-i{{|PfWms8i_5n<89K$Mg!jI^aoq}|UJ>?H%sVkg1GyB4H<&_FFJ;nYvLp-^
zP7OfLO>^;K#OkC)A=c{>``+Zqs3@moQyfyt*SLxhi}y4#C18w!CYI`sz*D%qZ$WfO
z@uH)qkS1++2a!uJ$xKF%o}YBifx@g=vyKZn{Ttb+4h;(~sB+&-Bz36w=5FfvXe11u
zr)ob&YBrTk-cMr~%6q;TWon)pMKXDDy{_zzbjkIE1ZSZCu@Dr7YJbm9kMB?DzDmM(
z<+h%#*1%(2fYt*&YggL0hW*H6il{MK!SD05%c=GMD&XVuIVqhI_0sz~`++C!M!MGC
zp5z;Z)x3b`3T@1x$q?wOz#U(M2~xYKVmiYiGVkD}cQpHQ0;JOj6&9lF;Bo;8n9{BG
zhUUS1*$W?0Y}9ZP=Irm$XW!x%;v}H;0x$W-hEVhQ{QG{6ru#5a<m!-}snMKw2>%)#
z$H~jR;@Ll}#Q1^j7KX22iR>8O`+l!r8q;9-0TrtaY5o16q_5P;EM=F6U@6mwcNmVs
z-H?8@DmvYA1TWnINQ1!Yo&cs@sA`L4*Pk(!DAA=A#8tYy(Mj!DM!OOj=VIR754b@?
zhJ_uQG1l#s22l+XxR78>RSoHzOS4K_3hZJhF$l|vJ;+Q?Pw)0hWj~^J$*Cs+14BsM
z)>N?{oqPVm0~MPgImY<qU`ykZ=>hLr;qA;S_7f80ALk2zEL^F(P}6;gzGB+=Xv}(p
z8Ni?J$o{v*cq4~}=UDZ7IJ-k^^AtV#lThJ>NP%||1F*WI`%`G{F8wPYMH|vS_mlB5
zRl-^9Uv}*N{rlgLsKzTqY(hwr`%~!-EMmYoxIghmnh8ODbKeB^H=S5V4bpnDGf@H~
zHLf8iCiDB5V#YbQ^h|^ov%6?klHZA@62KmiPQ3yu5jbqZ{Yi8uCWUesN=AX^-$*vA
zIVb>0KZ<{*dvdl{*LzP@6!HiIxmsU5jjK^;0>|c}TWmiDnziTUEMixoddwo16Rfv)
zTLDc>TqsE<2)c##pLX!Po})@igF~Zh^`;co5G~UC9gP|=-&?G=R@A(nOtn<$SVwdF
z<rtiXy&#tZ_;+6d?%l0JY)Gb1r3GZSI%zM+sA7*FPuL^8;`?XLoH@T<X#s+~;^H<0
zt~83I@Cg!V;_I($Zk*+X*90cgo7Z}jX>J{m`lid<&#Z&SoFQ)mq8~`mIaEn4?<v0j
zlZc;ea%by`4<q|fH?MjU${te98Nm{qDK&(V#wU`jdr8u~>0_BSTNGd1^}}@umvN*~
z=YTz!?EMeBvb5cdhYJ|^8zjfXH}u8#|CWJ6aG6wc%0e6&wOb^O!^Eq=nd9Ht0iKHH
zUjY~oMv<Mf(}7<T0xbp1?}d%)e-WA}taFIEP7WciMf+1=^}ETK8VswU(jU!!clWGN
z_+sY9V1J(D`sa}PR)?4nqgV=K+aQbLfY}GvUQ|P?B|~%Bp6s0M%^BUz@EGv;3%oiO
z<7;+Nzh);HPl0hSf#$`NF;Nvvl<7Yn(R__~ukfAUy3&`(_$kzBcutUK>cRJ?>eqG(
ztb-!05q$v0G~J%e9^IW?hl7Kqga<BTkwFqWlc@U{4k8uvjT<*c8nvO5t`fah4hI$3
z?@Rm3w{OZA83q>)bz;naNrYVU5V`)Qvvjb0SU7^PKxKFTBAr+1U1c;|ib+rJg;$~z
zza9|j9!!J!e(TT}>iMLSAupI)6<D2z=x84#HodEreKEOPiHYrZo=nn;UoYgCH5}2{
zHG#7SsNYmYT%7DZzbtyc7^Qd9EOXBeJ3_xTzDWpS7%9WwI%G>y^HvUz8RdFX*tBm)
z4ii43rpzaC%ql|)pp%b1qO+=wF^oko3`&ESfm!nKNZmQo>N4fpTw+npn%An}g$AAU
zzOn3fR>TcxDYTP7i!24p>4~QiSS66UN+?t3i}uxo#=cicV^?Q9WT5TAg9j0;qUJQ4
z|11X84ZW{}>oNB+TdNCXB)4X${*Pt};f`vMlXaE_&a{8{D1l4v1)jSCR}=HjKk$)*
zxrYWrG<r6+ccbMd_LDRFsc`FQW{|;w>59m^Cjp!tF#h>$2;!$P_Xjjk05nI+3ot-U
ziLbdNWwJ>$6s0DLV>`!B|3@z#cnz7jj4T6)gn?UBmYuU5@^fxpVf9f)DMppN(n<Iy
zEqpT7BrRxEu$Dbw9`oV)mu|xR_gBlJ9&lin`o(u4!w#<v8_;N;*<>-y0C0&kYECw7
z2FxkyT7V6k9^%Q)c>?a;?o2XG?oa5|lsr6W8-aZcDiGwnT#^YxMr%ed4B73N4SQ&o
z5-)Ed+nl`Y(^)8y<NaAMeacF6W@4KngS*tk$;XwHl^uAu?YN=)8*>Ybeh?NLl#FJt
z9E^!j?kB!5Seqb~a@M)PcPp#R8Ns%kD?DK_^91kC)SBB5Cfm|{+0Ci7NI8Q6@-if|
zHJfj_l9IA|*jonTX3z$}nwdrE&WCfJQ8c~-I1fy{GpN`}_gzIiJ;(*V`1}*@4Uie7
zBe-p@60}YEfCYewF7!qFUZ`8DDl$tFJ$dZbZN)!vd1U5jvMI@g!9{mJrmn*P$P6k<
zS%?<dGlH3ojp^=yL@L_8)Abduppme)z+{vLZVVR3-pg_&!obZL6!-`t4iaj=UTm1s
zQqqrEgwaZB4`^OZYMR`T0$Ke!F8N&QI#WPoF$=AJk&0hX*TDd=jjmo!?3#DZz5~{A
z)L3k`nC?mH{Ymdd+Dgp)rUf9-hTSjN)ym;_S0$YhJi*|?CScIbH>9&Wm3@M@Q)Kj8
zI)G`J(b3UGC)>ci<AA)BjT)VTNyVonCc2__iW_8B%O1@eX>*ctiFZ25graPdHuSMX
z2h*?QZH>sw6};*eFy#o(Z`Jp2o~pjg17juzl?0$pTV)dO-*5X`8ei1?6KMo+2PFin
zWzyV0Xk%_=ZlENp@mC=9=)DAjXseGg^)-Hrgxo2%{^$XYH0noz1C7k@OqU}=nx8#;
zhQ|mT+DT~R@x|QK4|7QSRUZO&fI#yQVblOh`YnXm<NPs}ljHLKgkyU_6zB0kUFQyg
zbFoNk$20s4UrXLJDPx0Pi$^qKs*)<VA8fxZ36hFcF6VQ`8>mq~C!iPcpmxvaV+6?&
z+||E}=KWa)#OIkgy$}T&0i*$_mX-32iebRrSRUY=ob^hL|I4oU0VQ}}EQ2QH#N**<
z9k+Smt6+J8N=S030TkEcxw<^{5q9K(i5QuZ$j(oYz_>se_6h(`cC~4cB>y-vVhL6Y
zq>Kk4)5<%;Y^W;6qn=C$`yl!Y^)Q>dlk6@u_uWfI^<emP0u~4nIUaw)z9!><8ByyA
z{B75OWd{&1Yj#(ucW-I-*H#ovN(=WxT~M+!V=P<Z^B{La0x^$rb%A@-i_5~W`=^>R
z&zzGh0lT+YRr8eQ@}!gHI^j#e|4~j?l0``beNgVT>WOzLtEt9<-AJu<Q;$i_?@e|G
zXVIvJlrOoJI8-uS6biY$c;oWr%T=7iVqn8G22B_NQiiSrVrkp)S(_RUx*=4Mj8D@y
zsd-NKZE13yAk+EG$c-V5`Z;KX;Zus@>e6mTrYrm9_ZH93%jG*yvpZ&o8_Fe}0r&U{
zv<$*Ak8kSjR2koA5w3d~JdKg1wRkijV~$Riv1WgL8ka1BQh(%1z8S2WT~pQ-r7FHE
zMP!Dv_k15OlXhonCQaFer)sZtGw=n7R^h<`cj(cjF3<KwZ(t)xB1lsv3%*xPQIY=1
znwK<Z)cjDc6o@P|8V=-a-QJHZ5N>d#%9Z_ZBocz%x(@=WThh=;qYWZL;_}eYJj*#x
z7Wxw~zYOS*K5q3Es$lKQ0mUiidYd<<cVgDODjMBp^xDuELzkMsWWw5s>;zUEF=;6o
zch@&gU50%{a1~z9)@p&dd)sUT_lqj@5hn3@xBpbpz!d=SVZJa|gA{e2(qwpLVc3d5
z^ZlT-wgoOevAc;*D4Hdm5v^%!pVMSw5NjQVw~eDJu1f?Y@K3dCjoZi5^1Za{_((dV
z-P+F9RvXy|FtPPbRgb6MW|1){M=)2RQL3}1C%*=BeiCkEhS3-b6=&;a#mrwiD7wP%
z#Kgw^_&m)08pVK}?EodMOzVkm$VYs;%q|*14KND=Li{VEz}pOz-Op*m-QoBTac0hb
z2MH&-Lk-+B528#wb)2tp{~4NFZ2J_Lj652|3j?QGFnZ6MKTMpe0mqWeS6ZbxhP;l-
zpvjZ+sK7WwEu*Q7sg?;_U(Lt{S<3yXi+Y2w0Il+p>2I%r)Mw|;oue{t(47?2{NLye
zv#2rNwf*LB_hXb@>b}5$^DAyQdtt?`KJK9j2VVa)s{adE1d<S!)#G|-_Ew4?;u!;Q
z2^bFy-9vgKWezS_TvJT*9?g^GlKub}#nxYWS6=>QGPjE&MY3u)9z?T>PFk)x@~S&B
zXi~vE>8{rH7=eYt9daewU~WR>*|$bLiTZeQRp9&jUwVg*KhKL9w`k5P-OaignUp55
zbwH|+<&&9uvQJyRld9u$$=eS!xA^Ro48Vg`<4d~JN;RsV?fLx?Y&1L&RIqx24mDF<
zHqXb1r7yefR{;j?)IkdxTjAcDTZx9Tfl$t3T!R=z5VCSm(gIt=&%kQq($dm+<3x$=
z76i97v@SHq&*q`rel&i=O9t@AtGlA+_x9mckEmASA?nl1Ti^Z$O>=Iw5smv1F(WRI
zQbe2LEfe;VD(J$b9^MdkWJ(LLQRY|E9oVzVJQ{%vzX5F<(D6&;<;keVJ}dQ7>dmPF
z#*y@o`6dgLNP>Fa(!4!`g1}^Co`ny#y@mz_bpiObEBb8Ir%{*E#2CZ7HD29@?mAU)
zdQif&{&AAMzQ?U+XjG4MU;2p2%8L-8JfC~q7L;znlNux_=db<(V4^84Q3Qo>%WE&q
z6~;8>Rss)zQQGJJL~=b%hoq@qt>-Sl`Th470~Rmr`F+KvOq$#|k0C@9)i2DhR)>gU
z8u7P=hVP$8bWnBQ_wd-b6p2<FcR1VA`<m&F>~ml;m{HJt3gE^ol}=M{)vYbmTLicW
zttO6<mx_>l1ByxOSY`&z!!h)*ZU)H{=o@kR;5>V4!Shawa!^_5(tnaYvk@tYxfK@d
zwOL4Lh8$R{E15WPWcGV_ebT9i^k^`W_zDQY+N8up>+Z2MZlqqFQN1+A$RgpO8tRQk
z@7`HHDmy5O>d{rSiu&(=om92Yq(1Zz-M+k7f;ea^m7jTw`z~Z%qbLO5=kS_tBi7J#
z2v_S&E)W>!o2SN2Ckp+!mGD>43yAmzGvMESjn8b3ANy@MMeF(50-xw_RM=VRyy58a
ztu%&9$24e<ogJZbgU5|R9!&W6(IZtQDGQDH3U#0V`DdWHs-mLe{(bw_xJ{ywh2q@6
zWIP4HfW2^mU@vikDE8gR+G$BSvTX-3(^5O_08;<Qk0_6(*`iHDZl!R4R}O1Qt!Il*
z6{C@cC8HwS`+l-FsWDV9^KR6#IPsHA&Zc0L4oBUVbcScLhv75;pOdh8;qRARsYquH
z3GOYGnje0gB6H=EwAqZ$dUWF8Y*as0+H|v#(Iu2C&^g{~?Nen%+1qf2eK5+m=q@i^
zsWOpnZ>_llwX`10p#PaOa^G=^73E$6UVcrPAcPX=Tbuc1H{HivhioFU{!>o-+A{vz
zf0TtZexi6UYA#oBS$1Ce2%6|7Vof~J!+?AYI0}g0y7@-zro1{H&AXeG(Y46eZ+*=$
zcGGGPF}mw-4on^bb~G|S@YG`ZzZCb(SJw--|4+3Pp>-P`N>hsAkw0i{j|=nfpR@S|
zvxxR}@<aow6zz3!aS?JUToG6;ftUro5u-g9(tXzNz+|AwW+O`ssbp5tx=lueQgbpN
z9=5+V;fudOD9A4d&;@T3OoKeY=!@yyfM8;p%ooBzl<ea~JuHm@^lZ9zGZ3Wc(z8^Z
zP9q2R_FCIhD`|ZnvcxF?+)5tOCXu$T50Wm}@q$Ja&IL(2Pgd)3R1GC1IUuy?R^cWX
zTG9r$|2a$pnHBc#_;wnVa1kjrs39iau_a)VCp9Uc;Yb=K%;#9RmKdHBK0~ALrXxdv
z(4Cn{1J`#?l}`Y5Q}uS}Rs+XwVD;?D80Lc8J{KB5vsM|_LMJ}lVc_9Fh|=5WGlim^
zo~MD8?jQ461(>Ma^^+;&gP>Uu`fDE=b8o*jp!)<vj@%dV5CCWekxyt7OH#GZ`$pth
zJJb<S*^F;1azi&5(q)f?&oMu05J0aHwv$Wa><dONZJ=)qP(z1|UpGde9cxJBz4F1q
z{_dI$X|fq0!cZQ7q734K{nSC1HYXPDfp>rJkx5GP;T~iSADxB-@vC9XenA#BH@yQS
zEAk2oo7|0hk_!&T>CSi4WQzJAM4ZRx+E|_y$7sSf*+b7MLG#=v8DOLqH|HmVFJ`|E
zMYb^NFA@_yk$iIw+gb#fBs<k1;6&9&{EFk@{{I9usHSgAJBm+rsEfq=g@lBV{6nM~
zydb^5K!0^^cQJMDpiH?Wm&T&({$BSN#54X!eKY0FoHYcZ`QJ2QFZ4R8M3BBl0P}~Q
z)#Jcq!fgWl7M?NOlH6^Kk|AqWeD3+a@5j|hE1bhIah+g*cvK>3*Loh9Osor%UudSq
zEXrk*dCpQ)r&UpFj&P{k;vNuFcxVXIiPZ<wpb!<0eSj)vWb~o%=lsu{8C2Q428AkK
zJ=PoHN4p;>PY{oKpeGJ<9TpE4@pEPm)uIpSQt!+Iz0tvLb9)$_al&Z(Lp5};Z2=Pp
zK%93pMXt2E+6zpJ=&ArD9ZhBNN`gF8<6@Nu%lx04Ku^Ad*PD622ONx{XI&nc{Pyiz
z5)a_40ULz7v~(uVQR)$`8|bNM6m>K+t3nsiw*?0W(8fdF3(xo5N(AKKED2#W?AqJe
zxq|$MdRG7Wj-Mu0J**&D{^dLgx*cYx$tAJ2mOa4mEOBnR0*WD*Lz@nxsiLh}F|+`D
zmgO61mphs+M;Z+Y!7*MRVNyQ1`Z)K!eCl!^5&3UO>rw$9kaENZW~OTB(Q9PLb)t*o
zEd%W6)%8SJ>7TQ9R1u%p)YL?{yW4V;(AP%Dg>JSD)S(Aa(9E}2YgCWIM`l4B;VoR|
zGl7y!^|gJ7X}x?c9;A=z^}HC=CD??@(Yl$SM$rpVQqtRzl<8_PCzdKvBhWdg03u0E
z84px-5Q(qZ5KZ@5vI3JW@4CI+vJ*WFm=l1xSRd}Ii&1?1m80^mE>~>5(MjrRDiL&2
zg>@#wr$wf9Aa~~8C^~x7y&TGM+f)JTa(u#c{l-}>fb5!>n5aZ@SnfXrYlQb7xgek$
zzJSrfX$PYs$f^{=-<o0glmE1|QVZ|&wHKI$@}bxFP_FYFP*5$vt70i3r{{AK|2x^v
z(V`q#_zgQCe}R?q@0VDQM(LZJR@+PML34DPvzRQVJJ4^7JF_VtInDH}TYUFRi@u_H
z<)Sn?4=1*Pbm#-&eD+V#O=$M>QRL<D4|D*_O8J(<b77i?>mD#7{NqM$*nNrV2oI4g
zE~ENfOj!P3c=%ihJQh|`K0A?!>HL8<QDo&4F;C-_X4)&#@7%=OG>OA!q3X83&M6s5
z3Hegfk3H(67{+g)<qtkxaTLZ(N`)2}xu2`3tgJ*Q`~psEX@b-^gwb5SzjE*KdEANU
z9>ixeo9j^P3}18ADR+VYv}8<nIkfhE2>r&U3<=a!=0`DnSk{dZ7TdVk*x0sl`+@p@
zYWaF&xekpHMTT)KFipfB0bBO9v;}3#)>~ow7y{3-!3ff0QC*|j*Y$Ds>t<#Wb*S56
z8l*+|-Z1~K4(CML(>$8w3&z=8AbRvK)xpaAE$FxVkl%DSXZBa#g9EPh5W|jdN-Zug
z2ktc+E1}C5Rm|EgE2NN@=<)tmwfjPGNl@3`n3UjA!8@isn&`)^cYW)GRddd)C>9iz
zzP@XMWwD0k628k(({<L2eJ$v+WGT;a4FiKE&03~ou8)z}B{|&c_>Pt@5*ROj-nzbj
z-^HJuS89TUHdvZ9G{=02NeL}T{ilBQHD?P?=a!b}se?=gW8))rUB$F7^E`aee(7+<
z%RLKt#96KO;^7!LnSA6fLf$8nmM=_Gl0`czCT9c*7|ddk;D6nue82nvFl3~(E$N$`
zspoM(YUBr2^R`P&eS)Xl{<CDfFY+zlzhnb$kWhAO%L-RU5IxXTe6&<QN(7@Uk(}|3
zvw73l_M`5lPIP+OMXtS5{`66i0akNHsRk`~D?qwp)v8r)t{);!O6QX4>4w=XqBq4{
zEBr(y+_U!{#`C@mi;q;l!s;dEbCYqL<#o7TYo)7`;@9WFZiR)NV0{RW_bf0t2?iy!
zbiMJl--RKc&dv!7To0A%&sqPsy$>_KxUw;@q)8rm3;WSdD1^@BHyBSjA{$vUpr-vB
zq}X0#*2hGCKOortn%SUXvfp2rj`l=KZW2T7t9;KIF>0hCBR4njn8=PK!03db+P4a;
zYyqM`aH1W(o&JEn9`aNrB;v=7#NFE(S^-207BEnyd{vAizIL?q4l?6prrvS-iz}4{
z`6%Ap`x1Z1LOQh7j*$Yaubh0?kEW{B{TMFj8)r3h=1fJqax2YQ_4K<u-c*_V^C;i?
zF{orCcpwyeBLnj&SNDU!$9%d;a-`Z%=?T4Xz9LP0-`_uZ63cgeZ9p~nt+uRq%E<Py
zh?FCV3Vydsy6{zi+asWF_{J|2=%qoa5!l{YfTX#(TA~btQj(%%6d+`OH0-(W#^VI+
zf$8<x^N!3v@>6=j;loVOEd<x)aV#H@8RpWL7E4B{Qx*!eMP!IMZ_+Nqmt)2wX#Oti
zi__nL5x>^Og)cziUE)uM)rN&}v8pp7ruTp;S90toi0B!5pE<0u5K!v<Nh5+(JRYoA
z5lMY|n~4yBJIlPGx*hp{TRS`X$X7q%sSa^@wGRn)S;hIqflBpzwT47su$p(qiWRHb
z79$B1uFg0Rjx36Zz_HP-@(}EEx!=ENh@C%xj(`c{BpA-_fNK3G?cz#nvL7;wa%SoF
zGt0y<c0ymzjaIoBnIhZ+@991_*9ed@#;11#Z&~5X$X&3Af;PE~+LU(E=N)0$O8-?2
zn5(`VQ{14YY9Ae`A!8X1YclE^17sOQar%HN=G^+5<IV0Pm8TF|5TbvV!-tI{SP^u;
zLZ&cW3Z99aWCJ7TFksvmGSz@O!{o|)sz8;lG0-sA@?p&TVs=}TIb=MbTn~Is$Vnh6
z_%J{su}iy6g^4U~iOBmzeu?~e@7}Ljy<a}7^&glgAx$snngjz3jOSP+9QHn&UWQC#
z*m$Z)1!Io;C}|@LuV2=j7y3d1LQE=w7KX>Vcz(l5ij)h~8D;)9tTk?jUm_O~!9kvi
zk(j%@{wCfrVJw(9E(LhWKSq{FLLD%<;#Xbt!^fAnD5ZP;a@r`NM8CJ(+KVRkjr-=1
zUR^B2h<X?zHL02ejWs24S$i5=4*eiZBO{VPIK5szLZ7tK^tYb87tj|8O-v>UGP`z}
z*2Z`CBjCgOD;!TWAYz`#(9dwhvctl9NW|)?%rz0DOvm_s4j*=C#P5Lqu7G}{)V@wI
z*YNcldAGIntA(BF>&j8Xw7-->LmM!aXL_YCAe6boz)vzvC~ajxEt#gYed+2p;(!xI
zb;vO$vb?+`qbNB{ox8}8fzpMk)%*^JQ_eBqH=xi@{22cqBaeoaW8|Xs&(xaU=KF8f
zkB4FvT$HsTa`-TjvvCZH80ZL>WOBA3tCw*(QVw+q?>l4)NUw*1>t&dFDqi&!!c+%p
zZCTiQuwX!;hX@Rt!?mCv`#Mq<(p69{;3n0(A<UFiAF|pT_h)vS&69`@=ocRSW|+ua
zGN*#+unxE6>9Ts|s2Ls^SqD`z0baRxPiwO2t6QMS@l((B95$@>5aZ}Jr>fe7uRPs0
zS9dB1$!Ez~{TqJr4U*`fj|kFl?5W5E^}I~3>hDXivG~7#LvQFIes}JEf<At-;2aq!
zgN8%sM+j{*nbZ_Ep^qRAKk|0rpTzILuSTd4uq!!;+rs#5ACor>+oG9EufnBU5rBiV
zY;l^>UN9zn`gtJXus$P}`>RkIjF}lH85fIkI5q&hUe;enuY5}l2^9js#dy8@Gk(rG
zLbDjk!sARYR-@3h#=wSN3Al!X%Ojz}hp=jeh5e86`cn$gz5uRSvxZ<7Waj|9la18p
z@Z|P@3wUvwB9s`(CLTFlL{Z@lE8#<rF+GngX{lds7X2Z`VVWc<dO?W;oPvHi>rJ{M
z6b)cz4eLu-A6-@uA`SRc!TQtbm1Sui=HbCNgdRYalOg=`M^aptI5~h8vYHxD5Ir5G
zOz-=|a%1buetjYf9iQ#qKt|$?J3)0>w9?Rw9i)AuFuegi&~M(p-Jw|r6-wur`Mv{K
z>85IXG;WVH<=a*4*raPp-6)ykJ=ydG!Fa4OP@um+!!yWXubQ;%pERY`vI4Jtm|L=A
z<Z4D4g2lXeq4(n7n{RyMR^^<Ue!k$dPP8qDoz7>LM)1TuQ9=1}yfNI5A{4nTrxw2&
zQ|YUU^H^iBp2KGagE#(0MusY+9oV*+{QL|gAd4*y0!!<XKuv*^k+ami(s%=d=)GWp
z24Z<?lNYrf5$?$f3!6{<8e2dPeIyH^<6ijX=Fy8)F1nNyeLKbUql?KvyKzd491)8f
zeC3+k2mh?z&j{{zYt((FzqF^n>lHJnqn*P_4z2VLSFB(HF$RJ_D1r-b(1p(5_4+b$
z3sru(1@s?tBJgG14eg*QV3a6J_zVD+<)*^vl&iNU&9dBaDl!jix#oR=KTK!94ex;x
zmrS3BdUdSLLdhdUz=Z)cVL9dl$bW*D<w2qEP9)`kX|mHa!>7`PX5&hWil(@&(gqyV
zmdr01Ig0@;U=gF)7{WEVavvF=y{6cZg9Ld3lSw!&@m7F9C`;UPY0HYylrZSR3Wbna
z3bvcWJzfD8Oy;2K8*y--ByO1|gPA)6y7<L;@tp}<5)OAzT=MxxNXLth)+^9_3F=Mz
zLGOndE2DD}2s12%jt-D7h6voJ3S~5mm;(s-eB%^J)L~z_OVjK*{b@<uLq`--|AJv)
z!nkq<IF3c8iWwczj!D1vdw1?%Q=G`*xTKvwziQey2UXJ46^|ZK48r{Aix)3~^>|uJ
z37Kx+(Nv)my@%c)7`c-6efZEySIzJ5I=AA{3hF6E6S^TElYC;W))d!q_#UNSo_z@J
z5W%8~pN7_^9V16Sp#)PdKvNbK7aKIFIuDu!X%3G@9wX1dc{7etX}P`m?&Djvq$OuC
z^+kM4&CdkZSS$Av$lFB`nzi(=FR&c=G*B+^pvVofe1OmM$33R5^a;=xQo|45lQ-|o
zQcD25Zy1c_uu*e@LqG`(j&xhYYuvk{<QR&BKmQ!EI-{Exf$ZKvLx97xiFsRDsSL6)
zsPdy@`Fc!ea#Mx<G4L-T7_LxdJ{wpPc*hqckei^3`Ah84b%YLS92cqF5f7)9kK#yQ
zcuZhy+;P$X7Z0BoBSfzpN)7>s!R>Wd3S6#6g8ZH$6tfp$0=jgitI{sJNvwp)6B{|M
zbd1nTz(v4x!mwHy%GdB$UCL;S%sO31#x`mQJf(ju!FxfKc9fxjVP_|f>pMm9$M+6e
zgE%2#4Gd?{JL5Jl0*%a>D%k7EMD|q<z8`NfrqYn~D6cWt&Eeuj*+aWWp_Du0e+}nD
z;9;aPnr==-adCKZ+i60(zX8HG9TI8q9_&a6m2y!F(v0l6uVu|aIdy>X!4;)LC?^85
zCzBp)SFv1Kml|>+o8fwdXVrZ|SD}MOAcs^z@O_}?!6JAJNar)NQaw8D6BwnHfdny0
zuSEbNL2%XKdtN|+LGx5xkT_2-kcl>Ik)0o-1k7S$^cHTuGsjVUnO*skc`=(gh~pgu
z(+}X1V}oEe$_ry#<+F7{*|sOg1X51lijVh3E1Gt+C5L0L_!Hd`M)eiEi&+ke?C(`7
z9I+8b=iqQcpK!<;<)<R?lsfA<l5rc4h!|50@YVY;IXK0sDM!DR#FjogJyR$8Kl;y^
zb_8YzvJC_r?q}{15eBTBB^tl#0;CzY^jy$LS!)QqqJJTRfFmP##I2F+8XAy-1b8V5
zyx|nGr4wf?i7IU0ju}#5a<bRtwA{bQ>tQHmG>#kQ0|36<TYjf0Z*He4_|be)bA)?p
zN3SvXMgLk(t_8O)YM;WWIDCn^&KL!$K{a(x2JOMyy|4$!`6xu@a<B&m#G8d~7J<`(
zonE{+H9=%H<?v_i19LxjoRBv-P6YJ~>9al33WT0D)2kObF>q2>W_o9C;T=nngfqWT
zhjZ`vwx!`-tKYmcj%AyrIBc^J39lJ$IogCVfv8<SAIfRvXosq*IKFX%WX{k_<nT@F
z1C)RKD|Q}v5JJf}YV`z)fl{hPr+aLCyo^guSM-L15+V7)`kdA>CUbL;U0b~K%nH4q
zYzu+qb%N<4X-u{Ei}9m>ys{yfkLTtfa&<kjPQ9a6+flrWUGYGg{_joW%A}K^*R)g(
zRg9GQKyQ8XfVUt6ei(GA-e00`=AD?qsQ>@<4hzl6i0oRMR!ECt#1{?=eM0})<A4z1
z-8MF4SkV?eMj>*b5JtR#a>6(pZ4gr!`N5*M#A3Q$3B6-@1_{n9nDGM%Uo~09g^cc7
ztwVV+yDn{9zf9EedKaidI4=}{o}N1JR>I{WUC}_s>{#7IiC&kw0SM$-E4}x}rx$As
zPSXp~yw{bLo<MaTH4<0tE-Pp*OUStu#Q<5eXoJx*->2Dbj4sXuACsV-JiR?Qqqc=f
z`G7o_mc?&<R2R4CG3t*jCpD>vj07_B12Q^}*Kh&7#q4_cClHf8_}k@31nFu-oq);4
z371Cb6hIH4q!UO{z7-qm0k&rCXia)emlK){QUv{I1AG;Mug|w(0KWg4g2#`#eieAn
zez+_Rpxe_??i`k}z|hc;#~ysq1~!2GKrSh6*U~Us1rOjWr==+Tu?3vg5~l=ZdW>d9
z)>CRAV<;=96PVeOh!}cnAQU?P(n}r8G)q%@jXr4tRl60nkD`{gfg}mUx=+ScP3`zn
zvIpSR8wMrxl5qX(-_Q({BaG0@kbxxl8z9WJMK;JM_Awqf6bgM-<;)@Y{|Mg{m>Dzx
z*MbE~7;T6w>ufEJGeg^;gv+{jpyWbtMvy+I<?et@Ll`(qJQ~^U&lm#*qVy@-9IjaD
zxT^Np_ZuPhauGA=y>X$@gy97ILa<i^c$&W>|AY-S>F^w#Dj~ta<Ekc30{<WY^fwGf
z&>z>wt$$-PwC*e-v6&$Q3GjED;b+O*V@FvxQV-L!qgTLZj@%y-m7R#8gs|WJVZjP<
z#Vd^tuSJhHjH6Dp0)6%<=TXS={q{*CT0c6mRVX*K5zZt>;iV4M%JX8T(>pz{rBU^^
zV<=!5=;^R1YAVaS5u7IR#2ifjV-E5;>}(&JV?jN%7l~ZVEPXo7Vu@so?@$}{*diD-
ztF>a^2^DUHF}(&BY6{HX0XP&tus1d~?m#f+ors~7BfgU{5*U<i)G+BgDzO;3e#1bE
z{_3ue*w<Wn;PF^-C_OUy5uJ4j5kDwFLQfEYtIHjScVV&VdZ#%k^fP5h&X9T0%<5al
z!PG5!?<saqmC>Oi5h&r(j_7>pyciV@N52ict#}j&q_ojohCpz~YEg>$>v~8Cxg1*n
zh=z4&;_y|kHvcWXUGuOfr8kL-lyeUSDtx@p%uFOGX6os1Q0eQ9?^&&5`UYga#%gL|
zvFm)n!DeUQ=HJLtVbba8C^35Vx%P7vI#sJCS&<1CtF0*}?_V-V#ph%Op%Y(Y5X<4x
z=G#Aff9!9)P0xpV%>NaT!H!)klRSa2gQInEL5T#tfti1iGzFw7oqJ)0B-+>`5k(XO
z<N1lyAVwH@6555+QO7v!>hAr-#H-G)mO(Y4{*wtd<{hi4Wx!o~_ACe8<9M@XOlbl3
zJVbvY2fZm$4lfiSCdd-zsCj`qDE~w*1NZ;eVQ~a#Wu$#aM(b$@K|MPT3w<a@Q5@_f
z$Y!C`!5C1fH+Sf1U3iYHQ4827BwcQnq7VWd8=}u?y$F==ssLmn*|D56Rp}l{TlB_K
z*XwOqt2WNG;UL=Y<s*XWJ!zrHJ78N-J@^|BZ$0Q^Y!&uwa$z({|BduScVHkIXhn3x
zW7EJ{$uJH195y=T0IDz;KsOs;W(B+86|d_kUv;?(eRw_|X=DPSpCt#o5SaifWEZ7b
zK*R%;Ht|H9H37u`J#!5*^bfXC%H@cnV{I0h`=>vFgOL8u9@qc?AP}kw@|!&$nuGFx
z3(V29hZ)^yZno7RMNgWF@EQFOj0|;ih=_}8#_WY9^Qm>fDK{t1RA~EFmQYw#1uoqT
z%&pycf<tcRf-y$-Lcc-*uML=uD#d?WUNag{|2i8NmYU&d`<zS#Pk4^s2m0i{kab~s
zIYKiM>!DpY6LSR;B6=u?JtiGgyqFy58}|)#_UH0<=q#8`uj%w=qh>_f&)}3rT|ZNP
zXWNl2dOL>Nbj_oWz2|qJBiWaCHGPI=S#CX2XJ`-~)V>7!pht;hH8Bj;IPirPm!c~|
zqze%o3`|`c@Mm}gGH_Y|Q38TGM8KM2hqLR!dOFth4Jmv2S<=H-gseMA$qe(;QA7ah
zyIX2`?lX!M<f4uOhctYA3UH~@^^S57r5;i&(-z}{SfVp?!igUgk2B)OQyDK=$K?b7
zI)mdXT9uGFC|)#Dj#@pPqUKsLp=TObrxTK$auG{8C^@5`p%Ga$#Zb0(^?X*dXwa7q
z)Jz9pcMZQdz0z%vK$aFsf{6@5A2qal?^wN#ver2_xX>dMoxrmw8?lSS2TikqN(IIb
zdUCKrrb%(U{8y-k_UG+X-Pj-<hj#?tw<VFBmMblXYlWUqA!G~M3Rjih+OmDq1;3%z
za>@|^dq+us0Q;Neyb~PCJ}${pCJ%&=ErmJIyS|1r9N(gr?jq~aF@0KS*}XymQ?c{A
z?d=sJFLT(Y=q`JE(sPD4^1t)z@sR;r^js)!BtMmm%<Oc4^B>8(nm$dlOd=ZsH)G~Z
zsBgqoI~Z{prEaHbhbAf&<Vwa>rC}uWc2p~G@Gqmk$0(r!<)7Lh=0Oq~-0GuN@1<N~
zq8)_IXHODouwTPLF&SJSJTifrzDDAVDZvqxirs}7B=LqC;r|MKO)%d9Zn+LHCv^C=
zs3oX;JHulQ)C9v^LD>AMLlY{I$9ho(KUn*?>3V;$UsoY=S(0oRzhx4IXb^)57%Y_O
z@l%facKO}yYcbkvyjda`Jhzh2Ap1Df+h*2Q0VYaC&j-T^E-csx0Y$*?iAIvV6R#+a
zoJ|QK3mw&51|-#X4J2LZ0DjlyNNopQZICF;aJU5=_^pP2DBBaW1Ia&HOMUI%5oZ?Q
zrHv#fK~usN5{LyD0sbtW;9$XQ^-hXTkZ(j6`8d;fGVEqeb|`(Au~yUsSu9u|SYE>9
z9%q)CcVZHCi-5uJfps=-u(Zqe<0UnOE<O%-=!n+(Z`dn@9Ykp0$ds2<qUg<Nl}n~h
z>uNXIy?^YxCOe12tEuyV)}1+H2F`w(x!X46ha{D<hAQA_Szwzn*xa}I4H!1BzO2*s
z58&Wn#N7E4ngMMv@x&yG4z|g|HFI6s9iaL<innzqhg&oTuso^4r*dC~-Kf<b)YoDK
zHZ_tvk}n6Mske1g-+%O(F@tlUYL#Z76+o?GrAsXY>T9iE-gG}pz8rDO9*rJR4krzC
zDWz8Slv=^Fd3I7JL-3(`H)t?-X?TioSV{U1Boclavf58>E&Z4BXz~e3Ms3G8ye)YV
z)pHy+b`>twHOI$eA?294bIUCC?X9JPl$Go@BNC~t{st2)D>9Bl++%Ye(?l`<6z=7j
zW%ZW{x<_C%CB!cVCbU1O!1G4o=`BiEJN6ro<=}@&l(5#wI$CNRHXWthoqR3)lhw5;
zwivm#MuP(;)*v2(rv^L%xU2O}%MwJW61flM9a0Y-Z+5|2E~vhK$3_u)w8^rVa<GO0
z4v2?xLNadAd@hO;X0sod;D@5u7juxsaw!x(n0gnf=pzw46m1fd_~5~8*J3^2<{WfN
zK<)0-Jj&szO(gE}6Glp`ujkjOr{a{bE~Y#H+zeU0@8E&#)DSvLfBVJ9hQ?wasMWfX
zJ)O-}gg7`n<SqmabDe^xjQ&Kg;1a_@7D>D`;NF|7x)V#y7fVtrZu;r=$mL}vz5|=+
zE)5|LIM}G^1dF9;Zwf1ib_wJV<F>T!r3hspu?bLw<IMbE6T%`RpI%rh$id~+0ynP`
zdR0(z1zMMYgE#f~5A`ziBQwC2ILiJBxJ`to^NyykmMyc%LwJYoC?vK8F&wD5DmDKE
zrGlq_TEU^Z65-^^&aDLegXmPp{|Ki;V?jcC{ca>k$gV;bmDv|isvq-O5iw_$=sUCh
zRn{6p${bX}Is!<?b-~pDE#%_)s`puD{q=uLybuF?S1^{r5%fBZY2x6GC)z;AH@Yan
zhei!nkZOuceu4dY#>4Iu+{!7xeH@g*3<Ut720}4|Pn2i!<(_-{7fdj1c)JIMbP^ft
zn#N&Ib_4^)#G|7OG)jO{xr<Ce*1}?JrMn97BWpA`0RPw8z5M(WI7?Vw{N6{DCFEjv
zQM8HSCrBhgJT<4H3K{~$lYa+?yi9N@YdF@A?yeBSE0irt$YH2ZmUSsQqql3ga8OR!
z5ZXHQ2Tle&5w9L7nnkf=DBo4`0hzlArp~?X7X695uVv{4oY#Lc%48^50n(09TJC6-
z+aWNMqFs!U*qZB@(+;vfL0%3W079*9fGA_8k|3h`rF%ah-I*`JV4L_`k<c`BLz5s^
zk_nmUa@uba2lE}3hJa)MngEIeQX~67NvIfPc&e&_3u9%caifbnNSJSEaA*NoYx_4K
zL#Ah<1f>M1S(l3Vcf=)Q%#b|C`<g?3l{XwbJFx@cMG@-9JRIz_h)aNuC!H1?*SQID
zBPKiy{mBy~Mp0CgFOVYUwMt(>udOJ;0iwU`!v!|zqC5p=7Hx%D5GWf+x|}P)$TsM(
zjS=VRZOc4k0#ZOd4cUh1=ti*fDaMvl3T_7X0OwM1Jn0Im!x_9+Ii!hXE7(Iac}jOg
z^k%3x5+8+k#}0sHr*K;D92h%ZVypoCdotq%(8au%O%y2?_=@t-BvbP$$0rllXmEho
zk4A>pr{=5H6_=uQ1X&szOz|!ZZlL&q4iSNr2Qxt2zPA6+0zH2E2lHqxfU0lby@Qx9
z1)WAkd3jz;2gMl=+)WxSmdAolX(sPernX#|V5`uL@$E9^X<dYw1q31Jwd_YRFb<qI
zmtIbv1TZx37$B=PbHts`y445m1P>F@$e0QoRZdB27=$5cg2qlQv@1bNz!!)A=`*qS
zSwA+zvVpHh6+|AcE!DkzPj1}3OQy}Jp-zYwv&|Ggp)m&|F6S$S<~<6nL}x{`0kW!l
z0bHDJaGFw$WCmCdV^V-~IPE9GRBRW^K{!iTcY*1!AAJ*O?^09ed>|K2PhzYLX!vT(
zE$@1Wu{_H>4MQ{|(OJ1uH<8lh6upG>ZH(`o^K&hkGy3-Jri~GPOd*X-K$8H#4k#Jh
zH<d!HD(q9y)Hie9;31uOEbjb$=VP@Gok+I)B*}h*e2SMZkx%A|W{H-do!EI^#6#xF
zqCY@`cS=w;!sIj9LaT+ZF%KP6e#z|jK5gksAz&*V?){-2{Z0z8<&=7b!3BV&#LvIr
zPq{V&Lu-+v7d^r3KifQRgX%9UkQyghE=6J-Bt3TXMlvoGYl~ydeV_iGJzp`wb4tel
zW<kinsgoZwq<ax1Qh_=kd&gJmVAlCNzxm8QfvvAq&|KluUNA`zv$ddI9vvRToTMDK
z*i&wFXw*P-Xi<drs=Uxqal<J}r4HFt#iB-+aFb|s1rh*OL<93Mo+hHUX4xC*9{`4i
zY`Bf5rwv}<EDXG!Z-tBknMvC9-#+kIBf~2f8$(qMXHbe59rwR$UcYzGi{NLjyPEV2
z0&`&>DQlg}&o^rD7>?r@xCVRaM4NCJMdjr-o2YfhXdXhLJK~Zap>#g&H-+gcOnr@D
zE;X=7#;bMwsy|ES;MN{rgP8!j^SfvhxN<KSe9n*2r0NOM7UdHXb;-i3%x^v~!)e%h
z_m)Xg92`c`s*t6qw)O;19~3X8XX4I|>V(?@6m4DC_TfE`GK{2{YR&I3!K0hQUMYQD
zI4{Cw;7Fa2IM^qd4tw!3Y6a;TfPfVaR}DIHL>66kBdXt2lYr9jG-WaAO9h`{d`%iU
z7F=CVc(X6epT+!2J;Ec0Jh1h<5W!_m|J!|J@b*XG_D45@7IA#dn;UN$G1+svK?&vk
zrc_3(KFO7|0a0~0ubw_br)5QK2JL#1)7^FBM0gl&4TmOSOuNwk(02Di^wQCm-H61^
zK`@vQBTW&dQY7(a%G(%(d)&mc!|<cvDWY`@r+(oHR8(%-dhpeBS9eYj8Wf|Zy7SAH
zx|P{>fO&YLMG0EoTS!_jR)RsQ@;6)Wa+|aWIW{f<a$sIW@qie~T0K8|uSw_1zVh9p
zCJlaT$B!36_g@0szX3yqzuPro6dtA>?AcWEn8o`AkhyUsB3%%k`=~j6N%5Cs#*6!^
z!E2ln_!|5Yc|$&OS3ERofS|01=QfW#*&N%NjH+8PhlhIlMZRPf5_fZOTyp)4c@YEE
zgsX(9)xNF-lKgE@x9*x2nzv%~5=BK}^mM-;UxE2z$deS@k_e@&Xh?8<*yYM3ZBW3I
z9xKs-j)ivz$UvH)PO)N4sAVdt@|WVyg|bW<!U;8HWJV})jwC2BxgtYCXUH9Q8Le*a
z^CAX14X!3(aI0L44j{zO-u)3wX$a;d3|Y&8<I+{$@ztU~WrZQ;EB|Tp2Ffi`3j-kt
z|7~HNKFXVFzej8P{}?<@6w!6-)5rQ?^et1A$o059gLi{uc~?DyKO{<%qBLCm;RcPg
zzi6LaH+5dbK!d>5Bq$>%X~G;(_w&H^(Su3@{XLVE8VPx+MfDB+Xz$-Qz2%}#P^vu_
zY6km9jCT(Jg++6pF#vGy4iGd&ZRjqU2;_HzmJK=014d@9hFyIM=U<T7G;&e76wly7
z&8g{&x%O)UIxR1f%M9%{^Akv8S%(-*pvZ_n-{}G6rW8LA$y&U}#WEt<*m2Y)%Rout
z(bk~XpwNgGBhsxX`e$V_k_Rgz<1l;Z$|o#as?&j$A7AqU7>Dm@_mq#-q<94v%0On$
zWwZ|1c}0<d4u#G9WV+vhi(+y?dGMfyzRdX-5tpA4S|;atX}M?Dchb6KNUenCHe#eG
zIyil$&ynoRg9i#5Bk~5`=edH<gdp@-x<9g475D#06;XF0#wb66{&i}`D|8nz;@uc0
zilbELg0$ag`?FvgTrn@AXF!9>q2E~Un+`ZvGG%SxOE4tM<<N%uu!m&KJ&-5gPd7K-
z47~;{NX$y=l4&b6NicQ><h!K(#Qz5~%9yI95F*tcdSxkKwt3SB)zhqxuyQE`)<giK
z9#y;F+&I{i`CtAH3jh|u*-FNF637yu#Dts=6jXIi%YH7Dr#ukbS6DXd>1&Ln+O6Tj
z%y@6*Rlx)pqzE;%`0%vBHY;oOXEUd;8Gc|(1-tUIvL`B5k|Ee26Y^|!4wbyW^Yy!T
zOLrIjFSPgDPMqALk|`9Wp1U2uHJ_HTn2zV5;Ud5^6MN0TizHee(v{=gv-aa3&x4p;
zO;W%gomQ(-YyUJ9HOQaAd*aF-?!EqKDuem_FRDs^Ht>xTD!B-g;TnyBK8dMD3L(8@
zf^AzQgZ{?$&oxiSH5dJQo0MT#x%~Y#6%S(BetA~go}^Sz7aRfTPsH(;Go0EPyaR0x
zn_W;%Bw#>Z;pgD*t2C9~d1p|4HcWu}Y-Hk8r}j>BeHI|xGI<04xuMEae-xvya(SqM
zL_0j|nSWLuhZy6U@&PBk>ZfS@-?p1;@Vc*l#&t&n@`ak{)PoGlZ%=@HPA5@%eJ<#G
zfGU-1n(8!DM<3qT0}U0T7$?<1CYNPoWIQR@hT5{(ovl9W{&-{aVUHu<78N;Rvf8wR
zj06+{k7pGJC`NQ0_$wHVGKL$_C<9ko+dpo=h(-e@=?P@{t`^D7N>?+0^#I|*)<c8m
zGJh)0Wx6uts?jB-A(n66np5vmTw`1JG5FV5?opr7KJybm(kj0OC#KbxyS!v~(F#?s
zb||L*w4_^&N(V1e*98};k6O|!xU@+{pzA*|c0YAVC8jm?wRa;KMi^X87nGWMygQnE
zUc`Jc?YW~;I6-c~^%I9gPVjx0aE3Q~xNd~ylK;s^9Nx(N)F;f(k55B-+zh=eUeQT@
zdT&<v3-XLyadv^_V&|0ZZg0=q<%ZD)MPJ8Md4wjUcK+;fbBMbbAM2!+_||LLSsjr(
z!^k8*-7$s+2I#tXMQRU_2D68b?~x)qJ$~?d*TIYV#}0*;<Nxdi-0xT=YdV^@)mYQ%
z8mQltOP52BE*2?5u9mSh{I^@WW9+I@N<0}R48~S>7*0fklxQL%L|&&eI0a$))>Lg}
zxl6L*%6%=>RZyaZRqKJ~JShFICQ2{&)^Sf&N73o)#PI+A`wxPx`sV_N;Q}LI_Aaja
zPtmzv`?~!i7FFWsCM7x-MbsD0cb!3zHJx_>Dnif2csCC-OL2)-Jc>?~olE~LMc#-L
zBD&X*Amo#Upr>xIeczOr*x1Q{5*FTG5AV{uZ}R0Se-^E)t80HyhydG;vM6{_JpK&_
zL2v(}@5fkm5758=IRfGGIl1TnO1`K4+TaFQogbrJj$ZPWQZ9H_!IKB=>$oR~W3LdP
zy{*Fsf)is~%Q6Yo)_qOozRk`l`<hbFK(J9n1xBFkAFlIcXtA$T0|+wYga_3o|FI4A
zbC#jGSLR)Hw&*2O%i4*o77$*2HoD*U9%y#PJx`w}Li)xSYatVM)M*#0S@k0Z-IT{f
zR;<v1c${AsRl)qUZLALYhr|nR<o2y!UJQSE1*(m7=3e-hs9FX8tp5?nM7T)Ity@cm
ze%oBExBP#1loJ(MyG}IHE>4Nh&COL-go@9T6)A#W0>N9fJ3j4)W|SY4o5NoLA-B5n
zN6-6ZV^Lr_c~|iei_RkpT_==pX=%zXr;!*4g=}g5?=({MAsSnYpzC-~L;%`Pxr4K$
z<15R952TojJ4Ufa$+%C;Hu!u>1@f=pv7PecM}W_`PttI%%b$-%oWE~E5O3n>LbH_n
zSgWL@<XJUl(ElNrAdkTm=2_3!ct*;T#+poTGqbA;&aH&*v7O?xE)`X<m1b>dLJt}$
zAZOz<e#e~E;<L84o?SVUwRk9EuI2v{_@$qw9~Y8v_+NrhAg51wr6#&UZVy^f&J8}7
ziMhG?opgW9qOZ<lTW+<Xq2ao9>l!9~1`tnPL?4km<04Rqx1N6*U%zeWf=*-%!Qmzy
z2E@L45oNmzUxBevdH>QrUF#yWgECekK)}e?2pqRkR<Z}BFIFO4Em_p>+MPz|XAVd^
zLdogR6UiSC+qW2osyhqpB&9>|;FBjP!+f^{&zF(u2LgT~o9I4U4ku^jw56h=q86U5
zzP#sl4BWz3sLOQM;uP0CPlgY=*N@;{|F_Y|s1!yMYQ*$FT_C#}nTPoiiwZ`*0zB=s
z-vxM7L`aM={`Jf4>u9lf?D$7<E+UE(a+5mDgU=f@gr`#;Q6AQ_sN1(y8vl2GWs)K;
z9c##m>H`Boj!zF>*u;@V{^Hkm{XITGqB1eUp-64nUNArhJ8x%o`Lg#cI4ukQAmwRG
z0f<u&DMBwSV=44hP8eB4;=Iv=_F*Kvy?@J*(Acc6|6gHIHZo@^NR6(tpD<JjM?S*O
zfdd6cC}n(qbLzg^3(k2lkI>&3&1_NUj!m4oW9SzSN1PxQp6*Y`8=h6LR#eH$I6{0u
zTresGdo-4e9bD_4mA~*OuECp_ftwgN=hFY#y6(6f_ct8U6fLWaLJJj%vPuILDYBwr
zWJF2HC=HSZI-(R66_qVSO9>ebQHiV~lI$i*>UZ7G`_%h-`<;KzIX>sS&+~l8J+Aw@
zue*?ReXnSm;|K3M$@2ocxDk#=p}R5+6QknAV<tJjbgImH5x!oUb6{kygAr$`OcVgF
zo%5^DA0j&o)JnLo)izP-pU<5~4lj{f&$lZ9$F$a%otPSfOm&cy`Cg7mIf=4Q|AGV{
zJki`+3uwf09I^d?;*Om|+`{YHQUm!`um`xhFVIF=JaitiFDMLbGSlwQezqslO~XX(
z)%W`}{&Nb&&L$<*U{5w+MAaT0Wwc(7+y31q>;P}r2NPH-fFSkZmaz90^OGC_4bf37
zgEW;<ViQPQJhqQsbO+keb<v~OZ`<mVNMd1sFj6+G$9-OAJC=3wQh4u;V7J$5oyOk?
z=~9Z4<k%bI3iNcxFZCp5=RZeDH3B7K>QIIw%}F0$g7k`nBI!h58^w1<g}<QXG;d2{
zBXE$xQs(>F^}Uq&`JClgZ-^c7?N~YH$QYI1gUApS3uzkG_lk{{`Fzf}ZviY#G3tuN
z_vRfOHXTj7>&&#(*=5c|^k|fg9=$35F7rMjM#g`6m-1*zp_dqMO`8@%^~jMU9zIU*
z(D|n`B5DY`1joT5jJHGGG_VE8G2@TpRv_HHcOA`Z#art^#B$Q>&5M=784pQEyLBoi
zZ@Hh}f-Of0-Rfk1=Z3+2+q!5TiHt0N`jp2d(4M)&EW{7oma7I6(uG4HHu`YOcIUV2
zc)3+5T+cGLLSYoa(a|%AAfCROgAi%_kgiDKFcPfVm#NOne}uSV=_F1I@SXQw*T88l
zEiElBHO9^OTa~XppR;cH`5)Y1IGdHN5W#Ubd@lfP)l3cB%(9)UU&k8>cYiwpTc4H>
zv~<Xq$B&3o<t$w_z#*mZxzv-$UAj;#-AsbDJzWlr89lmB_qUtBXi@*x)nUHdFh~yf
zF_Lp$+`x+Le=Z5#f;`7OpOy?tsfM+s=}AfVy$<t&Ewl|m;u`4+j}MxUB?jdBOxC+J
zV$b}~eL~`-!wq7XWrOOUWTrN0QnifY?IgkRXqkyLc>ak*VtnAWqS)6W{cua&kfBqb
zaXpm@9KlI>FoJ7KcR((xCQO@eHwPX^Fdn}#1|hQM&EI!|)|awxHUZ|`a_V%HZMm1i
z5U7%D1yqfUBJbmZPK-_91$f~ETC&?)ztHN#222_eMbJ57g=&nNlZoF(99XU!;A|zD
zUjsEla#}x+X*7-cH^@3X+u8XW2`pshs*M=~Iod?Ek}}XK5Z5cOg9t>t7ba#$^8X;f
zrM(&<=g*ycsKP4&NalYSK5*5E;nBapx%k5V_hL_42KqvcTAI)`bd+meneEi6aJRy`
z<sVk+wr)zJA`6kiO?BV>1kf%C8b}vWnHUkE8~XB>`MMA4jKCM5NMh+uppmT{wS{y$
z8va1_N8)6>_WAQ|mT@DP)^wp_AbuVw_8BF=iF_x6H1{&KPe3@U+sQavj@SY{iFw7}
z03$hr$lzqA-H-jy$tYkLNi`!V<3G35ux~5(7@9EMUao8-_&Idjk<iUDPT{0NM`IwT
zl~n0BT3sC#PX6};wmZPnBX<1ifNt;CVn22JcOdCoYCk6-b8kQ65JjvuIwM22Jzv>8
zYa#L@39fyo@<%2brUHh&8SdNW+U<-q<P5{at!es`>L2L0mjtb!$NFG2&@ErQc(IEy
z_J}8TUO~5EC+xh@!5^^Mm%G28W=Csz4l=xfnQ5QO&eB|<xxqC!*c8ktf@9ZjYuT<I
zy|{<2H}i%dxCVR=As|T4l$Xed$iBT0hUnPG2g)bB8cau?JjPnyJka|Lhi(PX&vWP=
z(m??)V^eMA4u}|bJ9HoPAzgD88vF5cqQMd%EIWMt+Rjggtui}{S&R3%x1af5l9H>f
zOuu@ywA<#kod&sI4M&f9{e#x~us?P9<70^%P_5?wOypsEzlT5@ZCUq?a6zhUyMKVX
zq0S_Xlyst!T-0#R*7|2bEI&Sq7$^L{h&h&6ILg5O&o81~!xq<d2L#7Mg4P1{Wj28b
zv}clV#&|-%aR^RJkpFAAyR@CR|69=QLyd2O%s_r-YZ1cQrOTIZG}E5LxfY@x5b`oI
zXhwf6|5l)NL!v@H765;@tx$4<%tCd`m0+s(NEIF7Tzx}SUk8ev1E=;0-zL1ezHtdQ
z!MCHm+fRL1M-loQGc&{O6UjkIl^kHLTs2K6I&-!iN0R5pGN%NgWc<h9XvmAQDCP61
zv9@5Q(~X7Qnq@oz4Ul9`LeYgkwrnA&qCDC1QDpKG%dT%&^ds99kxL(blUac5E_|Ki
zDq#6AOr5`k5|c{!B5))oaSapmOEmb}^GC?kt)Crq`R~D_X^jty%e~>#k^S-f7_5=r
z!o?(&$oFGZjK+_>y5_i=|Hn!XkPv@E+jC$EBM$_vWs?m>w98)vZ2~bQ7fP9iqajDA
zOuT+RMxj6QlIE>hz%g4dUcP)MCVB-%C<7fpS=7-?dnbD|&Yn4QChY=Bs4Mw?kS1Tp
zZ*L*ZMT?XFuD^Dn9sCkZ5<|a|vS!V>KX{q}1)o9C9nbyH6SV#`q_+55JrOHZ=Kq`A
zMp;jsIFWSrY#&`)U$P24&q>AuIM9~zN>qj8(=*@tiE}U2ULdO`dq6<db>Erg^VhS{
zlvivPQTy6wsNObW;UJd0!PtQa=I~h35t{bH_ML|>=1DLf12+<)iTAZXcVIG7O_=Jx
zkwZ}u<v?h&h>t!-IS?c)!t+EpH-3~C?o|BEf1w%XV^On<!pDA$_1&Zw%*%NUDlc{N
z1Y}Duho1S3#twn(%YmRV{PTJJwyJ*IK;%wi{*Z**P<}B=Ln}<%m6lNjJY`3ONE6lr
zPQsG+wJp51vHUy!BW<OtPyr<qA__UySi|6Eus(RM74yBoXm;Xz>)+k${=IR+!@b7A
zR?dym&JW6yWxtf#@8a)zqEy7c4OB{X-`PfBOSdz!5n>LN2q24(&pkcwI1xh39QLu_
zi=R4Sxdv3zTU;Bm1CC33M{RtdVcG-gMC$gvDK#hqKWbbJRhD{xsY4Sv2iWvC#zzl4
zcLCBcJhqMbQt;`TH{QF)TL8voICd4a7cdY4H5};v!9^dFwnn!OSrC7U)eOHt=bgub
zFq?V;li0ASO`o0=bvwaoIT*)Fn~I9(fuKGtvULXo1_z2=L^Brudue9X8P@*%4fgA)
z?jQLc$}}61m)1_I!Blgk2ua35tgWm<pNH7>XGG}i3LiMv0mZzD?mNBUpSthER2q#7
z=S<8PV+Ybiq3iG;Hr3PU{K=C%h>9$;11M7%X*R=}ASmZQaql1psG+IZ&-NZwX;V|5
zh36@=hXe(|q{?TXpIwZu>Pp<GPT^RGHh_R(Uw|gA?S5BDGn$MX0?Y$J4>>D$paPV>
z#;<3%2bVZFp+YdK@#57aNZ8%LT}JYQ@V<EZLx^oU8#_=74nFfMZmn#=#ofJx`F{-A
zx`?d;%J6r(|JY_kPaom;aHIxFS#w;%pc@z!5F7tG6Mmk%`ZKG@7-0mkzeXqo-B)$H
z?`vI9KtoLwb8MtXj(E=rK34w%m=?J2OMv{Qaa=>Wmkeb?+lqRen|1%^hfs=$mp_V+
zJa=w1&rhLT?SGPTe*T{V9&a=-?{utBX19HePQ&=hF}UZL()l`wQ=J&Mjnp7iGnLJw
zc)(6EaxIWsEAH(EwpTaf{qg|dS<)_#{}Ori>Q!2zBtg!-ys{IP&Zwdp-DiAjFAb>c
z%&(pq=t&9Sx46#t?U372up9lR?S9bP4m$c~g7-IpR1^O_O2_2{H@BtJrB;IYz*936
zILZcHeq&p64lqyp`~SqfvzFM|-CSEFXXV!cN@3n;sX;eL=rg{l6-uG2xcC}<u9-XF
z>q_xzB>RKs@j1S08Ds_Nz+!060TCEs!kn+jL%UWr8@vFXc#Zifkj$qxts5rS?RTt6
z3v>SLq5WIkl2Pjggs80q;DHjq``<~b>}ffvfpuMT=AZ?W?x=fU7CJf`P5=8hlbkdM
zgz7>VM+^@een6zXh~o#goj$&8*_oR-3;e%D1CiVV+n!kHAG2`#*S%kBfx_Jq42d#a
zq5<XUKRB)CqFuj8eA&PEJpsD&Z3Rhy`VXu3{aq#8{q^*T;zWsQ3-Z#sSoR{6)9!sa
z4?`W$RWDfu95LZ8;CQ2hzrk(0*fNT@c@Ik-eJs|e^{i1~-}pZfx&8FLkWM8DcCfLt
z9ioZE#KdNyrH~Wix~qHda8ZLv0J`bF4)E=wx_$F8xTKx5voK^>KNdI3LdY^e>@p_W
z=W;#OSlG#MG@HvCmgmu{bbrGzWTw}Fhq=EfMFc@1lyn>M$B_i?@WyE4cp>>|J1%qA
zZz&yRS4b-PUpho;%H7s)ZFSOzbs*2jNZsyWtz`L|_Md5%dqB9~$_*Mgi-$k!pE=71
z4=Pfb{BOAR(ut^Os5QDjaRd9P8c>;E2WrCbeuI(`=<n6~it(NkxMpcT%#sr3P7cbP
zb%L&6x7`SVMb?8Yp$ur2U~}i&QHxN)Wk}8P+<pk~hDwv1P_3K?hypcXL;j7ooQK^6
zw$jkhP&{!C%njgUbJ5S8FQ*dgrL}MskyQC!jx?c)CpO?DJL~b`mZdc-At+5?4T3?f
z5h9$-^gubL6-ti20@RrQ<vO59JP==s!gTj9-Hqh%NO-va_oV=?ESD@<XC?{CEwggO
zT9#dn+>ob+$gCWhGrSxgxT4$3t$j<n&vYp;lsH6$m`;G050pba)o}bzI5I9=xsO>o
zm#q3(wQ?<>V~vlk{ulRkv?8bS(K;#}@`m^=PUYl4IXG*E(^Ea0H*lfZD?tAjyh8vg
zZaeZIpntE!e6>yU3KTOMz>5KhG+{&@n|jz$bLg1|xDY&6j%WgAP%Ki3-&ig0l2y|E
zw_ajtjw8PUKW>&_4umHuJhel>e>UxbnVFeeMhxFIxZ>5)P+ZuK0#NtE>)z>Z&L2Uh
z3%*fl_}hV;=;%IKk!EFUK+Y0evk5HKBp6aPfx~`j-uSB3?Osi?1!72M^VF;mtr5DC
z9D_a{_HvA-X0lE`bhaG1?qLLdEmWTso6>#J8QtB81fK)39S};bU1C<27v(a7!XVqy
zC(pvG$B+K<CMSpcy+L&C>yaJ?Y&!fH+|KT=Hy`~l1$aG*FhfN8QjZkEngbZcbi?w9
zt$WE#?$sXw;4<MZ4630H09UW8-Oukyw5|VwTC@AvQt;g=FqIynz_Ch$yupu(A9n)~
z0*|Ah)jJ5k?&)&);PjtMbGtd6nFz&=ae&kDK6t$t_VtK%Zh4fZ6Y#y>Y!pAfWSi=t
zC%;cO%Tx{Rp7Y=w#=A7a_x*xMdtvz7TkP)ELL^c2`U_NF_}{PC$d>@M?mvelhH#&0
zBsDB342=M08>n07`iru&R{Xz%+WWJeVK(#UuhiwYB1s<*Mtw&6Wo>R2JWf_-=F9N6
zajd&meZ1@o)E4-EPG+92qWdu$E34IdqW>0CN|Qcdfr_xJB)Kbf`!;JbOMs#cE=VDv
zU9PV&hOcQtKk<xOEf0KVO&D)O9wG*3L-dUS4tP~O(+o5?Sk={Ln=iBO)g2H*fS5Bv
zP`-V5fD3LInNRpetUsS2k#U1wRt>^m5^<8!W)a^5m(t(uvyAr!P{sdS3|5}cVCDS3
zWpxF@jr;cPD^@Rnd~_IEX+`DOaC25K;l6!lz68j{_0q)UH#vwL?2pk7F}`|a^gbGh
ziv4VZ9M_}~>j$w1mqZz49Z}BZfoKoGXH7@i(?4_WrE}-Bx_x5UWpsCuo9S=X>1O0u
zs>Cd4R-V@>#mw*3Ts%Yr_ulh073XGmP@ZXF$QLk~4jR9T?!gYY+z=5twnje-{r(qp
zO8?V+C&I0QyMiWHx*FsW{1*J}G5+7u{(BRSlyJPi`G8@ez=$<te^3KGC8Aqj{#&(X
zUXJ>v++(NtwtF?2@@f+R4@r814WjH6x%~>`V(LFc%<2kS-s&??ge2s6qJ7R!;o9J*
z);${1{ij|7WQbYkNYiArMywyj*@E;7ORtOw-+(3{p0FX3wt!IbOp6l?O3vGofgsLm
z6K1M|ucr6s>x7Yhv)C(8h~`zpvy(iV)^ptzXTiXWkN;DR$G`TH=1b@uQTaPLZH^2h
za$xYDH5hBqA{f7n^bfc<^By?jCI{j61HOkrGHhk{4_pkM+}YD7<T59zVIm3u_myVF
zJJ=OFHgV%()#s1~bK-ggX2nP=6f5S_wxjzGK10kIh3p%ragDL0IJ;tVA<HJ~gz^&)
z%)qQzB}(eW-VHD{MJkH_Ms88_2q^pWQ=;Nncw}TU$kpKr1K3x~z-%w0qDi#<U4?GW
zy5a6)-3kf-KJ%Z8C$L&{E<lraGL~e+)AWUl7Y9UrgK!E@KFPcq(i?2%O=yig?DlGg
z05^ZY&DP#YK#W9Pf=WSb1Z@MY7pa0&3<=}{Z~H(m4;==awoiphKeUo`+fp^K^P31?
z1U^1k>Mk2KXn-VXp!y1I6&@NBvjd1}HQZx4xOlhYkpj~M0fK-{bNWtiD#o2WD&6&q
zr(@jA&b$VNfXB--zX<KxiSDsp$o8VT-B4%b$l=&i)Ma{%^xMH6xI}bLG_PIm;gP`w
zKx*OOKV8A|iz5Z#e0926qH}GSY68}BAKN2Cpkap?AxW^Fy$o~6S*oX@5<IA@WozFf
zS~(!&R5-R%`}OTrdAG$otL4-78;du7>*s|d{hqV0HzKmogUbMkf)!ejh|H6Iq_!k?
zGykf6x`G-^aqN<|6j*RfFL@QNu#I!ru0ZF1j%wv2y+2Oe!&XAS_W^oQ=YN@HP&&u&
zr_2vhV-?Id3W2#n3M?_@%2ELHs+DL~-oU+6=oT|LHw36F-Yw1X$^CBcrbhF&$g6DM
zj?{`&Qq>GtX3A<J2>7>t`?uJ99L8u7Ad+;?<Zu)rU{fEei1OVoX($w|CKC?uCqzV=
z4ea+5P*Uas>Isi~VSaAj^yaeSiw%exy8o6C1SA-k>|iHMrl80!-q8u`=BZYZs4oBF
zQCoV&=njzNv|{;49gcD~pj=ZLLrs$-O#b7<S!2n*nx!K)dC3hJh+(@h14)2US^-kY
z?3Piy^RZxg_kRbxIaXa=q9OznP~uQ82v?ZQzFj)#hDQB=VLZVL{$GTgfMHqzQBJzO
zU5Ui7X}XjF;{To{7|b5udw@s``~RZWovxZQb9_gmhtIrraZBT4eVuN<<w{feic++p
z{y$@+Uqa{MeJs9Y(V`>V$eAwISuIp8umx}mPI{*QvJFe$kwn;u>|nIElm#g`c79*=
z;W>tiX+VPbS072Qc*SV+Og{HI%*QU7Np<sTWLEIYAtGXI{4$`IP#=-&ufc@j`j?qh
zn>a!+e*E$e&`B)n_FMNBV6^r}qZ{z=&^{QRB-oo%I~lYeMeV?4mjJ2qExZRI;=@Hr
zeY7S_@ZWq{J)))YBO#SEN<JuV92@qPp!v>o`S*_<TuY=N!7SL;&t=84dlHV&{aBm?
zSVo=R6@hRszCeMK1ov`E;m>zKln7Lr<{BHchHyXE@hyEW>Uoozj^ARCalT8nOqMW3
zT&go5u#Qk=L$>>`ZhG%jrIC+Z>Y@pTHyNDE0+A?w{|91$SxEDRSamE@E>aN|ybUa0
zb5Su)%yzjRL*QjbXU@^cxZAaOp_qTSXQ2neRilB0zbdxJcKcxQA8_=PTM^VPWuVA8
zYLn~$GoPuYzANq#A->##519J?^X!b>|I$GmcYk}4avI@hd~~Q0G<(6z#kz3_L(g|m
zp<DGc8bnrinK#atGR33)AJDe?qT3niJqeBZWWQbC8ISbq$;srMvh+Jz9><@d9&-L2
zpypOA8-tAVAKtwS5z*j06i>`Ui_FhP-&^5vw>alS{s&rH`fdeH?Flyo=j<t(1N=22
zW*-=|vfITnjDlnf*+~$s4v2_xz=6y4B;#&bUPSbYPKG9D3($XMUc1TzWAC~j-=kn<
zU&n}6ci~ddOK6sowCo9vZF>)n&!q5eB-7Y+>p6#*$~VRL8W@zE|JM8EH?FC8OsyEk
zmef{_KsRLj-{$hb((mY);Z`LRFxOCF5rdEyau9MpBPqTbWR{D|u8Z03BE>GM-72Yy
zaiX{T@>AT`pZN}k)Ed5BC7}Es2Po&i34_2HrP>wK-laN#{iUfOYZ`4$TT?jT-@{g#
zgw&wfZtSnmPYXBmZ)xv3L~OXDfqLhsR{6^?DA_(vdV9IP_2)Nvv1$5BLY?&UaL~^a
z%z=u0$zJ*$33e5~KM-F5jehbiq=qs=mDGFyQ{MlA96Wn#If!lC+{Fy{pfm}Tf#yo}
zw2kGDy1kkyO<-_zbX@QF4l}keiFATe*AI&Sjqs!bd>1&QM#Yn8*uC<dX*PCtn{@ei
zGr<;BW_mMVRd_QR)P^^kg`pYiRO{wV{vgX;H0I4}?#fv7vDN$m2EEs+bvz=CsQLw;
zL)k=4P1M%2&k)_>pg8JF24i2|kxI<>Z~RkY=DZj%-;wUw9Dw}mzs5*=J`KskCHRf#
zEktA=hg5v|5Y?DBWI<qvTk#`iROlqsf_-W|TR;dN7yF#gfVd6Nw4<DgX2Xpz>BW9z
zD@5K6k1esb<_4xFqOg%KF%D82uLWZiWx@YilW~?kYco$>yXFq1idlk=9Ov@fgqn1Y
zR8q~Edt#Yivq3^1|K$TAY2B7g)EdpZ_{j8{sEr3i7I1K^%ZJuJiWxjQTg@to|079b
z86SxTy$RhuQhLLN4gW!Wwb*vHwe<@?9L{E;Gy;e&s5Z=<<yU}Zg<R7i4eNak%`3|d
z%L|8l)$MM%%R8A0QA`+Ig(i_xklKv&W4l9~5>PyyD!WsBpawUwO7O>BcShQBskpfT
z2G>#V%D?=j(&uQCvdi||t1)F+s#6Q(i=(9z*vkCKkYUrOl<mZ5Ngm_D{2&8~+cA(h
z|GTaJg=0Z(ne-c_64;{mNC0kQ%XXsh$jz)c??wy*yc>Z?D!c;SAoR&4Bf@q6Q3=Q5
zia=n~i>vp-71(YQc}Jws@zt-Oj>$EVO7QcJjn!cm0hc0Cn{aFl=Iy)#9B_Zk<KYtZ
zDSv+Cya`fpW^VyV^7F)$-uzQp=G+2O5xEq3dhW4R-Ii_8q9sd)gV7d0=D)r}r%)er
zIQIw9Z+bze1?%b91*4esqw~uUr4EWw?$rI=Y5{B#8q4UV8Hlkic>!pCpUrO~5#efA
zlv2;nBnA*a<ww$_-Hxy83)n8CP)5JxL7E{|TD7iUyiYLc)AD3{eXa@dB;!^?uz1C|
zH=vo;$V~k=+6pN|T22WwqnHy)Q^M%@UV%TdY?gpk;U3>%bbK*O%)(ITA$jNS`???a
zLp<`Y_`;A=2lYI`3{DdK#D)!n?7f>8skQ-e6XScmwYb*t@->G>&hCEs2joB;iV9fJ
z<iK_1bevosAE99@Cziut^IeXac|GfF&#koqEo64OUDx+iqypJIJ~{+yyiAd(j1$&7
z9l+(CUgiUFiqqUvH=Ryh3OcZgB^n`&u)f<_x&H@Y{c+^X1I8|0VI0b~iG+*?pk?{T
z9BdQ*wL>w(NarEcVj{X-eR)UBZ$XVOeq#uT0>tW;PGIZ(tLGk{GBMnjhNN^Ai4p?o
z{-a?>K653_ck8wnFHKNOBsg~eK+Q=bOi0#}9N-{*A1~DoJC6_j|J{TI2xc!-%GaQ=
z^4Vs7HxmLn^`Zp}q+2(9XBvM@^x0oaZ>fnk>P><y+@cP1h7F;=wLV+tA(Y$wtN1O6
z{CkYcdyH2D8TycvK`VP9?vobf=vxS`U;n#qvY(-{jUNM|pdG{SX!iX5py`{RqD{Rm
z0Cw3FJjAtR=D@upHlxtX1$LV#Xt2cGy)W4-VRI6@zt~qAnCGpzWWj<4^t=t?`MH;l
zr}s{|42YcX)utd;F^qf&5UWN;^WU50d=%_bf}C-%@n&rMMo0<LXwQ+Cv$C>?)b`Km
zm4$lFjM~+ENNm|6>tFADtp`FP3-nVJPoD*6Lv633QJt~ddT{Nthdw`RrNRws&QW-e
zX;%jEKO~4wtXI#++7hCy9iDK|bkA#21~MFM;Q0!+)VyuQKc(igRBDbVX7f!Q>gZ{?
z%d~)4N$*BsDEZvEWtUgn#ucRoQLsqA0<VaHjYr=%+~a!?JX_Ni@9HIhj<>9cVc|9!
z<yeGbrCBp)-a|9LxoBsvh!8B>a@F3%6y+YzbI+r9$;#<@?WjG`0^Ct6b_c~8E3S;V
zoSM1~U4F!x#`W9*=GyRvIn?%ouQm-+{@dmYTmI30l{7m-b-L<fz&j^=<fm@CAZ75R
z^Mmm!eG}iJq{z(|&H{{Gu)gff{P>+Uh_D)iBZ-$VK(8kh?3z&491hj6K<$MH`~aMO
zovpe`pEF_f)9zPj1CyFPL`p2Rv%M&eZL)E?jHi!XxQA+T+-r<+w<htRFBVU41WNo+
z#f|`+rFkN+(UtPJ!{t&|R$Hj<!i5W8BQroJOP!|nrOstcN3>FHdY$Pw(q^#^2||r6
zB`QHk=brn8sMM%wqqP$c0VS#PA(usB^t?U6G3gt;9IZPai=sciaU9#ZRx?<>_nBgl
zI&!@KY_sxHe!x_lF5gCjtGw79;bA+tM7L0Al_~6H2=40`f;;DYh)eDU`Uf)NJw;Qw
z32Pw_)aa6WsOuTjiS%SGHZVxafbB<lz5>PYe!Z(m3xFHe_1`HIYDi@FiU|za{T;ZV
zx=WKWi`uH=RA%Nf!GH_)Iqjg9O&;Ze5)63ir$@d#+PP9e$jCml47~!p^gDrmacZpD
zzdY^6HS`vo<l6j2U(IKTf#;y^4gooISkcbET|r7v;QF?p>f3*J^Gg&N(B9TpE!<#_
z5EA7FYR+{%1ki0`H>cm-q1#^{QpsoJeEgI#VHV}N%eOb(_?BuZLnDB=7oJ%#3Oh5`
zoq&uxbD=g{-b5zs+41K^rExY?2~Wcu|6H+s__Ay3U7sV<MbZY_0NeH9c6)5UVbh?T
z_RQM(Is0&5YMi8YCd&gPXAGN=BxoMa$vhcGMf}({5SRFm`S;fuIOd4rpL;ROs!A=T
z*}=WgI)AyB%{Ce)Mf&cfHw<%1t#B;^Nht#}ZGM1Wf>dYtTl;TNZFgTRzfC~kR;7P=
zsr^E0>tm<yk_yg9za-Xc--c4Jbio~(vC@T|Ex?4Tgkm($8vc6t{=q>V=5qHfz(4@B
zWXGZ6+c0C>5JDwkOWVsY8Ry{u17oS%?^}ctE+VQtSLa*o_ct!TvCWy7?1P`Ny}iZm
ztZ#Xad7*<x=i(f29Tn%%n5rw2xo#Lt0=okwN3>H$con6}%hMj^R9McRUk}FVv$C=w
zW%e1VsXq_R9;T+L`M6`?kBS?=;H-~7-@{HAcJ!e+A}55R(c&Rx5XK?dJzqh*03C5)
zv=^RqsvRge(?P?2&W_iV96e@C5#9V<y&?-gP;}|?Zida4qKPfP<fvv0yEX~KjDb0(
zLV~{I&ATfPhHY8j@27wOx|@>|Jy5AabDs7>dw}u!w}wmg>I{=^PXSo&GMFf<T^**O
zljx#S*`rmY3hBcH&WhW>!`J4-elu??j{||Bp`-LJA+UKlh1ll8f`W$RwN8-s&X|6N
zy;L<cE%xKtWayuj2gRLb@tmWOG%I$w1L0Smw1;^j_}PP7MOK0^!8poWyQB5IGjYmX
zq4!kNp%erIt{CCF4=CBk4tet!-4gDj0;=DHO#c{z_%EslC-4Y3d+Ac&dsU97k8@cY
zCXihWH3Z8|+;*b9TfP$e7_it9vV-lfp%oMG<g^=oB5k)>+jPhR{KIlRZv6H6bw(r3
z@X(50okrZZ;DZ}R5JSSGJQlsA6f1tZgbJz2*!4neYOxkKAzqh;Zt}oZAe<~%j~9qP
zUqjqX#)Aug$I(gEsozf~VA}99C@{cM>D~7YC2!DSF^@W$OQzSctJ+TUT)}xGrFWnI
z(v14vf?sOt-`NQykpf1j<#0&OlyVITv%KK-_0QaZHv1gTG2R82AM?>_{7#ue@V01b
zpSB02?gM{gNg=JUaXX1%9Ixbj@`(4ziS`{(qL_B`T<fFZ`l7+B6s`}DJF=mNtHFc$
z7nTL5rY;?`=<@1&8rv+(BF;T6onLb8zeNM)Zxfu^W2Uu(jLm4(xLrq-b_K<a4NwnV
zv-ahuYg>~pI;*$-uBm@tUz;17@v*w5GP*G1*R)vQqHCMO8Xujm-un1Y*5O6xf$Y43
z#adi69`kux#2WTEt4UfIuGl~>CcGaIDip;8dy4E9FjSaPR%<>(5bmVEtn8?)r(UI|
zD#t9=eliy@je`BGG!8>GRYof@;KwG2#pM6KPR&_ud*Jp;ddzFUU8yv7N=pQ$3H7)c
zD_5>OUo#bi4HRjX+1qQbF}ReX-uqqt$Sr3*00H_nRME7-#vY22GtjTuv&YF<E<I14
zg<lYLnK&HKYCxE^r_7AQB*2s%tW4uSB#xD$9pH&{IXakLfgm1w5Ok4WQ*N0K6A&nl
z4*Hl%;02m29gm!7d^FVsh^pU!YVB7G6i!Ujii{$Ly6WIcT&9{{h6=kRMexzaA(ev0
zPe7+UQnz>u5r9-unv{20ECp5ZNU;0Y=i{ptuOPv<9sC*gpzx36QhWP1b>5>L%rO4M
zCB85Na-Bwc#7Y0MLd8QsV1~<xfLEm`6GN*pwZI?zoOaY*BCjiKe-Y8MzWPu`8Nv*~
zdw1&JKZx$fEI3|6a-ye<q`85`$B6LoLGx&_W9tySr)}tw_S$VuW#BEsRQs*Op;dP$
zbq3Lv^qDhfA_m%)8a7!#AWlar&G{3eK>0fhgtoZgBgTvwb5A{Qern>3X&w$56RA99
zsqDq{7y8c-V3s>bE?YDx3BFo6>#5DsrT8oty}=1OJ-qxXjSG?|yVn+nipD@VTYdXv
zBwA`38hu-8(F@x3#PiOd(EU51Ia*oo5{NelDZ|ng26ld=7wwgpKT|E>tTVoK{wbXv
zCGPOI!WGVyH?}9YID}RT)`@z`sO8j<1aPRR4O(Z_J!jgty&Gtx*9wyR>UC3(;nI4_
z)W3+JYX|)B;<8z&Bq(X*&166Cy-u$sXiNF`1Z(e|*7j-Lq_q+Bgco1U7cN?q?G@zI
zhct}*ef5t0(4#i2JVEGMyg6x~jvRM<^{L3n$f>XLv&01ivV(s+%@OW9<oP+J=|7*Z
zHXK+zr1KD=vrKU2ZRoJ^@1K23!4kjklU#5WytGVQdj9rz`l$>&xn%Tg?H8WUQMmX$
zhD2l4TlYrmx~|Hbt&iS_<KG-Ad(*9%o0<rkdkw;mXD54HMbQwqGV+E(+H>~U__sCC
zeCuO;hek1{g42Ft;3;!UrPe&*Sv2Y9<=Ja$Xk0pb*0DtMHMS{N)1ZFbvF$zRsDwFx
zf@pT+-aeDvs!cM)-S(3$7<Oz<5t^Nkj}xRL*thFo;#Vcj-?TnU^E)wRX{ysPcFQC>
z9-b&QK{%T`T=Yw6Vxd~RP`_cXf3N;v<+2KBy`pdS%3bPz_rA*U^@R<{`&c~A`N55R
z&hrDEKS34`eM@0``_Qc20^88#PIGqeXvnkK;5Xs}ZW-E>I*pOwC-WKF-su18=lS_9
zb$iG5DUB95Tn|0f0DWnaWs$X;6zeGiV8J`jJ_kxHSL@ag+bZwlW^^=6&Nsiw?53{M
z9F=)7N3RcRLFes0yHF$ciDH!7hYMItZ(hhXLnE+xCqD`6|28O=_=x*<)ItbXGQT)_
z{Z*ZMnsZj2{02&{*1xsT86N1>SK`g`<JHP<m~9-Ann<qi*{W*iv<hWUgT7Zd22btz
zHAgGUo?|sNK|$@=gSmwYeoe{>Q55yX$v+}ZYtF~T>y$XSydx>XmM@saWijC9T+)BT
z7oi<em*<NPLh0<ny*!kWdA|q(AyVN2=z>|m|3xT8_Q|Yj1H8ujMWlt&P_8ofXy{(e
zCs2tV)G(Q(4ClFp{pd;N>_=LnOU~bIlU6$f_8n@hnWxBJyWT)b`<U%Q`^fc{t}8AD
z3sT|^lPpxR85iXmb6!IGWCzz(-o<;~>1YK~nNqig)XWU^0v~(TOyGSGu+n_D!&H@T
z>YZ^XRW?ZF;KRhXz;pWT^Mxtg&-b7v0NwPUDeh-D4>~a5vpG6L-?tb!Y~4lkwM0AG
zkWCDAQv?hs7W){~MVt?mn*#B0^zoFJmrs3I)%Bf#x*xl_jF@KH^U$j0Vt2!4*DM%^
zI&s|Z?Wh9sTzBSHC|dZmV>%6cB1;UjgR?|;Q)@`91?rsxG&X<-Ruwx7rRUrsK8$0)
zN`kiY(9LBP_)et*tHoH8;^zzMFustJv;K29lI9+hcF%1gT6O2)kopdGZ^)f~gMlC0
z{XPvogJ<C6HpcT-y3C<&%(1_e-Ipi^vX(>kH2_J2dKdI5^LRa2USq4ggT7f9J^4^&
z4_ux@t+_un`sIEkS=+HTWyk^U=!bxlGx=c^a>*v;2A}1bP(O5vP=;!rE&RE+!KKUW
z@va3RYT45Z+G!7kyIhO{xtQMK0|YBJ--Mqm>UvJJw&Q}l&0^x>DO={7>k0mbv%@}g
zK4cl>OcyQ;zg~G40LNIKJ5-CL6%V1ICwvcq3Z8JqW{}!*oEa}R%E)EJYm6T^4A_E<
zZq3J&i<9k!wyyzxvb%FP`o1cCGH<H-#_V7Ky1W+%^Sd>BE8U{06JDIS#DB&vk<#d0
z>q%AxCIO*p*98*M!2t=a3J(u=_;?Z-MsChrIB3_Sx%~tLYR&JiT;7^HynP%Vg87m`
zJa&ghl*I|AZ!Fi~yQON?`Hz#mziq4u!zr7g)ueofGzJuN!6_?xG&37x5vuHyAUdqe
zuGhQE>ZHt(Vil!2dh|;S39Ris`EKXUK?BR2qXPT)@0Z}NmojY%+?Vk5t(!sIjqWsV
zB0A*rtGv+sr&s-EdOuIt<{>o{GJDFnK96`-UJhm0$>Q0;b2PITe)oIbc|(d3L->nc
zzHs5;=C0LRPpgHhnHa40g_K8HU1xU_G84tz$#8twdAgGwS_t=wR`x|!!0}DUNv{@1
z??Bq~K_gF{V{0yDpadi)CMK<X3xEaaB}}x4UEw_E>{(II&v6J1xNC+nNLmiMNFNFx
z&4fgjVq2m1#%E(Y?+JV@@Hh3pV^nD9QoOj^e;gmU@e=p}f_ra4ymvp(T?xZoS?ucS
zdS~aO<5RY3#V{@?UPo#ZOpRhLAbj&}8t1CcOSZ3}CZo>trT40oj{e{9w@s*1Z^M=6
zMiJ#UIjv8V{oLM?J{UJ}(T2Bk@cwKSYqsNY1}!Ls5hekZ#_8yV2<UVA%o)|G1Fvs*
zdah3M2})&O#sv;%!rrH;O&~3~#8=)0{eM(fWOOM5$iUvXA-|9R@Iq;wchT~uU~ZsH
z<rWQf^@-f?!7$&d#<LeM7TuYNg&5oRWNFrpE1lo7_*H3CR1~RnyiW|ibLS3*Z_a<`
zXXC&zlu;WougPG3-E95VLtHCe0(WreDWYMcKUOzZ8?qmRly{1k9#UcJ^fM$4;424+
zDYNrAn_iU$#B94ic%1`=&myG)6eDIEx`lPvTR_7zGMHR<>Be{9d!4p23Xb$D;TLhQ
z9wBbhiJe=5C;?ma9n%nC_tAHrvUxtyw>z{eJw6WYJP3W#5)nn;MA)EZcl>v`@82c7
zhXFXO@ZMV-0Pdf~YlYCHr2DHD(l_mT)TOIewY>DG`+tDOXMh)?9ojgxoj0idb;Pr0
z&+zQy_wL<0c5xnx4U4#!DC!Gwh|G4;ur$7}(nx>nhp%GD3=aD$1<9Fi21Y3o`xr+c
zdhdR*7Srt#W^?8s1mV~mkL8(AjS1Z;r3S$IFIqQeb~)`!8Bp78jT6+}`4a@ZuC&Y4
zrRRt*=p1_#eR|8Ox$G{rwzi%;c`~$R?uluW#RoP+XHU0!OKoZ5{;=28fP%6`UzNg%
ze}_n;Y0Y*EbP->I+W_<<(J}#;v{T-`P7mjQs|Eq&&Hl0-==@gVe=LRw^vnE~<<;WW
zm@k_xsxa77#vHE7D$fh@RS|plLwo2voHfnwM5MsD2&XyyJmHWLvTVj+bW}LAQG^lR
zC`%&X*4Ea*iwV0zL4V<rB|2^SD3Yj78;y2N!&X0=m2_MspWEOS`xN*<Mu_j8KiiH{
z7}lHGaX4E`o*j70XQOJ(9__y$=qtB10@bK0ccYfIhaWN9ujo$XCz?u)I2Tnxl?+ax
zyK=c)LQ;})ZH`z=em@Bixr0!$8>~<*NM0Xn@)>Z5<<D1)S`m|pBIsuB%hrmm<-IGm
z`H(MFu_DiW&C+ERh(H(J?xKewH~9l%*uE`afOqBN$y;Lo*dR8<f=p#&L+p$!Ow%bO
z5sIcvvE{6-KuPD9<37zh+6v$M1ysg=-NFTwxF^?o?16hZYLQgi8S~DcKVPb$9BjCz
zh;}&)K1#lPd0h1t65zy9xIq@^vbuQ@&|q;fF~d90K`+>3Ch1M4TXc6(^Ct`6DJ!d|
z0^jVzecD(!p$pXbRJ(EyL)qd!tsar@-nM46CBWsR)t0aSP;jIw60G|T+?9>GN{wTq
z4)pKe9}S^-;R^4_JV^J9PP}wUqxN*9`=`=-T!WVfZ<Ln_Hn)|}jw97<DBrN0tU|7R
zJOIv_IkRMx=jvX&7t*$4Ao;u0M78sn?S}}_WYibvoFHM55Igyh+^XVbYao5Y!haEh
z$vw;?=HL!(gm*L8egbO4oAi#I)agM{$gzUTb$X4*LYH8@%=Y_)km|L+R};CI|6H}0
z(DP2J_D~=dV~}H;t^ucHf>Mwc=a8(m2G(l%{00mZ&sH&uhj>D;!s8~cAbUt^%||_n
zPA54nyDE^+UGc1&b$d1OER1*4z0XN&aWu3c*sCwzF9J;4iAq6Z*#~)`u;+KU5}h(U
ziL?XxBUD(`Y9BQMabL;(s7hl15P4(h?tQp91Ew4xE&u|=XWI{id9Z8f$Z3V@C%P7g
z=1rDIih+SP0R*E2Cl#XOZfJND#_kI7t>G8Me0?$-z0lCubp}n6bgn>jAasf9*s-rM
z7D?QDH_=xc;`i^{2mbY^K`M<J6reiId4fKcOdyb<JDvfoXl?b}<0o<zA*!i}Q^f3H
z9-~xiDj@&MbLW8%7(c_D<+i&g#ume8&}}EU<gQ}F(p@NTDYZ3%x4NOh4RyiX;h{8!
zeR-{TQ}z#U&Fc3jP{{_7sdE!Y0!$cwB6jwFLmbT{4cPOq1<C5T13EkER`7plWKP?f
z;^QAtY<uaYH4|2MmUy#rgQg%-(y0z+G>~qNO(9~_JAdSWSfx#f=`_1@Gr%S_l3r$N
znveMm=AtL>vj_7#P)9s#<sCu(XHf?f78Rl_+(j*gl{HFQXk(KFT;YKIoZrIgs;V@F
zA*N~+<4K-JtSx}T&y;On2OWMw;?Yr2a(nztD5ItI#IX|Wokg~Si%9^lleDH%UuNU$
z^Gh_lOuFfYh9&wyz8u<?f&Be7^x#rEyYB)jzLXc?`0oJ_x!lrHQu-+rC{Y9JdWMVd
zpH~K-zGJBpxlzTTL_`(+8xGJPui-qR$TzH6S+8p6n?H7WaVSn+hA--4mj@3i-<~)a
zWrtodZLHIJn2A17qm#?#&xcaN{<)$jbJ^9qiE$3Xi8nYsRGt2wu^kAjcVed??DBx!
zsj%!|U~(e6g|iz9##E(beoK9&wY5L0HtS+sXXQO;#T8p_Yx|A+g(-4(mf~z0_FPVa
z7U&tUl$2smh#SyDp_w<JtpO>ff1Y~|_<Lq!n?*;HMNOC*r2|o4&8uG@%y%x4{GF)j
zt>&;8KJPcQlC<3V1CQ+2k=cePZ}E)Y$}a4Clpq${*bF&La_;Ck?Rq9)-QFKEv7!H-
zl#>@g-532aEf&aM0OyhD5tD28YDbMsS+G49)#p18wgO4~{vt)C=-^ez_2>4TI=1>$
zctiws7S3}?=}Vc<^N3V`(Cq8HbTKQz%1@Q@L^?E&eSW4G{CdaZ3bfz~uEN?YXSH|%
z6qcLExPmhX>V1nhLb$;2?E~+YLi@h4&k-zNg+frceGY7P)s2<E1JlvFGPI(G0e>&6
zJBwliz&ugRzlak}4Qi^o)vRMss$CL1)VHlZfD8N{!_~J2rXgMFh4Fa$&bo}rCiq#g
zpC^{VT%;`>g{F&$y0gmbo+kTTN0EB(Kwt2JQ4y=H8>kQJ6GFV0S6wflh`PiW(Bt!6
z5MNwKPS#P?y~cTgur#nY<8*&L?r3|=V}9rzUI5>St*dE70csQb4)|}Y`hC_1;{;z%
z)2}x?4aY2=O}G7e=jW`2?i=bAeIb}!7l|voG<|ctQS=@5ct=b?2S=T(4!}JtKchnC
zY>)FmTa8G88%?w-;yPhw<*nvTR;q2^;_Q5}#laupWda!f)7sRMEs%!*4;k3AV<k$Q
zL=HSZ$_D8K!OtDdk}&yzqk;X+(bD|=B(}XQ*!&w|5dO}e6#G0yhY5_pBqd${&aX#B
zC2nkaXcLbcV;&FMm~e$}<@Jq`?Bju7s2o%I34TgjfPM>N0f@S8M)4Fy>@3LhUdn1H
zfQhHs+pglf5p+5Og*;zP1IDQScjgRni%}Q;6wm{cVpCS`(r?MPxQS$IQp;3jy>_Ky
zjkztH_jKG3rN{?>8|4pwAmdgyMZNeW`?@;5?QL-i9FA94l#Iam<OPTt*S0qzaskc|
z4-fB%=;UH)BPIC(iks0wnZIOMX*s+hBH_uG)R_DXC({u-imehcbtlHS+U<BIdzL(!
zg#<?^F+H}eT7~~q%p7Pd>)zpkc%UXs3U|K&`tvr~!LrjZ-5Ac-TGc#e$Gc$9y^ieZ
zFY;qzqOys`52!a#dWbZ~zT}g_gyp^)6e;lSJt9ynz8$0kIg6i#b_g4n5fgz@p~=I!
zr1UQ88aKc&KCNBE-~g$K)|TLe<MTq-Q@IPVq@1inET@{(-~$N{;hE`(On5xP8hC^?
zgkKi_{TONIneU~AG;n<cFcg5*YnwkmC3$@DBgz>4ZEt1u-YcooCro8cmvkLtx<c82
zGXD=IoDT_qb7IFkQLgPe$!v~&j*r2w*nt{0>zU2bE`(Vju5Wg58DPhCFeZURw=ETt
zmw^!F9sJ4hhjWK=!q=iGGFAyXjIkO^MJiK~xs|_iI=pv|>B=%^BOD6HFh&KO0Lb+`
zR2hljyoOSd!otF4<pqnaty?a-c%O&CD7tTqj@uugl%o7OHEi%Sf;Uz!kD3SzRh5z1
zX`b?wIpJ5eEd*K(s+>Vi<39$scK6!zUEkvPl&8$fnHXUG0c@e>99}Kcj)MS?b00Lo
z(J;=)^#IlWgw)5CjIF_-#@L|CTjGuI#*G_?-aEja#iAtGwj&V{0S-CTwp?B=FY6G(
z?x3M(GC&!g6tn<|b;0P-qy6v-N5|dfyn;0h5L~BR`$dvo^Lb{yZtM5gEvUc}9_kAS
zF&1$VW*D^qP1fB|c7}%KLf+JHMJQ;j+H=oho2Gj<{U#r2X<3}Ul?ePOvHh)9(MO-;
zGn~rf*%4FO6<Y%{Bc5C|$zi6I-vv(SF}IT1EM;H4zTT_za*mjX%jpCiHtavsdO(*`
zWcMMl-B@tgDx}ZPaE#s{_y6;3Z0AbcBwLrDGk?JXjBMOs<@c52rRG+`DnyS<8H%nt
z9)A%-wi?m`dxpsmA_WdImkNSnyBU8s#9_Io2^#;dU$`(fVTvo@i?Xt^L9!0J*_EaI
z&wuwfaA%<Z+6h~{o2Df^e-dWG$E_2>A~Od{K+#gfoxJxug;%9UYhd|$_wKFuG#DLd
zx#0@ptbw`)MccS*$WL^*cVNV$ky|9!py0TpSiP1hjn6^~XBa10O9t;Q0@A=@RDK4Q
zMgu7X*X59!!5sFamo9h7?jQcFps<j~J?I{MXb?_Qy$sUggsI_%v!;e|j42R)5i!gV
zWn%O^i=Un6{<+K%ief^&)eATt$CQxpeH(D&&vFAdsgH*y2VJG6pV|TL=$`&RAG{Gw
zr$9&)pNi@j#U*mG1!Xk{$7V#_kGbJdP-ET@;}_VzA@1UtSPRBP?vhj$oI=OOH56WG
z2?D&)WoRuj7hT5Sdxk3xLu4{PE_54)$nn(CT!$Uoql&^~x!)!Od)&kS_ghPrKm}-6
zX*7$dNQf;9;%w|;k~EYprQPEiS7zl1cEo3b2w6mi;-7t0MJ#BGUI&w{Coz|<xw#n&
zcbi=}gfuReh8T6`&%baq)D6D~e53oq1&OrGzb_mjH?bf*PKl8JQG&{jFn9*_PC)f|
zj7rSx8ifMPgBa>A-uiIDW9`@jNPGNeC>B%v5WL4}&Z$!tv3iQx!BC4L4w1A&4Ev?f
zZG9QH1B{1h{AUh#42M+I#Q;R2{BFzV=+3<tz%;<{!M%uJfAnTck&Dop6lhdOakB{8
zk)VyhRqre>hD*w-;=gKgY7~aFI!X&m$Y~?}PhqcxJCrqqdZ(bTpQl>G{2Jj8(v^Oj
z-rUUY{%aFSO4G#U<e4+z_03EY9L#{4l1CG$k3;n+><8a*3HpK(BVG=1P5?P)-jnB8
z2H0u)muKn;o$w`G<UaOu_0WhVeBJuXmoI!5?e5lZFDF*n*50{exIKOXT%=^ry)ziJ
ztM66ykT|T*fPn)87pic4p&G}vn?hjL{OAMZJ1e*)(<Tpxz9YV)i=dz9xE09HYr^zE
zs}zy#$+$Y*yj1NdoYrM~qrePOGnmDml~Wjh92cjsa?n(05%av89*n@<;|R(OEP4|)
z$*c4<8CI?lm}P}(jfaN^mJeQSd8*TRRTDNK&?2}qZXBsY^8^Imw7EBcQ|#Gf%+$)e
z{o*rFRRCxjos&Tl9IR&0%I@wJAm4GLeqK-ZF5+5(TCxPMi9Qe~U&6GQHUZ(`wR-a>
zK|pYEJay5Vg|ZIUVeGPNs;U7rjfm%N#NuweyV-ovrFTV;9aXy%EciAe9dV0)lQOtl
zC@7&1r7y>SJAhv_Mm6AXU7$XiM>@@WE6lspFU;c6ubtI|5o9a$K0W#Ou8rJ;gaOq;
z*mpDrq(B%uPGJojwqZW)Doddo2>W>~3$q=6-wNxCWJ*nk?tG(Z?7vP(&KoyF4Uy*A
zd3~6x(fM<O(0FDkfIQ$R*Q}I~Jx+)1pNJQvRz4MPzOy{97Bez9Hv^)NNrGot4}2dW
z-!R!*!$=5$iz7dV19GL|5}8tenJRNDpWTZ=^yponcG`cT3IZ5l#u|J4WZ4rqvby}#
zW2TX(GY95uEg@ytTAG^>%fGc5t2*69Fe&?nEA(a;uNHi1%ch1`H@gz*n78z;0P_{_
ztxyOn>h`UDkv**MzX5fOB*CIJ3{HIXz<!NPRn_emp_ssb<N(Fu1Y`CiO2GW@H96tZ
zC6I~{Tad;aYAJ(cwHcm;;R-DR<pe)mVn~u*JPYPhV5X;8aY!wi-Me42ng}*(Tc|>M
zfL4BMJ9k1<B_?#7YySZQP+#D&p`2fwyk%|i&yJ+%O<LXGaUC$?i-u@YdKs?Z!LHly
zhwJph#%&skt|D&aQ!PkFfQGxTZ^m)(?RMy*_G8gx5Lc+Fp)=&Vn!#btjg4MtZ*PxE
z?zl~89!8r+=f-mOf@-!G+jAO^|2S&Fe^6|$;d7DYhkm{bPN_ANa4bXnaPFFNHedz=
z>+|54aj!1t6)!HmGNRVbohbD@yQ_Y&>dzdb=k2i_EwKT=|Nen`ABL@_6qZB>DZeTm
zG7&=Vetv#<5`=B$iXLWh@&vp<@7@=w?!f<;MT-{;RqStnS$=aZ&y8L$D$()s0@Q-h
zQH+=sB5ifv9lk(iVyAgDAY6tKs!z-LTlAsU+}PjQ@%!h0Z<aX=Q<j3D?&tR`r>!uD
zYcxY;$dsgcG}grJEI~uyO&-gnkdm_>-7^xX9c5}d_M29B3kBMyK>fDcc+wEJ)WnSQ
znVGE+))nQ1X~>s_dRMU<q>Y+xT=-Mk)UIpn=0|1_)my*sGh??tv*vLiyXHFX=x8W5
z>Q@nY7|*<DlhJ_uL}RwgJISl+MULQssRP75JCA6!NrIgZX~N{&-ZL-60%hS}%BN*z
zf))D#v`6!PkvBlCDnJWA=Db2KQzBVt`(Nc<q|8xYs1<K<JT>W>NizHp#mlCc4Zgzi
z=7t^b(J0-?8=6-KuV9bD#hslcOwE>iLlfR;9R=&&doW2|AlQ0L%Y<Ax16TGEbT|<D
zfGz2QZn-!GHpmO345(pI?($)LtJJF3eAYrKYoP7o#cOrB#@*xzW?|&s=(R_<-!Ofj
zws5o%m`DFaf}~e>t5=n;57ha8<aFG`3Cu&4cp4^OGnn;fWANSQiu!R3;v{C8q}UR0
zlft<sd-84ASa0Y~yW`8y$P~c!?z`NH^Lj>e942)_Z!PKExz7Zjs<c0&zaf@8h_>k3
zACJ&yjZa3_?&!1r_wd!{Ey=VhndEU`P}4Vmzi;1j_$d??cu1j~^+bd(Q@LMb^k}UZ
zp7s1nB{OKTN>A#YH1GIJ$<*AAR-_-`A&uWr_tro8vf~rw$WuCp;E6o=VoU9{oN`<O
z(!btH=NZ|9f33t#D~st9Yqz<Zn}@U{piapNixk$t%b7lxI;!vMllv=tl)VEr>jn08
zZW{p-O)GRTKD#w3U(HX|cKIjv%sksOgsnh)q8Y~fSefolkFB!9q3_8b;`YA{i@()U
zONuFK|C}6e#s*XXWK#Q33W44S?JtRwoSui)C4zDNiG5Zq#G_9mh(i8{9^WWwFF#Jy
z7X}O>JZ>(gkm&KBA5L|*gsJZGeBMMLZ!R_xgY>+9gWno?=X?>ctorj{H8en7XY*uF
z8SB-{qt;PG9$IOPL<jV7b5XrV>=qj2<e!Is;j!Riu;5L$UiJ43#<bHk1`AZo*^FdC
zT~5cJ9HW3fjy4iIkOeK+c{{*C4?A~XGo9VJiP*WdNMvXbB2pGqj%fCb>ThAzB2%I)
z`lRfm-ml5ITr;v4W~5akr>!CGZ+pXs!Xg#(uHZrqeQ-`TVk8*?osqxc_KlDe*whw9
z^&|G<e9?Tk49N9@<J3buSG1kYk6r^Y+;g&wLrU+w?w9DgzuA7X@p#<SQl@mP-4z5T
z5P!_YMhr0N$|=biSeN_BlW|4#+v^X)a*dekG^eo?s+UhkmtmOETv5c^g#PsT1&*Q6
zhcmbbZ@8YzrV3HSO>`p&R~+ILXz@`qVvOE`T6`Ji8Bj+RzzNUlL$lFz@}kZ+dD(TK
z_>Y3#wP*+^nBvyALs3gUvM0=IsmF@|m$|>*`1-z&tiTvGG6x_fWtB;8!r)aYCfW4P
z?HDH^)=njgF^c0}V;T3jEksbyjnMP@17Pz;+ek`m&~qHbU4y_odjb{>&HaMl+_b7+
zk$}N`n$O!01#N&7qHV5t4ipGYz`u@aGZ*vy`^El=*N1>#x*_&0pjgrdyZ}^{O#C=-
zdqP>W)b$duZF>yMa@xN@b-y!!2^|BVsFlWtVFSWcEGEU$23)y{h#dhCN(h;Y+z0}p
zUQ&xUA&+ke>>8zReKo;#=n543Kjkze;|1?0lMWV{ilI1gPGFW-m84Z+7$C<W29PT7
z9_cdmwAVbGnM=jdyJ|_QYtMI+gP4`>RSM3@-Ma*BLTGd8(&y(|=)vHM6V$c3GDlp%
zq!dQwX?owws3|ttd*CB~DK$V9<i?Qf;IWt0i(H_)+E+9i$sNr@Ra(WlR^whnPaTg3
z>BE`pXqd+%s$N<=^b^ArVIF@Q8+8Lk=+V<*2u|2fJM;@R!o%9D?grhr-4!CQlhjcl
zactVkk`;i~=?ntaX?%tCMhl3cv<pp9+OVO|K3xtv)92TZj`kl0^^WM2<ME$jsGHnm
zs=$ZZ?`it1)N=$gwxJSb_&IatEMId6r@jNMAbH1O^vS2i5qcj#RE`7qM?U8_xD_`4
zRT>X?9#NAGhO4ZstpBJ=H0eAE51PQa)SIDRw9#@~)Re8~xZWoNiRnp~(gah3?Fm29
z-}hzgW9S=5&?C$%8@&ZhZ1#{HxZKi;(>@q93DNyRAcNY?vdgI&YWaSkJ(4Oq<~?7P
z>ZTKO{@o^{9d;lYsYxy8M4(kja0iATr&eFSMJ8?{_V|rkp?XaiNNfrR5q@Y{<v6~3
zpXvVTDx^T(7ZrkHoR49n8{dz(g?qN=+1w%kiWlxD>qFj*4i@qQ*UNxhaIORHR?Is-
zg(zZiK`QJw)5eR<qs}b;Zw3yBZ+4-2HTi_l<exv@ya5irSp5`Ajmj~Rw82-vN4`IE
z7dlbber_fvCLWC5AojuIui!+~*%xesk&TkB=U@zjzWa%5h?tp^iZeHJT!J2qrflTM
zcqZOv1N%&&E<$=hj4$&*=$+z$C<5wQCFh{Vf(N4d%RaE4+)#i+1)^K3(n^o-d}M6<
zr&zxQPneedPO>J)j?kzPoa8Z=((f=AsX7ssr4ia(8%rt)(f;-0jZ2xwPug9W*$pN3
zadi;zpXl=!!WHQ;k~BHsJE#oHQ-9m{@OTUJL$W;BW%SZQlXK&qo&vv-FvX739E#@u
z004<VI)%Qy&X}-1xI?p}L9?dUjMdcha6UrEqru^nP*6NLblN@<KA^uG^ZbhzPJAJJ
z6Sbs$5G+B<D!`UVfZ2C};0OrFQA)XT=@iJ{_W^hU+_Mij2;wi_l9AIHE}&P1o76$}
zU~H2zP=>WH1;%S5{=fgYRHrQv55V12_Oj{t;oEVeP!tS~v&2xpBjb`ge4LVi!XrIU
zF|nHdrk_xlng|#fe`(bs{;@4?Z3l^!9<dVf|HZ5K!omdV+h#raZQTmDu;z0!dRoKl
zT>s{v$gH6fgm%2I=gwae3%&t=RDtT3S`UFM%{1gvRm~e@BCXEf*7R+eQ*8bTTNT|;
zAa*mTuWK*rxZ+O9>CSi!3wf$OrHLAOlk9w<Mk_kJyrVI;<Ext=4p4#5pI2{&ZFx9e
zF*pqkt_akHdJEj8nLU21N@;XTBOpHk0fiK>d5_m6H37GG(iQN!12>*=p;;M-ipwB8
z1mk(L@euI=WAn<sAS_m@df!U>Y%@<1U9}GorW5VQ@CPEmxuOC=r*Yu~*Ce1+_sENW
zULfxRZiP~onl~m?5AG{4{1?>OlNJpC_3GGUd;$eFDNV$@F8h*$QB(8@5LSQ3d_vSF
zv_ym!FSoI2gI4`fX#q(i96PZBleCI$(Wv9$uTX0y7>+KDZF#t^kMOR8J0U-?d<`jF
z>#P<II)b{74$cqm@8CAO2<!mTg<H(g(f$iHP`wY~R)&Ty@Bx@5sxK!iAgO~O<k!>;
z=Ew7*!2D6kQmaTZm(T6-_B}*ePJF1m>*CWWPrr?wo#+$Mj@ac<*%!>g)Wgz{HbbE(
zxfA_xb)XRXp*e;DUc0=~An5V#-Mi4&I|+T3cCD*=a8yWA#&AUt+3>_uOlN3k?28ti
zybHumu0HJ*eNa-Qt{=_jir6VbTBSd|;J%9!P|v&MD*hI`_7D$8m{I?QVkr%k5kBab
z=$PFu_Blnn?PG^QAL?pW%t(uiTY+1YWbE@NxuZ3?<ImB6j`F#?bR~NlwY|#uyWDb$
zC&(>4frX;M9E65{e+2X@aRqKGR1wLPX%pV#>=ZpPdh6T%(Kkvo^a)6SL^(%9b6~nl
zWp9B|t1($8WWNB({wCRVKySdgtrZ#Z7K-dvESK8?xES>xk>Hyc90)<c{RW2~aMDpD
z8ewWsvL2ON9!G~`mM4m`zQM0VXu3N$ng+uDu0z~?tf<r&_!+>|5YB_&qt(i#WW+68
znnLjRs(&HRNytwPhuOdx+$;DNX~r>W4-SuO+}-|5Y^1{XLuGpnPgsttS?y1#PwmYn
z_@W|}p$PC!pc4{lPr)*mcM#^7)JQuJdH_s;i5fjgRL3e=A5Bs)O&}Sw##b8+|L!}4
zsO44dP%`^{UWmdZ#Ag0OJFI5Dk33JN=3izS$`nL#ZCjv)u;!vp2|Q}A$EO*;21b*2
zFiF@x%KKt*Vp>N1e$t$yTn<$x^U6R+OBAwbQo#sKPLdN)OB_R0s!;OotEWt;3}!lx
z&JoVoiwaT8M}P$@pLxoxzJFkR9p}?1354%PWn!?$Uy5z(_b1!lc*Z_U7-OiZ&^1aZ
z7Nbqz=7naqbKs~x#Mt$BJp=FryUf3dsAY@4EYq|}FZL=$eXcqd9OCg)6NAAs1w%}J
zhh8=!mHWsY(J%NiUAztrD+wL%$k<4y4#N3jLzaeqSP2baoT9lnFOh18eA(qC;YZPw
zVl?xu;o{+&;3Z<lt+=%-ZsK32YD*j1el|Jz2STL*o-$OTa{Ty$B2sfKs_VcNXdy&I
zisGpB0Eu>px|g$`J$yUprv8yZ12yfO%&G@-Vg?0#=ia@0MN=#>^dN>?+e?E5o;OH$
z^NQjqA8Ne_>x+;@6aGa56nYd#Uz}<DYAZ0~un)K`lfL&P0Qv6{oX@TAmkgV@A)J#8
zEYRD8v<nh668X8PwbT7GIAp>3+C5OsnXW2w90<+4u7<79<%q*osKu}%>cqH}ek}<`
zE<N++@UY#tJj%ggGDVVL^?u4NWRrugIhs{sVdQcez=NYD-u67AK@_bxYNJPsncgGq
zQKg1pPAVIU<WVRw4)-H3pPT9Jr#nMz>58afRJURtx&$6$eMYG<GIJUapV#t>G9Vf;
z@qJFh&6|41BMFN0!N5`-pBGq{Ohn%(z$RWnyZ>In_nkqD2Zx?DE)S@dY`7ozA7(0D
zm9{)}I?J>E8#35^qbI7roK?1;4mAw*^5{WWN70WTu(Q?XN5{pxj(iJSWd3m?8pA65
z|Lz42_=`;d9WOw@5Ka4i^Y3@rq~Yu`PooN`R4oRsY<BCk&%0hDHkiB;IF3PD5@usQ
zWUPeUOW<fY1<TH8LyEFIdh<;*pP5u57m(NK32tBqI_whLqTr`DwnZU{-pWaJ%?x1=
z5QmDJ=Ar3SOR2LvH<TePcF@>E{tfnTdsxR*c*#^`x#NDPSvtN0l{y)+X;og{uufS4
z3didr)##9XI9-ncGo`<KgV6GWu!{Ss;R*nK`aaGlaZ;s*UF_yU*hyctO*`w2E8mwY
zRxeTPH)z}~_S({@?rYbsDVfPgym1d#Zuo<PoI*n^Lw5Ec`G}V~d*Ks4%H_PQtIDOB
z6l4<tBKX%rr_W@e24LcpwIOPUzOYM-N`<Ow8+52|^B*dvR{V@pD`xI(1{x?QR36nu
z2MHSb)~9-OXb8kv16w^*w_W>!U3O-M=5++*{+id_KP{lgn&YvQ5r7zOvQ`iM?SOG4
z_Qi03r$G+4cJ(mGUx%J%{v{@@DQqs9X*cS9LW#8D;h5yCQpAjLKjgr=C~}@FJ7545
zevS~&?3FH>M%PCWEew0?{JQTTEeGn$i)tRyj+PAR_H6+HK_{yB19O^sN~Q%|vLpNP
za;|5dM@Vl1X^2YvRV~!hpyU?CZpQgbnWI&IQuBMaS3~B6SK|yNV4NWe_eLk)L%rqq
zmdDe#qfL;eMVe1iCS3zGe|#kA*QdYf9WS9sr_^>&46s87BEiERd+jmO5y!y%_XuQ4
zrtK3TI$?EZpmnN~wT2xBb~Ov5v1FklXv(^_qkfgu2&R{Ll%Nnv#qL@47%DmZhHq((
zk9YCKPqYLUbVj|j))a(?2By=fa&(dG@one_O8~rOQ|6~6s_>u)1_nRoH%>;oE#Qyw
zFxY^bPC^UF<UQpZy@?U#-9xVZJTP0$T1~{!zTPCyh%SM;ACtVSrb6`<5mHG8?1<65
z)No}I*91XK@$3zduNc6}Eu39j8qufPNW0N(ZU6p!g|)>StmJGvrcyT}l^)(8SWcW}
z>7MfMVkK*F;KiuhY<gQ=u7T|y44OqCOTry!C%KfI?2`8i2j>+~7%))(-0Ia&@b5)<
z(K;d&rRassZB`!PhRYv_vT**=s}80Bs!i&szgg|jDfUhfxl8oPJOhK<`aaXdyMS%N
z8)7aa42-jFLuJH~eMw?tr!0p%y&`E6CM=*Z5s>T;M0;E8170I2q*MI;W9ZpRV`BpA
zEw896nNpoDzY-^pq{Nln4zWuEIkr)Arvh32PofLl*@EiB<aog205_{DJ~2I5i~|ou
zrU&Au(DZaDE?k3bhGHn1P-Yag|H<jtiSp8?v_3R|6v0hWov6cQ9jU$WYq|@a=9n%p
zH!ngT<In}iYiV4Mv5qRI2@514M9Kd1KSEA+yHl6Fjd_1)a$bKN57+IPN))qHy?WEW
zEzfGXerw4={~3qT1^<a?h)u}L5}#=80RBeVPtwnl39q7ctJ%ADIE#%0{p*-WaPk8V
z=OP}yRp>(u1+?4vpE-Nz)`i9W#%ag6<FOtqpo9c(Z?N+rlnb69ydT)pv_?Q6;S&k5
zA@P_UBKP$5zS}T+OD-<k3jRg7zWe>1GJ!u~cKh~!wIC|7>P_I{c$%<oRy}|dLHGkm
zmXvk7+4ezm>ZGoXepH_~q<SB&sB+fS3g|WEGGVVULZYnD!0Ql3R}?%9LU>*f@|m{~
z^^Cy{1LP5XVGw3lIN~G%YxId4eoYC-tgiFv=V2NZ=Vv7(I54vkndW2=-Of}MEws!d
zR{2o*uIlwiVBu~vkr>HT9@iK<me{PZ;Q;kAzQe>xcr5sWNEjXFQaw!}`{3|wIPxdp
zA(xh3d53d(40ZCd4vX9l2u@_7MR+Sf0VJ&_u>C=@Scdaev3se!H^T?yfU;-L<YYPR
zzKY}GAxkW*6hr0QE8Ga<KVZpAVnFLO&^bf!n4TIc>2Aw%({`agu2DVyB0wyW-J^Ut
zwCC`esOdYm85kItC?}seGfv5EFXwO@S|!aD-KPWi-kpD&kxgwXmiSCnSqqTfcEwDB
zyj7_1H4J-D5-h#IvL6%Tz+F?kl6B@xEgH*UtXYKLaOA85BKvf5e&5p^!ncC=AeuGE
zeUB#+pqemHUKO^}klKWVQMEUWTwUuy;NLqVilu9Z%Yu~1*sQ0)NIZWq*pwc7N<f^B
z?Y$-%oYg7(r-Z5}6>)agHymk601-0J7DiseL;DL5>;XL;zjK(u3l^k+tW7OlUOwnB
zIwoDRj#J|<#_1U%>x2?j&GW8C?#T~EZV3ep^lo3Ad>J-x0<-s%Hz2j6^Z!N;q!Y_U
zi?BAYJJ-euDrhL4A#lU=k0@!W&Yvg2IawzRL9~7#)P*)p1`EWdp5aCaH8b1@W=H3H
zd;bOJRKCIT8T_XIj0L(J8lYJLh}KN!Kse2XWGd?OIX?VoF}6Z3d;{uDRs8Qy%{GD+
zaVMs3zr<OYnKVi)@BLC5Z3fD-q=}fYfXG$ga`H0c2cyIJfU}w-6kMEe>`fM9DnX>0
z6ic8GMX$f`%qXCuh5YA|X|UEg1M-t;qbV5Op<B?x@h9W4w??(~1z-fCyA!qvC`^&-
z`x#OUr;#<HehH7!2e2_BNb+ldcl6Pa`PEmCikWwO-MkZvqw^-DCMLK5^hJ?gUguGG
zXRnSwf{+2>>i}%=-`0DmVEb*dv>;8y1xZM-VyMNcTI~Xc+s+jk^`VzQ;4_4csH}MA
zO6YijuG_CAkfd{HG6PV%h@zf*D+O!K^ZJcSCQI=q4@4KhS*y*A!`K`2V91uacgAgj
z3f$BQUA9W2fov72b{L5N0#hTJxk9fSaQ+}CA?Q6@RYPN5gVb2X;u``%;j{(Fh>+Dd
zKOG|B>r)V}Ceg9^lQ~2%lAcE#mim#^I1kZUUfN~2>C@2OiVIR;?*PN1G^OE|u4}t|
zgH1L01{A($6sLN_n{cjK8r0DGD?YZbK=-?$!lxkux}#8r@Wg1WG_$AROw^&@@iRRW
z45-6uY(m-6z6djX%#1@hU%MLoMuYJrTKuR_mowZCXP80bz72kzp`nGGj~&v0-wT-L
z^%SQQg+s+q_r3xSZ=ksK62+I%&bW|&f#SPlixx+lU@Zri0WJ#tVFSNS^GemE5FJyy
zTnKoo%kjvptk=i5*6$EmKkCD*9o7t4iM1T-H`CkO8@ABsM3x*t21%1BVS%eo*e%@8
zi&_Z?>8Pugwlg6&HN4Nulyw$3)kZhB97kj!;d?)h3^8>g-e(jeR%$>oVqND2_3wC4
znJv6zQS8oM>r|qn75h8ixhL=Ex8h1qvGr$lg8IiV{BNM(?Q^ONEhej$J=vjl#Phc7
z^<8Nvht2QPylk{WoUGbCGr^Kos_RDyzA$}M_G#;s>!)3&e$G0!GGL{`#$!%yFE%DW
z{`h$NJVm`}KU%$idHPsPGCnkBi)4QYY4-J;O_AMsi<b~*oPTA-ZNJc|OOS>pled|C
zS7=uRwL5>TsY!KQu+{f1q|+T-dVV!eD7qG&)QsLp<*@wFurK(B#Y$TH97*Zu?QN1W
zm0bblyts8;2S@dPY+ZL;j{6%<qC`nDBJ?JyNF*AxLn7L=jYvij84W~3StTK%p{NKY
zqo`~dDHTG>h*Ak@D2?BBKkxJAchdQM&Oe9qJn!@U-uu4p>-Jb78uyVVgz>@`R*RCD
z7lvIz(0KZQreejN7btE9=!$0bin(7LZ~=yGdfsvxbN(gjJ$Q+`%n4TolR9SPC^EuG
z7Nm@lRx%rpG1$w|#LYNBcrtj6cc1FI$hh1HjXfb)>fJZ%^k_A=6Rq)iWddUTP;eZW
zv2NPRr^69Gx$>==g7Uz&s4uDj2Ga_HGJ%q)RTGQMkJ~Fo&}M?%NGnrQQ%A=Orp`h|
z%nk0=0<J*sD=3fMp8wTKi435NX^E2D%}GzR!F%=;WZ_2;QS?2@PXl>iU_r2`%9p+m
zo;a{pJAK{TCLltk)Z9ub5(3_kSWajT4GqVQ9v&m=b#k!D;rI``@wF!4L~HR&e=*7j
z=n7$t%SgeKjJtjgf{O|AnczgIlI#j6LhK`VCkkaaSX)}gYrBF)*_oPG{`P5vLliTz
zn1<hhx@y#i2?LR&JqsB!p`)WY=hYvbPRv$ZTL0k=qVM+pi^hWrvJG9zw^3jM6mhXN
zo*D_NzTc6J;n8h|5xD|q7&Q{?-_rCJ1V8ihyBwNX#bYVM`{=RlE;HNlZH<#fbUFww
zRj8#pIM}dY5L$(c&CECgg_a_GO41R=0W)7_ZvK`DLbJXdW>_~<i(C3Vt@yj|ckd>-
zTV;lKBaamN%sx1Jl*bb4Wq`ZqjN<d^`OFyVz|E-B_x*eX-C(_LzW@q9t6)YotZ*_T
zZIP%vtV9yZd36q5G-ZK8e|r6@Ua<>6>uZr@dAcLLv$?aQ96Jb)hldZYPrQk}0CfDp
zeN~haS^&vR0qTY$_!32EM+b)#)AUKKOa%D!syjYyO+%OE5}S45sh(xLc~Z=lV%Q@q
z?g6~UdO<OdRNB0qMEd1=r|-_8lR--8741s%LLg!<`ZE<1EyFwpPYW|5Ml<I74UQct
zJbz@Ojd*>+JkOB5d#!?BuCS%q>aCY9KhvBbpSrf}rcJkxH&Tf_-#p3_0A`8t>mZQ(
zt{^k}vE~-VP5!#aul9<`@I{voL0C$2>QxM}8?=OzUqr&8Q5E?bon#CUVC9(iV5i*l
z!(XPm7RHBW1v{%`AoH+_8xqN?9$}q~Ct^0(i)YRR_Q)*yBCTKl>sO8MoOcQHhTWeV
zg-~oce6do6@e6=?5ro~u>xo~ZQM-uhkFr}b0N;^nD4`YBfJwQMjq#tVl#l`Z*D1`$
zaWpzHV_u4yoc?oL-o>NI(I;gapm^2)6@8(8ttp>cT3SNFr+W|9T?waj4b9Rqn<A3|
zTR{l{aaw)fonHtf^zzSQZ%pGJugLF2k~1Apg_s~0s31WN8L{WgpXb_(+JXgYA$6V@
zM>{8;azCABlYcDH%gH?#6Qg^#OYR4KhTI$B$aiheI{1+cPG}?}w&AOhAL>izJ?ASZ
ztsd<ps8a-2^@kQqCv}AAw0~VlL8RAqs_vsumh+8dPWntj`aV(tj;ju2;n&lt$%f}9
zGkA%ax{49W{blva_a`jvvaow#VP)cJEN5(^rDB5iS8${{$G{5p*)LZ^yK>Xg%!T4L
zM(`bf3pnM#?dL^CX{Nh&9>JC!kc`SzDG;?HV{7-gDkzPOFb3RP-q|xQCMM?QP5anY
z+gNokL4ssztbt11w7=qR266ZI-&c(KT=94<h}(P%3yWW<y-ceReSyNKyg=c65x%@{
zR0jB+x)~lvt;MMUjhitjBkK($_6|nY8lh<S4@X1*!>AzJbWlpeJfiExG%mYK#@Ng(
zEU?SU%RGja4+T7~PgYVHHo~>WV+n1jD{<|4c&|oIh=MuJy%^^XIrI7v{*9=nHnuqs
zC&-~k&RO||rKJ$7S8qtt1tD?c+R>YG8@#&g=C)+R{!_W<eG4XXSN$WnujK9rjsc`9
z6Hjo1a$jwxjVi??hfD#jZ-ayHVQ6BaK%ozPIg(|q-yfP#pHMAwY6j!3pqAM|G)IvR
zbKgB#DIvi3)5vn6*HO7oo=|7aD)@*qSIu_<V<smbjdf(Rx=CB^u_?=Um05ymqNX~0
zG{k*&1}(g;GjRZ6B)<F&UxA|cV1hVh=WIYEs5kIMI0{2(5TDEHcS$W(r^wtEH~Ln;
z=}Zmfe2Bt?XYHNBel}vNU3VTKXF1>Pvz@URPtQHs39Q?Va}PqjH{#N8jFYO&$`O^J
zXlW-DZ!;3IM$?nqvk9%eHaq7h)q!nXP7Y~UrKG5s>|F=9;63;c%v(Lo+aV)aee3az
zeH*%*{qQlE$kcmO0jAKGle=PILwMUk9*81v>)S!d^MVn%r&ByCazks(P5wM7_1BH&
zkFnY}e9Fi7m2U|%x3Ya+O+C#TjcnL-5^M_0#~*Z6NLdDi1!s)Fp+^-ctyu;iS<1c`
zc~#$W;ph}D?tbDZ@QrCW;{}X1_rwFY!h-168G}+ci_x(Re79Xzvp&jYFH3jOU>cYw
z9>%!T9^BqAmNaacGERM#Louyl%YEnv+~o#@i^8+|v+8}T;!*hSWk~<#sdiH_Vt}kN
zX#^~Y`_1!dM3BPUz2+#0CW;yLSLIW^W(O9>_jBw~hwGpvjLlB%sJ92sTsdY(+Xj07
z#~mi~T(!umsYDLYNTww^R&jmUXzD$JYthHD*?s%-639K1$e<J5XN=klRMah^8=$Nz
zdUNN_%^i}2)@jPstJXsCePCMbFlduCFcb6I=`UYCaMaMyKuzvE9rjXN;|a$Bk<Ja;
zWhi5|WnZ#*ULMzu^3gQ)V_Bdwio0JiW<_wtr>43bV6>~C8sJc{X&BVanfCDFhQa*4
z!aJ>h-Y9RsQ0C%*x?cabZ$xj3QsAyr*B&{00I*K}(;wG7?jL(*&Dt}eN3y6bz3TpI
z^hyNr>S=3^u&)N*a)gIQHF)?3^Sh?PahKT3%gZM=Z@})eGedv(qN5hY>Ii2~nX0a+
zZK$cW^-HwrLKg~-opZo)-0=u*jNA*gjv);W`=M?l{yI*1I%nE45tGqEO2^pnbdcSo
zp{3eKtL#^OJyqI-{n8ir*6mNk_gM>Xk~$O8!uGx3ZxF{I;TYdnYb#>N(Y_64blZ2*
z55KI(pIkRdM1q_op7)SFcl11*votDTi(V;p>DDIFtu)N$#Kg7`*n&~-7*BZkfYM;a
zNt56_AK~MF>~(6BsftxYSP-Kzs|a~~E#kf0Pg~&V1lwepV{g;ltI^r^%Rn><)bL2F
zapm`zSFhgzv;4?8txT%$^qhh3_`CQ)k0LAKfXLFmpuh*CR9yX<-~|;om}EE#bH@;A
zarOt02}6yv>bV*>c2DoKeIH<U=oAFb^aoi5b><lt4nKMmVUFk5!f_R$BW7-0@9B9(
z|G5-5ifC>I&9gv2C<qp+hIR$enFyeTs4QRbq#A7Gi3`1u9i=3Ing;W4oNKY*gP+0)
z9qPunN)m%&F1bbDF$2_zHOz1Mbshjkx&<}Sz&2TAhsrm9>|(|Li5;rXGGRg*@kFv}
z6{o>fxVnqz?#la}fbY-j^Iw%if;qS$%?rg{?n$gcxyJj>MUhu8UqS}f_tt&#8!(nu
ztP9YnkUSO@<qgh>^LE$j2v7Vnx&nJ@HQ<fj`fK7AuC0B&jH0UCf$5jPwGkk-5DM}l
z*(%vEoa^LD%TXWOAEzs1IhY$$q#dH$pw%5X1#nU6`WAu@c-0UpcIGN;<Ue}kS~t&B
zKW1UCg3Rsc8(($ZiP{P_Re!+>zg5}M&M0Mp9=7F9v}{%91D$bAa%uZq6(UbVoEKem
z<hT@JKPr*31;n{>qshAOs8WlBj3Jssw=NeS_k>(LGmp2T5q}|ts}&(Yy-cd;TZ~vH
zK-V?_X`>J74X6#SDw>QP4ThIHCPP>>(I|Txf1meq$<Nf4lnduz&(<sZ)5fYw+yK5^
z3z4th>rWWoDD18ZNbX~fz5u-3nlXE}Wx(_4UjWvu*cCyQzH~|<U%{bjk3}dCj7*CH
zD^bpuj-?&&cLO9<uuQciYO>O}B6KXKRtz_m6-Ti@*>JE5<6le5(Z|@8_7H_g&;Goc
zhnNF}Qnh7SGpbg2A}`_E`6FteHVM745GAp5Vn%C_z+FOX<~x{>3WwFnk?bh_xQj(~
zox`WD+{en>?70c%)OgSc2AvzaRzpJ0EMuJ8^&DWX{^;Fa-qF_5P2_AnHg2r(weh{v
zZ@l~h5_+XtB!B}m8aE->#(GJ3^(!OzY|Ajm<ViCAU2fkPo$M9ckHO5>;@Wc`(N5+j
zx%u-RNukZeB&_?GgjID2d(<Xa{UPs3o!;*6#-`tDUmI+I*>ay^ikNW;=1Isy0`0dE
zKc2s?ZECL=E1w^?1qs`*ZF(63LZy~pZS_ZU&d+<r8v4l7DEGF_Byu>#=d!GhBg?)X
zlbgEswMsWzqlA`BWA=FfmM(v!$g$uM(Yk~_SwNV6F{7I4eCbCPYPUF}gRfVfvKdwN
zt3UF8y~^qEgmvP*k#>I!)HsM=Ki#b~(O628{ZJ;D3z4dMb+MayeeMC?gqMHPXf!>9
z3EK8<gS(+<FeY{Rf(6e&Q==$Qm?^@SvIZhX4MSOR!I_+Buj(Z0mTlOps)PO;Yue{u
zf^9><=`S@GGK#*DwY+E;*U5As<Q;KXty=qP&3{eXs;@|}3;wLW`;#2G?NjK&V2?ak
z6i6Y?WqhIiK$@a~4kdEIZRLYKKpc?Lc$d_Rc&P~C%^H$SP>4(2bhw^_fMZsKx*^x2
z{+R);j$L;ipdq>Gzha$!0$-r_{Twp9ByCr0f=u5>9nnY5DpitY@RCD2JlByu32o5^
zssO3WJJ`}B%Vf_ThcaE3ySjc|gs{Cy@Mpe1z{AUJjRP5O-!j8@;}3xED@i=)4(nd@
zoHn3@eAcQfIWpRa^3!M9g)-s*n1|0T96_@E376eJ<2}(8azf{?;^Sk&6uEP7W)zu?
zr)7`WXQ?3J%9=<gn3U@Ml0d<wTXz{KnLHS%BcW@qVVvJQT9r=p7G?wtE=<SHf=_5x
ziSs}5?@QYh&LlWr$Ag1>!W~Y6iR+v=@408eXD7cL<4;b{6}vofJUsDp+MOA*=p}?`
z5%cKdc8U?qhL*Z%VB<0pk~J8?$5(4|?8u=O+PqNTA`@nWE@86xS8fbn>H;gQC%GJK
zS+;+@iz)&QTNEJ+@o12TSaTo!7KX@R(zGitZ<z@-D9~f2o~9o*GUA$%mc!368H-}p
z1PNDeoLcIF)|Xzd<qcKOyFVM7cO@Dgdm#X;4)2Ss<TCg_=3t~{_vm*S85vY86*Kyx
z%E!l8M{#rtk$g&|X(EC*vXT*D<%o7^>V2gEz5C;dsuhn%gv>s^75(8+9!I}Eo^Vlu
z|50M!odX)mJKM`K7*DI;BH;JZ`HIYC;_;73BD!AZ#P@S6DA)5r8$Mb>2_2a;-Na=D
zl^-vjr=YoU?rzwl7Ml5qoZnaL+OJ=~%Dum9aYFa4=F8P3GV1D6k-%Gk2)YB|)DVGM
z6V(Yt=5e3NG%)!b`qQTspUbmf_VnZoHF;Y?n{@v&WIwwdZIl;Zg15zf(^Z6`Sdgo9
z2GGj>e$KYNbsyB3)c!(0z8#|vt!N8c_Qksd@vapB{~+MtjL`%?v^|xlR8cggf~+<W
zoOY%cuA`cjF5Km3T;XTZpPtg1Q5E4icW!WNJXR|f%pa;`E8(#t3A0(W%1F%U%rrjd
zj<S2%7i)=q9hmuILA(!cQ|2rdG=v#}^6WZlp_ajctcp}39CtHKq%gGm%ZRx3%Xg)N
zlOPZqOHd=n2$udNWhKX(i-lx+kK~I>Q%hY4ssbc?*`QWf9E;LlIJ{>F?Qrw3MyZ+6
zauIG)ToP*kvv8iW>yWOftq@V$Dq4nDFrnt-$7S40(;&20#g{@d+MLj*Yx(w+;&VpE
z!m2u?F1ru1o#bW6h)j<i5htwR2fMfEpko`ox%>#$nna_^A~-Gt+3o6X?;^nRE=L>8
z@H=pC%LC9!6rS`B)}X2YTwXsv=as#64V&efzQUBC1&Mqz(GBHntuqOR#fZ1UA35g^
z(g6J0Dh`z<qSb4b#DIh83Bqc2H!-lb0$FLNkiBv*t3p5fS@ts+@38`hQbqX8%A-m;
zqkM6L1TU7CMkQMUa<K*xIcNNGnJ{@JC984Sp{z`BqHlw_{z^A_nCOtHS{2b4KR3L1
zya>uTXRaXgz_-M>G=egz19!VXNOt*1zO<F@%PzLHwPB&{)$`})OWrOs>>YVT+#$)5
ziY}ao%aG4rb1%POiXt9O_>@+n*mgn92$=B&(Ova2vr5#<p5PdEKPV??AosJ<ncm|$
zV}yKX#f;?C_?%<LAf7@f>AG<hwku$uu$06@3oliNAf{l?7Z6??!H;PFCK~hJI9^%#
zVwr?l^KJX?F9g(d_sVh%uW_yQ@d6~^gFNPbMTK#Gg1NqQWGP>0q&5|}*1V+YN|b1{
z4}U}La~oE0VeU1q&gN|L68AVqT%rS#;NPz57<sUgL7qwK<_LIdzYsBS?HtnXauRXR
z$Vs3E+9&IZHo9)?@up!u0)8^E)D-jrL--&4faPi?^geKrph*<!T=B*!5j`=`bP(A2
z&@Sv3fvxXD14p3-la-a-(d$k&6yt^*S;QYE!GYUPUfAXO63O)?5SLO)qom7;SjO?H
zM$jaBg;ybpG2;XdwLp46_0#qVC;0nd=kJ3+rbs>JzYm-_Pn;A=qDr2mAD&3z1{i5W
zKe+<~$W)86Siu!;dGL$Ugzm?U?)vUBTMAyd#yVUct7h`^2YMNf4}V0&;W+VfJ~0gp
zPL>m!6h=B;YPlAbovH#GB$l1t*zjp~BQ`3-dj?_LOR=r?1{!x(!Cq|E&<HG^U_cOR
zZQXBro{5QxOju%biYLaDxhF0+1(T?uc5_i#I4Nc%JAyB+hR`J_s37euJ#peh`Y`$b
z${23dYzq=Af2nP3#&qvC?~84++MaF6KK<XrmN|>Kt}*+HHqcc=7%eiuChOy&Yciqz
zzX4SW(3N@47uu)#V=4AX<bw_?{n~icpFl#Djv158IdNV{86nuoY^GG{fXdRj<Yb8u
z+&!_4)J+4@Bc{UC-u`^IS6A*Cdw{-IPtWfo8T9p4cibz>094bLHgo06m4i`#j@60Y
zAeB3lrTZKp5=sS-+=pNh?)-VdaR^UvE1}_aKu24R6J5eh7+!&qa*)KxYHg>Q!CmII
z8s;`v16&`~zH3e*+ES;8PAU$qpowiqub(kj;Djcs@;N7f?FocBr;$0kDx0DaPwl!k
zz)>$8gM*7(LFP%&o;6QRF6;}59A9E2rIpp)--!Km^v3XT3p=}W+WTp5mIR<+ble7R
zGQ_w!m_OLGx9(=FVOay4i-0w{tjQ2ufaqu4gfK<i!!*{IFvwG+rQ)>p?{9?`rJxUq
zD$pYNCN^xUp^1^l1E_k{O^Hv?L}J!;*z;?xsz<O(%TxkvA*FgKO^NddTldzTe-q+K
zyq3rai|k@Rk3i^WY>it^FUAYT{Xwx5o+9AH8g#8<6;n}%WDEy<l8EpC1Y|GAcSV84
zhyuf=Vhkr{*elEwnnpjs(!2O?xWp%<Lvpw+yvsmH)>ecswI2B~6e<UkWYr*S{H%I)
zeD$ec!yGm}2cWVZZ7W4NWZ{9X3O0m_X*0D0N4sL`QTFwO%Q3Z;pD>}Q+v_(X^rugT
zPhvh<dL8Xu$bE4dF}tvz;e}q(r>aN$qH%X(^3prls&V82|NSXIFJL5!8jFkO)|xo?
zDl$()!{02|=E%{b7|@t)MO8?@>+pEABhb_q<30l$`gMgTo(NAW?APD!qo;gfYRj-F
z7p$<-^>PGv5R^1V&;!Bb)V3KrsnEM~cRRo>%1TP3z`mvuxwf{6fY5-f{zfFBIJQYn
z2HonlAO>d)*KVS&BEuEHem3SK64l06eE9;loT_fZ20F{N1^c|wWUVe*41d@_69^TB
zW(b(Xj3h?#m7JR-GYLyTu!zIer~TfZL1P_vw0lm&JsU<IIduONjNXs2wcstb8pqlU
zO;luRMEarj7BAQ39nyB9)S^KKf5duayQ}kkM_?CWXdv<o)rw3r7%%gr!`%0kBVcI-
zUT+5mmyDNte+1(@Vc}kgVJ#9eqRltZHP#os$oA<%G`NmI6QFdx@(XW46Kc{`ZFBC0
zEWiUgA*K3$J0QfxXH$_B?FR#o=hG<<{q5BVzyVwJ1`E(4bKx{;IUO?1bPZL2AAiDa
z9hUGOJZKr-%FCD@y?`QS=Fbow;=+w|JsJL9vA!ZB;^%gGq>1n0_Mc&Lc2}8$GLjhP
zYYP_BI=0+dr(ymfQ`dT70t{s<8i}LvSKx47p7~nw`7@|7bho9d4H^RV&Ei?B8nRXi
z>fBi4yhU@cy8P$?W3=tVyFv|{TkyXpLjPpaYCcAYW6a8cO(#OlDveTfK+4;`sOd{T
zUx%?Ov@g5?Q~ppsZquGgnN@MnBxU3MS;0@y1D7h@RAj<&vE~}Agc5Ek=lo?XYlzuO
zTSHcfcUj2(*GYGgNXKk>1rTr!`(BWW915K<oA3%^M)4#0mVL%9;~dbMq9`2EG~Ezj
zx&;ds^zJ+e=xI9ufC@93<WvyLZwNI`z*ypNkCh-W@&wLQc*sKbm_gy=J8B5Hf~2Jd
zyao$Fi(XS!iu%$Qfvg+ZsFiVVSL7Ogr=iRs5AUmSar#2Cg(7@YGV0rrvqWPLiR;CA
z6%`f4Jc&iq9`B}3jWB+<;o&GGHtxBdKN3$Lpl#P}1jh(Rus0Y59*_~?EZ)sMYy6WN
zl`pQ04H_*h+!sLCV%aVuHgr9DlP+kf>3zR|`Fra|SJLL*FJ8VB9X(UMV&!uf$N=3q
zlpgM*j&vE^1mED);W2#Qj2sLtXu!;bc~Wt&9lvR&4`e0)$iMI3=}#-4@NZMTUsmwG
z6}HAcNHP4QZ7>aCnlPNbnT^y$NZ17|r$u*h03m+gtQ)7MEDz+~jNfQfmqKUE!je7@
z6p6)*iV^7+Vt$o|t*J@+%`+ImF!BLqr)=6dF4D4>gcOGJ0s^@YJKiA6XpGq~0HL?%
znS^`jRhs4cp%TH^yO5ErGSUF~I%8s2&=qQs6N9*<MUnn9aGWh%IL>l(#zw?}RswOe
zXof`t5=rELdcAe0)4E~`lL%KGW)z<_Wdllcd8f2)fK}?lj_-6p6fPsp=Ip42V0|AW
z<UKg@QTpM<v}l=D1o=&aJ6V+yw6Th7lWNdc949YME#~6_yb5$KKN(<x#N*B`VuFc#
zX9m=Ps+j&9^VtEqX{UQh&sbY}{U_8fFF#-Bu=26rV4uGN8B5O@9#`k0UU&V|u8y|M
z=x@XVvqJ%XK)T5M<4Q1peQ@~*c-0x>RBlajaV2s&k;lD}u@`icn@S>7_S&N5R;teY
zhFinD;8=)~HO9pqh8`BegcDl3if>PPrZg7adVp{DiiQBp`gL@hqbfS10D2`V3B4W9
zAxCZDOWkECSur4ct05WA?ffZdvxgxuWJsUSe&kcTiwjIppR60BQDJ4o<$eGCAo-{m
z{T!CeUpMVcFX@mD@<T{eP!YlTL!D&9A7!3xM}||>u`Ae%8G=Mbx~6^PL4-s=BbMrn
zfo|`2RhWYHlCR<ULoL#P280b39)f_EU}~Ps`jW|ryQg=z-Kd+ugbo`rBxxbZ%N^q6
z2{RJV+S<Q=e{|xxKgt%ijac}wV@~Svu=x$OuYbWUUg+<a*KE+&*Y!DIi{Hi!f`~jw
zKfKizLs8TdoJ6}7tU?fl5^7CT&#p7rP$t6IwVQ3bj3bFrXhNF6H7^_te1Hl-%)__J
zPjIvt|D!W~cP5s;-i}i7f*Ie^4Q!6CdVBR777@2WVy@ges@(>@-Q;E$+H~8#W|_N8
z5z=rwwRf8=ax)^0WMS2;6hzs35okBx)z;=S>aYJ2B>#o5YuA|))Va->$jF~)N^;-D
zB1n5xTrA}0Y|&d+4kjfZy%b{WWfGeQ_W(+qj9FxukL&lh*VE)>W0_71V);+eSj<XC
zQDU564)F2-4PDgyP^sM<@YsvJkX_k6)!6n@#Obnco5{Y7k1+1D{}RkHHz*ZK<6+xA
z7BJYhwDY~Br{ZQ2Ow@p-N9(X>Fp_IxBgJ1_=cjy3y#$nzTYenegnrkbhxY-|M6CD>
zmyFV~zGxAUyQUFor<kZ^Nv?8jD+0Y_gD$UHOJ3Ei;9~pN=cj`{Qe-Nf3r0qR3j_8`
zPMs|z`)357N&&_&!;EA_S1gZ%lo2aOCmI#$A9vixFC5r$8=-pJANCqHB294ros_xc
zHow(lH}0p?78%WU1jTE+T|)>1_wi-mK8$V8y1MptA$7esEX9dnAX`@8kgo`zWYq+x
z7UMTglWEuHML^iCvX^}1LxOun4JO<O#hxz^p_&6{LqAQ^(b1_M6ClLR2!2Cg`kd^b
z^=O;Oy)iAr&{0T1<Y||7!4KOF6f>D#WPol%xj;AT>3q%)%kIrZDvmj{c4W9uPjrT4
zhHm?jui@3Osx!MJDp&PP0xzZAxS5R%BT~3BN6qAz(}WDVD*888+YMVii-N}HM(B#=
z2@1%i)*@ktimb8*w2i^G9f}a)3oBztEI}cQntili=MaQpf}OXpJUC#76Kvs#BJ;${
zXtn;B+Kinf771(Fnj{c-6Fd5-bOkTo$N|iXQBR3cD@n36_8oy0v+H+awg+8H|I$IU
zX-UlJ>U2KCP400301+=$Gzg&fwO=g3KA77&O&I>+-Rm7k#_Vn5nY{_zESj*xC;#ZQ
zeHiH3?RZDfT;<5xiNA+bDLtk460cjut^$R#C(JD@P<l!c;d}lj3Si*7_kssw{4lho
za|ouE-K!n${rbK)D|kMz9NxfjVfJcjMYtK8%PIo(LAh)hI41OhI)IFe=_bq1VODmP
zEpZ2*3G>C5a!`obj4p1KWR#IC=9rwI-Wz)BS~kg5cZ{jbkM2auGfuH{2(aG6H-sXH
zdz+)owXU_7L5||ABAKSS0dV{zb2iJn38>7j1#NKm=0RPD{rWZnCqVb(aM*}&s*B_i
z_qQ|XHOB%5y_R;Pm-O9cIdJv=M$7jPI`x)AFR4ABKY!lR9gOX&JPnA0t!=Wg%>!;j
z;*Nhmz40B8(|-;Z8>Df#Fr!HePos0tw2cL;HSGL`0|m)2yewX~yjhNjjcBfk`rf;M
z4_gUg`7@7eq-;)By}43@(!vnMDDJV+Hli~Nh)!vzSd|H)LVjm0n9C5NP;Kbz4JK{w
zhxs!2w6UuY?&?e1l=Fr95$zY+5uWH*DBVq$5H;N*T!~!O$21f45>^WA^Jy)*o=ZDE
z->$nUz_@cV&W^TFx2ghYv*aQ=Ipf5GW6rJtD?j#j;ghm^(N<+w1*!MwVy+v!8f5Fq
zho^D14)$W?rUHnGUH3~N((NjdCNx)GoHawP;)y{8;u{aN7oa_J{<<Fv`1lC$fM0EA
zXJ=_CQNKY!M`s7vO@$*yai4wzt69G*l%pN-trrh>;4DH%gGqrws>zOEcxQfg%rgX~
z25!AXq|~vN-C9UiRD`y@&!6{U3FYooX+Mtj&*oHVx4FgK^CjR58%O^{y!>i(mv49p
z-w;=Y6j0>2MShgpLPnN<pqk2va>j8ZS(=LHgdSttApu@!>?E2b!Y?~Sr0pbgKV_gF
z@+wiy@;{9=XSKR4{`n8E_|HM7G^9b?8*}{J8y)F(ol{Yt4+4J~0r6<I?<D375#)?H
z$-<lz81QV!HX$F3v+jP@3_?r`?=B6nxl0;A1FBWynh-G+(_!mAMRlA(c*VZqA-rU@
z*(2c?K~pe2ZD2j#tGUj>$|}J$Y$Yl=7EcY{BXip&+H7`uw(G#I#Apo?BeMq3RtjX@
z1M5T-bDg?R9gyyGUTshN9jhqNDnX|$il=-4W<D$`YGl1+HAK@Q!HCj4g)f{Rl{%f<
z3Lb4oZEjV!N2e2h<!A}+-6DE68qyn@5M2h?|E4PZiWLcBM(5eCnLc+L>2gapovWgx
zfu2pM>|G0*aF}}C4+hcxBpeJXlW_FFR4fDMUnb2e4cktT8&2r;Y23zEz<RbE9vx5c
zW{M#~<SC+BHkyA7M)hKrw-buIo2ab9*qfemDmE^~u#ZQnrXo2XEL_%hb`u1|ud#Rf
zBLMy8V@Hn;tKSCVZ?j|xGjh`RX9_$+$D?Y#vVy`tk4ET$M}u32`^!Hj!OUU7P?Zg&
z&^pC-oPhXU_TdF$V10AtDyGCN_Q4CHwY92rcc6QnF4*ORr;!i#B5bDPGACRhBZbnZ
z7`W8kN(L?!?9!VSdl#9n+Vkek8@w}gp%vOB)GtlUrGgusI9hiDGz*-tU%k6xB9sts
z5T_u{nz_v>mH<^DD~l0yb#yEv6T}JKI-Nc9zKY7q16zjXXkzFGOcK_KtJoK;N9Lk;
zwgoWl?mwPcIWn&nL-ps#2=5#&+<7Gt6?c8<)OFmRu&~HANYwOL<bj*5LWZ(;n6FuL
zKIXI!5OHsM`?ITlb4MU;k^oX-M%B2@1m9qEr<AC9IM=)mXjUg;7N?)zy{oJNRDBEp
zi-Gpiap8|p%s6N6Pv%?Vm)KS?w+f8<Ohx}D?h;{269W-?m?5)h%)Y0MGN=Ue@bjSM
zY5w+Bt-?B=SsYRwuCB%&#_S}tve7!s4LXk@LG0yKu(ag80Jqrpn-0*^tN*J!gOINd
zYltcw!hLutipQMhZIp!k-iMSmnje;%D&QOuY!n~Z+CJLb1hCYIIO>+2Jr;BeT_LZ9
zKftY=_1C@*hJB3#3Z>fjpzpmwbL1D$G5_95pa&qnW1?mPD2U`|D1X4MW4zz4S_*sz
z#~T9!SOsqxA#<CauO4*EQqgLQ`77^TsIJ<%ciF7h_h7RUW4@cz6~ylTNN#7(K?Wft
zm(L)CIHs%!8W~`+10hYW;01^fII10DRieqrZV>e&;**EXnCxuq+a=y65br(JgqLvt
zfjw2$(Sew!n34*}G2#hTfm-w^0DKG3{rdvibp@mzpsb)lPB4zlXMZ^G^a3Z}5u<3?
z-y0348%0-Rb-}q>gg40x@4k-Oc?@KPTXXxI0z0h<$fSAZ#9#_img%u5Qs`MdUeD;H
zhQK7J?3q>Xv(YN={-ZTN?VGyGaRmc+pr@*D#R^J;3w2Uhivf3n!`9__G*e{A2Zm3P
zskpmn`}Ph*9YYjR;ejAliv7lziomE+OmjXT+2zxwZ$Q3NP3Q@)h%68nHoB@~Red8$
zG>Ek&JUlA_?V|by^Zbri@W(c&6fx_GFJc-G0z1L-0b^>Pan{54>z!>PyrWu|Yg_vf
zt_?=*G(G@`Do5WD&&mXxvrxpDT2c56L$3;mY#19}i!dn24F!dD`tQ(BPd=PVem-QQ
zP}}o^+a)$R69I+SU-guOdKwZoTJIvE&{DKTT-^E7-on}KfM5z}O$Kn%(YMhGn*oX`
zmLRC^D)CPb52B4vxhru*Sv{W59zLAz)T&%CG6r=&e9Q%#<4S>;;9hC%X3P!**}qEi
zJ|gU=AH!H#M8NvtU}t1B)FqXtp%^xSMa01XW29Hv2o!KbH*q|U7Vx6iCU<ZosXPFU
z>uVI_(a-tM(=BYWeYUx|x#d>h12L+l0?HoDrEW!rDLCroIqQ_dZkI}w%s858Z2aI}
z>V0Mk+D`NG$~<x2jZ&9J=~sR70@iC%gn!Qh%#4`D-Yr@KsAsiEm#>e4uODOucF6|C
zKfqdAG79v1HQ>{nMVo+b`H2c=zpf*hsuUo17-dcr5$ZF5WptySN8&yz64482L&t*E
z4^_L=a|NmAaZDiNun!#)=l<ph2qH3kF%KOV2nP7c6Uo}%>otE~tn0ec>r<4Kl`Bra
zhICmfoMy~nr5NA{`Cj=|=q*%NukLQSFqyea?iv0DSNm<g0_i4{5jk~j$>ee3T=i5x
z=Bl8YfpTklzt4LurZS^&uqcE?%o5y$X2rC>cgXx^ByD;H*E6nPU(A*5Xkb@20d`fz
ziHwT+gPFT?huEJDTO0cbrBYH*XLlLIv+y{=5gbO&*9o0ZN_v!+m(skGFjL*%;zcx$
z5AtBX7IDwLlH6=Y_%9?bcESF-_N{#YvOotnR^u?tKCrVs^GB*U8xcfy1wZ5GM)DKY
zRZP(~OiqpXt4J!ONLX#sH!F`pB&ePMb(9KlN(<$}^%mwTgINu=Ig4=|v7_62T~ae6
zJ(2^V*y_R67{TyU?0PydCA$uML8Wft@-Ay`^MI^5l!u<K>UyY^`*a2iV|4-!Z9<q>
zZYSm@@Z*8&S*1-=e7QjcR!DKUVI5(6k}P_vh#w?eSo1gqJX5-C%fwG)TL^Vfw$jP#
zYeIF=`X)K)DF063BC00?uw_1u6WV!2trxN^zB360KFC?n)0eCdO99k_YI83#*~2GL
z-0o+e0ob$c{_C+AK`<+i7ISld{RlkbD|~$JxDzN~<93a<@}eSbc+&j5Jdu{-=_?*f
zSgm$o%*%8EP@J6B3#vC!dl%pR=L4^_t8C>!^f<WUvwuw6dqK`CcUr7Qhck>953QVr
zT}`0_-B>qj%U}a>G6?VbX&y$^I;GpAEn%q(-qvCrg8}!^c@|SpimAE{fbk8tcbZn>
z{4IBQQQ~=M+Rk3R;p-6QUtRuCG+?z7`{N0JJy0_YdHKa|ZzyIE0V^3q0CV;;=ATfX
zQt=UtPxhw`^`%8z*{CaT9k{lk7Xw-Z)K(YETXj7fWRlaW)KLUK+g;$3lal7jiZ^EG
z)fJsR$US&|l-Kfby2pPKcP^22Sk4uS=divnwhu)(>-1^YZE~?9d#@PGv@fu^3B>6B
zzfp`NqL@sg-TeNx7Lk_q2F5A*(Fk5eb{^2yN%Z5_vOiua=IYhW=m>J6KcP6L=3aK*
zREY@%N&;`W`+ORy=kk+)9^Id>iA@7|{;pm8_sOY>c`ls!(=c~M6fi1NB8-zDHb6=-
zh(G-&08%JVoH(FGmDWkaZUwAf${xO`q>5~n^mYJNRsU;us%RBzmiuJ{C=zEsAgzlh
z>K2FrM{OOITX8ZUcqeloiyM%Ll4K<p-(gRNW-|*EZo1@;?*7XcsDJSHDqmSyQGsY8
zd0}sa97crHFsLh`UL35=h!<>nZd>^IsMF~BNJL<paXVHC%%GFST7xB^bgI_cHN&$y
zenOYR){??5LNLAi!D9q2*3o@YaeHp%=y)nRdQ5}te7SI%=AuhA<Nfq;s5b@f2wH;H
zDY(oelJ7HS{;gOE-sqKIj$Y2}MP1hB^+Q;j3&g3FcX&%}NvyUAW}M^JVC<Z8Z4HQm
zG-2dj*YcEgDNk9~t1-$tg*o!60ss+Dt@`l|#4;EDhIa=M-Z_9+X>-AhZ@@Fn@g0$i
z%|^vYLOeV?c8%VRkgp7T#1;Gbsq!*{^qWBE2Gs)c8w|i#qWKd7TPmF5{J<IrQB5BV
z6`kpJG_U#$vlnYv@CRIh`&(5*5&9<5GL;I)JV07Vp<H>Tg(F8<*E$HbSGR2YD=V0D
z$rrm0%NY~C1$~Q5{;pI;8zz1e^*E|dHsKc+pY?;hNMJ|Y<cirQsLxm=9mWrjf6lwq
z`#{Fi0?T?mVp2#h=4V94MaYNtogR3Hse`Y=U17l=gayy1ExqF(v*isSsus+phFwO>
z6RI`6LNgEB3#Y=?%)Q8FO6f;{azAU5h!w*|5yzv*!2cZJ2ZOz}AacgKH>wVQ6?F_L
zDmAuoC>WVwpKXB{#3SqrI%$h!IHBo0En0&m+@q!}Gt7OMh13Ph+9z0EW+Jdt9NLAa
z=Te7*ztH{2X{HRn91hK#D~|OjWPPrz6x3=*jRn-pvBo%eUM^#8>FDT)otr?o@-$VO
z6`L~bNYxXpLzH^9`#TU4(-&aBwHfyv9<Ns%4X}@m1W&MdKR+pBo=7?nMm@?DA)0x@
z7bX^)X|N>s`xk=EZ|c((EBPW;s=`HDLl_b8I&p<D7Z|=H7<X={swvI*g^_hAQiUn6
zK636x8}^?z5xdp3&EIUTR?D%!djU*#RXTOF{y7_(YAACpM>C_kNEfbE5heW~K+!$6
zor#Y(jPTbLP37HbCb}DhfAeXrfnBk(iqM}1EUWaXkCOa=ppM#9oQ=W8?jnghQie!U
z`;e=uA87@6&h&8O1p4h_yqlF<%tpU1x3Co;(R?FMlh!!>u`ST}gUGh3n?PPc=^fXK
z?zJUXAi=xBpdm^lx53rJJXRF&9v^s2+t(}C<(8pLs`DdptJNOt$uO<P%34^O?q_d%
zbL$42J<6C!Xvh%Up)2-Sy0RiS%ve^SP?j}c90N>_Q*GZz98*<qaVHYV%#4ggzS!1c
zvpFxk?_33ta^6m{L^-Y%a~?+`?+DWQnA<<C7C9s^2cuFBB&~PJ(x5(CKWX3?93jbY
zAmAZjfY$*}DUHR`1J5tmH(JX!;`k9(tMuGycrbZfw_zX$hO)SLD~3P0H~Wh+DNeI9
z?G`2@nasuop=NbhP>NI)J8P^~>oW&4S%B7<R;>ks{NfSt;hS)+WUet;nJiqh-o+~q
z_eq}S{_;}Lw-8srC-7j!0IC3~IiAc)jq_ntkwAj%j|Hujm4L$8#2#k??kNNOVAe!5
z?I3Oc{&)-+4)96=v06O$6|`CpxxRs0ZYg&DCgc3U2f$Q#IYkr<gQwePvJ_VNP%u1=
zX@xEu&JSL{F`F(qqR?dvOF69<V=bj85LSOkea%joK(0`#ye7U!54A67*KbyvEfnv{
zlbgO?n7e=L4@Eqp*!5U!@cSHNK~Tn5;?e@i{kW;N%_H(m=*q>sdc(9h$@5793dn^8
z4F3gVMv-I@FWP&micy!)gFyi1+uhilqBx2fM#0afU<=QmQQmUkpTozP4W5so8m^&~
zWDvvGmZ39{0_R~;#@crkTFni6`#W@CFzx>YMmo`eP#+cp4Bmsza@T=kpstmA>Bk&r
z+j9{xI7aj2!W*++QgSkcaUd(Zy7wO(kRf0;ClRLd?^|A`t$2bdB$Sd-LQFOe%BAwj
z@W5MWn2)WRN#MmR0z!VP`5CjOYfV${N;*0@Wv||#@osj4PAJk9BXgXiO9|AEFTmJ6
zw=1CXni`_oieD-!s34cTNuKdgONDNNieUFM;xs$k{Z_+XA`S6!T2onBIbi|CGy~E>
zmH-HiYQ_XGbtNX6B3byt&W;F2PQ;ukwpL<po|n6(GdcHNv8^<(Jn9T!rYE}JKs*LT
zPU8a}Ih$`FyWo5r0{&yoWZI`nVNE2@8VA0;wA29EjDhyl1EjEpOpzVO&nz%Klnuhh
ze}6yIH+;+V4XI*q9ODl`_B+2~B@PySP~>63`jK28=)smKQ;-H-QcR#JQ95_RnN$t;
zMPDkt`?;?}KCb6}m4^P$au6HLex0$=g@|w5asVW^B%S4XZ|7`$(ucc2K|UT&^Ydvz
zxqbpu(Zd^PL%^nfhzvLdQGmoq@cUh-`VxpkV5XzKp{S9(&;SL)eJYg7oKpsOt{FV|
z%5`j{9R`9H@~-tQ8_<O$Xy1#)tjTGjgYB&vIiKpBc*P3=CIH!ao)}8r0Is*|%w49*
zCk8*Y5v1-3z+mbfyn<Gm`<wK+^aWQ~t7CYdxcKCX(Gamr;ue{f3bhp;&9?mhwFM)X
zHafJKj$c3KBcxyVD#_(KDmkD2z;yTM1D^iObKgF-;=vQ3TN6nz-?4%zL-X{UXqmS>
zFALy4oZwJ?FoK{URjsAv-B_J4hR&uJ$8-ZMr?eg6LWio1qK&exs{Tm6#SoY-loi*@
zPX|6bBt#g&lWQGoki|Ga&Ci4mLak8Yv7inO7+s*drTXvjvAYjE0~PF7K`toCQVOwm
zt~CsEJZ){$Q@(~gd0e9_fJCAF&~u~ejK`}L)U<=u<mXvUTcWk1M$HBfFbyOZb4!#;
z0@UHrtNeoEP-AZ{L`_kY8j|Na^UJcfFS0rL=abxo(0jcH8HPy<r3^7OOF9xFV;g5J
zX!_)Na)R|5vx<rb29lblGN+cTNg02b|Ba%p((GG`qcv<MC<?8L5Hsz~Id@vQX?to;
z=k5HTHN!`Z3e^bKJ#^*NHiwIL1DdDW<h$hk$-bQ{C2<8)S3#4mVO@Ymqia9!!KOkq
zPd)pLJ)u+1oROm=+YOGjCb>=XVynQ$u2sKZ<>(Vopv{l4HM%BYWU-p_1M~j_v(z5$
zA6Yr`|8Lw{GIDp^bfnb<pPOht|Lizu#0^aT&mrQd=#^M;0dNypi>qKzT;i^OO(JB$
z^5x6hoF>;kKm8CW9f{;|o96kcCJg0A7Bs@k)e(%#&;D~_5n&qgs|lT0<?#m*=*iY}
zT>Ppn=yUCTdW~E-lB_qipyJQ!_s(2?HmWh;<lEzOMM}dXUh#^*n?mTBV*U3t@#M9<
z%;#5L4^Ib6jEI99m3*p;#GB+|e0@>fL}5JkU>f`o%>QC2#ou?BH<xZ)7?cS5Q2mIe
zdE674TwizsEO5VQGa$eOA@~G7n-$KYA#cbfW|n57*uy=ZKO+!$JD<N_?k*=#_yAi3
zYE7z+=!ZS7`~@uzQ$B|D=<2Z?r_2Q1<L3SSXjws(XA7`18UwkV?L>8iKcQB$zr-Zc
z%Ct=xWTuiA&V#OPwzRVw8R-fz%tI81lHg+2ZR4Y+vzvm%MO^46?0~GF_6tp=vdb<9
z^=DNb;deDw5}+-+qFECt4gL3aiL%Lm(lyY^ozYfW>dyaWo`ppbP}QR)S8YR~jgsg6
z11YvDPzs1^!r<AKJ?L28uvEZZljy8RN*O!sGXm}}Jv^L9%_d#Dh<uZg%>qYa_MD;e
z*aL*{Uw&;P+5E(l%@U9-bLv-IM)4RsQmhih-kD3ABn8|8Mrd+FoDUi+C;y|%ieopL
z7K6rJ5~p9Ce6E-)oIo>hg>wmnv)m_}sta$i4dIentex<EH&iV#$*}+Yf(8guY^+}F
z@dP{Jc!eusTd{^XpzILfF&RBvG8)&TGD{hl93!5?2n+ALX9Y|g1-X-W*WCBKeMTCF
z$m&ymf$-ThVdYou6=2(RaBNQAnbg$h+h}hRAc^-8X6-z$q9Qo!8`Vq_JG}w4O=Rx*
z>NUi$7?db@xM<d_`)Iw;EjVCLFmnrQDMm-pl-1!T_5(`bQ&yHABs|G*u2C&8VK5TM
zNWlo@)VKraJRj~wG!bAo_xEE;IAFEEfkBdyBWHu<%RhkzP^Mf99JZ+d=moslOF9;>
zzeGFwJ5Kkq^25Ok)Ne0kKSK>*4584aB(A=wz{h_1j`)+S2_v!1)5dvy0zx|!6R7~%
zq{&(dmE{P{FrpFW9zz@w30aJ&Dhgo7PEgrC;K4KRvYO@w?;s|9ci&SzGpZmPG`|Ic
z%`*B3<}%i~(oQMhTAf#a#{7jPUa)y4`+~HJQ{Hu7#jX(2t92nM+aX~;8vonzkT}*m
zJ`hGVKv#M${qWDfTI?sis%>vVizkt*1%(LyWPT~SfF50K%o=hD2Y7e<;KDO-Cyp&Q
zfcv4*i-1-&lGM2Yw=l^*p~3cMZQ2v5zVvimp-l{KS#l)|#XG<w%4*v;<2MgsL^TQ=
z@;Wr&;gDfJ=^9Z`K^w|hZQv{?uo<ukXmelyA`7*tfYxE`doWj~(6HbpBBN0~ya!VX
zMR*Tl7WO@n=-W7(e?}ZmMS{&byd$cFn%9XDJ};>-hrNJFsDw1=p8$OL-wQ~s0Se@D
z^IICNVeIHS(0s*$>7n@0jE{T9;s?jUBz@b?6Qx3c#ALh&m-{%N60pVU>lj=^GOhuo
z{>w&$$m(2A$D=PDDhpMv1sWfBuo=T=78c>HwMXFvMq<pv%z5?-03D%b=0dzHki*KC
z%YagcT}*%+ty#6oRGvK=5Q7Jl1yZ!v!|$Jg?xFsoQsgBG+~@Zu0>-gL3R>PxZIDrl
zOT~C<cHTC_V+E*RDxQ25Q7erxW@QM=sYJ7cqe0de1O?_miryhWlGY7`t6ZiAvaNz4
z&0qxD()e20dvL)U0*HZ%I(OA$3?+Hvh8042H+&bN{LmJU<4-<QjIlMC{N%VH{1?HQ
zT5xVXuy>-YpUmtBbFp*4R29MdZohwgh@DUeUK}IeCSU{8uVR{9P336)?lEoB(M;at
zC&d~&$(JQ1>R_SGT~v&UJ(%8<K8rKvNKwEMErQtQ_b7n$BtQruaY~rdw}@^>=8=L0
zWJL0Lpl!t|FlPiJyq0(dPFU~3nF*omSbvOcTS`Oab1N%GvpWX;f<4k4z@oJnM7YXR
zge#bQdk}Epy2;pU<mRm)d=Ac1pV7gwglz8WvJc;R>buK%?`g8oNn+|a$AXcUVq<%-
zJtZb6-E6wwj^`xiZ%0P+@)v4Xp0y=Y8YGF-p>^a_e_i2kY!s1u(?{FM!C}<vs4A3)
zU|)+S9^l%z_ZSU$Ra^5N8;drIOWV-s0dx8sF(ZGN(apq8KW!VF#5V5zdVSn2kJfV#
zN>>tZcxSSU=Zg0?!WFx%UTu6yg_Rxz9eLS*P`*7OV04562{LbR2jWY~LIBdJKdOv6
zDa<!m*8|}*YqlwZ+fUKH!}t|$@kD13wkv!Uqs^lNPvYowPKuQj6o`5KJPAnn0)E+f
z<Y34$BX|#<OmYJainXgFN<*&P1U`810C?g#UlvXTbVK3`@3$ZL`RU=D>}y9NB2pZ)
zG_M?I?sL3=TiH#GhL1DQtm^59J76ZU>WDX}H>xi?A-<Yc3WX2*Fjz|+SE|LDo5ECt
zVW7U4ot=f<S2|aAD4J0Op~+bf1W)X>GDuxd&wd9(tbeT(C{yOy??aJ3)Ie~nPAeRJ
zl7%Q7w%Pm`&oG!H9<Z^s9bt3wI&hGO6!*}zZqi?}Zr!@7h%LxgHTs24<S8v8IoUfT
zk~EXF>4_F76PR0FpX-1L*|5m);KAb+F)x05G^<fbk<-`NR?JvC=_q+N%@+kqeF_gR
z5V8jcPE`a|&d2vUG>+0GHsv!M4x2zIn*ya~+-<Q-POZtU*el!HTd6>$ee?i;B9sXt
zMEUy;*R^Y<z|x6q8`VM-6XwN_AmJ(O=QmKR!L-VE%b#!8X7uzG$ms*!GRB)6kJtW#
z0YcFf;u#Bb4W}s)#L&Vmtn?cdz|tbWm2p~$+Eloa0u!JQ=10S1l-KLLXr1!8JDvLx
z#*+1sgWw}e(5$7WFF29NeR?ddR#0&%Ji9O&4G)U4IJ_v!^*AUsgA&1i7jG&^cKNIS
zEvo;QZ9-|;qbIeZmgYMTtlx8Fq*F?9|Ni~7EYdl*0zCh_$>RLxN$44_?zR})+1O;j
zm^^)@$I217K4R6d3Gqpfa)f3c<cOjT#;OH9MoCv@upHoxR<JyGcwX_)6(BoN=bW8g
z0?O&vknGo!c5$WXJ-X}q7c<cFc-YekVGMN+bPPQ!t`r%u<eISuPZt&z5)ulB^nrIs
z3dEcgl!)%S^9_FXf-wT(ST*YC9i4NiO7OV`68`FaIYOUQ0*c|+L;nsio=cs<3LEyF
zcL9WFs8T9~@Jz!7;5rWaWGPLU@TAG9e+<_+sAC|sU<L)hJsJo6cl>Z9%Snvg*$qf*
zCL^JH(IwXDY}-hFi}!}xR(6<S8hdUq7<JuOt;(%-cjBJ^C16kTvfrUwaxA~{WDoYu
z0sM*Z?w5zXW9o9`TLiyjechgYhl`7g<)LY*Jm19&qqVjGFk;i5RYPI)cmZ&@WV~MV
z7L<22!*w6~^8zeH5JI3E??u+taT9^4Hjs0aiYJOemHCCJ7O^tg^#Y&eNaw>YW@gyC
z0h<`vwx4-}tibf?(*rw3TmS*tqbWNNchdst-q;jaSUam3A=YViM#&z49*WwpvjgYC
zV`<Pq#j#i6x1+5yJOL%n5TCax$;l{I+cj<fL{|-OfyGBQRaw)w{#f#m&s?sX%@~lF
zoV>az?hgi^O;0e#v>b6n*px$rD1a)5)Uc}j-U%RltSKJU@@xj6zP0TaJSm6}=LeHV
zR>8b)NWV4@kY3TvjpgnvaB4dB<a2$?&DMMHmA%mgi1q|dqj=2(B_#)}Rw0FZ#VXvb
zoV=4Wt{D`AH&La@?j@N#BO#y07mEzs)4?OA-ZxaGn?ZHH1s;zPnKf>R%c9ZmU4Lsk
z!mK%G=3Kx44hd^!*9kr0X2tH=0OF#S{j|{ChJ;Cm^8-IgB<ndiS1NkCY0v_RzVpPS
zl!@<0*i^r52r#j>(qN1>_AL|xk9Gl(=&UVlB>E%zZ!it+MhqFK{QC2P{Ss)c!Ntq8
zG`*(_CsKj}hh}6MaoQW2<lf)Hu=hk2H5GKU=}R8)z9efRwvifG!d$G}ewn@F@|&0R
zzo%ifXns#mrUVR6M$_sd^JG4BrsY+p)MtJ7y_lM6hKfYUPxhLZI5A4yf6{~rb{J-w
zzWprw2K00WWIe&M%lADU60PWY=a~bcsU8eOSau4+_Z^ip%7f?cV@sJK4P9@*sxFX(
z9Pk;0H6l9<3=H~qy#$ks>#Alo-RZB~<6UEtO)|4)I|SmzcmUdb?5LLGY#pDONtK(B
zX7&v+HkhTJsHC*tXVj^g+dTSik7B<8mCt@z_kq;?+|xHu7hHeKI|g`9vg-aBBZ`Rr
zH&#tOi?Sv)+L88s<OLZ*=p(M?<;Lerb{u41LUfE-_pvCW$Abzx^9^Y6)66y4WnkJ^
z#zKZ>4kL_&;uG28M<4wWBS#8~jF4zCw+CEi(V|7gIdjOl1^eI1wL4!@?u0y;+u=|`
zFsJe!Fz9~oJQ>y-h^X$9ENq5BqSxS81V@eUq4rNmD|<T9%h-JQMKA`8hBmsJa@F7x
zlAPEbDqvRb(Hvo^@~G+;+JJd5H*D*MneDI-q0|Vavv21;y$;iHbz)Y<JN5^YD>Kwz
zLKW>DV-8H^3m%LJl&KU5`{~NS*Yu@@FR`<~QJ<YoBqm_!B>Ul#{WZXVg(IL+L<#Rq
zKGwJRV1HRfe~%R_J|m9a+py&<d7J>TlHy`ZTiZ9aQRnal&-i2)bc;wOe}H#oQgLDy
z_HPuE+@u>q=+J;4yPQM(g0Mi&?|H(RLul#N2{$@d;QOFr_kY_Jk@o5foyT_0tg9Hs
zt{9;E@gIK>mLt%C{-8A(p#pKVl0ydg+oPX|T=LL<p5`8*1v){AFbd1se3|@+Bj8qY
ziPP{5`;e~pbU?&m9a~d}hPWAHeYX1ow1c{fsPD5)wIx3>pSlP#xt8b4(hGpr1xG4s
z^1uU6$}B%jcOaqv9^|b0-_K*XOTBlf`ad=lag6?}cdzU>?q`g2`|-(3^_?XuPy?YX
z{%P0KO+n*N=T?6754Nio6qwrQD3Dw#_}>B(1E=9mCP<D>_A)xrE!|Lg543-Ohb`2l
zW2EO}nO9AU{iI>7p<mIUMFAaGEjyKstofC$p+$r^Cb!z~dAr3W(kpWzKTIVHWu<_n
z>uXQvkIxl#*wR@sUdN2CKr9TD{g{Q$705+Uk7QeR9`EVpkBgHm#+D!7m{E%F4yn4v
z<a_~QTLE<qD6>ZN-=hoxd+C4ShH_?=Ok*%0%!9&m0%}K6MPTh(ss>FQ#Rl#^6Uu#*
zM<c4PxUmt+>)_y!+P0Yqd&T(O!2w%vt{R9<&%V+d{&AJa<?sq>zLGLXSD?{dhgi<J
z=@U3s!bwbdwdwb7ZLzfq{OlDqzfdRJ`?;A(%>u;KK`2__T}@85r#GN^9CYCZEn#Io
zf<udwovS|~_oKkHmiG-(JJI;QVNi^Yjg3UkAohykf&CrLY2$EDlB$Ym=E4m^QZSFs
zdxLXF8ESOv+ONQP(~46BERV`~p1xNu*BxgC1Hu5r>(CB&L$X7{9NKmvniY?nTFZq`
ze0;ZfWo;GJ1z){GW!35g>DqZ1vDsg=Kxr0-;$q}XHU(6ljJ)Y9lDCR;4O~Ag+_nW}
zSuz;vnBOiKSq<yn(;)?-O{?bp&KG0h3JZYBEO|wM19Nln>eHjJp%%^yY2cwpJe}@+
z6|1>L%^)^?I_#J%O<$-MUGA#Jbnj-j9=`w(GJFBTkPz~fO^5O;$L?q;QDcDVIHPLb
z8Qms|+KT>Q$9@gCW&f7SS6oXpz%vZ#oRJ_O)AOUsD{D`uqd$B0^~moCs1=!?{sds4
zp^-n;8;dQPYhRy6pkl*wU+TkoZa9T?!?!VM&Q&$=POfgcUktj79^Nqk{n_IE>mc^g
zXlVV2CWr}IBHAkN%|IxFS4e8}?!X8g8_;I$_qm*n4srjcAm#&F#I7SfFR*h(`%{z;
ze(XO<?AqU;JQ<VOfYoG+xGx}LKFM#x(St!S>->kc)bHP8Zem>hXHBH$z}N(|gi~1k
z*hdI;^5s=2o65FTYVg5N{-^Gd_5o-`5eF}3C{hN|Tn>@=gI8#4)!lfIU|~MckM!*C
zOQubqerm%5$*>o}sH9*}iNEa;#l0)E1*i=ZyQmO}niL=Ueg7$fyk7qyU|8U!ER1u;
zXb(axS58zsvja8HER4QDV#-Sa)nzgLu!y4SBN))X&fk{GK7Eqi)rk$-$JI9=#25N+
zf-F4}gAqeA8ypcmw;G~XgB7w)9p;eGR#{p}hb6OQPJi(XUeIIoINoupQMaC$9?b;3
zUku*I&Dz_xPJ;^SwnEg^Z9<1B4S2GCvGcyaa@yhGOUI*D-_k}E`pZ&GQyXZt^=zEG
zUMOb`0+Tt}YZh5s<Cmw#Y+p%sB66S~*wP1P?L)kP8d!H57f5>=@CNjhY`AYAzUEx%
zhd`!B9Y0QnI)JBH2CIkx_r~V^KA(v4OSpREHS%UpgKnWkqSYWrGmAYO{p4-+6$J^+
zK&76gqhErI^Wms1_Rb-n5I&m0R2#Bo*FK^tsDEwQ1I(+!iD+xGJ~$Tu{q--#0y?Pv
zuk<X%m?wTbv9MPUMgxmR9wW<tKU%n*Y!4Rqa9cl89$*#Q>+_wBq5awtNv~AK@Ph{;
zxnNBq_5}6f>mEUfgza!{VX$h@Dp@!5B!fn6UR>2X9{aFaOf~lOFm0JWq8Y8Nt=ObB
z3*Bh0BNoz0X3vIzJ!N_zP1(SQl%88MW~)!mcWET^Xc@z$0>P2q01~&5`EEqXLUBql
zYj||mEKXfEi|Wz+f8#iWKU5Zz;b{@u3YwchM_p`>3hqHFiTwiyPe5Kec%#<@>n5pe
zP={0!G8D#PF3-oL9|EY%uu0A%qZHI*rFA;-><3sZ&=}XikA}m=9xnF>Nap+}Wiccu
z!I#wZ(hy5M0|i*_(+4=DWV|5h?+Z|O{Qc?&S8tozIjm+Oy}kT;(L$^oI6LtFHEY!D
zFVZ848N~F7CkSilf_FK$gBWxF{52=Q?t);D3_q{*s#{}o&CP*a@RQ$QOn=ino_9!S
z_^SOW(*SqJOyB=E3|~<lef{ICPP;;}dltroi^l^`;!4b(Oow~q-ww9`FX*+k!duZE
zj`m3#wPPmf*K1KfK+A-)HUUKN{%i05*RLS#mI}nip1`^+hH!+==Qzy~N){5WF^p$C
zuZ3HX%tq4_wsE8Bw>1x$F!!6-b-*nO=D8sHJ*EBsws{i@jBB@|Aeg5!W;i6Y>h3_D
ziI^4iU4<3CK4{<4;qYp^x@+Fv+_n*cKWaC?psxYw;b^o8Y1mZ*Y6Td$pTLgX;v#Iv
zWgAzhtB9hmBC?~s#_63m^M+MGe~10t7v-L9gAC&k&vC8l4Wbuh80Z~O(m6&#vo`3~
z=O;&l9My;!podGHzV3}gW^?xYUHzFOPXc0f|A3VM>wbddQSj!>JMT9SaFsGn`53CO
zzUcU_NhgARK2>)bG9zbhOY7nJ*<g7o$fCiw7j&g_U01Kx1pVRI-ID0{Uxz;l39sj=
z*4spEb4E~CG13c&?2#izy!m~$khJTY*l*AX1(i)h?oi)=Mr4=j#^24Aa=r_jJSww!
zwEFrOviZ^qu(AjDIiJ8Bd5e!D^1J(Uc}3Tlyz?Mgq|UIGFBX=qYa2_c#^cRGH_=Sj
zS3V9F`)#fN?->h`M5b|~^D8|vR({VSz<eXg4ytP<&ylE`9z2gNj2L;x4Ga4P^QwyJ
z3CK#Oq^X5H`HTtd&sW>FQIxx4hQKpW+%&}5U=aM@vzjWZ)Aj=|!7xXXIvI(F*wrxh
z?r;<xkZPC;ZDQR^!4V*;JwdZKJ39pw!hkbed7aM$>TfweU>ZcTR-myoZsC~!4}mn1
zQm&*`Ew8=5qpYg90Bs*}!f2wuZMyq>T--8Zd!&!x8PYdcw_B6lVm;dC<Zz5EM72*Q
zOPw}~9}8>K3MPC3zO>=ryO>SHtDg!&$U+#}1Hu_5xsPTIA*GxG-bLR?OiKDH)(H6G
z)x3pFbdFz+3Zzh$6Ly>!^zh3GCV*dl=+GfRX&U>bs_6-%TQ@W^8znk2hj5M({9V><
zFI$`b;)w8Ydgpw`d_ofly56uI{odvtet1Ro9}@X!(49G#0XV!syQM<^x7~z6{0eP6
zx4StN0s@{g!|a;_QGklhzt(h+37p?@E@6HrFWZI$qp$nW82SG<pQzkRPsKpLc=b7=
z@ma)u2W@PMyrEum{_>`WU8mED<48Zv_})yEtnB`{FmdxYvBmh5+BFG){*Bi9f7=re
z+w+u$IunyxK}UFQ;hwC1Y3o3A@GjG{zpr|QM+Pz59zqL>Y3hA+f+RLg!>`OOGsGhS
z6}4>?WZv;X>^gc6_UOQ``3zUmQ1$<cCVL->_JCTFCNIz+W?uvT<gnnqI5rx=VDGo{
z48GtZ>x8p&L8lcH8ynlWosUekq^KWt?JAsPntFIV42=G@a%yh2`a_TJVkPQs!e5NP
zJ7|tn>WU{rktPM`GO=4*l$%099me-ThJiNIF_jg@HSMn$H>NI--P!RccGzPuj#fSY
zj~fVq8`y_-QlMK7vtNb;DG(Hfv%>WDqGhcR)^$ueunPh2AB=S0=W{auvj>ruO=v&M
zdp=TmJRkK0EwBIgVsodc2u4n!H1k_C7A0KN<}}IWChK&z<cBHjg-B6#?yrvhCOibE
z1mFVjort%XJ2CaFS0g%I*^|OXFp(^i?XddZ4gYVG4lP4r3-vn5*tg%QOG#I*LLiLZ
zeFyB>w-^OXBN#pQI=r7N^%1@YRR)`pOnHsnSLJ?8rtv}SH_%Q2j^hA*+B@ez-d=&k
z6||y5GTXOfl)HF5b`3;s!obh$zzMX9vw}eku@_N0>zahT6jHoVaYRXI&@qOo=zb7W
z2bXSvH{m~i{(IH4<Bw{B*p|`crs5YbMxkpHJJ8>LQZtS`QCvTU(Ky1?NF3lbi80QK
z{N+<@`@w3m+3SUFqU~t#J?!C_9i^xL<1HoNEhPbO=>Kjzn3@t9m8b+<YVSTPd#w64
z_7~d|>jXfUDo|ze@>ttd2mX|b869AS5&fnkpG2wfUCv~WMK`g1{2<7t{+8y<$&Sv%
z@Epz_5|hN7{TD!oYmj?a6}<=XL@Aa<DRKmsRSiRLJ`WB|I1jWUGYsFM8yoaPhpEin
z<pQ?UXhPyk4>CGy@6Z6Z9G9oZVi_4ahD}<zs`o*a8fz%z*LJr|060Q^XfnSlX|h^7
zv;>U8Qpuf1$SDs9;-a<s68C=(BASZ!2}G2kV#eLta2wnTXmA%56vP<+_YIa#PUWZ|
z<u<niOzVkoET|%f!8+aL`~XFyaF#ZB>NfUpN*L9Go-wUT(t~}xR<}I%F~#8}_jr+o
zZ~?aWp=~hMsRc}`UPc`TOim!h4|jWSrIQqbKQtnHFnf!ythP$rsJXpuD<eoNnSOV_
z5WtG(1dH*%{?79^OEZyTk1cOSSqrwx=SSxdR>NQ;*EHQXv$6S_@gF;}pX>w_Gf?v^
z)2Sq%V1RNJMQxu3GfM(F0%<&ukl!6}PhX(eJWbP#QQ|G3+*jC$__g?Cqd=aQJ&oW4
zJ-h|U<K*J$DK!31Z@Ehg6CzI@LvkyQEry49o@dt;Gy{Q>%zf3lyHm}DF+|fCoR%Y)
zmp24fivGXDnMN8+bcY3ferr^oIL!#rPT1|Bj<QScQb;S|N36zS?}Y~&RG8)>4%IZ&
zLN!G2R}E7xSS_@#&C_m}{d&J19*!2EWG3Gj)m5R4sqVeK;q;hUv*qfr^mUxRP<#L{
z!480v#3<=ZkA9};6JwoG?~X?H#}jrpw)sBlcNY1h{$Iydla8rQn-(zX7HX-C6_wu%
z2I(<le~AO8{Ti$KpkIY|cxS_ZjrGbG_%S1KB)@WC?J&m1y%0n6sT}Aylvp;kw*0@T
zPJ*enF$c#-ZJ*zm=Z;Z_h{~_#^}82;WcuBBNj>i+cL+crYKH2J)+HyVP+)EF9vU&k
z%Yt&mQ#1bGDHp>jPwdz3un|nP`#!>##U$oMP93izhZ!wXoKJ-vFIJHVpc>sI0aW^c
z4cW+-bDk9VeBdpj!fH5k4Ew46J^gve+m2`P=>Y8|LB{$18UPex*Cn}QJ;UgdeF6ZA
z$NTUuR-(G51`zKdHYcgn1#Y(XbvtZG{vn)*r-ifRwm07>-&Xhk9>#n*qTULk+rK>Q
zeTMyl4f?evrZ<Tf%doF3)9c1Jyyf9BB=#c3IC|vBY`4Bhsan_VFkv4L8d8I*Q9l7~
z-P*(Bv9?a+ou8VjYU0AnczisnJ+E*W1!G!p7=TFHM~LyaT;&PUEN8T!!?J?yCoQtG
z`;)R=k9`O7npst&Qhi!H{uQDl7%$ZkdgsS9D37mj-zDRDyU#Gg*<&H=hlMzyy|I>>
zK}*ZaAaT_EeAq&!xi=ITy%eR+KCGBBAgy8dkeT%$K`MaW6^#1L^d<As)byr!p$9S6
zU=T{MST=JFV(P)i`-1x`U~UL(X}1gLo@;H3n5^=Jm^z`=-Y!LK-h&XefLTr|vp0G{
z%`eW{Kt?q&rrH5E!NHMxsN`VT51%oE_ZgbNh$FDZ$zJ(!lu4cNc*1L)TME<F!{a5j
zyf#&dwxKRz7QOOyR7@3ZZBXgI0Tn`pLHg2yT~f4(iV1+Q<T~gfD+XlFwtgOyI;|F(
ziERG3ud);-phs&$3$7DR4oIy~7|}x@hgK;b?MrP*=I2kMjyuXu20IF2@n4;{ISo%B
z^@C@rr=T>x5eR7+CO1Y~`cJzjlR8Zk0JpQ+AM+}V6bHf9@$nE_&LpW=-eQ%g){0tv
z5+h6Y+)lSN)9Hq&F5x9LA;qZeX&R!Hz9lDzK_$%(yoIv8V#928T>s2Xun{~oq%@+4
zaVUI>IPC8HFmQ({b3l7;A}YtGrC6OWT`g{Wz*F7_{2T$=HD~uYEFaJ5G)f?4#v4%j
zm&F?ZXxIAZ4Gi>rgbyYjs*6BfD-njIY}S`CxZBZWDWrK`GGq}tOvtWe76}4+TZ&ai
z6^od-{gK#pu=^I&EWxPXzZU>TgP;qy$9Pq78eKu$7;+y6AQL{Y{7+dkwGkmcyLo*9
z`PCTnIM`+pQGSd7R-G3p4hrXhqHKD@q~93Ftc!)0+{vP<1c0=hn1RI~ZL=E6>B)D1
z!61S)N$<s(o03TMe{_ky8!2yp#CK5Mk*%Ff0Ou@ZkD<jXhsjtDI*=SdDV4hc2FpdN
z@dzapsm#7P)R$xgN0{kne~G1jwiv&L?B7R2vd^CIFs>L83$utVMXx(Avd0ciJJY*D
zt2qBxFd(Fv^CJvV7`%E<WVx-XK&S+K=XJ9eZovo?9(azVvdj4nSj;VFIeXH=@Vt&O
zjA>92cm`!@7z$d@?0-d)fD;b8%Rw~6z1Hk#(8q_F1z}W~G%Ki@*AyO!!f6o)>HMiT
z&7DkO70y|YU?FzVEnL;V5{bfZe}~;pfW~njd>scf<LHxsj#XQ}7JlBVoNWrfF8YEe
zkkjEvv5r+WO$}iVIpC}gX6Qjhk{_YjaIZcS?XEsGK={nZ0u}oK3vW{oPRxJ{w|6vM
zj^=9AdrDs=J+33Oih$)b*vh(>P)T0+Z2bWfgZY)eI$L%i<tRM-NK75vVFi_)uo|EK
z5gZ!;D5^%Zfy2v!(W_9J{Z=WX9cK?7%(jByi}Box^CO<lHtTdYGohg*e?FdC5Y&lg
z_rDs2PK6$x9k+LqIrHo#(^OR(>fX&zd5g(BWn>r9K~22Vr!;DCh8>vc?cfY^Iv=vn
z7X%r0>yPq`jj%Rt<NpyL6d&|oy(iEb_YLh|)N+CcOlP{<{$jZtUJoPp7@G*Y4F|pU
zb&R0zXg|nGPoQ=i$OtNedwUQ1zXXGN|3}uAXKNMnX@NDFYc=PCncbDKnsNAFXu0h<
z`ieH*S9l*$c5>t<!YJtFbR1>Bf+(<7>imMssO#x9v<j}ceI{t1+1C)h<>}N?><dSZ
zWVA+#E!sO50(vc$;gb8=1KSllVRg(Svi`@`m4{Q+cJGo(laxlOG!dmDB`PFkiZVn}
zi6km2lqRI8G#DZwl0rpj5FtaeLc@tdrV2SyQdH*Oz4mdA^L~EcAMbU2*Ok5Zd7d?|
zd)-T<sHkY2p&_e&n;RDEEZMVpE`PxlFx{MB^xCi{xg4|>Ps#QSeuK>*=X0yo9}k5m
z{7B>kh_J-CnjF#-MZX0hPFk8Wx%lEC9`xgD`*odAZI6@^u3zy#{iN~lKtyV@>k(*=
z7ja&rrxKxx*mqjMP~?C{6y>tH2Nef$H*>U<Qg;W>vt;mu^<LQ&7pr`y0Q1o2>`nY-
zSAw(20j{&_kK_aFU)@77g!)}<vd<a1oJy1&18TyBOXu7Ij?nq!dX)EIfiC0GRB~3S
zqw2wG@uJiIN<9@_=Fx~|?eJ(gnNy7I52W=~!0-9@Jq{jx0|Y>J8QcJLC@RHb7#XfA
ziA1)Skf4u}vECo`%E?{-p1;w=+GC3b+%dx1|8(`_ZNRM)nY-Iv)0Kj(HP{asl+*Nx
z)(s<*Oyc%a)spSYAEct-;GcT}qm*4)*j5g|8)ZG0VTy@--9jApmloX`JI(1oh*XI!
znP50RG3PNlYlYi@I-V&GgFu^#m%DqCtfpjj=+W;e<o=F4^CVk?6W7T$;}_`sQQwR@
ze(my3Om+W$;*l{f1Vli5nCbCO)L5DG7+}a)l{SP}pT=Qvg0-zB%D0S=a%-+pn^Cp)
zpLG&R5^^efwkA#|FJ%7m$-jV)`1$`nn`uQpW?GT-Y>)ll?|lFzle)W`uTsspF6*I~
z?c%oe{EqTEOGH_R9bjE3b&T}S5TU<1uJVZp2LIn{Mk{IX#^Z@*cUQXpAqnz({grl-
zk2l?qjf3QZ7jgqpttZ{7u*(Rl%Rq`wLW+oZSf$()*sTBehN&@``;%!*GLQd>!p@2T
zqxF%@VWNWUUC(duAr4Q+iBy=zg5|XWhJG0$iw^(W-KC)P#&Km$pZ{Qxo^3$jhDS3U
zNqY3YzqFy)y}i+YnO8~4dYI`bdv0jOy4{@wJ5GBGgVWl2&`Usm?x88W-&|f6`T7sk
z`Tso||13#dFo-l<5q_3+d~okDwbrmC)&f|di`+D6`c{mc16S&Q$eWg?j`TYKyMQ2j
zSckbf#OMDNOzMr-IWk*9n1{PE{cLI+`r4z;fZ}35`T_aeA7Dsy{(BkimFZg?1Oo?z
zn2jn6sX^VIer<4z|JAITKOc7uzqbf$#z!<4V@#&x15(*{Io6L*Mj;=W%#EwG^w5ua
z6S;g^iYUd52%blQ8{%~Bv5cgoq+;KuhQnVMPg1-!x0h<suJP@ARDNdntGkOamZYnc
z$D$e%sd^V91Up?_Gsu7jaSiAo3Z83sNp@X8@liG6^{i16f@8d}o5V}ybr!VrP-%AN
znLG^9^Woyp9DoQe|N9B^pch)}5et2vfrK6F+2sR%qUS<x#{MwZ*T<kNwfAHiSW1mC
z@O=6abXcTyUSdMRpZgxJ<NsH2e^Wp5$77}+$(&E_G~Xp7LspIhMIuIv=+mEZKmCOm
z5|OHyD-RwV#u!4Zppb{xE^LsE9R3>_;lcm+yOC(&yBQu<5}l3pm>~gm`sr2E9{hSa
zD;Cnjn>U;umzjwHWnbT}8N-R?XQrCa?A<hkR~tZV@y-A5*Hbgnd9oXkI={}W*LyRE
z>1Z#%kFhe!9rr19CtX^_;FEAl-B9(?3Q-$<dVdE5^O3%d*z^7We!3Yt$h9i%SOa9p
z2<CVDk|Z`pU50cR$j+VHg&7~p>iAwj6=XziE2ty9!QMVzy|4P_DS#ohsZaib%IhS>
zH#>zCc;cydWE9E|XNc}u{qHB5Xy?4;tk?n>d5_)T>!$#eH)`m2ho-N(eyeslmDpWd
zF=q~uEkS$l33z57-TTo##Y^QeCiV<`C)h3&6ysj<bxF0sh*X}nV_)E~vJ(v186%^&
zX~iyRWt+YxVQ>PodB$+SC0wMm#8$-qZs>?o9$d)i@@NR@O<EhQl1;gx?*t()HJ&E)
zm4A(RJr62_pF*W5Mn>%c#AIXR=E<Rx{<USox$r7+fgS1Q2LV>&sIeo?$&vtjweQ-n
zofr44)pJfF5+CUY0Eaj>vh64BwcsdOG+2OU$~*4HVc=|UKmoq<g55jrF1lR}Scy$=
z`3+X$SQ|}Dvmhaxtp<m^2k${C`1k+&9(ZVf|D7C5k%!1A&81;dt0+JKcdVWcCPU=P
z&GZcaM3k}MYpBs=B1O>W!zzikM?}`yV*R3cYi%c{5SBPdOcFTDRc67Qc@(L%99FrI
zyWr4dH<0thGD2&^^tcx?J#OgR53asRe~dE+{r*v51CU-MUcos<60lS~;v-dB6>n9D
z*jfvV2#m(0Z{IsFaRslLu}RncG&V_Y-gW}0&_NP5{9YUy3!vcGLO-F5_Y;zLrO>Jh
z6l}0z0Hx{cAcGte2*_1(@$m0JksW%kHe@-E%>(0~5{J;Ecwc{hKeDgxH(=xP$ur|i
zF-%UrU<x-HjoMk^VCU-Efo|92)OWBZl}J3#<7<a#!mNjOJi-4>*)}X`9*^rJgo*^$
z5LZQJz1g&f9N9vUcuyEV4-^iZ5pE%mQIEJ8)63BBK}3CeA$NttLINc@<LD(&XCKje
zU*&t!yCpj=S5FZ^p@7cjIh_rdNel4h+>uF8=*NG*sOp0G*@M<;m6&Fwlx%PnHwfJt
z>A)sGnp(^WsAJ(+yl8Y@*mVQiig|YjWUghNj&Z{?2<VDML*D#xDQSRpj<?5jm;c+N
zz)8=?oFRm>KevMGcV|tCGY*d?ajV?Ec#7(yoo)&Y!A_T5;1;gEdl1Z7mlpiQUTKv2
zP^YmzgBbl1I8!|xo&vR#7591*%v>yfD0Q~qOn!lGE8^E!yu{0i>jLg*rC_|Bam9t+
z)cEV$cgn3YpfhJTEnB+hv&L(&f62VVc|wN+ZieM4pAj_NZmp-2*-oVZZz_S!f*I=1
zop|=Uvp=N?6{dO-5GAQE5cJIcbB{*0t~J4(F5OWHxK#INAE(bID38NaU<ngd4xx$n
ztr+YjW@cu!ueWrMYQuu^#u&BC9+;OD6exfoE2sR4f#$?}AeP;VSL{=8?`+2_0DI~E
z3Yw5~cuFf^nu||o3|ddT=OaUA+{ZR0ascYws+(4^TY5A%3r?ez&B%gfux{*oHAM)n
z8ma(>Us|wxGw)o$7h(Mx8x3$SsJNI$WS~HN9(nj|`4u3ALaoe#%#PPS-k}FM>w-4b
zo1?8d0_vL&g`nYC`}+Xa)Z)q6+$YOt)05HHQBU~{|Mq}$;+EX`Y|lzm84O7lfAVBe
zwmt}aFUpQ=gH4+<ZSuvQ(y{Eph@O{9#Eu<1a4eg-CqHU2jLg#cHhDs2r{-d_s_P%C
zT}R&|IvK7}Gw>A{t6`>4HyKsQ)1>u9vSpYxWHhz4TT=_M2kj$`8LCfy6;yKM*wF5c
z8*A!ej{hLb$Qo33iUzXl?v6bqD(;Bgd8+8QEq5P^_6@Mw<QpIHLMyY}^k|}vnsedw
zO?_~pEME$F{2q=LP~>Wm4J#gcpS_G3;DWx++)ic&<rdMj(Ge)xFsE~(6vs7nOFzW8
zspz;E-r||aHj^?Itp#xGdoH&cJY(la;lqo76clf@2Cg#^M^m}HTPQ~%c=KDzv>Z=S
z%aK%Rx2-E;E5OYoRgZzJ^$BVQ^zFP!Os(RSx}YB=_h!k5x1Gf-e`IFnrnUHbSyZ4G
zw1sTgwCTsPekZ02cGD3xA^<J^jo{#T=y{YX3`uG#-_YiSY5n?j4xK7!SlRRG$z?RO
zj*@BHHbp#gxA?vy%X~M9zqM~49~&Udvo2E!*hoaSk^0p{1xmW~a7q){@0H|b-XIA_
zX;cT<n{*X7dkdVX#;2xU{XMgjDeixBtr1+EOKdAec!#}`4m${JBWdHX+zrVg7G^9m
z<_=<qQE`ioLx5W<(JJoitP7kYZ5B@*PRkzwz!1HMH!Qq`SwOr2ovbf7k$Kz`k)spA
zrp5_r<08;6<6P%UIN%B6M~+*vf4546Dam>}Pu2&%m+b)Pgs9SU;&ZTlzQF`W{9`si
z5bBek(wf+T!^BIt^CK%5+fLH~W4{mvhdFS+>v3_58F6d^I*%dagT#08BwqNwIqlE`
zI|stEz_B6%<{y#KEsuCi_$xg97x(E;2~hEx@LprTFg-@PPQUxWzD>Q`*L2f?XcERz
zyc=B{?7)V%n&vVMSr8D629_}FYYDAh>$D#hb&|pxF|HLjb&@P3H5Q%aHerVZbE^+A
ziuRW%6kDBKFo}(B;##^`?l_@G;KWra&M4oan|>H|PT-FzQntohmf}gfYnSgwf2#Le
zu(n;nS<da1(pQem|0GTZqrS{AHh#eDo*$6oz(z9wsx<Y&bl8wJ%%hQZnAVE!FMn_V
zkJjP+05AFhU}yE5^9-{-9I;%s9$R{?CtU{4X4tS{6yNiv2&3vN)1xsHeyTJ8co!t5
zdk1)Cn-E0ejnq%ndY9T>Zsr-W{-mk2M|a~H;6p7KBX;a<QPIEr$#FSLvd_)vntbT5
zSlHIRKZXZ;3JMOgU88YJuvo?*H*)*hpMBI%UmT{s!(fzw$cmJED+UiXzG=1e$a;+e
z*%{_3#(~Sd)*Hy!E*m1S-hYjn`Vv1+&*ZjwZ9luxhX3AneEjj-2QFFtk~~mxFX7vS
za5r|w`@1zC_9nLF|E6Cf49=-}#E$ahW@8x(rR=+}RTfvrMj?vEwHR&3O$??LFs#OO
z1dKz5DQx7%2w{~k#0764&(n6*e%dn9NmSt(*l>b>joge|q|nKM_vqQv%2=2vkld(0
zY2m*5b<sGdn0YAEM%{HmNp6goEsHCk#jgXDgji^eu4#Ds4@U3iA`2^rqXXKYKl*6P
zN2wTt&K>o<XjuHJ_uztZ0uO;&o_wKf1|)xdbSu#}mK+fY-UG{efc20Fy4TrWzQMJa
z)=H~uB(#X0K4a{ciOSAv!sf~;E;~j90@Z!90j;4FI_x8oEC*2BoHUC`i$?A7!Ujj}
zm-n>y?AdeRK<d$c>YHtvv7UfkS(dUZ!fQaNGS@T&{k9f93Xp!{!MV@S>W*hV{`di?
z09FN@3tfX3zUj~unm!gZn!(`I>2DSE(1DF%8<ea9<&|lL)Q7^t8%h#_Hfww*Ev;MX
zIAri(@l>_n3@0CaNsi5>yk&l`Fr5|UQ-uls!mqY9lD5HMdcBs^*^t%chTaYT(~BM0
zh}quB6e0({QiEez26$}qr4(A_ymmmSEce6T|H6kmqf0sc-ZWL%zh<S-)InmK;pe6G
zx)6}X#>ybj=6`Y{V;GDQ)?l(V-SY8(tph!?&j4WeA~!eg?}Me_+PEqICXB0EcLJoG
zM51!%q|d%FF%@e=nCEB4qPqvobhJVTwr{SW7ijhzOrwJxz7@a^BL@RyxJIxjzmA9$
z?MGciM|mfu(Q!i+wsD2Az6!!<S9Hr#`@<ImMJ5qfLoQav9jx>iXjsyL>+^<PJC{D8
zPoF$+6$r@fIOV{`00*|-8~re(K^5lF_(OeVJ52j*Kks29%w5UlwIt-21aKhA?1Swf
zPUSp$7T^G;Xxap1CZd77tLWA;n6(#W!}|280n?U8@B)JrXV^z+CsC{g!GOErVJK1*
zc!Q>~qb36Si^Est9)5Bo9X>Mt85@1cXVp0=IEfyf0pq{gu4#4fyk+P-y<qb{ItipY
z$j0<j4Tud2p-ur%mlL+IS<%|h>*G_(@%=T!x!-S3->(;{Ak98_`01(9%+tlwL5!Dj
z`LYE2OI^|JFGyXe9vD{6llymjb-`m;1P-sjw=|R(t12P5@EQ75;m*&nG^I?YpX^A1
zZmP(*{qh5srD66WAmKS(By+hwWtdVhY;8kp*kG!snxkvKK1Wi);U!_CSJJJkAP>Dg
zqmO6yfHwAO_fY3-nq)K?Z+nAh3m%g#Kpc+eG6ZKNls<342<68rZa{|-G?pmmocq>u
ztt<ySaA0C`!N|e%)alc?xw-KI`0zX|G(0C<1jz6+eE4YYhaaL3*J)`^27zGQ>$iD&
zyJy3Xz_&C7zUl-TItoTc@K)bG*`?sv61f}mBM93yGEg>`9y=S8-T1e6MIe%WsF%uO
zV>;Y0ZfMeGu1L=Q1d(K8exUJdu&+P9{3+snql2yF@F9&OpYg0$Xry^7&<aU9&A@T3
z6N021X>e^gq^<c4NNc<E9znyncR7siaw$3|5OYeQTgCHB1>Fm&jDRIHQM+O<U)~IK
zHyQSEJ*RSsqpLNN(A94-{UIu}-WQL@sOYZq9%Hzo^5!!=9?P0$LR1KFPlDLK77$%s
zcFpQg1*`}b7-Yk{$ozM*@z<}{Vs`4~R$bzyj*F8WC?V)JgP06hscB#>!g=iHp2wK4
zq@pk>_;zyC%(t`A@(+gwNx81Uh*Jp%yf8Aa=V;>1)3AIKJjReij3=xURugL1LF0LC
zgLmkK3oez}edC5&FXz4^sG80kThq20zTN{y2<8_=0Xdv&82S$k8h&)}te&UeHCM)L
zO5m&ucmV*bW+T*umh8i#9BtLkE#4mWgSyZC%~3j-48Xaxmp&(LAYt)Cw|#c>F?W!e
zVb3zvyQlCR-c6<SXvEXRll}6kE=)OW{5FF=Q+}=k>JWP?3l4>0KixVji9_4~AxVN9
zQ&z|KT>)Y~sN5Tp1(VPQCIo4eQORF4bM8RcM;7&be5!dGl0~R!OligjgfwGtv%avI
zD_CYFi59B5L#=rcchF}#d-QO%6{|4#m%{iBEqO$AfzLWHj7OQ}Pf)IMoYMRDG8XU#
ziECql^E4le6OY4Nf}H(?#T0Tz_A4IKtw5JokOI-T1Z!9e^>$P==PNGD07%Qs*f_9h
zB7W%3()-j;8nmlNYAR&o!B359?8{OOjz^8|>zVCL1zXlKWo$9ThnsURbPOqxboE^M
z2oP^Dw^^_QRH)LjZ6H3ROn#XIFMe5(7LBZ}0kfnpm*&Cx0KIq<w2U!Cn#__`wtOUr
zCtGr#Ttnh`1N{N^@pjl!j#-}@Z;_>f-2m)q%GgGPtbuSMf~h<c5L6eWf;obC!=(tu
zO;DyXqPcyxPzG@Pv*rB4x&AIR1pY4e=+UDGcjYj^dK}C{i!|WE_STIM@OI*zhHR1S
zMAV>P+OOg;$J%=InKJJMhJpfo-T;f0J+hzK1ny}o9grUwiLzJh-{b9pM*0R}<l}S+
zb8;<-O{<$iL&eC67Vuofz0D~}unXw_O1lu%DAq?MuS8iX5jA4T@Cfv$Zaq<Z&&{1|
z^$}cLCo|SC>;Qb*XbLVCl_1Q>`g1Qax~Z1H%k+ZEo)_#)amE+|e4O+Ld2H0V%{-lk
z!Pq~k#3?K1m(M$svvej1CU7QL=i2>rFyk1e{f5mlf|q#I<A6@P=t`}=6y`WTAR;0Q
z^k})LaJv%=uJ(zqK(E?(d(7sUttHUTHQ3xCx$zv61&_dITcK}_a7(Lo_uwHzXjcI)
z(ldj1kz*QyOxCUq2gi%_RZ*<IEJ7rA9;N5Gsi@)y=)Zz{{D8Bc#69~1ko3sbXNsBr
z&e{nn`(YQw<xrUq%%qSYK?6dP#e4S7+8`RgLu4#%s(|KBIOoB&S5BYv$^Z4s5mM;R
zl|mZ?(Vp88Fwc46>y<i47*XR!H#awyDYci0@)XDYk9XW#wNsoSl<|hLn;sNd^;NHY
zi{*6T&NlcdW>ykXTw9{B0Jg*!Faf=^=@ozlw~AXj99<4|m;;jFz<upBe+_p(j;HWJ
zGs;+Cvkk29REoNPzi2Ml(<<LGg5W_kkmeQ=rvH65QpZ)>XJb%)nXAt{<nCQRhj;n0
zPB)-wO%%MaQ;&Ie>drNDY>NAX4HwZHJ`5F8xSo*k>Kwk^5(gCjHwk>;7n}P#f~yMf
zDp08fAQV;Ki_IC+_l{<&U`mfl1rb&}#k#m4)=7)Z!>RQA7d_99HE#toWr|KSsub(x
zVZp1=h+3^I^IGYv__rZdCfu`*`GT{qi@E5alP2X((6FyUPNGDK8r@$Ny+{h3FZCR?
zIa5iNzOWFVS$2QV)wTURvoWkKHSs)fP)V6;ue`b!q}fg-wV{RDQHOrPxogvrj7X(H
zBm8Xb$-0)7pNQuJ0{G-^kC1!|GBGxGZYW>6)%gmNOU%=Z5k9#@5XL!6PgzRy%s5L|
zkFX(J2c$tFu7@Qd>`RBSXsFJ~Gu1n1Bl@J$`G3JJ*GBo4G0)@YK5}568a)d&a6^Y(
z$nED08K^Ny0G-_P&RXSwS-c;c^OPH9={-c0Rp<*G0zMy!kq-0ucPUhj{_Td%)A-S=
zVZyz)^Ndz;v1|s`U>Ma%Vh;%F+xFj&1TU||;&N5ViRWOaui-uwxcB*ttX3-%;jLBy
zaY{4pQzHq&n2C+5F-c{#n2Dr3x##G&rSJmW$F%#mBD0T`um?>+u?A5xwcbKPlb6ti
zkyQyi^owXOm`{(#67(2_QqWSD;-<(i)6zRmjyrotdNi13z|Ca;QtuVW54W`(v0Yw?
zz%0P$IU)L61yZ^8R^l$5`BFeSwEtECL+49sG*zD>T*^w)zEfB~!&7Z)DKBbX=%Q=1
zssCimpIglmdNN+X9b*gA!_T4^hZARWPkcOb70)`1iD;!sbv+#Jc3DL;_~OZT@LNW$
z_2@pVfhGL>pNFC%3S>~QV+1xnpg`v^!b50zC6d*ESI^M*Dr^-{n#hf5kI_d{VVr%b
z9V8?M5!3SYG;EQ|qkOL0ss>}2LQIX2#U8IVyMY(jUKQ;QY<<*rPT7rdKkb2EfO*>D
z>al@qZvCW5-m%dB9s~vf$c@*qw^J3c_$pAi4FKllq(~(_x9OhZ{OOTbydlt!ACv`d
zdu8j+NuAC%ZilwsoZH>dpq9f==sm&B@ze%jgBcMo0yniJcEYf~Eke+Xoq*Ty<r?Wc
zEefj*is_PAiJXwo#AZ4YvCKrQWcVpG^+KBaKVobOD)?{q(?`=kno$<e^}V7qp1DmK
zB==ylTFs6psuoE3XsH_Bmse}>yyu6--QF`)ZON2*Y~$IpXD8iPo5yu+VpO##Bt49J
z{|S*t0a+FUGc3f?F~uBB*FkhoVYFoQ^LkYDJfot9onf_8s0&^w_<4{ARlzb&SOSSU
zdya@9*KVyOi+B%1Qc;-zj%>Qlh|6<o^I)Z7R8j#nFdoOEzPYOB>FD(!4-9pD0r%N(
zY2eA}f{0P4pFIagQ?oH+g=TZLa|`3w09QBcG|pg-WD?xCHv5{d*a97d`(A4N&$L5K
ziy`EfYgZ$r1v^;0v<RmAfmWORRsf&RX_mYhM@J0>rw!T}GO8#AFz4a_4>yj=<Anm%
z?bthqE<VkCs9)|mZ<7>f1Ol+PdgH_IBN1jEe;5jX&gUWEsb(7_9&mMNSPi1Lo3O9V
zo6ZAVzHtxPTyOtrnlVXl8({DtgQEnvYOgWpYg#ld|Bf?;Dh7=fTFkH_>nr!8rSPYR
z2#)8vc}r-|5s+b}mdE|Vej)?rvB!XMSqdS%ai9RzZKQ|B5R{&S@y|`=Wt}yW-C5^s
zk=+b4AZz@7^s@2gIlM227ZR*bdZ!=vH?|$f0i!Tzs9<7paxylqN)Ky-^+NCKvWP92
zTkm72ZZ$dH=6TT4_HGaQSs?OwHq5mQyAScTKJBlLB+>r_`J~ItuL#E+Dh>ymjzQiR
z``WTb^61~Y5goIm_m*E<JvpKu?-h?xugIyk<<nB;*}#-d4|fCC==G=l1yBr;<9jw!
z{Yq{}sn96mah{%MJfLAfacNIh$7lnhvzVTX6$adV>V@<&7WU;v=jj1#M}ze5#|p^{
zA|hD8Zo&?nuaNeALV_n3GEzB=IUv9&b<$)jATF)qHV#vF!mVu6dDo8NfCS`lgFTqo
zWf9frHvHJ93qnKmaf3RgGyzweEhn-`C~VOlE7n`sI{DK>JYgNLFgdi^W5|2o38>Pu
z>{~!Gab4v^$VH(dB%tsJ7n?@YCP@`Rnqp(Z3r*(Z-<7dO&oZg1u8uyVzbah0aAdzv
zUfG16(f(S#%kB3B2+(AKzMq+n66XbNV1-MM`uGbnx7t;BzT${mK(}3MdPzLnh!wkK
z-UlO#8go9RM-%7E%Z+PL*+fk$US4FK(5jaH@!*DBzd_=-h>!>;XAye_BUAJ@^KS;8
zj6cz3o{UC;NwmUCSin|bf6>6>O3qc_<o$D-9k@DsmMV(SSy=^GmU_^9u_D28VY$a+
z9D*)S#**2NL)^HT>P~60U4@Bm(ah->f<6<Re@1ClpzpZ}GuD0h|3;#k<B0A_1-4Xz
zpg&MI5`C+3^?p;lvMI{<O{dX{_wVr|&f!OJQaa9$z=Ux?>oGLYPBE9O<-7FHIvw|R
ztltE`CqXB>&)))LsTy<6IIz)O`6#m%1OfL393A9=N<r-3n~NKth{n$rL5_P;rblpg
z0-&dz{}3c_o0#*tgsBc78m8NKK6vl|>Ki*!h$B1ybfH>Rk7|RkNI@6`Dq9_{+`oVS
z!GluAen>}`OKn0Jp0TYAEXfXAJ+9-7mT=B!9-R?730m8QDHW`93QqO@*$z~mtIoE=
zXa(vBs!L4j5ZLT+WfQ&ND4TriQue^2;Sjr9St|#l_k$8^K-94x6Ds~4Ki}U9L7u-S
zqqq;r{47OlQ+sR;&Lo&Qoq5O!?4dKPSGztWlrcoY0#ihM-TKbM4nn#D&?t<Zv!O_)
z5SBQ8k>9r|yD8DXR*Jm-=4=w4lT$e7LpF0KnjUyMmE}lv1#M6>l^)gzE!6Z@6RNtL
zItV4w=_pN49lI{Y8NGYy4;ly^J|@`(tQhF1EQl%Yh3vp|Oh4o7QGOLt>3R0-YRy8x
z-)=Bs^I|SYPy!wPq4(R0o>v^sU0eYh2vfZU>0AUj@$a|yvekF4BmM>G<^9B*GLe5E
zs+O9#2W^!*chtf_{{4?JoC?7R^KQY+L=E|!{v(Ye`)3f}g;5F@{}T|g8+b$1E?%Vf
zfENP)I1M<;_A-ysVIGj+ovapsl@(?U(LkX0h^g|b9Yc)T*Bp%=mRAe2go&)eGaRQ;
zXeA`cZR5sp84nSyR$k6=v_7cZF;`aFbS0H{O%s!wL{jm1YFV*b{!|_tzTQlt2137^
z@!?{5q`A=4b93@AX1ualDX&yx95Q)03nS%9hQ>}rpe2syYKPo0Uk_KietoFCvH@sh
z%*^bXr`RAeMUUf0)MI<ZQzm0hDhUb*ij(I$G5~!Cfn?0*<8*-9yiV5s5wWfW{`pH=
zy)`5R9cTAE{hoOmSZGOQ)2||cGjly@_TzYV;ORzWFE3sSz>6GN!^GEbHp7)?(^76{
zOBJgpLd=KvAv#4sqoT?PQ3X=?<>w(d>v+lJnB6H&po+3TAA9p=-hd;v)eb`9m`-8c
zPDY=9)(}KwoYwedM%v-+W!aAoqaMj<QlFt18i@WzY`ilBruc2?34^nl6csz9qDm!3
zj~$C8z#sdGT*IMPw7yR{=W|c*;zBbsv`FAsD3fFYgq<h?K-e*$yoQ9mDh<Grn4|~9
zZSUUYw)Vd-C@Nzi)p4mSkglF&_p|aLN^i-1HD<ajuf#SwKj#=|LGDWd7y>^&LjcMs
zyZ|AjDf5CPNF=7=Vw%#yS7AKJQ_ZKR?)}iETTPLE@9cMS4UQbl09VW3fwLnqCgxeM
zXCvg-s_XLygr;Q4{NG{s)TcoGzgUW=r#ezdUBbeSmyyXi>uR1AJMVb00Q!lOC>|to
z5l2IyoQ!K`1_u6H`$J6lF0}~dU89EZ1Mmr>zRq(JP2B+&IVybG|3E2}P}i}rWB+|a
zx097uehX;5DognGi!zaQ_+Nto%0sUCcj#=c_ngg6#!SCzN9)ccj7~oRaJ%TH``9yb
z4FdUVEw9%z24xK}o?}g?^amZ)*0;Vp2b|RLIi3yE9p!Wr#vy1lKu<?G4wfQT%f%Qr
z`<5}kfmUNvc%a$w!zF9h%$ZtsgC~}JLa&<n`n|mKX>``z1qgazsyAwKz=ahJ7x5Jz
z=7DA8mMo=xSVCuJ(jz)&QFHW+M^xQh9s>OL`aN{|S22llj#Y+n`s9%?1^s%&eKlpP
zu^{)TcDxZ46?NwhE9i(;m95YQ@WHXB>;tBy*TcVq^GYvmVImezMI5V<bbN<F@ONz`
z$%;ZA3bdAy(<m|6s|%!&#kzIDj!&@#YmvFYRR60}cB9^=EmeetJGN$hVIMhxrNVL8
z$K)Y%?n|&s4&y58oZggb2!lt7jgpLxQ5!6<=?p*?d&<XF2$%jnS9M_2G~viN`eF=5
zu^JbE$~t#&kDAKCDc#zLL(=|GgZio|^Tltd)ysNC<R^3-$9nG4JS8bPc^}~a(qr`?
zP`HbQ0|e)4G7GiN`6NcPS4s7{PJsH>vF^11IaB!ca;o=+-{|~?0ePgIuB@usg+#`X
zzWpc9{7$V8aszWK1WadqIAJbepfmbf=DrnKgpzjD-5r3d^?nKJtau5ElkqA8^YyAi
z0qKrXxbxh1aw5c$70k;+o)*h)yuaHr&X`seC`#;95))`ZpdNMX8?vukH<3+>Saa+;
zHK;zKig3Z%o}f8#ua`NADpey`AyF)0q5cT0B6IFN9gQpBgM0vqMYsVXEx<VMEL9WA
zTZk0HmG$u6!RGC~j=l^?L>xV>+F_3VC0K2Y))x_F6dnDvo};&C@>PTj(tq`l8pYqZ
zaeFx)E(8A4)ynt4f4aGs#8A$PhcgVTaL@jWF(rSF^<a*)G;KqpQ?KHky44Q`pomPU
z`VU)Ewbg$jAYmrs=WSMF`QO8)++JU<huRvuq@BUplbCD3q+*oNQ29g65jCOd(|8`~
z^aFEc@K*$Wyw<Ir)6oI<1>a=Uv~L(K>wV_nG{Bjw5SN3JE!3Y_OQr3<p|YTXDCWOU
zPIK?-4Ajgy(N$@;b!)UMzB!0SlD{pziFViyJ<_29PK?3mkmjTV({JSeB@{YcXDX4U
z0u!>T<OKrq@>`b2C=yu7Q~wyAFMaEcA%}-5eW%RRrIDiYfLQ$ZJ<wZyDK&#_WGvSq
zu-gS3JPeSN33HndQN}S!D-@MnwPq^Xyal||cGewEQ@Eu}*c*?pK8-DTI?6HFX&WyM
z93>}dJwnBSN6COJMqT_3u=Wl5X0u&r+4b+=U;E>IkN^;%1{e9?N7IT>TpX_jQ-&Ah
zGMyZ`y|rj{@(KA15_0TpKj-iYY9P!U__vMV7m{|ZLy|aigph~zZl2Lfi#|d4GjyS(
zO$|ZRoPZ`Nfrger$F-EN9v?p0fjRPSvvrQl7QC`#S=j2ZMFy_Ip%-%;Kmn_+a{~Rt
zOQ?#HB_GcV1410wBUUeMHs&ZN^I(Ku`pTCRvN{j$;Imr(WVPZ^J!01!M%RF`hhyRr
z5-4UjDtU)&buArAJ>PlzlC6&3unJiM<*0v#3VF!RsMb8ues2#shXqVrah8gA-!Wcf
z&&qEB(_jG^tl10E<OwWqV&<D91kW|{YzaGPl0D2ZP%gTum+D|XQ69GkfaEmBH{Dol
z1;jbMzS_R{&J1=<`^SpT<W|zjcyxDFM~xmm<9j*OW7vd+*AeMjALcu9i#vx~ej&d>
zkCbTtlZWBUKTD~$XtUlb>?&W<>kZ$FZa;qAd<||b9HWdESm$UZIT;fp?0V)@N>`;c
z%_Y!4^*QP(Gt&|3n}1&f{*(1K3H8lh&s=M4JiROoG&?QKS(_XLF&Ruzc+y^gTawTV
zrf@5!h+8YgMOu_(xw}(2sa69RX6UB6he%PL16CJCIoZL>F}l<3^`{|1x5V}7rzxp|
z2cJ3hn+lt=j|AE@Eg!x#iY&MW&z_(${Q+wZ&}A}mh@(}Wc=xRk+We+{f)#0$8_E~P
zO-DYrAIj2P9tJ{Aqdg>nDku|4+t&g*b2lbtCBoY%Z$5<&wFV|F-VmW5r5yQeOq1#=
z`_o|0{~k&u<_!uDIk}-uH}e%KW(@c$A=c7MYiYQN9?a%{a1>^qJBCLQ(fxx58zP}L
z3t`9gl>MoF{J(1R0BLQ(A*XV*U!Rf@%#G^WSAlt-^FrA1=gcP^(nM$!HTk;IPEJ%i
zxIb<Zs6Ne%Z3fGhdE6E1F;R5P4`;#n|NZsg)~z+nVMt<{Vij8j+e4XVF1}ft=W=4q
zLw_K_+kSpK(n@IR=HFUrx~hQm&z=br<tfHwCuE|Inj)Dg$cejg!(g(LXz>>6;ktfh
zdbo&rn)$S_3j-=?GS!{29QQ~*=;C?|TVW$&#=y5=9kE=-7xr9&{dj;5rPP97IHPMS
zHIjFp;N}7B3HVb77}+hZ;PE<(BzX5wGDM{r^_-^fA3yjO0>n{(?CHGo;Eh;=I)-d^
zx7>Z91DSwkHLC;RX=-cm#hj5ea-s9Q1)RZaVY2Ox^D8@XQb={=wOP`H5Pdq$U7ho}
zyvy~QdGF_9u9DT!f-*_r^HPn4>oB3Hlh^`wn_u%?A<s+r<ay~zjEy_J+-S0F*+T~Y
z2~<ls0@>*kmjQ2h`!b)srdo7)j{mfPAFr7S!y3u;rMw<)DAU7*Ce{@5Y2s~*ZD^my
zY}j@_Fwoe<L}rwbhh7zonU1nMmQL^Nb>KS^QYG1)hmjH0ZE?8rG#~VG(YZ@@7R{ra
z&E_Njlg$wYTUuJajDFATQNm~YqrxdUQH20V{V6Mv;G~*Vp#SNR#*T?<`%T<ozNy|p
z%>?1b+@H32(yd#&5sr~F$fv#H3?gTK87uKWx~TUFaKBZ`2aqXQ4&ieKjc^8)Whhf=
z|NS~Qqv;&3(bf#-rfoN8a!G~`J(SlCp@%{bnF!X@9E=Tf5W>wA5t#3Qlu29aHHw})
zdL76Oqid375ezyVHO|WzDZ%FKYtBJPnJ5bk?P>U@MOzdCOo4Gx!<AAA*!Bbjf|tOQ
zd3UkwAkew~0f=*Z$%$^up+-2pK>BQH#86=Nw$w2{z)?Ag{0DsJF<>2<4spg*JVKlj
zY=-vGURKCmEV}M>9!p^ra2}wRSqqhvtw#A!&u%7KTj||4FN9CcOXitw+`IPW0#kI7
z6zxl*bP%AuL4oPqt1twj6-W6D>Jl=jOtY?|1#IYv(8={rNM|tG6s-_=VXj32{+iX!
zC*AhJ82#8swV)VcjCY%YX@H^1V3<8890t(rx;1>s8NoK@R-1M-fT^NcPVwzj%*tG1
z=}b~Ne4jknX?U<jL8;!V$7e;Gw`v{{Y6l`zrS(ev8`Z+9aq;VJPo?-Za|bgtr@lF+
z2_@I^tCHBa3l3QRpYb42%IQ=;%YWD0Tdzeqlu=`=xrJ5w1S@r=7T9fHyszFCRZ&(l
zT0vkfKTS~~z58EWCuZ7ycX@royjE9ThlOjcYL|N-zq;g)4!R*1XU?37f*Zw<!6P7B
z34gCE0cdhlvJYj>eU3cS-F}oNy%ImGjUxs&slDGob0XXv*|HJ+Hfg;o2$~E6t|V@p
z*he8Rt35X>43h`aDVuOs7$@(#z6NT>a>rCF12uXT1`1UZ(LR(a_EB4j{P`hirD2={
zMtJL%&!-i<u&%62G?M?Vu8t&jJ-6H2M{Fnb4x@tF=*fUOY3(4-?3J<c@p{VE7@JuV
zH>eLKLwzKsk&0oOTZuimBXmcC!Q{|)^Z^^uxR=iO0w9zLdAt5xk-9ARBde>^<OZnp
zA|VY6pP_DXPCqzu`HA&<eaIyA=`)_7Kf54r+NQMxRi7Mcgi#xv5BDJ<U5%5w=<irZ
zAw@o09X?w;+B<W|jIB}beNSd*b$-j5)Zs2J++n1b$?)KzDzhrJFKvTI{yu6jAcuz#
zodoCygAcc4dynkqoQ#tE*VX>IYo@TLB@i%pGeTIC2=}%WcTs_h&prALq`j(v35~S0
zkph~avOax&ngSuy)%I|AG-7rEjY^uweLHsRiOw66!)~Fz1q4o@?yXq~_pcy+A*UXU
z!GakR=ZZ?{Cx!MHZuzM<26a|oz8g|k1fCVv=vJoNN8fYUS%|q<Nj+<9u95Eej6oy{
zyp#U%dCvK?zZTFW46%?ae!K)tYwc~h=qtxd_9P6{XFd8C^#G}jk$>6mG+p!b6ns(m
zO4;#Yj@_(eyLUFI9+RDw&E&HyL&1|AX9xsg%C4*fzilX<>N^SSc~N(_z6w(^!Eg=N
zV>0rEGJfvs*NJ}zcVJ#}4a`ga7Ahr&e+Gn)Nf>(IA<nw<Z4n_8B>Sb;I;a#nICMpc
z+rg2cJunw!S6dhKczkP5c>H7V^&bv-0eR^16eeIO;;{e}1Pfj&al6C_!iVu0nEadY
z%hyW1CjtfKjF0~rmxg(&bNc7OD!opx$GVWrJ-G~%*t{un?AWohd{Vt*Ow7$cX1R9-
z4rs&@9MX+D)_lR$c$eS#iq|%rhNjWZZ5&*ssUrsSCBtRVTnsHlGpXCd`Hg^w8%pwX
zF!2rcB41lnBN<wPXBwtGw&-g8fF?hb>wlh}g5(u>aA59jv{YWc)Pbw`S+o&sg4&ud
zF^apR*U3=KT0X1k!Gk5M-{<Qo++fhGM*7-z!@BBBvnZN+(F1Z}2QBmh4aL|oV`55u
z&td@=bacr+TD!<P0G2X`UQY+qShDPjJ5~nyk44B55Q=<2$7nz9mYk>=ZI<SGW=VPU
z7))+3ta}oIldj^jq$Wxtq5EH9k<)5vjzqAxVw1R&sNPqQMsG*T%__%w^e@b`Jk;!F
zNl#2+C)$&|qQ2XqvqZ?_`E?|8&6lR5f&L(xj+|5K<QzH1SHlX<wIQRQHkU4RzSWbH
zf%|?!FDO*LW6yV+dfU(kq}<hID=~bbtc0B8pD70SV#>Ef0Y?`!vL{$3NtYuIR72Gt
zsQH|JKo^>W9@r(wVp3xoT5{+fQNzjlBcGaHq$Li;GKH8*je!Y^0l?(-xmSK*NB>~R
zXNpF_jSgVk_li~wJd8fD2Z{>%gA$bPqn$m_@wtw_KIg?J<@VbdCw<n8oHRk+@N{ar
z<LRi3Jn8jxrqicIoxbj;6Iqb7Y$a|PO)=)fKk<Kf31WkEXZAU_T_#3GSK(wN0p4{v
zC+r7nv`(i-i$Z`leX<$wckQuknv$!6lKSsKz9ijIfh-j(h4dh3(-(7{fmMuo$xv{~
z(7DMg%W+h6-U}G#Q0IeZCaaeB`f|c3q-i17{K{PEnRKAfjg+%0PzGl1bRXZH=5GNI
zSUP1Jay({%r3dhj)JS^wWuWRm4@O!zAt(cE846UNzq5ZQ@Yot8kRkUF%pqWt{GI|*
zs2|{mGA14B!M8w%d{O@g5nap~{Xky)?9N8)x>{Kor2F>LeCF#EnJ;45V^5_~VqX#(
z`5jp7DZR7;t0HDmHRU(|q35Hq4b+ZN!rN!4RN_iIqiO3WyA<_Nso67n4ErjK5m?RJ
zRsJCAbQ9__%-4@^0!wj>^C4b)_<x`N4ooHH(d+M6dm|+IH-ozDtVjbB6PsS&Oi}yB
zsMldi^M8Ft78D3cgq*}A9i;Jja)q-i_;hr>Ma(cXRjcLWVN6j2M$TDDKOn-{mNwTi
zYSf)##gH0-2}w6iD+@5%sidQ;4uApHu=N2U$-A3xLA=)9s+6P<xK&`!XIATcKQ$|1
zck}=}edU>P{?1nDAusOrbTEV2u~xLAwKs;;$XWyWJwP_+lFfiu03?O6yYBol(|`)}
z&hvG+Au8Yc-mf~Rd(P)H&=5JjWDzhh9eg@*ebT3&xsO#(MkVP;FAS)n=gpTMCZ8mP
z%7&Zwc2&4>vYARt&{{aK_B0~;HtVHANU$a)Nx{SI=k|anp{B^~0sDpVRa5LvQ_S#k
zvg6Vy;7HI-r4yfm;(y}b%eL=g&hF*2iUi;qN?pAI=LuL>PU`t~Kae-NcJ*KiUrrTN
zQ@*l9i6O11g>E`NUfAtE{V+D3S#ER%R`>mg9*f9aC5A@K4WHYiqsRD&X&t{VWUaDe
z6HPCwL#Y%fC-9Y-lKL~lSLfm{U#>(`W6g|{O9V|!O{Ky|A}2`p;fs%Q0+dcZ#1MbL
z57xk;<3t#P_zWQ#DxyZ~)39BV@otg)=Y1KPd9C6uH?4glE`X7c0H2em=cMH*^BG$L
z-w&W(o(mg)!6>EJ7F1!R^Gksx%|hHp2I~a>q1-{*Zc4qGj|<(}_pdz)1SA~tMQ0V_
z`ehEKTfPUNgdhjHfFoIDVNXgmj-gOCYV24k`*<|kx&aA%lWfkq=A}J4xh!eQPc(3K
zmBpX=Ox!V%&yM)Pj(DSyU2y0^Iaa;p%t9jm5XpKN>v3UB-zk0@_)4aBr{lWIz2WBY
zm|#f0u?D?+0%!*apFl&T?0?T<B-6W_!;Gyykgt&xqoB~s%S(G{C~n>jp)o~>N-2E)
zekV87ed~?kevCOtkJ1<?Fn0DjKWF*8Fr|0BUICXM|94IXV9Nfc&q_5OGo9Z<i@Q6&
z|M~u){ivFoM;}DLdF`O7Lp_Es`KiXbJr8KoZ`dZ&-A|Sv=%W_pI?d(L=-Vm(dWn4b
z2(^5h?D<;0|9SEsEU~eE*<sH0n1HDeC;vGgFX-i~Uf;|p>)wOt&F7t$Q*2yYSWv*U
zd+od1y^j2Kx8f~GrZ=WJYo{IskGusczv|13r|gC|(%$w6i>_`?>oML#Q6rnWhX(LA
zUIU6+u?FA%FLn|Zss#FcN5?lGWdbs?;>b5RqlqC7b|z+6M2pPr<%&}5lZn>>7N#8>
zodM1gG|ztfe|6`{)(abhoWna|)Shn`?<8vR0x{X?)29Re7`nM-{y@3As)^6@+|9=Z
zZ-o39T`UJ1!P=wy)hp3@3J~@rWRhl&yC*!Wh7~h=$EF!X=4j5ZLh(4-bTRstR21qN
zEyH&w7k3wa_ipGe{4Se5<rX4c?Q57%^kPg^w@Y3IQ#F01@1$^iH)su*?|x3v7SmzV
z>gllY4al*BY$D;bM<c2|qmS@m(|G`pz!_)Qp!5dL6ob0Gq=v?4w8QTa(2q`j(LRd)
zbWZOB*q#F0RQ^<`UZE$j6<ID+k2o^pgxa1xw^#6)SmwJOz%?9|TmmGvRl=E6?=49n
zzz(e0iQl|-?Rmxy^7%A{#lYzi5UF~50(>S>*vte_agy?2!ci_2#LUyy5qt(9hztPJ
zzO%oF>WUom-D6<*VK(*cUS(oJpLDb@pQ%;Y!F8Ma0NuBY6|_8&jgrCi@ILUfmzdA*
zK6W!Yc2q8doUeAGSN)X4<+&9yJ!$)5SX?SWc;4n${)J$)Fu8Gwu-r$FEUyF@JEEy&
z_;~pzQ~9jyY*^Wm*wtylI@{Xb@_P#%N>&}BVI@Gh0{*N97j4CXuM<hm=kb}^(CU>w
z;vw>iZxL^RTOn>F>dKf5$d7v6b*Q4P_jl9&qNoYYfcGt8)F@7JRj?n+Z=$N(1F8_s
z<BOQ{9nl>l8upiZ8$l-sg5FVsVYcDc^jiM_F)-#dpM4=bxE9q&v=!^nzw=3jVPs`s
zP>Nb@7q7vPbpsC`@}GGdfw56@+Uoondl*e-0H2})hDGhkoft}-J+;RK<dF%uZTohr
zVqLSmqQ3)*_0q0I1ZyXU9L49>YCM}%h0WKj`HA(0u_Hal3INCODQb8Zil0d*`5b00
z9Oe_Ib5}qO>|pW{mLF^Ux(9l3%zpgGEQ-zC6*!M3^K=5aMoQcg*^k?ja%~<mc24gH
zgMTW(p@7&towfophS{PY*rK8{a34neNb)@F`|x^63Z^&BQxb07npge#2dMUqX8?h$
ze`$#lJHVYCeWCX$U=09n5R-P#Nb<C)RuowN4NNFkG$aH^mh)LDryNWWVeb3sT(?La
zAR8H|(O;qz>2-A!8sh3OTz87Dt`5#O0gOnJ-$UTanGs%`^8@~K_Ct_Mh@(6`=O3j9
z;5rBQf9RP_wS4W0HmFCh4{roVrc~K=_qLA}jCj;7aBKe;nD7mjG!VJwQGTcFP>evy
zz`)?;40M$r8xI1o5mZylSkE(2iRtz6w9f}04}_K-V_z>Iep|O@jbD<bGTLj{Ye!lm
z!Ix2mI@P?6sE{cHj2Q>BQfLWq*mWIwzkmILl=u2@e20I*;>BI<-CF64QS@sH=3nzd
zuF{Fu@t^oxG#88#esN1dM=>k}Dk9^CZJUgZt^1RS;`Gz4850E{cvN{YNHj!AOb)+5
zJQcyR`(AwaYh*^%F?{qI{*OMruc5pQ1bBf%GLYxC$Q}e`kYtB{?WCm#*DWKL!|B3?
zV9xW&G`1m{xvPCsXV*yjm~7nWYO5NRWQMrD&Ipe^!vEN-+z9rX3+6(ukF0VQDg}s%
zHIP_BXYuLy7wGtY^jZrK-6)5A1m{{ge71$M>)YMh1PDy#G|Vy{;!6}#cc6wU{ox=b
zmzgawGRuJK8m{tBg556*mHc=t0^~O8&j6$M*}v&)2`b0qoXqm{#2Y~^I>qjO&2E>y
z2)=ti`Wudl(DP^%fE*HyYa_3t04>2v>b9l<?zNRUYfdBGn#N#!33fZT5N*#bb55sv
zPh)z!>%-C8^(Eos&(HLcP;8{WOh-9_?&~^!02X05Tsz_q2w6Cn$p2u1VjOD5xV|kN
z$SrXHro+Q(FQyHBXni-1fL;B`Z&&vMA!~cSZfw>+FPUX?j%P=7wSU#FXvXfcoBw+z
zZ&6&s!_nwmVLrx5^aSON(N|mFEFa!-6bi}TUaLuiLIM!7s5&T->mzegMHW%g+t7-J
zxA$nidfOg2u_cC+o*{G5WS$&&{(N+j5!g%eg1e@IDM^FBUe$*dw!?-?ySNe$ip;WC
zUfEc+F~hY0d!15_@|o3FFssk0_wl0;VWyZp?$X%Q6wTxL(HB|)Uv0yz|MYWwcI5b{
zZK`c%amFV);gVBFbI?R{sk(cs*K?W1T_GBGBJ|v-_BDl5@KMd*kBDgT_XE#<lKJO{
zDky%r*?EZ*>Xl&jifCE^bWO=%K99Qy9(NM9$m+2&T9M~omDR0s5=Ce+TSi8wWE_MR
z!%$8C`|M$~)|~1*K>C3g8B?WRM>h=e0E*~u`Z_uX4wPG-IR}VKpmSp8_QF`(+>tG%
zPq2@~sP$Ta)apN<@X0+RY42o*y+F)MBWVepfc#HdOV0!{Q>!%yi|8zS$ucuUy2%xs
z25*CA)TDZ&PWkogSE=u(z*$7pQYzyAnMq5F_Rwyb68_kI`yZ@`GM<s+EwT<R=O)>P
zimZxWM*^|cOEaB2Yu4%H*>bP-h(K97oJKBNx8v@5M$y<o18Sg!T^F-jE(Q)ESUblv
zwk;=E(6^3LsN-&H!OR8M<S}YUVzVHJ5?-9i_tGce95Or@vn_nD337nM0iJ4BAG@RY
ze_i!u9ZrSKA*ycrj%w;78zl$HOg+BXELLc$#P)c_Wh-`0EL-($*#5CMH;!+xGC0*y
zYxB|IXR>D6-mTZK_^&^|ee8tKvZK9>rG)?k9$7j4+TwSU_jMhgJ3sHkp1{8!MI~M+
zeQn;~em^e!#omN3OTMOUG94-Zk@e^|LJ`S;7r}CsIINpGer;2BBiuOAJ{XGxU#O_S
z(9Y!8aPhrRqD5k2V(nW!b`AOMV1-gW5#*T#Cu~S%nc}wyAlTWhXlXv!GcP$wLU!FP
z37et8TGO$BZ0p*oyOkyhAPJh#Ji7COd&}6#lVJ}>alms2kHKR3NL|`Lb$e1Lcn<{4
zgXT3z(#FDIdv^5&iyv)}V@t#*_b=X(?XPZQzTPh_J)H=+gDr{N4qNd|yrx>?1t@ij
zH+O_%3dVOhY7v7u0N@OoYkZE-?&`{Kw5T|FcHeS6px;x0)QlYuTI_!)&bDSkND2A(
zLvp9ChhK5fm{=OYM|nu@nkaE6$&9!oKHP(dC<`w91v24RV_zV-x&h3Js9b&eht*n>
z7p%`h9C%Dt_lS!qM#|)LSyW2RoDJ7Wc+T`e9}Oj}lcW3B11upX@-18SvX0=a=gVmc
z^p(t1EUY-z`_qZk;riQFpoNdh>e{-90{iU%H*G9M$Ve3Qer-jXzZpiiRIeyVV9H%t
z*_}*OdU2e*bq+8PYrW@H?-x4L61}eL&#tbTFCaKR(p{3+g&xqOM&whPXW#Ea%YAe1
z$bvEr$$$_L@aG+LMgW%yh?$Ij4*$s{qe+CtZ%^|ENTF4&e{J-sK6w6N)VQx^CL=;@
z>UA^}J>v8+0=@Cp)<Ja+p9waVXVrwB(5-R(v0H8^wXVxN?WIK^lt)uQ^q?_y#QZ56
zZf*9E8F3j0k6vg;*7b&ZY%+2^4AZlQgT1aDMpB8WZ7dER1KT+GMJ}S@;o;Op$}?#~
z$>cYB7yt7D5#n@|6MbYY-vWk+73YQYd`@Iz@Ik6>aT^%g|NW!cDNX|PW&+}X-N3uZ
zK{$_1ZkH&!<LRW>2+p}DH#N59hgL79odMVRzFx%H+4&4eVr=vmIacK$tC~4#*^5SI
z9o6-@^=JzG=yd_N;qq|7NCR-e5-y^kpP3|Yt!3PAsQ9AY*khfE`oLfhv7QSa_Jk@-
zt0cGROChziSvz$mq~HDD<H;*s%`32+xwm@x`1s%rX4}F??=%xgy?7rR#I`JqghTra
zd%H%cpiiGlheyeygu?9Y7LA#`PGr99s_7#cs)-RkIQ|L{G+xb&r?Ooy`P7sQ(4Ic>
zgm}E~`1z7Jp52O#Xb5Y#HVwq~$)tPtTHxAd3U2Q<4uY?e6~xQR%BY~5$JPK_B5rT=
zDYdHoLQB5gyn7`x7Kz87|COL+($dmORkp4uk3W1LRoNgZACiaJp)LC^6C5IO0F6>1
zGIbH{f{8e~&0Gw{m>|^lgKvKtZMq>+M{p{144i+$Ei?2W-K(BG2j^v<{>-@Fk#gd%
zss>@@qAKT(?5?Fnp`+T%G-S@7Dt2>gI8=c~_^-k~+(Y_2IXPPG=T6<_zX4|#K3T1p
zAaW81goN)z`*9N!lh%3?o^F6W3D4y)nEw-D2^Rtb1<eB<WnQeUt)*U3UY!#dhBL?V
zu2$UId@!Ak8Y%DiNy6jht#S9U8V_@;MS}IOiF!Yok`)OUXLB*1<+%XMvur#f$tl;q
zd2389%BtB=<RVI`bzZp>iWZrpc+WrllykpFl6O^XTDx}ieDyH-t*5d~b?F5~B_(3b
z&0(0QdJQI1p6Pr7)A<rikyi)QPn<Z>@_p2i21(Q!jj=km*&8&?wiK`SenqfJ!iMq9
zAM1x7B!Y48M?tw$b_4lnP>~N{`sMWWkIeCVjb%__n~0KRD_|G~{Se2boG=Rz$h`RP
zt*_gtyFVVF&USd38b&x5kp2Is4D-1=Ir?GfVi$)z3AvC}oz-*Q4Q6;|XVMNPVCsSs
zx7pdTpN_R?M?%nmL5;5z(|cjV0gPTb4HQ;m#?>b@fm%lRJ!={yu}yU1?;W2p5%;mA
z<o|nF9WD%z@w|`qpgnR$#4*@Qq-Y!10RsrwAUkKyB)@P31aL@B2+38bi=0!z0EwFx
zBJ7$8w!~9nu|O|9#|tGIB;mbaI7i)GegN%L*NHv9ZUE=kP1)h#pf4F9C;HUfA$0{t
zOxx79ZTpJlYX-)~3YWhk1;BuP=d^$q|5j0sm8DK3kZ%-8)~I}C@TIAyre@v_Lq8W$
z8gt64t3nT}{>P6$dwZmK0=!&yanJywik4gcRAXu*;#KguZcc0^z;j&OK=BZCq1qmc
z;@N(h%rHmod>Gp`4sa9Kmoc448@<5|4pul2NWp2dP{B4c>}RO}o{VElLhS7-xt==J
z3iZ6_MM54KUtyDH&7C`Wx5Alma#oNE6&lvb>EImL3_37$L{5iZN!*YspBFl+yzB{T
zniKENNv#Rpad-r@3lnd<758XC4=Z=-0yZj*=q+T<raezbgYKwk!Y22QW@~w+%33NW
z4L^L<^*;TCT+UC}LO;Q3Q^8yi?Bd}?WlzYo;r6btK7hc7hQ;TCDC+auxrhXWUoaXY
zJ&<ZWjlQ>O5d;_tmZV{e!=a>3){_rwO-(WP^!O_cGo^B@NhaZvs!@SC6B;phW10fz
zEg%-yk~|%H8L_J`MqY-yh(erCZe}z-aHq|g<;RDbWw~WwsB)4lmX1wUb!$ITzNhkA
z{qOz|Tyf6COBNwEexZ-Ma@cF)W5pe7xrh>dwy3v8f#zXwG~7!`X&VEKdDm`C*rj_5
zaSlm|=KfKumYaK<*bovD=6K;}$RPT@)Hj0-2*}?NxH+{1hRNcx1=$ocjDQqWJ&25)
zf*+7qrs3l97oDkzcg3}%4KoE0ED`LO8_pcw3Y4x~R%G=NTB`tw^({qFWWX@7f;bmZ
zEXR2X*rDr7x`xG^YnV&dV0gp0pz1J2*q|1he@If89Gss6Ym=Uy?%Kc46hFeWxQOB>
z<m}XKNkdTnEN?=<iz?7%g0<wQP}#w6@%;`~pssbe*Rzr4k6ExV3v;)RZ)#hX>S5JL
zI=JiV9_vjX37q)*t|KBjE_^ZN%MUm?jJnIz%=iOD>|$l*QG=)ay!9Q0afpTMZbgv1
zjEj&*)LNy*jC{2EBWexAPQmN9TcDaFpbK9dtT1jOx7Z(p&nCZdpv4M3LF-FJ1w}<E
zMA{F($flGM@4|`iIQ|K5ObOaBS%IFc&i7;>A6Mk(q0%$~Pp)*h5*@u2iwq1R>QQ+I
zBezpNR%Tki>rpl7ia%gcUs5w;-&B9wojbvLp8ePfF!tu+vL^uFH5@~Peq;m06!)x3
zI9ygi!KelE6XWBD-IeUWTX95|ITq7?Xu-_}@dbOo?Agu-F-a@;<vC`XV4i1<Qm?dJ
z1YmD#y~F2ox^o(rk>t$rWX4t}qr<h#EPbDSLEd3FAItaCD;gxxGsq8Kn}a@T>D#xF
zzY(j_1R^(9t@gt>${OsBneQ34U`G%X;E(Z+9p8?q7#kRTVm{i9;hRwFqL2QCz)^O)
zjT>r($ISvBz?s(7g@Rx`PsSSgVF^}czMu+ycO+?eg;Ctqce)j~-XDgsuj>x1PEpwb
zqf^^$rRS2-K^^3*Snz8gSo1=wn#-tMY84~_e8CHA`CbpU+JLbKxvMz{F@l1|z$WCC
z)5!C47g11J^?o)&8XvJ>5T{Aw*$rhH=H}+WmJ|#O)tcUkrfBT}AmTLnPJQH-Zhc*S
zr$=A&Sl};Z;T05xy%vL)ZdGI^5js%b71A*^H7&?KjQvnJb@v9^g!n3Wm=^dZj$^sf
zLBM9XMp?QM-Z?~Tx|Pj2B0|H7+Ym)W-IsK4pw1%}KoXmFaU}5qWIS})u>lhuH@g9}
z>1F1(B;%s3ot<~<%;7VorCZu6x)MI(LND=al1u3PZTfj}_15ym1qH;MSG*1qVKGS$
z&CYG|<9ojVSieb*Hk-`M#@j6W)#UIQp2)ZO$P%DZ*JCZ0(iwg$ppP&FT~42nRkRKs
z6(@t4p4kmafN|q>+gwDSPjm@uAdp+rA-<mwps&pKoBtcPHE+l{Ul&oDz0GU=0`4q!
zZS9*n*aAyFRn=J-dhsHb@l;^27ak?}Eat<Ssw2XPdE|nudCeD6rB2Q$52=zmC@Cj4
zD~3LvtyNy8=e&0>08}bAe;dKX{mlszu|N4D_ZhmhOkqn+fValy^QN09bZNk3#)H1D
z2{AD?nATCD1#R;<E9``*w;LK<VUYx+W1k`CTwFwHKtAu~7HDM1s+HhwkT~&y?`F+(
zuDyfAdWRT-QfMbAb|XauzQluBQ_UUM1^rh2x#ACJ-D@ak*_q%|$c|o~$_`kXzGJ3m
z{g9;P-j4=JX}BJRhyrdOit>t5J>MO?*o4h_0AS`lt_PG+9oGKKR{0PLjnq^_=2<=h
zj@Fg4!(7sXB7cETMCkGx*|F+&txSYz_$3xDq99n%%nSWOwK`!)s81B1-X5$%UVqa4
zd<Oy-@b+we(9a^EZ-<n>7IwP+z?|`L=DF&ZwNXNh9V8+E{Z$~E*8<PO2$}Iae1Ay7
ztjyb8nSI?iF)7IzO>1TP9q0Id2m1z?BcCsML8%t@TzY2LN55mcwFig}Tx((yV|Zhg
z<W*6M&9HoTIyS!d9ej~)gAniA7k0wG3$F89l6H~yaafs5K-ybS4mldZrwKZ%L6?vL
zuVZ6-0Iqi5-o4VJ)Hk02et*8d*3jBrsEqytCrs@4|Gb6f+e2YYRJO}xqSc4Pn(HEp
zy-Wx5R(=6l$VUJ<{y+eIh>ntL=Y$`l2lFN3FfqMDV#hTAWBe_8$4OLc>$^T77{Yv9
z`T*i8Q4=y2hk(sAh6A>>de%T&r~1v-+El6+^^Ofr{O334!*#}_K&hvpC&{hn1U{a>
z?N!*n3q=qJdcm}CeG3doXm;3s#;ZQ2I&H&tTGS-|biGI2(!#ZQ?M5UiaQM)~QIqI;
z5SW@4%xlLi!QfjG2lgm7##wa8*qoNzTUUUC?)|=ZZ{Ny+Nhr8)O|aY+;r986xyN0N
zkGDtDT(i;kGb-km2uU7xHPZUIPSg3W=6ZVh2s)aO3v%dZ)@UIjepB$_gXY<X?%@0B
zzs|2T>tYeY2=u)tH{ZZW4#vV(Zh;bMM)M!0xs#KVe8`yt$6T?(1QaZrAN=|A2P~jX
z*TwZp_dK6yF;Pz(WseDBVgQ`n?V2n-cdqKsMCd)6#DwDs=OKoQJn-++iK{l1s1!OM
z>Hnf5-J438HcorKoK>ttd9litNoqJnt(X4$!BK1VEV#!o?8lj+U`4$3mKLjkifBT>
zaJsdk0+k@dlZTpA_k(hB;erKrghk}Mw-3Fy1xryTF6dDC*@d%Ka#L)iMt}R2WY|CS
zAZvTST8m>g4!GhDZ&Nu<U8r3jwXU}qcz^Tmv^Rjuz(UNqc>W<;*tFMaQpeoaA@i>R
z{T;`-WE}c|e3+7az)nRMw4x1p0%GM;p|`>D#SFQRr{<}Gk1gsipCzmox|fuMLj2Rb
z`HA;G6Te*^3-HFh=r?WqogyWCxWNswnP{Fer>(0I<qIq%nbcu<8c~R`Otd*<nEA~|
z+R*^=tU(CtD|Yb1hYthS`EAI&xWD?msa4P@fW^_=i3lQfvEmC?#Z#%;*vW3j#}cn&
zF^adQGUt(0{s0+2=oo}IDT_Ph$suA!nW7xw9rnBaLOXp@dDlO5lgD*!v&#Gq^WN~*
zV*n(a6IJ~nGk4eO>8a~`{%&+B8@+j%;~$5q;rbDKo{g9L=1qA<>v-6pG(Jb~XiN)F
zZDN(MgB5jcBw&@6y4T)mHoD6OyofTqfttjWws+p3b;fUNyQj~4SvfF%>c25+teKy-
zEB$*i%*0((KQL)9scUh#!hG;N6!1q`*9nY8XqBD6(Z>H?S7!*nS=xy5>Z_3XFs#<@
zOAR(|?7%i6JK8tS2$O^_2@`EfEC_6)Awmdoe!4fRLceI4QEL=fD^%NmF3$&GM4TqK
zxUss@`b{9D`v;~IPJB@ls@lO`4d*uDjVJ+GckGB81LhLW9sW8>29H{0+cP_^PVShQ
zeYC2k><NP8k$1gm|FAx;@8DHF1@*ZCRO6FXivnI)bM`pZggPsiy?HYahgiQ_$_Kdj
zeqA~)E-5Q{FL9)Dd?-#>b65%t*?8Q|PlwNh-G`#L2EWYBB{H-CSiN#vKC5^2`qisj
zutr91SN!uc<DDuaDB2J&V6h<77n*8^uU-XqA&f*G1Y8@5DSlASRx+#$QHWmfo{$$%
z0^xF|tswXD1SX0#CcxZe@XBo=_R^ZBFgJM(CzLARoA1I-?Ch|6kRjUUjf?MB@fSr6
zfcL&(@TmJ%6cJ=c7Q^HgtU81^0N(Dy4@ue*9GvkO#+SI!&f<q`;#S;Hjt{*0!(-s2
zlOg*gM|z{DMKuK{pxMn~U_iY3=YuF={-O;-2lIXg!B%Whjv2ck{}SSB-29DnZ$iW&
zMY8#F3k>i}aqCT+__b;;+qo6<C*I@tBl^JXzYM(HoI}_g4F1j~m!;5c#!B${(fznL
z??@2X%h$8c0@*on^8!cIg+Y?+ZKkjRt|bk-3)qExLd6~?D=SO&_${vy2OH8)*Z>cJ
zFdesdSCyIm$5tL>nrnyieg+}!k6e-b2%2-lBZWEP>D~aoOgvP7+hEr5&!0a7#;Nre
z-GvPP@HoR7W$(s@xQGKa<H;(ynkVlB(ER(`$(P10*j_+Gcg{RLoCBp7FviN}oW^F1
zT=u)U@<*v2V0%(Fsk?J!Q*1ZF>ZR~0V)S1ob9nq3(72lxS)1@Pem?@avl7t8klm5q
z8lbHP$~aJ%`x|O4=r;iPav-O;8I*DYH&^<U>fHqU4kEz#(O6fJ1|IczR3MrN?95Hc
zn=av(uINySI7TG#oQywj2!QMHiHTU7z6O_$cA~wlZS9XACQG@(bvX+v-&ai4?QlV*
za=88ple6L|7K5<VYRq74J!l<+OF71WDQmHzWBXM9lOacmw40)3O$1(GJ8a*qwi=Y0
zu0XEp4CICaO+ye0sL`ebF|(9=_q<HbZn%GuSb6g%SwhYFmfwZz?qKmnv>Eom8p$1d
z!%HSws%aJmgfgAZpR*M+5)vFxf8u+C9K8XN)yYuPT^OHE+TPSw_5?zKlGc_WA#9fl
zrQdLo|2Now|Nfn@AR%5p<LAd*y}BGx^Y2{DApx-f6L&Z3Qg`GZ({t|@i2DVck%q3+
zi#1=rUOjGQR)yXYteuogKSo$AWqQk9{<rkY&(B9HX*2a8c;G<KwOQpL%0XK?OIxM@
zJV+C@Y@Sogq=_e6WY7ll>mZ{u@hgxuVlD)ur+E=^!aqe}3t9<x#f<=4S7JMJ-r_nG
z@?OqLq?lO+xd5&V8dDrWaCGn!*Jfv3Kgbi^UBne+jP)f8E<0@BJ~&>i<B;N>$HNq#
z)DOWAPnw0HstbIsilxY6HxYVc{#XF6ZSC#X!WJ%RMyd5leWb7*&=07qm?-lk$rNU)
zZyE;0uC9CsWHdd0zaqh6FixZCItId5c6-Z-ef{x6?(qMRRmd*dcSTV*2UPU)5;Fjw
zK%+6I(|IZr|F1vIi#Rf#BjR`i4?TXt1&5tGhcd4K?a`Uwi4V!PXJGQPX%D|rL}Se7
ziRu+M6keH!@7=fW{f7_3#TU(M`{oUn8Dp9AFRYDu_G3{ULdZTSKe`%-vN5X#f$dDY
zxM)pXaA7w3+wa&(o9O%R^M8L4w110D6n1oj=m3kivJd}7e{s|PPhR~FT49ojHc_!W
zn8+A$gC?cemjS6?nSaIThOup7?FGipqGG|8)AfM2TzeG;$`z0eUjFV%tff@V+rq*W
zV0T)PnQJY7!?+GR51|=TLsq@Qh9R_QV|ltKE=zU<4csdO-ku&iR~U!os5rtY%`qEj
z&=_5K3^zMI*O&Ph0|Rgu!Io?qt4EaNWr-XSL#Id!i!D~45m#a*hVUG$&lu7>8uly@
zYK0Xa6ffrLu83IjJsSPN6_HIS|I`}77+Ua!fYBCYDLY+WDKd-hHb2~tVLZ^Vs4~t1
zeTFEW!ex&jg(KYL^-sSsDg}kBsL*Q%Z;ek#R|SId^_bOWC-N)Y*lx8RAJU`NLslw>
zJJ{OJl$FizbjG&F_UtTHVPP~+DY7j|DUuNFU4QYL%FgQ-wElDWOhY{_(+W^rC<RU|
z2_x++3x3tGK*0dfNL~Q@qeD8an;nt|3LDJ#mt9p|vIjwWYd<RRF>zw|E70P+9`YNJ
zJbuaK3cWB4T*%Gu!r0OVcieXe0R6oC?zo<rsbaGm#Ujqe^jkk~uDfl8Ru~i;2&H^Z
zAl_p=F40jnQB!wV7&Ud@#J;_NRsa@d<R!V6-rvJ^Uf=s2GZCm_c2w@mJ4__Me@`jE
z>?J6+G&eV6Xfm(!;1wjY=XInfGCu+;_+}c(Q#lA&t{eOY_03ouRlUQNyLX*H7Axn}
zsoMT&cqqmJ*%kbkV$toyMGLx}_>e+^XKsBN*mvn3v}D|Y@9weY)<ii%P~9uX^Uy_s
z4RPJ4cpG~IgBv>$x+csgspU^o-BRGZ|7~M+6BLe#T_qWy6HOZ>;aoBPwZXz7YgdIz
zAuk5Wq<tipv65z`UNE?CKI5zunFj0}S=qBPZCD_IB@jm8zz8)F+KcO5tB!&={n-P5
z{HrpW;+-Nm<`&{&M>s2V;b2Q(L20_F=Tc~@ojPxm7MNc`a~+C)G2I&=reOch*KU92
z+v5#Y>i>BI{5O>R=_)`}!b>2eYoLPPX{7Z2o-YHMo?uH`SxHGOPeasbPD~9qW?)o(
z_m%Q2_>6~T8i*|kDv%G&e1hs;@7J7!iCNHTnm>2L|1tLE;Z&~g_ogBhDh(u4jx>{G
z$drl(g^-evgj7f>GVFv#rA{eQkp|KrsU))~q#{(xOvW-M^WMMp>_Pi{&OYDkcdn~{
z&Q<$;-)Fdob+3EfF9>iy-df|j^;4%}r8g26M_O_c-bj8cDk`e1eE8>zBoV`2V86s<
zTMQWty!mWZI!xkEbjIZdcU@VqHF^$qMS|UyRP4vl&)h<@(SZGU)SWxu0Uoy?rV3Nr
zUk610Pr%bZ79V1WF|#Km<b|cQHJ@(!^}9016;#TwmJ!H4T-X?+B8P^z*=#rP@8w^=
zeFGD2s`6Gy^z3zXjM0<f8;dx4Dyk3be#cD7x5vn87cfx%94;RMeU(5O!YyxYw?EC>
z=}vQ5%BlD7-=on~#leGyT&P7KB%s4=7VEDVqC<@O@dOJp)DYp}1cQIa(SM#HK}2G9
z<EoyxCW8WS;O5iTX%U!<7!v-)1|lSuXqm~s$96yRRl0i0R>@)<R4gp+3Zik1Rt`?V
zGaQrRSrpuzOD^TxADg$jqmpo1ywGf85-<p6_9FzB^I=KU%h)VQvOM#af4B(^Axv7l
zi9@xbrM=zyiTzr{7FZ~(JFYxRP!^*>p#Yg-aQD=dIr<wyP#9r6%nEJjN~VJtn9>Gm
zDY?(=i^R{&>?c@~ra!VsMF5pB!p{5Gi8=ab#e2d0RTwoSd<4EO!E~2fjjy4LBOE81
z=0T?r$W1;KD{DdL5VZX38LCjeSEDZcMRZrm6G{+7i}mt<x<ze;$TC|9O;9gsUEIE~
zMY!LwH@lFA2FzRSTCu@?NjQY)at#|JkF?I+fI4x5E&Bz0K(2W3HiRQYJp<dZ<0~ey
zBgH@v{v_kZ3}|+&T)7fyW-Gio2nfI;1^so6$;KchobPOv3T2U~b;UCePq}9{&q11&
z*5Mm}|Ni<ZWLMbjyZRocKVSJ7{#^W}EZ_0HNXXEATYU6Jn6@t=?1ds9V8DdN0({vB
z4+C3Xwmmu{E~=`k%5Q%9mY-@d05hNd&M=c<aEUzdGJi*~Uy!#XB`2hs!;mz)34=AZ
zodnXcu_UkxH~dsq1}3gS3j-r0pr`SMqasKKcEEu|MBmLJQ#p+9sI`Jz0R7}b5h}ui
zPkyKty=O-}spp;vr&AD4LJvJN6R|&(dLF%ap$3asvT^ytgai||Cx}GVz?AH`s_JT8
z>f?zD;)>9^W>B-fK81XqT-<XAkbi}7{SCg7JCXWe#k%fR-rhq{T(XnY37NJ0d19g|
z0`}c5F32tu`G2hf4u~xd6#`=e98eiLbW1?hyWq?$cV3bT=N`3gzek6>->+LzxrUWQ
zKOzlukOp`5-A^Z)Dk#^9{UCE4u=^6{>5w0S#3wq}fu?VTmGq~W^osD5F`=kq^vjNd
zy2;L3Fqg+-TRc@0fZ<}<&A;Fu!T|sKK|vt)>ok@hV3qB3g$BV;U|tq9EIfu3+?~<6
zHrZr!K?@XT&Wa1@puUYIb3)DG@L^J1a`;F_^EzwdKE7egIsaM=Y!|+Lj>F{Td%rw}
zfm9HMq@x#*HcS_|Fw(M0_CoB#hrKqF1DPm>f+D#u(-vDxs7yq-BoHTR(B@JX?s(%=
zcmmSmaPW7;!LZBkVmDpWnXbL(3)8g^yo50YYzJ%P0C)+az3c7(ZfRwR(4uO%1Icst
zrFD0r8#95QPry^YkJ1UkA7x;&z)O%(@<-U8GFvG_yg7=lOBpDVr%G67P(i81MS(Ly
z`1HVo6`&Y%k*i3WE}VIMFq3%<_QPFuKa#FD6UyR~8^srrO<&Ev=~u{=H{A8-%KNMY
zx#`X(z=d{nC*qC*FLG#eE!Ilq?IA0Lxa`*J3<7cm%*(xixb&whWqD)R&gfQk&(0b|
zURG}bhkF&Xnc{(6xb$c=R5H5$$G`i0LsujCMhL-(g;95J0IrXd%!0~DvT&DFh9~-w
z%COnobHgrnNE4L;eic#uFmmbd&+K>pFhPa-55Ldo&s|LBd?>(Y{f5jO<B|X@zx3y)
zNjdv2{)@ff3Yllz^z4*Vq*R4)CN7M?C*IyEez=L<>`-}kF*xHhda~8S4FN+Vsg*YQ
z*x~`J2v!Ex3#2H1>Qu`6v7;Ywjy#2Aw9Zw_QM|)Flsh~Hr-{r*D(ut@UAKD~4sQs%
zGm&Bz<{j36<@DwK`;CZ9<z6GauBSckPa@*x@1L-ePH6&P>~Kbb^6D@0?YJFlKr8^2
zH7tC9K`c<gAn3o@Yps9B5YY&5>GYowHaje1_jQPo5L=3v5j%l_ttFp6-5otw=cki%
zwXgD}igV#|B%AmMP_9toeH0gWdfrai3nYs<D2~Ul=LmmiklvNBq<5!QMZ*jt)EXeQ
zO!n})KuXGh1#72*lNPue@rns=6`T~#?yK~O7Xaby8XG{PyuOTV#UnPO*lnS4{f4HO
zdtv4wbaY{<j1?eh(1o$t8!@X_*XSAuPvG>P`Z#*~79rKiCR3<XzM><_2V0(+atgFL
z)O%N#5f(5g;iV3+@KtPOigng|K)69{>{X_T22FTgx%J_O@}cWyf!~CDWP+sTi}Z9q
zsCUGjVCA^UB#Pf$nn**>^H}o}uTmnyRQ7Bx+&c_Rvd9(>&}Mfb>D$0O8NS~Llpj{t
zz5_#^uc4qUdm#@w$jg_5pbwd9j9x#2Q$7X;b|mkV3!-J?gKEJkr$~j(LCz1UL=x0Y
z6mg=Y*v#P_E_9(PJ9q-Nd5L+2_=MXAn;%-LU{nus5R8LK0(3b`k6FEHAYt_uT!4=<
zj>!O1R+Qb0IXnfxDTEqNiXEplQK&KPJlKq<Y?~pN2?!5_vYm}dh~61!yY3OW>l+sX
zgbCYuE+634mz^Oj^lkKO2weFX)6<02^FS*TiWAaGi#SO$?SRl72w}zBu~zJ*_H{cv
zfk4$sDbwnvXOJ1+LMzv=e~ad*k4QU}F!l^SOySR$kTU5OCCbtr*qz$an+BtYHf|f7
zkF)bfOZ+~SeJ`?JpqYK=4W&63*9KVO=f9EikNWpZxEMmisxi!ndUT`dM1)Uh?TFTY
zfm+Ih7$GebNYQMyQ@<Ky=s`gk^ENhPK6eHSdHasEn(4^ZHDP}MGe9yHFvEnI%NCHV
zleHbcW|NN~x>u#0hrko7EtMG>J+)IG-oD)c^erWwY{*OY4arA}@i#K`nIH*jV9OiU
zq#}+15-DCZ1+9{V_x0yy(0NuWXzq+n_zC{>t9vVdAD*H#b$Jb{zNKbZq;#|tvDoPA
z{F7<ap`G$CM=!SeWLUKImlvQe_yH&bq=0cFz&ZJIBYcSt_$QUmR#Q|GM<5cOeQ)Ju
zCM)665R;Yg@8_(IDM35Mx$rh2AWP`HMoBbp($^=u)?>j!sGT|BEmeEgPy7=sq$42V
zYeSJJ<m#tK)@d5oH0%$k7MGQgUbL`*--Nx`q5w>Wi_PBN>3oW^+~v!!gAak;Os=5B
zR5Ub^DlFojHFA@&k?>r9u3+;_a%f#pgNilziMqAq##v92BRnD7U?=~Lv`OROnqbYi
z*%=lqF!FK#aLIRa$^EDBs9S0*U*6W%hGDyxMtXAq=OB-tyhomOF`9YFGcdpCQLM~Q
zoWbmSuMSsj`0@2w8(6z1<IIubq7rlnyDrFHC=xVBxw`Pyca*`2SmvfY)Bzvy($*#e
z&_)K0lmopR_eTwT(R~<p+cHuxb_B;}z^%3OI+LfMK<ZzotjPoJvL7p=;70EG>s01F
zkF_1=|LZu;=gsJ7Srl;Wmu$7LNP_m@pIxEzUE5%o&s}{-L`ZRjsV;U1_<^$o53IIF
zxuJMH7A%W>!MMHAqhD{c00v}D=&`@}KCTB(1MEW!E30c*F^t6egq-ourx50`?a1Kg
zjB{5DB>O2rrYHJSPN9aq)WIhW^X<=Kh??jow$MlPIqDipRMj*B#cc{h#m)S{fgUja
zv=~^NOU0ibck@8lo^^!q{TmaA(!JWjli2Z*-p}=vrVn7usZ5sU+q8x)64~^Hm7aIA
z((^HEP#6K>;`s67C`-ejfYv%~p`}NSjoD~$rDjS03MhJ?5q}Mw_Jrw)*pyQq8d~K1
zUg$*^k;f!F$(H>|olI|jHyt)4PE=avd^%Rn2E>uyq+S=MSPQ(^WYmGE%_EN>YH-dI
zyx~}}P<VvRNUMd?#a3iv4ok$gleOBywpM{P5Z^x|HVkz1^=o}|!LuT3Y>tcu(>eaM
zMQGqAG!^7JjGA4I0DZ-<Lbn_qCsiKWrgl)Rp)v=-9)^SU#|{F;Q&CmjBZIs653&#+
z$ahfBQ8K!8V&cq3Uw0>f&08ED9g&0rnL=R#g%HC#oBVD~XW2)Ial}3Xi9dHWoe*Jb
zOKU*`MquMz=rM@2yHp1!YX!TFm$tOD1dm~-d?uOdekTe@*BLMtexVUUqG$@?sG!*U
zCk7P?K?wJNUaoCFg5hj|rYBshH;|@O6gT43;33KR{nt~}BoQC<zsbL6wk|YV2Jk5e
z{Qk7Bq=N-f0}?F&36MLmKsv=(vZxsP#$XDgbF&6aF)rgzoM23ha*!Q)YeU`CgHln@
zSGI_H#l3s?PKEQL$&biV#ff5)CG{Da15rQ)f8j^`9MsT~-M_E<a}y=WyxnJDB*pG0
zirzCe&PoR{1@)60;b>Au^#TV~Zyy^d9z|LMpFCOOAg&4C3Mr}yqTg{53HUy!)dzEH
z!3-ttE|LH9hy`Thzmj5Y)Ik($u^nwF)Avzm({(OUL5Ye#fM@Vvbmb`1rYIG|A-i1v
zp1#oXHYSqu<WFkxczN^YO>`^4o&ZhfrJbI=m}MFBh*UKhf3$kT);We@BKQP9)dbJ3
zy0F?FkRsyu;pem2f1axnW8ajNK?<U?^Trqn$i*Pz-i<6mYw*<+Ovb%S+ro+wGY<nw
z*1K#dFRX6!@X;fo7y+9~LKGsj1m{MKARC4mYXyJl_G7jH>n~lnb?AF3#b1Dz`c__k
z&p2p<fE+(=*}qY1)_zNqVlXVhla)diMj9=*-A;FwRptO6w;TH&n^>kuEF!zx4MP)Z
zv|nz~3Pb^>ce3SycW7v6w6?X?(|CI)A{N9^`*Q)$<20QIuS?SXfrdwOk=tpvP8T~c
z5|l1BVM&oBI?cdW)UbcW7J3)O@&+axrzidE4351mc9Hi;-LdhD`q2RLCxJ>QU1sH|
zzYD7y&q94wZMB-G>V6Cq$TbWjU1OC%{{xHIj%ya`bHtsCO%i6h^S}7?GV<tE9NaiW
z4t9va2^NRx>!T*IpT0~WY8gy4D{1kNUfth>EbWaBRmtJGrhCJhfL(}o&$Go?<F+E3
zMLJd(<q69eC&#C!^Y!JyJqQ^%N7QN%*AXZO4JKOu9&#F?+XCT+s7)&=nSRIYBd4kV
z`|}l$fsv%~sE4k*hv7aWd(hMJ52D9tRCSc>ub8sKs);6~S$54NDn{jKIfVH@>!cY(
zJGpD@Z*3)ikz8l7>CEdah{d5{#USpZfy5f2h#J%lLBT?Ufw?d<zw|$k+a{Cp>*M-c
z^07c8WXx<4s-V(%cAnn;{apr9zPF8VAKci&ognC&0htL)dIOzaWZs(;WJCtH{~u9w
z7P%2b<vX@~w{#y5EYm45>1r_1uL1=c7snPHsMWyl1XDayq2ypbnMcL|Rhv4TnfIfB
zTUFXN#t9|J9tc!;dFb<;6~BgLdg^dU23)-T5;1ds5M(hrX!}K2ycKM@m>@s7%U$v3
zR)$6LuNIJtM`(s?hnXmj6fGqnJ&3Ll5iq%uA_9SAZd5sj+1%2<i;C>LekmG;Yn>Jo
zS6#Za84!cqQ??zbjRNTnRgAEfE>t>~FJC@&oj_nk7v{M@41Bk)QE6$Z56tX31plMh
zry99rBfnyF&Wc}Nwrtt7-yhNewuNo&Lh$&p<*5)XyWXBZrcN~d3Ba*_n<{Nclam#S
z0gknJ>yz8iEQF~;QwG2za8B;uAS_a}+2Kjf7|lw)0%fP^Ds#6(_J*5Q1GGAXc5ufS
z$}+TW&}lDS`xuQ6wQWQqCQQs-_JX25I7j@DRKJ4>YX%fgJ5Z%b>qO{AI>(a^79Kr@
z=8eF;Xh|?p{%AJOJX5D1ExUG&uLyao7gVpw4s3BiCkQWe_k6JdL+K!;zVc?EZr%!x
z(enR-kxvl!#B#TwopF3`jLW`#X9s5I=G%8e^YG(LSUUDj)NH`dkBRwjywhcFvHi|E
zJ?VzK3}g+d`YwvrevlFuZhZnL8hb+|R-REF7bSA$_Vj|R6DDLxEi8>3JU-LqWcawx
zWo5f3V7JKx5Ef&r_8Z+lJH8GYNs8c=;oFq?Dc*)syn6b?=2K_`ir=2Hxh0SV4r(G8
z&%3;6ySR!dQTQL<ZQuuz5A=g}`dxSr8ty$MTr&24)F4+Br;%kT2?s-UO9v#NAnf*)
zH1MN9!esb=cxlord7T9s1d-OW^6u|%Z^=ia{(g7`HIJ}5!g+NtvVjI=-CO^9%1URj
z-5)#%7F%ExvFss`cCsD2&>)M98v4rx)FX8u(C$HVga!+c5+v^P36!Dq-|02Ii?Hz2
z=|EIz_d3Jz+WyysP4_Z_D^NFg91(?P!OthVfzQ6qSVTCE=4){&K37zbt03Hb3N8ty
z#CfA1F@}CtiD;wBm^9$%f<I1FCb?FF4<M=BuVs9eSEi_v?|=Ji^@0ie*CF+qq5S6u
zTW^u=PN(zRj5lw-jG7z-2UvkJ>odBqlFvYML+qqxlwvJxb$*~|FRI*~R(E>7&~e?v
z=;Df1lVNk8f=Y<U2`A=wA<T8#j~`|*eBIg54>EzuDS;1mWCx-qZ7{m+d4z|QM{l(q
zMyJMRnUULQnOL}Zame5f3@Ca5F<A1LKb@Z7or%_geEt&nf1^nkT)7!mOS7BM`)4ST
z<Y<*F9+rZEA$O**$Ze#9MCRLz{X0t=da%rkZb$xtbK74Sb{NR3O%SIqscwN$I!iIB
zAVo7Q)&<U}Gixg4c^hJ-z|7WiZdAuG&tn5hXqinzh!`9B3*tDSU$)DSHp(%2V(0qA
zAnU*wt+1vXBx*eoUxIY=mJ^N!2(pS%bKoEe$9Dec#(R;|yVbzJp!DehFqX&k?A(YE
zC=QQOP@axI!Raf?f!Yz+w6Jgh-Q}D9!fpWa;fOtt?mA&kK0$?hjhv8rs|4zvJBEYl
zW&Tx%QRe5Tim`1-moS<OJ)hq>f9H*)_X6s|F360=kmb4s^=VXpa)-tGTxT*7=Z!2F
zvVt_TJp!t7e&u^iE}Wie^trrT9kQumTiZ!nIx(PZtHo-Ll>sK$h-nCC!ZnHiMqQuy
z?;G%vN$>v=Cy(y?73{nJawuR6saI)sQWhVLq<1FHu0V@HU8YXG3lFJG(GxwPCkw~$
zIBpXxM-W4<d$<nUbvONYuuS_ky7ff3DRZKf_VD?{cW^I0v|=zdH9;GVt^LU;^_Ngy
zkQ*;!OziwAS|mEs)xK?NzCeVvsN=ez`uTt?8W}S7?uJ$h{74;o6!xOvHEL7bN!P{;
zR3jwZ<IrZaB;3%N<^<QQwDTL$0^p}9v2Dx#^UNpQqhET#>whkN3gWF*Ba+Wof{;vF
zYPM~Q_&E9{rsPY=3|AA=cOXX(k0gVFa_02uXuqd`tmE%KW8W2NMqg=>DKjn)If2nz
z*vNav_>>g<b0lAAO+$(VcJ-!vm;O9vIe84eb)oZ(WF#4yBRBCmg&z)Pf_9lcc%pIx
z7pwa{X9d&aDh?yNISg89q1qsHQjKwv1z({pP?5p$K(q%qe$Ki%EZQ`kL3tQCy1|Ae
zs*s|L(J6&i!%q3909eu8HP|Y3NR@lEznj1OGLL`a!k||;zI{1{x6$~xbLXvj<NS?h
z;LYZRYEb*@|BykU6J}<_I?Gfr#ra$i1e@$E1TX1(y-{g-xmWAGnAgFMHmGLD(|R~8
znwuwRZ=}`y_(uW)1E&;4Aasl{o50>c<!q!E(~NqhX#EDGRxFBGF@H=aY>OTnM${U#
zsjCpkE18yKqDAXWT58HT67kVPIkh}SkEz6SKvclpemgBaD<Ux9f;bwz-m^vt)1!M*
zw3zHkrANsIlUr{|^FxQ<HSEwtu_bb?bHvdr0K_t*E1vxr2QZp_vo8=-uFht>4)nlD
zCg2Z5kz+tTx@OvBQltxb1ttgH!b>(*h8%#T^=m`BrVo*{pOoB6crT_h0keHw^c>_P
zl1Wp{*izJgNUxe6cE=Ma!=mJdza>X%Pk~AB2#pt`o~3y`JWEYGK$X~2-{J9l+8l|T
zFu1U#A*LoK#ei;pQr>cUlrbogKpYB}*MfZlyl3tu5rHwg!TKQc?&t+O7VBg5xK$Iv
zZW1U&vHQQ{ob@#G1GgZI)bMm|L>4E8r*k^gd1)jm?+C;sOPe_4IYAJ<M?%ajZ6o@-
zvuW<1?H5KB%HO-;f<g4yP!fSjjz|**K=U9N0E=6bJd@c>3~O&p_5&SyGvPiU&|e@)
z-p516{2;tD+IGiks4SCg{*Ql2D}aBYr(P=2d!3x}u0%$IsScirQvD}V9~?>Hbd=4W
zCM5wfd{cP;%?Ra3<v!1{-s^Mx{+4j(JwEeJxpBgUXJb!KO?)=#-s`VnRl&D^o|E|b
ze(9l@196*8_a8Z=@x0}%#2&}1@1K__pVR1ah*;!o+8`VC_O_77ips}FC2og__y=BV
z_z{rydqDe(0qwxgWNPve^X?wLi*6hJ;y;(@x%)--gr@fl_~@wj_RgL?8%`lxbD5G-
zae4Vxi2Kr7S0QcJl$4q|(^QM)<Qz`TKOo}HEmhK4+2NP8<wo7j6Ow{XV3$81JG|YS
z9&l)IaoM`czIiGtpkA^mY%a3aMf2y6Z?>$4jleO_PCblJ-)X7FL*W~w6Y+PX@#t3B
zE_;D^qYqjEMrGl{_cI3XM|%dMzi_j;`52$(fy_+ZOq(Qz`Qhc_`YrEn2yqzYg+XFZ
z!3I5gCJSsC0*1rJW!Ip!iD=_@+g#71ju$(t58lo_4;C(0xWJ-Ox&dhMGV^WSIO+3G
z8}xs?k}g_gRS5vUlk-CuO-4LNKErzi1OyNtIwLqZIH$e~n8k3?9FS^0m6wNuu{wHT
z!|Mtp0trd+uJandpL=ZP4}Yd1-~#u^uvb88u;|3`6DI;Pj-}BJ&@3*l$Ep{0S_%eU
z+da4N0S!Y3hLDl)x{M){e2IndOa*~Aslxs3u@(}aP$(Irc1Ury72>FG5UVhInCG<A
zvSzWLot<p|^g~4E*Qa{1oh#jC^K)_i*w7FLT^dJALFIPRM$#EtLwJs%fw}^rxI<*5
za)!pFQ>*L}z982?KE~uz2<L^W3nU`}G%7n$fhm>#8T~fHb<??hOGD7MHf73`j-k~*
z=pLw<i|c6GDT$Iz(hY{|cf3YBB6ye!qe4)ogJj4Kl5Ln_$n?}Rxj;I5s&Dk?8E8Mj
zUW}fDngF+5LV7N?wzhFX_TbFPWOSFr7=11+<?}FxN^bav&j}n(K{=6d74~gj$^GyL
zM!xOv@v%b6ArO(Co{mbbOO^*pDH7`Lt41x|3&lr>p-Hthu8N=};|D@@(B{g0wuH>c
z_41p==|KNQ`<^j;r6vN&XqGSup+$g$pi2z=xqQk&yW;58s3U)R$4bv$aFJ(B6c#or
z^Mq+o_;SU!x#%W>f_v6pu$9!vy)GPfCb4P0&A_@$l^UEZG&&I*S4)@~#m!H(J&8&K
znbG`b?-!D10KLCwc@El%3%^`Rg8@G3(1P~tpTilKI32GW=pcN6B2?7F?KU<#OO`|<
z_z3(2nsysPJXu;9$@iU>4njKSPZH?mzDo#$6LnlO5^zuyZ8|~9N1foV^F_cmH`EWv
z)8ZXCeT*{-=lE9c!$%3Zhd>L3?{8?$40!{w{4m;YP=q!Be+FL7atoP5coM|HTg<M%
z=J9!kvxQo~cHMwqQW!82(>vxThhz4fjA!dV9OH^H_2eukt5$i|-;HpwRGfv%xh8a;
zN7qZFFjPJ2`SahYTCV}A-~kjsz*0riiPUILPR`TxC{8Q(oSyZG6xBU~XtRU_$<Xu`
zUCc+G|8`Nz0<72?xsO#<+8P>hM_6XYFgH})wzsv+RAt8apLbwqQjm6YX9V9*N;>?~
zrlgAQu7$X`zC`3HK}+lXQ_;puG?#%4$n%I=#kll5TS4GR*zy2~1MTN2<mISltr$JR
zYKtuzk0SWlP_|uS=FIuAR|cVl#P>^&=!%Z-zxtb=I(cw$Wmc^^w|lOL``tT=s(#%U
zBl)U_e2HrFS^++HGS&>jo?}XiKYr{+*VEk5yor(%s$PSFWn^5u+R~+FYVikbZKG0C
zed4ZIZr$qDkauBx{}Yanz;+T*cenZ^Py{+_U%^)LsJD1%Tv)RJJVC7051rfP8B4Ua
zAN^S?Y&jWZ6XQHja`~YLZq`lH(>sHXQky7)Mzogf^aq^wBpkC^faA$V)u(|w@Yy#=
zw?-J+!?DT9UY1gt`w<+q6B-@sGUn~yCDi}JQD?dN@-7IhEjT}Y;N$wmW1ebRJhJiv
zv1I&E0jJNoasz5qHErq48_!2wgH9(9o^QZKB1c4q-r4zh^mNq}NOay=?7KV8x(>y+
z@wHRWg@lm4$*86Qv;$amPUqyo)PIM@`}`HvvEAC*T6*E2mytkVojZyEK>gOp{mr@~
zqezMW`2)Lz1ZKP=W2u5V1;Ij&9Q#xQZOMa%@&~ePDmb02FJMy<ah;-(N9Z{c6End9
z_Kb&hA%&^&cgbBjf}m&A;7dQX3g=Cy_Qo5P=@15$cd<=XXQvzZd}i^FZ{M~@qmNWW
z+c$dfRK(52<vm6HF0?_yFK1VLgXBG0(lZxUllkSijWwxA2bP3ep|^;7e+qb3U7o{0
zJ{cFDj)=-MVA*hX4a+p4jg#KIn2gjO@8qg~t5~?dm&1jxjs5~xtt*_0bQr;y1)P3Z
z0XlkToz+jISvF>95vt#f8<NDSu;4y{nYb;Un%d^yIwAP>@<seU)_I12+Cj8D3HLAL
z;KFL7(UFVh9qvAFM7aF?`~YA1crYrj9;{1i#aReGA-8Lcl#|8R^zVcN>!<c*=$OCi
zgxF?ouJ1mQZSoF=qe9!sPn*W!oVg`PU=lre{ZV4+s9M2J`r9IZzIE#sir2U^XqEv7
z_tdFztj{4_M*&`Z5-^0laW1t+xY=j=vk9%eD1R8kDT(TbFm*epPm&dMDs9q4dWN`y
zr(=(;E$Mf>{)yj+XcnP)iGlAD7A;Sq_oG5?LOzRWo#ICus?OLF@9g74rp$3`6;g0P
zZ(MM-=V14zn;`YTI3FI%I^f_Zgrzp)#N`>_?tn(uV{`rFD%pfBKhV}O6<7j;O)Fw6
ziGxj4<zt*Dzpn2((;G*HNZzt2?}2<Hza<y*3-eT;aB%%++CV46BwhihU<d{?J-@Wn
zt5>gJIUhV&0LzK=5EVkk(&YmBJ~~hU<7SaDSZ4M!3xhvuqEm09u7{$$N1O#lypws5
z;e*zk_Nf>xd<U`)A0ZNN#5Ff`fc+2l&=$p6D0A!>@{wYk1nIy)UFmDMMMsVtx$8WE
z3=YdsSllwRx?f*6rCVqC1WhrZ`YngMDwst&mWw1IPYvPcc;wT%RC#%Murh!VV&!5%
zZF^I{bLUQ$qbZz&Eh3%36?+vLTA(=^vP#x8?RM^&>5wk+8JlsPgYgtShL{Rt2Sd4&
zyg>Fa(AP&Q&3E@Z7>UJB9x$H?D0raPbv(0yU(pr8`49Y$IKLX`^nFjJXrYIWdtpI`
zZ-=+?&`%Dh?4~YAA@{NVQAU2?&XcvWItr=e7GP=xl!M^n$cdi?O>Pq_#`Ma>p1?9+
zixcZl{^09!&c@Kl$Vm4<_iU+wJSepl`WbMD$rL0|<R0e~idiqz-$@zfFa<BTkkE8_
zVycZM_8!>*TFXElV5Kpqug*V^2^s6V&b%n@y!$>MsW&30iB9VY1w@&;frG)h>4s7(
z>I6q!yufI$sfj`&of+hXF;irO`IwRf<egNi-{>Q*;1OunFq@bOEXi+?O+E>ekGk0J
z+*!YOSav$6?@-_umV6KwC)gbBiotHgZUz8zu}rXA4Xb%!joh>|*R_2uS&L-{lj#QE
zeeWd&LT3?)SCp16CB0HdLuPffwMCrm*u&vc1~sCIWPEzokP+>d5xM(PIp9^;C2WBs
zqkfx$tQ{E?3kQs9#+KW|TwhP>NP;<4_nt^5V)1I6CzK4wmZ7izw*GoIk-?LLalHq@
z6~t-h0|oHWg-JmSS9z=jdNqJGPqGRNs}W*Iw;^~|cYPC&@s2LmK>qkq3^hCZ=EvLR
zpq4D~-NR|EbYXH)!Csx2TM4KpI3$Ge9D0ypsG_VDzxvFygB~7*0hB<7b@3O0L3Klv
zk%zoX+oS4GH2{aE#xXvP%8}!mHK;ZGI$>$9`#7yx3zQzlr@t6_B}FyHuGs2B3xJuy
zqIe2^wJZxhtA+W>8$>?iUZrp2<uxGT(?=V$q3zL4M97T2I*`~`$?0s_R-#bAz5Ump
zc)#!xS!y!x0DF(bw%F+xs4{frJR)g$J{da$!3>8u{p!7x=X*7-3-z~eCGtcM(&37U
zPyI(CuCOJ>C^8*goV?;K0KXXNK9BNuC|7iO64kE(S5iboG;1h7oAot2$Z9z|IfcdL
z_CcBUeQQz`iq*OSU8a$PrF(T!I9Zh^m>+@@Gv;XtZoD;dV&T0AA*+Fqjla7SX%?yz
zES5nO7x{{PK*!@6tB4whclBX6H@AWA@20+OARhqZeD^z%Lx`d0?~PIyuS+WkvZNME
zHpKyXru`Bx)cC*C^H;)gGBBWPsG*ZrK>-$WE6%0q!QVo&G?sc-_wBJ6veY&0!iU`m
zaO4z}@m$#$!>g^SaUbW>&CM9er0wfnw1fNh?7=ABe2O>b29mCg(XVkMw*?~^cgNkZ
zrch`Ld`r-Dws7((!(Lz5ZO7hQi{#5WT$tMBZZ*1qS~68CfgF;$ec|6^1|>F^p0lQc
zdCxNRqC9Z-4{-L(Rb3H)CH~q#Xzu(TlcX?QA=&E?r=1rhDn*^MDlssVLMj<yf3Zj6
zxTHU9#H6I?>W?^CHFaduUm{nDasG}f>ds~9_3?ha_iKb7X(9Kn(B!!5L5A`r6%`Ub
zEBctxeqs~)@3Uy48K1!rUhbXoa1=-IA*)o08jq<;7eQd^tI;g{qni4SGP6h>vEPkE
zj+<e3Vci|fB9koISq<nzhGusFUm^v*0daf77IOe4Krj+`3Tww-D8TbRA){LEX?PL}
zRWK!m6iK+XUPZXP+`TBWl7r_xod=X^y3g`H1D(MO@Vl9NL5$z)?d>f$nQZvoyXZC(
zWI4n>@__WA%8rd?DG^<-D@mt0f;7Y4pWrZHXoJp`51f+w-0#S$C%m|_&DuKsa0mU0
zsGlF{`#mMHKj&D>3lt_1ksoCFk;cF0AJE+_dm$`tPhUwS)x#mf0xfR37dsbI>lQcm
zmvgcvFRr0Y3B;)h%}Ed{2?-8XfDJD6M{UYb9&=tv-hzBwQ&Q(}`!$v!KJbbdVhEy_
zn=MR%G=}}y3=;~4k~yag^Q;Nveqi_qhhWIqgD@XAZ=U8h7YyDG;(NoCNi*J3@rCqD
zVu$=!OG}G&fD5_F0S4gY6X#G_J^Ze^utxRe_siRO!~+01wR7H-;s9ao8AS&dqrEfT
zH-&7A)mM0LL|tjN8kXC(VMP5XN9qANzDH01j&UvlB(O^;ef=kFW-%&;eVn%gy|$oS
z@cpX#6WllUOEsYo1DFOU3$=AL`y&2grIBHO5nHl2wKLmH>!qtt{Y>TVg^(VkCsz|H
zaHdytzKEM(`alt!;FQ61@3!p|Crqe79j;`s2IR%p0M|yNt(oL-AvISwKim#+rq*De
zGkBw;lD4L%CAmj9UUX|6YF9|W;yP_~A`#a<eKn_88?i60xvc{+z3>DZ%C4qhf}VE&
zXPy0ZsGS_t?Q?CX>@SexWJ;{V4eur=tJh3(L6HR&NRkLGI$c{^yJZ2yb}u7OMQ8-=
zv4yPe>IvA5{zzJfX=;pf!AO4Oa6Uh6kn-D+GOeiZOs?JZE{^9n_Rgkz7!1n)v<_qo
zMW6%#r`!_AfI4Jh<L};r`l8VDDn<+5AGfZ79uc#jFHqm_<@M_#bn1!yy>l@$#uw#_
zMRgqgNCFbnreht|)Mf)6VjEwa{6FJtA6>HZ5Kx)WKhaF8^NcH$sG~mjp?P3?1xKsp
zRYUGNDgsxo5I!Xd>FP8vmu#sx$+dib?36;ToW#0=Qh~VZpP+*l)0+LzCW9K_^Bq9j
z(P`B<>DM>LEN<uY4cVG*Q0qkyvLtGu4X``_P-MA^QBxoJuQ+>reSySzXe$u(&T`+2
zn#JCTO=`H`K~tYn%<&F3o?!ZQkH(Ia4sGJs^%R#ZnH=&491>u-TfqGR6;}QtD?+h-
zi39A83g-K7KHHu-myMykBdz684?lnWxJf{mlVR+6os{1D@MxnCG=m{_L%=D)95j-8
zLAqr@*g1rGzxw`M5P+WC>Li))`sQ+F`=Rn6-z?HG^z!AAI2UWV$zH%R#@A+XFpoQ`
znCbj<mp?aN(7?jtGq~)i7clZt(9u~ot05nzu?e`nj_ZeqadEBL={5BHtRMx1dcg*c
z&(KzfQ-_h&z2XNd8@Y{RzGEU4JCKKt^3ejaFs$OW5VSxx{d|t8c<6;~@q0#k!J}dz
zh4dtQV8dn(##iwJkz5c;-!}wgDHef)=+%FK-C9S?SjmDdSuKPkKg}LKCv#K!?(h5Z
zSSIRl-(gDM;h>%xj=zrA0E^|QwJ&Zvx`jx7D7XPaG;))#Sh=r7T#9lfkz@W7J=o48
z&w+6MD>L8c;8SjFTdxJ3M(*t?Adh)yY**vF@kaAPErAW`zVQQ}18}cQJo%U(8MTZ|
zbU&EQsfb-!8*&Y~mG1i(gg*B)_LdC#nj)Mfk+N<^J4k*DV);2mb6zdr&reLZAHq<B
zz|>WG(#Cm+9s(TWDYt>Gil0qKXAJ8B!;{E<EcU4#Z~E-Bkfdf9?ehR8xJ@a#H#FR&
zW8=j6E6RxWNl_9bYj+K6LmWM({r?KEGv&x*Fi>}A-0QC$etjvL8F}z);1ew^`@Hpq
zTP?Z(O#bU(gb0R3IokFNRvaMF&4(davmp2o)n&^Tg`ETSNm9*iRK!MtJc<Nzrq4=`
zczgE@%N9E=jZ*1*=jxZpCB+djOkiI5rJcC)=|$7f4tIJrSTOPzao@JmQ_|aMX<DA@
zi0w#guYe5x%puPqPSLct#2WKEeA4w`m2jiYZ0u1A1|&14pW|uE!1$(W!53RdBNzgI
zx6S%^+}oNOTZB6p&BjPh;dGBNt)*@rJXizKB{e<xks!2vBW>WhfK?~p^afa)&iq1-
z=U`wV!{0ycV%+|fTL23v=%Yi*3Y2?#g6aik49nT;y(O1_dG)}1Gp(nHTZVmUM}B|c
z^QZ;)N>y9iI*jvKs6rNtRutZ=10?Jg&we7{uX}-BDYEOT%N>Z{xBwKvAq%RL%X~<3
z!aT^J_)&|D{(KKq*P4&%%(Jt%Z-C7acFG#T)WEr4IDU+xC)k7m@ukG&@3?;o2jCe9
zw&-3kY73b2DDC%Gu=w!gJRnHqj~ZFB;mYcfO0$QF=z|_UT;JuX+w>FaB#Rq3KL8Zy
ztj~Lpzl2DrPma4P;Gd}p0g73}d)ldMM!|2eSuf)cg2}*(<Z6=o*}XvrGWVC4S_hB(
z%&9o~9<@C?Xgrw^r|0L!A-WJaWj5~_!7tuE1qC}SEKI;?1yU|QwF8ZY{~Lkps<D3&
z#2z=4DGiZ=@zIQaGMhrU`8hdG^Ex?xRHrB6!vNQh+TD4xYwsOLY&-{T#3MO#v=eMK
z$aFw(J@EU`N7v13VQxzz%=Nt^yTC%PHlwsbEEY2m!dn+VLiM4q<xx-}hh#$UFY&Wu
zD{g|6SB-I2Yt|t~?LiEV`;#TL@RtF6V|rg&#|U{}h$3hLENt*#LKNjN17>HR#QXl-
z(=;+4Q5TfIqh1kB>C{e+msRgO92UgncIlp=K<@zdV245wK5RWB7N6KZ(X#^*7Z#Av
z#+V>IXj|)ToWs1WI3tE-vSe2Chan5J<>~JH31QXnurP&Tj?4GB90HBN_jMy_5!J&E
zGg<)>_k2N>`%(H}D@n!HtwW^}ypl^~ctNA;bp{IRPEsOs%7bUmdebOW*hP|!@xF9M
zcdWu`jt>$U{FIbFwWx^^_z$Hn#@#P;{ey}BJcY)7^!-HlPxS&DJ2R+w9A7qZ){dlf
z{C*31ro3sxgnL<JxV$57^$t$hF{$4YaKDlYh&U@NYhXJc%vgZGzdz3H!7PuZ%a*lQ
zXV|ompxOE(^8rmjNlklmFzl`0fYOgvKjiPGBLhenXk9u~g9ObN9I%nV5~fjTe@AVJ
zq_zw{0y#L410Zg$0ka+E$T|#Sl9D&#WIh!Yb>dlPY=-T0dG|p=cl9q$c2vX?4c$(i
zY31mW$1aCSAtm0Kum#%%3kv1}GU2Bm(f*$46Fr4-TQ{OrId1B6<Z(K~7l~}n+O_#+
zl>U|AC!pRlNsHsBY?i+WjR&C@N`ylRWlyyE_q@IzpZ>L|=rM9KOG_as?FSzCK1H2o
z#tl;34V0;A-!aqzHc-P+$EcC~O6)}45ob=uAdzVd8fj>1S5s2L<tdz5PALQT9ma?q
z1dk)bLvmTCc`aH!)m`}?K)<>B1NaFU0?nb@!6#d-s>+EpaNe7kg7~+@6kt&_4~&vi
z`oFx<evSctcgTRA=(~)?%DheKaizZV^0vWeB<qmK8!Y3ai^|bFcBcIvhiKEk=6)iW
zS7;jns@iBH-~%z#KlkJb?aKb{hP7qn8Fdr=HT+SyUczr^nRcYtYo-k#-#&gadg&<j
z3xVA2d^jzYt*2){rfmz%;DZ{DgNFxVzA%bRCuJm_)=NZANcD&f16P5sJ}OIp))zoA
zhL(UZ&9=uUBqYFdj9FR$uLYeSt6(EeP^di-5dolIbNL6R-)_K_NI#y5dG$n`8#P3<
zwEgJluz*evpF)*LEY|D<Q1IIX0FUz{aw<>}6D<-M0!qLAHpd4vfE2|a%nLLJ%<M(j
z^LmjlWo1Nz6NUwna|@;|KhPL#2hkSYuH(8Y+}WrY*nULGK+`<2Es{16xgIIapbcF6
znIvIOY=nm3vuAxFWh?wb${SsRks1dP)rP&10q)rOfj0bscWh1vIyw8GCwqSm44Wv#
z*TV?XBk%h=G4gucJ9AUhDMfti<uB$Am2!xhGjF4D)NJ$Sv4vyWU(`d<!M5r|=Kn_J
z=9m-`<+&2)t+7Btv1Pr@;KySvLx*YY)c`T;d^zoMjj+~W^R28Bf9AdJlTp0B5ek*0
z2and(g~kZ<TfKhsrVLVt+^3gf<tz}2Urg@~jW;~Jo3uTn`#v`ED+fQ-zmk#G+v+|Z
zz05lV{6UfvKf7_yIUi>9T|mZa<j%w?mO_lTmeMtyqEwl7%Byp21*dh80$cj+mBqqg
zlPWyx&mySDmaWmV1tbAQoNlSVP_036myp^ouvWtdbScFa@a-Z$eFQ@FU9*4tCQd)?
ztp#nFtc%Am@EBlR?Kpqb>dwBYhiXY7<*pTaj}7I^pzX_j8jAddGs!a4b@e~}El(Yc
zryY^he8#~xyICU~w%ue5Ob&`<nB#(p&}b5m9B!Bsqq4OMB3~e8n9pSWC@+0&X(Js?
z)n6^q#)#T_W0olgq?)HmMN`u!cd7+n;dqc;l{GZx@$m%C<8z>c)zAxQ)Ui3YN)}Y$
z)gu(KOdnJd^rALOi|CODNJ1KQc*v5dg$;XhIuy3^($dlhwPtnz8AyvcoI3<0Bqa7S
z5$XbBY>4h{fe>@+PLrzCZ8G)I>v`fMI$uCp=LV-kAy8CRC8?+aZ9amzVY=XZkr(Lv
zgN6Z8alr0LRY%r$F*i<E2{2FCFLYBimgLTxI~NGXhg2_4FD$J~SNszQoAtmd+1<$r
z(>nwl1q!Lvpzdx#6FQO#*zbI5Xjj*MGMa@)bT>?i23ZWGE;ipwvR@)hX)58k)LB9j
zYQ~sbpr{S@%W_yb#S@wTjD`n(%^!jvmkz}BJ4uXRYfFn)^?Nw)vf^6?|M9aO^DnI7
z6P^U!lC}7!4Kj~YQic$ie?S2fj53IzdwPEH1}B<+H%=cT3ydmhhU>Rhv}4|I<Y?wn
z;|=Af44=t?KpV!AO``v9C*&7#4zQi9TFhEkh{(__kb)0c0+LVFxR@gw0tom4Y=5-5
zzl(@@{_l$TnC|`HC!p>*CU`OlXbEcoJ!QjSZGNVEXrd5+wed2=au7?TD?=(*gaCrE
zeb;EN!>BR81P6n#ki0rMA-}peZutH#uaU_{B1ACvP6{<cc-}hsi@5Y;ZP1u284|ht
z@cV7NSew=2knF!q$fL;BQ9d_9MtH0ROs6AAV3flRO)N4wSf)IirQ#E3NL>PY_AQCd
zsyuK`O6SIi&N<N2hUXNcv%0;drK=smA8`jqd5}N_mV;*wEH<qspd6I~6G~?whx>j$
z1E?jqxnl~2@ctGHGUiz!iOozN=Z~T8u6I=;7R{ZT4B8;3A-8BjXn=TUt})<q2|Off
zNAUjR9kMaZY(Kg}WgiA`c!J_W=G{zW-c1o6nz6?9_U2|H^F^7Q2KCHUvPw!ywb#k-
z!y|JA0fQHasYVeS65v^dgtt9*?AX%@qMVGNhZguU+yMcQQveDEtPH=OmzkM~;UO9i
zcX*)844vJytq8%4Q?bV)IPNl~)zvztkxRQyV0}h_Op0z-;ZU~sZ()YS0Rdo{3r(Cz
zV5%DLgNF}0(p+U}J3K%`$+g4bc$74OX%MZfKtcQh*ZL5cLDr8i;D;7Z&Ec@4FSwW>
z_zVdM<ZUHy{b7%%r`ohLC>nm(2?4pz7g7dkdzkNCBMZ}#?SOE1yh9130L`t&a0uhZ
zEHae6pwOB{mwubB$_Lh0Q$3mhp6Y+J>Cc{iPExBijTvHbO%K8;oJR!IP@jF!zQG49
z4oPt5qOX4WcX5tqI>vEgR1p6Jq06kRsw;4<ASY)P&`zM`RMQsZah{siRsp*iSj(bC
z7LB%5GsVQjX3S{DeQs@iS$lluw8C^WLdnL3a5yP$SwwFIFyGOv6{e=93=B21hNfDM
z8H7*AmPieKc|+!Qx{|5n^mt@FRy!dgEKCjcagfZC_I>*;OzRY;^r%vLFwpY~Q5e7h
zFwpM4fKt?$1MR>RoF+{z;;<q8(l;R5R8yYCy2iev!11!}7%WF;PPaTHK_j^Wq)epX
z1M0~=v$%|mjF^}?;f9AiJaX(3zyU=G0|do2a!;Z<K@>^aQoxCU;G;`Ct(!4s8&GI;
zx`dAnYyv18K)iNP>s!D;Mj86p$X-mEBiv&_|M>@u1)$~GL&D37<Q{Oh*2zuA(xN7g
zlGrZ@@?jjr$1$s$zg1HYC5TpY`pl!qgDf6ePE}mK1pzO{P7r`cQ7Rr738=p@Dz~Yn
zG1><^YgTk%#3o~v%^wwJRj7G&+oPBZVZ#UvsDwrU4g8(gBpL@EZH1H?7^tOU2VoJ+
zy`1!#)1zyklx2dMO7Kh2rXjXc(FjNt`3;EuUZ!0;g3Jiw{6%?rCl8N?Z<07h`k<Vc
z&4UI?A10Fwc90fGgfN3T)Hk&L0XP5@)@k5*Ss`kXwuelX$$}0}R3F9J;5Ydn?OFk&
zi=#&WatbKyguG?FY17}dr!P^;28*t*kr`MYEeDOUVYi3JFO+dX;rD}TE{R4?pq~0D
zrQO-Z#TtXELEP<hH4=<L?hx(i$SL^|Q(Cs{5aLI^U|}Naq^Y@pREA2PfNX;?-W&JW
zj$lB%Qs_FbGprb}5cgdkd}%2)H(PYv5Je80Oaef@i%_Msnu3cOZ}Y_dl-9`>4HT?Q
zQad;qN5htmj`-MEfiP*7m5G^ueIpGVk1bD~OsLA}EpJ5wx`&TX#mtFJ_vCPX{((S@
z&-8+R1oS3wsh(5%Pf$G3SW4?&Omhd055aQWd047tvm8F5{&l9#mXpy?b^;+MYh@tD
zo+7yr;aQxCzpK-i{9(f$cQQX6H<Z$4jnM;irzKWfAkmZm<;xdRxvEATe5edKEe`fl
z$q>l0le7cmGDrfRT_wBYn-wCmhxhK0@q5IF1D{u?wvXwZ`;D8EB^M)dr2HhNVv%mF
z!`j>1;rKS}@sK>LU+vX2t3wS3e%J^o%>%3OTQ7poox`{tWGwi5Fj*B2tI&UsE~Y)s
z&afrf&&3klYO{X?I7;bk3v^ogwW{jI8adp0V9NYs$D%vSzGZM|sHr<>SM?b~c~j97
z1stwK*w%Yy&Y*d4-wGu49XmVvthL332psFX`hgPmMB0JwFL!aRs|H?<i68_y*xTEK
zv57<mEyNerAXPNDRScCp)E1D1kEum_l2Wo5I+Uh`ya7&-f<SF>5OeLQyg|_8Os<%p
zpKg=u-oSC8`p`JYXEf!xOC<ty0e>l?Qm+Hn;l@3e)ArZD4&WC)(k9bDYL2Dj`1H^V
z(O6YfWQ+<q2x?2he>69r%(c51F^LpL)Kq8VctM2_$aP{_-Y2<j;d_0bm}`*XI-$OM
z_dWoH`n{wxlrnPNr8(W2^2bm?Il5a94hpdhxvrf7z+usm4!ZI?%C%XiRx#Kuv$zBn
z60YGv9)A({UFXC=&`74+i_g%!DlZocc?0MNxiIWCnlI!gkE?aa0qY5LzX-^66gvS{
zmAwFpSu2cbM~4pjZ!1z46RGH??hCXY3iU&-yB)_DRMgZ=#cc7T{@GCSTGal`oV-Po
z0ekWh>r0XkG-|!)3cN_f0vO=Nn=vz3uJL+BuoGTx_ioHu^o1<}oc!gJESVo6e8YEm
z@1!Sp@}|jcyS#V*lfMe4fpR(=ov|weIT!GsrU%jpk=)!KZ+V}{!J{g!$8mgHP{8wA
zP*}4Qp@cNw`;$ygjb+JCq%S~t)yIh}BLvr0fbWPKn^bUg%F065$YE`76+6KbL)ZtM
zm3saDlh7ECi;wFMu5ZGYz3$4<@$rj8*CAWyp{xY&r16OrrF#x#)2bLwtdcM-yoGM&
z7RUztLSNEkv{z?}K{y#hWYOT$BjpB}1r!4WYgJ`oe+g*jZ52NLA?T=7D)q*V8&HQS
zE}nEI7gUFtGiSnf0SFOLoeg4wrHxHBCI^D^{O6pBOG++k?K=Xw`K*jLyI+sI%@|%k
zsq77>Yp8{TLB`N$+V6niD;f>R)74B7bfSMXMx`+wgNgfol-IA(T|<;B2%rNa*;Cfs
z>@+VF;x28l5kGzezp00S38=4apX1~2-3zbZ_AAI1F0GtD5<BebY9u!qrw`FEFE1}N
z8YHETW0LBwMeDC?Ipms4aRC2pH^VFMftls*34+xpoQRsjR?xyhckhupWPn_8!R8+s
zTcOv?ttZLDeYdW_s!N=8tPpXr!UGR%6iB9fBgh7p^MBCxdoF-i1u--+F%jm0zKI(Q
z4Fh|wY4#K<9~fAK{si>xy~;J0<@8?9iO!^FK_@a@!UC!(gccHN6GKm-bWZPL@mOZ-
zM<<*ETbHqD6f7UL5x)((^WU9h7%N;jy0J1r%^7wd!+y`?0=;<>n*V9ewmW9?zGtSA
z|BumaP2F09<sbYudzf>8uE#iRW1z5MOHfThI|dL;Y+wEF_R67CMm~0yi5+OX`Uwxd
zQJ~P`&rz>f4NUv;x;rqNsDMHS96Y%N(})NSY}5VwXI9C|R0N~_W;aNixCKka_p$Ba
z$jAtogohmN&i9UO?EX0fQ8Vjcj&IjU7`9~cr{W&S8A76*@peyWw7;1&B)-(qBvIXe
z0OiF%U3j5DtMCYOM1-)W7#&_N6^r`?Kp7GkHh3GvQO}>91W%Ibg7(gRGA6ttc*cww
zsOrBjoIrwt%%;CF$_ioAhnk2X1Q`46?4oCk-^GF2S$Gx_OY)v|=hw&)GwXI)LdQ>2
z{OsjRml|{gZPJb$G%^ag#nTh~$A0y3pA8P4Fk!;4-@h^I;jhW3(4z4pgiK;X85Vwa
zTrIyoCT&B>Q)SsQ^hLh5OYqdJDladGMgIKx%DOv^{nPn^Dk4Y9BgdqgmLG0H)}JMH
z;(7)um@#Xv$Wvq3>{%C0nb8*%ri>}%LR1wi296w<i2=56&H-=Np!s_O4_GQbg3z=h
zGcy%9ISW8ai~ny6GHsF7jO=$OQ9pr_n~{sY{hq%c8a~%9pNG(Ex#E~Q8(OaoF`!jB
zK<MTSy#cbA(R01S=lsQMBb#XbA+#-}&o~{)#=_3dP5^v7cZK(SdLTAy)>LBb$}et=
zfmS(M=kKaWz#QZb&BUN{01|w9??aQ}oB!#w7#r>$F^OCCL(_=ED%K(VK&`1p=Kby4
zaz#bxKaW_LilPEDqjMUq0iBGmbW$laD&h9MY(h!@7uwyCA9*Hxrbi%OBiy*y+MqJ6
z;PWO*z9LC9{ak*7(Q`@b{j_^x0c>nvM|DSoZUXBn-&C{;M@E-Vjs1rI=uSdjbvE-v
zB1yuDMCpg^M7I3+0V-pATN~y$)z{ZkQxp!B-;#9WBFKMFdj`p34nz)FX-8`9$Ly*K
zkbgn_z9#jZzNEk9CD>a*72go8+8Uc`C@C$br{WTr5@en1zKeQg5d#wU9emouo4cM_
zhQB$me;0J3V_EqbDr7jMh}B9<%}|TSc&BXPV1N*@_wPe~W$(Uy@2F3!ii<BW4gh+m
z-}w2Fa`^8`VKqsjV3BE~bFkIy_tOgK{Ge2ZMvJ_MCJl%7xukZzCIl)FE3PMoyuLA$
zgyKb2f$Yt>V;aefKZNkp6^pk3DJ?2GPj|=ubk^9k;jL>Zx6z9S5tc)NQESFA^pRxD
zL@gSTO>VkZ+TgGX=IQqvi1>}>my<_NF7y#s-zvt}0ZB;~BV1mMFFSr?W;Xw|Yw+?d
zepx+!YPml2?$@6i)VSDMlWp5sPUgT_r=|3u??B_>jG^r-SH60>$dxpUYWsXyaV#1%
z4$=)1XzWiv1c0?+7jc_moZv_vK3q6d@|_VMqiXT$`7{%2>l?nEKk6ONoL{*5xio}G
z5wM`+4>ifE>YYu;(nfxN<Bm)~Lp|Ynnh(zAv}%h^1Kb{K^p1f#0&-9Wf{rJ0@9gUX
z;J|4AI5$NU&>>w?Kd|vfK}B9303cmgS68%^LqpcpHAMdw(=c6GG?CpfZN!bI{K5%k
z9MST~oLL(eWfK01TPrl0tp7T)nD{>@*KGFAE-(t24#$Pv;amYGhg>4&;JAjJz$)4+
zQ$mGKqxejes6_OXJ^yhS*}(!hAhf8Q@KToDc4Du}6P5hU^4lW!hN3t4`wyjhB9VVo
z*l{b65p-M3OS?i0<5SgI_U>Z;Gv7f`Cg{+Nm*}nB#c1)kwiA;JGD!D##$TIQ7{)yh
z<9-%puz*$oE`AeI1Kfwhz_lD5H6;<O;70tf_xnAkntr?x8jc#C>x&w{9G^Io=Uf{G
zfSCTJ!FjEU$?!$Z(DiZ<;)Yx9)7RIp>+-~02de5?Mz6TK_e=^sIOVYl)0yB$YoZND
z{xI^vJlxmdzfro**kF?oA^!=HVd@2R6lI5M%HMpLCbRAI-W5-Dz8ykMrMDNHNr2`}
zC$hT`os5-e92`87qG?J8%l^lpxb3fmrYrZwbMT@}%4yB&uEq#g^ulat%g#r^wIAa)
z=FSpbz&Kriv&6+)J32b(a26xDA}UZw3^5G`VW{~37?iWQcIbl6w|93}&K}gs1(n4f
z3L4x7BcNxQ-MNRHIXj<eCaT^tlmcq%5Vs0R9>OR!91%KiZSiGhKRRVL{45|Q^eijP
zf`};n$VYuaIF65I3z093Oai;$8Q=)CdJRq4`ad!99M#!Rdh<8$23{Y&!>^wZx;kyL
zpxEd2m!PdHkH8%7DfDJs&&7<^b^kLi&~>MmS3=<JiN8uci4GuUw@lJz#BpHz-t{ns
z5-43XE?ZSY<0>n#LIw(U4RXNI{Q20-Q#~J%-7@o=e<onF7!k~bo+<_fOtiB)5B#s2
z>quf>YtOui9lIF%0LB8uDICqW1*C(XH4umd{HH@dH|5i#?aLp&MKHe5AVkt!LqUSX
zeVBwHGn#g>#m7&f>W54pwDv%$8IbMels%W`<OC|O%-dI$s$P>3MaoR4j-id=UmF-D
za&7rx%-A@{9T36>8|Q}7^Bl-v|4sXOh0&CQZk_v*qL@}iOqZ?8gozV-`}+ER{Q~Uw
z<HIU?er0E=v*}&TyOTf`;0Iv)|HnRL>K?_5AN|0!J;5#)?YHyBB&u=;UQ8RuPZSPP
zuP~aOb^Sfj159{W4rhv;=!a^27M2%f=%o*5E<x(^4#e4L54U2b<58}IT>!KhA0LnO
zATd!{GUUxapiPdhXVf+*E?d@Em>Gl)C{Q3*W>SuTPXEEI#h2rHxnU%<if4a+si_h1
z>bCyGt3Pu#r&>6{k=!b7hSmF+wHjMX+)u-97Wi=(Ne$9akf1SLnch%7<SR+BUG+OF
zF(h)K-gD9NWA<<VUbxXkIs3$f>;+!v^X8?#vPyEfP8;mr!Ref5U&y!D?g{_EH(>=q
zxqmgAkh8u(<WE_Zb{BpGUHWDkFSanx`dY;2NbBI<zyTCL3)_K!8;<8hC+}<odbk@5
z1Y~Ez7x-}4*?FPJ0p3q29MLf))1fmd&(RkUcS+cLimqBYBRb3R4rfSHh+aZ?ONK?*
z3VkdsT{>GC+N(G$nc+_`dz<vRmb^Th$be<$%_@XE2{}muq5mE4GBf7IHt9H`hzx)y
zdc$o{hLR`|Y=N(_6HIg_rI;{@Cx6ec;uG1(G@f17jP7Qhphm<hLN{FHC(}!63lt!@
z4^b)1HyE%Xq|F`32T|i`&9XwCkABm@0l6XG)&HF?&xJb5H_KJOn;!fKK(_`XqWST`
ziXJz#k%<msVgnhpKHAGA{Hq&K-LV9MnMvt@u;JZ#-&<%qJE*s=#dT;>4=)KH*FFE_
z)$aRz`@F{sY<-%uxjgFQug{Y$EaM+)MI8FP!NfcJ<G$xwl8Kg0o1${I7uUq`%}+aO
zci!J@-dWixg3gz+D^>&<_Tw}x8A<py?PpT2V_DD0HP`py^`HH}iG|t!?Q<i9<}9P!
zcU}16=XXORHy*2U>u&t90iiV_t7x5<cevyXl0<c<4z^lkIDh2Y=i&m6H3BFuna5wD
zy^+(=hgORTK5>rdl0rsv*F!fKIXt8~{dM={e|M)hCgs>EZ74aG6%`lf8!;*1<xs5f
zvWN{<FJAbV@a?Bf6%4|B%SUbb_8%34ky;fjS+az`6zqp-;loR}Jh315m$(|@sk~Dn
zd5ByfLqcu7%yV9%aub)iLc}`^U&y^Z#~t!kaQX=DkBV#wLc28eo{_J!@kJf$_@uP7
z>mRuxtI7vSMn0Z0RDaOx=c$SC=2fkmw1}~JS$&tn=_Qzj9g_FPAyNJM#-wjQdj?ji
ztlF?@*DBvtp{pLge~=ek$gScC^jiM@j;2V8-?E51>v#dcaH&t?QO8m}RxOso8~uH2
zn_!d%$}MRL3B|XA#3zh}9nckcfSDRAN}ihB(x2$F)CpzWy7*teei>^;JpfP**;?JG
z?Cyco9W(OnH+}9+e<u+6q3KWx+5^g*L*10KU6SUX#u+v<pEx$4ZnMI)GjN?|^A6%?
z+%?DJy`J4cD2if=ypBFJo?POVD!{#*pCWS}K|2};M~WAYkygh(#(l!fU%QWD30oGx
z*dRHU=G*+xLn94ZO)Jg{Lv76)zVVWff|Q;#s;@F4Gu)G{9=R&ZPCgZ=CzDa#(R*b3
zqARazA8%Rk<l$;m(9VpXer54#>2)%KACtsn+W!^|()%7cTfgz>SCp7G_~jK|Rk^=q
z!KOPaV4;p)gU&r(=SxI$h|65Ae7uwKv+jXmq{<rxXG8M=Vj+w-Qr`Y~4ir#3C(XWJ
zCN9ADq<q00-U3tsA=)Xx_g;6=73c>-*l^mJ-B+hg7;A3c>#2D<I(XLd%?MZ?m+##e
zQ}XlBkdO9$+GP5<deH}R=J6fK?Kgi@&dZZ9Ub*p_+77CI0o;7-M_U9_x*IWg0k8IB
z!!I24qYXG+u@2Cm6;rYcTzdwJNmvdMp8FNh1P;iuMGY?V&3Wy?Vg%wQ>-Z<AGZ#zk
z4t)|{UDu1e+FCa?5rKXmZJZ+y4ZyN;6~T|c{X8B0YoGqft0xOi4xXG>R(Dx-_5t0A
z1JGXI(_ON5*f@HX<I%KnT<)@b?sysYs~3u3QOUJl8@y=75@jHu+79a!AZ|Wb(WKo6
zC2?a7TLJWGkZX6|)WYHl^DJSUfMC$DU&iOQh5Pg;IKN4LzihUqWJp|GmaTEg(=7`w
zL*b;w<EC+JB5x+vW<KZ^ZtvBb!hF8GJF_@_P3U`^AFhT~0u6X22#!!Z)^o2p=Q8)`
zywBghd1I1C#8NNZ3GbRr&=1KX*(7WU*1v@wiuO#$aQ)yJlLUjh2Q<rgkAr{-%~Fw*
zM?&9z#}jekbM3C)9TTv^+&s-X{_S^nH3_7lI9FfxF56#Q*S&fY-ulHuG;X|ngy9&k
z6wz>hTSJnZ_Bt?Gl?L*kd$ep?=iz|BQ+ke`G-3p(M-Afo>AK@3%`y<-!<mFXU9Tx=
z<m$Q@j~DP)xry1c=4C4;x?u9x>MqY~AGsaDuqP*L)lYab2wL;iBdJajtL^d>T@5Ye
zU<6S+63ee`jQRHS_kfSTuC#>E8Ttl_9nbO~kj7!Z+KEn$u+(kt5Z<#-+!{xJC1Prr
z+Zf0??o*I4J66rLe!ds%bn-U0L7Tk@I#Q-%G^U1{nfrpgHg}fWGNw&1d)VIY@=9@V
z+3aj$r#c6GT#~+FwB*Ch<5Xw=IFu5tAQ@s%F`h4A#fB36B~-)<!o}3^My(bLrU8T6
z**%-63Ex@g7>siTWnktatz((bXLXPyI=!~!c5-quN>L&LT!ES5n-oN|YahRJm|0qJ
z>F(}ay9=${j-XSMHOf&e`0jpu2TC~GKhM6fMnSZDXv8}H^LKZ3EK05{Jl=r=i7(vt
z4@3(@i$+V!{zRMw641t&+z}o}hgiiW{AL+t_bnoZGp}iRFLOP9Z6eG%Ik61Gpm3A9
zd8McKE?d5<K=k6Le9#|(XEJQ)LQKC#_mg<3XXv+|lBLgyk-|SW5lgn(*zgV1_cOUm
zH;f+}00N4p<Z2VF(CFp~bi(gkShMPLul<tUGsY@sPJM;bPG+Jsx~UmzA3FdH&a7oG
zTp#&P=3yVlvggP5GVQq-CRZc#!$bF6c+>`?CDA_|TIM{Fj|Z3FmPck3)A6Z6of8v#
za_b?ACN6YhwL&l1yKN?y-~jSd#bjEg<7Av==gn@N9e3%SgSw*mu?y-^1yC5yX_)g9
zj!&U@E8q=sUzi?mGRq6^gD$}~Lvi?kJI|hl(s#@fB8y@Ie1)ofI`)_mj4~zOB$H_i
z6t8?}n)qss{Kc1#0I?e@l;UW6Xgu6}KQLm_p_J_=q9Jc?#W^d^C|$ac?pDyxqadoL
ztm2Osx=Q4Omqq;a&<Vs8Gg8@)^SF%b#_4QtW18|K<!q+>Q!J{b&TvctSL?fvqy>v`
z@vdTcVc9Ehp>4(5vLHRb44YN!uDt#3riP0kH=HTJFafLPu3GUYG=TEUd&60__Cm1^
zpam`+m^%V0Y{=y6TK{p1aL9fK&rr&~eK9(29{t8WgFDify#VFV2|l0NFiCw}kRDJU
zGhq}YL7Jj(TJJu6+$eK%Ey<8*g#f|u-lt^6DqWI3G>r%A8E<&@_db2N6L*JJNm%eQ
z0_HnBd7=tAU$ACP?<A6_0Pd6Neg9&aOiK^+EM8U3k@D$W!8k*f1_meM;sUPnN8pX7
z7-_C(VVvl;ZJ2quV+P2YxZt;n-P9xm_<Xb@zx_O<aeecB^Xzv5DV2{DfEP`gq%mWH
z^Hq8JN%BC9|9--ZaNanyq<JJ;$?H@>-FRi_aXy%E5@RrY+@&IBG~jHKu;rSn>y5QK
zyAVl$!b4w#$Dg#IAOoJw8-8s?@YAE>T9HoaG2d(tl7WA_mV?PwLMLuv<`yv1#zqIm
z-oYXvsYZ+*Qxc@-ugrS@rn{^KM1v&*wG;rPofmXh2647aDsgFPzEBX5-yRCx)CdoA
zhC$4<yzutB-c4h)0Aw~5<gB-so4n&CVi*#MURB0eRWz|BY%wf(YO4sQ+nah5(cD_M
zHK_D0i+Jd&oZZoPhHC-O!_hc1FFdQ9940*^)hb^3dz_p^;{P_=9s3_f;1IyO_o+?k
zd&`&5cdv~Iq&9VVj#UBO{Da>iwC-OLqQ8vW7BtzJ&6|p=RqUiDA;iZ6;{(aV%LkOR
z(dba=68T}y=>5m#&G%oiRSCB7W`|GHa`||1h-%FGOI!~z9OTfM!KH~FD>B(_Gf-e*
zcw|0w4^Z8q#{-{ER8>_uEj<DiU|}D%ch1ko^7xaZs0APjCjDmUKW%`VJB-4Cl*)}U
zra6jiuEe|y3%A>H!8E{Ic+ZrS2<3nsIk1!aa@Ggr^Y1w~L#YnpI#bR6``ya-9>Vkn
z`GrGZtzUnl0(u`Y7N26lj1=HoFzsz7|CehUn=ltoVHV<XRNJB&+><>tF!cz>g&wBR
z&(cm~uLmyLih7I;!9>tR#AKSuwT}*~jEsaG$C{skar){5iuDq&+yN*u8J#*l?LW?V
zB|tY|{~N9@Kph8l6daZJ!)PRnKsS9<V@0ZU2l_V@X33#YMc&lUn63!@@9<JgS^pp5
zWxB4k2+n+QEb};3RasLf-P9-B7>a{(HFd^@YibuhypwSMk8R||2(`quk-2sWP})RR
ziL}L0FzDFb-8d+c{n8bv{JpAG%n+Qh1gVxZU%@n<^%0Z0JW<FxEiL*V=D@`KnX9s1
zse#$th<Dwte7&c!PDhYuV7SDqhC8hwVaKdnd5*)Kn|*+*RaZMQQvJZ)wfA>@yuCKu
z5H24_XRxHIDjnRA6%*fX2T|2XdqD0^03vzVlY`TE(3m@c&9^l44Y5MGMk%AMSHR!M
z3F@Pp(+wqTwtzC`{1h0@-Yw6ZL-WU%qvvzCc2%qd2hW1#jM7F|ks}6P%TJtcZq5h<
zEM7h`!pOZE-h<Q5tWI4bsqP;5am6xRg}}_K<NcGZ7SUPsTNd2iZK*gzROzm(a`erc
zE1tg-h<GQz$jT~R;aGKtaF8D2f!kRm9?(zN(r}G;glyiaz02lX=Wgs1pJV+A%!{?5
z1^)lBUax_y@7A{L#Bedj_QfNwK)8#q9;@xq<fax~tZ*m5jJ{UpGkCy<ofis|EgK&A
zs#aGTnJ_U>yC6^|bs^-P!vS!plfE_eoN{I*2)@m)mU2%J!9s-+{Rl&AZfL<BS!X8C
z;XuLY8>^nqt%Cx*;nzigS+v}0;Lpe|?S)VyINq~LZL1(}N5m=C@vo(aDf6Xxdj9`%
z&Z~EEOcBVfyMut~^@=gaCGb`%Fzm2FG2Z(nRGcpfOfb7=_#ZoX1EijinX2#Ym+XMV
zJU9akqovffZu(h$GK{`6KMh56TSat*tnM}Lqc7Qa0oF#2`~M0yVPd0=;b?CCKgO;E
zoXWLpn=&PHrO8yr3aLz)DjH-AjfzI65;8@I-9(y0kue>annz@6&QwC7OevztoSFZA
zHuhF$@BjbLb$!?QzH`3b{l3q$p7pGCuY28#+e(bQT{oI~o!Y*Uw$IgZ1S0!8ntJ(a
zy1&+4$11J%rlaW2Tt56#ndp1!9vB27OcCgYbjw~adP4X+5gTJx?@p;($cF!q#FKc2
zv4BRglLL#n4tD5R6$ZIDwms3>Rs1i4B=dM?Z`Iol;lSrU4=z(<auSe+JVvsr+nHoG
zVK;Zt*LKJu1{Po5ObzkOCeus{9L(wbTx_i8cq<(rmNY>!{rm`cK+^Lu@#ZT1y<4{P
zi){v02fZ+LuL@xMR!P^GY~SGayA=wSSHs-4KvA+2B?Xu}3EEZVMR_k>Jp~A;*76W+
zZlYg&LPEf|%Y!YZk}SJGt(kJH6|%oh(D0mks@mh@UZUPfX6e9>Ls8sej}YKT{(z;l
z7*w=JZX^bsD=ccynJ9;d;prYU7|SZ>!G!CtT*^De<S-V)%@-!<qS(6zmotJdb8)az
zFev!&R9O-RqLoWw#y7rZ4z|Q?$QCl{&}B2?AFQn3eeX=SQ3`nawx~$*<IW?6J4Gfd
zDcOoXZte<La(Tt9xbFv4!x!)!o@k-3NZv?3IQOK4-k{sfvPmb7s!J%6r2X&Yia6AF
z=wNbC)!lM?9VGiOUYOi3Muugb7&bb==R*<omUd8E)nu;jDJQ6Jf&m_*0UrP$#18U#
z80w-cEfG=JFcN=?T$UM|eX<q>oYS-u^X>asqvo|R8;oRmi*XF1mu=hm^)vInwxiUt
zpy`!ygc_M~F9P>~{6anyfdbdz#qDN0de0KEg17&K6kpcG7W(ggUsbhNY+21t18NMT
z8}sI;$D%0Hly{1sb{?s6Tq}z7CNWu2_2kuEIDELZ>I+!JNJ9MS4!le--+TaSr%&@I
zTajYN-;oUQST7-jU$}Grl$>si__QC{w<VY+l8&l4*TVZmE{=#X#kwWBXNYjyc_Pt#
zs&YQ}-y7f2MJLMJGXTx&?=ET6`2&!q?wuQ#mbRWAG5m4m%xAY-f%KE_;<)jX!e&uG
zp*baN&B-8USk}N&`j9lSHY}YA`Z6L0^5!yDb0z`2hh*5ImfROYGnXA%a@j}Mtke7v
z*}Am)kr-y{?Xau60IXPBlkE$ta=ZxK+(9h@W27EcVU1`B=rPP<n!+*>h>OY`07ffU
zglW0&Ub}X(buu4=Tn0v$==6N+K8VE2R)r(#e28;K&B#r@l)im?Ir&@cRSD(v#yEiM
z=31rl)jS-|zo}aNxNk<C0u@o5XgCGFdN}p7d6ej%csTJ`;_<|jiDwd5t3}l3uq1Ng
z$3oZ2i(WO0aTJyUGgvCwY95pQ>jJ3`V-D<#V|(CiL4$4v{3^&MbseQ0GoLyUc8N2P
zM>e<a#Km*fEm0GDSnC8k{4atqYFIya9zjyWDoK)>NhhA8oURq6B+84mdUpn~OdCq9
zJI^aAp`X>rY9-^hk~|FQ+;9O3fCDOO!5t2G*;&KkZUyJUB}uzB++O$ZfkY|1ivBA)
zHr=}6J@R5=Nz0cmJq*tz%@Rn%Ht`m2o6M*fVn>JbPL@(F*f=kaD?z!M?!fkh<{ehF
z7G`WsD~NowtflnKWhXrW?;wETrkt%LqAcj#p0K;P4MRLo4Zh-*I1w+kLYfLrXB7t0
za+n>6-h>a@!7Wi^gg$^@1OqUV5F@=n2{8vhsGLOf`I^ta_XT40E%35eR;u<S`1Q&e
zS#4pmNC<Nlfdn7Cvk*nRHb@0fca9avD}tC73u7wsipzdkllcHlJW&8y^06&+*!-5K
zOxp+?6*O7)KRLuDs!0{2{|etb3pcI#N)Y3fP1?%QRt?fn;_9QoT^T?v`r@eaKU_CQ
z)>Rfm9H$W<B`64uOm>Ps6?a{T34FJ3XB!15zFb7nzYd<OD&Nd-<8SHW#|@TTK3mh=
zbs$seV*V#*zHbwgb!ixN(ZeY~hequQSG~$(@={P^xxX?w4de|4i+Uz+q(}|a^TGML
zcmKYH3|xKe7j!FNUdZd%t6G_xn?Wj8n2|gQcgwq(f4~|vBnl3ShD3e*IZ65I!E8NI
zUPEc2O=Z~bEn@Iqh_>xd!o;_#Ih5N8dm4Pm{n~Do)W+aSCSZQ3kFFkAioq`m(M<w?
z{3si;DWjn0TW1Lavn<GYsryFrYb4N~sw$IC^P^-3{~bZ6AyYIV;3-8v=WBafm7}4>
zvQtxK+IJoq>j~`b<*@L1xLjgYd5=CJ(}99D?j0OBki1|HlNXAxQrvO3!uWuCnXR9W
zHF+D|#9^rQ)UnsN#sH6n@V`5@pN%3F@@F#20FauwS^e<i*sHrxrZ~3(-4cow@~GQT
zbA}zD@Py5Z2bJ0)?$Z;7I13BU0NN)O#JI12<S}>nRklN-5N&lcx_|tG<DHxgjTPSb
zrJc1=`gZ0haZrH+xaWh}H6xLL)WO%QfE(03d0u})I&gL1p};$Vbs=gqfk7iUHIKsW
zrC5s$4bNS)>GW;~hssQEb?d203{D%tTFq~-BSKe&e#;w4wke7Lr9}#9iYPBsJp_E7
zxmBj_!<RmMc=6lisknMuq^L*~syy@<g<7K8Ss=!vr}Bheqhf>p$Mn<7u8d&h-)ieY
zYt%P|6Q9^}-N`q9)(+qHh?;*5o(3`qQUGtdpjmQ?-ocEO67~qqjXiPOu_bg=y1wz!
zj$VA~@<H?|RNb`~h<|$PbCmvy#CubO=%B`b0|LX0vq{MS&!vWOO`Lu6i&&~WlghS;
z?fGP04+2-k34|QQ0QG&f@<=90ozIlavCWEHLOMkgN<V&_c4`^;YW<WqNgJ-e*Ozqe
z@7|Xj@M!`iLnKhE<lVa+V#`#+OWk$;s!{#p;1Gg{ASr33jVB_QzW`2xhOr25-Oexn
z5a3qvtjqWB-7}B-^qjFlX@)MTRnKaj5bSuma3}HAYv99kngH_XOWNID$Eo}Jo12#U
z-VX*%Dj8B|_p@1S95pJ2XjGjkDxdlF07U=m2IeD9^cO&IE}zo*y(}VI;Ov{PNyROU
z!C?PLBdHchF-B;)C$BWKM=LVCbjWzPOqJHVF#`?|GB`ZdtMfo6u)SC}u^WP~7Euz_
z{y?VTS8H~XvDWpKC!PP3f*#+n&kpvkvJUoCUs6@&i*a?~29MArWgfqUG0MrJ*?BYv
zQa|BC7k-dJ{1qFlm4Dij%dxvY#2)iretP;8{s8X-2NK1P$#ArpN-8SEyqR|FoUyF?
zrsj_)8J9Q+E+xMi!z&1y7{406<Z}TsSi66au!p>vvA=T5`QAaaQAG<|vokTRro&Py
zqP*d59ssw{NIj7q9%}F^N<>G;Pv+n4_#1=J$S`|_8N+XN%zUKExTLIe#Vl#Zd30#%
zsgLC&B^+CN?Nd5WI%uZREM98@oWrgXXCXdN^+F%`i<47a(~ZIUE`R@iS!gmVg9rh*
z{+Eu&g56N+hC>Jrk0BG0@2(Hx1a!xD<i>Z)tHx8h9j-s$YGK_8hZDj-k)K%p;--%$
zM+<1Jck`*v!Q;oe|2PMjKFM9Kz_n0K>l?ZDWZ~#$%^O4y!FZWuVJ1GSBw$G`I-CIf
zZz-<p`B0X%>@5y0g_Urz<wbYz-aWOu|N7HUfy=IpzLNm8fCk9=gv9bc3ZPt_?AIO3
ze|i{N0OT93sh7`&I1)7vl@A~zaZxUxWlY3ut&)rw<S?YJXF9O7mPi1Gn@vDV>V<UT
zA=n>6<p;Ds{@u&EAed49Xuop&QkkWa%^Z1GQd06&P9m%NegIuEq{aaEenD-fv24~i
z2tEWq{bW9z!k|dM&;-hbK>eajC-aVb&4=sn8z)c%&4L)cnpu05Ha<8hY%RdVwvzm~
zMTd4E*KEk%+X<!P5PqCMu~aG~3thl&%A3{?4iC)baq%UID~hmr$=!+`MLp>m!$>AW
zjXxuV>g1A#>u4y9^t+)+j9vsKHilejB}r$}(7X3)R^Y)pkr+FHhsn&=ErE{$)TQBY
zRQuOAll_i;BMWz!^r&ao=wTof28)0R7G8W=m3Xf`C>DVaONAOy=m1Z`&;o|sF4Tj{
zY(hzs$0J`|TS-CgfTwY)ak_Cv$F?}z;WYUR!|E@;gOm7lbn>NVrNB-s;**(N(V9_8
z97LU^;qXdymM0t_{&Y$?FpdO(9E%It`+Ty{$Znx!B?%v47YY)U+DMc)!r2hi8^$>9
zZMZJ%xQuFBe^s?XGX^P4V?I7LN1v?KT%l4E;cKsMgd~1owobBA&Ir~UvWxBmd0``W
zp?$2m{6Ck}uvQcSzBAw$^oODfl^oy_pdVz3y5s+w9YKnqXG~u6!JRt)CSXtz7=GK^
z);q09Sr}?$5x@H>le;JUNDeyoa)ad}PO3o!Dg~(gRO*|_Mwu$FyJctY`W(s()oHGB
z@k`e)m3%%$5%@;`d=T>gynINFXG|LLh)U~dOpi8}o$wSb&GiMOR#1-&Lgk2K>L<?N
zjy}wXh!D1xVq*2V+8mA_9~Ablt%8<p$nYyI9RW95%BkQ6knGi3A<7(E{r^ZtPxwVp
zLx<Txs=G_&hjWKi4+~ZcE{h2kA{J9DBrM_`)bc54CSW->W2S44Ikc%R<}UgwEheGs
z6Q;=Ym71K6<)BpnOs705s=j&|g7sPl(*7aFhiSR|J{mowX?nrbcsF@l?jyKc$CgHi
z;cIG<?QpO9M*9dMRNt(vTMw@qOO&XkE|m3YATNnF2Yc#eZ+iiu3Q5#rl0KOF&F5%n
z??G4|*5QBEG%8Q|%s3pE*VW*4Dxt%9@dOpWU#>CU;4dl6$P;3dL5+*fk3capl?F|1
zpg~iMLzal5VSjY{nwmb<5H%3<5@K0LQhil;-Vv3P3WXaRn7McTC3cFTcPboYqhg67
znUye{mWKE9t@EJX5B@*9gp2sRt?=~7*R2GWinD!RiXQe|YVhuUHY2<ZYK3&$i<9So
zav0p&KN3ESgWQ?M9BM^!yBO!t0^%D;X6pPKyieWVx6gzd{6d><h@$n}KTV5}gb4TM
zYzfQKT5_xF`rY^Vxgm&$T!qJhBw1T?E#xLK2sD6?7H@gA)UFfFYHEB}V9QkClOb#B
znb~=2SqpIv@=?uxe5Zc_*6DRMcz3(czsTIIER($R&t)QRi!hWV6IhSsY`v|45ti*u
z*U&Gz+6*RTmxz)L@wsB%|GH5V3Mg!;#y=H2mb1HvX+-)LYwFL8bh;BcehDBYIA4I2
z`S!Ln_cbZ*HCet^NtgaIECdmRkLMrBDb6M!uC3u%sS(U`19j4_-@jL{roNu&DB{uS
z-LTMP_M{Wo!HI0MCF$AeGgkh;vQw$ScL4W5Sn515aL`X)G>CTRbaN5Ox7w44wj$#p
z`&2zEmngr7B<3-Fvx#g&fb?GaJllhY;1E%AA8sH)JS6rs6EA<qaHVVTu*F*vm(*5_
z!9D4P?lq!2N)s(qot6b^1=1=anD^kKOL#^hJn#w?p5KxZjCETakvt)+yliCsS7i&S
zDwXy2WkI=MDkg-tv#MNO0DOw#QsnLnq6B+P-h&O-Df+eOHVENgD$G_`-|j`l7c~;8
z_#zIUMwIEksWKz%EnJ1x)_-KC;OCf15(D3YxXM|L@NL5?*{*H4A0%w8$IQCzmfG3d
ze}($C$%my-r`#&chxL)?YE?v)AM{!--83Cg34mf*Xhr&R7tm2ElafCWjs*)Akm5Ey
zNd=u^QyHp+fb}Ei5qq)@Jf~TElC(p~s&fa$iSmw1wi@^=gGXLSIE(tG_JV9^kKft0
z@5>Xk6D!O?l4>q%rqfv(sp-nOuoiZK1}d{AJ8?FQ-Sa|2CB_UPnxc|A;V3Oaklzg#
z+Y3jE+>^OEKwMfd^_{y;xbf6uZ$Ey_R{*7*#oUz~KVzkf5vn_aa(_N;D2zy2Z34Fb
z*b)s^j}XmRG^5sY+hlm&UK#vq*r{(M2#hhcRC;*o4!y6Du1hWp#_-6b(E*Nu)-!$?
zkYXVkWdKblZ*sFhy!Qv=r3R44DCkLIc0Hsx_6>bCLs_{|&oSM{(8y{w)0LWp!dd^x
zDf5jt7y?$<QBK{!lo1alqXY@~ALWaizCm*Af9Fboqd&}*c&2yZkJ7ujJa9(B!1Q+@
zOQjF`F@k@SUyC5mvp)HMd9&$kmcy936=~u$^`e?PhCP3VPBEA6O0>6+@;P=}{6(aE
zez%Em-u7usCt*V16)GU|H%v93o!FHGNCD2ApmJyhKRQoH3n7We{)(w+Uh1_hM>RTA
z8iF054-x+akscK<2R4Pd11-DS!O=66J_i6ZdF|dm!UQfh3SdyU?EEwu0Y~JW;6kBL
zp-v&|Z$Ex~$syd2oLCavtx*jJnK9wf7R{ml_MY<k4}S%oaPL&Yu!~lN;5v}w3M1@Q
zzp5cP$EF4G;hU^-M%f(a21nT(H_rbboLp(sn`4QTqai20kwxVQHjWd#+_jb0!Yf~s
zZa}M~2TT`9l|x+Hv&ivbgxl$cH!*LT?{LDaC@#%tfSek{v7BmWen+4v$c<R0-$~c=
zvUxeAHqfAk;d{~0RtZ$The<U^yC<G7R2*77QUPaBI1R3`+;9_p;Lneq$e(N|7%w<Z
zXot{qAsOML0D?XhkOkF$P8Ku|ZGrYKcjp6T{PoXn{y#HUd<K<d`Ksy1(pazd6`0vF
z&Yk+k;8t_6Q5F||T@>{i0|&l@oOzEtsJK`awRFf@SPoW!0k;VXE2P!P-acX0#9#D8
zc7$^UurPGEZlbYdxF&wL?BNtZ1QVXx9i0SU$&$<k{~zl(AZr%w;4&K&wxrdCe^q#C
z?qkAA?hG`Z7cOst6sY_{;?-7SDgnVpE*%6|3~Z#)`v{f}^4~-ZfW7}aT*_<I?tC!R
zQ7dsXJsUbk3iq|`J=E&z-<e0;)HvDUgs8&df}bO?0rT0!27Y<h^hM`xB0WM17IXhX
za>l+U+DjJ<z1xUr9xZ%VKai6mGm&?kNeLj@g)Ob!Q0vAUV0Gqh&hRvkS=aiQssL%-
zvz!lNxl@Kg6by;FZ2qdb%R6&3l}^W*rMOxh8%}V2?4<>VMG(LgQ_~TW&xYx<1{aca
zP~R6+D6v?CI6>3ClvFK64+@!xWtB9E*(HBWk(vPbq`Nc~$6HOqz9eRh04g3qgLA&N
z=;1cxYvwZhPp<n8=la+&&I&bB8zX6Md|{xu{5^=3&RFDud1)<6Dab?Nx>#dRBk15O
z0wdD+*DbWMEIGbgl!+<s;$B({;RTZeD~G9(huHB|Pl4Z{bZzBgK18}0A*+<X-;fS~
zjp6wgEdR2N3C-8yJzO5SBPQaGU`JBPAMD;YUMW>FqHuNmRA!jrge}|!D7o$|P?_=-
z0G;9tQaP|ojnvpD4cK?#LTiWfLmBO_!DYAqU4?GH5fBWd@;|T>=1k4u27oPm6!0-Q
z{5K!~X;JU|{rh$J{Fh(e|3KI5RipbSn4rE>KL~~yA<2qZY1sr7BCUm7c*e1}=<-AZ
zz{cAaL*pUJ4atOrep5i10Fmeb3bj&>n)ddMC;oACoJ6gZlXT=?1mOfdIl7yF_>C^1
zy4j*mhQi3IJ0i|eL>QXn@s~A?Fl_Zrkzo*3<M!|zFf=R3wY6m!`vpJ5Hr(AmUpd<*
zqjwK;#y@yQ)cA*5P^t4uY8u(JgV?kN82`Lo#j}Dxrb7V8BK{I2oyhh27ekXpbAlK5
zoJl930C#ldo?TFlMZ=?&I4oOiGa}5^z5y`QB_0>uCRF}ZLgO2aa3EBP+I>Xf`~8H}
zCdp!@hdQz=##Ag!`}_#1KJpgVxBVI3jPQB=Ba!m?K?N~<NVo~LX{&B^MRxq4Mm&&k
z(+IUwz5%Sj+AeAAt0RTFqmZwMJadCz|4Y9RU6rlaEwpc<wWMw2YhCcQvP6@4>oe|P
z0YOl<CMT_C{@bE#Wbi}#?Wedy%za2;HPr_a+gqjLNd3qaNX}-DiXB8vL{cynWB(ul
zedUbq2aF3x+dAkcBp6tD>YQfIGrhsjdY3Ba5-dRY9V<Qbl~-@=m45OY6?HD5Yy^%y
zkhiiF`7N6)LfiWK@e~X(xV|zXcjsMndRq9AgW@?u=RoO~vR&_GgqHk4z6T%HQ#;xy
zp6@BahNu+q(s_WMO1$ch%Yi&>x__qE2oc|kiypznPJd<=0WVKRIbnJYS}FGr4qeB^
z1@T)l2XTIQ0ftp=lF6U3Jnsk5<SLIyBz*+KF=(hVhlhs;bOS54@7@PC2XJ=nad6N%
zDjV+ZG#S|YHv$C*2k@iq-;0X~;}93tBwSbu7>=yz+7MQ`6Qul1t0Y}|G>}u|DDk5v
zo0QPAU31plumPg^8pTQX@-o(6bMAt`a!cK!ew4QFqPT~<^3IMExyns-PYbDMrHe?#
z-A6V6yf_?v{ui@4;?Zc@%7yu+P&X%PN2nB&xs&VWwB6(P9x+onESx2TiSOr{G(teW
zW?#a3b^4mwC4%{bSM4hB-~}^80*$%k+D^^#+NTU<M{g1O`Yhs5+MH4TZ~gY$>o#oM
z=rxHz#h+A>#(W#CD}3+l9T+s17o`=`Rc@{(929a-9>qP`3(4@Xr=K*c26mxYA9S9i
z)m8MA!2d3pM1*9-(ms2A3wkCDp0#aBJA#(r-?HTqu_Jv0uaYXIRASz|0o`%Cip#gs
zBwWREWCId2Ag8#iTz#g`2y_Wng!q|+)k&{kn?_&K!aEQjs)(?*s0Qjl#=kaCp1Z-|
z_%|N<WfM72{A}A99W|@`akv-X?75k_rY8jgPCra;sV^Sk=AN}2H;SrcCXl(y+Ct<O
zua^nV<-NIzw1!gg<sfBLd8lq;8rEOH3&+g3`Vrt3qED5yT8AFW_}?x4@L_kBa#3-y
z3+BL~T^jRNx+z@ZP+f@ciLr%SCF3MMrKs>huJX4m%|eomxie7JE=-X3!(`wV8xq(}
zd;q=xZ?j1T*fk-Wl1DrSp)^U{asIR<*t^^hGm{ex+>g<3jn^;G1>~8&kkzC_+^@gY
zL9Gv50l!A40kCfUpt4OSx|R%7LC|My_Y%NXLr&X(hV2)a(JJ%D0kW3m0G|U+yz*c6
zg~MTYG(Yh-=OOm7?o%euL5l>35o@r9HRvYZzyIzFrw7t>TmT71f4oyUVW$(yUV@De
zg#V(rKQ{Ocz{)Yi4UJNo^)taYT-&w*N&rL6pHQl=Tffuk5K>7m!hBB%I+}wh=fh8r
zVBJ#S5f#Wv=d+Ovpa4cUJ45|k;ZB7|qwhj^d%dD33@_&IaAEhZ@$IsR?;V5G+B<+A
zm|)|Dv9%{WO{UZyhP;F)22>45d_i`EAy5w$J?3Imta7iP03kqlU8tZMt2j>1tEBT%
z&Ome<?uN`>d*jO#Rub`}X~f!PSKRH}=Vq+bTuK^ESYwS4k9QE=Qe#Iu<+Lxtj&|S2
za7`k^Atq(1WO5C&g|G%QN|*kGBMA`*HHugC&;*;Jd1wNrT8-C!`@@8MzW}K?S$z^D
zP1Kj%t~jWnnSGiWFNuq%p!Ii6%Kfe0B0+KHuDeEleky)GWIDX@C+EmtzwTpb!MPRY
z@#zo(9lb*BiInHS<OvWfY=y=^)Z;4{|95-n3i3J#&De|SErQ=UEO#lBhx|4<ERl-S
z+0aUlp-T>`7Mf_a1CSCunb?Es&exIaZ)YCI=bk>a3c}Rnih1V`D#91c>)_vi(}f;E
z`_FFycX(yVy}=Wlc!uU(F45eK`1trC_6IP4h9GFU)C{HU_wZ=5xl28x+T?s#)-SCE
zj|lht%8k1bx@s$vGrGeEEV>)z)z8QouyhU|{&>4`#Gl-NGt$y}+#cQlho>7leukmn
zCXR6otox?D^UJpb(nUlbrsy@yJfmGaG|zq0&4P&pdl|?+pd0fwZLAG$;h~(EtgJRA
z8A&TTF|`+xR~QF`{)*}<ku@o#(kE&&a;J-J#tD5wAc^2dWn>&(NaAUF6qPf@v+;63
zuhSGjlF^0R#C3gAhqmp$^c#GvG`6+(FyM30b^tTu=e}7;_1CWs6HH~os?p$dm#ImV
z)<V$5e-IZ@{~kI#xFUIfnE3SM+g|;EuowS0rv;_*@^tr<-0Ng0FBrR~q2I*lOThOY
zUI>%Mi~aHu6h{|6C3AOB*l8ZR6~I3uImht2O&V13529btD}Brf*p)i!F6Smsb4;b!
za}i(dj~21Dc21D_?;aN~JA8p2n>^F?tPO;ldYaq4YAIZat9|JKmH%T@9uxZQ0-<01
zeU5||<_!%<Xr5aU(ic?iH#A`BcnJG4O}lE$0a$QSszq<8U9XU9&bHfPW`e~09gi-#
zTq&XXsq;Y6{9#w5@<S~A@E3?&?r{0u+hpdKjxws^J9T<=BX$ikZhp`0ppalc^opGD
zKD{>NO_mh1qQ)V2jJ?4N_(nSpduA>|xS0Ba01!7*>v)5gy~D%_u%SFplk3vuBj#{I
zt}Xn6ED}D<uRj=MJbX>?En3zfdDlVB8*xqKulLrKqRo|Q!HVz3^gW4PgJx>Nz(YQ0
z%jDkm`+_X7%1g0+urdGxn|=?)5f3wBfk{*^2m%GviSB*;DI0tNBSvJhKNH?SqCYnB
zwz0Pea^cDK81POFURbgA6(;Ic+g-jBF=<QbDPtlRyXzb7Cl6hC3}Vac!E=VPdiB)?
zx-rV%Cmx?h8MJjntfgHxyzmpr>0j_;*y%mJKwfDVVG8A4H<vF!<M^wKvJf9ywi}X1
z)XR|vUybhMs>(`-HqGP5wwTL|{oS+I)R6T~$o-7=|BRu-%{AHC*Cxwk^$gl};C`%a
z8a{5rAG@2I8ezgY6z+t0leQ%}ZfQI2uD_VS@W?6=65oOrcx%|$^+S;f7lWN4yZ<H(
zYPQVzG5u3~MrLMj6uM+FbnCv9S3knF?4iS$Ls_yT6Fn{^Lnn0mTb_<u8uf#%CP>}7
zd)Kw?ORlVM3OX6kq~5-rWZd;0NH7rc^W_#3#lZEK@Xx5r&5PwOAyjdfu2HJE%*rt<
z9!th%4XovZewu07&Z*Z>Y3=mu2lq3iGmh_R<B*|*Z^zp~zqX$bYhHhCuOqm7JLFzX
zWrN>95D(&$C@_?kl;DUqBdRQ+#|-BM#m>BhDvWXZS~BIw>kctX-o~b;p2q?Qi%Lt~
zUQ81~^%r-R6-If%TeBppoh(@VI-(MDt0F3&zSuf1dDfUJxmXAq<+Fx@7K4?R{1U|s
z@1d8rkeFSstQck5H57-b#C^4Qa*%bBmN=7=U)nmYLxYh0@<L=W)RCZfTixOW{q>>+
zQ$1U^2LB3uzIE5$z4blBb`-Bbs~r&U;i@pfn`LBzG0u~`Bt5P?d<3`B7ME#=i7N@n
z_~-joJ=)(@GdUGe%&~8+F$a5_2G2;GT9$vn0&*cklc6Y_(4GBkMjiT<u-Lz*^O=+&
zx<wYVK3uLI)uRs<3TZaKI(s|Gc!N7n>aw%>SzJ87vL|w~th6%Y7Tqt108O=N3wG!y
zRH$Z{O(A|}9vIFWpecbiwwlT?F`WbVYMPrH8)G3jL&e|TX>EEskl9szMI&)l_F(7D
zH;bCJgR6$@V|*_tuB?Z61Po<_POaiha+myGFG1XYO+(!&YpyGPHynWLgc$@6#XNR2
z_9?ovBOmI4-szl`g1z~+gD&>*gofoe4h_@#;Ur)o%{BAGi3+h%C<{yCEs}V^hd!)y
z$%mFP8f1<3h}Ow?*F8H9{=iUmSJp!$QozT8mq3nzg(W$UC~Q^l<-~vYpZ@@bi^01c
znD!)={L(C;WI!StSMs%wTnI(<>L<@QZ^*u>EWb((%6<6%Cr&%J;#q&$dME;aw&G#&
z|I}HQJs2uY9?OzUh7Up%zp2_@95@Qpnx`dmkbfbo=2TtjcVTn?%gYD)F_|OM#U}e&
z=Bt<D-q)@@;)E<`4hf30nqv^hvarNDW*~PUOwZivXV617(;H77g1_wQupD=C`wFmZ
zaR%g{Bo~77`hYJ00b+DpAEdf1ht~IcYU^5+eXOoNn73~L4k5+Ia1eKa*4a=B;%Ob6
z_qr*&_U!RS6i*7L=*$Pk@Du;FC&I7ma)7vE%{MMwl~`;Yg-IvC8NRV{)%WqX-AmT%
zofjZKioeVsL9lyD@cu9|fRwDPB3ao)HKbczQBOaSrU`}Y7&0tz=qXW8Hs9p5eECcK
zBhW1~eGCs`aI%RJF5osytwiG7fupZmH8WD%%LUy(aJ1To&qtHn0vn`7Bv#IZr6bo&
z5+=WmziO8b63IuP1+q@cP9Gnf8vgku?;faM{(fLcF9UmeIlUT`jHtFg#07(^v!o+@
z?4CUp(Ca7i-2P{Xj5i{XOI8an=7)anvmf4EJt*cJUb$@#WGo?bhW#D>t7Y-qe@ee(
z)~DP^4-%$+s40HwHr{K{&l92KQuoYbTO!-~6uvL#7{oBdUqcfYnh;Goa+^Qa{Xodm
z>B;1O*thumi3~69fES{R1k<S9#nq&Jz(iMcba5pka0f;s5>4=4Te0;v0LHV!-$#D6
z6@bQXEUM1$a`R0$Qx`9)?;?&d<mM~5fa@_nx_&PN<Oh43csHK5<;h3Pi_N-7)YSoC
z2ERKF%pi*TF!E(aKq$AiU1)<hR}jMg_D{Cn>n#F3xgm*zU(PK*_U(Wzme}&39+{#&
zL1#}h5`MTAQ!lByY=NGO-k3w?8rnwi9f8j8ZZD1dyYyK6sVk5*$Xe&OT7*S4u}#nu
z)9E7?q#=Gy-T!$dx=J|wwV<X-{_di(E7fv(X#g%wc<c18U^Hb!<!(<rUjW@TqGB<O
zdBt7=CBD+R8|spNdCm4{eIMA*ieOpKs_YS@cm~g8V`AGS?>Y=GtN}5yBl`jjJ+r4F
z;=Z>efpEU}?kJu+u#ja%EdtA*L+$i;3w!ewLZXg$wB|t<nIItAgsP&i1ATPzcg@Hf
zJcHzb*Tw4EKnYIPNwx;Hp^z0WLw(speW1bN<Bwg!|98m;b`}<PVE%p0Y-AkZ(#~#N
zUdFgTpx-(7{kM>qLMh}VA(_X&8PJmN&VZZ;57Z(GcNYKbeW9N9xE(!P+$`u*q$Uwd
zLD?vMh%*$Haq5SgR(p@O_4(;zV%R)_4yTt9F^3R$<tyiLOS<J{LTBZ9wSjva4PlU7
zVy7{EFfeJ|oEMganLvL&x>8GTFS*=;;$!6(oiE^TTFQ*Qwvm-~*>Yg0YK6JZAc4__
zT0t7MI&r@jBj9}SQ}5qnAm|uiY@ra_iRX0~NU^23eX8C;pXJNB-pG>BsMK;KFr2%r
z!TcqzCgW=TXmuxXnI|`)N;T4+Z+8}`=f1p6{_qjLE5!W(bPY)K=gP%gd+S1t+3Jxp
zM*p#Z*N=>*#;?mpQezQb@{OIMBWs9)a)0)ZD(|!JI(@tSUc0Q>Q8ke*^|-+!1C{d&
z;$~{y5xIWx_+ufJ2eMJK*baTT_u#mNfx1?%{N{5ff+lgPShJ=zuCRK>QXMzv*3@}(
zZm>O4I${5&sdU$#OQx#3&d+<j|7%xxOt8@Fjpg|juCM!AyFV|AYMzyn;bqp>G0>aw
zuT3US=h9<*xm|AB&b2v{ZGGM<v=X&=A$hh<EdEP79&<@dcXoojKxf{YY$TDthdyn5
z=s5T7XX5a>hU%dOTwZOkPj&7%Y5NVl&ytQvqr3Ge`=RtfuRJGtNjBEJIc%&3lKM`q
zxzp=+145$g{|tVdrnE5qN7^t02ARfn*e03dt~2y08w11nQ$ub;TuWHz1_W4t4dt)D
z_s_Y^<8|9N@EW|nCmLwE%dpmP@+Q$RzX?+p`~%qFpc$(WxA{&XZv5zhqh<~As|Zj4
z3?*y%=PReRz9GJFF`9+KrL;LmItxp{Zp@e@zX<hhouOR7zFd_N-8Dl4HNrY6T;iT~
zL$<F0UHvc-m4niq3xusAjb^j$Z?4wa`YbT#+q0E&QP7R=f50X!7#k@qsL#RH!5dRx
zapARY-;bOVtq=yP3+xVq93w}N{$!R~a~)_W<L9oeyB+scFYWg2eM2$Yh@pygQf_4T
z4EPzI(QXSI`+2#BwU7-5=3*yEh68l(9fws6H<a|}=4jN!4;>Ug*!?MIV=Gcy2=RXn
z8T(trx8S!(k+etFgv$vKQs2-Qj3o+bT<i;}enM~?T8jeF25*t)xme6dNWmqnKA8jZ
zwyMhbl*Giw-T?mWyZ9~nyAbE6u;_dV*32sacDWx)X<XGo7w2wl2<cu$-@{Il{i#c=
zP`da%_sq{;y`@kG>l@VjjqpnT?P27TT-u3HG^-eEE21-<5yXP<#B=iL;J2HD$-t<b
zl!(nJP;Ew_XZvZ@c4#MPv>5qdxPq3n?qeu)vl+&YXDONh$%;z98MT1a8*|^FU!aTi
zA>YA)@3vi?%3lfKOztx@PT3B&UEGu%L?Y+42OZG#!s=(ZDSP$U4;Ma_bHmRe2~)@U
zo@YV8x4Ur6TJvD7O)+D9Js1zozMi!?6wJ{$Z&@cov!tpPu$_Np^E=}jI*3U<RfFM>
z^{p7P2Y7$a%ufnMP+XQ@2mPu9bh~4}>iNL%h^>JSxW_3@18Q#PpkQ2Ql`iC2Vxm5Q
z`&;i<h3stl+E+DT?Q@}tKGk!2^t^4yW^VPPt-E6*Zxg#C2MS+{ss^X}0ZQ*1$|2}K
zeiv#6T>VEoTHd^FL<3<p`{0+XYL46LZ$)6sUyN5#G(sIYcHXUkz-Z1t4yernFmBn%
z7pU*=5~R-}#9g#kH~fkNfw~Y97Idkij#zu)!lZ;}gY++N#{>qqp?H5ti`7S{A0-oI
zKMo}j<;-ko(D!w$Z6yPUAAEHDIHc^0krDd`*q1!?b10LV-r|!E{Ptb7!CRW^_;;o=
zu1<uC{MSrTG1g48dEKl^w}!6n?WgawVt9)>#r_k<K)e&$r%zjlU!XLWv39Y6@8kXb
z%+B)EecM<MN?{u`Yi)fY+l~~@qU&G-0bg@s*87_XdW-068TBMd%WF|g=HjTrc{zBI
z?nTlc%ejUGfD$a<BB_n$OJmv3AsUj<eB~CUD6!~w`CzlLrL}`8tT<7QPELCNHn*UJ
z<ef9|@Zlj5s!#LcLUMW2=bPWd?|$>DBVQucsyP+xg_y$$U1#irIg>YgUM}^k!32Sj
zd2lqQI|M0&Ut?#i_D@kPaNxdEoV0_lVTCQ>BjGoh*uy691S-SMMqhy32vVer#<5_a
zYv1n~&WmaRD%W5^9`_&Vs`2v!m71$x3f-5Q|5_hqA3TfnX-0wH!0ngf`46>PA+1RG
zJ=(9bPm$z@Xc=coEc6@fCo>Ly%LawRH)xK$8PZ<DVzgWBQM^*FNBtx#oL>Re`At{E
zIpHK?$2pcQ>+PfJ1G#Zm2kHwovaxhL!%(7$<r>|Xqa56v-5r1@V_JJA3Sj)*HJk2i
z&jkv<?pEe;a%>6I)jqIf(wucIDYgTO89J?q*1fQ*WO7~CQb(-X(%x+v&=wm%>X-9o
zvO$<7*eIT@{SKP>UNkSH-8(r_xOd`-L@7$dJ>ry-rVfDSKSIX7995e|+*7jVP7b6I
zE&~5Rc#cd#{bIg*=WpcqXtO?r)#Usw`YZkW4cnpKvVe|<Z=Yr!S3-&Pcj4mwmHS=-
zU#KoGPsN30EZZ(I)GD!Qs2Pd4t=~ZBjiDiwsg4YN2PUI*re+^vq#Xt12CFcrMKB@+
zV4pZ#u6fy2^h3|}2DX>(<k$C+vb>h`)CdWYNsYtXd>K|}hyEh<uJZ_(ec?m{db^6+
z@j_(Ub)oNPq-DwxHI8F`$*xRq<*6z2WmOhJXc;p0os0N!(*3%Gxv@vw2D$jPd$0>@
zE2}TwgEMf}@!+2!+z<>M*~3aQ2uEq%t=O-9<A^&$boZ!Ly=b<#NbZ|(9d8e1M`O2v
zdypM)>CMEjHS-r#4`a#UF(6_Nd00F5x<@qwH){6bqU?WkQJ-9kl14D+e@{v1>9+j#
zemP^mhO}<z=!F@(wjV-n-U#y#+}YZKc|I7_X*={TGy8zkPZPHa%nB~6S$T7g36fLx
z3Wo;;2d~gi&?Jg_m0O8_EaHY<6D|Wu|B;_tf}eAPDAcT$Z{AOM<hH9$NqCk3HSv)z
zyg-J+!ULgHs8c#{h>RRZSb_S`mWM;dnW^gMPQHK^1q1<qIXW57S$C_pv&^<R_vWi=
zgKU-2@TWjp1y$>^GA`MFklj_?-WE`^A73n8v`P~9mZab`oRT<<F@WxRYEDXXUg+#7
zVJ;LvMlQ8S03G!pcIcaXUWUL03BM-Kfs`dAsE9xH?w<S;!z7>Kw4|a-gACfnLDzfb
zetc1L#@+9F*<ghEELHawF(d&J^-ga)fvZh=nzpC2U7!ba=)SmUW_QBM3}Q#tL=kIZ
zuyDZwj%sgm@|QnbOka?5WXc*UyJT<_)RahSr!+9+Hb7MCNs$dS7(aBwg*;RKLI&*t
zysPO{&p_IV!-A443G2YDX<aIyQ`(u|)ccOqim|?%hQ>dbGvtq2ATA<ps+sKIP_jvN
z)-M7zvj081w4fkmGaD;HP)Z`4{%SDW$iT|xc=ae#r|OIfg?(pfSCE~PG|i8vEu%(+
z<q={F{yDpADx82&qYR!UEf^Stc+jv0qmtQ_Gi&;)yQT|qy%}9~$lM9juEaeYzhK!w
za43`y;IYmRKRKQy5Y~Et2W!U7A^j;>)95vZVa--lY|CurpQ8E&vL*YBrfZL03lL;n
z5jJ|IzYt1=#sj2_)<va)wqV0D))q))vB4#+a*BPVkY{bweEZXub!Sm6#_zbX5^=yM
zv|T)%Gy6uWw8|PpX6qh#Twg?D*b$?O;t^yg55j~kQO_zWR3#wVV2-?n;d5|7P~Cvd
zup4lnKn~*Y=Sk{UwC7Hj*^Z$ofPNfmckPEBwdwJtWF`dF??Aechvsv2*+{a5Ns?hV
zeUaBPdJr1uD9fwB?8y;&OO)xTIR%70&y5}zg=5^LlEB3=cT;+)3HkRUJH}F92!|9z
zI)#&y6>D4_;XxpAVHr1#;l}d`g%$3^q!o5=k&#oEKW2qk^TM$E6Uo$}TFov9-nkgK
zkr8v$z#Jl#vtWX+5QXgvn#G2Jy1IySa<2VE1Ol<uU+ly^2I>dk@BS|J*!(A830fvu
zWd&5Co2$9v?n{NY1PS`@w?%Wqq(Aw`!uO{Nki##@HiMsBOrdz2gTOhV<spz!2PKU%
zg|~YdJ|dYrqcQFTo5;h(hdF`DR)n$$H5wx{BTqR`o<eT-TAnT{3TdBU8TXOl@>mHG
zcL9|Jjs`~*G77h=I*tn@As%_o89c`Y|9SwD0}`Iuo9^yBWf>>O@Hq}bsi=R!n#D57
zwCKMdg0ORMZM_Fz&C`M6G%o^q*(0M7gwV2DNEe8QWkIVBcWjy%UI6YHS;l<0p)jKI
zO1As%Ud*+HwqumA&UQw&Cw(&K^y$-}M{}%^gM>05<f;kCQK)f9hzJftff8ma<wX4K
z${zAvZt#ZTS<sm2jA3P%pF^2fN$}SF?9?~GbLyYTxr$!L`+=X~bMroWD>|U$8Y!#;
z6z;zVA?_4GNB8L$f#khOfc6>_@EVag+zJVi!k)>QfGHz|GeM#qcJyOij^Q*4F0iaD
zU6qs5wA>m2m{!7PjeHFkc?`H`SviY-uEm^U2et1ZjT0FfO9Qvh=TqvO0^-xfa-)XB
zXBot@)FsxR(vmyf5aZE)l0MejbD##^H$7gB((iviNcv1pOLpG1eE4D;VcLtZ;zf>-
z$s9TDKR~sgA(Z)w(Z?LmIiZv;uM#$G!9PWyp;w8}9-xH>IFLvRwp{NWL2zDM@1F}3
z43rs~inKG3R@8@~^b09kwcy-TMVhhfg0YkvM{8@gs1V82MbZbz*n|HZXJN`FjPQ&?
zSSXSXddYih+G?zXpEIlvw~yw@$P-6#2Vnm`qYfoTE_@c72od7OH4|le6M}VnX%A^?
z%boswCd#uec?uM{3N*fI$s8b)1sS{ux^1ZT+JF+9I||fFNVA5|pHT?TpZNGmbGWr%
zM7tHQtZ;a*@b){S;}^``Uk-GW2(_1_quH(WDD72_<5dia-Z|j9n)wcjp+rKlA*69U
zBSTrqF&IzC9_)uOl;GQIoIdIv#fl=^mTm=xsn^0ux~CRN_oB(2(bxDv4voY(J3U6c
z?H*j&xY^($&wmcq48nlU65RdK&8)uzXP)*%i8D&rmkc_kX6|Ec|9FlW;Z1;%$Ce@j
zf}E++MORnqldwAG8+=rk&r2l<=*^o@hj26Kn8A39x(Kd6fg2JMrz2>uQd}3T*;`d&
zrUwgX%Y1Z@G=%{X#;cGqUkqM_{SxqfJEyHhv5OI@dw9{_V+r14C~qg3CNv@AkuD-m
z)8o1fZ$*PzR-CLoBzzF69u6P+XfL!Hu|sbewDZ}OZx9_%e5+B10^=`gUk_Zi_RPs<
zRgjFkNDNEeruZjvt{qo^YJhv~<8(dA>KmB}W7GjI!f;YLHu{*w#tvndHUtt{QJ6#-
zWyv<`4OYS%3^d{<<juIh*(F!zbRZioSsUxGrB!0DT-gS_6w1EM;Ii!5lX95$j8>eX
zfgLD02F&nSm$ml&%+nfmnARC@8HX!%1Y;W$oCS09eMI?ZI1>Qb(cOr4oDbqSXB&s9
zNVPm8<?uz)ytIJ$FI(v(-G)?88znq?fPx~;8X6}fIyI=>*BQ3$^e!i-{YIjoTrslT
z7{V!q<u3C^J?<l-bNctp!}lC5tElLSSN2tCN$rIUot6wQnN`ZbR1Uwzm>|r!5Bn8`
zItqY_#gx%w^bd~FlZg5DrnT$c(Mc6;tuTx6<_Hv{Cb>xRAIQ0y@{De<VLtAja$Gm3
zb$c6q1d_4C=tyui5<SoOgSebf2&ZHo9|;f?Xs@Ax*I+}Sn>E?feBk}VtolHC6(;XN
zIwy&;bXdnJs;AYEA9dQIQgGTUu*Zvj+<Mxk6ORUjbJ@M`Zbk}^qa~wbgF=H^DPp&Z
zp}f~j3nDd3=&!?#*ICkwuJ=iv%Q2|a*+4X!@dpL$MsGC5%T_DuAd1#ww)qtNWEfH&
z?Uq0-Ah4J<l3?(O(akO=<e7*J3y%>A>%~Bt`wt#y(Z6USzDNwkvrm&tWma9#km-eF
zDkDp&x&@RPvpjR4uoR*&W`mw%gI=JUQpj_W&d+-#E@_dLakh-2b~&MP9X+U;b8_gQ
zGn@W|=TGs39IW{<{u!5=-`$Igt0Uwh%eZW2PZ<smZmG|6D8B_p|Cs$DDFh4X)rPh2
zIzZ<-Ju^;%@sgD9g*F@@5Q<bS?CHB%B-f}fI0s+g`x9`CNkU0a^(v$V11UQjZ5iJj
zzn@WUwh<AD2mJtZ;sD=78stTJSxowch2CZ}b^$*wqwkt)83nWC0gBR>WZ*U2S`zSS
zj?G4x+=$(a6dpivkkQc-hkq@b_j0<37&k&C1P1iw=unO+c#8=S1;^#0#XMzwLFB@7
zUNe49^u}RJE#R<!9oZBz>g<cM;p}S&pixV?mz7m5WWmSeTPC80&V6zm5T(0sbidOd
za0U;MAaJBT+82FpUcppF6z`plM3J2^a?;+0C_O{c2RGnwaKO(y>J0{9Y#rx72H}cc
zmS7xS7bs7wnv6}OR&xBlK(L!NYc2^K8udfO0c5oRdxYSR&zU$J1mL@P3#8o)-@*bc
zVgFbJyQTC~Ym4kV%Nd1Y&dmmd_o^Ws6u|hnWMXgxajaL5z_GSFtz{oIuWXoCrXwhv
zl&?kTS4R4&DZ}(3DHyeb9zK=X+MooxrbW`dGX-B$9h$oxOs<KZ<4Fx8O4tysB8)6#
z^L_yF{M_k%zaG*32X6tE@&*R!YPRsggB^S)WZ}SMW5=Vs&2P+?lAH`QdCU`lk#)qu
z%I#mBmFkJsN23!;N6K0<K@d@7<U)M6VAjyf>8{3h?pr}c3^#s3za{KoB+8c{Qdc|R
z3RL8Dpftlw$G}cv#G#(#zlrw11dh=_>kx3w%zDCN)QWoD)#g=du^gpk4Vj>`OkTzw
z9a^|3+_VOM+<l`ZLv!)RaDC1N7sjKuzWu^R$9)E(81cl&Li$Wc?3Dc(HWIZ}u5qI;
z!$vl_4j~_ZhSzE6R|3E?MWv2Wc41%_gYTur0Vg#eNl*VG@(>-uN}mB@G==L;eg4Ho
zGC>!Zd=3V`ZqYA7)%P3y<dtDj-wc4@z_0C2C<Uhjw=%w_>}>OQ-o+?N%9grYTb^R_
z_@`Lf`A@(@1b$Yg-Gw36Ecy!l5!(YzG6@FrXi(@M5z;LdV1jJz&y<kiIU=Gn9}R07
z7r`qA5JiHVnP1V{I$g++i6?rJia*QAQAmIQ+Wp;(qZjQ7kd2dndRViUYfJ58AaSP0
zr5Prn26m5)FD+02{mk0>U3qyUeNC3*bIc;$->;{9Eu<DrEczLn{hTmN{6Scm<3MJk
zL0gvYhI~;6ar=n;mR%qMB!u6$RucN$FlzeeVw@dKxL%-C{F?qX%J`aJUO1LnEk~6(
zHZNRzZQZ93c@XUXi<_+Em{i`;RKhOu;%`&!KnnR*2XcJdYFD<S;^k@(!pqoH$6p%>
zTQ$i@=Rd;MW~540w*c{C+j6qG8h}T57%_d069mZ!MDXr7y52Nsho(abZ!&?PqwLDz
zV0*^Tj61^Q@>WI*4I+Tv#w+yeF7uU~>-6MNwEgLH)$OmSXuCYo2Kf(=1n+(Vta4jh
z2>_c=v%{@7U34`1J#fOTLBk4?qXHRQNL%pfz+#3;y@8$IuHCznlahE&&OqZsrH#^8
zy2;P>QGF@3KErcgZD-pN5mPi_m>y^L7`=1`#<~{Ud${cF+kGg6jEh!Ud&8zFL|&80
z$J~To>$qcQ{*D23z6NXZZw6ntwyLea{mp)Z5HekuT50<;sp(t0YSk*#HtjgPpfJ_f
z{%bMGJdrOJx$1_0mV}fPrlNoGzHOXPGPpxIM4ri0sxW|+RM*|N0;Z(z4JmZ<7eCDE
zyq@D)-<8#+v)te~V}Rn8m~o7*j!GbY=Z$XPHF|bsMplEcuQ2Y^$0?L#8kyb?E#nLr
zhi2>eMxxbqQDoTY-o5U0AEFjh(3&&~5hq3xh&axT$zx`b&JTjet{@Yf=x6UEb~k~{
z!>T>rcEYxQt8DN-qYx%o#|g+K{U-|Wh6^FBo`lnt)At4I7AY)K$oTH9L$?P3VqJ65
zKaH(eM2o!L4x2S_M=ad4rgmT3q47%21|eFEcjkl<M%f}-wbSb9yy;g(|Bz{Tj~v|K
z>#eOro7dmDbH@eTHq+y)nRvw8>u=dT_D4H)gC_0Zk}9Q;GlnH_RL{DZ{!MYY0ii%M
z@?_kYn53X`<yvww)Ck>Zv987}-SC+)pdz8y$w%P)(*S{*yY^=+<M<dK|8++|CfdDc
z4||G;T#SzkmY%k^8lJFLf9x$X_+J7!GMtQ$%cKPBlr`f9Cpc|FA)%h;*PUCitE^R^
zI8gS#Qugb~M@nOQybTRr{{{36?B1sTxRcM;o`=y?P@SY-xrbrwzaN4dK*~3G5F&65
znLH-Bvti1WE1yB;;yXD9F*<Gvx)<U{5K&$O<-BR06mX>CX2wM_S&xDdBNem30MmE0
zy{)ywnK3H|qZw`?>)}fzUg+1<lF2*H`1&{eb!?iWm|zDgQuI*92b<f<Q`7xIolW2j
z`t$^*3+-T%ze`^nHq+1ff1BwM1e)3RF#pa<F|_#|DS1^>H@JYw2le9~8HTS&U6>v%
zKEeLH9L0ZKGN|}(Jrm&RJPh>^>tYleU1^5Kjc7;$xNjfZnNwsy#hr?4mT~<|UP4jU
z<dUe^jy2ZSn5Ru&kS_q^gyAV;z{_Aia|%T{ccTy)#;0zT;fPngM7*lK>J)l($BrxW
zI2oheQlJ5i+vNsnOdR472(>7N8jag|2{dgN+NWqp;<h9_Yn0{8v$$RmmMd(<)hGnT
zV@4i+6`@H_LLDuk<a^o8{j>*_#~@Z!h1z=#`)2{a?xi@rOoFm|L?K^XM@yvwv)a+c
zMi0xnIbb9O^WSg`>VHJ^cmPvIOpmKF2@`WGz$2kTR$IYbqhAs+wZUnxpwDV)swvZQ
z167_S2?kzFUPZLZ`RN2;t%NW&8f3Oy^iiFhhF%4Z&7x*nP5%;9U*3ta1Go?eX<>Bu
z<%GDF-`au!0ra>`qCMB90M!qCV7rdo)zxUtJCLVf@taA2X91^*G!VfIkBBTinDIeo
zG_Mc+hZ3Hv2pMjM&^#tS(XB8>&riPyLlGir@SCu1_=z>6c<;nk8s5w78UGT$`rkOU
zr5VpD2bW9C(-4J+t262Rz&|tH3-?CC)S~vk!&E$>B-l)Mil<&bnc)VELQ6&(y`PLc
zV$wzo5d}_lQkM*Ypm{`=m;ys&vIJOY@$AlF7toNYWAcP4VQPCf*mZ=PCY0T@ZKiFl
z0iX<_OBnK6Ti*g^O<}!@>2Xgc$CeX%-^-EEiuu0>Om+XhChf8W7-{1$va^{}$j(BR
zT0sb1a*WPP{!>zrMqMC91SKuH582ufoL1gnqBme!`?v*6V63Vd6C*o?b}{5rXl<Zv
zEd@xXk|GWKn<&2TnGos7iW0t!b~#3OrUTlWG1HIhw=4!fzpt$PK?^aTA$PV2SJ|?p
zHmL;->-!B1m{gBkFv-*k!`H(COv6+j)9#1BV6%q0R$M+D;yh<G8^WmQ-w`HuJ|A5b
z5`v|j63)Y(0~mp%AdpPMxVv}TSK0Qiu00BVzl597Gt4e!1R%7{-&gl6$Lb#LeiC5-
zoF$K-J_h^^!#cF0@@fm5o~90AY)u(vd2Y|HUAvO5U!PF0%uKUB1l%#Q8q~TyYWEhc
zxi)+rAd>>p^AR`<j0jPAq^0x*Ci@eQC;;V1@J<Qogjm}jPa6CrF(}od*xiM0%hQ2j
zOd`Y^C?8D<JA1XH2?5L^1VzM{$ba;;eN@-}hFqja`IdlsZ((f!mPABBQ-)arownP_
zNn+t{gwJu9KYjgrA%QjsL1Kg?k#?u+{!QPxE#l=xMYVx}k&LhL9S+E^Fway)F-mRk
zB@#goSKC2ILZPGnXX{G|fea3eNy7NtDg_};Cg%SUYa$VL0mlTqdQF+p*D#2k&$<<!
zTJ?QJh33wJu$)!3AoNBGGdYdbcPf6Me~_KM1ciNd61vhJL}f<UOnHP3aCRsEgeZ><
zq;|t3*s8#h*$5=h&OvL-6OM!<BWp~m+}^t3v)7_qX{wgnKUrnj$PZ$M`oEg;3|5UX
z#6iDNGRpBBG3cfG>6@4N<tRvvm>}?lya=;~ymy2;=Goha3EiIqC7C#>S+b^)Vq(z<
z3?T9M{EugV6OmVpTm6!0LVIQ(q_vvMZ$DL7%&<g#JZA-T!b{)3=UgKPSA!}^Q!KnH
z%GCT^0FvtS<Xm?&AtN8J-3#KC$!Ts~CKswUq!_|ckhLs@zj-|JJ~UQD-WxEU+1~hK
z@gNjp%ERm_R*-$1D)#TF9g}^gBBt`Iu)dfJ)Ig$C?XCfEsoafPE;4o#CygKimS}N>
z?QSWAfF2O<I|{qjzr^?K1ecZ)kcA2H%Q<40Y_$WT*)Y=tc9D04(k?Uh&)07U_!rYt
zVBe<G3Q`@8`D2C0a*Rz+>@tMapk9G_K-4f{6q^f~5>pTwOoe~DAwy9YXh$Rd{&^`b
zBX<^PbSi6Sp#2OpBM&bb$%)NPakAAmd)0^LNGNRFv?&v6I|&?pjAIoQQ<!ebjT_+K
zq8vxIXJlRFu&&ccdA4>(vz%Vf@tKF%8Ha!>8xX`!6TD~lm<@(MQdoBA-hM-S4(3$1
zT*b6I!mvr;;9>#}77&9CyG=LPkv9;)8?c8P?}UgUMLiQrSkDv|>oWG0FRsPM$D<Ig
zvU7&IqYX5%=!?OL#o%<z>S@T?Q;orE*2_7FvI+ylWtk&5!%(16amd6Jm`LGKpi>p@
zWOWz%nfRut&Z=2dP0&G~R>K5l&rwC&&T#!T`(UOeZBxGCJ-$6@_9VBpfgt0vB8iTK
zE6Yytp0t5fPR0B84LETlq>u$Li+i^Hb(7&bYsObwET8eNxVSDvvS5rCp)uNy>!-u`
zq*UC0x>W4ANH<a$Fsx*L0t@EXuBmd#pNm6FBVUdsMF~@8A@iIlmI3!lg=9yqJRYUW
z`0q7?KWm=&+57-aHd5F}k+Fwf32ArWA>&@$M8fRk=;&cSieC}>gHh>dUxE0q{_$qD
zl8mpfS1)lvts${N4mCEQS*HbT!$NPc^$fe22T{JBdFU`>>yq@_h1OQ~L*N$IhR{}1
z8mk!>wP^o)f*M_$rr}%*a>bVrrmrDq4Dlms1bcr+{v^7Ye}S2ck|DyBm+4Ta)sWWk
zWNch*JV0_0?ooFA*p(24%ZWJCk*%GCmx%-=d<*&vdZL@a=BXrGTU*OyO$J9CYKjP?
zluUU>t1i0`UgT<izjl|+$qjQqB4!?1$`}@u<XS*X)L0Gd7TfG<`jMByqRxfj(gf|I
zQ%IyxW|PsF#rS(%^hZq9c;|mi6)iELX94Ovp=FU$mIhfQJjaUhbI!j(*ob}|I}PtY
zER~QRWWJywk}o5lR2ffmr_<U%{vZhGvk^-Td-J!Y$^gHP$1tNb0_k7DcnZi2UyKrV
zgTxD?x(&yOwXTsnCLmEw!r0O65!J)z1W`%ZlhdfA?8RM-UHS?lzMY(s!hZMzHQ5?H
zG|xfnmf3=HkBG0Qy6za?6f!JkkxZjF<CD{41*m$)Uk>$E&-NR$FfvQGMQ}explV4s
zsxki4nqruX#&>cLN?GWVtE1Vx==_+1iP_IjOy=7B$yDRh5z^QY%aA)5%>Bh~<mx~6
zfCN5&_|tpvV_!zu%p#RkHgm@j#;ZF0-_Y4>u`@Z94=!IGbGK}Op}D|#DK8(A-Uy+^
z0!&Q(G1N_s?A#(u6Jor7Y0}Rf3Y|B?Mp0j{lxx=<2w+#Vl(CVyw2UNO>!QRoUAAxB
z7(u@tj^J|&vU&=gIHRiuTP23x3N2>H9h4J#B~@{2%Tlc<{sL$8Xa1ruEYggL*H2AT
zUqg8G&ZzsAqX{5NxPwUmJUtn85~bmrU)grdbC73DOt4Yf|6+<ghj?W}K{z&OEn`!A
zbs61Nh@;0Fy9E4guV6b&Oh9waxabe2+O5y1`^@AqE7To6KDeHquBGxON@dF-`ekK>
zRXIW$-hbSF$yr^23eV4w*oTbYWkpv3JgI&WDh!|Liz0xxxJ*8@<%tI*H#DRh4H<7v
z@m^xD#_s>ISMVGrMvSt7+c)gk5&v~z<0fVgal+I(QXp`zyH{fR{l^bC+I15|F>44a
zB+6Xls^1Ux#0*skEpBEU%AAe%#0QfKLB>bqF!Zkh+j?Rs5-2Y5_wWx{_HV=!I5HWN
z74#4M1M<jl)fF07^vjnJ9-d>kQ4}N+C~NOiG3rKlCf~Dk#J%!)q5Zh3GiD#{b9yi0
zb6yT~)kt`n^?lDC{0{joybR+yqy7O$lp;vB^({)*H=VY@gz>O;7B;xA=Z&9kOW1GV
zDq1qkxP!&Bk^QFUB&vrR&9%*{q+J+|%W;l{sT5$Dy@UeH&5S+R5o9du3aM#M&MX3-
zLlc5_0*h?%Dtr%|YBm#yf5LS@{AhhJ<zf1$3y=mt>cX`~PARo^OvssyyI9QU^Ryan
z?bv}@E<I<1qWz51fPW{W?)taIc8rfH=zKK705-FyF@Txz51Td-yZvEeMa#2B!d7iO
z=O&JEo<c(R+SB@aOiZmBU9V2>kX|)p|HsGL>4u}Hv1-3<?b{!iE^nk3;cirvz~p1f
zP@>sJ;?{AFAXyoWMp=aXFhFDfK6fDOU!Gp*M=`qoI*y{&$#FFY!{MpaGulId;S4l>
zOS#FfeQ}f9Tf@`byb!HpfS>61J^^s;O&UdaK1RJ5jdA;FCd>N*PoaUq0e5I%usN=B
zMxl7|ayZemNU~ByK)4z6P2@pjv@T`s9Wq}Kufv!&sNee<eSMR|xKXnCBY-{nDLCyQ
z!GbY{&BtszA~G;&O?j9;27)h|5X{4i{B5`9EHM5I2nqo@=c;}=Ny(-$B-W0}9+|;r
zseX29R_&qj^xOjc2b)hoSoK;|Np~I8bP_nm(OUcdpxH_K{sg)yF6f4ex)v?%^hHEJ
z4$M&Ch7rvsb3%=XydS=YG0LUSTZ=NQzz>t@AoZc8+nr`Dy@<(qf7lMz+p;-rB{PS<
z2gc?w`4|?`x1E!{@n$~CSk+@g*^U|TldDIO;e|hF$gn7_Df-VeL-YmY051R>_$!vJ
zK`bk^2~-G5!maT6XabgJ*lRP3O(BdJ?y={88*XnPzNZ0=`g|FOeQ7LUW&}jhf>GmL
z_Tq2j?LCAif$~kohq-@K@!`hFjO}*y3f%svU{I1;!Tz9ca~4v_C#m2jvA?k$WdRo{
zGg;mosJBfKn-0sRM$V&AWG1eUaYfXfp61ziptmVuNW!~I+{WGOtAMtgromQjUg5yC
zO0(bI*u=Xb$2v(VcJ{>wS61ELoxbMWw~&cWSFBsoW6$qhd$ewy&9hZ*L8WVMkH4)N
zTPizISzkqH7KetMy}gIM$Ih;r*Mse@4U6<%3bntIk<<LDeX}h45qn3w@1W<`9&M*>
zMn$$AvMf^O+-{I#K_}Km^yWVh4Gb1sAu&A9(J@wwuojVyA6^1HfQ|KtzJGOnu8l-0
zTGZJm+qyOs%#&_hen7-uP@Y2z;xg*zZjoNPjl9oRa{t>5?k4}#1@gBRXZ5$`oA5{p
z>Qf(b;G&#s53%(1>nln^%aDcFKo-pM(7}JzS-B}oVxe_$K!oQb+V&ryZU2aVui>_B
z+twGMhi!`yv?-TJ`U`RwozM=?e+ZR~Y=KIsbCqpc#xbx8#R9gY`dln59kRmsqhB1*
zCCx+WxQL8!!L8*$IvscInkNHYA?lkn&W$dpf0yq5o$cZ#ly}>Z0~_=7UW8NR;HB`h
zur!*?!UJAaptVSttx1^xe}HG~BD4W7JE-k3r@hK-yh;qTAWu2@pev;ljX@E@9gE<e
zxkcG#u&@+GtSCoE*<Cb=&*CUPvl%%iX(^MhfyDaNBZAzL9ApYW=9A#GelNr<Z2N7H
z6>L_zIA5l5Ka-X9BMm&1C>O1$>RVe5D_|@u^<{Eyw{KT6Be=9%V^OqHYF>W+Cg;VJ
zq6mmnNfkj_2F*Ns(GHBIHC-Y(kOCnt!VdrT3|x1p5ZKc`$@wHcDHuaExEvogypGyq
zU@pnTP_p+pI-25`)VBe}b)sz#^#K8>^`I0S{p&tbDF8=~Mob?gmHPSgA~qr~@~*UW
zJN+?p@E!pKLi=@}4Bwj~B+ActT~~91IEo8z+PZlH61o`xoMWlximk3q^inPRP6g3U
zy+=dO9Y9!?hn`$!Y;3#+d-i;fU3{4BL<7<14Ef_Jj{q`P*vHIBeSH9e2F5(RL5!_M
zVLB=gv&SF6g+e#`)2HQJGxtDZk#gc1NkYLp4~dO!G+|j1E7-`w;*X9&9KJ|wket-n
zLoOJY#`(Sw6+gfYbTe80I5=%{2l0k8wC*Kryw=v%hJ59cu;^4K>tc>(p@lx(E&A?T
z5!azVg}yE;ur4Ri<}*3<Njl~+p5o>`%j9zwpoerd>TN%wps+7)N%I9W;R~E9Ak?Q7
z^{c6EH(D>iQdN<W*Jon4EL6KC9Swiltx)Us8xM`4yy)z8AEQZT9(uyW74Vv4RtRC+
ze*6mMaS!Oci&EVI&d_vSvxWnPAfMEJ*bTU%ud7*IM&6joi>N@7b6%7wmq!uNWTGFS
zPwm@(mG4juzx!HiwzK{H1D(4q4;_kUVkz$lkGy;_+&hp7`l3Ajl3RJ(8yr{m$!jLt
zI8%fcOO-G_f#YTe=SHFffnw4Db@7iMKThBs^^yDma;`5iGjto+sydsjLoZ%LO(Q~B
zRF%m#ze3o+HDRh~qA8ieM7NiKr@?94I*15|v)Q|%B0B@~S|%r%0)>I~>x)o=-9<Zp
zE%&v<7fou;Xg4Jre2l)q&P>*|g^Hj-A;{f*)v6P;b=42U*VLU~rNmQdPFOF?IF6bz
zcGjX<Pz)on^rR95<pn$MavmM^kgU-KGwwqOMAun;AtsPP)*?(;RF}zmh=SNbvOgVP
z1X-7aM~#HTQ+2-^JcaGRJKJ$eS0PKLX?Z9vmhlD$Jtv%)bx>_<B05R5hjcs(PP@{M
z4iiqtc%_V%B{L6gW#UYN3`fj$)wjRR)knsx;q(%4a=~^lFY&c9G<=SuXu=b8l*t*C
zo~G^yvKqYg1ocWg`WA&^i+D+=0^3E4q3$z7{}`AXFxldLWVA1P3VJ~Q^s@4Jl~QmR
zXLO(O<P+EKh8lvJjJ!0HJx&F{22_kASp*G!|2R3R%a7iu-NNKS&nMeXPd#=3^!=0E
zK_{4ejTVA+!e)r-B!gpUp+Odwc4ix3ZBC_Fkj33wY8P7`s*7W?KW|a+&N~E+&z1Dk
zr`)#vR|^ynoI4udQa&aSNu-e2MkZcX+MRe=wX(mxYz*8$3*>`+DJco)h<p_oE7ru=
zPfL?#dBVFLJn)x$mHd=jxdr4gu;QDViWP=?-$R%_sIC(bL+Cg%Ab@^)dCVI4OBR;d
zh@|dIM<oO68Ny^ecYm7e@wZp?gRy(6q0QUE;%$k036ET43rezZA_`0nO({l8zeM9r
zB5B%ufO(O$yCNqz?diN7c_BHQN>GHwdoT%;6g)<()crSjCiH#P4)5DQ6wA8xiq7S3
zH|Amb=UE{<<ZX8aO@BDRv_DUI<)u2=10-sUOBv@Fju1))$Mz~?voO+^4JKp-yCkCZ
z1g-%sc9q;2U2wDtVHc$@9)m@v3W+j_(pFD}J4G)Id&*BTsg8E?=OMnb8&)IHshWc}
zXvmDh2{C($D(Zo2Cf)PtaiQ6#<c8>iaW<IvLfI3K_-dL^2n&<VFv84v@paCm=7u59
zv{%^Iw*7{jU-wf91b5!)mRg`y;NpmqkY{CbTtj|*xSxUu-aINyW8YS_h=s$)l|;&t
zOQH)pzmPoAMxJn-`2|8anY<05*;-B_8S98MMW7B^hFL>a@y&ysDLK}weSLjFko1gQ
zz~rnIyFv$x<3d%F+W`Ps#Ky@ZNUW{z02fvYv3D~_$L3fU6N#5E`ngQP*7$^xu+_H<
zVe6yxbUs{~<ZDQ{O-Y2!Cn5!0<b2!>XIo3n+=z)YT=@a)FkrwAnPmJC{3*#{*bs|x
z8$1tAQ?5vJwFl$MtZX?uC!Z$}oEg*PWtc3nRh7rby_9ML#ZYhwRgqwjS7Ak7#Zd<*
zaIqkKzMQi!D(mW+^Ub$1d7UJ{z#_87+&{7~WA<%zwIhOY`WqoRP(A?Fp?tFG8sMA-
zBg+3-pnv==<H$nh8t@RH!Ps<2PLX?<d>D1l;8S!l3x{wHQx|S{;1JHv9>1Ji#+|a$
z|4LR?f1n$%@W=|YN);LVi+QScr6K)QbhdSSzutZa_=cQDXNYV1W}@+T8u-NoMC<=y
z@5{rfT-&!zMH4EeQqibEX{Ivmq;{i*EkmZQG>1ZFi)yFT+oVXwYD=jo4Q5M&N`wZf
zjLDc;WMQ$s^M2McgvI*fJH9`@V|N^W?{U2E&*ORS`?|0Dy3Xso&g;H#*zMb-g<$=f
z_O(GSd9fr(|N5Of<p6qzWJ|U}b@PZ`;8e=h<{s~!2CgNYjuacW=F~`8Oa(sH`A*`2
zkL(tBMM8W?P(gzLO@9Uc#pN+3_fcjF4+O04eX)JVkv#>y^@vI%l2B+;bkww2iCt$v
z9;@({+U@sbjpGJ&&ns6J5HziR`unnaRK9-ut`Wsa*}Fp_=94htohGe$Yl+a>E4{DI
z)DfWm5)HMxi5TRZg>i|S*gJJXki66zkElFWTO-6bgR`!I>{<!Milwe2D9>j<vU`$x
z>J{i6AIhwdwPQr%R`&8N8jq}dhSmjD@8qJx1nATsM#Lg){tSG?WreOE&N>(|z!NiE
z&~cD->A2cL<o!sb0;g)U?KFcLovXt&C=|&0yj?bR(^tx#&7XD8E;nQDD6!|euiNdp
zJqEkve<dIV0ivJ%kK3;6u0e`6W032`VB*Ys`J@e?M-n>DI+YzShvJhsuTuYlxpOXa
z-JiQbtF~1e%H<`7+asZVbJSDb$ZT!LpTDvvAceG`fvc=+AT(^uyuhHK#9Egzl;7%U
zc9Sn97ZF!ny5Zx;k3_s{AcVlFbX#rasb5LiP^oB5&`2kWfd|y4a(R(MBoF|b>_Cd_
zhO3v}?i#kX9ZtD8yu>34T`^NGuWhgz@YnY3F(^sw7Mx5)l)bFSCy;F39EDuW&dx77
zBZ39Bv4If1iUsQ@!nfsP#Knjs|L9}U9ob2tR3LJd5RZV=g3V=vAv1;zSMheh%oqVL
zRVc11#Kgo*7<$r;i_z|a(E?o)J&@L6v;>qv7Iw+yvwun@grrFmy-xsXve`IbrAF`m
zWd!-~8`rN-zOP1ykj#cP?Ta!(Jj){rk_-c@9OE0~s+W!sm+Jj<M4&*t-`EWUa7eh&
z@%tu{86HtYt}i4cBzQ;wu!SmgzFJO|SwcMcEroC0(rt011jh%B6v#LQnH7+ar7pE6
z9U5~&WGx{NCz|~$8(3+o;L1OH|CVRFIUz%GJqa102+qcJH0JorT7n0y8MH+?=jX?a
z3@lEjor?&Uv-hVGN&u#RU|N8}wDw%kU3;o#adb=!z2TmDR3X18m5=r`jY$}lOT`Sg
z(LwXOp<ZxDPAczQdpcoQS|0Olo|TQw@}c6gT-G!IM~)b)+^6owY9fJj61TJ!=3Of1
zR04TN$>{AGmpxp*hH;&!Rt~^~w$GiGwrakT_xAh7f8?8#U_`maP?=p^HYL80IM}<H
z=+R|6SPlYteQTX&F;@E=RJi=jtkGf@c`$3qD?Sr0Zt>eX#I-2CP3VkkJIsFWG}cZt
z6lh&%`{8NN$GW`c@+pcRiHi2-&BH%G%!3Td&8h89;LyxUS<A5a=6(4Kf&no;eE5O;
zLl|o*d(G%>uambtqjvb!|D+@%Q*<%SMiB5LfT5o5h_MztD$Id#J=k(`_!LybJ6(y%
zW;_K=uY#G1oOVVae!Y4D5oN+|1y!sf8|~A3J3NWR=`BtR8c_=nl~z%z*{=>cRaQ=>
zaGQD;n`f{(QPD~H?^a9C8KU>V<e!;?ILPcKiM0($*%%l&J~9q~|5;5jZcb4Mqwol}
zx1<`a?FbG0|LAxIV{KiMKFkp8eNAlzcxs44$8&$uaYQ==Gzb~#F?|E1J~~~r>ul7$
zybO-$=I0ip^jVkVn#N^=8G6t~n>j=e!&N)KC4X&oQZ|8z6_#)yVn>^NOD^^*`U-6g
z&l5(gz_(^czU#F3ozs5poa^(}nN&2hKPzEAg37WQ4Q@|a4a}hQ-3`Ih3}|~|q!Yp%
z&ZTF3`Ljn9#J%J!v@N0GS9Ex(^aRFaE?exjx3#$eJQY`KCbQ{hr*A!~wN|sZ9V0f|
zi!?&vz~}taf<{!9-P5gntI2`m>x^#Oz0%;VGnKu4RB6unp1Z2Dqf`_uW^j8Gd5<I_
z5OV*LG}I>@9AY~6*=22Ul0I5j$UZkNL>dlR{KhC(buN3PI2m9lfen7VKv3QJO#*){
z&@3kO=t$otqMfr{&}dB;lkI}PKf7R$7%%xK8H3Pi_E+B{phdI2P^jrfu)Wdd-t;L!
zn@ikh@h^q|GU)qswr+h0ztg!rftQdbql^gU3*k^xB4IB*uJlyzSLpE@>FB}6tf9~E
zG|^Jw^LAXvmGjRHA^#Lesx7jZM40U8^;8#BrsDjbF{p(Vm6tujB#6$~$U>q8URtGT
zCsbjT4H7STc@cBe9w@*iI2r$#RA60mU%2oKq98b@^GRytNuW0j2sJ_2urFYY2!&+!
z@*V^_ATIWECjQv~m+kzVz*_++2dQCSjaGveLKYPw7>pNcTms_bJ03W<HiKIYz#1Bv
zoql%Gj32U=m@hmmJAtTHHx7vR<g&Co9zn!9;gibcweTwl@;NtgYp88N3jUh`p+>9D
zj<?HC;Lnp}=y5rPXE#sx7z>ecV;duFZS=<ey{SEZtPh*)w7@HAXX>)@y9qhTKQit(
z>E~N;@z#n*I0oC$`kTSFY5l8NOi4iBOCBJppPww?#>HS)gxNwx^Ud!DTYtmEaRJ_H
z{?6Cu{j0X)4Ey-_IO*ok<8p3FM@8wIp>ZFlc*BOb;I<9zh(Fu5oY5qnF*?C%9fr53
z4IWQHsy!n|k+>kXMAiK_0gLu-AwWXAD`?hur|j*3suk#enhJA{4IK6r;yislY7eHS
znErvRI-tYXrpvomC1qo*<A}&zWTuhh*8G!PR<+bX+rV1O_?nTfZq(in(>v|8dQ_%I
z6a^8#TEK>g88@L`*n9q#9>cL&o$~HB>vZ>|e+tR~M5J!64nc{fB^0-CF%|n9XbM75
zQoyGbh>lN(y-^QQhFZVtoy!;?nw}|i_9z#Nb=1KC6yvCXZeeWvUvhBOsNX=-(7#G+
z)~s0=^--~3VD%rpJGT`ofM!VxPYBT>0SUkJ9t@c2DQii1dh0He-u_^rvuC-yixR2l
z;KnshR)u&~=h~)802To{w;aQ2to#<>l`wkaa~}_4f|&rrb;g8@tDTE-V}^f<0p>oq
z#$19l^J^QSvsbyi2onO1pjr;8Xx_hXA8dA4=Mi*$MqAKGp~PD_C~@9!#eH1%r!ItJ
zx3vLK2<vLMb@vhiQrfZN?Ux8vyjtii1;F>jF^)TK$8byCbTSFL<K`&b?yBm&o%G(f
z6<$9*w{C#z4ytn1^BTP-MxlOjG>9YOrs~X9VrLb~T7vHe&!lYN54jhj8htcjK9}1k
zUAQBd{n*XZHf?>4qa0OyHQ{X%Y6ST|`Co11Vkmh!bY~|G4UK&Mgx`V#=FHEsmV|(p
z|FsZW0zmg0n_PBq$<42&$>~SVLwP%B3lUZCILYT_2ej<0GH3lV7qpqa&Qq?o(#4QV
zI(sdRW3m4HNR6zmtj2WQ8%>+g<3e;m5)LM@wb;ys>j!fQF`V{-X$GTpeZ@y+40WVV
zb=S}nNOqWX9^4s0_ZwNtE$(Xmj6ro<M%P1)^L<C?R%$(fv9k&eV37XB^*&49Y75$5
z;9{;So=CqCg4mAO)IBh;8THj8ip^24w^F~jF*pKr-aGtpTvEvud5C;LKhXh>V^>#5
zxCR}U&-E1!3v4p9v<HN)#$@zr5B@lc_0SW$TS2!)9Flyriq3NGJj9ZtK6*qEW$>b+
zB4vX!zP1IYKu=T|n;yd@HmLW)?A3wxBO#;Vv*cjs!rHI<L)m*dMrjF3PJ!_9t?B{w
z%ed@Kt0^RD30e><ek&}j`p1t29T$ha#y3<3pyx9xegyEMc**o~ZtrmeoK5Y&{u0U^
zgr=&z&J7whpl+ubtF09+miGB7g%AY!^}xlE!j5}}Dd}{OyT;hco?jb8)ezbv>-^87
zUVB7Qu<=;ZXVJ7p&7a%dV}$$*qsm+}1-kjzg!!|M%UWWx)_62ZYi7!DaAT*LGMd~z
zDh6%xXqFieGV(X8J-d05k~bwnCt7&Z$E?oVg~{a(+av-NNWIaI4i~NM*yMbjdPHr6
zZW|jL8=Q5{W8<>DP~%h?d0v%^haCLJTrcddUZBF}Zl;@mtD)K>N`I~bY1^)Rg?gLg
z6qztCpCfVd$&)7lDF;jxBl4bIW7^fhw15<hJzQW>eTWSh=ZU<Y&Bb8;`3{c7oFQaH
zNoR<E;{&8_p<J|adF7*opjQ|tyI)%5wE~w%3@}AX;wmfD-E7aUP*w4cp<rS%M61Ty
zMTIK<_TsMtD}Z7Vx8WuPtsw}}bX=wh7k-4vT9UDHD5vFpFnEQynJ9lZEmM=`VtAYO
z7(z;(ifRi;3s#4(wD`BVJSiK~WyVI9GB8K$#|J*AuGw~%)MW*Kbdl$FfwVyg_cHHo
z@4SS@UvRct(Abj4Ld4r(l}s307?I3aq?X6+ZCrqNk!Hz^Q&`cnXGJ@5nEaZCR%*XV
z3+<>^5SFma8&JBCo85eX%rDs90@Z_oshf7(isR$sX$vASH7TL!B?f1Eyxrv_8JNc9
zK+VuWp9(MpfF7Nza;2G(q>sQGJ@Mlrzt}=1)=45zo=bE)LkH0<nN5>90GBA5;1+f`
z7}3?zs@^e}RKT3Rfy1@w38~E3iR%|NsUflRuXKwL;<p*8c(>F%7sFTDN2iaOXRHQZ
zgS<ffd~SwUMcJ-B0@%(gJ}d)F?|4YS3!5#(1ccE6f~5}JmNbu27##+kSO|1?V!X#2
zzI4LGmp8Lv;y#}P$rsX{ersd*FoECc=g!cW7zg6S9L~u4+dhlnYvva8Tqh0R#P9l#
zXJQoL{EnMdpcyp|Q1XUHK4lau&@ltNaywt^3NG=+gGge!VWQnlxBg$TPPka=41tSP
zggC6ko;3xD)9a@%Bw`BBk9D`(*$p5@QHb$ef*3aqR#W8?Ky6wH=)@26IS2Ynr(9i}
z+;NpK);2jWXhdr{c;F3I`{!HaCjjkF)8=dDcI6jf_Ql2UK%(2w@f5|s;sa8U-ORQy
zJ%ln!laLyV+l%N!VC3WNt%B#oQ8xT7z;#?>2zO~jrw+8yW|V+!stp79N{zT!je9DF
zHM<QRJL1;H?w;4^5?$631KKI|e6LXiSf&N$aXU$?poMfmH(gLn$++V*J8o|CPa(?m
zBKlj;NGZT{vY4*U++xB%31MGE>Cant?`A*R>(OzfYoe5ty<5PIBwa2aCTPsMY!B(+
zq0%#ixP8;dt`>}erg?$wsO_JHWl7mb2MiIAlz@8XKS_cExY=DTk)$QLnfyqz=(5gC
zQRc!qSxck^7+ayQsXY>0qU<LJefoIN<``nYai_qyLt{ZFsKB}qAZtmaB=t8yJ!`^C
zQ8q!bm-Ja<jz<#J^`&wRF!7BW&pSS+w1EvEGqynD;KOHEHZ`%AE9o;C_l0*I8oTAn
z;{Bc-YnvX^i5_1dv(UW-2XHC5tQe#dzH^OkI!3y5`s)kjIkJ|J^mL6m333Dmr+6>j
zw28|lf5`>W0BJhh{trqnP_XaZ7L!kqGgAS&i4OcmtKsN}2KV1MU+o&V;AI?oMN|M`
z0P2AbE{wF}iWR?S=(xZJ1#e-qNkHd!d{yJ;c8V6DrO9A)J=okGMHjV4_T?S$aibXW
z0_XB+o33#CqKneZd|^N%-7+(ojd2Y-tHP7#4#`%{M%K4#n8b80UO$LTT_EkgN(KmV
z^XN1>8{(wc4ae<^I2f`R2SYwHj!jJLr4Y3?1<mA!C^Oh5Sb#9VF*J&mQ2IF_1*$cW
zD_w3;SwmRH3pd^8_D%k@=3v@jCF2?L@|ZE}2UF=F6TB$KgGi@|>vA{PlRUP&uLp5l
z?!ptDE|)MHz$g|!`DVU~H`LJ@)32!uvaX>>vt-PIO|QAV)4xQ06P>X@Z=rs0fl1r7
zz?;-V;4L8l{Q~J%IAR90&we{cWA3oAiD*aef(pmLr=qvQBEtf?kM}$diLvJmaVZ3{
zz?TNJK38QvC({@~KMrHkw`pq|K_{C`p^0{AG;4>g)+Rhq^4`C9?^xXGnqEWM=(ED#
zfeGBaqZT{PFntAGu^$M`zoPr8{T($bb^KEx3OL8Y!2)6+W8^n6xUIwe%B%>nu?zma
ztALxP^`|P!a<13Nc8vee2@M`mWCXRvUJJAjR>1Gsx<2FPZ6Dn;jElb82&W%a?_A5_
z)QC^Q)_edY2boMt8Z<f|V)^s82hQLYg4uz{UzYge1u8$7p0urff!tH^knPb{d-Jaw
zgWWc*seGpEMQ->;VH~91K^c&QSel>J7L*X%mnN|Yn2!^<(9oN&EOQTxjdeuB<G>*j
zL%1EMgP>0%lrxPx2o@g_)b{jO;=&E^d!7$d5ULUe&@pc70QZwz((w_5E{3xbBd3bk
zw4-D{eCc~xOZ3Y?E?GHe=upfaL<RfiV2R;eUS-+=%tJJ7s>gA|R8i_ZHp!i>PoggF
zOaU^seoJ5^w{_Jb<A+e&FmxCh?0fjl(__LI{YkCvjKQnxcZS}+J<~#V9vi6~Gjo$Z
zH;V}&L#|AlGCzL?MTyNl*sREuwS-s(8lZ;ii!ih3tv+@F8eE7xE7mmQ10Z?(Svc<0
z!Zale%w#unT(@9waOi^xz@d$=sBy!cHeE!?Wb-F&@CiFgpWl|FGYv2!CQz1@O~k&&
zMn!#@=W4>uM5ey=UHUN1ek4%Dn0Dj%r+QvMiUC?UTpe7<5LqEM#jzLs?c!fZ2zJMv
zqPN7*JB~}ugjA0xu<(GJ`OoD?90e}s`Vlj@VYO>Dw{PFBD;;v{))fWwTxd~?jaOvX
z=~u|8JzeIDS|OP;GXmlJ#CfX%ZMk^@Whi->wtfTEA^-~lZB<M#FlawEgWqO?O9Am$
z0aXTM%$oC`9(ySwA<xCaf=ZD6%`7+H?R#hy<Qq{>Xsq}$rG1+xiKu!n2i8hB@2VAH
zx({Ctb!8JDdx;O~i2#`CfGK_($y{3Wn1N-l-o=l$PGjz@LV!Dz>7okpUO^#uKQ5=|
z{ZBIP>#oeOu?u!WNd>V1W|F(M@0pWa%}z52q+r-zRhcX))&R4VlG#+j&3^sKIQCJo
zP-<DSXps_7Be3%&+dZBb%3V&s)a~<w83dZHcNxJgC9j7!RQ22FHR=??cgSR#zPql0
zjfm3U*ly2CWqSwZe)=>za?KbFK?9ZRfW6uvZdvffHgX}{6&WV2;s_akjJ<9~^K5&N
zS!$p>=PBC5Co=Z7Vs7<<WgKQ>FZtsdqH3!<ALE_ID43DJ^i!yTPtwlzCn+8=^5(;|
zl*b90r3ez1+13QU;}7D7;jgjB9GK>t`DmKNgs|EZCr*Hpzf_Z^VQt+2WnC#LDbg*~
zZfj@1UaPGp3X*_G1ywIApx`C7X#ux=_!qI;zK^$5N;jbI=lf6kqsNaM$fysr=EF<*
zw?(se2tt5AFnMDXXNU4njvvP&yaLq|M(7+pG4Rp|Znpdbh)DIZ+!x@XiM8HAPf-NQ
z7f{MW-Y{2_R*hCK2%(4Xd3U|IxLAL#EXT?dGMl&jY9UkNMs8G`R5N=z&gOwVYBRaT
zjUfxX*T>U}USK)szXHgFK{AmQb_#<`{h)f|$&+6XRcTFiMet4SS4Vx}x-Zp4g@PIB
zgbU=#7!DJ&ml)zcAwZ!GSM%UMm`{X$&k1d+(|z79QBaVF5Gz{3(S<F|`WLAHR8i*W
z+&4x;uJ`&??KUXiY*K^a6srNRYpR+ASUekq8*KW(OJle#S|w8Kw5{d^q-XPNvr$HF
zZv4KC5NtdVD5O0;i@r!DGH+*l6`VGSy(Ys0AHrXW!EqN<-6pRa{vy)l{f*ukbk`oC
zkcIlw{mkuy$n0bD)*<kLFeBKuGjf)q+v@Xi1L4m>qzwGR)tv8;(F+KUCoAC^a07Rw
z1B3o@cp$0BU2tnBL#m+BkJB+IGR+)XjJVdz6URNnK&G6kswzX44LWr0p`imKECX8i
zWA2v9U<vSl5lfsNJ!hYpS!&T?4h8tpaW$EZYYJP)EG)EE4Rm&N27NbW@QgL~Bh^wS
zZQ^W`e0WED;qft6M~*;-^s#n`FeKyBR_`rX2QBai2qC{9?7_0zHBXZOtkY7FW`5uR
zU=_q_6a!XD!Bx%o;_F7JgAQ~y^oMrYAN)aU%>_;_ZKto+YIL>_il0JBKhwYlM8W)9
zrw7XVT0C<fHM_YjASXLJ7M6_NAsYP>Gk|TpIwwuToK$GBw9(O;NI1!h#>)&e6L3{@
z@@#A)p&p_VSEG-S(4ILau=aAUkoduNmR`ZSF2tHym|+u-fa=_JU5ld2+Et$)+@P^)
zG3d*wt*Jz6+_;VC4pwR%;f_{~HjIu)C5_lP2VeGrD&jJtcsc^W-FDS5EHRennymta
ze#+%9tsv8y=f+foKX1wziPA09532y{4XLE9Kp6rPcQI8{wMDYHo0+`wrvB1*@|OTU
z5Mpxk9}Gx$9Y9COw7}!IIS;*O(+>+aby>}OKvglLlero#D}W%~hDmvToI&O(B54eu
zwV{HVhN0#SV1o}~(ll`6<C2z?)0t<M9SKd>a{>^|Z>tcagVKljg%~p6L3py$5;n+)
z2Uk0mEO!l{kzuh@1lIcCi@|WCSi7R8ordZ-*CR=yfldRBA08h=hf>jjb0yel#UZq9
zzx7v+fh)n7L(!p#3nb)WD#!gH^N=7G3luofc@AIA<NE0?ie+q`^Pw4?ZK{Fw5Kx3)
z1G9L8_kwAD8%u7hZ#FWbLmd`PA`=xY5>og<qK%9kBxLfC#%hCyRjyI50^9w-TD2q)
zD+>N)6BA!t>of${_fH`3OU55xyY>w<>Ex}h*CA{OCxjDZ4_g~H9!|misE?Y=67;Ls
z<O^q2V^IAnp%-c}Xb;Y1<<@y@gbLmvMEd+$5mU3sEU)g{nn2#sIBJm4^*V67pw7@(
z?{NT`F;Z8PwS;Kf>$l6SKq!iCDe(H9MjrZ(&MTaBkdFddzXi17|7sw)-zAEj8jy&c
zx{RTELe4G(%~1BHsID*8a1OXS+HE<lvH}-c5MLRgbsQ;iJcv-5^y4O=w2)!F2M)j?
zic#M_*cIPC(LBCsmz|xRrs@NXS87XRF@f(8Hg2zPt0uC(0JQ3A7F4Hb`j~NfY39~}
zL&uyC7EQ!FZnOiCTMIbL7!}w+y@QC=NRdlVI7JZ6+pwCtouY&r+pB$vxVoK4NKH+}
zo~QaKR3q+A>$+#~dcQ^yGPO7&&N5K_O!H+wfK5#ey;kxTEI_5S*VR~ItSE%~%915Z
zQemX%IMkfrzA#ns1|}(B&?83s{?cfF?4VLiVZn480oDJ-Lz}`sg?}@y(%#IAe^Sjn
zb>rzCd(F(^ainl<dtuzx=0J?JWmJe6VK>R>HDJ5t$bHo4ZfY(C*5G*zbQXv!a7W*>
zpb2iB-d{I=^YH|@{(E|lL-2~A0~2?QOu-w&S>RQKS?8ch{^0vuF_q99WkZL`yAziG
zdU?uYe4zUp{t1)Qp9JlhHDmTi&l`b(qaN*<Dn5F^{K-M91}wX<=kB(ITer%##a=%a
zXg%a$@TR{9uU)P7eWm_JljUc>?{|!7TrvBvxH(Q*w`rL7F>Rg5q0j0g94d00R03Ff
z({|guQnjPIl{6NPUsiZ-5Tv1!p+vUrJTxKTwViKA_};j217m=#O+`;aeOlau@*o&j
zj2YiY@nbRqlAEz_M6&5Lnt#cfi+6P&-n=%!{YZ@8%`)JX@`kac=Z12g>KTQZs<ecx
z6haFOP2uN<G?Ga;b;jdPIYTDW0zd1C4rAslfi%D6+FDwkrtrQ8_PCnlL|d=9SZHUM
z`B-(invj5;n~0FkiTGcS!@%K~)9~rAiAjO5D#LZQrJCBW=`L3)h&wVizaD^Djbw7H
z2c$fWM-Km7-r#<{^hnJ2@85@2Zd#9jnwnmwc0WGOq@9JLrH<%uk^AH2<mA?G+$d_h
zoKWXJU0y0P-C?BK`d_3Q<i5hYv_}#mWFGArXqMi8C*telbEn)M<wU=YyKQYFPYrOt
zaN$CtWLh{NkuP6#X*coM$Zx|%ECk=CqzttkLRvoBplcd|H^no1f75*H(+L=Y;AQH4
zvkZ5MRI@lqC~RVKuFnvis8lK@$yw0QU`&z56~7}P>m_3QY`m)+w|A{MvC=p!Gwb1Q
z&5Kj9rTF|uS(S@6B!??G{JN{UhqI=8z6z_<xbjzz3=R%9fGdKs;)|V>D{>gwx_+H&
zK7Otc0a^0{LjJ0>_>d5}<Yj&v)rUxcZ?Nv-)U=<`p^~FVGp}JxTmox^$lbM?xR^hG
z{=B7n*y`1*v8g3*#*2H$4I+z*c8$+IrMvY83*qO*R7^{abevp++h>OR^{h%<a1UTf
z2sh`%x22?{VAfJ=*)d9fw4N>`a3={6q(XjcKCt<GJkysvb2#Z6?`BKM^8`wYi+@C?
z{&YNi_;AM_Z4CA>h4lWzhYvSm-U5+obs1HO5#2x1)6-q1HGc@!j9DNo1TlW>G?W~=
z*;4+3a~(TW8IhNp?zRb4#=kL>Vum{$rQ!$~0p;;n)4Pg4fBqy>U|brV5Rue0>U#@5
zBa1qfC?Rw+CyN0Z*rM6)CP>>6LU-9B%i=TNF`S${P=q|jMZrGIHgR@wNeP{-Cx(yp
z{_V5p3&`t7ZX)N%Z)5Pa0q)!GEMB_QyTr~QbI6b(X9rVaIQNzOP#%S3&xD$`D#V+0
zmuc@|Ns7dhCKHu}Zd<QyFEkNaW4m_U#3p@xbQ_z5vi;n6=gktQjgr0DV;ns%3s1B)
zPLPU*G)T1@>tC3=1u^%PM9t#Y^kxjGeTtv?r_|JhLa(xq(XzCkQ`hL-+ct8D#PjFR
z{Wi)bhlhmBHHxob;t;{)6+T^4cYFErr6&**NZJ$d*o}3+&nLeTfjY%XUC+H&NyK<f
z>>A<uT%3x#zUkdMMfPJeKltR<-{d#XhUJSfOI0iSvZgyK0uFJ8n<aA0Y!FI!*&?Ur
zN9ybeLi*A~;U4*er+qM)U=PW=)>75#zMgt$a?P`UhG=P->V7J-8mNCSYTUhT`wt&p
zz&TFi=i%!yn|>0EDa^5Gws7npKHlC<4`_*f?bD7``2^bXEUK@s2mL%7Hi{YCb8^fg
zX1Egp?ltVGR-yMao$EXVb1zsg^J{8ql5Ww4jUMG@qw{QqQs2JKu=XplyEygUwmaB-
z5P5d#Zqv$u8j;^J>P7Qs`7O2FWnyA-JHi+vP!K(qvm*L>>L#JKbR;Qoqdqswpd#=t
zT_6!8LXW9QKba3+O|)df@^L^Vi=C<+@?2~T&Fzx6I~UN4`dZV;HL^b%>@_x9#pGa$
z8?jPyDecwuM4tC6fts@7Qqtyp(8|i{c7z_>*K!m%HJOrddtASMlaT0<+JeN8kZ`$j
zt7Z7K4RV_v>D^0|%!_ldG04RB_%Ef-)VzeSwjW`lwr&ki#hfi})86&9Q5yM{Ho$em
z<vuL(yK-eusrn2Q#XZd;wo4=;2~f6RujwntvUNt|YCQL0IT5RG|B;6~Sp?!}xZI#u
z16@{COtV(R&vZ(7njw&qmX^uhC@}f=?s2=GV^)pc*N_kp4D5=j$xaCc4i7d4FU&vq
zF<7PHHa{!6uhZn7aK{QXl8_KYgCy!%@<?s|k|mx~*Z7xM8#v@s4;0_FSwb*(S=<k;
zFg+ceF*RX*dC@?T*c0}g)8yzF8zpw_&kv8}l7Z2G5~L>jxhevOS~!3HgO3hD*REZI
zQ0~r}*^zzt<v@`G&nha2d~25-I&<cX9407x=`dGx?$BMcYP&Rtfw3A_{r%f+hxK`d
zOuoMtT<$r^3pRYf(5c68BcH=f*F|K_aW4QNqmFG%#tmf+_PANs#mNF2A{eEn({Sx#
zmwbpJ!ObD>>J~LgxMqGpTK7{)^WsjFeAvDO5u%osb=muAV&f@!q-|~r@wNXG%oTPj
zRhPD(D__vJ$)8ZPsnXZd()G;2?Ko`CrZfNh-?TLsiNBwyNEzl`eWJ809NS2w57Giu
zS^9>r`nJtsnOuXW_IErFF;d8O>`rL$LwDzgwA!G)5BpqcT)Wl=s1e3|C>ts!rtm;*
z{(=SJi~?OxEv?6lA|=aqt#Z$(b50dI1arT~V#=MNj*jhF`MwXmh263v6*|P>n_>Jm
z*{N7NOKPW3lbfhW$;^uzkF^IE5TwKKk3LtweftKxDvY9E?c4T?%)zR&^Os$-VA-;-
z>7p7jIU7G%;S(g~4eJwz6aQ@NZHv+I%K86G$;$opq@-aVV%(=B#vMWAle&dEGX%8H
zBe&Pqx@@W&{9$j`5Vzskv!lDuvah^11Uo|0m|R1#DJ~o+(%Koym-cA{7XP&SJt*NU
zd89rc=ZDd}aN$B(O@Gdt<}Y)2E|ohke^~NrmAVThEl2x#(`amPbm&i|0AJtn6DOLP
zxSH(T36N(ear+e&cR4S+p#@c!EPnGXRng_4s069xWm@&=q9U)3S9Ok+OJ20W9Gs*R
zD#uOmiBPQ3*gPKSfZ%G%=}%8dP0d9aUNHBHWAVZOJAV1{zCZslJ|7qSnu|NFtR8i`
zeKdB}viB)puL*rzt^Cz1r_0JBS~W~aj_%W^=r^PWs2X)fQAlN{vrm<LMH_o|h(#g<
z6rf5Q{xRkpEQ-c(quKr))@LX&dETWdtIr~`#SMi*>EcJ~4-o7oX3D#p)b6*-r;cll
zV_dmG^|+d|-z&{&Ekm45`*tYvf&O>i%}7S@Dx5&qOWx>L3T3AG!S+PVFjhUP?50oX
zu0OK+JpVGX8+~UG&j&pea-KQ@C^A?_*@ALXI&aD`E{MNcZlYt&(>r#D`hH2~EyFNv
zi9~b{1Im*6dTmcc*50e0rgqKsdokmPOn7%L{fS8e`4J<$<{Cs_&9v#$h+b=2p(NS;
zN=`~jQc5e2hRdj7fAz@&M~?i$Fyt~98}Wda?2^#Bb$wqXg~7`J1E-{{kI2fwh(~ql
z6%W94A=^`cvhXjqDTzSUWg;%+Z<4fH$*{Km9vOwUlsn~I<6Q44M512&DN-FE{;?mk
zjw6{gcmC<4&C=@Iqt-q11&jc1)Gg4p^H*(Se23p7xx<~$%X4eb@|zDEooA!(5!Uxr
za{f-c-&e@7+N!=>qVmY{4IjegGy_yBU8(oHf0gv+3aX@;IXW_d6J-_-iR$wh6^@I>
zY$0+xA=JcmK7I5Iy?dVuPLq2ew7jXwE{)oL*UMGbVGcPi5qxB>dsv3#;Xe2OPjy0y
z50^v63ec+<UZennKg*89z>BCjer9_S<7v$Rn}0r)R<eQM*#3sceJ|(xM~H-p!_&O@
zZ^k}yG7-efC5MDWpNI%y$KsoDpq!J|87>I?iLljS7~Qwe>q9Lx7&zTFEh3KtaQ6ms
zu?G(xOkGo_6juI~^NvefN>{5c3#>G1`I>2%B~LOtu37Em^HM>X<)bv-RUUcu>Q!&2
z$knt6$N=PL^k)^XUbSi>=(+n_M)e)t9)A8dHMJ@$LOwYhQQ@LRA72P0zJC20b77rN
zCCV>SRjno_iU|X*Qgb(a;2*|1YZLqZDq!h8!yLKU?a|YzQ>TcR2b+PNg@7n1C`nD;
zm@FPS=~s1L-oD44?!Vv-C#X6zuhKU6^I%QHbJhkWcK%b7AOkRt@+HtE1RDh<X#Xe*
zED(w`b1w68-m7*?wLV>agJXZsq=4{{S;^s`zsLN(O_Ed=wwVkd6MoT@#!N}W-L$0E
zy6;5|53je6yFMd2r;5R_#X&-CU;FA!Q(cvoiiZ;;scSy-6Otzqn2Cwhk>Pd2oc8wp
z<##uAji81NTii8*%1Ecxj3$oMbq{gK{KNgHuCX-w+P9m(lK@4oH{t6)SRua{<|IPY
zw_<Y4HcPmPPN+-JkU_3&`}bLB=jVHT$En4zv9H<M{dneC1lCtd^O!nz!i0lV{t>)4
z8t^@01N(f+B`Mk9TFVl?^eZN}A#@eJ@AN}$+Z|5C^d;4Axs+a@PK24k#i@4wnbvXl
z@8^4;CSx<_C6Is_elzHYJ-VfEd+n*0vgq4`2*nwO&2Tr1SgxXyn=YF8;>C;ldMhM~
z8~r$6UfOsyGj^BSX4X?_<{uyOT>3uhE4(>g9+wa<V7R$et5yLe^EB1c!uhf<raoP+
zIc2xX!oceoidR)Gp4N8+HmDr6AyeI(d+ouYU0gp&0_`wYu3nwxUH~!$+lbV3h(~&N
z+Hg^zFCBAf_1(;$_r2wEeFRF>hOcOPy;@~~%;mhA6AJnP{XLF(tFL%=F1gz7M=SqK
zSqSuXUD2UC8DHr%RzuUqzFqONE&gcAp^gvGcsM#TilCr8EhWWQ$?iS82Lf-M%el>;
zB1HczAi;8G<?=&)ZnubPMs~KoK_-di-|u^TRVQ%F#EDf_%SiIEc=_^jwjJFSB@+-O
zx7*wvid)4;6N4ag;>Y%VDr;iNxJW9VrMLd-5zn&U^_}xKN$0=Snw^>7rRn~8XZp4!
z+ox25jG!17NBbGl^DKgrfRY7u#xEB^mZ#q3Qw|!FtlMra?)w!j8s>HYg1h(ZxrG?<
zH%t&t6Hr^i0vo^_%d2B8)e71&AP#8V=Go6SJ+_9(q_?-XSwsw`2a#_Fq;fjUNFuQy
zlXHqZgT;`=3SRvMVc72ttbU%jLLIe30^2IBxw!3425zF78jIi$Nd#F3YukZ+ijrRd
zB0<Cc{l8v?>OL&a0%N{`g{e$^VK~pMEl_bMa7zwo2_&#XhxODo;7gDpd>qfD@CmS)
z%Mk)NIjG0a{%utauMmbT%ch|jX+&NCW2<}{izTndM+lY8H(2Ul3Rq>{e8W!y+jLip
zOoey^b2Y7gHHqYenZmYV?@t?|6chV>0`1ps+L5L@Q!#y#>L4qFJkJR?H!t@LAgYEa
z#GP>5v2NH?l+IaM0prBy2`G!ph!BuFVMwkIgLWs=Wp5cvkPopPpZ-=Nj6@ajBA{}O
zGDB{c5Q|k+-Bv}$0Ar#WtuaMIhEMIo=!ZWShuOV#S>@;L9kn?xYm0gf3o`_2g@yU<
z>vaJv_(cB1^OK*V6wNA-Z!5^8wPqqbC3xb2oalzOmg)d+Z&WbP2;+?NjjM}Fb23xA
z)K-z!R&lO1Gc`<ByO@tppi2(SCw^W;h|lvf5#w(86cPUz^`%UDTLT-VDnNH(QzZuR
z5XJz@i`FEGc!QrW^l~+g)dI$BiHo?w`gAEqC4x=zBw%0p1FARNH4(fSq{?^l+v+?;
z+i_~*#K%4hZ-}dQ3JVFbiT~+$tlHn;e~7s<<EpI1l<SGG=S*q{8Gq<|k_GeEeyKcS
zUJ;>?|HhZ9G2Pup&~Jc_Myv)SdW8L)C+S=MSFh~CM<y*at!BA}C$c^)1T0Stox4A|
z{g&ULkn@p)4*<+UA%VJ=Y8=ThD@ahR%P2;yf)}dJ{#<QwdlCc7No)EpIGKT=m8=`X
z8JetmO@CD4dPVg_W+(T9Z?W%HNifk_F0b`dlBL=v^qt)vzg!M^4RX0Gn?L^@+0`%M
zL0OffH4ZUq{-*LV&pJ%MJD9Mo8axQE=WbW~4)c#|89O`2hnACzUl%RHRPNvadYqP)
zL9`X>W<YYK0S-o@E+TvhO*BnOMKw<SyrWhz2rF7zD2tS1UKObb9ylPk<|4`QMpM|!
zu3fv-l;bhbF~G-XyJ%i;-)>bWu?y}9i~#-qvq^Rn;PgwE@~SS#_~|=do#jtt*w7kn
zk_)%@jkbymM=1Eo8r<)y-y`SUx*+w(Be|g-E8|=q;k;jM+L4%?qM!Hr`Q|+kX5Z#%
zdi5qDg%dZ0d4Louj4}ut9DG0|*D)aGw%qs$6I!cm=rNA*+kbOaJupY{FFi2FMH@&4
zQ(K@xy`okz0urh7N5zzC62hC!`i^Y{P9qZQy-rKrBeQDI%xb-7&z{>%^r2-z-Nq|J
zp>I?~oH3tkI5vX1?N6cZMf@9oQX@zY@)6Zv`i^t%U!veE1fCFEgnRYDjX!p*u4669
z=&t(tjiR19lgLl-dpEjk$C@}hJ4ZS76Oy@>6d=DM$0iy@9&+JfSR9qsfOY3iLQ%Qh
zaBOiP90>oL<AqtHri_Dqqv1mSVTNH5_@$|9R;^y$C8_&W!M4hx;#PP;7f*+uK6$Z9
z^n-nGb;6J%Pz>Q9BLIW%qDsd;GJr*>EH?5R39gWnik3#8oOamGxuown)O9~3(YV%U
zsi4v9l0sbB2EnV9g3^qH@Yd9R3@8(mRjFW16E*jCC>XjK;c|CE_8mF0jk*}4QBaFt
zuxQbvzk(0u#kay0B>(d1JNXb<RKhJ}2J4b?lK;CJ5I`%MZL~D-Exd>4&&ED$bjVOu
zxW;1CaSTN#`E%9r;yb_vfCVk3vHDre@`}!?-#DZ*F-Hz|q%%}C#iD^z|M<%VJO{8x
z&AnM7{XA5v6j%(<UQ{TN|DdXXOWL}&;q-MKrXKa#@}FCB8{_%kHF!qEnl|<Oydq;W
zvrx7--}_h(1up>hHp~3T*=Yu^NRvI~RLLgJ(43Ug()CcA1+m|YrTMKpfI;58z3A(%
zW*oKj_u4E9=+A))O0#OEji`%)p!KR5R3+`?9Wm#cW3vl_d!%yY479TnG-U4kHjrfd
z=FOYn9A_b!TW45zkx0BDxu!F~aH=N!628{>`XKpXw{B;d#Z<zYW)}$e$Qimlcd~w@
z#WgnFAvHaH6Kd4=ww04>krcn=<>YLOks<2<BB75l7M%QJYe4hIwN1J`0#V|rI!bgR
zcNK{Uu&(Ea4<FE<ZppzIDIX6uTqUv7lNBQa7%yD**=O7_e0+5D#vaG1+Y|LIEiKuQ
z$eKGD`}Vk*pjJpS8*7A_Y2(FR)n)()_r7vPXO;&_)F^yh7*+qh2NFtd&vr5n@~a7i
zVg>)XXhSsy6Py@5<5QSX0IB<(Q~q^DSoEZt%+bc!5@o?3d+hD)O<NVan<xckacDS!
zB0}2&2m-7Tkrf97nvL3YBEZN#_8km)_wHTL;XMZqRG@4#yIi&h1lVnN+zlX8)@^t8
z*CyhX4Kf>;%-G#{N8=1z2#zTR9#yHoCU!3};C1&~gKSnirAfk7O`N?qd#`R>zkU%?
z?uhKe?3y36=_e`zT^Ut*8kmhs1*eooFL0jG-630&Oq9bGf-%2CNmd#}e6ppm^wQEj
z2#&yTN+C%NRh5MiZS&-m)6{-2%L3ZU)-t2|F{71d6Nu6|={s&h-(PI0hWx&+5z?<h
zkTULyNVj}`HL&n4A-VkX&p($LmObLQ-?v@dRU1bCjl`od?cfy$1txJ4Z7^QeGdOY6
z7sw!NQC_H|#C$#F+}(8Pw|y_HiwN_9h_9btPK3D;;iYiWP>MHY#2j(jXpggVZUBL=
zioWN7yy<erE-Ned$zj@}epYv~g__VrWqSnOC@A6H?G80c55qI1%F)SDg$Ug9hbQ{;
z#)HXH)MnL>!LH1=JeAM^3GYlWeRk)gS%gjHS1qlKpmko%`N#qy+nk%cXm)CM91EvN
z>%LR<>mGqptIZ)gA)^`2aGX*@1+}KSN1*(q(6yT0@>5`?jD-`E<UuYfQRoAiYZx1O
zou<~`64dOQeE&CG_zM2{`_E0TLuYkc1)`EgJ5dNhjx{!YDjDdztvRhnIq`QtttMpX
zLH6q&H=bLfd`a4H9NWe9NZ48gm+b`4>hHh*4q~0@cUc#oOF)o)xi_NQgk^<d5_qcC
zGIZCP7|4y92vU%>LD*605B=HKfkgt1)p-GgqSA%x#tCw1kuY0B7aJ7T1Wl;2+7HJ3
ze!S|*gz}NDC20X+1Q*757tm86f_M@1l-SVnfBqRjOUE<)PlJOOo3U@ss4*y+d_e2S
zwpW@z`o06Pwk(kf4o#(7xt73>vczw>kmOkKdI36>^cSn%?$54l1*ceWu54LZS-19h
z7fl-N2$!Zb*7aO-GM$;sKk1!WguFrYqkYag{n?g<_sV>zX(a<Bu78qcwjV?lG&Ij!
zyjV*j5jXw>K{7=9il3hoh$_ouCaN?>rQN$3klEjE>7O9HOrS5I`>;JC$H}DmF8%(e
zi%9)?nVXQt+M_myN#KWBEq49xR?+xwyM;<4ArrB8>y;piY%e%Pw!bEU9>V;t+V*``
z%Y}X)e0hW`O7MWWP+vl6<pAzi<Ta<m=nSom{z_%O#C=B)WJfSWX2XY!#wk4l{_ZYz
z_S|-QV(&QD@`Z3(TE$>NNo^Vq<P?8i*Gctpe?|~_eM-%XQzI$J7_4{&++Rd=_u@`;
zr9cYt-{%M|bS4-!MrRIVRI(iCjShF}f{%2^O^HQx##S-{+8cy^{T5zPEC@ErvHVYR
zCH_E2m?q4MrdyT04`Dv-cR@2GT_xx0-(sIDcEw5KPR3PV-xSVfxeW$vyhT6Wv1yvT
zU6M;)9Wrc#d#W-0E<p5@n%@Fplr#WeNE*@E?6)!jBxshT!YIttqgC!VigBkM>F)&0
z`}249TlWh4k@tY&lHa0|f>G*`AH(W!sbSor)hHIDKUZj+AOa?C1vt~)V97cqbiKWr
zuoOpj!2?DXEg`I$z*%=%UiFgzolMB@AUyZ1;WAuBgv>6rwls<{Qf{*gZjRQf{M}ce
zE6rPi0=h^{&^-|D2(`dmxgmnP(4Q)~C4`lMS$elwwr_S70`kfC^SE*qK}9FzGS{0<
zn0Pgz#-VAqbMfw7yLW51s`p3+eq+Fa(>q)-fys32!1@tue#RUFQiuT*m9!8fGkr@^
zHmN;rq?B(%I_H{1(w=trSY3(F8X!M|uxAB-#U{c*GGkHa)y)owu>#wu(i@#7NoyKt
zC)$1YyON$sEHFdB$b}OVPmq>2=hVu8^8+y<1VE(nU<mTl!?shUw{eA=dxY#=nC^jK
zyMtRd4mCm)^OD2wf@RgEMtIuDq;|;Lv@{2B09{Lp6r8^+3}pJ}Jl?PEyShit)|Hk|
zQm4r^dQ0u=_Yj5^CoaaEB&yo3g!znwWtJb@{8AyF5V}6fe}ZNCDIS`f`l0`=vVDtp
z>XWeO8c<3h)<!pJHNrIUAA9>>gaaQ@76}P~|3}MOP6jUS&!2VJNp_{{F%co>(uIhE
zT#N$IEYw(0ap0D4-?Rv71){<9qll%owY4XX%pi0p!lZWusf=uv_!q>4K=>txF@l=@
zk$0T*bzzU}vWKDCwxTrHJAk-NOY0Hq$`V8Z@7$nO$4{R@3UWe@)BCq7^SkZg)xC(^
zuLRr8jg6wP-O}_)gub`3hBVAGtb(`QfttG0?ED@5ziOl-={qK5?Jg)4217R%B%Y=y
zcF*=T0M@roFn>-8a`UB6vz;{`%$#DqANr?<vFe{_r&x7w_-JRtZTlZI&dL!3u4^3_
zODb~yN@IT}aV!9K$!4l+U+F?DcP9my1k+Hi0Wl%Q=#)9PmC&3M%;TW4@%`@rPC>5E
zmz~KeO_7d*1lp>WeTX5f*q8|L#0D^q65M}6z05wrJR0=qRc!+}WDZggqL`rie|Kqn
zV0!@V>+L-s2P1Mmk-ci~C;*^n4?dW&2C}hdm>It`KO&11C00ADa7+Nlo#Ft$7}evv
zS_GB%#fM7>Z&*(Mxm?iMFRu1~=0n>YIk_f$C%c(qc2DmYa)>*wNa$7xbtoNtLv|(C
zF#7<gtwhG^nCfKC-2zmHYN#j}+~W{L&Qiww9_nbZgJCu!j?|az>i!i5C(<mktgMWm
z{v;O|mi<Z|zTUc>a>cC}4~NyfooT9&pP7xB6^Fq>(@bJt050H<P{hs1XI+|pkjCQ+
zUP2I#Dj~*!M@t%X@U`nVY#60+2s+Ay#-=Zg4hz*rOI1AhLxKEs%`ykP?KjXt``D{_
zH+!eX6{n0P&s<bOU?rEl)$$Hsl3{{MLDfhqkE2PYFZ<8$toZh50r6jotLh2;broP8
ze)@^FV)M3_Tce-#|NV999dR&}5(&=KwcHC6$RIj6E_uu8zxLCMhqs>Z-v#`e$F(fa
z2Cs&T#PUHc%aHEex`mEnp)P7@j<2>hPwL;WUK_X57$LSzz?|QLVPaVRF;sI}wU=D7
z(sgML$+)lMYFE`>Lh8{7LY_B}@ij9Z>JE=S%3ne;w^^%hNB<{FMm_<WZ6P6Vs4F}2
zq5a}61UclHgdoSfm1s1KYAMQ_5&#W4<sj+r>?E_Vw~boc9cet2JSPIo)9W+P&Px$5
z<*s^rkF7cVy$JFN+1b9q^}M`r=9{jY)-}G0cl@$&wa|+7NSpHz5W5(SMjee-GP8lg
z#Gtbwpuz413Z2)bm5?1{F8YQhYXtfmKGdIGUbP2U77C}N|F7-yj&p<>D^e7JO4>l?
zQ8e>Se5YC1{{rmQBdt~FE6c^;so(o{uwBP9DgiWNtnRwjRbwVhFkr@gpV`WTdpTde
z>&THKdGStNIE^6tj5cxv+F)&vMuM!9YC2A6f{N#-i%hi5=Rdb+X8&&Myi@`Ank;bz
zcb;q0n=|-*Y><Bbe5x+k|2?Q?cDED3>e+8VOlx+d_Dj^&mgk$dxJf&I-a0O!e@77R
z=)2q|G$aI#3pvmc{JrNO5+3w^IP;x!>pK_vH=|m2Z2+JSYk7HF&*0zU;Q~mqukC0K
zkt7TiMBwPguujPPfu1!l<)EiLoUi2UzN+VK+lvqn_~(L1A;$1wzBPA#xVnBT2PN&p
z#j0&P-yUY=vkDm*9o-?M=Ku^pPt?q8O;EdD*P0QB&NXxh+OAc-+`miQ`42c!-1=%=
zPpjn*v`6#WqWrLOF~@wtiSH~q;d@`6K(~x*$rc0-*Yb+lXkSAU<QAF+3fE4xlCHt^
z`aGD|Hjm#n%mn}J?W5(AF`~c@WWPOMnM@{mG(T+)D=C0^D!6_;FDo1L<;$1o(B5me
zdd$OwuXMG>4SajSwpoTu2U}prTkX&CU<!9~xey+O#U|E+@Mv+*m(#>XcW^s#0)4n8
zGnH+4IJ%qc<8u9#9WHWVJw}cLrKn{f)2RjZx5O#-hCICMd_V5Jwn~=6MU>=GeM>l-
zRgm{UrJ$sVQ!VMs!=`hZdb)yGstXrd))lTHWTdpjFy@=Ep$vzHW75pP)%A`%Y!%~L
zKV2fQu7J{YM$PYC=<7(s2RPMsYty{?KL*DGO8cngk`v~<$IPtC6k(AIU6qx_YEt!o
zKZFB1^mP`#1@}pT6ly%B>1dK*bR`Vcx6z|VD+Nh%^N0*qMNZ2({SpJlp(8e*8vi%)
zRqoVKkG-4rulPfKs)^o7(d!qI?~T+ts6F80)|G{;?k&^*>-3fek(F}-=jrR|3)QdW
zzk9Uwn#Q16gICY_pC$kFL%uJL6ePd&YRGk7RiMfUt!Hj=YO;^tTk$a0JFUDy$}6;T
zx4%JLP}<M3m!xx=okDzj8-wART3R4bd?*7d@c^?*<dWU5-zxw5+<3SMe;lTY0Tjq=
zp0srS{MwIOHJ1iOy!4=YeMzu8mvkxp*(2VUiCqjVdKt2Qv_b$rLCp$Cm2OLOt23=K
z?Jy4aQN{~dmHCCj-w`)Vu}@ZF)9*+MAwc@EU-Sr{pIPe}8itlf&v2JZW_SF0czB>i
zIkOxMV3CofV5ptB^VIl0sD%4l=aGp+fNTBUZ!b=wIPqjpB}qfw)E?=MxLF{G_MWER
zB~8H6q@b~Y8!|b(0$8AYJmWeDqw{v5{M|ULiJ!5Z-mv{_sl6?`Yk@fU0i@gcRFCZR
z_=%wdk2%w65UW8WzQQAQfaX7y<dYA>+rSNvu_jzmH|C2h6iG=N?E8kAom6ioIf6q-
z!}Ew9xsJ$&7^swU_6kY_iyo-G+xJ*>yy7Ma4TDS!TTRs$diGk^&@dhh;S<cj7EmAx
zE`@qP#{-?zBt11C9po2!B)n>v(7e%aR`OvS33PQIEcMz1STQE6@>feM87@d(c$5+2
zndAxPP6r6P$^B!Xs>vqsBe97lz3!9e_Xx=>w#v!NXIdLbXn2~k`+k8PHrmI9gp8Jy
z^otTXKCB#*6F|O?{&X%JxfT%tp2^dLKcIG73o4DBe|1(w7#R2Hf-=?JCXooW!g4s`
z1B)Q2UwS^km|io4o`2Fzr2m(_e~sF&yvids6`$WxWW3a$Xfz`<tQi8|Zc^G>NF?{S
zzW$0{QM7LPz906jf7@De6NOVKU&k2+e?Sd=kBRH`jD7c<Y5`G#gB~6}S-{n&fOG`a
zdHk>pXz`sAy_tuz@Y1HazARi0?R9>BWVf=NTPi^$%B!Z=y_t=G1oOUzHNCTfiYBBf
z6oaG^`#-T~&O_-vI`s7Q*Wm|n3xna-zFVP&akB(mi{<9bA@iKooyQa9(~V^VA>l%r
ziTg6xDWt1HOKZY1t5U?XW{Ap>^XYny(%78&m}KU8Epd5LbyG^guYvjU_RJl|^TyP-
zw1RuTaN&o}je@QP$;)x=wV~1!^C`_2CN)3n^H&ib)>BbePtP9RX=v1M-zG{@;dEH*
z@cp#(H*efUgmSNvX5!c9uAXzQzj3zo#Lxae|N9S&EHc$CwsR(sZ~eL%XoJqEJkrwI
z2FW5h0at};NSz;ite4sRB^==(;+~~JQmH)bRb2PIIuhdqU7wH;_FjD>SchnnT^;GJ
zJbP;p<ZDP*DS;9AN!zzc4<u>QC2$E)nl5k1ZAv9=+N5)heGljf39zzA>hNuyjufqR
z-%7jwo!|X|$Nr$_RUe4~s|cABh>nl|xNK_o?b!g(3b5GCuF?$GNvaAId@ytIi~hg)
z{-Z~KWj27iz1%Oguy7~XG=$-96ifO4U$0K7iIkHT!1Dp4RBK17emi;fd;f=%VHOtm
zOHbF+&OeVNA`k<E<=ms(o`<ZaJ90);+xM&TpODjWDp{U)XzknD{x@Ae85K|J7_rGz
z5KU1^JBMlD_wUOH{9VynC(o)oFHKP^#&5pSzi|q%j8JbtzLZ&xCO9x@&@+0`RCluk
zD%+u<?rLLHNuGWuBl?q=E6R+N5v}TfkA&i7k39{{tA<7lehZj1>LXA+c@wTgYl{^*
z{-AcNYQt9b_*#;JXAWQdw*MWf`NZ+}iO`lrpDaj|XgK*{wG6RENJvo7wBaH*%0Vix
zKH>N<BCGO_cG$9@n>@T|XdV%C_B8b`A$?Hlkg!6D0wH%rV+$FYBXay*cw{7l^yjL_
zKgB{9U;5<Q|CN{ja@^2?K%|nw4|LflK=ROzTq$$F5eO%JVd3&-l+dkKmOt<R^V83x
z$3UALA-22?<mXbu@6K?K4#fa++(~M%wXXy(yjNXt`}S?3r!n)MSpPRu!ZPT0Ng{HJ
z$eK{+*t`q^gS#usrIqdLVVsl2mz-5fb8{?dgG#ge6g`FhkGgBaqu5yeOzX`Ox}J~q
z?xEKRdxeoyVVR&$?%lga=-CD<3^tgQR|Z_&5#sND@1uDz`)oO|P$77b_7f=H68Jih
znEg$^fYY$iS8<ZS->t2Mnr(&WtJ5D>Z&>}vp9i~_dTKncw6O&{osEusiW9y(?Ps_g
z6ryGb{5?SlZK~?=)ADN1+AvjVQ-=0_HnF~x`VduwlevFQAKLCAA%<a@WmeOtPe&)u
zp2vDb*5T~<rF9tNj!b}gPUQG>xw&&8!4sz*Ka%uE<j%aeu(yrE6+o#vu>oEIsZR(C
z(D3#ZK}1#dmvs4SpAlF>S1`M}F|$ohSX5!tW(oE9hHyE{mE~!@Srk@<lE-3WS_v;B
zxbM`VL+`wwHz-1O8voAcHPt~TZFx{bNpD^cMnTEuxKr&V#*c_#8x=m$D-W9QNBnRA
zl1a(o0QNxA5*9UoD&P8Eepu7!<yNKVr$o=}b+7RHY}F7_GZi@#`qVhRx)G!=Fwoij
zJUf3c)1~PDO!)Se@hgI^ym+zoz3SVjZ6MSAG&4t{H%ldNPlUJF<w&8I372PUqcyX^
z9-J=|lj~IL(c@o<V$v%5nGHXFge-L1-&70s=cK%~r1zuxq}ga5_Gw2kWKT_XaV3$t
zoRr%~r#d^@HMC1~eQrZ_)i_qi0xs6q(>uIY;dE~f1-^^YpkilIu7@6nhi~aPSMaE$
zJGHz4Ex`5-Aq!<AA6FkGg6it=Ms44u+n&zMS;@l^OpdgzEdcCLr<4cs8&tEPy#1EA
zmEHb!FI`7|*sa>~Wm4W091^RQ3SK<C<sL(fj>9Y|aAT#jNi!4Gr%ldPNlD2B#Xo#i
zoa}e)D%Q&*Y}KZc0GqWuJ@)=^2I*`fGcueTv}un&l(7?(`HL50!yMvr(L56s9$rv0
zc84^21W!21D>auQjJnxt7==yja%yD+GTXesL7hpw3&5Qd&Vv#|4Pv5m%;<TvrXnw=
z!Dqb9oyxZ(A<T?sqV=I16<pu<8+bSiYC_J9ENfHKpwRN`Gs5M3{Qb?SO1ktit6L!<
zxT2a@ju9kskZ$@6Le;ifDb?%U^1CVU54XL?N~0nwYlLN-wqRiTP(324n##CUAJHOb
zG%TtRGR{WC!gzZuV^`bO-*4@zs1kprwB+zui*HBIo!We;=*{JCr!@$-v-HxEUh|%A
zPHS?FL-RR#%!IXeLIn<;DCp=x=N!>mkgRU7*V1Z||FQd63+X&9T);|6W1gAvRgnje
za55o+83a+n$0SW5Vx!}bSV!a>*^aBfv4seWe0lIL7U>F-?U+xMe$dnbA5|>)(V<?G
zYgjg=^G6X1P*PVYOr8iO5P}9kKAhb>mHLwjuZAL;%IYOYfh<i@_~jA6{CCRNO+3A?
zR7(S7RiT9g(K}$M<g7h8faXEY74CF1TVj8v`QgKLdJ$H@>d{EPaj$x>H)D76^=Wy?
zB_M8N+DXqX-vwdnSI!l$PGqM%_Nl(KWE>ApY_i3Kwsg#(BPp9Abg0g=+}p}=+A^RH
z9A~Jdg^|E2d&V^ol?ipT@1$A!@p7IY?Z*-bqqo`j8X0>T+9%1T4zq!!p^If$lq(S*
z@O?k9u>YG!O(^#TJ;ivQ;^D$SJx$qFKuAoL*co~jC7zaaYR(o%wQl0!B2r(^9e!{<
z4-Rr={wl_HgLp@XOi{Nwx*`ae?rEyA`J;7Qbo7tuHJ@Ucb;*>OEye*S(x#fhkLLFp
zd3HB5sY-Lkvmx?SzUhGS*CGfzAkKLlMgzG9Xk^p~M@<#RQ!&vfy*R_P;Ri9nqGzhr
z?N(cvf1>ryTsAbqEmY1t_N}eKh-D3tO1(@ndycZU=e?%f-K}kPL0Btfi=hSC31~%_
ztJ>BW)ALjh6&!bX#(EMRZO>|L6pLCx`bmo~Ey3_-N_OfuIH$X(uqbRBM0p<&(k;-s
z#IA`E??Zy-?6DWgh{C9as`cuskKpQgeX12}j&`X2?al-T6-;Ch1cS7<w$5lBR9;t<
z{G67V1UU}X<w3)GO^|<&oLof4pQB~|9*v16f^d=<?@Z4#YI&&t?;dcNoOis=(TN#z
zcF%X7ZlPBd(vf`zWI^K`;MfTE__*ph)DVmuYc+x&)$(L%JDBIdd^<pn-0x8t6n|`&
zs=g$>W|M64Nh|GycYpQ@7Irs;W6bBmoID(TKx~Q_b=J_{;Im{&eA8QV5VG{k)$<<i
zK6lRHFP<FlyN&s8F3-k%FZ(!L7KO^p21L-9ivb1B(DIB7tq3!#dw1_<(~ym+l#?B^
zS$pGIZ?SGymMDqNrv$tNbuBp@`lvDJfFn38rd&8s*F{)-b4a<u;dn-(yl9gJC6#1q
znmF@+gL^aE&V~$vP`ij&P2cb7<y9U%*L`wwIQqiT;)8}pKs{Por%#_wb^IQK5DP`V
z^J4X=x>>B9!UGy1BjA_BDgt{6y6(w89Z~f{8M7mb#@uq@_`h4fhP8e@FIJ1LHkT>m
z=Je(n+Eq;15hVZc()B0C3?10&OwX#M)F&wg!V)`OPaJ<y$L}N&#Qb8%dzxgt3-}2Z
zYj5xX?U9%Dd2s(eR6WBop+Y!v<VZASfqT*ghb%}j4@EOEHtjAEWIMa3uvcs7?8srq
z=nI<+A>4Qq42ZPdyLbP-XR&qef<`nJ@QwvLmlvZ4H`c1`;?{9Xd0U9w@IElEA1Rq7
zJ6bPDH(i(%@EVPhrooHcd%fd38)@L#1SY3@XNyB23MmEZtN=4@x&+kGp%ZK^+mNL0
zd>x!#X0>hGHsAL%dUdz}=DVamC)IWzem;zV{Q!@5@7jgzC$hxfp$=ew9h&t-nFGl<
z3ZN@t))iQJLZ#k1ZZmI-aP{G%NA*w`0EPPZ$jqo^%BMCNy70GVCp3<YJN#9;Io)mi
zyHeiXH(Uixf?Fv>CHv57f~cU^nvEahrW9q)ff3?%j{a0@_9eV1$jk8X+8fEdoMFGD
z=H<>zTLLX?(9~?NdZ=Qk0BF8~G!JK@Uhw(?LH%w{8Ml|WMKHVGj!lw;3Vck@IoreK
zz|08UFRb!{XtzvyWecw6Jtv)wd;jg#+4eBoFEZQ%GU2uy_zS`pwRrumwaBO~XzMs@
z-d=Ud)u5oD=uqUR*}xatAuzp`0ty0qteWc0BFu*K-<!#Qz*`0{`oWvs9@yCzRQXb=
z5KIlEM*IEJphbhxbVaky^7G~#Z<x#*gPx|T$>D5CIjT^_-m(?G$6Bc}-Fi^Rvg2y=
zdkgV8oBmJ^Lu8)1vk=^SZsFLnmu^eP3_VeC4LzodE5g2wQRQuASAC_xpd7PCNOeJ(
z)C64#`;Azjyr5fAiZ1Aw*OIGHnAb)GxF<+S@%GS{&+F<A8)Q=AfyRdqhXY>#GKX#*
zlC|sU>5V(4-qM^&S1HURt&D*8V)S<KfCLhM%di7-mK>h)lL;EiJ_50z3mM8gt<Eeo
zSxm8Ci*XNV;mSKt`>q{dH_B$OFK_;KtogC0`KV0ldLCp8DMO-`ToC55wJ$~dQY7PR
z{+YzOMU<3D=rmTAw}m(w$)yNve@uRYuPV(M{vos34I<|%plnyAT<JCK>ERbgq<0U&
z2=>DX!ItC?a)h9bbe$I~(TJ(xwUf!)3Y<D%XBWjz*Y{K1#dxbD%ltqk(aI<&C|JeN
zR>gQE^|l9ydGfBrOw^a<?Ho+)j$a<k@}%%f0+{$P<eebgcY--h6+KE(Lw6+c<PHb1
zvU<lbfwf4rMxdnT)B*t-Ql^2<;5y=SYHn*f?R%Z~1O0ZoIyx5oFE926>U8!*lft%|
z(6W&yBKxilEBr8Tdow%CvF&F74JBM8&Gp5s1*G{v%t@a&r}=FI8teYj^+dlgp<H*X
z5;x3YDU<OdM5-U9nY{7<XG|)Tii~`^iic7R{KoO>49;=fsr!jiS-RruTfr$GBqk;*
zG)bcV{RTQ>(Z2$m_R~tC1G2_p<c<{Hwj%TUZ{Zq>Jzc8C%L~oxKn*P+pl_;bIcXwA
zFzR%d*L_xa!*NeG2naj#BHw@jR+Vb=ex%4iX@@mUse)1lf`ioHrOv$heT(OF<mCQh
zR~1rUyjV$fk?*fOje%Q{q2-TN_gB(lU5%K(vdpE24pezHpU0(M<)`8VrWHUEN}kwg
z#k&{#2$7GlG^Fzsc{|?Enssz_Z?h>GH-m%e=Fc5iEqT>jrFir@Iy5wL*+RfbrTN`C
zXb7553ERj-*85m&{{h~7!M9?20AK{WPgfT^2-Sicl^e*=ae(8y7fs}Ch424?W2oFz
z=s{6nfLb0npK;2WUVURM1}pbW=6i<hpIr={8ZyZFg?xoDQM$HloDSJc8;0V`W4s;S
zCBFr`%{T9EWTG)PM$BX=Z|m^)UtmcR-A+_8A$(gNrBI7rg0J7s@fZV~xlX#SvT|R5
zXJ#m@GLW~Gn|Byt<U@bDxd{M@R7YOlweX{Lt5f-Y?9qdB%Xst6XMD*pR=pC8Ufi9k
z_-k{o$Zn57<oJc3L`%l6aXLy+-#)MQ<83omrD7=egH(r+NV?8CedNtI{#^)RVn=1K
z>ClnDu+#}Hw|a#XdR&p<>*(kpgD!uf@PNj0-p%vcy93d)xrZ%7gU{c*Z<)IaCiaBk
z*B}!Ru&ic2Zz0|B&Gn$RS~_`72Q~w6eLQ1KuLwlX&zqrVU_kD;cu|Luq_?5u(JLO$
z#Hx>w-$=Cn?8AWOm%?gg9*^_9>z<yTSKBU+Bqfnl9>cXkH_D~*7%$;p)z^)LBO)-F
z`a~Y1$T54*Db@X{?l6+Zd_}9UeRYaQGyD4f{rh8LDz?wPrbAR?`0nMqcszDLpk@RX
z#4!D<wv2CX+S|5`ui!Btdp0H#UAF3C%2w*0)1mLWdEH1}Z@BM&+u1dd&NqnMZO^@e
zrpOYKc*`yg<?*dIk|AKi7=r1La|IVWGu=(a3S)GH6<_lB0B52+M9kKCA>F|Q_be$S
zAxv88w&WL&j|fF$iv^HdA`wsI&VUGh7)NwV0bb<&^lv0JDU5?-AX_~)m?zZAZ_puJ
zA4`wl=j{#nhV$mo`6u&sz64hCc7TOWCh!&|iT7IG>KSgz2=L~))i9crmrK-z2bvba
zZ~j}rUf5d%IQ(w`d!e8b@b^Cp?1k`4GRFUoa4*DlU^fub<n0FkZvp>XK%!;0`@dVb
zmktQ9xFB`=Ul`M)ljXlKh6@S(zc7YNC(VCh441y8|H2q9{dfO`F<g56P-plr7w)aM
z@V_vo_eReD3t<e)4RgBhMJtV0J-bQXGBRq<nVriNvZtS#zuhE!ot4)CT1w(dzocTT
zFP{H<Q4s*{;{R07yk9YM(}0!#9v(h;+gz==-<;{5NA*ohXWi2IW3EPE_K+au%Id<b
z7mS~d8<#!`u3Q>W($M_f$2_~tQYAn$KsDD+B~3kU&VXl~jlQ^Cz4<2HhhCqr<w!z|
zn$E|ksPOJ())pCyTu*_&$y9E3+X2x}bP@lIF0a(6L8QDml8?{VXV3S8GsP9n*SXFk
z19(Zzk^I<CR}(qlEhtcuP4xi|;^Rx*@Tm5QgotlMklJq)Zsn0OLQhioT`kQh=@}tD
zzLM!*ZMP2=O2r%eM(xEU1A*<uE(uE&_|aOh2mQd7BM-*T86Z%CH~5WslO`nMI>%Bi
z-<RrbTZFJ8>eLc^jO$OlL6<&)m4Ukjjwg67m9a?Ozy~7GRSba0UMYo-NwC{K_&6EN
z&1vD7oW78Mg>mwDam5m<VRi!_pM}}=|KVfu@dp1bC#N;zzvcYTIs^ZIJnXOaq3EOT
z_%=Y5cH~ka&74o<`9Ep3i(J&G*_x!&38P5FQ#S5Mx;Sm3sJOAv58?zBC*3cT<ePli
z(^K^9MP~IxQ9h`;@{t)k=jj83YwMEb9TfR6b&5}BJiHShc?=)a`uR>y+m1mugh|P_
zr{ReNegjh=`0zift^a~iHR!Cy)SWvKLHJ$xC;T2CpUE4~rTBjVVU$k~4xW#2uBg8x
zz7^h;{pj=`2WR39BwVS#>2f99RFwT_mu=?j_}PCCiuQajF3qvzJo=IS_QCAGN8ViK
zCVV>k(SaFRTR4Ajr3@e6eADFdoJZgL{ZnTC_bLDF%KsY!`!nS3UEhh*n(~y#Q>TS8
zS=t889Z!!A;G8|a|NiyA0y5Hz7cce$^8<`TlvD47h0)PPwp)NN5CPH)pDcd9O^AO2
z_wu`(cY-(qYuyDoPsgY>NLQF=M?ScI-2|0Xs9PLk(V%DNGw<(b*ELS>+qVz1%v+=R
zU5twi@c6~8VW*v%%Yvs5<dezGhGX3KW6W$@K0e*u^3r^2+J8QI1&*}vYCBBlIm|5Z
zxq9`9S<W(NCgbVY*)wPAhVM)QgOK@y-^HTPd~rr*rdSnGkXWeB$5&f?gpcp9t*>;R
z3i7?4?9jx=7yTMtBZRDHrF-%Fy~ki|9+?~Or0ymPp|OLOmJi`KUl<x{6Pfi`>;J>v
zTZU!TZEeFBjdUZD(hbrjol+`Dhje!%jUosLNH+q~jdXW+N~?4VNY^*fz3)f&yMH{#
z_v`!KgZ1OS)wQm<<~-+^V~jc09Or<8v};~7yQza;2YPbe{;wPF)fv#BX$+L?1|YX7
z*BhRJMS~$ino+m0{*b{Eq#&4+TwSq2d?5K7;0)TqBrXKQ=B#?z2o4=W7H(eVhX5IN
zJ3dgq2Mc;Kk|EXC+glVzUh_qxXBgBDP^rZG(iuYF0LUVhfF@6MRaTOh53MSK__+4f
zQL)@Y@>Mw!R{-LpP!>D%gbXq)@v+SAd*?&30K?ITpcRnlYj#rxlG`HTI02I+ol+KM
zWLair=G9Qs&rw-W>tz)lg2BF@lLbcWwuJ-<tg6>Mxa|~%!mr`2v%Wej07gW6dATR@
zb}*~RPkE;`K-B}4+I%K@`rAY5$Gh{h)6=JKQo#(k0yc=8XBnec+knGD9t=@IUr8S1
zv@0}(=(Poe_NFYW#<7ClJMcpwdASA!OHdHUkOmJGXeb}%GH|jEJR}j*2~MgWrdNC@
z<Z)ad?v_pFFdGHAtTQ+J;5;!XQI3O$C;7g_VND?t783ZD=fxNx-Bl{qJ3CnM>bnJ|
z#nqzXb_6=VM*#WP2yUtVaB*=r-kS`@J<4co6lBq`SnZFGm{M0)|1v%v^_2|p-&ilF
zaxpM5HGrIhFVK*7;iNn>P{%iSs>JKcy~zk7T)*<NzCr`cJ7RIdZRfRcU!B>P09kjv
zP1+8`q#(yO0=b(Y-B#W!tF-jdiRKnKz8ith3u+61-seA{cbke3E7Y!=d<_LT(C*JO
zTc~%-#V;}H!)r4A-TbXoA(s4V-EdH+an4c$LPqlL-8=HS^n!xza-o1|{K6ON>hbPx
z!203$Ja<Hp1o@Iv<TssP042}>0fel$qacC&Ci(+NgqdY|0OXDLzKzNcP#dp+`c`7%
z%VSx%6Y(H{78ba8c%bt2&qHq%2GCP5Wf3FQfH&?vYTS1JvV%0SEjT8<iURzRZ>sWu
zB4Xf5NCX+INR2`;^w%2lw~exyu;^NIF+ni;Cf%>k)`U)%{8>DxPc$t+s#k8Ts+QIq
z;CTD`NZ0%7X#7#Oe4_Cn1aijUHKDAc5_7&&QBmPFbvtPluI0CLX8e0mB{g`|eEAF!
zzSOV);t!Y|6|dxWt0sXADzVsr<73GLhu{>nJTd!OH6WQYnSnsu=)ER5+<&f*+X%V;
z%vt0Fy|iaqxSgWc>VNk$dy*cc>DM}0)DA7XT5xi3h+h*I0rmlB<7seKn}$?&vjm*m
z=CXqLc;NJz)6>z(Ty27}pE<!mzA<Hg8$cp~2$xJ4-mYJV<AW9BF=$AsSz%e(Ef-70
zVLXvdVM0PecUvaN7};Fz24(paiQNE=Z5ApdJPg3d#KaU~VV!&iQw)K88`yzB0-G&M
z!O(4AY_IgSWWho74LaW8%FD~|cvE+2T7a4-xl>$bqnU$Y^`Ls>ctlqSfdLrr8LijE
z`0#M}c>~a;piQs&TNHEuM6sG4bnhkTo=DIBBs~j@VnZcJmvr%324`4=4mz(cXduFZ
z?{se!WVV-?j7-pfL9J|1N07l647iX}zddl=Vz$xq)tZzbNI6>$otm6Xz=a4e1nF`T
zgG_)USx|j>#|0JW5UpVDTS3!P`K6i)%73XQ3}cH=eIEhE5-r1be0=<Dv__{bm2zlE
zH%n<j)^Z2v0uX9l?%Ro@|21*2n!hIQ-(83+bON+k-4?eZuhK~h&dqHuF{l)<GC+JT
zvHHyOa&u*_MuCDm{Vx->wY^e~g$Q~5MFV8#D{Q_2a!-vFR1l*3>{$e#7f9q*hk`7}
zs;P{D1o7uhjzH!jL^L%u6E8hLj_u||ek_w_rTIi2;0oSdXt)3?8gQ#{-I|k0r+wpF
zNtphn$-|v~Yw}jh!p+Uiab!YJ^rf;hmD@&V^E8rLp6y><odLuYo3TJ}Ps-89r>2T^
z8qT9tb?E5nA0+7<c!N`+s`+_&K*|mB{uYrxnxDv5&DNQ&w)<YjdHoKT;dvJ+zjMLA
z2XU(pn;gMjqvvJ=WGn%tcWZBpC7(ST0aSyVTU+w?fvTwi6xo>5Zntare;}?f9Z6??
zELIDY!1%sY$bCH(AZk=G6Q98WuyGdcWACD4&#g670yaaSe1O;iss3%C5w}tPWuQ}|
zg&}19lkSHD+)6B#fN|=}1EB-OE<kIi@dgHRAj_CfcQ91qyb9U@sUth!c(pOwKaH^d
z-ADppM%;Wubz@N336kYrEjMs;a>j$slLQ%Y0%}nUTUB}n1`AVDR=0ndcJY_LO*`K&
zr7t|FQEjVdqzLl6r4S2WfXQ~(18RD3@)Vpz52-%^8A9>1pf!n?jQP3s^#YNx*<t}G
z@oYv<4!FyV2NJ-nTh4qo19lxG?Gm#a_v3myA1wFaeS!Gw;+OwkYnHz?*1uo)fYW?@
zDWwF!h@oKt2A_+6+(~F-XmJUYG5^5=i&phZYZ!<djxP{;RUNla$x_eM)OY~NFvyCg
zvf!FWJUTi81Om)-bQ>u}@R07O{eNkke+e>gUri7*C&=X(M}7rxF6X=kXVNJ4t^hOI
zDi<VN3iE!Sx%nEXP|0T03!DmeqU?_!^?*d}2uuKeuEbG@u**C3`d^l&^0%c){$+&8
zOnA9Re&yefBj*<s^iQ4wqP2JRC!mma!oPts0}21Ri)zZsp8;teI5nA1Tv%JnH?IUL
zse2&6!$Q_e;yMy<U6PmiJ-51Y@=Koy6$ZpU#sOQ~9IUKdd@e9t%h21Zm))<LJkH8$
zP$A3DU^~kaKq<94Y6&YVtG5Vp*RHh)iO3tskk=H(|7DH&B(2Pe$0j^}<-GZ(;re{*
z%Yz%hJP*1mPOkSG;5<5PAfQ0_Be=ot+L#kEY^9H9qXG8IoJPT*f4}_#;Xyu-27jGm
z3&?B0jQz64T)<d^{sBK~d>`a`%3DInn#iAW8s!uf6_u2*9_a+25PU1HgNDpg8}DaH
zKmmsm?-omF6PAAozec5{I_MYRDfsfhdm`^e?)NZ2@KOMQ90Z&9YArYUefVs%D2ROz
z*k)@4(4iw9Aos%E|7a;tb$fjjPow0eNNgq8Ih*%#0+aN<U=dUoNRa-=ed@Y>si^V4
zUpfOQ6PFiX4osc`NtGsUN>6Y7*Bb%l&l>rb3lIQmtbZ*v^G!Rj@_oX8PYz;vh2Z6n
zwr-0$+pnT3ovEyPqMJ)#(z$LmR(e8FFYs)h+n8*Q=l)Lvy>Pb;y!@vDP?v_l%?ea;
z@9jMW&b}gXeQoW(hFnVzD3_jZ?#RWlwul3_>h~t0rWuF_2IFt{Os`VK!Qn9cq1f|c
zE0Cm>OnM1qIfA><Wl!gqjmX=`lS|;BJ96#}Befxi`fWOMvb0_T?TZ-m+k<4)%;y|P
zUDRlBcUqHDIC<N4t3!bTbCcpU0%cO{a;w-uA;5CPf1|zveT#a@Z`5IDHIVTbr5h@N
zt2Gz~_#C47?9I$_>40INn24$Y25R9MNFh=OJ|}k-4A|`0`vM^%Q~Ry){j3TBM@^Q_
zEU4Lr9VAX8r2wzF*$s3J7Zva$G85Rw0e!Wya+JFi&?KC}|7=9UZoQt@{r@(yK=5+A
z`v8H?Q<DHIb93ORC=fdX-l{BE;5;ZR1pwAYjI%VbOf2yJfq=y17J~TS5WLi6Fx#Fk
z&5f-B=4vn`zP_%zuBf<}8}vc|x7@tebzfgwdmc~`f+DFg^}&EB$Jl7PTj)oE(h)K+
z+#Vsn3^&x@hTGnvP^W?K=m0F<F>w0$>m!i!l+3r9DHGWb!cE>CSqXW0c~71^dERsa
zJhWs|kJi`sB930|&haY!gSAoXtuDYX{p||aTL5}=#sj1qkUZ>U*?fL}o}QjQ4z=jM
z#dHV->DDdH&HCk}Nb;Q(M^Wy8DiHRD7S{h3d^(3)qg5ODd&uFM3>y=wy7CRJfI*E0
zD+nzFBs<rSj|89+P7V*h6a@3nL^9zBwg;e`ZI+jlo!vg*!r~XOJa{NBn3}n{X&wn2
z&)@6NsQb@ex>`((kBjRVf<4UyTJ&xztvbk1?14a*>!y{JmUZ>z6CN2zzP)L-Sk?`x
z02`@af<Xq&e=FC*swKaM8a`EcTXuRaqd>PwuniNW2ln>%K4&S*$yo;!PyLrqfH?-F
z8x*lKe1|NlpzXk<ysSa-B|uq+kX;HAm(T>_fn)*eDT!8XK0(iZ+t=ISi>KY{SdxFK
zoMKH2;HpRHmg;xB>qp2fEDR;etgf!k%A$_M1g?l==mT8d?kxGleh{W3vCmH6Bk;YY
z>ZrvpK3p48l(Y<pjgY2P0rkwSbMSRszXJZnUzYvWe-0GBwaJfvJA%y<d~OHJH1TKP
z#PjHiQ~(7ag(@Y~U@!w7?oL5H$PeP1N7t!!Gz0Dd*h7b2ehI>+;o{~_WQPb_E=zMd
zu|b0h_s^a`|FA0wItBDu(eYvf@WlWX_N_sX=a$yf`}*R_p9Z)-OuDzq5!P7@_IAhB
z9o_&6?0kkmY#Ps2mv#?sDb^G%IJvmq6LK@p8{y#M0;V*JiHV7xEncv11p9S!3k#6U
zCkR-zoF}&e0HMFGu1zZhV)ub!V*Cuz10+(;qK|+b=L?Xv4N5+P!XkiC5;(E)*NY&h
z$p(ZBdTs=P+k5Qo@x3$D_;%-0hHJy?1op1vC9lYbZ>O31xArhC%x@?L?0=0vT9Loh
z9)OHO!?1WJuz^_?<usy#o+{Q|l)${@VLgNe^Y<mRYVNoDWrb-9R`1fgf471*Hv=f%
z_{_(126dffWO@MQfIKdQtjN7Sp$ylz#||i`vbf@&5y0lB9{?aKWcbD;^&2IVUnrTT
z{#q4~!imy4e(!4!H#awvBEWW}E9iAu?FMwO=8%dK9z{uIC6^nOy4e&ka<@?$AiZP@
zwag3lUqO6C{3W;4FX?ig{$;v{HHaHqfTag(HYz4YXJ-Np*e8I<z`y`CFkyZq1zFpY
zFUcbSgGwLRy7OxlW=Mw4)H>y(!Gia3w`+zif)Ebk_BG;XpzXrn|GS4hv-Kc>w5%E=
z+Ni0ixpr*bCc`*6If;q+Z-9MUmQ-FY19?eJ%^65B1KtQ=R~x)!sihXuRZ@zSrUara
zT<>I3f3);#6P5Drmz?JQMlzfUEQeQX$u76m8-N2Q2NKgqZ)C28fJ)Y>FrNVKl47dh
z4$$uHSqWG{;@aBU2Y$e*zZDh&c4N6@J}02Wz=SF+C>-pn6sUm!#Cu@XF1lp?(cLGv
zx;xFR5QInWuK({K5D%)k5^<OgUp@0WZ6kFD`v0FxIAsQG^gU1Jrx-#7J<nam|G5JJ
z`&ct;Yv6X*fGSJwpTL*tKkkLQ2!gN%>CI_ZLs5GT_=5G%Mb;D*9qoD6r@I>swFbV8
z|8x0x*nv0@sDP^4z_$8#WBA5{NoSy<Z{gI0t@WR>(`^TMU@k!x>D1OH_>BEeTfi!m
zQOU;q1hqtx+c0kn_=3$)a^fHNAqU;S#k)<FIlf76zop^;rF<5^K*5*mKd!TMV5q%s
zu6L5+Xu$=%3regWbpF3z6%4AsAA$9eEQSrZE>1u#JE4nu2k<rjpG&xrmW9P;ZMYpU
zr>+5i#`F5{Ki)$=UlWvQ)940DC*-EO{5?qgh&5`jgUD1O9$T`5|NP-M<TYv}BqZQq
zj{=JC-M?=Omjs{@yW2cae0-!l<+%q^7Grl`-GEBvRfqnmLfj)w-Zx;!_W4HQuiglk
z0DoYPA=IL%#`xy?N_+nk&|9~ZVzqnPV0Xw5sPJlngg|~9Fw+g+T>b|BAz)#<7$w$o
z1tquYz7eVxs734o#;>iZxw*N)p<BX-qkLV!Wnvt#xboL5zupD}7*AZ@eFYg}lH?KH
z=iJwz-igdfs6RZa`He}H)$H+p7pcj$_TPQ&2K?topIAKU=0yT<8B`2Uwm@9`*5Enn
z0#nww?Unr3Okh+CU2Q0Cps`&7^EtHZ!lYJ0Ht+Hn2&;M%=U?(jW(zdn6PK8;0fg2$
z#l=J4TyB$BHyZQ7ws))QZwa*pVf8HjK~@k@xAc@16B7gCM$xpAm1xZeEXw3w-(O2$
zsRLhE&c*uM^~K)MWEBYOWnIL8kh*h+=ig9*NXMdW8|nrcCkVvSFf1M<E?!d4tKJq0
zW7z$>NwN))4lY(Jh6?0Mh+pVBp50`Wc;G(kOJV%^sJ{RZ9tCdRXmVmT;Qj$LY;!&P
zAjBABeYuwE-56}$`&UQP27I7kMv1NB_+6}N5WGQRKDiV#Ep1MI{#H7_{a-?A$_^~B
z;~MoXebQiv2Pnbh=jYcQK?)cii|Q!;0@jBMFzQr;9q1S_$1OwITd5SVd#c4I0Mydk
zwg1uXX+Iz|jNkQX1N$jHf$bsh>%;Cxx8ekd^PH-_|0O{`;sKfRCw^`_z{l-q*$)Vv
ze(uim+xLt3!#i#4@cgxCJr-^YpgBs><mKn{vn~a*G{#vsgNTCDM#FzTcsL}Kx*OL9
zJRR`hEh%;P$ZZ9|qDK6`hcyH!Sx%j%wDH~MmcY!{f{I`-LRCOaY!{~!l=!VX*8%5e
z{<s>K01c9~%d{Q{n5c4&gMAQ1*T!ts-r3<=-MQp{Jo^#^Z#D+yFSkvE>heA|o;2)9
zyk`gdaO-N!zk9)AYGwwgr5mx`zAYtK6rKHOLuV&4*;{DX7}UL$975~=(8m+N0U0B9
zb6t%$VIwjG0mBFil>4k{uXbDik_EB};AXkcJ55QD@PU$4{J;gE_6HA=*}n!moZT9r
zhW;3vk2GN5v>ChV2?7yz-TkgbK%Cy4%Jf@yQZ)b8;(?#R)(gnaZQdcn^=X=jKe_ku
zh!p8g<$pdn4qgRxR<{Gfl&-fYxGsdQLBcHifeFZPpSG%6`b!?d6=Y;&YK3@j4+a2*
z2SQUpm$|cmf)tRC%oZ*FJ-p=F+FFa^2CjqkmaCJ+#X<Dz@1!?@m$|pXvv%~i#{c{s
zXd=ZE8XE8wj@+9vLK+wvmIEnJh-4}mfA6n848I2eWb<?HgJOQ-*WDn{NFj6sDuFv5
zcKk=18J|9_XfB?sEN}P&A6DwR8`PO6K05>7<(&VIXN#+=dFE4-roo;RUvO@G+;OY0
zD$A=3=sOVk-&<w)+kluZ0@ohc2y%3?%Pmk3q4h03iw@CQ&#lh6iT=AAS|bKwkMZ0f
zUjdJlC6J^`npBU2o^C~uYwdDl^q<dyUC*mer68&`B<KwGGlVwvkB-=3;8E5i`ken#
z57p`7<aU5NNFqh#^kVVm`b!r*Jw3UQl7_|%MeNRBh=m7%Z*k%&BI{raKptUw(GSvS
zqt1PSIsr38Qe*tLoXHUZf(GE2Ms#5SyBmAi4BSpI)aN%-pt|el?4mD!&$8HU3i0xz
z*SFHybK?Xy{t-7@esIX{n|^=02W+1KSy{IM;;q>OVHR{i<2rt`0q7rW%*{7*k^dfD
z_fztS>cc^P3Pp0c6an6YrW;VMbfe`zpV9)Qx0hvJ7kjKQMBI3SCHdI3;xA(ciN63O
zR3HXA#=wp)@I!b=r|>V`-TcB0BqFm5a6dBYjk5II04y&7x(%AvHB<+!?kBsR6=Fb3
z^c9o*u@Lobz{mP{qJ(vzSb)P*cJv8wSvI(jgGyquP40g@7}x;5w%yioR24A&0m@o)
z2J3GAH7~=D0c-IW1`-Ys)EoT9&G(l&bqoWQlYq9ZOICtEnR8HdZW{Ru__WzeK=5O|
z7A0b4X6D=k4m|$x>a^MZR3V{xO#k0={GTRUxXR4jymlo<lRWIcz(F^K`VHUTGGOfh
z2v9&t5h;>KYM)Ib)?of<d$}0Yzm_6NxK&vRiHVO%|F}n{2=>Cqw@Y_U97MYR<9C3U
z<o0)9ha5)4`we{e{-5h?3&^n7a!t@}DUG&2-Ohul+wy3ifDob$fohWfT-{2cp`jp<
z945d43ff+_f|$+!TnAIX9J4U_0-c7LzweBLoau|78`y59w>8Fpdi>9iKzPat#FguI
zYZ>NA69I|tLExV^Jsd&U9gwVW&wmU8#n0y`{&brUMu6lrf+dj6&tV6?M|*)X#wl38
zf8Km!v1t7Q7ajr$VIDFkKxAlaKS<CkS%)dm90+((g|JWjIyznsuZ0pRfa-fVa@8WK
z%}n*0K<kvy?eAY|r3|eZm)~n+J=J*eLiEp3`d2LOC1&xa4oz(xIoA<~b<JC=;&eFj
zp*Dgd^M6WWbNDB-2lY0BsV>^te~L1zq4z0uI&G~M^XQ-L!kp}KV|MLtIF6$J0l@vn
zs>UEnncb5*(?9;TnfCG^F;0%!Nd4He`2HU^Sn-6su*<(#Hs}bEpcnpe!^e)72yJ%9
zxk7^}wy#4!!mIZXkm1%A%~pCg=+OV!k3W8N_s2T~?hyEYfxt}{LSM?ncO)ZgLkCBD
zBR#9zKiTM;BLTu?UNTm)+dm2jAh9UB*&30ts44238yPqvu_!s|JO26?87n;#BP148
zGebvHG7dIYRwNc_BQq0I$AA4r)XdT0sgb>yjg_s9wUM<W84nVRn2n{4y^^h-0ccU&
z$l1)mNXlN%4T(k0$l3(7@`(Ep7Z=&>g~TFhX6a~TPsSo?spn`UZe(C%XoMsvi1e?n
zy3TC3K;3EtpYWUfEkvaM7vsPC$sGcB2;3oXhrk^IcL>}eaEHJh0(S`9A#jJl9Rha<
z{Qn98pD+F}zrurLH&c?JBFNo!hrk^IcL>}eaEHJh0(S`9A#jJl9Rha<+#zs>z#Rhr
zmm{Eu4ibC7<#VY^DDr>#Snq~+hrk^IcL>}eaEHJh0(S`9A#jJl9Rha<+#zs>z#RfF
zUc7nCgYoMm!-2OI>>VI?2;3oXhrk^IcL>}eaEHJh0(S`9A#jJl9Rha<+#&FPJp$3k
zOTxFY0O9baBWMccUtxi}|J)()|7!$(K$pSvzxZz)9$ajHaCrQiUxR~<{TIE)|AJp*
zLTxcDgA2oJNVi2(Ci>LU0A}}?e#fbNM`gmAMT#_UXS1D}QY5w+x4S4k?`oExH2`u}
zD#PXST(`CTIHZYS$b%%Ub3Cn6G1*0DZJ?qAMIvgzsx9$;jFq$=Hht#T(9l<rbTghm
zyxbfHa`KpzYSw*1^+&>lbgx%Y1Lg%|#|`})FGjOg=YMRj9@|e>5FMf?)z;QDlAs_|
zEexy_^RT(zTo_GZX1O@HTs&ON5#!`|64myJV`j+Ed^_PmN*?DAK5wV{ma?eEYIq+*
z2;^GF>_bP=Bf~T+Umq3e@rFEAC6IY3$G4{_p?=oQcz+GaEA!2h(1?$M<b$ujind$j
z3s3}>*%AbJmxV~G-<L9@@Ub<NoA}&JDpj)*aTRKT4te<#PBMLnrxYUn@dhWPjK|E<
zbne9J$$%r?jSSV}x$A~_*lUl(+ya+s++0M5Br%bq_a$<Pk}iJ9aPjCV<<|QrIgmQ>
z61L_^lI_)o$|nzL_7Og$mCW34DLJS-h7lz!?))MlIOWWg%43dDY4En{RILGP*G>4~
zitHl4yn1f0s2tK^Gj#v8KUVGlHepaAQ=}=2lp0#;CwgA}WSRk#iR0@QI>{gdn|L=|
zLv2_CQokB>)~cZVak|V-%lpdmb9zw~Z8ujvMYfWZ`Dp{+*T{HVqpOO0dq`Q2>VuVJ
zq!HQ`vAio<L>AgGmvta#eukB`JVKFUV^%2)`uwGz@Dq(plJmDeW2auV>%yvs*AG8i
zR^2rt?ugO4_`rMc4rU&0%Ta@tcJhk;voN>mSH=;!wvASy0Ch`|Ovm@@>5ZcASiae@
zE_&nJw<bpx;SF5ga8qkD5_)37t{WW4bl1nfO=(n(`G8mUMEcddF-(1h6+;<3{&jE|
ztpeV4{c(}4#`BZ+!_3Dk;+0n0;@TE#9Ge`3-dYrkxz*`!O_MXE653fXq%+Was)iI6
zG`QtITgGbGY_pJx5NcUJjXLI<iQB86yomC`x0;yT6nr%up&m@te`H=?Az0$4lgU?^
z?KyDawWWzn5fw{HZ0WD=#{*4`S2A1oa+zf;ZFa3PndTvqnSeLas8)<Ht@di1JNdiR
zXFu%9F3ElT_hD1J-QH9;etpO@r`JJpnw;$>NF|82G>6yy;o(!eYzs6)Y@6Komo<`D
zc#kEEo>H0-pbQ3;@Gwj6!SXDH!9=wprrv*0dW_oM<tA`0Uh5e=lf~`76<w31vodwl
ze|eS}mg@(T8p*2gJnDNWDzuPx`Z;;?$?MC&`3J_`2?FkLt7LSOqE^Qb@Acylp^bWC
zIuq;D<(t9sI<VJPcoJYgEve6)YLmY*EpQnU7GFG)Ug7GFo+axV5e(64#T><x&g^a=
zt6@O2@-G<H3lu@jU`3Wy_Te4-zQu22;yx2#A1F(epfj>V*YiZUb#SY$==pMR_aR&u
z;S;N%#qlMUmm!23WCL~iP)J*R{WqDmhR;|hgTFVle6~#Iu}Sk`a22v;PV<DNFJw5f
zty)u$jsI$zg)9z(fHh)zIT}4GL5r;$35&92kxBAg73-s#Q}1bqb29HRxj>f4f${@i
z-*lvV_Ow(niRJmoXSqVkI(QgSXGF&AQQsI@hDGr$EZ(69Qn-$>YO><yy#B&HgQC2Q
zL?u9C^xoKuV?wwh6$#^QE*?XST7KrEA4`lOSuZ4l*`Pm2#}bAg7Ui4C!n{pNg1c6_
zzrHzmX<LL$+vEt%MkR4?X7#NlFMQ>?2@gNo7~?(CSasGx7IN+EjQc(0ajGAqx^0g(
zbMRlMzP|o0Fg>kQm!>=h2gMZDmdO0$TTwWQy>-f)Hxe8}7E=9_flz92@fJGYR}DfY
z)iA?x(Nw-J8AX&^3EJ-%IGq*n-@77mO}>FN>10>^R|U`Ye=2x3PIiv}t%C1Rm$g~0
z#%#P$U8jYyMxzs?)Qe~{E?=-b#xU&U9YBNMwyw~;|JeQK4~iy136*C|W;4ex%Y~Wo
zdZ>e6bbM3Q9TBJvbN$&jp}SYKFprcX#oH&ZpI#?MDW4gwr?s(eP3jr)Mn6H@=JVJ6
zw{^NUHpSdQtNTJX$A@pX6kTpyUGkSbN=HL1V9M;QnCCJX6lNtry}igx#y(VqEd+iX
z+LIRNPxC4gVe^YMt8jHD+|!<nZz(Dny*5i9JZYjvJY@~JNGITeaXv8;a$IpH>aB#*
z^JDty=8_mJjTmE*%yY!MP8A>3Gxzfi6e+i#LzS#sHJ-XhT>XgdkARm+{JKA+d8+a_
z3^WW!v{B1w?#Ykv8@#}a{pnPh%VFJFyWX2zX#3NVF`<(ALGb)nQv}PM?dezEEeRyX
zOY14^_{kSOQoUq3r+gY~Xxm3tM^7DfUS$_!uzQB;-BZnB>A6TgG(=GFZp8EJ+Er5~
z;ogCm924PCpe}?Bc$S`75y3w{yRv!m9lC0xB2a{T5$@__+0U3N)3xmAUK<oOUS<9h
zwdlP2qR;){U06qQ2h~%0(rS(A@FPaCL{TbI`4tI&a(K8{X}^-HessS3UP4}HicfOy
z3o$Q>*Iw$y$D9ko?Z+g2^O%WvYx>!bQ93Axoyj1B2SP8Y&v@f&)bC}D;o#2u;Ty$t
zNnJeUI+;EeK~wP2*|nAyQs&r76zlx@u^?C;oeSr2HC=SBOXN+iZa^}jIr?b@N89Y%
zS38ytGG8f8f+BKC3eWM-5sJRD>%6skB!IWz;q*+3+oNn=IChRsSgevc00Z}`XBTPD
z7<1N`lpLW6PpIv%C;5d0A-7d|JY&)~^2A(bls;TIi4u+x`l20n64!l4#;O&c3<jH;
zkIq%)k74~6l>=d6MqX93JQZL3V1^{`zqffD6=*ZV&l5<@|NPS@;e(XajX_-Y=NY>Q
zNrz6bZ{fFw*3~Mk6kT06dn=LO?;_%K4I;$q8rHcBlp=RAvbSYEhbl_S8iw<=UOWCc
z$4`;@LC|*T?X~hY#Ly$^!tOQR<^HfU$GUa`-?F22Sdv$O|9R?+!ASV9aoBfnmPAjh
zy6eF2<U<4Kv7hB}zvD!bzOxQ~PsLmGC^B-LQh{eG4ugwL<~ykw-@1ROgRQh5au8lE
z`%haqQAo1$vX@NUO4gSIjWCQC7B4u*)S(|B#~}HikG4R0U6gR`>sdm7YbIL=bF3tE
zM0_YeJ>cu@wm7ZV-fio=PsGxQKl(a)3%MqPfNBM!{Ux@uw1G>D>6kNd)bkZNU%Lgg
z14ydELIr}jX&M1}7+ZHJ8ZTOe3>j~hFRi048K+00FgA}884X7+X}0Xo*JZLumY<iE
zp<ZsAxO{D`bQ}B`pK5vXR@Q8m^<lPlkh|JUo@F>jiNG6VqQa(m&!3};d){aM`Cg56
z=kA<mTIZBZktpA2W+qgv)gGXZl_MeskDHC-j(pwJ^;YhJ<J96AO6K0)n_t!%+YKIC
z5_NI7FhG9KCLEf=%NdY6^0u?|enFk1&?SptQuS<t#lVQLZ5H&zxr?GHjWMQsz{h5G
zmN&9LYL{kubOSvfx(>PrinErEcBqHt_3t6tBl{Mp^p{A%rf|d6*;)o?G0D(=Vujkk
zZ#oJbhc#B&`F<v-BeyA&@(8ngL7Jow*Nt&1e{R$J)tE&PJR4I?nQrx_LzBo^#KsE!
z0ih+|8LHz9UdrV+^ybm`97ePX(+9)%uzQejiy7mP3Agc6V-8;+IpgTBW}{3~DWf8~
zQt1~vcdv_J-(N8!5;#ZdriOaryA5A&EfR}{-y*YZ6L^Sj`z;%4UNG|QB%NpH#UbzV
z0#YorA1<@gXUe4KJ#L$C*cGXieBBg85M~5o0#DtDE?=!KOSANJSjXKbNa|4b@ZlP2
zpE8#wQKnXYGau>8ZnjmOVE#cCiO?m3wy(|YS=7OyuNBs}CY}X9qJEuWIuZezH}_V2
zC0PvxSUST$TEx_7WUxHq;F&=*aIEEU{YeC)@fL58Tki4bkSqzGZ2ZfIebi|0vd;ZF
z5i~zz63A-#@V>r^-gTBK;H4C3hul9M{Tzn+o*24etE-1#_cJx5bU?Xq^s5|?m~p<k
zuPyd-3%sxa-slS%PBQNb>f~8Gb-C;3y)Knha!5z!gQ>~YU*!-9s0Jx~1Nj6`kd`ya
zq$&HTY@j(Iq0iiFXvaHJvao|*&njbdrEUt!o1}k-!7<*YEu{>9hc;Bd{65Y8+XwEU
znAxTsqQFP%fwCGyM{JQgSh*oh?x+e62UIS{afj|v)+aGPcyw*GV$gx1#L;;J>9?2d
zXnNGww_OQilXe(#?}85s=KVNqO{ZJszOcdbdMoQj1kA|B72$Z{MpoqLorTG(t?S4c
zR->3c>XVk;O7!*AL@F12UfZ;wI{X(aBRa3E0#!vLw^qKzjwjjByLRF<cj|vVOOlW)
z7P5s2Te*4^LQb|THu<*pe0n!XM6QKF6sp(rxa}hnDcdehtFY~&F2pe69<|sP37?z^
zO@tV8X@2vX(;Xj|@yF49FQ5eJDz!r@zF$S;os=;rKgxN2p`Az7oUfYDaF6sU0hEFf
zKIh@9=Lzu(Fdl`0aC{$d-I-`FrgR~%q29Dbs_Z-ZF(?UQe3ZC2jrZ{w;?GW6;2l_T
z8BCoqUf1wD-|)=Rr6QtFDBX47J<}H-l8%zL*UfPivu?P5kK4M51>xqi-&OMLoLCs?
z@cLr{dL^0mE}@QGx*JBX)P1BXUnz34#;-)9hgni69tr-EZ{%yQq49CZjRb`EM$R*&
zf7ZLdU8tc)P#F1sf5p%g4)(RNj97N%ULO<dyLZcH8BQPO4%@!+2D(0ee~8mr_UWDW
zvoBEOtB1oX&B0nLH|H`d?~Iz8@kOn5-ssyRsWPIn*2Arda5UEHp{+9w)XzEZy^0<4
zfQGY9Z=IVF^tA3gKYXpg&|}uBj76QAadPQ6hn-p~+bSOVqwIshzBe@&14h4IBBY;o
z-RoyHMQ8$?KL?w&uwj^xWfxSJj(c8QUJi;6dS1>M*WgDRA1tj+oI%UR-V_Mi@wFS;
zv6J2k&i-*N)G4Z#fg{PyIxYKk06gw=@?tkGmr?UXD?bKRcqwzjk50|7D-4piFtU%6
z^X&I^#3<Rr=Q>_Z)+n4Djk$FX@}quSYUG(NeQ^Gz$>VwmohzJzXZh!ZKPoH!Ax*=S
z&ka=w4e#X$7Cw{YLetMIgSk)XG{mz`r^dXpaQj_#ipy<PRX+V+Q`|}3%?~>-B9cEM
z^6>Fy^VP?r(2+|jQ1wnn$QlZtE<MA)9OjRz!&yAUTcO~V7tyQ9lWuJ3N)-C~(fsPo
z`YFpwu|?2-^BcKY|BK(q&hx(x3^b^zJIqyMdQBBCY39%ecZy&cQnAcbMA6ojmcKiG
zXk=59Mj)*p-^GaD^MwB9F{-0K%ySA}3!4=q`1X<Yx+eHwYYM!pv-P_CjH$KUklxIG
z2@>sjakj8B4n?&D3Dwzk&wg3EOW;@PraGv=&pzQ-7H+dTzh=qyZsMyG5X2-1c1s{i
zHRKIm)NZVCedX@P_~KGg%iY1w%gfpA_;~Z%PpKr{i;cB0gXm6G{Jn+iHXX<IK+c$`
zCx+^uulH;+)?%2Oj>o@Dj1)@|6&YcMWy;-Od!}J{4`ISGfr6jq5ak#@g7HI{bjE@>
zx0H}4Q{=~EUKWb9__fi^7q;<+&m{>=AF<%%sp_I@KMq~}K0T2&wbNN&mIwWSMvL-8
zc6H%Ms!*lUq22ufYo2O@diIR#X0~f7<s)u8^vr1c{uIP`M3p7thTJ2yd4&G-8^-z&
z|97|>iG7Sv<Yp-}5}g_OoloYhd$#aq_Nl+ABVFc9#M5Y;w{}41NLQwpZOoJktxC*I
zEPXU1*%~n3JX#&`qRSUa3aHza5BpwP-0yR)(EHq7s5b-0y4%n<s%x;e9^&T`U4ww#
z8he?ZiQ>wr5``BaNM_g#r%INzBn#a(Xqkr|a_{A=-6N92X&HEfG#ySQ|Nd6aDFr#T
zb!%I+;tmH_X;&vT;(og*D)ooy9JoOkbUK$$Cph7+bA$+pv=jGR;_!rO&F17Ota2aB
zlx^iX6I~uB4iW@TcJgyfiA^kg3!&D?>WW^Vj)QfRjhilyjZMS<G5;LNaIg0r#hBv;
zoU?xRtNFN*pD<qBW31oS6W&`M<=8nEUp&#B3aIqLAn(F7-HKS>s@ze5om6#faLFbA
zK|2^pB+yZ6A-MMxyX~n{;e?__E0ufu$mBMPYb>ElX5<xBv2D6J_tEEMCsh|hTR3@1
zB+63pJiHsH3+YOqoz!(`j3OZ%7_2963<)}#leq^GG5LI`2+KR}jg!uVWm(H3`wE;B
zH!0K-=VR)lB84oIU%Rm?Tt^m`4$hs2ax2vlFtC(4zp~EmR3p`JrL|REZe~8QTUlW{
z4CU;+_iS?cwYKJ%iuRQ67Mv)fwUg2q>t4B?1KB2=A-SeWZULXUVw+}Wc>86_<6v3t
zH6t=Z9ZDn#fjy7GR`INFeWaryY{Y!Q@0zCC=6ld<7vJO#^2)_QRjo=Y9tDz~)TRu}
zD|>w3gW%@G^|pL_=b<oJ-UT~nTqEli#o@`^HFuA)FHW$iWo>eecPLW1FwizY<Mp=(
zxtX-R9ufL*L=kja16acpBv7Gc>PJska9m96rH$A(ds1Z@9JCoCl?&7!F5(F~9ITBp
zc_&!A#nHokdZLjFpI|v4f-c=+g3~i8j1{n6BGNvN9oGGQZy_n;p@~JZwG(^&I!fGI
zv^b3&Pz&7EAM3Gv&D`uWxCxk$hg6EnDwizIa$jH5euZtm7wtI`@`jtjWN3P%{DUTJ
zM{Geq@4M}8QU0>#_aWQe$Ib$32d1AAuu&o!5@(q5Ol7a1i}+E)@HIYs1$Rna!Ykav
zkzMVrgj$*PVrN|5v7?ULq<xyAWU}D-n#elk@i<(jlY)Y$9+k5C_7lhp;|lIV&DlYe
zYLxepMQKYSL}ZGrac<Cj?`bd|OqW@!c2_^x5E!Z9f*Ifmhckh&#A#~sALF~C%fdV)
zyTGT*4gC2~Yzr3Agvy5@mF|AU><*(%+?Ta0dvpw@Zo>UV$T0=uL;z$Mm9+vjW49~*
z0b(t^n&>Aq2mTn(pu}X1d{&K$nS|^#1VmvzEV>u<Rm})QzDRqp3GxxKnP~T)wtq)}
z;<>+piSpUO+$86?UMq)KkbfK!l9C|Af{FgPgNGbWnwoZ4u5Kd>?-^#vd*(<tK|N@k
z)gpi1y?QhfoWRxX7)Wayb4Zs!+RHKQmN$&np}Y8mF^v~I*;B)`qv>TF;+ST>w9sjo
zexK27m0`~8{DX(zJPHZZ&CZkZZ*%mYMp8fxb<PWriRHDNU4=E+Hw!*oEw^^AXhElm
zdCbjA>Q4~ropOOa#Ea}S0G0f3k(L<xF|@~n8*|S(XdlYy<Xyss4Z#;n60~Xr-1C#;
zu*{@>uS<mB=PV((zKLn0{>V3PF8D9W$)43JyyRJ?Lm5tFDQx5smg*u<)?$6CMr@d_
zTr03o6V_73I^Y}d5(f9}KG`$c_2=hIO}@S9)?XxbTWkh%2fVY;+}w~8>w7=JDjOni
z<uk?1YI7?$BH=5%c{`n0Cw8x+ObrXQTDvF7-B*w`;wrU$^%JyQe=K9Iu=Mxf$u<nf
zGzn?in|-9@bzkL!VR=rGNVBvWRGuBau>&@F`>^;tr!3T9>NiBk2#Y^T4(b}T?INfO
zaJH)QwM18Mk{629pjC|SL73s;JH&$HW!)FC!tqat-%sSzryWYps&WT&H_;T64Wsqg
z?>Q3qyGOFaKU%x)3=2=N)Ndd1jy!1U|B_IM9Zb--VxVr*nQNoFte8hRWK4FJQe=|m
za7o9b(-h~s)iAZUIJ(eP?A2yQvxAM)^pdRSQ>z5|ctNWW(I%hBu{!(->-}U1v;GsN
zqy5z4NoGS#P8v8R^i_#ZqWmgR>|G041Q^npP7;#l>itWF8?9O5rVrb<rq=0YJ}mhZ
zY<F5sr3k}q{@7Zu>@2Dx)ctl0`=(8H`MJ^}mM+41MA{EK=>RwijI}Z{NBC1MQAKVv
z@}ba=7QB*+>fdB3HPblbQk)MDkXb%FMQWMc_yC6=IQ02hiq*qIN|v&fmUphH7sPM5
z?3lMpO=tZK)>SZ=I9T?PXZ*xd@HT^!+$A4sioVxK+MGeyHm+1jeonqPn(p#?&L>G*
zNW0T}jP<Jz$x>C9!Jv^Pa`?}IGC5V((UC9JE|1hKerPM7D?r0Td1+1Mh(hvTFGAvr
zNn2B#bBE&+7RoAr`V^hWZ(l+?XYUPMzUK?ZuD60>Ty)o!Mf<2v%PsNH5ATtOVNTsZ
z$PJF~)JwNQ^0)(=Eaw|eE93<;3$ZZVpVYOv@CHZd)NXH(Tg19ZqUl>xDp5`QeZB5I
zb%{}87u|{Oe=)bET~UD)!Q1?ne}3VCR!NH9jrI+Q_JCq*Z}4KU)$y2lc0QEz?TUy(
zq^+|atXW3YdMiy;Z+<RMYw&hn#5X=i(E6d<e(#LyzdC3f|EGh-{)nCPe;e2<RM&L*
zUWw^-QanuIq|aWBN%kt?tyZO#eQll9q!me-TN+ibzHG!h>B8@q=U)C$Hf&ZVLmR=c
zKGi7j?28^SffH9x3a;Ic26iT5y(7pqjqHr0wdZ$LW{R*34D>!s>^L7eP|9zAJCfIz
z_<7_aRDW|lzjH20MecqxPsr@ydUN8GmG0eiW_Q)MgRXnsvA$tR9VDS&dS5ZdY}1jd
z^_+nK`OOD4nWhps1GBP;9VpK_W6umFLSjXWc*%j|db;Y5ye?f1QY)177S>zGaNOe(
zZcR-Vtep}USzkmQvp=jm>sHP=L{7~)=qh)8dD-_W-Ns7ElU`vi&fC0~Nd}cbN=5=J
z`}E{P%u%H!clN>4NjFvHu!7i1hWA}%jcXd;*D|mQ6>nC)PAApY?^f3`9PDmaX}red
zzdji~yO+2*7cEa*p2E&MQm!^H>_V?pur^x}#b%!OEo$oN?3jb^!b(XpWo2AH4>3dX
zL5%fPdZ#4gvuKx<?BP+=vGdI^3oEk*4nN}4nEHi))?|D)x49{c?6vP)^7H%Zs^d$o
zgJ$Uis~)}FVOswBmuCHPU22#ny|82ZFzjk=XHS!Q<#p_w#+O269DjaQv+NBMrFCr1
zZ9z}BNUE_NX=}`R`!Mws<5`aw#jGF_x>42IXw9|FyudiqmPW5(-nNchY-etygF30(
z>ksBs5(P>`$N`ylS&<7CjRzMEBS9fn^me24lcPVX!pJdYEfUmt=M{hOsQCnA(|Eq6
zjA_*`;*r;~9fafeO*0E$Ff5fFi+*HgN!91|*de&RIj6q>28YQP^;%JK!8Th-10Uv9
zO8;rnby9odPQlIhY42K22|{-D;b#pw`u^;;`7+!&&za>-^ih*d3pU*M+@C#*4>2-l
zsf6o`ex}e^q^%4=>5g;SeC9{^#Wa`B2c1*FBOTRXwnzYCTsc=`K3%)f7xXGc=d%dh
z4i9H!k=l4z=$PDlYD5P7A&o>i{l`8Yr7x}c2C>d~C$l0coqJ?gv#?R*j@G6V;oF0j
z*%=l1@$)&<M<+jpK5f1fB1}F<$)G?S$HQzs#^~;}ZKUzRqNe9(BZKNsN9;NFf8cKP
zY|N4SEqSrcjPB4GhX0QWYXt-n1)7_WJo9ddI5MC#`gOp%OSj^V*LL=6X;WRpBdo^<
zAJzm4lN$uY0}DS3NwFN)h0O#hX*$k3kup_jod!uu>P&gw|MbQ%#!SOBU7Y)|#7Dh0
z9y}_2J($$9cFEFOnY68t*+B!%)zLSVw(!X`EH>7~UJaYvWZ0NAtfG4QK^Y4~MPAnu
z1(m0gG1+65M@z3rRG#PZRKTy8AAQ|S^?fyfrVTAB<DA@*GFtOc2DQ#N3*}z97)M}q
z6^~q=*!T@%Wrd+EH@SjuIk(xiW!z_?!m$psgflNhhRy-M!8iDAjmmtsHE{lp3#gQm
zWD21T%*w{tq%y*vzv1}lY@%?@Z-wv+j!Z8HK2oLj9USbWPNcPrZl3Vf{Xo<sh%N^&
zXsq=#&0X@9-{-dtbo?U=9a#{j_$#r*Ug7c2Nqd{n)S3QJKK&z=pX)+?vPZY=!ly%+
zSK$?K(bN>zUMx+05k?L?XG~8Goap|Ndao2K3%b5<8{0pjMf(XfN<*VsdOCTGJ0eHg
zn^k+{a=K_ZjW|2xAT7QZsgli{?&#<j%B=(HG1Po`$oR__hh-kLg9Rjk(9BtK??hJ*
z32%0oS-mCsDH&ldNO=WXJvFaM%zcd&_@kVlP>On)`z6efucqX?u{ECV$0!s{zu({S
z>$M>E=~p^)GENVG*DM*gvNNxP<jFKyV!^$C_E~FXuI>F|-CI_%Dw4x@F^RqJFP@6x
zCe#Kb9#q1?ct2`|9`pI`5J8$6<U|pK+1wNCxhfK}>S{SNYC;^I3HO+OMo=%@J%s;(
zU6;mKwrzP7g~aZtZ7Yn<5n*<k#4NeXDYgls{J>QL-_M!1pG~;B;%=(luEnl2ci)un
z=6U53Gks>@BA#mSt<QW+C>t#t^s=)jIOLSqWx}D?1d7j~I82nLug4(OO&!k5&;+`>
zZySYCSCe~7%nUCMitZUQ0hX<A68@*L0(l+7wCk}PnLK;RFuxN_`ekg+!BrDMG5KPP
zN%xe?x5|l)?_W>yuTF2}#~$Uer==&bu%u)*5LDL>mn-axS{JbSD0{2#7vmv9ww|<P
zT{6kEL~|_@95pmG3(G#Ud5}4Ol0u)Ako;gLQ5e%@EpM;(3%9)H&v{Ayu53j&C+2fF
z?!?Zpc{a9#OsIA?_6VMeQYH6<0h+QV`wx%lqJ}xqs!&6_4BjXUEuU!Qt$Px~Tb~X}
zNY-WLB_P0$+PLC7>F6GI*&Md*7v<z_1xs8%b8O;%kpnvu|I>Hv`2^ye<ZHqL?eA7!
z#CdO$C(Rs-kcu$cN4=|rNOvLhl07-}wXv4ZAA}Qd>VE%GMl6&04C$*q*J{yvEZUm~
z<T%+iT2^klZ~G!<>LNOG=PwxjPDHZ~w|JjkFjej9$OtAwTi4ZagcZ9&KWEK8@(Ae)
zV>DOS9kUcw52<LGj6EIhXQ=q8iuVouEHh&=7&r62uOsDXgMh6*9E}RzXTcW&(Db;n
z%Y@>Jx%Ub&WnYSA3T^Qd$7Ex|h~0c6K-y!bz4U-zmdn*zR_(*1E}#DJLSaI5?K@Fj
zc)--K$eF13*4kwuX~pt2vfW4fa_=73C`sm`wXJ%6{_gN^a@}Wfb^2pOh!LhoWhq~H
zePHUz-KU-oXYI`k?#0Z_k;Oh4hfV!VW$yf=txV><{0>S1K5r$0&L*M$2cse>rnO4a
zjlJf5p}Gg30%~&^o_1WeQ?y1u3(Lfg$q>er^Ue=8%OPB?VY#1%Eiq1G`Z&Wv%Hjyc
zA$;1LVeN3v%B2s*N-h3_cG|MX{rEOnS%MXov^9guvyYxuk7)G#@Y%|q%vdZKLL4R#
zu!$=WiLdK(Mh9q_q&jUei(BnZTTpKl!@gsA<JgB?TjSfFHY>f64XN~L@a;zVNwV$e
zP}G@5m5wga4I3~Uv%%E;vyw{7?`IX$SAviCPAF#6`}*=cH%iGe*KH0Ld)v-14ni=H
zvRE(#&D`KV@$QGmhhQ)DEu%u|-LNu-_u4^u{6J0%#YbFDUv1THS@b(r;`|X-Y4QG&
zX$_uKb?fpWx@BgExlT5ZzV{Rhst^QD3NP~TgRm2{%`tjYb&u`!r)+lGn2&VJ=Clo0
z$2(P8d}LClHS2FOA}(F}6^UanNowF$C(jBqW=Ia@)gEfWXMFB!n<Cbn(o^G`{;ov%
zA!gT1OszkD#k#NA8>VgJNiBI6OO)n^Lc5)hysDba&S702bVb5PjC~&>I+MGLv2%qs
zqPLmaP^s_Gg{)L@K6IyM4xb|m>n~>Mke5r!^v$;(q8WmzctUa7^VK8sZD+f<P$710
z$=o4Lrz#G_o!d$((E+m*XT^QVq4zBj9|9hH1G*uaGGc4mT<SMVm9*nhq$OTX6j2KN
zU5^WY1@UHx{kyakvq5k4`AUbi_!NHB42P!OF>fR{EyjFvz7gSqR?D4-d;G8QVd$fq
z@VA%R{PRv6Ghx(4M|$e`Mz!V_F%ArNx8aW02=7T^c{W@rDjnNIw9W3`SJiRxgt9iF
zPANvU%}<S{zpOQHKdpVord)2*jTo9VZAB=!%0qSeGfT!Q!{TY-mg?rmNr_pziZ=zM
zQF09Y$e5RT-CijnVee6BOlyQzX0dd}(T2hf-xn~k3$g@A^*sz2PUBCrLlrn!$b3Vr
ztF|W6N~zRClk)w!B_%7`eGZF<G#4We-U%*}s)^^7sysDSNZYPB@tP-vAnj0utYZlH
zU2ap&7`5!pS)RERmlP?JGc|;})-lO&Hqctdx=4h_j8WpeOCnf$<HWmEK}CT~%YBqn
z5aw#eRgFE9B^6^FnbBNTOIsOYCXLp(XlKAJTu{79u^^a_pP54+fA9Dhx51DsbT_`?
zb8upfM^l+&Vd1MwqA%hbcA`N&3eihd$ujRHOENC6ln02XUr)Rg6dcW4xS24eRbEAA
z*d&7IPWE<FCa>=nN<1Q1=F(?Auj{y=33>5l*;zx-A$MD@GKo&UxMMy}p+P(8t+jlV
zTLL3fQ}NeoB?IF`H9VVXXUXDZd#-fFZ1~fio5eYuEA)w^!qxw}A>sKyHzaJV|ND~$
z6Y8_}b6i-SQ>y(GFQ`Ny(my^h33Zo7nOacL92n+T62&65wMfr}w>@I9U)`(nwYHVk
zu-~3uI1-!|^ApavSo(?jEw~O&_tN&Khnu4_!JFU^qtX!kz5((OeEN3;VdMHWI`wGc
zc*ITW*H_!me^_!bm51HywvBAO*}q&{=@{70RTjX&<QIJL2wi8PzPeVhi5D$$N%wT4
zV`2r|?l?Nwyd>qZ-|s74>yp7?EY{I!DkCJedv2A5Z=(L76`>JBK`Md`(<s;oZni3U
zR}Q%@KoqX;BE6z4gKSFE!}PtxgSbi-eh(JY%NCAT9`qH74kka2R)Uc|Oi!##=s!%&
zw`DTsT-3c15<X<W(BAGPpV9_jqpXVh^k11+K}Vy%bLe<}>hx}kP}~=GpR71v8+{Vz
zNF|oXkV~ape%|r4k<-{}5@m&LFvEYFRzB_>d6cJ7Xh<2E5S~cG&VzYp8Xmv87Jr?a
zI_63h>!=R{tTU5T5P$z46LXw++8zj;3+g!^UhdF@%T0gEsn@y;So(^g_Kp$D+%X41
zi!zce`GIXH12tpZ{Uqra<8p~rt765qIH@3gw)T5?Gck@u&lF+j`<ovMJP($#QH`*g
zJz}-HxG$|<GJ3YR7yi`Z0llfYOS7q_|Le|N^O{1*P3Q;l(t)!i;%Vb=MdWycMR~Qa
z?gxCIXKQ(lm~FxMnuJQ$KKR4)CBpfTqrxwha2O=l5hoojXYSf$ullY^G1z!oYJ`)}
zvB2lx)_8KEtDu9AQH-lSjw1@(wQR{|@Y$T}*9&^fh^8GP5KJ-BtjC`|A;l*#pKsW`
zKYy&>p-;Bpm<sjqDeDS$_1B3O)FznRCC(l{r^V@)eoAvt&234htjkIGgXG`Tf6Oh{
z+Q>lcP0)nn*y5k#?687Y+K@>?kw;%ww~TykKjEo#NSw+aeTJhlons>1-|Lu+OvyUJ
z3ki{b-(~fws$9P7Q71x8DZ6Y(nRtW?3eH@HW+;w*$Z|OPLa=l1S3^|0O*+BXrK-fM
zC?pWu<+{ZiY|$15kLyqdLD(%7^uak2KZJc6&CKRNo}?0i9<>L-x%cvqRxz=X1D>s~
zrXcl*V(O(-w5&ZkBS4-A4>{5h>vY%6e61dzx-6ogO;*(@7p^SbY@FdR<~QPp>Pvsv
z38`W~?ZH9rIl(Ze>UxO1K-*ZfUzj#BsCb-b*un~vD}>nmnwc)FX$RROZNuLp<PBO;
z1hE?BH<;j74nwVH4@j&`W!k=v#M8iF>-Ss8gfAvuo+Phe?IP=H+!te`j2R^k-9*qz
zIQ@(@<nxv#%KF}dW&2OouZ%&<Ih_Q(@r<l4k8zlA42dVO!@O}UFU;SJzt8DJ6!H1;
zOb)io!vqCaB#p8K|2<<yycpdJ8}IL03rC^wQ>{I=;Uv?~r0>1xCFgyO(`rZ-**0Z1
zlfat*8;#x`$4b}tAy8Y}+z=}S1;M+SJHEaBp|?<+n+%3%sD*#MQn}4ND70uS-jk;Z
z4f#^nOUE=i_IyXqIUTR#&I07rHj&ZaiEU&yZXWYLsEFvmX{2<e_d`USej=gN6DAd2
zDkw^YvMjGMLApRY7#BF<*rOC-RQ3XXKuln4SP~<zg5!lK!}s89PS|9pJtX+{F5CyU
zQI*EcHBaeL9pmuQb6xgEMvyw9$U>`d7K}eXjiQq>;eN~)PbK%x(cbY9<=QBRd*Z8N
zYSiz|>z5>{&L=bMv(LJV4*DX3b~Puea41zcMt9|O)M&)qyP_h$Whc?16>VAPn2Xy(
zgtIXFT$i;mG9rp+$#K`CgNFhF8stwS<Q~YUN-(a-F0Di8D_|P!{CLOiqv>MAN+sak
zGpg@hueWp%qtRA$6D^Q#@cZK9#z#LvCtZ71Ildke9Nr4zpF!|6^;7aMY=VgvH@R4z
z{IFEyIec5>{AoXmsgJkIus>docE_q@H78uv{e{p|6m<e=&#Wu73k~VBr&27dO#}1a
zauvCSE<0Y}_hnQ``PY7QP|SIZvy`GL<;)Xeb?*bSIF`s12Fru;X`-tG)#%KaWcGC`
z$bDMOw1h5+$Oyt(ySPrGU=y52wYJ0Z87Gm6koz7D9%5wIXK2Aw*^kzq&aM!5RtJ$`
zwL!sNvf$Dhvwdrawf}z@d#5eI+NNo@Y`dz<wr$(CZ5v&-ZQHhO+qUh$zp*ygv+<t8
z9Ct=W#Few>W4a!>yVoXiLO~D;f&~-Rf&l9dhHTl&Ns8o@=55VpCz*Am_4OH2V*hCl
z7XBI%Nv24CZg_E{mZ7*)rws@D<x0tV93xlDN$}Eu4DNHCq{)}j(dgE2@2@BsDF;de
z?Ge}V%#6HW6p9t2X&BX_R`&UNuBQT%mRqa4FB&6WZBN%RUgHktBQCPF=x56PCv{EG
zFKCVY&D+Wt($8+(F*E=M?{ErT&DExEEy17H==ZNj-`}8K3m2;IQJN>C)|%fB5Ze24
zjQpM+xsR6l;#$}r#mzH}(@oJ&M%&F1PoF3oJn*cyi<cQqPA(9;Pa7T7jxRXyp+-Cz
zCDiPEY?@r05^syaalB(SwyoB@Bxl@l-WqK@eD>DpU4Xtv^i;j`v0t$iIid5?v>yjk
z0jgz-Opu=!j_kfgKlDksXbI{(o8p?!g=Qwn4lgP%1u#t%65Y)8%RhC!SlbQndKZ;f
zoL`0NdYE0T$$-ovYdagzBu1vGwm&b*V41#|c*KXnGa<AG6j8E2{D}?Bg<OS_c3wIS
z6oY5z9~r*@6Zg2L|D_?=|D%r1$VmVH{q6C8sG~P)XxVLwqWv>9_j=~dnbWyKw@m@*
zRv@*D#%ro|(QM`CkZhpi$i)h-A%=fmzfT0ig%lc&i|9yMxSdY6X4yuW+a=5&Zq9NC
z3s>zhrxUY|GA@nXvmvs{%OR5~F=G-ZlOzns?)kAYbiKX(vL2sO8{uzbz^O~m!~L|s
zUJczp68(PPR`_;veb{+8;H`c+yYKAy^n|@+a(<j0v&Nm>eP1&5c)zcA@%XZQF><oI
z_<oqb?w_?N2=_tqvjkk}v}&m55n(TwXIJf<Qs_k;i`%wY&0k>MMI94cS*&j5$j2XH
zYMBirM#APP6{Zj>>_1?%XipYHC0v}i_aaJyed?UsHCK+O=%#DGojA)U4EKA5K8y9b
z^;}<PB0L%14^k)S+#V~cE+iPP=d%QU@Y9|VOo)}z(Jy@1rR%668lU*8m;=)B{dF)=
zll|H$fq76UM}K(2%-m3XixFpgT8D(#$z>P!uZc~qsI|g!>Y|Glj)Vn~$<EDnX~+&8
z6|`m4bk8q=9%#S3e5}@;VS<^A7FSvg05#vyO9u@lzUFDTC<+k4kz$Nl;i^D{I<>@0
z-+Xw|wc3j%F5B5Sj3~#$Cem6x^oI~68I|^mAf^MBM`;~pL&b7qS)EZzwKxTuj>gSh
zo!`Srjl91Q^GO^Nxj7d!3c^{j>=us>&iDvm4;%X<?}aU?A@_k@t)FEsqsuby_|=)r
zrJ14oyC0OXx%(tAyg@Np`FrAc>^?9U1kz4m!=V>LR%_giC6UG(*XAIid2goCs>9+u
zx=dQf9w_1|kYUF~*P<Cv4V1}%kD`)tB3>8*u$G8vDjMi0iTCpwN>*%$mOrC|c!bRu
zSKd$y=#bA3X%qq0l|uBb4VOx!bd;ZYrJy3n*}6lv82vAp`e!HuJ4#!<FF>$US<Mu>
z<=sbo&n=si-zE@<qINx4dTU3935j{0u;i>C2+R~%%|9w_s0gaXWDv=FmrPh}6MP%A
z$-Kt2T6u3#f<qmZXVGCxP2PQKubD|&m|)l)rO!y)TOK)5$Bf>j@fnNOQ)ferYaE;(
z+FeN9O|TGW_aS?-=TewY%b-Qad@!SgN9(;7{Si%8Fb_4EZ)FUy)Ppt4VJ2Xw5%BW|
zYz7?Vz6j!zBVKu2&?<-mSqgb}x~Z#AVif|n-V$vxtTeWlTD>MC5s6!4tnd_g2+*d4
z)w_zzMm}Twt^Ou*W@K?Yz?}A>T84gzG{16F=(uASp{Zf_9`LADc8{7nlRae(#*8EE
zmN><X!qbkY+jYayi%@Kp#h}!5um~STh$Hq5P9NpNc-5$WdqgJ$CdwbP=ji2%E;XO|
zWdA)0^D$JzQSsp(Q$_~FaivFN#%wBszfg|fk6(}nP{pOke=dP<-Vj$tcJrS~c?TRi
ziPT`Q(O+--%2vFo=xlWhC82x^<huf!`r6q3suysabY!4N?dF}BS5b~xunMyV@$+}!
zx^WK2z-+xQAKS>21p2n>WPld?ohjasXVhY)?h6c8fN6s#A#%q@vpNu0ZkybLW1$xw
z2k5PcPQzp%g{OoxHpF(0e`VSC#!2o*5D{`Dk3-CVj!CmJ5;ttN5J;{l9XC1CK{+P{
zSW?h_VdenIEON9<Wy&(|V<DK+u&|CK4?v9=)?Dm;{zTDbybgJbvUOXTlmNu8CAfCu
zYDlbfX}FDM1|-LSWeQwJZLCSSoZEhBJ|5ic+5e7XE){=7TbF>R3j$`+>N4aj97%xb
z%H22XBg`mYn~i^M6O0G{8?iOz&pbq{XBtG5sWtOelaGv(>(tznmj}kcMG|+(w!iC1
zNtnvPK%eBy=HA+ku7c*~wy1)xymEZjlHuJ$=ax5dZ0yPgd)qo}$XZKUXiudRL%L{D
zKazEvI*`y4ae2~EW#v_P4U4uFq%AP@_i(zOhicPSqs&p6g(T8(-HrN#QgR-kS;B1g
zOcbr|EnNbyQ9wux@$Dn-WW=lKu9ZHI*h|GGN#5slz<46Bv;~!ST}6sjsW4CmT|_hH
zvn`5ROviX)gf%-(RH;C4I1NN#Yjp*-pT>Y^^Fj_mq-;hD66wtzqh)$*C6NEAeRafR
z#3qMnu!HqRkQ+Qe^AUyCfn>_rDrX>Bn$9xUfHHe}^mvKxLpXue?<StuzbQj5g-VXA
z7f9K?Po=?rRiL*m=SSQRI*taUFm8@IS|CcY_O>!!C<*5)%od96o^_-IF^0Od%`Q*U
zdN!sG(^Tx}W7^DBo;WF(>$o6oBeHV>FwoTPwK+Fq-)JB2vOIl`VW$BPPl5J}j4hz?
zs0_!o=fp30^(9n9Z_bI?;cP<jb3d)ve8gFu+H=U1U%d+xR`r#MzdpV(amhf@H7k6{
z(6PcMyr@Q45>=A8%ogwn7dvS%*5i1Bq=zk8>7vAOqgC}7B-*Ksrwmhw6K(a!2qVJP
zdkAL3i3z_cae_Xf8PQTp;IzF>!qtb<?{z>&As59pBghN?4$p~*yOY{}dfHzU07-$p
zzgD_t9#fCpn5ml7)iQV_j8F1SW`4p0W6o&_HZW<45avt+Y@*~VDzYQ2WUWQ|q#736
z#x0QR7v{w0!`iJ=pD@;Hb!fJbmFaQvDb%d->LP%$c1^RXyi<T%FPUv$8w&lMugvsj
z%lvo?a=b(F_s~6$LMGIYuMdA=M^oB5j=tc`zL#5=b!Yo~e6png974>fK4Q2)-A01t
z>&ZJLRW?NA{1<QD@NnhwxHaXvvg`f&se|Tnp!@_trX=vp!1%QnebjRltViYs+O;2Y
z_Mi*2?>Q2&(vCFo*0#C(MN8@52R+wrseSx95zZNYP@<=-PrhXEQ5(M1%gFExP~;)-
z6^ai9v7agVa`AUtHdLo0`)%~kZ(A3465TiDI0ks+^_A~4OJ3gN?*2)5%dcN71D4`{
z>K(TKzh}b0K+o{MhPc;iGIoco2;KijXS7;npomCX&Ih*EZ=#4SJXeGgT7hV1x(00|
z5RGqQ{L?!)j^<(s+tdk9ig|wx6bCUMrZeFcmb5&UKFeQOuVKjuUXc>U_&v`T4i}K~
ziJ4J!JML_~Q^tO%zfIT;m@UdkOtJcLRnJFr#d?bmE&LE5E*j&Wzy%e5Jv|*fSTbHb
zj(+{6&{Gq~-8p+C>BM>qjK6MP75JdOpB_Fgx;e_kaF`_%$PorEV~!_2ON-B?sz)qt
zdq(%vFBoE-1TkUu3Lv`POU#C}q2yzE(1L^)#;BTGue!Y`#z;S-NOWP0C2ieJ2*?Qi
zF#sYNW8!u)0U3qJ`TKj*_AzN7t!)Tb3J^i;K7b|4`O0pct8I0gm#tZyUZMgG<N9*<
z%y$$1e%m9})$MQK#}kJ<Vb%S$SFf+3iEMBYO^yAF2ndvFhF6yN90^DdNjcix(3NPb
zXLdxaUJWZZ_CX8U1vCD;N$m+7Mtkg#8SrhBY-4+YN%`}4P;}Yk#l8}+WyI|1PvKkC
zn?@kAdTd@#zkxI1`r<7jT_G`2_~0bstyw6)ID16qPzjuXQ~)4ECSab;Xs8gk3=P_h
zMRkQvIUA?j%tgc1(Wumc?5|w=BgN~<a{E!ji&kp0hl1_WN6VC2NXw}+mgh!OPB?6U
z1MMP1{CV%8-<{rbug;i3VVr|R4&D7~*B0k1l&HDAdp{H*GKQfjP+7J!Fj<M41}IxV
z*vG9^@}afXi^dgtrJ{{IfWT(8N_DoWuJJN%=^95=$2G33LHHkbE0-<f&K34vrQ?#Z
zMh;%oWp1s_AIrIn1<^JnUwMCNd}V$B9O46BZ6|hAfbA<Z$=(gMX4cT)0lE}&Ewgr2
zk_oi+g*^>Rkc`!I&BBzP-EvN<2Pjk~wwcIa-MkP-!r<7K=Q*>Vz+rrNj3L0lX^(`V
z*eJ0j7$jOcnC&&nJ(bmts;%#uZL=&BYKb9_&4nGJxiZv)5r3Eu6xe~5><RRlUT&^8
zxG<}*GbIFd?1Ug!`m1v_0Ahyx0%(BnQ740(H%%M@4qu#EXca0l4fOX3&}XCKNjX+U
z^0LNg-vnN`HIcRBT2Cx4Igovw&~b(mh0>^jIFOT)Fcaf0WdGiUc3OY+!<@aJZ@2^z
zfsy6(YFq>#ZS+Szh4qs^QJqlFz`eBc<&-PATQScDmzR+YvQv>umN=;u)j4K(<ghWH
z>H$|v!n?_6uz!k&F`wXkr>G~bquhX)DEenq>8H3<Vg!|BNYmo&Dj1ntb%<Q2jp{Sa
zFnZcBdrAXGFqoz0LrnEB%??%^Lq~-(imb3UprXH9J!}5z2^&H*YBRg*L&hV?{N0D*
zi^XL^!_}s9;0@TLYo%|+K_+zvklq3;)kBZt-`{_^b06hCL_g9+z@x*O)Zulruydue
z=QaiBDFSeT4Qry~v<;fAq9f>2Q~s~#iQ_-lG3Nid1iXxCWp~ta-{}|3dxNHtVHZLE
zk$=5i3_AzX-lpftf$dCHO{1^2A+DLbUGU4pgne=TOyq{17Yvw482WfJWuHQ}3Mtpq
zy%%{;r-q(t*Cifgq==ENeT`&-ob74aB~|lqv@MCz;_LUkZBOjcim3gs4tzOyzm&ZE
z>`4Tx_brdg?4g1iQ9*M_xw$y7TUpa)+u##vPP=QK?0!<+BrZ))i;{O74iOY>!U{S7
zj{jBhmhPmHj6=9`PFd-g)uy6x99hkY5L)dmitRe#@4mKcidyX#MHVa4SeYrrkXLe?
z%L(LMsT4^FblSH+R=Oh1A>!@ICZde+#bA_h!ijle*9!^?y`P9;+(J2Koqm7|5a3A^
zsB<rHWcJVdI^?|ecPq8r^mld&v{oF>Y8e+Tk`7=1KFifw@Qni=IiTgjejR3e3g!-#
z039&<eAo@dX?0l+xWn;T?hFeK@u|OL)oGMf>w~z<l8c|?rrWHc;L8EZG$@V=L61p7
z=#nM}E|E)&PMRu8asN>RQKog6Vn<}&OBaXjW1vIz|E6vZj~kH2XNxG7{fmXIWT7nN
zDq6b0SnET`%-$X@i_84+adD6uxibYs>z}A^TbrgpxsFS_(|sxdm$~b2lPs4g2ly06
z-k+32zfH!c^Cg}t8K4Ip6BUL}r!UnnIzeP@eB&NcA(sx?jh_YRimuFpR~~S}%s*8&
zL#Gx?l`I`r#w`IAza@(gi@fzB>#A+nXL)FYgG!BZxc7~;bY-?S2d8A6ayi6^o2(_A
z0OU|y4$q(APz%UB(C2dkEHc>T1UTUzh6VAV%wtJj-gi?Dl3XlJ_1Gt`9Zd-={Nq0&
z>PSS1(ci}u4GGC8kuHkaEPz?P(hU$2JCT-WyaHFz{fC92KoJIJSt!brA3EBc`BC?Y
zKhz1E!c^c5FdETp2@)6hA&{*t++51WleF4Bo|Rz0qTc=!b5$`A-CdY<!Fu+tIL$Q<
zIxDNvJDt)P&%b{Vp%OKZQ%>Q{@|r-qnr%Wn%pffT>f_87A0SvG7y5$^i(yGfS$RfD
zIg8(LpnrC-{MtOkqBI{K)e(Yr#9CSSj5=Xb^ec&v6o?;xj-oAt<RrAMCw>P^%@P`B
zZf8gwwvvIQ|H`JGDx+CkO{6(G^LMO5DXfovcgJ`X4qyO$pSjYn#eJxnwN}*B;ajF?
z3yJ+`z@$cLLJ%Ns=q#UffT2MxjjNGi>GiN>CN#*=GJJhnSf*`fa;u<r1PK^hP;fyz
zCQTl1J>cuZqtsLuel-UgrYC{|(HnO)Sp;5&;7VZNZ`wA>INaA4Q~MPaT?i=rkwq0e
z3Qk6KSG3A{mpQ*spN|u7$wUc&6Ws*k<frl%!6;=}=WXCZWErpj^G#;huzD0AIR<)=
zNefa14kUqtG4_OHTTuhNbZYA5-}~vk`?qA4RWkCe4$SM+yDnwqq>{KA?iE2K>GE9F
zmF<_u;|oZdqk#(de?0jB8=5@(jV|Hsqc*^vG$Sz;acO2~nowJxgf3nKBW=N-qv0AD
zH+b|H`A!y}m^U1h92ObZ-nqAn?a$r42><EPD=<OVbv3qa2CD`kzzB<9zGlbQ{Pzpa
z@kZJ|J`fbYw_Y;sZEAs@6MbH9Z?m_RW%7L9)qhCaLnNg|b;?VHHui<JzbtrmlH3&3
zW9%00=%NQ^RLMcZiUD_i?~lRu=ps99JbcD_;^f{zy$#Q~n<?sOSQ>qv*HO01%h5u-
za>~IeHW{o!26VHOhK|n|ujypiBvx~a1OHyZGi2nqIz9ExUmmK;{{}7}+=P*__67sW
zpCMBVgT3MS*sjbAi7li;ht%<Z5t8u}_roW^F0Eqyfl&(mx=l?=kDq9qMcRj{8a|m*
zarY5Vm9_GID^Q8mq-$9c_g}6liI2G23kHn91*~((yq0so(qwWk+0EqE1Tqbp23yh)
zJFJhnzZcN5q<#-ho`OW$yZKUz>V&8PRXV={33Q9y!}l<+_kg%0pH{!)7FCBp%sFMc
z*fbl1wn!TqJv@@1qd^Es6_aJMynJ8ovO7E77LdCTi?O?8%-LGC*@F29xfdT<Pocp?
zPN^WYA_Tht1w;E5&IkJ=dSBm42XBJ@86UI+tJ4>@Ad`_z<3hYyPzY+cP~9`zQ#xLn
zNr+1+nYTN~EuUUUzG|p4{ipO$37KJcNJ0xnX9PbyTf9JVWbqB_D%XoCSv!W3s?wqr
z5pAUO{HXq|_2Gk*7ts$wNLScK7O&KYN%6DZ12cChL=GRXr)gd!OD6nycvFJ{eW%MH
zK?sg=9(`~8Put*d<_EHj{4INc8X^cQ+SzpngZ*wmFHxV=w~i<HyvdC2Sl}!(AF+#v
z*{NV;VA>>+Tl}`sN{&i|e14!vse-C4@F}h1&H~Z}wwPsfpYeh)<RpgLNFyS7p#hAC
znS8rNKaPS}pOR3bP&T^PiWA)x;f!N{(9NK=^LF{eCRrgY+Y~=Wx}iVRNBl1N6@o1@
z|KvJFSA`tnLIg)CdeFovgaG%V6J}%nbHO&k<`n{bKt36t;aZx=S!$exFkmYRRh8Uq
zxiwPC6~>I{sDx6A<m#u)EwuAw<6ETws|}Kh3)?5GK>|1#A3aEBs5&D%sMlVLv0JD~
zR{9!F<MhyS5XJ%bOP+nIHXLh7G*AEM2(I;(aVw&dL#j>Z_xHixm11b}&5wE$2r7?$
z1u1Z<@`yL{<OxRa-cg*=jC+iV^hUlB<Lz0N!=KZ%KYzy7M?YA<CXj(ld^&nK5`!tM
zOu#J{42oh9Goyl`LyflgZuzgI@8&{0aEWFwwWfF_2Rf}@mssorFvtEZ7RKtf<|K_i
zSjpbBPlB<4xK^JUPn3>BfRd1K!<#yLqN^%a$*M$ElA9!@a1&7rrw))dK&_wW#E9Gi
zrNshYx%y3<mJJ<&52=sBVFr{=iC5L6p);KRgOK}bYiWH!d(tr*{!<kF2T{+!%J@I;
zY|H=1sLg*TQ~r6>TIJ=z0A0dOGG>nKHC*9Y8y89m_kBpj`TO~Ibw;J<cDDhYN#et<
z*u{wwWK4izMz5QfT4P%p>{}dLUfk<mchq+{<HkoKGsZV>@yK9^hXQBPrZZG(WA1mR
z51)Z%0h2TJ=0gYcV3^bSWW79|931SJFx_#TgvNme7!I%F_mQc7vN0o}ZNA*RedCLT
z=-`Kog@^Zrr)~DkIoQ}a+@8LcEY}VLEuhl$a1iWY$gtvwOo-xU2?0MI*)mj#X85#W
z+Y8i;D$B$CLp5O3ys0<+!6iCOqxG7Wnq!pR<jMhhjoPWMq}$ZB8^Guujsm5V+{GF_
zh{_i}UBHi2+A^074JYhb<88p1q0}Mu{yM6p1kI;dph?e`$rwDBREamik{2Urh;4~z
z9vP{qwwLex;;$^HJNo2X<@1;Zy~MBy%tJK;vsOB5RPXfj>jjzz1(Z-CZ;t^emlmLp
z3P(3=Cu{U!D>-q&^`{^UX8EW2GO1EbI)>_{l!Fq58?^f3%FAou(5WQ_jctvbRtQ}u
z1g9&7`Su*%5Zr|g8o%_0n^`CBp1ocb+WiDh`H7YO2m?0S<Xc7%%JnUyn?qz_ppWiR
zEk{(D$|OJx>&)e5SbUn37gNS1jFb$k=81pBQAM-Kob5|W!v%?(b~mW;<-mas_+1ma
zt$zJQ7piub!4B+<Ht=uqq7Bg|0p?XX3<ri<emH08m^lg|%?<~tvdmQ$sw8MQRab28
zlqIN6GKQKJ^03>va`4{7T2!VL*#+q_Lvm=$ZeVzF-+RQEks0TE_qwp4X8c8z6Vj~f
zP5z)(%S?x4dny}9&D^?=Cr(9rKP^z8sOz#S3L=zJW*xiuCo<=rgOr-%u=kcHg`nxH
zX!(^#;0Mzb&B>@XazGX;%!lNt4OU<b#nCRVXb{{<#a9^P@y%SaXz_Ei&4~;dPdXNj
zJFzvyJqqs1NLx0R5713v$0`mK0zLJIOcc=o#Ts4_$nNTt3l?_-3}6J)I!qh!sJKY1
z+JW!2S8f^qK^};K<kpYJD}c0-r0Qsz$7r(wjEf23Z>54)4os$jdCF_o9RvVBBOI22
z321|K)`~!(hmU~k5Sj$bp1Lm{^_IhbvB;^hjNn2bGKG=Oh<f+W5KUNokv1zXXDtHe
zG0#%#ut5%)X+6|)kvfj}Nb;>+P!UA0TNQUp*GmY(7K?&)&1rr$3C;E2`brDb7_vAx
zCj$lPA=AuiI!-=G`x54N`ZVa2+SIVFZl(S!0k*R=%q@272IZ20PHrhV|3z$$1|3tH
zJkdfi6PhZzq?kpZm~rl?nf93Cy7Bp^SgN@}BX~)yGts|ECO;MUkD)iv{REdV>@123
z1t*i;aVr2?bGk)~t2??KzD8eniglc9`Or{^4u0>np~gP*_+x8ZZdz+nbxY+lvY*8K
z8MmvxM~RJggNuZkmWE#qI_YFsb%zRIiFM}&9s@4&`jhrrONw&$@@7riNux{IVzNY&
zO+Oz&G7|@oqf^VM;r>>Oi(}hib>R3;XLmCL@1Y>Oz!=au$D;Er?WY37sL^PokU6+)
ze)jtgUZ-IZ>2Nwghd!<{HndOIk`_lkTMHQwSqPZPru^d=n5IEb-VIx0`O0mI;Mfc(
zT>vlL=IIOg!_BD+mykLBKq@^@V!y3ZZd%8tCDB>aVeTzmYT^(8$w6Tr-J%bRualoY
zME=|zTrc2;3NWm*BWuJ*1>cU=s4L19GKgm)dSPCLyoLql)JlTN$9kD~e*pak31#0{
zeALaM5Yhxe;C-)p-rSDlp+KtL?$7Mk?@!el0ms+vXq}eg&v*JeSbDSXNs?QM5_vkt
zaA%PcLL>i>&r@qKW3^lC5iH}jp6iQZ?oz@Rl3&7hA_$(Z0#Lo2;+Y!t!veM9Gbk=*
z2=AY9uaAwMui2}`t`9Jc5WC1u>@%;aKhK>IVF=VHTCjxG^Bli>f8|&|3EwLUwtoXQ
zqjl?P`gv*Q61Bcd!w-I89!WWRoc20N%~sbWQK4l*e%|hEv4(&OsYTqf*6~(X$kfOW
zZUG(EeM=*H;qIb?yP2*Cz7U4UxPPV=%nVI?$FV<L!u%8y6&}g@bwO<tjDD?h7Ot6G
ztb#@#dK*$Ct^Z?8vC{t+z{^7aKl35on3D;ctPi|;{X{BU0r$k=SGtB~Sr4iQ%30;z
zYZT}t0!Y}rC<z(&?o_`$u*~A|*cwJ&R<C)4y4&VYom$KrBFrE67w^~0jj2%=;>g)|
z#2wqBgyBij5XqDs+roB6B@kqh#ggV<7thzzv35OrMT|W<b<{t{(}v@y(RyDGA9`$d
zzFi*Q2d4L**L=C%NrqAKl!7WiuLQi++xb*j;(5ZQ+U4dyHm>uhEj!*+OaDleTs4@Q
zV#LTG{G@tu(}*JK=#}GSE%fk!M-5^|LXD*~bFb&i)&+J0#zh4%Zi%2n$VCc7P0<L6
z1Y-)v6ZzJbMz@ph>Xw#<?Ln>z@~^dKBaA#vMp=+V_mAC36G01Y+**gZ4W0DEyV|;^
zdmpB#Dxo7^zt^!8YJ;2r$eCVw#@}VD-0mf7VxPPWr-U&3LAjPDok&Y1yL+q3nX8Ph
zMi&=RAB_C2)HFiKfBjm7(V0yogq@pdLkdMibihV{Yv;7>4UEQJ^;6WFDj3?X!X)5I
z;Fykn6|LvXymyHbxz;S#yQ6be#M9}3Vh48hlRt+XJY`x)5)u-ks7N`x$1P)O!2+aP
z>)7Diyj5Q-9ERKG+2|sXRq5r3ll-CJixpF}e3ksCYmy`q78`*l%O_W$84xVkA5{T4
z!?Tcl@uKp<;WJNRczKu;=Q)R?RJE}9=^dMgWf;z&9+|^nWFtU{8I4b1M=(>VP%Pd1
zR%=WpNm8<k6s2Pqk*`i9;RWj`M^<ZW`$>}G6K1b^L{tGSc@=bc6xfA9haU8SK+WUf
zSl2ad!fTuAO3#akwf)s;v}Y#KyK*CTt03h$f4IzvLs{sa@E&N%pLL|{?Z+~^n6oW5
z9!uc3GuJxAl@jE68(e5z^6cHHf#(?`TW4IiaUDTa?@)i1>$)78v==bk5Z?yJgZKDz
z@e|L3AZkkW$MzX$b`aA~CdKzPB$;JSiq=ZmMUe}La|4n`TZE8Q{oPIu0XY_1_Ru!_
zX~irL4e@6ghvu%GY5^&0Dh~j~S~@BtxoQcJas$D!zFcK=oAKiotnK%mk7(_xoIGjT
z$NY&wf!?wZ79X$U%VGX%BBj(*?77JK=Go#w)OMraTX`X9u%8GHO+^JAAe;v5>VghJ
zWOVbWT1C+v#+~4L<6N0eM2+u^V+{Z~-oebB6MHI>xw9~#?mI#<Frg~hz%}Dc4Ix8J
zCXSt<c^f+iE(Z*o7M1xf{aGFJ=HWc;2eWZw&xA5u9+duME2(#TZy?KxWrwDfAtp%T
z+C3AbIK=3~6hg5jnA6(3+Uda9htnUkfj8X-4a16si6a7QJKx*d8Vc{Wr*@6wSwOs!
z8>ylEJdfx7tL1`OUHT!Mo~YC~c3@<XrAC&>yy_)65~dpQ>Jg7G80L_|0~WSJiy6Fh
ztrDkPi*?!2!8ARiZ1pIzjwDABuN~7!l0Wk0+=$i-zW5}0TH;ScxE93+N%$gwQhEjc
zWqVOLYrKVlEv@5k)WY+^7yAzOH^%?5dk~Q<#4%Yw6rDTe{N=es1iAS*pMGZYHD^e*
z4KYf6c=mF`i~ZQKUL}zYp4}niou}&#mju{2Sjj^m%RVtpDWxFm3*`}h2wOW5i>akl
zwZmFfPP~TqFgXN$A_{_%ls@f9d~SOt#lvg{13YwqV%cJhElcE17fCyYWGoP5zmq}b
zI?ZJtR9b+i#OzEj8(*<NF(moV+Z>7@XFMlLpd9fTwM5vIX<4`5nM5x?dMT~hg3=xt
z?U8;sHL4OuZIk_OpXO{WCa{XOf5{<4>ygDJqMaGsU!iMqFEuU$4l^J{JwisZ)9Ok>
zrjk$Mx54&P=mW6*p~^>f?<bEaV`_?u=fs2rd7nFT=QB@LN07^#!gN?jf~u63iB4XO
z?OPd_6#3LL!8P57q&a22`!na4v)MD<l7;a|_CtH>NoV(jtNAY*_j^bw!MAf<*A&yS
zU1Ox^8-JkKoMk0RMUvhq^YY`Uv+?W%b9^i3DJF}J9DPQ>`alkT0?^c|$;#H=MRQ=U
z86T)@k=Gw%$Y=XI7LljT<=QpdG@UUNVrA|?=j6nw%HL2<)89Y)0^^*%YGB;b_<_>Q
zqfWo<T9E^B&uR29s00xP7D|72g!oBfyXZ&c!5EV6k5Gi@j%xIQC5x2eNVQ*m+s+~R
zE`~sB>;sI)y0rrduC*{V%uR|MeC$&q&r`MBB;sARp@1>B!B!v8W(!IKg9+W{7ikP$
z5400b%Es08#6Ak3`H}RHboD`$8I<cNoO4|>UacQdpfMEU`sMm`*uajSzW{G{(`<aH
zJ1!%OgKq22?MQ$t5Xy#4_clAoFFsl6`x_Yzd`J&gKBSA}TtvG-jFLG~NXx=6J3vxw
z!Fs8LPzkiJx7RJ*>aA7oxRa;ve8W4U&jLF8ijGkrS3Vz<X`ACl0f#x^oUDPDw$o1L
z?-})OZCH+2oToq_a;c#y$~l@^`Y4HEk0lt75^ZUcDy=p*elX0N%baHa_-&EBwW;7t
zs*uYmQp>46*LRi?$TM-SvXe%no4hEuPis%)m)iUAN+?Z{nQW*DjK`G^Y?Ny-EyBr4
zq`<-cnR+iD;fL9saQXw5ZwEm{N6P8Uwenfl1^?N<=?{H|lLLP73{<c3#)7wcg0N)o
zY?ZjV66%X#B(djzr<Bn$?IxsJdwj;iOjJRaxY5Dvg_7n7_8@lGeanu{=1w=KuC^QO
z2~XM`?;HA@ZXtgth~qqQGxL-nv<MDTDV&f-f(!^@%b58RmAHstQ-r|@Yp)&Y4vt-E
z9~OLqz@l$5CCt`bT&cT_z$C@?V0@mBsMBXfauym?fbn!_2tUR`zZ6tO_o`WA5bwb1
zuX{S$LPp=jp4BC)VFRZ7H&oNR3MwMBA#(N;$$JCg?|Vhrs^1<0=<s2~PMTqp7mj>>
z;U2?B5*jwg%&P|%*Zcr#AfALANNE%?b(6`|3dQKkE1>plKaoTAKudEcr8|^5Z{|GA
zPLj|!ke1O<p)<n3NBFk<LVeXN#f|{r1Eya*+<O*bdKQw-l7Cg8TCHbtt+hdCC@8Q`
z?+EzTZ8e)EJ4=}Si|cp)({!nVo!MDuRjfH<$pz^u^=uIbMeve{Q2IoOX>%(fj-~C}
z&)2~zIy;a*zs`*b9FsM`Iq?h|UW`rXyZ<Ktb?6~jC+!aoeJ(j{pNG+CzLcTl=Qy~K
z<A}75;+EIgB-)W7=Fm4opPDgVR3b{kEW=aZw{XE|(CwNul7Cjfadx>j!ErQ52?BfN
zjB86M?iBubKwScYtc;sAWfh5>Ut9>CGWai`ucxWpe@esuK!+KbSpR2fIIW@OwAqIG
zf9P=GR()MtJcw&4{P~JRcjKis@^i5SsgOk7bht1jiV4@|x!)}y96u5>$)<#93mqW@
zAUOExLpJpJk*6S8t?Umg-}`<s3bKsz#NzmuCE<bup^~5pV|p#=ZAsR`BK)JCulsn}
z4UH620TL+lRzdFX{lV+|d*&jh+*uOIk#kCs^T*@muJ7;a*TDnoOm7eOX#XyqFITU(
z*Zs%Ca4oxo0|%eq^T}|n-L(YS(j%*W6r!ZkTcB*3Q3Az4QO*>by>GK!%O_aaKVG9k
zPJ^|YHOH!i49dJ{VE*otg)Dr`5KCk|?UK`ON~~p|8$8x5oHQKkPX6ldYw{~$Ej~v#
z>CM`9YGyBY(%a0{j6M5ssi3Wg^_ju8eI=-vk++8RvWvoYdbC6{lanshH0T^t4m6&g
z#V8AcR9>s~-P2C~VX|wee?ac4#Z;rKG$|8RARyxzLUDizb_r!agQ(drvm=A%(z2Cm
z!ixAnq<trPGTu!yDIH&%ylX+E6SgeO;#mDq)tmi3ht%0>R->wSPg+4!b+Ulr4sY=I
z<qoSR%*<(oED9+fBT{&RDZ&3+^i=CZ0A3>2M1mN-f9dhi?SPO3bCqGgk)eWy3aN!m
zqRWV3qMMk~#2?Y^@UV+Gt4MGpDKoQ6V4tR6VF~3Zyw(rHd$e_{+r9s`<+1=0$&Ru%
zrE>z8PuW~D{Djf5yt&zZoY@=&0xFpuPo+8*cMcI4r(Al7xHmKg<4jxv6B!ImgR<m_
zUFvTXw@2SRk*9t*ae(Q4OP6*&b(j3nV$c9bV5aH6SYi&vT`<AAE*iI0iXdDS_#UrC
z>IbGXDZrd3j!)lo<lWiV$YE&$G#4q(BK_8F#~vGPZGHCB8&tph)${_U-Yi%V@eW1I
z-6$i|vRTNOG4s0P!%!7RcG;vyl>Iz^U&i<H-d;s}9hrKH`t@>zjvDqY^bcq)fl&`W
zzpym8eIPVcr0I*HtFaNFdNxok<NE^`S>WUwTd%WOOFOJg%KT2ni0%+#HIPGKpj8;m
zaeW%0Vi-lzF!x3pqZ4{MB*E4%20L-Z%e;FcNgX?M3OyRXsV`&*LI|{8)<1Vr3$>Cp
zz}gBzsI%bVL)|OsLdsXiI$~s2RZU^fVk0`(uSOaIC02}_UUGYpFbyYuue)rRDmI(P
zTbN~iy}^>3^(0r?BP7x04J)lm>x>2t-}UzDYGUbaC%|7z^@gp4Hf4qSD)$nt>HYxw
z=s3)k5t+Q)Uk@S-M0pYcvWs52eM1#L-k%<R0>&nL#G-Gq0&dbZngu@h80DCsMRqt1
z)kCMY#uNN=Xsno61~<!ob!nI)llCRs^42yAkAP(D3@P&l4MHWZ`ME(W9%08|Sv{{#
zpcJ+NtypYqD(EPV?`(|+jc5ra|8I@IW~t{`6Dt6>8T*cMCn=gQ84h!hNDnzagxA;l
zQ9mBkul6n5CV&<EJmjwSm_uk1m<u~_+7hMf$a|aa)sI8>Q!>)py|!KD#!slL6peH=
zD0aU%Q2QE89(i%k)mjV~mVD0HaJo=Bz>sV2*$)6F0d5Et7^=W7OxcwqO%AD%nMDEn
zpB&JX1c5q}ay}?_!+zn1zSug~lBUzN@~Xp7*ZQtPU=MJ+)jD7Wb-{4f%WPMCy11GM
zj0Geg&6k%00x95O1#yuDF_GzKA&pT%pc$Kt4Tc^JbmY|os-Y)<twcBkjvxU*)?<MC
zhKhc&NIq{AsQL`DEC1q0ijId<AAjIpTMe5$glj~49fbgc?HZ^x<{WoOIA4Hy&Vy^Y
zyU;<$M6@UMDShlZ?7K4ya?}a~`~ygJ4A|w);KFo$6e?4Ih*p`gp_Ju`ELi+`Xr{dN
zlC~ZT!5n!XsP%rbN1_(wuo?CxwLvo<9*(EgOf|jsOK;zY_o8mnA<a0XMojp;Ax9^}
zQ<025&tt{nAUR8>cFy<HRmO8JlBB<d{1$LAqDcOk1?g-(i^!(|s_^pOj{BcL&HWYa
z#ate2QWQYGbv0z!E<r2b>r7>BIN%gCxiXZsmtT{ASDv;*Z-)^v%oBfm8C-tre;U<+
z_;60k`fQ_p-e(uXf%Hu*vK(+aoO^b5j@<k7<hZuG9zzxEg})SF;3G}v0@mdL@O(Wv
zYo9E<4Jn^xtRL+37r}tmfkE~%q!Jua+d>SBv7d~=5fb8?tblhn1NsR?<bE2>v`1|I
zVtKk3&uISy#NuM{$fUxXd+yBP(|q@NmL}W|zssdrN{J9FIgq{d?jq&GwTy2bu=ro?
zwB+Kie2_<vcPqw~-M(BS<n}O~iI!9=?>*P3*E%%aF@1Y3$#$l6j`LMlbS@y%XeqK|
zsFu+$7Yh{PL<l1i7`;>n5lu(4Wu)CO!jh6WjOSZ*T*qXS6Rb*ubK!CkNQjfVcb#su
zDQc1~k|<w*h1S?mnuxoaA86^g*4#~Qs;2t%Te3qg+XTSEs~z@F%4c=MR$BI||0(Jc
znSL|r5gGQ}h9zZBg5`cfEHB$Y;%T(hP4hf`lhN{B^ks#v?DFV4qoiAD1G38$_)y&`
zpwao$JMs*vKZkZ%_k}1`H!YQKq>E@clDx?AmpFqgFgm}8&t4r2veED!7aWQxS1m~1
z)&b-tv)ksRop>pT5y!g{dufL#MiO(}8I5~cf@`{D-(%JDxSmd{PoC0H@pV${V=)Sm
zCc#~a9rfzovbTja?&4eox$#I#*wjSnN!tC`)ldI+%=WMHUzyEVKv$a^i=($Te`;K@
zdW%j~<oOC@6|nK91$;MnD1H4M$RI&7g#1nr<Q7M#ZOK=(*R@fRF#var=>n#kQK^DH
z)W9NLz_Qm?c60<lK!1|e6#s&;HeGn>pq@W9Uhv76^R<655x^I3a^v|TLVV^GL=^K0
zfmD39?(xmU4~WobITZzi(}u+gqWglKk_P@Qq%L=1Kj=HuMAXkDhEx?qTc%JCLTEOV
zc<*tQig6IJt!)hpt^`tFyq&wYI!dY+WDfnzM3BbvJ{ggtJ!F?j=2*85d6gxPO4Hhy
zuvyj*ay|3y@VkjZlS^)t`bYOFE;<UkwTTtO*Nrv6TxF9=5-3IJ-?eV}Q@(fQ7~9UK
zR8PyX3J1tJ@^HWK$hJ)g|0&1*1Ds@F<M^KgrD+WtyMKYwPmW%`U1Dh3erFLHP93Vn
zypMVP6h=s~2*IW!wRDP(gzOgV=PlQR_@*N2sE5ENd_8#tH?Q}NyIa@X9%WwtWul%a
znOQA{Oh6EE#-%<pbP!z9QjjESR7t<d7|Omey>do~-A>Gmp57OCm@zw69lN(11BRo>
zVCui)c@Dn!A79s#kNu(ZwPv+NYv|;t`I3X4s<@?7Fr4{tKqL}F$xt1$7nE(+i^D=@
z*|D;`+4SluVAKc!@?h}O;SmY?4eSixf>aoI;L=P5#@Nal^jz=P`Q)u$#y}q)Kzx#W
zMXIyGYM@d97-@?Gl>rh3hY;nJ4Nc|_=-O~B#ry153$MxwU#>%%;Knwxw*<p8Z7K^#
zI|U0Bl6-E)!KrvTPzp@o1wCx?nX${Q0ZFMcd<EE5;jmYey`#`3Ny3~M|L60C(y-E~
zKBAzzGpwDTEYOJV6CcP;+1IPWYtmmL-R-~MtDiftPmg;NOaGwI4;!Sdjz)5O4I3<S
z;Q>?G+cR*Fpgw3w`K}USqE-%dIvzT)j!nwF<_*aLaO)UyimootyGS8sz9u;W>(3^I
zyn^;abmI)g+UAGrS(w4dHOHy^xkA#gxqj&dw`!^`tSV~^q&&_T8h2%9<a5s61^)H&
zzh-Z}3ziO@9#p-$q*Ufs&Ce#%G@0(8{MUCE7Oq^oeA>uQV2G#wXULBlvt@h3m+kWo
zFT{OI49EaA{$1WlilY}Di9BIxA+^ZP&M1LXT6IZGCxc#Gu3*$SVa%TTZi4x9xlCDE
zr*Rb7)qmm`A_rl$u9D39-Z12x#Lyf&>whRAh;lbCn05d&)5al27y^P=B21jWN%T-a
zHoCpb6e$5SVHe(9U%?UqoiO&2rxe?A?9hsggt0ErnZD&O;)q1j#dA(d2Q_N;(Mp7p
zeo-MyFW=IqG1iT?@1j{{ql7g8f0t=LP-ZE$zCiVIzAP-}B@n!zTc0GZKNY?4bO8yY
z{Xq0Q^jChOkd=2Oa`rY1Ql>3zQwi=s1*fYuKmos=3e#3ckYpbMJ~E%T($Z)t$9wY}
zg=qc?oVsekF@i^Sk>0$4+Th`@$bPcYp|ysAQGLjvNpf$gM+!nUqDZ%R<UkH<+quT^
z>e`Nl->z<9n0>nnDOf07E8JG(Hh$lx**m?ch;Q8UGEV@p_Q6-0X-*2<pE||uBbSic
zN)w&aLqei7wYVXygk5SA;WG@kPe+ZFca&CW+OnpAJCe<!NR@aV;A(nYG)(#JFZsh(
ztObY7x*6!4$ES^EMQ@bJN`v;QztY~hCoW3Ihc1Co!MOap`c!U;qjjFM?MDpiH3_mj
zkKQZBs{54EcGPgygP0vNpgrG-`&2Ploea;8@$DXw$5^*O`k%_L98xoVPl+!Zi-V%m
z9NNVm;J4*YH^qTl%7N_KsfkNA|Dwo1J=eRrV{Cht6S!Gtv*LT49tEqrivpfeY!3Ci
zX&CgL@vi<M-xNa16Tl%VB#9Gj$_A@GNz+(@o>kbiu<hidx#(On3kD5zgscgwtG-8Q
zqK0Y8Oy+VXh&MyIHeu+1+f|F?QvD%tBTZ~%c3~QKzn8nZR-&AtAlKHWX4#-wZ0{Q)
zwkHfFBWeAm4U2*MLX!9`aBMdyyOc}Ah-t=2?MjV<#m9qI^n9{7S2yfCO1SjM^6|Eg
zB|R=&!U96m*JK1VD)H358|@92ku9+}c0+u8M%nH5Y+l0rVG@Wa)W;aFu|6cz?)wl-
zT&>l7-gJJ@I6r!O3c{frj1@94xWu&)K7IJ6dAcabFm44hm=?%jCj_g#WO0(4JS9H~
zGS2XAdVMyArrSoYZW2;OM4#TL3SCcl<JPQn%X}x?lPHLqJK;>xc3tDRP{E?;DGp31
zNLrH^hqpZj*7Lt0XEgWU*b-qn2N)4YLgXDRdf|PfdjV@Ze%tH4le?qm>~s$&Os=v_
zVRb(yU^#d=IX+Ar5>f%Nw%K1nN`UX-k1E$=l3MZ0Z%(2<P$rE2w8UJ4LX31=;%6z^
zMps#D#vIQzRBpMxc3*i-zAtH?ONq-(EJryA8yzGz;(Ew4JuVWN!}dRiTu^qZ_jm%$
zW$Uq(=T}<FZRv5Wk$e^EZR)wa9wxOjwP@$(iK{H78V0{6_P>O5Vd%t2Vu}r(H&h;<
zs8K4F2~WbHIgQy(-5o_#pUy9rl2R#KuCt{T?R2<1p(Ep3$8hkQzmHOywJS=L0n&5Q
ze|Sr<c4+eZ6*>~pRy)-{3D*lP*uBys;1w*BU)1gM)rJS5Pd5p73oV^0Yqc`}2=<kN
zF_CGunTR^M=`rlDa?9zM^3Hw`7W9@Z=;-FA$Z(~<qO!B$zZ%Ss!o3E~J@XBWUTy!^
zKxJV1k5)ZBBgg;Vs+an|hDOh&+MakZS`;{7fQ*y9(z*j#ypmWtHg;?%BSCnza0Ri+
z<n>c8`jH^~LV~3R=Bd06ggxMuZ|9oM6%sQ)NH&Y_4Qq^?PK@`J6sbaMY@6^va#|aZ
zi;(j%{XoX*wqC3p#l2dK1UywfOswPLrW+Yv=k=s%B;<j-e<lq<_~XmL;OOl3;|%5;
z9j8s&m*Y?V_P?cfp2Gt^=ux)S$JOKEM~R?ZfSNVHCl=EG!ibYUGle$St76;cYduet
zS^&i%2C_JeBDk1t{<=+Jt{ug+S1=0t$}vJQ`XhQbC(<#W=s0A?`w<ZvPK7-6C5HSt
zo^d7B+gD;~SkXiOR~fu8;(AU+3h6-Dp$6OnAyh9MziuBAsFgkohbKFdaKszeQXz!H
zr4=}P`=&2V*#pc1i+d@JQ*D*2OiOSLga+?|+Z~Mahw#jl?cA+3>%QT6j_}a6N#p11
z+qs0sVgbgl;BaWMUN(?hLXU3~DpOs~)o5IryWux%^d6<^zNFo`nu;nojTjhDL_jha
z$U)+=)#dy8d8VMBeGQq+3ot7m_xSVRi0*G=E19Sp&q(kr1p;4r*m#HnUoR#}AiKn9
z|CuS$OC($0k=<pOtD<g8b}AT+V9Uhr#}nZKrK2f*m*ou><7nv=hV6UNa+8D$ugqb@
zNi41a{G<t_q6|pcG3Aw_N_V~=b%&4cm1eszgTZ5ifWComnpmAw)KGo=n+e8I$w>SF
zCRk#4OyhwVr%*O)xg+zJ2&VP;w5bznh(A-<jWbN;dqA)Q4t~0oO@H@*Aclzt)^mwR
z(hz@mRo()uGsMsiu(D&+m<sI4ozJ2BQsn{s{*Y+054xAI@+`LhtG-?}hA8GBL~yvA
z!Qyw_QY_VJk9>qzuxm;}DZFb!UjmJhqs_PwmqtesS@wV<txJN1<>hij<5aV8vEEYO
z)QWj~KrUps#kX_{xJ3?N%tHXoOs78ca>~mm^W?Q9UHjrDy+gU=rOKO->t;mcCy>)X
z<*B`ePX}}N6LdA`3pc#0bEtwVNU$5qp!!e-$u`(x)sjF`yCsb+Vg22=$I&{<oTdb!
zl|3|Y7WK>#Cyo{-WE2mqQB@#{sh>a`G1{6qxB&`$MJbRM#sI||k@SsG9e*U$>`Ly$
z5e}GRFSF0dS$m19SK%Ee3+QkS#qX)+P7O3@K=&|Q<f8H>J>;{Rpn6s7#sWyp-Eh`$
zf$pwxpGXIE1Kz8I>28&TBad_xX)|uPCoQR@?=5b)g5^S@-&s5y&qBT!=aYCS?)mQu
z(e-k;I$!jMRwsg`rBF|q)dVhd(p$MkuRosaTp9P>SR&qOL^tlrhWXB`A7I4pqvL-*
zEB~d_IavN@onEY=75jg-X|7&=Yw~8=z6xP|cOEQh|BC?{Y!vO>wqbs~g_e%igQg?*
zV!?0sX<QNwWkXg89g*(r@nO5CdBZ<6d4GR4GPLy;lO>#h^iZQgos3W_Y3g61u2GeM
zr}QsTZ)~41|JYtUxE~zln|4g6HI**i^~;)WFF5bK&HZ_KyB|DU>>OR^L-X_a)Q#X3
zD2GtNymNYDM|D5*2SX%h;83H)^o`9tT3H#b?S5`^e|z=^)XJz6h7iP*A~q6+PN#l_
zZ=+=dJ#`ttAlnK%GEaOS)azB9fG{;hf-N%!Qc&PfsLM+uPM_gn4qfLiWd82{W!`k$
z{^_b@`?sc{cjm%ZKFcZ3ryl?5?d1y=P_lgy!Kl5eq1nUuL=%ubiAvDMd~+mBd$RsH
zyeE_XTea1PN76+&4>k7t!V@Z`mJi%~m*Bd(c197(`4DC9#PMTCr=T81{J}Wft7$jC
z_@^k{sV~c_S3UdXMZM{n0cn5O&-aio)UFq+9<a60MsCy+i8($|`W&t)(p`YAJlxn*
zxb}BZ5S(`s;y62#yg;8Y3SaWUxE?%)&sxvSssF)QWfM${)27wraCPSSK9x_fZx3Z=
zeZp?Fv~c%Z`AU{Gt7?^3dkDb!XB?D>JC)%yH6CCQIR#pX+1<|h-$kgp2Hm)98=zNL
zE#<^7*gAS(&6$}lm9G`QwI*Rfh6LCoek)tcI&A`FPtmA5o6pn+Ve>g$m`NS;)|cl@
zS!mmariQ$wpsT)U3QXkt0y(Hj_F#FM121y2<@1uSpxE#&z;-mT2|g;?>x)L7kLJ@4
z<?1g=?m8{X$`#wbs7X$Feyx7k&&%>P(1PaF=*23!El@=H1Xz|1U>{oXkdn>bKn6l{
z=N6osD4l4YQXIBpJ8Y=BG{`D~1S{6I#|CWl>jCo6@Q7QKu<)sc0w}%Ly+05Ds0B1m
zL1WkW7sx;i__WHnN_dbJZo)?#%7780#w(?k;j`ih1JpX~wPNWr!<SJE5i0=&DJqK_
zhAgIQl!0mnIay3`Fb82Yt|ra>kX=$7M*2pHlnYr9QLsLS1m?3|nENs9RIv3jqskMO
zSXy@oOmhd3_1!_`*n@1b<@k>6{FTme+65_rAcmHG;6T<Kuy6oEMW(>^DL9<8u@L#&
z>Ey6}fX3=Lm<ZaE1NGJd3#T^iTLnDgr>g<FUs$P}y*>V9Ng1M-Y5H_yGlkw?$8wVj
zd=y3<XRlLvg)+fD{4{3X(ghYVSU2;4VzHq^05|@S4oIuk&74alAFQ_Sje|>78nYVs
z;2GM^8hyb!B#S2V>QDS3xaMTYfFNXLfrOl`N#~EI++^6UMu1liyUs|;Qjzgm21w@p
z2D@((`-Xg3_O39yl{kaOW~<eFr%`Koceyt0>of2_cMaCw6@Q=QUd~7f5`^=EFcVBt
z#TfwuL2UBMF7(<IB(%?{WEm0l90X>I_@-dopd^4n<^NIcp)Hz%CssOGLFHcmhQR-0
z?3@}iYr7yFn@?=pX2-T|+qP{x>DYGCVaK*@Cmnk-GpFytw|~GsT=zm%UF)dt4y)Gc
z{5O}|45+p)%f)4b{X5}8C7bECWy#pmkt!(D|0D)Zc%-0Dj}HZ^N}sxCgi~F#s=qv~
zv=Ds}-QaqdyHdz|pmmmiBB_m;VuPm*EE!h~cFDWu_tyoI67!aox7In7XEOG(M_=YG
zDzBW!&{%Padk!1Z_APa<^gd4)S5pbV@3>yZ<O6h6TbE~q(VDDtF2)|ZTJ+n<*1wm-
ztbvjJnB2CVUBxLF83wu=aVTT?)`+Gv0-QZ1r3Ew8<4@5J0^X26C<?aB#X)(p-0U68
zv*#P`yVN^leq+S<&2$hNyy(g^>uo~a)?fea*u3I>J?T5H@E|4`0^%eKIr~X@snd-o
zocGtfp}a5oW{;ilZC~<7$<f~iE8A3GX1`=xPNn|6QMg+4>Q;^;{GI%98u#_f;Q?dk
zaqY>wMS=fKnXlBJ9u%URx5ao}sL_~MMQJ8)jG?xCI>&X%Z!vFhS7Dv(Ns8B4u1SGV
z*eU2q&0r48RB4pQbMnB83V}$0z+vv!@)j|)9(pw+F<gJ1De)f?<j$(a{79EFW1~zW
zC3x2UX67;2Zfz-7>Cf!$dODV1>xU_o)gu)F0leKiUg{iD?gPBZXS9<SZG~y$aXxI*
z70h`Vm~?B(Psjl@UE|c}&<V*}diL8co+K>Vwke>}8QL7`ftOhKuEar3o!@4-lfS#g
z^!h#qbEgq~1aB|%x&by}$0Sc@KsS+mdZskxIgq8!RX1Jri&=*m{cq{xX0&Qhf~+@w
z{Xmhd&SZeN@+#7QdbwcIy19pPojp=b33vwXs_b-WHijK>tyriy3H=4z7W)^i8g}nA
z4_<?iRG2zNI}w1K4v2{rtS1HB=w+edP>K;FDIpYm?$OO+Z*7$+m%2Wp;1i}L?+BI>
z3HEN(f5`~1&>0wh*hte;=1_x5h=yKe+abIH8EQ-`hSDsyR3PJMbjC2rL&(^FK+@y(
zI!-9-Y5yj`T9h{oWbCI2nhQe8e|+#;T9C5JScwdlLluTGl5|ZU0U6G63}}F=24jzY
zwqrgmf4u{u2zJsQruFuzBERvFd*Y+Rzd>9^Pe%Uh!Ds#t`x_HG+y8EV(~$m2BSHFM
zry+4EMzm~W5JVTIL&yf;!z{AR3M+x7h^&c_DiKwzj@R8xUegY#qP`}AIU=$T9Zh>5
zXRjU2x&AHUF#p(n-B?;1vL+!;WTFD#>C%y;ig9z$GHJTBL~IWen<o3$ePK!zwkh|w
z#OSkKOjUMg%a{%v@V8xPuDP|}?YF2U8j`0{H6F~VNJ0?PHub(2x}k2lHCe9rINRSd
zHQDJ^N44^}+g}vgS9zZ4M|<u0`B0zuE&zN5XEvN#VAR(mM-Lh`WX(m@0hY??x)v$U
zK?dZKq?UAO%2M!rNaEdno*y%$3yuqZ<>%J(1?Tp0t1A9Wf<cbZw4S8Isr9H2vq<oH
zeIslUJ)IS^faGC3OdYmUzT~0k#_YDG>SFs(*|X(mp~M01dEZm_Y=BOAdt{YY&pLh1
zaQ#JTM+sf5c7|=$lF_6c@jlNYt_rCrk$AgVW}2L2<8*lWMWr<i3uP-`%w{XiGr?9}
zR4az9_GM@A=Y*{-ojX(0Z<v45hceh_#E&qok5}kV?*VW&QCud99M$}cyrcRoL&`JX
zP^kweEZ%Bfz~$ql@q*b-jJ^r_HPIK9^y|6ax~jfw=rr|?m0<DuKWG4k8<-T2>&Ey@
z8FERN7n{1GI^Ry8dy&eWx@o>VcwqGySQ&+W7vjR#H@FqmtiCT^AxJXK#k9G9mvv>-
zXNi%};1~0Ky&_zN21iK>(oiTzNTn*sRu!bC)Kj`SJ}!t>?4e1YybbaCTgFmxM+OP+
z>97=R8XY44G#DLY4kpBJ>%o#L&I1%bCfY<i0vZ_IGsm81lN@7;jDiwOjej?2+MJ+^
zddN|NKsD6CdpK%dO9=~6f|*l?n+RMIUlWG{!0K_E{2^~Ws`H+>BWX>wj)BoUG0Db?
zlHg2)(!f|5Zjg-)xMn~APyr+WVk~2(!~h3!SL_6pvJ!#!5Oiz>11je|WL)00#j4!<
z2fAiBt?IOEE>400qk7{G4oLlW`##v{zY6hQ@xxL=VU|hJbHV898^@I>K&l&{Td;dE
z-hs0S1%t^iRz?L}o$|&H#Oz<eGw>ro&Vyfet<4RZUL8%9_J4*COu#}t5r$;02A?fR
zr4Qsqwb7bVFJ9J7$%d?S?u5k%-$rQ<S2=DZXWV)BZYi$E5_Jgsz<7N&9JOj-;c};O
zy9YsB8bZOXWewo1K{XJgHjYf2wAk7rK2W*&{(-sZNaSpW+bME0wWa2o++rL?8ADY#
z%_B#Zgu~s$$>Rot2!Ryhpa2Y~a{<M3&fLXoTNDx}?83r=?~{}JvUBJcjb&UQpDXCD
zZ8qyIS9$fuiVs5_?`OPuBUNU43Hf2&j&*Lu0p^aJ9iuu-gUJj*F~LM^TWleL$D>#h
zFMWBU&#5qtC?9^azJ_Q;cgbCKA?Sppt~=<_uM$Pw_cbdWC}uhW9x5Tqy>M#u4ZkI8
z#DfSUoBL84jp*rd`G*JgK8VzCvlz*il>_|97A<+)zA<<nNXmXAaokLAog)B0Ku%Ib
zAX<7nB?4-czb=h`_k+O?w29aTA!FAA0sU>^6{Z{IQ2mu>!79h<P<R(RF?)!>Z`Qn)
ziz<~QM8S>Zp!@VI7XTu#hm~OkHGj&v%h>CUKT5<r^-l+nI9x_E?aO8t3Bru779R@t
z3E0sio`*@CKi>KNE|hXNBY*F?9w6+wFxc@e8G0rPp6bak00&{*_{|wC?8%BBfBD&?
z%J4iyvb&hJ`#8(Zse9t&(m9Q`L|MI#Jp{fgbuc(k^U}m-L(Ojoi|OKi4gbfChsUPU
z>KlG5VlV_a&&^3RQTC>FEvu|zshGsJCF}rFmSF`<+IQ{k9WI{KFCoCP7`A6Gp&>96
zr>^$$f%4<QVt9c>Pn{OB#|^ITV(+}~H?|%VP+UhOGPKzgg}e;QJ!VM=00~6ac!n`?
zUNzKwerd@}`RR*(VyD!0y|Q%3U8yn~FuE7+hku(^M!R(GIV}ougK&Sj&8HJWG89|~
zKuz8`k=n1q06c5kF^q{BFF>=HQw+zBZ2tAJbo2SLNJiv}Nqi&%Vi>~I^o+Q?u4f17
z4<1A<4rRs<oAb#BKFrSKZ{d^7bm((A@|MEG1b0Fy7cN9w#FnlBEkVJnZnx;J54hb=
z;qR?@i{k)rHr>mJDP!6x$x<MWxfX!;4f%(>PLcoTGxi^bC1ys(|9KrYrMYE)$d2GY
zqrWFU!&Y=U(HJXC(E4YqFx*d<o)fp84#s>~GG=<!xs_t=^7r?9BJov2uBJJASNqdX
zvJ^m6@i4wv45BR%u7~UW{QPo1Hpwy}YS9uyXA5i?JkeEXow8jE(q8x|1U=gH(B$PY
ze&_pP>%yNQjXP=YsEXk0>EL<sdqs!}t*7(jX-R$O&)3O*dR4D}cZ1pJqL`2v!qWq9
zON_&OrU?`McCi(s_nNH!xdv%Pk*L1<#umB1IFFgE@@Cpbx6Nypj<UJB`bH&P<-Z72
zrQ(V(c+-LprV?amnf7&%XZs2E-l>aFsR1%4W<yow3ro*d&295+-yU+KFFSFUgd%Dj
z(shUIU-PG~a&(ewHI_74twF)GDx>BNh8;V4C$7$NP54nyai>?hap6FQ-ifxABE`x8
zPt_&b!CLASQBP@X%GNboMYBf|-+epPHJZw?4&nLH+h2H{02dPCSXWf=h1MQ;K3134
zDOmMCJ>2gzn*N+5#Y%y(0L}tvwGEbH9E|=C(pkB7&hq=+<4XdD)I?iZ7}Ed*sT5-7
zW($#y%T1vYL6Fu1xUSGx668TJ`&gRR;on%0_UUp#VoL*U`i){_IU#-k(f3%&^iLs{
zJ@l_su7VS&;E}{hLy-;<#Q9}}zu-I#kp5lmgpn}KAeRa01%~?=Ss+(8*$m~0LlQa4
z5A3SzNkbacbTN%H)^fcn&O-itTT~UHYVs-Spb(Q*@}~`7e^KE5qPFCi8k3zW42?5!
z2S+=n@v})d%R?>=8371ea>4m$5e`RaHQHG!VN38WTP1X|Np1n6B#5e?IFHPwm)XCp
z#*>lKe9fCG@GaN|)?K!(^zHF0ymig1c7TdYN6TW;QxfZJ*1Zx#-c_}irt(2MV-yPp
zDQ3Y742eI*jngHGlCn<0F3&%XUvmv2=ar^;9!^Zn^{$tkVuW(lVTt&#j>)V4?8xq&
zUWp5=F9_UDj9AS5D*9Urs=4#w-jCrwjko2>iTmOAh88rv6f<p4Sb|;VYy8Vf!Y-h%
z{k<vX@EyY*-jc?LrK&Jz3*#iqZ|52QvYPv-+vU3iT!j&uCLlRXHx}7L4IBZm-2kJ2
z?GK)SF)|=}rcu%B5A*V<By)fG7r$VI@Bv$PmM4O-HkxrZo;&3EP|*qBa9wc@%D`Vy
znPcaWGk9KarR<Mi>&)L@nMt?h@&n;G0zmlS(4*6=tV)4SfTZQJOZd>ii*ETuD)*b9
z7e*YjUSw1@ab%2I9<eP7hpTgm_CIUsRm&KlWg<V}2}tRn*AO#D&GD_PZ%7rnFxlwD
z77+8pds$&`0x^T(bf1{Ahb3#iYBr4*vT-nO2#cNTmmN|89j43I7ctLUDs7tZ81-ax
z6x)vVV6%Vf=Mxv}%iCrShSc^wsB+L;IkcvarUelyD}5@IOu}KV+Zm}_i6zMa{r{B7
z=9$sQ*NTW8=`06y9|J6>V=%FUyR+&8P4!|x3-bsQ0^wwYT)k8Z!lVa%ny{3ai9IIn
zdVm5+u{P6N`ci0gn5nZvY;_BeS2N^cin~U1_4L)VjYL5?OQ@@C5pvOFvDVieEKlYI
zXvV!~(>XXXEuN})&P}JfbuINtV?u1BoUW<GPh@H$h;HMT^ZCuk5~_!oFy_}Pt&ZJ{
zS4|c8#@7c_EA1dSueNacVfO}~J`|EiLS%`uBi|(p5<K%T?Y!~|o5K2?!yz!+BxS_E
zD#Qg`SfOa4c4b3<T`0eWMPN4mighgW2NTpX2Xt>Yb#J+s+nM6>ZK7F#ajx;l?RH;u
zb1V_LNV`ap^);+T106!=E-{iyAV6E80#HdlxqNS1d$bdxCHztp<@@yeq3}^U<~TgZ
zf#x-%!q#!0wcdmaAV(n4sz{Nc-|ZkS!E-d$Zs8-VfeV-yiM!bBiv-6odAPVO%cEcL
zta86;OOBt5h#m&_e_M3&ll=`?qTXp?EmOv!{PfY1uUL&7ZLH3vo(acJ8-zc4@%qL^
z!{CI|XwFvCNemga`aX@4=0YOzYu&Yqw%z{FDzNq!iQ~`S=_U1rN0sn?6P9J7CR#sN
z&*R}7mJE%{LFYapc>h;aiu>mB>a0TRim*23bGI8TJg<&joY?Ddc+$N^K2Ul1oz#Y?
zi@V2a%4rJM2n3U%Qu}WWDia%d3_3ipq%t$4bY|9u+DO7ejF<O^rTs6M%XCXIJjCt8
z4gNQ=^*^~u3(T*s2iRNov@?(ReXM>$Zwz|&1JM(mzx4(e)cG@-aP84-1ZoKh6y>AE
zi5a&@^4}PWQpJkJ6worb!o_=UfnSB0?G;G|0YxAsVKPPQO5gbZd}$G~5F<D#)davG
zLpu&8&&PMr{yvooY{L&@LqBGJ-++#O$pzNeEWd_HU@0z(f}R~d<_)65sXX(Q4{C^o
z2>%7vVqk!qO+Lt>%a90`7bnr&Spcb-g$tFu{k$cQ2))vW=;0k#g1&KJJ=#(vS17yK
zzUbu~CIc@xHxF<k^g{oU1H3@S&!#OXO>%q$TZ+>4c%fS_V)~)c@wegsBP6|P?95e0
z2_pj)$Fh5{)1y5ioE$zNd#H~#!sUHRxoQa}MV((KsvCK|a)raF8kXix-86mXdSv$d
zMUNg8({-eN+*sLGkTZuv(3YV5fk-H^Oz>Z=0?U7y#u(ZDXIl7=6y--5jQn-|!{8E^
zr2}K|$-Qtrrbp2)bGCQg?3zAg#?A;bSwyW!O2K_Md)w&h17Ve%kh&zb?TH`+qebtE
zHqQsQtVgi8{Ty@0pkSCLMm{8!_W<ZrgZxa!DM;8u{X@(iC{9_E+j<{YLs&(x2AC66
z$civRm*wgHpHJ!0>L*QH@s^_i=Ijw@c>S-B0|x%?H;SjPhj@ZsZ`ZGND4ySTuVd@Z
z46f74Nk`D*dgray4@b2NuHZr7MSGfC1qzRHcC97cLmZLuh3vjJx278w{Ot*7?DVPm
ze{9WYVO2LOOHk(aomJCKe`aKr^h2un1A{l4<$h!TT>tM$%{qFyFW-SaZdQXpq!4Cw
z5k)b~1HlF^rcdHXLmza~pyCwpSQs@5fB)K~X_OR3;7{K0e2v{=#hDIR^x=Pmpz&k_
z;;=^|#UQ>%vivSiw`O??*xSBWHm_fy2J1P2D(Jr#|4?%~>}y!O438BVpnQyBj^jM$
zZ1nqZzh6ySwXFRTrt4j+>QuRUyfPw-^@}dt_D9F_*XUlph<%6}hL!;gM=y?HKjvf`
z4;5MLmPB&tcMEicB#c?&^0+f!O=-DCi;ewr$6nxhF)X7TdANwxdH%FVW}2Mzwv;&}
zl{SvY)aO#csu`t}JlI5hjdxdEI}=sqz|o}do|@9i-Z~o<FgvT`Kkma9Bg0I5z0Uwc
z4!6DCH3EephGSw6AxaDPa&@qmg9yGC+G9in-`WcqNdrG63mEy)F$|@qHzwPyU5~pH
z&k7EL;PJXg*w3C@-GiN4^Ehah#aWJ{K-E|@Uz)k=DCfaI{dHj#SM$4~DMP?TU9-?c
zgGu!~%ds+nj8WiY<dBw>QBDatxy%@s1T$BSG4D9#EO;@XIzUTnx|2z^C5-jecB`x~
zDSNK$&JAP{GQK<f6E}@n@MHq(expb-0WH07hkLP1+IR+!ue~-J3zBM90VEs8#V9Aq
z&9*OyfuOQW<y{_hx7QHrgp|NLaxKINm>Zjn?H;N@cU^L8JlBee5!9H)+$3{&QTmjl
zs&i?<p9>H?ex6yX4J347_=BfA?k_T#sutVK`n!ETl(xZLnL(8+R6c@WSOR}!v!5b-
zfl#LvNR4!qm5CYV_R(oneOH0KQ$db`muI#Qbrqk_*VDxwOBi84<hQAwyM(94>UHZz
z=b4rd4kPJuokvui9ohTv9}wDR^m2PKV_F>Jy}Bk`h=IT!+EncVy*-rbE$Wr-tx8Bg
z)CNM{ah(Jrtov<Bsg*q$hoUYTo0hj}90?w&@KvKP(EQFj_9%!Pgkg&lZ&cce*OJQ$
zeZ?E<2neX?YuG*Cpi<J4vy^9O&H&Aa?_*z~Vp6HKh-W%JU}co}`hw5<*P@l$3VZbY
zL}~W-_}wHqQE$v0UoFr)KEu|n(IS&qmU@~|aM~wB#98pnKw*Fk;Q%e=2m3_kPza@K
z_b{uBo`soaI}P{c-HU}jQgZ<lkg~@{OT^V|w~dKoO(zx*iJJulmLnK8f_OnFr_tZe
z2AXWSg1Mqm(0xF1O14TCpdSx@eZr}X+`=xfVZ@`ZCQ$WDB5`7nZ8m}Glt@ELndte&
zn^ZTuh$f&Z_QCq;W^9ou!TBI}kJMB1=8qAFRn#>uP_$C`HBg<YzDb+c)l&gaTe_Fh
z;EwmIwrxUCH?+AWsvnjHJQ~JNRzTDeZKKaivcaPv?y;g=`z{p8u86l%ZzgbO-+ht8
zl+@dTa=-9t{4l|;LbysMKi0^ou<5oRJ{(R8Z4SjPzTx7IbJrvjT_3JU@g%`#3_K7~
zQC?LCk!j?gG$qrXDCWU{O*(i{sX?+QAjpf_fcbNf@MZ1Lf?53XYEBeVS6)SR(GG1b
z7gSaRnwHiDIpl~#rwYirtLsM47hD9!4Mr5(iXOT^@*nLLpK=Amqh6bDeC?BKhdN|x
zAzZ?kU~ga$E3`Jts(}nTSFm82T~!^o&70wH<M3*&y9KPrjcs6E#NO;}fi*oM-0iWY
zhhvFdZbD;CuFFX>^gyOp?3)lqwNlGQ=aPr=nVUFdIwUPP+*Iz|Og-k>VuqNUF0m7|
zY?le>KP_L)P8}0P0LrdjzaI|A8(@Bo^tm5_PM`g|@>+~4ea~cJbi=W*yKNx&a|kPe
zS3T|Z<BWMG5DW!4TmDUYuU2n$2;ndliuu<KgO^&@c$l(xcR`jpmIi~PDca3M<tgZR
zrINao*T1^uSSwpz{27mW|FlF}sZw*UsG9|@X79aOG4tBbKg<xg4_)WrT0;skT>*OM
zc|Cu(X@oT^Qo#|P_oK9GWZI}Sf8$_9bVOir$CLN5>pGxoscL+Y8EUSqv|@#Sueao0
zEUv(4|EfPk*C;sAOWd>_PxJd(Pv1opkscfU<+IQEE2l^B#}Y)56#L@+Ix)J`4@MqZ
z{{D`X6gjeMH(%@dR2O=cCLYQ>u4WS-5gzJ+7NZF3%VNEyh&E;LK1>#N@3`ZF5GtiE
ze}*?77ZVu_6Q(%5G80MpgLyB*=D4uJ*?&FA-zCiWJ(e(7SwvS#sA%y*%<iKtZg_|N
zrEq<dJ`)GA<%%2w*%n&jEz}1+unC&hZJ?d-o&4;(o%6UT+YYLPlqkmh`aSjXaL(Ip
zmUF_J@#A37dXy`G0OhrP75qe&YbKvx<IsgBCo$u(QgE~47+U-FnCdcFuH_3>X)T`*
z+%TJuDCY~7kN&jfKl_URa2zqQ{_NNP-v`4WW@+PM>O{yOW@G4LDq?DEZ(<6=#|Pu=
z;$&)Q3*)gqrLOsZH|hVECQY)Wb@4~C@Kku6TQCZ@hNJ*IsjuK~%ipwcA*Hmf!q2PR
zBU-W~Hy?HZy>>`<bMv0e+}AWOns_e;a{ft*kr}Mm?g)`AYkR+I=qsw9RPXU@if!+F
zOX4w?Qr1U*pZB+Pb&6hja~;L+r3ns-8My|3uYdawhg)4g`2sTh)Bb#Z^rHF&Dk0SH
zCoTUt;V%oJU8xlS;5t<Hn`$&S&b@L3T^A;NudId+w2v7SV87wA7zuN%>@(#ryJumR
z!|_qFL~=Y+@1`kw`P3%<#LD#X#_tw^*<nxSxYZgYVt_blX&m;!1A7EZ77hZ<X{WuC
zv!t72n60HO0ZF{Pe@%^?$ta)e`-S&j|6nt7nSMm8h}?UXzkpknstyDO_@DXSU%0|$
zA%l5T!C=<^s^(-5t2_$CGRaUA!|@*PU);QF^Cc4{(ss{%CGOQsIksw!$csIZkHIed
z)dwh0P%@w-Z$*{thCt34*B{1EB0XkkFlS@)>~yUTn1xM2R$4^!BJ}27BM)wK&>tG&
z3B!pSHe-C(v5b(D5MRocz)fLI62>*g7ePg+y(97PiFt+~DXSW2@JP*fR?H8zN=|@$
zZNl8gCa!e6Qe~Q@cw_zUmUfRig~Eo{VjR8V%Hh<{K9iw5_FvChKaVZ48W-9*bkthp
zRLnE(qIX}ll(OMh7;l1muh`gf5dy}c`Ee$P`}#PL(<gnR^Dayc5}rb|bY!&XU2PQh
z<I1vfR<#VOFj4k{djIeS!C|A?B^hMkL#4rfe{ZlFspaW5&_;91?smJWpks;PtR(JS
zO6624<1}+x4!JhQUE0S=ZTLrO>!-lWU6Qq};~O9f$P$t5u}5mVa+F3iJ=8U87{1yd
zH|2ULrmwX+7DaPzbk@4|85{v`2)uz+(R)1adV2F**~0xc#nfl`;UdmOQemSiWumit
zJ6NN-%=7A%i0bHG#rJ#jL{(vJqFW6~Tj*M5xpxu;wD6E_+bAq~bcA7Hwe+(z1IN?X
zU(J(Nu7(Z4WOv#LkBm!LLq;UGfFQ;a4*aSKtv9qF5>qi3VNl1F3Ey8>^nTEW5;LVB
z{k>i$CBbYy1TS$H5#i8Fjn}oiE~2RBd6N1SYdJVGaOT3z&{Vy-bV$W@;D4Y1M_&Yf
zQnjlv2m@6S6dpv`4kv$tipYRFD8f*?6pxJ7BOgjY&0Tq$U(qb*FZJsgI`=C$bHR;?
z;Koh0Q%?Q$m&EuW6L#h;t4j~mw`a%>q{}Hn@6O&rs^t$nyj28hKIVuEFW(YJ2J7iY
zcvEJ?j_<tVRS<Zv7JSl`dYNU2e79~1)z4p9Yft_t74Wa@^iC=w=5c9QC7YVAibt{-
zKpN4*8ySM8xM-Znz6VN#&{4!PqjSXEg>&KcCq|naT=XR&cevFBJXBIN<9HAh>HMOl
zr4AF8RFN6i?ovZYi?v*-Q8(+*V$u_Pu~sjA!-iiydoqz8N~>*<3G@Q@l74@kRJf<?
z*D%_qZy<HbP)3KY9T0djT>qJ4H?q%(t~o>hU4^LGKb*{+gvU?TLkv*2M=mc&mF%d!
zE~5Qgo7q3+!J^fS>3DAL4n4~#c+X8W#^!eKHN{hQH*-pu2pyX8K{0Q2mvhNP#kWSg
z7D5eE7*{v~mf{u7y>Se<Wa_6@S0@{A?fyC?Oi4~#2j=VGLWDl=lSqO5;prpXuJdV*
zfw`d@nJJ+cbhVrPlKTjY9t9v1+*a73{|dNy*F7)`&*OSm9~+|)pv^YwBYl#GUiuo7
zN8?N(>p~#KJPqUOcdkB+qwP?>YOB9oU?lz1*{$24vzgGJCfzh>r(R$vrGd!KvquDs
z7{ni3Pd;)vzqWq?iIeTL^l^QI>PcM;{%3vX`VYk=3(Nn!Vs6p+{}zVp9}7bhaQ0r6
z9{S}VcJW0R`xP@HW&(Ja5lbAl=08d@TW*(LUvd1SjPb5EzX(bEv96Omo}{u_3#_@*
zvi_<wgPRsqMX|<=k+}z6fMc<VOu5RO3vw<Tm#lc%g8q{pOO`F2F&3E+dBNWrZjbwe
z@Uy);ivcve#*7$p>e9jB=NrE6Umu_);=Mo9qQ^(JDto~;FPY{*YSdeerWm3u-c9F)
z1NVz{UmtC(2I)_A$F7y2O|t00*64^ptCsKYNExMN=H@2&=%g8qjg|?>F~9vO?l7by
zd@wm<l;D2GbD@EffHG)6_q1%!d=0y-0AWa|srH777$)a#(Lat_Ux*+J;(pUKF_hs*
zAX2;Uk3xD^XHOQMnG;)Ev-|aoKHh4sJ9bK>2!1MIoWR3Xy5rBSvLZ>Dv)il7Pbr3b
z-#Y1`dPxoLiWM}8xSAauyM-YoIp$%Wk>@9cI`Ha=Zn|?A@TzXMp%Xm>K3VzvV@Y#{
zf(tF}v>s?d*#@1@?+47>bmL3(Dxb(-ws4$*YA0NOdq<WRmfs&DW~qq|=Wz2FrII*{
zpDw0r7Zm1%3G>py_pq)i^p;?C`OY(ULnzkh%aU0N1yl_fILLcDl5NGyX4Jj+%U{AQ
zt5&Aq46gZc=21}dpB<i{pIAMEzKf&Oc(TLPV<KZ~3(rd?7P=q$HA5R0M&4q4{lOft
z=8u^{YfU_gMbPnyXHF%Lu)cn8>}9p)t+CkRg@@LUp5xjdUlJjnIGkVbVK+m{n?C1s
z`{DA!Y<|Y~YSoRmW~jDk9vxMCb5*xsm;$cunf?0h!h^S`MAdsE2s?A6h=-2FKBiYV
z_l%|fURmOih!PVuoc~IWCf1)`7MsESxz>7K*<Cu`c54n?>5sS#KnHKay1bSEHys^~
z1k9%2QoY;Q09`&MP{M#GU-FYqg;G^c(9iH4nFabz$mkv&ZLPfALH21wGw&d;-!;<s
zxfjp#`MhJcPY?YJ%Pmx$khf!HQqV~zXP@><{uS8XqFf{??y1(e&{+ak@F+C(Ph#1~
z(&5-jvM@|G3=bB>%xoeRCEIcTtOD`8!BJK1>>N1j)Gs>cRX29q`9|D|ASwfqrNh#Y
zcy!K}Y5pD?>#FEAQev0aOxJeaLKL7WQpJ0e2KdGG@uOXVU#3J(z=pS0Emk8T3zwCn
zd)4fXV1WRuk<lBiIh+RssT3hVqJ{xA-$m6sn%{GCh;F~(d8lp$VvFNerh{1-Lb5nm
z4=5i8-lrh+T987D)8xLRA#|Y~aq``UV~A_W9Duax*Uf!G=uFyQt!k{Y-LLo89((x^
ztcCz2$%1lMR6#wWVXU}-UoQqEGY<-?PU^kiPj@^^yY0}G-M_&hZjm~55y{HmnNevs
z!{>@o6Kz<GLh(ot8lee|?MSsd8Y=cj;(xbImZ!|d5CvT8`J#Fe?>fy4#3q<CEAq`y
z61;8nyCEYyhl6wfs)v_~F3k=_QmY9B(<vyfz7YG_505gX*HZI$bny62R;)V7QYVp0
zVmxeoWH@BLJ|t6tG~pqx%XMCM=1z1H??A)uq`h@NPVAi`!&2gWUAiyRYh$c?h_coG
zjiK&2&GS>xw^QRoW_qDbSF_scY|Cq>@N@7k&CUs}TctgtYOJR8FL5O6T67n3A4c5{
zF8PWHKDTl;<8R(Nm%0!=U|U?xJF`=Et%;2;)s4Ic{}T8$ZnBIA{J{^Ec2yg_e6^Bm
z4DW@D6SDfir|9F&iT42uFTzFnU&Wa9Kbmc3R;K@1r#kf`TRCh;?zyR(rDJBE+7m`C
z0ppT!GfRR`&?UUsutRW(c7rWS^U}8R{=88(NhOt1i~q$17HAkTWAx{u>Ud0eAi39f
zM17}s+wb|ynF?T-fyymycY~5jp{dCH^ZK?*`AHgH9PhM!XVjARUV_b-yfyUnB8z(6
zOxNOBZCvuPB3@tHTzpQ+S>^e?T^WUUT6_Dyf1b#Yk<mD3zH@NWOaRL7L_(fjrvHI#
z^9*?NBesgdd>*-+PY35mG~Z;6bov(Lp7=S8;8Se<$}1-i`v}iylrG);tCI0G&i4JX
zr;Ml{D3X9Ff6?mHd!XO}4CrsC6iZ{9h?QuhH6Eul*+_m{%e?`)&y?E=(BG)f&fV9f
z?m+#@Wn4PD>fZxqz^so6r}4%<7hl<oc)&^*z)bGc=Sr`+aI}o5)f2Nc&{|l~RS;kU
z;Bj-qCp|yUP|6nj{MhWtr0#&?lb^zC9nCkcbqOwSY$3?A>Y#^Y^mbeFwPq5R&+u*2
zf#Y4@!B=Sgf;`7jA83^93!N53vTUM_o5fmLaTJPg<O%p6J$5Vb5t`LiTllTABQGK>
z={1hGHd*j`t7JMM|AESE{2l@K=4NZDasZ@2?ey??DXPYG|D3_#F?rXWZEvva>T&_R
zpTA3J(mw+!=Y<k(bK9akJbg{NR9JRkK$MU+3|8Leyco{rOyi%hAZ<Ac&aQFk+jrzK
zVSNb>`$|f;SA>P##fG3+QRHVH2Mt&8Y_zZ5I*EnS@?h{G^y^;SIlH^ytmBo)%|nYF
zYUA&Mebkz&$a|xl1LwgPq?GB)<6VtsUX&Pt`S{Af-0`slMF)UDyc*=SIv)mwWRWhe
z%E48XpE;p<k#U4=#)5YI&UC3KGE+b_#pzA*mW6M!yl0A&1hoy9*nPuEd|)6D&9#@1
zP&i?)(d<zKQJTOeOaw3uWgaQoC;)-l?G8aUMl(;R5Q`ABuQsltP$e2;RJGnKPe`)y
zw>NK;p0Bu{;Q~@|K2$0kyspD%CGIXk_+H<dP=9Da4!V?})cT5|M{c_AlYgm87#|;c
zcIC+wM+HersK6?DZ(K6O;<b=VvdxEsdb|b6=AU`udMgPq74LR6`Hub}i_ArH?g<iQ
zhKMO-ZRW)<W7dEqauksKld&0q0LGv3JTdm+tQMDfL?#j6?T1iIM8=H+Kq?+vIECO4
zE4!gIR2tTEfV+53FYV9eTV$3Q!EwVSA^3>G>M+<Fp=rhDN~<)0k-^xBq>G(FbaOC1
z9`Rv!jIoq06F1k6M^MX@NaO0xCfVCFpyF3@xtA2<BZrUE)7g75c$W<nR450!l>m|;
zEk<QV&Z!WJaF7Ui&%B!?;5c(gyT1g-Y9SoVoGdK^f{37#%Cclaeme=92VZ=N#8RQe
z$<-ArhbSmu1uwvn>TBZ=P2_tupdvHzzs>RPLL3fwwSmhJp=usLbGcGV>U2DUW|&JX
zsreL=So05}RmL8jafu&*5liiAOE=5dXN|YXJ&`#|@io5<5)2j0LqiE<<kCXCV@_^O
zKfoI(2?9ab_i`vx(B8io1;CAFe&{W@$bYt<dvOD~+6knx9++_~by4mFk+9?`HguEp
z^17*B5#`q<0!#b&pLY5M^}?R<YT)&X$EVv$+r>fL9$u!O)!OD8ui}CWtI%_;eteh?
z3c3)X2+&PZ2K|wr7Ejn$P%1dwrBrwTRdS1OD)ZvSrmo-T>H(zT@_v4}I6%1zTAd0a
zkw8Wz_W5lSbV@7ly~IslBI30#NIw*3?(uP@(xQYo$)CJIS-@6(AL?m=JAfeg;M0Vb
zN&#yZlRGF<l9zjka;MB4M~v>&*sXnjIkmo<>)7QMEruF0Gv|u>0sE%!Dxa?xGmJ7B
z9?brt(?E_s78vP=@lN;i`D3TP()Xxip&3$)Kg4ex&kC~C^8U8ij)h$7Oh1!Bbu`Jz
z9)CO1EH{lf#`>jhK@N=?*B^B49(5P>EDOZ8O<E<!%E{X2;&cYu;>Vz@ofkoDp0en7
z>j`(B(_7k+F&je=&RxKjzs}qnp)edz>nB#-b1w-Z^ZVXL?;;5C{b{0{)yAq1{++@2
ze8bk1!H}7J7~P{NnJhd+2mH$kj5e5SsL0wq1Z565{;KB&A+9u~OXkh})UyBT9aIm^
zH1iwF{kR8Bd276o7}ACPse=hkq2o}8gfT`oxR*)*d{HYbD<A01besbIO;XV2qZ~-<
zyMaWKT?&(Rmz!Eufgm(R0gvj)zmGmWKrK!Z8J*fHk4C4bsAqk@nc&?;<~_@c{J^`k
zqlrAFv`%m>gJTv3bQq&s5kpNKqwlkBl>GNf@;WVa%qI8@5J7uC{<;F9Z1qTQd-WCp
zk{Xv#^)8_D&=D4n?l`HBErX&l=?Li$Y}AP?T#6u1%G9dpJzb~6s7!dCIg*KnVoHb|
zkZMrBGw)y<8Yc`55G2&-1MJtfuPQMp`h*(bSMR^5Ih&dgb$lwK2(>UwJ<sBR0_34%
z8bIU7Fiogo$8*oiu)TVWGBy^Lc=rITfMv3|wVqq1y&03ajYw#Y7pOj}+I+YbELO9x
zC_LbjWO=-H1N+y`JCnk|@ljP?6``zUqW!vO39bVdS<jfWITqZ4SO_T0rW83(fVAcl
z`Hre5u=x}RrVbYPs9w*~u94v_Uh&BTgLdH+YCt!j&;l|y7_dgnjlp~#pwuHc440h>
zz;*qQyr>EUJuB6~{|+N<Pzp2wM=jNLx%M~Tz*SpHTipl)=OuLQ;ruphs*JJER!xz~
z_-HYKb?{`_njW~}JSdTK<0Ct5Gc~m%1g1TDrfgyiVFso#JPQ+~+`7e+=9m^b3S&q;
zQIPqB5jd6ZqWA!Pf4YzixR{0d8}mV+%J6NitXSR_9-hRdQN0P1P-RbvZ1i?z@6p}D
z%isMZGRFB3W$)&mDk$K1FXo@F{`u);#)h9dbodIJVM1C=em5^3{eTX%0`*x71ynA4
zUOX#9*WQ~2@oPJM?sYs@N1MyS$@f0=h0SNf*BN??c=|T_7dzZKGSh)m_>S<WgT)8K
z>wzUt^|fp-vH#RH*IzD56SIF_Eg$w#M^Q0G+#L4eq?NlwkTaXH)J+!%5OCV)i_+p_
z#BXROZ<;1so^dtIWole+{U$qj@)(O>mD$MK$pzTLH;_x2=b$2I`-^28H(#K@Xj<$4
z*}VOS$%vVqmGOVwlx}JLco#VVKbqrPi4_s3?f9DYV2TEFX0{m-=kdwFB#UgcNEZ=m
ztdkXcP<SO$*|A2J7M>OiSfD_$qPs_w(lJ?~<eTZ9-26Y?NyLtSEo+lsU5Q-^i^)|g
zO?srZ%W~%K`A_(}yNwnfOSwcjl?6y-ydFP?>h34nPFQL!#^Jc~sbm*c>}zZ3cl5q^
z{hjX)F84=n=zD#=?~c*rIzO(Sqn$|>UbFT6{7yJHixNwoo{v(Ep^g7i4SmKnwI9Xc
z>{k!Q8ZyX1Oe8<Eh*8<xs<(D%pL}!YLyIjO+4fMStzmO~E2MZT(^~Vc?!Lq*Ah=iz
z0Dec9(JH1O7s)X1^!Z`1%@z?)NqtN?VqB`Me5IFRxm~<>j&Xd-6Z`r)YWCjJX1Zvm
zx;Birz3g>r)m6=idIoR(>g{>{&1r6F!UeZ!I}<3>lJSZH=dm2w94Eo4yz8E8|5Xsx
zQ-l86T0yg$#goZg#s<gIPEn!3-8<#jT|KVaFcw%>Ik`NP?h|^@$Hopa<Em^tLooi$
zo=;;_9c4X(@gaHNXQ;;xj%ySoX$*CMkU7bJ-3Zb85RO9F<8*{WTQklZS8e}qh@B>F
z^-+g)mM$BJH4^fc27w8QFRoSd=+Q}=b{`!+A;Umf+9*?uJ+MIb-UiFk0TL&)!Ro-9
zdkD`^NR1s?8xCj#RK0kzm`EvhbjRI-RRjwiopPXD&MZd5lw>#FIE@O@c`Bh1<CMvc
zmTWR}p<OW#X8|i*yxlRUoav}k$dWCB8>J6UNareTiiy3WRK(&5?6<2;)ParDmB3(l
z$e$#oa1C6Cjln_!d%@7%Ga~W+Im2*)>QJrNV@OLu<K!jBQDN>KKbT^uEZVxZn}1gW
z+1UKxnt{X5q8;ld8*k`cIZk2Yykxx4^Y!e&F(^`GQ6#gOsCb~ru=hv~=IQrM@#fxr
z0k_REq@dBXrj469f4-B+X<lB0K~QsK`NAu3R{1wUWS7Ufns#QlVmO<V7^hGq(Htiq
zxndl4V<3lVF%KL__jgING>~WRWpg0gwu_{pA<?1{sIviaK{!rVKd_-0xW#yuIde%A
z;P6Y?cTnD0;8N79?qj1YDmr=NaS5*30U|s}(iTbkWy|Kj#gqLSI9n7?H&n5OJ!_3}
zWf@sa(?rb<v-J)aZ=}5E;pFBa;^4c-1#Z*w82XXoCD5R_Fo;f86nB*PyNVBY9*0lz
z0i@oS|CUmOF?BP<i{-o)dS3m5jH~Rono(%`rcl+n`!41d0z-{_U;3l4Hrvcg-FGE(
z_e6L;UAYf1kqvImCXvYwC07~C3<u$gIp%`{AM#x8DYlnaZYU764gwq2FAg-GWL(xE
z*7>gNR1EU@&Jg;~1LoGskQ2tBs3MdApdc~rCDIq<)<%r8u=1bjApfF_G+8gaFYBB*
zAsFsoI--dZ21rkAJxJgv8!(-Ai=r)m9f^J6p^wsZUuYL3yXCmO_O)FX>>1}qYM)Bx
zSPPO8M1Bn2t<Fb}!m0BDKnR9>yx;I4zOBo`KjJGWa<`$0C!6FLz97r;9UX%(!OTT8
zIe1l;3!ZdtvhYfs6|)I1Vxv-*5K%QK68MCQaV!hAb#KNKp!X{iV8RLU{G%KD{yNtc
zO8kM}e}2QM0{gnqZJlEGOt$xp4G=cdOS!rvCWq1OH#7iyt2pCCK}NFT+CHw&?a)+v
zZP7of8E+izigCRh+TC4zIONA#gS%$RZpUOSY-K5AD3zq77ymsP5wahjxGNU>@hPlN
zioZa1`-F!!j}HC@Qj%ivTcMyg{bFCZ9U*&p=}ozO_hp))PPl{u^YyE}4#*Ie;p(8a
zPb|7uQg(|42pWU5%5GG&dZLd`&b)KW$QpR0Xvn)7L|IxN#HvX_BH*O0oQ0k490LUn
zAdjp9PZb&j-^RRkc@CldsM|wZd(+TVp0!bw3I(iCLQ?pgi~*f_vh7a5!eC^g2LS>9
z5poC`EGExx5wB|02klr-6x_tC6$z|8bQpN=JzV3eh{efRm&rHGMu83zuW7%{uc&~V
zhV!fQp;PEDD-^$lp<@lbS{INi=9#8Je2tIhZ{+albX;qx)rqVI*+Vg5C-dQ#=QMQQ
zeiM8VL6CDzVr&32y~Xb2i5}C}Ha-&B4$j}Q5NpH)kn^1h0|aBT%*WDSL#F&V)K6gL
z4^wy1!nhTX%pQ3yFOd0{H|o*~2pqX{8nXyYqZSDK6An<#xH%X(V1#L|H_S$Fy2r8|
z<KIHM*1@H_0E%F)k#I@W6Ocx$Ir&0VNboq)Pia9k+!_F8+1-m9F#PR67ds!fMF9MZ
z_~;SDJ8sybGty)C?)H`Rc^D_qP=GKTf+r5*KNnDT+op{I!tpJ$z5L7%NYtO-0<!gr
za^dPoZLcGn$a#M`*rr-v-4ENB+?hsOV?BofGY+~VL8a`aN{pxzW8RRvAlIb@V?s^7
zQ_I!t%RuEg<#_1Y#_Q%3AQ@V*nfxn1VFmT&dVv^qDF;C~y&m5`pht}4V#G3XVF>W)
zs-QR=&LNW;)_^7RK1BsIDoljaT*tGNqb4FT!KLv6Ofea#d35G+cfc118uwrbHmu^}
zwAPjt!_|O(TeHoGd4#S0Q|_@Hpc{NHwAnYB%*u(+blgqcc%rI3c#;g-mqVr&OZq2R
z`*0K9)j1Ps%jHO?K~cMOzN_*CV_}4Cp<|v{Ao)o`voxoLwQ8gYt}xiPUB&v)seU#q
zeCDYYZpq>gz5JHL_Q2ofGRRcmm~Jnak{ua?ZUgB!Is9B<X}wE^^r&9(q~|+mT{T$P
zzTO)Uir-^=fF`Wq7GLDBv_(Q-CM0zY{f6p-1AvhK#Y_Htx_q!3YbnN>YI}?BD_PJh
z#3Mcz^s*<Ke3g8f@ishB0%=H2>NuaL0*<5pqtV76nCaOdQx-@Hf<o7a@vjWRjj_Ag
zDM0=k1@yO3v_t1cjZ%1lPkMk>=8pks4!YkCn<m!EVpms})kZT)@E!~cumsDpt-80Z
z1N)!KR!+9;f!Nz^O6-A+{b7C~n5{si<``j;o(uZr{?+I-O^eHO2l?5kNaE%v!T`Bu
z)I_%rcz2jytS^rYTY}5DH|%_clO7wu%&wPtz>C<%G_mk{On^uWaD2z8qd}<BBgyks
z%GFbkutN)vxh2X8{$2Ukw`k*|?}4i?9@T5cULgc*{g7WZT;#(__p4Hky6|GvZN_9d
z`G83FU{2Fk*G>N|+&ub;F-u>@uC5JvnQ@c4s)(Atb|Kb{YUSHf-2oeC%z85f{OO2|
z0q@fa{5(jfJ16e!M0Zt04r4fbW()e9ygk+#dOMK&2lge|1<Y}x%tiW>iw<`6Dqo37
zEzfe#<zSe&B9dfVOS8sbjz56BE6T+eOVpiR1N8LQc?rtip?`{w%4VOP#ez4~JU1{T
zvTamZ;2&SVtN9LY>Da4R((Zam9Y(JUbAjP6Yq|6+Rn#4@fE0%{q{zSNE}Z<H2(-FT
zP5-My{qess651PC!SM3JFvytNnY&mJGP7~~&$s+5Z0q=q;k)-w7{4l&eD?QayI4O5
z6duVyZVR7J1x~SgiaR+GGDV^ao+{CAZ*u}4TGv9k)|g@egFY>K_w6Q@!OX!i-=Jjs
zx-Tc^$0eK~*TTd!6><)D=;X*<fFXTSw&*`UA*H`_F65+nKA(deJH4N7+(jVrc0Jx#
z7(3lC(eB?q?)Pu@exJt#_Psr?U*T;Fh#dZ(=kVsfA3B7zqzUe)R76j+Xb8AuV1_U>
zISC=qq->I4iV+UEo7H<CQ_P@%ioA~yCb}wYeAK)QC#l6lzqV=i7JprgFWIdecZx=8
z{maSm-p&o%V^=+!kN0jEAZ6yUm9-XMrR185N*m+|m9OGECFDXBa3e{xA>tnp1j(4l
zq)R67l(6c9=n*P%zj1xKfP7gT2?hAAg6X|cQ++!irrD~X-%NuLycfZt0D-!jyBzL3
z06b^*musJ#Jl^f$;v8^<WJ}$-%@s%0`l4nL_*^<AqX?N`){l!Se^x*qf?7EwJV=9v
zaa|Mmg)y)JMGfdp<?kqdqo;1lfevFo^3bIuy_V|Y#3ogH6X*?GKbvsSJOw;a21m#)
zFcvL@AGEkvSiti%5*`rjxVv|B^(28zE`JOT$<nSe?=h`LFd(g1g6fJ8tPol_Ny4Aw
z4q2CNUMZDW$f$cIkEN+HsG#O5z!k*=t1Z_ADjB$|sMWfqP7uhU3ktwR(yuzmi$J4;
z%-C`sjuxRK?CfH=kEr@B@`r}w>lL`RO!ErF2VX8W>gQ7H{li`sy}Xy}5UOY~C1`BF
zXT1=M7_ZJ&{j%@c=RU=EzXs*CL;w7<J6X^-omnCz<t$%%-wFc&A#wR<)i7FP)m<rZ
zc%8bFl^v_UY4*jjGpXJc`V94M;%|v@U~q*7hm(tD;X)<RF8Y^C-B^F!SX+V=`zSz7
z3P*zIul;qDlPw5#+a8i7nhK=0YJ(TlM1Fl=y?*G(h)>JEaQi{Ii}m{2^tdY{_g)+J
zfeQQ+yNyNok>VNV@IDgd8J5)|7nnm?ps-N+=6WD>cXXm#@oJ*_`fKe_<KP5Po^7!D
zc6jmDQZUGhCgz0tpyNH62AWUecZ=nX)QRB0=)vMatPst9p(JNBmOr>kUh=_9-J{o+
z*4fP`9Y>Sm1hn}P>LDe&q1?lE!w2kpa7<8O$diwhXzh|Z2!wX>3dj!jsn|-6YyB_H
zJ}Ej`#6<I$4_{zhLkiT8QT1wwq!^{_hMFlvG<HvJiE@5xnX>TdEY|n!6@!Tmy_<vN
z6Cb+G;xLGvqt%)0+q=c)wuQf&p<5(zto!KZ`qAK!(NX?-@DDJ9K!3D)I7p|z&rh3P
zPTO*_GT>^vS*ssqd-rN{a9&3(ReHd^ce@W=PVj#8RLS;P;dPHUB1G14?h&k#Qr#rG
zeUhjY+<>l4HgH%U7|q;e&MQ<mJI-BLBcDW*1JO^R*J|k;_B>PMF6~ZsTG$#XOQ4qS
zx*{{fQw~MiJo9>6G5<Ip#39Ae&QgMG2K8y46V(gC<b%C@ogXJuFId<UH<$wpQ_4+x
z+_=q^`aDxDZeLX>Ow%@XKR=Z-;b4ZK&N1P}cC->b3oK4Xj<qSqN0yVQgi)_cXgoSm
zH#t)O7QC6Zw(6mxdSz+VofGU=*3I<o6UN=Q?C1m5p5RQ9x;)rPP?<c=9^7S&n%3#3
z)?)An-LjX`h289zt&MNS7k_=7yn2`?z$tB61jKvH9{To0*`_#tQ|&j1@od|!BE1c=
zN+|q!XrLsAemCEv5WoHQ#Ln3zUe%M~nRq`U0R8$Dy`3qoVemol@&fxTGvs025~USm
zuwvq_ontU-!1yLWkIrzN+iM%7`(*wal6zSZz49jl?`bdZFgGW31X8#}Q2gWLc3y9b
z@gsP<WsL4T79#i*+)E;+QdtnTWf`3GTM#Hg^BnZ&FN3l<Bk(CIZ(lA`Ky=6w+e$H{
zsbc?j$|t!c-eJup+(DaIS<@H{fr_=s<}Oqsc=fX4&o6eF`VBRT?pX8(>-{(RZ{FB~
zI93M|D}@;VC1lv*=wrSA3aiqF-nK>=<x<3a;O_e;v|f2>1dUtiAsh?X{D?wPY}+zq
zA9Q<yxT(Fj94%|2nJsMbv!W=tXh_J+rl0aXm074jOarD_uWzUKG9VEKx_uaVR<b1k
z98A)<pg~VpIu;28ypWghq<KJ?B70)Y_!wLID()1U^*~kxff}>mdCfbAVf?p4$)yZi
z03OltDdzc$K~Yhf)GhU63i@z4)l0Vo<f6_Qqg<?^wUKA3O-w*+IJRe)P0S!l*$Ca{
zLr9!!r>zRkOA!NC-Ze<F-lQyWoTUTfrwmbth|W3YX?qMEU1n!AU-1!lJXpBG5AfQ2
zk<?#~e{)zT>tDvG1%G!J7%FX_(P%rtwCgEL3*J#!2{$fJq7Pv*F?@a}a_n15BxTqJ
z7ZCP&b#CsP+lnDc_B<*`$L;Jy^%cO9PZO0seZWe2(J#?=v-Y2-UY!l1!iGlIluk6k
zHd?kKs6}KsDbpV>9}`I?!6Zb3T3dcCxwIkUr`?U~G?(Moit|~aliC~1)n_^n3BZO;
z-p<tsLyT=wbknK{m1w{Z&ML5Z4bD3!<wwdTFLRSn29her050^qq@cd2XVyzS{k7Zf
znEn|D^g2J+U>ZYSL@~mxTc4MTx#C&d+z4%$)NP`-sNdPZoO7A{GrIgkf;<;2CiMWR
zI{{^V?o4(*`~K>sZ@!jz{r@obPEDdMZ5L(PwvAo3&0V%_+qP}nwr$(CZKL;h)6o%q
zF5Z8zZgS<5GiQ!b)J+F0pkPJEh|-~Nu@OBJuG+?l2%02{6bUhU-rp|W5cB{sF@57a
zYKM{0>Ps<~3J-sLNoEv*6Z2b5UCIcVLTO-19q0VfSpDFn6g61yc-kt=pLQY}`-}O^
z$cSdwFG@^o_)X}*ZDp#jXBmoJ*q$LsC@4HN=ooUn6m!lT3%Jcs4BI%)J2}6#9CNtu
z@kada*ydztJTZH^Ad{TW90NG-54%T(%%vQli*_(dg$;dV1l#X@oT?86T#3-5e>})W
zoidW%hdVhqNfy)AyK<+g4+mMvmB{7>{mVtSvgwuF&Rv)}OV6%$HQ{l-I<gNnW1nUZ
zVW>An7H@;sP1mkQ<BUN%K@SwH%V!%9+#JH9Jwg6jyuLIOcm+6GWkpuyLf+#K*zz#h
zTz)zU$bZ{Bd8N>LIHq$%wp5}!pDVHvh4iJdH@MFXr<?!X-=6)GLP!pLxiiHib}TX4
z|FcZ_<<4KbRz@VTudk`3?9({`=1a4Txx)BSx;vByhLdH^nzcxO+*9Ct%n==6>+a$p
z)wuK9MUVL|Eo0$REinD|HK?aK#n<b~a>bI<0Q$X+20kXGp7GrubvD>;=78S^PvaqQ
z4)fk{)9>RN3v3hzJo$APEage4%Z0@|_8wp6_T@gbZ??P7OSKB)LwNWN&;4uD@}JY?
zf8?JGOpO0?{yD9-?Rdb7;CotoSC|ecf}djjTmr8ZQ*^#n#52HCoE`-b?-pS_!)l%`
z^ZQyuz1|O(ALP}u*Gzp~<KnyV4SBW2zQ%Fwh36V0!+iol;&s7w*gohL>d8BR=Q_f4
zx=<YCQajR%myJNpS^^pKXFsnn%js)!@Ackp=~^*H8!68w_jB#r!Nq|Q2X=PUG_dx&
ze|2&Hbbjw@O$y8BllKfPhl}^~`b!Y|Oz%=*GPkFBsT2?@*!D_VC0Jf@t3nG?-b5Rz
z<AqfG8l+vV$hUl5b_4_o@OaPpGJTsp)*CAuug!w}=Iab$4bdfR5Mv;S9q;fDOqTi9
z{NU6nJ&zNRn_EsIh3U>eq)9BR8wMiSp;8~bsfoOyk&&ZJn4KU*!)!&7VT4+rl-KQ2
zlUSpuM%H9cY&~{ut*JshAN?q6(XTB6iirg6zns%Uc21!tPMvI(M)~G=S#g>BWbR`e
zjt8&u8EHV6S3P(uG+D|N7P+vFfpKB1)GzR}O0QIwfMSSAoP+Zs#e!~$S=vM!J33Rj
zDq)~Yq}@H`jE>J0-_br7wnbjy-B93RBAs5W@HVswH~jmJ0JGd6nN*Z^Me?}>;QpnO
z=t}71MR~y<$=LlqiCEKS8Ai#JC+2}$!p2I*R+&^-y%U4^@vX{7fz4O|M=)+LYr}d0
zL7tC(Gbm+No+|_7(jb}kMxxZ2>4y>HBt(i?gL@5Cx$p^>%u87PEXq*MOMXnaw)h5V
zqtkA&)2zgSiHL^IoWJg(4MeE&Sz`mhv2z;vf^Zf-v+|_UNJF^%*$}P{(7)~dd&aE2
zF$eK+$&1d$SYLEP>zt57{5A<QbEkqm;{xZh)xlN*i)l~48x}V<j!=N-F!G~5A(?J1
zTY^f8`g)$XJUza!s6<ROtuzCAc|5IzDVP_8hRB~^Ey!9&SU8OPexxf5sdhXv4WH;B
z80?fueGE<AdLZ_LUqM79@^`(L`Z*jvut1%sU`4%!^x|vr$3v6H({kI9u9`Or_Y--;
zVHJZxB&}IhH~pef=MeboL1(x2nCZ)->98w6rd@4kB2|@3KA0Paep7tsJ+i)wh|g=3
z?s(HOpmqzlEn0q$t^3#2mTgnH95tkP%BTt|`It)*89j_&nJ16gT`p9-4S&PN2xoG2
zt7+Icew-v!2Zu`1=Oqd#9|klEUn#%82jXoLwZ>8p0c+-XbM^!fMYPdG8sEct`+7Nf
zMnZ(<NaastWU{kJok4m{#{G@dy5q*J&<JYCD@^Dc0xXW&2s)`^bL9Fv2*X5X<b<_V
zsY^2<pyr^CqqdTw_#gunj5F}f3)WF}ksE41a6&R>y8?iejeZz<^N^*Xaqu#&{h@^d
za!saJ?Aak%n!R0vSgCITS^>U-eF#P&3+ijD;{72rMT3+6+HQ!qBwS-g;jW{&G#b8W
zGJzD^*DxWJx5d|Y&Z;EdINo1g!)Xz}^GhP_s}^B;1QIb5L*tAygwQQUeT0YN4AIAC
z_j$}U=f4PDQBFd}C1`zhw-T`^BsLW;pCX`H$tf{Aw3SLA^oq=?V~|0f%2(FOknnR;
zt!djqNLvPNrnU?pSnbnPfUWC!l%e%spPcH$tM2(Jn>H4JB6!uFV%S2<(Zn>r;n@_C
zWkl7qIDUNJ#RGcYOzT5i61uj$Bi%vB!smgOkb!%_w?D4d5X;ngw=03nGPxU}5_S&w
zf|t$h!k=U5<JKLodr?n1c`VJj071V1!IB-J|MBN#VP^aPoa>d`?TiWNWDP77ovfkg
zWC$4O>Hn{#@8IY}z`@G;|DO99*w|SA*PE$kjotslRQtT@3h2>Y3z7VACkV+ASjSiQ
zg`i4<J3$8!N-vRU8oIUsJN<t56lM=Kk&2`wg|3m`i{bJ4xN_$dW|0yGw}@iUB@MR1
zzqz=bl0s`>9I&xMQQ`L151fwyK{kkwLO*P6=C?v2AB%;@?ZhF@JJJOdC}ogxQ;rKE
zF*lJ+*%0bkSkIRP9<zF3WP2FUR=Nq+-sN-0;;V%0Wg|{t<E55DhO7!AxLKXv^OltB
zMWy(MKFXV&)d!_FpMZjpHZh1B5Id!rUAhz8AkY^DCdbYB6T#yE?S+>wL*g!cAqEOh
zDufM)85k1lE&GQ;4PZ=z{e=w2LG%9!5ePa0WK1apY{*aZ(6Ab_(CUJYViG5nV)mc*
zkJF0=>?{bwj;^C5VIi-2Sp!SZP%FggeH_qG1BUz@YY`;U7YD{2=uyknNJeEXQYj9K
zu2iARVitE80Ax!cO6)U$le&r8$J2l`#Z+GqZ3C$oq*Ab|57{GX72oeN;;QfTqZ3af
z)KC!`9orP{gE|2w16tJs>K&w(8-@d<<i6`+hoF_lBM_O^5RH=z!EQ=0u>&AMtT$H!
ze`f|JC`N!W0eGZfL<~EpOOG}}Ez9!oOWYC5i6Y=v1qQTTlZM#5MrMQbx}RwvT@!*=
z%tIq^%P_0j^vo;U2ifUdLNt6I!}W*BGYtd~i8Gg=rZRIuBNL&&;tEm*8$2SG!q1Xc
zKmmM_hJz<R*0hHrZ?-#uFZg;I&D}#|C5$-VUad7vRISBI>I1xl0;M&H7;QXPz`=`n
zQ1hpUdJ@-;0YtFV9fyKu`}qmaPBX9f01Eco>;M%u<gCuEqusz5tpDY0(!ZVS@rgJ)
zy%NHX?Oh0pKmL5bIw;bJP?YcS^`xGNU4AWgtNT6l-KAvC(p64;fL&3LPKAcU9FGux
z&)1l`IYD1GKwVqt)UBM%d|Yk3JxGdf%!45s!X)F&Eo4J+){p}$PQ3ni{!If+703W?
zZJ)voWBVuZ_TLQdOOHYD>lD8)eZRWNEbA|X{?(e{Gqv)s>)`v%;Aoky_&dU2mFx5Q
z<?39I^|MVgYJ|PP5R9eg^+?iE<XTVKj^C0g7LVnx6CHFRO<0g&ya71rgsYZF;m;z0
zj`N<YB#iLaO>yvy1<Fcir4J3P4=uFOkgCt<p6!(`E52pCmlf0GE&M_z_@{4Od#~#T
zeyjVq+G{e2pI9h(xYM86B0|y(xbLQ4nuD~y=SOX8BZ46f(*mFb5a5#`tvy`Xv1Z<L
zKA%i2w5FePEn5V9*<U#wsUQt5cwmMXzfvPCiNS_T9Q%6?q6f@|4ZsGA@57qb10`k{
zD`uF7Lj_N|lqt%kqeH!cM{|)ZSp40CwYTcToY`IFN2`ZwcMniCfr%9Wa<*JCc)v4e
zxK)zvBEWWY<s66?x;e3PRuV@2$ARg)l$(=V3OVU7`ef#M6E}tL?_JEVjuu_&E*f9T
z@%3=Ofjcja$z$b%gG>FN-Yx0Gn*%)^@AA5s$cnC>_Kunp@bR`@Jd}vb<Lgy?yyxZP
zeptQEj9fa?$5geQ6P;1o8q`;ZYH4tbo~0&`<CRRw$>>s!?c3Rg3@_@9pBmK8BSQw+
zM(PD>NTHXPe(FgKo~yVj@eTtFKk2Ig>8UyfBsQF_4e*`1O#`2io-AAUj=KK+pR}73
zLoJJoNQ%@NQMhA{at$cB_vU(<ORRo*@er&%&J*=Z(xX1-nMO`k`xfpxjz+UcH*Ebz
zHkFAhgJLD$e=P*>7BBsv1DI3^SM+Xj0oDgl+BMcw=*vN!qQxFwZ@Ke*UhlaveCCj)
z88j_QmI?Geeq_`&n=)Kd09oPp#j8UpfLHWL%Iw=<hhQIy2g7XW$ytfr8j*Z%v6mh0
z?#wIu=1w+4=g=y0Rf{o?3zX_a5eR!Vk0qBIt9~AjmBHX|?Y3)xHf{8_$13uU0cLWZ
z&()Ni3b58>7EgaOBiJx~fUtV#FwJxMDE@h#>M}c7LOyOAQ-ss!iV~|PzCM=rg!442
z_<UTyX!nz}^*bgRUjPGG03QH=eGbb5>#H+Z|LT3e?L^?r(vOcktaBw@bkL?G5g0AH
z*x*$if-V#Wh0pRBZoN~bR@{opc7@IQ=HfCORNjy8*lONzt6$h!@k8q8)2?T2f@f{Q
zX;gbAnzvrm93>^BbKOVy;q98YYxRGwWB0Ap9j<)~8$_9B*Y0Qdo^$dK!esVC>-@T8
zvX22mG=fMyTV$1y0gml%6xY_W6YyB&W?x%6wg}TttYfI~&TdGtm3RC<R@84E6RZ}<
zq74)bljrra_1+Wqnf<;&$TQ1so!EQVHN%^~;3yqk<o~h5va<e1a?ZfQ`ac(Qi&Z5P
zH(6o+ub3;Jvu^wr_s85rVo{nb+h);b>t)V~Q7gQKWqvUJ57(xyZXAL2cxSrHLL{U=
z0YE?>c$(;Dn5b<{kFIXF+q4-XYEhh)=%qA?Q4(3g97*(w@{{QlBhmN;Su(TF^Udk=
zcW15&8+Y+NU1h*1-L3A<c8%@rF7fSirzeSzn|tJtCjxl;?(WZ*$@A*RD0(gOKaN&W
zPzjQjeF;0V%2|ZR?H>xpoRaxtPOHLN>uul5^3CVvQz|eM^C2Qa2xDrAh%9=w$(M<T
zU?~CKSa#viakIJklZ#bbXNP-$sxSwL2o%;oReO1S;R-1U!t^lGl|5cpB`=3&Pd;5O
z@&SizQOJM^GJ-6h*&RCDViJeo)<Esb)|I?-N#3FP-?}>i!%Cc|Hf;o5V{=b=*v1!u
zKV~1I@D)_gOt#Z|S|^8-!V*DY2~`9fu`AWiy_I)!BJ=~{%z&;9$MK={wq%dH(++Y(
zXA0u$t|n^r+&9=`Yr-QH)TIK+u`g7S(=x=^Qn*ikO_&xu2iS212!!TFNn=q(N(Cm1
zn@D?kcB+?wJuSKiy0_x}!(W2`XuP@Yhg;Ao;%(TPlo+oaRsY_w^;@3?tJkk>!<fzK
zv?@l<1f+wzO0S)<jR`fn!NK1nz+nd}&KHpcmywxE5CqTP)!Sf!02T6@7ybKW4cE7M
zSra}?I)zppquteP%)qx2SNNmXDY)EAHCN4}KedwZZA?`SBxGmU`CnH#)C9DG2?2~w
zw$`=82}bFlDfOy>7R~d6VY}DNSVeeW@e=<Wo&vDV9%fDG-+k@EPzBFt&y_WobSPRG
zXQ6Ql@6?5EK=RXb+TX^p@ff2Ww^m*WKv91Jm5N`n8YbtvW5?IWSQsyAtV%9WI!ovX
zQImX&)}^ZwItG=UV}EG)zY^I1Qn7QY0Rc#JAa#K|@7TehoPtuVoP%Ma@OyGit)0Q0
ztpbKl#hpu7<t*oXK@RV23te7&%bTQ6Xd!G$YM7nd8XOom%a>V0(J(2A<@wi(M9M1d
z=)4=&^s?QiC&=}8S#*?Grm3GzQq|O)tIWB_*E`%xF4~iZTy%GTvsy}zVJZ04<2X9o
zq#Vgo+-V&|i>6H7ZQ?5zuvu8Z>aX)M#virIfuR?n=W#gR*wN5=Has9;;<!8*bPG~<
z?r<igX}k4N=pe6oC&Y5l9b)|Elp!P;*%Od*$lX9_rDWp)^(ND}h#^ni!(a+^{eKmW
z(g=e0Bu!;rAH{OB@9vQJ_Rty#h@NRdi%`&5Slc{`nvNikw|ay)dKEvLZ<&!g%PZ-<
zP#HD&rNvCSob3^R>nYW)L`Ka!m*K+q2HmZ~&0*xE{z#krlweNT_sm{@_$-}TISn<u
z)9elSFm1vhdd-IUTU(caFqDI@nE?;&K{e}Yd-CKD@%TIkx>sDM`;kMAxBcbI?VM@+
zQi@mi+R|qlKdc#jR$Rr$*0yd&-`>06nHg|`+V=`Vx9ncBBW&K~iE%!kUItrjsdYkj
zlbsi_OQ4VSzR0^6DWK@@3M$BIw!$cONXuhCx!G<9UsQ)L*>T14t$yaSw!=Z7H9haT
zbL6#r;NZgcgEa+k%O3}>bV`E_3KUrj{HlmM%%N%}7U&)2369bSHD4<)6jGuBSfP%3
zn$xx-BBus46&xavHkChY$5r-{6g_M?hiRT%RmXpzgI(IL>{JzN{w8dqI?dt<hdyes
zjS90F=IFe4+^SsA^iw}UERTNZ#Fn~zG4cqvYotELl;Y0+yO|C!a#OLRb|1|L>OJ+M
z)>#PHU$4GiYsklfns8#Tn>wdcMJ(cP2voOSq`ZzW!R>PyyWY7>D(rS=>I$dq=WyRR
zBy+8g<2Iu`E}rAAR#wul+_LBHr6k04O><EMnFsf^WD=?g!LpU2=E{WS*-qV(H`+)K
zWy0Ol5-5e31!%siN3)*|$k}xucv-gL?Z_b$P)b2+p636}HA@ftJ$5*aqtW8Sn6W(b
zwSLX;dF6w}d2``#I=)yI5y30l3{K@+r7Y055>a~5d+pTlws@iXCiD(Z1NG|mpIgm;
z#b_LC|1-MOTWvDovJG~7xAsm!d%2oWWIiE9-1tUGAz6AsS$!%Sl%~;oMyx;*zLQHA
z-w)uqjz=BR%b&o_4nX2chO7SbIJ52Z=lOe6c9*I>Q6xh=`*w%V#iDm|l5$zO-jnaf
zN7^X8=j-Js)_weDm5t(o`u%lDcYB$c&NdxwobP*{6<Fr$+_T0-S>w8qry{XDj)NM7
z3sdRRyMs=^8M;E1^1*u{A%Y-tviP0I{%LEueP!x!J3dI#u!F(n;ngJ5Wb?42&ZBAR
zDRkEwR~1&#JI6He+swX&k$qNC`43|o%k#Q(Suu4&<ar<*g14;hrr*xa^1j$y&A*yr
zHYoIBYO7ZGbGW`6FT+@gD=<}0w)Niyx0dfn>O^`uGTNEzvAN`AO3C1wF$vFL+;fdD
zaBpq4h7$k3yaCe$7#~>!b-A>g!Gpp*9>pgI0A^FAhy7*q;Em?H^DQLNrh{Ca`jO=2
z3k}q+wg62qc^Le!#FI}v7fpUN)_hp2^%Aq6%g-qAj|?N#;JXdkCC`i9LZYi1tJiQq
zQez6BX5Yxj!Wzj}4?JT9YsZ2m93ikB-}J#eMhBgK+D@q~gTg+13u`OajPHcP|DJEq
zwYcUSSLYL$jN4ymyB6$ku7j?D&1zg^-f9o<c*oDZ+XteuY6%2MwtQOExEzlqw_7GL
zPop82M<t^|DHAR*nxgo?OFBYC$S&$+*g=@R#{||nmp+Plv$17~NA$zdXCF52A=d;K
z0^>{;x$Zfga;bdxH)>h7Uhqa)mT)BK0n9#QTuP*tt7Kf^dmRR#J26HXMa(1`-qyI}
z*$m{DR4$FrM}?3rvj%C#lkT5HK)~_{5$ER1gCyFy-|h)FI=p;Ud3U*MycxsBB53K{
zw(}qVGZ&C#8=xXkjZxRnOJD2f>>ZHO#qrOI&?B7Nq|iz34HUT5#XGr;0iFMQBXV+I
zB}Vle2OD9rFB&)P<ve$1R}eE~4pHJCq-5m5G;r(}SX>rx>Q9-zEk9nDBjFDs4F)Zd
z)AQGI<dI3USmxD1c7$jp4>+Ok%xa8(jg!;y!mAF-4WtV=VgP{P%t#yt&`%I}<b}DO
zRa?42rsM`gL@0M=M+{>e-o#2MFO*`NVMZiU6(txAI+>CkXx}}k{YczNqvU2;e-%D|
z*f<>JtlazWqgNa3nYs<}Byg{s?c$qe8{FA3Oo(R{BL=e&SRiCb=r#C1VPJk6m1R>d
z!loNfyzY}jUey3FWB*oW?r7dPh1njHUXT<p%O1(!P%FUDq`27Rx$#}1;t>{4IA$})
zyyG)h(+pJ=@wM4i)mWOCARGa`TOmfO2`63p`YC;shod_LF+v!2jlH&I3D~+;O8h~Q
zN>Z&jjj*r&&Fy72Rg;<xUBDN9qohe?7+4Ib{Cwpp5~zY6Jnva45qC|RuXtqg@?wB6
zSJI`4wJs@8bPt=bV<EQun>1uy3i10Mpg&JkC3E024p|#NwKeHQDiZjMy&xwL_hTN0
zM^)O@P8w!#FNv_AI3er0Z%5IWZBs(MguHKf?3Mm*dZrnNw~O>h;T2<-?*~`?J`Y1N
zr*+m5)5_vgR8q)m!)J%?-wyt-bCXILr&EBYd|F!=mM>~?)t2F%X$1~6UgkwDe$$6c
zl(RuH#VPQ6RSlq71gj`~1i?>o;&AL)qRF^oa;OxJ&pd+4;W(1kiyBo`=2_gJmf)Km
zt;Q&oR$+a;uCiB&$|)%kfZ<b=OQLiwf%$FZI4oQ6E(-jD|L)rJSdtj+{ulw~7jV`@
zm-+i-I)_97z<#=$Yh8<00oC!q`um8Dc2E29m(>_iC}i8u+2R0_?vpAZ{OG~e*dL;=
zhV$puGYYPmaJuWx*O_32q=)d5R|rACW0LiSkPP5&gWo+ggdZH`02bMhOs#ox_v2A|
z%Jby#_t&6H&`v_uIeG(KttYaP_LOmAcu*vk;~|FHYN)tzX0YfH`C~G_>S6n8rf^JE
zy3P|e%S=XuW73k>rDp<@Fh2Y0063w+M3C@cV^|jIKVpdn!9WMrvMGslpliARPEym6
zy5T!(jnvX3g~0(bqyKzqZ35|c_IlY>BxyxNe1^Fzu#8C!3PnYRh*U2~;!t@h0GFm8
zTKEYUupmI#M(<C-@kln1qxv7^`*w==;Y88CQyy9L7~}M_DS|KF-XiJPD)%RK6X-u)
zaQ$VioIPPw9bw{JxurntcDv`PV05(p(yb{7F;&Ias{b|=^i?y`FwfqnLv%9(qM%Bi
z{s$&W)TO@L_<qNWlWp<%nKVZA_gw4pyLJ%e0EJ%S#cnek)n0;(;q)(7++0`$)l_mR
zJaXGHKz|WU%k$C<11G0F-Xj*c%`ZPvpf96DNUA^)>4s#Xb9%9P3oPpEP2T9awe$O7
z-J6@&TtSskMv=)0xzh!gK2eUomC1jp30ugw%zv7LC|-}_CFFEsQsNG#F323jWc;uU
z+zqUr4*GT5n;-NPynR8-JG?<N0R`p8)ciAn#U(t<&AVo!XYFN^cEt;UiF9#_%3RPt
zoa3WcMN<NruEZ7YT#}>@%6%>bE5Rocgyg!m!+$G{rVO=w1E<DD6I`O4KE)R4<MaE%
zClDJB`CntezgE`&zpKSc&-TA&w#you3IBL*-nq4N#HEY{fC%V}$<oUujEN2whucFF
zBg!rK>`0>pszlvHEQY6-j&`U5@(XT_&E?#PAQ}9H^WGoKB#fW*#ErdL-{$tu2Qdi-
zDn-FY4lQnJ#tvj{QjqA4!xk^B`R)>nm+f&R?cR@rueCP1G&(+?+s|2$30*&@d^N#h
zK3?yKEXB<C+rFPav#q<@y}Dn|!??2{ijMI}g!K*S>GJu(HDQV-tgk?hB@10~Dpem}
z<<=D?wUg1=iSn$4B-0KO&s6~>?rD#>h(qFt9BLV*QucJrJh4Gzc6gli!08k8lR}-1
zo7WrG#+scCwIwR4B*#Sfbh1l!??7a;U1Rv<D<RYwIssd2q@T{qX8Y?nErmpZ*J<lv
z@gMqZ?aIvI3O*IBDj^CC+?7eV%sN?f>|6`%IYA3aI5Q==OQe#)FhsZY<pCtSQr5a9
z?GT*5_*POy0YU0MY#h-3w4;qiOm4mQzJ}RS{hrUEFI(^m^BPE*8SqEEGBjazCI=5Y
z>vZd%k4-VuLkLIl3%8~G#H1ucS{7Qk{&AML<53oE2T+%&7A<Rn2giXi(DBrp)xofz
z8`rL}&%_8Iw%q5J*30Yu4zALVMGbV79ha`p-#xV|Dt3N6nRI<^)a~^N&6%#6$<djR
z@$4DJJtg9sAj#lZk{J!$jaaQ3h9gl59U_T6%{=hN-RnZV181Qz9gaYjtMkb}nIqnf
zG@m^i>n|PN4S|_0?|W$KT<;yVEix5OjVaickq*zp({SI>+ZCvc{Gvt;YC@ph_-}sp
zHob&7hka0;`Nr#eo;l>N9yB^6#oo@1uGWe>wv?NyF55nz<+Yn55gKyC{LSSl-R}j~
z7sc1mrW@O8=4)#)7=wH3P2cUL(I>#wOndDJDwo!s6|88AzzwO=-89;zuX%ycA;2O}
zs7@pC6ih|nOq&Coz>y89=NKLaE^5oUz*xn|RhjVEY?Gx9p@}Wvk*cM7C*CpXQ;vq2
zvX>t1E==Z^D%x{uurD$IU^bU_HkUWsKX`wC5L0fgMmxN!w8p9!4mM^k=NxKkzK602
zIirOR%ePnE<ihB0vT-UdrM7D=Tjh?xF`mC)Hl^R!MM=N}xlw^3#v-Vd-7>D(3kk?7
zX1g0nTaHdj#=p@w)!AnHRX&Zafj|OtDHqv7DVlJrjvBnFU0I~I%yse%{Tq&SR|Scp
zvdRYwLV$-tRen3Oz^lO}PMMFRIEwRkq?wcNam(~v@wjC8BLkU0dc))3ZxEEYS^P!P
zWLK7Od7r3Kw%H9B)r*pj;<H(<%~h^fTB7r{s-l_x_A6w>I@*G~@W%_i)XO~{&`uJO
zrIsoLoXF*%X&}?MbEWGjRq(xA&?1<wgh6{{5xD}XOKFv)TyWr!%c2IMp2m1$`*?M>
zeY{WL0|A>bM41AuDNm(rR<_=bTLm**>tHraCB|=o@ph(n_k463f7Bm%P~!YCYadV)
z#tvH=p#LYFO~)wFKLS7#(n}%Kj3uDQ7s|sT5yMV^>%Cu6;v_L;i6Xd6Y>)AM6h2)M
zSMM6H{xys-Znvkhj4EHG?B;kd6jeB$tF^(a)MnGW1Ha+pG-Fpn+dB4nD0)m<+;iO3
zGP-~AfFp5dXvQ*AQK^g9jgCO1RGW$-iwUA%!W+I5WB((uo(VJVJ%AeSiSGx(>FqJl
z*I><=rD+NtHGO}TtHF2W_)jLl-CpUc?R!|XqAUg!SKZm4K;#ZG*4!yxfhj{>Oi%-n
zndIPV^=;Hvp>*U=({PUE4-~v+-;aU6*dHgTYI47EG-3{&;#pN4giN4QK)0h`NRA_F
z@2ur)!KFtR(Q%=iCpxPjuC4ug(KP`Yj46t>H#DP`hBRl_CjK-YtW2LcPo;CxQ|S`R
zso@;+1hazUEn058^T7#-_s|g_TfIkwIT=b@z_&zOa4h-RQI%1%ola7%%Mhtge5xf%
z@y3(;gqZ_tmAPNHHrL`&{fG_j2QUj5x+JrC&9X1J9i*MqZT+52)kr+!HJzV{k3+-a
zHRHXKwFyJa59{aCbWz)ciQw%N>w;22RGm$@JQlET_56-HP!NuWkrT6N{s#$689wn2
z{o<l2?G1Pp5I?*B8`yI-Zt-@a!=8GVNmESi+h=Drra7xQN3T4HX#V3{DQmsM8BRl`
z4Q5LUZ5+zngWxGt(f9$dcNRaT!%3|_stUn8U$g~AB$4dl)F8k<Z+Z-2nk43=oZX1R
zMS$WB>@I!yae@6bpoO8_AbwZYkud??=U6pW1&sZY0DmS!<_!-ZDR$_lYo0k8_f!T$
z%?}4jwLlY`Eyu(xOiIgL4@zF%jgnfv%BE)x%>r9facpRS-%f=s*qHI*&bkb=_)y5C
zYNy>~)AOgv>Q&m;5${{ABF?NpyDQ_aAE!NzZxDQXnOX=g2ly%SDvm2K#7K^`{o-Bj
zZQppYImTc(zY}YDYXr&$Ot&+x5Nb+Ak-d(Cok=K59qN`Uq1=Esi7FJucqEi!V$A(V
ze2XEDFtGdHwZ~_nfv<|mo`_S=89w~}=+8zFJG!+<1~Lap^t#jC9y_!Tt=p4(Xfm0$
znN0JD@DTcV>YhT@CY~Ep6INI^F=qQ%bgttqUBvF$fo@|VafP(<R5L|E%$V0uGjJ#x
zc*FK*KVpMjI=A0COnV@rD{Z*R-VAHb9M;;n{H-k;c5l@E%w4*akHtnQncSP<8Zfx9
z{t+d6c8D>Wlak8cr?yF<^BZ%LSE}^-8CD{&BCf6I!M(~b2+a8Ab}nXQmBD42ln!wx
zn2D_s0^v%eDm|q=`K6{T*+RDa4{R;-t>{{Cnpmh3vU-M{7yu*1UnT1c=YQzRF4A%-
z5klWXT|J+E<8Kh?`P4hFEWi=YOPD<$<2~JXPgov6n84kx9YjHyF9{+sh@Q^#k3XqO
z!16H;Bt27@za)VutO#6rqFp9ISiK#~IsOc6{zG@bBVk&zQw!X#I42Y+(xz<yVzT~W
zF}&R7B9=p6Pq5o^q{+Z9M%<0P;jR$iNWHr^Y{t5L9k27p@E&+C(+_tgu}1A4^sJ1y
zFBOEbHe?TSMp;Ju000vaYVDHP3#k^u3vY#ryn^63O5*q2z?EwQwlX9|Qy%6%<B}Zk
zIv?*rqo(Rf8z<CsWQV9Qv6<de6p~E06~Cxn<p8?2WP|wAXAgn*$OvRhs#cHR?;dP_
z8OqkG04mC?_8JpNwRKskCN)%Bz>96qwG=-xUCwRK=KTscL@#rSy@Rmrn`XDe@F|(B
z8p4`A9QTm|Q0?xxKZbre3AOBPl3wPWz4-pz6*y_LmaWCWkB>Vlp%5kT0k0*P{Cf7~
zycPO)@IQjR^oxQ{hJ3Qv6mYZ1zsxu){?imrWY>Hsp%rEy7&Zyb0$ksj%-KLrm?FwD
zW#W9*s;}!mGBqCDl*Zl3V*Yk%Ykh;Sq*Rsu=K%U&?g0kI|G6}<jiv6mIr6UE3*>#4
zlogTL9c#XwzEr>Hq)?i?-bIoQCoD*Ugd_(bXHl~FyJJg-0PT-Nb6TN0G<?&#>s=Gw
zC>_!$OZU^&^>)T$nl2!!kWORENXFDpI-q4V>HM-Bg~ah|^1d3YgKPz+H}ofn1(XEt
z;#imW=ksBC*-HR42Qm5tn(+SWny>amaH{9)YLbJ{=;i+O=gU^t>*3<D73~rf&MPWS
zP524AB93(dnWJfV!%!`CI_pfm{=$XtExQJee0UXOu>zuVvu-VU7q!ztbM=w{1`*f+
zL)_ThJV+$Z`EvhHK4yQIxgd2Oe^UIP$5pjG3M(A>VPd&cLIqUf8P93$96>_70y(ii
zjQstGgj>MX5f#}4A|OO*rjdY0kO2uV+`Z`pWr*`$83d$OWyA$>fRV7`KG*>DIB_U5
zb^b>-3fo9a1mG8BkHPa!F9(16`H7Jz3a+e4kr^NYwL@T3C)&}3QYHfAiZEfzFi{oR
z(Lwb4+ULY1P6R563-{7zSgK#r7~*HqkvD(^%Koj{0U4kx3fxEPA?^Fy%?`?nnyFG}
zVu*zYjv97_e+Y|vf<J6tB>D!*!9|Gc09QrQ!;7rS;nJmJfGC(CSY(P_&_iTbupX#m
zf0n6u3&retUN%B)`m4(c^@4ty-GYIfbDtto>G<4v%Jq#2B+E@r${Q7?JcYj;yMP%R
zi^2@UG{9Pv$mGo}aG?A)%B9kBiD2j8d11pI7yRBp7x@dA;8?HqXJX*r0f$G|(r_Tp
zmcf?rrlU^o!2*Z``d>-g)WuxJMoEytI)w2PIXf=Vm)q_u*nH+|9<=Q-Wgl-`V>Cqb
z1^n~sy>p_(H+ItNAnJDzjDLI2XNT6m)7^G|q@<9u9z^5gzLcL$<NENAjD0BSEbX3h
z+2|e%V`;;bWXa6R{W^Z+uvqi}u<clYT!3Pf!q#?WN2sEd)xOMxK;wzT^G}lkDX6lm
zphJXg34Lxl2z&~}%_W$MQfW`@#<mw^dg4OdAo7i+wm&_zuf80TL%V!%XH=elTvVOx
z@p|bxt96eFH(lMEXSU;dZlDE}**>}Tx~wNU?k77kkTNXpjK=>OkOPS$ghf3#MSyI$
z!<zSvz`9i*5u=i=)GbHhDv#YMb&R-h+|>{)B#E=9NzaAeI4zL9Eqqd@3;*rAPxAps
z<_hK~@}u{s$%>C!Uk@>XAyVtVb%2+9-E=^eQf;}LoSx#VKC;$6UhUMoeHi)OIXb@C
z1LU?-y@_N^>)CBZudu`GE+4OyR$BqY6eR%&9>oq2@&yq49?z%jd|~|9q2y~4L5UPC
zHxh-Fl2~vH7jy)~1${193Y}izL8d@j^G7E3xpXDq>>P0?y-HD5Wp=&49C82Ccl7f~
zYdevLw5FA6y_&QNjtvr1JO-aE=2uB7mAy}S1*wV|QAHgdzqRANM)4Z13cH6$d?Vww
z7+2uG0@e70mBAPsqC^w~{x)R{%ejf749Mx?Vl+=WHcx717jA~XcJ>Ht|KVqK@EQy6
z-g^%VUTC#qb2&&0pap*P{;?^9jf8TIi~<>k90<4ywE_DFJLdAX;=@O^YIFyRv`Fn4
zlMXg$uaV|KgW2e<4PyK+F_h;_hizw?q;00h;?XDMupt}WlK|B|?J<TtfCCO~A=s*W
z>gTfblFFNnWM<yr$J|!(P9+V0@~w}VcHLmAvXuAP*Jy<c?GEZbeUbM9l`|-*KJHvo
z{GE;JuID(NpS?)`$KkF?Vm4>T$5VQ%4h>yT@0s<Oo-ld3^ea$|>L+yPY*DtWQwfMc
zTtp^OXZzczV>%4oa$FOELVwO3xWS6HgiFmpC1-F(O@EP#mBZ{sl0z8o@phj&tv4X~
z5DYCp+WctEGoL`W>>`+LV8>XLWxo0q6+u1elHO&kX(7&}ASBWdPUW8Wd{^KfIv{Nr
zv<Kh3Oed$rY^qTbJjd_ScR>Sg&&N<#If&IT<ZQ|#ID>YZk+oir$LP`BZeSlA`bK!A
z<=+O_tIvy03a{(W_@#uf_sFtht_o63F*tKml6&7>a$=d^Ml}tKa2m*#(rd};6FGiu
zlU1OJNJ}59@bg@h5YF)OM70uuX~+(?*jC`60js}%U5WH{2mMno>UwFhdR>%J;NX*$
zHLs~oH*S|#tFJpc8frA)ttY>UshPxHhoYG0c;@@<4jOD7OqEr&bwn#t?n7~5^XW^1
zL&#h@HNgCpl2qKsPvoqt?I<D0k@x4k`peU404+4#{=g@Z<Kyj4=byLi)XJXwWAQD+
zs6Xw01kqxECq^FWtfPJsWL)R*z}FYUYY=+np#eohesxSrx36zdX-$is1ZQ(+wblGE
z{0&n1JLr_GcY}vmY8omA`$#2j8_<CHkD$b0Y219D`$^=Q(pqIroM<z*KVKDZ*<0~#
zYxUjLf0gaBC5tqLP^dq&5#(P%mGSf5n4vIKrg$||M*<;(x`zPKIQP%0wn<-C>S}?X
z3{zso2E3m9D0Z9M#m34ta&5g6M6=5wIw!Q8+zId*M1vW^6VK=p!_6vjoyOH=S{RZh
z5JJTbU&##<Sm?4o(>6D|xjI6BfHo?cJ)<QD*mIW-KU?9k1)=+r`c>eIk}qEfpo2Qx
zfeW~C1a5z0jR%gy2?yVH1#El7k3<|J)QW$)^HzuVGb+>rElY9Dn%?yp-Z}kx-)yq?
zrmrL*@2F862v!O~DadDv!@zVaahK^Sufg92{TKFZqPEZ7ujbjA)Z}~!9HXMMz<e-N
z6_Sfr0Egm90|!o<dZa*m`Nx$$B)>>*R5uOj2fPV}>E^%Y6ZZcQ9x$-8{LkRZV>NZh
zrGH@J*V?&4R{h~DPC+r{GX4k@LR7;&z&{e58P8Bo)Cnsw<M*43O#j#%8isu`1{DuP
z(@$F4Jd3-lOjBGhX|IR<-Gzah@a&RZBw~Vy9)GE1LvmE~`;uN_7;l4T5BWhF+u_oG
zzkUWiW$}43rh`p>5=%pws=rO)bEnx)e_ly`-c#{!_%gaVKh9r(NkbPYap`dfDr{8x
zi8s&v;)$rmi1M4+zS%~1LNnlMIWK<Z`#XW*CK+Qr6~Lk#WrsP#+>U4_fvHmSLw@SX
z5BcqD8FnuARx2;s8sjA6B0+dH4)WrU;#!34j)(VJ<z6=Xl@Ue|Ic*YI&CNL%n}~n=
zX&;Q}rN{xnzsEOw4R>Yd`ZB339j>O^AxO*)M%c(`L?`{PHf3`hEfOI9iMc(tBR-ax
zQ#TlId{bA9>tx@;Icm5Pa5zcVl`d~h9^)8##U@2kcIvBlxRv}@kRyMKlxrFf6Cw8U
zG#7F?af2YUI_RO2{P?N##|xYN5ft)meKwbp%R6wg-~z^MTc^c+0-EROE~n&#G>rdT
ze$;1@)nZ#%nQ*^|_XG)GPBh|Rr)b-UL~pXz#?;?~3^^zEhR&?8shrlWay%ui0D^E`
zWFA|u9z|NGt|sGVJzbFeIHXtCExDMp9rjT4=b-5cJAj3p(6rEA*jP)}@17i&jOj9x
z!8FH<hbwwB`lk0>cOK)0TQIJc^T?@(6G#{+xvQsU(wKo@l?FHBAxu_X@pnEkg44*i
z<&C&xv!3CU!GshNXm^NQwhVjd_;)Wa_!Y0R^ZsI#jmHhTNHfqZ_%9ul2uO$j(F-ta
z+K=yn0WN|NPKvsl%7n%<I}7%wsxKb|_=WZjl=B#89;^#R>UA`JGBC_LTR2&pE70BT
z#z`d28Y*kuzrQRl*T8vO&S|NI#uidJwVAt(CG@zSx;tTJ0ue-rNV8jZ#&Rz#i>>US
zm`7g?nAgl@*r1s%WrArsr4AeT1@alj@FwLKnmPphkRbaa^0kcc(q6?}QALxr&KKtj
z44-z>{LDL?*ocoRE!JL}4F$Wn`x8VTJqk=CyrH~nwFxJwU8@gvN_aa{r%-#xM+7^5
z#K8^x#o5;&XbH35!9FTRv0uj&480n;!c3k*T{QPYIUQ*&OQJG&TwH&imvF=w#{)Rm
zNGhUcZ7k#amWw<=gYYN1{)S)~j{e*s8du=;1SWEZtaAlxABU9<kKPodj*yV%7?J_8
z#(sSV;7J@c`mZzk)r4KK7+8SxND<swaPEO^b`AoE9bdH^tBEawlp7LBX?r0L##22L
z=>Y_hN7W#B?9!Mtx4rnHS<#{H`eUs!&GlFUE)aaXA|vjQ{F38o%UIk=pOp}^IgI?i
zd7`VfuJ|Thz^^a_Zcqn7IXvNVuRah;{WjqMSHvWIY2TQWFfLk8B^5FTc$56Mk~>R#
zh&Rss>kQ=-xogKY;5S<=%@Q#nqAk@L9;U3@r^xdKQ`zT<nh=0Qr~;hh%a8(9-xWhL
zI1ErHOuQm2Ab<zXDLJi@6|rBtv^Mds+-nz;apbpF)`&MPOn`+ZtyG|3WP-}dEW&4C
zk#_FidH^SIb^?w3al++Ru^n`V)flS~zxOt2XXw=C(rL^Vn%3hMzoXZor`It1AN==9
z>t^oScI`sykj3O8F5O!2m;Pr#05$bxMB6NNklG1zOG%{b7Nk;U6tb|&7X;NC1XXVw
z`$CX9p(y8e6bDx@J2?H(QDWR93r_MnZcYqlESv;yK{S|ItcN7?J8rSIg<~gW$)w(a
z&cER-oJ0;@I9iDu^{I3gq$uDF`Rovo!vZNLZqgw7s*CvKDy)mmjU862E7cUp1AE~x
zMzR>4%)^X8RtWOI+I0$F>(<@RB5|Orexn|@Eg-8X?Br}ysN4}i>1V&ts#?mM4E0*C
z0WQOvYrUg<c5o<H)ntEK0V&mi>`iuQDFazLNJ)G$Q<NOi0(lsSIFfAjBdl-1YiMqN
zX?`ZFC8yT+jE9$BbR?_nZncNa6>j?j#enh!mGH-kCfUA;rG9r3Ug&6UNQ3JNG)t4z
zOrFcI#~PPg2*T9o6QkW*c})K$qcp7<&#{bLe}VgejL0>hJ656j=_o>HcMcyRIl!-8
zAJsw9pq>=wfC5iB%B~sDp>uxU*pvNW=#MPY#Lc+8i9{D=*P|8!cA&R*m#_&=LX1G~
zM3xtuoOQOj;z5u*y6pTd{>}yp4+J3CJ8pC=0YY3mS-adu{ljuoT2#m@5|{+Sn~+(%
z4)q9NX%<Cx5p>8krFtA`h_uoLH?EHiW4FEM!x&^~ToM7}5=C!ebI!_8s0@0my5$1_
zhgbJqUu|87=Xj9jhpwV&a&-y$FJ)`J;RA;{>@u6j_maol>>4MYXWj%a-i}6wNHGzm
zo-Q}6nV2|zvE7{Gbg|JbC~g=(*(OKN+Wdkad}8C5h;;quru<(3Qx=y0nJ_S|A?}FX
zj@omn>Lf#GDU9){z#@~JxQ16swwZ>BZ-uX+FnJI}gxX*neY$!oRLJJHS5QxujNCcj
zH@La`3b1gXC=j`7>YJ6z<4G#AAGWB4{&*>PCMYglrZoIj;l{T=CBrxK`*4L?Xu}-M
zAyjruurT$p=t}p);QK!4G=iE<FImD=e13G@>yt*w@_s)Zt2nZDH2k{%sr&nLn!4Nj
ztTbyj`F8+>bVkD~50om1Xro@xpE9u~ZP%dlUHVu&e7v(9Inl!2zE>;6zjC-`XRlkQ
zO4UbL25s~1xMnM<Rax5_tyf7^|Lj2Wjb?E6rKGDepf2m8zK(`Ui))}Q%$=eWwL~uo
zVlqwF0J61eaRz@Nshh~bGJKA<MDn}9Ekki=&FNE_YU-sy3-3q=LX!FT{^1($rrEyK
zfes}onQTAjs*aAi=h5g_^-|h0P`NSB6wO8qwf@cer26)9^4XdS-_T+xvS?^*2l-Nk
zfNTWFoG9q3#4AtB*l571)H8Gbaj(kRj^#o-I#w%MwC3g08l!VlDm}Z3ZX$BnLEhJ#
zT&2?b*TK8D<0>oB<J79?B*;Q@3WRjxuCLFnSA~QZO<Zgs3C~D@xe{j}+AGM$XTlxf
zYVRX^BI*Jq?3rZ7k%SwcC)`?vKm@0m*d0U3X#2>0r4(1*judvX5>eBX?$q>`<I*dP
zLo`V*m^lQaIAU_M@fF|PJ~(!VDY$W1i0Qdy*HgWV=hJj7;yfhEtW0wtoX~?>v`8j&
z3vz_oX+uI#-4YYIJjz0wUc3ycTk|0c@;>M+1CyxlKyOq{1x_MioIHMcQDNZNC|a04
zNM7DfqKc&(!hk;2pQ+KMM4>9pK2A9v-F+Lg%C$O?p6dZX!c!Gn%zU8z{G`qsj>eh6
zu#UH}9FeFuS<V{&etFgvlQ5ldqk<<yf)y-P?4!}mhFnA#RaUI7RAU#fxhHXm{hoOR
zoCe$^cjbfI^g9fX#BZ-q9JqbV<WIjEIt9lPmgpb+iqItbth#)UuKRP9$?;<kW!nks
zYsXF(WD-lJ@)E0nh4aJ?6|q-i+>?-@Sn#)D59R6$k#-nh$dNmzNo6%3hAMMPf#z*i
zJz$~N&EX40H|oJ7X?0cmdEh06G1WIzEwB}D1KlcK)l6&6&<xJqlE1q4oN9^EiBbqN
z9}+(GrE8KU@Ti|9^e><(A>R0C&(npLIp{0s9D`kacO_&_^{bccn%Efi6W_8C*<kGQ
z<s%Cc#l>9B3V<81$+rf!I~d9je~T)D7{K;smTNK+zPoaUY;vM8TZPR(E8k+Gk)X6K
zk}xLpsBq7j$oVcoq3_{V7lbQj5;sVmZb~$chxXB+qy(e}LaoP|6m#%{!>;5J7@~rz
z?3{hoo5SjLLjN_Y5KJ?JM1iZJ%Jb;xT1Z1Se}S9Q^aKn?C@0ni5NoDZ{rcDbq{vPD
z%||p%z;GsvZ7Pezuj$-rb=Nd;D~badac4j1PhG+lAQQS(oFWLWf7f?hxDr;{({96G
z8Un>9%maZ@^9QPt2jPiI)CC8lK1%Ij6p*`+5d{7)dvB@N`FxO`mg@~hIO807*6bL^
z%5Q;;g})?As%bgh=K&$aLx8k`5QL$Y+iTsghvCBmqgHvn{B*(KRV>h|?$oY9POz6U
z$cr3T-)Mw_n={{BxF$|<8VOEy7!ByNt$7r)Ju^7<)PQD?dglk_(WvsZ)>yONIjaAW
z#l8MrG8Fo;L&a#gr5qouQ*~AkRm1}*`J6r-KBdiGiQKf<^5pX*6&@<{!*oj2a<{BW
zRd)fTbFXcG<p>1wr06u1$1_!ADLW>9b8v!>nj3;0wcg}8UXw_s7$v&q`V_Bi#GGF9
zt`;|f4m`*Nuy6URsPtlYn3*+8WVY3oK5NVpJ|jB`go?$<m9UKB4Bm)k1_?3V%-8Pl
z>4{V)yqtq~s;21BvzSy?Qe=mTvRV`d{!e>Dlm>*BITFakdg-*6$Y9J6yvpGXm&pm|
zc;Pl~2Xg%SbXvH*Gd;-874o{-X}nBRC+!i1rj?rd?6F8Pdcs1q`$ShnbQ%hR46xl-
zjYDbBbBY7ml1MW37REn6c<`BMWi%LOw<FG_l4J%T({zxj<-}(BDW;J<L?ag^*A;me
zg!jPL2-%II4!h)&+JSySx4*MwI8(ctd8M<BWi_GIVA5RrVz)8KnKPM{My<?uY#RzY
zSY!<+S&-cwGUi0K&q~)48q<&t$nMuruxy-S+vhVwT#TzXVP2ptm{JmPg3>mzrY$nt
z28l4Jkn8h4TgD%1gS=-k9Y2~j^Fkl0<=AH0;M)5vkZm8yedXrTsh!oYUeHKlcg$Yu
zyQhGAvUy14TlUn<i3}#3JcZ?Ry8~ujV0VuNkN1xgaDy>m@06I*x{6#%7%qF+l&M+O
zMNj1mbYR=F5rT>(svnWJ=k*6b8lByz+;%{d#Riv8f7PWCtTboKH{onxS~HFVnd#9#
zG%|j^S<545$&B8-z}2QFgzRIp%UsPAJ5ogy*jjB{i!k_W>lOH$8T{eQtiYU(8-=ZI
zd{~w~-gkb#20pO$QYRKeoAyVWrZ<Q#+MnJ<n>Izih>}gxmK2IA-{8EP972dd-avTZ
z3HdC5YxOC(o^i%CiLegvm@#Mdg_6wL*U1-&&BZaAtlu$$`9c_7#PwQ0$K8F=eQhi{
z(9@<d8v0?_j?!VEUVs%g@;)C$y32Wly<}rX_g(e<@+Lo)%jpbQn}h8^)latM3y~o5
z%b9!n9gZi5Wf!^V>Pd(Q&6zz^COl&tJ`yYYR$8NnbF*+Bam!@~de$~(wAH!2Vo=kO
zJDPl+h<6=i+S}6%kJszvy6yKUF`sT*?C|g5Xl3w3Qec+#o|@3&<CmDZu;E_6t8mJ7
zo}DfjEP+0Ab#TK1cIeB<J(B3icyQcZK1Cy8Qm6Q_o7AlUarhNG%)>ocZOcDv6I-6R
z^>gBoW(IOHr`sm|8q~ar_lBvN0l-1R#T+^~cc#(GrW~r*+nK%c@Z^H-o7E<rxA~lQ
zTzd_(UJ(uK5=*hb#$k=Mme{nRPE6(gCUMR+bv*!FJ)X;_<VA)rqVVE|CUwRVs5JZv
zDAXU>?VpZ2NdTHlA!Wej@cG>V$;BKjx)io=*TLJ)^4~>SZ^tp2I7KKl8md=kp7~$V
zBgD|>QeUEiPD}Epqpaafa4HViJ%?pp7CEq9V&y~Y-Tv=648Szt^1nl`I1_Gbt{N7`
za3*n_1s#iM3~uFn_bWa5T(99vu;jSLRBMBX*!G9N-GmLI5klkG;#p}5ag%YYM~KLT
zkHB^jiq=gUwH{2yH;&ChZ*VlU)48Y<+rRX~2BQ5A13cDeC8OsZpWYmNpL7$vJZ1km
zN&Z(2nvL;)u0da8tS4*=-+A{6+K$!bOqB<cDuGgB1EHXFlDPv=vQWiZBdRlUx+DhN
zT|brmZ76cXbwKq$5W$M>D7`hz@UV>afS2h}%i7{FWqvV00_VO%qml*5NdU}OaE~m0
zO`Ihn8ndbay3qY`V^Ey%eY*R97<<PiO}HjnvuxY8ZQHhO+qQj|ZL14ic9(72c9*8l
z#C({D^L%*!!;Z|gGjpwLK92ch_rrtZH0Ez2Aw%!rGGMn$5EJosTI8JZ#&9M;+t2@0
zFNPSr3Q8TAX%1zNu1$z?>E7*>&jR$!8&cquIQjr6nZY)XbXs4#R<?&<C^CgB1hs06
zuPCZ`OI0li$_PW3>)>7vJV3|M!WBdEI=I4#B>szpvjd=o<8g1Ry4+Fi@HA5|(znA!
z^=Jo)<(+o{$C+WR4mej=tq317L?>tPDIIIk`U?@%EGk7bnf4Lo%VHsRkJZ^{&3zhj
zn5740c<XT-2mI5;vixgv%UUU+N?x0|bq0XiYTv|$eFMz%?@wfXQz)-qWVZeYtpQ#U
zq6>xdj!6L=lwrN%jOWD?2W<({DbCioBBkw7QueKUY{zYe(VM)E00Yp}PUa&&V<toB
zRM~mbukYCm(C70Kt^J5i-Bs|p2P>MAkse?Q2G}WH^+KWbOA?so-f4N2-D^+^c2ffw
zaHxzyUIwm&WQi86P6x%AEf-?@UqSbawID*IkWIVc*mx7zM-PCrW1wM67+3;5bpqHS
zcl0F0n*Zd4UV(5~-^-S?Aak#MTsanSqenv`eLIioYHVdbRMMHgbtN!>%kGn!NO`up
z&a+5Y=miSoK%)80F8!@_fAS`ZR>|uIvF`RVT|d)uKD!wUL;V-U%*TR8CiMpqh@Gce
zV&-RfmFP2|JVwWRZ+d12{qPsV>S*%k(T%evDp{Pb=Ese+iQPg|6k*@+Wh$L?2JNIX
zf6#}iW10b!<IPN-t279~!8wLmX=JU|sd%S2eRsNLZZCv~Y__~2*4+#hbZC==VYhtw
zrMh5sY@87F6|F4LvW8Snlioj$o*uo{kxi}s^~PY3d)u}ZY_|RqJ`)<-1}0nAzDk?E
z$$KlUt^loG`Fo85mln!#?;sQcOR$4(y?zx^fQp@G)_Ta{*#ZXr$XDq)M^^o!Z79(p
zh`p!Ablkgl^X@;!x5l!>EnNO0_vBLgM{_?o@}8^dX!Ewj?n2(vAQcpgK=Plz;LUqx
zhXLT>D{-*&SWID<u+m6j{0G7u!Q3m5dg=A;k!v`X(+QEr*56DvjIiWrxd534D5<G+
zL46-^^Fr~Lz;l1kvUd2Amen~KO~~&>9qW=eB}tN9bxD#PS4o#p?A?>vmEj7|0fDuZ
zVP^kU`cD09=|mJ_>{?X9fK(P`+Lh!$ZAbDQT}SrbUyyA|RY6*>HnNXBFDA5K<u+~>
zWAHNX=P<8r;0As7wC+iYj|UWEWxX`nIL?Kf9Mbj~Z~sVTmrpp3OA8H)byW(9&(wS|
zQ9)YxO&+z1=VT!apl)0ep*u%fvS8rcaPL^BZ9Aq?G(cYDlwLW-;~9&>ps68ZOY9Ar
z4l|7r7L2zRkiI=CR=k=k++7xPGl>hujy4gbRHIV0Xr5KWZ5|?Uk|w5_pg$mx-FR9d
zFY;vJ<xe{8qFEO~YeNBe!$l}Kz?>z&?L4}Qn{y|Cc+B+=tL^(b;rvTd;T_OxX5ehR
zT6qD|XR&K<S9r)=VxYqjQT>|0q$D(3i6;rJlZBVf(ME1`xa*POt#v@8Ns5R~F!uVb
zvz-k@c+!79Q3tK<+tO7>)y*etTlmE3_0fMVrc5u);AeD%_OJu7zu%ow%K&prT*;$~
zYGOWt*6q}oh0Vh_+8Z{pb8b=Yg{~)O-<CVy4V8;^%Wik9;3U-(r^P;0zKihwg^Psa
z+Z>8CVp$oa9i$G$lQg?3Z;y<J1Ay1`@Mp5NHj2C~nL^d<pfe!M#|nMm(P1)iW_+DA
zZu^jngh$xWM}{N;*rv7N1coAoPS5T<)~c_cZ_t*joLR>g_~bV#Lm4|Z+$ZY-Zmv=(
zwOiQmtAsnkg=8UysL8iPo;T=gdvq3@Z56uIk1TqrZ9bX2Rp=J|t^?}u%mlbC;yD~W
zX3prRP=21pXdmrd-~tA}iWE}aDjxq%8N^-tH=>5Yce6r4W8#rBW51!HfKL~OZpJ$E
zDhZG6mS4u}dTdw^w89n{$mLkPn3z;sdnUW)B(%Ekog(bw`?N@RZKbU&^48#6(n3>H
zyDZvWQ}pWQalUj|5S9*cH|CeBeh#};o&`+hqHh?py)Sh1;Y+r`EWPcxxi!dXGE8WL
zNgY1z^z4)T#l&rTp8Kz6$o*gT(|^ZE|LZaEHuk2=rq~m|VUZ;G!SxO~d|H-@&P`cq
z1Yp4!P1ggg8PAMcp8D=Oce~|xZL@PWS_HNMo%QQgr{OA_jwIXP&$Gq%`Opd#G3;@+
z)p<X3X#D@fU!=)qECDc|KD{k})<7y^PfHll$wJ9<6E!I%_IW+|eIGq}jiSeE%OqIJ
zEWV8(R*cDv2nM_#GMmx`Ji*@ter#JR_IevP5-5+2g+a?S(}lw0Fi#>GnX$y4;1WHX
z?CNCzyv?1O-rU#ob2teY+W_9GC-zN4c_*1=&d$omuzLSy;*Z@UJqrV9UHZ!H<_7AH
zNMvKr`kkGXFnS_<wtiP0$89gvEnd1q&Pfy|VGqcwYRPE%8JFD>cIMj{K{U~rBjT|!
z$7+e_mq=!DW~yZ91b1re#p6`uI?_K=g+#PDQLEzmY~JeM26WCnG?yffPOdHb80m`{
zyi|8Y{r0PG@Q?xg%nc}NI)x@q75ukp5FY(y%WcBja52eQL@};M4R*2du1w_;F3t*J
zMtOnDz~cjyLRyeO81U4o&LZkK(|gpJVskb`0Hky#*eYlMbU$*IbRwB09-6G!L}svT
z6hsj6+E9{wrT4WKZ9PLsC6*ZWE%;kB4bDO+8Hy8cVGYt{jMKzmZ7egtyvlXcH)<JO
zDLeRY75woQqT3j&tOPAfV0tMQBkm>vrb9VoaBnk+X4&NZojp-t12A$v<JHW2FxhA@
zc`l>`B9XWzGqN)NVQZ=AJvrz^u=$fkA%FZQu6i{#Tw!C9A|w*v#ip>k9axb!wsEf&
z;3<Y!@h*F$U61ftr-!1XtC)0cP}B!;Xl^@5YE34b+1b~wLR@H)()*M+rixdNQphN|
z{R1I<mvUkRwNtQ=URf$T+o=xYc9Xr5RnkFEXlR1@$m#fj+946IBeFBW`|ytC)aXaE
zdZ}{1@EPWhVD}iGcJUl~=MgjRQ<~HFGSIF@qRp1i+S&;W*(ear<LJm2Z&U@sv^lc#
zH6Lu2-$Zd0kt$SSO*iV?Lw<tBf>snv!55(q=(!VCT*x9b%F_M+q*~z1zr+aEL2(yf
z6}(QNO;m!ernIfKI@!Mx<%Oi8Nft)m>rgKMBx$J?9r=ew$%Rs5e2R15Q#d%u;Npn%
zP-ckQCb9!D8xP;XjCRVaD+K5qJzFavQHVq|8k(hl7wPl+?ia5HcktqYp^o18Q*ruZ
zB*HvM8$*BTYT>-lO_G8hh)jZ14wPo_UuB^wqqCI^Pwf9smV6s^g95_+Y!M$c|NO-A
zKWs4kwOpeVFcaV^gND4wDD!VB(UO!VwDjUs=S5|}M>fu>)P23r0rb&tGm^3@&gdw@
z<x0A~2JwnN+dJCxLq?-H^1LENtSOC4+D3i<mP_lnng)Q$W&_W%74?3bzN#;q@jKw5
zfnKNewrz-OfJl+?l@C{<X$&Ggs6=R%jKqBZ6le^jXWk2`otI(j+GQ+5Uqg-`>(k(&
zfVL0#{LPib;&dfGW{JlZp<CFcaH0Gz@yv2tbz4IdvC=)j9&>z!k|!Ltl5y(SUgky3
z0!bQAn>++xvTW6XdaqSiS$CeDNtqT#Ja6~lDGg(dzaVZV=rIH|6=R*{-cr`eAS|u*
z25iAN$5ly>(p(lY_+yYKRRaL>Jjj$43(-$R5Hb2g!X|`bu3zEFYO(qsAi!R465>Rk
z=T``=&UySv`S&clGz;h63(U2C@8hj9rZUf}<41nUBg|*(xb3|7Ks-z&HLLW3s56kY
z`B|Uif+d7@JL85XC$*9j)usV5i5E1``Uu;~qGd~h!v@dSXMRQgRAK<{1v_JZDU%K5
z9Bz$O6F1AGNRAy~5D4ZP^=Q33N<(sY7a;A~Ck$<7_2-s|e+iiHs4n++7ZyUepHi(e
zyafxhWZwG0(QeLL29Tgki@}Ivo{wXNJtYeT3nw)!=k`*aBz%TIfMuGhHWz8kbFFX$
zO*U`T!`ytC5y_`C_QJD<0)496#53g>tGxB+RS7SZSay?(i~F8LEA$Is)`0l?1>vnK
z@jbA@CI@Usb5}k%^gyr_LviXyjQ%=rxfo0>;_2s_&akhcz^EpGpFt5X0JA{zHA>Mr
zGvwE4+LV!*kcf1#zr#;=dmV-JAxY#)bI_=-=bQ<QPg>ERzAsM~Bg#*nUAu=s)t4Ef
zHg*BBwq1wrYX_pbc3(d2jO35LJLRFzY)$=DH^*0j&LL}AfFg4y1#~*}fvYr|VwY?p
zb6*MbAQODSiF0@kVH}+4-80t@G^X97g+7;sfS=bIJaejkAKc$g9l8wKp{C0o`ZhtI
z`C7V?JY4V?3!Oovq`&zOryp!t`#%+*f<7s6N|}6ikKgAjan>PSGh5wsllVsoE&jHS
zj?)wM;R61kgNGIPu+x*hH}yLxzOGjDduwI}@~_t=ZGq!jI8HNq#2bmt276D;mlA^6
za9cK5?07AIKmg~NqztT_m4m-$6SdFBI@o5MK?RxUGXv8VkRY`%J+CT3V3R61ZpaX#
zwK>Jg7i0I&1+85&oBi!w>XhkEz^Mmi&pmizHsZ^E<Ld5%|4c7fXt4e~W3NA+#OrQ2
zzNqhH01#G6xW8Cq5kdE0;H-?IFd?^5B@+99iONoX|ED1M|0h2Ee@V{F9GqPLt00)x
z{>K{OK>B}KBePmi`CwjW1O6S-6{&4aDWY3=R?*FCQpGeX3X6ujh+tB|Z1&vvr_6`g
z!FvpiU2l!M4Duy8Q4)d$euDho9s^c5Wn}VRDDs%eQY7T4B4sh<k7qO%5mGKZ%Fyuz
z1%3R1&xhw>v^WKQPoa~m4GQToyWL){6Iijcf$oc#4Tj%L{`Z@clZ&2<hWYwZ$PUEl
zMKr<S(+Ih&v~D6j%Gb9ov1#IQ$!gRN%M^7xVYVsRY^{t-iS-n5Dw#Q7IoBsMxps$;
z%K@IcTNbpht!s`J?#d3;$h6ebs<o1&E5l(d#pwy<wpZSryAG2yo-%P&Lb2ZH@kq_V
z8&CWCabY!}>BY?v4_qyRBs&9%F4yG_E&9p&4K6<?X<1x?=JzY{)qbj&Ik>MZt~f=v
zB=xcE!;h@5Peql0p1Nvx?KahxFCW3CQO_G<l7&*SLBQyYx&~%@C+tOc<%Pa;PkZon
zq&J&UN(z|e1Q)X0Z`K|zD4Sw7TQG8F=ukZ7B`#({#Tl<*D&z=e+KQ+H8kphl(wKy6
zI_TAAaAW@88ALSaPBHFSpUl>u+kb>XBGY0tUS7A)e6UpNbc<@3Yrys=T2b;MN%Y9c
zMPfqL^%6$Yk!?~Goo=_+<ii#9qh@fx<hT~YT!T}Y!VV)iZmSpa+<K;LIB6~>;PHqi
zFRy-Up+8F|Ci|(nBjWSrny@3F3Ws(5&Hj*YH)T?0qH$-ECWK6Vk~S(PghWk%b<|2j
z+NN0ZPVElJ&j0jZ*s13jR)I<Z8f|$VU*^S`zJI_pgG<gn6D9pUIuYHxQdc~gDyS!L
zd%MaovLA^&kRd9FYSIJqB^F23(uU)FWg@_`H;xL?9mseqp@UL;As!`bAjIm3nn?oP
zPpd-4T#61T031^DE2%nROf9OyE>z3;cGk-R;2MuXe@*%pm&QZ$`SM65l-Bk$Q#cnI
z734?5fAm-QWOWT}cg2mnAgk1n;{O}f2?<chEA-#e%+d&geenG$ASsH%y4_SIbQU2C
zu`>Eq;da9D*+^$Gi2(r_E`=q(=Xcf^_p`yvtA6=A&hpr&Yhtx<+3s~-F&KmQAoMEe
zjR@I{gKzyXclrdjfVglz48ie};UcIEcpV0Q4_dJ;k|v0tIw)DoMtmws#-~X(O0DNA
zfRa+F*m{HV^uswv0NLRO^RQXDdWp&mw7Rs!LfBxQ0ZGy&*IO@v)sS%vUNv{vVhc|N
z55#7$G`s?+LSk&)BgeAioM-0Lp?r<2^S63mEvw)!hSN^NvQ!cMbr^?jo_6uhhAkcS
z9eeYGS{}NqX-7UAV@9QK-92`8i+ce>^vUK{iE%Q6G+s~Z7#_oaf7Vq(tm)WJ-!AH}
zDH7LrsN0uDnwKmZYez#A$47nTnV}l&?x~-q273@EkQXc1l+>2ct-A%5#+DZU2%ieK
z^iD^yT6SN<UtJ5(wjU|u-YaRU-n#PKn7-s0yh9!XKIbWq-xl0T5-K*h1gAyfsiCno
z5`Vpfe~9TM1%;xEaHy1_cQkc9>Hyu5NXR(pnZYako-whEVo^o9a;0IE*1IDA94GHM
zp(d*P6nl)GZ*6&6$~~Tp_{=hiLU1AxVMi-XMf$hMc9qZlnSY}oNrcm-At_VUmjnDO
z07K|vgw@@tGrh(QTMwO)7hWVIU-FTV%x6@kN02kt8LmrP7(dgu@J`ns!Iz08x<!ad
zZG=pB)2xeMZ^i|ASr98_!)FHRKl0Bso>pbM0gY&3m2(k=xh4^UO94)~mW$>PUt>&Y
zNCWbad(lk=>?ej5FZMVlb4GL1aRm60nQXgP^XHL?39xC?T+Nbr_z$`hgm(CwzXnr3
zKFu*cFc^l3&|;tpiGybRM=&t)6fR7^%vKdZB%~tw>kKr5KilC1OcPwIEZ~`ZYWN7B
zm)X)@Mt)D$cgm3i<37{dsebW68oia%ntlRGFa4B8%vcGtG@7B~=)FH(y>o7FkZ>!&
zu>YV^SlPBt8-vh=;8=iY6up=}e#VO30dvf+QOTp6|ATT2hZD%rAojJJ0w_n~HI2md
zY?;aSLYKr)*eF+2GY^I({H2`zsGs6^B<c&$;=f8*e%|oe@Z$-8f_@R|4@f*obq%+E
zU7)+kiaZi5T|E{VHY-jGVuo1=53rFet^9o%i099^7HjbGL^W_Tpu5Tuzj0Fv>udn^
z{MH8{ctg~t!Jhb!{>I6~{C~S&nOV8n{@4C?tu5=R%89h|p=o7Z5E^bl7GDYui<5$Z
zz;7Qt8PV*70@+3B9K<E`amo8?e%b5i7-zbdTc7(jgD<`7x65Znz5`7jK=8?i%_qRD
z1BMMHEyoZ}VGJ*AUYn^G#VIO=*OS-n<p)9|W`UHeMzAhz<Nh?cxJx<KAYs~S17(mf
znz3~EG^<H!8`)6MM*!S^cRqRb-q6R-8{nPog<t7Mm@Xd&`)ca806Bq?<=>H0kHI<V
zCunr|6_m1*{RAGF^~J|3CN%IbJVc7FPE6nKgV9_)m6LQnYtU->OhQ%`?LBy&%ZuMn
z1?w_}Sq3)1wC+J&HWDIfua)RLZXAwLB`$$7^q-gjqyq)Rl&(_A31JgOJzfsvw9Dw-
z4Pl00!WsA}$TzIBt{)+39=P2}ik7a{^A(T}%WCWsW#RS0NsgoA-VF}lxUA@HciYCL
zq>I@qR?y!&3-Cz=U3_q)^qf#M&rhydOJ^d@v-Y`|utR29CldS)v0H~P6`m;yp%<ng
zMSr&!hM){xIHm#=$>f4xhBZZCjxd~k1sA@mlpyz)_vT$ip{#X7BIKDd>Wxsfrr^t*
z9+L7dF!%HU`%Q@{kQ@zbg)ZUEW=0>Eu#6O@N#amJ1Zx`84pog0tTNNxi*LY&X3sh?
zMd=|%oTnKxw5rAx_#XT0&5bt>s>X@jy6KFQlTn9dI_J2l2>-k3l69A=@t&!5rD3c8
z^SpBA^H43OBVDtC{b*)-fMxav@BpUQj~$EmEDeiFr-Phz-HuU`kOmqPIvFodfxj!u
z)27Qqb6@^if<OJNSM<aDYY|?|@IK9`BhpI-Qf`60{D4{CXkDIhZQ>+usyB9ujHKlY
zb5<uLs#;CAp@__?cD%RQbq&g=B(onf@?$yMVZiDDvomWq(MU@AAnuD4rW_Kf2OM>1
zPL&BR4kQYE;om+c@ngS;hcH4GJg2!SQ#GHhE+{57GM=~KRUYyXIw{sxP<=B}%{^AG
z6)>@3f3=4(<h8+onj(iaZ3;V^wj2NNsNQL5xs(-?w3cZsh|sp4BEq%*C>uWCkfV*b
z^VdGqJ(IeR=|Uz$=n@2H*CPEni(SOxI^dOaa^2-urUCmQ!~$oler07{f9|h6#Zk{a
z3}7&xd+Q0yd_~93LuL;yeT#zolSM`bjP{zu*p@D2xK!LU$@TV{ce+61rV0M3iXR2R
zgnm~B>7TxC-TVqqhOBjaj7Is^xfTv@1I1V8fR=83c|eV%yznDKnUjk!sCc+zUs3Qj
zm3CFKPQ(097q!5!f75k>V&%@Gk?Z&V8c}-0NB!qx61@vNY)c^Bh4d7zG!odu9|t-I
zSrh4ORtH-Y(;2N9bgNEz;a4H+D<Q7vCUrPy$o!o0Fp4_)bYST-Q=IF}xZdq^f5nWu
zrMxG=CXdK4AAS2T5nD?=^z>2uh%KL0IC%5Dv$J*S!(jcruww*>zV&FoeizVY>JPTH
z2XPB)_!*M~S~;=477QkPIi6*|0UWtF^G9@uGJD;gjEq%UV65%PG>+R&_<Qysn~%0<
ze?Z9y{X@CoW=|PyWmTHkuP+ePEDwGy1YzmhX9POItC4vRd4EQ}LIJ+U*K5C;cEwyk
zW#k=E;~f=&;Wy@vCJZ7~R#_WNBFer!?k{4eANi|Eik371x>5&4yz=b;7b%x`QyAVG
z#!rrm2LnzJL0h6y`7sRjtdD;55->nrzq2eJ`r)vdfb1xO4WVVscv#9f*_R&r?G|0b
zw>4{T=ahistw4Z|-$d?1P3)z%0Jy(F?a4Y|@hX%u$JzY}SR7A0=bAmwx4EEQmK5#B
zFixrYDc19Ty!8vwtP|$$KWq4Z8OXT)_Zq&fqn(I5h8(bK$n1CmGU>4ki3I6nN1J@8
zmCzj&tQCNj+@iA~t3q5g^#1%_-F14lPTU#7;ju?Kqg-2G|M%i*f4L?`GqC4v@_D2}
z5e!k?Xh23+e^+fnLqd1-S5-Y=Igxcg%NXEl`hL}QI@{~}@~P6f{=8_p8}L1StcN4;
zv3dIQ_wyaFTagpU_wi8U$PoDZ`|l#94t%H0P)Cy%8%w$h@;vygi`rv`5d;3?07Y{J
zXLLG}ML#VQy{^Q+SP&R>=$SN366MPLBXCVW5t5|CX7a!LEPu3HeBMuP>W`@+@zO1d
zF@L~0v&xP<`c+xa-DK4i>1_JvTER5OwA|1>q_yE-w-CjUQDKuizIf)@A~!8k>op?%
ze|OaVsNOP!7tk+$P3KG>G;Xw+eNbP|f~N`(DCFN4F2b}+9;L}_X=;zxE3F$ZWM@z>
zCx7ylv4Q@{OkC~yQtFuv1+>hTw}E#{(GvZ({W^13Jjrw>nIsn_O80p)f<BvHkjS<X
zdWlq{6c1~tXZ-dTuG~+r6jPGxl7SpYO06FGdoWhu^=r7RftCrT*9Vh&HQi$h`oKfe
zKk$AQZxv=bDk$EU=`lt+OXKT1%4p_8kj+6@eVX1<1jKKFX@FHqFb9S_MEjgH?HLkG
z_7w@fdTg%n%b(<iO|3{B);>R|8Yvr`s<+=JE1QKeRO!+cBY!DTG+U!b?^7xt)rMx)
z|9%9ZNa1M<&MNa|f#9^__c!neNQ8WOD&1`2eGt?tUO-NE2G0c_j<N>}iZ>$2Vr4^y
z7T6G)atM8q6etZAg*NiDXg}8=X8uG1t5HG&D9!~Rfkh{bmTkPQDycOO`><?EI>Y<2
z#=Dlt=P*Mz98NCTkFZbOq)b^y-rl_x7&02HpZdtFf?W1&W~eqLdQeB&or=-aAKRAf
zmZHb95{{bpEXtHhGNAx9?OCf{Ssuep6OsMaQ1ME+KU=^dMR%Dc_4Chf60X9I!fUyU
zm=NxcJEg@g*Ya12jaOzlL=Bu@M~=pSLkbc;lV0sK6~<AS^<nH}^Apbh+2j>Q-I_7^
zB0jVx{k+r|g9o$17Ue+8w6N-OQ_1BBXtVebBCbW5uSmn5MTu*NoZ)yU*@)yWX+D&Z
zASglG7jmmB?}(scAw}Ork>hmgLmztRpy-EK3JYwp2Y6;YMH^XSOL?f?FG_Zb^n_ga
z&hhQP@vWQayx?ezkGU3C@?<=LsmMubT{$+jW!ARbkVPlXtws77P7rABDIf5QyOk50
zwGYaa9pvoW)<Qvx(Uu8zJgfMp<wuhS(xa1@WdubAT}#3^tN(zMbwwT3GGSa&zo7uM
z4V`Gp&|BsvpM9AAU}y2G+M1y)O^pa%NW=NASRPgxm{Wx=L$wwi(cDO1kckA*R3Wra
z5BrHhW=9R5#$H<lR&qwVwfk>gEqn~w<o!siKXz>{A8>51cmhn?c2B4$Uv3;e7+QWu
zf3{yv73#m28{lm;$@`xfTHN#U<61bCB*j)KBrKE+pW1tJ05Ap{ULx^t9?c*X53;8p
zri<4rAS3O+<1Y_Eiu}8*IU?2l)5dwTDf1eui?N|t^oh5aTX+a&)_(0aOAMgPv>o)C
zSjlh|7cUg(&!nri5Y9j+PLTmii;2XfiOxF<ikPhyetPS;6|ri97M7skx}a;;o#Cg8
ztE(~OO%-oLB{{s%6j6pix4VB&a}6(F`t&)zXr4o0fcE6D({FR57H<a2d_`98D_SlU
za#dp<b+Gy<4Htym1lhSur7-4*gE;p%mU9q#KPwz?)y4QlWK%y#`xuWP{i|S=vq-BT
z;Hptmi%tr~t*S0uaEuhIf-~SG=T}|ZFV7y5JIMUy@C1`rCzsdFojFRdlG=y9-1KBg
zE8<t;+zqNiN~Dp)p~J`ko=Ea=`RfaJP)_~Qk7bIl6C@GEmvh38)oGJ(IHH`-Ri8zA
z*LIy+ghe>VwytM>#TIm~<|xn<EH)?RA{49ixso~IhBnUg>F?<qdXU2ldixlHr0W;A
zjsBsl*y|55`6jfw1UXVo5?$h_813PQjN&yG<drPj7K3TX;`R<4s5ls@NH}T~5^17e
zw@WZY&oVmYo|Ec>TL(_BwmvBAa+$Usm;OFtz9qXp6Q=%xCIwXDJka2SzoE-A69r5e
zB|WCJd{5+zD@Bdei|eS!ww5C|)1ax7m$cbyI@CdY@Z8%l3fE<E0lL@NS*@1E)aM>)
z?omYKx=C<X5vTMng}fp&lv#$S<n%Ac{9eP6;WcvOu^a1%rby83_C1y@{Baqn_B4@;
zrDAcC$Ih`=jZVwIFTdxOCV?xKPt<pwDM)52GwO1C6pH!~dBp#O4(AG`4Q;!)i!O1>
zgWOS&2k;!O$I<3}78^J&9^l0Uyt6X(S)$e_hNMX6gf4SV@1xbmn<$yS170N+#Lwi!
zBJ~a(6P{`iDhZN~9}K8<(_VKx`vU}iWm^u;15L+lrT=`n0iz(2N^E+qm!j0&)b!<M
zsm>L9Iu;FGGXUZ^&(}>p+IH$LL??`i!DmZd!8$5mRB_U+B>#+t)8#B4$j<SUO#Tcn
zuw`VrYkHi}o4+B1=M4rg{>kA6r@I6VH!-)k#9@#y);lL1i`M5HSEpl#k+@B44XSTo
zr-)||%cae0<3VF%`rEDwkD4t|1PPy-o6q$$^uRiYTqy`{fRR(`78&z&PEkb%$+(~u
z2MmQU(UB{EFW_5ULL?&gl||npzmAEyZmn33T^M-+9L{DrSHfrG&XzS2m|{K_)tI84
zE_%{Td~I)1+!pb-;+Q6!KF0Pb3IzG<L{A9NBvUB4#uF2!|FJt`uHX$3V-?9#pLF2X
zI4|DFeIcNf-BxX;=a|t`c~5c*pn%Qu|M43lhI3+2H~F}L`n0C&R~+N}p36K<0L5hQ
z5v~v{tfq7)C=yDw!rngi{7X`3W`f<0Kf>HqlFM!Bjn=005_I+lJ1u$JhB47@`$8#z
zkH|ms+e;|P`#15_^>A85L_8bkHKd!Li~2O1q}=tmK(m<JtL(6Ed9sPo0U{R^Ha(@0
zNY!kp_+^Vne)9u=Q_r#}FJN)~-iVp;&F1fCS~73`mq112kh-q(-+etuPLMoC{}rSX
z;fo#Rp1c^}42i+xpYPdw*9LqC_dU_in8G#L>V5g@mZT#GYV-(gyE&(y<{j1ft<tXM
zEsl{@d5Y01R_HK@ypab&Xr>P-AlUoLB0K=}@dQW|q!{On9<OmgQEL{;?#Q8m&E6Lg
z`kW!RoS<hd{CVF~^AkEs+yy0tX9c3tSdk75qQq9Ds{@^|+30|emy3w$(SvDlN-R(>
z58SuQuo9lOW8;FeD7CSqa3CEkwL8@_qa{;=k~)>!cZCoka-YAa<1l$ul!w#WmilQn
z=vx7zZsHav!!8}d*VfL9J8oy*AMihm9^J{=z`Qqkk9zk8v^?2<M+17gGZ=V~47|*{
ze*ZL7I;u%x%T)Ls7cnEB$sgZ-wKCqOJ-L582eOTwbw1;MaW#{cQfW!2mdnkIf7-)S
zHWb8iA<g8@Bv^-cR9deuNhMxx{7#Ote0HG`*lk><qwp|Ftq(yG4&(PjXDDCR=(OlU
zhff+C0Rfg-IQnTzhi}KV-`_a9%9I*sO|Vs#wo>caC_KqTu^vP{VjQ_ChyBBQkh<hd
zdQC^oNzUi(@!T?G2oY<5>vdf4R{zHx<B}mc@!QpLgLDMXS6Y6)l*0m^N8ZhE(P{z;
zi9uo48m_6ip{hmo*Kub-21P2&Zt?d$e-ZD`wD;5(x%v^#X~Uq~=t8xbPLzYlmd}^h
zn#<T9klT~z;Q!L+%>QAaWZ`CI|6gY;$2!}thi%A#uZ9Ede^@EOcEQTWMK>tEGHp{%
zww{OqkUGWm6e#C0Y5st|-Yy_15H$Mpn+3UrDX&p9C_0ST-ER+Jy6GF%?62dCL{3e@
zKw@!==w&@iPS32ltdwk}s^%Q`m+cT&@2{`vUBT}VWn6bt4$37bqywL?^M(R{|HpUE
zRoVKp5NwwowQL51uge_UGK}5t>rLCywSyMG_m^Pr-^Wu%!=fcxoTp5x0Vs2-9E&Mx
zFqEv;^=X6(n;pNU5TYtf#1C5wjA_x9+v;+LwzX_VjuFLh-zJ1@kLllAf5)ny{V&QN
z;XNa-U+Mz8882N7-Ih(EFJP-NIdc}qslPI5qH1t|eyX5AyGA{MfMxa2s})V3t4JWe
zd7qtQ7B-7EpAeVp?Y>N1Pk(mNmtUBs5a8R7tGQaorK~aA#dRHOc-RtY5k(N5Vdzmg
zlB(8*djjH6iXPO);|YO%Q&7`cIfalV+}q5^Zi|Hd84+~DMc_==6jA<xPuzXw(eoCC
zQ1T}Ugtlwxgwtj|7>FX_eaN-2<LSxhEKKfcGGCN<iPA0wpK{QylUtcEPmRGGLdHY0
z_K8_quz@moS%+>bkSXpi$9{WP8}_%br|OGE8&~W*FXOcd_uG5nB5BnF2dT^fU)<3M
z`40om+%?H6=u#t)5j;Hqjv*xX7e7~?<Ady|7g>V^1}U_>GD0MA7>IQ?Tzo*)%#rJu
zGd)*VtuCLXdH`d&F~U>`lCOxBuSQ_e_FOz0d+Hu6E&06P1*)&m&p2QnZ?voFciQFw
znnza0$bk3<v#&|woMat_T3q$KsM_pN!B%RnjrIQ$B98Q~rEmA@c-u-=)5xnaHM1ho
zRXJRkvkm?dhv>chfRE_jvTSfNkx83)d0oHzlrNl|XYV{14v_|*r+b|lRG6pr%1>Tr
zxlrgf)y8-Rad$cUGjIyOe>d^``5TD$_0NWe;GhbJJgY31Drfw5JZyhZCe!b1bGKhs
zgcIt|I9q<|T<7)kC5q2_l`8FFyP+&Mn&=jl7cWkcsa)Rb+MdF4hd-|ZTw};_$aSJv
z8<@|(*9Wlj7ztwi!9|@Q>`BU3_`{7ZJ1%`hi0?j7%1_P?^vZ2=jcvRWs_Ze`KZMFx
zPP2R(DUe#zynkb~iyOs>UUQ*+@t(#uT+FA89TC@pk7^E(^Tw^oS|Gzc<49*l(D%2Q
zIvHOgcQhiaJ2NIWhSWu&e+n9ne9_z$NW9!3gxDvC%2kSCNfnYk#(3rw!-92&bSQaV
zyB;w&I%PA!-EIzjH~CzBEr#Q|fSHT8ij%{QhaNj@dGQvJ`f9@rfZe4iNB2+D$)N3%
zfQ<77$q$>j47x0Oz8fn<QDh+S@5J$&PzsxnE=15B!&x%ZZ<pTV7=j)5uUo_&@I09H
zU$aWgCvA?_gLYWPDq6}rM9+u12OQ4`uCv1{no{zNAr<mncBMHISWuceaqU0w#gwMq
zf(DURDg(otwkton?FP?I1zpQ>{v|u=zhd%+;Z}sg`1*W_bXm@uy5bhfI30g;P1bfw
ziFJ!oyduzRrP*9?O)Uc@tYAiIb|i_M>T>8Rj_7OV?%-Q_!SmDI(%7sX;+x9M|K58B
zpA4bHTo6y5akJ3RJ#OUZzz?;yotsc}#52s#<&DgZo={Wpbigg*E|~I1jP|%by}@)O
zs(Gf*DSn2}pPGEm%_Dbk;A#-+uS~j+=jj3=g&N**d&>gHf+tI^5KS<NISwldK2#8>
z#Da=mwcEc-z3RfrsAt^p+!Gj}P|e-+%tr;Sa$p%P!!B3EGj@r)`DlZ!i+2=rd&@9c
z?T<c9D}i3JW0?e4^2q};u8rG-poat$p0ZgY#btgct=c0H%VtVvGrw^9OWbu^2NR14
z;5A-qORSi7zgP16TQ~zF&!A5vpQN}#;q&0(zC4;KK;xsoIVea2zc5#*v4shPa6x1=
z0+OKz4NQEN<D=)d%pyQm)_KLcZbQ+Crq`@S6Yi62**VkQPhZ$lW=mzSEdKzA9$=g)
z42yO0U;%~ou$d3<JxM`|`QmvNu4K|^nuG+jWgk>hC8%^p3a-UU^H_d{93T2kbpbfN
zdfjys*Bf1!xz6cgjW|RS@3BGZ--AZel8p@ebNmI`GF+FmIp@s8p9ST2P}1-FN!?IM
z2P8rhuRur3r<|HUoj4?Db1fPqhd3vxe7e{5y&w4&wqDU=8o;M3l&kK5W1<-%ak~~x
z&7Y@&eN_ZuV{SaIj<7a^_f?T<Zdst}{vxBgDX?K*$3n0_l)5KC_P%MyHiN`-@5iuG
zN=zX+C8VGbN|)>)x`ici+X8vqfm~HRd264eEO3WFoR7L7ICs#Cx5mR@2}uUHPLh8>
zry?IqvId!TbznBl?-uiZzVWXY!YYIKemURN2jmZxzK0{G=H9bbNK4mtpRrXjaLdQM
z!X`@k9KQ2kynK42s7@5brw}1$6rMxNn4ldg`yx{^gWJE~2Bju+3xCN7)_v0S?E6!@
zHuLl&xok^iSNNTLd8IlNDU5<RS5v2v?4TjVi|0~&&18jZ0}ow|n@OTxe%{hGF%1Eg
zU2)3Or5?6`=ZbYd1YUVR3@N%1K`bDz?R~c;!)Xf4=ON;~5_7@<A{!785#t+TkzyID
z*MbVf-<&UQK;Rl4)Q+|)a<sr92GS<S3(q#;r#{u(as5(aF&x9Yt~a^KO^V*UVd2vM
z*u7jf=09Vn?`2BS`IuHOcwv%>oDYo4Abm#gGdpsI`s^!k_De||LqBm(Q?9aQq9ODM
zWaZ@{+f+oZy`Mu&99NX%`uMZJp1TP|o|1=(I{%QN#!;dc`j0;T9M?$mt92-nP2Q0^
z`hz$kACN=BJoBn7-2~vq(hc^s{6y#O(-34+21bX)CjyOI1iEZy4aQ<;P2Ha?Y4<<r
zjVfGDw}PlBg7--YSU2ZVI6@m_yZj`I_wIUfhRo>^!uokwLp1N?PSe{@WUJ8jQN9%O
z8?T5wrK(<oo5li;&u7>0MA(@8`5wHNOrs_CoNGh4>(C&RLFNgfd*RQZ)KP#bmcPnz
zg%2)mKgXDjzfLJ$O@cOFp0dLrSYbU8Ype28tWY}dv%U0GU41R2Cs6K2)tYrv&SYu#
zj4x^ZZ)9Jo<&F49Uyyd#Ju8|T_A38nLl=ay+r|Gn&wOdzc+x?)tZzuGIgpAF6h)Ic
zi>;?(cay58ip-uSbz%9-x7<s5)|#4KJJt!Jzh$?j536OaiSc~CCwgxpV;DjFu7HKY
zP*Kp|qG<;-ujIuM1Yxd)oZLtodeE{TfU1HdK);37!#hw@LE;9PKKWHb`xsh?do$j+
zy_-6s?m?2q7}JXpX6n3csvS8&ex=kf$(i0)>~fh<4}GuX=!z<dP=;|iig{6`1PO~M
z>Rc7!;OM{EW-ZETXt99Vy$8NyJM+lZr(}vu3sdtYs(Ve)CY<f6YiIkCS92IzMWY28
zYOZ+GxhECeuCSj?tJb!3Ty3{$^5dZcl{$H-MBLv%JaLkwbV8i^{>rX!w!ktGVc+bK
zQ&{w`!a%}PgTwo8{?>b!r`PFYWm6VrY+Q@FFJyqjnh0vNRH2M)>FNoKY}Nt~_mLxW
z_Z^Rhg)3oJJEiP9klL34P94JVf*2%vtrEmTOBc6DmyrkcM%bpx*-}4HFI&Q15wH<R
zyE0bCfE=s8Kbg&f4d8LLZn{fAgtzak`AjGK^r9ljjkVq}{1REk#{!3IdbU_DX3RR2
z9>-H;#4plS+ktc}5MigkXoSTqPCQorqNvS#UU$}XQ{Pmt`5tVQlj+fL)SvzvtJLBR
z<;l#Cv%377@dwUvW&gYUg5sBMjm4|eS<IV5TcnEW$3?C2P{j{0&=0tNuGD{lH#wQv
z|L<if3)BCObwB=Ry58hK`*#e~$mwlEyat##N>WH+NFB94P))_7icZLMau%mn`1U7+
z3d%8(mCs#b<xd793~moZL>wOP0;BB9{q%Ew*rOhxVE!XTlNvi2HZJBA&4f8VLgUR(
ziUaM&gNZV0(9IX<{QZztwLao_I!nmUo2MFV*(dP#a(;5GVQYx+G_in>_g{wpM`%SZ
zyyY>7F|qNyhl)%%InO|Tq8S0yi5jteDXFbT?CZMA<q4@69Hm)QBa{9y6;<57UPv*V
z0(2OPPAZ38>MSXwa2hrf5K~u%%*;2(e@mevMTG*zrihgy#$D_)FSbI82Q4GClJAtI
zq2L@lrQf2cnWI{4sl8}M%bC{ZI$((gZOUGXJT&G1V}dGHAv(X0L5r>xfUizOqNa$(
zb$}6ulDFi@RGwpTiM6qE<CAp~ZrM~@<EgxCTuR<)j~!fkmBAf5J%FrkGP^|Tjjqz#
za+JutZ0ZC^Ky-l^)v$mFkcYcwOdg@e=NW%IyI3KhNa_ZU$KvzFf(6NtD&_JFGdl0M
zCVHzpLuYi-qt5mG{;8@s%TpnW8CZuph}gxTK%pB>1;$>>_L52Rv!a&Zk)<#04rT7d
zZrqu1eK<Nex<#$=%S{TCNr+>C=0szb;+T9mde+UY-8;U8o-%WD*-pub1~KmVs+Tgz
z`#euYR~5aK3Wgd6X%8s(m+J}ELs5m|VmnJGm7nU*9>*iiq{ymSiBT1o7OQ%}vE(U2
zI^RIvHvmm)-pB=q14a%-x*jJO)k=<1_R&?78YU0LDiaChbZoLyC2`aQ)?z3RO_mIf
zslx;^{k!S4WB>yk+pe9>^_wGQpp5xhVay1}Yn;3kxMm(DieV+m1>3a6Gs$VWc1OQ+
z?C*lXJKe5@`C$aZWjwC#%CY+I@tof-j0-l$2DL3V=M|g3y}1iDwaYj8=1$^q>jf=v
zf83SfGI(6&AQQ|7x!=;KF)}v0buhZr5wusz_T0C<hC623Z9O3?Q|$#j8kOI3%r)45
zfu^e$ba*-pD*^!8t|cp#z+~p2b;?B+g^~6FGD1+VdPjuKB!;Wc7gGqC>ohHe%AM@!
zP(}?XT<tN9q=m~1@$>>+$p!pR=+D1!8^7h^P*&~&HtwE3@<K_^(RP}b0t^~F>dhL9
zk+|n5m7`b=j~X?~`TVAO`0ALLHeeS!f>%K4vMK<g=0lAQYJ9)Uc_p!VToczTc07Dr
zSL?_t%Eyw(h--B<N>*Gv6|hZFb@jPbRfH6zB;ttbhWi>YN7urcZbaO#tU$l{coWjq
zIb42You#}KJLA1UN4m+Z?@z>%e(ZNVWFL`<Y`Rzo5cv}_##YX!%5h6`pv6k}u*}%B
zO`!ge)Wp+9IP0`Az<S<QrKr+F^(!CGgUp4ASQk;sG+IexvEaqQx)Jr5wNW5Z?M>iH
zlo)Y)4zR8&<<Ivl)IjnDfh|fuFqsff=3K;H`G<>yj|78|9Z&d4HnyNAa;afOn+d(d
z!;*#;Z2`||7SsFo)-#2Smo6pAat%fh=wh8S(#1Jc!t^$>n0RVp-<#%!2d+*?5$a+`
zJ8EFBUw{I`CV2)eZN7BFl!ZeAieE+~TI6D<0F-FbP7v*w&n&HOV%0hQ;UE<~|CZgb
z**A>0FiK2zsA2DfKr!=l1cpj*Qnfd5N>~d@%JIPAyI@osJ~PHb`-b3@LO>9$56n9s
z6{c31#L-sRz-t4!VMGU95K2ZS4&h;bSUjqR1#SsD-LZ!pw+9L+F#=)AN*GYU24Vx6
zZ>uI~!bpA4L#6I_p;+X%$P(LK+g<K}ACR(MyB`4<ap3d<ap(y@lQ&|X<y@xFwC#DQ
zxhNuE3>>g_2k3-q;M;ocHbO(Ytx_}{1f1$Oqk`sHJ<=6wVL36CH#Y9gK^(Kvq!L86
zAlFzGnk?K#?oLi)FfTJ$Cn@fT?FNR-UPxiHL*N)+`$?`?^NT=a#0HP!)IBg%;2_$%
zE9+nk=cv7yFTiGIlXUIVME+oy_l>8n2TEp0!mQmEFahZbcJOvmX8mffnP(0pRUf>X
zqw8gYaYNvrax|CW{&ITWMz^KhNE}pET<4W{L!6f6Fp>p4(~=wn9g0Y`g*cN_<E{Pu
z7bGB{<ONYMtu3{D*qoKb5Fn7_-X@>)h^<Zo${SIgPt6qn(}3B!mnM2m{u%pudG=D>
zu7+~^osr>5B%LO2Z*?<s4f`REUsYYETl44ET&KVC)HqeHC+t!g7fr+(r*gXMw?^@=
z>M6IqNwAqnMLl<#h#~OpUj4K&#r?N?kBnm(E#!6nc0Ema&MqErpf2A`Q`8B(xkB!T
zBQc5ocK58gGSyhpRM5H6eK&b2mI4qqU$>cV!mVlabA)kNd@C8RowOII?mA+^J+(|<
zQ^%MFL9AF?y)Z6_8ys!UIy7EiysFjO<7Kuv?-4Lkw!qE&GM>NMNd5(h=?yww<LtP#
z4-W3=z<BA?vfDx_M=Jh#KS-f(LDi4ASU&B|#&$s9J!|DZve~<n*0b+QDUiRsSRTj!
zFY3*yVa}#6&fLia?!fnrmwRW7KREB#pSGgF-8zkxn%bxrvJ(`)f=B1yhFf=p|2B8I
zmT%>)t3Q-`X~?iD5=779DuY%WYmJ^{Z&U0QR(wNiq$vJ{u_rG-UQ5EDViI5++obQJ
zZ^729&-GgIf!9`>R>pMK0k=>T-%87?ci;Le*dBJU1qd4NF0eO|5<jzHw00{$!1#Nr
zU;i~S{0H;?zn}!nob3NI25(#Y(shds=_kK$aQBXxTReqauskgb#2=_k$mdCRbDT>+
zUtuJ1D6up=xGyhVqqoIhUV5?g`tF#nu5|fE=40`O;CY7cf}iir!rAFb2+TkvB`I>p
z<QTk&Qe3JsX~*Q8|CJljk`Yr7kKc$5Th3Hk%xWq)I`8?q#J7*`hH&;oPo-eQi~~UY
zT=4C|m%0%6^3{TEI@R}zH5T~yz3U^f0H+>uY25jV$QbnIG5;%9l1xEl>&QAM<Bzeb
zrlq@1p%Ljlb`cv`#*(Q4ku;PYIlL5HDhBwS))dN#X1};tKuN58y{2R1YCSr~76(a$
zy#?LNwMJx#bknZebEkl!tK>?!O=D-Zjh&l$`2>Uy-NQ19^@Uz~T~4OXiGyP1>LA(N
zm{q?oxQHTIoK$|{-ETWuwtf}!KVGAst@FoO#Gxh`rz*=_m*DBGw60%WgS`ySI$aUw
zFq`5qJOha{wJa+g8jTQQ;4bPo%jB<&D>k&uQz7b(jqXLcR1!WiDPnNLi710vV)uC|
z3TX!Et-!qW%HYXYYC!wgTO~oxfx#MzpS3)Sws~zUYmFH3@bNoKO}mas8TC!Pk!HMF
zc^uUOil<h0!5zj@C`^6mt7kVa(vjJmHmNJ;x@^{rdZ1I<mvZc2w_MUB3}u{OE0#Ov
zqGN~+ofqE+dx@cSBEUs(0(?Zo9j+R9SUyNHnS{P{K>FN|Hpg`c0X_9;GlA5`z;pY>
zDkPwKY{huIwfk~j%wD!wT#d+16LGwa+U52<1K^`610Vz-cx3z+pT}1pyS|{sVM20y
zlhQyX9{hOQR-8os?&cY)`dnNzn=u)hqxALl6Z-#hl1K*&GWUB)Cz{Yx6w1DRI1s6y
z1G^79q=JM{GE<8aCZj?YVvC-bg#yrC7Vz+ihni^{nVq`#S2SA|wQjj#!H|Ja_U6fs
z?tdPs^0zjWpS&#Y#-4yD9q#06eWAAg>Y3ZCE297Ny^%X?85oWPbVVn;c)LyQi;rY@
z4^LF*azdzXS9*V^`M*xlI?G<@q6PLed>6TLSVpXJIz;T#TK;`W@A@ifzroT!N${#s
zY{p&nh-4-_*BP1~cSN$4Si9fRS`6Xk{$nLL+`$~K<%!5PSIW`0!LvN(o{?&I!O-p$
z&fuQuX;|ey9Y&_7!1nPW8q#sCP(o&oAS1%TewrH=ya{2dhZ-R2q+O}8>=?^$=q}$C
z-=}}3Ta})5AEJ_PooeVuU`&a)FSB5U;{m~pT)=mB<Sb%#)aYPz=edoLg#__nNcw~(
zar`FZXnRXuQ7!MdC7<2B@x&0)K+^_%rJh$M-}Raqz=FtZO@>=>ua`-$PjlWa{)YR<
z%r@S-@b1@hR=;ACEvjEQnLfiZ1d2Saai(y8ga*9IX#YRFML(I=o2>mo{NfB*FjtJ>
zehr&0srYgAT62xYqFD+gg_apZ)|RW|)o))U=$|?BMceq#s>SkxlAkT}TvVtvbctr0
zDgne-OQcbmxqyprWaFZ@qwiU;eg6gRpy>xsgO4Cog7_8%AV)bCiF~6i4F3y4%&j$A
z^W5vul$fDcE&JDFRfN%Bw?^R>xuB>b4m<vl>~v2C7^7h=MUSIs#pJ3_(OHf%YMlm}
z%P5dn(~!TP4G&Ex;Mvd-O5tC0R~g-^VDFGh8M@@_Sv8A;mHmi_X{^TW7_hdbjR>ZH
zTYn7V@@eY6nyZ+c&qwjm)kgo}>iWv}ND8eSRex&zXshqdEc0GDCK@g6&m78La&{SM
zE8}3;!837g8VdrRHCKNr6VVKq*u8>D_;44c03k6I04vD+koWux(?6{JYNF5X&~I)K
z5t9WVeyIprW?`f73}<I`ZpXWH${?3PXuU26+~@a`!(#0l$DtrzdKW<ElQZ-E_`z5r
zO^a54JI~w2$q&l(XNF-G7*abzHt+@yohM(^$4=fq9tV6nha*KA)Hyk&$cuB0kBU5f
zwMRME+@X<^C3xPM-M9;ZR!EuhY(X59PucTotK0FwGFcczl6$clHBvNpfFwu9k2%9o
zNV!_#@cue`&;V7*61mqCLmo`QZtAzvcvc?y(>?@6(hn7`-oP0IXI>+}hXT}^SK(bz
zI~*NOE!t9Vg>%T#Sq7&<$&3Co&Q#-B^TGmGYWpf4p}+VT_GV^iWeqZB6eS5#OsB2^
zv~3+5m@NfayGaB~Rf7gv+7BTmv=&m*p9CkR8}L>LRQB(+MoQUjw2lhu8#Os>>$FR}
z*c#<ii*XT`|G+w>fQ{#m==)iO6O@_x{z#l@pig72?F|0h^kG2$nC0enF&57Er6Grt
z|7WB8uN*WM&i|R-vyC;GvekCqYbbn6o*NTE0wr+xMg%w(qn318@r^lBapuazH{(9a
z2J`H8KHgo7cW{!r5pAxOQRz$)0=0UA{etPCWqkT}dwD)ye{>Yr@)B1nRld-_;mh)o
z&V?|}^#bVL)-gU1L2F!+!GGN{?)d$jn2@vnJp4b5y=7D!LDw}3Bsc_jcXxMpcXwv6
z!8LdY?(R;4y9RgHKyV8ln8DpI&%565uKVLz>wfd2yH54#)xD~xtE={@v-e`oGFy*P
zO$qhQ`^)Y*8*9(wH`<_p>*>jJT`|+QgU7apT+6r=a%Ih5e<<UMQTX!0l;!ZT4N|GM
zuC`6SCa{?JpHyUI*ofsJ{^Gp}r30WN?a)|Y{Aj<Z93&V3WcR&oV7a1?RxC3(d*8LG
z(YRS_$Op^8`eQ0Y027hRzG(OeAh4#<^Q+6i5kf1@3sdcQR#w~0UvJeeu-}01PF{6+
zPAvdsW;L?5V#Hfj1<qSx3Xxh(dZqqm{o4Ktk%3q?$U1}ZT`s#SA<gKf#*yMM7gtdG
zW@kFrC^a%qovkI`7^_6$)}|F7x=)-I)1hAHHcyQHuFS-W(u?{po*YJp=QMVRrf(|l
zzqQJ#@1H|Z02T?SIH)J{<y-9aKWp}i89NQ@G$_*%N2N}N4KP9r;grO3<&dQN;RgrW
zG?!KEn8he>sv3*>Wq+G}(o+H0X~Hk!Pm=U$(2RDsZhWHT@=%$jBUB=bRebt(Xp3c=
zLVg<mR=Cy*{iCDWPpLaIF~1mEM%s6MjtLM?qKbc`W5C26X~WHybNb8k>saECy40ff
zvtzLhkdh?|t(@as7s#i3X}NOgx%GfWXFXHZ8LniV_JPIgVRb_s#fh9X`6rS5-^>>@
zzi;wG_g&&Bq1FZxrcm26e-N2T<fucWv1fLQ^LUu|(6s8iOn=k^hJNORoqv(gzk$ZH
z$Cx3<yN}AD!bWyJeBa^8rhKe3tCp#=noFGOZi$3JY+iLIJH3FE+LQsnP}{s>sv>{u
zG8D)D$QQ%qhXr8zjGy=VYVt*)aDvTbd5L8_{JXP-Jmje0=ERFF!kGUNgxbd1WPeIg
z06=Pe-~5NBzAxnXN1MBFDUkQ2%<jL}&5{XAqsv~eP*8YDTL_QDFB(P<o$I!XamU3$
zY5o{OXVpK7{Nxe;P#_?O{97bX;Ord1oOsF53dDc<tyhCYfJYcQeYUOSW6hq)Ec^~j
zG|GUu+?n@)^vO6gnlYq_R0yM~kQSs4rWj4u*ze{1%mnTbabxB}7an_*ow16MnUY{;
zd379<K+)kL4qt1xhTbaB3oiR~&Jo(;;Sr<oFA-f36Yn-rmYPFZ0Z}8j_|vxi%@#M0
zU(8HABpx(D7m}n#sEJ@99I%d&<a#nd8rvGZ2UyiAg!g9|_F23Hjx03;t<5EaFyWJs
z0~SqnmhEGh#fO?T<K6$7m$z{@TgD)_B?TAm3c*m}+>%ImN(BM22E6omMfmr>3&DS@
z|M630Q+L__6B%2Zk+m&GIe<BY5-3Dqby_PU^j8{Q>QML9IdlS{pe%J#-Jnk^9N~65
zHjRD~VOI&a!7QH<J|JIbMr-Sfu&RHhPT=D{RKraFBku8r-5c5P>A&x;&kIAmF}K7f
zw`U3!6c(4-wq}1vV=-<MZaxLa)Jtzh$n*^_b`uZNq7{T~QtQ+5<~>DJPNkqs(nmJ>
zPjFz7?-apR04OP8fasG+K{K`Z3}PrEHZ1>OF1Mu&O$lzi?%IRNXvwAsu&p87D=kt@
zB4+rDavMPfJMOy6kD3$yG}y6Zvw0$}J=(%nz7RNH+mHO;^2Db_?7fET2$z6y0T<FC
z${=$1Pw3ik3YY&h1%s3e=v=?=gEViY;_)S<|JMFSE8TCrgrcbO>aX5I((L4P%|-~z
zCEqwF-JC@#K@X4~FxaUKru?<WM_8%PZUilJ#Wkfv+0%&Ly7Fgw?u*DY9UfllH;e85
ze794W;B9q03oR#)nrRYfM@)FWY^!Jd+CUKHk1_-{q&aEo$iq0oPCVb3W4i-GVmXs*
z+d+YCF`Oiel#G?2xGcIv{i>cLQv^Ci|M(^}P(D3l4-L9wd16}*GFicy6>9q1s(c^^
z1=-aK|5TK}E&xU1C%lecG%)ph{A>D(Gq#QIACIVStwMyqmS#0793rCMXkN*^d3*yR
z5FjI!OK~Zf8-FAuonez*C8Nb1J{QV2eo{qL6R;qrreKHp>ZtD`bIBTQz33%v4KOEk
zS8Y40ZX&N`uX3<ceKd2Oc=rMfQ#lZEEYU8j`0WQQF#qBQ^wUYxkc3DdL|+n{SQ^Gg
zF{tk}Co4CHv`{k**ArxFAZ7#jrgJO|8#Qvvr3<+bHX$3bjxdZaJ*4;yT_vO_twc3=
z)GBPck=9SvBvBO9TIRA^f@WeDL)YTrR+g8;0jdpeQ;h`KMD$j0T9(lBY)?^TzCg`L
z#IgFlRB)E5E8^T)KBZV(i1Pk*Ngdgp`kqb}igg430vZv@u}B)n2Js-KWM}fXSnIH~
z0wYVI@%EW<P@a^7i9wc|mX<>LBb$%acpBp0dH`u^#M2V5GW0{TU-2|Ov4Cr3)`hh8
zXUd%r#81Wajj@`<jDBN%Yp3c0mB#~9Y|c&~#ErLVG<8?Zi&98Eq%^PTjBHAcbA{fj
zym-=o1f~*pyim?7ZE0-v)2RXUsu}Al<6jKNic=z0me7;UJ=0}lgCu6hZsBx2G1RqR
zCs{e^dK<OE61Z%A=_mKK;7@Wk9C1(-K82AVP-Fb|ZKzFbYA~j-|JR>B-?}R@J(Zss
zbYQNn+V11!TIDpaK~*Mrg_Uq;_&0<BpjG<FXnbCoeRwo(_ABnl!uclY-hnTX`}2B=
z9u~5B!osYpBgoa-r#q#(=$`b-<JdX)y4}&dYowFdcr>U81A-amcM4QvEOtZDEFQW{
zPQ`p(A9|?aVf3V_p3#4bi(sypZymv5IWYCSH`EWYRKnjte~g_MJK19i70oz<__%SZ
zh&HTuC^ZiEYHMvm(c>7`BX=h6Pm|yx8rzNTpf#b}Dg5GCY;nf68bXP#edetrN}Qy=
zl%~{*(8Bf$>m7-u-oL}$jBg<P*J0CtBmLs$_`l{k`AGZvpR{K1M<m1ElEMd_eDzPB
z!upRI+Pv1*?3j~rc1&t_#-v91H5IPca}#q3gC8{OzSNgRc%KHY*WtZ-KbRA4%kQm@
zIwg)C!Qws~Q8;>A=DUao<Kh^E^3Lpc;i5+x_m*=(Z`r?n90_8iB&ohctmtgN-(0oz
zS2Z~$D4jG#diZH-yuTzg)uz6!7i@bGSiMd1@xM>jiFNt9JE@+kM|@A_-`Dg#^%A&k
zRmnT;1TPl4FIu>;+-A~!E{gpMZ6$p<Ml>x1G^R%cQzdYyhVb8|ujP0IZsQ#&%F>$W
zj$X(iZT!I(aKHP@d_g(ajM*ZgFCG$ER7t19%eWxSM2PhHx!%Vy1DiUwWKEdN^^^+A
zryGi!lCloI7#Yg)zRY0t*EG=`_Rm1w>1`bYU!7xbnJnhXT{Xtzy*k3DWnb;XNfo&a
z7i{p_*<OE}&L}-1U;`82R=1T&SwCMf9jW!>-FBfCT2RC@lKuhd4SG*sIxeXxjk}qq
z*cokIuR0PFr;2c2Gh>m{p}RC9sRb5~LZgt$`yWXko7V+7ZJ&wNBRyUvQrUuu&cvPE
z$`YNYe3+j(O9IFtnv2q)>x^pkjkM#;>3Bz%`zQ*BcWdeQGODxa2-Z3=D`}uraZ#UJ
zrU`j+ooYiE2IS_8S8R-tfP=H27q$1buW=H+Q&?sY+TD;jc%kQgRX6NAjx@FTZ=im}
zem$;%dFYj3k|KiQ+)xODyuo*Y4gO!s$SGH02`u-MN_ji-I{!HCCWzNHM9s8+nz9W2
z23zaejf!WR?z0}!dbY8i?S(t)3&cdFYPhSRmdvf<bBK?*Sut^YUe!p@0zP$4kE7Pm
zt4Sa7d1w5UDC!-HX%?mDX4`J9I<{_C7B%N9Fc(soH)w%0&Urgbc+?_8wv6pFqfQ9&
zVp$t9j&ly3+Gra)U6Hiz5nrGtS~GQs!qhPkzd;#NV1DOK6lFvbjM<R(UiW1Dw#D7{
zuQ0=Atwt$&k(2MUH<7xQ35*RkcX<F@=?QN?oH=lBh_TPa&q~KyS5W*2fI3b1WlckL
ziGC)VSrY^LczmK!xm3$TI$Du!jAC}*HZ+w%g7WXOqa>jO1ZB~N_Q!Mt{30|tI&?ic
z0tat%VP`Al{)HtxqwRMHZ~R3R{o#1h#qGVq4;44YvjwYdnwmkboutpQN?rWA+3RV^
z_XTjM-U8`o=~a3}OQ{Gag}E-J*}nZagCsvTtgmKO>&`R1o3JF{Hn51qL~qgF%HH5L
zgD!_E-bYXUH@mr9#ugp=XNey8jYJ!-)=O|OXtBysjW<KXSh#AE7ZqHFq#J4@Pz%KK
zMZUQOY*SWX__X0FuBW|MDAD77$R&8vq*6>MKJNWcu7{VfcT#8#87P{3f$4lYx7MB!
z{hQsmrAzI;{LylEBiVBWpA_<UG6D18XFPo_<Fc;@XAI3X!OGp}obLfTPV{*A_4=gY
z^dSOT>TouMvTXGCatcKjz2>#(>{bVy<49E57JEPeBv9$r?A32Ge*J>r#S#&oQLoQD
z2||@)M}+Wr`fL`S>D4{igiS-2gvMV!jp%@6%wc;fH6bf&!2&f3x*4_`_NdKbLj45<
z^lSF-UZ_S@tSgQRqpuYrf_1odyu%723+{PoGTmfg#rji#n^E&le~o?kZ-iW;A|Wd&
z1}*JLFt(!23&6g+Ok+u2zEESM)}FeHo7Dff4#GjF@ejs+kVxgFam&XnHi`i8$J){N
z&ga~9<q~iHl=tZ^=%#T);A321@PT=}2iSYu&70t^t-L3YtHJ51_uDjUr$2BDyQ3fw
z+B29WhxJYd5_xEke_czV_MxuqYO;Ni$xfS4pt8>LIqw}x;j5SHf9uw{{)?6^2S3OE
zHCsi?|J$u!$hTmG@I`sJ$<ryUE3Y-zs(K2OW`{6$U>`^S$W<+S??#6CfsmbyHQF5s
zg$N0J9%2ps4u*mko9Z!}$?v7nhecvg0Mr}9E5n5-g|;K4SN#>H=D@V=LC1{f%@26E
z9bC<o*|{o1>+SLtGgJRINM`nWc>=jT_qw`{Sa#l;WlKNL?>zv?ry+-ieJ*9uP)sx6
z9{eIFj$X<>0Br{ir8S^azcH}ZcHq=w&VlBc>;x%iXzV3)9AU1`-V|M@Yk$?WF>+kR
zDh|PaPy#++$>mBT4TnAvJJCX+K_%%^RLlA`?qU-~BSzEQw<Y48t8x|B6Ivaf`yPH?
zw-qYSt|H1?N=p-E(tS!XT~z~b=1N<%i-g5qB1HL6E9=m8b_n*i^T*BV!y#i2Hf`!=
zl<`BCAkcpm5|q6T8{&Ui5*-T*VBNTZc5sQV$1s<_r^4lDn(wM;><&kScINI_qb^M5
z|KqiGl1Nyf(0aDm?^k|Jdf}<6{cGbFj!aYcyijMIF%gI=&1E>T@DI%I_N|iv{fH6H
zuE~=)64b}Bh;D79q1`dM*Tf82l<~+XKTh}`TilCwL1~dGzkwxxW;ru;9{x`E7j!wI
z`DE<U4-AZ6OfmWwo@SCqTRksY!=-pSWmo27CRj*Y7kG%urKcqI@$V{n?5iWAQvDS4
zfKh-AWMj~zZbYadLd4yo6>{k?Jdc(6jFw_~APV7Rf<s1|vB%xlu`^pHLEUR5V~8_!
zJ*9lMKtqI%%;jX?XUAY7+eoWTK@^Cf#BgDpYi&a<E<DIi_jQB{8p<rqy7q-y^zB|y
z5w6ee6OS<<gyYAvK<aEx^95{UA2_XtBgWR)Q$kmoLM315m^x1!`dak@TLKJIRvVD{
zS@Y(nY>2?&1i54eu8wpy==EPRtJL*WwvzPQmV(*7S{BMQUQ5VN`;()_V9+lrED)wt
z)mL}`<l=iz9;?_D7J%ze29pv3x_^6LchguW9Ce$OD4pDT<*GKoCdo%$Fr+#*Q91jo
z1%k7@fcESOb!~E-C<(<x&JEr}Zbv!5xi@DB7QpP#ubZ=rYFGNZ4pTr?Oxn6J?wwsy
zKh$UEC}!QYUsAubvC3<5C~J<WrzVOUoi|_{5acY0_eQ|ft`SqL42*MsT4@j`3~4+q
z<({oGX!P3iRUK2wBUe}r%i^~XhB1zD@g>T}=!c~D%MvS|!YK(E*m$P;Ua1(R*#pmN
zzgrc|1p4gPek2f^HJOH?SXiolrkBJ<n?IfEvYGdhAXXpK%1tsa`+f^)8!h$-Bcj%a
zJ7<WD|J;<*#b;xZbZm!%G$k)N0*^dYJ0rYbFQwx{n*Y9g<!)+f7E*=3kaMi`{9sMC
zy%I5Qvn{YFhJvimj35(xu>i%oz!5t62mbeT!RHZGDdlH!-!n(wUsHT(quc%!*{46>
zJRIU7(8})YL88e=&J4l)_!XZ<nCoxl1*~!;yAbCif+gIpZI`46lJnDRTHr+^`U_6g
zl$j;Yr&j}9L?t*|uVTC-2oDQp`8GqOel)D`nd|=1x^vKquxaaA^66?K-ubCApa1ud
z&IfojXy~zjR#8=#)ssG#qLdFFNa<Q0s_*e*<arU35XFVwYO$z^VM9K0eIeaLME#<T
z<EX*ukp%?Pq+IK|vxyhriNrQ&o23>%h1GU(E>aLL-y^63S}`)&{!CgXFDKog*?VvG
zOe}%xJ_wN;#<h<gOR}^O!}_$!v;u{51kFhH7tAT}qqQNOKQxX75WHMx4W=@*w&q(c
z@4J6${lWTK<`?IPSAaG1etJCkkU0s1!HV9jBH9vEF!4JW+;YabHJKuM#JpcEsECZr
z#)5y1BArN#NS`dS{`Snn<B~AP)QI=_iLfFB_9uF8t&;}{uXyKT#c723{2j@hhjQ9Y
zG`X#`pg_bHug1zGg2fioPse;^G|qo+=Goe+XmiX%%=+vL$I<z3;Of(qWoN-JsdK^R
z`p7yR72zsO<Kg{V{kWf~i3MXt53>==VX{jD`<tw&pTLQ))VLSJ6v=m#Dq(D8wNenk
zfka%&4MK#@O%I5aAznVHrZ|BfQS(db3x{F{odx!2exwr)QEl=V2&}imuree~^+*;>
z#)nF<8jebRQ9m3>h7v_vYo#4ad2;r<XW0&hq^tYBo{vi77sI*gmWCy?T&mj9=8dD~
zXqvd$Ec<0Eu|sDkOKAgXk;qyont?(T_N~Be8!@@HA+An$&ZtGo(p~NM(M^#UZk_#>
z?m@x80=RL^V#l#Q5|mL14wT2=FRPp@9vQ38#00dDCRg47&!B@~s1fFx>)ux)%nS25
zOHeJS{imCeuEEE&Kv~~wSy4fx%N+Cc(i+}`w$mX&iLH#HSJ-9t<LwICJ`q~`uix7L
z!VJavA3lx$`602%+I>VG-N@NwK^7hWX@HfpH2_Ie6v^Gg4PfDj<g->oI2ph9>w_$2
z$LNoE&LH1UST}k0Eb|a@gk;sb`ke2es`ZJLWUKMh<6gBS8O>tmb&*cL%m!&(@UU;y
z@h(6DFX&C2t~bd4>EhD8A9o6UCe>J2ge}x`tc$#oJ*I4Ww=5QCe@9F$C`t}jjD;Xg
z`1w*Iv${6><e`2)8kqC+`0WF-uBcjv%eXr&U0WR_6H{uSmoI8>rN%E(hT`lnkI$W#
zJ`#8O5Lj24lMl1)kb5p}s@>EJ=&lpoVDdJJAp*O#Rq_6M+WVfI#OLGPm4)&)NoguF
zf*cu-TYSa#-1@bDG|<QZk$KjCs4-O<{~Po9dTo$CQf1Ro$wBwic7^@ZCQ)sk;AC2W
zq$jy~d!2U+&Op!NG(ur(>ZwtXQ^rAUUA)Tba$TC(ilFH7BCvg{_0~6CpdV2OmUB5{
zQ?IHj#70a@rt?E)7C~J@qUq|V{oTRx34h4nRNRY6QWlA0gpN-%A%)qd;1kfdu|drc
z+OiBdmvZAEk2Z(dUmI<|uvG>LJ`r)C-2s2d9%|%m7C8!fbA83F6P=y-kb$wH@4eq{
z_2TQppN_=&F_zy-I$Q}k;Suc#+210*st&mGxwNsLt>0&x6Rll;y^_3*#-G4VCWO%o
z#w3pC;rxkTe&sP#CiRuMKla-bS|9Pk%*4h_r3g>X)kC<m34p@S{4d64BnQtOm0D)i
z#xZk|@r;nEdAfsy55)r45?(6)FNx{1u2+Hhr>hkqEKfImo^5ND{qVlt)Tu{ahvLLS
zc^<&DLry$p1U<F|^L=`}u@tr7(rkBkDdndF9ll2<?j)IfO5Z#iHcVhD2CDOTk=&2d
zxJLY6c5A@)iX~6>l7ZEW)93x{{rZ$>?+7MV`kjd%5&y!LezBV~VMWz!zn4Ikg`BU)
z1elc1@{EJY0_<p$W#ky=xVn%|M4wtnO>|}yRB7g`gkt*u%XxSB&8UA_{ETIh>i0$9
z(kWTE+(;9wBR4MT_Ys2;<>RrMj~qL%%@^CGJVT#teL44L2q{z8v6VkRzJ0&8u7aWa
zR!ZMrbk9&XHXtaS&S!l5WYv8p(%keVqEK@C3cQzbV$?1Msg$3v8>STO?$XX1<z_C?
zUNaqaqa{FyM4furj~3L%B4I?Z0emN_Lc){0xd!oyuv&#U+#{eQrwMJixQZ|vNX?-$
zy%<hSLBU_bXt2sfm?`;~N=QN+6}!6N(yqf2qxaWrHr)S^md=u{uH`OeI^fu^0e(A%
zF>QrFO1PLNKZQsG{yHfyFbZXeO6hGsRjz*C*VUK%%qK+8RJmL!ajOM$Nv(h&iBKGZ
z7#Hcf6BS3I{8=b$dd4Cz*QNTw9Hx)e0TJhmU03_S(}g-Dl0Iz^?Ij+fR)~yafb$SB
z?_5tdgOIBx$qbLWmU?y#Wqr$HdEHszd0D*?d}3Uh@+q1?=`WR%53>4$(6qGwm^8(T
za;mDkrpsYkAfK8YVufs3S;tN=#RR*zIHqCwaI%DQqCUCVI1suiNnNQdl1!@a8fW|`
zqdA{W*!O|B%&n={n~@?_9?dt(c2nm`*f5wI6rgUSe)e#p5S67U&Tq*b$1jJY6cP?6
zgUi%ulC|yGitY;hErI`}+gTcP5<d}XLZF5z5dQ^7%#=%GS?ybf9yqI~@l~~MJ^Me*
z-nMQ~A3zJgJVQrYM9OcrZto&3Z5dmP<jG>rg^ov=_prZd-?b_qtl999!ayzKg)t@M
zMZmfvMq7C|G(^oxrU`-bQdcPjMqr^N&C^=q%zSE=L@;wpV&x`Vlp^|B<QhU9P}3sW
z{u$E*t<&M)Q>m$Sn{Z5s!zQ<!8mDmexZlVU{P-3@ABQqOMvf}nC0>4+XE|>nOv{T@
z6uECtX%yV64#S7?pvA!554yreptB#E0NDC;vRG^R8|zfGng4AHthg5b+*8|1v5>Rf
z;Ye<szox~<F}n7wr3XD$di}RyKF3QRcmCP}+^3GV6nAQjZ)L#!my)HFJOy1N)0HPI
zkLJY~1zoz?sME+3Y}H-a@k6EoEG3LS=hQA~ZsJM$rOs2D8d_}mU%&vaiNkWE(EgFv
zl7Bln_dN}Q*R};CN9~dEN48S{H;Mb#tcT_bN7rA1tDz|CCYDia`D^h>C;{yF*X;O_
zhg#kopfQZ)Sj>`u+lIOcT6=URd4YfP4$eFieY#fYlov7s2iy{vD~%LjFEXvf+{x6i
z!C3eNgG0$=$saZ#W$ENiXnt@a!59NeZL!?oJTx~V;F}HaM!?1h=P3A+bEuJ$PV*5x
zC)QN<8V+k$&Iz0=ddwk<Sm#OW%29Cf$k8oAAJfB~l1;9wpRi()_xjYnbqbvDPj?uw
z#th6)>giwG_`;v9bQM@cR6e_TBp!WCY^aqqFGTX?+tR@h5LOQ}oUn}!Q=LJhoEz(p
z#xZ?pmWB)CY^;4O6w?j#<!iE}OnG(T-hvAim4u3KMU}V0IB1`Z)@PiB%`zOTg3gWj
zOnsaolrbg0A9{IgWQ*?0OsArAVO8Oqj@c5bhqOgs;X+`X-I&MJ{qvFIK?FR{U$G9K
zh_vU?*$u`l%p@sad?gi{8$T>zF)cwN7Z3nDrrTjJr;mIO-9S0tvek<>EmQ%iCP4vu
z)v@*`wC>g@chhyYzG42kYYlXh2E$2Dw<AC7j+O-4{lDP*1KF~OIk)J3x8hMecTzHD
zI#si4NS)4m=NGgU*S_hbqHn5ZW9L>z`!IyRq7EIk^KY<}LsV(62mZpPNRHz1MCqyV
zw|@7+LOo|+5Tzdo9(7G|t%*?IBOS=uOC{H5Js76xh93BJwfyPfGe6`0KzBfIIVb$o
zlS%O*qrYIb1^N>weUYqq-~;rc^YtL{!Oxm3R6OK*DqGk^wmBXfe@OM*iIf9BOsDQB
z6nTckzMC#S&EBopOvzuDCXY3<FHq*K$)j&>xN<v(9el+^NP!7)IYlGkEZ2e9wJ<pc
zu$eg9F*dw+4GorOU|f#dYu2LaAKwe~Lv_rD_G6~I#sbB_hyuS4yW<n-;$4QnP_3{C
zVfSu*x%FP(p`N4(_hG`ryRX(konobEGMTK5t(V-Q@)1kOcKv!`s?`D9fcRm&B7v9A
zo16l+J3K8n`QZH$X^<5@-v`X~c3@G1ln*VsyqN-yh)=N73m={bH7g+tgPIGFYsX!4
zLTOwXw~d)EW{~m)zOf=47qZgw{5#b<HE2nUn6Sxz3$w#|CL;cwZZ&y1(6MSBWg@ju
zwF(^2#hQmzQ949$Ut5`S$v_N9Em2R76R?jnZW1rN3`X%<UJrA8J$Ocd=C@7@<>QuZ
z#NIchxin41`hOTi|4f&Q3UXu_BfX$m1DmZ%cukSZ@m|EAl(0&DcXF&=BI_U-iZ9(S
z#1*?o@;obPB%~5x&P-K7CpW5g$45elhh7kBklHT1pNm>eTd8~dv*1ECBLa%WM%;G&
zFR$+2*SG@25B{2i&k-lBu3#oK@LPIl#JTAOltl{JCRzCfyWhZTV}Ur_9kIe>zMo-9
zY+lZb!GKe&xp%cY?B{VagyBcDu;Qe3xKQkVwO@sZ7X|sEFJxXf=!pMqc+36Yh~fA@
zG9dmhhqrP1<H;Yyr`@x97h@HK{ZI%_J*y3LS=z~Nn48Jk(aGef#>Uvm*$vmX$Fh9#
z_6ligD@1Xored115%RLz+ua|6oL{*Ww4*r}44St-u)t8|CVHN%MB2Ee&VLE|bSH0h
zw+Y_K2_~&%%iixB4F4wNWh$`Sl}mn1$t0)!Zf*Cw{`2E^#M#+$)%WFlC5?0%CB;Vi
zapVuJv1TXnI5}_r+xyDM(^thDPTimC&gXhj%yk(qrsEHMMvG_L_8ulN$TAzJc2Y9a
zR&t}d1(iMa5L>_O_rG`i-D0Tzjj>{nahpqDK1DsXudlAPz;mm7)?GkuX;RJ<rf3<p
z(Pq@S^jp?_)jJ8QHyUlUXij7icdgkgD;2zG5Mklve(G=SUKd^Hi~pm`L-wpys;T5y
zsDt*abT5x+WXn;tWb^l-R_Aw{;(1$0*n9;BqtJptR4-Lk2&~zF@-nxH(A?im`Nl^Y
z1yJaZ87<@yQ1?OA$L!6`X#I=s-%V%IG)1--QthND^@-f$YbCyLcfB3V_+w*iyFf1x
zNxmz}8Hv-^NqP-@QB4(#)NJ!`?kx{<9<Tf;_?WAKk~lP4+dGrlQB$Fl;A-L4K@7++
zX4MBPizWPdA+s^2*K$hDPI1G@s1ZfFWc*HwQW9-8wwPCKn2WqJY427BC)}|@t%9KH
znQz;X=iH_m@N%*{S!TtsMab}o%V5Ln)SgeK(##EYl89!=7_Pl7--&<ytMMVS>Mx4V
zCJpu76JffrBK?cEh#@V->OAPnqKnjks%SC22r2!Lsix*%4SQ8BEGj7+4SNB>lly`;
z0rm{j63|RPiZ<I%J$6!=zev5vmdjHq+R2+F(0p|v_u$uSB_{aw`H-;OD9O*RzG>9O
zA%C9B!prZ(%$D`~eD64}rc)1h{>rW8*c5Vm>)!lz(`MMSLdGz8t4W4^;s<=3N<bq5
z;V$i53C=zYqle+Z=a0Bs_l;Qm%qgq8bZt#Q;jo_4X|5waIRZhTaJ`gEyy2b0mkGs9
zw7nC3)K3!o*l8<N{$;PrXAA0!jhr;ycqfN$1hwVPGp=z;y$Lu@6N=^VECLxA)v-vy
z#mo})NF%xcidc;!?{2~MwYT%9U+X0;zvubpKt<IvY3fgsNCY;*x1@m*O&sO=-tg?V
z4;yodlv=Dch3Wl?_|Rebp<C>>yad=(EhHNcY1JKKn^nJ1FXOOCugr~xJK4ZMTt0o;
zif1Ik`SyjR^W$<7NX)@2X{BCUZdpqrfjjhon~UNc1(W!{-^=IBX1hGhF6xg%P=8EH
z!_yU0?c7+2s};6>E6OPr*QDN~Y4c_E3Cu>%CABGLVT3JNEV#P4v(6hoqE@JjC+ezu
z%!hbY<SY|dkRRi&_fUZPnnWMcXD=_ScfCP%EHduUza+6f712%kX!m1Jr>w+v`ChRO
zv#<IN1qb|Xb)a8v#Qr+*W0YrfjTe_G`Gd2Q=9n_qJY5*R$;9Ydd--bQ=d7*W6+`0p
zhi}M(;++Pfxe=tuKfQPR>stw!#1}W?Ry6x`RWL7-1Pu3l(JBgmMmHe&iSqnzui$^i
z$S|87F$5YKyEVkfOuKXcRc4vo_L?4P!Aek1ZJjHw>HXZntMEzNL(6o`zsMgN<7ZYc
z_hM)s^8DVYHF_uzZgK|!Ju+jJ#Oem?904d@e%8h0B^7+e5H)HunoQ@c$Ho2dkTv-h
z*6q%MG(7e5)P;$hH9`W~b_4bM`k(Ye5k>JBTw7u9{@<(D(NMksC>d#R$6a5eWzied
zL$a;tln-dY8z&L9%oXXxy2y?;jG_IH&TZe6)4o>`CI`P+_mldD#mqJbdw~#!J0JEh
zab`Jn8_KN3zpdXCcfAz|>TC~wJqRLFl9iM3)48kP98P|I;xrN-!j{DO)^Vf#1b+|k
z6n?+S3Y)Yu61RDB8VFn35|*o~n+@zN!$luYrK0tLONh8&)cEdREBQxPf~e;iC>KLi
zUxH1^RCY#XhcbTkb0BlynS9jX-yY5gv8lP=q{61_$74p!F)J+ACDOw(-mz0%m=)JN
z#xKGFvYJg!ndQMB<qm%Y%47t`wEr>ZAz$G=4OuT~O~EnY+GBj<g4O#?nT}??@P%}w
z^ZU`EN=#&!1`aFbmCw$iW$hO1*V_<Xm@4K4*d%kavP{G4h}152cnwviFS|R8%9;WZ
zhzglvk^2OP69n!>=q^k$g1=Xgsg7i=QIgdjD5&ES;6dM>atW37iDPFM@qN<;t#id8
zub2Vw63%*YiwFIV%d@vkk*BPooF-WZ+rYCdT!jbzJjK5(Ye~_XDqVZv_ltOxuMrHW
zr<6Fg1YxS5)q`yt78>?g-)Mt(+KgDyo8CKeUi5J!X${mb*7058HmT8UwzgjQAAj{c
zqcx*K68_tVmh1n!f>Hac3xJ$WU(M1UVC8|trsZkr@jo@P&Tft$kkZHYe>^yl*wnRU
z^jSYJCV;AiBY>QqmCer4<^yxG@pJ>2NlUA+x>y4l{?|+0!Umw~>1g?ZM*W`?6`X+1
z(souJ<Q#n5|4C}pL}HVaboL=P`cK{g7Z*G0hrBz-$BvJi%LIu{6X5Rb>1G9RCl?k*
zVv}+PIlF1OSXcr6$8@m)y13a`k{hw9TX=jRUQTS%053Z$fTo<J$p_%|e`^2Z`(erW
zc*tpad>~@~Ys)%2d3+Q<U@|U_|J=-f)+K2E$727mKJp#@*W&t*^`D0N&oR!Ad*kEz
z5B#P{{(-2msk=E_X#qUQjmX*5rDe(4v;jUI<o~%g65IbWw)1j)&>{Z+Lvq#KD)Rt=
zcyqowod>vyxOZ~h#JzMlLvb|4@2pvAv}-*s;=$)Ja_9d@Lk&vh9uP&7Gcxc#%cIlx
zzU}Ef-qr;%y~hOUJ~zEo_WC}GfS(rE&s(1A-row|@639ur+Ndw_6ITbcAh<g-=ASy
z-(TCvf|j;*rC#2Tdf)HgrzGD`AE(U%$wVIRZ~YR?-c4`4?G(}8@NkaKUu@n+x9>Va
zm>znsf)2cO98LT3^PwBW-fv5+9eXD?I=WWhUoP)2JE!e+-ak%VT0Un~YLb7um5b5|
z<RiH|*Vuo1mCF(dOjxtF$$Gvg1K6+j+OW+#d9ywkMC<Y2bxa5J&1$;8rniWyr3bwm
z05i7z5(J!zkONcg0dK57QO7mOE}FC#wp-GS7GIxdIh)>{0`Ap5j;CWjZkA2|W=;)y
zU%mEM>j-pUuzEQ-s7M%1EpR@!>7NWt?L3Fr^rO9KT=ldU(1=~gjn9IgmH5PNY*wE!
z-!<=~g3_nGA2W*a-k86d{WFUW{Fl&Ju>8~nL3?Tf^=$THcVF0yKS#f7-u%at@AW^P
z^p8Pr9apE(1@C$1A~dm!lOX#&f3o+zmki)k>B{VTHoao(tKZc8o92EaJ=Kz0R`)Bj
zyFu#XW3a&?r>taKcah*ZpVPv{y}cpz^E1O`$#Q|F*wZV+W7B;>vDouF!}B$RC?H65
zv<zZqy3F!?{Wbsoc2DL7ApB;0&PDmYr}h06{C<l1vTTNniU;=u-tMgNefNu*Ho<}<
zJXt}%I*aMnm*46>xnFID;T}`IdrWG8w_72*_HPR;ay3EgR~SM6Tj&4dSnMrdttM!@
zaq=+eMaHHjXuB2pIp{^_e@b<v?=OjREkWz8HL~v>o!XE0j*ZoeuL*5Li*EtzJ*scl
zv&8>Xx_H|=7pQt&Z!MX=QfqlC>FBjPW6pjD<-EUD+&CNF@jQ?@JuE%)o#x-Ze3Z0~
zx0|Qd-q6}|bVLb9N`zhvh97jLm1`b)17FwQ-MN*V|GmE^*j~OrERXWC=M^piJxRV7
z%^Gx8k(HzpmJXM@H=DJ*Ezu<3jU+PQ{PH|w7I<caZ?!n|6*t)|nxZ9iQPH}Kt1)Sk
ztpF|t+mCB5<suSTxu@p4c>9Tsgr{cXLUb|MY&V_ytIbNs2*BN;S=Q&}YqQ@RDSBZh
zM7t~26bIHCMuKbFkWHlc^h;62`*;`O>J4Ln;6K6LmK&d@Pr!)Y6(39|tB=<l6aU0H
zQYd$rWxiC$nu-%;!x>A>8u9+hX6fi_UtbFCtUal(%~B6n;kGJepTz-X9EzR!K|sK|
zLy<PQf2+~FM4zl2<9%1phAlRIxi5x?j$iwDE&n)1r}Yr0<J66=oY?!`IWOwy)-TDr
zFP&~DXVutfZThz8EiFf6rPW(<v98cJX{ySCAJ-cWgIOBJ(Lt>YpLBm3x9V{ao!;mE
z@mnFR8Vu(}r4HMqlj`0&;;GV^#(&$)P}FV9Sqb9P1<|KCX0db0(gam~%2^LroA}!3
zv~*Ic?Fde8aXhJd<=VDxd+!Rf{K-r(ez)}Mpk=4BRDI3v;NNE6UEYmnr8}zJ3OXT7
z`I+5Ive|>9-@-<bg-bN+Xy(kenjsL=2ce=jC<rhDDwKpNPHwYS<nf08bJX)SNhP}L
zXcE=P@fzvq?tD4TpDsNHiqjNGsemqxn#9-|m*{mD1@}NNtli5<;3tbI@M|Xoqt5lE
zmNy9-Sb|x;${xMAI;v>tu4Lh9w)8k_+0-OAi;`uHqyIa7>aaepG!un5Tdd_#ke=?R
ze`b`<2^Y=No0n$~dIQGUY8!ylGekDh&&rY#2N;*&&y4kHLJI{95Pik9tM4(IcGKko
zUY|R~DDhVD7Iktge3b;LwU$qnVx+=s_AAGK(NDuTCB&jFCl|tMsfE?H%#AJA>@_%7
z=;-Eq1yp6GW-aE-t<4zPiDX5w@ap`uKIKKnnpd2hNRLwIpiSQ0&Qwshx%<bI`RW(3
z`nh|Sn%4z+$yfR0(a%WDoK{q_Pua)`LmRoM@w%-eGjo!*qk3TnFuUSUv~oC0IPO(<
z(|%6qK2vR7)iDSmW6@_)W_d0XIRiL6AgmPHxBkwFGPSyU{3Fe)wi;6VsRUBb0XJx%
zgm`Jw&6YBuoN3R?ehMzhj4J|i_|?8n7Sx$ohb$PFL=CNRTAl|TrZ}aTz&L;hm7NAa
zKWK<_QgV=&v_c8ir<DEN3FFp6IU2^*R*h24f!Dg(CN#?cD+e_EN#NU3hZDH6fw78q
z-O$!@C1(2RX--uD>-Mtol&W#InE>oJ3$n@_s;=kQBf`CtBkDc<Y4!u<QW}guzI0RM
zKwINWp3*|A_vmkJa}H|%m&$eg)z6YE0-9cLpHLHcev?t&|3{i?F>TFCSfwv(#hPr>
z`as#*GNs1f$kU<FUNc2g*I~-x2jfZ9hu||XCaTFd$X?}fq+KB^i^b#JI@X}&QA-+g
zeV4XSF`cVLHg7s>aaw>#%%qfUUoC^7H<wYFsPv_R;gr{Uq+?{+NlC9Q3;}4UmyRVW
zo-#@+hIQJ`kBZOReVxHukfw#{w<zFJYQZa8F-v$gv1I1wYjZcnloh2SX70LqTW=rW
z&YKAr!DA*Jn^nm&4ze6PnYdj`jXZnKy;B5D&^FcOTd)x_emkb-eScse)Wzi*9EfOY
zZt!cI-fwQ1zqw<!85nXj<Z)z}TrIXvnBlBi6?(SxmzpdMBIpO960ER?JLygFW7wq0
zoUoJ7X2q?#lXxm330*fcS|#L<S)&N0-o27JDymZxjcKnj0|dLyMKYoINjQK5&iqvk
zeS}SD7ZAZSb+E-<eGj}KcaR{YZSu6frOA1z43?;2o+z7^f7KKJ+Fu}C*TRj^zQObZ
zpuSSd5fqwSE&H1z6)=GTuJ5ko^0Pz)#-Yk0t}~PAnv8zDF=Bq8aQ&UOntLDiCi!eT
zE%q%h&eD{M@cH--;24~)-GJ%3hG*RBKmNwFE;_>;8x|WvLUtl~d^5jYs&Mn1iq!zm
z$qrDWvI)=T642JmD*G??3SpbWf@nP|*!|Es191jX>PV%udaG1yeQJC=+DL6A{Lsn~
z`~uZtr4|Ch{3fk&`ladT?;$ZmKYuX$0&{Mbq(^@1qZY3|i;VzATMFJmwKE4DA(g{!
zuf@!r^_?bRugM**A7_uA)SV-+K~!i~%!-7Q^C6Xx>zcWg{8G+*s)=a9@F+qm;RKDd
zVH`i}P^?KNCv18(Q>b9#?fRzRs_)z9;h#lK_PT8<pm`%@LIvRY%qo7enmNqSTek%r
zX}1K3F;hn-A*V)!DJxPeAO-Vw)fn7}{{zSsNUc>$tNT~u24&TI&&tlhXqL8v=uhk`
zf7NGY<$rx3AjG`nxvJvM#5nZpcL71gVmmXk;b=QBSLXdxX&&&SJyc{XGB4C<KSeG3
zOH9+P&~hK(X;r2&w*{2!fb#bvGuuI0V=I|>j;iwP&CkCHhxd8iV1eoSsJbe)R}0;f
zzs)TsZzNVWn6gld$KNKL6#VN(K<E_-)89y@uvLtbAvyE@+(w8o0Z8nKEZD8#H9!6R
z3f{6Vtd0gUZXw$n!1g}Sh7Eh^7c;jnsSGf8-x!#i(Gvrtxa-|^Gj;ndgY^+gP?|Fm
zO2=m)VI6Kp90`*CYJ9N-Gz*IfuwRVuUvT0mdy<b+27?@y4i$lt3j~Jx>`iNF7_Erf
z*KM@EM2;(z9v;3cl04k4hPXs=-=RP6m9M5*_@?PC>yEQFk#o54@))Hzy-%<v2IR@f
zz#tO2hZ|oJ)3(lCYGM5}P*x6?gvbEpKv6($KmCA^CDT`%g9WZ<{C{GGR}28!;Bn=V
zY0+(;>yhTKfIHbe=&53MhT&hU1#%^Pz)}(+2VNov`9P~BcN?10Plu|`FU?M<0yb09
zrCi5MLG~x(zzs%2<$8mN1OU(LDJNEraCrAt_HEW3vuUmFv{y#)_==z%Cf_={`}b2t
zECikFw$A)^JFS09HeOMjzPu^UovriSHlQ1t1sS~u(5%{Ya26zdm3kU>0EDAWYqY|{
z+bn+f_{~cM*g`>Mcz?iQfT=Z;Nm;gX2FVsXg}@HD`SoSi++#&RABfO$H5~WyU{U#5
zE^6OpV_2-7akq~<*ECxPI8@#{PtGyh_{#3It%vV4nh2h;x9Zn{ymLBY3lb5_GQpC5
zC$}xeQL!wL=#a0jt>mF3*@Ag##wsi06~eV5DqSPf9_8HqxNd0S_Cso6!yK#jvBC9&
z#AE5sw<nP6sS}XD55hB}Wk_~|WpYY}1=?ncir@9y0_Lo_Z3B*zZ_yb#2e6v^u&jXu
z4rS^yok#<q_@HSGN$;i2;k5c3^7ZM<Qx}LNgX)gCNfcm22z2KJYx!}+Hq_v8Oi!5-
z(7IZ9hTX9H4dbxitpAC`$@#(QqvbBa;TJ~2v64Oqh5ftVO8I78-1ZPQ&UqhAWz18K
zW(g?eaIOEb>Jc&mT9<R*!u$4GOdMVW2W!E#p_Ik0NULvAPiQ2~&T#gPQ>mI3Sm9wx
zLJ#50R8SS<Y~GBgCxo!IA;pX38ZefBCpRc_W?$|_N+=$m0<8}kfJz2#rlymKSUZ0a
z#l0_?%B6#u)!0+Qm<$q^8tF=&|46?8ii$*KASu0$syz-iw~_JFO!sXzlDl5ZAXKuc
z+Ghy9BW-t0;EdroZYagKQwD>X^q#X`9->Y-E`!|?yUqO7>Isji#fH^93Jw8};IOn#
z6UNS=6Xg_h-Mmi5I|Erxtw$5fEGOR{!OTJdyy_#aD!!PVq;gO#jb@5!AuhOsLo3tI
z6W4H87H{S4TUMpf2O)MI8llMV-2hAPliz~kCuSCG*UCe2NC3`FgQ<?`Zy!kx6F$!t
zgr<5n$6-+ewo2%r$p*VthC*8YwkFaVLazQ}<$?4S`T@@;XBuXhm0Ho<Kyz1hB9K2G
zb6+N;bSGmE1hg+bRIA~*EU<KiC+abou-@Xy;ZbVkv)TK%K!i0q<zmy2*4z*WN*^5A
zGH}Wgfb^1Fzv)8QB5v2iK?Tis1G1Fhu9&KDJ9On3zZl+}=*MQ>VSnYRL{YSG9*4<1
zuK2$pj-j>|&ld=naj?@FNHNokahKh97F2MsgC>=27O)QNTHYfT02`tJD{^e!&sOnS
z{kuZM?Cq!7FK$JPC((~~AOAB$NHoa1r6z(z198oFq^UJM!C|runMDiPfwL*hKfKk@
z-YaX`>%HbUa23F_c*!0Q6_WLGQ^Z&V=fq|0KkQVbj$hx}HH~RwNPY>JpKEYH;qe<R
zt0X9~p0Yj?Wsb>$=7Nd4J*Y*Ql#MtZztVBazAz4|nSn&M^cb3em}JlAsSM}%_bBGu
zmOf8qS<-yLi*q!33#nnYai)bczXvUUUgTGvbvgwec};w^K(vfbeIT9bap_vf92mI~
z{>CCIqMmb9^Jis`*V^}$Krz&wQ-G|cCWmk=ko>6ZSPC??f0%XT3g--(2?%doteyt7
zP}LgtikZ;zqUDI>pr_;<m05W8X>T4u3-PY<jvq_?k)q=*-GY4&wP|(eCc6At5LeNC
zVVgL(5+gEIdXLdrn*is&a5a`eIz?(1Mw~p^3Z~$_EoK5X#vYzjMzv&5$Gg!gR0kq2
z%*lQ6q)#MJ?a8F{0e<v{GrDZ6`-GX2^`2xB${<qQQY=uM7j=LTF67V}6ncZUPSzR$
zHZIVyY@l)N8Qk3C;`95dtf#mBg$uE@)PYh(itcZ{Gl>O`FR5)^;M~jvh{ko*8HP_A
zze#T*It<HEcm*><r#rpcF&77(#2Cv_z@dVLdNo17O9c60wVT~K`Kp{&WeGsb6##wi
z3f0bG-b&i4%WHAH2nZr$kmoS}^Rchmmnuh85vUyRnF>9$;CP?b<;eQlT(H6$eEeA3
z**Q!M)QzwEX%bD%MR4lgq=XVoQDMXH9d~R$qFCpv5321?HQ`D9wX(mt?%jx%lCCl}
zK_IATk!j5S%!sCB2RKx`$8%`)mM4+QRr|ghyp0aj5t%qv2HsUC9ErOt(pr(yw!!kW
zx0_I%iNMeHpq-B)On%LI4Os|hTA=8^RL*t!o;gD@Tey+sqBY@hoj3)oyz#uTCTJ7<
z2;w~TUb8SFl`941fH+E~j4O6aQ!Cblz+9hZI&BVz#=Odo$p*9jK**j7a>RIUgPJT4
zy^h6<h!crYPO$7-{WtMB4A(e<fL_kto?(l+9a#w$3oA7h7dERa?4nJlBE(px)24Pd
zUaGB?DUMI^uU=bK6n?{_nV14ZwG)&&0nqiP4(VLCBXoU#ytpk7zE^?SMGW(vv~Oo>
zD5-KtXa(~cz7v7c51INJ0Te+FCYz&`sGA0l=bJoBi}#KhDo^#1M(KM#%d03y+yFd_
zU(ofs&-FYnOh%Y^$f+h+dclsDg3s}VA{)hSi^$fvlNu^2*ad)r#DJj%zR!}e-{ZG@
zNE}jM8xAU?>f!T2Y#9dpE!F0<&i=6Vb^GZ}EljOUhc(Wb7gLULgt2zb{`V#009w&J
z*T1-+tOt+QNyp`80yAi3Tv7u>qjfxOSE7@8$H{Nbcd5CeD;&PO!S?*V3|_GvA}x*;
z!Uy$M+Owr|kzR2MmonxF{O%mO&l+_1!Va+VF?BFIPJkaIr$I__4E-E3w>XOs#&$^g
zp^?&?%imu|@&}yvfmTY){{68kFJt~-5Z%Dxm8BhZ-w!`zLL$(FfYyEckFUNJfwBim
zIjXCV>gxf9{!MCEZ6_nNw9VBZ$<4ixNBo=+`#YpJ<emb04gsNv;b}e&Ee<~205;+_
z`PD}wE%<=@-sTnSu&gC?q8t{3^T2=VBQv-io{L5ZHuwki+Wy<x4$0$L1j@UGq_n$2
zudw%v=2cugkHMA&_@(RK@6-Mt({z{Jzkc1f%&QaQ*LQ=d*^E%ktwWBV&#fw)tLv-g
zW(kpFP>nt)s!Skw*j#tTM+VWcMNZFwjh#wx>o+)jztQhbY~Fg_oh#oA3gd3wytX)Q
zS+6(oTW6JGaaa)-(X<E;l^dszO>R+32tnC9sL>`OP8>L5IqY2SHbr3)|NNerk}~%_
zez<`#Lv20vI{mBT5dX@-|2#}A?_{a3(45`<Y&56_rF=62&Uzqy_<$&LkcV{HZNoSK
z)lU_IqlB)Cz$p}VG38?+UAh;W!@KJ=O=sjh5(IrmJc@x=e;RHXt4~=7O)_Gu9vs7~
z%ET#s=?@msc3z`*Jgu4U>`^INnJV-)LP9oHr(!p0cjS7M{7|!M0lTA&vF$83ZGl)!
zN4Ax1>u8&SyWDd4plHOBBWwE+yEMyz6}~UXi^SAN2eT##Zk7De7Jr{8R%_yAEtp>R
zP<A$UZ9F|QtV`VD&&sZ;a`0*S^5VQ+W@8PGUfBWD?Cln%NEeh#(WsVI#1|Z@rIDLY
zKl1m)PumQzd;`-d3$|9*gkDh^4<((asH>XZzUtp<Rh%Hd5zsa1%}A|a;`@Hej|8RH
zCk}IEcF(UO)%>a$QpW32SQn-HA)yX@C2o;N2;$N^xhdJ+9I%0*g#cXOo{!;$%w|TO
z*bgS4kJ?=SI!*oI7JJ@oeX%)L&N~g9ae#H;Ome@gY7a+za}T_8aPXXd0;PEo6qQ~b
zxr3fF@hwj+R;!*{pM2biyHZw_y<36Re=|8mMJ_SA$1;>vdrNd;@v+K?lQS-~xt0j|
z{t56WpvtbWUK7_x=tVb(Mt9Ptx+rp57X)~QZY58doT%dru>8(s@i@yeH#6&sJyjK-
z;oh)~TfI4DPZ{|L0$(cM6Ge72WHODRs_=75zLfd7^A18?I1oVG_SRYN+kh1yZ@m?s
zCCZmp0aqF4EROBVb1yx#9MOjSU0Bf#zHfPhNuZ{r*o^V$I`{%N??~IG4cr%ZXV6M(
zjIvi{rv)S%)Vvo~RAXHN<UqMcP%sOyqV1F!SdjXjHDFj?dk@JD*cZ(XfIKT{No6$Q
zu<F-Om;HIU9xL0X70R<?%mX%He_6PMkga|`YL1;BQ^F}FK~F5YdKs*K_SzYd9(q5p
zg{>F-(;Wnw$gLw<Vbqg#$&YH;EIJ0qUajoZAwxFl65|@TfYrTc08WQ$@;;8&(`<2J
zOTxpDw7m1%$4l&<hVbtvSsi_0c(j2M4Gc$9L*;n;iQGl2*=IT}iu8=)el2H6)-cA-
z3RAk?VvXLOWWRA?3Md^Sw@ev}p?(cSRpa)8cLUl4)O{s!VYfo<ya?E!UL(1@dg(Rv
zB)Q-cf^?9(*NbKXb881|p?8rEP4&2(E}?21?~yn;+MMzRd*o%ur*=d!cZYV~O!uzA
z6=%!<aQ#MY-PivIgFt-0X0E-lal7sd8op2a{uWile>|JsKJG5=x3-sjo@SeHbtknQ
zKCeKX`*^OY<iv8J=Ryt9tQU&<0CO}y6PmH&9#=Z*5%vz1y_Ms7b$@QQqfbngdfTzJ
zxbr-^bI=;Mj}~5orPTa*_&$m{tSjF*Q@pxgW69%gho-IJCRj#(68N~`nZzZPI@*P{
zc*59*OeW)0sIlh*Gbw@hU@bpd=9~rhOb)B=k6nqtXDp6Joo%*zv0&<or>1MjVuLFd
zS70?^>+PhUy+f0>?i#5pzmv|Xjr%w@YrS8HYb&tDCK<4g={HcDJyuI-=Gt;k&G~73
z*dpRAw_<@PK>q73`hk176K)S74znrmo|(PwPcu@)-EcX%m=`Wnc7S8=`#c*TWZ)Y0
zKKb?$vRORtBbp`ZF1DS!LkA9SBDn+X2jB~Uuj}lWnY5uRHi=fh%p|Y4pKRR10k)y;
zifav|sKaf#xc{gvIiH_*DaRYPRJ|q5pjgL6SNg4UXz{T%IV<<)(`2vT4IaB*X!3q~
z$<=dK4eoomhcoB6x1>o*_sX@m)<0024=(OlD<RttJo{>&Ds3HdFXuH%ej|9=cX!M+
zoxQ{ZO|B&^FNsg)g}BagN_XX~8Ag*^d%X2_%Qe1AoTS4;(V3jNdv+HBy1wil*7{+}
zI>(F|<93J5Q-`8@;jG+g;6~QWN1XVpw&l*<u`drayL0k$E_mD?EnPmod&K@akXxe;
zr>t9Yb9D$Pz1;^l3tpysn{T;YvgOGgpUwL0w{>;rP<K_&wUx^(n7Q5ki<6o1sK?K}
z*2_d)TCwe`30rqhZ<FdxE>pJVx?hXh3HxF}9^9}P_t)b@4k}G=EBFX<QNT1Axn0Cc
z(c`k8bIq)yf~jv%X1=q<wI#o2VB&M$wxyvf;}I%tTxhbFn*lXlvVryn#<@bU?^u|g
zCG@i9W~f7KJimXSxeRhY3kMUv>zOl4xV&_N8kL%#nXt8@9u0~63khv7numbuE!Pr3
zFP+sdMC^POTs|9qg`=_cbKypr;yQ(Xem)|~lfB51j7zP4Wuul)-_6=nzo3EVLPETs
zV58OhJZ$2g+a0oRGlC-zFr8O`+c)yL8lvq3P5c6JrMZpUa_+Y|2B*dsZZma&Jw@JT
z3$1&>{LD7@nY@!&iTlFEiT-exiVIugJRWJN<pZrCBV^+vH<}Z+tzVn*6R3d0R26=1
z?>0w+JDRp;0$1JZufd@F*3JAt^>W(VMQeJ6ChYVwsXk_=RjwN>jQcSRxe3dDp?%5K
z;aued^~~SpN?Jhm+vEdjIRYx*VWA<^_Yair!%KTTVfED$fj{O4YWD$_3@HR`xS#!U
z2j%?m)z2V#O0%U>;wjghwnLS6;pk^PcG{Q@6<;+xocFr@<Y>lAj?r)H_Eri*yzg(9
z_{`Wc*KNcMuhq_dt$qudUe5q&xyy+ncPXc>t@|uyu8(v7Yv+qlc9=cBo8}2Z<@&aX
zU$1g$v<+0Q)EnI4lLs_8wfs67yZQ<3)#~#5Xt~~3KYVU}nXSjoAmDM^nq2kM+j=)a
z&waj<cej9S(>av5gASYx?=TkBxJTn~7r*+)!nG2bW>mPC$JzBxC<RlUP04znlC2#7
zb$9^4Hv6}JT`zXQ9I``>7x|udG!#z}HZ(^FOCktxrNFQM$%k>*pR2O-qT0G&|7$ZW
zKbVON@!N!{ITO%eaPiOJ4lg|*{yF|@jdO2@CVYXt_Im5IC1TS@eC@}Dn>+Q~ZXCBm
zbFd8fUEPqB^=u*X7P_Q42VBeVRUR8TpefNlwrEKA=fq#&=cC1KGvIbp+5JyGPid3c
z?^c?q)Z2Hq&looo*CFMJV9~fkhp&APl?KS>*7SQIm(}63%0y+TtJQa57+@sjxXise
zanzr;k>1qD*92vJcfifbhO0`uCLBXgl^Sjjz*)J(fuDa$d7!W_I{jGl{@fO3@oTVs
z&--(;y^H@mUB7`g6y0yaF8y@LR@vo0-*G+Gq2S>@x$0LhB=iEqhF;n))aKV$rQbK{
zH^DZqf-4p)VRe_I>7UCj$+o5cTvomczq{u5e>JX%JJU0V?MqDo!KJz3Uj;V@XTni^
z5@N2)WrD*1UP9ps1e|_l$G6@0{k0h;!{Lr-7QkiV*MF{Q8dY@c<wLCLSm)4s$IrE?
z-jeoB`Viwlb6ej?O20E*!hox&pD1kHwV1IQRy0Wabb!7opHRfFjhlU6E;KE!`gLtC
zS6ryuFOHk-lrxsleB8oE7p}LUee!;DoGaOSq1bB&?i%V#MMLrH?q>6R+8*rxp-8t|
zY%Z@4SLMrA^V|59o0XT@OZDGQnYW;fb06O%=(=zhKBnatarT9aE9TsWZ*m_{U0!g_
zL~g;Ow4=kfeBrDk=T1Hb{z_<)-$%EtTzDo?N&HfN6S=Ilb19#1eP3vc9zVA4@&ipy
z76!bxA8ixf<=p9)TX3v%F~83yx8NjP2k`2A!0mVQ`);@2i(nEk{ruJgCU900Iw@#*
z6Qe~84w%)_c?n~*T-x{wxSr4LnmDuv>fl@Tlb(DOj|R|qGkNjenF(KhZT9-C3Gu#U
z7;WN(X#?b9Fwo@MEs6h}M^G54UDXwbk3=CT>o(pOz6D(FiSs2O=TH7~E%i2Zc5Shx
z#=iZW+}#3Jowp$Z>NxD9eW09p2<w{728HWhz|})z19C3H5BBHR8=HiyMOP<ST(t7Z
zD@Eq*ka|=HY4=S?=s@e^-r-Q2zgq<RN?0fMod>+jX2ktHJ-Wp%<Nmziz7N!yikz@_
zB>bF9kGIZ_cew}Ah#ToZTkX3?*690Fa^m|8WP!X;aK9(A!~MX`uIhwm!PT(p{rOVX
zmi7B?u2%x1aDShnD*HYV7Zsbw{+R6p#b2&173(&m-Dj}QTc6^32U@Ar*3F$x+SR*g
zY%X5;=u&=MC>-o@XjuD&$8i*U_=@d~W8)yMOX<1yV9=Hy58_hhxNKDJ@ZmSP$NjZw
zw;slYn}nK>#oT|bKNwyK#8avF(=kJ9ae*Gc%h_&8@oRPB+sF4IZ9;5^v#<MeTl=d2
zJTImK?K2t<cg@dna}A3)HH6iPDv5%PyRrE3$p>!lP-}NtRThCo+?VTj3-lvE@z_26
zG899$239IaiVNrPg1fX^V`mQ@=xTMqF}STEek-#k<dVqpur_Jx0~a0%d;{wiG{aOq
z{>`m#ZLZVchMK26mDp0#2e(U#a*si^y`a{0+c1ghV>`;#qRgaftj*lS%R>v`#w~$!
z;U03J70jOEKOd<u8LZf`K>X$|Tt3rG^_huU^U)a`MqFlpAzV;N{kq}1YN85QXH2x_
zE}o2{O{kDHFcZA$!G<Cj4v-Rz;!G<PhPnfj+|X(st4zJh-Sz%jDtrH2ER^|yQalLJ
zs=24s@CGYBysM4d&PJZXY*oA?p?-0Udx>#_s5bnX?DGb$a29hYY|wFZzCR^#6>_LK
zC7Z{CTiqix-KTWN8)e)~dyak0Mz}Y;6*zit^ZuNDBiM-JdYFetL)?W|yE7eLp+s~5
z*qlCBOkF9r+^Q#DRBD|dAqE1HvfZCrtBp4N`gAho>OYZVwB9E%Q^Mc&o;i&844jYs
zTq#V*;TuqL2c>aVPhbin&xBKqLqazZBCDhp+?w0W^Tf^Sw<JgO4ts>eaVYqS7#wjk
z9(XP>6V67~2eRGZ@wviZ7KjevKi^$$$?5Nj6=B^TO<}8j<8v4$TEoU76u|ay7RiW|
zR(jlVbHH6R-?F>N4DRWbafR|i6S)p$%nR*3drY;m@AvA*aRX?AG)Kn|)P%ONr&dC%
zq7Xu@w(u}=xSPjStKo(KYXct!_aVPO?gjISO~i00a=36}@C7f{mBtAV`zxD+oP9fF
zX7uL4TgfBN#vuzE8u+E>`rJ4rG3^rA!aK$x0nLCTy!n9AW^p*L>?1<V^eWes`#hW3
zfUo=hbY0!ckQKOufNs8vf8xaTu<slD@a+8(m}p(J*`Vg8Z${)F1V-_}?IU-r2!2`!
z9>I#<!>j@>&MtO1R0Ql?q5#Uy58J~B`S5xOpI~EbAKg)UxgDDB^|2$=X2QiqAfn^t
zQFHU~<ip`8pPw(zn18uLf4`7ZbcfsuPPi+_i)bEe-x1p9$%#skNZr!sLUa8*j`AUK
zMIe^RM!2c!!T;J7kZiEwcJI;u^5dEjgJ!!LJ`$RZ=AYqQt$d)eH_5Go+%0SDHsRNR
z7tcz$LTEU`2A~~<SE@tl9^Vkv{{DQL4B-)1YoC%+hj$+3KiR_H4mQF!5r}j2@UVw*
zcfg@ck8!wG9+@@vBc_KIT=%RQs6-Dq_aknPOdczP$(k1>7aOA<=TNG2y-@sHCf>tk
zW(jPnlu`8l+Sc~s5gTXv!5H*tm)lSKX0q-HB-nw!54q53)Zq?SVBqN614wcit6r%4
zM3Ce~CuAu(VI}jRySc4Z9hw3CfkH3=^v^uuYFIq#Q6#PAPeZ#9PJrv}T5dlH)j2$-
zW%&#oH<KOa6=jFJcqB<P8^a)J1dpr!NovN40@i}J0e9wCjyrmB^=Oisc6&p!ZPjCJ
z-c=Oc6h~aN<dRE2@#C7WCAOmM+sr8Kgy`J;^K<(+8{=-D#keJ>Uzo#DY|8EWj6k4u
z>HW36)=CV~(ssSQ6pk5zJwG*J0tg{fam2g7HVvV8!Bs!am?Y|6Qrf5MM5-Z@su^e5
zIIAbl4eD{|)`j>E{qh(BsUB^2rgK=+M+LZvGJu;~S_X{;f_M#_&uBgn^4vqV)>-Op
zrxPdvxN_w}QWAGVUD)&s_i1G}z5SUder!#*eAp_uPGFmibz8QtyE?qvq0K+V!JxO6
zyBol*9?i9N`8@ZYL|OExuX=lFCT?fc+uW%yFj4SuzzVvWH!qO51H@D)DR>xdbUkh=
zT|0oG@qo~Hmgfwn(|9;=Sarp#`c-en!}Qf`ogj1U)Tswzxh2i;5r_BVCCps+NV#mv
z-h#7wq+C1_Z;g}SdqA1}LK`l1n0}cKiF-q9D^+ym-U?5{ZTFgN`&p*ny!tTVSOh;t
z8TYZ(eabj_yONAB!}!x-C+3IJ3wOsn2C?al<HAZfrC%P9BGf)0;fa8vF|#^XNtaI}
zTR7V-**t^}M|@%Kbgut#MyU}9E#U9~FCL|De<0yffVYw9p-1q+%w2jQkvu$Lak_<c
z?lo=y90`}u5(>d4fD#e39TM&7mj?rkt+bA%r3}#2+^X&#1cq`eu4Oplk|*bq5xT?l
z11^&J;My{tEmx1I<*?^M+_`;GJ;L)Eag_kb4PGxfBj~up#S#k4bEaOykjK?Dm3bg{
zTThhdm;@b{{^s&--DCQWA3hc=Q|^|Vevy%gA$jiAiK*Q^&@&!FAILpBU`x-Uj*B^0
zFF1frU8mz0&PhJo1@iehZnX|)?x%P6#%6p;ee_|Q{Y|d=t=P%v?gY(XNQT>hRj#D~
zHD?Q+$82V^E0Gf#_u=Kk%f-w4<8zmn0FJ4YbUb$v&)g{N^``6M-n7G!?yNu$1<!eM
zL{kEq*;za+_o}zbiBB|4(k1GX6^c7KdaJZIpbodT+qfwzzM7aRGVfq)yhjNWUUKIO
zh!ZBT=AKE%Pm-VozV#9-N5|E%xWqwD%nRL)RJS~nq55)hGF>%5>cSIQHIlXIbCRXL
zCF}Sa0e1wXbI6Tmn8sqziBjk3lW=0v+saUsD9+YP>;rcq)p0ojSl6dcZmS>K=BIf)
zv6oYidKFYSR(Cpb6C3K=Xwcd2quU%jG40De+3a{&aaIbIQg0VjJZS)Q=g<>)d!DfN
zWlPRkk60=maGu~UJ!$nrB-{o{1hIiO4}A}<xpf2;oCiF>d@f*}Fx<;y+jIQbF4WW3
zjr|QA>{C1aWG@>L?9Ck)QbU3AQ3Kb^Q5=5F)JME}IPj!w<egynZQmY<r*wmA+EE=N
zjP=541dVqp?8R3P3ej$B+9QvEjYi4{cdIyw1X_#xvQ8JC4b@{Ij9b<ge>r8xouK$9
z)Y4Bx|B??(oV!RS&}Ir<0^Je)pnBW^xsBQKu7^t8?SKx~kgLZJ$Oq;O9ywP%md(6f
zFjH6b(DZ@moQ0p4-tW}sESq2mo{}IBLOX%GBddUYLiLsvU&q{sq}@t4%Ynm@D;dGN
z{dS`7gpT_Jig0o|(`T%6*vdy<J$`IH0q3Rj40xCPbU5{#a`Gp9gfQopteZ|F+^y3^
zd<6GxS^d_~TTJm$#;0DhVD-@2a_!_KPfiAh(eC2OVchA0J@vjt%|t)bdGtKN=tT9#
z1+V%&`*F*1&*D@Y7j%I#_KFM=TW?V|x(@xUVwh>sSs9-P)5h5wbf<z2HK`Fak32eu
zC*_Pdmo{oroOk@4nem<Sl2CcXNbqzQZQio2;~sI`fnxs!s%&34&BcR#f1)S=d-{fU
zHtJd2>Ip#BE)`<FL-Rnbi{JI=2TE*kK^0j5U1@E0cBl!b0F~UX>1-$WaZ+}O%B?|&
zg?_So$fslbOq%&wtk{Mf=hI!KejZHSTg9p7%qIx=5rOX0luX#Mt$v+8<s!R^^NcqR
zDEm^A;-`Y1F36p2ymiKVjof$azU|fzo$2WT6}%;!re9|#COD>T^?-Nt)-XI)IaJc)
zz3&)k=q_@f!VKBn552kp!NGSuHR$*}h4dYXS4#asb)Qv3%Kp@_Ibc5dSf<>kP>;0_
z%^O-yJ^q=~2tH;Dke?7b?CJQ_F_O=^C+Tet<Jj8aM>4BNiaPKruh@k(=yjRx>VZmp
z6mRAOs43hl=>48Ba_U*2OitwdAn+d17@%6@)-e5ci&ObL!W!)?P}mZ;UK}vJ2J7Lc
zXV%Se^50bHmmU?={poHGO#GbXY4;z`VDtUVaa(~y#T+O}<E&5iI>WqzF2XaMTsyHk
zjN)gd2d29>aG&SeEBA<p$NJd^>Sev07i$wE1`D{>$#}zN@9mUHj1_dpsteugEKUS~
zy6xL1c2*~wWAoO#+*!bB3UTO?%qx_KtixkOoZ;41K7A1Kq#n9H17oJ{Ssb*t+XZ#;
z;4n|Mpo`EA?!cIrJ_gpFp5=)nw|yyoQ-?+z0%)C)V90(FU(?{;dpY`mdg7U+QqKBB
zfWTXW#}@YFO0C>3bUo_N@um)l)bUTuI#O=GY)LpdsY+(NaO$^3ou)=HJkszyG=ph5
zRpJjjl~2E9`V9BTuAgTx({P*`T6*U|C304ko~LSqYqtb=fR2d1;Z)gKaT;Q6aN%{F
z3{~a62X3w6aN#h{A%MbBjQf)2HrC*_C|5Y3lGgq3Al>^^AS>q*fyq7AgGay(C2(-#
za*Z>LI~O>phe{@JL%+4o9w^TWVIIerU=F#-IS5fMC&5<Dqm0q~Qp<s<qX5Yga&mFi
z+s^KIdpe9PNKVy5Klp)eyS3`{;1AF=Mh1qUhkx*x5TPqV#REV<_bqx%2;j_9i9xtO
zj{yNqNe^8f1;Rr}=%u}REC4P^<|YZ$!!|k16O_Sys(Lc6p{3ulEC)ez-x0cEew&H0
zZrRmZ51Pgz)FW22;nM2um{~Df4^5#w2m$D!A^2AF7#WI|>M;G|2GPvBPkBO)uM77$
z+1Knl%XjAYUn2cl@1fZaZm=*_oyKi)r>-*{&*pK<n!(df4I>?_p?jVo=XSd!?IY%Q
zoBLKfiie9h(K7m_%Y&4A;)4(6hONJyCOrp%5a%8>0-DaagKG>PH-h*YdH6WEh=3=n
z(|vnuAC3CUY1(gFZZ<lLlz0yJnKQlgatD$0QhI25GY(9`Q>W>c`Q2j?>UlT4<_77&
zBGi)@ujTzF;MSy)g4W&6TO`p-t#aTw?$IQw$AG!jc8M9@QJj;GUeCA7+qkXS+C2O%
z^qB|3gAjrqR>E%r*V~$jqtXvmx%Dn-XD(ik*MWnwN5Ti@{paC6#_D6+<<1Uw6K-e3
zBrhQi9DQ-VufD6869HG<e%?;sl3xzawnrr5`tyM%zt7u=D7ZkA8Y}Y?hGIK#TRn*k
zGybkKnJr4{xWDf<AYQ2npVTIB0A*e~{lGET$(d-T_`JQszBxtH(+nspQYmIaJE$$x
zC1~FJgL>F!r=QM>FZH#C_f=4vzuFJo{Z>E%k3(ns3<3AePYrXWV8$8GW9EAhmas1d
z7NLisfX<Y`ScH0H<%lpI4&}^c`zFVQM^qD}Z6@mF%&f7~c?-D8(E`{Yy1yPoW;iql
z04J+@q~y@NTQ_i&sv~GR2M-Cmj2jeJAH^}Lm4J5<Ck$abVse%|8-Q}X(&=Xs5Q?xJ
zG07IzeFiQ4jP`7G-7c`n5Ok(I+pfR_&1(_Ev*R5(c|xQH<<@^YG)@c$J|W}e0<{MZ
zvwpY-M+rLitpGduR50@N3%ZXyne(d9+f$e5q3gCaiPY&V{Rn`55vKZXey*Y%_OBx+
z=hY9kU#>kvE0JduDnrgkJ0&m7jqTSEPPb&|x5#O9Os;uv+6<14_sI-8b5{|#V|LHn
z+#Lj5-{-5Hn))|}pdLM7YMO`8#Z?5qUy~;=15;3s9<WMYbQpddSjM4JZoR}_=#f=*
zB^=f57<VZ96dnA`;r&23hr-prdYaK(fjA>Nl8ZR@l5ou<cUHTTo)Z+eTT!j+=V?V^
zew#hGE;V5h8DRU>>e=~^)oNXP$M@uqlw6sQ#taQ^|IO2{vE#X*Cc94sGXm~G*=;!)
zXlLrlP#tGkGTl^fz0H%o`i-W%_%PXvoJV-9l}hdO(40q3u9Ro<Y`C3Kluy5i6Lo$s
z<3pM2=iNy9?avdO|5UK^MiNxcqOx|*gIRu`4I|*X?fZ_Bv-Iein7~I-{&=cStV%)>
zI8DbxsO%PH)01g>Fu3WIIEnpI-A7K^DpGG*Q;=)WJb#@Y|Lka8QtJ=u8A6@rzV1YQ
zC`VFC%_VsHdU}RiGMtuEzc=7CT~?<@+&XH2WcEs@J7&1F$Q}V*9!s-&Tq(JfQ;(o`
z(hxtY9fJug?~cQ6XT6b3UF<N%k;FQ;w@RwjL-W$>BxKV0aJkepE?oUAO)_)qZW5}f
zA3-TIFX}<}IEt6tBaKOT_c5Y|^jMz+`x$3#;&%V-LSOU9N?92jXEqN2aRzto@}WC2
zI*VkOm)ZQT+!+N)C#y-snspW@Td+cHfRZ`w&<xfxi5jzx=4DD+EJM&OG33{)coOqn
zjzT6oBZbW7r-Y}q9f7r-lGDklrtixn_Itcv&HEt0kk>>>t_iDq%Q@B9;3hA9%&=Y3
zbOtUx>VSQ_pQ63#`ThtEFfP@Y<@he$#s+ysoYtCBg<u%vSl=A%-m(O@3A)y;7x6m|
zO<yV~=^0MG?mBK+HiGA&{nNAUK~bEpb_Z-rI|9?pmw87_JrWbNr*kG$`w^J?8h?}6
ziv3usJA))c%_?3|L#`&!Sf%$!N1#byfuoTS<zQSGf@O~kX)?Lof2KY+OK<zbBhEU_
zt_}|dby}JcX&h?v{s>$=q;6Sq)}BF}Go`CvU_YPw9TqO2uOl#V>i$pysU8}s^`P5x
zpgN6I0#1GB)#+NdU&@vIL(_?2a?5%jg}^kP>=*Z>cMlptf1$Q89*qEpns%wi9N%)+
zElff^0+RrUw}wubJg57La!kUrn(Te$9g!F#N&NLYpVMO|A1FWII`U_4ZMRpZG3$9Z
zxl;)wMbLriTQ|9yT93Fq@%DrAah?};uQozgt5HLmj*}~%?F_h<y!r~=?TAY)dDuG%
z&)7?py_1drjlBnrfO1a#8ky6b$^&uc`;i}7bB_qh%sw=)u}p%^6GFr3NJjpP`2GC!
z3v=nfyj43rWATWhI&M*v*Hbn(+9vnl&a<L-CbNM%3K*doZu|JL9GHs;x5?I7oY0VO
zo%Wh*{61YkCc)<=+UgnAX^ID|JtNpN?#1+rRA+LS_|3P-U8voohLy_@x0!ZytY;vl
zi9NWF0;VH(h9@SW9!1{F?_7qk3H505#09#o4NsTtbmW{NoMWj`@BK)*C{tp<$G%-4
z9d$sxJy`JyuAlYfUP<ji?`O+RkfXzL%6A*WCFj6Aq?i)%#+=H$EtwrLg5Gb5`ws4X
z!oB*kK2&=NdbK@n81$pbn`xQP%6))-^e<iW1En81sq3!J#Ht@HXoQh=>X{Lprb#++
zZy#rQw(nMN*;(h2l<HPRQpx#EG|Z%n46KLF`(&uP0b>(j;zYfIWsI)5eMvn~KP`DG
z9^gHPobEJy)Oe09Zu1(7t!d%8RN;{;5*4KFIq{U;C1^UhLeX~)mDKcbIo)SnQl#gl
zx{o5D8C4nDr?#N!DKoJMHO5<G(|o$G;||#O?bz`{bb4j#@emBU@*BlXuN=ceZ~Tdg
z16?gaBWZ)vjTdQcKh9|;PG`gsaGHU`CXQcKqioB?(Veo|f$?${UQv9^u-wcC;?(Xi
zzmSDHkRLQPzvSGKdKf*2kxI+qa&IFKSw0hTl|dZVytEA6BihPUap7s<XN#UYR1;3`
zp~Ks!x!mgnmTnk}1A6X;Ib9#P{WJwApGuEK;dFlng*bpgBUOjf#pM}~oKbH773KB7
z3>vsR)Ob7N-&%ai%MYw(RFLY?;>d?M98NQEAQV;FVIK1=KPnhO)01TR&iNUaCLhSP
zP-pwYhKFZ&s<F^5-xi#Q_6)<ce5-4pSI{$n50p%+hvqfD#a@8>oIum9YE@4S9CVCZ
zuZa7cc&gZbpLTl|d8g^*yj*`-=QWtFr%5>&4w{=7%V$bV<o;peJ}2$C76i19dLKCQ
z=pL%>assXS5(pUUs&n)32sNqoRf6Vn3zhQ#W3>DG!RD&N`RVE4cE^Ew>zp&**T}87
zMT}BkshUrm+mJ2W<GceRw<KlbK4@s^K*fAK@G}Bti1zbhX1X3M0&co)()`BK9TyWz
z8Z>fNm}18Jh{2wb1IVf0RXtP_R<;HYytA7ClDh`|Q5IjUGhQ=~TNhy*v33s)O9$%V
zoSnx0*1`-7sW)?9ePIP^e6}+!J1CcV4wMVmh&oqBKw1PAEb(quWG!-<BfBD93~-Zo
z>d*{&2;FR}In8)Ni<pw`ZsIhA{357i4I#NQ>=u*^ABUz_#PSe!a|I2%8@k=y1R81A
zpeI?A({z(s!><8G*+`+_bmd%LdD^%L(uV)bp?d2jHT{uD=>@U~mBxj-+J1P!F5;ar
zGY^$p!CD~_rS=PLcFh{}Q?2Y+6eN;G0r!aI&r|ROG)tmu&~-a-A{aUyahkn|zwaI=
zPV;LZ1i!?xB(vYLb-2fg(|yafA|Zffja$~V`dFGIuOlsw9u;4|WY8w)c8?RM`JG}0
zYK(Y>+*_D|8j|u}w16%mMI1UWB~<L8yL`$nEujYNl9QCUlO<lnuCcGV2s|0HeAn9p
z{OoE#y;MEpnA7$d1k<5bRFG?~N85>SZ>OoSPRN|3q5m(BdlQS(PT%}>Lf1sK87>m)
zmaWq~99_$ArY4=Y4|sqD&23z$j98p9F%)!?0zBxxeVC-xJ~a2C7qR<C4tXS?gi6Jb
zKjF2?hsuoQDI+f}!idbBaNlaTu2`P7ei}2CbR^AHo|qhNA2m-F@}cp#nipCYAuyz8
z$uI$+Dlz1G<sFeIxU76=&KiRH@6ZP{?{h*WBoLbe2}j?tK<QP9#Dg+M$8wW>VG6q6
zNprWETXuQYzC+i2h~-x#4*k?1o>tCjdI&E<Q@Z1c(|NmVGawYOydJZNO<7}w@~kjP
zca0V5hJ0Pzb+3{%7Xz-VN2Ph$1GdPwa%fJl9f>awm0Py$_#QN)4lOQU=kGrSN68tU
z(RYD(o(9Y}!!_g)A2d&L=^j1lg`C+$tvh8Xw{eFi)mv6PC%sd>UBwwP;kf<Ud*<X&
zxn<RF6CauapUa)==<zkfodef8gI(RC3R($BA5<>f(i7^fpbc3;^Lp;m<4u2H>U3P#
zBUZ|xIperXcb=e<py9%OBvSnt2bh8yKTlVyg(avxi;fTV3tNY-+X1swH0i#5CB<3L
zp5la?M&&F{Y>iz?xS$nu+#1y5jUF2NNK!nSdD8u%S|y%$#G?t<Kw61+lZ&Z%j?EnL
z5&KBuX%=(D<AKF*A810kjdks3aWbm6F@lK)O*2nCJKT=p&EO?a88vBqVuBxb&ZK#J
zuym|==cz*q{=@OGshtbnxHAVlogtxwWD44Due8q{uVMAuE1T6d+6r?{a7$bxi-vA2
zIQA`xw~)6<pduwYmI)FBe=zAmbEj53ThU5-i<*6)F$5ACjU30Acnj&W43(YZo$<H?
zkA}Uo><4<xg9G<%45#&^dY?=}x3wwx>8xz7WgJ>>wa2^JVa#RmE#?`6xNoJHI0@qc
z+!)XK9m$-W?mFUy<u;Hs4`Q0#+P7@8?>MR6vW<(6hekFTD2|j#>xXg+FX)B%XD93l
zCs)A|>ZPmiq@gMb2?a)tk<olLWE2=V-n#B6hzHX?RBmgg^_8Uic8uWqS?}$FhJkT%
zC?(wS5=cCDpyxcONplEJ=7OXjxI)m?$YcUE`zD4PIdv(1!9g$V$EUW%RhOV<%xltl
z+ZxBA^U$4iwjCdN%{!OS<4frDR*R4Pg}1j-G{nh`r43ygD-&}j>Fy<l`}4|)ds+IW
zw_|#ocbeyKi;fZbvcxiyS>2pUZAn8p7h+##chkcVs5M4BqaO;fFI&;1x7J=;&$1~c
z2!Gimp(Gr=<%nl)3n)jMNpH6%Cxc0EV~bO@pz}7iWYkM<Yr=6(&~+i_Aqj6|o0Ej3
zw=sK<J!u?1LC-e!p)oduTArHkAw?a*lJCpe$?6RI71)!u+`WUYOFf3rp_;IF>G>K`
zOb`)Rau9PBB&2V?8NGcXtxnMU<C%tXt$8CrJTPuf@6=GBo{K8z?J3s_R?^#5GxsFv
zt#+eRC%j!y4_O{G8Y4PpStF@jx{6-9lTur}m|R;4kXh%2HR-vw8MG(utN}^arO3~u
zw+o}?oc#%RFSlAvI+^xCZ(I18U3h!TwU>MkDbt8&IqDzi9^=wGVB4Lyg5Hj4ndJ|3
z&3<VN<B7$Xk2WscQ|C7{)vx7i2oq7nbGKjGFEtT9A8H?d&9^gdy{1Wy_E38;sShO$
z;HB%*wfjp~OWBcu?|^&j17&D^DUm%B>V->XKU8}u@=`7N<%WeR1kfyNeM4mY3#GRC
z4O_4Cg<9`u=h8;RtyafFhrf`AN|Tq`f?wVmNPyTLx;!vu(4JK4X9PY_Mr#k<easi?
zo%A1QmUa221ogS|@CWo9VBfUj2LDnIpYlyPU6~K<b=}GJgNoE!b|xkI`X8u@jEBy-
z%S)F>{d=hF^->|L#!FS(OZm#{L$&ID;pe*xxCw&rV}f6bf&G@Pt9)0xxqM&SAE>8<
z_@;f-{ZjG?zBM4K=0j&|@&~$mdFgU5g>On`rSa0$wLf&X(f&a9m~Ra_#{POgo11mB
z)|alQ0&Z|UBj&f99%cfv=|lSuq4a@k?fp=~a=jIdE#vLL93pSMW{;J(-sNG2zbO&4
z{y_KEcMNbpd#SP$_tK^N`{nonXKgR_cs}2>&LVFO4fE!upLOp8Pu|~$YJK6D?d`z)
zu{?aBG?9m{-pe1TXR`gKGW#<ZNPp1v57d*=eN*jYzGX=m@m-b_2=iNHZgL>Z(O)W4
z-Wtr_F72hd`dfq97^^<itO<Q<7$Ps6U538&qZ54K33d2QJ!9GfWiNT??pyyr-M{3U
z>eOe*u?zK)n1t>ubv(3sj?iyP*oQyRRUe5-{LcCeIRQNDY7ZTrlJT2z)A_Ao`cdgO
zUF}&oh~n9cK2%13pJA8&4vj6S`qpYvrM<Q49*@W+O~QC&=JZ|3_0lz;MT%SrQ{H+_
z41OyHn`{ZIzjT!Ktr$Sp(w7qG@~vTRd4`;<39{=$xsLD$y7qW$*loX$w)a+js1AKz
zFb3iyg|H8G>qBcV^$%3bcxxbt<4e2m^E_1QvrZA_v#z%W?27)-Ia_;cShdZ!?2`5f
z3}SuuXPqKHK_71od-g{iBT2w0Zw)F>1q9}QMoi90*f%uO<k7{5Z$HN$sJzqMBg&Vz
zUPAl4U9d%Fy)_WQV7`>d;16A0<{zlaen&y(t}opl-tM8Y<)zxSyfvt5|9(AKiO)R8
zO4MhaA_i!AVH=NtCI)DE)+wTZ*7DNP>Z96kzvL+oCBNod!}24ohptyqB^+qmZ-0pc
zJs!HpJ0Cql?W2tmP<A|0lSr?lx0mkmNKJAt41H-UuR=!P(MQoFNa)-zbdS_;I>vfC
z%XYzA0kK!)p=Q563l;Gj<quuJm*u6~{Wu=#88^NuF+J8xw|)<~Vkzpk(A45*5$&Z*
z^{t`x(%!O7%eT-};*OQ?g-QwfRqI>Bkoq=8H1)TFsqfF%eDn|{#8LZJ?V<84e_A|J
z<XgkmXO=60VfyzMtOQkgzILdLZ=tC~?|3E|RENGb*uS=yvH{gzDs#WodhjhYp}NLP
zNrUm$u&TUFo@uUrb?{rahpN0q_uTazMm1UfHimP=cUkurEq|b@-%6!oDC&A?tNJbC
zp}Vw~66Iz-lq(ngrK>&@nz*9%TOEV$DsK&Iv~Lqui7ndtTLYIJU%K1#ewtrkdZ3O+
zERfEj?-#D|tr;l!9L6)^By<?x%9xVUVdz_Lch`W2lFII(qs?z+Og;B@eIEj6)|alb
z<gLN<nDSD#Wa~?}d@oZ<dIx!g6A2!6`vYCyYps&vVLu`hQ&C@7$0MFG6m{uKmwqo(
zN~F>9$V@_wmTzTDi8WgGTf_7d&4-d@;GwHmdHZV?e|c*l@>72(nPJ`<&h5W%%HH@p
zWL$9lei3AT==-IW0M9R#G3rBGe;%0Z5AQrD{ll|RlKx>n>lFDP=6)++e}BDH*B`pD
zJQT?{l`7?}fna(sCH>5NsY&yf>fGNptda?0KVwcRi2Vr7<)QNHOI62PLz~Z(Cf?|a
zcIj`OV?0AZtkLpqz(QB8FI}yUKhRyDRf=e%t-O7;CFT*jL>q1EEvp`o>Y>D$_@d;6
z=+6`r_Y@m1^-vVww6!tb8q5Hem-a`xdakhNMUWn1zMVDA<`He=hgk3L*H~{;Hj^L4
zf)o)_Ub>q+3mQoxWWD{;A<y(8O~if%oIDYGzO=Ue$al{~RNt~~=(qW**oi8Cpi;ig
zS0&zP>3^X3*FSX5XK)BN`bayXjn?-O1YhEpYI~u5`TcE*tPk(lDlU{?lvtzvS))iC
z(I2&d%n|(=aZ*Q&XT-@JF~&=KAM)1Fb(WWoIluS(R6PaVAE?Z4>s3i7upYI5i2iH7
z^ym5Mg(hqsaY-n8d3&H;-T~9E^{Ev?r|&PN5nAh>Z&{Od`=zaG{R36~TA-jD0I!D<
z0D8Q1moIq<oujOmu2S?5)Pvr9Q)%@}UOL@%zjTjQ-x{>E@dvuv`W{J-vH7LdJN-Ub
z2G(!G5!g!KZ>^>n)tB}*#(wD@^3pM8k%!vE|3e?<n(L*j&1cL#cu#w&jPJu!DDgwy
zvU`1>+CjB_?~yv)-={^b`2D-`Qqr^jfv&wC*xEDZ*of~u$3}bx0Tc1t{B<g&ycLj^
z=IsGdKi{%z7L|vZP-t%r_(Z%t+r92TP;K?4I@|buX@nka@&_ut^oQ2o`yZ(G)gIbr
z*9Ufw5A)D(#}44=&)g>+#rN6=o%2s^Ezc(UP?`GHD}6P2sjm4_Qx^71)nxg4sl1Oy
zLJE=oQr(Z#Bq7D_FKu<NKT!GhS%LPh?GIFSzI068^`X-vT0L}s?=L{tc*GE~M~mu9
zS8Xrdt<JaF?OA}ose1D8ztZVf@&7<qdzLB!kbXO@INiO!w2l4@Iyou2{(-J<rxoaE
zW4`r{{q3{@ou2RDn-b8szI07_skBx8K=*n^owO8dz7(OgU%DO{O<D?hzSk8QSH5W-
zyR3({vin<>D7bG~Qz{=Zjj8zUxB?wR=S!zw!Fs5aXRRU*X&HZ@THfx;tmPdXLMzvo
zz7m5lq>my%6w>xkqL8-n(puZ29uS7KmzOpVWAjkYVg60~{PtXd5+3>wRK7h|pp!h$
zFV$JcL)V<okc+20_yhG!xZia5)&D?sy&Y5Xn|{&l;Vb??Mdkx*e}<gY6#JPVq^8)<
zT19G#z1|wSXEFMwqicKXCCS@*=$yJ=N}TRL&|Sv+Xfn#99uSYTmOs!{pCKn6X<dJy
zyFP0b5lP$g1rw6A&H7UE#XR&x8~=vy8tTae<$-u&?w797UTUA;d}|<q-&>YMByU+M
zbF7!H`OGnLQ>^mVu=nb3S;9SiJ+P<n`u1FLx}?3-gvk5@-SWLsc{ZnAU)sxlyU<N_
zst^5k3u-3=&mZU<I$w%Q?0%_~`cmoTTRFJdMZW2t-;OKI(_PeW1>ZA8)Nec1vn*8o
z1J!TGmFAKA+xI$kg5H1Iu_p@rGvvfT@8b_tyS#MFQQmrI9e;rGJxNczTz#R=`cj$m
zeOg?T<)s<z<1bYE>Tg-qZ@-mxVwZd`Q~nouMw|>4GGE&AR{dY8?(b!)lA&Th14f35
zEq|b*->N~yT-5#ts=De!O-0kUUTJ@*R(5%xfkX)UP<@np;Ch~xn56X?@$S(nUTS*5
zw_;Gc=WO3E{c*+QLwoDrMl5vq@%D^l4sQ*!e7mhU#U=frtBr3X7P^PNR31NK=q~en
zm4bT6m~X1>+ds?cW7Jg=Rjh9_whq6XykEMcy%cYKdFht@(mwXLYEa2kvBw{%r=|U-
z6JL^-Ztd^0kyqoTQr>A!aMIFV%AJGuQZ3J*6P&a@Go9e1b^L)!eS{{#N$a;A3)R+M
z+U7I=iA~y`m5SJ;y}oqxKHeI}vl9%R^ZPsuy51c~sElW&B45S#akNvN?WKFZ5`B84
z&;C+b&q_tUiZ%W~bv@bv87ri`wADw%lCnb9TW#<1z5NH({T4mxm%r((??lH|)bdu#
z6_)i<ee909Jn;2==x(LGt&v{pElYZ-@ApHP()w-1LRbCP?<-+S-+?4d>AR>9ru196
z{y%^HTlw`Lzsj$F{o8;0761Ex{LgBz{HnkHoy(Q+=YRjtpZ~>C#f?zozx>y)|LLEm
zH2o^^|DXT<-~TfU{V#t1=kEM3s{Kn>^Edy+Gym&6_`eZI!?cC}FN*(*KmNo2a{lL;
zRht(u?UkXjKmYZk<O!l?Dhtj2=a0X0alD>jL8S5|$Wau@zy2seC`iS+jw1K(Kgw`j
z!1o5C%>Db1G8LP#6U-vJbHUG2H*s>7y5MK248xm+^6x*(+?!lk=z<@mKAU<`>H~h3
zBKbVAQnFC~{bw11prb<ELKplj#cf)yt7oD7`_G=|d*aOB!$<#odFpqC&Sm|4T|Wz6
z_M_Z=XFm&m;E$5`&HXI-fj>%qUf*B3`Jq3{e$rw8Ec>B9%jUcMOVJB|mb@?YFGWA}
zN7>J7{aN-0{aH5O?q7;t__O4F#eXUKp+C#!oBm7D3xAY+B;Wt$_wJG=17cX^9v>pt
z`3RChndt_@{8=PXl)x8o5I68gk;`}%4)AyV8GwU_!D*}At;=ZfQ}#!@pQ1ln{*?UL
z{-@~A9z0`|;`Kk{3Dzg_`j2}5NGbn6J~i0C{@>offBi*JpE#4h{+-Yx|6B6?iedZ1
z{`RxKcgcuKv(BHdJIna@EdOx3QR&~cd&I*(+-_9*ckLcw^AEQhmHu73M{@ns@Ap*t
zx9z$V{HI@V*7~<CyHK6~bjwle-?r>B`k#KuS?k}mj7#Z1+bT8wZHvw0od46UMy-F_
zvadt``13}sf7`MzYyb4el(qg{%cn(OfBva<y~cmPRsPRgJxkd?+w7;#|9QK<F3LaK
z?x)UYd3_enKd-OftFW!9|Cd{EZsgF?|N9@&<p1%nk1|Ts%7(A~(NM{e*?~Kg<`G)H
zlJ<Z&lCn-P86X*;fKBy;D<Zah=bV}-ju9^jrmH)XQBFW`BFp2pL{P=4sWA(QmWh%m
z>x3#^o>tsGI*)PKvf)kIWARkh*}<(MUd(J=$@+A1Xh+Dv9{E%VC+{Ka#q+<e3x$Je
zV)$ifZi2~ow14sZZ2dyl2*s1)M2EN(L^uWU$jK-EA+h8u#w6kIJk41^zh_QJyB$Hn
zh$TNblJ<afJ$p0@4<~*_M6#W^IFKfA-0mfRP2LQL&6EBH+=Si~zp;Np2@*vfzcj+L
zNuSxv%?&a_t1a32WA_pxSlWu%Ys=HL>|ZAUT3D%`Z=yIHBh8)+?@mOr-cA&+bpmEO
zB-ZzZi_`{$2+PQZ1dne-EG9;n_WSd;akU`Am&ZCkuzUVTB6qEbC`y=k;;;i{MA%KB
z-zg`+R+}dj9I@jYffERim9gJ}o6~2aW9>-kO6d5(;WM$VR>YShNW9-C7td?jv!oeJ
zRuv(rG6D{P;~PiM#F`>nV8Si-3J4(OL3<BWd;t-;B*JyZy`SynW~8P9El!=qGrF7r
zfkcdN9L5ohDdPFUHb>6HS1M~{!RB|!J(N<!kRmR;heaelQjai~K+mF1aHGh9N38hH
zIUJ#lGVwZ4IcydWUVCDG5h{L?u#Lz>neLQO@gAX%xI!b}1k;f-I6?)L-m@X;WE3DK
z&?+NKieDY~;6FK(xNtS5u)0%S{#+~PZ^Y>FghT-aF+hwSPZ|rve~Sbi1lYNR-ECs)
z-~t(}PI7Y)7^g=HCBnm3FSjIOPJ@K=PY-l+j4>j^v-L|n8;NMNFtRQGA{<Qwz#}TW
z-xMbt&1xqwHBsSB@_NBhR>_7Z8FGJlESY_KvpP{$x2I!V7m}M5xIHt)fnuW9p{|I$
zNl19Ig8tek(&<bnL_*k}-*0B$wR6%&Se6|jO@V$v+#|3=7DVE~lLZk-&enwX#KWPd
z3=2l)Js6qeTEN!TmXhsZGD8$_k;i6$$32ON`~-q0R~oRF$Pz;wc+!jkM<!JS;(NL?
zyZK=gK|ySi9l1=1_D*sr;C642fTUFdnq+F(N&B|#S40q5HGeIh(1-XTJEHs&0DcgZ
zpWq)Q!d(;Tz2G8&t<)mz0FX>9z&Wx`T(XFHKsb-g{tn!nsu9g&r&Puvi--pV?#Nit
zM0qFq#4loTM07})CvG6-MvsgZ1b%NEs1aPlFL?*_d;!GPnEUo<kU)+Y8WN#H3GdFo
zdaR5Zs}5Io<|$~habzbKM~B3@D6M9}MM!ZV(U-Fo;cD*on@jg$Hqoqs9^fgMU;ZX+
zMX&iRN(g0`Nwtzc*6p*G{zv9V;=B*e&j>?7a_NAC*7dg6h$z26g2n=gkGBIhLazSu
zxDpXa8@M?;BY=cK4cG^{*@+b~v}MWRSH6iA5m{$p#F&J&%z{FM_s+2y!5>CsmjtR0
zI<ev-!!GgN5wFB}$c!^YcPFAA5b=^-x6i11NvV<SgxKz+1n@`>ZDhgiA<ziUK;ZU(
zIyo)#hz;}PN+7U%B>*k~8+HnLT&*}PBQ%3Vwn_rKo9y+HCh=na;%db;CW0_{YO@3F
zDa7IMsHz8c;`0Exp_nDRN4D`;3y~v{xb6gnaj2;{;bp}PC#Hf#Qa7TylPemyXJiLJ
zKlbb0#7-DBt6T)X_~nrkdd-5;iD5v9gwju7R3PV~u-#`C+yt2P@CcC(o^bA5@&|U3
zv?ug@y#zuqVRD9ui#S9t0BSC{>LKg>=cbj%XmAb3zC_>}5iA>hox8w56J9lc-nc~T
zP-n<S?%=Quc%mJYEZIFZDPavpWDO>gyAKyp4Kh$FHb6yMtEa{~0qTL=wFBCkTJz_N
zYiht2VMnnR3g=-S(4h7#D1JA};~D6M<Qc8Nac4H(RKzO)YQKY|gR?N=7EC>lPOSFm
z&<ceJ#03gxI|5l_moV9hSuoqtXka+)1I-*eLW|Jt?(Yr^r5gJMM{$^rNH78P3)}=I
z;N&{sz%p^T>b%geZxfcFMc`qgw&M=#aF?D9EnG=~t?v`E7r00(0My8~LeO^n7lD1A
zPZEaTA8bTKYiuf9=A5m$vP0^(w$<|=%Y}=C0>DUwhw<fCwh2+tBE>zB$p+gY{|VOa
z_qFw;L0$nHp`!KV-pzuIn<?r^TqP^)xIj3JyvDHYo7kS8@k9v6!W%>j*K<8(2ArH$
z|MD0HYh;ykaVJYXUrKgobuwhAC&a4BloSFm$wp+BbIj9P1lbqQvRnOxqDg$;Kp{mg
z@eFFDR1wcpDSkp>D&5GRo8eg;VlAGZp9E9BA3@C97LY3^r(`6%fr))ZFk2Chj)Y~}
z$r}KazAxnK=5!=tP7{=U_Hz_9XU&#}e{uxO8z*cxOwO}NkJRG?WOwxoC~aM+B*b9L
zqlOJHUVF<khkO+7$@s7LjvNL+zl%+zgjVCtaCwG^1Hc)nUV)zW9;od;jZRVp5I&*F
zk=$6+l`RmDsc?*Fn4k&6bnGl>>Yfd>hgpc%&7~QixkPD?L~5Ah6D<*_5uBJX>~$mO
z2;JU~?hF*-3;;9R7lu1K39y~N9fW{UzjqVkgO#P?#mE+y@bs_yiW_7TU4Xa@SI%_b
zjZuf0qMUtzg6;*JCk24-4Y2X)Q5`<de8!%j4&|cF5eFz$ZZ;7H>w$V#cfv|8KZkpq
zH4(V5nLHpwWa?(Wp^@+c*xS(tfSY^(z?r?o+lhDrKw=~X9EETK>iNiK{@ez+`k+aR
z(ANZ8_j}gB5h*`_o+$~qB0(J+8^ha{eUw5Xsuz}FLQa&C^~s`ZuuNO%+aG7dNxhqG
zY#Ux5UDV$k3(_R{P~zBiO0rK96L)K*&H$Ro^`Db?Goj7R?myQSiQ0gh6ac^x76{?h
zhn!i3li&#0{c1H(*2+5jBK|*6V>^uK>W$C=M8${{Auvyft{!S}<gdx_)Eq``BEqP9
zf(6)!R3bnjvjuR*1{=FMZgm4C5*+|-339e0v^#;-nT_f;d`+gVF)-=6=5LkD)UX;a
z3^yji1;E;ffy9fcoeJ%dW0k{+i3s8z^`q^Cwk$92PMv2n6eg%W&NSX4#t5<g)gAm+
zw$uF*HIS5I0mmdvfO>Yil0PSNc))#`5au7IVYK<i6k+}~<5dP!qWn8tA^UJ$n%#%I
zMLc}p!Sd+~oA(*bbW$J6)*_iI+gg5e+F{rHxsp)BnS?+T&qU~~=c2|17dfr;d|zB|
ziGKHa%h`Z0JuouV0+qM`4kNjqpH~puA84{Jp()RE1q>a8i!F?x!<_mUOnPR9!4Mdj
zSVZ<eux0COYV>oInGH!Ju)lXn<AV_VcYBY1-Y)Gy)(VH~J|^kF<Jm{gev-v31}X9y
zu%*}7?D7%Tc##-f^*nk@U1BpwYrLK?1qqH`h*s}!H_U1|M%ZB^RIN>t7Clwg$o31|
z@i-|`H#@*nW3t_*%_Q7Arsc6+fQt|SXa26N5zTwXOdt6)%M%x$Z$cBAC4RnoGnQ&L
zwEbqZ!+J}a!npeR_(rV#++X2dw12u`DfkX0LHzwA-|;57y~5Rhhb1%u%(`{NU2U;_
z5A@`aE=*b$Y-|J`p%aO<f%`lU`Po)5Hx!Uzy=8r)gh-XhjPARj`^dVL+8ZY<`wVZH
zsj6XJVb{1!EiK_18qK6ifOVxu+~rro1}eK3F3eMmC{DRO#D4e*0+ra<zCdVbf1KnH
zh}4&ob$-rPtKWYH`VMM9sYeUa{6-0e-r2_YH7j=UfSxhlSGU#@Hb0s2bx@APVYikw
z&*2Vp4;Uy@auNldThFi=8%5hK#Ivq;Ig*l%-WdwmCMsFLMZ_>)=d(I9g2q1TKx7Ld
z_Ic+B5h!6M0R6f%aK>&t+r-mVYxcYKJ~W%4EA*4w5cvh=Xfj9dAnJKvp$D-5$3Qs0
z{gqIO?cx!%B8wB-%+qZqG^TPU8vD%ESHbS#ppuE}@NCt`Tp{SWj|=Krc9NSYBvyyX
z?GJiIJ-S{@Aoi#$A^BY^jLhCZ_9|dC)JOKM`%$8qtl95{Q!UK*)ejTD*X$~4KZ45l
zx65I@eb`1D2iDjTVb9y-u8*Uw8S#Lg99;c1oZNK*8G@r*pFJ8j#~b^nur`j^1@3wV
zm&1`3(3FVkV`eJzQ61Sa_vCf1=|?6*;35J)u=kM-4Zq6_tmSM}05|(w9?UIvtk`h(
zo6K%}L|!{K+)F=dbM*`30oMqNq~@F8#>qHpbJ21;FZ)bt4!LqTOx-<NGs&Lw*Gg(g
zn5WMSwvIWhu`scNZ(`RSBuzBLXUricmzO=@U5TgRs>wPApM3;p!69(DUtM-9P^05q
zG?9AD#g)ip=<zJH&;2ZGz)igUV{68{<&Mcdo`q&B%q(d}`!RhvAj!7v!#VzvCXHuy
zefgE;hya+audT<LUUF7DqL>dd%X$DoxevaJPhd>!$WaCq!t7&<YrQ|$&K9V8TC<+r
zR1>~>O!=Yd4a~thsDv`X=>x~X>;gz^lfbZeeY;?vgX-;S#rFC*GRcBXNg=@&gp2wE
zZK%e9D>k8iI;wWxwFoHv_z&8QFX`txJLw~Q10;5RfZC6jDIZL*WwQ5dpTzzoM;vv;
z&dUzP3@G7t8Yh38dUQhSgIdYn*vmRYQFabEH%PdZtnTF-V|I>5e8A7HRlu`XWjbv<
z3P&cZ`@R$#3^SpgEjsoalQrU2z#_pLhbR@72DYklndD$wa!acv8wUNi<enP|l?eK5
zh~;()$1QW89YYehW-|*R;V^Qr!C0ky-%Vl<fw1K6S^*r9Obe*EG^i)0*xAmWx^sI^
za&*i>s$O!2i45E948usW>u`DMc!xvgQJy%$DddI#dS>r{?r`CokaCO<4t5$r^F6F;
z<k<wWy$$POU~$?rNOJ`&b7te7*_QoJiTA*9e%oiCG_uEY{?q!|khNJGl4g+%<!l*R
z4@bf^aB1-9u`L@Bzi`aXtr^Z)C0hq_wsYI<&i3oZdUi)O;bxnOEn*FOkJF<;n0$|O
zyJCR26oN+su!YYDaB<MiuJev5$&C;~<ikd)CtL_2LcZ@==SXtSgv|4Zyb_{8I8asJ
zE|-xjn?aMU2bS0l=J3haFgBXQp8>XsjSsvZgm7eC9ZuhyJ8+S{I=1C^{v7?q_uFTl
zuy9P(2#n9J{$95s@%zQV5lM47yc)=Bk8#{C)I#<Ljt}k}G$?4Hz{sJ^*_9iqu*`xB
zR2K!Nfvz4yyb(QRo}-HuZ*Bkoti4OJCAp4e+0Rk<1mX$*pAkg6q#8AVY-BYuN+eo2
z{o>jHc&;7Z@XUsbyytGVoQJAPrIH{CY%p+DJLJF}E8xu5#Aj;rr^Mreqf%h@;}GC3
zb|gU}_4xUjr|pb{v!(2?XZbXsALg4aKt&V-!*hwFuUd5swM$rgIu5hrofY=lSz+~-
zbcVHj7oNkgEVl+}VaG+*8GFUZ;25{FD*^Msn(xy{orW{0?YXfyHR}lA1G}4)xK|QJ
z46_{+(3DdHhg!Zb%b{82)dTB{`U-jK?MTem*c_yFbnll=r`$B+AWc3E{9)VDj6ct5
z_lQq}Gjye!oyU2qM&L&B9vCT$^$fTr{XEX21<1^_OvE5AGZi-MXMyq9c<k7rVXHf;
zTqq&yAhDGwxVxPXo;luD^7Ms5-!%buYYZK?C63%#**4ogrhhS*+99UDZ2D|M=w4d!
z?pC1QEuRLK&}ND787N#B<{Rm-C6@+7>D!1OH^mAATWbN&qND5hcKVSnY_PR0`FlTn
z&=DMs1@t|09l^<=VJRt96g&g*G}y|BDi_JO1e0fA1S&+B?`P!OY?~oLlK$zo#c3>#
z+}Ypo9IK4a%}##L3}9b8BYv`n!KEe#!)#rW1WCbcK+tF~ZLJAhtZsD>Ze4|mN}Ca|
z>R*HAKB9s(>UdUYoxpMYtz-!G!d|8yxJPP3;2C^}j^<~M>kI`1EA2wMH#8F)*$kM=
zv-1K6*%~y8&K1wS3g<~cIlnnv?$IzuzZc(a4}(gb4rGx<ss;x3RcyiqPAnE+%Qg|}
za1C=77(g+=%pTnj^)U`vk0`svwDCfxXw<^oBaudj)7%+a^)(x|ZgqDCjGwrfCND=|
z_iVP9)OZRCrX82D#WAzym~qo%y4OTu6mU4r`D;pkJJ^;l1MVKhW?B{}=7xmViRTH`
z8VFwrrgkNzb7(FxlZ_iQ>{x$Q`7)5L<f1d`W})i)*{uAKSCgm8cJE#-G0~|v_mvSP
zi(-9&Y1#6D1#UHO9CLJZra)gH8Dkt`-C!iSdzx1I2-M4P%5mW{+;=gw5aJr6s^gLp
zX0{7ATbn?hz~PkPjwLs1JoQp|uP2{r0d-WcBq|kmxE0$gb8*hN$*p~BU~;ELZLhHj
zfw76Nr^Cz^>R)D-dXuwz$#NJ2cKWf4Z4WOsd6rU~!OFgv{F?UXLRH&KK&7WKh$@=F
z>;^J$vADs|9{a+MvzZL?7+D$#M~3{Y7NXY9SiK51w_;HFW^>s*vSO5MS{%d3Y#YB`
zEK3BlZ~UO;jcqR<uh~CvAj$)*c4mXT_ZfG^&MMbOi_ZD_)?IdG#4wmKcLdY)PSn|P
z?t;zzKVfI&HRelkguM%r7p$vRwtXWYGl5{;Hm9QN&4?<0H#Bb@c19(^elODpw*kxb
z-VZW#a*R7e(0#@;bZ&zsdNuyY%lz+_FCkFwPIlvGj%m>ri(d;S-`qmXyO3FdHS5T>
zr3WP4A1u51g_3upJ1jYF`oJ_i54o1c{j#Z!p=8iL=;J!aA>NYYyH5{;Y$D5Gb{`XT
zyn8ms!!g_|2^sIqF^bZV4YM49bQ3u<u4ZcuRJ<#3o$_$>Ei2&W{VC&FuJJ&lGrKQ|
z8TP1i@=mq$;O;fB$>}G16nF?ZMEOtgZI$Ekfaj$MmL1a&K6-d8EX4%-AjSY|&r_7R
z+}Rm@RWOIcJj&e_b$MVJB}(Qq{tBirq|xn`+5)ioGft9+EOQ}E*<aAqV+=W)4nJx`
zwYw610rUjfu+3JWHjrkHYoDRhoLX^XkL!#)Spnci-Qv#2WT6V{F&9m;hM1G>Ja0L!
zGdDidY;(<YpYcq_T@JMjDFM&TE&1(ssd+n`9M>$%0bw1EMb_5Ed+N5&1C!S(TXa5r
zv%<nP5mfxV6`AgW)`0f+JTp0Pk+2@vY~vh_?lb<1l^rJBk{^V@^91qhV<1}do+0<b
zuJK(S$L@gyTx`q(YmBRl93fTkbQiZhFnG(*n@C~VEfR#-bDyc+=OgUuYOdCMdh%}t
z`7j75ePq5rKX1=mtvQ9UJK7xCn5*?dJT4%c%&-&#9wF7uT&<6Z=hFns5q$o%8@y7$
z!YYGpW2G@QW+Q}RB5Yp=b_U=rnA~{^MQ(k!Y+U-L8Eo<2mtrqQz|$X?PuSe{xFvMA
z6PoeB=D``j!8E_*0Cn&Dn8QP`vLO?25#S7rU3#Rr&j>6SeQ1*l#5~!LEwthh_rQY;
znB4a}BgKsqqTJ)d2oPO-N;((tr6@FOuZ0tYh~0s5f~qDtpT6CJ=p-Y8jv%Ths7H#s
znk)ZSN`Qc84Vy)ayU;?VEULqfb}wb4Gs#~)@iuJY-mtWrSZ<wz^C`|$bXXd2bI7Pf
zv%aGNJtM@J7GK~W0(7YhL)%RE2G+zuM`>;6xVL&13f;mM#ws7WQHqD2F~slTth;)%
zXJ~6j4Ma;cdNe0Z$<wVvC!P+_oAw?V?utt?CsorIgQB}vqqXc6^nArbt$1eaeXn-z
zU>n7I<<j%{s?ACs!FsiT3ZfD&+e3|bKHPE(j;@Z)4&iQ>+8oIaVfy_3%-~i*D!zpd
zJw1x4hSh4V_x{%=#LRLg&r5gbID5%gGupf0NxxLX8twooI`Ml@;L5Hg=vhoPQRH#j
zr5o#ZS(nH`PhUxjbD7!w6QO6OuO1~c@~xN9nkri7r{30BjC1z7_qz0qOhq;Hk>|X0
za0;OO)%P1K#KJ8F^EA>~UCNmru!S&6`InOP!g5OY=POw&*6D}Mf`eQL(CWTwP%~EW
zO&;qO-P+aB3H6S5UiG~jEqMma@}lR2o4}AuN3&R=GZD>OK`FdcWfAwyRndi~kWSNY
z|6t<7mlrDcAE!+_wVcuK>8aVF<Jk40BYo<{;_4PURX-2W6}&bjQd#cR&NQ$Tdxf;x
zt#>9i*`c}ATXt&mTy6%wMu47vOWc}rw2u!x4JIFjPt)v1E(DwzM)x|gn3f{P*h<<L
zj-`Tqdyo(C3{$$L?#oB9MH+-DKlSJYtk*b7z~^F>F@shzxQn(?Zw-~)=c3uIG-y-i
zaGX0xJb$S^A!xnd-7$8Rv0A}z`E#{FvS(-sk84pSN|txY9=Bdx4Aic9*Lw?7%X?s-
z`&PTiY%5wyxlMB>wj`SF8#s`3s)Dgsosn*-_%cGX2OfyxQH>wX-0^{Syl%45)}j-1
z`4$PL72ka)Xl^Q~aLsxzCDR>NSG0F0`$turB6);v(MFm)=)})8D1m4SUpx)8Z`n?E
zHU%*5xo^Fr4vT*rqeh=0U}@0aY(}on)j0YR?Yq|d42xt`qJ5291IsWyFS%E0&`la7
z=$YF-O|!`$Bua*Oqih9G-PBHHwD=%~ZmWhfb;j_cU5BEwr|IG=AM!*K%rib`R&^vn
z!0Bt%)Ph!0b%?gL?!_95HDGQrrVb+MhJ}%?G;h(ar~!GMjdiOXq$Lo|cg`Jz({+nZ
zogO|8ov7i9w%+d5CPMO}InVoK=0%H#+*~oY3}WgC>?Lhb>-TCK-EYyHz1m6W4w1bi
zr=9UZO{}(Iv|T%EYI9x{tREE=G>(C>+!jP1Sg^Hue_v7770vl`ttz*6#qzGPf;LXI
zfvtZk=%}HJ=5l}E2ho<eD6aImEuBID+pM=z%r+~UTiB}Hqee>(>q@@YI`8y!c&IjV
zgQ7MldRAQb7B)hC3R_;ce;kU&ooFvThkR9%nh><O8xr*V-|n}pb&KYKiv_Tcbm<xS
ziWM!_T&5FmYp}fURLH^)RY2%)4?gG_sk3T8i}%)hjMt{OljObfhuRi&wD`>U?yY34
z<+9PO(zAq*S$FA8EAV^XPnTdbk{`%#)*X1E=hM$d(+neHX_Iy;o_vnay>5lzLtBq-
z&xp*LL7uxU+6%n~TrQ@*pcn)<&dg_|m*?;OeU{}r^#$Ia6FjY-kDbc>dCja0p?!ZU
zsFTo1=;t+_iBV2p_$cpniz4E$*n1)|ZnbA-o>ug(r4;F%6g}V5h)--|Oj<<#baxlv
znLLJ~?^_%9_mi0tmPfj9;pBnW_3HcH>CS97Lpfx)RL(Qd5+f}-+m&!QJiK)#7j#Qn
zf{o(g%QkejJ6|^R3@f+ndEHwrTatTDiB6(I=lCon(An0}6R>%Jr>Jb?sky)xX((+E
z=}QnA9ODT}+eHEs%qF|uvZ!MijX4jY?IwwdA(yxsluQccFY}xC*9aXrf<arE_hR@<
zW>Md=5F{1`|9syrgl!yp<g@P+CTiL^t-^}y5D3!w<?w%>i8wgvOmz2J4T|TN=|992
zm?ibWw8Wlubwdv=Z5^!=a<JN`-a42yrt&q`TlUN}OE-%taPRgoQ=4Fess}yRZ8Kqt
znb)NwNpFHp?ix*qw8%(SLxc}P&qQs_-PVlmr4YEG8hFX~&d$AWD_i|U2YQB)qF2*W
z7!%MT)d{&9Wq)G1IxW>urQjQsj~H0o+((ZG*Q%*(nyUsZ1PlR3EG>XT(i6m&xEnI`
z7-H7q$70R5?6Hx%Nq(%6dQcM68top%2JM+ghq_~};6x$<nEHpF34f|-@ge%qjQ|O>
z=W%cBG)vW!rQ)tu(DSB;gi!d@K6K)R51o16+udp>Meg)|ueRm!Ftte^BLZ82$iNf&
zvrBZ4D*q3oV~kstJzLeVn(JUdPv4({x!sa%U2Nd=2uNV<>@zDI$s>-SjeSmF_3$3S
z$s9Me(Z4RmX;+sTD;^6sU=A#ReSc~tJ+Ldk#5l(mG7cO-x%dQ~!ycgDX$g7lI;_3B
zhXV<6#3fi`F^Rgig8ZOq#`h>!_O2#uD@V2X(c{Mf?QVEwx%y|d>uH}D*9la>kwbA~
z!_5oSwY#)c&T?;+&&6uyxE$K=SjCDElO9^#6$IL<J;Uje`(CVVdtW59C;|JhN=tVK
zP2j-QWGX78UnaYIDl#LwR%6&pYFKgv-m=Fwth#<4dEByRVPT>?xz1aZ<TO@6{&81l
z?k=LX;VH_wl|)upamT#GH(0E-aT9Kr4wNdUUQvv59V3qB=G-;*IiZVCE9guyTQE3b
z)<E0X_ZEnSwHiwO+*ljU)Q|43$*yBMM{|r@PNx@Bu>keLp$`~Z-DeXzIapA9<cOZt
zIR;nBxniA<pmt%JhmQ1t3+Fvxjb$f#+6@;EXznqGG4rxeBWiQUHWYLq{0<#{5(dTh
zwu`pt+XJ7oY(?*kFj&$xwd$EX<}F&!;H}^Hx$7-k2ScvHOYhUN4kY4!T6kx_lo(sl
z?Ayx7X*bj>6tkMjEjr|Y4IL}rvP1eI&=xy{$Y~RUC`usKr5qvwQ~6Z82udc5cBOXV
zDLB-$wXaXhaI`e##t_pgc06JIip}jYLa`1g_QE;CZ9#KK(+WD>KMBfgruIIQ$No^F
zKt#(90|#v<wkx5@<!?lZN`U4yV`7|a1XR78>_&3d;%yiUFD}u^1~s(Q&mD}~P24Mg
zEWAkwV6P@37Aw9(k7c>kw8TH(tDz!hf@3X>d$l#~jVCsO11irH(bap`+qN6mO3@jM
zos932Nt^AyVu)dH08I;MT`}L|Bt@VI^kWe(#nFi_oMA4-+f8)3MhLWW3MtxWzGYiR
zoJC>dcm7}%i2ngJ&*-460`co6?tHs*wkW<LqBC}_GW$npNkIGb&4)v0!7l__=cgIw
zLf=9`?(0+UB3MOYa67i+GW%uOCF17(+}TDSNRjh9Y?<xH1<)4l$kv|S)|pA7=i_h4
z_B}MkuF9d>XCg*_MRr=C3{u4*v|I8_$jyqjR&TA9BMMPvjda=?j$rdggaY&!TXmQm
z-i&W#6C!ARsiDaf^bBTOFQ@f%iyo__SOxW(Lg);3Ywwm_gAU5su*#mVEi4I?O@_5M
znkx5saSBrnHNu-j7tSKk>TxK!B*cS~)libxP?BV{j3#$56CKtWiwBRxO0o7jtHyRK
zE$FHOTNtkgl5<Tjj1b!Ud!26sbMZ}Z;K1V*;fL*_LeO4V|49yFUbeJEE|Q`vK}SRU
z0@}h(E*$N!Z38VW-`hGXFq1z#ruDcKXH7lpu^IBc+S=L_;lL^1YRMXB`iVVGEKAm2
zpV0`45S`kNWZ9B@wNT`QUyUaBV-Vfi2t(_mdp{_g=1~NCy4R&u;q%b$Q4pdiKuD1z
z)&&Qe!ePeI;d5VS0Jcg~{yak0K8FrxSFn-WbOFhHuTIF&7VRH+u3dDY<$8^R8$}Gs
ziNyF&_!+ws)AStgrFPZ;!P2qsTXsvs6B;&dk?_gXtxng15!|?u07K*uO5Q%BAw>hl
zO~NS4n@cAVnp2{5(YcEJ?Dby8jNt^j(Z-lMBpm_IVtiQz&6V<?=M`n6aR3RNI)|Wz
zU6%-o$fRCO4-pTjTIFN&K&1T0*>wFBTLp=XDWSd3jjl6r8?Ek;3_b5Z>lK^@Q=}C@
zzmeEnNzBA>S~ST{-NG67qG6f5@hXAN;<gkN0SVJ4^=+|&ies_XZmq<3MEoIf7SJ-w
zy%?cNdbKUwgQ2b9{D(@;p8Z~|aoi%x!CcUeR+{LDy{3X8iC^-Rd{B&0n5XP6L+e`i
ze(?=8T3WgHJN<=M56Xv05M<gx{e@?;w(eb)OJ0g<zYjg2*g`9WIODWFvy12lMc-7k
zx)>`o%Vz;vV*OEY_nP*7<FU~%9jIp@Q^aX$(0ayh)o>*mM9-6}#m~^0FKumEi|5{M
z1^o=(2HWRaq_DK#moe|}@ftFkZ?|I|h_lHwkAnLQvnT7$=^hXjwhNHam>NXuZJXtd
zKpWqpyCxeXIfvPo`e%2liT2V?ik<++f*LSqRvFOL-4KGx6=N#G$xXk4&||4a=PtKw
zIfJ((ea~z5(}scqA^s-N9w&eWu}8w^r^K~BW!pql16t;%-t*BiFA-+lqT8~A$1+2@
zR(V`HuxunNyjD!@>>_fs4xQuE40AF?K;wY;z{R&ANF>~)n=Mc1J={F>4D5`R?4ZV8
zV8QK|^Y*D=t>`x6UvjGfrX2GU4#s1tT8!;m^sMKT+7Va3Th`QXlcLi~z^XhZQ;r?!
zRzMh{(Yj-9#lISz4XKEEjGicJucGoPcgo~sG~L0K<sfze;$+*^B1I>TG0+u9>V@Ut
zcFpH;zh7L_`9P<{r2D-Z8oqkA(b}zG7kuS_wfd?Dy4d7~mK=j1+Ken^C8g$?h|tWL
zH?;3Nn{&3rhB$Q17@0VoDtZxT5imY*A-Vw_Ek<Xg9LqG?%f9q1aJO2ww>T|?5=bXn
zV`t>hUc8&1S+U-$DK#xHt#aXM5mVD{I~GHS&NGm!g0{vFIxrlkbiVVw*pib!^ct%~
zj-bB*?Da$U8Oil_TD!9}TEucXN!obpmyTggex|edq!z<pH+p*cxD+`DlLQYpp~8_h
zD`)81(L)v+ZmNKF*S**=b*A1~4GnMTnU|7Y#A(eK$2g+*Yc+R*1ombl$XUzhThan$
zp=Eq-YtXqxV-a6i4-$qfQNYn|%MhC@TK4$d7LuYyKQ9NZ$W-+jO1L|VPC1?=9Y7gG
zED~7`U?0)*QHw;ELt2F+jMy#hBE7PrES*kEmfwjY9Q!Rx5Vx9!$go?s>J$QsRMVv!
zaSr4X_w-$lLR4N&MprEE3(h#z;62m_qO2)8mhXNY%0{!%zSF&hK_KLm8|E!3mm{$h
z<`V8noQ*(hTKD>=qu3mEY~}dKHVc5N*Kl17fxFujh02Rvza~N(ZHhW$iyw<semAVa
zaR;xL;&>@KyN_U1t#2LdyiQ-|K_yxRy=&5{-rDqXQVS2z$_F#8<EOfPV1v$O4W6od
z4nw>2e4yp#1k)%t)QmVOUy7@*)tfxp8#<9`k(mFE!w={rXlZEKyzUb(LZpFn-RIrO
zat7?!ZG;iHEar;wx#9|m*NL(Qgf$ICP(ieLv>7xbOdwkOxMc}ex3bqJfDVE+p?U3F
zwi0q~^fs3)<J0=kRD6G<$_}3@d)miMzKYS0iigBV<D9r^iBa`zr`xZ4wZS(hIy`tt
zRC?~BpK8w=F9v%Z_aT;I6``)xEjS1~0iJ36(p~J$pbCB)_e|Z*X?Ri8B5gLo{K+hq
z35gO+F$+KkdLC%B?$~mzMPtS?8e31$h1(=_g%edp(w|7ZAs4}U>qR!e=vetf2hSs#
z`mOgYJ(FI;S;j_JU@HqVO9?1qKE9dIhFlNxt>G9;+v%r*iFUQ%J%XvM4iWc&=CA{+
zG)k<+XNv8G%3ddj38FbZe|Aul<<&T2mfL2~iQ5Tu#YSc&)F2y6K8nNGh5}BPH3pV$
zz1D+pKoy5EvF;kIqQvnFhNEPafKIcYheW^V=7b*_5@yhjgN)I9pNWujgC*>*s+<#+
z@8fjiap}aj!Kg>si^?+qn%ae@b9TXgzYgMHIFS~NtvBfG)+{~r*N^G^FXE|;&hSZx
za*}A&#qXe5$8bYik<qz0Y4bcLy{}K%q7x<2*eR<`M+}0k$16nj->+5UaIbPIiCm1R
z>q9`<L|B$bWt3ZMM^M3qmG5m0c3h2yD>RhNO3@TvBb9X_ZYQGAV_6n%8!3^KzD0?l
zP|<ex?<(s;vYLpxM^YtLjoOC7=%XK^Oqr}Fov->sv*d!Iqu&ZLE7QMBMmpc3g=L-5
zaJ8ykj<QC>!5KK%v=hwHh(rVyo0nRREhXsk(JzKBd@|UB1+lRbbR>J-s0a27b-FF;
z=^UV}r}es8ioHr`Om^)qDl1`^&6>Mvj&r(WpshP>dnkjH=xXudk_^B<@q3sET)jDL
zQ7t}0xDY}|3`X)fvdPu9hJub^(-H>oEnACsZyfo`rP8i}w)v@{H6KxGi}Uw7)By#L
z-CZvn-j!L7<)G$W11+Q8Y6-}(YOzFv&bI6mLpgZ{Woyo9+8-KsoBF=buqEgu8f$Z?
zWWkixD0WUN0och}yeJ6ADcMZ)&Tv1!WwQi#L~p-G!7|&-gvI`$X8`X;U7`$HJ4ffx
zG1{%SWtPdJu_lCSuMpbfT)c+vzi8|aLU*yLsod2}MaZ&dr?KZ(aZ{~k4RF|YpLovd
zIiG6%)Ud4Il+Y0t(;<s!ySH_0(XO|Hmw|T>oO+wXh)e=bvZ08kd1;m=C^U^2S4kPr
z?ei@uHi_$!<EUEvTr1`g+@Nvm5)IF1=p1GXph|h`ZIvBz(FkdQ?peHSO1IwL*+zxV
zK5q>}A*!Oco0B_WWvgLaBNRXNWfK%aZnuVGOxdK7dvsDICoqed>@Ynj*>Rkvac^u-
zj-R0sG@UD}Li<dPRTrxU#71n^%W>olQ~n)6mn4NA3(;W;8`%20(Kzv{8VYQ7$|XFI
zstt%nKVvIXLbll{=k}?gD<^<xnzyVSPY!8v-a8nrEQhxVF}bRJya-Ty=ow*}(b(av
z*uI4KiRRD6&H!Zd(=qA7RN5ZdAnIMbX<;-R#*sO6pe<qv)O=WKtT4N0RSW+U;Iik!
z1L?xjY^dUZtpq%aIIA;+r38^v5$!JL0ky;kqTbw1+dbRiEqMBqO-|>n`@*p)(^_}x
zdX1uET5Eq?eH#0L%Uhe3(xqewHtLczP-4QJ<_pgVYqL`->jxbi)`*7lVPiqKhtu%)
zgsyEPC3=iK==myab_xT!X^Y=%0S+72r8C*&bS8#qSq?O^Ll-&`+-CVC7hZ#**tJQa
zcZ{@jFV!<vwrKb5PN+i7M2qi1G^8-`K7p>=2@u1?;t+N5TIke9xM&@>UUrMUhIF5R
z!zLyC+pZh}+?A6D!SI`gZbm%OuxgvmXNO&DxNb1?e68Ecj(&@tucuUl?)=_3w`P|^
z@FKiGg!+v4!m&I3Tr|$=8^`%<mx;#K43vWzU|XL%<rRtbTdvcTV)b(&oiz%kdVf6G
z2LaaUlL2tD@)2$O(uHAWw3H7exrAu>Jn4Du$*xsKbO1TgRHyezSppLED(c2Ula}W;
znB4|x_z@9~Z|P&=3qTwbj}y^|lW5f)6hvb+-xNS^86S6;K@DdtN(hd7ug3_Q4qfV7
zbVSEa;e*leJu6l8w8g&FYQt4<#D|4QCW0iQ;p^3^LcE+tOa+v~72t|c$R_8%WpKC|
z23%;8h{j~n@JejeY4HV2L!}uHaOjMM6?CwJDC#qMYo5_VFangIt6O`YPb?mRlmy<l
z*G#w8me?!+jxaT#+UXPZ44Y6wt}Pkoh;<26M-8IIm*|bO1PGKUw@@6@v?lRKPv?2R
zdAxy<C=rbfkw$ic<rUQX2s-x?%b%wud7ZtQ@Ih-V0=>$+A^?u<ucmdkLz{$AtO=V-
zqY09)%xFYdH56EqH+S71N`gdtGuM3!I=y@2FLCQV*T`N&oESnm{WBUHx(y*sx*D{^
zNn}GwlhZKKaP@-91yD5NSsP-S*gm4Q-x}s7{8Thd_eS6YPVCSe2N_La6Wy+s!&w+Q
ztG}EayZDXBjph}!Fd&c=!W&C8R&0%=1<}PL<q8xJK-dk(gw1YK5>a3~XqD21Xm7V{
z-8j~Wx?~ZQ+#y~=-nRx8HC{G?iJ(fpr8HS98cbO=lhd(o4MV2~qP^T_AQ__5+HbwI
zUNxHcExVNYQ8ZnM0D)C-gJz(cqaV>3UzwD+ayHuArCFISMDM^jr2=IzoJAWtn}#B4
zvLP&H-3wotTds~|CcvCC-WM{PID3hXdmR)tw?VhC8?k@6IgC(=Hs$FzjOI_RW0d6d
zz8SRJiVc->L_imG6`PsPd#kqr&H~+r$~nT*sG+u5y|on*8FY9DC#cOIf`kUun|tK>
zp{v<oiF-~BiVB1^cV98TlGFfY6VdsBq@FZ5Vpt-rL9bpxi?LF{wDt_WhSfl3gKm@4
z`6|k-Rtqe_^tpkq)jG6!?l|a>Mx1CwKtR{D+3CEpPa~YaR4*IOh)v(2SzWWWTkV-U
zymy^X{}yF|Z*Ql$ey!+=4&GuOl7VlZ?z!Z+G^cIGz^rittO4=*7mQA$AapYZIGX#=
zO`ZtRwr>SUad-tax{`wChrOFx#x2{JD+P0mUhIuBN{jlU9k^}unxDa%&1@Tb=8f5I
zY|v}H3XKg)=!_0tV*93XP*Ti#pQYVqL15O4)^#6tYhO~dc$e!OnBiF^8s8XHs>Pzo
zr~S^!V>9FZjI`baP@M*dMlgBj{I>Z>nd9T{oGW)9J>r15o4T!6&@XJYd#BCqQ|oz4
zK2GNPxdY4sdo{WEkl2&b?s^UJjdm$}SR>r!dB<18zr_@BpTKRi>kLD)C)p*&=DxSO
zHT4-yx6Z9Z+M+QBc6^a=IT4NBYv^7+P^Jjt?koYf_l{KkQ|c@Nx7l82^z$hQDh)r;
z`>;rLCLb!3Cn<#<=<OZpPeezup>l*2KD3u)w!qk~jwrgkgLC+IYZ!`R6!ioT(5YQ6
z(b%Gdj*`8G2qx)hcG`r)q4>Rp<?T53r)f5+oelarrc0E_FBEeqa8#R_hJ1OQW|Ht2
z^)D2-n%Q3Cs`zvz6Si@3gvvn2O}g34G*+B2K}Szs!ECqMR=Sy9*MjcSj~ZS<oV9jb
z)4Q!r$LjY925D3b#`+LSl0>8R!dZsdUE{tik_Jp~>Qz3q9>dqHv9ZFlrIR6HPQO-c
zUPGl+A)2BF(j@qMWz+9MAEt-Y9UNxq_ltVU8tCm=bM9Lyi_Puy!dkX{b_8IrOSgCK
zL({Fbk;lhqkB^3MpESIN*wBHJR>o<$Wtr#gCCvJ{+91z~({w4@zV;Tvg|(yVS+ciG
zIA=qZ!Q5!Q7ds2_>8%hp7IfF-H6%A3JJO!j7O7)I1VvZqn+=p>M6d<ar{<eh1z2kH
zk}KWc7avZ>2+!J1l%V(d5q@GF1;d`4#^=!_kMCEy9hvaQKD=nZMM-=;90ofJYyY}O
zIMa86I!6zB87jZV!oj>ZEvcH*bjuQH;}2Cn*WW0(I`s<3p|FxykYk=w6mAU~-vME~
zYxoOD^OhxbchKlXU7||X*jG!SvlQGtXub7TpFcS7?g14+Xt3YA?QX9enC{Q}@KLA7
z22k%FJ7e^7SLd_#=2nUwr-YyT4TJ|zm$s^Z%RNs1priO8*+cB7qxo69!F!yFKv(fm
zvfr-`+0UB1Y*>gL-OpNs#-ov2p$>mw-^J|1Q@0uPgAOPHD@H@NrK6XNEpIelFZmUk
zkCQz@^PnS&`a-gkVxjvtBWJUqgl9RumHK*dTnVC1j)hUYWmCLG2@*%qeD8D-g)Z9m
zr-HSn*#IO`zn9u=`o&rdFsOM%_iiy!p#`$MbF@43nwp6JKDL~p#C;l#liJP_aGA-4
zhh+;|2|09HKXhy}-O<;(HJsI!xAxTugYM*`WV@XXEzs1WY_ts`GRa5D)<4fS{2fH{
zPuQ`;YtVX|X720{c8v|>K(n`ZyBFGL=nfb&9g01r1r2ipJy3a>h=v)3B53tNvPGbB
zM-;SHe+~7*A{*9V`A}9kPJ^9pUYwOrrO4d{jlLm4CZCsE`P}U3y*^I1aF&M7=-n+E
zo74;4ZGe{htzamSz-YX1ty2T$TkFcvjpFnfIEOc3x^?b4i@&V!B7pAb#-X~?IrNz`
ze(y&FUFUJO{2Frf2+F>O({bss_<Tx|OkhyXES#20&x*PPO_#R3y}j%DT&>|i=5*Zq
zB`rmF+AbX_wV-~^03FFz8Y7mQUD?iksIN1=aU9UH+qjYZ4SMWWpO@<F_hOsPdZ+bX
zOvzXc`cC4&m(oAx8lUY@8gy?jH|7T(sWtnR!aZW(J?oQYjU$zztRw!$Yq$*$+>L9r
zceo8hSG(7`MVl0{1mg&5;33}J45)qH`|6|n7hYpl8u-a~H@RFoTyXgz%7{eb&@-O;
zq*OzdV-T?58~Ep6-ff{{xW!dt_#Mg5NsP|h3Y>@gOK@G~^Qs8K>`{Y%xr0m%a|Lzd
zU|3X+!*VM?S~zIF6_gcC(*ji=xT<HZfSu2>1f93acE;|Y<Nltx)L%kaUC^C!&|CD(
zX3a0}ISrw#vxDY)3v*_--D_yv3?c_}bfkzbfO1ykbltK9J_hw6aUd6ZdvcuWLuc`Y
zY2@33<{Ar_TH37vuLOSyrQWiohv{5B2{ZKE?hKl5*%}{nNw?Nkn_ntijey%H@tS=B
zHHwlEI-D1d9CKBT^G)asE4{iDH<U1<OD{a5cfj}<2{Io(D^<_D13k?>;=am#sy)XJ
z>g(uH73}Kmoqe+ox@W&M2R)O?@|FpC)`5EU%0btM&J_)o;$uPyVa-b~+}jtYHCz}5
z%fJ}>QhfTX0(X44Ax|?TCb0vSfJ{CP>%FT<s!^xq{(j!3jZ^iZJe)Bzh3_Ji99B-#
ztzavem{YGo`szMXmK}XYDOJm*0qOL^VYH#L)B4)9>}W#-L#vxzB@XKcQZKbdS)at9
zBl_|cg&5>`XensFH%7>54jK>xKQ7j8(X*Ox`@QUYx5GE<4y*U~r*rkF^xo@SJ}s18
z?$1{nQSi3lPQ&eh5}J+BF~+3^QBoQ^Dvj|-=?}`Kw=BYD{t7%n2ehQQE-jZTY}%+B
zt^<{-)h_LquH}YLiLshh;Eu*#^Ioi{T`z5yjy*07NIwZU?vtUASAy1{Wb+Q%@7<c8
zl)+oUS(dmVP*%T>yEVEPZTA`2PmCI3phev+`*&ADSYr$YtORwV1<Iax&=IuMvZCG?
zz7V;qIbOtt1U3eAqjDg#SLda->{(l-csuvL2o;N$lWzsbF1alpUpYd9?KZie_YU$v
z^m6HiWHWb|ZnX$AUwW%u+r3Pn_r|vJ^NNQ0+L)f<)akf1o;GBx2iYKIy(ROdB=8Ly
zuqY3%!RTNlgjjA1l!y<3&KP?l-g1x;V$ECh!o5@6*5_KiyQ2y8mK{o3>GWP}%)KwI
zL8p=^IxRt0a#xcP-lM>gR<z-`x3Cr5=k%7{oL&d@CqpVU&r3_toNlf8vQg@#0r44(
zmk<^-LZR^~gv#bXC-nY2CN>g8-0mH+c2JfMLGMj<CAo5X;cE4P$a_bvzC5!dihb$4
zl>L>U_h#{Wy!1BKlG@id-DErq>d%0u+Gz0-!f=se;ce~~Fq56&fos-)VN=Vacw9Pz
zqIdQ}>9ITQLED{LLrqOV1~DA{fI6~uSOSi;Z*$+0wyXOEXW20^%`mDJom5_m#Cp7$
zavxwLSo>1Ry*jn`rAWKnexY(Mb=rbvbUGheP`Z368j(S3P?ib5P<+#Z=Aa10zqBKa
z_tM$&r2)-<_KP8D=FrzB3McOmBa82PDLPa^XV8s^(@Xcr`-`GtFxq@*mhg_#e(7{C
zlRs#Y=*vU*9GAABteD$FTXrv#Kj~bz!9TRuwQkvRsgxGJDRPT1o!6@U30F$_q2|VH
zFxVy{`+*~^FNosKr-y&w%58naO#G=sU&J6#luG)A68QMg(__akl#siJHf{EO)2;+s
z4`uf#=n6UrHG1i4x4AzQRqvOQqxl!ge)>cAU0eKw>%N76$gX|5_=ggW`U@R*taE!?
z+H>>M<R2>4nQvOdk@gST(%iq$aOC`h_OkAm{)2`O;U5%-_?J$l+Ir~NcXImBrM2Bd
zaXNbF4u_o|6fw$lrxLdcp);44=K8ta58T&K`@q5p|3hmo^(|XxnGbBW>|ZDcn-A@r
z3qEvkuJ+Q|@=GTNIxpow<e`hhgomCNS%0B-C;LOmAklYu=^ii4{VrgC&~kV0{-AT_
zm)06*wtvt)=L1_`59@iAU+8DG{DayC_Cxp1@3ly$I3Btb67^8b>xbg4{!Jy>J=#l0
z+AqzR6)#=kZ}O8CCI9}WmFeK645bfk<>|U8ht~8~i@*PV=%C-cRGEXn>EwQ2x>|eb
zfls3&!r#*`boBa7$sy6le5nN2zUeUqOM7TZqyIw3SZ@uLIKh{$*~&xDqs{!#o>uyW
zj#XdER>lk0+7CRdnf%g~erW{c4c0?<P3djMO`yRSML#r+@z9*p{)Nt5`a?_2{TI6P
zdMT$D?V(5n)tAmPUy4lJTLbB1%0ua6`Jpra(hPI9@>Z~Fd!J$Lr|%O))6+Lqa8!HY
z?Bgw(Yk%e`;wndZ>G?{Y7FX!*)5EsZ>EE)g&h`u4bH4X`77^*CBd=OtI@fq9nVZT>
z_bxA8%Fg^vciwN4?dRhoRabj@xsrXQ`IgNieOGIl#mX;q<vt%;TV5}1spPi?#;g9&
zUdsH1u9{v-^7Zjh(q^@nj=JkZS83%Jx>|p0IL71mOtmCI_dQc7+1^tDf_qPkq3t&w
zhHTP)Xnh7I;27Vt)RIJFek*9owfRj+2=&%`mT2j%_gw7TG*BkTHGZLM?Y9lc<@VB@
zc7JQwII(`wqJ$#lr8~8kqI~|=a7>HmLn;4yCqHt0>ZtCa&wNDU&|cmf=+6WJoaNg{
zB*AH|=@+_ded+A==?e%;*Zm8f?P+m@uv>r2?%mdRSyD>$?;(@yv{IUU=<M^Y;Vi<x
ztuf`y9rdl@EIGa>DYEnCvELe$UHu#OI@^BfV;GQGxWBy|!TetuVb1k!iYW#1+P~1f
z%XsMS`)P8PiTfEbkrYkeb^#fnTWT*YDa~K#$j^w0q-cJc97)mqy+k4dcFW%jSS8Rd
z<*lKQ(%%|Jc_t1tMeFwpmIUCfzRy71*9%A8Z>>mvzO`<tRsBgxEzzDaQ~6!K7qHOL
zpGk_`5`Dh)o?-CYFd!j(8}+3vjc>z1ndirTY1iKDL)UzooCFi|dj-2lFtPG4bgcH)
zu*!NLZLPKULwDQH3udMGj6y6F_qS!g5EV`N7rIk>Yd{<Fr4`+Vm(KjmQ>cpO@z&6G
ze%c&W(Nf+TMlSs=i_S)SC^;_6OLuK=4XfA393#y{OK%N(J{6E>;%PC`Og!TtStj~p
zi;-obw_oUdyE$1VMtbW#Q%`+w;Y>2)tpN#x{?a+iTLV6FztBC`+X}O<{L+=zGis7&
zV$GNCy3<<&X(zs44=cr|pR-Wh&pf3(mM<*r8KBV=P0u`qsAx_vB?@o9bmgbXQ5Ahw
z3=3J&Qr>643;(SFEs6Jj&ue%lD#c#6x8BpVk2Qu`LwiO|32}TcVApne%WrDgO1^1H
zuR*4qxBY=3i18;};~6uRq4&L6whZ11PS<~IgY8Fv4_)(@TOf7qOOMHt-)C53wU;95
zF<!d%cav{ciu-%e+)4}mZ5u$x-fs=1rt+38JADtD8)?z>JyC6>MRR^@XfwYT>ow1}
zf@cGGCq)gcht6ESO?&8kroWvO-T&6G_S4I^w8Ou@;EjIy^R+`~`yMnm`6ixD2HkDH
zHI&}VOWA=cFP(k8v?;>&p?kEKk_-8*VSH~x!{*b?w^BWP+xE~s-=gGLf17dV)!vJp
z$>3>n7K%IlLiheQD!0;;jF*<YzlV6}O68>p-SmgHl)mRF=*UlpMqV_3n`6+G=UW3|
z0r{aspS?AZl=`KsJnsj2(O-VQ+cPYXdgA-Q2}lURL)pxnFJ0~17~7<tX#1_Vre59}
zDmeik+DiX6#-MBD_aS;4?WH4)`PM*a-Amb$%`aW^dz#uLpP0|!MD~eQexc9){3iRv
zdWKA8&3j>P&+yDbagY7dvA?IOjl^hr#!NIu)3-6U5gASEt)ZWx{Cx)Uo4;l8o__E5
ztoYMg1JY0Rp(~g1*02yme<*w7Z<C#I;Tbc@KT+39Gy2ajo$dR0YLkHCd0-_uefl~1
zC!UFt^b`G=r$|51*INPm`{Sj1{BjGUkNMJ>XL@U}G-Tg&)%m^e2i<eOboceWhTY_!
zSWlaie_}m@CTqq0ZI40sZf_0cIXl>hjlO85`qtUn(*y{OrtbwTbmaWfQF8l*uKdhW
zD2<l%-s_oSo<SF-(K6n$yOjOjFY*81l<X7r>0)G`sLz;5_KB8XT1sng4QBy7U;3OD
zZIVy)=S7fwqQ5tLESqO&BlpC3e?QuIFE!J63=2|E%=FSFf5V5aIo|u)Z9d(L+!O0*
zaB@$q{?c5Y1AXY)`7K-8{$9Rr)`@%mLTCD3zHXF8Q~ibFZ~xHVpT<FG^cn3G7ymxO
z#J%{XyS(sx`SraNnIqn|-S}00QFI;aGe@DVSs$|i)zbR3IP$NL(+X5lINuaa+P8+f
zdwOZ>kJHLVLc5P&=<MIC*S!^QS6;dwua%7){dk*!ER*@A&w=J9hrxUsY=6y9C6OCq
zeHo2VR94@zXRKQ<J+D0fLihffpr9N8kB6S&^2-hrUEe+{(B9H`=}6z61EM2;H8^yZ
zZ=V&>mDfvGEBUQqr_z3*qm1vCL~yizS@ojd2g|_xy>JAURM%VUF^uv{Yx#Cr5nXM*
zw6&h*Lr=&5<sSOz<E5kYr_ISfG0ID4`#wB{uJmmZi;nM8JLoRod!(Z4`?RPeQc!&F
z%|UbeUI~eQI}`v*dD@&c;@i(zBR-A567hTaD~bfrL&<6L-T~5|Z`sk)o)0Av!CM19
z5$~O?n%6IMm;BP*OZ$Fl=#Q593!OF9hvr)AFLc*Y9$I?c53FbToL~Cw*a2+y>HDOi
z_@4Wqz5jNtvzi>g(AoD}?=hXreCZzjrN>xUFWq_O@zVJ|8ks2OtS{Z`88yj7vFb}p
zUgH-!zkODqweIo@-MhcEoyN%zB|XS5bbaqHKu3FqAw)>i-Y-4F`AbPk{Z@Nc)$65a
zU6H@9h7f6KztB;hsR||1Z>JT}RqIPjt52hoj-u{g==gS8fwt26TW?$6PAkyfdVVRg
zzWJr2&zH_J(l2z4r`5?xG5SjxwAM?<Ge(n=Vm{w1IVt9x56x|*@lX<?zh%o<<1KrP
z%4e7!%KX+I+S=|fJ>$swh0gTMRYPISztElD+nqh9w{Z|y&M$pO475leLjWn#@=&Bm
zOMB^=JRb7^Dbku=T57GoP^EJJrnP^2u0T<V{)NtO&lPBw^7}(~&+Vb3_ovCpNzvyo
zbdL1WRY(1W?(yE(Cg188B`d`&ztB1R19N?voTL=%=^!MfSkGKVQi?U+8cwbIrLFGr
z*4y@enw+KLUN1$j`xm-WdmnAK^q2?B(~drL<fqA{U_O4KD?f7;nxy6Vg3%-`J-<{@
zzuz=e$Nz-y97<w}XSg9RMSYr_ycDgx^n8B(t)Z{B-m>W5zGY|XZM<~!r;m}CVx+f*
z6?*7@(2bh-pEQ%1;@fjYbj<S7tf<$AuK7JvWpY!j{8DL8Jn)!K`Jvx#L1JtFLVMf&
zrMSedm(G-5ihk^T=}z;dtA9JLKsWM$FP-^&9h=c6&HER+za3Yg8DX@C;@DeXI_uNq
zB&ewE7rN_wY3nV$_4eF;f%ChQ<}9I}7w(>4I(vVg7T4tY(yRROlkPg|TXyenzm?Ng
zgnxcdQ<)4E^J#HXRLuU;@*~jx2i@y?n#!c9SWknIree)s=-l6?L1wwQ>o0Wgkspd2
z<Xi77zf?O{dY^%W2>YS?G46rmd0OO2^V8z!lji->V;6iI1{s0UZ#NFnC(Z;~%Xh?`
z=&J3#Gm<*IHT3lDwjzp4`a?%)--}r2YWt=0@gs(=)W2sb=;-UEyL|g+i9U<EOu~xs
zy^PIGNsrb`$1E?!TYtWE&GpjS*0*VpNm;SlFLbQ=*3j`KdFk5qeKzuIymY3wn>R_*
zzm%AR^3t82Mn|kPKRq3_(%gQbGe3hSYNh#m9Shy1ytMSE|07mfo|y`<(wbj-dTD!W
zXwObCwD<4xFz9%9AfdB8GgX&ddwc2b<)y2?BRY92`ZH6Jw_-dq6?rR0`-SfDSO+Ao
znCYb@KSL~;E9Q8st#y8H|3UY9i;}NpJ+$YyqqA1r>8+L!mhsa4tUEG`#r;?ZsFtSk
zUTv+^-?F>CRu;0Q`Fj-$J>8){H2>0&=$5{d3c97=#`XXG*MCcY{g1!WU;qB!{;$9M
zfB&EUnGZ~V<-h)o7)$@>fB%o4|3&Br|Ifev%YPn|>96Gf|M}<t{vT23fAjm_Z{~kf
z?XNfbpFHuu+hhMXq&1GM@W1~06aQ!Y4<&yzF#i{S>@?K*=f9td|DD?Z2T8;^usH?`
zm;H|)f43-d#rBiL6Ac|N5`X<s0%^$2+S-fUzyByhuz+?3yv+Ukk20ARV?`G>Zs&rZ
zrEcovD0RWlQX$G4h2rl&%iK7wD0IP(QlBk7FZBUGOOd@E`IIOWfB#tqHFOM!Tj+wH
zr3k1cbUg~i-+$hDyeEWyI@!X0zC1MHw&|F~+xl7VvQM%5#(tLjz)!*Z+I|-Nz)!)?
ztNY^?KlDe@PcH1AML+aM(H~OFpOP2;DEKo(?^E(aKSlr8vb{xr&>uzPrT$U!!XE|i
zi~XbIhyExUulJ9V7k&ypW8MGZHy)$Psa8Fx**Eb&?Z0H{g`KJQ$A7-e%P7fo-#?24
z@)9JQ$ef=4C~}#X`T2ZN<}w<B1OLS#Q*?a%y9&FUe!rg*Kkxce?B{)d%Ko_fPthOu
z|0((NDLzGiKF6o*A8+fH{rNo4`}*-B{%DWi>wQ|}e|~B>fBb)Sf&cttqR;@a|N1wC
zQ2r0*mml+p-{+6}y;GJ~8g>5cdQnCL%KV3~>y`f9b!~w9hp+3E{@rzLy!(f*>y`f9
zbv1gHfBXe|t$%xE4aWcUD~?+K_R1RE`#*hUuk~-QtjXe^e$7$q-(4A()_?Y@)cCho
zEK*+or?2X@{_T}DC;sEl>$U#vl{LrKfBMQ^>)&13u9*I*Ykq3`uP>&las8uL{I^e?
z|Mhikp8aR9`>FH4zOGH$|Lk=?b^h1awdwnxz3!*Z$BuX`gx?K`Z*$`q5C4xA%K!Pf
z{I&Lf*h}d6pC6sx+;v<PYlW6cGEoVsHsM>?#0Q`f`0QCQ5%I6M091Sdx+!=9_oy-&
z2-&C(<q}rAQUykplU$|Myo`IWiCQBtt^4CeUJm88FDnYJjmm6QB8wfD-kmd1618{F
zh=z23=HQ>a*`eJGoSgfU&wP1eZ-o1+*2c<_9!SPF8P@mS0r6xL9LZB4q1&HXtCRLT
zRnxx(Du}<5`_U5*pxfU#@2^M{-<#|SnTm>1l4nZ?4o>?kUevCQk7$26^QxM1w_6g?
zYQ2c%lENtV<MUV9@QJ%w#OBo{t;x&qzpJR^-Y40_%2ni*+KT(ziv#|O*0LvTMy0=T
zzRxAj6J>`vk%Oxu)#a%vfP*9wRUy$OU@7EgsdxmKW*%(e!ctM<%_gsmK7Zpz>QqV}
zXD#vPN(^ov7b;Yr%k*JN(CG84Z_3_`{zyy}hx(Oxi5Zs}jU)YvN}#6`MTcKe`oOkp
zJos6Tp0`=6l4{4TFDaT<;)Fb%1aOgVk8dvI-%;l0zf)z24nQ>3`P49&_6I3ZTt%ia
znF=?W{DmXVN`iv!PeUXXC;1guHO~x+BELfOD}NX!(TVnc+I{NUp7;YcIK{6`+3w>y
zr0dhKTN^2^gS5FukfBt_274zjW5um0`5=II94Gh;7h~KzvOH>)U1m?AiWvWC1HB&7
zNyTkiQeECz-WRT3iducOa_@6zjy1XY>4M(Ci4Zs?vnOalh`*8vrRu->w}@9<)vr%$
z`s=|bjyx#z4N`(RM5K!mae1~9(Bx#9J5M<-qD}RxB*&)G4kMp6-8YdN%3)IgRY0o0
z5I$cSDk=_7m&RUR%9*|bD&H(agE9si9B1S;*oJ?kL`!j6UnOq6;%Ux}41M9%#nI8;
zC*vx&N#5uvh;M9D!AOXX_I~9Ilks7@Og2!V_f^3LCgO|qJ}D+CdAf%8R-&ZPmtcfg
z#dNxXQ(ai$^Tk=I==C!;&F0;W5UFjKi7abq?vs;F1y^lRkjQpTCDo<vVavwOb1jpR
zMsJ@8+^W>Waod9=T-LhV^n8J&uhNxk9s$qST?IEjZUL1#iTNQkppn{d<P53gg>Cl<
z(eH1R#3eAUzDan)+*GsYeMM)VTrrwS<{lR=jw>tqV?DtiD*MV(!YwV&(}kQqDL(Y%
ztNVrUb8FTdn`BB3YvHJ}N-$tXomGh$s*)$w_<3WV)4jWxGE(|XRFqr2#a(uk^eMQ1
z{CIFE%i}0Bn==&Lfn6yM4Niidxr&s2j4hPIH7B13*Z~z+&+MJz{m_>YkKj7l6L<ho
ze70HHh51p{Z=6kXafuZg%qR`(LcN-vFA$Ahy~!1mI`L1KCY5FNhb@jJ>6V^~4n_Sl
zV^r&cp1~OLd<E0jEzyJcbwkKg@3FdyH-h?=6Uj;@t=*%}zSZhej(qWR64=>HUvfOo
zM6Xbvcx@{#KJ_gytgdxSF0wCF^_vyfUQ(CCp4cugkC1-nAd=y`dKwXQ^fPM1jH;dx
z3^qB2tcVVK?r3E6`IabnZXrd<j0lw`uI;JAP|@$Q_fQw!KG@>I#$mh7X6r&}237K@
zc%oHA^pV@q)YE#oa3c`GKwms7pbTb?AFGa~XIy<KaK1{*=L69d(fHqLR7qZHV<q{>
zs(kXQ@r*kDdI)nGwj-I=jAq-&CuoLR{kFzNdO&c-L`;Yix)4=aZ!A?9aC7EZ6?p5m
z69G;u3#Pq%N+G_9vvnr?<ypaigeU94BOeyJ`Bgio#6%bWpb2h55u!?a*@sbLo3Ju!
z(H?f(C&@u%C>3(f23+dHIQ;h(a1a2<&u48Z?vr4L({eIWbKKHY*cmX_8gLP^1uWTx
zw}AvZ{Z*0k>-Uy+<Cu}3d)gYL@pVI}nwKC%rtniaX&F;alS<mJZE0*^1Vd16E>AIs
zPQKy@xS8(RPY}e%17Gb!KJtKyjIUX#c>*@sj}>nt@2fz?#aFUm%@!p)e#UCFl~de0
zYlSAOu@vhDK=OR*MJ%_<k;v!%bRj0ZfOu4@<ox`6;f<&dy2oX<!r|$cz4o}FJxJcb
z4UJ@0IqLWdhF6Vl69LuPkXrLp8A#(R5?;o#=K(|mKe^SGJX;iun4d}zho?eD6(8pk
z{dkso;qvag^DT%fjJ`7Ixf$!Y0qDMSzHDr>pO0NR;<m{S)onR2;||A8W<H<kc*L{J
zjxkI(o1U--Df~?Jyu@D276TrGz&m;GadEQWs3B7xapPyznW}X_pWvoyV&$ild>0bT
zqmpL^-xp3YCDkmq-R<x+RVmg5h9z$R`saN{QnP7L!SB^9|Lr-q(7^X^ZcqPYVdMg-
z<lBzx_A~VAr)M>~;-`I+exzcFdOH>+owa31_N=-n-)E9DRI;3TnpvQt-1(LwSs1$a
z6)lHKc7wg+w$IwV2_}}v4$nkI=S!Md#Ry5ly`YjjBVHl{6j>SgJaaLd0L}Y~(W}uS
z`@Iv7q^_Hl0J%9lpBIf=<OUHJB6vz~mSm_=^4Jl!(o4nDy`gtsw>}$u!fP<}b?9ov
zQ)x}D?`-v#`tb1r-!iq&xPM;=H*nePDId|eZ#w2dfTixZ>{60>wT=(@KqXLSNb~GI
z=-U_4jL^$Q9`WnFbb=!fP9Ku+C(~v$duj0<;!Z{m2Ck!Ik`4!(8hi|U*Z6tS!*ZKx
znWYI_Y#OlU?b%Eer0pBaMXugF5fdWz39Sb1J$XW0Ah}M}F>-vL_>mFB&S<W_<ZzLD
zfvc~vqonsLzU{coOx@;(z0;1noEkg0_UiX8m(MNIg>bZnwQn!uJ|bzUG%9s=rUW^l
z9yW5eGv{5%+FOl53p>n8fMZK|f03-mrx_mHM00pP)@>F_DN$ur+<H}7Yw!Dn7%e+p
zB~fSmnk_WV7^|!gT%<9VEmM6CWT}NJgKR)gpom_5BGCf<ytNs(d7dZ9jv_XCIV3(%
zbHP~&mnb>eCt`f4rA9%SPcU`|oNZYRg;~7CP3~zU##x)%jsVj26*&ePy+*&nON{gi
z%muqd!F)zu4Ob)iY6aQ)Tx^;?<3)>KBkd)4K8+G3Yppig5!TQGDo9<^Vlo3A&H9cE
zgUqcw1NMvxL$!pT@$>V|H6H>$w?rb-=KV5$eSVW{B5LLSnj~f#N3!HE46dvW%#PW*
zkABuqyVxk7?{&69(=}ffvq^#2tkn`R+I5H_8fsutPz`5UmKp|Wd^^MsO?ApM{Mq&(
z3iWl`HjJfqg6wHC$$7skE^#apGn&V)BYxX5nhTWNp5hER$+n6i-Yz>6NLDZJOsWC*
zwp|>S62;-@$StrWD!%WEE`1?=Le;qCEk1d$YHN+QcHHJ_WC+1Lk(q69r(_>v+3N#G
z_I<=%++JGvMt5!SHs3seF+$QlY(zQBia=*kGJXFX;rT?;1ipDKXB`rf$|hGEBd&4`
z1Y7oz#U?IOmp_eGuErYsWwxBEC16jaOQ>Bz>~Wd%O=pcQX(qRtaqv^;?Mkvf1D?Ux
z?x^<`a5hpUXfMta{vkK7u=6UP&|Bvk?6_4r{O6Xog3dXh=(pMSB1;0S4Zk>%LBm#k
zXG!6c5rIKtX(h>UwF8c6%f1?0waCxw6=)UgsUU$YDy`s*>u9w=Qqbw&JTXBwq_$=I
z%OLjj=m8|~b9NfQ3?V+~sIIe6*LdNZTnNrL)c6@>dzkPnueO=Ywd8T(OlC5l<%)k1
z>@OuZ33rPsk7+&hInBM|(u7LeC$>C!FjDi1E!W%jrvXkZ0h%%I5hZiQB&P>?`cl~T
z&THyu2Js})uSv_09abhh-B9z#0UA0lC<SWO`YwXuQDNKFZ*io!G7;YTyb^EKp)5Jx
zX-x94*{pHT+&rSjvyPqH_T>G<1h{=>aw8hYfG*2gD6V|Eu+1W8c|W5<ayiNxIJUiY
z00iZ$j%p<BZXF!Bk<F>nPxt#F)JKb^B6q3m@JuMV+i61a=az|Q3ldTwIjc%k&Q#ew
z*);GBP%>q4eGho1Qd{jcjd%tz(W+VTZrTI2U4fnJ{#z)?SDRRly|dzR#@=q|4cjh2
zMP5PUG6m|4A1{n_8O(m32wTZkd#4MJc$(e#bF~Nzu|gu7Z6;Lpv&<#Ky)0Xs`Su~|
z@r4uz7_RoYf)=NPHGPuFq8#6JKEQF}vh4>n;*(tINR;5lC{Xztt29$az@{*9CUfaW
zJ6zS4d%Y!pi-q(xLl-J71v^H`mMMF_lof0=TkIX1$-f3dd77EbYsQr)4(YtC<`xig
z@vWP^d>X#T6K=9_R@G6WDWqROFU@8yAQWFXN6})|vo``26VB>kHSf8YfSf?ptO)nv
zuqYy&L2hlEsepq_d(6?c07M~PnJh6PdRtLrUDI~8a=X7hU#p)(#R)Y+@QIhi*S>ve
z1suE&ZSi^1a@64!2rk29#IMh-726<8zH7d8I@mS)lx62gXuv0RT(ZM9o2Gy>*)a)U
zZj{_gWNpVJ^};jg>QIABwZpx0nf)m@J+~&3@Oa}h?=dX~Blep^jp}v<a_}|VA<5cx
zhtoDju8z^u?kj#>)sfg_&NKb#7pwz!&+iJH)eqeH_#7UKwCeZ+Et(5A`3iu0#lwuw
zQnX5%$!CkknMu)3`n<y&(K&eaa!1k>GI*MG86+Qfz%7S{#f%uYz7_4wT}HB0B;e1N
zZKvTmMsm#wI%jq}8};`(s@~UmRKIaTa)zDq`Hl_B@7;{m(Y-$sjWf+aV=Wr>RRX-}
z%grZEw$tbIHH@#qft$Pp4A&%pci210l$k8ZB;fEYB5i~_ucgHu9{a*f<;G5d!&ym6
zz(^X-6~K)%17mZ_<|%;KduPC7xF5Af$+IxB83duE8E~<rv7H>-jYaBRQQf+b9|NK3
zo^tZ@@#2H~U_Zgn-1q1Ki)<81E>DgN)aJGVpS=4G{341Qw#^RjP;=L|on%E?6g+0U
zSwG#A2&3;_sx9>D8MIqKEn%_Hs_%EB?%t{`W7>%6Ymj5cCXq9Qw0Ooj<kE1aMn4Tf
z*<EqmT2}et8E`c-@%nNT7G3w&?TCbQyLOvI+jKi~-@auH8{KT3ShS$(UWScygkkr9
zYuEwF@N;k^-4(q}g~=CNXtyVMG&H5(ip$i&7bqL8-1}+8I`EFX+^{umir(RBfS7w*
zcHVx>%pKPmY1!uDv0J@cPjbqPZF96|PejRGBdNS3eD4IPux;&4dhS$o+|Ss%Y=g0l
zs1ut_tBLyKLi-H9I_xE9o!#dmn@qFg`54X_4Pzwe@lBnzRVSZDRlNU*k|nDHZ}t%U
z<K+-rJ`GFz&dXt~I4bD~w$=DK;yF4a-2DU3EZyx`cQ$`*jA!y`KvaBQrZ{bW-Fgi;
z7)u$jlsV2BipMZzx;1uMD`_%0Mv#AxXBCy*r-2T-15QrYs>aM7kHfj*vc*y9@pGyE
z@ICf0m^i-dw&>|8CMf(yyDyGX@=<3Kwvse~pR4#nVUswf4m(oYJxZ26d_m$nL8!HN
zQVRGczOyqb*a1nN?id4{)6AgcVP+_`*J;O3r*T^p#PHN+WKX`GjZV8_vazg6_ZPuI
zzwM*47a5N@6_AakhmVHCRT&Vm9+Mvn0MVGPNfzI9wq-|gnC<&?$g>x9UsgKJMiSs+
zsHt*5s{wN(hXBW6quV!*3VR}jfCi^zrU!7ytDz|CYm}^t#MTj8u_oH`l-pWr)Hr5F
zmtAq#K;@<~*b_ZM>knUH14pq%$?D#z(Pb~yPD$4EV=16KJD7a7zYA2^TeVXX@<~*O
zi_~V?{4RN4r~qu<8u#IFyD7d11`oE#^j)?N{;VJeF5Qgo;Y)9H*NKg!-KytLhq?6{
z4n_#EOF~2T{x*vict&8oiP-r@g<V@NI!)<0e%%yuEg!CNuY4w)VkUO)_c~qK<+aOa
zvk{E5CoPEFiv<C81zT&8<QoAyyLXOuZqmJ>IhO6e<>gcJ)}d+ABy-xkT)4KasM&6f
zXWR`x75OPPFUcZ{qZusA;T}rBDHlX?kZhYhP+qTY)snJh1yqc+M&;I`WYNN;OERK5
zt2x_@UDAplS1lOuZ1cT+6xZc2L2Dd)uLLw-M<ldsiC?3qub;1-O5FKSXtfN3e6KTR
zwF`xFC|c|00!C&k+pd7^>Q-N!_cF`8k+7Wn_ync(-2A+~j}+J5a*UBVHN4JBz85=h
zpYB8mJMY<C63$v+In3zSX(o)5QB>A5b}Cz(V|f#CXghpuGSjWIR&&oAX<ej1YArdI
znds=Ncg-+o2=@_#O{r;<Nmq2S!-@B?oJQk4R8qVwS=3V-iw^fP7db-J15VCp)U8Ut
zC6^+o<vmtgTxi*G0ht}!j0<hWPs@F*jA)^4Th~jDZB~+%mi8j#25a^dqUvm=+J)Tf
z7`U`H#!qdUb&00BZR;`94evzShRT}5&uC>Eia^_El6>%pI#Hgp_p^LM3b=GCOj@8d
z;zFJExR!+OltZ){{gxkh_pF7nmer!A`uc5a`ZTK=ue%!1SI1}GU2&63?%`S;rxFD>
zcD`@x^b1F`vy!shIRp*$GyK}leTpx)lgee=-tO?7JrvY2e(W3Hw)=!udO{IE-nu<q
zct&73`<Cz>LO;ErEP_$f+zS!NlwdBV<(6!Wtui6pYaW?&#wYWR9r5Ec8rNc8H9f_x
zBleeVimLZAXAreh+?9Pjif%gl6YQJpU4yM;GvOH+%`}DaYuov#E$;CcyUF)_lfN@o
zT-&BhB~^g2;M%qZtSMd(3ob%S<wxa5=fIh44S0-RTY(~$UX!_upe$1YZoYWU%_T+2
z+OwUPq%r-I9vt<WCjjY)TELlY4OoizOB~*dIdS325NsD)0b&_thsSCe#W!<H+|WeN
zQ74ye#)HpUZHdB8UR(yRGn*&STMoFZxp39u?%7DBcRA&7)arf>+h!cid&N&`U}p&4
zHkbNw3G(O_34iUL*2~TiY;*hYRVCW!Qnb*KPRdbb&Di2v3XNNm>?mgKCik;P8vP8x
z<~Ke3xCD6KxUUQQj;r^M&%FXNT8WY)&uSjlIjp2fb`~4g&nvg(6~gFeRI&kP<`pI9
zUd2Nb^1V-1sJ@Nmn~A8R=Vfztiz$*N7n=QY$uHj7S#sG*=&t2qr?VDwSX@3}vgF#^
z9D7rYw~Ovvwyomwt+Hsw)3vjIri+X$OK>hfe?;{>W!v(KolMw}e*<S~Wd!65cG$Gw
zz6ijDs}v5u%lDEHLT|-(yiRAf?Pw&=6JF}<ctv<Ia47n8ORF>LJnd_<%6StruFgiX
zV9UP#MFqW8n^n#*>g}=VyIB(m`zj;gXw8Cas?Q&UvA3CRe=7`|O{|M81KafTEcDNt
znG8+YF^%0ZVia?6g`F|jw%pxE0OfQ4bU&9kbC|6Fj|uFf<>uVcj{DMZ=QGsJ)AaAL
zO@E8*+D%&lo>j?;XE}DUtpLwLFh!kH8|_S8D>~a*U7Zowx7_x<`Wf5IyCEk(IRomG
zmEq?{ycx;!va>BmO8vlvpM|WvVnZG@H}`Ky)z848C9?%ji{xZ0M$yU1yGGJwmKBs`
zl%A@&8@%97b~?5Cw0Q|C0oSTr8TCRNo^ggw^Zx#<Q|Gdeb=MhmkOGB6)Nb>i*k(Pg
z;G$CQ!=6uf$VC|Uqx<`ni^cTp?u$gxDPJ81;l6IYTL&#VvGr3f(Oqo@Z-{XjJu9$I
zIGAh_-wlFhZR%@aZ1H3`y)`?OFLSP2w$yGbm-Fv@%a+}3>C$`aGidGIx@(_xm7pzm
zb5HEyRVRAB-fa6}t=y~iYUc)LXim3i$H%iOrOXa?R+MSm%XRM}s7;f25Y+mCbK37u
zm+w93I>WPVp7s5?{@TlZ>C&eWw{{S1n(tneLm59qJ`1h1FP-v%kQZrLLC>nHb@Nlp
zJzsi8>Fx(&<JNE%zDdjLk#oQFO!n3M>~dST-nuYG3f6LN0Z(t=g2(*pmv)`nmL*f9
zTzb}Y1=`)Y7I<FF85#=C);>{);>SX*sigaRB||P}3LCozTUd|<-<0~f&{D`&(f;qJ
zPMIr(&2(S+3)0*FDms2BUMp>nLquWa)$<5?OyFY!6s%g0U=bboJQhsll4sB&Tcr?8
z*Y<_lGA>e{iuYB?v~q@l(sh27gR=eKD0@t=am%8{ESke|fnsxYT7eDgd%Jy&GZz$W
zf*-egy6|bJBJrvi|Kurq#-u_H+LUj-bD+h~ouyp@o<2JX(&kvde(2KE&fWTb+}%hy
zLCIX=Yi?RB>0Vch0YyXUR@d4&#vayeTis=6cXAM1<eQ*tWOp}*md5y2Of~q@n%SG;
z29R&PlqB=&Gh?KCTTR<w(3RqqGhX25Q#eH0@x~a3*ncn5UmiK=P&^ooP;&iHc`z4k
z)Q?;5B$>0lvDt1bOrjE1!>FI9SmY4avxVgWJyTmU`cY}Zq^#%CMS(Wax?S$FYRU73
zl@Dcv<HmA4LubeHm<YV_v0@tOgwS=qk)kDhz^NXtBffaB{x2EurAdzQGSVsn&%28W
z4-17|?!t=sN3ES~e$Zq8r4J<!-d{y4nt87^v`Ja;ap<?&GeFbc_)_G4pMV%>QC!1s
zwTrA@e`WHx)P1W(n!H0-uJhKr$v5>^(WY>}WjCpsbhXi9pdx;HohH@Q;w2Ce$(ibb
zZ0^-dm0zc@>L}6eSk|2pPO!zh8PXftqQ$)=pp}#;G)*cW*qGN#`B%6!i;;x`XlZJl
zvW&7yV}TAGB!7UC3*+7Zdvc2=ICE>>K_@u}pvw=Z1ySqQUHNG#a5WsxX_h?Oy;t@C
z?1k<0sdb~RPOUA6kuz~mt7^jYeWiRUiry5BKqg?;$X~KT6>HFfj<oKe)qE=`h%oE9
z`n8>ix@5PzNO3q?MsbP^l-PSY->Wqaxr;XKV``d&*`0|tD@RYvqO8SO^kvT2d%a}`
zi<eC5pdBqU3AO7gT9P0BK+o59hRD1Wg@2#yd@QH?WR@gp+>0$V9id&y!cI4Gh}<p*
zq4&u)N`lbXo<n~oHO{Sh+&$LS4?F)K_8!WN{<#RbvtBFJWrwBE{Q4Ygy>M-7FH}Wa
zP}byyn;mI>Dj+7XB9lR?!w;QA$Yk5&`_eT-(y;9KMtOFX>=7sd8~7SztQHT{0nThq
zNb>Jn5?O<#qWfAH3rFTNO=d|TM8hs(?#|nj{9?P8KX%3G!Dv0^`!Q>2nM}}&G^3YJ
zOi3;_?zgZBz&nO2Z*k-pRkahjCAc+nowr^hHHW$v2X#SfW?9IPmTKykP67<+p!*J$
zY`S{eoUgRstC1_A={2O9F72J;Y#MjW7}HKC=I-q&R#>}-QX<PK=JUX3eRy6<CJ)xG
zABwjjFE{Ue;p+Bo$pCPhhJa_ihqJW!G{Uh3PFLYQET8VsGgtJY)HU4=Q*@&-kvRd0
z?-@Pwj_iK6eexdkyz(<kAnckvj!blNAR@Wf<>Lf`=9-g`d(FOKEV@Z90ribJU>mrt
zA*$N8kDw@;7qlj_rY^Tqpo`TZDy74H*r1~kS%$uC>plyiU7paJKt6cKa0?yGKGOLO
zdmBabGoWW^D$6@*mL8I-onx<CbkpG}nMeFS|4V1VlJj-*4ThRXwn+UPFc)lpw4KlQ
zc#9rW^h{o8M(6yg;21+^Zh{W?zkuR#wUAUhL%e_b(1C4aqU8J1CWmoa2&sfi7YY>o
zq5inDGp#dLa{d+T8<fiv-e=ql9x<0{>w-T_-w%b(g2z#|1kCqpgr79s!|(%h=xVra
zsO8e9hMmyd*IV-7I?;g?1@&qI7G=KY$b&~O@##R>$U~OgkGQiMYUvT!n7WXdRz2--
zw-LdC97R@4k`MR%rEF_)OpRN%E}~kI6rUw}2kXIY_0E-|+n$wQ#-^4g`TX3%>DWzu
zVeS-9TZrD9>R_mJ-?C$8#!D9K_48z>vZ_xflYBhr%7i-cm5p8Qoe|ipdc&s+h{w6Q
zIQs(o(p|cJPt~PEa@N-O(wDPqU6+!<#lEN1_J>YlI}jD$)i*up^K@9`t$X+R0l3x6
z2@r7C7HI9wzMhE-Pte9@Cyg*izeSnCmMVw?SLiV=vgV?-eb-!ccblP((O+&^k}<1>
zTKm1(V^7!&b*!s=@Ag;@HePo3=>(pyclcD~ZUds5%@U|`0XsdN=DZUO72J%CGMI1<
zg;CSp4@nfqI89qD0fTaIZ*xW%oWPB92j+Q;Pg?7h(^vAA$A1f2GwvN&{htwxRUzR6
zn0qdY0BkG^WT%Xn99R>Y#Y~9F4A=*&F%DDfxYr*%H|w4qwmEPiw$BPD@6YdNL%4#n
zIk{O#|4Znu6uyvluYU&Uq0Nb7Jh}Nsr!1)L7dlxkw^_~vSVWJ(9G$CsKoNJFa}1Wu
zYJA6qPEtN_R5RpT?rxN6`eH|+;Lf<_t+vR`yYN}=_?XbA;xpzMT5gx3q*V31{f=}T
zlVBpTms0b-^^(r0Xi=GazBT9^gGK3QXlMJP+xqidGbUp~Lu`*{C?+`|C-Q3W5(6+H
zLZfQ^d;Xnn(WMANHi>eq>`^dpTv+XOO$Q`p5P~RnIFv6PNbPf9t;X9Kn9oZ)KY95f
zmM9_dDm!~L!XWhN7+^F)^5Y9A{-9m&H)UTK%Q(ORgnJmH-g+DUj%u;yP7k6)-$M6@
zod`6{8@d+sv7SElqH1NYC%F%aC_$BaiaX~Z7sbXOZ$0}-o+ztHwRyV_wwP^c+gW2g
z;GA52oNd%tqCnXPQ4);kic=@U`W$WM391OJ!oFtxls#51OI4h1O720~t<ka+n|oYJ
zY=~-@DMpk@K;Tpjrz5mmwi2trh|>D3w_XMk?IV;Gn~02?xX;@@!ZjVVIN2KMc22i~
zQHhLXVOv(d4>2}gAw{lJ_xDp?Mrws?6k$IPTda2F01wLfp$64DZUsv1#YiXHJs@tE
zx&b935JmiFYDKqmq@2an&qxFmX4KfZhfcyr=mpc2iJ%3CG+u5`?(-zN@whAsTaC$0
z3PRI9r#1Jzg_beEi=qUgcd;mBdr4PYVP3N+#DS1byJCl03t~LI%Fz1qmc>NVsI+@d
z_s(X)O@vRc+>+-C43^5qFfUyVmwui>q4-XB!ogI-t^1gZj;xV+eI|#^9&*yRRI2O2
ziE@t1wqTtxq-?x5N49OuS})ZJYvDk+g!#4qCA*aEzq#TQ-*_N5?YZWa@AZk)(fG$=
z^%iXn#hZfLeEZzMvCpa+_Dr0<qJs_`MzjlB%dK}5L@<i7yT|uT>;btjv~}*`CD5sp
z1nS*=-h1GbSTF1|2W)mZvrZpYn=zzqlBh6LaxY`uYEfD;5r!29D{O)2xRPU$9sQPX
zlZCjsg?f~Vs4Ehb>As@C!s;G(!W4@v;Id%$U84<bv!l!Pq;7YOfZ9QKR`|gOj%^`m
z+ud1!y~r0>X^+x}VrH2-iAR+~M{`|_R$P1uI&-oEr(@T;Jp`S}<Hb|@+%2+!!<@}y
z4uLW-n|nv>k~2K?+<J&YI3;L5n;Ou%1|)RKlY}TTk6X#w(Bm)`&)!r+Nk7YOs#Z46
zx7d^Hp^<U7y0oY0A{plk@=gwQZ3rBmN<b4o#5p)lhk(18J_9Gf0Xv%>QdvkJcs>m|
zi>5R_4?9@A(-V=|xJA#^&wbNBODXqKr>l0+@FD`9!HFXjZPAQd!7(BjB{(2o^OoIn
zxKnJa@vMa~!`V#S$GvYw_p}JLeNVv;&6=nvOG2BVxJ{Up#&lIO(K$VkJi=0;rwvVy
zr!{j^5ZuzuVAA|i^sLO6zPvNq`xwqFp0`N^rF$-16iivSy~f*VSQAP;BP(R%mbLi4
z)NczY-=lnm!$ji9ph!Q1rEwnjp4J(^o%VDqIDNnr9?rz%XBvfw(cI(-jI+tUbReQp
z*t2cXgs42?6Hzq0$E(3R3pbM@(XDi8ks4ybz2|=BSaQXN${tb9qNCd3XNnVO=95yp
zik>yzxJBE-)r3>V9>?$)sSS->jz7CcOF+><)oI`vz1l5{4yNe2zjAMP7B-hSRZZQO
ziO^+-jO<|FTUZ>C=u~#@`}^~z=;4T__4~9)+Q4>B0p);VQ`704G5m5OgnM<7cYnMb
zLI$TbI6L4wC{7eAOZ4k4Tj<yNnoe8XtDPYlQ-DzG72^p9SnM4Zb4|H5U~hEjXDdc_
z4EnS>p^Fci6(M6RM||63BC?hsghz7f*5>Dd+00Wb9&2{p8fIq;rN|>M<`&i<9J{J)
z`ZUf`D`4W@$&HN)+)uceySEDR+3Bb@CNUzk^Z#Glx$J6^B*%H*PjQwri$i7HJ+=f$
z0^E(%JBkZ|90~$7lm;ZI*H?Vz;hv|es|lzH0W;lQR`F*>-Xi>BZtnTaxL1^XlX%e)
z^9w@(lE9^M@VA(y$XKbKN?bF%U(ikBnztw%EE4JiT1HjV<ah)`F4!lz5zl313SG^e
zrkxAK5O+Wc!W9)uLO0Zts<OzN>BWc{-3_l}CR{(44}hS*JJmCXuEjD?C4h8ecL%j&
z#&yp-hOG>`k=tMuUC(<BXHnmt3al+w34Pk0aUyeB5?$^Cm__g(!=_D7IPK9&9^g#;
z@uZuICNHX*SlLQXtaQY5)=fA?rG?OiC8@klaDaHF8sZ!5rWx7*RoyTfU#TNeb|#^{
z#`}RUxZQs3na2WM2yn2Hil2GW<Jm0*5soqPjy>1@C|`7p_4l~Q`zmp_f{O@B%2N-b
zmMiWe)I>gk_&>rTBH1tU2{i&GOgdB|MLuDpjf|pf=VJ0HkG&i#6*`f%crYiYOWp!4
zj_OSR09)s&YUmA6vA#Xk0|QpB$d%BE`%t~2FHz2*cl<oMcLI{6tD%v5!DG}GIY{TN
zUWpzbtm^=0t%wDf53#T5$@Sn2@X=N>A5|pNyo9rKwdsCW9&~r4nXAan*U@M5dM^1S
zlZjPDBK7iAdTe+`O`~%m;F4LQhAc;w_D|2l!fkBsN<BS~tJJwZwi0HdQP;TKJu3UU
zM+fZ{F;JGVUEYax1dp`dU6%#oQTMfD_sH;Jg*_xK^z+s8RC6NGA+C<qL>E%*T6~}V
z&Xtl~kI0qqvqfln+-rh1lw;3E*OuRzr{|s7OY9RieG4m%6fsI2BG6cW(OukYI_t!Z
zOwi9^6t@M=h8)WeB9fPV!YbYpU6*}we1}*PkdYMoguunflhIS!cc%HwMEhuu8xyY$
zOgz;IR>Tos5ScE`t4A_VHWTF5swV6`=QXz*Bkd_QrzQPKt{}WzY~<3N?h`6T0vFUc
zUbB%KF3I=dWn5*)>q){>@<xv;i2c!E&<}}s6va&}P%i@!oi27!&?Wf}Vl!7=tf*9s
zim9sF$9HU1$*V4<yk~0&L4lqR_byYa8brebbi_9bDvy)@XjcRkbhyCJqa7{hi*@9M
zaG6WPARiPT=(ug?%$8PDJ{I-z4fGiALZ48m0u{Z3ZwSSN05=L?vUV@~S|IhzfIGf&
zENVGxyr})E2?~{z6l4445`&g5`U#3ZvExV?PV^IE9P8QxppyztIs5xAtGCk8AJ2^m
zf~iDKWt|Z>m-_;vNUp_hdDQV-T@P#FRfBqkE~&h0WXSKvG>~@?uN|meX^-=C?`KZA
zL5I0sMadp{GE?cIIFymm0G)&$Sj*46^TI?^cN;p-g=nlm=(8g_w>@xWF@nxwzXL23
zN9l8}*|U$TI%i9ku%L1FsM7I+hARhZ(`|So;T~wCWW?PY%b{HX5gzz_){9O)B3`2>
z%s1C9bMtjrD|rP&k}RAo;^5hOj%}kv%t$h2;+{zCtK^j}<^Oz>Y>66tV)-2NY8(iK
zjjQ9XOLYg`Fj@TtpYae}4X(=d+gB$;=z5%|cOFN4q$p9UQnl<ET@qF5+5D9an~-W7
zg2ehh?%GFXjXq)RhU1#`uBfWBBRiTyd!ta6ufrk%u4?=0<;0=8R#c(byxjHK@9dFt
z(Bs;5=S1P}&#RbeIn;$Hb<l{^N(VynAI~){FY5P#MP)oKXRwu$R@AJ$cYvbWc7-8h
zHKA;Cc-eLvks$DzuH`|A8EID+BJ|2_(mx|m#6F6LR-RauX##+e%&J^(M}?wlr1K3k
zblQt;4;+@g@|a!p1+EJLvOos`Z-25^j#em9TR^+=^Tl-?+6^NG5dw6Z*Ka?)l$^j)
z_sFJZ@-<OY{!l7D{k>SPCX|j(Br*x&sUNALKv%7xsCMe{BJMNf8cr(fcw91+8#Dtg
z^|{fdR`UP_XVgpBS1h!+c4~CurR6(N6yH4x)Y{0C9phm8Zq-W5v0WCp3b$`fkWi~0
zKs`qB%sGs%2rTb<Sg_QQJT7qcF5ff<Xf88_SC_M0^=gEii|Ki;rR`3U!$>4j#?iY6
zQ0vMC6n6C03E4A3>7cWF72XrER!Tnfm2<LwtL}x`@XBM-i#gUeO0D@%++4)or^ioG
zP=8S+`lK9tYk@#py)97YD6wbLCt!O0@p+$X9^rTqw;pfV&N+C85ct_~YiJd4HACvY
z@=>6(N;HIWZQD5m>T>)uwFq&kR#>tjjs{3TZihl{PRr+t1HRvTaF!@nR)@~{qPDMi
z{ff0pBz3>xqlE`gZB>zS2m#()?9sA(Jiw0({tmRZ;w2Jw#MeMfhX9gnyNH8R>n6cf
zU#jw`<k-85as?GcqD0m8EzXM`D-Eho)pdvIqnu;Mu6A{cpIGqO>DrZ8k8Exe7tq7l
zXCZp>uuJ4P88GXef_cOpd54~umIr?m-y3tBc@LqUl`=jCHv8`1U5Nk@;2Pe2hP%}Q
zJvXagQKMW}VN?$Pa#z*>sSks+M4F7HL|c)Vh)TZWa%m$lUAlC)=Q{P&-J#p)8v8X*
zb~Z&XCTp*yNh1L`j_MgO!bqW$z;gLWZ?zuiGB0nExX1-pvnz5KelDO`5z464mP3yf
z<2|=JS*1X|-v(>Tj0zm-$82v4g)wq37k_Yj%{M(o0`wX+oTex|YqGhz_lxLPT@bWZ
zZ#Pk+fKs6Nm?}q&)LcXEQtBeOsq;NRSJw`{wHb`9rda59&7I_kP*GW{*Rq#Y^RUAt
zFap?C4n3xo4R3<A@~9L=c|ooA-#{lBk0!EH%L5!YVX6GJg}j<j>)tAg6i8mf?%msP
zS?Glqib5Z@lqNpzU1-!Brr-L#9eWmiMqr1j!ata--g(Sx+Q)iFD)^#O-TN)5^T~a(
zB4`CTr}y{ZThF2A{hxLPPjjZMoj&xP{rS4;?S=CWJZAT}kMP#N3!TgF^?0J}qSIx-
zT<n$h6gyEfS9?9q-8)z7=9|L$a@hUs$5lb=I|B@CSN2>%!tX`Rs$2pT;h#iphoWxm
zj2+ha6T8xte*vxK@!ni}M8qC+J-#n3iD_IkdPhYU3}ILHdXJswN5RXbtSq7zU2)K-
zxGhw#$pN~mmx@Cj$}=Iha*N8NY7X=~P!|>(<4JU>gPdg9->IHu#amLq7D{k6w*qEn
zeLE@O?*3_A?+w7V>o=Et1B9LJ^fsk9meAd*I)CW<ZdE0R7O`37K9}p9b=nUMSJv5~
zOC|K&dmZ51mut9n^%W^@E7vyixSd9Ue%x>ZK$o|3fSgwX9G>Dvmp<n@RGaI3;5`2v
zzE_X*OnP-i2*pcQ%R`>c#l@FOh<Jk<js3KPms@8obep@UE7|@XFL2+aUb{k_XYql*
z*CN$-oq14ofWha?dd7|yxZOWZeHYM5BXbY5Y|A6mDY^-f*(-wt3dh2oVGFCROJv;%
z5M@Ukt;33eU!@pm<a|5gnI!(IX!-Glp`OVbjyE~iJ9g^DR=8;P9IIfkE_lBeygJ4b
ze$=s(a=NIK`=f}4MM4o4sjAnmc>q-|<G9xM&TK=Vq^cG_-b7jyQ0&1yDE6W#lbEN;
zBAWs_yAymCS_Oet{ef6Vdk;`W4645N2YAe%h@8%uXs1hr;x8e-Aau?fqok`$pwwP_
z`Yx0lh1&AEz4Q89*Q8Te%*`i$QFd$<bSt1{LIKg7!TD%Ba!NgK&cLt`mGO$;FQ7TQ
zaJKcHpO$utcDGUE!Qcrx-LI}M*o0S33aUo*x#I_V{;pVXg#iu05(V2=jwD^Yy#6%8
zxgAwM)CiOV2#OS_+7hbZwqCwU$NVZ@a7zxWC%Vb3lk^MWZBTlsc4r|*<zpY}3F29R
z83}`~keFB8p7-d<vF8a)9bxXM)CAqCLdDB_K^v6dS9xRC8ts(GH_wIBT{SOwt2|xF
z)u5|%lDFNTqU~5y>0m&UD&CmcUes0Sl(U<$Rzt<o6|8dA{)8>jO~zkj(z{C}aOi=L
z5pOI#48*2Vv%N4^P#v1(Qt;y0QYXv`P}TUb=cLZ+>%`O9R?g}Cx6|@k591${`eWr=
zb19(+2Nt3e04}wr)5riBld7uO_g>i&|4o7umSd41@CQCa2LV*A7FA3g^`jjxP{dOm
z>KfS{FHrdgSieDU1f^9053%NIK$)(mAcEu7spow>hq7keEV%NIM@^4!T@vk5IMW)N
zcbg>y+Au4Dg}6gUd=s}y+=;Rhouc;=jV)0m04f0vtvit<>*|r{tK|9YV#8My+4rSX
z9|Aoewr7|s9a;k2Ppe+(--8U*)qrsLUeofOL(NNU6EsYoxZ$X=(D}j>fh|##AhgQ8
z#gm;T2Z)>Sa4h$CldCE8$t-TC&%tQ7dOfJzpyo?TW*|F_4p6bED$9=5Pf=%)#@QT7
zf<!m<+Kys5fn)G371nMGhY{4!iV)W*S6WD4%M-R4Lh2%<%;}#<#OFZgx;%Y_mM@Bj
zKb$%r(GyTJ=joRcX{wHb46-AT5=6$kXn7P=R@OV-NBqF;M18u;lt9lCyF|{0fNqIy
z1mK{)^j!ZC=R{wuO8@#Ddwl+>y!U>hMj8~EjH>Ql@sdhjwK_e#M4k(D`}#F3c=jEI
zPG5eo9diQbEq=_C;rpGyT6)Q)54sYyMimd}YpOzY)}AF*iP$h8Nx;O1Ay*=<JW#w4
zrHTM-*e#%E4PRbl9RZX<Cj`{OROTJb;^BG5?#GR7a}Yf0PJt-<b_fJ*1s6(>GJH|c
zHGILGYrV(qZTP!8rnQ_y-&E0v7jyliK^^mSp1xP>T$Bb-zLV9k`9YT#lc05zpyEYh
z(?mHk;C9-~*|-xk#9+0whsQC3<_AUYw(AwR%qvmIjQPVIv_=r&1Odb#hz~Cy_78GK
z@_yh8Zht+Fd3;yqQM4cqC+@@w2D-LyPdk_}gMuBzPM|5yi24m_syfGqxKkdWPmEcv
ztj%}qsE=2o;Lz?wO>a0FBI`XHz0tLo^&Oe>fr@8wS^~$-vB=Z7{r1GiJJ9j0`X1q`
z=gCaB5ox@j0}XUM+vBneH8RXYR93G73E^RZQ-UW$CU&C7T{CAp?5hbZ)FgN^@1477
z?S!vHsR_2jfJiFF3Hx$bo1k;J0>@c>?x#0Un;V|kD;9eZs0}CTw+LAc+9rr|B7)Y)
zp>PqLGV(m{rA-<r%Gg`C4HD?85<C$Rq~@j3$=F$R{mT@PByqbg8b5$;<(s**5d`U>
zV80vk^mn{tdF4%3!RVn90jLGr)VL%^Vz?D|38T7o;1X{44*Gntx!oja?P-a&D#;VQ
zGqEP2#iw_iDzU^m{e6T*Om`sh`FKqe)tfz0+RZp`s3nN6<d)XX`-&nSg+07;`)W&r
zI>mu)=m{Tjuq}#k!tDuf9Tb6Zp=-W(@c7uiIg!H4v3oaPDUb6TOM;3)o0q-ZJmjFZ
zNMa`}Vv)!I)dmf<AhM%_eWp8OFHJZ=ttZp~%#tG;O4wrChfy|YtUyGkikUETPG_CJ
zrz-izq8|(6MC`!ZbW`Y-+KHIgIf1GCM^!B3$)R3aDSB_@lR#N`g|070u(rIWY6fk7
zXmN5C7n;rH-P)+S>Z5yZo0T(bwysA{9GXH;RIPzjKz1ATfR<)ivlq`+uZp{=7U9%^
zM@R#XD(@jyaOu1cCpui4aHyZ1xwuUzehq{_oorI`+a!wl$TX+@S{`vsUvyA>Oz_yJ
z+YRfiSp>D>PHKrJy`Za$YJ%@#E3e6(;?hO?{o;w+Bm2bHIfE@Y*kd7^5Po>VdIUAd
zCStYa#j*rp1F?tC8-W>UfWnWJ{rij;oy}FD&I|jZ$Kb!KM#-0pHSFqTp_#t|JvOp4
zNGe`%p#8aIle4nLu7DfrL+6eh)xa+rZp8Q9Y7RxU%xPY$wLI)}Wvmrhl)?QVkgFI@
z6la6ZOMi>To`4AbZhu{(j0RN8rB>6RYz`DRATSXyiuT%N(1Jm4w<Ja4DL|peH59+D
z?nFy9Z*|#=#Z6>~B39O<yb#|w-jX@Hr)1J*;1O=`^HleHfTortkWqTc-yg5b#bwVH
z9qTkV^l`nyb1P@gYmCQJsOZe*So++1BflOA1JHH*bhMchI5(tSG;}XOw<lNl((`s9
z!KUQj6Q!U7W<Fv+-utTLwQjMBm=*mJ9VL!g*KHdZM@1jBT`k;bN&|`>j6%Jw@0kav
zMVfr7xT$%BDtRWAWdsv^Ak#vAP>n!yMMOcYJ28)LPkFXZtjJa0-zkhTyglR7Yk$$j
z`z^j$1d#{58O9AlXSp2co>4tka$ajVAIwS!?G4&9%qzhkOFG>dymIT<s5y>qx3~93
ze@D7@UsFr*d!>YCZ}*u|a{8!Nn>R<v)FU8t22ZYpRRTxRJiAm!jn5k|@b;hCk~6z#
zQ%$%MxS@GD@B+7sL#}9uDpYwyrRhg()u@K52hBHDBeJ1Teo4>PsImZ^;+mMa1bN(Z
zV;(7^T8#pAP*WQZDInzmj@>$?m-=P0s7onFxU#4ORQ)i?yTy7s5e{`F%xa-~G9?>p
zE}ec?`H9#-bB)}(U(m%9)z+!zAu9?be)*Azn(7CAl{iaO3xQhqQ=v)N0_|PNL{_*p
zbL>z@I;AylqdcLse*IAXETCDbWmEF3+}8~hVNyhKhecWCq-6tkDoMG1TR(y#|0uNN
z4FsK3@N`*o_qk57V$a%4l1G9hfiecPhG8>{bS`y(9=Vknry%{++Wm?_DUcwimL*J+
zLNPjClKu*wBOnzE<k^CyIJMB|nnDK}tkXzL=o8VH>FruK^E^9q$={Dk{6PuKnI|Hr
zZw!5d<dX@)FCm{4iQ9=@DC{#P8Xp!QS(6XtTH{fRbb4MPQmr*KGC()7CS7f^-#fKM
z?cu0q*EF5P9(aX8SHvMf)ry@oT?walLjGb0QHrVL0bnJZ2OxQcyKha1S6IwSErk5!
z<Y=_$CJit`rUUeR!d=SOJD^VSo}$nDooc6Igb?ez4buUY#^@s7+?{Z;fAEeHBdvL}
zXyt%yPGqe|&B~;ngMtjgJQk#WP_ZzHb&asiC;c3Vk`Q>Y`ksvnz$x@vz+j8&JK}0R
zScg-q^10cBej?KZxZwd1`Dx(^PFgykXhJwdY2k~qryP`fF=NSqV+OSZpYF!(r#fCO
zVGkYY6Wu79fIdOXn7DoC-N9d<5tN*`eeAXhPvq-@+jmT3f&*crOz#H#p(upCGd=ry
z*GQPgIaJoXPCsUz{<w<g$qz|Wv==p0n;Bc`PLwuq>%*fPms2pQXrXG0@CLp25}k0q
zi19oTbDC&{B34f*rQV*OLJeZcP!IZRf==8Q0+9q4>>b)vdBzk<a4bQU-r!_?&pg0c
zsUAH=+wZ)(5n&x-x<7e<v*=Ep>-_CC?kBt;6itY*L>EM%#ODa#l~B=|xaTOE5b5jf
zz1(F(@3En3AD|W+!t66YFE8Krk0vrIfD@BLd5s9x%%OG7_&)m!8A7$>RcyqjgW3|z
z5)u$n-MMIe;DD0Ym@`&As3R~-y81fQJZ@8o2)Uga+cv1dEFtA(@gd>$!7S<Wj#f@#
zpehDBQ=u1h5<2SYW$~h@0SjGYywgQWU2EC-_-3AM4ZN)`m7kudh(5|+M*%b7LUW3Q
zefx43&brBVRKMu!f}#c^w6}Qe(FrZ*yX&H9U%JDb5~R`Yb6(z9yTzg!8jG}Y541K+
zH;x-35>v)V@!jtUb8pT|dS&PgW$cOk1?@q))YL2wdz>kDB~*x8%_*D%U;tEgWO&hy
zQbd<HRhUrtr`_-QS*kC+V$ukLdSa5IJv2x`X=)DIWZ_-Zqz}*)4qSOQqsg{jkr42l
z#N{C8ldPSQjhHyYxI?5xTgX6wa=AmKgB{qp2j|lz<-Nj&QU=d&cnf~^r0)TOCa@lC
z#_NHCF8VBxDSC+AfI_rluUrvBnZ!YjeF<9ojC+XZN5cvj7GVWX96$IaU9v~-45nIG
z^4hlGOMouqOhh^eIg+`=q8a*h)>=6QqcNfHxUJ^dY*I~33bRB)Wg-<yqXs2$`+K&n
zG#7br3!3UQmAMJ+!3CcZ@RnsDZP36btQ6jj5gzSCDHP@yUunBttArvU&6#>ytx8Mk
zsOR%}@=WJd<^AUOVfBMf?gtNSMc!rF&Fyk)ZM{dJKb%lp>K-Wa-ugn$5;*b*>miQV
z6&$u4i!9p|TGMhz{DI;P=S<B!o6(7bue7ZoEKE}$#%p?qS2!sVL(v2?y+)=9l>Q3p
zXi-B~X3{G{iY6*#31eWZKa9{4&1OMa{=HvebY=k+msYIGaQFcw$qrGjMV@8SGQzK6
zj@%yT8o`<**P!R`2TV%bj~S8aiTek|#I5I*zn^G0>vPJa(8T>i(nJ<%@Qgs|+pblj
zMGf3UE0lIC><;$cf{kpgUXU8<d7x5#?P?@;t?GqXC=wdHVm(5Lpu61t>1p?Jm#jYD
zpw87?b6`O$!~gD*9jbYRibkYfyd4jHVPVqESYL0@`9KjT)Yx$?hlM&_b8qR`Z}&N_
zM@{7Q?Tj7Vc~FI%U>DW(PFDolKC-)yFtp6HZ%WDKq8PAUx^VBL3=J(UpwW^>DEd>f
z9I`>eB8D^EAb=a2W6wfW{F6>UY8CO!d|^&u!9@ri9TLXY5Vi+Ju}06lM|5~0(21l8
z*N_oF&`qhmHi0OYSdbp!6Ls%zmSa0I9l8p8CwAVdp>+{fMGth)F5K;x+XzRfv=y+H
z`t6_vie!JqF3a6z;S6rw8jH{<7!Azb>)G!}k(fDCyDjkmw;Dl08S1fwf=)zDxL64v
z3RN2jx4>v)+Y&W&4mcT2)dMZmpE6fEX!bzsL*sc8^g-hVr0|iU$7HfR*?DEJ*YrH^
z$r?3xGGe3qIVePI_nUfOIfX&N*N4)oJuVod1@i=7poU29qIm?OCZxz#L_9$$(-Att
z_glnY5j7$8SNeMu=lQ#vQzQKWI1#HJDwoWgu&;yBQ<q?APmnuQ>TQ*Hwm_EdBD7{F
zmDSXyQz&k1Nt9p&*|d~ZUNnzjSK<PCB-}uu0&{?!jFyFS2#t0~3pH*yb{i}c5_<OR
zoLCouljBBR@1l7G%=Mx6ik4FtA`cE#VYHkAbohg^Bw5WXJ7(rtQb;D*e5bJBeSMHN
z>fw|wl&2_HAo0>8T(i`#u?aoKk>%O8*g7PB%I!mZwgYk;XD1;MP|D3hfkJlD*l0@&
zTBs%^!!m<V0tk)3Ga<1eV2U7Y0=mMcSXf?gQ<csMp>t3OxXKcZcW5Uc`$5K_=S}F!
zkqAX&AkjfK@Ls~@POmPhLZKyBONekdGDAGXnOD15`R)`-Uf3cD!gE*=$XRrcK-gq_
zRdibfcd<!_3i#KHDs(;6;qjW>i@YA?w`3P;I1ZqSBcO=TW1$fDa%gDE{jG7EKMl#x
z#Uif0Kj$k@e!*f1W)~+<ODZD2avXrAdq~Y?o*#3RIknw$;K1M4>67gbDrcGdT0<~r
zL5kH2E=|gKm2nDDH6fo==X(VUA4f+}pyi^+I9hr%ZIySjNV@0~u@F*Ra(9}(Cd%~g
zP&6bEH5qayu4}v`w6Oha8o4Sm_wy=pG89`^q6S@rjAsvJ=7m%!W!OW#C|iJyZNL}B
zFttEgK|K^=mIqK(9V^0!MU6cXDxhLiQC_!@t3cF*H7QpK^r4!siszjjc?DFsXsM&-
zHxJ9}V~Wt!{@xjM3QbX}tcG5^l~7FB8AfQ^FZxGNlO`h1ZD=QSQmqGdYlK|EqAvgj
z_)x><8$kv&x;Au~*(1lE&jnu<Q-55Cd$C(mWD1&f2cQ!+BCN%1IJD>>!G7!ZPsy_#
zm!r!_IIe8m9Y`e0omHCmCVF<(r?|IkiA{57wS*Yi;?s1~wq{@#(A-xg2$qvVhU3N@
z@@vE=+_Ig{qq7`(te_=3e72Dzk;0R!y3xW)bkkr*W@!ZQ7D}OTrg6Sn2qi6~LfMi$
zZW7xgiv!v%$|D3*XdxAf-UUGm3yLl%YsQD_J)+cnnF#I5Ijr!J7D}PWXy8j&$o_yX
zR85p~m<bMognqyJ%xf5EILhDdA@9HFAAza~*E^$6*rI{Ny$LmPZ-5J(0%nkOj%7~8
z!PP7^M;c2R522(@N&AT5t}0rY_vc&1-EzxklrRA<I401YBk{Co9x>d~bQO|7<tK<N
zoVSIjIPbscsKL!&G>^cx!TMVvp#r)%oOqOD=U8HtLdi?z*=WwxL8(no^tcR<&|O2e
zXVE+Yc4JpFkN48%!fO=4I=UgOEryRirRT96y6AwXudY0?4b;6;aAr}vEgajn*|BZg
zPC8CHNyoNr+qP{xZ`g6id1ITW_pkb^_E&ZG?O7MAX3bgaa?CNF^<eCy7Fq4fpD|aq
z(Udwd$$EJv#oXK5rSpM$+d<QE)m}p&#LXWVG86(*PkcTaio?z=VM5WguF%4~Md+?7
zh~%ziT3kIQew@Qp&(E_L04ER~D5TG&33oYhAPE+QM#yaw%|S3%H0_pr*~657-IFX4
zQ8L*~_G+wI^kO$yI5VX-n9WV8%T1ZLJ`n{DiG+U7nP{MItEwG!x%KiO(w(*M+MUH6
zVXL3hcg9Z8;vQBO4wTjY8Uy(yTd*FtRwJ`AbMIx|AFdwGBuckR28AXTF)$TzED|<{
zdy!MdTgs1!DHL63ncsPxN~$5kM92~5hsJM*69Y|I{?Fv&hqHX;ulj4sT>Oh!NxnzY
z$gFL<Vwht{Yr_5n@)pl_?_umiyiQGcg&*@WGCJwpayf{x%8NNE0OL%muELClDZ#=+
zn?tG-4$%%S*l{Se*gtXxE-5xjLDuET?6$BMJ;-?yxukh;9`5;z!!axnqV^Ph5~jlF
z)gtzETyy}zxuM>}>84T$Ui_N4I-<TK2`+@>EBYnzRpCVl_7sblI8+q*L$lQ^Y$I##
z3VT6}-w1&=cb*0ei}Xy!_9q=b#hfp3ip?R0qEfo$qosn2IqY88;h{yav;1Oa=Mx&G
zREfk2NK#Qno*yL8FjH`8N;hOd1Z=6Kx74Az=ML_R65OO$9ulBAZI$y}87O&E0;ga|
z!1(JP?O|kZ!kletj;>cu6-%`XX>Atw^K|#%-tGg5?tN@RHvtnexR($72*g!{Wn>2Z
zJEjn`qs#n*4`eMZR3eBX=xCs4mL3a>#K@kFLdU2yS}Hg`W_2(wWCX<yN)7(<EoWty
z;YqoP6%gjFt(<Z#?z*Q5VZ{V=%4ZqAdfAV;*RR3Q>2vKay)@l53f2mm9{6RU4{M<Y
zn-(?B24p0%tkqJxkcg=;shcC=<h(b3M5<##qVtYQ&I){MdRR6is#M+<io;fK_GzuD
zW(6&JoJOD|*hyh9I3UZSY6SU_u<!t70mc5^%y@>&^K2^roV(aeqyO1v+8Z#oy*xF%
zYM^4d(iN;3H6hYNILG$7zm9v?RQ}6xeT3yB-}Sbw?_oHP7jF$vG!3+gpYShZ2~)Ep
z`-CpTunZ-a@wG{!w1CE{dSU(v`B}6FclGWP{A#I+p7ZQy&<96yrn;ICnTn;fdwWBx
zvCh*bZLRR@<VQxm-jYx_$_AI2=t7&8xe<i72)2%wA+}atZM1*jx_Z7lJ!44Tunyke
z<^=&Isf@y7<F0ox<ndAfn8q}T0;x+QV&?qu-V$$cIOvZ_%d{x8-wMSx5R^X*BEK`!
z+u%zeGmrKh0~y^xmuvKJY{|QP61d42$h+Li41%|Q&l`tZK3O9ovKGTArJ!>rV=AS^
zp$@B6GHjmU<oLoc8}O$;!&r7(CDLT^gf_c-RGn^JW|^W=CA0@m+D6tLRhG35^RN$*
zPOB-<#zE1<_sWs-4{}`Ct*<w1sB2%+H?|dlfV!Zv4vaOS*s}49V1yW`?h|%%c~Pk@
zPh6rZIZVrLkaK|&h#ab{lnH=gmO*Hk(|pMyHppHDd@`ow&)AX6RYrxZ(Vx)U`CJ}G
zX47JjC13TFa&19WEeknOE9nwdgzRHY_~?-{8CL-#TX^fD5okcC7(LmhX-SaxWO|x*
zPV>T*&6Z{{A(+`@I+TRglf^2l)PCBf{z63LEzAvoYF;_PA>~0&R=6qpBhcvn_x0Q{
zvpH=fwdBYF@xcR0W0AVGSB$Q#!)s0I9JSbdWZt(ChBz4agIPQ%Y2C;rCSxvet8m}$
zB}8k^?D#-49txPP3Y{Ycpz`FTib$n9gBFgZBCp^rWz8%MoUs6jbdsW<pZ8(|X4m1W
zKqc4crMC(cK@EAR%*Z4t@qQIh{e^Wb<*Sdy@7N9_7HGlcv?Gbnv|J1FNB$wP!^l|%
z=|Rw?D}am_6r}e<by#Q&b=HCtu8=~rr67bA5SACm8Ga}@7RwP>hyNbIWGtt8S(iAT
zFk9Rd*U951^d8Im%XO!PU*)h<1se0Hq<TU9E6>fIbsHrC_J$kgD%+x6W1v<7%Qu9^
zX-BpFU$KGCYTZKPtLn>;L`?@D-0s;`&YkbG@M%x+?7>!ehYq*b$Sd;zZ3lA8U8lD$
z%bTwK@|iMwVlWY2PeC+&7d$b9<fQLCe}jRPlefQ6d6~9}3tL^E+E{o`q#-YPl>`gM
z+Wz*~jlSVAM70)Z)K79Wc-d(~&W(k?ksB!6X1?oGoHeQsJ}CmfuOWQGp$JI>`_^He
z4ibw{kJ<~ed0lXiM2@yQ0yhy~e(rn9g?Sqm`#fP&mY(am`b6SGKirDn&2<ch;e2rS
zLYJ=a>*VieD^ud_i+0Ou8g9PbCR+T2Kozh=-z+7&s(LZ)?1{+}1H}D3+`anj`Pcj$
z9#7Srg#RbCxn5HnV00`bgn#mW@4K!H?7z-yD-`A}1YHQt+`X^^Z~2JJY7rgM?<fRU
zJ|`*>=I0&P%+fnm)x!|}9g(UXU@%*;PpirX-@1ZnXB~I>aMfq%F&XKKz6r56#Vp22
zPZcz=jfv@Op_WNDRD8-7v`91!-og>i_^PPC5r>vr%jC^z*&t!QZFjbHK((pcD6gt%
zN5pPG`+!(Q<E^H#MpzYBs+qHg{^nJXlJEMM5mSmTH7|3@Vbj3nM`8=|@|hfuAmqT;
z5A>mB+E7m&GHWiuQdgwMNP5hxM&_(zM18^6+1Rq=*Gs3HEEKyW1kt0o=K?Dn{agZO
z(jhai$f33XYeUzxw}mdT5dZ!=fyrWa3)=##m?~@N4TkPtS!}|JXO$8IRmPNvs732M
zQcy6ury<(w!u=)EvL4KB>=wwl!6Lg%gNE*X!tiHxtr!l;;hZphgWsmemfMtY^zFl!
zX}Ho8coc0D;1>3#pKkr7&|rUcN^f^+nMielwH>m!^iL&}pCS87c>N-86x|i46HPe`
zs4C4eu`sNg$vUA@xKbxdpD$s;w0+7MCUU_^<kw6{=0YSN;R<YHF+tSuS?779tcB7a
zyo6Xzls~B@D22X{qeXGo@@>yS;|-!}U4V9lHqwbJp6>O~rE}`V*8~)E&}M^PR6fd}
zYW^EDP+G`UEBEjqaMJkP^R!>H%rQ=ub_TGK%yKA3{80zHV?Iy8sZp}5d1mv25@`O*
z<>d#e?{ixhIyZJx56Tb<97{aS$(oPoixqzjnzG7QwVVORaS7Fq1WjlH$NWL3TYST@
zrbK$gRraEewFyf?Z*9Dhlv=w&I?Qhe^XMR8Q|bAoy-gL4?HS>Tv0{S1q!AJZhp~><
zCn4{9k4}rU?o}y8KSHOee=Gk=)`Y9{@(&ji&QXnN_l6ZQ7%w@O88%PKC`m4??>i2(
zp?3jX12PY}l5xvMoh7QFEtXu$!x1i0)^0kl9_<(Z6sjiP2w|s%jY)x9-I)m(d;S66
z=Mu~ryZKP|_$1^({PpD|0mTVD@|V(rn*)KftTfQ30>Bf^f*&=fxjQ;jj0u;p`Ct#S
zPn`Y+5Hk>76O~5yJwV9=H~W`v@54up?B9r>vr#Id&z|sYWfDw=3xU7nQQFziDI{mh
z;_mk(M2BqfEeQ0QCb|snZdu|eAgO=kV!0xzP+dY)$?ST^LnX>F_8J9-B}?Q+ZVasS
zT$2AnV3KAQ0PZ2A%<MDqM9<@ppbOXov?x?_1DL#edUORIJMPG&(P8;l(G0J?-~qm8
z64^WtHO!?RH+*-_1=C!Ao@Peeod4=6SWp*)%LBeH{@!^409VwKO;-8mfmn&!6v{0E
zJ#;Ra+fvT13&H?!CQ&VR5UeUMa2X7pY4U`gmEyu#Gx!Jk#Gn3)R=Rl=w~qG{s~qvd
z@x{CnO*ZY-&?K=v;nZgCas2hII~TV15?Yv}5dGJ8W+8a3UlC0+df^aWG4W)qgsdl&
z^9t@#Di=X-bS_C=TBP%-R{~*t64m2qQuT#D8EcjAlmd#=v5W?}#Bqzp7z!w0G=)+K
z82J{Oec&a83!u(I&}o$KH+)57KT&lPw)=<*4CI;Cj2AA^Q__AEK?)p|{Rvf+x57p}
z;bCjzW@($Li%Oht4vD6Jh@q>5?BeZI!d}Ah>>We<ab?gc89y&`j7S^#8U$=T1KUIB
zsYrYX3}N4gR8whBk6Re_hL%mNns>H0Y18OR1hZ!ct3iVVKEP(2xI%bp8r8nH=683X
z^D{0is(CWTDiy$u8w?czSx!0GK0*JE)roRFi+og9cfs(wB2{W^HF3>1BP^eqiQDi%
z(je>yq|!KphFa|}mCn)rvp300VX&J9n{Q6%c|<io<bzH<5gy5g(?}6dZ8{un*enK1
z$tmY{A1{bX1-Eh&Z52}hd4_2zqGw&6<jKPvgC8Wc61&I@U%3p%Y0D#6k0dRaQodW?
zJ!o7V)I+%RoM#EUzj--C6PVj~NiwUxHU<WPfXO(`b{s4!RK1yL_V(6>P8Ac>vE^vl
zOid%2Bi+M2y;kJ7wA)!VDZ2u9iPmBvR>Mv3FuW*A9fL7MxggNlaTr$!8xXAdxP_x3
zp*B(!VhJv>OkfRuTaYof@&6JlR8<-?;TaJcsz7zTi=q#5lU^octkZ)5jT#8M*Rs{k
zgV&5Cin}!S42gvMv9s8c$;O>O@rO%ldV~KWiylA*^G7>|B!R??RlJ7YGvt6dd%o+y
zj~oHcW_i8Xy#&wq*>&fQ2VDb6ZHQdc`FEu_>4V?1HhT&BhR<Gf2HVN{tK~DL@d#fq
z6ygOB^+#Ol$%UeCiiVrWTh^<Tb?4q4Gz^u#Hoj8c#qT5uAL-3>Rfwpmy0$Fe;&m`;
zR;sO#Udf_r#huiRpA5j(!*bTA&Q#EXvN?V+$J(9-hIMbdVy{Xl*%4*4Gc=_bB-Cmf
z$Qhy<(bzT~g4VwsY!zyY9Sv_IJ@XXik@$C5`=%KiKA)Mc(UE(gs_7j1M{FV|Jbe@w
zxzu9ERp&MpO8n583>0X}z8=4iT_S(CA+J{a^4walzzSyLmm@M7$`)}ep3ChQGUV?@
zqx!qSgVE*!+UibBitUpITY```?FCf=1u<WA9y4G-D4?j?&2*w#T>D8H8l`>N=|saT
z9ghH?3~HsEBA+3gc{_q>ZyVR9ohkv#eLRdYJbkC%(y(Aw8%in`45JU7dI!`6WP#3x
zyuCC^5gA2vmsY(pneMecB1g@M?ZLa8Efw+m9A`tY;gh0UuLqx#tJgZ0{|29?<HKGK
zsp~aqRA1_#ov&LIAaQ2O66I&Y6E83q^?dH@vJ_1`+yj=5^U;}Zd^~g#6j;S7#?Pb4
zCFM=5XO(>%CjzY>us)PMU;Yi-S)LX@iNX&cj##PnIwB$#^dq51_v^ox7werW_d<xx
z!Sq@e{nyNy=f5}-9scVng<t5Bhwr46IS_Zd*aI4cV=7Tyk-eA6HJ$&g%oVkMkL7UR
zm=s?2&bkV}eVcCWEVRgZ0l+yRI*c2ren%zSoQ7;wG}7I<;_y0buDMTG#3R0>uh(~7
z6nZjS_HK~}<5lt+X)&4a&dYb=8aroE01W1<=368_m+T9*Ud^!1Qoe@)&S8JY*#(l3
z#QgS^9LK3=;yq0t*2rwlsnr-6ppaRVB>Gb#J_pt;IZ6IIq^EOx0oaMv`ubOQ+CyAG
zrP5Nyx|9m$KuRoL$oHC)=4}+g_qcWdL}1#X($9ve6sAR>FN>)Eda}?~b(W(fzRQVb
zhR{!N!@DYOxl&TFU5k_k!6TLuUKK2~q{{q!or{P}orhL)_!lEom$Livlx^cvFj-7R
z>wBk;kTU+foO(#(UwHBW{?>w@X=TFVmo24hG2jKTQ;2&FZq;rfQQ~YPPCp~DmM20s
zRwST8t<E=<vhO9ZZ&B^8dMHH;6a<jMs|;47i-r$vyyng`+~SE|S>zA)=1Gg-T^u_=
zP|5G>>mWE6@<p)C!}`G^e4_hvIi(7sMGq<ku_=`x)I45AuA87~P7$$^H=l4FiggR^
zkJDsewwAa`*6;q6zm*aiE^ed2&yp6j%DeJ=JN+(H{Jv^gji5CvsbI~MjT&R47&dX2
z+MvrL4OG1+j4ZSA)SsABfWOP29Cu6HQD!?qwWRe3n%7hztl*NPOUY85Pfhb1L@(*h
z{W0wxRPChRw)#9YCnqovr}vQ;2?c@gBG5ibcw*t0&sdmUUcSP&+Y^gxE#eDrF{v(K
zQ$u+HpfFAp5piQz?%>_<j>G3g8=YUu>)7Fn=7&oqipW5Sz9u>%ny8Tu(0|knlq{{M
zpxu}QCm}BapZQmj5gZ3i9UFVs$r~kV9DZ*02_+>$eP~_iX4NQ4N_|fDbo3d(sN!AT
zvz%%v=KQl84YmM+Iu6-{-EEUK=nBpkPNJZI2X>TDv=jJ;B%k3NYsPuL05k|o+l5&5
zZ)a4m=%|fld!1n%h|N#KHQP%@gsbFhq4=W-C`aj&JV(=#Sjo(cphyAB`SMryZo4N4
zhk+NA4(H6vE~oywKXOHC3^^zdjpaupX<1w*4dSr@%B<T3y|Iaw!7ufeY{iCk^9o^}
zvN9$>M=n~~-XbjIbQHuQD*n55lSK89BLriaW~_R&f7)C@o>uUgUsS~q8Y_Oqny3)%
zS<1!k-#k1ca=P7fm-v-*5~cfCtH){=xHH5qWhr>gG-U~AogevYOe<fz3q?I+eojzo
z#^SI)N68PwsG^5LFXhr81@dP8EKMT0i^77QYFHbaV{^3QG&!AyCpT|H5Sw||TLMQo
z>$?4wu+EVY$tE6TQmKh>XLJ_lCaU>4>SDZ9YMd$L5`Mv}CpFQO)##B?I21mU`=&KI
z{Tf#M1NS5fqh17oz8DeEBF<9yUe+UwI%_CupQpQ96k}T@I)7JGL6TVj>}<*T-nyd+
z^U9tP&U`ZZ-Rxi&{Segfcg3vt*ipfNUKo?5^W-=fBA4CRDaD1i^EYkYgLY;r*|=YM
zzQ`6ttA@59y9sMR8N&#LHz!yrTv<SoViovDdVGFf{P)+b&&RBi|I5oM;5t8!@M-Y-
z+r>`?Y63<S>;q5_;zhAgnb-gE1$c(;_rhXO<c@w;e+AY<zpWU4Z$r0UT9-P0e?Ifh
z3%yOQ2tI0)8ph+VF(I9QzgvDb6WqMg5a)+XT?u{fDH(oC&SQT7h<85j0oT$GvBXx-
zLoM9@;sJ(_?{oj};ry%C&se~D?x^ze*#|xtU--*(zW<w&zb{b<*T?6}YrfzW3i0i|
zkl)Al_1ELG|M&e)o(ceZ=j!zsK+|UQ{*uxa8t;YkxMcX5L;Zd%yz}!O*qZ`)*8RE(
zhCi;acGu78zx0wJnsMHI1$mCIUk%?v?MnJ~u5b6nEj1Z+mPdT42PSu!_6eZ1of6*f
z^z{?&<N!A}3#UJP>hF<-d|u{<Ho|3#pY6NxF>AxxgaoIHFMj_`a2|MOHBm=cJo*3%
zeNBIVeOk8eeAfGar3-z1-FJ-i)9o1UJqvw|vOUwi_%QA5bk8gK>^&0qC|6bi{J(jx
zGYq?DfUAu60#EUEtd`fS6Fx#KM#mTR^dTsA*gCP|zJz;<KJL9jIREV1lf?vL>pk*+
zerO5Kn>z|!7;xP)3GMWMmd6YEeC&dH0QXAcG2%Auy}(qxy*<{}V1@iZSjNInzP>+%
zz8=P}ub<n8`?|MT*4^ube9Gl%Pdog-mC_$(hkM68kq<1Bg(k}u)RA0$CY>>UF|%72
zg-kH+jS%0EFZ49$=y#3+UgBS(4L=^f&+n;N)^eu0{SJ7m{jSE<-%2vtKc{5zgRgJ|
zZv5Xzm%iNQpX1->9`AE6khzeCSS=a8)U)4rJY+()xgmPp-|Bk?1&@(lZ#iEsGwkbu
zifopzHmjnUbV(6FmajiHq_g*(lm|C-O5@{Oq(Xm(`v~u&ZH%+>ulf7gLKg`>QvJUT
zorTX>;mqyKoL!vEjcxyHWp82)$HK+IL&8k*UrT;|I3{^>J4=^eB%C}Pa7>a`zg^6o
zNSGvl8@rf`o15C3nZpSR!u>zrdu-UME7<Qdqx1r`Jdjg2z(<KgghLZ9LZk`WYGqqD
z=qSqYBYkK>#J)e@VqrC_i6W=~_7;C%LOb6dyRuK31lA(G4lic&HsuW0WZt}YB;4I~
z^z^Ma5deSE+6a^YN*As6(Yb_czFo6z3LO00cWtza>5FkS`7&dE>Iww~$ld<!{?rl%
zTGNwzs<~lRnGOy+)q98BLWFGLjZ8SjDMD>RTDJEiBjms>(Cjk#O+2=o;!_Ri?6|2o
z4YemNmJ1N8#DtofsJ%ByK#^N|;NdpVv*EFgr)|M{w9NkMhW{bI<c2h>@1gXv`Z##o
zU*%ZQExnFUr^1L|w_e98zOEqq)6-Co^V4bo|LC*1S$?tmrYfqf1!!lu=C|Xs_G|Pt
ztM<-^aJ}rqjjB_Lhd|11zr!ameDQVz6zVn%gq(mxSz&}I16(-jhL#Gv0ipm_ng?VV
zB@-GWU~5#>gTygFyh|U=S`|b(Jq)C(a9%TTuoH|~%uC+j1!2gf!PF)3p2uZoP@3Nq
zDNGW(7h%|H##|g{js_%}IY*?6#hAN%4pawcu0V2X*r5`lz;3uxkGS5w4X3{QrTw~o
z&Y@KtFC2lEr_1T|7ShDf$_fSG(<5@hFh9SWL{rAQ`bv{-ry^aIiJoU_C*osA^Yqsu
zqV!s><H46rKl<NmIkC@PvE%xj%FS`#bibf122n@u=Q!=lR5od$AujQ-ZK{_c?Hqp|
z`CpHm7j-GsPuyqs&_0$&RkY^u&iFt;Bss)Qh!-p*BJ5)l+(y!jvx64w${y~`=!)WQ
zmbXBMB}0&*E>wRDoZqR=jokPklK`^cDaA{v>&?Hh1We??h8DOVvZJ$1Z%rq1Lqhm^
zCv9v|ay#9wqSaA<Ah5(&);2lMH;{B5iaEF$4SISH7yK_bgFgNf0ytU!I|8`ac>V_n
z(D*L|VD$XI2yl}I(+DvGitRUSR;gF_ikfsKl8BHJyj`yzNI9%Of#3<n=UZM3Zhd;_
zD5-DT&QkasUaaJFIU8omJUKffTwUGO@e<4PeXy2sK5~0|nK1ck)y@uFHEZ-)XYUOr
zY1y_4ty-A*DlH!OS_weDH@tJ*TdESic{B-dy%$<h6D+q|tm#d0HP5820nwb{{j$H7
zT`tkl<8z$#Xf}Ayk(eFe%CJ#?^G<DW*lru9G%&b8iZ*pshD}fSYC(IWY4?0_BOW(c
zyBj2w{eH8iRucQ(dk}zF0`_tGF7-9P1BNa}-hBA45XG5^=+kRB`8EM>U#&bioS$Rc
zskcv#?DAcWBnlc@*-MUUnSM)s*(66klZ#FIP-V5>*>b-|-MCcI<&K<vLVxH&RU$B?
zvA}Ja=na@vR5G`RHneaT1&xZ}!AGMX!C^sL1x-?jku%PV6^85TyEk@8LQeXqtAB1k
zU|;g<8{tZs(t0A$S0X|3Q`7`nhQRMexa|(G7_S1Gn)TpwRi#&x<wsZqaV-6M2DqOO
z5Dxay{SQ;xIc`GkNknr(1E9T=U>hhHA7Fuj=eP~`By8i3L}o~}L3JW~2@;~Nko@JX
zaR)uOTaG74a2}|0*%7;j3H6J&lWND@kd740uMY3BBJGHus*4=Q1F*&~NR{Pa)FpYO
z+@!gQVR<KLpgA()IWRdDB>nR!sfX|ariwnlq@AjQ^HGGP@<Uv@y47=B_W4t9I#m{d
zlA(fJ*2m7pWcm1*?^_ikKuOwsd<h+u`C`VughJ1rs-w1f>Ymxuv-bBTqpSQ@uN+Fe
z51xx``P78}tkKtiGSq@fURU=e*n&#Bul!b*_a*<1a^Lw+hk!Eaf=am0o>L(JH7eh{
zruE0UEvB8%#aqrj%w9pSmCox+?~^>;=lQ)z8(nAL*!XSK>aB5sg99NbNr3qca(Gy|
z=YK>C$A3o)4>R-s3@w|yC_U#|4dEcCgB(H>s*DKWC1&6}58w(nH+69%hp-XmVr|H;
z&+*MR*|v#dF)I$8e0&69?)iR#b3Xn4acuVzQHHl2LP`cX96N2{`&V__?mz<qr5)1?
zQ-1>iKTM%=-)9`9{XWlM#AAC8Jd{E)+nb&7t2Ob-qCSLUIYEN^U$kF*=WgTAqZhCI
zJ-K0yREh8GIe%5>q^8M?_mSo2)BN{c2%ZkK3j3sK(Hn@~E&7KM#!dY8NRPEFn1<Js
zuvzzUm*Gk!gj~v}u|M6qDWEuo`iPYrpPBnEqK{N({7Xj<th}hh@}pnlE8~}#`=Z{3
zGD=g&@Aq!721O@KTG4!cWFVI=cG|mXj)@UH<UGBE#0yXLSgP%I#t)1)N3Hn!#!P6B
zWa+uGHBiUD>NzewrthaY+7&ySWQ25?sx%tac}J&El=VO;1EN6mz#2dZK(4t!_yY#?
z*g=-n80>FB{(x{oaD!yw6*OtWg9qSgLlJ;fNsXvSBLw4x*?8R0XhhLHq_K{UHi=4(
z;^J}8hcraB2V{&oq<P`P5@BKd9u-Ce@mhlfX@}Y!nB>8NF(AMq+_0`R2jd61ZVn+z
z_C6MAhG2)`w82wmTGm+#z#Ftw0Jjct{li}Hlf1bM@!0x7Vaa7}RjqdAF{S&vZD|xm
zmu@*R3IcwYl4F!%)aNj2Kj{W>)aMRSTXx!;G}ufABTjNrWx7p5KBt;vECpXPl7Emu
z_K-#pEt)q<gBfuj5-9A>unw=RO&6ykWfRw^#TK4~px&%O<3Gr9<Q+%bf7H8Mwo1T+
za<4<{z_zS~cfm2k*2y(m!=5?Ppvg)?g{~Dyuj?r9DotlflW&=_YvCa&^ZNT|{5s+e
zF)TQ4TQ6A-58Vxu4tj7-)ICRy^6lqJM(MtdR&evtz0<B0;rOmaXT6xOcV%!^+IvzY
z!?P{zPc=V@*BXsTovl%=-Cb!qd!wK33vgeKHK|b0T6eht=>+dljQ!O*&{N14EIEnf
zf`3#Mr&;DkYkA7fJN9sV&UW|@C|CCItD3%h=f0k)-gna1zx$@Xo;lxlx-a|<Ex(~N
zKZ(_!{O8xd2hG02S>J_1F23i@zPYFXiz)56r)f*IV?yry;_a5jd`fCwDkrT)>0tIF
z%wwCuZfnF1(O)N&J$>CTZ;iBH&}pfsn*R~b-2WZU|0j{(sF}Cjlh~5tzq*9khV~SD
zi%f<_0Oc{(2j^$y=>|0w5f89UxyAJlq0ugeJQLZPOG>NFMpr+LvEB(>h?%PzVtV3x
z9nup;gU6>M-i-0eRTqiE`ooH1Uf9?0sW(g+z}NPz%-8ZcKXg6v@zrAsKf;J#n^DKs
ztu6oY)5e+k_gic?-NEI-uM8Jck*umZ&hmpY*3U9uE|IC%q*C)v1bMx0uJrk+OS=-P
zv?<^VoJI?RDy}ZQDMrU=Tfn5Unz0?cv6Z7ZWOOh)78dmsCOy(3M6yhnq<%rTC~|AZ
zm5FOCT+(Yr#as6&(}riq0BhWs>K%@@Dj}SQv^F>Z273VYav;!js19Im*^b3qnovib
z8DZ+fJon)d=y)}bJ2p!FH9>1{vkMCp3E_qZMSh|{H<HmmL7xL(WitYbSi~LhPY~!r
z8UzoKMu(iic_`Wtj(Z+-9?TPB+>qsPp$_(w7?vI+m(MuD?MoZo?LDOi+2OyHme|Y$
zq7U5=swzP$N(+cNi1Lyn3M^29aVJM`qjIZ=_!W}=n8fxP%XyhkI8=ugBK3<Cggtd}
zujD=*@%*vpTwDT(h4Zyro&icq2y*>+>QspY#HtE%MKu)_NE!6-`Tb+Hs7?OLJ5Hsn
zonx8kde6l>n>?43`w~k*CEh#F#YRA~qEFt#7Ep%CC$Ht-#s*O4zmA?g7efKbjJ|oS
zSNAEf`Q`F&{N~N)xt)f6@1+N}Lv&ug4^_@Po6qwCtyhJ^=&LQqZ+OJr<eHuF-ecqb
z2r+rnyTseQ<q`i0EnNRWDzb3!u#)^=&;OnSFLi94vn5f!>I?^^I!))EkxCRSH^vmV
zt#bdUloRK~K+8=SOM+o0Y{^r7UTz;CLEy+p51s0bnZgrJUpc?uT@a1^=?$$-54M*I
zZib(uNA^sTwPA0~Xn{zFaW;;n`ZJrr>UHP#FuV>q_){VVCN86-fFk|v>fPRtDAzE5
zO?{X2c0Q7Ug6PZ5(@%rGucN2q<?->Tn(p(lT#ISWUp)$XO&AnjT%CSRmE^1|ZI-3T
zh_C(rrzz+N5*pGZBqW)(E~kVx3Ij(pagey2>JWVpsxz!gQwkX^e>K@V6r28!+6(0B
zX%lZ`V-(2A%ik=SSk_h=wMG-Ja;IF{IH`|>jw7^8As<-91h7|FVC9!TD#+m?Ya}4W
za?*tzs*#XY_EPw47WPOab8?x@NE@wH3QZS1sZVuH2({uGsOhswt|DjCNDv7DRWOeh
zwX+<`I7QhUolAIHe{zcEn$l*%%qUBVwz51E)U-i=JPO|>w)sfn95R2PXzXu6(&gwC
zcgl|b0+X2RHGX1L#=c_aA-!V08P~y)I@aVyepnd+_%NjQ-m~>`#-V>ocSXpVJCd3c
zEo)>N)`hB>-ZK7_L>KbhIaXBNb%P;Wt_<E8fiXcRXWteTA7wytKw63#RwqRtmdUV?
z@o1k3iu`#q-eAxgD1)<oPPtY+c&WoklU@^*4HHOl%yEe+sp8VeV@=5~Yw&W+*<yCh
z9^_d;UXX~XyfmZKgmy%RR6KJ*%8Hguk#k(x{>0iLqIw7ldB}+-1fSO$3h4j>Tt^r_
zK2-G5<QQ&Uc4}M`2O!9V>tU`d$pbDtywcw1n-xmfC~Esi&=NT;dj~r(+4+S-w?>Z@
zmJ!uB+_uanAqD<YANus=-fvr{5eyEw+HCUVKaallP|pEf=jvxviRO_+@=)0GODm#L
z=TY>fM=D4qGL~Sf1T44LAcll`v?U>*hif>WEZKZ9KlT9F%>|<}oCDy>34jPM3vK+P
z8RU3{16-7b<HD%4k7<ewG3*J9=@PF|F0P@<qNY<;hQOHv8D<+3xY^m&Ok#nhZjn4O
z<K|hw;ol9H^S7fwrQCR^eq`A*vw?YtH*}{=%UB?I>x@y{BZ&bn_*cL@RUil9hZBu7
zI!y$XGG+(f7fcy(Rjq#ZdLG)(Q|W=r8RLOV3+)MSPiL6=HfCw;Rl4f0V8?5_O;0kE
zGPEEt`wV`K%ko4!w5H4IIqU5A%3-?fkEhENvzM*4Q534ramo!lTM66o6cfnToAYFX
zkCk-EXcOZR^?wN%RlS81PX;_iKOB}{b8|66M<0r_GH1MbXG%E2A3C~X^RotzmzzWl
z83~UCFEac6_I6jtXSUDU=QLD-LN?KMNrYFV$or#aOJXuSJjZ>3lk6+$Rn2s~%Bxn)
zVJa@Z!4JCvv69~LkLSIA<-p6Dr)cX;tvo(c#!mMj6#~fRFpdgdFe)a)f;87%J;sJA
zPHS+>G8?G~g^WHuWd-fFa}{6_G>B@kFUEs=qtBxXP9;y-6_3w43T`fbLo?2Z#yLy#
zF`sHT_eZiE8@pti;Vs@Lo5*0gCsoSM6@ynXJDKg&v6%PXp!*Y<i6q%Qh6lh-IU?Ue
z&%h9j-6?)?=Rn|uRxW>fOAn+Z`+G2;f>|QnMtn_cL52Y}YVM35e+5w^&FsavMo*nZ
z#NzVB#->`{Ds@ErTiOw>8lB{Qj$XF*^AqH8q0Q;lly=ULd}~<{==R^K>^#ur185%s
z3bfl7@^)q&*M|}N6L}B^%EU#{o}J?{1q3o@AL01gU1z<)wLaW{?a*dsHia7QmKN85
zZB4FaprVbE2<8i$)WayM&Ab}h*A;kPTjwYIQjp^Fe2P$`!e-xubT;EAiQ0Yd8$(q6
zMOyA(9%CACr7z^Nvxf{tkpY3crQB5BhlKIZL5*)w6D1dZZx^RakDoR-x8dbewcK!a
zg%;_jmtu}F+s(UeSd&B?3yJZ?yI*b^e2$yd<@^@-5^tejZ4pPUPbZKMUp)5NTym}K
zKUn<}LvEkY>~^0Xo&B*M6K6ix8`-(}1vsU`&eO>edyoZoA&H4&anQ`jk4?YH7hm{^
z5x>C1;<#S^BNK7{ciqOq`R`}m|B#7Rv~=w8xlsI8>(``kTYvrZF1-=MTa$<$=~XWf
zM#E`IYr@_Z+a8WuIB0k85R;vDtr?3`X<;H8HGP@oVk7Qk??C!>^LSBxba9xoz%DI=
zcT1yP2aXC^_|E~#xM3!!-8l_SG7igF90(LL?9uhR-yU5E3U~QBJ33!o<9k0k%Dy(t
zC+s!!^LAn!UCTg#MI`68X&4P<j<#4PQj)>DXGSQqbtdFze>_lg&t<UepT!}N1IL09
zp)m!GCnrNAs+r2f#a4kiPY#&Q7}@nN4t~4{FE4tav9L-2p-W`00qJFFn4ySF!pZCu
z&=pUa`yMK*%2vg^{w1BrC<;pSMS%c#l-%VWt{Z1SAP_QncBsqH1%bJcV}k*QAB!|X
zyLEfO1yLiDcv&x?HU1#6L*jRyc77)iZ}v|-dCJVNi1dIevwAxnXeYX&+phdZ0~n2-
zprMxis%+hl4M<IPW<08JFkq!T*`I`c(>&}yBkQj`NBG7UYWC^9+U%)wsHkDHEgGYG
z8;2kmQ*&+;M)(V3VuX$WJ`2WDb3Vq#vw}A#4Fj$jfY`Lv_5D(;10m<uc4xgmEhJ?Q
zH^=3{w8g{-H3kLd@6UP`3ekuinV<XmMq&e%L6H3C#O<EbKVDl=i8wp_!qoGG@9E{Y
zeu&vw+Vq59^Fk~`Pggzb%9eoMekZ;#$7<IwTOYT;gawUs4prQoZ5`dVe%*Uf2R-My
zS3&z`;yX0w8O~<h#@wH__FxzYd1*2=o`oE}?h4mx5y5A;RwFAo0P2<f+x_XI@T`E{
z_M${JvX+T>l1utUKB}9|L~>v*#>%GEuAw&MRCZaLnoe@P?TLCSPnr7C(F(`c=CbF?
z=Z;!g-Gh0imaliGKmvPeU*FczY%)4b7zzH+!(inEa>jHQz*ARoGx<IxmS`ubghUj6
zPKJU$?<jVR+PjQ5w70tG=jU5uIbQ`l*I#L}Umv?yPi0rp2=CQh*7t40LXUWQ=HJ-5
zot0N*JSLO+dZz7FlPem9o7jEQV0w;b6^bA*%~}wdFsGuDr$kyS>01A;eh}V<pDk<s
z+)wRFcgZ`=L|HCQe^6b!LAf3qoq3l2c$)(X)^bel<D89coJcm8DPm;0%bT*-d;j%q
z_bQCP8-#D6<nM>+x(*+h70@uXO5|@AYKpP5@%@^rt+IWX>Y;Jsv;7flWR5zu-L8Sr
z5R%Lc#Z5WTS;5Dx!;FL?&j~YjyxlfN>OqCeZ4+Bjg%sX?RyUdlTWl40Dv~o}r;Xn&
z+`?Vrfbd;&)>KoykL3PTJ0}cY8UX})F1Za9?N<8s;cUZ7=^$&Q#e77F{3UZSCGa&8
z9Yb3DT*SqdK^H|o#bs4Vsf>lrXynaBCQX3Nu-&b}08Ba5R6AbUEeRK7t}{t!PwHdl
z+WlBLYJZEE)icpfxX%xZC`14g1WYqiRkoZEm6*4K3&!D&{MjtktnO2(=1V5KAYtKS
zi(x?<<TzgEkp$<|$^^_R+s_IjWg6&rt6gqY4TC1PJx~4eaL?2^da@5U-~f9<<DpPN
z5T<MiA{RDYT5-nk<lL)I&j@>=fpVPE%VB7((mgPqInMlh0>`!?|F7bjfftD|{fdBo
zYXxPT=$-CmO87A{kqRH+$Q=s8AdPwrGM}U+>M1hIH|SK_hYm$=ZJ1KKufNi+`1TBW
zE|&r6Tv0&NQ&o8b>SgZQkP>2+AXwSIJ>CgvJ46$A62L?vE2i&PoI`w;$EWB&!|bDx
zYNkcEH-Q(bH@USpwvpY{TB%vjXF%9JGNkDrusn@;I+3c<T4nQD;wwe#9;yBOXX8l{
zJbPB`SUvDqR^uqUPdS=!USKW-jjmtum<NJD5xuugQ(JK;H17aN8)vItEQ^?-!ZY$=
z9&H099Y&#FE;L|ZU(R_+NMWI32&98%<g&AO5GzNB__MM~!`+xDrT2a;=Z3=Ek!`0;
zZs2ke-jeAeUDz$=g%KTDMWyx>TZbf%R~@^-TUo0)YP4m2jn`W6A*~}p8fjV$_Fgkc
zmik*3pDP~i;qQw%yO%J{9`Bik@f8&#4-~fBs@`TN@K%tMnU4mxMQbC^=4#Yw$(?tT
zH0~*350zoqY=#kHtcg$DAxmZT#IL{cRArMPm5wFOl|0O}@Ig0S?_~qed^ry}r(<zj
z^V~XJ_NzVA7LeB552{4w82O*bjGO(xlNl=;JIDW&%zCxuk~ce0`vCR3iE6mUw-k}J
zy%c_C)R^AHd9?%=wHV<+k~69$nk5u~ekE|kgn}8J_=8#kb&5gBVq(MRRSX3pw5*TM
zGkbTfot`wZ1#wG$xhu}d2{cMJ(Nfu~Ka=k%@SdON+d_SwvlVe88A;-11S#B=*gM~k
z<K0?9hV>mqv`dFxrL0+B=T9q}4KH_3kLQMex4&+%{^f6P?pB0q_vy|QIvaw~YB_oe
zV4XP|gP1Fmy9sli96Br{HVB^u2q<uruG=0t<h-l1N;TH^XL|>m-OTl_;}v+8w_hx{
z@ehdqJdRau;*_tnfM1=(KYz2V%<mMat9@JCAgwGd`)p;4^#&uFn+nL_6!JG^{7@Zm
zwPX1`%U+?qnN7xUtLl0Qg66rw(ZQU|w1}*}&^O$xvCckb|Lj-OQ`>>ly*aFC5u7sp
z@P74Mao3D-G26)gcf)cJ&D_GSvEK8^G0t3*Bh|oSZ~Vvwg^z2UaXIpQ!=saLCtGvD
z8$JL2J`wni7{&S1pgr!{c4MWAQSibVPc<k`XhaPo&R&afHMUyjEZdPL6~x{>j!~)-
zXqc-Dbre%ieh<`)1nDuquhmnTfgW{GDQbwhW`d|(3r6CESj-4D=bek~+goApV0hi5
z6Yy%S=JW5kuWhXF$an(r9{5J?S!-tQ(9%;!cla)MrCqQMNS|vj9%`ZlR&@*w@kWBF
zj>p3gVpS?+)y<JP<Xd0$nG^HRIZ4!Od7VzaTqKkpu}h!Oa@LBmY1p7`P*rnla5;Dy
zhXXRuYi2Zu=JP*ggPVvj-tU<)7wJhi^it463+r3D*#~8VeBn0K4V%P+UhW<i-E#3v
zZe_neFB=%b9f0kZ)8HV!%+O5x=wR-fcZcR-cKlPg1h{lK3$h+0*~rS-3_}UL-wnV5
z^fvlnbhE-|X)5V7mSxvXvNXA#-!{%Pgu+#|p~!Vl77e4o`7-dIO)dsVew)5a8vU<D
zFen5DEpSg|Rb25eyU*1y_L<|^@Jq^CM2^jPDir1Pi7BY00vAu_d=ErBNN)4^lI-3J
znq$!;1HqQa{A=i7(5<^Fz>!Z(bFi}%F+%jyS}ncjd${1_?!NkK{GA<=O6*CDP%8Bo
z-5sxK+LV&pNP5fK`YLaL0=q!!?1KArw505~i?Td)ChCcJNtT5857Hre9Z-L#*~HrG
zpgvraV9k<uIo3R{I<m~R^q2`@dw0^ELMjIK0ghZj0$aIrQwSO$@*IXdHDpc_ESK5~
zy0)YYl)Z&ij?GfmaDVovmuYsZZC5@^3cN_g6wsP;ybaDd08WBm_grPi4ZN>i;)TT_
zkj_Ax{#G?}SXnfTu^L%;W)G}39Fp~s7L(l_cv$JLogGW(_*^uuh+RHXr-!gsJv@*U
zAT=aweJo2Ddj4`5Yfhc`2izLwnji;zCL9lAJ|=3Qie;@685~1mQw-wfFFJRMCdVAu
zN78||S=WJFYTsp;vZx?Bs#dS(?Xg5fclZ^^zlJUlm1JyYJK5U#WxNz{mVAVyqJTau
zuiR6|(Pvd26?+>&qz-mG98Tmq^oj?!U^>6}5Rv0OPy)%{FnoI#`1N++DO<(r?gU<P
zL-}sMB7}?~;fK=(7mDomE;yCF1j_?`?0)7EO=10RVBjlGM*0_?kVL`f?WBZ4B4t|o
zj0_^pdkiqlzj>Uk<Obrm!Miy#$e_}+{)q2ItcJwiN4G088!UElXq*NmD;<sT1_cMX
zO0;t2mq*7JT{&84+LVRO?W^9L3JJFMH`{Ts^zXi!tJUfz9G<<KxC%nYTw5`Rmr{`1
z#7UWYU`%UJ<+A`R3BMn`sXtR{4ovKE>S;Lt3$rnpmC$NTlY0JQNsL7fa)RWDW0B>m
z4TXO1JLWdSWH21|I1e|geYOPPymk+ybO`WWzfI0?Mso{mu3(VpJ1jM2?h~LA??(Q*
z%DzM~f%2VTM{;wBhU5<2BQpbCHLeMhbUi=NtaCi<d_DO5?5mN_5#he~0686|RB(nh
z8}msIi-hCFX?b9F4ZTiI^y4xB|M_B{@(bd;KNr<;h&wT$6&J|rL<;T*zYf)>)+Nmi
zh}>l<9{RQEc4;ku2;+7ZHWY{)I#BTA>jyDz3VJt&4S{6WOym@<-vv4?&zg)A?Sg!l
za8+X^J+MTXZFH0QUmpgBUrZSP#YOt^VEpBmwVzO$0)4(xKI*oB@)WHPhmjObb){gL
zgH;xF4(*Hn<S=k@&Zip#tuDfh_d|(=z9U?ufbE+H9ky5;Bfu7AQYJB)<Sx*u6UQe)
zEfKPEJTw`@*af3wVEBic1HoYs`k)W%OX$pd>dXUsZ|>Ca0gPmf=D`HO-)8eCU*C?q
zFk`gE@L*Q6(I3x2yGsjDrhwBnbC!25BV%3H-)=+3*<EEK_8ItF4iHFcw+#YMTykZ#
zMiHw2EMY|$PE6jpNmykQsK|(uKhmk2$VD`ul%KQSWPVq{$k5E7DWj@RpA1xMxoh<s
zLV`HKQAZO*i418_tz4ab*0bvj%kkyGR2(&<Z|mT9HrL4fHTZ0~@y9PGIL19%6s~-V
zd{zYOBBve<&B5FiCGD}@VbGvcm*Ol--j*M1<A8u6L+CrzBJf)7Ic8F9&PW%Tb<c<X
z@N9Ilhzv9i^P}Rrt9AxIAYrcz0rkev^)ayHJ^|4{vk0$=P-PQR6_X2DG$)Y!;_v5Y
zS$*{&a}aD%;1O~niY}OW3fyuYk4Nuk8Tk3uOerlFt1EPSC4)~3k?D(ui7Ro$!wCoW
zM5)_(O=oG+xqKyzkRU*{^QRv`hoO|>cQW7-vH?wU1ZfV5qPrY4yt=c+OuC)^m$=nv
z@5hrTXQk!Zh>#_^Jm~y$>Qk*j!NV!-rp2u;Lnaii`kP!)>&b3Rbl;c|A8f^K4K=)$
z$zE0Yc6t`b1zF+qH*k5P9QS5v8RCbYvdy>sL?1gD&{EP&tNqEN561u9y*2}{Nww1c
z+b5ESHEDtWDYAC%Px*}=0<LA7z>Vj`*Yi4>ev1KblGl%OVH^1`T%B+lt2`klfnp6g
zqM?}CrszS(D+2K~${Qu;t7N2(wIj?DD5y3jlA`HvlyCT3B5>9JAiX#_ng4&%OWo7K
zoP<fq#9Gb87LG}ggysLv>pD5RknnJ^{(t2`tjsLz|5Ir;s;d`|&V}IzsK3E4h`QY#
zuR}p=_^pSj4pAxFeS^XYUQ(K=fu_4*xO45%)Xm0+&MF`Dqn2v?sq>X@{Fy@po>HE9
z$f__Hne2{fN>Ya^r{TxyFJoJ?pb+Q4NQxfkgw`Rki2x29HW4ut(gtQLk-UrsGA}>~
z8VfF!vTEKVDo%@K=%H{aeMU&zh{ZZSv>cW2KsCrC1nsr9t=Zoiu6?n*Xw(f_+OmNp
z5;(X!I6r(3ixwyS(G*ja4paFl3|`aZ(8!Q0?rB-H7ax;r)e!#X2C}9Lf)WH&eDwil
zT9YXd=F}BRWyW?dV^DZC3i^Po3=~jma5;zZjVwg4zq(~L4TH|PP!|y2A|j6LI<RJ@
z6RwKfsgy3lAnUY?Sn3hz$}-&5nSfs-^3C47L3eTI1Hi~fTqb>}e1ydSg&@%s@?d!=
zT{LnD<}1}P?64lPHqe8gmJ9*Y__KX$Bt4Hckg&lRdnb-{BQwXL&taH+Fc8yGh1+UZ
zVZp*p7p#ZDns`+FKn<hjiPBvJI>}=Qz#jFXxlzAJAk%m#O5woSAT~rmw5t;c)W7L$
zG>VL(-y{J!WV=e7Qj38A+9wdZ0k%J?Ko;*PajH3sU~x%TJF_3pSOh697UZZJV_=hb
zsani|IdMN&3`a=j(_%@x$q`uc(HkX^d!zm|f=i#BI3Wb+i*siLS=g^hhFIh$!=0iA
z7ftqhqkEFgF;l>pzI7%r`F|pMhEOm|gj61lBNP8ry4u;9i~v<3*<$EhvhGBEzJm3H
zGn?O=@vp)ql}M+g$Sj@a43;4+C}S)fhq-Q(aE_j(tVjoFuDnJ&k<W$rxH<heZ24$G
z)k-V>)=&AB(Mv~thuTAvp3R}er{hPKpMHLJaMq*0nygmmtviR^9KJ&BI5t<XGf{nm
z7iK{x&1ULW8VbsKj`z>CE*GSQZ#$?M5u^*J<rIXuZ@RI3dM|)+3D~`5J`};(IDfMI
zQ<6lTgxWoDZ6=&EBxkwppHgf{JRQ=6wj%glf(*b2Rt!B)8d9x%)QALf^<txVS_iV#
zyXYxSJGqc(7edp!gJ}EFXy<2ZZJ@Li*nloVHNj3?;Z_LzmIBAtKz6tV*<B16TsIlY
zU~Ah=H4*2%8akjs*K#rVU4$F2&=O9)vc|HiD0D#))NW|cJ(fo+gp_-id`8A)HWklh
z;e-!gDrvUTfXd|)g&%8jwoXSOKs0I=m+L=V-hnUUW;*T|601=})bIgn6beeaYO>Kd
z{j!*XA7^EX5g!_xauvuB8x|X%^u091LlkuNH!!~Sfb`~_t$Y^aW!`{-c>H4f-{@T~
zNjqoAUrgk)X8d0p!n%(Gg#8G1uST6U1(!8_wnqMT8M6rJulont3XDQJ^xqj1;VafZ
z>zbS%b^Vz96XNGw7#oCuCy|jo);J_aaDr&rhj784QSkq+{|12xR8P`F_mvhjnB&Xq
z9nBS}KeEh!jrDaM6~&SY2M;xj2@ORsHHqC%T2@;N-vJhOG4Y+(WBQxMEYu3Qq-EPI
zioN7gitu@+GlON8;wK&;yr(xj9w$^4o(8N$0audQOrw(EMi;kC5?Lr0edN|uD<0e?
z$OL;J+$S*aeZ|BAlWt7hFL-V)T(w_QD)z*MaefD4X<i_=&>*Zv@S7GJ-Jzd%@AaH^
z#f^?#(Ak0S>P~f$h!+5%ymUIp_tX7xt5;S1nVo%L9b?Xqm~QayYNxi!fq=8={UY7G
z8P3uU-@96sF`Lo^+>hPcSl0HGF37&}6(8bh7mn@@8lbTGlW{DOV78#2P7c^PSMyPl
zK~|q(4dj>g>eZh**zgPt*`BW6$J>=$KHDB{(b#PyE$b26MqPWN-uzneqDUXsO{|1N
z^w(*VKF|I-Vsh4~r8D5seot1qe!E24N?ywO{mOpOElAR))ydv(gX5Zcj5e{STeP9*
zqwY8JqQGZnWP{%x6QOj_O!ed0NKnnwgUhb-ZUhf!(}c?k5ZPCqM9&{c&poGI*j-oa
z;J3pLzrCadh{g<v-s-|D&y71CZen&O;Ft)|miEr0D0(>b#ni?9oTm*ne(fdGBK&LJ
zPjH}G)sC;%0OgSJl!m+r(<f8@>3!lPC^7~)%{tOWe_A8g+&fa6F#?D1!uGcmGYb~I
zFdDX=fsg$xJGG?`!kHOv!>->IF)_$n`xm^QE1N#?Lh2UQ{21U@;1L^fKO)eRQx4w`
zGemq}t4CLfLCdDDKd)4v|8Ql)rtf0~F1Abqv*~W*$U>`Yd&4c=Ll@bhxh$eMr1iS1
z^?1(JVZe6!cB<B9u50@i74CK<@{~7}=^mO6uz$IUmdpk?uw<O_pVpwPHm5C01W33z
zMcG;DH~|mAb|}76<a8$C6>U5F8g(y}YmOv3b=B{kYrraUkdrsM*kX7Ra|TQQ7h~_#
zE!fgTZEoAPUE8*8+qP}nwr$&HZQHi(uI~OW`r?ey@A?C4WL{)M&W!m?0tQ*<WWR~N
z&8jLT;M6@d<UZcRam>)`dCjStnHedsb)WaetljtUDM{Zqrm2phm7`pADy$Dd7bBU!
z>{ng|7mT3N#O!5%+}<SKnbMBAc2IfUs4(IzW=)pn!m4$p!?Z;eJ{c$FDXf`+F=$TC
znS<xoh~OA&R1D7de7r8bS7I0D<*az>gqfdG%IOS0bt5{Z1AYri=|`{4nAItZ?tw8_
zQZ4OXm@I>q@in|g-tZ}3TM7T8pJ3zoPbr6m^?&UbQZ%HT4%y*)U+eA^DkRrPNF@)p
z7$wYH+bq3WJ9)}7cqou;nlw|J6SJjHR_Vn3J=twvZ%9ku0QmO~!Q=O{zC-PN-|rq@
zhQ`PTo<*5*CY_tMZ4R7^5|i(RZCaO4Ud7GH)@O1gJ(|BCHc!Jov#29l^YH9$iVx<$
z%`CaSKfaz<=?Xp(%{usA*?xC^4iA-(=v+u59Z~`zbwi?4vYrBxR8ITw^P1JGd!vy5
zx|g!LF&cHqDvZDA%fB|%NXN{acGC3e%+X<_5oFi}@6yI;z#VQ9hrfF{IiWvNW96`~
z5;)YJk8BHSO?q@++HAtF(PmDvER?tIO54EOXwCEc5jT$S>6`rQX|sloy4v`aJ}kME
z>PZX>tc`~<<y@}62t$_^i*Ei?nrf1$V)^y(G77A+2D(TiayONfP`@{05F9`NV??lc
zHgdAsnAm$T9q!pSm4R6|ackkaoPA8@<Hdt;h0G+ooIV|Tz{zFALY==1u@=!?Vj}In
z$@+h%07x+Pr75<C!yQnXL%u6$JwMeLHd+Vr10BkcQX<aYV*+h2D}9G}H~ppJ>JLIl
zW=0?Ny8)OkW91iO!LP7I)mfrjpI$tr&{az*7xi1)7RE-Vtm3vUo=)b;K;7AdsdTC3
z98)7PjsL1NtD(Y{T4O)xsXY<5&b?D%*=2XE<<8|u+iOcyMmF@<_PBkSXYu%|)i>?%
z<^_@`jl;E<o^@`7-o;#xAdFBZ-ITx2tUC9mQeb;`bL0#Oi>cVT5zqV)4H}c@Pe(tE
zM}bIT%E__t=<v2DuL2ns0)(!Xfp!pByzmFA2M^o?x4SF1qUryRlwpj^P&K6ac?M^d
zz-^-D(Q>BH4bSXv+etp<S8089Z8fLKxPjQm@nU>2e|S*U9U}$h@Pr{@R{+$*`vRC@
z`6bnEP)26X<HiQG!{3Z(+N)wTGPnffJ2NJ#P}wP|#eRpz!jmJ#)mYv4?@kA#AU)uK
zUCi!pnrmKqcA>`VcLRE2!}!;@V|e45VP<}_Ek3n!@5RD;6qQ3J`Pq+<ra^u|wMw$^
zwm<+Ts4jpe2`K_LG-#9c9S6?bx&=D<cet7PS?cEz+A~q6o0@;lUB?|glLP~Xqvn$n
zE?L_>;Be=N>4N5J)QE#K%ZGtl-o=OJ;(}fAkrO!1hG22)rIz7^(MX(ywmr<PT<^@|
zxFTAxSJ>!eEmQ!nlJ;NA0Y?8BACAzMpiXnAmSkiE$Ji(uGAQja7E`ECbJ&P>lg9f3
zZL+6Z#2ojihZwb2@>(mK0`b+ewXg$MG>|y|#}Khj&4f%Sni+DhtiE@bEgJk4Y#f<g
zz@<JNS<nX`M1OeTSC^Z8sKWLR>0ath6y_GDT}D^2!Iq1y_>0Q{N5Y&8l1$~?S+877
zeVPm(_F|sR7Z(35S{zLY>G;&|{1KoBH7;ToR6?3AN|pzP=Lmj0+sLb8I+3n~fg0+-
zA8|-14})W?q&*BGr;cQ9Y%L&?-9Nz7Oq<hD>IU^CYvi(drNfaj0kMDti#Ltk3dJ1G
zdqD`1{T%>o>?SgmE7wY<u~+4qC|N~#W0s-Q5Y2UxDh<MVJ~y(rhYad@-FCx_u-tBJ
zBejV-<IZX(3L$d792<r}83~LgOG?8$R(l=AIqhoSIqz6{RnlYv#$qL_d~eRVNC5v2
z>~%-_4g>CWW$6T>3c(E&gonag&wO5O7);p&*~a{u@}TBYVCxLE+v$&vfNz-}GEa$8
zh1mdBpg_Sr#iuhl5_Q}GVwr4(cqrU=yYAAUEDnkP{<=gwprb&OY={(jv;2W-qN7&#
z&EZkDeN2WYrdM-{X$Cn8gcZ7_<tYzoOu%<YO_=}wllB1Uzeg=6$eyk|?1kY=5h^=Y
zR0Sz$5^Qpd0Oms$5{|ay#LSCq)YP~LUhUix($q?1o{oN5(*>(63_Qg42qi2I6p%%`
z2V6;(Lo!v}m?zGMD2~IJpb$g#3dSdP(fV0B(gp6@)%>d^;{xcNM8o(|7DByhq-B16
z31nx7WWR65h4hSQuNM%rHCA<xT7Oxq3(&pHl{Xb!;*c-M`Dg-}vc|sZ9?%3%v?sDu
z#zc1@5<%c!>JGIGx&3<6V<R7!Q4l=2QwA*O^(Lh@3NH&_CTa8WZEERvUlN@K77`k}
z(vdN~LAW>24rP>Y7!UR@Nkc<bq_&S6Wg8-P#;`b=dJY;UC$_BPADCxze7kwk+BI(u
z5+}o+kmPS*7Na^wuOj}o6=-*8Ew|Q`XQy_UOebjapnZvYjl&-QcQF1PeIe0v8(*<1
z4g~8=8$>v|+^p-<-QC-h?IvTa2GHJb?hU4&#uA27h&sXB;<@2j=Z?sthA)^x0(2^*
zL%>X_;r_P?pa6v;y$UTo#Bua{Qx)_jJ_`xT+lbSrp>5sHyJ5dUsoG%YrAcGBQ|ggr
z%W&e)yTWne-;ii+hGU?{cm+(iGk{F@^WR=T#X(%&l&X>5&KMpcLohI-a!nw<<`{Ci
zU9)<?B`KpWtDguad7RVVIKTXA<3R!QhTyHHF7;99b~s2`6%d}Fy>D8rWS|>SwV?*v
zuY24%_wt(Qo$X8T?0o}>@4*rl3d}Da#Zaz1Y|O|ajY&GNG?IV#ep|yxa;raizI%b{
zjJ1u=nQmulwhNwrAEadm^UvZBv@vjp<=f~ipzkuy?*xzjsQ%))T=@$nkPUV%BRRU$
zy`nFBjdNW0^1v?nL^VN*Jc8;WM3!FYg0%kDGKq{fwuH!~U8Z`7?frr@u366eZ|roS
zL`IX$tU8hB)fNMZKMLz4d+r|#jVF_4=n6bt9aCz*s3ExU3~e@GZ#sOqW;grNA)-fF
z4LFfKSih^bxoc8J<uhtmmlibrW1`b?X!gnmd;A4Ru&R;c^ZZ(%RJ?bb|E=GjQ0_e7
zMxSkMJ=|0|E030_UG4li&IK=|>-)&d^V7G<4g#rJK{^Q=4Y_PF^bieM9PmU+2@oVH
z-t;{o>%8NC#l+ZhN!;o74Kylo%u8)J2%I#wUnOjB8Yl%_{Rub0lH}B0$J@1VQ9fPX
z5g0+rMMG;7gKR_Aue`H{{1>rYnR9z(_IJ0%Wqa`1^7iqw#Lvfj!}pC&u0O)+KfAF1
zFm*99Gyc!hA=g+kZkwa`|Hc%FpQvnKyc&G22_tG$Jz{f4pQLu3y#2-J1c8C503+v1
z`g)r10niA96;lnTV&&AY;7m<Vt<2DZ$ZyHrAL5d3P04q&oml6^rKrf8j)Ny6_K*x{
z$(lxvY=kqoeb_$tvz!8507OB=q!sxM3wnP~+xTX3f4w<NfW(B07AxY}KaLOHr%WH#
zN8aDFR?+KpJ6G}bc)WfF33vx`QS<qQ`3)rL#T=x`luF|Xh7md4rU9ntE5bvL#3e!R
zD9{>$h~OfTI0-Qt*1T=@*sllyK2f2s*v6FkGl?tRn_BcJ(P25!3W-71EC+rL5uac7
zj?M@ZQ=j7#_Z)!jSwjc!h_J)2A)rDq6vm3$F;QdqlUuMy1N{>XasQ##aUu{$X|Ck>
z`aWz?2vfve1|dOs_M(-3?BBzM9y-7yyN-oCfi_4<60TkpsYfYf2L8s!a=?HX5{wS9
z*08(DyVuFI>Fe2iP-NX00T-F#c6L1>;$s>eh{ZA-C_FQeD!NR=S0LiD)8M&izWGOb
zRoU@Zmi2W}j1*krU8c)1Ir79V-Qu2!7SqNiboQ26g@Ajw<y#?*@`e&h3rqr>@)a(W
zq4T9iG0#T6(*7x5%`(hgIv?zzP2_YavQgFZh_70rGo4aG5}T4qNeRi)#56X_6ide-
z9zZsHz^27W6qGalqraC?G;m0vf$d%4l|k;28#{W3kRhQ!Qr;2PusQ523sJ109rZJ|
zS=DTh(=RF%+BnG*wZN^Cj5_7mBWc;@G4vyocsyZpvv06Gi!R}-)iJBm43bSnr54II
z1PePn2x+Lznl+xd<x7mU?}WEa@=o^L)RDCM>3qEtQ^uajU8ySSZFL)0udYWLf=vo%
znA6i&>Bcv`XtM2d{E|04LssU}D!Kh+(m{z9Z^$@Z6lG{lu1q{#z9|tt%(Q4I>DMQ_
zdZ)UyyiPm8%WOA2Z(X<u15hn@Sl*<S4mkDn$4Ial%UnE~UQQYyvIon*T_9gX+4h1;
z3nNq0lWq>SZ`nmd+1DZqHOlc)m5bMa@Y^TX4lPwoU2}t!KA5+6RFl73JN(Q<qwD=?
zbc58NEcVpjz#5L{1|v7f*77vwxbc3i@qt&Xwz2d1MwRp5SoUMvQ>Q4gph1H$h+#&_
zJyDGq?qZOTi68;@Gau&bNNAujgw^t|aRQnNtT)re$WVFx%Xz6FjprsazzQqwFfNDw
z5ZN19c+5*g@A@n`DwbrS5ytZm%inYU0BMHeW_W&I-~u2Y7`*h%n8}qT<``EOviPMF
z!ubB3bTfp_{uvCVT)9HnM-|oGPIRF;?oT|Ef$GbQgcvJRdkw5G#v)!LTUgeFYNkW3
z6WlH_MR(`nMno~2a9qh?CNDDZhi68K%3gZWI1Ld21RERanB6^dj#5o_$gU&UTRG&t
zJJaiqFJBPsWE7oR_mj75@*nH&Cr+laQL=wIM$OcZ-CL@DxHMWQ?I<`-N>IztFA>+z
z;;GaK1B1pB+MkZb6y%(1iZBuU0KRUWsK!%mhk|07W8jR(1hWr_0fwkV5ea9Cn|(2z
zbbu_)ia)RImGs78ol@}>eZ)nJ%`!=2+<c<No6qYy7Uk|ppihUWM)iiMCYHvvFW6aM
z!K0<UbBrF=eMkyO$r@`y4B;t+3^Q8!QE+jAOj2IqYg?A`a{+yy>aN;hBdkfr!wN#e
z(wB&Q*Y@JWoe17sqcB38)B-}}u=DcLV7wb*z`^wwA;(FlO-aAG%wrKt`ZJD#EJ6&2
zN$(d&*Ie8h{TuMe2=v3%H6hv>2oW=_4-=P3ejQU9NV-@6wBlA3p@LK^pH=Ju^yBtW
zEOkg?+EzFYGbp<bGtBht3poOIH<fM;$@oJayb|g|oF7x>?Q9m`16*W&aW%1BW2#{(
z199@9AJWP925HU4uNyRXk+kX8!Q3M*vjISAN?mMBhyHRy<ROMb_8Ys_bhqyd)q4@|
z#qv)P6o*Qr=wqJ7arosXgm(T6%J%k-pH#*xy5OuK#_AoNj@vq{PqY@EX;H5=33KK2
z7>1MDj-6R$W73-cLet}i_n{<kYU7Xc1+mgo&j5bLJ5MCOjSqk%_|b(FhIsXW-!+%7
zsx04<$}NX;9YLt9f@OFHDa0N0I0S?JBTMvRp6u$dlo8S+FtTHThX%7_zT7@CMZez{
z%HeJ2-w{m0>nxW3k~+Z1@{ELM>G{wokyRl<5R^$12rMzIt31Nr8WpPn4bmsYatZW>
zA3hGajXlh5H@r#T#EuAga(|p*0GB>9NUSLR$lZJBwSPHiq;iu@5k=$Y{d*bTVcSVE
z7qB$LjtD0~TkVyGu1;Nxr09OTwg@tWxMol`1cDPI{jFx}%q!0KQr~!-s~w>G@cc&H
zur=sX2hjn)BQ3x>I^xRct=aIU+64mjh&&G|cire!POhnaTzUfTP|d|a$}@qQW3%w}
zl7wjJTb>caJI#L<yLK)CjK7p@s)~m~HQ9cW4Cb-ur;Tj5J!_Z7zVqH8lf2t`LZ~oh
zTCmT@0ng2K{`|A*<s}C(4Y^asiLDX<Vr1L$^ujNN4LSfl`r$r1I=gg&Z2A)g3vzdZ
zfNSa-49hF&9i^q}Ws#D#1#PQjlgQoYLYJoN9Mx#vtlNHi$w_&0Va!nD%e^8)9%ZAR
z$gEis!9pU@EvsM4K+k5w3AtGdm6TylsX+lqM5sm#4AoyX%>I$*W!=<Vf9;GxX#A!B
z@-niX<p!P^VrW|A1%F#8&2t)Ja}5O6cKpG{cI*bAG*Oc^agkt3En{PAagWJCPhDi`
z<fEh7ssp}rvG5c(!uT=L%EU<RJqDi|FsqK3Q88+GEfqBHroR!#1aFRFtm#J4ASB1>
zLd)GJ`CTSu_6^*c5(*ROuEl@~9@FdRdIMIo%jRo*LI70_ZN&yMhUgfhg6_NgTA;Ys
z^gyhJ4l2cp!Qan8s<2rysm;yqWYh-D<har~)e^io>?}~N{UEI4q`KHZp1d72TDvPI
znwGTzt1XuEU>O==yeYz&?lujit<xcZeZKWkE37S+h1(o3fy^+Cu~I-4!L<?sxoe}h
zcj0+xpJ7Mub`$B-MAEI*7V=tiKCzH0bG0F2&_EGoF-}mu=Hpa^2E50EbaFaTD#;V0
zqIzu6X#YI|mbL9qs5(8>f;`r}V_{w_uN^W|v`ki~qtUr|1Vg#|3#g81zWg8U96SAg
zh-wV%98CYKsJ4u0<+RCq<>ec~JE=9Dm|QOAwXL(dil^f`wI!!RSGJa1Q$=Hvh*Nj{
zvI~wBPcvDYQ`)nAEhGpA1c)F9w7;`;xHHvsOKt|Q3z6$dDMgIwk+cO(Bo~5sSnxo`
zbQ?pbNVaZXO>`qitEb29!x$;#_wcoDPL2ASr#*La{us6E*K^fBi!k-&$>_=Tem)RI
zfA$E0dPoI?NDYlS40ax!q@1g-Js9gX2#pe@xKEitk65~64EmO&F!^oC9BX`qUX@j%
zd)~<6C+gcZ2Vr-qa8Ng8v2uYasJbjoH!F6iF6pO)PD7}AnJVin_UOWa^wUgJi3B)P
zxG~4tRcqUVxRQXVJc=r%-|~>*WKkJ!m?`?xm>HL4(Yge&i*gOrLL_`anx9HRZ;2=@
zCmdLkPak%sczV6;(n34q!v2bX_;BIx=~XyEZz_=7S<IO45Zn_jghm?SRZ4pjCG@3-
ze~M2r5-8l!{8s`NS|?8U>e6ViKZ71NJrQZv7E0&h!;&NCSnaa0`eGSh+u;py3>48|
z5DB&OA8Kd{%v&q|d0;BKhVd?Tr1DLld(7MG;|kQb>H+YfdR(mj=n{;3^dl-jQ%;Ed
ze47YS`XQKl&M;eD&{PSVv9qIMQZ#O4?Si9fmv6XT--0^jA=HZJmG}Al;h0B`ColWk
zqZO4)U)%pi@h=F5LnkNG9f6{Mo;xFhFOpZw%R|%d%^%VcZhHs-TBWxVAnqf6T1qVm
z#FwpM3bbKJ;D`$2st=m|0wd7G3HEEnSuw>si<2Goe@v7mb#XL*gct7`rnH>b(NIwL
z`DN*}<w;01XqtyF$zK<vt$*OoCUw|xtYOohjiD&*=AnUl^h&?O#GJzo#|>-qo|2Yz
z+fF(eE27J-hf}>2H<gPj6@}WYs#vv(Q?<5n$zcX_9)eIn#;;^&F32$=iGJe&wWVYF
z%b&2y?mw5vnaR@S0iv-M3t!xY`Shrnpk&$|{rUC;X=LrjYw%+Y{!dsV3oTng;730r
zTs_oiDlATdC$#fy(VqcekCnVwPbi%zPoiBTIlzA7n!^5#B`V25UG_Kgb-~Fe{Istj
zOdvzCxfsUCzxyFHwyFX6w=_@e!OD0rjW8&6Mv_iPjRUZt7c%#$iW^2qCr=@i<TpAt
zzlN`O?2o<nb;})EGS4W#Sjbhi&AvA1)=i$ZT1WTSheeSpfKiLZi`MAhY)v?FS}JW?
zB7PQ=9>E<9P#KC5*y+Dpd=S<^CD>;sMR8$z!->vyCp$xH!e_K6C%{g2Dp6f33ajzV
z@<GM>w2QW#r$@gVe6g#b2sAOv+ZR9{@meWExzK>+zu}nftEZb8<qoU;olUd!o4bDq
z@jP9(H$@Mjxu0gneZFxcd*+t^rg3M+J?S7O`vMLi84leKzI8??^DWw_rk9F|7F?Ub
zAbe<THKY2{#~=zag79<U-fma`52Uq@Fnm)l2?#gi!UEVeVukbf{KgxQ+xWN61jKex
zKvqr4G5ZOYNGq%=U#@#r&KTYB*Od{GCDqKXT?*=%XxUh|v;qSrkNQ(-S2@1-ZticO
zF!{)8$<S=TcUDJHDl~NrTKnpnTFFdH>r2uSBQs4u&(#bl&L?4ZilEsaVw`MK$v^2u
z8%8<CGM9B)S85`+X{?xfp<jBA>o9fp?6=}#Q9yfBo}XhZ{6l@oeZgpqkI(jNp4W1z
zh0Tr;b|WFBfkD}5ZLB-o1Ol@|ZXafa&?*F9bk0B5uxJFuyBc_zRaCq8`y017VR9=S
zde)?^)R1yA@LM|_5M_w57QF6&1_3Hl3&;b`*a`!eZ(4V>N~ZBpD)Z}n$*345r_aLH
zE<1PD-aBzD<I;~AU5md#6BXFaP)}V*Ls>8y#PdV%;Liq=JqOnc>zFYyD?`lOyuiNZ
zo;|e7u^9;39-edw34@dyrN!#uC;w%){|)dl!~v#Zk9yxjVm%n7?YWhh#|+LI#_JN3
ziz%2!&3EQ7qp&|(q0_hO@jv2c2-Qdtf(xULHo~DNN$&Ew)YP|Cjj)E<Yqk&p*~U^9
z44W{^K%3N!vz;_fzKSLq3i3>$W;_OM(8WbTFeG?~xr}6P5JEe}%U|X%Evx=Rx@5v2
zKTttf7K~K@X4YyshQOA37L5(k6_AzNI#Lt*1)nj_=46ZaS(h(2g8VLkFTTHx1Z8+Q
zodJxufn$@-e<O^^zSL8lZS`fn9T?VS$ELD~vMTjqtt1un8K|0_$V!upzbDvSH?!3#
zM)5TDe6f*CB!qH#!Efu{H^Nwo*^;CIJVDu%sbgE!X8=^Pkvs?&z?w`2Ekh2ixqF_q
zQpr}ZPWvzR!(NhTfo=$m<9W)GotlN1=P|l@1f+ew-C3z7O&PMT;tVqvCXA}p|LUZM
z;fQ73@0(&dq}%aNa)|RUUOd$UTQJKXEceRkRgtFr$1Es@()h+EQMTzb%6mU*+z-TI
zh%}c)B&L3Dd;)!&k=T*}I~Cu%nLBUC7dp%#+TH&u+YJBT*_!`@)?r{_WcXjpc8i+S
zzp{<cd#QFO-ohG9j(`3(N9KAVd)XYz$wPt(OC!KKhLjPX%#*?VeXXM}5E@T9m6#J%
zuSxxKe+^6kTMLBMjrYZfg_9$RSWFV*juLGe5Dr92ELZ|1y11t^F2eXl|8aemJl{+{
znoL4;kUuQh^YwhnCq|Aw^e+f&T#=Dm84>oEgNws1qwcS)+^?_K%L&du0@lg$#I5MP
zKc|}NS3Zdh@>)J%6K_K>eFj(&sJc~c%lzPA1nos5RwK!h(h;*t?u=D2)7AhVn}~-R
z4n}5J+7zKU(e}CEzV*V9$-eV3bdRix0xM4rLX|!@w-tN=FW!o$BAODiVLwDkAv&}Q
zNx<?b<AkM;e>w_h*dpqWgJ=0;)sH$Fizo2EdEvzK$~m^vbGB0p^}njhwS%Hw_4B^f
zOTiH885A-E@yWWf##G@4IEU<X*iOP&!c0MN)(!EvXsH%JqBqQWuo_r_Zq!REF`G<Q
zJ=*!=CnzdIB-&_6)eUe%qlMWQ9JVJSR%%?R{zIZ_k;ex}^t#A0gAiqqqDF*=4q?iX
z;(%pBkzxkA#C5h0?>HJd5_9a{kd#wptOW#5xr(~Wu#p8^qX30~)m%xQ;NE(J3>x92
z{?jXsEMY7wQ6!S$iIUO6yOwH*O^qALKNc&-#WX5AmWm{j8U(1Mh{<)!tXQS;a`qIF
zT_^w=bmXg{lGS8K>&^WitrOO}BZDI#-5t3s6nX270IDf^I&85xjqPgF{R|_~OC}xE
zrOyG_64bCr=@<$1#1+q4j^ZS*xNK-+FZ6iNrZRly{&FP2<hD!3*8$J1${=Khzu_@Q
z46PVSB(lYRSouhmBl$n&fFW8uzudw1n>Q}7RKZYn142onLTPQKYcMALY*iv#2<hmA
z#_k++<ZP<Q`Fe*pIjqlCu3ECLbwd4WaLfIu5=J4tJArN{&j`E;f%E8}W%XZAoc5kC
z7)8QR`}j7Wm_SvH6EV1$3(m1wS&I9^@oS~y6FF;=>qCFND$f{f8S^n69oi+?PCgB(
zibbcDo1(SPft-LUY%jd7fm4P-B8XP7DN7+Y0D+t@6{BeZ$3|CMpb%<R{5370iGytb
zQB<)l`2|~THBN*^aq;YdqT5zg_@E}k<0IUj7i#s`ba};FVxLdp!*zH2HsgbaxMBO|
zfM1q%Yp8La$qp~Rng688R%Hp8_Ga{d7le=eNlcdC2RpO@Q`i3M@@=RCC?!I|?Y{MR
z;%(iJwMi(rx?}@V9IkcVvM3>F@#ylg_Vck8$(zl6i4%>AXzd2BK)6|X2ikj-uh5}5
zaG|LKn$=3Hd+Tvw%>2fDw(tkSggy%isfr?<8Z4BCh>0C(4TU6yyWE)``>Ulc=_WxE
zGAImZUmjB!lKaSkb}K8BX?+N-Q+znk9+g<v-Q}ZBj}IH@roO$Zj}C74dQ^=D-%%9B
zoYxo!iA;VQChSG@@Ftqcc2uvo4VW4zK~_!IJ(N;9v;Ui)MJojF^q|+NSMcPVsl+`t
ziM#NLvR;CMGJ$=q*va>Wz;cvf9lWOJZ08t%F7MFjbnc}*$n6D02liGYFlc9+hj|}W
z7*y#K(xHLIq2}@4$?TR-`^Hh{Y|=dmm!Pa5w%zHnM(Z-x-0fm?=a{d6#oc<C>iODd
zyYECHD@bZ~-|?}^w@Wf?F|~V*%I9zQT#bhVS&v7fl<A~k9J)tC^_UGAln;xqW3K(5
zjrJzW4tqU$qUq`~ILx@9iIndn#q^3q;jNg5BH|~f+}$$|U!7E?uU#_WJXTih_BDIg
z+OTD$B42hHRh@G%*e~}l{TT{@1LmE#M!=r6tK1*_4-ec~^2K#`4HT0WdV`nVn(C##
zc5dG1mx@&TM8T!p^{^>}S?Giz)#)#4UevywsY?fd4~Q2&%sc=Cs61?jUBtm{uDy8Z
zHzEYM0pnR-{n-uh1I$l<pHtqoYsZbcNfBwHls8JxTq3jGo^iLH?)_~#jYi*IzVn8@
z#zD5RUzWsFxc@96|9SKK=a>0^zW)I@pswY#*@@`;TDRwsRE7u?4?uT*L`a^-<b>y~
zsg}cAj0sOFfOaBZOr@yGX!Gmk)*zKxj7sU6Xi=7g?gR)JHPW`LtB<(_#`t!8aJoM7
z=q70@K-S0}9kl-MRP8@g7q<yqXLRRN$bc9*UK(~}?E1M}zEn$#$erHl?Z^%lmDSa0
zxcj(X*7NPME!&-=`g**07`j?Ly>(vV6T&8xfgyV%9;6r&K~JoKDjsKyH_i_nhaev6
zFIT3>0d0b<{@vV2b(dw=IFXT6?CnxMA70QmnCdKyApPB%Yom&is~L{6YHOYq-&*T-
zKBj@qbm^+lcvMu(pu<vY0F2peHd8g(zH(G!bw}?O+|UI?ND(c*dp>D`m6W>@Q)6S9
z_(Q)2ZHxi+=VRfX0*4CeWVu)Sgv-o|nLBvOTAN8zyM4j6l5W|U?V|@M;uHZeQJX`v
zr94&j4Vn_r_RJgU59%2;5*f_7Vx7KEYM{wXCW&$<38KD~XZ--o5ZJ;%%G5=_V+)=8
z6pjv3MZ7@(<|o6hHFTzG60icm3FZtnwx^l3BhHy}gnX$nEtdb;utgjPyE^q%>wV}@
zI3<8$i~7}kJUifGX3%rlP&W#p9d+8Zw4|U-0MZneA-SYr9Q?)oUr-tv?t2KlR;}l&
zSfvEhj~&~!l11KK{cPtw3ov+4Ab8(yaQHGz?)Znod{AQa0qLmtRW*0?^B31Urrp@9
zYfLLc1B^pm@kzL~SZOLe(e_wH#L^*%RR`>Sg8dLeZSy@N;OVua<FcF65}`*ctY00h
zj3#S7TWJ1CU^t<%&WK}z>Fb(53=alLo}*D1YGa%imPTb`=BQW+q^9Ck%sJMTHy;(R
zD>*bd&pUw*O7m_0<JFZku$5f+2^N>=+HFoA16a81TPN;e%!B~?j(SRi4`x?d&u;NJ
z4<7RyiK}BUY6;;+BB69PJ!>5tL?|VeiL(P}0D@KRJb-09`V`fbKokRy`^D9n+3AOD
zY1JW#W6TPO3;1R?SA_zy2W0-mfMxlK#G?|VMB89hd}!iFlKM}jJWyo*6Un3Y0NfGI
zqVe}1nY}}v{!FO1fKv=bB=;c8`Y`exid4{w<!TPW&y}R^5SVmQPjj4#aFO^c9X~mR
zAnf>XZ4dJ$EmEy09s#=@HdW$i+eJtv8S#i^fh>Dlscfcj;<Y%-)t^Wlbd@8w>9PIz
zO)aM_5u9mnitRMnbz)<v{IwMu_>eNXjNq2N1S7-R+Je(2%rf{XFM|H1IJ=X$0S$Um
zgc`*L0Ob+Ee`W^+COGz)jX=yVZ_RmLc6{6^<?_k~|IIVDfAiKE1mLt{sb_O&Q7xt_
zFi~DbUOtwG^lM5{;;Ozk-u&^5BL4Bl8v}7qg%#icC#F^r<$)4=Qmzh3y%$1Rws`mh
zNlLvk^j$?ev8|>&?QSKTffE}qr&+&H4GPUhsNVZKv2o!=;*c8%n<dY|IGGb+2n<Tk
zG{j>TAf@k6iI*?7Y>XwaKa^KSg_5wagyc3JS_V5QNRg?pdWK0~SjKAPX(>~Y<9qZ0
zUhLd<89GA9=YZOVv*l_Cl;HUgs|gK@*_;h0YjNqda43@;L9)Lgo9n?!4@gwpne%f#
z!-XY31A19C;|ebWOn2g13SCD0%0#}@_tX8MfR6lLewrz}6__nO7Bn~boZq-N{r%3a
z?(}&$OAg#l`@U`65AFHGZvz<vIl8-TbJvkq>*sM$+tw82r|rxZ${qd9&D}uIM$3xM
zjytlpS}c$rOtdIb-!yP>Fe}HBsOPFkL19J|yqZIgXoEGa{g{Qqs7IGLM=BY?E<?%5
zI;V~jM1IKrLKGfFVuhdnYwSIX&&!LcSWlmy5y(X3A|9+1_K3%egRSst42(i@QNDOQ
zIWT~Pq~7mG5ab>blyMn03X58~KlJ&1G=lolpSGwz3CHSP5e0!u^&UR}qfrB1Z$W*1
zPG833Cu_;!Eq~=n&NrF06NsH>Bf9Ft?dxbBq?Pl86Ik>aySeTuB`Uy3wtp-Ydd&SY
z;MI2sIKB&FNX!d<=i4>nIwu7AGL$sF5dLd-ci!i3$s8W0dwZuWf9xTjhj7Cq&#{)}
z?Pisc?&t3DkV}RXVb}0A3f%p2byKQNt;Y7*7#s$@gCm8mv|06quJnuPgFP`A(Dx4&
zO91_+-eZkq>%Op4M~4l`;EmV?%>ZP3yS(A;b&)9nCE~o>JZghCQ}n8n$@#wHP-{`d
z2feet0ywxYlYa9Uitf9j>l)oYxZ5uje$Edo>1z-82BHrde=h+)>q?6}-T>@eZA`2r
z9tDfzc?|`KE@}y<0OsF_;-Q^H>06Vw=bZk?9hBGHHcvrlV_}gBitfExWjTI<6sv7p
zmHV~hsp{2r_p^LeBN#37PoF9KhH9;~??d;eLfr4_rGv`GKgZbX^US`m2<`RtOq%H-
z{_^{87biT%p;<a^E)No3IhC(8<IhIf>)EHP>pi~71XI<2R_p(OX)!YWy9WM`)%sXN
z!fC4=_5Zo>EteY@5)Vke(>K&aE5+tSA7=)cNsibSk`7xp$~2a+Pkg;{?*VFSCLftl
zDOG3-GpDxg?|Vfho>55_^5uNp9~@0|>?z1TB$v#OupU^Dh)5z47DX|7;xi>=|2W>C
zlAkh9kZ+cw1phM?x9iOMzWvmp)A@4iFk}oflb8L;AiI~N)9Z#yn$^qA>+|}2n6dhL
zPh*?a>uwB|L!y%ue2A97#P!>(I_oXL?|WH_ib{$jM_D67jA18-EJ@g-Y?#X0j>tK!
z2>zy0#8kkH(!{lq`*~>*l&z7mz@w};nykT95>ZbTON1gq5i$CSbR|p%Mob;TpGX+8
z27}<^FjJN;pAR->girg2^~cY6b4Z;3RLLZ1SpO@F7e&)=>aG9eM3fX$+f}-J?DaTR
zQFQlIp+*eobbv2Ecb`2I&AlE?x=<p1rKe|;n1b#ig1KLR0CE%cBKa!*(NzX-NyRbY
zh|1i<(z$YN(z2esIVGgScKw%XH?!*U$8j6y0x9ALn{Ncu_j4FNQ&aCo+UJbcZkBp8
zoJ!Pj&VWp}rjFL`E5a6ID^ws)educ-M~@)^F=1x*H&SW^Td{BOwtD<)UdZq!Zuj6r
zjG{?09H6QegP8Kw+P8CsB~Zw(#z7v<C!na1r-bTd@yyD$a5$qBZude7<9#NF@;M`K
zGvg<GsPxP^o$9Ob{7#F_w{WF%iF5!Ocnmk5IL19U3FU6z5FD%9)>BIQ2-)zvg&5rF
zvU1%ULfeY(^fjw<(FVVn!6pA7O{%9x#tw|8ZP6ut4c$gTrIYcO`ioY58}B0K+vVim
z>tsa<72tr~JxB?V;<>b(wrl7HDQEkm%_}^1@(Rr&5iz4|-@*M4`i^(W4~K+0ZnEZO
zWe3lUMz{4CGr2@&0fJpt&drRh*AS}8Dh?(HUEd~~3p(`~+pY9!#s`$2`a10V(e&+k
znU~5}fvxT3ddT*%g0*OrX2zh8$9y*J-YrMg8Ih8^E{@B&7JXR@KwMj=BQY9V*iI*@
z?5CFTnB>4hO@)*db8PGJ{m0zl&ZbpU$7si)QGK{t|KT5}AiGG9vXlD)SItdqTv|h_
zE`NR!`6sp)NC#P?_L$4kVFs+p*jgHNE_JJ^;aIpFxJ@j-W=fZkp&NbDR$}j!?52d!
z3cJh;vTW98)wY^0(W8o>oMmOzzYs0C<{>rkT`jgbiAeYm-eP9{_)1*3*Nb54&Fnw=
z7f51SZiVanbbd1S=CF3pqG?K%sNG%IJUpd*c+NcxA1EmJ!Fl6w5{ZR)Y<E0m1JC;Q
zn|lP~Fv{6$@#=uoW9WFw3+-#S26A|8Ax_}w4rDnBn?-n+lDC6UoL*Rd-03^F7cZVh
zt=Mv{zyNDnv-lyW$mjd}T3@2)ml^m`70d4c`#>OFka)Z9c5vX4cW75ZHBmWWC}m4t
zJs5M&gF{w5cA6*)G6ycQt_8RTqq<a<i~~5SQi3VzTY&vbxO5r+Y$7MFBY)#sWxBl4
zT6Y7@FyS{va0{2;zLLBcU{mkj<K&hPr16f@NM<aC8<WeojAllj2^2u}x%%}X%z0C9
zGAjhV+|96?>VRa6#47d_jEsYO!Uo8U0C=w8f?hY}O*s9z`zfO`-R?t50>`Jpvk(fX
z!ANkfjllW0n%TFQxe#p1E_ZB@AgJs2Lb@GXpTg|*d2C%wgwVr}2|!{s8uk!j6FVkM
zyGRO*gG-P|v(2!$6=FHN1p(8)e8QBhdY--d^*E3nKs!p}IvlZRd5|*&UVy31vUO=j
zRu$+okAqJfeNAf=0O~q{MmO=q9`}w}JNAx`T0AO|QEh(w%eDQp8i-qO@z9n-`*0W^
z)-wcz)2Qe&a)24m0q-fECJB0QLC-ZN2cs1C$XY-(?L*PXB>|djT|0*geBxn66+j(I
z!+@Ei)P&Mi?_~fFTPg_l7q%0Nse_7_gYx`70nrtspz)byIm8>7z`WP2rQ!2aj~Yho
z6GrH~wnFS$hH3!M$)2bl0g<{cODP=ym2-INKkMKo|L~@vna79`xJ{MM9*PqcgzA$h
z^j_8B^%)U9cv&mBPYhtIx$&n-0@)d#d(1HX_yD&t$&H6OgEdabNp=2_Rg8zifpwnh
zCey2A3aW>x(pI>@A|h#sjNP=3)-;5RKt4Nm+BY2V_4gKxnRp<AOr0oL(Ix=Q;QR(c
zVvE9(7-Y{`x{)_xN*R+w?~y>vg~j}NCs;5S7g#`{;*ezAhwd;=_>)eqFmJs1?V5}U
zkD~{w=`AURNC{5*+bcZmTE8$xOYF6jcy5$-4@z>=7hTi^5dCI*Pn%zzpXY`yNwp<{
z-yLPJ$+jc%ai%l=I}_GG@(1-jIzejcEIn1CT)`ZQa3`!+OG~re{X{=HbSnZ+fa}xR
zysQI&;;)%DfxCc+m9=tQ8X)|h)`lpu@eSU@2IzO`Bt5$jpjesrWDq)Dl6$BG=}3Do
zCzvxQ0B4gw97E?#*)zp+x-SYzxqPsr-a3s?k8(V1_-vpjd+}M~fm=N9q5hJ?rjYn-
z-*$l?x;<uqFWx4qM54iMli6(A9%vxXY~wE4ztpSQ=WU{1^*)*3KLC$hLEm4zii0Dk
zyJL1?ufMym_YgaDF4K;1DzSB1JC<6<!KR35hMEzZ?=ngsxU?Frz?)F5T+e(^S{Db!
zG`Lr<HgQugB>raX)`AM$2&1e`xGq?O;+P;MNx4Zt*66%yuReC=LV}Nw1=tWs)Jvmy
zNw@p!Gg*WGM2f~Gk>&kjnhH~zva9xEX2<($vFz+usleeRUZu*rU0`0A_R^Yb2prGd
zO?jhFlKSC=89UKddx*fRHMAGAC24=uXR!?f8tfIMY^N5pd0$O?V2W^e^V5(QxLHA)
zTobWt+%A;Pt_T{)rr=S&vw|1`3Rp1}YrVcAz~>v}Q#Pf+*{?<{Y?WL+)2{f@utP#M
zg>{M+e_^6~oma(M7ah3kg(18uGYjPO2A&I1<~X0G#bDXW-?6CSN;~^frd`q;WAH=k
zLkqaa$6wWt+kgGTF96C!9MymIRgC|IU1VhcpM~^gbt$I<c7)!?KhD9K`Qoj{T1+fq
z)?`d7UgImKYsguJ(TIDRMoq1d6TCm4t$1M8cz}Y}rtkzEG;JUEE^SL7G6({@ntV^1
zt9q1(;`;j>|B7eACE;IgX$-L*#EZ6yez+Uw4zZ1%FufZ)J5FA!*PCAFPB`-GEHAE(
zp03}YMIn(NL)D7DtiMLLW0!T)+gm{n((T|xXEZpX%Y|qSiiD97z)XSVXklRKi~*iZ
z>Ri*47G;-;^R!GXVAqaRu!ISR3-N=_&1N!qa-}7TV}SQC4CZl7BonC)oD5cl*2Ze+
zi`E*<?9u2J%A*>V=R+h>zDf&hA!In<^p*u<8gQ<Nx?IddF#kfGl7V;YnTOlyi_jXd
zCz}21^uak8pF|2sm6d3sGnEH1u$8Ff0YqQ4O^hu=9K7M3R_mmg;U5{{k^~ykVN)BV
zO@99DgZMg!b+NKQdrRvmn1QLTfR(S<Um-`P+cUUjFZ&uKj+6WdVFhEL>J#OeG|zFK
zjagfd&cR5Y4%dF^hPo;+!y0kH65AA4Q%pZvCnVo@e~>hiQS)N?8|-ss7oDWb&Go9m
z@BaO|s9pW&X<>|rf-DJFgrMFpFkex0UOlYm<*$WDg9Di)#cK2ARg=W#2KfSZ7VF1%
zn-OD~41)CvYlLm~r^~n<@-i@VgngVDw>FIKT0<8=9rM#9?xO{`V5|?v&Zl*$hqilu
z?SM)0Bsz2B5q$A~ew+EYrk@>31G{wny$3<ljQ0ULInEb95ZBF_q?eu2`HDnh<_*Af
z0Vz$^vSKuv;psNv?Ceb5*&ks~IWnwHCATMXNxC+AcV^g+gl^Q-jaBjJouNh1^qR|@
zGW2C&aWTP#X&R9rysNvgn(W0nm#n5NBT<1OCq@6OA{M}@EQZ1IZtX@OpERw<KLf`y
zsp-0){JS{M$T|gAp$6b0&#7#O6Ul*K-J_sF!0TmIK5{2rY3yOI*RrB>UA6cqCSH8%
zUO_3^gQVKw<sHaA34?0;abM(eF`-)`Q)g>c7JCIi|J-q`QiO4>F2fa@)v2S0U237~
zYMEYl{fkDAYbV@lA(UeEBle;-eagtDl(R7Sb)4TDBd-<m8#ZgoOkO3xh#%9!a+Ohv
z@5erXw5fXLx*~?f)PnX?rIbUcO736jYlot9)Q-|cTeHcZGOvQz{FTz-LL{gY#$o^0
z*8B1GJtI)x&qX6ef)M_c0v=qDd3M;WeTQ-5+`GMwHK#6{HR_SO>51G*&jo$OwZ_)1
zjwratDcHCIXZpZd0VcJJ-ViW%azY<8$>|=5#GV&DEf@*?7Fvqy@wQ@3yfLCTZ{oCr
zP+5%`o2RAj4}r+MIk$STMd~bb{>o`nk<zfhE0!}e(ykWUt{^@v4z<6R%!tyQ7=~$W
z6S~h%Nq?KmCMEj&zEj+#ksYj>eJ5%LX!`u<^|wpJ?!t<BbFsciy=^NbK6r4J?4k`T
z_Grb$FTyg#q7F-X25(L-t<ka|bKNti-*S8v-4ajfGh-c1-cZIl&43X5bncb&o8f4s
zNZZ6mIvQ>{g8eiBumvgs8hV2AzH>h1DZ_M!UI$oT)DIeX8T5{|oxlyC@K@RlO)2Re
z$V<hwrBF@AxMRwo?zKY=jLz$ftJx&jTc?lH&-?x4=tk`4fCgp*l8jzO8{M$h>#Q6)
z#O4V^`8&9TebLW<9;yGq4`gIx`+sk$1poiSa*2YB+#o+f&uv}0QPwzPlioF+ie?3V
ze*0*Ur}<eRiGRb{{O`>{Du_h)p2f*buaDgWm;oFF`qgxUZv~0EZ4CXMDFLjx2|E>R
zZ9thYMM7`;l&^&Ot9dbMnkYyD{>uB^(+X-*GdVMvvB$<DSkhP}#&u{=B{lv{)jh>v
z)_s2a4iwXss(>w10SnHGE1VyXOzgJ``^#U7H6jG@i3A8{!ZgVN=aE(?3+KI4{zS8m
z;FY21ZQQOS-wk23hxnp|u67BBBr}Hby7Y-YYdG4-Ipjd>ibA3Hl~LqsK_QO=V#_bF
z9(7w;qK!9O?x#yv2A`OGUABe}&6T{~BoEOTpb*HEx7#RJmf~;NsNyil|Jr7z|9k@e
zKW@nXVVhO{^1%nvBZTee$T<+>dOm%eYN{gt8~t)-f)hr>>{oyO{6j=0%f|ESh3cHI
zpHy+HGg4hH2~-oW!4+<n-)~+6c*yOEOpvRecboZypvaZ7+QL439ZmhQXCasvuD}rF
z9F2pPQmbPbT7ch$DJCH7;nf}eeZ9U@CfAo-=oy!)-gqJx3Mn4Tf1#g{3THEeuQ_%F
zdcaDb^Pj8zk8A*j|6@8J_#alwBrmQ;FK>U!*e?JKU_=4B|6f00{*NtXWMKQBy$vP*
z9Hv{WNdLFUf>xOra8Dv&cf2RFPFdFaVsL4C$`5M*YsGOCd?cRx^<k3_h;$^cs;$Kd
z16eq~t=C7G$xXui{v2BC`|IoWc%d_Zl#2W}S^B1zgh;BGHk|6=s8*7<Bs+0)_trN{
zJW-BIlvHk)MTcYMdgwa#Pv%co)g~hT?$fh7duGFqR;Su+u{e~XF?Um$lA<}_>0B^}
zEERDEn5@{|la$*($KE3R3r<FAnUdd>^M+4N+_Aa+d`)`eK-*X&2X>oIiA&TK2oBey
zQ7(E}qB(6E3|}_^0l#3ds#M|QUJ4jw!lRkwyPDFzrWfx{7G7Q#UHg779=Kx{vUG`C
z*hW<wUoyG;O5|R;Xui0FFl~I!yQk%3XNz#5IQ({!&)j+VX+P4>Me~Jb&>s`6{h`JE
zz}*Xm9q7I>rPa8SCJgVNh38xV>9CCfsz5(KJnnljf_}QZibCWhTW)t6i~{NfL3=@!
z08Id@(z;g@;9{6F;{?w`-lt>4fdYWoW&R6i_FXM`8Kr$)aFXiVp0K@(`Z}4eSCSNu
zE78N(xEww+ZZ&jeaUyrwzvAMm@!F;o##dU*G-hnBw{kkayMXvzXnwB^LyE$yv}rQH
ziR`<E(jz<SN5T_0?Ok)3^7lBKjU6!dlPzi|P{YXqwwUo4{n03r4qa|zP6F)J0UqH9
zhEPVWh~vq!0o`0P3|m=Bc3_LKcO_PBBtS+=yug&q*|Q`-7CgscfC75`6g|+Ar-3mX
zKw<h#2x@ogE~lTPS93zsy?B;kZwJhmjLi_VO2w}y^1LhCIC6|gv{MnH^YkPho>%=f
zh_G+OlBC~4j68nz<$<00HQ6)|;-wMFM2H;3iSN<<sweG}<N_%5L}uJ*r)w&ub_&HI
zY>*I>rRTX|U#y%g2B?FA)iWc+>b@I}U7$GG$du01!=G(-y9XRA_A2XY0)f7mbbG;D
zKNLhltl8P8b!Y&hEC$LTYONI{cV!l}#WDyW=lvmOoOjDxCU>}td)qh@yUR8)B>|y^
z^%)D$443h!;eji>90l+JAFIA@Q{MGO9t0y`<3Yq8Q}qxcb3piD!Hc=7`=$FJW1UN_
z%+zfYW#Tbr*F<)03RnbEfl^e-Uc>eIKp_FU6hQ-<%DDR+!sy(oZ%k@5REtU+v1j4)
z78%2OTD?Fqg&TCbqD=V>uH6(1Y%{286!GOM)Y`>_C*vuh-2jiKFf7$N1AwN-`ga!I
zjTJiL<dQNl?-RWz1V&mf=ybE$7{OZrD~DiZ20SPhDts#`LxZ%fMW_225zgiK8+RTL
zLHo~)Iid6kpzW!$?r?Z#MMq4hmMJ8vAp<Lai9kY3PP>ims;*juq&8#6*@Kh&G3rYj
z3A#rEFC3r;_MJXA;FOTjAD<hvm*H1FpheuM$T@=Vh>-e}M9i(|(`#FhK?}SS2rxF6
zZkBiq%0zsa73~<@*>kfRcv)ClP53?UvElGQ!DC$dbjTo|7<92iwQF9jSL_BEafOAX
z>rnR+=e!q~!&MLBLfrmpIB|}>mV?a36kk%e78k&fq9+?GdlcB0)iA~X2{xU6hK~m-
zYdti>ZdfA3W{q-hy7p|s^pj+B4kpo+#v5n#pc;RiXNa8PoOVI259<{*yxsHmQz9AP
zn4sM-QP>I%wr643uH=iHP8ez<QPl6IPCbjOu^b-AF<@MCfMxtZyUf$^qt5c|wr-Q+
zrV^Nr)z0U7hCG-lWn_XC_6D@M2rPBE>Yaghl&3vPyy*-<i0AhSM6gY)f8V5S-42&J
z9oUmx*R?NUFge~@>HV^(&v-RvwSi`HeG3s0{M?M9fuhI+><+;X(8qFDT}_A6oQu9B
zr4dcQM~o|(0>Z2`QX`|es$2vLg~btrsy_|KmsH!-|NK<@9ZcD0j6OMw!~ILnB^#0^
zzUIE9FOROQOFg_{>3qhB>_f2$&J+wdV8-<5a#i5rTX3uKo}Go%AIF@vfmaQfZ&SPI
z=nEL+BG!j9Dg0}$RPoJPV*83V*|W{%+|G(8`P;$(--_LUPlMU46rd=U?6ILmGlYN{
zg9<bHLj4%)Wj9K0v~it{ii77^w<Oic5mLj5GJK<*aN_)i6V6E-Sz~;FJn>vDPwy6G
z$faBRZ9bfvG+erv<3AlQ28m3Do%-`;Q&BDBG<Kp{jhaS)afq4V9kMXH+5X^u=Us|1
zyfBUvIR-@{%AtpyHUMIRd~H}{=Z5wPFhczax1<5+K5k~9XDJpjJq-1lrhG(Au0g9x
z<%?|p7Z8<-70@pg_DKK~T}?caMUXqFkX#K8fKu9o^E)h#Z+Gag3@P=@s@aPZ#IJoq
z(yV~GJyG&kEJpKA37YY25hud}0gtee^fL|9C6K4Seqkw@);&5qy`)YXVpBM`?ar7r
zc->DU1q@cNU;~Xt?RSia+mx<jn+A?W9W2^Hy;TP4St^~)F&QDM-jXDXOZq|zOSiti
z$yJnrLoVFIaXU!X9Tu(BZP&g3pLSTd&gv;#6eBF}QeXAuI{Gr!-aq)Uy~ClFCv+fu
zt_9=kk*DFRDzkdaGQHR_MuRa1zq9cRdjs-*LUOSwK1B(y+aIB%1zDz!mJ)h=i#sRx
zDdGBA=;SXBmK_m0FZ7Tsk|OV!cK9P8S+RxKy*6x7B$H8r7od)Ix3)j+5mz<r^Z#-U
z4`<4DEffbDbFM(6U*v2T5rvWIL)OkKBzSkez<*x>q-~at3~HAv#rt=!m!+9C*ykg&
zIZ;6y+W2w2x-EPA8cr~$K20CedpJCHV`$BG+lu}##@?w(w5^M_Oxw0?+qQG3ZQHiZ
zowjY;wr$(2eQ(r5Mbvrt{=$kCG5eTvwAL+rA<*0F_SJtmWAt2GYBSG;Ubp@j0BvUa
z(p<U<{+TI-H*F$hqp?%ZZAUVe5)KS*hn-+p_O%T%*F5~>*UXcX{?AbJAIXV{iT!`R
z;h5IgwmW1)^4sklcy)%tYWzYe2L7nivEHDqIXArI(}9B?`D2J{J>Fd8zS#Q{nw20!
zDb?V12Kb7Uz`}_$Git{-fc^^I^R_?zTFJhvYtcNWku%|3-=S6JY~Vc6#ul|~dLJip
z@pO<|ncLHi#1q#7nNqI}MS$em{ds>qy8J1RE^XmX#M!BykxZ`U_j`ODvOIn<_zeHH
zII72|U2FgLvKY1;$KR{x<9>SHyFaA(n_bt8vcOQYWJpOn`G8W}ov;us2b?)DWDPYU
zy5rRDGzy(S`YO6e1TX8RwLIFrx&IXEq>*A1&{(m#ize><6^>5mL{@cC{sY@8k&hP0
zoC3Mej%nI{)4F9(dII5q0mqaqGoyr*f$6>^5IO0gBm`Rog5RQUS_Pr?W!kJjgNo)x
znIjv_j_5&`E~kvpuFpC9L!H1*u(eHaE4-a-^F9a7f%N@2@J9gt>ARwCc@;wI2R`JQ
zGQbAx5XP&mUU2JNz*>%$GKuNZqPr$BOJfg3)$rFjMF!`<quP8hSjDg956k2)eKQfO
z?YL+a45Y%*DDm%Gb60I7Y~*d+f%+^1lsp{RH2T0`^$I|-MUVRR{@d&DhKuZ6sLTcT
zSBsXWxj#p&?c~csT=Fkt9+r*6eG%_62S@v8XsS7AEAHJQifYuPA3^;`!)YGv*+_G1
z25fBq!ID15oxpDqjc0<Ch9yi-St*1WOV)|T0hN5{X@HFC7Iy=`dIY;Lf?^hVrOo<Y
zTxNsCq~&(1O9=Eg{4t{3v1C0pp0<0qULkL9ORO0Cmi;OLv<Y{<fEQmsnM1^(mQsjc
zZa5xz7A5VSLlNeVbOt%R?J6p4EV0gA(OWn7DCVH_pE~Kts=dq|5P)ppJ%^I(z3MN+
zzkw({pK!g>d`1dbG<xvlP_uznUSyhvRlK^!l4@_@4V+eJsX?q}3}flRxj@yI9XcJP
z{Q_S-n6JRmnAUy%LH*!B;mcT}&uBANNi1Q8I)RkD2hQEqng3jeUYR4qK~|dw>iZ;R
z5IuR3u(3B6#Bg0tl7T0fawO>mQ_JZoRKClT^a<?-Q2wsYd}9C_cNJIJASI<n>n-12
z+RT-^jg*v`%x#==U9PK6RmL>91S}ZkVb)IxpM{&ZlNWgCZZ#dZTfCTB+s$Ech=GIc
z&o!Gc+}+||&6rnx{Ayx6#X$Nj!X<Ma$%y-MnM?4;BbzCD4sV1SZ~HppGw%bHq9cFD
z$yWXO_&R6H;OP}+H@kE_#fkyPH8{x@ZFQouKun`rxc>}UV+#z$js4gMhB5@{LoskB
z1?vaBTb<{65%$IKa6ta`Hg~{`Q6c#R0+c|Wvt)#Bs?)|%j1tePxwKUF+-FTQm+Zqz
zpveSt30n<%4Ve^u&{_l)zgTBCyJf~41C!Z>0IFC-bI^{7;jaQY*}pUgHyPi~DF`Hj
zQPe&CS)~5J39uKiyK;5DQn_)R)%9f8<^og?Z;?4dd#CHEiUKC&BP<;x)P1Qn^!aIa
zL$GV<54CSTUzC;dh^+ykF>tax1tapn!scJ%kJw~ojU9U)gB`GYvtY~*oL}_}c=nL+
z)?4?k0owhVWQIdvq$Z^Ox1gsi0DdQvXFmq`71#$a@l7gmz$4)+@-#Ua4ZtxVB9j}q
zD!X(u3|jA3VCQdU>KH-Jxt}&`DVJBm8J>v7@K?dfIY1u$<Lz=3L=zmGQY=7Pr}^Bi
zW4q@?YVaNo+zXt2FxT<gyqFI~9GwSwORm(zQ5F9*U5a$fpiZ-u{|f33J_VD=)_UwT
zS#sXsYrsbBX2V2dzN%{L9&r>7Xxvo~0N>ep4@eS3e2u_(jy1lsRa`vweW^vTil&*Q
zmG{luW38Xsi$4|Tp)#!Dyc8gokh#)J9*L-dia$;Hc|hvUa_ziATpNP23WCyF(o0)n
z1wl9CF;PBQqKc;lE``zL;cuUfi;zH@5CBpFPdDOo&e~J+v}e?&F5HxoK{I2AhuZF}
zcu{~F05e%M20pQq-wPG>U&SQm;j}?3H9;mKC2qbJ7azpcmjm;yoK#!rxAv-_Il57}
zvq}~M%JS1gNyG`zUd)Bii@z%N|L{itK-Og1Z=5a}swThRVHl9ngb97LN|+K90sST0
zgi>U=usQ+kSxnYE864~}cV60Pe+Hq$=r_T`tbZQesapBBgYY$;2rA?x??~8zjrcrp
z&D)D_z^+a~HMDr<z|uo=8bea=qyCBMUo5ZO-)p!5C4UUkq#J)LK84Tj{Go>+ir2?Q
zmu@}2XC@ZF1k2R&9guiY`k5?_<c(EX+4@fC6b(4ru>FxV#Wea*Zs2;s82a?-cd_D`
zRkM)OG=XblCC;Am^4DPteN?3C<E%&+_ktqh^O{G@G<rpXiWsAEJ&3^9eu703;9kvz
z+EN>i(tiz;l8isw`(|~Y===Rluz$D0+nqA*JTt0_Zk>-7SD#s(^`m+SHwoouTvSFi
z=I~L{6P2x@6UbHW{23sci9BAX+0_-I@7rpGvcU@#@yN%#W*AS9N=t1bFt$7Ic2ZDn
zbXI)|3oR)>OI84`UOZl{mQx|Jjh-{!zmNG@_$xIk!sKo_?zz&o8|(U<QXQqdeS?Nb
zo^n5_*kj-LoFXfxoIIG9pbA+gMO=W&UMfnCtR$p$kx<ejZ^vTNt$g`CNsguOqfrEK
zQz>_O;$<LX8qOB|_?k#Ya3eI%vYJRa77Z7Y-Z(DQ6Dc1^EJ;F?{4?I;YTY5!IS|0Q
z&baL>+j3LE&RPdSpaF9SvdtNml6)O)e~J{h&FXIoB{E_)Yb>TL(!Ds70+VC@;(23<
zOvz3?z?OC0x=ys&WiVD=x7&PBX;a)G)EF+1>^#xz;1L(Vk)=mBfWRM~hM#dR=Fnxg
zQXxcHWO+3hAqun*hn8Hu&(E@{Ii&*xFrH%+V-1um7RFu-#gm2J0$K)QAqiAL2F^d<
zvDWKbwr#myBX<mN-+=-U5)w2wdfOG`e1orjE@%1y%lW^W{%2tNFT({Z<Nq9({;y~3
z|0?6>5^9<`+>>NX;CHqi4ywnkEyY^5RU%W(|3ZpZh)m?H>*d0m`x6;&WM~J22mBvl
zyqmCfy-wZm?#_3U=j%^uqv5}K9nrMjQ(dBQZlXx$eAu$_jTeyy^?X?gJJq{O%jeDe
z@qSXTCRM2MC=H%}_xtdK7}l2WTPT~Jf7j=&siU*|^Y2Kmua=GpReL>uW#=xR&bQU`
z<);>DsW??OlT9CDb{SLW8SbMDox0}HSv{!R);-U0g?g2%x>wtQhtl;tisZ7|wSDsu
zXa_j^BCdJc&4w)RlFsI08-ESH4ZXY+|01ty7(YhRc4Mv``_ulK?FfB{6ai$wY^w;S
zxF4<CXGzBhi`jTS4j#4N$(?%33l?+^6XSl)Ig7aTR+@`mtCsa2(Ac7t`y9@lYlbH0
z#eH}4QCu5#nn@F{t?k1(8In~Jt(hoMxP2^(kjd|Lp$8QUOy)V~P5yLK7=CXtM&y+c
z#&QO#9o>H(@p_z!oHtCaS+!=`u#fTr#cd()g6c3kwXEa#J?Z>1p%P69qL2N{9;^Iu
zES4+TVq@e;SC;&<5m{AR&x@V3%U`8gCZkP$YVvkCszgna?%Yt%(AFVRjRZ%B5v)#4
ztdT@~zoM3pj@E@tsQU&2lYHfVW0~||q|NpTPPCJzJ!Mef5C>M(T`eP{mA;Fd=sLfv
z*y+%ZI>?JHHRG^M>KuU_mNKo6Hxr7@I%^%g&qy#1f3C|>CK_Jt<MDbxM0ud7Z^b_Y
zwhrr6aj;YVEZpq@<BMGi&cNcEKtNyRzK$DL76<LNSQjaS3fC{dunf@EpY+>3fldkF
zohd8)Hm65uEU`X9)9IOVvi;~%jY|Tsv8~I*m=VqN`H2ghdDvq^8Ev_K94%G{`>=!i
znx#0UqTpAZJuU1%ZrglIEB4j>-f@c>SfVnMy5F??Ve9rK2fY1$KeLY}wn|Q9l?`#r
zRm$0iEll#FH-hi&Sg2S_qOkD;)a>DoH)NuzXpv#&3a_c>>~E?}CQR^?X8R#VF8jRa
zxw@zd79-=aBy~w7Ak3xoL|~Xz-ZX=FX5+4qU%i^}%b_K!+qQam7kFuoho4YdmVIty
z0aMb$BmwYjx*+&>Qw>W;7l|J@aNj9dFD*>%h!&=>PEsC`dm#iMt<{eokzg-%%30GC
zGaFi)akI|=`MeAM%0~HvEode*X%HDw=SL;yotIA0b@mjE=d3!;K@nHtqMR)0WrMN2
zC)|}6U6;87GgG06j5!xB%^(>s$63{5G}0P>vNj{fWMmKz9UC2##+VENl~rXS{k8a>
zR38O=(1+@4it<Yj&pm_db&%{^Icu`TuBNPL2&^($T`g2HFmVS)D3A&dp!Kh4mT-ds
z_nEnT+8?`5nQ$7DL&^Pab8JuwWPqWeqE^`$02*)+WgY8SVqXXIwc%N?ZW#uv(Z-}W
z9Iz{8NZ?6gaAU+gEdO3PTuG`teVecaTZjVA`?^H{xUOwNWP6vpxhUgdO-Owi;`@7U
z2!@g-$$qQY>58zIt4{=-$5we=I&oc>5?1iD3FFH!3k`?gFW#_71@7(=nLFg5X+mGF
zS=Xa5r6|cXK2ccx_>5yJ>=V0lCCkPY={REhtUl9;EYFjFxY#h$d4)ZXsdxaLK`6(U
zpeUm<rCOda4|J|jhd<n2=WktiiBA?$ZeG`KC2AMV4piq!^BVykW^|Ci)rRZ5$C%zy
zTj~{k&mi9Czns42HBoLiFlt_?1!f6OiAXBGY12rY$udl<SBT)*HF0?8ZD>n5rr5*J
zwcrO^TvE)Gd)7S~J>UJ|LFY-K;*%_DXp84uLy{Zqw7<|gH7Buxc$iNmT_u|ogx)J`
zYKx~GBjV*=(0g-guFP`r0th$G&CeTqV3dT3NOwx+!!Uw?bm1Q2d-5&|t0&lSS=_Dx
za^kh^`n$TB8r=os%=X*@=Zo2ZYQG-@w&L-nzw@#-^fE^?I;104V;sMMjc+i{MSG_I
z9YdO*lvI1Ezd5qFIz~)Ul_#CH#HZ656<ed5#dt;u4K%L)$Jop}tG%sf@1PVqA1ToT
zPq~u}X*Q1ay{?|sRqvHVD%%-;DYej>Y>YA<DIDPA54p1*D=7K%!f1rXYs?t6kOBzP
zsf%sFwuoNIk^%Q}>X5Ii%CWRuDe3hWa~*H9N_aDb|1WP%l*)iT;%(Ahae?N9S}-4(
z6@XjKK{Y+dfyp&BHR{wwmJ6Xs#-(*AyrGbN`yLF}2@eoRn-5{6dnRixbQN?Aj8yLR
z;g$2B)DsVbg2HGA=@6vy6-S<n<ayKbnYK;fp5gRsJtShs8kKTpUGFM<8#S*iNKOPY
zZ#a}mXUl$|kOU4@5kyT<00$^Jfp|Nol|=NiER{AyVMe1NoDOMq<RPwVC*5hdR2&$z
zH95&ZoFov=T#_vSQGB;$h0n8I{yyFh4{LV~=|}C-z^MAv67<w(BehzcV{Hz<W+!EU
zojLMEWi#fgwuYlJ7$~a@<qO%??0QH#9SDs9GXRyFnlB+%5(Q*Shmp?oLSD}VAw-%w
zKnJBa_u}I&Xbt9;D-3C)!)OiNif7e;hpVd&9HO~Xi^iCAPzZ5qah^uQ9Al#RyYz@L
z<+ewu_H@j9m^DPHQ%6hE_WO!YI_do_NtP0`OCbY&#u+C8hNfT6j(=LkF=D-};}UIz
z^csiRrkX(Q8*UYdWSTwTlxpO-r}ed|MTFK6CZr7of?XaMI%A3<B(3kZ_<Z9pg5tC3
za3*+|t;fHQo>e#<HpdTGdP!ye%u=2s6D2>)Rk6m=#iy?$D%-h(h-#cnEV;)42DvX+
z-hnZen8lmtoKyM3APj)$YjK=r6}g>8otTw(^&)Hd9iI<9w(+*A-(3ywOZNI5WXMSs
z0w#F@83=iML-F8@#*3a9g^u8GV44Tc)~7`E#ytwIh7PxxO<X)s6lw{x*sl?iCwQ$y
zVvtc^-}GCG<tmen->c4VkHpIv%irv?KU1G{X8xwvdRQaT2(9+Ic@Js_5PS&BioPBV
zwjDV+EAb%Wx(eWep&e^&mUzvhHg84-59t3&zvKTsfp2<y-7kUbQF4599^H)xvu{hP
za>|xdDh6RzuZmOP53@}KTvC8#8)6KLn#Ol@aMTdgkopT_{OTq9-F^gM{ad_SkNUzH
zp$nL-x?{VUJXUp&=983bj5yLw#$LSO>0-b4|0Wvn4LZ`gKq&P~XD0r%XW+B%NyH*V
z6qqp;mOv94JNA~Sdp~T-H$ReGKYhr{J4D!}O6J@mXT4b_EglWZ71CI09?ZWA^APn_
z@BYLevm`|PuYkzP{QnW#|G)4pGvoiRK&#fg{69Bpzde2-{8Ve{$I{;3Yrrk-vYlZ)
zbFlUlHDN)9)b$Elj<lR{FS)&hVaXJtuCBKo0{W3eVS^q&rcKTcy?Ie~JKjEzCzV!|
zXsPPM%Lga6giAzekqUq5ZC9fFOtdqi{C2+%E3L$A8loYEY0LzMYCAq&zUA5ReS6d&
zvP`oQqfBbdUB9>EZt0t*Z25lfV0%8!N8|Xn_U^uC@$GuPe+H*Vsl|U`VR2;r_VrEC
zLlsP#x!p14bsc4+Ro(fz)v2T!Ti*sJNa2}pcZ!vNhzc*6ggCQ=(X9y`3uv??tRlvz
z4@|+G?buYIucR_<Dn6&temd^9Y@~3uP%G)=Pz{>+wHdNgDl4ipuu4~sd#Y9qoU*Ld
zhFd@kAWItH;Z1PQSw=r)<2!o0RHfW)i7S<ow(b~ft;N6ui~yJZ>UA4uL69jLcI%r4
z4;b)fVLN+8e0b8_yD9S|J9SA)a-=yZ7X&U!?UyDGz#w|?8RVJYw|Z}+Rn}-D`*o@W
zSQ4g?@}N`1=Vlsp*IanT)2nIJ+hDUVtk;xPb~d<VIMY|u(6+i6Xdfm5;kOQ7u;Z}0
z-Os#tJ=^aGkBs6thZWdX)_9h6oIZ|>vTAGmy>~qoTQJ7AIx^71>9BO!LJNbY%~I)u
z@%Qp>x1D%FMGM&%IdSwB+je%!A|;wetsgkJF!G{l**w*UoOX>V>^$)gvIS1HYhYTN
zs7;P6IqQPkdh98`fD3zeaMQvw)i_B;l=gZ?Xa14|3dFoJ3WZQF{NpfuO#?~$beg0-
z9j_o#z9ZZ9`pJqj<*!HtW2Bq9SJ)Dl+`Eu9gHP6;uI(WN5qmk_mA%Xg+I^BFEgvpn
zh$5dJP}X10D=3dx{v-l~?3tmk?euJ$3Pr4ar03X7YW7r=hjFHx)muoku|XfdbP-$x
zl0qywjfr?!uFq32gq7qDZ%j=p9PAo9<&@Km`z8Y2qiyX_uy)&ew<OqNPVZoOi(B5l
zo!W_JmeE=BPf!OkSW1;x-9fz2cRDH?u@fFQy7HchZwQecm8*?NXdJG~H+&4!jkk>^
zB=5SFN{Aw?@&wJW-H(+@iywJiuYKt4P}VVB6P$VAw7ORtg+q7kkm|#A^D3)V!Wn81
zD+u-0XoEM2VF;(HNPp@MKIv~Z`_Ubj2!Kd32Ia~#jU~8d<wR7DdHs9i-DIlr)|F}N
z5nOo?L5tIN;hh{z-j+QHaKpznyQ1o_yEMFFv0gV-Owvjr0Isg!!toEw_ESFCg9`rB
z(#w0u6vmZdFFYlyY$1TKtU_o<m#w(d*r1(;@Yh4frPzy*PgslwT|{^fBjFov=9C92
ziWzXe&t2vOfN0zx8(?(h4ByX7v_uoZ>i7+W0jV6-RKP6&Si4g*h3++AI6vx15)TdK
z8OyH8W&^D>E2aa}xePC<U9Ix~laxJN7DGB1mTDOHxr4o17PGsr+dOX5R?d`*yaOKR
zw>O$VgBg^<R$Vns-zuT}#RDAARRBGx3rpF-(Cv=J+#5ofsJ+EBSTQw}GC^vSw<eM>
z!+#uy@hkj_ZOVDr&SWjZmu_#VvqAT;t&P5d>qe6G^xiH<hxiaEa53c7EA?8xZuj7w
zN%zTPWKGmIoDFN;O%wDKD*A8PyQz=%d)v`YFzWC|<_u@28de1mH{X73vzK&4zNfbk
z1X-WqF*F>8I}N5#u~9din+9-)2)KmZ3NM?S^&ey1sy8x-*a{Ixg-c1?y1j~ZeqvI9
z)cn4ApDPz9x|fkPsM*hxxe%yP1GCx}XX>FB$F3#H%#5qjevbWZ{VbRD7J6+ZfmGFU
z+x{H8Af4O3GOjmjV5Dx?+aOa192^tyIn=OT{HPA6be!#uzin<X3m~lEdp2`8Axskp
zH$uV6H?PB%z@^}6+%TlM+^xFu5a`{o<mX8p7cNaX(^}DCSdzt56#l4{M#UE)e-Bv^
zvHg`0Ob;_Z<4Za@05wWS=JAXDR`-xsx~M!UQF^+spLp)Hf$cn}K(fL69HJ%wX)5kh
zTWhya6LI$sX`^Sjbhq@ix!j@!ZA}3u{C!|U$d+a@Kmlew;_N`tYsYaHdYX((y97@a
zziN!1lse(D#tOkq1L@d96_7c$xF%m6=*&UIV6pxH0N+6u*;K?Q6C@#s19--kH3%v(
zLm40qfw$t?U`9~Ye?C!^dQFpwXD>%a6_T|=+Y)yA^~kBqS^r?M3n<+~vJ;l)X$m+J
z&tii>^>G`sR!J8FPyf?$0!XOEe+q}*B`A|`I9ceHN@#2VzyK|o2bj-4;}YCn7qZO{
z{jol%W_!EO-#skHFhlMQ^#4ExoETKWg&49`<*Bsw<d)=6(;x8&FUxf7bo(nb@<y!E
z7AsK@HYcG;(bWGGX|kr;RJhI6S$#18jYBrM8;7a_HdYq?os4LNa$!6sLVUCLeY6{8
z<G<5SKe{=_66Q}V1UG(9O9V^5Tp<gnGtc&XJhzh>018&+L_zE3a|h4c3#iw_q5=C{
zbhQbosA!{DokRn|+*(cvs`rXajx*J7lFz%RQ2&SbT>XS~)UXckqSr)drV~k4Gk=D$
zgNTuH*W`!*o9wu4Qk@57xy7HT#cqGv8BWK7a7v`?_dKcG3M7l77jwv6_S0N3WatyJ
zEpu{P8Vo~v!aHlAHUrmxD%)u#J^fQg`)1P;`|*9dUf00mTPg#<tI*lJ-Y3wo^8vK!
zQBOd`CjoA)w66#U56Pi_BfQkrKjkKhZU#&=y-D038IWxtbxrqiv5?CmGRJ8b+f)o*
zrw0Kz<<*27shCj0g`;_c2rPti7GFjXz)`F~#-7$)l@GNOV4L+)*kbJjICvA?#Gz$%
z8Z}wjb85e(-OZHI)`LI0A+ZURs!LaFNc;7nScS!fm6CuKFHlNPA=>~gOPlL1>Bfs|
zHnBxya~w_W#WCr7@qB;EVv6<5@kMl$5sr?S{L~|esUmDl?DIvdk3`Ji=kHj!av0E4
z`U3UZ)O$9-{a|tXQSE_0p?102#%Ue{XE&;Aam-~5e!qC?VlC61=~5mW7WUFiG@V(6
z3>c*tLr#%5AweL%o2~xcW<cFwX8d*>ckc7nl+p3fB^vP---m0+CUA;vkT4qCYi+#t
zTA(cWFHXdIfA3(dz&X72FY66+ARvviLH^O%qgf)Z#?6DZxl$k~OB~%^x%DaxYA-c@
z<VK2bjjf@~o*5=;5JK1Xwb1yn0z-1dFz*ZGC#ymkn5L-upKUy$IaMF`;ZhgW2aDG1
z%C~Jb0n~n7RNyzO;jY13saW>iY|9_2cT8Td_11!1({lh0#6#OoYQNoSjm7|3AF3b#
zc0Fl9Md5ep5xEX8xM60Hj`@`Jb*30fF%bM)j%AUwpm@H=pVM~MXn=U^+-s$%F<~{T
z*Ku6&o0m652KaJ46E(Ztg3kfvT&n%=cuRYabYBoFI2{)kpzI>wgO0<FSpn-?4vrKT
zq-d!5SfD=1Rxbt8#@b0$2kY4ql#||($oZSn$5Tez@75@1>92|N`{3JxzCE_R*!0DG
z-<Wh_u(<FrLV@8Lx!W?cHn3!wEYhzyBxI#T?e+N|)Kq~DcC0Y5MUZpmQ^I;Uf8u-O
zp|;%venpU@Uj?x_^pJSxV1PC49P95Eqk_0h%Vx+s2Z+o@<FG!QPAbPXfF)+GKUBFD
zxO@GBE1=pLI<Y^efaW$p+JNWPvD^^z^^K+kksWb}t)uE886ZQRSaSGkYJR=3a)&Sa
z>~IquV>(LDfk7E>K4&lQ+I1mWLe)4nZE%;eePCF`!J!0+OW1yJD&Q=2#=(HLT2zLB
zsBbR4Sr8nlT#kH-x$ryQXAAtfdBf!Wnb_R;UC5M}!T{+pbiO^XM%$dGO$L1;^ujj^
ze!8=gQh_9YlhTSmTm(rcf20Jn)8eD#Q)5M)Bet2oOwXYT8f85MJleCHtI-?Mk%uj2
zngTx5qA84;Q^b2G$>mCPIW%&FFe5yu+SJ{7mF>$y8c0qH>1+M30KnL|argx&9-H?$
z$2rD<$mGqxAEYKwXNn)3!zN|zs}=4B$MKL_5~47;ZSc9T6;MV9%#0ovUXWYw<$U5~
z$YDv`6_*$FbF2mqOZ%25nQ)KVh#P;st_G(we}RV40$cub6k-4Wk0MMA%>Q#I#{a<=
z{*5C4>P_ct%9bF4&;c8$I#tJ$%~rLVC8-pt!ero3K(%Bjln^DUXKLkvx4p-kvL{X&
zUI0<>wqr(C_1OE=Ua5ONRx{~a?RNZ7cyx+sSQn~V+^9|d)U!ojx;Jmv`bLwOMOL%d
z@6V&}?%4i*?`B?Khdqmp@O+$XX2Q4bZu@>Of0SPP?f5NOQi+X5kAgMjp=5tZ+Hp|&
z{xYH_t!Mgvh-s!4fuZ}1ZB$vzKA$Tbl`+m%Y^9j2pt&^XB{!lBIliG&qcRf@QJqAR
zGFW(l1n8-_c-foLON)GdD6tOj?ANx8yda-efW0sC{_s_cs-UYz_0Cu&h};Cr9#*ia
zFYM_H+IWqbdoEsYIHiJH1${r`%C^UMRB+pViFKVz+O=z(tJJBNm_N^C2N`1?n{0!4
zp{`RCryYY|VYZNsPnA$>X{xJmkPqe(l;n%LtqE&QM1@+X46>;-xsj_~ORDgDm(g=@
zVH~Hmjn-i^u1CuLMPalh6=wA(Cv|V)&6{*zYNfNnyN7-qEga#@5TfG&4Lj+MhIio@
zVy*z)E1?2JktCzdmtcTAV`7tRztaoHrd*N&XhQA_2iG4EsZhg4WFAJP1EIJBjo+8)
zq0oLEIF{k?9<)rPb2#Ql6ItX*oW{)ki6G~Q(12NBSw-MN!~GiMT26RPmn$rpp7{ZB
zLDGqQ@z1WK^jPIm+wa4R{Yp=HJ5(`kx3bp-0DCg=*a;ZLuW!vWv7cKD0QcO*B}Bv0
zRJ<OoZ(=WTc9XXruc<Epc==<sB?Zi=zOWx>>Ive`TzWrL`bRD=@Xx>~20V;vpjoM9
zwINq(S*pZnEk?NbR0R2x00k%?z*UDvGUV-Bi#b4%Ib&5#T&e8<DpLoJ_ObrGFn$C2
zDmB$3+*8|*0%uN3%8qtFhyc&pM|?K49R>|6RY&FQ+kWBJm^nOKR+Nr7+aN$YUJurt
zsbnr~7Dmn?p4ZDwK1OuN?ipY2JGc^R0rEycF{Md%7;2DTNKT>3c`#|B`Cb_C!7R+C
zXwNbl#iRSwOdJOnM}tQg?ruQ!+p<S(7zC04T@pe@8CVme&2Y+1Ge{0)zs6_lrMfum
z#N@>12h0s8UC6OBx0HXCGaWMTZ7VX|hTX~EMdi@p9F&sMRd*_7(>!5@lYF|Dr|ehv
zCSts4WS$WeR|e-PJV~kbrMG6M?~2EEdzRW8%`btqj1c=|thO0G*iq#b78qPs=|da}
zIs>E+pa37(Gr0Gs2bmf<ah(OTu4Ja2Y2X9m5QLh}Phr0~%6L1=Xs9qDD`g6tJsi`^
zkff+3-epRCrwlzxvE6{3?PyLi$ld3ol_#7~5LLI%#IhT%G@{E`)sWnUG!z^K1=pKx
z<Su^$J_oYky7<j*6@Mss;T{#9hzcyN*$dVA1#^T5U{XYum^P|5W|-zCHUX=S;j}JV
zSTjk((~<m6_Rz;OX6qbo|8qE0i9}C2NFqUDpiR6jeQZSb4S4V0K!+U*Fe(7*-$0!p
zgFs*_f#6w@XHcmmR1g6BU8e8oK$a&hUUZdS6y(O`2va4!YCI_Nz>VwaX;>^%BU!YW
za~!gKWWQ^uhZ7b~xWThSi{oY|Cq4kMpdw16X&IUKQx$el(vWyp$j~?|LXjb~3r)zc
z(&+wWWY0B;ki1kZ!&s=W2Sh(J4sWP#*5Gx-Phh+9C#r3sQI>76xQMft*p;U4ns`M5
zvsODvTI&)@hH%wLUH=!QT)*}cQA(d-QMbsCjx~x)FD^)ivF^d!+|JOg%!FS?vp#XP
zYjWT8skKhBu&)^J>)FC!%<^bmLtQGmUO%G7jb8H;W_^@L5;}is7w?pzMM};PqJ*;f
zBxUo?L%L+n9<t&N;W%xPewO}&<S4MjY{HZ-%_gT9KQgXST8olB1T(LUoMCFbfG%JQ
zvz83nx;6k$<P4@CalaF(5da<h63&O8#=p}*@H&zRwvSYeP{n-;+<F*HnQGU}nReo8
zn|zG{YWq;gg2QiyP;;r{qLn8yo$tt$0G1jO-OL(E#O8AgB|UHgfLZAho24UFJ}o=B
zdh~wcSu0WCprATc;Tfj)z^fcGZIWdlK8x2-dy6eMLT-LOWVTR`)Yg*+=xS(T9Xp4&
z`8%MNKyFru$Q1fh->JCr0gO%l6<4;1{XzXx0!j#TQj+8QP44d?=~VN2l>GPU^bsK{
zMq%97ha7_qYf!?5{)QOFW?3q<x`njNlpw<#*Jc@DAuDCHrjS!aHobJ3)xtB_)o9%~
z+-jrGb*;0xoeEL+LbN-u66L91`p6^5Yh_3F@Wp2Sr`sQDT^~aR7GrUKjg4$flAq|I
z@0?QO$>gYO1l@6Sm)Qz?n!`#qv^`^twI;+Q(kLb=@(3}~Ozb~_#^2bHj1N$FNGGKG
zqO+UjLhnA$Ka5Zh`gmQ0soIw52$FTW=c}=&zsS!Alow|4I}(tf`FC;Jj7xut?v;{{
zsgFgJUyG$$yq(Y(Rb0DYYSDn@`Xo2^6<>2}`3xxSdH9yHyolEZme&MmjX;yKa(r7Z
zQBvQmjq{)$OtKeJb?LuWo#V>Gy>}Xrq?;LT*e^EY5xU|PuHJGKt{8F)kUC5by3T2x
z66^vHSvb5bk9VWHs1^E+f0&pB^c^01fh|+P49eLWbxt&%TSVFSWAxnzZ#OfqGx-57
zqTglzD@CyV2L#E)#PUCL58ImmO43IV|1bAoZNZtt&Jd;|<tX`3V!eJ`M+MFe9v?-D
z7z7~e->Lii@zm9aiQsKRB0dATS2urlKDrtQm_zfg(P4jibG@>5$B9i*Lh@guLl|Td
zazE*Swv6d(QZ5qJ-E^<1-|q!)KNSHnjc_EAzF}F<&&f1=qZ|K@OaM<l#<onz(D<ia
zub)%WgR$%DebD!1*zGOf%++-!{;lb*_p7bzYxBu36n6yJfy{SQ35)38#21%M>_pPx
z5#5b~o5)YC$!0B7*X17?Ew5^9ujV+G#|rh$5MViz?xhDA2r>#}pdD$%)0=M9M&}6v
zB9%}PFI8(AP*N0kf>gu;Nsxjm<*#V3Y%yz&0?JpB*&JBt^&*M#sid#X1bs{fL^0{u
zgH=Th4kb#+U_y4~l=`FERE`(0$n*=)UAx@k{TI;0MA;IIVkQ@DOw>=>5{aUrcq$Q6
zKtQHMje$6^^*f4`JgbOgr%q3L&b4g}UFezQhk>i%?M86!IbB|3nNcLkL^%oyNgp_;
z;Z9IG44yQCnm2td;7Tb1h%vrwlNeW2VSRW6kSA)BqMiU!W)WtlIuiV5<4}k?9t%mw
zh2*a+!o>hW;!zX{2RK)iQ{YEt(UUau3qp6LeuRVxfAV5EwZ;^2Q9mC|-p$OeiYwVi
z{o3ENhYxz2=<N5ds)-ri`?^YWo%N;Il692LE~mZCC0?>kY#UOiMY@Ly0pNcx;6e~>
zF?pJ<Ym_U2L6=M?N{C`SzUB{$$@<Yrc3vZ$E?of!t+L8i>a&%uGQxdG0IK3(&*68o
z*rG|MpHVX)X;*_LInOm06SAswj@5DM{ZS>pG%fMPIK*i&51*e($h#ZS!p#UcFERk~
zivdRPQ}P@G={c`jJ#IJNON6WhH|u+fb0`qPGTa64h;CCPF-}m^OXhNb@%`>3r9%nO
zgaT&)XHV&8wWGShhz3VrQAfHD?-1CrN-N~3hkkom6GP-V@sPk^dmS>S6884n&5-rk
zjow;c9n=@12EK_qw4+(WBe8L$i8K#fRf;Pf8gU=QoZ*#}#3UY_m)_2&I`l3l#i;~s
zvT0%m-aGD|w<&&$-u4gq-ZYk%c{dtkn0X#p%xxpF37*2`;lb{~1=MKOJ#l)uRLLbr
z3#*q4N5L+GtI9BV%tu!PUdue|nO?YE+NAPuFsY*G5#q{`M~e3&1L@r~(P8y<r;na0
zsPTW>x$uWi6QzzsOA5kUFSbq)M7#YgT47O3bP|}0AlzPRQir`WwG^fZhJi?s%7h_*
zz2n{C)ZkPpkD4pG&=Baib^t0R;Fsb*Mlc24#NSb)f1txp5TFK|{Fga2r9o`%6YdVW
z?!4rC!7%CS2)WVFF*(rU7J?})1}tR;;+B?JpgJ?asM07=XqCHn%kr?|x@hFMRMcK3
zPH|u*cw{dMnp41LwN3p+lic?#p-J9G=|w;jH7K{6ZD@`QG-_R93>9jxu5v54l-+Jh
zJSu2?vTLEmznP1x9Jz5ucv!F?O3P94iU2fk4Dl>Cqwzp|YD?#-bO{KLRb9Spo2ntQ
z4qh;#et6^3oKf_Ak8;%(`r#I3bP!jbiVb{(8lqhMm<%<v*JNE@(B?elBqLE=x*a#)
zrjDs7NoD*#$YrHhm(E%2XV5|S$j7OG$pXzjF+z4|c_(B5nO&>JW=K#jDFJj%>G5I$
zFhP~25BA<<m#Z_j9gh_kbRVZn##*aECv1=NU3KWxVhaMq5I-19CuF~aiEXi!UtXyR
z9Xs&~MsH{+?Ie`k<~RTbKevVA#IxsE1oUBUqKbw0M!PmV<S>JnR-b4v<u6JarLiaN
z6w|MOpdZbo_j6f`@v1gQKz`05uREHp+7DlvXgh1l@+J30NZnYEDz;$kP^EJwY)#4^
z3|a^<)rH$%DuB>b_r=j_G^4ege@$f(;exMydbUrV7g`TZw%LzrVlE!qPk-mES6gd*
zkYsS3PyJN>9GEpGa>PZ`Q%S78+a30mAe>xHZf}V~TEOEFBikqlPkhdgVdPop#)AV5
z$fxhSij##~jG3v>Fu&*NB?BHyKfWjmE5Vpk^P8D4&T*4u@aOL6A#fP<qlAc>GTGsv
z9-b?iI;cExB8R2V^at8n-nFP7b+%x@v;d*dU$%77gHhm@3b3<3y7T{mf5VKK3(+F!
zyVvg=JNtrX@+=doKv_~|*%-Mlw_%<_u!laUj~-bJWayLDkqD07Ub-oxkio}|iDH83
zjDoS(=;$bWH%P<sKS%q6f2wWm4}+4j08J}U!x(Ttf%^!`%&IRqHxI*j9?1uRhsz8^
zMep@6nAEQIu9I6>SybQK#5x@V&6~q=QuIAG=u#gX?%h;qjtS;A)<4^M-N8$}FZtP}
zi@x2Wy@j{rk2M)m2`J4jt9^PyVtE5SfJVinw{afXfY0+sc5|duc(MQt^QBayryqMG
zF7OeeLXO;2Kpb(&b05%MpP50EZ7Kv<om?I4I)+jr43>h0HxD}Tr#H0_xcGeV<mpL*
zh$>1qcY2!qg3nWMyOdcQ&4j~Iqype+HdpbxHZon=DfkdM?+PK8$VSscSC>xapP1gb
zoU<qOC#Wc|{ze**H@(U#O1$M2+h2ISS@anJ0F-CH;}1M)XA6JD*=5vwO;txZr&|#C
z!8}cX9RK33GQ(p&Okw!VB6J)2dh~MDg0~0KzfxmLbWfzCJ<hju$=oxG;HA})nw)Dw
zY1JC(M!@ba7VzK?+nc+8n3?iY7G2mE|9;k8dZz;sojrTNdeoLpI$Sp%rI~~wqRoqf
zmkVHVfyU?dMSFd3&L2@9rYLB`o)U=mTRG?<lF%hlF;LB0v`T}DLlne=Qrwn2otK;L
z3GrcM3@R#L8%qsz51+F)eQM29RSsqgH&t}MWrdVplx3$c$)M%KnjOD$hEzF56w32*
z6kvb)kLsi6M`REtK%L=SzqiZI^~zs>ZN5yj|BOce1v9h!??~ve=0DN0*#AL7_3adf
zv!*^3n>VP|Q9Bz_*49NhIXMMLND5_uSOm`!^m=*W{W0t%q^L%{){rIY=C>DTYq5vc
zZ9*H}%xY_Adi<p8#$lKeC$28YjF^zMiKawdo5%Q0P>2I&%!CfUro(dmIN4cxzn({5
zZ)eO@_P(y=@#$t!XKVLnuKw)?_Vhly?*9HfUgG<HUw_S3`uY8)HZuCrA)p}*dA$Ux
z7ck?nOfWEE@$E+&%Me&dkcEo{XI@!?KB=9Cp29-(bL6r=d#}?@sNJRPef=nFeBw`Z
z8m^;mR`fJ?1tgt{7YS%rrtdejKWi!pF$W?p2CMTzieohDPV?9G<wPSea>0sc!HI|_
zCWwGk<{WjRAO|I5q-&G^wJ##Yr4&X=#nphD3P3Tsm$a+oW|&(eNR})r=g^N=mn4cO
zb9P(`dI%9Trs^$2uXiqvCy_^`J6;k255bPP3$BGmJWxqTG<qno>0D;vOFxo;_;GYn
zuzwn}fO6PT{-Vx*jD`$kNnMXt7=n0Are^~LPlFHmcLMF7WSw<wxnkeC6C#=bff$hp
z+S1Uc06bNa_>+hq;+vn>5aGxHAd&V(TPEp|TmkH0o7^;hxhdVNBZLx=6AK#Sij*Uw
z9*<EZ1}I5AT^MEQlOrB-WyU~EhPiC8D&!K{ylOK;f@2T9y$gyGA`w$lq-l<-xTWUl
z^&jC#rj$7)2yzxg%3nz)g2i*SkUyx*DT_%;u!><(m?je%3MpyVtWJt+;cKJy2be0*
zo37yBjI`XZ4(M$>NUN<pGMQi=Ihrrca41VC;%=;$K6sBLyRtdq3&j4;ZM{)L*jzO1
z{26X|cC+p9?5WP!@Y6z#dkGq^1+Ls|pkp6y_^8x=!~UZ+%2kHBrV`NqqRdGdgjqUA
z3Hi#!w(6$3*!)xWN)RJJ3j16P8@lF8)$*h4k|gKA5R57GLU-;R5{(J9CYlIqKkqmI
zdQ}~-Fr)}^*3yzkM4MbH6$ZA<gN@kl0FjQcej_l7-u6ya3`r`X#5zBe7_GX$P#M@v
z?v*>`DeBBd<hiWH!g7h3jxAF}Ab?t+0?)}7Y&?-F^zRHmzpzn0FH;67psDE;Yc$?A
znHz$UM+?YS40j0GIzYz63=y~`g2AjwmX|iCsbRFFpwZ<L7)6mbGcZEdnGn#tjV*{@
zA;U?<WQMGza4lk{=fh#!Q>4qjiSDfu%4Pdy39UjzVE4Y&S6pU6h$3$2aST+<Yu%uN
zkdT+;<Mmo%iNHAUP~|C#Gtp<z_mfspv|f;1X%T&SDnmNS3`6BdRk?@|B<*ofh#0RY
z^U;$|h#8B0_I;2gHiH6L&rgtun=ilou~k5D5I2QEl6X`UFA0S8S@k`9?<%`Az4WuV
znSjLzVtxH-y>!%dwsEi8X!)@PT-!Wq-L~9QYz}Y&9riiA{Z{2_81b`W%PF9<Gw|-F
z&Ie{B5+HYXUMRpU_D|`WVV^cV?i{Ij(+-u>2?^CkyEcl}JQ;S}p^dWP%*$+=)aR-X
zx#8ryOr+HiI(DGd5VxoUZzV)-L|k_pZ;LJk4f{3~dE7_F-tYfT5W5MygKU&Ylfuso
zsiKEJ)NFLagB=hv1`k9TiSGCH6?3z-xHvu>60EOasLs$CzghTedudP*ATt;L?-wSL
zIyA<pq~!xcCWhkYoM0zR1}p2Lis}f9L^%0;dnXYga}68?*eHj;ndBl_u~en2Kxn3+
z2gp}@95~dcI*Zrgo6z91`*T^N7R{M_c~%ul7-t7W?bl9yswL5<qVRMfI74agSfuB6
zAi-dT9e#O+34Qu)e&YkvvBb?v{@3>3GTqrW*8s8In4um%!%`TaPb(l_EYSHJhZrrG
z^5<V}x?B;M?~!Y{gX{r!fxxAt32gSip6LNr=hBhmy1)vWx#}UJjr33qnn7R+Gc?hu
zjvZg&WK#@i$e_GYPl4_Y?*XNMrx_@)zpu{FGx)W$eK@h%;uy5B(isVV;^$~GYiIa=
zA{s%z%(;<*06%vUYWhx;)(&YKcV~W((VCEZ0T1@L3bO>uBgvy6sgT0ojl%gF;R}fj
zf56SXprQ|wpro<;uY*M8N+Y+7t@Hu5gs+i!r#{5F6j)&@(C}lj{ID3pI2^wO%>XvP
ze|QjKlOb;VYY|{m)AIi*`xlb}LIA#mv@WLKi;7nSyJUa)OE`QzXUTG!O}%sx_<fd$
z=XfJTHBqpf<2dOSFrX+{IO~^y@uQN+fuR!u<^$S-HpY${(aPyHZMw#&pPU;R>CM@e
z0LqB)Xkd%?b_#qqx#M0;Y~$RtsVSJa6QTD^?+CdO_St^u!)JM6W|c60`ZQ(vO{&4d
z1@V+3D8o%0kis5&A;Z)F2VI{Fwwg9eAlFK@Q(bd9Yd4TYZ)tHgL!qpK^Ln!SpFoD!
zt5Y=ezP=`x&U{YA``|E6?EtFLG65CZ?-%+n);r2C{BJI7U0^S=0J0P|NrZvI_V1n<
zvZ)deD$NC)t^6$Zz2W|}{)`%iYrH8f9}6??=6r(I=@5aV&FB;88;z^>HQ#mLrD@p-
zO1DuA@`5kz&$L*#4x00EfndaApaC={aTRK-qxLutH*9Q(vP|NFm>O80HthO(Y1VSd
z0PQu2@>8qhRriKAv+}f14#?cxuKd)f(sq$H&6^QCM#~1@%k;4trVm~4{UiH1o(~4}
z@Z);xjIjoUMscb<+9#qsxnL1W68?NsFE_hsJ}Oj)dhDz-TLTY(c;Na5a;RTTgNyog
z+h&@qI^9=><6YBla~0?j)ASEKZ<+P6JXwcdGE{Xz3q`v+=*#rjZcI2xspjJn+tZN1
zN3_R0F?*H{-&+%G2xvzAOMXsMg(%1*`F}*>k}q2{*`6H?k}*Y!V{^S-RM1-Oz{yR$
zI0!`rMCXwyABiwF+OcSO-9U#q?e=fis%&Dzn40jWj7H*0$y^1zj2-=Jo<pWc^ym<)
zwv$hvZ7)Pkkdc*Kwb9*NmM<fV9m0{gjpa<t%Mp-U?nAsYL!2M21V8U{34wz?=K@5Q
z?tcW=8TpfECl&wH=`4kq4Ei*_<M6%v<zBJDZpEDqR_v>z^hrb-q+?9_aO6{&_!eOD
zBmWpUC`9i;2nXTuKe$UpJM-GePrVgY(IXw}DB8JgqMw=B^scVy=}Sd(z%Ac6ZEvj_
z@MF`b@5Edr_u^0!LLrO+V8yD`4CSM`G7$3o)A-FTyfbLw&_M#|7BM9C8iG8x#$8(8
zd?alTiTxSqm-3PXSUmItocCJA{;vSY{+~{e|JS?vKZr+c<R)zg7!bNnsb9>4uBx)v
zB1k<U6oCX;^33gM!4>i=f&n6+zjz_R?iJ)4JDIteA$t88ph8iCO>hE|W-Otb0>%O(
zs)yD9zWdM@#U4Rr2bA)?63Z!4<SJK=DketPvm!&GZQ?DGwK1ZqvSyfahN3GxkILVx
zD#qh9l;__X^QdmWfV)^dN7>XlB}g^Xocj-QXq<_Z&FpB~>%yx}4cfVU3;UCHgTtJ>
zlzd@$oijS01Yh)cYV_b8@Ae&hIDTwix!>)c0|TKEhU6Ik^Bet_N}iMTe<o&@HMD9E
zTM&Kudxhed>P{XCtUj0i-4+62=WPb{WPyZOFQy`BI6K#`em`gL$2SzYU)vyAZ=@MH
zx?}qi&%7ptfrt5n?{?n}-?G=z-}X+65-5?0;(*qE3XlpWAJfb%?zy0Mg(%f6$JP2_
zhk@8x^%G<IQhc;*T-<D1(i-9YIFSZJJh!k{vws)cL{-+JOvggiQWHMfb1ZiZdhGgP
zmlK$I5Fu;pFwhxb;)M$g(*Uj(a2QB}&EWL^U2V9~&kzaTY~XnA7>0tEzm_V>=Tzg~
z=8P)GGIz~}HoJyhmUhYq!sM7Tl_}LZEwq5E@jJIZEvTN_OJkGOQ-P&{`w1$X(dna&
zOqW7!-_Vm^!CZTQ=Q$A%zDxo>w06)eRy>;*ETajOZu1m@B12Fa&NgjQ{<kbyyzBQQ
z${dFx`lpU3meg=Es98mZk`wGj*PIi+m$JiBx)D!SFFUG-X*x;2|HflzL1N)%K66>K
zKlG$tP+D(8);vZP(^kJ3HBYM9R%=Q0l8Y$#*4%*bfgmUa*+k8CQi>4p-nxfP<fpz-
z$&?1|+<WQJ1x18y@|0LpXGtC{(kS)Ox_wS_d`j$XKJu?IktWP`2C+A6Zceb!5zXQF
zx2O&TT}Ms12pT73<ae1_g6coL<CJ+x%qNE<-ikd0!yl$FkBgw4CQw!$_$+^A+Cp=_
zW^OciH*%)%Y?a4=xU>mkGBjXKr}x#gDzd`lV+Qm34NZr47XR|VFIId>$vw$84sYhk
z0PE&hq8m9>BJHK-c4s)xOtafz+C!P}@Dc{EGhETVmlUYZi?U2OIaftBUK0GA%_Cr1
zx_(tQSg{K>kV7d_{>&QIsRd=wM{T`~eT9X#9-5EizQZ}rO)>h%QYobqwys4j8Q?e3
z8#K)MY5#KhP8^L6AvGKw3p-7A7=C9O4tegWVzz70tmE}8D;1ET(MQZx3e=Sjj!o?v
zV-60j<vIFlSlHp?QfQB^!ql1z@QJLJZ6dIf1`gypR|c3d8?vDaiziPcRHj&KVQ|aP
zG^xL1Mvkra#)bSA5=hJNS(Buq(Hz&XzI;buDv;KfR;L&N$CM<s`tO@~fwQbWUpH#8
zAD5Stv-F2GD2G6B@^tE{BmFOLfvzcifBji%_tH%i=PvYrA71q?i_!$rQ8<z~zamg!
zGc@?PJXrX5Ah;RKK*S~shPIs0V0U|d=lkEhVF|WDI98L!lcVQ#zuqU!fC9z&hLC$R
zix-RSdR<ODKD>E8?IPxt9l|BqW|H~qHZ0p%Lbq0)RsS^~vuF@J7ii0%o+zbIFVyCi
z&EGF;n%b{b=3bE`jLyE~T>*nG=%H7*<C5E)Oc6jTOkj1E<o|e-dE+@rlFQqQV?2>0
z$!*1QS{*IBPgKt7%X#vYq;?xj9!AdTw&ToUu)CkZT5m;hlH(~Zu2%&!-{{C6AV?vb
zRw&&fMn&%l3`HP=cxu5wbUjdt^#}6`I?VAdm}DSfL1cA6>Tf3b3I=6*a<DbJ0-=c?
z&eY{|*j5YzDB%D|kT4j!R1vr>kLh8$$c+@I#*WgYlpXI0?YkuG9uj#foIgQ26-?(V
zY;#|go%jqtSVtnlRN;hm)XF`lAlZOXB>4mb*&M2c7S_J@rRMQ=BT(3fAy*}Sg_>}Q
zxwBhWuYpMmW@lBA2@u0U&G~VLT!zmC!r*_4VkhGKD5)4=IdNg`vC5hXJGXX008V8U
zalRfH-J~6xh@ZTmG;rq+9MU?15_aO+Epc0p=yD<*&9FBO_m?A<xfCl)Ohf>1JDZm~
z=g;zj+W>X4sbrPGvB&#<MEym43=9(m3*IvsymsZ_)F+Uw@Ux;vKa`Y{CYJDux23H2
z&7(s?s@t<a1fUw~BHs%_?SW_d?Eb4SFhIO}`N>ViY(y006ZQ*RfNoDL>nHg{Fi#TR
z$e(@&>YhaL9Dc^%McncE4dHhff&4)24b(=;A?sT5_n3HrSoi-S?H_`3{l2JOICipQ
z+s2NyW81dz#I|kQwrv|bwrxB4_WRXabxz~g_*b2(yR%w%Z_T;p9OD}K<Z8d#l=tEt
zk$iq~yf(V<@Th*q?U`+krq`+n?DF>>W%FM?EVcTH8f+1ietL{;r4U!Pi_^Vc;gWt>
zgXBoZ)c75yx_@GCz5OFTU=U@y>Yi$Y_pj)`+d;p?h=IWFfdY5blpfsrJ9tbL*o?L)
zA-gV<F(f-JV^P~(K?;auLm_?~V1~CX>?v)el%jPij;+rdt>#_u51BVlpCwfj`sLer
zI|Pqdnh1Fx%S20Ipp^scurr*vICmGTJUzzZBM5j=`fJ>g<tQ5CJ3iDMJ!2A}wfPIq
z%3Pb)i>q3Wgi`{c(ZUzD@F%M|*VUj;n$mR_`{z_c8*-t-?Ts))j;g<6{3bXiyJUUo
z=u+&D>w)3d7uA0I;}WB^$I%%stkhsN5kwfC8_g6skniKuunWL>pGrDprv8IVOI~%C
zkC-KL6auQhd<d0|(h{SF4>l^dT7$U|*aN*4B?m0!LMpM!=FBU1@-4cQ{GVJ(10uSi
z1ws!Ff3R9PjbdH*`$16rg8?BK=1K}Y64XlGVYb=2eLFRF4>Xy-x(Cgk)tQ^T-y%~p
zh8McU$s!Gq+BO7Zq&87}t4zE3{l6f+1@vJ4v*7<1jh^xM|13M&Qr~g>KL!6qufJ7M
z3AW!B#Bl-g<y>mRVtuHTZNc7#t4UbGn(8kS)yLzGnLjvHLW1+jYI^5cpo{}BtqaDE
zPbXruDpGV$_v^vsex|FUaEki0GmDm}pqQkkU~z<L<SE;+r0k!m@AqkLQGl&xL3jWa
zDn<9#!_rIRMdo-C4Ys2wxl*yk=hhNw{N3d-gKh1O-u`ND&CU$@)X-0x1(_N<yBf-f
z`sa=8{R`jT<Yn&8&((YP##aD@r`X6ppTn5N5OIE33$5Z5A~}F%!%w5pnJpcL<e+OR
z#bzUy^5}twX0LW6et1R>$zCa+knZKRL1I!zbN6GJZsy$CBR7=>RYl5crm{go;HO2D
z*ImA=(!d$XX-T9<lYJ)9e;m=WBFw*(-BlNGt+Z(gjKsKzGmL`6SF-@I(@-#TI}JEI
zuA;V0L{TxpgqcM(>$3ez*!Hqsxx}`$J13_+Z|a*nJHPCL#1lb%d5Zh=VZ8tzMDJ#A
zC!i^SdeR@<ly(!*H2yENV9PIP@2nr1Nwi9F6OSPJpP7kB*bke;tBtTD`?n-8G|C8|
ztCQikFhm+0<f(x>!mlV1MQB41+?tjRr<)i^CxD_W%VyI_sG^;1<eVRp?*O5&a9`vG
z8Y6X4^o@&5oO@WmH4C)!%{fG{>zaN_^o0}3F+(LlUJvz8QIh(`uS#QKGH6Zw%YZKA
zsL`F`TUOUj`{BMW@E#uHhm!SW%={TT_Jkw-KJ*<VJ%m*o&GwB=la0J`7$sbU$jbx$
z5i$hM-dF|Nb~0_)rlmJfzxY!Y_6n(ZkvOBGO>us19~Kn!Rw+hVn-(E+QPm+VeU&T*
zg{*y^M=X(xCj6+SI1C_6QU+5JP)2)E`d_Euo!}}$A6hrFg;H9ENw_smF?2D111$`-
z1RYM=UaT?N5CV@7yAkHD2~R_Xyky__$(Txvc1x?#q30)B9y*Ggw&h*K(ZPG5t@LXI
zw0yHDp<Ejwq=0TM#_;Ij_rZ*T?NVQ#()La_6XBueo<I7sV>r8B|4g@57nt~#jFNgK
z`voGT+-}*p8XC-AAdv>o3yX)>fWdV>D*X02L9@g+BT#DHT&rcxJBCwC3wS)VEMvR;
zCWsoWezn7A)~2ScG%PdHL5u8QHrsYX4~aF678t*Mh)cKyzLwTXO}~E25*GwT*YYiQ
zfTY?8RPGL3^yezV&MJ!G)d=be>gSoFr}^}8T0A=e2bPwNLC}0-3DUS;l?URHKXDE`
zf)Tw%H8A9q#m>i)P^XM^Uo)mDVpwm$ds^SKkC@S35r|>*|MHIu5a$rT{6#+UTCva5
zG?F?>T7!Qd?6_q}2gd0Ex3g{JiM@X1MEG`jg7=7tMgOcD@JK}G(iVSZd{H$nC^0ip
zwH|CR8<!1P^~<UsY7R0ZmOo`2c+3pZ8rW$W0gYwGRjdxtki!!`*#)-V3D+KbBAaST
zJT7*TNIs#@1lfu>;<gNBDCmqaK8kT(mW3%n{4cjxYlpUzz*!-<$3IAvB9_akQiO#)
z5(~qODHo{eie*_7Y|?Cw*3&<(7nD{W(AS>y<vLy)5~=?>(<;%gsb}L$>D(XuPB;KJ
z<QgpGl5j=~xq*&B&(l^%&k#mB@dlx#0jVuc_~1~e^?IgYJd0rBKQj8Q{`yGX<fP3Y
zS74(~)@p#D$@l6RfpHHpbSRL9gnF11<t;$;1K}P~RWeT~x&%?&h7-NbzEb@r|Lyhl
zOc&|V;YyR<X9}LUUzN;0>GTi!S{H}YIr045^}#d{MEx4*+E2rm(UDoU2(D{bAii^5
zx7a4dIrJhayceH(YgI&nhj68`SF4zm;wKMcBFg*s^4dk5i?%?15Arq}=qcm#lRCY(
zZFyo_cYV^XOEg^tn~-U+MyRNq(G2?rv~(JK57na1=cb*CA<`m+vzBiq&JrxWb|Tpa
z5x+CSg?P+sJO((+38WamliJ%~C|h0%vEdS1rl=4pn0jb6YHYsN%G2`vOk3Zhii@@3
zS&Zz#fkx=i8RP`&3|smvZFBGknITdn>hKngrmok$6>O|)y(GrH1h{pv0x)`Pahm}-
zC?Vrc6#q%w1qjCmP{$j?6PypkQ90WHK%e^X5%cnkxkuG8TUkD$=%*QG1(x_69FuNz
z(#=w5WP#^5Fahr`KB;D|=h+MQbnnzz^{du{p&oTel4DBQG1Gu?d`0bj@}A!wd7sy)
za@54qKkxU)WNR;nH{$$7frfq7AcNgI@!DJPB9+}~P2G}ZtNEuK<N3ZJHFgNDPwOan
zaQ_mlS!>TX?IqfHmt;BsorEY1^;Da?7!T21fDhh<-nQLgE8);7zh4bsDpYh@#9;fG
z><1sm{ba8L=i4|o2+1tnmG}cR4Noil0n+w%&Qt!{C||==aqT-$$U|35?nDM~uOO!f
zSmCm(9E00lfAAE|GdS$qPp3a*+>o-@dxu26lOk)=K9nSjSie)nWrh0R?G4>=>$Z>N
zG#6FeDlfdyitHMe%!-xQrOIjx{rwj6MPnT){2_&pr6_(qKK7J<{oD}b07P!oY8Che
z2@<xacXp@k(vAszkp_dLeM}khWXbQO00#_Q;Ll)Myc?YM+nje~SqHU8o8hvzG#uP_
zn%7LkC{{iwzlJM~5%`LaUS?e%PEmh8u$`SggODF!D-%iX|EqW4_%E>$3+MklEcpNP
z$T$#uZ}j$B9sNa<jo)BHy-)E?pP%hIxv!q>IGi9MR}_$q@VJe*ryqOy6pMQQHZ^9a
z56(zQFCdF3l0Fp@cQ;dX6U_AV@^*Sv8lM*vM>`#E)pQZw3!*?<2;R7yZM|ti)6HNQ
zkbZ6Wc7L2?J_IB9f6jKY_vZBWevKY!p-7u$%8xOE?6Gjd+0UgPKJkzhK}dQ`zzQN!
z9R*V(O%!ckp^xY_pG5YV_im_@g7D3*^JuM;)kj)d@S!hd^sL3Mxu|KrFWQ_7c>+73
zzBD%PWL2)VUiDm_cI(DZNEYjYaESq*SE&<aQzMPr)hM{+!a5Jkl@sk2urNKigKH(I
z6j{H;C3~t~uwzXKs@wZJm?iYR{6>Z5$&PE;Ubi`bTh71l%*gUxMSTlVm$j67Wb-85
z-~QWH1&DMuNv+H2;76}$EMkI`n#DyEHf*--vxT2_iXrz=m~Y^B4#X~NMYX_=iO6I@
z%wH~YL&%Zy%5+XislN0-PAMEXgrr$M3_WoX!J<PhJh``5d(8ImH8d)*EGI=0@G9D{
z(Mze~@4gAMLkfKVA}_YlJYN560~lsN<X*^M_pN*5<SN`#TmOf9U~}2e0&5T_F9MQ^
z^Rar!covdfN$5(!6?5@~a`OLpB{e3DR6I%-k5nFR>Z7|Rc_$#J$j?>4-i^HL6q3>I
z93V2c)m^3bC8eiaM3>qy)&WMnQ=Jjciqve9hl^tS-T?UwQb@+`rS&;!7B?caL?BOC
zGZ6u=l^fg&LUy3nALY5Wz7QtwLWdF00=Aafzagxs$0xiZW)3HEsq~WPxVs@Z3m<Vx
zccg|RW~f?aRk7KUq%uXhER>>5Yi!nFD5;S|Ps+aapzXg9Aa)wS>a8SFnL8#7`)t0_
zEb#D-hiJ8VN7@=mhRD13co7$wuBStqteUL59*O^{j6|WeF_`GdfuGKRP};d03(m?N
zADZVvwZO$MrbtFl3<$p7>yjmff2(8g6Ze+XfbIyWRo`L~G~lR^LJfwUpp_!l<``@|
zUDv4l$1{czEYIyn`zl}%Jis1(D*{0hiKx<8GSMJ_NWS-2lft|=%XF5xiyuZPxrFIF
zj7Jx0!ZoI(Ix<$~mXqR0n24ewTVfU{#dX7!<+*uXtLi@av-NqK-nhM)<IRD?gADZ#
zu(G!qGRU)&L&O(mH;KZLDx61LTfcpN`R<r}2&xT(2BcSdivq!7G17h}FkF>3m;!D=
zn-C-bZajgDu}Mcv#RPMAhv%sjfcBV3scAXVj&+RZAI1GsOS`u)F%T8}cb8Bo!@dZ2
zywfvS6p?3ifkD>_acHML>g6!=V_W2!DCpHoiE1r2rJwmNR*I)NF|gU!xTkS{<Efq7
zo{HrpMvUDAql=8AVh3WVhq!MuzH2iIbW*2hX++vV$jAgi<ySfbgT(ockNXP=zxDQG
zx#8$|^C{WfH3aF&xAXT0@(XYF*EchYhJR_!<k5;Jo4*6*=xB`2vifZN`&+6Ou>(is
zip%P0If$MrJ;M$N8&3UM0ITLHzqxU~u3D###v$l##sJ>{KLwKw@(i#(5OMTA%H1=M
ziqEe+L1tk2+Xkx$z%nYrM&CtGb0shSqCxEA06rN=Y2X1q4kA~@0n*c6UbJ_S0(U6V
zIV69~(3trU(E;$HAZC!x8p1o1HC#?WKYzdH1SOU*kkiYG?2|`tp_-EnV(9k>@ghj@
zd2_XOw{5tNlgq~6u@-Xl-KrRlA_m18rGeO0Dk38wi@r(qzIu>df68beZHz1EE#Sro
z%nbfgPl?l6ao^e5Vv!=9b_jVS{R4&>Vz(9@$DG%Xt_{Z($;+r*$VhX`!?L#^<23&J
zcUD9Jaw;O)PcFhL3@-X62#_T^_C2tzuPLf4!@sO`Dx*)e^%~=(kHT|^TfsLUsz+}%
z8ldJ1qSGnVt1OloPe9iGOPL~^d)Qx4h_nOGca1L9Od*%hU{yvW%46A25Zz_@(e_si
zJV>3jL`C|eOkJSkK&$B?wTqmC;tlbqUE9~^FnuH?p;4ce$O8=l+CFu0dXnJXj4L(=
zK|Z{KOWfOB=budmLomXb0J-9@n52u4aUC?$%ViaWG}K~5K#)EO$m2ZJDT(MF5}Q0c
zp`+G42ZD%LD}!!qn!Mk_2{qgvV{8Hw2gINChPmI)zfCEYNPUhK|6s}ZBaN_r$i}#p
zurC9%?DVhV7?DYZ9OGD&`HL5!`t{ib&mnDU{KYXLC&lf3+4V&_@B88Y_ha5V6dF|)
za*}aJ0ZjIfnR`ohfX36BF3l>~boX6@-PikEk6{;P+Ia-?d}zUj9rN?lZci^n#sHSq
zpYvUh?00hz(a!OZiT%Ts=^G4hiWdyi?TwbyKbZOCW=d)1a+BhZ)9d+Eo5gYde|4Cc
z-XV6%^hrc23p=$8H&=`erd&F!B0s*Fa-L&ngpbkfL}|AG@FiASZOu8yDU$GhO7w^i
z%?EvCs5~wj^N2?^lA1R8ds{~<9Cnl_iE^@udk}D=9Xgm4@ZRT%60+;7Yq7!6+EvJ1
zfrmvrk1u>!pfJ}{^<0-H*);LHS>a~nSH8Br7scVMS7i?xW}ni#27_JWE}NtTXeMaI
zb7K$jKi$J!Mn%M1OC)$blisp6DIQ&z2^ls}`n!Pu&A+&5@){sfK$JbG8;rZ}Aw76F
z)Ni`Ka*R*{z!|er()*V*F`?4ZXVh4PXm&v{We@|Z<AM}dFiLJ@B~=8;21VYS8OYNj
zB!Zzs?;(H$vyqK~T*1EbGR@`;Z9#(Eh^IG$B2k<dhLtfvJJ`gSo)SU}pp<`|pw#Mx
z0Nax$82<^RH`m|Dv{6!X8%Q4NvvxOd4xS7>(_WpPoUG(Pt(TCasoS|a6~v9uQZmls
ziAxCR8;M2pEwUnaWk2h)FRJ6OPdXF9MwH4~SoxHo-}R*^>0QLOBHkA45C4kR{#{%i
z_H>7)6xZg<|9wb9UaXW~n9AVbV^}nA6jfR4ntLCP=DUwpYewT^{_Pq4dS>YnejCvo
zsHWlR*!4V}*xXySm3_1(2*M*(wlU1dCqM6-8@gHoD8ak7)8f{iHpPAU;?n>CpLqDw
zB^~`pz5fh^&}{FazwT#>CyGa43f@!7CHYgM{H(cyObF+HN2L1YMt#;glW~kZxG}qW
z<1YT(#Ho6Vj~)t!cFM8HY7WjDhM;^OAO4K3?;fGWwzon#%7#bevx>*`4M@w<Cb-2h
zb5fdRYo8a`DvSF{<~!frMu`FsxmxN5;LXvIH!fR8Y5NFnfBwSHU1#~vX7IoKYron4
zXXSaax>Vx7K~m4nKmS_!eVEj3^t1&}jzm_(LAAzOc~T;%hye03DiKOi(?qQ+=oX<s
zbe5znm;mDG$%`|I<|W4b{_IVUPVdKU0y!-HJ<klr#TM~6hDJuHLI&OP#xR3G*3qPI
z?a%vOu*jyC>Y)L|zoPs2m^vzVak6qmJySqvy62F5q;NZU&>$Z8bnzlLn@e!JF*_@F
zK}Hiuptm)3aB*>vz=ZzknDzGI%Ax-AcxF4BJAHOToUp>n*JanDPkX3=j$^M%I!jG2
zsOx~79W24|;e9t5A)<iVHCAgF8EXDUkZ-Tv%tEnI{Yt&T+f$xj+K-Tn!`h@3t(v|T
z6ReOlATS@G!7WNl+hAD#Cx9{cqr%%3L2`0e{VBUD*r#u$vCMO+o&6sKqobzdZ$w4n
zQVG(!HxnD`t08J0A3fy!ei>ibMRv260y|56wk)Vf)#)3lhvq#a_*T`a7=>tM6qzEv
z^`_5C3)e)4Wd@B;Uw5@QBAA3S1cQ*)i(>OD-mjfsV)M^MQ>uFu>s@dNOB%aLPw?p`
z%yVmb8R)Xdt0?W1g6NK@YwoPmUZ98ly#<t_D}xwjHB*=2f!P!Qbn*28%2OZW7q_3q
ziYuD62%GOjYu2`|R28`v@xW2hN=O<HFopB*oKk~v2UI+7Wh^Qd%cy5|e=!&T>Lb-%
z?ZU<I<#_WEVbEZf_87>ge9M=|NUp-FEGNHmO*#!mCr!6_n=K6pp^cbw`TVWyLk)Fd
z>v&W9G=GY!e{nE?=9>zV;)jvq*s^ob7!P+;8=;R|rxG~Qw^>E0IsyQhr!Qy7x^?b&
zchXwX%rR_{`YG*fqy#XDTSk$=l>E1^zO*_Eqg;souGDZ=f|(si_7XuwPRa$AWgX6<
zTc1QA#I@Y!rJ!dbd0946ru6{Gd_U<gWii;&DH9)Tp!4J`{!-P^g*om{LG*HH0z3I%
zc=$_-&#aa=bsz;*z1x1s94<7Y!~5E-fvw$(^ud|iVrh5MkgQWE9Qy{U$cxI;y3P%J
zxKwjg4LJ3VYjWla>;=TqSh7BhgZz1t`jmMDnI5=`wf(X5O%y!GN)Vkdfn~nkYfi#R
z7O?!m?tcy=gD?!TP@vYvkUV4sTBp!}&BzQANJ-3!UpOy1D@s>SYboyZu%s@dzR9sH
zZ&kd%Kr1FnSjO7jQ4MmEC_mQE;00(pjw3v^7(uSzJwQ3&F>BAoOH3Kqc7ZU)&q+<q
zkN223UfnE%h(PO|=R{pFw(NhuXf?_GrIp@KB1DlHgtfws9>jy_kN>0Yt$$1V!Zv8|
z`Z&$bVSZRr*>Qm^iHpKuF%oE7`wBe<r4q_?$I_E`&1h#VyYRTFthwx(hmQ;Y7GbDm
zKu1{A=W}c^erUaq=m++Y12s&GWEa8au&!pvfI0qL<wJH-ni#mNR>ynElD;GD+%{#m
z{`%|fhAr&?3WS7!8D&%Qk26bneYd)Z0f@*-M6v;#zcJ|ByfQ<EpH{T~2_lv#Dg(bX
z!cOTNBM~nw{DVrTI{T0&_o?eE*EL`>0*&S%5}EjqoOMBblzmzqs>$Nl^MTmo`4_4^
zFr-Fy>KgH}0;c^vzI<E@k|uwN2~sixCy}y%u?}YJHAM8C4j2YB!r8A^L!-Oy9*Sy5
zp{!IQQ2um0z|Y6)-N}!@Jt;UrK#wnJ2RvllI&#9n=qiaD;Rq&mWdWgNJ}FSLOnPn|
zt`eCPd8mRqnR>q^?hB+uA?zY))NnuO*Rg=wZ-`ERMjImQc0&P~Bv4qkf*Ob_rG&n@
z*gSxhVjt`-$Ls(Lh2<JluEs~cm5x53<~uc;P_Sz2*Ai+~-$Qj}n(Vy`D2;3w0<e}G
zLilm}-5?m&F~b<~=4XMb7G<=btBMvYJw+A#4wrij%|GpriER73H@7`*e#?5qh)JP}
z*}yIqq1x+eIP4dtmdNnFR_fwPJfDmJSD-uOp?u}|)fhg`yk3y`4Uh+%z7f@Y0DKkw
z<UC>A??jo2a%O5BX9e_Y3}L=PO7^d~CNeV(c_KL2CPcBU1R%OR4}s?Tvs&{?{8Vfe
z?ycVN3MlO}H-{Ze$_z=VSu!{e${fY@F7ef~{=&KWFU+aLGxi2okysaaKaPrPu1J2N
zc-w(tu7#k;X=(GkUp0OXD&><oILYb;Wx2?~9%4#nmCwA%o9l$rU^7FYnxAaa;rnrI
z_xZlGKfQ1ry#nV~>VKnxr(WTk3Pof^h%?<+>!Ly>B-@@f_(?T?Q=#lki2O+;(A~qd
z80Z+wLS(cMLeFWo3^D5DQq@DPR$!AQ97q*0q>;=KGTKNN=K9;?WNC6Cy*h*qD%U~b
zqS@!kNuzfUMK;EWdOud^aQ`n79nw6B&;iM_jBuu$eX1@c`9lESqlau5F)(G55@=3t
zKEAUj#Aq^6>s%$<LiYxh5iX-Rptz>@7MTFr<OEYBJ8#<26(t0sbumy5%0|5b)SA$Z
z@K+zL2I(+$j1d~7tr4*skCu>@M5csyEs3E$ZbH{+|Gq!C@Q<#r7jHVI;|l`)=me0Z
zn*1rpzM4rU%DfOD^lMSZ-ZOses`x38(S-#(lbPhD$dG@G4ETAWG?(!o8pKAj%qBud
zbx3I|N!l4@s^$3`T-3;k<L_|G;)X%NR2njyzYjXkQsKL4HC@4tdO{{wpTMJWuOYtE
z2vYS;EuTWUfpM=mK#7cQgqxbOL4{cqx~_so#^s;pwvrJyxU^srn5RpXO=O2q3mT1I
z^lSBJ`V&Je$!zq8qPQ8XpznUv-{+5Q(LIP{%fOSuUz<JcW(78W{lh*62<wIyT2^tk
zEsG%^R=6MNE1or#5+Qyx!wFkH;ai?0xA$x5V5wTy7)&CE4|H>ddRq8NS}e4Mu?Cew
z8EOZ7Tw%uo2}!W)cntPb*g2OpDex7Gqq8<70bB;#I}B4qWsVnHSrkg5JN+BW{$5CE
z=85-gOKj)#Ye#QT88w-u@ET`AoC@6_*PHCzl^Y*OSz$(z|7w^1P01JuZ4E79cz9qK
z{x5lq>G!{h+W(jWeWfm?bnq{&@I>wOR2eD1^=b`;JFZ+_#3gmYphO*kTp?_2)f5ku
zxw-krTZ<tK9wmnrR7;z-9N^sAV^hf&o~I-{SGrxBxcwZLfLhF<BAq{uYad)Bp9U39
z&g_KB6ajwytB831T9yxeeu&5A9h6*x<?a3>=i(~LRmY5Fd+UP-+t;D6d(ZCCzKQ+&
zwb++uZ3iEw$3{|Ix(Gxriwr9>)?<iU%nqBGN};t{ZvI6V?~n8HcF~J#Y=A}p<vuQo
zWc)4+tjK)RDHf+1a*v!mvQaT=1#8C3w_8Qxic}HqA#HHzp^c)H1E!&{QWB6|L@1_$
zzX~$<VMJB=LoMAczlT%Z5&@-P!Ck#oquwUPs(wuSU6{OC6B%&fh7kEshDZQJl7Vb6
zb6a;b(>2(|ce0;ihrn4^g%)*ZWr%bV;qBTzMSRiYg?pDp1!GiqMb{;5FybLW_-H8^
zXV+Ec{AjIr;g-A?W?<){$X+f5H!osb8qZzXU1@<T4{UR$N)Mv-{zn<!rs&mO`Gvex
zC-miF21`q}#?iWp%Bg95YN?G|GCS<g11{RPD|M{xv5lY)KXi=A=xwiQglF-!+r!jX
zQDeVi-yej`jG+l@OO}4rmS_;#>qX~h0C=S#ad^ewCi_L&CJg>smj)5lnNPDy5$F0W
z&b#nurSnV06&ha0+(TWZt<t<BOL=26l`Ch>HR+H1@sU5#4psGPy;G_(bH5i{q=Z&B
z8UzExv-ta}KOA3GYq1kav})FkX;w6U<vCmV&va?hSJL39FfZyg?t7KrIU@Sj{~n|Y
zW({A_Fm;3JRoO>oVLodpALDV-5)g9{t!^|C+lovr@BCTJRji6&SE@qLRj8-lk*>$?
z#i%?%r)g2GRE$uc0A{>)`3PT|beANefvS{A-ldA9SJl;=-U2Q}`CS_~fQ%}&jtxv2
zI<4yNHF{<|O+bxh%2>_Ht1M{rRm3K+x(if@3>)UT;_nH^4BU2yfb71eJY$_7S@FFb
z3aSm3js%PU1rf}lMNjk^K$g$wxw8yQB}HjU><n{tXHX$@_XyaFXs*rkuz`U1u&9|N
zUtFP8UdNHDLDsVL&tBW`SAi@24ej5Io*X5(vG_&WF=vy<&|IG%w8c|3M&;@q)(IHI
z27;_{V}aXwSG0u^#DxMroiHnT<$$l7T6IKzC~?}a6#1j$=3Ak4D=|j#YJUX+@^we^
z;-{?=sG)o6kgQhBiY9NbbI~f$9{FXJN18O0Hn`UyBteNArX=diMq!}6D-0kdBkBFP
z;>$?lxux{J|2d?5%iN<ZOg*8L6TP~ewoCd-6{7R*?Q;M3(WJ6dh_&zZ&3b*1yxs*;
z+z|*?ph+r9EBbo=UKdXb9{O!s4ZUVsLCqOcZarP+MQ@sdEN~!MX<`^|W=Kd^Np9u$
zFHiFZCJ9sv(C*0KULpa~E~M2s^KhoOU3p$$kLN4b9N#ktOh^vkiSR7z4EPak0<JIm
zMGInUv&_U)#ERq(9oQDAoxRq{k;smC|45!!KVUyFP``^1GgEkuD+`9M&+SBt&2<r!
z4wHzWYF)d<6Md3Xe1EDamsz3vCg-JP$-D5ZI|r5dJh?y6V0M~)T5>m2%VbVgC;ttm
zAHy8-HS_3{UQNI?aJG-IeaYLfjlxfF1roRoW0uHYGKhYnMcDNN>=plY`JZ>{f8?Vu
zG5!9Z_Yqg>Tecf)h`#?v%djRS@*tJeWcW+t*jfXp*0BpD&<Ztzc2v-sD)Ep*VE(6@
z0`ExO0S6<d-q>7^_FjCO-Pe_cBsD00N3$1BYYT-eT#T2F1xq%3aY!}ew@Kn&P}KT>
z3-`eu!4}F)S2Q4|gw5KZ?d<iXkbQ7F&j>oR8N3v4a=FUeWzk;QlY@izI=i>8xwZSi
zAhh#}4-)OjoRQ<)Zf4`Bit0<xp7HZz`eNIisCzi7C(0wHPQd0MQ#XNu4*r>U38&>7
zV2?NnMWLuT!l{s<bnHQ@-UQRJ$ad)(9-?xVz(TsPtoK3P&K56~2m8y%Iy}@e$sZc$
z1mB*BP*c_ZAi{E`Kkn7yd}qlEa%g@HQ$mj@Fp73som)or#&&>E?k}C`mwoA`!`)xT
zX4b5fV3jD61#Iv%7r7us98G3JdjT@FGN;x^T&m*ex|?MT^#@3GIUER<FGvk5(0<ZK
zZqvqLwayBv3@aGNLFx#c5=POfSgS7OLbw)Zn%D7Wn=Y)d+!=nIn4FI~S4!G5d4T7L
zl>;~7k$%E1yA-;?pU8cwteOI^<??%QNOB8CA<@XY6Os*<rQa>fhA;D&Pia<5ew<?n
zL}e+Wf=NV&v?={lQT(nz6^LegO=(?5(qIahC*o6PKHAY!Jr_sE<-xh7^K>z`11ymV
zuT1}-w!=jv<;eH$1!>3#CILp;QIM_rm*A!gz)0dgVtFUQ{O22jc@O4D(_c<LYHUI#
zG($uViMK5*^?q?MfV!qsoCK<b0OKC*A#tHb_&lPzyDhsB_Fe0Atn@BnHtV1B7iEca
zOI<Rsuu&aW*lJ6)FKpTaLWVfRcxI+PEO0dr$rK3PuM!JuKuvFWtqBM8qlq=VV}^Ks
z8C@QRrr+2?n<3gsdG1lky!P}b512MvWbzZnh(sbO!aTH33+&d{n~7UIcC+fdm#$fJ
z%`c@md{fS%h9JsZoC6Z0<+F9<)`p>y7Md_H$Ow=E7%NpDET*IUhh3aJO?~(oSl*-!
zh>8VzdLbq%9Z0K{Hdo|iS5P)}fre}ZJN<-1bQ@XnNy6Evg#rZ9;a^z4H^!*byhEnD
zW9$G@2iHQRJh2@)p`!gVGdUsv)Txp<p6T_|4rL%p9(w?4O@a0pq?OUm;s{NkYwO2~
z=uAY@!V+C=-YXaDS>j$7rOJ15Y$9R%-2#*%6OEK$_wA)BaNPhD@9tpU^{A-E_rSri
z%93VKdv7l!uZ&LnQQ!WEOe3$C(JEe@(b_1h%8K&~KG5S_Pr8`>X=$^tR_pY~hou<y
z)vb^PWzOfCHudLcTQ&*gH6kgRJe(xm^@U@SE&9s`F_%Pz!-I66a;I{AwJQWeMLI<;
zwsiYoxnGz2?oGls(GR`5GZ23ZYgO3~W?b~#jo&9noO_hjAv?sF-2G<E5Buzu&q2Ek
zZSc>Qb!xQ&PmNj)YraNT#BwCp`|YotV+`tHijN$?(tloA|CLa{$j<S<wt>U{aw-r0
zB@}%8%c<;`2mp=)k~m4Sj&5aZ&XK^SP|<31+?9_TjuN+&uxygG{PDy-g&_ESb;rZj
z9tH;Wcpy}$x8LvG7kD%K?d9s?PUnVhOuL{6>6%IS*FS>wpdRLWxqm|*reyU^(gJ_G
ztHa&V6Q68P?u4e??e!q@=KJyBmn&a(Z|BG4ku$DL`A~#?u|G7jLb{IANchc>GdX#s
zn$MdgIqfuy&3ihH&j#)uQQVH0(xc4Ae4f$k-$m1AdKx8H6QvXGtf>t)-&1rjlqg5&
z-?#i6D(*vy>URGOCl@8bTlVPqT!F^Vq0iby`D?tpp;>7s9JCo7vy<yhClaW-yV||1
zZIEgyc(MoDJ&sBGJvQ$Ejh%%Yh+HeBB&+o_wdfyd54~MTv2sq$ttvJ_cL1tX5!5~$
zg0yMbe3*LGK$q^pq|wl&S}3MIwmsT50N#*77`+H6CkEGTNK$fUUwOv8oBBq^+xbau
z)B)>FKs8;0<HSiekA2A}^#$VS>={V3hvp~yg|K+V#lguwfe9LQOSe>Os6F_Hb38Fj
zBt7Hi>iubHY1C}!SsI+bsyOQgzqsWxQfJ%wHN%s+E)W?Vi@%GV)J#qFh~BoEvu(J^
zqK*=qJ@cfgp3;2!cih}~gAKxs>#;i-vOkpci}#c(<N*iy^|A({LEhlE!4GW~v20q@
z0cSJ%)F%pm0HOugj>_`b&yBb42F<S&%u<Wj645YZ-{P$gCnibDd09bZoEi;B;UW~*
z_RGe`lMC6QwZCA3Xw2aIahToh>~wFKp|GBp9YOl(ihnDH`z`zxMCKQDWZA)N)dB_;
zw$mys&Yo{FK%_f;)&0Vm+b{;Bm-}eZne3$G|9CR{=fm)&L3-KGkVFg$bwe})^Dm#O
z?Iy0N2m3<AswI`sn}e<<d8)ear1b7{zeD5+zVyiuYkRsG=b!btZK+(|8cEE;EO8Ad
zNtJZ#;Wdej9;>^HnX(<yUkrR`w+WQ_JH7#22}`wDJ|JKF5SuiB{DlyyKx+d+Zw1Q$
zv}t3wWVnMl$#0##--GQkcFoPM&r9EPnJFbZt%p5IX3;JpM{5X<b#Sk~r(Wj#JKq1u
zc%wSsjB}2qhHHnz%XtJd16B(d^wR2X)h4ImZPW%5r1G2aiZxX;4ya(;lQVky&W0vA
zDZ4<~sEz?k`Kx~nvIl#_8-%kiFD|F8;+=b&TNiq+WI8R@BVtqr0CXP!llO)`dr;W9
zwyJ$%a^_Y&%cwL3w<!V$%#u=wav|DNm$uG4mb7QcKwG0Qm%(_X7<TuS11$;W95Bt|
zXu&3ZFqK(aP0z@X)B;Ald8X5w&V@|ioB(M2#IB957wAb+3rd&hS^oFq?Y1P4#&GQ1
zrJ+9Vhe=?}WFwnBwI?m#;yJGCjV_Pzt9l<`RQK&rEfo&42rFfi)#d~ZHU&tCz!RN=
z)%VpFa`&y(KS~^xTlp1gD&Gtz{<~5KUGLZ^v^Ypj(=xcGF1a${#K6bO3~R4IT&orn
z3n=#Xp{Pi!e36QEMx7#U)^Q(TNs}T2xTj=3OBZr`Md`A#ZHX7WOz`O?(#C(y$iT4O
zdmBnq`|DC>v0XLB#05B27(+G^DWNnH#*Guq7)}wHQ)}AEK>X0TpofVS!sLESiY+F+
zg0)AQtnrhw=HFB0AF?A+&xn?*5|3A<_Y+AIFt*b&fqBv;juOvKlZA9?zl*TLDjLW6
zNC&$qQiyYpW}N$mH1c`9RU`Z$O%3oWes>OL8B=gfQrOFW^OZboQqfXoGfgjrQ*iAi
z{T?Ir`FWUU;LJT`o-vCLO1Wd7cj4iBE^8X9(h`MU(0s4jD29}Pi)VLj2OT1T*=E#u
z_RZr=W!oDth4N?2Fx%%vGQT}QY9I<!>MU79DLpQXd`~&W)HJ{^B=vFsp}b2`tfwq7
zgh*A~A@rp-w8qw=ho@?NV{IL#Mq@&xFH-D3YwZ4&^X~NVg>miS?D{;phS`LL!8RJ?
zz`87;kiGUMPNSX+QO=8l*F?UW^>rl`T=X8rJO<2B#CKaryvdT;{^Z@D0c~A>ylf1<
zK9BjAB3bYC8Defl{;S(3boMhgsM7h6wgFiHem`x-)CN#&A$1X7D6>Es38s>fnkpV&
z@811}?0wm+L5xRo2Ja$B&U&Nms&yD~Mt8t!@$^;|9I|mLk^B@+odx-rh%DKdzA@UM
z;W1B`$-x^}xw0zPH1>588LT_3pM&_+gWP;xDUZ|qDOxuySLpJ9**wd>o9tXpomt4n
zYV^`zbGybWPiPpsh!88!^cT%TYreQ*hfdF5=|~l9CF;(0#<_;jg`b$GQ=`{A`T|GT
zt;7x)WJ*k-bPuwE;Xy~Nn1ajrq$V+6cyaH$ZX-T*xd!yAa`8A!DigHYsXNGie(n@&
zVsq#@H<!{%r|q`dW|jY<r5&S8!+Zp$%qC&=1;`Jtr9L>;@K*FrG-wpiSg2N|<9z}-
zj4reRK$DQT4FkqsVV4wKRhK;{@$Z0HAtj#aSsW!o&5qWS00dbhv*$lXkU1k+Iu4}_
za@{yu3;9W&Sh!KX52}}n1t|z)N`SE6?}|Q5tk9>a_Mm_a9m%IQAq76(0?W-kiK2eV
z7zATe{}hjigRn7_fwHram+6)5m!AzS^s1sV#m2Yapy=-6!iZn-VJZ!PGT+GT9B9Py
zoW`PV=#eJ0ghwj5txq~=&((-+q2Se%PL9OfO`>4k;G98^kN5(MK!%k5N58?z`2W%0
z^?#m7|N0HzS-pHQtS~Ge@ubB+M=fj55!c%~%XIRW^e{Da6kKBoG~=@9B0n!UGjM=$
zdxsTmf*O83X$N71lLvO8$y54-c>;zW4Q=gi8ENIP0ZSaZ<@kDq40WPJ7R{=H*!mWS
zXZO34%2rAALO?nim34g6*>rAAuiob7%rRGTy{aLbTy@lr&TIpz?xxSj@$n#;_s7O`
z=CJh9oolPVgSV$Auae}jFQ5GOk3V^2ug_0v=CAK(*ztQw1!JyYPq~q6dXMXcD=WDz
zTKaGI{9R3~@-a-J<Ot--ps7gn_V!yRBc;_`G4!4cp7RNF@4G>(V4kuvxH)HFzVdt_
zlnJ^{nR4QZexhZt-g4np9+K3y8b>{f7SR1s*WB<&*z2%AGu27<D}O$X^youdJO?hn
zPOxn3d(BG9gBM1u$k3Rw<^9f<w6aM_%1v0*G%=Io=G4^wx$f=vpL0_&)GIO6YvoWm
znhAsfFklF$a=()^^{guvDwEObOjT7J3nG$`eh#gRp^u)+t#fiY$4X@J?mArwYltVm
zE|zN2C{TxtSsxaOI5e7Bp07%9uPcvZQhqJj@xk5+q99;L#KmULp745BoNG(Oul=*N
zPjh``RB;J#jv1zhEVd{ln8%~JH?8(Lvwt)I14%}-X)e6l-9VH)g2`02q?fl9i7Z`i
zc<4Ph4bRtYvrgt)rZ4<mdLokv{YW6sh7`1=n+2@`hvS^gdF2zQZXT0iL&E?>hUTou
zYsHsZdR*sMSRof#XmonEnyrN_H}dGhk^fdCbr&@2tvjA{(MqNfSi!J#>YnYvbr4R3
zt>wjjIvHNvOHfw#Q&%!3o~6?nIyoj^@H1ylcF+6}xsf?arO{NImUn?yf1{~h0qC^;
z!ptRS-ab_k(^Ivv^A}^E+~0Co&8Lwh*l&-<I0f_?D0Pr)&v@alGLqB-jPI3;(#hl1
zmP)KWZo3|i>1auqWSPA5aiwA(Lqrk6-$uSpmBcVKHi-c$0cGS6(1|RN(5_Qu*d<Yu
z%{yNblHFAkA$$SM_M6w1=;d4RLt8RiSz*cnT<J($M~rVp%?9=Q8F4!stkE$LbrW6{
zEk7Za9~1OGGT+}TlCAL&`_ylM6q`iTIGy+5zupp1@85h3dt|T16W6|ZHkm&wS89t_
zht19(>}mxDkdF<CfAe@}&_kpGL({=_(TQwzkVDuf6Hbe;w2^E#=TxL<XZdU0aLY;#
zs*0k4(sWdw8C3M)RpimuloQOTZ`Fa@#LP`fPY|C*L0-=XBxU+oI2RGK3JRb$?L_Qp
zh1L&TR1qDi&&y*nk)3XI)RxBhhi#q;wG{2eSz+_@mI3QC2hA}IXHq+VIrxnnW{=4t
zxam=~g@-z`FL0F&em6zbKiy&Q8@PaqX6dzjG#GK7b1f<+M>o~wCCf;_+NAWcJ42IW
zd1&U(F&fsI%q|lAM9V-d%$*nArAV&zXfo-k;;b#wHCZIv4yNs<G<Ba?b+eeA7t1;r
z<acvbr*`*6a88q*CL|$?WLR4ICJ=`ouR|GOegV%_aRghD5i#vF!DBL{AuGBpgiwvV
z9vtPTK7N6Np`vIvVy_ritm(Cy`2Mksv|sn{(nRms5_m#i5HEYTKy2HZ3!q&}7?VnB
zs}-chC0!1!I@Bjgw<r$i%K*DBjej?=gU`d(m-~c&WRlxpQc;hW@C?ff><i{$i*L)@
z^Dh84MeK9s8AdOSQ$o|ndw<VKS=0&!42IR-QwuWs>zPB-S#r0het~&!9T6ej$78z_
zOZUN8L&hTYt}O{Y2~KU;ZQO}(B+cT<H}uH^gKQazz2kBQVc?YvwEs~z(HsNfM}RjK
zReQZgTlJ!xAIeC?X?ZkpqXuq1a?CuBUpE?@m&v7KLGU^oLnjK!Hf+nr*RgZLE<9eH
zvH(7cx(g8mkCZ@)kk|(|cQO7Q2oEhgM~zL}ytaJQ$9+*jU>*Vtc0lTvKu9~p<tX5d
zjM4|5sZ{o^%sQBr!HEjatH)Y-KM&fEML{|z3BjVd-AT85LzRU&>48~UPnJd$-gL7E
z$D8HH!d9)fp!IT#RqRg)2^(}ni&-Pjc^DP$d|#yO#{iSls10|P=Z3^^gi<v%V|3)K
z)Bf$31cry4Ptv8G*{=f6ahyKkjoK)VbfhrR2ekbitHq`^>wjWzrHzS)j@SFoc#@67
zm|^f%tlgUE-M3e>W+xJ&dC{UZ_E!Q;6$Ll&2-OX(%d)?3E8T7ko`n`}8#Cc{;A*4&
zM+gLbVh_o!NoB@XK`xsgv<q$24lSpo?IAA5?j4ii&xSL;@8l&VhQa8m$@Q7}yi&bV
zpDminoSNAu+|hvdN;~|f4Um5qB=<?<wE1=Z=b@DQ%`@6ai>$)I_(=z#_0huGW%F;(
z=SpOdB7pq&mCx}H9A)O*YLuM(AKW7eB;IwrI5`KB>HIr0`SS})ciQ(v^Z3~?c&5rG
z1cumA=75rc!#(a+=wG3T+=ECS**x4`CNt$Z(AKVJ&{EjE^`^iK>7QivBhv5v&Q&I-
z$0p_}dt?V}JP&6o&lC)D__pX6e897K2>(^^**X8;rgvp`I}<_%IYUb&CjbnCEFlvk
zBMgJ6g@dCLAtyWY|Fef;WB#895Yw6p|7T|E^HRHO^dT>K7~e&Lh$^mSOjidivH-jh
z!DAe*1}1eiZSQxzbHmN0L91bJECKvHw{MIW>$>ADhBwZEZfkkKk!Vj<wi|_^NR)v%
zQE3d)n9r4QD(rhpK2jBshMG^~4B+5G%QA?^(yd3zeZuIf^{!TAnJEV@M=8QoxsPK6
zCZ$9f1hr>GUA?fCk9il&>NtrAGPrbufTqTsgc|X!N>L$eTz4FQ)eH?66RvSP8FwRC
zgu%VnRh|_)D0kak+bv#y=zkSQa3loFjX(=P3$X$_;f2hef+baO3<6dJA?}(6M;S2Y
zwAxK#nkY8_iHdmy_>turi?aGCnnU*^gFBPDi5D;-`M?&{gz8hGcgp804?u%SETXgt
zRa}xThy2bW>=RA55UpV|VD{7(AD%pwvC{wm_3lc+F=Fi|NQ$?a;wGdawnqWsBoNa>
zduZ=V8V3+=fT;^USEMj%$Xb$8X$5d)tqRxP8M-}^$>?L>(kZer8RJdJYN$Yjg@-#q
zkfW@*lY+<!3sCy~5|qa^5R!b{m7EH;w&Pj?$C)ALEXwFhJ(MgFh&=2LA)6s3&PJ7%
z^M^tahYxpFq%2j&guMvgBlnjM2L|@H2#3`Ib79Su^%I#Olc0pFi6r4qx(EP;0`DN7
z@-w)E9Yi9dF^?Ov0}{~LPKx7-$~B}F!b*@$Q4pT$Lsz6?RG`8_#i&G$ChTl1EYxsr
z(5{-Ek1v-+{lISif^JwK23cT}6ZZ;xv2~LCmaA39*y459ujcIay?5(ad8|u`TkNzi
zlEe9^)tjw|tvnh&^8)mwogZ|fD0%1j>SZ=!>6XXl{=8YmV4UckG@jFaW-~0acVvL!
z@m2nrwoPfRjC`9mEvM@JV(IPT@@@RI=_v!<rE-g(VVr-1Da^;bh9F=-ms0;|&=_CL
zU2YOFQ0>;Jp+BiKto~EEtwvDM>c&qzv*-`ocHPx#!dlfGtvOZS0%qfeVGRe?y|A&f
zQm66d*8JMY^@od^S$P!WqVlMKtJ>%LQQ)V$Zpp*BvyEdVS5IdOgMD&oX}|L8W8+B5
z&E;g1Te>XLuKyub&sV59Rm8>K)7$&GO;J!J+*y1&F@n^v+=lAUT(Lwv;GX5|o<D9c
zd<HN4co2CDO8LEy9G)d0ZcE`(iRTJ=jXvV)w|VL;Tg3D}QCs4nuAEw$INlN`s!h4;
zg=BAM_il>J=7#t=#+O7}rlk6SVM7$!xBflr3dLg1{OsM9(p>2$n9NJ9_J#y@)a{51
ztO1t4-FB=UU(PeAjoTbvlLLwQQJL+W_u606{zpV{NH7a9u+J{{9W8Vv)PR>xzgG_q
z!fl!9a5H^>IXr<-B!N*24Mq;ebM<Af9h2x!8~ygt(8By89jTTid=Y)md0&1bw;_9W
z!-72|rBwp^$??)nSUl)_P?UxxqKoIFJA}8IPS)b?E1NVO_cq%!6fJA?c$Nk%ZZ7i4
zzL+$CqFBtoEC!jU!TX25K)Dz<KkD==X;wD>$aK@X>g>~UiNOxNx^2~%xL(4AEqV$#
z-+8KZfyR(bcBPe`*U9WojO<bRHc(2mg0VN`FP8#ZGLN9jgZ<-pySGqw9@ZQglImlW
zYf(?T5+XNNl{<d%1RdU2w?-T9{Xk;t;2a%$Kw^B(B#W&;T772{l<xT92(Ec3o&_im
z^jsERYKyBe*PJI7`^)FWmbCUWyc^+hMKLbFmj?Z%{=LrFk{(G}G;1jSP?{5M_tm$5
zv#xTy&GVm|>nLFY+g*a%S>wYSw_G{b{|bD$YZ@K~Y(ArUg1$=J9>s*vjYD}o{1{vH
z;br<h4}u>2F_YYUIBcxO6tCJg#9i&jvt#))%$C57fWi^&m-N9uQLUs<4@`H{U*yr1
zz2KRVB>Go|>kru97a>A!iV+WBmZ$cCy2ddoP2V$f2GSTHiYG!@Aqzda7E~(8N(`e!
zCTm9PHRryR(7}|8lcUAw@uk~5?w0U$Q7}AmzIv~ae+OUhn@q02>&7pu9d_{gtWn(_
z_^!%q+`pUlrAeMURO2%`)w))dd`GIDQ^j-&fbMTH!6-JdFBGWDjwcx2`e~|{-li`X
z&+Z-%=jOI7-<|F%-o<irE1+YLwIJi)&oA##7)ntn?L}FWA2!F|;KBVW0{``pW%_^D
zwg0Q1;AHy0W$S-{xm>C3*e)<4`rPOl+zCd6-L2JUGu|;GW=6;(Y3+y-bg*3!11!d4
z%aQkePYk(r@I>?J2=U#*uNu8ywp>pd2gd>!9kxAr&$~H5Q{sZiBWQkSpyY=|p_;^#
z7Gkx-aLCJi<SWy1c=K^|<HF5;5T(*$!T~%v&o3xBoD10rLR@$-d%D{$W5sTCvqTKF
zd`YPiVWzd7U@}B_j}e(N%<0L_T80BY&CmDebZo`CXL<P>qkxbCe)oymQ#%WmxOR)u
zCk;I@0fQ@xGEpi1zVYTkJsJ^1q&$K4DXPol84r$x3k@3@%R&Yi6BbY}r*u@L9UggU
zY!)~Alwrb}CR*;tht~%iO9_Fnc;tOY>U_ZJi(@iZAd3j@DicyR{`|>Sog82e`%6=r
zOyvEH61a3OK|i%$*6C@@tFdvM40u+sVB=T2dhlkwEo2QVWOW*68mQ-09GHZQ!^B5w
z_V>12d96=X*4OMrqwoyECshSS{i$AE`7!#59qQ5Y!&me-!rHHao91?(V~xQG3yOY{
zlm3Ta0n4)+7gbAft%=E9*Z@DF`J}&Rh?O*dM6*WERFoQky1_*HpeXOD7||Um>~k3?
zRNA5PadG<2QUB~KO9_wtsZ2NYk*#{E0P9V6$({Mi>r;>)%7yO}5G#q6wkBz$Zs=O~
z)<ug(x#{BC0Lhh9#_HQ#e<n={pzwL~x%Vb(lGx6QCZUc8Ju}%Sa0Prx)l{?B#LJsk
z;PWzy^$>E~o+IAvll(c{)CM1!IwfyrxHr+MKA(7IUa5)rGt}>(D!1@=I%)RHa6Lb+
zTsYIIA%$P5W|KhhWB_6`&<?P#Su+YK2g>7dfo|XK)@s%qiT#u!WVZkCQVXKU5ar<O
z5NZy@{W9b#gM(jf|A(4wam__HX>oc;Y?wlKPhr@+KQekT`-la?mX3a)UF!7rUv}%`
z_MD?az_Ful%iUZiM}{Xx!3piq-7D;I-(e5DXll12vY@D4fWPjr#>j^($u7>tJaD(<
zd<l;r1Lx6!kkrsiJv-r=&Rj&KOG3dFE@(H1o&!PSqO@%+uMZk<0xVF{;bl{9lr*lj
zCd$^JB`bcoEq9(BGR+lT2nm@D;ma2=uZ2+JVyx|#oC_gtOW&tEl4fdH>PKQs*TYFw
zGCM+&0HExd`tl?F;Spn6@ka6u^jhTW95ZJsmuV4j8XxBaL--1JfVSZc|7)ku$@E`4
zeO4CM|8+q&t*&Le#RlgyTYFpBNLE3*gK=p9x@-uls;}Y4TO^Xd%0)5Y#o2txqEY|#
z9ADtHf52iMUoE0X8(%;fQSc|?c%E&4K9Z!z(zlDl<GBkto-z$p16#+sfHd}RgQ13s
zFouj1mehERp5A9eF0xJzOwPpXN_+NH&C+P;?}NKp??PLdUNXric;^|FUL7gF2D8}(
z(hBwDN!(P_D5<$=l0DT~hm_Ph56J%?WA6}MSp#k1#<p$Swr$&}*mhE}Z6_7mc5-6d
zs@T@6``CDY<92rAG<NHZvDTbx?r&E82vmH(vLTab*ztBZE!{lZkeVwsJ&P^|(#O|*
z?%DHUi1Te_NN^6X!;!k37v`s&Jy}@;X!D$g*m!|e(i(6i-7N8yhdDd*ej`)W3mjBe
zCWQ@H1CmIHL%vo9B-mriRxp}j4l;Ip!lhtmh+cwHL)eqC-%&{{FZ8KR?8;1$4KEU<
zmOn6Ji?9E+TvPUtbNk<*Qtng6U$Z$vni%_$V9(Mte*en&7-hL1o88$_c=A}PN?m>q
zeMj9O1FR9&>7ZnH)H(5TL{vt4^HLrwgkY_>mZ|UjrS+?^httw6VJ79pOiXOhi=aY>
zroB&^`T{l$)=Slh{v&k011hxU)rLDd@w*Ts4*#Q6Q0pOG#K(G#8|?``)_wxp>vM}6
znIdCHL~iAs42x<ZjJ3mvdWW<b;b3x5V!IpeD=e4KeyJql7=!4<;?^YUFE0Dez^H;9
z@8#nJF(Tr|Tf<6Mq=h~%{{r+sw5^5#^)*U+6{zB#JFEi<{ze*(jVRdtEB>F3=Wu~3
zJ5&U!`Yk)>Q+sH9E%kA_=#{JXSOQ$0fV*LBL4CX){tXAJAk;vJ>jJ9DKi6>y`A68}
z>GhgWStK}a&D`%-6%r6rW*`jw$>_d~p%nOzGyHinxAxF`X_N$tz<3$sGJ~poXXJST
zJ`7?l3F*P+{d;7w8B0Eh0lsff8zhztJh?QBgw>PSkc#u=XeBp+F5hz;OX-((s8!wn
zvMEXZ45xhlH-pz~VTaOyziLuxMsxe!U6bGJ&|jkRf-$+joT@6f^{Zq}g$AR2Wv5f~
z5Z-o`Twrrh&vlxv&~O|jWML76oZ9bqf5<gnPM@r<0a_4ZB&-@5-1_x`aZPL-i|oNt
zUI}e{A_HNgXvs%FNTW$w^4xXz($#P;@Li*b@I8b>ve)d*I_wDW%7G}v)!bla616)-
zcTq0dxZ8^1YY7b^BA`&&sY#cJw^+5jI)lAYxQw(D1X~`7jcn6D`5)KU_r{+?`f2I|
zGI<|T-%?nFqqk5u>73I{J?pWetgEOp@9hZ8zR`Cd1D@YG%GpD(N*L5)b+c8N3~1)S
z#HD2u=Jg~YT3Gbt(pW@6bURcR;7q~lgx*=dH;vbjCuapfD}%=XVu!}~U+x^7l_Lpc
zy_#ko8({?XOa&p=>N3e4W<>zs4ZG9t*{+1#{%AkRUw7mzba_U2)x~<y`ckzL`Yndi
zj?vwr@Z=Y_46tdZ`JCO0Ii-W;l`r`@o<-zq6l~cArn#gHhVQcT)0T=eHlIi<eXcS+
zib00*3Rfq7Dqne7l%)&e=b=e^#@Bj2-^@UlCyG*l2}~IPlmkhaG6P6Aw<}G-7Irmg
z&H|`{6CxM*QK$XsWY+IcsNg6d2PMro3mflnV(|vB@dZ-c>OvcKT|cIVm@8%sTeTnb
zj62;r&@?Sf0WY#75S2zq%Y{Tp4^m(CKr>wv4xFgCvpA@>ikuH>q7^Jy^|R&6Xnd?-
z6N?1`q99*Oow_?f-NJM&Qvc33`92l=<2FttIG;aJYC`G7m;@EXhab3fZWtzV<V3Ws
zTj?DVcvYjEH&_C+Y?9eevWV{<Cpg&Nz5~p|t73%e3`!hx{?JpM)RlO>BkPuC)}PT@
zcOY}RYUA=-sdF3W<)aH9zpBn2WVQMS^PWz&*mK4gi;@89!v4{yA7fh~yF@9&UfyYS
zWPk8A)=#fQZ`Po?&6Ue)+NhvG9t_Yf8oxJf^lLxKYKg;vCo^UBEn4Wzg`}KD<7o(*
z*?%7cN-o?&HoC=$yGX5wh7VY3hEu{!us~ylo8u#f=S+f9OXd84+<6t}{O1b!AK43L
zj{iOD?q3Ag<)8KOKiLbZSt%wAM1hToQo9`KMYD|<o<(d)NQqGTGL*9@SKF6WkN=Z=
zbIIY3^Wu~M#1J@Ql26zl5*-N17}uMLpPK`s7?M(&7!kbgMN}+Q72#AGqD_z!Cc<Pi
zriy=e`!p+vH%61w^LZD3I;L9a^?E<;F&p6Zb~1A3`}{Rgk>n>J!&*za^?OZhL4oJM
z-*RR_%L+ecb6VWd>hD!~t+xR0`xTU<E`bmc`5{n{3N$KS@W}i4f_aIPKg#f0!6z<i
z4&m2hg^0T(Sfp#>L4I1Zm5|pa4$uOT@Q8RLI}jBHIkPpxrfuV9&~<{R1lXvu=eum6
zJC@Q;=Y(^&_qFCOqoE^5EEi&QBlex+(a36EVhd}s2JBWea{*CtvTC8)O$uK|un`Ix
z{{7L>nv+?KG9~A!xOHz)JUIJop7KYWO)ce;q8~0PvIK$q&)sMzq4bmEa@B0<J4s&!
zVXfbrWwS%u27BORl!NYV&1*b1iH8OjY>rQu*AIUaPfkE*4~pEl)7HttiXN`2VGWK$
zsqd<o6E1G=m5D`sBtg<&iA+LEGP}lD5OPPb;B_4E0Bw_XASgyTYmRl85#-?T=>;UO
z*m7zH&1>0$UKbZrXB-yJ^lVcpQPov(b(IcmZ@KK<u!NQ1FT0-i7qp7JKO`u0^GsS=
zOU*v^J5CG$jf`pv=-H~}%Ffy^=Elxm9n9kq{XuO%p($6KH+WmCt0?l6Cr}4FO0boU
z^PG*&*X_$q(ZWTYx>EJJ=lW&2(QPme*T9*S;S@=0+jP0&6!C-tmoYgp0+i<W%dYmP
zw%(@v0bc=ZBr(Z;HdZ`u_IwXLqxd~lI0P_i)&49yJJYZX`nq$!UqajBMX;$uWs+_|
z0wt!-(p8tXcHX}{dTeicE6f1gsLXsuT1GfLsB#gzPmNl16}p$^H7T-m9$Fk0(jG@e
zwzG42fPc(dI9kg)AS_H>hm=IK2AOeOLx)2}+9<(`+1Jt`aSar1-DUKr9bBBwN_``u
z<4&tIF_%KV>F5?}yM7DWzO|D(oamSo622M^v8Q`^sop$CoQxy@q*GXqa}zdpgbC8f
z<M(Hnap@egtU1r`RO#TEG=FO?>QakD1%j%QsJrtCLoNBW{ap~bHtxA^Be2+Jd7I?z
zPB?nxlH5DN;4fUH%N3X+@ai8Wpbys~`T!j$PplZo;Hz)PoEYCd|IC-;II%Z<0!C^+
z5;e%d%-TR>J0V+gv%w+^q^UIR5Oa+YC^<SUWo1vVWi~LFgmb1K{1&C4QCP(ykXW$j
z9L^R?7wu(nQ{Zk~0EWE0s~AwUg&VKV`2c28ZIf=_Fyk7@y}lf&*svKaZsW_cq?mvz
zI=LsF8KrW|X*=!{j)%zk+d%yKIbdmN_I4p?!HspLh5NVTlM}D~9o*CEdL4a7^`Gfz
zmo9=I5z!n#m=~p(EhlaeJ}@9}5IWW+d9K495*9jyW8|0k_IE0n;O_2qkGtcvEf)O1
z?Rxmzb1l>a#_xcP8&Cj|s#y9jM+c}msH~tatVbp<LZLw*p@ZX@Oe9|VKrlq2LePZ0
z2HMBlHsP&PNS~PyL$Kw<T+2RCt{zjagc~<F=wMf}1k3gmI6ioKf@A4dJM|(PL${r)
zFD44YbG+1o%8zqn72atMXah7rbZQHipL+vxyeX*w5-#ajAQqNm0O_=h!);8~PU`x5
z<nE-~7$<i5M0rq>@V~!=NlX&+Q;D9RbqXBL&4h@c*QeRD0AHr^x*t)&&EDcE?KV+P
zfyoR1zL2zUN@q81@cMc{7h{aU!rsHGIdk=7ziHWFE8S)k8CWUK?n?tBc)PQ)^!o8}
zB@;J*E_JpdO)u5LleNv}51;ac)8((C(=z64c9AoJ?4&YRaa$!nRXXksL5HVC)$VD9
z21YC84H>E7viqwA#<euz?83v65z@GUqZvt|RHi(*Fk#GWJa4#jeBl!6m8pFdQ)c0n
zwR-_1v?gELsv(+?sTyrMA}KIgC)%-5;^?lOf0lElmcwLA2BB~7Lej|QDFqE+D123F
z-<;6h!2M{?9J24Px#Ta)7~P@{RFkgSafrOmk_td4QcXv-5~RMiL@@0n7wH4sb2q5;
z@`^!%!M#DpBG#V;-p&Bx##&WQY!k{K6Zc)+xt(%(I<WvCQ^R8A{~BT}|IuL2%*Fn{
z)<?@&`YxO8ciVlz@k=rAhtWWp3WUx*T=JCDdVc54tUQ8BWM-_AiIq~^&wV!_u;XMk
zQg7{g*95k7n*G-%eVA;Wu~v`om*H>g$Ej$_5-cO?Y=+|h{7kh_09jW0Jjsd7jQz<^
zd$$Q2uB|GyG956E{o?NT_kH<W)x|ovqSX|4raoEJ=+EcNW@g&m=d_}O;l857&e!XC
zpTkbiXkGe^zu4aUao6f1e*Z&^H#PMHn_}Rm+4hbE;m6s#tYS}A{?|<mur5x66qE>d
zG#Mt@JPXZ=B?hA!$j4%YvP_bgm@aC~%=c5N%Jt1S*q#@}51qV^RcOcaiV5An+)@b2
zfSw#m>aS+8fAuNxq{qO{K~N&X-k<3#M@)lyUDLka@1Lv+hEy`9%%zTp)H%I{<x1>d
zp}$2j1MWgmz$rtQNuP>hn1ue`N6B)}p~f;`Q3gw^Egm0FDU*c~DP^fIXJEc6@7-(a
z`4>XB1YF|vg&Cp>LF9bt`(%KVp}L2EWqUqY^UF>1A^%*|d`{b{r6S^gc=JXoHY6WD
zO=)aZ304Ov!QMJp1Ap_1NB#Q=<U{H$`Ssh<vfaU=Yhm5a``Ec;-(>&u+t(;4n~{^#
zl=`!~?r303ck$|cN4JA71M!J!8eb-Kk#LR4h2S=}t|C8~<2N!u6j>39EKF)4^fpEE
zfnD!Gut{uGJ%F<di7s~4aru`j?q_Az0$&D#8M??AAz!PtQ_bediY;>Q<V~zYoCE>x
z1|$QOvP!%*AqyeO+h6d4eoAi<3+u`gjFG=L-7)&T?n)S^+sqvW)=1cpt{f<57I(_l
zLi^U|imYW3fX585248dE+z_VCOa-E{f_zk}Y;d_n2f}#h5f*tYCj3Nt%v8;#4^<PV
zLgJa%SKr?AnuZ)8#Im&~1#8lQdvl+gfy$w$)=p0N;7O-SuX~Peu$HI{D-B99RVN{T
zxFS;JPu~Zto1#u^d&BNN?qe=!EC556)XA5HKc>Ak!F`L1x5rcGj@uW<jXdYPVi$8@
ztxjEnySXNJ1h&W4=#)X5I%}bm4GxjYp=6>*t1pzo;g)$<;c$UqCQ?=@)m31$+Bo<?
z=lhe@UuLUyahlPkdPA;CmNdtr^3$l+4$jS9(G@S&loP90QPVtmF|L_BS%g9XfyN8D
zNIczH>?fFh_fgkN7IyzlGMht0=OAvuEMOE|GGFKEy`&W?q#1x6wtA7Gv?dddGyE!s
zMly<Yms!97d-TUj?PK#Z7HXjEn5zh04vUFIv-;s*t3z-{04VhJgIy|Pt^@GKsVb}q
zFxB1RqB@KV<Lpv~wG)X@2Zapfteg3wHQ<2Lz`{`|V!F0XTK%HD-q;(7GY6#td*SaM
z3l{r?>Jr(VPlQwGt;m=CrfbBlKx&}Qc`oFGz$XTg<OQFR$95i8yKx7tQkcr9c?*X;
zq%>(=)i<(-F%5pHOcn}+yE92!>;NUu6akwV;<<T?fr7-u>pe}Ay+|Y3ey?(~GUMl^
z(UowSJjT=|+jmE8_aC9J5F1dud^{PSgn=g@%Ib0wgqgfsFo$|dTC5kIJDAo8JWLzF
zx%=uhp3X2n9?NU8d=192N3aM3g!iMG-piyFM%Q;J0mbB@eOq_G28zs)(2f`~fy_Xo
zeY9bX9h-$)_U3KV&s?0gZiyv42oIzbF>>0g?5Z~n`~dE!?F#p}>X92ikMdLdd@7Mv
zh~PE02d=7G(0=81?aP6@Ubw9A0;oF48p5T}ocA7S$Zy~U?eh(=Qxb<w#5Dq?0Bj)l
zZ25lrP5plVEn+%pX)txE`^xR#sQUSwVccViOiG%mLC%nR=VLvHC)2DuK3?aP*pT<t
z8&|J?Q0KZnJp36<BZ*`x&Rnp~*by*3X6nr8LYxABESgHbU3%QU_AQ*={3y@ryg5AY
zxXYKDqWwF9(9H-71+C2pr!GfCLf;Ct#wnS+K$fRD^8whnh?H;|h2~faa1@_PQFb+f
z!yKk}kp{?oL%rjOk&&3^j$amvq96&u%66`{)|Xm0uxgxiX0=LTNu^F;1J2mP0&KN-
zHrS;U&~T5W-z#eSk=nl@ll>HDwu0tyD_}TynecohRFG@7<iyZ%;V)~4QncYr?>j}2
ze1hm|jd;LDaIHYSIf3#JKUmiGbkb4wl{$bz0dDKwTxg7czG^+gt_pCtF0fDLXw4g-
z<>~PsX;<oX-VvO@@oxhoEw{X^e=WL=TidN%JtH^BsvdpN6Lm<Y)7Q#xk{S_kwEA4w
z^cD7(p`IAeBBmDxEzZFU<{LMqNbTnXrSVRRl<zHG+YJv;p#X6+n1_9t?7ytobpgXl
z1MKAZ-h0}|0+wno7Bt3dVO)LU9~RPA%qkq_fVQ_=x;n)?&!FubZveJ_xPo>qD&o~2
zCdvMU-ie_6m;(4hlK3a&yM?yZzphHe5qe97VDADbPX2)X=Rp214h9p`|BaddpL4sd
ztGP1~qlB%otGSrDse_q0jDP@)i>tG_u^o)(#`OP}=$BHLX38VmufXF#_c>UKY2wwO
zsc`l!AR9MHLuny9mTZ0QD-VJ&OCpus$t<D41O=JvUyZ%4!)1#lAi?qD`#5~O{ESk@
zw=P1>6I(PVOOl!hRjMdDBkBCdTMmw)peejN?eoJaWyv%6c|CJx+m9Mu<qYWj9DACU
z{Crx)wJi^LUd(5xEAu~x*(9`gs1w}r?d{IoiMuWj;O<K*@@Z8ItpqFBpO0*ep7|r`
z<6o+6`9wySf16^)BV4GsoLq5N4`}PU;((j3+_nhXHfds-E#n3XK?dv#;V@gm(?~%)
zEV5tvZwEUT9Tp96FlNJeQ^&TcFQPRq?0M=MZ}zVr!#@AC*mS$bk(7qRkRq(EA<%|y
z8}<nY6-Xfk`XF1Q{P?BZw=-;*JOpkql@-G+Nr$SQs$*7QE9(_AJn%X>1?YW|kwYK)
zF;{i&0fw4-EYz2;Xxe_a2&NbGBV*Aai<PXeYGoblp5&{|bAK9B20FDn^Q(<_R-~-p
z9pInrRS?OH6UV{29asT>Cm_Us8x*g7smODw0z>k`i8&6c;**5#MG0%|B|%R@g=4qf
z{?bhQ`|Q^1#Z*obh*IMQiGw6E$P((VZhMd7o5hR~|0S`aSQ-zr91Z?@Yy!MN*^U|l
z`!J;Ch*G_Bu-EO)*0S5A|CVa8i18F*-mqKMB;7K+G7@cf!P6$h4FpaO8AKmO7A&&_
zKywtbyMZye7-_{e%1ZAFb0g)%yVfAZ*u1mc9GRZ<)=1^D?Rouk0WIMQ_<&+i@a&d~
zR074ur7n(?)6U@pwc07b&^dmhk^gRAwWy!V4v#b<6juR|OaNEBIJK)oRRT6NC>pdK
zbnEGtIWEEE4{QXPt2R=vO7Hj_<X8m&;MC-L&+(bNJ;;GRwsezHX=eP?xK?9y3aljL
zgm|t_=VT0EW`hILbu;OsC+YeIMLT;ZV?buq%D<ojd!iqC(g!1_SGMIxp^mK8r!=8T
z*7oe&`Eq|Hl|aKZcWs1ALTTn(+1h%!G5tdP1F8cB0Qs6i+*21blRT4lNiwTeRBSV6
z_Q<nVG`Qw)xv+FVR(-C_kWv*<b%@-=ILsT-^$=e+ZGgITe^p_FyBlDjO=eWUn<}7K
zXk>IB{K)`Qfk?laX!cUh<Drp;niVs=UKUc=>`yOfv-HBy>}s`vabyP2FCUu4+BNCM
zWth_fBvCO0vyo%o$YpH_r@L#;E%+X(5~Y$T1hflV@)P1-P^2DfH-Z&fn3KgDII<xE
zO~<W{PDk&;pFaytlZ#1h#t07+P(e0%-MSGsBhzcLC+4!dWV3zk@a$n`Ak^uS|JuGW
z6&r~OUIOLDgLYMgjad1jS4utP)U}mjzsM}CE}-4yQt-HWKuMN49`;5^6mR+U63U~<
z*u6JR6V?-l`mP<CIc{q!{vCy#_7+)@r*}9q{q!?^I~<xY(NGdc$@0z*qx-7o#v!gC
z&k0()e5J;IG)~Topa%}@KFA#47=vIR7En^W%SgY5xx66?_JZMAk+&hLG$pH|LLO>z
zP&iUA9bk7g98DSJ5e=c-H2N!i*B-y(f<)+RbdA~u+L3S$jN~{7RylSCfFn(?0sf?Z
zk8y-bniF#-pn8L!*-ujHWTefnd_Of#Xtx^VVF<Vr!xtGsmd5Z&x)a|=Ra3i(N`>Zw
zDlz(sa9X<M1%)t27>i#1=(z(K4Pa+Y=~X#5K9d#QaZ@+_0|^3MTH7C%AM15n0%E$K
zG|JUByJo#*JKbHs;sOWn!W{>h?qRw^`Sm?srQN*_Y+Yx0tqmnOpdRiI@<f7Kd9(T>
zZ!cQK`UI~@j$n|=sq@mB;xOW)rGT93M1=xe+}BoK^oU)~EC~%JqBOYJOzRVTweU|F
zZYMk5;B(oWJk0e<EK&XvqriU6I=6eSYCF&EBn-E|-;nPC58xGcRuRRY^yd9wva8&w
zn(RIB-E8*SelNi};ZgWwdaU|!PlftRCh6M^k3?io-Fnry3}Z*upl^9CJ>xDXB^;&&
ze&m<uEA-8b59iPM12R`(s~gulGo~2S{MXzvdc-wFOt6#S4WM-`Z9DzESF45tG>>k&
z5x-kPKWoI-xqjmUtr@98Op37P%VoN$?l_BL$)C*tk|FO+y=#;XfyC4aS`K)Rg4RKU
z>bXG3|M0kb?6Rk1TGvvw8mz0puRVymVozYFv*hpn3GB12$p8V<Ot+iiXbBWKE~4~k
z?Nm=h)rt{(FYnvMMyIFNRcb@yA2>StK^4_H+)JLXf^Ce4{4b9}25FGOWH#Tu+m@z}
z1trXZvu6yLd<u>P6^;T+nnTVrHoBIz^Ppn7s>bR(pcoIH4{@OitF;xxd0kaQ0srEq
z<HyVNrk;04rx<Y=q_lCXUV)||66JcnXf#I-+2Rec$)@~XbklLlVF5FGSfwH`m?w$e
zP{f6=ck;;Q&=-sgvpC_5QIDNawE9CrPvcHdZ&Jf86XulOmxc0uU{bACdXB3EDm?rE
z&mRfbe%IX^^dCcO_8*9vqn{p)k%JECsP!K%BxAocZ?c|5Mh-VeYF+{Exra|*!z&|8
z<FtNw%&(~Nbk5dVV}YOte|^wwosfVYfhEBWzuOCR>s`#G5R8a@L0L-PmjBuiSpN(C
zz{UE%m+b#Gg#X8Pp4MX8Axl0RuuEpkT}ti*$XV(UT16&tB}!|S9K7y&2-8soCw(UI
zIU%A5^SNVo0reuv;!J_y_;S5_xw_sPL=h23p73DFgBF64vlps>7fU-LJJ&#_7w;<W
zux7b-o@H2-2}vxTrSB>T#_!}J{??mMaACoskslK5^~Dim!Snn4Jj=Cv_k6zS>J!ZK
zt9K~3?AsW8J6jL;N1XE#o0pMEilCia@nNtWs4aUqjq~Q@d4Gur(!r@0`lr`n%}|iB
z#|B>C$CqGh3IgyP<(s8~%`xQSzkO>Huh3F|dmH0WH-&U(xxithvqTWX>0~f5L2!^M
z7i5qjN87O}DD?j%EimQ#)<KBEm*!s{5?21<I=MB<3nA{LJBz>{ZARub%YmY(UV9uI
z?|KA_+tuNYukUr>sVuO`j_o(ssKI!`{YWaRjeELWOGEXojP1C@4BxR2Bxulug}DrC
zK_U+g+Ag-K_hy@&YE<8M?oX;`-tH~eJm$2;DG#F_zlEW-|HqiLHJLoF93AD=RZej;
zn6lO9S5(%!dMr^-osGJv?l)G~&}`@L0cwTSavr39sc8-BHcC%-|HFA$Sdc|E!1TQE
zd#v3*n*dnf1as(7#5Q>xTbU`MH7R5J+^u8T8J=Oa_64hYi$Wz!T%54AM&IyPG@7_r
zx3z|jT&wh96jRNrs-50=Tc?us;NdItiJRUS;th{K)+%|tBR;cV*7m|K0}BOA-u4y9
zBuCHa!ZZux=-YIW7rL>xOVfB0pmlquY49o(&Rq-dpkT?K^TRn8!$T|h>QrokiIHOc
zrJ84yIJL@-TqkzXxq-?dJof9Vc0LFVNsQ|dx64U#f$1a9AOs{hd?#L<ypTPy8c{}v
zt?Nao0L)U5oR4{|qdue?2U4bI8T9&a4&h>o;rf`R=V8X`Pqz@E3eKKkO#l9J4ri`<
zz~ddN1Sc2r6S-bdXt-qriZhpsA2vquj(`Eu74Z3v|Gv*hSYL|avk}D77=a`_6!A+^
z+D^Tzpkl8Vz^cImZ^_{O$pQ^=cmYM;r0~Yl8gnbLK{mf0*xMAg%a=FsuV>HG%+6=y
z*k6KHpUom`E~bQ}+?6}}mB@cj{T2KK2No#XC1E9(DbV1oupOfLiXdieU94o}K`gZt
zagLlLA-Jk;x`i-1{Bm`7+s)0mNneh-aF-$HECF?afwb;M_O{9HTHGT$G~m_0GJ(j^
zzL&!W^nkn7Ljg}1CjK%&mL3DW`%EK=E=StHq-QTp@5X_=fCP^mK7N^0ctwac4nw20
z4SP3E75ynt{9T*{2C!+8#zQ(yZstBR6vtvD79iT!ODWzU@b}VOQ01ymrw*Mh`V2|k
zjneIs0Phy4H~SW-R>++S7{T;K2WuJyzD)(m)`pJo4LA>0X0+FQ;W2;m@>&v-)ph84
z7O)$Z!gkA0IvTKYu+bnQxHYZS8cSP?KHokW1Vv<84SVM1B%fh5Bfn9WaD&P3<S?Bk
zzOC%hzqDOS1zB*vk&177sw5@COdmGlex;Fg2Y`-vs{Tqd_5p3&rh~t(<wIzbK`&HP
z)uUaX($8Qri0W*$fH}6StLe84nQhftFE@`~z90(BaY>4ZZ3e4%LYV&?fJ?BEE)pY*
zI%DeO9A>%8;wtcEMtJ?PbUfgOQy|ELy%_^WOs8H9mFQwMm(n}R%6qg{58r7wFiGCQ
zOIzvB?Mqf0q=xGt$}!2hZ|0r|*YptOI?-I``7nWv=3_5tCx4p@^PNo80znX45_AKl
zN6}oS5#>@$_X-1wtq6-S`9%!Fx;l1@y*H9pp5a3^s-P4iZsXstCjcsw+^o=0Yk+Xz
z$X=_dpQuj~+@I}=1srXXfk#p&xs2x0yDxzKYXUZfF4#$oX+;{7qn#H2ng!9ri;Dr4
z6l^oWwc%;MnY9PXe|4*kXFE$xZNc@i)x+SCD-V2?W1WS8ver0r=90_+^pXW8RF8bq
zC|z6-$2}9QE;?l3dQ$Y{VL;qASXMQ|Y_WUXJwQb1`gi#3-+@`;I_jpqV!)!Du}@~B
zG@C&v5#Kp$LbNgNB<2rXikQC%uYy#yDP!Fg5NVCfkCmx0<o;ev@2QwM$|nZ2;%6&r
ziw$xz^L`_zh0aTy%f^SNS^YfB30$0`67JCc3Uk<ya*^jzq<#R?nl-#;@ZWJli@{;n
zC4r}}DAEUFQ&DCgDJQrnpJ#@DO&0TrK+!K}?AEX_t2>kmS`nd+S{E<!THAG6E=48&
z=b>Q=P|<!P&nHutZ9~(aQfSDf2~=O|UD4O-9EyGBsR7H=<@}z{XHa(5;cs-#rkeid
zN#I;~BW4z7<t>x9Yc{?<KY0i39U-6DUBJcCYjdf?C0*W_zCgNh&*kE0!Ls}UqWl6)
zu|}_}NH1XZKZuXqzWDz+eE%y?#QDF^yD$I0N8kE#B~}c%?pMdMaJB|rkECX|<!07e
z#AdMd9HkT)m85OMU2u#9DVem-^Q9i7JrJ!+hQJ$LqTAY@{*Ujyu7QDhYIqbzV(D=a
z2c~DwztYG_LQ)P)_YA(H5kx(`aG*c`+)$syTS!a60KT)QOt{acSwe+xt~c}me?i>O
zz!pZz8%Tbf55e5cGLL*gqvJ*1;{DG{{+Ld@Pfu4GPZwumX(G{j_)`SZ?K7hK69}@%
zj1+@5y1Jf^teJeCb2G==Rp2md(0K2h<Ukcgv6x}HyWh#6^rwZP-wl-1#?qIm09?Y)
zdo@x`1{DP73^Jmm!-BM<tr{HpZY&T?=eA#J2~2TJ&eqyx9yM!0w;<qTK!<mzsyT(^
z+?=^y9*?E?J-Y%oMT8JR`RHzv;mG9<gtb(+?<vR%BULXnxZ4^t%soNStCP4|vN|Yp
zZmxh(D>}om+2Z0nWf`aefK+Ug$dd-E@s%#!H=J2NLW6?cgs_yZySqnBcGBjYfmKgB
z{YX`+_ss9>97h?D69o0Y(9}eo4V)Q)(3awLv!G|IoZ6)}g@SUl;0R6-2CHj-vzAD3
zD+XcGIX$Rq9K4q_wHW_3$aDrIV~}2|4FK6vr{Ji&*HWixwpw%Ln*Ta7JG8Damg7E_
z85=5w{{SbAfo>M8sl4pW2vi6A7%-8H?TtweWxb%)nh~?wnm4bZXyT!*eadH}^=5C=
zPaFMh?PQmR=A&rmm5grKz=O}5D1vWegJii`b>nM<ObPU2cenuSKDDYFH0sT$H)rbJ
z0p7T2AR$K9T|C|S$6h^v)ITK&362#%+yxhS584%!=FmD%F`f2pqbiEe(xn%R{FaF<
z#{)Y^w1hpb8;<mH40Q;~PAcf{y;u_0MlAK%^B)JIsAegY4Hj>vBCdLWG^QqbBYWZb
z;*@!KbH(rzGBRdEb25gg(#jLB9!}5ufq{!nuRLUt?<8WgI{HwR@Z0QkDabZd)W?YJ
z`7|=Q0wiW=JZ^Y6YG#6W*8aoc{J0ZQbMgv?5<$QyPQ@6Ev}kg&J^OZS-K!xe`96@)
zF#3rT<aRgscG<oR{GwN>`#u+DiWE^%+H~OICQ{d#H53&@Y>lx)wQe+9z5^3M_?}qY
z-S0g>OC*SpOESY#pM|aJhWo9N7>^}T5gc5C)C651kdvc4-bJ<~<!TY|Wtbw;lx<p9
zWE8U4!c5~75hPPf7|zP#LX<p40c~<K18a}7&%punS_`marc*OBadIqlW*}av&{?Wo
z%}JBCeG=4{<sY2Rni^nY9njQNcB<B6m1FKWXO!gEjZR~4s@%~TkcNz8s>YPgeGcJ1
z;%oRmU{{lISL)wivhgZXMH&DPGDk-Rup=bLpF$u!3kfWIGJDtjNhrHH^eNM{*Cs8G
zdE8G5eJWd}wzK+~o7*uw=<Q+|%2vP3jn<{Jwubp_$EOz80}Z)wNDU5aF{m|ege!|S
z8|WOUiuwSK9k+24U?9qkL(EOiq-=&lvN&@~nI_7PAovMeGVjDkL}dg8%&w^dVp28t
zPc6OM$?dW^E;6<l_YD>JMLv4vP)qLVnp7OP>YzznSSRZi)MTrY8%2E9^@cT5s`MDJ
zeNmH@?RPpd6&+iaF)(-k)Gs@yU{mHC47V)w?20l#sx)2OI~f@=vj<-5T3j`pPzFWq
z;y#Bsh|ura<=gN-+Qe%ZesVxY6gFWdeIf@lV&OMhKvYO9V;(#-Y{Ajt0x@uDU0H_9
z9hEmq_CaYJagEnz7%-vQ`!ah&C?WD!K`UPg=oenvdjWo&O{fe{&mOs&p(0Ph4xi04
z9;S#&<6fwVwglJxg1Kq#g$|3*aQu2#kbdlR9)^4r^`;b_SQ;<}WAJ1OQbe3PC4rRt
zbhkh6$}rb)p%Rx*e7n_&hZ7RxKKGX+s06LAvf>o)yZp8-tZ!cy*B$n~{eu!ObeEs>
zBo;n{UKfBy&z-A(#(1m8Te}??7*gZkC);9`ayZ@Rd1o7>f9%Y2AUuIVBi76BuJgV}
z$4Fmna59)f2z?mguex&u&D)J)aF@M;#>${x`1HRN?Z7K*8AS!bWm_~%IbjEEhFMb@
z(TN|IrK+59EK(YkwmKzV5IJ%oSf~hgRzxdlfDA7dios*?=_7blSXswx51ML>U;E=u
zY%OLocK$@T?zVdU6fe6AkKnr$Y$w3VY<XbBvMpsZuy|~PnA1;k*W_qsl^Vyx-LY*3
zg1Vr~S8)2ZkAFF@Ct4)^q<EkNuFuCTw>?S+oPBZ2Y@yjc5ueNz{)<V-&GsLedKM-Y
z&i`%Jc3b=6f4abauKR*_WoFm7_@r-H-W=_&YiGzfE2<QD4?S7f#Mkj;OJo+(a{GE=
z>H0OP^skS|(*2R|HXz*tfZR3N+TpEkzmMvolQnNjpiq*;_+T)qgv5f9kpGF(XmdyO
zd#QW_#$LatxOuaBx&<;57|%8FfRB%CLJS8+#o3M`;?tu?J#n_rZ^PZ(Q5HSD&)aR?
zoX*$V>CGQKU(fg+PQu=S(c^5yn{ryA6KFRb1+@~gc_sx=ix%@NrcAJCc&elwwfHye
zdJ~rJiKfTGiQH#8Xg0@)(EWZq?&5z^a*U54(Ij809yId6Oiq=JC^?FWep>UlwSM_5
zms*YEU*_43O8Xgvvms%^8H#SgI<8BqA~w1-#px&GtM-LZ73LGAd=~tZ$#6Q4U#%V;
zi>nvZL*@+=lVyKoayq=MVoMa*O|2^<e%Bp$RNt3<%Gz{~LV(Ogj}uZjF0NgJ<?I)g
z4c^%^J6jX1&@<Fagx$Fg!y<vdqQrz_lGjdFHr6bb<+9fRaluURsSwu1AYTFYwq|o#
zLxDm3X9nS70J-F5BJC-+$YM@Ax}YhE(pnny91<Ks0`{usGbc<OEYOd!^G-#FZQ=3@
z*%(>~evqnyD&m_k+mbvz11>78gB9+U;21;Hc8#vND?)=YD2YOSV!5a_Fitm^O8Z{f
zhe_yEj(<Hzf`?<`N3Kt%FIvp%IWTPHOcR0bFYiF$yP=K82^e3Cau?i8`++GM_5$y+
zD|<wX=<M!Cua%#DY@+fXm`PdLfqO$Y6(~{M)&udQt~Q8!sM6n1IbxS@oR@}#`8X^_
zPFNfsyj4?H>b{T*4M30CuoRvkgA3+SIQX6db>+EYw72~8TiuKmUWFgBwj-eaY%yQ|
zp>tG8`RjPjq|GIl6;@8y&3L~t-445Qc#nN{y&x&rT1D|{n`4YO=Z1I+$z;~!&j8T1
z`P#>?lad2Q((BI<YH4#*w%CE)(vI^o8<bM6&7lQMA+i^L{0^7zoCrALBk)csmYkR6
z(>ra?+<#hrb$i{eU8EDdNYUt7kpX?gCRIMV-p>t}{N*Ke>*p2iJfc*aqxT1+4%g?R
z!qMq;RAbp&|0-s{ma{6&zNn+hJr1#%F9gK*RKi2uZ`YIJ3GIqRIzs6e<3{c<K^<;5
zad13`v<bq|qJtmqWp=0UPo|m!Cgl(RnHpv&izjd~mH{=sSvD#w0}1$rTYo!=V;h(o
zLAAp7U58F*U?>WOQ4Zy8F&Ij$1cSa6yH|97JB|T-1i@uu%Wof?FdioF<0&a&zfb00
zyAa8VvRzv86|(|C8xZJsDMRcKxY>1hl<2^bOzs77i&jSq!<Hw)moL&|jOSIF<B3Qt
zNl!M_T<rUyZu(dL&WQWW8fZSYN`n4jkUk_&$mM<RY42p(_g3*=BAcs{#3%0$_|vxC
zIB$CZ-cI*{;8<<kPnJ)<+!@>b)d&bgF5|y2ciE;F;Ev=kd%p-Y#tNeP=qTdGj&4F|
zm_^Wean$?1VxK@fdbu`qKO8&uoBFqINfRdGL{9E}jsr$Fonvgcf%lmUCxe!R5}}fy
zp++N<n12Uwtf8s=exdl6ASf<OsGMwz%D3aQD%<F49?{V0T;8mt*2zxC$<jA=X6^zK
z^mDb=Hf1o+qG({>%?2ES%aV<~S-T^F>kWs(GFQ9VE)0&IC7I+fWJWILHylgqT{BNL
z*%BZCvkU`@weU4I`7=kU*30Zw-NF4k3w9zvBp5{5_{U(dH;0YaufpR|A!o>?GZgWZ
zeyb1<AyN<mPFWI-Gxsp`JZmAKqo?tjolY|3+@O=42Yi<lpYWVWWbyCh`}#91*XKn&
z*v-mLoUIWsWyO}~v_WgE>)3^={6-s^&lciIWH)`!x{`glt2l(VI+t59Ua2FSO7F{l
zKbE=Lp0a$elWZZS$Yoaz;}=Iumn|)K^$%(ryrQu_p#$fm&j18)OB)^?*QX|Y8<7h3
z1@)(EwD2o8p`ba921k9?d~Ih{l06%52D?B?gJ~QIgtD75H_$@5>F>ocbVC;086h9l
zvaDuCQ?hBlbqoj@g30C-l9=(aq=JKN*QF3`qUPiWPmH@p4!~UJo4xm_wmVLgoH#+|
zfbS{KqvEMx52UA#Go~t9T1q<rn(pjRS^KXDynaNyd=YZgB#)1HJUMii>Ty<C+h}IX
z-E}`V>)?9EDHQ(mD>7#(_{;qZ(*#WA4?ugtR_tguG)FEvP|)p{&1<R+W@3`8pKBM@
zI-F&VZOo<J+|Pk8RFV8=TYfI~Gs*O<*}CC}gPdid>oKs@*Nh;XT8h^(L+aKKwnE`f
z3MN;T3qJzvieY`T3l(CSu2ZuTFzp6943G)~G7GNqntff;*(5Hm*p9Y*6$pg?A|Q{o
zUO5f<g1H|rjMzlUK^-K9r865(tQzA+q;#&!EFl@I5LG6I$gzr1R)hTX{id&-P<`^o
zH#)3cx%(7bc0F_nHqX%kk|^>S&SQ(nokNc^T`U1Ioc*8GADczgX<q!?u*HoAk`Lxf
zD`b>0ch=VMAfp))0j}oqqJtoSwV|*dL|-R2_M_W7cpd~p_sUoCbjT6e{lK39ALiWY
zm&J6smw?mIFLGFA9ul^=^0#+CZS$A4Y7JW3>Znd9>h2GR1-1{lyW{(pGtAnvA7JtE
zqq6@J)@=Vxmib=;ji>(=iEeVB{GUal#bq2ZyJXvEqS+go9$7p33LdN2tBZwH01MK@
z3MuWGfB;xJAs9H*F}Hu2Q!tRVk!zENUC6n=6h-s9@x9rExdu#G;@A^xtMk6-(D>99
z(5W<8jKx(ME+i9|V%9i)-?NJU!cGs90pG3yy#LPL>4E_P-;-~BcZUJ#4}|qkyAPTV
z*s_NK-A<yr-=F*(1_A$kxei}&mDLAAMaB{?Jv|zcRESMSyWC7^$)!wL=`_dEqPw4G
ziubj-Y(ZK#0m1jfh=E2nlSHIY77XgFM6$@J>BKV(oq<O-lO(8Au}u~o{I84I4J=!z
z$s#n9qOe^gEs=6)2=e?~E15%wTL+ggWD+#F!Tgkc6Jf+^;6@TCO4l5Iah<r*_<SO}
z8g*0nRYP8Mi$&rX2|xvux6sHLc=|mN12q}7?QhxCMb((G%}(kalP2w&`uQ|*RmO%~
zhqgB#TZSrm?KXAQfH*vtg<@ZeQ0IIRo+NBimWiS_2)N#AOUht?=e%p%8ep@vvtppE
zmj5?F+Q_zDq-gebXVp&4wx&Acq>rtRhwi@_g)MuK`iA6Wzf5~mhkf(qwt?9-b9V8)
zl7$sBEzdI8EaH;B49NMb-oYMVM4Drn8A$yaZ{S#QNa8S=hBcD;G#{fQIJH-oa&LEG
zHa%rZ2Vy+>p*n_5sS^h`QCDf>0kQ{Gzmx~;9;1)`?H3ks>0LC8_dbq3pA+^TKi~dU
z^T6gHsu;`2p#d4RkW~^@*saK+dL?jPxFFUf;31lRs}`PjYkFFl<qCa}LhiRrw$h7Q
z##+Pz9Gh|*v%gkVc3tVMxy)%rd=1C8Q`qLA-el8I-aX5i;5;^y3ZY^N%?v%FG3~v5
zDU9!Q`dUXnmjpA2%27K9LQ*$;S}~WeQfZ5hSd}V`I#7sHS#N+WB5@oJJvYQ!tWoe@
zFs3ocludQX{6AShKTQn%CVb5e%6+h(0)^rhSl!wWaBzlpC?x8D1)0nB43fN>AVFAk
zL!&PObVIl(F&gPKlacoUbskVy61M`=Im=z6AyX(eA2{o;gaJItz+Y)yMS7jeoK?RN
z@8EF6n|AEZ6J=hHAX>sZI96kFL$;xXf>*3wDQ;Sqd+66#4AS55)7U2t$_)BA>3^*y
zoz<7I`+Ra~&HL(bwdjcX>aK`KZX-13*sg<sb-mdYY~Z(vjX*qMx%{z2+mn0dTEL!d
zrY6SY*VX$H5DD!b<J$k$4n*qfG7x}95<$tVKq4Zl*FIQ<Bm#>-N<yu~Pne$f9XW8b
zC-&(FyR2Qg(BHIsupcGjTeGwZ$KUL>Y1KF2?eLIS_prnwAVMJ!WolFy98kDkv3=(9
zZM1T)7e-LtHsEd)84>>Ea&*b$LfS!3i(5nUCAx+tk2ETz6GUBXE;NMCu*`}niF4}k
zGxtK$ib7aZxB%vObGq@3<>yaPfVBmx(KpNL92StHy|r5(E46{JBXuXy5-CBlvdF)n
zSWs}#)zP(eh4jpi(1D;HQq31H{>+ZR9+ouo_x8G>lDJUFYcQJMO3kZv91lnnkki!Y
zs*%TS7ge}VGGPl?i(^b}a3l}pjbco0xK-;%pkSQQLNXFyH5NV_vvVQpwtMdLgE%1*
z7|e-E;N~8Swq&AwqgyHJ1=<-g2<}+_#w1aq!v)YlPN@;lB2uBH15_d1Ja@0^4$82Z
zkj`K_dB~I$7J;jjrMT3#M8-Pt&fw}^+@Xk$hGh9q9Yya8j$g6x|0cfO&caS5GxNxa
z0gg{vdK_7_=-UaZIfmYbQSw-whBD34Mlo;M=@1$<7KiP06GOApf!7#`=cBuPduaK6
zu&BUe9V&U_ZKGdQ1JfwPUfOOCDxLApMX%9!Mn^Ity*egVA^RJNBNdFKf%zkx@if2f
zx6vH%^MtXp%(selclyY<24l$Jq%`NBs5*b)y{}JG1KdHP<hK*WlwxUn-8ZfF1N71P
zjW9aGPuw=!=iV@bondM1_zu@v21p0RT)CH@fd@xCjJrRfWTkwv>Nl2G*n`g;_x&ym
zNkRUeDbm!+wE)vbG;-%?0|w4Q9j|J)GP*Q#yes9l1;QCC106(bEJDkzi5MV?Vl@wh
z)_C4%)!952Q$A)TxQC2UKD@K6KzJV4Yx`j-u;x58e?1Fz-H}()cS*L>LuSB+V#d>D
zl!Pi{15yGmo!dwnr@r|S<<{tZ#hGS(lo$w@BFypVObVYrk)yf7IuSL~h5lUK+noYT
zP)@&eOKNn>yh^uN4cxCzi1L5)xFNQs)>fzYwD$|@j%4}qmd)<L8b@}~yqqQy5=^d;
zUwpcAdToxA&Fev&?j@_5-FG00Bt5%L8s-?~(@=$#_PGpL$gdT#CLyppIF-(z^nGF%
ztENlRN2*ABdhiXSb@Q;Sm9s5>pKA%EK*@kUsTD$*EjS=ehZEX14I6xe3L7#R!K$6u
z6bn<t&VlBZ4W_P{0>gJYe_rkq7fd3+i|{W{zr-=af&VQf_Fzb0BPf&4LP>4k{7n9Z
znM<cLg-QP$DNYN$142oY3j1nIa<2?h*(>2^#I9%ljrb=yn)5#=KG*-h!hnT|>3?k(
zOl#XYY;vJ|=NT5pvlun`F3WcD6M1Ytd#}+yHwqX5b81m>pu#4WswEwoMSs8G{*uy8
z$!(TySo<<jA+zN2eVuZYTHK>480bn4m{Bz7P9>3&#<`=)FoYn1Nr{0<rOBuZ79Tg|
z;m-d#9nqCo9H^%vMpi8-#(k^{$p2;uaN)>w5*MABFe@X$`Tp$7r)YpXy$N{xhdu21
zxjoF@-3jmvH!LUg9|o+Sg?k(#Go3=WXvE2xg|Z;vvl#^`iIMXu(Gkm)4fuktT;3M}
z*|8wtZtEOs<B$~)q2z+l;;9a*9l#nE^oO91LKn`RPyAKfoSIqFD2o|%uW8uVVBD<q
zpMt@F78$cJgZ*a7Qm{ZSE#&%%lFwl#IZRYmwNY}<P1kIVlMpXtc$7%50k4=oZzW7_
z;W8Vj0dW$zDwQ>j$m|R!q$rIdQzAj}sA15|Wf@TSgr>cr$)pBKtz;L<Ogl%B4HM@#
zcG9}S5ygkjVNP?$X+3V7uDWLi>PP_#^_9)K8o6<6;f>Qn)B|}ejAZPnI=_IhbjZb)
zZ=G)s)pzeR%oY!U-%F?<ih3Z3<I&}DBn1@>_e^BGQozFES-V>tG?L^z>>Yl7YK{_z
zih#xmbAVsk-q|tF&K6N56sJoB+gfj026;}T$<jV>k`&UV`D<plUnGTsQa-hs%Zaa+
zQX<V0NMjrd`kseK9VB1|h^v*!N-b}9etm@qN1-+*eOfOvoSJ;i5QMigm6lV>dBrKS
zTC&e&T(y$cU|U(33B#%OKma?IUp~Tj>0)%HL;|sNk8aZc%Nl`%TL4PTAx*BZ!<`fv
zkBzaCDA!_&lmZiNu5Qao{+dWAWGqP}RU&g|-!KpZO9Ix$oQHZ03l@!C474(<tqZD&
zyVNte$-ZLWh{IR3+;TMmNX_mAV*<2#0Uemt3+18MyU)<wu3o_dS-gDoZkU+oZf}%T
zQzk6-bG_1G_TmIBI#Ym1L`q>jL=H%scnYGuNm2hqHOM(90ZdijhS21&r32PRi99Va
zgQzBMJ6&z<SK}D3Y6mHxrrhEZLZnWgFhT{{Mu9#%*IG#hD+)9?8EBNb`?JLBK|ER?
zdCYm%q*0}?!g950U-Uwdg%0g3GU4*Ib2;OXI_x%&<U@oJa5S}f(`>8%m_{ad4Ls@f
zJFy)I5?mz7pg<qTxs3@lnR{%U|Jv|Q3f(mCs>zjwTMf-7(?*%vb3NLF#6O1gy}<O&
z;e(M2>}kuK0ksXfi=nhn5Ni2a3Ax@Pv2gEo6^PNtT!|XE#;MkUT$b5A8_gW$HyA_7
z;c{POh{K|wL+k{GGeL?zNFZFLfX4j7i5rm;Hv?<|HDyXJmn1}!DYi=UEmA0Emfrf>
zP4ELT2C60@qf_qz|680={(@czi*54et%sU7=${e$yfnO$o|+U#Yt$7`ya+ILRd-bv
z!Y(}Zev!CzhfD;K`X6nF+wyH$rHpczHDL`#$xjm3QQ?w0#ITIvc;{C5U8iLyQ6&<A
zQeuHoZ8dbxTw^xyZp}7KO+JSgZJfQcH}UV`+>(DAe`tEh*D1f^7<OSdIKr%@Q+vC>
z+%YwfFh}{Gga}ie=)4L?*GK&kys~sE5XJlZuSK4&uUiC1$niJbsOIYdcpPykUf~*C
zWyS3wq{>t%p<fj}VFGhD9pi`AD^-<-xAq#uWOemj#&HrKwoY6E+ldVLVPh{0!8wo5
zi(CcgL$V^J9~T58XgHYn5D+>YCuNzw7>t1Yvz!S-hz?{r<&R%&V%#@a5hHlcz9cl-
zz0S;KX=(`MjjABmfhQVrt;$v&@U=g|k!1`F%7|a}yRGndC-{II#j_mtaFU+|&v-Et
zNU#P>!oK35Qq`*S3Vhty!iuzFKwZ-|ZI!e*5eb~{(yxDSo#NS7VAG;}g2$i#680p5
zBy+6!*Wu%fh#J3pYH7r4X3i=lN=oeW<2+XK&^&3U><M|1NGtR!4xuf-T+)}DLOQ^p
zleJ?~+YW^}?(XX{`&a4JBc^yux4?P*+`e;y?szhbl<oJw!+ttT<-c!2v2E$X$Dc8M
z7((=AEBFbKPRF#MrTO9)Pr2yB!GP4fe0N;V))i<+7Zxd{4zZsSHv8UBmSdet!@Xuq
z3v{gB=``m**Hwq~L2LJV(o_BYAK?Gs>0R=N*}|<%-bB@JXO9v&Wl1|??~8QR7n&I_
z6o`Kx<?=H<`*tc;<rT&g{%un8;^(5mc<YdNmjlu~ON0_Q89PhHn(fWL`ve~C^RFb~
z(Re(+spq>pWw>FK1sTpy1M81qTi^XAm1-_VqiAJ}cwo<4fdhHw*TAXa)S|t=fF7#O
z3R|wJt-B)xdL4Ah8Q{J?!rjh-T&MP5mqc43e9iuqU(|FflfEH9`<OJnRIxNe;t?8?
zKWa*{T`ZgYQz)!7Trr7iDRO7*^1@@X5pyt>$C(-^+wj=jCA%@CxSeo{LYzUltPka+
z@HsoYH02^@DXwljvZGx4?yd?mo~6-ho=R#q+1j4Zl$B-UIjsI_(>oD+OU$GqEYvp3
zQ)5s&`MId!;c!i-XHi<AC0by-nNc#;?5dhguy$o<u(omqAGi7FkXnXr7%k`1npIOp
z-fB__$=SSeS(H#;vF)PCIU{2sd2^Vc{zaG)qN4U+ivs(9^n^2WbNpXth5rSs=l@{r
zo!TpFxF+1#wr$%hwr$&XI<{@wcG9t(bZpz~IJ@_EvJdui@XnvGu4}FuV~(o3)Z*k2
z2LzGAZokl7wg9fRt=+9-V!q_k8&Y_~ftFU1v}pc>*eLRIyqO*kF(Cj+EN}pv_E4Ma
zUP$vK2*!M)VGFu4Za;oahVmd30nR^1%A+K!Eaey1VIfNrJtbFJ%bq-Q{<~I;)BF0n
zsbHai0?V)6cQDr8Q7fAY-adJ|5Vn_R6zbxaZnW{abR<96!YI*Ittxzu<+_esNhQqT
z-_{Ifex>`G%kfdhDyZ$;_qqmCWxr1k=AO)A22lO0MFalxSNV^i05d1||7-&+)in6O
zwvxH>`At(gb4jHVGBh9RHuufV4LBdm=%6%9od~+lG&D}Emu-)uT+<1s#TK!E2)?AT
z?%l8bT(1Ub&%5`id|!hd{q5hAqC_Yd(CCoumx32W@~mY|WF7QD9!6yA2SeBShlzXJ
z%6D8aKlax~#D5&$s(t6YR5y(qW6%%{p51m*<+i(2IO~nE|8)l>y8j0MH!1B0LX{Zf
zFvO*r^MqErJ1$?h*<#9I-!tuI`9gkv4g2kAl+{>SUB5AkuEn4wkV0qU&%*r@#dx`P
z{N;dny)&5U6CGl*ZU;GQ-d=C}0a8W-ts}$RRqOU=?g(7-AY($bs3x_NvuC|OkP5v}
zVz3SwS?5$#pA!-fE9<Y?u{fVAN-9G87FWZXF*dBzU`vMK#T<MFNqd!b+&HC!`-h*G
zlqNQF9Sd3;tE=`X%@xky<^9P6ykYi_wtCsqI+v6tf)^~Z7PFp8x+Lpn+?Qv!_5JZB
z@Vc~TUl-1r&8pTh`at1yeorm#ss1xcBe8#gjN*nHkdv5<Fo!8UtxS>U1|;FI)!Bh1
zIJgW1!-}<FrS&<<-4C1WEet?R)BGR5@~q9PC67n+o9<@zKhY<ehcJVCmfc~l9RaKg
zScb7hf8Bs*SAcQMkM9o>F{>)=wp(Y%rSQI+C5!G9-2<p^*4V)d{IS|FT^k(6@I&n$
zMLIng;50o^vi2MKSX5*q(2g^OOPd>>4CaSgN}~}Z9%%;ulJP9GmyK>iKzEnf(?sO3
zjB)RGb7$nT%7UEqw!4<-BFfSs^mlgck2Z-lQnHjI%Xr1GIMjAB=B6=Z0(tTV%wmWj
z1PPR>I<A2ozRq0I=9{CGq3kchl`ztq5oKbe3=h+V!W+o?2AN5V-E1Wx1ZB`gH#7f4
z!eS>p-jVlb6z5!8PypuHA^b#8-g1o(DOuoz^WeOoBjJy_o`=yA%5QAVcvXn+?KttW
zGr-cB7CXs<L%x-R0?qP;=rt!Or6j2P!x|EmH%j41$t}(kjHph{)}g>>>CP7TUTqB3
zQP96YAaHdyK`970zm~GQg2(R~VII|vQ|4~ZRD%ujA9)T8?^sp#wt7S-RcjRIJN4zS
zFyuO#wik?10wDL+_6!|>WMO+ZlMox|0=JNI{BuS8jG#XsKji=M%AU`ymqb_?`gP1E
zMe&Y!31O9K*YXA*@mcE@_Y8xdmkoAlkC5<XB?}}nVHqRC%~Lp4IYo&iVeIS3m*ErA
zGn|`^(|pPrT^^+)Y@Ma0(oCu8-dKN7Mh!5DG==e;G9hG;nW2MTA0c6doG_|zNS+1r
z?vZQy{N~X8zX1m3?{Dw89SK<J+<*W{<Nfg6fPg)^z&jrdAoFGKZ3Di1fIr3Q$IKW9
zXR2=?;>YKbBO}e<DPyG3^*z?vv}jZa7lI?nLwHt-xS{?tu-&<4T)w%*19<AW?V<gj
zQjpYqCHa@ASu})v#9c2}KH=s|#OLHhV$`yM-=8Q6oY69CjE5;B2XtiLL*%D)?}IQ@
z#cY!FdNg}rV=+OK!wOU`A_q(y{*?Y;luchpzK)!-JK^O8sQo0MZhfTHP?AQ8^K-Ja
zCX#@rThem>lpxvG-7Fj3koK{h&mU@&FG=GcYgP!VCbroy^Xa|@=a+&pO%V{c8UeyU
z1wOIq!tbodoWMKRN;F_78u^pcdcEuF#!W|IxIr`y5AYURU`ueDP_C^kA7#*dSEBtP
zhpOYA%J*pYmVhjMDH^Pp-M(`@{0p86QN{LTs4AXcREL9-dD~3H!uiWWgawOKpT$F=
zNgcv=;yS)wM-cP@F~!AR8G4NuQ&l=?&tA2l%l>GQUeEEV(~alV##C;1(VSgM+)Dr*
z?JNxQs5w~Hu|n&IZm*WPYIOXPEU^G$RhnD=y8@sln-4O+7l}SANCFbA6T%-FPgt{P
zMw)&?UAbgW|A^LlrsNnY2Fg^v<qs{8w+V81!{T#ua4FhgQXJ*|r9|{6Er{iqFkvq>
zAvY@z$zp9SF_FcEgPIaXHql&MuSdRlc@a2`A#^hxEfv8-_gFAkUEf2{!=m=#0&I``
z4F;}7gkCN;@!HIlB?FG)%$z;d!n~;O+-_(d*<JbJOj>>3tt0<%pvlP-h#G!&cRrhA
zKb&f#h#Y6Ge=ib`+D~z)2W&w$yStl;c?iK9x|DN>Uwbn<F1l*@bh6-$+Pot)ck8My
zT~}QY(;c1)+C0j~y)z+&y%&kR5v6pFY3FFHgr#M>1#TN6SAl_>><&BRZge+jIU)AR
z)^<F&#Q+witfkvqIWVEW+j=W%fw)bVL6ZC?&xP0qihbGW8XCB?t96pjODKg$o-u00
z0x1<rflo9VxLS(nY2&9rB@trre~Bu`e~cO|Ox$e$tA(Lfb2t9~y+Z8v39qInN3)Yg
zB~h`T0oys^mLSqItn-zI2mI63S~_l&)@-|lpRcf-LVBxQxrEE;Hy^f~x0CF`Vna;P
zfvIXjhx)wjy9y{6byFvdEql;JaI2(7DyiU28c)p8Ag4C%axJv)*P(0;M)gas*um^}
zfxZsk=czdd^`D-v^L06rGL)ngF)Ii{L15)IISpjr$91^@%I;Oa51uqSMi#y&_xZ1=
z)BUbLed0=67chO@_?v&0^mz9i7hR8UF;(~IqV`Rc`au($VjhWN!KRGnu7OK_=Gm%S
z$1M)76gwTP@nm~#!`-55*Dt%q>^&4*<YO<FdG<+u@2(!_#U-qX+qr@|=j7#J3L!+9
z4^m&W#;~S2b89@-Y%g-Aa3>pftZOZI-LCuUzQ$@LpyiZ01^OTy8urIYVN@L|4IS(G
z7q{oOi{y+eYMHEuBGk$I#Mz6qUoHnCXu-k-pdy>?U+W)Uc}_$#2|O1G#B6&D)>7h3
z1&#2D#da41%5cIkGJ&aKR`@KKr@6tl=Q*LM^wch8++#Eo6%s(oQ3d@+T|SBQ#Y)us
zgRjZ=>y>b@1$<R|3gI`lY<doiq7N6-oUd?sy3W-qA(8s6Ds5t`a<L6fYxP6awyVB@
zEq1V$-*Z!Y+qUrnG&xAGHgZ3+WoyV$D2l*RA)9F)wsS9RTCJW)F<amLZjzG;e?v}K
zu-9VwJD2Nh(;lu0#8k1jHmd9MJh!vX5_EH()KKx66VB^wW9az&_`fR>UXIpWzK|ep
zd6<a6H7=}a&-4LN)VE^$pzvUI%nG%ixB#bvlQ3pb882;4A%*6QH~a$1!xfi+6tic%
zZR3L7kYBadJ>M9SF0L8Tt>?GfVBk<vTs~u`bo?M2GyOMT<zMbxir40<tO!H3_nt7L
z3gXQ2m-Q~*R(p5*Zjg1F_Uj8vV@Tc3$a$PH(Uy{`pF@cSdFWswCh1t8-{1!r?&-qP
zgTV2iHG7&zyP&2FT3Q2#U*-5%tl_S&<vHe$qfoFKCq>o|{MGVQL|g-=2_X{aXc>Ch
z_p4X<m7z>jVpZXI#@;o$Z_f7UmGlqC3wYV6)7y1(3sqKpkUEw!jbJmEnpDzzbX*E?
zGqyWBPh+|*7B*r{<Q{v*92s)xB72*hTJ_z1HozdwnOj%A3p%dMXtGKoZ3F<TMHhSe
zwMmTReN&y{DI4(%cqwD|$#q`V)g@FpTX0w+bZu3)5$H<O3NP-nX+N$#yAs6ggFU%!
zXoVdKLxD^XqyR%Lmr|ZdQInWhl1wNIDLI#CAp>CV7t*%ZaD;V2|5@O5>O6ek!AkXf
z4vHBhqFG?S$zTkyb2Q8<<HLqgY9|};5pmN8RHV>o1KW6mOC#aaNq7$#8I&<}P>bPv
zFcfE(3W}_u;s~%(s_C@(iz1(ya<H}ER_L$AA&INmPl2MnH4CC0<rV|Jag?=0&V`29
ztDZ?OgC@PEVdSUPjm)K~U5xsL5rRr|mNmuqKH&&YuX9=5L&+7S9VLwh4Fd%Ii>cn+
zS}N^vdLEs-7gDx?h0=-!j4pvx9u*-~fi6L&8e0mQf|<0nQN5cH#R=C-Wk3~xO78Y(
zKT?aJav2NU3{xuz;c|uSU@53lkKOS+zOgt^sSbxRo(mHhok;HzGGXm`vw)BSGIL;I
zqR1^dp8z^h1wfiwclvZhZLoyT2$JNcNq*$Xcixe$VobaUzlUj&oFn-FQf-qJCxE66
z6X6N2siv)4y`<C6)|L~FN($&i%NPrNDwP|BxNM}+#b?Sbz$7udi-Ef4XmOIPq7;QZ
zxt44Knqs23@O6WL1?5_^282cd2p(JH1Igy0r6sBG>+d$IdlR;hXw5)AP#693+M-28
zaGme2yN5k+J{t0UT=sOwI{OGUboB`)smed7&BHux-=OvMLL#08czoXFF)^1&&cPcC
zI9q)qRzEX-so^Hk)ozy-=Uo^NLzSKxmie$1_$WX2G<po-KJfrRl%e8)50d8}jC&kt
z?T#@JWkM|s?ctV3dSOw+0}M;!D|fObZ776?1a}_aLU427-Sc&)y{h-w^$~vNwa_OM
zR+c&Ch)~a@MDX|_ky9G^J3DB?O9V&_l?U#ED{-(CJ?_3VqYZvVnyW4gMhUfT=-8q;
z3zJ`}8*+73ZZc$0GGf_8S2~k)rj;_iQib+%(Qnz<bc4rQV!Vz^S33gGY9tM9*S7Fc
zaRGqGL-DU0%TQ7@Bl7@@SyhUS(L{ps5*XB0SQ-ZAembj>1oN&5<S(NdYAJ@dcuHix
z!%ErK5o1_y)HS_Y3zHXac#O!T-+RvH;n#6!Wmc`ti1PkdDRuHe*xwpMFwo0Z4YnV8
zca^29jC~JFeGc{fy16%uy6ov7m1Qhch#KzNK*q2I2$?KYQNj9<o7F$RYU7?5Rfp-g
zfz5^x(V}kVZBJT%JZ#e>Yy1!XajOXN7XRW*bAEZydbZCJ)v9HmeBBjta(qt|<^SxU
zKJXvpL$rOIVov$EqduCZfd>wq%yoNrzdG0_a<U?6zp7d?J|ccRBD24({w<LP(;v*w
zDW+-8TrIZ|jqcV;{Jibhz$r*vTt>OY`7wMwdKR5F=I$n-CIX75bMc5!3S5VKzsRmk
z&JFGzvPtl|D>V0MyK1~{#I?qqm~-j~weHKpe<R3tNHg!!&sDm^Gq{8noqU+|_ig<u
zD!u-Uc12;z|NB7tA)QuQE=|)5cI9wy&dU0ZdqTIL-?9IL{vM41b>`0syr%JA^9SdD
zStvQV{%4$isg_Lg<|uOSr3O*Cp~@6;z(8|OG>J@_E;)DN3gt*CPLPOm6H(ZpiN`V5
z4`6!{KqRO{J4Q_~zxa4Fwz@q;wFa(qH<aAB&FA&bk}5=A#WxY0ZAbJ-B;$fys^!=W
z%^xC_sQ=~sc(|DEq&gEeqD2w)J3wr!;<P^R>+58syIL(t%!p)UnP#f^?({F?PHnIM
z{loJn=B~=>RQ@i1pV!yo=HHH!<Ol^P_Yg@{^1EUp1XGF!8ib|}G$(CvPRO-O^LnR7
zI@@P?_|djCX_0R7o4xM%HqPMA#U;R*A5ri5n-AB9z2dc_9O+7zy2)HJJfgasLuA#m
z;;q|U^_!gavBlP{n+Li^p}EtDn&M!p^ZD@cE$E+E7Nx<}?c+cQ|LgZkG_$50q4IcP
zvHTST+E(kaAsly)9Ny&^95kXswqHHu1caP0Hk<Y_=~`$v%V80%E474;3)8A!#Tt<)
ziBqm#UH;-)PmZmV#Z;fPz9fbt@w>`pkyIso-KT=#9qlH~nKx0e^kpYP+ER(+&@;}`
zK?_X>yi*UlqVu8BBBkZfF-?bY78%uCc%ERlFh^PGoww6WT-3miP77pa#8DNR1lK{9
zXOEi>PwZb~>aF%zlUE+&wUZ$+N6xtWqG-`nT3w3Tqa=}&ov?x7v%kQ!A>8AWR^YbG
z4@dLzdaJCidYTa};fU%1Yrl|-;FJnF;Z_g(leI$Z(_D)9NQ{h6^;g^da(!5@yhgs-
ze^buLD5+Qe4UI3`mPhF(s?EfkR3>{{2ia__UBK<vi=rZL94~_TP$&+HP9<)~w;TA>
zoti6J86}|_AH>FKoSwflmyx9;{9@E2oi1?_1{uuTA$T&Ko4AV7Vqj5U8r^CC5|mJa
z93;5g_v1_G=3Vm<H3rHesXB6=aeHk1S%Rl(S9^%ad}AedLOAQzG<ldzQH6P3&CuEY
zW$tM@eTDX(A|W{<S<9Pb;h|<q99)lc=F`Wy*Kd&J_^(m8oVuc+02Qy+N+|)l7#F3D
zl{eTL7;DtA;fzfV^uB0c+y~E#1N!fxWyD!z_X23ev=;=&t(`}-5Iyx!K9BJ^FBs^J
zt0xICITY62-GXi@3x*oS`H{m?K)PJ@9U!D&$B3D2D>O$&a3~JWKg<VuCg61du6~q#
zp%sAV%KNlxVImJ99U<Qp6qj4dcr}c6cBH%JxYYq-dFt2=6e=7R+bU56C!S71UJ>Rl
zP+@--X`T-;X@xJ5Uwe3fMqjW2e;+n$so#%})O#e5-FRZ<*ZgduPhjZfKObCK|0Q?u
z%RRLMiDlx9CONW<+b7Y=V+qXA<Ul@CycSn(CYXP1R&L@ogR5QT>zl2$l(DLQvuXo;
zmkM?_h5=1zPFv*YEF#RO(<Co0f{9?{o+Th~hk|oIqFznn8~t92#o>3xnBwLop6q}l
zzo7hmYXecVF4_bg%{U+8Ge-J$5U3$9sR0Q8b(p638Dvc?|AqVFBbPaXUm@wX;VQ8^
zvB~z@dm02Tfh@tT=HhQy+|2Kw!bv`Kt6}V5zxih<f)GFL;eDGND&T0I<rkez{Q35T
zMx|h3hnc8jet??`%2`?0<M`DTK@t!`pAJ(Rd3j@$e8L_y)=qV%*d!gWq<JDELb|Dx
zr)mSFAtE(EC?Ui_mMejrKUroF^vDW6QNChTG~XVqUnX7nC1rh9y+VS8X&Yhq%qb4%
zvx5;{O}&T;3uDFGKCq)u0awl^&9!#{ZQyEwgWvbsZPR8z4d_!e4Sbs+<IF=B9Sy{F
z0zDi}3>p%dn?={|K@;7{oLY+RPm)PEYitXYN7r($Gi16Iw(z_Iq-3}*T)@qM*;5Lb
zU^}dtXj_{2R`nd;k}_m${WX3UuRE8*2tnU!3mY<+1EiVuaPpLsf_Q2KY#ae4s1r`O
z?#Rf-C#;V6PRR@3)tDQiQe0{->gNL!%qsXLTYe$TIonvnNw355Cc(ZO#(p#;(+=#L
zr}A;pG$6|hcvKDN#V1NsvObfuLI<u`66bp*U<5(8abHABK@I|z43Wx*%$%Z6b8G@D
zGBc}@Tjen`kkX8%;MELNq8=p$2Uu}dx!4f=*fNMAM%>dyO>F;P{qei_D!irTA0ae*
zdeHlG8qCKf%~^ybq22`Ny5|S9nSnUTqV9XaEK!6|atE2AwbmtqJyk$#p$L!@q%ld?
z9jXixIoJ+lkwr4a@xEdD7D>AOnKuU}97k{qG7Xsqj&hY`reLa^5cK%YXVpT0nQDkh
zv{Pny0#P6jDw*RMg5Z|W^cYlZT88lK8JMV}1ZYa~Fd1oIodRN2W&x@!A{V3n4HWQh
zCKk%Cw-`RM846-UYhk=z3dIbq02{3Q46{NG1Zlk3XVm1hzQg-7EgN2Awl9`H<jiEy
z$cO3KnnZ>k8fIK81<otsCX2vGQ}=MgCdi<n3lO?w2p&u7zcG%AV#P@V*3LOA=))sM
z1zH3twP5w(rSa)V@VErS_j`0ZJ*3uNjTM>WoSJxg3QF_`<i%U9BXV40-6lO!grBo0
zWo_X2jf!<g8h_(k<cL`Gc70m+L-ThSm8{U4`0Lm>z+FR?XG!4YG3EO3a?Eg-4Xcrr
zAZ9752@cLjQ!`3hfWnT$O5Z+$;h-epgnD<Dr6EfC-&EF8sm?w6&gSgHE(4=+XJztV
z1j{%weaix#>_=^+WCfoQoW=hD$*y5BQGlw^+BbCIar@*LsW-T~!`&qVgiK(xF>TZ&
z7pCZ*6(Sw~2<ZikgAGcny(#7JL{tS#vT*$r8Tb7G_x?6%*5BfLdou0ncBS@-(a%H^
z@^usk=cbEc1l5XSL*T);&Hp@%&r6h}L&3r%+9*hs&T85FeJHV2D!Y)U3=N^4(82Rn
zdVY|Nstu|M;>`^Ew7EIpj3>rGch+*cgQokq+vl@f&X~-lrx{A+9MKY)z?8zcZ@kHJ
z%G>Qh{7~~G4gyZF6GJzdEAbiH8@nwB_-?%Bwb-skOVxOorKW$%{~6n0C@-fC2yKE<
zhwdrrvdd%t#Pb@E`Gb}g(|1mmnvfiiH|lpc*Bvck@-ky%!1~_gJh|rE@j0lrC`%Uk
zgqt(H_o)zqMl=xrKrK*Gl>I|}2uT@TZJT#D75ebyah~ZFP+yPzeiUv{WC*G|tPRb5
z(Sz$^xGKnM74&zn=-UA4X=l(b7i7)U@li6x5VqVnSV7ZByZ2YjOn(+IGR&DQ7MwrX
zo5cN9J-8*D4`m`0>3OaBAApT0H2?TZSP(Ws>(hvYR}q@;ER3n?GlQb*Qh|h236<Oa
zOq$QVHm?omR^x)S;h<{Rt$gZm=--;(Awk2Z6Q4X@Q_v0jW1t2)!mk{$*s}%<6groP
zEAIt`gmJOgOXt^<c|U=`4}na^ch^kt+kJSd=O(+fZcH<VI>j0HqivvbK6PEAO#>f&
z7}m>q5|?l}0@S|ugCmQd4ao_KHt9PfQfrzY(rbu{G7B+)A6jq~L4ue5AJ>Y}Ros6q
zK4hXgy67a6(V2uf>m*fF0BjSoW}w39zI6WLD}0VED-vpzpehHj%38QpOhwGBSe|9v
z6gt5t2Vbqd8q5<RzX1nB@@=%<TgZ_|Z<|d4qaXv4n)5of*DM%G)Zo!Nua(cvzJ44|
zeZIe-K&Fb)v-Jx~g@0e4%?N_~D5E34fFfgQp%{e5;iSJ|Z?jzg6Y|gbAEz)D7S8|q
z6!w42EDn_aBQZN@VI7P}9J6_Gr2@!WvCCifrh$`)BGIE%B`2FaY<@mC=HdBGCnjW-
zQ+4T(zyfc?|E?OeCtKo@FSK(S@NfO@K9xd7uaOkP0oamciffnBDe-t~O2juN#58Y)
z$|t{0{Uz-ExNs+SMcMs%`qDt${kr=K-;786`Q=f6<O*4;RiwfN#5vxg<C;95iEI%W
zpoHS&k2Qy_A@<Y|-YS5P&=!iSO2GM~!(PtRYHRQHt7a9yw&MQpgsQ~mOrJKp+iH7R
z%{30?d%B3G$%ane7zW`*S^4^8?vaK?V_|6nOSGN8txY)y6kk-XW!g|ZCa*0**gZ%;
zBS913#%}v?+EhvF@b*Nr{=KQNyj4%FjNQJ-#Bw!A;VMfMwjq@YY~$H?Gb>#~#~}!4
z^>44E**_M4LAE*dY2qXqDvQKdSaHc}HBk>w&PFC>bg`es6h`>WL_XcYP3g_dW3G?l
zy_H6tJ6E4o&OrBF;klFR)hB*`MG5@7o>cWq8Gjbcn7o2NIje;-wIvCfG@xBj!`=pz
z^V|XLsym@z%(o%Um*Ydk?DkIbaU%rp2<TRX*=s8xKg+Mfm1k&vQl(il-LMdWA0lv0
zSE3S6Hiji@f}G>k=PqxtCZ3jeeHq`_qB|)Ye)o3jI!@Qls=WQC$5-wWlJkDW0xPp-
zJY;3z&7nKCuIK%JbMEsZJ3QOK0m4KwjmrQPo7NMf_Jz}4zYJ&P6j?*|g3j8mrSom|
z6h^8bnTf$jAMH*;=F^hn(F90DUzB@Tj0PGm6%FYO#spyoAksYcdQYl#-yR=9fN7O9
zTq{<eadmK-ej*@1$@3O+T_@@Kw><I*4&<KkJ0g4D&w&&HC|4Th#GtFgbIl7=9-rT5
zNq9bMSKhH0L^&nP|Ht#P^yigh#c5V6E!Q{7G8sX#1hlLyU0CrDexCE@g=v^=U*kd$
z8IpjTIrG(`**33+^>$lla-zkMZy$5HySwgKSlF@i7DJb5V!}WC$`L_{G{ZH$6gkUD
zFC>Rw;0K-=jVlMiAh!G#Ddoxy`>v2;?IJ56kgmWlDK>(p8RQR;&;_8dO<`Yb>mMq;
z?7F=BFxou0uPo1Wfj5WtQq4^_{W+~*`N{4LFRk9`&Q#;8V15cX3A7X)xv!j5o}4U}
z^(ozIk_BmJY9qj<B^giNd$r`I-}v2rl15lxaP$E~ZO`(xUd{nSPOnF|2%-#r{qy-*
zu9@}$gc{u?1lbGtD&chT>tT8-BX%0nTzDnCl)qzdbq{xXv`pI*p?y@!0$TTY23MU|
z8B!dA1Gpwm|CJXHv%qciI_F%kqa5`S7V6D-a5Ew^Wts(t<R7M`@?39wlYS=|=FdQd
z72|ILl%~9!UKkzv^G$=m@e~2dNWmDT0FEO&8lW%|3d8ND(@tEgw7_ne0I*cpC9|!@
z9<=@@!>&DIl+`^X%sO1#;y0wDoy=sQkw8{BAti|X8Vj^25I5FChRb$*$C62ZL}3!;
z5thn1=*a2J1Alfx<b>;{kf>%GEuTfUd}*>eyJzRW-5y>Zj6@4De=u??8xi36|4MOi
zaRx|g>BA9t%bk2Jl&Vcj|9sirWAaT9S+bQXV$Z@I?}cpSY%7l{=Ry)=k_m#;h4?$i
zflQm<VF;$lz<hzaIeEMX4UJrjs$jn4J4bO*#^mjZ!5-uB<*S7Ke7$Ym(p3Y|L8r2m
zs_ywAXqyU3_5TWD7YNUQvTD?QZmB6pJkq*xVy&x6BJcz_N*QSKv-fdrxjpiWeOP_`
zDmZ8mdD)|2cI#na3zZ2*Id;?Py4v!JG@(K`G93hfbz$HB^?Nei_(MmVe=^X^T=LQd
zs@(E-u%B`@UEK^3+yb2d0PSP|CyAq(m@^LLnCBwxMr#P0mq!+=ut}32$dZr>O~sqV
zdTY>;7FAiPrR<5S-g^-qp5P|J8AdjiZb42(M|BRCej*%{cf?*$1M-k1PgRdEYx9E9
z^7lxR!lfFzP`~~z5?lW&yS3J_qb6&aQU8n0Z}+de|6K~q&TEqnSbaZtuz<^9T-dSL
zsHZ%hrQ}u)3o)g_<V3Qg{n;<aQ^vR<*H<=4|Ln@wI2<Te^zvvP?ZB<z<f!_u<kP6*
zOmXC%hX#JN=1nQ~Hc(|%EP`@^ee+X(AXs@bwS8XYJIw`L|7vO{3ePtBHfHUdonZRu
zm4j~h-i<?*J5YWPkUZ!zk!-N1+j(KqpE=46>k&~L?}*$H3Idc*K0L?T{z>9fOyk}$
zbU!x03VC=57q5@>;BUAO6N&w1vPvlWum<goZ&SNi#>DplI^A#R83p_aDbV=$)aUn9
z1DaVF{ElmTOAzkB3uiK2izB)YZuekaK8#-obGnjgWvrg=fR4^KriJS)6S3>=Qh>u*
zuDiY*r?n7H3=TJe{|=060{auX)uM^kb97Rh)x4mY-A1|$GthER36f?yZo9k5g2QF6
zt0%b_j?O%Y9BJqa_DgtupV0dr1y8sO|5Zs)QqRW5zrb<zDr4(kxAwywyu0V&23UC5
z04Wd96Ljm8mVgH}WvLF@cHR_cq4#wy3Cusl>onLr^@>Cvzp!52JQ}K{-1f8c>s$xF
z%cFI2_!p#%lFc0?r1>u&yvv~zIh{K3qO^!b7+ub)F;gZR%n`B8<D!Xixs&}HI9ws<
zoKPh+BMd{B-K)3c9mQ0)KX41Iy(|y_>H;d?M=rSE1_eOBb6~v4WdC`XGj!cdVhNQO
zkFr=sRH=L+)A?Ksw*2mzQ+u|$UY@s4u<(B26yuZL9`bzuJx*YLEH3YPUNuU9>8A!5
z0iWgQ19RTBxq?0!_^wYluwp0YD=Z4H)j(%=?d87n^DDhPc9TvFh6kVX)GUVMX0Mu9
zm6YCc3R1tkGDog3UEhd-Sa7|G`2+x<_m3Pqqxc2QPA)0>ui+x7Uac(;L6VS5^+J=_
zpC&-XW943#XWvu2DxpygT?N<bTtjbJvo$y^`U+NmfS=^=EC1_LBiDan#5lNE|JPIJ
zjOM2EKa7~~Uk%3x(HTiLL8RYo$^TF$i!H2HmsGw^d;}Rmba7rvQj?A4i~j%E-_l@i
zQbiqB24GOD%@!U)cu1!!#IxDSgyodwr@$^KjL={6nX~v9JbX|96h{jl7Y0267zvHV
ze1$n`DeobwQ++h?h6sY^i}Lh7xj6=L)O^%ciyg~Bkpn+Ze|KgtU-$2q?RA?go}as$
z<)`cV+ns85g$N<$06~CoVJ??yabcrG@NgzZgMxY+dJWpJeRo0?N@wM`%c81L^m){F
z+2pMmAS#1*lps^cAU3zThG1bW)hb)SSVrxzqNe4YikyoOLszx>SJ-oCM|$>G0!bDZ
za|SYtowx<){KR;FLZ&6S{DzIUD4#f~&jyN<FWIoa;v)V5xf3Ta-+@Xv>di=hHo?!S
z^id{GttFJ+9c=MdE<=;}TX0vu-t)xL>4slLiCW{dGq+W<u(8ZX2V1Kv@nUz)Z!wQ3
z=dc;AOcIs_$!1;frDB`C_`%$cWjlHp`0mqPA|ryozQf%#7ESirni^ZCUMn;&)ctM6
z<b?>O`h%VIT%@yLByr5<AZcE`fPu~&MPZ^>b+Zj~wRNn!vXE<St??#1ci!P!qvPr5
zSlXSmC|L-hGEL=HRcyq>K`f_cj7=D)a;ODRv-ONb8fCRe<`U~icn*Le@iYi4V98@@
zk_{*sY5cm1;{%rNMyxl9=2r43FPKu!(Y6S59&+hWDdg@rm@2Ik50{}~p1(%jPF>n|
zo*MkPDQucm4t5bTk@Pu4O2?w;x889PM{%|GdXRrrXR8>o;S=bJiqYs+2Dt@KlMMGJ
z42r=0uT*sy@6!&i(&AOFO&OAW7Yogm_SxRLXikY)p6t9&uO{&^B8y5;i6G3!HK}OS
z<ym2bLz8gvK#9S%%2HN=w=c~UiiAiB?3NGcPsgz2nlxCKq7Sw+e9<TY7%gOzAY0*?
zS(=ZhAjx>wyEATo>^OnGK33=eBn)2Ei=j8Oq=()~^PwEL$SB(Nw)j<^-*v=Hb#98j
zWI4ay!qW9*uQh*w(~3y-zMN6RK9vr{_?RKUEfqNg#tGo}r*FnpPL_}=s;+>&?+1xU
zkeFUkRYXl&b!%#gK~cBM8|(*rTH9lsGOoj~9!0nylY0Zr{UNOOLvecJKXmBN>Mfcy
z2N&uBi*NI;o37sOIk11Fsj(Aeg-;vzo=Vs>KYdf=LE3!zvavnnR;ma7HICk4PI3@C
zex+SmZqDB)u}Zz&6>_Wd@RTmk3O0+&@A|qLpp4hPLLbWq{VJKgl6+61+i{a@VCUd5
zy@Q<MRY<y6S-``aGJKPRv#0YTfeHSC!%!I0e#m`%aW)V~H^?tP^tDMf=v!IdQT6Yj
z>oTif$-I%(JPPI8arz~U`b%@D2k<N5B0uvf)}tC<IL1Ci4Jv8K0*Sb=lV4#K$Ax-P
z_NW3Zd+eZW5;0P!sstwXdZcS}0T_-gLh#`Ag|GGiXs9ZY6^sZ+27k=6_@4;bs3|&2
zgC}*U1TsNQL9vHfBblS8_)4R+$4;M`yRKgoYyU3S!I^;Uxo~k&TXohU4xlFP-~5C&
z;-qVJwrZ#w5sqL&{4;KZn!lNSUEUZhQ9K+S)Svf)*A_!kyM`7rNVHN9a%a#3QBr_}
z8*FQ|cS_Q88H~VglM9GCWO6sa>CRH?!tyY!1htsKsCW*+ma^idy`fdP23I|sFQgwS
z%PUih(?PZER5>qU5bc6YnfVBvPkL}K^}yK-Oi5kAl`u$FLb@S-HN^EcP(9qO!HXEn
zwtlg^9CAzdI_#LMN=5b-ulDBE=NlCCLa%F}6D0DFp{g)C3Ln7`xYM9Z_o9%?HxDAO
zLe8h$Q-oaFxQ8p=jS<y`7g<w{kjSZRrPZM%UTS$T<AHA8VDqJE?*|PgR%x_~+=!if
zHjyhB`IR5q=#_-6<k62sRVhWe6l`kuuBcs%(^onF71>x)KFFIC_I={`fb!U|1c1X*
z%UAy^3;-!{);ZoCfA(1b8dg_rX{ExAcqqbcL~ErorL`HFmD4Zv_(2>Xq)@ay_eGD|
z`l&zWHv`V0d39>xTF&wrIv0Y5HK>ZXH_->8`jwL$u}NqJm3WyeyxsG843pS1XAipo
z4MtF<6G0M|F;%b}gxR_m(82>4Q3K3fwv5Xf%Zb&~;I6;8@4yKR!!|mQe;6=5W%*7#
zmSErqy*g|shLM+y2uz2xf&Zt-as2-?M$GL0N&)}x`5$Q7n*TCJ97z9{F?zsYCVQT2
zrJhHV%ci$!Tl!*Z5fhwpt(i^Yld|7H-*(tll9)*P{6hvGMl2}$=l7wD2V(q@(&a|c
zN_24Y{Xbq;@gRyQ7+=FE>l8fQE-cjZFB`trosBJW25fFW!RtlN`n=dV2K^oa?yd5d
zMT@z-K0n`=3vl-_#(a~*&%u~acQ2=hr=zP++fNDF`;%49Qw51ai+BmTCPoH{+c5#c
zTyJ~Kz*E!<NFXHOkitAyLHLssA6J@;5iI21xz>!uac({`5){2xJ+I5!Au4bibuC?T
zFLe?Q?6E4k<SU%+WnCka#Oy?~M1><VEwbfPb>q}JjHbzGdUYmY?ivYt_3jiFl=nh*
zd0-WfW5qSj$KR@J)23C^6_aK25f!<n6}iktVg>w;HqYXmu%rh*MScCt#+`)7sKoc3
zyXi@x2$6$o$Hs}f71YL+&YyKSIi<4cuurKsYE8<OlQf`uAWwjV0epS32*|w)H(D*-
zYU?8~q!hyVc2J8v<TwzLj7^OO{U2nYJ7M1NGxW6_w_otvt+DeH4=@?mf9Cm>8buFV
zp5XeH`K(H;?eRE<XqYE;kTakfd@#a@jdY0wbz5qp$m*=~t%Ht1i_}7=C$;oTW@1X|
z0uWMZaKg+_!rAhxz&R+R`M#UUI=thZ@(2^hW0Cr0mSgSyj6#20BqNkV+u8L{0J-D-
z>>G-JLGvDO#H5l!5KotFv=-W7B?sa=2co8lC1IL{OW9l#s?~TX0Ki}!8bOp3=Kw7v
zv&*vC_i@Sulsr2~DCs_eE4nTVrHQ`DsxG!tpTkTF$%O=e2wbU|IMQI3p$-jAi&Sf)
z;6B_;FT&#Bu3Xr+%gnpr+b)r3-3wRfVs*P!Tc@;okj$xGNqZg8SJgLrYgjQ5D+gAa
zNZxd^`U9erjZz>Y<)de;A%}9<9HfAH$QE^N)B=8DtAID)ja|e^jFiRGEaIFRR`|&3
z0yRnKNQ4m};QKb#@C5>||0a9$^S@pefL{xL2jrZ7guw1bEp9B>?+ty99{`i{0fkOx
z;eYxWq|B^~5^)Cd7n+<FDLNee$?3#H+2CQdZdVQf4Sp_cmDk?(M0l*cO;Qw{VSXFo
zdfNn|8v~6Ow06Xzv~Mn3j{*fb&xrvJV?D-yKEKvjFLy4Y@lqSl9^$R228RXLGK(H`
zeKxEjYA!{QHS;9#9e@|y&q4EzXBXf5?SH4DzDO(Ux1pmwQ>+cl-o!Au68!|W4-|Ep
z&&yE81HAM@5*GF(K0o4W80hN>dG4*}UtxsW+{mf-#R%lmg!@o?Ah6{N$_mQ|@^a0`
zNe0{yXg~ZAt`xs>BG1Z#CRWd*FUrTU=4$}&?R!mLH%q_dS;!f5)&n?;k8@-f9@j_3
zn>i~`n5;)U9-N&qa{xgFtE>MjZK$K$!M?fX;P15?KXL8VBo(9Ac`8louQ>8@=ra}s
z>B!%HwAB7;?11+E7~f~ZDu7+9VmU|caoTDU*G?dEJ7RwuxW3(8VZ+v5!W-OfSwY{+
zyNg}JDEu=U;{Rk&j^5=ZTC{g>`N+`N3-0=NVLw4+`^wK>gdjm^*G*(gsi(kgodjxC
z4f9srELya0>uIySR(IGXQ;(K1pPSGuaeQL%H{<csP*u%cX7Af5M8(2;(p-i@Q2!>K
zhT={dZ9(M}`rSLLfZ+%T&TfS!Si(#3H;M+KX^sL)=C#CVW`JjOudS|tiILW?*u@t>
zD=ZKq35sKTeJOZDH{607GA0?^2hIi!7%-af+TokIT7Z&<OKY%B-@aS`&`$x!JjAN#
zJQ&tsM)|g(Nw;6;P&;(D5mfI_izl-bkq3Hy_%S#gVSrcBV@Rwk^UM;Xyb`zr4R{yI
zL35cP$YCxGaLl0z(<ma>h%8Ggg1LhFM3M7LGZ+VR4(o5o!+Yya)sg#h%FW=#qHqlY
zXsz844lB|${-rKu;No=}w=iN@ZfF2%6ExRldF_a#r{g#5<PHw*e~Ab8f0!|tIobbb
zGscXj*8dAy_WN7EN6H>+7y#tSCTH<ySblMutFoAn92=2bm_sK@)$=t&{(ZB%KZtEY
zx~UfLIvNN#8o+-Au6(*xI2#>oU=A5R7Q~Aj`D@O@TqgV@4voK1n5TKz>(*O{>y7K<
z`YvOkLSnRpERYZgRpj|<wcNp6qCB~hvhvc2B~v=T>vo4AS3aK*zmJdGJ5FcpRz7b=
zzn9;u+KWK38k}5dPwG5o>|txY&8_RUvUq)<92CO1atLo0+H4y6Uyjb%qLF9*aw}bs
zkS8grA633G=Xl`ugJwN*oU6tM`BO439D=s|g#{}*2g&fo%Li8$NdOM%RC#0_i0AZc
zseDOqJBR(Um1gclESb{6e>mDWZ}bh4hB|3i<P+g#HJ+e<Bxh~3GBa4-*7(GKJNtaU
zaZD5u1A_CASLFm5;SwTq#nkj`C^_(sxu%y$*?rdM%hN%uENxLRRe4!1^HwzgayZ?n
zT98$uww$ZtvbDFvCN@QBCNnAl3PYP2Pr<iM+don#E-0TzWsa1;z451Mv_j8P^Vzh{
z(UvB4kFM@2LWJ9>uzE%7w<2xo=HXOP3$<wgaB@&TNYC?Ce$#<;szTP=oTR`!J(Q#*
z@3?m76iKkia=by%h{hCY0aBq5+bk&^OeEpkE_WgvJ~~R|9<q_3*_{r@TakS9?~U`a
zq-3HCCWk|{ccMs}Ybp_N2wE%(V<ZX?po4uTSp-DNZf8vU3m=Zd-yN#FZn^f7F!_HG
zY<Xf@$?6Vs{{Zqsl-R^C2+o}+e`&^yVw>u`w7l(QuUTW)9<u0HYNvX<LU^hm-I&}H
zWo{eYf7Zm*_m(ca1YNSs(n_k*=B~Bu`4sgRtV6e)-a6eA3BViyoCIt0WN&ExLNrJu
z>o<NOFf`Mkq+T<465IzF{SvG-5^YMn)Dig^C8nKq=}6(abYn^3d7*4Mw`d5;g~Ja#
z<M-~jBF&2BcGbK|1(En(p%4%r-_g2`o|L3yG)6EXOZsEO3v?zI^v|Oun$#%30ux2(
z0S<O4@HsIT4gd~ORjzT&<E-waaq_EkRvUE-EM%v7j8Uwwtx9_BssaHM4w<zDhXLfu
za-VhqGuMu`XjhB92P!L!(VyyTtR>94GI$s9V)o=#h+&wdC+*4W?LlD>MXK-8?HOEq
zT}rb{1?E=oSko^){?24t#VYs3DTwQ;>*p-ZcM9zCaJRwAF#0@sIs$KcE;*fba6IMf
zX)$EstaXY>BnvMIfqfRUZCA}x?Z`gC#ig%~rxY37N9R<$r~#Au=?&|}N9rJZ8r<jh
zJP2g%W>jwIU0xBP*SbX+t+I*Pi^^ObqDtu}=+rJ2UBe*{;*E(CRH1>CZOd4~oZ3@D
z_p)`Q<H#kG=Bx;|y~ca@|E*audBN)NrzWqA`-aZP2`GCHt7FH@vZ%%9ga&(2hw5Fj
zpI24cA1ZWSanF<!r4W%KRD@!)@GPM~F;<L;?*-~uPp>2z7LZEhoyZ<p<p^3ZDCJw|
z_y<y~W25$7_Y``%wq;6jzjwA7muxdKbZ*bBUx`q#z@?E_FdzxU7NYzXG>8nP>1QuP
z4t1sRREyU}5BS8gH*QPkaX+8k;KCLxOoz1fEzi~B8v70@Cf;}*<u(8`s|*R|3<%-o
zzI!hK1628R@5y%$>K~PJzfnzPnF^&w-12EC4jhN1hp(OVM$Ijg!8+L#kZt<B=Ft2U
zWK}aWEgF}%#s-bXQ5g2wMg+!V*@<|pUZkC$#Z|w^!(K;S$~zY|l5ftiMfat$Z*Ant
zvH0Tbee)N~Bj5zWqWTp=>Y#>(&S0w2_x&P45a$w1P?;<j$Fckb#fS3@8l*i|+v9zH
zCQ`W1HXQtg1MIAg7m;t=@%MFOn{M-UCmr(oku?%rq8!**m*%}-*9DJ$Eh~7W&^(ab
z%GB$pxCNYO-|L{cPa^tvEN`vZ?)3)Ap@<U-=su<7ZJ^vd$Q_b5@W(YIR4p)0Q><;E
zw7Y8kjU`*f&2HFWtqH^ic7JO9SenfMxoF5pe3T9lsr)aab-Jh;t;`xMqZ|?Fvp?2H
zWo<i__F7@=B+m7J-a%He8FWXrf;%}`3{<A`W_tR`>K8&WtnI4H=|vXkk;RW`<Xv1z
zRg4Fi)MD|`m8qTpD+=jnPPLOvJrRu88U^yS8I*Sy4TO<t;;Ngk>xd$1@H9r^t>3Y}
zfLS-$q@NK=ym8Njk}(6vJ-1uPpXYGHd9guSP(s~NN=^XUYN<5Qy|7CyYQKm8P%y2X
zvG>E?ZJ!QS&&KKHZP{E8F0kWn{#yENbv&1QzI)%OP9^?;!#8kZ`SsF&>g9hSoLE`^
z=jx*4e_iXF9BBXaa=w^v7>==sVTy-nJIRpt!m#cm1tQd#H4aHen!MAA>8bBlF+7Q0
z%gMNW3g&90gz7<8Aoq5DDNAdK=duU?FM<L#d`$otgGcRsE4i)=4?|&G%c#ob;7O3=
zA419WjdH^9zStiwLccS_j*=H<KDfE*t(w@l@V@-MvA4U!k9JDZocvrk1@ST8xyCWy
zlg-m&#ILFQ{P6NPO^s_w;ZoSNLUYFXZ0{v38L2%d>e`o#nVK>E*cX5s%Z=|jv!V>W
zIkXU!Ng8M&6rQ=<L7qNlUC3AO2o?FXN3E>ZYER!wYtu_2wB{-Br(|F_4lQXi+brHx
z_;{zV1Om#5&`yR3UcS*&-C_+gV4M|pCC*Wtj4ngdzbJ{K;|Cj~WtkPTZK<iKB$WeZ
ze~^%=4sA^n+YV<HZ5-#paQ~EeZR#B~UHQY|>kkfB0_naE=x8L+=A_G1^iluoQw#Kp
z7`45~DtVPxb2E|fW>H)W9J~{W>vv`1g604a?<Mfde(=+88iMd}OO+Hfn(T`1BJ|sB
zR4yhMoc&FFp>{^$gvWZ|2nV3E?MGl$`>t9eMFgp=k^N7XwCR>jUcsRqpphx$Dm`vt
zN~~o!AYY`jq4s55Y_{!>Hjge#6~(eBu{2xo=(`BoFB)PQllnu{@V9OCo|2fh2ypk;
z-!vyl9l&TVFr<pQa5!U}(#<KL5&3hAA5@u{V`41SMQj7*mG7d|3aWAO#0Xw!j3-9G
z`B9066k7S<ZCPen*Q$A|%74U|r4FUX)iSCpy%O1u<cCP;OjMH-Y$64bfK9Ai@i4Sn
zL^13_vydqb-70t^)KQem4}hS(ld5d&s$ZN?Doe(@MP<@Eit0o(VC-zA99F<d1!T*(
zr2bH8aLrKe>$+9mg9WImqC8k)cw9I|5-xd<%NRaxdxPC@4;Sx<#{&VK26`S76oSJ7
zC7fDSQ6E05$wjFLa)MKUu92APfZtfP>Kt2l;9U__@wEhb>1;V$Xl25j>Lp1Bv_kEB
zIj>&UHj~%Ah^6hE@}$VrjwMVCEj+R>^t2Ac(zVe;-xm#7%Ha83tXA@PGxJ9L2vKfC
zyf$Y=g-4s}up%?=^_=rq8xnj)kbU31lQCqG%?8HfwHZ5+VOWhVT$S~or)p_q6{X^!
z9b+9ZsYFf=78UI?Ml(?t9Sr2;l_%@uU{J?}-Nmlqb>La7?$>6fF2&lAf0a9Bj~8}Q
z#yGd|in1&X>tL}~#QKVK`v$~w=T>zZE#XPnYCjML|2&zkc^pCfK&OM{e3g5c;(Oen
z4$KhZdKh7h!q_XmsZUP@Fx8KKB5cjS>G3JI@9N#Ccx;BY1@B%Nyf7fnyuc8Z<xw5R
zemUs7Qop;^eqa0B5J;Yp2CG9<O|cT2R`UiU*#FqKy4Rjr>c}ocF1b@(FtFTh`8Gf;
zhdHh|mp8XJF^1(Aj$%Vp1#~Xie2_jOucc>InOCgLBdAw)4`g^`W{GPX>o=MYouHev
zkpXNaH?G_oW%RgvLZh}d_W3+w9}5QGb0ndhrSQkHfVmPt9>DmI&$s1LFc(Tl*!7*#
zjF1}-B;s0v7w08)rlZenGQ`4;z}TpK%`XYS2@|0^^|r8LJt}reN-l*pnneX8r*QdQ
zw-pAi1M1KIfZ7TnhS_Ff<zsYOXBv$wXUB7T_Ua!x)id|Je|G^+%E*V_*kEsGoE9he
zFLEzY5yb4a^}%jY86c8?lxhfVR?4iwM!z7@0Op<l|Nr2B_?-XRS-6@1_nqbcD~F*{
zpu<88il2B&pm5J7*jiGTm!cGjOc;R{t0F^}SfuFd`A5+UY@z41+Bh**gf(n@eTig;
zuLH&TbbUI!I(y<n6O#TDAu(W!9YK{~Bx*_Zuao?D=3NNE5C6yUQ_jMY&cCI_2@Mpr
z`QmMEPJd>WA--Jq4>abHCk-i&m#>>Ux381?_rt*zxcoobuP49H+vP~=iI-?Cnmr#t
zL8~82s`ThRp_12Z(j<wZGF37m0`3LUp)5w3t6pt}S%zX!x2H5kgJw*HQx5QN{m^pd
zC8KL!E~34t8mWc{DsV(`ECpuOR1`PHhHPff`J8n8>!oyZ&mzF8nJka?Js#agTXku-
zP!~fdNJE2H-!<(wLtjjzLpvp^#%+7aN%TzzkCMQG0^v_dViD3@m~lHP*ZXVFJP8FS
z{)PT>G^(~reHT&lOs;b45SfA+I*``>t`4EXT+L5%iH%qZ2Z1}F;?hU-TEUDDSnK%A
zl9U#&YgXIkcX>oPmq|jHlfDP7ZsWN2avo9ZKiO4Y-E0%tp!q032*M8|JNQK<RowIt
zh#NuGX_DVYzHEnEy!(Y6;Dm=ds-nAc%~3@zk1lmYpA0P}8Z=ecmMyn!-JEQkwBJzY
zqIPiF#$E<)Q2SnxHJzlxG|0!AbR=1MVSa7i`phx%%SI3&lr^F!Yp*_H9GVYAt`Fgl
zQERfiKLd4~w}hlihNyG_=j6a?tz;RmS0HYU)L7HJ^;X|3!K%eQb84MwH#KX$Yo+sn
z9zr;<-#l7L!HS~YirWr)A*bScW-3_Mj<0M~in_5H0o9t0ojK*5XeDasi=}Wu9)|{b
z4RTVfXIhHGV8lff<ZY*gfZ4x(iOL%;Qfhx^5+8742=|5M*d$!T%FC)@-x=+1vKZu;
zShBLKq}c(mMzkescS<c`PG`Z2pK){CT{>1U=)AObe;QL_ECgE9>OTw#5eAEXPArR-
z9DFA;6#fvyW24a|_<Cto?#dC#0OIMuCRg4o13vy^E-@?2h-L{#9RVtu90QR#EMqzZ
zL;@Zucg61W-9jWlC&&a3jC^-beJI+&syBT9XaIKZpo6~6>uQ(`vZE|T=)d&!%cCDE
zM#G<9|B2Z#Mz~074AW`<R{u!w7}>UYJeHwdr0}~*ScwVaGnom^TCnqqQB<NmC+8P6
zB*#esvA6YaJDi|<mU5QBApb#Xcwx5Ty<~v4uqmZ*HchJpB2|Rt8L4!5jJyIeyl~ja
z!YY-`Fb?X35WCahp{wDZVhc0Ly5zA|r<NJj+38{9vUToFV2KFSX_e2`s6XoTeGj75
zQ@&`q67zn!BD5M>$p1sxJ2i(IF44NNZQHgzW81cE+qP}nwryv|&WxRtb!uOns=Y4O
zx%+;=S8tE*(c|fHh-M5&m?LI!O(npMDthUsxY||Xm5x0mRq!foT-}0foW^H?-lFk;
zM}JZBK|_Of=(se`HM#%lJ7O6#lG_=S?N#N-e!e3!5gV75<lJ5X=}^2#XkH~@8}lXF
zYau+5I%Dw43z7Geoeb@~q|ARuz&PVi+P5oc6;dA^9F0j0DAvo7*+n$?io?t*)`z>d
zFQ8!}^7xF(p%H9mV)lMQ;rZ?^&v<=aToS<RGVo4VAFo>CrvE)4V}1*G<jWE;{w4g<
z+bFo!Q0X{tE_SK!f`TYFDchq9rMmd!!3iM*y)=}X>#%aT-@CBJLbq?Z?riluz7|sk
zM9%%|=qJr~$f*<tGJ!Cq6&R}Va7?3eX1N|QeI_-Q?+Xp8n)R|*-V<>dSriZZa&SNu
za{m-<NvX&Zg#@|+-_10@AeQeEqyV?E8UqaPjLvy1So6ia55eE-GcX5@Iu5kY%-4>Z
z?!l_@xU5{KTG7<(s>J?^t~b86+zZh(0ht@v@=QTUUHa)VHAQ6K7|`D~nQ6I@Z<F(V
z4IJD64H3L<V^8>5cXUR8>#44GAwVM1x%XQAYjjo0@CEL@zu~soO3`JZ@OI1jG5Va=
zqBZ_|7Z03mJP9AZ1Rl;)cEArD31`w!))DRiCGo=E$UtO{(aV~R3cttB3$w*O#v=dh
zTy!eR!7iH*BM!pt%>q{L(-os4f!I6Q(&B^h#NbNKMI>Wu&b2Lfjnua*Y?juZhyfpw
z>*38C9L4@^xBD*Rur2mI=u0)6w{ME8<hJ8>#@(%iQNn=5jrYZ|+V^+a&c6gc1q318
zkn68&zR?MYe&SF|(g9x#qADoC&~+t%?`9>y@CB?QabZlN7y`VE8%*6Dqn2~YPAJAe
z?P`jNX0lIF!gw$^<tunw3_S%#EaoBdPKg0hlp5Wie)kHkvV12tg>{h{G}qauWP%%4
z@$>E09>9y)v^4jGKsFS^D#}MKi9u?)gauyJ<u*-ZBoRmZv_CbgydeI9C~(i5O3)uS
zMe~?rXf4akCBg`THuMOQ8MWoDPpj=_fk@0t=FELEH-Qc1TrM@%S&-BNh>TWPG+VOv
z_oFR$jjsE95x0927`wf0Dea*t4Fw5@W9v}3yby42Hs#lCb$2&S?&0JTnY*sLKKyRI
zFS{W&ZIxdWUINbL7TKp>!+4~Qr+w4}&_A%LlMw&@Yd!yOJrxJX|AF894+fToR@@;A
zif^7?!dNuRVZ)(zh`Tyn5L$sw1B$!4>bT4f8<C7;_LX(W>n+X)PMHK(zb+VL49mZ3
z96vs{E}nHzp4XF$sFjtSo_!JddZL&`{T5hbEE0!F_iT_VnLDm%lBxA4o5cIO-}kof
z{pPr-F1PRX<eg2@-Tmh<ie7C_ulM!fedc1T>qA}kZ5Z-U0&0sOOM6{JE!*{Zuj@ky
zM5vxff=J5s^4zv48r6dll(7U}cMob7a7k-trDjuk29D7ttn{j?l`&2qDOyZm1{iTX
zxI_p95}ih>QQyU1Cl^y+Vcfn*W<$lQr(uVp8Vl#<C6k~ZUn((w3)JW%CCE6<WsL~_
zN%gRel2s93nYrB418i!Kc+}<kN`OHgGuUByr{Ka%nDqB4{GP8m8Ep*tv#_v8Fbj<~
zw6N2#>Z!&$LqQ-0p|?J^UppV6uFP;W$RS5A1oSkVp+8tT=BRazo6i9cugTiWx$NUo
z&H6k`<!_6aS~>8Kb;j<IK3gn?8`2rS(l?Hl+F*_0B0r(MKoyT_v&wcyf{a+e^?to5
z0wFQ`m`QBClIE#FEH}&2%7ytC;h%L-50;THun7nLv81rXxk>ArLc9eFp3kz6rC59r
z2bQrjJF7Y^45%%E(qF`O9kSSeok$_ty$cgRONFWB(sHBhrn%1b#|T^}Cng8HjR?F2
z%DEt~t_dSQ@4COrm%*|O-vMnllIS;Kk%;O#TBzloBS0Kd*N9)g@+|1KRWLeGge=pA
z32?Pl$$wzJ*UA!y##y`Na$%3bxmLLW>p#D(@5lStF1B*$pHh)RvOhy^KhU5gh__$t
zgDruO=SmbK(7>YDI^$X&tIs;|aOV;9Ahz<P!nvG(O#^v-!ZgZTvPN9(NCpd>523$f
zAoj#zL0iBq+!z$pt`2|<|53Spc((yv0E{L|PD$oUGmueKm|wJiK+b5C><eNpZjDG#
zZDkfbuS&uQKs$(Rapv9az}qUla*1ti=>!6$qsk!Rd{jN(m@C*VZx^;o*y${sQIelE
zd|3!~SU=7L>Hh&<gMEe3zrQcA)seFzo*SZfrU)x~80;j}1uz&dlHsH*DH?#(t166H
zIBB|i#KzM6zD^mhUy-Iuv6NntAyEH?XEG+{x142?>97-2nW_6WAL#&tg)(*p1d&>F
zS%E}q8lD*_>hf%220Q|SIt+$Cd81<u#PlVLOK`R;Zw<I<S=1FloNG|%2dx0Y$>C0=
z!$$H<6sMjj+Qj=l8}cw#6gY+&5iLP!8ob+VaNEPg>=16hrFk4@c=m$f-&0?yQ}hlv
z24gCzYK&;ScD!b~+m{h2{OO>qS6A!XhXpIh1a}+|Ofi^5L;+8CImHt2#OY45?u6sa
zwXW;)(?H-uI7}evU~jJ_CQgzW(|vEUDebXJMq)-8$Jp)m-q<c)z$YZ~J%}s|^!2jy
z9k9i%Pr~QrC(8ydA`Jzk(YmU;$BR4br~euuud-IL6lnMQJ_GLWM3JjX`-59<PgNtw
zcsdoSB^_u5Qa~m}soZU?ZITvZx(D|Jw)MWFkjC3e`7Q*Vgv{cN!M42fo`M=9jYRyI
zZ@SyMGeCd;p&*kHVr5d9XqxgCql1E^4j_hU=72N@rJqH~YpYeNZ(#6X7)9KaAFBjc
zOY8HzdFDY>_HF@_7jH2)9dkk5R08uU&Z1LX4*{hoXr972bigs8e41>!8DxtM)r;Xm
z{!-2uCR53TgAw3RSM&3y91U~Wv>|$#Ast-~t>W4xRf2m06lq4(yEyi*J6=T*29cB?
z`bAfl)oBPdwf+`IR`s=wrq^D54R{&lR^^%1^29WbTL!8(LQ?yRNb<pR>&5r6v;PK!
zC}#Yr(E=c$_xBrky(7v<;Lb<51Lcs2{WFpG&G8?ktuveBT3wW5>8<jX8tt5Pn~ft?
z6|MI74y!!qk(<yLKE;?c#p^9SN?_iEXfsDq=O)@*<pNt~;|F6bk5K$_5gXQOiNA-W
zW3JW+&g5HOEi<_L1=qes5v&!mYn!^Q{IU{$-i771e9QZXQh0uv8hzgg1n4xoG*^Fr
zTzchrwuVvi>R<P7Ryj*W5{zk3o#T8Lp6U7~7zroq3#>7e3R6#oX??JQBMa-cqY$Du
z((;K&uy)oVfWMG#!yB0Ox?u9oc!riT*x(~NT5RnBNSbxBtL3Ja>HY>w$f{XXv-BG&
zi=oT40M{}1_^mH|Gg@)?4aZ9OI%++*T^T=<+`r68BED~Y9T5(WYA>x0y$yO4#7;=`
zYcb2x*M0j;_2mMkwdsxnylVz0O|@mC%Qz-8r_dzo-iyz|BEG#qaay`P$BSu(dQRLf
zwi$6^$k{uEM}<7z0(-`e^MaEZ#Za7=j0YB{n5w_9w%|W!nv%npyUu))MO&R8m^e!R
zbTaTW@Dj;Hlg(kyeSR$5PkUQN^lk!I*?Crmp3a5q$?9T~iTwR*+Lxm-Vub!NG6x{P
z@C8b7@q0eV#Mb!#++F|uvNN=V`t85}E7bOX->TfvkVx8UMe3ca+o#Qr1VsiIDA}cK
zHA6bxPcx$~d&Q-ITFf65IbmXce>oo=v?1}BXs1evf?!*Sxe=wu&+p&z3V3-~sd`$O
z+KTt0lpsd)xXc+NmJ3FN6gsGB+r{UTT7UXk>+9_<5#-7c4Iywq2}0n#e4^+7YVGy#
z>oFv{!z5p$Mg;x7#kbw9+uh#o^>KR`_3r%o+PQq%`sw}fXQ~6L5)oKThdiLoo7v@l
zIsa;N+^MeP(@{`oyDJ~#grs;HWFadE@9FLq;)(;Amg7M~tglbly}9t)t31>vJQgsc
z{DiPgH%)$VrKgbT0qMqZXv1V7Q9#0GXJ?`|&3*V*7ZKdQy?8ew0D3rwP6bk^ypWZc
z)b$QQm~oITgEbfac{pAKvJ=5PVDWZUnu~asZT-&S&>$fu6^w6JkLhNRg^VF!l&Zfl
zbD^gmuVi@c;@+4<AHo2S=)0`yR%*#fG!rp+a7QzsLpDbVy4~=GytI-8fe_=yL^Q8J
zbXHXwRYG#&N-NPU64x_17QO!HYzt2DD4#9YdA2d5m|>Oi-dG*!X)ax9XP_G)<BK3l
zgs?<w7=`Xj@zAkxz<ICPRMS`sYF|&2&v5VmQuBEkoW@Zv{f6Sdq#sozn%7%8&grXe
zJ0rE@L{^uCj2Y3E*L@jTEpJH>E%>OGPi?NkF3aSgO&)^@rfp7wvbK>m2X^3qWkM91
zCl4*FYwa&YeVsR1fv>^D@6V{vOM5!2KjHfKA=J8Alkj6-cF+9)%A>)*A8)jf;I*Mn
zS)ccQE{#(%dF>tg?cY(hSJ}Jmk8n?d1rnhM!HtI~KY&r+D+`QYY~Kq(LE`KN^>mE5
zvWjU)4V~dulVwmAD+odqpUi<QHj(FPlE#1xc9P>AkFmnRHkswn<3uKH6At`)BGr5{
zqv0UGMANRVG{F$&OfZyB;;W|y?9~Fs*nLP~G`}q1XEjlJocOksts`ZSMM?0{%mkg%
z<YX1woi7c8=P!vVH1z1OfR|eKh-WP2>o&^g!^WV{*<wKu7Calf0D6-OM`=+Y{y>I|
zRYj9c)3c3n?o$m5vhxPu-z8jOOMSAOO@QXHS_z$wg;5O~vl;8(xj8?hQ9~=C;zK%Z
z%r~O~JZ08BqnQT~V{ojn+1@iOnWu(awQ29H4|#3}6Le8bOF?a9QNeOb&hj%?n8V|6
z9E|08ZuDS{x=KYO{Y(_J3#}9T8SP2IY*Fy=Dew*@7EGk+NlV>w*)G=<UDaXs+a;kt
zdM>TW3ha54f)+P*noL^7r1+1LO6<*|(_b@0#N72oppGRJmsXXPU9$BWDap6=l(0ci
zoMF)_Y1|ysMIJdbJlh|*WOp|8X~GbX>~CTNMtESPg(-Q5B-Xj$Nc=olO1wUJs&BCW
z{PLb!a~#*NIRgJsYSUxqfCU4Gg;H4q4W=m>Cn}axoe%7+9;z>9VQ!dqfJMxqqS8^&
z7;DT773V-!!Db$sjUm^v{-j;}h0Y|~7ZCl+!NRtF?JSU(V-bxkM?wf`!IV~5rFhm?
z*~J}HgU%|tl*@)sU<lq0;!9aAjn1`2!BIyfbYfpn81QJe7C-`jFcm#Aavc&<7!LrG
z(KN$$5dz4Y!AVr?tA*XQLr5}5g^Yh^L!74LPnw@<1EjSfSf!HE+_Qb4i_uAzycYA%
zW_N?npuaS?HqvnQtj&i`OuwvSDu}AFK|!2&rUgrdRsmax)A4WdFcH{T!}8hJ)-rCm
zXaW8?vZ79A8)D_5mRc6MrP8>VyFta!fvfMO>AGoHS<Q8}7xDE+5NY#$JgWw|9DzqL
z*uF1EZ@AJvfEz<jU{5d?aBYR2F3-h%a4=(*6hCeca!Pr)!O|h;+VLHz8FJ<w=TydB
zPHi%n{+AHH6}@t@B@5b)1E(KH1c73oA|P0}b;q|E66+}Pk0BvCHTv3L7BU=ygJ|SM
zlH%n94u(}+VWi<VjTaI^4{2fXu$<xk7K!4X=3B(KmZ{{AaSV0^Qnzu0FmWieb2q4t
zsvBD<dS>ccZoKUdz=DJHUM_=BOtk(#dUE+`nh+24e_~k%h&g`5h{VwB#r=FS<!~>C
znhq-LguUnL4oC>;sorwxHH+ynl?-k{tdWClt(cJjBf00pCfpM^S6@kk8?Mxm=~2r7
zs{U~4;cmZhHX|WQo@Yz7z~Q5WbQDQybNVMVnN<sjXDbs}0p(Vrqr&{iMr@a8t>&_X
z0j<Vt(}wSS@53-1MGMHdCQVpR>Fo-w09sTCG`a3VbqZra+l7qr+77{uLTZRY!4fF%
zE_bkj#r8Z{F6VZj<!&p<V*wJnzM#a)9ab|!<^K)viv#{#oU74M)f`WaF(}0PBh<&<
zMbm~8(w+8xK`s7IUO(Juxn-xJe%J?Is`Dd4So$$U6R~7l$MBqM+Uz(DHWC&AmTcMm
zcq0>CtEFSlw4{d=P@-fNzdR#nwGB+j(a$Mcr_%oJ_0WO7#(Mi{jsN@XS!b_2xM-Ae
zMa&ghECsZr&-4Ye+#*^8P<$Rqy6UnE>rvrtdkJ}Z18++<wI;M~`DdOxd<)OGmVy7l
zx-qG8vvRwDYndMR2{<32y$>YxvgxB*VWX5`Cn?smt7qhrcTGENE{U372{xZfSyp!;
z3+>|Oh9oq^#OW&w=!(83xW@WH6qn282jp4|F(T{-{mAHr^=oUV<4fORz<8ic7O&PB
zysP?Vf5rjLTDgcO*FQfVATIzue~L$I2f6b>FZNL)=D&#!M&|#r<;uzS|0BvF;dIoB
z`b&+uL$j>J1`-IInvpdu8Ks+wSa!0H(V;7kuw57`84{6Fe7p4kz*{E?OC*ZXPC}NT
z!N=Qu=T)@3Rk*by_uI|R&SB2<{FNPe9$7i)!nqK{LVpfym~M;afD;p!TEffyo({(F
z`+9g_4S!CTSkae%&bRsI__aSx9=^+<KU&`2ZufL@zB}K(uXtca#|sFQh3)G>Wr2Fm
zf(vPqiy6@ev-q=mUe=e7t=xRtINweogN!Q1x(Nu92Z;X)_d99jp8X?0s^|NsJ}f>=
zRHv|ebA(@);@Do)%MbZTV&b}Fe1v2(I5aOM;H0o`Jt&0KG+cRewbPYxksI~|0*Pcd
z!vU$TZ?KzFODy>%welomvRC3`+&I*x=qlrDw|AN2(+b`)<JK{G$NZDlX!o1WGuq31
zul4Ep1*Sri|6ZagRk$vwZS3CMDyVIp>sB&}N>rnebqiX&x-Y%P5nf13!KQ;k_=*YU
z*7e|cQAOkJZLDaI;p&%BUYT({QYp)<;c|vn^C0HLR+;R^LNP}TthCzi@1(&ad(YUZ
zt-H6?S&^ZHc1{r8Z8?G+_F@&f(+qBMN@!GYXOuKd=<I%+5oBxh20f!OHA2D_q_1A2
zd%xFYT{34XxX&)a2VU&_j7n*RLq=*ml1lkORel*dtPC=ryth$fzn-+Huj?$j=Zm7(
zb_K|9Tg(S*R(-=7xQph+%G&H(CLmNE*B9uSoa&9wnL-Rg90ff@?NpEQA3oRaGh-$Q
z_mp(61O$R+!DJj{PK^!Li-W>1G*rMQ_QE&Ad(u+6_A3b7RQVa|@Q3zo1@Ua~1Ppgt
zC(XwfAfTcA5gkv%R_YRGktr}ULrX@uM@xorn>)j=7H==o4iw&R6U0E8?U8|GvUaWq
z0vfj0$V8eDIn%51shvvBWf&QspR5SLjWY^Nq=+RiA5YFj2oXV)Sa)8V#4E3?ZZYI>
zV=-%BJH{CDyg>$c!b>92;aXev^SO$#-U-KzrzL6Z{xTNJ@!}&>>BltC|A<hZ6Gy!Z
zhv3^W>XFS;FT#<uwYw5eP=neG@)!B`I&1=)2@nye{ayVx4tQ5>k%Oe^M`6jm;ap7y
zt#Htr%x_4y-C7Y**gMA=p?s*rzgkiTk?o+*Qc8XQ(DmBP)8x)DyG!^HqFi$srK!8V
zsNQ`(OJbn9Hd<5LqGQKe2m(ka$cH3?H_q(hmWmKB`_6sM+2$`v!k|l;evZ9=m!>TG
zgDK2`{?FNLck|Bx(&(6>ml(!GfQvO@)nR$ktTr?@$d}D{4V(M;XO-+<b5&}P+KX8t
z^HXp6oQE8c>^HOl6IKpMmU|Q|WV|v}*^S@xZ|B}sZaA<63I{}YFO`ZhEGA-g6!&tI
zy*~7Iq}6Rz+mCveAK>c^jYT##VqMr(@HaoZZ8<sA7w<~kgDj;yk`qe^$+Ys4TX(vr
zULQ=7I`e2QjSbque!Q821iODp>|36!Dfu)2LY=7@IOTJ4az)}mA(ILYb)_Z1vO7pu
zvCEtIajjuW6xe7%L~-7Y8T2?5rEvtGVXF;iYj%eS#TnvU*^!XFh#yo9m^#*$rRg<i
zz^`6Z3uo#a!koZEJR0p<$Z{uxfim64X^;u~&TX}`I^#7%t4Jvi^zSBA)yOJp+gkBD
zU!-Wf84($y$b<wk7ys}ulI{V7BzbZp8{tA2xK?+}TCaZ=*$x2Sn0Qbm6Qk?B+(JeP
z>5qATzmNb@;RowIol1<e{2UU3|2Eq{Nn#LlKY$iL#F-S~X66KYji;D{W8U3+9eU{S
zOK3J{0|eSx+AVoAa6YuMDwgmX->M1pKa6)_k8m-GBS`FPFIDjH^fs8bDHb_s05P_^
znRj7w$lr4}ef2N=dWm*;Hn5BP>XPahh@rv9f3~KCeliz9H&y|CX%N&$@!`}zZZCSr
zuP#~PaAATf>;iU7sIxD))qYf>rfB?z;7J(Y^Z&$1L#7hY`o9ia*dJo@>c_(&j|rkj
zniE#}(7#xQ;2fyi#+@n|TOdGl=un1;?5&(461Qwkbr34e07y`Quw#W{A;lByALhL?
zml?f_jjHISHhR>zb4for3A;R5<ll%@nsO<EB9NjHD{$d20l<B=ar(n%$7@lId~I>f
zT_qdYs_4}0P~<Hrurcx86B#KBj~TuCv%=&DDEMc<B8w)5YYbdU%BV1|FMoop@&B0-
z>styu-#yT6sS=?Aq9uY43;xt{7<c`R4RWOy4^@g3&$EDtW#7lxr;ZbX7`lTrWSXPZ
zfuDx_^IzrQV?&KZ8~`gQh2tPaO+a@_^#1_Jdwr#TV!~KWX7<m$ou#27l<`N;!mGcl
z_18(?gg>=7LCks}h$E2)W<`f=pa>39XQT-Vj)=WH%k`&$>R^F#7PHXa-$KK6WQR39
zae^S~%zOo<se+2m<vu-LCwr`cy>}^XE>ec##bs7rwNp7f^Hx~z2q071%i)snrBf4l
ziTo?^!3n#8v*QEZ(VX+n%^;mG0+a}B@~2Dg!BXpS?3-vxt9&<}9bEX#maBwW{*sHw
zGQ?(Vaj~Phts{FRA?9qa0A|+T3`GAK#O;m#e4{V&|1SQFoWBQi7zq9|eBj}MqF45?
zHzA;xGqhB4wt=FTC17M=_>V{6=;TcBs|TRyrA=(joXrUs*_r<L!+M%BcG&C)J(p?-
zXQZ6zQjX(<8W@1tAhtlw0tjUST0`I?yCkAElqpArUoSp?HY!v#wJvv9H~$#Zxs7Ld
zJea;V5N4$)97W%&Zdhka%6^+3M~e+<ij+Q*h$NScQQt}hsoWtN)j&-g)+t>Lg3yaI
zHi?s<ZsZ4?D~T^6t4s=dW>0FcaG5hw8gWos<-iODuR129;px!aW5^_sk=*a5N-W$4
zO$)li2wD@RPZq$85Suv!+$B{@OLw>C8Jp|Ehpc9g2y&cOGuGpab7YNZOHOhk8OTj0
zpc~*PKY-HQ&PNh~V`z&ramPDo&+8u-QX4KBwymBuy2v9!)T7Y@&9m#$z*UcL8fHYJ
zH#s3J7Yz~nKa1)M#lf49<a!S<fQS^NPe7o11%ic!s!t1pP@pkm5-Tz$1>K7x%t0i#
zX&)8d4>L(MLs1EXD}M+W>X0-mCP+=z0pl{Vp@9&KCj`Dii13v$5tZwcXmetsRtJ*>
zX%8eI&=n~d;w%smpOC`bAx&kAhwL*e>zDSw5<Z57Z-fi9sZWDU6Jk1me?o>34N|i8
z=jO-4M%hqOww4ZO1Q(Q`qA1lTA$fsOET*9jIV0@pC&rr~8Qf<Qr3W97>K4+YwmB+m
zhZ{)CFG!LgHk%vJb!JD-mOyGM0FtEkRu_ha3$u9UN7Jb?sv}J7gVQiBGEs>Z6alIk
z0Zr=nMkcDA4FV8CXrk3*lZAyPU9@nCL?S7^^;G)4J+YI%d(S6UG-Xv)HBCgJzd&iG
zwH&$rxpU{!su=pCf(v~)<JsR30*wZ@!&l%g-|ZDv4KV+9GcmbTQ+P-dO=Tmk*P}H(
zCwGy;rLxksHdgsA|I?zl$Kv<;VHJbEH1|~O2}*sjVM*zs^(bBW^5p2-#8H=aS*ZN(
zTS-q>Mt|y8p@euJ&P(A4Cb9%7I;F1$f=zG)i_qU~PY)nd#(rSVwGkY2gqsy08>UmX
z_WsTZn}z4>sRkjkOByrTt%vY!1>d?0eAtI`24(2C?g(w|0AOvwp>`vex@*g|>4GdJ
zYsa2**&iE<F!i^j1TRLDN0b7cu2mlyOn*qH{=ig7xXS1;?sU?$k_XUkyj4Hx!{|?`
zmFKox7#U!utli+je=L}H0a}xk&Qz7iE&ylBs^#j`bs!7E>F^Na<l%p`moz1S>%Z=}
zyFZ-KQP4{lo!8o)iJV-Ab~N(w@?R!+drsGJ|4@WkI2@LtG{$oRGr3Dm%aLb3Bb2`&
zZBd2{UkWg;6b|oQi;cRWcN}V!MH|+!8?tL=>J4U6Hi+rU<m~ur$8Td(8@EXp$KDPK
z^pi&?)t*W{HM%*Q_XN+`L&%rirsf6d&DkR~n6%gild{ZS<G85WvdT|GpF`S!vu1XD
zGpE$$RV`C<py^<XI;ju+xa@;kfgHA0oEy9}7|~?K+J-i_aSw;=GsFs6TIDl36*InJ
zo7Q1#!{DtV*!GXEoUPT~*?6b-rcigAtBzxiTj51yB9dIC`Qd@7ADHt}*uyd2qK(D8
z5`w18lvt_F6>O7KA2Kacx?7G($<B$Fm}D~?@0?c7lmD)=*hexh&hF~X%VTHN(jGKi
z6u&;dYd*c6O<LYNKYJ@{=ipPT&ZXzj^QxlW-WoNR^DG6tbn&(2%-ENzBu%w0=&LTP
zZojkX=*sZxE`p-oKP@}cC$r~n{~a%sCpv8sWtP)q9QDe0brc%oGTjC(-z;O!;$BRb
z&g-az_a}8ugcj!<JESczE>@U`L=^aOjfANte<kafMX5kKwEXhk&+?hz>TVF7bM*X!
z{IvCW_Mn`OetcY{Xv2EbKhPhIxfr?#gK_cAY17)}gjW($dxyKkN%!*FBSv?jIIZb;
zS_O8;#~@bWz*%l{p4pmH%T08`s}-jI%0@^QND=%QtC>&DhlQht<yT>gM)|khcwfds
zz}~6+6&kO-X0*idiS(^&-%1EODQd~d_5|K=jFQupTgZItpMGvZe<r3M%Ul}5$q8%U
z3GL*#n-{H9zC=FeTEYCv2S;%uU*;{~g%X*C%06bG1&ZG6P5#Jvalfz%LaR3rG=T-4
zQMSIbOLI9gP?n{Uap0Qe_2SNurMotDl3L^)wmp_iThJr@7DrcXU}_n`|1wi}KhXlr
z=3rIA#AS-5>7TX1_*y!PDb=;EzQxWyxps``Mw}dwne=m!k32h>tP!P{X0J=M>89(b
zm6dRh>g{(_z4h*!(=2~xFt9a*1QWygq%MKQ%3xdp#dfuk%ZKym2jFI;H}QYhN+t$I
z&i_xX{6DpjiIJJ}e~p9IXlVQguj;#}7a(Tch%pm@2}4z)O!-{f<QspDnuE?=0t6W^
z3<^SlqgkQ&eS6*^0E0$iLJ5x)!1;bt*VVwW4$AR*aDrczr3TLkt(Xi&6}ZBtPqg14
zz;QA_1@8-85pDPL%N|+*9ymY3IKUw#0wj80cKh*uc)0jJXDh@BOD@<7hsODSzxk$o
z|NcJscm4HZdU>$cwpWAa=lLPK&^@3UsRmD(F&~TbFPu!gbg=$)jh2wadU9L-eVy<@
zp55iqLEARU*W8M@p$uIR1A0FcxX@)7MuMZB3*bEO;(}Q_Zpof|Ijh(DVz&Eaq;zsq
zi)e_5o!Y0!*90O;g5y9$97|lEEZx);6Oj_(#lu2@K1vKyG@IdPm#~Sb7W#qPS+=yA
znaRax6+Nu*CdC3{;NT)P_@l51y_)E`<N}Lyl!9uVL>BTd3+v~*j{(gl#$o@`GRNKK
zZPPAK07aB40i|JLI|k<^7Ftl7lG{~r2;sN(v)8G`gB8jKtg1>4=%90<_#lFb<Aru8
z?`fslUWcU!r`lJh2yKn88lmgNOtL%*a^AHEKer&I6at!~D{WQiAXK?dEnf1}N}U7e
zkYbm14N6`t8B&)lB^wqnvl(nG2#FdEgQSJEd$qQ-IR9>m%IA>$rhEs=CZnJP)<-hE
zX&!O}Nj!=-S<<OX{IDl(E{|PSiK$iFwA{~|`AP#Jq+D3_o_{925KpRjyhiX`qjteK
z34n7sFu3vZeLNJfs)VqCOxR7AK^5J1EaTUaIO{cJDVLw$g0wLeEg%6IbG)K#Q1|nv
z>avH%?XnY6Y8oOXn60tsyN?TxbzS%$Bym%kuc-WOQeA&?774w($+~}qD8vZ}KZdHO
zzUCyoS=t2J2;qq3)Y%`-TSQg09f2jCcpfh_1XK2~yY>Idke%@y{r?iRlxJ$lb6YNC
zxwO(%HR_G7ErRLky839<g14BcmI~Y=`3@#X#2RL6SKBP5OVo`<3nz~7>D(P=@Zi@)
zfQ)3nt&K9J+OPp=(%IOkRA%OAcv#>&nQ*|UtX^g-j=0;!!glqv0`Es;Zvy(GmZ9mN
zs79NGgz7<?#v+DL9V1>hxvByDg@_*N>(#9vQznXpB$BvNN?)-C#Aw{We(SamuQ^W6
zZY3BiMfKk+sW<Ykji9BXOB5Zx+^W85mGQEtaGBn`l7AH0dMkG^zb|BpyR6*!G;7#y
z#F)(MR<+gXt~jOsH;Nfa`EuTCcW^)ADTx$OZ~WcT5ZdX!<l<`wec?$_#YbhOq#UQi
zcq9YTI*}Dr^EqUo(=2N?NG8b)vwIXeRa+jVI#TjX6yK<ut+sv?jI|n7-<%s5=DYYE
zh?|AJxWN7)@oF%B(R(#DcoOFtvthZJMhgHTm7jojnS1Kpo@?gAM(At?YF#!sp2ToT
zhR@k0x!rHeU0Kk?GzqP8W;u3D^Epm^kbg9W^N*tg0}>@ujmb6P?ATurLXxx0bFe;)
z7()fa$aT>G0?Y1J_fAWg;pkbZ0duqVQ++zXlkspV;<#eJbJucW?<r$e7|C_$-+D8c
zM|h19GMcfm!EQCB)Xd{-HSe#oMq%0JXteUe_O8dJlA1VTy0eCJ<>tn#a!e8-HkSnr
zJwlFGB2$)Dg6vNYh7W}Uw*xd^$@)9nZM!d?VRZ|Pq7i;j31~GXck;H*gZ}L3q#OWa
ze49P>#*&4rEQtZxU!BWfC(DjBj9h3Z5JAY|uQVJR@q~%uO0}pgfFYjOtYX4KNnsH+
z$P$i9#!YPO)Lk+UuN20GYkbCGLy5sUsy?sIa2C)JwaE@WSSBvCWdfg0CJyMjB*@b4
zPjlJ3j(Uw0gkmaT-uTbsQe61eUH{7?LY%}fUx+}DBr=2iGLhR*Sp14H7?RTeD9|E4
z-H7N|kcHDMid>>}0f2%Gbf0Um;kw>$42Ql7VQfAMm;~8EmTy<oVX~iPOQZ?e+;n4_
zikXm9nsN|VgX+ef4fW7HiZI-89QnAA6j>61PDVf+;#!yk8|Zt%M2V`UV80v22C53+
z^nGX?&3a}L@rgiH?u0V|DSYvt60FAHEQSJ%J%>T7HTw$0V#h{)8@E|&pB_6lmckiJ
zqoc%Ae?W#!Yu?L<1qtrbtZ0eby@RYQ#q2bHaZ?sYGZ764fD-CclLzy8r4(n7#YVO%
z9#qx`n7e@#Fg?wYrDLeB+0x}jt$Mko*2X3*`4c#a`rLFC{1>w{p-_JC`htNbVm*0O
zjRu7bKt|v%#twSchWG0(rR_$I9p6@VDFH79396B7;KaTtHKWpN-cM58g)p!cqLsbR
zyC%97>kp9N<X}7O&wzFTy$7}2b5QA?7{wo<hg`0Gu7H2dvsGxjSKVQb$Mz>t?z+Vp
zJ717o?_mCKNCjBNM5fEHx9CaA=7@4$R2Wbo`UoP^gmn^+5>+ghgCGbw05D@Mh<~4~
zs6oN}vVQcJ7Zly8$w%&_Ywmsu6ovskyklhb3T7=w2x#8QRdh|eUPePC2U<CLe;_Eg
zo0Aw`k!WFCWm&tKFIr}t%FIn9jb3i;xORv!YK@qd_T%gv?<5)une<HOHUpqY<c<%q
zyY(S4Y1h@>37J>lr)c9n5@Iu2N?`lW=0x!|B73d)=6OeWjJ{vdx^ZD0Y@=8F`~%+B
zW?|yqgGwRtvI-WHP}+ct{0}}KMf8G4UTqwdgczBlPUcW~s<FG`Ra73VSFZ%fOIgL!
z(*ET_xIOb}H(}o_HenjSxjS!bnC(PKf)R5Pwr=e9f9HH9=4}z$zdIOiQDBa_q|-a<
zoG^V58Y-DE?7aqMeuNA@@}qY}YyUT~!t`HO|Ckuq|L4a2T2tyj@#FuOYT1Gm$sRFr
zG|%QrmJ_knc~Nm~Y#!0bL99S2L8`*_b*p0!*egBZc+t%HXc$OPzpit2xRqjIk3PSD
zJi`7ERac9_8bB!OggdPkFdvYZDPIxz#`H>h6v!Y-##~~-(f9WDzEww_0CSo~PsUtp
z?*8!f6m_e2`1*djwx_+j|9X7<bPP-9^Yry})b{44rz5ezD@G-h0>*ZiuD&6HzU&|x
z5z($VaHp{iI>xu-eN{X4Hgby&6I%iZ9ZZniATSz9fkxAc?y6}HI&~;8EZjsvv8eCz
z?{MsXU3?s5+6mY%xnHIy1KW*1dx8M0$Kv2NOoGTLG*XJ)glm3#$oV%#ow>-M_2e2z
z$*{yJ8Av63Wp}1lY0g*i!s{3N0ReivkFQ_GSHZrNJM0Ng$QnBQ?c9ua;#nS<u^&8C
zGmSwG)VpkrX3;`_pX3-3ZoV>}!qXHLlo1_a3~cDfx%<7q%dU-q=1+}T(taEW#tauN
z?8ARqWid)&Zq8>A9WuhFN21ax{6m%K-LdtTW6#cR324mCA{R{}<=XL)lcVX0lOs`N
zn8A#Teyina_eG;+R3Wub3c9}P{dfyq?Km8mS*YU$Q-lSFq~i-}xJ|&D@0Usr2ujPV
zMA^$)J%ph_U-mo`tYMNzH7)tssik5OJLK4xS&uOtdM}cCf;y{u<DaR=-zM7xbWZcU
z^#umveJZPKdR*OWz29S^3y#!WXc8Q)CgR#<m(mE-vekxUn6J;1U1xPuop?bGZTGTK
zdhc1q8MweU0srdJHi)@z<?dpAb>SMqe_^TSZre4jC=!e?Zv8IS=+h-p^9}RIWw^V_
z&Ryy%D8vNxlx!qe<N%4^+SN@%!vy;Cn~suc{ZL#(1l`0X84v!&qjeKi%iAGqK^j9z
zhucD(Ji1B%16v=qv2P3(hbKpY$j~Z-`fgdgiR}vFLKSm|bP8fA26=5-+8qbzPb|r8
zB1-fy*F|XK@8p%EKp{^JD$_typQ6jnKP^!~1_dN-ca>{BhbIoO!rZ;E2g;=SV7aw?
z$Pg*0n4@+z#O2r6w7n)*#Bt7MheAh2bL|1>4MKmk)@Enp<30U0oT?CogD>g~ALuGx
zdwO6TAe6Q|R8wcJ7KL;e2VC6lYvBt*N)&L~_Dzl_QrMO;7PAeKAh)s(S1zTma*@Oe
zsjKFmf>_r1;E&!ddy*-BO}(YUA;Hd;O$E=sa{vm}xLHls&A0Wv#|lG@|5nd~y=}IO
z{G7EM&NRs}m_v{fhjhbuf|SLTiftLO?bF~2Rf6HSxkBV38~laLuhaUlO={PFLK?!v
zreaF=m9v3o3u^p0>5s((nDGd^Vl~Y6MO}f_gFIGo07xGvNH=2s1@u~?i>Kd%kvTA8
zNO?*W=Uyh3#qV}IgU6@){usg!fcV2D9JE9}kTf@&mI)_#*ghPVJM5C**m#5Qwu#WT
z=ohXIpNG)^M-&cPsh>KSW8bo6?RwkvX$xRAX?!VamsBYk<$dwF>+^g^45u?cKHw>E
ztuCtDHyi-Z&(JONrHpL$yYd0F53K$45KRc+TrkSP9K4UqqXCfc10Jsg2ok@Mg}=IS
ziX}8m4o%p)i$oi(k9Pbn)SyYu`sjaO#{($<=p_%}*)wwpfMThQ3=2SktkCC~@37?E
zX*m@iKHRVXHv+U7;#|#ZU2L;}fQZ`QnL(D#Zi!n5OOk-%yj}EEQD9wU3r3JRc>K7`
zCKA%8=XJ(aN+c&hDR7{W*UuOOsQJT@f_usz*f})(b%^MRo#tJA&cX`c{W$6Cj*rG8
zqX6HMkat(&IwnplZkq`tlggbj+DYE=n6~lmoprK;K+R5~HGg$k-%!~9SG6KUEeo2|
zC`RS-OlCpZEx9@jYY9=tsoV}F@|nub2MPtM3lYZ)l3}Of%Ybd~xTydtb5(G|#@(Kt
z0C>B(yi2;tF~NG#RZ0{0Vs_QJzIREwK6?yzG^GZ@V+<l0!m5&$oT)p6ns<Lb!D-Lo
zm5mahM46%pLJ&?{tF|enivV#K?`{L(?(=XnRkZlBF#fz#Nr~R(o$=*^ZC{53tV_(R
zag()h<wiZMh*osq?c>8~?W(`psn0&Pt94NnTlT5JyvZkDaKPplKgoz&v^fk5$VrMd
zu7ek05t@s*gr!>?7|xY{Wn*tS<s5NLL($HL<X+lwnT+A9_x5BRm}H|hmA$qanNOms
zkeBhV7mdTdEsJq?EfaD?TpB$5qT)jf5u5-T>*!ex?kycmtvVube|zigE(H@x(4u~7
zSNUj53!n_}tc#7VJik}q;BUMS+3T4b<FydJ-}PE)n8(E+#Sy;cd$%`6%7$xz>=4Ab
z1nr{xeer9BLonywX|So7-5m|a!DbkY^61<k!UA$7btgK+NAAWFdNzlwygvwhNQG=K
zP&XFzZpyk@HCdChW=kQw*1ivfmT(mO-3)mS+mgpAL&22E8IIDlHro!YAY4c%#VYLh
zzw(jRp9kZna}x7SGQ|sK^89q7iK&<&Z~)s%hrOWmQt3mf)1qfV%kzTgYXzzqLA_YT
zfXT|QgAN>42!`?1ry3^FlE&an6@1ZZK_t>cv`9eJd^kG;Eu(jhW@ZwM*50)le`@4^
ze!DZ_1+V&LR#(>s^P?Pnk^F+r{5~KVo>L2^Qm$<>-*z=TE>_1w$Staw55q>9oLXgf
z^yJsLhj-B4-tN!AKivCy`#pZZq$8ri|4nHy{}&V_6T`3m_+Kem8d`2!!U(=|b@S&*
z(+zE2Da|N&`7-%D&GDH-z>kmPtyd7vjGZJifZyCrnK~MZ*~RCuy45SXre=K^8*iee
zyR${RZsdD(a(H<X2Ns3Ec<xg*<Un#0{_+*w$D<!BrwR;SJ)Z~vFpE4HVkkuh2|x&U
zs>}BL9uDsxy<7Jq-?ig6gg}>n-d4`v{%)LaJfN4&*`HmVqRXZI<`3LWjC22kI^jB!
z^jI=$d9cOi^4n7>n@ZxB!?2d(J^4>kJ9ilovbb$lPPz>OMLLeBHauFWPO|%YLhiQ8
z$WMqYKhNq_0SG<DbGC+-BS{y80(>wZCKO@-PkNvmA)z+0v?dbKTOYd=yh>Oc2$ge5
zo1ZbMCnP%DIH65H8)Cf}`vd&gdwzlJdkCVWAV@8tzaLwHsQ5XR)A|jTBQW~Ya+Ih~
zyTv8wHjVj&TqP$Jv^P!5n+@H<Mzf&LCdy?gPA|o#wLu991!?f8-C+vJdU!{)i%l-F
z6wj5k{fJ%CBZ1A<^}rZQ;_R8&+-nT7pAE4<=%C`lIc=<6KlqooV{R64LxKPzqCpdB
zjLv>;!ZcPJd^P_0C;H!lp&&>!1YZoO=s@t@=WUQ6m1to<P-hP_oH~G^*P~xiWa(2D
z5uG%(-9g$68eqxyp5Gy|=#rrq2JiIiROX3qPnje=3Sv@zRpMTfi<29O)8J6Ti7FuA
zDiNT1sC+C`w5I@I`VU2)uTruKo|#>x;R`7qq211saFj(2lDDlAGJxFO@q#!#0-N7U
zjve09LaVimI4tVx`)bS;3-AZ`wg!9~C@%Cn!0q|q7ae$~36L(b&I7d<%qF0@FHz_i
z#*;JJHQGF-KK>T(h~IfGu(9ofy!F^@UTE7Hx!Z1l=gX^|-J4vlD=RihCTP)#tG2x?
zkU^WqEQXOUs0XucA}m?vRJ-IM-Oj)i9c3cbP2$DOGH3+O<MmR`J|bjCzw?XZ7KsiK
z+-GTKh%fX;*B)r2rB<<4J;Q9h)+P(zcyE9n$pqQP^tE_Qjzz$&i<VAlnnP%h1ESfQ
zi{+`d4vDL(^@{jIF>p3!5MFaAMe4F5{1M9H9*KjKK_-+ElXFc{A)R2;{1~zLUC(7>
zz0DL^a$Y+gc_lujnv0^2tNLocFK7xtpQ&%e5ZXD2+K8(Ic+N=8l21NIO0eY=9hLq2
zVt$2s7+F3nQTiUnBFmWXG6>csn_Uwf)%<u?vu&mGislUnGY-6k)gzR;wxQfuC>wFW
zaye!}W>2tB1{!Q^Y0fPMd`w}`yEZe@;L}Ql9jQaQ7bMPHFOQEa+tOg(p;mNU&|!ZR
zx8-5#+N-m^ZML#EkYv|z1of4EP;6k0R&%-lyc%|b{AT_QLCb9QN9N(CQQhAs{3CS(
z<3wo?H99B8@(MqI`t^-t2~}NOcrGj$<pjd`8~_zgU~tgB1>bwN*%u8rW%nk!@-Lf=
zfO+32)?f7dXhN)DxsQ}e2VAJvmJcU`ATo2&j9-RBxTbv(O_LQSxGF>%nZ0Nd5(Xou
zVIq<~7OpKJp`AE8W$)=HLka^$zo_SQT*Ngi{6QluIrT-+J5Wu@f`E=|9@G9jO5&*a
z;{g+^>sd{-S~wqu$B~Xa3gw7`fOQXLYl0~~W0Ktp0-N}r{gb9m7I9_Z2#FiA^(wO$
zQkI<d$M&ifp#tF$REU5HS-x!3Q^~uY%8B%ErbN^NPpB0PhTo2r=|DhVvQz>CES7C#
zBN)lqdmk$3M5dL=+N?k2X735E!lAV<{1iTZxu#P_b4-ITv-!j)yu$2ls$HI|cfeP8
z7^nYIW~`gmr8#)(uXAulfGQP(@H_`-%d*#Ukh2EdIn(LBUu1`X*Cz()&}wRVCiNyu
z;|fXiZb)VvuJr+%7rFH4NQz9+?BVl67#jAKdc(g32WHF7$%^{uJ^7n=W+OM=L0f^R
zzW%TP`-Bwk)GXB>i~e}heWcPlrzSZd1v%C1JYJkki4pct(NDWRs@;5*!^IPdnn<;h
zqMlU1UD*~GBbBi3X#pO1YoCYhd9Rg)k6!HnHIn@Cy#qkUB&aNE8q_YqHS?E`g}d%F
zjYnE*45Ub2mCo-W7LmR*;&TS~l-g^cqszsAPgyST{$kA^N0uEfbvOQaFI<kx3(0ia
znw5|_z?B<zQ(-I8Vmr4)Ndgn)iAEuji}ex@p!*c;L6S;fe+9Lf_+|Gzh4W!pMlQvp
zxGReH>v2qDRlN~Xu3P{ukp(;$6L|>ZO0_A$&)^`7LphxDauzt<d;s}Lk55$jg(k_b
z632l%ctj^j6_~IRrm~3KG1@V!iaN#j9X|1r{cojw)*(0>sQ;85%jj(_U$6G-8ye&+
zKP-^4og@dKXYhMA7bNxdmhOUe%A8X<AL+OxImSr{HnWY*bgK!_?eH6ix`zgQ!QoQ2
zV-^hJCYVZ9yNGA>=rz}KQ@200JQJW1(JeSjsnw;abV^0tFoUOW;YQ&avdp~LbFZy;
zxg=zuT~_a?bEJRd71M`p6zt9%7K;3v)*I`%-ucVV@MdtYBLZwJ*>t9Yj1JD*Z}hO~
zgHWDDY2t2pn{rtJ4xTwb@5!KTw{j4d8ecE(e8y-#$=BWFl5r}=5K5q8eZi=P{%j?h
zz0#V}y<_H}*&e6Q-fO;{!pZV7!Eq^@yKnRZWGugJ^Iw(2f8%TZIx7FGa_CW)vfC0x
z_}x6bO0cYX>XjfmBeb~62mRX|7>JiRm`@@}G#xe>23&OQcS~bNnbb(T5xW!A(=@b^
zVQ1!whd^vKI3BjpwQ=^C-4q|p!ZH}Y5h?zsnJJl4##HKl1Ub=Bb`04jojuF9_x0_1
zU2XU3>dT#o<)PG$e%J3k{Bh(o?sm3R+>fJ+BdhR?K96qHZ`PNWSFO=b%wn<`5W}48
zu4X=*(>OY`cRur~?l_a@(c*nu$*gHXj~LDr^FUZ~l7n0ls}hDh%G&sC(L5erd|$~3
zhSI1J>+4%PzKhxs(i?G*pZ4FpKl3%v@E;`q!{{rZVN}+DxLoQ(3f%Ts?+lF5?rw@E
ziXQ2#H<OIIKBgJ@3q)0T^O?*f@^Xtfp5jW$RgFi(&BeE<3d@syT0YzJKK#05<_m15
zgFwkzxG__&j9G;rxkKN2w5b$W<(GeX|7K6y@S4_ldGJuZXbMjd3QtfGCWgW@0kg^E
zRQ=1nCRYRC0!6;L_&P8N#&PoTz1YR>!L<`V-|yRG7CSF3{AK4r^*dz#alc-aPJ=_-
zqJWl5ah*k;%UESfD+{2)hfaRGl=qf7UZ*(5L;MbtDj*9y@BJOi<uEa_WBweY&Z8Rd
zn=GhUQ|?u<ON1z^CL7_o$L=kTbXQnH)X8s0Eu!>ki<uzJEUn{Kv56$nDO`6eVX3)X
zx!-BVzojrHkh^4$skj8jn4fiRP1>1Ld;z*g-|yV4^ffXjTe0$l?pq^pm>7~vpe(WY
z*RDxwtSeUHT2qn^wocn9kh(Zt>&q3h=_b>|ADr?9Yeg2oQjoeWyfFNem&d)r<uVkv
zPlxpOb1&sI6o<%@Y;Y@wF6?3O{PWxmm`iCN^H6!ND>1dlabVgNGv)P3^GBiuX?9K|
zNh@UA{dr--1$NV$KK7zrShOeNJ~lEzJj(8MMGV#z0_Z(XCXp+mc$78DHaq)6er%&v
z_P%8cwVf`~I?A=Gas;N1;)r)1uncGt2#hXZd<$nDh)|<&3UK>aVbv5kUPW8CO<Npe
zEYiM1@@j`xrP^M{I`gv<bnO&vf4p2avYUGs7noW$y{&Eu5Z<#Lwi~e)kv)`BQ}lsE
z#<E;EKDe^S6C`htZ7w7#pR|#zCduZ)RRxxC;&1opwTreYz=Ulv`b?A+`7><=yO$>w
zfQQ<Y+j1v5HYq8J2;Rk)!s5vvf}c|b$8|IP*{}!^OTn+x)=YAj5d(O?ztc{E7B?UE
z_?%Z3a@i`Kee88hP#+ls1tkn283DKwQ(q{UPINq$Qo+xr>F4)sZK%pX=9%D^e8CvI
zYyIob0i~#qSt0y9z$ru;0w)B4HC7|xOb#RBxHfPFh1ypZwx>Ho{>3(Vk??c44WtLO
zlRi>23R0nMOQ&CMeP!K@px+{lH0+AMC9`qM66#X6_8Lb8Tj!(@dRqsV|42Ll@j}B6
zPL2o&45QbL#GxO}0w?7h*7oYqa*QNs2S^=}j!wjj;&PpIx8@JXG|V7B6Lqk%r5}6z
zF7S3)z#Mk~T4^X_pJf1`Nmi*&nuG3slCci{ugSkXZM2q`-WKWr*{NPeTv5fLvv2?p
zZ5tx_g<rl2r}E#PLze$aUt#;730N~4GD(N6h&}&Fe<kw<{*%eimPyIEw;d@n=51?%
zPaaN47cVAK*eL%smpxtyL>wtBcs7*Z7>FQ%V%F-gV{Bn$Rv~5hc;*)#KFY%EcC}LL
zqwPzL?;o=va>R`y8a3{lCHYFU6F2DP_+6C}<(x!G<$k=m*q&_JP9sbGc$cHR6>O(Y
zaJY8!DovDDR~BTxPI^!xNL8J#(ej6}#0w!GtSDug2Q;~>S0;{lbW+An<mqGuqi;8l
zWqHgjdU;n1tQ3OHI=1YhbQD-eDsz0(8I#baOh*^5Ih}p!vSc(up5<h8zX+jrKN_Ci
z<8NI}T3qvyMJa9vk`u$FLDILnHS_LW1WlbL)Zd#O2w<cxEu%(e4@PR5ZL>U(HA}SS
z=bNb_Q^Yo(43u&As!aTEmMuV040M&)-`E~)E-82eZ(;h&WhCSaM(V;t6H4}jsj!_P
z@(n8MR~sPho}Ap-u1}B2_Si=VyN64@Rca%R&M3cgxKHB#;6w6_gsVh?CqY{#aiR<V
zKa9O&lPFBEZrQeN+qP}nws+gMZQHhO_io#^HTy)|8#5o?h&ex^vNEflTnl2S12xNm
zMrsMaaD0yoVfjyv37mXOG$^J#GuQ__>2yPk!>1nez83@43a=W0q)8?Pn#4PtP_bm(
zPL>Y0H7X2#cf}pd)!yZO_a>@_1Lh~a#@*ks2aHj_o#n9?Yen24-*XUQytPn$kzeD3
z1An0)hN>dq)ZcKWhe}ih{Iro)4mm(aJAAFF4+YdPj&bf1WjB4AsXWT#9pZ(Ca`}}?
zT$PboS;bm**(hAPfSkpkjm<?NMk5$_1#~^(ZKYDc0w7wCbDvGWabs%uI6OwItOCk<
z+ANOsTST6Hu0Mh<A8DrsCU8|R)i@3Pgc4-K2nu-S{yJcNEVCK8$inT4=_uaJFo=cp
z%SeaU`{Lf&FEz2Cxtc^*;ZWkow9S8!D3`-~sVzv;^x<mu6!cAqn+X}BtI>519dYzf
z%pe1lg|iqoyS9QGvyqErp{rX!XT(~^8(qio$tjH?NOC&RkS_p{Zozn58EHpBMpf{W
zxWC-F-*_fC3S(J`N4JZ25FoBFelw`JKy`#zITw9{wbSu#BoTRM(2^t&){3tM4UwVu
zm#_V89x!GeF+{p>&-8UH%DrtIgpG7gTH&qeKnpSnhs{MU(m9<4Eue}F96HjRVQHE3
zFHu9^KT!-LF{*AI>H2t1IkIg(?2LE`BSLO6O_W0JcLd=1d3e7b3$|6lpM!@AK!y%d
z$)~o5%JeN72gmTYJ)lbXKMS~nb%tS2`!<it!umq8?Sf||lh$qnsRm74=OkA~SfZ$_
zsf*}079}|`=2Eb3pBZf<Wj(3X#x;V6`e_ct%)~eCVI+yh2w9YIt0~98-y5}&%-2Y0
zOfWOo=?`*c%<G7xe)$Int)tnNEiIZKQHvZ0ZidqgS~$pxi|DjTvDC7m1*4G|BdnQc
zTmqJd&D-2PYT>&S#kj^FQ~B-%jTc;(tJp+R?vtkOy7Ey=?Ao4B*)&Pn9xG#W!~R{6
zOJnf%*{o|D^SOZ%CkFtF+E+HK^v#3rL3S~B{fWW}JjU#Oh^E@7h-sh%@q6IGpOQhj
zarhhVgpM~_HKNsk9Va4YjGFq#vV2l^n5|=`K&udqIUT>q?>&8N4<0dc?I4FB1$5K=
zC=Nl2mida9N1NYdzc0^&=eiAqJ?<&ZjN-y6`)U-4H%iMnD!1KqewiLUXrjo=QYYgN
zDU+e?em=~jh=89sptaK5aqJ0H(i&zvMlzOPl#gD<+i;qG-j5e_Q;4|CW#2z-H8)o{
zvc9n|u7UzusFcY1YkS-vBiFLQfrl_&L!-F#Jf|?^?-uy%?}oZ8g8Oy5+b^7gjZ-)2
z;y5S)j9(aN4ltg4p5_P21qW627p;0-Lz{^BTqjM^B$o2E@jbCY^Q*v+bK@vmk6%jD
zho;Eze7QN^ujsP8BlC5?Q<BSy?+2c=T_}ezfNGoW=Z`)O)ABz+LK82M;(C)l&Gf4H
zdZFmRH`X}EbHKYBD29WF0;N*h-vU{BPuE+hU6@QX#eKl;fA~IUM!huuzSv!3q)2ET
zluR`8*%pM257rcT)eidn-P!KQ7cQPFFh=L|veg;dZFz@mV1jb9`D2^h6?m5irP6(q
zJKb`5n&6*$KwzvtjGpjU^=x_!)kRFxh_i8BBi60KD|JFp5L??+#*yc~w}rbU&C@nk
zMQ!RD(h@17eQ&xR(D>T-Q%STix3HAom#ZmkW`^`hmaLfKkl9+_taz0sQ!@ItZhAD*
z1kDua65gY`Bnohj!`5pEU0up+bVef0ngDZ!W6*xcWFF`E72qUW+qZ)Vi}d)KCjqel
zY?MpCgJ5#)dJ~GJCA+f@0}3md<CS)}c?uUSnyGxWSz?_f+O-f-#Cj#q-7kzeX&)o~
zpuum28PgZR^{roi2W#(km>}EQs$O&QkXA>?_=wfu2L+Mmkr)h8n5RlZgv-2cWP<C5
z@;Rc+Z2zb9cyU)VZA#W+2W$=jOMwl>B-#QOIOlCSH`5K^&+!e-3~v9vgH}$eXt#m=
zMb?wF#G!D3Fx}(G!oLS0AwxlMtB2nD1wS4x({jQ0*bVp+{3yUYY-WsTq=!D%BZu<;
zy!l!GBU#4D^1tV_XlceB{!5nS=@0OP1XXYNB-#vH6IfU0fHWjr$;AU;^D7{+3I`FH
zCTw65j{C{0swWx!p|GUy_gy1h7k6>V(N75#e?^Sz?elPPdt7k~+#F}h8Yk)86rqZU
zmcv*j>(my%HLA!kXoF6=dwcHteHz;RefhY)mUr{}zTmDI?P@mMA`wVtGetw+cJPyH
z?gyn}To5XyX@aEf@}hXjGV`NuY<G9;iVq#z78azxEVP@tai<Or|K8fDv2pEc>@7=B
zRc_PObF0x8EdCp&y5Q*q%SgRHj@%~XWR@I@Hpr9$X;iwI=Ha7{>KhcFu*X5Y^$n6#
z*|+W2zTYeb!Aw+@A>-0_wzi~b%thJ6cA54C!=nXv#F2+sF%N@sm<H9JFR`dG?F24e
zGAVZTZYoql97su_3KL~wWCqtVJ?_l$G>zJh7Xjq1FEOys#EgUp)vAja0r*dENy6MZ
z!hsKG(hg89Q6gj314cUI1N~Q_HtKsgkxf>WH7;nT8#Jh(&8e#@(pJ&UL!HgHdVLx*
z`Y6>jNT*gi(%p_KNgKS?X%vWwtonxs8sdzU26Mb%vje(~Qvxt}XxuPEwuEuvsiQA$
zR<`H;2idKsx9j~FjC6z;k#e3%M|3@zbdRv=8a7sV*#B8GS?G>|X#;)X<ODx&vN%pO
zJAjNMX-^vg#6p?jVRTHClBK+#EaJgYkvx)=PwVQzr<b<VXcsI14!1wn6?002>+HBs
zZ*&zi9F5CZ%fw(Wk<wuJdKcXTb#%q;eidKlk0)`400|xuOgn`8Y)Mz}r9R!)01k@i
zuONvFI5@FAPMXo=OVpgP7s9z=+b2gWf7GSD9XjOc{e2utNbq<+oospWug}9wJ9%;V
ze5)q<%19L@q~fq|w~H)%D8MX?0^H5Q@#kk@5%ll=u_?7W4R1xU)y)VRuzs)2z2mS$
z@Nf4r(oFQkNyrOlEbmNS^b}t^1_afe1nm+}%uQ=BuN}ISnWJ{6hTLsc_vGuFojl$S
zdFy@5XFcg26B{fg{%PWS1ZIK|%7jdNnN381OM?)Y2oy!Am=nJ;t}fWnY9kq3*AJ8_
ztW9s(nnSEID54sPq&rQec|Zne3e9YBOap=P9USLA?m~g0Tz<L!e6;Mxvx6Rx{{WTN
z1}m-uq=$H6%KQ{ut@WW<2K^M&zr#(V{adXbe9j<SB;jz^aD{g#`^2d?%Fj*^ggLk_
z@TNAL%U10EI!!${NtxPJj-rD>XTTyoFeM>hZQz@rZwT-P!~&y62rU;IKRX)8vga3+
zwY6<V)f-NN!n~dakdsQvgBu`MGov}r*{ICHFRB}7FGeAfBTV?MCZWfHYSUw<J18I`
zA<{mxf(KMR)e(Ncxb_ZstryXbuO(Ql^I4FY8~otFWKZ7-I_7Sldv4DcsQVhbT<-BR
z@Y5Y6Gc7G$TEO5G#UAE)Tz>`{yA%xQ3jfL(S_hGaZ7>`9IX(juj9U+`nG{6oxH<3e
z8t;Z)qz^V~e+@%V(NcDalmtamsv@yRX#aT#JqO=T4(oAzNO10gvP;kfxKBxcy2s=u
zBcpR!Z8ngPTqJz(c)ZXL4-zUP`{ex4DVZ~DII}k*-^2q9qNWbuYY>%hCy+aEUn@0x
za9jW_n#OS+@cB-$tu9w(*2NuNRkWD4lGQyjoPN>Evim`m+xxI`Q9sw!TRM07yhT;M
zm8Lr2nTIezq76OJXp&q_8ceHxcl($qg9_GpSlez<2xTmVrXY;*-7?fgA9i%=a3p;{
z)2#x&BjAmY>>^M*x30Ox95S*}k`8=w+ISW|em^-NB_#kWSwCh89MVYN<#dcx#lJN*
z$=H$($+aj{Q0L{y@mD>aZ^^kF{GK`@ZAAE|7euIHP@>cZ?blRT*@U#-cwSN*Ic+FS
z1&c;Dm_{c+1}R#D68pxnE$k42);8-^?t`Jepu3Osh9^rB9{h%IwMf?ttkx4JnjRaH
zT)H6rCEV15#4s^Ykq-UFj1V*E7z|{Vk`4Qy5ya!nLHUL?YS686-iKwGD-?(EIS{YS
zH(Y#NhADOs2Y*dH6&*uWhjR<>$_w$%Y*eT;;NQem1Fpjn$>Fb+j>;!#lut<=i9R_(
z{;Kprj_6qT<Npl!a9)|uI4}JQ08p#5;CA0k<%-d)#eX|-WzUR&*W-5l_tw{&;-dji
zmJ6R?j~_0P2@wB?FBMI`tCG{B=f1xfc2lANiYY#y6}cV_g2z;oizO~oIPAMx(wJ%*
ziVHCAP1ZGQerBi^S0ibtpogQS6bNDD-S{ya444!{<e-%HY-m&UM6jn7S;4P(9gwYq
z{>u1`23UVYS$)~sDH-N-AQncWE1cT%mb6M`o)`~^-nH0f{5s1+pL1Id5I{^H)1Iho
z)|-FPk0x%eT4sW;jzfWsg)Xf5O9yQE+;#4_`;nSvYf&*GwsjDbzl`VDC|LjI@lD~2
zr-@!d?%eVn;ur+>493J!3uM9YFI19p(YNSs|Mr~w8aPLotfptBx9{f8eN~$tDFHO)
zqU9y3T@m5&&t#1{W&TBiem--ESpE8md=I0y1vSDyZ4Lu}Mao&a{yc7}>d8FyXCo}5
zMnVn&+h3_cDYKF><=@Vb!bmRa?zrwdq<rqT-sl?yJL&o>{|B8yY)0Z_(6YR%0WQ3_
zP1X0weV_n^SwlXI>pXsZtvNNhxJrv}m4*R=2t;R9G?!Be0oMXH_I^S6B&#HYQvh<X
zqiD8`F!Kb*R0#Y+Kvz{daMIpP!uiMEwKhIn(q%g?LlTU&v{zGE2VVaYi29BQea<&v
z<>S!Y?WT({`u1w*5~b1(I8Hur@i1(wA(DAWHpPO3H>m-^>khdL34+T2EqT#mrNy>}
zebjt4O;wgNKztt#+&ypjc?X0nZ&)cH@=*wQ@+e+OSZ-J7@@7b;#nzwj&igFLnGj2~
zNaI*o@^HrtaA!S41u(UzswViU_7IP}%iUe~Q>hF|&~FJ#T+y!p3e0a$t_M+oKJo-%
zQqazX+DAJNF>XEn>r5=0mSQ~0xK+~fuHY_M+JdHIJ_-GUxSC$_DACLV%Uv+wD!sJJ
zk5}9nz9WU}zfZC4fO)1pm2m0us<+=9TG5(=0gTcVUXm2JWop5T_b%t>3dVM$N(8HS
z%c}^NjdcPGX(;js!qPo<xlh^Y_jT<Yxox%{`o`+%r32qwt0b;6kRs127<$$}lm^5G
zWs_}FF215$76Sgafibn%oMs)>j0h;o{^rc9;eWj99jN7W6EXDwBOr5ht1v_JO1+{b
z#Q0UnK9PopfTCa+0qQ{sH6dkwZ+#CIutFQmf+G1nougU-K$CwTL#RBV4f;#03_t^f
zGT*F7f~rj0XTkhL_OeF`qk;!^9aaUWL6CWDUb5?bRc$Q&58|)?W%A?U0D3<`0h{%4
z{Z9A(3PB>PAAbv<uAZ8%z!~t9Fn*=q_JTczJ^;{UBUS_ue;Fsa+{9`xjpzBj{Z&V4
z!GFKcbF=>pbX!)H0979KDyyT{t~^!m_#;!HdSw6`D;REXVL|b!KGXZ+mE0G4W3K7|
z%Nt|5@oX0ysgGLqHIIfo`9n{%g=x*tX=N4jaY~6?0kxUU1=_|i@En__S__c8pC=LB
zHgDSkjWh5aXf`^1FPRKMF?lWS{32Ak>@URPZ3@V#?)N0b=ILPZ3TTk{fn0Bpw@Rx*
zsohsyLN2miNrJ!J%ZH7a)#_+4{!Uo{v8DI9-)J2MH+jATh8=LQd|IkD)@WNPe5#Fo
zX{6sf*Y(hIYkg!*72L480KyCJLd5vZD+eKYQq+UrE*HrhExhu6p%T&&>1$q<hp*@w
zC8#m-5oy+9a=>ZFSphEv#Qim)xtr-rQ(0LTjPE2sWw@NhM+o(Uf5WH@_+LFe>wmoF
zF){xOOZ@L+5&u|DsQ<P_2FhUA(F}DUop^l1+t&Z)u({eE7q~ECN)?jCQYsNvY*%wn
z*xv%e3nZYt9&Fo|(E~vMp8EFw5m2%}RJ@%tw?8qbr!ToPLJH$~I<b}y&P^mNQ1W<8
zd7U_A<o)&dSxrle9__J>wnP#YDB5<uPrKcj>-%-Ua}f2RoAAgNy3qf6bM&1h`rY2W
zk9)=6>3P4I+3f4d`9(&j^ZVL2`e)9B$A=ALwlXaw`Kv=|&!dKY+uFy+^RoQ<*rY#)
zQD1`h_%DaaD8G;%jQE(^#oR<E1o}xFCrM;_rdrX)_c>-9a;0Jd^w$H1Uuv&lU=sWi
zYql6YU_p4_8`vl~L9B^gvPJUWzHRZc-@T`S$Q^o=mv{;R*UeW-T2@|pp+}>3HBvfb
zG2)&>&`=x4Ua7jox1?4a5v;4o>_cQQ<D<!Rc({>$FEHlmW_2h-!lZnZX~$$FET68q
zhn-o0pRQV-qphwQhh!u+b`9`#PFMz9MD!scu?j-mHW`6=G{a07c#@F|Ix`(bK?7E9
zM@JaEhS0@AhPqhWntkgzC-W?0Cw&qSt6nk3O%m05iLBNNb5s7e>LQ72<X_0|MhVzY
zIiYa!3S%Lc@aRd_6VptZHs<VB@tH+ctvwq0G|AYds^ZB+(cHfv6G)TdaIlmM3C?h=
zg;p2jE;+W5x>^Ef4-7AtWR^;w5Ps&MbnJG&y-XB|dZxz(qkYkhSyb!xjzeU$i%=G8
ze)M|X{Vot$%L`IC&mY4eOh8#h(Jrl155pH*$*#jg7vmNoaZ%W9gd-E3(~l2W!l-!a
zn%gSt2lFN+ta@cs9#VX~*vpc&5nwjlAzTGdj{>^T#zP>CZ0I;ZB1Bl4-S5C*Pi?%z
zmJ245e8c6xKp=v{fhI%opigOvYaukFcqi~6U`saz9dwRL$uW{Z-WsOXe++ErP^FV_
z0`d#|GE`ldo6Os`Uj&2urT8&nMqMo0wo0ew#tzBC1qJ)l=;B4Iwd1Ws=OVugD#8M~
zrTN+dCci%B&rJ{7ZCc$1OP56e#b7bEK(lgGXbuphijp!=t@ZzECh2Mt-Uus?D}nxC
zDs*;)@v8g3M#?wQPZCiq$~vJ83~xTOY1=JGXf{m4S3Q9+h>%DY({QxeoiqJFpo(U)
zl6h``k2L~EdgjqFgBKwP2qK|`Z4X}eNj(T#8TA6IRfZW-4aS!%ye6PZiR92t8HGG9
zNesJGR{}&R$S7zvwfA<}K%L1v#h=&w+)7Vbf*gmy%X2{r*k_Y)IFnxKNc)+UR3{dO
zKJ-efCuNPgfOi*>?9U4sidM-yjMQ2{V!d-yKylV~et>p^q@Ce5?U<6Lm&?jDtx3x~
zS|l?ZrE(^VO_9=04w*$=lAUU4)x1yf@~uzz%*=~<Pe|BMpIBzG+1eV$2`7hHNdUpL
z5RY%NswBiAD&jwk(lwK!i$zH%=m*6(f=NQLFxe0iS+J*06tw>O*dTe;(>!2jqgr-W
z#x2VYq7NU=>p#{jBYL6GJRB2g{`0|V7=`ivtOH?0$Lcm}aUS`rhI=@~R#em+@7CVl
zOy&oKvUyRA1Q~HgM_}M~!#5W~#2Xbq%=ma8#9Ne#N5{HI5bL#tI#hO*)~RC_=8o()
zlBI|AIZh-DCH}FgH^uI9#O|^!v+S5;E4Rwdd}E1NyeZGRu~4r$YT%@YmsO|@Qtgmt
zn&PUNzh0pnSArD4TW6sYdy8195gt!V1BmEB+5@El8a7(f@1EEYu~!0bTbUKSeW>R?
z>Vb0DXsCf7Y4pi--e~ZG;H~9s&~%HS%JQeRMfGpY`ANNc>3BBWv>t+(Sk5HS0Nhj7
zBbYspt$U<#$C@vnRkL()X1%#$%_%Jr;L1SmjH$unb2^)_EcU|n<G)Ae)dswtRU`ZR
z(HeN_NFXY%-C{0lq4RZ)RF`&gdpR9a6L5IgKyMM?#1;#f4|H7-@&H`E+~q33n}v%F
z?ravZkbE++U|cjD+?yo6wH;Jt%Xz#S-`}!n$Cy7z;a^65rD^-l<(PU57l^nS%-Baz
z2(q^Z&9QYFzem(0#%oGbK#jgl`kh7@7>|PZhdY%T+65;u3$P5r(^JJiEC*~9J~sK{
z53!D0R<HAcKvh}BcD@Nn1qcG#er*L1h{;~)>LyE9jHWw2vu(PiWgX}joY^+nRP@pv
zo7qZn1j>G)IRSEk;#FU6Wazlsn(OVSr4vJmR+nsEDGdN^@{RXr9VE0s*l=0(OvYIg
zQ;yAA+pB!?6lEw35<gVljX}AX)-&cy<c>dnGW7sn;r`l4o5h-~7Z|L(CfpY}<ksg0
zjbAQ9FeD;T@Rj+a%Wdcyy~2E}P)-xQNKlpYdh$v!6b^u7rE(Q+8^E00m@QMUB#eT4
zpdbNdeuf|csN*!4jhJg(1l&3!Ww0phnfM2gGP6rpK9PW+#4IOEiUw;if&%MC2;5PA
z1}Bm~b^%n1$CyCM+1FGI8?QLLa(og-XSqNUMKhyB!wlf+AZXIW&Ovw%5lnqR5F)?}
zE<5Xh2=EJ4{%Vpnn;?86;0+`3d0DX9b=GL_v<*}h1_VYFphf^fJBLDTCe!pDLzW7@
z!;Od+MVA>-sf3W%^wT_?J2b321Y6=BI7MNDb?r%8_h+LU@Uh-*3U+(j`k-Ye!IHge
z!`01LZ9w{WUN<f=jDgXU6HRP5+JWB}?%ok77BURMly6am(sXld?Q4Zk%86Ete!ilE
zZhw$_X4K(8bER<M&ni*|!gQVwXn<N2CDE8YntpB&UakEU@wy*>HWdL;a+P82&fq~*
z$<M~5U4}w9{a3(Fttpd+pf(a4+hA4LlmtEnzF(mC*Zo5bzcZ&J?a$t3g?689hOh4M
zw`*I|<7M4^KCfWe^urAMsqiCDM)hyH1hzJ%f<ot5tSy`><43k<Uots>u^x#F8Ytm%
zvZh`>kYo<(=vmO0M}#mu10ITqcVyf*LdP5o>rmk*7QnWw2njak{;T<1l&V&`S#28D
zbDgU3$U7D2I&LTq#@*%^rbD7HL;mh>>Vr2K(`~TW*^7+i+b|M0R>Ax)cs|qw?|(L1
z|FP}D$in!)&Y&J^{$F_z!q@zS?(`jz(qe1vn!m|ey`k=D7@(yBMXL>T3#oh}^04IZ
zCmemF?1bxqbSt12Bt6;O@gd>dJIYlw&ehlQN5d*t-@l?SgaSpCy)$n(xSuPqb|`1R
zr_-@44C2zMWAJ(~hwtn2J2irzbGb8jnue#nd+Zq80Z`znEOAtb#^%WY<ulo#Z2y?h
zbX^7?cGG#*@?(Xl5QTfd5XwLpD-Hpbe)?2K%Q*)U_8cG#A<AMRhsQni_clz+(oj-}
z96$vmsoVkIF@0}XnCOm`#hzO?py3noPj`OjZ_g-{VZH1xWVo8uj6dC}lx;s+yV=9_
zjuOIg7Kl$#&BYC&eG<)iVI&ZS8=!?4PYl2~ISPu2o1>OD6)MtEHRM!>9a?dtIO{uy
za^Pwj%mGW*<hp&#OmKiBOE7=7WmXNQ<)1WzqsMtR+PI<#GKoo!Jgll6jH=50EMoXM
zrXW`1;4gh+k!7LaTvkQSHu%F`m;_MGWujq>yUs>O_?b5kF7W%}Xe%jd>RxWiejdKs
zAm7G+r?5_sRsQJ`s%8Ze?>eGuw<Awq3BdO`k{zxn`_z)Qhu2n{mz@D05>EvZ+1Xy`
z>=`Qm_3)q06=>8z^}7*Fv#lvIZnWBaI+K};C#M#jAP%H;N*~5`tawao0DwA>N*p>Y
z1jaFrtB2CUOWn0I*_fTu*(;6Q6tm2efj-f>xGTYK^4$h0E2CQ&$c-g$v?4*##B711
zvN+QfYx{CGW_>G;EXmyf1hsRtA<jlq#Q?tg*`~@qk!7lE;Yit!A$d%(gcXzuHDJTA
z=iw6k#iCDV1}I8^CPd5+fe3%z7*G)9U(!DeDaLU3I;Mm@_cqln<9yfy>-g0(e&&FO
zs;PT^Nbn2MI8#c*tzDW^Sa5Is!x`baH3<wxaeouL$hjU{JJK7GEoFf`nNbXJzqwQf
zR7A%ki@fb3*wEom-4a}IK^t0_gB;<$nDx{8a3=UY^5rZ^#s$Q~mQDMIZJ0O+Owlx&
z?ZXCG+Fw{AJ`rhyTb$V?-wcxqH6QlarOW`)t+~a3o7LN()7Zv#25hEnvWnP%hhAZb
zAdbX(fcghjoxwpyq`jcVy*8D7$}Mh}p!_0{2{t;Z?I>qvdZYS&w=S}ImN%nqNQ?@v
zUNSSBZ7kRb#Im8kGHRQXJW^4FTOCpt1XM*O@6C<5#E6q;5PJuf;GP$>{?_F@0569^
zrUMvk7L)ji4^=|14|036%1r_RH0DhnoLSg}OfV*0Ukwkl5)v}pQ%TLqt&;<?FsWH<
z$%jOlvt%cDwKqTn@&OgWgFKA)6q3-+)5hLRVxr#BJd@bblhA}~@0J)Ve-c}abX|K>
z>jjU+DtygqI21Jz4~C^$pR=%$j%wPkh6X}}+5N47!_1GBJZ`#@l?yqF6l)vDiTKw*
zs1y+BN=VP;Rz&=}j}OACdLlnPXIxyJS6cgm#SoA6E<vV`xvg-A_^)t01W)e}a}&}y
zT7$~>{T4Z01C61vbA2>deBiz674e7;z~@YhPKc#PpwZtQ=XLS>_12Nu{=Plrk}WEy
zzkx(6#NnF~a}SP^Ta;6gyP6|iUZ-UYpM?rRASE@XT5bQ0X6vDaoMCBzl=T`iRW~$C
zpJfiArrA6xu(0<KI0Db}5oLo|VcU1850kXF_d?7iEX@enuy4g|kC9{Rja{2|d@lhG
z=RvWuK~7lRZ3*^~)DJuWzAcepDEidZQoN5`do3d3W5(Hyzti5cNTgB`@gCWP9RQS9
zUDuf=H&tYOZ{)8<GY!!!UB}!Rouxp(;0HQ!owiHVhI|7PJ8>L7f54h4{5iESaV8lp
zfgk_DG0`f|`UUGC$maRay8fTY1=jyHa$yEj*Z!#We!EY|K4pP<Mqbh;Ax-ac4-3Qs
z8Q4Q2Qefw}He@s9h2;7j2j%@bXJ0`}O0VuPs_s7Ojg>i@einu@w;DOu+h^4Ldbf6t
zuC7`_WVV=rt3!{JO4`{lqrt<eN5)sC#PIaD<L5Ly@yngCJ@veA!)&$LXJz@g^x}4o
zzIgw9v-IYp@rGNZqkEC6VcLvIcrefr62+Ye$uvD6fDLmBYRw$_^ULA#;t^VPZP|y0
z{j;&|f;Bd(x%gc6dB?VLvs81j-Lhi41=R(x5n9=F6FeWa=BbTN!Se62W@qQ4_o2ap
zjoW9hh-LgEWL-XdO~xeZy(VDVBHDSIeV@-~Xea(uE0rWNi{c9PYHk#d%kv;$JknrO
zl=Vz)UF!-x-VXgl^98V$_;v4QTGot@e_fc~HCpZE<X?E(4R+uvA<=HCiS~wJ>Il$8
z>6|2}`lgLl=q+5grTR^}YWC&S<tqVzBUdY9e|->a<Td#%2YRe0ZUh_U)?9rIIal!>
z{DtaznpI3kX|xI)tObj?G^WwVki$l}YbQtreB2tiEdcz$9DNh1;#(jBEsFT|8X(0z
z%ly6vG^<t?BJhE8rv-QoQ@?2QT;_SEdl@NEYnrNB`>L!SH$whDhuz@DZPaaDF~WVv
zJi2VPMdzL8!ejeJ3|0bUd!3eT`JeisE-G%CVS?!XRRW{XkP_p|!o4Z1GU0)5cxR~r
z`o_jxgX7S*UOUtcy1+;u!P;P^)V7Pa#93Y};yD?74xvYEL0e?wG1*RLQ8Y#^@zr~t
zla=a#T2Limk5@Wffuxz)c!C<R;`1NG63RQWfUfHb*m=WnBY<x0y|Deq55#AMI$Z!=
zp5x>o4MJp8r!rxWuB*+r6KDQT>d$S<o2!2?jjr8G%Ja=`!(W%RlUZF_k3ajxaLT@h
z!+ci4S`ss}s5oqk<&+r`!Z^CxT}aDscvdOB_DDlb<6)B<t0`axg2cuz&a{bW;enl5
zBApD|tJT`up&-tr!WXxLi1<4OII*qlUslW{0i0h9>^5n(Lpw3d8oqzNfB<F<53yv5
z|HzRImUKPv?SUueY_^ZMn}o!$5d1N75=7c=|Cu#xOyC;oX+tUK1~fQ1$=xWq)a|OV
zVxo}ocXlM0!Glg%cTes_Pz2A}-yn>Hwo`WI&MmuDQ_jLn{bLbqxCqh(>P8>L{kX^A
z%^{nO;}5U1ZL{?v>wi^-aSWz{{-N6e^>n7=IL<dTT!ZZz-jjw8^lS`AVhd);kBtj4
z55)`*DViFX=-Kuh8wh~0RqlXb&<!L>q0b3&L4-Vg!u-&eOUEjb%4}LnfzfLr4pq`0
zK@ZK7B&_KlRH+vqgzxT6HibZxQSaI5r{8J~4#IS|&hPguQelER70g4_FT7&Rkr-l@
z0l7XAmXExJIb}jv%E52{(sc}VjOEH%pkpV{;VR|-1-`gIbES!G?v4}455JY+9weY^
zHN}#r`0JcW%s~fWUCKc*&6K^eou2HL{A>P?3MJys2w%XICQ}j5b8b`Q`6E=q=uUbR
zZL-sPdDKvH+xsI^?WdFH!-0*djBp}g5x$DU^l9W3`bfnLpihFtQ#vnsnYgmv$DcOw
zR><H&v<!uEN0#KQ_$C7PP0ecZvG3l0gQHOd@1DKu%=C!6L!-pkWmRmS`_H|O`UlFQ
zv=59*gMt~-BLGuWg^XF?_(2K^7jhyA8Zb4Clm2ihiQA4<Jx-CP4V1=1L}_f_2GsQX
zwd26Y5EPTbjzI^=D4!{V5cFI_P)I9ujD6h0LozS~>FHt8AC4$*V2P9JSRRCl6avu{
zWDUOv5g4b0fMco&5a<scQ*gtm8<`@d+Q=<80le^H!4HI_B9J~)ToTdpT(Y}&2O|j{
zFG;SE43B3f_487MxYJCejJ(@7ZG^cAcBIlcoRbt`v7EtP#THPxKcDg5=}c0`aXuU6
zel9C9z(h&W_ZXE;%dU+D)s{RSn7tkY?cn#f1Sv8eJ)WdED1kNS+fu_9u$--U7<ZzL
zH>Bvb24-*W6=5XKpa|a_o`ToIbg#_I8*Rd>dAj8t=n^EfUa-fjXyNT8TrWL_?_I9v
zonHbf4SIoV_rt2ex&0Y*C*db{arVwScQ-eM55!Si%28&8P3uUKA(@eUVP1r{l)qi{
zaB^R>gg462kTcVs0j`D&rQzsytz47;Dos(DyE`L?>R)Pvtg3fCdHWQ~a7(jDbwLS-
z<A)RxU`Se#Z%r0|DCIyD{rXfCF(&NE>Q6~ZkHd}500Mp9u1u2?7qy@wk)vFS)$h~M
z;<VIzqWX`6L|;uoKXp$m(E)wJK+PcbF$vU0e=d!fU*X{dexm;U-A#@|{?FfW7Bh4N
zib3eDF4P``Q1IRYGfu}pbWZrDL|9!46wndJ4ODx@pGsPl?RJm?JW&=_WPf-8d^n`U
zilm6S4w{{TCK-W4bcglaFI0ZEwL{h4_FFi+<NiJk9WrKwuvU-T$~<F#;m6m;J;3bP
zwtxLX)057#5ScyNc^mRRV+8h6OsjM+aSPTCy#7EO-XTQX98I0<uY%9UFAjM<-2n@D
z(PWL*O;7T3g@QX)!Ruu~htmr~iAr$(*cK`_ah0A3N>qJ)>d71rITDV^f}$gomU1XQ
z@Lf<k6x1WE8zn!d@!=m>qH~THlO9URnR2De!IdTjor^@uuWU1jhbA%O0C~r9qTr}x
zn9enZA)UEc1wNXYu-Qn%FB0)cOf^$inGSzJZdOE7wqIohqAK+uEt%3R-Tx7Og5Wpv
zzmNF-2k18SVC&8Q+E3s@8~eqM{eewh(!QdB$110Tl1Rf+qPRI-DoH^mDChhIkGL8$
z`>zIu?Y{<7cBcRJio>;r=KriX_^#Hii9^M(dEr|s6YwGx(_StAl}-QwC01loj!+Sq
zboD%Hb9>1bS!~8>7fmSiS*T5I;%45>?gF7j6r2;&mlcQ4Q!+{zQDY1#g&syt6fz_t
z$(&u73YL_)cPB!X?B(S4xj%THXu1!UOzigeD>FB)O*(&EN+Rdy$@@X!^Y-y{;={F0
z;Ni^3jnkL!^LkUcV*@8-On?G`fT3{{Ax1J40?`I$4x~6hqLxl522(a%aE5(aih_BA
z$Sr=HYA#<g{nDm8c(6hxdluCT(PaLMZi#EA+A(|KouewqCuJ%vfYre)#>ieh_45fB
zqlLsQSI<<~Rkoj}Yu{KnNnbSVf;p<$v2%4HR%L`268mxFq8?FLodXeWB%~39j1ePY
zWAFr{EN+!FQk%Zwp&9H4NB~XYu11z|s@gNNdj&@VE5PQ|q$rlvJu#?Jaydhl6$Y$_
zC^>1%;{)t|(ibX!$4*z4HRCEHtE6{&*&JFY=iEQF5fnjcKtnT@*g){3s;3X4-zn4P
zad9NAZb(Rvmch?JZH9SQ20zF}ftpIJi8@*sZF~ayC)n{App&?6=*>(f$tLxta_2*}
zSBn*Vw8VLTjkZy<d2|42P+>Y8vJ#{|fKNjGTFoT{`?F=59Azl=qW-*F4m+s0&tw4!
z*R9V`HcKR%{g`bpW%YI_OjA?E!GPt8C}%`dgfkcMu%8InuEw1kj-zZX(3XqI7b85V
zRcb4c`{hODRxrSNoGiQ2gsYVW@Y$}PoG=gqo*Qi^u6sKNuF1y!dK72V<q+$!S#HQi
zceZJzC|<RGMMi2FzfIbCv#iPdb?f=s`N_^102UbvxJ84KMLTYbNgZh%wot}D7R#2o
z8ROespU`-aB;85F96?dzp}wNY2@RRIsL)gwoE3eI0r#OktS%?~^QIq#jn9Vn-5znL
z%cBTVwcf(eBs~bm_m>U0%jMx`DuT$MYx@-6kE!Ev`Y0-nuLfR0eOJ{36!S#RTSoYo
z<AK!R_0HI0Z&Q==QE2eDcS}E(7e}>Mrwk8~MG;84qKoR+(f-+U&34fiij{FwO*|zT
zP1*M55M<CmaAu5iz-2idsakbw@YgKFCy}F)+C2TdcBlyQ2_*5w`<-j;LKs6x&Z)(*
z$7{Z^uu2Ws^BLw{ooOJ!xf+3dex_U_byk-wbjJW}d-0Re*&kA3k~2ycSt5nya8p+`
zOw$Fh<#>>t;-+qHCVMhIW?SmTE=|9*syZASXxJBv3z|KMsU*p!Ny=Pv#hY8tmj(so
zZ;D^}hfF(4Y(20w1*a^k{YIp7cwV8#eUJ6(xogJCfOj4*o&mV&WMYq-fJ<dl%jy<0
z#fdnFWhm(SAE=*)N2|6BdaySN*qOK7_LapBGO+1a%Vkcgg58k&B^qa$KM<ymeg&WC
zXk0(~)XoRW6wa?kTePuXts<V>CpgyJ>FBzwF{&@Gjxc%JR#k3zk8x<LcT$4eDPb*m
zlpO3~-af_snlYFp6)!*BRzgvnAu?0xS6443*y(AYC~o}PH+yE#vHHtDL|3U>qH8hx
zMq0P6qB`~>`YF=v3lm$sTFdt&Znrk$c6!E?OBzTkYCu#eO5bf@q)FRk-uLKE7)uN=
z`o{ahmsrqUuS_j*cX22TqvB9*3&DO%LB$FYFJwy~{Y6O1*os}pK(<`y7D-1gDK%)0
zaOG--Sa@JtRZ=gJ2x=fN!z{O|v#5hp$#COf8)DFg2QfQ1JLYh_qPLMUkV9f)YURIo
z_Y`l?O_kds{j^&St3(0Dmt?)8Fw?i+)9XOP5xt`nz<BA{6lZial%>IJhUY?#XDc!W
zfP7M!u^f2A+~9`2W$xIN652Acl$&F}p5^C*FAl>&1;il5OdYqAq0a%MGtJ8davoVc
zc**NMhm1b?WF`VM5WJM|pw5{03)k@>B!GtU1JEfJaK<JQpcX{c)L^uW7ic^=L}<ve
z#<q_hwM%)70|k_35<?Xm{36}nvhz5-9c~<}p?V&ZtiyE0gn&z^Tw!`MZc4bcSw+b}
zy$QtFN&tb8SG#WPzy<<?N$mCcL0mvz1k1(J+!i1N`wYq&?Dnh90C2Pe@{~$we;~f_
zaJKOF5D*xjFtBDM=a(q91&hk4rxAk4#>;zZZ72fsnIO(QSx!Xop@{|LVcktu!35^u
z_&V?e2AO_g{MeCTCL8N@DLfseg(5AC6=6XG0<8!?4DNau3*0TR6p?FzP3A4b;ab0)
z2?GB}V0C$Ioik)2viu^0AIvlV=Vka`dLu@L|230nMN{(678~0Cm<Gmc+YPHh5>((~
z$>p+(N#eGYm6dZ<BCQ%p$#gb!NVqrc^TJ|e6>3rskK#MZB@x`GUR@!?R}o`-xjv2l
zygY%LCkRm$NH(vF(nNsBBb*Rz94%y}Q}urMKF%zmgH)}v4MHT6fD)OlKGx;+Y<PQg
zXekF6Q_Wk^BH{FPcDuVcb{H}3ba!+3y=+g-_+7=GqMz2WQ%@Y4pJuIQeqp)-(H0Xz
z7(?MZh(w!^hz3_#C<D(uzRW6hQMInGp}&~6JHmS`JRRTd^lqVSM;14Hs=U0k*;2jo
zHt}~gtiw?wrP&_3US&UsOqVzzEd&#Yxl{b*_Bs$`wMZloRB=C>5PGATn=0sj=0erI
z7NN@;nXT;k3qu^O!UYTkojh1a=qnId;=q7?rOYM>L@#TtmLdqW1B-+P=FSZZoIq)T
z^1u^1C?k?+g<iCuVSBQsi>j82L!#o?wTsT_jzUc0yY5aK4ZCld^=#*M{`8<9S&Ca{
zzxDI?=Dh;2Y{4M`SWSnKgrbEBN5@F$?K=Hfmo0fSh>@>lLosD~SsAA!S;U$wK5=nW
zbg9+?%gsgYTOr1aAbzxC=%3;kF<L_^oR1bs<Q+x+cLWGOD4Q5AzEk#=hY$@x$Y$fd
zXfxo{iAm&axSY?3JpNbUV1f||M=|3K_uJ7Ye$c5DZeg<O?XigIQX!xDn30btHSn5B
zD*kBehM4S+cL%4J7n>-FImH0Y0i;-ZgoSoq&pVztr}r}9U_apP^j|Dzg$h13(Ym=G
z4-!PV;(t-Z)5L<Ni0td656VJ}l8UDXWzT7MN+lgJ>j(@vc})eP%<)fyU^4%2mGJ&(
zXbJ!zD9jJw{CpSzRYEjE_Dl){Dv%<6Ah3Y?Qtt?`tZK@H_~C(3P=CP$10Ddk^GJhh
zR>F7#95V|=dK44nz+?m!e2t~xLK0a*7)YYw2>{fH(9=S8YoPJbU9^#2n_63ozKac?
z&GY6oRpoQ{9k%ayR$)J$q6$qSz$}Cdrp^hKz;~8GpF$>SGSH#S&+&tviILa#CU<T}
z_ByvFuF=C%iU*=LeOFs^e5Ti$70ofab1XwNAJRL}AlzVcsX~;k6$pMv-W0&oSt)Ku
zpjZa!U#RGufAwbSJIv~~X}d>gPZPO7S0HrEmK5Np6bS-R*rSb9V3@wT5WphEMGI4?
zye5x_&JDTtPqs}Dx}&3ZdlMW9fbq0WM$oO>NqV3$Tz-d(CRKd@FqDPWOmz}KdL<F_
za72=knjj6<@f>p1&_L-LnopjK7VN<$fP7D%<rr0)NYe$K*Y0QNn9rPRuds!WSAEUs
zRvRa9V+-7j7X?`XfPg;+FP%<DQG<1UdBR6>3m_<|5LF+(y$6qSU<T{`7mUU|2)aZz
zML;ta4rb#`Cx8V~_@+ExTi?#Y=wU8fQXtb=7r`Mo$>-}#d=Ir}RVOQs%bT(}2}DIz
zZ_Q}IorY{{DO)N!END=4mK0tvk}t)uBgTRI-&D*w%-HvdbDVHOXr?ouwdAk0`=i&o
zozMv``uyqn(Gw(`-=eFq+HS0t{O@zI*z<e<ZbhK$sLO&MA7vJ_oauem7h&zZX7;B5
zj8q`5vc)j|v3%mdQfiLZaO?APvC}Hhst-mrC4edt9(L;_a=(Imkkce#2~C0xPkptJ
z$DFRlN*p=jy&~=D>dGDbC`}m8w@D>)eoH4Lr$JjawRP2FeR**d7F!}%Ngs)={^-E+
z!u<hpg3u0%0LOd0x(`M@DnUs7yTIaY?Qp>{S93qw;=4WC{3>Co!?Yk>w$k}e+rXeQ
zpUv0fU{n&Byu)Ka^UC*6qNAF#!sU;8T-u<wdd@x%g456Rzz>}{aSo)u(0q1Q*s(b*
z{xCRU#ar;M0rF?bc+uPf$}qJ>)#B>=6pINDx*GikPBZQ|!?7-+*y}Xk>QINhr>Cgd
zlLooCTMMB3csL|ea~=A%QEorDb5yw-9ZS2`5u8rV<X|X&efEwjMd<DFS6FnnQO>>o
z0k!3_?W(KFYVS*AFmw}l1TIKmEQ-cpOm-v482cyPAU%yl=BOxPWw2HHAfE{^+D1Z1
zS`9so#(XDz8Vi4E5+SU#cK%gmIRhDoi831lG+UXZu_GiL49{0rZ@%$PoGa66<X9gk
zJO<g_MUTPW$=1~T9p<gIGN-feo`K;%=Q%1FJk}z#JF&;MYN#=<mD|MoP7t16(8xDC
z+~K?(VS&t7EzOOcG0TCqhCqQZEuj-hRP6IMJny>Ns@K~rXf~VPoywIi4LALn-w&OZ
z``bbGD|U^Wjn^OYwxD8CnGLr0o4VrgKAm3|#PXm^1jY#QBnK#=M3v@(J<~`&stH29
zrQw2|*O3<BkGtYf=b6yy0r_p<D*JH;FADWe=e4+$`%N2wc)}>+N*$JW2+D?4$Z9wc
zx)r5SLHanOanJ}ghMi+Dqar|+$g7ktpdd?yii@(PF7!qph(64ZvpeQz^7yk(6$~&@
zN)+S(uRt)ytJ%xx{~~$EthsG{VocO4b#hcNYOQ|x_pC~w)M-hf^{on9dS73NSVw%Y
zF-zubJMNJjlEN;2pAKI8qzYv;qQ9f5AHS?9G=K<_7)*UWES1nvLQ-=VKu{aRfOQB7
z_~>`yPoCp37xzt{SdV=ghgg$RE*n7<g-W24gBIe`eegik%SA3_yXueLCEqqoG!isO
zKDT8_Y_4!f)fy4+Iz~Rye(h&IAIpjS@O8{EV?!`i`j4|Z8{m`Q6P1l`AqOw~{aFF4
z|J8^5qagmj(u4m4a^_(AKc5N&|NmO#6*XD;O#y_SQ}wF_#jF|MI(1d}W)@V`V6;I<
zs-t1K`ta*z$H?;KpO;xrT;nhW1QDU0;jH`VtPFO0H`#d@#HWr&!{u{$3dV7RqI97m
z$^$6L0tQ5+7$N5y#t6`xmtvxw{egvn@~V(Fb3ZZgv7sfqILQNC`0dB5Cxf?+PoJj}
zegh!_d;!$9d=vy$*7&C3ZXTY`PJV}2>~+w;l`osvH}=BDP|UNL2IBGn*x9iBote_=
z1K7FA3^XZvT};F^%3&%jh&C)qNVa^Bn4V`V6Mu^MVz~IZP!^Fy5Kq$;B0<zOC?d%u
z$2K|=_il;xjZ07_iCM0_z3dWPqLPDh0ChE-(3cytRjdq~Iia}K5nxa?-Zv7|LEdSo
zI(bFIUr&6Z#4!-U_vo~os%2+BOj>9fgY{<BqZ~X1NhH<oq<1+X(&14CZSR9SKeSSj
zq5BHB(y5B9Vs>x43PUw>K_1`K6^1o0WRhrf(FC<<ws#rk(4r`|=KBvpY#Wz~kX(`6
zpDBiL^sdA6S5Hp4v`~JlKWiPqCJj=U5>udml><_*88#=TJ{VA-JEGWfSr4f`>BOg9
zJHS~|PgceYB8}|cYS-W4+57{O6zy4lx+<SU!Ci6|6htW5l$f#-`t^c71=?z_%_7da
z#UYn{F~}ViTU(vmTCW?rd(s@as&2H)|IL;)PBijL28z3FC2TqRT_A!DX5QE6#rtXu
zqYcZI5ve?Yw)IlvwI`o#?Da>h?4`p{|HSocVUi>WTo+jTrmf=pp!`i7&C=(f6CAlU
z2Sl-{H@exKr>d@pXd|jeI!koV8=3WgL&>UNWmEP0T&65_ZQj|Xgr{sUK_s`*q@?Lw
zo}nD|4IndVr0t!2>W-<zf@WhZ$5yuN<qxrGg8leaA0()t$~X}_aXBDq@VCHmXY8+~
zybi}T)ut$%Q`;XLlLv_=1}*eF=9(s4Kohu;K~8ipym}Zq34%-DMqkkkSoPT8M(=s(
z>}*?TZ&<F(2X%C72odIT(XmUrI^x4lfSd|TTB#cB^1zrK#%pr&!)zr`KBd_FTb*F_
zTYD=KuJ-@#3|*3Yll*`R%}&Gp=W+c1I!OL6SDp1=e)fNzl>LV)<v{u$RB5YoO~^nH
zz_@U$Rgx4FtE%M8=qzOhFv$QKI|^k|GVjmpl0XoOq_R?WPR7P!00IQd$GZ!=L{V3`
z*e(6p(fh~AL75w7R61Gq1W9K{OmJi(3o+Bu!VK33(jQSezpnR^T_hh(hRnOS)lMxm
zSL+RLmyW%}(LOa<>&xy+?6#jWw}^12M9NwmRWXXB`aH4eJOWfj%Ix|OWWIFQpkNtA
z9>4~X(V5vydk+t{<DbemMtgP+rZ-=8{YI8r*VfT<UAMOqG{-fS6g2cYnF!`fD000l
z^39`d8@(MZ`Po$JOLw=e2f!vtp<uM}YRc6#=!Ee0?erJ?80<g%WNFYD6>nyT7UydR
zS=|FGrYi@*KJ}z>8F!-_(r9iX)ZRqgub(*JL9jeOQk3+MY%+Z5SRk4qhXTI*+&rdY
z1Ot%&xOprDWg^`zrC*aBVA6Xc^rZ)}D=oT*-RLre%6Ayb6H|Y8R!RYKC@K!)`sHBR
z=xmn!ZAC^8ZChGF*W9G08Kost;nGbp7b1krHo5eD9_bQCr+kvq1Q35cy4nFFXOS5O
zZujsyn62E8;agnR7F4u1PqeXrYWI!}G}B!U4$JGw;sJ?4bC}rnlB|FtMfo1D*94Uv
zG|(~g5d>Yn>~5D8?WvtpsE<||QPA*(Ci)n2q&xUG&G!#X*3&Ml<mhSXrlX7)ixd(h
z2Q?z2>8?k4@_k9tXLynKR;vBGUACA2>2lcLj9M7G&`1;Sx$M<GO9j6NPD<A|>){|L
z0?3l|k`Q7EWYM&jHWX=IEs!+80HAoJ3n_V&<diZAcEj9X*O;LUPmOgN*G}Xnwt=#{
zno2;&h9|VQW<%W>NJF73Av5HJk-1iBgBvNw{38ISB*;i#6*w3|rX>o4<^bj!G0RI>
ze3elVsL_aed<K!Lw7qYvx@OgBxuoXsW2_<gA!53z;`|x;&(r|D{855k5W-&2W@~69
zG*jNp4eS$4gsVbU%A0_$I!8&v&FI2u;M76@OV#9lZ|MDv<hLb{a^E*nZEyC5X&wh5
zr!Anu`3_N-juQ})i*b{A?H*RZXb}rlo#ukGDjuJ6Uggs-r_V4SUvN5+;{w9S&N4ww
z0tSeoEPD5eAhNAccMLoOCrnqs;INvuJ%S%K80$HuBV>*UvUnaj1OL6^;n;{3f!Z7~
zoHjjd?&5lEMsFLkI-|%3E!YTwIAOw8d?)Dw+oU&MyIUlLrEIr|e}y(S96HQIAa(x1
z!C%Oj19L(4Um~h{8~p<%KSv2NHBe=UWJngp$Ov4zdInjwB~H_gxiC2uR_T5x8T@+V
z7C&AGfka!Xwik&s0G!U@A<UDpA)r<dSHXz1FJ2sd(N&J`=fnsbxgcV7f?5PxV4x(J
z{CTS~MTWqzSoC-ZNbm@<=$OV?`^xPgw)ttU4&W}&^A(`dXJhkiupu1HTwV`03I)zU
zD+z?i#C3dLe@GBQg7Xzbq8Z4P%uL5%l2d|pG8T-Bl4}sEEZEO*$h(g-D>B?qMQNhC
z2BdgkglnyUz-fGcjTxG6N2pR}d~s!E;zfso;oA-JfD@EC%&#HiS$%#^hKZ;wStmF7
zZ{_?h*43MbsLUS&HgG2=Nx>4L&1MpkYsz?~V3QygxcAu?*FE1V)H(sEywXIb=S-8!
z9m<62+=n#fynmNGnm#yq8kU<NDgZik+##NQ4A^8a+0B!~qg*a=mNpkZ6Csyu!C(PZ
ztaU&fcw$V>$R{=j<*dBfxNJ2OPg~A%o%VedM=$Uk<x*`vHVN)g_@&@}qd9>0|6%N{
zqUu_LwNW5A1PSi$?hrJ%JHg%E-GaNj6WrZh7Y@N4g1fVDxohw9Km2!$J?=d(OXkCz
z-PK=Jebuw8N}qX(X9<E&JHn#BT>@9(S_9>a_)PH4&_!PyYiX(9syeNOMn-u#O1I(o
zY2ud!JGoh<1Ca_o3b;Bc-o(Ud8Z4K}B<B({=MN4ZBw?KILH!N>(;b);_+3}E4%K7>
zW956%tEuFLCqhy-56r3ebwXyD+$C%ULNWr9+t1@-?VO_r%Ecu8sd8uEcFGsPhQUHq
zB#JH-u%iF)pyV{%5FV{+L54vyzO0QjI+d0DU2KOWX{uEa-_Mh8gye@b1XC~MKO4uA
zpdDUXuQV|p@9Z=5i*k<J>D8=NsHaFKh4alQfmDU6VJEog-@3sEi)(qq0E^aW({HJX
zgIm=1*fP@X%@GsBE_vxBeGC)&PL(=&xCo{Cg&DVh`%=>82Imb&1#f3&9@S&02$XB=
zE{~O1l<~%>$YWV3kMUkmn14$}lK#X9slp^y2c^*Xv}l7EbG!c;u(*J%X-f$l%~)Vd
z@J;r*G?yz7BT^EydtFV>vM8NMeP=zcKHeSxI#SIY&FABr_wSfuKIBhs>Q&mT1l>+j
z33>M-yL0ooq+KIoDe>KBT*mZPPjsw*PhU;rA+KQgxKVd6t@L%^s~ZjnoiTcexV7Nj
zM!o+EPL4w|{B&HY!V_wJ#Ix!#5_O6WT_X4=WPts|dZSinY#tFo8ZLy=%Z1O|OtS)R
z-luDP^TdU->X07$QajH(qEo<x-BI0P{!FA=XPb)qtcIr;SFzGLYFX$KiF2e=UZbvw
zu5pA?*VVKuRwirPHZ^6tqrAdZBU-n!GoL@><u}u!R`e?07sVGuA#hs6R060nktipB
zfE&E7^U$7@IT}1}AiZ~h>JiJN3{V%_O|Py$MiCGpEj)8}Ih1pgINcUOXE|tL{l*1@
zoubAyzgx{6O+?U<UD3|A!!~l==Dgt;j-!r=^tk__4RI#GZCc+gHgl(s_6v3X4ixMk
zR`}K<b!p>?+r}5HYP!ry6()3KHC(s%1tTn65He}~KAdTF^C%zZO^HB3PWyPswj}zt
zb9(wfE)-BJIW2zl!IXiE3(w`K@o*~3+s`JfBJiE)6t6_|+~+H_6jxtH#JI<)0OU#i
z)-<GGUk=JP#Z7{!Af03H`JxRyi#YcYjL#LsiG@g#knHTaSma5q(Ds(jhCbqe9Gjx@
zx9S!weZRt4+!6u{guYCPUJJZ>(O{>V@EWF%pzVqPu2E?%;{w&5ehAgXzWt)1(ByLi
zfBL}S{Vxth*h=hpz88Pq9NN!6_;3a3cGtizYaN}QgJ%?Y?i07l2cwWHTg9Hf|GoU&
zO#ZB!<rucx?ckt@C3aodndDU{Uc0)sC;W-+BlN5+NP^Q?wc*asbw=&R!^mr)8W>es
z^BVGj*_Poma%!ad*Sz<`d1S67ZYfm#>h6G5^f^B!3`_U?3QrViR^sh!=BjoulQsB>
z>C=yMac^M8EDOt4Efm=<zq`Wb;yzShC%7KW&0bblb%nH};N%F|UU-J3OP;QB_(-g?
zc95zL%xB7QePo7fIm|Z)dqI1VKPLb{^fz{6Tf`IsHbt2Cv}Te8g^k1YQ4p+d36=c6
zcqI3KG_qyoX8&Iw0-geuyx(g>{f9?VtJBkOK`N&VUJkJxY*uFsJy0|^!e|xP$}1K9
zEIOa<RoMISEuVUJOO=9#NV5Mc`=6OwyrD6bz$67M!A<?Q>jZ-wiz14Yv3KJE6Vf)(
z^r(HmoVS+zZ>TQq=--$Ge4c^JSu@LmTU#5SXHHU3dOpv`xu0#_AFoqkuQCOg`gT_v
zD^uo=UELop-oEbccL#Ty2Um}RpVtYE@dQ|~2ymT_8!lGijCI=v!YC!zp=d_#;^?G0
zRF8Q1Q<)PKg)T}>m+hc*$s(tj2Of7OH9bCl!3EyN{-X}P&e@k)+i)YC^KcJ0Q8c=c
zaPw6n5;0N=HwakCKPFP~-^zKRZqB}CqC}^Y4*j7ar;&kZ=#?8K<7R>aB%gYGS(%I|
z6Z#=Mn_y7^Y%f&GB`sLtjZJmB#Bh1A;9uRCRgO)UJ&H}W8fuR$uMu+3UP05M;A~be
zS$-z7Snb(oYWH8q9xjdAs3y`t!<6iwuX#<^Jkz*rI4v*B&A#X{IS{aq&uI(Z3oNCN
z`N{4&(!~XB2Fqo~IXNlK)li+qxsB1cCLM@^8VND@40TrJ$4jFcV4-k>fxuU+6f2ZW
zFdOYyG0IK0*vA@41wOk-W40?d89NmcLiOcL#?*pwyCfKLe3rj+Pzp6cNL<A{N%G-a
z1HiCpUH~g*S*Lrtd%WZKgXh(D<kZq>^Hv~3u}^@HL!cnJXoI)(%j3EN7MUeV396ap
zKs>o6ajZaF3}<A*&#p%KfFeq`G+d=RH0$46E24Fkx(Tf?Y@=9H&hKk&pI5Dy&&uqt
zRrJW0>W80M?ISfBg05$3b`dSc%qunT?bDSdU;@F+dwx4cVw#s(ZfVWW9d~-9(8znj
zq*Gl-TAX}9n>L-~XX8$vmRuV^tLdh2i6J*oB7}weXuuxbj|aF)QC0J|T7`&y`Kzg^
z?K_u|nyO#S^3eu7lOPr`&XFW5>eZ(pTs-3zA<QVMH%i`z^CjOd73!Yl@gM&TBoPDM
zioMoMhg=G@zoE%sGLhIjWL#FKG049vJFM2KtV&m?ql*fTRvchqJII9g_zV!?8t{!b
zXm&+(1I9S2m8h?5e;~H>VbzMgo^P=}_CjZuOm4<cKr2imW|ovydL-VVVp|ZSx+Fzb
zOkE7A^#HS5R~gs$&7QJpETfr=I*uW;ER~#Pa3a<`VPHLSnl-WOOBTjrz{@-W28DRi
zQHSJ=MRmLjPDRsGQGR`Cs7-g(l$=<JZ93m`P}5g#Ua@U(nqw~)*6CrYmy3}3W&FVD
z%Bl<OYv4%`WdF(b9Hf1Yx>^#k5M;|bi~@6`HKo_Pzm#cVF9%)OhKkkNJVEJZVjW~d
zYol?X3mHmd2+7Ho6)TXN_k6>+o8VW}Y$(s<q;PDUTWI-SwZV%T-@9b&GNc>`OL?gg
z1hloj4mM-eM=NS3q&O$*95flfC5<QrtJAP>*GmQ<M3yjAsCh%q08A&9oziFcbja*-
zh=?K^9N^Gka@F{7cYKD^YaUD=?bFbIN4IyI#K%J{TbI-`>;#+>1ET~qIEKGJM$p~y
z)LGn3%?<xHLqPY;t3BN`OkNG_T5j@|1kCSV_4{U$W+yrzZ&Pk+=esCk<?x^j>A~*O
z>ZFKbeo4Vo{?!B#*eh?F^aEX%w?=R#;#+C{u*1}JrBL$E46&{3m>yO!Ez$MqlCSe7
zX3VnT)+D6--Hoc?d~$^$l|a!$qlJAxeUqkkNAA#ij$e&dswjiSv87l^UdfW>F5+>U
zp|GJgkc%=+HdT@kYN#Pkm<id66sErST^WY5kh{^tmETQq9ix35#4914Xu)L(7#<vi
zYn6k&7|dGAe<u^zqY2TZjRz<<uD<*pb7y*9v)>FAzgvRQd@cN5sw+;ySDz4XFJG31
z!oN=~OSK%r2Ft(qvs=wcvOjQSvT>7<x2;;cSyqE}N3wV!K?HpbsFf)|w!oNK=CKx0
zCDfMrb52MUydw6_w+pek_*?ATBkU}!A^c1%D%4bo*A4d(yJEEn21*X3$T(%<*C$@1
z1#Q|QA~A^u?@U99uh+Xb*%&5Yj2)m|uG9h6ZJ{m+&E_frtK5g5));@BY17ZpSsc*i
zcgf-l<x$P!@5EskKpVY*<4ul6jROrvcbfNA{!YI_y^31x+3YB&ZBmAv8!9}3P)Faz
zHBd6k#vITg3-E7<ewk^h#LAu{)^!ti+eAr;+GKZBHk{_{by&%$w4<R7WA}?pRE*`5
zp7QEKYg|Nz`JPpfV8eM)Jhws8P!i}XOGd<t(KMQl@?bx0J9z{I&jmE+R2z8N2_AG-
zY^9m9SoYHc<zkh<oja_k&q@mj9$<u|EZre&f;!=t3Xrco8ZJwl6q@HL1fSp*SSKv1
zxfXs>?Zl`55+OTrN{-a%@JJ+)MQux4>HG$4^MIWBhAL6E+V@u;!#@5vl+tQjqt0)?
z5jUN!h$atG@OwT(zb*3RfsVd0>}?$b$!_saDlqahssJ1oLWK?|+1Hkot2=%Z-yMjc
z&F10qbOpr<@iRg*PLX<Xdea+nb(its9xZ<JMU@^mL83R#2H=_)^4*7<A&#g#z$p(6
z^ZaWh28M!_^-LH2NHg7EW%7FDe&07-+R!#In+N0zM(t)_W_lmr=i{N?n|o6pq;QEc
z;5gGSCl`*Gd18zHvqa_f91r(b|73Jw1%d2ed|H)`@zGuL_p1lg_|e}GRD4PnNsm{0
z<6_C$W3?D$DqPY&B$Od8zH)v24BO=yV9$OeFXgSionrXo4(l0hpMq!w^>)~ePk|P?
z>SHX6ANCVt+W@5)EMxW(y3l$t)w9LUXQmR#tH!9FCH6k?0$#{2_AkY@E5DXuV4m*_
z3eScy7>Pl;Z3cl({x<<dU{UEy4{4cuD6FvZs3ZX++a}94x)nh>o7~?L3HeIj`TaA-
z(FLYw6B+ms_prQM+HKd?kjf5rb@YrdP;b7%oXo>4iNxWGYIhaivO2aCjJ}v$6PFNC
zOocDbM)6yKZq1W{#*5<@-{%&b6Rmh;)61j2Ph2eJh<2TF=mvvQ0wTd-{!o?E%Vc8)
z=0NnTe3$LPXh^f#1Nx@-6_Ta5&i<C?MF)-@Q2Z=-d4p-$TYNB0{w;Cj=FPWyCZdGb
z-+4Z#r}RYrXpiE^x9jaAH!4VJ$Yr^ce{s-68LnXU&GjztYI{Tu(KK`71F-E!BeXAD
zDK5?YgHZm>Cl^5!3DJN*y-o)U&pAwMz4i;D%(C#fPBW;vQ};eVX&QnT-T@EtTa|Zu
zc}{t#bWkwc<T6_HO!i7XeZbgI!iX-CO49A>7P?}6?rB;EKOF>;kjR!2wd35HiDH$X
zAOaf{=!Cg^pli<z4oqteA4PSuaaImY%P(usM5iJZB0K+#H~b&-0y86_JArBf2IB)R
zG#hdf284xckr{EufVRI)mA{YCg3n%~1Q`Q=(M_f4^J&^L`n?iRSz~jy>LMnNS5mAA
z2^bQxri^V43vN=G^daf6$mMJY4*uHz0shNvMAx!Un}(8X7GXX)6i4yk8-W=u4M2ZM
zk_BigS7)WJmvJBF*<=Pro^ldp8E1r^*As*P(^OHMP>d9_3WAeu!^$UXE&cjn^2s`N
zo|oh(M4Ws`(*QO1nx!cL?spN5v?YDAS!fhQEf_Vnd_D0rwb2tN;N!Bls;-w+;oyog
z2sMZ?T*CN6aCT@(;eI(O(Hjifxc@!x+4S3eTRG*FIa_$?$Kl+jI^d_EUm(^SUa47i
z!GE&t|47YXW8wN=r)Df_Yddd%>KiQA{@t?@kNWGEVM&-6z1d-zxv-?24P%wouyUT*
zFq%^R@j{3KCgt!$*@^GuK4Rfp9}oBA71+RP7^1m<_tVGAZq{<PDKw6hshe*WO9mVx
z>LF?8mU*vT*=$;`?&o{fGGx{)Gbwq2Rv+nR84;*h0n>8rmY!XCA0F-O!B5Gq=Yxxx
z?a#eMg>7Bho_4>xpx2j)nJa_+ti0`>#xp&GnX%xnfy`#t;O0ff5f-9tV>NL?=XpRX
zjmR@O=Z@amtbzC0t%GV`wk%yx1r)j&4%9uh_efD#a*n99_sywv(;&&nWb&J$7f!`T
zn+-voZ{0Kx2Q}41_<|!{zSM%9Gp2jF?cVqbVJ%&PPwjT~M>fBtg2-pbzA6r|hBR?*
zBE}Z|#tE7oFNd$SNumj<%M5=TcaDR1F+DQJ#^GN=#*`dv#6xUw5Uztax0?vG&wSBn
zYt_}|u+?jhqBX*2G2Tsahua#-MnU3D;LBaec5{!lGbfc{=Mc%3b-VLS{~fPbltqFv
z)q-w#(Oq<kg9iZnY;%vVT%d*2P&X8<<<CToPNy6~O30VYjiL6<+5ofBo~WBca*LA7
zbi^LRZtYTe9b7)l0Uu1Z!jg)Qht{Cdn$h0U@J_}VdJ=tW&A|$|y&W`^a#w}0Hz|0r
ziYYJ(rzTJc{5u@J(p4Tf3;AX1K+X}v?_93Et()BJj?mbvogzzDQ`L5FIKXty6;*|A
z4|-puoXj(+h2<-uKud#aaANqNn|RQzRUWy0yb(FIO@@Izu6~WW0}9k(05dSi&GTUQ
zv&L_dD0kzvw4u9Fq59JyO`T}`H0X6g20QMH6`bSV`1{&|p~hvYw@5$nuhQh-8ebB`
zVc^cgs$(|fNZ1e)q~eBPg<Y_3ZE4d3L&o&xo@j?BF`cMbb@~y4&Ez&#Udex;9OSn>
zYW!UY#NVG9%r9N-1`@tr6;0brUw`RC2>*Tgqa%!!<eMx#!u|_irxv>j^to85&w)iQ
zTY+OHs&xF%TLm`9GH4f*UHyT&qHgI-oC~}v@ii)g{k6#=@R)t#U!K-4(tx7nXsJoO
zdE&$|Omw3_`5gAdOd*|&&xC$CVR$iI;WklP)d_+@tlAnT0loVOIEZdD(!EIG)8f@j
zkDD`l?)5<lZE+Tqy;8elmgB?t?#`z)x+$uAL<#&*hp~JE6_Rx-6G&iQL$u{6&_9=J
z6vIA0%5lGaE1%4s`zV-<m9<N;3aOd)D4(-m=g=3Z@zkZd75Zipbp?1=+MYHqjJJQe
z-8#(ueLb92@lam=lzp`}sM7oKyDE~;gigUypR-eC9^L!%PVQZ!O*t!C!yFsV?1hz=
zJ|hoTp^+x;fTZdoz1kjWxITk?tf2!a=G9DR1dezH=mbxlc?26pD!wL}?z2VV5NLeg
zcaU$NSifcYW1Xooix|>J90s!@nzHUfl<qpiWzUqcrZWSEEiej~4UbcjlkNAb;`avD
zkF8f<edGN*@~gYBH;pD{>)*+Rvcymi*a_>Ud|yv39oWk<y7N-#yz_O0Xk>R&ybmy*
zr)ADS%`|l6pC$Fa?H^1wZDE0GTEmgE{}N&7ziT67ofF+UEG!?hhzYO+v($#DkGZBc
zlok0}zbG#aF-Qz68pu$9#2Y9-H%-pgpDWAx$1^SiS$$!2mG<Iw_jiSY!NOSCAQ&n#
zi|J_Qd>z|DNZ5<WYRO;j;cyh=i-3qad;pxL%MKU4M54_PR<ONwacih0qo9~?FI54F
z!G4tFBu=!bQja*0cTGR2r$->?*Y;uIGg<hvi(uFw59!1L5^tm2EMyotJ*C>1pgex>
z4jJouq4#9`QS9Uv^yUwx6DSiXq=Qf;zKO$Hn_4i}csC?w+szd^G-Ta^5LhP*2p{MR
z87bs5xXY=b!t=orre*ott13!&uI^EY9T^v)YL|<d1qCV8mP*4jtpZ*Yx<`X)^@@0_
zajEVm&Jm(WI6<jjXvv&xaUXrG?A_lHxL?owbCg+fQMeHc76o2Yl)^Uw45zW3J~h8q
z@u5_=4AQ2UqudKSO>r*7PS5WiIQaXmn>*RvwOfG<4h7P5Y46AK!kbjO1$<rUAJgEv
z$1!r!u$r|+9g~!#(eL1TvQa4%Rbj*Rx3Z*(XP?LmI85LAr**v^5%1DY`o_`e-Gc8V
zA&%hAJPMq@I+RglkATO;i{(>E!W3Z>%OWTrxf{QxdLiEDntz*FPp~bASl2g|P!_o-
zpIJ$O6sr05hr{r<837!Uv9e-3YGC+%{qfX=@=#D<wvZvt)bX?W{K9(YpuXAnKYnUp
zxAEP(YB$MBbl>rEMuIx3XkQMSHrN&0t}b=ZF~DmFW)8x}KM(7wR?y=@&U&S(d14=3
z)l{orm48OT$)ZY%qy0XZcrBFfhaO9QOkt<F7X8CTr^+rFUQE%H$w<=xA6IgjHpXHU
z4}~_Htwg7dSbZ)<SqOI2^q1f7aZv;utl77NJeWSn^3>*3*LR)pZiZ|>yG~n*c>K}v
za%7<a+HtOP#5gGN5IL7&$|OTdHMB|KRy2B^-*Mm=a(ZZkni~EEmH*AoOqB)Vl~Yzz
z^*DPaZ*4)>a7RgSy{Y-NUBC4$POxt*Z6*Rvxlc`WcLXC9c7Y2%$z!gCChjq8w~@?B
zC7)MD_b5O)d``S48fF|@Iey=mkTB{`^Dh7NQC3Awi`vcatI4`4W54U1jGf6Q?cCTR
z#-~3S7V;8A?p8$B6$60%%IoDFOPocA@C-^l=dGoc$od!_ib-(|LMFLsPEl@hdy!p3
zGBTInCKh24a%w_jYfnk^_3bn-l-aNx=QcOd;EN6SZsuGdok{+lyRS<9hT6IE_PWs2
z4Y4s%UnM`KiN6c>v!8F5d%rpT*lTTmDvUBhwpl3qt`?1W6oq(~a^tOB-!sb&1}xf~
ze<}_x5T0d8-P05eH}HfEPMQiY8^0fSZ5}>0SJp}Ui{048BKWoCIN4>%J2o{%YrI2!
zhGZ%rF=gOE;>EBLCh@*1jr{`4!K<clETdOUbwePT*&LoIQ%6xpI#JbQV_SJX7z~$W
zGB=8rih+93a#?3#@~G!AIttyIm=?*dXE8OR5dq_iHR3P3f|L3=_bl`>9q0^3+W7gj
zydOYcvhVj6#g6JK-_<nZx(Roq>Zd6v@J6r<*E>ja=bkc2;AX!teBs99B$DgmSL59_
zT=H?AysECWXi4P4P*qcm7o&S`lg+070viBu79Q84#qKxaPJW*4^2xv~j$a-#Db-yF
z;(sV*tCLyJQ`~7Qb=>a6Z>0nWFehqsNkX8+$Jc#eND-qW$%s`oV43fF5}EHpr8yaW
z)T7#F)a{7A<n~BdB|K-lkEF=wzXD^gaf%v4K3R*B{{Hkm+of5U;S^M3d*}L^3%_26
z!M<lrIH64VI!UK_<5&jiGYJ6ee9E?qK>_Ph^@(^YdAIeZ@0k|(WmEt-hmjijn}>~L
z*0-T1Rg>D+3ppgd5BV-aO{6anSO7J^zazAHe4vQbk9&%rFAw0MT;IK%^hQ|Km}TOw
zj=ck0X}L_^M1Fn9kg}T7WXabzLkM=IxT`JL7j!W2cUv(0YwBVmxaZuWPEITu?~u9P
zs*IRrLIV#c7kOk#qYWA5G47L!i`JM5hat@FoLmBV1H=z-QRBB=_d-~_PI*0;*FBk*
zA3Ids_ZFP)$J|8<-jV5f-Ry3c#>Rw<-b&@1+~+#z4+I`NZ#RjzIYa83p=mcF5XA@l
z@b0^d7xAhef!4|Wy^%qNdM%`FEm}E6-j@3v{LP-%XO`T%5mhEEf#A7iBL~!=W_SZb
zpC>zRu(PwT3X~oRwoSZ!P2wK+F;ww-wgkM5Q>inQRkw1-ccEG??<88tT@x6&)px^T
z_3imvC+Z(xmuwio!pEL0NSPI#IWpq6-36_Vjq&~yYa@ItCcJ@x;ZUISO5<nTvq*+|
z>fXV6<<d!CK(QRFeAhMs(w<%(VkNvc)Ue&$-q3MF#Qh`m5cAr$lPCRIQr;|jx3`ox
zMr&uFeJS6*n)5M}JghmJ+W@2B)0EVNyaO^Dvwar%z?MZKT-R4<sGd_hfXrEC+}4X~
zn>EwmS19Z8kCKVMIS-9jm99yF8roN&PQV-fNM%o%llOoaTjZAyDgO{4pz`$C1^e<m
z_zWxNqV5o8J8vRS$GrMeA?!NX<yF6%J;3z8f)}3udR4&A@xMNao}ww^yz%d=f=LF|
z;lbf7la>bq`*Vh2?T6GNnenZ)Kr`-ddn7zKiC*L>pSjQ-Jdo7V((>d&`JJ<IwVgj&
zIyCDSne4apyPtQ+%!ex^5y&P_ty&Y!ic_RA<|->MPOV90$LAZHIq9FDCu)0h{XecR
zci&F0%&mWG-X4$%_;)^CFW=?WPJ86~w6_hl-_1OAJzY->hfPe}G37DM`1QW5BNL?{
zuF~YN+6v+B|3tT&#S@2^F4m4aTPBsc8~9rj)nIp3aR`k=n1g-bX3^4e;&^Hh*TVRE
zyRy^9KgU;H&FPD<of|naS1|hHTg$E<n{07A{ZDlc2Y&l_Nq_C#iNk2Bmck#MS-#a_
zzWM|Yc14jzRzkQVJ}QJSmkbS<laO6JHyF{y1%FxY-lbOP7jYi(`joY*miVSzLnC);
z@ueIxB5_RDZS<4HBNvW8m2fYM#*#_*dMINOX(r6I{CHK2)fe-JY9u(^F8q!wQ|Xlw
zJYPBQA+KZR9QAIJZ~D?JP|Dquxj7~EGpM!-8&-bW51QsQmOt|LA_{8yEL^l~?49{8
zZi{b!UumKW73IHMa6H|wVZsr{q&KCEic{*n9HfPD-!)?8q@4@YI-wS8sP6dC&@Hs$
z7n=qKBI-=22Fa6U#?8l$;knSJKSHaa)S(zU^kbNpPx82*H7uZ8VbH$<$O}`%il7hm
zBsMJ8*b=Bl)-EL~NQ-E1RdIM2KLG67b5+t}B>*vC<?lTq<zWebo>(YYNa+YSm#Ev*
zCIe66%SuOjLwic2xnFCEjr*r9d>L&%-#C~%?Blk1eKtJ}hA;>fX>3IQ3IM?E2}NU!
z1BKS~t|o9PHGVK&3D~ajbN{rAQ~P|93qL0cLeP7pfK})%wGGc4lz$Yxy@s+tbe)IN
zH*^AIE8>eG?Js^`Y%jR3`@4iJL&yML>cn61X82=!I!M8jmIAkN-=vRndcqlb#qYpg
zZ9ZFBB|a?mt5$TNb@-PaW#JMrMkNpSqQkN0;f%>dI{awn!jNT8-g|EQ8}NiKo}aFZ
zE2xy7U`&2(_3Zw-q&fD{5n)a;NpVe<-(e(Vp3^oHqa1Q^MV@ql)2$7p)BVg3#9tMt
zET$n!nBC1EMNZ;5XA(bd6m8G_KEuzj+nS9ipO1AhmQs<CI^-+66T`P=OW`!;frBkJ
z76!GI{7<`4EAV477Qa?VQ?$@j8TYxrqE<5R9nP3<2=T)If^`!OR{9h#{Mg|sAD~u6
zlkISpAxoOM;Ftd;eO(;FMs(x)y;NK1=-!0%VsL~?))%)x@^s?DV(Cu}>kQZ}cln@*
zIMOSP^*;KH_jOa}H}kIR>+Jo&06Hq&H@!gZ%L`P*37$vKt&6te>UFE5Q660z;t7FW
z8y7@u?fr<LV_3{Oi&!XD6-!E1zCPKte4F&55b#el42-1jLB%a6mlJ#CqUHRIPm(_!
z$z^PnDW<T;G|I!v{7}cZ*PC+NWjhR&g~dqWj2IJbErUi&?OV}@RPBV1qhTOlKo#ig
zY&j_sVL}2h!8}?A%~(Ax9E*ujsPnkV<54_=$0?tWz+BU@wqgz?_YMy|9(AO~`0vFw
zT4dCM13jE~I59sKCuS*45`;ZbXI-xo@PuX{SmC5k5_NfUr&jx-1Wef-<Y-{;H>hNV
z%EF;?`YoNr;cnsHLKk%<noZ30A<1%0%LD78;F$7JEnq4Rmrp;1v~oqh8mr1?1XqN`
zRw#8|Y5Ck~{08G&gFY2O(DXes=8@S-!{c?iZ$I^&ge<PR8wO(Iwx<h|1r;dXLQL?F
z@ZW040cpAvCEIYk>{=I_>~7Cn*?&x&U}yf4$3r1pYf5_K4nr+wvHWf}^`cJ^3O%hT
zjH{4eie|x$WI=qu1HOkH^{wfOh)mcj<i2=*TX!`Y{OjI&%UW`e-aO<ra}*dd(~9&s
zOUr?<avnHeRr$r<s2@^L!LN^T6vSyqL5u9vy82Vy5G-(GdO5Fgd_HlZ1e|MCv?W$W
zi=`G%oyUtzTKH736hjDLAT1&c<S;D^RVb@~l@PH?(3R*=r`z)KgJFF`D$auEr#_WI
z9?dKVogU4s2m@?u>!^_wT}9_0?uahNFY|4UCTJfZ1vVeDA&sS@-TYFR8+>)&C^V8G
z>^-gv2c!I{PAG0tDLvh8w_Cd^fu`wP>(5zCtRS%T6*-4A_m12Kq%@5!Mdus(L(6b#
zp%MzI!<gc6He@dYB>QNEsE2Wt!YNvfhxv)4-FC2PHr=}R0-j{i%S<w@0~V$DxWStv
zcc=}qqa8427hG=%heOM88e9XvFk;o-g<u)x0A{u*f35SHA8@h?@)Ii)-{YasZ+*lw
zn0T^6v#Y2B5ar^DveTyfq?D8@T`Xpyh?t>FSaHxk?;{t9J6MJGEoN}9q6zu>#_L}o
zSO8QdFFS`zFvPMOUWlPsYwbu1><Z86!#dT5<iHHaW`-bE4CwVaF|bpBWEW@P5k>H5
zT#ZOTMO{&G_|x?+Q}{T6o=6I|G@xCH?Fsi!8@Unk%6xx5MS*j>u4#pNU=)3d5ha(K
zdtKuRapW1&2=*UiXBq&&Tg7z&X>OKQ7Bb&fz_t3)XTd>uA`(&_l1p;zp%gReMV+-T
z2_3*;Kv!HxuvSYL;Mm@+UY9+EVuXob<LS@CZNQUOD(2&>{=y5>_0jC|)l;tg{6hoi
zE0xp|#n=*rV3)U&9<cHE5`|r+*+EU`Flg-#ryUyr*nf~{b{(U8GkOfuK4VKD`dB|h
z%${>E;|sauE_!IjITe&Hvm@bfiCwUH&AaLMM-I1tiaQn)jrZExqz)jvBhb%Le|`7&
zOr1$AV=a-@>Gh+c@q-TzkV-Zquq$HwFqfK6SSt|4V^Ww5V<3Fg5Du9WT}oGi+gC<#
zn!1X?-o%E}xkI8|pAVU>*ZN>VeZ*%(h*9rLtEohd{|@E?CZls#04uf8)XQ{w{l4_l
zBSGz;c~ojQt>!GU%;VxDA|q2tz{odCGWOPK|5y+Q{?UHJzvUoUF8Z5jX5>s7)+Ksx
za77VW?x2rqVW_-!T24_{_)VCg3`*_qRoT70LKy<^Ms77(;ylguc1s{tm!svIAzF=m
zHyD12_nhwEr(c?@z)*tf{G&D;q<Oa?9_}SRXHp0eZ(4DByv5TsN*iq8y}gn08GXu+
zQdi>N=b?Ta#P~E;J_bg|%mi5~FU5f6q}aH9YQoCc8aG+hFuwjeWOe%wgv<lRiT`AI
z|FK2Q!v4Q6!P~4Zm9)T#-ZN8sJ*)dW<R?=Y{;$AbbehpWTz`qzGVP$42-f-eug~sG
z_}ecC%|lg%pA8rk?B^dbXDzU1kJBoTvMNh!i^81}srVu|xe_EvDbR$ABRR$HNidUZ
z1Z@4hUm|CHgaN9;9wooVgn+xt=2x6dwd&S27L#ZVT9v_otD0>A|GV2q^Ga1f)-u!E
zWoPSEz%6k7>@oHVLrjiwK?4TdIyaGCp$psiwcC=PNEf}OwG}Xft5Y3oYL8sJd3u*y
zZSQLB+RZT(0qJO+ikizdbFap2ed*##J%uJ%nbk5^Ohlx$-b5f>@lM2^5rH29%qN-H
zt>(GkM7!dZPb$9(CRR2xz@_37o0nFAlb4K;UKL750hK$EVNLO*<y6jDmAX!frWB15
zJ|K=1ppWJ;v~2T=^+^xMrXV$*oGlIpFN9l_phY3&%gd13WDc*&!kF3ZFCI-%Xupjf
zWxJ86I4&r*5yjVSVx`h44k*%3E|MH?h#-Cl&;Nn6A_M-3E`SLqrV1b{M@L1^Rs>&!
zi-!Y?x2)PkuIVlB)LZV>KX1VWYY2wgAA_VobTgBEg<33$Wd1{*L*%(zoYldLM9r?}
z*Sp-keNz`wEHV$Sd;#XCQJV?DWpkuH4jnj(Zh=9#yi-^0c+g~2k06~2K-QuLGtd!J
zc1rC0Q%N{z2+hR<^*+oGjUQ_w$RrKpZyL9o>%B37bBPsXLq0%%vuXpZXCtA|ko}fY
zZEM<#lbt$c$0FQD&Li__9KI3f{H*)1^;qlir`Mmc4eC0rfe0nhTK?{-l;Vj>v`@J1
zis%g<oVLM#@8`{DJMf!0f}NTQ@w-OrVVwDQo;j4N(QDM=0ynIfJ?pAJ=0Q9OD1bn6
z#2XUAiX%k?R=4RqIQ*1f9}7nTF}20CJ(lAQJgZ(urC|*xVJL}?t@iSM^~giU&}RcX
zm^AV+g6wnv56RXJ*2@hp3sfH#b&(gl*lv4dp~c1$8Ynp&m{Ll;<=<Eu#St@Ltgw5J
zArWGRd<*a^?idAs27eFKW_KbHlG89r4GEpW$l1Pui5=?g7vCz%QB$xV=sD^mr>Z6v
zz7Lu-p{#NDC;lNG(@3LO>w27w&a-nh@Aw<b&nSNu)d}Av;a=QR(Pz@da@aX3RUa;z
zny7l9_bB{`tIeYB$y|+DEnvvy#zwuMx`OwzUF<EQA|bAB>MR*a9%jM}bKo(!`|)*R
z7!T9Y(k0;G0ah$VyXCuw8gfZahxZENX#5rgPHW4`j&4@x*23^hZWg);Iz`twHg^re
zyChAmt^IhdMpkB23x>|@l|sqctSLuF*%$otm`g9#sH3#3%HEn1CsT3kHY;si-zU*R
zh@~7EsJ<(9MK#nU2<dKTN1X277h@kA>crCW83DNDG<*-1fso81)Jcx^D1k{rd{eq#
z?0?`BeozQm2;Nb96(SdJg`T5qRD>200&vMbzTSP3_)K1E44~~~Su(MX^)GY;0|yl7
ze9bG352%XYjifv6aW^2<*Q~(6Q<*gy(=$(@=M~(8-A(}DeY$^^GCPppDe}0nA`{0c
z+yY_{2GRFgtu`6~Q@*V`;zy<NS#o*2*29)>9gldhUE12%d@C2)+Wxb=kB6wj!+CL%
zTI>NPe>NJ$43Vhq{%rge-_Br-MqnP@^7)FJiDK!=Dpd~m_7^Hu%0_F1_!Kjf)?DWo
z2LM>q`Y=j(wui%NY@DM+3U135a0DUHPdfJe$9>>*Slb$UX!cvGN@B4y;E4LW-=%i<
z#SH4IK6wR<%OMRUQu5{1G7jl+gS+le+zt)We6o!Chtbo&1nVL9jtTVN+lT}^77tLv
zuqkZ1%-tME*s0hRN=@fWZ9Q^UKF%UpFlUk`4~P<{x7A^YmH)%gvoQZh|0Fi{{}avs
zFA}<3l_u<hm{Gd!Fa!p`&A+xpdCQ=MmlYI2=~q<1pi2g#6M`4K`F%xVP|BLGF<ato
zG!}w^v@q?GG{!_hku>Nhh}|b_$7dO5Q4BSS6p#xu4rIu3WX4(te&2YyqfRCQr6Mu6
z2GrOm^de%-k><9?7{!F(>5}<lJUimsan%DU^v%ItOP!Jp4RqRZ7IcW;<H5OtjDf{J
zRKJAZLlayxgU7msIC3&Wfyv2hiN)B@go0rUUH!!1K|qL2O}Y7CdsvT=Zb@vQ(h|{f
zzk~y)#}*kvOXsQSAsIzZD~ThW2fv3C&~B0GOHKv5)3W4a>a<B;H+5aMI?r?;!P8#&
zqPb{gA#_Svo9RFQ;{k1uv3R^b3Tlz_&@wX(=AC`E_#%;2F0{E^e7u4tao*+g&*ex7
zuUXvGDcL%x*7fbJ{N)U|T4m}zslvig+{ZIsnSb`;f2<QT&;Q-<SJl(Ol$c3F!N|(g
z*ae<R#ns5=-z!qKhUTX5OzM^<E*8W*%*^milBSmC7B0lx|NIuVba7TRbrP|+b+ETH
zwF6y+XA-fuv3F8&Ff<0;Pt?@S(%4j4LKt*MIYVb_VpbOBf65Lh!!!M#)FL)^ZkGRn
z#{Yk&-}xq-kLu$4r$1KH<<djt_EJmBgUq6RWsBCv8q;o#Mm54~Px&uNA=q#@??4h_
z)d#uMIw%wr2nfS%;a_!-1DV(`kR;OIgde^>`4eXM9=Ycnm?dV6P3CxZ6Fv3^ZFnC}
z=kU6FPPZRzct5j}1fvMU{(pSLuT8uF&!|4rnR?$6F@*Uh6UZbeI}^%PN&&;e3>tL?
zz|%!bOUvQm;i~W3rVr<<UQ>Cz?u-@!%T1Q^HNA<1{9YRF)zg_AGBZ?ERAEz#P9qt=
zHhag%$1gjt1Xh1sw(Z5DQOVb>#{ZI=NM!(}4u3w42x^cF4a#P)9#otj9cJ0vZTGw%
zPGvFac9?9=_}%OdKRrEx>D1|ezASsGU9C1!akn&G^}Jp_tlOt4kbiz$4<ZM2{hZ>8
z*8jLo^AdnTCPaJ@308MY`t^Mp^7ezGQj1|_f~j}%B*H)`P;7II2=qwMgtTsguC}^6
z^}4-2-?u+w#H^`4KOTD*sc>10M2Tw`X}O}~8_HV$=Ppkl!2KkJ3ezkOYfN+MKruK>
zSGI(N1k9IFF)=ZaczX^rtxd-2EiOctA2z(Nx{Hg81KNLXZZHcE6>M$FL7Z}<{e4}I
zCE#6X*mN8r(A3n-E`r6R_-!_OeSO_CJy)ub2a=r2$y~8i5+ySKef*xK5$Mm)=g}}E
zg1IQr;7E?9b2{&)c^5}Ac-`&#YSg&6qEYJ0+};13<;MqbN!z}B*m|K1m5L@%OsY(?
zsjF)vNBv^0Sw>${x&8IB69ECCzD&KrY%0@KkdldM=5$@-Y^i#2t>^RO6;#t7yUrRT
zYAI9f-8&oZ?cMtG<6T|fxB0;D5RG11(l(cj9104`tu$h_(cD<cD9v~z_N1|bh)%4#
z&Coz8maRFSYuiV~cMFEdCw>sNzW_3-L)yx!;->j49_LWO5F=>4YP1$xH{G&&wIYp_
z&a@UfMsScDoerem6Vz+<rV@hxSsUHrA3@MB&F-77SeO(5pU?M4t__DEt;K5XR;1Xk
zU!t6{%R`b=Qp&yar9j)2o=B8KPM9hB3Enm2{QSCc^b8C+g8qIpx%}IHuNTYu{_pRP
z+n?u+Gdi0G3VQ3k{{B4fT`#AVo1N}?%!UIXwV3j6t#Ida+7Bo6Je<r>DirSqNtiVW
z1U?CIaS({l20{?j`~beM59RHqCMIPpM>BbG5fKR#_G?Wi`9Ff(uC{D%ot}$W!`hvX
z#(BqOKHjhVCs;=(OpAEA%I_}C>!av}q@|@*RaHynbM!hBK?p#?<CIOE%1QHXGxKqb
zJzlEL^1E16*Jl_}*!JZBkW*19<a`Gf#tB$0lqLAk*n%Fk+4+d2?~{;|b8$)mQaq~*
z2AwuVrp7j-I1<VzrL657SP3R&B_*X4wnvBI7o$#P9A<-GdW#cjOz!j!PuGU;Qd3vo
zbMx|QWiB^6m*0HZ*w|bRH;Ml&YE@r`?kp5wU|@iNa5i5Whfk#E_k7S|yYBaPGZGOI
zv1m(lwxe#TW2Q-dH{hf(AEJ=UcapL>q(6eQeWCBS(Y{|-A_)9=f4xDjPZb0HzFjrX
zQ!_4+%QEBhJAFGh7d)FSlrS|dcBJw>uOG?YQV?;&SAR_V!}fWm|LOHSJ|(mTQWopl
z?j=XcTM-4M*+VJ<P1?OFzxla27b=~cqg?Nr{QSO~=9`<F-uDZ8IXh)8^J)CM#vpgW
zs8~<~TVSsDrDL<r<EH!e?r_wU`R-We%`A^_Y%w4$tE>4)KSs1gEn@~G%~LCZWgX`!
zZToHtYrsu&gRQ#fJa#Eog7s%-<xfHV8TBfr7P{uE<L<FVD6c(m273BQ6FU&dIT+(^
zN(+onm#QmetW4XrDnXUs@EN|_Z*_HE^{k&Dn7-Yg%x~%#6Z)kzxMNTFgCKio&*gD#
zp!)<G6VQR{=E@Nu@tWrE&iMRXGSOtU2>*#9qwO*jj(P}#`TV<-n^y4MuIolTDY5ry
zn5pOeLEE|4)ZAR~>2HwQR@>8AZLgd85ytzL>#}aF^V414r_Cen^QuO#pRc=}^5gxy
zLKfIuXP%hY)@9!3TPg3v0|3uH{Y}1hHS^|#r@-^>@TIx;b@M7a^=3((>|~|BkoOKX
z!SCbwNUy$3zlWD=CS^;jzGtOoLBZhNNgeALi(Vstg73QT3#GI7J8(H>g3a&EL!Bay
zfVU)SYzC8m3kwsZKeca?wYLW#<HA-eayPDyGlW@v%e@DDr$cdL)SPME@~Wg&DU~O<
z<dCkio*f;P+~Z_={oDPxRoU5mI?D9eY`u(Y{b_>y&6FoK8H(9cf+q`GVWSnKw*O$E
zR=+p&E8b}d6K(IyVl5KEch!UGvr}#&JUnZkv88HlV5Q4HZUjruW2d|{*9o%4r>W6=
zCbO2m)+cpBozT~!s(ALIR?p`~?D^M#cc)aG;D?8lR7C4ibnOhP|MP_Y<SV+zO(je2
zcZyrB1VHZxuoh$#$9fFZ@0hYKBQfaC>Aj{>R%{oSv5njlfy?ek)9rInbW2$4Df<_#
z+nl{ZOk)!hotGVF_-D))q*=uRG?k;1y4_c$d43g9UkSc15Jdg79%Qz&bbIVe1~M7@
zeca{o*7VgIk924m*!!kzHCVX1Ckb&+d$@nXXXi|K0FbZO6;bF<rBz;tnZj<4k1gKU
ztLy3<?vH10c_%zbq=*Ew*ew%NQ%lxx$7XG|x;)(k58N)+4o(STzQwz<>vhF{K3=RZ
zYm|Vznfw|2{WRTc598b0-OS_3yaZA@vcOY5J|Yxc#~Vl=V<^;|AvU>>Zc|Ehc-=4G
zK&og(KVKk*aMyg4GGp`@rAHdg!k8uc(NuXC5gExUSk>Nc=HnatbXr*(U7IpoNB4KQ
z)@-}S7ho?cU&zD51As6!H3d0!_Q-=7|BuG`3G++$@#|%G^@H}g&h;*kVrky3or3If
zcZ8lv^(mJtFk7NNV2=Y=eOWxu^AM-hH_FGc)*Is)W4qZ}-MBFT1f6Ev^`*<%Q;@k6
zc&F_)n#Db*^b`D7US%M8?c{@06D3?@wxDBa+OqNJW^aV|L&(|dtfi=koG0s;bujcR
zoBs%&U7O)-ftY-n++i&tZRwe@oSU9+wt}o|Q>T~Fn~I<P=>VwvGo`G<%Mt|z#Z}i`
ztmE|Aa&4{Ch3Avz`SkCaQ}LmCkuao1=gyOotbLus=YN!nl#WhjT2Y+na;1K-CK{_Z
z6GV|t%fCP)GV|Et^d|4`@8c#7#%=4SSlwQ7UPWDuKdtZl4kKix;j!D`)@^Xy1ISO@
zUu|#S7#uV$Yv}3aex$sQ2!70Rk4b=>*^TvbjjNW^XgmoZQT}RjZ@89~n&EWX5zzUn
zE_9&-fTr?)8YCh{sjm^!Rg(5ATGusU_uSBkBjk6fm(Tg0(Oa0!z3~R?zU|v;d=R=h
z{dzM(lr?Gl`fvdl2!j5~svj^T@LI+Ak8kCD7`fjyPY@*X+(&)H%X2@O8}4EdZC;AL
znnrDJZ~p>=)YOSW6K0y{$5Gq;keTZR@+C>ne4lT39VRX=E>ojB+gH<XIwwHeEs$#5
zv<6kulBIxffXx(y+o^oYcQb_clGe=lc2b&`xl9Hkj+L17-$CYEo#Fu7`uurU<M+{K
zE$b-#>EXY#%fHaM%{lEXkm`9p#6uUYvQWAsgkkEs4!n8H9CH(}-{RA~%`nep@f=yV
z>y(Qnp|tnfA4R>(q}!W*eck@tdgJ^7Krn2;R@*YST&~dtkPYsSCHX#TnoVb0c2VU%
zSCN~~<aUD=@?J(+MTK9Jje}Rh=YCcKb$SkDaX_BA=YCG6L5znptpmgYE^5-f-zyUy
z7n64P_T&ze($lfg(8vV)A1*hu^78cAoSq&#+0>JkTRrt{)Bz<3vFgg_MqL{x9zyXc
zwOI`lT$Ppl`o2K55zUh$P~c>@d1@V%ss1e8;dWs)7ckMOUmQYzI=mO{{#XmP(H<=r
zdC(Rd7>>cPR;|sI)DAL`(^D^>ps&+J|CjkZ*)1i%v7gU0TV7`jx&o}Mb0$X|C%^M=
zv$o&Fu?Tsbig?fdJY9l_=)<PDB}g?>-or19y7=~2(7Lp&OU#bKaYg-FcrP+`H)~4_
z2SdG-H#=M_yXF(5Hh9Kf{;!GnzpiiJ^Yy;nf?nS?6rKz~kaw0xVKH(3j99KzS9t$S
z)wZke3^^Q*!UBHK+0|DX0Yk)cT6EbdIXQn~eJuBW9L(3w31b3x|6NhAOC3O-I-tiR
zSYx!n(b198`E0YZm9gAj{FI&7?VR}mhuu@6G{*D%8v#M}gdlK1g_74EG<uM@b93$Z
zG}>1yYzKnjUMRlevR`kSv$zeqFCLFF68XQ?*PrGCtV@=rQ9-U>AH=X2d^nP*6ga1^
zJI-p*7~!3=mwdeJls^63DUYP}Z9pvvYod?tZP!~euu^9%oHRhJ<MG~BXn&@zj#=H?
zLRmRcN&;D5V?hmpn!q><yff|)Sj!sDaC*wUxA{)T>|TQ`_@+4X+xzj7`X8%P`Lmie
zL)*inO^Qy@I9i=_lEDkaA87@L&sQ7e>@dwYJWJ)W4$W+<_uroHE~`W|sQ|IMJ+JhD
zSLFNiRp)Fx76FUF8;Pb@YrygmEKmBxt}k#=z2=heFR-?xtn3)s|0!rM5mx%(Nzmu%
z8l;FGvjrdTm$D1{QUB^%eXDT_@_2H!<T`N8y0-aM@o~$$wtlhZ##6}6oq_oSUs6)C
z3@YmTMU7GPn-1&N?L?`*>9uCl_`i7T1L5uQ8AM@t;7NZr9Y0*GZ#I>^J>6~;G;Fuu
zX*5k`apBpUST)?)tu>9A2!Oa3;MKpQ9SLM(toZ&OOF()(HjdixcTpa!pijMXsO&ip
zrL!0Vc%$AP;-aFoTWnSVF?gHp-EmAkxI8@Hx8~-tZze$!r?LjTNl~lin!JB(W|WO#
z(Jh^=r)<LWIqr(4FtM~$vI`8+l93J8)Ng>Kop0}J>CPdseKCLXHQC2-mNrk^%1Ei!
zKLryMl*O`bZT3PqAK~(QHRTRE&+v!+qdHc4n)|L%AcC-*qhQF_D|x;ahUcz#CxMG&
zFH=modLqA1Zt6q=(9Rq7Sgki}2U!By*j6=x^Ky`WoNfhmaQ0?SsM}r7b$L3M#+h&&
zx`T*QhwGV@jZ&|_JEhg_a3t`>je@P*?tKnq%UnI5fBYU%$mUjxTSoSsk5RI3+kt*}
z2mrw_lWX!WM69yvmZQFGxp&;4yVt^%L2K_{Af3#Yp7HR#T@U^Xq{A@jTJ8If*DLoU
z73>+{KM|KM4CLOlwI2e)9`NR$z|_^5;GYN=AQHB@InfQ&Jiq+vWF2Fq&T{sVdGpFM
zfTbTZoxx@<9*wP3tH(cb6o_+$Kw@fT_Wv5RMHOZwqrT_H4nyn4?n)e}*Uq-fIjbd~
zgY8P)j~I5&b!V@<eW6CsXr<{-cRn!Ed6$c%lJ4>M^TPsRF{|eaVi$j?5$1th<TNgk
zT4GX?>lmJW*X)#QdSc@4G=~?mVAKs5g-ONj7O|e-*2^RQPE;*`1r)MoaoOvwx7fzC
z=CGHzUu|&^hi&)rqH#PM4SV@^($dkfeaJEk#E+84;BhT6@%X*94W#&O*@4Wr>llai
z68Bb@@V4b|kYAI(dPB3o1iq+Sn$OY_>2H-P5Ybz0d;?*C)h*5!L{ybf`U@`ab;qZt
zYi?Ddao7M;sji2-wO#72mHLdI_n0jI@Q*8js2j#95JKT-qF#0@S;mTG(m?R1g0bD4
z5o`VHdj9KhT573xJ0~L3r{2i7UKi()tn(FASSzh2*Qwe0NWL^iGAc6CWxI?j&qq0C
zwiVKSv4rst{VA|^Zn~gDDyTACu)fSu26i&1669>;vuaTS#pJeTPlBF7&LK@TiBgVl
z)iVb9IVHr!oWSqpr1XX2al<(dG_Gu~`=ZacBS8{VC4C=`%Y9Hd1E|(wb-&!e%{F;|
z-OQ*QJNt)%E(ReBlpVFZoG5eo_a3MAo&L;bVPQF)v94(c)KsVly7Yjgg~h0&(c{ye
zWBg^ipy!D;W^q=nMc!vGhO4e@6O`S^n|AY2pe6tli9%n&>ht7q7pJJCWLVXe$z`8Q
z#alaz=oa_poYMRGenrVe0-^{kgd+QOSJMqKIBbU?<ebkIhCPkcKgKkJJXkFy8JY1~
zljVRPN#wJNrkdYl=l01nw*d4)PEwVBFRwF)r*GEdup4kUis{r7G`7Vlbq<R^Lhiv#
zbaeA^$ngiQbe3HX?MB1!#u(dUNmSMUYB}IY>~f=h-b5gcQTHErT>n1<U#6?=GP>2F
z60R$ocTe8aHZ=S_R%JGsZrH_BWSrG!b<suSd!KOAj}HpBS($oG#y}^eW^X<!_1@^6
z;5mDiL5686?E)0l17rqoFErK)^xmwZlP*D<NwinZ#dK5Tg%WWsSyo+zK>zn;f1aG6
z|7!-OKB)z-DLjSZF>|%jva;w29UW3_H@DqS`YrIlPYEL5CNt~Bsjq{b|HAYw+WByj
zB_qWV>U(opmNnTZJuXP@Jt`SP`(rAg&5k`@X(MA}-6;vAAD?%EpT<rBJsrHY*g}4d
zot`5fk&%(TZwE{r*97hxJ06mtH7KvD@_E_`a}9tY^yqst90E+dFY9?%Pnk6~iFH3M
zqGMs5$uE85`N*2AXll9#Z3c1h@Ma76u6cAg(rSbJ;Gxy+%Rg+?YNO53E0?MJJ~8Y6
z;qE<vqD;52!BIyYWe`V)MiE6uLCH#x3<@HG1j!jhau5UwlE;9k4U$EWAWhCWCq>E0
zh6V%yL2}MWX3y8ecIMu@wOh6I*H&@oS~UH=>AWXA=Q-y~X6d9F1Ly2Y9&U=Lp(AkC
zxrWkPDN?wj^-_T`qH5<xT3Yr)l(k<mJ@4TZCUx)USm6$}Q`hvcSHI@!*e2!nlXrNe
zsr*o*KL3f)4lW^2q3<MME~iNR01ZlSj5loRu+;~=$>w%bHbn+c207*E;4tlp&Q%80
zY*Y3+iMk_cCnN=2Qb~M!6SF&~9cmX=DY{FOqabXx8@`v_q1V8k5B_&q`8&>{p2e8~
z_ldq?$q1*F@i;jj8@%TQtywQQY4%G|Ug}3;kB-`-I#aND6t{-I@}1PS_!MuEn3t;s
z=2K&xZp<1VDD>4rP;HvsNx?YW-}V~|`aoM9b#X~f*zpe-_VgH2E}x*F1S21vY<CY2
zj;!xKZshD$GvQTz^u(bd4%4DNLF@C`#f<8Ca_Smhu_7L4{KB86n)}AnDJMIlM8Pp|
zqwdR|`Sjwri;SP2->`pX&0=PYM!LO|XQN1yQXL|2;C;@Ks>(|@=G#ov)QG!NaA_uJ
z7He>DT3S8QDE%VT4W|dVk6|brU$&W<%`%O%qm6uA<wh*^BK2--t%&_CfNAx!fAG!J
zuu9${pAa7({w$Z6Io$uOTODopBc6eQfvx!%IG|08Dj8o=wO!xK&{Q@cv2?cnr7SC(
zjF{zoQBP&Ich*UJ_8l1XCppRT>)t;YhNFZjA9n4@=O%U>WQ~H=CB=yurOZ@)_;AEE
zuW}WRq@Q!GluXmoYC>$n2_^qzsp6amrz4w35YaF$R+~cu)iv4uF;nk=X6U(sg{F$3
z_f!4zE$TMGCZ1*G<s%i!Uz%Kmd9DsWXWZAF|3E}SLoJm*%|J&dB4jMbyr<p4?|mhZ
zozHM5KN$UTIp4cKSfX7ye%$0lXJ(8|aaFazy(xoZ0l6zhtl2Sdx8|=O>X>yvU=Tn@
z2y#XM(XG5Z30{$vRU@gCbtfPL<jjHjRqQQl=Zcvm1_wLpG4%9-?8g0dhrUBtCMGI6
zH6A!M!Bv9RtnH&uJnBv?BB2L}BZp7@G(b43-6YQU9l8PK*)`*sR2#PhAr4LIYD1Hh
z3x5RX(XD*_w$`lFuyP}6JVD;(%J0dlxcudt5EsKsGiwIU6{i8B<)+vjtb$wZI1jqu
zERTNz^<rzh2K|*X0RT(Qq~~5)q@?kL#avJZ7?_%muLCVYn$Mt`H3}F<gL~oFa6&Ju
z!s2fyVB5990I$y_JD(H?z6`F?Bm#Z${HBD%bS@dmrKP7I{Qbn8(F{|!o0$M<g3p?0
z!4F_)ruqYBQIM&<u=9-=Kz;brViJz7@+LqyRcrWGDtZVP>yw%o+rj$S9b$;WsU)(v
zv8S%anN`Fe$xw6S{C_P10sx-R04JsLEkw^QF~wydNXDHsgR`S*W%>P89>yy|H?3B!
z7sZV8K7INGULi!f&J7Ww5LVkP-Qm}~o|C3i=?zFom8$b}<%OC4+FN2`bc~F|PwQpW
z)ELqOd2KD^<Z$mF?bELd%GIrY4OoEf7l|gisowa@Q4tZngV^vi8qn@^@t3)|muBh(
zG_y_Bt@P*Ms4Dj+sO7RyQ%6>nC;U2XGmuC+(oylU!QLl<iRfNTQ`@Dz^YMtYsivkT
zF7C#CKT$nDOYgVYp0?kj!WFE-^WPkL*0Fw4@HdoIH0JMm!JE&|Gc!xsL@pMuFAFIA
zMr8^vKI_tDoh0rXe+Z^rq>8A5wP{Y=E`N1e+vVtYN&dy&<ApsDQBi4vAd62u1F3m<
zRKK~!9~k@n(4h}%d6u!J9eGB{wxB1KslWoj`+BjC=Mk6c@#DSl{C?#^yG(v!0M*c>
zFbV3X54|4@)D4|_lD}?ZCn(5D#;k#0HF5L)0I=f6aL9f7FogQn>%q6CAZe+BznS|1
zJRcw2my|#5{mne#C`?n6iAqBhjbg-^HUYTOe1J+)Rz=0Gp6^EtpRm&h_|6B6lJ8M7
z>I)Y{I+hUMq1=N~AC~h5K{Z59q+j_hR^iIMHs1y9gia2sH~d6Yd;Cc06?iSFA^(v;
zEI2BoqSE&6+P>CnpdBkmM@?WQ|0Bor-!oAE|Nm5m2i&(~^kMTYT=BsY!A+8!B`LAT
z1;HCO%cGl)ieLI>A-}GgOn&y`m?H5?sr-|PJoAbU0gu<+B*iZm8~IX6_%01kiIi1*
z?g~9lUc{EG+;eim@(rU<atP70BO2UJ?TS+~EtM%wtIa~L)}ht>_lKs<PcmtUCC9Af
z6_hB**a<%S?rnyCSyau;Y2RH!!V)A9y87f`WnJCtzT&g2X?N<>Uy%Cq3J7Q_<O`GW
zU$`)E&^Rh11IZ|1h|D^gm@+OhFk}q=apRSzg4*=VKz;QeG#nf)FNv5()aGT?uH7Vd
zDpyY(T(6g=mZ|+O|LcxPj9P>&p6;UVk8I^j-yDB*KzNjx(`)g{r^#u$SCpP<<;OQ?
z=kK2m(KF>fzsYHouE&t)Bi1{4kuB}JcXjs^+(?)K==dsRGAQ*004u&97Bteqt_6m3
ztonxMJ(e?<Q}m>JHGEj6R9gL79K507CaH+k{@Cv^NoNx;i=<vem130sMBkNW5c_TW
zpD%^s_tX((sKUO(*PMF|BD})OvcgNGwZBSt$E1$M+{b;fmg6)s;i02o&pJ)|iu~)V
z!y?9KWf<^Z#6R%Zv0qW7)_LCz7;SV*Hb+1{rGlhis3v>Gd+eIRP}Ngppv?&gK7NlW
zE-*^zAC#iMxi4>V`vZ9DA-JaaNp24C6!!(Q?s&_$KSuN`_Ugz#pR-tcsTLcWb((lE
zNqpBgsn7q$U^kJh%_idBGP19L{`l75$QXzAqWLGwy#m9_M&|Z(IuML^M|#)V?n(<9
znP7`oZ8Q2|Hw^y%X@%GBU@wfgHz8X?Jo%5|m)aYTv;T+Ti<<Ah?2nnUbNriwa_d;|
zo3?8S@OxiRa@7po^Z=M?R`{Myze8v^zO9}4^^tIX0gu!JI3l+9`7B!7pTO77rWkyi
zbipt1(Ko`cDRp2%uRF(b85L-lz330$H6Zn`yB@Ffr{YnWeQ3$xG%;iZoQirt%FaQF
z0Su9+Gj*UAkrZPq!h;8_1yXDHr8BntG<Y3mXviM_QR0{Xvhl|R+fxm$eW1cWKaXH-
zM>SRD{*LAXnG{o`+1d<P#j)?kXpI8jFl1p;|M=wPAYKv-`8RhDzicoY1Y)6pe(m=a
z{_dX0PXg<359{6FgFgjJ!nGD<8JJfZ)=Kxx+5Kah87Y%)$RYF9{+HWF^}z_c@^}Ka
z<ljamBC9h*E?ooRHZHVb2I>b4UXdraH2&igi^t6r>>A0@|K(m$pMTCs$K&-Kg!&@k
zx!xR8k~4?%_9YbOh-QX3;h*=L3{B1@HnSska0ZWeU2{mMKUc!t_~|GF&~1YRxA@py
zuf|%SDL}MPytXs~K7x(LU#dluT@_yEz()J~+hZ|hhh+$N_gC}Y-Ze-B4m=6fd;j+y
zT3T9~<YSpPgoTHuwY0QSQ&anyW0QpMu@PLZL&FIgg>5%KcNW^8S`WIUB?uYYp7>+O
zkGmiC8x-I$>vSWWy&sp|Y|LJrWR1d4o`5vmCL4h_V}Q^3j)qaQs6F+DoWyWTvZjiP
z%8*}$^Iw1cg<F8EjG6LJOZGnNzE}Upct<gy{rRGci;IF1?DCS~H?GFrn(6NDwg}=V
zx*l`%<Vjke1H|OLb&}mt!uJS;>`-1srL?>@5L37qQ_OpA$e-onQug3|IQfn|z|*s9
z!5fBUpZgppq>&Mm`uL}lyXHCeHJhFTT(Ty^M{aJv%Xgufi$0%&^~Bt$+EYG74a5#e
zbo$2vo;RQOk1cx*0e2|3)i6BW;?o0_J3yCs84S^1!#QnjZF#Z_TyorwO(|L*Bd&tS
zO?aJR&HL*^lNpd~3j2*KR!zG6G&E&R@ibB)2^xO>{sUJ3kc7Qn#Hy*Ryu1Z5F*h|I
zE7k=;E-;>qNh#&3pBHN~GlNk$fgTr4(JTU%f^<YX#C>N%eSE@i)V;a<vB?;)l(?#h
z=;-B&e~}#7s_N=;x0B(#T;L-2oNpCvTHt!npJTTO3DF?y$oh(k!MVnoV4fcs{X@0=
z;bcObabds3)~e+l8)4#&4}Mzr$AR~rT{lHVD;K7ENNK_`7+lAM6*!cUG>6W`8)Lzj
zWn4D|#*nO%(!<X1rx;;ARqcV9<lch)(D|gduTR;$TUR!F(A(e7PhL(gC|ghGZR(&m
z(e*w9lTvE`^c&Z%JtFm2LoHAH*&C}`cBfuE^r|9PMl+D?Ikz?cQnxE54hJ}>f~`A0
zD{EtP%b>gB@KDt!cDp3(1NRW)S+#J0i>Y3}8vT29`G=z03Z%}Pn6KwokZ8!?sq&>w
z>JK-&O2?Y8|L5JmJ4L}`&A`j6cG3buT=26c(UnmUR!Y$xKXq!vNTH}G+Q)W3?QV4;
znBeE@`=`}1&D?Y9nK5PcPZM9WljHXeYy5!Jg}L9vh&sA0=Q#hijum^}lx8!bF^7{}
zv;k6-{A-|8CCL~oD7@~$egA$plz-oDQ>^6sE4;l({v6%<_}A?EVE>57NLq9f++g!z
zNMhmSY?MZR5+P6}6{)6>OdfBPj>KT#N(%CscP81$NJ_pMUd3gbX(}>GI^L5$MYsr0
z$S2V~pC4{A&N3X02=w;GXk6^)z2EhB-f9;_aF!e^fZFE0(1N^)(zV@6(AV>Q4Zr{b
z_n8}`rLyhs-Mx!3BBL9^hL^iiI>ivxhYEaKeX0BnG{P<|>Fasb-6Z42RfQIe_c@VQ
zzn|g#qQn4GrKPQ0?bx&10{b47`fDH-kh-X-sHzXqX=!PB#0ri^KTZGnI^n$#`)w*z
z*qt<Dw+=Z(m1!n@YPT7Jz>C1|2-E++`Pi`5T?U8x=|T%8>!|52t9tBbqpFN<6U^Cl
zZ()A<gc=TfzO^G#yX-3b;?aBT(c;~rLpWa)^?6GMqn_AZE+3T=?i)oU!9b`v`=10h
zCB~V+WceQ5<v}2>@y{j;{$SXQ!2E)jv{Xo_Wnt)IipX8|_8%Vh`RhL)5^9cvz{>#B
zxvD`ZT`;uftA<@(5}||Ud-T{}D8DC+1j3vWQ+y;VZH3_QVSCeJxNEVy4-OVljkyx&
zvCE$%@O+O63f5z;gzU1^$OJ)1embYM^q)-*l~Tq$qfz*O-VLhrv;W+T(0_`=|4~m4
zJKbvDD?XLD|J*@(<+ah#(ZE7poqQJCoVEuR7CM;-g!%HTn=pqJRC9R&LPf+}tjUnq
zdly?!z>7X}COs`JA*XlxiS(;WDD{WRqg7R6NchM=7sO-yAx5YfFkfR{SI@Dk-by2X
z*>rAMC37$eQB2_2&@wY;50<FfWTJ(M4j+c6q`6ixGeexXlxd8;cf-jVj7ZmE!JHeK
zJtK3Ly5Gt9S_X4()Nk5@qYNt5sQ)6GR7z~WGXCQtIioVC<aV5E<Pr#=$?0$Ic8ftC
zjKBR?5qc~v1TwVSaRW4HozOq}|Kia?xj$(If8l#n-~P+JVjzEv%5<^b<+&o$)cl7c
zL(SzQnl`MN@Max&EPOAl?Y}<1>Yp2I_G)@(%X1#XfIkj5+;*bdj$A=y6aV_^S%2{4
zRw(_muKj?aGT5!g;q#bcu3Z-EDgLRud}(UrL@a!F!oWjRy8^;e%+otqAm6(01OxCn
zp_+wR5o%dvPy!_jgAxW^283Y&GARNTO9je_-Xda`E=LCO7nTa;m)J3+F!P_E8$%#l
z$RB{%?g}^H6kj4rjB)o4AVF`j60U65wh}}!cua!OJR)IE458T-+<Vs6);8aI%==z`
zm3LYncsApk0Zck6Dl?DJB8xrOWU(Io0(7;r{QR8-N=8OT00tivpn-J8l`B`MsrRE*
zI(FYJ>Q7RlH*em~V#_uFCqp{y*y;d~cemsKufz|<D(J8_A7#%r06sEZBT#(+$!ehY
zTEBk%dd!rlsHhUI0hPJS2vB97;Iee1;F{hiq^2zQ%5%8E0hk^X1MWV+7$(~ZNFE9A
zYu$Z~sXzUIB%NBOkvtG3{=9~O@TS-&2T9`g9y)aBm8_4qHx38MEG_AHRYHI1>B3(C
z&^B0w0eJOV`+2S~5`-MS=+PP(6B7gQJ&V<JZ>16&Cs>Ca>mlgy8nR@EPKSqwlTh*}
zIOv*UzkE>&Xv?=TgV=-dlF64#w4aeIIXN4wZs*cb6;%Sca(?1lvX-H#bSqBYe-Wrn
zimXArwrPPv8X9K^{h#sjb__&9M#De{P^;DY@OXcmTaaz4!DoU9_UXb`3=jd0h9U6|
ze+A6-IB`3>JaY#i{gg?aTb4969Y0A#1H?IguOQlloJn8?_?YFEF_Y9LiH9Uxx8ii4
zWPd*(D<dPLpkQ4Vm^vs)_XoOvU_f<t^5Vgh^h~za*7JSz98XgROSGjC^G@kzG#teB
z00t|4Nh+D&Kq=s~0x>T@^a=OrY1r~vrV5V>?&VVQefnG&2-T5MaCX^wC`KlxwWVmU
z(0UwNB}oIP#6$kLRwR}zQ-Q(mX_$tuo%Aq5?bns&Yl;w))eNt%sbO<wZbPfY7Xlu@
zYse#JFWrW2(D0qdWROvRdVto*xqnR}m__$xJlEhU&#E)-!cUqoJ5Gk!%`sKED{~=y
z_Ek7I9mNM-A(B*grM36d?zjtLHyM@WAlN=9i=?lNj%-H>F2YaV?{+nOJykLW+jAet
z?`d)_X98v<?>;shp-dQ6W1u)A?mZ-FPDcYjQAnaAO_zx9L_0GL;6Sfme`@!=Kz$-2
zC@2W{+wt@Z5ZNfPf%CX4mJ~E9#dt#`BY&gDw(I>Kt>|CoAeDF+<Bj<JGa5>4Zy1ac
z&hFMn4*Jrfgx-E=6r1y!1IITa^Z^22M(p_nBORATZ$;OKC+_vLZqJ|J6ce*>wt0`a
zX1LoWKPIE<HG>1%DqswKU9wO)k=U~1oZ<1U<U80qnFPuhz*3wuie15T7e0h2@aCf3
z?iAh|1xL+&mzpPlS-}cmKf~v{)cp37pXu$GLA*x&$qXXyKA6FKRPDR&9-$36%0Qk*
zsCnS?B_7@#tBK6{z+Mrdl^us30fKQJ6vR_J-P)Uq{%7|Wl(|q9fo1|-j(s0Pa8k(h
zn0I;F8#bU+ddKzsx5t+MpY&K`E>I)8RU#qI0viSQgDyV!r{mK^@!3vC{PR!=Yyff+
z;G~T|_1MI|@cXl4Sdil-=0mqV-10;0nlM0VG+04$v#E#iViZWA=jSc~v)8e=JOOH1
z;_;?GhUcHB4`crK6Cm&$@ka!nBSH9$NTbT~u16bVF&6J8Cms1hpDf45#8lMRyJ{iU
zJyc~U0H2E#=D63!rc>bwd?*g!=NuMT9!&=TU=47R4zdK3UckXjflR%+G}lC7H#>OQ
zamjKKCHDwS6qRC{^ZMGb(@|wcAt!P0^WD52e2b9|C%<B{Fc|Jp^|kUXZ~`rW(U3!c
zW8ef)wKX-P4UyI7lFrFCEwNQXIw#a_m^PZ3tbH5`m#QwZi50C(R}VXRf)?WY*xud_
zq_7nzg0Nh&W^sbq#Qcr`1H-%Vj?Sz>AYG0$#p{=Im0t!5i)QLwm-zH`@H)n-jID=H
z0&ezR6{TMAFx7kwcrXo68fKFH3oP$|9-KO`j|h+LGbUa!`N~qFFdD^j<4AQ{X;tyG
z^GR)n`;%SQq667qpJX%moTQ!)q?75H86O{?bLY;Tj%VV_h>4m8{7;8Ve$Qf(h&#Ei
zgTH*df)*>{o5k#g?CeaWTWQ?=a|ITAebUlc^8Kq%OocEt7TC@}Fr@t`D2SqoTW)%x
z$AeiltH%5!idlofr@%nE@*7mE?I%KUqEu7wmTG(OMO78RR)JZV+}IN$beq8z{t%)(
zn_kb~=wmlOYRGatAnsNjKroz@53QVlUYpc^$*|%4go@nL(YRi~Gp$l{ayTENWV}va
z;Ne+L*njS+Lh;eWoJkAo*L~UFy$AaGDvj!^s^VI!g_<#ElIFd=y{D(A)0^FdJuV%C
z>;xXBo4?Ifp=4eOL16E1+DJGSrzj`4fWxFo@Pl0C-P6Bu<5{IR3sfR4*8skI;zHqc
z6}{0)qm;nChDYj7`$w{8+A$t{^KV2o;{Ze&K`#JY>#857C(Whf2%+Iui3+OfmqgC}
zVi5Tai&GtRC$FbM0?)KIAn1wRx>cO1*Zb=|EVchk(aJ*FV7D}<Q7N^e-|tA6X_9}d
zTO->5Jd`t(XU=H!IV{bO#aG8$(Ide287?E-#ThQ~AWp0*>NJbabK7nxsxZ_G6>vTi
zX1;+W{wQU)E6)nw9aVBL0!8XBpA$&Ae4T!o+sV9wylXx>AftA@-qg2h22%8D$<2U*
zt9k`!bT`O4Q?(BYengUbOSm1SYQWA2pt2m_T6}U*oCW_4QKQ3>Z!TT)5SNs^5-Oyk
z5}3`9=eTS=Q$j?>gmY2vc=;0V?HBF}Z!)-1r!U`qdS;?d*3-?~w}pHD7`OTNr;8L8
z78V7%SRfqDlmG)VZ28!3zrf(V3L9{^Jp`;rmHcMBSn^f?z-y|2N)*0nz;pW_c&@gn
zqU57Vd`vp><jF#y4sv7m%@EXWTda<(n5hGEKzV_=%iQn>fIN<z%YNR3&jMi0{s)`@
z&1d6*@0pUaX9;TI<;eP>=5RMGcN`7%pXwf=Kpgd(x!fk7P1aFhn|>zuez)KZ*LfNm
zyK2m-D<+qid10nMPNf~gU#OM=@hx-Ao*m<m{%TA>;;xBi+=w)iUf^Q^gx%o?g5<vz
z9vjb|c;mFX#MSv|WAlt_cXVz{tS~)2{aiXrD3lk@=<A|LNr!lv1fY_^;)HY$Jslm?
z$LX@K0VSuZw$_UMV8A+%j|Ujh5VT+;B1VmoVph3Ig+!GP_=~=kM#BQN_B;!VjMNg8
z<VfiKC9BTIh<g~?0Hqi;O%!g~7v`#1ug>vF)-r#FVD5AXK4-alcPD1W!Oam`1;r=l
z1GyFF!Ov9D4?L$><m+{SK~okB6@1CXYk5v=1K<-=45R?X=0eEBl6=&?rZ~SwkymH<
z9h977^h0dRBH~`-B?i!5@ItpBGF%<T8!mm1P$!gsb$y{)g`$xbT_o?afI_(Lbg95*
zWx4Tc%)%6ykE4yz_2-UEwPU=U*O$LpbC9d&TJ-ry@eftzZ`d`do!FW^s-U0%qMjRF
ze5<pKU{7Ew4<QfHyKXJ-l@#bl9-2GE21<J3TXFH4?B~H3!d8qUP_Nep@kW@C>oZr^
zVraruW&{4A=jxHS_>I~3Vb_7NqT?jRS@mizDL8T0g2IJLWARP{h*Orn=ctpj8=ZHx
zn~GlMILDqn2z%qk*LNBzXgdp9c<XSa8ORq<@|BR^XwOYQM?+~pdRh7#xa{q=%v^C$
zOqvfAB&dd-J}$+Jg;`MvJ!1g&AMk!vv&8G60<<thNv@;IvN=H+YF#r_yWp#NGdE@N
z!E&PRcAQNtidOi+%XGn4;_a0~w2BEzRKM0jC%xtxaAHQ*_6c>Q#=K03Z;Dq0c-$F6
z0q}J1;59jU3gXE=v)B^wxw|^VpiFaQ!Fl!~_hBFG9IpaELYl#d5s;XoElkJmGBE#Q
z6q7rFsC`+l=7i7PX>$$xF?Q-oAi=hDxu;x293%q<Bcib_|980b_Y;?{3fLe{f_!RU
zwZF0GD{}V3X#5@9-mZF9sNvBrrdoY?JAX}f$u@jCbavK~K@9PT%y+OLw(AUSaZM%P
z*7=)fw7zB?r<nkPS9^w`Y|oge0LTLHmOzzQLHO1fC1Kp05Re+WoS>X;G;h8!vkD}(
zAkN%y-mte*28oy*oSNiUc`2I{t1mg9<2s0NA3fy!fO%^P3^Vqku?r|_&?g5XcWER(
zxOnvR#;{XTd~?3^&$EnM5O#Tp1I5OK6}B@mY$q6(z7vINI?4;wyUv*^$x(A~*ep)>
zK|vG6r8`&Qlpqx6gK`YGC|q^+rfwkE5tQLYYP*4bI4j3Rm!TGoxzvG=OIDiaC^p1k
z{}za4a7lS~j!dt^q3M@;SN9Q?KaSvC#b&+zh8G1k9DfMd9eZ4Iz9x|mE~XsX*!&m)
z&RuX7;=C~oGjCzwMuzC3#fDKRN~)OAnz6$#Y$u5k?|&n?ZZY}Pvxod9tqa5@cpwYg
zHn5f8ew}5|=v=?tlqelSqROEKRoHMd=JdCfgF=SE497q$Wy+Z&svKXBZp?Qn7>pmD
z)uK|3VeAZ9|El@>jKh~_5Iw`lY`<`B=D%Yy8QP}D7QU@cAFp@4r)p$?ZD?qK%f!OX
zod*=))7+NNpFe+dN$YVX>l<*Kq1e3*DESbp1B7DN6erJeP|@YVg9lI^4~+j%a80X&
zKodExESR6avCG>hKa0B8=<tDZ;>`G>XU5!h;?7lY_n^i{<D4s-VS`Rx5SO(JD0c0;
zcPWUVv>Sy&ao6!OGA07S(Aov;Ro=a?S1vt6R!+>nePc7<3G-wqTdo(L<E6AVNzM|U
znVtsge4gK=F{-x8kG_vHG&r~rsv11_mu}{o*9NkeP0L1!OZNbEF(4p-_Cn^<r%#c}
z(l=1#|MKNwZLQX`x>+nkz>xly`yiSEg}B=HyRM#W1nw)iH$9gTTV^o1%b$;-W@t;T
z{mbg<)cr4BmT7rre2|>=3cJ%V{?Y*gb|B7*8(}dnNt3=4H-D!+2ip4%gG1N*M>G<F
z6%TndUV*1^18kG@JS|@uAntT@Vr8x_F5MdgrGWpxp;Th57}{bx$efdwV+VvR<`~;@
zALwg!gR%?K&`^|rH%-Z|2f5F~Nsy(tb|K|i{xM-@LKF8lw{bH=eo0zbTWtW|s65$;
zUKsCUZFm@*Cf7i!DtY4cvEZ;+LX|mq(MW+~P#|QbKfQVLCZ_DEi%VtbM^_RXh63&K
zBl1dD1ljdK{TaC>P?t~h+0XlmhC0rlJdbykwgUgTWT5z@_nJAM99v9bUDT;Xy&n?!
zJcohS555A>=b$MRf!tgmyMU8(3~&NeXj&fz4=(Bf{32HzAJY-r?Hu%DVS&NvNbVyK
zIY&F^63JpOfeYAFc{7s?;uE<xP+C6P-Ee?SzOz6QxUDT`vkV?Tckm(v8uG4o^(|-c
zm-xU#9ogs&^~>hTNZx8T_tmVX;R%IK$U?sAM|8d9<)TZvKWNRkw4laKECk7o#?v6h
z)Q$cbrX6ixxGjOQF|7|D2BRba)qFtD#B@U!iAk{UvIk!rzpzOVAyz_5QiT4=XBm8E
zZK852PWrZ=P=51N8QJ(Zn>PBJ6%`c*VD|IKe2QMILdCz|8TF3XBCx?MR*>f&uAiK|
zzsO2QS7q1@!BkA{$^0+Jy>b%*;4t8itCN3e;x0LF5pmCsI**$iQ8b{_wg5!(7}VIw
zEVY%P#V$WP=|=`|PQouru%Q0=oItCJc_f9DM~utH<@f#=Zo0Kb0e)y>37oh5gX-E^
zmv?laK^?mu9ZXs>)UfUIlZ{q$N=(}ycNaj?vfcEyMCGM<$%jQBuGhO3+DmdZrzcvV
z6bk^VH3V6I3=Ciqcb`v@Vjhp>qm!YgFHzhbjh1!3AtF*83WdK9R37$K`5Bo)N$*9u
zG`Qg-K!>H%&t0AW3cAA`f64c5pFQzSv=dIh6kRVd?eHc0M@A2O`;EFZt&a!FOO=3+
zn|ICtDJ5ZL9gEV_*N<;G!);kH6n$GtQzgT|Yv}=mPtyZHRt`pKKs;$zbMde|Uomtc
zDZVIQ#qH&nZZ*;j9Ox7AqClnJ8rjHe6O|(8M6KpbR4Hn_8C~r9qJfh@SLsobkww$Y
z4Pu2f`f@y>fFLe<<u#)!MRb2(U#$O#svofEt3$zTK3MW<O27WicHs0`x`4~t&;E++
zC{!(t0frB>IdBi;hEij^GPw^3zU&4=?U4?)!iiqf(J*b`Nn_D1-MYV0aQk%@n`C{?
z3zDyXk}9_ZMN#z|%jpZJ!CkDqrWg~~DGEx>C;I_|fWhap0p!Pz@4HU@m-STdIS=0N
zma0oo(&IFr9xA%$+7N^|czygqYa6`Vqm5mobe?OYm*u6D`Oy+VSr~`$eSQ<1XmK&R
zv>Vs2b4se@+z%5sYE6q&NHTGcJ2MA6Y3*|B)-CyX9GHQ<(v`sPW)AtCgv3Vv?B&bS
znKXsn&rU(Sxl8rG6fQKt4pJEhUJ2}8gqS>FxCm)T!G2x0hl80RRFO1mq#4?8L~vtA
zPI1TnPRXx!1F9f!rgadJ$a3T<ULc!>>N@lVBt(~?VPcX@o$)VTi5%z0TZ7DuZNTur
zGhu-jk|tQor~x4rc19HqjZP>MkPpp;SmGWNrmleQ?it!|N$Na~{)n=h9s8=N{tDt`
zmO9OfxKZRl)O+a6B8G24$;xv<`Ob$u#8W!p3D=-s4Eym;KsyNt^@~;IvK+cU0QXu)
z`}2=5ycJid-o@5gHIQ!lwc{fdW1=036m`Yo9KJf&9SWH8^ZaEYr2)=VQz%BEJo^p5
zV_sleoP0dAR2gbHteN}PM-uWF52vAJ2JOX*mzY~)WiPnYM~B$OrKEiJKuJhQxW9b4
z)Kv}&xwNbdxb}FRZhU7u3USIDoRn(LGUckC>AnqtD%7Y9tH@aec)WVm{Hy#$tGPS3
z5#Lfi`U0tK_duvKv>YM1<{p33D-B%U{_DvxF%O3#{pGsLDV-fa8#|@`BWJ^&vJJ+r
zD_i0!=7QVmryMvf7EBTiBfx&~@i~HaG);O0JoBlzf=>~lP8SEDyhb(-P_=PSDxspK
zVR}VsmnqC``32}Cl;ns8jtHIg{74HkSo>MAUV%UU_`|3m)rNFe@C!w9q3)fZ7lU&{
zNJKn>_FJJO^z{0{N+(a4tbft=2>28A3Yt=UTQG@&y=+o(J!fUe&&%6B&YvOZ0?jP~
z&J*?2Nj^dpi&d5MW?mO^z&&LqgUz}j10Pevm2f4r3=E@yt^{*ieR^$Hw5aDd*8sOQ
zPa~t5!49A#`9*}Y#>q}gD=J|%cb^F#*1r8~6I{=@!pQA3S~fmW|HFaPiGc{a|Iy9k
z$B!!}s`N=}wT`G}8qGlgN>+0cmKl1(7&bn%*&mF-U=W3lgPJ-#QfzFpeUgwNKh$%@
z#R~ny;XQ`xR6U0v2auWQaidV49iNXyTGSkdikE|UKl8;l`K=iv94*<}9{oeG-qme(
zLR$Ir^;g0cXZ>WAuD(#T3gEDPI07a2@}-7^X(UE1YyfNaoQK-zqURzzia10ETIpoH
z<LIz#Bw{W|F>ItMM*7!yS@a`GCO_qrJCEqj>1*J3;KDYJsiKMsW`FSY@3mXJlq{^y
zf8YNg$D9Y4iX(@#sG3@kGZ`(EFmOEp<Rz&J3DEpRrKC7sk=<n2A)P&zmM;R1_zHb0
zkjJIMsujo0OiEq+6HC*nNlKH%HCpxRR9P)nm}zO@4>UnHLGPKDw~|8GwXq+=843na
z!F=}1UDs~$Q5Jo`eE8xw1m*D3sKTkTQ^u1q63jQ8F45^jEpEAd08~hDj){eOd105M
z10dxoTzw-`>1@Kh$}IzH-NUWP%ME@h)eoM=j!)Zr+dlBYJaZaaoUG+gV*70d`0yif
zSa5t6%taCQToA2WLXz+%bHSeTc%X;HCJ5ERzT`5+b65MdEq_po7WQln#Ne@vD!S*y
zLvI`ztpYM!UBWcEqn;kkE@6kmh`eR(G#^SDLh{YY{z&eAWpIzesoGMsgw~mAT!_eg
zS);`RzE8V0d2pQd<HNfE!@}G4axOsFpYbb~>8Mz}tE>{+^E9xjkaDGWf!Hz=gj<F&
ze)U8v&VGKBw$9Dn-AewMJvT({!#oI2!gd2K$DvW`MJ59a=Jb4&lEEdI*B+eR0_%(Q
zU{)}Hs$`pXKuH>>V}+L{k`}8QB_PkG<D(9{I_XUg&fH63#iTFnwhf9QB+*j(+FWQz
zGa9cfOaZ7+49%Wg-XV3Y%d<m0-$KMBS#8GE<i}`<MmaNca^C3w4epH!(YAXtERRl7
z&9<>_SE+YR*SEMDoar4L%x5)>17)@bC1s?K*Opp}j}Q@k1!86V1(PK{$hK7>_^@>l
z-L0f3)7G^8$dv&v9*LF}QU{+A+Yi*Lk1XDwXFz4%0GJfQ0fZI`VZ?9OkkTKhZiRL(
zxy@(jHG9BJAvx&_bzkQst12sp020Z2tf`=Yw?uQX+m^_#9`p9fxa@MctB?W;MLhWD
zpMTbb3aq<xbS|{kJ4L;H0%SmyzxBhUynwO?O$mYy9-%mDT{TU~V=aD74jPc$u<3t-
zOmXh+c9a*QQ{C@~MkW1lSXf(DD<J!umWc^m8faGHZdNc=7EfUT)fo~1Me=?7OeZQX
zE;lAf&1W~e0Jc+L;qv2q9Qqq^BllKHDN#A=UPti)nmB|+iS2V5I6}aEk8^#v1~tyk
zP{^8W;ib9qoea_m@zWtek9;;Le+&DBfbs_xzWv*2yph)!xVTBrW3L|XIH-`=21+E@
z@yyzOK^>nr5ZfVY+LU++ZjJL|pZ~D*G$s>lP<)8kt6XT%1HuFWnnt`sLrF==ZpNTW
z@=k=DYVfy`<wWNbfCi=<Dk^~%<`F~>^=;C+)liTxf8|&?i-p24bm4;LKjU$M8DIp!
zc_3PW)b&OrZnZ^(Q%N8F>JPm>6zUH$kpE`eAE57hVSP4QAmCeI?7Me&M!%%OK3Gx9
z1h6@+ER>W8Zb(X^!uByVLDRIhGWV~%lok(CV?ur2zKy4DP1Y3PU>;=FELwr~T2|Bz
zJkb3JATp4lKQ}J60y>;(vj{>Z*z~Nc&G`$q{JrWGLQmEt0B66V<KW=H?(7`_;K%sV
zNU{xG1AI5br1jW_uyza>Jl`z0$P4Wm5|NodZ_nR5pn3HPyz&PMki%v{CMH_$905oQ
zdQlXBKOOXvFq1QaTIHVkvW8O+A=$rQ3Ag^j!$Z+@rPj<T9caISwGkN^mP@lk(Ae(i
z#*CZaY|nVusVMcdz?-bEGhy#O^PoD49G9V21hgArR^0z@kihUW;iSOZhvVUSjqtaC
zA3)6;r2lzyxajD6Q_IiD+EH9Fd2=CMrPyf|+DTbMXN`?asewEFO`-IS01dbLMe>Xk
zPXUbj`4q};pA&!p2wG@q8Kk1DeDe75#tJGZaFfipkB*GYG-+)}x(}XJ_PwvDS<;9!
zL2w}>+P@ppzv?N^w;HjYspZ%jf{WZ}6IC4_sA+XeX8mEu|0mgiea`u5<pW13Vm=Mx
zFbGH>`!68D{rrKr>I1P#%dp)O?YXi?h#<g{oiS_AnF6RF_nl^9Mc8Jki*b$gW^U5T
zZ*(-dc-gdgKuB5nk?XBAlU5dW`ym7A#aEv^SquRQEt>)bV9*fp!Y#X}4?1<|=s?M$
z`*S*GW{;OIy{DisV3Ot2vhA)y8MdPgfj-M~0qM(pl)M=+4ncZ~EI=rUMuCElbMxQ3
zxHwOm@-i5R)Erli%U=y<H-7ENHyObCG9MJ8CmywYDQxT-zL^*pUTQPeYyJ@%LZaIT
zJRA@sfK^LfHKErWq`87r*cceZ%YenwKGg>~)Y}EV?ze6JmeDq;JM%cUa|AM1Pjrwd
z7<6?tX~gC;jr><u%q%R9*zp5Hs2u~rv?A%-Y4aeJjFgnytt?bjAEx?XAfvb<ae)W@
z;)_TqCi%@2uAo{JKJ9Q<TEK$sQ|JbhlRYY;SLa%{R=sN#n%&|4B190v&k|KKz4t|}
z4B;?<=7_xo39$$g5*Btb5?g{2=$OlK`Hm3u0;oU;13<H?`<u)$?0Sp+YC~)lx}2Ai
zNs}o~I?==l!pZB;AcJvR8y3#UD4`6hr%-aBG0)0ad*m}%zKT%jH)y58Nkw(7HZdpX
zgx{`%fou^L<Ig;xQ51CVL<6%X|9I#9;b`WB7bDP2U`*u^cjn;#j`;VtI==i3xrc0%
zKuF!Z5B$*`Wzmbb$Lz3RIOD%E_4V}ys9NHKdJCre-1PFf+>j-l(}${$-ugez-QZ*_
zupU!k&u(v5lfxURQjF6!7X)X6Fl{vY7+|s9v1yaAlgExiL~%m14+n)#Lt-8$Po9(%
z5Q5Gix<=r<EwOTQtE^eh{aWKkuNapL?D=;r8@h3hXTzWvJXf7<=(DUpSZ+LPk!RA&
z06Q3ykTBYk9GXhg*p6v!Z5_@8_|RhMB(BjP#i$ao29+$2SGWLQi*|w1Sq53?ptiSi
zBXVRrx&=@gFsjOZyI)jwG2x}UywoD2Y8I!{-w<iYKZH75BXI|Zf|XVX^P(x0_-@Sx
z+jKtKl*COF4@FAQGUo2YiARyks9;WWOrSj6tv;!1Q9bz-5Ji*L0fT5>(g@(~f?(}M
zupE^*pj+MC+_6XRx55QS9ce{cMn=BTzlI-$Kwu^9JS;*RI|b3os|TuEQSwDacQ3_Q
zI~*us&LS(|jOjEqMBbu0xuH`J$ZeaEOl@hO9J>QC4jS@{wzGrMJqy74Z+fBe?X#%v
z!elp##o*_#G}V*kkWv=!7}N+>!;1qz=j5BaeN!2Bdjdy#l;j|km9aoS!vNgBgtjPD
z8eRUYQyfVcjG#NbOQ0~0xygSGj{)N!wpmt2ERL=cae_`wewXZL2VqTENCV6bfI~o0
z9beuaf6(m+Ior)*g@vk)AcD|7)gY50F0V#JM8x<poo}Aqr5GJ>Na7lOUKUIOG*Dk(
ze^7fvGmxijaolkpD60z~y&z_wJs&(E*3s8TEA~;lxVT($!-uqqGMjvIp$`}`=HS?Z
zh0Fn=dny|Y4}<%vo2aNzIfkq_KA~n3VkEqs2dLo<7;QX1gPa`27w!~p%c)kf^&#Hp
zgV`V~NWhT~qhG&%z0E(`+!G*|53l2qbG8^pyDZFKjWgX1)LV!>ZY$csc4Zp$l`_nQ
z*L;MJ$InP@><Ug1m1%y#VdG{Xsg50AQzL8eM<3}0UP^rO1o;Hm9b3qck}rPUcy8v_
z9ldrJTa_?v@pfPzK52p|*DX2<_G}tZ#1xbRYG?Lr<$e;`H|YchlS;G8X=%s7t=oL+
zc4BBF5fZiq7$7^K=%A@$U9#&jy4!T=7<opOHAllCQSg<J+!5#_9P@Ux2kg=fsYX~h
z_#HRWMaLC3eeC{gYrdOTkUhnqQncO`O_quzw2+k-Z_!;7jj)trVnuTfIS*6+?B1l?
z&V_HMn{jj~MtT87HM^>r?(aLNu)Z}5v@)Bjdg<i-cN&{Z1xIn4GDEc|2y%@;pQB`W
z18{`$8j(eFDq-bxC+2P5^|OE!Y4rHJCJ6JXjop=AB)x+@=SD#VRnCu{xk`Hb`&DKW
z&M(tKy+6V}j3tZIz!^3{awZ1;#Wp!2e5*PQ{Zh)FAL);a7qLCM(mkacRh8s9D8^a^
z_fS3xCJhE5&P^%Z2MO&>6arz3AzQM;E{`$non?crLXgI^gfub(g9)@EbtR#+mgnT+
zvIV`Mo@Y7Ko;w35E1(P12kC!Gz=&MGUK7H%44i<Q8@Fjo?Bh(JX9Q4O+$mUqNK{@_
z4{eEFQ3D&M&$Ewn064dRW44h(OR*ui_wP?^obEH<&8UX$K4YM)oMqWem>#Ml$2@$#
z7in^kp6)@x#R4jv-2&8cS;_y(u6)ffJm=F1O;{n(T$O7NSdmd3a1~L%^x!ida4TuI
z<<Fo?DI?;%$e72OBOpSMOaWWB-xc6^X4*w(k92<$YHghrK;_~^s%*|b^YtP^F<h$a
z1%y%}kj-n#x`2XJ&&>_oiSWF8{$jv!mXq^!BOxOCxTTPQD!)l8Uebn;fQ|u)^C&Ki
z(NNRS#L?a6GlY!Fw0^o?Elwi(bZcv|v$Hm<^nRRjx?U^kYhw9YE@+D3y<z$de%oeh
z4f;KGb>+E77DsofZ$Vj2Edhb+M7z5`y7HiaQsVkO(&T9#aUA(M<H}SA@UnCppgqcg
zO|!SDFjh^mj~}y!kofsErNhD_^ZJVbqo*UP{zDY+IGGzk<yl}Hb7OrFm+TrOX)_yh
z-B0tpuAbdcIpo07$1e^0OPYb{wQgnbnv9y!RjKNGWX{uuLdBLLuwV-9=7zr&uLC2v
z_l+3?B*bD5L)YX=a|IwPyv)>em~tTx&--dK*48wn3Mp?*tC$}SV}ESuw#=BFJL&<h
z1efORg(qilS;L*4x(-@aBf=h(0!a@8{QWr~QIpTZ%x>1X0*sjk7wEErTPU>u!5yp$
z2%~$V5U{Y%I)n4vV%Yuk>96b@+NXZmyhht40dv`K$EMM@rf0i~go)5~CpDGS(KHt>
z&>uzG<;?M_z)B$913m>}0s)|-6(xmI*=+KjVaYkjV0y-aSYFP$4(O?c%0mFmjZDj!
z<~Q%6P~)ch8XmC2wh!Z6@+jhrv2aa>R6au@xpiR<7`PNx$>1n)!^TvexEx1kXfviE
zB36ZiR^YJ$_H)D3&aX~fE-~zeZjf=gef|BIP2XrEg2leeFfcctqSNQH8uk_MfLzsH
zEjW0{c8+Pp(+kQsAv~ugR%om@6m0;>5|+dbZwm>M8cXOEljI(-cH*`aT*X#Sa8)0B
z%b&?k_kK`%5Ys8HY9Nj441aiXuqhNGI;*y7jgGT&9j*M(;)Imna?rC;DlDWZ!vOo&
z+`!XA$9wA^?K{ZLSOAQu0J<61ZJ+5h1YjFQCz)o_-%6q3U|lf(psa{`6eUnNbUzZm
z1ESz}B^KyTZmIx%2dDJ`3<|1K{@M3ZH)jKiH(U;Kag?7g7dr1RO&J-?FI?JXzegv_
zOZh<AQQlHI^oX5~KHHeu3%8>w0dhF*g+9>uhsJC`$?5aEYb@+EGzT_s7$CQ$*v}>i
zU(Uc&8OMusw|SGIS?u)6XSPGcN<P}|vp!I!wYvcytBL?kHf$Z8#vX@v#lDcfKBxM#
zGqz!w4JuUFsHscGZAH1E^_xs~c#Bl1e|`L6N<KTBQW{kAT47o^EM-peJLKeP`KG!S
z>9nTl8r2Q=ZgUE_`wRdv1k8&qf`RT-LEfSa5PnJ{;}m3X06wGCWR#?q+o0q$+4X@X
zO+DXwi76b42L=H#%cBVO^V4Nu(2=)3&0(qrVGB8AD(u(?AdmQcQyZ3I#^pSpso-<K
z;O!jax{PF=!lXTE+vNO$ydkdsO8VLZ-Uu`i7dssXkv8b$H~|QP`#EQzQAd0%`1Ry$
z2A)odlYaSJTVYBfN@N4)10r|R3qwee#YKCIfvexs(=(Qo7!Res!vDdcYiU{CH2Zg@
zIk>9ur6<G+XrogVG`;`|d3jMCq?{@F7w67G=Lw$Dx}iScVr<cMGoNmgLAHty58T+{
zT5SJr?-p*g#d==drb2Jo&`q~%>tgRvp*R0vhhdFWDdD}7ZH6_J?%*Y2>#1T$+nr5y
z3<h*-P&kAQ%f0pa&NH{Fh5lbZQo8tt^+uh~)*#Z(TMpA!mD4bk2W^l?w<f+}<+pYf
zVLoppqqjbX>Y*(@8B#liEZ<;k%ai_p@Yp>;1qB5l(3M&0x}vSqE(b+kQ6TLhgAzAg
zMY<h&zu;avxazB2Pe`CuZ!J19Dhe7NI8JH<<OfkP6qRg$Wz!-H?qvXeK-KGhl!UlA
z8=V|xy}iJeb;Bc3Dcw}qVmI$xzoJbn{QyZJ#jyZT35iMi!efh4pa{rAeSJ4ux2$X0
z+nhQlZ+po>?MYT9yWzbC8b{Ga8pp-ngKybZabF?~kH4jnaH9-t%dj2qFKYZ+g-*Gx
zeD%^ztr4Q4Cra0TmA-!O+J3`7$wrbgo->$L@2%RC^}dRSA)jG1t^NU@Mx{=BM}YF{
zjsPy!V&~r;1+ad6Xxkt3ErhFu0;MYrKZc$z66Mfa-QHdymx273lLCRsv9X7~E{^8!
zC5;jfI5r;0@3?fQ^G}8<4E*%n!G4mXfv?3{&dtt3;|DwNVZ7b}xK(Y2h4(C!?FaT^
z(|YH#XU`r^fZ<g36^5V2yV%iv5m79u7tvy+qb_6`6%<5{Q>A1dubeLy_Gi{SL-zd4
zh^&Z6Ik=4<3!x*BJ~A_ugun1{*#IS9;YiF6cn`w??oOaf0>tyYsLBkRmzQ^c0*uAu
z^9R-L0DDgB5vZoLi)i?Xs>8f!6oGLgzt1Bt1c6qy`>Iz^#PN1PECP^PP?O?xURzoC
z9jxYE{6P4PK=~cebV6HOT7Cl1{jLAVNq9%ZVvhKinVpBmio+Jj<f_bq|2owP%Dmt1
znZKqJ7gIV=6Gp#&eD!etf-0~X;gh=P^hr%zM@Mj(qEmNI!$Q@^zg@MYh7S95%CCMe
zkDM*l?0=iV9nB&?)^<k2`b$Z~kJ9v*xKEhwgJlmVARTVO_Ys;^vUJHUP5MliNGWJ)
zIzvaxblQaQ@N+9<R#sM#Q>8Z6*3Vi(%M{Nfv)W$?j<l2~*<S~zA~%+zI3O2Q-Ol^W
zU0f{NU5m)4ldK*ScMP4Cl~raBNn^WM)ZEF!`a#DV6Rb-a4mT!du~g0)6RdWj*joY6
zqI$XcD*KfyM_Mrf_GFAIa+9n}&?n9lImyi*xhCcQ1)b0+a65F+A^eWqoy(t&OZD14
zt0D7zX2XUYLYhGvW8j;rPZ$O|KX>7ear~z3Irq)f)zy>qCxaysX;RTb)$(5*jzc!J
z;VhA{UgK4E4i1u5Oc<CBSNjVRfxaUrzuR=cTO)G2=yf(k!b|wLqAr}8pBSGFZ3gtZ
z%ektms=98^!^5M#4?jQprmfSLQMClSVg3k+DznAesi~Jc^TCE~i2cTY4IeVLb#$n!
zs-8;b{HAZ+0G;4LD98mqcziohkH|YRNeS_d0gbSqAve&&@*xNa&@ae10IYm*lo*8G
z^wK(NW69)<fsTLUHZI5liWpkNT?rv0CkL4YAlxe;y4a*O?b@0u^581WP^Q6My5s1y
z7&&hAv%t4+^=saEt=^N)%*-SvCVt|~ld>VqZ5Ywa5PtXuou8DJ0JM><1C5e01pq~E
zMEQnnZv0m)@+JO!*v3#2aTav4ApvtR$}#-RE}~J5nAU?g<8}V=`)?Rgb_v}o*7Oiz
zWx37!fb~{liw$fPnW;&DtLO(;aQp13;S$4m3U7sdhTl~A4*V6iYmnEUkwNIKU=Z|E
z_$#rEl`^PUplL|?{AlZi1>r%|n@7NS5KZG}zhtmR@VEHA@!l9Rz7-?tBB@LF(ChO`
z%OSX?TT{tcfMN9LWIv#vq6$p-U=u;m@ZY5FHrOwY|AyLzplnV)BZDIG??w?Kjq_|R
z9Np<Vb0Zk979$$D@ohFqIQ18s>!$BS*44=47uzK)*vRSkVtWmV$Ivvip;dB|e}x$a
zfwXUJw}OYT4_hd|3i5ki$DK8As+z4y8Bj>{$VELSI|Tnnv$MK329P5*VcCGgFkN@h
zf!hZ&oMGUd6)V1<`B-2Xs3W?K(?~ODQehKq8Vbk%9(3Q<-*2k7t##Zg#mfE5^5Ovj
zJ$SR)7QdfS<{*??2Umr0<Wvdy=RBlQ?k=_3T>pE4zcbjSnMAXY7Th!-k7h~TJ|>31
zd)S2mmY{#`T$+J1nx>EApIfbq-NP8@+se^RhCED)Qg{1!HiUCqtT^ENmh%7DRbSy5
zCAQ^X3=RKnIx3$rq9?_5^RkW+%#!;AnkG`f(bA|d7_Q*f;cniyE#XES0lbRXI#o+V
zv8l|)ip$cO5p1Tv5jNsa*#Lgk7rV|>8g=@yun*>c?j@T#O5)|eh@dW(N)^^MbW=)~
zjx(Kqx3~bkKpagY9I&&?GNh<ZPV<Aoc`B-^A4f>o#Cp&I3+9FneE>WYeLUlI^kxs5
z<YJqD9XB`kV`15)@3KoY#KPz!@D_N-B?Gu_uX|^))f6*eN>W_7EC+9sE54chZf|Fo
zBbbLPIdx(1d-$IylFz|l6C#=UeZe>#;63w75=McXrN|J&)mdntOzSu`bb9^{7CP_2
zgPRPI2Le(0DWkYpOF#-M@q}!WCs-xK9xc0d>d#WIyYbJ6u2x9sl7E8x$KhA0Kf8Sv
zgZGs)>T|<C^7swOz^!<>>sDMuA4wMW@d~J7dyak@MR*h+O|d$^79A=^F7$%L!uCzV
zYquSB;-`bv$@w>d;fC@v{;PJ2|8_S{OQUH%vi+L`du180P(*v6{<{Px-a*s&^ZncR
zzatVnbTsN;E|dzKv{3NQNpEkEffr-o??ba#yp|aX|4%Dz(#G~S{M$-Zj-mA^&z`RN
zr=Ic^zGZ;b`TTENz?dlP^C9?u>OUix+T9JcV@YsRsIv{`_cZL;n2#O7om4r3lCS_-
z2cuAxPw=J4?*^fG6`wI(bSQLNNB^XA?HB@o-#8rl32;8PROm(Y>y=-I&N3fUu|*8W
zmdaGlaNsohX_T1%tYPCnVJf?rjuC$2T)8yCeP>+UBSZKLMv<TOOU(IAD~VVbaGiHF
zr;5?5z+ILm&7hLRlkgXYT*I?Fwi0i=H_djMfu9oF<z4nm7wc_$MuK3ya&OCl?n_NL
zCO4W!@+XZew<Tc;p?E)i$L25wZ+K}tdx+PFgqu>DZSei)Sw>ngj+%mIdK(zX53jJA
zpP&E0dp=?D(ZSl<TId3Cm}Da`fCH`7#RnpO7JpiggG*Gn>H{pndyw9-JwuA)hUV!t
z)7oWjvGMWM>qoaXQg34;amnBi>KwNOXra{^d9#?Eo!ydv5kSPPhlzF9R5Ud=Tt9FO
z0yUASCFW0r(f(^Mc>PbPkP0h-1T!O}z$^Du3XFj(yz241EPou>I6x-0a#ULox_9qh
zL19?fqb;Y&@)CvB(nWM!Ai9eJSTBJ7Ew@_%ZR%?6L$fdLdiz+M*feEjk7v!ak{ptI
z!=fveV+{0!C$_R19^i(xi8CVNDSvq40ld&~A5#i6p38X;@bcSSb3iJm$^a6abG|7g
zBvfDnX{$}cRE62kh3t8pk^Mlh4dxW`Ub{VHU!eJz+8rpI2{?s>US@9p7J9)BGgjb*
zR$f(KU*Fhx1cJGbD6{wbub6dGwB`ajTjLH<i1|y|FLFl~VgPq`*-s-GRI%xWF>o*7
z1HKqsG*GC8(a=rcT<0-`IZj<C=;4-oh8+pOVxTl?(|@^$);Na00x}E<U%Z0Z+1XE@
zJ{{oE70!x}#U9lQtUIXDjM{Fyq41ALDhCH7oV!_7R#sLs23-w3IGEIOF12VQ)PS5C
zNM)y+O^lE8&sCQ2GYx*fdYCAPXk8c#-7T<jZxd3YZ}!lW#A}|I|7D1+XdxvoJ1Z+{
z%Fi|Q>Jwp-h;-i~QONLVoeKaP))@w8Dr9fcB&nOXqD3MwmKH-73p8Ec+0Q~)ALI>T
z^Xeb&OwB`8Qm@^@X5a?+!(47gBSskQmUW(17XRHaZE1ONq!<X^928CpB#x8CtNgOY
z#>NH4(Dnk{gGsql{H;&}R^l!q1JwrQ-<_q&HCJAE-0GnRB0|mj>VmOJ<HGlkZQc!v
zolQ-oPW*h{buwV1=jP|(YgZ!QmP4VUM!%075(1t|^?`RZm;gn<j96}7f7}t6J97Qu
zi@#vkd^}1HYfgIV_2YxPOY8KOMKLqN!OX)B(gSF2P25npzw;9OMm|LMmDD_vZrH~}
z`TW@U%j$M`JJi_N*s4Dl6mhF{y|@rkF*l)C^fz0GT;<mB!`-1n)tX@;i5VCenDhSe
zej+Fu4jr-d1x}cY?(p!i#cF<0<CcPOMTaMWN)t2Fsx={x>F@97zXjpqWhY13f=@DQ
zb6if(1q>S^)QCP!<b^_u?+=RGauyhh|7P3*beaA$x$UGApstP|`XI<9Z)$69ZeH)K
ztCMW`Wy1vMVzp}I!I>siyCcf`7r7JizP*Dpo^Wr4(MihbQR{{7?(S2aA5va#UENu4
zPDGW;Q<G*{k-&THx`s}SjE^TJCb}yIO|bF|fl4r+Zv^70MX)(Okl4I>bS_s+z8t0E
z2B$}Vx>+Rd6o;uB$?yH`X}Y)W=>K8_Sdbq18oV9mzzeNS?WBwS@BoP%<*C_0EWRFW
zk(J_W8c2oIiu-`ev$9vG?aNQw=%6$J;_f1;+0Y5_-y<R6h=!tzN@>vBH!Z_RZe(g2
za#{Pv{N118NaSzFeu+iAqtX&5BalI$Y-{y2Ha0$py^yf;ynt{-&kwO&klMbQ@ZnB8
zU<kZxLUWk@&tHJ)rVAyoQKAfioIi&_Tc@v_vBVH4x4J6;!aEibd4y(P1#v?XA^6#C
zvY!b181en*2M00+%JmUExcU=DbdJ#AIUvzi205X@Gld|!|7ZKO6=Hxmjn!kTLI{><
ztHKuhm~eK9!s#TQ3cZ9AXvc{5Ap!lC`9$`9$GRg*Tg8yTJCJ}yitP?A(+pJbQ)VHU
zav&1~*_L2vb`A@qan~psbRv!X1h$_Iru>{>%27B>7?e{*+Yy{`Y?c86QT><gI2|bb
z5ycWoFz>`)&@@AY^M)oduagJ{RPYw|k#juG{~u?41TE|{t41*Gs&oTyZ^EZUAUpkf
zj9{nfzhFddL;uT8pVPudxDg836iTAXieT+ccM$NhkCk8s!>I-`deI&P=eA2M(+vJ#
z2=r|vm`yxj3#~|;xFhz55x6MhDY9wgg5B=NEzK{vlu9%MklC>C{VP9{dou56;Kvl#
z4^lm7siC+t!=t0T&<;mTa|JAe%u>}8VaT!iG}LM@RMJP9H9Yj}X33I{<aNx*&wooF
z5vI)M?h?Ax0tgkrli_rC5fAF)nR!b?hp9cnw5R70kLgPa6~t+UA2l>Ji=lvSV(ISd
z^P*o$M!yw1SJdBMSx7y8;Z+FlqVxjrn7bSGj|6kggwCfYcDJ;?S`7|f=BcN?Amwz_
z2~wm$(#?8i)2X?<RQxg7UxAIuwtsBD6i#5J>r}U5$7qXwXQ5u3ui`|4GO3fqi61{w
zvkPa-*KMbq*Nuf2<^-~t)+ba$*H;tkS6j-9*Xq~TGZ|Ja)a2sgBtv*@m&dYQPiKT(
z)C1PD<O%Q#t%am(Pz*|wP;FDn`DZTdN6~znaaq`#%!~|(Bag<0d49rdp}V5+4^jlR
z5<7;*V`R{n2#*&bKw?Bgb8Bm^QB&LkV>q1(6ep5=bXr@nh36x$<1L1|ud=X;Lg0cw
zx&e<r&1c7Eq9qx`sde$ZsQ30yL|kHa_78Y(3nL>V0D@5HdtXXQ>Vxa2crm~<C)zUO
z!4o~9AqnUraKP8t-*D}J=6GUcbhI5Z_p-8$P_pCp%qoBH;hyHUHgbg_=wWupGyQBK
z(RsfDz{8Oi&fpVe>$Ah#il7M6CHQ8NhzoHGbUaD-PmYiO3Y>>23Kc+U@1H1x*WNO%
z-+kto!|k{<JKb0P@fW=V@D8L2wVeA}A|kFIQca^%yDH(8!<}oDkdk|?|2r~FB^5d)
zwW@RN8@ymTp!mQ$Q%Lr%hc_pwzg_hll~8tcER-(90k}^p!o0rLmg4c*D73JyK9HS$
z^c}I66_0wNayrf1cknvQinS5wC(^tg&NQ)CwC^*#?2po6@<Yy)Ezq_<dNg%(bQBgA
zLP4A3e5-!X7$i1_A6RC@#?sClxAZh>DR&XTH8+=iJw;CLHrBOHreoc__@f2(s@ws}
zaLqrL*qssZH!(K0hj-BSxI@}E=YIF&M`@<fkwDrgdqgha<nYd3N9V-HODQMsA=%CR
zt1>6&AqQDjTi4?u(7E0T?>%Dd%zrrby~h1o+|+{%@K5+JknR5<8Nfo_&JJ6a62@Av
zp9knIkQI6BxbRmCESE$ZDF>z)gR1r)x%8vSJ?cr*cRD1c{|EwFcqepC3(be=9e+wW
zu?jtF3Sd8xobJo7K`FHp&Sp|l_243WKPeJ$3Nq5<lxZMsYb1sS2G4&$w~CI*ep<IN
zXu1WS>07R7dSrV^w5RTW2Ipcq^iQ(2;&ym_Z?}fO7P8E<HERcn(?G`5)w+;(1T%Tc
znU;FIFOjOvMLqX9A{a0xzd*&^$<jp~P;$3BdgA^B@Jor!Lf}PAE;Umacop;%)DhwL
zG{?xy%;Q{uae{{3P$aJq<A`-GDKiB{mDN}Z6p(z{uaKbBt$sGPof{hVAaCX#U~vG#
z$>F!*;KS+G*@N7nFO>b5pYOM=0Lh0jz7OTV`^f5{k8p`5P;b1Qo$q~djR*ItG|UNJ
zA9g_P4y2#WAHfTLtM4mkR~OgS)q#r!waehtv8>YpQ^E*jW@s3mHud}yZ2u{<Tp$4i
zed9*5HXers49eY=LMEOxyfrI2nKd5}Y__(VsYO{<GR}EK4sOxkL=>F&&oZR|hxwy!
zN7COVffNTXtyy-dpQRN>8c3Kwn1y%Hl&J!?^_>x>Qau+_!&U=)*^?iyy~a;A6-sYD
zJ^H`cd(Wt-)}>t(Wr>PnKoLa=20%a>$&yth=d8p9NS4&(&>)H+ARtL3OOVtyt<XpY
z11K39XppQ(Y@+1Msdnvei|#Y_8TZHiamPK5wT3j#p7Wh=sPI(PQ$1C_JNFL(;R8#H
zzxbV^b^@dTq%Yf8Tk~Y;deF+sOF{bR<K8szzcyfUe|-A>IsEfkh@3y4^n>bjq`o;2
zlLt87UMB5Zz5Kg!=duAV+e14}BnBT61-~|+p)`poEFuEMFZ43lH*fH;NrQVTQC^#B
zr7f}k=S1DF+P(OYc?CQfj}e*$!$3FEM81KpaXd#fRa@^L(l2-CM*L0v#zll+00#gX
z5z)=fWxDxg@Idg6&1p}VIBOnwGoO0F4z~WRhWj~7ATE#XC(Z5J%7?i246}Y^gMWD<
z-;0|Pg$i1vqG99GWDN`q6d_^zr|O62VzJ>1F*ivCXL<eg;K?AxBuB{$LC(CU3`Buj
zK_2F@y#H=&X_dg#$1l)G==*gu-&r@;I2Be>TiB2Mz>R1ZehZ{ty*D2tQ;%<REDUle
zdhPvNUvfiLt}pxrO^~~xxLkC>f>J{N^S)&Fp&<PoyV2ggM;SzKKl+#~6wsapH1<b4
ze$K$v4~{PY9@wwX#B1}tjx~i(|Dvth<nn7;X|{W|LyzG+kUW!teUa+Mq$<iy=yDXg
z0u(r@B_%>#PLOJfxT2w>6FWJGndkcZ%)t=4W#g%(E~9XAXc!C&G-oSNgUfdDa<{Rv
z%E`$I`dIk)`v;9U4cL^-IYmW9Y%-{3o!P0W@}2Ds-n;cnQOGMki@g7MCR~}IrjRa2
zHej#pKewP?&e-_#9Ft;2k_JXb5?}Ym*{^MnN$gk^OZTL8P8;6?w8F933OpQ#N|+VR
z4g$di%4BMqK$197GI-oBb+$_8B0u=b<$Yzu7o)Sw-*Ugk9x!XxbX=00bPa|iDUEa4
za~sgf3*`69?AIi515!Pp$>;qDFKxJ9fw|I-*V(gYzgO%N3@FX`xd(=%SfPN_9YTrX
zZV)O|Oc57jc+{Z3qTM;oGYedLKOS4IlKS4SAf862xz6;RDOvy-d;0J#Ns!y%Q8-@g
zv|zundPgLg_oqiT#%5)?Ve?V^$v(o}=0}rGREtpa!auMS?z{Q%`kf*TwcCNv5zJ44
zrVQQ?d3kCtsaaleQwdC9N}b-a#R|zW3a1o#45`!ps9Wy2;4n50H$k@jnSpQp@Xp4s
zva&#-%>+F>)j552H9`zHx75oW;pHE0;uSlmBV$XX2S89R*ncW`Nd#$it_R$z>4!5w
zK-3)$F1YQejVwv9uGx7nAY|-F+sN(R_b1Q4z(FY3NgO-mZq|<|0yKuVwrLf7t`WG$
zE6TGwfT?%%^@#$AG-YOOV*~nVUU8tqYz9lUakY~`SZd^LdBP}B+0rb`z!d2sLi6Z-
zo$Chkoent4X3l_dfI1T3=YPbQmNa7brM;Nrs41xoX12Rv3?4;MT{>;>_w|394?lm%
z$u25#gWM}ro(l(<7NO42W{D*E-D*DfyMGS_%S4a`EDrXYP%&0H6omrzoRyc?fBr^d
z_5IPj@h&Kbc=819B~9vRPT}D*ZDUk6{k9jTuS5h4Cy~K5<>C>q!Hwki5troJ9=`?2
z$J7F}Gk5JKz@J}h@nx{{21_e6oOF*o$P;gj=SR1M%0)sKJ7Q7Q7Ul<hPyI*P+Gql_
zbnXGqTJpSHWX6RR__yg`jy!jW=~F#|F9+-E=ycIm&OOqzE2&YHjl7hzYTyf<tMr|e
zyu8X%pTb?ZNto0SZ7A^IG+L6bc;1g};KZkq8W)BJknYh5^*Pd^bTpjn6Dq#z9XT4G
zZo$jekWzxWyU0<{{xUR(be*b~*9IJ`wR<50&h1J8QgD&fI1u_b2aJ=(1(|*8fTT!E
z!|yNlfH^`g&U<WY#;vLhs_EtefduXFnvf0@CIIabAc<!TVeUt)0N3l@F9mEqZ)Rp@
zxKI-c5^HrT0n7f-iL45^#y)DmVfg3L3FyA<NRdV=M-vei{t!xPunoiJ|4R`7s1<0r
z<>@VODhA#-8KAR7l&Uk(vHn<IUtf2BvKX0c3nF-Mb5xMG<x?6Au)cj3T3{ZghcOdL
zyav5ohUKd<Snm-TiFG_vN>*9f1~e*Mun=E}gHUodygriV)oP9kie2(E={DUfK-w`{
z&w<^9%QWUZ;E~=RsRXN8ONhDX%5?(v=Xd7aU{{$jegbL=jb#$%@$C}`j#Gcx&RKW|
z*rO)(K))3)b?V*GgduqVShEhC1c`p*P$T%o!VmO=z4!OCf_)B>+hbO?wxO|96|l>c
z0T{J2RwK>cEQ#St{POxr)lS6@4@7njyOA<XP)|xh<IQ<(PJ)aC3ovO*4cxxSOKiT-
zT@|(b=o#dseesQZ(dsaZ^;tqso&7HOjZp}Bsf!nAc;q7I7e7H`VaAiF;RqQd=NYa#
z9IV)C7i{(IDGLcx(vzu5NXu~Mv7uYP%Z=wH&k_Nn)t?HK*!%{KN4=*Dy8<M9fpVZ}
zVfYF=Y+QZ7d^rr$d6%;!w%A%eK%LWRe(MgJlqu3@(#Wh1Z2buLZOop}Mdw-D+BQ{f
z&H`zb_$kyRrsy_la$<X+K3Za>q{U2g4VJp}t7K1auYKu6X7MJp3@iXEbF<;+-~)=_
zu#qNVtR;o3_!8;nIht=sul4gjq~Top_0fd%GCx1U1%Lb(+`2=4pWOgMX=ZoV$r)*8
zY)ksd-o%=chx_U^rmIY(A4Nqeximc-%$T@ttC(1unL1N)X}Fj;|NM=dt+9m}C6|uX
zU1v)sp{rLZxn#|(EG(Uwu91GeY31ysYKD}ww{@_$GqZ!=QgTV!-?K+*I2fD4{cf4L
zTA7;3AdNjJxfIRpEZ`o3JlstG{GsHMwz}tRhGgQBzGv)gcFWAv{;nCNxH#p%2I}53
z{eBM>85!A!7uz}ZM+9HMkN@}2|DA#V@6Ev17Bcf<R_y=o5OVYEI)wi^BE0_(j);re
zd*lbEljRFE=RTfay?;MLu&}4c>a%8vi)ZZ@P2ThyO2W;HlOcO-GYKaeX69Y@yyG~$
zx`&F2s6_3j5zn6Ydae4VaolCfpz2ZbDyhBKJOouL4nD-0nyRbiy*%4iDjoi?@yXbw
zQR)kMmxEqDxSY@u(DFTY&jf2kTv&O0)2U{mq$b*|k^?uR7K8;J3z0of%5`Xs_WRJl
z*>f?T>^`kXtK3+DSM*+7#u4-huXbC;LA^!KH(O~=>QiY$%AGqF#)@iGdwG-8>mF4t
zW$2^m545W}mRzg4_~!3pXCiMp<C3}1nU<&qldg_rk!MXk=>{C#Nd^8WtD`!D6rbbO
zJ{3=LSjMkXo71;V=Ih!9a|9XR^QkR*l-@8~&Stw*ByCpqgR@c4g{jLUJ85wI-emh-
zw^kH?#nKyzTKVYz?UG#Gy(Ir{%0+dVn;>bGjGb(lc(}RwVI!#j5|dYX_<4TH$^WyM
zoT@nJuBr9*^;S`j&(>1Kbj4P`%}T|<Tl-#H^z;|@=7qj%Ekpi~(^@IB5A&oy;Ne&h
zJrQvI`Qc+s$Aqsd-V{Fn+%L|b+y5`w`~GUHT{&8X1sWQZva(w4gBD_<a)g|yp^iEs
z`{e-}j7HXfIehH3O-=L+d!f1>${|1CtlEc$hM-{%^m-C<_KsF%2G!0dQwb!#&6`zV
z^SHbTf#i1S88vzdpYrY&2z0g-4Yt1)6Z*-=rlnKYy#z{9SVAJD;lTAI3n(4h8Jt+)
z+L`A9dug3yM?u%du=9*&m<6D6RZ0@(KRo!wcW1>nL2`@KpV=<a3fM3<i+nc(*BI-n
z02DYlmQJ@>y&Ar=dbA(~+9Un&-Vmc2i_#G1N13XK{`jlQmFzE;YrdP`8AoQ_j$C9k
z<0N4(5WU=jHq13rEmXjv`D#1db7h9US*Ec5#Ki~tI2Iq3kDJ2~&lOp!2xXK=ZV(~c
ztHDe!;yNrlua@iuWn1E#a~}^Oimwj{Aw2D_l56YeJJtY#_hBJBz}@6lT8Q@dS*zzM
ziO^g3VP%og&UzQLY96IL>acx45u&xd>*;{Jg2QtUs+k~Ghy(*|)$lBIx~_RYoG$tq
zDh?BhLPL_tffN`4-p<zSEy%y0R-^h+ePBN=re?5xPD02C`f8)Rz!{>}algMH^EY5F
zr|Qwr?XBsrU}U#hEQuTMd!0;BNY53({lWqv|6Nw%^ML0hr&tfTXiF**TPNSj@Wbn#
z@>Th`GprBbON-AkZ-5BdvFFzTfoPrVxdv=;VXTbI2wj~!MRrNYAL^bviTOmDqs8)y
zVey^Kh4qvQKqlAd|GNIScU@y6&PWfUZ~yuVxWz#XL<K}i##`;+<Q9adk0BC*xfa`1
zK<mFrx0mT^B?Q#SM+>Xgn(VJ7dq)>rCS|;SJx5!$`iOxD7!leo@$~Vrw6Sl@E07ng
zdtK2T48K(+(zDh8WNWk`nR0b5A8j5!dGf@!^b261;v_0*;QlKI)=w~@t)VCj%Ft*m
z#~}YZB34F9$^}TO-ueRaIOw~B^1z6mmXD|aPTMVfNlD4*ds4)9vP)%oVK_=f09dS-
zMIn9%l!NVJcJmbQH+YuI175RY(`kT!k$NJR7pWTh2n4=-D_&3Ba=r(I1Mr}=wXu|{
z#Wsccl5zl#p~xW~;&G0SZw99u8XA_yp?V=<<%;1Cl@O-cZ~8v2Cf7zefvQWZWT8Mb
zm*3HljpR87dV-5yd3Jbs`1y|2+=mG;O|R-@W@YW)DM=8t7dV^1AL<uIoJc<HWAHN>
z`Nt2a`i+?UNMacUf)EJSa{+y4Y3Jw8@mZ@RQK69se{VN{-#D5_bw9-OJngZK5n#Jp
z?D@L0nX+RVd!-qeraX!RpuyIcl}8Yk-|C<Bq0Oir0NC=uAbn%#Xl8<)@Cf_D{QLck
z`hX->dPP;0jkumL89Tm~nue_ck`(faB6MCUI|~cTs&ez{>y9CrI~6Hm4$ytIyM?o;
zkJ?g#R^Q=0Q1&qG|2+8!c@&dB9y%yD?6lLJq@6q8%}9v9hFx^~yeY6R{)I1~g^l5s
zK!5zSg6iD0Rv(Gwl;BEzV0|-hfN%!!;Xf~5i4gXfA8tUDHO?Ly`2kdE{%g!d%E^iI
zMd96F!QvnO`L=lIFTBtLJl&$JOaHVCY_j*pngR}t*`i)+SOG-h(97BZKs^|5G1<~W
z>3Jn=YV?&06;|z;Uw4ew90Al+xjZ{vGM0RZr^A6F@MWh2X}$d+9XK$bk-H5j<x(m_
zI_oWHho&n_!>pAOb+Yv7C7K$@m0DF21e)$SHSu@-)!ob6B5GQd0%cJ9dyv4wX3o2o
zP$O!10ui5_ybZvl_S(GLfXCPs5k|tlR_)>j0qgVftiCf|s)KIdJE@9dU?q9Ll~IJ1
z@Ix%Yk>%+HVvoM^>N$|B>v$Q)3SigRp5581s@!`=RrOe()JU_XLT%MnZxspu1a~J&
z@SgSsww)!TotUH~`f-Uer!i%(=ufYt!PEUb?7OnFJ-D;rNd!U^35a3D#Ez&QQMl$E
zP1u<52IqeH*}QH<D3E>;fi|lQSi@}PT1#r7H-(-M!oj$$G_;VeTp#yVyiC<@h2XAf
zGgMA(2;cj9V(}()_5%^_G{~*UHH_mkOs#Du-|7PoKfVcEU9_tHB2M}62H!&NH@444
z3~MXCJ8pyI*BrjSdtEWyNr-A^a5=g_cZAF#jOz)+LyC-gA#%ag#aQAU$4NpOyNjXR
z4ZO&TKKkkt&flTdtta6ivW_PtM7iV{@?}OwA3y`k0zh-@=|Ec;lL-N_<y(>7{cmmc
zyb?ei?4w(GEpzEF{OOH~gY%fK=PH2HKQMYGDRjknUQPRZR#ZfUvG3)fm+-wp!E1nr
z1BPyERRSR=lm(e&i$N}?C9l30wUoR7C%Dd7FwX~OYk%E~*mP?j%NOD?Ljs0XG5HJE
zoys7x^J1_sEk4@?OFBqU@K+E^MJj$Z{up40`3555<3gmKj*7BZM+HckzT!O&YG!~Z
zBO{7Jhv0l3KyGozu0sr{`|KAR{K%|X!OOm-fWv2oR%95(y;oru7rB*zW6W;PSG8Kh
zm8p_4;}!^dy2I5kcu>pOaHaLS>BXTppaNpLd%1kq6iQ-~MvM<eh3C^rK-T8^a|=+h
z=Z2eYd=YhZbrzZD3Rk<eJj$>GPL(Xp92-Xi&)OBAuYLBAf{I6v#IEmCL~8s#Yx#We
z)f-SUA&Ao2CvUq|2v0+{H=y-7V8UB=K&@lcqnE7Ym`b-LhC#DtRp@>{>bpuVXh){+
z3}v=ZH2fpJBGN`}uIf<?$gcf8(?F0q%?SZZx=mYBl8yOyfN^>RXL~25FABbA@+ug+
zk0_hs!TQmykxD#{ptiiB2Bqm!{#e03OE#Oe0IdfSEPz7(!PHoU3<I|wkS%W=j!t)#
zDEnHpVJ$HRyEA{s;Q-(8%i*!H-u7YuM(6rR!$fATP8v6U0(YHC{3K|Klq7Q6Bt9$k
z7o4y*(~b01C-#4v^U{SC5{gxn<2D8|pT4p^#MdD{5R#(b#6z(s$OVb*)$b$;GfS7M
zz*W&<6GoU_8Y7u94>c+CSw9RjsF@|b?C8+k5Cc*{(b~|_NCia1q{q_=sL-T1^8jGc
zM1yv&)ja<P4;~l+CFCeGu!vZA-~KS1{%Sx)#s}wN4JkP{mJ2{gg2E#Kxng2lP+x5E
z;#}cI_r9uroed$Q6Js$I&fg}nRYxk{h#5?ZgmLa4JemJAoN(9+`prjGmho&!KZ2N6
z7cPW?ah?Xs8@0{$>R@(~JHJJ>Z-=P+<dnFPcNo)(kq@^Ju1_2`>`Gyhqx`V5*|PJT
z6IiiM)F=3Udqmr)K(dIK1ml5^DzS4w7&=)<5TJ!adj(Z@yPl!d*DRNR4Ll3z4o7pq
z=*h45<ncQoC1kx)aLA_&yG+_$=&J1{ta%D4E4cE_W?Eb*kar)#S^(|t)Q>qhq!V2`
z>)0Jnpzp=v?=5U_42<ws>z+X9j;NA&MH)W=N3*8;_CI6qtyaEEPNu&Izzxv(AIkUL
zSZ{~Q@#_f?@qRyG`xxB=XU4>E{vG+zze@0{24oXbaLAgL0XT}?icSh(w8Tf#LS(8a
zP`r8SwSS$=o#~Wjz?7%yNfcDHrYBo|(65Pl(PH{r5Vg}V_%%>p_gL?|F_D@J9oexI
zAaN>IB$LX@7TJWZ%-cdZaPL+C3pAms)5<wWOND#h_c!*;+!ppx9dpr<l*AOD^&AiA
z$<zch{xuDGJWApoKt)DoIbs{@yw)Ar*AqAa9UFj7u^W;e3s6q}x`BIN34RJf%k>pX
ziPyMQzloQf93aJrF8?)Uc)s!>*?ER%SJy71x7yh_I5s{qZhze{za>9%suLy{G9Wc$
zLpo<HXop>4mw1E5g2>>T!Mm<iwRi6JEqyq|wYaWg;8;VmoNG`f5~)MjyV2q|TeY)&
zt~h(p`Rv&>y*k7iO>%6Xus5<tYWpmo<-3ZrFOHOr-v<tw4ax1VN#HO*9D|ygS{cpq
z1hsYN?Y%oMnxJCHx@+gK;~3;M6%LkL9Upf|k61Y-??EKl=;k`})C<K^%U8xaykpZI
zx&PcWK`Ju$-WTO&g<fugASBa%R$1Oy_L%fei87|5jreQ11Rp~W`gfa~o9sp=STru-
zwryWA-ll@Y!LE3|HITuxb|5Hcd7|jPx>B)aOuwv6trU7CQZG=Y*}iJ4vwRbf;?V&r
z)j)$q=R;g6orT=8Iu}cp<C}117ME+4*lTY^mf%U2N9t5lr4FSzrc!L$-Fy=OQPF~r
z)GUl<O5Va*`ux^V$@;a31Kk9oj)9TU*BqfcnPTv?7o~KLU@d{Y%bF!3eu|dX;%#gd
zOXGLQujLQ7?|rMp$QPpK`y<5Y`RHB<=~7j`e$Aqft3N+Thr|;>@qLSOYJAQ;IwCmW
zesKka<zi!A=nYE8FALv&*BXp&_N~E7P6^IxRrK{)=7AnnT{ib>Aer$|ir=()hjlFc
zyWZV<9LTIFULU{ed>G^@5Rtio-B9j1;vRD$+P3++ko8u<A!HdgC|!4YEJikc>pSJ4
zLul_h@MpHcN+1-S90slw?9puq4+tCuDuSLVZ%9>Ke^D`|8b*!Ko$l(D6%`7kI&1EB
z>DvHUm)aYJ+(JU#cKgo^HA6xD&A9Xx`kvf*LnmPGT(5W+%%ydwH>1b?OyRa#XLZ(h
z1vNE1&^`j<Y^jrsvm)dYCfH>#zOFayR`?w?(OcyZea_wze9-^&@&O30=6lpZz)Pis
zC>ifl2FjCH4^lo`i`*Zv6*%CfXxA|M{vpYMtXWBV>ky%Fn5B3ph2HA9t*`H0`jIP(
zJ@#=4xCRgVEbs1U&Da8AvY+<B9<59u!)y_x#RRFa9$-~TZmg|+suwQj!X4sXfo#30
z7S^*h^`LLYW+=O}+QV#z!Go}N*k3F+(;yEzj%^|?aB^_)mZ*X0mL+bI8^2%o)ML^R
z#^5LpxlQjU`<g~@x5Xu#%*xCR$X_6tJh7YGD^*94=|6h!wuV4XMcgzg&bz@@vh+NQ
zBsUhCq-=)T4?Y5!P?!9X2B<idb$&`pJ0*e-$*L*^1GPyUJVu%!MlSO+ub&M~;nM?s
zyx#Alb*FNP&s?B2a8|Rj0wxY8^i*f?&lxc=!qu~=lHCtU?q2%rN`Zam>qo-)AD%pq
zv*&|&|KUnB#IbwR9Uy=7i1QimnUb-a9`@uEhuuN1fwRiC{XN0+gd;*XG^H6795CK8
zk8(YuF6TxgLBH>}CZ`KVc7Q<HbTyj(DNsoLSGvc5NnO&LKz>y8Z|ZLV16cR$uiaSb
zX-oEku!AkQh-<Fk5*J(AmRJ@iCHijk4nE*>*-NI6)6EM7+QgFn+jk+M`TdG0uyAc8
zaov8+1iG?KflCZZPo7o)uS91aa-y(J&hGtki@dDtf$aeh_tQDwIx%`OrQnEXnM$~T
zJFIobuHi|-LCAo{Y(*PHY)I@ZNNiP>|7Ht#-3Sn8AOKz7LN1tBK@k+}NIR9mkevp$
zRHH$NO<A~VN{Vtj;A@42l;XI~%VtTCy_0Gg?YVqp4%kyW?`?u>xR3EuIhg2;pGc|<
zHCTd5KMsnE#?pSiM`CM2#gAoNrG$seuKb5|!Fvc0oG6dW7wadfP+(}5`Z-<=Lqogy
zSNlpH?dvAM8H6~9(Z166(_h|vtTb3B^pLjV;0$^z5gW8%4KxJiDWvS<$Jl(1jDFvp
zt+%o3KG#_=b>7FzTTEjRboD`gA=0gNZ1$rM4k50}J<ihlPDkc|Z!g`3;N^f1FR+nt
zGw-jk<8w9reBD5_Tz?T*nU_ThNL{ue79H2LJz}(7Gu!wT@`$D@#MZ~t=LI&mxu(ut
zhA|snYxpSv`F(_`==iR-7$WPKZ{!!i_YBISLb_BSx2bCD;og|@n3|S6d7IpyCM0@u
z(|hA1Wt-=_5TMuVL{mQ~uYkrygPckwtO-oW+an~L<DJySk>O|w<ICfJ!G2AESbr2N
zDf4S%ixko)5ymMWtt$W>5`*r08Ea^5|6SzCM3+<(d*<-)UHUUo3gZwioE~|66G;5_
zR#ZUjr+q+udzn$#IorJh((9b>USB4)@b=<AI^UbiLT!Kp;}>uMUMuuP01OIypl8gF
zb0s63ck-)W2`!4;z<xP%stUqqhNvY1WE%5XchsW2HEdC?S_3oa71CnJzux~$@}3d`
zXVcX@Su_iN@?zObgF6Om-x+t@1!VZEPmt4wMn{nt4Mz(hL7kao{NhIQLbLP@hRcUr
zFEQ!E^i_sSEPpZeqtxjUjJL@v`DEk$>54*u74SMdHBNzwbVz*Zwzo;<$SWR73qj0#
z>JMC3R&Ew@3W(pnj9zJv5Q(>U+(_Bk9J_Gdb-%|)u>5A+G6@PyR720c&2>LAQ6hO|
zwNj*BW3D2WtuKxlEdd3mdxaA=b0Gkt?jE*hMDMUfr~k8Y>g(28C66Y(H<BcYjUP){
zz-XJyceHvRtv-`4(|$D>Y`k()g5c5oB5>=ks_r?H0E!I)!#3xaxm6St(4Rh1ANeUS
z*t%}s^7>}^)Bpu}?|$1NBoyzbL6S>y+CN&*)=yW`g<DgVWG5M3`h2xor)~pBdydj1
z%THh%kO&L=Gv!aMF3a#olP2+Tl*W;W%_d9Xy-r;q-dp@a*da5%1U%nHPZumV8lWv)
z{i&;!f^;QR9{uME(X!wg2tqpcx|f2ovJWvQYicLH31XS2LF|i)i6RYfrQ-B2i;Ih4
zZCnRRL!T4DWRFfxX7FZ(49R)jJQhGcS<Z=<9kX8{xRB;=<o<K>+p+q{ph5l=9}h?k
zn<~uA#O9G;%zU)rcux>OIa<l5i!E;@Cc4($Su2Px6Cg`8t=2M!+zxPRp>7%ltdAgR
zmF4Ayk;Af=u}fa*5oO;X#fAbnXGDNr<qK4SZwj0_{Ahk2*-Ze`6R09Pf!7?ILf`;`
zys?V<3sT;~hCmcYVw*!=1?Y)QCZCk*DH9$R9xj0~m<AR|pLNIz0PHIF-iZ^<DF!a|
z1ftZD=r!BOk2E(dCajW99}8d|BLNcuhLW2XX~~!ruVHQ+B|88Pdk|Z(GrSw}`q#eK
zHeHDJj~uGnC&|bVln&yp!2Z2(sL-xo_{f1o@=EBE<kV?qZjQW8Mn=2unmy3J3d9ue
z<@~%aTbh)<ITE8oWLqIA<bN(`u>x8kIWA%MIi~%8Sw{nvjT4oMlG4cipLb8z(A2z3
znGiUT!0`3(>O@wGUFigiST~$Kts@6Oosuy%CJMD{_VrYF3BuPO!+|EoEhh|aN^c(+
z`gseodf54SRoY-6e7C%=<Nys^2BG1ZXMA0X#JOW2w)eby4aBaTC||G{Fh*6$<a-Ga
zzeXu87==-g-9ms*r#I{;dg#9OB>{nD3R)8QYk7T%)G}v3gZuU7JV>V~019C8y=4FY
z_#*&j(J)iuhH10$nC94Pa-Gx4E@XfYR|Cesa*GN2#?5a-QNzU_O|&JCtKEE{->1;j
z@%7P*i<-1Ry@3Rlhy$pF2(K?voIs0oGF9hv$xZ1Y5y&#G(l-W=pElKvc}=W_gRJ@M
zY%$2W@&k9Fbbo$(q)UQ&a1%PZxJ`XI9yF^z<B0r1N}&w0H&k8%n_t#@3;_3Zj|e&c
z?MO<b$4UO+)`sukf?xQNM0@gNEsb>EQy7Nt`UGqFUNYJmDMvN^lmGhb&yNnj{t2!C
z*yZ0>l0N=<^?$t=Ens!Kp7iI(|GfL}4|7x_+x_hS_^5yK|1g^0Z})$Al^4pTmKgoK
z-^P2b#lX4hqhZPyUpvIx;>(EOkfF4nf9*{iM3f4;fBomv&tFoK?Y>|rjchC-q%$$%
zUtjp=5^xK`Oj5GhF8;OxUq#*26l6Fy3o<e;SU1c5HUa=U^*c_fVY3a{e*4ZJZ%(@}
z+m_bltd-f$m3e>P-)~PwE7o9i-4~&I@>==J|5^2)>js{beu*VBo0#dJ-&kr>Got0>
zTsp#avCwJyj}h(u2y1wuSxT&Z(GXGk*=RMUts>Rv-tG%v3visx!G_Enr1Np+i{@6}
zy`*~_@b83m&&7wV%Ud~KHqF>QgL}WGZR<PRYaw(krLJFGdJ-|S;wAFie)^9&IB<e?
zX*_#kKQW<9)BWp<l(~8J!7E;WJn&ecv{-ujfMmLl6yN;_3O(_EFDhw`$PjdzGSjK}
zG*$f$9eECmZk>!T-*>O_{SXR4^|DHh`$Fq`Q(lL5*mPE!)HCcp27hc?{gX_h1|100
ztbKeXlcl>t4qlh=`<ZgUA9KdFwPweLFbEGzlt1Yxc+laDJF5NT;P1EP!<u9c<jrbY
zHg$NhjU!at&sh0&y+X|etNbx<9zOM<o-}Ur9BE~H6|7!zt9$*|5t6~YHK8@_I_y54
zgs5F)+lz&z<LT+YZc0fu)+iMkK|x2ibb2E1*arp{jHNZ>od%g*3(a&mqPGb|7%wj`
z^gwk6E-h8V3l^2*H=t`ZxT+%~#xrR#3ZhzvL)TsK`8+{MEZ#41$Y-caO5MBW_xkPg
zf-8_uJoCNYd8QB8`#5XS)w-ba&EMKX7zFUG2UQuT4EV6fAl9T;MUx^P6WN?;hTrUm
z#3M7Ob`Qt<aY#r34Gt<omy09=4huo|##uDb&)wC_{=0Guf~<c3bC&-(nySoTM{D_Q
z0>e)Ad63~Q`Qp61Br|or1KdZH%&nyy*XbT$IHY$xZv<(<HyBmq<tZg;n-&m)!~|8w
z$dGlqoG$MQt-})MLU*0AJs<v}X$Ee8`|kXj$s9-~d~8iDK%On!r=B$mSO;g9#|Cp6
zC1}>@tWv;v>$z%5+|*jeEu|PVUZmpL4LWF=^V{mBb&qZTKHcY@CZr!Rg{0p1vlPI?
zRC)u6%=egst{@%z0iN<2jJ?FnJW<fy-TiGU&5$`c*A+icAM;`7wAEe3-zNOpYvhRh
z(kDPaH~RmERNT_{KsMO;5Rb>NztDH^HiBN_o#h@*9<D%n`)wlalPpK9`#ifKF-eAN
z`;Jes`l_eZ$eoS3r2NmnME!k-7v<&N;IP8uku?it?;)#SON7){i(!g@j?KJQ_p)sk
z4dEr5?iUaP{<YCaN0H2;F`7-ILkb}4#p6N*ezQu*k@GX;hd2yM3)&ZbDb2>DcYeUj
zi#>l(mR%IP-q^Kqg^{z<b9MK9_G$(te#uROC>a7^35_{o95gjLwD<vHSrY=tF)9mM
zmx*lqpse7wmDIBAUfC_P?hOPZX6YWwFLAN&z|O=5pvu|m+rZF2`U7-4@gb08Q7!gB
zq3?tMfHeoaX#3HjA{=fe?2Ej7#ObtN_P`YGw%315$0%f80)(KNat>Utt?122;&-U{
zpj@1Xm{C>EBEa06uXxnd)FcQ3d*nse4u?M5e_Tz%kjQIeWL0nIq8Yl#dX3&W^+2M6
za_Qqf!4~7dg)z#9@2>ZM1%nt8B(>wFIxz{_Re}&fF-V~f6*ZvcWM_}pi-RP|6jDDp
zqlS+BkmQY>xjnQS7mpcm0AfS^(|9<58iVN-g8=V#?XBiG_;?`=$sbnjhHwjo8-sBY
zTZF3D(i6EkD2kwTAurL1&+#~}HaMIPSjl(+Y#1~$%m?sVV{sXKmClee>(OU1`Zl<}
zst^Y8p<6HRY`klb3BfUJm!#xage41kj*iEr|NP<p)muJ$L|lj&blpYvMq7vQht?kI
z2%hFPx%Fb}`QD8DkFt(<Q5A+LIiB2dYv<U(RJB4?oqH~iVq<3xh)S~y=c+|CIX7qg
zcGN<_Hvnr~m!4#z!h0>1#U&C`yyt`NwL{eRdRcIEymF;vy0-2XLkcoIOpd|<kJ<iU
z(J78&`%O)J^S%hA;thgUAuqp<6(|n@Tc1EHh3+JvR?Ax3eXo)D^5Ed~<sSTA-aGcf
zegUItmFl|O(nJw6>=AD}=ikN@2#vj_d)Rg$BCK740)kWBNyvyrrEmgs{9}d}zinFo
zk92DE$76hXd3bJm$N_<9sU;A5&lZqUD8$bOcLpScjFwmsIp&{FA=Awx`xV;*N7J^S
zbhOOl#WKY-lqPqZDE{O~%mzd-C3R#*AT=CPAzH&047=sPH6l!9K@@;O*O6`%VNLb4
z6g^l*`IUCHT`fvxR!4kcOWf{B$723UnYSj#+%D-18{j$gkfy;yy2w2f)pA9Cm?fjj
z%_Y!5hEt*#+@Pp{l&4RRr5z5$g!EV~`N)fUv~=j?x2At)e-s|OehM`&VP+)Re{;8>
zJ~s}@;Ga`T-8HdY2wj|P*CR^E3L)VVO`m`i^dRbQ0a1|nCexNux)nO6t=A>f4KEPC
zzceJz=zXBr-PKCD%cj_u?J}HExNqLsKogR7wgASNY7z!&O&In%mi(6iM-uw)x|vpT
zAVz9#ZeCwA28r-lWq<SEGw5i;2a9IPk4R$#6qor>4zx==>byp?1>U$;oW(qj5+ktN
zS|U5(3(8X#?KZ~k5#ye(Fu4B?7u|Ikj?M^A(3rWang+tV3=Fbk#}Pzs^|GOCk?=Z>
zEb+Nk%jBv#`iuSA+|t%c-FjP-{cB6VZHYt+1Tjh5D5qMZWTn;e0D{Q*GNvnnA~7HP
zY;@O^|6zR=TRs-1yM!4iqKX2$LS%;iMreoInx$VIwrhC3pbooBT$4S%0<RFQUG=&B
zNu5d%Yn5brB?2aRZO>;}U~VpCsN7W5;mA%am^k&@t2iR!b%Zd6QP0`upVwD75y{KO
zpVfbpk$<SzZ6gq6@nli)x0Ap`ZO~zw5z|I&>QTcY)AwXf6#W${U$C5<rZU6k<+U$I
z%*?m+Hzh~*rvZ;|cf)VpXd{vjcLa{7>9Jb~tQm<D*#c&@4oY;i{bsR5Os07xi?TGG
zM^0o95?5T$X5I(J+Z2Zj#S!Zb<~<Yj=JR>}jP+z}zxKr5>$l3X_3usNJYgQ{O>-~*
z80OWOWll?A4;;MXB~s&)fS|}uY#MIH9)0N{w)=Y-W!dGvSe&Q9Co7$Em_!Z6w2;!!
zj!5-EwhehINmP87jv8vWQ24`g1q<Djme8wzz-p<oP#W#5R;980{#kfWw<1f`F4J<)
z3%fGVeaU;&S8~-j@6T!t)j__pM$GZMYuUu?eoh>6VZGwl=Ak5W>6e<ut$&cWS{kWj
zPA(6SMvXY7v+fOz7}a99NMPHb*WS}+@GR_<SB%$(U)q~2TLBifM6W_e7u8CwQrW&3
z_GsI=jlNQy&Ks0b$8c9KL8b{s^~*T@Pp=SjidKt$NwIUk0=h^$baX8t6MIYzJIC?z
zeQdZjvcjgavi4HfF3CdHAx2wBA<OZS5F0byb}6&oLy?9#nf+)eg|k|C#ClkSwyt<?
zn~eqUo&KBsF>|}*2N^LVWK=6cyQ(Qmr2<X}9M`1cqgTxs%RRORWJRZT%?Eg26KT>i
zV&oY#FwrzJm{dIHo&ae<%_BPn%Z3`uG1^Z{7bYchcY6eXICUfk!7~$K-9`-YZYR1B
zvh7Rot@<5$@+iuu#Cg1TRf7zN&IhhfyJk}#H&bG1sAaXJ`(mICFaL5@(n*c$V*hDa
zH2a11>%YGYEry&VkUd`#$`ZIe=AX02xX(9s^lC}>_jt@lyRFJwwZfKank?*#S`J2z
zbjiFkIJ#>c0MbUP>B+zj=KOINLtm&M>cQNx-&u|E;!=G+me0)IneY0-`p}%wjok<N
z*A)G<zltq0iY*7MG6-I=XA50AE~ZSPJ}Ag@jPQZMkD8xs*wp#3TX^W>5GAnvvM79B
z<2<pcYAv~HUY5?NuX6%z^$ju$QzPjt33bC=fxgV%yG$zpEIXv=6=SmMt(JO4n<ClS
z!`2&eux(OXr*pcZD-aY1TFb@z(BVIg^M3{lGj9&Uvn`li#iyN!2~Ib^-_al?iB~I%
zrbx-}qRDVAQq@7CZhX;Z+~q3&;f;~$^MM%Z#2;%T_K!$ERB2B^)~K*{$tBKX)6FXf
zY}T^Jvp;F4$ge&!zI~AKx2Y|Ftr!-MTfZrS8L<8oJ&Uc5GQX-t-Ta-kQ|_Edve)AC
z6Z4P!-E#)H)eF2k>dkX+i(9M5iPkMCGjGn}ce#Av61NOUEwBswYioK?p2rOAC{~n?
zN2HgG1Ul8%_n$WkiAc^hsrL!#Pp;#`3XUXyz^w09><~#CN@KW)$)3MUgaLUDYA=b8
zNOP2or{*U<++Uz@v{<rI>tf7Ytl6wv{Q%kn<*xN#X9Rbo7#2|b+|o(<{t+(B0J_hT
znU>JU?4qXDy2X5ldSh8==UKFQNRs^H=wNKXYj3{6nNviKgH^?@tdqNsA}z*_c1a%1
zHK&9=sykGNiTxZVf?%I#JH@W1SuUOKZvPCX=XB0$ZZNvC+GfpiZglCJ>L0UAbWuxB
z&z!N0v04Im@yr{29iJ~}Xtmb*G;o=JyJZG_xcOP3Mz%(W`PCgSEZYXBwigSp*=*6e
zCFl4bb2O9YF<`ayNZe{^EpPcn^NW|-=lG+RmRra1rX~B2mVTC!5_hAyJJT0eV*1&d
zYE73xQ-j8g#!+m<qQ25(UyjbgdObmAr9*z{_odtQ!X!XzGBSkjB0mTd+&aCX)si7$
zR;dJ^qMXOU(Sq-LDJB&bV%J;gIVr$8zrcHGSdUvhM*W9Y(ksD*CG)lHq^7&i!b>N^
zp3Pp3nJd)aDe3tASvJj5w)Fh`bA|O9_i^i?hE30JUEz1E%>FW+W}TW)boX9QB4THB
zMCiOLyEJ>Hu6jJD<8b!QNV;Lu<dW@kq)M{<>X`CwFP)4QqXsVPNqRzQdS-ppDbIN&
z^a|$>O`k7p9Hmv&ahbF*)hCblzfQd%V23Q%e4^>CmFAA+m9{RV!tj~RYN0x!z2^Ur
zke2igY6LY6H%+vt2c_6PIkekk_Or^aRD6;MuZzCsu+!E$u5hvq*;Oi88MB|D={9v6
zW17)}AF!w&FuV~zrdzHnLil}@yAH7a$rneMs;(ikzMS%_t&tz_g}m9&?UU#sOTQdH
zERBqTbS{Dzqdlt}3ayCTeC4#?&nPw-^x(Q3@rj9A-QCJ}&4hO9i%hbpFT5N=mr?TY
z@2`3#nl(SKCX$eowrF`@el;wtvvp55^bkI`437A%A)yngS*5Dl+AIjVRC~j1pP*JZ
z1VR^FrqtNakT@U%PLagv5y}(rj|0m9yU$VH=XFrSmpnLdi|Y;)C}Qy`Cb~Qx!u}L0
zXtz&Sr$4<2tVcRqYMLxOJM5zNMl}c;B8T!z2%k43EnBJWhINpbl+6)_#<!jUo?MI=
z{e!*<gYv8Vj@O4fU%3U#idtL%0Qm?cw*4DGu91Su;L`vF`$&#(VFCYL@3Qkx=AF0b
z0`k@$lw_mhHh<?2pe4r<?dLlL;Y?SzzxiJBa)17zLtV1>`rmqf&s^*)v|>*|>tsPf
z?enMhBR9CUn(no6x51I`jNYQ~`YAf@6VtuWsLI`r+vqsJc1{y%4D_5lE;5)Ssqfo~
z{k%)U(|MBVHeD|#inD7ibam@<wd!(mlxu0{ZKb7y<8v=hW;RWTxW4NAuxeg!xH47b
zQDK1U7`H4eqTIFE{~@gxMZpW4J(9X_=1A&@RwhAYuSfP3855NVUZpBy$<k3oPDt=0
z=ra%}E9dw`^Ja34t#iI%+UwRk;<_d4**Zn4Sozg5MdnSSZKMA$z=y1O98eao`yCo&
zRoA0iG(F}IKh?HAY8~s&&enwZk_1f=QiB$3e4IN&g1uQvZkO;_)da^-uVdt`%*}JP
zeS6AlD!e<aPg_S$3dY~dd>1o^R{DJdV>scM;qK075PG$eFM+MV5zjqwcPorunIh_z
z>vHas!xs9QBTj4cF+o8bC#bb+tElHOUvfk_dSzk1nr8IAEz6o}I68K_TaiX%Sr^s8
zFKw;Eh-t+a2=4ZF-bvZurL<(Erl)vTwZLfL2zwSILRyPTR|8#hb7HO*4TAkb>D@Zy
ziFu4@c1yi^{+se@#coD+-hHxNUi?^W{vWDkw=n<^Z1MP+7ySa$!ReV4K2LG$8d;sO
z47CGVPubphC`c=Z@hYW6MqYVi^5}coj~@-RTu75|A8xJbR6;7_BY#;dBvReeGk8?Y
z7UY!IWj2@mIKI?6G<Bt1p3X&nVQbw-$&Sd~hV<98!kM&W+&U%wQpdZa&$aDy2W^6~
z=tU9}*WtKGrc-n1u#gs9{>tsIN4B&XVlB+h_TM&q*&yShqs{HzUM?=nyh%wfM*f?J
zrjP41=s2AtBXdZx9}o71M|}0&qWzo_i5Jy0rxZKK@|5%{dM4fjIexkKT0M}LeXe1a
zZ%nDWXE>GrWj4u|d05Xw^@-+NJa^lZCnr5H%H6YUx~BPY?1EUVKB^<B-dXz(@kK=f
zRI}%CS!s|?GM=qPG01`OsbCY<ecc<E(`B`S!$rq%L_I?oq>`szNV$;aG<xah2}EfN
zM1_tp!vS%1nfv?gETwu`UsNSW+389vn*V08T}F6Pz;yC>P=2++<1v11M3p36BlY+S
zCyUVlstIc78<LaL%!l$WkIKjWePgJv<Y0jXYh!$&riuBJ1*|v6QTf^eVI!=E%Dis(
z!zp>!?1M_uGShA8F&y7!+g$c#p?CXM_d^0=LPqb<-&BfW)gU)Qi^kEdy6RSpa@+R5
z)xF?mt1B%nudW`+(hT$X$)`$jSmTPO$)`nTx^;{wqNQG-^i8?EVVSEz{EYl^PDR|!
zSN=x2xu~0x@@Bie5PiNreef4+q-8MA6yLWI8;*+5$S}{(C)TZNp(FW4*u;IHb$s(V
zRO6GE)Ofr?rjj0yi+uSe$N9nFi28#3Q2%@7CN=WBW?W^F^bA7rlZ0}^Knu)077Y~c
zdT(c7Tgh(M$^k#;3~<vKZ<WAS19&8Eozt+Ncr<22c8a~Wuek2#vibi7{??5C3P`xD
z$K%7<!B6c8vdnjRyKft&VGR~cbEk5_M49RC#J?lR91<nSpj&^}=YQA^j%r4@q3r(1
z9HcHPc@VcA*)~NlqK!2edqo!mGDj|J!lK2M=efC}n?9d)+mPG|j+>xBnTPLmOAB<4
zrUDoW(mT&6<*k!N=B)|8d*NZi-`K^7S_gdtb(lAqeghtgEw7bX8ooP{N~4tw;wU`j
zHlS|WK9%jKHE+^}0cl5e&;mE_v)mBn=U=N=e;oQ8YH)dnSTw(ivuZwVHMOG5oLv)=
zcya^Bmvy?by3-cD5_{JQS;4}mrFC$y|9PGLnCEZi{l`8j4z~lG4sHir2=s)XCVMzT
zL%#;j9F$<tt5;fVY;_8;d9BJXU?<*^;fgy)gPj+0ZT!HU{$25LZ2iOeSNATSKii-{
zO|6}g$)aE>pz*_ywPc_kTZ9fD)zM@ubk_V0XNQa=q?&0)=6JRN#m?Mu$EwYGVWOdk
zuJ+8ycd=iJ*-%Twj*N^;oPu;4Tlvq=C}b(-IpFc>L(-v7&I+0M`FoT04Y6;lgY5^e
z>9gh#7nN>SR#r3fOuL>$!`sEuHy0+!wHOOSzIo+_pXx14-i1y4hdW;*CE1}?#vFxa
zj4Gq3oH%Cfy!iF~y@O4r@+vBLWV)kckkjb3_C>X*L_%s>(uEY0)Q<tBrqx$v<VS7<
z{rFC+9EPNhcg9Q%NhvSkLFU-b;m}MPPjm|6rrOd=k?%Uy+SxS$7T%31IN$*=EtjXi
zc}QuGHyq1DU%e<~1ugAWgSMEJB{IPlDjnrLZ(Osv)?ohW3ZZ)K;E7Y=XQqpxf7JqW
zz+u-w&zzs9Q}?t{Y&Z*xaJOQ7>!q|wxxAG;e|tgm$k)!AAPj!f)qD^1`M7Qes-~yw
zy5ZDyQ9lM=m6t!^m&laA#L7CMSmC-iG=5mI{VXlLsO9zxzXggT&slfmYl}=W(xrt9
zmZ_uChfK|{X)O0#)hvkPD={3>d`C?0RYa@mq_m1<I{mI0jt`T3o;1m9dtH?AxA@S6
zHkaw~U>fBhR!%|aq8>?l9R`}Le*`M|dduXF?JLrv9yufGTYc4H#;0+-SndR^j0`B&
z8mp+t$ZC5jXZ|X~@-9B?QGFEd-Hd+tZVO?THvvHjP>Xb>Ve8HZ`psPqU}Vuc^qd@S
zo)6_;8O;uJGqdD%Ak$&jD2D|z3c99_;zL}6y@MYuPwE{9wR@S{xpq_SVv;|Lfekx5
zhcbsbhY!d3w6ZMAP<Q)Q`Bf3on7JhF)ZBE}Ka^oadb%GNkqYQ8Q5J)RoI^6KO8T#Y
z7EBU`5(k$;rh6^&Zlw&&TvZpg{91ne<Eba~1*$hMX=}P}Z<?9BrA{lpP`a;-7mNtm
zzw-R6KF;XdtH=GeYeTP;i}*jAeXYmSO@C8MMy0fkxUi69u2@$mM;$`TnZnCzOpFHT
zI7#`%MH$$yLy5$+3a_SVf|GI!gVV6l!8PqC&ewN5;%y|XGonlw=__uX4G8Ye(~U_i
z>+CPzR=pD@mseCTHUh_WiO(FlWX?{xOYH6gXIwOb17jbks^Q#$csLs;?AAsUw6Cje
zR3yniNS|@+yke1QXwqnD@ooIwo&{$n&0&pOiwQFYRWqZc#a*5dTPgE-LoF;VZ1>8m
z-Y`)>d>|(<C8<JwVNz@)d9My5gn6>GEH6#`dpRlZ|5{~ORRB&cr%SWJop{rKUQ>@J
z4(Aygrkz1x6iF`2s!h^PiTF0m-IOwm5EQl=8dAIX-8tUyCQbu%8m>3CfUnedL^emc
zdu5u}?Yb{(s@!4io0{5*u3!cI&^y?K9W;uh$)>KvtMT69+S!N4Kfe0QdLu~<f)r$Q
zd;=5RV5F|&og4EpTR#4|hNETe#I8IAy+ZaoQO_@t6r0cH2UuXsq4F=8#o4Q4I!29=
ztNCb=4h1QBuNCAc*%|}pTP9Lwrsh`&hqp!=k~Y_s652yn%;h$n)f>dRa}@393x&IA
zJ=wnnmWBNrQz7kYYkfYfdSI)VkBC1&2?V^PnGkinHduA%7_0P6mZ$vLqLSh~Z9jJG
z=vl-YCTbYqY+~aKgmwJob82&0goDb0tAp#MWw;`q)0SO1Z!G0mTIR8poW}n-Y)ltm
zLhlM~Mt))4zlDFHeUL;a9SoRV|M?CTUDSl%6sA|(!VDviJWe~Nc@88-W5?uOR8{30
zlLX;o?1-L{_C{rTQ^o8E*B$G({<qP^=+m#Xz1wIq6rNo%f+;eY7S8Hxk4QHp$Q0d%
z|E+pG(N3sU;(uH5=f&CF>M2K3dtV(%eH&)i8l3LoF@swlY3TP*)U|W2e!Y(*LE63Y
zejeoz*L#fZsYxfMB~lPOxe@Z-OOJSsxMp%2WMsmT7QCE>IQ<LE%tEXdgyy6;vk@U`
z@tC;<cAXE{R<)q3ul)ZlR!H`Awh@?yX_A)7@cJ<Wm3i4;V2>pyM?QUh6W01u*qv)2
zEN0h*B-4e3*_+uA?5EiAdDDw6z1Kh+UwCOy)_Bs4pshr{zmcT1%z5RiYnY*&!pscQ
z^A=TD{r+J$Ae?JU9@@;_S=>>?>;~D7dE6)d2dw_N`Tg`e#<S_HT92g8u&X?JWOJz#
z3+FEOieo&toMPNL3D0O=4Q#iZFdut!&OF<No=h#=t=>-M+9LX%8s`~44c0}krNs7~
z-Y>?IwByHYoEIb47Gg2C->C2ZdtrfBUVclPv8hO;YqHhu^dF0GLm#x3SLcTFfhKI<
z`3^k(qKp%>E6v@}JS7D=xz3=aI|a{_<BQ#g55TEUbHkQgh=WZagGKZZ+pXv@5>Dkf
zgWd`H-p{hm(>x0x1~!&L2?4345BQe7+!ofKJc(}7<5`|Qq=~}alp)AM#fz5A$Le1h
zHGjI7@1#DqOVOw(WZuLcNsWK!vsryggUl%HRbS4v_#}e}ZAM;G;I=&!NM*p3dPt^f
zN?5^G!(g#MB@@-tMn7LjI{8IA-lex~R<L~yGW)3#1wA$UqU(MLZ&W!?1cQ|iS}G*C
zOSYK(n|-$aL5Zp~>PiKfk*XK_G(X<><uH<lMO7okNxw9BzT@OOor_>E+4z7AeDgsL
z-F1m2=={U6$y_z1MA}|W+O3Ql&7fyZNl&>t6!d&YUop{`5ro$&Je$w74%V3m$K!6B
zbiR5C-k63)R2Nny>hcwRr%nGd?x_Bffv9@T)O72N!n|MRJQ-|_fN>HmPJh#F&Gi63
z0M#Q_BpZ!@);MuOr#Wfz61NpY`wPFBIL{zu`lM96c1o!Mu0}>9VFwiqLKsktc=r6<
z6YIieI|fy^=}nTZzOTa?6<_6MX1)XNKu_L-#`Moq#Pm#VjpUc<x+vVT%fAg5Qid45
z)u3bMm^EEzui2bZpejMl_Cl4Th9_VMk<7|qzhukJ6sGGAO*G^c+~TPbaADHiT=;IT
zA#P1PvZh<Ec}`0w>aERWa(-9zCQMZL*~v>?HVf++85vXeveqrvzMVS=ka+V_@<@75
zNu?}p^)LDNhbka_ya8sVrZcYoNUB7VKI@yTTG$HlL}hf!nF63nulX2&81E9)9-6c(
zkm8c!AqeORxalv>GAi3G`-u$OM5rg_)84V|74EshINje+XHLf}tQ<mY32~7R&s*cx
z;`qkM-tcDK-)?c-vNEQAxKoXPmqKvBUnQXuI2=qz=|`i$)+Jr?s=tdgM%aXe<Su|V
z1SjhICqxI263SELkc<=;R9W6DSR16Lwe6`6Iyx90N!k=r8dBPNzNauQ^JT7a-35VB
zO&OuHS&CsPB7*0+9Y=_S9O|{TGcWfmnzDd`N(+s<bdH{LFjbhy(<U}%_-j|*KSt`(
zsH7C*R94S9(#XZmUf<MY0R4&kFb;U(i8K%GkDk+a7+*+ci(o1@9hyP*b$6pLY3C`n
z;nu;5zeGWm*oDplS>nl-%=UIot(}dS##l~E$awP)P;Qx9OSDM2_A*7>{<~tgZ+}yr
zdE{yCMu<@As!;2qoPFh<$(iw>Zi-=U_6rlwFo|{j@@V$@*Fu#ask{(q@hAqbK&}pC
z(=d9Gilre-2r_*g@TFsE9-Z@eRbhVy_6B^;0vM(Uqe`CT4YPP{3G2Ik_qP`+CuR^o
zE_t#=vc1hAPaR5`trGcBAwBaHJ96ByBRoM!K~BPoyHJV4qTWzreq%XK1cdX!6N$Pi
zx-)X<@R{D&k(@w#Fft~y5guS^EFa{2_U{z6Js$$0Qp~WQr#J+{+O=<0-+ONWAsqRy
zpaRJq{EB<p{^c=*t(qDeJ8XLjZDM0$T&f>=<5MN~qeR)*4$@*w5%FVkX6&4pR<)M9
z75(R()7kEGMWoDXo>P*&omvr<#=^<K-DvfxV+-9)og^$Qv~(dZUS{XpM)O}4x5vw<
zSS|T{KHFwc`|da(L&6hJdL<3+5JT>uJX`2RYvuA{AIH6M{*e|ezbaK@qwL|_q1b6-
z%3@1nz1zXsmu)cIhR;$-3Fq8flC6K<#kvx5Es#k?DeUa*L}=hDN`OeZst?;wH!M8j
z)LDAMk<_$w*FaU<x{OJILJA+xQjHNglg=xw7@zT{Yr2_o-uaSc&pT6~$j7_jC`&J*
zXMFnLZ28D0hjgxbu4#tu(Od&P)`T3L7NLg}&F!1(<KxWWImPo9x+YmQ;@;2bV{;Uf
zo0rOPZ#qiwzZ~s%Tu{P#`}N;9AbnZu2C0m<PYNh8Wct5u%vP<S1YlFAG&MJWlA6p2
z@pN_$XN}28Gp^{5lw($%472NtuC&(5uO5<W0Aq??(DmEb8w<>yEAbrSY}z^9azk;9
zm9vWN=#b#Y0^UyL2P4$+n)FODs~I&NZ!zg2Zb(mFxx?eLeApCWX5h~?qee1}4!`Qr
zdg?Ldp6CCTvM@KCj@~7}wIy_y8RioQtv1gb3JnU1u#t734Fx=!s^fF{WeyG-Xm=|r
z0;x~>xV%zJ!;vGYnZ^}#IXZ!@fW+D0)q<j<hl`7=$1UA-x8ChhrrJa#Sy%_L;uRtq
zSwhpnT-INl7W`DT<SF$yA~-$xl|0|ZySRM&t+;{*nrRz4%064Q#}7FhTZ4bUgnxB~
z<l0B1(+FEr$ZCQ9JW8)l&zW1Io%(N3@V*g?O7a!r7f6MIj;l-PuZSMB4wuJUNV7xc
zES9~8+H=mcP$ob@i}{Gi70#8mcIH#n$%GIY%~W$oX}X-8atshz4l#q*J-LGc_*LWc
z>4QnH?0fF!_n4;_(bJ*|-WLlRIt6-jO0$QtpU+?awxPx-=%hn3_)r^-gwZ9ns8cpe
zcsR_~Vw?H%=QL0q&Uu0jGNEC)5uN_KXY~#=VTR1J(AyMlYX43obG-($*)J(4ckq8~
zqJMz3>*Ig9;h_Qt<@)8dvETAlsBY<(o+&T3q<1Bu3dX3fM^YyN!H8+4TjkKc2B0pB
zOilRdQQUq4j}jP0<NxUAd#jn>fe3<l>!Hpkc>up3`&$y>1oBx;8H(q2D;l}FL{5^t
z?Rv4M<mo(rY-k+7{sJ*`p<z8qlZ9_It*nfdeg(299hMQLkfD!q1>heBWz%`3-y9Oz
zmcK*$&a)gBt3)8lu9wDdYM?@ITLzi+xLVLiQ}a>^l7FU^Wa2#q&^*uyH*RS%Bp(D-
zs4Pm(ZA2}mnoBjT5WDt25?Z;m?(Tx`GGy2wDgahb&Hi$>oc~5hguuq%$c9m6fhcVR
zcza>byyDLkXqwbTLPK~q505jtR`t$#D;Ydm@Q?|=hj#Qa*?H-gvPu)@BHTYfCB`Z+
zm6TnGI<Un|Q!RD$+_ht=u`(sKw}wQx5D4)R_5%<(VW54qP%SlbR2t>lmLcc2eC+Jm
zNx>MXi(X>XG4CsOJU1tNKg+m+_r0y)tCimKi{MM1qK^SG(3)3w6h4Q$Wf4NITQfx6
zetD~%B}$6tY^9JP>o`Hm$iuO*f$Q3NquPt}tG)J5Vj*5W>YLKGUSJ8vi)rPAxYyVF
z9;TnKL_g<7)d9Z3y$PNYGfoR4pmaIlr(GxQ+IpG8WZ#Y=D93_Apu)r;1-cRoLrNVN
z*P!N^wS_oGyzo^7kq(7)Gs}18`&%cojl3J8ahCn~Rd=@-)}R~YNstS2js3jI+Iugx
zS%JhCa*I@kj{2Hc7Hh_&6OLUZlxzr|ePXltme&kxwxEX+#+YtA$CfqC0A^7=t~NtV
z&&jEyJ}Z!6W6#2WSI4J<Xj$nmBjaX;XD6tAlDhc6xVq}NDA%ofY_U)fP*RXigH9zJ
z8io`QI3V3A(iSQpgGdPo2t%jhkP@Pxf+7qlB}kX#P(ywD9n^d8_s992^E>w*e0gH;
zwbxqD-X0zoWeYDn3BAPMSjIV+3V>dc&O)*6E8ZD~JgzmHSJ^OJ7szS%rm?KzoK`Rp
zX!JpMhVFfEKNR|kPQ5vQ?a5)>A?-Ul0a#Z&Y{HTxgRHT*VV3yR<jIPTQn`ukXS~5`
z35m38M`_r<J+GFB`a9F8>Tp`jv8|a%Tl47DSa{_PbU><rq=ezny7h~v0y@eH+Lrep
zI%B#<vGGmq+e?%2RLjOS-+}f}UOP(7iZDX9$(qLXc0yK-ZF(u!AiU%%{1@1GgWOS>
z^$|+%66ivCcjby&=2Dx&uRX0HCsJZC%C|ia^ZiIDHk0#FRxaWm!r7|SovEioj<SsH
zPr-w_Rel@CIf5snY5MtzeQnqz|3<QPSy{$YYL4@b*BRs;H!*D&aF$Ei$)$1p3zqs&
zFg`Y$RbE>%PyDD_Zl5TfLXpYKAe1<ruajYXi~B$Il4dABHv#7PCP8Oh$Cpwz;Hdl6
ztG^DfYyn-AU@9aYR6vA-`dPugv%ozBW7m&9rEu}uFFk?1T(n%$>kz#t-~h~9?X$Mg
zIV9VmknHW&-nNWVL}J5<r^*viTZiTpY%@0?dmwP_ZC784+?a~Inqgj8oSU1T(rJjB
z-<ErL8Yi-DJ>%v9oMvcf$oe{P=WNM;gtI+4^$OV3RF8d40}H8@l`j}yj+X;N{{`j9
zpnHCK>LjsEDf3vhcuFS&6<X)!Y0sHfZ>t2oZ|~8@5;>Qb3o9pB@iIGKZCuf>TQaxD
zP6%lwzKuO@64(K$qE%FRXxK_Uz<*p_J)eUtUs~K+{r%!74{~L49YH3nHreV17~6We
z?LR5_rQr=h9v&4D@L(_buRyzw7HfE9_TTI`yc@{rLMD*^0G#tnbS^p(=QZ~O9)wLE
z+lD?ea4tI);7sx|kBxPLbD3hu^BStjqDpbF;GQa(tL#NNwEQWB;M)6R_J;ns8ktId
z?H-LAbK<$z01{_T*2^UOy$>dalgB<4@Mts<)D^z{YxK3DKhTWQ8y4|a3jS^WD?mto
z3^x_E@6-x@@PG;=5>O|l-30aDN1GOk=S0slF!0G#en`6%9WyJokKr%ziL7-H4hPxc
zg*!Z~0{S)!p6+U&qJlU6vg-!F6GlOdYLZUI5r+JBW*s(72hA6nUp1-Fxst&m;2^PL
zQgzz=Cf}s)vz-1K?hrAK166KJc>R+i(AT3ZfH{2POtyhEw?I^EV2@k8{G(gy<s8~%
zrgWF{O%f|Sd+mZ3KZrrWmM^<h><DN8k8iv5pDN;I#SNY7IGc!qqf~|^4(sr4EZ#{O
z@lh^V|4jA#0fqwI^T%76ae0y`9nxp9SbfwwRljFz+!NtfGCD*YFO}naDOo*=w#ze7
zll%R*J~5o~ABr%e<xXELv+Azde1E3f#3bETlP^KTv8$77HLU~n=?Ny`qQ8<~WgCYM
zi;T=LgzIc+yJ?82DyXBlW}S)ug6=xl&?C@j3YlKt-}VOHN`jD)vLim2I(|QtmZ2kQ
z83_(mfZ8RWCxQ^GUUnPrhOWb7UvO(snEL-r>nL5bnSxt0*@I5Z4jgMh15>iSzV_h*
zqwi;^$|X~=8AW#;xOznZpgkDt`trjf8OjMkai50wLBSNZwp<XaKTDwHN<7<-J&lPL
zgMe?SMe*#vpo7rezD%U&e_R|FO)_+idC_VA{|>L+(HkI5qpv<tll9b*SMM_E5x4U?
z+>aA8Ho8<_FLq!ag}Ht%g$mVMg4r(XJ`L2r51hbw(H#$e>xL{>=ut$U!yK;oSsT8T
zz1Lngvhv2QR)lEj;=gP;zIZ-EJvg>SFJ<wdvho`?2ZrJ<j2+u%%m1sJdl12c5dqo~
z-0+86z+Ytz0Av%`j?-1V)NsyhjJ%b*lQkK&N~`Ag=I-0NHg^pPIjn}{&@DWQ@eFZ(
z3(5J5k0sgYCKl1Kr<?*^W6MP%-T5;&0v<fr<Xul$#ELj5#y<5j_h`R-CEi*q&Vk{$
z(|_LSAnZYBL1+&N>FI>E3+8nJOzO&H<KSotHD*uoMENdz<VY&n7;e2ip2*4$P_HWm
zO}5bdXqIkqu!>Y@S>I{bB`RgkjOs4Bk#&d9sI6pophlr!bt67#4DbB5yr`b1lMx2@
z|DS9n%Ok1(%UyI<z?q2_^~icBm?yF?EeGfg-;mA`&~!mSGoS5yphTINiXHGs&JRkG
zb`;CEjch<5P8X(ner5uK9Y}4r30JK(AvfRk?PE6ma$ek3fKJjl@PR{Nx-wmr)YX;q
zzHLKjDc`RV$t*HFP(4yCVu79I;WoL+hg5#ncRUhu9sTdhpZu>N@n1YU)O%o%lu9;k
z_c&Y(qK(eX>{>8ihTKH9_B6OLY>)~`5D4D8`^J>~`kS&vdy2}TqW+nn&KOxh8{J~-
z^GRWjk=J#{7<NnCvXJB)s?>~}wsfW_GU{H3IELGb+-!lK{f%@dFUGi>&2Hmp|4sGW
z{f`BgM76jeRvjTIk7YXm>r8bp6V`$~Gvu;LELyFU5YBTY9jdgq9W^@zZQ+nkKS^%1
zb3!CN9ypU?M&7UgGj+9p@_~`^oU7i`#)B1pIwi3*)msnWdiz6Vsl3)-6QAEI;WYEU
z`wb-prM{JQ;WN(fiR*0*o;lcpAF8Cv!dQIjbIB&t-wQ=zzeH$?VU(>nWK$iw&fJLj
zhszW_%rMuO3N%UOCue`PBO0RNgi_m-sh<<Bu5CH7!H0o1`s7aKD(32H-z+dej1ixj
zqP8w;h*!mpjK-S^xY_4~uj~lxwd?nGsywC$Jrm0Og--jtCtFv74m%bbL{aizHy+V<
z^!P~u3lE&kTRFs!*IOdNAUix9tMc`EwMF~qvH<^^kfXWq#5WT)ffH#RHQQ67WX2Li
z2tBx>wtZf2jR~{8aE1oF#{Xpyiawrn+v2%HA>+nJd-Ww^czi`}c*txBA%vK-FZ5U_
z#av%p_CtqhU}%?k!1VK9Qt|)S6X7tmt03v`ao^ALy{rxp!TETa8=4CzPUi5%5eZ2=
z`0_T@d{Lx;rilNfv$OEikVr~{a}|y+>k^j%pg|f;Y6WbzO4z)t9#CbI|2hA`^@6ej
z#loz@2coZ@zv6uT_;nrrCKX&|Zs%TQ%O18Jt`0QWUr`tNs1HpFOL;o%rC?ZlIzKT%
z%@b|A1*XHw%41}|PKty2F09f@yOvz?Vg<oH#|P+lrc`bU1uG3j9C|42()`qSF3SZk
z;p8cs4hwbTX4-05Y<S~J{Efnn)OcZ6?w_I4IfRfcP!mp3e*3H#@x8*)6X-kr{LC<`
zJ35{afu=gqg*yLvqO<!=!1Lzou&bbcok#(LQ?a$bo!ug+oKT;s96KvnV;wFA^_h)v
z2KkRrpN+5uftm|rlf?6^Nw@3bNYUt%g7aCIji!AgrW`l4xWf~;e4DH&5~5G$Q<YPF
z=?KsO$&2CQ*U1|-E@r9nhK9-k0oM$?F+eQ_VdD2_A@}I7Hv?h#ZEjbKU?B`8ok&>+
ziz-Nlrq2MlR!pAEv79fS$dV5Y&#2yUg2xO64F2;!m<_up3wmE%MZ4mAx2H~gQgLR;
zQc?13G$H6p={?ZH>5Z&JQfsq`mxSY`Q?%JXsQ}>*3R|VO%Ey!<gxc&-RLE4w=_oj_
z^9N4<zn8AN4`=he89cp60SICLC#(GpkT_lbTtdVQcumj~rnfsBf}49)yF#$emBQ*n
zLDs0Qz&_h!Ia^me<4rc-{7~CcdmJ$pRVC>;$8cZ%hR~eS(zJrTf)g9=Vw*_6HThXd
zQ*$?+O;~y@?MM+v^;pr{7kVe)e3zbq6I-nPAMufN<u%}6O1Zqcf@ERCY*h?~no8pM
zG@D+Yi}(W!-8A9mv$d(CnFqH~YMGV2F883KzG1su0Brci9IfA~<9&GvPfbE6-Nu01
zeS}XupI)lR%LEo9MkapYT*KVx`QBnr;sfPljDj*?p74JC1w1Ki!cNKpT)GVXgHtz5
zQ@}wiIP_N&MOKYa=>`NW@V|Kuu~2X=NoK3k>7t6&Duwez7-Wmtw5OX>6Z|h29F?kC
zDT>tFsXPCHm9fJ4@wfJYfH#*V3i*AOmIv%UcP%?j31rx1%qW-&fvml(pE!jD5`*@#
zlE+SCf4MZ0L;&<afj1}6ut=(-VU-|{aq@*7`;$dT@p`F6A?O)JDW>9yM<x=O@c76~
z;TPh?CaM3HzBSwW`aGQiixRV(f9sLd4H^lBT7s(6T1H#GN$j1e9=z||nY8Fcr;X`<
zJ}DG#l&}=+=p9mIo6Qr^tkWedJBOc$J(<!0pC**Zf-mv(C>5|iJ|W}0xfVp>KF*$Z
zBMukf5A#?IAX54qR4i@WdC$ufcKE-)?>~?q$#SrOI6VLg1~qNUmTn&Hy8i^;&t1Y-
z7$1lOIfLX0yLGrDK$XI9pD7PUS#r7bRJfGj%-y$#rg=9RN)d%QIk2iTIAOb0L$<nw
z%h$SkThB{4vW_zRSX(RBkZcEt_D<=Pt(vgtGeFI~5F_K!%e(QGUoG{va}dVR9#uMZ
z<hlwfaryOM#)_nGWFi;U@wz~YxDQA?VDUK)^5pFIp8!{qmhpJ?M=d|9#*{s#&MYZ~
zW~<Sns=EJu0Z7=MBC^i7wEWSvu+Sg3!zj7aMNG$^M8uC`$@_*!x@3$4>v#k(5jJaM
zYZ+?nr9=-y0B>y7(=-xbV*w=#$Yw1T4r^WX0&QTnsu92!6%<4F38VjBK>SiSQsl}#
zI)dwXn~nuk@DTgStys<P+QYV2KW0z%q!>Q|-sARgv1_x;<%;c_ADF5Gv+Df4PE-$f
z-d%Qb%M=I>{NbXe0M_i6A=z*{ALp`n$-v42<RyGq#Uk~NeC#*%afFih$i8Ae)HpMv
z3lQl#9BmjB2cDtOlkfHZH*}C-il<I)W5UB_AdpzWyWHBUH8ey~%(4Y`k($Zbs9NXk
zRW6FKfw(nd?r@T?PZa+i{|V{Lp{#H1Dv!NG0!;!v2c`?%>O1_M%RWk7;bv~k;ZZD>
z)Y<y<sAjEiqSVAI<fn>XPwV83fzswa(A8PZt$2aYj)*SKdF2xJARo(rzA3lTPf+zQ
z=>30bk<?BAZ<v<#nd?+xE>9T72MqNPUJbm}O-w$5{S=`cR52ldNDWUDpoTrZi4a|1
zBeATEfw#Z!TRm33QF7JLi4a3a4E3g^shVvj0W5)xDTpJMmz5;_YP5X_>(<w%mD}0F
zz@d>S{*f>#rEfm+_`u~10iYM_<C(M({W8hzM-mpB)IgL3*rlhJf;F@JgOPV1kmQ8U
z4%Fx;P)Em5(?Zx2f+Xvc#cRD3^Q67qg3R%{jLplpayGr4_`sbOt~*H9WJ@ZTQ_roL
zF1Z{3$z#d0AO#&BYT|3jL{=JQ(yy<o>!(}4oYZ^BwwxpX!OlnfVVs7|r<IEoq+I22
z6*O7!F{)TtWq`jOfnP~|J4fITj^4pc*<3&V^tS|tq$^JT9DeRc#tG%27`qQXk&PA}
zNL2(T&oVA!7BMj{h*dqWOonm=tcq<%aTa!KW4H4qi0uQ{B#AF2^0&YF?DNbpIV4*z
zv$L2f%`UI3pjodYgjcH5jhrsVM(Q#77vk%vnbe{uTKF>{6@!5GT-Y%fA=vc`)%#!H
zTwb7@21p~ci{07S%u@aU{N1IarRlHmv`=bdCP}*FOlYLunSHp-mfnppWE^a4O^FFF
zsU|l9r#%jL(5_K9*IeALAMXQ(_66-)X_@s@VgoVvg)82E`}=r2kB`Ix>XV{^W>b`Z
zRRYo8t*UOA9Ng!ozTOVQUhK+(I~!3e$r?Ku8#4$GqnbzoS>GHS%<LQB2+4)`@u2c2
zUkSvMR+qns?f=QjyL~A>)Iq83snHobr%Bpz?$^$yojsnm5V2lJNnACbc^-tAczP@|
zs1IxnR%Ml%FwLEQ7wAw1foC9K>7T@)8vJE7$SBK9FMu&rNHfnF$}Rmd!P=$L7ufZ0
z{)0juG+os%+>x!yCIWDWF=sB_-UjU%BoZKLDU3?<rG38U0}f$i@YH+PBX0N>To<H$
z-*v!b6++>Ze_dj#^*8BhSSiBe<rrTg2mLQ6pX1d(__QI07%3ozf@`G2H*0-?)#P&u
zT=E<+2`sj^X&+7v-1ZkBJD*Pavc4A6A&gO?n??r3HiSh0D~?7e6^|Xe6;Mz(Q1$k=
z0!i5EBLHjmXr&*oq8o8d>XtbQ{~Zni=oW_NVDywNxd(n;YRyNAK3nucJh;~=z$k6~
zk(WV=Tb)P1_EF2~fG0w?;W-#xb0lpIk=>kDtwIdpCf?;B#h%i+sGC?+vrcq&uJL|X
z<#nuea+@RBh4!+LpF5&Aso-p>`hR^f0BtE05zRrY$<<>UAGNem)6rkXhZRVtsMqU&
zwN0a#`hP<Z{tCYx%o-oXjy~e?XiqBaieCcZ6MTwapU;vq3PUf{4FmRq=+s)CoBawX
zW0fm&c_Q-=XW$F#gaT)VZ{Y{Lhuo~6=%_6db2onYB}U!|Z2om0*B43sArLSsDUWX%
zz!<5otsdK8TWbzAh7rO5`W`}j3j(PF2_yM@|GRd7RrTFmEGTKL<B7xJSfU#yvaE{&
zuplCSPb?q%)DE!$1(uyN7agt^Cb_F36)3^OAA<L8Vxi^;SF-+DSk(&eL3TFV=6vFR
zAHXit4*ke!w%KaOmiwzGvUk4E(<CwQEp$!~%lOIUbY!+wgGIs|9AT8~yqK*31K?`w
z!1n*Bu}|ENeVUx9K_pxC8mjsAZdVi^lky3ep*COi2-=n1H~{DYY!G+_M986KDDK2K
zwezNc%r(dT8G4e1g`4tU1e+gtP6#s>Gr<Nz!6cH9$`>cNqnIlx2FNgg&}I6Im)`T?
zHRJoKft$%mv`$=AeFoA#{mC14QdvgM9?`xP){lj&yYx~(hE8AK05i?Zj^hkCJQPrD
z(6wA1Mb7wD&rDc6N8E9GLX7boiSz?Zh=+Me12fFh(v13hNClCc7djRr-qcs$`2>MR
zOFDqU5lb!jtr2}4G0%bihsd1XlS{&FGIrvliKXn7i$iCEIg-yT%x}rfmG>v=?R44K
zRBdyE*~LA!PNuLdj|GXJh1+t@`D<zG%U+%C`b-G(*295Q8s-evFjAjx^&a!uGDqCG
z7rmThwX|A4s!c-G!&Zq<Gg6<H{=5m^4W8ia_^1dxs;2k4O%gwor|02*7Z4&lUo?MK
zG$^~^+BmWX9IA>9>8^hnl<`Q0>R+~Ps&K>IG=DNp4y(WE6)v+G3jq5}_Te1$?wwH$
zf5Gx?jL{+GzQ9=RV>{iM*KFTcPuYyCUj{Qy^ce{&5-ihacl`d8AFUPttJtr7VnWR1
z&!fu!XTJ*xC6!F_W$>_-g2fqO{&eBt{+DQ^_5%mVy#v99EAB1ty!)$PbO1jFMHF8g
zW2M()0A|N4m#=Z;n^^x;5QPH8m=G8^Ev#iM57~PQD*><@V|1|3E3@tSB+#o$g10Sa
zHs;hf!9BxQjzA88Lsre|*I%2b*AbU(H&lvX1$9vQ#n>RjslN~ZNErN;8$3y~OGCa5
z9uQAwfkavk4ho1C&B-;Y=B;%|^|S$sdcI6(%-Xzvrrg9*^OFGBH^8~M5SzA4w?}yC
zJAn9d#SiR^(Ftjxlm*;l!cE*2=B5j$a@Utj2PCi10pbAKX=1`RjJ?D`S~nG>QE0<~
z9kOaN<YpGHxMhrBabT5?@6n1o(xJFIHxT;A%1BzowK5x%bfh{5u&a;9hw+~`Hq}dt
zjE#+mkLSH8aI5<PS?D1`a%OZa&TK2;_zMJxdl6tO?MjK$g||j8ZWY9i@dpQngqj3Q
ziYyi?XCIzQ4(I~Emr%`iNsV||98RQ7JCJUn=gUpHn#<@^@Bm4agip5l&SpipQh@&z
zu~&HIBZ~z|C+6&R@hM)}EcBlsl}}Nw`C_zLYD4P5H+ZE4y5OCMdw@Vp+DD_Stjz7v
zZkg6Aj|yjKUh+8X>47MIoi8+$lpv6F!xh~Kq~pOChrTtd;$QSz$wouvpc;x1_e2%C
zr+1j*jqJrITe8X~Mt3HU!~zzcirPvJ9GLFm*?5exTgLl3F%jmsFqmp$>-AtSI&~XB
zufPfrW`*=+Jp`>jw${}}_mtw0G9o7;1j8t89uqz*?$VNv?f6S~20-)P?pV`+Zs_TZ
z^gk00m`Hhqw0wrP5D%SRxjiFilHP<liZ#%{#)BMs7|>s~dg6Da;wrT2s`hvd9WdLr
z!atj>x5^n9#7VNOR)g9JjW%~;jGQ8vIAjP&FFql=(<Xn}1X8x9y~<w!4UwVI`w#jv
z-g@EIC!y^qGP!Kszx=bKon4Ltww3d1lGKLB&Usr~TdPN)P#M8AOqiNCy!xqOxGhuT
zQzI$1=8Mq)@T?a@6aZZYsoNKf%;6Q-$VghbJ0)&r9Kwo(SNu9uK;VKw%St?uCJt&Q
zTGwn@L1IwKrTOWh(f0NR!E0xsm=<EeWz)d8$R<LpjF=~i&8Z4+`<2k*FPG$J{s5_p
zYzZq5!V}bNch)_EO{B)3(7(RoA;@eN^kBjU$|2xP3_xQ6cL1*E?-NqT4AnTTLK7*0
z#3la`XaN)&7IxfYb*?Qlyldr}MAL~(wFEt5ep*V(n{zu0(CB6;n4|K?J7|D%>M|)N
z!(&M#b@+Jb+~6PR#P1|mN$knZG2A)yW6N`6!1rxGuC}qLftT-5tJY=rB(*nR&yIU5
z3f7ZZAJ}us{F=oztB9*7Po6cuka6a;%av2^b+Y>o=v>(fT=gao<r}{<!F<5KYsue7
zd;>ZJ9bA}q&0j7EI&J#pTay|L$<_S#U%{b~GfqY4Z}%I@<Xngs>JPhhn4#s<!4#o+
znXR&lJjOwtTVW+nd6k*t-)GI03O>z=Go(AuK*jTOb;9l{74gHtIr^olVO&E&=$J1{
zu0D~Lc7IRz!`X!}`P9^*H*d7J*Oun1(p59Fqm;g(@mCwEEklbGx>Juk;a|PFP(Hw}
z{^IfZwd)@>c(+4$KAl=8gFcApT!VEnXfC6wuFlWLS9R-`p}F}GSd4Vv@(bazUKUkX
zOV38Cze_F-4i4TlmX(w1wsJt1a2gTj-f+6=cs!**ovJ^PvKn=XiVA#M5z6*F2lmTL
zh(Gvs;e_GakK}lh5h3^ZgzwC_KxaBbkEjaqZ1^?Pdf3*M@?)13SffgwKAf?=VwzLv
z{Pw=Em8{J;5s#&}G1eZM5U%>iC_%IC6KcptFQyJnVH>8!9*ngd&yLDA6Z`Y4nfqk=
zJ@*CO74A>Yf*bwa)jBbltelmX7nSTX+DH}85CZ*eOW(b_(&0P5wDdvyxD6WDEVI+s
z+iSFp8yp;j26(Mv(EJG+f8R`c4U6S~uW3$QJfStj8LZ<1zu>lRl;_NCZVH;%S6u>)
zwbD;AT&D_Fj4E5)KekUX$Zjla@r#FeI9SVCF5@yNJ;5&Vj8{O#KTXhLfXJ6!UHt(i
zk;PiGG2WOsU9;1<XAg?z+tT%7*&$Bzn^gn75|=Jrnw^iZ7+D<sFqIh|lgJSvJV-zm
zIEt^4Ivv`MuqPS9bW;YJ+BZY0FerOndlBz#Z$I5r@UOjn?<?w{r}h%=;9?pQ)YsP^
z>(=g=r^$rTFj|xj=LENozUKNN86YvdoiLP0Lq+wo*F;nE<$`Atf2iHsuL{@UYu?`L
z=^5Se6CB5Ad40fFk#T!`jV^>d4*srRaHEz(m2xYtR;Gi8n|oU?bjZ5Mv?jpZX&hS2
zPM0Z~m?S+<?#|R;TGs}BE`1pMJ_Xo23_C}toioc7pn<j_imzjRX}td2Hes?Y&RQcu
zVIjh^giP4bX)lw3+QOIsx1n~WJ5lW}`SB)!bi+n@`h42F<{*NNSu^&N*3Fxgn=A6a
zijFEgIkvAWZ*j}@?l^&~;Pdl2hDXO^N-R=cSNIv4(x_0mjZ^8t*bp<KNWygRl{7*B
zC{f~1iEF=P`tG)TRQ0beN>x~{#dAg8p#HjiJ$v~6gmbj$)M)U)G1tSQKH2BqbZW7m
zMJc7l`xQbzWH3i~m)Ht#-g8Rz!VGMzebT4gy1gF->!-~jM_J|FcPlJq{$Qz%lT(3c
z0T$~%Fr9a!T<e9#fDsP5J_X;04nCfywI4rbr0^mUxkAIF#3AnA6_|iiFFSnEhY>q|
zkyzvTq}ncRTcZc3_zfMTsBsj-5~N=Q?;umL!`Tl%e-mzRIp?yhYziSlxLdd9RY1!(
zX{r2E;Pd=|>abk?Cv$?<v0F7<mfZD5j+Kvgx%U=_;}e+;`5~J*M!mzL#D2v`4V1<<
zJgrn;{0SxduNx`RE3llXv$SG2t&B2I;cA<^hmCJ$7~TCG_@+#;7ei%LRa^dr!y_ZC
z-!c!xxA^({*Vn?1cKQd+&(A+Bn(&WxxFjM%!?6P}fLD-oK8c5ULre+nj#+~f^}Z@`
zQS@vP+XlBqB_!CmxfvH0f^o%U{TVc`?1BV}T#`QN@Y#MKrtI@#Y2=IRNzl^gi0;_;
zLgy{P%AG9LxiafXJBnDO(<Hy~sKnc-oM=<}Z49S7F}PiVz4qN+&E?uy)2&0OQm2*D
zqQ3CkK8o~BpP!0`^yz9h5`5*NFOe=T+?pP^&UFi08!(nv+)e4o5kn)R(EAX%JWZK-
zcyd#C<DWj&Ty2n+mL3}EiAf#$ND7d2ovBAK%57Xf1#QCMXO<Z>LF^=cN`CaFNbBk6
z&!WzAigeRkM5BFAqs66<j`^s@u{vJY@a<fkW#+rT)6J1_*g**{e;wsWHg004MA;kN
zjZ&6dnX6*vTO}+UqCnZ^YPxu6%Je$dyhc67ihp6nJX4hFBY*hl(>?6xprdi6bM*Qy
z1Xh;utg&=eQgn^t3xScdw_Gcw-=1q=3+~F%BTE!>&zR-8QSO!wzea4V{%XVP*RO9>
z*Zq$1zyGry8$LS=3swx1Ou%M!SFPBPpPZbWcvfa6$B86S*VI{_!8vC+gQx)icKp!L
zP+bE*KmVWIwBhBJ=N9D-#?4d?X8FxdjDC3(@d5RWobk!hqNI&uF%!O*;+1)M^QIVM
zBuB+L*cb9yfnO(;hiA^E!xKf$+(1vd5iVXL@u+fymYR01K4HM$*w80R@BLi6dJi{Y
zdNyX?M$fsyiL`}p?$+f3I}LQ*MGXV3Ec@w?9}QJ#ZHxSt-24w}uCG3StWhi^s$aCQ
zq2SMQ_^kdoFa_f+W_=l6oc}gZY)`}B;qFd(=;qCv)79^x`!{W4s+gFVXEJZ{=f<0a
z-4u1oNaq{c@{lbRN>?S0pUMJ)DT@B_5<&Fwb%WV~if&WuxaP|4*~gvxt&#mP>S4xB
zTaP841`9@!&mGE{)70rLA7YxK=+m2~d#h@yu=lW1Zw*)mCqDmNq5LG)95;#KHJYmO
z#?}ORKi-%<yq{i8B~sK7?M)xy9tCSjJn4>tgn~r#(|pX#SLs*h=ga0iM{ZFqDFm*M
z(zv#K`jk1#1C!Warts2IQR%KWP*YPc!*37<!QV1;n}LP{qZi(#LLV(oWgm5^zsh%L
zDE;e`CnNLo*XuqQH=w5qUPyfLZ@oPZM=SH2s>|u;QPd3nff9Sov=}IL+dd>0N|`de
z)S)d3G)P`?V8Pt7d3#@ve=8?<^p3Rf`G?(iv8-CPXp2%Kd7G>{{G0JhC&f?-hqzn`
z%*t`!e}DP$z@nTbnQ5EF_4HJxWeDX%FYn&H+OPk;+}&Ev0Qb>nBgb)hsw0O-)Ac3^
zZiKg{uIIF&s3Q3+>7z5ux#fK}-v@mq5Lj7xRfn?D6+XwPZ66jr!uCGo{G;CUI5R0$
zGOgAMtTKkX)5B+2L27NOc*lV{f~aV!Q?SO`(N&Xu{g~-O4?hDNW@V1g^>o~YPTQ4Z
zrO7e3o%>$VToloE6BH66+k^VTgP}l;4ll#$t*lOJ`@8h{MMYXO(PIC;yHl~+$jiq^
z#nS;kA_k`O9+j!7=AWN3dsIR>368S-Uq{LUU~C9oW9nt=Z+XM2o5mY3Y-&OkdQ6U;
zC1vDYKm}WBbMp;n58$MF?Gq<Xc;49G+*x9j<(H@hCi9+xbEZkSt1JId*}v$Nx-vI0
z@mccs%6v(cO<UArA5?RRlFzyDV_|9h@>3j+^>B;7*Z#0*=N}U_4%s46hk018oF^QX
zlYNF`qjVrM<Nh*Li>A%czk;i;p<?kj5%VWBjrlm&aPvti^ZfK$Qyj7hFQ6_vx7eil
zcH!m=E45frWfq~!$8ULteXo%@xCa#|lRLucwsFE$aOdaWjg6w+mR+9dj!XjotNyK9
zB|SWYqDfRV$ONL1W<4F9C-h=k4o&60t4<v}GyQK%p(osNdo(OXu@Po$WaP0xj`KjZ
zr$>j|@89iX>N<D2>OE}g{nPL66Q06h`bTEoxzuRCxfAtX>b;K!IrsAK4SRY1Sa3^>
zYU|#$kosR&)pBl9P^UkZhC^sbW(pL>eJ)4gp2FDWd1hq{i%M~oaGt)~X)-blC}I0i
z&e2Q?BamgJHMuB%??%Yj)D-u?NI$^2;&WAC?;m;*MkkU}tzcu5bt70MmcRa#U&*jn
z+uePK^zyFb4bp9$&^VZ8Qu0w2%goFS-R0BS=idLmZu#_73~6<cl9%p^w_qyhnWdhR
z&tp$0iJm;S|L9%0{mT1k?p|v-b(i_B(o>38@;`j#xO1H$6X#6+Y}QZCJ>eP^GJVQ+
z89ij<tGz#VYNNOg?<yn8d^xk6&tRrFp#Mj*+PzaWu`VuxVqyzl<2**j^6fAoC$zk|
zv|U||>HnN9JPeJr{`#ExkjdPSDr63zjg4sFw%=x8Si$F>jOfPIKAm1D6p@zV3ny*b
z>rzu=i{Uo~{Lx=z8H&MtdwKQ#$Xr%#ZfK2{!(~#c?os58Hx=?ftxidwKXp#mElU2#
zWUcQVroDUK-)%W}{qeg;nk;#u4)f~!ebN|YWo&Qq^4%Y3oapa5{=>&0xn#fpul0Ps
zZ9XnSw6zHH8dG*SVLhL~tNx+kHdz%lO>Mf%)r!UMc+5QrCw<rEtIt&Mh#@y|aS~gH
zE@t`ji+Shf<$1WdJ@X34afGSf2xK1)tE{T3AG{s50?qXYv5eb!JLYMgMjkL>Xi?d_
z!^+aK{^Xzk{5L3e+@wyw1(U%6US*pXK}S~20oYKz<kk0?6$uHwc?NG_UQsJxm)AEx
zIsK^Y9vDdW(cZUjU%e&FSP^{9$>%aPG<3<~`zu<JN^p-59c+JxTWO!rBN0b1jboUx
zw%etHHUR<Rd)Kbs`KJDpUE*GZr~MX=#=GhH)DgIyH<Obftv}N2c@_*;E#1tgXo9Di
zQ$e{6HA<hJFY9{X#t|C%$JpEFD;e&s*1sJVcDbymrNuIvKXXmJKV5b0QpbtTH>6@C
zvoV#N=u2OtX>M+QrEzfZ-;+2P5}Lvfv(!y?j>}EoMBA;I1O)|o`}nZ4vYtMD`oc$2
zV7575n)L#^8gog>^9oh8GDg-%@yDdcI{{A4o`#F`-DdRrsrgMN#QZ~eE{*wqGP{3k
zWnH=F6=MqQu8pN)XfMx~sa&ftWolh{Ri^qkqL0)+=Gu(=R7BO`uS&T3LwMpA4H*pB
z55U+n>Qc5FQjH+{K=<mW#0}^|ntU0Ck{n=XO0BFC4s{zwfRwOUC@kKmX$3BADpWdy
ztjZTR_0e=TcE%&eo?N46c>fMxY>yxxcn^6-Tyuz1maenZ-cdCnpt@g{J2!AEC|^fS
z!x$@|AAFkJ^<;DF6t*!e&QJyJ`_R3zOYk*e^S5rBR!Z!pqC{P_ly^Dph3PAe3jX_H
zFxhD#RG=E@k*M-=Z-&gDU(Q}M->>8bV}ZyQt~sEcAZkxUTA&Z0jz$a($rKpZ=otDm
z6VOoT;nH-5CYTkD9)2onS1G0FRm`+Fytmmlk_~g>?7yLeS+n%MS0=CJIP)cx=@qi<
zPgDC(#)TbDi4x7ZF`FIq411V=Z#NXel|h4?!Kdki!tPq^rw@v&C!obp9qLTK&oBO<
z!*o|6%?5vTOibr3Lru-6Pt$^f$&8}t{x`E^jX5Hxzk>O-lua0s>-7NVHiw8vEZ^@_
ziX72rin*dbN0>1tHA()9I<(K8o^x%lm3DZGK3QZHJ(s$kq+NWYnpAB1`g6JBZc0RA
zjzlQa`+FLFvGVzB-^?SXOr5Ty!Q^jZ=%WLYxtiSm!UHH*t~Kh~=vs6c4N>9UzqikD
zXT<Fr<izC-W(gKv`Gzk3jZFz$ud`xen%~aR+9s{7UG}V0F)-*Za~&S}5;+S!Y$0Yr
z_odp}+Le_Rn35CQRsOcDOtLTb1NA6!N~9yGquq6wTaUug<q|XW<D+Y6QVR{&m|qPq
z8lZZ3=Ip@&lgYPGFGMNvfA%`haBwwNlvtl>by$@9*U6sNJJh9_MMV9i=!$EmF~L!c
zLuI5T|CL4h+X1gyG(|T2BcX*k7|WI}_V@+1_;+=H&#X<|cSu}v>8o9n`}|--#lYd3
z`$TYRuLtBq^eibUA+MU@*xe0cnxT#hF6Tbs9;YRDVb&P@z&c<9JPJ)u7k%N^f!lEe
zb7IPR!_!M;WJ$gC;j;N7O)=Cd@3}0u&y4iCdCLz5Eh_g%rSt23rmkbdu)uauH?7`y
z(->P}%`o#}?_O4n)_6XAThyg3*8!>N-yjjUO|8Jg$$3FocozK4jwcg^9SxynGrQ@D
z-=DSLNU&y%fOaA9vW&y2%@W=CzR+G)w!h9fybdjQ|D7(6U&H}-DRvEyvqJ9~=@oo#
zX}jpWHj`TXK^I~<zoW0KXWAjubK}%S3jsYQDtd)~4Tq2t>u?<pL@_oDJ1f2H2z|aK
zae{5gM*B1u>0IUq(h0WwWNOaql#Ez@#aRqZat|Ij(Ea^;9*_Z0Eq{p>5D^sg@bs)7
zuN?cx$;~aU(=kCFODFO>T<@O#u+CJugW}Q~Uul~tz2{FKeYAV%{ViCi;DR5)gwzsi
zLW<s@S%vk6vJxE;^2Z__gVY*^lrL(wJ~E#D-gS`vS1a7s)-lKW7}byi^Ga!g_PBE1
zI&@B9tls-T1EzZlR}6+Y!?r0=nX1o3xerd)q62q0ABg&F`IO{EO-5BlW#e=jAip0a
zE%-#F9X?{8@%F}7pC(HG3*kW@c5#5T50aFJkQ%5X*Wfe$CIxDB?OIRI0~R;9d4~wE
z*K=~~&}mzbB%&l9aKnMrO_D#Q{l;4OT$5!R>HvgH3Y1LGf;Y$I{qzj&8VzWRpOWDn
zJpN(F_GPDrh{h-IxFX?|TU;?1tILq47K|m82)J#>_A6a9$=xr#&A+Gik)r$r)=@!0
zNIC8u{<0#_UoL_i9Bks^KR$6#>_IJ;9?ZKwd)c`};*xFU(&wKPC|{4_E8<=kw$j1?
zR67#at=w&+InZ_Ol_A*EL!7UabHm>kw0>K5rbE;4i1xfmsKcYr&zus24<5Af^V<t&
zfNB>3;ji{Q5dA~3^_3yjwML~vFl6_20^VrU+D?^;uad=85bFFrD960jX-BdagR%Qx
z6Gx?8Uo1T~yS3%pdyWFtt3to^GMf39fdKsI{95$X`E+pY!hYYH<aG7;*OYEQeWE1Z
zvi3$WfJTqPWQ62fD0b*3;I#ocxM76JA`tr00Z%NQJNQ;;rzR|ZJXW!U^M8+)1OAY`
z%&~#3sTvXcr0)}Ly!1$<2Zqu$__R6?S2n*T1-<TLL7I;nZ{}TY-8!-t<Kv4m1X2V!
zch3}SHf4t<J~5p+KRD6Lm&O;D<&xAmHA)?=!IU07AB=sbXmqxCN3($xSBmU{3RA8Q
z^1szI+^_o$Z>Su{FPNgmPCDs^JHoR@<lHgs7hEsyCL8#K1yUVo*l{&yg!5T+G{xDo
z<D=du=@m+xR!p}eS6{@<QhpR?9ynG!-Svf*ertwOdq2J7R?IUjYbz~1^hyEzga=xS
zd9JcV&E+(5?(ohv5u}Y`i@1m7P)}>^L_-uYwgKq<=M=s7VkD%bm<<IqSpRhS4bBr`
zg`-5d=?a_E`Ju6Wl&H`A%Q|#jquWl5nlh|HExy{+$ZMx>?;ccdu^t4C$#FV?DSRrn
zDfi&?XwW`ojP!ZZsSWX%{TG7Y#7#MSC0jL@BpFl#uqCuK9Pqyy0=IMk&d&4S^z8Hi
zJ7&dUxu|+9*gEugnGkjdySU&JJ&7%6)p8*K^XYPWfuQN;(XHid0Yx2@FAvF)UvP_X
zlhTE-C35^!ExISk#Dx&poQE2rPhXx-Pn<S#z-DJA<1SLe9#Ww6r`FE(686qG$Zf62
z<B6MtHfUFdHk5!M9HKvK(rQy!YTBLQJ#>?)HC4p<$FakQTiV);a5i{oWe>qCKR<u^
z`+cs97hQo%g|WHGBF;(K*|s(|oCTV_4Nx0;dU)J;ZC(%MudAhqNGTZpU)7N=2I)^g
z;E^LMb>gMkx626+<Ht4o;vU^zh3P^tk7e~xv@pEQgQkF8qUA+Jt-wVukHACNn5d{M
z6G?c#32)nk)EPybZut30-h4vBGms^zSakIpJq%46J=Tlr(5ypu>d-2oPn+nr;KD`#
zANbbmB4ozxTX0~Om6uPX)S=DLP$kyJ!jV&^PY+V9#YcRI{(Z;~pc2e~*G&0=2^Z^2
ze{J*%H5ZbRJK$kt9(1a&eS|#!AS0_C_cJpyGc@eU(arrF_Vee@#id^}GZSTR-ng5X
zFa~79oKNunEr_OKvD#=fv&>nUr)9iJhGi~h(E3oCy0xth#z);bfAixLB}K)5!*>aF
zj*isz;+RW`?66q^R2D38aTF-2J~lXZfrdj;_7%NlOfR0A-S?tENfckc%8DrwJcnT^
zlrCLo>&`<nYoSed+^B$-$gab$WIQJEN08T3?-RbeB7p-(_9w#=OXIZtdW(fSC2JUo
zTUQ5qK7KrSX+OLDlh68kdcX`)YXC+>d)yvsKK!ums<X56dcz}fCPs-{yqK=R^zw4C
z6EJg%GBC5Qc5!hLRD+?ik`iIx-#b!E-Bg$xHHoW?A_MYK@dSDJ0o19ZOHae-Rwzgz
zzouxavzs37Pc_^H;jJilTz3_Hx^(N9?#2UUqo8U|SKPAEL!9AhSI|NrmbQAo=|{d_
z5K<m!4JM5`Ciw3vyR?hVR&SoIue9wOxg}QVG6g6C4^j8@^yDy@Wb$ink~4Gz0am*V
z^n;K(`>g__Qgvyys5?;BXboFzzU(Z!v>PFb@`i#n1(Vp!QMe5zT5hLZ2N<r>8|VBA
z`ubu__!e&X)m7xv8K~&}=Z^2&Pw&4K1d2vNzdib*g+`=GG`+trNd-X;l=17nLkO#7
zc$1hlXfymmdN(~x4~%`CpKj@tUt--R{!0xL!A;;!L7(*VEG#<7Rc@n4f*0TszK6V#
z5gqyQ<HvP1wn6=zwAeh^8u8%4e(`&A&!*OqGF=;_4%JKlgBcv~pX~ALZ&m|3f9Nc+
zvTbj>h`d};Ik}YzvQ_PeXU`~n%A5zJJ0g!BIWl2zQFU%+=E)<N6Q;L;0TxYCwV70~
zBp(b-{fVl&#?=;<V232qKRi^q!RLp|zNGADETat-F|ktxiG=oGqCgOyGX2)#(W8*T
zFMh>9Jn9&r{nwtd;*$tPguBkMPD&J=_1_D%;TrHy)85t#Tn*NAPmh2EpkpDwq_DKO
zsF-l^;>8>eeI1<^omLoz&~l2q)b!QAcY8v8eSEa8R*elv?dB@nf)U|s+~p$}6Y5<G
zg)rMsNyBf!Qitz%LiR|7+80Vs@iAYN#l*YBDql{q5Wh9dDc30KQvK11qnDyu-82fN
zIS@te9u`*{iQJDU_-r-|=vP6X#ki9tdT#%we&BHRbeLgL<&Ov3?yU5bIUihp6IpXX
zX!M4h7_*dA;>hI0#C2?fAT;j(rGmWWPG+bG*z%D?xaXe8)nC0ReI8KDqIWMh!0PqZ
z+gr(ix<N0KHX(;KY-@ccD+jpODx-F~i%Sf_6$;|G{ymuLjS5KelRUb)l4ypuqs1;d
z6q7q86sYWf^uD1jhRR5E2)dDV+-8{9(wxL5l52i-@>F#Gay6o`2ABHt*@jh*g~0mJ
z%dO2KhbdrUT6TK6g8ju{g&A!vEyeKSd>xsaACF1eTU(>ZA!g6Jz{5K`jT7YLnYj%r
zGv!yu*1nxNbB5QX^32YWm))m41^rqv+Jbn7rrq-P4T5r7VK{dkdi}PP8dH`3_}co?
zSxA)~x((=G3MQWkT*{agV(XFhA??ex=n){|?qSXQ@iZ2v=yh*B-FCJI;@g(WzZM@*
zl?SH_SP=97cVlAg4%H^dXI>>L?7>gAWy%E)lH{bM7cW$O@*HK@4KsD3Mg)vXj**u-
z4-~A7Jbrw*q6av-xR>qlC<iY6>enhq7^^=1t%LX}B>9TCBTVi4n}&ArHhGsG{{X7@
zwe54<z_;v`H$ge}#S830LY_l)!Ta$*6*VN^Q&9?km($Y)4=SmeD=I2l<zOzhWP4Ya
zzONgM2Q~f$9)dhg-1qO_$CG?#&wA!Vns9RZjjbA9Wu43(%<9o%2l{wt`-}_0ni9c>
zE*=EVN6h;iPSxUto#}wY+qS;^YfsL5yr;ltz4u6N;wGV&{QRh3(H_zNgu5K7+WW*_
zyCec`Dd%RHOqM;63^}EzsHv~<-1`<U#3LY3`0)+!AcUTo8HZxi1HY<Arl(cNA#X3Z
zz-xCqjg_S(tm8F<6rJ^sYJXoXwrY;GvAX*DxP*i+Z=pkp+Aid$yopxaM_+Zzcqogt
zwP-0PU3GAx+x@JfT}~YuVdhR7ly+($<m|`<O5&soA+f9E6e@xia6WIv?12s~_Ihe{
zHw;U^OyL;kB`-Znp5Lc*+1Xz2;W36X?WL@5H2Ikd3peMcrj#{cD(3z|t*VB`M&+ES
z@USpl|B|e%$0qzS+{=rL??1rs)a<QiwgUL>H`bTsp%?Ls$5BF^PC@9fCr`HLcDBQl
zig7p#(lyX{s6a(uB}>ZZdY(^{D&0j&Qx`>L>5r4eh=x3$c`90QJ7pZK0iRa~#E;E7
zLv8w{St5=ZDOqaqRvpy&V?c{1b>4sWcMlvX{#5i=rzYeclK;r|Ztvvp*XsI5#dYs&
z@+~#leYSO&<Q&JbuW_P>KI`~%b8tK=I47=?Ep7`K;I>4^l895^W9#F`j-5%hIVmC8
z4T?aqjEe8d%z9VNb`a$qgoWEh@Ejjf5oVYKK@oQP07_en;&-u*L|blGTr|{FmT&AZ
z!nurjrW`EE(%ewv657Jm9%Da?$mNRVi2D;JUSAXO-{qn@Yx0fW1lDOvG)2ZOT<|2c
zD+C{m0NT3Ul|4ciwD4svCMG6HA=_5glpC9ZS~62VKRmL3;kPuV79Tw|H`jbpf$qY;
z|43#7km)5mbb|h6cN)w+tRZI*|NoNO9qNu=dg<9i98K;cpX(3#Pd|BbY!7;KzRUko
zwTXE7H3MyUFG__EJ&>4nmX>yC5GEqp-4MHQ;VE@RrYV1)xaZV<jrix!+g&laxje6z
zqV-Kqjo#b^rb<SzA_u2~v5Vzc2%xyip@}(b1Q;bIvSJF1haW-YJ~-VQ$O-8B2&QCh
z)v9M7P8ic_o>qtNGA+x5ON0d1=w=2-+<yvygdcn2&lZ>!F}9>a4+t2ip1KW#nm?}@
z(r%5VA6y#Tgb7E%y4ONOLqX(M`mk`|z=6f-`*{}nFw+V@93qmstwl#ir^i{0#$Kz9
z1T<W@A()jkvJNj&jmzCeUmKm4Ss$zgiellzxz3c|0mlOEeP|NQCpBM`<8mP6AOFGO
z>hZ|+<X(8(`p!Kk<|tC|zV$$olK)%a%pQM#rWDn&QyKzM1L7#SK?a;{TV}U@4Q5T8
zLzciZMri7HZAH97oqoJaP*tz<zQJkU%@oF_=4Oo_LgLon=+UETxw#TfeWE^xi*u@j
zu-Nlbx#{V};B8nioR8Lbr%*$o4kh`GCwdi_)|jluy|HRy@mqo36KQE97Yzk3Uv`HC
zA)_bc=;(O<Jhr`EhXg^0C#jqCO+i~mitZ@+_!oI2NP0WH?v(c7Wk8A4+dtLgsKkSN
z*q!_#_-3kfRphcAAPHzV*O*;V8c!7Y;CL#MqueHG<`6CmRQCIZQOaY>1j7ESWa-6@
z$3mzX7}OLr+UI2t9eM<8_Qk8C11v0~FyCs5-A6&8ez5lb<cL#$1q_aqq7G#gb-7ZY
zDY!5X9y7Du{Ah;vTW&)I(lDf%6DRT)vDV+xC{d?4P{G)WTx1;)LSk^@Q2sGk<uICD
zF{nl>W@~f-#vtw;g+sPdnc>0@qvuHh(q=0m*mWdt@~u!a{)I}o>UlUo@D6)&a&md|
z#A#~k_qn%~7(9s@?sUMdW*5O^t^}y2Y61ftt?w;pawcvf!$a{#u&$Bc%R!{b%gaj$
zdANh`=tv)4oZ|i;d>1y5YL$V;<pw<zyG5YD76g}6R#w8yp^)fbh<oHy!=MecX&Ujp
z3}fmSBB<!`EtMFIdYGrp)gUOJhPe7v{kOGVAp`-s7#e)-?p7_Dm6OvD;<#W<Jl6cx
zcBJsfU1TEcMPS5KYw9+(-}s1L@54Xhx2(|JR{WcR64JffQ`Z^e$JZ{F*ibsLv9VpY
z`nonQvqNSSu-fmn`YTrW(j|(m)bOte{!IDS40E2crG23_s!vHtQ~53_FQwo!pv6AK
z`7A1JBx_;i+yhZ@%Vz*3`p`&>6ZLWCYOA1`9fg~|ejjM{{^;y1Q91-x6A%9dX9>{?
zf_#*jX6p0;|Kg?gut>QKUa_&U@mK#i;KRq44@1zUiZ1#<aBI!*^E`kD(+Z7%$rEd@
z*}eu6hGFVSU0>y~yf^(0pHPZnSzqrUB@Grtu%EGC($)X+jY0Jk8_G0VG^CPv15UUY
zs;fcFuvsgk=a%I&<XCKjMzI^9wwrn`UGt5I9Wx7+E>>G&dha8A{``4OXi7d>oPRC1
zurQnwbhTR2A1w6RW_v3m9D%F%u#d5FXGANkq7mw6m*PWC;`EuqL#PCMNq-Vl1VxH4
zG;r~6K>t;=5wyK%_dK#Z*1%Fj^@H^(X~hMeJallZ^4RqEO3(3ciS4e2Iy(Oopm9pq
z+q=aAz*5%~=@r699Kk^i(0Zd~JUI|%tKJ0w9=;U=i_{PY6mN0j@ULHL<AWjsByO{#
zj-*+#iin}=|CMq*O?E5~xL)|SW1af3ml{NKyt&#Qn*ltH(R1XYbj8JJWrP-KW$=*J
zhH%QLy<RL}tD-8NLtu`ay)Uv|fR>SwUyoA3-P6;*veX0oXKophk=l;qYa1JFVsMWZ
zhwsR@RQ)O)2;kx3TF)gDo?3KF6$w<s#}6eAC(C+KP1A|gCw$Tfwekwk-~trCn-`D_
zS>cj&$W=$$3s(`99fJIOa$|5zu|+hh#DPz$0~<A&3LK~M{mrkN9fEYyJI;M&_|tQh
zWo2tHT2XD`n{8nnJ8|K0LRVj^lM`8jFcIF%IN&*6M@_<b@^hivaMHaW=L>_t!}Mh2
zE5cUXQg?BA)uTm(_2jQ9Er0uVz^N$Aiq~b>dE>l2kdpfmaBC^z_O!BXGC24Vr0H8&
zJ4_CrrQKG{$8^>Fu=<kaIv|I*mE&d{*f1aK*<Of7gtP;rffhCi9L?SuuEyGI>L<yV
zF%q_gY#C=Cl{k^LNM>rhQAmC#>_f1@3lYIHBlS^23>tKaQ9kk03|%}U6^>szQ}vK0
z8_tbr&Pm&9KOzWWBpbt`yigm2J#ra;0w@|JI_;uASU@ZiOoZT2)_2(@N{1AlFum!z
zkSwTh_tS*;lcY}*OtojK!n+Eu9lRSqt}1OrAag)e;a3y>9A?}3$C+kERIRPYH!%3|
z(a}FPBhuk<q@Z3NNU&IOe>EoNyP^@(OWApA?g!sFagY{2`Cf*2J)$Zh@z!w1Ot<(S
zuI+2Pdf!2uO@MK_>N>ap5cqSy3b;PVe@KLYQ`hw-O38tuXNt;S-i7c*fcNAOxMez!
zlF<BBl4wpzloFhT-O78%DfHk$b_2r)Hc9TVk6BwV#tod0=YpiceF|AFHLfH4^-q-l
zWNW%hMnnYKaeO{r9S-iqfHh?5Y-KL(>PmnQBo-P0W;upP+2yYsXYwriQUfILUjw}l
z;qn@1V@H4r=S2RW;6XWsq&L#<zh%Tnd-}PB<=Is4Z;3ZUh1U;qY;G<!CrWs+2bI{i
zS%IC<PDIRyx9Q!xf4)KOXz9q&Vz_GGnO8oSPKuUUWAs&}QCz0dCZ<>*RoEg5OtMEf
z7k`P6L*|t^i;580)<>*Z%DbdA#iU71*4o;GzCJ<uZ%B1!wvqCeXq3Gd>D4paB{|+q
zQ#2ZH11id^q9QOhbjfTlke?*<dkq)^dpDe~qt)Bj-=Ez}DUwHYFfcefw+C5G>Aj(2
zGHV3<ojW%0qd`HMkW&SH1R;n)(1Z$S#KqUeW1F5FxrblWfefBpo(6U?lg@&;>(tB-
zl(>DNsYyt7=|;uvAaobb(9G;7P);zSd>+ie_g;e4yS=j$AwTSU&6Zx5=H>Ag*<8Q=
zeCkb^yN|H2aLjKVb2kfXzqZlQiGSq#V65CDn6F3B5RHxq4gK=U?8EkYb0GXQL)>$j
zz-@t=&Bf0p(SZ|sXmm<$u8PFBt}OKmx6#H5&uN?eJK%iXja1IXHrwEmYe2vx+z&{V
zHDK6#jD_+Cfh@<5?+qp$Fhim1frd+e32kc1#Mx1A&85?)Un$UezJ*YinK=%|_U+}u
zz~IT3KotICLP_M-($>yiq8yX-Tf6W*cms@QqE}DziHlcAckLpS<KM3u?jXmVbjYQ>
z2zQm^Y=mtugW^aF2dw^!%=Wu7cVF!^ls+4(m|YZ@_#CQgBy#-Td9}Z}(P&0svRXPK
zetdB>cc$vv_r}IX7;XLW$=Rjkoz3yH(JKH^4UCN1wjZjcGl;s}84F5<5^N+hIe_>&
zXHV~P<>B3{>w@V-H|iM1+`of;bb?E)XArq!tXh>%;d>6C4iR=Kc7++Qp!E$D8!NrI
zH}`QauBKD;(cxV(Hh+nq(C{xW*}4Re8R1?Rmz9-4o5@!8q^1Osx%TX~%f&drXx`wA
zS~ZS*vw#CK*r^9Mi85&G=%}A7Bkp+qOTur*zu`C3nGT>dM$=EBrp5fDR`Lo#fg{r7
zcpZkC4Z*^Q8pV0f3#`LmA!naCt4Au~40{P^zx36_zf7E#)`>?Cw^O~39NjVB7JB%Q
ziOk3doYB3ltuNsZKXLf*%lwfAo_|3P*U3)nfttP?jJ%oY=|a(kI6%Xo-Nwbmt<7iW
z<@vqrB9(1_e8@m;tEos(oH~{3U2p@C!1lv|Xn3m4Yy|UW=w(=Q#s^DV7e~J&2-)iy
ziP*w)_`Un~(FJlvvhNCN!lI~Y?M7Cl(t!w)J=?e+4qYvtdZz%Pda~SxY?k}*G|Qh~
zlZsVzB@KxE`3FcT8HdqgFY;pe?TuM>o}4r|ME=p15&Ty+0KnKZ0@i*$@)T(D*dlr!
zJsK+P*wfCgB?%*Vb+xqKlTb-_DtY@=LoZ0*0Bmw%5YPqYc#7;N*y&OFs>~>*i(OHo
zA)zg*Iydy-dINQCkXOdmcg87vtZe9^Yj2r12;ZJAe~L~^D=P)mC;Oya{qEH8a^or=
zT}3z8<7N65@*_v4x(ZDlS^Ou{V*#ak!f@i>c-Q|N=0_>w$p->K43+z7+K5vnZUmy=
zeUGi9moOAJ*f1sEAI4JJ2V=`bQ)uRbMCy$wgAt_+QK(?eIzH}h1@WBaIk6gc##Tv%
zdWF2&PrX2ZaUGi+9bLY&!`koye*K+|z#WZz#T^I}>hXeq+!52wc%mwkt}$8iN?vSu
zSJw;a9&<VXIylS%F*dpT!DkXj#}=U;$Lf08IqW2A-#)CZnogft_i;9)MDWiD|D(iF
zmZP)-RoW%H3aH*3WO`d@;xozCI{ZH9DAUjpC?sLv?d@O1UF{*e3-<RI;cEGOxaV$P
zvnm=Z4@l%At|i3A(v&acm;O*zR`w|i?W^=|!{foSgCT{iC_Z8feZDnU<WzyGh3w!-
z`Mlz&?D~?JCUB4l&oShg3M3<;BTh7#J~Xy;#|0IOu|m9=_pr#*h&nXt5W+?wAqT(l
zw-PIDTj#)7q}wD-vmL*QfAHWzx%)W3QAu72^r4|Fb?CknxV;IkqPY0@hYugN-cu-6
zOWB_b@b;oV(Zkr-cvtq3?53|5NrOTy4dw6euf|Vx_N<tB-Q&Q`VWz)d&XBe|5KXp1
ziG)|Uy#ExDFdJ*DRY=p7Q5GtR2NWpN4Up_9Y!mdZA%rN@lmlTN?|dT^7dk)oB2`=D
zGm51p<m$H)qyAKqrJDm97Nco4NzWd>;*_4xRE=8%*VG+MPP3k#9_^Vkn=f0TBUNrm
zpTz&FN}H4eS3{f?Z`~3uS7<Y!=TPT0#7)%fZ0&+zNPmK;X--R)F~|>0g&ZstsGryN
zva*tCfTcy0vd3Zj>qnYgdG0xe`csk$a3-sl;-8VYoDkKHXQDZ8d&ub(UeJHIe=mw&
z;s5x0@3^M2HfnV2V#5NcNKvYE1?f#u2u&iLP)DWrj)XQUHW)#=NE1Q{H5!_T0TEG(
zfT2oPItfLPfb{zu(3yF^@80{*`}%YK@Z_AmpS_;-thKgtzYNmk#CQKwvU_SYc(aYK
zRY%8|wSL$7UJU5y_IrFsnzy(2eBS@tSm~yP*XqTi4#(lXk?82)@O$&~p0C%Rp1%p%
zX}2zJ+-6pIwuv5Kx|%AD9YE&99ASsl!&*K|yD{8IzwK3}jR%e%TB=I^hxO9S^<oL;
z+7Cf634%ga1!OE=NGLZsH0=_47!_oC+Kr*>x@n{K@TYkDFoak0efX;r#=l`?N`CEn
z=JJo;*Is7TvUCRyFp%PMm}*jv9C(xks<n}i;N_C@o-|@1RB3p4c(5dbLL2yqkr9v~
z*0=$oFgl`ZK34Oq3T-ZqqVq=c_3gUHMXYSu4hn^c*t|)bpi*;-#{nw1vhbRg8x-z{
z{Za<XzNHJ=+jBxgH5voIh8}-Zj=^C5nT`ZpV9AH(0NPpx7V-W9K`V~1?ID{@E|5g#
z+vC%(7PB~I!e7i6kChIt`dCO_?0#d`2+ItpLCu88ePJ99ovJKe-uvkLkKg6u)nFm%
z2S%@&YJ>1Sy(q7u{MUN(`)K}SpRkR6vdIX&3mLhH-0U@j0!&W95pc~x8-5@QC_nT9
zmI#&W>TllDr=I;QjdO`2QI)<ZDkgR_@)uej=bSzr@Eqxiq2T-@eBf#sCB}T;@NvPc
zY3Q>AP`MGr8;WjxVs2r4@Z|I1v&364(u`uQ+WeVoYmO?uHQHO?%(m~>6KeYcPcR=N
zfVzNIuBkSGtoR5SiP;f+u`NWmcDWB!1muW-Anv(M%{fwGp%@z!OUzUc#>nXVIr&W-
z083vbtfMme5v8k2yRSX)>$slK@nEUtU)<>vE;%)pm@CVx2DZ8t%i|l<Lb)(<&4Ufg
zw`nn9?Vi546Uq0#n2C;%dhMJKV}crPIW8$4ZspBiyUDOn`(w7saDC6?Q<WAXb!54i
zQg-*DIiL}1uR;Di`_o&nAAEg%(;J^YJ<%?p_kWFxeCFrp)9+v#EB&@)A+Yw(NEtUz
zKvwP}WV(Q?1lHbx*xenwP<CI;>d2cMqQx6SXvAx&BC5!IZ<C!b@}b}Se?7klIAm8H
z(}r#B?T@-In5rm`xPjv34I6@RB=a{1=i|82@hRCcei!ymIiHzW1wF!KG0Xri-3Myk
z^Gd#QA(2>{26jSwf|&1|<<jJ58ntuiTYZc`d-59?%)K@Se4V>BJ^kkxm_Kg}qpe(@
z8K(M35EMUJ9_b~C_@C%JdGciY(_J|oX^<zL{;}I2C*}Oj-O0s;g}4VYGBUSbc~K3D
zc^^D@(Au8-!YgEJ@J{~VLSEg&XI0;f&f324{JAafN7uo#pxk?e8wjFateM$_6B_U(
zfHFnq$(X_MhNGy6k{PF&FHPwn_5XTS)bnf8`2pFcr~GaVm`Z06DGG)8Wl+!oj<Ipz
zGHoJwjO!7$LE%tnGC@*-+O%UaR7FLlg{Yydtn6h1I^OvChXcFpMU<0rfnJz*?IK#;
zdIipnVqVmhE1%m7%t`h9TM`VIW&#fO;cIO)LD@z4R*_jhT!Jw+qJn^BZ`(BXVCJm!
zK1<LTg*+r;2ggrsL;*iDKzxQ%O(En+hX`T>-P%6s4W2#9%$yJ#>$&d#VM_KR<2Dq!
zPCdR!qGl7d&#Mt_p;X{sdjt583r*;3Y1zJ4S66q0pyNR9?KN!~0?5}Mcu+S>tERA^
zpkV$Q8ck2{(6dd9iz~Rmd+5VKgWo!L1#bi+BS;M)5RrnBVed0&OAyJ&GbKwc_n+FW
z$44Kglpw-4++}*d#af>u;~=z3>Foj_`~kCsuI#aN6Pri4>fKW;vfW4Gr5g{j^g&3-
zSz=HAGn_t$zB}xW*jRI0<4vw<{N~HE2Ze1<w4cYWzFxg#78(+glb2^o!WO^bEp!fr
ziEfKSVPPRWd(lcHfgmDZR8sPx`rs)EynI)4GktRyt8A>=mY=69?eT3K;9~wUK+n5S
z$3^BrihFst>L#25_8G%oyAyT2DK0nGFLEq{KMej#8$CFla3A(jk>BwjAg*|Du>a9v
zRGht^)272<|29<hyBn>oQBU?C7oF*szx1mkwR3y<ehZj&pAh6{O00C@%zIa$$H#u^
zRPORF21Z75-H{_lJl{7o?5962WSqXRKa+IDra8Hh>(`q24X^`rgs_G$ACl~M2vS|N
zfqZ}62IJt_!!UG-gAi>QX%P{$c@iW5-g=WnSpvJ}`@-G3cmH!DI2~Zg&xoG*8__1n
zd+^9hOCO=@KRz^>4y-j)2KoBiAp=$9f|Ph4N1K)f($vGRXYC3l!a`V;s>r1leY@cm
zI619Od_?BoAJ@KK07(W~<ING*ws&&x14==C!pXS=#uHg+hsVPBHR3H<oIY);x>L@D
zc65NqI$l=;cTgSxi4uK{fx4$DGJmlLo7F9Q5Vu?tX06J_v(vzao8uzyHh4vD1m-Wl
zUjbPK)KNP1o+`i#<(k4~g$aSiXM%}<o98lvq4D#;U-NGGN|hOxkm0Hda1j=C7N3|i
zV4v7FSi_d?aM5?0_t>94)7eq|;LBrnM62}L7pEvi`4)JXj-D_&o|JQ_#ZNt+E6jX5
zijhNTYp51ZPV*}2UihP?QzI$mV;E}`yNHlb1_$=#^JC)Rn_}rDTJ2vqt_$C@dv}Y{
z;(Wus-%?!(iLibXcxHq_N7zo-#akuy@1=eqa*)3UoP_6<TGr7Qk9w1rTw=K${-)%K
z)bi|}4=DjRLTaT$J|%xEam+?!J%@L)JeE2f7a#ARefk)JU5vdiBRa(m+r-Vzhn=K6
z(@VO@d7622225_?^Zwoj3F`0palkSO7#5^|`+4Wi_NBVh0$uGbEyT36JBOf3=>%n;
zSRVUlPdvs^GQ-tb^6`2|yBVaI5D^)9;wa1ASFiNoUsPwA`Vqs}l3(bX<;^qIKcH1T
z?Na&rAWMc5RSi55*U8W(aCz?yC@kxjB{;1-N2YZieyt*vFUC6?mJ?FFYlB~Ed04-`
zi!+f};xvd|k(cKJi+tqYrWGJms0W#~t^OXno5R28lIpadkz=3YFTXP-T0SZ$m<Iju
z_;r8(H<YEKMtP@mh&Od>ezPS4_=T)Dxtb@+*_P(uggqR6&A8{ob6dzi(3`(<P{h;K
z^_BM&^DhMzNY#Z9RGZc(QAM8H#|@w9>v%H{=n<xr-;z)2g0%Zo(dN5`xtb%78we2!
z+waoc!TsNBJrb)aGI1B@WPI{WN|ClZ;$60Uf3Ez|P+}E%MA4ViL{fu=HYa!M`=2Z@
z-kKe|4t*U$-+?M5Ms<t92{48O9meHYELOg92Mn&i^PB5HPkLZeD^FBTP~Vdc;$oq4
z=s*##wmMWNVJPv@OnBZf@0lCx6CZh0%R4^g4~Y7g*GN1&f=Gx=fiUOB9BCB@xtIJ9
zSJWym)xBQLnk4&lVch=q-ms>mk+5;WAKOs*`CMTf5fN?}4Bvb0u3obLvuy%nwm8aZ
zWL5`@$z`C-$af`r)HfPL*}p`uT`=+D1F0mO|95e<FcZ(tR$E#8__-s|I#WHYDRK0^
zDhunILNwZ86U2qQ=`tO}8wKd;r-Mmp>M+0Cywf9`nar~@7GA|HpFdg;yRN7WB}IV9
z$xv_+tb>2vB(cvfg#8^7(%huARQUberiA&|m@K?1tyDzMAh}KMw{$XEsQa&cxS!<%
ztQ<ULo*Ys?w7JbZtCvaO3p(uT*-NkWURy0a#l(N%(F6!MgFe?E_8k*{gd<cwHw~DN
z@srL;l+7{nWY~rp@Vh){`ykx=^n3E{0H+F&6O*4kAI7hD{IQdTHX9-JEGp^_gL&O+
zpJm&?U(zNx@y*jKk28<{(q3<ouAAQ$mhdli^YRs=`O(G%M4*e1TiQU#U)9i1Qo4N8
zr(F4I!o9yhOM+>IrN(~a!Tj3mXPc*eew?Miw*lFJc}8W{V~jF<K6{<)zJp;|^+AcO
z!i;4H=yDN&DH9a-QJs|?xa+_2ua2_~)Q*qe+Q~w~#z^rqVrr$k@xMB(vyxkqjA$W#
zLG=xb6rIc-pRg3gl-p2Y9D83ZW^wm_Ju7q#jmVioo2ITtpu10$+F8v{Q+LC82acs|
z??*fl6*~F`Xos&FvcV-6E_PRxva+lV>LQI2$KB<wFy`xuzG)42%GhdwnEML)Z2Um-
zSuP2QZ{sfkkU+)84vn4!o%Pi6kjSPWg8hNAh`!#iU9zmtZz(fiR+pEyf%59{3>v(w
zyr5pae*NWiIp^X19^5FKX4lJa$u9T1X;kh_nJNFHd8+{9favdP90AJbb;7S<))E?H
zWP{aDbyQZ8Zpt2WWtgC4w1apZ`sSyib-OT9%{Rd_4I+3m@6^3g%pm5NyhX84#sJ3J
z7qKo3Y(aqO?_w(8a{RN!&VwFfhX`!VTkD^wydfMqnwu9)_&SbB55Z;Ga-&`65#OTc
zCSvydp|WA6K|12<`OfGOu1E!O+`+~x0qP;4^zoC${&UX*l<e~TK++wMj*#{IkMzJX
zBGsLpMAt$S^KkX7{iA25((7WTeVO318M-j$le;+<%HJ(4E7yGFMFt}bj5hm`7Ia-0
znW2Ht`us1_Y0F;oFNC%uj*rZRh&_Ao+Y$oLZUE0dS+A#GuZQ$`edW9lajLg&<vm!r
zq_H2KzeQTCuZn!6dh+m~Eiff|%p0)aW=i3x75JtbK+;JMIgHBw{HVk-QmVWA#M1oD
zDCeX`zeR)&<qj*rjD$K$?%R%9TLVp3S(#l;CvkHUFyeu#BBKO#wpgymzYoU_>t9j;
z8><XW^)Pc7Gk_%-T<&l(h-|10j=y(n+AAgVJiO7+Ovr8tTcr)vtwH|L1dd@m8hh0)
zCJS2D=Od{K>lX)Y7llW`!mo7yP_HyLx*GH?io)Nocp9#pbfI^DLBD_RLzSaP4<*UP
zc!S3OP4t9;uY!=VzW2}@__Ow=YboFWQ@-F?k><|-l$K>%!Dobu=m|)PP#ZjZK3s~~
z?Q~P?(CHgiWk4&P^0|WpXX@Qevu}!zDD|^9s%J6Xi3-YN##3V2Zt_Y2>5<ACIF>wc
z{~3>{O&(rWJ2lj-a6aQ;^74EUn}qEcq!eVK-ISR`=1t2#xIA!s0^`u5Cr$)_Y)jYI
zZp)bh{1h1_XO8F;!Fpz($I$yahT!|3{SpB~?EY3%)f<gnz|0~@<Lh`w<GM_PKa{^;
zreATTM>y7=_#(d`StXX34y;a|mx$K`{wwMc9t^2g-@#WYl(RC%6i`{hCUI*FJbJZ%
zXBTvbzUjyZDp^YU{+?aRygcA)Qg7IaavRICVj(@0Qx8KHTX=O^oJ*e!?mtiTFQ>O<
z3UEYBHA^r>2;uOABXDs18)sc@)m=C9<URTM3*ZX4_?8R@0C4)P3dvAN0{i;rLeU7l
zWm1{(<>>@$acp#ExWSlj_Cqy{0l{d;Mzzo1-9fK*O!QX4{EI8BB@jIn==w~$(W&kR
zfks{604rWn(!0BhHC)`>9KKnt4bH=74JNa6GVQ}Sezve-Ho5ubaUo-Kx24hi|Jfw$
zFO^4F(_xw>iojU&3jY%b)Ox*TX@^$OKKMNu*axq*Pb?EV!%AWo=pcEe)G_m^LJr=H
zBUCA?zK>$j5>$S#=#L8b2iI!u)8t<p&r_`)`-`<Xub#-lwD!f?`<>Z@8N$MVN{5Hv
zuddF+5FR<XdduEKt5uO3YHETweq!MuA@kO!jl{Y1ubTaMp<MQxEV=m-<h5?+i@Y%i
zTb!x5E|<9F8<=8!^u=hVbAb7U4r=hMDAeO$?x;<zd{nsN6A9}yQoJ!K4c)H!unh&)
zKDnK`xAl!j=)m+>MJBv9<^Ws9UpR0akQdO3LbLrZ0txYUFLE7r9zDtemz~}k)m#Ss
z$fmX$k?E|h{PKX5{W5lg)1nxE2Ny)12NRvh5yacZ9j2mO!>OPUQ6iDrGY_a^D<8ov
zx?}HDUv-X&4FbG2lMvE-->fUkALZLcU!|vQfNuB-y~rGRWYTRU8@2^FsvLlJ_f!1D
zA-G=YMQ5+z&aNpYXv8(`?OSGS#GcHBY)V5Boy}Up=3^-^g|;02>wa0|mapue@(9fF
zVJ6aS`Dj@{!oLNQJwaJH%7cNgJT)_VqWN8QR~H}rPUJY-Kc>c#5g=DZR<OW$0q%l4
zhg!(f=y{WGL|0iOkweIuz_q7ha3NlEb_Z9z^B+U5pooI_%Iorxw-pkIgB_v(?&2C^
zv8~S6m`2aQkp|~KC}0r>G(Ug%T^wT!?~z@o=3ZzRPFnE9I3u6>>Ofi$++btyqqia;
z>bLO-uvxkC7V8fS^bSr{OH19;bCqpCRkr8a`^`|0M>z13J%NZED2QOwhnB~~&u$EB
zcax_^{imLV$W``$mohDR+R6kzDgOMqw)%(5dWvgCW!Kk$Vy-P6_=0Ahk@JaNnM-*S
zA(XHxB%~w<K@|6DI3_04#+q$<D?Ad_VKM=NvW&p5_eT)44x0&^yV^`l7`xCsCHo(I
zc2+RwtL+)5=X_&$rF0l*XTB4_e6@9SyaV9CGR_Pik@}vj{6jv%F3PNmj5tsy>>jJK
zO#RvYjT4?XIM&ae7NYPis^kUA_7U$neehEA3CXBa!@y6c$0+`)ateQB0o-3>jov8F
z*d@olNf#*35)by#hlBpU!u8A#&`A4ac%gD(cp)T|D{N(N^t*7~Ya1h6e`WEHwNipD
zbgNMVzts{9ZVd_6tn#tQaODxs!C;944wXcO8m~`_VUV$OWjWAkZR9xg1%0<aQ*bjC
z1-);QmsN9xmLrS)P<Wx9gfE)LL!yxneJ&_nD^E?F^G(|?3UDoCL;-8E-8O?yr1|cV
zJ%gfRQ6lE`SEkQOV{`16ZN(9}#8xA7uR1@w2xgbix~-CTnq0qrX1el3rD?*SE4qjh
z#G>5$;J^Y8_Q%zE6zjpmaHmk`@IHvoIW^pKD0I+b+tvAFyHlQkrsH;y|2u>td9-fu
zZa^1|QO><0^=oOH|Kmm9ksB>KVi+R|QbIP_<Pe}HCuyNzvMlL4yO(S>Rf-U--5-y!
zOm#NAMj3`enhwYfKqoffSA7xkm{PwaPjClZ*S33VuKoUs6|E%%$3AsfHh5vF-X3~#
zJuY#pNCXoKc=^wl?mjKiu|gt&HfYF}fNzMS-w*GWkwtz#;_SQ6Z#5XGJ$*c2ehs`g
zj4kwq_pkr<138jaY_+9PGeTDN1j^|csbxcLrtM)LKJxxtK}mmkaOOkL-Hw6?DVdu&
z0;1D<vc6rOTwW|m9ihg^c9t3x1aAat>=>EXuDPFTsV$eD=JzKoqo5#JT6W8!)Bvn8
zJ@j}d<?FBK#5TP1Wy>}A$pJ<oM<=BKe-68EFzAd{T-QT*C}eRQG}@y2c8B&0rU&uc
zKLxAHOV{5XA6w2MP&aABYw7UmYP&5wI(8JbV|&^W)Ponm<54@7?L_93thGE&DSntp
zNQK&;+?Sq2l`YPNgk`7f3iAV9$JDt)`PtO((u9Dv_`<mD2GkMoTyg#z>ZK=EY#jss
zvghD`ggwtCBcmWII|9YZGs-=TUshE^<bvL>(ev-F4Z7T3P<x43$*6G%Rm3JGB*Z>(
z-3v!QhWN#y1CT3Ze4N>_{KdboFLHrSS;6k+DgeI=clD0i?axr<yY=wyTmG2#>VhPO
zhOfHE9V*Q?=@Zr|De#R00iQK(Upb|6seDrDNTs~?%a?SJ^$Sr&hubG`Ka~Jsy1i7{
zgw!cop@W@K<>f2T3_)^1YJNTmhO;1N?$*-M)C3l#VHJX);J?}1Gcz-X(C2n*RbXj3
zINVSOqd+D)r64z#Iy&k-_?|BI#Q$OL)3yg!5?p7bh`-?H=hxnC15W;SvCaas&~U+a
z0UIl;tIsZEf2bS!@HjL)+~)7!$7rjy_L&yu3{6Sr4neE!D0H&=<wLA83@1)ZZ6>gt
z0IqjN$BED-)@#e&98o>BXCX$aJHJqRu8X%t{ioAFR8Bnp`2?uHBUEe7$sgLDySGhU
zAS`e;`y_<k1;Tj`-DmA5J7)w*70inDvju;d+1Eoo&UifT;w}RyFMTHU(ITKvn&2Kp
zifFf|AU30DK5f)jDpUo9+K>M5{yoSk+2paFpkDVp$HAc#NK8s%W%cp)Zfa<#7Ab}o
z_?TUh4dv<S3D+kM^A7ECuaD-tW&iEnJqW}2*_V`@tYye1W_{p~Pi(&#=B&?YOBD5Z
z=iXJs@Is?@)n>mbdjJ0NrAuT;F{lOl<5b7Yn%yHwsSr(;vDGQb1ps9&bF{wlr#{Xi
z`N0Dqzul`ZMp}8zrNgbB?NF4#@F$~vhJNxB;0{|4rM3;p1peKcH{lF+6z!nNLqn%g
z56%T`fGTI_@cTz?l_3vtIu-?qh^zH0-d#-q8Uw`H^5L;STQ1=`+q<-wn3(-cf`1$W
z#iIX^3e9`rXPWEIk3a-*9{Sgmr3|Wq|6{JkBh;gu|K^Q9XmzCu^79-2{l%?mOQ6Ve
z=tK^Im^jgx=+VEBsYunzjl<9ainV?Pf0nQb<uO1D1+dA&k#cvXhlAv2#3yN;){Zg}
z$c?wH;k~VcJ$taDR5Y@lqS$?DzdSisv*_|>%+Cjz+vUJFJOS+aY;v!+`vH1LGk}(3
zWHb{J@1VNsd+33h?NP<sJzJ-x9_E6u@FBs(bJw>k@H)tOeSN!)vYin<)m^-w{$n^h
zI#er{=mBxnb9QMv^^PBpV}>sqQve{Z<{w5h3;9tKkT#TdS5G0Js7R1^#uMxA&H*rO
zh{Al`nTZJ0CUKQ_rmV2Hm`9aG`Tl_y(3{xN$YCMN*k_kuY@oxZvyW}?^b2(erm8<?
zzW+1F6Co!f`e}|W+&2SDp-ui20|U~KL}j($e}MmH_(S6+$l7e@a!SAjgikuI&ieG{
zs-S?tF;r1;u^TZpHI-6VxxOqgdsH#_#vy*zDjzl4YKBJKAfK3+7|-wF!7ns6ezSC(
zMm;W$wi>yMDIyXhKlDaLMa}q$Dn!)ALI$O7K*{sxe<d?k&o?zS-I9VSQn6v-ix*$7
zV}A=L>V-B#sdiIcVg_{U^}IQG6%|>xp%!Pt0Uut!&UVMKi_6Wk-_fW0<B#%)LznkV
zPqylI^D->Gi7J{OM{rz4Y0|!D>5Uf|IpCj>+MCC5O(0)Nop8_YjLA=?&&sF!p{tj4
z{^DTG{P(L@!Fy&x)zsDHV&d*cN007>Gi;U{aNGPJTg<@&xHnd$EZvu%{e=ZX%G6F(
z)uR@>)*snGe6(X15iUE02&!*l7J=^Il}uteb_|Ej4hh+JV=7tRZ+ve|kBHHWi&G~5
zhxH3;L%I<E7_g%gzJmayvKUz!Nxsyx?xUJm^^o`4i+x8PLT1ud-#6MkdofefeQDj~
z0RSr4xzK_5&t`XJu_qL5xWe6WnmT(X64avyD|I^YH=232cJJXgOqur_znOFwvJKwz
z%U0~@6@1?=&cP*hAM5Z#D=RK`qz%UZ$dC1EhhZov(E{d~gvw_-X_1Oo#zk(f>6{Mw
zc6T8Z62$J@!szK)-ngN2b0mZA7|-NDr9~pFUFDCEt<q-h1hFi0U2wS*;bV1-fMJfj
zFU(*)#g|zA_{~wl1qcjNgi8Bp(WERy3!|1^nbN3w_3FNZkCg$58e^vSpnxyD1X+Dg
zb4Em=<$qBnYd5w`cn$7#os2hn76iO?8e_`djGrR%(YvVwarG3tY@O4wOspWQ*#7#f
zFzHDDkpsK$r{35pw>uKL!_QZbLSsA6T`O&zatNU+z`;I0n(tsG8FGc<LXK=v91#Iw
zDY*8I?gb5qQVa@oZFO~Z6%`d=tHNc$v2lerD?R;l#DAueQO#Fi3+Hx6RaaL}*v}pH
z`{(zy86&OJFCYEU2}F<IY?YqDedhXg^)L!_#0Hd~f{A12orqN62#&yQ(>9VL6dc(1
zoP0hX&`G~4z<D&gB`rVwZp(XZWsH3OTySTLHLmFLStq0L-w`Nxdj1H%ZP|NDr-t0x
zR-F5F+!t6l1`8hKps8I}CMc*{*yj6Q3k4ZVi1{@b#;kr1%_=Z0Q*w}p>*#WeiWpFT
z>d6#-_(HOkI@S!v(w1tghS+c4APu8EndQilONQlG8vp+_yM0vjpW))EW5+I^Hic+P
z5L-R&3<<kNVQy?}%y__yPS#t|*G!U$(bS*^o5S~O78Vv$br7}$3n4A@p8>XA3H~r7
zvt>A$?nK#@ULCactsx_ESTN~^bTf4L-4275BWwBF0JI4|n^9Z26~;RCv+>%9Lh8e!
zqk(%`!$RzN_-!L4b~ssa35Gnl;J{F2Cw1~pMrTW`=Bv}sL%#}rSGgsNwdnoBd57c<
zF3wl-gSK!hKsyWLUD}qM3l=%aW6b;R6<YmIucf=MnHU-G*@Ux5E7AA$^|c#xwQf%>
zA_f3BPfAQm%3sKR-WLb4h!d&B#o}zvUtV0AstXJZyv5XM^IPcvr=0}sq7g()$Qbi6
z0)~ZRqLG}3-uwEpU2)7?bZOmhL~2?5xGNbUVTJ;qco(oktVM)`wT|Fo8xVyY`eMbg
z%c^#M*nM7)Ic}9irRA}ST>5SdF+<*aYhv62-W=&aRJAhL=XUDV$(=s;{5avC+XeG1
z(pjk-gLG&vkeU5JC#B?U#hAV1GQ;Eh6Yx=H>Y4ru0rQ!BAw&y($N-nccUQOr+n%85
z_oyjZKH;CasUx`?7OxOr*;x^*?Nj{xXajvZ*2-H*X$QlP)cYRUmJS{yowAYN!g5xr
zIli^`LwH0<K591&ZgI#cPJV(`l$iJ)alsNTpg?{r4-u6t&4Vm5@dmzWYfo)8cuY@r
zW=V>Q9w;h2NSl3>-t%_tYk{kaXVfbb5JU4-q&akov>wOTBV$n*WuNU9@_MW?3`+J(
ze?e}P&WdsO)qW6pupu^W`rDfAVjk}9<2o>CnWLmbTJ+GAUh}{!H_l}7bLam2pN|5x
zsyPp$<=rMXypYse?t?bvua)!XI{E1Dzjy4d0)ZJcL|LR-z|4}Lk1Q>mvdjIfAP+yt
zwv0FXRmIdH=4h8r<42okYH|j%f1+lt+(C#2`l<uXf9ri(ZMS$}&*MAZUH5*6@e*}h
z4fSP%Ay`s2xLz7?V2bz_ZE>9`zdYZ#cpneaxKVU((6q_zYNrAy6iQC=<j~OFES{^+
zxJHu(vm%zYL11J$Wa&=?+xijo(#nch-7o3WZz3+mI5aaev*U}1+65C66XP)RNriAi
zoR)(3<kQtpa2*SO+VTA|+F}<jr0>EFGcMcSzJ1%%YI)>W;2CmCdp%?@=7c5<`!KBj
z&O_Yyeu#ZMviI<z;N4Y9I~9VDUatX+e0V%!fezv@>?X)fQyKP|VIdPZ_Mks|0WNYt
z5Aiz*%p2OmS8WjQ=sA+nZ4ONlkhDms&(EnxAz@+2*k`D@T9pxn)<<Adk^Kuc77`-T
zR7hf~p_db?xN!a5&R5Y`aktOS&EZ(_%SuYaV=moAbbBDynaAKHVlFRp=RWxHI0Hj1
zdN-!70IsTKWMuR_Cn_3Q;CS<9^>0`R@qF@9*n3;zb}pS7h19aP9dwZ8<$C(UL#Zzh
zAe{Kl(d){VO`34D%&5qGo};Z{F|F=sVDveXy46}y5P^2!j`~?Ap*2Gav%cW${KUr)
znM~n&BG!1v#)=E;`$MUKq|7Uw33XwAz9*mgG@!^YA#sM6ch^$k3uMky1v5QkJuVn#
zYz{5;nUt$VGT}~z!_2oeZKGtQL5<OCpZK?kFRbNxLGz4XCbkpo>;$YE3WdTNXlOV<
zS2I}rRe&)1kK={B2cf`XkoN~-ub;PrTu&1_MfM5qJd}E6r|C{+UUc*VT};V7nA9PG
z%T&5;;_$snWNoXjDe<jq(94#HElnP4ZI;!KYo{OQ1erEAdI#Js=u+ZdjUBd$`jhwL
z_rLxcM*uxU3jv{n$gp#ekw{S<^0SbQX;*lb@jR%kbu*3@gX9_O>)U?9BnzyU=@2uz
zSN&|h)6Hu|<Kwpc*tGk{Wu&AYx&{3{9_yL%c|lJ)L25<CAR3Cl2FN$FRZUH|U)&1`
zfqSDNNDC4Iw_iTQ&XKRM&xpe!@J!?Bsowr*1Aof*h@%+{CyJwPq_d5kmMZh1hd}{M
zjLsAtScjM(AjOezj7H77Ko5g@719MRLp%z&-?<YMu<8;4MbB`=t@D3^h<~vyFc@U(
z=w!J^^eL~=!XhHgjg42)Xu_D!^A|4;o7ZT3ulw+ORQr4A2W;Z+5r+gLM8zpNq(#$(
zl!LIhj+|>)_~hnC-Jzjl&o8|=e49S?iqR3<E6-8gPai@4>5mwF7#Hb?M}b2xi(P|T
z;{qo7&I2jcW=QGGmkNRF!drRDfw2&XU2*IhcDc?rNMXLc6QHQ3Ty#Q$fnMTcynvOA
zi81yPV!+sXo`1%E-CqO|r>tHTxzJuCex1Kw#*KsaITdO89)@bXD#O5Q6#!mVUAD9J
zgXUhL#huq%Mrmd;>H-RCtm#fZ*JSvYHS8Yo+6~$sw=2c+=dz1lG~N-kdcsbTCm8PH
zg3+zrkvo&#p?e{QaSI!pR%)@xWAy8z!(~Ov-D17>>I?nyLI0Wyzy{<T-qfyd%9o*2
z8XL`oC^#iZ!+Z5w#pvk+p!34>+GbOesMl4y6CTC--8pyU+Oe;bz2=nD$})Tg-jwul
z=8S%_h(ZkI=@kSAUD+QGM|-Xzi~u+F1;AV0|G>k`i%bDD)IA<<Zn~p(YIWSc|2Io=
z1|i=NXKBn;c6quvMvDdoko3dPIY1W%oy4WS7+~%U^1p-J;mC=jPj~Wd)7bfLb#A|1
zZG{F(jKax%k6);}pJC6eU{h)t4tVqfr;3n<qc(lEMf4}I0B}X;<+~RlCpejH4`_uV
z+S<~^Y+3++>Rn~ul=1IEg@%f7UJc!hr1|IL{T?d=7yP@4M{y!T6?Kuf?)$<<Iq%>)
z2{xSY6QiPzLDvHb9=9R978wIzGiC<wCGVY=ax<l={{)if;#24bKF@30y{Wj%ca=%`
zPhi9_xV`P?4^BFW`9EX@G+nfN9WsX?t&Fq!B8000PYF{(M`L4zV+ZMONN-t!z^@2o
zP@lX#Ub`t?{g?k&`IV=OXkJv+?6d*{;jZl)wzh(U6K<bBU@KEEou--Fi$s>}yLSd5
zq)q8<IQ|$ro}$j0u7^T1WfW+q!!Qbj+Q0>0MOxEVh<SrB;8vvHK0BpJ8G>}MKTTCs
z<{)%I)mK>@6=5>66e1LV1+m{f+4RW#I}Uvuk82;I?c4H4Jc2$Mh=lpJHZCqXxsHY(
z&-1V1Bd(t&dOv2LIwpfqv{33)>5P<~KGbWRFS}s50n>-z-P1>mor_Ft>|RuK#u;3w
zKwHsnqBKIoAdnU)eHWujAMh~Ws1$Gvcd@jzG_|%Cb^oShy{VRjUTqi3YKV3eb*gex
z_2K*vxaBLdt#Dk1IgS+ekqqMqQtBjKSV6_QJ2E+$0fvo$q?#lcx}3SDOwlXB9J)Me
zx#2Cw=)J*o_Mqjsba!rhhNU)z8IpRfAk_PUk0%_h1A#ZujLRF9BtQBP8O0y+$!%Qa
z34MA~*16uf3pys{`hH6SL6)WN?oD08n%*An@mvXsn%365V)1oc9*@N+`0-n;N!qBB
zULRK0gwt+an%!=Gc-N&%yKRRLsdJuv`25aZj^`f97K(iq6#=^#pL8r{Rfyl~JLnx~
z_u68;qH@gF+}EykX27-kYXV&=wIg$~-%C(vDm^bPP3&NtS&FQUvw(hjawg81)xFz|
znzl+`Mjmzb{F!pPCe0}NuwaPt%d<@IM~2WF+Dd!3{-nSV!5o&OdJk7BvqL?e-^C}U
zJ|`c{TeZ^I3!G=p=yy=`>wYL`J(FmUKgO@IM@)WEBWngP_%dU!ZsG*Ve|{kG8P6G8
zg7QXtS7v8_y79;PHu}UcTsPbh`_&<QjowD;m{KP}Hu~u;O#>AbYqh%g4<D2ql<;x5
zibkvmN<Y0i9tRI}^v>3^g%>Kzvyi60;XRnhgLscH@+dy9V7f&k&Q#b?%ijKeUXrX%
zXMq~#Tu@~)iAB<SWoF=uz09}$e8sQK6Zxn;1IB4%s>Su@aewGLN7Irv%I3=lyeUtM
z^45<l{_yt2jxEj-`%hKg9Np6t{Q2=I`nv9N$*JMGPc6jW%uywb1geovhaN?>;zJQE
z*fUxF^WFkJ3FVO0*%NU5I;mzp=wd>&TCJh^z*l~`1Wo~=so`m}j~)3}mprQiG3=T?
zD)+)}dE|Pt(@Ol7tJL$OkK6B)+E|o~vklVL#uxf*xP;QeH`Wys8<#hJHi`BPY)t6(
z3!@2Z-i3MEE-tMRn%$#%dpdwz1r102Yw!Hp19_eCwB*~gLntd>o?xuB%mHQObDr7!
znDAxA=QQj++7eGQhEe!XBjaN!h8Ik|>-dEgqLP#2pWbp@dwrS66iD2tKG~FbO+J39
ze84gATV;|rT5*ZYkHPNB{!3SKWTR7(w6C38Wo@A}kqJHE=U$e_zBUkJY_QR?|5P!#
zbv&Yn2i;2}=gAXeh2&3^)|@mV+*KD4_)$=LOFD-`oJBu^@)j#UE88peKM%LL?1FzS
zz<_ZkFFy%t3@m`<(0Xk~JD+fKSqj5JaS5i_=)}6(+TKFms~sO6KVTA>Yod!0cfaz+
z`(4oJajNCoe73&gYX82mIc@YJd1kdagI~w&hIU`(#&l_1ov4>4c4DGI#M%Fcc-FS{
zXti=rt|+a&?T1#SrL&ronjJ{mjxnOX()P%|%@!eXP$2AvwxIg2eQ-m&C+$9t=v+i|
z-+GHDRc~A9dR(`|OxH!l6C)#AB%IuNI-8TrP_c)sv1>kVSmz&IS)FT>yjtQ>Qo6B!
z=?*VVxd@S4(_inz8;{SYY}hQ(ueWC3Y_X#?D^jQC(s!-2W=^k5N!YdTn+d+gYBb|L
zM(x#c_%on*W~ZH#-$&&nU&_$9>q(>V|9;pVf|L^V_yfqo9Kna^eeLuw#^=dFoR&ey
zdcen0NAC|E4a>G2$#ut)sx)0Hn0Top$0f0$)SwqSvDb9pE)7?Tw63S!(ZSr%Uc;%<
z{LsQ1-xd=&c=B&zzg0@EPD>`B-jwObF0i)yY7&ng)d^>5s%>Z}$gYcP@7KN|l_N$y
z2)pzhIX0HZiLm)qcS8Aq755-jdFohty|zPTg$i04+bu5*sV@1d23A)66Rns%3+w~T
zK6)K<R(o!aO}AVBB=%$b?BHJGLaNuTHFV%`AF=c567NQ-Ve0Gm_>EcCteRHc0Si8A
zW#F69c#Ay0S>LRNC<(aKEb4pJ{FVFxi}DY$XNsPmMj1GnT^gH$moc+5yIOm*l<@1o
z^U(dX7q*t$pU6YP9}m{jsODkyM{7<(ojZuKDlTF&OloMTlg{T>o;3+CFq-~Emn7j;
zDmWFq(Gw?0`aW>VQhx16v0{LIjFfJ9`1;rerhtQ15M+_rGvDnz|54GD_9LahcRG$r
zYiaEaIxb0>a_auWA~wvjn?wZ<AFmZ-a$}iiU3<HkJx?%FT@Ruft>vNK{@wzt$xO&z
zVGetSyg)`f>f^y5?MpUIiLo(}krhJgcv$ze^`gG>H;?rP3Ya%L*$0qx8;8sN^@f+&
z2QZmgYF>2Lia$@FHhf0lRkodvCt|1OE=-VEiPp`%g}2DJ4^rvvq=T?nYi)A;@@_*P
zbeUiRtZpynFqy~O*<?`gUbugK$CtL2q5P--Ue!}ExP?HH0UaO8r6lfE-xAT?qcEJs
zo2S#^9{J8v)1N%80rSNJ&VduVY;KVbJx5Af<p#SzQ6>JYo~gYczfOD7a#m1<K?fnZ
z=X<?ILO(VjcX}NiKRnNB_Imhh=2es>H@EaLC9gk8dCf^4CH30A$<|7?`Q~1yS)7~#
zo@QwN=O8dDLWg$qs%}1#*OG0FYY_=|#o+2yAbb|cxTY`IB-4$F?O7))yxCc43nh}~
zG$v5N)-=B{S$NGSvaT1Mqg;dMXlLJWZEp@Fs&o{hWjn*35trR=mY|acFR+@G%s(%w
z>ZD6G?sM7QK^?($yv2Gf;l<9(UpG_u|1Msvqf9inBDMWN!MG{bcVuwYX7USJQC{im
zm>5EyFpgWf+(#OxbX>@Dw^-e?`Tj66XHo}_@TN!&-ik*~{QLOrS{M%Wk?dIUiDlS_
zPFN?&<&AZRDUV9i4XTAxn+6(^2^6dy->tNIQ}1O9#PWW0k%y>AA0=C5<>_LFh2YQT
zVb2<|BY0H$W%)=t*o#VRa*zoR#P?kH^pA+S`r~ZTv9Zt$)@Ctoo$)G3$f3%=x#DtE
z%U}1wnwyv^XbC;Oe_!)xt-6LRaza#NZM5)tU1CXEKX0nY*3gVCHU@v3N)Es#mBG5>
zTW@PkCk{Kvi}P!Uumr4QvL32+P%A@ENx$nEQR`UsRm(NxrpobPpxEX%`R?ghLGSfi
z(RyOe5N412D|HM39>H?Am_*E1n`CDkdrE}ns!fGEL1RBKYt@E_(dJ7?XV^N;n9Z2<
z!|LtrCPwhmmvUHWNvHECDZ=?Ku66wUP=jas5;7T8G~!qiWj{_R<lK~p0!C{$l-O7{
zB$+I1*SemyaGdd8y?ez1L*K?c9~2g`N~5f)R0KRF5TvbW$9N~7(7yeB#YqNkBgx^6
zhjrd;I9A$!wj-W)A6G$_Y7Tq$!P8A#GWS<ksY^<?C{Y=xAV&ec8|$}VNV-*79t(#u
z!zPyD@Ot*mPugDV3D*8ggO#|ccT995`6Z1A_rnn#vs({Xl!t%#`|9UGtbF?t)W4MV
zCF5;b&h@fq*Dev1gXTEL+VHn%Wflc-8J%J4>j%M*mA>fxY1y?o?cN^16cNP^V=r+i
zvAG|lV@j|!PlPSQKP62$Vn_Nm{`ui1nwyrL6O}mp*ik#KAtO9q%IyX?gtB<&YKH5g
zXba$U-+JKR*l+oyFZC&;#>g7H!|45yZSn&$I_4By2DtT}idu#AiU%yuudnKC70~e)
zM0}(6j5R47w@{3e*??Q~1PU%N#>S3V`=;TPU{v7!E@gdvn(jOuC+dCWtbJnMz-sr&
z=K0B1>iR%Xehl{<LlC7V4c1OyW~Vo9duN=vD}u-ZC(HZ8a2ibPPl~UvrM|yn!xa^O
zxZ5rFeGq~U`E{CAdBDd=quy;o3OQX_Cs?q!czMIw<c;ZMoVtr@nIO+EPvK7(a8xbU
zpCa~~n412*g(9_Ww5!Se4k&q+m=U{0E|&a-7*Xk@?+MTL3-G&JB`G9%egEpl`iV<J
z7LLGvzJfSf2esdxhvcu<91)^QyO1rov}~h@!QrOv<!v|oMul}AMQfT#A}-tM`h0Ba
z6VUMcI@3q}bTcvozKBoEvU)rg`To!7QZK*#^|X^;UREdW?<;nQd-LE*iLH{-9fgbX
zOiv(_8}3S}CxP(f{U#D8+ZN$vUqokLqW*ON70`#u^*q;F?4V5K7f7n3@$pUclxEc%
z7j{o5$o#l_g_W93qY)~Im_45rHzu*$vR>%sP8K)d1?3(1br=czOz#nh{r-*d28w5R
zt}m~YTQ#mo*)54?P2kN>xz1TyySaU#OJxh|DDg3E%RMB=;>4`C34&@#)-c#r`UvK-
z7zjlraF|?>kT95Lm2^&26&>DT-JC2$e?l=(?pHm@Mo#|x{KqnszEONgo2y<2EjG<#
z><2Ju0mH6aAU*#ors(qojmx2)E30CCsx(nR)oLPk4eA^rVQvtuoc$87N;7PBy7r1v
zjGGCg2<4_IiU2f%XU*HOk|gmn!>UZ?%zRTHvKbKwq7ZhdBlz0ZQectw&U^m+?OWKE
zBN+wUKRiZ1(Jk(PEUj$C)QpT>TL5Jf1{g&@S?47x_w@A8`WXcbf>-E3RMPpemRIE0
zZy_iPg~{Ag1mtf`%6gWqquh<feVJ-%ZfJ=0nz#K>kOT)GAOpa>x>=JggIJ8q8_M4{
z%QQk_Besuld~0b_Z316vl}=dE!3W^%pM>ACbi38fvysApf$zt|n`L8@($QexrG$5H
zbL`lU%BM>`gxcexv{@Ao<iJpql(Pu&SPH#pRZY#dk8nM8cWJ|(S(XhDS$cVS^@TOn
zZluAD1h)R8X31BE#Nu`cKaf|qI7eR5sT@*^R#@7$#4Vno$%GPF=II38JqP2?en%0q
zfrbqKh4Zst08Lv8nGF{CVS}FOnu<8I^=OH=&IArY19xx3+y2>=bfs%%b2Q`-${mxa
zJxW|(7`gh+&};J5c`eM^WV}`!uhV?=lPB2mCt~4hDABHCO7biRlMhpbtaQTW>?P4A
z{nyhL>uze#JvGu>ufi`ncM-XJ(O)2zs#XWP6gP+6NR)CLgg6XI+_|Z#skur#*g=_7
z{86X}avSiYWNa3!J$ckEC`Z4Uf+6mK8RRSqvXIEu>haMy@d_Ot5mb+?x@CI@<15pq
z#LpzZR!gTYf6FD}U1&~QUM9~x9+;iRJ5>hV%<Uou&H6q_GPWX^saQ>5bjtt`Co!Sl
z8Da*%%nl_Im&-gB>RW`-(4qolv$qo};R!ZaLhq#=%8Rp-l6m10KF4z`W9ISPp<^U8
zTWI9R_i6A2ur8*iy(HskdF;Y?7tSVJfwC~Z1K{yy)3f|KTxRd&n<>qLPU%#nR~Rf|
z!Mv37U4O%cVQ+G!P-**gVBIh;0=}G})4?Fi-&oIiiAIp9m}}v=M|~)n<bYBxghtEF
zv4xnS6_@q?{mpJ4y`Hb7#f%O6o*nLP0GQ~I8;9FbY|+dn?>8U(RL{_mojwM9gngyT
zjU;gWk9&l#gUaW3N4UBChiCtmRy_FxNaFrdWf)M<`t@@S?yb;eZp&U(xw(mu$8nb0
z)!GBr0VHH?3cyY?>&u`Wgf<ZqAK&z6=Fj|ot)D5vUlggAvt9-EKAj?}ggp%UQT)o#
z!cBr|^+>KJ1^W^&SUlb!BD0G|oPVe7mgg<pPQ+>#7#YScy!9zdgwCQ{_6~3GYgqU7
zWPu$<^}$GsbJt$mH@k9Wy4>Bx(b4f$%0)CUa!!qNI%*>)KlX<;3FP!Qp?2Smf9+Ra
z+^%C6FZPsAx~qpSXo@+jFA5+Zv!d|zc~$5LS&Qc3?IKTOj1tihS_S|*KRIqfaAB-~
z!}kgIbH<K+&iZBV`qoy$Qsm+>dY&pB85#vmJsC$3kWTruyk`rfJgpUIW#rGu$XImJ
zt^-H;a6gRMp>A|j1Hnocpm2HXEPbr9g-`igOp=OaO8S#cH<rd2Wy&hjI3#6zPJZHY
z7z=BoQ4a}W{4Y|&`%_9xa^pY6uH_OOnY=?3t?cqi{;4)sX>H>VB}*xuWV-WOtWhrv
z8Ys$TJ%izFa^n|Rp%N$By_%#@!Y-WAlgcq<J-Z3pEu}hz)!^!|`E79`7KbWIak#rD
zpCccg{IM#{Ico*)UPP8HtEh$*ll={^Rp3%#tmoBacb~BK-DgU!<T|O9s@C!GrB(pR
z{1U&z@WY>_=056^Ub^Pja`v_VNUIF!k``Ycb?9e8-&Ao)k*X^5*2`)Sbx|&(qstC4
zC%a&L3siqL^|_ZHteBZ2_ur-^tDtvmoq8u7Lt%tFu>)*YcV^_hDAD~1$;n|`XY$sv
zRIhz&TvT4}2Qmd2H-+ai6Oa%DOSqlYe`+_<_Ciq(f1!+&`};WS!v1JcJF0flF@6Xi
zl@%A~gi3~V5keN;uIq2o<Jq~i;)^s4J{gV8k&n1I2Mo1KVM27y;4y@XLC4mO^~d={
z6%xq_mTVO7^s=i;TAnYvtQGB)qzo&{iVl0PA|}R{<2uHgOZRQ?HPUeV=y*WrCYsva
zlNJeH-GINJf)?i5gM{d~IDsqNV6<LaDu01IqtS(KApoWii|QJ*U3mSg$@a}Tq8>_K
zdbXD?x$v&O5<(f|yM`d0ZW~`Y&wH*nvi1qdOa4@^tW32l{5!=poXH)FE|rWT;Vg>6
zmsn+~)$;pCXDpy(G>mc*gUj!jrxKWr&at>F;g`FZjb_G;*HPDjtf>87gdYYMC)Nya
zIX_N@kni(e%d@kyYpcJWYyQX!JUZPFd4V!z=6l$DDO3_mOnZlce`?3cPw9Q2+wtg@
z&&MRC(NM;X7u@dnGz7NwG)cEeT7<Up`*~%r8ahXC`#fg(R&)#NvU&(({ITf!F+ijp
zuW&caE)hN(#-1?Za^wlF*FDNBfby~?eZz;$D{Rp{mv3sT+1X7%p8KS;He{AH)z?!t
z2s^e8S_M%6^(93YXyyu(A-sH<`4K!$srY0?UKFnwJ)0aP#!_Av^ATeO6%Nb#%<O4B
z#*6Od7O>7+WXX1}4^{M<F&E-5tu@1;oLiaDQT;#6WnbSSZ^IZH1pT?;rtI>qi{<?H
z#)ZfVp5P8>J0$u9wl>Q{&J<nHM+nDh^!tMoaC*XT3l<>X8Nm|1Mpaud3{?mG55G>e
zb|J>a>m8v@I+b=%IH@r@X2kQ|KpCfgY)5DF%9m{G$c(sfiZv;(bin;u&hgOtNWutH
zS-_|9DdKR1F3(itOr%R3w<4;6GV-9I{?jLZVQ=QIKE3=A?VZ8b9^e*b&TOu*tyRPj
zsvroxcOzX(g5)VpzNl~TUMh0S^$e=53^8S`bQ_;jJu&6pY(1}LKLe3+{L_Gl!@?(d
z>=GUplTTYo1ZQIGPtN|{9F2-tlVNYn&<81@<jkvm`BvBFrsg~}F{IV+tyzWY4elc@
z(O%`bZ5Q4yrW`h=%3kVz&N8*Jg{HQCzqa7=VIljJ(@sua$f4$;G*w6I4gmNyCvf#%
zOiceY1Q^LoC6_pQhfuAARxLMrrTwxhD@{r#KZbiWMTD_UJk*Qr=x#VJb6;&ba!+mb
z@GNoo)JV_BgQrC7(ziKC;2y4ztSl<e>aEOE7!Lldf>|5hgg?J|8sG(sr6}_#qO?7_
zz;Nd?>z|n`<1cS&m1pI<p{!uZ5`otUzZq3@cN{iV|EZZo&i1Rmvsn;kpbVctgY9|L
z^vLYoWBBElE$!f^yDkY>>L+<DhXqbg%vqLJ(LX($lejR?I+gHW#`#xW+WTu6eC$ed
z&W)}UdkQ-nRsup8Vo4dQ9ZU9E#{#jhTY=Il`H2JWUC?aslXavx^)eJ6wVA1>hoeqA
z^qxSaBWK}~v2#Sv$TnCJbGMd+Sta-tQGw4^HJ82-i-z!oZGrb{v`YzlD&lZNf6n;j
zumAHJSb3Pd;AS526)9tphCzG}Gb3dxcmRugmB~~2-LVPC$nI&rGgW*A78R4v^U%j9
zH-P-3&-B}KUn}v&%~dH>diM#7Sbw3?+9_hDRu^}8qO1+UKTdGTdr5k(b=amplD)tz
z*^@b%uqW_s@aO#$8{ckVVS?+i$aeU>7-akf?Qcl&Q%A3k*ya<Wv!NktNtog;_S#ib
zV`B*$DkEL0h9>hSusvj0d4JOG^zK=^ZoRC`3DRts>Qu(fr^5w{SyQnO0$38bG%B<!
z%mQnHAh%?di56BgT@L}~!Lr!rsW4i9L6s-zaCh@<8hnS}LjWVs6UNX~C{sMyr(WSd
z74mc%#%rgtmZ}Szx!3UvK>=fxo@D532<dURY_!i}ukb#cB;P8HlJTk+?u$0OQ{VC<
z%{?@kxt$69V~o2_l5!76T$ogog^335SkWpYk#fp3sZ?W%E;ad(p7X#6?)DDqW%vKt
zms`c*A4c)6G^HocMK7$L@+u>N=FRK>09RVBj<ou~&nHWJ$cT$y?&v4k&6*_HXIg2{
z*BhT3*zDNeBX0p@`mJ~M%=k4bm_jKMq`WU>P4{qfX4FpB$YP;&Tb)Q<gOr2>p<JRU
z;pNL0x#exo4r7IfgRdod@NWnJZ2B)vO}_4+{rGT`Yg@7@fl|0ftd276Ci4vl%~lmn
zJ3C$|@Ao@hjDl1<LPnZKaO@v>CgPjvGi42*geVph)nWm4b{67(C@h$Z^E(WwWQ9$<
z4RpZ-0I62?15l@|PlHvLS@L$CV3ia9OG7^>Q+5jLj-o<`+PAF_rMT$HE^m=Um`WBE
zO&r9snpfJjMzrTf#tAAsXHA)B$$p`WjtbGRIvLuu>>4ijXVvQ3+^xR60pe0^LDEN0
z<26{v!f~YyQ3SHCRV`iW!dqapj_|=z5&XLzu&!@XZLy8oe(nLa(>Pf6Vy)1t7yHF|
z?3(Bn?-yEu)!*rnj_`eL*Pq*KmyO^se#6`LD|U%EK+`*fH-{Bm`|a=6Ea=?#7w_sG
zuXDZdrxKpfp!@KD7u@D)p#G}{XbX1p0zMc%fXa73IpiU|NAa<W5;jK3`lA>3aSoNT
zwYzfxS0D*vB6Y7@o7??S64kM_!T3#WnJf6CFC$Gmq-tqgof68e^sMtLB)yMrPl(X0
z*Y;Dwo7+AA?_jAWjeJ8pMU}8D)%_>BU=hxqL({<oQnDo->#is7RB6M|ycrPG&0h~c
zKV<xD(JeB>v*STiq=>?^5mV8ewdYKMgKcdDIML1KeWrm~X=!<5MPp<vI~YeYOP-9-
zG>FUkyn_m--oJz|1E$Rt#KhY0ZpHZ9qDAAUaod%h7P*`?(amiylaUjL*wsbsyqy}#
z#AbAEpsTnO#u_)qutGUmDu#k+^HK?ANvh9`JLM_do^8d*Fszr<P~I}^>D!`$u87cV
z?|8dx%t~*6b+1#%x&J-z4%viZuV8DC^*dvC-@@tHH-RI?eNnHpA9RAmlb450%w!?g
zKx$0&cp%Xtcd{l$=0KZ~-V1(uBN9Ibrb1dDwMnkf9vc{Z!(`>HcNcSDl7-QX+su*=
zpziVptFXec|IfDFI_m$OeA0&^Q%9gV36Rwqe~f}5-Hn%%671~24XR+at6m@OCz0a9
z3ouiYog(I*7h|96t!4|ya9cD8X)ww5%o95b+`1q0aHUTV_?Q~l+pBml*y$P!v>3Cb
z4&hrpi*RwcZ#@6ITk%prV7Au*t!S->@1wKmk<UJHFAV;vbgVn{2LKd-;3Hqs05(%E
z!rm&%|D|o}O}(i-#n>2KU4r4`dYN0FO@Iza7NE9p8=9EZ3ZtDZFackKKaUUo=Zl_&
zqq94Y_e2a1I+51tY#z%+a>uC}qOiH(Ss|zh`#G&1@hZo_xWl>kmTGeZ3_TdLG2kH(
z;x{rY|B#*C>5=2NkPbz?`sv({LMwfPQK)TRz5Ax;^qgYa_h{}tc>V<&!ZrT}E{A^9
z_6$MX5bd1?G7@b)p!jOV)(@o?a97uqP)!vR1-?f)roKTaA;}{hs9}c~(obDeehndh
zO~yQDqoA|S4M#AtK2of3@vE0GOUx@%>9HmN=cVO$%H{M;oW65j-TR;E>u=_A?-rQ6
zL9S(LCj_uKwUJ!>&gs@_{e*gc>QoZYGswHW)ncmfTD+-?U{TJa{GYFQN8c%+62>d$
z;@HVAYJhF^bfXC3v<jG>NMW1U>KLq*#$BNc{AZ`Y^r!^bPd|53J6)yB^J9UAK6bhu
zse+}lmH66hYaR4H!DGk5=|oE%6*wGV-}CDfcI0_K>!;RaxS1qATLzxO^Lqfyy`7!C
z?J&+xNxHi1#Jq}Ty43$3cNZvvlm*QK7bmAkE4K^JSH`8A=iK$sX7&LAFfSLg(P~gf
z`;<)Hi<9+9r%1P4O$f`g@KGHTo~YAVQu>=;;|1(WS|WvvI~gC3F@CQ%@HH}{m2$U%
z5|E);eCGc+Kk$WkcRl5tC~E#R?zrZnHd@SoO2OEULKQI={9K{Qp*A}lQQa@(=Qg@9
zrPa#Y7kxa){(--|2tN~J9uJokboh}`t>LV)zRN|ee{A$F$zdKpzGrMe2#Qlg^_9wy
zSZFcr^o#HV|J`=KA=dA2cik519gG?3k?O;5wU11<VeR*5I>$jg>XwcOJEQqNQJ-nm
zKBgeO*h|^M#rQw<O$K<mGgxJYO~tXJf9t+a_tI=M41N8&I;PzZC2F^{@jd16f51zK
z@=6iD+~=GwPB0W#T+yX@M&=s=!M!rFzxx;KHbzHz7#m=*ncsELPwLfGRUZ#vVWZ0@
zS=M#l&?CLlYlC^qVve@c34Md<w)18#vjVvYY;J}5Bb-=1uttXdXN^pSL&ut5N268f
zWcF8SdYALcdv&#&b;G0<KwJ9o--5Hi!^gkVxRjX(E5duq^GXhk@SFycbC|AbdWJmJ
z(P?RINle0Ggbl+w&KDOmnhaekZqSYC8{y+e3d%pb?glx~SP#L+i{9+T&%Ujo3kVlz
zv`&){*yvC6wIfWfeRb@jzP;v_Y3tF<hDs;CPZ`v~*pvIp!{`iXZMG^Zna&-@8Iz<f
zk_NsQoy{e2UcmtEGtbJ}bWG|0r*Kveq^Tesxt^^nuYaGUWCHDVZb7@|JsXetIqh#5
zQex85@_c+(k*&sewmnzv*fG{nwl6`e(7y-;)YrRrKZbchN87LiYm|!|-U35n`q{%G
zF8}>OJM<v`j^G8c1mw)eINX`6(u9H;#Oq)AjrRC)lOqqz=DzDZ;p>h@sQV+?KXv_N
z>LLk?GqS=)fakijkK_TOp$;l%g=Lp#$(dqEc7G|#Axr;1?{4J<Ud4~Qi=~*ilUk=%
zXyx)cynos~CtkRXsg_>=c35E{n#%OXR*bvmNVWtgV_ba@{iKf|7NN9@ilSh1m6dgM
z4^H&@y0AtWU(myldOS)>IwT{r`^CcRCU&1=Q4ziVKmT9}{s?d|CmERwR`AYU{||d_
z8CGT6y$`CWD2gB|D1wxvAX3txl7fhIgLH#*ZZME=3y3sGNJt4t_eKTjkS<|^utmDN
zW^M3&9(jNN`Okcr4|B};!9#HG`@Ujb>ssqvaXPK>sJxjJOHvwNeD$gvN90>%g3P5f
zW~$V%fkqP7pPXD>yj)C<m%cQ#H*emM0UNQqAkk3`Q2^_eJ&`(ilHzX#d|;ve$6kLW
zHb-Oh5Z(l+gkNX&1(fZwU8|Y-4Xfg5KAo8gcTQnpjkT!)`TLKUb}xoBcsc}XGe<#Y
zNGJx(7AAoyO&)a(l_xXmLCgxZ1`U&CZpRuX%`<a1eT3oq{>Lr{f3+Ay9bCf?7f>Tu
zS;#ahZC6ZnSTMwt+*E0a<VeXAkP87*^U3`%_n0ss3BC`(E#{nzQ_9(-p<V6j7Wn#j
zD#|)jH!`xh@#iVJCF%a&em+WU;Ex@w%;X1r5%t+ysjIGwzMUG)N%XfJ8d1f)ymy&y
z(AV}AkOY}=$ZJhZMC!?rS0$y0i+z$(W19i@f0&QX@Y5isewlnFj?fk18_4^&GACz1
zZi@CBugN(%$qvtN&c@oRv1C5Pkq332&#V&G7)sOM)Z!@0v>TF)SzVMU_#MdsyP&6X
z{v-2yi>0mv01Jfzluw3tX@m!AyZ`>*9+^OIFqos{-Z#kYbtk@OrKsqe)f7q}l(=yt
zX2-SdLxch=Ve_IGOH{BOA8Tuw468I@hKj$VLwEz^vFsYI%xc7N#VVPT4iC^o!kzoi
zjbx&z(TO#lC`vA%u4uDnj4JzS3%O3bRnXH>54+Ldc|-C<@=w{jX#s?x!mq@%vdCDf
zviZX4pLb=58zQW<xJa`;YC~ogfjfCUXUO3R{C=zox<&k@_mq%XdgRC1!~n^aY9r~4
zj)gQ(W4Y|I1&q@dgJ5%0f|(0a19a0<>n@eOdK{Xor=H=<=&pvS6$AIUOaNaEjfJZ8
zFK>KnBlw*vt)&$s;Q3yTQ)ojP8@jNzvI?uZxku<*L(j^P?$?rERD={uT&|axU`-Qc
z;Tsn2{*}sAB4oeT_H<+=&aH47!gg>s5gIfw-DtgJM;zPSroC2LoggM=Y!DiTnM`Yg
zj54)M?pMtZD-bg#kJJ#E6a499TnGu=tq2{IyV{WM<kVE+!U-B#R7wi23ZZRwkPca!
znEpj^+3y8~h2j&z;rkJltdzvWg?b$*3Si@-A)J|<eAP}*QQ@P-?7H-;`RHDboMqb=
zp<ht=8tfFofM~AsN!yP7iFdCTq5-i7VF8L*w`Io=tc_XuaJ`6chi}9OayH8PdOF9D
zx0MdG@kaKp9J9Jw<hUc&qKwKWj?56>nTps{Qx^WszcSr+;ZG(UT$<;ei4CG2uCG3(
zi<2~;8CmI{t-q$79UfRX#1Q6?T+b`P)PK>gfe6mqioCqqqD0ApyMu?N*0YZ_HD9M#
z9FZw_W#59E<pL#T)DRVdZ%rTlmp*)-1M2BY)7W`a-`T{wRa0)`Jw;AndSAU+mOl~v
zDQ{9!TTCoVO|K0?d-i8L04@LU>uX`5o~5>#6B)jmvNcb-f9y8B*SUtEFaP>&-x5Sj
zA^h5{9HgxuHz<LVMH$-VGLV*C*{;YSou5X1$+Lk7G%F_1tPM$0ks{7-6&excN+D4d
z)}?XbUWJSVZuZ<?0>2Np{-uMK*>Y0LC}QRX!);2zBkg)HD2VBA#ySFRA6_Yjt#`~g
zUQGAAh`g=o$Q04v|H^Q~Y;wjiainolkX#}(@dK{V-%cjtIP>Hz;46)7r()(`M2|$<
zC4nS?#C?00dpudaAW3?cfxDrCCG%S@pH*bkX2VmtmG=#kZe;?I*Vb{gtG)h(OkRmm
zyu!vs-<Jk<yDBG{(JAP8=2eWjN(V9P8td!7cl2yeChl^>=>F<V)v6sEH7W2^+c~k9
zyQv*;E?)BQ%lEfPerjN%<64{R-fZIgY|4V7Y2meZu#DZxAF1JX{g%9_vym8Z+_GZi
z9chJKkwE;nV8#q?e(cC(!c(5TJ8TT?dqMxW)d&F=Gal?IpZ3b`nhj#5n3-iHu}qgV
z^b?NG=G*H6tuerV&$?r)hwUZUJzyMsuKhcEH1MgIi~t$_;A!FFD8?~NBBTyRC}1!C
zzIOGG9MNNewxcsy@|MvQx%-21c<|xlkiAQ;1l)|8cmFZpC2A6^%x?u7bFCLnyWw&~
zj;y?rw|m|xJ`q!f$@Pi>tS5gf_~u69kJPo38l7kLp;pfo>{+eY2k183xy3`v?h^`k
z5>9F0W)|h8E8WK}`&3^3$W;jeT{_v7R5}*8ON?)_U=T;{Ai<omcAFeQ81%PQupa~a
z%VkXc*7o_~`iH3ci#@pc*SNw1;l@D9SKI;QQZsySC81CZrwk2GvUZ-&bv<->on<i-
zg<RBu7m?3dDY}pXRa!#*l}cB>|Ewpi=`>c`K>)<`*HwB>16c|-IO@v-eDB27dDpV2
z`k~xPlAx0p+4>SEabUJ&i0}XPigzrN=i1Jq2d3&P2ZIM_+J{-Tn9Li6*rje%#gG<c
zlNSK&Y^rLwq8%_4d`P7BPtji%IVZcs4*$rSwotlIe#*;}cKH-rD~>0Ebz~|)G`Gsk
zW^!FSY!u@R_IMB3)jpMCEFrhu(j(v9UI#N*)&?<*NmOq(&=MkjWT<ncFX1lNN*5@e
zM}{l@s9fMib@}?YVvsYCLnf*`FAH3?YZQ`9s%XqLkqjuM3G3p9G4C99Xd$_#6j2d7
z?I`<U%jEmKt8l={I2!yrrVL2Pewurk=-45*XS%YYtcIv875Llg`tJ?>EojqOdBNjF
z;We&rE4v-{W@Zz|nFEwJMmfC1M7lav2Dny0IYe(VXVm^Z);1HlspDxn^zM5jD@Vfv
zm3@RI)VspPXbPP16qG%aTMzFHX$7%^&!iBkm2*Xbp|#ZX3o%kxwVjxBNGi}$w)SR=
z<fDTOQ2*GkHiu+Dyg+PlcTl2WP+@&!g{p8`#PWyF*zb~gme|?6kI*sY-@D*{n0s3@
zIVjYSWx+N4-`Im3COP<CYI>_TyQuco2AtmJllVHy=+}Q`MN;S@OL6!|w)zEE&SewW
zY?wTe3C^PQ_2%K4G+!{_QJ>PO?UYax{MXcAAV!2RgS;U)A3GzwShk*aBvstRe`@lm
zOwR4Li8Z-GoZ-sfUf%Ngyj1=NYi*A-R=JL9tllkK4GOz!@d%l+8`h=&ms$KjHt3F8
zr*G%1*PKrCMO7UN`wf$o3f27nvnRPo0$h<1eD^}Su0oc|UxvIyR|2s_G>C?4X?vCt
z9#LAECw1mIAuiJIJ3~&6oo-zwcm8W4UM!QVvpw7zGqI#Hj0&EG(=-X4tqCLQ0%A80
zZ#zUESwdEAIh~ym-@_0xpCmIk-goU1;#+V}qJP$Be;jg}Nww<cr~V75{0}zqA816c
z?QU-BK#yzZq!*g-lN!bDqMXwk0UFg#%abowyE~oML%WI!y&3+;dj~zLrm`b`x^l$U
z_nfWN8JB#)LYljRc!H$2`=1laXMuZ5sNdq%)Xe543g#2vN&e`|MNl@_)e?_Y4m@--
z2_iS+C}(=YDzdAGvT_OTr12++WfIFO*UD}6%dxB;lKg#l{jb&jl9%d}*i%qE<f)zt
zV7c;|T;-0dXHY4PDyKnN8yxlKHiwZzH#UM0YiCcRXX{JdI~&{Hga>x3DlP5gn$*E7
z$rVCE56MFtZV@(S3FlHMz<VH<04ZOV8En`VAGH~f%6fQ}a17;~z;H(uO=lt{MH2Gq
zI@8+DtftA8jhb{GekfMRPP@FRtCa!svk}ARG^+{e51uG7?TxQ5Ah`s(0&F<KXAU4j
z`#X?j%{Q{8!tt`)o%4qW;<uq%e$fXZliR6U%AO7pbDjEfL^DAIYqG*S%jA0xqHnWs
zb8~-n@>7STUDIpAW}T_P(piW8_q+~WkPyt^?ZhO8Pd7?rUPwY9J5dIm*((SM>ovEs
z0MqqwSOcOiXe5$znh(hHwgQ3ndTL_g^lny_55WLlXoN5n`yV@qQo<m~KVJ?w`QQwG
zfgze(Y6e$zC+3)jz?A^Al(dm1u(lJpAtr-Y2kZ^6s9>HBc|#ampoWu~0~8uhh$h_$
z3RrWX_PFE1h91P;UEP$GiT}Vca?E@X3?$c{izJ?`G_=M)i_arL`h+WO@4nntr%C2s
znp|Z-qB6nuNZtN>@LC{$O<6?+_lE;CtA77s=mtgNZAu4oAv!woq462(^GpvS*~Mm~
zgb=Qt9*qB-3G9B<>tEe?(%BIaecKXm<7s|AC*m`(uQvNbKjyBhz-mjf%>u2nU9D|w
z*!~^cC$a*FVTT9ylNE@i4^F@6y<*#S(z-)qMm#YdRODL*86%?)rEJjmXbdMkJ$?Bo
zztb!yH{JE?zx^k3o`Q1$_2lxiz?w+=?=wy4CAT1{c>YonRk&r_BQneUsR(aftjds!
z3J3@&EiL`|Q_rRL%*In_>T76hYnxlb2KjK(@5VCO$DEpD1vReSf=<x?iqH3lZ#MC>
z*Pd{Wg;i*ETs)dw@a)72xvqu)eiinfxYku&Tbm&!EhV)~?2=|)zYS?6UlBdfsC}gj
zI_vziNNsm7nA$x;Acn&#2E94AP21HQ6Fg?bcQgbx094B`2W;ErIt{CZq49$U58T|!
z`kt#8XQ}2;|KPSlqqHb!;>r)8wm)L+^`mxm2VNBtlJYQyy$C!TC2P$}4@B{t%s)wH
zPEJlm>k>kvq9g@!mq*@hvzl311%PK}t+sgoynm~vwuf!+iT1YjF-)?yiV6w*X7>aW
zai)KZDArwLX0m?Uw%&x{UIy|RkqvGz$HCRb@6RqOoW{8b%^l8ROe#Ff+MecaPF0P3
zYOHjG8HBWVbrqPjQLl`_muAYUAG>Y=Vdq!G*6z-hNLV{WijTi`?Ej>TBN0t_uMVe1
zkwdK5U6J)DKA~yErN?dGSz;fMciiGVr5Yl@UIHyY+LNWbUU3+=bk45PVKcQ<_A|8<
zJroN)K_JFzZpcOa`en4EbJ1URH;NvH3+3z!B`N-voI)-x1)>uiz*`;S8K>>HKc}(T
z!7@mVecfJE3tcn_cF-AJ54yt;i4}$vMm@WF2FU&sD{yZ_0yg5-i;2*eV?V2)rG*es
zoEBUIna+>Mba`g-WFe@0cuOGJboF;@oG``47267i9^}7&o%?}Fj2gNjS!LO`HP??D
z-#O=KXU6&XR%+kJwV6)gF>{F$?U1vr5lg{GBl#T6>T1u(jom~vL3Y<bfB$oe4j>{Y
zP%SW1Dz5C|r0ha69cZ!t<TMcbTeJaxI#3xPK0|(uH5x@eYa8N(MyOo70G-We=aN${
z{5cG5J`r`{f!^qBf*cFvrgaCm>fjuO2lgaXaV$32zm$Tcm(Mi7tx<po4p4oy?b<D~
zjZKGgd%9X0$7w1b*Gl-Wih2;RI=U#QqoyDk?wxD$n^=~hXbXF$V#><}V)_~PpuBf(
zX~D+ECcV2coD~;6(AM_lEH<NafB03;5{(2D;L`Qga=V)BkwJi5Ep&GEg;3dbR~qIv
zj)X-tFiq{cO9K;^Iw!D0t`9-x&_u<KF3Z0sd!1S-U=pvyaE`Uzw~?}2h-(sUvrP2>
zZjtL!G;<Dd?u+;LCQ>J6XJZ9iav#1U0J1J<gQYAi6LE{`$EB3R$44?ZFYi(U5aJAy
zu^p9%HV$3Z7ww=AJ|6#xR3I7oa8?-DAmc9A98}c7JbdmXQL$3^=j))YktYkTc5x-`
zt?9}UB}alz;@@cmPOxLBwZrK<1Ob78b9;Vj|5*i9`&3ApFW@2FTx=5(GZSWyU%_U1
z>p+IheNJ)h{M(dsT?udD0=vRt)JqU8;9?7*usu)me{~2IiRnDZYoRILJ=0IpQ&noq
zBZbai6UJ-;+#f>srxMMZE=#~BTGDey`5&=q^Ra)P51BT5VTc#PJTxCq?l~qNWr_6;
zNQC^^t(&o5|Cu6u<sU2x3`8<!((CO{8zn4*R&kRni1Aa?C>k=v|8{0gAlOKh26h9A
zXzSz@Uc2|M8eD@F+B|#)EU};da=c7B)jEHg)Uj-ddj{PW{gFuEWM<EB8v>XAhVx7~
z)!|uvx!h%|9@HxmrgM0?sl5FCod5MCth4ivzYEsc@-C0@O^U8durHV8la3POn<0cR
z&Kw?PzJ}tE?(>@z_*uEu`SVvk9b++5M-w<A>Hl`|HczQ44p9y6YOFeT(3|%N^(Om+
zb1A^Zm&$+G5)R5qt>KgCzxagQlp(~sFx3&6E9-yG`+vA(e~%JtgPbG1JXxvZN&8%q
z1*yG2an{x07Hs@KSs4H^2qOEbxyB_ksbr`$%~tGlRXXAeQHRX8z;rQ|pw(DP{M6hz
z^A+V=k$JF*g#XCbMdWHn!s_Yv_$kTg*|vn{!~gfUnJvdmdWjgHFs(ud4TfW!7_NKt
zA(upmju@Y|{V7t?GUT?JWz4bKTlgV|X82p*P)<6j`3v=o`C$ibGoHBM;hX+($t_XS
zrO$`m;=hv}uttMVd%|4?|8hgLMJVgxr|{h(UJ(XUC5p2;mu2(mp3`9`gLIM1R5Bz=
zHy+2<?2yw1!NZjMA}oJzELH{mD`p_qNzA0xZAZej4-1n0Jaa#ZRQ5T^4WhqK!M6kg
zYWbE!-dY>g)>Yieq(g`0zr&g0NHN@1$eI4v)ZRB*|Aq9iXYFqZIXnxn#{NlI*jJ%{
zoy-4z)qgMQ;o1JrPahuAe}DS$m=24~A!+@;o`4_W#<%lYny<e8*60RG;RL{6SFe-C
zmAG>@>W06d@Rt>Wobb0`lR?(L#6&P)#{uoeH;F5Nogf4y*ic!XupF)^%Pu+^9G)^b
zIOu=lREaQBTz^bw2QDtqb}|O16y)Tne6)ln;Mr$hB`kje7EwcEV|ZBLP?g8eYQMK{
zi9Du!*sY2nhyA8zF=29R+!B;3DVKQcZ1T^Q(@=aTd-WcwxId7s78c$<_wA7pu+33;
zZnM?u`f|7~|9*i>U`yuB4j01WIT~urBSKib|LmdEDXdfx?H@iWShjutfEJG2!2)9|
z@a~;*DTYRGy-;Q3c!gFZ><r1xI^9ad)@>g>1}JNY1dgJBD#G#4&v5S11XMlx0Hgpz
zx<FP{^PbX+@ZKq^|N3hzr+Ki<DO;;dSzFs<b-E84P{qc^3RQ}wNQLfKh1oko3q6YK
z2Nh#GX^g|42p$8$56pbnwgfs<eQg^q*f}K{paFhUc@mI`_OvoddYm~$MIcxNoI>`$
zA3b=`#IG#wPg<Z`b#jC7^qSUU4VT@o2U`mVHC$w^Ta7=zAu+eC_KUoDrg1{MySu%-
zj`KeK_4px_0RSy2lmL*CX%DnrUf5@b4DbX(4dRi@iyC)MU7<^19&2v!*rVVudcBc&
zq#!kwVZ96*5?Xj7ca*UmDg!1fjG*Ypb(z{z+pZUP!qB?%7q}2Cuu;_TOh}u)7#$Cd
zTk`gbp(osM=1uJAtEhmUS*`fVd(;NAcDd@V8b}LbZFUfv_hf1mP0j#u>nR@oN!ya9
z%6PH#nAkcJNMP+A0EQ)=!kxX!&+e)cyXC(=5e&1%@jGQ*<8+CcLXBY>!(mObxLYFF
z5;HQ8IN!|q)#a6yKj4}{_0K&g!z3~h-cRDu;N)!+6W;UIeHbT{Y0VF|4*_ZoSjAoE
zVA@Uo)JlUANI$>R#l@1x`3)N`mY8oYPb`Ypj?8eZtgJxnAx~DvYYY4G$G|MhA=nT7
zIFCBjj~5Jm3&GVMkh^$dye^0)BO?Q67FavHPwC8AT3Cz~JtbuSkqcXdXboE19(IGK
zcXSUT;#J5ZueQbtW}tJyEF8A3Z)j+D0~&hX^7GWSw|RM8AglOSR90PF583L`*RkhF
z`5i6h#cUS$8n@;uS=ZRT{dyepMZK=Lw|{=9t{x>cygbpdKQ|5gGY@vV$YzhK+f%a5
zsVoMXRM3t|(z!%zPigFXUq0OT9H)DATS%Y}V+IN(v^dV?g8;g`?hg{|KXUpj+?>2b
zWSlJtihd72zF*v@tE8l~`4p&Z_hrJ$)XWU)!h5l~=NY$#6(xc(+oQR54Oo6iYS0yK
z++T}cQwYaaU8Qp6o&Bhm2VU=3Ae007ZM!t<QltV8(8T+PGBWu44jtk%bgJUjEHCu&
z1i^>4Pp^>3Y@Nz$8?-U+-ZfbtrMg~fJJZi4NW*45Bv65AiQ!jA1JV7vCurKe@7>%v
zr*`78j`rB#ko&s$zEC_K8&xF;X~ck6+uyH_xa00F;DhX0r6F{oojTg-Nj?v9?U!Kj
zqgW0daE^>8z6jjXDV7_`A5X1meK?!<;e)q+Z>#9RqkuOm4R;dQ9H}UtfDoTLf$-{a
zv`>Ej{%!%qg$w#O3x2}R5?!iqY3W`|AP`y~hL&-zqg@#)D>aag<Ff^AO+^t83=L_0
zZ0-nU^G@SL0P*ebXQl2Nwsp90_$P6QZ)(<!{45~v0<{ytVH|L+0_^C8{UKf?A(;|R
z#)<NRwsQB;@nf0HKh|c^m2ss#&vy0&u=}H0BCgcMi;G9$IJMzE5^XZ)PX|6#@aGcV
zq;PMrP<owRD%QNo@swux>q==UUxnb!n*js?R0r~_oUNUSJK0}g9whv-*{~gE)`@|l
zAnEX%U8A=_+dZX{+LtCer@0})Yjhy~BdDorzYFTw*Li{GuQ)Ii_@<y1!$@QXe4)S*
z-R!Kab1qFY__=Ne<GXFypM={myJcBB{qUw*u2&UTFEOS-@G{&6Y#K|)+J>W8vD!f7
z!V<y2uoqb>y(8D$MxwZXZW(oSop9C)G`q{hTU(2=dKP|D%k?Zl8L!WU{py)DthgSb
zkx(U21R}~h+=z64F7xgu#ZW3xUWhAT8!|R19@u-}psm9yh`tmnD=o#6L9PW{<<0`I
z<43DIu-IeRKueD74%L<)ZG5a6QYG}1*Z%jLYpDqd32zAEqnjV8hxDv|{s0~m;3)(>
z;J(T9F;<@NdgJr9%xSLl>-AyR6}S#hoFTr<{&FikSG>prjy2)z(f$ZmC{GC-*R?h=
zIpxw+wSW@dKg17ob>Cl`yg>iZ%l+xZA_5}v=XxG%z)OzMadWGfKXj;m%M*t_c280<
z3AiAkfM~Oj1NK&^RiOIYH!+iw)E^BWqaV9tH7yh3_IY-d^jkY`dK@p(0wESb$Tqj2
zU<~%#7sZB<%Yx$_vcjQ7J3<uJ-nGd|NypJ{lbwteMXlT-wz3B*<y;II@Dqc|JvKWD
z@p~aeY3xf5JL}WgGrF-L1`}80Hmp57gs?MakWy2NM76epaup^FwTGBJum_^6k4c>#
zL-)HjZ+o4-@Kq~Cbarkj3YPJFSA_Y<F{5}2o+o$a_-w~c;lAo|R20>&0-vwG;uUs~
zbx(ZVkGs!*!QG6_d;H~j^LhQb!SP?eCY|bWMPZfP#p?o@SXqJo9i)Y9=#Ruh(ST8N
zR5GrN_JHkts32@Y_f_bI*4^qEbzXabK+s5h2x|UmD*XrKF<6~TW(&8j(`Jmf_sK0c
zY|U{0n}BPA{mV5ev$pg52h#0x0kt>>;740ir|_L&<|AQ1gSxt8=?)z&_r+gnta%T`
z!o<kJ62qqNy%PbHanK4xjA7_H!8%r<eTH)koL??*%N(WXc75(U-yiU*?n};~hb%F6
z5Q`>b&<FaU>|ytOv(hZts>wD(xPC-joQ{$b7PmWhD%;ZI4A1}4gaC^off^2AW_mha
zgoE_HiBb317>JB@w)%Te!}Y<kLM|g;PXwIaH$JXBhpJr9FT#79*?Uq7l=~(L`|*um
z+?5+I<3FnAJf{)af5b^HfK@KK>^7sfn|;sI04?%uo1fL5&@$ky#x6!@_vMTKd=4nJ
zs9UF}E$kY>l)=)vHu?iQAsNNNqn*<-Gh+nZ96$S86>a&^$2xCd!g+=E%^-u{OUoUy
zo+}wTGrxZigCzzpenSNf7Tr+uFIlJi3y>mr_s8PR6B85TxoYgW+QiZNgyCR{-^ZtP
z_BK^ei=kLM@5I668Vrn#%o5wy*PR1S{`rl-QWJ++8<y^x4F6zA25D(&E<gR~I}<$D
zeLQ>H+rP3M*kvNYHlsCtOLw3xeBFP3^`@|}P<W?9xPWD|Q$f~7_2A`%1`9vG-4@+j
zpx9ez)Y-MQv*VEP-k-o(v)LadH0zv%JL8_a<6zuu=wT_XcQ~MtFSJXwo$6*e$GZPI
zzc)!q*SC^>L$~kJva-FOth+zm&|5=0JD;I+n$=?gxtMrU2h}t}yT4fH7YFA&dhRT@
z?rpb1c`T-;XvlR6*0Y%40l>R^Gw9vST~6>Zu#curf@)hKMn2Q_>lHhok6i<nV?BUf
zvJbttgN}~#hA<O+O44f^cAd?BdojGyeWIPVVh8R{&lMHVm6$pAlsMrO=ue&)D72X<
zyh)=`h_!;@;a*791%lmSg0tkFJ5xFXYXPKOlSrLyupmjcwt0ci7`<M@z3k&YA+57H
zRk>^apAn7ktf7l!^?zY7eQU9N4pLUzu=QKVGuIV7zY1?x2^$z2SFZm={&|b2q_Oe1
z4u~14uh7O=@KJ(d;J!rf45AUESJA%Pc3=hATl@yGRM*hJ5dY^L#&954@3&RBt(_D$
z+FoC<?J<r+5I^orPiKNOj-2=Jb;s(0R+WXoO~cM5XbEW92e0MZkGI4O*ZSVW42{Qm
z4x!}uM|1#w{vlff4M5;y*Y82&yP4nw=`iNgYm(&k`G$?eJEK5a?m7R0;xRY`u$H#y
zjR{##Gs?|^Y=n&0mDwQZgxleRn9@MIZce0BZgyoi^muO9dXkO!fCGvB2GLDu#K@PZ
zur%a0C+7Ub(sJg4tiZyFJ@>sWG~ff+7mo=ot`1mVRq}wfOb2^)dtF5b!Sy)-nqvRw
z6WVrQoeOUS3$vxWkGv$=+ztb2O^J1tqSy<~b~l7k1W`a_3A@{MpTTzimDeg%25g&g
zfz^NBp?d;Jv_GIfiwF?U{>%St#Gm<r|Fi#V$Q#+0g8%Jb2B#MHjtVC?8{!#S8Q9s|
z8tPhN|7ERbhR4pv#zlFH68ld90X$YkmnVjltg5nlW`_Fqc&rKzdiMLjxo@d!Y>3CI
zVrpP-LdkpU79Ojlp{cQnJtYtJ-}g-I?H(E0idb7dv9>a_g5Tn?idb7%+bTTK)raSa
z8akTl8%o&fy5O-sFtjp;XK=H#Q6BukV-+{Gus5`&WEHp2wKo(s)VDS;#1j(2`!i7I
z{+TaF@sAui@-==p&H5a7JpA+j|MUN^!2kPJV7T>&Q89h?|MC#>{B;Qb91%8lHje+h
zN93ilwB5jMLe!MXykxW2t)b)X^;9y`ucKZM_D;U<SbqBG%{NA|25O9i2ZDq^Dv5aj
z7eD^Q>dMS4?$5E?U+Tl_=a%aGPJH2Olayu(-eA{dKd(vMB*w0DeJ@Fo`uT^*H#|G`
zD>DPri`zf|v%AKbcq~PQM0aOtb9-sJ!(B?z)d6wAAg}S0cARI4P-RJ^me$aNy_h{4
z$1i=~$w^L~Io;-LeU`(?^tt}i9ao_^!ro!8;v<a8zK<g>PSp<{w|Q;DLuN6O+EWp2
z^19;Dw@c?*jm$&ct?1BeJ?F}X6i#O4K4T5+W^m=XBh}!=;D6kZ$_KgVSmf}i_s9{-
zx*=H(@h7ed-9<_l!ySxw3cgyE-mbgWV0B5h=c;F~#qis3i$cRZ!P5_)2PvofNKmHT
zzvp%5-d$01UlY`#kf|9lGWSL_-?n*}9BJ6_XX`S$tO~s~LfY2E4-=(oy-e36<jTyh
zO=^$UDVrldXp6=vb?i9u_)NdO8*|#&SBm`Y`ZKmnnM+@eRD80FrFLz@d5E()ll98!
zUP6VtexZ>imH#412u=EJU9Ps@(hm2iM7<`CSRH$!5MUqcBPo2Z=H67NV!T>(R(a9g
z?~WQBeFG_`bZig9&soynBq^M2zyt@@-0;3jC3@D4#Wm>x)vJwOOwWZxF5kCbL(bRK
z@Tr}VQSPf9wtM<&f#oN@{O|2?s!#8}q>@LLAxXWys#1P9aw+BOPBYIC#b&jA!sm2@
zS*e=ick##!y{N^1UMy(N)rhVQ7FP?fE#Th7IpfZx9$?b4CH*UjD&jn4<qPsfzSmKd
zH@_UIW~3!BEn78`Pk9P-@*0?dX1@ml{axSZvsMi)NwUMj%tX6_UKB|+UQfuv@!{*M
zU@84f%-Z$l*Ud;8!6k3jmqFjBPd&YT?7P=AcPOL9aZ2VORRK;xb(`L2C7k4E1-$!x
z9~YT)M}P3HpYM<%OV3Uc`ILyqm-~b%`n|}#Q`CyD?OjjO30~sqcgaUmW<-5m-oEcX
z>Tu)2{SOxC6E<ZzViTQs3QDqVy$A*Is~PQ9_=K+>B9+vR-=jI2=onc4wD|<5m;5Ol
z0g3ZN(K8ALpF~WukU?&ilA9`8p?iH#$V$#r_f`9Rd%C4vLSHAN*7I;(9c|)$kwT0!
zK7-8fUEM8pP0X$KTYY>@Jw3rRsc#aWeJLPdRkS2HeRPs=gW=dV8BfET53cr8gXb^4
zXFZCXPBE=A2uRhUE4jm)%+`4HNqnoA^O~lU?GX2}bg<0j6Vk2*^ZsAx-chgnD@JMI
z>gy)9uA1V%u3*^`dhuo$p|(MAlHx5*JR*LpG&bPP=?K$zaj)JSqYq^Ia7p+LT}}4*
zm!~G0vS;ywjaA13HtKA0`~0FL*H%*nC3tRK#>`r_DrkJ`>Ao*OCAZ*c%Huxl7c|H%
zjpKQlBF$j10YPU+iYnARViiSZS4<fYr}A6uB-f(5q{!B~n@nn4@ZAmK-6+XM#_v+-
z(aEPLUh{Pa`8dtoTs<26Xkmwh;X;~Lsc`IDDay_#$upF13J07Bd_DTVPOD$t4OtjU
zSlBsPR_d1!S`rxkmHA7J(AO>EOVgPt+*bnyN<L5iUc>~du7BO5c)WaG>wl#VZ}I-6
z4*#FDN#rH&ft4epYiCZ$&c@CTZ<YVYMB%)}!^69;fd8*el<7ey`-_;bPZwkcyzZF1
zmy{4q=R9%rjM!1JdmIhPj9%1Ff*w+aJi31QUeJ)+k%?3CA_OtllL${y;g`58&r{Q0
z{#-pL*i>RyBjcFYXAvXNN^$K+OyqRabj!9;V-wo*0+4Y3pZwu9066o;Ym=^YC1BP5
z0dC<)mCxwtXq*7#4(cn(XT9|HZjj!EsC;^4Ho#owI-Gjdr*O5>q2tLL`bB^xmoy!!
zXI>AYP_aS?^+a*}g=36z3DpF`+n+nU>z+WPrOvSmz|$bXWDU?t$S`3RQBH1dY-$2M
zkU=i=$rjU4kK|)wk|J--2V6f#HAhb`pxV=OYlhoRJN;zr<)NXWMNuG<S0}uE-pD1g
zx6836R!2uqPMTo)1%c4xHBkdUe57W9Sy%M}D>6vQw2GRVnv~m0v(Q3~*(`?7Ve)!&
z+jL*vj7u7+hu%d_WYm<Qp31-#eoq8INDo{sU?z_V#0*OCk)28RgH_*tl$r_$%--ZX
zrLnQ`LS0U3D0;>Pyj%`;_G!B`(xqw!9%8P!#_Fs>4#(Zi(kCO&mwTl`7#13V3p$vZ
z-L{GykX9RB;P@8StM>i!0@o9K2JjghQI>sq#sp`Al}s@ZL7)@1U^aU>J5-?o91^CE
zx2=aPiln<D=*7R}q`K(PITpm^?DeQXzI(FI_|#P6!&h~6_>BO+zP>UAi5rV41oD7?
zdv33@H3lpBiG`B$+8WI5ZH@*ClkM<!mpNIRn)ZBD1LA@Xt8gabf+#V|vi%1>c<>rJ
zDbMv1s_9ixL;NJDrI41BGXztgUY%gZ!p0^yl+iygBWEYb_K9KO%2a2vG>NH)j@t~^
zZ5q*EgP(l$^z}nW3$4+y2j8$9{`{;x;ih_iY9@;Dl^A(X!U@N48{qtL2?O+{Da@;`
z{3@{=Bv)Z<8bcVX77dj2<djyYvMXJkoh6Z_fO9Z8>phk$<mjG&WTbZ9@1k>+dqP()
zQ0G~zdY>a_3Y;1nyRL2D;Fbt~v_sy#V;;2K+ua!+9u5@VE!HTq0`Sz|*Z0<~1&GwJ
zy#zN-Hj4B9>vm|la}?*KZ`(Kjns-SL$zDi%3Q|$sMJ=Snc|>tr$40bBP#`)1^=88W
zqzOdUl-Z3k0#EDe>L9ScJcmY)i-9K=&1?6;*`nrkY%8t)j*{t+QsOi?#(+_h3VW1i
zXJ@N2#vxYwBG~i+HFg2W6p<+4vM8$V3emD~ioAHS5NaHPs;H@lloB425cBs=dDoh3
z=0_i@`S@{sc`BTU6uUV<jWK<k?hOk*1QZP-+9kH%<|$-kWI}P3J+3}N%Qtsippifd
zdG2ll%(c6_t2YV&8io7r#yB=)a^_6qRJi_92~#M6s=)Os9HeEG!2HZWQS%DBjjxt>
zVmYeXNG5g)Qc;HQ8VunhosTj}c*349YD3}?rrrY%I)-ATp<->wjZgo+<zQ)n-nUmL
zAIcY*cF<MvpK{UZh`;*+7;bvUkgTXDHa7J_z)irk$|J5_D6%d01)D-O<z~TU*4Yb<
z)<*osjO%+1l%kxve0MjeSlMEFbUd}t_COv|<~YL?_F&cK+c}1w;F_^@s@#>Hn1#xv
z<~{j>>R>6Ua>p?hg#e20YtwxSyB8n>kNZ}d7?PNgB5;w1%?;E}xQ`-}^ucPBi2>{)
zk8<B2t0gLeNLzQ^x6oE4Ms3oD5zn0!l`2oG{(RHL&p5c}n)Vv2pE+I$SllG?v0l7Y
z;kt6{8diqw=Z3Vd70D+{ozyO`++LK9WH-#UK0eu*mdFvtV}XT<{dw<eA&FVe9+5X~
z1Eiv2wcorX){B>x_S^jSE8#V)8^6>G5fO`Fp22)t{VJ?%&`kDZ$FX&sr`F0d@(0Zm
z>|or4YS)zRB9diFgDEsB)Ggp_S`HM}<8-DgHGWRwbDU0f7XTQdjT;Ds5(FZ6jO2z4
z9g!dd$spJz0Rt(r6jj#ok%mg?0|~zd4S5=-q?wIbSy@dl1Eoz1mlSNNG+-q7?&q~O
zes&e55u83P7hwCV>52<;2@$PSzG=q?m!)4X@h&x-tBdp8?fWG5twql74eap-5F8vo
z?u*Op=`S&v9IjpJI1`16+*zol2(|94Lsyy&6lO+zc}puj`D5i*v^9uZT!~fpYfbh-
zt<YauTn!Z|lCN)kYI`r@gN2Q(D^|jYPPgN_Y^}}p<fJYk$Ko6spM-Yw)%TkXVoI|)
zNB6gR5|q>J`J)xT;pjfGvuj)sd|>#KmvJsAxjJNJrwvfph8GBEB`L1nxlZrBX!^nK
z*n^zyZ!CaCwW5xYu&5`^9gQ|uifT)6K}7QWxLNf~Tc@G*-1Ii*MbQAmHi!f>)vm0q
z^{oB8l#Outw1)Q7Im=iWnO(6O)q5H5dR-N8qloY(NtTacSjV?_ehz3tl)>zK{4?7K
z;Gw%*m@3b8_wL;W0=??WTK}slWFG)M+#K3`bgd0}Tt6gG=K8M6xMdtDh|5DCuap>A
zpqQGA%@Pt5L(N5}?|bg;IAM;+#tCT^T0|1DI81idFN|`GdWPs`I2b$Ym=4jG)5a>c
zn~N5fDra=1kmN{WgwyA!1CO2LW4+I2h@f~~Xx1eKnM+L`S#FHLi$*2l%}{y!34mFx
z!h4$IrIUca!};ULlCo3l^a}T@nqepVy^_2=9j{aA(fgu~`D&Ps$A*|icSve6+({^s
z?D8~%GiN$BmPI!ze>y8%Q=MK3yQvj58n?UDDvW#fa{RkRh{V|q_BcB`1C&iudm5tE
zQY3Z34GTzJzA`z4b>OL557>{zP60u|Z?jFGjhgFd^_P@PYm^}0k8ES;^XEG_Wh@w}
z66b}4dVaO5S^QO7QcFT3EDK5PypE|V*|DfE&~lWSV6~n;Zj7f1%5d{4LqkSoYj)R1
z2|$K7GMQrBSajP!t5OKMa@JC%02d+|du_*JgpE>qS3+2?A<y?I>5Z7+bfpYEf&_Of
z)!i-h-a|-n9mio_P;XiJ#t)n+IKhHWv+rz&LFtovT!!dI7;YByQd~tNrl!rSJ-&bL
z455e*Vb#$duOMBh^e5$F-)$TRaS6jER%jW2o3oM5IhPK7oPy>mJa$T-O_cBLZY{n6
zS@<wxU|BYMcKSy1u2&H#s`1{tXbtfM3Xk=>8&^_~D6Q&7-L_=%ydr+h8Bk(xv?!^d
zYyKtXQc$0Wyw$P#3z$~z+{$R<A*D1Sq`FD?h-x-#jyfhU!4Fc%hH-;X_>Joeqx8QN
z@7%dl>9GTmITgqiokr&{YL(`<)PC&KDOXBYWOI+A#2D-9aR*JggaL&7(fuq0FK>Oy
zW`t|E_Zn%0I<&WZ<}}ist&=T4A3Jj2&trGPSXNfnB?Qu<x0$a*XG<+*kuEr4$_lLp
z%^vds!Pl2BU-%uTi^2E;Z7VZS-|skhrB@|t=d`sSeHVIfyPyqv!{VP!+y2P0wbiRv
z-)Njen4#n`umlKlgAn6}USa~sfNFoj$)UMWQg#F1XFArAPD*@XVwS~?Pl|iO7vohe
z!uO|$Jng4nLFU;Kzx0!9p5{av-X*3woYqK=+6Y`UOxPvl?u(OGdV72m7X@T2A&Ayc
zwz)cEzV1}rr{EzFvJ*b+qRxI>2dj4AW(%K7*RCiX;mEn%5)t#!vM1&lBDpVaKFnF(
zkMy<$gY0VA9E#&sl5=Ri2p;9ck23R~o0zQ;|Et_#yIM(-=VKI1jEsyLShb&pO<U(g
z4JP!VAw+ily6ClH5ubC!h-W<iPT-XE__(vEDyOhUC@GhjsJQrKbn)Uhe#jQbA)sMo
zyw9sfQ`uW+KOyfD0tvLFy*&nddh)B!pFbZPiRQM>Dwp2ia7XCywVQtRI^9ro?twOT
z7DhT;uy>Cheh0dhCvueZwN64y)^Y<`{D>_nX@UKd)0Yw_=+^!DjowKN720)*CB_Yq
z^fMDTa~@RhK_Z4Gx^W2to!*`pT}R~P7G24^MSdJUkandsD`XU0!@k+sa*5*g3YTwN
zT;8T^821*lI9mN|1PoF9I7``$pIT62`HTu0xu(B^l;1UfmwBIa<nplr^`B9{z&vVj
zFawgq&4=zO0---3vw`Gl(Bi$TeFO>qI)5errVO+(Y~GbTF7<8I6JpYkQ)-l-3Y&pK
zB-6(3Z$`d)CJLQ5bv^bTyQY?=roGW5CR|6^!%)SAjWR}om?7{4YH!!=hAL+*I-TJ{
ztyIGjLT&j(f)p|eft>*>#$^{`edmOImn~HxVqSbbrTQqu<F7)_2z0NzKPr1B(R~2|
z4-KE+z+I^nyVp@SfmQDt!=Wmt6=%(>t8V=wvIcb$Mo%jqWK9y>JZ>!CSLtuGjXm!9
z)#>N?Yo}9|73n0GzLDto((12EF_g?2SfXZ^XKt|MC}LHZ+Ed}}`4=N2q38@n?JrEc
zylUgowv{`p(e6Wy5H`O-PaQX-fIc14WYwmbEpGNX!AK`Gz)(&D6!XH$N;5aQPQiJu
zVr~BrNNdY#T5NlNo2T`!E|k6|UuuxT=C>2_j6iP;Ap_2Ve`J)c6XV;;sjTg>wtUdH
zGW+5Zrpo16$@4`TrfpY#JtJ2tl!=_*u?f<k-3KQIRMhp-)PLTm{Sch877n7pOu!sK
zH#QK?qCtn=g$S8MsoXAIx{r^KyWumiNyj%c1@bgH=eH`nS<FCOel8)SZ1&{^a=#iB
zQlSCrHf8QE43d#KBrXtKkAb?0Jur3}aI@`JZC!K}@{DBkKYXx9LhNhX4KK#-g>z&`
z%hkLZR;-GV{zzI9=do!D0a!S1-6fo_TU@rQZ+%ef{5ACujcO;_H-$!O1WFiaXc8rZ
zsQY9SdcnPzsDU_$63Pbdpv4f`cRW;ssoO2Fs#2@ga(UTO%x2qmy*b)CkhY_ut$4zN
zuy4Q{1NX(^8?bBGZ8w90f@<8w<90GLHJ!MDhqJ{7hqf+cs^l(nPJO`$cBkuyAdU$(
z<02R_H(CYKESe(N=%x>Cwd>|`+TyLdE;7^z#9}@D;ylDkL!Qv;SUA<ap;~++5kOY&
z@e6N*^GsG&7H{lo?GQ6|{q$kmwPxq$0>#LG!bR+2Q5!%$f*Vv}sS56LcuLW+q!44j
zHnW10t6X}UB@Zouxpospn<^k}2RH7s+*Dd+|7W?mxnMzj6W7>HLvG^_O$*=QFrqgf
za7eM%G4<Pb7Fwnlw$g<1QrxAbN~^nIhPK;8rmn7~+f}I$8NXjtH;%(S|B|mPH?>bC
zIY11P&G}n;{Q{D&ddeKaW4s<OUiwZOS0$9!&4It@AG!OZ#113Jq>^RvSSNVY;iLm8
zBzg&K!@*B4m!3w9mHlds(}`y`^Hb7)5E&WS(E3CSwApQ4?4XSA$QbFQ6X*Um&#s`8
zb%aEw8nW^_9h|U-->#66G4EvO<>j?##4L)9Ly`uCfb%EFSR=3PhpNRKrqDUD@y)Jk
z2ubS8`8XpOTUn!~9O8XHrOMo9mf|p|oYV=ZSg5&rvPC5IM^94}SL1wAtgvU6JB<}M
zkDG;5dX3gi`WnSgbirm5e|?#1J9wvu8%`TXR@B(o7}(cmR?j3~Mloe5CQF5mO&C<3
zO=aU(1I3E7C6y;U!eY<{*hWLJR)JY4TVg(V5W7^{9VxOHb6nrrY>uQwg+|bYH_$&J
zoK~2q+Buye?~+poDG^s99zu%uM0;{hsz^CXXnS4+r^kW%Yr^U$-7DY)E`BA0bPYCv
zZP@bAK<Kz3g57Pqo%IEM90kx!F^i(%eCUfrWsfmvK9|L$k=oVaFE|*9D#&YhFoE;e
z#W2SNW%60${cq9S7Z6L$+<a?HttMLOifLpik73Op>zJ<6m%n6AFAqxyvi{Ws9_}?-
zX66UJis&m}Ln+W_wvj1b%X8ps4dy4CZC|K*P0Y2GuGj({|D(@7kSaN{qShIa{9Xvz
zKQbjPq!uZO4o!h&x8!F2dgNz>)-{yp-mW*g=A%YYmcSD~BHCPGTMSD4A`wx7wVB=z
z6>S$WH14uz<@!k#@Q5)-&AQSJ*Y&C!0=Gb|TMd?Cz7unern@YT`EB!?2leh=jZhR9
z_YMyae^yZi(dj)#CvM(`?!u{-&#3lkS?4oq_7G`bG#Dr_H*R8R=tzXrrF#?Y)8-Pp
zYQfuqn%Qq6GQ|8{^dj1@+H?-Jl_hWo3YL~-m^e5THfH05kfeE>oyKx*ZV12H(dlXC
zy@2+0yDTDIxf|B1w=H_(^Rz-+`BL*tm9%^1&`P)LnbW(>?6y_MV11OVHXyz8-KftC
zNGFqfC^Jo2T~%juGo<nvkppC$?Mc`C%4mO`@c}%(#R_mgx#l}8goTAMhL3XhzJY6V
zV=oNF5!C-=@QQM<OWZu113<arHFdnRg($&pBah*r^7R4eC>9~PVK3WocJ!7C65yg+
ziDJztCI{_CaP{ijZiD{rRI}n$@2m^=YhX6B&StfttT`RyO9pt%x!!0l^AGL<fJyww
zGY(u{cvZXxm!95@_uC$d0oMAzOyo8+?f`jo$k3Gw(eYsEGF1jKU;z!3PfvylE9TA5
zUYEXH!{_qg3WwVpuzUbU#YDTEcrsR{(I?oA^7Ho(70wSGN(DFOftgwQYqFe59*^zf
z?HCaR#dm2&&-cR^OcykY_VnCCA+{#pq3V^VTINyh<=DMbOmf%ntCM!aJ%0Rm`La5k
zrO`JH9dD(<9AFWa?y95rc409aQrZCN$COn+eE3k%bs6i7JW4K>7U<7LZA$G5y;wxf
z!tB@gXRnf-4Nim$0D~#|1k`1NpTI3+vf$$hskGI}H))$LDzY3<Mwi?O-HL*@O_Yn#
zSl5U8UuOIIc|ZtQn9U(~%(k2}u$R2Jv~=>v&1;_lPRXn}iOj+pygB@J9=2{!Dfo;K
zgBitC2dJWPk$g`>E)?ebID}_JH5Dte7`NRIrmI)bE_cphOFYFhAz&MWY-yyWqnkV%
zImIgTI>#hS_Uw{+XL-iroTtZN+a0~1KZkF^zHYF~MT*T#u5>|JN%^(!=XK>Y)D)M&
zv$`S7*%J<mkf)UMjzo+dAM9n*=T1#c&9m=B#5X*?Rqn2^>aU)MJ0JP_HRs&hoJ=ur
zBX>O(SVS=-68I0yfRXy-FH9|k=SP_G>Up?^%?YMjE^|eir>VCjG*Hy}8k#gQ{Ip@J
z%%msc>*9X$?bW4FkzcOF3i@GYFj~0c%UR#R>P$g)0EYm%jW0QnDsG?BjuZ3uj7)7x
za0Ea^`aapJ)F|Fn90KaCqSVfz3OCU5H$D3_#sxFLdc?S%j+ove%^B(H6P8lJTo!PC
zA}{}TA#et6UUEMBM=j9OZ)Iw(zo2oR;5b=l!Y_3$d9YE=oIR_!YoZnTCOtcQ4hWEK
zV;)LLy=v*H5C!OJ_0#O@K9#8bS?E2h>#1T=&o>Fs*=o`TKrd8>Gn6$wx&ma5N!Eo5
zVMDzq-8FxqiU^?AGotqM!(ZSW7{>*uN{!YAhAVrq5jRgYipmfd_<*ZXUGj67)R~Z1
z`S}en8nr=iz|+Jo0OVf26ZRCMp{a_nvHQ{EORlG?+6!Rxx8u;f*<*X%2GMh+sQ?&z
zeeY)i?p*Wwio%BnYEYx+?7WUme%;{cdj82QWCtt<0LKZXZ2>0NV>@}`M5bEaF976R
zVmblKV7Lb9;keW^G<FqDuLiUD?O2rZ=!rwN&4`eV&tg+uT=ja#)3--_C~jN*xD5BC
zWdWq5ndKg_EZ|fu>}oJy%x*gF?JrD3MSUqGWlhkx?vCg7g3q>jl+%!l&xv~m22k}H
z>`u#$5PnTf%`7EaavO-kH%^2`MARLlV*ij`xvTCmQe9J%d&l<^!}-W=NSGN5wuzh<
z6iGOy3YQm%TC$1^7$<6!qN*Xjiybg>47x?Yb=eo5UJMRkW7&JKuk!Z4k4b?wY@k2M
zz1Y7jdZMtYsp%v>sbqFI==vP&2hMmg?9BjbEi}>1TDkr87*)H9iAn0R9;3pC*E|9G
z&j@wph))HnK|eH~<y`O#Pf5REsD(a-U~({5$XA~yJL7c<67Q+)Rv>sFoq}?5aHw!u
zqJzh}K?-URt8P;}CDSLs?U|-M7x>Rma}>e_<;ohxX-XvhLNev`>(}6Iy&{p!GiruA
zxAqtn!2SPU_~Q_=LLG)Pg=8*jP(AgLm72+9klhLEphQdqWbPsDu(Q3?y0_8l*~zWr
z34k3W(8Is<40j+gYce`}Z+ERS%o)=!f^}dvjV3|AT%4K;*YS`;f85!e%1(E`wz~qC
zBuE}DG07O^;hM9<KT-cx&ICL8zxjz-3fpS_p8~<dKfMz3`=@RWTYGc(ryPd!|9lTy
zU~~8<V%oz)%2fwr6`~0;l9I(POGe`2Cl7r&$N~7DA)B)%e9wG+egw$4!EZQJZYWC3
z0!CHZqoD0d@EZRuD#RAi<o@7KAmYb4cI@ZqD2Mj-T{fGaq7$o2<E;lR{Qu04D6#^m
zUM3PZuw~#fMT=Q*jvhiV;s+OiflW;h4lmP^<I1GcL~|=P+_6<L-WnG^Tj4lEdU(_>
z11LmUl1xrUvxl{->*gojs_F`W4?Z8N2V}Q~={%2uZsbr-&|vcvS{A-Ky1ldW*Sf>n
zHmO1l&2sR@@bF_q73789a~)MIk2<{mO-kTbj5UFuFm=@k5>DH15uN$3r@XxtJJtkM
zzY|x1l|-=p78F-BV9PXScR6wt%JiRAqf6nql6TQMhUgU~pf(!n>Ilh{(C&Vrw6}LB
zqgf}G0{G3Q;kucOW_<rB2UXoIg2<zDqz3WBLkhXCe<yMyJS#FOIjv1=3_$*U^er&F
zS{G850sY)qjFX8IiV>3%6*ZUu#J2h#rNi}r%<JNijLVx%e5uw|Ztua_`REgGwe2g@
zvfztoAi72@tt^Ydn{#rp&|$hYDoL>4+H4x&5E#M)|0TVt_b9b3fTk5U1VGt6kd-Z}
zCO%#DhL5P#xIPk!kbAgaZ(l)}nVDr>7~4v4*9n<JE+XVmtPN{o8z=@e2`zgrF2Oa)
z7EQqIQMzLn-bfAH!R~cx6cGn4k1JTsV`Byfn@^i-3t~zm5pc2P?CH~O+-Jw_L5w?X
zSfZk$T){nr=s+`UU+*VfxP7dE6BXF%Zz!fd`l!lALPFwCO;1mssXT?jC_(rGTLNT_
z9%^l6rKE&(>Jw<sJ%pYf^??9^EW$>QGy#G=arF*g&kkg3zmD9OuFhZB^xgWmv=AY2
z2Y2JlC8p~N3OspWY_O>3&o3-Yfabn@hwK$xb1dMQG=wmIeFwsIUYnfL<V|rA*m?jL
z+?0bNt<tAc8QG}*GzEG#0~3?q@aqG43Sro@94c3H9Z=v?@_l&hk|qweHhQbbL{wB%
zQnJp6!sGYM3`<*EM@JrnD2|1J-tU8p|K%AHfJJiyBc|5GE?l@!RaFHMOq!&A_%DEe
zr8?lIK#clxgZBp~EC7;i|12tdCgca#CqMwL#Us1d^Zwean0;l@nRV^$Ck&(?SGVaw
zn1q%>849eRROwhUIP~{rWVno|>9!pw0CoKu2YX`Cp5#ZEHZ}yoFl#__RkES{a9>3v
z2&@3$=68W!O66V8hQnLJVK6U}W#{9B5Lm@TTo8<F2bF9P3<(2_nFpY~narXgB331v
z#M8kI<8!~Eyaff<C!lRA8Ok!7q<0`52DtdUH*cOEXh%_T;wx7eD>fLLu8@$>_cA_>
ziilvit7s2M7Y+T({dZ{o#9X;}5vaOoCTJGHmA$D^$Q*|r;2h<RId@Hf$9aAja8+L)
zmDES?A-Fu-mxl{W4h!V>ni2Hcrgo}HweNW_m)~Z8Se_|DlM0jX%P1|M>#|9a)ga(?
zkF_B}hOGo6C$|h^jNaK!fJA)IVY2`ofzOsK0cF&h(cAld5SgXMRB5kY_a%R~7!?cE
z4FaCK?nXv*SmEnbCLSzzu>-ID#3UZc%^%r1!q4QF9-?i>ODmynP*PG7?)K^ztZKPq
zm3upL;9>){0(KpPMy?5aFt1WVg@#_y4eh%w<mA{Xn1w;UF6c*{GDpj!B>-4si^Xr=
zY$5Wj!|T!NFR@Dr3=D+wE9H!tj+DSI(fuu<WqP{*)Fm{8C^J;DZ(dXf2aQS|seQiQ
ztmjAt1T3DWu8JYHRx8{$*#DfF{cp@PB%nREEUUcnU*0qCN-uDnQNPg?9-1w|D!SQc
z9Cx33WoZfd>eZjyXaDX?ZDzqLJ#2+PpTzIyC(hUkuvXwPGdaMChBXCZH5V6`6DLoq
z|Mu6}pXr~g?3X%t1#?3cQ1=A59+!U2*&8}lb#%m`cmxF25P%zAyOCX<3v0Ijr9U^G
zdS3$6q%EzkIyyLzL*N?g@CfVN&QI-xtxIC9#S4c#4AMV)zkeK+Kbe5l8*PnZJlhQ+
zK{z^nW+)3QIXPQfTL?shtFD{@VKdqL_y5#Q9!%;bb5uwOClnm2sPsUr6pKt0#<045
zO6>&09^!>eG1vA09zM844=#4DGO?GrNqOFY0CwC<IqUEUDV3>JRtOzQtgNhn!JuW_
z`ryV9J@^(=cvKW0#BVZ{H3}{IzJGtjAIa*L_dAnB_}AuGuxwIC#x()wKOGJZjDwAQ
zXbNczwji~87pNUKXa3Gu7A7c#^ajCzg_enA|3g1@@Oe57d#I@a#d1@-d>z7>)0U{|
z?u7QMN#Ml1@zA5VdKLNS1Y$?jCg1>T3tM(tredjuU%7ZPxq+GK(#4BOU1ne(SAyF6
zdn-u$AfIEZD`R|YjMII6zOz$t>m4h2Q@^3d05}SEHa1s)!D~S!X<t|XqJS@-`6$+t
zsi6WFFnChdy`R3sjyF;<GnGQz2@(p#6HU9)LxXgR_chO-Z^Vl!C@26#&(zro1^`$c
z0~_wp7c3d7Qxm()GL$l~eLfH5YX22^l?Zw*{T*;cD_xd8?6ijKt10G#+Ji<NZZ0n1
z*|{Bj7%TPpchr_)AYh=_I5@;`nTI7dg;%L5LSSvO%g)j9+}X4HG-*Wp604XGQqB$~
zk}T@W1J;!e@f)_H$X7%cF7P`{CbgMBX8#4mobbL)bg=p-suRFknV+9el}~N}{Yhqy
z0$))caC|#kTQdmL2UHsUy*AIqz{cVv^(T~rES0L}a1jkfKD_T{1(4g*;IXOv@30jI
ziDjgvuikq6^l2(Ao5;5OeFX6(OEhY*u`%P`OT7J*df3rFE9fhxATN&{-qALQHSx~6
zr+rRH0F564{-N{2Pq)?SeaS!gw32mOb#=AlO#f-$01Dxu_v6y!#Ft`(J$FM@v=`>*
zRg0|1u`3sXbz+#)^^n?Sj+qc;2e1QTY5_tz={czkbacPKmUgnSfs$9MzdkL66qk_b
z$<{$PH(LUv$P~fK1x!e=9>gr_zExfj0YA{$`{$M^&qPsC5u%l#*e{ZjrUz`uM~Ycj
ztZ#|}S=<Q;5@yx?+vC6xPOnLT045|fF>4mj!v&zop);A6`Wxzy+Cx=eJb%9b#DimZ
z>EhL^8lX&NWud&-DcuyMXeuWMoI&6qo{)grDsSwGICv_beOp@_k*L#j@B3Ueg$yw<
ztS1WUb~tsUvHP!$Ue$yRdusCj-pi12Al6$&TUYLc0TCCV`(N4sY1$_hIFN~RDnK>`
zH^6c1Q+0=3msLB3SeSOBqoVrXi{O4~vqo*4#IEdtMsbJ-TDGB2x*`Mwfq-_A7GD!2
zE-*$8xh@R0DQBve6c>NgEFt>q^S)pUZp@a=K@%aX-cKW8u~Th;H-7@1GENSxg=xwe
z4$Hwd`{x8Z9+p$YkZiNEuwY|r3ur`;{TX6FWnY1v4~E~|$jHcd<9qwoHdZ#&37Pu=
zWb+qa>iR#3d-Hgx*Z+O|oKv0lI*C!D-4Y`$6bfxa2u+qOMV5-8q)1tE+G9>CONwxs
z788jnlD&mOq_JcRCuIpGTOstl?$;Rg{ycvF|K5MRAE$cFYq^)_a$VQ+d2hGLPua!P
zNg=jsBLEWso)y3Vei`NOE{Nj(Agj!z?i|B(7n6zLCsf`vBQq0$4HZ=jro-DA5!N3Z
z9Qw#lm14&}!)fAsUxmn<62c6nhEygZSec(<poExw?l2fjMa8LX@Z%h$`ZRfymh-FY
zaB!Vqt5V<g={h(M=j7^?`eSRBELrkB)7->>CI-~vc~w;^ydEm1C=)C&_4_7IpALWQ
zfKC6tq2Zw9Z|vAY4(H#0=bzUuh5P=w@5;_h^fLqXS5xyHin@U_PdHd)jsRx@J)V#Q
zOaxgP7A}?(^7FT~wNq{O&t=D!kWd1X4v%BWvm5twN3e!`e0^0Uicm8(W%?P0sT1Ey
zX=!3P+!S^(+#6;51xq1$4+%p5lF~Znb8n*^GuhI_y|ExP0H+Faw=R&EpY(GPBpX6?
zDEb#biuC1U?3rR#Chi~dW?;F?7=(W~LK`>wqUL28Kx6M7c=z;q%@{UBHb*s9{bC6W
zmM&d-4|umyX3sVUhntsf>|&N%nJ6P*dcsHCI=YB*IW&0{m1G3Qmhp#C8vp#lX~yIe
z3<i7*G*eX>M_2yx!T!TT9e&Qr;cADPJ~+|omh0#^1$8STFsO1)pG7$l-cGPnn)U(7
z#lrahwW#f!dD7qJCh%oPU0s2?IoG@^ZyjAj>h~C=%N8O(ga9(KvbYE9Uzj%4HEW3U
z@Ry0UK)W3l7GXh{l54fvVxA*dsJ!}mn;U#EE|Y(bV3FTE-=0em5mr&h5>7ED-)EGT
zl%TLlBi_t;dxia|$!<MIi(IwOQnR@HBx7<ogMr8m({b12&bA#pB!Bs4%kN)5p^1?*
z#^#vJ3C84q84RKnVkKio>gwx1BVo2+iFcEb2zX|k$(@79k1-Yoa;mE4H~9a|Sv~ES
z_j5+3RL)8sKTR^P**($mcrrJyqN1Xz>L`rZFW=i~fST1C{ITQ4HTrEcPnvJl0~8Tl
zkrIm2U0mqYE9OOgfAWVeZD_s6M0Q>3AsydtX^XPCbMuim?AW;zuFc`{9b~U&R|npN
zIq>I|RbpoqsU~A%KT1#c>w9<hVdRo3POFtoL*@F=;*GX^B!gFc@0*9ny#Tp)EP0(>
z0hEw~rXKj6k&AcvY8!}zit_Rj`=4#r)xCqP5$+9nPvMI7y!)s2m8E1MwOwp&mmRy@
zZHjZ<L_HfxBy11&?bW-RfKhvpC3y>JYEO^*10yq+OIyd!h~{_C&eZVJu7|ZntOgn@
z!uf=cw#`MY8)43+x6h;OkAFj_A1cNc{GY&WB_JeXDThP<6X6z)0Sf>LiHO|_1h3W3
zSaQ?^Wq`SnRz2A0*4EZYMD+~}NO<24nP7Q-{vCiVzl^jYtc*dp@)EPVmAG}`n%$VX
zMhG}#`HslC{C~O>`8klttb2+LU#Ll46cijhM}=jQaisIyxd}hFrU?u90yZER<vY?c
zIK_^8|I3#H*u-Gof7v7)=Zbr#Nb9go;aAZrx2%LF<~!6|A#q+tRdtr|L*!_mkkzu;
zFM&KLD*NIw4=Q3!6>vKk+fnbNm`AHkv^}YPW45NxqsHN}SRYmCsNCn-<<|Lh8+1n{
z{Vu#AcZ`Md(xsmvor~R>et6q95nTHwiR*|&sX^5d+2H8dOuo(V_4TcN`*xa?x&uEO
zJxI7&MEY0GdiM0`sKcNsvmsg1#zLRH@_C6LnI+%c=g%(#RW*!Mr}^&vIVYJG$}%#y
z-9MJVT<aSe!a*SA%IwW{rB9!;YKrg$9iQ+Vp(RSP#E0v|e@VfURCYl0fi;cu-p&ap
zBx%06JNBCdgb{t6Z>}m~WTtxJ#$nc8caXATR!2Xid>_B1#S!)g$b-eF9qRaprfS%t
z+)n-@xA&u80yByc!RD03iCEP>u#&7Th%+9xteh!)H<B<RtvRo~5?uwHUx{Pvo{LIJ
z?Y!@hkNp{Q>(;1gHH~J>6cLd(yK4@HN#U9nNor>jN3Cjx7KiH81}0<GYe8aD#9)`x
zKa*o}`x9Uwz{Ej4H)>sCW!Ok%i>s=tii?W}x{+3MJ9nnJ=ObhU**E%`ruKXg%0!CA
z-TDqB2mYb@0fPZ|R^q69sQJ}sXdXL_J@EZIJM#OPsdHRb*pT5!1|uIzw^$(~GaAEi
zZ&>vJF0hP$h)nlkI2qvXYV12K2rx&Zu~r#h5%F*-RGTmeB!$u8aUFqFVq)+Oqb~=2
zH8PZ$Yur0+=1}7%6LRH>IT&v!l&-XiI*eWz@4|od=n?WZm3`G^52=rJf|NhJ^C@*$
zB?a2znqtlGh2`bFEqrn&n>w1h!yop1`J(FmRbU*+*uJ>m0x4w&k{n3lBZ|pv$%xOG
z@eVv;nuqh~j-BSfFDom<fQ2b+e<dR;N^HeHMOT(@yuqj2WJ%&i0+VvXTv<*okDl&|
zU+#-iO@sh*JscgR$fFJoih&y(P5@1@`u=X5j!unP6#zR_jT0%<=p&&Jv9Ym<H=8qe
zE-?`5botiW^pv=`HjMAjn>}EV_?7Rd9VqFfMMlDL82w(BfbxUHcXZiS$CD%hX2cYu
zi6$O+NVx;w>c@9l4<Lk}cZ*@%vXLaR<9kiSVLk{654VK(MjR4;m`g!5By=w0(9WMf
zU(|f|3dWTzA0R#=I#$46Mx7_il?K4VNYn4_alCt^l}Q<voTukEEV0q$pFEF!=JaVq
zxkQpat(iP<C2htGCOB2}9T_!)xGZbvYfW8U9h6vjAhsX}Z2}#Y!x06HTDRo!Y#7RI
z?cpa+p8Rwuix>n~z_<9~IHbG>4^l7)mDbc)DlcEo!{PzZi|`mT2{U1mgl;x&gV2N8
z{CvN7#?a7Et|4)|hu4rTV99|Oxh%(g+ShSw^vS>lly;_l`1rAkXRqTOMdlOe-whqi
zCp`G5ErJ<C%aPtMU+{T<V3mz4;&24`$*+M?aux#ytcA+C53Gr%ZiYMwhlGY*YWDm&
zyw>cj-Zxg5OkfoR_#9Wt;r#vgZ;P^(&I^rS**~<MkZ4qn16yA4Rf~H6dr1B;nb1<A
zz2MZ*RqH_y4+`pPW#?#N+~6m~6fN#=k_;oL2)<fCg2dbN3HBb@flVM~M+P|A-8G|A
z8=w`iT+Ff{t%8RF(r<ZZ+pRhPO*+)_R;&ejBpdij`@-CxZ&yN-TIF&jIJgu`pIU$i
z2`jea(E$gC#rPmW3i^^t;OEYro0XL{O_}<H>!=b*OHK6!2{ig{Ai9k-`!|`JoFi{b
zj}4BFjs|F!Wn4eHii%<w({aephYvl-Jk;M%mOGNH80aSF&6_v6>n<lcA_0Y%VfH)}
zqvOSjc$iCsw1_-LZ>Ew!cHM%*_m3YxK5aGiTJdN?<-7OlXgU>+RSU(Bn7j!X8v%5<
z4w~BruA;Is^^zp+=cdw0%*ykI3%R$5GI?+4YHBFQvAJg)6Zpz>{pi*V-$M*gxQcVu
zKx<sXJo4%M9RQZDsvd@JNRWoNFW?<Qcdx@PE_W-_Ecb^Xk^;ghvv~1aMCpqbjUgM%
z(jC1$>|Ug?%H_vLlI0a|12j;<H1_78;bdyuyGr4vbR^oL_NjW|h4crT5$N-9<Bv%b
zCIBGV2^u9!%i*cqLLB?La!=lkbNr1PHh>@|gOv>teiE`9B=!g2_7ht%fQZUe@`*vD
zx)`<I_UJCjlw5P*zrSC}HB6M8HA`;#GFF*fe?3AK(Wb)T?OSFsL}|3!CfpL9fT~iC
zriraF=IPfy+&723pt4aiT5ll?M{O6i0t)sb<|K>h-&#m5D%zH)t6MhD?1n~KcmCV^
zyV*U=^vbdRQfcn*;e!zPUM^<d^xV{#Vg#%Fe+e(veFn^569D|fccy8OnXSGF63bc=
zn)#6!y`QeAXmGF=R`|T6G%C9!b|V2$_k?`cD6I{kK)<B4f`Y!+Gt^1()a&U;Vn3tu
zDrYBZdBh5c-%w_w4~1U)7Gb2KaviOf{sluX1m(9KQ}_OQcheJ)0E=`k)xDF{@N)m!
zaWtU$YseDgoJ?$hSIu&E-(@tGx%EU2Er$reI_(>LI^H&X7yrtYD+?I`?(LO)K4nv_
zJdhvC%EruCq)pkZC0b-gHbhn$%vm`I-16o-0Ctz&|AwV<`Om-7HwWv?0M?}O{rthF
zNhRAS_ZX1K<{1t7YT|<j-Dnp>tx1^*Ae3%fWCPnNC*yFC(A!yb)(3DbooBR4#pg&n
z{f$q)x;8)#3govd_7q;Zd-ty73*?5kXFaa3$ALrq3`=Es`Pi-3H=vJ>|0K%g%#jIf
zLY9<sNq|FbQGhaPK;t`%4s^qZ5rATx3Azr5ZtC>u#hruE01t!S&YCkP;=OWi6jNf|
zmYfgFoH4=2R?L%>yxZ|40c@azK1Bu=1v5|?@Ly}I7ph3ej(60=UXOr;H0daUYf<(#
z>zS<=Rac`P37^?^t-uKyLm2<kez(_<_TBPETeHab%JVdR<fbONRILcs(KRv(+i#KO
zlbSVN>BpZoc4QxFTPiPH__Ve*6Q|F$ULjj-Uk$LYw{PF#BhFHGa%R9k|CBvx2AF<I
zA1<bG&h~iATau&$M#(3;h~W3isK&ebd(O=*a~ZRY(sqO18JIhWm3wfoJoJj?Ypt0%
zzATWDUO<11%cb|U0;vTFUMTMGH$_U_q+K1pzHGTovGvV@=6Cf<hUzf|7RFVQQ)jQ(
z6jvVbY+<sdm?`MMf&vjqmG2I{k&4$g+hCQ37O$NrCMF&I&ba+P2+1YUfy1O0PYw*j
z<^7^N%D$y$EB`&1eq^S0KT2tCS)H=))#!s+Z0<Am>a?u<QhdsCp2ht9%nK$Jpyxq%
z+|BA2vD(QqN(E9|>4fHFTUENH%`ThoMm||mer(<yp4N{+<K2f)_}2dJ-2!W~SReUV
zS6|(U?10)^RkGechJ6n(j{4dtD=WK$14Ov~XPt@^P-h>pKHw3$tM#J)o&9$sx=?SK
zOIoRwC_?7*Dvm*AgHwI9>yZoF@)4Qd*^|{rGQ>~U;cU=4EB0e(?%p0BZKHl8$*yZ9
z!Ig(*t6c)jsarSGXxYch*y?&IE#6uA^ByT#XH;T2J+w^Q4XQKsn09n{I2I0fISyCb
z;(q8*cXzjODdkM2=e~F$zGT}?3?x?4AIl1c_t<~^C)@q@;DW5R0X~`~Aed$iFDM~h
zx8^qDJxUZw`RKPw!l^ck0GUzV&W5S2c=l}3!i7|Gc2-@^;UmagkSGGI!(BzL@;P0N
z=f8H;E7cyvuKq}K6TS9cuFB{&%yGjLf|I-zfwr-O=VmBN50Alr=LB}4x@~I}^^USM
zSfhI}G0hOf;w$7Xf%?50<$i4A;vWu6yiquz?sJ~a4&JZkJK!F8BTx387VJ>|jdLeX
zp1hS1+v@F7lKNS8%avV)CGc}e$dQo^XB}*uM9gxT2u-Zb8S<`%b4F5oKNAmr_chcv
zvPhg!DGI#J_gP+yeURVVkhZGJ9}?UODn2XQit+*CPQ5m<6&sEasS-9vjy$Sl1WK(j
z^KBy$;ddTkUI7yKB=`-HzgV~^0*$>sWWrqS$^M3Mw5yNrUNeyW!_&hMvOcn+_KJ4C
zg?c`==kTpQ&tuw`5+uElY_wna1aVYJ=F?di$tDukLWpkKe@qE)|J9z4@}k`+_k;!-
zYyx@vyeo%0>J<I=gG^Qr^DQnc+yN}yy#p^A>N#;Z-=?E%b{nKoB|EXti<;%oi#uA?
zjkDN1(nfaT?|R$)MK)i4rG^lKdpu`9vt1Ib>=GM_DyFj*u(h?dQ6tK2O51~%3A!~i
zy$P6G{`MNt376`)89>EW*ZVB0ohvH3c;|17e#6_3m}#q(l}*xIGPho(*FaUcywNAH
z>zkI*0SR!bJAW)hZEMz!JJW)why|u1Ej^uWN+n>)NsQcV*uD9I!`LIl09{&Mn|erg
z{`_^r{zF79^-@uHU6El9N+Ew34}82B(iMKap<XR2yX+M6<~$?a`pIGtC2r`i{4lQ;
zArR|dwzv{2*%IQ>egNDgm|+u>w?T5q-=1g#{QvRubcbGUgI7VfCJqYtgh&FyfPv?b
zOH7zLRSFQv!Mm7Ly^f>d&Z10|_Y5}0m!b0$hS2pt9^!BL3KR!^+@Me0yv^LC4Dl+I
zj=bxD148opa!R+&UkrQ#aSEz6v;(Er$$dQuEI&U#A7E6{YdO+&h&`S*XO6;i)8U+n
zc*8M_O`9g1`}(=|`<oy4eh(m4IQazQR*JTFZ!5`bX4vvixbVSYzjPHKCh>cimk>oB
z>D`%eWCmGwPf3i;WHJ$qc8$AF`M@_xEl{lAOkiPD^fs9%?Xo0Nz5@Nnl%I_!aIc~5
zg23@oLw0-dDFg&$ZCJj1IZ}7Tvt5eR=P6hzuU-2V;2yRPxo8GW(UA&`D<@(NQ@}4v
zfgON)%=X4c4*_gQf%~eo^A|2C0WJU#60#a$FJd`Yeq!RX^V;u;Y+nFZk9)yEE$v_F
z*{A@bGMjzQOeWMXnVFerRZqfp!hVf=!twX_$HxE_Nq+T-da(#|l#*sPR31R>=*ZiH
zBTz>-R^EF!G+AoX&!x+IHAi|Pi3|$~A!;acx7LS_Xp#ve7F1@?5XCBT<LLkd%-l}R
z5}^*B5nR^7VX$+Sy28IvPyCKkBp}a(@a`asC65~s4k5OZlWY_6ekXOs97}h!_-s&h
zLm)8nPO)Zeuqf8;?O>TGIacy&wuGh$=mIL+6EjbO1{S2Q;Rpj<eVrk`bcb9Z4Vp)i
zoKbJ`parZ_&i<uxa+~Xaqn=b_X(A~L@ckXc4hUnZd-YoMpqN+zu*!%Og^)YpFf^)r
z-e&u03165JYtYs_LOwR;8MPGl?dP95bqZAE_r5;-1r_>-(+A*1!CAs@P>IE-Dh>zj
z3D&Gx6RmcLAS{HwD7VEyW=i0DAcLs<vWt2(??g~%ApkznFpTVQ>I)KE#BF@uH8mYZ
zZ#4Xl#~nh>jY4@)OK>y7hu*MK*3_(f`SN8&#p;R$)FM7?$48=$r(T$~oOm7bRH<4H
zWDjMI$}n0KWuP9wU;s2{LO~Cof{2fT0if);c5MfeC=C6qmCL9Fag3gZx(GK>;zDpI
zvQj;EuwO-M*|fg`4Nf&F_Mo!^kja?w>=?e{dVsK~(fytPS`u9C@`Jmm57=qpj?ARN
z{2-sh7@2E8^0i4$!IhU3%@cpayVCI&V&&+`ibVS7CBCR}2B95&?b<&W9U6#w93H77
z$;3GpL~}uX;7bm|Mv?SHSo1kr0{>M<oy7-S0H%qc@gvs(?kno$L{lTbCWY&WCbmH0
zM*J~$R4b5%u91KLp3;Jq24d)Y&&QJ^GZSdXN3kp5stGt6Ofa?_VBkl*yq@Rg;`h{I
z7^1Ly5MUj!$JRO&*Hs<Ez)`rj`Hp}5@rQ84hZU*)S5uwtN!*sX3D{eb4FOhr9!(<&
z9(n71+bk`sp|0qH@-brBQfglIE6PVMt<)^`78-_wC8sSf(=Dh8F%1Uws@ma;5C;QB
z&25xoH8v;Wzlg)XQ!p4;!^$1fhlb6E1S|w)&7x_c$NpOekwd9A6+i<Fg0@LNRK6E=
z{K7@&Ye@+Dezi_2za)|d;CUfZQQrvaOWv0O+x_|TXB2U#<{MKPbag5tVMi*boHaB!
zDE#)ixuMzF+1fYMj&@N1fk`~Da4<l?@1Y_eT}6lz?_flTYLw71_?~1xHk1}yB|Xq0
zS-1c`#h}aSx!O|TIk-PDSzIpc1mGflin5_r<eEEZ(@p64=1tT^(`@-LkpcF_2oN!s
z)XZt%+vf7@v7D(hZvZ7JtEr`8K0xRaZ$&UhZ!Tg`;Lz!A?<wIX!JCv(Y3VlB3v4F_
zfb5dweGqLzNn+#E$ru*MqC7+iqh3>-A1N4vaaEj>)C4dGl&nKh6}WiA2SY}HDDQ0W
z8o^S60t4l1AUj~<tsZQlAfDEBwB+btE<l9^{jfIbPAE7D*n9xN|GiaH$-y0PV~;I~
z$;9#rNWD!(5ZJE}WNnc=57L=%1Z2+QqpP&8Q|L3QVv%pa0}#$Z_%X*|(^qyViBsc-
z@^!1PZU<$le{}pk$q{^ddb&3hW`(K-_)8uastom1V<_c?%P)M~*s(zPV1+0H3X+b2
z_9<>K1>Jth1S2Iwrg9&U2;i)Tj{K(?!PpKI%6lvBHw@7n;d_<>0nE9#7ZeY*h3WFK
zC|rUTOfaP8W$KyVKc~Z2=}}N1iDsCaH=k4090JcSyru-ulwP|#Hll#m;c-t|60!}*
zIv_nNSX}?IpkScr5C0`vXgjjH+ykx?5o&d70$L>gc*9DztFZsNM9cRQcs6qjyU;tX
zSvhR|jGSY}_6@Vld{;_Kt$_=v-%vUaGDM6G98BBG9lN1wjft_cW1mt#cf_X5tzKMg
zO*q_$=f{#<k(tfD%sTN#U*>pU8%j(E#sNE#`YSR+iWM10yK_G8!R8Qrd2A_K&Kuv^
zCf#NeCD?kF2k6!CxDZELI5tohoqgDXnat(2B3Bj8g@~^Bztbm8qE?F?zg;&h#t(&`
z!m$zQ1&0Qb8EL=ZG2!;0vY|KthCyI<;={vNuipwb3Rc?Uvk^6+@G$7vgXNAwb>TMY
z{tEQfNjV%z1LY7AIs5qd=v}5GGb77{nsR&e<2VAHBz-^@Z;;E8Pi%o+FvqP8)=KzF
zJ3dwNHn#6TtsgOVxF{8Bie%8o!cCiskj9w!vY9ife>YdIkTHs)2U?>*Qy3EUE6K|j
zK*<6C|MKug>eD>T!A)t$l_UQm$0d_N?nYVedu0iOBS%#699$(~ffBv|ai#qqryC32
zCpN{=($W$utj_n_PK7~5{EEkA(dB*feUQ}RC&ZG5N5eTTJS>NuOKmEswM?-{kp&R+
zg?bW^rlh2J3fMUOtmzKY0Jp{7DD)5xPCQ(_dAEo>^|mS&%BxnrK`Z*=;^`yzFmMKU
zCvm<BUr$|vaFpySFo5)-$uiVkxR#2R7%XIER{IFZixCzCN1g3>8KF}<x(~pKS=PmQ
z<fBb?6GPPX1XgimD)||2fOvrJ0q{uiVqm)WW6T@4Uj<|g<ve^!b{qdWW!A1Df)gUp
zE9L;;haW)pL$(yLobMl0=bKY+0Rv4)a8C9vK{SS|L3O)A^b$&;T=ZBywqC_9O92*2
z7M(;X(w*Eudw?V(AA6bZwQZW<D^Z&kY5WJ}kS7N+kkif&JWza1P56Chf{!`wI;}4N
z`EdbG07KCyLI%to6>ps(N$4%3$CQIXz%J&C&NaJVUN)ejOzPdopwKJ@+Pzg22F~LK
zD5mYNRd>4vvcz&^!pbX<zvjMLIKt%yFp8{LtX6%c%raPS;WgV~C)ciB3)PNbt(LWu
zqO06oX#q+=*2aS{@MMrtgD<iwA{3T6G-Mjc-9h$h9V3)sNG1Z%ix)2<%tbGz4D9&b
z1tCX2C07t$4(AM@wGdl&40{4QwvGehNF?1KINQiBJni-UQwm9s@g&`)3JRdz4KVf<
zvkIXFM#qGJ3m5kK+d_%b6X{uZ$1*u~>d1_Eu`L^}h+`UGp!E_DyIA7J%!N}YO!zBZ
zpC&~E`w58_kr%=h;@0@O<Q*lR$2Euh?8%eX-2#pro>7671mQXj$9G?4==?M5U3<Gv
ze;0Wd@&m_X$BrQ*6EH)x4|5^1Z;eZkKW%{uQy{bnRw<v+g91d5odH8frvoYlWDe+S
zbQSc#!CkzHJ+tkY*eF1BCBsRhAc#Q8ogLRFwjdVIjRKp8b;v)l13qDjzN@QiXEtc&
z70WUzjUm9VM;G(CZKye>Ytv&T*hJF5HjW$(ATT>-T*MDkC6ozR^7!(lOU!#a)ciwG
z*&PrNu;O42Vu=FUr;=Nf^+niFxM^x?Hv9|pe>MgT?-2hQWO?C{cw79p)IEggFK?oO
znN`$tDF_3VIe=n88kk=(HyKCBI55&syeIV&m<Kw{bBv<&Dn#yO?<@gYe8`8BZ1UK_
zGrA<@-ST<?O@?NAoB=VMOjI#~If21p#x=fwZ;W9>7ie;eUBNO2LCdPpMQWzD_=!+$
z8|<vfGug3^If&7n%D9&_3aOQr^q)D2)9sAX_7q7_KP*XK$3YHPa*7XrV##?@c=KQx
zk5I)pm!ctGQc{wN%ONSCLs*1Z4IM`aa5`?8o2ZQKeTW#3O08B<h=X{sqy^m?=v(Nm
zEi@>T4|#fKd-amU5F6Sum4wx)shC!*Q_I&PFYFMP{}J4f4K>zLiPf;wC%Ch|0Z}8r
ze@!zmH2nJY>*umN!I2{`4s|F%2RlD?h(4(ohhXQ@=KgBu1dStK9_R4@ok4nbQ5#Ii
z81@XddKv>Tt#szI5vQ{1BR(v_IA$7o_3n6tCTzV}mx_Cl%i9?3(!dp@&u@dXHIpWT
zog4fEY6=mh@puqfx4Mv1+9iNpSx-F~ctw~Rde*B!GQjKP$Gk>l5em8eTvc`T@Nwrz
zLe}8a-qf_pA@8s60(;`P{y__Np~l*^IJpcaQgJIbe0!WY35g5h3L%d}l%3>Uu^*ro
zHZ1zUnZq+#gA9ZgYsvWTuCCa9iNM+JUnIGrJF(9FJ~#no<OOZ*wp&55AgDu_Y<OXw
z)*JKtXg8NgU7iG-;P639vevLy9=0Hi&bfC8A+EsEV)+CT)z-qOIB;Gg*0%;SAbAt+
zYrKo#85p(~oH2h+fd*Evf3faa-DjrH``pkVGGhiDvE!4E*_R9Zf;Y{eCHs@W7MTTB
zsl6*9y79?ZXy5=U9F+k#3p@{PEE^Wi%;oE9m&%dqgTvdfkXaS?2=fSRWdaT4hA*08
zLBvNn7V|R00pv-!%e%1QaZ_YJz%LM?n#B>yLyBVMNKgz#jG;^Tp8`sE2F)~ukv-6=
za9;aRd*$+xL5ov|qWt!j*)#8rB{<s*n#X)LxYlee>91;pOoAnGVSbEjM>j$NKmnN}
zlq7#e43HXOYB;%>p@73beF8+c7q|6Kghw6#AqWfry|YI2-=CHNeM4#weUKpTFW<`_
zw0TR;{K0P&s{v#|cL-n{pD67t6Xk@8P@2(&9cMuaW_W+wzhT2uSbVSy|6I$eOvQk#
z!4Z@$I~7+4I9-^^wk+kKDnfitKP(h=Lms#GE6nOD=Ql)wbAJy!#6Z_eB|xaGBg+;%
z0}JTWpN)x|3vgieY^PlzP>D@Ac5Xo<4<R!oO(+}R|D3tS+XZwWjVKqx!nTiA^HlFc
z(W|&l5vm}NLx$Wfg&sqxzbKIb1yM+ev1KPny@9&@Q>RP`84oR0$Y2L}=%yprx6RaE
z{t0Mdo0Z95ChJTNltl^LP5=X-Js9*40uan14c=om=ZvBR;J|DKXx-?}g`F2j!AlDJ
z{fqXQ(GZ5acVGqIyilDh<47vQ&SS%1Gg)ZrP)`6720b&xDj@ns7IQO^&?8$UX(v6r
zaybf{n)Mk2^zhJ7Y2QH&Y~x5*SIDtE(ekI^9$Gg|(v|Ok*$<LYXurPl2|^rQPvA&P
zqHkRJT8TCh^lF5sK)G*@$b+DnK$niSZtWQ;`Nfpxi}=U9K@ARR4)M`&c*~AxMuK&w
z{M0Y#RKGX-8!C88G7;z-*DcL_riKipy}ny?`Yvd18|pv~H6Ic!ojZpHaRK@a3IZ8c
zvtSS9aySqJ;{YFYRU(9XkaRWA{_LPN1&GZpNMs;5SUPq7;k{KIV*4dQvOe;}Ht#wQ
zFBqiFtmC=>=v6NTkWW3ppK+`07?>YQ5Cgx}z&|$F`9fMRWV4<giQ5Y6FwSF((MBl8
zExON!m%Gx1fAQj9ZXIYd2tDCJ$z9O>+x_O}=huH3Wlv7wPv@!yB1&I06)5DAVBvH?
ztZ@dVAt^ruk)s{+4q#^gMJ-T>6VPE#2kiP&j}>)Otdt3TXyJ9>Kw{QJoMyTq{KS#x
zdMT3X?wix3SFKw0f|&!#@IR1=Do2Gwl>6>v<xQgzx*Fe*%YmL4JK<|*Cp@Z2sQ=K;
zHlH58&O9VOhtNUg_uqd<*}{duz&GFS{kwejLr3uA?j1l*dJWbON2YI5;v){dWd-yp
z^1U4J>5xCoLEGmGCcalue%fsCoj9Up#2NwpCcXA9Rg8g~D|_cP*vET**WW$`Hl!9A
zVmnPNFz7Q8c4RF<;-w=YgJw->6m_Ubs6#oNW>NCnWEy=L`Jjk0ki}nzA7F?5usuF}
z%x&r&Et3k6z5qmaeRvEKZRo#~(<SY)(2Zni)Ew|~cDwyeZ-cjcF%7MGlz0OoFREK3
z?tuT-^IoTO^f4%T8jdD?8P=?Rb4-y*R6${(=o%Y{HPr(ahgz7#qEH2ju9XRlT-&?u
z35guMWIc|XZf>PW#2GsZ-9$TkF8_JaPu%sX+-;3j{qY}pKj@1WFE0H<_d>>ePY|=6
zA4H3no;{Lep;WN=g@4$rwwD|N@j>!7?y#?gW3Zf5=*4|A$7ht=^pxl;HFEV<uTFvQ
z4S&o>W?a9b;eOY1s~*AzW|Gz}b>8S`8ML48*}mx6!B^MRWG>iM+rMAwr2t7O#s*Az
zWw(cxP(NDS(yIamG|qE?TvPK#^GYvM8S4uOcrMMhKvslR5T$lVyItkx$rmtjICvK)
z!t6qZxFM$NBJQS%(sfiL;o;3i5wg>&LL}o&<K7?+H5yvUo3o;b)0K(MR@{4tl+*lw
z@Qjp&45u0B>50710Ywl<3GZNnW9e(x?r1k3GiD4tlqu2J#-sn-%@;_qCY;`rop$R!
zPKQOoxc)Qf`qnq=hk|7Q5_wq-Wd~=Keq24?woBq6_Y0asBD=@B)Xa}bik#uEtDA!0
z^~IOzmvDH}zAiE$bd~|-Lf0<XxN##MMi$`=MR^u#^X|pmgt;r55wT)-zz1!VZ2Imk
zD_f$V(1si!=sAne$;Cbh{Z^OdYjwpe?HejemWMIWqm?jUpQlcoxO_N67yo@}zyE~`
ztGmV;$ozw;cm`o9VK6`!ga!ps=~OOf8jBBq$d+n@mBzL(HO8Wb)d_y93ijc>j%e%>
zo2(4w28^@E=SXl2gC3>l&Yd%73>^EbCuK%++VpADXyj-l!+O!1-T6jWmaWf{hW8F{
z!d(wZGBn+WU=$7J<|e&3!-(I*$DD}Eo8&<_d7RQ9z}|YK9c;wWT`y7WL`K{g+T^g;
za4I1cGnXTx-z)+B4^Q1wT2}Vuyi#a7uQsOWbKp|vsMZ8ky6w$=wddYmtDx-XOKj3h
zPYFC9i?%ay*YP9r#{f?Yyh!UKptp#m@a_1E*gME{3bI!gv^{cubNAA7U+^mKpS`}e
zUws_!p3o`lSy0aDDp!LAxi(aXny<~@&YV4~;2L)C)yIwwrO4m5UOloOX3NwoecRv~
zErEG~u&eD2fx_@fQXH_g0a9%s2g#TkpfD#mrxGbeaJ{i2WKdtd(ZhC8Z92k(Ll!;#
z-OT{sLPEIRljmQ*3z%F#I!McRz~pGyf6;@zhLJBfNZ~M~v^M1P{gQ%#+{LulEEk|e
z%(HR}7Cb=(ELJM{G}5V4*RJR5*H>+i;9G+el0$}G7$Da#$sEY?a6&l&z|WFpPdyL_
zo;du+agb3WUpJw3Md7gFo~>Iiy5ry}Zl6L>!PuZ%bFd3?4Yfu*rx;xwrG=eq*B%=j
zL0}N4GvHchq_+zylgs%;edfn2y>ci~A-Gt^5*;ui({r-4quLd36d!T@hrs~zUE-<4
zSOo7%0*pnz{bxT_#n6Y`Hy5COpua=X=qQUaS9F#W8+|VNk}IqaJsNR7e90Qfsw-2#
zLJ?HW1DZ*BKOUrW^@VwP%MRv1rI=SOI#7}d)b3W-QI7koz7RiDmWhbK<=<XwI>LM^
zMP9Y3pFL*7#*MJf@siHqxk|jFnk1;Pg}@EwC7}5oltcnQ0puT8Lnkttx2dE8BR2oV
z3mvmtpe&c$IHHXNDy0(FUVhc#Mxbw;=KS#<8eN!wzJvp^`hZ-LGSrIOgR?IkA3GuU
zv4A_(wm$u4WUY$Oh|EH@I?eqf(SDODMClGnm2qNKK;{(uRqJN2_nQru9Ry`C&=+#?
z;%me@=qP=(4_$+{PD~`djj*!2G7;%IOpN%8rV_@$sz$lUA)RYfiY-SgufS5wnM*`H
z)EWVtRive0e7g|txEB5bAT->1JfzPU3seGL-gdnNN|Ys8)6@_Ax2`dhG>;q}<MouY
zCxuGZyAPmSAMQl#r3^a&xyJmjFUg(nAX9YgGVkwd1m$`C+Gc=kjoum>8q+0JAD_tO
ztV(J@Vw!BZZ=|jRzbNDkJaolR)!orYl527&X!#^;kvxq4eVZ6k)zCG52t!EUMrUng
z-@bLr6y+&!P}9_@AfIsLpj-VjKM5jmQwz}4u#K*K&S^GVKX6iSSC<4dU4dGQ{RVoR
z5Vv7TpLV)cw8EA-$I@i&^xT!8@otTLiY7b2`e>sVFB>V2NjwmY4yX<=JW<Q-wb!kC
z41<nc-|P3|o1eP)f`xQe3T>$UK{1MxuU)x9sZX&ul3D=sj)oaXalY>1hlGR>5iFo&
zH(}uXBRkW3I-6rXOI7plg`4VLD_^jf`?7GUtEH2;jqin{MFn(&B5REqUz?VXV=iON
z5ZyY%c%$(XQKchKCNEiXWZc>b@<-lr5_fo)mvpms*Qp;lURVFVL#_M3-wgA^VYO8q
z-!9CJh_tAjy{BpAn)H?BVHYp{D_1a*r99<a-owXvoIHQP#43VC$3hWUGiV{W5)yjV
z;LugmtWI@X=HwJ|o)V##qlFkR{2PiTRX+=OUK4#HaZp$H!jiCa9d2MNY4*I{*wFDd
zH}@8TZ;T@ThXx7>v(C;w%}{*9Iq@&nniMJ{OtVMIV10KX?z5YxAs{kUz;TqI<<R%X
z&g!XYX>dI$W~<F*>8w?3(!G5!bc7p$p-tY$WFj?+^8T!bKdD?amqfsIuRQtl51d^!
z*!TWIp*rjcO=^Y1W+YORCXaNXFzKct#}W1ZAYkfQpPCy16~-1{1Kd~k?KLW1i(_<o
zT!w{7OesXu=f)^F`;8#`mccmt^xc2|EyEdFcw~GUw2z_UdthnN3#=dN+uaThP_Gzd
zIfq}q47mIu3NP}EF2<s!c=;F}w~?xnXeh>UU-5!P$C^wNL!Cq%UIl+)=?Y8if>y$(
zZXtho-9%dU7WLURLd-1pW6wwA%~#Kxmu;Y{JHGxGoyU_NJ)%leKt$$nfZy2JWPl?h
z@0A4NlN*cevwLa`z8ei~Ral}AX0R<!FaW-mqWmATGBsml0^)6<8bOgD92JQ>B;c9u
zReH|RE+8hxP%i5DoY(eD{lpejKIhE)fEgJHR}NDVbiMrr4X{^SmBUEndOsQPwz5?A
zSzAY`??dK^YK=GP46wAN_jaz?H~PS%Hhx3NYeh(*O)G%W;G7SOF09>F;{<Mpw8E7|
zmA)eRX#xprfT*Ci*b#geo|~+~E-;ciFSUjS^AqQcEJh?4rg7!2E^>5TQ1JM1<Cd@F
zISyG8AzOpda!pOD<OM?n88~VyF)4yy!Wg!OZkT!{1C{ItyO7W<qkcvhR7jj}?y;so
zH&OPRSOlZ7^|#5p27nGu$hmp)<TFGd&MNVJaZqXo2gk(3ti+EsEJ=w5u&>tgsF4wR
zLk|xQeh)1)T0a4GVD0oDIx#SG19<ftRyCGe-o(?tKl(00i??C_Iet7_D;F%y2oSK(
z(xeAXGtom`dyKgGbIvML9Ha;^b@8pf7l>b2s_>=P2Xb$bR`hEAjJ6665s8ypot`v-
z=Sq5r#v41@L(u$;%#R0p<Sp!t=OcP$YcLFMlAin?u?Rm_&Y|8NbA)#nAUwpjTRn8#
zzyI^b=^8y~yNqm)wG%tN`Rd*%rfn4#5E_W{I`15G;MM=Yrk9kGD(|%M?KSNbf9jIM
z>2)N1xLNIEH3arW;?2bc8-azJnwStZcIdH_y75RYG0+YG-NEDgQngK9$Z=7=+xx0_
z1$vZ38oof?c&{@bi}~F9CFe6omaEFYd3l?8o^|(k6Q0|w0(6PJt7}4;3^{0Fb@AP2
zoL;NLhaY}5#HuBWhD+L@yvohp4MhRjWcJwYz^QglT|iKz{5el?Jy7SL*FSmW1>Vi_
zq01d!DGG}Z%P(1CfzuW6a#E6#v5OrEXwlO`jwfgmqa|NdK6+wgjZJ-Sx@ps?N*e(G
z1YC|Yq<8l9_2D=G1cW%i&}#5KS{=|=z$itkx;;gqMNQ2}%*f;Bgc`;;zkSpR|AFa)
z=eSKoGLplx($W;vj1?97NaP1eBOyg%nwtwWH7ZVM+%*7usnDS_f|r_`Sh~ZeG^lC_
z>RI8c(n?B8aa=QLQ!m&?xtvDD@aX6S@_`^xEO&K{9lEqgN3d{2*_TsVrvFekNVw`+
zYR{bd;q&KVapUKb&pAyU2*`lLKggIjGP8F4E*j~Ec{$7A0z>6BB#_)IE1H(kR1S5G
z%}hyIEtr2hegUogqq}8V%gE0xEurXdn69*70aXi!S#sRX4)}BW+`+s>7jS?rHFFc$
z5*kNO44?h}?FJ=!z|KJ53Q%vNYg({{@JH5yhxsbI3vKDFS0AK5$85PelGm7<L^E>h
zP@`2<R79wG48hx`SCMgbCr4(A=NL*8%S5#cPBP^FX^5}HB}`(8VGg?L1;#mm(QDz3
zqN0X84M8auuDOdIWD?4b|Gj%hRbn7yOY&kPym|;d6@H38V>l(^(1pV7$~<v#N`@@B
zC;$}PYa!jl<+Xxo0L)eXgNpjdr=4eDCY{{|vjhzy{1Y8(U%q;E*xg-n^?s@jmcW|f
z=1E<91XLyKK<M-@0z)IpHnV3k0WT7k5i?AFI|mu_>FJun#S>r0!Tt<Y4L~+N9kT-c
z!P3SLA5<ZtNKY4R6ggUa>WIPZLpx)F@SI|_l>lMMy*>ndU-+yyZ#kUo!A2!iCWCRo
z_rAG>$e*<I5!*jsI0HS$X`$Pn)d=)6o6O4+-PI$J=K+6`pCX*ifpu)O8L58q#1`ls
z#vE+rGgygGRc`%Axik6XYFNVOb6nj&za^Urb6^f{>EYplPFs|IhpdCY8O?=P7X`K;
zX(OD+3|joY{{DU_V*6WeT#(nI^lD@pWN3g4uOzm>*x_ri%aWnd_pqBOKRnXG!@O)6
zJ0Pu2K;&B1pxUbn!RUz-e+c)@pH}WT*}LJ<Fz7DomHFY@O-#yAiAstK3Uo#X=`{yz
z*i#PHIl0W#7zb@i&s-eUkNSbIyNNJK0q6HRI5>n}0NTuG-@4T49y#zMB*d}9!~e9j
z%Z<zmkD&n%Uck;0J7{eme6<d#+@JUC(3~CC_$@l~#FS^=1JV!dMPkS)w`qdMU#NTU
zmf5_v8y3N<Zqj!Ir`N7s>+0@~^D(JxLZ`O138}+JwLWFuLp7`}eFgpl(QSZfdj@T$
zOKBZD=xp=T1=Ze8(|wqb`b?WP4X5n9=}m*bkPYnnXE-TCNiRojAiSs2{=(E*%huK3
zxOSsiQlq&=O{Z9GKqBgfI0SYI0U{C-1!`We_%m(!;gn>jYDbbIQO=P)#ZR*`>B*UC
z)bpu^(3K16mcd|Lv7atpy*k3>nU~4YNDn$GDX9?w&J3FGF6cst_+DiZC2GJJ$A>cP
z?d(Ju77tx!3s&>MhejRXok9~P@sfihXkjJ@3?P{6^6kMdSBx(8PzD3x;xUw7gQ7{4
zBfaX%O@eF6M>kA71Ik6Yz#g=ZI>}fteg>_4J5EclG%<`=Mag|&w771HmcXHanjto`
zXXdRx#pp$&m&J5+UK&XW%*y~i5F#1pYje31i0(B0J`4u>FMK?HL-;Tp;0c*V!x1fY
zbqVT{9cVCr^3SOl9pMLsKjk2d1oeh&9{m>+9UIA^2-7NY+6iIP$q1K;BAV=|&xVvv
za0N`x7fe^V!)V*MGZyaXC)B{zB7qUnz}oEJe-9kzh)|uni#rXmcoORS!otL9(wL+y
z)J(f^v!E*#?t<Is3IbQ){W-N6%_432T~nphlbW_mP$EXMArl!;0!arBA(|%_Ft5&c
zM9C={4L*|y6#A*MU2OCjm`v(lVSk@;?xQf6XZz2&bCfiV+=WHyaR;&ccTpFhW0$&`
zT4Pfa{t?3uWztCo@=Lg7m}TlYd>oj{2NM*2LHLxUrL<af!R5#y2YAZabc*rqQ;qB@
zf5gjVMUwjz!FGa;po$3nC$PP|y>XN-j;^9Qf{<CRqPI3RHMO_jE&YgxR6SxqJC=*A
zyPPE=91xEr(j`TV(T|u7_NVGT15Do1RcBW2DIEP8LpF^@1GnP>kv!(drHTX6?kO=7
zqi^EGpZB@m;U;MtoRL3)7^v_hM-UJw^)|1$BU}n^YB^m^FY(IHIGY?IXhi}qArK%?
zx9YZ1Mj;Yh5*YZ|n8KPQa-ML`f_~U(p-U<o#9{#=hb)b}OeqI3k2Nq{pmBoLgxHtO
zK*gCTLk%Yz;VxDW9Pz@)5_HZ>?0%dieChLce4LEFW=%S{Xc4x^rFaVz8|g{@gZeHK
zYLgLi^tB*`sTF7xIh(qI6dA-JMYpz|D>MyWNiC4ed}?oJd33KKo=0UDX|_F0LOCzd
zN%%(C)j$Zhi{h{;%3aj(S_rmro3s`az7IlS4iv7iQNIz7Qp4k-sPp8>jX%3N^1V{H
z5fKq+uWJ~CAv7=3rVrrA=6Pr>9Q!0#XdzTMjMX@aUQUk`N@6A+?&1U9XP>WwUO+e#
zNT?=HnuPq=u9EQ;YlD`o&PIZ?-CX%+WW$w)3K5;vz|YM&&^?2ETKp^0spTW|`a7e9
z0T6771nTkR^hiqyJg%C=-}_PPtNFcsrEr;Dw&#Oc1j2;g@7NOb*nRAKmdSdOUOE&z
zkD*42B*_i{Gld9zFL_q-K0YG5Rp8tc6GysWk>s`vfEzOe{U@l~n6hc&;#ysKi#DE_
za_4=w(8{W$5i<BR14+?&!jPINrXe`tU5DO)SY2ww^<fH-6@08hw_3cmWpWEhp*b+=
zp`jFcOzhGhifGVv3Ki~q_U<L2R;(@?hh3!tfb;Mp2`e?dw-Z}1bmtF#okXT>a*`vO
z{#NGn6XPJbBzX&Rzus&dt3rYfRxt;qa5ISB0Duj(m04G5q=JG25R}1SP!|<)pzb2a
zEuZf)5iSq|2R;lMgNgIU0?AiFt^_(nN9P3ygwba?kf_eg<=&64DcQE<m8IE)*OGL6
zk&`1Fo@591@PQx7X2`n8+?NM2pj?8Sce1_jZ=@K4yf*hE9RfWU{YSsgB*XJu@2|4*
zau0O%dG57g^vfS`Gk!0n@EP3Y4DoC^d-nap@+;PJIIz%ZC|snz>KR8n`_g#f3<O)c
zqsb!@cr`UO>YpETfS=rD?oScMJ!NZYsOpU1$Pqa;iTBeP7v^fnuOg}*;i$yta!#K<
z4f*Dr<#uDqtG?M!lDJ8185tR&xlN|A1I|pDT>`0RLxUZ)lB-E;8C0H}bc-HQpz*kO
z6oF@MV=(RP?Rj!enw?>Xp`Gh&w&gU7uOXG`KuONU$9<GR1ES1>je78D2sm_gC(=Sc
zx~n6L12`cZnm4<#;J1t)KOP6s(Dl&xfbN2WP!ROo+jEW7mB&7W;=8~Y9#2tSz53ZR
z`={AJ5UL>)hJC)(F@E+k|IJ*LWy`ikZrTFYJSZwEkb&G!VOkva_UQ1Wp}GPZE$bui
z#?kevmx7}oi1%V&4zms=Bzjj(JjSVY0$#E;8*cHmTAa_>hYy3#Atv5#eZ5gCA|%A&
zJxvO>{@HY`YokaRzzqsCXD|_`CCWM7FDcn3#woYz!S8_ii@q<6(pa<22&*Ud%Ll-g
zVZ|BfXK^%kr^^W<QXK#1*s)-ZVgiHnJU5l67{OfWf0iA7*Vd+RHs<Ed%KHp>fHml*
za6?N`xD-~J_rRz;<=8c!W9WMOOBp*lI`*$#6~}XQbK^-i1AbNU8OF2obkV9sVJ&1U
zOZ#8LtG24@0$%WW_NJyL{i&=zG~o-U$>wfEzwHgF51sfZK4hQ2V*Yw`t?)S<e8ji5
zis|g2%OZZ3pc7DRv(2+rt&>GZu>$=?esOWaJEKb-2MnXurgk;2%~e{4wql*6u^}F2
zW_Ecn7t2H8$2923yZ9(duDs14`6&t{j{68_-;BjK(i$qOs<MbFwPluUA&0coPp(Bf
z+scg^uQ?41)B-CFDn967`ELx&1}7@716}`F$@T3Er$%ie%|MR+8@?=l(D8Uyb6Exh
zkZv7N5975|D<xtwy3I-7v;fXs;;Vx8i1?MC*b5&Adnfio;o*_#N!+y&sLgAa7(UnD
z-ridL5pacN;KeIf9KS5)fKXrV0P27D?%k$_Vra>&jhbBEdiEX%hsv57RCbOi%y`^h
zOJyxBh}at8_7rH+MeY!Xy;B!+ou1_A<aB3H)7!z}1EP8~^p=p4%7jvD1S2hI^F#4M
zezT}#Y?m5}ZUg5><@~3cqa>5PLJO(|U!}unUUg8Ox2h_fs5o-2_aIsq`bx_L1_P>}
zBMtuG_MDQ(Tbgh;1|~x61zCsLqmWYMHKbJZ-OKe@N5L1O)_C44XwE<jK}lO6_<tG!
zrBqaNfE60BjvSEk$=?uhzs0*FvjXipW@@MjSKI>&Wt1H=Lntuyz5yFL>g|u40TrIo
z_x159JC3lnQML-ijpWjYYNncdb#?tE*RKtI5<KC3{sJ%+{BNyxc6PCYZ$cj*dXg<W
z#fLw3$n8h#ka!U(dc>gw!>HbgXz52Y>b<<1OIb^g8Z7&ixs;cPau`l#g>-_N50j0<
zh)lhe^rCcTM(qETsqAL%P|^$%Eh*7)-MYW8+Jq&x0I)khoMd<CR2CPVMX{^A8vC&i
zxZC?gujy~zyj4jDhhVyS1R;sdKW-Osqa!0*a{Z8X#_z%BHoB;7QZ=iO`#$`=BW_TB
zh?J?nn_3s{C-FvJ{osdll?OnZu(J;=TsW1)I3RCW?~}12Pma?|_O)>bI*Bd^XrFqw
zX=f<}DzTo&FPuAfaZKavB-XCk5)!O0i%}<aaU+xoEN5n?%jq9c&+Ej+l&biAX2w!V
zWlsKYRh@YG<F2lr9?}tR{rcZKpdi_fZN!!JU-*nZ@EGk16#A!-PTrFDA%9idRKIsa
z+04%<h)_8q(!2V?KmYjrr#66xjL-T@(bu|%xa{({u#ZvE#q6%5OiiVNW=4Mow@On<
zo(NMYSK(tFLRagJd{$t3XkH;wHx2)UVbNZ%n|NRDL6KheW`v6?js!s*mA}BQw(MK_
z9r5|0HS6Z^KvwGH%DkkC&jns^72Oh9G7=@kB6iHDoKRy|*B)qUUM?U7(HrkSco28(
zX4kmzDOWaS)Pa!lM>pyP1IJd^WZW)=5j;)GU4#pp^^BwSHe0=bNMafJPmB3`TXgHi
zPq6kgdlmFUkwx2#>m=uMEkI;#YRIvEq^YJ>^`Gu0Bp|`_Ov}He=R4PeFU^!$xpgJ&
zQ+>7{9GJntB4JDyUd6$LF-5^-Ph$re_<DYXMjy8QOHa>~Ns~&xO=)esQl_+8;|oi_
z^@VD=x2ME1e>s%bhbyd~Xj1u~#=1r+brhisZ^z2#z{2ALyNRyF!MXkh3hW<BUN2C}
zA|+46;C^fQ(N|HcmMg#L@MkfZ8GbdeyIr63Iy=&~aO%`hNtV^saT#pqJPn_|#82iX
zb*V>$rF^iI$qr0Dq|7}93JJTI_f^=RA<7R#+5G)UDZWaIisYa!AB3aV2REMZYkjds
zVM{WY=l<7_uC=M=gP;NSdTW=0&l=qcTkE9t=J&&}KeXoyQZ4vnp)Hnp(1NZA5X?c%
zmMB*h!pOaclZn@5*4$oq&gVuKCJ6~v&@*(296MPdBY4T51*Au1g0-1pXgn$KY*IJu
zdXa7OA5jV=8V|~BuxtIKr!e25o95ekCx%qU82=#hP->Q&Qd(fi#26DNPkzR6Fs2Q?
zA2w}-koNX5h5DKrtIXP<v8>$-3-(^NkqQdU-plRt3=LbWvP-1DaY=vb)=M!ld;9AR
zhDtKelPVz#m)n0(f(|NTMcfEr;KJIjJe(YGfsy})E?+NIizR>2x?i7xa+sXl+|bLH
zTO(p`-gLm`1tPvws~pH+qj%Kv@J%7X!8fvwDs=buf)>8h<XL)?aeBjP#_7$YtFxFY
zFGTe_O-#&|dA}$p`qsM5SdT!{0Y|V@Lw~zoIh~YE|NHNMSXnaRo7_=%(aA<sl(#0V
z{A;<>V-EUPHU%4LI-Qv(j8T%T>1lSklMqzd*(J(d`g$M7wz;b#%fmzo!?W^O27&~{
z-V{8E$o7_nI~GdAuA^{rLf@UtV)Aiv?(oyeN+k3LLFR$3dN`H!<;xL)GCkIoJuomJ
z9}sl;@);5!Gc1x2<Pn;sxOhZ&K)1A^K_-=**n(Pb0=OaTkdc=5Hz2!pbSTh4NNeb$
zQAbAiG!ju>wr4_dW{U<ZCc=phPa%vc!nHlE3Mj1t*wF3O+uB>WviQorl|l$5X|AHg
z+`0SDj9l^T;Gsi+htMfxG?%5(Sy94&g~Ud2+C?1IC4?f&XbA5}$01aFk(gES2?xjz
zX^aj#D)eJ8TF6eF5PXt8xVWmVmV;9{`u_fXOkwI&A<$@<O^SiHi|!`*CP6+Eh+)cY
zazwKi;Q;S@!Qmj+n?8HRSR^w1fn?%@5|PbN<&PxlfFp(81^PfX$+ZCp>9A2ToeunX
zPu*|A$&B}8<CMl7%n$?y6jWwIAauma0+?Bl@Ho?o3Jcv29Qe}J1#H7SDT|ICz+P@!
zGtgy8IK&V4P&6qT7*W8muSFSZRr2$7Nk}EQCV4%3psx=x3jj&7CM=bxJqbEy-*7xt
zdP1HC2Jq*eMAAG+gb_0${zdui=%bwY4Gj&D9->Pfm7jN80k%#3AB<yPDQDsD%qL9a
z9aR&___U${?_;1gOr;1-;eBX!q`sQLL}LId?^IK?+`9EODu8hI!f5ifdnC~E_V%XW
zdqYyh#~$+zV(rl^4x|FUyBa4Ektr>5;v>adqpsfc;lt={+}Jo?A^*vf1quq2N#gmy
zq9?T(4PS5Q`0gXcSCrAP3>3hP6Q@sqK>&lbGy@|e@dzmtf(YN)Jc-NYqPYOoor}6;
z2vS}OZW1n2QB(jP`~K<0T<<C5NeDhs#*W1N&WrOx7=+2b?+7xABS)I<V1b1ron$Z&
zz06ZPNL6D}^De)}tB1?)$T>@p{!mhfYj^Siti$NLn8ah1pa;cs^0H+#=pYUrIFMIZ
zC|t)Vcn0f+&RG=pIfJHhhhcNHF%&x>^;pUg;YwC4;()0(7TrzE2YH!S9ALsghLH&g
zui-iNyQ}XcFvX9-Lu8N*_LA({t#Z!HnL=C=^aPG^CAc;9KJGLrv}%OtWR#M}NAV{C
zuwOE2a{TGl5ST-YS^m9=YS{_Lz#V7V14+hH4n$cw<ikWsa}d!vS3zP5XFQi<`UY|z
zl>1O?ViHx7Y_2RVJtF9aX^Ov(4m~*O35W-EjXUWE@(m<s(y4t?LUZsksRg{^jZrk_
zIjMlj!hVg2pqe95(-510+!V_X<tpUM&pY#>K{~gw7X0MsOeBSq1qjj3$Q4RieNXxP
z<+8FPo%P8J$)IKrn2v)09ARx@XMK4(8FcQM(=IKeEAa0Bu0V5|hDU2iWMnPoObDH0
zoNYi0V`R}>@iejAJUTEXnP~@q0f|M~nu#fOUTp}gsikaAnjn}%s&oEe=16%^E(c~A
zb{udfiYh6Lgh{VPUjZn_>kmFWD4cu6Dh^sfJS8wDWejUMh@@slgaE!2nwDIS%3b&l
z<ROoVefmtp@hN8#QK(0qB0-I!<+I`M**{RT`yBWKb)U|o@B!t(uh`W&!W$(YU=tZ0
zX%v2vtoof%n_@~)qv!}TE_@%JV-XYtSm{D@xYX!AX*aiPj+mGbiz19ZhOD2Gx)2h~
zh@9xSnpL*>r#;N&B*zA>hAtSL&U!$jkWTa^lun5<&Ol?JR@y2`6^3CSKO;x-RsORj
znyPvh%zXSld<+Wp=C0f`g}A;myPZ;og}aZ%pm6#h#eTypsekOnu}-A8_Tc^@;arxb
z(IsZh;>s0Jv@CY$zkoHSpgF-GtM?1PH=Z)APf0UJz?d{^P)tWbF<G=Pof@$UjY?eB
z;n33~bzLIf!5{`ZGnxjPhmyXfo%%gzVQF0(I-<pz=e~00fl+nrVEEgc^D@RhNd8tr
zA7mhp`HUj7<-0c664mCMp>Iz(n+E7!`D6X}X5o}?+fwKiCbJR6v<5Zel$NKf!jvme
z$)>NThX>?575T15lC&_<xJqfouaML@;{l(}tToN)T5*Ut-@1%cRhjrtr*7T41^vD9
zK*yq0I^2_a@~`-S=c&zD7_9jV!Q>UEs-t9y22dP1(Dx0J8t|G&$Bi2YiOLt*Xdj`g
z$Lyl7?Js-Mxegd|j%_<ySPL0GT*F(sGc5!C-Kf>J_UQYE!v3vP<l^Ci>P=8KYe``(
z=1!#@B}J`hV3oW7&FFT#L?r%V3<!XxQrAGA`x3QAFJF@0IU{Sp8zv3<iPsdlf;DS^
zBN@cUCqgqUoIxI!*@_T+1VD_}9f?b9ZK+0c)b@Z*a&S<TL3KFZWZrXp2!}I$uFuYv
zb}b!!nJ@;#2@3QyhkOxYX5u=aGq5`}W7aHT1cfS19&yX0wJY(t5n6s(iH@VCA$Vg}
z4q7v3VvSJ&fG<KeL%;%xJ}03a5s*u*bP4p2mn(qMnL<0a15!elOgu_(0eRc^Bu$|J
zPx2nU$)slgx;g4Egm#tA>VfLwHR%UVE@%XpMfVb(JU}igYdJrA<?W^lNIy2Ma^yp!
z=JEM$RaXk>-$M;73c!-n!Y1K7B^7p4W1||{Q-7@@ZC?b#&Rqhjd{RnE^orfM28m`3
zI@77{AOPXdci;e@?1Fcu;;K}LpK<2n7pVMy-o-zBneB<n3RJgH<8hsIvPGt9##<r`
zl3BN?6H2BQKS9_Uao-V=lM%{{iWta{rge37fr4dq)sCg>kt_z}Q10<oxSO9N7Hsy+
z8QnE;2SQ1J68R-2rL#&tB;TcKrD4mfOM!(oMYvkNJWTcH4S+}NwKS=F4^WhfCoBj(
zN$>0L7uw8R&h2}EM`4bkxFz=21P2w^+45nKe_pbu1smO|=Aj}PEttzQs1z&-SY;<1
zi}-2|Qt<%JVj@IYNS7ZdjUMj7TKRd0rX|R+%#}%TE!GzLGxFA(C+-@Jtm&+W;Jm-#
z+ljk{QIjKGFn{jY@OYRwR9y-D@tl+v_;@7ot~;%k3B4$wXeu5=aEaVU(P|{rf>B%k
zb5dVH<N~VV)ir;HKe#obyfsezH7F`8M1jGb@S3|AZmYx*wuvGRg*REbedtpR`Zg(=
z6CD`jcNyg)#n~7Q9L)^8W1Hx1(w8G7Z<S(rSQx5md$fLZYY7L7$L)s)x-VRQz{B=~
z3X~+}z(Otz{yvN>Ka@XWJZqA;NLIo0LLc;tfzk)l<mv8S_R$?r`cn}BYzod$&6P+Q
z5h5b%w<RJ@1~bPG?+eNenIPs4CAy=p;^=#;^dp2>BMgSsgBDCf?sdXbjxImGS38O;
zJ5U=-?I%*3gpQI0KU2W~MjH<zMv@!s-c2p^mdQ#PU!s5XN1LU+6DEVPv)Rvca^wQ2
zYC?aP7aRx*4<9~^wuHjHfcvC?oyVa~l|+&wQWeO+0S3A2TYh0AVVJ#1;)`{Ts2_Jh
zw?bWno|A>jP3&>L7pe!0m?7v~jxDu_h7JVwz+~@w#t9v~i9QXz)n?MB66jzPNIQ^|
zyaM*gt-{5G4r=mCmy&~TD5B&3ICj7oBU?VopnhJl;h?8y!U20Py53Dwb|BilNgfOM
z^$r^*O`5b#v=O$?JY+Kv3>06H^M2y$9E97V{0ixVx|rBZq+Ln-&<@&xcc1;t=;uQ?
ztlG5R@$-ECrv4p$QYNs`LsXcU;5>kGsOBJ#M7)FFPjaRjRa173wczxrQ*r@8q!pCE
zvI{xH0tCF!1cCxg`Vf-`!4n(5Wbq*?3@G;j?LWFSxZ~cyEn?`!*&<HmBRtw@Uk<#X
zBF=~*h5&pn{@pdR*9?)JN+NG`!gbk8Et57gLr@Hj-So-Aq|TO<s0;rccX$9_Mup$w
z*+H`*p-VmRZUefp+1kECSj1YP>*Vwil`}ZY0Y)<CgUaA~yXIq`->P+<%z0Dp9z5%7
z`qfenEUuVBAI=k`+Sv*+F&Bgxer@q7iE>w!pJbqwk5S(@ct~vixT|`3%Ru!=)uXlF
z;>9-ybjh}pl2Uwb67!Q3+MT02<$G<Rs{o3?KKq$E&}s-4vD(#o$Bus(`J13;36mwD
zSHX`CBm;|vzseiou(YhM5(U@Xn>TOD`(5lCL(4MNK@zp?2Sf<;LZ=XN^VVDr7?Tc=
zVyhmGvX05DUu@V_&fB*;{F6}bF*!!JHWd7kg9CTMTl8*j^pq-3k^4VfkBO$eCeFs{
zZ$U*t<DJ92iD+rfmYnt1Uk6YGrfz6g*<q$c6B~a8?A>?ho=3MgQ-3Kc?#)fOzp;as
z1pRjnee?cho<55TilwTu^Avx0ZZk8JG4ewSv%RmU3Dbe*p$q1AefuUJME~&`esS?Q
zL^%qB^KEyI&_w>W*;bk6?vUo#m0SP90@GRVG<wU7;aPGy4m~*~VH|KZJ3}!Ii}PlH
z*NB~lHkd(U>_Pb$1qB5yoTq^bqyBeYQm-%u+lQ~k^wmV6KJ*BY?Rs3xNvw~zrppKZ
zaf(q|N|ah7W@NG}PF=_H$vV^wK<WKqv}SaAK+gX%4cxkQ>wFzl*+i%>LtENuqUjF`
zf+wtoT${NTB}RH4|KVpJ7B`N+><aTO7FGH;HBm1)4Zl4=svKUmWpv{vGSH`WLADE<
zk^jg)vYlu*LCqo|k=IcLr*IP|R$o(J2GZkL?j3YmEx3cOjky!hRiO1^d%H{jS`@6^
z9UhOve<jumMho6*v*fOTXQLt<^Rlj)-*?zDdFZ9%(1e(4ai%8`Pk-Ok<@^!BeCg_P
zR2wv!G$@%(n-LPa5k4yR+NSo|jT0JVg*Qdd5wp8x4a<b2D>gmpUAGn$8+5&<EiD^L
zUI8c_%DLTlHtE`dkgSR29jW?W{@C_1XRi^eN0+Pu9dm@#Psna7XERW9xRgH9DM>Kq
zW%2_!_Y)rs#{e<Go;ALNcFKPU0MFymfe^_@Xhq;#3e%tG2Ax^sMFh=x3+y6{8xbr>
zA3P<ke|eU03_!g#{AR3QBPX<f)el=hBbUvA_Hu;>PCxHQakxT8o7~KZX=*#nne`vY
zQG}-x;@Dcd)HW@<TcE3_uirSK^;^T#D_9-|GI{TYI$#Au!sn}bqxc1lO__Q$o2sEJ
z)6GKOs{Zwo{ZEwY8=O96jS^0NbU%>R*KBz%VDa4|E)eC3eclHp)VThw==xohiB9wj
z&gOY4#vb2|0w`)g?hMxTcc3!;%j%hnkG+AuEBAUMwx3_=R>;Es@{`YYB1dx;IA3w3
z^Af3qN+cpd>vMgAfj2N$eb7;pKETyNP#5mQTzK^RtQ%VKSyCEK|DyhOlLopS@(Ofw
zqEC6Ttk)%ViTeQSqt81}Sem0>IGlxQn<TaK7Yx*$-EQ;<UC+z@ql}hc;4#!$z_X%G
z@7Lt$72k@vqb-URG4RT^F*alR?Miy#H8tbFMDz>7*qF?1hU-6g3b>4~*7nT2n76Oa
zp*sovEV_VSA5dC|ipv7Q16=v2rPliP_K-x)I^l-E-V159V@4V_qg(U5^XJYz!r}H=
z&DT&AGD{ZQpuRZXr}iemrh)|!+?pD0>;J>x&n>V?d-kw#w+bk)QNuUtd&p)E*V2i;
z9y7^if*$(%hPei%n8ywu{?xU27z&`?^YasALEgE005ret(?txeEK_L*+RV9pS$t+_
z|DK1AIC7jL2Z#GJq(?9;=MSoRNyk4+hP@%M8Xt%ryQW3umvaKsyL?y8JCZEjYk#dZ
za}Bk8iVA6BvmU0T6c-fybXi#Q(!3=XJI)3CAn?3QFxRvBp9>cf2Fde$lKyaRtW^W2
z`9XZ=OWEOvq@Qq2XhXfnRV>G%Yc#keS11dTQ;z?7$Jqx&{Lrx?x#21SPf=>R&0j3*
z_nwzZ*?F#7=+&M4RMs%JJ8oXb-2)H$kO{s#i>605FR0meA`R`%*-#Dm$oCN|7=lg`
zGK|$nyFT=mo2`#Uz0<)BPhyesu3vL}U8G^mv8h_1db-wJ1t!zQ>^x<cF7Nxu{>3!4
z*9e>1d)icCk|M5V>nK^1-uOQ!r6^j<Z$ptKvaQoj$iQW_yqAD$?15v`G0U#BUcgff
z1=Z5u*BA0*glc(qY_)JJQ@2^M7rA`nB{M5_0GvAd5fof#YHN!K3ffjw@dda5>4lHi
zo@BJNzFYYXnaH0{$E!Ga;Jn9`Xg<1V(Z=@b0>ATIuRqzBNS!4t`?K`dAlYoQe?o6$
zJT6HK$q_2@K|cA@Q;_WPmp>8Yd(?~Iwg2}|1cU!|F#h*XBrp2qRsZ)-1S9$xcm3Z#
z5qv?AmI@~CfB%GIuzz{%|Gv`y{^|dp7X*S|p7HXz;obcxOMmmxJxrwnA;MB;9u@{z
zF>4uxC9<+QckRM`oWghH<}ANtZVhIQuH^g?aWPDRwqG@<;q}uNcDo}?!BG-u2XM~9
zRuP9DIKT`Sj(2u~WZh28UejD3l9i)4py)FG8QO<^h&Bty<CJpq5R8#DY5Izn!GB=p
zAfcuH1LOgjr&A1~v|4z-5Zenlcs=B|%M|Y3LgJ48I!z~&aiW8)*>d~`Ic(6sF}K&}
zE*Cj5&RA|518u_p{EvAWM;{*>>fIh`7#13eioIJ1Ks62oLB0cw47dEy*oatd+4QzJ
z07jB1@q}P)x@0$CQ7xQms;&JT1^ZWBd&Lw2Pk$qi6Kh4n15)<?vG-O{c{SU%C>q>d
zg1fuB1$PbZ!QI_8cyM?3KyY^n8r<C-g5SwnYwvy9z5m00y{(;k`uJo@QM0Po7`^xL
zn*`_s;GTd<5@44qP^E;0eF18LEmE9hNV`)oG1b2O1@_p$!#u9QssR8QEa5x1Vzk(g
ze8xZ_Lp_jH_$U<*B$dWRodzR~AA#6ngm?o`p&FC?nT~|zH!lc4P}<W_07N-=egmV~
zSy{RsG#=czE(zDYQqs~!)#>yO(SiV74(e-TuQEedI1s65>8M{a8mM~#SU7vAW$XEL
z()U_oCP~05n?k2WvM^LzTMJON);heJD=TH8&4GQG<6mHUrp332ERebT&R+K?WmVy&
zZ(%_L@nBSBALuL@pt_B6bLo>Lb@DY}l(ENUR5#ji^Zs{8@W1FxbGt7B^UUh_V;cnJ
z$bcf@-{-6Z@FC4YKhSp?z8}c)dKzGMm4Od@WmIqq@B0m#!;P*T1bCl>hXM7d2QY({
zto}txN4x<nuT|gBe+Wcnb2*5_AOm!ttvn#mGZF?&l7l+z7L{C>>%ay95ZI5^C_d4R
z=9}b*i%HP2uqaoz82zY3<pU52-|@jg$-ls9V*lWvdbSw8?&udlB-eM!t_1KL6gLiT
zFE165&hE5SWB7piYm^WHBTBQgq1B)V%&bz%c%W{VwC&d~E`Xob0zG%r_1~wFWibQx
zZch?%Ia^p|(Gf8*$)X?btRCLX%K$-gpa2+ACim$ipz?Q#^4pqkz(COi$|<sAb~x;W
zdaG`repESOhf9hE2Gc+jK)t@}m;#m<eE>IjM%7a-nWMy`poJv^BY`JNegpc$=7{`5
zqVL%f{eSS1=4AmpGvHP}q$L$PBu3!RK+-#kcwQUe>uL*p6rG^Sxdg-zj=?~&8nc=s
zU_dPH9>4=h{N0XyMj-vzA+D8l+7LdRc^VHD4REF%B>}>cw+axy^>`{q{gUD+qc`!B
zeRvas3s7PLQF}EP$-9sJcVaBSf>M19{wPUKYQ5YTv1FC6=|I;4(1apLSukkkwUfk0
z1fs&n(;_oJmOWzN=Lw=fg{C|^n!A;ycOWA$#>aMG1Qh-IA9m6RK<w6=a04Bts@v8C
zs_6aa5PyIa`0g<`K+^)2;IROJMHcmMWQz7*$TaQyo$d4oIQ{Z|73-(U?eW1vG6W3A
z$7Rt5A`5W-)1e-K>1#A!4N#c9AD}n7`~~7|WD8ov_&&3~@2?Nu^wW-XAN4#QOBX3L
zxIKW5zz?XI$zV{&pV0izr6@{BgnHsd!ETrX+ucBX0t}1+Pa6zO05rwLz5xIisQmXD
zjDf@dKR>YqqV`((*+6C4QagZgK=!y~Hw{EsZ*7VIFT6YcAkWT%($erE#fttd?;m5d
z?CkRfQPtYb@&ZqOfD&0vzB;==ABe9;b<edM4%E2<cS27OJjltJPre!e6x>Lwt!R}7
zEYN}47LC_jBE0{N0#jFj+v`^@RXg7D2r-+?w5SyycvBVlC?HNlsnONe2;3=VK!nl2
zbTlvqFz<ZffT@OF>%;K_Kweee&H(cbAe=sBtE{`R{0j!|Jz-;-PJaNwFTnn_A~Vy9
zC(Yc{WMDe<;{d3@&#k!4#-~(5YU*A9F4cz|X_UanE_Vcg$&?Zq0OaaK*>wp5wjq-w
zr8a<xR=K0c(aQLr{*Q3d_4L1hphk@g00hTLm~<<-mTRe^W>AfC#0%>H8*CueL8WmF
zqyhymyO}=dI93o5au5$HtWE&qG9ZObNx{(pNilKofBxMLXkzgvM}V2h9zS`7LPv=S
z;DWjL1t^odR{$({X6pdC8{p;8UjlEx0YP_l_d@{Aeo%AxLGO}1Od>TM*o>9_Jp&s7
z(hNs-cxdR-5o!8XKtBseq#NZz6g=Mn`zBzmNHQ0llR;4F&{nlJo>vG~|3!KOtXF%Y
zK$O57NI(D2aESiplnKy-bZP+4s|o<(z41u`HaD5lm0ds|7-3*405QE?X?kf%NrQVJ
zDGeCyVF7xRA(SZPQ_DRGcVKVpn46QMkI;Oqnxg5@X!xJ&QwIhan&m1WWgS&$1AL?J
z5^Mp!=CMoYBTp0LSJiX9y@MN5U<<qVt~+sOHu9@(`&-|C&CYB2Jtse@06``ZQSJ*D
z0AT^fCII!uwPhL(2_FP6y8!b&U=yFBZ}Q<iW0#l!EGz%|mw(I~BCWtG1@PE`wGo(!
zAFreKD4SZ1qyRn7f9Nmc$m;lj)m1y-EAo%6vJTKybsy<h0gwL1H{f^5Ux4Hr_6_hz
zg@+%n15T?2FBL$-)u({}svqt68=xg0=Oh6dz=r@D5Fh*aB9O2E8W6wyXa$;$07D!Q
zJpgUoZu-MMc4_I~pRsg(m|B%40Zahk`|(S3=^w(Rk3*E|(Rl)jPXLMlP2%FC(o*h!
zc=5oe!-qi8kKF)FpQEA_z<2N8r^EShr7Rl;1TCd}y}-Z(((O<Pxc|PQ;a@(0wGj;q
zur|@gevq|*MAlSapXg_La&mD2U{A^hC_eUyKdu_*4<MSGvkIF`WiIF~0BoXs0lpN#
z0s_JgYEi%_(hx`o%H9@yLHf_vlL1>>AS6o00ss!}t!aOMf1vgsO#I-`Pz9!LgV}d)
zXCT=Nn2e*`#C`MtJfJ#t0BpQd?7s8x@Bp*+@D@O51$cu$o0}En<==MRs!{+^9y$FJ
zU^PAL0BAJvQBiyO$Ub_q?+MWU0=bjTogLDw51|~eO^yoqFa>$%1=zkW<3N4VQoHXT
zA}b)ojY_KD;rYWu@}mbs*#SYo=H@2Q`2f`jIC1&1HLz_4s``e+kNE|a0T!QtH_5or
z$5~ha;>REY0iWjEGoVETWI7*aDgfXC#`KRJeL&<6pi6-22W-m?&ObU#Nf(ITx&eET
z<c~9%yaDB~s;EYR^j9jzzuwpf>_?Q|Dy}~M5Hqs^7Qx<pU~y;%YPdTEeRRpV7T}P)
z_plAD|HG?9AD7_70mM}-KD5)UoFCs9&IBI#GL05<TV&vb@!Y&X<!(U2PC~~6DAo5r
zeOyIa3lISV_;deXnECN3@NeyShuSysTZ1*VHF0uwG&QvOcx7i~3CqIF!a~GE^zo99
z50+8Y!`_sLQB%Rl($v@)mQlsU$oXIINZA;go5C_`SeQ6}BjRRaf@PF6wJ`tYOvLr^
zTExQHNzv3%)Xv7<&eqfxco&vY)Xv(@QN`ZS7&xAosjG#t=@&;s4_HQ7Q(JT37@RE3
zME`EEj1m^s&Zdq;j1ty{&Zc6f#&#yAumS?G|G7~2<+C181P~CA-rTn`J4DW0;N|~q
z|F;MJ|JMVXzd_7?QP%w9L&*MrJcR#wL|E9kSpT<=$bjZr;z|>W--CYMm-RW3OF1K+
zqv#!n<rA{Hp+9U$@&xXv7cuPO<s>eI?>=sIAmYk=C#O6|Xgvc)b&IcOP~BLZaGigj
zqX^X*JD-&*LX0P++cb3<lXS6da2Cw!mgxj2B2S_=G~Qp%zx1s6%VCe1m<fd+c)rGJ
zT&UZv>(do&MKqooS<or0clbT`DcJe?y2KxwxICttZFIbXWTx0{_zATzx=)$p53<<S
zCQ*-N7TNrfTXTJGuZ)_oTZKk-*I!>%h5Y`2MIw`KHn#7)PsDCK1A}^B14o(ovai@)
z_>*X<rrNIjOMcRwOT)&#t_POKwI><}{j~zCwb<g>nAt`GY=MfyekKI}Uy<BA7qhhK
zC8M9>X#2iij3N{(qASyfG>V8eb$QdTHcGQp8`h&w6wyaJ4}*Eqt)Y@c3*_P}s)adV
zAs6-yh2tkK(}4|1iX(gM4sOi@VhOUN9FpULJ%8-0Lku!2dZd&m3TK>AyJ{vSoz6A|
zQHQ%9Lw$Qf+D#z6xDDGxOLF~of)}7EhK20oL^U|rr`s*NRYUy@Z0}>mtSs&0kx4wH
zoObzBzulZw)7;a?UyPOKiy|PJK!R&L-ZXw@xy2qy00)Z^=Y{`Vj?pkPT|XuluTcu(
zR_b4m=~LJUe=uBuN>&qr9$nQ)<idx75RM+akT=Q;;a_rKgP)HRSH!RLQ&$h|lf$;k
z+N#FDKE(^ig3~cgPZCp))%xh40Gf3f4YfM*UNvE&2CaG<4oFGy;X;Z>d(9108TnZh
z79p(z7oE@0&2lFWaNY#OLTwWlz1EALy&L;pXA42K?6<kALclL04)Ivbd}ck=#MlrK
z`6V{nZVV_V`lngkVM!A-#Il106$c$`9<SU<mPE)J4Goo=#Wfv#0!eMtChFJwhp-Qb
z%R?i~;{+(mipsp&9nc2w%m6Gsmf8y+q5UG=B3x21@x&#S*m@Msw2VQqT|MlzdPgp|
zPpoY82YhCc#~75TjqEv&M9IZrjxU7phe+OtND}1YStQX8O$`(Ofa**D+(i_L7&m%g
zuCE#j7r#Rrue07+`>_IQC9+r%vfi4F6STx4+e&>X2GKV9w^5qpubl9)vSb<UqE;pe
z%7}_jWxBJjA<ZJ?s6r`&jku{z-xUsV@|cVMT28me5Rxl;pGgUjgOP=C1t<{&tBT8C
zo6K;mv@n5wW3>^n4_fdURP{HKg!F$#c9f(da2-fbfQwsk(Zf)15a|U=DKL*ezoDM-
z$rtHW(}_fh(GPzr{@x%KS%ygEpUXY*!&oE?gpm`{<mU!g$_bx|4^+f{EtIiWF!%W5
zIvwIGjzzH3LL2tkfl;<F%P*7$7S*r{Q3HhRmLUG1q<j1ChgT~pxx;Mj7`HrWvQqQZ
zZ$!ROT;zd*{t*O227*Qsluqp8SJ5~az0aRrFl8WpA8;_CJmxvru6>8SYe2T@^Kw{f
z<FQbf9yy}VZ-$t#Afj(S)Ay>As$wys@q{OAqrIDtMLAFPN^yr1lFK<hzKJZEhzk;2
zEYe$k_rR7~+o{$mc)biO2fc^aw1^Az7;j*|T~mAHk(@l*LDQCwm~et<@0kkoH|Hj*
z?5(e;r)e7!=oocKNJuk^34`w(doXCP4qwaKbzMXc`w<t{(zNsAQI?l$bpmH!j6+ZK
zds0TO<OB0@7<nU09*R$bptGXwrucDMjNa|*%pG#3kpEX$gwbeom7Zw0GrbGNG;T*K
za`xk<`d_^A1m)GMYI{8D2tz|WCc<IH(tOgvhF4P<cY_*kLp~&wo&;hpQ+55)tkB99
z_&Vg0OL9xq59H*piy{PBPYa5jr8PY<@uy5wHpG30r<zhdrFC82@TjpbK?38=C2||Z
zo@!STxfK3ECn*g_r8Gu$mp-ks>Uqytl%M^+xo@7(Z&7kiB?vvl!kijWC$PJ3#=Dma
z2MM<u+#;l-Fc;*PCYwdgaH5Oh>AM?jymB#%7=xL9W`G#S6^D8}4WYib4uatR78pvw
z<{hZ~o2PdAuRM$!Kj`;%=7ef}ToG$(3>E2Va^&k~e8#WdI<S)-*DAhh{ndoIO`Y_{
ze(Ikvn)gKA7NN-bWDm~0Wftdqs=-_XYc}o{3`wD@FLh->9gwqBJ_#Z($o=};XzFag
z`2EFoD~~9F3971x!o4nHX6l_iHy`%88A6TqRI9hY&yiwNoB}k1I5Zcsdvt_2c3m`(
zqK)h5L|;Wz?`}Y4nizjgb}MBZBldf6Ds9q23%P9#Tbu+9DQ6ix{+Az=^O}YDJX>+e
zn`F7)+@V46`5qB3ZyHvi1JV&(m8oTnf4{btX@A3+3LM?Ix>O<UrNxa^hzYiL#15tV
zRB+yzCp>>!h4{i>v)he8Q0-{k(e%o!Y^J>Rm^VrVb6}nZ6caT^q(vAr7?ETtzW+^K
zz<%i)!BO^ynBey}WB9poXS$^gC3iM&FFXQe4$4daf<yPUTvQ$>=wU?pTO{p{X+s4N
zjZZF+?kr}r*?V?QKeBe1954TRvW?>fMQswS@t4S?vskI2-{f?Y%qe0q1&@+MMbkMB
zgMs}Brz13<Q`=%MSx{eO12ZO(*%Xb9YVXQH<M82xYc9dI*h)h@<@bI9KcHdaD#SMH
zW!<$^#{qBc&$?u{kjFpFI3B+J8w`_mBKP-#`->1f7wc8u;Z(XQ4?lsY1jMl+>Ci$L
zMfk|AQTOd#26ZIgjq~o7nL!!p?Bm2TC4Pw9*|bU^Zu3PfB5yebd5{%Gwfy%2j4lX)
zkoe!3@+wH^1NIq3aRQ?8*25Z3A}Nb#V1-1EHU?F(>BV$6cH{1td!EjIg2r6!Gq;p5
zh+sXUr69EEvdIouzsDZi*H}~u8Q{tYzWl<g*D`ZQs%mYC)7q99%gizzY<w+`!_I&M
zgOe-HN?Wu<<;2APWPM<K(4hGu(aotJBYvLH)%_(rU_zozX0oyttt%%r=p^9mAPezj
zW-&2IU`CNOWi&pNwU&v%z>kM+Gw_<v;Lsd&#{=ieI$_37Hi756ZT&QnNzd<_Z$Gto
zB5gcZv}jC;A-R8rfH+C3RII#-|KbncdQow>s_mRO$sp;}8rKh($Bi~NCpf-RtDC%v
z|Mm$|YZOA$62=0X-nIBCQbjRqSvRL5^WLu}(@zqNTZQ}1NvN!gOS{aX!h*76WYMtT
z<>@b-LRRvIa3C&wbK4Rul<oo-#9uTI>#v*H!`109XK($xH1Y-B<*|<v6RVseBkv_n
z-na?-bYy3Z37G*?SSHhNRTRD(m3l8&p3PLy{|ExP{{aNDGqL<{L131amea~Gitj^>
z!Fn!@MKrg)TDH;9J<SzQ*^j1#axNrMN3nc(Fg6hB?YAdqg@?^tqBa@@{97HLhlM`(
zpV|Wz3U=#jT^`R$g9GKr;^i~*uYJ-b75#-I93#iRZBGh?%6a|satVXe2VL#YmygF1
z<POdcH7m~#U8}y2VmZzFhYt%YejD#^D^PELhx7%#JKLTvUN_c#**rZ%q{K1{y92s}
zhhP{y(O6Yzsf{cQuxoauDM{#eY2sVriZ#~=4f<Z!dk`q1V$Oby7ntd`fTN7Y_^vgq
z{c^SOY9<f7L$jC*tGYX?b*2BF0rkY6^!s&$a3qRjVry}(`kXgGWYmNt6rRqKUR8PA
z<K;PUm9-@r)xwu}F=p4}xVl6qNRvIALl2rF5-q;3W}~T0UiV_Tem#C0Y2uIC8~oqW
z%r!(-p9ZS%m=G}772}9Ic=+PJK*1ivvY1i&Z^f&Nk&^;?Bc)f~Z1gvVIb%T>&iL34
zeCHC!vZZWc!VIc;hpP{AuZPQ76`Tyv$|*jqux2Op2&>>U$2~Sr*YOLk_x6>&zh3(2
zPXv>lx>#@)<quf9Mz;vGvv?hg1J_7nW6+NL+pkth2m_3}oMaln!*VgqP)orRYv)*x
z%A8TGt6M{<R%Yzpafd9O%01J4{WCi7!Zp8V8_Z~;kH$ebAu%|*LWb)nmLC?FcU*1-
zc>9NaqxO-NU6a9#psI(E|FfB%Apcfw(!R^fDcxT-NJ!4gvST+^j>K1?8AUg_AKj%&
z!L)JL#K>#U{6y9gBPXMTkaU-}+j3$fhdw7b3ibudET1efDv3M~J?<Of8fANaS;Qs+
zLf5x8l0YP!?*zpZ1jRvD1}0E0eM7BB4?A1v6Jhl?FWa|i^5v(?YTi>SOLq60TO^Ad
zf3O(JiNUds=ERwo&B_?#cg3)`y;jhQ2><8>7d70m)7RbJOk`QW{v~o~BJX$2MHq7N
z-V}3MyR+v%;a8ya)WC&KcW)M*n!L-yx^s}*=$D5Q?$#1w6-`fFa5`#CiIKo(v=c1n
z(v@5(rxHUP&AxKS_U1&BqImrj{;7&cC4z5cq;D@U>^E78G+A95;b-`QCDl#y4Ta=^
zbG)|!ajcegN$Qy-8Hm4!39`zEvLP=o>BzVt!_iy4okYPXG@KP9`6p5RR{D2TxE;k#
zX|Pq--J1^sumvudm=L-HLE0<FBiSBw8#MbBwjxDGUPI%22wq_zi_6(K2j6t7r9fm<
zjrQ4(s=0q5Lm^cU?4om7hDCf?Np3pVr{;eACY^;rGRcLq_D2W(5S9+6`h1mL38zfR
zrL_4xZNF<!3V$><7KdoqBfkFD1Jo2YltnR`syeJ3yo<i7VwH3lq)22){PuTGcM1k&
z!0z|@LJ}Dk)Lm>x3CUy&`Q_FGMJr8X>Tc=_a%zhE`tonT_V4gknu>nIT~23u;tH`W
zA`u325=<sR`!ivoA_)`QfsL_bPzf;iVl!|MOnV3;EQlq_T9WwI-6xh>Ua}as$AMrh
zxaz8#`caTr@Do>2+Fzp0E@W8|idqDu+3fuAw@T0cW4$(w@>MdwSg-UH8l@;3gMLK%
zQ07`>gk46QKh^bjh-EdFCfl}V($_LY33MHuEDO1XOsZki+sSNQjjxl>)+64vWY)$)
zk=%C2Yp^WkyVCPn96GQtL3FAzwKfM79v6F&%#9RUGwsTPpyR0#JBahScw?X&d!MkR
z7*@ZE3_{?=X|Mbtz|S0{T9wxZ=fk$G(izGcShhJW+tK>9-Ci4ElQX&7yNB@lsl@==
zk=XGz|20h#^gZtnrF59Q(0;T*iF{`4#3Uq}`F+HX995~bqRrr?m-LI8xvpBu1re+p
zdQ+ci|2qW)rB4T$oQ4Cr7S4oUpTsP{C3FxOXvj@$;?42*tT<Jpiku0ZvQv@{{HWcY
zBbBB=r$6cNI&%oSt;NREgL6Q3tD7!?EaEzXZqW2#OxYT1g&2UE8lZk^M0SbWFQ?yf
zxFDC@ZD00SBkGYqihhmT)Ww5p&?KMPOnC#DBU}%#0EbD`Of?};<Q*;v+VgC-qL-+Z
zM`*QG;WsKSW_*z=_6HeB!3ao}rvZx}&gS|G6^Tdf1-BriSwc--nN~-whf&Vw{2MEd
zi%6q?6!R~%oT(5x8Zs^BHtaG^9@_o7TTEmOOT6wgfe<<RPobU}mtx|1n$(@sl5x7Z
zA0BCP7}O}7a(~5Wet*sWskU*7k6lU5^AO(W@S1>(9Oyvh1)osQhC0Ww2Bt7|`j}x7
zAnXFdib}spUl;^MWKPB{pTEl$@Jfw|P}uTC<3>IG)Z3G6rk~#)RutDs6@K~*BSb7m
zS`>+mdJQ#@$5>xyWz?}oH~MmtpRl6%bGZ$OXd=eXpi4a!qVo(h-m_EDZYN{J;qp9A
zgkYsN$e^orGf}9&2T36mD*JUzc`r(33C3H9>)hK=L<H6g-1<qx6m~6}zBCI~)#sgi
z@Ds>M+rG;uBZsOQdz@G_e-&6-Vx000WZ!C)h7(W2)+VW1MWV_~ACV?VX^9kn9f(z;
zCvhENo2J9z&@eqof!tMRcivcw!Y{0#o)JLH`GCH1mbiu^DY!SzQw#moNH5D1hIoN|
z^q!n~U(=p|`2KzP#>khgOSj!p5ONKN%BShz`8NZYWT+iigWRbdq3HAdSk1){2P0X$
z6n7kq$x3F!Y0AdMMLX}OxXjeC@v)Y%8X`^e-sRx41yqb{oSM_8*?l|&Y8z3ODi64j
zqh0u3M4`v>Bt1?AH7IaVr$5F}NM=|BWQEhcCL{2wco*j5P;RZM^HzaY!^L}zaK;*(
zC2QH3ZOe}<uI5j^xot?d^SxRV#H;ooNPJgVVm>VuNW|s}$kB7k!&^l(&dqY}yWX3W
zMVqI#-j|;V+?Nu*n{%_qYf@b*lPKnE3;gY`itKgMW~;yLbLID3y7H3TNxR3Euna`!
zdaSa|{)@p-NG~v6(MZLB0#bdtW~QIAK~3U7KY>(COTjl(Cy6T0NTdfIo!NSf&l3Vv
z8}J;)^<=xdJQwu1z#Fd_!g4=nkpDtc!>MS@CrQsAS|A^S^@Fi86}*-W-+J}&Y6UvY
zYCUHl5s?~E;_}i|G|HN@-5>4Q_z5>o_S=OoqVw6lwtLT}KHbjmbMs9OU2^CqY}r-e
z4m%j4l4*g=v?t0)Wv(q1u`TdMJDJo$=)D8jt|04RAG9TCkUooMr_X|j#l)zy!vU=9
z2SOA`3VA-oD-FWNM^P!(5)sHZU&pYB>hPv~N(roNK|<eH2%g7GC8^EcbwW0f9_#=b
zXCM{NKjA`Fwtv8doZMXhTU@xNHIcZ|iqLhTzoq??RZGA({3&(8nSwE~-FAh|qMntX
zjC+`)Lp_JQC82%tzJ^iAzkYY#mSgsQ=Bfu>ERYBUOi;+tp)D>-pV7Ca<?oS=HD=2&
zS{9ux2V5eZbtW{<%D{x@6*mh@%Zn#q?zT>o7F$f60iC5~ux;M)wo`$`X(ynkXF*Gv
z6+T9rF?`rJ^Z-~oPPA#HSm3YebQL!fuKNY1EnEotigaQnkZb8cCTs9uXk{bsAc<?M
zR=d&D+b+xe5Vu-LzN1%Cms`8DQTpTa#Zg|qAb74@k0D7gHCo*4dO!{`b7KLKyRbX2
zJRNsPE4b3r%}c><H0JF&bUk^#^T=Hd1vBjIh@0uSna_?85!u})D!Zjh7Z5cT&P_(e
z^u`o@ccP~w052l<b%>cc&%jl8%{3`7xoxsIl7mS=zZ>ny9V#G&RNFgiLb5tRbhY`-
z7S?)OKFfl0!VsR{IBYHOqXai83pRtMt2=TPyOwl(a@dpRQVvTRHjz5y1~57|VNSn)
zUOE&0tc8kyI`TTRS?ReBDgX7mQ+&2XQ>7HQH8alL(EjL(Xe1wD8fl^ktz6V6+D+8Y
z^K#l4y!f<J56y3zYje|M<yvi@5f)6Fd7L*zXuhHqX>zztYR`z(gf~7!W$aC=^)vs9
zj!_q%7+X%jR}#ld*Mr2u^->Jvq$Z7T>$8NTy{5aCz4VLK4)~3^_1(?=ub}Pc-=I~0
zxovs!c<H8>o@R1zp~o$>=*D!HHqh#5c(S{$<kQ2v;t!M>!rD~tZRM@4LMDt*y&QLP
zE@#-lIa3*46};K7cs{os5}G}5h?xGqo%$<<gixYEgRy^#z4<v}B;Tqwybi_^SuVCu
zZMp@qx^0Ax8Z#nSJ)FZd2ox$O%0TeFR@laNwcO1rpYGyHkiUvhyb))%{}aZ}9*w05
zN=B(2W5eO*q2XdS0TOcDGC|848ff(jnz|~t4vPilv-eQ+QGawSwYm(cBh%F*Xm*_V
ziDu!`V?N));IawHAC*D(8bw~0*;xa%Q&Dt7dPc1P6Dn+bot~|_(`KhXEXcjqJd;Fa
z1UT3`@@DPkTTw<G4A1pt%mIa|aKgfnx`NrY2a6D(eJ)}*EzD}hZKN%goH6}SP((8n
z?M`knS#bR~*u7&`;xzLl>S>ZfE_(aOu+-&Az6w@cjCh10+)u(>I?qhFP9lw0WW>Fv
z`4_rBFL1cACyzJtvy6iaPIaEZr-+KuVDD`Q`xE=9{1v4<UDM+u4K`e~@!N#a?K<L*
zOh}^5O1E1(D?w72{4JNPh;hD?U>0)F-XMP$`0c?FO#9mIeisdI+eJv=M;4@TG5NS8
zM%OU#ZFvFb6^}^;-_=XFl_$CF^LGyM7US)isz_0+dLPN>%H3je6>Oo%qitAQMH(Dk
zzI=D=QMT82HE)3e_o{LCbjVcZIWYY~r4l%oWfiS>>@GS;wxg13cI8|PMR$TATtkxZ
zAEE~BBC8hYtn77J#-2suVC3Md_5HFjQBa__J8%&bV75EC?<uTL{-Zl4xh<+j;`o$$
zu6^tF6WfJ}S~P$3UWL;cmB+Zz$VK+t`={rIAfVK`&m9fE?PZb>rSw0hjSwz?+AFbI
z{u(7_#ZfgkJ5%w6ZW>R7;P_tkru4<zz5BEUYW&R$O4gX&;U(fT$|C5!#^IYK)2<s)
zsPK4k-XMrzbMzY58<~_CRVDAKr{}#X)QcBWcHs^-(wv#7s#I)A_zyDty3IYVoGHXO
z+ctjxwit@bag(n_h#9lN4gJQ^!VuE-P$;3PwD07<Y$=>4M_a08=paY^Fv1C`vQZts
zA>O6u9MuH9w80foU*;^7;pPR38KQ<_qtiTS_$M@VMt=V?&l^4TH<*fmx?iWK&(AAO
z0;;t0C2)KUVosQ1x0znrY@u;>z?6$zVK3cVc_+t0#r)7ddAC}(r-(BRRx@B)L#&n7
z6>0(IJfd5n8^PKRtMBhl4VL*Ne7lY1rV&}Busxqe**JQyWuZC~0SISa;7_}z;0hBe
zR$?EHGdopN2HlOI^-{M@nVBa#KU)PwhnZx<69qYNy3q+SEL|0>u}$pCqdGH!ZSkq<
z5Frd2N7UiK!r(WKo;x}<8ZHCFoo^a+Yt^LNt;Fd=>yn2c@`*`*NUbiAnDU5b+VWqk
zS_<kOFk^XDrU+RNq9ipwQ^bEYF5mh6nM@v}$$>T>+F+pJZ15o*!wIh300GVW^SRpD
z=*n3_`JT<}WuX~s8bk~%qT3>*;4M}hyRX@#ioMj1Wagt<%nzC*OzamQrYU4Lcylm>
zVV72et4C=m>Vk!xYM7^CTaDAUydPL5xrn1w5~rOT%R%j=^+}A&?3hZmIW^^lHFB*p
zyjaRUp96Q_7LiS7>U%g#DDYUzBlP%K1e1&s#zp8}wpNpH!5Db1rQClE^-@GWvmr|}
zRi<wKWr`_aKgWtM-$bD}b9*u2QtS~_<-~w8Jg4qJxLD=X5t$H{31B2ZOz?UK1N%Fs
zQWU2ahc~XeE-0uBi!hX^2wF7DKF8>VOs`f<JQDUb<2iW~+{$g?^n7*hqIn=IV*o#n
z#;99WrI2X=)n9LBaYxpcHKpNj;c5Ad?PRZkdD59zD5ZtK?jcqyxUAl~amjcMMv^KG
zamD4PL{##wi))Q^1;>9-(7jb|cG5E)V;P>P+PiYGR<tgDoIoz7=W=Z6d$-hhx2e1l
zmB4jgCOT3pZIX~0m}!*UAPo8?VV?m0-NUmZq{C}AL312b-Oos4A@BelmLun?D+@V(
zcA4oug^b?Hs~NYNjgDdzjaqZ!Pt~@lPSyBB5nFAM3ZtU>T2GyhF+Qg6`H*gf%%Quy
z8VE}GBxqmk_d<}xd=-3hRHT)V3}(puX%pcJS~L)ZQc-KSs!^KkN^g9$*B{r}I2HG4
zuczdqLGkvEk{%j6xt4e_dA^$;N1<#y9vgMW!x<`Au&+Hq?gUtHZV<KGc~6C1VM}Cr
zs#a@bcyW#+p26`S{WktW=S`&9n)?$+?F1c}qo6a}@Tuw@A{Uyu`@Nc54_M_LSoIMn
zykj9_V+4>PPQYATzO`&Epcx82IYY&Yas-J(3_H4fMOeD0`>e=bl>4c#iAY1`wTWAH
zItUNs;HYh}kMtnI&Lny&AOL)A*5Jg|3*SP1b`JG6DDaJESe<v_TJ*x*^FRi)Fhs_G
z$9P&j>=h0yBmlIcr4vEtew8#VuWR)30vS&*i9af2+(t5TCsxLPJk<ETnp9-3oP4#`
zrlsUS5o5bV@Qnk!2;WzgDfeEZy%guhc|-E6MoV|YA!1MlXx{EbsU7!bJ0~aBdpq{t
z4SK%7Hy<@!?>8DOD<We}7UYx|U*$c!Z|6I6wpT~nc6|@uWT&Wg$NK%eWm)oO9Delu
zR6Nfz4&<tY@ach-B;P_CoiIflD!<=qtLtG6V65Wx7198LbR3s{8m$Enbl{p1lo(;1
zJ+MA<u|wP0`hR6$-z)NM(vUo;;r~`-<Q;>;#4xxMsd$UxfQW7n;oDqllj%ryd#w;B
z#ENkH;<0S8(&ak09-_bdq%reQwR3%*%Ws?W(IQ;sYRw#Rj%=Yrn@(4RnBRJA+Ts_o
z5!Y}PoQ;}vzzGhxpPimRVSum>W!vG+HW=H8WHX+7co@^^<Ew^N%_bL|y{uuF3!?d*
zNG`Ey5%2JZH2K$Vb)VVoAMx`)#8p|D*#6gX)iYcfr<Gri+4_0eGz#=|>qNz~qO%iC
zOY&Lr+gSu0{N1QFtTOSq@%0PuorHR1LQ2Z36Ss3{ps0T@JM@Y4_~<UTYwXt7J0H)X
z0`Ejf>C1Fh*BMiFpczp1jU3vvUfK&$9h`rnqB6ceZq1&x#a;+{zc~NQ2V;D{d%L&u
zy`7y{K@wc^en0mTe0^J+D0yhU*zjq6dN_;Sc<t7L&4CLJ3Euha9{NM~_7hYzCWIj@
zX*&@jhB8s_-L@;7RfU-@AqRP8-tdeyhnoJ+?`3EBjLq{kuZDi&k*1opw#{u#)ES*O
zH=TEIoJAWV$-mX!4>#^&4QH#*T92#CE8fU`nLFO+ey-G(b~ICg2a=|1=7MD-%7HZ9
zc`v4832HXG*tqIOybMN*=~0n!J$P+!YS2n7Uguelc9hi^O5D{Szpj+u{FZaEgL#c+
z)+Mxji064a{jwSGHTmTJD3Wfh!%ySQ@{iB!Hf4-vbG@hgX-7CTN<=h?sj5<N6ilRF
z4vM7+3OyA``koid)?$Im7S|}QLa+mTuOyauFEKf?DQjG*!I(`OwUr>-8n+SjmZanl
z;fw-u{!GPvvm=5j5x>r((MdvYjn2l2!Po#A!N1XvzSM12ygUB%S^_`5=U{i```K!M
zlY+6UT>mLfqrpo-x5($?h}|!{<@qD1o<(?u?|)y-(CiN(oEG5o5Y*Hz^cDJeShvz&
za!}LqJWc$qodi4Ru31<Ayuy`G9_9~tfvD30LAqSa)vVl-vq8G4TVM+exjhSgf9aFz
z@HcZ2&@<i599T@)95)n%_jSVi?Jj=z-W*=RpV&=d>x|A+p-9Cy7p_F;agcQ!4Qwl@
zAYqJVSgGIs#7>(DS4fP*DCUO`dF!5g8MCK4M(Ll;frTP|ac?b{Am9RvNiy7$3PD;k
ze04ty6UyP#5o8ZS>@Kt*J&bow{W^g}iBsa^dSnPk`)6~DLb}g6ppc5MSJ&<nArx3e
z>$&2mvadc);N8t|>88ESlh$r7N!OF7Xfa!v#yv;ux?NC%x4zl+Jr<v+jPh}!F#dX$
zmo4=9afw1hbKwapw^26um8FOI<4pkw2DOJ7rjDNL7-?NKBr%9EprDcMDWaKjYeJ3g
zqWqKm?asFnPn<u9wvzDJk>N>rl#5`cAZE31R<YcWb20o_v}DS3qCSP_?Uf|9+)&cw
z_9z(YlgSag`H(C<aoeD!^ZBgbpkrVcGi<m+!s^|rPTsOqxMQKF24H}&A1}SRyH>N6
z@s76}BD)B&q8yR6xgwfmP=Oa@6!cDbkbGyT0>mGdS6GwIWCbx6vbs@X-V|l`l-^J(
z5K{Xns3GnX!&GPj^!kWt)86tqQ<ltMvY7CHkS4V5StFn^;+9eRab9T<)Nr1u!wLCl
zXrb#)&`G2j5=^_=a)lsBynOixP$Pa73+1QMwGv4V8))|2&|@_olV}-*R!Vgm{VWN8
zyc+vO#vix5^9#{)R%Z$i`Z%>L6WPuH{n_1%!7c%eWXR;92n|CxI>GxlVHHh$s*(zx
zTcxEkh9iZ2ahz5gqp-$R|8M0?^E=!1N}GEK=G9vP5EH{#OEF1kui|Q-TKl2J?gcky
zr3)kMoBK_SjY&~=&;{MFZ;i|y5zk|1bSFt-n_zww&^;T0Dl(W819IW%_e_|*lvlJH
z$sJn4-ImQi%6>ympQl`VNUnbBl_h&`Ve9HIoGJ2cXc5FAfh~nS<e5i>!L2V4<gAjh
z)$wH+)TnO9-ZV`Y>~$U|T%$<|59XuXwAq*}79);ZxtKOQ@(6>}`GF>OVL-X7(-AQ!
zqp|~e&LUMHYvU+*z{8avIVAqjpqJ}URlbwJco0QBW$+b57Qfyn8X^_GACx!!3u`N{
zp-QNALlxvCAtigD5#@>rtzu=$@7NzZmC3}SbB@EvhMH^7<{HV`R@98XxQ$Q}Zbd5e
z*k?vCd1?uJy<wsU=pYj7>_@~of~UQn2RszMYxU6Wa(P<n$jtk*X6|2qN@Pc`Ab<A2
zh49g}$tfdYaV0)fg5`jZ=jDAZ9F7vEW$0&7)az#+=^+qjEhVuT9>dfPH&bQ4B-_o{
zsp#4Fhz}zFqTHKXt9noos3@SeQqJL}h%2%<V@kW;DJ*L@<}*uh+?vkgO2hVdW(phk
z^Fmzc%CJ3l#gFN&kdPc|Q+58W5m66i2_kv%8g#c`-R%q$6kPHPVz_zH=jIf!i;YeG
z$6rYaz~5~DnCaSWWpE`cS?cdIp>}(5!^j*sz#s+*-7^I$K|8U~<tvZ3d__k3EVB7N
z`IEqbXja=k(z2V6N5y3(ZFx5@cjz<~pB9CjRFJ9I2{cEDm^gUtQ(2X^2kSMaqrk)u
z$E!@@sDVgD)mlY(KQq<w9-XB{xFXD?`vfzOV8}dwJf7N^@5BcRtT=RvLSadX;}I&r
zssq$lze+VCEB5H;DHiE7A(_=7W=-p)P~aJ@EW)%XzBQE(s(gwfMaXZVf>BZl00)aC
zg{)PFFBG*H9F__VIJ*p9Pn?US3hjmq^Rj}4y4IR!%x#~+jB+@7g%^Wd*46aq8$D#D
zSZ&)LfGtNa+SkYGGQEqBKpCd&z7uB`Oe+(Dz+>M<!m^e)Y>*H|iRD1^82L+`L0x;#
z(bxHNLUxV*ua7y)c6+GkFE<{B$ptXyc)Hfj*6<>QNqbHewFoHC)p1-*c;Xs)Z@jmz
z%woq;>o`&bL!@7fseD<27DUp!pZML_#mF24dLM=^eybhfM+#Md9>rvMOc+2LJC4^a
zMG7euI_7#tnu8Vz5I>S=2(nQ1jZO&)F6q=(!9as9oi9KJMspd^$?bGQURzg8wvOLA
zvyjI@57X`ji)dq&xu<*zrU;?h38|u*Ov;Jd%r_#btdWc#=9G*#(2T386ey$FybHE}
zohPU2GidXye|1C|t#nKM2B8X@5U1FgWzuXVPR0(l1u1qm^^(ShGr8c3+}mAnh2QI^
zMO!3^GO8Oy79<m6ANCE2{l{oZi=`BU-Drs;lBGt9F6a8^31`vX*ASnxu|r)L@$Zba
zGqj{CE6Epq#{Mqi$wz8Q*zOsZKcwtiw=A12N{Xfw!*z$EMjU;ma4yof2lkw_7AoW{
zR1Wt|_U=uNMbjRxLQiaX10ZCBCA8km!6&s%{2~@qt_mr=_zx~v7iR4!Yf>l>@kFz+
z$5s?N3Xn8^S@?iQfj4`;m*YJdT((^M#0i17Tl5VQX?4n?@R&QP3VniO<`Ok=#-gOx
z1*76y3JuCiyrEp4(;O>68!6g=ur^QKTWQVGg3A6&fUHQsz_zL8SmpCMbrG?cbyO(m
zXmC(yJ}j98I``2hQ+APlhmqL(SDTaEhz?BXpvmISTrHf$<Mb~~AWK&u^{H?t<fM@i
z41+oZI}#;+i(nO8;2>-Fs`&Xv4tF3^P|6LygEK$MbWK3@6*k<4XUrX}spU%l>Avzd
zB5)?^2;LDj%TXgOq=YPGJRJ`CP?DJg%gE9Fx>z^}{F6c!i9J7*+mr_Uqe46>UnWZ)
zYCEIG(BAuqI(FS&{I~(>MoYU7d?o>7&+F=X$RZ`;EX8G3um?@iTAo+5V_k}~QVQx`
zN}OgmGTWEwI__zQ5TZ^K-pLU$zYy7B_dMF}iQ?R_{oQs|SVarT8&(sBoC4w_dex?}
zn~IeH7Y&7?6KY2+9_o9M+IPQ~Z-35o_X=V)(<f3y7wq~3xmffRd#K#)pNEF??3#<4
z{nY2S%+VTF=C>%Z!jhzYp@Jd<LuwHONMELQ0`b3Z#fc;BR971n8Xh6hvq4Dgw}&I5
z=x{mum#;umQTl({NF(^k>ec<Km~q+jrsZ_2jUpZl!O(QGh5}aBnQk#BM>AhnK9na-
zUcv*Cgn@M`6L=6QW0VbYKHBHvYc+RWNE7TRjG~P{m~K0FHkMU`FGpqDV_3<eSu@hb
z$d;;(exjLsFpXMICtwd0*la(bh@*=f+unK#-<0A+aB@8@q3nSs<#`)?jI>?m_P!SU
zy?wKGmhleaHO<`p4|@hCfS&&UX2k=fcok<GSVnmw<`2@lgoUG%GZ8l@>wk0FnYo$%
z*M!}PuX3??oTwcq8kec~<7Rc5FPiXRT!<9RwNVJxjB#rLNcWmoleQ!{xCn8t4<6zZ
z(QWfp>N3ZD1bk+kQ6^XO6-F4^>q;Dx?gv@aOZKp#nwPR8ijm7bp&>tsDWHWR-VZ7n
z+fnqV49pWKeG+96O#>tbt$7f<M3Y=V%|r<r@*8q7)Cg#mHhCbE(f;U#egb2k3e=J1
z;fSLtI2I3DGU<W2<}UTkoF3(2Im|6=fVAVom{$@{{YfemoLG%pn;}%L>u+C~Zfq=|
z^f}5<ii9Jtje%!80^3#iOZXRBFB{>{(ybECq5N#de9UN<fvDs871AOVhr^{@5^Hk(
z)=Dhu;i>fJp2R69)ue;T!HBt-nSCl0Y=tq=-!SH}@L)O*-_Y=cu!;8NpmlheS;%4`
zbJgOM9bhQPlzBzQ;^yPh=HsTg3u;tlM6MKRc5CaoX5v#dN)MCs;+c+Mv5AQfF%3*%
z=I0buT)n$9Q47E{m`DbeGonpO8VqQt)hJ&gswboQ#mkKie;zhegX)lgXu3O2iUnj%
z2WFCG)sUeT;6~${Nm;5mPWM;VCW_jr=ML9Jg#18K<fHyQWt1p+Cc9{mPp27_)F*J<
zW`ar^6MUpnsuSo|m^()jNdZG}g=k!HC1jSU9N-fmI;z84C#K$nqC?y_r3v;q&FCy7
zM^Q;_69YRp(O>|-od&69IDc*!<$Hbw%U2LlZGtMD0VW4^P7>!`5>p4Dp5FTF#n=p+
zpx@ti9k!Z36=#k=J2c!FF0VPG_9w1=Uu#=**DrM1JD*psXV=0m7W`82!mvPTyUi}J
zQcZ%ULt)*E12=p$H2(Ii{?(mxH%`}odt{u#4P`XDyIirmK!WQL%xnPDIEU5~{o>?e
znr@XEw`$wX9`In<zlH`6bKrH_BhSS@*7F^Ppw<hlHh0=dU-=K8TOG^kk<=roT=r8p
z(Xy0C#s0y>)BTyFm975{(n7IE^NX=nHiiHQ6Ws#%;<FTpXf0Ch0St&+@aFN7c27PO
z$a&y$XxsO0fqvb~HhVWFNSXX#LzUX?tnBG9vT0LyR)kGm`Mx%fmakRJD--93%09=_
zMc+H_Dt7Zdo-ZA=bKX6hpDywreLEf-SLf7LpZ81lU2cv;kArAjqv%cRp}V;}+UaX$
z<vEV7aQi=9$9ZbW-TYw_tlp?XsC=4G=&D}OrfX^(-=Gl?^@OKCg_j>W2w7^FKcrB%
z3~tUR?kA~j_gj}(6Ti7WE|*_#Z(m!l(wJD;C$!T$>9Tw4So3|9ne}dM>w+A7ka>PS
zi+Vl_`=M(mz`0`9@bt31aJN6;dQ;o_TC3~p2|9ANf4&nm>9oP*u~BH(Wt-=iyar*^
z9MZ2f$ei8qV^;x*K&6F$aD&6xOAp!k&5=06IA;QXGM0dDhe!biR=3m)CZPwxnpMAc
zf+a+z2#svOTxK^G?h|2EaT%1ql~W0t46n}7prqM^VyHr1U4eB@Mi{nFr(vO8YWKmH
z+06s^d}>XyZYsETax%juEy(!%Mf~16BHU%+#%1ncm`V#ceu>=eO0$%boN8mro#BX9
zH_MoQW^{Ooh6&<ipS`1{(>Ty(1gu{+R4N6t@P<C{=O`LyxK%%7vSYV}La9_(#f}Q7
zUi@%&Omk=AoO?J*RB;n12mkU5qr3jebL3?njKw0hF=}cE+W&f-C01A<HMn#*1e!1a
zCy8lJtc91;m&BV8bf#vX6>a_LA=(3^#U0eq0?dwAMCrm#kzscykk07N@nE;MWoUlv
zilAZ=sjYWOUAoh*Xm_%CW;T!8vM2_*vcFb&+ahKP0i*<L?SS;hr?`aKILi`e>hZLh
z_^~FvF6l;*C1l7kf*%f!$t;92$Au|#MO*`M4%^Iil*80o2iqt;Lu||_w=$H3)W)_v
z+mT)xT4;=hlV)oB0u?NBCTiymG;Q2k1DCQ4o>dbZb$Y#6QNuXWBA^b9%HjOV{N&RG
zg_dyq${;_aA|9iq4H=2EzEU?%_;y-Ap*CmAv)VwVczYxZwjRaPrt=FPVlHRZTi*ZD
z{=`=}^~i+j=kbZd(~tcX(=8?=l%=~qVG@Z*+zf#TC4k@s8X7-|{gN~nB$tTbr{L!(
zqB1Pu(WK)d`de&au1#;g61r@inI;>rYW9)4i|@BlD;9rV92$K88kqo&G=>*cuDZnd
zNBEVolx^W)mtz}(Z(pQx18e*i6e`*-^uneCut7Xr(8-S}tHf+6lD>(2TEXCnoFVqS
zQ<_moo%|iFOdw=o1S*jolbckWo3y_<Q>R@lO8oPr+Nbj5;erL}k8igfe@>_i9=G6t
zn$Ln7fBglWqeb+I&i$Fs0)HmM>6(<YO5kMJ4vsSIwJFHCGw7<vxpo>uWQx8R$^&Em
zZ8Kbn$IXIMw(EwMIjx{iZ6NH@fprI1k7sjD*8}w*b8gqkgI6a#zafWJppa?HI+S_h
z*cXVLa_PD!<8m3az@wZ@2?25ZUvJzwwBYAGSihiRe+}j2BR`1z!vfFD_76#QRwgd?
z|7}ve<ExC*>MxYeXAN2TymB(SE0PFNWz#m-L^+NotEwveKVG?JR0zP$vXhZ3`1%Os
z_CRqJ<TcU!q(mi}`sU2p>Fa-WY=}|wcE1k*kqci%N7Ga(vHI2y{SlKQwO*qljpmH`
zn)v93iVfTMt2HD(jzmKpN?8+JvDMNCKfmXr`{$e*+u#_|_}U7*tp4ee9Kp`#qi4U4
z4rw*#=N+%7r=uP7=$EUV-#PSaoOK5#gA(Jip>OU38p;j*mp}G#c1A00A2csqwehRn
zF)@;hW3KDs#m=1M7-J9Fm3*3M@4vq{PH1~tEju<BEW9Xk47dp9o)a_~=m!#D=B+b2
z&>?=$(aN5pxp*e{J!!nwIFMI~G}rYzWANN%K#T54PI$b|3@6)G^a1;{pKjZP#!OJh
zDft_Bh6oy1ot(O)5rr70Jn_1kD#l~L`f>Rs<HIdJhl7UO{HJ>A^b-Rwn&o*UQ?FTq
z>vGVQXCu+I*hk_GtVJ&NoDP4r#)+Jl@g`SCe@8hMSP`6&M&y2@8y)mm6Ka<Oi19ju
zJ0aazk2ae__8N}cpKGq2R#_Ui^<@vHKbpk_b92x2ZI{nC!zxcKRhmq7r&Aq~il*ha
z56>kPc&}9LXm(`#aq6_^Il<@YzNdS?|8hbV+L5W;K>|3ggNdxiB#HIlaAP1HDRQ$I
zueAIf4lT`ZDtD^*xEt8P3KL$Mp1AYR+1j=grxZA2(Il=B=`%_ihOeJYMn})SZb|a(
zTfP6RDzn1&wP2F^>{FYv?cYFhau6pgE9XWqRhbiNXkG|sFs=8?wNurh-!OnBUg-*`
zWpUH)VmE^v)nwH~`fOtq?;4Cz?W67@voSWBeOhh}2dAlp>BI1_H%WlNnGq~~=@6<^
zlq#vbJ!crcrh6=3?{(EHQ?R{aTxaWKXMHc@U>{ysseWYf8=c8Qjy|vf%99O&Vwts+
z<+M%5?p0P{%fl<P>dQfc)&Z|wCuv+Uvp)4m8fYrlAQZ}vcNh4lM6qwkv0n~!9ZHXF
zn`wjk!U~ktxvU;1b9hg;6xUtQq|v^`6!`Glu-hPR)u-Ws%sGwaV(vl(GgsgdrzItL
zBdkUtQdaqPY&pO$%<NR_Ket#n-~LfUoO}WoIBn&Q&8x$|qx){4Y>&DCRE|pbyy!8_
z;v^u3vbdT86`~WLPdbb51jVWWv6!<j{uUNjz1HJkTuFMJgMYAayH$Mw@5bbm!h=Se
zHn12bC`k}y%ApHY<?xwpmB-jmP+-IP(blcdOsh&N$U}ukR$}89FI)Dtt_Is=!1ZCt
zwUzK18BY7@iCi7;9*&W`@-Gi0mD&QA1!MvlsW7qoPMFWJ{rxIhyoX<qF6VL>+4ww1
zd+u^$e+6@9AbI3cZKZuhix^hAHfTSxzYLs($o^?I6SZg;ZSA`-cl)PJ0FFn`A<tsb
z0)dn^cuxT=cOC(ZAV9$UHV3Zas?G7>jEW*}lUo?E5m5+1BzkN*U{4P^*=+r+b0~nG
zbJ7V1XYW3*xNw>~Q2AIm;%RNtBCr8360=1BZ4f%1C2`dL2q}ddV}^-Aj;o~8DzY!J
z0m&XDq{@LPLq<l%FqMOqWynEd)OmCb9^qn=*r&LX`|FmM7A%7XU+5rr^A1Zg2qi@y
zM5TaS$~H?`4>X@nw<J96&ZYg!sdOKK@*dcY!Es-&W0|3oskR#~sbw_-ES=$KIjlEP
zFtjSWITBHoSPraaif1x|1bJVnU-s5Am<Zu7GMGfOrYC_gr$^Ta^Tc}D6Z=2SAwRgl
zOsZz=Q6B}T!6%A!nlEV(gU^nBb%QC8XQ*f|(?O}rSY^ygab`H$F;`R%x4}Xf4s@6f
zNxSP}s>z)4VI;T6ECuaSM;+W=V5Zh%CWfcMR!5*+yakkdWT;@}b(8me>rti9AOWv5
zP5He+E1z?R>FN2@+0SxMDu=q9CFO4CW{kfR5lO8C!|lNTwLa!Iv{ao#INVQ(Hv``^
zhaj<Wu6+BDT5+TUqe6~zOp8WIFUJ=;jY#U<bxzQ#>;}a@*kvXPrz;mIwg#N(S8Q3M
z_vdu_!((b3D45OJ33~EH$;{-cf~g3g-Tj2bsh_T&pf|xoL|}a%rp?TJ9HqIN<s@C#
zI$N#nxWaWpOu9M^Zu#->wHkV*U??V(jWP1UTs*8DFOp2HU~wcm84c;7ct|zjl6|!N
z0teC%Z?RI*;mHZ;ZlCZm`WKcAob|oXBuhr+3`RZKwT^1w@)Kc$=4HQMmF52kAV0L%
zJR(1J7)T~}9TGT5|FrWr)WkK_S-5a`l4bMKx%Wa8&jH2etsdZe9?4IqGW=HHfY3m3
zJgn#2>1E_ACWmF~Qon2P`+pdFry#-Fpj)$S+qP}nW|wW-wr$(CZL7<+?W*aRb0%gY
z=8HK0zTda`u9dm+$r)@Z@IDB_EDPgt4;;oPs>^-Y66#5^iIO~nCDgvj;Yexpl0So&
zE^gy%;+BdCob(j^Yr>ur)%gq&<MR)`f|Okmt<|0R_4zg(AxH`2zl3&loD9~;WXU1Y
zyE*D+qfL9k-n^H5)gDz99={XHN$1W!#W|+I91Nbi*Fo;Vp*tSqr>xg^wHl{7hi#Wp
zAd!pW7g|ucCV^kPD}?v_{iWjDW}&2xt=ztrL}CLkOC1PwVVUE2`6fm>mq2elqaLJY
zIcqeY>fZkSX(@ShlzBsQ@N{kzQPxqGI??az>j;1V<%F`At4SjWgj#TjOi+l4#|VPo
zenP=VE`))QLcxy>+SP9|=mp87$mawmvtU_S$*n9(cz}f&FZv(1)R$3OdUV_Y6PE>+
zY|F)Ds;p_X!1S;0ZH~Lw-+w*-rQiC{x*WCEuq~5*?iVlp@{{3`J<0RMM2^XXp$=&I
zi!I_U+eize23J%L&cdkio>z3^uJO8dkH-oro#5Z&MhY5C#05T|o%f1C_9+x;3knE<
zY4|&({9PdaD&7v3;g~T<O@QMI5H**N%KA|<2TOZZ0`evEB`F#pf&fENZC#dFAq*rq
z6T3sC>OBBc-2s!|M`2ySm_s5j-q6c@M8@7wyHahS)Vgb1XB#H|v`UETEG_UM0uo7u
zBx~u#W~6jmSDL>SugbCxFRCYWQ(1p|{JGy(7X5@$(sREkLt?a@a_O=?T*(DDmK>_(
zMJYlnrr!^t%?J1HOBR!xq~kf+$I+X<idxsejIcee+I}R$mgkWE5V$Ylg?UM?G%v{a
zR+5pMMNjgpkn$T~(#hAPsd5<5iJ7|p^cdGwLBhSf5vc8@&nQ@`8Xp7TbcsbrM-M8T
zCe<GZ94j&WGpV%a$1rB5I=O4JaaYr3<7eQ+=a12Z?pO-Apob0;(jawbKP!oP_2mfQ
zg*^B2y*cO@$qkP=eQ(Zd`&rP8E>ot?%Oke8)P<^!eY~nJP0rT6gvh1tb;r_}d@mC0
z%w?sgP80?A`Pc$6+epG&BN{xKzlQ<t=IdR-L;lQg!k-2@qIwlM3s_aR1T9XKRI17G
z9Aeq;EE+s5R=m%mU#YaE4oF~-Mv(b$R!`42P;ck*Y*rW6YQKi#?z9HvezJn-e_=~y
zeL0)@LQkj-gf}gtI938w6{BnBq#dFfjXu5)n=P4fwf%j?m!GXmnB7pW@+PrK`3T=c
zcyuME%|w`$2}6jpF8z_%qtBa>0~SJ$w*J>*I=CHaSZT=cns;au_q&U6YdA(R2eVkJ
zAw5IHg-yVZaL+SE9%WB3EW2Bu9{3Rma5VaT?`Iy&&MX>eJI7733?hZZ-$k#+lj;l9
z)fQjk-R;Q?c-uYbCJ9#l3LV_Pbu+B0b#B*FYb|qfd^^fYT|EAqH8)3pN$sL>q``pE
z5D_uXDf<S=XDh;n&I3sTY#-NOB;v@$7i2%=d&-3A(8aujRE7eQjP4B~MgdUVj%(<G
z_WUtG1jJK#X{^OvtDY#=TpcEBQP&?gmr6Y~PtL(La3wb&h+P@q_LBlRP<spYdMh;E
zuL+YCPaur}EW?VT{#K5LvDGg&#CUjuAd5LlA)YN%teb~(IN~mhBc|Jas>#m)BAayK
zH_GTYR1%2RxB|Mhx=;qoSF*cC85BnBljY08>2@I~kwco8e=)hT{~pMwtMI<>!uqRn
zD2^-a!C(`N(B${)=Vj&>U{tB-`9JN&e`F4r*x1<rr}kn7OV(+#_2FOUfS7kq=uR?z
zNjWhM-Yu~exob&y1DJLJ2|IEbnUeeGb$bK|MRG|g@x;UtsT~0AGiIdr&6@SFX-z&?
zkJsZv>drZ}ktOy1>cX`2AY*~?aAUeQwFy2UI>W@w3HkBtVZCd2NB8UD!<r5@?#Ju?
zVr3+1cQ;oPS4XG2o8$~1UER+7Y4cMrug~ZG26LtlqCZM+S1<S5<@s?*{m*mUN~i-e
zu<DJ<h_d!mu$Pcoc8LX$FduTPNCHE_()tP*TH9seAH<(auMeBZnbxPnLd(Y$I;$>G
zKgS{H)F!8oY1>Dg)<8{vXrL8<LdC9Gi~8-H2$IP;k!Bh+8%1^40lN#_24at`LvfHl
zDF%Nk2nSw~fiM`#S%YAQWHm1*Xr<*9VY6AoFYK`-La|f(>o%2Da@h#Y0+~}F_QR27
z0EeeOhLX83PXDbE5>Oy+O2`l|$yAb5e_x_7vCUsDsaFU%DjAW`<xmeqOPV@WqysfP
z%h>C~;hzLL)sMq0Z=JhnZso`R9BI?Nm}nj8in@t*!AVuuO?d;i)<gIg?yw@WK#!Ur
zR2qQsIpZ)xdUFi?IG8$%m>Z#*pneA~a8g&a&3Xeg)?~4bG0#><&+iCCi(aa;KS7hJ
zB{=|U=5K$(X>PNC&R?dj;F-9IXk_?pq=pZOd4wfQ8}<hu8<xfxMzELT*+4BbscVbd
zU3B5t;1fvHtS($0d~_BVIK{VumGYuM0OJ2WlKybUgIJpL^Yf96ul|E{c4}5f&%0V}
zz4q+e;%Xdt{;I8I#Q<m(F<MHS-9e3aT@`Iy05TzT-$0{3x{$tf1v_JVR<%yQ+B{kT
zO+f5I*c&7fjqJ~~;L;lku`sV@W@zCI42L*ChH;{*wF7SP`lT?H96-m#57Mff&mQ!*
zLEys~$a0b=yAZ~I$2hCd`mz#&`PMRAm*tgcGlf;~GgBfAOQIWW@vM=QMAEr!wu(|R
zD!qiTd&?L)tbaJCPa_dU;6Yxn8sI4qq@Eej|BV1s{8~pAA-Y43C(3&h$y}NKF)@`k
z%Cx}_i)${?*`7V)8kr<s7$f~!t;zOl$AgznYnk*e*Q~3s2Vb0jhBYeXel)2|fHxM0
zE}ScCc&dQrRTjEU7!P2D;!fR#l|D891PC~)O)#3xV}$?|MSQZ;1VxM>|GDtP8jevL
zXj-&!PsUvs6csOy_g+SGf&)ZjEKxi^`U=56gDQ(lhoIdI<}G5>zwaoP!?F6Xzl#`-
zJwq?mrj0-Yt0;*Rh4cF)?Cg&}2c8H9!AkM86h6^8PC-!@j*dpeo)lg*dKq^A^Y;8#
zFBb$Cjgx7RF;%|tF48<BbRX}(5nMU<VEU52wBmhh`71n~R0-|Upw{Hi2KJOkJJTSo
z*2?~lPO}$&&OY_<7RlsTgb=#0r$-RHq7XzUfKL59ui;*-GJ%x%473ED07AHcmc6dq
zV-R+oA~)Tw00U=**em4*5w*n(EeVBY$6S_?WJEF~iI$_grAPZVt#sn@wLi7Iu-pKP
zuDAI>I=2g%6`=yO-w3qRIC?3=K+h5>B;A5%xUp?Uy!|UV4I|$t%F@k31l$c0goWq$
z#Jl1S>+tdNSA2Kbc-P8e8E={USg^<jA^8oSJX~8V?iQ*f!<c9;eU?5SJdZF1LS-BW
zD25Z-AO#Ov@VS{Q<xwVCeQO302GU5o^rf(8@0N+1&K33rWvDqkfX1l%1*wA&uVpnJ
zNn;(pzPHFB1}c9=Ua|#cnYA($t3()VJrI-e#<;)YJ>|aO26ok~eQVN0OTbpAIA&fV
z#?fx8u(I4dUQhtgIgo^WYyupa`JtxW`mXZWz#|cB(IKAYsBe$ya#7~KcOo~0nL*6k
zpI2l`W~un!h_P?@Tv8%MX76lw%pCsGL^-h;6j=h%tbQ)SCkxDf*(tG?;^^Vx_>bH6
zzy9H#B6;%btW3KeU*D(cE$NB-+7#}fV}s|RVHjT}2_L^$P@@GO4GQy^Q*p>p)vIF?
z@CEr=FLlBLZcEWzIRX@;1H6nPQFceeCu4BV!uY92LnmHu^~?R~_k7j8ssdkF(wLg=
z(=dNuyzTegoM%lF%Y#S2lf0IX50dC!iH}6#b>v=~3dMD9#*S{x)#e@8XQ6r%lwsk3
zqgaCsFUl*4OL&q6d@F7l4%rqC?n+s>c+_H~hb^27V`*VD@X&>?^pdS(tF{A-5rdpd
zFb&0O_z0Bph>&($s3iI0Bp6*_`NWEsRhaPgT|u?;>pFI=A6(UG6pT(my0p-t-6QMH
zsjER~zk2jcYRqVOdcZ_cB2EZI*Ea06l7!!BU`MEF2yU=}Xv>w-+6`fWQe9Gzja6C)
zs6gbmxfL(Z3JZ^moU}{+fH6xJ>ZsN59Ox>d_QuaVgCG;h66|dD2X9grQT;CCE#<Ay
z{lK#M^~5|=NYN1B9Xul?=^8F-W&lIh(P7p^tz<&+EpjS!O+`jRg)H`&a1%zDH|#2Q
zR9IS^q(U*;HuiFEkTTihQhtJF<FjrB(A^mjhovw0$0{c2|HZcDTp^YubQF}pyZV<q
zNx(oWvs9Q#RtpZyI$pS;iFiR3{B5I0!_NV$zB7df)*EUMaToI^n`0MkDGUh*Y>8)e
zIm6JPx!gI@<%TXp_Ud#&4+;n%hLhkkh<y=sWs~4!iF~fV;!heI0weCDir}xv;UiDl
zXs{2{+(mSkTTwg2Qj^lx10cd$aWtxoI=IS5ie&EnfKs32L-F<b>L=tFv{zN9|Aw5J
z-pVuW7TsHvl{-`H<5)7q|5n5V%tWf~2xBh*0Dbt|CKM1{67>fQV}OXV)m{q2n?!@Y
z)&d!e$)T^^vRl~b427z*Fdotht~}tP`Uyg(A#1XKV;2Cx=C<wc5i!9eP{0v*0gbg6
zj(IzF4E^;pv<Ks}vo>HIay~!N<)47Ng%F0A7-al>sok6YkraIQSlU(*_Vrmi&h|t-
z)ulT6>LPjBah-sYu-K9Mq7v)11*Nc9G=YjDHwx`a!F{A&?JfZIcIZj^dr^CT+a?{b
zH2dC1=3y%&cU)DDdrcytq65LPvNh1<1v6)tP#U1Xej*s)s4;H7a0O;i2Yn!eKIrrp
zM8m|VlkKQ@m$T(pV{zPpX6i0O@%cKMugMs@6Nh-KwON3-_I2U2O}Wi;!E`p0w`hmP
zV*Net2T$`%F%QMdAUFe<P?;E&zLGkLx7m**;u<cDeHB3;&%eVaC=_&+6}8ehLsnF{
z#(&NL)@y0eKA9e2F?PRr%Iaq2#iSY+9(sZ12aV2uCyM)Ql)DT&v6FtaqD7lQi?@Zk
zA-5}gwG@AM%uSK<e35c)0+)#XpthkPxYOf7a@Yc2Y}%QJe#{do5N{=TbQG_f#O5XM
z_2WDzyc;;A2wy;r&8utHnfE5Sxz|V2q0;Ve6T?w5zaXi%G$r_EF>ZZw?AU?#NCBQ-
z<>XCHc5ZgO<Vc23ncQ%v^OAjh&)HtL+BS@Y6Jj8qqLzSGeF|oAeD!vFKEMl}V0r(S
zcxC-B8Wj`A|0cHmpMh;P8k$ar?TEg4dIdbiLMwkTw7XlW;JcF6aqUcEv8=4{m?+U{
z1eut5R7gzfetkL@WFr$zxSLYQ+5E+d7RJ_KC)Zuc%=&siJl-#!=~t&66UezTXxW*Q
zrVX<oEDqNlc|X;Sle2%l-ZwAxlBdb7spNuZ4H1yqce+2%mmaQ9--X7Q79-N>4@AWB
z`24=!^XztiS>$b+yS;_0^7wQCPmjmu)G1z5jAiAP6@)C))0kPA8zhuiqg2U~XUwLc
zW<RNn*3m9&Zcm_#R8^LrXtQV$yGk#8v{-EP;}<F|Hanz<cds)&WqEvrce&M5eQOKD
zkPW&@UtRRt6GMAn2#BIu;q5#B(MoW2;JRzHo<}D2_DU4IGT<(P1Wjm|-oRGAbHaEx
zOc#6>c9D+HmO4GRRV2z3S5)lZdJ9uzmlDyNWlGf!m{oEkN%%O+i*fZ=Rfa2|u#pN3
zd9^4KMle0(Vzk9a@mJQQ25+u3SkB(Fxi%yUls<7UiiO+H)0P{QNG>!UK-bIro<*#+
z+UWY-SuEL%=N&kQA&`us=j+b`Z;Qv9f-bFXMXFhGn{6Vwu4IL-@>SY?=6WwJY=sjO
z_sf%u=;FN3KME^6phtIuX(KudH{>~Qpu<NSiv_vo3Vpt_J7QRbPF^(D_;0@}099xl
zk@2N{2i`|D68zHGDq7^Z{VHQ*2hPbiP*~UN*8{ktc9ef)andOZOvI)yiHz&@3Tn6q
ziPhN#Dg+XR{UO5d1X$h~jE*S8c211t{fUW!<d#}2iU;C}5W-vP1W&^{l&FxL-?e4w
z;ZG)lLp${I;AakAMf>bDy-zE3BA{|)F?kyTW*962A!|L=e`&ROgAUnk^<G%Y%LbW|
zZnUrnwtU}wJVN^21eHOj^f`blEg=j%e1pV3sY-Mm=mv`q>wb;Y*=RQtnbHiLfIccz
zn62`skFrmZsWUm?q+E6m`^9cE?RQEYS!+URhEnu=IOQxWXoB2;_F~ivL1+6F<n-n3
zac0$FO-SX4;`g>VZ3ad+1d1~Ycm`TNfnIKchDkO$&?bd3hYjV71g<%1>&Npc<KyJz
z9D=I6Z?!VR!neFqTfl!3+aq&Jsa|a~T$%@aRl;M*q7oh(A`WHds+EW&+iQ2wA-KsR
zim;c78!(9RV46zT6|j(-p#$-hsMp@%=EiC9Y6@dtP`iEOB?Kh9Zo$rHc%_S@fEjB#
z8F*B-od^r}lTLxQgg)RStY|JeU(Uj@6bg<wX*y=`R_X`Ljomctp9rfA+{ZaS94|dH
zdtJJ6#QO<4nscQ2(*G`r48ZW-c;OI|Nf*={u#lLK!%*CV59^8sdFpKkX9IIB)4rwd
zT(b>;0J7?XHR`V+V#b5@7PS{C0U0szE3r>IMCE4Ogj~5}#Nl+Nc^A4_4$Pmc*hM9r
zZraO@9lIUB;HHfwV`Z{X;l9UG-@5Z^U`;+0U@hU=ZOJMUk;z#dJII1Z4>;#9&53}z
zMv*NHQhA-Br6eOU3yX`68?hE=ilN`ZkdnyPeZze+YeVFQRLs7L1g|2ZB@fTb1FokN
zgN&(g7+v{uZHP>10fD*z-1TE>yO3EMnwPde2%&I+fS@LUQ5EonIH<F|K5C2oO`|ez
zvkRK77uW?^D$od+xp5cvXATjVxJ32RKYb4e-YMnkhY&E`ej2_BKQ-+;Hqw~(k5y<Z
zV>DJuFpWM0>B6eLnxst*_(2=OzY=^#jQ;B;qYpdC$MN=O50b3&Dxyoj6s`j=9PVPD
z=xhg-y`;>7YBe(oYawesBg7S(wL-jxd^>NUq5gc{0tYi3o)GvOly&J!leZl|0^9HM
zru~i@%JZ$)>o@dPVaY9uAuK)&80teKE|lv|5p&&0HzxwcB){Z9tqeGuS{QPgLTpcZ
z|GLw5*Kyedryer^ZRRd50h^*+s}>dkDkz3}^01}=1VNIOi5s@mebL$ty^J<9LCH83
zG24o$c>1p;>~->CC`|KyhRc2{D+Cw{oKYjav)Jp-?dUh)o$xMe0P)qNw%=yY--dxr
z)ojMr+2(Mg+Sl81G0<Sp-xwbR7`WGS^FVu@5HV|g$%4{HCVr>@E@!h0u4drr(2$rk
zg7))(+5>VtaLnWhVjq5c;JPQ6TsJahD(e6>9F8~1-&gbvsT`=9=>~0pb@9^12LCON
z6RobD`<^|?6}K|P0%gkP{oG(j;%0&~%-WEZ4(TB__h@)_caC}-UEBr;JGiXJEF*W#
z0adb6|5InYIWlX#lKo$Tzp-*OK;a3|<f}8<!kMbIJXZQSkXg}Sqm|C1qikQt`gk<-
zRFYwC8x=#=^^1bJo{~Fpr`4?rdqpf!arpmS860bb9;f=hD^sT4Sj6{%WOVZOh#Eo%
zGQT%CKMxcsb4-LuJSA&7Kz2<bff`b8@ckw|_ce9ucv%dRah4<$dhwN+E%bgQSOH-0
zy*D3#gM$B9SWLeAz794myNFJ^EW6=Cn^p<B^Ab_U#Sg%!Fj!zYdIOtvhomk%+~YSB
zmt|1eF`_>WP+}UX7GfsT<uLz(qJchYpl2+oM;k)Iv;S1rD<CVNmC4i3%!*cc6^(|=
zH@qLao(9gaAE*#;0&z+FhyAx}Q^sN)H=a)~-p1C5{npdQ@y!QUIcX6nf&Z@{)eeMC
zy`IEOk<2gy)~Y{iZarNrOlIPdVrHv3qy&-DpBQDKTqhA9*&632(V4+(!U4hL{Uv}O
zGP6^mAqnO(z!JZHIHqW4j0}ThGfF<Sc$E(iQYrQA>KwzMD*Ke7_jlvsAhsd%0&%si
zbQq}Mre=jJO><PET#$TjGhQWurWv%|0zpUbbA!X*TEMJ)>eiIQd4(cP_yz_X2ztNR
zB${^M)!7JWIASBqMAWBn+AfgX<09#0&M1}7cOp{22@;eAw5SM^BNiFv_HV=Hkb#A!
zPQe9<Q3TG2C2@Nm|ARchLwYCLf|1V`KY)X>1Z5~{%y9o^@COE~MDhY@k~xDlV%c3f
zy>E2%X#(#5l1^+4|DW98{{!-5_$R3T@9+OComw=u<2E_q{POw+>A^|<{#hv_)p3B|
zzG-F;liSRiFu>ms-IvS8EQ-x(%6W+#`ni>kNaSbAcrt1j7p90HqKy1XV3M^(&g|y+
z^l*83<1Aup7&hgRR?8AGj!48TP$I2zd2k(IE}URe#njQ}@9}<oI9Q1?nMBq3a=ia~
zA8GpS*0$`>(8<<2ysAU5@*{daxp<zVpQFz^_-Nwrxc|8DeBZp#i_4Q{RHa5OpaQPy
zRvSn+;wl<Ys83uUM#$8>{<}1u=O`OJW54FoyRB;C(<Q?!h3>QDEpmaYnv7I|zmFin
zm3UZP-#>1@jLI(Cb2ErV?8%0U`s8suEdxWY&xkP;u{hw6CaO!XEEsVAu>SncWCAvR
z>H95xZ;>IzuHXk;CqYnMgaLlB^7%MpD#d>@cy{rvv*TrBp)!WCKbqfZ`TFzC#47iv
z()cfN1g+BKP&lIEmxu*MNu)VOqrT9E=n{cSkpk&p<q6pA0UWKz)&NA#HDu`WZ57eK
z`N7F5=FuxdNsyA<^e~7RO0>Iu3<1how=*ar!byeVK!Oam#~-g0JN{~?ZV|sS>q5>F
zhhoUAP9TtcY6WO`E+#~ITBc8bUEURF>=ZXLQkkET$f1C)2iGimWg8%AmCOJoM6q~w
zLKvJh&>Tu!J4ULOL9i03$R(+^PZP_90UB(&B^ec_I;LDMR(or07I)eNl&!z+p3?8y
zHB+9ffROf83}`Zp!CIlE-L=rcsh3&ZP)AwaGi{=J&2V0L`&%QfYjhc!EZpg0wXN10
zn3FugY>G8xJkaakRy($@U55g~nAi*<One1pk5`2C0iu-hgEIp**`d6XV!=h(cx4&9
z1~GiFzJHK2*6RZB`N4eoKu}_q-3Iklyz#+&*rFLALJht?_X>v_5rI)p@P}|2Y4Ama
z`zAq{8I@R@PPzaRJj)M*EU+Fyl>1jO(Uds%MEpkuEW`_<3AHMIJX^7NII#mbNQ~_v
zu?ZxnAKV*F5?ksi@%}JOJXu&rBaE0KjDUqR5l;UFP$1GDgRmqMLcJ(sTLr8gF%ac4
zr`1mll4LQc?by#mh8%HfdENu$JF;HbMUAyr>>0{vghp;3L~~ItJMJw1e84T~);PRs
z)kH^i^ridks&@lpwf#pG$tv9@I-K2?3xoE+1*v)Dd+jwsR%X776ecjEvg87a3`~q3
z={=tsFCfIuI8_K93PG?ifsimhkRh2C%z7LJdt<Ijgxut&3gqTw(LnYS0*b-9G@Y0!
z+&)oFo-9-92*5#2SSM7{FH%lOw5KIjX5HF%4~}LyhBr`wpmJ5w5euIF_rLtu@&H)v
zGW+=wlY0<;1P#OHxbfiAS-}u&jE&o}wI~fMGI>MMWO{Q#7y%FyVZ@20E+V~UwH==`
zZUgPwr3RLc_8b~_d9uVi!Z0kxO6)YSA7)4=h{JI0Mpy~LoVeU9>YL&Nu=rdMGqH9j
zTdqJ_m^FS}sF@hdo9Cb_6VH1d%v>1nZ6w*$Q{X}o5T>*G*QebBKv6pZR5yzu$`W9N
z|J(eN&tZMA34}638h9yia9~EbLO%|I<s_dtRr-Be$MRr{tWfy`9!#9#5h4XvLli@}
z)J-G+tb!L-#xkWb6A3a;Mm?|lbNt)tm&<rFuZ|9ev)0AkmN;PmEH{b3iuX_)@$fPV
zm^^>wY2rR6eToK8jkc{;$m?H19cQ18na_jdmTbW;5Ozl757az+x`7$%UT`dqIus*2
zKQDEk0Z4B-4yeOfMb$^lh^f*WXp+2Rn-%>k!Ri*H69Hy04#a;;^}-%AJ&CMBstbja
z_XhoF_%A;GTwh((cDME>YtDC}_V4}@9`akAZz?A~>&xT2*7jI(_r{@(^ZL!wO@G1j
zzmwHJlVZ`ZvPqA=ST>6s0Aai-=zcjI(V4+D-I&20&kZ&QWWXS)>{(Z#-SCeM3>s>k
zUP<+DDv;<ZLzMBBW4kL00d7YM)04Ae^Y0Z#$VTU=H{XA^q|45JZtTZ<>?m&GowbKZ
zapNSY4iQnEx#EeW5UPO3Yg1F{GYKjK7gyUn^l)(6t~!18pCqvlk)h4qeL5DvfUTp5
zcUkBPy0Quz-9&N2OJ5ppd9;u4d0`h3N0&6P(lfSVgGVthve>;4D4FjH)m1vevatQ=
z?smN<5-d;L`iPJEa2{Yyr#`?s9^m9boh}d9u3)bCmT@2L8?(|Ol2q@s*LSMyBFzDS
zMhj-rz~&mF;9bLk+?UTrCO>(mM#nGgyYCrOJ=Swh2RuQ{Ud00pFg3=!w8J7i|3(gX
zm(B63hz=)L4DGMW6Jquh^lzo~bdjE|yQwa5dsRTa3}z!QyYj*r;-l^$^YpTC@Ijs^
zGM>V}?z?)sI!tf^?C%+sdE2k|e3jHv!bo5rR@?qN063_=zJ1k2H+sa<b6*AV1@Sa-
zZNoILPCuL2mU~@VL2wn-E$T<{>(jr<{#}g3oy}}t`6Bx|(1-lfk979YpmSQ)^qUKd
zhj?QoO|<WmM9Qw^N0T7PyF#Xj!b6(WD#2#<`=f|_I7SSl1I<HI?UdXv+mumt*OS8x
zS#QH^&Ufe46`<%XIPA99W2cyqCroS|c60wb(59yjD*>gGVV@_xwRV^f7yq%>`8AfK
zF5jy|mm@vB$@c4r0c5B5{c;Vo5Zv0|dp+MwFF&{hUP}jE30nE{JqNNeK;HibU+@Hu
z>9z~{pTgol@`{Ws4F6*Y(6xq^-6k8#KRD!kQM4${ZDPHPfL60sRLwQcZ3QhDS*lGz
zr+a5HZ0L2{ZFtxN$%teQ%io83^eCRAj=MW;zD!T-I8XlRTspHjU5Wh>l2vW5sTT6)
zvSKn7D!Yz{o%g`|_pswCae9mx@XhqaU~#D6zdCci-Vd(t-i+|=g?wwqcoeeVZBgv<
zex5&%Ngfp6$5UqQA1^bMSKjZY4jU6Wie`|c4&$g0+1^z#Zrh6owc2xOO}l0=%Sv$`
zLI{(4$WllnTikPIxH{&bpTuz!>Gps?$acJ+$F*uVbK@ZM9x(h;1WVzK_UwZhPegYu
zXx}LelFlfFO>U)39*ZQZyblE0h*99yRAex6d$worTVBDSx&*{Tnn$3~$bhLDKU9qC
z5;?DQ^T{6>6LV^zK3?7Kp4bClaDgK&f#0{WscbzU5Em4&UUsI(WyPHfpNHRZIl52{
zA^eEKJ;alPSdwLGe*A5C6-%<V)%@K8(uK=Mkw4v!`()~iDy8LXCMo39%ihMrjC2wd
z%Q`}_0Y4lnV<BQYh3I6ar?h`Q-T9(rvb~x6kW~6*B+G9%!{|O998-+hJ*Djjhun{7
z==Y*2{<1cnBiZ~y^Mh_hP3$O(KE4<uU<erK@vW^T*GZQ7<0|S2qb@kSE{G_f&^pbx
zZ;>h7X<n^OcxYC4Ivi!KREk~HN=;41WFe_%<$~szg_&u3{2-<g66sK5iRUl}3X`TO
z5R0ol3F^PsGUtT>Wv>5ghEK3lMqx%2h~p`Y+NqI&1d(Kz3&RihfD_rz;rJ#+HCx-v
zUZV;#BPa!S5IX)*PpIqI{<JEH4$!gQ5Gp{rxbwv-xA`Z|f&A_u5<@#Cd8R+8DIZl$
z-j2|=)abj24)REHmTUQ(@!Hp4wFH5rK=LO6Ho(ddxqzm?_J&QfHgGidS!n4duZSuf
zOz)J-M#!8#3*^$tXrZ&Oxz)mW<>Osf**GYq+9p=>!xDIK=V3-vzbIKEI6v#|$>D_u
zKUZ6sVmd%yI#6Ne1+ts_n$6(?{MX#VvP&zck6A`{dw%S@1z7gEydB@^_oBd`A?*FH
zdhdRJ8NMmfeen^LdECP(!7!Stk^u6Go$JpfR|`Z)3p54l!oNudEIMu`QVG6}nkL*b
zv|4<t9JWTp$HJRllOMpsECRC2VkqMWA&HkAN@mc~)F(EYGY&D2ijm7Gijob7@CQs#
z9pG;w8E)*i$d!+;YY&Zlp@^*MtK{SLL;fmm=Ky!^48&~teP2E}*CWE2Ve8n@KRP{P
zg2Mt$gP2trhjE*qa!0WjP{U4_=+lL0p97ULr!CN7jPoF{IP+?W@l93_5V9h{D0NDn
zt!+Li2RtqN+z0B{P(%^vkcuX`y3O}+cbRzMv)-JIW}zL0_%f|xf8Zxc8<%6-i*Fa7
z%KdI!&_7ZJ$C4(*zl8U+NnSIEth05so->Z_xJHR>>Fby(sP2J{b9`qIC@!0Lf?!_N
zpLxx$xTAvv)Uf4XVlCo8S(hVz{dKSPs&A44oFRex%+}d?LNSOxhFLVFFu&hiWeM9Q
z^!k<&CIlE}JmRKrVxz~=WlGm^Mp~QR&4fOMbzj#ZS(VS)`HQ5j`Yac;)gWeH8Rm!g
zf{rL?Y{A(`u&ITMgv!)~FA#9B_XmkG3<;k142UqF%gr4rGD&6Q`QrzdoslrKjcD^V
zA}cv`5w%?yo?CWM!6Qi4f)O$uI01Hfzo`a3R$veq?P4I!hyp7ZL53t*j4E`_w!o*o
zXwrJC*!+>%i<+TzL^0SmHGO+St<N!t+9m1AQd`D~WZ$!E-88M8Al^0(a<^2Sos&r6
z3b9Ke@vd0m(01Jq5b_t6&7JhWD>}#j-zqXPa<ctTtzwMYKP&|s!oLLiT_UErn9$vk
z5f?8>?RB{kdxl01-US{7WFp7DFexSXY~3vov=m86C$B*Rh-=@7Z<jQhH89$ij*qr>
zPv0>OBvNKD5e!<EKxlv(p@|fDl*=7CB4Ny#ajCEv9zTzFN8Yb1dj{mpIQ`@1=KJ&Q
zYm!95$kz#Ci{CVQp2Ek&^W)fORv7%-cc=0!zn@R`SIgV;cTJ*Co0M{?u)uM*6y0?J
z^+^C&G+6^s0w@oJ!BC;VW4dDsNE8sAafwU5L>Ds!%F4zOg{hTt>pHU0V6|eC_Xz9A
z@ksHm!p54WsyvZw>5ss<o&8998{3PD+VRy^$fJCipgq37lsV2=eSf79%$_6!FPAK)
zhzE=&c@Z`=xWLG;H5m^Xc9Uh3S)JvgVV;}g%aF8}vR8_xDjicwye#W}@n|vx+Kc1j
z04h@f*Q)UCNC?RbVi0y4w)P~}lb%-pib2<uo+9CUatAorIxho&Ro5F>9JDSiW+EZF
z3XWn;a+qn}4yghXpDcreV>X9JVK`4wI0?OG*CSH>(7rPpo2pLnPl4b*#1hOpQ3OC9
z8U=nqGz)5@yvAyQ0lLL7!l8Uo(apPz)A%MBDy60UlCfxN8LKV8P02D8rZVEFP6sP`
zVuc0D)J=c*xIWe-E0n21Akn`2N%lai`N$chvjp;alq=tKxNlW*X;D#x@7juy)x{}`
z_;cT|SFt0BIZYyA^cMziJjKC{kVFR>0>0x~=oo#sBEJCQP?QBL%QXZ%L0-BkGa+VH
zXus3nd&&c!bwSyO;Phre(N$gJSS1#ePP^oUN|8TIEOR5t&{kM3W+2>)ISEd=F3x7W
zolT6}8D*d8usBe8<i8_e94utOFzA#k@h}!Q%Ii1rD}4KKd}@E3EqeuCgmYUlLdp7r
zf_{~3c1Qxjh(tUg23}bPMEfyQ7*AiMfP7?}L+&1rTir+mN%koR3p973J<ANmgehtV
zC^va_0ARKT5a;`rv_WuyEslRdXhK-In<vNIOZ<l~6ni~+!Og85`K@Ua{5{h+rc+qY
z>F^~x>aZyExShAz3~sqj3{;8N!1P&Q1B{sh9FZmMP^3sBN}RWwU5O<=l?q(-0yw2{
zF~vya)pr$OVH~KZ2h9!(*E1cDX*+anYpxnwXC`SaH)j<Se8Epp-HUlpq`Cj}D8i?o
z!#QA#YbcHQe4mhQ+u4KIe)HJQ^bS;*Nw0tTj1#z)FWB<oZLiCERYJGdvwgarPT}$8
zE&}Z`VO$?`X}O@nP+>P?SB~HjzU2iDmm01boEpG;4O5zJk<Y%tF(k<hb>ZV!%U(2Z
z7Z7T-S77fGcGK`wd+}89U;S(BgGLb$tuV2JQhB2{=2cE={ew#Hh)F2j)=B6oYGfYL
z3oZ{;8)4`n8tST-7-ey|athuw9ybai{-%5(pdTqi9X8cWPJ;qQhC#K5AGrKEMUAw{
z!L-YNrh4?rQ!lGTz?~9}JO1vh=h#=*5`9<(jQ`r!-YwSjJ)3)}eJ*RS_Lp)|XGlN2
z`1`W7`U|fT(!kn*n?cc7>S3h<)8*Z^LsR<$m~jEU`(JvNo%uhiA15cn|G4^TNIM;}
zA^od<_VNW*w7~#xgQTh%+d0c8@s%#xUKvLK3H8A+_)w6sW1p980k%pgq;bl%On4wz
z-(Q_tlFs3YXJuW+%)aiOpFc$hi1g`unpnKN;!JUiN)0lo3(B}Yk)i`PetsVw9Y_wT
z_K-^?2Z{*|EXb$OwA$`0<h*CA9L#@zEzCY|e90G?R&`>M{BT^{7&vb8zBPt_drl)F
zpY`2-81ua)`RO%?I2PAFN27n1U-6YODN7v_yhf2rCo|~jgJm#Jfo{Jpnoo(O4;{F2
zJ_cfZQ%11Zt+=-ynb`i}n$9bGTekt#1H?I^&CvF5JY*hXu6)TodObzijpEW)Z%Z6d
zMer5HQc2Du^#(Kxq>j+@WDe-_5tGrIix0l3ZS&r3bnm0CskSU-Ps1NcT*ZU(vT|MJ
z_Eq}Uyf!t&kDOIyub2G=HlBfRdNka+1&wWRG8&W~C}dHjr%*6|HPyd&3b7JJVMzAL
zgKkDoea-28pnT$)Qvw=-C_QQmbK+&B8rlhWIS$Hj<r7JKxiWnjX4N_g9RiC$4P06_
zb}ug;4bI5lysvrnVI5MbPj|>7Q!+$4e7kp3v`6rtsxX7xO8`E;&vLe^rs=Q4!?G)4
zbNp7Z$Nc7<y^B8kK~C3g=z<fYt6;Hf77l|$eZ|2NnjF18_kGJ8G5c_AT(lc-IJPia
zCB78uWvIHlV$O4}1D#xN0-AgAbJo~v|G=_^`I%k6e)9M0nIRA2Pc@ozh)jJgd;*RG
z$D#3z6En>^k`hC{j4!~xjoP(7JUR(TR$vHBVFL7GBRXyLvcZl*5Xb`~a-Iiy5c~^Q
z6V-*mro+0U1h$nLwY<|xikO8LkM7rLy8k-}UoY!1VQ$mu(DBjgoLI8cd6_kg`6mrP
zta=#`uYe?^qOEHjd6*y7O@FSU$xluYdXGqbwRUF0xj$<NdQMp#u@M+`lduaZSM$D$
z!x8EvB@Kor*Tt#TZ9=JFb@c~S9&((W$DlE|JlJS&fi%vw%E_<mz1LVX6A*fT{7^=!
zKgguwOfa|X68!f^PNgIpvi}TPk>@SQ=_@N)FdsNGTaZr=GoXy7Ec+<d)l~wx+<jjH
z{By*a+HS<6#KKO#R}vF446)jXVr2DY{gxN|3uXGmN4Rg1JC)+S-Ce+4@8R+J7}_3y
z5@~=W9Cf(;08p6->|4=~(P^^t1fsR7msi%^1S~QM(>rut$7B=$ov=yIu!%xBJT$s~
zoqrUxzkw9Czc3F+PG~mxeyn!1Oway1$zmg4KK(b<IKtl=hIkBDT8|}RwG04=q2^jc
z_AvYn&<X@99l!-O%e2=0YiYCxV@U^;c(97X9m8&rIvhkWmMx(}gA}iiml#7suCl*)
zx2C-I!70z<yXR5~rmqv*`WT=csxbPlcyUM{xSx_W4Va-RUmeoIEgMc@V!AaHo^XB0
z{YuXsiRCbEmJY_Nh8G7{E<uMLC@sHIO1`LI%p8Qws(9~1YpYZN6)k!5&QWnV1v4A&
zAF-}|$rdZn=$$?WZ;Swe2B^*cAozXLF2H14|8+@e5fEB_%;M=@v6SG=&AIJzn|ngb
zam2mD;6Bd6`F1Y~#QGquZA+2co2?vF15Qx%LPNB~Kd^0_LI*kajpP|sP>}XPFo#|J
z*plpaU|^acz`me+b!fJHQ(0Iqk8z|01nUUhjHdI&2!sZoep3T1H$>k;N}_QDC}C`4
zbpS~+_U)qWV&j827jS}5PaTXO<VVzX2I88ymk>02x-~4W6*@gAnBAV7;MKk(pAB&1
zYMG(MmSsl!qBr+sk$w*w?F-(C_X;LuG*#?w^EBej-QBm0p|ntgPPCz22Zev9et0HT
zvZfK4J|DrWDgrtjZoWv*0m%^|di7I``!rX^It!VD3GvTh@-q2c4uEc!bnLd;J5LY^
zuP~CmaD>V3(~6lAl?x4FH=VQBl-fR#u#Ug2dS|$6ckq?m_bB+`Q)$jy8E$^FWpq{-
z^j)r73)K5$mVz-2H6>e~Y3s`qnBO6+93eEIba`->F)+crQPSuH)&%S+`JmI*qSvx+
zV1ns0Ac3taJZdsUBPU1MY{CODUH)xiq{NJ=?H8Jd_U%bG9!2~~!FpeTXRqovR>H1=
zv+GG)ux&wp;T3V@h|wdf{~B=ZIPu#9Sf4lXm4?z|3K}Qtp1^5I88ffS<qNbTERi9K
zcKf5~MqsSW3ANv7CrgJaml-Bb$|d&4CM1cxMndW(H2iL19<UGL>e%b!y`hdnS~bm^
zw&d@f4Vu)Qd;LwMX`^bO6amF5InS`L;ZQ$CzvgG@xgG+GBkLvtdaDK5u0?13gn{s(
zeY5HO-ckH%+_K9c-8Ttv6DIHjsRiotE8X6EU0E0Ztr}o_kgy<9-A0#uAGspzCZC_A
zgg*Vg0`|R1ApA%0sz~3?H@KfX@Qkqxs_487ln)2V@$~75^U9D?Rh1B^dqF>qK|~;S
zb_!=6BCrorTw23W!v*&|D3G39Kt`A++6U{l`&>yO+Y|tM?}n@8y^ImRFs#SbC)QeT
zs#T8~e*H+19^@CfY#&E(2)*y%QDyU1zc|%wUmq<$nzm_G+sB(wy-afaC6kyvOw7`%
z$}jjNzXivC>Vp4rMa-Q4W3a|G=7jC0*i&s?deQQq!*xkAS;aDY7<BW@O0JdXfK-Yo
z0Q8hXYZB4NYoDfgGRi*<#^|&-dMrcxS#5PFf@|UVO#_?fdvCa;fKc}N=|W4O2T+m)
z42Vdwr)P%aJkxjPG^m(za{SRiY&;p+U*252IoZFyOb^5Mo<CEcyJBfed9k~YJ0c_`
zWJ-XzL1D@j0bs)kVj$prR}h8a(pQX;B+rbffSIKLe#kSE-54>LB}3i@YpNQ-7Aj&R
z8;eCO0(=N6+Q`Ll*e3mack@E(bVW>8`;7yjDL{B!uY-mn-v-9#b5VC}O`8~F?y=Xt
ziuafs$2DXo^uw2p;dN{#*<D_3Q{Fc94Y4e&pBn%ohWM*+;ECppi!RAqB37T;4j`K$
z0aMDKC>tj44ydSAL~J;gZ(zcIH-N$Ml8{{14zRXgEgfC=YL?FT<^*ZU$AT|Lh60Ms
zQ{kRC)sgmmL{LYR-pa%04eu@sw7K+%(;Yd?0KX(`>09K;=uj|FVZ(r+!~Ok{!B;~z
z!KxbEz7@dIJ$;W~{Wxj;GJsjo_|TTcjV2kw+D?2CLnOKF=|XpTG0fCSAYl0+sCE*+
z*r;457eoCH&yebqzwawbHn}n|$RyJ)Z^ey7`<atgn7n(!)JVp36)Z$~lxwsFXUmfd
zlgLgsk|Kncxf+1NO96VgxcFMj8ArKGFqH#&S9(8R2Rcb?L@fB?5_66?ho+(b><1F?
z8$v43s*E~ROXO)KuCVBgYiE3Q!M>tAoN4zhvC<sMW%7rUtPd1d0L5deXSYD`5AU{8
zkr1Gu@@+g1tKSCbShAi>6A*|$#$w8l_KeeMVXacF4Ps&Aw#=ScMM-;Qz<g|5`^AyV
z?_GA!3>#5RyQ&*J1XZ18gK2a)`sxaiq75+UZnEknCsgfO=%#tPCxjfA`qP-<O>Wg*
zs<K3(<6jTyIJj*KWOJUCOISVKltVPJa@b+b&{k8M5_1K6nhyO71T5=kx~E4)TE%EO
z*m(CC3~8EgU~n*@><#xElzwwAYTd%|^mRr&&TIfpr+V->(OchwgsDAs4v*#ut+4<q
zT+xl9qRa8cxEsU`+QN?jne%ftU6m=Ow+y#J;$=schptng=vQ{j4RcFOB5=UJ7MKAl
z;D7d7tNPO#=galk`h9xJN03$weYU4X4_?8A!PWCg0U?#2AzbX;*CXhHKG@KAAvhJM
zo>ihp4NI{&VP7L!PWRh+7g{BR@ty_IsAo8JG114!54on=H9iokCE$?&F0@K5PU!<U
zMg85S;+<6bq^`2kQUhMzUwM+JtJjmoynO)ajOFNB9k_t?Yv+hcWSN$mN;*O%PR3t<
z#82CoqMqOxm+tux#vLS^*1!G2+0hk0-nbF!vToD}>+>lm1Nz?35c{ifJ&ItYkXd}(
zu%<nuc^H(XB2&Al;AS*tqfJ+pAj|e#-sRorI=#ZHJNBka7{7%smD_gRo$Lk6vu6>w
zDj+k#XBa<Cx=l2?-*^i?xe#c7sG+F3KwSKS1TP(@W?!9)BCy6AYwT*i)&KmPz6mmf
zg0h9X$zyPE$&a_$uRCx3MsKu<3_~Fv=Cpkg!d|t~H^$myR{e+e&__&GN*|-(qGLMq
zjjx5oz?0Q4_0hCyjur2&{;Uj!<lrIxE{p}DL)sTM&*thyb}8cOCo5?1y6V00k+etE
zW~ShLmH=~&t-S0eElyGkKS|p=Fx8V^pQrK+=h|UYVSDh7V+wTg%SK>>tcu(3J73)L
zl=Oxf_USodD&>(#@4B_NXdZ6_>*uROu9E@ms%kFRv!ze0VXI^*?bgrB``63*b4Sx4
z@(rBuT^s1XO8CFCrT;HG>3?S+7&(~#$BnIRj7`T)ktg1Nw@fo4*u7Rl<T519fhhkS
zi3`A7<Qs29f;OIzklL6Jusf}utu0i=;a<Xqq2Kwpd-%ETq<Nv^vHWRzYu@T+PZWZl
z#IYB^-t{E1M~R6*mBOy&j^3+3aP)ljJ^)#CzlQ-Iu_sW-{rKD#MsGHDnnbC-97dm1
z0afnj`{jt_^n3Sp7iRS4`Ej{VlKgY^G=ho8!{ZeSke}z{6nz(ktnHV}FbeScba(c7
znAZ&AwLC5(iQ`z(eg8+#1Kx+rPp|rLeFxtQ=I{R+V5E}^3zDGfc5W<v5-Hg8J)n>h
zrs?K*q<Qi8*2N3AF`#)NV(=hr>`{?Gv7sc9gbOu$bxm&^6?*e#fF*zU9WsUWsH-d#
zYtSe8n8oofgg%wRh^R7FXG4%Btm>H?6#^khm{(8;Tjby(JnRKabH+z^2E)O{lM@a*
zY-$w9MfUGOsfWWiY*Xa!i!5_1BoKgsQ;8CEcCSNe%%*0<avZ0vW5Y`L<t$H7Sv7U8
zvnSk++e9$pA)|E;uTJ48;-Lgg#0TF)&{Cpnae@Ly3`^V)Y0jWe(AhGMa|msTyCiuI
z8t$I;jxP>c)7X>KJJ>An2v_Zec_^0j9g@@f`e}e;5GRYdvB4AC79Zui8BenmlN$Xt
zR<a^s+SWUP07lId*==y;>9u*N+P;FHQI$J#x}@m`C4?e5;%%clgh>l%4C=u6jJY~M
z8IB{;ZSYlA)Zo91d0oarYy*Z0BD>}3Do##Ke}S@F1X<;2d4cDmB#%oF`4}l(w7Dp;
z>7Sz>G$x=^c%nGA>Bl!G;6Fb~2lY%pVBOm$9&j;p05??VsnO<^!zk$jJ>Li{xq3p$
zo}$5*thhAv+hc~$68ThcO*}SQG#Ljx>*wxsE#v=?!PXLZDiOFREqgH0Gd+R=XKV<R
z{#5=k4vF&7L=}B5CB#qk;rrvUk1!_H=-^=xPwqj(C+AZD!bAt^$<+fb;^;an3wvOl
ze%=y9QGEHg*pubBm4sg4jG@(MklSc)d%B0xOsB_5N;-#>j{CNdn$>-Z`;L;1jNA1N
zy<Nc8_rd+=$sxN-x-Y1KZZwp!B=<c+#=*sv<q$g0#*pPY<?QmUjAk{iVS$6ITx-fj
zX%`J^bbySdIbl+~%WOIm)V{bZC5>3eHS1e9*9{_;_rTAiw8y{HD~vbg--ZRzTIX-x
zkBo2Q56lH|pA{;~*FGY#!e%|h7gGQjv0a)%D$((-v4z&kdmf}Ku<`^Z2B+N}wOTUV
z^0ANzpbFVZpN=prQ~?-k#(QU__4=aOqGgNRVr%<IoKD4xfNRBqa|a-iMQ+N8ME>Py
zx$Qjw3^?DJ@sC{!@~8w#r95X%w;$`i6arYn&(DgnQ^6VC&Zt&f2g8-$?9sO-(pw*k
z4Xm9L0Br}sBwJzGJPH(+A#rIymza%~y1c<lgjjK_^K&XUGGIgn&i-|Jlm6(5_weO@
zYp)sTI~CMu^#=@arka?`5`0)GK3aB`uZEvY5>i0N09JEnAeavE<=;B1v$8{(;*$LW
zS&`DsFub<g7g>IrF?P$bwpvlPnFkl61-7<1<E5ndSef)l!`Ovq9?2>XI3glEIB=|S
zC5aJ9pRV;v8jLd{BP0-01+K|OkfX(+AY5tIJvmlOmX-4T;KC^(U0uR*)h+RQIe%>%
zNGo>pGpfD`nDjkE@rCoyjHn0%+ZsxMQZ9!Dk1BKl2)#5aIt~rs-!x5fc7d98$yF;I
zQ!a$LKf;+aBX(=DU4fRtzYeZwgXto`3boa$PaNV#zr%QwJ7iTZ3D>Vio*W7R)Iy|9
zLut$j;-yO~^Y%nzIfN_V^y<TqZyN@UHe79iF3-?nLg)@-&UAZlA`MYZCVaOZl%V(A
zjl)bG6Jo6rRaiNht?do9`YJU}pV0wmoqnu@@1ZZYh2O2ZB%$c%n!2awjpNhbnpS)>
z!0`Mx!sx|73MOx}q$Ruk&jJu!<lV4HNpWB&C!2*-uur9zy)4Oli&INFdHrK<MJ@@&
zydA4dS|+LW9*3CTOLjb=!&#&X&5WEB1oNt*(OKcv)5_txLikyombuf9M0R%fo3O#X
zzBvqWRBBIn6%#Jmy)AJ#vQ9mnkJmbd;UCoofikebqhQh~f@AnxOuk_dI7CnvHQ`TL
zo1}YUgiShA>YGs!`3SQJ%fgjUxhDENiqj@N<Amfa2ud+MbsN~PvXwwY*PF^vWsjn+
zuZti(tv2p>O5CO*_G!^8SRxM1u|@rs-_GKYCS07uaH?!*xen&kIm9{={?qpBwvOJv
z2}4`;lD=r8B48K9)cussww*P23dcsE9B-*{%`B~ba?DlICTgE|mPqshMZ;B3EjCQs
zau@z+&0U9w;H;7wzbG2yQQE|tWsNP8nw_)kH`CT?=@dVJ*J(II|M~i||Htdg#PmPj
zJib={Ut^_9^*ss~8Vq<KlB`nZo9HD;UI%2=vQEBvB4dZiFexQRo}8yPH$i_U!SzI<
zvcq0FpqBn#w*WiHPEima{x@4L4j*C3a3LHivIXbZc9FrPbax<GBWG4)K2vuVv_u&4
z(_3*u>|EK|ogDc&ayvgh?Drk}kC(BjJh8PUe7RlsZt8L5bO=aKEY9qYk9mKM6_7AK
zWT5A@2Q=OmR?gH*Z>~2R0YO?p6o**I5K@4#aFHhK{Bvz64t#=1QaDV}5~0U|Fr%8S
zHwn&pO$h!#Y_c$>21@;9VL%bN2R6{<A*MUovs6V)lS(y{V-W-<@gcY@2u4tN>ux`G
zGx&Odm>vKx7J+|O1zZ~IrzLYm^qd0Z-*)u>#n?M^3D$K{mtosBGsCuR+qP}nwr$&y
zVcWKCReYl`s_``5pKv>8pR?{<b3d6;L1K!WIUsg(Xo1@oiyC8Y*+TYq@o4QT8;T?w
zO(=PL%};vdpGT*6V9>9m6D^84Qy}9i96K}4X-q{gm6nS98Fn)z8&1;G?ZETpSZ(%U
ze-ubXt(S1EIDuj0b%Fwn#vB5YX;+^WXQKJCcZ%XVCGVl^EERByIxzgZUhlsBcf99e
zS=MrzRw(ZCbW{;il%*k|>jaT-^c#dl1<ts^It%Tz`QEJ8qPGXcu{=C?N(0$nXE`II
zXS{afAG2gu(wJ3zg@)(mq4I=>anY#*O1|bqE?!Gc+2YBjS9hzQz}4&3*WWrA{dfd4
zs8}hmKc$gHSkdX^ArlTlf%^3_2F8>X6-qkf@)_Hisv$tCqD`oh+-=`!f1mf`P0KBz
zQ2nR729YZ_hC!CjX^%~^H665!$X&=)U7#)ISFxthj{nLqB&&+Y+Xks^WILD%5_wak
z<q(#zyK?pk!hd+z%4SURjE00sQsuixuX2GUf?$$0YyWLL4kkrmy(m&sKQCxw#CO(T
zwkS6l``rEvpEMvsRu;zFDjJ_nF`gF!<*}ZyasHDg9JJFisQAqg9%#1^w7P8qI08VI
zX9sMiXYoAkV5tmcYmB!vw!Y5rcb}x!dCU4sUHKWIDfw+wDJeH?rc-#Hq-oGpy9GX$
zyA{%65vsPZX{meM@-r*t-5@SB0RY?p`kro>J~+Wce{>Kt(EA%HZsgAvI9W>Mq1!*j
z3W_Vq@8i?dG3$t?HbqYz;ge6xO4lD?uu$=xb6aU9pXii{8Gb**{(5|{A2XcOuV_0x
zZ4q}+Rm&0s&79CDm|&Ne=GxpHCX~m}9=p!tIR1XGo@<ntw@-`8`fuF!w*U;>U*%QM
z_UU@$Y%7q_T9%Qk!^_XVTitXFkF7ff>hYiPy&!JCBe~84%qp+)ep>Rb{_(6jFp%?~
zHHkq1X~4zIzKsnz=AuO=Z_3*FGL?``f<34>qEZL9LX-HAEY1=@GQmsxoqC}ipx79o
z>LV(MIutlrODkY2Y)xJum)a-C{ILrP?X_s{1;~3qV`xr3rlH3+dLa!oz7#G992VGk
zR|-fj9Dtk~hqO+URw4=RNdP97Hz#gW@tz8ov%;KM!G4_KNbhlgemj|u@iM$SAVvbW
z{A^-Bxg`^taodaF{51In$0FPWzm#&sV1j1f0t@3d!@pbChn1JZdfhfYObMY8=2R9*
zZNO^wiscTl!1D2`-IB{SbTprk0C6u+|8+RB|NphF|1X}+$oM}u+oset9WPq`&x+ky
zVrIkswm5)QmTY^Zk4LBV#6|n}*$E?x@C3GevEYKsoZlW`H#C(SAhmpp?EC?GI7r><
zUzc9cGb8T&L|1g(ZC$Oc5Cfz{H3#-~UFo1Kp?ngiD6*G{F<#*Q$H$!)-<}R*rfceh
z2naaMRc76<v-gdU%+9kYQ6|6iGpYopw$3*ftcTtA>-Sn#x{q)C6-89Amz&-<Kj>Os
zuV{F_@4YQp9$&}EAKzStVZZ0Er&Qnew`nmu{_IKBR{9HZ20{7B+!#_QC)@JMqSe}Q
zi%e9@%||(^aRXgdag`!ToONY&^($Z2Y)BKGad(viA%!+n_x2EaOQ0FKC^<s_4*g{l
zB?Q=q7frn%AJC66wldao_u2km3Emi&Y&NO-;?9WmfSDN4)Mqx_K@=u3NsM=bbc5(M
z<h%pjIHC<9u?E~B56W6amIJPmrsxFQsBn<U<S^JgTkTJz$YYcnlvhFfOpuO@sZ@j?
zuJPPBx+9RpYf)c(wBunC5{+Q4Rl5I7<6HjpI8a*Tsz(VOXY*$pSb!Pds^b@*lnf;F
zL21gZHa1etfXUq6sR*LH#>+E%VeT$Wh&LtU$&HFM@&?&C0fGn-Kmbb=X1LoXWoK>U
zkiEB|0IF}#iLn%$DNo=DJKkp=!Gc@T$yEX2EYX!BKNb=qNmevHLMm0CFW%&Wu)}KU
zYXT~~c`)rHJI=_KfoI6n=6)Ubp?Eeo+klVKUIgkP`o}Un6(VAJ9}*N5Q2*SiAo@hI
zvgPm~n;%s${3tA*hV5!AKS*X%6D@W`5s$#dutxl)rpnEN*HO+3aK$vj?fue9s|xfj
ze469%E**pav@Nn>H~ZzCdFJJ&;ylUvB#Dlfy2=<akMWTGifEEq@r;n_Dz80;<hc@6
ze?RhFx*}<~)r^U@$3E+@?EUA#B&UkWbE%-hqWag|GDg5OS<-mUU_M`PK3PJ3FCj=|
zbK{Gl`WIW>C(l|*+mqq_of6+=1?7T&rR;5w*mL<`%{7avnp+43kt!~KnWSVDa2mqi
z^oPa5dgmcKyz0MNqCuj-HZzoe&%GSZq+rx%jh1<kqafVLIw5XlioEY+OjBd@M+pmp
zM2~aD(LuR|5TL}PqA&MLw*3-=I1E}KLi|<>+pU=pkrhP@CG#Mo$$*07qMOZ3sfCG#
zwCo(WH6Gi<1&{=CvfVwuddJnM#6gC(6w&P?4NUEz!P^h~i=oi9#eKTm%#kGSfL4(x
zfy^dP?xa==@&I6mIBns<mIw`alzDsdlEkM7VvEcQkoL<5L^Bzx*68?gB1pxs6zs{U
zA;!e3z_{?I^WbM?LhqS<7{LR8)t2^G9M>fO2Ka~w>rG`4lfEnxf}I361rN1~g8DgD
z>V%-<%iu84=oEmNSZl7wxPikdZUVmRlFI&|$*><IJ5U`dRCTWgsVj*9mla8YPB=lq
z67Dl;Ug;0(^R4S~oujL(g{z6q^WX%sS}LwF#l}P%t~7brtlF}{);e_Y^&>{AUsgKs
z7(;?mcP{d*Jt(AvQHB?sLXJ3Qi{Svt;m0*Q#|3Ar3Q>j0L8($z1uqn})QJQHPxX4t
zXACNBFm(powArkCkO*;sJaC?H*+E_A0CN@AUjdq-Y!YQD?8aP8lSugWN-vefC3y^3
z#NiRul&|xVG4<7R(c<6X@&D_0ZAQf=!;xLJ(<ar-(pDbx$V6kVQ7ehWQEY^T#1I$R
z>K8P-5;}!;AOi3x2y|d+_b_4sK;cFPKsc3yd_|7Agb}5LJOFM;zY#Xf2Z2dkmaEYk
zh$KW8m`C(+4$I$fAYZOcY(%G>Jw-Zt|KqAPs0XW^!R-fh033qkm7U3b<;UHnND7p=
zRosbGXGuVNX%CA8?Yg@m&5>=Vyf+1hJviGjgJOjY!sf7=sfX1J!Gx5+S1i1wV(dPN
z0wB8K5ckTxN+XMQRx+7tZk#243peG41^02=t05~Rz18hWx-|7QB#VxF8&U+oKS=Cn
zYtXQ&lV(pytX4~1k*FkH;ndLax?x%D^b$DR;U^7{*C$I7H;Ucv_>zOupk(T9co&Cs
zR)e64e*>Nx+tUls)>z~GwTAiD1(1l@ghaSd(+U^gBY_Cte5rQ{KeO#vhL|ceVoe<a
z5)yh_MDzVli=zt_Gk`x8SIg>P8eAk;POD4HOklPwVZVCi!&$y2cg_nmU4p_mbF(EA
z&(umlHea?ehRqK<&;!OOlCK>kGWl?|glZ!Lfw=@;7LGOheQ#l@@N;jXNweUU)D1Al
zS!4n1!QjYq_ggctQhulXPT#ap&~`Z6mfm*n#yq>YWF`1G4Tmsa6f-map_-3&e^E@>
zj-QCsi9b|Alu>_i9_Z9KC%6#fH$drN<ybuza=-S@)f1wmCkXh%Jq)o9LzXVf$U7qT
z&kyR+PhQiWf2=`_R<k>@B;V48zwxT(Uxh7g+oi5+ID4BvZuN#-(-YqwBEbUt*rw+o
z`!fW!M|AYya6Aw&RZTPAQ@C`wNq2v$->`RveX1DXi2n!41EW=5ETk}(-U7Q2*h9ni
zF0#<}w*Cj~GjM}*s?}z&>}#D=WOuut^WPDn6Qq0fGH-SfocZw?L$GYqQ{)a#xo|?F
zz?(bUUU8S`wpiHN!CG42rg0|ncn1JhO?^e6eKu{Rjmi?XO-!@sK}e$MoNOga@+CFV
zr)uDb!g}BoLLj?T@?}OlR@DfKGD^H`#uy9Fjw(05VT$T=C-#r+@Pqd{>obA=4<(>y
z`Qwl5$U3GOQ}^e>PV80@p5-S-$A0Q9(cI4ay6wYb2`Pw|FXf=?kS&;Ynaj8pwnn;v
zjPQ2N3@VoN`1gJv;Xwb9$+mtfb`(d)?q#V*go1;V#n;6)`&(Qw`QLhw!&ijs7J__J
z%$JM-@6hCBXqNiMZHZ(6`giNpz$5RNOKNGVEzm|NVt9ca#1*p;l#^2~<{G0-gU)A5
zo_8p-Z&|_!E2yvzJQKFlDk!zYgH$25C}rWrMD=u9`&`Vnmguj~M0AMPS1>qMcNf2?
z4d2Sca<7dYI2UXT+&dM67cMs~0_-Oa;RihCsvP}4FI)ei5uj)J-xa${YSzEVI>Ps_
z1S^&Sfzb<*5^)gBNsYm2u!H)6PS0ANqQ%OxnM5`rv1Rai!}U<Ug@W8=-o*N4s6h7c
zZu^FPP7iT*cO$WuoZJ*W>YNX`x($l#fv=2DBw|(!EK&WI>yUoWh$$r^B=v+hnh8Ev
zi|1|6%faXE@rAsQ4T@aW*Xw!W^ysj&F;eUiQ#e7cd>3y*l+KzZTh5i4l;c<dY|ykD
z;@ZRct*3OFocuQoZc729LI{u>4iKY2qC}I5L>5Gs82Ba;Gbl)cI;Tp@{rv0^@K`qH
z5bL9ma9TD-loyXudBe6ANCN=V8jM95-6{lFXoQ|Fz$3?s!XyVWOOcnKYDhNS?XG0s
ztH0j7`weKL*ol!`+^g1&mrH8TA5EB4l0trV4V9_yOQuR`%Zi4MQNpErC=~MQXtVH!
zcZ4u3<w88%P!@?hq~ZNjbF&Fe1xER;!OnZ1jIvde5f}d4v&{>`C)!L>geUv9I6j36
zEfNgriOvf*LlLDQfY~ee8Ou)0r9s2>Oky`0U=Wm=R1Yk6ZzU~MSgmtRO<09&ESqPe
zsb?;w&Gl$#L)l}q40PI~Gk+B4k(0-=>t}e1qN{J@@lis}4PhazIO{AGss(GFnPxkv
zn-8WTj(fV~C6Mf{%22IR2}5Y4Om$U2SC&Uzmr^ET;f%ewwVY8~Jpa=RC-b6+bOsG$
zn9z|MpOA@b5x7?-r;NY}c6ZNSiY~;;Y!a(I4i!R^>pl?LCc4MnKGT$y94Nn?S(n5A
zmoXbK8uk)EU7%gpZ(+Idsl9%z?Cvo;EH6Zq{>sJwqN-wC#Yz?#3xWpRJQOFnAfB>U
zFseXKg;D<^aY{$KT)FxFk3!E7jx`0mx*7>-ZnXEg2H<Q1fr`3sew%$>n6b-lgG3$+
z50V%s!+(pp|8*4-asu;EGF#U=sy^O9%?gclR*NY2&cAIN>YssG+&ZsoVy%kXS#l!G
zzGe}1%|qbacyx%=JB+m^I}t=^h&1L4yg>-&Q8qz|3eG(qirEl&D|#leDm2_${HFaA
zCj|u+WaUCOfQx>#(bmQU5d<FWxF%Tkl@BiKgj4nuoVqL}0>PQboA&X3Mws>B&7o$o
zN{{d{QK@#x5w(cp^(Bs{7KtXM>+8NggR!(@B-2N*C=z6|gcy_#vRTWH6fy>5b^f6C
z$<aVnqFsy(#jSM2zl)~ng-Yd+f19Kbp#%Xgu+v65A^#B4$mTy%(+!;1YSpX62atX+
zg7x(1Vb}Js^HE5cwma{{_JFm;3cyFCWY5gE)Am!k>FLL`#u#m}DdyV(_UXN&JtH5f
z=W8}QomDz}+IK92#ePda*_p!`xs^3#0|n*B*6}_4ypDa?tFXJaDBDyIXGlA%e14`8
zu?CG64-J3roq~geRZN^a9JWBrTm1gEbXHoX4w3{~Ev-K_Sfg)=0}n<66lxthFLV?!
zmr>+8V<vQ<@L_k%cY?yxv~Lu*zGdRTbZ8k{_bLq53Gsk{@T^}BnoEI5u?Mb&@-rW5
za82Jj=B5wBiG8aX;()dHv#W$vXJ-bwVOzjQBHOmKrvmh2yc6)r<CnYUT<<^}p6v7r
zJ=%U8nldYTxf~R-#!Ph>(H%YfFwpo^VNTbUn%Lyk3xdp4WD4#I(+|Avj>gvu@#g>_
z-#{$uWOe^)9ZeUO3nU2E8e9;B1Hzjg-@JW|7y_as|M7ucF0)2o!TbTii5ytffDNEh
zBrp2=JlRYm*f0wcZn=vby5#Y1pvvZ<YQf6QE&CX}lqX;MFSm{QEsa^PqYeyln8VgB
z#C;N-8DP+3j23RY0OT#+!_1^T5Z_Dx!FWo^H%>T{7VHSfH30_v&ngpjP0PqL>qg4P
ziPiCH2Pa!~1an?XL$|Sx5tK>S=+p<B$B@b)#hnMKoHlK+a_`sx4|@6t0ZJg@Z0EK6
z3HaI%2jmjfbwzr!jOjO^B`m0t3%0I{rVs)!&`Jy{s4{32{dM8C>Ij)Kq!{~sa1#Xy
z6ge^M19ycj--=0cOM0V9xCw?&6Mcw)8{ri?(yGBHBOkw%L!DGms<G{SxgZ%$>?D+0
z%i4amAd~Xfy8itl_sNor(P+n8o-{AK5VCk=2e=<tTzyE+Q}`VWG8Bi#CnVq$jpOw&
z<MF|)sO`iLWyxL7E2jCix#uuwxGVb)!T~l&b-p1Nj?DhO<cm6PX$dt2eb3wDPW{k<
ze%>Ck!n-bwEbZ=pj5!9n|Ehemu>H>@>lK!^<3Y>aR&RiJ1(zY=>*0veI%erp@pNe7
zE0_DJ_Hjr8MW!oKd03LSOv%^P|M5NYE8c5gb9kRc0s=)2T^sZo<86s>y}ut;W+u|=
zn4ua|MH=I5Q4ynxii|{AB5zR@s3McTozb%Hc-;xyR)LctmyqPu(fNLiyv4Gjuaqj9
zzk;~Pl#(4MUPR){GP1Mx^89=QxzPT&`(&JsO&!6jg1GrSxn}g;Ro2p)eGfuo`SPF&
z^sP-i)b@0`g`;HIi87k16GexI96YmdG&z5X%4%<BeR>e>YT$*7Q4R-(Ia#O?C7AW_
z^gE+s`^gd~Gf)>;BT~!yycaUL;6Q<1Fa+|7%;oDHt#@aka7gS)8yw}zn4y(Ia!+*V
zpj7fX-9`{y<i<&nrTY%=AhKjg^uzE;2OB}%Zd5<lm_4v=eC|16Ir6-*!d`4uQ12)`
zxn$RFPu`p>&vkivO~LBkOhfE<veHg$IKx@)OAgSmN$w-O^@a`0BCYX7f*x?)&6ucV
z^*0VdE+92|z^ZO<qUM=$5Tji`Sf{tz1g{cOQ(<dtFb@ToOn68px@&KtE2}F#1n51C
z`=d-K#>?n9(www?iB<jz>qQ%X<JWD@aOr9zzw)MqCbs!uIyyii*Yp`p&wBbq>7J<-
zVNI)vo%@!mzjZ96`(w)9f5_lL<~6SM8ffmJOf%U0Q3*P?h`gUHjNJpe*9seN`B=HA
z8N}89%*W-G=(JpXa(6TL(C9;K6X=PNQcZwqWh06=YD+qCu~ppMe+)!`=+a+=3rVIJ
zQQ}UcM<T7Y$`&W`7&gh#eua4Jc#vlY*g;sidb%cA$Ja3}?G?*pi>6&A{*fMU!<pYe
zVt-_zqujQQsGt-u@PBIPAQiZ<@ExEOBlDb>%Q7dJHK?VQjT|@+<8z8(UMQWzl|f;=
zWdIDY<;+1cLDJtdOz|I-ry7=JQeyue$5l25y8agw#_4Pw*3?iicUG2W)na@#x_w;N
z@$}IixcT4`^t+2vs$`9@WcUo!AYEwb8o%M0ggjNUqr1W!W)b4e?vpB_qzMEqlCiew
zOBIPSP)(m+Gi6Sta?2yc>kmI|tfe>m-V4r}s9!qgA<f#<oYj>GkFThvwYL@%J(oFB
z-2GJ>)D0+dqfo@|Xia@a_{)({pSQnG`DZjlME6OPz^S+&c1lKC9MyQqR*K$T@sf>G
z<}@ZvKj9KQ>_?K??V3PW0KyD1H_RP^R&xxmY*g!=Cqndrm~qzKquL}XAvk{vq`>6~
zY>{FeOkG^LQk)mt0N$*aJm%R^#JnbT?xIcB63hDDxWnRvL-1OHB-e5$y>k$@G<d+A
zGmPwTk~o)~V;{Su-+4op`!$p8@hLg;bH++IpaK}7nn_fX142F6j6P^S!2_u;gt(m>
zl?Ny!+{kQ>YOC-<xF3xGI748ZFM-;yXQ8~G1_nC0=V+2(iO1-rDI9IWTx8#fN#h^N
zj=x^j5hl||14fyuQ%I;|8+&-RGJ+>Dr;Sc4lQ(+aJh=8waZ~k5?y{<Etpw5k9tUFB
zVZG7)dQ`Adqx*Fh1~^X+Z_JCSw_6)Kd_;AJKD@fs+4Ty<Jz5Tply|NZS$*dys<jx1
zQRgI*^k@<=5LDPjhkivM_@;2&QaF2YChjkwkt$Yn$4r}pHJ)jOG?<LQ*+D^VR8j)J
zXkCK@y|hL^jUd*Z^Goea+ZD#FF^AAZC{P^aDibT=;-K~x8xMa}pMYm89Cteyu>^Pn
z@T@!w;Q`o(lMpW+xdtf`QZ*pEtNXzb2Lw41SCLc5oa|Dl1!<e=Umhgq@!<~uiB>BD
zpcnc0c?Xzs7p}hi<ox9kwICt%pi6*NWa^><A+lWqp><jbg=iHpFj>b+J-X*BT%Iw&
zpAN_NP|fBV*pZ<P?S-1iielRd*qHAS*rUEiFO{0{XZoP=lLwEJScCqD#ekwQwMpl=
zmAIiL&4dP|U22CvS$6Q<05#5i*~kMffU)GX=hdMU*x(8auX46K{{DZ4$U7}`tM`^S
zKSa31>xbVJx>Fg+7~(_)KOHQP+hkzpO6@@*ghABGD?6$y9b|Hq*7}X)*XM1{26UGa
z@;AEodscbC;^jNFRwBAL=bP57SqO+K^G)u#RdPS>ZTT7l!V*ciGs=aSJxlHaB|_!+
z_15e`8`C+JRcs08lV3plJ3W?gjwNF4jPGW}v|y(cB`iObNRb*%zko2UFKXP1($Tpm
z?_t;CfJnG`3G<T_%2^99x~Z=xP;TQYbc?mvdH*Qw#0K<4n5_eUES!4BIgy;VUkp^7
zTKsz@^d7D%2X~$Ywwvk$1;gWuUMqc^53v3ys+*s5tk}1o;y)_HbN8(s)l*Nu-M77W
z=l(n9O>x0P5XTPhz;#$epD6bRw%SLl-U?Hu(Wls1Zc3BOyjrDprO{ePF<j6zN^(GU
z#Sr1Fesr3ggrdYOe%%F0b%}|w1(yZe7B30D3JBr@Dl`31!k1b&a*c7?CB_hTy#7M!
zQPYZUo`J-WIx-EACGJO(g+V99zwDID34L`Gjq4T*y@Pj_hzMO0&DcE=ks)1e>qIEX
zL>#Jr;MIGul-2&cIJfQZhaw1#sN+H~@bE=QhhIh<DCld$!JyT_zrq^GqBYCe2uqzx
zHbi{Lk2#*86wF$kO9-zZ2d09EcfRf(x!+U7^|sngxl96#GUfPUl0hn`0}?u&V}?$m
z12m3~Z&B&5z$I-1=j7Wb;7k--Om!=vehJvMza7By9&u(5bH@aRPQa8T5;C^>l*76W
zInjd*<~QAL0Ld9569|q+rZO@I@kq3GNDIdTjR&z0e&bj9u2h1&Q|uOi9uOTP3IP*T
zgxM7q`j!Ubh7earxDMLVYB+AJmf?(f=)p8JspLAbh5?0Q@=`DbsBgjrStJfa3Oq!&
z_)Ui=XEv@`P@d0ZY{mXXaGL*`S0!VGNRmt-rui9KjWtF;C_^Y5mL5U+Y@zctPzY*e
zz*5ZD29M?xo#qsEqB1v;nCo>xI(ou>jxs7t5@$>Dtsr-1e^Sz?FBW!RK964u;cr$=
z(>XlvE7%aY0;Uc)7IC(y&1%DpQ;KiMeD7XKpF3ujU>Z&K?NmrVgDw5OXa=^8Visz5
zC@sRZq+gu>8JKkf?6msU6^K>EWF*M|$^2^1#H@_q?@=K+2BM-UC2sX(p7M@&II-+H
zQ3v+$RizJ6AO)0&GbuU<l)<20M(icfR7o)74lC2{)0Z2nZC_VQAD4#W>ktqYaK-Jc
zWls0(q&@JXy$HD_h~UIA4$fkdwx-JR6Dx&}lMc360QA@ekDnOTw=**o8YaTCG9M1#
z>U7IXp2rsTg4xCp&P-TEqVUp5Rr3ZG%RssGr*64&hg$4<dS)}draThAc*6-}K|C;8
z{@oueB@+z_vx+}&Gdw@2oqej$io3tyC$qvxVASVKV5s{ed!sd~?yPP@z`lX_FZYE8
z0F!+)2y;q8p50PQev{^n3j0?`G<DE$E7EZgyg9WN?||vY`p<p)PcH;(0Q?vHPrSp+
zf8J#O%Q3+EyD8~^low5@ZN&TrHr_9)r}83(G_)Ov8z`$F;u(2nv(a;@BoJnrEy}0+
zYV#bC1z*o_?8NE{Es+FkM8SZGSa4%|c5ZN?n;Y(%p7fp#S=wCaTAmqaVj|lP7v-bS
zfnj$9k15zU)wG5st6qpAo4)T4OSQB)-RZPOqhBjC^`G7}LASLvx;or%pzGVTy4qau
zi*|hkMqE*PJghkh1AnI2(sv<;Nf+D5H#dp&iDtJO0Jgf;*TfBWs?Pt|W-j`Wev@&V
zu$GM3C|FgusIhn|m>m!BW;AT8i~w1w`roE{DumEWB4lO_kDni>RsXV+5=v^B4cIl+
z^5)4JiT!<z>0CyTNz;>6Y=`FYDbgMMxmFDaS|mOlHusuIMx-gWSc1%ohV9k&T^vhM
zSwOsKA=~yRPfr)<j3@JH!S!}DsP>a?NNy3rD!VP_hf32<RQVT8O@rT(3r{Q09xYkk
zSx+_cQP30#Y2tY97X#Gd!i1;fIS2{j)m?_?)~3DQ)yZF2!9eYc%aTx>(IhJ<I|?8=
zKzkUJ5jT8}VPerSoN0xW^JLu1#P9?42p+T4i&A}16k$I?TeNk3njnzSLi^NuS5O+-
zKZa0utQ?@WU--RllV=G=|B$LVHwZH354tTeFL+Db+hJK>QX#TjN!ndR?#q{Wr_d2k
zzOP854d6KxT?W~r84~iI`e)yT%hakk!9mRyhd^T{Fgjf+q5IWwqCdW$mSPSM{39$9
zk6&bJ333d%BcN;vL819AnNaekIq~i_Uy<7ydh7Y98mxssAcUP;{zfTxd6fk-9k)qh
z<9vmVyo|8hfK{8(H+|%{sBv$i`9k|hnL#l*xq-Q%`hyY)`T4G?Ri*eG(`l;r)yDW8
zP7!~KP^I>a$sV_m=_}^jCDu3wdhxA;=amK~Duo)FU?tR6BDIYg84C^)5|;xCX&30_
z&b!4}k6fR}WAM==95`=o^Upw|f~nXa?nI52vz8mOXrZ)5osJ7Co2G<eA{~Z}Crjx4
z^9L_n<yE%s>94>B8t0gC@g$LW{qsjn<5E$xqxr&Xq+QFjua?4zg|A)QT2Q<KEV4_3
z(<1ad8t|mX!6LL1z<H-M`y7rb!&{$~%B7P{-<{OIm8t$St*n=^UxKu+K~JKNuTvhy
z+tx?f&Y<`c4CYI~q#N4oMo%S_p>7R7uBc@9R0AObZ-#S={&Z{M=m+ND4q=$ri&1z4
zo^S4)ckqV(Squ=(QFynp8=iuYI~?Fml=~WJydPv2E-t1sYaZ_)F%R6^uMSoMM|2f!
z0i(*;uev<nBy|vGQLHrd+nvfgQjwz$K*iv%K>E`5=iIWiKLX@V+8|#r1G=3qLm%@P
zs3Ih(diqpo)*0M^b#Aly1qv@k>G@U2I_1QJ46>vs@~||VZb^}UhY`4nssRzoJ6+=Q
zLM*kh*pm_3_-jB6zN$&`k-8(Nut~hxuwg;-GaVT@BLQt)#GZ)-y^&|w2uMs$pPN?)
z>ro5^Y@_)00=WDhECKKIym89#an#dRJdu-WWi-uRIGq7MPGR>%_rVImy`>l;{T++|
z{^q(yi?xD<a4ny4B0JuuLRhk8Lhbx<Qiq-&d|+W!XR+TXHR$-B0v{-A*#(eF8%PU$
zlK)vf$U>lBcvyi~U%AVMNKIWjOPXsiHn&;e1)T9pzAiQ=x-}>0BEcvy`k_Al;h?w$
z4q>3?%3}j!16_O$fPxeSnamb0<c$f|iNy4^KgV?w-i;GOI$eK`hgq?Mibz8%17?k6
zVEFYC|648Pb@RrAISe#xn`~?3CX<U9y1v=#Yt*Cs*LYQ>@^UZHBY5u92WuMObi45V
z1F951owAKkSu$#PgpEU4;6aA>i$5Ynav*2x$-!yRM$6>@dJh8@^OqArr)4MwfyG7D
zrv6eK;-G12xGqxP>Gd`Qc$>ZK-&nhD=g#ey8*;&AOx*-Mg#MtWmBQAW_Ey01r)Z#)
ztSlKoeOQ9)tYlBy-G;*&2D_*$;Xf`TY<^7ai#jlSxg>$AXdY2>^l_v)F1r9kV0v<+
zRfcw*a3i6(IQL^i5I%Ww3(bpA9SgBD8U*Z3qHQ{*iyH!`q|by}Via=A%7A$5Y^H$-
z3J6tfY#>kH?h*gAZyqF9Ph4LhYE%G&k6**3##XycDAN)8IK`unAg>ozk4x$clrxW}
zi2lyt;4mWor9ae*U&S_L$Z#>*KH<5EO16WJGNL=@AcZ%TL$BTy5{m%v7~^4htm2d8
zKjD~+MszH-^&22<wizBsqUND=SXcOfNn;_KxVU>V(-7GG2HqawDE+SqN&g>N4?QFE
z|BSX@sr{nuzY}sr^;X_O8p9PSv4v5hvS_SG=G@|awj?p1M8e=tJdOCq!P(J9=WYNJ
z|A>rNCR2770!dpR-X-284PT%#UtnqA$U^@fG9{2;)G<dI4RAg%DRaI8=z{R=#V=U?
zZvFmw3t3n~JWNc)CN($e_3`z(<mA8w)m|X@$_SNQGI%gn8NWApclR0!`RV;U;f6A{
z_qaAejDmyj9ZF00d3px2qs!C(e1d2QGt7g{>qGxGMuZ_3nvR5QUmW10Srbs{gS+Fh
zXvFbq*vCf=t408l9|anLRbYrV*$TWt+XVE+rk5aA9>@Tj5{PdL-Ee_6EQJu*-)!~;
z^v4M&e~NYJoLTu12+s<|LL&4>FA@ESWo^}WV*&WGqG5@pJ|K2B&UIk<02)?+27;A-
z;E%+ku0lJ@2+PFDQP|L+Q|2+V%TloA3T01U!tGHJBDKjF)rQd%s0)qkgM0SxLcv;s
ziJOKJPvYQ3y#zHKqF^GW%nw3PQYk)O)>}CsaCh#J7`80&Ts8)Ue=aO|6^RbX%VJTq
zz0WQxxsDQX5I7eu>_`_=$42!{mmN}ly>;#s>u-P*PP<eeC?^ni<_%k9)OTpq{L}Ld
z4k(&eplIsKscV;u4p78dBbHz`%GI&`K}2LkV9c6w4@4hZ)8xqqHH(8&VX)Zm1Dsgr
zeTj0oI0W<iem?Nv5-O`?6rSpNY6hMuN$Lz6Mae_9h!1FWL{NPaaQ<2&5bs{vOzX@y
zeEzY{jfuOY{UyOA!m5l^w4rS@Nq90V>Dcncj^I!{$HF?P;>w-APwsmq-1Zyn>9q^z
z0QoL36MeC}=U@kx67K5luXZ_0mM4<v6yqedDSDfM+PkB*2-&rXUPoTrYS8Q1&0>Ls
zMM40D0%(I^4xi}-jgva4KGO+*oSn{TM#j)qX%wJ&XYP|F)3$XC?1c){S~Jh~VcGU@
zn+{6y(-SR9b;wQOR(W!A{3U6SH}=iC(DsSg7f4iG(zC)?KrY*isF$YM92ut~his~8
zhG`<FO4vj@y&uIUP|35-CF4^B9g5~g;F<*l^-&<39>f*MHk!a_wed}0_2+0Nzt5^Z
zeL(mr42je)Ia{R?qf6o5d){~1KO_SnkgyjFfYFMMnl0~+v^@w>xPsx!GO@aKX6%6z
zp5=W5%@6kx71SjQlseT+_j+x+>CQE+Tlc76s+Zv(aRVQbA)T>L=x1$-o_t?&V4bD|
zt;ervFo5-mJ&?Vp0G1clobY}z;v}JURd?H>HG#5CElWAT`acYKoK{j~5tzB7=sK`F
zLXMi+TmOK&Zu|o?qf)-+ItQ*au(uDy2WoYj{N>;Ycj#2HncCIsSXx7e7&cdUZv&qc
z8d00y7QWqOEnGD6ivozVZkKeD%Q(&0lUC-<d~qsDXq;JZ9WK|Db)SJBTx!Yo`w*lt
zWfuSb@C+s4@?TNW=0*pRg*3)y*Rv~=9G^Q5`=q7a7+b%wTl#7+r>53_DLMD9d5mDB
z$@f!+tKIPZijKSSn$j*cs6rfVnm9FaJH$5x^SobT##tk<yl;YJ(nbAR$GXytS!jIw
znk5=Z3(&0HQgL?5K7g8A&>5DKwlq^SGZ*NHUH~*5d0v*&sJ3%doE<Y=w=@Jxn8S(Q
z@N4=G$yNNr!7Kath@St_!4mW%egH)|ZW9K4og)DDfNtyX`))-XG}Cl3q0@ix8l+8D
z7&vNhwhL>Rn2Kj-7vG?qe1E~S4n8=81?&9qPzH?JAlRq0q6@1prN~?bi!Vol)CAuj
z=?;N*UH#|j{9o@Ec9#FSJ396MIVEnzRH-mz1IiY0$jT=#-5NWSF;!_Y*yWI{83#j&
z6`ng<5_WoEnFx@Zs^&(;Nx?G_W=}^}qD5HE9I|H*uf|+c$xm!Cn8OLl4%8jolf$Uw
z8HklC4jqyw6>Gu?N*R;q9zT4$zF#)*d9tS43eJ4DcVA8vht=NSzdBcTz91}fX=tZ+
z#OO^Z{*PnTJtlW|04ZCpR)FC^19425Au#|Vo>+ZLo0zslv@4BBoeYUAJP8xYr$|ae
zLXSo9pINsj2d^2{KYHebQgN^hMv1mvv37^?A3OK^3rJuxCVj|*KJhxC-*xU|iN=l$
zKSGJ9Q~@|ddgt5mERcpl48VgNZ+7>Ip}YyBZpX6;BJo<01SokjtepUee5A-S;yJd|
zCtji0QHGqrwhFdP8~PEG;p0PuB@_h@UbUosl8U&R4HF?zX`kVcG-3e0$&Tb|pOo9j
zQJOaqS@erS5b+R1fkd?~0FaM#Fne9RGMMoay+V$a1sIhSVdJ}0h~_OxN!L(YSrF*v
zhO5-Huyoj>fOtyoMB!Vk`lQs^^OGN;4_tQpoWGFMFR>j@`X*`Zy${FU@?^P*I8yKg
z{zHUy%*XTy5*@ysK?f^p{qJH)Mt9vR?iZaSvAq;EIV|kLoOpCpaexcel&*?O(^vBV
z3IGX-hA`m*7wcnW1RRH~{qALXlG{SPsKAhze4ofH=%(@foYUT0k*Bgq+ukD+^IA8>
zY%|zZLx>0b09);~)bCX~vs2vHQ-8D}iD#Zfc{Mi>zRYo+dsf}*V2^YY3ysf&q-xoW
z6y90N?9)?*JL)K3FTgd3PuT|hYc?IYA2&uYC+S*fuk>A@<`P)}fDCeE)A+@qZ-ekS
zm^)R1Z`Wi8LPf1o>2PG;Ic8!Sh`U)Mn=aEST+e1~uB<u;JdHka1V%?0V&h%G?&d_+
z+~n4w2<Gk_)V3?Mv9a{$OP@jmJPk}=8DH+jM-7D4NrpKzy|n%Oyrc7lx56dh&3P^5
zq97)_Z2lt?>2YSkVLwdt6o+-D`6^`eL^ZQRL^LYOe?!0_e3)sduI5{^uEMWrAX%bh
z=gIkgbN-`M^g6;uhLE67eSmO*qKoZ{E2wS4(Joa0O>)`T0W|d}UgA9<9BGtDE}qYU
zTw=Gx?6|x3f-NzQJBs^GZAJW#IGHz1H}$fxp<!?AIDMw#q**3I++@qzk1yUKttG%2
zyp&0t7={t)?7yqtzo{%rh3~dG(?4%#PAT#kHa$<D*6m_J%()DQ_)}f29s+&Qk4%_2
zQ`R)xyrLoYJ#~zljl`sI@oAGVIw$eM{78gGv@vn5fW*kQ-&cBUrQ1b0@wAaG>ccrJ
zzs;3xV=X=fxskS1f3LzN)wh>x(C_9|-n*Lz7uiiuhymnK_wab{G>e>d*O2+ZY;<2I
z99|O%^rf<3><wsgf&H{Lk@T72W^Ag?V;5wZIb5I1>fN5#_VuQT{PT6S#;I#yhd7Hz
z%cPDp)@}Iem%+cERij&7cZw(a7L4#pvdo;{Tpq%{;h#@amEECqi9(dz6?(i1!uewL
zx$@2izB+xpXx}mL4e98ZM{$FUn4}yZg~;k`dU6hno<o9|gU#U#XiC*l6fEL3JR-il
zSfn#a*P$nM#4VEjg#hUEyR?)fWT+f&Kv-c?ZdcoH1>35YCJGdx4ftqr0&`_KjA}Fe
zmnSWqFN(sNi|~TYROj#g+}PQK;>^46-paq2(yc(IjVAkmmAkR;Z(k)<dF@?VjycrA
zby2TQ3&~Iy)+$(Sb31@fuU{U;OBIzt2TtF(%|7FA)nGeP`4VGQvX3SV(9Ure<EDeJ
zBDHN-5nivgI&2Go3vZ5w9_8)07nNN;dT^gU)+S*7wUl2ZtY4slZ}j_o6TVe#@PW%h
z(gkA8q4)_<7zJdu>G!GwKf!?V{JKhje*1PHssP=NU*COMJ79v{%_^!>GH$yFm<qU=
zbl$8*ZFNEm_HV+n?kWcU3di<q#~_@1o2sm6yC&dpIiqIJUtf7Q)@JehwPfNqDhGIw
znnlQbIUjU)4yd=G+XtqK8uG5A_DvU9YMXab6fR6GaN%Pfwt#ZI=be<1>u;_aMgX?f
zc;NTXezu*0TSK8^(VbL7aC<4mTkGyK8A+kpcW3&3KT28Rx|okL{@F&wSDYpOlW{q2
zICr@0jd+qWtL);Uyq$7Yv2#RFa>u?b-Z*S>JSXpI#Wc$(k9xB+xyHx>79Q1q1lq0J
zG=;}oZ3PU|w?_`&JN>**bs1Tf^XKkOQo;zpiA8>ZFdMXpIodDS9li=pg)!gxz2xEV
zAjR~hi0gdHHJ3)?fTGam9OH-B-%rMWhnRz<_+tSZcd&ps9qUxwpkTvc2A&XGwY&X{
zmseNxF~48;kDNPH(WR(*kHnPcgX-Mo4HKEH2RA%UUs!{xgOm54L0_?$SEgTku<k|W
zaZ8OAk@No5@G^XcL9_9#n<qr7XY@oj0{R)n4-H;i-wwo9g+(p*jL-Xh+~AeWS7)*>
z%$sKy7heyZh7>*?n^U&P*AhIuoQ*}!4`uZ^;*jlTdjxH=D{X+)J?O|6{%)@u(SHvA
z{}f03c9H+b@bc>aq0O%U4{fHx=(Pmh0Atr$B$-*PZE-o+P@reTK4&3WLqZN4^7Vr2
zo@7-Xo!GGz2m28h$C2O)f8#czEzm6)>oEf+OCH!wOxcz84-md<Y!cbO)QGQ4Y1`;L
z`y*gtFLUgnn^@!^+M0_bAP}na<O6@CmMqkC5b?KqoV+Iny2#7v#T?bn%Y*&n1W`HC
z8xxxUNIpU<Yk%<i2%gn<w_qB|=Ib7GKr1+mk;RS2%gyd(jQ~?N_<uK#ztwa+T&c6K
zJkISH+<5i$QamP+e|a=6N@AptaY-~_Vs!t2uXyx7L?|!gi<z)K+o(pK!eo5YhOfmS
zc;{l%TBjJ1-4Fpt5XdNDnh<-22_W)e`Ry`^9Tc43z-Moftb_7=<v*WSObe+WZ8Ay#
z>%sg&m{?b0eW%-qsSblgg;{oJOz`@)@Jm?5!ZjzTK4^@l(&R(JC61pOryY|#j{=^&
zAhwGbxb!k0iP{T6aWksoH_OK4;`M4Z=ZjnWpIgXT-Htk3cCBP9C(&j0AnMHpvGH5@
zMz^dRkFLN<1hYGwa9_uHg}px<K;lVInR3dNL$*qD?)K(jUf^_w$EGhYbPghTC>cC_
z5LNEI*^Ry&6uL>$)!k|CQSoa0IR%XVRROc~xRa!3#!!3(u$XI76>&K>4ewB<D;-@^
z_mZD&6e@2`0zH>?e~6P<sgT?fhJ*Zf$e`&yI#})w^6#I#bT%=7`4W{2PCA_Ky&@93
zMnyBBD~}kc?R`7Jm5OZE^6M|0_lziG7MSQQ)DBCp^^WBhRm<D?;2Xk#ksyV#P~4SM
zIVLRSXm&?oFotEA;Og>2juypBJ6xcjt{9jpQ3Y`U7zv<1Jw`tjPEaY-l_(tgvKY8!
zRUEakvkLA)`0~`>D9aGL`18OcQMXwyWA;-#qr%hPJvf@7^oQDpW6%(#Zegr4O5Br|
zS}B&q8#FKogxh&>#{8<`tK6ruEC5V*OTh4;S#N{+bOBcPEo;Y%AtjQN)`FwMoDk~`
zZjZbEHnvQGOo#e#NXpCRu-yrq{QYIaRXx-<L;>-G6=%Wf>suC!b&5=Khr0Y#he#p}
z{%N-66TMCn7BA0i&MJa@4$&9V*|H!Pxo2-UfHJVH*TI>O^@lb4D6MMR@qXo@Pq@E>
z<g3O(2pikdM49%X*Wd6oLdlxpXmK}5W@7^<&gn54;)2lZJu5lepI+6qL-~VxuHqUq
z$^8>CXAvsn(buG-IBhCDImAn~7a!=^kpg5%!F~yJcH(`b@I`?%FjNPM4RKp9`4gH|
z&2aEo!J<3fJJ(9lvI%s+)JzL!a$QsJUdjSMFk!M$GhW^xzCrKRkS(6igd|oq9Y35Q
z^6;E|(dB|rL9=`QC$oLREGaLLAW>wQHx4`7NDcuGBk!k<bOJ+*mq<_jo^<KFlDP~9
zA`WWJRt|Gf886beP4McO_Lj~aB)Q03*4lh^8QE?XXGD78KJ?+{tHnPI2w`~321r7s
zw=Z+}(CKH-!?SuH_v@+r;!VcIws>*}nd;JYnI>?`NjfigjhAyOTjt*K(X6wzQAR#f
za!Cb7E>;cW>Yi+umUXNkROogi`-UX1Sc#e=YGqLOwSUa@>R|5S=y%edXO7IqM%J#x
z+EVJMd1KJ2P`sHu#n1X_qpVOq!>KLi5BKUCuXyBW1f;Hxr#oNz9Jf+fV>4_7i>;ik
z9d}=W?K`(x9k;f@DM!88+*HwBv7fr(kkASa5KzP>9;7qv!G2%i(0&u+(EgmfiVodP
z(-h}?ptiHQkeH&8Tory}Uj({o{nVO^<n{IfgzVD(*Q#6w28RDAtTM9v&wrh+Ff4yD
z;oDAM(B3H#%Mjb}Qx3hYccFNL@)iBl2><$Ip+M^Pge(?}=bc|hxTc8A{2W6J{4&y=
z<KB&Sdm<_eNhBu*Z%PDQmaXxhf@Kh4BQg=}+_+o`Q#$qNM!i^BL1H<J)#ve@Y$K%X
zNNtRcY>7bQo$ZY241x#aURJCxSH`H(=f}ffX_)up`O09fuI%=d4i1-m!-SCecoQHd
z5IkiQFOz{hm54P_6WL%h2dSzx?v=!o8;b|;eqRcrOi@vkV*!#lYNAmOmU|Zy)^w>5
zV+Mq9MO0Rl&zkaRb%Yj>j@s)SQn*K2;VEuEs(8PhNXzzKL<ST81atD=^TY-X<H)f^
zIu@PI@#+S$%JUrI{)tVO>pg*aq4Q-x--A8uQK6D7XgN_}f-;c30nn4RF3L(g$-W+L
zuTX{dqxVU|SA^8mw#|#nAN6*4_yU4&YdKo{S76XH{g^Hh0cYyLcOYDx-KnH`OIZ4&
zT_t35()n-X;I?RAaCjf62EjDx;5?Dx(Tr?k3^LlyF0N^9f8AYakWGVl^isrOV;wZ-
zg9xzRd#sg)(Q9I*RQu9!%?Spbx-zfBl#3K2<QocEJ8KPn9(g0-Y*uu_x<FXSQmH0?
zgRq2<l7LB-yzzsFNy*%C`|BG?;VyWDiF-j6LN_g&2v8gFIlv>kwH}SlYMeOHg5$}d
zO7t3g9a(G}_;(?^D{JBK-+%`QO;#olYPY;enO~4eunJn3nXzo-Lpr7uJtw~I;K1sd
zrP-9&XmJeKdU5Xra@{33T*i)nSAGlyDLl4mG^X_S6lv+?1$+UfoNXoh1V|KV4eVq>
z63I*y-i8|0D{u6R0@}-$$U+zqW<K*pFLl7Woun@d90*q9m9KPsxGn&pTT4=m9Kj<y
zF%PR($Bs}TLjdLZpV_!&HAPV|S$(Iby`3!*3}WnN=cF0ldd{l^v39xpcp2i^N*k3j
z)6#j3HkrQ<#S511j{)P1MHG1G1Phel^(hJ5X!GXqWc_Hx=O*);yA<!=6{R=PAl}0%
zb=oiMh%WO)WRgf_9X^_(T+kN6c?pYr84$FY%$qK~ae;J5NTfVx^_90fhQV4ggNG%L
z^y&tdUW2ZWFPW~vlak@COIxaa4drW~{<t3fXi`Lb`ibYt2f5QE#Vcsf+iG<6;Vex3
zpXoyBqKNr1mR|kyh65EJOa)U$whgT+VichSYlT3@lbUhd43ODMX<S>t*vT0%Q~E^s
z&p)svL=bFIhI%O5$jgNWa>g>_q$pr?#NE=>TbhI9MiNJ+UwEOWHfjzlu(^wIHP%+I
zUWi>S`G|7K)m!EDjDAgDEPp~<7j4f4RGnXUk6#*m9G`+^?vAc;^?=%A$*Yg8fiXVd
z;fK}EWwCUR1xxOZH(1jucJzGKJO)qbP91DluPi0q6Rz*clq67PZRan1dmCjK@<nWm
z1%0<ZgL$L8Fp2NTqkS&*V*=u=D4yWI5Q{kjHve&fo#)B)m_KOIo57;&aG$@k#C<@=
zaLed^)$FM4cn0vqT?$lZWCsHZP~ec{Io~ZkZc^7L0P!$E)X^A3XR}U3wk@=Pts()C
zHG$OGM9@_$)6h0R;?p8*WIGM)g}FYk8<m=HP*ntw(v_H@IB~++nEo?i6ZFA;QW7LL
z{Ac>4q@IQ`rM;qP!`?jpy>e0h-kWn>n=@`(lcS@Zc75Di@7+q{`aM&;Il-H{Hqwkg
z(IRutI1}i-c3OM&qvn4wXbFpaVb>(tt*%YEvcj9UD3_v@040NXD)Zj@h1^?=S0!is
z4Z7l&m-AoOYX+A8IHj3c|98q&ZOd^(2*LM7H=p$|zh|?x2-27=IH?7Be!qiDj$?8t
z#n7J$Q&Q%P?(^o-T(+gC&E?XB0NA^}sC#2;=jx=o-1SQSrgJ+{{^qb8MMsEes+l#z
zL%>tQLZJ9w1o@W?XuIyk?q+izH(OjC1}0()D|_~OxV&6C^;*TZ6ZTd+gHI~?zWr3@
z-Tz*@z5H-g^Zs5?=Z*V5eVdTnbaQ#5?g98Xay@zdzNh5zy<{BrdwE++jC@!@<B#Q-
zOcDZA9lzj7UD+N<2^n$7j`iq$!6+-ndk7*(qL;N0!FU;d$RUNU*7H0yS1a;wjp{bP
z2{ojzku77I4c0D#N!Liqf51tdhwdlWP|Nk(0J_CXB+(=EI6f@m{arT@6qf?qpG{>^
z2Zzzz1w0(C2ju=E00*>_bUSK*5LPUVsbQYQ>@P<yR*>h;6TXmr4oc4TyYc*e;g{76
zssRPdLIh3FJdbVgx7yfM>{=726QQUDmO#Yi=M$F^pgS>x!xC+5<3&LDFuA{|pg6J5
zeNzuFA{jId_bkgjUJ-NGAAVCdU?7p*XV95WoLr20LqQ+caBF8$r=Lo7ESi}zh&NhE
zJIy*L+r+X&0I=HvvyA)An-EazBKQ#z|I`u1$20EFcj?}2*nlJa?dz1vUuD<{76K!I
zamyS8+>1-y6-E@u01^eEY>%lJsfhxV6eJ^$0)0w}tX7cA-nc(%N#tYT1GVGa=S9;T
zqNAwLTJH2IiLl`;0+GLv2G3}%l;gvV(=jBcFz)2;?<>P<X^nz@(sBri9}_emoFFmx
z$2VCs&A(BJFzON8>V#GKpfU~L;9v>X?2H&OQT9!0-jqgA{R|GaK>lMD1PKipx>;){
zjpUeZO7{ES75^e?aR{;l+5^oK_Th>~G+=Q#Y#&o0YrfgUpP2+o07but$vxouOCXp)
zl|!JfpJIa7>Ft%s4N5b3!BV9ZPX~i|;gqZq?PbkPSCQxcLnRWY8QAC&K`ljO0o}E_
zKN{i#7NQVu@euWQ!{KP7zyY~4wi3f1^Rp{_#JEQY2?MClq=iYz(vlOUO=nsv{BI)H
zM?n>QTUaAxLVrQ0MMQ{51s5VDgl&6gkyw;_b7c|JbS+B0{34on#I|G!T4rTImg3JJ
zUE@Tc@Wf5-mC!s-xPF0Ck=<*~Yyn&sC<y0)CaVdsP0RqMF!;d_GD-vRws^6+zmW&;
zPR7?VI0rAQMF+|j5_(N{op+Jo!a-_m$pR19A^@8c2~`PzwkBL)oyo8MElP`PunyGW
z&;g-=a4T#$7yt?wX33zp8Kx=t%TOwj!pSBof=2+x8=D^)$6n2~*$|jW)$IGeundlx
z)*wP81l502zL=)|FUH;}Sg@^$*4(yj+qP}nwr$(CZQEvT*S2ljT^(^d;zsm&_+Qua
zTr)Gr%rQRr^710>ZJ?s{{l*q9Ifa;5pYDCI`EJ+OEtP12wS&r%hS6p*gXV@NH*7`{
z-~zFc6@DQr{0*lpje00dmD%PIqcj=mcD=$7bPmzIanQHslM(S^2TjD<A+y!h+_ohy
zD;)*ikR)4&?3HIl@c9Ue4a4I258$%UIXCO0AQ8mwQb+k+BpwnOx3!4g*U%q`DBb1Y
z8qmAS?==+3kA1bj&XAotyOab2g}J1oG;WUnWS|!LM-(G4qjaAs-uoRgWH8g9pnI-j
z$YAC3&urP3Ibe?L7BL`e9jM6qEdwUT(sfVl)2xt%+C+nXy9)YwRgu4WrUf>wEuQw(
ziF*%EH%{EbvDV49g+H6q?i@BcIe5GyAPLUd+?Kq(9UCV4?5uYL^;=o)IYq*8&nJEH
z$5WT$@vZuEFV0)z^X^}Jmqm4sy8q^G(Gc5{aA6_bE}X51+k?USIKE_v2sLXLj4*fk
zt#S1DxRt9gPjX1&V}Kd^KE3c-NpxmspEEcN3=XKj!9V(gdf2X?@t|lDNNhi9dr)F5
z5fYzf;|1^%T}~l(!^z9BVC@@`l&+umXbbS)%N;JSZC+<V%VYY`sq;XSYgEzU3lyaU
zg;01F`6deRSNx#A7WpappGUvd@=@XQ<j1Fc>tF6jPf8b^`Sb6MqL^v~mgvv@E5`7G
zGhG$j-|y+Ap1=B+3)|!@#r=RxH#|81RdbmC<F%dRUxdT|D(RG}?f$<8Lcdr2L2=7y
zO+5d|p@wi*>uwams4}<cB3iHwGbF&aBPpp~qqE<gVoJ`U3og-Dq(ze3o9^4b?ubO}
zp^4_n!}X&DWdqJh%eqdoW@{oAMI~i2)Xdt6?htd8MVBfTX+8dT=d>V6J)K^ua-GJb
znfI&P2Jc)TfroKbs-~ms_v2=VkGIjUhkEt%@byP=`BDt9#3-gx#9_X$y-a@=LtGlM
ze|LlEn6<BPpS$aGvzokFay;@tLks|75EB4VY!Rl3NoVGpb0KOUy9fqV;NLhMdA%Q+
zG2t+RyPSV>SEK<-u8!;IK0&15PRm&|aFSIr=)uND+Kj8J{T5ZO)mTz(?e$n|ogI--
z<3s}d7XUg7a@FZ?|J@rnLn0->xC@FM2P%Z!?OyV~{XE>zktO$Zd?2dsot=$_Z(bU7
z!*^Di+f1X!hFrH-4-i9Xe@u?|^0w|>ADoyui?V<D@?AVX@}^=^qv#TD<#3{|aToJz
zNI09u!Gz}Rv(~yKH@mji*~6@|im8H+#EabQ10m+jPRvt364Nk&0RvbCw-cfZAOzXn
zuG%)Mq3I#Ny>C8HFy={47=@e)-vDqWaX|Qt2P<}mQ*P=kxXiYwD+BVaUD|HwY&5S2
zgaVkilPe-#uWNDM?aQuQs7&&F>s<}et=3=i$tbwXI2=gQq4&4pI>pc<*qskf+)3*M
z9C0F4BNH*X&mvWYs`fx(RK&)~p73%adl=7Sx{tl)F}SWVGSkA*dN{Jk<0*`<aSf(&
z<)s*=obil#go&9_8`|n%QrA|hDeUD=#{s|#0#~O8E;h5iqz`Llml!HG$W_`YhOLf`
zzzjHU?0D-xR{I;d&gIqgEK|yel$<k8%yfP^*(Fqm#BIeo4eza2_&K=lfDROwUldHi
z!z}&<24^P<Ze8}!DnfN&zJrJ~ON9B4i|&ZyAEAH)$)&binijQ1u;SmUjq7Tz*3)h6
zjn|U0=19MXu|R(29Jl~M`)d%5v9#*s&m#d?d`+4B2;iBc<3=bH6jj7URM`18OO1+!
zW7av+lqAEDBgv<3nzT@xjGfgtE7c5fU^&Jt11sy|njLq($U(k6)I(;da_*mU58ZYs
z{yWht_*-)a5nz$U!v57c_w#!P;$dM*7lfG65_+%rq}x&XBKm+gC2|x0`7{5A8<>gl
ze}{^vYD^{_wITM-ssFWIL)=CYM9>ie4>z$!Ic$)-Xn;$Cr$AB#WpCeC;^o!9l`#pD
zc9lfrL$oGJoSp_33x6c!`uZa5djEO*+V7SR<%pn(36+AID-V`|m(&ZXnt+>I-hU;S
z9S;BH`E_1*di$0Xk;M6O@j~#t6Gs^>Wwm{OxkLFEJno0!i1?@0cCO>HTRxHp#V~P<
zath55B{<=X08S~C5+|pEMaG$<DzG?@-VKD&5-A0JTi$FS4(0<;{-UmpiKq758Sg1E
z4d*#CtP2@drmUysW%|lJhYM}xIobA#{qLOY7uR{(+!fVh#0*b^t>)Y^@{U0{a|W5D
zP^RQ!dPoILN0!qC*r%*~o3&yXRgfbXu<5J-JdpslY>P*Z!8*sflS{d1pdED<W%o<}
zqh>Izs0zsbaB_<19kBf{cQwk&nIQ{{N_KUnW<;m!IPSJ7j#$-Qmwxw$)CuBXV>aF2
zEV?kPu5+_;rk2Sn221QbS2{+0GwwZ7S<zw}Q$e^OZFgCI`B^HS9W{U7Ke)A`h-Q*f
zTI^-#U@)yRH!1;Qp)XU-d&x_L!wn0h8k(zXvB2A}t*%i+TL-}Wn*op3cjQ{k@j>ds
zmn@TIa>cgV{R>_WsG;I0Xb5A~ZHj0?xsscUFtyx@lY7ft(A6Q&btg%~;3qD+OE}mV
zV?PlpF6Vlc%I+;GMEbOpfROWdV&4Zokj`2Cu?vTP!WW7TvFi)$G~JJ(HGO&mXjCXf
z<j@`Lf@{d^1%X*XXF#4g9E$fwGxF?_Yx#m+V&kzSf)Yssd1<0o)orbPYu*!xMMPL{
z6vGvyLPC`oi6mV-qD(1Dbp)ig0xKz?{JPH~N3{6f*_40Wz$*r-JJ*O)Rj6dtN|;Q9
zAxhecC=7`&KVTFXbk`WWmY+8z`Xqt69ET$J0pLP8L^Te^Y*_+6&_amiK}bN_2)uo)
zKPt1Yf>ea;crqsxJSu_&+zYdhQ1BrEp<LvhwVi^73Z`8!jsbx4Sy&i0?8DLwOyZn9
z*QKkUO7JZ!a<hxOv05bc(4Me_mIq?R^-Ri`N+gmLZ-^$%l^Z*QVurSeSDscjkFi7*
zT%T;aBT0m$;uMI$u*UG-%!=~rsJPdw3wI=mb>drU+vs^<F2gNvDuXjQl$wnbXLRJ-
zz!~ZV<5@r3$<RG3kVZciJrNu)J*D0r4Lq)3P04_1U|f4KadhG$%!uIloTC^rI)O@g
ze5e|GPkst4f^{2a*?igehHh^KYGM`{(t@CYD~OCiE3<`h%h?8z+N5TL#n}O#E%Gb|
z9!5l0X#lUo^aFs)A<yWL5x_9t9ZM)>h3J?yp8p`t0W>>!r%I54++fH<IILDuo6A(b
zvPVK=qH#Y>A?a@wh?-3&)vgz`XVFUg7}og2H|%bqIj09V7$n1L3ata|dXjLAESYP|
z%cr?oKC!IT^SL>-D)OQmVv*K-Ac6BOL3Vcc-HO4VDvJl#o=8X+D7bu|0cv>#$YD$z
zM`0XC(YBMw*$rE`3sTXsgA&U{Gn}A2uZEok8(BRKB%To_5-$ba$DVBL3;#~nQ1>{$
zn|o{2_3!^`;~#uJ@AJiy5*<F&abFbI;%e@FgsB+zR0d$X{${-d4IU85@KzvxrwnAW
zllqn*KDLs(u|5RwV{X=?NAnGyB%Q4_(t=<B5dM3(zMg=qEwryDuVcxu7ocmb*_Nd8
zz?F!%;_f$jPg6TaNqp<Y7*{Op)PUanh{Y0Fh;j`O$Z=yUHZ+3zCXW8M)Oau2aiq|+
zz@lV$6c`gxpkyGv{lqF*4mF!VL(9tQZwoy-$S8QG<I0O|<%CiK81$nwU;(id<QpJ;
zPDWu3^eq3EFUwtAjRNTMy>~p6;(!5Jh7oH%Dn|^JexD2EKp+DGfgoqw&?JE4D(eIG
z@gNb+*uWRZlepKD2R<BDJ1I#a?)UP1T%P}TYAta-dNR91qf!A)Fz;9~L*x&Q7cdO?
z1mNSAn1Acc7K44Lmingj>U~KdG^-x$nRZxf-f)8%fRow-m_w?AGhqh)LgCj)Ppe^8
zxS}JyT01Htm$<9|l&n`S${@X$$9WwK7-%>HH<aJ`MkM>UF1)LFwEDiqJ<T)b`wanp
zmpd<93DfkOWAiZPppC9gQ_9DPIUztD9h3(ufmM83)aaNJP?2ZD>>kRF0LVr%q2(Wr
z(Y!_|bQu1x!`*yeOvE#?IeL2_Knk;@2kTSaC?lNGz3IkwWashXLLk7y4d&VGJsEZ1
zzH(HRR>PnnpLVo{aL9`1ZhPpI1R@kj=O8qpc?L7@{;K8r`H&FmQ!mIxS=xCXfDV{j
zUwTl0?xAttx7l7Qm3c%WFt};S?@eGnpoBOH)vaa>xLYFV{1$@%&;!Gt<pXC#ZE-x!
zo|x^NJKQjIsB$#EXjzvAYMnK}J+2*KUy3*Z)8m<D?sSF2=XD30qx&Q=hC>)I{Ri|v
z+aJ&;Jl#PGvLaI$u!q<&bpMWt4ifRJXPiJvJ=BT9BJXWR*DP~-VB{hX;|NFjXdYv*
z+ue%Kzp=QnhI;Ob5`3$r_71R~{DdzW0;EM8soPlWQ<YfHrDI%)8I&aTl2TM14^$+H
zdef@O1;ln-HqGZ43-lj@l+5kSPyLMx3i%CtPo8awp?5}%??FU=1kT5S<PPam2UFc~
z$2wPPuUto){6o1NzSkpiCc_xHpVfq&1p34&ioo(8OU)JSj)`9Wkxm?<K=A2qn=`Fk
zdiy4R>51E><@6W_;LmG2AQ0whXMe0FpmrivrgzSs`7cTs;7@8Y<!_$Ch!fvK-*IiX
zm|xi&X`{sPZ`>UTLDK2(u}RcAVt(^s4c_tr#7?Fo+$bDS<i7gYyWPF~9R3fux5kt1
zf3`K8|9@SKiG}%p=5o$xN;_|LAog9V-=S4W0UwGc@EjRb*VMGHbZ7r#R#JMyW5AI(
zj*fdwV0yl+mIs1ykc77-S9P#~M-t2fU~TKW`|r)4F%~WC*6q~P>uob;iE~b(sxN#H
zB2h_|5UG$YcwzfSc#45^>XAl`y<T3bjMZQFKX-pVcW<up>(+Y+9s2KRLrNWP>hJu1
z#5ew>=LgShcXz+F+wJ{+f6-s<ul4tFB_H)r&P2$fn4i3+^so!9t59dw0u4l!mrSEZ
zq0VlicUAo<{A{v~$VFllqWp(vHcFI;FrrEFlxrf1USa*?Y1?XeA6;3sJjQcgyk6w@
z+H2=v)822Uxy5(mdL24lHI$YreqZG~SXaE0-bD%1XDjn-6r*q9pF0{}HVjE``Wv#n
z`7mgAXRyUrb{>4db5=`~SxS<K<hpZdwi!ex)U=+-eHG$pqeG*c>rlXKX5BOI(Rt!9
zbKnL1aPz#tVpjj@wyPO_!sF+BMn3~}Q3|;a)OxwG@nmZI#Tkt1`n-0t|A*BLeNEml
z8V_67pCV7O?aA2ou(o^2n|_(EK;qvC+NfjyY}(ci*f!Aj-$)IuyE4oTgx71c2S093
ze27ZUDXsWcez({rdhFo1itpktINo~;x_|pocw4BWk3Mbk9AVd8E)cWmKT{6BZdwl_
z@@)X6g8hzlWC%v9ILhmQ%~1}1-sADdD5oNQ*H{Wc679`(WE?!~+RmDyl4yP#4^{^D
z`Xtw2ir++6LGofbGu5W?tACxlM)Zy**X5FdLSwKrbuWm@y=651bfd*spX?N%ypJ8n
zw-n-z{@vm-0PH3r=9yNsW~WR%v(1>iTOJl1OI^^m;hY|DR=$AKhKM#fa$2<=KyPJj
zY7H#7|ERp`ArQMMV`7a8D@aCj<JBUgXg0CWOW=`Ja8^;22r#%M(t=<Jx#cnnK@IS|
zS{J~{ZGbETv|WH^>W~ea7!BPM;J&)4xm0W#W%RX~rgh70MRS6{u&{%-1NINniH#co
zQH<J0d{VphN~iFVVVhmSaiN^w?0OX2CmbU9+rpg<uacpP0+x2q87KU6Hp5|;>B=Ot
zmes3Y@Jv06J`~;z|Bo_>h9H88xM*%;O%U#8$_s1e)>wA1$j@FjA3h7Jh2+j!0@%(K
z<3YTMR<lqD64U(`wwLYrchp_DZ8r_zXIac0qxvFe=u3_;HwP6}Q?BC963fXy9=!No
z__7_dDPSh>5GGic@DY|td&$Mre~4t^W7(~&k&-{9O3R<_GFZMw_1&iGq0;tyZwzK4
zLM;Y)kVI`dKII`#=HjBdJ4ZAokw>zFc9<a;nkWP#{YR4Ur@<(t%V<5Fjg*ULxNdyb
z86+ZN8;}uwET&r0?GuPbCkO4~mjkBIB3MnRBqWTa$&V>`K_n<IynHlSLF4HG0}3=c
zR^AA%S}68{apD&4Q1v8BXf?E~BQ_e#@ScV5Iq8cu1vH%mmhyOzQ>84>Mqs+E24zw#
zoOf1!?dcp^E4R)amE#3w4rUK>5CP!=3N6~+JUpl}T56+)QOZZ*G6F`kIIL!mc6nyL
z^73U~hKd;`trZ{Ep;mVLMt*I~l@C|S#zxq~Lopu-Z@SnLh}J=nI|>#TLqrva%Dp8!
znxiCz;au7)Ty%3b#{o&;Kv6xM5I--z0Jz*c$2n%Q)6CE#j)!SYCC~K$6L_Oip<HtS
z;xu*3b9BfpAbTdun2C4*YP^lBi2Q9LhtD3Gz?(1ENcCJOY`jhqF8DQe_uq~|HE4%Z
zD)8-51&du`=wd>oCddE;N7IYkAe^~U!r!-x4k)(pr;NxoW_y5IlSo+ms~2%`PZN5#
zS2XeVYlD-<F-a+oIUsQ%s)2*MEe|;eka8~A*4bgW4u!`vYfXZzxH;h9yGPn}%1A=?
zz{NR>Y;LbTR)y>h0d?1S?a+x_;WTMTn!o9BmFWaQQ*(sb`PNea<G3L?yX%MUY`$nN
zcy(pa`wa6zzcn~_g*X$(PIo)r>tdGBC~I`rwz9zI9O?Xzp+F3AxKOXre~@?Cg#+cq
zSh+Ko%{|7q;sc9~Sm2rh2yKTxMN8YMuX$X*Y%)ppYPPW8-&z12FllJQqeC88FLXLk
zl}t|QNXayX?-lRE8l9o;5tO#j9oVm=1nt|<e57O-kz!`gL0nj`9dxk;V@(}mCBU@0
zp`E`{O%B(?;PE0yjdq`_8S9(bb30QTywzb}@navM8E)||<NMEBQX$z%eC%xB{+5PA
z{=t;tS=0sugq*C%R*nlb;*~l#&xsj6%&8cC!Uhb^Mw1oGhbTb6j6PnD*`!+su+N%C
zp*JDIq9Y%7!iJKVkQMg)1{7-fJ2O**&9)&d)X`JVKp<IpM>>|916j}BCU}kXMA_)a
zzzMOxF+Y%XDuUQ=jK9WEV$&9%yAPDnXo4~*gE*`47St?^IO{P&fruFs5c6pv0E{bJ
zfnH?Q5=X+5DtNYZz?8tD%%&CYo2%#qNeRnweT0&yH8{w-w4Pb?+Gc6ahPBAM@sB|q
z5CytDw?-O6j`)R$$(QV-rJ$4~1Bc&hHv~E$)CGwF2-R$2BPR*OEd%ibA3YY-%~4Wu
zYv@8T<@vQrKr=Q(0}N~;h;=o_gn5fjGr+0nq;M-AxFX^hogkRIo1P*mkwz|f@Fc3+
z7b*|w!^+>iBbNK!A%q0cjw&sg#E5gbF)--b4+AR7R3Sy(phPta0;PtOCY3qVOqm*p
zgV~o4=an1E0l3JtVBHs!iolFK12Z5wR%pC-_Bm*W<;RH_0AvmRd;dY4G7~DiA-W_X
zaPkWE018h^lFp<RTk^xdm+6a~cd>bwQ~Y>w{c*9h%y~PUND{pVHmr<_`n(*zYta(Q
zs2vDmvVu;oDJ}`uu3hbdx~>0>>o1>?7bP44oh4p*oN9A~Cg93(jAFnU7~%@KblwDs
zC>2ZivP`=Flp=@r?K^lFkPmYqEVj5+uF^E{fdnbJt)9p|JzCN`22ZLFlrA_0-tIj9
zVKX=IQFHf}9Q(8j@MS&?R~7Il5q5tq^RctCO=ojVsnboL0n+$Op62fo6M$K`_WSA`
z)N$n0FJ8U#TrO`4w9x(u16zoxfspC`gWCDfHwqAMTHty)bGyQMM$P6=WBJwRa3Lkm
zxi*ij18~W0iWxGhpB9g%6v^>#@y4HWNhU)clF8_x=|26z+&X5(?<dj-h$=D!%)0b_
z$j<X!zd1d>|LW^l{sUsd#KQhRZ>6MaO2;4m>+9y~?`)R^+Qm>vrG(_N2cP7?*8~%u
z$RLt%EVA1omxrmRZqC0xR2rj@manhu<W7n;XjUGnIX7|G*H6OX{e6D!?i&-b$!t~X
z*yh4xRkE8DEY(u7Prpy{BU9G5`}2SQQRYl!MaV9=R6PM@c$UlQcYojCAN6nW9+;#b
zxO%-QRMUL?e0)Bhp58CZ{pZIX<@NLXegdxeOlSti$9a?ycfn0`SY}QL>leEeM}E+e
z2+`lgOe$+a1vZFmsSrut)YYN{NR6pdPl$@9-Z(F<1NrDx3#{z*z+<RTb={K{0Y#eC
ze&`$3)hLt{!IDy5uo0^(5iIGTRn)VF;V=!)Ufr66<u`8Vw^w3VR=)GVy?ygq$i4il
zC7N;UX4s5~r&R(wM0LwdGwt)Z_kMXBETWD$U?yrPlZ*RsU}@pNKX_s3TiV;hxfP3t
z9=A(#E!3ChJbRE;nQXL|7KU7m*t(}ZmX}6vvqL-NVwk^=B3fN(B6_p((%R~wi9J>?
zNUfArkXlrisNJH|OcwrVB^BWKfYf<9I_yU~&RUgSJ+W-q<G>i?N!a_S=Al<wRXzLa
z`{^!p{#oimxMY5HFfL6>Pu|WwbMkqVYC(*Zu-3EDDB{(yXjR`^=!leuZEq=RY6(Pm
zVPDjY&|E!lD-C8~ezEwdwxjQXR{x&*6os@t#-8oJ_fw2)p}Ks=jG?>UDsBEwq{}<j
zmH|SpM}6akhm-L#P3je_zq~eK=psO`ERxHU7l@Gxu+EZz1U+IDYzH!TV1m9Cx2K9O
zVcyVP`?d(hK?S?o0T<r<j$3Uv^J3hgM!!HR7Pzo{uqs`0u|Y3N1%|lAsVf6!v<P#=
z4v8yGONe3MoGlw(y8%0(_rb{I?a_fs<k@A_eDG@H{$&89eifMm7d;?zE+aMn9kwQs
zKM_VN8&|xgjd*TYo(<7i#0y`LE02@#(HMGg>v-~R?mu?yl*!ggTEl2kf@-Vzqxi+0
zu{e_Z<YAW?k|%6j-xZ~LdMT6l^$n*28iW9vVeo(mYeTUXNQ;0R&SM!)+|0z=1-pS(
zKKv$k#8k@!Axo%KiiPxP<Fy@>mXL`gY3<Ke%ln=}A(%2;`kKTEV>sWPYmv*NO0uW8
z0u6pR%&>T(PrmIx4k8`&g0mOGosG*)iu8xlZlT&HOC^mIvZMWjBOu29u@_ow71=%!
ze(|VDBNg)Xq=3+|Tl<(?8$B=IYnlg_BhwI!-XL**2)o-5NA}Yj3Z6%tQUVf{_tesU
z!oGQ|T(olFR_p1Ci}O;^9V{lc3%K0DLBG4ax7HBrcM9;U{SxN3Gw@6jy#q+fw{$`B
zlLe?7{g+BUzP`v1LLz~q`kw+#L<=>mH^KMXxlxUk1b;-HuvpaV2%Ng_ItO(h&F2%4
z=Jt}FyZ#&a4kHc>6z87$RH*IYnOmsqx3?a1G}QqXSq&dT5?qgCbN3oZ8O?)=lfI+;
z$aAhJ*-7FFHU{H2+D)raphY1F&h$7P%DkDU1|R6gA>SrkKBAg{%EoowT(-+QKZXO*
zz-EsAAdrF~q9Zb!-GY__gfH1l^<KoiDJN`y{!R7nct$QfEbnC#?u{i+uGB7N<8G&E
zKSj>{MmT*~5ln?5rt$sGfo_J==gF@8I^2zcs;XrcQ$mE&7S)L$Mwr_7hb<cUCHoKb
z5kdX4=J~70<*pCx{9$!_F1(zW?unR%&~at)RPri-KC{~sIelMAz{p_QkS>`zbKl}b
zVGwL=x1qh3xG~NB%|UAfjz?>7jB&kJi`M}&FPGG1L(=XNl(it80suKZ8wB9E{%Mh>
zc#*oN7MspxYxg=BgDYS!IFFO-LHg!Yrai^9%D~A5Okc=iA%V>B3)?PPllA!WiqDdu
ztTf>_ROT%=AFOLB`!%m@Q1_EX3xN83$v<8l&8|z3m2jk=yATKhN^=vXKz+}F5BBES
z=&H&HVQp@ifhfLpLl`8`sW-1Bn3hrsK#Pn3Q}nHvnl=I7Eox0!t7l=VKqq#Euus0}
zZd1%C90^3H{>0>c^FUPWLCwiN45=KBilSE`tRM9ifydWLH6cwsQL9Ee4oM!E)@y|G
zT#Y&_yVQ^Yh=BWHEJ%)VzbXr3fi=Jp09NiZh%*pP-!fnzERVK;`y0SK=ey!-y@*Jz
zMz6#Dev7RqTWptmhA-}n@H++sz`#@xbU<|5b`RgoEHWfqf}B(s9>T@QwNzy(MR4DA
z>Gc8aC|RbY3d&pqROpNG#6|9pr<p`%XG||NML>5mTp*sH`JZF%q|r|84>XQBW8Yw;
zPKxa67a0JX+;GOV#4h|_ZhQ?GAY>Q^dPX)9XsHp)^8WRkMZ`G_rUSEpn0;jd9aa%#
zV36ZF2kVoV*c-MPSH|dzqKf7gja5;71`K#G4ScHWs;rLoSin&#3)k}r-fx7X4C+Nq
z<*I<wgpF=?5S?O+18ymHX_w+*#40(_HYNn4T#k=~&zP2s0oZF(%IdUz1snDJ(xw26
ztklw}HKjH2@w6aMkRK7mX9&)=zB_Gp>5bR>+g)hP!5771kGV`@lOl<epy5?wl$|g(
zH1F@r%^p!@I&X%JEQnzM^}E!xxg79K`B(NK04Oyr6;lxHRG2jh$1>HzzE9fx34_0e
z+}vD$%`}_Y4A%`YgR}2rsg4LpcSX_u`cL5OA2ZOcu+M^%Tj^wW@2eLyr$hrzaqU7-
zDj{;c2h5!qQK99>HFC(*69`ufQFnkox9D<N!4X(siGL$mIUSw<egcW4LmW->3`%h6
zZhd!CfP1K+FK4^?d2+F-nc0<mNN)PUZ4(^CV(14kr%^yt`d7ENm0lg=?$QcyTLmCC
zN6|=6Cpk;~wJ|4R<3RJbI$^$l5yqo){YVAFD|knwK*ovemw8Kl|Adq?v)c*8jrB!O
zI(Ej#StY($wFc-0Esxusklw>LNCwC^7nx-jlO*GF+GKbnR}(>M)8`!ZHl8)04Cp<+
zROxlQs*=g>f%A)9Nm|BAo8UT+UFY6-IFPy`wze>_h(&G_kzmw`BQGM50)SX~ucErN
z=*~)wyG+Ae@V=vQIg<21`XdSJ;u|qPF`O!sx>vVvfiYqEv41<+M?UeeAPcXuf%ztB
zQPtoM0}JTs9)gUnZVgoML?$p?*Ec-6UXqfCnwN3>arl<40||h>Wq+RNtO^K{XTd<N
zJu{^yxSj!BP0G5saz#Q#feJ3&RIO3WFc1FAISp^zT+R8}OmhEAwg39~82$?+>w0GN
zpS|CISV@?eIhg+Ey#o#D<gGTuzT5h}nXM|^2!c17d>n15$;4}o4Jup{btq{gH-vk}
z6Jzyil^^e}1z0%FtXO5K8N&duc?%c+ZzGf~aK*dT>i#XgUJue@CJNEaQB8+#conn~
zNwOjmSi8$TVGML4)YWd5=%H8o-JbQ{&(G;w{eG|Shww|?9I<`*{;#`-neNM7dw-vM
zDn)BPNva~2U~3%gQWCqasJ9b=D2gye6bbiInh=?(y|&O^_Y<#j6}&8K%$s^sydN}C
z#u#erE=beK>AP=NO5q)MdoJd)GS4APG&qkD72nOQI<llZ3v=Tq4#djaw<p6JD-NZQ
zGP0oDfK8K6yQ5)mg84-khPNmJB-6~=l6WP>f&OGaFRl$K_L95>8IFWP+(-fT^NWCX
z!u@uNSQqsrx}&+j6nJBiWM=W<FJ)O*!G*EU1^AZtF^=COIbKA&nKMqt{MlG_DFtGr
zrf)w6d$WVQR@R9nDnzlyUdeUOj!s;FcBW`}EIV3d18orIWNg0f@+wQukKch6sW%a4
zB!EKDY#HdG#K$+0?Jw7g@Mtu$F-$AqFls;?ioH3N6U3M#p(u_j2{i*k?O-xvQI#_Q
zR>ros+oQhH%)N|CC@<c8Mv{k*6C2IBxi#(QPJpubF3j1A3D7A{$3Te@(@blCEye}h
z>>k!>OE1l?b0BE(Zd^7bRd8M1?AnER0<0f(jEA7+z)Pa<Q*M%kOp+HK+j!?Q@YAlu
zR!%u`6r=%!C-b9JEaFOZ6F?1G%@*^R)_{LYR}&Tj1UFa5J_T=A2T<38qf!T`fN?#+
zolXx~Z~|Fl_nMek4!2}j?4OZgaVic#6joH6i^<7J=T8$y0t?IfH(ZKy^DMmi5$-Lx
zoFM03l5J1Z$YQ@Ou)NPcKavI-gE}98jC~?giBc;;MzY9a^nhin`>JV!R{x&4D`NV&
z*W0=WS91=`&9ixNG%)77c-wd!T*3d&QQ)Zgf%X4{+3(Ehf=lgkKt@oPPB6jMf954#
z@*vO1uWox#X=|~QS7*-5E_jbxW*TvSiDZEd`!>Yv3f27H@ds?!bIPH_$z>i0Q=mcD
z1l0urls@F_JV~EokmF=7rT`lZ0U{s@L2b>!B1(Li1VZHyLb`wZwI@=)E4}jZ?;RWt
z+GxCkG%ofB0ic>fPI?JxR;^GtI|KWzbKS6ubTFr$aG(^I<wwXEqJl@OrM4%JLw!R3
zx%T6f55ad`-qqaB1r{|8olp*I#oVxsYs{^~|HB}NWtYcOX?3yz9?nXnrg$(I=QsY+
zi#Q6=$k-8G1hB>=Ym6=iEYPMPRSTU&E9Ax$RlwIdn>^1lW5<&<WyHaJsZ9qu>nt-`
znvjG%DbMF4XwYyt)Xb$x4!XH{(J%yY!m9)uQeV7)k8UHl@fe^<DED}9P6bESP<u18
z=IacVZp}O?gQ9qG&yonk_iu*QlK%*^h(U=<6}WT1@K++3AmCCKs(WMcB%jazN#q3v
zaUm5(+`)r@L+21CF@^#JY>zQ}s$vQWgB}fwLP|`@X%xGI!=FkX;sq7wBq^$LAJvup
z08V;zS{vac*Jpubsgf1r7rfk>AtpqpKsnuucrO|Yp;bT$XHOYJ8+`WYx|(JmF{LSq
z7w_8|vcP{tRw1E$IS2?0u68p*dKeG99vIG%-?!<QJ;&q_jPBq<AInRX4x4~TauF{H
z9uq<oPw>KvQN~i4TqbH)p&ooxLcu(35~Z&FZn40ezm>w-%MhNlz*c)%Yv*x_5L^y`
znnYn>TP|$&{{sim5biSM+PQCyqmcAvBCxS6<fSvZsCp9HtW(mbjL}F|CXmBmgX}^$
zjgl|F6%qx8BS6;&9<cxC6DA?PZt9HiqMaSP5Mi$lO`fSvL`#{O-aonQnA?N@<R`E!
zWULtjdX~Uwmo_IfRk@5>9lU^x3V9Z9_tiO^vLf3o`WwJVbquZ=VQ(|f_s>j2ft>(R
z#nVLY792X%c1QSZS@FKo#IOgvERt$_1|y5|AT8lDyfxiH681>sok39zOq57*_YR<o
z)8C5xS>ouQP9QPza@|+B=5L=8%1pLhwAb`m-i@qVky5`025z_&P;QWGn7DB#ackOX
z4T{O1YEDfFO)yXsXrBc?Mx&%?9{M)&E<X#`;B6g{wh01UL%AWf%S^9reOkgo(gbJr
zEuUzxpRs#1oTvj@ep{j}cnCVB*%Kob8V3J&8o+6VTt20eu+mhs`PVsEDOfceQ36Ko
z=QU!UEy_}<)q$>P2Pd8dUD7g?2oC)CZw5ry+w@A?B1SO4Fa5O|FQ=V@c*hv$>`zaS
z#PRIX11dSCa#5TKB6Q8cLL=<aRokfx-a55;`OOtfAq!AR!#m1K817=vM{}bMdRkYM
zz7wp8{TK`bq@MK_+1NU^?|*Z0f+tj~?M35WV9d;^q(c1nqI>Lo8LY~&(0Ch*do?n7
zV%{KoNkdP}H+wd6$tLDVI%t4b%WJJ4+_EB#QK!$tKDzzG&wDHtqG;YH^0iN7HBx|b
ze<-8Zta&*SDM?BM_o*nPt=o{E0i<dk6?$BG1ZJQq{#@4AeUB)(v*wEV??f9P5AMt!
z%JsF*`^i-C*{n+#m(4f%^G)1HaS^*-)XtUD!soUtoUMxeC8+$KW}9ALst=~8+Mw1D
zE@by}_he*FYF^;<Cd0C<Ffe+^J{$KvRJs`h7J0frm8D)wCih$*sKjwHHQob_&v5|+
zxEil{URNwahi9zk(4^g#u2ngSh4hq-rMY(g)<32_%F|cH{cOFGbj2D8V<)Vc<m$C6
z>Z=$6CHQA#yUX*UmLpz{_*^_uWY@sy5PMZqKol?Er^vI+YRiX%)9TZ*cv}yJa|LeD
zqyJS~zGl$L=cQb4Pw&T}m1kg35MmSF-*zUrsuk7`DrrM+#F9h>OC%VwHU008Rwkx0
z;y~BR2<b1Eg0uOshctf8@xA?Of+SGThzJ02eaDXbiUE<HhvVAaTv2|9Fr899y07Te
zpHm{i`M?Tgp`INP{DEP93~BRSX1=d$_oORaunPNkqAv})JHL50-Il^0(a1xS@4E>4
zquIAn=mV;dw|e-{0m26x>Z|^KEr?8DI;G>~lh8Ho*Ke%w&48Asm*3FTW`BVze}LF8
zC$azP&shIUn`8LjSIM?Dw(Sqc5dV1*2HWQYTHEorNZ0vT!S$VB4@mf)3p=<`SX!GP
z9)(dUFLR3CH@Ul+sS|nDDe1sb&Z0YTW@l&r%#PcS6n{pD@9O$=_jo*$DJsr75zDsd
zh);=dF~*Q{Y1I|IHYz<l96r?7??B|muZg5F5+frVbn5y4rst=xwx>f^deShvb<T&4
zuiEeQ_<p~9oILN|4?ozm+uQd!AJB174rt~{)fT@yyW{h|p1(-D>(bNzcoObu{GUke
zB$dXH>Ixe=99>M<FU=B8>A|kt*cJ4CpUNb0<WQilKl|d;TP%fJi@XwvBA6a)6%L3*
zhGfzUqq6Ny&m-;*<V5(l2{L7#9ZCM2I~IhXO9QP4Nt6k<{Jgq!FsJwSZx%1l3g5-o
zZnBF0ds*zC)L(aXUG`7~5&VaZGS~_loWz|1++-=*<w-JQVi3d`Q>ydl%hV}bn@!g0
z9ey+yR;n<ntFu6dukygv5<eq74p7EdqW-iXTTO+%P_IAPGT+4LeA5(w8(9dhWKu0=
zKt%TI1CF~JO>3`dOg2Tj5Pw+=q5dUY&g{b2dA(2!qDqJ6+0}SWjEsy>Yp;ynD48_s
za!O&r<N9?rztP+&y2X#yQEenm$TD74$*XsiV<mW0TzGYP>!t{UKPV@`KSa2%Aw-7)
zI3QlQtPsDFB$axGUAS2*(b-_RiEh>tsswLr8mY6mHE?tF-%N#93btLpT39Bast7(S
zJ3&ocd{|ZB(Tf`BuEXa{w_(wsu#C#~Y|W!jWQxw`of2Q8noFh?ybw&k+ueCLVKBq2
zy1K_lqgJTH%D*-3w3o_*{Hp!+4X2hu-2A79?1e@oVurw!?2BZg3!Rt*1SDi?D=t(X
z;bK4Wb&kCQE)JA;q8DdQ7PdsiM52V!u<OO7F={aGK~Yk*<+sfq?}dwdhnC3RYqY*j
zsNLLFA*FXu)9kd9lv~l%>C9ARn5tXkhNy&%sAWdKCn~dys8W$@eWSQd+{eX~DmYgf
zaqbVUdL6M*{CUt=v5WJQ&6Eojqy*xr$Z)LhOU&b?3A_Pb_8d8dy2<p$;@0>VQ|s_-
ztW7{Ud+urDz6F_a)xmj7e<m&I#*{zZY{bdD<zx5j=?f^#0DFU#e*-OP>1}?4s#gN<
zj*bCbwn9<cVTP&l&t<ji^63>CQFINa^EL!Lr@Fdm#S!4)M#JcYl|vO=&BP@_F3Ji@
zUXvB%^bp5~8n4?rR0S;Zv8F!L<_CqjFI5G-nXaU4Dsy4t^(DQ`Q$Z^`1!E98IaZi`
z&aoIJp`h5+7T%=U>c;9`EyZd<EQbCN`JMRlx~!1-2Smh>Pu$d?oAJayb}D67GYN?&
zkA(!BD-O2OAJH4)T&gxNp!HXZJb*et-HF3L$n)-nOdRO>8y@7??~NnKMIY|;O&vgM
z62;=XZ}gFRPkH3!Hzc~d&Cd%F)&FK)g{g4abJtH40a`9S2~(7O%LTlu37e|DqDwng
z<@I*$)KdqoTi3MG!TnRABi}5jXH$M)65Z1s^bsMLhz?}BHBFFzG^MB@0FlMC*(QJj
zWPE4CI83b{57!3_rL>%jGS@=#`gB{NWf#Hd3R>OL{Y?ptmhJ|E<fZjBNjT#=0zWY}
z*c&g+4K{EFYRbj6-JL4&O_<eYXuS2%^|}M(7w5jJ83?=o7R@*}l$hf0`;MCdqeuU%
zfZ8~JD@X<W+9}L467H)uHo>P~^qil*9mIFHPB9`;4AL;z%-Sw8a?o{W`uSe_H6q(|
zaud;90rlzM-R`g~b<L-?{-I~j${TGw9(Z1iKA>Kji}}>*wvk){D9ROYC);V{j8SWt
znMZZ4TKPEK>MYCwsY1Zdf|-zHfoiwit@nt4R($<END5E&5N}tP4?BF@=;|_$f^Y4o
z6J8CWEtZCX3xR=g+$BWVjnCre<dxps7xx(#Fb(7r@jRt?ehvN#&mItr%{mM){-Z^V
zPC&WfARr0QGrrt9eoy#s$Qy;bw=uM5_H-;GP}9M+MI*=`p4hQGX^<z`EUKNwW!P*6
z{il^VK6Q#mgL`fh(azgX#J<VFd9*Lh<be4AdCpWNIb8OXHdo{W$Bh}ve;s?qMx8DQ
z8Gq9(PNonlI+J;j)j%c#NUm=3Tjv_ucUx7p9d!Egv&pP2xFBOldr>LX+s%G36Ip;J
zr095SszOJHHHk+3@JAZdqvdRZLqF2V@d5~;S*laqbTcIi+I%i^vMd;@X`Vbs>ES?g
z<Fi|D{rz#n{X)%GjDJo?DyNh2pP{0`ARy{9<WT3RDMF(AdW0>ZoL=<-**@koC{W=`
zlbq1K0a;MTeR(?eItT*SkcqOL@P`;LI;)8h^Nwn_mHf88gEC1UEFW<%Qvh8URo!%Q
zj=8RhfK64uwnvn#eA<7R-HPJ80F)_Tell%!o*lM^=;tN>n1&p+v;|}xXjmCOh)zq(
zm9GHLYu;mpOrZy4Qh=f5mBRbC{0|nOCg$})09+nrU`|?r-*g2RXSfCrcm7Somy~c*
zU}Wt)OgHq9yixK{hg+p9EHtbEWM`lO2}rA(;);-vpE|%6S1gUUXWxa_a3d%g-QRHN
zvWlT*IpeGyd9xHB3fU49-~a&pXu}9pa;kg)l!P~|-=8Vt<axK;S4^@CdNa|Hf%mXU
zJjW<x;j-vLwx;c9ej9`aluR&ceVE-t5g7xG$U$5-RU2VYBgYK$r2Ogd<79kMcHmZJ
z9+p2LYj5y(s`u;v(Q+`d{NLbE6;B6K0(yBPD`gj3D0(>pMg|5bdNE5UXBPrac19@r
z{{t3fWM^glUnQentzCO;c9?%wmBR67QF29o2|&TX7SR<TYy}qF2HG>>6))k2m75%t
zvzOJ|;v5#fPan>FdRi^*zQf4BuZbw0RtRjf3Gg;W<L)qT2)HE3K5T*`At-mcG8BsT
z36X^70x~aDc+LAC4i91-e;6FbxaB>Bz#A(Nj8#-rqsEX;aZL|_&r9a_gzKD94w28l
zAmxuL+#QFT6ZDV|N2{Y^$fXQ>p$*laB5Q~Q`I+n#YKSS|6`##2U~1$U>_pU-Fxiz9
zIm3wH;K+udW_h5!G;z}O2`F6=VQ7`m2NS3SV~1k~QA1+@6Y0Acz#LGhQUD0mQb>R#
z>ZMWQ6;Wt}2*%}3l`}eLi2(Gd21IKmtI*3t(*L^5Bj_Fj;wOr<3Yx+w)q^4vSH=N#
zju?s^^+aNxP<bcjRt4lubP+gt6dZZQ$~qM_bebShQXesiO+j%oQSO0>jh0AIg}0J0
zlv05ZgW{iMr*y_KKvVaC@Jk$AAkQoQRj6QIr)1ls^{y`)SZ|v906<F!t}ixlauI;y
zVu-Q=@F6BA%`*_4E!B%AYetd}i4^552+)io5+w=Koi`0&sKXNk<deQj0??et6J$3o
z(1Um3h>(K@<5bFpSuNrM8BFC+HH+GZK_+@Ttic%ShDp;|sT)_C2?chJlmZa~fw^o#
zfIjG*g+?q?IF3SuZG)&FHA+O>lK7)ag)NCX7?Pc-0xTrMQKdGL98t19Ia%XrNH-8j
zP_mnGvbyWXu0j4a;5G7eFg+fBYbKwsHrRZ8FqV^(aRQ?lIdSw%Y#!lH(2wWsaE3^B
zsUHEY(T!Ja9?;gUZ@?RTys0n1GF;Z}9pb%f1|r0{a~{Nf)RKp%##*mF{d)R{?QEE}
zN<S;}==SqG=EMAU-JgHj;rPDlv%(X8|2R6+T9#U6vC@;NPv2j&GgrTgS*7>s`LloV
zu>Mfpsql^XlgQB@Tr`cE7e=s*x@B3hHrfz~tAnw>I%KO30ifK*R~imVXaYKaZCDiQ
z8=cCwlP%N2i8TwtX*%S7$CMYfC<qE)LbUDV)}=wixH;;jt>}HxRuoM8t?IOr<NtnG
z_VaUYb^lu9*{0*0QD24jTIadmW0j}>_8|=G9ZwpXQCk#LW0Oi?0-((jsst1b=v1B>
z=e+dQ;nU4x@DOgz<JqdCLHZF%#1cz7$%W1ur~znigZQ1tKnIT5_`-sD(0)CVP@v(w
zFc@jb{l$)Jo#7nk)|BFwYm+M+#=TBP;=bPq-)>plj2g(>Sj9~Q6?|AQRR>Iki60+X
z4kK(c+-4ycf{Zp0V*3Y;76K74^k5$`VM>AQ)2+vQD}J)}^9ztr!}{dV<9h9ZDHc8u
zhYTVxHq>^w19JRLRC>JV5i$DYu-DWd3#V5Gu6x>qpU=!*GK0cw!A^rVegQ#i<)tiP
z#v%vkh3@ZgF5~CJ-0=h0#Yw_iq-Gglx0<vBH<BQyG(5BzdgB5-0tp0I3)pwiq-F|S
z!J1k7swP{ouRkSj0SmM%d79hcR`Y^Qn!_P(9dVt9g-+yVB=t#Uvjcal1%Ik9rb0vu
zp}jj^athHRV)eneFu98l-cC>*?21~PovmH1n?aHa*%qKY7f>FL;MmP(nU;a=5HnWL
z0XG4~AtA+$2mCpuJUIK!bqp(u&iL?};$yl4NX&irNkv`_izpt^s^Td+Cr}=ffwCQ>
z>DO}_r@S`DORxWijC${;<s4k&vNP^48&v2B9O}n~3j*qqN4XDma+{7uuPaf7$(^pg
z<Ys^w(YuMWB)p3Kf~~s!$HSYTxFzIvn!-VCusDy=OW$kif{NFwU@4B6DRel~`wgFG
z*)pa(+O_3Zk#7dU($4rXA-&2=>$YzkGamc-04;tXHhe@)8p1|y2931epDf2;!~v9J
zq@SxI2|J#sqki+zweC6jJ1CKO8Tb82@quJAkK1Ndme-^5!P*UwNp2Ssd#tP;;A<gO
zy!B+fZZVi{(w02CI53hCK*%c6zmErI+6)qnh5<Oo2G0sF1@4T8gfzGa4~Pg;2oRg(
zY=Ob1r}$_gt!nZa@Gpyl<>=n>0ph?o?z`m*PlteUEYAzG^IpbD+YD}hQ5f;T#>MCb
z+gDs)k#uK*nRnhkPY6M(f0N0%+LC|~WZ6i08)?)u;$`hW6F1^m%E`%9g`KtHZL%Iu
z(%I{ucHyhvPp6lU%im{wT%UJMdFS3wm%_=ldqr`S07ZAKgeKGK(EB!n_!>nYIM}%G
zIx*yN?-RZ}z$KFNKta@Ngq=?**8|A73Zh5QE)zX+{Ld%|Zg6%LFCtfd0Xg?~AO545
zGO_*N#q|FxmYkgbb5z%ihHd=6EP)@t{-NEA0<-ShOVo>SqE^*J3I`FPB7!4nHgxAm
zT$}Pv;6Z!ymzhRg_Fg=cx1_bp+;wwvGxI#n9365FPtW_&gM&%vAlIVAG!?RF*C7+4
z4v~y#lSZkX5*A^onG>Odr-yfTX4xh4^tnE-J`cb5i*R%1(WTMqdi^;+EuZ%zP;*~i
z9ReB>n!~>gV^<~6IBIeXi<mt7AswWWVGyg6M&QobR_12TQl?#$GP$eFMN7wJ8YkNP
zUDax4n*bDHrXL$eo8qy*26m6zewQYqS|Nk3dVe>$`OWOK=hI5f%<+aHh|iY`pO?}+
z&rfwiTQ|Erw2w9^E{<Q`x9o@sA&}RD<3XqPb_eSpb1h|czs%GKrc|~5NN9bGpEPW}
zm2dq_K4t8NzK+4Sr26k3vya8IxG9Tz;<B{3>gm{TxT@kigIgWb6w0OTkxO&bkmu+y
zlTbW!u-i!n0{->WU#|5B?A^%@iOAzd6$?o^o8jgUMM^`6(GZKs4sLm@jK}5=9jqd!
z#WL-Hj+~CH0JJU7n4DflK`BZ^!A;qeq`(}44=zej5CKC~B8QRuv?4!;>_MU=P4cu~
zX|!P9NupL}6ZK|q4ecyzhyxLfq<|#1#djC_<VKA6r@x5P+@)~ERU8oS!Wck_$YuNT
z)CLVl;%~ZygghrAteKWAl^fRGV57@1k8K|{Wi8Oux;yxlKM95HC-IrYpB~(%OEOa|
zgTjTu2$V>zll(KUG}<(5SCZC#dd)NDLXZ$4td2-k8j>KEAq^#*NOcwjkP?E<LWb%H
z1xtyIPGXVgI%fx{FwG^T7+QI5gbQhw6>AfBl_Sps@TociWj>G?oGu!gfBo$Ah>tkU
zitV#)5svVZx#jCTE`W*4r85+=U+=`Rw=a>$4Ot>y;Bz=D=pTfTH;}|eTHsiLC{~J*
zM?V|c6`=i*?sokb4#aZe0eC_H8P3a~J!qS!Zxcg{z_BgdsbJ}vs*kFjow2poM74RL
z!2UC%j>4W_GDEhziDkxb6P;a9%6jahdS&BOEv21wz4xJz)<@<XDo_1kNFfDCcG9Z_
zQW!rd1ZQlS#?EJrR!DLE*hQNbZhI5mPOF^Vv-3Gl-TVhW_MU!u<F`GV0U)kcO1UkR
z7-nctqb*>i|Ja+NypBz~(5h*~<wV6moOgul_-cQL;y?%{)Fw6z!UK)1y;jRSnGpH_
zV-^#~ntZ=))MX8g$EMo)x8YH*i!SH=T7ri41{7q@CQ<@I2BK$BOIiCukT^<cvQ!vT
zJ2xD__NO|VO(2TU2@XaT0PJE@;Qm^qs0mbN%_d=R^2)sWRj@3xCDOHGeH6{Wnr*r9
zDL$0g?Csx|)@?O$>2|O!NG6C9;<UkgU~z8SmB=4}Sw^Q$Dq)pyo1Tclv-kl27o4S0
zdk8n_#Q`~6m~qv8_`RlBMSYOO+YJgF%WUxtg1PEMHIyK`FMWOT+fSGGhd`w{bo72b
zm$X2FaEGEK!cx_yE!{OoFvT*}Gz*8gMM>MF=h1d}P)sc_g8~Jnfss4n40Nm*ov^Sl
zBsfbYX$s-KLc<{9=^vh%DPTc+FxW^~#cgC{EW~BRfZgCk{uqEMjuD7S9b9t~l+uki
zC(jH2oj;@`6Ww#S^%(G~eW3*RI=$sAzQ|98cR|0C5v%MNoi54G8Xx7E!mgj4@YeF0
zVAVg4_C3b@B5!}Fon*H(;TfPv)!=y&{!j`Aoe`FC$A8HGs;i}Ck2Lw}EAKT@;`xWP
zXs5@rt^wV^az1Qbmk(aH2wpc=-Q@YiFmZu?gZ%2?eB}av+r6wjq}YQdU`<{7?=kM$
z$4m#XsCi)pP4If{2-E=jK-psDt6aPMCU)6nBN)Y+>Kn_IHlmYN_TyK;sc$xj)PIJx
zxBm8E??*g^h8%Dgm@pM4kVL}C5I{^SVB)0E@Y=}~=$%@zD=LJIDC~NlN$BJOpdM+!
zoA&UUYzQ22enu+H8rE#%WHqy5JKsfOQT6b`2X)P7jsbxD-L_2G(H;7i@fluiP`(JT
z>nOk0L_t>2A-4b4&y^wfrmH7P3BpbC*!Mk{{@xVp_vdgarsUq6Ks8?Iie;K-Y{V%!
zeU)r1#rAD(J=*Gw?=Qj#I0U*Wjur=7ieuu?Vi{E8M&K;)z{=II@_jbu?UlwdJK<Qo
z88*D^`nFkPUZ+D?$dRw}|HIyU$5Z|P|HD+0RNk3MC_;8d_GlQ{MUt%?Bzu!pD52;$
z)`>$4DcK`CWE?wWWY1Ez?Blo|FM5y9=li~X*LC~-{=6=Kyx-?^US6-~c#Qkwet$d<
zr)U#hUJqTeYy$}|Y5MYf-3fWSz2^>(tq9|vco<Bpc%85>HTqOpCo^a(Nm#*;c{m@D
zf2>GYk$azT<+!E(?N@4oFJCktmVC`_lB{p|JSIng@}$|NoG<y*_8ns862$P5`uJhz
zCqfh6c?l90b;S4#H}+z3UGCS4ANKp|K4iL3!O2jTSntn0JU%|rK49Ve`DbWC)WR!C
zR|=1llfjkj((<z#VI2y+Hf)|BWFJ(<Q@?R;bFd1xqm<FARmyFwrRln_L+t9b%bbz2
zd1*|?bhW!pW4&Q^e7n-ONFf0uw9K)>LcX!F9k)rHL8`Q3>S($`I=ktDW8F?76kb{h
zk#M}S&mz;-7Ebx!WGGMdKk&dZEF@T>UMaaiuPf=T^v&(-9(yRY!%sKw*Z9~Gkuq~Y
z%Wf;Tw?t^8@W}jk*2Q|$SEX*ae}xAB_BBHiV*juIz(UP2V@zMc!j>P|&_y@C6LJRv
z(z&Rf{8SU{@TAuB{Vs|<q#1C!i<UdAb07^nUg&Y(_{zAxm3KmL`GH2Vtf7(~IxR5@
zt@5%BOD8v9OxkOdg}@RjTk~4tcRnkb4*acXg0wDA{8s7O?1s|3v#UBSS);{%g{m|?
z|IJ3Kux<w{mozUPO0oE#38|S5_V%U&<$g;)Oh06=eT(lCAG+(Tm$2{A&!08XB=@+g
z)Sf$6tENRx%>LM-qfw2r=s9sei*ByGg8G?DDL=)%nV)^eSVDf;hLgnF*vCszWboI7
zTq+MH9`<n^e<?WsG}h|?F-j$fEw|_zqqB}zOVF9EW-5NcW3lWssE5`Tku^t}ymjot
zgRy6{#p8L=UX)?-t&{qk9pWurrfw|n+n2Mgjb@$*Ds0en6Xxs(s(Bo}vy*>HQ+`q}
zPCITm*41#%_)C&wiSgK7PTW!JfWEzZWdGP_#hDQOfF&eD6f<0Qa>J>@<H9emnORw~
zG~K%nRMBPcm=$B%%taI!RI+rj8fm!s+E4c8vlqE74@#Dy;^}hin1as_$QjY!;cVB!
zbGWnwai{XJkQaTlQ3|4bMv)kNpo7dZWG(1d`{n(E&mvd+rcY4J@`vB^o<4eTFm9UE
z(j)!7hYTh0emynkhH`FckfQ4Ki%R75r_Q2cQazT@&;K}Suake+&%q8JWB<poi?#l?
z^QAKbq(`^9+MB%=*j_$b!G9UP*XSen+~AKdZ-S54V29OXkGnlOq#?zyP8rm;VAb`<
zBDwd)?=1pDr?yre)eXxC&@kTVc6*fQc<;213{{Jlp<O|gg`db+0!Dy*^SlSYg+u49
zD1K^HbkI{p{%L7t^{zAhN37AegZ0rY_-k?+va_sZcI9kxxPcj)1GiiQRaoEpF-05n
zy&C#4_4?=L+g{h#_2PCiA2RzAeVCIuyI5v7F6B=T@J#w*BSg7HgWN7W%ac2%_*fO~
z-nUYzmceoPOZr6;lH+-6H;x|A`W&^$+K_+2<i$b=_UQ`gl92V|o2t<r(y3ZPJi^!C
z-nu(AcrnFI5HCE_^5W*LPbAKVRW8y?tr+J!2Nu;@KH$|Y%$y045%#_G`YK;H<CGG9
zUw0_m;i*GEFL|fy9T=1}8W<ZSW2lbxKRtU%p=$r)zLT5>^7e6m(n>iU{YQl|k0;Be
zwU2c<52TIfo+TbLM^X2#g+9JZf~8Ekyuuv$l3eYM?4x-h2MT;%TwCv$7dL88{w2DL
zX7iWnm>84VVjo$mKYP(9>3f9H^oT0jPs8dIOYH#78vawYhBQX;ZqI%D*zZ;5NfV}L
zo*$>#lfss`)+(a)x#+?o?sMne6Af?A%IGrGoETKj7)g27e`s<|;u4-Tb7K6N;Fy#a
zYrz-iF#9^I1CG)Rp+e6k*uJB?t;r`GA1vgAW;>hq4$2FiQdBw>6FnS!0?X~5Ekwq6
z_z#uDj9`t${;od3GBS<%O&6y0kH*(X4;Q>V5%iGlR_LeODIaf;$EN#MuUiO*TagS!
z(+?jzAj+yPDPqQQ>f;6Ky^~I2t&+ixM<_cFus%w)$zxn%79!rew6~99{|8bENuhc8
zJ2`_%j$Q<^w{?xfDUv}k3mnG6x6fUDIP@sqVNYC;B#9JiT&$8roE6u5`IH<-o{e5c
z;^n2zJ*pSwqJ<3o6w$Bm1!!xk;7aSyY|VXe-TD!mvOG1dG4AH0;%wtd6PLvA!?wSO
z>sXXDQzQR#?K5*?D0kl{d_0~Lyxk*p)>T_gy{Yf&-o*K~@<>P6UJ+4b*{-lzS)*uv
zcqYmuruB+WOp{)7*o9?f?e7<jrS%w39Ts_LFw@sox^7wa<BuOFWS+brZ9Xe-Y3Z$b
z>)ZA+*&bXV%ZD&G|MboxGp0v9T>WhiVksTcOU_!}7slAU68lcg`!w95?YjDGf?QxU
zKJLDiL&ML0a~@mMOW#OYV-D?We%U9zcE9V?O6#i(niB`-*fmJCLgn+`A3Mj~d~$)C
zH0vd2cW=1To`lyV6Vi8nguTd|>L(v?j63s2YmTa5icmjsyXVcbw#_4Z&rjPavANXj
z%j3K-el<z(uov^an=yL*W}WqHN{4L)?Y@<2s^T*5Egv|Js<J6L`-_Aaz7k>lN0*N6
z(OhSFCjq9f+vRq5gUL;LFAW@#ip!<uxVEyk{psz+KEBP3Act9;B0WjLS=Tm0S@#K%
zh;tiZEcOl+u?PGesOG+U2gMp19<4dUea}ioajnqeX(z3Eo%L~Q>%I&rg8ZHF&(z0O
zAKCR&TgrLgBVK<b>iJyVq5DHGvyflISv2V$l|zySTMW`q9D|4`N6*N|M?7g0gC#CC
zQB!Y|vAsLhuzp*$$%T{K?5rcDmV<Blf}D~OdR8cI*2&}~x5$w1ocrxxAN_Rs-zO0r
zmu}iv2j9^gNj-S%*KGLJST@eO?@iB_Ito;IuSa$29$xN$r{u3YFHtt1>O<0zk#qt(
z$#Am8M<XWZ8&QM6Iz@Zr!N29^bK?Iv>b%Q=-vAQa!rb+sppM%u*Bw?|#qOq+#X&(m
zYn1EllS1bZX8f{+wbgCclS1&9vJVPgw03p5V&Qzr!Oqda-ooDXq{KnNOAfXU&N_}a
z&4E={uyD6Fx6oF;2=~;u>0)zISXc}>Vy*obt$t2iSa|ms4nn%yx>`7&6jZXk>1v^1
zVeWvkI4CD~(8bl+;-=j}&#%)1<Z-miBl2cwHpTP%ysz*Hg&w?BNlj<?h}oNhhnh0<
z-aWGmp$AqDX5WAQh|E@@^Oz#--dlSWj*?wCa`iQL8S}|+{)zqan@tYx4!tFP12%J;
zK8Z4uvn9Fk<08$PLLf?${Qp1y$18A6K~hrk8=>`8gxsC`SBOpV;(`T{RL9PK+(Sxo
z#x6#{xY%z0G<+vlj2*YQ%Vbl1iI_i;;Hp7*z+iIO*Dg&bwIy77_Cp<u_jKWqU-Pqb
z(qoTBie0A`d(0zzm%r!@rW{(IDI8*(tMVr<C50CJp((dM{+{^m&X>29b*JK4|FAX4
zP0@XKef_6bG>>kII(GE%9+ES<ep}1k?}g2}jXm;h)ruS{w!{`QwtjVJO+8!7ZIz3U
zp-XcbtKHgIj+gexSmiDA+ngz^e@Pq;q|){pUYuymtRQZz?SIw8DCK&4I^ScqU$SxM
zNtP?qy<gtyHYQ;LspLF6QZxqHN$$17mweW}5cLpVFX*+5m!2!_$WS;=_X2}&Y-AF%
z?SbhXJ4Bp}cWC(4XqshK>E34-yIlp7-Lz<oi6+Y0*cybT+jM7SW2c5HDpn7WJpa{(
z?{PY_g;OuLY<{MZBsahmOaEX7Rlf1#o&5HCYj&kY8Aft*I@dgN;zckE53hGuPab`9
zjz!~-FYg%MeXweZ8?l2$;BXaoKhdDdMbqO{$86P_I8(-3e$h?1F;+0ssL0vV5j%@A
z@mrthq3m^Ol@+(`xpI9aJ=>Z;XLGJXUQAw@)%VwAb9|W&2}$w=A1)U>M{E<X7(;dX
zu=J4yT$7HLtZv)aNDM~A?rUBM4OjW*Y)K{2xiwK%3pPl~?Z@eKivZmB2b2tVH$&5|
z;M57%UrxIVSK8p^w|GpmUdFtawc%My$*=FqG_ClVQBrG(myIJKQEg>W9q$|aqGxKC
z4Qr6Hra8IXZ7SGF>+Is<y*kTWB5H91mR&FX#t1WC3hQ>J+5Ob==rbia^>{J6TbZ{i
zwpP@f^X_Rr)c#=29}ZT<C#IHv=gUF4<v~7@d(^3z?6E6S(pVo6s}`BbIP*~(tW9QF
zm-LMQ!w})Gc%SX9RYQj|w`tu&Ok(MLv-&z{@%J_8`1q2g7U{alx8`->$ZfPg6ec4#
z@|pU9+VivPz4Z9gOKdvt4Dx>@NV=4UaO6jyd4+3&akF|;xc%(n|E?;W?9=)TzSx&P
zg+=ym5LL8_XPppC*4;r3F`c#c9z20KlxthnO@jAWpNRFQQM<!?m0lV<tlOEcFS5L>
z%yo&oKaaknGfk&j#Qhu2<8B4~df#A-W^MSTh2%mEg?ilC7pl=bBNX;!%UzXf^Kvou
z*I)9v3xkbQRs@L;`Q%Q+bZ)|Q3Refc>nv^ICYsKtxZyS^2VNIkQ;J4BTP)oc_?ouQ
z^uxnUY!S_kjbW_xgtIb}v8H|@hvLK?2KQ3Qv}G8Vh_pzI%~w;&ua9E{%YL<&`)+Dt
zg*lcPB5_vS-_y7>p3zPl7ufZ!<aKCLa{3V#>P%ffE8?U*h~MP5i|#J}(9Riqx}wS1
z_Ez^@y=b7GaR6nO!u7x+mv7-LvKy7;thPhtTc+C^-C%!7&JEm^D2LL8+N-aW=zgdN
z@tcf!5DDw+72pk-Y-+@<)gdZ*cW}xDnnMNrik|nw==npFoEOM4?Nf2?s#~9uNBz%t
zWf*S`de04j?=hJAIL&vly7%1V!nw9{#2U5d+jI}-UU_qlKZHu2;P~+#sd&F*wZ0XO
zF8xsDGx~G~E%`0iakd^_y^KSGU(5!D^Uzvj8`w4&TwORblX`_ts%GW>kw@&|65kSB
ziaq9t;~1^|@`5d2b5Qt5Jrl2F!$Ny-63L||kqEhs@53T};Hs2@rSru<Gnj-?_38S#
zIR(}lW0pLF&Ei8R5({5k3f^GbM{?zY6n4}{XN$1Z${wNN`3-l@rYop!pZA$;Q?j9+
z2#A`n;dZ+B20Vs*=Z7m*W~(D7tU?Z~yuJJD`_aM7$A!&eeF@&PMcgd1+LKaPvx8`B
z7SG}P93Mz>i;nNxye(M1b}K9CgRVAW@@FTXh!6IR?$aq^aKBK<9^ug{ySi_TO7>UV
zJk{z~o6MjZYwF72V%WTq5CUwYO5?%xP8}_}l~09i{^t?zH{(6-b?PW=1f@HnVr?|+
zWm$qoyqN4ZVM*hes7eked8X6CXwCW<;>N@onE%s3JMXXEt)BAR1eXfGt>fqJM4>j8
ze|A^Qm2Y;vH`IIcV;XEhX87J-ZT;3&$Ooi*_uD#c)b!@1_=?auERN$x($3lUf5=LL
z809Mb;0FqVgtW(O0*YL68bbNtsFB4t{i=Yi)%}N$zy1&iL2m*gjle7EBOL8`+x`zu
z!OZTD)ZIs_9&qH7<Q5!H2M_)*7T)*vY`DbeBN&B1Bd!VZ+VS;uh=$h^CSM<TuX*7~
zC1eY|H#D)KE2R1re!js+n5zoiRRwNn36!smR<T_91)e`w^er(>9cHz`t{mRv%7qD8
z78xaO&P5~?$9)ngnO+1NNfGY7zn8q$^v#_xu*<^XxvQLF@3B4V9JKoW#N|x?b%A{(
z&oTHbTr|`5NuHx24>lUgTH&v_+{k)<MS2YD8PiklvSuliCcRkCKGvGT-GNhot`;kR
z#@jp13A<39m3~(VPHj+Ns~@yJ<n;{~h+DOF7ltbaEaUA$T^bk7BV@%0^WeQH@Lej9
z9IG)J6Q7dVe9lAKi5?xHKk7YKw)&V+MB9g-hD$SDFVo})P3E<<_O=(7M7r^)+gpeH
z8w85KzhqP2hN)=@_cq6i7us~&zkB&sKi4u#JK3q@SyA(~ivuMdR9fRzl!CvE50ISa
z!o1L&Y~amQUu&%kX7yWx%zw45qHMou1ZBQRVu%%f$O&;IYB{kUL7UGn70*6cYL%Dw
z3=LRW#rPNthT)nZL9t?|hqj5Ukp}mA^$&k7uw&YO>bp9y8l|IIG#biV?Do^W*CLkL
zdpet9wFBH^H9%k(tICxN0YI0Sxwcn6AnDqXA_#;_tkl!}iYDGXGs@)&2?_HHik1GY
zIB9Wyn=zcnpnj0FJKc}HIgTS$0h;e9R)!G&H7mJ08o~tzDn1ftc2x@}<rbFv3xdl7
z2WC2x>ra@(k8}lhXvxbG&xZ{+ORWtDigY4t_1O2*K`D6~*l1ZuJW{UT+hCNi(12a|
ziMkQT&%J%lP2Tq;D9j0xko?qZOn{d?gh8xTqF37max$$b$NieJ%bl)v1|r2+)$f2}
zD79{a0XuF`libtfq`|49wmEvv5xXmh`L{p6>{*NE)jM)R!m)6`JxsmSedeo*00S0h
zwJtlWDYULBF%q;^GRS9Ax|Wuj2x0g-YTh~(cfBokCZ`AA6eq+k+-p{m{+iy5{(4|)
zlAn~YM9-9_=fx{$MOFtqp!;~4eGIm4gv~=(rGT1jz+(z_!vj6E3dZ!1)e|Cqs*T;D
zm7HoFznT2*_L8!Lg9w(X$+V6&xjU6--BG0{Ox2Z)6|t&iosIT?*#(#j)6@$1#bdy&
zyZ+UWuJ@Z@FA8E0bFEro8GIuHc@zDK>v?=VP4AP|r95U&$S+yp+^)}eDP+`$TD2ss
z8L+Qi<}6KI0h2n$sQ*goNjS4KXKARze9V($LTo;nFXIBjm$;`>&b^9H3sC4v&Me<(
z=srb4f>Zx0jg5n}Cx_?IWe_krRwQFR%(}g%9Ewwam}o&*TY!ZOCv?oyXvv_*OFXh7
zvcL%y5=S96@rmP_AeOUX1^{8I-XXQTdG7XLsh4NZ`nRMcLe581p9Mvfza1=Dt0fnW
zx|p^U*;2x2Jd^%n^~1wzjc?=P<C%*3?KL$u?-uGUe7TSn!8$F?h4q|w>aj1GVT%Be
zs~Z{I@7ke(ULa?7ZJX}Rr98gf@6yU4^M>;YPJObgm0;&LPVydhQ0`aTC6yF^uys2p
z^nTXu`fTyaTwAA~c>r|RTyS6WtaS()HYwsdWRifewFT=)Na`)QIzBj!{S4z-XPAuq
zfeMh#8IMH~wkp19D7g0rOk8=R$vM;N+;cNnG4a-~@1G5gu2kL%>EC*7Qg*lUP;7)I
z=O0D4R08)OMy7uE+@o#VhY`ixmkBQ@DYlp8FX7aOV+`#>_?`x_et8rjPn5_peCCK{
zxG-N_rIP_B7phGAdW{q-@9#H0kZxS!&ZEq{?_DJ;aXy$fGgJm@1(=nR0H&ly@z+GG
zDQD@Hj)4^RaYj<5cPCiT<x?Vlgf)K6?a?s7@cyAYQ|-wkq)9%DDyaRlkXWyoIArCq
zTTzi2J{;XnZv17*Q`2g_R^UWJ5@pM!a{`c|E%!8%ZekmBjNHEHqRHu9-v{!^sAa7l
zWsDhBJiF-eiEVXD#q*dF;|$6jmbJdZiHnfB&d2vrJ#)WKiaG^Vk6X8~4tDWtz74yw
z@;>j%-f5I&O9CUK3^+G`%SwK?36ozvot_aoLpK}n_Xsy2D481alr_>K7$RG>J)l+E
zq)~A35cyTE=!sn21`mokh%66x=80|;CcnEni);85%Wp`ZW=DSbc+~5#H?UG?)ws*Y
z>mP*dp>Xhy+DJJ^2_{`SUqv}m*OjKj<rB$FLL!BAbrb&bPETuLBe*)4DniD+B|(zg
zUD(<`20{DI9^K>J&mP?#D0Ul<jGsusgd_+Yz3j<5vO7KWCFMPphA|DF$d8=xgd~8J
z9sqLy$)3ahMD0QQSIUnB(L}a7+S}WYAI}=AeWKR=_E=G@&SKa_hLye4e6{rhX~#ux
zKP9kTy%{g=5Z?cmyQh%|S%STKadDANZ~Jy{j>QkdwY5#F<YWx~`-a!g_k~Kw13Miy
ze2DAcf?2)$o$4O=dk=bjd(fK1(^*suKvR31L(u3$mPC9PpKZ#)MJN&!EuGt6tCSxl
zCHanOAH*lPwyP)-!wuRf4q*La0DUNC6gxJb>!$r-=L@qa`0%FjHe}Ob<rEp{+^5#R
zhc!7IFi=(>XXYVnZ1oG3hw=K6mdaKq%wd`zFt9YgCcD}%^2FnQNP<HofzVLpcYc)W
z+hyO4b)^UvW^C1PWnkfl<-XB%r<)%APRg9SyY}|&m82md^V)>FKhgm6`YqzQ>Ygyg
zW{FYwNjQyu8mSKEl5u%^62qzSb+ndiS;NL%OoyyD>*l3#O-F8%&ErT&dLf04cPN*v
zNj*?*oVqsm5N3nNa4FZy0ET>U*DzMV_&Bzz9?jmU%;i#lD%jRZqrPQNQ&j^!9F+21
z>HP-P=@~x@^k+@WiH2zOlFlvhM$PZ(xh)5&W;;@~j*B~?CB3GSW1;To2OCDeG77z-
zI-S!9e||mBen3k2r1OUE+43tLNwVJV4>K0xUo5+T^`P<P3$;wiBdr!$UzIKSh|g0x
z#ozb#swWdyWT6s3b|LDa=49I?su{g&c!y|hHk|qo4~LM7)YeZSG&<9$_D#MBnZ?Vf
zrY5Wi<z|c05vuP{(5!z)tpLE}H__shri2DRL&7ret?t<mKi<AG0S4eAZt7{iCZ{9h
zbH&*|uC>*h`mJ;LEPn;;OXu2Rb`&CqnDDo`(k0E~(%AZD8+R}nwx)5Rx)U@--0rB4
zkl&88ICfrDrn|EN<@lrik7tY_vUjyYO(Y^*$?8wYMr~gq0?4ks7#0<LA7Ecbz5i5B
zE>`s=gigf|^z<YoP%1a{mSyQ=?tyZZ)oVJ4f<}91J~MrV5uHcG2R+RTFNXT)$pdg{
zQ0B_rTIqvMoI?A7HKAVpd-^weW59#4PaZWy^Fmqu*>`gVAtNwD4<#CL>f`pxDlPjU
zRhvb=^QY~ABIFP36R4t3Kk0i1MV5@ra6NkjUzNJj#=B?dY?Stk+H~>B3`eR7a8K(f
zgg?5h2Z(Cf<$K!;L&9ULjn-T94Bu)POiO19`dZ{Rky4KS*PgHrJjzNWS|O{LzNnQH
z0b#eeFeoh%NFsNWIGm-s^8hm_N`6NL3>2FQ3d;1F#h&)N;ZASne!T$5kYd0OX-y+X
zq^H{eQht7=%-~v)1z^(4%j?PUa|h*L1E{A^R+=SCm=};yr1vL%%HQJkLhdB`RY3q<
z<^hr35|6oh8nSoFp6AEMQ=s0+&(b<7Nz1jn6ZMQY7s!t!cS9;2^jT__^W89ldNG*g
zE)t$}hWKbxG58iso{$Pa#Wm9#BqVKkTPN|N=SvaoefZa_`fR4r9WPZdoD<J6Ar+$b
zRZMLtD^g1<3rYmYICH%k|IwM&0c9%yJnCuG9-v;l*G~DbRkMypL_JCm0)D`t?5^AK
zLP;n&E7)A*fS?k!&t%)2E9`exJ8CIR4DT7iKAt-j<~R_cCRip4nT@lnwFM^~Tqkb#
zRrHd7<YW`p1sEsg42f%JYNk+#@r(j~IV8UVIDh2{f_h(JC{8$iI%>V0QT{E_J0?(b
zGK4Gfk$M3nhK1^!T$0=_{P<iP6{}q7mCScgW|eub5&@N{zQtgrJ?xWJqpKqB0y1Pb
zXDgH`CoDF0lr%F6-BLicKyrwwozWea;(t&!rCpBDWN0pfEVB!dUu?lB9{7erDQ=m8
zQQ6#Jl*GQv>TXr3X_enb!mRRO`RRajlAdleU-Q+>xiGlqYt(v1PKs;UfaF(qhOP5-
z&ATOLOmlK=EP{&ajb7&9NR9h!(Kna?XOETE#6vzXa7cW&EL#%+fqNk4%{{AGVBNtv
zqT!g+hTo?tv+!yD6@t-w(Gf;z?B#jfDcR<AN|Q3{BbOd*-LnORIzkFN%?l=;@rGua
z=+OT`Nam!^;qi}P{zg5i5`cz*oq6rC_4pd=1F02uy2_#Oz#~$(!<MeUyZ#;u@6l&A
z8Nk(jf2kPer{$Vp;v|KgwoQq=4fx4(zoztz=+Gk)5*`CC5eWOVtOymL$Pzyv&N2bJ
zQmz+FzHgH(`8^vi8s_8?R`L8cv0OWZ?<4gTsw$};$PdY$oBdD=Vde}5A$f9q!ON`n
zyPz@GtMpX8kUfdr&qI_gXCHMiApr7AUD%rgTl|nb0T3(von|J7QlT6N7RIa&P3Z^Z
z{ZKu^o<Nl5lE$eUbhR4V7bVx%`Pp8DTngL;b&Cps*A~fd34>+c2+08mI9&dVeM1zt
z_4JNqu{lS-;LR*;7iS~67utp2H{`A3GY$Ws?G%nFf~Z-w_fF@Vt8I*D=-z@qT$&vT
zW=(WNfiL6Tg>M2`ix6}G_UR>`vzY>T18&3+BK3T!+mBA}+vg6r_f%H^5ZC)U^eR~`
zR%wy|R;W2f4jk7J3Mz@|tm@FS7$A>KJlUY=cR+mv9DIUhzOZ8J{gsZ^#>PG<E_3?v
zv{qZCP+wK=z0;k6tHn~pm=I)KXZyWsA2S;I*b;8)QOud(MpMp7$ETfM>2s*yPb5D`
za^C9)3ZW9%hW1V%b%_K{9k72`>GK9BE*i<BobzCWaKs!(cH^!TcD`=CpwB+csD@)6
z5OMTnE}8=cTqEQrPBnDdFHbs4znH7omZOQsU1k>YS(_L2JJ06U00HbukmTcjuOD&d
z;UR}UK;o(1dtBV!=*^U%LEf3gi^#~`w|o|A7})nheG`09_A?N=G(OM$z+LLmddOzT
zctgxndVCB$ih^7;QU933)>*qh&k~8Ru%1<|MBpmcm4Lu0m&O9ArIcs`o5wA;k`izK
zR^ppLP0|~>3N7Q$QrI=hgJhdu$&544LFS9cse=U&u(V`e@ZdsEj`4IiVMtF-YxjW^
zX_Y^A()r8m8XKE2+qaTrKgPjpawjW0wR;6PvaLH(AYgD*v@Oh!R9Eli(aR`>inTLo
z6cAD~a1VpaB1FNO@1I{$o)gJ7<R&u=k>8rq0?^jDtxK@0BR_QP=|<Gp&}<NE5baJz
zr#tOg!PZS+1V+X|frMnD1fvh}WV~qb?$7tMtbL_kOQ9Y3eI1$F$*P`BK%r^BQ7pSv
z6Uq#vT4vdb)WRp4K8QJm4#3P2vJlp{k-|g`6_Cv?KwW-8|HeH=<HG@@mmVG}0z3ii
z+OM+JLFpdvjaGjmX7rOk!o@RQf<n<+#ZCCWS*iq~hrm2;CN4v$9<H&xo$E4=R(_cD
zMmd579Ab2iygOAZ>0+}9;pn}El3IjJ8h7tm9rE`NMa<C#@F5g<>X}Js^EVmAKi+nJ
z<9EirG}Fm)ooXkTDhiBfzp;@c*OZ)ys;^Gfh`&Cr$q6>uDc=dTM9A;ihYV3|)yWgl
z4SW58ovjy8S)WDisblsqLf!aLB6A;@IQGSP7;xHSn5*Ul_%4mVTGaZ|H!iPh^z>#*
zdV#~b%Cjoh-gs{)Xa*zTAJ&i0uz)Yx_;Wx%gw2t<LC5ODl~mQ@<!So4lTi2w=k<o9
z%lL0^4t4b4ZNKJmJMXy+Q95u&H&hD<X(q2&2%CMtF4siJfE2*4_v-NViff6t#_Q1w
zv0B`IT$rilT=;l|6#jGAlv%2%o+CgDLUB*#5qZKqo2jfPR7~LJ5qx-nmJCHmihX+$
z(4pWY#Vx3kiZnU1#=@oPM}i)hiSzgszVz`BhRroV^^lUE_@n5O{e|lpMnwqFVP%?@
zC6kV@$i`5t1OhYsc+I22f18MBje9^AH}wVKq((^n31h5d-%h<a%p|7Z&N4QaX!J(Y
za{0#>bY+s?hV>7JTjOUVl%5=)7#}|^ug>-aVX^D)JgwE7TvbS|?)Myf!Y!vbF3@v7
zOt4&Hq#oe4f>MFQkY88_)%CS{H=_>ES#726)_QJ-`(~1<!V5KF49diYp!jW(Rv8iP
znG`?$fiQvMGkz-jm-G`fIgx_t5VMr1vsPXwD<#SK-Wim@hv9(w4#-~0c-BR*L=ob%
zrxWFv-cm1@@s)h;!8p#IA!{gF#_t24<l*FgLiII9Kl@e<P~usA_9ZnH%aAi$o0~O=
z^Sd?FPA#gQ#gNeODamUR%kS54@C$xlAbV0J!U<oQn?#VQ<QyNte+Y6*lbuU|I0wO~
zl9Kzou!7>|J>X^}+RyIW7ROgTUSDh!^nfzFUgxwvEmx9{Gs+R<4r-Qbn7KMu|Hlc1
z(Sw?tuYP`;bNUAP(BND&jSCx{{k(hdP(FYu+L<8S<-nsN?$bS^pq6D}c`k!{R!1Br
zif&3llsYPfMSyN%2C$W||5_3ff`W4yC9|PM!C3*W!s`~-2=DB_y!{f%p%&?LHNv6n
zPFFf(Ny0@6cVQrj`*Zf};~fD<p9k90ADAD=Ud4bt0|uTL<lo`&83NE9H%g`{Qbr<D
zc6{H@ZTJH{ER15e+n$f>5CH^3GC!((0-a^4Qih@E1XfK(qNi7@egCJh{w&V^`F)u}
z%Z=6a7tStRUPg~vq-$^A`!qMD2c~CQ;fIbV1|4P;K221J8xJl=)SGj6Iv>fr1}sS7
zP)Er0-{S4bv)8?s=PJx-EbqTP_av2F!whsn7N7abfRs|0wpqtfY|_un3ON}|U^vjE
zz%BYO=6)2j``QTzEFcDzh31}2IoAOOgQ%E_C9I4jBo@UO3Q_xh31TCvJHz-}6Bbip
zL+|rt8<Ia1%*V<_Z)Zhxyt(=c*>Hrb+II7WXmO}qz7^Eb9C=~S`Qdtg%f(p33mzTy
z&RiJ!lG$RL7a6F?&t!2Az6=A~5iWn3uO}EOT1Qt`U7VdWN*5atI;6H@eKNjg4)sjT
z?l>dsbcCrho&%s$2>`;tLh1FN&CShE60DBWOe<U6<?RNnJ})xp+{iyQg9<p@t-^()
z-EJoW$lX78%~OpoU1Y2-q5|r^QF2xg1`JD44G78Q$T$ns1?VITP)7DZ6i3)h=S9s@
z1{2Ry{FGuYKds8^qi-(aKwGFJw}YrOGJz1ltd1W^(P(96w`=1v&xhxXeRRJ-rWn=;
z@n0F{hBCw*D%I2akrdKzs_b`2%-f&`po7WPYJrElcd#8Vv+?~EnQsL`PlFghyX{bU
z9|-dO(##`C={K@p_In6diqBUNCIrd&Gt9WeduWhy7$W_)*=LRuOJGY3{;|1HuEGeb
z3<%fgO>h18Jxwil93i7{C1eArnFH*fnt1n4R@hx|Co*qG3ay0_SWz0yf-IbRm5mhL
zk59noOP%(a231D5EIO?JkW@|M4TJ=S0;&E!`AA>@aji~15%3!@a#slX4dxc+p@li2
zMrXak0<0p2#u0EnWl+FqAio5v4>$4g`x{UG;A=I5MfPzd996!LbN|XcGC2zFq9suY
zDiQ)Bd^dg-fyUUEZ{zXRI#n_JFl7P=U+CO<wz?5oyN+j{fkp?%Iw(&7Y_SY9`twm&
z;mXxk@OHX8g)c!quH4IO#)YZSb4K`%_n%-t$=mZA^>c9QWp|fnA;We$lt$vn;&5M1
zf_p=8!uOTHRaAH7rZg6WAU5Zr!73984*)6)RLDN$W<?u9Q3SP7m0Klj9#mQ&QI_c7
z9IGf4J>vtmkPw?nbY}4yM+eooRmDj67)JPKLQ=+MSO8%l!s(ex&2l}eACbG~+hTp7
zQWl)v7Z|+_0FTmSI_qXNn{w{!iNz^wpE`%@*c0)GsehD;4m|*zuE~WlZ8QQVn9TSL
zr?beTo@6Q>OD>FHPqwOOeqX=|x6Z3by?@t3GIt1n?d`QRucgUK;>D_^&Vw8@lxJ<;
zT{yc2Q6{{C#g@8cI!A4BN#|f6eq4`~%-8ar?z@-<4Z~-m*3@3ZK;Y!&zS#U}x^2)S
zeS~3TFSTv&#;>Vd2wy2{V!D|oWe!8-iXRey{BVIPA}>feF=&o-3k0``?&D}vJEsAG
zDE0KDqkH!sI`h51@R1o^+2k{u$f0l<aq&V}3W|sfmG4jih7T-ld`t2VC#)bedGW5w
z<NEzY@WWKhd?X;Os4-2>At*F@!N&uR)Mfy6P~tu_*@9zYGDS;c3FHX73w)MFIlDV#
zs0zTyY3f<m+=O*u_E3|-Z2dlxk|#i(DFl!Pf(dI;%*BP(6CEWaxuwSujwEi|*kj<_
zdZY{h`Hj`N3gDK<b$BmwEp7VG)%yV3RwMXvm@6$I;1?*n!K}JqZ9sfmTG=dt#XU&c
z0L7-MpBxlT4utIh{u)!NK@NK2J<`3}rF*Yk2w11!bOzf!XWfnhS>g>?=KO1A7BAHN
z0#?fHj_r-CG`_{qXT`rVk``hFO<77a*3{AzS}NCpztv7K%ko<98zRP_Y_8;5Hs`iT
z@e%_WOFc$zzBeq?97{Pj6QAZe<g=vVsqMefk>qo956PSQ0KsnvDL6M+mI=5Hyk`!0
zUqq=l^`6OpiF>1cl+PgVJH%w5S^JQV4n%WpR}(xOGYHmK8^J2?PxNU_^%49Cq7Aae
zKo<ZgodQq<*%iS9!w-J_?4FxSO(HJTvF4tos?^>9T~R6g1S4M_l&~$&&hgf0*_>@1
z!Bv_?NZ2&m0iYS#%U=M<Y?uq{*FM#eDi6K{G(uxEBz-8GO<-t<9tAKa?ykdAdH}yV
zs0cYOuGHy_B$!~cJaHicsJ|k+z5=iqzt!1+KyLqysZ<KX4Sp!<D<QT?0KE)t1V&J$
z!0ODlC#%hik=c0iaD9lbX}$&tshYPM;`m6!9ukuCMcq&nSFvC*N<catmfw7O>Jm@=
ztt1&Q=hv9eo_aSRJ3}hIur}IWR8&My3BR!}26D>$)G+`h9)y`lkvr>)6Hppbn1;PN
z#9wxlT}d4)oO~dau<s!iD~kZ?INZQ4J<G&vB6_}I9&E>u6MBj!Oki88r31l<^;W)(
zbQwAvZoL#*m?Ur*>OtlbDWuW`#sIfrj$qRHcXB|Sr9Yv&zILk{(DWEk>ije$B=;8W
zKxzWV(^-pbp`_=lC)z@A6$BmpakCOkw6S5&W1~K?<#-0IcrkwG(e_JJvZx2?aS&7)
z%H6&KI5Ee<V4}Q1(Z8f(JRr{IbH4+`;B>>n40XPbE-jK=r6g2R-=AO0pvZNeXsA9S
z%}4kMBwg%fCMD~(cY2vU&$7UH7-p_Xz<nA*!BCJLL|;0GMES^PNo17to#3-5trt82
zAloa&i~#kPK&U+rT{iM7J?88YKiXady5-z5$U&)0ANp{9o4B<zGv^fO{sB<X4>cO|
zG?5j;C<4Yi_gE!3DdAg25Ig8;0z9+|VS#%~%?(<k06RbC!-H|nK+XNQX&a_#)%@%W
zRO$0aJj<aGq9XkSgThz0=^mj1&J&2#@15!Z;}#wP>i6qgGiY=m0vVv{eq?EJ5s?6U
z_y(Ztu((xOPTcr8!ld`Wu4JZw{DkAjtIEvzAdy^4)8-h7gItUh#9ghSQ-F>#0o6_4
z(F6e`TOC=Gq<Q(eIwl!UN9#B|6uY%Y*we1R{|esJ@F|B`%=QKl(1kpuu!s&IL0xY-
z_!Cmk<M)RkDdn6;em@!RDSwmyN}6_Z=Nx?$8C1{*P2-c(`350{mx4@0#(Wmzn2jV!
zo38Y1%j-9id{^{seQ$rwo2q@nw7Zy70GTvB6KdaOmK%Jv%!j4T9=!#XpT+G{T1j$S
z;Pm|8MU#D?P-=xG7xI`j`iUY#JcFS75&|Je8~Hnf$p(rL`b@x$-(3nmLM69094BNB
zf`#gc4}w+gh*sb$7WR_Rv*ZVyA7r5xjFudtT@*kH_Lv*2^sj)Lw&fZdQu_X3hz~x*
zc<Vk)^Zy$^>-78vqF#&RKoH9V2>`kiQp_Xax3pA7Z9go+Nt&*7eFUm`Tz(BTB5WmA
zaUcBqQxoWivx{Jnl{$Uy)~82MN*P{9*33HFtTI3oCsk7%3!)<wg@#a}BtZb^$u<Y3
zezqs`A*`?yDO^fxYGN`70f5%VIMHVj561qG9GW~9l4DIm4?9vrI3YQ|pphUU)Kw@$
zn6D-?ZdnBeww~1mb`KJ6i}PN%;!HSVjup03J=tS~^(D&V1)d9|J?k1Yoa$Nq@Ht87
zE6C038ja5{1V%>q028sOEVtaL6T^hu9?frBF5^Dh4gnnNGmpf0`9&aoocAIj@g7^e
zkhxMiW*wvk3>AxrrMm5W02w8pEb`e)EJ!WsmRM11?ZwJh$`Rp*iICHOf2N;*M>pNl
z3oue7l^+OHw*pf&anD9|e4yGgKWGgS^&qH5B+q?Hrc&Y;ZDIg4b21{N+VbpxbQd1E
zj9U{a$(awP0o^>vuyZlfS+n!$hp=LH1{bR3U_l)1grP#DP0hq`>n;-OUVeN_*=3UT
zW6ya!s1DaZ9Mb>bc#{wZ)jGGkghn}3;IAcoafjem=Px1x<MJuvV%K#@kJr<0pts)*
z%h!f8_d!4y)&o}$g|BP5c`=lx0w}=}T5TZpmRL)$u3=m8v5M#r3ns%E{wGcE%VI7m
z=^3l_JS?=O@E^(&O)s`^t)?V_lJ*B!)VNu+;WqDz|K{?~LS$36agfC5Ep2^gBiN9i
z32rd7c;-Jys;+06$U;&J(Bv!z9yWj!GRLE>aoCj4_(#akcNEFxFz8v7oBy~M22El5
zbpkj}lR|sr`94V5b$*E$%qz@M<mWS<<f#$}m?n^VL48MrH#vRpsvwGNO-mq`r5Jav
zf@RR32Mue2K^PGN-4mUBXJwF2(2KP2nPMBn3xeh(w7j<cHG(tY9)n(1&~3K0x#~ZO
zwTe6x4@&s-r(__+u5qy=GkS?3eJ(Kwd@&&fV!QyP>NB@{p!vppwZ5gf`O80U1qIJ3
zcM`xa<_D!>IK?cXw@E)<+&@QA0e-mlp7)B5jr9Q1c5;;;`c<9*9$PD{OnTgNU-Ui@
z1mGoC|9;{cIB_R<&y9se$mZaZ8%Tp0*ig0nTyO3-U;%O@&55WmWF<BCLy87<Wn_Z`
zN%*-&_1{s*Bavq+W&tG#YG+-ckgWI033)#sF)jV}-9f|eg<3-42y8UaBzUNd&O)sO
z&7v{GricF;Hp$=1`}eEg4{#L*$#>^b@ap+roAvjrdzwnf$N%e3VVu9OgI6R|`8UU?
zkc9_Zl469GvYprP>i4c`?*GTX{f|$<s{Y>H^M9@9->*n;kAGX=|N7tl$EUE&zke5b
z_4k93SJn+i+&0<mp}#+Qi8%^S`HuR%Jg&mX!hs1QYsJM5Z@{2;W(<1Nj&(!*$5=v*
zOjJ7q?CpO%8)o(Qzy1B{ALH4b)qju7LX=571g0C+w{wSGclwO~#~a+^oyqL(Eb`{w
z!TN4G#A~>%V3hwyJNdV{SX?zRN!&Gs1`9EV!4t$<N)n3-3?6yVQSKtz3=DqT^v|6`
z<zk!?4v{1${QeDnuCcrSG_)({xyB+&kX?*=_iy|Cx2^o!r~kIT|F*Yq-9OfMcfbGh
z-*AsFuh*vxDJYyyUc;Q)eJZlWaP6+)?ri(x-<$IH8{Ffvv4aV75EfL~i2H2zxFM6<
zH4?C!oy|dv1gzwdrqYgY?AXc9`~4&ZP31*9bI%pf{13<3UG6_k3%m~8pKKOIc+UW9
zjCQ621KhpjPgC5v<RAa`KR*56a_|2Qm;d|0QjBTGu>9Z#)C*iUb{~)UFyc@<?t{F6
zt9PcdlSg2^A!bN1?!3`d+6gqD#V$+bN@^)62s8QHi0bWZ400X&&V3Mr2SbNTkz4)c
z48L#SgX_SPT)8UEI27=U0{ro1Cl6h7<j$`kTMK&&Ht_r!^h*&dwzuXWvLXcr^rG3$
z4*`H3_~7&k_6$iJQj9z9q>glKgSgOr3JpPC9Vex`yP2@l$mX8E3t$MU*l^~^yP#_1
z_TkTn!ebO(8+}oQfwMvHK<<r&cDMqP=qymAiST6T*HDJ?clHja#xp1&ccS@Dw4oh>
zvI$iFMAIVInXhheP6AC@sa3TX;NGkMxUK^_7n!6lTEYhub?N}e+}|`orJL~|7u5me
zfHMS9rUgh_6WL)(qPr_J();%Sov+g%&1iacXQZxT7h$BgIMw5x4&}N2=(Mg0rBhMz
zo3A9#1ilmi5H4S>VQ2AnOww!l`>Tja4`JwPzG59O{#Q6YlWAWx5d#VOec|Bm1O={8
z9{|NiO-4T;*r8g#a!}Ne?oa5FJ|<$>1YNN??yorhy2YCVNaxp26w>TI%mrx;5uk!4
zpzCc7)+KD77Rm!-<I9zo7FmBfls!W6<|_F7p>SFn!52s;b9R>n$T5Iu?O{KpX_3Y`
z1ex7TdifV@1LK4BhQQQdF&zlMHTo$en-c|@K6`t1pkx&eWFW{HD489>r3wWL{MWA_
z4(U`1ByBe5YouL$+kI}(7ipjt0&sN~fsxhF3ReiC{Z97SHCWQ;_0UU$b7X7>n!(h6
z>)c4aRg$lm*fcCi7yePt7Q@l2Q6aEWXL7(u9yUH*7^`Ctbp}`jjC3Pw7nIMPNw<Jc
z!%%X&e>o0meb*1T9ePW9?GKRLhxI(;LorlZT6#Lr6k2_HhWq>b`I-{td%^3F$$prl
zCYgXT4s+{VBYT8Q<lXyA&=CL|y6ZuRGb)OBuT0-)UIZ42lT`*P#GYYr33V>^<A)&j
z@MDnSJp(jTdnX2ZxJB288AU8V?T4;i-JNs@$q@;AXL1r{y(hc*pi^1PPZkRkk@is8
z3A~7(l02nH8Ya@M5eOWKMo1@O7L)yln=sGc{*O5M-t*`OnbacA<Mja#-y0R>z?QF9
z@26r(z<Dik86HM*>_Ijd@`SF5i3#uk=J`#)=EOPnfNE=Q4lQ=Ph-3cc>2c1^&Qk8v
zaYKdswuPXmns;z<aY2-%8>f~)2e-1TCE#Hj;Ah91*Tn$6A2|h$39aC(ECk1%tediT
ze?8hFz|kX<a-X8Vf&(%k0Vrs`_ZPVY2Q1GGZ3D%CpX6J|yx+l5JHaE=446KsfuA^7
z!f}EHs3xqwnW#xIZiPfGk8cUE^@iDnfA@l*7sUm{5fo+H;9JmFQ3l#eGgc(^HPLYm
zx!-SldbN|%V0K*Wu(vy*55Uq+D7+O9DQ73fc5=VW`QzwkiI7{$ySZm47a<$88(ep+
ze+S0wL}BEE-LMEVI^R2o^6}6Pd=&N2NAG1eLGi-rh7*5ZhD1K(sXI492DLMko#*aE
zOBmSSS6q20#W>{R%BiiovZbhSHDv`YY}5@_(zbJho$(`U{^y*)PF^C>>o=U)85R<(
zf6o$mC}PjZ=ECEzWI%+!agFya)zPCCXijBwxDC=Bgk1gCY#@{UhHglx{;w-`O!7Q*
zrTS1@eSbzpK_Tya!wI~J{J&iWH$o=1`}Cc3_2;VXCJD$#)Jh94)!>Sn@ZuDoYoa^&
z0U7V#>$N*HWQAa%|GMMujB$@`2s|cvypXS&KKFd675pclfQKXadUphWE&uP^ArnM4
z2n#(dw++DT^TaoH?cgs!^8fhzzqjPCoJW%L1s}B@s`QhMj=#c6E`G@}rp9*T{b?Qg
z*c<d+9IQ9eZo55U`Fm75nGn`z=l@7D-`ynS|G#&5cWZXv?1W!rH|}Y7;iJm@l&IJ(
z3hB7>6+hK#48^qA_T1pqxf5$;J)0tWQCo@kILDs}9|<)(k1#h+-r49gPQZU|j8q^D
zo;7WUU<F~ZdfRrew095b?;-83)-Fl_L4;<S)^x=m682|%wF|&J=|EBap7`_4t8gBd
zr3%PRjR(b`%VnI3>iRM;;F&es(B8n|*93Z3{5Ut#w=qAAkn!)fy{5D4(#@Xl3}a^t
z5DP=brw)y4pg5F#*<J>QqR)=I9*A!i!z-Zi8e!wN;Y5d>ZWf}A`L6+qTs5YJoFNcm
z^_jmAnfd`BLRDa_ps5M4_XJZr<j?_wU_tnWB50gcQhXJJKJ<lKUqD4wR|8%0g4SYS
z!=XW6wKa!dm+V@M;mMkCzm!oo;pFdz$(ktvYjC`X(NOj9anbtb=J!`{O`&wW?r<#N
zm%H$NXmb)?=>v84M)RLe^XDXX?dc|*$Lj;oW3C2Vh&@kO=F56q6Qtk>IMGwT20}1~
zODlY<06JF?N+;+3j&=UI*t?-dit!elNy+YE`wGhj;;xRI6c!}Vt%PkxpgF6Q5t66;
zDfb%;2!{>^!P9Wc4KIy7zckrwzBe5DTX~rfm;b%vJK$&MAtT#!HL%5yoj%cj#&abZ
zDVbjO;mY+{UmT}`q@)g#3JuGz$Sh3Afm+*a4Du|`CP=uJ(Sf0#@o#Rs3R|QY;1&EM
zC#qgwJ9o^cI~r0vzXNygg#KNFly;-d*U5d^FEUQz9#gLWfB`HZO!Yh)Pl?9?o@-n*
z=}da2zHX5;==Lvo&_!+o>lOaQdZ0L8_sD+?_OG$0oCFo*89+#|p%PHaN1HwZDnpoJ
z4z(E6((_oPaRW@T(*O0^(^uhAJf)8Uf@3!|CpUg(!U5C8>nh3bFd=*+z_0Qf<kiEM
z-2UgP{C=z6gY;7@HH#xhrm~q=%&3Z?p9wi8G8Q3U4likwJOBbfidsh+PvTpU0|L-+
z!C%1!hs*>u4-}0@svTT@iNMwQYG`bT-(CWL683Bd-g;c#9BG;vTAAtd03|^sB?g0`
zF_A-BOV<iPwUZdx1|{7fGMRx<2J-eblg~i%Ac{o2InCf>?@y18qHiMT7mn(Hf`~}|
z!p_iTFa$%cmZz}fhW^$0rTMI3jqJHY|Jd+<%oO7QcqrD$S!HoA7cAcj(ES|#wE%g5
z^T>jwp-9aFF%RjRfOa*%wP0v1<LM0BSgqJ*_FF?HOqd6_6`n726y!e~kc1a>yb&%Y
za1KtY%{xG6IRo}992+!hdmT`xw7(URq!_-yvFdtr@PtiJ0)y)a02e^P2{}aBEvGn(
z0v<jW`gt!`M%HRtLO<Tf&EONxk9@6m)7yVf1N#x$I*{%sr+2j104)OZO@^bVB@miK
zTGizq07{nu8Vwo@V8@aJ44(I3!Wl(4M>;_!pktAQ<9U~B@?Z%cl21cZ0a^16=zFhX
zkC6KDrqi?&FejqE)Z3CT(F5nEB_}ET9O>GZ_TcWqcfQke01QJx86ygg50?hd+2*@W
zX_vG<b&vOpXV#p`xUFh!3=Wn@TeRyK*H6><R{i+Wk(&_xuE8*;ltfMUebiJ77fnt0
ziV~Mjv3Fd~!o=AP+)BmPr7UG56KO0R?{$cWEQWNoAMfv_ksAPZO~=b~T$QtC8U<aH
zN6Uk3FMz^zOuCNC5*p?JNFJfu1m&LHd}C#X3EK(SlsfSPE#nF-F~bixUuXhGWE}PJ
z9YI1y%M1}KKt~4=w`}DruP1AFy1vep%E-aphT4b|&|)`Wbvd@j^vwSV{85dXoXeE8
zF-4k_FT(^vAE&|5J2D2NKc=%Ppl=ZpfD{~!WZR3Vjb;t#seRt7h}exnWpT*?6oi>U
zoBcP4;r5_e_!#DZo&$~4A=W^N19=T6<rhIQMLMe%y0gk4XRu>7fbXl`8?YYXUyKln
zH%^Y+1V9|YO0O;>13ei9+*L15g3~-|T0+Ed+uj^7*>cc%N0RhFC?ck`r+ElNM@fWf
zvdCAwE2vO1Ie;KnUR{8kxlnUs2S5Df8Yysg3{Dk3K7{lX12@aaf{BmsgD&m*ETqK)
z{jC^e19o$RLVM;SzIkXQVN2Zu0Y&m>dTv#)&pxnQnxZOo@NNb{lM^g1ps79}hcO9{
zh$O-(5JQ=)RLEE(LSTCKpWV}uqjPICQQ@*HJ^U+b6%30eJ%)TYpsC@|anUo*Xx7co
z!~Iar3D*9X`9ok|^p(XT1vi?M!eNz00uJx7u_GN@^Y_dIrJYqjS3RH{OBI3So97`6
z?Gmb<-V^Yp2MS6@pfO2ixR%M`&~^c6uM|l_H*TY+d_c<`A%&h5U(`py=_YxFe4eL3
z@-$yu%7Unj&N|K43`YX0TA=5%aPtrxJcv|Ke1+j=tx@BA7Bj8Ki=EX{E$F|IopU<Z
zr>5INh@+%UwLqDtH<7E+LKEG`<n}}}#pl=AG=C@lI$RUD4ZGE+6VUENW-R#)JS;?=
zCuc89(OI==`Y%T-7yBVq!zw^$uH;BPh?$7uC$k9ECk539Xbh^VKqMC$O)IE$&`d1Q
zCw@+57*PJ<?Scz|`<cY;`4&r7K-}fvMuB9u3K5&r$IN;b#i^BeLQ-B}+3?-o|60(U
zw<!s3!ca0q5RmS7l@$7-<LVHs%qtL)6r~Lsfy0H4kbe0$Yq%z;jVO=zLod~QJ*B{h
zlY~AvNs)4c+sLK)9GZ83aN|2fvn^v#D}ua4`2%%JZ;Rg$zHFeBL+0^8B%tl0)_+(X
zB*r>(q;W_k#r@M!<sfo#>-N+5tD@{mkI&qeLer##4&fiO%9lT{L>fFjQ2=PFDrDgp
zr%tJkHh~g}rIZHsN&ZAHp`i~*|E0YiGd*AN=?I=?ToWjYY)gPnM7`S(g!LzE{4>=-
zP(d1p>K&lSk4N1>1pe_*p={GKX!Ay^l!0<crorV*7dr~aDpzI&p#x#Q9-3-T5&ED5
zC&if`=L2MfsptG*@&2dVOHk^cBrKWTFc>U>&b=BBVdP)~lm*bx&4eukedBB3Lst*s
zq0|K7uY*yW#Yzj>A4WOWSu{R}#=082iZ{Bc(yK=1<sF8vp+NZ=8>s5!ge9Ziy;lDv
zGsWZH?yNh)W-2Z*<JLTP_t<+iZW{yb?NNhpX)Nd#T#J2Zc4bbi!FU|TasBMgs-XJU
zktbZJr&M6|_Q`YKb{vmULHBJ<S8NlxTbAaW7@qWkr7dciO2J{}LQmz_RnH_{;6&)T
zF==d2;YR_Q-Gh`%E0FBI2JZi_Sh(|6M@SkAa3I+^cp^@oFhWIU96td^Q$y9?LZ=Ox
z#y1FFS`J(y&|61KWiJWRY*mFFoM~Argx)HZ#ZnM~-XLFlL4dP62ulO!haLzTOT&SH
zTje%j$&Zbmb52M26U@gmhAduU@OXSkQy+*O(-#8&3nT8lO^d^+XH~Pz((Hi)x|SC(
z0F+En^B^uC-JAhb4{b*D#2NT@4uiIGFBED)8f%kL#5qzA1V{aoWc%W&l=*x0y*D6y
zsS!J27I!^;z$TC`fO#!MnIEmN0ntR=PjDI}?cydkVdN<3c)6Oz28Qh4JPNWKw6!|I
zyck@kzCHpIR8ZOx^-0mZB;<&Ov^He~2^xJ9ng`$NWtM}ix`qxUrwV_3LiF|(&-0rr
zGu0KcfT1w*ZsR$UkVzV$2_SkyF;YhePOa5{mVl7oA=YP0<8=!3NdZG#6C537E2M-V
z#qkv%RZ+WBJYuVii^O+LOkI?J>g{#m&q-0XMGK9YrqO0G1no%f-foi|U2V?0iR!VE
z*bnxO#R5kX?jjLxOjcb|(1gd93z3JF_lsJ$kMC@+6|^)X-%x<+0lB~092)PBVmsjo
zJ+-|jOp~^#*r8Yo`w`kMPN>Vv*aC}&G_H86LnsCUrOJF3OmO}lf=}0V08gB9zBHc&
zKrF)`|LyZvfwOBBldVat^0pxP16<bg*>Z;Oj8XLafP@u;yf=s3grRBZqUX3Nagm{<
z^}oud-Kw3(nJX8N9He1>;S}&h1V8?o>IjtG0Mu*@ycU`iI;j#t+8r+nAo~KbtiB$u
z$2rw9A|87d4dsgX&1iJY?gql4k#B&!A@L2sObQ`^DeXt%vv3;JAfhs6(1iVUG}i-E
zx96HE5P-;J?@ae(LmxGqeI;bo8&{ck95U%V6?}qv{R?uY8xE5~dX<#PfmrLFgyiDC
z%bnkreQg~2thWb+)4(YV>6D=92%r%3NNIa{f?O#48UJav%@`JREy)2YJCGuu%vRWb
zr7a(B0)xoCS^aRlRSp1VKq4X)@@sO50CcL4L;4)WT7lTpCHxHSKl2v8AtyD#S=AF!
zVy#yh|E=m!C!0g<iI82nF@Bs=CGXCcJCJwCWX(Z#jg&8imMHYA3xMjcHEE{6(&Cfy
zpo~NF;y)pIUuF^cZX9~6=aBXnw6dWSj8k8?pa;)q^x-Tq9Tr{bLESIV!-=+OaO^xL
zM3eJ|+z1+}QaAr^pgVNN3Gy=)ZKzgV(m=xYcxzm2?CQt8R0^}{Q4UZ)&3=AW2Q9Wp
zZ|;2S_ODcVx?Bt#gw!Ig6jsoQ_8}$(R~>w20AycvARG~2;cxi}ICCs18JY-GSg42a
zJ}bReBcw=6DuK?KjRuZ^R^KOQImYM?&4IN7(>iygu~2OV1w+R}Q@KdgtS0B-8I1?U
zH!*%GYvKl~Ta&8R+&4qJifM~Hxt%*AWA1QJK2R}`3atrK6Q*SF#ZE;{Q#LtAM1>@B
z8Ea^M{u<9VRKdMm%*pR@TXoP{FUv3Gam-nZWH<Z|h8F0+-G_8%ATwBR^$!<30m#=A
zzW3*JDWBa%IFDU~v~?TL0j2;HyQJm6p#xC^sEbCzkWx5?`+G(&06?g(kN6Rx(PkEi
ze)Oi$)`Xxx1^vayG+(IkMuJa(KyH|DAEA@WU6I-gj=F;89vq<)N^pmaO90?Jnkr)5
z{(6o588DAm8>F#N406lZf>geO`&aXgn$t|V4Q{**ZBg*WJ=33+S()KX55eL_D7Uk^
z8c;3m?f6ee@PKv2q5F%LaMs!CO~474CZQ3vhSQQFyhD(;5wxFpAGR|^VyQB$+5a-3
z`F&_6MERqyd+<icp&}vB0}c<eZ?^nv6mS)u5{AY+_yU9@Ak+gb4<dld##4-@6n81R
zKODAA$;NO8fHw=s<O5yK$vOnAdi^1!cuJG`3=G9mW(Qcs0F;KvMr68zTz|wgqCS}H
zKOBW1zcNc!{B}`ut->td<hq_=pS4Zq%B`P2nkpPh3{uCMTrG)PS~KbT@0ksmXKD&@
z8!dfOfoHBuo&Wsy%C)Ef>Z3&pMZC^*doEXTFdR}jq;mQ4Ns7}!K~E`FG-vT6LE9n;
zQ`K`a!An`WxvhA7%W8?A&2`gU7w48iX?$;zxC|4f5_OhQeLrR)3kao8H)JOxw~xFj
zEhRwnCC0ENN#5U8BPBC)?je(X+yBMdo5w@BzyIU!)4pg`S<@opRQ4qlIYlIfBV}JR
zSz|)7#T+e`EQO3UPK4|s`&P*?MwXDBGWIoFcJsULA*u9!|2~h$_m6XE%sq48*X!D^
z=XG6=PC3rrA`t+TKBkQReCA&GS;r<Kojr{8DheMX=`8ZGxE>ZkN56dew|Sb~xIuxM
zTZaK3`_Y)wK`7%%gM@@lqTX_Zzvw-P@ASL0=llb#Y%SP1wG&gBcFD`U#D<z<qy*e!
zo^NROW|!S|rk*MJz~-+QLESmu!UGYbxkS(Xw&&DC<=#1{%MtVl!ACwHfJkB6gNajc
zGCH##0%&r4PR4saB+~yo5mb3c>TC}*cZHp{K0!~jX`S=L#00m~s0TrXQvJy3R%6M?
z@URo$H{arpjep*S%Suld<VB@{Msmbma=g{46Oo_q7~UJNv(SGX-CejiP}nJ!!Mu2n
z@95hlV!gr81Id;tXVCy0?{!uAErPz&nJH62%1^uQd|F?9$Pr-ME^`k$^!NaTY1cm2
z7IWw(sN-1)un60l6%*!89~8CylKMy-#PJIOGCGqWjpSQfBG6K&KxTY!?$EWMULg63
zNV9XNQzz@8Y9<(+#B?WSw;o?04eB0I_lF<0ZGf~)rk6i7Tji?BahQ3RpQ0GJ#<PDQ
z`09oB?c26-aDQ%J9BO9{t`w}-9IvFzSBA}}l+p7AK~_KuHvt`ZWQY4(zCGFU15$sW
z<i()LIbhy2sJ8$bcMBoJ<SLL1m5VMcq*#T|g)Y)vEqQdPK6LK@+no`IcB@5hi!u_4
zoImbYg@N`T<%Q~&7R`clZXHd!U+x#l@Z~c|YZh&L%_T>ei6}^-8?*N2sR-!zDzVQS
zHezGJcQ9^aWYm;*blhXrXpxDfeSJ3UwY=yAVMhJUSO1*W#7#Fcul1nxM}sVM`D8qJ
zWVUn59OJBL&$`A}lo%D&1yHcZlxxMk5c&Yz%71`q7Xa2r{CxbMJh2@s7cy^o5#A{Q
z;IhR>{dNUY`6GhbZz1av>g)TVy*>Pj!`VhT)X9^wxLyd3x-{;sO%UGn%*=p*07DEW
z0X$LYB^sztR-xoZMX?2iTbY>j%hIW*TzjziC_O#Bv9S?BEQ9kGr+xnV^(f5^!giM{
z<HT;~=jXG`liYNx#@J|{ml26}eF}B;_2UQ6-T&!msRu0~DY?d}>+(Ui-CclTf>3{r
zY~{zBgk}!*RM|ql)1__mrNdrKyZ%FmdwABV`#DPH-I#;p;$-v9aqgKjSp?>K(=8?X
z&vX@(*Wb>cPcFuANRp-fVmD~dpWqCHg4>!*w)^+5|H6>ST|?<%jo0CpWsLa*y-RTR
zN`i6MiVg-#m0)P@TbS|f97(_gdYaz1NaEUxKak8(e2}~LBg+`Z#AN<d*dX5a<s<e$
zzyIW0R0vj#RAG1-NSr}55f<ir1kd$;424_#<i6~+F}7GF?Dsoz^yqgfUB$~#Us&UH
zsJu}X;i@-M-{t0J)N+bPm3*@*<Xt~{%<X-H>R5f2+soZvDyH+Nb!zL6yrthswvW}e
z>A&{cLnVID{_XLKBOTLQcSX<IepY2x&~^|lFNxpi`gFjHU6Xw0x~_s7iUp;Aj=o5b
zV~bVWqb>Z(pEeq2cM9Xk4~anCHK_GpsQ4R(5b#Dve;j=$fz2xkt*#15DG^*$-O;ca
zrLR_6Zz*FW>^Tr+PflS;HdS|b4Z3pGLEPUzQe@Qj!l>>cn?Tw}*7vlmyx1l05vNVN
z+BU??)-^mi5HRwer-j5@|8FV#u^z`~Y0^a_ePgeuh`u(~*Y5hKlwtnc5e$*kdz~!H
zuc%{V0S=DJB6#VSZ0iu)Fa1<LFI;BbmeTZ2+X$hGYyvIy-51j-yA1|kdxS@^vAnt#
zUMiHdXT+G7;plgzz50BCIaZYU_t#zA+}0F!c>VjERy6+`3^0vUB}Q|nJ^umowkNtt
zGDjoat#_g3%mN87h^ppEo%80GDn1v_i23Gy*snnUE^usHp;zw0C;uVgdM1Ue<vV|o
zTExT|%iytvTt_LCn8ya)f90vEE^2G~|BxTd2ZrLUgdG<M#~yXoqWs#9-Fl`w@{Z6G
z|3}GKkWqwtdK-4aw=gWw(?YOkw;qja%3;re2QZ+kZa`hDMaIJ_QrPlXd+QT_w!SQ<
z3t_LlP4nG1*Q>e_h96f1P`tX}#74_7%v}BPgd=Z5rzmClqpD#-!-egBa_YVhls%`y
zsmfVW@^^}Dt=ISL@#>Ke6AcOq%7#Gfg2#cV=xFn{ylj9U@7wkp<)kQTVs5Stv-+OP
z3hEO)AwqvA9u?MRFOu(AQIg$e`;W+CRmcYJeWJIsP|pH3J$n{u=}E$?4lFbU%nNe;
ztC>8?g!O}m|IBmSHs!F~y645T;>S6yMc7=&r7T*+43k6I+wAq(8Mh-3Uoejql_Rt=
z7UC~BdRaSidd$Z8h`;v?w|3IZ7GjZ|L=BDQEk4EvVs|FybW9}1j)omTv{GbhYARB+
z@J;U`1+Y)=GFXGR;BANMYilQ<4sYV12AQ;T=gw(qX<fKb3`x~!`KNK>6Hu?Z_JjRj
z_6DHnUdqJC7%b+Hp&S6UjrxYeySG60SQ`7cii+!R3=3;QT-+rG5@e>hve6O}MV&;y
zqGm8`d={vhnoqYn{uLTklgd)wX>TQyGHS9Qap<Pk10zq5gWId=>y_|qxFHm)%H~O6
zfule=JV~!eZ*U8lSsPeK#p$x<VCLvMJJD(;l8COGm!qJrjikQXqKX$gN<X32jF&;e
z^dMKisj%lpV7x!HJ7~D|wbebRWWz3cjA!0SiYlIAQjr4@yag6^mp%#UW2~v6HBCeC
zD|wC39OL}EhvE<v6wJF;4*6z|kVMs#qc@O%hJ}%F4bk5OClldQ|MBC;+FDHGCulU<
z6;cj?|I8W0J^s}y{Q1Y^)D-l|7U}uP$jJQscj}442inFsnL_5_V)A-TWo7(J?$%bv
zrq2@-^!&krWS=%$Yg2X`d-B%$&@XRxg!tl(I5IJpYnp8&A6fiyJ$r|r1$n;*kzU9O
zF>g<=i-NIx?+;|1|0)AYg;xFe3Cq*a-zZmMvA8$Sm$_%k($Y2)C2<2jh0l#}($A;;
zcm<j;-S#Tbvr9^rkdOBaKQdP0!RWfhZ?~)a=C@ah|Ip4WyzV>D-n2CU@5kb0Wo!|3
zvZ7r|s(|uM&$}KJv%mG`?nXuof&?#3_FjY3AgAGX-+Y3o2PUpfLxUN5Z=zZnp&0uZ
zVoPoo1Aq<@zB$e-tgaI*_jWkooTQ|FN_Tg+j3Shx&eGH5*uH8^GgM$SfRV26*0_cy
zfJssk6B`lH;d_M-K*j*KX8(XWof~{KhWFzw8zwTd6{BVa&#{tLzPAPSOe5I@)_evg
zO?}NlnoXpxcVi5Ktzvs<P4jgitZ)}%aNVLW?5T2c>a<Qsy-1?)l-JwyLP~${p0b#F
zS8IFiOhqvvRruvV-9Ma?`Z|k3Z&3QBaYYv#>q^7-9=NZ_iW@t&LC<%Kc0%Dj(dmOi
zJD*2Yj0@Uoap#;YnQ6XAUWn;>8aEn-CjnR}wYIkQ*7O&>Nt}j;Mjb^JLPKEV7jG~y
z{`NN}{K-FY@$t|ek@Ws~Ru;^l1G&HbnFrEZ^(!>|#+ZZvRFOTnfdVZ%51maw19(lu
zedg-jUykZ5Kf`b?xL#bBWMry2YHFXj73R=v2<D_^JB3f_MbGn=3<%zrmcX`Hpx+yD
z?;8}Uw`DuEl`VMD*)3(amxZE%+E~4+@T=J4*MJxGuV1l2k+&k=&mF|cyr%bfeqOVZ
zotyhw$0(a-k-TUYA(io^M7Sck3Dc=I<WDReIZKKtAmkJrxX;T>m}a=P<xB5TxxCNC
zTohKF5E}t}5Bh|$Mf!h6ZayzgPg0F4?weB}g{UH;ulWRzk48r)B+&dKia-r`!`{Eo
zopU}Fw0`r>A7diwzxd_l40J!WR6jJJpv}-sboP=0>ViT-+&TsGFs>~K0Q3MWb|gc2
z%+xP7E^ZC&U=SKuUOm<=UN~b%dH=}irEmQCql}5Ov$IHT!~hRjodzK%h=kLLFBIfn
z*9`O;d4=Lt_3J}?WGlLON|J;uTQ0&c>@4<5C)r~sP%N6{!n%JdJ$SkP^(4iIwXxdg
z<dk));=u+@aL6t)*85IQUa$lzllvPWK-#&7P+*A#W5h#CVipw+CePauG_awxUe{-I
z?iqPn5mP&pRZy&jN}eOE`d7_dJ!TmPW*OT@#-yDYx4#X2G{2QR<?j$ryOVnqe?Wil
z2~Jkke}p<(uZ&ea<i*>>5g@2W^xuc~;o{=u9MBEc*WdpzT}@UN0qX5-8-w8nF!a9Y
z7lB;zkV})JVqyjgoz(JdN7*vBJ~$8A`Kk~Y{;<YTATg_?B<pc{8ko4G;y{90yuE`%
z*CsR?jdX~$eM+Ug+q%{nBll-{+0=9ZK&$CFa*%w0QhZ-upGL+F(7!=^*B>rc0gjV{
zQt;c$8I{@PLjxfbf6O#CHrDyC!#<U9`3niw>XVXv=)1fz)qL)G$x0oQ2Ev+Tb*y|=
zKCb3ZOZ48>XMK745xiVw9FqE)y=vw~n3lr?9tJx@h8~oDcD1h_Uj;S^gZT4HCa?OE
z`S2e$WSOMG98;o2sh~?k!$?)<7rfl2YWg1EVvOf_O*WC<hRv`2IG%}<VM;v}aW@c|
zK#gwk`;vP+`#zn*a6qT~E6-1OnaJauN!W)N$M*eLb>o!0&rh&uZn<)y6wn$HlKY;{
z4!wjt2Lgsv*GJ}$f|Z$$S<9J%uwr+$@bP`y(qFv-VKTsApxx~Z*vEy1g~-XZPnHL~
z=WjY@l?^1s60ZnQis^c0u*0)&PH+K)o=Xj!6R&vYOTgqY#hl#&ME5Daw@$SojG3kz
zOF~TJ$nOQWs{YY%UI5>PtJ`pzH1W#jNm&ZVj46kd3CC3v-k{)MH!yQM=_ap|ss1Oc
z=Xko^Q(2Fw(9mNC>odrjH#j`jYv3t@ZiLkNWJa)yvWusM80?gr(S*-btwNZJp3hji
zn&Dw#(K}phTikTKEP6hmXVUqVz-jG!^##32%-931*3u^h;TV7Oc!)wFajp5}zKxj0
zaE)$w?gCoi$2AlMLMAsiLLM=zg0s6}Qp$UZv?WqVrc|(KUr4p)j)u=;g(f);-o3nB
z6P~2kOMa)A8XfNeP(0rBz)%UFOu){cFQ`x%wk{A&wlXGGR9FbBsaYKAQ6OZuWpy)>
zOkN*|ND1*2)d&sov~Yram~fBR{+HlF#S`xGj5RBJN!%9^zpXrsp1usx)AjjF9=RKB
zxidFm-{l_f`OxuLSt{ZrxeVidRRA|h!cDZdj7QdK-u_KokBIA<7K#qtZA40>m^xFs
z4W|-5r+TavIt5`)1!3otH!g;RUxN>l(dV;x9Gdo#?Llkek(NQ*9iOA0*D=9v)w}iE
z<g*%VD+NbYcT7{I4IxiP#$_==apZNPj=r;!{!9&-ol71Z?$j5_dVB}&rzE!wFlvXg
z${2JTd7mk_C{TWB6~XK60}@7N+2!XGlf?K<5lW9~$N^Qr>agB-pDW7`<EVp62TdzV
zn6!O0)S0<nH+K@(<!x2HMdxR6xhn<VTre6z>FyEND!C3_4{L_xAbxAgS!8_NjhW^`
z&ECS9%<%k*|9-va-fjrenUju)_S=(hKGV%@t1V+-)l?E=PO`{k{K5;l0^8viEF|EX
z)9bF=dwP^~<M)b;#nq^Oss){&Vq|jt`nxO4oK^Zl4B1)n7G~y}#To}D(>Xd#rRjV6
z!6xTBr0Ks86SFu(Qn5)C!WTOa#s5Ky)MaKR`4$o^gz)p}fu5r8YfJg7fUGG=$rcq%
z*N#&S$qZ-o5<j6ucN6^l9mJgJ>SZT$lmkk2PA5gNq4JIey%KgzgDI1>)A=c;eC@OG
z4;Djo7O`ND*NmbO6~3m(xbp-R=5iL}tSFW3pVnq1#0me+)ri3HFj(5^u%GGPu#Rxd
zuT41LPRVf|+jfPldZXiwpgwsgn(#my+rCF6<&G43pz*Q@8qH=eFLZ#E(&BP_6h;0b
zC2K;BTyDR)$;HV46NiN>CJ7?-s>4;7eFH3VUz>T@USmAZxDnHugnPO&?=j9iy>^aX
za4ffSx@`jD;=MH7!@~jlVl+@}n(9Z}_*HKe<Ag4}#az3p)i|~b_q3yMM!fJf^+wlF
zsJ~T2dVE(CDI@K9orN<tBXZG^k*tD()`0jzD&`1^MI7o4Bs}N#4eK;N{*Fy)clj~U
z(@&FO3i7jEj=c|GBtgj$89U_3atn@QI?=0*Mj9)s`e#EyMC1kRUhsG~$N%AgBw_mc
zPx#feDV~UlN|$_;^4eHJ;@vy-5f$`R>t~^mi<x+$A}Ob)RuJl2WM{7#;+gUxq5Lk7
zOOdcnSiSD!iHX3pxuL8_UGWaZf}r&|WfyZ!G1y&t;bZ7w!9}c*^AcT8jhGG7Eh<RZ
z&4OcRil@)=?YnhlPc)z1r|YVu@zX8=r%OoVM>z$!<hm(yxfg)FS__@-JD~P$TObwA
zllFgH96dy_(AE8C;LO^r^#+s27etwUM5+2Z24%2CfZ{pY+s|g0H3MeK5=uz@yIOIS
z1<m%g$ya{!jCLz5aLj7JjfWSu@%Xl;V%yIqa=}~<Q(0M$c_TBkA-aMu8}rY}$ZQR;
zU0WmEBYL)AknzATGw)199vT`P^lbY#C1n(j#MU+hsthWc9q{%1VM>MjnwpzKf`hqS
zMYdLj5E`yoHRU(BS!Dfv^GQSKgyxM<EA*F8UU_c4jx@%XB#+!Ty4ft`j_UIQ?>X5l
zJ-ZYJ20SL{J)+g|^Ion*dhh}viB?-bQ<k6f`Fed)xgxWxroNjOXvwcJDV@op+X%c5
z;e97%L1p-Zy0lYf>d8BTKXBit{hEpBIr>M+!`58od&CE%US7e1pPtd3H{LF7{+X@%
z!?iqKVaM55_LWsFE#=MZa;O$}a$-BBhix&$m8GTAK78N){e6YbK>(ykc&`l;2fmzw
z8iS*>^kx;i-%{xvCJa!hW_-{P4c;gCdVubD8<G!ALz-Q`IUgRm?stuc?%-l624xb%
zXsLGdK1lkfU}yZ3@37@{iO&~_-}IMW+kyp_UzF?CKQyPRstzM{wCC_J@lwZ@BtDlW
zPCoGn$@md+uiwxA5P}kOuL>!xqdREjHBjtq$^lv2PoHn1MU~gH#8yhuqW?{kQ_#?e
zWhPq+ezY(A%7#1XeV|mt@|cb{I~a6<ft!jtf)#rR**SwP3|6-Xzvyy2N+uksr@yAJ
zvqO04ihMo>y9{s-Q&L}1Bg|zz6D)kVJmo$Zy8U^==jn4_Xh4l5Ule=vd2wKttj9QV
zi!XoxxNQ9(AV5(_>(K7W)@PK@A7G~TAQNTSrm%*IY06;DWGCXPgsP~i#RhdjnKlCp
z-utzD3q{2}5Ff2++3pdunvZPUxG`7_da}*hRn7htsvr_SpQn;)m466#h5U>8c3e_^
z-n?g#-52q8P*`}cjy$}5)n}#mmGRcFy>hTIGUz(MFu!`ix|)B50M(M2UB$YV!^s1l
zw$jzQ!J8;bk4v>+kpIcCgJ<jB+6=S|JqCoSu|Z>szN;e{YCHp80C6sVz=G5+<knfn
z3%xJO#`!VjTaW}NEj|USmuTu2>M6C1RPDPUV4g+DdNtT57EzLQ*i200L`-ZfBwkpV
zNM?cF2TwlMZ#f<FN<c>e91hD!HbDr%ZqoOB@Wp3vw~;lmq3zFzhl5q8bRQP%@MA}@
zDr^-TyTMz*MmE<0r=F;!;w2jIWu@(O=7og^PX@NmwjE7OD)X+3LhZ~{i1Uc$Qnt0=
z<gCC3J+U=Ll1KG*wV?~uPe}_xkItqRx9)^`l7|5vo~T*K31A1ISh?Sw$-5J*tP=c4
z=5CM)I@7hWq5l5GOBDFEL?|p2l#9!1!SPQDaiP0Ey}4IWKdqH*8P%5c{Q3RuV!zX&
zK<BYB%l91H>9}5D=V9hYuUDPO>!mL#@BL_QP#uC-<!8=Mxrx?4R!>i>bb~{aHvR7}
zG71Ma%ZJtdGgR+5>_*5sJhbHr(b$-?y2dtIRlYU$s7T7kIWO(8%mfY{1q+L?P}@My
z6#K*{o3}$&LEklCs5!JEBQw-{yyO)%-B@p~HYNnM6J{GQBNkaNk{RtbG4{lWTGy4P
zHQnXW7LnVV+@dkCQ=3EF*i3dZ_{-i=tG0FAgc0vO7fHn_ps(|V`wL@v5vj@O9?Z9i
znZ-t<Tl>vjC#I%!VCIdU0VD(B9LeqT-9n!^+2sPO>gs+=OI^OCDI&7t1&_xq)Z@~z
zb^tbdkeSzo*T^dS9UngxZM@yCpflcH9y*F2aI^$mfJ;;tG&iJTXJQo`Pvmuhk+Mn}
zTib&&phqb+e=WqNzQv6<c$&fYcNBX}p5TSJ)-7G_>Q(=?K@jEcCFNy6Bi8d`DxR(+
z_3xKjH#ko6s%II{Z{Ij2pqfR9EOwT*5&U$Clr<jY{KitGRZO}QBC>vC^W#1>Rey8z
zV6=Q#^p7a(2YqCf9-Dl^T;T^Zsn3a2prJ4-{U`<Tz%BJOsOKb2zdi9rYmQAPh{a(1
zA4rUbpZ9ktZwR+kiHDeOtdy}tD{QX`hsSIg)#`hVD~08lbgG$!jwkIQ%q|G$oSe$h
zh?{y|6lG6g(soG^O}-whEo3?_^+wjTb1aV+XXW**y%3@h+R2K%p+%j9v3AiJ+u#Am
zh=OwyzI@PmXq51t1tz?HIW^)q;!u}s@jC);@?&r(zHs_-@6i4HF2FdD02tVIn4f<f
z<~{7djzh`+p{(fIQox`b)8Di>vJk%@lor;)>)U!+P0bEcKCZ+r=#fxRPzX5Gnr<4f
zi6LFh+FJ0x`T|#=yQGFx4E?gBOzqPIoK`OkEz6n3^=s%RtEIUt`7%MghJ375Ow!vo
zO824tLSJO|T0Z+!GyD!oNlI%f4TPG8CT|+_Hqzv#m}(~7w)XEcHa+QIQ+wrTOhLiL
zM=87GQ{+tTE;xsM_;5$$!Ua3aZQFMCLxGjhpAO{;&h0Zzd!%NGcusl4@oqPZMfU?K
zDI65j-Cm<j%bwbFZ7HiMRn$^A`cEqG3y#qUR4CZ>$1z;EpB-AHsyv+NzCN#3<UU=l
zlZMk44SbKFQ-wpSXfii0o?4h7=fQ-f0yvS}qf*eHV+)9jtSk@=mt_O-{RiPGaekJP
z@D<EN?7^_Xa2V2qbsW&B*R=Bbg9?NNtE6%pG~n(vLa|Yvpqp=WU%q&OY#{)6&?N;0
zl>mOt8t@n-eu-T9kNSz!Exx8`kqd+PO*aQ<{U*9aO0Mu-(@#w~rm&Wd<Iz&C*T~+c
z5$_0DuVejbDQ5((5|J9Td-t8riCs?yc)2V!D)D8;qDl!d!&iJl$?Q9cJ~nm7Pm%(Z
zG(*{AUViyEEH6(}1gWq^vLShV8>Q5X9=su^hAf2fn<2g0MNJSTx6=AS3@duaq|-;x
zJJ-UHi|Q6EMQFm}U9s^}VRaP;4TNo!_)x<(fV27xWfswF#}9p~{nrYwU*4B7{RJ<B
z$O^mP%p;w|(RNWWrvDZ?No7BK_UsGA$kg<VY*_T|SbLn{P8*KzboQ*m5cJ67ViN1N
zJviU*Wj;;ue%7y$PU%^1x*()b*+`tE+f&VkhTtUfX=%v!Gc}Mz6(wIDJ3$TH7-^rL
zb<7D9P0Y?#%cnImHr7*lS5_vNxHB7d=|+ns8k^62vfh(e_s>&q%E@Gm!<={N)?4@`
zTLJ1QCgbblP)L@<9Ty2@?|~Uy_G9B#UGylwr%xkGma;}rQ19tawYayL9BRrzqQos%
zG!OXs<f?lip02N}f|qD2us#@zV&Q;J<7lZcXTpqeZ$$HhJ0*m&d^}bAbnX)1#`jzA
z;3ibsIgJ*_#m2f%(O8^F;GLz=S(XRNInH&;49p8zg*O&MXkX2yFUIiEk73f^aOKHy
zAvA^4stPVhQVq$Ct^DD8<R{a_$k=qqv2zFI{k&an_TX>*GwRhY%x>uT&=<wdUBX(-
z4F$nb`bn`irGkpr$O|T~znaR}kSH!L-erA#Vn<{0Bd;hZDY=Kv&aN2O7k?@|vd9k)
zarn!-81tbfty5Kwpt?us<B0;3&?1tHkQjZ>HEKETj`eLnABS?#*niTOTx2;bh-Bkf
zD-V!0ePe=_Os&TK*juhVZHX-do3uT{Wy4;}?#qyzzvVA(?G;XZy&bs5wym$`Ab)gz
zU!ja0@6+Gp(SRgNIf7sF1jrNrk|YRHOh`!~r?i`WY3H~^{FWk^FBUXBXodEWNMiWd
z@%m%kKVC@A#9eqdI_gZs*03`tVsd-j!XSwx;$`@-D>UKc2YrrV6OILS&CnP`AVcqM
zwt_4*a}~J-1I8)NCrghb%G5bPsNPT&7N~`Kcb((7rZ1{+&*)m>R6HS@SgMsM#j=r1
zPb`W-l2D>-wX0Z3!kcbva)aLBPX4?84%~}RvdRIwP|LPWJDB;c$LM&?I&KJSI)uI@
zlh%l59#iDHM@vr4Vzsqjx&B`%=YtxbD-oK-9w_t#6v>*^@#R#WM3K3X(Z%o_$QxTD
z*=cYGc83fP6G~I|pl>9(&m9aNtoJo{C)|F2hMuFortT^ViT~v#BvQhHgC#6In@6AE
zy9Y#W_5K@*Xtc{)`S`6mF&GTd!jO;Pg{Rq{@`5yw?7(w7$*1z2ZCDaoYM56a$B1Hm
zbl1YO^Rp7`qXdhVuH(<7!Z-=p$H1<bW0S6QRloP|xFZzRAg&O>jOys|-LI|ATqFyj
zWo*0AnAm!%J{lb^5AtRYZ|E$vyFRM@8`(m<X=?KJ$ai_8mS?4c%M)Tw&0z6dOBth|
z{=5{xEM3WsCWJ{OWgkBEK829=m0?r;#K!@KH%Q)<7rYxOZDVC3XzcMW+}!4}u3H<*
zIt2v%dt=XBJyxPRJ6Y;;M~rYcWo9PZa;KYSNcqP^hb~pf$tARy-Ym;W?_)jGBMDrk
zPt=*|hd(=)b}I6G*ZAL<s^zHM_2)VC`VFMwk$v0WN{V=r{2jU;-KDTJm*w+*E*|5*
z<OTlH4%-xp6tyse@`2n?`Am@k<^9<8xvc-CQ-0P;T2@xNZC2PBta4&$=l{yVT?!-$
zcL%3cWTSkI7`r63w2M1~KPPO`JSnZcR~=9BIvQ+JBB|8(*r-BU9!a~InMEz7+mM7m
zy@smo-||o+5mQqWL26u36g1)3l6D(iu9kNC8iz)cI|_ng6Qst7E2E0sQrVDT5fF2q
zk!lZ%^*FZBcJc7;L6zyz+~@st!5gT<Db4^DZoI!kqU>Zw&fj#@1NME73Bx(XQ($+_
z!f%q?ey`GBZ8%Vn)ruRR&AIV^W~=(9rtID*0m}PXp_N3Q)H}h02NQkBk4=L)H2Fj8
z{!t(VlM*-GIr<S`BdPP)ebthh!mfzY?mtme^FjS6Fi!fc?K2krHw_UHq9S&7+4iY0
z8-K9TkFz8j85uAANbexwL@tzxK3$)TYu#D$R&Y~wso^zIg_*sz!jSzf@mx4=lg?<j
z&r|={u5d;>j2D@U8kI6soG-Y#lWr_Mlw#tNW{H)%Z&6^GyiuD&c2YagL)@-~v(DpN
zsd!0qtkNFvt7kFhKw)TiVdEDR>Nh!Ws4Y=d&Z-Nk0=oy5*6Dw6F$B$I^*<TXw*?dG
z;q^w)`7hbdcb&nALb9-TjV@R`%`5LtaB}RQ#>-vxIcrtUxBgz2J!Hcz87udZViU!l
zmt?ss5I>8GfW+DFl+3WfW`wq44f`PUfKWWv!kZ^v`SerOv$SY|l;Jne5k1}-LDfl1
z%th7bQY>oZERxcBuddFN7ER(^p4+ax?>;=_8Pm9-YNM1heb5YCtOwcCN^hqbe06Q9
z3I%0%+9AWL@dG^IdvJW#tQh^)(m~F<i*$Pyn{ngp>;GOO(B%_|MianUvg}A<ym_p;
zL3w=!7sXVrybBj8*WE$CJw=@>1@V}Skt7Dy_t(9Fq-Ec}y=QM}wX2j^Uhd-;78V7b
z5|hwiwZ5urN?DTCeYKRgeeV$P7gbG<vAswM^lafY>15&m8q-o|@`F~|Kk9kGQ9sN&
zKqAn?Fdf0^@D@Cx+2~bLmOmCF$+x0{(w`398LHD-41omc|Jj5sspYkde}2B{P*dw-
z$kQ`cE6$R^6E{@1$!mgOu(jZ)2FV%??h<NsG19I3zJ5fkr{<Uts`~zti1|TS6j7qe
zlhI-19NN(#DhBdpnVM?ZdC{&Rg;fdsqBdJr3HUL%1tP&*W_gZqMSUo~+d*;7cF2WT
zDrPXKkR4xFiWE-lmIEV?BGBrB%upjy+^J-$R;uC@L*Fiqe{g5$<dl??5_?}+J~jL}
zmPSm_@Ea~{OeY`yO}uf;?BGEyLZub1Jk0aX|5*Xi_4YQU+9gvA+6&gLTr6XN*e`jN
z^~1yF`^dKhb;);MB^N4t7j@xwM7XQt>dp)7p(BGRSvH-yG$|keq7=vVL`gk3Ip*I_
z|H9<d^_E`;i67If45OKQP@tT1n6v+rd{9ptZG9T5C|;01zJmgIk?oLW<_qu|snNcV
z^c!KRu`jVc(qApN&zwFV_$L4f<<On(W`{j%{2f>ei(S$!N9dK7RP)-h!8n4V?QP1r
z9!@c!6xggB_pkZICExgGJ7&z++}6}nNGyBzKdLQ9<Oos%go7IEobjc`A8H!w{sjIZ
zY%~TWeCNA73BLzM37MF<hZ8~ic<`(vVzQ$oy)8V##~!W|1JKnIKk%n3+VUIiX##lK
zP9l|PrDJE_KIgKbQpbV`)q`Tyc{<wEWv<9N6__I4m8AU50n!(#!iStx+tV$1>}|tv
z^s!14u<xzt;~jGsojkMxJ=z`5W`*pfDo~jle67uyLsRV-L%rV>VwXeJpF?A-l!}Vs
zgT_INA)$lkSJ;B8xPGAJvaduqSg&a=_$d%U<_SGCl9$?xMt5#|{Q={d0{-*^qw^Bd
zlZdAS7!^(k!T{!@B5z>e9R`|~$SY@16D?*>GoYGwTFYDWpL%Z|QovAc+707(rMqI?
zyZB8!*>GL_ykRkSwwiRt^K0jR(=#bKso!EViTm!Bfo@hro}kjJck~e$Ap(go&`fOk
zy>+6}C?~1Hoa~w*juM*M{Tum0fGAXl`)49W4h#=hw1krjkKf4*TyeSY$OtIL|G^|#
z^SlJG2ImmW!VUm{gzoU$8W#USU(<8f(`3(350%k{^<9?g>av}s{sxbji2+J~-~3zt
z_HFO5i3tu7I5t0W1%?A4X5}B)_UIpBB=c&8c;2xlm<Hrt@M_T~ED9*A;1P*YG@a2F
zO;y~YFIRrSamUtMZcs#%JR;{8dhtGk^vNk|GR3bPm4?G*P`X=fJiwMlbjD_83JD3>
zK&v&h@M%Hv9vaY{KE3{9@(|=Ve@L-BvoPfo6B8qdZCK!S-8F8*C;4RG#(Qr72s018
zZBO^rJvpufGY0l-dw@vfYJoyr=pzvrb7+1$c<`Wp?^f6I&j9Nrmzn`XYIJ$rz-{=U
zQs5=z*EckD{7>5woK@$=tt!cqiv!R)lm&D2onU3mNK^XBQJi*11Pq;*XeI%2(+7`N
zR#a3}SI>bKK$x4-y9<zm2zzevvv>*s)9oLIhowZk11MLO;+M{Xke=;G)Ddy6;1jWo
zYAh=|Pt9%ZbmQYo?R*{uByYiJJD3Hq6e-Xc`)onn4@<!9h(^y_ngCn^p}zR;pzp-D
z%!~L4<??<FvkQSur55*j=$N}Z)XWex&vf+)u5@?xO?3A|@kw`uJJ+E6x_|$EnYkM$
zPo7NADSvt<Gd^C{>57)tSs0RRe{Xiy9wFN5;KC>`NKQ{Li4q;r!Gn<G6ho^};mms&
zf}*x?TQ<Wm#MgK4S3GnPO#$c&yL-!q2^|s)HQz}k!zzJAEaTBj6=!FN3fg-jp#`}c
zhM7R=L)v#<KzT%OqWv|@J7NE5FgG=_3p);L29FJ%l3Set`x3bxK9eWt7;g}aa+t+>
zvv79x`<LP-CngfK^Jif!Sh%poHI7hZlM`;Pxt&GC5_2oOA4cybqN5WNIX@b-wzk@*
z!TbZn(Od28A&E8#NunkohhYvGrGtC1<6%Q9Ox%I-YQDu19l-^yA7BT^phiUyKYLym
zOiHG*T98kDTU%UAWaI_!2MG9;&&@#h<tfW+d%Y}p%|T$+;dEbWdgsk4px+-6iHe%B
zget`3U~34cVTre#NQlr#oxW;G|JB^Ci${3PUtunKNCv8Cd(#6b`H^v;=N%Mb{0a9t
zuOCEi*p}_l-Tite4jz?zQY0BFi)iBAP}(7f2DI-qnV$IvafRhlp$!M}c0Gk6w-Y=)
zlp{v}R#dD8F*!S1{fOWjGzCpVfULGrS69D_y97hY#W*+tQudWoB|&vhn<cOaLXXv?
zM-<QQYI<l#i;jwFg}rl-+eCAO&NP&q{R$XwygyrqtAwFp%p#o(^bMwC=V1;*pO=>d
z|Jsz+kMQ>PhB3lU0~+nYJbLHhlL$fuDK|f#x3-`wjCq#yUN8+~XxJ>I%mk&HGlyJt
zI$&d1aZv(b=jXFIIXTgt#%5+_aGZMhBUnFFO=BZ*r^S-ZNs)62-?h`}AIX-|)O7!+
z-4^|vFYGJ-ZA(o(DayTXo0p=W$Q|KIoYK_?#NA#4@{q)a7H>+O&cfA%x_@91bN``*
zUTN6e?pdNR0&r^+17c-mCN|+z6_bjJ%9;7SNMoKrIe=7&?&k#BOt9i7OR_X~YzUrE
zMbKT*$&}~J<WzeeRo3qN5)llibr?i=1LhNxxqX!^Ui~Gn_xT((ZoEBI_9X2(Nn7l;
zpPis}0E@n&j)eNLw|RMgUi&8R9AJU(GS2jS)^pi-Fd_kxaY7Qa-QlC&KG%=vkh7%2
z_6quJp|#*rJ1{0l&1}0%f&~gYw2oS9lN5OP6vi4rQW?5jOJLrOgG}fQ^aU@>wNnb7
zJ$qBcjdZkf9=UMg!dI`lP4@Z=F&toKmc?SX&&a^Yh!}KD;H>6_K{zN%3JR6bkqZ05
zxJc;WBp?(_CvTq#k8L=~#r1KoW#j%Emt<up3H`LtwX*C5A3@5YM<RH&WMySdOyat@
zF%e1{6)?iOlM)>J{JEs;MPuVwM>34;KGRMQS>9!}b94;7etrJct0S}6{{AwNEJ<=&
zQc@E1ts}DoV0XoJ4u_d9uHy5s-SvGSFS#as@z#X&Nbbnl?1UE{E|b^*m<;w@C@C(^
za++<O^Iy$81HHXlaORGVjtA}3mHcv!_EJ;0wqf%IH?UYkS5F=gIlAlv;8-q#*G*4H
zb|G%4yxPZmz%!j=rn%+v(HIDBne9qk*ZT}Posq48oXBKe_E`yUXjprG;b?h8z=uwV
zi~$!}sZ_8*aIB6YxiyQ<Wm!*WK>cA?`vV^oE$vck%93y3f*s%U5oS!bV?)n620M0`
zL6@cC2!pzYM&tf4A*qpE0iG%iEv?={5$m|SCzT>BuJX{gK)0@f)0iK0lye;@&g&5y
znQ5CU$OP;2=FOWghOwl4Pft%+;6zg0W(yT+Ry#g`NJ5eHJ&9DqR9|1uuB&6gePj5+
z`}~(Bqh_f{DX|AtlqoXxwp+WQWxK--LE#1s6_w6H5vy+cfESYF=kxRPu#yc2-J2=%
zk<YntGbEV77CL%LBEx1xdqyGj>=)-<w{G9gak|vM9V*#{xI#xVXxKh4K#Kot$(QZ%
z@boBI0cPOP6;5B+UYrL&Aj)Ji?rA_kPv;*nne?Q%4ZzA-r>l{D^}Aexu0y#AIF#-K
zs;0gD(JwZe)^F>kIY!2Z3qtST2h#Y6o0^$v!2($w%A}6Qp9V0Pw0)++2F8#y*Vp&V
z`{d^2bQbC&lf*8DAtPuQ4Dis!(wwQ8aYN0P1nv5428XB)i$t{~ZTdVmjoFn%VWj?A
z&W7{as1Ma8MvtnGPfz%=j!fVDqMH=z=?UIaTT@KQOK-vW8qnoxIhYt`sx&{@%L_R-
zmKq+089C3M<ICb~-m+a{n0FwepaeROOcMr7mz(L{=y8QPAGrTUZ`qG=aobx1YeTH+
zuBxaBbdQHq=D`L*M^WGKD73y(G{tLSD}eOal>iR4y?8t<Uu!Aof<~Dl4OmFF_h;+N
zsDi3#P#r8ZHY>)s-D!ZSEjMKzB7<shqaHdKaD~4($smo$6Z*rjEb-Vb|EEtADHkq-
z;3mAU#bWY<X7~N<7$DhiHzT>g^2~zCMCcFz8TApL{`BdSL)$oumuP=|*F~1Zv@|8|
z<N9jB1~5|Yvl&w5{Od4cq<(a~PWl;}oEU9?m44N7@=~Y1-%M@H*~#4Y1(4ygqRk_H
zeW1oju+t<jXQmp)?t#B?Tq*jBii$M$->|~jzimTc_5|Avp{0|f8m<*rSwga?A3=}7
z)~Pvg5x@oP#AWnvTYz~e+x!rK_+>@KBjW1X+DYf6%%YtH^{KwWR`~@1qr$(vCh>9v
ztst=NIe!UQ+eerauduS?D>V-X2WIN=nbTnM)+Z!z)G8q<L7-vit5=Tpr}DC;y-odm
zw)my;T!ZFLNGtzj$G_pG%Qw1a+hQVjN6UNf*SprHd7M!Xq`xj!Lr14lhz~Kj$bX90
z#yZ>oWJ?8yx3}vWvMmlehi;P|)KaP_sWMmp{`+yM;RD3%FS#^JcKet_PsOC5P{1@6
zak_jTK78mj9?~xNP)kZm>TdXwrCd6K6+cK)faNk4i;OOS=`~D@e!jjDZT-g<XQ(jv
zrB7WVs?0S$PEOYap`oFxnw$12`Yz;&#k4iR&gjD@PsT??L?kAfy$ah(4Q}?C@kt$s
z=o>Wgt!Z^sNAeQ1>77`wus{6(Jk8|1Y3N{-rHRSTzT;dKeozf#HSDR76N4;|k3Me@
zTk_yz{e+g2dkj8%G<VTUvWk;z<p4OnqGGj%4+vU5QiCkU$4M`~AV!g~QuW>4cZ)Pd
zHISZy$JmME4_D~2r&@KX`GRVimcj&?y?#)1o@~E$TiL4d!ZJdIN{{c(&)RX<_6M`M
zxsRr?v$NjcugvCWK<r2bJaAuPKux_4JNpMidDVb<|Fg7uTaaK2)YK%ih%Yz0(9}QF
zBp+;aOb$&q{m_LJwX*AP(vwOnQ^C*w#r0G4m{l&e4196Hk@Y?rN%k?BK%*m99}kt;
zA9JIl<LBHc(RANRrArUE<j4DFJsV$Ge{cU{i0;ep(D0?&)k?Z*?R119CA&(If;M`A
zUx_YDY5gE1z6}8-!C|hyYhp6ImLI$w_J>As0N$%&d|qXraYf<!^=F~bPa^_gRQrq@
zjVBFTn{J6*zNl?y=gM4!(Y~XXTjUddT{K#cZ!GA?6NM5dBpYbmdG=6?c*vO&ZU*Wd
zSwaILJd4!4SGacBQziQoAz3zqb*Tvdp8j8T)JPn%W?5wkm%Bs^I+Q+p@L&I4P%u=9
z3=Iq01YMWb_Or95T{N3aI{`W>WKV1#4r8@9G2KNxexl6JFDNL|uvufSGjGw}h1rAn
zoI}o!wXW;I#y$Fhp7R{J#BT}iQWe9z+u0hm!kV>HZ*vOhekvsq1m7+RnUf#`*(ON5
zTNE6wRnqxCxN#|cy-K03{Q7HJ8~G^mDQiZsBnB`q(?+WonK%gBM$F7O>HeX9>3!mK
z9rV)EZV@2$4Vyn<P41YR9q6Q?`B(<c!O^|VJP%yF`vEDy5d*j<dHISz#EgxB1fx<e
zq4nC$NSb!9-hp|G5axSE;KCM$GOo6Lh%4(Vq@Hs#Im!dc9k{$C0!m}br$uwA?*%zB
zb_nR8LuR2LCJr1uF?k65qQ?6AlIw@pOwtE>X5v_^YS2PjdP6YEp#|n;XYV~mbv#!`
zu?_o?Czt8Zq71b!CT0XwI(<z$e*8F$-#`{8k+=(C-<i<BE*P9-rvLRAxT3#g5=Qf!
zAI~my<uY7)LPTb^scQ|xm0@8lGu3TvZLoh*8v6yt)mZXgMDUstfu4`hs1JtfU6Om=
z8;j*TPFl{+DwIK8KiQGk{?+@kh`;{>%bhIcbx4jf6pFT`DS_p0=`u+^TljI>@2L6?
zYT2u#qjj2+dn@b3$OCIf?&UfzYOmf=Z->wdF6yI}3ZeJ~u0ZL!0fzZZ#_Q~GAhC<j
z5Fys;-EsQNnaRQ%v*m_oi;90Gq(opjWgbTTjqu^ht7dW}2XmdB<CAhu400kPYSSB`
zlXNn??1j!D&0Twu67Y(3MH-?Ic4O^>;dqQ1$lOiU5J-YZVKWRXS&}xpBa*cwUCXBd
zCmiJS+-bYf4Wx4|;XMo}ixDDKR`%FGMz97hJVB$nd{PGwa#})P<OKb8!q=~L#AUKA
zI$MgQr@MCUJHS#4B>^@=vWu9dazaBQmFc?04)JcPR+W=-5)u4?ZmL@wp~q%bNdR5m
zw<2Di_L0+Xb~*8Qa_Y=A)t`S>#U?Zdz*DBS?l_=8XD~k%V)I=Dm!F;{9vm)B%8qhm
zg+Trr7z;eH3FhpJ)7@JZBC13$%{T#NRbC1xcc?1FD+j=c)e+bj^;tYHHZc+Q293RQ
zr!6`(`<AZY--17~>3?NignW6|f?(q-vg}9Q071aR)DK4$w}OQsY`hMG=GNsy;WtMU
z&I>%%@`A2#T=s{-*ScPzF96RwNx!}MRM>I6lBKXhC1iaqVOSZfVO?3PJ&m~C=z}54
z=Q*O37G)7A6;$%EDE&ac3!MEL$EB$?@c5I`N9Eq}^4;eg?b$`O@11mh49tw%L_(|n
zZv9i1^mlHJp+kYV&l$yhePp;~vRgImy&X`MeZPh#BG+LyxwxEWmN$A~@QU)E=<2>S
zvG%Z+V#pvvn94P?%@o;H2-9+ZBhH?u7#4><TvP}10t*cpTmq0SO0b%efq|i4LoS~m
zFl{(kUOp)e4XZH!^*0A2cdI|9YOCp(&XN5G4)hn^N0LYFuxX%ob9rSY1;iz?x*Nu1
z{f_UYkmzENvJH+Bjh$Z@!Rb(JNaE<Sii(dmI+q_+POiwBoWq=&nDbw0p~J&9miBY~
zlk`HH_{^F1&6xANWrHF{7t#kq^?^wEd`sozCWNWSq-0%)^qG38U{5SfI#+Vg9(+M=
zQ^b!<oOPscMGM5jQ4JpI4JxvYp`T(46Ra8z!Isa3_DO=jzrW_*wL~|RDei0g{fh91
znZ(YXo&kz5?4z42N<t(!&w4<aYwMa4%+LtuLZ>k>LJ7~kfxve8W!2S#7xW-pm?Vt%
zRQC8T3-hY(^CNQInlGAgw2Y_sj~onutyV#7Y;45^7Ut$jZMKl2UAaI03bfS6>gp3Q
zY4I}h36oAk5OQsA-bsTei8ysU4r^&yZ<&E@W;cA_e@jeOyYWZ~4}&{T<$Evsdq$mx
z43CcgU~&K%1I?uB2PX4hCXEzB0W4Bcd<pDy<WP{DKLsVQZ^}W>#@ZSk$aAEe1JYud
zj%3I&TnH(Lsgha8))Ge%o1`)v?Y25L-S$>68+ryarP60Vw6-2J<)3sK8(AD4d8?sq
z(`v2%^~zd?<PFC6Ee5(BqbRdouxIo(7CXWw<N#X@Ck>Ekv)L?bedimD)MXy}d1xDv
z9pY?X7`CZvzL0%RN2eX2^3<Vkn}k|#DIMPZ>6;pb(Ejq_s-rQvxr&ccwj?55XnGp@
zjm<7D*Tfz?*xWVP;B(w?DEO9`OL*Q(-CX8aE~{`a`XbEFBwT7jlj7`e{!^Oe0p^Sh
zHN$?SJ<#(=&tjm}GbQ&%M;I3gNrwRe5`qae1w9I36*4pm!mg6=#fg|MSC|=*^&8bD
z8vE)MGP^1FZBqa||KQ+_qz%6^-R+H_aBKtZOR`|!n!=q9+G@seq#<Y*Zo25xr^5lx
zzp7oThO&r*%r85;A#9L<WbOwTiB>yjm9$6&C?ks1<toVKGLYMbM>Ue&HtCYL9v=ev
zxU{o}*tl>TA6l<dS(SE+D39nzsBrl40zepXUVjkDrLqt*Lj*upE1|M)`%)cml{c~M
znlIPemuw?daFN2nuWq=W3Svnh3|VD@zW=%so=DbLS3@fR@8hmha)gkgs=uQ`+X`6i
zPe3b2W3CV+-3It4I5cvQVsV(Gfd_c2^gl_8XT5pL(=uolfJ*VPakZuD0pe%>_}xp!
zKT9Y`H=WYmBC^sgM}36vk9jmgQC_F{6r`VgW1wgE#gt0_I%loVb{{b7od0v?T{ATK
z)1C@Ch6@WUHwu#D3*QfE)gAi6?WJKOsx#fsva^FABW4mX7IsAb!TBRI-ZpYU<rb9p
z*;bR9xIKHcU~0nY(@pM#i(^l(VzJz3FnPBvq8k7DSIzr6vHi`bts>8!oe;6S(1Jj|
z5G2ZtmY33%W7p3B>XMai(y_ab(-JA_tx^}p3F;h3#na3GaTkIiNXg+dY!gm^>PUgz
zmJRC$X&FyGK%?eyNK(t~Ph)g6OvVZl<(y}qfIGjgVjCvFU>)^(=V;!SxrXOB-@x+T
zK;*F?P2FE>W^DX9w_!Yd>)8ACg-y1VJT8p6I2aE*e4m1%H~BQ{;nwS$5=JvXd>KaL
zl?ceoplUwAW$Gc|OzQt++=#W)*t^eaRxm(IEU@B-EyW8|Pb?ua@+v|F=QoTZgw)K-
z{79!Dkwo%`v3A;Gu_PNJIGUl2AxO+=D>u_sON~;BFh}Voy=;dFmr8v+W~Bxzd2gYX
zf|5nvAYCsC%68{ku_fZMt;iBpGptm6eu6^&%HjqMYA`Y2K*o-+Ge_73uEZkf&x&Uf
zo4al}n1+X`DA^FHz6rvvETETMYkzf4mV*C-xvnfAJ4^Nt<mx*YuE^mEzGaoX84L;t
z(<e-}uk;8nQS-k<PfPCHvOGwf<zawk8!+Xa^ppxQfh$plQ0|dOpeqN@@J;4@ZQ>F+
z;5Sp0@ZORGwN%3UZoolXSlGnz+DEcc-D2dfexldPQ&3gHx0)RC#&2(4c?Z>PMl9|3
zHs-NRBK^j7XnTA6fi5I;gUatm{#%QIp>h4^Mf&>G?3EK+I7<Mm{gb!Hkc?zi7KJ}T
z2y>74SX|DacmCf|rm{n{G)M<D_0?5mlH(KN5mqG^E-S9pcgYj<eDO@sGiPUKM#_%=
z8PC4U_4Ym>*4i3_Bov;|irSp7teUwbf{c*>eC9-g+EXq=Bn>1wK5;-SExd_C+UcU#
zJO`P#D?o`g+<jEEOZM+?fQ41z$#=1!-Z{e1ziPB1tN`^8*5+k|Z&!FqX0hqQQ!O;&
zzjZ=VVB)^ZiLO^+oc>mSy)rk;34QXw*b>UL$bIdSM6JH9X4aymFenJa7xc#2S^fV9
zO;$w^a2ALMjj-lQ;Ifnu`87JJ5*7$rTvVsL|8#hT>|We{J{#9@TH4z)B9VOm)*V-g
zRKPWWWx|$Y&jfZM&<viVfn%iDZ<s2d<O<s%*f6qk?%(d>2XZdo9lZ8it+f39rJG;w
z)t4ZbV49jM%ld}Ta>hX2EbLRZiHJncBO<`$W2m8>_9YplBL1^Wg4D#kmt#<(86W$C
z(7k(JYidT~4L&L^=|3x!peBx%Zu3|!v33T#X;bCR|C#*rR9-&y37Sm`rMz<84Jqry
zVBfia8bBvSE^CF^U-N^!)~`28DEgxq7>FvWh@$+Vvc9B?*0P`s5-9z7(W#P&UH?-#
zmA|N~u09$Q6LYQ-3HK1g-_#^w|KY<|qFq<l!EgNdgPa^(<xH%25X<bgB;DUeovk7l
z^7o(Lyu^+_CFBaLR}8^Q%5C)J*RN@oIk^f04D<>9w2X1S2&T<&s#(nvpO-4(eI<Y$
zS11FH*I1fo$rBxUukR_)Rmr}enSGhvsCV^)WcXoFR%N&v@v3(#;Kbj_jo&fr+k$<4
z1Id20qAHv*@$quh2pP}WdK{puKbsu#(3YhlLjiA4Sew#fNP*K@=O;5<=+~Dx)jlyL
z^gTbkik8PF+ldpkbC5q?%P{@m2%^;bdyT4VnpL<Bd2G*D%#4>zFuQSB&u{_TIZ>vi
z)(F~{DaL;aje=2QpVrjQ)QJQU7Hv@6<LBqEhZxSNIs%F~u<aVgg1Vlk*6O~g#U)4!
zqU1Y#W}oI_h`urbCb0DawymuQ{VMj^y~j>NO|20|lnM}=#=1zf^PBmp3Sc*?rp-xN
z!Dh-><Q{k@AnSd{{}leSC%yhQ5oxufp1pZ*A-cO`In-U%O#u_splk#4l<U62fIQ#3
zch^~OyrQU>0;}y{#Z3RMd{}|>o4XmUu{MHosJa%cpw!a8_t#+_!wTkn=Nvvnxvw}T
zD`IMN|LWDN##m$kHe05#kx?KlceV|I`ZIF$)Kgq?1gKIZAt~_oW3tX60v_3-6B7dn
z7Gw{mZ5n%+?Yd=ay&?}(V-m3k9~{i6&1=w>XNP|egUR-s3W8%GnJNIi5xUE$A}O3r
zn@qdFzDz>90!c01G7}=SN8%ZZKgkGlkMxdJzK;XqZ=P;VeFMvO4xK`#Sp#5dt($)G
z&+ObBY*zXFlKA}jbL8-TeXxBawOB70T-ay*52>tdnLG$DP@jj&Ax9sY!o~xwm%7fI
zko`5s#^RCs84$*kW495$rXF`0(mR)^<p8A2Av#xnKn$X&U9r^c|8fAuPL0$8X=aKs
z@`y~RXZ5m1JO{7|6f!M8zkL8h*c*;j!=QZ4`AyK0-&U*l6{`Q|DDFG$G<p?MPKqi7
z(>Mew3CRPv<MUwGpqCOMd~$Ehu<9M~w6{ID+uFKgk!exh#mzVIH^bUG|J=Rjf{+y2
zWxcdy=RK~#a%0L496LaQlwo8cTP(7s2w+Ji%i$T-;RRVxR8p#e)u(k7Ao^8`KFo*T
zb$0sSO@7eSjldlvFFQEILlkB&^2VypzF&Sn5!R;DKohEz6aKrEWfopik`bd2y&I(*
z*!I9>@}Ii8I@n24Z*W;tb8y}&Ej9InOJiHU1Lf9ab(>V!N0`Xl+}sS!y-8n0XRt*M
z_Vx`&+8||8?eNT^;1CS&^c;_jb~^k*NoB#Lo2hXedQcWzW{<x)>t5pE0*Kn%4|;Y?
zB?=>UySl*yDEw7iQgLLL>s+Agu0tQg&qC<_jss0_41lGVQ(8xdbB{4aqKcOP*-;?d
zd2f!3`h-S&aZjG?TiDVcfjHpH_H2nmmRNm5pd2KIobIQl@+~trLo3w@*`&~m0KTba
zzWZ$M)kmQEB}~2>P+@^(je*=_R^qEy0?kdIKK<+A%K7J?RVXV!3NC-%O1!b;7Pmw>
zRdW58$!Nbyd#M%xLy$@F{3K+>22&v{Kxx9%vac=dojEh(J+495%bpyEnNcaHOpT3U
zQHJzejkA-rO73>%%Xh6Gn-;HFq5D!w0cjvXJcaR=mPC7&kZeWo`NxbiY6Sf<?ii`F
zv^#fDnIExEl9KY!{8^7%NQD#!y|}Z)>Oo{bNGG`*!?+;i=$HsOv+6$&&}{0dcp9I7
z9T+e_Ki_h_^}-@-%^TRkHK!1|qI#C2g(a!A^tI0ww(h@@$OUYslWuYqhKj&4o6DCk
z!(5RFaRGe-4XF^jnpR7PIu=j7Wi2?1wRZrJ590!$r+8uIsm#aAN=oaU|2i7Jf8}9T
z=@ZG$xmv6%l&?i+dn2nE`C&0Cq=Vjxe+zw5_Ccu$+=4o4fC8~o7c_+YhC)eB4K4s%
z1Edi@A0KRlJvgZmSy+gB752S6xctq29H<&c0uJU;;!Xdeir6;>U3siulPy$cTIHt&
z^)@wYEY&z~6<n6Lx5tO}gg3`Q#&PC|;CEMrnGu-r4W={zkZ9KMoi!fcn%MrALri6t
zUrL(P!{+Zt)pQp2^_h`Z`kCHhq*Spw`C*8Ymyp0LG9K0*ZP-0>s?_JuWomgGqWPB;
z$WScV5oF@fFEWYz4PO6In<Ei3^4OmbaxClORR_7sch;_}sreUTb+(>QZ!|@=FJ_^C
zl-sZ3E7uJN1BebuK2rdy#r)9_hL@KI3l6|O#ToI18u5ieLvbaF1E4}c`ZO<4Q`5S9
zgS%Ix8o6<<W~AuR1?hd*d2nd1p@iA#(z$bQU^TLRanvk)LatuXCaz=DY~hk2UwJTs
z@2u3^e2@5{?tQMPG+00h)d6iiGuYmzDZBh-h|H~ch2?BYBSdU=*MS|!<>d@uI3Kip
zZ^7c=yB9m?3L7%aG-PCmu=M4S?WaGCVFO0OXG`Cu-dIGUR+&Uzkl{ku_t<$TQ}ZJR
z(p=-tyY3v$8-+z};Xy&hH~w8DZ5E$Cm+w5g@!fv5CA^(Ppxm|oL6cgl2MP_Z72CIG
z4DdoBYbC}lS}dU+*}OSl&$_4Y7ej`_BttxBD{^3uawO$VHB9<1m>b@X6@mRYFjZ8?
zG6dk2{hKxO?FChyP*bC;1Tv7u%h7M^?c2~zXusE#56Zs_(<S0bjTk^n0k%1k)=v@3
zot)Dj_XUrvj!t;LY4hfG%Hou(&>T_0>FxKsUJ^RQ-s`txDmJb|?rZp6P&q)Ok5G8J
zX~U=ALBe&Sy&PHXhOY#ew}ovQRMp}Df0)_@=p}ZAu=!dgg_P&Zz{4eRp}LaDuaUET
z3b?p)-QH)0w<J$>z$|_kL-fS}H!g%LvOfB)94wd0)~>e`;<8v_^}{{IbgN0{HT7x`
zM?D~fpI{6o0$m1UvT%q(`8q6`kre;eys%u|hrYw=>bpO@*vpUebSgjmY|phljQ;0%
zj5f?3`C~8R5rOmP|B2pxMmB`&h^&NWu-V?st@5;f88XDnM(6!d<{wds1JP=3HOG#{
zrU|g+KFw+)3tLQUP4?N`vh~&pe`?sHH8)<ZxW7$oq={K4-(jTwt5q1f8yTQv&;&z*
zTG-bGd!p{4B8f(P@C)(-evw;*6$IdzhfnIBMWeFqcZut!nYZSAnSOI`b5CWsIIJ@S
zzZd?`o4vt$`_`39gK4Or@h^PJOlb@e)_jTx1i^4n_D6XD_tKf~pmSIpjm82V2x*~Z
zUAhqYa^wo?#a0VwgPF0eXF1H`PV;~d$OU#jTjc=@smL%4_cr&pOZWr9pT0vMkkjT}
zGD^-R^V(E2@uUXJ<sD}jO-udCfXl+D;?Pol)foCQ$d^u@4<ow-;PH4DQSrL3U!yy)
zd{CA;lKlkY4IHHih8@`y&b3lz!YMN!0t#+5n`J&I5r^6=JapaAv6Z`9rczNAJfDKO
znOR;6y8A<vH+<$v6)rlQUyz`waxbS)`yH#X_D4<=k)#0bM?J?>rje;AMXwYb`<0;K
zz*73#L#z;;XG5@@w6ttkWK`5=7mC#7u<-t3kHq43Oj%i3N4q?-0?==LUobyqu8?v@
zE=1HW^~|Ne;MoKi1Q@6f2|rwpRAKVM<v8s2pT;ZM*w{P^E(b&&8KngH>*n&#(-c6U
zRm#nEvWxe5@2!GM(VNf)U$h4Au(M3=z!D-m>P?GEP54HmG4CqNnqn?|Tz(4-zCjA9
zhe~5pc$39?YB0#6sKN)>JB^JsuukZ0ZrQ7hjGWjF)H^vxy(!seaGVGmQG&}qM9CJJ
z7moQSceZ^7@cP@%gJ*7M{r24m!0z@OplSuCcEY4zMijSpUIUtK)d8q(g9kpcfR5Jd
z^~%Vwnaey33ek=)rlO+aykHQ4KzOXSI4(}%pV|q_D(F$kswr=Oz5bjZ!zFa|-($vn
z?x5w*@DG8Mmo7N*L3@Ie&DLe&bZ_d<n(ncEWix-AWxVI=1D`rf2b$;>?w_vUk?@UQ
z`Hj%s(&Dp<e{yc#$y~}i4mQTlpoufH<wI?4XY@BZBYep5V42Ss%=uvS>Q;0&?8q20
zQy*)ZhD|EM{6kQW%UJtZA!KuDGcrR<`^j(#iZApjxN?lYqgBxS|JZvEs3^0gZ4`&2
zZ9sZd6clJs6ct30<TQ!`Du{yQEICWgprWEEARq{m3<Qy=B*_9w7EmNgk|;R|P42&T
zHy|_frE~9D_y6u%$2DtBck_n5Ygg5-da8DnO)CmPaSYs{*1Z0Arb0w(s@K~My<iu^
z3g^SBd&4Ypj6t|28#6OA$2|y(2g_Z1XgjY9t?{FcczRZpcUePW-d?}T&Fpt3A9zLi
zNv?{Nu|E&uHMGK8(kc!cREO*$WRId;7Vtc@r@F9E8>SE#6zk_PqRVixM#$pmCL(VV
z$%!a%YX98}fdXPbB}|8Rx>L7_@`QeX`r@KLA(r;cfiT5Ru!gK~e87!0E%Zy7Y&PS%
z&m`%K`lIARBgQln>9=0sw!$f%T>oE;x4Lc;f%_Sgl9IrW1|v2^)NKJwYZDWboSdAt
zGf3GHrLKBsQ=`2gPDQ3a{?2gmf57(=ft%{|)}Q!UFY!L<MhepK*7Y1wYmq-AX|UCY
zqk-E*1jsJ7JJe$HpI49(6{IJ#yfRzTK=)_2jFY$r`^>j8B~*L}3)m&bY8tgqg-9PH
zlz$LOyeipmm{$i(wl?rJWsUUZjlU^wG=R0Yt?F{pn9%)5(Ndh3k=Z%<qrjOTchNnU
z^I;<M9?J_7ONohYs(S0IMFy$?Ef{h7rBC5i)$jR2cZgf2XY%Hb{<eWC#E-E?b`Dy`
zCMI;wFEcVenp3YY;M!O}w(E7!qdzv7Q@`gq(xfzg`7&$(pVYNfyl*`Slh)OJx%qfv
zZv$h!-idPr^#bd|xT+_X+t=t*#WgD%TX8rmEBHk!)IO%x;tm(It9(kO*uTiXbLY<B
z7hPO3%mvnHEH}D$)m)aTNH`pQ)>R+CNuATjST-AT1^d4Uu%m5gYcrFPdN4LNhTkoj
zMWPgG;f%PDu#kNKr@|`y*7m|&z2n<Y4lnDSkkYw>4}3$lEdPjEcmLwi>D-G@isCKR
zX0ewan<&77NN2tiG~a%tiGsXn{G>4#S$vm>dwy98e(s~o(!R4zeY%s%lS}>U8`FuG
z__-NeVMh=3_+$cFs8z;zS;7C(8128@Apx>3{7Z)plh%8K-XpE|>Q1?V&(>Iv!p$yz
zIHmNp0=w6mqCyIX_=<H#L9gR}O8gmK2Dfe)!QZc3<E~YfCytdUJcz&H=6AOmfaEd;
zG&#-lmxLkzDrmz*|EqU<#^>hd>K`+)v~>8kdBYq@9})kOM1R8@$<}{rdfC$*)$_*5
z#*RishPmVM!#S1UE348&wkkGANritV?ns-+R+ZNJpEmtT!qBB;t9xlEZF2tdRbfq+
zB<6>cg^r>);%4~vAxsw$o@`ah(cpR>#scx4Qv4e1>O3gcM!G&P^0l>WyKZhC*Yn0^
zfL2G3C)81%tJ_dTVpY({HYgh_iF!oWEijvng17~K`SW+sQD}Cnf=W`cr}o2bN~y#l
zQ-~vd^TvbV`}c+U515ZQP?9M~q`W6H`I?4V{SL}yO}P2g<>keNed(5U|Mko8BDX;>
z?I=k%-Oc!dh*2sB^0~^b>pZwRoOMe7b&*KY0?USmzxlJCw~9^G$|@-uFA#YKOqI1|
zB?%RArvC&b_MKg$|AzIsXQw4XC7~F2pYKSCDp9a?j}sS$$dYI$TwO>a=lw~aZe#?~
z!on|z0Jn^|&bhr?=}Y2ZeyFv}57;L$h$v*iX9Cw*`WH9+6Wnv+9<zLPPD<mdg_EjC
zWF9cI*7=P%Y?5t99L<LBFs<#Z*%q$Q*$?{4EkQXpXXrLtQD|yv3I|@+C%QVvEjMmX
zK)Y>Jg#JT@^wU%EnVE_t`h_0+tVikf`DF#2uh)hpr?~a>>C=FKfL#KRcT|Yv;xF&k
zhu(_^^BzBbEDTLs;q%3ljPU9w>{__2rx8L9CTqH6T|Apova_9_Q)p-}7uYax{>xBC
z3U~i#dPW8;6iM&=y4D!x^8KjZ(9lq_KwNAr^U|woUwWa{S)EU*Qu4jcX;tj)2>ZBB
z5^-d2s6Z)@$AJL?uHGKC;1eoezNPoJSrL};hOwv<91!=<GWLGvRdRC8<KQn}zQo4H
z(s66K&-Wph7YrEB*<g}awt>aTIIez*5aO#xX<Q0sQDJL}cI~lb6be|8pO<4RAkX@5
zx<O{kjnj2szU=PE#zSV|_sL#C2!Lp+K@)YTf=P)J<~Q;%I!_dgQ=zBp?B<n8!cAcs
zLzJPIIz2KPMmWU39lXN6^Ed%G{0yxFBz~epo^5T&DqhzV=p&hWA@tt3GZfHvE?KLv
zAUay_EIZEm$8%`TH#;K^wONey86Y$2V(0)b*m74c1DFkM%;8{M7S(2IzEEFZ${ljb
z6$wQj@ygz}$_p!$6NaDUF&_4^fC|Df$L9944NH0?o!7|#2gHDe5zb^<h?7Y{B<Xf9
zoY@JU7zf}*QBm<EnA?zW4()n<tE2%rjS@WzA)<St2gKa-*mu5vUjYwhC4-5<=|IXO
z(XejwC`;C1v~_Q-F`N>7^=fN9s(B2i^OD$YX7bBsC++g-*Tj1h*?Te-$YQ4xAA=UV
zkbtOV(+_<#i_>4fe!Y4*$Ot?g+CT0TxTAHgJ>6KIPFxcj;&#loeEDKc<P4MoNRUPS
zNA@fj!C*rNG-K-^1`*&o6G%`6E2&eqbUXI;7+%?nM2Iu_6RKV}m7C|BOK!b0kQ(G6
zCljv?7qGD=FkWzRaWUvOP08N}Zra$M4=5hP>|k$CSxngF;ZV`3G3D@N!r{r5Gb=J%
z*1n&(Q<(dY_hdYwATxI{RaI3s<Ghsos4*c)<d`#L`W<ORo>osfDzg%`4EQtT=mwA=
zFATV~PhGO??Cf-=AUhKrg$)KI=*1h&fW^GF0WqqqhSkM}>W{T$XFIq|bQN)Aud9Gx
zcai<bf?Rq-t}0jto5HAS$i^RXc=9*d@6IY*Hb2Iu@`O(X&U0s1r$gGLy{#><Ad=%i
zIcgySU1{`v*6v-F7dGxA6yb9k*Mix#I=-btP=-^tR4<lIIaT|*DBlMCN&HAg^E!aG
z0wCi@`k5B(&Yz5-Q7)xThwHQq{F$T>SM|}RE<x~Y6QiRK*Q*71jKF#Qj@Fizc{t$f
z#Rx7qr%_bM-zC)?+F5!NlxLwchnC~_JH17yi8l01{XPi+Jq`gO)4yq?J8~KtBB)1e
zJ{+PqH8P^9mR*zAn{^#lTz}s*DSUg=bD`)!q&|8wQYzC~!GBVsFqKW(jzc3OZG|2L
zV4F-h(+O;V6LJ125yAhrOaF6&Y9TIZ)RrdjXwpOEH2w>B*y9FZ$gFy48`y8nIw9^?
zBo+BTx5t!!mx&F59KqX@Rg#LOOdXFmD!El_Y|W%X6#CDny-uN3J95X1UwUxeg8Be>
z)tSF*)Bm?GivIYxg0)BIG>id>+(vtlB!d5)g+UzrKQTGt^{~mI?FJ@Cw8hr89wc~Z
z0nm3F@}T7@>|`n<@!JN&arHZF4rs;xyj}e1J6!nB7<i<un+Ok`Mg|el%Gw~LM1yF<
z_YF&WH%KyW=^$?1lp|SO8&%kUV~Ugb<UcTt{-Ux-HvQTlR$(#{#a3N45-ERK*>x$E
zE5BpcIBK1`>sSAGmMY0s{+~+)QHmbIK95F@X1zfKHo8h{P!L>1x>m+OIdq|U$A7ut
z93jQ|=GOGz{y44i?)t&(f(t9A;AdMv#<TgpXmH|71*y~(p@-n+rsn1lx||xL&Ctk?
zN=ukOW3M{%QPt*~b&%+;HTl~gR=lCT*25Z7_vb2KzkFG8>(M?HFpwYZ1FLKDqqvPt
z8qYyCzyzkX$d0u#+d%q7J=Il+_4I$s0I;B&Ik4@3@}uI)rB;n%IjXkwc=it^x2V+n
zF;uk?LWak;vkE-<NR3r&<`MUv6WC8CSF~8+D7Jt6s3$&fzY%)T+P6t}5)=Q)Z2Zq@
zefjtLxFl~ca72h~@n10HI@_!p_UkM0PrHT4bAkbaU%CM`*ytH;w0YMC&z)EuQo!;r
zlL+P*gBs<%|36{l`OZ-G(h_jDh|9X}y{~@%?^O~}rLD_l^w$xB8|S}}l9h(}H2*S6
z2Fm$8NxA<oW6AUJM{!zv{$-;haVd#;NnH9UPCz#CUpL-W<Ya^AzUrs`c_;lJh*0&;
zQxJi21fzna-O1ivsfiTHg#<B$1Vu<}vz+Yy2(}jl3GP|){@p(Q&BWByWhg#QLj5AR
zCjX%cT`YLL>-{^c9*N#`vzE7zNwfN<U%KKMV}OkXI4nl&5I|N=Gx)T!yf{2G1PAVZ
zLTkx8G7x^W1U-WGwfz^|MI6k9!Si3$1QbtuC9E40Ri5Vv&-Vt7j*d>=hMEckkuE2t
zX{xoZ5iyvJ8{&=UufZMu84Nuq@w4^?P$#3R+6j)l2?+|WzTy{cZw!Np{ZQHr&b*#a
zrf~@iZu{ufY56)i8HJdLifzr3CQB|`#&<>?X9G8?RwNBd?r=i_lH4zZh2+qJd(eB8
z*4=9GBehnwh|}Pgn9xwH@WlJBrA^Gl4XaLXi|WUerVUd3zKh>+LfeQ-$bO_;#0f(G
zI@Z=t|Nj1k0XR!~Fm0qQn;YK?D|e^DGJhs`l}PJqo~%v`3IQEVgVIle_+E(7At#2n
z7W=ub?s|?NK-5p1ET+Z-sjgb<1Lhpw#t7akz})ikfkc@5*}MpBLvI0+I{vOVnODHJ
zHc|vOLe}i}_fq&0C&)Co)!$t&f}N==ZyOm|Jaj6;>TwZke(APCCLr^raGIypP_ry@
zU<GpHBz}H|q9$${WMpJFo{vtJF<?3M#yj%!WxnOunTE^2>OUS>;Q<en>_#LwVBItU
zGv8YFL*Ij~VwVEn=#<<Q#&`NL2hvPO_V*iTF@CuQ*of{*VUk)FM5V6*ZNJw=KttTk
z2ic&s%_(L%EiFp`(YNhkzAf&syfCxxAKD>m5L~}^5d3REE}lRW9<Hze59Hi`eRZWG
zw)f|*SOq1ef$ZrOp3sLA#(zw*vayYpVFONIiTDkw`WT#+dH~HQKQvK%|McV8+{eyg
z^<4X;&6{YhJq-(^wXSs5eFJe@a7pxM9(i@{Ss8$0QUM7O6Jp?>2MrvZhOj0><u}OC
zJGEc#Lg2v|(*N=y7Vz90-05tOmFWC@k>TTD8vs|0o3DVg6}S!BgHrk}L5Tmry;@iN
zSo>3o<1rPF<VgS1d`j`Yt`Ud=S$5nu{nejeN|udt$SWv7M8V&zz0kwMaS4b-J{fAt
z75#)+{DLu8^kZgmX9xg@s}+VxOE@^-A^j1{AnFPUZgW}T(2ukP2f0IAy1AS5q}XNd
zKK`O7bzJ7|&o6ppcQc`8jl#fJ$h-4vPli^zHe+*r>>dJtL}e2niJpvJ@PwKC4>9Rb
zt|K;M%8-Og=C+l1KY9#$8asH|+K`3iC${1UDx_XkEJC*xVq_)R(5k6}=9w%+O@Sz@
zB>HkOa1-MzsBjX#OTrK52CLi=%clO0W65U(0;!u8KrrEw4a6k7)l=r@6<7=~QL=y`
zyho=7lD4*tfoT3^K3kKaAQ}|rGPtz8LPj4<ySy3bZ~Gl%ygQItW~g(KeBk`wMq&>p
z8t91I@+DA*O({t=1G-j2-H^FSKmnk15cY@Ri;73EDywTW7*j#^C<a8XhFwGkLN?Dl
ztF>Uj(tjGTGsOJ7Q>uFy{7|n6%ul7@-~L9hj`XZ71|fSh2$8ZEE=@t2BZ5%+DuYfh
z{&-3Oui3y6WFP-F?q~e4XrH_RI#k*7<ppRfx}045(YK7zAE7yC@pDL4-o4|DY92)I
zCwasn$OTp@3kwVU!TcNVj^98c|M`F7?}4$hEISJT{fmooU#K`51&ONqGBTe4`^d@7
zoo*AClq`cLU5ZJn3Mwk0BRYUuRp0_JNo=0R&@!m6udm{C2tl4Mhl;%Hpr&roM>evO
zI>;-GavqA`l^+7DAT*vA=z;|V6>@~9$+5z9(}N(|s2mQV4b#LVpm_1Zg#hyL?_LZw
zh6UnJu66N$Fg@|^-DYh^=jE9uNuL8ehvIsFGXpB-j~)~_0>}y@9tV(CV0ddHFFCXY
z81$FKRnS?_x3Tlv9^c|Co>!`>rr!1&hj0YH)ge(sk(Id~$O{s&{ZR*kXKG@?!phns
zUrTG5XZ0yU7>W|<T(Zu{v>$Et#9&duppV{rVPZSN0BXHfw4Wa|GqhpdR;DJ3-$AM2
zeY5BHlsCcD)LCQ{;t>%MkZf5KE)O|;S{}<Ykv6mAogSTTc>Vsv1S2D(@~6<}oBEP8
zBqmy#oBL#Gf|ued%Vsszd%mjX52@tGNcC#z=Z`PnTAg>IE-iYptR>W-;D4Ss*ZlzW
zE+lM)%t8gBlcvK^(yyei<>Lct7t#FeYynqs9jGQ098nCRFD+<IUK@4Hy>oH1*U69X
zt?o4`m+_vo*hCW<={DO^-O>wHf)HTF$-VS$3b5SnGTCtcT{T8vON@;U0)akNL)piN
z4zJ^G%#O9w61G}FH`lqS>~ZR~wWa@tR;BsrDPy^QGcJd*wl;2RHdU$`p~<jnNV*pd
z&pbd&>*C@J<+4ID^{e)=%AQWt$mLy$9c6c)d;szCiRHH5(t~9-^sCH5dY~hW3Y~n1
z5>^7xNE%?yHYi8(-9-;T!0mos4$)Cu`WMDTmWGrHE2~eivuDjwec-MY-gD9`fJq=*
z@B*%S+P&!cMb7JV-8_wNd>*KgIcfpLYr}1EZfGy_VsPi5o|Bgq339=H58DM`_M@5T
z8~7dE2Gz|-^>Wnyg<_0ZckkW>^t}4~K#O72PQR@5^xNalK#Q>8FG0U8dS=6XH<6#U
zpsy(D_#Z(^(=W#GH3!CfzxW#3BO8XBf6Kg61}-2qE(HMj{6M`kA+FsYwk1@ORSf)P
z#O3C&XMeZlHvSR~IX+%8P{`v_bGmMsZ`<N|=GzsLzI1O=+ur|moL+#H`W@h53AgC(
zJSIygXnV{IjtGCsPk=YIO&dPf;rPqq-ROWww+U>>-_^mKdz!dy(qaGG8Y2<tZpAKr
zDbmO(`{rNg!UHJeND+pC;^<RgJIDhv{Nx1XMM`e>CO&rYa>0*%^5CNQ@mB`sB!bg3
z!izuCAutbOHy~=b9OEa}kqvin%!5%>AV*R6S#2UJ!QltjA9P;P{!Pkqm7qL4bzzfb
zvId5(&G%wc%Y;bG?Rn^|I6NvbH=5O1Lolz)wCuz&M+rN9QS!dY?>wESb^V7N1?drA
z%Ha|5?-FD#$+9XY-aCrevaC$dre8WF8)`A&MQHC<K*AY5=tZieWz^UzYa93S1lw9!
z?f64=%mji=pdU&Q^lQ;sAc8Nu5V^Zx5S8dVi0=h!>23rd-wPM6ynmUMvP!E*r=|{E
z%+WN^0G(3~wzg7)ymj*6d?IAUgmN3+t*?Z=y%Q>Mx{9{-Rs?d(s=T|dGZu=hXD;qI
z>Co<~vzz=n+=<$2x39Qdndu4R@7?=ur2qKwBZSvnmETb2>8U};y>b$Dfd22HhOk9(
z;;UgR^WsxT>u;O*3tQ0CoYcsn^4N1wzynSz_Y!`Bd`+|gOLxRVEln@u3i%U4d`o{}
zKy`I}NNG)g)-h0PqQ>fbmW5KXdds);fhZ4AZbLA<U_pZ;Xx$6D>SAd4DJKsMH<uz+
zuLv75zo8F<Sb=Qk*{zVu71dG#^)3DA1isi5!-0Z8neTpBReSI6=zadbMD*@+NSl~o
zHn<NZ>ny8kfwg*;!-X6yddvMoBOnBXI-*TBf*p6;Mv*J2prE5t{_tN|lY?RuyoIxc
zKZR#dVxSq@@#CRjo$eSvzp~4HjvM0Qm#QvJu8|cs1!c4;v>lEMIunt(GSwmL-2o*M
z1R^tx>zPe2y=h0g2hcyS^cJX)yFpIV{d*_UgZRN-10w>qp&%X)h%$hoo7zP+fEfPJ
z_tl;1#;`5yPQ8Z}<mIQK(hpA9G4;)x&riGRHkTdrH811!)%@o-=Jh+?;h?M5j{z6J
zu$trYA<M9RCDLU)KPyWB3IPj<c0<yo>lL{A#=-G2v6V@Tvyzrp6d@V-(4mSgOD}S+
zgkdE$%NWd6xEPYE$hD~8YVC5?tAXI#%|niU)5KXofyl@hz1~%{0iO~EQV^aN6#>>v
zClpAvQmy!`7f_+X&}T`QwIrJU+RFFCs-fNH9ruyvFJJaRw=R=LzuaCUE)lmrRxQYY
zfbr<`35G#<tldy_IyAEIR&)jYNY;B@M5Z5Fe=bXkPWhV_utDMn{Q5{7GW&p`;rH<Y
z#{|vQ;S}(OLIe|obL<jlSh%cs8_Mmv2MB8bDosXKE~yn(4d%sLlWOy<b|_u!Z?=%-
zvB*cfl_mgLp@zK=AVn0Chpol>48;~zmL7vpm#Qtc>IKL$QgWeopME%zr!AzBi7P;p
zb+MHmu~G#+NUav=Pf^e2id>Q<{FFs7;B9xl&|0@K8^{{<<paZ<jO_jIit_TPTGO3-
zR-h5b+jH?xxn&~+?NIZ%P*m_`tUWia|L4mJ*TIWgu&oVcpC=CeZYE5%Ol)~78y*O@
ztcndXaWhif*di4*H9KsS+|0S)6vVK3Adxr}=3qOmp!{!qFl5OZd<x~%l@R`}*16Eg
z<Iqr8c*ji!&^R(Oj^2F8W%Iy0cu>ogqP+4Sb<b1nz!!m`Tx@oRmYPvw9xG25iK|1l
zptrI#KPg+>1`Di|qA7zL|M6oEN)_(Kp^R@$Xu;QE|AM$=KO0a{ME<H1V9z_HLNwET
z0vTLyFho*6OBos-7Mr{;d;6Icw9o_9^V5@IN!zg(<;%MiEL%zw=lsJf!?el$88ws=
zX#y|8MvSzIKcaw@@lR#?!Si<Q956w2C0h=hi7nU<`FYg?*~FRsQHW|ss>m1sRpy1S
zkRBKXjzz6&KQ|=pI30Fca*KilDERlY)(vL*37}Qj-xo!dR^JYgNtwse0@U)|Nw|cQ
z&Fpv2AriMv$;ZbB6%by6EUS^0>Cb-dr-TE{WcquXAszwGt5bsNn}ofDw_hOhG!PTz
z)E_!mq<6_z1Og85&uwZXt+sTuPpEVLQu9wAOIHZ|RF9Y+C5?N<mqBdB-_P~%N>U6I
z+r66v+T&gpD*alqWI85E99uag*WzJ*^$<q81{sF;j=IiQX+f_>8zs_1B5^JwZ{B6O
zLK*0S{&dtq_g>tQ+983}+sja85-2hxaP}cs+T8G_W?(PWm)I+`A}wIR;1g-VJCwkX
z7b8rI0q3DZhji1dEG*p>6*hq)ANHV#GT9CM3{>V6j_Oq14h(sK+VAXJFUMOX0aXLA
zNnHt53F9`b$s14eg=ijRHGh;vi5kU<L{`wA*qd7MGmONve8j=O@2W?-fNV<?@mG#H
z(u<1Xh-*br01C&djGll`fxy|`B__>0N6z7wZ6HC!yEzg+$i&7iEG*pe_|VwcNXI~0
z`n%+-uL}we?)aPwI~)8RHd{J|NvIKapM5)e<NwZoz)e`eFIy$I>s#pA*jnrBn4?dY
z*G#r^aq$W;axkKg!ou6xuR2}VXJl8Fy=J1XYrCCY-tL+$@t5=FI)?h&*_Dj-Y>gN>
zc{n(>vrFk48yeX%a&n<BXN+xaF6&#1Tbf_Dw9vP(W#r$^E^cXNX)S+UM;D%WR^Q%O
zS6|Xv$7wsew7!KQdI}#eA0zSKc6JG4Gh2OYMs^7^9b5gg`nr~S`rAcCx33P>v30a)
z(@vml@w1h9OG>_H|2v8N?@|4m0{^DKzbWu<3jCV_{~QGhcAhcSJ@p@L&OH48^yW<5
zkN+QR&T^7xV8fTuu`y-jf|^9|Ub*$X`V==0FZcg&uO4Z4Xrr^cIcMe1LjMoST_Uu{
zj$tlbsUyEa<tr4u{nfEU;ad;WoH4q2S$x-{V5x1l|A?lJ*>Ui<F9(b2{t><TaBRRI
zEV3+x<5h0Uvl)DgQ0}0nrn+XVK+L19YKNLZA-B+?n^-RQ^mng-!{`6~-~VL_e9e@J
z5Cr$nl3oznfu8oH6rJk=*mppwaG~M;(MXyV)Y;q_LN^rttuN7mLOemMId7*_)eLl=
zi11jMC*MLb2ky0X*G){M?U(Zq(aeQtEo7A{crwL*LoUlyBmw%(pJHTjM+fZ%JbBx$
zi|eXQAXl+mH4Ivx#MaJDRk0)BUyHENBnAyy+@lRpA%0)ssd7j_ybgs68?TmGaG2Lo
z3XNQDjeeqG-@<WYOZ1%soYy``(3*d(7d3%XvD{}zTLCg-TFilyVDD3v>FwMop(JB+
zjN3=9suNJIxK`PZ#c>=C2}^PrfzQjGxp_S1iKcgAhC_$b(U${H0R;lYvW=BI;hRC2
z`G}Ozbx&izyI2NITml4wW}FxCa8xd}gtDYGHVZr<sPOo9HhC~Ht%mFJXe&gz=vCT<
zjqJWI3d}(>ayWZu9rMjBv&CQs^~rRWkDf!lgUfrB)ZajDg`*w8w+?w8ff^IWL6h1$
zc;LkHltCf%SgHx(e98MqGMqcMYg9SfHfQOv81~uJLi{eppk1LY00jGwk?q1Z&e4r@
z7JAGZK;$74J-K<%txxH&&(BdykKBZWD%GAMMdPe?dsggMDC4Ts6Ab+-+R&-rDIcfQ
zZ0G{uf_y)hb|&Qbmu~;#gC~zD#;`i%P3ZyC#}=7OQ&3`as;V*8@iIMsJ`ncX6F5`@
zhbI>onhIACETmQFJ`iqG!_%bBtylhN6Zw|g9@_{w`>=M4tC0UX04?p4_;DxPQ8`Yh
z$D)fZwKY*<g?B{{jqk#lSwI^Wb3Gv}>kq_ZfHy3_!U#j<2%Rimvt|x2Uh;d6sZF&-
z93}C>D<A13*m{dlI~q&A@I5ji9kl@61Su%q7e1HTL0B0ltU$fl6sxk-Nhr@8r?!{!
zz^QWYo}%vYs}~EGPS`=fttkW3U(|08DFKp=<$(AYD3g5v*4QSBDS43$U`HRVP}3{b
za!;Kc3uWnX+Hp9F?IXN`AgVc!V{>qfG8^h)pTiv_OPjS)?k<YWg45<ok;k)U8E!kJ
z8jS)o1tlOH>ow=5-*M@JyFDmZ;D7*Yr(;Zz;sO9`IuYKOhJ*Es*){{o8I7tz*a9H(
zSC{S&0Ldg5t}OW9p{PzDb4*Rmhm*-c1E2GUUMCE-lTT~guxcHO{<;Z6r!7r(8<XVw
zA<7eS#-Zzp#r!NpYBMgux<49R2GoMu32<?*qmJQFZs&z-X}{dfP-vq5cEK?<gx<bc
zr*BflJfp#cPN3ZLquFp$R<b(%(eUfHNaH;)PX}EKwEL89yZ0Z)UG<V4hphSoP7_(3
z?m{i%qj2)s-|My8+;hj&`?>Xqdxb^CU_Ez_+yqW?TOQU*QKyGq44J>4#%H14r*TPa
z*)~Bz_|>J54#}E|2~ZGRlj0iI6-WI0mVkk@XC}2b35}-Lvy=_MF^t~RII0`3N<?TD
z0Ikw0r~;WCUQiiQ#-3DvH^D&2nk!rg6Z3HIEf9$q7Fu%Y`vl?oec46O)yfUH4Yh*+
zc!H4D7pyi0(zyt5|M9r+DPXG)>LGNIFG}V8_H#2=3t;)XZbUeA2nf$XzV;|+xj5is
zIMB@ip$3ic!DSR}s+2@}X-|~FJQOSNJ7Gw3GY6+L^few--}H@5<g=X~1W-QuVdk0J
z#PPs;9som$_C69%e%I}-q~*rb(u%c=D&JY^OYidK{Lwo{TR{Pq?75S%ClU{B)F8S0
zw2#-$;gcsM;N%G$RAsn5Xr!D9wE`v=hEx2UvR{<T$%JynOcmbstC_g`y$BE{yd@lt
zMYPiKqPK@~9}w4kogDO~YfIo*Bdzh$+$1d3BIt>4P;1H->K(V4BqSAAQH<w;dYXcB
zx#FT?HQS=b0f8m#n=X3+DLKcI5>Y}8*K3cJ=_uBwehBKZjMLbM1rh8Peh&5#eu9>Z
z0K6)9H1iwU88jCGR0qvm$cx_@P&@XfD1#w0zf!-$EbSAEQW>1zV2TnQFC@69JU9ix
zJ!pwUnCtQDgrS$1wq1L`R<PJUzVbs-Y^bx4fFReV#*Xp>?ENdRkLTXAhIE-fniuc5
zX*Ww1uUrQ`CER_FX$(|UWrU7BnQ;<NK(|5LhlVSU?$>jc(bdoGIweQznCf4lH`DPf
zPPP|%^jz&YAuUizF#h@DGZfGrmAlB+P|vdmoleHm@+LOzx2BZ6(vEt<q(@t4s<gvS
zQ&@&`xovgw7N(^7@Kl$NE61I|ZV1@X&qnDq;vr)Im2~MhR0%oc4uG1k&BtmF0^Xx=
zqIk!^r!b|a)=o%GuOu9I)q?`v{URbOM@1*^PdKGRYI`~P-EB4@v>0{=my}etJ9yoY
znTCw3=rq5|ig+ceA^^G{pMDkXZKz;zE4NQ6U&x8fjw#GCU{}-zG8sx^aEkl$^)R8a
zcTDwegMCl5H0Pr;JWjxlkWiWzJgj0Eq2N@87A}Y|1E|&c!A`WHSuk*?jzJ{cqLKD+
zeY>uzshG6na88azbBO7Q26TvLB2q)=&R!GYV(Th4>Na}$VV}gwYhx)>ZRZ~EaLGnN
ztG5-*Fd1M&kI(0*gI2zVM_|<SeWDBY^Xc1BnRHV{YdNJr21hfG^hZwtm(+xW1dtw{
z{)0j0u*GFipYV#iN7XOoZ)4YJ1kfIm3FnAR{m~*0*$2XNurR>`cj-khu%v9FPDAaR
zbaWH&^{$X_0ph9^UWQe>*!P-PMTmRWIR%K3HOr38o#}$~q7{U2*T~&*nuDw}u2yZX
zR%daDLT)d3WxDJR8mRV1SN%i<?m9i8?8Tt-gsjJcd{)zNqLpB-YX@V?C{A8|(b$Qj
z`pV+ZL3t<pmexC)T7Wkb3J=0Y5n`GbvY60jzpGEvb@C%RL~e!ifDxfCdNsHcUk_9m
zao2d*QC;t*a%zCP_&o1nY^J-QfIz`?#Q4`4E5aRi*XN~tmj#5+WkZ=Tg~((M_s+Re
zzSkb{L2(#3q^r6d_p$zx5Sy3SJvPle)Xr%|>WiYf)yJbzu3Rf|P9aZL0AQ9}DkG#2
z7E->5qiWh!C(@*=U6O!9bW~|NJwaq&6<n3{QB*U}K-IAaEjU6g^1-TDTEQ1abrqF-
z{Q)QMrQ8yBe3RA_HjPp_bq;$eluO@-r}5R~RtN^016RCk*Ud2%+lz&!WeJ>x#y&si
zYds2PEsq{h>G{~E)tC0*7-IcKT(T5OaOkqPAkN-`yT-<j%D&t7>lI0b4l7W02ER{g
zTi_tcjXa@q2AL5kV1Wc=+}PoOJ@e#I;drrgPOba0xvaBOh1y}^i<xfA22N0ke2O<2
zZ+R;!BGvrp^^+kKh5%*dKc`&Ge~wb)1YxJmOCR0^*gH|#1*l<<ZP$OI!P}x280oGn
zzw&<6Zq#cp6^_Lkub`-+57Co$ZO@&~3wuoGY<KL*^m7~0Z3t{g{RCC9qkhb4f)etm
zj%fuARs7po_uXlr$-p!)nw$8J2v~B|Y~~4%juX-|>AWF{Ssmn*{LdCNq0i;_OG2YR
z0nK)Oe`Ax*ISjRf(BvF+jEh)!vjk}_&Pxmr?kZ}z&BPECW+Zm6=noQm3Vm%X0vS}(
zy}QT2LB`cnFWp7jVKfSi;~;WP5w44bPG7N=x$-GEXE?|mH7_me+5zTeaSg{a$BYKW
z;aOfC-B**F-I$2H=xV<!2D&2@3>{&2W1!Nsb7V)=1`<|!r287Wxn{JVH%i5ZSy?2!
zEZ-9s7wP<Z$`_rKiF)XbnV!_{%kC7{QF4MnhHRv8w(BB<ku_~UpKB<5w^gXO+?a04
z)mzTrW^b^E=oPnxBIqBb4w-VSN$GOEDBH%yCts!!=L$ws0(HEr(I2oyp`&D)DK(0g
z=qMkL%>qORN4Iencd0|xN|50)OH;j)lz)bnllg~<!V>kL*#nVa`9Qu7%xL8VtV`wS
zcUG<WN)3m01*hF16C-doQg$XUJuYhKTri7Vh_DiYf;n+{qDaZ@-UT1e;bD;OcIOme
z{Bj>uV`LFZ9gf=T)rc;RJo76bCExU=ed1O$3j+bRC7C=-vYY$odz7Lx@v;glg~MQn
zgL{w)7SfisAfr73A_g^<k51V8siZ7B*Rmf7o<`Bencu-WaIY03xb_*>7dsYARkBLb
zS)HBn5yNNK&e%RuT3rRvt_$pi&h@G`pLmd(uS{aydMGhI@^}Au%!4$Ew;c~{)aW-y
z@uP5A5I#x0eK*!%ps3RAk`MP@bewd7f9#<6-BZKI&tS!tuJ^JxKtYkIBTv-jbous*
z{`T5%4h#^>u)d-?5P>U;9MH6Mr_hl6C4oNhKN9he9$Cbt_%n3eHY^0jHR&{k@O^_=
z)bOBeyhzeAPor`V1kwg)PwqK-i6SaMgg%V}QgS0NyB2Rdde(wL^wzNIYmQC#mReKa
zLNMzr)gHmwW?kA&?YAAyS*<1s<(A1N9z9E7%JWP;CqU|<;K$Ltgap>9;U8e-a5G>P
zeJ3ESrtW0&ER_P>o`j|E`g=MVkB{NfG#6)_IVh}WEZC~KP1}w<bed6#yfrANMN}cv
zo2UgLWr`^aYl|F4=Y=``!RSC|!Ac$`>^?&uz5M+HX@cHjRtr$vCQ6#Xbv))r5o{=P
zsnKARIc@%_9rC(iMO=R{@D$g}9X)l4aa+ybeB9?B_YNEo7y4HmC;Zb!@5qLxzW+4U
z<weMG)}E0~%`wmJ3;qVNXHQ8K={&k9&<V!WE~jel8h)mJL)+daCY`qKl;zZ3wB$Zz
z*!oNhE$g|h$yO<#t<O>Rp}r@Vo9J{u+}o9*HCKFR2THGya93G?jgaqMkzUS>;`m#Z
zLd&7}A@-ndE%E0~5LLceDM`6-kHGrusxVEXs$&z??9PDhF;1@xttXhH(jmXAjDU_h
zQq#FqGMpq@ob}C8fR?0Xr#JDrFU^&|-e#KF^k4@PqIs9pLDR!tpZKF1t}NY`5BYx9
z{fgf~mP8C~yz7g_Gfo`+eM6LjKc1D-=PI11l%0gWWRez*!{4R)P<B>%)K}*PZWoIu
z)E-k6N15U@M>3Em%pG}K1unsGj<vd;EUoD0{+bBd=>q?r!ZYpeS}mikDU*^-ayp!^
z{q8H9CS^E_?3_cc{=BKl(kFL_Nc&9}LqSXW>Anwbh-XmfvXT~*q_Y)HAn{hYCAH_S
zFRjH)9$uYbY944IKRTfee%OPhE$6N9AtkLTRi0jS)>3+*vZC|7A>xA9KHM8(D2UB1
zeUM#PIY2%J+e~s^%iF4xw9{whkfN@^nJpZs(BZDpu}lB(VGw-hq&!RIq{);IY?Ind
zICU^wu0`H6gA@<{Mf=5zcj+RpK{?EObPWnkGoD~dM4i2qR;P}2pG%Qf_7br1+3|hm
zD$O@g>5Mh^Kc4}I)UXd*GPKY)jS^6uhRz2ve^gmSK4nk1Np7;%jwiW2zCwqzgg-{<
zeh@2i>}4(9$=m$Qe<$;B5+3rH{Z2?!UzF4UJ&Jh40vHaGfEz?@vpS01Q>>l9Y0&B&
zEaLD91t>$OlW}kiXQRA0DMG_+?XTh-@}jIW9S)?^%`Y$~Rt{MXtaAeO$Q}W#Y;_qV
zBw5cG`@Wslu5XeO4YQ(5czJ~R5?I3Oy<?EBesI^MHYIKe_QkL{x7N||BX8Zk@9dzK
z@U@?Z>|B*5!F+H?dnoK8aVL174&g2)3AA54@CY86qTUZ?zQs(8V=kMtJOc!SIkFOm
zF*kLh=$yVF|E{LB6E?_h;T2Re=O5F}Lkx2O2JzI$A)%VoT7;@fhaARdxy-G2L+CO)
zQr(&OwpI@gW{J#G&9mwIM13;+UgX2u`YJi*)BMhWF(d#Cou2*=9)8Uo5{DPK^zMQ*
z-Qey6>|HA?QP_8>RuUG|(0%pziu_s1smiaOGFnUmW~Bm&zKWszpQ6OXxCr+>ra13o
zXeYbC*xxg)^4dYAeWjgwIF1IAWAi0%o_i|l=G?mrw&}Zv98=Ym-Jn!k&rJ-DX8J{X
z46+qMw5XVmNf$H8HjV1oHkcMR4c~pd(2(N3cYcRS`1C*?81Yt(S7ziIGCrM9d3)H}
zeAvP-)-SE7H*H2SDtc@Or~s5{nrP!7>HdHwb5vMFxv9y2zVNhMC%Iw%Y>&&tL++Y;
zdGUM7Pg&dD7#f!W^Jz2De0tX#aPbi4heD)dT+?e0^3VuOSb>Cf^FteDdbU$m-XldP
z^MA_izQQT;4r=s91*fADa7MVqH&Xtnf>HLkuAfLSL0<iO(PsZ|u>Y~nBsZ2*?}Hpf
z$LGSzEozqed$&y7Ag8>iTLN6nUn6Te#3G#q4WyqA6&Bbp(yPq0G6>olg&W=vj}UTo
zF3oNNGtR%-QzYVba@?V#%EPm%5BO=NMb#{4-k9~htUx+N%_2J{IcghEGh~M{J-D*S
z<s0w9QJul%c5`RJ`h+>AGCqPo9@aGo$sCpnqU8gk_-v{s$4z*p`H@41lnUEz>t%+W
zmjDFd<-_v2`ShmlA4nBcHR`5{<=jS@5<G?14%4uF(UcbdWQR)#f8bDJR5w_6K6L8W
zp=c=OtF&?pb1+!g6EUo`P-_i5o&xaC65m#`i1=TBz?ON>WA{S~zPVHZFe2=r;nWhC
zl>4$nlx_`<q1sXZi0c1lmL?PKf@&XE2rG}jUaD~@kaxnv&dS-cC9t-UwyUnb;r2x~
zIN+~yTUMa&F<9PjXFqP0oA;FHZHaR1zAr2iGFIy&=iwP$fT?jFEyH&Ot68WxRUyxN
z+m<C@S>ry>MDXyBCr|zcPE*Ud5g<tr-`u31hkDb6P)Lsfbr$CLrzPLc)RrK41Sj3@
ziG18foHYLUY`a6JUo`5k*0dX!NcW0^$%if`vg^lP4m(JY5Rw%fj0UA{2r5_$5SwnS
zYW1K+c)f~}$5R5NTPfM&C>SDUE_ovF5AQ5>F?=U;RuUde|5uZeo|TQ7E~9X0MJzE^
zkD7BZV~Bj%#PcMo#@pq2pIX$5Pin}eE3wJl6)Tavotk&$9|?@sTgr#BEx_Qpv6uj6
zC-9hE-%c>`)W)6ftMds9{b@DRK}RZfEQ5Et>pu8E6WUpY{1=)$2Sv~6Y589uOfnEY
ztKv>%04xgjs6Et-^gS^EduV6<UdO6@Xeb5Xg>vuv<Q#8-cDEnTy}bL2khitd{i(yE
zZZvSFg_`9LSbJafQ?KnoHip0uq9a_G<jjR$VN9}A(`|mnP7~mz%cM4E=l|l);sFpG
z-XiZ;qP#3`mruPd-SMul(vW%=XAR&I!y4?U)*l$gpjs)3s`U=@CCD6d=xcXq*1_!F
zg`Lu#kw0s|6#td&ckl5cIdGOVU`sso!1)tT)7KD=ai~gxb_!j}8UPmdmI_zz<&GHZ
zA1tzT8&;d4oI>#G{8lIn;QSUXM%Gt=J|XE=r718uwRroc!sD7#@eAG+b{ePBtgr=;
zmTkv&owJh>tf*X~uH;O_Cq9f>fzp!_`>dh#q;b+JgC&Aw2l4UuJ(eHRVE1PYGaM+2
zP4TDW?V(eY+#wTwsIWAC*f~Sa%O!QewcLE92mwGre9F2$Jeg+ipjI@K2lnpG?F<rt
zES=Rb+&!EL2W-ym%K(?Xy@Sh)Cd};Z`IN=^AVt;VuC7Z{PYUToUFRUlRBU8S9iKFe
z?6P+qyC@mzg8U{vvb!Te`qoH02-FY)96hVWAS|T1#e=Y1FG$}9X`4(~%_gNsfC{8q
zKKci)+6Y8QyZ6>>hz&@kE~~@Nxywz`KE%Ag2-pmqES<hIIsolvlAo}<4PK=0F8Vbu
zNqLkjbod1&UOVea%24LeaPd#uE!n&i+PzO->sSYTeqCGcSqr@fAu+F|l66yqowJXH
zTDsAdUfV`6doIp=f_!n*Ny)<SM$~+@p*;X-2wnq$-j-H2R5DzAEz1X^q@o|>EIC)s
z<R{(t?tWWl#{r>qC`xzy57c7V)to9E(D!&@-I+Tyk?sqF*PK(!Egwk;hm5uQjTe<4
z_s4^Y@{ez_y-C#lZ`M_U8Ng*hwQs@KZ&=*wA0?Od24l0X{=*;4!0D;OyX@X!N)nm+
z-H=aFGw(j}%#-qUUJ-PVe?#T*E!YzSd0sS5%}^+#aMEiTynC(r5UrJ|Ckd9*P$~1B
zb2lnd|3u3@yQwT?B{=R|nnKoXjOkKQbR4LSC{5YpzZ-g1m=F61I3vU!kX1SXK8<*Q
zofG``Gz2FUm^45jKwcWV5B6(sjJyh%DRY3XBuxdYnY}r8s}Cjhu`=kD?Bp}E#P+8>
zxJUOwVO*9O;8w-O!(yYo8y2e5F7kTuw!<%cr@9cnloKlwqyf%kL{CvUJ6=?7U~qfb
zSP?eCS5|fwUn+$_o<3b<JNeT>R~XZaY#CR3_k2g7=lmQB6X>1Fi8!3)jfnztO+xhf
z0)9No3{54;f!$$gvP_cK&+mh1y-NxkL?(i;Le{<Izd5INSw8wI65@!sO^*^i_Na1R
z`OPa07(!+!X1e9v5@43TODoyQxk_4Xc|IZ`B?jRoLq^Z<%lEo&g*uz#QfKM=iXiJv
zGV({LVdQ<M&X<0oIX3R>s>?mi=>|#JP8`-w5dcHX0?;9u<EP@b>E+P01}c#;$slHZ
zCMZ%=M)jj7G?$lLY7&Kn-}0ME;K~@1bOH2cR$5{;4F?>+U;pATHqAdlvBB=l-8Nz6
z2|lUcum<U1Poz3a4=roD!|^QB#5WRsOiNJF*OW;6cM!TdRLcdK{c{ckz|89eaMdIm
zx$?vDk++AUGQT(Y-xI9#Y!GA01E=>|heJYA<`5nBZkd)rs6LCO{CJ;n2goQ0?IP+5
z>Hv@x7{y(W<u47+K5T4pnt1in*IU>c-24M=7nr=T;jU8^{@>10GQ7g4;{b7%?;S&4
z)37tjV%LEgy3S2hE+XKv{gQjP8Hb8xzD^MWz7f(4&Uzu#HLmr-HX?|Hlfel}7J(u+
z>3fSjZk44?(aCA?1N}v}3Qu>B;HDApDZ6zK2VN44PiBi-h+}UEEa9Ej7G<~&;GVau
zshCmSvpLx-%T+_Xv)Ux(t-yS#L1?K|HkC0W7?U72>_xGE_kb^SN>_uPn?hxt42txw
z^xk5l+ae&A=%3pO6?dJSuS3H~NN85?WAdW6)96&5*<L+FKLh%pWjOSOiwPj!tcj^-
zdXEES|1AfnZW14(3F{leP6+E<GN8Ly+J+zr=r2AMFYCrK*)nlR%tR2tO}UmXzytA>
zUdx~#cM%F8YxG`EVyWF_SljS+T~Sl90u1#@b(;CXCYFT6-J+@2PRas9^!6nJ{ot4?
zTIAVHSyd-<&nn6^KkN~9Z~m+h^(qs%w3f)F_wUO)PD_lf0MtQT=I5Z3;%0B|Zr{hO
z)L=+%WQpwnzt=y^4T0274R-<Skw~7|Guxo!6%~ic)i4F1dkIYV-NILCe9HG9%n`9u
z?(H#vu#~q4zf-RnOAO9qX|i!fdnsjWf=5MtJ*Pe=AHu@Qr7Tj5wzOMxG$5P&*3+vs
zwp4;+?|Pg?TCT``iWCvzQU}jYZy)Z&9PQ!$RjZ%D9?#kae8AN|>=XW?Fm9SZBMLJ;
ze<K%g>qE7>r!Nfq2S%mmgH#eNVoodXr2~f0)!io>6WChesdMXqQ?ew3Tl%a9+6)z=
zFwaIsP@6k;*Y7+0)k^2x&V@{$ttr_r2-^>BXvfiED2N@u64vqMBD>o}5vGh#qT4J>
zxu5Is6w{;2%`C4iT_OB(WO2WQ9zop#LIZ50MH8jj;3}Dlt)O<L^Hmcl{1_SXyNzds
z^YG3D9_uiUOsg$600hj8d0g9<d9%E{tnjcK&1m__vHmd2nqtq*9-bfCMEX-_ho`#W
z@aLgw3c@(@7KW#8XLZ1qK(7*Q(V03rP~<Aw1-nqe+(&kal`T^cQEYJcP94jMuujif
z+Y>mg1iNmK?`CK@nBfB<?unvu8#jnVjSs#whiJE9xRr!#6SajUHkeU;`87&@hHznH
zbDY8hmydguMh2Dkt7!QZEY#kOL1-jstp#vYQ>;$jOI1MyooLNcPtUZ2*ezP59B~8X
zD`}DaS%UUXkixgHhQhN+Nj3YXv<D4ycxv`r15_j7TS;M=b)c!{0rW}9nYo)mM1`!Y
zfH@OqrNE01^xr1yP5o-Snzf!B6`|_~Hle0$5t5c)E;K3n3_k-s9Fki`*XXW+8`;--
z5t<Ad$3-ai1mKm_t<KGiBk-`!w#iWkTn5+ocVIs&JbUU5M3}Yg8Wcp@-67w+T{9wj
zl&GI;hmNg$Foqo{G0#%t=*!^NQqSke8d-2_f~r8MG<URT>hv}SkDI%LJT7csj6u2F
zpv5m2x-Y-Rn*j=Vsre>4>KK%zo~m60{YwrV6*%Fe<|D5J!3b=w5T}_wiiKrD7<y*(
z8y*THp<p*sC87b80=<veGMit-6_T5wc!-9{VY3PZax-NitOEzVLWXGJWEP(*#ku5M
zD#L&luz6<yueOCF1$_)LCsfmoptzBL-+i0bs2tu9>8uVgE8-XYN_7r^CdMn(Y8KNV
z0I6Zx+oJaQ(|5>VFOq$J!8~U$)?a@|?>x#Q*M5H^3TnecIMI$p@E8c$${6QxhE@uu
z<Y*Hr4?Q&50?E%MhZNke-YMGllD@;Qfp4W!so+aJRTdi%5u+You0L350Jx1|rP%GQ
zi=`6;`&>=IGEWuD((EMPL$cveMr^m!i&VAJ-|b>C5b<rTr*(xh)7T%2&1IsB%Td2!
z!6oag^f2xj#wv~_7C^7vnN#Yf$=L?YF5?KgF;BnJTdmQDHiPt7e8u+qv{sjy992NH
zOxeFfl&^7=z9gEjI)R**Tp0NvDDFDtBW)BHVZS{pHHcu{tmBoDj7i|TKT*>P4p9kj
z8aTZULKk#&!$e%{RcWOg)yCmaY1?SNqO#5%WGCu4;-P#1EYBfn@|@-D&j1kwj>Woz
zZa>V=pK~?KOA7zf^M!=%<z7B1jodQnCii?Cg-oQdz?31Ok^yT8N)1IcnZwNF+^(g|
zK_pz$C$KUz0h88o?(tKIBzmk2c-Xt(Q;$-}?wa-Xk2-(50D*x8cP(7727q2>3VI7$
zd8_j?QEc{nr7J?y@ftdQnrdu6rg0#bTts<6yoKFkzDmzo5Dp!fvP)CqPWWF3b<*HP
z^HnWmFPxfu(-bEJk&R)uUj1MdX6gB2YY9q6Y+%I@dA{05Pn2CVqo8VHTXz~W<9yX6
zpH@0n%~DybJ<^czl3fIW+u-Jxpn<PDIl$j^uU=f4N%2S$mCBrYyw!N{85pQJ_I{zs
z%m2VFO=u0=Z-=<^o)fYmxey47DP_tc%@V3GfS$_l7MqF#)SU5UxxZ0FSCKSdkFb)Y
z&e2PfI48YZZV$c&!^kHdiBk>^Vr~W{LFlnb-9~0mGh(nv4%))@p@!OXLg^@V10uzv
zGdkd4&C4_QUq~5)@J2N-4n5(eN7wtC4H%xEUs1}AsJZ)0-Y|;HJ#gkYp+Mt}_&eYC
zeI;v(YmLCb2bB;#T4@P2%fWVA;MvBRxVxymDM}QPpq+=B*vj+jb`e8Mw|1QTG1$H<
zM5Q8mi->YFpHzNs*+`>17dBY5^%6MGkp&*(v)`%#M~^Ah`D#91l<2Ulub7d+rkc|~
zfzj?8Z2I$Ha(d<3#$wXGJ)G8}-yF_w4yJgJgDZHfG-o45isYbk(MWRcSZh6g2J9)6
z5F`wk?bN7Hi)V~bB$#)odUckV2TRr`-fwMS18`C^%VLAK`-zz~r-E<KR%7_-scG^&
zcoIusfm3idJ4gKh6KDm21-gkS^vAH&g=z341f{sfOc?+%WmTIzV+i0TQ71u0NIPXQ
zpbIXg`wc&c?J?-y$^0NGs|_9C^-KW_HiIuN#ck9SGJhhW$h&<Qe%u|OTWN>_a>F?Y
zvde@q3?1A~n1T~!v3*6w1*Zc>acRnn_eGRVAdvyBaj&;ogO`d1m^qxC&LlDgF$sX$
zi4=7!R-{swFAzRm=FIW()4SsdFE8(jL&1t5`~ozA*vcrEJ+a9R_KZltO)&Jfbx`om
zsr;Z<IQU60I#bofj#<;G-U{HDPawXoAm*bGWkdEqO+NDLAxIqwKxsmrHyrjB18_p;
zO|8H{xUR1Nb9Fe{=zx7h%*ehoF@{V6vS+mPLKwmvt^gT8B9<t`IUwPMuUex-atJK;
z{W#$j%X!rotu_zgBp}8#9PB<~+z9Mt4%dHZL8gYz@4NrVO;xiD;YFz0cYV}G!*k2d
zR#3+!tUGJ1TaDv1AmamcyqLA49UQhv&by@$GVoOtm-elg|F(h+Y(=+q_KP^%KGL=q
zD#>*6q|zBxqW;PY$HEcxFwzOZ6+d()Ovfw*^g$uWrD*xnxlJfntR1ih$lVfHnkc!L
z<_k>=puR*QouOpVj6pbTbj2UyJqRm7QJ<fvTi;K*81Og@6QmXwZhfIob*nHZbHLZ{
zV1X4*$rU19wc_Cyba?Wa^$#GY8f=`bSSjKqC}14KPzMJt3po7`Pip7d!_gRp<a?U3
z?@yjL8E2{G&8!g|u8t6D3XuwMOx1u=f2w^mX&0VU`5(43uq<4*?;CtsD>!^Y_t5$7
zqS@rByiI_BG{Twthxg$Epw(+(E9cm_?BK(bjM#n8p_XNG`j-oRc2>ZlG<(huy^p;V
zkyFW%@K}s;3l?W4efYd^^R}J)xuT~`UW0AU4f8ODoi=OA+sE8KiS_($?CUr<BmWlV
zl_`}$O@C8(GkpU@`U@a#5iGhnSj7O=)~S>@UK^;6={j$C2Go2TK$<$zhvRKpfvi;w
zl`u3e#(=}Pk@S<``cEA|Vf~%=4%<n|ip*SfBB1kVCzVs*!(-l>!U;?ULg5|RLnU&z
z3*fZilYVmWLBB#a83o*N296|Y7!MS}`+aHoYHmmcd=1S9L~fBlfJbwYA3AT|3GkRg
z$Fvr(&z7_Qmy<UCcjnEizJddKe)K}KPzj|F&~OP~IwuHY<Uw84i@_&2xe(F5CVvX$
z9sspJQm+9j%pFP%QTU_jWN4w*)PUJ7ejNc7t(E8a<_)?bI@1_nF$9CRxWRgZV$4OE
z8pt*Rs~(l9oEOu8HMx5cUETWf$i?7r(9!yn8sHMz4Adm$LF{ygMc!oT)Xt^rS4kHk
zeAz4?p)7IoW$KWX7W0h`oOtlcxGjjI9Ez~15jgS#xYhx@N0D|jXm!=s`B-$SoT*d>
zU7?x}Nso^oJGAIOX(;I1Os#%7rw`Cei#9Zu_!!_%e5Kev#6qT>b1-Pu_9+N^*}u4g
zGM-HaH32Dki-EEkgROix9+Ru{rJ&<ZITPQVlMg_Kez(tl?$}u$5YO2uAHY`%X+?;)
zwxeph%pJtAaVa1{PV=QubmGB071UdqdnM=sd5%d3(3O=r2+KH|_uypCtI|*V*i>!1
z?;i~op?M5&C<cpbi>tZB|NS4s=g>RfjzNO%3GtEnfvE980~G2%(gFuzKY5OR4j@E5
zfAmT1t@%Y>%lLX{$PHjU$P1Xn&^Lrw_oeK*2>EE<m?LP~4YUH*dM(!=p8f&cqKTEo
z@h2<ta|}?lV+?SsIaydgmd+C^M+QBn!#e4%7$QwKSnwdot*@|nzc=08oTUPeE+>nQ
z(xQIzrlfRU%a=OS7HRxaF)(Ky(ER03Y|EAtrU)*evr(4Z5&>!;^LlMwgfvS|&HZ?k
zHUV0f0D<70=Nf|u)w@;SU4Oo*;yHe$0a()f*xNBrtmrYz4nS%+8CCGGJKr{UfgX-`
z4|%bO(1@PeDb<M1$eVS@$6ex~QlXCZpo_MKEdl8z=Y>rJkbJe*;`ffoK96~5l2Pg@
zd<@rn8kFMHGsj1m23~+&bs^$5wo8)kiqo-9fZDh%JMt#9w18o!3JW1VAzbd+4Y`ik
zDMqX}#_6(oBf!NaLX&{dgh61dZ`_*!QlQ>qz<~Jpmxmv02WS5XMI>x+^TJGP4}j;v
z7_ROTg!I4*ID~*XJ;JQsieB?f6*}$Y<6(8_gdHdD?P4-%GXsY1V|aj^y1$GjsRMz7
z1yfznl*lEg1p>%bP0xQor6%Ht(G;C7!!0H=2zQvG_IN4bGc+jlVagmOV73MOVE*)O
zmfT{^F>tQfRmLG^&498(Jv0EPo!{}H?)=e+bnT+wNW=p|5Xby#)(GqdXxzm9avWN>
zp|s<1SkMRT1@ERhW%<9~p6IByDOwu^3<^m9>@D>Eu?qT`Fa~W%s`q=6=&`mYm*=UN
z6W};-VsO{Cqu~=UjDdOXD{vo?G^8iJS_*?6U2qI`9-e@VC_ucHqR$6{z<f=-^PTt>
z;$opO4psl<^^S-jzgG0)`e#^RV2aCsVi5FKV3xJ{9)nRJCRiUcx}Lb6#PNH>`$&@{
z@d`X@oo|Sj!J`gr-$246>z_%FA-#(D8J=41xtqkMz#YUPKZdDn<dhF2hC|dLzW)O0
zvfk6=#iYO7-#AuA?E2qV-$NWLx*M?_JR1#X?K650Tt@tz_<quGRwqWB^Exe77xq7;
zERkeHTC8&<kQIGin;Cl4+FM8q@Ly7Pb*R5a>vWm4qU+E6ML8tzxQ&#9uf3;NDYLd(
ztCU$^Ez;9q`l}RJU8pr`Aj<sW3F6a0(pH5H?nSR7a>?qbV4>D{?yq!qs=GY#Y{TEt
zgV%U+m0G_jM|{cuj3><-C0U5GBdzIcjU5a&Gc^muH^E^3L}Vhb5r?{A6`HNbC(p=`
z5obdDTjn)%-|Dp)>V5R%>BjZ3$9jrw4~^LEw4;Q|@>zNQ?5gR<>Ib}~69nz6FXGZi
z`|3b7RfRdEf>O`nP&nz78libr$3Z;^7tUiE{E2<IIQ+0#2b?mIb$H2@+HmJ|Kjq#w
zPsm4Jae4j9I?LdcYH}s}Wyr<|{}y}A^a>?Z8cWS{!q6W0J)_<?;fp8O3mLJ};jOVM
zX$Dqa+s9CboV47zpxlgfoDBfQF<SaG9f@JwhL`&Ap4sLs$1T{{lm@3QKyUFDCMGPo
zUx(DUEr*9MXnDXng_6s?mt!#KjK?2ulYh?OL7Ayl-<KEKyCqcU>V&#SYU!K3z6`cY
z+N`*{J#(kau$7aZY)^VHYWTU+40n7Phzw*O`~L-Bzj7A&aR3ABvC8U1ejDXGAdn|W
z;_pY8vc(<TBngH;uIeJvHijgRsM%O{Vh^an_4)%vv%bKy#Fe&B#3WH6zF8L)fi>A4
zdH)wZL45L0#lF+Ht3x<xIxO7#AajWJ=QFlYj8mQOJrXk$)l;i8QAF0ba!rQ*GbN~u
zUPXNTNN?f5#Hz$aXgwzVNF3WQ4Mln?&|h0^-#i^@6CldZBaD!VnoO$MUzwh02WTg_
z)WttqRr>2Q#+}+oE~3!Wy2{!mo+OT@mq@8pjh<Ed{7I3)s7Pw@zKD_Hdn9IN+pw`A
zB4TR_&R{bPgg97GK}2~b>5XOq;*B3YKTR<@lO~Jy0a#qvGlCDT(SM`nA~G9Ma}@<A
zTR$Q`hsS<b{+Ra~ov(W1Y(@9OyhyZ0tGo6&brb^&w5m2&X$!Xzw}9RvS8H@z(Z0Q>
zpLZ7pPe1GVNt*iU3@^wqS1kGShD1^euO#5{j~!E&g^g;fHz>SymGg>nQrqQ&C))th
z{g>@zxX>_DW0I7bLU*URY+GBPb-L7LmaT7NBuZTt2CZ})t<$PPUsFg#N@e29qK0LF
zc9~_5GqMGuD1Pl~cEFX!bm+#PH{l*63k!3^1=K|w9iqxAvpCf$uM%rbV!SbG?Vd8i
z&wocMpG2^Iyy#It|22Do^chK&cRdv8S%LIpA3A|4g#0!rY0<CXAq{@t3FY^m=JPkJ
ze$hr7b7_au<%T;d^$-efkIO}X*t0WeG>q8SeZtM|E|vp<z{}~n-%rPM0Cdp^<@-c=
zP-8=ojn$D8-`yII5{n-eJS#?#U3N&Cf?+`4PX~ZPt_oTK)J<YLBF~_CjDCjm6@G$4
zuXf`ha7zG)@tno)D=|}BzNdA}>k-+_oPa+BV;Q~&tK`#jItZPFsOqFZ+2|t6Pt%w7
zeuwjOpGf4xneD}9YC%G%r{yN%fSxMMb%`Pjr7jmtT(C`{uf-M$n#(yYVA7HXPg*=g
z{0wJ^MoOJ`I(;W-x?u=`?Z~<Vv%F&TcSH_&;=Uvct>Z`-SdxN(g0I49A$6+46UxwN
zekR$mYWgS3Jn2O<U|Flp83G73#2vb!gC4Tzj$k|Be8L725I1(gXxKD@Akqa9sUU(I
zR02ZfmrxQCbV&e5O(1KJH(Z$cMxUsOAZ3HMLxrevTHgVv8Z%OKBV~de<pls5ChvA+
z0SsS9s8+Zw@)}?lX~QSlVr)AnobZ`k7O)c)9fu;8CWA^>=2hgIy^>rpo3ZxgG`v}O
zLXgOLO6Y%~;9Pe@Pb1{x9rN6u1HPA72@wi07|$YecNOaX!#dNEyyj1zm+QPTmV%5I
zwg>F*x*&l)A@8SaG%&_2_<<?-;{fKYdWi`Vb_xO_$5u*SXYi6&EWWsW@khkT^T!pP
z6aXD{O<gL41d9(S+z6h#7=Yh?{N491B+l60JcGN6ajr}Mkf$_NFi{f}-8fq7v0|g5
z6O-^e&hG0Kcz06XCpdm|5>jKPK0)ezI;1vEJ*$2xty|4<WPHcCIylQHOaH>XoirR>
z01b}L1|#2p-T+>M#a@WzRP^d~`yRo1o~<47P*@Mnf|ZZAt%9}K%oc3HON=2;cK;vJ
zz5|}hwtxKhK8;gRrxGbT+Dbyp$idStk;;s4GLp=)j$IukyOON3ipo4$*`7nTQ1;%G
zy+_3Vd)?=dp5OES*8lnVysw=5+}Cw|ukW5N{O(b4BUhaN^;XvF&W+C{5Gw^BKH7Q0
ztK3J;=(_lotG6IpeU{&`Yc>M(kS)%UDp09qIXVaB8qDXb9+u`&QKP+jkm#M$TCSGs
zKKmTnrQZZCZH{gdVLrC@JLIWH5mlu+2{9=stInpqP_{tZyH`4?>9nZ)P_J~ncC^q_
zm1v<W9q@Z+rIz2_C<nZ%(N6E{Mp@%~pX)%P=YEz!@NYl%$OAX33ld8;W8b&n1@$_1
zQA!z9E_VRXKgP=mxk)QCh9T`pIb#?a6fO`iAUB=d&A_NJ^d+BiTo+~Aj`Brp+978Z
zu#HWpu6bk80?S2A$GHk@qsu#+uAJK@KQl!GPfN|<8N^JV;M|D##22_ql2_a=L^pCX
zW{?#1+)D_||E(Xvqr!1fXa<51Vn$6w^m%bUWzHv1f#{<Gw9!BIb~pB=bMlsQEo+4u
z%E~{(RGC_c;eGv#vO-H^G31xVBrwKC<;)+ZU+>)0&IN;c42=XK#bd+($W{bi;|zRj
zk=q!Q0YTRS^LjR(HIEWQJMH^>dY}~e^cD(qgSunL1eSR|<(0?`&vTqX4g3l}1-T-%
zhMoXcX;dg?%X1t8g^);?IOL(IYE}t;r=oeg;zM~37W)7ux+Y<Sqcb3Zur;?z5#&n5
z!1V*U@V(_G$ny>^)Bu-7Sp;&~5UujAWPjQ?Di@B*y1xw91i3WRFJ+(ORI2GR8i51*
zNw9)KPJTdn@X%O6R1m5YyCB4=1lF<a3kL-GN>d~ssM(!a6cHr1Fi<TpI%MvA<e|$W
z(KUMycGIeUN*_b)tnCrTYcUx_UkWo#YEsHI3HnFy=5@L@3|7DJLQ87DSk2Tlw+ZNL
zbb$?@z|sQan~NxCnjQ-F_7^N!Uv=UsdJgG(M)I&iLOf3N)3Fk6450u7NPWCIMtXq1
zuKpfmr+E-AeVFk)yw{HnF~D<oGQ@$c>bJvI7#f24IH~C8RirW^Frw2YRxvx{q?ugW
zY0G783HptPq_k;d>a7Tmg>4dwn=z=;gfR)h(7J|6+~lEM@@8G2EB;#7ypB1l2{HP7
z%C;3k1bq*gtAyhIb_$$h>!8hH`19>+c=M5aqoe@;e<3&`J*dYRFF4Gxg`n^BeJ7?j
zDG_+<-rk6@Xl01^%Tb*tvP;lQ1+8HZn1FyaMRhVhQ$BwNu}4*1T>yK8)37yD+yGl@
z2!pS)tf;D`9-k=h=cOrdUZmxW77UR%;Fd8wCGHEApYp-93HT!HWrXzQXdb1XVbs<{
zdm4Z))phBXiI4LreA+_p=Io;H2)IDaPJ)z0!7G_ZxO%hp?j^P(<K_7j(8aKuW4s(&
z0yEU!rNCbFtLwUcP!8T%9y02}R*N=|G#M4s5U=c5)-p=}kGJ{=uMppW^I+C+2Q62K
zqGkl@DN;r0nh%u^&vxS|+)2$Pg6YuKrB5?19y__*0jtX(<x??oT?qh0;NYF(1@2c?
z?1rl+DZ(c;1m1r00st(AD{{^%C+pRXLRLvIchO$b!4yOm$VryH$XN=PQKLJ$9Cm2!
zGcig!F!1y$G4s}0{y+?23A`D&(JdVl6Qo#04BGqG8P$`i*J2(5rClcU4YjYrs0Z>l
zc4D7z`$ltBSPK`x5rbS+m?}9s2LJN@L(L;Y6^aLq{)%rxlpmGAY$gT@MbPtkxOV_p
zHO0+?2cH;!nM`eY{h4BL*_qH+vuzbTA!m#?zZ)%4bCOW}EzT7u-Axww=q1mIse_o3
z*fmY7)`etY6;@cYm(Lw%+nYx*pmxG#BeW#ed_CF%N*qf*SY|Xq%*>*JF3eG?0t<^q
z!MSM`mm!WhtB!dIvF|7;;S)b?->sAbD63uwNhtmqHzPJvaGDvpaw{>0gMcghEmsfh
zi`1)r8P1l}RRsMnCQL#_ltVJR8VWGYQBK)k2sR3MmQV3@SLxR8*kVa_T1LlP2Jts*
zj*?X7KHo>=hmkX8QBp6LR*Gytn6*+C3}gXEgWPf4s4IZohimPEW;{ZIe6T0~iku8q
z9-9pTMCjwh3liwd)iiNEzBKw3Y{EONqd}7^;7(46E07VobkNp~K_Sp!xWszd`NzC1
z<GA*;TsFUG$ctn$>s&?XTaMp)1G_yHL%R}+h>etK-8jzr)sxM&lv-P*kG{B~W4%8p
zB2X(I0R)NNU5*o+aoYe`t+XyHYV^OP9-SlTi{Nh#N0hA0^Q52uYTO!?@1l~g3mX+b
z&I^M?><f68otNBzYN^F)KJ1d!9DR3((lftFNArLsN<qu1eb!FAdmzIDGJ^{JCn262
z*W_sPCiU>v(guZdsxoqZ0yzASdqu&L_(wM;ulSfjhv;?fM1#7cr=DhCZhz!_%`$HW
zT<V!`kA~0chyV1rM!e$aVUkRDXy|KgwQ=wmdvY*&Ki#PT!_g&%|06d|(>x->iQ|-E
z%=vMJ?=)ya0KN`sFZ=f|oWIkcI7~`$tNbN_+w+=iGH<Rz{NsR82WlO#LU3x8m?d>Y
zEwog$UTqz;&JIemv@kQ<qt<?Cu%<Vhzb_y>SGIb_S+W_=W)o9S;93Jdm?Jvwlh*6J
z1dZCCn0cLxJ@3xdSMyaY^BraH?dNU7lw6Undz*%4L;96Frf88uU#i4*3v```9gaDt
zJ{0_I{M-188mHtYhjv{3u{^kU6{`>xDg0L;!0mYG>RQ#<ee|vZ#sm;Alr}{OCSXUb
z;l(F@wo4eERF3h!c0obKib}P{EGo=zMlA;hav)TUShKM=B!TGD-J%o!V{3BLa`aZe
zz@Z>A^^DH0YklryDy!J7&>_+cASHiXzt}BVV4n-lIBJvES4iTMC)TWd-S~nar9J7N
zfKwSzA~HDP!@6z5Bq-B(8GP$X@M*TlN-_3s2hSDPHk(-rlo+jY8M>IYbI|Ybc+jrr
z7+C<`#%CUk#rfMEwg-kyZU;7D#6{4@<EzX*Gl4&!c*~@d!AqaOQ+I=7aV-2t@L}I*
zpDtn|2NRM2qsjJ!Gh#LQ2kUbw35PTat37e$jXb~zV42XJW8i*w6avkFgtMaVSGzb|
zw1S{tRC}?9(D(hROCOE6d3%XTXUhpu5Giz7Auol$DK@R|@1aQ0SH%n7iS%3dbxLRb
z?PPV+(V(l73=5TZseLt`jjC1czb{O0O(*U4&&ny8+jzWk6fg$N$<VeM7Ud=j%)K`w
zQ#*;@6jPnl*8%tIz@%*(4br(c@O1^WZ19^azl;`VtZ#Wa>KkpA`|gSpC(aTgCp)=Q
zZ=SpHw(yu*9cl(Y8a}9Ho#=@3Pgn`(15}_*%#`oJCl)%p4ASBSDYG%`S(IThpZO^*
zi>#M*-J3gUFD6L;swS08f=da!2?xb$coYTTV)d7+7ZA9knoo_8iYx{0z_!Q5jpB$t
z+v!&hxKPvE9)BM<8mL@NMlLXd>mY3>XHl&#C;^UJF|LF$pGYyE7aGdZTHTvJcmFJ+
zw5w*_Lu?Q8kCCh`1unf0l-b+3q2)xvM_m2QQA0dFk?x5}48_%FC`^*lo8urEMO4R6
zanK|o0>>np(Dh3<q9sPI1uq3S97ZLSct7A=ruXO$+n3c(O>04}0`V3OZ~<vc;VeX@
z8r>uK#FsdxSXQ%Gob2^oa3=a~1o23OfoDXlW~>^C4`qVBzbYp}e<T_ZpWTPAQsJ<s
z`iQ<TglZ@DiKBgs-zCkQaK)Key8-!d*!RTj@ft1IWZ)3t{Z{xQ_IifEuY8Q_Z9KX0
z`fI95)3*Hq>hZ#Pk(Co9lj^(ekR-}wLZ*6xs73@?KQb6<VEmH?>HRK-WVB8ViCFVo
z)pPALx`5YTW&w&LTNP!sojerOKw4~jw<nZoETG<m+OgF@#=au;x^>wU$m$g4f9t99
z9lZ&I#C}5L&kss9_|H~vYirY=EbsRGu6rX^={!MyK%L{6n>`0vpw{y4-KOhI;{zZ^
z87{Dzy9B8r>ann$V3upKQ1ext&F#MQ-Sf;=hJE)9y#$2{p_}_SN%it=YkZfz8Z@%m
zfft+`*cXXw2UUon`%tqXE$M5D`zaGh!|QvF^K`Q}nOFY$0ia+Aa+aDdbds(rQkR|M
zHN)>sm9(Bw_3_7vHh{S@2uM1X(Sqn6Xr{*QasG6i@Lh4=7jw2Gn)i|3(E_bjn)}c6
zefX}v6WO57q6636_b^KB{qpgX;?2+Fq;5{5%3j5HZu1VpF8xtdpa4LonODehi*1rv
z(|a^^_*@swhAoT_Kk>t)q2XI~eq6jSU8rHSJ3>-jrnDYkm7)Eh7oRu~=JN{Idwk$<
z7%uf^t<DCb4{3%gaOTWDzqS+Aue)iNMoHYC=vwXw$z0`1uwaf{MrYsmV&cIbT705C
zKrL$^8zTvD1yHF8zPPx6?D!Czi;Y!{l$fwMIEdP;sZOAzA=b#SK}g~(PPv(^JP&Z~
zHlvj(^#;y_NU_8fThg-vOu=BJngHI?`TMCtzyF9DpzW;LZH5n<<5@G3qft`YX>-^z
z`teqR>1~CD8&;|*Fc<Ik)O~@|wEOZxX2me6hoAVGrax=*7Q-F5mT)RzY#QvwKH4UA
zRU7<GtHUqafQ7LhSmQoG6uadJen>>u&Nvq;nW)4cx%I8INsQ2!DK@2}_$sm%L4T;S
zGIe(0ZO%9owX~J$n}HFg7(D^4)7iX8!%VA(ChYMS#j*=ygmid=wd0KQkL~caqTV~V
zN^p3M(-zD190(dn!Rh<c5D^585_OmrYUE6WdoNZg(%uc{H_pA)0_+~Q2uDj|jF;qS
zylXynlr+a$PSG_`o|b9h7$ce6I~RPb5L=NPwj1^mVQ<AC{a2$-3hnzr?;#YoRr}|R
zk;3Zj(+N_ulA}Z~y%<8V<CR}Wp&MpC?{WQykZ?i!+JvtP=uG@HxytLJPc*JqEI;D%
zk$9pQyQgA~ulHV+mKT5+EXc%JQ1B*xXs~d-ty0c@_6ip#K77$6v(pG9nDSM404&Mt
zH`>O$p>hD)jio@H6zTy0UCBLARtN;i0UDJgqEEe4EFZq=*a-u;gdvxzc;HKqzZutA
zv!boOb3d3i2(E>yx2Y@FaZW)z7XSy4om!Ushn2v5#9XKSbB$u$YC`e6)#EewOune;
zGVO3dnE&qNIJk)Bpju==HV#@Tb<dvJfjFf3O|#_4iRZ!~3bjwxe<kR*H6I$;@-A#&
z5-1cWkqH%4mjN*%W_E#~z6_|af~1YTN`0t=5%Lg_Lv+MG1;YNYdoTTfTuQPf^?G_W
zSKm>5u4TNBy3<)=P5y%%o0oOm^ZIK($EN<JVJ3!*H+VjW68@3u%|Rp^U$X%>nOg%1
ze1Rw}e0nAA7M!*XR8xU2gV+p$ZHdWXg|Fu;9LEha&|YZG638wDe(8%ji=t3BcT;PB
zY62%@FNfB<a*8}dw^Fn5W^o;5d~5LVeyeo^SNojw$2bVB#Sv>Zv9W%p9Q?W2qi=4E
zhHzfa%@ER)H;zGRAjHZW0t_dhG$PD24hC-^w?FKr09bM*TThCO8s&Iz0cBh*VaGvy
z*h4=ld<w`;%*|8=3H#TMpB#eLO9pQKs*t3<2aocpy+qH#_2;<aMm=yWM9mn9EU6Y+
zuKf>S{+`=3YB$MaU?yMw&L$4QINz)7D^XE%0R=i8y&S|2a)qGZfqMc>z^7E@I}P%9
zKRIfZB;9hx#W9=!g>f)K3>vt35rs{^A$0>@xi`s{ll`VC^#<jyR$9pzX@kNR3~16u
z;0Q!sr0JJOz(oV5KeB*Xnv*7k9$m9{1YXTF@LsN^`BTom=eY7Vl^Y+~;DrP^?ny{$
zLe;z4%qRihQd|@vQ246ynL5V`Nb%<o>Ax^OzPHX_pXl=!<=RU{N3{2MylakBKHYT=
z3Kz7>f=<}-q`;lQY?S9Mv!UKqwa{NpHf8X-b~A)MH}fj3@4EYFXK$9Uck^Z#|0}q?
zQodv*)A+roc}~T4y|!Nih(1+^tL+@>V)o2EI|rFJrjJKgxy6K42t8x7-Bs)_#NT%{
zYhBtcrt6x%9JPP>sf6N~4tMpd`gl5gMqT4BcL-Fpyf*ZQNgsjXfAl_JVIeAa<_u_>
zLt6<YrmOApm!kc~SADiknZheo%bvTLsVgQY`{)BVelNL0*yG8-A0Iy{G)`?ytoox`
zDc?$n+Lmd2;dV7jbu#od@B5G^4yn%|vq&bJg_$}lkI70i2mTQ~DMzenp_*F91P~53
zMx{*Y#l2FJ8)RxdapATC+_&Z2ba(bV>dc-!p9IZ`V6eJMZXMU$5DxgifQzpR*o4FH
z^@f6njj9eSe^0hOQQJbP=%iIRKE~CP@p~^wMpVOuouqd61J?Zz7jd(1h72;p^RK|^
zCm%XfB0J&b!F3c;eMIa89CadJN^AdI<tCvxi*uU4&NL^_Bh$#j6+SO(-Yz?HC{;Dl
z#<{p2ih6toG{a_;Z(CFErMeq#jxoI_U;H`s%z+XsCr*iP?Yz)<DMivTZ>w{p$!t(h
z4|vo0vAXeh4MI1)M_rC*z8iHhuk$6A+eUw?_J3zvSDJE3$ggp#Le15bU!8sa1=n12
zuk;&oVPRdjw)AYi?kkQX-E}iRC})bM&%9Evm_8GF6<oUqJ&hqP4XE#S3dq#gf_HZe
z?tK0fXYF<It);O|X*E~zExpL$x+7Ni&!j4xbhH-eW~#zigIo_}N|j89k0GvE!$Z)o
zl0)Wx)%m?se!v?rF`Pq@Y=Gp*LV^w498f?G1uFagSf?0VWp}_zpz+IjMQ`+N#CC1g
z3?=~6421fmbKA_I+C@x$CKr-}c&vvYu?ZBYO?XJMuh&f)@l}K}Rt1qH58S8?1!K-$
z3au;<*unE}Fsh;6NH8Kovxr#Ss0bIRe296=`U(i$9nwsfsrDsE^`6PGIk;_r;rB_f
zfRtA|Co)PmgFzb*6@nB32tTQ(bVB^fkKte5`u-8f%%GTIQNa-}*~9p-yI)IY#eAY)
z(T{0)d0QpaV5rYk4MN<ee?kxlE2HHBQ=yRc*q3t{A13;Y6O&&P*>%<*R@N31Bg+bI
zq_Ys7?yR&i=;F_(G6->xKp`Ww&b4SfJG7*GVN}*zDv5o4g|7){RL^>j;wO%u#zY-K
z&T*5uA%IHmnhomLhN{?6(=@1Bg*enVlsz^l0~1UlhyAa@N?;X?&Yw_iTWxt5FE}ab
za?M&G^$TA$*Z_5ZQQmVofqxchpPg0U6&E27wLnM@5}hIKz_<G<;+>?omIA63yGKd=
z+@H&Ha;7)NRqwgbN1JLW8t8G>U467PW-EdPFrx~nqx}L|>)nvsD0guvzWh>A_a7LN
z<Wr`}0`8Kr6C|c>mntLfND}nRF0ZJ4!ve0in5`x=c2qNuF6^dx$|1c1ycY4B)yOx%
z(~ic#)yM%s_$3I?5yp+UA0L(w%`r^6lVRjGokq-eMIY>A#o`6}IbV+-ybwSrR?qOg
zlvu93;u=|?QU2xBhONV-{^}Ya_ul7m7#QiU?S?%C(^g|5csW(lg)5&LHG*Fe*+=)}
z&96J^;>`$iFnp2|CXfrV_G{!S0`I)0+i1b3f01^L<+r-0LoK#fQ+b@0g}ASrny&u|
zj0BLQVVDHogqAqygDFt!lP8h64<8mwhR3~bT<E-5FSs4k5il!?_F((I`#8;KJb^1p
z@+i|BX|&!iLY)b_CrBIE_(kzu^L*f=WjOHOg;`URo8P{f%NHrlJu*&uAe)&h5{o&u
z;U4Y)iv*on<FJ62ua*X0@NVqW2kZwOyL)s$L8a-z$&6b&@a{%2r&eKihUsBoN^+>*
zOEC=0K6J~H(5_HV+4Ab{5g|OA!x5<DHpd*(=s1^jZQvEtGw-5DT?7YcVVO;z2>P>X
zmEE52Cs*VW#V+k46dUe~t#92ufn**dsbho`+bgYQ+EwsUtY*`hy%|*a_w`Sf%L(x_
zU|cM?rFVFy7v>eW{-8MFiBLMccStT~Qo7ecVzphlfA=G3fmz>`7lPp_BX;q@;3A&Q
z{7l98-Md4iHhxaoF;aRdvT0zW*v<Noqw7x+B2PHI@cAZ~Umq{Pe$H5T#{tJ(Vn>d9
zgD(Mb2ed`P(bNNxIF&mg{x5=2uLH2i>-F|O$61G;A}T$Bb1<}3^9MK=d)vQm9=rj^
z9NQdl4S$|3^Iw-bBT6|-xXW)=KL@wK!fB$xL2^SAd?wAyJ3$#am&%8Kk=U=LqayA4
zw97Aa*Z@>Cx*3m@MDq8!Gyx%e*c7YMIzUhBlmNmGS-iP*`_7l;UW*SRkec`lyzQiE
zHD^?k3oNY!hA&8?TU-Ka9;tnEEA~QPI^k@>yatQY4KT#Qt~y@#2te=IN5FDpI%uDE
zOZCo%Ot<V*jjpnDP<!fAL)ow~EMdzhQ*|4Bm*q~sJ_0Cai|-BJ57zI+1J7|RfeJ_G
zwrQ%p5wG=jy~b=^F6L9!xwRj8ju)Whhm%d)Xmv=;=bqa!0yh)bh?Q#n-U^!q;%Zi6
zNP$_MptqM$?qoRd#Ns%nb@kesmBOqO+*8(jPc1(TrqzQxzzcwWkkOt`ah3%U6Se=2
zO?l|8f=_v!m3zvTdwZV$a$4m2aTJk*M=(POxcwxl&C+f*_}%1Jd^^X?xNZz?W!QjU
z8=3li)aBQW%e@6PH!f&XEyeA+$j;hhd!4B-IEI6ahOrIOBQBqJfkHUEB8f_RP{jAi
z=4BsL?~mj1s!ci{gHNapJ23zh3-BeDe+37l*2>5tN-(I8t})&Ig)3t?zx2M=?xE#k
zsVxh893frgH;5rR?TP`DVovY<1c}lkWuxG`)gv$(bUfCBbbS$#8~OpS(I%Oijm^Tt
z=e8fj=V~dYOBHD_UHC#-&6bTt3Kofs=c4})kCn0_xQnrRXFJ+yJ@ux3IL)~y$7)xy
z=-gPc-Jsi>M{N#6(64VOz~@SW82fy8kdu>N--J`?wu%;3J^<FlDTdCBQ4{MY%3A0w
zQ2tZ$5ZE#9om!vtI_h?Q4{&&~1oq*p2vXILR+GKnAKp4ilEEv*;5^r=rbsN8E}~qU
zy#`J#3{kOj$w4m@bAyM&x}3f><@SX%%J+beaecIuk`CWKEn)CN7&^Yrwd5oe`*j1@
z82BRknex7L8ah&)4)g5r@xFVjV}|DOi4TB6>=B5{T-nh_t6RG31n^{vMM=9ker~4d
zO1GHI$Q#URU$T*8Bp_j7jdvl?9&0RiIO3&H4#mwzLTM)RW)JNw>?00$<Y&o87!+kq
zE$5Sm@HyRZx4aC#BeGR(60SGr{8@r)CUb((<>P@;&IGAH>mtS2apB27|HB6hO)wl0
zU?_~95tpfZjAM}?+hpuv4_%qrR;=R&dK4pG1sFtPBS@t&f433{ONO>4XVsK<)n5pE
z>jvy`kp=)t()+o<BntB(NJVzBjgvUJ!!(X!VHF0~c?VWVTltH<WwJ3JOTTS(AI9Hw
zH7brL`m_uoHDEz}<WtseeBT0I<E%huT;8NjgL)g26llDgJihmi=lS{qTXDlkp8<H;
zqgQ4v%~*gdTAu0gIuDMi0U3E&&JN&{R@n7mTo1lsJx&VXt~&pdJq2p(2Iar2)KT2@
zxpku&BInor7TtRS?!{-WC3w%OP`-_1f!8s&{8LXN{9u^+LC0$;!!p{}s&{c`lvuZK
zPLL6J<8;h#{)KOwr}>PJHa;aD-0uvB-ic7KWZMoM1WD(rk{+h_iSk{|gTFCCA6)Z`
zTq*HBxS#^1s=B##heDX&U+>|g(eACaqCV}GY>BDhm#!I;*beIC;yd^FIF7bV9k<`$
z*O%wp%7v=Td<G7L;XGAm4+R=6Te?@W0ByzI+&u9z{@NQJo%XI|s$#;PA=2!M5(pN}
zt~}PId7Pg6g$@T4Emcjq#(#s25NUA%&hGBD5zuZh!o}#r@?C7BDst!%_+ey}`Hc47
z?ub8Prn)SoDW)Du3<UeY<nLuch>j4ba)*~iTC1mF%3(g{yEVOA9Xb6#i?rOO9)IiU
zzjWUeLN7CaFYpA$9H8ymExcOi<;)ACI1=r_W1Mx0wTwq?FH6of*G=bQ>-Ts}_f6|B
z+k>y;Ksl$~#lD(<-p%iNCBumHP`xs^6E24GUzUU(q&4?8g%@5Rvv3K{^+(8mEzg9|
z2~%0GErh3HJgVTGHmmOh9H8|)mbiZKqgQw}kC?E<!1ZMziMlYCyCL|25OgbFU6}M$
zy>+ilfd0PN=({ng#;qrpJv$Xznu>Z8r8mXi4O1PAt4K0R$8*e2R6@TAu6DWVt7g{C
zDTL(Q&uzR~s{FxE6}Rc6AC0Zu{chy@UTHtMAAjW36Hy_?qsDt$jZk2pB`o7F_!*g{
zNjtFYSa&Y{_BJjFpQ_a=8@VQ{TRf|*qkU}nhS%E{^lpV1gdb^nPYUz7l4`4WHplbu
z$L-37DR*b-QmBn)_1&BF)XGaPcD}A9f4AtUl&Ve%1oIKzfOB*o3KfQ(j|)rtTCrAX
zhSvK)EnV$E5RQp`iLL>W8Da6sL8q_dq}Yoo+=ezGdfTfy?+|!L5(|+og83xD%Uw*7
z4%8RYc)47CR$fuu*02YH8QW=A-V!CU2u~N61X1qUAIaTXz~>(@<rw4y{?Je(0?Ria
zjNdN0&JrDP9GfIH+;eOuz{qbA@-6ra`qQ;fJp~PjE}0Oqfb`sTtFVM(rh8NrA4V?m
zt>C}^g@8od0<95>ljBPgzUG(bJinR#=HSoj7TWTPDB1^~UF&s>HSBa%tZB}1!%Ha%
z`fO;{2^_?`dsyg`c3C-BMBvKP1y8qo`#|F5q6*eQliS-3ecDq7YI-$1j9S{(9n$tr
z!L_S<z21HjqX>p7aGL`!7?K0Bc#5362_5z|6ANb=%8~%8!PLHdiu5EYrj&KGR+cy;
z$tE!+^@Var?Jf^D?>6-}o+cRR7|9;2RTN#?j@L2o4vT%_&nds3o2-1ACq{_23$rYL
zDKwrUeC~S>|7hd?u)O!jott1Ta4D#og$%E_I4RG~Q4pf)_F$>9WI$wZzHDZjEkEe1
z1nnn2g1@ec@;)t^S2&lx&ju`cb7(%nV)BH7mt$+5C;Qq%uXa~nD&Oqyd+Duo^WJg^
zsdDcR{cj9z9xpeo(>*TOZ(8RoctpC0uW!xr{53pNEB7g_;#sxExrJj*A<voz+jpM4
zroGzbUL;#qEU#Y8q*-G^=8>l9SILdq{FD7J)4Ll=C-JyJ+X1FhaqGcp;{h>8oM$KP
z!IP`euPqYU#?NG4Ejh_PU;3_(;5ugXD;)6zC;x$v_NGW<i1bD6z*u?Iifv7Cx77CV
z0kj$N1R&3=YOXy-x+5p$iQ}Faq5?ww!sPOP+8UAd+H-`o-NTZ@B)y!M`5p-A81nQ9
z?f{LFvrRknc|DtzLdFB1YoEUN9k^dmV7m&jL55`8N8rt;tk}j9cAA+jUhz&^X0-#C
z!t>v7_dpdTv~TV}yT!`vQRz)%s#%;edrPl5YXoY3w?I0?0(vMV5Q_6w93oQ_o_MBt
zcjg~q{YKkqw>*?<<u7dp(E|7EYo~{HZ{o@*#jfL|%SJ{<!K)6oy=UvDoy0#@eu>ju
znjhqhE3yt4+v%gghJVrdAIJ7sQ6GD>U<{gpL?fhP0`J)4Aj^<0mOY!T1SU0Eb<)3D
zZ0~}g;66suDg%2DB^{n0{L$&*C-eOE>S5CN=kA*Ze?b%=r}Tl~8uyr})b=lPOgqNJ
zv>z1m&}#nN9EZC|?2@t+Q~GAHj4AZ`29(;vU>3JfSl6?SHq0mW5$9r^g~5bDRzxLT
zf~zuDC|AUhn{b3T&Y?D`Nc;cA8@S6TXI{R^FvaIyVC29qM1=Aw-m~#KTg2EiXCy1G
z2iSk!x9(3rjE^md^Ji_elMmS+;xy}Ddv{MaFg;TZh@&AK`E|XoaYc7W4_gU1vuxU1
z;sMJpAjsv_<1G%H&9TBiy#DwwC8iWV`0LQ1KpFobrd0UqdE@n{Gp`0?%4T@KgrBjG
z#86`?h9$=onh0Gp%wTm+I!xf&TZ;0J7jOcdGY1lQxLDJRjM79o+1c|W``d`*3pda9
z)5vy7zkrX#zJR^-^G1qxtpxc?L{_J2bbC6TA0u1RP9C0+-$8Kg=3S{hObTG;m3!QI
zQoB}6sUQS14i}Q1Xd%a*zQw7G&avPA9=OSkMwyn4|C$E5N{eWkPq|`E-R|^Ud!F>B
zbJSr}gj4z<`Ct#G(O{qfmQ3A*Kl@{U*{sL5tqHh`v|UJ{T+}*q6;1~_@W$XqIRib6
zT}DanH@qE)n9522I_Ra*4>C*?<DCkV+G8N2*`Vh3UzlWJSjC(0U8NyAKPmqBaCX!_
z>9DxWQaO;YcYZWPq(dGMpD8=m#**qfd9AGZmo&j%q~L{<c-o=NMC`Z}Kw)t_{To5M
zLRazwg;ihI(dzt0jbCFYU(kA#Rx`UAaw&v}DWx=y%%(0?<VHxd1YT_^aiRBGkXrg(
z;BaWK&+mBpiPBjop`@1Qb@91nY;>9<E95~ekux1u)G|@sgF|soLtlzf(+I9@M_I7N
zhl57mXJ*<DHl?Hb_T;;r;4}UR5g9S-iRGK=v}_zgW`Bh<$-kAx#C^Mzbi(FhZ7oV9
z0KN4c`x?Q^C><X1a0@iUBO>(l^*Og}p=X_v^o}~3C+~jDH?4H$Wvf^q&TT$l;NeMr
zE<w0OgnoA>ppl%G(Tu#^{c(jzr9&b*7rpt`3TnW+fd<(VaCz-A0fj~O&DT1~0``Aw
z9^R8T7KP<pY{Dy$QT{0u9HM^ms6FZMpu&L=C{X4#ZH6A_@!CZ%;J>b}E=bA04c+ov
z<$J55Fi+bEDaVR>y_9+@V?|Nw$B!S!Qh9#t?<ADcDKoc=Hq_X?*?xF4cmlT!g=+-%
zf6&Z3IgC<y-oE9Yse>t~e?D%#2gOo!TLzKhw_&9bQ|)A;@Q&0Km|fmRa~>zvK&R+&
zC?ge8YW14X_dcld;zJXpIjvl|GIt%6hpw627_FX(IVkceEazR>A;;i77uUfJ&*&FJ
zBC0|Y?e=teG>sY@rA7c+0wLPr?bfun4l{Ewlv})s`byN|W^P4wb)I<=R-h@TZDp)+
zppE0fnZErl{idUxwv0Sa>P8qFYYmM<QEL{v9@)UHgJQki(76T1*9^br`guzuu%{|Y
z<rIG9AaDEH-v0izj{P+susCEs#cMk)<N_=AGJeYxR5^)Af~3sUWeFwXnLq0@U$uYk
z)l;9TgP|GOoNO-fL(~vRi7FBs0ncoa>-AVI$ZElaO9N0mpUjpswkxNv63GSK*W~s5
z=_-U%#QXQ}qoW5d!iWd*pegD=I5#jjX9$DH581n$f&#szGfFM}_+<Vuy(+D<FtVKC
z0a-ml3jVXGyM{NkFU`e}doP{Vl?qmH+`m7C=EOkPVz)`P2@{D)6KJX+YBOcbT>3UO
zRdjaKH)zjy7Y~VbCc@;3L+9yO=Od0kYeDtiD04M%rbUEt6P=s&w?to09)W`NnGS9V
zk*1jz@X)84e*X&C@Gao1wYBy6w8m76ZUQIsDjaU`w59kbn8Du6ONA<0EyMaRU%o^?
zh8{9K0M-}2?;odRWMm9YYLLk_4xcU$=B?mh-9KaPW)-*3boO0+uFY&|L+R~`PRqH2
zYy8|yH>F9e)JuK)HZR7KS=*YLnp*F0T9qojk#~?JTPycTpz^Z9>Q_X|5$GN;pWZx^
zKI~33fnHa${mpazP^!<BW7*#fSonl&ls0^J01QCmA!tQ0V)GdKh{*6!A3a*hbn02I
zUf=%tHCqE``tHxGgt4q*v+&FM%4OT9PJ8fl_nr`_<raq;U;A-I`|(=nKN>Q42ens(
z@xUrg+ft$F^;*`dii+pf#sb?pIR~=ZPh60Eg5zHuy?T<g!8N_ZAZp8nt2+S!>!Ab(
z&5+?7w15^@Ihk|4nO&LId=^l?aq}JZqMhc4<YW;}&IaJ;`)?0;HB!!8jOrusc3v0>
z+Pz=e%+xdza3C*$_F=`YLAZ>s5+~4v;fHIOlDw^lf=@Y)a1~RYZSuhJUkJGRhRuJi
z<>hB`2IFI6k*cXE_S%IeBQ&Mfm(CWKO2ME9;0x#&d1^Cg>U3@M9lnqU;`~-aEi~gH
z=x#V_^B4l4lT5rs=60BpG=FY`x9?D5qM+;sGuq(EDeGbA$>R5fJ!R;Qll7hgyI1_x
z`Bx|=at%?rsS&~+a<YXTyQ87hVY%K68k4I&%3j!1wWjH)^!bWcV%CC6P2Ht|Lac|R
zjy;t=PctWJYPL(vW=mX$rYh~}?`?lTcPtnuRfPE0ty}v{B3cq|)h8M>&rN(hx6S+}
z43iEe`ofUjSr6_wqvi}_W2!A{_H~hFhf!(aW{2=08<y!nvEZOl-c5I6GryMdZKz+(
zEBTVM+y4pnz^L@TM}HVsR#naQmCotAe@twi{oH&#$@qqfN<G<P1g5>{*hh}RJZ}vP
z3yUu@`^iC1$PmOX79;1mZ)<I78OQj4K85u_UIQP8jE(ju(*E7NIn?*w3~9<zjV2>N
z41KmP=BKi|MLg<AhbB*Dz;0~@hM@A}b4EW@N4eM;o;r03<`Hjc6Z<(<o{YocH(41Q
z8#5NT5UdeA;19VwQz{)M;y+IM7SpYXGm1mbA8e*aB&H}5`NL`x&}FcvbbhvUS@T+E
z=1pTc*?lr@sNN`)Sc(EXr_$sj6gmmh&`lKB2gGy(MCb+<saDK~JE3sKvLx1~510zJ
ztj}}AI-oL#n3q};Gg}z5JbX>!Rp-?C<{webYX)x*x8;NqD_IXHXgwDKInl6-W9!z#
zyQ-sA8J>6o@A-$?lcYH?;xH+!IS>v#e7>nvq9)Mx)86)Q$%>LhzHf>dBVI$|7JW5W
z3RU}?(t#`8f~nEu><7(-#!AL4z-;0BVJls8d_TAncs26WgQFR*<gTK?ShL2S$_Q&K
zE6L2!;NW2B6IH%LfTHeubs!3HxHUK2V|Jp31-^%%F~etZeQ`P;ZXZ_OGDtl(xT{#|
zym!1xEw>|GmvGLfWR6|TMDxVA&&Q2-%Brh3!-z2xpohELIz}Ohwp1wwnDi$n7ERpp
zQi&v<BR8S8<AXG#uR{AUsdjSp*$)y}u>%M~pXYY1o9$X*+SuSn!^Y!>4_}=wO-xU>
zh4AZ^fy(OYkMZ$Z>ekSoem#XoZGj|AOn5p?kRC3k_4|w!jwF`}*Js#TPmTZO=jR7l
zHR|iZG`HT)@8~UPpMOXE09_<0_~T571{eM40>wERCfTEwxg%2s^@#_&-qNi*+P-~q
zg|suT!?Xhk@%C`nNY&u6HZU-N0A|2(vz~pc`BV~l$-PeV)0jOUUPbwE;|SLdHld7F
zB3;w^2tUwwoTFP?TU*?2LUmKBdH-!)-PKKodwYE8J^+EM%x8OOUoF-lvm0q)@}Dtt
zgiH+~3OBv_H1YI`Is^R{oGf=C+1Z028xOr7nHcI3=8zxR{zE<t$Mir=oXog*EO-N<
z;TN;=w5t=tURNH5+5}=c=%VUxDIkMqb54Eidd(RZ(aMcga4=F1b0~n9cW%?;42!vp
zk;WTeG~W5rC=)>7OqIb>wHG|UM4xO~mTxqhAY})G2%g`RS>8Gd3mKSDmJG5rml!WZ
zC0_y<6puCLH3RHdEY+ZgTlR>t`>$`8c*DFzD-R+MJyUSbx<||6``I`8HzVb#l>nE{
z8ScEVul}|OD5bBbW_7+mkR;=XBnJ8SE%b5&%!K=GN9-%u50q0vw|gU0Up$-A%Q>EM
ziX8UhMF4e<leW~dFT*cfHFgiqxC5UkSCnFpWn$@3*sU^dA}pIC5->_8L@E=|eJ?nT
zAWqAZ_9kk?QO$`^Tuz;y55-Kl#bq8^853P{hI0N2xt2#^6I}AemOaUGz69Q3%|PO}
z>X*O#!(=KXw2d{Uzbn0ZWqB{BJR{+Wu|K8}k~@>ht%&W*R$_YXcI8C2k!vWC4{?zN
zJS&8h9q{d9e`5qz_Yq#M-`zErc1v$eox8Z4@{kbxyCrD}>)P#{7_q$dx0gL8`uMA|
zw4Ht%0Cpa>NLdV4U`6d?=1rXeP4}=~Y8gwJ_ayHG!`h(3fYwG-TFtd~5Be7RVc5;=
z?{OF3)_!uPCsQog(RAyNktvF3Z_n{V&VxU0V6a}a{PseKwwzy(`^O<VsnPw;!z5-N
zeYZP`Byn*ptnkbBb}hN*HG1MQJADI;jP)E!WefD$mi%$y_m?<}eN%B*ViBX*Tl=fz
zCJ!jaKs#bwfvgPXB!H3-+CpgVb&&cTfm1Xr?Md33DRh!0k6)}%ze8}Heo0|l8vfdZ
z&uyojyurSI<;5#E5|S}kEUw@(+M;wwItW)3f=bnY;aUrspnq`4-wNW`P?m$%V1zL&
z1%l5Pt6N@ysSkebOH4vAqPuBpYFUx#y#y$aov(21W8f9`23SJiQm6a_+$&x1_ASO7
z0nruWa4uvkaAz#RF6R-_osabR%_4F|a^l(YPSKZvI-ZHF%}o;%6M|>1>*(lo$ydvr
zRs}CpQof_3qx5-ZZ6X<zvhW`&I=doxEuOu$$@LKW>j!UW%g;mYCuBFAswMRG^c>!C
zWwXBL@*>fokDyDFo+Y1R9iXAHUZ*85E?c&&btl)BEr)hEztcDI>b!JU@!QnbK>tl9
z@YUYlBReFe8yg!Ng4ay8fc{&$zYTSIZdnoh`gJz!gv$Cf-M$Qz@~)rJb#qm<T|Cu4
zJKhD+k_S&uZ%(_VGlBPWW-pu|$;f!Dc1kH0x^`StRQe1v$Nh#K&0Nn1@jvE*;D1U;
z`_*@t#;$JW8vA)-8vO}Nc7PyA=)R%>@dZ#|5^1jkE~1R$E?l_i1;KFS#?UkFsG&G4
z(;8@(K~~tkE+GVrrheU?lp~Jc>g5)RRcM8ytc&-T!;B$0pfnc#!>$~=b%rkstkn_H
zoN{rb0e9H%4jWUlz`RMJ_vC2@?4Ez2^#Qg=+9KbHbm@k9`v40_Jz}@!mry0z2K@)2
zD#N`B6@(|0>HAsWNgZ`V&n91Vn+&ng`O?DQ-<)UA%K|P#l@Y($i$z>wi1ed31m;KC
z{VZXvrANdl_?JNVK1v$E9iB%*0ZA%4eG5bpOxGZq{q@Djlk93EhSFFXB49ZKVp5^F
zey*i3gL^x)bay}})Z=2e!hg{o{Dn@3@Op4^qTl7)PK;vN(2IYmTz#tiSN6u@^S`!(
zZ4DDEjMG+luHX2#Mc|*mC=-QOe?KHOgc+onG(-8rLebzq3Bwhg%;Lmrjr{g9Y8~-b
z-S2<V5c_s2c*z@Kf%U7GEQyX8@tM+D8S)-VFqYml#-Qj78vd@ZmwEIHRuYP~h`B*W
zNI!^5xu;y|Vsr7yMaa=Fi5gD4W%%2k1WxS6Fo*@USb~R;2XOs`Ur<0>LOZ0xlkvdv
zzwP<&55Uodn@A=IITmhAdKEtG*?n^+*x!;pD<6Wy_0NzLj+R--SddBq+7ns%`;#f)
zr8PWpiGepfX{7A$kj-leefhol)ibQddtfx)kNl^*;6DH`A~D*m2ZyJd_{>`R-rp7S
zPo=AT(8%d<K|_ktttS}^oax&_ry<Cd%>t*!5QP2qcI#J$%O4utx9|x93t*jdJn&nn
z&VM+7-Zb}fW|>Pud_3}*ffEZKV+H-~=$N{Sa*2o~E_a*=ecd8&2sUQ$)H+UKf!lv7
zDpo^tVzzawTp8|Kg1%YXYC7TYOmCC0BgQvDTC?SsL$urPqoO{$UQ0D+2(^e8_a_$k
zgl-SC+b;bx26Xe77-&j~n*QIc0e>AJH-h?qIAG!g8ceJ}-k#C5H0TU^AdiZiK13xI
z75|@E^!uUR9wq5rlaWMUg8RLl#ndEQS6^?!jEcc?DbNW&7PR{%YGB6+`3n>Z1=&Gl
z<JwqlI=mx!@N8>P9yNxySIdWnhDJJzvNcO!tOvvr7YGTv8@V1)tFdp-_FO#49DOZr
zZ^=2j&O^%C$*AonzLR6qe;=)yYX1FeC>pTT5eXhOAtLemw{&Pcpjyt<2hNJM7ScRD
zXXot0c0PK)&Yu546cdYvC3ke}`Ld(R^-c%sDP0XCm#($Eb}@(c6$P4BfGo?n+3pzC
za;KKYfwM+p#Rb2iyHvi(1n=So$!BsO;4ZH}*9AWVltyynB_Sm!X!ix(sPy!oWx=Ot
z_E6BsHG2chG_bIDR0$e3<Wrs@XkWOhT~%3G`82L?@K9$tc!Z!?AgfjXXQAfe`9v^j
z>aLQYF~!}kL|^`*17Cl@e={?bW12i6h9uJV*>rXSL2$>RiupTg7o7QU-|nhg7Ze*f
zsU{ikh)V;BsBLdAKrTPJ<4Sd7<F?5DrY1GyH0k0J3QY*T=-t)yugz(&S|D`LUSnhj
zO+>Xb1<!i7yByDW*X~j{eCEJtOUNFAX@$+gAGWE>6^xL|VzrC1tZx|^F{BhbgXF_E
z;tgwK_jJX5rKj)<Y$F^+RTw}~a{9-Fga%7$%gQcqut<fjv1Jq>B)WqdV-x6(B@JnV
zb>Z?6(5&Ab96f{eANeO+AesHpSPQs;=%ko`Pk_N1_I$j5|9)Lv9rG|at)E&CL2pEM
zD}kBwPV^N!P9U8U@xEs@P=Rse6HI*V{r-K6k<z(y=WK0lZ+1XO){UhmbQ6s9Rg8Rv
z@CR7XXgBgZplst}t6C-u?->lyJbU&mG<@B-&KlZBGIqhRvpP=fNhhV6f&(^qyedY$
zxw$z|EC?)XNWyQuchL?)K^XF~!<Ok_xVlEIAxtKa;A{lGRG6(egu2bcq-S20#K`Dq
zzBJ?`7ZI0bs38Q?ct*ky&mHhB-UUvOT&kP|919#`!hx)GdPWA-XqgRzC9zfl^}w9V
zx6=NC3J<^m?9#m)OM+$4C_FY2v^V_e)2Ht4?%<dQvmABoq+Q*@$@%gfy%hrHH`fzc
z$BP<g09`Lg-<P;Akm$+KJ`tcmXPRLCjoHH6m#7}_-K$ud=`X&$9q&FVed;&<j@11V
zD;c-NU_sc;!3qnEJyMX}Z%N!2z)6SbzZZ0W6JJl3V{9nw0sWPm@Ov96=NYKP2!7IS
zBgS$6VS(MhOHynBz=s9Bi!K0x+cC@#I^X(b<nCZDR)5$9eI^+I`~@n8nJ^Nxx|R0E
zy37$ZF!^OV(1qv{sJ~y2;y(Yy0})je(*N#gzkz!bh+Bs+tceRFw{hW4mu~d-P>>T{
zeOPR!uoOOUosk=vcf##2$Y(%?-|htdNlLJMvgU7RL2CfoVY(W86mj|CO)&44!Yu4R
z?xFi<JcT}4L<dIZl9D_K#vl}mVPG$8G*SWIb?W5k?{Uue=l-S&ZQr<XR)l%*ZG|I@
zAR7jXMP3CjWMs=Qw!g6D911&w(iW)36Gvk(NF)>hr?JEQ+ZTYgqp*6l^b%Mu1ze~4
zklkd!PQG>;#6G}Of|)FdILu=E=CwOyWGu+ZIRx|x9pWt>jeQ`+h>w6pKIm-x^5q7P
zhR^MW%i3%iLuyIJt=w0Yx3aUZ+B}_~|9JJ4uSMmQy^<IB>nXq*_nv<LJ~h=}^b=Um
zl25bBW*-C1Ja?W;`<CjQG&DmA+u3v0R;>cF&juiNWiNH6qH`#yz$s*CGz0`{tXNHL
zt?tPXzER6B(38;D`JJdq7qemhuvQ28use4<`LOfYlfyR)hP60l&SL*&048GnB41^Y
zWM$WhQOJW|a@CdJF|Ct9_wWHNe0)3W-*^yWH-HmZUhcwuF5)iSz)~E`q3Fs7@Ba;x
zX*^>=(+uJCZ}<_08w{?t`1`_=jO7=u7;Oc9hsXY{30W|LA>J=GNix&G=$^4__~7?l
z!#{sRBk=MX;{A61hVhI0!WIWe@eKcW$BJBu_gj*q3SW%8gn#hwiTHoAFO7|UT)*ur
z^3p~>)PU|6ZT@|ybiuvA{u!k3>jDL|2}h66fx1BO59uDH|LH4_?x262fktpYwC7LD
zUV``X*A>uzr{C#+PqsOf{n;2t|MxG5D8zi6OOSb~#6hqYWc5S!c>n$s=n6%C!%T#$
ztAG2;Z@^p1JJ9cl={g@e|B!xsxe!s?A7iMKIB1$smgcBe6}ft@p`zmG(W4B+UwDC6
z)q|+86-;bW8A?FYYmu{-!RIiZ=9K}Oi>$2d5fL4znGs&(^mMhyukNFf`s0j0AA-tb
zY+^DB+MU0jpP+utlm1W8GMSN%412=f8TuO38+b$~Nl{(hes*g3%CZL+V{u~>6AUsJ
zK{_Bqp;C^!DV?M<RP#Zz-PZ4D!<ngE4kN}jF1&9~dQ(#q(iF##A<#sHkOA0X+K<E_
zv6lXt3MKr_CIu-b`nJ8En-4a<x}p}8)z@G7Asxo+pfQA+nwt9RJ3y?#{g7gzu_J&I
zp@moE3&<%LM%~?1bxZ%=Mmhz$hWQS+Kr=0fr67QLT{>yPXi2;$WME)mXx|~2xVW0%
zNKX&xmqJEOvrx>N7$5)1{7*)+=diyP@n%m5v?G5+U>;(<eiz!dLc2M4F_xcTETyKT
z(D~Xsn_s}`w16olmu+6XdZqHw@U!^%>^s<vCCn>2$)y27Ob(_#DDEq55M2Ll#hL5&
z_Va)J^(VO0z?ny*(I7CcrS-|-Q`uj6mV2DyISEHazEn0gzPi?gad`GLUB4&qm=Z5M
zNNB9C)<Lpm>(z;CD<FopVa^f~0}d!gUx%4S(CQEZ?y>U1O{>@B>45_4$rKqJO!GMV
zY~z)6ct6ey&v?>cxPs%FqL&Yip{X5i9cMnFys<GIVv5E!FvS5HC!Qia-wpD`_v!BU
zBi5}CH^=S}gbFQi8>*Xj_#M;By5Z*L29{J|@JW-M7rveba|-M!P;iU$LAljFr4IqY
z_C_5Dpa$g>Q(*|hp?U5&NE+Z^(g!`=x8CmEHx2?}T@}Uhaq!S2A2hlbKfGhOIJtA%
zsm+ZNx0HDYE1v)SNo$cVclZ>itr6bt<o{LqPili`3s_@^cfkOw%D(Izt;{yJei#|T
zjML%TvMYsCy<+$bm`m>ber2az{n+w0o7r3uL3UO6h*%OBKtoy3>%SwTUY`i93sV``
z95MQZK?_df$mYpr;<6q4(qLiTSM$MmD8qK<m+ZncHl8TZ;2T9C>{;l_Ok&??1SWOz
zU}9+K9!?-)1_dX$+e=y<VMtMVyR~cAZfTPa6c?ME#98#$b&iJoPRM=)XySg^5@K;8
zXT_l<yzdPtWc@xk$T0R`2bkcIG_7r3(*X0S$k|XH2T27-vcV49uxL=7JhVuoJJ+an
z#>#BT1is<2P9@dcK)4Hr?*P9;+*i4ubHPq)h5-zEw{IV=Vw!_?`S<(ua5!wX*m_El
zH@`=}(*i33L4b(=Ok<)!eXPuR7$#)_6E%;}9-={AP)nIzlnEm;aGA(mMqh=_%5`r~
z4m2Xy!=0)u@H$(VntqyQggksMv-&Z@ygYYSWLiZiL~V`8s*TeT5){lekzx=cm<-@>
zq#xf9+!&5{E<2!Eb(qxvTr3kN3S?(zvy4bBzz7xxW)T#A1pW^WyZ7P)*bxTpf3P?(
z{*V7J6y|@DwX!jb++;yxDg!D`N`qV&P$V%1#ZY9i-+!qS0y*0DSHV90-H<?^!)E_s
zduZpNVjNFG9R}fm3+izc?mYt|{z1EC+!S`vfQ=YNEMVKB6CAFUAu|zYJx;uJ5?=pn
zqf7N3OxI_K9qjw`P5swEqgpL;M25DofENGGaKR+-l0E%6|KIFMzKsUH70eI%&(MR*
zmAZq1A|v3O#a@rLPsfkN;{!P?)j8<W8o`<Y41l}9Hj#iOy?hxGed!xWW>#hUzyGhU
zbtyRDciNa`*)Zw@_dmabe|Q7!|LYCE=`a}j{Myd~BBMozgVa$YM>-0j9sjfK4QPTT
zAVXQWdMlIBP<VatV<tiF2a2k)+fISS1F<U1k@kEz2!wz>^SX8GK)Myp%Bl_hVTc)0
zVq#)@cL#Jff?>m>LH6!gC!d7lzn(P2vi~3?x;VJlAYr*rS{w623G9%R{ZDbOTFBno
z*9MU)%4Ua&33G77!8>aMHy_(^=(%O4o}S(j)+<6iqak~D_dJ-)E1sjr+0u+#A3%%?
zSf{plvKCKB^y}9Y(T*9)$j=kM;#!51I9vO&$pxAWPlAvyz-z$<!j#uChyRswE1><K
z5RGJH6!X8+G=IltL{N03XE0q1YcPch_{}1j!);?t`t}hbhu|Dposgl9J^6nl-oNjG
zrH*^f2yvo+!{-niCg4~4GNi#jlPSrWgPolo6{Ne1FiQR#{a3crj4eg~POeMn`JWj2
zKY{3%k@AmZ0bhaAS}-maI4ckcBkYbHekoZz!AV2=!Nt#89IS^}uKBM2|6xgs7st57
zf^kw(+BSM?`hWD9{F6|4KL+7h2KNJ4Vw}+Lf48l$Qi?lrUoO4Zzp-tqWZ@hY<kwPu
zxfDMzLRpHVKtq2&-gj_g{NhE}6~Z%1NiBjq5H+qsxAyf5P(7Ygm<sj=B!7asl>oXb
z==%kBdl*{J0-Z`;qlr#I8X32GB<LpK`(}7r7i0vF?KT)s1V)ypNg_0ZdHQt2Z+s8h
zA8eX!f<(ec1F6v37|o&12K5Bm3xlE$lT@wAl`zx@3Lz1AI)_M47AX1I)gc}WXi(zV
z1_K^uK%<StF|S>VW&yR{6P7>$@df6NAjPT*Zu`|yA7QlZ@Vad#Fr4RRZn597h|ElA
zfc4AI0g<Ec(8y6>SvsVOfji1WC^9`=Y*y;~Kx67Jh+-gm)&zYo4i~1vY+Dl#NK(FV
zv!l@Kz~3I)s5;ckW~Vd;El@s-O`)7c*O!Mh&k}K47qT1}wB-B_(#^{#^c-7gfTJNS
zJSES#j73Nr=6P`8oj!sR-arvDsN1v-hE0C_+ltRrz+O0?_Ss*kn~-+5v(VTWjN1S_
zq!8Z;`wK!`M?S8dvr*o#IjzIo%xr~;Ax{Y3lv&Y4&}ygPXb+~QL2!w=fj`$NB{4CP
zWT2j8Oq2U`IT1$@l4jXxc7HjAE=fZ88afobgdlC7c^Hy#=OD%g#wiqWI}mL86RJnj
z<`=VrB{rk6Sk5VMvD^+dkYt<7kqA;jKNf5pH*A2_{x91^&3QRAKy%m3dwJZ9r@_tV
zd(LKc*AmK!ajxph%1+#yig?S>t{?O2L)JG=J`a1rr@EDIFhQ@{Z++-{$kN+!y$M22
zhaouwE>_J}fi~~w8GevFxNe7nqT(bZqdMX;R?SgiMHuJ1Qn3*RRJT%RWiF_{#2wM*
zvOgth#UW(KTx~v7uLW<10pQWbmN9DSbx1H8`GO6T?Xd4yLSo`g{=pXe&lhfkW!RHn
z`z{V=Zeal#eAdr%ZrWa-%7mbWo{<rLeiHgxPshZ+wuENsr6_f(4HVNo1i3E<Ra1^m
z+I@ud?JGjxzkf$_k@Gn0!3<cw3@lF;wGqgq-{-7jHv}wt6GSie4TGlVTu$)?@`;01
z>0q$*8Eq026r2DXsG1;dBb+wdQQ%2`4lUxf6@PQ<mMw4*#bAYrii*ZbTuZT@92g?R
zmcGdazal2f;I%w&77Q?+Hp4)eZdH66hQzDvwCVyvdrsf6{C&V8{y1fDKI9>fD*clS
zgAX^WB|$T&uF}AH=+oZCZVu+JaY7oj30ks|v8@~&*+=kpFeJKttxCdec{0J7MnleL
zfCf62q!o~@VG=VME9c^F3FdK|CGZyX5nAhA%?~hW4b#V~iVwewiZZx;yD-C?Ocrjq
z&wdn|VYRuR$;OTY4qozQ)*4_}y}iBJIaDBU2p}0WrkG8Sbk=$c9X<L|)PnD;P8Mib
z=n~-RL)DOLE}i)SM&fbbW=2AV{!xX)rWQTmbeel=%F3vE)zRlA_M-Aq#@ATXAfFX1
zJ+Q@Q0HwH&o1%TemJ%V9ySVKhS5W3KV1fM<2Q9=JU&(|1A#&xm73Az@CH#a8VjsZT
zwH$&DMb=QCu=lhs$_H*85ho7?t-f;mww_*5)6lxP>X+xyjJW^%PiO`Nb^Gc+4kp88
z!8M?%UTz#h1GRD8kbw(11C~vvz<1J=ZVkL79#};S0UTTF8%`MqC(DP9LtmbQ*J5`o
zy_;_SG@`IG_Og6J%-5>St7&cqmIr!DU_d5>^9tJIja##z=^Fo}m<`OptN&;ect|t1
z#viEmn4lnAd`swgrYaL@X=$o3Ojq+NB^RkU<)4(V%#c)>mmC&M?2KZ!0)_#jERT!m
z(27PNp*t_tk+dON>BG)RF<zh-rQ(|^EKT4rLe0>MV~cezX8)({Jc6%#rC&xTg^Ct}
znAI4}JqDC01&$oIxgJJb+I<nVgpNvChiH0J;Pz5hR`&DvM|clUIVwh}d@OO$)94)-
z2!KNHq`jlHU#V$ck9p<f`y=AVV>TISsx&zx=MqW<k8_4AQn&6;_@KFtOWX#Ab7<u*
zxBvF-8~d>q@aDB%h56SfWMpI}62O1enmMxKmAGwPyk5u}H3=vX_xw3xU2t&iKY-!q
zcy;z3)Y!;fNFVB{TB&(l_BIs>fFexl@i?gZX-bQTI56AoGTvM5Z=RzL*qCD0n`@WR
z_Dp5#i;o+hJ`4}H^-7%cYBSI;QaLmR?VWV0N?}5X<D*A+U701)?trw^QMZF(w8mQ?
ziQYENf3P9>$dRy&jHWwJHkm*$y>RZ&GjLZ^naLlI@*3^g$s{j7!(;ICKu&$li;r!(
zkWt_}9;aU$+pw!T@)}EYeBUW{ptNJx-=SGgWT8Pu*-v)yfy6}p+GjsyK1)o|@Gvcq
zZSH_6wcu%c;(y*Jcx>ZAFF+A%7!0VduU{-C%)VtVq7hsu>(vN^!CO#blRVhEPz>Z$
zpr{Y-l2ffNTGTp9b#Y4;@gFwJ69q0`trxZ6uko{;hO(){-oVh0m)igI5itpymMd1-
z3K@SyDCZ&H2>1u^vC2cV9R$)%Uoq&D)!GS3f32C{fll^MzS3_>6X7KGw4A?c3N=pc
zkbv`=@1M{z;%lM!sY0Ny?4tVm`s^YLaL&$LdhoB-7Vzcx%1e``^>vk+Bu}%2hfm1Y
z{ZMrN_~D?Dyc{ZKN&2g+<kbw+hdfe*DIC_V@|a>UvwEciMdqdO4GsC$g+1S%2pW%x
zHv|L?f9cL8A#T=FVSHcPK2hQu%;jj6hucgZ<Y8aGjxbZO{{7cP#<f=Vo#<89ZOlzf
zli%?ZM)1M_VymPSnEH^+IUjk*e2Ygoy^yACEIexgx=X%;6%5TZ&)X%%S4FFe_B`Kr
z!YF-QL3q28WAfk-<c&<@U+lG6Ak@!<_&KjCgp{Ebn2XbeFZsj#&MS(F76lKL%iQ~5
zw8JBc2baVEcc_zc(>w>Q`^-TjX*qyPEu|`RqhbI;KS^Cki}5t6s;>5)A(%tOgRCrz
zyE=Gk$xI0e3CHI7QwAXKg45>5n6AWr>HLur``^}&9BGEr9=pO(O!KpW@u#skBf#7|
zP5AD^T_8{_3)Jr$4}sS5^Nn09A)ZeCakwb^J)tMjy-4S;9EBv136aatYb+MXr@@1@
zVfU<}JLF&;pKJlf`p80DSoj-v=$gr(&ZyH`!Qp;lw|*Yj&tLHn-40JECUBV)s%uon
zCnoq{#M<@qpkdmB3O8z*gcys7=kGNpr5+dV@C}A6z+b)>?#xU53r35PnVIuIF2&+I
zZmOuL1bkHepmC;40mOcRSZr|cj@hnA5yhSV`L<bAxNL?_Ok^rAcS(d2k9mJ!G@@w+
zSj|YuE78sCm<%=QydE#`r(@P-KNOYG@pI;%ahL-Cn4Fy4Y&uH1;TUMp1jSh7>;RjQ
zf?^*~t3aLfHQ%H08_=;!AFimV*u+=y=+UE-^OZZxq3v5>vLk3RFnZN>n=wR5hNKSk
zR>uhSOP0NOM2FmM_0TB$P%=mTh(d*%tE+>$?Aki;*fz~V<TpPf5c1KW`={2x8&IUc
zWjs>{WpWM?N7*6eN#sb(a)Sk`LfG>)1(R1zFs-@Z%rWy12F)2_X^CNBT>w3suJ#8K
zT7l*!PzMsG>Qw&^Oh2Bq!|P6z;IU(k#@TB{3=<6M1x=zI3>Q}TMYm_FflyVN66&g|
z5<v$6fVgkX1XhVi-FT1H0$7&F&}(b)c&OiAtN^iuktCm;c#R59dPUk7jJ){2Oa3KV
z!<*%1FvQx<U>4;mKNv88K!*BM(xC02b-Nq*#!H7EL#EklE@!k@P8q7kwAGjeOz@sp
zg~TT`b=j9=Zd77^l$>p4Wp$U>CNbA8@go?iCRZO2YZ@DWie`?AwxUWoc*VvJ8q`bL
zcFtGT*7}yd8XF&n5+k$P$f8ynY3SX#={_BaA_{u&=zYR$WAo3(oddyu$Q;d6x$hcn
zM2NzP64Ra%(1a46k$izeLyaqA?9BS|#Od$9+DGGt0qAJN`<cv@kckGbdM=4afr#em
z5{tW>Y|t<ozPw+X0RE>e*DXiHM}13O|7bx&yjnoDMK9JJC)wKm+_zc{im1RpkB|P5
zEisEiH8Az~G1MTZ@<hOJ>|R7D1OBsl_Os_Eq)`V{XB>t5i5tzxrkw+7h%R;U(7*Dn
zXJBBiQ+=*e|0zsQ6v~?dR&WzPSE>H9G9VxzmOKQbgBqbobV8oK{L@68(tskqRNmzk
z(6H7Hf?6)_voNFH1|0H^yJC{soPiy>Z<Tex+0RefXMXr_r;qW?E7fIXPtvOg;f31>
z1j4Z{2>ARcm-t!UcLN4JZ?RT<Xhnq_uCxd;b{Qc9*$}m?`+w}cV~{0X*EU$TtuEWP
zZQHiGY@1!SjV{}^ZQHKu;#5EP`^8Me+&`XQ6EP8Q#)%XA<cV0hbLGy3y>KmeCEIjz
zU9}i3q<$MwrtmLI&0Ws3?Uc7(5C2Sv;rYAt)A!0T2O5**8&KhIoNgKS8@4urstTgI
z^C;cw+t!=@ht)Ax-+KOn$M0|Dm>8ZniuTs__Sd)jcfNa=qT{kkRo%h3m38x-KNwF;
zQfudZuC+?6{cZZ&8+>chr(m{CC#4wAoqwLL;dk-p+gwT)is9LJZJWn;*GIlTqjDiB
z@3(kgw<|)E|4=E)rO|5gGR&;{1}ys)md9Nj@RHw;P|@LS>$%HS0iO5mqCTaA=6_CN
zI|HOie+PX=kDLCvweYLjVO;FP-oK&R|IDhW{o9DE_Ia%MZY-&dq&@Y|ZSC2%?1n`8
z_||lFW&g8&e*YB{Zy?CuXkSoYING?{|18_YuD`F5RaGnL_kVve!~WKNhNgflgBkt9
zbH$~{Wxbpaz;93Ao4MsCFBEsIiyou@pUV{g?YnFJ{4u%xF5|YS_dcj+tcMzP@DD%4
zzP+KVW4^CH7jhBb9wlqX(?Xfj>ax!+v>wFJKa(fv{EiL&ZKHZYCQ|>Mi!1Wv$M;&m
z2W30>5B(F{Z}490^N&ZpuSb#4kJX$Hji;I&)qhwqdfEN;*8QE5d50kXUHY!_^i9L_
zTcG=%mH1HNU~W*h{0|qA&abb(qPJh7FFJj`Yu?8heEvF8fr0B_&prQKr{%9eV9Jq;
z2>FijiLI_dL0$R(!#0g?o8NT&Z?WRHiQInhd&~Ve>+hm?3;Blk<^Ja>C8>$Rqy6~L
z_;29l?^?P2&-b^R_4oDsXS#d;;fmrZ?)x5k_e;uG^S8B@_<{Is28;5HeMhb)|1*_u
zF@m%H4X!cv8TP#=&4C;D9}@pVXjVl_>uvD%{N#lEjp$ot9Jvi#`#iV%kpGn$|9|`x
zhQ0cVRl%!3nb;aTIysma*!&&Y8CpUyGP2R*)8qdga&tq`DZAU7;M1wg8CsecIYH4W
zIU73t=Zu7nftd*uotlNQlQ}*EBRxG7ow$jGnYq*d{1CKoa+Ei55VEtex3e{|b;4(d
zq7$;SwsTOjH!%99C~V?lVPqoaVBii#Cv9SD_LmAX13UhI4k$WN3u`A62YfnFYXc_}
zVG|=eV-qM|Ua0?>s@vLSF9RR|06<^vXPFr^UGDere~<s=!2fdKe>w2K9Qgl+18x{@
z&vH8dsy~?iANu2eWIH1ZJ<ES9+XpqZ9M78&eV(i3w@br8_kApi;Y%>NWX$55c+VP6
zC5nZ{8zxl~Q6lcj<}G|Z(b=GK$wzM`*3lCP{Rgj_+qY@?_eb|Z<hnXu9^d!QLiCv!
zM#L$}?n5CHkb6n`w55-p6W~D?#q&<3#kDtdzkDpsZ?B%dGq~>0*Q-D8Y|GskzRp~?
zrMq*!p3Xw*mEq7KpdrWE8KNM$PJlv_s%>E;M5p&|Fjd5{Psuh_GwNtE$}84hhdmzI
zvlRRzugdP&^b*?#=f_9ZFffG-Fu4h)a3_9+0N#TcchO}`GLGaA3UQINsk1$yQ)HLl
z{Tb{&8S~KCe0zKISZ{pXOhjp~(yea#JVir$b~SNBLLpLMc&^zJC4g5ljJT#Rp2ws>
zYd1Lt=d0A>+ha3PqBC2<k#ASa#tr~s(fwoFSw5cB{CWOZe(rCcaxUfjbzAo2v1`~_
zwpLfsv@5d7vuI#@O1IoTSl3mZKQL?wtKLkfBq;8mNr)zaoP;PIFyCSr#(0UEE|$P%
z(|U3y(i+A0qsn!>T*3g~PXY6y3F58&v0vd`d>2H8@50g-c_7gz#J7pM{k2sy1af?i
z#~Y@uNv9Fg^)WU{41JfFvjv2R;`xbsX>w@Vb&3rNA6~5_v(uuaZRPhToJ4Oh#V0dr
z2Y~s<_?%voOsjT?R(Drt;A5{u)Bpl&-HR#D+rBmLB7Ut-umLN7#nPAfZ?&b@9hBcl
zz%8QjZXX#AsC!dY*n-#lN{8GNI|i#R-ZyXNCqGL%rOfRi*Sn!Oc93@N@t3q<3j)}1
z?nv>cJq|1yxGN8BBC_!f@pJ{}A<d8!qyUyOV;ckG#`mi3mTfzq*rMh5TJG;YRcR`1
zKNWsd>3X+lC|AJUHPZyPOtx*tx(pQ;C>ixm!azefq4&C2BCXA0Q{RG_TnX3@M-KOu
z!TZlG=zOSas&Kd7?9oQGR8zB51~_^gxQmCREklw9X#Tu%TwNj|N9w5gF>$emG&05@
zy`Ll8oE&-*coc+B0X%=9+L(0|9}&qzXlvp2iyF4uMXUN5SGZQru%%jUTBRM94z1nv
zSjh!{ozu9-X98m&=!TQ$LKXTtUR(z9N;;5(vGDe7)3km|6;e(A=HrYP$r}m>t`uRW
z4!XDX(o2QFZ@Q~exRd-y966ru&oPlRdAT^GbUrXo&X`THk)#(UHBTUTj27J(#PTBS
zpo6fKDLr-J9cac+8s~t5pa47o@PP}Dx)q>WwGuQXtO#Y+X0FcoqcDW*$rc!^-ye&l
z;WTJuI0EFN5vkIWH^3y;g7k(<axP$^nQ1FgUZI$9{nDne<fY~~Nloa9Gs-yq;_TC4
zKY?QX`w>;R;Wz`iXHonIaAJ}Dkbff}I>6ldR{`PAg_j>h5sC`9Pw*IVAj}b>c(%&r
zC`k~)^hLz}Ia--h#uBa>D-ME*;}rrsAULqNLqO`Fb-crvPs1#X3d7jE2qh;*f$Um<
zYEueeb{e-`NoC08)6ar=Ey5d6KNAA%IxV<(Qt#+HVLw7Ih#`StD1Zy3pNE*MWP=e3
zK`et}q>X8Bx^l`DCw>cqLmfvavW8xTXu!u~v;=F1z+(7g@dDS7C=aLT_iTE6?l9n<
zll#RUj}c{Z6p&o}vKY3=&lL4jU29M$uK#Gy!{@etke3LQP+1P6pLQQ@axnDz_5fFR
z&2SKZg0C$PHWgL`<0?7f8Qu3LqedV(klGS52`P2tJw0RAk7<pvSb&FZgsnQVD$2h~
zk&F2TyFPmepBqE=6vKx;K9CQ^@YgSnFpb;*3dV2&Zpk3}?5pO0NdlKdg~Gg0Z%~&R
zT?$)Se=!abj0BtxTM#%x>carX#$+BWYi3m#q}(NHk*j({SAOupQch%Yl!&4Lec#|(
zoDC@lt^1y+d_|=!J^}E2Wh~4ttALnszj13xpwm<Rac<kP=V;AsbQ0{VOs`w<NF<FX
z2kK&S444hx9Sv;xnq$`tpNvFJ^BQa_rYMRjq%J=hpv(Q3C<dqjT9PL^aEMq&R<}g|
zXUx4+(?oB6G@!5N)+hW{da^*mUnxkXfXcN3G^2@g=GCt3gr&{1)$Nz5Ii>^V!hR=a
zF!8YDmmTfabgol6uQpd$g5Vh_;#j5Lidj^E5^s+V0~HT$kQj<}M$a7?*kMXD*UwZw
zF4>U79>!YO9+>;ga2-XLeFe<f)g>F*ocq5g!-Vxc@7mh*LF2V|#^Y8<6!FkPjGRVe
z=Zk_;ch+<$2JT8`Oi)OyLFR*K!QJ>%{KKnZD9d1eu}FlE$oNl<q~c&W*{>bQ&5(y~
zh8f&~*qF^L+~xb+sL3_k&tIz-xv7l9Qb32Z|N6R2t@BI@GlcuFj<IVK2%OV%sc1XD
z?z0G(V7p2wK?$4$Mv97|bXj>pa`Y)M1uiiYEm=rB%l{!K1@@88Ew)vbkr?!&eroYC
zXetKLqwW+<@IWc{VeT{{geVw%o^}|{!cR1>W_VxUv&I$36>!sksw9)5?bUa352xOd
zIz~lLrJF~v-A(RcxcAC{d7_<da4?knnXJWrHAlD1Afw-eJKW#3nj<_fqK7GtZY5J8
z-g4ZAJLdr2Sbw!1oH0u94>t8`mPWjQROui6QK@eIXi6u}VNl?^JU+*UvK+d#39EwW
zZi+QHEjEGa*_9~B_N2x)-3WE-yzi~STRC3Wx;NZ5JE+(&j3x>@b~Q<^y_zos`9gvr
zQ+cFPNZmKX&b6iItBm2#{e?GLpA;yvt5}=WhxWwByVfuK)F0VXcLkPQJU&8EaBNN6
zaq;fLkoLya2Re&MM&JM-Prjt}2Yp$nK#TCzm-`1R73*}{>{M1^xfH`YKtlMvn^&{E
zrbwAkkS6>VBVA4D-2BiFqCfiCG|2~NegtOh#OnJfJ{_=n2^1r=K^7vVoA56NEtL)X
za=id$dJ1CfMDyl;V$$<g_Zkne?Af~7CXf@mX(f$Yw~mIU4Xkjy+~cpa?me+X0%z;G
zJRK~ZiMv~3O))*5S)K(E2Fa$g1~w=JQhzO1z~bz2$>(cqtBog&GeNCdGa4JbT?=s{
zALegsJlOeav?SJX&tK4f6jW`j8h?EmjyS;Zk&MWB!Q5`P<=-d{nr`}nHsqY6Nrj!S
zAkllOGYZmqTgh&K$15)(mP^F$aRm1qtyVHt>b;I<ukVn-Z)3Z^Y(c1faah$;qks1Q
zIM+(i4SJI_&+y<2@Js{$3dLuLCv++v9!bO~9e%}05sV%@8X<6a2M1NpT|Fw1_do21
zJ0P3u;L(?EgREH7`(0R@ez;0}{5lf03+45WM}{KF{I8}V+rOEHOw9kSY1pVK<+#y;
z*mbUU)1;j5e@m(mu9W1w8slhRr+fJ8&bFu&Q`~TZpiwBML6Y*%<tbPXIQdi*4HtMM
zF+{E&UR(E3Dk&SJ%vQFyulv(pG)g#usAHD4E}0;iq@`eyBrV(Murh@A=lx-hUWdVg
zba5>=S-RB9z@*1x3E$`A-DBuIsv7el2ZtVo%>1(5R)^=~)#IC*?=v0k*VE_bTXV}M
zCM=Y^A!C&?WA~M9yiheSx9j5hc)i{fMpzM~R1kh*FG&J18@Gntw7s@D=>2_=giwng
ztc`>BQ@>*)j<qq+P^&+m<crAcNGJltlyQ7uWa9}@FY&DVk++xHMa1vRi>hdDh&Th&
zI%&+lo&bbY-?!GBJb}7+jyP&L)hJ}Z^p;01E#)|NjJFT7YYw(>EBmHX=ZbXjb}maP
z?ytkMs`qe8sIQejl1k}a5b5$^SzifE)e_7BOWsb<FB{2`X6yMR3Y6#0>)V^&X*Ang
ziF_~7Rn}_nMtUjB-OyZ=uu`Ce2_ZiU5be@4)0+$x5vz=0g3;+Z00c)Agw&$qr9&J0
zsz0XdH>Hh8+H@3}mnUtl*mN_dxlyB_<MT%Ke5f%-*wQ%NX~C|!hNn@7(RKGzKI=B$
z&h0*msQT{hUXQT=8HJJK?%u-;+m3!6nd6KplzqL9x2;s$;VIm>_%e_m$$M}g3?Y=e
zS>Lfizc#I|oRtCvAZMqMw~a<H<?E}04D{y2t+E583budIaSRbXX4t(ur|twG60G1L
zJI@DWw>Q~nUErF5zBLfrd#}8mBagsH=_!SpNrxhic{1iA=P}RbejcI~b^@bAkj&Q&
zRsaE9O=)Cn^0nEXd1|FJUA?c@J?-=Lr9Beq?2Cn1AL2oOvF1+3Zlbr>wUqZQ$y|=2
znQ;Tsh2fAxm2>_|hQW^;93w1$RWWRn>`TMISd&Y9sxQrUBLK&P2S1*g4%<i+Ai+Qm
z<<f{d1{w7S7sa}bB#{V2Oarh*t!5`y$3e^Ncfc>J-Pzw_iaLnkxy98&fct(w#qCLv
z7_{d$4KrJr5opiTF2h?e0hsNB%~Kex!9lSlV6q@fe#xXRQE*WrjBpyPamTrhDGvqp
z3pK#4bmWqDcEApGIf(Rm=2|X^sIm6qiFDqwUAC3>UDWb}D^EZ<UK`g0U?EC{V7U`A
z%P$2R-%>8@r4o<&Lrn;3W+oOo#82~h?Ec<Fi``nHI+x;O5S8r#9-Z+MZZ<aBCBDNt
zh0d&KF#m8?(kB7yDfwl&A-<9bym8UgxY!A+R9)E}6Eg_FunHa)%W+uRo!7X-tvYq<
zU*Z#XfjIL5iCnG4_EyeTYMCK0y%XiN2W=X*B&OXv*T+X#SDwam#j|9KL({&~HWZRn
z<+Lkr^Cr_i)P0tvx^!;YeY$OUG}Nqf-P*LWrNV!pmL}WHWoBXTUVnY{^uygSmT;{n
zSg>}!m2C`M;ErM3kz!6})usK;avKko*Es6%g#KbLD5@E^b=)w_<2QSzcc7VD8yXd6
zMAj?LLtXcl%3<3@x!6}5%$j>>4~Sgn#tN$NL4r_+WN$p+sdW_XbR5_iJ|l<`5F8#T
zplw%^ieGbe<@!+<{8%HlMHBEV@W#gttlt(qcj6`u2q}brrkBzZTLOIu8K9zF<QvkI
zO0nWyVayB0opge99dAh04O7<BG8Obh%cCEzqqV4T@}N}35h<}L@**#&w5$2p59TTN
zY#{FBI(D4j+Ga+Fs*y?9uX4N#!&H1i0Cr<xW2qT=;zPCJL>%(90(XRJ`j~MwZ^mbQ
z(`raKK4EORB8pjraG~dHW`42-hSxvH;4*;;itE*RgzEghlfWZL@Muc3F$n)i*O}d&
z6r@Dbg4zrx8#@^lmH`S94qp?nfpbT%?uM0=3}MQYW@9cFwA`TYRMW_-7*|XGNssQ2
z>Eg|Eg<M9mh6k-KyIp}={;>jUD<$+aP+N6MZ?aFbwx=6n<f)>aq_P@3_}1>@pL5{G
z#EqUIvzqj5(5@>-ZpT=_(;0#eZ=&)i_%iEpaEqizBFmqe9|QWZ%%+Lp*6DER39`S6
z50(OmTxa<>BcCfh&Hu(!o{b2wgz^4Yy3Xw6-XJYYJM$usx@QBmJpwHwTD9JZ`moDw
z7~=>40O_r1NlO9(9L#L`@fdiR<nO&s+5kF1@3bW}_P{g-9?)l9M|^>amO7}|kDE!{
z6@^VuT!0B``i3g9S~%_Y{L1#4{7N$>Ru`RK(*3Q;$9Suo0Sg|SJtKJtJoSF*Jv07W
zOtZ=$<)EV%0#Lq{Am#U7IPL0pwI1q6Zpc{Bo3u79XMuBydG>V5nG*Rvw{6G|&<xDD
z0`N>2-^gR2>l>KeQ;Uy%q58LjKXIOIjg8%&)=%5L;nQ<9=&HJ@Y!}B{ES}ppdy?dP
z_9<z<MC`+9uJF{L)4~M!5|7=dSH{O|E!V-WmbKrk4&cyQi6*G~8%sjemrJOB6|oJv
zXjL{HR&RD)zx`ZUKSWNq5eVxa5`j@Bk3`^2b}a4lhv3;rKG(v_YI3voOhqLy+^GW?
zB?Pz}T0-+P<D;75=`)(cOF=j=QNJ^+LjVMRI1=K}qwb$^QC#LDVs)I%?=)?M_pcgW
zBRe6a$ah<x{st>56rN!f_en>R>2h0$RAnp|qv|cZP!`MTzbuTRYnq9#S#hPEErWE*
zvl3qZW4Y@PCoJ3JYIBSyI1w759Ap4Ks8M55TJECLM8E}koKV*l>N4}n0f+Fo7Q>16
zf(z^g(Tyt1=>i_#oX$M^0xi%`s$TsOS~sb#o?(>Vdl@9eV4@vI(1-`5{2A=$5m}`L
zhK!OK<qb@K3puXM@H4ENk{$L4+Eo*bh)Damq?cSKY6$hX=X&+x<h7cm>tHYqyuVpL
z%s{x_NP?Kmi5p^`*H%&0$~?l`2oJqNWEqf|!UyVTCRP5?hH*+?;Kw&`sTAD8qcvq3
z64)fNTc+je^;il!H^LPR^pi2E4l|g;%}d}e$PuP83$lG~UY7tsnx;ASP01-WGdmJM
z|1Nh(4$J#A(=7akwkfR%);rR*&(7xylML$(Tzx_=p}Rd(vn?Kobke)qQ~!+~kf_Cn
zEJp(hDftk(Hdvb9UOl6sJ%*O`G$VmwD69=f20d?ZwAhR4!^esro+K$It~%Q9vt-N}
ze30@6`C9z1JYq*oGz}Oj?|7cIG{$#Y_(Mm5r+I)35Kf4}F3uilz;uyheig80UlFgj
zXcV791O+d6kC-wX`L>hRn@e&$UZ8%Aqi+CJ*5GP*A19-0so^~O1f3TV9*uxyIo1bi
zH<oJ{;wN}+w;vRDLG{3<rau6h;%=|yK2ST}_^$al6aUR_WM}w)+l>qyOicf+-FT^P
zWrsbC`cb7<i<bhxk^VfD>kNo)ABAGRHu8=_My|RixfY;JdB9k%#oXO#hTFcO14}V3
zi-k_EpSo}})#-Y*pUEXu-yUUZNB7y%(efUlkfCA7vb=i7ip;5`h)k){q-}ETXe%yY
z3SDI3^LSLvG=9IVR{iyE`se8E{91P!DvhrDbDXUnFWI;4T{0pMw+uoB%d`WF0E)~>
zpy1GY7bHk^H2{4;{l`G^^P+Vx>LlBt?U1yO8rvafW#yT7dqp+8O{Y!Y*c0o@p)x!S
z)5*NFb4WFE7p7n=Q0%kH!n1)Vgx*>4?<*G*Z0}XvwL*gOFf@^p5*cj9&O8si#+5uY
zXfqSCyDFZ_@?ssQCyY>3LKq|7!e2=p2_M;%=?+pzx(`qpK}r$y-;kBOTJ$;x3%+NW
zB(O|@)h_y}Bny%bQK#HPamvQooDYKh?JS`2)nRdWDcqfVhVh<L1M+bM@sGz}o|55r
zpnt4F?R%pq^*}9DduS>eqw2^FQ5fSSgJ8Es7!tJvlzR<N=?f4A4c$@RvZWT=JyEp9
z7Yv&8gB{7l=ae;5gG6GYMMY{&x%Nr7ZDxY9astsxZjkd#g3hJ@#}X`Ng(X`!Q95PK
zg@q)}Q6yEIpwgTI!ZrjEwNfAMggqieYGTF+=7uhvvK{7-s){c!EWvMY01iKxhEzBc
zvIHcN0hfau=PmMuNn#?_a?Jr`sAV?>W3tHANe|fFYQ_^Vg(X4Gsk>>|G8i|d12B5)
zK?llS$+<b8nk2|jDN=8tf;s0kfk=$%{gvA~V|cv6tM)kLNTVnk>Ug$GjbsRqPoIo~
zteB#~r3BkBz>JcrTuMG+=*aV}V`qs`7;AmPO{tMfnZ_B*c2UKOY;c&SDo2{()ZBfJ
z`82w0%4=%uDnh@}R970G7pD;Kxhof7eQ4Bv4%@8Mlxu^sI6I;*jT^faaBWNic<sqW
zsD+yC2qpkY;M!<jjt!G?p8fG|hx|-5s9p6ay_Je}_QX@g>o9j#y8iK)^Am*>GgJbS
z+Sz?G6S6ziUL{IAiCU9w?BNe3G8Q1B$h6NTYiUQ!y^yaZCnpPRN%5b&vh^F9Og>my
za?0}=+}%X$gHhy`9g&~LSORhY!HYcd(gr?rKSsJ2_9z~+4}7sWZb8t26ocGNGkXO5
zUEvBu$!t{#6!VCmzE?5WpVxl!*s8jOF#qxg?XW~N4@T0P)6|*JTb1v=JIt@@XwV_2
z?4#JZ4h29zOF@3RfkiqK1FGvgA^UHI9_>xu?F~nE{Qw069%n%CpN{L#m>hh=GjHmP
zg%brqT`fT0bQ6KYTV8FCwCiu+V{}yp1$RGS&X%rX6$cG#)fZY21aaH;dQrn~<7D1q
zTNg7H2F`?ohiBrS?&4f-k0P9`VVJ9xFdg#X^&(}`l&uD!eJa3=ePnrw9f$)L1J;NW
zl=bRokKM2BcvM5nWa?RPac;5b?+bsEND)N|7IhJote)oT;{cthxy^=1<jf`KhU)wo
z8Vf{I2meg+!ybjI;<b45N|E98r?MnU3}%Kqsi>eos(*O_PYXK~$;R(0NZdAkktQZn
zdG-j;vzv^7H~7P<4P~3giZ6|(=^b8RVU*w>6|0#imTJdzXO>m6%q;!>h{TFc^LJ*<
zEb}+BBC|WP7Q7`To-+K{W1=)XQ>+f!EonH0gQu(*rV*(0aM{>{f4Vln`{1JeoUY?q
z*iB)sF79iB`1(@XAm@H!ubHU5oqu+$V_S4z{t8*!_fnQuD@OT&(vg}x{%+m5LrToh
z3yYe0(y>Jf27SR$z_WW&ca!*1WAoGf_xR(hP{)=;*0$QA9A1F5*3$V0H#J`y;^d>D
ze6A3E<}Y|Kw5P?%CXLB{OQl7hQZBYvb#1$w@tMwrlX`pST@VJ90%_ynFN;HS6Qk@5
z&sW`6I^PK6jUU*0AGN?5E#qA@Aks#{wYAJCR}vI((w9}%Pv@VTMYe4cXF6J}n?Vc7
z-Yf>gouRuv_-+AM{o6~2GL7)n@;8B0Niuw{uB&A^Ra-uLRSFsVTOxqM69Eze^5%}s
zv`#H+j0Zu$Lar(<kz1d_bo<a5?M4Qh+f9iO8KP8bw;!<uQ1APJVb{JJhWz!ns%v6S
zd>-A7`i9`^DB`UrQipmBW7D3$XbQ0#CR%V^RZ@jxr-QHWw|yI?PUBighn|)P9-kN*
z=0@wI4)sLZQPQBkFwrhzRI54?oX#HcOrd2InnJfb@rj`_4Rhebm%VfvO-j~znvr(P
z{YBBms0f%HFt9jT9o#WDNP;q>XMmzKrRfj0b24+nkuxsKHFH>E%aYJtl&wG0bMaA?
zo$aI{WZ7q@ca89}F}{{K2W;W)(GQlcedD*$t@M>(W2n1IL44p&q*p^iwvI5IMVR-8
z7wG_lNQRxG_<HqeKi*%ljp&tHHk_-ozJ_<`&;q}jh31mGvFMV{L*KSjQwQjzrG2<N
zH?LS1Dc$blyhmGZ<nnT6Qf&#H*s4;Zfs}@0nO`hjq-9^QfZM-*9&;d~-s%w)P7f@O
zlI?u(0o2?h{;PUt`4{!h&dl-Os`p8Cjo5Wo1fOlaJh4`FM|bhWpa#%*{ER}-09I$*
z2x18HjHaNGIo0@Q@!8L(DSPtj1GlV6KxR6$>ekSHyNN*Op1=~aSW_~y<2WeBWQ(%E
zRZ4lzFrrB}6!3bJ8%|})td6V?2Nzm5R(8_4L<>wnWY@FX`q1vl?2b%nlL?`;x>4pO
zZZ8L)&-dGVGCJHZ3G8|`y6kSZhzr>kHNSf5xSu!|z8TK;%S+>QMIMY?f6j&d)N#Z3
zDTaa}2r(HDi+T+m1I{If?RjAtYnVp^hQKNkw!4)_^J0TX0#G_iRIZH?8L~(Zx_n2-
zq{;R1nO?$Cf8zZHn6G!F$Ok0Dl@2zBTEZ!i>1;S0>BS3AmJTrCwe{WBvZqf980@hm
zuC6turgh_4%nWU%P`fU_1m0LbEe#vGmf7gd=>Wp%kmD=E2Nk8FbIlaU5d!kXcX=c1
zOwYRA`z)GXvt~Vljcr{%bYfh<VjU1*E$$eZ3{zOA*;a9`XCQG{M+^|C1)D@A1$hRG
zRwYlIjg!yLih~$hne~$eH4Dd`fBN&lV$rUKz9YP*mz8SK!O3k(fx)uwHxO!)vb^pL
zhpuC6V8w+~XrT7C$Dr8S9T?Dm7+KFiCX3<<i7rT=C2W+w!HWFLlpsd&tfo}_wb+(R
z(cD0D{dJSJxj|=1jsN)R__0ZI&R&I0*OfCRSgM`$si`Q-<1n_*0c?ESL+CqIQPQ2s
zJ{eBASRF?~v5^6L-G7D1R)RQC7%QRIo8v5**;GAOj2Sx6N1ijYfPnJJz`vKQFP|~l
z*m6SJi;!BOFJCyuB9lx&Tep%x5MCc?^oFu|k>PYl9}V)4`;in0|9fh&xkOt}i`d36
zG8%d6Wz)k}ZXhQ!v$u6q<a<bRvdB8ipw~L8^P(7QLi01etiY@?J!H}xWG5wO2XHNn
zT5*AZSwvk?-cN~;z$bG6;`F#Ej_t#MEKWfSh;`hg))xDtCDS;qY7vFbU2RnnS0+Tk
zkP=;)V_b~)S(}umvtr$kIB0SAVJ7TbX0Gdjz=37`jIhqnDrqDHj60CZ+#m?EKWG{v
z8r2Y?>rS!bgrNROwUzB&i&WsB28=(DX&inv1N<BWuPc|soS|qwsw=fVU)BL!x!ms=
z#L&S9I~q>6D1CH7f0EcxniQXwhItIq%<h|~Doy62vNlQ=(WUmd=8&uopDqR5F0U$v
z4$HB*aF~uA1fM6nL)^|i@{3?sEBsRdC283FTiR?~U`V)A9MyK6=Cru4>7=Px4pk%x
zqw5!=iNLGyE|H_rdfy9;nT*13&w%l^4&tv0dj&gMNl%#=6Ky!=0W<tZiLNG=m1Pw1
z;{4Ri{CtQO|3w6)nPEF;4qYkNurUCc@;0J~XBBUiM{Q~M`MmX6>+E0{JY&?Znv^EM
zbt!T1n;^-NKc2lQ00b)eZyVP}d&XO+=*^NUT%y(riM)G`cv|g(uSuYhv>!-3YHk~Z
z-BX?pJ)I_@0A4vJbHFchR<7(;4?V_NH6YohW_KK?Bto7x^kF65S$?q)biF}Xbm`Th
ztx^ssz8NQT1sY;C{$z!8;?b>G^a<IRSp`<41dbzW>ZfR+7{}062OcYf4`XIH*|9v(
zN8r@i<?MzQXwkc>dJlOu8hV-(`x~$}Z~RFB?3!v<xu|E=R$6il{T|iZY*G>JExw>P
zZ>BOyes_qG#@%IoCX)$v^D3JqW<Rr*TlD&vLEQ?J@Y>e398iphF_9%+NpzHg%t<@-
zmK#>PGS;WKTlhRbC5Com^)|F5VavL#P$6mX9Oqu%kDXjVUPCp%F5W5(I~j7$6D=wM
zAd2>@=F!7>7^;1}r2;|-yCRn^zy46|*r&u!VK#O)(^Si?&VSf@XQ--sr~JM_2$!d#
zZUGzlP@ux$Vqe<j53+$90s0_D_G+=WYa9f}W&4l{jZ_oqPQ~^*_<5O6VOc6~i>e2p
zow0m#f&(}NwT?XFS(pL8?4M#|t{mU@F>53_pgdgITVD<!^|p1+iWn0)Mh`k%SX}ud
z2$240TLps#))YS`vt(l(B7vjf(Y7>G%N{aU-O5571OF$6KA~U$;LGRTNIx<-;`)>_
z^olD2=ubU(Ht;3^DO4d5g7uyE_@0P_tf2dR#!1-daSzx@nc7^;&V!d{0U&wg_}n6N
zZpmKu)C6yeHE<@OglIr&==t*kZ}<jUO);X$DtInWv3PSdAwbM{`L7x?q0UPHcdhgd
zv?-$6<rP`?4JXG?v^YLoie$l6%y{j^^|^Be<w&Ggxx1iSUYaeT$Y0Bie87k&7hw4k
zC7LJ87X5hyFF@p5IL^92?j`#?jiIQvwi!@~l##9&GjLLyj{{%x;?78IHpaIKc>`mv
zJ;3|Jw~6vv7LyL(EA4G|r?1z!U-&a>&MT9t?yIgN&XGL`eF?(5Ev!#^J!W5kEV<*k
z|E7vL{-4R8|0gk%gW<pSw7*qx!v1&iPL5tLo6;ymWFYzlyUPB8Tc@nM(exRG)iN51
z01_1nJ879lwbGNF92?dz@hVYI5>&p4+(m#kY&|^lUt;TFVsh18n%X>nH0^&_6f>oX
zM9LJ_k;^a+B`|AN7Npctt52EUGw<JSUOqF?>}RxRT-%~IyH0<)GmR&>-Cp2LoAJ?f
zdEYtbhd9J1-ZBHT$VMq*q3ist=0J*)%5@_qOsbO(BS!^A4v#_eRQF}dpvoR<A8Rn-
zLKf**Q>myEwMDwlXe_f3a8<tyqvT4za2;Ywo(v?|Ixpv~+Rkl=wy-b-?z?YUm3wkI
z+_L*nGg%(6Vr5ocV#nljpgndgoz(!~&8`Cyb&XLmp?=S^5sZtb=VdebNGmYu;83?Z
zvwF#M>tL?9o59o+Spv9~Ukb$ss<hXK&MN?#nL^2@>F`LSOg=KS?w@-&#Ds^fOy*#h
zigne|TTrqujttNK*j;Q_p`cJFAZ?TEc?n=m5+@btGnZ4dc|HUU*3Qz7H}?~ftezM|
z=BHAE18mw{v<uT^p^;fMv0ECbppKqG7vEshv0E#*V0glO6&{GBx9~ZY=<7P`l#fb}
z^UF&vzuj9t+FEfwZeC$hNJ0UkCqIE8rk?-VY6sT@&iS0~*G&6%howO=;N4AR8pf_I
zjwtqCRCbXpj+72Bx7(OGuo9A-3&-u1JvdEWtIq{Oixrc{K+jF3o`;+RXJW?5EGt<p
zM~uHPrDjTM2tC3%XqDbVr!^uJ^rjY?wB1tGwwQNro<xWqPCTMi8w`w4!Bo-PReo<M
z5FP`&;OC%`LAM3OWnw{4Ks>2WT(5T>C8EYfLf~~{hKiKovm_%IRDs&_+$Ap=WG%aw
zgqf|B#X)vSvZ{U=f+hD2)F%~Nq;c~SOmH#?@JwvLIDQJCCy>6P;%JVTD%>9AI50m3
zX?NzJhj=WO?Jyb~pnzZo4931^4Eb43Eqx(`030~f$3-~1wFJwacTdI2j0A$>E{ilM
z!;Bu^OK>{o*8XJV(6V-o461Mv{y+e=z-AxcW@oMPy75{CXd?Xi+xk32p>xJio=)Dm
zw=lWzwKY<_fnmYFS=*^~#3XJmy^NXUMUT0rd9595%DKhVse|MO{`r25dnSohK<1WE
zmyOv##|g03(t5d>1ng6Mqjg=;sO`a~ksG9bR-OR@MdiSL6IxFr7xzgX0Tcw4lzn=C
zU2p8e+a?p*u{2x7EUd7S-|`n~!_i9tennh6-6eK@%&<!l00xRaa=5TEfyR+mTRHr*
z9co(QAyNpp|19`B@XD!>(#ghQ{7f*_vpqFrrI}9zCqyPUS_xp{D+EauL~rPuY7xoN
z9D<QvRkYrb$WdTauU^6ikJ6zEsRp4}tu~N$`OZ)9qMvFu9w~Gq_@T$jXF7NanXPKS
zBrN=;EaVu%4yNRok@MQwo{M>Y8v@$$&@&xQfKNn1&HWbsDjFM&w;&(vVb^8h$BrmC
zuNg=(+EH>0S(ux_*$vGU7!-g+e#c3$HAgUC>r4@Mt->-^e6(*%zw6~tp1LDsMw#M9
zT2kru$l@iQS2%U`;4)P5imeoFMu3mBFzc96KnyYY-d8qM3^K4(54xQWJbEiMNK<1p
zk+{;znE4H3kg;{V5}=<O+NSZ=*Y=RGR4Kij^!r+Gq==iA0DhvsHnbKVWC(3dtG~3-
zj59_4w%DjW8aJcO^EIyB7F~@`?}vhXRW2k+9-3Ii2Wp*4a^D>xlE{#7w#e^rRy97a
z%DB%(AO(0Gk`3e=I+X^Cf!2FFN^>SKZ|b&it7L!YD-X=Vof{cu8($^ZWb|FoM?3ht
zcABsw2|$%X480N>C%=8Nm2vkW_mj15H{aX$tbqOQj^+jmRmMPDg;1g8Nvp7WDg5YV
z-_u722*6kHSd<difmTP+yRt>!EDm8=x=45c6F9nRZ8>^hzq$DaB^J7`Ph<)~)atc6
zXZgTs6LYK?dAr9jA#)O;h5#y*Q_s5~5T+s809N!&0of?)kIMob)^P9>Y3tD61W&T&
zlb#{%zx?pCAN%)p{Cu)X@L)bmfODe!t9PLMmp_lmP=!Ba0|!G|hxV@OuRO4RR-JbR
z%?<36q3TVAoJ#Y3a-+(1e>su9e>O7+=vmWdK}R%CJdXT9>{wEVsu0Z&SR~SK_xU&j
zp$&M^bG-yVOvKuMrf{7)Nb-+P#r0)JCN$7M<IU9Y{kgp_2IV>)>1I=7QMb!bpV#1a
z;TY*xB~Hna{Pr@Ol_x?tokmHjwSDU>b^`PIWZZGQx15zH`o0GmJ?GYJ=Q=h6pLF~)
z)oTOs2Cl3rsp#B##Tb%SE|vp*d;xWu;1?8+57*X${Hb*~O^K&i_)B_2hu_t8y*$t@
z;T-{jxhq9n>oD)~Ysk5}QVAP-qod14-;Xx?zM0P7un$=)o7uO46T6dfKhCR!K4DIf
zw(5u9y(WrYjwbu_#X61?*63c};b*yuZ(xA?$vTN1&@?{3u5g&bXx!@Gcd>`%1kD%&
zB0qu6yut4NRh9pXSIx-E&iLPY)q|RP3F|GezT0wnVwcROFF6Yht!din8%4{eYYW~d
z8=FoIO-rg`$xqboY!b&;_2|Up8i_U>5_qq5$@oAZfH&ZN+UeFv>0Yk)mv={huqooB
zjMLAJ-E$za$jcy;C^f1Q@RSHnFQ>=(czpZKS+RyHNz$bd_QX10zl-2fs<*dkTk<KI
z4_xw?W%Rn;@1GXem#=j{kol;;Zm%xeuQOY)u~2e}nY7D{U)#5dQ+PYSG^>`!GxTgR
zq@|FPAq2>c!Vlu<EfN;5hGv6R>9<HRD;JDMyTq^TKA##T;iQa#PPha4B=MDm-SOUg
z)RWa~YS5<oc0ho;^yAH~X2#wKs|n8mFX6>sqByd<cco)UchXm=&w12Y<OX?>cvVJX
zn?GLA4RoPQ`}vZ0I}i^$)#7CvKaX7UC}h?h>c_vF)tvrZ^Lb*93PKt(CQxIL0jVA=
zx>Kv_)PH&G_v{>fcG0wzg`u|xs@<%jhuvAN|0u8WDjnMsu0S6)@x9-N=PRHX5|zga
zf<%{QUSZD9r8`_nVQw0_KPFE^U_74#EubCwrG|f9t@lG7nZ#wXt<<yxJ=2~^c*l2R
z#kdz!fu0iSFe;Gh71sC6w0pcFr!+s%lN1nid*y36o3ZKm3y1(YDg;sWiM<$E0=2d#
z{}8U&Q;+-L%}dQ`u<Je1)kKNw0u^ENO08H6V#LdLyw0hSM$L~b`^<ZBZ=_B{bJIz@
zbWCRz)vC}17OCCFAN~aDy@ht&R&VRaMuSg3ag@Bzg0gV{`KZEt$63>)L4BtVM6rCf
z=gA>tabSm5a=th*=Lca-d;aB(Mbh%QLze4qKMy|8n8lb!U_;Z-(eTj+o9G}Jg+P(=
z`ZG;Yb`b7zEng6@(E<^|!Bti9WVmFzINgxzQVMS6$+33h*?>c6<F;v66I9_8e}>^4
znfc1w)0P@0O-l&=VtJ`<V_OR-TGmr?w4N{wxEilf^H;MVP6)5^VRxPplXL-$2=y18
z9er+pe(-bp;dn$#=|r@RCV8X0C=3pmvMEW0o5DAOZZ4`pSll}~Pg&2AL+wW*2OPx9
z%(z4Lk@1#jwU-9a7E@N5Dasi~Nw7DXNRA4sdWrpdn}K~-tQX4R9A3_ib@J4{R1cn#
z1o{ADrK~mgte{{NnMaZ=?Lz<XI>b4MqHwSQ;Zux$?PzBfN)lVWTc=i@KkNm3OzGk)
zma{O&)|^XW8nPHH3`#oq%C)#X06YPtI@NPkBZxV)Z7Xf=FUGxU3%g4Q7F~i_O~BVj
zyf;gotCu5Zibz+M{A6}I+{qk!R5zA*=%0DCW47l7pt=F(mugsDY`1Kva_4-oeV%b5
zFMg%f^LxO4Ij%J_+gsCsVKV_WQG;X0$29Dfi~LSl9J{B$Uwzz|papVKPOc`n7_<o6
z_7OG5=-O)!TiwiDUP4Br)HVbmw-!-fytf?10DD(lUTI3HLc_iFWnlMEj=MmcL)_~f
zgOFk^faP9-AXYc9{lH*9&g!74e!e|n^77e?m#4ZCAx;rd6KU$PUO}kzz-!)y0kg1N
zl%C=s$%_&zs~7>Q^`O%GL11v1_6GN7LMS;RQ-*}=4|s!eq$!e7PvsAROGhKSu6Oj4
z;W7O33(fCE9M@0aW|h&yg7NYX&ouTCX;{~@y<|M8*}&DD5#qnrnYxlC%6na8xKHx3
zFyOPotml4Qbsz}3Re^qwLP{x_y4C~BwaD9#Inae<NEV=i3jb`Q-Q=y-Dh+pm8!<eL
z5)Uk!G1*7Gkh&ltOG2AQ$s<gx{)u@~zK!V6B8X|An(eU!dwJWZa`Z^8Eq--gVSZ&*
z6dlkdT9<kZIVV(!&=q|Hv?a(H$D`}#(+_qRHXwTB_h6L@>)b!}*p}x!!qA)hLyi;p
zoxR(-m?=Lo{I~_=e3-D0-z;TN+H8p@<puDj&YIEA*+lo4$c9upkfa(2p*g~y)xfTL
z(xP68G~jXyuXgh`fzAA~<VYUW+diWbzG6l?3da&ZMKQci@LiD{3OB+-XQAQ-Zu3z|
zA@dnm`Bini6>4Mgk5ITz6PVJkvxh>BuvqfE8U!&!-7@Gyh6ByA!9ugJAlNZDfN|@n
z8f}3)lYFt2J#}hwJ&x4RA{tyBvT(W@U)+1|Q%{9w-Yo!{_%SSBq3Ar>I3ROC3XI8j
z#ZBhWBxrNUMm<jygv62ptkB4)RO*Im-)p#?ep+Rgt6}Li=dYqlwMB#IU0|ZIDiM!2
z`ASLc41e1O8lX))Dx7oTgsw4x&RFqGYmb*RS?l+Z3mYCwTK3Sp)ezRgvcL}KS^?}(
z*UTBW8s`P)dPgxHahS!J)5Y51`S4K(va=n-^BnR+KTm)35-`1e!^xVJH-6pBaYUHC
z<p&O+B*%5dum=<Dvn!i>6n^?{D_n1y;)$fFNt6W@&B43s1HfXy6ypg+FcL5${cztk
z*atEO0j?b4V<-_@$c(BlNnCV*CT*SN>!>7(P&Dq`tsSG2^gy;e|HrK9OaTtpMlE6s
z*Ihp$+}oo)z|eTdRF<a*kUGQL35?ccnN0wc<wSv7c1ay{N>GzL)tnI<x=Qg8I*r*R
zhg2aO^2TU0xMOKI7=0c8cEFTmDs(wkXs!jCYL1QQv&e#2OeYt~5Nl;GHvsX{b|hfM
z2QBTJPk`$810qa_!O$r@d$ca2!06{xwNQHY00X)*yFAcq5vb#!_A{7aW&=heNInZf
z9nNz4TSX9ek=WCR2Ao3R5vJtq-}h<jU-}-1J?|u8@Gch$`(74+uq?n)(W!isT*%3V
zrQsT{LU|a>e)Q3*oIue##>rFY+DMklR<vahpjigH)^ZZ&$UPWXf$@4&0h^%dFB8Gv
zc_{fGhX|P%Vf)f7KzO9Ev)r=EtK?ccT%?iK;%L911>K))*vZY|+jT}7R!?#bX7Yc4
zOs>iqU1L)`_M^^YEc1<~mQ=n8R`ZdCc|)VP27yRCYp)q06bA$|(Fq$B%*zA*i8d9f
zL7h?il!wAiZ>N3c8X&h0a$XCNC@0(-zV_z(Ep%Q<rt9P&*8wrG6^R}z%R)?hdhcKH
zehjlx&p$zlu4$Ak(fBJ%8QL3Q=HcyVG%2E4w0IW>EcVjCKDf>92pcSQ`H@2}pYfPB
zdNuV}1I3e7howDKC?!^|iFGGuG@#7}v%LE=e%D%<ziXD&Z~sA5QXWj<4iRs|y)>6Q
zSke5efriq0X)}}_;95>UXc->iu#`~o;{zV%0iEIIykhY!*4A;7W2>zK8qLaMDhfIe
zq;Tma8zp^RXEiAnuy5Fv@9JYav<Z3ykFgiw{HhhsdKOTGn9P*7K_pYo;r$lf6~sA2
z*JOk0ZelZEFkp<`@*Ulk$a_Pvr<1Rhy2Hi9k0wra-~m)i{03-49lgr$i6#)1EvD(3
zZF)npFV$$S;Ul=<haP)hWHZ3KC@gA+SlR|UHmf&BgTPs7jXw{-XM(o@V^a+m5bUGx
zkqi`H`-*g!XI}V72vhW^0KMW^jgsK2tK6>Ct$~X2qTlaNUOU7uGU*rZ&W=BHh*JG#
z|7y!}{7Zz0mHxlpmbz5mbetDO@O}P{5P3$>J?Fh)K`}KTV@Q(Q?3DoZk2K*})f6to
zLt2!4thYZCXpseNfMB@okhEpmeK%QaZ+=_vpEU_8AS=i*El3y!ITt6QDGf8N2n!Dz
z2HH&(t2cRYkVcI+XFLRootE>>Fz57f`aJn_dhzgniWf`Dh&!F#_2PWz@iacVpW661
zfMY<bk6jTmI6#eM(Cx@<gC<=jED7CgowjglEXjdeJw^j#$g?FIhE<|A$~rpFxoPT4
z;;f1W5RS;*wW6bwl!V?&;VMf^b=IKXHtLbpHd<k0s^YPU#xR%<nkE@STmpOGK3#^=
z7sL#GHdmJM_Tz4|cxsDeeF6u_m<XtXrE&<d)@2Zf;$_MKfp*L$+d2v#PtWXeE&7g`
zV&g=ot@O<RsXER+fH@a>;fN=9oj$B53=#{9WJ$f|G{4~{Qsiux_>tA~S{aGdlV4sS
z4(Vs<^&SKSB{Nbbkui6K(alNUI^oY8_qNfwBy6*!x3*NbU<P2SBHZ`)A<Pf`8+E1u
z3g=fHeD!NP)<b7i?KWTZLq-u=*0|)v`_i{fT~WNZ)A!LIMY3iEd}5xZ6CtQXFbre)
zBD{?E()Q}rgg^#mJtphSn$bSpRFb~e^n@@+6fHnMkAecy9CJ{~!wJ?J5hl!n)g_7d
z`*~yn-e7J4uXrHj0aG;D*beRNw)Z{7lnC_|`%?(PFe<-6vFJTJaK2u;Ra~Y8A@_{o
zVFd{`xAw^4-@7oh4?fBFo?uO<XvEh<KI|D$F=BHzyMqM`gAMZq199qvwBE+es<S@5
zTq9xaOR*e!GWt8ORea-gjwI>>M{mQ@^!v9EdZQfqH(dD5`V){#Pzjh%LMeL1q3Fgl
zFZf(z0^q{Dz01HD?4u#HN4bI?e{`_RomX2QzA$NVgle~AZe%M6o20V&5wu`qyKP|%
zXPs=59_77u{WN(*J!0S9v=X73_+>YioD)C5RfMWqDl}Vw<%W7Ma$RPCtj&L0UxX|g
zSt#i(0AIt}C=LlSLA?-v?E&Rsr3QP5GcP}6AK`BUp&-t`1H!nZ)USouB9`hRo5s$&
z0|!y*Cp+@eSp<61O`-C<Vc%7O?O4dRg>z`Tk3Q8IpmW<qUj2L5jf&3ghX!IDOiW8a
z0aXOCwjK1bN0tVh(TK?0FQpa6`-geP*|F&Aj1QjZpzA7e)g$hTkoDAU+YPp#IX<!%
zK2&PnE}<{LV~r*IPfuRvCpFCZAp>)&PT?4X%C9Gyu(wL{A}8#L;;(52^?+>n&g6Bn
zl>Lw4ME0igy=9x`X@?dHy|W+jg;>NBl%k6B@*iFph=5R#6D34Od*ptJwu=0Uo*0L@
zI^;e2n%iUa%1S%09Ypg%ovTb#$)~7JJbs3b`@eUq^p^uO2@UH?72?t!R|!|yF=5bt
z=RWiUDzyknv8gTQj@lfVr5_D`T0(O5V^zj21cIaMOYLv50@?XWnuDyyQ|$EO8@>V~
z$ydQA3d-7`9dvS{<j>T-knl051T;JIdn~<K`YbiWiYZ>I;rv7*`UT_n1LQccp|{j8
z^OuL+<L(L>@(p83UI3GM*B@M*3Bh*jH<MpP;($bE;t12@cu$%@Ay5v74UIPfc1p{2
zr-O0zmMv1LuG5x%zm;NA!HCYi;Z<QoAX3DI&|r;^`auEHi6JKy@C~hgDE-!FwMZPT
z5o(tD)}K*%u%Wqn?Y>@Qa0UBWEv;D?>QVrNctex)FQLwU?V>nOFjzl<-_41pa&7W=
z;3*X<U-a>|Ug$%XHi=J~!cO~8F}ukphLE6E+9s$WB;ck=KW42YIh_vjy6TBeQPV3@
zN92dv<X$f<`QJNe*(@HPhWf=T`7Ye_E?)7-Qh{{KZB0<H90zL78po_V>zVn-fLX;X
zXi|-JyY7oE&4U|EI`63k2;>k>TgSnRIw<}~Ec#THc14p#s1pMHsd*D5fIzB!U!y|-
zaU-99e}f)W^Ul-X)+6r{zt*-($PYlt%m6W%!HKq)k@s@*@0JOoiVC{oZuFYi>l(ME
z_I25g^Fw$~d%N|2qxRzUt(QhOvBkZS>BoIK)muv4-gA_zYt{4(Nj2$bcncqGy${B^
z#k*Jh68IOT#PqFx{_jQ^|0h<$%KTq9vFKFSaQvILuw1=Uu%Ss05CDEw%=CL$cGH;A
zrkvL)UOEJgAQCyw$xU5?>(cHT2wD=v`6X-~Q>(?5?;@P!v5M!62k(RVN0v6kc7kvS
z_d~JcG<+w2P*6lqM)6igE#1p(`zak6deqn!E7~ye0Z>?~jv3$A%lX6m<7^PP(}a-&
zGo&CVZ+A-zcRRPQ^Aj1Z^p~6SpR-m{wtO@I62?7&9oY4=we{wQ*V>Y)l7d(wFzd2m
z?rpL}0U{E}7PfIQk2QGU6<p6KXr#H3ibOWNk0<qC-k40Jofw7auUWq^Yr=?7w)x1B
z`@HabNZvts6<tmJPDVk^^2NeEkcYzI9Yf@mNJMz(DzuBl5`N9y$|#bwS}Ln>8R7vM
zZji;{Ii0<~ZMpgJ+@vV%?6s)W7gypW)4v@JatS*Zr5`9)SKYq3$<pm4o2?wsG;JH+
zBW2jtNF=jCOPTeNhYO+)BuKE(DK&DF^W`ApN{D%F1u-G@(1s@ZN1c2P7TeA}gPK!o
zbQJ*x`?^^E1SbGkkOY(x*_eXaLj$+Z$;6MO?$$6Z|Fk9Y0vFh#&8Q3|u#5e<`+UJE
z$DHp!z6EQIha@;6z`<$xpu^%54i6WALzL+H6CXb>h}1MFP(<2shh4{*MYQrPY|yCD
zCI6zjaC8i4DG0~a^A5>%SA6nuho;CYxh%g|;zpc%zH*Frw~#gxE1(EIFtV~d!$^BF
z{j;hei(;Hgg{?d__ZWeqoh#@DT+kSBPfmq_$^QLke99n;OGa>(l2j6K1D4p~#*ix;
zzWN*tp&<eotv9ow;~1h9JW>Ks@#>Ct9O1O)<QO5(5TGtv?pCsfC4V^Vk|2d=F~#N6
zo~p_u!GzEMMcFrW39@iOmTlX%ZQHhO+qThV+qP}nw%t`VeP%P8xA-Q1L!KKMkpT+G
zBH-Q8G<SpC_XNmiyD>e~`}Y_k`$nCcLo$+w5iY=2UkmVbhM*WM_bl!y#kPq?S*%p7
zvUSr`KbQ=zqGck;k#`$KI@Rw3=D5?y$mUY}bD=DnD<p5*n$sJlx8Mh>0;sOInd)xY
zN*Y1!K*wAb)z=Bq)S%MGtVTAi`6tkGLkP6#c;Um80##;==J{jl=DiYEEZ&WW`ERpC
zmi49W>B_XiI91BHvoL|oqw`7|wDqbcl?`I`yoMH;C7KR8==lLO_U0!-yHjDb(}*s+
zegWj6#dVzEtvxd^A6uJLYfQWBoLJS>bI7GOGHG@bA`)S569;Qfo}j`gE=A2J0W7PV
z8$)<@Q;=zF$F`C1LLF2I=j3}?9m7>SgA#Lg2Cc2wQD*mhhBYRG?pBy{^H-!x<Ejyx
z+0?A3tuUNG=LZcWEQn$6V6D*l(iQoVgF9v`7l@%;IK$>u!!|@Mb>&kT`Qb1XL-gq!
z7jtJ{-pgA13aw?9-G~fv!pIO;zLc1V*4RJM>q5(qr3>$cHO-1Hng<wjnm`TSo;^k(
z@U}rSr;pf^w_<0Hxc9lT!C6QHMRcgoGVn+UVFo(`qkPM3kg%I_O<ek^t)lOZQ@B;t
zt;oS&65rh#l&UbZH$XdMI1Ym@gL{KG+Par;U@&-1aBj!(f|o!p9J<Y;;>PWxIPS)a
zp);4tK=1@B3-vj+vZL0*{3$v>IqSvI9)Fs-gZzG@66HyFB(s&P=+a!^cH8*$lO#6Z
zBK%~1lyn(5jkG-XZPyo^)oXki&O7MJpH-vr<iZK%IO3f5S8=2UA(SeBt9Fx>zSI}=
zL{9eit&l|k1?hcWu2PS5%?62y4h?RSxhg;~z6v(pG?uL0yPb$Az>N+T+3${nwM>}^
zgv8r^8m65<Ks<f%PYHmD3IU&d<*V`BDLD2ExcB*}w@jmS>QGugU+lBhHbRTV7G94h
zfd+}Pz|KybG;6IMN#}0KtL6r7g>LOv&Ip&aWB6r#{qleomhbIiA&iWXb^{juC?t_t
z(tBE9;U#HKilnX)$iPEMc!tAHUf$ws-EB*(3IbnMUjnLlntUi<uH|i}gv?J{04KuB
z{re~HM=pRth>VCduG1pjeVoR(z!@-9>bO+9_lm=Eo(VyDJa>Nbtu%dZ<R#s)+D$~x
zv-r0AjiF9v^RM^F?Z0dblfR_jxRXGC7e69cn&kW5uB^oTPXS_U`dWQ-6?dPAbpTAQ
zQTWCD;`q1kkhf2_HimQ#^V@j&fIbi*gWvrl1zf0sYC?_lSjyz@do0(5tiIs!(sF73
z6T$u~kID4E8}D{BwVjVzQU6tS>}!_{4<5oXY|SiB*l=lgSz3FOG4EWPv?JLzEsP|d
zb9Cjrw!HxZDuKXCwz_{M{mW-VWIyFV&w}P}xQaFL<NG>$J|A+6J<*VNoN{$_Pln8<
zEQd&?>drQ{5!*!)9z`y?xcq$H9c9PM&C36Lem=f@WclaD#Nz1oer&xRxu^SkKIEYI
zaw#EJ{*%g0LQw1$5n;0_Ct%g7u=&>3HfrV1=5${^AJ5Z?!CG02GZ8|Vcu+`r);3%m
zk4Hn*GVGbnKpKKfub8^HKF6FD-I%yOA|3E4pP~yBp9}!$Ow9^7X|qrBeSwJ?Bw?EL
zyy^8~$p0)N5|S(u#f&}fE<zUB-Gn$2W!zi;9;qYzbtB@288nU*o5f|F{S>euAmJ_W
z(xqWnuX-uW_GD=r>C{MZ?by8r9cd>Rjl7=t`?cWFkN4NP$Oxw|ZON-8PQ&Q==7Kx^
z(pM9m9yosJ6Q9bTz`5#uD$D9<Z&Lp@%F{!C`{OMAhp{T^Di2Mxr`GQ)k*c$z79CVC
z<M%VvKU=gf>f=T>+AaJOEF#;(@1}|*BFy?et$#RBvfgzbYxP&ZO|#nU#HTJK2-)s&
zoWBv$T*OdKq>_Bg|Jzt=Th^YB0}xn8s_A|XELxoSEVgdG;0cG{&~->1meMAb6QNJh
zHN;bS&QHq-8`U|yOZ%O3r$uXfw#w;g#Z6oU0?qxY^U_db8y7z4$&iF{HHP0FJ^SNt
zfzuW|_8-K==vCUE{j=di?><=O-t|7{t-dm8F2gWnF!Ok}@g}3j81}U?lXSt}kr<se
zPC3M!E|sx^&haDY`e(!#YA&t#Ox#r6`&vFU*A}hAtZDHoYW}kO_)$ea+ErD>biW{X
ztY23(+x(L^_pOtaMSs8z(fw70;WAlW8gtPh@6$KPi5eTgE1CLkz@2~{C&i-vUc$15
zgLJ=l-;2Bxd)JQaj+A4+6>5H4)x8vJewGBlfBHzE4X;^IR^je|ai^A?*0d)+Gph*3
z5u^On7FI>U=Xry`zjV#1+S*8=78-6}KB^D!ba5%;p{v97nJvPpQRPJ{9LW~YF=rIR
zSs2{Ds4qKpx@)~cek+Utw^LT$)Yo?c9#)=5Z^&aVaN+nNSV_!5?l<HrCflcB^>O*;
zHp<l|JWMwPvEHwTV4>?P%eMBvp<bUhMK(SaL&+;o4_lWQ^D|E~6f$s5pmRQ}p+GHZ
zqSTl2uiZ7}wdc}5HVTj)Jcji-(N3BSd|cGl_)QqfJ+YgA!j&aGrmtxf#oqv((1o7?
z4!)cW2JG$Co}8YZz<;kQEh<Y!&0D*?;pxy>F)Du}CM+X*A^afB2M60Vf(Y7R=lOxm
zljx#AZMQVbz(r87GSy(7ow_zv97e)#Yj9L9$Y-R&4K!P3J2TOq+@e`F2tet?Qw24l
zZ;1eBi5{oigv@_Et+nM``z{WKPjNd;gR%*!*l$9>OS4kfc8ew{<ocrwJzBL`c16YQ
zYn2{lKnlyJ*pPOFXL+-dS3SDzdCbu_RdwrtSeM$MVT^T6#EM^=EYuGK;1H8G*l+Uf
z3-H*@z>KltYQER=+&=!WWrCogaV>3;MT*VfGFLW`yITpMiMhYM;?9`8p9i%Svvt{*
z(h10Go~1rr1I6arhZy&P(#|x%8>kEc?^>EGQS{fF6_QN13#DqHl9)mcIoF3{c{f=o
z$ysDQ{getF3|C+8hWJafY;$5XUs)`2V!T`eV;WiB$Hhw~rZgu?WRorkQGxDg6lyy`
zRyjn`$^D5jZe=}8t{$ZPbT3BPC*V|%uaCNrV^UHJjAd<i%W0K2E!-$U?r()3t-+rw
za@-YxgdI!!cI03)&~mk~N$O%qW^XA(J>(>rH_=}I1q{haXd@Ck+RoQn(X`*Y&Ff3%
zvmy;ikF1k^<05@UXLStx>&V?=uT-OPoLUAsL|vR!OD^F}yNy3kC2n5d(TYt6n9=22
zLl<%fJQ-<o!nqS>AY45ItaQvk5#dIGBAbXD!p;lX0&HX?5c4t!CR;8>+?)Ykc_AEx
zTrd~Qcr#zXz1%P#O<eDMR^-?y@iTpD@;eG{YW8a9li;)h)<m#Tm{r;`SL2=>I9PaG
znmwHvL=Lg5JkHSJn<ZfaJ~nMooxUxn^a3A5F=!6w52O9tewEN#Qqd@}r<K1C3Hs7w
z?Wm)=F9pI#1>FqceaU4FA!>`1xL(4)dtSHGrV8W?uU(r+Yt-4hlCrhAn?6pY%09K=
ztGZWQSuq<eB#PCbLDsWK5K|f6wJ#_-DWkrmI&le+mF=qeKJ^Xd&{7HZ2!W6N2O$<5
z4Zauy#fJ#Rr&0g@1HRxn1^kAF#ZV$fhA{PBLEtVzY4WrohkP^UCl!rt4%jym=O*N4
z$^JT)%-_Hs%l*3tIOM4IUibDc@o#BF#&-gpIU?G1{h!!KCYjspBO4yZN!F5?UlD<m
z7z+@NKq02<vpQv!oUsm<?HMj<aq}XjW;d%$Y#OngCl1!43bn0LEapwxm1nh{xX1zS
zX4vCHILnw@t{sw=s|6a#5o0WLN#lOeR(1Y(W-t&5D>&qu<=D$-yOoy_6A1K0PFZ~O
zwT>AA^QH?JH}qY1cC9aACAn+?#P}0H1D(%#ELy0W?NI*xz>BD%GM9sZ$((CAZKF|A
zM6%}LTa0#{l&M+yU4IOgZ1t)SA6I?ae+mES&JrHonaMi7la$A%on=W7e~-2OLbw@C
zI9UHu@6MrgoS$5}&NGM2(pE};?7nPN*M>Ke3~QsOznX6b5f?@dm`b!)2c*OJdA{bz
zcG^HJ?IIn1K@@elP9K9|<$bSb=X-rVK|%>3Bu#zxq1ivVC{CMtN_-E6v3ClykGIIF
zO}xl4_KESqk3+VP8&7wq)nqQ-+$(zMw({mXb+0a~JzXfv+wh;w+d9j<ARioCu(kO>
zDd#_z{j8sdw<qT9#aMiZ`@i4{;3w?=^=ZWTUy(y5#{bzpv7>P(f5e9Iv)ebgx4Nr%
zwC8BRUco(pdlkaAZ289v!n6X~P!>pX5>{?&kB{3;8cC?2L;@M#!f4EI@631itfc{w
z102HV@-N5bH69a`&p9+vzXen|KsICq26z!9hC8Kk&(eZ?S2)H;s-cDy1;morq#`~)
zydNfR_VZ^*5nW>a+8K~5JRE#HA094_|KA}$jBhs|m%kGkR|Mbyf_uc-!KcjJ;PJj(
z8!oE@<<2_;o3ua}3Q&ln{bLFqI3bQ>F3JNh4v-k6364Naprtwb?4TVmh{`+iKHozX
zIVxA2j-LJN;7ZIi<%SyM14|wKFi;noX-E0vg?-k4FLoau<m%`VF5%>Fp4|4}0(C6F
zmji^>KWknr`ntkH8lsdzJ~ITOGF{udtZ9WHto5=qzA!NBTRMI@5ahw@1oznxma#ga
z8VS@wCm&NW+Tk1Ia*&VP%)P;V$q0uU<9@_KMgOo?{P-PZl!Jrn7iN{{tF|&>5#>LB
zkG#Xr(hK!}OQg?h%}GKcu`2#D=ur@F>j$F^5S;BgTVU4x8-N=9i?qpIcA(et0_u3|
z^=bW1LB3BWWedZNce^VQ$|3B?rP!2u`{E_W<DYy;fZ=3Ysg@XDEj%>J#iAi>8yitr
zY=JD9G5Zq!_%amzQr}1WS$2WkVngY2j4i8gcZ~(jYhp<)g_U)&WWzFm89ZhC7f{?=
zyH4QA{Kj`HBZO}$i~U9aPw2U*ra?SOL9D^%xF6>4SIvC~yA8wTy5+Tbu;X#%)1!-`
zfEy)Fx~Qv`ny~X#&GOb4E$a4FZOAV=qK#JMn~){_$do$G$QiV`>{+M1EG3I82ZnxY
z6?@+-)Q~hsFAwX1%C*yPti?4g?c0s?ThC}1wzW;@pw_=5W;6tjfrW&<D{Z5Ui|G4@
z`0?h-;=Q^%{Ak?qyN)|gN5uT&g=ahkl0B0?MT4)<)-G!<Q9P5z{kga5U^;q2Gs(=6
zftjPKa(y$9D5}mZaq`XP>TI*YV%{S4jd?qL<*!9n?-)Af=WJYscWYH0_JiF)`HY4v
zul$x1jc@E@bU&Ir)1D=C@ecN4I4X@==TX9>l<EG8IsG6hN%QQJEp5SrkD3TwiTQyj
zm-2^x8<@v7oX<TKr~d@uBPqK|EVTd(qULszBZUtPPvyrdg3*nG%RdtiUrBY=OZ}Sa
z)K%s=9YrBa!`?MUZG5R$$~D!NlJe4f5CSBz=}!}j@C=tmFADl<VdWqAN0x~h-$L=u
zaK$uqF8q{eI~9%O)xRcqRA-KblQ~t0p=HQ~F6rjM?Mf3Oqy-Nlj_eUj$J^aVu>&s^
znNh66+pUWgT_82|L9={md?MAS`?GQWG3wXe3DSBwh$P5Y)ar(RWMnc#HW^rJ_sL?(
z;HHs`3zvrn<ZR_rw#%#A@mEQN0LWgV$g8IqdLSDqC_D*r9nL;;M9aIQd;xXtq<X`x
z&HVg{)Fzv)Ol_kU8SHD8h@w`koLazKX;sU*QcYX*g`OKrv{xd<(nc#fA<drh6_9Ia
zE~SYtmF&XYh+2HgswDC^8~06cktkjWjT=#8B=H`)Z}aG2pLEJvS_Qk#2FjoC5iXd}
z{_`6ChnI|zo%4UchML;;TWl!*h~j)v_Xv#SL#Y{U0<7f%X%{IN6b&uVN*c0j>x!1s
zX7jrLUV+3^A<<N8l98ysQPkndJ@>yt_JW`uYj|&ZFMfFM@Ht4h9)!>!D0_|vpaVxT
z1hZ}IxE|>|DiTHie$dxB`Je0XWqv;XpBobor=Q2Wqh6Dr`xE@Ww>=ntUZ2^ZORgB=
zX;2$w8sK;@1^BHT-XMlhwATo9hhfwg4#kN2dF)w(zhfkQ7w|x2e<gk${9Ch6Cv2M~
z3VGP*PcMwnWM#dxO+JV`yQY)*9+o};pQR4>N~E8>P-bM0d{azEHH17yw2Wj6r1AwM
z*V3AP1e|``4-!^{_W>NIw=+GhT4(HyyrWze$QRq>A4K@OM>B#}ib!(LsJDdV2gr!Z
zH%obgIhF3UY4ix3OcmO(c8ZrczRqeVBnA<YOUTKlJ}Bg&VkpLbh=7jJN^|xsvR$N)
zQl{&$3knI1bZXMPSDJ}MWzvWNLQ0hpMOKX~%`G()7>3uhiFgMG^gLuFhcTJQyR(}=
zG@7)*Ze!GMW={yA2Iq5<O!h^vYAev=UBiP~ID*^Q7l_!2Y0?i=*`i7tF0w@#u9T_K
zH3L#Aac>HD6ecOBD>SiY##Wc#vChl>a?Hl?UJQ3<tlv6vF<&XuV9`MJ!!w!OUOv$}
z66SB^6ci<ga#5Vxo`GvZP`+gxM}Wm;Q9$xk%vp{ahY0ckkwmb71V#*&sMa@-go#Eh
zCW)9-jZqfOo^BqfP^=wv5@+s>M$@pt$Gy3RKm7DKra^Uf<Uj#tD+<1=%yd``1`$sT
z;T9xt3ju-z&H$S~1u21Is7hVLxh$gO1QjY|6`$Qouu=AZ2XebYuc3stI+Y3B8cBT&
zWto<GidUc-!%XyBJlH1DF)tZ&F(D8?**t==?4l_b_f1Dpgy#xv7L4JX`Zh<HIpWbL
z&AFL$Oif<7st{ZZ02@PPD^)6_a##LYUF_h42)Pzz!XL5KAp{~G)bur#MynhmRR}>*
z4xq@cv$${8yg{T?9-~53TP%>~mr8(u`3=c@thQ9tG#M8(UDxE!+!<ku!cRqL9_IZJ
ztQrPGk(6RHXc06S-X&5F)-6mxK6~#H=mR8J0Z718T{ol++8hoJzN|dXJt1gRLOa3$
zi7nT>)f!IdpTzDD8dM99rAdfD>`c~l5pOzHXS9a5t+uCl3G2F7cP*Y4V_cO{70fvT
z#*+FiqC0NJ!hDD*T_U3dJ6IsHEauR?T$E3vO3f^LM}~;kQ)R?oE>pF`h^SS#?d%GD
z!h7#p4(yZ=xCatdi8YYR7F+-;LAVbRo>@dP2l;bA463CrX1~nRu4YhUsp{eSYOBl&
zl?KF~%?S?&*|I_!e}-7&QzHID9=malID#lr5KTvN20(PL%0DdnE~(9ZF2*KXYyWtW
z<N2rL$i&GUs0k?*C5D78J9<%YZ?LfKqQ+M5xT<P{0(YxdQfqfBYEz{zrtv+p!(!;5
zT<*4IZ=|K|V&x+`u1PMe2fpW#ML?ranJUG87<<O66Fa42A8dQ_;E|$7^jg---<5UI
zPdI>!Jirf1Mv#XgIXl<+D>N9W@hYn<3|rU{Q=>Ajh{vf{cDb%ng#b0p<*(7fPH!7K
z0T5T<y})^Z#wht|THg<n)5UM|Q_@S5z`%&4<tEEF1g61T5l_pPkwfA~QN+gD1t)zB
ziECBf3pkMR@3RHG0D(1QVXv-$+)Sq5HgJX!&<?<?&xe4+s{7F8WqKJn+F$n5H=7fX
z?-5u~E}i<8^$W&`1}Y;de>lpk-XY<pefp|9$&n=ztP?F87j1v$`My`6sw`NsTcb$!
zdhaY`@LEua0<}3<g1QuETCHEVQ&>0jN%RFCVBegJtK3$zDv1!SbrJML1&l}q7bkBY
z^gBe~cI=kgd^R30hWgbmF7uYrBILe0#<4ptd)pQCI+Ied`I`aXyDU<g@oc7osO@BB
zhIEYiUKcj6eggkh=h2^XQh1?QUM34`dK>W>B84SEgHKbV%#T^7r;A=HSl^eM`02E)
z=3+eN%@}uh*5Kwp14C@wHW;Xa9h<ur_Z%IvqYo=AxyR}V>*@(#EQqskQWf(8_MCUd
z7UmzcG}b3H@KwJ6?~naFp4jM^mkXB>0z?fz@k8pg)whWU8z)lU;$O>ZBm9!bboDy7
zrg)X#XR`_K;8~=fE99~qTg>a+RF`U2omx12?)3Ke>>9fLxD2_icK8Y+re)OOvY@(5
z-w*0_q;V}ax;L7N%eo%Fy#_|4y0cG3WD9h=E-RN}C4MkSK69z_MOXdq`kF23vZG%+
zwWYBhw;Hav?8`c03G~tVUe3=6_OjGJD}M4}J--0R<y~g~g{Vyb6|`dGVEbR)*((~_
z%3ES6erM|EHKij{dd=vhvgV5@N{4c#vfJUShCG944@g@;l3`&oPg~xvM$*(cz+sYG
z;55zY&fl}L9A5Yb=O?rA-#Ony8KRI<AgGGJEJj3!4FVjeivBpTTr_B1W8dmu;UMI(
z#v+s?7BC{$73lo@zb*dvW^K}(w+f+dT{`=-0|y5S(M7>)^gj;R-+TJ_J${zsHwN?o
zVv)t!;y30Ge4mS}S+@4L&+c??5cMO7kr=`m2@z2kltM)dHe=Py0|(tv6!Hkf0u0e|
z?tWWplckA}l0=}V$kQ3}6e6Zbj)RKR5lsga`qNcAwf*f+JIv-JA%YZdMG;46)d0yB
z2KEX4l>#>vnh~A<dWQD}1#{LPKDL&~U?S#0{5YJOgHiK){3Q;QXjxxFGKgLNpaZV8
zCZ!O-LE{nKF^qXxuZj>f3T5TCF}QSPZZ=|i*^q9*Uhy(6=5m_5;9o-%h)w4ey-j$M
zfdMd~cAkC++;s?>kKKCO6ZCKL7SuakNkuea3$2TLDyodq6%A}OD3%GPlgGz4yzdQY
z#75SEH14heV_FASx3P}13E0r0*cny33x}bNdcnpF*kTFyDE_=E%+ZJUW$f)g+EmlO
zaA3<dEm>)xdU4Fu!8DxZruQ@ye)`J<>Z9iNaGz8!2zaCWxy3Jtz)wYsYMF@beHcMH
ze9iav^FnRk1$C0y)dF))S@C3_jnL#^ac0%;1FbVFPc@`dX*l*qX7x&fD9v+vD&O9D
zo0^{~=cLMa>qd3c%tiay5NZmyw?3YwN!zCrUwX{~QmOw997&}kuZXVrD8vM>tvZUd
zA_^jl5ZF3;c^}PzRDD@}74NkbT}8tJ0831AJnt)kf+&auY^@mZl$cw+jDj3=+2C*@
zkvW2pWh$HRoRJawbGAZ4oj-+I$t-nd%qv$l-A%?Oy8dF6J}&2$i*`X%)hMK2mO6W8
zC!c=Hpx_A2e<2i4wp8s@XRIs4$E`_YPm)8eE4!`7mGcRQ%d>{yNm(a_kVYHmeJhqv
ztU;JKB~4UFBZeaFj>q7V+d<}65`f4l$Aua4iGt-PCUG_wUvvfkq52`%)nowaes0T`
z)q!1aRWHH#x-wavg4}-mQ{yISo}_jwn&$;#)F8(!XZwaD;ib4GL_`<agGii(9}py;
z+g2%$c*0U#-4mi_n574R%1YC3g!1^dIH+pJqP|;X5ah(@O#LBNLXaDpL|Bu|ddTL9
z3?oTba6rn7Z+?dPH=|(izJ+tc$l)5D>ONi*$cGbCm&>Iq-Le31xWt<|cl#KM0cC(E
zC12wV3eLA%=Li}~7-N`88A1Cs1Hko4dCuyVl{O<6MI{nrV%WSl>}`gCd`BYM$aII^
zXk~xn4|s>jdV97Ki69sT1o#*S*3c<+asz?9-bPVV1cQi;VxE}2A&1KmZZhn^zk*}3
z7USsj8{Fs*E?N=++05zkRs|zqstl5dNHm(a>>|L_C(@8Ch|zv9re<LV0(W7(0~p8b
z5OuFBQ7tC~j6>B>_0IH-^2z8}!T~`)EIn#AXCeo+eKcz1{+LFt%s&DQ(6J^{6fYGk
zHW|xSzYiaCf48RnPn)H|-Jp&9aHZ6L3?P`;$BPpHn3oH7gbIr(6cMB`y78lBC1L<k
zSU`w0iUWlWAPXRiWjbVW?9R=W7$FY@9bsk6`OX<OIJE-9$~$qS%8yA116(fUf+MMT
zwPXm}Bu)v)?Iunl8(^U)va4rcphjH9zioQxs4pO%L4j;A=1bU92t<VDr|vV-y)miu
zld^|#Y_=N(9J6#3I+4y_f~8yg0mepjgqIky{8;LELaH{}XakaMTnn7crI(z{{4<?m
zJj?hsQ}m*%rYUvDxB&U7FAX7hqp9tho`qZGwm4e*pUm{j@bYI@<-PZZ*`X`wFGacN
zW98IB&5#e?oZp&TD6r|pem)z6VjFqo=hh5({T~y^?iTp)N0Bz=7Js}XC{#F^=Xplg
zhPq;+*dB*07d$i=v=^(z^9g0y<0}~?NUHijUCtO$t&*~cgYk5P*B1V)_wIQ;!j@TJ
zTs9rzRc@bMjm~gBfwPBao2Yhvd#uOGt}4azAM(5dixvH`O$r>+p;%$t*C_MULb(`g
zEdcL1>PmLS2fbV+Brx$g;g{60zhnNZ|CmOOa;!)^gS#V6A)NG36NLRvo)ljGU}}~F
zU86BV;{ORD?C}MO`Hnv-{7d<tIQL)7k(2FzW{x)+((y-ZNWK4<qwR^!G@>Lg@--nO
zlb~B52sfkC`UF8f>E>!H9-ic_oP7Pq&Bu75iS*3{j2GI*l%7q61X`b$I_3SV-oCH%
z)5+=Sh%}yw-e{x7@UGd}#Cr5#2B_h&E!15Ly;54^3Re{H?^s>mX0^wq|L^6Nzi)$-
zkpI)o+g{DC&Ck!FDFwY|2qJVL2D=6vlgxB!R=+M*Jq@DbG3{?x&s%Sa`W)}e%jxM2
z>{v_S7%!z}|3i8Rjhr!kojr{1DB^1w9y~Kz*!ER(Ti3^#v!Wc66J+Q3J+dShOHsGe
zBs$Ntk_BCwn6rWp(Dze9C?)yL&8A%`?|QU3x;ThAB81!;Kd7RE^cO$Cz@rzQ@zijY
z4VoErKHVoE$CQJZ=L^PmHeuca@IjM50KI<A5y8DbYxFKxNUX_aO@{%}`gO<cd~dI>
z@Yau*zbN<L?JFzuW8ZMmp)+Ulwg(=vSWO^sXEDN#q0DV0sJTMKvuJb|N&m{V8@ru8
z-U_)lc`d`4gC`=Wv;JRWQz$%1W@J+@{ynQ=AWtMH7@#AJVR1bM|Flf>fXmkTmE=Xq
z0THImhc$EkZns&kuAg7H+rF&rFZMHF;Roj>x0cG7ZfO39MryrRn!bNV;`29C-)gBH
z|JHl?@EVFPCVmw!hIt2YMgD@m8ocG0dA80rt_<tX_y)QCV;mggYj~muA{8{O&+-_)
zopi@|=QPno2QC!I=w!`~>@Gk8!1s9S*^auWf}2TqD~hc!Df0Ck=`utvLrnLH-z+)Q
z61%V!EfEDfZkSX%7eDWkb08<K)sFRKo?v6L-BPf0e6SUS@SpOm!ZGKJk0C+i2e}vE
zM5`D30BpR7-P;H)I9|lC##NEYtxpfENwYgNxVYZD16)iQcGa6)im};e#2Yo+DJxmY
zBvVbFK>i3#Dw2adgc~_oJCwHzh&Q(&hTUy?gqE$YM&4Kz)4LsO@;XW0kY|XEY}NT|
z35-dK8~Se{n&6Zsf$YF28WTh;3TPH}RliCabY~{iA74(K>EvhC7WPiwWm7q6(GKm0
z;j#ES?xzrFsYmE8x#qSqB<%Cd^Qd=NEX7JN@#Q`oOZh!{n6NZAyBF5&3IiO{xqoh-
zmD7DyfoF<WRbzajjE;+NyVTDzAFXCC=4h=GaH5wo+MPVPH|$2nK2!0p324{>O?<7I
zNCh@`TUT?Ni-1QH5v&G~4}oD!X<2cTc|26OOF{QluJ$nu#+bha%ChhtErC{Ov)E%1
zERoxyf8TAEh=?c+6$foeVs8osQpON}kRxviN~T7LKBkTCkvarG!0z`kpJxhO^1XAo
zm4Q7A{GosXWiCKE@5MF&sQY=YtleS&c-#Ve>1m@`9@;9~v?K;<2p4mtc~Sly5-~5?
z_;{N0^#tgV+21JT+KZe)Xzbgq+q7IL6R$UyOEP6W09V)~=k_Us2q37i<vq(m1~DLS
z`6wC1W6+XuO|6_oz(3`iBL_z+9rFZKqA{1ofxZ>au1s2qL?qy&eW|NF*-Kd{k4(5*
zD?)Fc>UOsI`z<zTWjTJjd`u{3AQI+G!+>LLE>*5qgMKhk*FY>(bugfTt7`6}p-$;O
z2O_I3hCvv$R64>UtXz9r$#&I@4Z({pGF;9l1H;GT1Gq%L9Wb0Ut~p)~)!68+32ly5
zW5zEIBj%oe4tGFP6ysSJ3LBiZhzFFOX_|;@h4Hqc-n7v<>86kWI(RSdU}!>)y|l8C
z1*-NHtV4`+q2k|31eLpxACH5#6--Pdu~K4_h>G8;s8`ySM7H}9R0@gQWPN&l??W=-
z;6K%`7{bh;igK<JIe1vU;2urtqcrxiz0f1F@ur8mS)(L#bO(=Yc|(P+a|j%ueU|dz
zy%@+#|NiU;-B7+vLHLGb6m7TMgjeVb-9^Hjl}-6&VtT~<#cGGyXluc<e%w%KEu>yk
zL-D#cnyx*i`ilUZBQe_B-4VoW#Du%HG4>q(+j575V@V6Mbm9(~*@&Tf;{TyJh(Rf%
zF}`#bXEP%nYjcpt(ct?gh`coqS6Qg*@9NwQ9et`SLxIoM;U7bazS}LvDS+OIR&Kf&
zOH{idkkF12P!^1(eXvTa2$wI0bSLq(>7sBVQUGi096-bN(pE~0&uwuDR#v;$vac?r
z{LlZz7kb)q7Y-?1_R>Jn;3y?Amtz={FUcuwoUYZ|KZ2vndy1Kb@=#61!zSQU7HOaw
z<WJ4do;<x!Y{_>jpKz<<x@ebG6%>}NT}+iB(koeHMCEQ4PU}AWeNjjzI!Z)nwx365
z#XO+=T(fb_a3U2<4OT1aCkl)uO4k#2N}n~wIf3=4!$MYcmttG>=i8?+hn!xgf9$TJ
zZVyBcu(2Tp<j}jTLhLmm<1Pjy^$oZ&V}=o7-lUzxO`lI00yUH-QYR)-5<Vkh24$b?
zx`LH~HsM1HiuZn_7~9C`q9RLf`yEQj`))PGnrX$$CK)!-EZ?zHq{Sv3`Zfm(_qpS9
zA&I*D)zoQ-0kaE6VMoA%ZAq!@HXRI_@sA^8Y|08sL$Wk!4EZ9J2mz<*nNZnX9i=S6
z;c($PEuXc{`m4JZ@mlnMVL$VKaa>kb_WzaR?r8i!)wpM*T0$|UxfZyU=%N!yY|Rrm
zj0L_~hC9e~BH4soZ{hnD_fa9P_@Bv46a0Ckm@*f-&-sAUC8y(v%ifDW)|-B|{Y-=e
zNrvMn^9XY6EU10keN_3Z`|iQPf$`^L1k<%>lnpCPgaJ-wx_0fKN!HKj(~!ZWcgvg*
z6UO)8Reo-Ex7W{`%j=um59625i`%Pc<3*<`K)NvCqjw26y2E*`3_iYu3-|5kQW4>b
zYx#laQ0dY^9)YeLncmxL2bu30K8Jxo2%=13w)*YLIMv2T#_~8MAot!Xfs&jE2BB9z
z0<r*)!2t2^JD!t$4S;fZCkeuA0@I;l+%h(*y5~X=s@j}5dT@cmj$el!ohciwJ|6Ay
z2qy3@`(>xCXH#7L?F9xYAC1P^`}wEctHpv!e;RHb$rVRDwAfHy0@2XwWcb!(3+5u7
z)n2TCepEkZ6^Aff?qMhz5d4Le%p5cHBQ=hhr`@c#5X3@^9uc%H3%^y-`-ab@fc8nM
zj5dy7(y0dYH3O=*2j7A&_eTIwMt2<Mfddv2lwKEzWuI-6)Tk=b6w3}AMaegE%+saP
zK#2Iak-$)|Rtb}ubgF85Mhb`2wvJk(l}}EwXLtXFN{M9Y{?)SI<{?I1v`h4Gvd}2B
zs%9J(MOO7DTipvhF%aS|*DyIr``0ugIiUSgHVz69K?fhCZ(tVx9pXN+LwVMFJ~=l<
z6xGD(`yEvUU6#k7NhMz=-Bc3Q1hdYBc@8Z=-YWs`&NxwGvdYa;+CqTH4;`G<+h$Rf
z<L{1`kq<n85&PH+f%Ydr73b}RSfCZ&i(rWlZ{M(|(naFG?(9dGA}Jl!HB~}KuT*qs
z>9iv?2PzJ!5-YPygmW1X5{S(|I2_8#{^LWnTZAo`D-vL69Fb2eulWbW=(WRaMGL%x
zzDNuhKzR&#S{qEd*Z~<~Q~3it00=(AM>w!G>QWtm>KnGhj(?_I#%GoTF88575!53^
z+!sWqaMU04T`6gIl;}`qSE>K9F`unfQ~PE)xBcp6I`pPgHM#*&Vd)WIM0V$o_G+L?
zLPA|m?~9Q!-}k=<;?*UwjAY78;h0}7n!o}MBS(PLcvv0c^p;f8IJT-dfOY+`S0S_(
z^S~e`iDrB_uPz+AaA@}GZOsagG6w3kWdX}HMhD?yO1@UxKcNmvL9|tXOLT$w2p#|*
z+J{f3jZG%?pUSeRrGt*a{5GeoBI;;0><vo3^gZc;BPF_7Cam57VG)DH%s>&koIFKe
z)yGVD^dwC88UQ{MTt9?bq6WT;e5!C0`xZYmaB>QC-sTt2qmuT9Yn9fmu6tT(k=(yG
z&%yi}qM1roIsla@tT%)n&jvItfA190?XWiqHdj&qaQG9TIR%2lGbd9T?5c_KXkfD_
zu-D%`^Xmnl2J32pfP^oVj21i5mX8orRvOuAXxW2}JKv{|izcvAakFgF!&?mp&1oof
ztxm45?%1os9wkG|FlGg^-ZTy@4x9KOqUxlNA<$Utim~iI7mFSOOYcBvNR=Gh3_lDx
zY$Tbhl*gkzL(hZcxJ&zi`2-P){iA91V=Se+gx*}GYa0rn1wV>#P$5FSt!o91B-Yja
z9KdE|aka1Z0kJ0AhWGQckm}t99WY!J1&(u)n(FNz*jE?z3ut}jPLB@&0S~$Cq0dfG
zR0h;IMO-8Iv=eFS@5uAzEUZY;ZHs+9{rLin;<{C{;fI<h-*WSPvzE+#of;yosai<^
z?-hYBD2>EN_K9z>WezN-Px(vgw)!JouCW+g?-^6Sk0{wh?O{RudF6$3FxD4~^RLiC
zu+^Y*O7axxPwOY(@q3Ip#CGlA?@{#Hp`_S}{{6t!&H*@mw7?ME-e!8m)zkFpcc?GO
zmsvgZ^V7%(Z^Lfv*3B*L4L(j|x5lJOuw_Ud&oZtZ?n^m%2mM>M*tgr#sNsE&*&JA^
zPm}r5PMxHk8Pv;<m2O4jtsku*BzpKwrn7W&B|h+DWHyPV(#ET2>#Wh1_hoLujARBH
z_-3bUQr*^?0<VzmH~s40?3w?Bng61g?2P|2%-m@1{QEhH;6JBd&^BQ`lTaujwZWB7
zy;(?`U{ui%ULr`S-ApThq$)W(7IkC4_aGccQYLw6Cfsl{Qe1wt^WcHsm)y|ySpW98
zf3`7j03BpZFUAPp&aUm1;!Sf55k8$&vRi>KLLrn)d9?of^3k*rBJ#79tlr0)6EUCH
z<K@l6nbpJRgSf~Soj^FvA;ExSfni+;d>!Jo6lZ`1fikk4gi4kAaWuhVu%3R_*xY2M
ztXgT;<Ybz+)#5^*G&s}PxY_LXiFZVh9_y=7`A<P8mc6E-T(Xm2HIvN!&TY=5N?d^-
zAu^c8j(?f_+OA4Y_rCSw<9_!L=*L+n#w&q9=7m6EKBSET?vdjJ{v0O`DvSWvqwP5v
zp#QSEu!%=3sK*<=A1dEr(cpBHLrj>D-WTJ%$p!+CfU>%q)M)lE+ZQFdwT0l&vDx<Q
zCpX{gGWAv-G-M1z91est^H*yXhUa?m?`EHR&;$y?*r-gag4XmOP|W4%z^EPn5gDGu
zt<J_EozS?-BU6rR_&p7$XOO@sND!GNc;2&2IO>5t<RPiJI<}hr`}}_WV<R6+bmp{B
z=@q_^#awlEf_@(PiwA3+c}&b^_B~k+#>9x^L-85kqigIEMpZ$0)!Jxh=Zb(v3>0Y{
zzu$LZb=RKMVUvs<Cx`!jUdpxOWrdY5aGm%#8dOGy2Y#4-8JB@GeD6K&(9s4&2Yf`g
z0t2!~K0h{T8D=<UPY#NM8{9dpcS`P^NQg4*2;0s@yA%6s8LqS<nsilETKTQ8LbXGq
z_3Z(f5?_}i?Ck#+RC3a0dqK(hq60J99-={;_ujhl6+#~|Bp8m1VVL#bVD904EkL0B
zlUog26)ONP-QVgKV^#Tx`<7T3o*>i@$K)LH4jNq1nt~h~HqR{!6teOmt$QdKp3_4#
zEQHiDH$68M=O$VwoGKp8@#Xb!9qXCG{vr(mEUp&%Mz<S*Kf+$FN^GfC#}J*U`G#sI
z>c~P7clSn%F7>xr0#1U%h~fKK#2pBVWO!iXn#X)H0^86=<G?(<%tkH|8vAnK1^B|2
zW!n2cIi#4NO^%M@58tjCyao}xHq8K{!`=r-L*MrhiUG@`4`!`J5ZPU)^N879snsdc
zJ>^?bqGVGesk-XYyl#L%(7^hzT>7xJzAJR<sfh(iBb$&GTw;<}9I+a5w)+9vwsQb$
z(=Nein%`M6Z4C94CcTv=InAYafn_9fD@2yE{a50=4t*);uN$ap%-{kQ*Vjflm^TJi
zu7Qz}UddS587C3>-2U>Z)tF&C7O<AO^Yd0c3aVtv7BrLt4xpLP`=L}>vJ0*b&MPM&
zY}9U#0W>PMO#xSX%@oJm#aVeA{s<9}Mt=$W6MT&h6vwk`Uy6%1JCFE@-ykud0!#1s
z{xZ)JWjEd9V*eW<A(c0KWO=~Oq`x`ZW~2@90#(o09P2oOL`V8k!nBt%z3DXG^^BKY
zZjfF+vL--5r1%y4s$s2?&AnsW;N~*cckOBARgdRTuf#ezDgi?8w&;!h;erFaziwSq
zs16Rdu6r?uAAjUJ>Ud_?*l1llDDNR-Tpk=Rh#V5mQb!abzJFQaSly<`pAhOwNpqB!
zkq5%<wyBH5J01&yeJNiV3HK)Wg0BJ9EmB@rPrIIAeUT-N6(&)3w3akB_=xK_<e}Kt
zi1_u6{MDgfc%uoov*CY`XF<@<^4;c;@!O6DxT#kl?_FZPE~)>kRYpQ#jUHAGU?Qbx
zGpTdi6HH^_grdTcLljyOL`Wb!&q+i96QsA|GQx_MupKs%WC(mK8ls_Yt7o$rE2^o0
z#{dYT#IB&>($Ej5I`CdAEA`bYQG1<K;~vV?4`ZE@(jnV=Ia+!tS310>rnk=L-UYpa
z-WeiwrZ+xCrJniK4pX<k;VKwg<F1uLlONnb04345NYuzjwU4(i*$`@6frC1FwlV29
zOZL+w9*^TCxyjY#63U#l78%I2%pO6pO+*CbtC*|JWDRn@*;=-sKRM;9MAw?yMLr3@
z_5Gsl&%o4lg;q6AXUb(;M`dY=Jmd{JUbVcF7o6ShE`%nG!8yEbLnDOJIy-b8U~r6`
zwuM5CfVliFdKRk?3~CN+SFt(m>={h|!b%_r06_OoNoD~<3-tIZ!6(Dw3wmpgT^TLR
zvi=r0@F}rk5Kg4M7W#uZ-<q;oZkk1k8;0th8>VCwu;0%?@OutX>+b8~lDo{FZH+aI
z#xG|~BD7ZBG4%Y~8nB}YRB9<kV8Filud_!L@IS*GEdRx}m^lAGGV;>)M{EeaU+VW1
z=!`SzdN&oa$vZB9MX=k#iLg+kX~+d$9;BRZ^%nl#;(rotDR+lu8!g~_L=GG{vN7S#
zzAOgpG7av)_+`M4oQpKah2=eMIt@0BYaiMf=(1)vQj2sMyK{e7_|l2Cm6HYgc}@Ez
zid;W`zVgb<(}xnBLW_X-a5{1wW@qQ<<>+V0%K8q5pVuZvj&*JA2*v>ig!Mb<uKc65
zUrQ>c(rdZ0;P_xkY^XC#w44b^olEvNG1#QRTXekX&cS7@Zo&VcB5K3<c~&HUai%eu
z0GiA#*(Q}7=<OSS{tRb#TV?HXN<qxo8oPF&OPL9cghIm=*PlGo1qn0OHqnGJIXq^O
zNGaCGRds#xAE|`SOq%MAs%D#Wa89XD;F#MNs-@5+mq#Um<D`2!;kR&+WR1<xEmT{z
z6xyWC9bLD^W!u7!7>~kt)Cs#Z0+Bbob6q6kttrH+cw?(vh}=0#(9UcjRm~lO59uE5
z<MB*J7uaVbD-oS=4B4*~L;h$UOD5G>hi4f!H@GhUWd(mT6Rj>e4*Y75;GNJ#?v_P`
zrehsPEzS{!WsT-alCE?im2F#$Ei){CU^s1U%mJI{*FJe5*v`E%pRBtYAnSCKk=X7c
z?x?;I=(*Xz6ik#TFG^I=36_rSc<|tO3~}H%>!Ex0aqtxvU2CR=`kH5i8myXN1-<X~
zbz5myJHFbhp!yo+(<yq(y3@->b=2YeY>3O3{=}<U*~xX4IqhU!a5l1p&Jy~im?Zag
z5T+|A_YQRw(Gw@z7X#ao&69kPRoNh$+L@oh4$nP(qVQz0s%d{5^2c-aKS~VJBN)0x
zr1Qe`a6xy3wf#|SC-QCpcEO!c%YM1dLx}kSQJKV@2>epV-hPDN;GeK3->)4W`Wpjo
znXmbZJvgK)M*f5D*Vrv#s}$b3A?Mpv6^R`6>*;>M#qU(@W?k2~eJ^0itLb7tjwnF<
z>^5H|1PdGTJ`le^(*SyqW8<h`2y8C`kvdsmXjMhDJFmhv#;K83P;LfUWo11_HlYSg
zIxbx3siQx=>JsYg$wpkGy5SVI>FGcBA>d>h!G8UfRGUPlNA@s{6b+~;^SXq?E^66a
z4vrtH(8Muman%@=d|Zs}nC+TGq|Fu(?5Q)WX1bw}?^RcgZh`hM<$I-whr;%JGx~+(
zh1@>bW_Zafs_p=<<yG{)yK9fEwdo;;0;a6pg`Q5%E5FoT9Wj6t<z|a-AsDA6uvnPW
zvMRRIl0cRi!C{QdTg+0OsaHnx`EVIvUD<pK$a`w|3mQfmI>)Lo$v31Va2Wb)&rDN-
z;;#Ed73K<)fLEx}^>l-jhivmFo?{B}E21dhODGM$mrU;JmvT~?LMq;_+2D(``1_OL
z3-tDL^zNny3_U~#pp^~ww$PIseAYHaG}N|#`27WVoo-6ih}wq+Tn^S2Hlp9A7i6Uq
zKaE3uhgAcD`KdcK;gXB~M?kSxKr-n}TB|J%n+1T7Z!H-2l$ACUy2W|(tsS4j<r9D>
zB*b)Pcm5Pxksc-iph`xX569BuS6@qkR2CvIP1vRd$xuB4h6rT)t2qkdf)XQP(c5>q
z(9poG&+1)@Y0lvY^qfU6lW_+}qt^R8$ipJLeNc0_?z1hXGRgu}C~qS^CWSh?zInXR
z!`p_S(2gSK-|qA6QEpYc%M9W66|ckvw8k_0W@{qrf3XY}mj6h5bNoBH|M!8S7qhf+
zF?AxK7qc;RF%>a2wl^__;^TvIc5yN_w1x86+`-t0+hV)-?Gv(3jb;^&T3%@e-6WDs
z2EukRJ}!{Nv&eddT2aOAa*Tf6;$Dt1k-VBA)6S>-G*TWq(8*5T-r(N$bbI~2yk1Zq
zFeW#1+y|YtIvekQQ0;Gj%vo9WJn}+}UN^b`-}e1J4qp)xxt*Fi>+@-YNbU0ac{F;k
zdpTTCCLJS5nQO-f2#*VT4i)7(8h&0_(uk=~=lxuS)<gmKIDZj>IQS#0tTJ(D)}`Ua
z_iy%D<H|LOYz7dpi*1iZGI5Q(c%LRt$Y31zKq3#cfB7s)G@yAhVCZsEgDCea&6$3!
z;ya;swhdw?s#7pwL$ic-m5k9dcx$b+GLX?o1sbudROV%L8N$-*&{8qwCB{YhqI(xr
zHOx!0Z8>#TrU@XlN-kB-qYj<VL4_vcZiM3fGoWmfYvRqV!pjY57)x8GGYz^Br7O~8
zat0b<YevBlw6#U6&=*jSOf5KaPBIc}K$4HRje4=UU2>lS!c-N3_+co+b5Q(Y#ycmq
z!tflK299^n{q%Qbx5N7CaRXSMS2O3^jZK0S6aD^75-p(0zL6O^JrMpqkr)MnCNSU8
z&77ZKMe5OgHUxx22j9GRSQvSn6ts{$1v&5+ED1;vG>H<^Sj5J)$lQ@cgokS%6oJnX
z-_<|C_wMY5MiihQ<A;|_FznU`D3|XPEWeKb&#?|V{I)twVPIayC9=v#*t=6Z(cegr
z#~cAVZo(0rOMCRG0Oz-|u{;9=R7O%Frij1(SjUG0V1)1!F3O~LXUlv4-iCp_gWoUt
z{h=}(hOC`affRULQ;tDR3>o>KkNvF7NP&vX$H$zvqcItQT@Qil9H$oU=UduNw%8wZ
zmAR-IN&fPd=L~c?W;hi`mmA0mV5lGTAHkEr#;3Hezq~-bJeRINkhMFwohQ2j>f81?
z{z>!Zv0(<=S+AH$UoxLZe8Nr^m@cX3u2$M06QNZF`(d>O@hqanH5<j9)FZ7S+7Y!6
zdn5|h)FNf`SNo!(#{-u6pRai{1uU#Ac9XA`>ce1~tG1i_ZMINcCIaVHuM(;K4@MM8
zNu6*e9VBZyUX$+3`N=7kjSH=U{b)9q1ime(3hKb}GL*Z$f01>!C5Yr9$oeahiFEOS
z8V&FZfgU&ETIHf=>e^+jwpZRgmx=mE&m{)jts&k$!H-1urPPR~m{~u1I|kuY*2<pM
z5HO^k106XE&^g8v=+@^SN0j*=f#5Iu{ybMXEmrJUEg_g{+ib#ksqO^ZvD!E}33Z%1
zaR>i?ge5TLwE0?RA8UKBl-}xwf{Q!)TBr};$#_uQ%&0atkSP&ja=7TTa{U%|g^$ys
zAIBK80-UYSyXU!Tr-`oYi|=i@N%4wtwJ^3zwxP$Z1l2`qP3bvUh{m&18(!`czLb`?
zzuf3K-Y4kZ-FLB(u5AXjkyXUfN4RoV-+M4s14wyl^ZQrk-CfR%rocKiJnYB<yn9vy
zK9xWFlp0CV<(M#HSQXiB2DpN}skF+*I-uYN=ms?)$k2Vw@>)@>AdOaM6+54`vlH*5
zdwK~VoWMl<?A8M4U}y7o5_e`kU`+8L=}1r{p|l}0R*y0U!i=Ksn8z%v2ev{IY!yhm
zzcxP=wDj%YKH)e{w0ZIyBkZ_3+Rn;{R-lLNj~s*_qGS_ekZlyFqjcajNyr?bx?4L0
zC|9TH6>FeOTC#SwlQU<hXbBc#I8BL@Dq_^%JaeoLmeyvX)%VIn*Q-0E-1=Q4!cX^<
z#nff+97Zj*!jIr8B@YQZ8hbc>mJ$e+qY{OPg5bsrAf=b~E|S8|;DA&4h@HRW6=%S!
z6FXQS0z3pLja30{irml3O<NG)3DMmedM@k>sjE8KI>dT%iu4c^1co#V2bwxs4NJCy
zB|9!h<H(A(qQQrB*<I|5oe4MMo$wqCWn3xiYAR=>>=2rgT0uqipgF%Q(f}wm*7p+;
zYpkAnH&->%lk(kVp>|~o%-g}<;#v#vc!Mb}FdW1shbg!TkV{tNYK!0CqA}qjwDImW
zMB!B_krccI_)w~bkm#YPt6=&K+kjRzdiPS+2F{`p{;8B-Bu{k6RwSr6YbrEe8C2EY
zQY_HEO|;W?)0FC{qv<elClC%@#Rn75I;XJ-7M>$LF1#w{#5W(E1%t%E#SMb@g{P@|
zcEh4pdmgy4Hn3vy8b3Vt2d?$4<HFa%j)MDYRQ_ehvmrR<XvS)K!@_I;DEwyTUmI4C
z^?`SusckpU&*K4DPnf{xY1=r%W<8$p8GJJ)tI*cv<^zvqe)8!5yc`U}M#K5@k>a|(
z@zj4H`*R!DKySTA0%_A%R2IawVgBhnZL1@LZHL;_E2#MbJKm`!Vz;FL)bs|WI~m4S
z23j=KPPyqKExuK(-%5IGC6ys|W(?JwX&){YRt^i_x$IfI6xsb7uJD1~+%tLop9uV4
zC1C7qtp76tYuMUvv%&o2>L1uU2Z$t^Owdnp<<ln7VI$aRh8ao(3%RfEV7^Svyps07
zZQ0|KPNHx16icQR>h`T8ktU*u@)e8pbcpw;bJNed!^Mg2lafGsp6ASU6FmPP%HDxT
z6eil*Y}>YN+qR9<=4souZCj^p+qP}nn7&^!$t3e8bLT%)rS_`5*0YKrAW`SJ<-4Hy
z<Gfz(KSu(M#1gkz(EbOylD4Wz@9Xyc@HV%0&6m1(!nn4TO868T6D$8G2PeDx6BvBj
zmz$T<+hp}hp|n@x-Dku#SDZvn=&)K#M=x1f){aw>h_0CZd&TjlIln<YR>#??`{4u*
z*t&eAlYk(JQQl54Ev(C;&gfK-!ocTp2oZ8%wz7$R<L#(U>Nz%~ZZNWs%V0Je_J|}5
zsBuDX8jS`3jK@C~Cho%zNQPHT5+%UhfEBs*T}pnV0xt_p;a7XPz6eWw89tFsz}?X9
z?WO@AVC$A6AFE_XlSH=+W5x^ET|Z_>Y4K^XKpTvQsH?$&DE}s3Z;_deIr6f+%QoW(
zu`@3>lxI`(moDSoK7ck!!a|>$!MZ52-E*DQx@!j~gmuiPCCLVYeK!cXNUH6>_+P0j
z+ZEavGn4B>7s)cw$4{4y@ND+J@Ohn41{jF13rHV9l1Wk+6#N<!F+rHjd^%>I=pwzN
zb%V3-GtoYRsa96UpOUAstunWC%e{)R!6gQm9f&k1B}K#?=B{0qtScObdpQhSZ7QTn
zYdtEX314K`+1@++C(C)}A}iYG)l{!*vc|20eK+xg0Hi<LsDAHAEXv>mwlb)sK+!G*
zmI%gdqECgEiB>+pCb5aln@S~VLIctml&|tqYqI;Z+Ch>_JgeUrVB{yWN%YO=_&?pU
z%Bn2*R~CWMt#`}$(|4&R&7<QT{SoUE{byw6enPpcUm8?ePbNKvzaJ*eny{v~8RPjA
zH$Uf>X=Zuw#>oSdv)Y1k)49wq75IwdpHm<lxMRkWgXR)#%%UaViUoO<+J)&1=DI9J
zBUPqcH|NNAjxYmgVynN3yrZgiL;hR2ug+51M14J=Zixqc$BJf0Z^|IPhYx$j9sEd9
zPuM5Aot-Mx$Q!)`MD$?kq^~~=4H_|6w?Rw6Zg9iG$ea&$k)DmHf<E}DfTTW3Nz9gC
zz?y`WbP${!JX{<z;o)=;N5}{Jy+t$CP1h2+3?z(I10a`VKZuWC!=?&2nZh(+ZtsBT
zp<d+^ok|AVb3S*um4?8??w)S#3=&d3;)Y=psMC9UtzRNX(ts*d+yR|r0&-vB9@gK^
zSV&+i;8y%9Z6F?$_NAtcZ`juGRa>;`Osw0y_@57=Ogg*CRY+L1CEKli%Ox&tRS4;z
z=Xz*)4*Om%MxcUG`if|d)3fxm&XOc~X5D6q&gpv`?eagCk>!ZH*5NENO0xi}snr&B
zt^B4)tAs4b*JcNGU`hsovP&z)fG5+@@PqZ-?gyrliyQJVb?BklUy-C_)W;6W&pwIO
zGJ<u(!-G@l<zP|W*pM<z4Eok)VhDK#vdNacHZ8rT4zDL=$D8rd-;``8E&%lo2O8}#
za5U=r1ByAjapZ1?4qZMiK~ejS!4f(ul{`bwgM>fP{YI-2Q4fhCQrG&xd7KbBYq08d
zK~Y3_;{p&bg{~8w<65atf~V(!2keVzRjZ9_jceDN-akN^`TEtf%-V=?8kPNA%qgVH
zOaiPWO{%z!b%RKR*C^koV9IF;M12`}9OjFwjNb~nzh2`Nx-lGE*v9^M;5C##!W{bw
z2jB=x-W*Ux3vgI%YLv}YTp<OuSqrPUVlz(Bi*4iMF{-LG9d39I=f)cB5*><vNw8lf
z<@3R^HaH?8ZK-UU5e2{ZQ~>s!b_4L-lZ)Ag0azcoP4)+$W|u_wduE*>=~B%k)Y&HU
z{UyORC|8@N31nm#8$wxNu=M?^Z3v`MtBX_QI)O&9IQgY9FnG^aR(XH>AKXi_e~dQ!
zmU=3+_Qs1MA%RNrXEr;7wPD4or0l{bsceW#M;#=nq6|(~!_JoSj;8=^5@CQ#Y$HZQ
zgAFJkdYe2k0Ui4Q7l~QTRCCl<GEcE1Sw|W%opNqT7ro;?I7k{5*TDURirTXIMYmjY
zv3vf(hNk-b?shNWT0L9pS4z0`aD<S^Go`<PbCjO99DDQlCVJV$f|p!leR?tYLi}=1
z-vPDVpW__jVw%43D*G0GbCe>3BDHgWL%TSwJ$2gIH=rv=6S3HW7+?!#j8?nI9l%>F
ze7OFt(O%d3gmE(fqE@2}fcI)uenJu(*AC4f5mM>VEecNpv9C~IijJw9C5PK2ds@SQ
zk_5YbI367AB;pos51m!xjtA9J_`r<*HXPQRG=8t>gNUkB=IkdMeuhSz=n-*9+x^$G
z$@(9dGA1S_mjCUuIfbeF3wgQQ>ItyT*TxQoM*_(_6Lxf*=CIgeDV=sYPB>FQA#Es;
z$d<@CkY3o?)}@g^Mxt%$mM*0S0;EBU{Mtr^X9I-q=KHX;e?FJGWs%UVD27A35I7@<
zVL4F5p;7DWP{2y5f0)pCxSiVR{y503L0@6l%helC3$^|6aCZKBf1jG_pF%&8MEM5v
zMO(2&U#Z#2`DuMUUQ#V8c`Z!4(-?R{>xw`H%C;9rG_xQ&;NZY{$=ogf?P%~gpW46t
zO85KgP5@Mlfld^bWEzhq*v&iOTx{gW?k}0P%(aT%_kKxFPdabZE&;q-wQQ;19+t)i
zX@7*i4~YzgFpA=1D7xB(D)4?T6=5C_4xTE+1o{_ku2~Z`d1SY+k~C40E9}552$rr`
zpHaZQtbL<kVw#@yWUt3uqrV@X=UB6;V~Vbm_3ntgM_{h$!qhjxN4%mCf2>^3t7#U1
zX`ku>c<L|@@zDr`@KLRspKMy3g2$`&wgcN3m2ujh4~SJ|_~dh-<9!5p_~xRdrUEJV
z?-5fyj+{fvAb9=9k-hZHrB{Ivz6OUtT3?wH0^|Tl*%KOFM|Gc~1Zagy^DC(&Km9GY
zC}`nT^Ul0@16i5wLW=HNpEbL8D*Hx3F{b2M*j)@jxuF7lZbT?%4NJae0ff^zF6EX9
z=G?;bg`<JMTI=LQ!8OjB?sBrV`Z0^>P}A%|$NfgtG&D6-Uu)KF*vO=xe531{I5`;I
z*h7J}eAhIEg7F?cIWw=%-k}(v2Y}}`qum{%`d<G$R7>K4-wey9DSTtIlc%SbS3K}w
z1loIZ6+6f#EJTXQmeoX6!5S1s1AVHuIveXhV_L@jcxbBN6On{dO?K6#&6Z2$#fJt^
z(Be6x(}6q!)`hVeOk|Y=zrJ(ykDUQ?fRNx@_}D*Yy@x@;9qe$4H|l&fk+(~Fq6mQn
zVIV)M@n+u8QblekE+3Aw)z>Me=D{2Z-&f}5ecuZ3_QZ|=vRBh)O<RCQlNHUa))~OP
zvB9GAy@|<o<B=Dq{&K6?5wo{R@~tDAiO;1-12P%!BXgIu79<?E;$uO=NC^JL+7bwA
z&Eknss;~K@+)#1ya~YN;1i5z%JV$_`*r2g#D&V12@Ilm(FI`RN7=sJ5NSM4{i(hDl
z$llugU;@Y|xSrFQrhInC2yD1p2Cj0FKr)O{#qwuU%rna2Ksh)#KrbL%{V|=6lTJ{9
z2r|MXEqTyTdSp-!FkR`2hgb#%%8NsMmE*C?tXLyN7}&dWIr@F*$9XQ28cv}94mGto
zqRNESyzL#}1$|U~=MYuy7T}7m=xolbt3FC>&HBS-luL{+A1&9=KHxFA{;0gGXeIQn
zXl2-UpS!B2n;DDN{;9^@w7wJ2!;F_s<i$`8te_(>C5eaIdFGC@7Yw~(o})99V{?9D
zfMCwa7#3)k#FP(`|7n?7RhmSC6DJTn=^uj2E1^wA-KIJO?65bP?Tb_%grH5|8z1?!
zCm}PiC%W{92tXsVLXe&!nc+tCL0&9g9r4-|6^<7On&Shp(*{7kBIqX;MIGfSHml`(
z>sQfSVR#YZVe~GDxq{ADYNYoPqe!4mArWuX3Fr%!W^o_A=9TrgI3^>9Na<|__TS_Y
z4vSZ4c@r-I+cR%(DIzJg9nr5!c<pJ|!yG_DeKQ*`oD?VPHINXvdTySvEzhYABgeik
z!-#M2h!KJyr;-YxGLmD<D0ZL`#$eJ3TDe$T5gJMm0Y3GR!$uhl0kISrj=n(Loce=_
z>9$6UFa*yB0*zNKxqtIfFm26&Ob2t@Cu9bhgc#NbgThx_BQW8JLX{$Pm<FI51+pwM
zEkSko%PXx?V>;4ROlHf?^WY4ic0V-Jela-#IWy^Do`YebPB=(hmSrwg1<M)+5V9vf
zMsd~0Nw9yN#L4}sVKS%O4L>jYSFo$N&s;qFBiU!WVAXRh{tzjkNf_2yyJPnv`4w{N
zyK3|LoVRsLRPcosd@=8|=A-Ns^xo3LKh}$3D9q(i`P9L^vhfNFoA>2g<Fdi(@O;qv
zCXa<-lqeS(GK)>#<H0{L!)8cK=ukqzaZ!CkZZO%ZqJfEtrmA{vr4`!n>rPL-YM?D5
z^#nKh>qjZDWQ9`{EkDJiouayy9eL|d(4u)UUH!pTdcHm_5gG4NX+Sp%QJ5ZIeds$)
z9gP|UaC3<7-7hLjH-HD+?A~Qmtt;C33iPFt6yz*?AQ%atc3)Ldr5<>LU!-D34KtV8
zY8XO>bP0RF|K40!6BVMHe5CNCN{uaOU|>8EM2g^fK*(bDG;mtyNYp@W>eZ@SF6fd3
zHnRdlv@9dVxCj$Q-ZX_Q!Lr_cUS<X)tun@jtpKJ~mT_F|oQq&_OWkA}0BStwt!QII
zs3&d{rQyFEExZjBjgE3;-5!i!6k%kxB><OiDZirO50EL}_BTW`kvIL>d^)e{2`(Cp
zAll!2X(gN14T);bzOLhHYhK0^_F2~x)~RPwTdM#KJ1g&`cih#tol9UNWkdIA7p=vu
zuA@4=JSzz3CZ|FW(=QkbL(e~W8FRUIR3%z>VlK;|hoeZ1(rbTw938MqvogO6Ad}V{
zj+U1U=@maRe1O^ush?nZ&@dRC*l;u+ggu`a(0+;qN+6&)TRtFzM8m|sJ<&W3fcz9c
zH??7f0$_#uv9IkJZ9=VLo07lJ1NG0Gaw*seESnImh^8km$mN@{Gjzfh8r%rgg^sqq
zChHrncY+sS=g@a30A*F0?fNtX=ltIiF^CdrpxUwRZjvONimR`!H14a=PRkCOzx66<
zHgi_4$=uu)_J-43<<uB=u-Ccwvi4RJ*bvopE3N+Oa3f|(c<P}!4&3CBy<lSuuD)1Z
zA(0bHMs*4@550GgwjG47I9i<xFZ)lr>_BGI3$mv?&w-1e`)YENGX|Rwn0q~dy<Bsh
z|2YBu2j-5EiTQtT5YqU+H>0P&o6&(#2}$#oUN(v54)8)*T@+C)kO<KYphpRkHE7%4
zepxuyB%|_}12<Tji0h%EkgS!F^u5`|bnZoW_E!TYj-LA<<OBu#BN@{eg_uW_{2V8;
zI>!U3Ggw=4{OKAD*p}ux0)hxnf}%C&yMv8fvecCU#MH$jK+Kq;g6pHDrH{y$)Pu+!
zdxQIn*RR)XME1O804bqd>9fZXhije|CvTdSBjZ~R{k#;eNyGpJFyUQfxn$-kun@YW
zpfCS{AR<CDx$^9_-4mmKo3EoFv+fZ5N(9QG4%YAi%rlSxhNSC>`gn@|u^1zYxX!hK
z%m#aC!0IFr1J4rQgI(aKE@Co(j6w0pfNU(El7CzqIoXLt%elXw?b;ft?TbsD6`zCJ
z*$*}hh~2k-{g6^#CY+!S-_Etud||xEcm+)3PX8t>=@XbpU<!62NV@#oXilkUos+Hu
zn6&X?kKod;UBwF*qq7=c%O=!C*0C#7g~0aYdlhU$9kwteKh%6<OZ7+}XzCI3c!*5k
z2NQz@)CRe>F0yxnKDKxq#Ig{_`fW6-=N^0PM<bac=lRQu*6yt)DD!G@^S2%@9HC@t
zf|!NewU0>RZwXP{M)C^G^6pylA2*Bsz=-ESY<lF|2dt%g%`^JU^x|gGR?=p4!EpIC
zu_h}6P$K9$|2=e@7NY1DxV{7M`jfY*cTOQbBpH0bvP2y=Wm^4U$Z6X}VMlZCWUa2Q
z*dArsHS?#eR*0&ALb%C-k$t3vv*{}(G=SmSl@#SoCGf5vdh^2s78$OYz3@iOIiW;C
zO9WLeYRlBxh*k+5YYan4f@PIG5u*6jhH>KpN5^3DVy)#@pnr5BK>^Vu>9R36=91yZ
zL8$00ypqc1i+imvI4zS%(FXvB>CX^vZU&hyjEFMnVkCLD7ORQ2&!hT?+Gia)4*I_7
zTfpit2&xz;@zNDKi~>AiK-x25GYMc2MW(lAG~b&}H4|Mxjy8r%7m|!0OlifG*bb%W
z$5jGZ$0Vu-45!PT`gB1nP)g)|Hs7)sv;qs^A=Q=3Eu+_Rk#XlKmkb#IWkAH3c1}eD
zjxWVdfEdaRknVVmaMdJuMB|fD<qzf!geV-kAZ?*jbF-!!GWZxOu>DDtUzR_Ui|h;_
zsiy9*MQnO!!+Gk_52ONG>g_MVZ|qz3Q#<6~)7>YNoQX?Q;h0WLZvH^x)?%k9Qh;*Z
zI7X=LWI@7)h~yk4%~g(-Uru>P;ItEdY*(5loJ;#D?oh%LkPG?z*+|Z*BCYcl(G_zy
zxz<*Wi<(ClgvGoo;sRUrLqe?p##SkFbZs7UXWopHIj&*Xo$_t+hLv0It}dt#MNvqA
zs7^4N#p;vY4gEGKO)|TLt8f_8ehTeaVDMYkHS7_zAO2*N6R%00eAmzBF|VEQp}FX=
z2lo=Tr133z`+l$fe6QxQN8}&ssP`_gS=vI!OXc9jw;}CU9(}QGFU2w`o-V9%Kbw_>
zt*iHBJ?ebOD*oO559HDU=^gV{37KM94_O=W^<Bs96+<(}=ynlW7qimoS4QJIPiXYv
zkr<4O=|gTPE<2_0xui&hReY;W{I#R4%t&j=cB`+LpA{b2`jNSUh(6$l9en*yo;BNl
zAg`Dh|NnG<ZrL6-BmSR~g*M&?3B_F4cRR}KDCoG3MHfGy>HJ;GGgkV9R!s|U`-o@T
zn>gl0srz*>FN^q8Qr6G3J+E(5_r`EHC!hPPrLhaw0CmNo3ZAjT%1nu4#gyZyQcb6(
znCgQf<A8>R2AY$*%!h}_mJ?gKNS1DopS{mVnlZP}me?KNpF7}bdf(6ID}J9=H&6Gs
z?}N~Z1$p(IE}zu#vnz{Xw-jJU-?mAnmb0Klp7`IoC(Kw85@t0LqHz$2IJG%d=#QJn
zYW8Me)3&H3kF5?(ENT)M?xiPomz8tY`4!8&VpXGy+;|QRlBMRYYjD{??dV0d${4z<
zWr=oW?!xvAmvwAIFmi#%u!ah%z|lPJoH`^i`s#xpDAtFg<;4~oD#IpjRFkUsm?(q!
z8(W+=^6A9&Q=8tduRO2?_>yb;m4Cwn`UsRFUe1-CT>KqP_%uG%;U-2Mu#ko!+J;|4
z`xUsGP1lUdPbZ36k}0H+-6lI1ncl$QRB#m+0<|`$f>A1Ha#3*Vx10=X9^Wnv%Cy;c
z9<-Hi|F*EX*IPnXt|l${{WLvEG;5h`aBYbzdnYxAq`6)a8c}HWNZlbZqf~n8_lCAz
zl#OVX)GEcU@X9hKb?`KfQEMlW)xv=^|F}z<;B92DrOC0rI&-?U8L+K@Bg}T`Gi_$%
zOuIOZiHQ0D%X%!$S2g{WrIBgbEbj2(DG#xK^6?U9W3(UEoF;9%@V6!lG#l?_linY;
ziE9MIGi@D-{Yrt;z(Ixkg<(^GRANr*wrbz3_YOs#UN^RnXqo{_+BT^p>i{Uq<588|
zPbB!4=6zDwYYlEFqni-0-P-d}xeKC6Ws^odv4uptQ5Fd(ijtC)>$H=gdP$=IDamu1
z&0ohBHI=Ir$Y^r$3(I|A<OY@6{h6M5G+n4L(;!u<Pohtc#)E|?*$FkvTy7)aafkNN
z9eqVjAiS3_JI&L^_XKAJ&6ZyfN~o(CjnlS%naXfOlxJunOE?!#ut&T;0P5D&TA(%N
z)OwMFIWX_r-L>E>vB_)SGzsh-KbgSRg&#g)<KmNsBdCg_nE^d>$}zh5PzIDt;)Q{U
zvePt?i2<}3<sP95%za<6HnL=9L4hs%8U``@ji-wanFdN@ivr0TO{zlWE;PK3_|$vp
zTXrImq1$wY=Hf;|X-estFj6_)K8sbd%6eA?IB0V}q0xj}_99)NR+vA#@4zp}7CqeS
z*TbnGT}Fljkr65a3=70k8+w;^6ek<^X~ruRYo1|qPmNcAkzWMNyIrcCG&MAAX&Q0;
zGtURBWB15;@K)4d&8Y8qX+6WuH)n|AWcVRGc5?!wAOXw}r{&jJcxm?Rw0W@D;=}46
z>t84W=3FaEkx0jRm7p{YMoP`LZA~V;T&VzW!M-h2>!S8U;xQ@y+jL`rd<AA+XB)rI
zatPtFbg`9PASHv75+a4u$%!Wy=&BFkudRzMuOb^rh$et=hE)e3GdsB<wuEI*K$1c-
zQE=3{Arecgo7Ho>xKWv?CpTOw*b9LH&;_X}=IptD%H*0}9VIeutR}*ZyOs6%RV*}3
z@KOcG(FHVoj45a{O2T<^=WDlGfc`niyj|wfq{e`};}zX5kr1q&W1kYi1k~y$F)q2^
zw~yb)AeqA)U;N`cK;j4J=D+S2X0O)j3g$K`()A`w&-Miw`cr;~SArtZR5h8X`4rH6
z6lsB#v9>kC@qoEDGdY#$CV@sHdH8w$QHqRX%$fqAk(LeFui)4Ws}cQ9OL3NRY;-Pl
z4>He#)T7*_i^<yTU(ME(=COs1NAeFuB_S9MNqL&yli4DUj1|;s^BrZ>G`gG~La3H%
z0;vK5fm~@^Gm+=7NzDuhctekWJAe6GwY%kp@SO&r&+1wCi{wi_87>ti96h3+#Bk0*
zSB+J*qO;L)97H#y4KRRJcoYIraM8x%2bkRo))||VME(cpdbZ^=mc<t>A$MAhd7HYc
z2X5Lth-N_a={kM4ZAF%HG%u<ChH%R;u1HE~wb5U}KNORAr6;=;5q=?Okc^i}whHKB
z#y9r!%GX+?H2zU+2VhbTdywdGd}#QBtAhFh95jKixm$>{mL0ZW4c0;ndlec8?*A@K
z!3I?U#PmaqV5=@z6D;R@sxxlk{dFH7LG(lT0MP3(;l=jA3uy%4>T~x9pY!fQKLIh-
zjl+M;2BaTL8jJ%HJdORCEw>}6Y?gQvYFz`zBAy@7#5>xVU3Ly!ZW51P%|tN2qx{0v
zErGWec3;wPIxQ!VvTnaAD?2D8^UG<KEkmoL+phV%`$;~{aMV_E`91d)FCHvq-2C1L
z_l3^MclfiZn9H>5wZiUkI&k{4>5qKdT~UguH^x3E&pE|NJ1aG$T<z2qm;~JW@9A`4
z7HHhqoZ-yP2{tRZxfUy~`JAxyd$xyGD(j?Ag$QWs8~hF_p#zcvUp|O$f~~ZJePNgQ
zWBF>0+hDYT&cVBn<JLYvKXpKPBII4$?v@Vq>iL0gB^_(yboO(Au#_o8dZ4fOCQZ0g
zy*MUF@G=zX@CL+$NN6gR7CBU=vfrE)fetSGLQyyaBfAB(Ik7;=R#|!rPKMmS=n&6D
z&{lFuWuPjCtJ7v8U;0x)CI(Msw|n#^wE*a32UJTk??5Hsf74@1yWY&>L&9sD-VRB!
zG*54->3r}Q4|?=Iy_)Ats*+eTdKG;$dc}PMk-p@pB{P5RoJ9hQqAjaoZ>RnWcYmaR
zKGhsHuN(`SygkDH0S&7iMN|L!s7=Dxd|Oxe6BVcM4Q<Y>KKEbB?RUBR|Ek#d6>k4)
zN=}!iwBv>tQul@Wu7v8~SSO(IL}=B-vqxpY`k;)p^;-2mLP)6tP{0rX%CXR|tFs(%
z2){UEQf<#TQ-;`_oa>w#{Gu^gp+r}7UtJxqAGHXQ|G|@MS|Q~lld|S3kw!Y-+mj`A
ze|Elpyx3w+c}{4gAWo4Y*jN7fxf=1M+5U9kJc*cg6QetkT3E|$vEAwM`8j@n>iHbp
zU%7Z$xry5Zg@uWd%O&2@2{WEnPB>au*4so&E}h**>t4WV?rvZ*u*5WdPNn(v8q8Nk
zG~u^3Tv@jQPax;jG}*Z-WYnq*a;N#y^J{(7y?Z%xt|U!&d-;Ex_1O0sPlu^+*$@|g
z9MTv>9ZK1|64EIXaWn=%m~WvWAkbi5O7l%)1`F?O-i)s|H@8EsR94c5VmElC?FJ%_
z8!*daSjv`QD$+=!X)lWF0!9at7d8;EpITiwE52WYkt9!_Uzrb0L*JTT2p@=N7%r{1
z1J&KC<e=qfIOaZ?ikNFQk|i`g;nBw_B$&02G%N=xpSiOoFQjJCTC3Xrqx@P~XS>>_
zgC7mRXEf2=F;ExrjN9tU+gSFONn6KxuC==1KiB=+^8B~%y5?$bZBYW}DCDvpuTsiS
z&_gbA2kzN?yuh+6f$hl$dr|T}i8?uI<4g8_b&>w&d$*cBGIEEG3l20tiWEsJ#ca&d
z)Biw-TI2VqBjwK<>ycze=KfG5dXHl8k)Xs1WTtF)3ucsgc!~Tx!LIj{SCmn@wiq09
z@AL`}hl9V4RLezz|E>0B=HK!6^?hSb$PFQM)6CXKox3d}4Pz)>zwotBZ@|=wkulSL
zz-PfN4}{%Qs1-d$6A#s~neC}$=<bV^Pr6SAWp2MOBtuxwtPsp}P=h-7d+-P8F~-z(
zsvbsV8C0q4lj$lNGo<gfbhg@TA_+O!(Ko%DgIqvNBH^4TV5@>J>4I<@FC(nM1!f*D
z$QymuAc9=rjpwF}e*}u0XQ~I$RkOM`y!fPdqSXII82Z)d`R_;5!1h+R#)hi)O{Q6&
zx;Y&4q%=#M5UjXEfCD|+rJJ*1P5D8Rg*SZdlq(R7IN0XuKTa?tgP)cDe^4q|4O__u
zea!+rrzu2Lh9?SZE>V_3iUopF+erOBogK`LRF*f(EW}GWcJq-HHhodRriIoj(<~du
zq4~9#e&#9pndCSq$4U3$j;y~kV{|9{sfTt_hpD*%_Ch1xTy`jr2jeD+>ddCj$HGPK
zpHP#QVYMlV7Zi>?5s^)g7-X*tBSEK_(CP64$Oc(O*4@EagXkBguTvBCsw&T~skohm
z1xC*X+TDcbO2L_z6AX5jD>%aw+eI}y35Mon5pwB5c2j~KS_Y#`4n{*%2^-g^ZdS8;
zy(QwLznxEnt;fROQ{x-<NjvN9+1NEJ(=%{8Aa{ozHwuWL_&l<E$<Ve7KTi3D`c}t=
z0I|KV7Vr4QkYI=WBO^5m7aV{&X9-i+5qPpYeJw}JAboop4FL#*RwC)U#w`YJ$(3{*
zT)eW?q+{k`92t%h2JYu$24+!c9HzR^h12T*8hD=^0MEw?LSfT;M)ugL!4&kDv=hKV
zx+Sz3eUfx3m1=1Ah9IOm_C$_HZ=g4*lN1=8yn@memHn3=eecn>8Dv=R5W1Wk`E)X$
zcB@zCC1LiN7KKKN-cE3t4@tK&FpxnC8O%8hQX*gm$d>>!1n!TtH3*DHFxhh?&Mb{|
zFPy>0l!}y&)F=iBG?SQPD#z+|rQQ(8y-+-nfTildn9OPOaVfAV2#ZiJu$kGAv4YR>
z_mCB_kWv0<<%@|(1Grz*QvWtl=OpeTRuTi2NCKr_Qh4{bK6FZN8BI6X{u~Xni@Sw^
zYIXeZ452Hx>;pj@J4oEVTadMa<cuVg!PD*<AZq%Teh-!hl!%lPXK(WxdfRQbXy?jt
zUv&KFY2?O*@!nHjAS)pXKA>ynDqwCZboG8R#00$ROBw#KUj+W6Tf}GZe`N=h$#Rz5
ztJns|*jQRx7xxyLis_RGo}Iy<Oyta0feGU?DwI)%I1JSumUE~ku=tr{Mlis@jAV%V
zLWU8x1cIL+tlKfHG}6WExDZg`9QMu#R1-2iEPy0wrgFI=HV~LamUwj{uX-=5Mqsu-
zOER`lcYv&eY)!7;My2N<3#H(_n%0!dN))u4r{$=mMSbZwtuUN7%Am}i?z?9OTLTA%
zP<ZQAjDk%Uu7(Vuc|d?sO$>o?(igN{t1O2$v`apVCzzp-40|0&xFIW_?e^7{(^cqu
zx<>@d*1g282E9>+uYST1{#3E2E{PdnalY^M_h2kS`a0klo#G5^DIHYMbJZQ>7(!|V
zFCJou@y0aE3MXg%43F3aaN|n=b1G!1cuz2E_(F52p#&5jr*>99;CV}m*8uR^$MRr(
za}of~USdfUr}Ueuaz*mIE-KfE^`}l?>K*$a2UND~U$YzdT5u~TQPhEJY28#)fR*U-
zz3uYh0?38y28YhD9$f(3k#(oY*c0lwvmSkQPL2MX3z1b4@?mYxPKLnbdQ3PT_7ofx
z41);8ft|~3fU9oGeEwIU%sjPj>{2Do0W%!mQ8Gj3KnLv=Lba6v`FUeG){P@6J#;{0
zK?=Xz0+gz0he<|10@>(<Q%9;UArK%@){R9*+bwBbc^hTNgp1@6Da&K=$f1ZjB&csf
ziqg#Xs@sA$@J}@Y3aUw+b6X9z51GtS!XX3m&|{$im6}h(Tn=5^5VKLdc{f$3E12rP
zuf_X5<~5v@xXXY2GjRT2@wsLy95uUe0)gG<=Ky`uTn`aTeW7V%_*d=J$U}3=ghgg(
z2Ez7@r4iyxrF7>wC{bv-SH%oVt(_cpN_i&fx#HtrwL&M^dUg5lc1E#xQWK&0B@m@~
z<ghxB0(l4q9l3=HhfE}0b8#WdV9SY+oD8iW;+>5EL0m5{!rTZXp{<gMEZ;gO46d&W
z0OD(cHkE!Demj<)cX!OP%M78U;PGZ-+1hQvyW-R!w@hz#caAiVFE`OeN*M8RB*z1-
zo^G-XVt{|ZJIf<JN1R_N{H@bVzF!!Lgjp#fRoCU?{>mW2&VA#oE`HLK`9aKm>4))&
zWm%5bklho{V%>OQX2RJjjgAQ5mdDXY+qV<xSP5eH00&25+WzNc_FrZ^R;K@T`Pq`j
z?_~CyHo8#HV=u4M?}TLe#TRyxakl>Zmp;8X<{xh!sT`pqx;|~RzOxPb+Z!V*-8?U|
z5h)QExn=9xqe~+*OG<RQbULzj`btU)hG~VOZ0_RUDx^dq6C~0A%7xDyO*xQQu43%P
zmMLQR`FQxYuy9i2>*;-O$z+lc<;&-N_xbdj#67>0FEvYaZD1dv2faB*K|(I?HU5_{
zF$xtOChlLN=0_b~LesxaLC&U4Cg-a;^hUI;@de#dFHybFDiyi;rK!VFnDpzDkwR4?
zv8?jJ06;S^(bBLuYn(c;@<|?2-H#qRauq%=W$D*Z0FY<2W>}#DFrmgbz?{?C)+YqL
z?C%c)KrDn&3}`D^G}sr)k6A^op@s4q5D8QP2{)9xq(wCD6{Q}7T_sIR9}S=9`lL54
z<QGHfBo7N93ncar0!tIhd$V-<j>!Y{@(cvXoI5Q^(IJdZUc61oZ<wRc5j5CjO)J20
zLX;`DCzB5m60%-ydlF)sW^>>XN-`oYD2rho2+$(OSnj+-mtZbJA*%A9Q|>MQO(epy
z+*XpUf+FoETR8j_MQ^Ib_eV8d<_ER`1>FKcBcjWqkzg7|`qkd!p$cyzHTdDunr}Ci
z)rN4Lg>{h{&*n|NG#{$5(&y)Z+_w45gfcK->-oSZK&=`IVI;r}D(F*1eHF`cV-heA
zf+2ymu}&gCW&)~WAscdor0B450t!XN?xnJO0H4dJ`^_68s6mSA^C0_XOK{3n7E@na
zrYDWXd-mLYnf|F-R#%;ef(c25tTGrHd0Fst4~w6T?wwQH_?MI7W2rUO{zn4stdolE
zw*(TduU<A(5E#NLG#y0)l5eu~Pm3Y}!>qN(w$MX?KzA7@+I2zgkZIGHH;Iebkvf?@
zy7lTUh5?6?1Ovr(*p;Ol&@Do9<pHBY0VdMy-{3Gj2c*%CnG#CYI1%7W(5onAa`y%q
z9PyC;AxN^QCl2>a^g!<tSGxK*)Ty=B(4G~0mnF2hVL;jy(Qb(c>6$!@$p(L+EwXpQ
zuR3*yj9WSzt6=^7>0DXvR0L%=5rV-vtEFj7vFL?QuC=5md<Pxt{mxDbHtL^$PBWrF
zIwtN{3=&;GrzzQ$E*%djDLQ3UC7J#I4)bWI=jEpkfuv26tE5k1v&^+Ytd!N@?19w*
z4S}zeIji`J7z!E@{}F<ehpUh6rh#BO(#uJ!P3|4gv0hB})Fg0>3|v8&(r5sJ>n|s~
zNak*?#2B|Q!?X~yf5m+~^~~!3Xf&J2iCfCA-RIX|z2(&wRa8FA@o%g@x4;EN&*6e0
z`lsU}(*{ll^go@I%C2kvKveKQM|8n7_nPGu?!YFxy<Tx`7u|i(P{wBYoe<#CCwg78
zzY#Y}dw{|AF<cR}yoays7@T<|0?DW!g8&ugJHDyCRgF$K0bVg}+AA7s+ArMmO)XGG
zGtWF2#v8Bd%GHLaC6)bb>J)VFWp{JgsoOEw$3P-T_Rbcy@uprf>BI==00VNc1VUIN
zIRTo@`4fC~K@U4(ND}Jb0#8=6gHo4%hm4l_-Az!WEp5TZOQscmC7U6t+uLO~6tUvE
zgN=N-%zs}NcI>biEB6&(e{fpM6?8(;$DSPA-2AJ*fhts}q-Dtitv&j#=DBl$G#>pj
zKfeT{E@hRN7ylYCu7JSI<?#^^2nz9xdj{Nt8*(<JZmRkEJj`DBxdVKc#5m{8;9tzR
z$RveP>YL_=pO%2{=i2VIP$RX$=w_yW4vTsuM1dwckG4TN1<zk;8nAEvGYlS-c%$~g
zSH3@x$ddOY4Mjxo+M|SE;?Wkt+k~>2U-T`tS&{iAce9%?RG^(bk+Yt*5hK;<+84o0
z(;iaVJp*za8=Kv84YCnlC|;BiYV)9Xy+<~2><bTWwtEWL@wekK1?t1sZ{*5i(|)D?
z^CzUOWZxI4u@m`0Fo<M8WpaKekwzD7X#bDXV+a2>Vm`dh-iGAH(dI_sSaT+xv!zdF
zh_3iJIxJ5Y`J}EdFxMG>(SN-p?Ee+4Vq^SY2dkDe^!^XN;iYtT<B-)<G6IFv3AYwi
za)TE{dzTl+T5*;w8zT8gMd{mlmG2{lM~V%V+hQJV;}6NXQ|HH>n`<X=v>Y*37tR+C
zE-x>VWKtr}kp+vcm?V~DJQ9^Li>`?6zKH38>w;bm9%F`WT^$WEB|5>NeV5zQLe0WJ
z;$4t5)<aYjIaBf7-rU>;dJTN6-lso*Ne_MxpO3{*bQw6q>@>uYKq+D*lH`1rXo5*l
zI8R2kz|`ZUUs2Ww^p+*J;a!!^3_pQ=TXp=hi<N9;WpMQfO(gmdd4s^YWfudf0~%~X
zL~!}KVYRdXok+c6BU8JOTAY)1nXWVdR+TvE*zOeo{iI3$h9N8y650OKd^agjBZy8Y
zuBJcL_2@Beac8~Xm6_t3g%R3;48}^krT$Q5j0N#{`XRtDWUO;rwOqWHYNHwU6Wl9U
z(<m|#r}!fGBv^;*HH4|+<c9MW!lqaPHJu5I*GlPJw;nk?KIerU*E`2KA$brJnte6I
z0aZqgSV~S2(rZIlX_7*^cvE;v;SOHpd<O+74ItOd9wBlTih`UMjy+icj1H%)7Xf5(
z7!qYj*hcQ`6Xrw<*=>@s`?ly}?soUfMGP4Ys3DxKoPMFbk7AQQ8^PD@8d2!qy(o!g
zVd@q25}#hB9M3M4W(a&t;V!_VLH{D*MgACa%Hq4bhF>G5u}Ccq%V=?(4|me{@hk3l
zkGr4OJHgnB9Ft7GNb%le4_+5BZV(74=vpsT*UE;2Sda|<fi-6EAdiNGwG$_UU!cLl
z$;aQIN3YGqZ6(Q?y*@nKM{lgmApyOHSE%*tjr|wd4v0psY>up`78wJB-IoRX*cWZC
zFm$Ktzg<lK$}_pUgpPv*zrBLgK?qYfyTsdh_!pLq^?X0fO1|C`dPp3>(r@*~vSjVs
z*H=k0iUK6ky@~k;*3YX^$+GNF@f_y{fA!7F<DY<uS8P305>DOq+jpW393K7Vg$bMR
zQ|9qmux$A6Ums7*eJ%hUKlaNMUZ=L^+lPG|?(M}G$D|l|pxYWnCpD|Dq=kFOM?RI6
zFUM3~X57YA>}X2^r<WPy^7%J;4E2>IptSkQsH%!W$Hj6(lUchcS0gaH#Dhb{4V&-n
zi1u`%R0}$*Mv5b&h+sQkSTUhjO6jaxEx>o4Yh6&a7Upq8^|~2B%rOs2=*ER9?7!6k
z{kY2@EEy-++@<!j`peR*dK@n08lKRrmCFyMmEfG@k~U~#A<#XCya{vx*Fu>DoNh2?
z>mK)1HVf=qo~(n2Tu&##H`9+Z{cUEQ<4aR<{s|@xw5fI<mp=itR?=gRD-vwgigGP7
zhu<7VKx-`-p6qC|v-t^KLzmwZKsiEt{a_C7d}+^oaHG``^&(cxfhgYPk@JQ-O&&wL
z0%+Uzmjz0on1zFV4!ZsEF~@%LrL}<3I;7v-0A+bt1LNEBuMD8~I-VBH@~!4kxpQZ2
zBW^r_8IchP2j%55u!7gJFLy%f)^ljt7hqsP%{knVdv`~+DtSk)>A>4zp54!>c&{^}
z%rG`5V4j)?i<vEI02?{v-iW(Qw{TuO<k~Wx`+#mm=V?{}JERcf(A#%$+8dXS$@)QN
zeX@xOcS`8a7P{i$Zi+Nlq_8}frh7jR?HU~{Fh-tJMsMpj_g>a1%f@UolhDddIrfP)
z$(OC7n|?zOt$(ktb2I)7i?#`bFngbcGaFUtZjbjdNL@n^)Ad?od+DXu^Yrdr77Y16
zC<cp%oy&tXdDoY<vP`g{sB+kbCK{R^V_63**w0R~nu1`|Jch<tR>UZ?4@6Rg877qQ
z(Q^ix-64^7a8uY?V;nf+!?@_FuNC=g^X8M}H@8ma#hILfEKGZ@INNKfv~RjRDtoa;
zceOt7k+&en`dMO(?+K#LJ5mv$I%fU){(<?39DH-nK^-!$0o8}oj+A%(5k|3_PjaG5
zHarYk8oJF^JV_wuLI)DFl>m5>y$3I~@`9AAZh&fDXfBykK&ND=T2h<f<0@n0hAx%J
z^FG(h?c!!I79pkDuo;u74|x{04RmJpuc3sDsOQeQZ2YE`U>$aRRt6MZsjiur+~Bfy
zy|vyOjRa@Q!+r4K0s>k+@#VyS0DjsEg8u7m;`k4y0uvM4|N8%V>bD?sV;J?f;<=Z2
zDdVDxzB!$z$obNq`G76KluMcXz+Xb5KL`W`DaqLNSLfK$1Jof%49l&L1%l|mT;_am
z?$BAdWX@eqt>xBC|LHbeG)>W%a&~b}gv_QahfJpG&@#O-;~|eqd}y0@a&dRIIlRnt
zt?6!adwR~b-Ra`Ybrd!8?RkA(+UU{AZt!|;iJW-AulatzZ$cU1EhbX2sD0FiGQpk0
zj8+l=DFHillyfXdnmkTGf;;=Hh&3=0w+v&NOr)A@u$m5QRNUr@vma&_3m?UFbP6%2
zZU9_ciH|x5G#ZAO@??y(A@@IWw;>izH%nk314hg?CYPLs9o$*?vQNM#ji`v(d1B$8
zx&#?mG|W0=GkCEW*GMpLJ!-6D8FMzQ67GNKlAo{$5qqEvk#y?SP=Q%P6b}L%uj8K5
zAkzn(U`l5$XdJ6vCTMS`)_ts7t9UQhQFWDVfZ17M<7s(0DtPx-`tuISS9#7bmy+pR
zWMQ+f(6ZJ{du~$4X=d77#g}Y9@nNW3C0(^aga~rtc|29p{_ouIh_qx-Xq=*cx~RD}
z4r1R%eaf5DVz^a+*)UobT~}V??<h#)a)y(vLar-Sc$d`MA5tGJyyK|slSCld;Idxd
z?4VP@y7VKF^E!XDp$g>dcOwMe^Ie@6fCKU1fCUE6X;WpX5YmbJe5&<|@?2143~Je`
zT7Smo%+WGAwRGf4nf84Pp&mfsU>=cGa%lwi^;raVp<fKkYfXiiRa=^N(#w@%sktmm
z^4a!~y?vVEVAP<!;66r6HcvT-&*`vsgJvv6u2~qiF8hRovy4Oh-&ZfCTQkt*tf-QF
zZrTc%2=n@f$muf`_)a=SfHo+UNYjHSYV?1xToz{RQ)Hb%wTWq+_9X0ikjQ-Zw@!6h
zOD+Bo#}I{)^O$qd_~GO+YNrog6qEjb=fUE|gf!}BMCuV`q7qll^kNReiBC4FRtI*-
zIf6Sb%E|EJ%z7G77qP9fpldn=i?~NP5Q0wsM%IihU8SS+3;(2Ar6P~No34uQ465lM
zOe~eN;TDeK#1Q#`7d?9kJaN|JQw5M?oF^IBjpvrM5*Y>g*^@3#O@+D6*E$ku?u3`d
zM*|croL-8JBWoSbG;(}ApI@Fd@CD38)504h44g$it$pBIG!GX2_3uAkD<szq>=6Xd
zU!&^a`QY=Lsv`%U4<hPSMLe~J2j~iWDEO3Dwgf6hI=31Q;GYM;T#L#?X8V9)DHAhS
zz!3<HhpP?xwiMvlPq!lJI?Xai3|m%aTMBBk!@+jGv<VLYMYjNlo?>(r_8?xg_ocQw
z0dR^Fx~IQ=X%(yI8443(NFY|{jye?#RH=KV%?M@cq#dw}z1%F#;zg;{ZG%;o#T_<O
zRSnF!3gP_|UKCjXN55i{Hn<4{+IqFhqjsf>8b8`i2)T$lM3mEI^+m=Zch#KVvfBg^
zyHQWaZqM%Z-AR4kY)L(Xx3f}ddZ+BxWuaDiT&IT*h!XwfF8@V@)4Qu#t-|JGW78n4
zjBeX)r;4NkbQIQR3ZVyuuWX}ZzJmQA0}VuMn#t55h~b?dcV{IDSr1N*8tLno;iJgp
zfGFRGQw*aa-w}0JsxJt7Y?J`YlV`E-e<!#M=VEXg>7r-yjT-HvMI^#^GzyU2yex|;
zhs8T^*@b@6ZWT$ex^nW>8)}VR?2U=HaTHow<I@K5n0iO2_TozE$X|cLmy2`;zgI^W
z`984Fm^2N>!EY)-&>OcJ`~CHIzxicekS6>*2tEd@V2;-N&-X#K$e#$EHe;x!0q^r_
z=TOexf6eWBL-esR1~qLp<P!{T_|WoA=9$rfRtXE5qEh>MYnO;^<L$ur6fs<HWujwj
z<$gQay(2KE0>#oipYT}0sU9ueF08X%p%1g8G$BOViq%2829b4a;ziSnONbg{_sHTD
z(5KADr@w1=j4O<8Q=F&V=+>@GtBLBhV_lo}$bd0#L1Sxi=l*3GNRr90w|0O-pMb>R
zc~3u1YbrSKzoZMWUM~1Y7aooCmc}*vubd(sN%Cr0vv4O?LQ4`IZ^CCp1g*o_u|WtN
zI&F!GkYXkcqQQxCY+8BFZ%#DB(V)}%2X^7A11q3_B9{^g?@ql^*V}?FE2Jp8;KRkk
z2O)>;hi)oIrxGE>)H~U`rhtC{Y7$txns6rUV*5hNRal$&C)?Jz{xEJiJ(|?DN~^bj
z@*W!zf6l!&3?8?MghW*}(ySA04>lgH?*uR+N*y463rPpe%Cx-OFrn``@{NiTlD`CV
zNx7(iql~@0M!Z!5V!n!LyXX05!A%u-_T2vdb+CSX19>Q2NwsDt^YT>jJ-JE_fSpo!
zgZyGIb!sL&&hoLq0v~ScE6(dzgSrNy0l|PPG={X>5R*f#HW*EfISu)?9WE^iX02l%
z+kFoCDmgX{#w%Q0foAszkv45B5Y$-!wg@#q1Yp)f&N9!KTdC*XSKaZZl3uon`~e`m
z-Dc=<HN^1_7(y`R-7v3Hi5CB5FCw<mRnk(OJykmn#^-=@18QMm-=OJvs=}^2Df+_L
z4t;%9Kv~)hgsrR$mYn_xVb`@EH1J8y<=l}rbt1=<+N$1B@SMe8egebgR>ts^-wXak
z!C7^RiA1~Q;T9iC9kE>Uf2-$aPLj?|c6xTXnYPZ-McH6lL=_Rr6dP^F9F}dpb8y~3
zziL%pV1CoZ!Nyn^+;!vTVq-v|^DEprTt)D~kj6@SK<^FCFZkdce<}JASzV{&4~l#B
zr;6ZzUb2N(1svn#LYxW&xzbJD?qP?=6m~E^_}LPqhzU7T{$Bea!wPGkzxg?$xs?&~
z5)WI^`v~Coyu+;I)nnV<^F_h3cv8ML%omQ0G(3U!k8XLrHtFcFVILw=bBA%0U@XF5
z0rU!XFy6}k5&oq%f5Eekb=PU_IV&zkIy@1fkouxurLZ40q>#ZHux$rCI`DS%X|T3<
z^GliW$S16}c@nYxK%YDFsQu@x^dCB&<NsdO6a0SzQmj>F9Wxmax_%pN!ptm1fd)tF
zWDnS~EyxlRgiEgk0oP2S!hUNZf4Zsv)e=s<`Zdm1Z+Cllm>vpXyg5DPx3+H&A&yz(
z1VK_hlSqYyA&(mhMnrr<6a-oOpz2k-W*tTT<(V)?8b%o0YJdCu&TsYJv=Z(MU?|M|
zC!t~2v9YzaxwZZ4R1Mre`@a58gmoR_@q6>^n&8~aIu;6I`*!is$1Y)+aqt2odI+lq
zio@5Q#(>U|*}}a|S9{(NR8=lVc%$N%!<6q!M3f^EM0QhSkW-FNC<rExC=8&e>!rZX
z0mT?wbE*O3TyFOa^>kLkt^Qs-P?Rrp&CaOXPQlt-=<V?FZXF0te}{$l;Eke_KuVe)
zaeG!LFQn3mUN;rMn0hcC&Whuhuz-=|?ELp>y#*2!baF3%54MKY*ocMPn@wp#oQoYy
zGFC0rU={V;`LDXNYwOdPhUV{5w8DJY!*1sN_krRYS8ptoU<fd04Zf}}PS6TXz+d;I
zud)zl%r~MQmSUbs#vWe3cX(Xrb2vCHb!e8!o%Sp2C7|6VoIipb=<^(`nyNw207QkF
zNrZkiHJijK0A|tOMweP3;5<oe#I9jHE3&-ZyU}K<Qe~7m=hb(n6M5QR@x?yX|DZO5
zzs@=8UOf?lLdOn3<;+z8c>_dbb<fVil;yF@q{r^V$Q+a4EQ2@JcR`swvei)VpK;^p
z?ACN4jh9Rx|DY8&0U`e#;P6zpER*lOlhoDO&f|)(7-a-@lAQ;@D$m`NuC2=srVg_3
zP>nIiX!@Cs=>oZ<E(@w_E24SjfGrCIk$O~5UfArkaJ$nY4$ed`@`7YNWSLkXRg=y&
zVii>r#59Z7x|Xj`7r@B58;hmjI7_-SwM!SXPS)|*wp?-Z42~BVl`BITy8V_pF7q;o
zQNzzpCl^i8Ey)}eLTbS2_R(@<5(T*evFONw49pBAe}J1pFIIP3rZVqD=UF5NaaO4p
zk_M&@MAv2PRdFr8fp}1h5)<}ZUA~2B5Iqogml_yk@5Wc4S(QHUn5dp%eLR3@f&y<)
zi_W!~(DcV@3KVb2qdo%PnFJM<@zKc2uZISraN#A`1zn|S{!xQ?F3@O=!8NhS$_*yM
zBjZg!Hp%%@xx!&AwDYOmNn8C57#$7q`5(8JiIwgD;j~e4w=*H2mp8OjcCv<|mm^?g
zVECW=>)_}_z{$=5MK5DwW9DQ|@Y?|Szg9v^T08&qOXu^VC#X+<B}|$ijyFa~5!ZYQ
zuRjn`0^APjZiievb-bE=1#tEK?k!&W&xA6Dh!XS{dX5x3a^SiXFRm0*Fh8z!lRVKQ
zCK{<X9B4$TAv!H7tArBWc{lDz2QowmEspAx5TM+55#HpMeTxQK8fB`862&Bzn52PO
zKhGdyl9lWeC_V<BRG=23X=xA!ne`W0igKkN<sWW)n1ZEKN}!VPysr$lDurNL!fLiY
z)!#byegd*ODJr%M<VKO)9YJZCc{Wke_0n+Cf2iqJs9fa4QxLjH!8#6pVgLkX<n$16
zrlM93kXiwV%#Gq>($K)(W1^IryH^_T5)@+yqZ~(7yFww!#C^dq`6Ot0u1B4yoPU#y
ze$$;{2^#Wf)`j{6z3q+eBT!;@gL_@<Nts|t^>?#}|Hcf<NbULJ6}J_A&=rvg!n5xA
zGR4*22>g%*?I#^TycykKfX}_hfyHgBvK#Hh>g)|Eg4RGR{Gq0E2u@Ht`;_V<bn&oi
z4dmzh|AM0FPs%~jNO10DqS5eRiBM+)s>!kS2%CkQnCA$NRADfHAU|-|1Wcv%UHRAR
z3Dobxgux>N*NUPOxZ6n%Q>d8Imicx0_d$X)3lqeXqXanhQVc~iIVA`2p)^MPcFIA-
znM*i<C?V#E!2__i^FT0=b3xJqb@d6Jp-LGc)Jv5ONz8-mM>2--wGvv4|6i27Q<G?0
z7qnTnZQHhObC+$~wr$(mW!tuG+f}{4i0+7v=yUP*-HQ1GR*pGxW<EASQX|qfPgIbX
zBFaD=NkKu9?=rZ=`m{u!F;u(X_hm?;d?<U%UzTj#89csQbf+$Y5TE)`T4~-r>2=Hf
z0aSTLcli#s_gns5U_t9Vnh5EFYnNQ{g7s~+PD*J#l7XNjzFvXcy_dn_GH;cj^U%2x
zn4!saX;F(smNl`e?;q<<sy#J)b<6D7yT0!#1%@95naT17cC31J#!5^JzDhk6YEf0k
z7R?(nrtjNz%j?5W;|EP#sW&>M9C`Ug=SVdr4s6At=$EwGjOaqEunNoj6@cp4#<*~h
z^!1&nI2m;q*}W@3boHIMKR<I0$Xu>&LABlu+S)!!JrPB_aYoAFCWTZqWt~n==yzFy
zHQYvF!>r-0oU{nNpV!}d%jYA>>S>KA@0T}vI+<N<Bi}dvI6QYxYONra9i-_$(QvLD
zMZMsbz}8c8)-qmaw*tM_sh&9ciNh0fgVf#O_(ISFz5oMTsQV73t61Ha;j7(>gZ%XT
zD}OznxIelXV`v}kF&N-S5l{Ag%&&wSW=1wxy0{meo&NZ#_#iJ!04ul*W^AzXT;goL
zh}v-^+6EB21{1Hx?CpE^__rKR^9fz8pR+;BOpr$%Yc@}9JW0McdaLpc1HW*R8v>{W
z8X5WYuRV$Aj9u9&(cp08(Wo^5R1{UcTA9=MwWt5;;NjsIe;@Cke;6{-$4%_+b|v<L
z<F8)ISLIDUJUq=?^5&pN9fv(wJ9JQ`Eqj0-%2$b}Exs>1t3|e`-aGhkep-0?`ZCs;
z>fzP=JHPO5=&GK<qpdo>A{<e2^1vc4Sa<c9t~s2YsR((DxH&np0a$=P{<`sWuN$|M
zqE!e1R2Xg`->#Ls`VqZ33GSeGF3!?SwSM{(A_oWOk2d)R$Nd9#4Gxnjz9&6>`$=$@
zPrhctb=CMHwtr_G7Ht8@^bEdmHoxA4EQcF#6(`}eLbm9zk_wy!Mr3>C^9INF`9jOx
zzJKOm?gFjh0HAR$PUrA>Xs>r<LksrZU`3eOO+4S?D8Q-M%?aK$v!^1rzo9IYgx^yd
zymA)Ut)ALG=NN9@CgFI5sf3ta92@p;DN^a(;NGclw;Pmp8a0fyts2q<ugdF_RJ5x#
zt9T(qL%s+~48KP2?t?WJzwC2zJ!AELClXdps(<D0-|&jVU<A`1a?0X=%0f4b>rR)(
zAR35YeMl6VUtKmp?uI>C(eHRM;Aol1J|ZONBt}|C!#ibUdwBd)ay|+8RPw|x!IonM
zx3L}@<hq6Nx5#0Zag<+?Jnc^?4j2rlD@0T#f?=Y~e&8qLDOY#G1MMcg-3RgU^;#KD
z=M#q8rCg5*^~pB?z|AvE$18(Y?P>Ytv-T!~U1zQp)c7joSofDtj^}e&sY!qt7JzIT
zYw~4AA$v=FTc>I~5;`GluIkGc@$##7j$Mk_y0ZOb*R;;*^<`%}mb=53^bvL}L#DX(
zbI#SbuEJOu6YNpob)0toXSdX)RL740P^t?3P$ytxPK(|yZESK2vCRmPQ6ilokns+R
z8XxsSiv(9o>2(#Ao;pgrwedCKHvRN#-u%|p5n}4;+Q6kbc$>qTEb_U6`FlF$hY+7*
z5eih&U9{ei?`vnN{Nk}QTa_vLuUwPXZ#&OsJ42xa?n*)?!U`fbv{6eB)d&2<hD!Z^
zs4g?-f9x0ksp|jieU<#9V;!*|^xW2s#YD5%9Ej`U#A|?<BUI`16TtQL=d^X!OiMd3
zH31*=D!Z<qlPpRQEP<y{yS8h`)lKAxoKi&ecqLtpw8q)-{k9Y%PD2(|gyYF0&Jrmm
zlh8h%HN^VHXiDDr`hEGXBYk`3N9f;!b54{h#{3A&b=+QM&i116P)TMqOi2}z6o6hy
z5?vIRVva+_Zv-yok*#mzrjH#&v7|~EF~bY<%Z>3q6C*YAZO;?&Rm{UxKJJ#K3BLP$
zLuwAH70}<n@kk!J+LHOC)3xcWbZ{R&nDOc;^a-<9j9lQ87So``C?^J(1S15W;xzki
zMMzqFxvorGt~@P^2R+$IX7cp-Km&J+!(A}>M>E|}n@ly%QUfn>>rXZfjDxrTF<Qov
zZ&Z>~?NDL{XXPZ8{8O0ua^!}OcQx*6g)~_!R+zFay?jU-&)hctC(C%AEMgbF^78|{
z5yR|Qc0w{2)qSsXQw*))#vbMd%^^et3m=uqw8*kJP+WwE_qDBDby?JP<L^jE5*cHz
z!feXMO~d|Ti!_PuG;k(aG6eRf+Tt)0MLAISj1%#wer~`XXB@g+o|v4zYq6r6)$GWV
z1nxJJUsTL}?M)9BhOCDKg#5&r0Q3OCh;}Y5`NW5$EAo(ke|yxev)l8u;A&M7fcE<^
zn++wfAEO`HAoDq0b4^rxt2S5?2{<s6l7N3fo_ax&P&%o$&%)@)?ETi`mSHN8cvyV*
zD|N0XZX#(+S=27a*u~+%Gv35OyVDmp)*-pXM-)up^vI;(YH|Y0Gvhj?l_O`*Rn7Bk
z?(W(F09ez%Rj^zL_~15?KS#@=APEQ%vI{_T_Fos6wnS*GRmVgLPVYm2qVPZV<J5^<
zJ=tYDT+M-X^lhW1&(8C@Y8aP2$`XQ(vj(h96>gIvG~^}XO@2(l!!dqFFh7+H#ga(;
zZ}qARXYm}$y~Yvp{<17TtuXF2cfYV+DqrHP(?+Lp$$X#987*8q{qNuAH$Eyl&W9<v
zfiIst+a~0`y12`3(~~C(z<gnoM?^biWv%?bt^H$-(MK@d-_4-(xs2eNu3anA&^oEa
z)I(H8_BR05=3gHQ>|f$(Wvyz>gD{$Qv_sM-v`0ku&7KtmMB^*-;iiYtHPa)QqF2oz
znUhML(DmD(7&=v^qHEAc?ttVF;^(u#AeKut*#<zCR_D`e6BmWt&mAO&oqZVOqC&!*
zdqRB|tb(RQHejHwIQ6A@q+^x$gODzE_memij7<q9UHOY!s(A8OPnztobakjImh=$B
zG3|zNDm7#k*iNck-KpYY1w5fGSN)=$_Irj7D>7@<{ua9NhbK^7Fcf-47ZtO14Z)2X
zpQgYs1K{D0G=V85E#3FzRr2<TMR`jxxk0I<1qJAr;H(qFXc2a-6Y(MihZMqLPIxAl
zk)7eYd1Mrd9k6eK`If7eO>(CC<>c#ARadYK^)%)mPX3W`_S@3CtDokrndq35LSCWW
zHvL`kfurB^<@e*7$ZIW9{-uGY_w?w?`$)X|;$?(Wojc5xVw@Bpct?L+1nJiNm?XGT
z_N(||4+`s44@)w0$Yv0A{W>C%GNBVAQRweBLgxHvV>?IRn+6b_Ze~jJqA;&|Y}|oo
zUk`Rf=M?I}2KXfIRuL&LD6N439cUU3>#x+O@uo2hvTgA&_i*%0JJs5hXD^@{z5F2M
zr1XaT7IK0fXyET*lYJ`Q)+V7?80|^nezZHMws(f{n7l`TSa(q84)%O;Q$MR4z{Y|`
z{dXtiJ5qd&0OG%=IZ1E!j}wcOKT}clrFlJUAs(gRf+}y>m(Bd+FSU+amnT$;_(D%!
z-VUVUzVeqNN1Xo7@pnjBs{1UjJS=#d3W0xjQMYz>gwiWWO|Y1}Y(tp9DI`0!@Ccm8
zke64jwP@qyD{cAu2t{?Lmw~*SbO^P_Uk`NHDdEwq=oIP9??c^=7V%haBAr34(<OZw
zm5HDP&!-7!Hv)t#FCf0@(q?yGM>u&o`ylyx;zoHu&YI%ScL?XZ;j)1v2M6k#Iftl?
z07=;HGS<omr}(H&You6QY|ai9mImm%z;+z-b9Xi12U2)(QvacI?^2?jgq-z0*SG<U
z;VLwSN<Fje7W=UF0$yFJYGB=O_os;f7P#zp9f7XrdZRlGL+;30V4~{V#dQ0Dvt+#c
zyLRCkx@rWL_JJ8Tn8pz+#l}u#<}_Xy%pt|T;2H;CFPu4WxH&ky0IsWpY=0WM!g=NF
z$JwUvi~Rgy{L5eu@(4l*647gWre!Zw;O~V=*qSzkFaF+0tJ@Lu219%04B3vCs~!=j
zn$_v^JFk7Y+0%*^T*AY37pSn`lw#)Gbe1*B?gtDzvWxv6ddkA^UlCSjPR{?i>qzs`
z?ywE<H@8=4w=&-fPp@zY^bbMz4R55V*S4%;erH$^(PB%zr#mg>y7v5Uw=f)OROI22
zqk29B!U78?ydz#O=At?F{Mlv{zmA^wa|}uX*&hc~9Xq0eLJ=w<6C|CwBKQV{uZPjg
zUhikvBF~nX^JIt-|M1P{XE$|z?OqSJ_5;Y6cCxf8k@>eU>FL8NH+8$+9*@uO-dZ|k
zOg+Ef*4{s!=u6T7!5Ohk1l7T$U09TL?ov!O@joxbvS;XEzyCnkidesR;M%9~E0Bb#
z(DoV)(e~5vQ~lLp17Az9tQRQeauHyzC_fs-ockHE#VaJzjb|O9gC0#U?1&IkUx*_Z
zk_$9EPR4J_tEWDiJpE~l!6ZDI41C&+G;A-9np>?l%lpoYc=nC6@E)oem0#`FExwyG
z&Oagv>ca|vk1gkSY8VAu%)zIrm=Lu}N6#@Ol*Q50BTU~Cl=M=ID6egiiAT8c1E71%
zzay4WdF;Z<8mTWy;j1<0nY+(#%Q{|fM`k9yT62sk6|#$UH&XGtyoB?bOi5}M1IPgj
znVVf$4?ER0rQ->w?Pdqb3<~`DO}yVU$a*8lL`}m5>SIr)XAVkiI(Clak*lM%1Dj7K
zy8GZ(?l+gG5))0)#YB)9wz@?DzlVcy9Gc&Vqd|aVr_BgbNSWCy2exKiDLDyax_*V>
zXdBQ@f%{V}#UxXWS@@J2p8QWEf9$EgX#0=cr72%qL8?T(1PVHQHmm1bt}D<+al&e;
zpyf-q2|^-l&G_K;yHA^eIFWl4t|vW~MFt`oXZZM=W;<sxh#2UwZ!Mdufcc=jH?O}R
z^%*p{Os?7<p#_R*rR-NX8+%7d%T~POT8TZMxSUx#RbDu5HY~SeP9J_)#dHdl-EYh}
z@ml7M^twM^Ldtfu_5NLipU{Ja6won2+{OzFd+INRNbC8dkT}Si&h!hXhj;a;j(AE!
z=xylp%t(?xOx8hGhaJA*IjpZfdL4d3x*w4F<UnG79Zs_u1?Z3nt|J{J9FRgd<*nol
z>6Gff0f|3g&}Iw2+USU(@2fPsB!=}YNXQG1j@C&4+0hY*E1qJ%XD3C5=~NxKz-YPP
z|J(uUAd$^V2|;ZUq7tU11-~$=N=OYQ?l7?_;Nw$@$zf25plQ{~GcU29AlzI{7u<~S
zRGXCx3LG`9Ll$}N5UjM_15!y$!4w)msCXA+f3&pZL#*7;V;-hQYiy;+8j#@xCB3Yy
zH3@zhNsqwn4pCXJkmtw3J|{Y%bXaR&EmHziXEpM#O@`k(lvLGZgMZn3o7K=*2`Ex$
z7~eN1!Lp^(d(3}!3mBmQBBRmZaTah1n2IsGu*geNa|ocq14%=-1xIJC6jBf#6liJz
zUSkE4r=!>`c9PprYuC@3v&(ab!odlKIz#A;?l!xfpsDZO?8}aJ!ozud<9rsdfqxpd
zfu;w1Y%4q$eL3Gq0AzqHk}M;ov{7$EOX1c{@37dbTYa6upon$hLqc(|W{2K<!p7Pw
zHw!n`*>8UGy^_C{zuL6B?hPz-SUDh5tfjy;;7OVgQ}c?~Mn%(_L)F8J=5Z8(+CCIk
z6A1RgMP`IU{zHvwIC~%QaB4jW><btK@D+(QaF=-sCnl2tAhQ}|ym~+o{*!CfJx#f7
z!_}6Y)i-nRyAWnR_RX+RXliy&AgoV<pIoP}wjMpu))E4qIhKP8BF&2;p`=xXyZk{k
zZiT9cW5_n&W*c|}MJuG8-HBb%_#R#FfVfp{W^v$fDUi8+|5)VV*jEaRFV;fUc+D6G
zz45k5tKMJ-2r>^hym4iF7_>(Bi<~k6fH}%#E*|S?J(Hdng@ZR|<?&*4c;G*d=TSWP
zDzsADM$be(;dsq5WV8n7biJby^c*H#w01GOy)&O%_zb*s>s>mz^fm<Yg0iSm&ngal
z6(rWIBWYAAxQ_7EZ^;o*`BdHlm6O*qhsTC#>{@v}+RlIESGc&k>2=}WDl8wHLx_)(
zOZP}(vBiLH98L+th#J1uvu95-h@fKaXTlS|Fn-=KlCD9Rvh^^#dZEg=yE0Q4RN020
zD|JNuZYVk2X_N_D$@2_N*PymoIKD2rs7d@)&?a~vb%PaxsXRUm9dSAgI$3pYn|g<7
z#u@QTR=_LB0?mwu*-%fvAgHQIULMlJ>vAng&1r=a<<FQR29?0T3I_%?&hFFq6$ArW
z-fN`I$2FD*j1kKXhFN?dJ!D2lhP^Z}Uw#y@aZE8H)yellZMqasj9&HU;g=eElH9t5
zzW_(wt`%yov$^I>#Nkm0va)1t8~n9g3<l*8f*Jboc1ZDA;W7{<O(@2j4vTGaNIeV0
zlrw4{l~SA0o1Um~-936`K4_9=6nN@Io$E0f5&}1GCl3~hJp~?=EiWQ+n?R~4CP#8+
z)+k<fPG}fZa}SF9kU@TrvvuE{txE+~PfeEJub}*f(i9GHEKARVHa(3hU+AV*)E*FD
zN}8}i&PkR>Lk}{@<w26QxOg6>K=>Mr{VxJ+C62673>TE729Qr}9cE&cp)kZl!}^mW
z7hI)fTm92qeoCsv>zhS%eGh=)2-qoDOQ>0ATlY1HAe*P}KS2G(6^IDoKv}d`C9Gh|
z7*?*AoLj1%dw2nsUb5;l=ZQ;sPUk5}wP}xh4hqS`LK*TL5p#98NF*55qSR)>c4z`5
zPMkvG64x<EbH0TGwAf0uoYztbU)F+~wykdKZ~e5`8!K$_;@SxuPekFid8{WlRX-Pl
z2lW*IfS9OlcJ4_3Mn67pSOj{&DIL*`s_f71m0jy^-sF%cc{#foi76awWkuNa{!z@Y
z?eOzwui5h-)#L(VZreq)2Fn;LFv~ODZ3XN@8C6NchkO6lR0o3~DmZ{h=i5kX>V-BS
zK8K8k%*Qh1gdhK~%8dwf%NRb~ehZE4opj>MwoIf%TewY+`jhZIKIDU)EV!*5fgF&V
zs8VW_p9g~pYibmU$M9L`WgWov19FHp>hE6Qjpw*)>`;S$H9@##!|y>~F(H5QTa=e=
z8f-dg%Y-tbRQYSd&Ix>@`L<Z>mt~tdrF+>R_6R)_uID_WHDR|&qoWu`0bna^l5{}2
zbv-eVl@N#Pd;z*6Sg?!TYt9FMY@-|C6;1eN?$693H`+3Eg^pz94}Ucq#SyDd{0pj%
zo`mq91Ka=GcKCm61x{wh|8-zH{v#dte_|xJb#wDkBQSYHDfwZ4Wr0w}3M!nW0tFf@
zw90AiNV(T-=6-kfrO`|@b#HCsK|M7k|Bmg$nDyA1{^igl;qdUepFBD^rw(i)5!RGi
zG-i_5$b`rxni6HoAiWi55eCtT6^#_P)BAb3Sxr0S-|cSy;&}f)r=Hn(Y!lEZuCLA+
zQ8Lx-=+i|K%P1Bi*|utyi2Z(gscq7gqx<P0AV4E0F@P};LJD9eF=5itOE*Y`a@QdZ
zCMGjhIyd&cAJuuqSWh&L0Lm(+Swj*jmf|X!;@1!8)rnP(7GFJz=q<SUsPn5GglBzb
zhagOs*cz0-L82YD?F{5i@UmFvd}b8OE;p9JEY#qn?WJ>ZkzA6{i>>G@x?b4TK3;A~
zzftoz&b?G+q@4)yTAO;c(>jf$l>`w>T-eiD<%Oe2NF?cs|A&}9RLedo>D4?r79wV5
z#Y8SZNC2`$Da3SW(nZ&E#(Ys+lIMb28ZMk_<u22{9&vnqy5cMKVIV)nA%>Z7?#eEe
zD1|>f!axK&WC`LAY?quL+cXO33jtBT53P0XiQFUfu>1bI(6p`MSLtSTxKr0v#eYA-
zOS?gTD95ks_w-YwWqS~Ga-|yOxY>Ftnu2bk^Ht<iEh|9|MefXiWk-k~l%+JJ|CPG(
zsbc38-uxSfcze9~JqLz%TJHjFlM%)iSU8QJpPVrO3^pC-ukz+|g>a-(!%Hyp1RiZs
zVRqxtLtVCMxCo%s^UTqrsdwE1=ucA_$j=eHQFFQ0WfX~6w8f3^n(Z5lwc4tDIE?4e
zn#J2^1z*UONB6xgRp+vPrsToVy^-9+m7Dh|u2ZwEqv(q0(W-m8B371SdP=eVW)$}z
zAc|?PQ~h(qRFn&Ywn5Fl3sTAqEy`jhyU1IaOqjaBvGUH@c|-bx2dflmo{5K(zx_3X
z6U!g&d46t#wyEbgzbt!cPg6wI=O)Tqpw#+Ag0ocX%Guv`c&Kw5!@Fx|l;Jlotn$S%
zwo1|`^>CYExRc|PF`n0md}ZY{!3^;2igG_b*f)p(ilEi(Q$Fm>`3dN~2q}hRHtTUN
zY?Sr8{QS{{TR3YN1EkaZxZZR&fv*CujSyV%>JmtqC_)A^L5Y2A1x|otMz_YPhA^6E
z6%=4<&wML;I_RiGlo;XOeu=E55I-|u&NAwLv-QWU2@gpEGP$K*g~cO}9ag$Du|Ou?
z{b5niArg2Y(dfWClZ_@t5|B(C`~%5MUziEZFK7&fiV2j{FZIt`gPoraT7ZwN4l_vi
z{wgo^>hNjGYp*{%_(dtwi*#qM@!^RXJn*f}5noXJ=UN$k8!hwxf!dn*M(zT$GXI<x
z36m=@Z=*$M`b3Ad2H3>rg{Z{b1vM}23A^n421x1t-^!ibDRnq&yvs32rR;5D8^FsK
zOm#x{cfOCBhfa#sHyK4a^-iqF%Umauy^G4EN}sgiuzyz(44`mn0jd!5et|f^5ejv1
z4+I!d4}>&17Yuk0ge}j`zBsY{s;=7YNdSqT<GNwj_H~<W>C5SA207-N5!+A`Ao?}a
zu<d1*-BkeC0oZ1#g|Qs)2HE29O0uMDQKk?O)vzdB@md&$ME2o+7QLE*0`Yt@1t=O(
z!WRx=Ips-wnK_?EcTjZB4R1=*k@S<YQZg(e)Uhu6)ZwUE`S{4A7B9?cMh)}e3#;%b
zo3+3SxvAwe17C)#Zb+a<{h+?B^rqcjb8A5InoJO+BP0T$Jyq(}HznR;!lT=a23@ZO
zMrSMN34#$7@s)-5CEEs^zm%-k%9yj);#n;M1E>dWMY&Z0A^>oJ;MLR&n3y$N%;~TJ
z%@SA_;QQJGJIz&wH|J8(nf_LFmyhv)&C$n<;!w<ao-j?xiZv+^l=jKyo_y|>AX0!r
zCgKP4)m5}I2qY=tW`%;9LPQf5JVs-pf5uVCaV0%CPKhPq(87fRk)}Csuno`z>^=4S
zN*$)weq=C37Di&$>Bi;)1{(B74pdJExfvr;Eat_Xv{x}m(2iyFIybwz9N1Tau5F(H
z>q{U57!EVw>$)^Bxp1LOBQ(g0G@t*`vy@6PXNo92T0PB1u&<k!lyz1)8Bhbf=S9#`
z2pa+7D;Kqjc|?VBTE(gKF?i0G$X@fH{*c4H2GDJOnMXr+AfID$*PCY%3qO`G7cr15
z{nh(eDvmd}#s1@SLxhk7TNWx05#Iqpcn~`pr3<(QcKDwB&-zW<#rH)B013?zodRhj
z!!%ioalsP@4-xo*ZrfNC-2V#m>m@({CP6tL<)TQv(^8fH;x+myKF@q*>yGnn_4?WF
zfXjgK$b`?E?a=wvsd=z4U-o&KFD7Jgd6wV-+ryI@9bI=K#;7zLnSI3A46@>@(x037
z4oIGej1_x<rCYcj0F?*2ZKTFx|8THHJ0N2j%F1-;$ly(Up)J-RnfBz%VUqIwOI`vM
znWXhOpHT~Jq;u-wd--#*sE!i(@@?v%J@$FBbT5U+kYZxfLVZKCQN7Fp8rk@`OI|k3
z*wq>Y;Dr{l^7G5w!@Gw*ZV&4B4X@z9IQ3uR&iG#$QTG2`+_yC}|4&BL_qA@1cv%Ai
zE>N7w$DT_rD|S>C_qjxFL|8F?8d9{9Bw0iDU+*+1TC&7P^a0KyI^aOVTYcx+ciR&8
zzq{1qPSg0#;c37fQerUA6Pc5Ia4w9H5b(LI=}raxfc}%|!{y;=G+B-bO+>gpOz@g2
z|F`3vqoX?$mXoAUZv4DP=#n3&RGXi@qrJ1U`_sW}kRr-k5_&H#wujdeVLnMjTk=kQ
z&1`p8shsP!3&;AW)vSIg-eWKU3LOFoW5K{AnwP0YFjDF$9ylyXf;zQb&+BI`Mkkjk
z&As8AE=BZ<KtJMH1}I51@sfVipXd?7lAj{{hMeBgMBl!jV~Za!FZ|brX>gmt4LCPk
zAqWY5`Iwe&#-fuU#zXyP0Qs+Q*SG3UUf<mMSak)vDNe``?)^I^9UACoV-~dN4^fv~
zcCOhuNlO>qs?zZ)r0uHH;AjP`R#zxNokS2ImR5iQaV<u{t5ssOPl5?Hika9`*cM8*
zDzRyG&==G~CO;xQb&hmYi=5fjH?u$I&{yZsAvvZx7f%c0G-kF6v|=znKv-_ttgRPd
zUtPd@P=;4_`@{?DRLp_<g8tj+6=GWn_KgN#g`ROa+J`c!sT>Td8~|>PR@bEZ#hR6_
z=FLNF!<gN=Cz^lu)(SH#fO5;jk$?NfWw7VULw5Rc6_Au8+Q3&j<*pfG2?~q^GXXiQ
zLbtbMyBDqt|9U)d^1bOZ`i~~s+iwUW=nDc3J{S}NkjnIN$nS|mZ7pGQIwS-Rer}0D
z8G+ID=OB42HK~+!zj4CQBD5cnSR6ipm+f}WhOQwD=r6<}dQ^Cod6coW;P7|%YHy^a
zuefyJjXlZ7?QZE22(<yOd%79ls+3lJaG7HS)kPE+;!EE0IhpSO2AljTgDz3s-m)~x
zKtLPrT;m9r>^tz<-jpo4Bf29urD$e1ggtoNw=HX`SUBH?JZoAu1>UH^Le9P`|D_^O
zWhByEU@sAjUi)=<SDBresMIYE86Vw@=JKz<vB|9Ls8$nA$jhj&n3f}cOLCsDojWh8
zvBVz|<L7w&!7?sLJ2DX68fW8ykwj`E`5iEY7_&xaOOQ_cE2I7)Ex`?WG{?j2u{d3p
z9<_DgeBycj_m9PvsUs)Tdal)91dcZe0!{qUd)A}ks4U?~6qy(H{)ovGTU%ZlC<FZ4
zTdiFe4V=DCd~?g?qhpUM8n^RmPjG9p<<=ebV8#d$y|rUp2oVqnFe)aIz>pu1X6}4G
z|IpwqMSO_$i9eLe()AJfu~Lv&@3M-tsICFR>npmW&XHXz`nsKh-_4HJ<5SkrwyWxG
z#uq=-`P2e6OE3svkj)U+Lt4!vGmT7|L$-m_FRDo-=O4y>H6FBIP9@EYteYHUz!S>G
z0<;0(Txe#d)Ybuo24$lk{_oZryFKr-3AMKFs%klD@4GooO4~0R-&k}l3MD^s5Xq1w
zLfWJXBoG7iFQ_IkAkN~EM}Fu*_#6vDwaD@EMoWvRiLtt`Ju}T}wiD%(zQO|d-5DEp
zkmjkb9?jL3hkX_amoAgf(sz>!<(WQuAbI2IcN~}>W#N)zg?7`onmCAbIq<XKGFMVh
zVu?55(G5!N%u!5@fodwca1%^!+)vkQwOu*nqh}Z+i~LtGo-{;A6M!Mh7O?Xk%KYsq
z+|Wp*iqMV-Br}o611mZ!WhGvjzgKTdy{=~feQMpQeIbAeOdkg@wRxBQFY&y@;@S1y
z{W|XhCLHEX11FA)E*{e_$yv?+C}HsHwwu-UE=-GzEd75e*^Vs%N3P@r1u~B@iX)qo
z)Mu95lP*l6bQ8mc()aeX8V*^dL8YZ*-rqGZ$xrGSh7oYw8tgOW9@*d?Y#eQ09Ns^p
zcskF>K3@15gBo-yTSQXbJc0&s+*oHw5asr_xsc*@c*Hvf(vHm=_%TTX?1DVjl#thn
z5`bfgz;Q$k-{8OBE2mmcY1CqhR+U?8lEg?d+m>C?IYatrgoAG$5Z#E}wbq#|9Yv_B
z&;%;B|Ltp>bjq5Q*}YjVw^mw;UEs?ME<to;FQpEtOw}|~Ab#Ug`P|U{<Z{X9%y2t1
z1l$(-6Q5;x%2LvE57T<*oy>Q<F1<9E8>r=IJjdz<z{Bf4`c6X6E3%EZR9#z5SaoOX
z9N8_uCfET>a?G6=bjU&kAJ}Ev>_d}nkZ6@{o!mq>Xf=cWB|PvAC{VL-(Al5xxW{ZO
zl@!QRfa@$Me$hv5ZYmvPZuawYfDD2^ztau*2as&u@gpcksDH&L7>CuezxxFce34WA
zFHvOr4;%v{+y9?3;{TT@N-;Rxhm+G=B*}3vPO&JK!pDvZE7u4AA5lCm8uhycf|eqo
zrEa-m%aGHP(1*ODdwFy1@5}NEcpaRtI-kGC#v;)bN{kQLd0IFuAQzPYBJ4h&;(pc?
zqQI0j#(2Bc_50oLTupk#_r=SR_2cp-c6ZB*Cy6JlZ{BpLm@X{plZ28$Vx;We@67ai
z_3?B&t6g0z+Kr_Mhvm6&kOPw9MN~p^x-pp~76yOT_qhs*t`~7nd7z8)<y0A;Mx+Gi
zlHMcaw7pQ=X$C8cKo8m%vV{@%kih`0QkDcMGw-2JVKrH9%6H>}4C?E}Bz^`Q(BBMZ
z8mM@~S%h9L#s`)ZcYPTRBE{)7==1hUmBfi>h{)E6HKbmyt|}yVH3!B^P$09p(k`#0
zSoNr#eRQnNdh~AY0=W*{eB@fq5F-7>v0>qN`2gF~^d3TazE+GDmGj}^==BSY?%|Ec
z?+n{zf9v3^$@*)lTW;uTO5<&q>V=GmEO^-C^)n#PLZH}oIEh^36knU<Z$I`4X+&})
zH{}f2V#0Li^euVpshG-~J&1_atRfnuth3i?u}VLgG2xJ2a$}^Zd$>11+fnPhSbi|9
zJoDgf4{#bQ$`lMt1(7!23j_X*0`!GS6NmsYPl>?mwA_Z_)6O16LavC^s`)pte$m2L
zw=Rv%V<axzlKNV%G+HsL^~Q%8@44@A2$De*V9MN`3y>FTAoMIZBa|?LXvfmmJ34x|
zE{(%nrMX&Mj<Pse4&6H?Ew;=4Y%XHYtX!y5IXuz7SrJqY8YF?~rBs@uKaLC%3Gm}=
zBf7ZGcHx*x-KaV{2x_lQL8@OxF>m=g_RnSlOt!?p;ZN@6U$OV~!fMq3EQ}<~%xQ6V
zEt;JD)(0uo>s1<^)k;<TF_wpI74BM-0vzePsLv%ROhtTeeSa7j47=$;<nq6J&Q<uq
zM@%-(76pMH)YLd&>J~j}93XKwP09Ie`kvZunw3Q$sE7rL&#wYzgB4F@#2i5rXdyET
zJiXiGnV|6|z~k(nq|BGAKdDv}_RKsNMBR>IJ#<y_6N)v^u(HSPix0xvLgzqtYi(0&
zn1O^6T%oZt0{QXPG)&EwD=#w0WQKkjzDlfch~2O?)z~<K)f$LI5A>5z7L+|F9}{4S
zI`HQFBMn4O2QP*eY{KZW3=6|DHTh2&3-cFJ+#R6`uLaKy6CEV%W^C(G#2o!HSrQK7
z07_}(p%bDtmDJestQ%#{_bg#&pe`#_p2#ZPZq_w5Q8BrS;_{RQf0&k>3RF}6jI0j?
zOOH}CJE$PM*J&eNN9p_X0D_i}ewez|+Nl&I)c<QuORD9Ljr{Tqm{{jvGUm>zlYMuU
zTZE;vaa*yf4}%V1nUv}@Dvjb4lReVAG;O$L7!e5+>?5b^-Apy(cO4#J6YQ?h7bFz=
zT}=Q{0Icw*@&+!N>eWWiaM)bK`12Mn*VG)t^R@+wmTt%l1VgvwW<N32K$DPIo3&uQ
ze)RpKkSG#hfstPlLt+Wein{Wz%ufgQPrgn6I~NuXTEEtuK@c(JU;KOY?%Q<z06uCt
zr0N%}1_O7}4D|{+ODl6`pZBAi<D)&T={+nLdvUdO2^;K2IlmJuO6z$ZNcJ0Zd-gRA
z;4~PPjyhzo9aFQYa3nep1Vg2GiMFoc&L9bIaK7|JhTBpS8Tv(;(~kCGOk;c&m>iJ5
zOVCFNzLgF;G%a2B0F259<|V47TOEJX_X()elF&vq`ExM+WA<#g>ad&80mjBOG7(#A
zq6_}E%e}A@*;|-Z&%Vf=*#bZKG8nUvG0J~rU!+VtjS}c<u`YrwTcdehLQ5k7!F7t)
zb!XZ`?jwu<%c8B-W02CR;8O_40ZJ&HKqXdU!{VvlJgkXpoH;a{L$$2RI`t7v-)@n*
ze5YumA(t!f=X4fQju<?Y&_@aQr@%>XK4|a*rQrRvj%GnHK;R1~m-M-C19;p2dbcGj
zpJ`S;xrXnSnZ@ZFu(p*ow(Uy9y~U*}5FPF1P4Q0ITr&HT$rc20?b+e~!hZ{){Qe4g
z@92exx$)Cgn_48@xPU<0*tr38%*_$43ol$<kM|RHVQcU3VRvNn*!=^DF~G_F2;N3c
z%MUtN>{0z>Kn2oX^o`0>`3JEt{p--DcV+yS2(>l-zqz&lzU&Mwp_o{h{-Z;miG}fh
z?$EFK=d5r2bJlNq4f!n<wN8ZT58~l%4O%G^Sy*9E&bA5To40gW|5y}-`Pk0?T;n>1
zCbsQdebmC@L}59M?eq9D3pWj$_M`cAy?qZ3KkJa7Q>jEqput~;$e=kS2h&8N!&eQ*
z1;`{G&%*ol^c>1empY_`COT;h&3p{z+4Avk?a<eWOaT=P9IIy6?>#C-?4A6a42EsO
z-~ReG$rSDFf9vsOJkw!F_6em#a_HjlZB8F}<=yyEuU-!hw?Y#$Yu$w;69xf`WQS!$
zqYd8-%$Xn!ep&aoLxdzosnM^X`+e6a#hM)<oeA~VX%Hoex06fT*P>ybaU_V1>W89K
z_ZKmYD=j@H(S4LYqd+X-Cm=9!0RBvH+#D$0TpK%#eLp@>8PK>d`TPXEx3nEhS$l(0
z#q$ir{K{%&^N&5}0F<7iOzD}gRy!X@*t=UpX3D2LR{M2(91f@K8Nf*I`y#zhcJOo5
z^In%uPj*oMtE0WU^$h`|1?FEOs=V2&7b|P)_q&^auA+BYym`1FM`M1Q^?UQ9Y?<AR
zxpBza?dGo7*9fhU$_dr##)A4T!FClZ_1Sqb(al(u*1kGkz@t?o4C}jl!D)8)xO(c<
zmdj<uTb2p9Hu2u-{Yhi!>{0dxQzVE%Fd(uBn>D#_u4ac&{8l^Ee*lJd#6uZC%WyPD
z41=tuQ$~gI9}B~+e)EqK5zKluQ9ecOsUZ}&&R=7b71t6Iw8djOW3hhKnD_0`!C`SQ
zVM+Sv>{eAml+@?ar{hLBV7z^T3X{E-P}M~*N-xI5WRdUCd7xRl@D5oCk<Dg9dXp3)
z1)=TTR4>utQ0hN~=7ZK(4*=U<YYu3fwE9>P1X1C*dc3Jhpql{L-mpC*yO}!Fy)@fC
zsGqpr!1k#RhRSU<X(*9AuD#->=E@t#1<r1SImX-8-+k&}^OdjjfPOfyymBR^LttXm
z5++;WbB%d@rc@TQDDNBPjiua+O?PZTHsg_LzssPP=Yo?Spn1WRd!P1h;(%p-JG(F3
zlz7s)bFxY-T)%eiZsg;4BgM4NGIJ~hgG7_r`nQF`1)Y`sM3pqT_MQYvRIU#Pa4In)
z!9!rlKu5&5)}#XP4kEr(SrMH&QbO+t5&~fow0DxS2n3XZIJfg8+YmF@hXI(anBA&Y
z1dA<H<MLa;P5iKtBtEn&pSu919K<{Ad`u5YdTeM17amR&Xe$c2&~1g_C9964eN#Ig
zAgQAy>9k>1q2r-62gm`4ve&L<1h^cTu86v(v+IrRH;XgQB|xd+s|**a^G3B2s!%8e
z5v(d*S;3)k?_JQZKnEW#M?~XELml2F4k^*a2K^pNr;wUkCH2AmKpa*Kddr;EyHbSd
zgGZRmCpm<!Bi1y&%N4bYJc5Zdym69A_$s+1B#2~YBvZuz4EMCkhC8DORKb16v5PSW
zEuo3?(XF~90bK1nt>4x18kjpEg~R@cEC2^at~V7o5^(?CRB8Y)TJvUc<u2SLfmQ}W
zvYMlT<{Crhg#*9yGrEEKTH3%hR}Ri<AFLgmH`Yl1iYq8WEp`8gf6K~a$&)CIr`D@v
zr&4<0h~7WUUQU+6&~?2XJgm5-B9$AUC>3qtG~YEex>BHn>+n8RVDb}83>}gf?Ae;{
zYM62##ssFi3hQBJO=C<??#GpjC5IsMYZggzgDehcEjzum2QxcDG}x-m5XVLVklhr{
zR@7IzbLj{S4JlkqkXzz+fWaoIW#Bcz3m*wLvpt!gLt{J=;tOpFQ6YB5s@6`lm#{)G
zWtIUCj0k500v<+q$^Yo$!mP$T0st_zJ7xb!_pc>4t(f(a<aFdc32F~-=baoELh?q+
z=6g~9b{0cQ3D(Au0FpEn_n375j-Vw?&V!0Xi~Byz008EnE*v|h9P8&=5b9(}--m&1
zW7Evc!jU1(acH7|?15Fd$s%m)YVrS<MdhWvZ@+3S3s4tmpZso~U2{y1pvE-KxXa*{
zRa!G*9W=FK&PS~0v>3!dnk^v7t8eNU869PQfUuWIay_9ArkLs`<E@iGi-N!F2bd;Z
zhzuYp3q0y91y>drry8o8RMy05Bze}9&hqCVFm7g3B0)rD79bX{H<D7A!o-w#2Xkeu
z2iJG1dz}+(t4RL@sP4-G{K~Dr3zVlrokqvGrG)#ygCi^%6IDb4AD@#ths8;^+y=8S
zx#r4%o6H0G5L=DKU0U5%czMs*QGieuwwI4}s@XpSqe<4hXuvZ~fPnWshf_HcF58BY
zf&_<>Okz+|IRb1$j#iQu{{cBR)kPIH$pTy5KYs2an)aV-F+z_3u;$`Iu2?&<u{fwb
zKRc{e%rcFuMs3*zw}=)dC1ASH{tQ)GwypO$tSW4B*L0kTGsVCY3^+!Kz(lJgtOp?(
z7yy%>7*A5RIxr3-&|=F8$WIp37x`h!+i?K<oe|(LGXO!jamekrA?)WgzIya_ZZ&#A
z%vnd}k|2aXBBk}0+W4^<O{EEPrFl&AaM3Zt*RNsDjEA}~L&KkNK0{Orji|apFtaXs
zu<4TyXiyEHx(Yvv1Z?%xX`T{2?(-8_qJh-pLgNwF0tDVZcshs(kE?+-lx_81;y&-;
zxR_8ln1~L1>uyk)K*&+*ZIebJ!j5nFvInTb=^IGv{$3LJ-QFh$uRqDMT|EIpAp(KK
z-bTh|!3<BZc##VAS&T6$0?p?HVubut>_4;GO(s!vNroYa<w2FkQ_c?3#ro86n1Y_V
zpEJl){Y6C8s<LO_ux0I~&aGUArC7#A$k9wf4qeO=-CvlyWG7<JLBpYi6}PIc_NuU(
zMHY+A(5i}sfahZO6i-GrFI^hANxo?zK}B9tPc%raTio}_My|Vn9q*}wJpJ;!puNu`
z3{JSPADm!)AzUP2QE4@IVhOSn-(Zf!AJV~FNV2<xS6dY{8SOFBRtz1J<PLd_8HD87
zGn|2BoGf!U^)iv`Zbh!U{^D%6KDP~SzCk|SspzyQIm6Uxm$7?7k|cyw-a_i8<7jrv
zq0idUF<y^5xQY7NLsj;~qQYyk4wmON)m_zSkKYn|t7!LFh7>GI+93j_(`ECj-4oo5
z-ErIC(MjAddq*Qnw5f0~V*bLw0+xJyatoEw$@O%&_2ehYFQXu(h_Aea%$2f4-GNO=
zk*WI?>!|L!?QDng;U`sm`fpV60~}i=*7>gy?%(45|68Wb%FO(~dhOGg>$Y2C_d0t0
ztt9{vH~-T1kPr2cSE4{5YaE_UutjF2W}%32jPT<Vt7RQknWN#v>zAO7^vuR?>Mq;p
zn)VPHcazI{x3xQ6hcxgH6zW2tZl3~`1QDu32PAEqp!h-w;zS)ivMA8M?~iq@x4S>r
zHy?}J{5xGy$2e4SM5z=47yiXk*9FvKuW!HXw~l!Bc*l}NG;1|sov0ScX767UoZff)
zsmx#~XNxvgE#FX}Jccx(gEV`w;3GvWl0ltN2we46xgn<P*3-}bGQZ`MjUHEb?Mt?=
zdoAv5%g3%1E_y7ixNl=friH@J!lzllxinE;PBy=>vQ3wJD5K9k{Xy6r)`xfOucafe
zg)ihSvMyHbkQ&LS1B-bgl!t=Z$BHmATzGP_HWxed!~>&74ZH_NC3$y%w@Ad#rsCk4
zxLqxzxLWl@135RW{b|!LzDI(=9b_N^2c_k_9gsUp)8rAO7FZ4m#LZdc#&y`06Of6)
zu^cr3E#>i0=~NfMwJ+j{>bIy1^dK_>ITf9SevK+(5&}b}O3FCJsm6+A%*BUhe^F+V
zN-EsbIYLgE)1~riYYaj^3wF3R)N7`T_GCnP#?v?=;S|lkxY5Fj9B8)LZtTGACb_q?
z{FZnDbLC77c+gsL!_=rO+%c->a)U-!uEvrZjY~1?Peq0#(mfbz7Gq$jm2GG+U)t9X
z{Z>`vKpomZJ8dTxIG)(sz*bKS&cj;BQicn}+OFN5BjQ-7gqK{|lYPMFMIJc3SLe1*
zs%qs*+C56;33Nsf%SWq@k7`qh#JZFK`t`t)Bl}_L<FizUy;4f{^fR>ccZvP?fLO4`
zVQ3%;4kRkq#96IAtjvv0OPtRxsd7mY9YeviX;O*4Cb8K1i4%;?&H((dNGa9BW$H=d
z^{**Gr_yLuAkZjN=+?R$(11NX?L#>vS>W&m;_@?D>AJo|as?suCT<5OStLx0eiFeT
zMeIojhrUM%&FFWiNTZlqMDEX1v`B|YuqDk1$fJkZke!h)qx8X641$2ni6$vSx(HAS
zqA2%KTgbpA+sARoa9b@X;%z{!>mF&FAijozd^}<6oT$-$a)fLa@#ot$-y|5zik^;v
zvY|)$#%KcR+f~pOK2eOWvM6<?d}+9lW9CiK0tov@VXTp9mLP9t#Qjar<b``-;EUup
z(V<9l-kH)pr4>X^2|a}yOyVjeVl|yXaGknHSI32G%=|g0g+wyOQbp(y=4;ZD0%W(>
zQT`gk;bkW^KY#2%a?I$UxGCAOi|$^u*i&M%O^lHDVcyoqDm9KHfv^)5z(kU4Xk58L
zfQ~giL1d5R5V*gg4O+J%nVLZTWGh`Xwn~l9n&_u>!Ekxw)(rJbF?IViH#{_)#l400
zgC>+pH2buIB856f#oQr$zq|{k1@+ro=*|D4O#pAC1n!Bg*I50jB=i4_2p;b;@IcD0
zn-_zi7wS8(hDq%bt3Ln`Br~3`4-1*E?s9+%ipkJ4ns!p78V7lpXs?I?5Ir5okRF=Y
zRZtLgUdqwS4?9VSUFu-ET&EBF!!7O8Rh}w(@6;69NH~*3+0}-uzq$yipDHd%7O0h5
zFn(LV!3T{~MaR5aP(R!u5TFmBthT9ObHxx&NmYTm6x?j#qWQH78a@<8i+YpQLNvx}
zJ3X~bmvl+;_$S!DvvJaP8YnL?vGX!gRuUbNV{biHl)DWvj}<pju%9nA0d{F+Mrpx*
zk%KuJu*_DFzuSU)%Rx+Pna6q+43@q*8XaC=3w~tvsKKoN#&8`@dPaODzQ3UK2ePtl
zXMmr2Rn&IUg=xU=wd1FPdlGK4Z8uxss#_k-x7qR}1L#QN@)wh<QRSPa(L)@Qcr(+2
zv60eYyp&=Mx9>O9b=sR2%$~UvVUWF^gKIe*>NVyXNqqKcm2k3FFr*D~1Tp)!HEeHT
z<x|g>cxkh<ie6or5u_A61tUxsQ)5d_-BqV@2-J}R_kjlP5As#r&$Cv<Q@7VpfitLT
zAN-OxoO#2;&McxkUSGGAWF~Pq2XJd?IRmP~W34V@S<GxpMeE9?^p#!5Gt$c43AX|V
zk}o++hfM)C|KP*^;%lKr4BaoqUx65x*rCuQ{_pxkoR@F6`jm?B5=xF{dy{)=F&8eh
zU-|?cGN=Qox=y>Y$rh0Nsj{_1uQU)eqZYe?_9x|n{PA(0R6X?m>05&V+waTM7x1@J
zn9RKjl6T<8-z}T)?*WrX{$Gg)46}-~)4a7XNHBfx`LK1O)yB1VAO||AJgr8O18v&=
zd1t+hPM{8oAV+(^^QeQ0hjx#1N4VD0b<Fz}Yu$D&f$J2r^^fiJhJKQLS|?X0=4NGj
z1mZ@Fg=!x1Qui2)ai&fEngyUz1Gw5sQYHKMa{MVNtC1#{4JAG*Ksf3%snG7@Gh(9y
z>RN9Up`JqJ6uJf%`dPXBC=-VskN_>iuGy5s3CAYt7LC)5<sy%PvWj@Mog?ed<c^mL
zt?EOk5Y!?aT50b!8I1@Vop<u`-_ZcVELd)jN}`1MIWc7`8RK<DiT(J|;A~G;X1Nk_
zWm@S&dA=Bpm5)s5)FgrD%<TE5UVXM)-vZ;}nT-7QnX}&7J^aZtbcuL`4JC$gP`0tC
zIs^m$@iFG|!mQ8vxoHsB6*YT(Liz7{O}jXwVO}XrZb<e5%{+XEd;t@~NG*B>$Q+hd
z!`vY6p1TNhS7kC{eS1jP6BQm&FMq!!o~vbZ|N7$1LZ@j29=j<u_#Dk_TrC#=i;ovB
zXz=`quq7UZVyn~rZTScq!{0bPfvxIZY*5xTmcIZ>icOrCTHhGf(6F=pzVGc|WkvI#
zFDnilpAh?{Pq1{i3?c(bd07BVJ{<WchL)>eqyA>tZ!)nk`Pd)H_6wak9&8}{b5Bh}
z;G+Q>qQ5#<+2EF=8<s4Y+PRvr$cmLmh8JUlN}pnX{jBXRDAye*;USHA+Nh~M+bb#*
zsjMT_>mz7s^2$=rfzk6sJ)*+-8W}HE=rMg~3OQ!P;j$&yYyHF+YM<oEuilrVozLE=
zw|5Q!1U*nU7U???jNueSI@-3=)Y5n%5Q2(Mq08xm=%6q8(|63xfq$E55aSNfimxE^
zvX|YbXR1l2^M{m-NmzY{j`mwByso>L8<YoH=CT&thyM=u%N}0@se_&`Zw9M}Wl<cc
zRb9^fNn6js@1HH4@dKeK$=A{*^5BaJ{^3SOk>5f$oy@_FJ&-XK!GTVMEfX0UR8~9`
zUrF!5AH<Z^k{g{~an32iaNU{$Y8i2WAJsSf<>q06V?&3yi#-20hOocMkBCMRyC0q*
z2-~=@$Ab9xd}F$3zSI2K>i6HT;TO==;q&G{$HxC+urM&N{?B9Mzozd)Hl&_U^*c1G
z1_t+sgeAP}WO8n;COIZWx(u0i3dqDZ(F7<-kLqnHUoR~4cp{5GFKe||+7dt@Slcna
z-rGxBUQ3@pSDWkWRjW$wj1j<i?oBD=K=R}M3Kd=J3tqQT1exCH-<Qi)Vuv$KXkj7@
zU?NwQw?7{DhgT2V#uKQ|9x`Ogs0zQMl2hMNZs>Nqx;;LRm%}8<KUdT6>Y+QBE;ul;
zX2Fi>W&a*-5D8B43=<6)Uro2z&sY}){Du1JDo^5~bsWWx>V_}}8Cb9f!7vdS?vmeU
zuRM#<T@v-+%K%IM=oVB;9pC>3$4TZIA30g<0;9UJ+KRy0q87nGpcBapW6>*BZmKg`
zHqW6#>ffx@AZe~u$a*HL$fA$Iukz`sAJ|k>lG8Y<d)3NWExL$@dP|sAtr5e3lSFB8
zYu{GC>63LV3KK`yRP&a3WQD78&LMx^am<I*rHK%(O=kB_v%2lkyY2BisGVG2zcs7*
z2SCwJU^*f>XlPx_&G0~^(7^ZLArL%E+0e@Qep1298CMzrnHt1QrU=OLWWLQddcFm`
zQ5#8CKrPCRJ6TkE+0g>l!X$Q?@O~>#Dt66!`a-5~;W3QHN*B!G$=#`1%^r^hGZrC<
z>Y_YE;0CcY$XJrREgwa^sY0V;k31aluW5o?d+qn&SA-q;9!<9Q(&5tqb<oer_PrS>
z!;rq2$jzL4op$2+Rt(tXXiV%=4hFm@s$w7Z?dC+iC6+5r{`RR_+~+hExvTFzHp(<f
zs)(PUS8JY`E_)&z2<ZuxI30nkwXZ!;JM5JDt#VLIuo}ZBa=u^u%D!E!&Ml3duuF#W
z$(mWDsex}JV7za}A_J%7+(#Is4YJC)M6zBK+q>qgTFfezAvAhw5>DcW@}AwkwQ1Zu
zKjwscigDE}mRYgto`ORjNtYi{;sYIx=~Lmr+UQN%)?llAVjI77R*a#*tNy<jd#CQq
zy0BX}w(UHzZQH51Qn78@wr#Uw+qP}ns%Y2u?X`|p9lXC_wl?O_J^J0R4%o-f$07qh
zp<JACPVm)NR}B&SI8A`azHw2n*uqJT=eCqngD?${to}n$O|Tp12r(b>`J-%V`fh3X
zaA$Qz2Ja?;dF@;yHD-pChEbiG@wZ1{f}j2_JgW3Sp|B_(vJk%HY<e@c_c+<%vQb?h
zgQ4+pG~DRww>;BG`oY)<r@#6d4osBnZ+n3RuW--4iKifcp=8eHJE+t&MkU}FRMvY2
z33(0>EN|!6Sis6K%-c<W!g<k#GzyULxpi~+an8H>EIz0cg`4lrn-uRT2yv*DFstt_
zUSBxO(k))Agblj;xga<sgoT^lpI(2W-6p>$SGK$;9s!f<5W8s|*h#L+Lm%)IKV@>9
zbW0=uZ<E$@$$+*C3p3L-Mv!oYIDo-uG`ziGz=$ZAequzm_@RoZ=+yTEaI+`^HIx~r
zoHm%IMOY%Q1R`jsrgOg8lY5?AU!QXz(J5M9hGaL~Ii5t4a4=~}wQ69iySA`6^R4|6
zH6|=j>~57q;KngYjjXVQsmrJ=ag(oXjGmL^QI|3|xcMD$iT=^)6oA`Ug!4Y%c;EQ+
zaHiPsr=C=NDoRFA-0Vki(D8w=CPOJH+#lDQML{U{AdPD70&VBj4vSbf^)fSq;&TpY
zuu;K4n?ygFJ%&gefe3scav%$Otr9M|m8t&bOg)LQ&Q^N?Et`=(Rf;x}Fh;>uX`j8g
z)LTq38whuq$PE{f_AO%%?gR#4?KlNCWP*DmP~wDh;4EDKi5C}Kg;a)`iTwkSHCl2s
zfR-)GxrjQPN%iDWq<Wxzgm>4*%C%>(ZKxz%14Ig)M*39(>Lx5gI9PSXNZ&c@iZEm3
z%)1r`6s#^nL@4kv`V#4h4az~BQn3-;uo>Xu#xt8?(=6*f4h6YVIlXE{M98GVhkz$T
zB60lOpvphGK6a4_(V$8nRcFC+Y$e^W9$A%L0pDwPbet6|ivSFvhQ;XYo}J0%wW&?>
z+;oV$3aF$5^{IXF=TgC`e%qi*4=d;ddA!rnXt2JruyYV3K=5~-#ze}iVPONlk2>^T
z*Jz%eI=<amXKn53hqx^6pCzK2M9mKuJ+^Zd3>SiYvkt2I;AnGMHXm~C7kX=V%e5gj
zJ6dXo6GvP-e4f9_=I9Q+;cbqAVs+W22qB{wQ#7u+6MRWTKqcCHN78%}%(mwc2L_xg
zlb4sz`{dddBXjtb;8uO<7jOYd3(tb4@a(;5ulh&a>YZGMuVf?4NH2&{eM-7KtS+&N
zSm2Nn80vU_`kW&l2r5AHNpf<tW5eOg7X;F^W_ox&9X;CYRzvuRtF;ds2Idwimz&xH
z<l3zjFrSk_d&Z7Z3b+_ti><EhRd5vuLYb04gTVI$e{9ldcl-Amn`?DSMkfo-MpC47
zjK}JptzBJct~_f_zjvsPf|&RRf5Un4GH>PgH2qw^Nl1r<AuecbFO_#`hkWiCaqTEk
zKoyp|EX=(lSf>!0?6A8C=C%v9C6EvE2{G8(_}SzDJmA`QN+^HbjDR+KHkHLpj*f~U
zF3(g_2k~OYvE@tdA`!(0`TEIGHYK`wSu_kr!F$<}%W>qlw%XeOH!}L3`&w~+<gAml
zLCPZ8c14G%&iTkb<(v-MIXZ8d$ZXYS2_?_~CAzMQ|MQ*L^|6>HIl__^Y~Zpg{(<tM
ze8N-p7Qd4S@&mj~ALm~HDOQ$$HhVJu=RlCF|A}h*-$0OCjEXuv-dAB3uEy*lmMf!@
zAI^@v1Q7vbE|f}p19cfCJzcQ0l!-})YV`|;vyr0ob@E;vKp2baoQrSnj|WapUm1#L
zlEk0{oC}r`U@<q~K_1;K**uu^sX>W%gNSAC$Dupqwz{9J__}hwT)tP~^r<JP=a7I!
z#&E%NArgxx1!9rw_vzanEzVH>F23Z8C8tN9K|wm_G!n4<C@%g;sE4A&-{KkcAcVsb
z2wy4VZ3Le_716N|wLFMje*Jlp&2q1#ag%pNr16AY3+SS5?J5V*FMO`rG3ic`V%Rlt
zajLI7hd-7E(|wR_GDzxlhMVwJ4^mwhOPM|xNf`{|FxwuA(kd&C6giy;R+thh9vPfN
zVhK~s>l>wMMRe5g#gjaCiiXA6R?--Anqt6XFoOqQc%3Yof&XHRPN7}VYvLWY4yCd&
zxDrf%L>m<%L;12N7qg<5G0DpWYgG9a4CeoA*Dy6tSk7cJ>|fGx^AGQz6z^jR0X@OF
zs!`F=<XdKp;`YiYmC{a1zKcX2=kW>V-vil-_@Px~8GvcNmtSdTLd-!w;TE{F6g_@d
zu7fdU5DscrP?Sl-))H_I@@)zes#e+*(OrRCe1#jj`HF4FF?}U?Wj8P9X}Qp=Zy08g
zZCHPx-5YL1)E+7ei>CO~B?W|oYO$%Lt1u}xJ+sIfPo$7<%F<X3wW&HjY8luv8QpS8
zP9jVPm}~ggMKaM^8n81i%L5$D(lPdW5Xq0YXQ0_q(ma=vd9R%9B1g$gxIN%xc?FyL
zDKtI^#V~*rXnta26?2q*|7UkuI`u=rSp6e44F+Rq2U~+7WTFB;_>D|j=}+wvKj$dQ
z20O^gOD|C4k_4>nxTsSw>f_r2b)msKR0j`w$$FyG>5?$0Pz%Gxb*mHl3~Q7Q>3!L*
z171-O*A2XJ_E&<MJ^YK{*2K7Dj*8zdL(U1<{D8C#t(BmZo|6zE{SlMu$QD~`c4Eqc
zob^LmlRMfe#u;o`(rI1Ws={e73<!j#rQaI1z3{T67L!%ytbHv!VG?wMFEb~3MzCkd
z$R1q@Q~As_%^7O)s5_-&Q5r&g({@m3$6$7I_oZ(DEA~j+`GRsg$)!o^CFnO8>S;g@
z@qQ2lozt8p--I+|54^N;@f!%?ciF**`ckwUNRn9M5E)8S<i^}p)QAbcT{Z24bG-3T
zXvD719?cDEf7q+SGQ<WLy-2_Ls|qeSx!*J})gb2RVqs^~2C(FcT=|?bT5mHg9CD?#
zXyrD)rCOeskg&25+>8Y7!3^%MEH6}87(jp%`F81cqkZG&sg^7V%$m}9#FyRO%UDJ^
zH$RVXl0G*r!u&p%dn@(JP#b3INCedrbI(ZGMvbvTM6#3F48)4YryH0g7Fj+6ojC8{
zC{b(!TZAuHbK<5Zeog(;xrT^YVD2}8ZN8oUReoIjD=JfCsf3V9oqiB~-LzQycw~3+
zjG~0L7Li-ypN$SqFV_NSuMsf1hoQhiZrP79odcPl9L7Gj*O75x-Bym@#G1PM8Dt*&
zT!9fS_)k_l^dl4on!-_p(_ri^Am>IiMe?swfyAdP(ee~L^6%7>z=!;~?t;Wr^5w3y
zK>I)w?~d1$y!TyxUJ+5=oCP6@ALN_dYS_A${iB}LSn9mX##z7Y09t3>(W-ycwWP#5
zJFdkeX?_q1*W51-S4aqnbm}lm)b^x;_Ds*rMd*5>EA8rnL*BcynU(O8f_Ovb;DH#7
zdb@zN+)ch4lq)`gydi(H$7)J;3xs&CG&Lwk-4Ys$KC+6BN?_O3C}ENCdr$p&qVs#v
z1Ky0>-mnGkYS{ttu>F>}DXeRGX>`g$eiSsCw7(Eg8(6W&u2HEPpTYAZ)88|P+zoT1
zJ=gpOv*!saTf;=iL=Unq@{EPg`g<P_b|HUGnnf+96W{o*(@gg8Ngk-Co_<t3w=d1v
zi=m^(tPp3L(IuF>FUv?~H$ktdkxt|N%t|&rkSomJAgu*&)uy1i8dyd&MR6N^Dg0!4
zNK2W`iyZ}G7V3$`60CKbwI=bHe_WDryvg^>DVL`0Y86-7LRfU=YA*amH1rHpj!VI|
zjSO7pIYQjRw^XE7m8k*$%KbwIaKlbFHk};SH2@QC)OjIE>H4P=a^o@y)UGoJDX@P_
z$4Y`FP$VC7j9?{{bwa^a>Co9qqfGxohw{pKG!NEN{DLNWG|0y@7kRwv1$&*zF&7wd
z<2|b3OK$tCCnM$yym5O5=3hlJ>py&qEX?ez|FuXS()fG&asE${e9G=%9o3geY9g0j
zK50n9#*_KlD20$Dx}l|9q*`?8=KIM$3y+fbl$O=(g#--}Enax{v;zMy^cle5)Ai@<
z`TSkCQA5pyC7sUZ27_7?E1Ri$YTu>dwNZtu=gZ;!<;E9zQ+z9EDjfwR=v7DH`}uih
z!<DewwrEereC;WE?Dc-ek|xzwpws*NYwLaI`{|i`WnyFJ`z|!mbx103?J>Y!u@|x%
zNe5k#G5>1c1vW8>Z*;|^t>U`<vuQLHeX@`f%De1VMz^ZN<shfQ7+O2x=u=;_3A9RZ
zRScNLczi~*AuK>7Q@Rv~pdx=zkzy$^fW=spyS78*$JbVs%^&TiY5_<3m9n|&&D0Hx
zvggK_44(~~UWCx@_0D`Yg3fu^*{O(8g3VeGpHD68CzUIqg_84Z(!!JtksDELX!ib;
z>=xmrK;*1CQO1;uO>4bV2wiq#wHtdHDeLqx=ZfA5WY{4XnM9V`LL=E971!h4BV%MX
zw2jGhpVu!(u>2mL!l~`}6`rPCrf4m&Z$@a==qySJiMwAXHn3xtr!<)!d`^_RMKqGr
zMfK&qH58iwLdd@7K;<Hh$Fd;Zcvo>PWY+x9Lz&IA`o5OZdM=f&mlNJs$L3C+#*6%G
zjmw5}c7}s>$=mjgYXp&{*y#e24zvI%ohX7UTB?*EuCe4|E2pYQ6*lL4;5{}nk7Q?$
zyr$HKX^hO-2T1~8*`g&_sSQG{`iyLoT01SQo_u--vwNOJU+Wui>pA1z+SM87Svk^_
zB|vQnRGC}gV#$k9rVWZ;ACjwhPWnt!u{g2aRE2b-Fsgp5wcs`@%om95p47cHGv>Bz
zEWvWcUKTF8c5qq8W>IL9en=pPH!+@7c-g8fREx4b2u+}Hf1$z7b-ceJt*8{$edF3q
zz1%A{-3xAf5ch~^>(mh5>#tn3b@xIuNcZr-r5nSMUJv21)Mt!peN+%)C+t>RtR9_l
z+i_BWPtATZsv{xoK!gGtbzu1LpBPd?94spEdsydp+0>O6RTU#C>xCirhew7&U{VgW
zPL;SoqH8Pfo(m0P5XF#m7nt4dN+|8-<LOkX<Cr)GMyer8@;>VeCPy4h48*Wn+?BP;
zaYnR1BetIYn4!GwCuA|H>(PyxA*4h)0lEjtCOd%3v(8t8P&o0H&Vt>v5vX7Dxy>>T
zmU|W<p3ipl;B;IccWUIe*BgimHm1&<QbSEj`XLWMq4l@54D-z-EJysZ>M&2l7fi7{
z;du(<1WSKG8mZfB<6$KVV~ZBRK~)<w_FS#*wGHU0Fwab8%d;QE&U<N8CM}6HY*TIx
zHl(_&w*=UDj;NFyj<5C^pF{UQAY!cr<sc=PmHap|lg8b*M|y+Mt#tTg90pr#{IW%#
zjgfHrfev&_4r?dT;3<8eXDn}I1W8(Db5plgqqKokW6g-ei-|^Sy+2rMMIc;MJn9kC
zhPJ!tQL-W9i$_>GWH${{y`X3VI=&uSrnV<oWR_{@d6;#S!3mtLJ~q-j$8)1x=)VU$
zdgry&^ExzPcRr-6PNSe?{e$~vu1VZOx;!mpG!E`Q6!Q2P=)@<ySJc9t;#S^uOW!x!
zlwa6E#Ph%djGG^Z6SKTet3*!&#m|?oNsLGmlmz^uG|*x_lZQ}c{|7O#ENz@t!e{|J
z4-f&z6%Z@rOXV3~7=>iGM;z0~;xracw0<_zA8I_H8~()7RkO1=M`)p1Lc*!}+*b<-
zT0T+~w=#1BMAVD>*<UQ(N!9OvV@7`iiI7)#@d0-#YQzDnQ5gR5=gcl@r3UGCv36#0
z;<ydGX>BfX%5U&3b3H&-QYK{r5BmipS15Ql^R&As87f9z9@U*T4h7yWOq~eDC!{SQ
z28K0~wg!k5>)X*cK`g+cY1|MU;i1p^1s)|<q$GcoQ^&0&lVVJc9|kNR3(R07@8N_J
zod?HW<1_f;ktcNcA_V5X6_rH^NN>-p{b}u*#%5Pw|E1EWc^F2}X|SJ!RNI`Xn8a!v
z!k|1psZ(c3iXA~iZiNvKhe@!nvXL!(6L&qp_`JG;K=L6Wi*!~zasv&z*bG>kEi5Oj
zEb%$zMp0h~TMqVJ=|hFJ;(4T>4wl3wgP0fgqJpw&Il2ysIn&qv(d!?^O73sEz+^|7
zHy*xf;tL`)LJOGiYLyh%=cN}m31=XQGI)lft7(uhE4oC^Ip=S@bV2vGdJ)y4oQgNn
zVN4t{uc$oJ;vV9fJ(MjCln;jEsv86uwPGH2ALbem7DYE;pONvU>HzUPZV<A~l2|=@
z5V;bLj!Z;DwCg(P@9_54^+dDxHo3j>%wriCxn}I8|G5}7CVb+M0aU2y7hk(#R}W1^
zsXpuUXa!JR12dUJ<~AjiM2*T0*Ug4t<Wb#XCB@*z$Mg>@m}Ls!Ew~`K(4?-qew#r)
zr{p*_7Oqz`0m6alYv4wC)Kf)GOR;1%l^1JI@;8`c+6`Is06~7)_opLZtpp!lE)LL@
z7db6DD>pI<GdguwWCAuFaEM{zO$3`gah;VMy>TkT%!1)YL8+`OL%P%LJ+8Kf`qGsE
z2`K=|rBLLkP6NT)6Bo=t$m)yqdTH9c&Nzn*nh0%QPknz98EtJH$ZRLHfPs8J_tAa&
zgbJyNn-c99qsx-q4;BFJDdQljq^yewX=t;?H4RdoA=|$!RO{YS^*j|c=B%LLc1POX
z$fqxVSLRSqRrcfHPm*_*K&_UheQmwh<w2}K>QW=HFZC}Vfv}ZJM&!`G2pHfvyaf_S
zFtvwaU?{p3ODc%iM7HG4xS@nxEJ~4@Ioc8&nYs}8lNsJ(TBI9sBnUvgi{<Q3F4Fqh
zh17q6Xk_Tb4~?oe%0`dO_KQV{8=YaBR2&oAF4SrnGN!2(#dk2FtDvCTkslY-^w*)6
z`K#j7$k}frcXpQ3PqGk5=&~_m?$b7#Fzo=>pRI@_ffqT@Oz$NdiorwH-xqXtSyIQu
z$z2)j*7pEV;0Co#$zj-bM32rq4kQDtDy6j~L^Ivtm#@xhmjQKjqHfTOjBu{%g6XT4
zay380spi0l*aqro63GA%;a>14`m`PeI{dI_lZ8@D3=%9sT$94PUc{yAS_>%UT~J=D
zb9nsEQ{kXt`4u2T<Y910%#XFY@t(p!cfb%KB2UGDfsWg6EvM_w2s}HLQs(bgm9tc?
zL6SrSki^}s)J#sJFrv)F*S9QkB@k0TZP}n^iP)S2?YAJjljLL#CWI~2Hew+-Abavh
zt8>B_D%b+@$#x_1bZC<y*~<#z5$LmqqB6T-R%vO~)$~06H1Xk*Ii%+w3bzY)WUc)S
zPiC<&z|D$L+TB|1KY*1AmL{=Fyd?zqJGghl08>Z1?UFS|$7%q79U!|?(Kk^RcxV<Z
zi}NHEw8Zs4Zinui{ff}rZ5N8g=r63HQ4bEn@&HQ-R&R75mX_=CjdTE1E36ii?1Z-2
zxP#NdO1#v&r`)+GE!vS@v7(wcYu*l2st0HTkR#Cd=yc5sL-q1t<>%zju;1*^DA&V2
zA3K4$72X^886_p@539+*9fGL9n~b=@(mqaoZAy7YiwpKM`txYAN@}D-^c0P|Qqt*w
zBzfUd=fItD<d8w+zr4yRsE--sC6x|VdSSeJ_loT|Gfo^PVSjV*TKjV>DA@o~<u@jz
zG^5`8s%-sUn%`87TB740S7wutB`jPU?gaPW>Z<3jI9fx4m47g3rJHqGV84h+@p26|
zef>oj1@O6!;|ixglE{j_wyu_ZtM?xRM9hF{b!=<+z4s#F8+p3({IvMqbu|4zk<M>G
zGloVgePh1S4{eAYRagy!>tJE&_VOY}cd91%7I4D$q5|=&V5_MAW|t;)uiM6V`lKjD
zOEGtv6@>Ho=*%T6zTic4xVk}hWlML_xAc!2c6-IFMVI&zGyX-ZAivEz)tSWUaCK|d
zRHBZ-|B(L^<YAF6=ldwm&0(QPSt4xWdO`1bSD#KFTUm4DVRlN$kW)I-U~`v|_Z0JI
z;r&*jCH8f7u2k)$ya}&nV1nltX5RZitjL;=g!TgL=wY)KscWHuyKYk)FUQ?g8>dHZ
zHcg(vLA0KjaX>Q+=E3WX{B4{LqEhWu0a}_WE7|Z7T8+6_9|^?>xRrg3XQF2aM`7Rz
z9!KEs^jKr#p1+O`&Es+O>)=jS5srY|?hyHI*1+%0e0=lOeU2Ijj1L$BkS0H!%nIF`
z3v>i#GuhYANqFB4n_2Ri)o05kEAR2Rvs(d7lQ&%Q;9D>*!!L;M^n#kdA~|4OuS&2l
z;+6qD2E3Q{tm*3mbNA4<R%tlvUlkD>^Z(mz#>Bz;pA7?7>Y9In@n}9XHM51CL$p%y
zNsYL*@Yep|G|iKq(t$#C$)dINIiwt8A=fQ8aiR6}@!5lND+Q}Eq{^YZEjL<j!r3&T
z+%>U2Y4}}kLuRKzq$*lm4KgFjF=!DAMA>WoT#oLNoawT~1=lk>-^_g8ZwFQm{O@N&
zG<(jj&{Ab8c4cdaZ?dTPiW4L079jFcrsLiyK`>(`D|bfYWX_e1YD4JV5^o^737s($
z(6F2CofTIjm94|67o_wgA0FA0g=oKi87n0%F~N+MpG%JN<_6J%0sjS9<^v%sF_Yux
zO6FoqONjy*1wkx>$#@bx;E9K0GM0>?L4)i$jczN&YK+&8q@S-0v$`|w`D>a0>iLpg
z<&S`o>FzYC&(fF!>~vX0)n(?Wfe8-tz)6VF__RH3P(?%JLXX{W8=_4snM`)y{i2z0
zDJjAbQ?BFueXOF6SoOdZxQu~t#^aw2Hh%%X<7Ew{*IccD4G)#osGRc9O{waRL3$01
zsBJx{05Q!gHN~8p!iJUj1tko?t_o|keC3$VS)g;+XI0H}Mm}&?BoY(OS;2TkxA5)P
zk+YDp?>87AZ=B{x{9P-njZC4oDPtZ>r0`x}svLungGJ$PI;F1c1FMA$k*st!)B)uY
zLHwzD9eoX*4$a=&-d!ivsSSk!u~@C)+>>{_!hNT-K(vt4ygc(0an%02cCEMJh~=$O
z*%O7jFX;0o`qBVO$~~fwcmkXAs;u>h-gLX(Drxs-VkmkjOJP&wB}Imgi##|UTuxC<
zEOyF|jf@xlZ9gFJNE~RqIK4hF<2+rN=}lV#??VMl$*%-!QP3NaVE0by9gS`@tsq~(
z6?S)5vN8!NzVzpoQ3a*5e78@45SVDPoZT>B`@}qs!xOiNirpA{3m#Fx)A8%oa3%^c
zW1$BAHudn0I#8d#I5Gnk%0_xeA?0nL-LAzyR0`)Ig3wfWvgjID8|g4v88UE%iZ%Pa
zSI>eVfTM)~(iXg)4IE&Do_}Nviqy5E$+AAvR#_HRk9AG5B=>0~iKTMIN4=w_Gf~hk
zYjZ>NbodJG!ot2kz+#1@J?Px%i)S4^TgvI2skip0ZD?J9d2hSS=r*4EutBXsy1ITe
zBpi{0+HvI-Nim@75B-4~UNzz8^6ciQGsc;rZ*={4U~K1XYVC1|vt7xCfWxC0)DJGd
zY^Hk(`UfHYW)B?<#NM7?D2>l!+rO`z{n)*2WA6-UnbG73TiYUbDk*jG(X5Umtnl>o
z>2?h`Iwwz~JVR7QUL~hn{Kz$P$xcSHp+a)${*>6MzZkE^_PiMOc3x!rBB#37$t@hl
zHnvJzOpT5<YVl*6TFJEk<$=<_<8{zQ#xq!*a4-ho$eV;}B|$zw=RvtSYW@Nu&K(6f
znPAs?g6}Z~U)fkwU=(Dloep8GSSd0`4v|$?+0oZYK9*Lb+dS=Bq&rXpgxzu<u0~0V
zX{aD8p}1Frn8#flh%Z@1+ZE-NbS?ad&7lKAY_zH(yy6dksWyR})kNYRz(Q=L?l2`J
znV<!0J<Y089wN**8>d|)<K>!Bx-kutB$i7g<0B3b$2i|@9p`G7C~e+6gc>;PsA~Ij
zJv`kOw@A(8yD|lI3*P$|sDTme7AIkgzN8ajD2O=;-T&%bHHN|~#!hveq~~Nbb}M76
zu)D|v(LEM}kc)C;kqg}c8p5YOkLL<pE_E$KxS$)gY^G9h1w@pCZn^W{|FyMsruhQl
z6&k>=_ri>IaCQw0T8wJa^kKj&qYA#_=99Lwe8zy{puS}>@`+!;xH5AyBjiO}PGB8%
zBtHv_W8tnw7#cZk9w-(B*(#c^d(Ml_kPk=}2-g<KStt>egZnvQN`Yt6FFysfd8Aoj
zRB~5L`JwxgFSjXbptsJ0!1Vm4firwfs#@SqCx0XEZnEs}k~+8aOw!o_M`i#*itMhP
zIBG5LkHM!j^688JLP0Gwt1k|PS+lbE#y@OhLWs9b7(Z4a`9uBVnDZCA=oOrg<|4%I
zw?B}Ax(m!?aE|@N7kQ*~+`#xv(;EL6(d>S<%R&1c^dE_BemxF^(OXDQrx3j#A*)5U
z2j@i9NvN-bC%`tF0CU3*7{(<bl;4|ZNA>+H2G+~1Q}4LOZJK>Ut?Knl{u>4RP4q+R
zo*=YwgN_TW^`dvhF^2et52VBS(i~>H<HTuGiROw{Uo}83GyKtOe1p8Ay<KZv^EThB
zg`U_bdw1TRFCTf^+Psbz+uPvRFW0ui?>BvtN7u9hzD5FYNrYu&+b4(C+V%@*1L<3q
z+d^u%yuf$gAghF8!T&rY|M6_d%*^qh`R`lmTZsqF@V+}e!nPDGEpP%)_Nh)L3vHw;
zS`25=x$SFf@z(3hS+rCMms?%z-=2a231n3(@9E78m@**Hw-P{Lz-b;du^v0QUOaqn
z=R8!MX@kIsooDl?{-s8OWh&cFd)e>Ahl!DAJKcX`=2|(Thlr4){E>JrY6P-AdA(oE
z9sd93no{b#ubITOchlTyelNF|y-$G|^8Clr8{a=>huI3DehHk1((VVQP50M0e;!q;
zT^AB{<>5@undG5_^+6*PiHjbOGf_n_{R7_ON5G&-_I|Ew=6oEK1Ky=gXrGA{bO9Q4
zEK&iKqMx53-0sWFhNy}+NscH`;W8h5R5D$Hu*I3o3%J=FQi&K`>#n=tQ3~7VQt<$j
z95@n`{;V?yqaQ0C4%Gd<JDmB)-ldi6k`07ID7im48q`kVbZh2mFIKml^az+1f#g=E
zUm(UWXraU-28ikjZ`R-<!xcm@0xo{6m|rQJk=DGJ3FaxfqBSyu$U^|f^NN4{WYdq=
zwVGIvy_qCvL-9#h858HoOIQg3c&DMgbC2Cx#<PawO5TIl7r7VyT8H6bJCNM6x`3dV
zW>re4khZZX!e+zGSjr$;tgY~r=`l0tBV;Uzf<jfCN#lj%G0b?04+WyTr(^xRU>{1w
zF(v10-8_D9u{h-b=QjnN%QKfs|L)E<Vzb%iHi1m7;L}03m+xoUN4e9%PG`P9k@^}h
z6A$@nc4-vT;P}bC!A&YyXVaL@Nce1E4a|x)lB%`w*`Emxmyq!0)v;#MdvJQ8T3@}@
zjmgPEqspnIYOc7Y#L5YcDatfx6;0cEP|t8?nq7QRPzzccSIy3bfKB^s9t+ah+n!&O
zX`OuVLO<5LKB&eyzTh%nKU%*}J(c~@#TVsdt+xKaIy{D@Z|^5)D%QZxZVx$!RT{ep
zN2}j5?Sn%#tvSV&v(&!cBzI^6(T%IU7TqgluVVU1W4+&(Gp~v3?CH6-Ebp;_WG_m5
zn>?5PTZT{G*FW76&ig$ue5Xf2Z0juF$3|`@K5T%BEUSD^gGKvjh<##ShS6qStGa$_
z56~zfLl*ANiPkXtvwZ<!3FXfAwsN3F_IL^yHi8mBZMRpZPBGj8dQ>_^LqZ>?`Yot-
zru9n-!e8J2k#@>iX)h0UHpDAgg96APAjo!uj(BMe7lNDoXdkN+i*E7n`&i0rRGg=q
zh}7hf@m+HaOo-kg*UD~WF_;!iKL9LR8H8SHP=}Ht56Rab!4-;38nEDZKXwG4HKxA}
zP!V+-y;HJH{)OmJQTI+QX|i87we)qA<Rt)fgLVx)MtqZGKl-diT(|H50jkrI=jwx_
za6ZF)@Nq24=M`>v&%QCit!=$q;?>GI;RaS<LqI~WQhPIdrMvre^~sZPm8;8(VQ8G-
z4+>R7*tLNP0`@jS#LIE4CwF;a$V(9?S{~{a<9l7C#A)?cI7{Nr%WJr0dKnsQgW>N<
zhvPj_tY1@m&O9OVe7AF+`%WwPt&vkhb4RN^hPk+;eL_hC*)bVuk!X~v-lmoTWV{cA
z@X@i<e~}XqHS-M|R}WldzaN*38+l~{Gzsv|!R49weha_9zAqBkE2N^I(}sVvENmU2
zvh^C8({HR|yy0`{3XcP$wy9}(SiZICQFrk_OT(f)uwfs0JXctTtn10tdjTD#3<YDv
z%6R9%Ygn*F<n5>+EQ&8d!e4gSEFr7cosha-xd5Y?2KR9_zo~mnfnHz;3nAXPwF8u1
z`RgkvDv!!R{C)sAA6Z<ShX|l4#v%=S087M>v3&^UNxVj?*Tg0icb?w9jXanQ3Nqls
z5gL%0qK$Q@>8&`Ca>)|}p$i29xvbz`E_4%<rChBYne#`tJsi4&051eE9v40VQD1^&
zQrFYL`Rh%PC@)3Cz5lv!>K6A1Kqv7zy(kH2?`t;SWtox?mbF(&Cl4&vL6>AJ7V`F7
z+gOLDS{9>qATtVYnt;KZUMy2-$4O?Q)T(4nH<qxZCEp9XGowVEezb2Y_hTdrm2J))
z=<R1S6KHvdgln}07*Zi51y5fzwh2{i(H>l}E~n(V&<a?%VsHG>^XuDY8PQZCA`!kN
z(v-J{WS!O@P?Ri~O^$hY+M~A;k~HJoZzt54l%!pXo1p%wu$Y+FJQZZS%RP*W-eTOD
zPrAxzKVbP1U>580yvvyYRX@@<9=DG#2J62NG((igwe>QD!K5^)y;p4uP7M_Wzj}F0
z6qN=_npWfL(f@H?qHZV;A?a3PvvLDB4p*%8Y12*;p#0Zt0JT>Qzer90X%JX?heYs)
zaL*|-SZp!##ZP${+zO96vu|GAB{yVm{@7yF2p>fCDz|?1saB&Bw2!UZtS8$^5Y-=`
zRq^>$XM5aNwSF$zWU$mdXZQ##@d$b4b*&Z@PK+YBh(l|#K<py#bj(YRV({*VIv^V=
zwJu&Q#<XG_Xo&Umkbt9LFeI(|!Y1_PwNeooiwNU)XvR(;j#pM-8gQW{6}V3}dhr<<
z4q`_0;Jh=pGVU)wwCl)89zGbG-wH9e+B>E4(!!1(e|Sr%G)WLPl7Ci5!|m)M^JufZ
za)|RpU5TX~^#c-QS)4Xq4M5-!<#ty5#y%Rz=^C7_r<e9pxIoJSAU|0rp6=>E1p<E`
zv!E!iWXbOtm<#@XmjRj^Zx(Vb^SpWNmWj09DqWp;WV?pTTzQQJnWS^xe<b>JzW8{y
zbg!kcMu8i5bM1HGmUrfFq3<R6o+PSB=!5{qz52@HtW^ZQ%yih$o2lOz@~M8jD9I50
zazUW-dPxPAWyG)nccpwmGAz6^vnnp4wFa;eHsE?;s9sW!Eii1Ujm-7eW=DhUJb!0n
zoeS%>e1|omc-|fQ=A%QO2nWrzqC(r?N!<5rg~9DfI<AB^w)3kN-ua2ze<Wd$22RQi
z4)mY9RE~fD83U&<wL^881z4S)BWu1LGZnMq7k<GmmkD%xo5(+Z{~#jV=@vZW_2~zG
z$G?xBtN7P_iTxiIHYRq)|C~bie;*ZIYG${sCQaPpB`dTctwfXoeGzmIniyCSRh+`K
z6NyF?LUW(b_&TJK^@*!78|bS#!V%;%Z*JG2ecOuz+o97l1Jh$vefgLLamv^5w!Fh#
zy(dub{v26usoFTi5gvhl2iG+MpX{w}-xIa2Y`i>NPwEdzgaAA{Km{UG-(m?uvlk-_
zI6O!V-?8#dc`iKy-;>YkTcwB}-E-IoRzDbotls0Wxda^xL9ff1;y4mf|AgU9$-{d-
z;DQDP;(|3u0GK5-D>+hnlk|}kDu#uO6R8O(9F@?XD(bd?H<keZu5bTngr=bp6!*rV
zm+6}Pq5j8FXtsG)meM|lox>~}YO;^%ns*k@;KoY<-WcQW=bi>Hrm&rl_`zFeRn}*E
z^TKPws0<XeIGiDex>dxdw^%;9v5t<G4p?8oUQDyyfgoZCJJzb7yz~uTnW}7|p?p|+
zsrSBHTDKgA^fk;5mKL$=NP+{vgL+EX$SipLgkB9PQ%UZu3^w(=(G($9q|}9`H8?1T
zH3#pde49cBgCf`W0u@zl8t?7RCQU9TA+9E`W+A97SvROWUg2KEuJ*SN-Ns!lg(vo`
zmhmi1W=VuRF!%`fWf;VLsbJp%I9K@51maTYv7AD*M%D8#P@rHW$U!!o8;f-&?V}On
z*)BZ+{I&McEh8{#9BZ)ku1-%B2)A2(zo8{=qiL|4V9rv5=Y~s8>u@$gqPDX@wG2qR
z<+Wb&8%^$uTdT*0AHq=qmhiqeK{I{!k6X+5gH011G?0Sn(N4WM@!|7J17rXXP|BJl
z${avKr{LioHu3WA%1blpH1$n?1*{#GkIuK!`_!@g20~YTyt%UrXcQJ}7^pO|Y0A_s
z`!&2-UZE2&n|`^qGw@|;V|{xA9bciKQr!A9tVms$3=0(tk80=f#3PQ{nY)YO$OZBA
zFU52tM^7I`>-1wj;*K>3|KIVH7j?pA$C%H}Am9OJvS^TDLL|dT6wUaCbR(|UdF$h~
z=WL*qPEbMmgl}jFrA<X%+u*DPl_)h|^G7K;OYV*XmUKTs0#ccs#sZ{lk<C!y?udxz
zmg_dG*s0)l>Oq<rN6Ev_<u7C7?^lND6=^(lAK@l@UyO0R??F7scVQQ6!Mm8fU*lDI
z+9doT73oXZu?O>krcw`RqhS|uU-2}>dOfT$D;|HsKC?3}v`|&FVMM-R1-d7@%e$l=
z)N{LhnhhJIL?ym0dM$2ZW<(}e^mR6)2=z&;g@GAa!h3At*&L#m#(KDoMXKO2kAort
zEmm?e!2eBDq??I2c2QYj=jx^1c(}-9+nOD3vUj4fp+`1ms~(~2O)mV|Q)q4RlM7vj
zBp-2%TGg)DHoi=%Av~X1A>%UM#)$tq@qYQTk_{o1>W{PlJaHst?0jof)7ij_#?22^
zv7~i%6|BF&4E#MssGoBhO;{?CBBH`OKrJD)()GnL3WL=~%Db@5V^CH7RA(h?q&ewl
z)<!LweG4#wC2?54g4?KNGv0R(Nd|j|n{poH5Q3R7==Yu4#MGzR#KQdIJ*CyB8nSp%
zguN}6)JYdu*ZYU*^xua^t5cWl7Oc=Q=^Pj985c6I=tY4_cAF5jM~+mdyFCDno)(y8
zYm*5bZYi54Ylq>NN#d}mifD01UcZ%4JbtkAuWwLbDzsR4dZUaBhF{Y3c((^vGs8$B
z^)>&ti+&!{-~nnty^kvm<7>a@!d<2yczr(WH+yfa;z%AN*iQEq(mT6ftwp%GOBdvg
z`<|znRV*qGEU<<GGm@Yb?3ujCm13EJ4pkHtJ8i{s0xB7CmZF1y?wMi?wZfZHL3c>v
zNX*9VgOBH58%dvhur}l$hqX<T|4C!qF=cu{6)4>}9XrJQv)<2(eY{U>>BECPV_NAw
zsq}u+i+Q}kv&(wT7OSRR$QQri&k~7}&J0RZvHRSwVtk%<^;l0l2bfV{=ItZ&XY$(L
z=d2jgH#)Y``8);+vt?#X4_|(drHQDEpwt{NHs1^s`8Nn&qNLWW0nNH_Kl4lRz!eF3
zL4~0!H4Z8ge-w5^TZlO$SG-nMoR<OFGpJN4SVvh^p-LQVsFGOnTkEfq;uP~VkM6YD
zN@W^+Kw4SNhW+zg`ez-_^`9~3#cErBG3F@$TgS)GHBm>Pk?~z$S$lF>C9>Mv_cn$4
zrI>4m@oR;zSHW(6`-n^{q;yR~obuUo;wcThy-q3E*TU}aj%4;!*>ryk-R6bCq>t(~
z!MDOG5f!Qx^;w{P;M~j7el`3#c@d~>O$)Chj^s473rlSO9@z?A8W<U3mO5ahDi~r;
zwtv-EB;DH@=@=Oa9clUQ)-)p4>hk`c&T&Oegs@jJlFEQcHSB@yZhpM0x%N=o`tUJy
zjseB6BN(#tPZ~nTB&N*7oj73zf8*8T9d{sj<X+wW`1qBag_~YSJlf}#;7P-RJjN^X
z{)uUQ@9TZsA5R-$xObBwUa#j*nLsHg<PX8$wr=V4Yr)znx(LV4>zc+8gLaW_`lVgB
zmbVDrcm+?qP-^+r5y5-uX>ikp4@D!{Gfx!xdJ}oe%}s<Ea;Z`r_v)pr-O_a#;o0F$
z>VZ`rBpzLH7<TcajrmzNXSRh4_ToIUy3%{a+AjZ6W8v4uoQifN-lAEzsS;E{@`$^e
zL_!5bwuW&9MRb%nJ_35#G5O4vn(b+!hJV=4)%EsUoC(2ES`?{>_5ByokY}NK3;V3)
zPDDSg740{0sU@4Z)Ac`+0q1sgPnYM65;v5dj;?jIUNXAMoN(<mDtu?zQ;%{BVM218
zU69sr+Zu<rO%{R{tD2yVws3N|Phy3>naF(b<Xih?cB?9ymZx+>3YzDdT0k!VA%tKN
zE7?W1%@cX!1H@~)NLUk$fLhSTd_k)D^e^IYW+?npfe71SCCz~%cJd4pt4bwB1)z%-
z&S)y1^cVS*5Ul($v7q|erLW)3ERw(3VoH8p%?d7`!sCs{g?}}O@;RB^rq}bvh|*V6
z<K|)qEJk-_A4|P}jsLhXsex0XXRRTpO!(<p#|{dN>4zM^?>7vsxUV;MPf3L`_&e63
z<>6eFZZK=+ASG?f3~wddDg4J67pr|4g**MpY?274pLV^1z|SpCfW&uxhDepUYLhXa
zO%{rntT=77g{xtehmlQ^A|-E{PkgAMxi}i00?<;^1=3ZMIC(jNHh1)ipM-jV#g{f>
z==dqDOfwR5OU|$OA0LwTpfc^78*0C`4N7(1c19wE{k`mq_B~bkP@uAT(<2!4`cDY1
zy@zH&BJdbG^_;2_7!so-^xPlm-YaZ%Ux?Glj}#K2c(-ln9-#t!2}&s%htB?Hant16
z*#Y8%H8q`P&#L;8U96US9h_Q*wzhBtFUO3Bftrsefkh*Qr%_eU#dy_9B10#LLDyk1
z##l*CGbN<`<B7qNkpKvz+bYF<5s<w77B(V~al<$6)#o^GU<AD^_zrk3s|Sk^hFGpM
z!C=sLn5kn%SU<4nO)>qiBVjA<!al^H2+Jn1(}<Z=d{Kha_~?%GZ%70?C-DydRH82|
z!OXV<s_1eM19cszf@`HLALU*+>#ozT?ZcsIL3Q?+`i_xi)LL)O;p3-CcxCdasY)|+
z1&bON6VGF-qtuzA$P(=Pzz*<^t~SNUp#&^Fdr(iDmdPjwiN~?@S0155{WS5QgJ>aK
z?LPM`h`T{*bS0EDfg2nf*1Sw)vz*<b@r52SBrK+ttn9t4H)5>aspFS+D@|dOl|^(9
zkEjz)XYZ#A8wTXbi27=4!O?|!m)pvHW&ot|ewGi_XT{sUW@kMMfXq^cKH`4)*?$=i
zKS&A_vTm63Pz(?FUa@H_iRAMUa}D9hF_8h-G}?bZu(lk5^}<m{2q`<PSSt*#O*DEe
zQ$2QZ(__i>Yu2yo=n7Fu%AR7^yU#F2u@Q-b=aMYu(_-ua#s5GnZqMV*f_4vTV%_)4
zlfj_RH@PKeBnT3F-gApl!L71D^jMt%uJN>i+!)KRuSj>uo<gDAV>ZARNRAO5rGkUJ
z)Q~qBHhxzVEGE;iZOz6xAgPxTs}Ce98;43egq7R?c;b~EO?5(cV_0h$T^l=;be!>X
zzz`7lYK3z`Q69U{s)NXz`m6KHvr{&0l&9x43r-R0Fj)BI;n*RQ)lCI?nM|`jL6cF?
z+HbDiQ5jeB3Y8eIx7wEkgNvNS<un~I7ApY}dp?QsPQE2^8`Zk4xS2Z_JxGGpg3nI?
z8kiT7O{(~4+>(mGEG&<7B;%Ca0&clD5G<}VIP>z^StYB`mjQ7P<&w+0VM;Ber7~;?
zIAnJrVv@>1LS_5-F;1b05ScJTGJ*KnQ1;iv&jot5cs3HQ_RnNVrN>#)VgfWkse3qo
z^HJ=*xkZ1)mq7na2$INhjAGhA&a#E_(XQu=8F+!*+7|F>Xfm9|yGw&%y7FgF=Mrsj
z<dNz3e7PSS^&gVJEwMOJMBtnbW28>nS#N*eeA11RsY&hsa;de!QyBk>Oq(tOsn1d{
zA|RH!&Iy#8z|Z2-lUryb4ju`lmk@!<ljgMq)OA?{<C&IfMk6BV5n32qQNklX$GQel
zC1UStd6hUlXO5=DnR>$yVJF5<rc2o0TFR75tvP@1r2=z3rX=je_Bx1Bjg(AFlL#7a
zJ&lo?LY7dF`OB9$hO|AE_Mi=1l~_<(pFupF2D^mHPz$5?c6gWl5kI=uHb>HT$=K!_
zuxD&{U%MLvq$*}9qCP<buHl;^DMT>o@7PoevZxi2Qw(o(;a*co5kv3nPCX8IW+K*i
z-%SkVOI#YCDnop%d0(1g&*!G+*WZd$W;KlywtXLlC5BIdu#koQ%J6C>b<K~?lC8(=
z86~UC=e}CYXA>MeE+$?zgDJ=6p>k?WGrTpny52v&1_nB8$vI|-F-}Y~uNbm-_BqR?
z81mZIwP{!r#Do>H-&jBC_vZz?+{H8FH_VPIxN;rl6>-R_OjAlUw|!smt5$WX<Gf!F
z<hX0yw~(Zg1~$B^(ZNhL?=L=goRBF~X^l<ap&E88Rn|G15PIwSe*T)4FZ#B$4uBIy
zW|A}O(Gqr|<$*e@_t|2h4GBmkHSM4<l>WArBHvD$$7Iv{K?TZcKsA<gFR?$vAPt%J
zS;V;gT7Kss&iVdJ{$Swz*cYNug5qD7sK2tu|HG$Y<M=-YBEbIy&|5hfI{+BOt@NFY
zMU4$@jf`RV`C<MKsnBh`Dp}ih1ArBL^FY(;RFb}L+c_a--BhkYv(#};9=SRwKCrvy
zrvCL=D^<y^lmT}F6SV#CdSnL&r2^!?F?3>O?|hJtR>T<niDiU~Gzfq(4C{w7B6D;2
zK!dP56K8U`CamtMx=s;jsQJaAf2)BijwDtmE|&lZA1(l2_XTe?6iN%~eL@XeSWU9V
zK9O08jcX_|DGK<_MYmT5K_J?lLb4mp7i<n=A0)JFc7&OK!Q?`Kj!gK(IuuhO0U?(a
zeW+!VUM4H`S2M4B%1cA*DyJOP;rCXJHBzW2ia-x~$SyxiS;c-fmQ8C)accK-XrRs|
zQd%n6`8Bl<Ca!Omq0Q%YGxZ!sId%AKZ2otP%-CvC`F+jb#WoEK7{+QP2=O_`-tc+-
z^6!$j2@*%;A%dJ%pWZNeXa+G4LIh>t#U-LCp@*z+W9j2pn@5|wzn5g_!#;1;^!PKW
zu@)`l2I*3PPHr>+)zVS|h)}J^9dJQ*gkeT|n)Z<po4Wsc5hM>R>zXRYd+f<v=&3?e
zp*)B;md8opsGkhJ%t+6Ao3e^x5&MzMS48|n#AGzo<VV}*9^)N_z<b`n!teAAY{#98
zcNYwO8&0sQdlsLQT>;AO&nCOk_$4v*hwHR8Ra`!r7w@EzKb-o_S@oJqsC&%5WAUHQ
z{b13E?_etPHj$8UO_pb9;F?KB-U56|Q?k7sL@?8#lQQxWE?M;z%`@^fgfiZnshvZA
zbe(wKz0a2JIL_TWr!l^PbkB|u|2ZfBVSZ=%Yjyjtef{1_6E^*fNS!A%UQX0Qui39@
z#Ca;_kqUCf=+fb!@c^N4&>r1DV78wV=D#``8X5=;Yay^?J4Y6Aqqsm}sHDgwVS7oD
zw1_9cB^_)1Xp_E7|L|g={`Tn&gFz-4Ovq9kD0NGTIrb3*4^bf!KtdTvuc3@Ou}_Ff
zIP<&=#Vc88SwYPWWq_%^OO5gqg|i<Q6cj<DI$QQqp!QhMLxzgdw`|w?Wn!MP;jQzC
zyy~D>nY>kcT2Wt?xrGn%B>;?x%zJ)cxnW}q<_D0a0|Es}6I|Tx7s0KH$u4%k1to{_
z+PJVt%W{C3SYE1FZ!NNPzSbCtuGX5eYLroZBC;HxGz%TY$rZKRy1L}iifg*qBqtw>
zl!>U;NUyJ6|JuGZU{wUvi%Yt&|M3U_(^4vuJbmPNvUi`QQRc9zILXnYRGE8n%oBho
zEZ*kW?#QhF=TG{V^^o)bh^YTz<M=uK12#Yq31YXS$3RX$6#r%7uXY=+kL0wf`8Vb7
z2MqYbw{ihkwhqLxyD4cSU@vWy&6Q{jrbbYPbe<~+B5A>_30x#<+Y^dSYhl7Wt6BJj
znQp?t6Sh=L-Dt!@RoGm71VU6!GUgY1XDO%4K<j#UxZ!1GYg8uLv<SYgBL(FcWw;>L
z8{Kqj%qNNJOttMv^v;gDKD^swk$JZbbe9r<6`Qn4c;Lenf#?k=TY)F{U!%bJFR=Lk
zovHs}6j&4_e=#c9oiO(bfdUy)g75wFBmc(0GXK~0u%sbldm#S5J{L())zY0<5rj-V
zie8XD6DA0Z)pR_w@MbQV_<G5;yRSc0*A_}v5eZk}sIv6rYv<L~)hEA}N+LVJwLhGM
zC}s$&8M@zhR|By7(-R^uHiBni&G8kdX{%Qv<<s*H?^iB-;bbs!Q=+7EKIa>?>u55!
zjuL1xKI>_gk1A*gVL!goY@2_o@4Pkbc*hy+g)L~k<74IHNzQ?5LW!|jF(i|(_lMxF
z-u~H6jAQSxydUpL|J^neU*ApVk~(Q3))kRB-kdv+&W2zuk=4hbsix0-SY2LYC+C{$
z`pSX=lo0FQG{ZB)jIz1s#={+EnjMkjij<TjiskRM12LTepVQM38?aj9uul9P{(ISY
z<7zy!>jd#=C=Cae_Ii|yL^#}OgRXiviUEtPpZPeIvY#(kC@{CYfW<;#5QI)TO~sx7
z)5+$YjAXoZ$bt2#ZAT@FqbZ1a`{s{vOcHTSyU*~0C8y~F1gn`pjZ|8FxomYz(oizV
zc5kA&{L5U%Gz@;0zhz>zmrpoNX*e6LHX1Gb;?`#4&~=@f#L<a%jIDc@c(*60W0qv#
zx#}cwd<)H4(T)Enj9i%m$#dg$4(tuKhOu@a=wBUL!fPP)BvXqXyz`6YPmV6CD9m+8
z?%S)Ou5E_1;uDy-&)Vg>BQ+9IV5<{VM^Bd>x4zIGwYDRvek}~F|FrzRN;j=O6)JSk
zVBj(s9K+EbyC?`}Pb|`$>>6hYqeRpqr#q<wlvtn2^k}%Eo?>OvAT$kDvk@BaK<2qy
zJf_O3gKn%kBPcB7OJ0at^hA}XdH@DE<xIwDFP<sV9e{~+(!H{f{{5v`Nquc=B+oaj
zQtyXI5`&3SKsJQFFM%}PrQ<9J9_(TWZgpfkU5DvS@n**uHoBTLtz9myZzvR<E@A#r
z8<cgJqYFxYA|3o3zhIBe#5V4sxh+=Yl4eFdHTv|DdY}n0W7cmsd)!)dxgpKwy#Pt`
zd34?EY!sc|2fExeR4!&tayTXrhv}Z5=o(sr0@7}lHn4aeldi@-%II`o^<#zh1=<o1
zHj(WEdsE#YA}^NhAisDoAKGa;>dmAGgUHil1-A|BLyh>kpswqeyt~2W`(yW8tdmon
zKhC<{12Ba|wr@d!xzJyIfF!<XUW;?jKkpNa=hy~D!3f{Zl>btRg7`Qq8RM+ogJO*Z
z{h8lakD#7O;i@V=xruqwk2HatDmRra?MaQe-N^NYNk^IV*wW@x0F#?6SLs3?ShU8?
z<CV$Z_E8ukRu;J<Bv_OLiabm>z}uZ;w&T<lq+qQVWwwN)5li0M{Fj0KWMxFLY8;0q
z0zm1`TY*M+_aI?HozbSQYY1uZNa!IKpT`g?Yq$qtibK$jOldpl&ULW;0Y7;YPRNH<
zvBFUMpKsl3X+OHlGnRY@!eyWJdtib3PA4nH{Vs{?zPwr?97#L9N|IcW69*`yagJ?~
zTj@NHvvKVi0H{J~XAM^nsi3_XDgPg1@7SCP7j|vN>e%ksw#|-h+qRP{wr$(Copfy5
zw%&ebrfTNHT{X{N*j2ms+Us2BIuIiEEO@m2EN_di?zx?**??N&`Jc+*R_bYlT3ggb
zq5IX95Q@&9m})b)_CJG)6cL4BLm}A_hp-y{eTG(J3S>u@FjeiMIEpzSeqW;igR$%#
zNYL!AN<)jEH32rRb*ZFc_OrOcvb7}W7<E6vSU^V0^xZeF88gN&LqFTiC-pPpdO?YJ
za57nc$ug7uqyRD$lL_!S%lT#TBLvk07UQh(9-vA+W7NjJg&K|f2We^(I0p6N#UWG)
zFr_BRtH3-)gqY!U2{i1Q=qOo^FvrSHjTtmC<nB%=+b*aPZS$KN@1M}du-scw<?A&#
zh7ufz)U!NRw5F^*SQ;~p1ctJeilKR<I1KQc1(!w5_PDqv*_CBx4Vo=jZL)ohSi1q9
zY#)4z%<>J~)xtZWg56-8s7XXB$`gL3c!@~k3SNUlcG{NZtz?SaS2>rfyyRMf4x?g?
z6qzgWra`E(li_&3sE)Y8wGqf;Y==H@rmE&4TvwdBJo|5<YoWoRfE?{*95*$D!6rBf
z<f#&B@*>CEG;Y-J#KL^w^sRL+M|{}R)Px#(HVh!=MD<}KJ;9b(6W{bQkb%>s4DxUa
zj@RJ9hs@Xv-elH&Zw79RttLYCCdjLDWUPcVh{~kZve*P)$kX&42_^|*+tNX0N-Q{n
zB}<#9VLy+gQ5Ig93=7j5@oyXIP41wO_n?FOTRcZzWbiMqYN{S8KJ@^#K(H@LiW}ak
zb}$Sr&siXCL9pQPi44#7AWX1U1SY~MGRxX{*(r|p&Er)SW4|GD5UTKBq*o<Ol|dBc
zEqaZ`zVI*cr+(;DZI*6Yv}KxbXJV6i!JRj+jL1E~0zzPxQsE)Jo2ECe-LW~f&}rb%
zB^L(#XE7l7*)p$saL!8>+_3DZFuT|@ScaIdlmfCsURl2Jb<(2@9Vv{v;~x{*JvPMh
zK=s@m4u#N|@)ui?A^d<@U?j$z4y$}H%DDOJUc`6CrDZp*mYldO8#^$#y9eg{^HD5o
z#Wz;@fTFiVrmF)7yr!bR0^6_3VWMBnd-!Lw(^0sJo^vxhBDp`)ZYt6oS2{l}*tR`_
z7W<e;Gb)I^P4g<;koY*p_vkNX6H~nMW5fdcu67qQITbVHKVb4OG_%J;9zuZWN`pT$
zZl6`mXS>Dw4CzIxNK5rQ2x>@4ppU>ieV=U1_;zR@g|H(K@?gQ!JX*DH#a+jDpoej<
zM2;lB4bd+o4E2GKopbvKsv>*!^mw#!YYZsj874D;Y*jNO<$C9bv3(!jbo(4~2BxBS
zvMFO{Du+iS{r3ETib4+-U(}uk{KK&_+dR*8r|e8Wg*nq_P#9?XWTL|p+%`injXy3Q
zG$dHXouR|Pgr??^tB0w1Up@q_A-UuMw}gn(DW=$;YaSk$`QONF?%Jaj>0Hlp(TU-C
zV5^ZiL*r|4`?Z&`6TL_ofGkpvt}7fPBtJ+^LwG$GU~4I-Q`?V^p6^~x?;SV2BaVY|
z*lBT02y>jy!`1uHCDx=E`L17nUJmBDT0qDyn?U{}_=3wzb2{O=67ZwZYHaMlyEu3U
zgk8$hcs)^|g>$27zZjN{u*G!rdSO@cGB`{VyLpb<S2LVRd&>Rb6+&D3Q$75Wj&;5{
zi&o%QV$>5O;A$tNKWib}WAYHZpGJ&BKHfk%_=Yq&egPWuk|b0yQ>8<8KmF=+-jx1J
zy>b3WCZCCk`G2k6wlpVQ|CRV}T~N0b$!%itK<hL9niV;4m*pu`=h7I;0bi0CpL!Iw
zY^dgl^_v|IvJRAKHXgp?;N)Xl6G2ib>nQ60&oEuVn%2$nc5}F&Mo=_PSq;Qk8&)BU
zkRA_|FK>txeP+W5;rlq+AGPz<UxXjRTo5sjk?dN0@pE@@b$CBsi=xi>^(SXqS@5AT
zclPdfY4+}IEpuU~_x*c+DOYar<^iL2_WSj`G}ZPS1)CN-o%sTFx{YiET~KEsC3TW&
zs$o+*G}E*dhCQ>=zWUp>QA6#b1)9Is^}<8TZh22_jiWi8MTBVzKGP*@QwCFYYSDfe
zd@su7R}VB1?hVr)KtGf*>Rv&jySU1ub_0QER3NNDcF>N4VCmyDroy^uW5DUCY&j;?
z(vaqY#n;86@?Q%T^1D0T$+bRbw(<zcFf=d5elpcDSc5-qLUWl|jUo)0Xd-TvcQ-j%
zx>=h!9MddDSPKySY+(_kBgPL);~RSBEq$x8_UV00H(iGlSi?y{J51dCZ%Tc46uN1@
zDpNv`iwh=Jyl%|hQ$+;_%3Lu_OsL3j)e)(iW02}McL&0ec-{phSXs}a5`~C7Lr_Ig
zb*}IbD7<L&`hPdYZnbtZzdQgrcVXx&_0vKbjl2o}dP8JevjdUk1tBjThRWhCvMu!{
zK>`l(%Z4VtP%8a3U~*Fhxy38uu}>+;0>USSCQFEi9OjS}BRv!RISlQB6#_ah&lmOR
z?uIomO(5b-C6XpuK6jiUp<<m|p(}?yf(f)f`DttUo4CzuNMH#T%Qd8k>&P>hVZdSx
z!L)4!R|srziX~pt``CveuqGKf-)Q9XrsQt6Ug;HMn{t@^Cg}tiD^RoGc00;xpmX2r
zrDvVbh^$s<E^o%>?`{jkQLMP30|GKX0^-vgfF>MF1)Jb*uqYkCZ7op=ABDw%DXLUy
zP4Im6C^md$EW3JffZFzABD7XpKX!JIsRaVwGa|}Q(*GOXM=_0$cl*SBqC`8_93T}P
zRRtC$u6{guISg~(OrAuw!@4xEHQu!aAg^rDZGFY@P2TA6>8)7z#4oZ1X*h8KMTd!?
zNt>Ciq=IU_sT>QsgN!2G&^7z|%1k9+)8ZFbdcsvb+G17IUsKRWqpUlpwAZ1_1elsx
z*KshA3Y(~JE}7y$-CLrt7yLAr)zz2?i)n8$sYz6UpgGBc5`iJ%S@)HTzJs6#(83+-
z)!3#yaUs7{UXd}moq+i{HCn7&YPq#+T33oj2m75?m1nwY>0NhGmjtn^=THDCA*H+^
zJ8{%H#M8GNWk-^n)b-IWX$1u6gX7;+r;Uwx%}_}MJds<*++s>Ff7!dX$W6>hic*FJ
z@-~X>ImbI13yg(E?t}SBX9reQ4hx23FLBad23`HRv+s0l`~DE<17|{YwL>nD4-Qd8
z_QfDdJ{Kh7k+{SpAquD%lwM`U)Am8Ts0%AL%7nf|I&14Y4{tJn0#g&i52tv@C7#Ih
z<Uv*l#H7ptl>IQx4$zjqQvl8NDZUl?a#l|Q?&&1+oJ=?0SH+E|r49)dZa*6w)#CNk
zE-bZLYD!*5e-9qmwCh@{PR8z8px(DHL|-$fejEv&4Q0^vlf4SlIcxW$B5e%PoIjl7
z!7Ov2q;4RyRz;+*UWIbhZR+TSf=E1g;<cb8!b(tk<56%G+@-~hA7GZ||0L>x9nslM
z^jRLAjAw=t{8kdHK>&W)rJObJ%EMp7AEV#G9}8UvO@5p+`W{XOrfmjpwR^|gqlghc
z#5{|lwl)Y}p~=_e;ae(U=t@N@pnu-rWV8|>BFcTpvQ}s@F%HcXZH8^d3(g86T7GcE
z5}~j3fL2_St0UKsk?uHOh~5$%fNp8A^qe@+5uj)W9hF37Mvt7f<*jXr$!(c&n?L^7
zm65o}z=OZU@N?;^7AV%@vc!XcFkzz7l37jOnzpZR3r4xA%s$4v1QR&N$HyQiK~&d{
zFlDrOe2uEH@!90(7c!qkg)GFKv8617dCB7rsJ#<HQdw1gy1ILWQw*{-H@4jZ`|5nE
zy_nwO!rWE6%2Acks`X~O+2zSA$_iq^!8>lJ`QS5bgfBynU!UxP<=;`5SaHE2JZkhp
z1FwZ($o!*#XH2VUPOY@`ve#8Af~rjFGyFnJ<;NJt*Hk(es-Yt<R+r*|_S%Ya#$yzU
zldY-f%+2<mOr7FKk=jYM36ZxWHCj#v%gLF3>gt)@ZK#%0JUDbqT4g^6mYm;fbYODd
z)wOxK&d^eGoq!7%E;v~Bm0f6#$(_-?ZcfJ_U@~t;&YU#AEwGH!qyw!<J3*e=OiNPF
zO;VTsBkq5&S|_I2ngNBTw6!vHBCrh0uZgKsZHN-=zyM8E_zlLTre1`Y@;+n(OL%|7
z79bSu<ix=!aaVYM{B=wRzcn;bCwmTALV03iJ&f@YzTn}tS7fvmDxZJn-!6tkMf}D+
zc30T){-!L?k_Qnw-Ns$N_tUF4fJ`#gR=-t`_RLJ>t1xygjKNNTBU0gB^qNXp923=6
zIGVq!cHKOR$HU(huv;4-irX`os3BMt*4@uf&Ih9Th-m9Z3DHr$sV2al_q76bOV>od
zLDhNQ`wpklMpfEp(COA4<HzBA;>W)T;6vP|$tenF7Fl}JFFVsi0nW_7!nkci;ExUT
zx<9lcfY|#pGqB4bB@HfZK~yhh`=w#_Qg41wlM(Rq=PL=~=vG_&+w_D>PB6akIPjr#
zcv(TD`gbG|ZMUG@yfKFX#kQQV!bDYUA%#<Q#6Gko`ofaqt<R-vg9fvg1Rg(Pj)Pm%
zrQvWGg|#0IO=)cKY)tMu+mmH0%}XhoOIJ?o5pQW9--@`YMcR)Ybk8ek1fVvUtvuN9
z?+w`bVXW0=p9yMO5C`pmI$mfO;rDn=c<IKLVE+E#Lw0=N9HETC=bDB&d`jk9ZFy@4
zwZN<0y^bBle=4E>3T!Yk{Xcfu|NpGvN>j`3pbhPRwiC7R2##-k`(4pjThc`%Y)dlR
ze8yqX8>&exO}LFH$FW5}HoV;9X(pUU#*JMuMD>Wm$q;70ym??}Y?-or{2oKc=Q7>E
zGN%ZL%csazuoBaEkqzh?#(6ri8KXG3@PCDi`}z2Ba&~ig`8n{Efbh}F{y0C6j<4;=
zeZ3s5A^3evx{XG}l9}DieywV_>&jlfx}A*vCinAwh-v%|(I%iFxjlGU7?R~02pDF+
zfjyhXZ(SHwCbTKibJn1{t{|#wrn>NKe*7J|)K!L@X=uy>`u>kYZ8#Md0-Xqs=+&yZ
zF>I*(nWfc0{%zHwxqQ?G)qC`<;D+<?JwjxCsavb%*|_x8a98`#1&xY7dCW%CK9ebM
zOQ&}dSr9;*4TL(G16Fi^k3_{}3x&Ms4qf3IFaaf&r(mdt$v|bb^l2kR$$+7|fwgLN
z%l7Yzvw!ox#AbTekBvSw*e)ME_D~-CzS`;p7U>@+lJueyM^|q_DGGh?YKlQ`OrAnc
z-I{s0;5Ijz=b{^h&C}Ax7YrjIU7><3lPN|^Bw+7_F>Mix&mXo~oDgvL!&D<eIEl>5
z+5YOy>ig!MgH6b40-ISFkXowm%2*zLC$#%x{}+`vKc1*7yZh{vk_7dR8L>WvP@=V}
z_^2SmtXoo5Wx_wq$gs>0@#SyZQ$QLDqOKzA?vU(R<3O=}m4NDcB~TE>=VoS7ZEQF`
zT6dmZH&ZQ&MgB>rq!?=vCSH_D&S8q7Vf#_Yhs9)(I5kPMiVph~L2b`38uBmfuBdUH
z(U>53=@PC<HAx}lk}@-|M+Ryo5I&Oh$%%+&jrQ@N9y9%TUJdmY4{86H-TVNv-M>eS
zI0SV$LNL`8OGabDRm9SSC+eNGwgd|A0Ume2MjbSkKj|Of2%|$i2}shvliwCgt&bHO
zqi5OD6@ky=HXZu&#yIHs{w}N5C;G1iGp%6;xxO;A%?l|?3U1ZMEkv8SWYeEi2A^1Y
z^rpT45f*13Zcy($jZp_DWd=vmZz5NMP+@N+gc;~ygXh~inJlrMAQ~EM(wzPd_bwL|
z6FeFZnSGLJxhajj(Q7+*c6>Z5OAH`{)rGPEEw^srXegA1Nvmcq$RiP^sfxUOKOIfI
z#}oidgMxtxd=HD)_f=qVM5;p!7Afy{-krez6@Ve&Q@q_(V>!ADKwQugAjK?-O!5Q2
z^7Xa5C0z{(U`C3-gx*#L(LB9@5botqJCsKsGYNFA5@hjEcgUl#8|=Z9KKwD?e~A{O
z!%cJ_dukmu3cox3%rv6_SPg~8pe59<yqQXTo_`+f8Q0<x*H)-Q=g%zGR;*65<C`GD
z7LSjqTm1Oe8*`8?_Rqabz5xV9&Qga2j2(|xY9qTmhY7-*e=1iiCNLH}@L|a~k%jz7
zy<E(4yZ0HjqW7JT1Y%`Sktk=IFn0OB_Xyk~x#3k{u&9m>u`0J(xKU}0@78g{dc#-Y
z)F+rOgV$<;>IDCc*uGebovIj2iq*pKn_z;FL?zHukzhvsIs9=YUHe>2LKFD&a>v$i
zD|c}uGe1E{Pik}}RVNB^osPMnNikTDufI;a4Z7Gd^BZ*j$)7EX{pzqqPoOm#ydzS#
z94PVa3!oobsCLKVX)~M;yL=nKV<NDQzCZhQ;e!{)<s$LrW*~M0wk&zGkB-#{-YkL)
zxMCbf-r3h*7MrK?_UxL}M61WVxd+}hq20z~ChCJQ!JIa$p;WSq1!bJ;qPk2unxe)w
zSU*5LPX<wQ$(m7(>|AQN{la6|ZT87XiNoKbHyiF4Cq&Fwk1a>#T{6Gj!alb=Yh^E7
zf689|hsbNF_|(#0o`vo@zd|3@(TG*@4#aU3pop0*QTJys3K7jMHVSO!onc<j#>s-d
zYTD#GG}qu6nF+y4$Z7lgSlnYOCkSJ)GUhaT23v;A%Cj_5Y8ygv12jS;DnXQWI)8~j
zL8~+uV)0FAVm&+NHg3dHH{&XIz?go@(zCYkPE=L+5W5(!OE@L5FR_D><M`zAgxN|Y
z!0DJbYK)VAlA??Q7ARJ8#<T9J9Y1T^z3{X@@Vz^Kc;c0Bue(ta*G-ksn^!40mift`
zMl8YfjjZQY@bF*}`$PS2(TVL8S7;BN8@_Y!mxr%#Q*ORcd=stvQ~^=>{!m#K=%NBP
z(G)$ac6XCj;*8+^1;<E_KLH2>23DCtyj|DkLSD?rdMZ1orK)dHS*9bBEv&NEFa6h(
zPD6Cdd~DvsC7-<&L4$}+2`~_O?{b(I$>E{}bs80`Dera*AK(X@cBZ~~cnqwbsDXg)
z@(-P=3L8TY#Wt5jqxon7l}4Gn3cm1-zS!W}l8yL%LS>sxSCrDXrYg_s2FJI$icMOH
zA}(-n5tzx^=+zY<t#~pP!J)qrc<?=Ob8E8X@*39EpCo#{m48=Et8&rOsy(vRa>$4^
zpS&K`&bUTep-;Ia0mNi)W;Hqlolstjr^@%i>y53LIxSplsi;S?int86sGLAt`m%bg
zF22$*#y7v#aSxn;;oJM(X+@q|dNWu4@mxN@D01RzPJgPuahl)tbHiYOf{o91CHnmV
zy#;P27%wZ75RF1i=yAdnt$5Fsu^gF$cF&8$Q8V~kQ^8x=P}Ew<I{06h-}D4MorCK~
zn6~vQZ73gvDp=tlkrW?@S2k+g2yyrhE@I%IIOhn58Q>?MeQgp*cuFL|!cnY6+MfDz
zKACN2V7sj@P6<r;BJ2284F1I5IJZwa_8{gFVR}GWKu{pDm@h3}s?IhqfxG*MRt?!4
zv1MU*!uDs9i<XwTh_XN6>I(Q}{~=LW8UAC}fr*3Zf4%E4{ZFE7aG?KR1qvWTG&e+y
zr^2NhXRh_k(?$JsbvBIh)|?6}sf&v9rsy9#C-NkuiYeBmk`W-H)AC)oIS<*?;fa%A
z<o_BJ`1n1af0>1gX)2~?&7zQ@^pg%~lbkdj`b?nme!G9in)M=91sMR0!%R@oEyTJ%
zKOz?%)XYXvsO%>gxN@Mv7ZzG?@*bwOR{T7l&+Y7Vb9&!as6SphF?w^pAExbUr{Bd_
z6&8shiK|h)`=BM$1cc^95QCvs#@~?Hz8zI$Yo{D}XsI72=jGT{Y`o4Z2iR&>RCKH-
zo<?HS(cdO~)n~W%!k`dN*hu~m2YYdMU?u$p=ntV1n!<=nJI+QUsOxi6aYYLJVc<Zz
z>-rMF!e93jiW_rgf)3e+tHQVsF$@8NjgjDWC^F=<=ZIiVvOu1F<#FW5j^DL6{>>;t
zB;)BVY^ul;g<RC&zWA1mMl8p_b6D9?byGFwedFBS)l(js{&g|n<D=#kkn$k==-Ug=
zhZX!_7zdiywRX*>x5<K;@0AzlFr<Ml=L<`6I~d>RM%<57;h$dNHj&No7uDevQ3f*<
zm=_2_*9#~?v)<i=T0G-eMMbx!3RhCIR;Tvsfo&)0ZFm5>M7v#3t>&OX9WAr#IT25O
zVu04CUS00z7kU^Hu>FJ;cSV&fZbD3Wvi2mfnOYkJTEXI4^1ymo<!4LX^Ej?<SkX^s
zyYvWOcK7#)M(s>2-|yx{MnJ$%mZUMCT&j3Ut|agQ8Uuz!igkps@8d`7{3#8+{*-=#
z<txJ9HR%Y?nXKMC067c^UAhw{&*W}rOCMAbF(Z62N~VX9a~%vJDn$am&W{t%jvP`M
zR0$LfJPU>4ZM%lP(qW|e;GvK3%qN1pht!(W2QxdD=Zm-wo#}A+6-d!iJMgPq3@k8!
z(6lWioRs*Slldf{=McvAkfYe=d8#G~aeW+u19rlg=j8{Xf2U1tiT5!`ctyW0OO2Lo
zRHQgqC0+GF(`o$KLpA~lGF_zdTm5-HH<uyiJ?WYy4M16!MwN5NFuFER1e@dB)f8(b
zlsK-LxVHXhG+Ls(tMKSAq<(bqocVyQr@5&#`wjDGkl|9$s6-lCph*~ykzk*=M13Yh
z1=2!Jy57X_fad8hvvNE%4%!}Fj0LXR$)|~$0d*2FY58T147H%Z$+)8Q-}k;y7KjM)
z$KszHW9VV}doQ>EM)4-@uWo&*{x&h5fi5^mKDas9&;?pXw;(NO5sl8J(E!kYTyEOj
zcsw8lt1(Oy#rP?fK1M|D`(Y}*&84G7AtDem*@q?Tx*t!asR;|6ZFDaX?YeYiS3@j$
z^0HWMod$<=`qXxtnTd++Bn);q^dJ&a37L|MrP55P+f+KGyG2-;oDaFH%LNw=mG8z}
zhVVih=%2ny29q<E$?Vk!)r*|-$5%#b)#UKLP*`48vf;@iPvw@4+Z{2ZO5#YtEQ>xY
zKjo}wb;EEjm#a&uo?GjTT8zjij1?<CO^pKyZJU|}(%%H&@bPT3Rj9snC2ullnRN3f
zy<2q+JMY{HwEj(fWZ5E>-X7f)YN?>i?`KHzW76KkLahrzmn1^?k&>)!ykuO_OJ+3X
zs?h?4(=`al>nNm4TCOq|*7VzlzFl}ZR0<P)&gE(3_<=k?_)^&>8ff~=u%$Nm+BM6Q
zOnR$?^?0Y@m^|%7Tx{Udff!Hj#*5JD@%9*}@U8|!`=Fj?!;n$s(NcNbZd>hNjBoGn
zrbm2iyV)Ef>8Nm4nFXv6zYMSuVUPm)Y?j>{ekp1T&r}{zx)HYk7}myv;Rs<2?W#5e
z1{?L~b9^N>VKQEQN9ca9cNhW3n?B+Hb{#E7{&ck)7>42?UGe&hl>l;<9T-Q8KJHCU
zyPL=f7BiP*@BlBPRJTf<=o;4SFMJZHoIXE_3If(FVeX4SC?7A#mY*X4!xFSfP<S^n
zvU?*BTS2H4?0zfBa<D#({|^k16=AY-{8<=mj9GncmJseNvaAHsWTUIEv2Jvvp_^Id
zwqdL|jW|a)QCXXj6|(P~&Q5pJ0FY?>T<B0g)!kWaF^HyMVPk6xT6XCfPZhOs!xBYQ
zxJ!xj=@Dk@A<jtbyk*y;xxU<v)(P3M5avhp-3!a8R;yVmN-7!Ec)Z5lI8HakrqpYK
z%Z`uAinj&<O|GH8k96FtXh(KeQ)yxC)sWc;Q+~S@(k!{!S2QhPMOr`-fTBscn?o1G
z1qvV`dE;flE*=M>A2Pv?Bljw!4NV=ew9_O}fDqg!S-0mPA}lJ8ypGNvC-fY82A@m1
zfw$bX*W~31gEtr;j5Z}nQ8)wJ0V^4<%B_%>pH?M3LFb;_()g1({T2ZxOPFxyL+ARU
zGEU){YvYJ9vjHM1)XrJpsm_OQ1cClpvx_1y9k81i<tGKQQX2%u=FO9*%6mBs6ocHR
zKs%F@5?>E$Whuf)40#71BP((go_AakZqSHcA_5d<KsFJ|0V8BElt!&AJVY<wt0==T
z66X$wPVQa9{_>k`_^70e>Qx>tNnDx`8%#i3=g_%nzdm_L_2U`KZD4n(PZ+v%gx)}m
zlX{A2o{S-(ByU{<hMLXf*i>cXr?{#%Fo34IubE9v@b+$fhKZz>og4Hw{D%_K!kr5g
zy%@h>YP;*7u5}C~rZ6Zk$R?VE9%LB6a%PwIS{hIXA(=mqxtXf95VDX**Yy@uf7)iK
zaSCP9lcZ~aQ7iu`Bqm>hL6swGgigt<dOAU(FO-#J=16+CiEz*%i!tpiu9ZxodCf~c
zT3xdg4~I3y;T|Lw9W>7+0T{Ak@qSQ~pgH>=d(7nF!EYby2*6V^9PwV9QYjC*G$FAz
zOS?rx)-8LT<ryY(Ywz1R-L<(<yR1@pLYRUYPSoO$aqC8Nz*gJasYWc%<6upbsoLop
z&?&u|@VsK24^Te}%8jC<U+>eZ%M0j)*jx%CY`m}=7)J=VFCI_8UA{xH-v;Z;WPw&H
z_(#fam(KnLRHKIWS%&%W{E2Pz)b`8mNU0DL+&G+Zyf(+pOK$vMKoDgX0Mw5`u#z`$
z<XvC#A&BHF``THcS%Lv3aqqsA<T8F9y+_m~cI9|`zo&Wi2IyWeZ?A=o35ToJfG&E5
z+m)YQqu1LIX3+OQUo}<-rGRpx3|iYl$?t)(53wIrNNIKGuyW>YZK-gDT*Qh*ARvqo
zXwkic$MXk?ocY!KoCKC-Pm0-j4fP(txkKdP_Soq|h0L5`hzTO|z$NhZLevZIp$(-G
zy|^pxUV1SKJPsh7Zk#HQ`?n!Yjaqo!6&0B;O1`+sDA%{~$5Y?;w(Cf%HOhhi%3ly1
zcrlQaQI4|<3cuf*_xNcf;8+URwr5BwtyY#yzng=x9UbXof&a_=Os{KC;(g`HZGQFu
zHEXhxeoJv@h$lk;w^2-|+!bbyw+BlfjRru;>#s)UG~MGiP3TpTG04sd-(9;+`l0Ga
zZ`%r56}kgztAf~q&!KO7&06pR_dQ%0KDd$^2L+1}?s796SOix$2>EMBhrUC~@SM73
zy5M<vbtz~t(j1wOH;*`;=WeYlSG3c5d(+ho3(OcAOD<D9a|%(lnbnLa=%jO%$68eT
zUF*OeSQszW!d43GI`$X>4SV6KsvIhVkKI0H<xO%EZ92HCbaT~I<y%8}+ND@xrzRt!
z15Y34BC=w26P?@65EIfSh@SAxjo8B2&Cova7|VsB)FS{R?qW*hC2Zj<5as6=#31L;
ze`srFj{mRyr|e;GLO?HPXsP6E14A!Mz{tSxKL((qlQRJ)J2MQuw27^mvpE4HJH!9F
z&UsqH#txSi@!Pj|Kre2cUN?61pk6_lk`_i81iD!iqyj4V(LXGqGNIDnZ<e_yWqbP{
zcV?q<5kcV}<H*D3czAn^v}2bj_59CAZH(n`L4uJ<YC@aHbCFmoz&b#$zMqKGpe2}N
z&N%XtX`G1kGu8SsVp2s%nvg?<71@O~j5eaMZ#YL(yhujO1y#5Zqbd(o_3zl?0E}Hx
z8oGJ_oryi;l7vtTnn7|@OrKnY1DR`5p+i0(d0Yaxmolt=z>aDzC(udqDjN2qQAiAo
zG58r(Iy_oxh>Z<vZc3&^kR=zjFtS9<hPp1na-K-Sv56g1%fu^4P%&+f>{B;GvBb#c
zkAov_2GmSUdANUKdIr>Mq>)ipppmUu1DCBMjVjGJs?3~&Dl_rk7RW`Rq;Qcm1WoYA
zl@8({p>-u7S#&HAR_U^wU_NV_3R*yjfeRR)BokI4)zK+=R|k%aILMY*21b6K<K&d2
zY<&6>AatNzZ&@RbO3Jok2A<`Qt0dRfq!v&oGeoEWh=R=`Nu&hFBaNH`_AWpVD_WBz
zR6^KIHYAXS=~$102s`N`D=^HzQUNy(Cj^0?DSEb)glA;dLv>8g<OPpAhC|U`4>Xb;
zCmZ2O0au1bH$nO(id1C@3Wjfqi~^KTio;wK7Xri|N!z!iU6v0m?wkzk4Q)|e{{}2s
z7*<d%U}hg}h?NW%0uom-(#FB5;cQlU(ag>1u!RK@dkA%nwQLL>B(i>40d0FB9*e&9
z!d^LfGTH2P)^XFz<+K%f#Bb%+(?@*pvcJPF#nblc(|(%c|8no?V->#6LEm9Exv{j7
zIeBzNwz6TjL#wBw!=w9kpo<-!ufvin{C5~`8-kE%6-utQR&Dp$QOQ+07q32Ae~wpo
zt4nj6#dfdjqr_{Sw;$rn*A_-w=qaso7OPj}37kXUG#BMr+s7b|*YD#3o2JvQw&v$j
zR!%m<|N7C<r@7-g$Bm!I%hC_wTfKH?(|65f((31BDOb&I3c0uAt-Ddj?R}*(l3&*k
z>FGIj1;dXYP;t}r1%s^)Zgc&0SDG6iEMVobVyD;3*P96?A3cmRu|z>=bq{MDi&Aoq
z2F)UiIgFys`Ug|$PGS@qkilDVz(FYJ9L|l)5A*`c<0cT*WD3{lp9yjW+PXX#pvB&2
zb9HG8)*F;-^@;<*(md@Y!0aEDR!9x5SCQ=UivCkB7W)j{v2a{<QuE>YWO%fD0x7pL
z+3P)lum!0<qXg1C%B(M4)l=b8cv|?hdu}zZ<iY@^{?W_MS&~$JFvNc|ymRa?c_+81
zzs4>-(%2&h80m-B$R$Hmylb8Z8;Ijw|KcDtWt%l0JOCh8v&N!`*%xV(A)?=dcROSu
z4r?QCfyePtTvH)#X#hf*A^J4z^E(hQ=3|2<Rom<C?!#0H2!1m<r^uk>qsHY7tN4?_
zm7ZKRxbP$)%XxABIcnovuRRop7C-#57vbXZGRm^NCoCJJTdSF2(S(dj!4f+$8f~|G
z{JdjggiF&#U{4<eP;C)+7_nu7gxM^jBoTi7hBp?El2flyfBqWb8MHfl2%QiJJ}yV&
zn(G-qT1WxomN~|a(IA-blTMr;ZP)LVES6ERc7qhZR8@A^_XjEK*sf6q@#vffHg^!i
z!MEsrvuV~8UY{>@?rKLL^@*)9cn{lRR(Nvlip4At5l4&f9!W2m^4}0a?$U9V59@)=
z{tgLSGIdDL^!0h%QY*)7i@I^^zkV{WixBO6NPNtR<m!x12R94<uurjf7yC6md)fne
z{iHJco4e;+C9p8d%$VSXblN_}dp)8Tt}iHh2J|3caqw3T?#m^9pxw$|gdpyn*<!Nc
zOASY7WJ>*egP;41Qc0>!fg34tB0^Q}Feep4AIq%2MUJAQ(~q`f#CPKU9Xu<fm++uJ
zJK1&)S@g$tnG^NqUJ0yp)<kG#!5MQq#e6`!qDzQ@35cPWh^v;;*R{#?D(_i8L~1~z
zOa34n9kJO39;x^e4hJ|!dcJNBVl4g~@5)mr0C&9do`Z0y!OwyHi0bp&=*XvVW7$0f
z&Qn;H3xU-|U%hhgV4z_uV#kMyLy=m{>V*PMdy#o_fb~O@-D8+>+fNkTd;MZ4nasaL
z+&I_v-z4$?ATo|qgiz+DG*^druVeiP4=Rp%a3L^RtEBaPZ1rbHi^(tQMn~cqfwiAU
zc#yNn$!6i@S%;X{R0oVT)qTi0F<@5B;{2o%xOBp2$K)GSu<ObGKVC*whW`k7GjcHg
zuU)6D{}<z<TG5Qu1udJ2KhK5bRw7Atacs=gB(h;msz^x5PnP)oOo#97Ut+x`NygkA
zMgVnb$nV|RS7UFFw7a`>IP86YN!ss#Sg5XuH8l%L2Kqmy6WD`^JvLJmSOz{!_+YD@
z?6G*`0ej;aA=)h$_xsmt>)T!C%JY}WgMsU#rK__e$Zy$_n-f1CuW=1MU<{~V(C^jW
z!;g8;u2L`z5y^pM@L?LT3SqY#X|)Eo1?nYI)qoI5libOooCJn+`~<|5aj0J>spVfy
zl%*u_3nndPjnpeI4w;88s%93&&4DnLVaOq<Mq)V8?Mu%wOxAD&B9Am2BoZPH_F)DF
z@WmTQ!XF{>$U)vmT8Y+d7K-aGChfinb++h!@y8bFi_z`fE}1WlD)p}?eT@OoB&cv;
z=y`uM(N<C^{qqJEuVE#6o2!d<uFeMb(s7i?-Wt?5TzhnKi;LNLs2z=4;fImXh%osk
zBlTBG*DA?pxH9DUWHlqP2Eh7~aSh4?hdm2E(oq?7!BK(sqQFeBv*f-#&#CZe4QjC-
z7chOn8WX#6(jUnFxaawUCUDyAGeY!?rjYk74;6MEt8QirkSaQCa*UFP-VAmGBq4JJ
zBm;<ULAd!Rt#;aF2lbxOIh%q(t|U8~R2<;ES9g+5Wa&d_8~7vCyLD%$Xpj-qzAhf+
z4RL$E-;n(6gE<u75pC5RlIhM1aR2c8d~HwkdH}pHrDP|V+VN1jTKkk50ks__$OTA|
zJr^lke5R9z7DB0zmp?Iszt63krF+p5s~<yVd36}@3eAD8PsoYGza$zZCycX~Ei|?w
zOT&=D7nKG!Fd<Absvw{+8xxm~nLIkEZ;-uGmd9HG^U#`uq-HMk5m7|gZ=oXgr)5`H
z6V=82z0dWO4HZ=P2$lQ$c<8Hn#=V~`HgHYVzw!L^aqGzt(BCZ934Iu{zSfj7G5+zy
zETZ)e`5<G{Y#tNx0W)cK9e&*#NxC6^2b=AsU?M~D-t*(J0SF@3F8Hi_dtNzzragNv
z>Wu^Jt^Kt)QRZK=Pu#MOh0|}YI8?o39df+jDPBl176j)75leLV2#nhBgnQ6L<EJ5|
zZHMe}dzwwxQcYBsOk7)*)ZQmtO|BQPNYq4qEcDJzTi_H*pg99=9?fg@KA<C!icrwm
zhmY{ds`5C{by~#t{>(m=c&*_>x2mM*xnm(iY`5W8FSS@Jn$HV<nIn*XRJ_#mNT2_%
zAgg@9t&?2hxJ74pPMJ6AcQ0ANG8n`ASg56P&iw=7=9WzzQ|>L1y(@7KY!A7-eIae9
z9h5HInX0L@Yn&OSy5ZDKeZx?~PBsp}kkOZR`o$ZL`*@qLNkY9sGjp}z6Up_L_08o^
zOuN*$LDJUSugo+xl@hG74-C=u3O!>BUA1)xDt}#9ATWZ=ak_aKANmx13|0GRRJw7t
z8BYumY~tk*RAIyEa!VEKdB(e4k)m!19Jk*{XF7BUBgj{m|JvnBu#?tfWm|d4ZM>9B
zMJ*%?buu1ZqXqaV+o%<Att_;;9`pBjP_lX`kzRMI7X$`5;hJV5=W$3ncJsU%C-_jv
z2_|jkgl0F7Be|>~u*v!^Rn>IJ(*N;=0-vM8Pzz$JWn_718d<oh^x28`WZJO=drJjY
zK$F-72N8d%!G{klg29kluZgl+g^2Z{g4F7+iH4I!q?b2)WyqobcE2(Jv4cqL`Q)nJ
z>>9a@X^*EwLZU=UIov}=wa?QFTn~V@AOa^38zlVW>;h=Jk8_Yi>XE=0Z~4=G^^m^{
zO5qL(&XO8@b=bGJ?n-)sZT1RHD2D~Ljf%$MWNHD~IC;3<)`@dq0HTZpE4ixKxEr#)
zOT?wTNSB~Q<E82pzCJmemMaz_$Gh@7UeFMgah~fl4E3mv6?FXpI#jf$!=}=vW{xA!
zV*%$TFen;MMN_xg7P+5Q{x+|*UT-V_$`+Uw*JV%bGfEanJ+%6Z<yzqtARn$3eLB-`
zR7(nQ5J9`rk@EDnRtsTvaJS{kATs7u*<-tj<0(hMkcmb)H{Srv8XRGO*fwledAlk|
zA}+*(EDw}#_MT7j9#H0IC{ejCE5WJ~m1kR9EWmFTpavisUhG1pZRLrKkT9Azqb`=7
z#Sl3e$$*UwB=8!4Au}3BOWXm!B7#XsUj?;&5#gS(bTdCu=P|{Eqpv`UmOX=gIZ4X{
z)K9t)f(D?>W0LLe@~CpiA}0kG703`?*Sl5`fG0LU#yzFu7DG4ifXZ~~#iWr8A=6iP
zDL3nVAy!!G7Ug8vvX66=>jIvz!o|DwEOh;_TzIsI+HP9hYI{e8ao5y{!D@O&G8KNA
zWAwt0fC1CGOAY7{4S^Dwr(W$`i+li=I5`E}gsxxz{`JC>A@>bBbRQJ`UoS1=f6aF|
zIsVsN-7QUx#ElUo-yOYxR{P47znf|0Llq6#&BUoDlk=+hM3u_w$d#f4!bAeqEsAe9
zZ^wO%1Z{`-e-u3k{S9WF<#x81N#5e(4o4Gj=wx%g2;2=73)BA4)gad(Q!4)fNYfpU
zZ$Pi?KW+`n&HCmIJ@!#dE9NlL?C16TOt#k2&FJOIrjww^&M%ZJ8{eO=uw#1PSz-D;
zy?@Tq_w>5otvLOB9dD%8`ug5Y#=3r?AX8w+F*k=Lm5e8ul(S6~cDB)dB2N<Oy|q3i
z`z=rF{r<m4I%M9qvgJ9b500v-xu9cRRp&Au?bVYm>cv$Mm8g1r=#3p|AGNb@hFb2f
z_6!xoN)Ji)&_pT|m`@@(UIJnoLXs1##puYL{YiZ-t@tGC^%Bepq$I^c?s+>{<bq(z
zVpWnRrU@1Zk1c}ZMoA;55Jx-W@_}SfFA_=qXN8|k4LQ<S@o)o;^}!-kBERbGpk4H#
zQ<|kVWA3pd?brRgf}r5iy-}7YkfV!}>?f)eLD33n&=R~(d!zpzeuu{09qz6uL5EhT
za`d7IjjAS=4^#m?I^5`Xnvh@}m5)4s->^(l{i#B)-uK2fNtTci^Xe)n^a>(X@{H0S
z5(aCOYFZ`q>^qM_(7;;&DXsn<WLsjiD>J7Z^S16qvg41g$`hgbkQSEM^CD#}9w|yP
z8BZD%@+h<b#LkKB=`rTr;t*v8adGx?yg3_M5JB?<q?E^N5TOQ}#f@8g+&|+<9=|@2
zn=^;b?{NGrE)M#G1Bo%7nYiOE5S{*~`Hz8{P=j*t>r;b~FN9JdK70*O&Tj+@5!g}2
zDJ9|tn$-w{^*6$4ja6C<N9#GtIGD5sh-;vZ%fpDcbHfUy9M#Sqi?!uyqXW%lpuY-I
zkh%=QW-qr<j^Ybk5mhaU=0(}O3_0$RJSGLjU%SW8Wi)chA_;TCZmwL#+Lqg=x#p%q
zz){PBN^R1`rr`pvLU5*Zj1R{hx8or)*lm-fmrY{rEI})Wwh>E9M+>(T*Fb_-1f^4;
z9L)z<r0(p|h%pOt78Pyo1AU~wYqSvHwC`X0NMAm-duHT;2U$Ujv7Sb&`H}yky@r^N
z2(s97B;0@G#oI?JjxyupLl7CF@*v{SWWP)s_l(=Aem}t+m^SFTJ|E~7MCHU9Kk`hR
zD%rVE8e~^ffu;r`<{{<bLSw9GW!_iP<n>GahT{_s4`5Hi4Dc8skL5AU0db4-6q(%V
zshu3|a_{tHM~~o|$5fQBZgh9_1?S>9pDxd_R;zS;>i6p{$0?W~eCv(?*RvT$W&}}2
zrU;$jA?mGnB_AjXbdQ5o^ves4(20GL_eYa4*4mmzxOEgbu<p)CRjsl$!^&+-pfxf$
zz$|0;z~TVxdFY<FopBLEV4=mvdvlVI7R8JW6mYs}q85^wP%sr>wL;dZqhLvos~N2D
z7Ka8t7mvVN%G1VR;2g5<H@gmn5O^c>I~b~&GCp;tNF+I(AuSF!FPK=*^LrJ1(e=_+
zFzGe#skvlK!AYjW1~iqOJEpJn<7mUF8zaV~W$H7x6+Us%1-NQ!3r7l0Va_`#EwALH
zkUBfHyU@vq=z#N;^VtNE!-5Wv_yoR72YM*UiZ0wEMtgA;quf9RKV&86az0!kGLB;`
zb?l{_t?|wxzJoZV2FPqo5JL>E1o($V6WD>Xt19E4Ryg!k8|e!D<~2lTK?{02cO9~6
zlx!cs*i_70eX(i#?J8Y%o?<^oyqN+<RyHAD7v`K{519!4uAE>qM@-|0uVimpUXs{#
zKbvV-e@k0nYTWXjW!IOivDf>>cP7hbUI}VA72iZ0)Yx68v;LA%y=m9-MXLyk&V+8q
zw1or~oM4tQqA$pLR(Ev##PqB{l6*P$ucX{}X+_far6GzqkYXy#y?)9r$W$HF*}4sO
z(_$)0lld2X{3BjENskAIz45?h2^S_cA4cVRkL3&deB$J1gEoN}`cnGu+LlHuge-(Z
z0su83aZ)W?VLHC9WQxyGjES<<pT+(O5Uaz2ufW8$dk)L&{KxYE-run@w221XX*{cJ
zSUWD<;7jl*cS&xcw|!}0tR>$?0h9YfXpaaO0X8s?7iyMTASBbuN{J3+;%Siyu;JWn
zqOWOn1v9|)*`Pzxcv)lgwvYk?sE&}y*pkFI$eD=(NvxrsFrI&vRxOsG)iGg?xi5ri
zD326iUe$lvHzbYa$Kk8gm=Ro(on<0~Jtvr78`YUlvo$A}qm5O!<fQ7Mvg~<ItXCNO
zftYL!>jJpz(>U*8khAjS>>M+z)$Y!u!%2KoR%8He>ounEmQAXI_?-7<Cu>&?I~|yn
z1C|px^RMTjfG$Y-)Un72Q|m@bA3Ol%1{cugyP38^d`khysU?!W<4+k-lfCSw^ltLt
zJ)N2&-$Zl(0+xt1^zZq5Dm`CqR9dSMu6icct-JNfy9KEEcQ6dg?ptfD3L7^>F5`w!
z_V-Z2<(845Fsx@{nq<SFoK6?_idc&w#+@(}9b!6WHUzPR{ojkcEYR;0Z8z5Z*Tx5B
za0pH?<<kNZcS6re%J>@!kqmq_g<?O{W`b)0DcHR46cYF@s73gGV-yJZVP%7mM3Bzd
zj#r9uH|KH5gsN6h+2^Qlfj>8*U!fxgp5%bm%nKr1S^AaS?#_h%k6q4(WJ47h1;ylt
zF7An#j=yPLI*TS3&Rs9R+V1(X{C=8fg(uSCWl!Lmxy(tT7LNy4p~M|fPftix?f{2Z
zNytKECw#d!XLb#0ZcICq+6c}VsMb0q38Cj-o3#VX4Vj=HdyY>Y|IXk5O9=GbVMED@
z`Cg`nM!rt@yZ5Q2H$kpN2dFXVTaK$;dn0f=rcHVWF8XOD?MYIarMt2{Qdley|L3up
zS(gvB(lBvl`lo04`ElcBa@g}3frC3B;~Nr#jZxj(;SMw=9_31<KW1GwmlmsFY%sn=
zP>|U3I?;3l(v5A}n$?3b;|xeW5;R?QvJ6%*3cih0d`qg6Fd7?{MMUX2E7URv&#u#J
zzE~H^EX_g);_#3nqlH?xf(hW<)vKAR(hQF%sQxU>`b7W>2|B8E{X0|s$|Tz`Mlz4O
z7(9#_Ng)zci@?4dE4n1hBgq8MwZ6N-^oN_!`dIm|t;1FB!vK73M98^dpm4Mp7OrA*
zkLuu1km`=-ias+=YVGLkbg1!;Uag<^Zq=!Bs?+TNeu@0XOggRHC4(B<!aVf6)?lJc
zymt2I^S^KYENH5Yd2@&%g6~$5`Jqdj`!M2FXAsz=@bqRdNNF4|K9$gW4&n1y&;it8
za7m=b4=fG8G!hY#0cGKYp1nmGh1yeJzTCI-W5ysY6To%-E%qsT<*qx^pl(kP4!R4>
z@qxeW2zT53u$edkww%w4sG{{Zzb|jaS={-F?rWsC2j}dLU)hcn|E-MQW-os)BoVtH
zeEb(<6@d-ibhT|_oe;~u2FW{32KI#wWW~c}IysHZ0{0vEuvBCJrXTN!N|YXHQ}_En
zaxs&ajoeFTnouY)^UFDm3{OxhdrO0hS)!=Qn#e`Xfk64@+8+Qhx*9p)D~V|ml~sL2
zrJEq~UEJd+yi(zV4bU=b0!%Vhrn8UXdZP8@PH51|#@<0$NfhnlrJs_FoTxDAc9L?m
z>N9mLYl5--tlfNB3ZScm8_)WM6CM&`!lPXe`*wVQjO9ZSPq=?A60*7iA#`0HK{My3
zvmr!kdhHQAXU^vt8?WR>{<6y}1~sqidN$aNDLSQ;zt&Ax%NvsH9GVPD_qS&EW2dvF
z6_b>F{Vj3K?<bqv5MH;#$W4hG+@jo{pZk~$Z=gUh(C0;livr}YX*eod2+uiwwhk3@
z3v&qe?O$FlmrLG`%U3VwdyAU6v(B2ifG;7=Up8%i02BP1J%}Nq_@4~+O9)+ow@gJ|
zfZRUJcC;%JG?Ic@>Ez5HisieVl+Yiz|CIj!VX`r?bN=tyI9mT`q5oPGF4T5g<dZpa
zUJl}_9hV!V%9}5THYd4iP(_58kdfmg#}dAMJn?}T30~ub#$~T^fM!@sJG?r)dakkd
z52^acGYfx@iqq?&SR)5T6)|Dykx7#UN&(s>Mivf@CKwb?-G^9Z`J-ad<n(fSvAaKC
zIv*AekQh09oIHKsC#5zR`FxHteZOZXAMVy(FCT+`;PhTo-A?Sbv$K2e__u!?-_I}g
z4jTW((Q`-myWOWXhJwYS81?6w^MgtMrO|J|hzsGm_5mc)1o;+3VPb$xW1dh1TmF+s
zqk@G`cM))@2kZN_S)uw%Y2g7xO+rFX;i&;g02z#h%_bzW0v$>Si9$}JUVUFq93n74
z=zq8q=OJHjxAJYI{t6PpgUEqM3jlW_;LoT6XbrQ07z!{+(+uERHD*^al(2@GBzFl?
zT!mqrf;9pqh^oTuvll{BFw2h{c1NM*;#ljv&Jhs`63%e7GWGYRZ1hP_1?|nnw?%g2
zM9HHRKg}eyp7NT32PcN`3rmfYNRZPciT*yUy)%uo557eWkNLmsjeRKY)-ogS(C89_
z3ycTTHi|xQAt&nn7A=pn4D{Di7rrwORgZQ{Cb>*n#t*n*IB=OFL5?Ji6Zv^igj%2(
zCJ&Jo&l@UVnV5fz7{ba@&^<U2LtN}k_LjjpUK%)f?UQwnOnSylRnc%o-ZN60Z`xd$
zR58{>8L7{S^xls>TyG@l9e3h(JSM!`onZpVN)fr&w_Nzj8#~G-Zkv^Wcp@D&r5n~Q
zprGjOCdSKZJ!`UPxox))ZNVMhJLar>Q>Saq-oiiRgK#8sq9}%#2D;I==^@}ybn6wP
zEBN3&po9VEJ)!zbFLBG^-<)+s0>5hEf-#+$q0l0(?3rzbKyj}Y?#9ehzIQ3HB8!L+
z>oxo3AR$427x`=r8Q?@!n7mor!-lk#a6ehOUZnC0_3|zuo&XE73QHkUx*s5qZAl@@
zBu1ff;EZt;OmUW6z*K@bY<qCJ8Jdf*ittCY>E~Sb4$e~vCMcxZB3$vC|8kECNQ9*6
zOQDTRm{AQFzh#mb$hCPmfP9*O+xQI|q{W^F?_k9$-0H*)vW!TMF63{+ux%KYk3}gM
zLW!x$x;v|YmmL$^gUi~F0SjEB0zC@G80W;q|NaV1u5|hd-C}L(?06w@l7c`)6V!M(
zI5!JyEmqMGM48ASOnClIxdWe1S&&uUCya@|06YjF93(X7_J|KaG~pFw#x63^hK1uQ
zkccoYicfNu%X${BY+hohgQ>jIcPHf37rQ8o{zkaUv_eEnfz&X_g(w)w?-No^iQ+=R
zCn2!Ag}L6<0~qTY7pk`Vv)syKI+#1Z>R$q*0mn5}@p7f1nV3(~0oA{$Av}&jl!<AU
ziGe6z$h#8$YSJ1rRFF&~66lj;eJVpNfzd%XZ?TW6*@Z(vRiNSdDaChVj0m+)wV|7+
zUAn9s?1)OE+o_Co+^wUls%p8Klu=#w1wd&8P&5A7{G^fop~jQt=vxANy@tx_g2s3e
z{<VVydy5gG5?<6<jqc~31JodMzSf=EU(pg!UR8)`O3<PBbB4TIKO9EZ?efp&-q`fj
zr4`;OhD<OGE;Xuy;-yu0?v0p~AXi2P>!&loZL|CP&G%<7UeRzT2(EA55rst!-l_&I
zn@SyVKC#z4*ao(xaen5#V;?fQ<|sPa5jZieC&aCOm?bF8dOud|>12ey`I-i=dEZLT
zNl6uB()+G?e6=FeQ?Y#sbrHPk2+sJDFJ0?>naxQn<d9C@%M7eTAjRw<t}a<&G`SU>
zdu^cA`1=GFIaA}b{gF4jYnZAb)tg~esT{UlF|oyDi?r;#iz4$c;*;F{+EhqOBHP--
zpi~st!vA6H9hgPwq9)B_+qP}nHt(@*+cxg8ZQHhO+cxI*M0dnQ%p1{P{e`MpJ7ew4
zCqGLmJh5R;B?O~y*7gRcSvL;6^}!C%FzWiN0bDIe6XDkP<N!8nlEF{pM7GLd-%%7B
zCbw^aw*(`N6jyzx0A*9#JNu`1#X9QL$c2F&{K095O9+o*3Clcxc;+HwMU6ukwP4m$
z;g-p^jMTR5n9RCwGtPqkG;U313@3eLU@c*#!BP|pw{Qmi?RGx)8{6w-!*aBHgC$gJ
zKf<PdxlX$ET>%gB&Yb&ZXL_+yq2+i|s=ihjIqS`DjLe?-@JCudafFDdAdH~JjSsr;
z6X9@pz^nTMBq-7lyb)0TQ@M<EJ%deQFV4X2-W`0ULvJvvLXWsB$PrWAcp^s=_3zu&
z(TQ+oWzuJzy3lrLp!F6O1KLEBnMm{9c>B&SH*)U<fF@_7)hRb};-rMr)jvjZws0f$
zdYBecb#%$Gwg4cOmVv7hcdMYuy;e#8R5ocJEr0lJow$Wry{SovEECz-$#lQRa+T3g
zHnQV>I|PIywgL~m_KBML8+X<-Ebc#z(dl{YoIG8@q>}Xo>9C#U>NXMCbEX&?ThUc9
zpxITO&_r6XulzgH&XRydGm!e{>gpGw^j=!49-4T8B~wQ=s^EtQzT4$R9B<kEs&s#A
z7u~!Dtw%;=u-`}dc0AzMb?HYoz~-2_KR$)A7o02XqoE%Vg&CeNW2?#%HVsJZ=E1K)
zRo|;j?jM<z4Pwew(^~tR49HnTxOf>Wv)aliV`l(rlReQi>4PlueZ@k}Y!jn7EfxVc
z&a*Weasal@&NC}{vHDRW2n8VVEE5;i!7!=@0khEVN)?q_qi*SZn&cHW9o;v5^{17V
zmkSoaAY+@Zb=$rGjDQgJUPdp}6#;H99!Ae0lFew%5cGDx&7I^9;@dw2+_^wR-jBTd
z5b{_m7ZI{yb-c_k-uas)L{objEOfhwYJ?z>`f*iddABvhHv9Q{P64W3^=MaEbAZSj
z>Cych*7T;jz$vid1B&vBSLo8-tHcRDQj-*Fdhs<In51MCK0eQ;0|KnIfx+Gbst`%e
zK<s;Mu+TvJiKq3NK%LpKchj{_$I?SUo8)Cw)DW945QC(ppD5fa@rv6BIBLjLX0e_Q
zsy{kyt)s=3PzAVSiRj00m89<a3|!U_qClT?9z8BpO~-5i$3W~bfE>+K<F^gMcL7i$
zI{oRO$1hE!%;Hu4QU%S20|^X5ZDR>%YQAhVQ+zi&=fSj3KLb(L{fxy3m;aIV>uYUj
zp8=04nWL(Zw>|V_g*}5@r(M@VlyX_CUG~TSeIhL;)F%4Xjo2${F?VMmD#+?az*ATD
zF`YMZl#93aqKk0Kau?r2xO3c}WKEVN+~b9Jgr_Euv5I~|o8t1Kipp#s?3$%_X40aK
zdNb>|Rf^_O>vv&2qf+#i4xC}cxgezS>UL?#w|<8-#npLc>zpq-)@Hk|f3z>zdM$Y#
z6zgr>FjRI7+8%2tz<nv^=dP$vzi>+=pe8s%!<mg{q4jU4E=VBS<E14aW@-1dP@SZ_
zw^O<>P!heTRN&Ba1%(aUp)<;saEWSIU8@<zg<Rd%(M*)LF7NcDiDFnLgQ!G*cYt?0
zWnghIw{qPH+I2A1M%?R8UrKtf=n`Yg3m4)M#<VI@Yy5tTjQ6SWdZ#lI%(C-VPuXL8
zt4aj{E6atyDAw_vNaL1f`~?w2!$wr*CFMQ8M4N1Y#V2?oQ>)uMVPze4s(kf@m_PjX
zwfp8;GI2Pfg<@bNU8Cg9QlQJ*uK(p9&xdmAj5T%uz?xH6#W%wso2f6Obor`+YHC)e
zH}*<1<O{Uz1(`=4T>NiH9L%@s|B}y4|FJn|Wcinu_<x`OL3mSr_g}Rng72$de(Si_
zKq88$<QhD|Wg^YsifMU$Y6}KNW_|NSBJsFIqrvB<+mRwvF`GqF)0qGBFi~9B;r6}z
z4<8l{FcywnPp&K;ZaBr51X(+bshJ>2kcNbiD9#jG|09Fc@ze2DsW^DF=?>cn5)vs+
z^qNkb-xhbyb*_w?hs<Dj;<+u2h5OeNCPZ8g*ZckW54?DsnRxH@;n(^O{U|>^FDJ_j
z`H}~VV^|^$C?{;%b98ZV1Wj7?{*-I+A0rB=N5@7RY@AkV8!(=#3-9#Q3W58QoCcPv
zl;%Hv`>wVqGR~D&?CQrrtl1G1G|yIgTJ#SdTFbB^==B^tWtQ5@c1xbhXSm2=E@>Wk
zi)lr7&3L)_p;vP81_`NuG!2jjFg43k-BlKkR3p4jDgZZKrE5<^4r$bGDs^47D!>dU
zsm**8D_3D_Xr8(h+^3R5rpKOynkU2#L<t+hhq8=N=;W8A_u!yD{)YGpQqsEwDVt@6
zURvfu2foTS1dlo}{H#pwbrdHJa$8={lAR{;WrZt@0Ogd3m@=i+l8fPuW4x3CPAUO`
zs$K3`KV0f$I~U79MhVO?zeNzMU`fbr0UybLP$qXpZ$YE{3riB-+qj?~@FI>d$p=U$
znrgnH{5qcd{I2)YB@s;RO-h3BX?%BCgO>)JWx}&@#)dE>HSRPQcvfF(1wT}Crw+jG
z!}HL3-3T}uW)~UN_GaElNez<g?Gh0J+2$4x25Xx?UW~5f@_R(Q>~0P4b^kq<w!&1Q
zkUUa&yyf1N_}7FI+C}ER*gb@4AnBQk<fSfV1#9Dh(G2@9sPVVMi>sE|t70?G>X{GT
zP+<@Lfgr=XU9G06h`=Gs_iDp2)Vj5ou_GKg2Zmd{?$UE4PlkT}+3!wS?fDNXy}69d
zaNb>|h2J{I$3pikF1_`f0mjFe`s|!m*cSCUck5is>4*{Z>W4S#1#ssS-<pXRByet&
z<-3BdFs9z^o2<?&U!H!j9cF$PFd1V4SQ<$(1QmkSHQ<tr``Vd@Nu6r4i13*ih?D9;
zjdDz&P2sj~5Y+U9Qfh}<wN`80N_gd>*eZGjXqkmTwTuae)-mi2S0$9We5bz<w}#iT
znNa^M2Puou*(pNY0?Z5ppS{>%1r43U(V-l*xw=39w!8m`K~08joa;r2Dpv5-6qmc7
z2@c`uK%UctHi4;0HQgt|8*0fh(18Tf6Mi?>?#VYI8o5@IN4v8C&cad!Qo0z7fCb12
ztwN)8sVJ*o$bc!g-o6B9H`%W;H^~h^BaC_~whlZbBu3~JMyydkhrilBCp?8(8lEMf
zN3L>$zdE|AgFzBK$p^&34vSn2bX&e761G+~%v+;lS7i?ik)@O{x0Pl6*mNYRI)ht4
z=3@rma>OZZ(^F1Me&Ylfoi-9^mLB&(V}O(!C3@@%gqEJBe4oO}TJo8B21KO|ckRkX
zwcD4101`b*ES(ECi}wZE9c9{u!}^OJsS0N)<y{+wu)mBmWt=g}20C$n9Mq}Cr|A+G
z<js!xVf^3PEUF*Tjr@oa@7CgZap*nrGNE)9gsYCEcb5Yeodst4bDE^S^{vk<EdN8Z
zQ!NUkHlY~pl#FnD&<Ajp7MI3KJb>2B7)|o)8vhRv&Rf^tdYg?{xEJ`|^9SkTQ%foc
zl7CD24gTp)r+9<*xk}lRCs+HmNuCqCsjkf80wGQ%+>!fOVez3fQkc7RclksI+spyK
zpf4ynj6*vIJvdoVjZ@$xJv3ly(22YBFJF*{Pp?dL=<jk0hCjFAb#5!<MAZj`5>`*?
zO+RrGm8(P3?YhXll!j*|b3n&!D>usm;(BL?zd*^bDeeC`%Kw9}%EA1<C;9&XT3M|6
zZxh0X@V_=863t|R<oHSBg4HH*T$t?+Yv+VB#sr8d48;QJMMZCOzdNMpN+g}_B<Ke9
zvv$2Y3ei;%#I1$9k@K0@GmeRiJcy8#co8z;YLF&UNd+XiMC0aCqxBOSPxTn3R2fsX
zhvT1}@6I@nS)#3r?@zu9*pcU~6?|W;>FAAyY~SH-;R5@%zFzE^y9dX!pXbk4uZFJd
z_<nRh?oNeAppns1lz=3GV+%VLLqN7-k1FP}x;fL4-m}iXg^G)YPb>!tDTs1KMbS<L
zOtQ*3iN~f(mnH@M{KQO!kx3(oqw`_-!|oUH!mlbvcZH;2I_5WVi-(-!{;eV}UHwN&
z7zFAp<4Q-icBwks+H>F=@{jb{l~SJ`IONfyv+kGO5MdFQSwq?i<EUgFVYWTCY$YEH
zZ_B`oQ!1Oyf`hg0G}h=Sy?ghlkGlKSE|oC>5<jjKDA01Oev0RMA|J{}-5O+fW3i8m
zD$IZ^#VU|%cd#m7UN+wxXhd-t4rte1ZoZ&yknY@vlWQu?t!H%&J8UJ5q#L=eBW^WL
zE!`KWR2M5~v@>>YI$l!~jbnV*I1nDr(X3@+smj{hZGRt21_TxSvFyo3EMz&acKq9|
zr}E`<kLQVD*wOCt7$D7CN*J*>*U8)kkk1UhobLJ#iZ9g*MGqTn6*g_H^oYRsHkQp_
zUhb$?mhrf5^IuRZnX9DcP$cl{3_W~|!Gk?l0eLC#wN>+*AXbd*2>1>T5)U5_F-wK?
zf3zW|egTH|$%CDI*j_MNs6(y;!wLr+d4Fe1PjI+qm8kkW1N`<{M@_I>Tk})x3P-&<
zwQ3ZZ1dU!-!tfFrJPpulunf;4n(VyNpLS96_4$#jxG_sW12#KC3_1Y8<>|rdTFW~9
zfmaq<=-ToJM}m!kC+%};nbta+q4UA-j<m(@=2K<F*fr3Yd{{GWOloxn!yNj6PdqS2
z%k5q|>{PJ5A-+QfJ6vp$?azMp_b^>UL_1y0s8HaA;R$F4G&jIUMBHnI8R8AOa4@1?
zN*qiT4K#O3`F2VBXdkmo>xh7Fl!u*6S-co6D0G)6{mzbe6*=~65BNFtRtf*>hh+W_
z2sb0=|DkaI-?i1WhOE<J8)ElAFa{}V#l*lt5J0~X&);>a3bruZU026E(zOFPQA^PY
zyzftzIa#Rqv$aO<#&J;u0R;0d-|r5=9(V1#;kes3Q7^C8n=trdjqpIp{Kw(#L7{X}
zTj`QXLx)z)uSClJl;57u)2E4p)W3V}Q+z9(M#``*t2MTjJ2olQ<sL6RmKUGG7$pJK
zT7ET~;=LpA_>RpKnifU(+>j%rG~?)o`7F#rG>$fD^>z1fY(LmE@tIdGT<twIUykiF
zFFO2f!rQoYTCW@yvYTMcUvI#%@LKGTIpY^-p7N@Thl@7PM|XnFyZ9*srrSRXgN{v}
zuT2jxUQr&L8895joXR0WPTTpO&erWpTc-H-xkOHS727=y>^)}8lfQv}iPtu2A+)%z
zTdCbzi-Oi{&TBQDu4NZ!>cz_ai-QZM;ulwy!p%-~cOJ|Qnqp+s<85^FM{o*Cfrp+_
z;nrntKM)y@SNKoX%CM+Ej~btnaWyU=&G<Ddmvn2r_qrn{T%N((I;{?{0(r()vcDt`
zQfv7>_6Ft4S;qy&qP7rMuGjK9DeF$W*1Tst!U$e*czjT%_I_QFVu7S@><r)=X`H?f
z-w`97E#mI0IaB^pwvlO1uK`cYt`C8L6XvH}kl<Rzmy9?<x5%qshu2=_AaMSVnKgJ_
zHY<(n8jy!M$)NEIxS`0g?z``7Q1Dh_AW?_}X$Tk~Hp{Oyx8gw|d|Fjf)4_G-_b7}$
zZr1G8ex%NWzH@4g4r|~U_$eV1LJ((Hr#EkHHgeK{zTJ|6jI%kF$RX$c*G_J4Tce28
z^YFeVLO7@-X}VdWg?n=v<21ALUz8A@Fn^(4L(k3=DIWGdoczz(7lEOg%pd1Q^#vr&
z*_+*{!KmH}yET@3qj*lLXk<a$`ZeVc*W!&WFg1OHfQ>_cPdyv@XQSf7Y$q5nAYunr
z?`Kg|hahPiDm<>>5IDF!k0HsX5QVkKpb`%|w%fXca$U{kAb|RV5YixV?K7}H`hb(R
zZN%ehT9<Q{#*N{17M*8{cHAZxHdz!_VT*=g9mc~Vjs`wB2<dq3E}vb6*jqSPr?3`|
zJjE@4jy$<61thcQ0depyTP*9?$NfMn!y>R>kFX$k9+dJ(n{GRA9${zR=2<%{ishCU
zhH~RxS+7G*oDQeW!kvt3m1N?a-_CGMWhG#7e?H4Marnx$y+>^O$Yk`((%afpiU{w{
z0LuaWV!=7cT!<)K_XhZ6FEl_P)GTRzEKj2Ts=!!hTKczB0UR-n!Cwp5Bu24PR^Y=&
zMmlmS!}{Z2W1?wNsl{eI-M4|Sc$uppf=H}uyz1MdxVj@u`R9**P!7f67igY2V*rG^
za-bA|icwwNP}b>#gzv28we%|i8X%(56=n-HCSESB@%9jgXhX~-hctlc&dX5!{_$V@
zj5c*ZuFhoEJUT4eHC-Lo)Ev4DSC3Z-)znuvBy>cHJyZ3fY#>8hT?1ERV1~NH-lHH<
zRw}LSjvisfIIn=jD0RR@r8nx|vB=gtMF9PUxF1D~2?#hy)ui}HG-{y!0ts7fmo5FS
z^vYQf*R#V>jYw9e5LiuiimJB{5YOdE4fB;SNi{8|gfBu(UC(}0B;rhCDW*&(S<e4T
z75>Q#P%s~+!LcIk>n4)V1Br(!ffj^o^Z@MLi7`N0<Gf@G$7(a6QJY_q*Pc!UAQ>-<
z0R8=(14?>y%tGD+%0K!c-~=$<Ln6JyJO<=5amZ<tjSFf1WvBXkV)W-gl9eG$dZr(#
zw=`s9luR&@1C)xp4-nRSner&<YxC_08^j&3eY*!~kInd(z3u_af5<q`jqn&0s5Z$z
zMpS|(>mScdX>(IqDegUTK-@`7FV$LEG2%F!XueuU1?eED?IG|!LE7;sTS^shHI&HF
zMbkGtk-1VLC-jhE1%GI??2LE?ufnW?ZiWdNnceuH7irY*`r>>TEAbzB=xMYw-Oax?
zI9iVLoAt<p_+Hs0RVL`tuy(hr2;UEAw&rd2Q6ll@%Uz~UNaCHEBHgZ_fh(r=Z!+g$
z35jF!+m*0TzXD?H6g0cr*WG~yPjvCU`{mrCSwNd%&7TU@&d(j(oDrxM;uE8nsLia&
zD+DKI1|Xn2Js;Zmb0X?|DK(9xf98+37uxgNK&ucc`X)A3s`-t@Oh-l!^)jMA<R;B5
zo+-=mdlL_CCO@}AR^PnlKKGz3Tm5;WtUa3x&D)EMehZMsEA%7*(!=USZ-phN`4dl^
zAt>6UP&sD6TOchgtnVXvBX0_;!z3gKV=fDc;}N15M31bb$@Dwa8p;*Br4{rm$-C9Q
z?{bVLx|2p9nXB}xJV*2F$wX!$3HN_PYHpoPBA|kgq?M}5ozvHGR*m|lmu+u-xffE)
zx{4OSFC%|?(a_A~bQ$I(QsRcyl%(SDNp_~B#`7@{dJ4?!9eSW%12u1HqFoj0j((UN
zNgVK!13U}D)3hLO)Lu~66{gJ7w^9TkL=->hs|%9X!sj&j==G5(=%X+}y%rO`6uil7
z$mM=nT2=a&H0kuJY+x4cr+$N^_f&^A`0?sbLcH#x5G#8hMjU<+TA4T(@A&L9B|o&y
zY?w}O-EMz?A^k5T|5FRF{QnC%m^c{zcO%FDT?-ulkxtrdL+qKY-6K`an1C4hRE&Vn
z(b-I&-jG~LO1cInRV1z<RYW6|pYD}E6pYNv3mY}EQ{qej=@)Y=ynkLLuy;+}zj_{i
zz8-E=|KEr#+tUAo$l8>A@$G{sICG>R!RzYuc)Zoq)28BU>wX{I+<%U8qxb8<Q3?R(
z?|yqe`Dp#es{PuW%>D6we0e?|-PFtJKB(e`_+Oy_t~bV}`fNIT1!`glWU$tM>?^_o
z#C@_(9f8wJ!z{bY0qJ55*P+C_h=XtkJinKARbj&X@XOhR6DlJKq+`_bNl~HA8n#g4
z3JaQ!s($?%u=lQ-ukE5L_Djd6ZIq<DigQDHPX^7?*B*n89uiLl+OI`ZwYN-FRl7(R
z&1!m7Z0%>=n#W^uKXtnYSQypVpf^j;DOwpKAyG1hHCWu(4$+)_v^3%BMuXBrpZ4_k
zaW_-c=L!rv@4*tOaxzq<Cdi;XXi}8q`3(>gm{J`RKK%e>DP|E6JCcNC5->Mea0DSG
zhlrxQ1kpXOf++ws0LCRf2yG=@wQy>W1+>JL5ClX_;wB4`ZpjFxa7u>(tTbCZE#=vY
zYwrVcW0ki{S53Lf=Ye%&;cT{fqIedP!#_lT6Yb{h+3Ol+!SESY%`{)rK0kYQ`UTrA
zO5S~GAFI|3GpEPqu=9Ut`-`I$F?MT>UOTql;!2jib+^ISGd1xt+$Q_sdxA^p>$b46
z>(?o1^*tN~03svQKYFreuOAqKG=GZk+}<sZ(RVC$PBP~>YI;yEzvy_@+dTrlzWL3L
zyR_dwb00q)4^kN!Gk%7>?w=bd*pDA>Cfqt^PT$QF)i;?~jwZ)*f=ot8#r~Mrp4z>m
zOaaj=NrIsmnUtcU=m)~vmh`L}`4_>02`t{T%9Ds@q`Ma#YEa1Q8x-=-qa1LiT;(Aw
zK?@$UPN6S9%(U9kjps4;2kQHv)Q0Oc(3F1?#>bNj24;gmIvVP@oW^5m;MRw#`xa#4
z!HB$+asiSjs7UmM77gJ!ql!S8T*#;=G*$M1gSk{IVU+v!X<x6`fG~xQbQ4^q))0rf
zl{EUci2*TnU$Qjc>8~30c;B&u{f_U%qcM0p=}ZyQq9#?wr<(#o0+U2PAXdFh3qPTy
z?0hJ}x8zfgZQvCXESV39$`9J@5hUyj@TG;EjD2|X)pk1`GikHyJ_7i!FsI~;%tK-U
z(dJD-TQ~x7%mBgn`@iLv&!F$$)KCoweB#yi>Eu@jt?N?qgHjlJ&gur>1``ky-2P#1
zBs76o5Ki?LTe|y>*N4$=(zZUO%Z3s6&<W038E48jP>bdVq$%aefQ8h6({&N<-mDQg
zEI>zEeHT_t^@wjNBOC52X8!rJ6tFk-u=jVDx^8(_F4Dw?xLij!TOc2S00BrCs(*H1
ztJZb$p9z*lAHW&a)z{v#`DRG4N1689Ln4hDp`Vsp(l7#*5b3M95Ym4%wx%7Gzi0y{
zha7fE;^v6A+63jzT|7E)cwVV<SGfhUXj*+U1q+HO)cFCJjPh>)C*ljn<SThFztp+;
zNqePA2lH;BM+zh0xIo}TC?hA<Y{1ZQCJ$%(zSXq@<;TaqJJuY)AFiK)L}K<WDm~yS
z0EMMdPFO(!X(P;B`FPaw{Il~ID_s5xEG;L!#<#HPL8vk*NPi~3J6eDC%m?Y4H^adx
zhE4DSpOjG!7sbO79vDTu%F5?bf8s*$U1xO~=ug2Q!}IjKWMW80RM?l7Hf0JKKICpY
zOFISN^6O-GlP?ub^Tvez3^kATnorGPBw`ztwB7qQ$C+#pZtk~r0vjM!NJ3hrVFUOl
zUCyf_35k0O<1G{Yxac}5h+C7rv>SxT@cqxI7jA|rg)moVF6?^bqk}lMYIYA8%Ce-+
z5nM33`oU^rTs{&k{BXpv%ESt%2I$Y6`FimRxAc7*saTQd@ZW?xccRT4GCrksu7^JG
zsYF76CuJsIE}bOuef<Lx@T5{@KF;iiO$P-c;(tA~c56!>$1SQwI4JM#rnf5W;A`l3
zm)lW>PBf5+<pQ?yKl`eGH*DOsj}&(vdQCbmOh&TwDv%B#jf}&#TbV5PUaR`3>0C5R
zj~+WMVUsNoH>h`cfPfG^*QoLxOdq+5o)6Oa^I4Dq<N*}N&f;@-an!T1+tlyLx0Pg=
zd+!aS^aK_@H7&@u*;1ALX#b6iu<5B%%T3+S%3?AbqedUKssMaM>n*Zo1KDawUTs<n
z@yn}_R|z0cB61S~Na)$#n+cOW?UovuM*$dZIfSxZf_-~O$3a%H#2`dtPsF7lsnAW1
zE-6!fz_ww$fUQzwEn}aoRo}98N<<XK3j9XdCUuSV>S&DTUf5cEb=_wu>0z1Lx~o*r
z?12aIyD)Jg9iq_+W~VDUA%#ZpQf{j>gS*0;qQn>eRO^TEYL{3OSeP^u&w`Y5**a5W
zA(X%jx*xKogbL**9Beh<!sGg2d2+F3HX_eL-S=F;h+U`UrOtex<?3o8R-IPUT%NDX
zs6M1VP~m<sHSF<M#cefNTkh@K=4HyxW<eAxyUzX+(X?k5sxgi!FdEUwI<(PzL7WgU
zpjfMb3uPua$Y{`;t(YgCzuPQx>1y##c8Pa`)eCW&&yY$q2%~;yfic4{{T!dtz7W)+
zjbnugmHM{D4x8f>PvMNbLeagpzvaKX!v*1Qp6oniy${R})c0*<a3|`1iYJ*5my7AN
zJ?g%^Dyk9YzbVbT0Qo>K5)s1?F(~=-iC5*K31Hp&0HiWSI0jU}hNc`O7})LpKzlQU
z^}mEcL-5g2&dGyMCc>~q3{ToclM-8=M;-&^ed}&?#N%AiPUp^M{pjrA-XkuZogaLt
zH1UG2!nhHr(BxN=sbhqFHlkV1+k#buK$wO+D{~m#^^pjvM06JDHa{6hmkK%N9(X4W
z_`&U8JVSpa>7O3pj+C)H_KlFz8BD{h9R1b;86txTrVXIoz695xrdD#ezW!al<`}Id
z0p1XRH8s_yYz~hxL47VDA#ASj@)NF|^WUlfLR4sE6aO7apSSt3D#$D;dETRIXABoU
ziRh}gARkSswnWI{r@z?f+{t#X<A>W8&Yc1LK!8+|b)4PCg_o9z3pDOq`$hDSyw<_g
z58_g=^P&meH>C!PGN%UH$mC7g;CD?0bcTyw?}7BAxEk*!mD$ok&=Ng3-l^vi8^0;m
zNs2(0WLi`psSw!MMApbYl<08M2FoBf;eF{X$Ow|WMvMHl0Ku2ZZdf+7<8VK3LFhC(
z_|^r|4N~sx&H(lV_obH%A)$G`K?ip&a4n<&P3QO^$qIcqC6Uz-&Au21q3_sS_SX7r
zeM#1|`=$sL;TUZDbZF(CS-Bz@CTF}(57`cob|<>uAIiA~GSrv@dB0F&7<wmtdUu*2
z#YosbVQiOzWn~%2e#2CXh~`qvjA*oMtOZzf)&*t>VX9|dz#PF{6P@3VVGLb#+-o^!
z31ziHf8fXkKUQs}3s3;bF}D(*FTs^kR)Q@uwMSPl9)saVrPtlj_3l~w<|5S(cGW7)
zFg_d}{`|*#>*MZR@6Y?_b>l@xnC}~*O-eVrrTamRl@~|0Frj3~#3UGbFy70G#>Q%2
z!8-@}LmMEK+v$_G{%H1syqt9==h0`1Mp2(T1WYhl=Y)I5N==Rkg+!LJF`;+)B0*~c
zsR(v~rxoICNXDnJq}@;d3%E4(_4=Pmn&Ur|G$RA^|MbRkSYz7lUq5i~Y3*M1lNVZ~
ze4x1=yu*q`<EH7S;k9lCo_RzAhH_Dk#QL*eFEETi0Fuq9MMrly0td#_shz1rQf4q2
ze@{nR@5f6uN(3R<p(c|aDUo#Y6vSVubo!)SC4;l|+ebg&&lE-E1e0ip@qjR5&&S0}
zz0*|u8B^_f6$VE(WlY)Ho$rU-T)kY2^qJl7&+n^)=jCJm_@n8|++7`S4BhwXT)N$_
z%k5A6YnR$G^g9VIF^V9c3<h$FNysF!bUT62%75=Vs>d>O-DUo-v_93Nv*Ob7^Nh`B
zj7xi4bq4lV>+Qr$jq^rrCEV)-f6H+JJWS1`i-USI_t$FckEUz4UiIXlEHiZ{sv`-h
zDoI#r5>w)E(KS@ANxeoPY|C)XTCz_&HZscFTGM2M*g#)_aJAx}B)X^`fJ<nx#rO1m
z3&q#$-on=^%ikYj)nw(A!6Xme&rIw}^*KRsT$K5|c$Pql8O1o8?U!!$Lb=+?5<x`g
z4)a&p+Zd<^qbmkf?)gWPNF>t{Z#Q><teS|0srgCc5y)vDfxtSMqFr1Jd<f<N4?M$%
zZxC{6gq>V@2mPTV;IPerK?(ryPKIg-)_08n)X+$|)a8rl<8vr)$hd}c2d_K(*JKNK
zy6`2>sSi#M_Z;Mk<F~n2jBIuMkYv^Gxu)Dmr{tr9!Z}B`O@IIRuu}<jmdQOo`{})3
znZREsF^=hkYrk0l_yxyp^v;Q#WWF2@_<NGY>cy_^$`n#tl()8Z)J+%A_dq{kx^aDO
z$d=20Ca@j+ZFbbIZI$oAeT3lL7!#_6s0BO|E?xyADkVc1UkTClV-j}9yX4bSjlHn(
z5Kh?6UX;$w7dUqG%K}TTt=pCB9^HeF4wx_@Bn?+u+XAG_ml<Z77oSdZw>BMciaPoZ
z0z16PN~VIYVZpK)2Sg*IK0j*ju*rhU7i@Vo)=5XL0r|!|?iKOT1b|HklCmYc4GDW_
zd)7m6^+7p%X3juMU`{~HB9u6z^$S9ffHNV{JihSz@uODELKLWxVQdzy<-NxW`i%3u
zQ$(i}MC6LTFrP&ZeyABB4lz<edp<hIlq&_KtILVoj?V3io(e1Pu>6ac6;cl4u+5aG
z%>IB8kp@LcrhJ8#nl)=Hd}Fi>yA7S?TJ|?L#Wx9i6;j(8h9I5@c<CJ5O*jCMsKnzX
zm+OKDQGlOaVD%%DNx+-!<UJj6q@w68Vi@u~6e&#_wCYaRb5T^jjDNdqeL_bUk>bZX
zOgiL*ZSj1P(ZEgnA|90ljssUwrzq;XWgrb$7gop4F-m<siP-#-0$iQ^N(V$F=XEyH
z8x3I9>9QtQ0?})|!qaQZOMzn0z6}l==1Lo<ypIv|RSo*dsdjsIfb!U)8J2M%plWD0
zzDFC;(1HQv2ZThNRsD&cf$+NI<&rC$3zKuw7T8!m{4?GQy4R=$Cwhe2ZdHT9Lu)iF
zcrPkjo+^SBYPlZ?*W&Oyu&^|0?yH8rrZEIGL?*_Og}KNoV6hgTeeARuR+5$^O^-5=
zdoK89QD%U8Jy;58%(t&<I3|Jt_;|r2OqM8|7j28)FQYN0i_Hq<w0eCUi&@V3_Sy<N
zVF-0In(9lRRpLC5tD#YycCWstpy<2R2x0V5-iEg1^-kM<)1iW=beHB`A^pF?HvvYn
z7r$s4LOcvxpCG)pbR36}g?lU`BjGX_>X2=r0-_=@))seR=4hCLNHwBo;cUoiN<$D1
zODfq7V?e`A#(lwXKB<=ka`Y2<(2yj62j(hA-aIxoc&<r-3}g|8KpT=bB9Sf!b@-4V
zK0q4g2oiJUOdx=X^#V%*AZX7;t_t_jv2Ep)whJ*#IiW%X^)jW_8*%!I_gU@!xmfL>
zOQ!m#G}xE7+&#!zkdls2)$BnVeMbq8vDSajQ`akO?Q$4WgFPf8YOU!WyV~Kt#^$0$
zX&Lmv!%FX5kOX6ttGwi&9vDa=B=X;`28*&b*em6N@yRrWSjWRuJk!$bQ4ID*FoB^j
z80dK;tfQe4rFoHVy@S7^H|h<m00>Zm?iYC1%G)u_X0pg-@Nq%gSrfCE7Tg7t{WZ#R
zX}=FS?f&o^=zHQu9gz84w(B_9T}E94ZklRpf`ql~1MZ5AA+^|6bS~hZ1%1M&eKv(|
zxCZhBY!RxAk*gl530inBuqH{GZ9@DNG4?~1$Tn>EuEDwc73WNo&;Z~7mNUiM^9F+1
z9xZLbLWSH#OOZV#r;JS4$BO$t@~&yV%n}Fa@DTx}>xpE{t*yQ&9?81j0`bQLhc@Oy
z^|PpxSy~S`S}cNWGWCNyjjNf0FuM{EEe)C+OI4x%AyW1h!~Di7c}s)?45f0}z#q~R
zJ!3eVe+x2Z{pN)}(KBf>k<lp8&0#QdC8p-%35*&!Bc_hk0sLWF-_3F{8aj3Tnz=gH
zch*}<n^Wi`K%-5mPE5DYx6*3B5kS*~t|@V!vA@d_R53Xyi?`34tVYbD8b~WFIXucd
z*H>V(prcfD&+ozb=ca#@Oenfyo-B%<JW354+Era1F*2WlWF^WGB9)iMF`K==G-IEN
zVe)1P9ImLoeA)49fM+JA5liBizK!&Z!-sQcF&@aBqmlfgF}Z70@did)E{p@iY|`8C
z%c{&ECaFKXsd;35zgs@@###heBH^a$p{J>K0MLvJY!6#6+0Zb(LL3zkr#4`XlJj=O
z!O^L5x_=rLj~rZ<im{EoHy+6c_A{1vuu+iE6#&s=rSQLY_Qiq_cs?M&<vR{NJ^ot$
z1vr5`{qtW!gylaQfY|=WLS$J}D{hkw%{NysUmO#ou`9}Y12kip?F;~RZp1E$U~Vm=
zo~ec9@6i;P=<g?<E=g2_b(~J?d5;E*XbyL;lY4Ms7=*Z8jqYqs&wGMCu0;{)-}x&(
zLnfqu6{n&sS|Y8C#NqklDiS!oUf;Ll$xt}IPb<8;(`1lwe(v1w44v;M;xiw{Z2}q+
zdk1CahG;qlvm8Va${73MA~FQ!k|g0m!I$POq0fybYBD9!4ekkS<%RO$XN&~Ctp^g7
z_K&G@IUo_iYsWx{G{m7Kl35}yJ+6}!`E_!X*JDZ|EOx`baiDmFBM699I>IAk3>XP`
ze{B(?ApTcOT#YiAF`|UGe8(tMyx9c8Y6Ho8qv1*sIh?71+Cw6oIK%nE2lKBKF`Y=~
zQIa0e$R?5`0y1vGb6GQMA+Qy8#MlaoVAE4UfWNRqfpDB|oTOj}E|de!0Pb^{N`zrp
zF;-eK5=aRIuColF6d_L$%<#As{<P|cHE`z0xwozwa>0S9E>Byl?0N3wRCRs7UFpg(
zkrv@#`{HJn%2WbSp$48^<1gE)5nm~i#rE*!Y+;7b)?-mr&h4nmkm*{@9vomaUo&=z
zTKd$t#e%q#|McC?Osdtd;ki1`Uza3R&NTh0=EMyMC}A2NLw`jT(R`~|{FR)5z&?pJ
zMpMgIj<E2sqCCN>Xt8Qee_Wasqb?AoaZuDlGYU2VkO&ZaUqm@X5={PB<xYlXt7639
zbvg!!MMcj3IpD9q;|S+MK7HtaVD=trr>VcpbupdP_4UDYQ|Se1;l9AwV3u>0vIq#6
zicu)_8(c=0{Y=$+b!-SVhzGOCq3X?hdOE)S3qiX&A_Qp0W@laZH)u@ve6gwv+|+y9
zK8)<2Sc8k^5Y<5xqJDh<>{mD;MZpJ|aWQ;v0p)cmIDV<)t@gS=;Hv9-h!tXS?*661
zK{9Q4Z$mhV=2McF!l^c*4te*9K#>sU!hl1G#NKpMP?+82!Fck6(_|!tL}8B7WHXh=
zCY6NRk0Ceyf!pi@H4(TJ+W?0UU?Bi~Pk<8ZRI2eA@Hk^D;1LA^U8xZG3V<Gqp8-+Z
z#@AC9{GINqIYfc0xnaw;Zb5tnfT1v^lO{zVf?Tqi5P`(=qYw^)iJ1fy4rKDLj23cb
z-H;)rQa##zr7FAVp!GhIeTN^Ja_P70q@qUa2hQlK8$+;bCLZQ%@2GDlmn*z$#Pc--
zzxsw0N!VH`=K=&Vr;E8zqXrjY_6zkx``g1Jr|!Qj<R|tcbCQ<a2)79fzdYk}t|)!F
zBz9wj_z7YYM{7i?YB%^R72?Bc*%c_`XL?hi0*IGuzqrO7SU}#&3GLs0D0?m84?-9j
z0MLNCXshN1>x7&g%Y^a<;w3>2PHeks1<Vy;q>?}!XL1Al3@<YP<$2Pn%J>Yg{Ph$_
z3?$>#&jIU$!IX}e3?=VKlBy}QG>k=Pdib|n4IL;}$~&Spfa-D=6ll8iXnX}|MhH#r
zC&PzPkThaA5i|Mp=M;?*8iq)`cNJMAQ?ACRiezs=7CtQ-Uog9Oszw*@Tin?ci-mNT
z$2>qm2qhqwfzZ}hLlytYoWDlZX*m@oZTX2|U*~fKC>jvF8h_EEgfkQ?4CEa<Or3j8
zzfh({0y%I-{Q1}CWt2q1GV;fh5fOW5PR>i<wxeLrY}G6dD;SW3NmFY|YwAPmy5T=h
zu$ZS=Da2{UtH!C2S+$=Ez<y{RI?5k9V+zf7I$CrOHa$%znx3zm&e?6e^dN(x+&&v#
zYBwQ547zF9spa2ogOBs@C0Ii`QaBUXMSb<uE|<~uJWQvZR#qC;@cnrS^&h@lZtow%
z3klekT3irTMu$y(QdN6cdhlC5TYQ9%&ITl}U2q^+F~{?wfJQ$oeb@+thcvf2GO@<L
z(Zk7}v3OOFf!!C-4?G%Sy0TM;o76pL7gc@L*Dd(#r%vKR-^?W}ae<&gzu1S{fKwt$
z>YwYt&NPf|Sq4_S;NLPLQfj|bKi8LdyBrxj9eGs>_nI^B>p_6SQ#Rh(V?uYrP$3{B
zAd$$bih43>+`U_=PUB1|1d0Y`O>Wxj7c$gaaABFRBOHFO0`5oqzt|a#Q|-p>w0cse
z7H*$-ugZ;=2I~1IeAd{mL80WKmMi~yOiqho!ag-4b5mCdnCic`=c}(F-Jx?~(KEG;
zED&9uxV8MvZL}nB_gCFVD#?#+M*8sK0YS5!HVBZ|;4VUy<3JzWv|SZeGhTLGwaaWW
ze;}K+eBFie<y~8Z24~+#7zc`rMQ=6&jkQs%rsvjgc+!J;Fie<T9Kn;y>3d#;BiS-y
z-a9sHT-vcP=xA*6N?5&VpSakdHBX__Ph#US(ML)9&8lHbuZL2*hxbL--*RFUVJbk?
zUg^#{$K@~#X96SLrl{>yAFSj$x<v>ZSp<NS0rcY^{O0Qho{s;thP){`DO}sr<Q4hU
z{r)(a(h-Cm#&Dl)61@6`rlMe;CebF_PZVhrgF0xKDDW!Fm_b?aOBhxN8Q=je6Sl9f
z%M47{+ikig#G@1tAOAWvEFd_4vz4^xA|2U2-HeSJXYCLdKptw{jqrOt;iHpgQAHTD
zUxb{PqmowN4?EGK9`Ssf*KYZjVtl0#m~eisr^4I7M!Or<*Lww4C!b_bWAwH&;@R8t
zJf+{V@kk;BbvpHYcHDMwSH+y)*jCgNu{l$sBCK$}Oi+ElVz93TzxG7+=Hl5*{B`Up
ze++Z!IILHFzx%yvD)*lNgSsHle=KOvJC#v?RGB@G^1LZ|UMnBjCUlQ9X%DOngrfcA
z%w#uVE%Nkw?3a8S4qmNnl#LNKeLY5$2G3!v|2(%|5gAB~RnrBD)-!y_w-41{lfBZq
z+Qi;OXb}6{W(D9dCVkAWv;(x7-IT|N)2{;FzwAf6l|!5xq(I~vYT<9G1=IK)EzVl4
zG>3mji1bY?b7ET7aKD*VF*^eMi5mz0c<|U;&ZX?%Fk*BUN_gg{pFUaqPoebxC4~MT
z0w^o<|Cr7)t?`e(&xYdrT06ICOkgT8fmB?L`_F2U*)rZS;$;~Xkbo*4PTJnV0@3l(
zdK({>C_3zr-2|$Jvj3fR>wPLTv82hd<iW?474woS%S%qA*avLpdhQ-@C=V=D+QXd5
zeXC7>a0uU+`Z<dc<GBeW)<p@9>A5%`6~x|n*m?;0T7JpMl=BAYGAl3l>+8dSrtIeW
zaJqhUxA*l1zvbwB_i=P*(`rnnhx5;K%js3K6jKH$AxYA9%RShl&GAax?Q>OrxPEpR
z9`NUaVe}EgWKIO}z8%?&nDrWz&l(<gq?9CYfD-xC@oCja@WrQ3<0!b#cCT4ZKaU7y
z>=ykfm;oN-4&vtxPpZ{a;OvKari52=DV;%%Cv7pX!7zu0Ht@FWBsi+f<;tviHQip!
zd(9-uawN%L>6MzzL($anJ?CRdzgPg{2NKh;*%W{pA|DMcZOyCCqB<3p6aeHUGjx07
zhH#Hc$9Q3&iw5o2Xf5<~)31W1Gwh;jeUQ>-^UqMQA`y_*4E#T$ZX$VcRT5{#&)Qg*
z+$BqS%LpnkBBK5U_==o)G9yBgwSFv(v<ElI{kCw{SXe3lRV-w7+R3xRNRoIXj)23n
z$mDE8o8&}w@PmJ_VrrSZpTrv<PIEa7m}ujlZ=NMLr4zTL;pQ2HXs;W@{dd<go?)Nd
zdW*}IB-b*z*Tu0vE6kTLUxp8pDw35wO^`3VZvA^qroj&{d=HWcg2+q*{rHVAEHfjB
zF3@efgMEK<Bb{Tn+s5cKsc0Q>c8Z=+UoI*|T!J&%*Fia?<k3>s0%N9;7XQGWHl&ym
zmQ!D`qSipYHc(V6=UEq|pV!O+`~+>L$0^X0<b-(frA$5*ESXIWsUqo>z8lW+uv~Qj
zB9g7p$E3E68cN33O=dO;rOTGW<qLsVhsl}oJ~cn{HCKh@*W-Cs0yT6IX{1Y{fC2lx
zC4+1+{|LZtX*bS2!w}q*tXrlVOWLZ~xAsUL9`T+Qj2(i!N;Rq<)PeSICPL%vyQ{`(
zo}oD?r6cdzYCmo%ojd|9=Hv72!P5upix%O6DlS3MOOaAT<yY+UJQQ;gGyq{S!{WMP
z=+`;d(;Qey;KeyK?`9=KM+mV&XpJkGN@gc7lZE5#s4Mzo^uw7TYBfqK;`7k+$_3aQ
zrL;eqHp?99iZnwbWk#7Wnc6Np4y_c#BpsGiCCXU`T6rRM+Ndn6H?T23(`N(HoN4y)
zxVJbo5l}bfCdX4A87S&j=GZL>i;n`-5D7A%>q){uoEcpSky8pL+V_h?kv_vpX#j5i
z8PqQdu(ZG``_s!xZSz&$4aRW|TzXdkF9c%S7asdz(KbT$=L?8R7VL**Lm0ifI>SWg
zI$gLruUQqy2@ZYr3*wdN97M8^x7yOfCdGvWGa*Q7a4=lKZzFL>RQi%_iOKH3Sgy`j
z;H2g<%yT!`ew>kp(5|8=9i1^86D3}^db2porc4D{il?F`hO!B;BD2{<^1bBn5b>f;
zd;PD$bU@V~tCYKzxu1Y}*;AlPTTpKrq)@b2j(nB(UH5%KDrfaisc!jx^=97`+dadJ
z!7x>ps!g2)-M<!-pFHE0O?W&nd(cU5i*r5a_(At*b}xSQ>!BP7cbYgmQEiFwhS|hy
zL1={e0d>2u@F3roquBPBE@tZccmj9s7-GA_7wf`+E26ZiacvpXgCd}AwrIP`6PLOA
zi|6o!-DUpwE*hy1BRE>5$E`Nae(0EeC4fhj6%2Pws6wJ?1UwAA`&syS<r1yR3p^gJ
z$NuXN6HR6%yf%-*>T>RJp@bLdS9f-coH_=7jYCNBGwrCY*K5uGWP1!;PT=>;l14*3
zLXZFWe(H(?#W-9ttQz5xR}@BU`xmih2jakro$FryZX8Zkpa^rec8;%2eP##w&}Jd)
zy|v7D_brSV=$GFReJ3BWG4Ja4J&3YB4WHBXoL!B{_8DtoG-0%DTYe8Ip}q8Ugo-r{
z^_gJ44gl`dudXRmT1YS^B%oF}?oOVR{LA*l`~c5qF=qVNfx`M9##$!U|GBZYSX0w(
zs{{3a9Vl8R1cUa7%^=glPF_9<D3bFdt>y_JA*N!72300ZiC1l+zh3YjaLv&sv$CK1
zQJ3rPDRDS>TOd4d#|P8T!;{qgjRZ!TG>3bh=w%#pic{`IL6<gnlq!LuSB}L-c)Gt|
z_aE0`djCKwk9Vu^cC|lWtFLyozYg!mAKR0a60#8@84kF>@I?ufoo9i_T6>fY1SUK)
zNS)rrMWzlJpw#JgEKv@tHJrLNhBJe++K*b+8Cbpl+IjU=^iOwF+zFUU4oEa0`;9XQ
zB8+r*H@2UlwLQ3PJ!5j?dV7A&LZ(|?O<L(Yz80_1w^U;;{o-=Zy{)KPr|2I1<dT<V
zQcz6ZXWJ?HQooO-4>X!^TRt}YWvLc<^?TR%TgTYFtZ^TVbWwF2vJho{{k)JO9vj@F
zqSWzC+5D=!nC8EaRmsS1j6mhnr%349chJwAO1M}?>1x@RT4k<VH#0lfo~HxJp+-Ug
z0(lSufXIRQ5aZI-@)TL;ViWDXrf#*YZ(>u4U^Ah-aBIQ`g%<-&VwwwZXSb;~vRiHR
zP_;JZ_uOQ$4GfUjzNdS<`v8^gqJ#l|FW`iv?@7EnvD4-=wyvq}C@@lW!#ET)hl-*V
zemc(kW|-<9oJ~{hxGF`62B=ztk>CzXXzizRTD}XU#=lUuL9*ztcVC)DEP#PCPr|WM
zJ<KvbAR`MC&_Lutfj_z1yW3q3w(bwjP(@3;%Q9=*m&wugGiRwHVp~}`oC<`nrY>o1
zKQ>8=s_WL=^6+9M=!#uft>z{A`d>tLo&TMsp&wSKLoLTDL712PEP*vCJhkFkj5f@x
zDhVhuD_dGj<st*+?9ebcSjui3uD$eNv-Q#r5fD+?y!LPCXqhux^Uj&-VTo!@CweTF
z;lx}OuT_++YvNXqu9BeH0uWRx0m(B19knaNh?^K%m?J5V)@K|h*#WE=7_b=$m4d*}
zA1l}oI#@FOv)b-Rf*ETpT9(Ie>!`0yLn@&2VA35bBMmY#LvcjmcQC`h8+LX$gtdGu
zVzMX}MB=GQ%!2v##oMsH@a8Kvy@ZyW(?(;vD7Y`z5yb?}{B*pWbbD5DlN(YJO@CL7
zJYl37G^~~iIZ?#%|7?87$$7JR#Kbhaz$Mm~N?xiHS(^r?$U1S5HrRZ)M5v)kyCMZ;
z%ZNM99rWDDNIp#>Wm`;@L59U0!5G3VV^}IgjeF$oDt3smz8Q7iM{KvbIZfTt6%zj$
zJJ4uP&Y>)Wod}A0h1S$#9=m1v9as%7;44R~5f`lFY~LWZRiHJ@<{gD)kUlu26gn{n
z*G8ApEol~MA|{z^18OS7aPLDglYIrvRSOgb?CwDUn(4j<ODDJZ^d1h1l@Y*)df{Jt
zNDNG<?yt5X!cJrh+XF#CvED)*1-HJAq1T}#Y4`*x^!fx!m&TL-r`sSu1XMHXcVHIC
zWXK1q<I>}q6bQ)j`8>o(F(LgUEydiIeJ>vwlLci!sMK0<S1z4(FZ1FQL1SDPgG;c4
zl}8w%R0*9cJj=Y&C`H<Q^KQ>1g%t?FEb-@2D9yN}PJ`cRvP3ch4|DE(kGqv-ad@!z
zpA+TfI0qr?A--(H&>qAZMLg;nqpP)MU%I$QB`C=Xkj3<grASn|5KU#KLMbD?fFsXS
z8iqnAiHsMV9i_Mop`bDkS!zQkQnAcYcH!e~w$95X)1x4SDy%8s=A1PM91m3J{G#rB
zbdkqPHan|RxB^I7Io2Wb0wLj%%7LIc<+E1O8<Q_E1Y%bNVqN1I;A-Pb!1f6_g<?zX
zV+NJ$K&<eBNHIjbgbzVz`oo8{@fg+<D6ViNf+-n%mD9V1bU{Tsyg>*gJk#qr=5;;q
zD4gOnkO5FKL`I2=s49XuGhzy14l{ciBpzZ-TT8BEO}9URAW%L$554ZgzaFkv@c^}D
zP_g@-J0|znOp772gh}jo^BO=PXGQI|pH$K!KWZrQ&*0W~eO8yiy{yZv=RY!yeAm3*
z<zxRG^2X3=Oz3ioq%h{l*x>-!bfpG8vOeE(1?qz~$V%o#gEa8L?Dcqa04fGQT1V}H
z;cRSTh1o*chj%%zVP1UPw7a*8Mo9A_yYSj#FtZWVj_vHH^_KeS%&Y}L76l_@Z}t(7
zmd2kN^-(RPk!C1Ck*%Tn1LX2DVcsyP_oYeO*jmTP-6ziw1t!xha^5pvgj1MOEk5$D
zaONHwJrZhTn0iH;98%o_x1d5I$jPBF1X(~e4d`!0_c27v%5X@=Vg~4AJf9>7aV~Q@
zQ303&q|RhInbhJZEaXwXg@!ArXC~bc_Pt=pzn0LP5=yph^#BU|xpO!JUUj~v^u5}T
z^x=t7%Y4wCf{OJsa>(Q!lT^rN6KT>>(O7S#-p{E7;f?w70o)RhCbcgasFc_f6T(FR
zNLLnMwZ+Ubs>`VI>OR6AFf5)T8XpT1qZ$KhgD^Qn%3btV!U}LeUChe4&oe$FEufw;
zgtSsl8JeFut%~=z-gereiw9^YhUd_w2jZjzpQ1&32B{6NcB><a7$w$YvIkjn*n1Jy
zlz?J2!3-b{91m5^1yrEVp)hj<QQ=3*Am@@w0m{i{S~BeaV=wf9c+vg#{&BE54A2L1
zW<JY3ApnVKT@fX(=Ux{g85M}qm*>^m6xbK^SM7u8&HWzTb2<CmY~Q_z@lcfM-GSBY
z9(tA6;#NXZZ0Ff{8yT|>d2t|B1i@ysV)*1O+>|>EnD4I-c#{zo#Y)XqKg|B;P;D?p
zb!(ej6fr9ZyBgQ$+dKj>4cy#AdLJxOLWdh9*q_g^<%MQ7`*0b(1~|0$=927Q9ytO3
z?xJpuCt868#jNS1?Z<PW+#={_k^hIiw~ngv?cN7b5EKNIltz(|l<p8oX^<`{X%5}3
zfQXcoba!{Rba!)*?(Uo$_<Y}=@65Yqet*uad0j3Sa-MU?e)hihwXb`h=Tz-B_F!sK
zVz`It{E2leW+fsYOCeo(B;|MX?6(wxew0Z4>P*tNs8khy>r?c>C!}$GFD6IO>z;j(
zvmoYg?c!m771&xl{6?1nhA3)Qd`F_3_YF+*!e=Lg^{w5|OAHhDFEJrp)MIP!5TCzm
zgtca#DX!}3iaUkEprq-4J8CAG*p#oOG;-fzLKv?kYGfp8I57J29d4ZDWx&_8tz3K0
z+&u)3c-QlVuzKVlFH9bW4qAU$FL#tXR;>|@IZ9)!V;WGJB*HGr%;?HvPE*cMo9Gso
zVw((o@@cY?o7gT4UX!S3x4VbKJLPehV-Zjt%$N0aDdmhX<DV><<;$F=jW62QHjm|p
zr{K?utg1sXn1+UtT;W)^z>|@%buQ4`7K;otNB0yUZgp(WR4#bPES2cf3@=0$GOgR3
zjp4=h-6SVwBatPND)x4>KZHRC#iyj)JPU^<Y2BZt^Y<?_z%5n%S~uo*Zms{YOz4nz
z71Vcp?U6)n8HnEad`bNs+hBnwO`2=RqoKW<Z&Z@nb8u}UE-VWhXHH2s4?9oklEW>y
z|0Boz51*!FU|{{<7A?TT+wG=WF=savKPhHm2_Lwg#!#CGOQr~!;W1LRAw7c=WP083
zECC}m{ifDVgBXTCG>XGi3L6c6ZmzEGA_^&Hnw0Hqd2?%XbDvw^_nnYzWOw+J5Z~J`
zH2yj9>WoRtJKimBMy-x~+`3HE^F=yev#I0BR&;%LUL9||tOxmzgoDf=`~AK1W1pSb
z7%HAf^_#Qi!>#B*Ma>JEompqQtK*B~^xfU5Y&*BA>s^06JMS;<5j)6h1+nmTa#=wv
zgI4vl*k^Qt4l5l@VfCXepQ~xZ1FvakFnf$bSOu3$*Ve?Jm1WI+WjmsOkLwz5>V7=M
zC3a$iB2W><l<jOMdH8$AC}9m!<M3@&_UYo|U=hTjbcYRS!dEgllC&ol`OjWg`eM1*
zsE9W)&8r=)p<PHqA|Cxju7G4$(JpT!4ARoLWzp5}pngh{>y-Zd?7_lG`%Da0sO1CC
z{cFyHq#s;<tz9z1*HP&<!DD|Czo6x1A6}<~MFxF)d)$)S9m?<hQ$MG>gwOM7Mc+ZP
zDZ?Lc;&=YgLW*s_$6s4R&KBMvBMW{!gu1ovwzj@qtAA+GQ5aW~ZyQJ6(4V3g9v|ZC
z^*&m(kZY~y`vX`W%C?4HMEcg`$a=@tNRJp9Smif_xIx+|Q}7DnT`i&!sqX$bO50ER
zK7Eru<_~_E>h1{}>`yPuAU2l|k7ZRh$`Tz)=wqfaP!X-@i1NFk3t=7oN8sk~uiz<t
zEM+n~EOu(LI0B8<W+7tmFEu^Yo0~KbG|L=D7ylS4Y<0EL`Bl8^QE+7r49p5(eNzCt
z_qoLU3oGL(5raL#^Y#teLQZVic{W+pudkCnwjt9}_vL-o=Zz<5P<*T%YwU>;E?UNZ
zafm~Xbh<_M;Pi(i;dC!D8_OWdh0XrnW_}&zu~(Wn@%kodknH3EzMV_c1F~#F{q?Vm
zT1r1DxGUto1{a`kpxM4_V7IQLUH`1ued14NQ%7i1WG|AU*B!;u<^3AZRjjMCGOGIQ
zxlWZ*4$+l*X^)+u+(*5X7;pTT)IS4UyDA^`lv*6MQnypt^PU5-3f^L$vrpE}@9lK7
zC}tc2_m41N`8`7_v=kceo<$!H3fFOjtL7H>-$g3<***nwQsEa2enycUuOQAT5`ptD
zo_;Fz#r%-(45$`!YqqhDYm1=6mu~bJI}X1!@7ee=+<SJ*_Q{jDU9@Pvhu6H?9&P(C
zDt;Obt(s9su{8xj)8M#bFHDz2GGG{|2P<Frf6%qY6OVrG6ZXwG-1=1Q2U3nf*Q)K;
z9}faP3;d*g;?nQH^_$q{*+3z|-iR?%FB2R2hMsS*wii2|E^g9!;|H&TUs~@-6e-(M
z(ubc%^B*^@<4jkyNyM~DaQQ{A|6vjFeEO3ZgVMk`bT0iTF_YgQPF)Bbf_KqxxM7=s
zH}hpBF~{nHYYmvKRrHocc+LG214Fw{1k9m$0c++0sFF79WO-kk7_F=0&|&bjtDlXN
zs}f0*)HDdeb0uuh7^L9A?P<Li*{&mBKwCFKTnWEyf^v`jM1LmiKF+NoHI7R>w46!<
z3hk5CfJ0WeUY*XT)(&apB)Cz^90}ZnD`<$U4i?x?0{7fEMX;YN7UC;yYENr#QI#E|
zQruK)^l93|C*o5-ZAod1i3!JkHPKKAq1cJ=^V9jIG09qeEP5n}-bUOQye<Me2A|Q<
zt?0<^!9XPWeGpTtzEi;gzP3N1hA}f3(K^coeWTPe^)q5(<Kr9CBgaoY;L$kGngl5E
zn9WGPFcM-jZdKgmjr`zO9h7e!WyCq@7G{>od4&1p@@Wn1Q|9OPp|424`6`@jK52Z-
z9qVo)<t%bcNXi#R@`hR*SA}WbP5ixESw;W{vq$uZg|Sd@nJ7*tqx^aopUW>2^8?ue
zb}Q56o44cBH87@6uzuE~Pr{3mJ(NxCFH0YQZBGeU>9g0KF;h3zG_@0)T-<njI<<|k
zG8`@tt1q8|l($P{f#GrEhA%jG^F3MG+8hH`uVjUy`0;jDh3e9PAiqt4WKmwTJ=^dT
z-t+>@#$2`fk+>5fdY?goU)R`(9<c*87&Z?wkgimvcf_Afi}p!*1~DB!c7MRSl?~mT
z?iXD`Z98%;_Fz2AubX1pdzS0tvuYZ<)AwX?GDqbrfpV!wd*?>;+Ql#i3I|U!yk~ct
zl%O}26Xo~ft9t(v^xW^mZ&#n`#vi&=%?b~M&%FG?${wD~sOwRJ|EyCeUGg_ezQ|AF
zS1v*VV@{7p)2F|V3QoOjPf$*Drw){hx1BcE9KNttzr;;viVY1HiqDT;H;XWn-a+sD
zBZv5s>_K|4kTMSaxbjQNseB3;gR66AN_y|s5p}HU!$!&Vt@BpRrp;Me-qS9YN8z>w
zXz@OVp{lNw&CNaf>1lc&o{i8KyS~bY@LUjRo*`{-A%(E$Kf1{FFZb6}+1d)LsXOw}
zhFovYdZQ&d#5ZLtHgxScEYWq@$Zt{C#szJkx5afvAdHrCBmUkmb<-#}j4+knC0gix
zxA;=P%Nqj*!FWXzYu{&EpKJn|7AZdXrwswNw%>cZ;p9(YG`zgcQA&rItM$qX5}fd%
zTILk7UnZ14l}`MQhY9xAloxZ$CF&yUX)>33XIUv`w>2w=JKk8921~;BsLJ&@^M^SM
z_Ze4cd9l(U*Xz|Mxg#&@OrVg!{eV837m!)-%s{a*Y;La|E))x_s`eq97ZLEWjGWnu
z&og^?Tr0mf50YKI2`qd1LmR(bZdy=AKTZmLm3olO-}-W;ZDGyBL&aoa99cP<7e?Cr
zTS0su;Z>)k$5tr?9=UeuxPGM@ToY{hlJzy4O|ja^d@rk1HD9}pnpvUochUyjcvg=~
z%DRe*u4aggQpNLLehZx#6TiA4)xMFJJ=`gQ8Yd2=blB3N`>T%Xus&JEk5!3CsJpD|
z%@X)4Ns_+O65C)|pohv6tN9e3e%`5*QyP>Ai9pZ~T<thsm}#6uBtiR8+rE>1%qem?
zu6D@tE<Lua=xh4I*{PB3v6RuV^zMHoP3-^I-duX-|9#Rl@<JJ$^qkJr^9Le6D|{`u
z4!TV`e$uX$G83{8BR?EbFe&9w6E(xan1MjaT|=omI+v=HF0^GA?VwD|pW&I~0*~90
zg@u2ddrLX{aJHTf77)Bg^pSq)zc<XrXe#u2qA=xsPjmf=#?|%yu65h|&nQTMIXAcK
z(Mne>f#R=)1=jrBtJ<Mv&AsJMJ4<`lQ_CNSu5Imi=TDCgVh{E@l1a46m61d<toaxg
zH@grG1WF5z&R5&@;dOpT{8Thh@YlD9IE{mbSNlU^lz+%dFT@io|JL%@*Nv#ak}3M}
zg`rb+!$tb599Jxy#UYPk+xaI#Ngkm!6Fz6eZja~8zqG{W-GhEI%zqerK~CInPxvOh
zo4QF<A}Cr>27hD0!Pcuf+d+EXlpS@C0;*MqgVj|*y-U_x6ljkrOw^1^C+=jgrxY5s
z^@@VYEJ*SwSExTDK~5N{?OB`@oIxfN%&M{ba-0K>06m<Sgt|1<@8a*ydM}P-3Uh7E
za*1Ljigcg56UJrIKU5o}mhya&7-E8JWR6i}wG%m%FsxMk7MZFof$FIcg(eO(wfk9~
z*ozaHL$CR0iKe9SN6x|uh|Q74-tY7<d<;X_JGmq0k)3F`H>(CY)kINO{H+V}Qx(50
z;oISEm#9fLmy~?C{P=zN7>ni$r>A}QyG&IM&kzAszs_@UqmAYf&JPZGCMZNU;><6G
zmIRtd9{4{%fl-$6b_iYS1fMP(#lT~<ew?N(`u6x`onm313NGJKkXiz~+Zs$VANFuR
zJ8PGDPx3?!etSw8qenu|6MaH7uJ*xKQJx#rQ%BO{Mplt67PhHR<z4t0-g_@-a6U?8
z(r_2s;+(SSG5i?naOmnAsXUW}gjtE^NiY3}p~b*hw2jX!YsA#^LFkaE&`UY}tpl&3
zGFl#g@EJebK}r^6--pv_FNq(2c?}njIf)v>VLJ5gaH@=ndigmDt!(MrfuL@d7o=rv
zzaDx~_FYy@)+MtLSb7P+`1lq%<*0wyJ{ose6&glmWce+MSeyy}jR3xtS_tx>j;DDw
zX{_%1b(J_`7(Ea4ij|ip^f|e&<1UEx6kYLtqB;hUZ_#XQ<Zg8RGTG=#*c08hWv(+*
zsurjcfA^uOK^6=4jrv#ZhPdd!KVvL71WahyK@Lp@FJ1BrR;!*{+X*6vKHUxZU5=--
zGb@nvNA79oLxELyFUavP@Tq;O3Rrg1Dt6{Pxnqp)?_(#j#(Rn1?eUF78=<?R%<IXu
zKhVEPK#*Bqb7zr&E0ghnefld%%<yo``SY~((eatLlF!0Gx{TbcNyf8&QFg6Q9ij|b
z0j3C>tB&v|1O?vui-{KO=Q?sPHzor3h-DKHc!T5-Gp+U&P_tv6G{ZvJ%Lo-rQQ|q2
zy0}D;-Wtl<up$W6R&|wq7KSH<2sY>o<dF5!O-|c&J@oQv`ld+wf}Ev2W%uA&aRRrc
zg7)h!q*W1$K}NN=dbIo23M+8Csh>12_g?I8#e8&5{Bg+kiDMdFb?>taqC72EW&QS{
zWNuITN!qC0o5iS=i)>uyN43{1sFbH<+3_wRq;Bt^7z5N#M!7~yzQhk1hL`vv`D(xO
zx2rJ2mr*eru<DN-f6u)F8L=WSZ6h?AX7)ra%-Q|seufd!BxJ+Wj7QgX@v%>`SEgx$
zvL?>NWL*PZ<FlPo)j}ktEw4#8ZkkY8R98~Tg+RSe$3>GoT=Z-`)>5F<$7BVe3(qYj
zK1EqZ&8~-?ar#8PoH(j;Asb24uXS1+Tg!~j%<^TKIv#7w*dLtyzNo6Qgg2{u18?1s
zJNu~I+w<^=RjmZB26^X>i9N3k=UzEf(%QUXL*oNv>Se-{Ka2L!wj*MbN1U@h-tHrb
z=n@@hdN<|VxtAgm`T5lhmcq+vnl}d8<#@YCM%_w`Hbo|iks4kWb#hPF>SNX)oDVUG
zbY=W#n#xAxLjCj@Y7p?b<lL@p&08~QrSVAR!1)`x1JT9hHA5nLIR_s5Zd1w3P5RR<
zGs7fRS}drPk?s3EuDWkxVI(ik@V%9eb`+>|<41-(%(yFfYBSaC@|=>d6=SukBvI1F
zS-Gv8sA|e?V1JJ-fBsL2hvh$1>-5Zw|Jxt0PF4J$XCLh<7?FAI5udJ#qn7EAn}wA4
zj;tlFd883Pp%J8ftxJA&G}GeAkUGAeQu}!B;l=D#!TbQ_{6Ka`i?gL|Y(~Z>LCPV6
zx*s2L*Iw&!4GcIxzaq?j6?;v8v$v_Tt8Gn#DNc%}dny)@-{%&o>GG>LWw{W|gk$JU
zAOtddeZK4*MY?ynykb}EcQBP5VZF3}+0)bFb0tFz8uu>OmS?Cwvi@o0>-{cnL_m-~
zL+QA^+U&Me1-`IM0xn}I_*<y9+BXru7~hm{Sv_!Tlq2T6fve?=1D5;O^^>PJT+S>W
zo)+w<3Cf5(uLT<|V}%cpQ`Vi=#1S4y4OT=i@8ubq6emH-%ec3~O)IpcY|B-KW*_RT
znM{2ZbKxuqGaHX=D2w?{?Hxb<v3C~1NYkwBsG!`<XC|STmQz>7!$ukg!7N4Qr@lD(
zlS&xGXz39CKyK6%n?~zIREJr#W~<;2NMuEeK}(Ou!>FvYJk*yL7zL<?3XP>Eh?!nf
zNz`{d4yH87B5B4}KkE;<4pci}{wlm!$eybDq4%daccv;L&XeAcC^jC{(o3Ht4Hk$b
zr^wNJPJ<l;W$@F*(qcO1MYzXl31L56Tk9PtN;RfT<DF+r_Smj;dzyvycryLSdO0Lc
ziRO&JPVa{e_tJh0!aw+_JXaN=X#7DIU;2(8>4UBTgI(gjf<9h}7kvzgmtF&-M5I5h
zZb%D%KRny~4Ex7<;vrTIbmIj^x2N}nNXuHD#LK~-hL1~6b|>>}U#|62Jetowb>)!b
z|6Oidob4htFGC0ot3}0|&gz_;wYeyM2>ak_)$tsWK;yTKsz01GjK?%={Y#bV*RrBw
zL$=XH$8S8)`q~tzdIes#uS-<FWlkgZCCYjKcz|-7fP$iy5s3q7@N%uJc$Sjb$>~$7
zibru{H@3iyiu6n#AC~=#-=uxM7r1IE?MiWay5)BGgm1%M?+J=|W788S8!M_;uB}y1
zn@OyBk$hG8whT`rrgvFTCjY7EkL4qU2S%QchuB^rzw{96VJNYd!+AA>0~h%vYE1ln
z3aJk>ve(V0^{5ZdB1aZss2dZ^ua^pkVekb|mVUi`MEnQurrf2PxHZ>tGG9f7s49Gu
zUAg#`gmRpZt=kxn^kXV(XAio74<BnbxSi+UiFW0FncKvA_A8R4Gs*goi5~9F{4=Cc
z1Wy6g$U5?ef)=k$Gsw6OAGLdP&<J_+zh4j95_6BTgsQ1<W8v$*K>g0PD30t<wvFU+
zQTL+wDUJsRG^*2eQSIDpc)bvBF$3@TM_Hb}PB=*gx;wmVdLmEm6f&}3<&(irla?26
z&lkPs&f2go`kin*?cBoxY1)O~aneQ^z=s6=-sYkHrMmiiCrzd89X0p<jP$!SUiJ6C
zqI8+^+#kMRQqDrMw3DR&S$M`Jj{2g&`<odX=}vd8sU5eBV+@ypLgUw*LgL^$lnnap
zh7rLz+L@|HFoh#9f_9(PXGxqJtiFCCMuT+7Xn*gnADiu0ugCLzoGLeE6#6;~<J(9J
zsj11g???n?1*8dzZojgsCr!>5s3Ux2ydOllco|@{=D4?U;3TawV?8WLO6<XXNxf;R
zL?2OtglM53@O?<mB^EB&R+NeqmSqYnh}NCXIZwQrKJNm}>_m$AEuS#UQ#(_?Jz8Qi
z*KY#3FwaNwYz5lhv?El*`+nfxwe^-%n`4S62x}(8KkVH`81(LczyG#;{>)`Ud1&FA
z=Z%^!Zae~y7^yJ-`>U55A(AjTuNer0KGtL=m~0vOt6&U&Dm7DmiaRLG?(Osb-OZ-*
zLzt#s-obWULd+ET!=G^UO3E6?pYZ!x(oD|f4iY~wdb^Rd!n-->x@Xd$$Gt0t6B>_%
zc`|(LcHKo4v_wsXLQ+TsMOks^fD=#pYJdQ#oHwj_dlK{p@7kn)x^pN!&BWKLngipc
zgq-pWOG@A&BSXzkokpHfH94*2z14~B4igiSnmO?ugE7Ue-HJJV(VN;`T0Ok{(CC6<
z<d|95uTIm%ogq8y;`ui)u;Kdw|1qch{b<ztFSb?!-)ya?V~t2BXQOR>_f^-Trds-X
zh;)jEy5MWAnHX-r?pj38P~X6sn1Pv{8Ig|P(Ar8`&+<LQ)Er`_XJ$>zhDi4wVgj+0
zGuP4qUzIJOXKSdVCo9Yk#+1;qG6rvBx&45n>|fuN&CYQDoRI(Lr-W?IoW%#~UA*0O
zYx+gK6u@7`Lh3XaEBi+5XB`cOjG3W~C>sB3RSeT`G{Lg39lc7gF6wr4h`OIDuekE`
zEgbD1{faaW?_&<%<I&t9+EzU2wul*Xh|MDkunO2X-*J<=#)kRtpZ{9mzZUqf1^#P+
z|61U`7Wn_01&GIDxFRbo7CPn915^<I+C6dK3kF^D)DM^kD}C{Fnswvj<3=M{KX4ca
zi~hO_W~}u0@2}@a8_Qx4Ce!ml-kaAdul{;{j$<T8Im8(!Fp&4UAML?kuZ$&&hO{mt
z;nUO8cZJiGNdNUxgH5`_f#SX%3(wUdI?|)RUQx57l+XHQ2o*>c>Qkcl>xH>0xguR(
z@e4Aek@Iq|hkw12$9c3d)>8tR-zryC{r9&F)jF+!x42$AWc;7rA}WfoUA58cp62Uz
zQ@vbi)q(Z5X(|X?+WIIyW=K~_pbPq8&D1!QlK$J2ns}^vIm!o#T~TDXOc~{NI|_cV
ze=SLEr#DVQrNaD}75dZ!dVTfsJv^OyRYTNYm;ArvM>P^zSro`bk?lYdJC(v^smsNA
zvM_pYyi{c{_UodD8)J0PLY?=zdM2p%3zm%K;IQ4yz}7IpC*ma;B9fHkVRN#b-u(l$
z+ZfFh^!qHuB~xuQ(%Hv7wtIB~JzLLqb-ScD>k2PT?51Qs`W--k_o~tx<tk05?U=6i
zkT;O$T;vP<Q53WBV4nJvrrVX_#qp7%#lZ0nbnk3shRC@<MQwUVak>A%hX|hgSSgpT
ze$%w=_Fgm~(}MBlTMjs^4HV0_<uIvu%r|n&m&F*(s@Bpv!>DTaJG;1@Y@^vMy6aq~
z@)~~6FX-2u8--xeM3IsbRQyCLvRs<yad9j&99oamUl+tm5eZhMRsY`Pg(RP0uDN(~
zwN);IfVwRkQcq~SA2Cs4I8vYm6<nUEI@$;hbKdWm3=Z?%tOYyx)Zose(|U<N@x3iz
z>mcXeQm<Gb8bgWitEV9KTTOptE{=Ef4@30($gYm3GT@QdF)5^pH9s$`rEiGGa?mp$
z2?R)u&bplB4yKCfMXkm;ZEug|kNs(O%3Dmz)wnvrgOpTMplnT+_Ka9N9h<C;7rG22
z3DTuxt6#`mVBOAWHg4-nDmKpvtdXZS;XFZv$|e1$_GC7W$<Sm~0{P<Dl!$4y>pLb<
z#V_OGj6apNxGR0>)GAEmMYt!^R(55hiSNER7VEcVc4*QupNUCCAGdYU)k`M{?Bz_a
zdOkv8F+G<Bw<5UZ@bGSf?sz>D#i9T^)~C-F0C1?;e(NtaPPcZ`YxypCGGhm^?VH=K
zufMc|y11+l59IWj9&c8F`>e|<)9CrA+HS}B{IF1s+x=oAzf!%*I!89u5(}H=1Ms+=
zv3$j?$-w2l_?toNJ}$XFLU=HvJ6`{E<q)BL0)O4x)*e=Dvzhaw`kOeSrBH?Z>uSaP
zK0?>i#F96yfj7)2;o6O!%8>f2&zZ7z0>zJ+jXuV6ImAR=mMVwTcSka|3H#&5mGQ*4
zShNw40)v&i-Jk?MzuQ&&6bk-0yUqI7Jax{Id`;FFx-!idnywR^TRW0*T-(dhNGqqC
zYaQgqMBGt9VWP*YdxlItYGIV}YPIJ(&#-Ce>9v---_$yt*#GfrZSjM~$;^UvP=;E+
zdE<jk8*_CucBSdGHB!A(N0n^d6J4*(dy5+l$V+pL>NF}$VyZOMig}wzbWGQWLsCM4
zgoVIIO{Yd_H0#X4xXnIpL}zD3dR@9I3R|NzPWfJ<k~|nhyvYz~ABn@4DXcWf*cay<
zDQR)vzPcO^2tV9!ZEgLWT`&zG8lZT8dKMW`cxY%dEW&Y5=jUE)0b3L2OYuJR`>ivg
zp`(Ksd5XhOp<Z1EK$S)=%^FJz*&8@dku>l1X|o=bv%N7xY0pEwn&pxBX?Ns8e*Lw|
zVLD;?i~_ZO8&kRz(Nu|{JaCgQ2&kuXU+nOh9mzI`Y>60`S2oi)?Pwj^0+4}^ArqYU
zv=O<T&inNNXMN+hF<rK$YRh)JjA=G7U7k{*cErdNBuvV~7VP*~b}Oz^J8bG-N8tLl
z-@kP~LN<ZV&B-3bu)AIFHYb1(Ot4tcr}|jIyRkIiPUl^XA{IfLA(QMlDG=OZI8k)A
zIzR%#(Q0$Nu3ah)q(jrGs)`Ee%|U!PfcGBjtW>cGT;A)kFe>bPr9y+VeXn2vvGCt?
zDopy<m&ymr5`k*%KXBw;4UkC@vT8N0((4EX;Io^I7eG)6CUeQ+7owc81hHz=DvQpV
z@PbRXfCC$?KS8~VOPF@p$r8h`Vv;XVS3DcT7}qODFPj%KRo0ncwsg)Zbn*bM3zIJe
zXqa@a7gaW0#nsyC)wZ1$tzd!0(Po1wh3t}Jsba#M__yXUdv<l<7F;Hx1Q1WPJg>_!
z2$SkMOSxF6>Zp~j?v#2C8v*&8p)IVt{yGY8^i@g<z8T0rB{M{V->3H{^wod<M1i~9
z?g;S({ez+&4<GG&ao3$^#YZ^@81F|;);LH2Fy3G4wZA@FVc257&w2ai(BGkN7U%>-
zB)A6vqM1v*GOIlry}GbdV;KQBK9YB3P~rO_oVpOO460Gq;4eFl!b7YB=~7bvEE*1=
zV@w(UcuI#K`(vAJHOp#S(;=aY^Ns7uVb<UQqAmTg*;>_Xh1?hx*R%Z-v&j-lzjv>A
zan8=3W%u7E;BrXy-4Y*2#<2aA$?m#3iE!J<z4-xdLhFw=fL=@&>1{#X`r_YeS?l5n
zE95&KWhuv7m7B$KY2BI)X3l#n`WL;<hsCkJIEfDzUfY1}RN2HEO&#Zg6mY%Y71-?U
zGL|XlLnfh-^D$F<bfNxuGCQL8pur<{BuifO;(GzWE+)uq33$MD{q^}!!aL2nn`=&F
zyiq#UGVxLLus3oXq)WSDYtc$|&VRx-qfd{g@;A+I=Wah(?u`vUmTCj*4EgmdXyF&Y
zyBg{_kWH3kp})P6sgg9mmKsN-s^a7+<aP#=I2B`pM>G-0XMIG;mYYkD_s-}LmJ8`?
z?~at)t|?ZVgb%t+ItR<lk=N%xN%wM9N~zT<q|?v?sCSC=kht}4Cxw}-DuX~l6h{c>
zsFpjAWRZf9_{Gg>cDlQ>1DKCEZ>;&guJ%G}pT;?n@yPi}(918VM6)3>Rqc?G>=Wzo
z`rfw-O^~hW%DuXa>w0Ww`JB^+@xo+44MbVR7V86vZj47YUod~-i_gg)&r^?nh3=O#
z2!bR{DxS+`{kKfAx7+y|xib5U&c;0`*b|8#mME*$@{PpFN_<G`kF!dESwnG3sRSfK
zPfMF`i6b$D)7C^2q_v@Ngq7Y{EQ#@h6*;5P91Ft+5ODC9&X)U)0C-HHb*_L<7?07a
zR#`V5u7&s@U(n8=@~N?unW|zhy?XuHq<SIXry&XU8`;+k@%V6D&h}D~45`VGDEsp@
z4x@dU(M86=B*EgYCZ6+E62tnaV8cLhg}Th=m6vYUqK74SB-WkFVN?MGsk@Hn+7P*>
zW7M;=KXY=$VgRJcXBcz3gqAwPdSlq@T(9;AQ*Tb3v-rI(0A2?Vtaa<9iQ{P2oNh&u
zeHUL|>@W!AI^LdfoHA|kMBf@Oe3^rH*vIP&auTW}tJyT2EyL~IS~R_jYnIy(84$i+
ziXfu%XV>dI#fM|JHn8(TJb9$ZfR0us$k;cO9>~4l-n=Gsz1K!$iElewvj!OU0F%VD
z^^im~ORj84Uuu-|#%R9qP0Px+cG7aw3xzbR?ZMRKbmZHaU=~GtqFIsINGN2IPFrrS
z*0UvkhM<0?I!%+v_Uv<~{V|%WSzm(lk|DKNKcpxObPg>kv!kAaSRKe$3L&F|*YDWQ
z&htFlx^OVB%9JA?Z@nEf$mbTNXXyU89?93Qeu?L>DN!oCq3P?5JkY0=XMp922*PwZ
zKNM#*(bapkfiAIVyEO^lcL_+HL_7~W3GhwnkiOa|XRrZU<kM*@<YeE`<hJxi^ne@<
z0Hu9*;0M$0#9;%zKw9y0slmYJbmb!cZ)lO8X~$2}qH8k{Ix3&{!W1GVY8;MIA#ES*
zLa7v4&M`Lzl8w#77ursI?7!Tm#~8EYO}42Qw~O8{7)TaQQN>d$_@qFww=wFnuH6kN
zloggPK;6c$?#(h7wb5yrtYLyR@2mZ;Ss9IM08=_TI#w56Tt@rs2cxPtl}vLr4nxVl
zw#O#2JyFb_D>d3JnsMy6!@~rfJjEYp9kL1$@{jA7sO)$Ib>pS*Ltp~r_-&LTyF?rp
z?GT2HD-zE<-~%X<7Jv{vgY<)lZd?N-RR`~u8t@U0W(+VIB7Ffjxue5}-`8yo0edu8
zHLjQ#3A-01PdARkhN+G55;y|`Vm0u_RmBA5VzLK}Ao_r>D`u0qt(}l%Ddyu|oeSIA
z1IT|!HeIaG`Y($L{dj@voFS7Tok+mRA`mivxTa$_W22vzP7Y?oW6AkOCS$bj3;6US
z{YZAJ51~}Y8~MDCGzBYI4HjC%G+ghhH`bzy9lH3kxAsF4tbzyFI}z*98$UlSD>dZV
z0xdyvQpPranBcA)z#Ob~W^GjqrLnZm4?7+(2KxJRTg-opx|GlCC^x%0Z;|G`;bIC^
z?Jrpqdw`xNw_@?HbuyecU9=3OinXm4KP9?$Nz-S#;~u`A(YUX47TdLC(tdH+@bg^)
z`tIfQ<#=lf%UMu{(Bv+d$BL|0enLQ5U~T*b0cG<kX-j)>Gi@uEy}4qq=O&2B2n{xL
z>=P#;t&I^Uj!;aI4;;@TwHhv2O;u9K+~ke+LDW}|lN^A!O`P_Ocg00}`v(zo`(*qc
zDr)pM$7LFQaI+O}2wcJe(Gu@707Ob4eibkzFGNhl_QCoaY_LFXPFiBuLhY8O+IPZ%
zpNp)g%Ja+p@1QS#veHUNsI;*oq|NK+{9r!K;%PJD#mUZN3LPwJrMGQEP8VCFI9{f+
zHHJzXYPl+5?q_|G2kCc)%U3P00sMD;*r7%)9Yt%#^Xu1ya!f!cs@qWR{BV)G{^nY>
z;laDL!PKS?A8Q>)+y%}KR-{DP@joO96FRUF%N*)cAz^hRe$xcb{%1A;kP^6f)2*rU
z9+o@qT|=2bJW*khEG-+uZgmNABK9Ji$S&R|v&yAqeBnIRlguAYWdLZDHb5dia6b#g
zIVg_09Leff@kSubcvirWnMuy&aJLxZz(@u?=LX-_;y#wgR0A#fhUuKQSxw0o&?L;D
zRZunhGF%U)B0Zdkx}2zk#ejSF<$^2#f5P}aA~Z@r_YAu4&y_=6ZT!0rPWP@J98k-q
zhy;*8t@>tgya33$C>=lm?*T@NBS<H6D>d4$e3^hl&Hxzhzm{=!&?ovJ9Trk|CZy!+
zc)W=fv<iX<iApArFdqK(d7qn#5RMzjn-BH?v7|e6!azqy529y33o?CryAg+x#xK<m
zJ8$Dsr!AlgYdJE;nc`Di`K+UHZ&Mt1bSSUeH86_N2t&>pf?`2AL^J<>A%B)4G#~#u
z=#9nW{+C6Vjg1=C_yApL`fzyA_Z}9F8B|+JzJLnT5)WkOvjB^<`cuKq-?HUHr;iop
ziS9qrB?(IZaUKpgo+VJ4Abfm`gh8>}jQR!04^>qMTo&^p4X^z$sltm3rpj`W9#0i$
zT@@_z+V8smC|c}`=l#7TA5xE1bGD54z-qSIE}Q(m%3q`hloyqa;Pm?9ofc^nhfPzM
zgYW2K$HGIj+RJ?;nO0~O^J>g?BSN;9YvA=!zS|>RE;pW*dspGwJX?wHikPUlx!(T*
z@8U4m(3{QN5kih`l?-{s<NBFze0?O_oB3w8)=4*X2^6{tA@$#x&!+nUvvg-C;;QJ<
zLAZ02{a;>4`Uo*7KLFEE^1VJcHsebIyoWePz-mPjN#w8z$v^o=z12#q`zf>W=;JHG
z82~KqKaSttzI)LW)GP1na{MpP4#qMM1p4A;Ktx_2NNE$UMk^J6Hsl*m67=&v(vEcJ
zO}PP;%E)5Awge9~@KZFjGLRr$R{J4w_DQx4F@b~<yZ;0_*XNF(r|v@@Ankn(GuWEc
zd*EATw}Y>cq}PQ7|JrD}LMC+2Zf6$%+07pxBE%~~5Q9!aYueh{n_9PvYM%oB=gIpa
zdYeM2;8QF|pxd!>M5R@FELY?PHwK`>7K`n(b6sjhuX#ex%7N!3V}Np4@y<R5-_Ov=
zer!*c$_=H_$>(bvg{zwN&ez9+YL1Zec5s-WHn@0$?kHyF908C;5k-cJ^~<S3Koz8a
zC~<eeSOl>7qMtSfh4dWo^o_~VA&|PHVmTZMjwA#y0L&0u0vF_ubo0j@8S*DmW@SFh
zRt@Q^dhcU<bGDK~$X&&AqSpMARBZ7S+O;;+#$|Ok?wt2}^e<jF1ts5G0v42Ew?n8!
zI$7u9!k6TYys;5|cTpZ4xfB@?IMmOYUU(s48}%p5Hh8$Of}%|QP~v001~aiGm^C^6
zav=E$2&uK!o1?M#;l+G22_7Z>rZSMIZ{2jM3mSdo_NcJ1@WBn_fANBf#B;LiYbRGF
zU(YGz^OKwT&ewyici5~JIGV5NdaDZ|*-X2+o%P1@YJf`JxDqmd&{vJ1*#auA<e1Cf
zzB+^cYQ@X%^?UJwy{nkdN_u0(;#eV$K+(zmm>JWOt5IeW+3JsXw7RzjBH?&*Vi4HZ
zJseRA?{0q;1LtH4d1}BHxSh)2UgNSAK$LueE3Msce?RW@6z8_qj6g~j@?R$rM)6x_
zzNj2yEi@d;n(N>_5U-`cE>p;LOGj2sX{Q0Dom@!$I?!QgG`M!Q%AwRUN%m=~9B$W_
zsJxf`<nhi-rjtZIC|h;Zl6;K4Kn^+1xX(p+xe#ptT8hg*J_A)si<HCslRn`_^kaM9
zn7SvoE3HV}oUJ|cB{y2>irqULvU#<aUs~t;4S^^Ji|ZhgF@2Hkw8YRY1<wlPRHOYd
zeu-;O;O+|G4h?Z%Ho#&6xefd5Pf7~xgMo+VWZ$u9*4gi+&vUZ8E5<0`D=UR{zWiQ@
z|LmNW&5cJ72u`4y>GdD=b-UWo1fo&Z=sDSP_Zg%8WblZHD0@tClMSHgL+~cEmzd0F
z#wlcOAQV>O-DvG$0gWpIRi?6Lt$Up>8tHeZ_B(^S*FDi^Qmli#B98rMWAv#&_wc)@
zH5`igj`0S~Y(oGrLvW)kC4hUEm`rflPnH^8U91j#HEH1jW@-73nu2`Pibo?E&yz;x
zMl-w!Kw2}g$>qTwkyFknhz*h({31-21ju{<=Lg4{yJ`7mi*-do`fA=^LTdg{atl8K
zSLa7n4Ejt!F{!OPCj+BlIpZ3j-U&PYo0}6U0LM;N2aYn+<DQ<{nwAOmM4{fYWz``O
zpx#{K_)*Xu?Ec{f8U$jT?MPPG$qqEQMG%k6p6|96f?pd-2YH@ZHUwG@QEiZ*)4ZPb
z1)ssAF1p_~OvONVJTKLp0MKHVy~7MZY4pI?@#uWBGiG2o^WgW<2y1YR^V?I3-RVl*
zd)|AxbK`{El_Y2dWc?A4)u+s;up~BqS6I~RQrT#0^YfJe5+&$n1L(=^Y;%Ufc~wp)
zy};d2zA@TumjE)D%GvP*T`0RllZ5$<Q9ob|NC13qu0i5uk1PIP>jLDz!B@!n@<#!h
z>3G2_Vwsce*!*ss9-Y9E&XMM{{tGe`qFHoj6#MmInfs+xhy$_-idIx0LOC3y*-gtZ
zRJ)){aUbcH4HB>wFcPbnn3zytV|MJ%eHmehb^g=jbxRv6kCltl%%;j{UusQ(Dj7%`
znxFz2%TXqo(RA3Z^aeauI;7s#XdhJG-4vhq=F5rLi&&%B{{bhUWdfgvi}?pWjYp}G
zMTRR6+5n8zy9wtVGv{br+0C777=Ly80%H16JWN^&e%W2}b^inkQM|VaeG7rSSm@{>
zEu+3Y(PO&Y*Kypcp2=P|FM%d@u^jC|Di+?3J(!_f+zR{(B%91A=hvUX>KVSXTP_w*
ztmHuM#TGLrxwIdsYy-fNxpM@l6sQ4^gUncf2DINd#4<<g^2tAL#x$+0n6LG2;nq1L
zBGkI!F7?6`89?hyf#hAi6duiJO7Q}yZqKM90i>bsQGy!J<=Wu{$jeI=>aWPi(rmYc
zNydIfIf4LJ$?hnJ+_>^-HUmZXoa{clOd^4n0%%%_s1ERR;PZS*s66(pAi1#SKN0zu
zDceG^0N7PT_8uMjCCyR0h`4ycy+uqWYMrcrj^t?^=yKs2cjSJXjsk4we+T#H&tIN}
z0wmn28A`L?n<^fsl}X_909gVR?|91tmJ=AJC(2oN&;#Jdl2Mz_zg!6%OEJs|=m4Ir
zMXCbIHNGUk{upxvfbo$;tN~FJ!r?tN{WQ>4-`yGJ_a*QQg<#V3jxNGX*erMN0j&oy
zgZq}Ejwa_S7c*7-bhtddo~{lDMNjA(x%0vL5x1>UMW}tUAXx3JLwb)7Q5~!KCn(hI
z#&oH(1(aSO8SO<mn;Umz$&3AyQDG*S!Oe4r{if-9M*j~l83xA#V2yRoT6{^>+f|I6
zr+^*;VhscVz-57!fDq5JJykyQ4_h+X7`5q&p!+cn3fYR_ke?`_wCa?}lM8_KbvR~V
zQelC@|5ld~+h8;t9M^97t}Z?dLKdWb0L~DJI{EBykXJw%<;Eijg}b|TLF%_Uv3b(o
zkktF84@^3(q3!}Ssw}8f08XqPqJ@D8I!}S>g8YpL4-an}`L?!iO!o(xT+uzBL^U2S
zxc1VXlau1caooKnY9F0U#_~)@7pOrt>a2gT9PfrURpzoWNKNE=szKqV2&gTl@&0YJ
zCL7qzD$|Dk6<H@xR7AFc+p|U`-`fH_2m}F{7Ac~u@2F`xX4goaZ#$c%F60~Q9sQQX
zKBro~-G;mIhgwG|O$vA(pmro8KtUi1m9|}9)weswoE_)`v)nSR?QT{xHRV#HB{8|f
z^`pqB%Oi~tURP%~zLLpOqLF(J0JGQt)H?W>cg_br4FzZtQA{kvGC;}j4x^ymK5-WS
z(kPnVCBRfTq8l33vbQ<LXA6NAXAF27qcs1NXQ<h7D|<*S4O#M3D?*pA1iz=KI^lCW
z3XFc~1{h~fY6b4%%U*}>U-6K-%H7mG;E*)TXI#=DP#~%Dm5|5h5L|ujixqP{o`C-+
zz~;R7mm+Fmdx_yuHt`gaEEZg!h#d0X46)@1G`b)^Ld5vSaJ@NU(b(ozf4Q$ifo%zh
zG@xF(ldPbC)GVXFwb7zhAh*H>KYhW+OioJ@S8$d)l*@gqo^HHLlei$M_*n@QS?SN<
zz-SP-bHs5n{7NrrcE^fG6eyF6VUxnBm0xvE#PP=8%3<yp|E!EyYJH|!7?g|9x_F<k
z)iF&sfI&cnP$ZMl6~J*@8+_Eb2V{QCEBV`Ea9(WDvke@>ndscEsJ+=})OnSk{w*09
z8EX|R$Q03KyGo#5HF<Pf_5n*Bt=A{VuY=m@mR(QI<=0ILNbo?07a3<n0N3i;`zR+T
zbxZXc_m1vqD_`7jvhbKYtY;tZ4xvKQa}sa(*IAnNW}rMmz2P|>UH}=E0(IGQ0h>x;
z6C`(v{P|eB$;Gb8Z-Mbz|7a*c8w&F~Y5*du7hyf0$3sJ;K%ho-D+J92q6oAC(KI>U
z2-H)9$e;>BWj?Es2?-30k95jMbKI#dXQZk8pFF{pX*?$R00C~M+75+kpg$1`_c{xB
zMgf}3#g<}YIw9b=LXod##`4uUc))py!Mxqt;GTBIt;0AOIar9t^Gy5`_c=f>q#cDt
zKqY);wo9WrnwHA}&T6)&E9q4&T7Tf+s4=UT7)Tx#(|`nQ{pD7SHrjVcCac$K{yt}r
zU$bwCLUvnl7Ks+^XT5T{KfwticiHa{2Fr`(bOS`R@e2Z2p8N(>ixl^5g4AxNFc45F
z<Z5>L7v?Y?fpULBcm=Tk^X+Pg`VTokZ$)BM%_`Z6#{lG;>bT!D&&CQX$}gQgEWeP?
zEk8ff%vXJegbDtXR^eA;pgm4-e~kE;LF5;GI~O7;%C`hInTJVpFr5icUm<!DKYr{9
z`<RsCn>d44<W~Xjpu_5Xbw9mmCsds69`ghCjbfw9?=i)8z3i12G0@pN<eDFcxCM88
zK`r~S-Pnxv5otGCYu0Heb}nmef5{JVVgzXI0vz^0PXR|c`D8>0%x8y(S3rhE;`!w2
z7TpgPIODj{qi+Yqj@Acmd&*|_8FgNEH$HV6T5nf^>-fh@{~>PGR+(z@NWWE||Km8>
ziE2C5m(lh|BLgd!u^%He9M|;YY1+rM#&R$A131lioxb8Sqq%Mw?Y;(V3Vr5I?0nWn
zshH<nM)=}<Cnx97fPkHxyjaNpD>MurQt{{f$B*}C+{3DnS}Zk&SXPMbOQ#EPZUv8D
z{~;`!=i2@loIj2%iaANtNye%)BGjy<;oahU#Ctt`{hH&+U&U@imCk4TiPaR=Q@m%3
zLGF`v|7c*}ctPL$)!Ns^#Zf^<9P0_a2*K<`t>!qe3*~Y%o0rjS)&?gN4&m-wo=CLz
zYpKB?o;X(?jOE{~C;D)B)m?Aap>mnu>@6U?zp)0{cut`;fMDs~AML|pV`b``&!FLO
zn$CL@1_YO}cC+2;mY1gso^bfR%D;evvm#DRc_ZV+6jM5Fm5!A_;xDJINYcb3D+~r|
zMzSLS!~5J7zEfs)<+{?_1El=PBE4o*UgwJ$JE-%;R#tx9`K2Enk7}Ls0DudywPc5b
z&u+Vohk-m%rP?j?{&-^`FK#V%BpdAuYK9QeYp=4?S%*<7_WYtJog;1wB*N!)-J~g>
zwziGf$jip9ny$azm}`W6MNp|$X~|=I3qLT#*NfMB+VwNwKpH34`<5DV^#n&aztaGx
zniLg@sJt=CrwSZ4Ms}{VWqx#Z8H)L4REn;*CncMkKZ=FfY}UskG-=(g&-B;F@-M2@
z`4M<8s1|_sgy(!IOyu(WUi;Zw2c^lSy6A2b65Vu#h3TKb8q=v0MRl8{9hB?KgFcK+
z13)yN=IQcWYz}#C?zUiSTDt~ABGfD`7FtgNh<I77Uk)$&rIFRUT%cl7er)d=F&td1
z1KN>k%Xd-)luP_c`a^<T^=hN#?o4o$TCrKw<$4hY^kr}wa56!5e!PWgKJIc+cD8t5
z_1S~$(03aOzX&Ka>p%_dgaqvejs}3@``nX&N-Jn!Co4G5-kI};f4VAns^EILM}$eG
z(7l5(^$1&Sm*{41Wy-8N)pDsbxGqh#JQUhW1u7?8=kAmg+-==k%YIW7i*Wr}O$o%4
z`L!WKuOB*>XML;!8xbXoi@FNAj;7|M#WKlGhz^u_byb~bEmfps3YsMbaZ%3ChZp-C
zx~HcEvAy!g6*Hu@4p*_?m#;zsp|wsQ8WKPE#w>^2OFr`*3Gc4@*vH;luj+U~R*%No
zVv7?UkG3fn#WgjSIr&w=76DUD+(q^$;G|qX0E+R^^~_-p=sjA9aiRe_Y0sTDXR+49
zs_DP$^W-OsiISW<EYQd|OUGIluf@>09j&j_Ef{vlBpkWv>gAV#rc2YE`GT!r=Zy3d
z9)qs*Dt=Z+TM1dc+f8qld^Xm#_KfwAzwuaJr^{t;EGOKid}^mW;66Le_e=f|3#b*f
zW}U&>pcEEeT)p|9>9Ye~`?DqN?dp6PLVc|5y6V}}Jsl`OvWa5o#p%LCB8l_ygl9j!
z6)Pwom-~2c4o%)8VL7vzO?P#4<Q|Oo_pm|_2gOLzC87_nFLrukFRnqsU!vP_38_<#
z%!ia&tsJkItPYCNxhw{|hLG1YTP>FW&#wH`7$fcTK7AF0kOPE*lQS9I$W_VmC9|cC
ze8m{6w*959auBu0Q;_ao^m@H9IEfq8Aa;RxbxTNl?!>>;NpKMO@aN-ZZy?P74(1}H
zBqz5T&9MXOc`KA49T3p`dLCLj;qhv_sR_j0B*6nL>QnQkuZol9QoTA)N&s<3<sM!G
zlEp7}tM&Y_ZYka?`-vhi&`8rMwh<ZWldV{v{91tAXiT!Roh^loXejEk%0^Sec~8)*
zQamvrHGK!wY1b+)iT7$v3K?(q3dotDU0^&rKL8~*KHhb5b>Z=?JHmXZ(5S5vbkWX$
zKs74y9Kl&*@@#)@`D{!+<hT(*6{{k5rrNDjQI(xNGLW04Vi&sLg{{G+)kLKgtyE+0
z>VbLz>`-xb8$Ssjjc3mBAN73*qrGhw0Cx-MQt^=vr*@CJYn{q-l?v?#Q=Z(8^10{t
zbT!oVLKCpg=(gUP^YvAw-_Mr2BZJ#eQqp&Jz*$qJ<<fU<4*E}gMBwXg#ug)JuNFJ0
z@+lg$=m3wyVN8R=uLNyUXWYvRfj2#E|8$*5gvPK(eeQ|EP&S)a1mb)B{P)gBW=!;j
z=jG76@7B8~P&dW<bk>;zDx2kOO8Wsp%-iTT&Z`9cJs;`X=1NgheW?=AQ(7aLCM)!M
zaGLJ|t*fP!uS&5`Mh7*9Y)GJugzH!yjm^evWWLU<6QtRSu*_n?Y3aV?BfdUd8%mZ<
z70at!UXM1bDs()$RpxD*y^*6>AyF~YRaPq~GgYbJs)JzR^Q$cgtj|*%3<d+~8_?8n
zvOU9nC<S281j@sH*n*5ZoUhEdH>wKBQ#qRa5Ju^BWvT*-j#|+4kyo|e$$N9=U4MPV
zYdn(Wa58g*%UV~i*M$JYrRbl7mHK@?;Ov7N=+v5cj-Zc)|Av4AXoNHTH@HR_z<U0;
zGwADo2<)Z_IYa;?fDW7dv-8h8AmB2b#x$IIKmdW~`k9XAbO1I1Di?_iJku$j7wb|(
zY5wn_G8sOJuHLv+z{YOpL-B~=(T~|}O`za@<O26Ti>iAXJDXpx@!WY)G=xmevZE!@
z82j874J~UV+fFfKB@t;1@W_7A+k?f`8n3!ca2m&AXsLO1Zeu(wQUPjjeyFQKzTK?D
z(Eo%KK;rt-`C@Dh`!v0-W%+Ckyf82D(a!5JP{b~PXg%BNHqQ6Y1YIv+H~H0h>onz;
zHxEs=CPjf6Z+2dpAAp=sz*#0Hx;70kvo8+y&f4{fH$V^;OPAaNhB=SwR|Hz;iHX*6
zR(4*^YSMQiSGHmpeKN*_T_A}d9iKc-Dw?rh&@3uu9A4B>59#aZ00*Cvs}~$Am``GN
zfyn1wcOBFA)TJvC<!jz2f7~Y^{0OQT-^X4a4SwwvUkyN>=`fuvK|Er&_VKC9@5cr?
z{27t(T{ObHG&IaSSndHl>%_EtMix+s0(90fb;VM{p;mOV8q3Xb9{oFPZjS(sBeVc8
zFHZ{r2M0|wjmWHOnt5uKHPrj{=j)UJ1Ka$a&ZZ!WOL|%k9T8DgBqy326~~jt05hsi
zX1xK4aRWs<0G#dNO!Wv_0y$X@yRFG3MF2dRz(o^p>fig6C0Y#vZ5N!QGJ|^q)QFIu
zMvGM6KtGlKcE_t?kgny@#5ZTTQ_W^LA=O)>l=691Aj^WrmgVZ5ys>J#6H>8deZ5pp
zvHSfAz9A<+0Qwey{`${7*x8D77l1eO-^9k%K_l;yaMJ2<rVj5_i3LdE#rpNEu9xEx
z8o2D*zxx4416oXKM+g+e(Q#*3yy*;hjJr!0P-f!a{s{wZc9-SPy#des+~df4DCNav
zub^=5`G{y)Qupf&=oQ>Sxi9SRSCXXnr>XMElMz=ujSQv2d4N^`_-Y)AspPXTVxlXq
z&o|~50`2avz5lnxc&W1w^r_xdFY=cf**V`_mD3wp0Q_IM3_+hNznL5S`|3x>Yw1KQ
zlHy;>?mJ_Gf3aMZ*CRjemB`Vk?S6Dms_)-`NDP=Skm<Bwks!54D$lLw4%bS+J|0@4
zTP=5+NB%Xe`@>&s7S7`({9WOw7|KK%PNw2~X4=39Xzy#6`%8l~Gc!49t2c`Z47~uQ
z#8hUr-MM~4%t~)GY?X)c*Kk3aQatRhO+Nks*he)Vr{-c4CQ|7u8lqvB`qx$N&g`z2
zbKty<kR_|Xc%@lzI#Q+7NT^eyJDPQ><?o&B>+Q{-1t%ebWmC-ea-Dwa5{U%|p1FR1
zHzzN4QRQE9%YM|lI41gD!3Z3+`8WkoCXixi?|Z%e*#%v#lU-nFbbB|K+V9v*g{^=B
zZo`o*FRFhVOc4M^vk)fecmDXByQ8-k+)VB9Uh$~Is(-(3Y(!D<37-6I0DP|?=d?2m
z2%Tedy-fJ`s}#YF{VG8fN~g&K5<i6#1K<P&+_&0s*Wb>A!8cWg8b_ohCNU!u%AF~f
zuCUFcX5RQnwaj^i;m(c!G2f%zVYxuzP6=vdhWhrZJ<x{dwe|kyx>lX!FZ(8ZN|T7j
zcyrt875HjHC|iNX(rym$5^j6*2;+a7_O=C=wr01fpwbsaT4(16I!_+Zm&9R$Kqe72
z_wetTy#_$=iMUy61F^5rjGNh}4d6q3^_pbtWF^yIlk>g24I_|lo0iFd5lbM@<8kJz
zE;`}$;mok=8Bl<MOb$zX-}G|-K52UZqJD?u<WVf9&ga%D)VMtz9Z#pW!36a-8z;1X
z_kcgc1flLWBCja)UDA4UoR*W5q>3gjH5D|jklLxboprf^4wMPZzgM2L3P4Lf%UU#L
zRRF*eBLj+HcJ3pj<&U{4%yv{GSu=%BCsKKTJHay>K&mbOcwvJ{rfsKaHb!%4b<u&)
z4upXB;|nJnV;2#c0Tq9*i|<Fx&2=@%6Mh)6plO32q<Byd$3vSx>+9zz6{bkV$AwWT
z_PUH`=T%wG)=Wt9{AIIe#$blJjXaJO1T)Yq<hl~edP{XeLU7D8fQ+B7e9)2{$YVKU
zBi~H8GVe!sORI*4PB^#z8iQ}3y`v+BK~Ec$X0dL0Ti>ZDS<RdwE4>7uv|w~T4e8@%
zO>(=L%~dVWk*CTu8Zm~wwr&x`mfxHz9|LX>Yx>t<39I0a>;N{Q4)W}ZuuZlM$sa)b
zy->EF-k&PnU)!<)vLt9(v;ZD>bGd?t7!wBe-o$+}?6zI`9CYrih5jAQr2U}l_y81Q
zC8igj0QTP;@ItCVx)1h5Q`B$}oT;+2-3Vs@m~uO$Q0q(mf4bZ6CZkX<DjJlNvoo2l
zIL^-Jffi`>9B=XBv+Y`aUe?lw|7-F8ANVn$BK9=j(?nU4;2E;=qw~ice}*3nNW0Ta
zIlw0U`RKvpDamfb$E3S*b3oqa0Q+GPX%c4z`J4C9?WJqr(x7A^<O~^DqhX_A824#7
zV%C?e585SQmQ%V#A1tT9&>2{w-~BrApUYri?tTG&_em15AqIn(^#S+`-f=g`zrFgF
zTu%})C>jR*T_l5HHexp7^T!4MyZ?We<^2p+vS6CZCwyl7W)HzD!x<wP515T$_{<)H
zzu*^d$shd+x%~l*aW@PY>Hp?cK7W<FNx-$>>dGe#F`?{lz&ht1N!$&$NH%;sKA-=j
z1MKa1l~3+2!f<DQ^RFqTli|TdF#Egwp_+(Id>>;d`h|lI;O*facFvIm{U7Xo2Rzm7
z`#-6)mnM}`DJ?A#iAqK?OB-!P9h+=PLrW5}g(4M|y(KL>TO1;?N%sC<_vdpa$9SG^
zzwv*)e$VS6=bX=d-Pe7M_jTRZJx=u+`IEMyC5a##%LUu8UkNK7N-SHpR@Qhtt&#qJ
zU_JlG;{QqDf4Kz8#bsrGWj*tZY0V1$a7*Oa{Fwgkb<qZW)lqo-NXn||6{?A0{-*U#
zlyi4F2Xan7H~wYpvU96fR9hcp75l(rD=PZcVs{Z7Oi&##E;2fQx#uV60Mwvn@v}UB
z#2&g}Tf$U!FY)Q}=WIm}Phn<KFy&aA;8=;!QOV18*@Q*Q)=qhIUPx2)@sz{~1qbf3
zDA>6VWOy#Fn!YA^LC99M^xjHdiyc9)2thGhlKYtU)#+VR$?P@Wylkxw32Yc+l6;;@
zChVp*S6Nh&?8NKj?q24IBjJ5^41xd&vDb+&SOBSX6#<nkBTRSK<-W`}I&V@GV-$O0
z90p++vda4$_L+!7n%@k-OMq7Z&hkGKUc>t2`JMc4FQn}c9}h^a!=|2N$<F5`{#txG
z6ZTqm31B!p3|RX`273qa9eZRt6OQiiCcre1)L|n91ZEHzU_MYIy>VK03Wn8Na3Yq)
z)+n4X`zyRe;RS*V{u(GjPe5s|8Npj4pY`*DCG$B0<KylWGv*DWeo8jkZ7u|yhpqzA
zX`Tb^7Jc~{%}^Q_)M3N;!-9_`dnh{}&M!WO9Xz^<7_Q@?g#!KTD1Ly1C_*B|3KRe^
z(UU?JCUhVGRH)3ONC~2We(*@C!v=!E`XN0=mLTE=WP~J)v4aE%Iz;jUo-xU=-~uFt
zR4h(mJ>gAIKVuvb1hZvLS=k+a&z=d+#Crp|A%YP7$e|U6Lkhx{5q6G_f`*MfW<Blu
z4;*-P(69Z;W*?|KDW8+N&l`8N`<p~Cx16~~YqitO=_Wdzzx`liR9}gSJ)CfgHfX4|
zUu_#Hs>6o!P1X(8%^-91V$!yO>VK}ETt-q-lKOzI^!|XZ#Hkt2{Wx#$i-u{HMxN<r
zD2px;40)*%YaE*J8W!^UdODo>G`_?B3gs=-#2{7588aJ|Or$MLfq0xoV`56LN#S)G
zFg(e%T=DpF9-d3Te_Vrt<Q+nEVc{Za$Mur3goCd|PYm>%!I_Q|PI5^7cuIk#AGqeM
zhj`PnS8a9i#+_x4CEr_r57mG@D_3C_8Q-fTP%8sV2<5=+p>?f!;hIe{u6dD<k*VTF
z{=^=<0L+|XSbL~0EufM8Tpt8KrSl*<#kAAt&&BmF>B)j{Muk^gJqn7RJUq&9{HPhq
zmpcb~x58Po%X#(v*4EaR)naGPBtGV0)79Cce0Zp6D6U9J)*fra90#_8CP*%_hTVOz
zg2Qd_niDeJ{rjD?3U4LF5A@xDYRyfeMux}hjGbH8xAHO9*hnm`@$@iHXmdbG7p75i
z7DY9B5`;__Eqv_g*qw7w4$61Jmm6U(?}JiPwj%HDlGq}4KJ_QjHOwsCFxlyHi*70k
z7hcvp`}-sNqH`CP>O6(2WaMNt3cwx66luW+*5q(_k<WXj99<jN(|CUt|5dGzcRs#t
zOn^g^P-b!L$vCk7*t=y-tT45P!+2^e-fSS%<Yd7{ofHIegMb59cjROeR!;P~_a%A1
zMO*IX(*vnu_u2Vi>!Q1xTk6q{vd-q~U%y^4&o|z)CsEDQOwrQkJ}l@g6Wpv*F=>G^
zL(@S(_eBOnp&L-o>T}gRY}!@%E~29${AE^_BJnqTB+m{ze4fEZx6>^XU>8DX^Jj~>
zGW@VZ%RQRq=V#juznNL5RddX8KN_m;EX|APVU7%5jIvlrw8?mr&fz7ld<wR|+WPt?
zwIpY?e-`O~D&y)1tAV;jkj{(pY|$*Khc&QUGV^!B`qXOK#XK$qI2O6Wkky4RdZ6O#
z*$KOTU1Jd`HL7yA(_x#rGqWBlTLe7zV91y%wL_l{IUz5fdJ0zkDwEwURGf<mab2GP
z1fp6Ld?tETOcf5<*<MhZcwq|c{0-NuN#v^g3f}wS0>9&ib(c9WWOyw7y?QmbtSRvd
zCMWW^{ASSV90HC$^UBrXf<e<X&wh@pNQXn5mUA|4ZP5xRD1Uf98Sp9s)@;1%4}=qV
zBBQkj`ecHGw@xvtfF)X}=yCG!D*!gNwz3c=(z*tUYZNyGd>4%G?<*a6y}pXOHa|Ya
z-4>vH)iLo;?{6)wUCgs>c6YvaeQIrP%7^xIwK-A>JNobPKA6^N&(;*nU)V>-Jpz4d
zSnyY<l7>VD;xc@RF9)#;Df|qqbu>Xqv0&D2nDN7^;L)T$DXE%A(W`S`9ua{pF^3++
zYybRiSu6&H)jhUBjZ;Qsya@`s72L%b=`Blyqk^s!RlJ#O1t%e%`!(^F_rva&WJ!y5
zt~;+oE`Hm&Oz<siFy_BWJEt2Qz&C(RT>}9;BZ5yI4=2pqF7(i?n0v#9(CX-%Xv2&*
zea-^SZEM1|XqQ`C7AFN~>}n{%{D=e!_^gQ>UBN~w0WJWQy$(rBOJDg`e(lZsYk-l9
zf!FG2y^1d$0Ag>HVbRuXF30y^H4RaMH(Vh=m^tFG*GAe{bU6(WBeaQ~^6};}k&_{*
zTC;EBboPbY`Q(Eyq`{ie(e=tM!Y0NA8*}{?R;*hGNJrF_yKY&Xl<w|ITuzJ2gL54H
zGF`R>s-~xlV=grob=)r=52?!I`JASkgcZ-7`W_N~_vH~-QJh-zEbNNCPsw2@#<1wH
z1*^#!+GAEs0vUWJOm<Kv^&A90n9svuHXiuzvWl04Edc`F$E=q=5*@2nuC#7|)8=yG
zv60O`KToXS7Qb&?V#Y=5^@Hsz1@HhsQS^mwZO+?@itVRAjDvNaP;iOz+JY)i-UpA*
zT!mxV-X+CwwgQ+0Fw}WDCMIj|vA3D}zDa<!zB_NvGFvdgD#=F$O+r)273N4AShb2$
z!W2%z>IqmPM#fXfDG-VyohXB@*?D*}!E(wQPWZFfb;Hs7sG}iAVP!{h1NX|6ZoAxn
z-L#{LD#IF3$d@ky1zx+aOG;hSuZvq>(m26);f$rl^X|?!X|0|QXkt)bP~kB)Qcrev
zy_Q^bEl`<rZAn~1VKCnhIT34*)n~s|-=6&eD%qYEaRtIF#w2<z5WV<)cJbs%P%4kP
z8r@Opl(M32Lr%j&K0#Ox03XSDVPCv`B!de@n1PH8m7?#>omHLrRC#iB$vy9;l2qqx
zU<bC28y1a8Z)-h1!*TgzzLp8F!Rc=7t)MtTZp98`5Ri|!)B0nx9=cZN2Zaks(1t9O
zg_5E!Xi(nA5%M5NdMzLEE?XuK3r*R6*>_@Z$?a^u=(}{+AIX7*d{O3oN5Bkjams2}
zjft&4`qi0%)dwec!Cq89N#VtUt<6A1Bi|8#uG%+JcTBca7T0flUR3ZDEMlyBYLBiP
z0Dd4F+N{0}B0_@mB|w)s?Q>@HmkaT0uU-1FQtjv0t1*h&kud{ehvdFFk24$Z{-E9A
zPQKJePRm_PW=x;B_-Csfsr7xD?6Bp~ywCDVwZD9)DPf;vv!0T@_RBZ9K3K4G6VFZd
z(BdQ)zL&x>MlOH0DOqz`9$V!jZkFtz2otQ0_Is7;RWk90QHtt01A7hH!TdFqZU)S0
zVSWNrD)~HEKi#@)$wOGz;$_{T`>PYS_Zl|!*Z9G{P_VugKTqii9fPyblES8y4o+tc
zb=30%m%Jdta#Q$Y-jKU~ujf{Bzx&zT`?+$f(czscFS5<s#CQtRlWJB-oFtgLTRlzA
zkaRZlPdU=Ovp?+eUM17%O3w(gwFltQh*zGT+Yc=YxTD$96PJ*EC}(@54V<01+T5|N
z@>25M_@@EmHMJKlZYfv2qr-M{?Oulk79uwl76l}qye}vubb^p*Jf}{oUoG*Y`~=b2
zJB;;I_w3o@dr=e?<5=4JDxE&uU629C4xc{deR(YSS5v#pfdiKdetSGvQe?ZNrvrAX
zUbwJyqt<FUPdL6~5Yv{|S>W#Z>3-AQ90F8id0Kf?kKSXC*HuK|SfV>!$*8b`dzzBD
zc&gN~U8nu;Tz|Tskoafvy1r>jV5D*fXY(UNl{@J1Jf@Q)V1GL|f7!=Ut1F1<NE?BL
z0Vt}!TFvvbP6{k)ji7E-`i$Ev#kvp9fQ^2xn@*pEK9y2h9G^VNgo7Ng-h0eacIl3H
ztABiHitCZBPnUzk8Y?U$Y^IncFPQK1smbjzc&2s7$-PP{+ao9Mb?}_6tLP}VFrfUD
z|0CBoriuGS_Bz0fD*I594Ad$Uvmg8M@b=#fcoh3p=4x~JsUrExkE#T&88_D=uIr?d
zJB{Z+aq&)i!nM9}T$52Dx>0%imE)zbXsJGdsHhla+Ll|@Ra*feie-kt*-tqjEcl#t
zn_0J>%O1}^H$M97usu_=cm3_5;eV%_oBy=`pRH$PlmATKtF(BUk`9}r+|e2}f0$@o
zb-5Gq6{5pcu-D3urSeMeKytkO^DlD*#B`>}B)ykk1%dvwu+$<|IS4VJSvWVQsHiBq
z@l^Knw)mp+d9N<<spt3~uD+6V`?S%_C9r18{R2X4EJyS~v^m}Kd3KiKyza-m>%|>h
z*9g1fV#<MD{+gJ+1Sq9gWlvwUh&97p`@*lMaHGzVyCr#{b$cZq)Ne2qsY-R89ewce
zqpVe4H7u;=hpP|9y;-P4@NT_~+mb@*U$iAI&6wunqHBJ5n6SI`C8xO#G$*j61?)4?
z$ntIO?-R->%vdEofBfQejl~^j4eubp&1{7YuXF!PB_xS*k3GyVIozXaZ*RZDtot!c
zuqyhkARxga>TlTAb8DtWfLaDx_>q#7{GzGHm9zhFUh@z0jVA?oe@x=E=-pi}s}OI#
z+poId?V%gHDyQC6sN~+C0gfe{FF&tMhL9*<sg~sSxRUK=0OnxpJ>O(y$iw99_qKW3
z_7Ez5lX2vnrVg7*?}mcyhm4w?TyG=a1*b56v?b0%j4;lUOEKO&)>1y<ciCRp{FoiC
z8C2RgFmqRd@uzz)V1+Z?ECI51*zY-QN%z$Pw|%}>>I>poJ9PVdcQ8j=z#iAPLi?^y
zoIdTM;EJDK|E0d5P!QYW_d{SWq$nKQB6`9H<eWCTqu8S|!@N4Pk-y|BAEa-#-)wl}
zPcGF8fnvtIH%)$?=8L0)TRVT+PTXepr$XVsRFvLtU`vD1fc>nt+FaMIY8PZ>FT}ZY
zmfg~aNo#6KGpDh?Zh|(NW4&pDZI@>a9X*U|Z7iP$zk9F_L)^9e3dR6vO_4VP{6T3q
z*NUw8<-6W<)1i+D7fyFsJDA^G3v(I27q`K&i>>{5<PNIzPFGrG0?<>g`9AC^1dd`c
z403eX^6Vcsx2=TQ&)gfwf=klQUPMPDO*#ncj;@21{YUnksnYA<FpiDhLpW+Sp<ckE
z|JK`7nVm~BLvS@V3#~?AgbdEX=`R|b_a=k-(v~@D)!*BmhJ@rKZhP!(1h?E&=x*P3
z?#mBJ36vU-GjQ9gCSISO#U{6t{RI&YZybCo*OhoCr_<g35i)K7+T@hK2($|5P=v}J
zHV8=TOizgPz6msLt9^e3vj4?!I;p-9cBo7%fLj50KI%BainVv9pXuq%>b>bw7@BAB
z9jZKU@4D$w*P_j<&v#irU5f>ml+~|Vw;qCG*iM;!#V7{uM2SkdrvKVP(BA@!sbSNV
zRC+79vAFfn^^J$`y`a^QDSNHaYb@R@@a5r?CvRYP%s2JMj|(I`wmyG*xzPJf4;*S+
zbt(vUu-qXYW8Zo8l&?Wis5FkUeCAcYsGpcB_JBX{zRD~N=)L#E{+(%}{5;mZ4#n1$
zY=^rS1wcjD<2e@r#deS9z)~~+A%}WU@ZI<cMK{;-_Uo`|Z#sm=byQCAQ$CtK$<I9=
zc6qkj8MK$ur{E;a!2!u8h6`Ls^?%a?TQP-B1%dl*Fi^YnBf;G6dqYXzY249%8p>j5
zV-omuEia525~ABXcv4nRQ^LA9y6f~a&!<o@{&8t-IX5!4OUc=f7g+37`dR{`=sfY}
zivOQ-UjEq__r=z2E6<w>+Z2yy+V@oZXOGpvX+`0hcL;Y;_$xjGXpq<#aqg>?W$dT>
zCOZt&Ucmah-S;Is;hriiM*o)?G|Y8lEjR6P7Yv0mhj&pq1Q;t_Hi**#W`Y|!9>89V
zsReM+R~2|XBokulh817RR$Z>tD!$$H=i`g3t<r9ew&hLx-EU`y`>pZtw#wy~mFU^;
zelMGI9i9=3o0V)Ya#(A!S7|b2U%^iUa?+$01(-5*Gu`u>ncy}qjo+bAP5<|4#c43b
zgEaE63Cb@n1xylQ3eNGLuB4u$(kq>7)v$Q4!%R_5%fQD_zd5Jo34V0=y#LzGoR)<!
z%{ecI5{-prWw#EV`F0WZ5M1|R1EA1p3Psb3XjGqiD<plTh+9J4usmsUqxT%PWobIK
zmKWC)xaCato3SRY*t%un+<e}XicJ%|xr&-<#_jK%>Vh)#b;LRH)>GOI#bpKcMK@Ps
z3%j*eU((Ybx@uLSI@s)~>AbKdCHU<{y+0M<Rz<{j=l3?ohF$L3c*+T?k{=t_^J$i>
zSB!?^s@e;_($oxxoz8DrwJLYoq!sr#_yy<RhpOfd0sgm84=*-B)5=p$P(epTFJG05
zi-PqGC>O#1L)TR(Q^4^Z#o5c2-P?32i(|8TIP9@xZrFGtY)!{TKRG`*0xq1j5itJV
z`KXmViqUY!M}NV`%S0XXV{ci+UM@RmX}c_PvRA~?YqGF1waslktWlreA*wAruB%~@
zsGT|Nh%QheByM0{)jvVzKKpO0iT(<xgdqCft$osXd2y^p6u$|wh=rj=kds;(RHeBt
zF$m(8VVC&6YWeaqxZ2`vwfco$WtQT>@3M&duyw*&L4pRC*YTHahgVn$ooRn4Tp2jN
zV!^L;GZFCyNJO+mqoLwHDf$Q;$sgA)2v=~dkBU;u$dlXvcO@ilz;_KHm;e%3jgx5=
z96lDF^4pMeN)sGgpMKG>uRAEQ5l-Iywv09JJgp4vPk(Tc+;B5)YWOUS4ZIz`2E+9D
zsO>JnHOp-9u?VT>v@ka`mh>Ggh=sEY3*Y3U+Q_~mdvD#(DHm#rshN+j(W5LKXkn#6
z%Q<z<YBwjWoLyq7YkKdy`?i|<mIiWe0&@(4(82$Dm_mgd)zgRE#qM}-`Q2xP<ro*Z
z0AMi*Dqu`&-on|pxevt_3U@zS7~1n)<181SalW#l_;-yI$LY+mX6=Q3XTRlIirm+(
z&Ll7%2kb(hA;|@|pB%dPeaU97^3uqQh41$H#>#rizJxo&e3!sob_?zX&RCeqds!ED
zY8nh+d|+ir_zM%NsFb6#kw3&4yGN5CS3HwL@ZGfMBCHuoBs_rK^bbPL!YKn#s82~Q
z>}$Pog#ejXv;{^79|#jxA7Jvb)*wcm_g$H&Y)D+>BQB(0JL|<a*lsqyz7rOLSo%dj
z!KVBbQGGiBjo>{0<rAtFS4j6?Vlt)J2k6dYnk1L+^TylzeBf-}jG~i}$;^ae&YPcM
zS476Q8bBKDQ<MQqC^xS7LkS6@rEUv#U~`>9HR7v_!X@mvJK(sGuy#{_4+k{2X!+F^
zD1@&oGKn(7#Rwag*}PHD&GHX_;-J5i=f#DBUs3vX#_2>j;-%nQ68AgtOSOT1x|{Q9
z4)$#qy@c@n{fJqwh-pg%D?IyaV%~-ImCYQ`NVI+%xQk5M962{^QG|6W1x@XS@40p^
zuEHkWI}FiO*v_D-IP$Dsv>;gPo9VArVvF7Jb!nsw`SrUxf5f&V*S2R2^vbE4N2UtH
z2^NS*%uEkeV~yKxKNf>qG@e7L4OZF`!yKMG(SlR)x{Yyg$iUJ|LcR9E8Uw!8hAM(L
zoTP!`ApyUN0NI`88M`M(8)CM^gsSHj(@=#KB@i}jZQ34{Cj478_|5d`iwk&pqy79i
z{SW%d`j*Ig^d1iBj)Zc|8<_*H?O`VPh9_JMy+!eTrSD*qY1z%XVi*VRrd<vW`Y~|d
z7g&^!MEu;3VW6?aWwtUa8w=j@*s>bIB8r`H(Q}!)i~7h%B5ge%AKJQML#{B5Bu@6X
zsSd~;OP4wHLg$$YTziB$jTh4=TTTuho6Lx2GZ^1D^@{-1Q#{iJp!5K%B`fr=g6}A}
zOu6LQ0|yBOKM`@?MQq_>#7$p53hmx<7u4rkHQNyiUk-=2Y?-nBV&<}rW{&MKt6Sx|
z1an}4ldG-W;fTsc(<9*o;**{)D1T1fSx>-s_#r&N9h&6NaH`mO5wP~AOt_||XKHCu
z^*7!d!u<nhn;Lyy1=fv!Fmv49drxn#sP?*icjv|V(idh~Z}pmV#{a##;<PE7{U`Y>
zo4CY2&HpCL#pY$+r&ijOuSvQ-W9_LaXHNT;?Y%8|^)@Syffe)Yb8bcHJEB+bi&)um
zwW`OgFTJjw(><!HBr?9MIodPBtZBV+RP|NP0RM|UaLUN7z7URbytI<jPqg-2+Vw?F
zBsR$i7fRG*;`nNAJ%Cm0x9A2)6~3IS5i{?Pn6j(WaRWziaImdug@TXx$~iqL&6f77
zOWo@WWuN@cz9XC)4j=And%|D*@I+h{c8vr^McyVWn`3F)y(>A@zM&eP(>??!$yLK)
z0zI~<$xN03Z#44rRAM(u&tCznbWvG8_pY%7mBtTP1pF#=)V=2^S<lI*p5(Xm^^qVI
z*o9oUh{|YA#5oI1I1_7sz3U@vSDp^!n26Q@qh5E`tExPCB^L3^<z+`>--)iqR)|Oz
z6)dpKp_U)`dY)03Mt+=L1?SDqs_U9XVJ{ZjLW#}H)7byjt5;LDufTDGu!o-+$%Bar
z`~8q>{s!Zn#{}){vj>7Kh&m^24XWjv3*H(o?vjr+Rt__2IWf;74AKDU_RA(6jj`r^
zoVIn}4!=5TT=5fDwmzie`!h3Ts<}n^pTdP7_;0J``e3=GKbr(h+ePA#EM~9x^T71s
z?(2cPrJUCV0&Lh??4Pm0QBri=OeE#{<Hz+TKdzbZMpb*u2`56e*B#%M=;fLzV~JR1
z!hesyP?=l_crTN;;@(oT9=KrE{N6y<ue7?Bmd{JtW^>D?id@x_TeU0q(W6*6OVaeH
z7v>qiYjZPJ&FQfx-VQt#T>FXRmT9epQac<^QLZ?E2?dicrZ`-E%Os)7ajPycWrw(}
zRvO<CIrr}1#KapKd9RFn8uC;6o2(|ZJU<Y*npd+xVW0=LRepST{)STPWhY)<*oz}=
z+kBaI+Ly>GkI54z=&BmmbWP__=^-|$z}cJrkV_&>*;TF*{Z8@C_fzs;n-t>rrD58j
zwb6wA``hQuO>KbNU!3kNL6Z9Yeo;@y0;QVZz7*AxNZ9MXw(}6!=4!b(lMD~{5;JTl
zND*!mX$eh(UAr!xB8^$n(M~J&Ru=|;PO|x#vh=C32b@GUG4HFh(#dBN>3rW2PB8l{
zgx{Kl+SAgw3=0gwI+cImq;^4YReXO78#?gkEA>;=+(=34YYDqi7~CbbAN)yrV`PQu
zu3WIV(;sf8yLP<1QncO33C9s+#h4}I_I$he>E4e3LSe>jns9tf#H2kHmb|K}2Cja(
zcL^ltw82CjLJHiuAh)vJFYzGUgqa_u&+8=oGE&#VC_|X-^Swy3-#>O68GS(aS-&`F
zak!mxqpC`TNry?vK};IR7-`KC*Y;pz9oxI875*d>6HcKFZsr^OF8+G-cagQxJ^SFb
z@42Qs^KNvMis*k2F|p!@y9@k0RrG%UAT$Y~evv+qJ21%~G#}_u3ces?ds;@stUIGB
z!{S^=<B}3R+=4A#{v@yCh1j=c!#g-;np)cX>rxZ85t>RN&7i(h_i$eR*%R6P>aCH^
zTbtS%eGqLXrbF>pib~)_W=_Z&+tyaN$TE9fI;K9vEhbB^?~4Tw6$cU*WS<fKm0^*z
z19wd(`82ps&trb`x1iG8Ax(Dn{^mWSk{y02JDK|Y#t1L7R2Ju7X<PBkfa2|<p^M-+
z;_>3o>F!N}(@D|{I9Qigm9+EzrJQQrA@on0G+68_k-1#l8I_Hj7YT`ANJuW2Y^XxV
z{Tx1{j0zhXZk@Qrz=!DN;D*E`5nJt_;;`?5d)jpNURV5R8ilfCC87MetB%#N?C{Hl
zd-$(m*v0g)o0!DPd3o9|-O{hCEoAY#87NW7SBD?Yn^~ReC)Ve^wkP`tg$qP`Fc;Ws
z;1E58J&64p1N-ELjK)op^Pq8)THv&Cfng6)x~6dg^bqzS`ZZibO4rneRPGrxM0Bto
zX&RD;LEjMTnt|R?LyQ6`JdY+gL>=pa(lxaqg$Zic)E+6^(=^1H8f0l$W9s<nEe-3M
z6!b|ZigAw`F=~kQK<S#=kirCoXKIgB?x_t)wDGZGJ)&ROA56>kvnGGckO2-7+E}$F
zkMj?ce#puGWB#Pu<+Z+m9A2)Hxr9XG>>n)bVAs}*LXqJUt}$7DH(#gtz$eL87Mgl(
zdHlF;P6f+sd7fnL$K8?44rL#7l&yBP`mn6sThpv+F~^)Bs&6iul9rY$Bd{t{##e>^
zTl;Sg-|MoIy|%1O_16c*kbkGa7Et1N)EVjuuvv(QIR4b?2^5rzn!HsV4UYU91cN_5
zPL!XyKnnd&c0o(rY-yfEM6bc7N9;)hMFT|)<k0h8m<~4i<biAw?y+MCf+VfSW=Tkt
zo$|(hT)l6-FGZuG7zUgce~mJkMSM(nbD#?x!(&1`*prCQk)8zN?sE|f7@#1Wtmej9
z*yOVv_O{~T8fOp4VvkeMw3?eNGhBXs#s6LuLxh%i?>K0dX)ByOfPZm=fn-LACjrUg
zojnEvm<{35$1URm?0O|6Vn*>lDhe7S`o~r=L=+hGr6RLL)`NT%l#J;(K{i=-3K9b5
zVz{y{<REoDfqla8FUi~>0s~b8MJyXcALE8eD`R8;V3-V^L|rD(XahZoGy=pDWcOo+
zyvwVGl-Yswu}AR+3HQEZmW&$~vgC2)c9QiL)lZ$3qg8UeKOS>-R-0H+Epd1L1}GeC
zS4(#j;{vVI**R|G^3#c8Ty{_3DlYgB9^YcWH1sOB#V=<yNKCuXaFK&z?h;g4AhkPL
zoUN}%)pdU5X80<J6<IK5dLI_y2jN)lZ{(h1BNlJomvu&1kL?uf9mc63n-b+ey?O$l
zMtglvHrFngnffs1>lrcW)v@3+OFz~za$EJoC>>JC(`)#ygoz4CAN4^Nl-K@f8>qb3
z!CQ>Fe+HS>6DPh6$3sBibZCLQsecxs{NC1dWCIL9Lp^zs+P1q!$LyV*Ukt~+n7H^&
z&?qzJI=E)NT*2`T6p_}TPqNVflps7tT_HhlqQC<n4uUVbhrz;dQ$?0C7@rU-0iGdB
z=m+c?^IqJr7}$&+o5WK6I*(bY#LLct`WOgNFpCUAiZ5L+X8tI(+HqET|MogYCV@d*
zfmGpOc*e(!^N0@)W66^PQxg>)%N|qV5C(ubBm#cSRKO$Y`N9Q-)58^u|BdX5{O}JJ
z7tHIhID=y;k}Prjg&a&9$ng<(&@U!lhR6VJ3-FISG!O#aA@KrAqW)-b0oNMh8t((V
zH(#gT7kh$1WWh0)flZ;-KpnM1KNt>R6cMku6=8sdOakkQ=3!D$8<PQfx#AX%i|vVm
zD~4#Kp1^i+1EYtEva0965H5OTpX9&CBtYC=$pT~)V{l~*4M~Vanuv?3ldNqr6eGrw
z*JD;s^Du-5Bt)ENT-@kyOm=99D4js^!t9F#u6UDE;#V-dBKHJ#1-65`7U~V-YLIKe
zISfE~u1AFOwtNc_gZd8E8L5$tQBBs~^`VRVzNDqa!k#F@#^M*Q%RV|d#>Jm{?_*@N
z7Zz@I-f?p?hnpziLP-6huq*H0MO9Ro3f<0kXpLygI{{nY9!ahon=zr$9kS)@|NY=x
z>5HkI7jyg-PM>~CuzIX?DC9NlWxw=5)D|``&0DabNQ&gLU6F$_Dq5FC!thszTa4vH
zN>9kj$-(t-!33)u*i6f-y}LjHu${tJZ!zW@NH};>L~G<2U!hR*`q&#=2^aUs@b8{t
zT=!ZedCD1qu*-#KR!=~l2|)x`4=!Eg|0Ey9%mOj2pub<bH0VT!^^^DzQ#j>a$1S-V
zb{l`2o1_K1=aXcop#Eu`VmJ@-u3fw{Lg)|g21D2u4a~R*_n6D<o+hz-_ciyqm^o0f
z3Bw57t&jy!unCX<iP*<<oetxPXoyjhZw<9lvXLR)hv1Vae;H$c#6vd{7)mQ~f4m(A
zhxjkI7wfhpZqh7?jhbv#!>yy!J-#KnD>o=ix_wK_pNSJ&w!qa^I;2SH{@8skqJg;s
z9x`w?Xp$T|bz;(vs$P>#nw5?l;<IID9pK+=P`}}QFG0oNSpGR)-R~jV?_*N!R_N;E
zSjr&x#>Yq#g)u}!<3+vGL_$*fG}sDJdz{8#$Hh}I9L}i_i31J^LMizPK1IY`oJ?;p
z`e8PQ`2?izgG7UKfG0%w-?;n{MJ9;4NOb6ZeAa>+FwAAJ=^bfAIL_fvGBnd3f`Vj<
z6hFgYA6F4REy4Mq4};{-2TPVJ6>8dm+T%0^OBTJR;^*K=3+A2Ji2Vkof2eny1BkSs
z9{g|IgD~mPrJ%^I(GN<55rpEnfk$OhkR36E-u7{oTr5ID3jcIMxyCJ3D;(2v3uSh;
z+tg)9q?-aRp5@p6RTeH9=g4v>Pqy}|ANE&<)5HN)LgOi-^}%sOv08n$E2g3>rlNa+
zjY%(EJ3GHWveQP11aoe+kB-@~Zy9hu1M~$b?|r@WB8<$&rH*^4LD+SI_phkQe#}g>
z+>48^p}(SRa_YE`!?GM2@}6hN%JG6AGI|<eU<%EY5x^hBWd+xX(#al!R!laSrJ-M1
z>^Kb|N*ep^<VK>9hq3y5w;jP&+phi=MMj3n8@g~PlIg9MFr2tU(x|gc_0>^@t3MJ-
z-763;G&v|<oI-7^HeXUSH#YtdvSubNnaVou#9c1ih0WeuNw`-H+j`t4qd!R@x9OUG
z&W!F@_3eR-tiaQBsF}*@Bf8kYF#e%liTjcsTj=*|EEaHCk2r2(<RM7p#`Se=mXkF-
zD*@xDNP||cVpqbA+jK8!b0W?#V1YTugR)n_>Gw>pzSAr%b!!-wXZ4`OL{&6ZKmiQs
zgWgtVzbdS;n<*DYAXZ<_M}CcXK|g3+3I`tCMbJCs{tyN9L*uDreUh!4+yQEa5hMSB
z7LLdd;r)<H&bVRFau3TvWGj)MEhikhtT4K(lGoGX^O6rAG6RW`v@%#5?%LUC7(Xxd
z_sXro&X{d67U07N%^lo?h%%w~ZXxh>2+~Lr*kG85VwDDLK{1L8Q&Zq8NMeVlfyZo6
z3&t#B5lc15e<0D)MM&~GA3P2gF^pA=CJbatB&sC9fp<uZ<AEA{m@F<(BwjEhb%-UE
zNTnDHNTU>|qO~p(o+DN$%}G>o0HX|yN{7P2a0A2iN(qJ}jbiqJe&MzNbdiCKJTM9i
zB#L-MgrQ&(#R*|1N_I13Yh>)eKkj@sk>+8{Fo%Dbs$LwiNAwLOd1lPPDJ<ZWsfr*w
z5@Lm{$|1gi^NSCal(djw1Jerng<A^MH^?KvI;N2$#XSd0VSKU7+<8}u;>-^IMoTV*
zB0d{aD5gxR<{~C>M(K@CY#LNj-dZF-mQwR1?%d$w1Lt4i#OwDrr`_)cQ)GZMNf8)T
z2Dr{<8}M(wTyX!A!qWR3p6|^2UIu22xqp8mY$fCM-Dj%TRFc+qY&cm$mn0j-WjQI-
zb)m5Fxm2Q<yWk}_-l1iKN~I(#M|Mw%5;UlTOV+9$rI^n?n!z^v^i^14kt7p3?G5Si
zoojXJ!1piFaDBe2{@A-Cy8+2`5&I+fH937DI}m2xRF|X19FjEbl8H%1RiU++Wg;IQ
z)|cRJL2iuN4Qn+?g)*=?@8|m%S6KJ`MhX!a)uEn`J2PHe1^&sVgj+5}YIx(J@c?T#
zo`U_LCY(*I;UK+|9b&9_MgKhAfYJ5?Du7aCh$PHVK-A}{?2(3qzkxX<`i0xkU^hg=
z!G|<dih%WGqPWoD5!^~~4q?)?nDPW|tR^D|$pbNMf=!TYsB9ZfjcTjpWR#A6+32|d
z#xfc52Dv+xA2LLf$1vEYac+SdlC-JMW3n3ym<=aI9BqL#dVn-KJa8kylJOUPKz;&?
zuw<#zsDH@w6lyeD8uC1)Qwp%q(8mpe(gqJ4xD=ON7BY+zGH(6yU`f1jW8{@HnNL65
zDD5e^-BD6h{~G0JxG*=b+gQ>C*P<B0j%TjO3Ejy@;&$p)Ka_DTY0YDy2^o}5sDnll
zndQJ=auhC&&$19?*(EC?{4+FP#b6v+m{B$Ya1izdoDiE`GS0!Q_heiz&uX&KP?dy>
zmR6~DYc`k-z&&5mciRM-V!}AhLNYE<J%9_{ku7dA^^k$`+KnYOuu}mS_ekQH2uH7-
z=6Aq7kz4$bx!@*qJ1y)=lbHF`ZEET3VNQZ${YZ>$?yThHyxl8(9kwVKVlxYjd7A=s
zWrJPmIHFP2!m$y0vZJNsOZ+xC0XCP8p?Bxy#DSp;saIrSA00*gxXqE>4s#pCCziUQ
zhba0P%P2^ehamxGy7WN~^@4tAy2mDBxSmk2Sa_h%iO4#q04o@ZdW>tt&j)2a?XmN2
zzCEjnqdiV#moz_Y%Ub>9gcmEPn?E=bLytkbP$(h*9v<}(O7!!;oZ7fsr0Xl1qTsyj
z;eW4I!##0&6!#N**rdpVGB%1psL-Ux23sqrrh*_4M>1TjxFm4)2DPT=0>@jbG?<n}
zvJ7-ely^FGenV?eQ8w!J+;Ik-tEw$kwSYw^$K59;o%M|9H^rBe{Ywbt_Ti$e?sG)X
zuMzI&k$>UIA>XzuuD;^e43tmci^HE-0cLYf@@sp}yoh*nX!sykQiRJ{@tx~JvrfNF
zY7>~`j}gUo%)q*VB4j$jJG;zj8O_TdZ26wwW9_qmE=(C;yMEc(JHL9b-*|Z`Vp8)`
z14U83&Q2)0=2H5FMd_7jt`QLI`eb%0wvx8YA<D+q_TIf|n84BW|14d3k1AuVV^qJU
zRof7Okt?5rl#Z4usq!bAA^BfQ5>9@D+64yl6my_B1;z~WbY?0zlU<)o1WHiUXpKyo
z>i!s2aE(F6fL@{>V3vwF<dGn<q<BnT4xtkc<~sPCkK{*q3sNt5aHcI-VWT4F{2*0w
z;0E_1F-v6_MJXx+G%ewM+*>hWAY&r^;IKj)#we}Oh%hbEdkhYo6zSn)@gb5&f>Q?@
zpu7P?B}<J?C*%hx62|xgW+`uw>;WT55*U!B${izzx+j?cMJRNnsbW=lK=3y2>;7<)
zxbUVcAsJac)|Z%!KHMSSy0E=jTlaN8>oXbG!*+&jrTy*UR*}M;BG*|}J>7}V%z6x9
zyD`eg6I7I)VGG%2L$;clnh!IC2vdH!FWIq06TY^3-@c+^K5TylDPbiI2G<kW95ZDe
zN4wsxAde*=3SQjyb=4ZLE0!fQ-onjT{imA8vHDM7Y6zXN5s-&X^$~!xYlQNGM|~E{
zao|87c@kfTn;+qyof_@zkh}Y^m#e?>8S`~1gkStujI(1QS;WMPxEjW4lQ!qPJaUAT
zZ$m{ebKViXe3|hr1XzRE1M~ovyjSpQ`kSl<OD5Yp*^6){k@&%3$4A+mg$?KrC=u7+
z3r@?_aa$SjZ{6bZ1ua9^9M^*rYTw4fmvxXchOY?sz4t3?y<ecu1``Q5UEnycn!7kN
z&)Wwy)aXv~Op-}ZERH;XK>lf(b^yjlHWGLQ)+aEx9BS=Ui$`IADtnCX@^zUq5l?eb
z5offPQIl1-W@9Le(;!;!<uAK=7~ir?l^P;L)(Ztm5CHf;Q{5`moHP->A9%`#6}AWP
za0<48Y!_}E-@9qRYugp$;}t})LMe{a=won_>>O{)k@#+-O}xQyTnCKR&%<~>EKj)Y
z5Xpw64AQzEVrMw*CHJ`;^ATS?p@seK-O?OdZ<#P5`?`^{=Rof<aFn`ykp-zKJAu2e
zDi;*K)xk0co9yi<w8}C+Riijm!yO?2In5ve3kYzqPmm=8_Cli_@+vMPFF>lJ+27G?
zg_kBV<=`Yp8v}4?BukenVVZ=6`X&hl0eu+bCK*sD287<A&C$r~8|XdE<w-t_KZ81=
zOHqdQ3^@#iAQ=GY8#9&_dInZy=o^!{p!EO}D$xlV6!NcQ6J=o;H(F%S<R6?<3=XI%
z+pzjYPm_PrO+g6bXp+MrgT_ROjRn&Qxgn+wEZEXTA&N*zbRf;4#v0gMfqJ7ngW(6r
zp0T8m;D#g*=uKRgqcx^CEiyZFY$Fmw#*3I4Hci6i0Obd&U(twpuy)bY<R5I%<Ow9*
zn`8pmfT;Y@Jwpy^=uM0g6*Eb1(RqKS%tNbPon99!lh03Guxq~W)4jK_aoT*(m4WZl
z+VLd?zDme;pls$#Y3WilS5=2Ksq`Hi_Qon5jnXS~+@Oh^e9Vr!+aIp<|HI?w<~F|)
zsmIQRC{`c=Cu!kpm(gC0oG1cUr;nWHy1RG6P<$CX;j7=ojwiuNHBoGOa(d&!hL2TA
zY#i|IA)072B2g@034ArgZmF*-$gD;(VmgSL^a+-AG>bduqE<w)`xmo+iJXb+Z>=v>
z#z4{cM|0-ht*8f@jChExfLT*h#bpb5<082JG>O+jewgER?;iJNRSWngq$9_>;{Y%0
z)ZrSuXJ4A8LxJ+231HKFUh|Wf%_|&CZe-?)A3twF6ABZGX&OPoY+by+DxEseYZ|m0
zY8!NzL$M!)=Cc_d9Tbzo5tE@mU`^;ij~vbhJ&rU>Ly*qB=pB+|$7JIty+rc}x<)jg
z(+8>E6hgVQs=APAXWQwEMPbjdfab4E6qA+pgz2k{i<>uCp7l2w6tOqOKk4gMMzn^_
z;87sLR_bsDZo|yF>lZU@Er3&@C{qKq1`^Kv{Nw<Um)G<u+M+yLoL{@_M#DH`=ZspC
zRbyix8mj2<gyAzIx&#P^hBaJ13<Qs932{2qR%z#8LsBeU#+2wmJAotVk`7#`k&liC
zYVJnigXVLR@n8Xiv4M<ZJng_~&<T*Cflw^(f1~uR@eLffX$`ACvakosk1mFSAmMXr
z48`SxO;ZMC8q}AxGescANdDuY6gebE!VQB#-Z6v*Rry1RlXk;VQ;s1BF(Py!denfA
zTlOFWplMG|4rmwzfeG=9fdZ+?Q1b!Mppe66*`&ELRlbybK*5gSz4z7(lbJ&OVTMhr
zAAXnJG8tJQXh34QGH@tracW%C!S43h{TSGw9{I9lTC26CHVwF}KYj@9lq7G}pKZW)
zn)6Cg7+kvhVhD~7aTTj3ZjY;H2tDl$wK+E08~mY1Y&<q=8@Rc4;L#XkQtb7vn=|93
zDE*T$d2o&{mOxCho-ATxw46c<PfD1Xd_@HnW|$L5Eft(U-{lP)q%9EVb5YTXBYxQS
z6f}WF+f$;2kU0G#31`$dIe9^K{MC`5Kn&9*L*-5{TI5HXm!i^`kLeT^EZD<hVL^oU
z`3lb~nvs<khPVyWtR0xEhy@t1$+Ex@ZCq-(pO4qgl685^*|l9BI+T<FR>d0LsprC{
z0#P<a#lP{Qy!Ce(*zhgzCnL<PM+NPfp_l;TlVov}gm||B8+=6<N~B2I_O^n<ZOG`L
z7!N#nAMR<;Tp-?G*zQ^yL$w>SfJyd++M<qNA(KEqp0(9Bt1lqpL)Hn&E(*EIU>hf~
zLhX*S#)40C=*C1Y;GsrC8e)9HA_^v9yD3C5_-3<o;SRS=i+@U^6=`~U*mGo{w~>#w
ziHUg%lUCRrX;nBe53hUReS?9!=_!6V)z7!1ua!G$GJh}5BG|XPpz^HLl#|izQiG_&
zOB5;KJOH+X1_`CV1di2Bn7VY*FCyG~a)8nxN0MK=tCDDC<P7%!f0KzQW2H$E2eD0B
zQA9Gt<w#K_S$+&F==eCs(wI)#RIO2lLyM1;5s;X`&s)<=neq%x<}hqz9%e{L(KI3%
zk~WY=M}<GKEJzlDr`(A6Kku1FW(*sXmstjNiC)rr-yk)TUg9Ud>7-1hL(}sxa-k{L
z4iP$Te3$@+4W-mF%n+^rN7~vI%}lOT{PVtPr8dO4F-xGG$Wz4yB1Opo8VUJ_q<-3G
z=wk4oQPcMP@AZX<{txMkg7ZVUAHg+ghyAYw-pgy!u~hlI(K6!*kK;lI_45S%-1EXT
z^gnGqDAd5YRW~oheFv*#g!P)_eT|BZvns@9am*3*=E`TfHc4mRW2;#e8^smoPS__#
z$hF!#@78;(h5H)Eb6tCw`t^9Az4)WP*^gp3_Y`G>o4>S3C~4{;x`eiv?<5fR6p!bB
zFQdgq|Iee9QsvwclL=O?KZg7x^(YBFZw(*+CyhHS8PZ=#BF~sB8!|r+|7GLi1{mma
z?LouE1~2(&C51fws2{GB!k)Mdy{}{-6X2QA+eko<R$_fEhgP!~$hp`W+A<u<|BA-Q
zFy{PQw*O6ITH9h^fFRD{>~f_P&M0d257i4w{U2}qFRYTv_P=QSe<nQ`0><Kr|IVZl
zHX!Gh#6W}|YC98n|1<Ewuy`<7euHUSNBltk3W*Ch-hK1oJRZ6>-nZh=QH#pr=#CB}
zBl~y3w{EmeIeewJQ$p3Kvg^;EGt$#>SVyE7F{;Hjyzi+uaqo#I8rPN9m&6L6i@)sX
z;2zx+XA%^>u`KWxd;@*ZuD({w+=*1zWsH^}2!@^u5`%r<DH}utjNWZJ5P$grJKvVm
zLBE15`cmG#)1AMV3fuoYDmWCu?KFE@q=F(kI}PJnoNP?_*6pf0uJw?@;vNP#WaSF}
z!w?w9^D7iS-CH!LSO&gqbNe>WR=8!~?Z*+_9Dlq2gwaD`nx%D<`*|A<9z0?F%f19|
z3S;Y4zi?s3+aJsD*n}c7Lt9urtbO6ZojQ(x8Wyo2xd$^FBqN5w6CpMOv!^%#O)kMe
z&covU?FIbJ_MW{Lo69eity!&h)hgOX{J!}jmzPGRX2Ow^aZ5wl2@1UQ!5qCB-qPCi
z*amtt;wPn@ZY1`o16dDYA?si28&7P%3p&S$At^K9btId>5-jWg#7@yr{saHPyqX#y
z8L0|n!su<X{Py5!3c@jz1`WwsMpHCoI{!WsB3UXX^?##P%mO8;i5A0wjy3{njyEhM
zs2Vy2#@$WlV3abEFj_1|$++#M=(D_|E0t6DYC}TIF)Q^hi?k_<>t?QrIe`YKJ)M|T
zpV@2m31!-1*I)S7s&U|&u$frHebcNhe3R$EDk;T7f&;4hg*!J_4OIxm?r%pY2b$0G
zFJ`~%ThLLpVZ(;|puZ%UZ9Dm!gM-XP*J?{6g$6!#eR86*N^TL(yV$lU`8NBsj<^nC
zQolE=xW$k9ID$K)%%4Jis3}l$<>BxsiLnV$Xg<Umjx@nws6a}4jHf+Rb7vS;Al)Bn
z6xe4}j~Fhp%ozq@$fQYbgE(TyGZ>vVV>F=%mg+bEm<g_>#H9G%l!5ON&Cj&9>DMNW
zH)L}c3b$>!+gbZ*QqhAYLg}6&88VWiIu=7sVyFhTszl)09X4K@t*ivT)asX%J-36C
zlP=rnc)?IDl*mvlwkEv3{xab|)g#v{!kMzf7_qp=`KlIuuspugo+jvE$UyBAcT$9e
zTD8Mjx%ay}p3jQKE(TsOX#X)j-X$LH?EBAYT3Pu;*kqneTfVoOyE8X6Ev@aq{xMi^
z;)|cdxV}xYhe@Lpfot25mj0d|EjACS(W7FLjDS8viICqUf%r&ya5-ShV1bgG(}W>-
zYuJ2an7Tr-SZA@=$K3kdmuCgKD|rVtzb7W~nm1>I)KvRwIU($0{U9`B$O2?S!|4=3
zM{Nza;z9d8x=)OZY*$9WcB{p+Tt1lgHh+WAg}E~#I$YqV9U;TfOVXr^!v)Y$k$?gr
z{}>5K*oeUq7C{+0AHqcRDao@L{6ixOngLNpjIzDI5FRM>81q8I#J|?ZRqj=|G(ixI
zIYt>i<Mi=wg8P52QYuPNXi;JZmg<gOrNc>+;rle5{Kp`T%yVoQ%CL$?vDoF5$;Pmg
zp<&8H6q6f$8f?S3lt*QPYFG?#$eK?bRk;{q040MNJp-Z)RP1Dg&G<rwz=owSdwy~6
zm%jV_z}K!@77-2Ib@9#@Q_PCR+zeW)pTgxZ9TFnlW%~JMALb{cybz;vJJeOun(~TV
zDxye;fm4#uZ64ex6DH#-#AXm_>w5b$O#2f``y1rGj3axDm21Qqk%6vKT}t%@$}Z*H
zZi3VYXw8ZiwYUK7?d8&CD=K#pxRyp&dB`4RrUF&YbIyTIRw`#0*84rcN8~-@O2NYn
z#!(65-1^r<Lf(3$wsQZLD!CG^58-cUc=ng!C$=u3vWO&Neeg3*P~n#2dS@(8rx*YQ
zH_62NqvN9;FzVM?J2p?CvZ;^~*Vn*7RUVcO$}S=0;7g3VYd%vyq*w<M5enuunZ*g=
zCEff$Z?=S4>$jc*v=Vv8L7<W42xWCLJ2b()n5upn8hAN-WH6wn%s;p<lPS|l24z>c
z7f){B?l0$!>w`0M%N8s3ay}#*Abp=80Z_i!N5AgOSDo5yHb2m10VkeDo@MFh(|qDs
zt{GJ$MD8gmzmnU$^F!0Ee*2|l>!6GKT=NO#jnN19>OS$2P-AG)n9?8th`8D;T5X|Q
z#+_Kf?Xs;)|6<@TdpX$$dQX-Jg-uC2eL-#JFLLS?w~-&dgQg^N@h|;+Ws+S$8SNj<
zFlfh2uj$dzVhl5xe-g~7mQK$UX7|I`BdHAqmgrPY)((0Tq%vxr!!S9aRTij}0kqQh
zNy-vN;&<309>s&Hc{80(21#fvvp7b<nV!rT!opusG^50gRzp~+cbKGzBF!->nx5{E
zSIfsH6<XEPsD@0=T1|1p%pPNUEXF%Bv;mp_Hbswa-D6K9=-q<;6@=e0m;*)Y^ae|P
zHI^I!%<8YuA>|rm<<pM`e?<*3!#zb!6dAS0T8()!G}5SEDM*iXPBjvl5mt#QxP<47
z!wv_02+Orb>IVC4OmwDPS`ZXBr;3s!(gzCc6_+pJ%oztIuc>sUPUODRGh~X}b3dP}
zio^?vsP;I|MKzP2HT4GWcbND_!U(QUgfmw9^Q2UdT)JdNYR~TB+Ii_><5A2!%QNAt
z^ISN0(;Gsl0o&f!k47nZLq&gdHymld0&YfQ0hQ6vbWAvj7YZ5RBgrYoG#Gjv41Psj
zf*AY>84Ca*M;;7qmN|^ELUbQF(2;6_^g5+~1#FBAY=1+f<jT-+!a%+`x`B+$C(<Vj
zW@Ia(&ANwSX81KPlHkUYQH@+dgINi^5*WM!R2})DF{N#z@;ataGqT{v79vIhC$uC;
z_6rJ34@OF`4|d4~ydFk1&q85ko?CV?M{-;babQ?K#)OKO?1v=^HUN&u!)($N>&~qD
zc~<vGKs`pBQ#>kU-urt)c<&F6$uT2m=wp)p6pgc=cm27aR5Fagk|Z~>wIBomle!UF
zq$yWMo{kMeaBP&`iQn%ZSz6%JBM%=mG=>Svs5Xq^^~fd)V;}7SJeeN`Xh}9UHcMAF
z;op2AHS<{TryD++<}L9f%V10<P|m$J{5ostj5M_rO|v@V5q)KBn9l%l=qC?ja-5N!
z8uUNf19;taB=;OEHVk9SWKq$qFOuIH+$-efBL|N$!uDum{wq!~#)$ggBbCv_hA?(i
zsf_l(m`vp_q;BLI{YV}%T5QJ9_J5C5#uQxtdrV~{sbZf&kQ&Vr#^gYwjd@H{meGR7
ze`8b`If?)rjPk(PLhUF<fWUb4P&*dGA1%ii+pJ?aC@1BPW6LT>8S`I?9Agp6{~p61
zEi3seNygaHsZs3uFB$$=0>`LmFcuyE4Z~MFV^Hzs{e-C=)j2QD8noZGUp+ObZ^qnt
zX8r9&nsC#pOK<Out!lCyBIXBV(<35v)dG*&#h?BVWmrm7QPlZ<e$;`C_5dWyzTb}=
zH*UYrzx~{W{xI0fbbnv~_TDyXY&9`5YKAYr7B+0W*WcH3n0SpneEx>x(>S*4rMaBH
zaf9`wz~9j5l<LlAx%um#<OeDj)$S?#@!ssMN9?Qm;^#8F9gW17+U{ZSeVK3=nDeCA
zsNIc%BB(SGP}1F&!GGoZwnzH~bSh3AP%C|R{w#FFo^O(pni_FI^*-ZGI@FwcOw)Dk
zZ*R}uYKr*T&Sb0gE7fkHjPh|~gHQK9bEOpp<SwbHVbRrHy-_9R6nrtpqD3?HYO;3l
z@bH{EVsCHnxKa6#ic0;vJB*eFX)@ihgbBrcD<o#j>|c6o>H^u%XG(k|L}cYSkJ_6_
zn)f!xWQd3y`WZcC>dXT!+v3Xkn1y<OT>o(A&V#6T`C(I!s7@v?CNi9s&sOBOYstN_
z?4{bHKq9;T@#Ctv>}_DvB%!C)&!>fyALM7sq)toYy-1%fKsEAMp8qN-r_f8;Deo_G
z!Qkh&n0MDmz!x~b-bMUf=6q|0+0~EX7VU+e?}8=9*BdmI#NN4U8y%M)&@!Op^!o9L
z(wX$a^~>vbQw|ZBWfbdsS{xl6Nz2vWeitWY#C1Pe|6riy48~Ft{Vq__v{CF#gw1Tk
zppua8FG<(!udfru)Ya8<s)V(@;Y)ir=Eh$*ll|ODLqzgeVPs6qS7s9r=dPlxtl+>c
zBZ>_xZH*G9P-tFt;peCO1k-1);S0{cxx?ixXW@S9fd1F>gW~)5FA;YYu`ozTulgmc
z=Rhhb4^I<IkCyAQ@e}oas`llkm95DV**e@7I#iv0xH31sp{Vro$-A%j-fI8m&cAM*
zPxpHGqT|38%MZ8ICe{yxU2#>`7~R*vUB#Xhi&-&V6ek(PBj4^rCr!Q2ex4Dk%?6t(
z?v%$&nDA2F#ambEYmK0-QET<*7cXAewGVihM4gP#`aW-@jlWp7fJL;VBO*P80~I6I
zM^;MdH@^L>!;w#@%WLqsD0BSiRj>0Yk82NJE{S>Su_8!0ddrVl@Fmd6WA3xGuhC*;
z{-93{_)cP_W4jL@77xCV#`W~Nhp>s5M#=l@*)MHNBX#}YfJfWn$dHig!05(%dXvC2
z_U)5UHXNI{7wc!s@as^vnlLJYJn|LPD@&{??R|dY8PCPs+6N%|)vH(U+rw(s)BYei
zrYJoA*w4<4oh_|=ofGfRTlYtK<SGS%B!op`i}iorFCDoe1*5-M5_3#x_xqdE>(=&f
z5?wrPVW!C1mgjymU+ubT&n*LAG{0@_&w*y!riqL9K}1MSxvOk8rlB($Diw2<OHI-4
z=@>A{4RX9%{gicG|L?44!OZ2$gw0O%RrU2Xh35%$#&$g4*IW%>;>WJF489=BfQ1VA
z!w?@!YO8oOTCe7Eyt&!hCA;hLUdQ(a-pb1IW+G?}^j)tN(HH;3!Z#!C;<tkeI}`TZ
z^_V8`@T>jm2-e}<nVb-#G-tUM8-Gcs6Mw1cS%Hx<mWM0%y@f<VX~}yCsuo>Ut5*53
zMC^vH+uS-Tz9#zm`m(c@hHJ{ro%!a?Z_$Kndi4eGg%j}2J`B+xdlQR;ldic6m|m5*
zIHLvjS$!R$$#SVX`#CLQ;*DA^INUx}@nyDyMcc;(tH(v`zHmWY*iR+e;Qaj+5SK$j
zLe@mxDum-cmS=b9Rqr(%_g4V*#;ZSn%6DJSTGu7|^047zk$$PlQV$`+^$G0LI6ah_
zN>8xn3+Q<6iuEz~b#`9F{P~?h!+w62ztnY=r|Ln6nK7N^-vca89BMVP<XgBy=c?KB
zUHY}bNca*<A-~J9)$8vqWd6LxtozX4>iGralP5x>e6nUtThc8Wy1<nE)byps)$hAg
z<Ibmil3c;78n?2RwdWHD$G8Wpwzh;fvi==kJ9zffy#Q+#z%Ea-cE1!At=`Vn%z3AB
z{BNCUoiIx?Hgq%eTI0t#VX-va99}$=>Zrq*+gNItb6w!}X2p3k)}IKA=+=0!v`-*`
z{oc|(Ni&0GD}G*+T+vh%E;Rqv$+i@}jYW?hJ=&<U^Jn1TrHLVaH<sy<@2o?LI#JS2
z=?>~%s`0yEM6dhbLE<$r!Unz%`?KWc#X<%G+vB-s&Jb{e-Efhchx9^|)?uHGrKG_#
zKFw1ky7tJ1`l4{ZPfqjZFP&sLamS9cbulc=D~{C5L1J=raT|Q2J<ANW6vz97X@5nv
zcRF@;S6{EFsMycDN%vQXWYA92s}@Y#o@~2h>fUks^8@}B(~BUH*YMgGpb0S~#N_&j
zi5imsX!9tv+)*9{?%UdW&EcpQgez53ZLX}lGnW=GV4D8?fb-ER;+i!_=Upm*S<0vC
zf+ImH^R)ybBX<yH{+%J$O__4|$j5Of!Y)r!`98&RqLERge#!mS9#5dl<$EW7t}cM9
zad-UCdiQQi@zTFz%JJ?yj}^c~!9np<((PA2@|#NI`I*?^cEiwglP$)rAM9?L{eFLX
zvtTcLo9x8J>0Y~Q8fX2T!Ip~Kn)^PQ9CGnn`4_wE?}G+@f9M$J^*+@(yEbMu3+s&A
z+u(bNY(G{1p2=ClsU^_Y({bfg@v_AZD=l3kdi^>0x=TltfLKFu(!6}9S9SaI{Oiak
zuS_Ftrhe<+`~?pm3MTxsKYMo5J4tOVclwFvj=sM`b*mf$4lEO~D4SRM=RIpQi`V=5
zLfd>JweR8aLRoJP`pMq;wrI`9U9+_Wh*8AsO#?kgCP#j>b6xW?uH!yC-$-Hd2e*0|
zw|a0_lbKYWzfTxxwwZL@^>+7R()>X|h<|oz(Zl6?>A4Fmk%5C^zOxRKa6sHk!nX0*
zvwcTQ^Gjlz#*1EunR}j*#qp4fvo96w(7lk<xD#++%q85cCtBEkWS*d>F@O-vRd|-V
z@@pPeTJgrD<H7eBlw~p#&Ok`fFLJ)SAY=TrXsxo|0{2^RgR+E6#{(fk$AQtNm}Bux
zH13ExAKMZ+S=qqx<6znnSjjQ)`&I*U!s+)`t82qVR?E$wI4#<=>rsLG9r#*IahLG&
zxklU{N5~bZ>BY$S40~l1O_fws+&p%@QDiMPH>tH@yAf*I1VxJ_cX@9L=v9BPJbmP~
zF}H!5&yk>|%m-Ctu729eGrFUU`b5NJjhshwCFae26}X?}aNx<~JspkT+YVc9x!PSb
zLBHr(tV!wQg$8LZm2ZVOz(uA%=}vWC;UP6j0b{rWLiC9!lK9n{oEI6#L)DdP#u;}s
z>eplyevdaR3QJ>72G2N6Y+rk|z{*PLt<Z9%cw?V#*!ceWZ+6bJTJ5CLf;Vaz=`W3Y
z@?HnAuDSk9dO_}$pzVNzrwZVUj_(sEFWPcWD<S9egC*@_u(Q$XrQigK?4c_^?rVw-
zf4O$bLJ{-S+y`$UhhaMZ=5k?qR71hd>AVxC^<+tBd~9?(cmDkN17_W=von0@{glBg
z#^;LYA4GQcp$TRTXc9h`8Rny_W#_6iECYX=BtLQ8;fnk#rFlWTlNPo+98tMVytHVC
zj#xrYRr->azc+NgUSwtiO*acdYATOL*ep;v%C@OqBPe!mboQTWpP8%&lofT7|DJK@
zB*5MW`<$Gd{WCISO~PT$B6zRmy9V#aM44AGF+NprpCkWsW9mH)e!)|<Q26+Z_BF92
zLfiImC6kaL;YE=^P|Tk_>x+|Wb5q9MRF3cO(lEHNv1sSt0?ilKI!a8U1CCqTt+<tM
z<lj~|^-LMJBd_Y5O_z7S4=8YVgS#XiY-?%=_cQ)$khM4PRsBgcBitYg6#yu}<rszN
z*X4amlv%GFwdPcT!#bx=jlDbnNE-MW3F+5XJJ0<K^N=^<8>FU3R(YH_mOC!hd8^MZ
zAK`&s7ceB~ZQ?Xn&-hISRjY+e#C(n5F2#nw*JapNwFkvUYA^Rs%7m#kd^_?!h0|gw
z@EryBKUz&_Ib(Q7{N9&4+nV}1X8sw2vy7fAB6OSx-zju3V2`DhRc_m2EhtvM)naqq
z9=mu`eR1Sg=<?<avt?N^g@r5FvvSu@S12=J<6ErbBR+6z#9Y<?hU=OjV)vs^)vxUO
zdU^h&%3Jw+nUi<k%k9BqhRyc~n?Di#IsE6ooT7^&HZKj?P>|jzbX7}LM@RA5xaz`S
z2}sh7A4KX@W`QEE6m4gnCS=x~Q{es#QeC@WX=!QN3qE?aoHk_R%klp<%J!2`UAFm|
zE0v?0(n&${HY>lvoOlr(jL+M;{0j5)J-csCgL6u>QPo?PkXY-y@}ugpV~dpLB><Y|
zTclQ|I3*>@)E5SCJyihph@{5#Z8h0KSuuC#3)gJ?8|lg;--!k74Vs5nF<oc%lQ>_X
z=3yi_pKxZN-(u!MgU;Wt3f%nzm7~R64Ewsr@sD8*gC4ILyn48jFJz;#V3E&f&44+U
z5=+0TgjB!icCu?0WzzbRntJrj4;!dVq&aVend|4JChHB8AKUEM@l*W1@tSq(&SrYG
z!q<N2)#bGny#3(Rpmn<Jn&s)9jwX(1-CQV!O18xHG#YgLEk0qo!AD3qq42HBsp0p>
zyJ{}mF3qWFgUa}QVx)RWjFGR=)TuKgoEVIfR<TgX9m7)Ci`}RS^F36f=pj+;QcYoP
zG!YgqB!lGk?~j7I)%tWL?PeQGDQW4u#j+zN_NYWQmRWj^*YOA~^S$D1Qc_ZUjqNNm
zr_H#U4_%(s4g@3=b9rmhL1<h>=Q@n;UKq>jhq^qPMZJS8VY<mWUf#gCDwq^9Yj*v(
zPJAn%Y|;_OcIlG@r&&S6800iMLsHW(*VJ79G~Ik1&kiNC45y9CbALN-II;axxt`KA
zj!U0bNt;(0{2kx9c{rS)%n{)#0QIBV=vbqco0j2lq3injRqD9`YFufZuz2{l?BgjP
zEdM22^m@9v6~8;5tZYa$>L50NZ4^UR0Hy2JMq;p>XGvdoo0N2&VN>iSm?KKcab7Jo
z8!!a7d4D}fzAD{)oZjA#b5-MBo$<eOdflg>C;l)$g73n4RdDN9nyXnCs+JrsZGcR9
z;=P?lf2@zXy{yUh6L)EbrNLt9zhifqYk77kL-9@O>kBBcD{z{AeN;N3`X&^(ua}xb
z0`2YXGjHHTUl7!V>>oXvbEyJV6Ut9chw|W0y;$QmtreqZ@1v)&Qi|S6a{RMTLiOL*
z&XrH&@bTkw+dkBM^D()fo#xROp^^3%=0&jISZppFA1q>-IAA+)l{f0MPII|46m8h6
zz|&v;i$1uti<5*qinsP;6r~mh`+d}uaos++)W1?{dbW&d>o->B>p>?vF6-HzFO4Je
z$>w<Py9>*Y(Ru$|jdMQBto$<gpsD1ez=DHvY16%?r#Nm1tm=fN@QqnNpeFN8*bxFg
zpX@fBib)Gx=z`qf`WnK?(NhV@M|0kQ8A2Hkisx;MEfTHY+D(k_C<&=vDor5B-e<c~
z+W&m&KUWx+Dd9|}m&y=%ESCxh*6=Tw&#NMz(aeAO=z>lCb}LwYuC8|%G7Q*7UJki9
zIus2y)G}Q*Lt?Jm5Rt5udVhcM4z2{g^~YP!34i8YEBGDIvcSx9-}zfJn}W3OoKKk_
zojiok!8O_;u)hcHl~81-o_HoLpIEQtiGX?;YXl0lxfXY@3R<s9tg%>~=;7DK&$~O*
zT=`+W_4douzv2Wo#|HY%Sc;QcYYx_le0*`x^Vy+Z$EBx>;w>}}%~-kNwQAdY&#Pw~
zrT<QElTW~U)wp6Y*r(y2x<w<eRq`Opv-^K?3jF?DyCPg8KP;lt`mc3XmuEO3!p3fe
zgxGmUw$)Q$Dys5xE?k?w!=(MF@Mq7RMtj=q@`Ef&+A2p}1BE1FNOj5X3Bg-AI>*he
ztgPDcME_t{g7x9oP>p>q;|iX^%+f8P5r~2H`9Bqt+ZJ_YKAj1>7&daWl$d0%eUQG2
za%6b8e1%c}XBvwqKi!dP(e39s*o+?eMz#5gG-xc22-V2neq48nkU@mfOCc_q5RLpi
z!ED>q)KZo0$6p?P4e3Ng+&?$oA#Cn>^!Ur`6X9Xa4q>omV(#l2gNE<N`|d7}D;Ctd
zIPi7#z_DAKt1>)w3*{lDi8hF>EFXCR7FZh7!l9LuV_wr8!kO;Y7Yc6LZMEm^xYdbK
zZ9Po`WdY(Q4WUi}w!x-dbB;I;@$11u7!%Y*{`@TByyg6vpWm-Q5mgY%`XSHL)6#6V
zDeT{$To)9t=(>BqbVj->|4Y~-<aB?NMgQdm9d*@NxrTNlG5~uqo*X(q&p)Tu*;gY{
z+}$ycGyx%90QRJQC*$euY-#?@^)KUZ922(p-30EF2PKb|`rvg?2XXO?*Mj2WuT*C{
zCnq6;dLOv6y#c0EKVO!954q&-u}drUi9+^s8Kc@4kq{+Rf-f9{+57_4lP6A^#<cf9
z(0%O6A+~+uwCaJp6C5cicQs5+U@8W?9u^B|C7f?k{~l5!%`04$b$7&}bw&}B#-~U;
zii_j@-uMB>EVRQ!Zg0z3hFalTTAU>R^4z1x0{@S_?~ccM{r^AJDW@T)6d{}lAtWoi
z5=EhmBr6gXZZfkQBnjEsWh7+pEqm{goxS&7_x-z^&*yYjd0*e}@BQf4Ilulnk8|g~
zulM!7Ui<kP*DVbV_UtDL<k^lrjdCj5E%CQ<U-FwTl$85sstIF36b7<Ynr+EkAQq_h
z<BkG2ab(UCs0aXFm)$5cf(d|e{Ox^j|AYw&au)3D$}ovBlkoTR;{&m0eVB9V5R#V)
z#ZMuiY%5vYyDoiR35e)WXO;qhJnHK4_L7`D#mGp_YhDoSEWn3pAF?Mbmh?V(uQidr
zK1|Y|wKvY2EzGi|*LhHOO;owF55%9BUY;-p1a;<f<ySk`1Qg633X87Xhuc;b#+6t#
z!X(H+yzlSh14v(=o@W0{OiV&>b?h0>sJ$J9S!|Uwk^>1Njjqw+AYgEFV@noxAd*Ym
z4@CX(j!+hb<Rv#rY`rPr=UMoZrGzOj2B0pe^8wewe~498vcl$NX14F6j$c`vGzX;u
zAS-hpM5YYbDZnm#Mk8!=VhE}YrJ76xR@*44Lu~8$YYnPBt~lGjc=+0+GiBgYUH5%K
z>}rOt{Hvs@>KEzv$E%a!ad_?zy-*UK=*8*Zz}@Yxe|)E{d&p6na}0*+l|*dA7OXj4
zd{(nsJ{lFNQ3hR$d@;qN&J=F_y|Y$|&FuzC&2E1a+yBKb{~mV2^pzmV&e!XpK&qSn
zm2`hK6@MZ3@3z^c9}}crx_`9U`XEJRAnwzskxwlCYmx$@2@FH|dvi{F+n;~_=@aN3
z_^T897k2-as)ZqdAA}arKe_~`3;@*Q`Zo^#FQot@Bhl=)FXYvU1HkihvCpqm{vWLL
z<tKh^k)V3najr9SkOO03bl<(y^2A^G^}jFsKj>1TB1^Zg3uPhjwH_#@{8B?D@=@iD
zQ^Q!re-c;2XfmXChBB?lhw^zU|ADiS2?Pg+BDch#BI`)U_G^dl|E&VF$xuh`Zu?H4
z&p{0+`A!HF{Kgu&Ai&GN*x<jH#a>Wt@6N}g6*tovZ|%3Y-)SuLn=#tNgdoH32PC4D
zGS(JXxt9UXaq@1%)becC>U=Z5e0J<(JQHz$o@Gkh{le#H^cxQMwWb*utkmlaRHP&r
z%i8T7uk@OH6ykBz@~G)(6xVZP^9!LNB!arE+Ptua006%0$@0#0mbTQP)NL>`^r9dq
z{qOG47tRbqd+^xTS+@o9n_u^g8L=e}kdUt~Bqm;-F4h$nLn6rUF5f3_?9cTlpk+uR
zk<Ya1qCt8V4qI_t1w<s40cN6dO0wO#5B;CeAngxE=g>9S5)Vikqyt2dv0TmwG)b|W
ze7K&h9WW<6i!dq$O4EnrGD;mC*D2k;TLUjY?-^0n(aO94(*ahL7I%0iwyJ8>otO4L
zHg?y+vQYVJ2^!5lUT9dl?aGFR)9hhw#gx`R@7WD-E-@e5{z~7$99)O4`u{i-`%I8?
z;D(&Csc?F;a3{eN+r|i4>h;hYwLzvLv}n+E3qWtY!S3^H4IZni+U{0y$4Z@0E3ias
z^>C>|u(};(Rq2eZPTN90E6etjPBRA=WW$c^?$n89*aLpiT#nGF^znIbo8^tlvQiIZ
z-(F^}9Ol&Re0N+$HU;DFY)b!DA;lA=)jb&HXe);ou&Znd#-bS}x}`(RYRfaX9511v
z3AVm79F`i!q)Sxi%^yA$3+h=$NR)OxJLquoBpQ0GyQ4OET9SW$$Yv)?r9W?PWQ;&p
z^0*yty-P)6!a2R?S$Vx5CP72IGQzs~wU9X(=Q~Q5>3!r%?Q{Q~b^K+|5lI$-b+$Dz
z>1<(Hzq6mjn}sL+eq)5kO*PWQI|etMDN;I&<Z|#RGXeffX*%CX9Cpe3lxEdjp#$-|
z>q&;aMY_i0E7Q#?i$~Y5&|LJqFgsY`x4h7TBi8B6v}$<U?WICN?lOEOZASJD3kqUI
zpBX6^BT+DXN+0wZV%xB$gX@ymHJ9zXiH=EyOoe(8F)3$WD|Mw2^H{mq5+BNV^|%Un
zbYr9(FsmAMq@?U`n^W)QYg9(-w8Wc&tIgXI)8dpeIKkM@(YpSoiH?l9k;W|DF6Ddo
zEGIG(x<0<&HkoBo6RWz&H<eqnk9y)F!?2)hw(DGu*<5Or)p|OK#mWV9cI|6jD3QT|
znB{4S=lMW^ICPF{ECahj)KVjUDPS$V_9BwO{#$<QZ+7|YuGOnCSgPQ#!EUp@N?*<9
zXi7(A6vRu)_^rX<-j*_P(8uuLH5wZ&qqUrG9Bjc8cRw!QQ}a6O2{Rh(w<;4XlBC^3
z`$ocFyWO=4RKx00a!r(WW|@Or3H@<K#ncl>04cKi-Gwc3rX*faqdM<oAmT`hX#x+r
z=z09O;!@DlF|)t=D1UZUF)oPO2y=E<z+Q@Wqxahf?WRXFw(nA0>NLl3-WJu2;y1Tq
zka>IK9H+qFDF25TG<By*DTGqlE}La$M9^$v!YCWDw7#f978k<S_1c}nd`x#=%j%yi
zc>elHs*6Q`@lst&=K0OD&gx)rQ3?Ou;g^=vg+U<(eQUF`%4T<Jr9!b%ug|7|A`wMP
zqjAuJr(&)X45U75B3&cudXjcWdSfM5m%B_`a+GdbsCb644k)T$enJ+#J9*Klnx9Lv
zE=Mi*0WU~Uzj*obMz(8J*|u;&Cvd06<_WKsHu2};Mj6Ly@!3T+vjv?#eE83x!Y>^r
zm3fA}d3&(u$&sn<ZFKAAjB?owmBYrPQ7Hw*I197343p7?aWXEQ>nWIyvV)AC$K~Uv
z@(Bh(w$&0i?Omo|o7j10dqTRSmNdh}hq?7@PSf0a?R&c;odbajKaI%$ve`)8Nw4k0
zw>)o1u!KosLXz=EzEi(lYVl`={PQQNLx#Y^9UAI^83qJZx}e&UP%RL!VyM-YoW$>f
z9AQDtlg$fWS_-75o(=WGFkNR)qH|5PpDRwi7YwqWhbj6?4ZZ*he~Kd@6ls}2ht;Zz
zv_y(lbXw~Z@Pj%^&{(y{mq2NGMhU4U5t$k<;w6E36X))TxEiG`ge<My-tNdSdWc?x
z@wiV;*W9sNT&OL1H_yh)RWjx6+l%8*khk#LO#*+9$bZ+1HU}hVj*fxrZ)R{EZKb-X
zZS@2rE9((%1&FAwIt2N6KnCGUdf}6<g_;I~WYs|!i!2t%gT($`KJkg&KOMbmJV+MK
z$Ef6Bq*z(;#~|nlaZdDpcR$2h%^1D9@qVkZ5X9ReAL{{$#&Bu2WuD9N>N_(a=!xVy
z02p&u7?8BJRfE1r)LxrG1&hNq)Vw;wkJ=uiQ+sCmMnPvV6O6Dy5hr~t{w-Lt*l8TC
z7uRmhGQU_y_X4a6Vnm&%$XWqrquOReulI#_vW}`ks6!Z@P`#|CmNz%7t7LfPwfpF4
zMMcUsDpa6F9zt?{QtPRY(d4|GAWl&QK^CX>wJhWD8QDnBXvO80X9q8Ovf$>*S%_H_
z7D|a#j$coo>#bN_n{^gg)$R@pv|1{qn-yHstMcOpl?K*YGkwh}7SrD6DM<Kl<dOpi
z=s0lld`5GXOy=3PL-01T%UTzna<lkNshf(k<M`DqvqrVTpp&LYy5Et**m(%*4a)$Y
zD6e^KA_ouzswF4vx-Qr6w_|?M7(m$JI#Kr*v8%7`M9zr}IQ-NU41&KRM0gd-GvTQf
zSo+uzB+O#}(?qcOBJ*#z5^d&kK@fxI^+1)sOX=#4_XdVTHBGS)rw75654k{4AEG<G
z@3mgy-Nw31bXMJxtldKQ`Vp<GQ;{PXsW&=;{9->LA@J<??c)JMbU4Jt(Xj{@VsZoo
zs6w0oi0SQv%0`D%*JsL|E;6T`N^C!8zu$|*52a=C$&4k>y(PEZb&&lO1j9f`?>)vV
zJ@JGmcY_#XvOO(-sFvV4r68s?k&BqgXR+VQA!k{maG-KDMYk)}U~r%#BL%eLk9(;Y
z10alTc44B!XKpk_yCqgNe<UV*1=O3{&JN~qS*|D$(n;K(F4Wuz`3U$<%tdFc6c!n4
zHAjoj4v3F5MahB~t*6l8AidOwkLWO*_hKQj!`8n2-1>g?Zz>%Ft+6iRZV=uNastct
z++#d~JwJ*YPPBu1uf}9Vi~`E-I&fC_Jh33N&rlc6m<$og18N7En7g8yq7+OO*=r-^
z77Ge~G)HUI4XwlNGX&TJAeEf|y}<C#(cPavVP3)PHjVbueR@LneI(wxIPADFB76;b
zYs;=eP)PO*As6fYsnZG&`W(t*x4mt-qJBREG>RxHq*&=XAoDC-SoL$1PTL-pof9ny
zk?UxMuxC>MD5YT-1WK3!P+SyrkKQs2(rI6EL~+sS-eU6E<GhRQes%<uv4Y5P5ggVq
z#HrQPa5LZ*z@=lk(OUGvVkK8`cE>)#`rH9z_acvt9zsLER+ec9EnXR3qUEJVw$*T~
zY6b>oh+BI$zLW2$LT#U&U=;f$YtIdZjpCx&mTdsezxU4T!s3U+$j1eTT{UQRW{R#5
z>U5-g7>gjg2)|vS`YH7OD<TPk%Co!X!Qh?4<Vt*%_6u|MHy`VT=U~$ZKfYHgMaj0)
zzi5h93J}Bv3gmgv8{GOiIzc``%MDb1n=nhmn6b@V%U}O?bo+-)(}cSA640B7iDinv
z60GIgd^PsQ7HX81?_8Shl?v1Xg?&4d4>m>04K7aBUKxO~L4P(LpZRow=k=34%e_g_
zifL6!X)_70t}9piG`clkP1e4X4+C)^pN{`!P|Bgx+Wt~MedGMk=1qK*SMD#9TgO&L
zMyNuVGh{{sB?T(h9nv6;AfszA|5HGc2a^JisVqKO6BmrCK(_yNqKTHEpRW*i-%h<Q
zr1_D$YHe+yn^!~pErXjVosyzf({YFA6<!u~{U!IRUa}HsEV~+yeQ=*?(gIL=1xyV%
zGDc%q2lAlQnpKDC!#bH#678LF+`F3%aO*-+MD^j)=G#zu7crL1DZ`P3m<uj>i(P{1
zpobo?fTj*UutZCJZp&a0#-%LRXeC~6Cse$5dN+XL-(R0+0wZ!-<8EeTTbkoj@J+qP
zo3x&*wF|IZm<GlW+?p7ZIp0LaaaZP|H7U_CJykba*0{J1kPO0t!5`V!m8B3BZs1?D
zX<QOjscYWI$=>U+*-5Buw%k7m?uNFf3%K9cZFe3ty=i}lH3z-$1|M@%c_)TCfK!O+
z;4v9#z<MgQCSp1m?x-4#G=RiH0c)4_g_TH0NeC83CZ4+}@RpzdR2eK8c&!&5)pDy2
zz#eBC#0v`NSp`Hji^Pw<X-J<J|BRye1V^@?TQDYcev__Z*zli>%*2)AqJ_<8WyYuY
z9&%l=a>>xF3pbo^OcJUU-*fnorm<_4A?S_TT>LhE+so`7@{WNsnQYSxj7wHWfzpJl
zXjr5*c6Rdh5bX~95nW?s9|fY}C8UDa@V4&zN88U<iGz6*f(hsxKY2=Os=$#!+8ujR
z^HB=cighO1Nx+glJu8RMijCJ7+OjiEXKg%vhdMIC-lL<kw+^q2Dr_uS+slcT9j=?L
z7Cd+z6Lir_0W}CCQqJ0?mP9V|ew+_SBnZee!Q$o%W33q`)qr^0@QQ${(HpACpuB%$
z=<St!N~`+~H=WnV!^0>nSBB$;FRrY7h`UK)cDvx|zD1gK&x}cmnjG9Xn3gKGQtyhb
z00Xr;EoQ~J+Vzw`{H~@(+7P3X0un-?bSiBmVi%2=qGxSZ5}0<u$${*}A^W6frGDAK
zK|z{2jJvrtG3{Qllem0FEa;%ew8VEiaw+RHjkOBqh?zVNM~Teu&OA1=wxUon1d-_b
zgv^2h8&DAvay$l<EsD@0JLV8Cy({*J5O(gYoZViX$2zux*d?8^8B&h)q7vj#5(bjq
z!f7<p*Fbk&Q8DlrMfEtfD>*r=*Xq$)d?BAeOFHnjkX<Lajl(;Il&do{8`(v5bkx=l
zyk_hI^S(Q%McRqM@*syaGgIctq=VA(xsfvAYGwAEW^>!lq~?K=06G>>>Xd4V66?eW
z*(KA{wE|ygNB1sJJav?1vd6*Ur58F*!<2~AVNd-a{<)f`zbI}v1EuA_|I$RPyn(C`
z^#$FqFEZ@Ggo>jqECAd<$zR`<X=4K%4r`b8XeB{#ETE9VEB@>(I*uDm_7np6gTR|0
z*c-9w2_y(1{3;3E^G=8PES65|qn797&@$GUG5Q4%6}2S`3)nAY?>)dEQ-gs};t;Q$
z7rrt-N~?kw6ztH351~XicfvLTb*WRbkyjl-{?K+;KU!HyzX+*NDUgK*1`oBP+^ASx
z=J^?5NvIvAE6v5DDtO=)bsn^m#=Z+<%k;y97r?t)uKJ>ujwR(7amI5|$}o;U0wH0@
zq8w7Oid~IyYfTso1BUyRQr{E&NIU!ty7@5I9Bh(Z3p$Ga<ggR3S(u+ZNHDkwTKC=Z
z5(YGUb2W+-)H{GNl`XG@+JMi{4oZ9Q@J9r-Kk(ly;#*qkC2%HN?CZi|^T>2FSDF-b
zFZnJEWK3qWnJ*-QiK&^OyWpnhTu!I?WETkOym^df-02DFB(3;|7qb^y%fV#8H3t@f
zS%5i#F_W`3Lex(Pa!g8I@Z#-B4K|q!=twuhfTRkr3$hV1=}SFS!!g^yoFWbZbcBhv
z!+i49(;B`Vd!MtRwIqn8PM+&1cy(qSt%%bUOzPPy%vBqsqJmP$bE^~N&q4fWIN4b^
zhK_^g?e+5I<z(&Q7Y2kW?R7;cty(-DSBxdn$AAGuZ+6h3;;8aqHX5M8lAy>Lr6qTT
zFh-~Hfn2BugU2PbvbG?ovGw#lCOg9N1|<(Y1hic3X5zY%I7EZmlAxN9%?wN7cRt&K
z_$4~oW%}GQFTo<bnP&&f7l-b(fxu1Kw0}9JD_{v=NgEgLw8Tp`y^~)UPl@EzRnF0A
zBM`)`1zN5qlY^Ws81aDV$hPbSmk%_>;6O4x^rDwl=IvS#IcsFCPJpq(Y<B>5s0@lm
zUJFfP+EZ0mOytcMC(?{Y7rQJ|XotD17FzN^i%2yn_TNxx?FN#dayz%fgy6!;cGO}e
z0K0{9H$T~%#mmateSEEvjLXWP*ReTlxlP4tnOg`VRR-aH`+IHxW*}$+eFUHhj^B6p
zebDA0-F1Qw<f9A*t4<K3qDAY%7h5!~&2Ezs)_)v}Q3Xd1U|H#`KGQRGa?c?W6c9Gj
z@Y(vxXh1{PLR*rS8%_Dr@*u7uQjYExN#QeL(2mh*h>i{myC%~>NqH3j3qG>tr|#>?
z`SWl{XKj<$A8ncxKM&#WA;X7;3>>$oik&F{<{7U?1B&cfRItvJKbN_u=?_*_?tsYW
zLePw77&LJ(^`9GUj5Kdz)+l^N6W(<%3ji7B3v22k5a$iylelSByc7(;pP%SpD{p4g
z?~4_*&W~6rXW@O0PT*m<KWlkFP+B4p;5+i!0r_?W=gt#LTCQ;2zRWW0fL6s(EkK;h
zM;aq(7yI0tf6R9cFk={Zo&pFjJ31)~P!;|$x2mMXB0S~?(Liy~baq!09bQ$vyf~p$
zr}!h_oSC88`!MNM2+r9xPb=Z~0?TvaR@ph!mQ2=JJ+f}}>Ut6bOuiC=JrC?S$Vg=L
zj=(c={5gakL@t4$Z`f$UT)Ft3+WC1h8^jJE1mtoI@L7t}rLQzYB$Qz~cF$%d>nKo;
zW=mV4Mj244WK<FPzA37!4I(`a5fe*Ykb8Hnfi-5wZsMvWh{^;Q)W2pN++x9PFmP(1
z@y5zr%9t}cBsW}LXp<#T!Uqh0!x|l#eBEZbh+j2ycM7DkS}7sK4I<uuy3mE2V39hS
z?;FaJv*yQy4#s4A+l@Ah$YpxEiq}osC8S<SO)N^GYjdtv*X5eN9kD8WzJuS*kaH3)
z;nYKs<W9aa95xvg#9Ub*eq8ZXGqWg96#0r^xD%-6`s-xCKj*yuw_Qhh+1v1puSUqQ
zQq|v_A7RmkXaIp=0Q$s@u>v3z-B-+;j9KqgMknq*8wz}AfQ}i1bO0r>_yKyUQ>?7-
zozPj_sl}CMcZTt}djoJ_5AhmTFvjo+-YP2q17Jb>kM4#i9j%moua~PWMYlagw@o)^
zoL5Nelo`8KxhFoQ<s24_&}$2qUd^**dG^!T(1$qxonpvL3*jl9i~xuW^;Q8_fdhS|
z2Fee&5-aPM76Qi++$ObAJQ*R&*c>4njOn;OKhn4}Bt%6%L1UglmVsuyb3YApd7$9Q
z$JM08g39LOw+P)g=Ub38{q>%+&(yvAp(f_&t%eZM7-5^S>3Yd7?^o#Pm|&1tQb?T=
ze*KiTHCac;0v*dzK_;-)`9g>=P2YdD!$hk>0^+=pJyUm6zylTjymGhG?7^gTQgrf`
z^e@s4(>QEu&Vim~a7F9gyXYLjX&=^`VsBsBmoqzndAE;YT0<cYY3uPLswT9PhkmTS
z^7it3qtUBi+yioH4YGSNY;(6Z=+28^dyt&bZB49#>7jzaI_O$=v+$88ss-Jyy|__d
z0cv8=s)ISV5et7vJvY&w2Ff{jbXt8u)yN5y`VzurUY&lTEA3~}iBkkZ4<hsk*+`LK
zk!E(?&b}hD7+rw)fty!08qp;KKLG!KeYHyfw0C&B&+RswAL#`8$QP6T`c8cj&<)|&
zlk@>^U35=<Clj6{6Qj&Q`{qm?4&-$~t8t3Oau{xX(h8k`x}D!m;$pLM+p)^Ds{ljM
z&0eksce~Qi(td!k^~n4uHDo-Q^|;PY3lt$Gpx2km{-4Zg;j{j6dt`QoYeS>}QMbUn
zeG@3XGKXjlf-CG4H=P|2<*vnKL<QqRARuu4u5UJGxA@%#HKR!oXgpl`?FU0}MQe!P
zZO{dtg5IL+w=d&lSnkcoqPOH+-fqdTmD1@L&U47@2UjTSyV3gNPN)N#$h9H**WlQR
zbEBVxgze4I#rbBH)}N6*XQ)b9<f*2>F+kSdF!@+yxjKppETZRp9dWGY9|uu>(dQ9G
zoWS&-D}le*fc9%C+L7Tt9^fm2Ss7q(I=1%wWX}3Rfi-SbWj9d;cBZ@#SP_lM&Zq?#
z><rn7J8*V`P43ipf*u_?pw4y^XCDiuTPi&dYRsFJn#ZzNcWg`{2D=##fFKC6g{A*V
z0aW8&34YQHz(zMR6>mxqhTd)%@)-=YzT1{eITnV_V-8)<UaR+H@;&dDO${bv0E@c<
z+8oEDAXZ0YUyz@dZ}j0P2a^^&-VZ*i*Hbj8Yy{wMY&9C57~nPk%7yW^CQ$InMn@|}
z=}4+H3yNvR)6m#^H${>m5!iS?xRY()JtAiS;`7Zn1fO><YO_Gr5y)^w9+<CD!2wz+
z06@1-{$4}4v;hcvM1Dc$4CJS$n$!B*Q<$n7ib(cVSM57!G#nv&5g!KUL3Y&*%|^R$
z3r~MJMMXUermK~og3w|R?-UjZriKYn-F5<H=-_6T_}9oM2=hJ?Szx46UzI;2)k`NZ
zb8e%V7p=4a!}_2g8FO~(g>zHn!p{zLf*Iml&<UI+qJ``hu0~y7mNCp<9vpOZVg{F7
zG!ilS7!T`@`kO3}u?@WUQnA8}G?Y?Wuf6Xn1R0}%g#lmOVL_|*US~m2>yia#(JQ7K
zhCdHj4dX8JU<8HA5~%s-&io50Zn68*3<mE4ePl;Tk)&4W3aSI#mdty5t+8|U1jUZ#
z@%MV4462JATbZ-xKpg!B6a{i*UuSO&y0;ye84jHCN}VZV1m!KO6B&~mLtRL{;!I~(
zKo^ygmFXNSt}aB9#(}7k?KrZiz&v*+hepNYhkEu~w=;v9f+N11QM|zpqL(ghWVBc^
zPvGJ(dG0IYDVAszc*U--rVxYj^Z_+0hLO6BqI1v;qp2g?mb|V95_OcLnpvxkmyn$*
z2qk;q1}F^lvVU3P%~lNh)_Hx^sElx4S>}$>5t)RiD&Ahc!gx7KEVSWPW4N#mlW?0@
z9{KjP?c2|}Yq_6fIPlZcw?A#)cZ}iDR^iiqwEFwf?0?z%tpBR;<=A68w(n6pjS)S%
z-<70k<lRf!L2cI;gnRp1Mi*5Do=;)}Gc}8s0{AM2YN_?tPe?_#+kzxvoEF56kxg|7
zp{Iw3aVd=;KJMN)KBZOaK!e07v$>~=(){xcDmv(33&S=#A7Xjh6Pgn51UEEf)t=}}
zsKypsm#CNx)0}r2OEE#`RX#r1Lvm>5jQP-OW^eECTDHLs6I?Ninj<-HT_-xq=24_l
zhR9nS4+b0>k*vW)%dGJ7On*GxJLPrf1$5Az1DCJ2X7}PGs^_0}=;a%QhnolMwdUX~
zWfop&`p>JVSZ8(YMF-i-n=c|eVns9AVbb-i>W(w{#k4-xz=5Th*Mnmyp=ydkc6-a_
zEjskeDT7QuGAm7#QzR*8wHjYWM;xPVY%pm7F`x$E{A@0x9CPV%{RNRCRo`O}CiKtv
z^=Cd^*d-4diSz{&n^jb37#W#-@Tw{<@lR&Pw`|kxT+0?m2g|g5r<$)|AfPDy=F(HF
zH|SjVYS|{~yWoo2^4F6hxGAIg&9lMsS`ACxInLk?V+<h;qB{yx>wP4n=UExIRq)KC
zdNoi>hVABx2jdNO(OR8RC0lX9&8XKghnq<Q+-5DjPyy=vlkCya7KGR=d*I3k60tQ$
zDKMH3fFW3AwCG6IS>TT}j?qPeK^|d+6<qC8`IA$MK-Y0zL5K0<0sPOjT1v{Qc5RNp
z&oepZl6PGTTcG)bvvv>~?)(vYb>I>in5}2ogS)2t)-ZZa@;DC<4-HCm_{Ma!ag?P+
z@@lb@j%3V3XB98S8ELen7L0(999R-C)pD0K?<>IuV&{7Ny6~fOvY&~&gYZ+6n7G;I
zOZPqe>d+r+!Kmpu%?43RR3AM!NMdARfww8N9HgFMAHZGh;1?ijiz#|u3f?PJ%uDlD
zsPJrJZ?SXLe8(%09O1ko)wj9VCYI@YzmDEJL|WFu!W{956cVK&RClTaSoE=x*g%}(
z!KC0fVU(#WBPs(c7`<wKla_O+_;XUJR;z7f@s!b^OHg>+&sm_Aw*k*UX&fyJk!6mX
zGFOySSuej?EES|z0&co~!lyoK?|}{<ze?${q8^-`&)Z!s(q<zn8mMwlxc@~LzOwbY
zEH6rSLe6aE;cOM7@45w;Eo?Q}>0is%iCq_!jj+M8*n$?TNl;~*^cWqb^?;P)?u*%d
z8}m!t_bl{#DOfLNbUj(UsyFC=4=o3zz0Bj{%-UV)Kxu)-wdDi`nV(j7QxwOLu+g^h
zqQX;kJ0qp(7AJ9Q<GPO}8&7mu%--^1UbI~%>BGesRIh_O(Bd7@=@|5nZHyGMPy_k0
zn8_T$YkoIFLXLt54{ay*QyDoYm!RP|j#4RPAcyvSt<K^oF%K<p6JW}riR@eDc>HF3
zhSP<_9%Wrg7Rzc1=v0T6X-?+glvAs_GK|*)t=C8MFSLWpk{v-0@mM979HZ=uQE+ux
zq6#|vwzctfe%6-j-OkpTciK}23CLRwPNtNUiLDL=&j*X}gV9p67g@4GWZuV6;+7tT
zgjDNzq9VBAr^&hFbAa>mUi&pLZd{#9zE0$APKUWI8Y6Du?H${ja$Q5?uKtV@lY+pt
zH2u<d^3rw$))9B=MFf{cZKdc;_}Axt=0&IHpW&Q}Wzu+jIWyDG>Zex663V<-<kIv*
zN26Eg1F?n%+G9-3%Bx+(Di&8#3FagR(y=QHBqVk^Zs;6@UX*fH5IBO+pvje%Y6e<b
zOxziCXzO-oZK<~Q2`zK_5S=eqj;<Cu;1pfgt22@|%k&=)C<<byw~sdGXi1~9vT!3P
zhTkSJ^=zC$ZX)j>V`m~cIwujNQ~$;&XQ{AS@`z9B@S85(F08#HM8y*Xg@uFq`p8*X
ztr@IX@2zRR01?Z1XNvgk=<K`1Xm04W)jGw_rNUOKoanTHh(ZUIMGr6JupeUQ2PHwZ
z^`#3#SChxm_8lYyJtm}rR)!$HQqe=j3P`c0n49!GCcvfv?iOaIjkub1udk$gco@&Z
z1syjtEY_b_(0suANJ{7R#iO{jg6Y6xAZOVLaFpzoyS_0I5T}3~O43L0=&R#gX=csm
z?hmDpTKP*d-oq{rDj79;EOxe;HN6W5nyFU{#@X21K?L6E*=N|gdV;am`4-Sw<aOK|
zR0!E<29V5@Q17%{nLEs?_8~Gx!gSC-G|+Nk-!YgF4*8MaR2mB^Q@E+~yirMn+6`6^
zr-z&*K&NxFj{Z)i&p{AyC@0;O2Zr%2>2XFzaIsGKoja>zE)blKj0lAPd`shvvPnlw
z0%aUX)^{1!Pb#Efrk8pxX(bM4XtxAO3be#pTT4w77=y@5qmq=8GF-o3EJ0H;K^h%}
z_XsqX7f3CqmbuegF82#&Oc!;50RV#&8447*Fe?6QB1PR|9?aDguK>yfwuY7$v}Q8j
zm~&iqDpJVK+(9+@%;SI()B5;Z)u7j5JQ*Eq**)-TjPjL;%W~7H_P}BkTaQeUpGE~O
zka<dB_VyxshJA06j=4$IswbmKH@Ln-8!c#8_T7E?Sm$HXZwwnFxRUQ0ov#j1vcosm
zuM4@ikz6ho^t7IDOz!cXTFTOQWH%i8g$$kcuwBgb!?QQo_mP>G4+iQABnfHU3YCp8
zG_49}xSDyKx;J(lofu+!;_X_kEw_UJOUcXMsXd{Yf+6Vh0Yh<xlhGMebD##?@sY=W
z`&M~?@5@vgu>p(4jud^z=FoCWw4QLe+*1|jcDLlk!&){?AS8-8+pom++yW>$YFI3T
z>{;B;Shql#3m4mnWrMNU#hh}Ez6jZJ)WGC}jLOF8e76=mnvXQUc)WvO6%QRZp?jC<
zX_G2$&B^v<`9#N+lpOpj4$X4UO0+Nz^<t+`8;b(ayh3hUpS{$xJ8d9yp%s^m5{~%-
zbkp(nQFE%7<R0@g2JVKbHR~_@>kll^L8$x3%UM#MSQIqHSaa<PQ%W34$_xhGcufN6
zK&3OuTFsiT_ZpOu^Lp;Ltp}65xF-|AHRAEk=va)*&@jr9Rf`S_gQOupFd)puG|X~5
zNp36{6^lr(ZfY%;d?k2CE%yiz;u$Nh`^){rI%t8g8+U_64@O$9ng<y)aOzyQUFir3
zLv3}gE=8v1S&tVrOg!h;8)KqNS37P@ecXZ?NL}PeHei;$GB-?OYe2!1NY@vGw}%qz
zj#g?z8)!3Afm$tcTLTH-1*wwB9iZ(Z2(8t}dYy6JqstNUY_z)uGr>rwvsPT!5{Ed@
zf~lLUt*uO}21g?zqn{`5baOMYS-sV4G};-55>TjTG|^5!I4cW+w^~lsvu9KEy72cC
zrhGVqPVAOS#AL?QOuaB1<{PA$32>isPkyd;lgIgXIjFtChO=qfe~_Cp?mUI)(|~Cn
z!&I*Jlt=pg%!!uhD2I<lPUOphf#>A-jE~ch?l7Nkqz}zg0ZshzXl10caoCS6R@1Y7
zk}cpCK8a-+dwT`u%&Fon7`2ACVdMU&XrW;nalcawf;pO<W_Jf$;(F^NqtoMMJ|-EB
zilF3c@u&hDwzkroQ?~rd@eu2n41R)6TQPtHc8L&rFeqV~(E)gtwXtjRX#&R9qfxen
zf<@E5%Em3Bp%R8ky4KE2=y-Ge08~^n5YRfv;0`Xg>s8BGHR-Z!OB%)5EZ0&mCaITN
ztfGS&mRW-f@~Y-_v5e7gGeK;0p^NK3@mRB>fo%AilJ#1zI48^mZ{z?{riRB@+jsG`
zR@=fvmbr<@g-e;(mB@`5Y2l9)L7Pr2TG7`+C6|gPBXN&rngdJrZQ$SooYSF?xy#yG
zZ@Q-$w>G#Gxc*Kpe5-Ow`wU*&QF_-~>LA_bj+Ace3HKWm{*-MyziIfsk!@41dK+58
zp;i#9C#eFa>$D}_;ka4U_ZTmSxu6VS!vUP*p`iW?Z_R09XUwf~q84h=eWN9~fz+O!
zo=p$a-gTVzN&hM4GW`d%^fT*vg~i32x4qG{iSvy;zy3@SNQX?6+07Hm$COt5(v7tE
zm(_7Q35$+jJ7GTF>H%b4O3D?Nl?Spyffn(qh0nNDO#Aya1Ib&%B&!$3C+!`xXg<C=
z9w-N9+UdYtxPu(4539GP-qY!Ro}V^|eY*i4QyC~(TY6~wz+AVC3~NZeUL94TWH~6_
zTu7_}hP~1)2%{3L8qNT%-}LqhzgSHH_n=>vTA?o|IxEY_(9lh8RKiJc`<%k&mvrF-
z%8Vfj+8|!*S#m)ElWc^d`L5b?pd5&Z+3zMbIttO{7>AKAD9)4I>66<1WK;lWk-&r!
zE}i&V2+!YgLB7zQb|O3D*7a5a6;pJ416DB8l^371fn^8lmC&*zaQ(~7c2po~zbh9D
z+gB%Znn4-a3Bh0d#$=O59AmN;#I9cTiAg}ily*PA&LgYSnivgSZsv@!I;!sdx60h3
z6;hgfb*pA=J<tHw!KHR)D@(oB0;^*+Pw;iKK0cb=txhB6B{1lzfSno&&gwGLQr#ej
zC>rDd<M)D+gU=I0$tS2)`3?6L@eAIcdJw>$0phNKg*M&Q`an#yT+D)!Nyl>THviRm
z^FE+4Hej~?a^i&j5PlZKxueoi@dBm$!C>H4WCaL;ig!Z|-iK-LKu$tH;o`_v1Ezxp
zSqAFaptiHAo&UNpeI8Vc8Vu(^m<yQEF{m&gME*y!!4+mZpQ(@heZuud=g|cSd4Sjs
z1m8e5O(AnBgoTC?hp(mn^hB6owgd6rXlY|arU2IuZR&6D1R4Nnb;NaG=9a>6@z)Y6
z%vrRg0fM<;lsW`cKqzAf)duk(h(i2ti))bC^Bbc!gv|X5sSP9C5PsCxqc*avY-5Me
zjsHUb5uFZf)ZztXt_c){(P}Ub2Mqfm3or7HAsj7;Q9PjJ9ftkBB@-p!1~ZonhMW*;
z7D9X>j4F(4gUC0hf<2Fl3U}0w_4N4Y6K?kP4N%i#HnU{dJEVs%*vWaaMEuXb{+~a&
zYje4Ld=7>k2&bo}(r&hJ`?*qMLP;JPH^Yc$&brk|tJdbzdw*jgKZIOudI0~8l)jn(
z!w4P(8zb)*t&jrQh{MPWgxDcJ2*#;{co2+!gkUQerG*Z)k=>GioOy=P=Knt9j9QT^
zgvdcC5eQj<co2*&{HA0ZMkpaf4nhS(M9mNnf{~SfA+=#t@HYj<AW=IHI8P8v10izX
z6w|=SHUuqUyfkFVg?JE*`A2?`C_~jrHffWI_GFF9vRmbyFS&G^=7w1${a5P1%?H^U
zmAs(eqqZ|E!-@Q&eN`y|C01j8!~;}ca|=wpy;8lGlpPfIJ-k`;+-$xv92A7K$Exa>
zt(;jc-_!>Oem>OHHJYgZ<QMgqrjCH(+;9nhp2lr4FJHa{cZJRm)#|)|#0}<^a3{He
zo@mRd_am$~MT|!px|eReccZkvlVvvll5Nl<B;=fL6)M}&w7{)27h%^aV^ngi=Fci|
z+%3S8T|Ufrt^ENN8=KMc3>UbYKq0l%e12rQl#bGD+n?i^&wPSOT-1@YG)4LD-=8kW
z5)=uh$Gg!<o-#fjdjE+^(#SX{-3(m&045uDr5govZQJx(&iCLDkc4{!Huq-*l@i%&
zGd83gYLUeAM!5B9UK5PN&=d&?J@t73Ld=65_?XeU&{@zAFCM_B3W(E@VWI`f*0r^@
zrvfWpa_UHdygAnnbapc_-nQ5%M8JQUUkzNwV5CuPMRRt~5pGjWFJ^zC7CX{v{>3@2
zuCbq%C244A_L7_&TEa=@egI8NJQk-~+8%FOBD66Z55j$LmFO}aTbirloqS8Ntd;w~
zN?2fRqNB~1Q^$2t8C;^vhfSAQa<*QdC$?M_IDEnYjQ6OoTQGI|48=Ey@VkzHqW|!R
z^OWleXCCipbQtyFGd;U+FK#?VH0pX%#i}N)CyO;3&Ou${jllX`%>!?6yTe5hk!LBp
z&^dc$FkR=3@px;a!}K$Id(+9Re&LfAVr8A+p9ZcAa&j7jdnIhI>9nU3Z!MzW9HXBk
zLbePLN{J^6+@IC5cEfII$!Xh;IWq1UZ+dB3X2qpeQ0ho08E`_aV5*pMvhAIG{i8a2
zFo4Q>y0FQOQsBsI^Ugvr^y+oZ?gelu#f>ozt<LEu-GIoOZVP6DSxEicY)i_$#XsI{
zcv~AL2}Z*VjI}Mw=I#QW8@^oDQ){!-KJ0h=ta)ldW5V3rHafOta2NRub6fl^hq<}O
zw#saPW5Kmm`IQ#UN=!L8?5o8+=+KB6TycCPWZ(U<xbxrw_p7f$rKA@DGsFTlX;Ehn
z_Ibc;EF`1=OwAoon)NsF4Co_PpJ?~q!Y1j%-f8{>9r^Gg<<x%Ru!H?>!|nPv^R~)*
zvjYSZslldJ?dLL#7l~Gw2?RdSOg>_|-tO(~ZRV=-;`#IZl&;Pb?G;9&P5c{ziq1xY
zauO4D4bmQf3UQq-Y>b<tadWhexDR_3f&bFS$BnkvtjAkTgrYeuSIUKFZkY~vt8QV-
zSs9U;JiIA~{8}>r!MM+`$-20s#JjF#EOb!GbF<#N_;?4t#1JioX^6=YqRF~YgU(Pj
zHMNDwE~TUSVD$Qt8v|gV3urGgo&BR)__LDu?<W{tA?2Q8HecpP_3CEvf^h1R567fm
zjd(#p!B}Tj*6DIkGAwwOc|)HIpRyQjirQKvfUV0kMUW0)5+EEk`Z8b8|0A*Yg!uxa
za0d3lmUW2>bf8vJ;T1=Q$t01!zo=Ugad#iMa^TR3*_zh^k^QJh<!#-$A(s}SS0aud
zw&7(nx6UtuqXon6Z*Z|%t-XA7{N({MGBSPtC<U&A-F>2>_FLGx?wvQ#r9g+;GZfaD
zy2;yzTIb(iXjcP{(0h#9$6UIZoYwX<HX&xI1dU3-=fb?Ey+w{&i#p?ar6(!TdGf$S
zj-UY9)(RgE{{1N*(v5D$`VIl^!pe<;y}WQp)a{Bcp_^<;Fwgw}27S(9(b0jonFj-9
zmR3ets0_Vm#Yypgut`nTSXLa&eHe4|=1n2ylecRgP*Ixy#~iS#r7jYnqT+;gK<jq=
z!P8*kAGJyFdCGCZAASZ><V3;m9%ytm%8a+23S2BmKNBYxtw?&pxH`aknh4pQhX_zY
z-s%)PNhTc3Epe$nHz6V&mMz0SB_=7T=W6{!Pu5~5^Guw{WGBT5Ff!D{Y&WugfN<-j
z8j>*`87;d$F5$@2$e5J*Sbq4W0o;TTI0`Q6V``@Yy{FF$`tV8Hi6)%l;^LOq*htWM
z)gBgl`g^rOX*g#7^&5SdxlZ%)ZSCg(<4P#Aj&)=(?sqd8U%8Rnr(%`XnDG=9t08tD
z%o}n3!LlY=sm+4H`GMg23vX{y`_e$OyO(qOBClrY7^H1(x#4tExD6Nr@4IE?M$M}&
z%kt_S=_dJ6HHmlJei9eJwC~>HCK-s0uS7>5?e7GVzisY)g?If-LJ9By@}g?OrR~6z
zfL`3<MnJ~-VPNwl32+XfH2u!WP_LrxLyy7u`9V#WxSN3lDQBCb2?AVm++AMObfEPV
z5bB)IuYLRkpG;jnq)*cYL_fPzE(8i>QL!(B3--3P7i=Rij7xJFUWLm4?5Q$U?wh+L
zbuZvbak0r4M^pc*B*l__FKpB%m1$P`7TX5nfh)Nx2C4jcyNEF^IoNsopTx7Znv;Lv
zstuRM^Cf%ew5IFC$jt1tDQj+w;JW728+07+wkuT^ft-Z}-=-?+N_<B_e$`tV(dA)Z
zUE?2X9&>129Wd|*YOkA!M8M*L%@fIstDHBMnxZD3y<ssLAUUw<a__}$3<&A?6nO?@
zuIh^@$!IqG9t(Wd=wHcY8@Uu_>nd|nA>HuGDU+H^Q$-OGF@ffA=`e{@BN1l`lC))D
z%E>0O@7=2mnzhD_TuMhU=s^$8i(wbDk7_mu8zbl}8@9WmQdVY5-8Om?H;SIqFDv#9
z)e3E~UhD?UCyc=kRQBd|EGD;M=Dm?#947YMpf*kN`l{RAhNt*;d+X}5D{(wqFjP2f
z(9cPj4u*D1t%kRt_!R{k3hJ?4#k7Ih*lD+yR^oUrrTquc-t+?(t5(xoNtR8r&g9P2
z`PyJR3);dzMQop;k`1DJ^dr}=kGlb4g~d)XQu4&RjVFPY#d@`7E<gSu^Mq#A^OCG*
z0NwbpeWm6#zknOEN=N+0Bi~<7Gd875`LGiixN$z>^u;LoS&1K%0mgH+Pgxj*&)YEm
zewp=}QV2{4iZBtsp+^X$Bs|2!#icIYDDp+qD{EZwuJqQKe%&1(ZEbg}IJEB`2Cm5i
zg~^VLax|aFVU6{BQS-`lHuW*y;sL`XFjjE;M_aDW`~3PWM?pZC8?}4-I!aVL0<NtB
zvyOFpdwlw3cYvwk?eVvSju0tk-dFxX`C7dCnLg!mYG=N~dZS<#Et*t}=+3vZk=c=;
znIo0Do%r}m!Ux2jOy<=3G5z**6uVQ6%vQDpiXzA|uq%k)crOmJ`5~_LEAx?O_df>n
zf0*Wcfd9BBceid%)+v1!&5ya*Eh$wc7p5WP9ofNZ-_S1%M?yo!938Ygn~4p_K48H}
zJEpmK@TU785EHQkP>OO*pUaK1gb+hyR2sB-NR`s}sc5G5It{EYSQ?!y2W{arKd@Bz
zaxv=1yahZ9)LE0l=<20so0<_g+-T!0_t&j08_e`P0+88ik3$q)+6-_!wzd}l2Z3ra
zXd`||UF|d-B!}S^<OAK1!J~<H>Q{U?v<6(aC*5sanaWS1%UJ@C0}pKoXAzz-njc@)
zOI<Y{s;PpJLqtWBN0M~Ze|o%S_7-p!XUZM{6#1CzJTbE$*+*L1OH1Z`%B}eXbT%22
z0K@x;n(nR|1LKMSzX-4}Zs@Zs9clo61>VR@R%+_S5wp1=qKWJcFixDz#b^hT3O~|u
z&T6mMN&gnHs<#d!&wUVRzw5q8rTB$9Z~3cNTL~rkJev<x_;E-5G#L-za?pNT@U+{-
z{Gn31;a-z2)%+di+s?hBM$n)8I>k?&c9XHN=0(CQ0Smi2acNYPWWKsF(%|)Mff+<a
z1bT!*b^;`6#2>OPGxQey*@uO7e?Y-V0(~d#%fIT8>2sr};GVEmuuhx6UOmDrbGmaA
z9Ld0lwObOi1Pt=$BAQ5K**q$$sSYMBJKEcq3hyj(vJ}|AC?+PhM~@I-zE9iBwLwp4
z>MG_#Y*X6<1oFT^x>cE+7)7}Z<0wU^kG8}C4mIN&@nrLD^^GoV`*y}C03)QX($dhF
z(V+s()R8nTj0dvc)>B|UcLLDwY3H01uaDjAZV<(RCeL#{Z|>ab+R@Gq#>=3E<BM(x
zrJ`a2-)%LSkjb{Fw#SY6Dq!^5vjTB9*Zuh&aUgc+-19fqoVFh8z)3e4QRGfrKWEw6
zT3Sj!sy=Yr%3ijbkA=aZ;@J*7cYJNxpIc?Bj=D5VWU}!m&185F*$~SqzP<U3(94Xf
z;3bt|1-o=b9@WUBcdPfSXNmU-OfF7jT6ZVS9{Q_y3WL8>gtfM{X;k^`qxmT!l2Jr;
z^zvKvpSY$gxg4qE?nu{rU*jbr%AOEFhfp`A<(iW+Gfl5p{IWm%uz!c((nl7dRQurb
zg5>t%s}y8f#Y_sEY5V_nZhv(RAzDCP=k9|En>cVsxAV4|f|V=0x!w7zc6x-oCQXh7
zZ@_R-$E03+Lc8`|s0g#!o^;C$enD%|1BXAvzP`&|T~k-bZsU85iHX8qJij~s8n1Fj
z(})@>rzjQF+|e;TQ_dn1{?L&u^8E54kYqgEwODTY+{24Y_b8`jzeng}443)BabhH$
z2Pe?`vocJ0y;!U{#PGFS511aclTV+sy@sKprVh@?pq}8Z^5HP38E0T@P9;NybnZ3t
zb8FXRWDW`MxO<n%d4z9%K8ZfAIi(BBIAPsQ#m{e9lSNAEltF>c(^BUbj6FPw&YNx=
zCuO$_$cAYA#1&`0cz(JgIxX#9jbnU#Jf;g{BJ@`Q`y1~-;koyA+m#1po;ETnq!yIL
zOt@de1ZDCa=*j}AAJ!p}a?gXQXuX)MY?MRZ)`JJT<Md*CdV;QD)ZRb3VX?T&R%}Pd
zP+uuE<|l-h8YZ-Sag^pZ2Zzfs`k(iDIl3f_cjfTXEZo(~I8>bVM%-NgBT;>rU}IZb
zzbiW2PnB#mI&h4BYN3s>xN4e*$Li$?)w9RBt|%xF#yB#oU`Pxd($g1}_WzAGe}zO~
zs>j3%K;nm}@W;;ky-5K9l*J84A|n~q((GFjGz+!7#@kZ%T3Bgl6jl1_OzcGGC>1kI
za2(He_6_2etNeX$j{9+8ECjPBQ|1muhKABJXZ-lMdauQOBO_b$JXBnRoy`xoQHH&F
zv%s}=c5u+hNXdDUxue6MoE)d4m2tm2uUTAN6UZ95K&JI%rxMVfjZIDGs64v$T0$=}
zEm}wj3!n1zy_RD&o*y{7wrovcG2>3}){A|D4vs034$|VHl9FU4KL%K>gr~CWw7PL~
zyPafXlX?=jFuq#MPhhdW;!#miamLzWrccn&SlP0KD<m=T{ztdPiS>*~bPmbBF<kmQ
z+sP+1q#~OBr|#W8RlL`(T#f&AV&al&=Zneg)#}s2@i)%IrlgeXppy^yab&DTT3%kB
z{^lLa`NrXHzqOIZ@|G6DoBScKyO$nk9m&d4V##u+oKGAZ8)KBK%3wlA7oO(e;u4rH
zAf&&kt(5tJuVR*q{V?T{-4QE!>98Hw6{`7j&G*~Cb2gK?e*L;oK5L3CDjwthw@pn3
zfUZ(So#|HVWm77JPUJ>$0d8(?v*60Y&SvwOQabY;0-9~fOqkv1$k}Bs-Q>!bto*kw
zMMui9bTw=3v=vk1zq__HjYlv+Rwb04!3*x!qvGJ;7}@D^A+!w*M!Hmel>TA5d5S@8
zkS+a9qohRkj^^^=iHV<VukM}BU7nqFB415RNztmIoE%=B2KpJ?^X|M7ey_I(xSL$4
zob^i(r>iyAP79SCQVx@(w0=Nr^f9-XkoMA{p15j0w#dxP#O&;arRm3&m70RMHIP5x
zZHh9d`?-XCsnU&hJv(lCw;UQY2m9F1wXbWwMH6Qc+iWx}Rqn@~K&ihtH#eu3`YtQ0
zk{=B@(r9pXe}z$NPrXM-hh^q6DSJ**KUn3TaN`V^eIvZQe9@EHQne-i5tyo`EF*JG
zQ4z4U;?-n^!^{dP6aEnqNiG-qpWw>^>jo}@YierJ?npNUJP6zoN~@*!Ec5FLlMa@+
z#zlNE@TAi`CL%pOC|<wPhoe7I&hG4M4;mWAtM85-KW?#})VynD!3?`B$aFpN1dDEG
zg*V$934C=;$uZs!K($aVg|5#@ub!uD6pRhHmc7<k?xoQFGYS@LqshBSAE<CKv1h=E
znYcq!S64^GqAVk~Xr(pEJxPXwFl~(HwU^QyY%D5b*+{=DC`c<)FdVz)1Dw%P*FxQ_
z`IXkB$kpuofYE?4n{-6PJ<h_^Qb7ba5D$W>O-9zNFntFQ8A^zTfh{msAHZ~WAR_BF
zFnv7V3H67Pvp+_uHc*!`!wd(2!FQ11O9<A2p*Qk_-e~^QhEsf9Qb|!6E;es{sJN+e
zr%LN6h``QLfuStH+`)Ty1a6)}cWo2T)ZQ31A(DRj#>q`px=TAI=!5&Ex?qMwew$Yk
zA5HPSU+aDeE!76W#)vtTBj|L%LiIv@Q~D8E?7lIfLy6kRcnidXzFWgT8m2#7IsWzL
z|3I}Q!1xlIYR&(QpMsDTbYvUiK`?W2zM_+ge0u0OsW8ae*P}Ls5`k!-`-;?tu!Arm
z$v=$RFhU7Igl~u1s$wvA$H6$}5Df8MoIcO)V5%2{PWu;%6Ov<eg2wF|vrfod4nfI(
zoH)U#J!BUFLNR`0)c(A<Zn$v&)!;C4`;Vjcw_}|EGK|btFu^|r4hz#z{Y}X>j8H;|
z+}EQv#OeJTsg0~eU=Y@?NNv%ksKC{z)4yz9)(4=spXSMk=Sg3Qk}g{*yub0{ceVH#
zp1OeHx^m<Dq}e|p8-J}dfpP-T5fEfa{61;+i-jv+Gv@@MzhPVun9u<o&lbs}2YFcJ
z2f^q=hzEVgRR7;&s^3$=U$%-+PARgqAS-j2jMVp3@MeD6cT#;hVG9%AqJmw6LaA@d
z*dhAlaJ}3x2Z*c&zM|3mjC=nH5&89RCpKmRBRd^1&Mbs>Lw*oMY!BCD0O55&Z20RD
z9EL+-7zc43;1omns)&s96-TQ>sHA@$vb32E`}J?%XSaTd2cfQ3flz7?x&|r)f)Kf{
zM{Pt+o%e${g#5#pZ^TD^C)8h&qu&f5P{LvZTOXGG_W845(>eVNdo}5kqRtUF^q9Zw
z)qyoY8D5Vt_$G|G`g*Jg!De3)@lT1Hf8O0t-xbdnCPAT9h7ZA3Ur&3%_>d5;C`{0R
zj+uctJ%|UvxMbgyY;US)|9d9~G7-KawINOqLX-c~s14&%KoH^UQ5!;5kl6|*u7gk_
zFmWBkej#KVf~~$WQ~`kqHs43(e-r381eYUF5&(JsIB`O%w#@t9H}gHm`<DO^X7ngp
z0LOP<<WoHQNAw)MVqj#Q2I2I;1pmkn`o}5Hf1WRb?6?0LsSTk-zAev_Muk6z=%>ET
zhd`H((_WqTCGOzhcVqWq8@X}9`OAJ3pgX9on-;^Fx(tJBEuTrJSWTpCefYy)H2a@R
zK>)Nz)u5}`p!I<EPT#7|_Qz*7WgpP$d+s5a^Z0EYC8FB-mzEE2N~-_Zl<%1Bms~rj
z5&{^&0FnESe8+6RW43?P+yA&Lt#{l;iXk`d3xi;D7;g%~J3&V*As&QQhy#I(Au4m2
z!ixW2?;d~o{hIjSI?+Dk2!Bn{M8w9yEVX}{RDnw>QISrV5E{a>MoAYVD$zV}H3me5
z%2!mQpW*F4A|@LyFT{rbg%||jPzc6>vAz&S9FcL*L#~j;4u<<6tnWQSoaSWb<Ak@9
z+xJ!<E>3QhR`}V4m#R!Wq{mikX2+4YGFy7Lt0(><ZX<P|Q140s_1g#Pxl}>d?lQ{0
z$awiE=-9EKM^f|wL2C2?R3c0=(IQ74`pxe3i`!F*;V|twsvR(CfgLhr=Ut6fH?huH
zBthWh@1;$flr}-|AHwl|_g24jyuaqRAmjyEFd@>#L^c0^YHL@RPzi#|kJ6{Ece%UN
zZaz2vZ7W7?t#(xSYV}c%K~kfuSFJZCUeMhPdu=VyVL{)rIkF+!DgHb`j5a)|tSsH-
zkKyg#mIpiNF^*J+LR=92N7UwkLHKl=9{||uzb_dKsTyIp6T&oqJ!(Vfw0|MBzk7UN
znt1zdSsoaF1zF_2BDEo81xB|2W-9&`&Os;<n6MgRzYt+Hgc5;pdmtW!Rx}53ddO@A
z0|G#VjWDPW#C~C@jm%as7{E6~ZQwk?s0d_{gHaI>L_mjWARYw6G%&IaLCdelG!Uov
zkD~T>*Z#}A1CzCZQE3qH9!e?#f;(XvV32hYj9r0I5fHKsqaq+g?#pEU?~lJCwP7?l
zgjN2E)J8tNe<QUa*a||nzaq6EWCcdH|6$aIP$Fo#_Pe5|d{3S<+r08mwoUxfsk>Qk
zb4R>GW)y>~%2u{-TF7@LsiR;jxY^+y$b^UdI6sf0;2<F(*_2S{8jT6L&odDplt1m}
zzJ}!h7q}sQpE~M$`u5Kilz;pLm4*9SX8Jplw`r0CLN6n$Fc_5pqgjxx@K-bd5LS3o
zJ@MauSLD-!co2+h3}LWfWDa@1Fw}-%tACieSKab`qQ{ps2~=Sa1oNSyBk}^=A#z$U
zOpMBQ&>F~qs2lS}K)VCtx6Xb=4T`9dR3*h=wgaIck>498%!UXQQ9<HV9lw{9ec8@@
zQ|Lj{TN)Tu@%8K-vM^wvWE^=N2xamu3Dw56xe$sUQ7NHzVd2*@cxYLeZ%P$Iu;;f$
zZHUwRru+$r{eD~cHcGOMsHtNge7^$hONRtySw!C-w2~GPk^HiS2g`5N*X=wXMSSoY
zzne^eR)f5J&R>fj{=A8ZWZy%W4Zn#d<N0%nO(X9OM6dUqt7ydA_;^g;CLJLkX~C+g
z9Sq-^;|!tvHa&pv(8FfDgxXMDh;ZSXic28a>gxebFvbwVM*UeyeC88s)P^`c2x^A}
zw`D9+>SiyGxY9UiRGRg4+gV%U)^8R&=jg7?*$r0uCRr@k&5t%ubmg>#K^zi<Kg*-c
zz))sA=+6}<8R+BPBFMqPL1UP1IDFD#Y21ORvb4OdC%AW0H3`lY%FC2(-Alfr-(N<&
zGC%4o6+t4fwy@ZROK{%2FnU8+A!Nx_rsaB;S>b&1qA2~Wx+XzK;+@CUZkv`|L$6H2
zaJ-$%Xyoyky+H~o<vZ|6U%Ih%m(p2t-KpCqWY^gm*_;%2?YNCZNNAx(b%4B#`((oA
z*Sl|=7zFzsmA%oHV$FBECf9XLs8xtS99t}y?75`d->+l(@!@JO?$XnM)Ly4<^F_?!
zPJ=I0aekU>3pK8YAw*Je{jFv-MPX_F(|s)pDOO9mxcO><)xaykB2Q}Ut=3lt266R+
z6<-<|@SEv@Lj4#9vzMr1OeU2JaX$7Qj4DeDf_*iwS?DAK4xOcoPtavN-ac4AfXI9>
zN9m!kx-dS}95cyUbcDzuYo^R0bm<P82H*a(bo%<VOSx%ZXutn7cIX9!xQX^R#i|N9
zY_lU_qoAKBILu1LYa{VaKA}vr3Z3(^(>EYK5Qd*+rj%ign~=g)jJFA9_qkqt)LvlJ
zJkO(dBSkU*6`pr*n3+j79QRT7<9v5LDf79Y>2%NX5$pbf3UBAtDRd4?bbP=U@8Eg-
zW@Yor*Fq|u({#i{6Ip9Bf!ovU=jP_J%u`m^P<qnc@z!RsbNGO_?QGW8@F**GE;G^%
zRQmSs{}O8bKNB0*hF52{IT$=?XlU@~6%;S2ww}zIKY|GYtSv0O9KChZwDd-H6ao%7
zO5eRcEUD{WLb78|iP*<0D|5r-XX%>W$@iZ{=Oy-c*1S&fyiKA}Wx@CGyf^o3$`OoO
z&pGp?-(>pJnfx)I`>Q(%W7~dAG0;>#PQ+MyQoD5yvz5TzD}|;&T+wn~HhS~4J@om+
zF)8!qoGYmbmk!Iz%d+Rcx;P<cQ)3^cAZW<_8RP$FHxNp`cWRfs@VzLrhpY1gOLuWu
z6$NPzEfGt=q&`X%XpB}0EL4lze=vW2(OG<{TchIs(E{;H-X}{BcB7&OX(AdU<$k1p
zl4cMnw$!a`X=M7;x1=h&5~bxm@>PDAUKwTFZ<PvU?O1{M+=!7QV;(BXPfP~bTGEh1
zoI0&9o>Lo-8q3%0yBT-w%zFxATXb}uc46$z;$-#1_W+TFux0i+I7>GNbT-dmZfdG<
z7CT}mWS4!Z=nUPZRKWuAcgk6LM(EJhBP*rTjH!fWZmfk`6wjZv*X3uf>A8AqJtcG)
z{}v0E$(a`_B^6_VA&;3APtg^K_w_wDL2LENxZ*e4mu-&-VI=C#$VFQmDG=X7#@T?{
z;jrEm!CSWFoj4Q_@;G~Sq75jvsgGx(ccY@0!^iLs-<N)&lDgPw?rK=0)5d9+YF|--
zMQM5ZeO-Y$Crel^`OKYm-Rl#G6MpJDR6m$rHn;Df2<xUe$|K*GQ{>I=yWYJKYkEtV
zBvQGlAj+6>v1T497jr%8k<P4?{^PRkc}jnu%-%O)VR+<R)V*ckiNd<i5EzPzjxKCt
zk#|?<)DXU^7Q%5?R4PsOax7inq~-k{dy-P#Ia#?|KawxTmoq2P6>hp+pmHIc%y*y`
ztL#ISzbPR3M0=R4$&i4h8F(EgrpkZ=D3CuET+_;M6Zpx8fF(G7pU2d0qHXcX3nAhI
zV{vq{S1@YbdBvQ%R!pR}+D(}&BWopo8YLQ)D%cq*bqThi!`4grZd3+Nqs@ZEhCcYD
z4fYjxsseGhYwjdtEG57<=rm<&?f?Ru8(<g8)-dic>Yz68kyGhl8e+#Im>~16za*|h
zXy}8RS4j~WU{^>y#XH1KX?FXPx<mq|<NBE}ZmlLE7w$?*+|v0oVas{DC4~prb8zdq
zl|DwJo?y|(oLUm6x&^@&ju0erXa$gZ8(2&i%AW~q$y}BPKOC+v-tMm<RwfMopLg=`
z;jItB^N*;MmsX6Cd)FW4Um?Pr?V)vI;hJ9R<#t)7Q4AKDe^qa>(9#9ABPp<qcP1>&
zY(B5VZ(t{#-_JdZ)VKltGI|_|w}D~;E!Pk`8Ak-AcfB9K`Q4I~VJZWk-XeSgp2~|X
zj$ohG%jIN`BxYAeW1d&sSr{`Ft+}(8q)zZ;Sj+4D?cmjn$Hu{~G%88){09>6EP`KX
zHAj;e`WOsKI8@vz3fuyI9O|WU9&Chfy5Lp4Vh(n>gr(4W(wt6Y;IQD*jaxOL&vD+4
z6?d5BXI5X;Yq!LUs!JG*G{l_=^Y(1wJQ?Qg&;QGhTfqW!;-r-zgTx?2#+6O@h?t1U
z$2U_*Q)7XEdoF-Ku+1&v7?<jE)}li<fBjEHo*v<Tt}EOfHlP0KdHAk;us~S?ht^fw
zkGAdY+X1TspHXOUmgQ>kLP+Y)|NM00_imZpv>SN#iQm?H&U%mo@On6(LfVY&&zZ4?
z!))N!FOG289X2a=(t-YU_FCG{nNw@C)wGi}`A_iix#uQ2z~*~sf8YA0nJ2z4fY|2!
z_f^1OWEy~b8ZrY;rycyd$KfK7A;1bG`a>)^?W&Rt;h3KE(dg;+Tc1{u_RuVIr(Xy;
ziU|s8FNm??$emRE^m;EN_)NMiQ+}2FB=yoDu;xs8HQi**Vmlj8U(M>`WIUaR3>T~M
z*zIbi-%mi+)RRR8BPrHH%TOkM$#>yZSN<RWI8fnz6X3rx7~ld5{>bv~THJ9{yX?tq
z)+2Vz$J;eGcB$5uoD*O%BTA9=X`?qI1`==A_I!N*__xghMFAY7*|ra#-v7;AiS?EF
zoRtwq(C{6iuhpH$edNAsOS~X$y?8sJ{%v!qQe(pD088UeQM&cqZ7)BO{F1D1y3uGh
z4!atveAQLmF?|d})mi3IF2UV@-inUw(}Bn&g<lCSYSYcp*?56~O8-@vpOi}~m0n~D
zIg_yQ45fN0o7wY+0~d2TS=&^1X{wDmi#<y^tf4n2QDTchtuS$8ltRk&q`MqiP1?0V
z!q=~V6qH}4T^JX9IQxItdk?56lV)$+fB_63C@4r$1Q7`mB}YM2K$M_FC1)gOMo|eO
zK@pIw0wP%?XAuw)1j&*!NX~hf`RW089d_UMz5AW>zvus*d%@YWFw;+Wb#--xUsd<i
z3a83Ov^6EH56m|hM><XN1}<MI|2pdxZrA0IX<Q+fu9B>h*@s=TFSybDcKO&-&5Grw
zr<x7Rfr3=2jl~hmQDs&&ybPE)c2}-`lvqDXwrcQ7Y~WnQ%b=k8q}u@vvWfDAj+ubx
zWSKh2TOL-ZL+#K?qn8OS(=<#}PVIUn`fzcgouMyIDNj=~KAw`CM`*6`X21UJ^`{)<
zw&Ooo9%Ltc<)GxZHEDcT7w<7#%~AVQ_+{W6>&Mx6agVG+6x^QJk|g>5#`IdV@X)q2
z0o2Z*_6*||7K{`-MswUSE}}WZFyPktGsJ$*6&ms288#*=Pkjl{m>c@$m&i~!buTfm
z$F12TYCfSniby5P#Nwnzk~tT!3o|60s+g!0h|vn1Z7^s`&`nl}19sR>s3fam=j|O5
zS5q23=vFm=7yBik>oJ+e@mT$Z=&1Qn9&?rSA;TGpD}|0W?HRdj?6sp60wY|eZJe%5
zyt6T6#knW#lt)#CMj~vK`j;PyqKzUgBp)88Y@>v?`DK<SJE`q^9;oNr23p*fe$F-e
zA>YeB%K5`eO1`A^n5}e1%e(e0Q`KVE_0y~d-$J>(N68l+lSOJ~n?;ne8pK6d$)fi#
zhZ;8yN1<FJ_-qarzlw03%pI1r@mY-%7F?PRtCH(8XBm!YIrrSSamGut<g_`<jIm?M
z9x@3VpS(-ndNaa-9-7}r)vW9G$g4;DVr>OyhlsH*6xd{nO0LmcN&T(*Q8okLV(*sj
zooLVY86da0a^*_@XCJ>Lc^;$2ciI)c&(xwPf265LOX`O^8ZFrSP(+11ucw9KpuPEz
z_jl2RdOqY_eb@z3)hB*m$Zhg%CyE^EjoUH|(L6WkTHf{B`8zIPu+prq`tfCQj=HOR
zj-#lSVLf!Kf)J8~JIkYH`s5sEh6|T(YAUq6Q_VDjrAjWnI+NL7I}OA8lA-4K&@oE^
z%K;i_pYzB}d457PPwh|-Ql}BdZF*`h$+bE_BY<8X-ZTFG$~E5>c~7HH&2?oE=FnN>
zl8<O0ULL2#Y9gKsD9rc(<?3X<)E>&EP~(&=21OyWQ!12C`pVm=%b669!+%8OOg1!6
zlt*8s%AK^+jJlU-3IDEqKPowMWx}s7kZ(X~P)YGM<x}_-nkx3dpSri9N<u)!VP>gs
z%J6H$3o=x80$pv32D)%LUoltFVaD(JB{N0jUy<C6GsSYVnW}+w(o#G?4bWuVOafZ|
zo&4#CzU{>*;R7yIRO8Z@?Bt$=O-GIiwj_UTVdzNcNEnxH4VYcP&pzjLV=>0r<x}$4
z56|i3ZX$y?h0bygmUky7u=}w_JI>ev!Lxm}GC44IuHuRiK2NXG8R;hqSA$s&TaXm5
zb((`_e|Fjfco3wzSM}%I^c}f#c}_Pv9xbU&<pPmHy~w2GzDB-CmCI3V4fwz?I7VWo
z7!*k75IWm8Ro>n3B94(2W2PAB;C!OVyI3_)HINKt1|%W>mZWolO6eOhDu@;1x!|(e
z2s~eGH$TKT3A1FLcd&oYaI901(d?5`%92|4xlJa2V1N;kss1@pa4!GC)UN_XPe);*
zFp9S@%xVz%w?^0T2>8i(Bl&jvxs4#2(ZPd?)8Ni`3oFad${&V3el93VM9r%)Dt`^M
zOhczx#v;F#!~f1B@5G{aS*O7}S7xiqSV=b8Y&L~vWK~GBjR&DMSykY3x)II7&A_6s
z%mUWUhiCBs-(+=2v5g^ojo{YTjT7w!E3Dv<mU-T-<hP#cDs=PBH@Tg1_4Hp`KL2`4
zqbrDr=NSO=;91s1JAJokN!1*yhiaCsZDZJ#z>Nm*elrGy#vQp<-!0~5vvN2q4Y77p
z#!+E067x3ku6tHG=oA}I`)i6m7uY2|yPr<lm0x<+ZOAHM;K=10$DHwIWoV_T9?(fY
zn;1<@C73;sc5A$+#HI40fh&fLb|3Tkj<GztVbga<Z~lx9c;QKKb%E)N+tZjw3Qu?k
ztl*8qFGiKW@1%~1k+A-mkHs*~=1M2LU^Uq1V)mR^UKAO>-WOWyfVG1y*!&JTEpbj9
z!dI&+)8jwVm{08cR@F{yT{f0F)&FJ_83VAKwh}5gTLYdC?C}W$sOEAXJzXVvCLMAK
z%bA$)@EKaOzK4X|bp5Xo56!yFY4XIxKYX|~UEe*&3#>joJVQ-kT^1L)S_ol341r`G
z9??17c&;ON<vAHA0@?$L?vIR$#1$|6m6BDc!0%@c4S4#H<<_+I_aoiTbBcgRY@kU1
z>PdrRfl$tBfCMYZ^DyHxx8W;a8EH+s3NC<LhF@eku`hyM_r~`l*p*s2wWFK40c3_S
zUu5)xR~86Bh|mI0B8&t6!D<T%3SPdTqqivcSybxjaizZA)y2XRKyr|K8<5eS3o=T7
zU^r8GfP8~&;Myj~&q!angZO`zBxIV-7I%BQ9#PM@Q>3uY-18Hd^{@JM6ZC(c`bW^W
zfy4r+^)VE~;=Mh4oa*cAuLv~f*_h{?L<WbG6)+x*j&Q(1v>zb~nZpN3W@O!@@yDb6
zg!WsITn!cq_)%)wmhrl5x6MfX;Kv7@Ec+00H?FCBJ}xbd%tGV)=Ja7sz<B6G`i55i
z?AJ(9Gp*Ty@G>eU9=p3<`|{>$p0WJ;8ewF(fB*0&3ZBU>8lDSxM3l7r0s}aT%62EK
zuZ14?M^m2$Pc)QRAPR+9!lN2%Gfri6_3SEiI(BR@pXeV=-F&KR0z3fVf4ZLFvv<2q
z+p}`Z#H&N*ViG^%ik%SOzhR#AD1w*ddKI~tUv89#urqJDcB*5gPtZxo4j<vv1{3^s
z<i1KV{2S?$o@iUR?da%86!D|LL0ioCdEkO~pD-HwT>;P?rw0(;Eh1Pe^QRCQHN2I<
zw<Pfs;>#QtMpXqZJ;4q{3OWr8qk2n+aitJCfVG%)`@FGlEl7lT{@exeE;E-jhHE0O
zT9XJmT6brgOPxSr%DnoaE8Qy0&w&5f&*oOz*btTmUtdUzLV^OG{W=?vZMM$F%xqg-
zdy<VN@!i?(iW*$Tf=F&rC7Uv}A(w;SeJR`XZ1T^0$|k8UNshRh$jiIhE}B-LGt4^k
zAoDg8Vzq(R|I$eyoLdBZQ#EWzXl|xiCq+3{%k#8Vlj4ae=h^HV=xWFe2@XjC#%nXC
zv1Is9yW-8h*Bkz-*s|{lQ;I!yT^JP?wFprY8J5>-SZL5{=sz(Vhe-sC4EoMQ5trrU
zP=kSF#oq`S&vz_^bzi$GBO^G(Q2lboQ|=|M>VievqCakJY9B76s4%+tH;@$#3*ewV
zV{ewnIJeNEeDv60@Nz0O-~3=(W-35p4HvLVak3GSD9mkYR@^NOyx*cw0Yll;<w!m4
zv97DEf-RZGEkuGdb5Y%F4zpzrb9#j{pP9nA3-k59CFbTHpmtquIdMXn6zjoSjMk^%
zE*QKy_)l2S2AU`S`TR^p*cy_8&XT@*HH=-i9-`nif&VD<8-@zhb{1%!1ciz2;<XU{
zp+FV64x1KA@YfeFSpE~4*a+}K+H+<iX8NZl>SL>lGl|n~nE(=LH6*YC3FW!m4JEl@
zCt%<HJJ~||u`b9k@Af@J(WO%4Jd$)<?pkCxhvmR~fE}|t*zK!d*4>N~I*t|AbR5Z5
z&lj2#g^o9^CgQf8L6pL#KJIvgJGTuh^QZFo>(>9HC465w7^lr1e`)wXyWd<$<o31=
z3k1q&W3xds#BgrTj68F*r;XZa#Llg#_-@V>T-n^szBeD|GNO*yu<?IByxAR8cW&5H
zXayS%Q2@Wl*)W>_Pfu`@6G$={VfKdmgJ=unk4>f`TLZZ9o5S!avM~_fVUsJH+wwPu
zH+DE8n*VJ17ITry)&Jn%rg!tZ^I!J$m$U%i5qe-mpa=dyY`P>Of3SiZ<v_@gMM2t|
zG%v<%HvGf?;4=E<DbbmIl496=9GM%4#O5C`EFv(7KZHzYQzW-69}K(U^B`LnS26?{
z|7ukJ&~v|AB6EPwkm3Hbd?MRnqlN_`;+KmBk8E-~b@0cgzS<ZPru&C=^+86032s!v
z{(IX0x%>j|_Fq^Pg!7vt{m~LJWMwkT9e<4f%e?}|0*8@F0lPOuZdrJ7vXuXcqaVez
zu;hOHOQ&9+{&Gwrw({7K7o?5>#5<p95s4Cw5D~?mt&iF7MniKZW53wZScWohyo*x>
z9&<wjojM*b$IWchYa@p{tWwEr>GRjttIdV_b#`{}?94IWB_eg`=ewOdmSo`fd);_F
znpGAu1ox2FS1!hnE+qHY=r<dFmmhn#lagykPZ9o-%#O*h9q@Kf=uvn<6L@|Hbnsn%
z_hx6vAJE%<;~Vh$qj3>F4E8E|FY;uL`KOJ(ceKQ)1?+-X_^aGa0*de{xuB7CBcIU@
z-0^{?k3bQA=;mlZ0E`B0kfGrZgmxeW287SIOl|Xz>D0j-`aK)3L7=(_ANt;$#Lss&
zjrV=Yp@&DCU1Ks&w!{Eo$P1TLO0Lajcz;ZIqrc6r5E85#cWl!9Yr;A7$Vh+kXEJPq
zU=gY0zp<de8X=s7^H0WZFkr`T#=*=HYBsryFxHnGI{U@t|BE$48}p_}@7#QCEYYti
z1KqH+{$HuEPBrfLZyjT|@uY#l4rSsw4%qD0>b)Cr>PvCZ?lOt;;#%rr!QJ=V_Ffci
zkl!u)eaBTAN-iQ2zw-}vZ2$Z(;=pCAV3vsZcV`#)HYSa1Yl9b8u42)>9)+4H^&GA8
zdWSioaCY6k#`u5(1VhQ{`C1je%oTp%1x`mQ<3Dc=U?6l|o9KHj?qoOFk>xPAs+cH5
z_}QU7i<OXa74zxwO&)U%Cv;AAA|2ZUg(!m~5}`S+>o=q;2WMy7H3(LAd(F%nu9Q4<
z>T*uk44{{>v}&iT9d1=A?{ZskkH@3r+C_3LgK{qy*X%yZ%sP<O9SfQkl>Ce0q9q+&
zg_^Zrz8Da%8gvz`yuUj|N}6(5_F$p4KOL1|9_r`!t`Ls6rHOX_<Mq>whd&GO@GN!O
zb>la;FHLoOE*0(J=hnNX@jfz&Lw_hPiZn@OVP)vsnBDjf^6;M#7ian(ABcP6U8`Q`
zNbC86f+wY;z+qytQy6%&x;E90MO&r5d6V?!4d*f|cC9-CCQ~9@?2=Entlg5Lex5~0
zRl)j{y?d$0(nM!|f0%g@L8R@%=-}0$^9O5a-pWO>n|1tzY6P;EcivDmNO;rKf9osX
z6G%N$U0vPL!QGi}H`XxhW7l=!WiSr}+4$*t-<FsAa{(+QAlB~xoY|h`l55<uxO<;1
zwV>nlK;>!C_johJkFF7MWt*uJ5;NXf!NZ<-Bu8`|%ZjRr5Jrv8k3bzOHD;#51uF5-
zlljT2StjSstC0u{f6$GBk^<q&1vH-Y*N*dp8(Z>lU*U>gu|}cBAxzxI7s8zECExeN
zT}~~iFB~1_eJN!lquqJ8axb5eh|~n$A?rqZ6+F??De-&WqD7@lliz>{of2@Ey|XlF
z?lu=sZ{vvP(Vwo5T4Wd7&1W}R_6T1a!IZNZYpP9F$NL%!oo4bbav3#x(jS_?m1UCl
z5s#$vB6RsIa&|6`Te`X8{?2vi`gL87OTJF!^T)ZlxlMAxja)vK9%m@POi^Z%aDJB$
zGdP0hQ}mYfB_bx7PrwemGs0rjk1V}wo|x%!5`XeU)?Sdd1parJl8;WQ^Ad0rkOX4C
zRH-OFn;YWC*fm{qW$(Conc<NzyIoeb8m^J;T%*T(i>5sjE{rnz`ogQE^|}vrXArY$
zf-!mb;j7#mW`Mz97YSvi7pbWKVXjQz)z5dyieNxKavVFROUq!#%bWbLArXx>tO{lg
z;k8=zr=7B%SiJ$Ednt<Y``5s#NFnB>HnCm6ub#QuU_2|bxxLYv#X`mFwrO<=()o6%
z(`+RYar20TbiBDl%`mYi{%&d4v6r(RFU&64PP9=uKP`=8<l#|PZ+K~Y>%#-Ypzw>5
zy!`o2CS$+bd_3-|>_V+CsQQehfQ<PGg)R(7DI23z&zuQ{+GApLh`;vCmuICQ-m?Va
zYMpsD>l4}CPc*&i3LNGRQ7Bd1lV7i7n4?JW=XFK<e|+?Vz0NBfcn<TDZaZ*;L5}JO
z(4=5JEL-_O*PGwkKr_s6v@t$9?`_@qd!5~gFJS9N&sMjwP=<;(4_sS$=-Z}pN$SuJ
zG{(sSgff9z#DfquHQF>+3$PY?W(;6yvBPDBj)f{3u2o|PEr_cPL06Gn*F``(rL_X<
zkvHz=bXe5+PP5TwU9<j#h~b^iyueBaE~D`u^1E(g{8fOx!j?MNmlqs*$;eJob><f*
zr5K*%Ki|6HFGak<zqTp9c}<Pb{#A6AQl0#6k<fgX{20~jlqPc(v~u^erGd(ax6}0=
zo%HN>GQW4}Qiw8z)+Ef0mzNjcm;;fbfKH8E%iADC01p$tBvPg4y8u$u6Y}X3m$W_C
zIVPsDrEYH!Vp_$HJR8A_nImnPS1I3)WVSfQ8<Us(U6$N&Wy!zh>sPbI(FLQjBk6_>
zZ=DvMRXy}oGYmw(Jh&)~3MD(KHMTq>-rCygWx~PE&VKmh4|S}dX0fXx(4(Ak6#tqh
zGfBjK5Yg*~b3=!u?L`w_-AMsar7QLzoSN*+Kh0_`p{LitCwCKzacSa;FDiWwoBF<Q
z8-D0V2`SSzmxyBYWtiCp3HgkSkAZs;lm1%VLqS+=@4f~;;oj-qsmCWhLDhE|AHKk8
z-sO?WYd8wJ%ZsGqTLk}_H>*(l2i^W4<}lzByM9e|nzoe9&+nM-NViO*rqLEn@6p`h
zL$}VT=iIw;M=YArqd^}AR3K1due9Yl2O7KNVHmQK85qpk;$&wxUYszqO2+<`Wxyni
z6O~y}lT#f|YZKXxf6%RAwPgb&;cS5WoM!&b%EH%eCsyl0F^{6dkqrUKtv8g9VfT*-
zV>{-)iYGlK=Q`DvqpPRqdy;p2EcJkSXCBq;LAj`)C?OXU44QYMEi>6At}a=%1^Cyj
zUz=@?JvY$u7!trq4s%1IF9+afXei6PJLP7aZv?*`0fysynzjdIA@^a~zGs<Edb17{
zOp7J<jI;CeCvs;GU}vl3c9CdRNmMw&+#A`gj8^trGJXl6F}1W*!K!j$$eUP;)fiv}
zDn~%<N~CWtvx*$1H$X36(62d}idOAqkP`PeEZ-6{zadItS8!R7k&MfNd4(I6xw=*4
zW?Ji*HOd9&A(^n|$LG61tBsd?+#2azTOi+OTK%%1+ckeVHF6ZVpRPBkucOoWp>kz)
z?9=M_3M_&I+VwrmT}k1Ekc)~2Hh8cudTnqpd$MC?9dh0p`L;9lu^c7wKYL5p7ROg?
zhigo2E|L);rtTh0P~t}(*DRTXuD-s{NnXgA`xUPy-0LoiV30-IW6<$;n63Li2Zr&a
z?%8=mWmVZ>WrF5Hq=3Wq+GOlZMe32USO7T%T-E3|*wLiy^vL-K<d%8uCdcojoW>}z
zwa}uB=YKvM4Q<>sQaADWjDBnBnk_pka_`}j*Vl<w>aVJgqI^liZ)6(IX=Oy;^Co3>
zn-fhHE-QnzQov%q1u*zEuHef>2{q?h#kHvgj5a0eyDXwVK2SfB&3)k71L;xrI*SO<
zheu8oNfv<WfQ@&?WDEq<?|2Y3ssmX(>9NA4y)DZZ#=gH^2h9(gx_LQU!zo$$OZ<1h
zO`mFvPJ*T!JM^Cjc0WwrrQA<z*_2SLJ&R7)ANEM<Zcb6}Wxl`4#9B;wLCv&%h?FcU
zakv3iAuXue!*%=zp-tD<XQ$T^g_<@j=|?)7ke|zlbr>Z$(4w&Z3i8Qk?uM~2<%a|W
z+>A;U0~_jrS5+HDIx{lbnqHA*sww1XEj(a;X)U6%mI_SQ1xLm_&=E;I*UtQW_gjaC
zw&3bea)KI{1nQHZkw7d2>1$C#)ieIl_rN{To~1<j4}ohXc~JTNN8d^g%_3*|Hzx5j
z&w_ahYQmOB;vOnSLALwIuA6tB=UuZ#BZ}Q-tT{PpRh0Q0-?R5Jthc7+n`HF$xYrk2
zPCZ~mUDnfdp1IwW_;KIi@sw{;iE>dqVnR{C9g4!Eh!8YWyd}17l>LrJ&u2uE9PQ4T
zin}E^S$)q@LzRhFuTq}f6saYRko8X|bG*g|LgDdbJ|**l8I^-k0>|rzH3A5NCX2Be
zCj)3!dmb7BdG!WCr1X4BpzB{r9?=xlP+8U2%2I_L2IfMMN6=WQ2+^~-Q^nX6(A*FA
z6}$A6`DJ2N!3jIOa<3*zxZ-DQAMDxpU2K8B4J0MkJ6!s#Ids18b5WfLW>!5}ntWb$
z(eR7B{pY7kg-5Q-9qGp8-Yp&dA%q19igdbdG}bJ0fw%qK*_RA#d+2?JLSK0KHi7f-
zPQIzTx*>1j0DilfMwe$o!x58JaApzrW<}81eyJN12!~4^IbRfB&-nNcXv(l0mX!uk
zFcx4ZJr?n_*Zs;B7LSvU`RyhL3hz6vf~%@nj7mR@9qO9v>suNhAD=e9;Z4>uV}1UO
z|4N^CE#BLYF?DJI>~$AHVAIr{SI;{uTU^CimZmK%ajjyssS5G<(g=#_7gmO^KGpa-
z@z<TvEdMZ6<<3`X>&E7Dd`5G7MDkiQ4Ap%+%Jsg5Dk{AAY^w@8!TqFXs)pdy#QoLd
z{I&{%B8SL1*{H%^^4VA}bH!H%R8+*2bmdqie?0n<Gt~G7Js&+RBu%KCF!MXKv+s$3
z;QP!%4zsV6QzI3J-x#+9Lr}pA``AYj?4-C*h)urfSC!bicHnhwL@MBiKJfBwlNfs^
z`rBM)>g=ZJJV?&@{<C(Uaq7a8m8>o&^UGi4FWw@w=YJ)X&~>NCbv?n}<juRE5<t|>
z-Kz*uY1-a@A<E6oAuy`ePdGB{KC|nJu1-KM7#_qhX0^n?Ha=eXqW$lJz86Q?JwJ3m
z!*Xeox~^!}5S-Zy-QZ3h5cB<HHB>$CE8?P1D7x$??CR5?)Rt*X!Ttrxf;^=kx=i;_
zJz!)r<ZOw#X1&e;`_OY`Zdk!ps2I&Y`Ay??>o*8i6g}Rcuth-2_Z%c_@Ru73Vy5uX
zF#0dnta2}`SusjEL%T1OOS8<`%4unGKQBS?QrDfS{f(`q9(GJ|BZ10!73Sj6Qho<d
z3_N}Ml+Srhe?e?$IEoO_xzV}wx^KpW>rU+#s}A8vpPh3U_=2%FH|r`OJnatgYBa*1
znwFTN^+n_8U={HCK01~vxc%wa*fhTS9WbbT*y(R3X2u|J?~Zac&(hjdIjvN0ih91i
z)T^2l4M9y_83+k{yWN$NZjocSeiBl5C8KDq7RPHh)If@K$j?<d(5XZaSAg_r6uUCk
zFr^#_2?<eTIK^uI2nnMccki7iT+x|qolCB}noJpFa;Na_IZ=qAWbV9)=C0#2y8hcO
zW^UbanMd5cfMnw3Er=wkRI4iuDNVg)!nl*$Q+MpzC5*{QwG4H`QWYOE?JnZg;f)9k
zJQg7KHfK5M56Q{HAmUv1^u(01>qVw>SNijk5yy?vqq_Pn6zULxn&&L0x9Rz4_79ok
z>jj~_cLnLbL*&Ol_~%>3ehJ%UbMuC!11OzZ0cU$wR|*2z9H7Cpy(3R>qAnV%NlSvh
z_)CpfZ-~!QchvlNKx^#P;T*G{-(MFVNoTVfvIZS~PiFLI-EK+NOeBi=FZL7pxIYDc
zlz;<~pO0)9_nCpS1t=`b)B(C&&+dyyW_~t}HNo)Ud?X0fPPCu-HNFWlevZwU($L@`
zLg(2^^s;H}n>QIwOTmeU#j1jsN2AOU8Qz~Fy^fx{YaDpAZ6n6TvBq0<4kEwtt3H)S
z{GiIN+%Ot^1`yv=m4)1;J<mJ9OL}x|<^?0k*kT6M3dgSsfP(~D4>7(!WI<4L-a9VL
zdWzoqFhH^~`*Zf;LN3)RXC7}vac%!mOu5Z9s3#15<S&g`wEc8Hlql}|X=UYRA*=rY
zx2XyldpTk&+Z({Oj~<;<P<VmA>7S$_{>h=qnuzC6IH4HfmSt*HRb`F@V#YU0`q3*l
zo6j+3Z210LtmR*nh;{nCUbKN!sp!MKpADmXi}x7({7&0NcG(iaCPMlUM`^+kK$IIR
zba1_ACe~XnQXt1}>T_p4YX7y%RP=mVh_x({iQwbaBH%VtV*-tuA}#22lq@P!N%}SH
zidWIx|IQc&ne2!O!zcMH!A}m(`FU9iL5EpTrpw9Md+jx8Dt#z;_}catFKfw0pmvrk
zYaUY)lC!l{$`D>$wrI_@Vo*w6b^h_*E+4GC$%WIWqi4-+AQbrc02{$+=zK3IO}FZ8
zeZBA=dd~eMl<g2OeScpQ&P#P+m#9wpXs{XnwH5KxIrKIQqbr=L#y5;N;0GBT7f9-m
zkrAxrcX`#spYP64>==0qda=&ey=3HCIJe_OTRHeY=ps`B&p_D-K1qt+_?rwZD~13U
z$*%}NC1?U>Uy5e&e969sQ$<dk=x2d}T#oZ!*4OO7o)KKTO!teX`w-JkL4f6X<VU`p
z2k4a?dc~FHiEM8}jh2|^zXbrk<c+`Gva8%X>SZc?Dc?{Y(ys}pit`By?mo;*U^j*N
z07Bauh0-{5qF3Bwv3VsVn0YN@)}bxKA#D_EGuhz<ahoFM(-S@cH->OQe(Y$|fn6lT
zr4kjk;L&rwN-bu2x{O{LYmow*tQPQOjEm7YFbDBxDKi1Y`}d|1B1VC>y6Jwu;m_vz
z5ZQTVb>6w#<MKc!OS_`kV6Hv>X2q&Ai2)IRpj;D_4eDdxXUx|E<G8%)PV@6S0Tc#g
zeEdbmi~9Ovzz-#yF>>L{^8a^~BnyC(t0S5TsKT---@t(vaH^n0b$6vnndj@eA-iv(
ztfO<)h;nyM0LWn<R;M=7ygk#novzsr0GUFoIEg!AY9T+5BYS%<r<1K~nTPPL10XZ7
zyDSRnb(4u({()@wIMWTBRjW1QRfAg9H>Nxs91TE2I7LW^Of>Ur*sd+Zm4LNqE|A?G
z81yk9&XzJWn*{I_kG{;$%eNOiJ%h8uwNXZG^H^1Y0Mm1p!#5=4Sc~zE=5rQ*i-`WN
z(%`^=Gx(S5tBN!2^NFd8`&;LurIrf${lncB95O1`$J1B4W>2>4efzlrf_fO&>({tM
z5FvvAF!X5)o3<U@)M;HUF-eMjNU&ht_#c&u)#`kjzxds|y`jsd;Don`bGu+Mho%nG
z=`%N^>wmlY6Usk?ob6L1YlZEnU$xmWf^Gq9?X+w5=VBKfef>@NK;Y#s{?9D5JZ`e2
zUrQ}^B>_aRC8p0s=Bvt?lnvm=^gr5MkHaq1mS+L%brDY<?3-ady>lTOkX8e2CvYxS
zg;5LLf$|H_0s=T}Mh66)iUge&hlQ~PD)RD+<Ls@4PCO}UjBWXLQm&W`a30X@vTwn0
zgiw#*nkO#_Whnv_+5`M-&@fc)S&RPkcn%!h60UeWwSW&wUSt3D?)`=s7FX7;&&#Hc
z-(NkT4J&e(3$@$#T;|fH+aNMV;jN$^%XCQUT;+cNoQH&q^71NS47t`L)yp&5fDajU
zR}KswB6~?>MbGDscnC89M-$IHK+Tkgx2cr@j0gOxf1e_acqspsE~ezqaz9CsZMw$p
z8b-xuefi7Vsm{4%Ie+V-)r9w-A51jHYbWVtT@9{jLOERWSFo{9fROm<i2&s_Sv`Qm
zH&G7+6rlSNe^V+E)zgx+BDS${uT=k26=DMZ&%kxifSMKN4Gs>c#kkjdN@PHG0*kRl
zoq)WIHnhPpa1MRn3L_@gqL()pD)&LyU+WZ7jLBo;=g;SUd~K|m6&x7lJ7NcaZDEOo
zcW1{S$S;d41nT{$I>|QpLTvR<<)|e7%9q#o0Q!J!1dJEsrgT-oqwA^;X)q3N-T+jY
z;5zHTTEOC`%IF9n9JU>6&g71`w#?iPDG%%2h$4}ITNNDKn67N>CJ^9f|3HAjLp;EH
zbh0f|&@~Kj6R|IEna+JJ;P)5sJ=v<+|IQa|q`N9)e3G4qdP}?4hNGO)9EYAiuL4}X
zmRV<YX^IIDXNF-=C6`}9+5?Vgoa&42N^F6sI%v~bz|a4u>I~-T0pbAwS)^3&3~ktA
zv&igOru^Ie*-^!S-=VNB{O%NvvjZhytHCXcV!M0o(j`L^)RTrdzowxu5mrqh1pj70
z6x-#O<i!z-JtczRrHPzMmA;)=X7le&@EBx+A>^?JjZH@D@#$Mw3kXLEzH-Jy(X<$o
z`5`(#@+mH;(ro}3@8;OP);yaAs{<je>2rJNU;d;+kZmuEo{y8G)C&-j!L)bd7yW(G
zB{;DjWv1^&2%(gsMrwIfBil?D%`Ip<{?4EsYK_vf_dYK}{1NK<@bKnhjLHj3;VMX+
zInFQ3+0!<>{pO7<IWe;2c+1_`NxZ007`eX!!km9TvRl6iDDLtnu^@4Poxd?%T4-@{
zFs3;XF5nOYfW#4!(7?cjx!QP@s_~B8iQM9<H%a4fA>9&Tf~bL=1O!8h*l@xVv%X+5
z{FncQ!MK+3r#I0IWdw+eqgNI5KDAW!!$1HiB*au^twB+_gYm19&K!$a2n0q+@~nol
z8rNTgEKYxA>MT7F;!P%Z;2K+9B{xKFeaSh?YI=Hl3$-iYpaCB7piJ}cs_@R6-?5sn
ziquAM1i29s%cCr3Us4py6t7M4*Jbe=HSRY*#F+%pWD!VLQ-#&s&_X;W+6A2bX|W*y
z>kuw~K%lQ2$JGopoG?zeKz(6;K$DsBcX49-9gwL&u7AVzgp3^_FF`E?o1)E@h|A6O
z{e1+YR}*T7@@$wWO62UoN=%FCmAW_m9;+=c(_M4)`Lk#8YR<Bt+zkBx8E=OmuLnWs
z363tm?zfr%6lP4Ys4m85?<Kzu$<NV;xsaUwJkzi~g|ah3J2rf+C~%-A_>-~2m~{%%
zwt`#hBmiRIfd+Z?`XA2Hn+%1|OF&31#OWF~*_>>gF=5tS)T+AskS<KCDQ-5smXzyl
z)XX=YwZ{_O1C{;AmeD~Ha5EgHZ7)WtrvQTqe4&Qq;*3x-qu?rzsQuM7{Z}hmBCYgA
z9%6--{_GJ*ZArFI&h&c#i)jG(fA;Lzkqgfckky#$2oFIDacugr-JfdmqSxY>n%=js
zh>kurKh@21_h0xOFAe_0)smx|<q=o-8+bN}OSOR;5uAAgq)_Uz+Lv9^nPMK1Ft}FN
z)I51vvFCu0k+blK2em7a<*pAVty&N8>G@g_NKybotA+8_=vDoE!2KaKMqG_k^N??z
z4{cnrPWfwy@Gk`Fbxj2SC4O=^IyFDj|H0lQqX2FDMG|6lh~@5-m+M<usz_C9!(vOW
zl(O9Zz&4;}2`QXEb;V0n$U6B@n_s0_+0Es+zMw%XZasKKKD31^K68uD^{?j6KXvfn
z?VIHzdl&u9!yE?#C%cYbeh&)Ydi=+0m17*KX_$hsmQ;dDI3j6_NR{+O*sfYd7{jgt
zhlN^Oh?pV3q=o2Bp(VP4i8g(DEcL=a)RB<bbwByIHiG|%pHHNa%lS4^$W8alJdgm_
zPxAe$&&#sM2Znk2s8=BSB5pY~JIL|plz)(4&@%9ent){?M&xULwo0AE;qOM-&Fi;o
z%V$Kn0mDhGH5cpqK|ocm9bM8O+1>{Q_(L)e9x<{OH$g0YL_5cz&h7L1&3(L}58JZ7
zxvXZWK}J~i6L@Vz;@v2$x(_u#nd9x*R}n=;ZZq0=L5CD9_ge?`00dqVk@~~8{1?No
z*Oc2nI{Em(0p$wqXxi?grBXttLPr1NNc8{~v-Hh~EJny|UU)g2mm~TTk`TayZ8d;c
zqO{qi?(FDeRiZKb!5q{<a>>2=Q|wj!`hveYX}AUna&;TCRx7aQ8#pVV{do;vhyu<;
z=JD(y?ukN89->0qmhE8CQ1d@5yq04T4TA#Izo9>Pk(Jg37|j^1ps%ajS>PZ8YRmI2
z90a3<@c&c%-{W5#;YPB$8yXtCCUZCvnW3#U)s+g9F7+Txwy;VqYAfa6g@0XMp3me*
z{*m1|I+dR|S7yK^BVwNaWPBF(SKdwpt9i=TrOD0_*1TL2$!nDdHi_sI2M5Q%IZ64t
z%6~+FI3&fqY>z_v6r4Gq6;h{p``FLlm~Tx#%ii5L0UNF^T1xU-w|*4!l_=#`aE}w0
z<abEW??+Uq0I*<mq%vS1BnIMVZbpQJIP`x$23nM6wYWvOnQq<!pg+hlZ3T&N3?!0C
z-q0~^nM}khTpGX(Kil2ob<z6@DT$`2^y*_~jzni$8$ML`cSPr$pGg&|3gYVi-~#y?
zMA+F8FwqgI&@gp-D-(7K_Sqx*=^j16h`~3C|Eq36ve;6G3{^J)#6Rp+a$lbn7!YPv
z?$5SXmo70FbzFzDnb?)#NMOM5NW(0c7P;rCLWDIVLp2eCcSNvvQi*dr!S84~F6y#c
zK#TZZD@gjluNbqM49hR?>aE-W&LZN2dBQ&UlqwCSGSPYT8?c@4R@ln_k2JOo6aiN5
z-9gEh;b(g^l8Z?=tG%tjLEMMpYZXPW&;6WG4t>3{4raBB#iEM?PyhhXy->TyR8f2A
z7a7!aNe208gQ+e`ge|iOTllJ~D5y-G>E$A65Vqt)IT5fW%JoTAs`kH$qUsZVA-$X4
z1I;vMEda8<P>4>_ZaO$Dmuw*pUc;QE2>R&JYVYHDfO2y-)*zTB1H*8une|9L?_E0<
zVOO=FurO6Yl;eWq0?5Xn6y>!-$d}Pt`Q=NF<3a%<s*`=?n!ud$kvYSE9r?#7{z?*T
zI*uZDpsI?NDG?NK**#U(!p7KDdf2R?J*+z)Wqbfo!t-?)aH({%9*n%wWHlrZiPJ%T
z37h95;zGz+!igchdQ%I@ABzGk50K;~ig1g5NyLGywV54w1_>C*jw!Xobo`4Tvp#tX
zdtPPCr$R<plAj>;RvM=gMV{@1mQUENLdRUYoqalxdpu1)`H=7}*RlpwN4>U0Y1qGR
zir_zk|KsQU$jyY|S|}+(^hi8AvWaNgT2R7HM~_7S;Jhz#Y`WK_w<X4Q3L~i=V7WzS
zZerk<vvKk|AT$ra6iD$C6t5cj-7?h$SfVLPn4f<PoSUv)o*qzOafE~i-0U_~tFBt%
z`l4!}6C4MAVQ-EsKq!|n>5S3*$ebh;Z6J=W7%280O?>PE|L%PlWzcEwd?NU>A8Q7*
z6|FPb$ejBdU<F~vU+4j>40M{e!M;9lV=Hu+N^<2E7zjysXrMejV541n+71WWAyX7Z
zmyh5EUn#4;e7R`H;#(|vdO6-{Vf6OIQJOU<dqA{%V<>2M>^T+<hghe}9?l!d7St<l
z1jKcfF#=0J8miFoR}>l$1hXx9ogKU8hHF93VIb?afWlM-*D9cfKmY=RyBGV!t6i2l
zn;~_Q(BGVGKK0U16@UW6ktiNBHAWU8CxBf*hMckMz7y>{WTYt?X<8$Y7lH%*EA8fF
z(df?lj*gCme6nIqcBS>b6Fuelwu*Pe#_K;yFqk2-Ru5Hcw0kW5sIU%*u|ya;SQY$8
zGe9NZ_HiElB_uhvMJqrt$sdBhX`t?Z1c)9yFD|Ob0H>)c*`@==1IEe=9m;lDAL@dt
zt4&l~>ej%{Yxl(l=M4;HRo2uP-SDo+?sla#6^5Eb#q|I{yq{Ji&Q!4qbD5H_7&Tg{
zB71WqhHg8{BBZ>YUz{kEWzu>U6h>n`nAp-<M~HVmn~(h~RO~M-{!Fmrlk)A73mHPS
z&H(<79&#vlbs-;6E)eUt1$RFwNkyIcN_D>n^`c~Z6DUVTtPXfb00cdmNVEWbo||iW
z$`;%v!aek5;a;)>piuWCHAR^9HTNVr$hWQ43NIlZg$g;ODVJ~8#z^-=9Zvoh7Y;Jl
z5TqsJAk03;cHlv-D=46WukFF|&$@>;PHjEc*See@o{nT@U|_gNHWw;x5&^*<xTiAD
zM-c8j9-q#iD!REqXgAip05ySWj$Dw?Qdw8N1sM_N2ly6RZ%B}^?9(wQrPPda#;gz{
zk;9_n!sv;}bLNFyBI*Ho8@qOb3kvOiWjz0~OLou@ois}9)>|I01Ne5Vg|(@dmyCJ!
zVqd)5%@`0j`%Ky_D8$P!YD$3fg`ZG7<tggzr!mfwl9B+M!36`&!?9;si;<eZQGU1;
zfZzfCjP@H6#4CA|YterNm@7`o05StW8lqH;M*7MdN*=jKE#Qe;49j}u(c5gu-Ih|g
zj0BvV`i{uf_q3y?Q-FEs*J9AAk){CDG)3tH%3DxFq*`(+ca=w-zO;XF;{DBZFbtVy
z>ta4A3pgtTNy>Nc&aNw`id)RILS<|c=0=7|t7S{d#4eJ)uM!OQ$gVJeBtt0~!_BwX
zj&6_#-qt1bY@vZ4U~?}5I~7`DCO4EY)V=?%OcBD_)H_6C(dhWP!)1Vo|41KhZxfKd
zbg7jgsyfxRVhr|Y!6zbBnEcPKLe+p6X8<diorm?S?S%(Yvmru((I3XHfrAK{0zccx
zTacjxU9#cGNV}n^ic>>Ke%IopX^&ZvF%+X`12#OY<9YdZ*t6&O(jc}39nys9h*Qtc
zf{QSr0K>*c=E;VJ+)~Upa`SiA5{mhXcN-YOD*8WG>MklIJo)3;#K2E5mB7TY?qc(@
zZlmU;R`4?&9cZ(tgjp!qUZDbXTo51C8S6*6_Bu$e32uG>w6_RS$pr<yH{)cfcr9;%
z133(CFFG(ODGBoP`(+96=;_?OWTz45WE3hHBG(dSK)W&`@d$WL>vdA;T`1HIHqG7p
z4*zUTmjXq?&cQL}Kzy|N4pb37l&2I}p17br`&Kq$zSo2LyxK)<T-0HeM7p{EYGo9W
z`%^Bi<MWd4OUwjVB}qLW?l2Jq_ll6R4I)8zkP0|SExsV!0k{qnHT_h4{E~@zJyDui
zIMS)@SW-K->CkK@3aS_<A>)4RrT=C6JW&WJOE^gApE{VmH0j7uLNsY(bOnmo!BXKm
zTZ_t>mA^~0HExy2j#KEx_HroMO&)&3#KB=z6CMI^@U){Cxkf-+;ab=*X1_NXS=l{r
zxYmM#$ISYk5e?gG(U(8>BJ~%LOrFZ6o+V{g5wO2^6UM&0fFKl$q?@QcpAq=vZ4T7G
z#`_gLd*;HLns(%5)l~F>BLekCbyfQz)6oQ2Yjd(DBzIka0;IkPv}Kp0@yDJLb3&-S
zWi>B@Xanx_fxB4(;EFvc3>qm(um|<{F#Yu<Z_@J`YH-KSn_6Mb4<Lh3Os<(e<eh1(
zoTQpP8`6W?i)C9bb`zd%)%O-!&$!e3Y2^T^^E@oKUTsZnZ4@`a@=)2~M=h)e6+|Uj
zT@Ly^Q1;}s<hl;E1a=UyF~r%t*Ou}fPyHrm0inzthOrgXhu9Ulr9h_O&fK~DkB{7S
z-%#klJ~m7#Krf=OYkH8{ZOb+%OY(HvQwi4wJsi9~P52Pom1!J&C+%pG2BNwX92du*
zMMO{zh{JK>62@y#=LF%#pbw7*i2c;Z2fN)5I_NJ?FS!Lo^t{lAitdIuxp;x7P+?R}
zM{fE<tdMc@)1D}~5O#}CkFZWEoDdq-K94i+?kc%%J%zD^GrDdJgxoNK;&rxKl{1RJ
zYkeWZhGZ4Clpl-u$VLePD`McBxbV`K_6);@>@IR7;|$u^q4R!QzRj2;BwHYftEb1E
zJb0rRl74VA2YBVBQ^jNmItK#z05rm@lr*JhCN%Tc&h$W;f}2-X>3w+j&OFpOiwFh=
zb_!VZRX`E|uKclXPC_3!?Y=m<z6#graNYg%GS^Cf9~SIQR#)Mg2N}UTKM)wG7V0Qk
zPdsx~D1>LK6bhpQ=zX97dlpPG8M6wJgp9u!?4++%=J<cmaP;N<%qN9=UhALRi73l!
z8FNA+AfjxJ$f2R_pKT6obKrkD2QYF@OH|>CE<-y7%unIt`x_8Fz$>|DoPX}zu2aga
zmtViS-RPU;b}6Fi7~Xje4<q^3*y~4?5(CwhOo^mk8HktYwl<fWrHvaMRr#d*_~R4e
zH)_vCh|Q$)LPic+g??Of_x~7XC!{Zg9r(suYc;gCw#L8je>umur@zgCZ4Uf@!vRl0
z)M|s2YPQ*rT&t^h@0MWJHG_hu5|acP@AEb_4IVqT=Iwo8X}ndQ!?Y@B0q(m<!uD&T
z)*ZeXk9QRoW@NleNEo+PZYfyoD$EzcqEW_+t_w|AUL_prUP)fLVyPCvL`z$wnD}CT
z{s5otxPM^NU|9FKL;0~_O)HshTJgQja6!LZszxy6v<l%CPcrpJ7OqSgEiLan@x2Fg
z-eOW`&~w80LaK90TmkvEg)w>ygTb0@`AdOZR{>cRw*Cnh4yF{2CNLGv_)w8)hC`U=
zXlYq>{b%ap(7^e7NjL<3jFPJhGk7s}&5=aka&R!&X6uLADLh9A$^(O^tZ?5K@fnVE
zavcA`zqVM&$D2DCa+I{MUpyq_1ecXr(F#A@U^$sqK~+<E`#QE|d61P%-9b0ic7o;L
zV{*)4C}%#-F+{3QfI~#%joS_eoq1e@!szQT%LXH+rzd#@Sj{J%oi6v8CFio@Hpf`6
z%*Jx)q3z2&!)o|hY5NWwJo!VI$4>fotA@vZoW34qJ1(T&yzzYOcIa_F>+3c)4h@CX
z9@H59svzmOwn%@XVuQLd--9?$yenRr(@fAe*J@5WA+ydNE$MTyX%X&P;XjYls3$`~
zF>qnR$;QPIx#%Z0UUMj`9kUpd=y52&NLtutCr;b4gl^1Y;p#m7UP7wxjpp@<ExI;K
z^k>eH9yPJNhSN@@bbiuBx2#J-OblJ*qRH9Gf0UH8tioe-GJ<>J!ah8lrkp?e$`6dq
zReFVo?~z$LoLa!1I{6vSF~Kb}nY(x4G$c>@Fne7|DcG~nYka&8n(p4Wn3_MMFYHt@
zg41k=J{;6-J>o<nE$tN>dr2m(@YS80oi=briqA2ltZqxocOcntHt9$ri8pCs*U|PU
zhao8&J4oI<vBI4-c{$YBo?fK?kS@iu+BG^lo1)GdpZV@;u(uy2lg(M2E=8ixp8au@
znwp?H?DCZ>=b4#}Itq=O(N3M0Qj1pKkZkF+4(caPxms$~y=kGa8_BWgPRu-YI=UYO
z!*@4Mmj)7vN`sQ)GCKNxW@SDwqsmHV<7E5Z$G2;6xnrOP`xx$5I<Pr13yVpT&)jga
zg0X9R)`+gIQ^X#eKEES(DN@b^kU)#ZTVwD&1M0@h9JS~#j5v*U$h?k$n?bHvE_X>_
zG3(^IEpQ8{k&Ic>IbAG<H3E0=?-Ed8wC{*8FI2a#a%ghi{j{F>8R}1^F*yB+G^{N5
zcT0`S2&@d8*3XH3qlHpYF}eqV@36~Ex&AD!zZ0quB~)0tL=SLELc-S?olCMcZYms8
zx%%M)I6S=nb`kF8l0*-bB`D_Vf6)5=%J^Y}*y|jKM*{_LM*iOTwQ4B><*KsWRrc~i
zHm5!2txQy7Xi$Wp$~aE5kF6IRW*<!#dd<~b{ZZ)TRO$|)a$0kSF#4`fHSVzKr-g-C
zS?U)$tmJs^%=VV<W69}3_k_cd=KUve`nwpm_fQ_Uv-ulPbc%|)as0NupHRtM`sGh_
z>-XZct9i@Q?abWAz@U`kJZxRjoS?1dIFg-^SQUE+r&YL&2O;%kxnde;b1gV}7RSXQ
zES%c>h^K2=AI_`){?YG-L-e1P=!uE@>?T%|FuNIK9NCub_4GccX~$Y_8Mi1|m8a}d
zUZm@4rK2npYj<!^5V^B0pgLk(ngx!fP@F8{dG@SsW_jk}!ifc7`d;d5SFbWNGe2kh
zJ@SrwDd#l`*kB^A3yI|mCz`USno8oRP-u@nT;Z#E9W8aWEi<1DCbB$Dak^$@Ab{|p
zA02%bFg@QMm(f;ZGZm=gMTnw?8`LhK{nsYfM5mw4^j8p&;*5K7ILoyCIKg9c*)qY^
zpD)-*o_HS!2vFVYM7px#;D$T)n-UqGPNE{7x-8RGSU1^)WdKIpc>Kt@#A0wd6nSwg
zYdksUvNDKWfRL)XI$ZwN(i1`L{9a{ivx~M#8Gvf^I*|bK$vcwHq+CQ&kZj(aIEata
zokXBv1C{r3rI(*D#-$Tr%Hj+5wa?F(6OiCE@;t|HXXfQc_;^a@DYbA`^G%?=gyO@8
zJOw2xVtSm0M5dCyC#hf#A9U-%9J~2-I8AkyP-r+?_#Ve)-q{yn`E)GxTvkKr43n(q
z+CpBuu<z4^d-zzGnTb#1cKG-l6CoT$eKaLwJ^I1{y<Ae<fo;4ghlgGGE{<Xq$7xDq
zBoifsh-wgvx`LJdrE*M`X@ea4IIK@+p6Gp?mLiR9nPcgX9xcOu+POO=K6egESmS#9
zz<#ie;E=gPesU<Rm;`hYq*q4jqE)AZ1DU#yP~vn&Gzc4mLCgFUQ{vU>S||unQV#Y$
zjMKo=61Ep8_$T}k#-RHxakoS0;}aaNYS?-S=>E!;#r?Qj%7C;8ZK*0I<}Bz$LPFoc
zdI6`;S0_#gI_kJwA$fhhBZzsd99N1sPv_=d9c>898F@>gv*iO;00qYZlq~BidvdI5
z>u3CMS^DMWdWmea+d!@xuKo;-m}s99e|@je3*aniT+uOj>wcKhI-wqn0OOf6P7+%z
z=grh&A*@27*FI`A{|XN6JHEA)Jk*Shp^M7kNpqNso5kgKXlDLoUbLAZAoycva5=?k
zm#B=*Gc(r$cjWn&C-M^#K0JNuTT(C4k`mmAK6?UZESkMwm<}(}!yRx8?6^;9iOAR0
zl$DV(Tp>wTi9Je2_5_zxo;v}XJm6~5_h8>-0Pci|cKe=WH0#E^xnq7xSXd6%Q4u+~
zT+WI)b1O8b*ZuX@u}T${)sKwJ3=FY6+uE7yQa*Nn5Isw7jFqBd$SoYg?;pQ6mfTv>
zRR*Y_F79HK-G_879bH+}S(%s^I_it>RksphKk%%w#80LJg)Q7l#7Fpwqqx}Ey-p+)
zJg(<)6%@_h0`zRh9mBrWF^?~}OX#^1vD!|ygp2fDE=#6qIP~qf&{Me7bts5F&pJ2f
z!>s>bXz%xF+2ce42kTyJop2Wl+wB(X)tsz)>@O@gm;{ru245*OE^{qXKy>S*;E2$-
zJBFu(gy7au=`9~wDa)r7t7a$NG4ES!ukhTuh$IX`MCWB5iCIu^qqDbAI5Y{=`>c(P
z1*uAUZFxux!jR>DYHVC24{E*+6&t5{J(1^GaJJG@oY{+H@H#JxK$&rWU*BZG+~S81
zg}vsOZ{|vv!J!_T3d(G+BRH)+#b86qa?-HmkDCiE-&v40Z%@~+fRah^*D4WSIUt2_
z0XgYboi~{`=BuLU2{8#~J`S5>(h7(42aDEiYsOo#Y$iVz@bPCX81f_SwGs~Ei1q!m
z)&rL<jEur^+%>@4N__W&RT75~8Ukj`qKuMQFBsCerD&_}<ny6eDs&H|hk(MHjP)oD
z&T~W&zEsR`SE1=G^S;60VFDtY&pcI_l+GL9GVE*Bk9tmp^8wMf$)dHjxb~M{s;h5h
zo>R8Q`Amc}G2xtX%jBnqb21$+OG^;mUBY#+B#J@mu0;tYyR3PCYB>Yu_#Wg792|_#
z;F^9;Rw=)rf`w)ZSZ6lKlwM_BSAx0AMzE(AW_8;YJB@*_0l8gU^s*=;4wrBU+5UB#
z0|*C16jo4UVW*xF=$h$RSzGGfHAQ|i*D@kt=i<Id!$yPrDj-0>s^D->mxlG0oR*QU
zwlx9kO&cxijd!F?@9OIkuqqqsSR2rAva_)fuwK?R)HkrEVdvuIB48CWw6?mgYbkDS
zde7WU*UXycA_1$oxrw>u&3kvXp+^Z_TSIMKDa*SK1gzI|&Gey{3vAqMyfnxU0js2;
ziM6gJ4XdQdU29zlU2St6T>>E?f?uTCw~d$VL<Ic4KQ?h}KW>D-y}a8T*yg}C2evt|
z&4FzWY;$0n1KS+f=D;=wwmGoPfo%@_kL1Ap)`z<{&@LKmi{y31?VoK9Y;$0n1KS+f
z=D;=wwmGoPfo%?Kb6}eT+Z@>Dz%~c|r*puN@bE?j7T!8)VlV&yNW5cvQrjHZ=D;=w
zwmGoPfo%?KbKt*~1JigEF;Qp#RhcNy1)MU`KT0^cI5;*cIR8NjXRNZRc{?`|CQ|dY
zA3JFp-I?co?uRb7#az|Q{#?2@Y3<{W56OzGDRlW?T-Sa1ZUh<Uj$Oy|^c{HSJh|K<
z9=dnz{@z;FlwmSCvlC4Z=v6FFR4a>~-~H^e7|D_9;F74j)t)rzmoJldp)D3xgOz!f
z(mfB57ZsLIt#^)fYp8!yAg-C}D)LH6oivZ^{nC)AY}Yd7PTR6F_pKq=&OEXf6IyI<
z8CKSxOd|Tu|NcR{=C6F}*Y`D}%!kMh9wcWIxpG!twE9?`+dl8vzWzP0ucYPJ9S>Q*
z{)uryRn4F`KdK|}>Au=gZ_VD?sa_t>j>B~Z-a>{_A5Sdj;QJT7G$KyCqak)Md&RM(
z`WCq!k;@xX?v<tVvwG(XYOB=m<`txjUnduO$*uaP<<rp>agL`W_f#7>NNt(3lx!YK
z%Ckxyc==Vv{I09J%K}0ELDz&lMwz?mt0(kQ6r)zDtZ%cX?-IE|>iRYRkc#QPu`r>a
zGmAOpj!x|Zqzcyz4SIj1m<7DH5aBJ9l}}{|*0N>X7npXWnUnThuV-fWV}@Cx4C`0P
zibJM`dF6Xam`ZJwbh%TVhDT0i4yAlc<O<y@ywqtHqNv_@*UFI3ihb<8V!e2*M4Tm_
zqQ&#+qpw-gxo*&=51%C#IJumnw)BC;Yta4Ru!{nlg5Fy(<A7}SLbGdzVW&9TP4&pL
zzo&eCKr(nHi^`UgJVLX4c<+$uq4UpdPJEA_sf^WnbnC{aTW+51GVAe|PGx>ks{LNb
zXd6Ot&d%HCG}TG1XIwuzO`mE{IbCt^h`2aaUsrIS=!sNuN#3ru{u416wFmurGk1F|
z3GMN?<>_J%miRt#zc#w}EG0>FLac;^8B^E~k`GVjc3tv%NX2%g@93?4uFt=4F20%X
ze`I#s^~u9i<;4CMMDgst)ITWiy2}2>&~ZGoS~+NM;=$JIHjyo1c`Pbd51#v^&qY`H
zY$dd3sdJJx^}Dm&9$h}&X<gr{cGGA5()e=YGKVki$bWY)t&;5R$C4w75g&hCxfj!H
z^umUmL^k5%&fsf0I~ujamP`hGo2wiAJdR(_XUfA2yk;=YKS@9$=QKXLQuC=`^n)Iu
zp@>U*fSM~QWtRNZ$EG_@=Alygeero0WTX?cB?L6+eMt^BlZT`UWRVVabU8k8FRp#%
zc75;NFBko^k3FQ)yd{J2Q|2Gj$nBL2S?M{MT+}L2CMMS7pNZdl=iC{wk<)@%gZUSj
z$=8w`%K9`+OzISkgRh;P$;)J0<#;s2m$;M6o`cjQSXQv(T-HY>PHL(d{wTxy>ee-<
z$2y+A34WUqcRcQV=FzpC749^Wv6V4&Y6Nko{Y~tiHm2>og`d6a9Bt-s6i<I|<oSVt
zvwNO}?m75XgDFn_Y~YnBwmRF~vIkyXS4Voo!wD_bdynBqnHCDFRjq$`YxzNBFz2Tr
zz5Vg&f#WxMnTf@ZUnofJsk7nS*LslRaN`XTX4aFHXVbY4)N3-XolJAxd6%H$itn&Z
zm!{8SmNU}l3cjyfOa2rfmb`navC;mvqg$~2{v!i6<4sYI3i;DNjrdGiykT7>SENNB
zFb#2~<$n7>k4CnCLAVc(X_4TG=K={ak=ikdSNX*KYiq)N57%ifR2Y`FybuuJ<I_nT
z`6`-Pcr-fn$DO@gUh&1P9#Un)ily4b3}!saV~vI$nP~pXHd)Ns<ztBk0&WBkx$7ye
zR{I>r^Lt_vwZbcYTTOO|f+|JfvuGLPGp~G>t0?u)+N2j~cp4%lrwp&fu}2Wozp&#I
zxGKk@`9qn>;+D*{<RGG;gZwwXU!+$&OW&f6-#fFPvLk?`S8}=x@7eAm1N)OX_wq;X
zJ)mM-dwZ|<NPy8u^kZrbkp_wUZq4ophX9dS-+~$jNwi-<Oq>I6l)l2Sr<I|1^#$?%
zA1-nZI?bLs9*He!4F`*TUezf}3|*ec6&_k3@0w`3&5b(aS2EOheByS=Wd^+d^nQ<h
zY6Mq=>?2O^$Q<ICUcY$o;`u=PxcGKSB1*z`^+;8%<@in+ts@d?sjnVg)GBMNT|n*R
z@jD^&ugnQIJB~T|?>%*JLrM%zGRWMuGKRwnJe)A@jo*h5*m=3xH%=k^gF^_NOLmHb
zS!_KhHzLNbM_wG}esTB|mAt}>Q>Qcz?BD-Y^!z(d&-?dZ)#L4<xOC*vWfI1ETI-R*
zrSIa>B1Z6~9(NA1P)jR))p+pHrP<!~VrfA2%A{R!67^8E+E_7mtr~@U9`L`_%5QhR
z&4FzWY;$0n1KS+f=D`0m92hvPUqc7?J+F?p_GjT9iqFwKc!E9WM(h)~<rluFTF%78
zB-2Q-uTQcK_ndpr+n|>(uNS*^3K*)(gcZ%!3VUVY-e8m?&1K$2C%4+d5aq1njWg@}
zVRbRm+H-gs7cN|Y)6O5{aX$Mf6L-nGWmd=9WwAAm8~5mC&WJ*S!iCaYFG-)s^uIh}
zKcjV`My4@g!JGp3^m&K%XiI8XZ$}ybR7Tog9+C26fk)7*W7q%k2;5|2Jyp1bcTfhs
z*3HGLacwJo-_8oT>OI)Cr~YK<S+$&L_%a()+{5Yj=HSLtn{T0hY8;$(uS5&i*RtDj
zuZii=vx6&QFL@6+FLi!AOeu7nU1xfB@W2G_xqcjW{Qvl?h?#%5vvqscf;U;MdZtlj
zO$3D>wOo!h;|g{;QiA&~e5HE6t?3gJUy@DH5s|&<w7C2uEn34d-ohDOeGqk#ab!B2
zx51qdo947s16SFjM&sMyD<52YAG=dKPfzl6&#%qZTC`^+6Y455GRE0Wv>mPFvi@lo
z!DnMX7>IR5&s30ezH2FP2!M;DeW=!}fClrPx0C`oW8SQe0_c00h7Fh;p~A)}SMvEX
zlFTAZUN*@Agl9b7FJ3gj*Q`zbNOS4Tt5rz4)my_qRUI&zH<8tTPB~>BZnaq)ZGtb)
z&e^^1|FHKSP*G*ux-f`hK*0zq0=5bU1VjWR76woWq9{Qnh=?dimJC(20R;q9BqssM
zQF1O6$yt&}D00rB2vtS>a~G)CeY*SH`|cg@e`6emw5?KWuQl`h=G<$Vs|KYO7FWYP
z(wE9TE9HUzVsm$Ahr;X@XLRBGrqvD3BbS2DTBznSucKo;qOtU`giIiehr%elP&Asa
zu{r^6Lh{utK);Ahmh%%lAhIx1mwHfShmZ{$hS+IOZP(R>cG*|B<1l@O+A>3h?!w*I
zaLFx9nabuCd@uZHet&LIj!|d*IMz+K=AC+;?E<KP7~x?645kc!?_Xt}j@+=otr&gR
z!4U?A!-VHNm@bK%ZB0qlBcrD)J(cmnFjYs$7v?JI1Njiq?0uA~hZ<`|xFa#f=B1K$
zJWs0_uU+n0>OLED_SqE75U&C~G`$njng4sh6Iwg94|kq%4obP~4Xk4}8C00+Dl|RF
zY(gf@X+PiuJ_zbmotmBY3qzw*2kXeTvkhKOyEuJ)c!V)aM9}|}`rp`IS3X+qm8J?o
z#C#pyJ&VDqOJF<VAILr$ZWOB2xRI(m`Gi6IRYPO@IR+0x=X1mAK(WAu)Jh+|xO<g*
zZ^*qxVU}Rnfxn2wfeU`tGM*K~-%_u69P{AId~o$!hOsslt<e}yg3Fz)5-!iRn$G%b
zA|x%TggdU(gU>FAYO1E=)DwXXOG`FqW@2t;<HvchL}Kj)wo3up#Q9pGbY__OgzgVW
zm-alqtx~)`@~YibkMU4{W-=NLmuShm?&TLX;}UaMcHCYF7fE0yI=C&iQ`ojY25u@7
zGaGtp&?elJpdnbv^id-ZWX@s;2Y2GPj;1qJn4^XVBlOr%fv<cd@goa2PPrsYdmM{b
zBUrrf18aFgukkJ1E|fQ;&>bQ<ewk-{4u{zxoz>F=vwkZV@GVOT*!Gb;!k8S7nlod9
zP&M+{S0XIpIT70BeNpAsz&xwLu%aM4{Gd1QL8S!sEUjV(4^Aeyf?Iojg6$wmBV2ZH
z^KejGw$8!=*QHAry1I4nqXfxAW)ojO)kp0(hjRkGc%2AW6RAkFfBpLP_^#jm1P$MK
zdv9G-Xs9lD0~+6vpXVb~y+{4^4rVq~IoGm1-LUk8JU{Pq&S5hMm>Ox(3}*9slwSUv
zsEO0qG1MP~$?CIP-<)wwDK=53+S7hr5DGH|H&wgkxtlbpNhUQd3R#SO_T<tFmm_FR
zbWpRQ=hZ`{s<oH2CykA<_xRfioiBkgfvGLuYJzbvUF`z!>`rf)$H~m<(Qmn<Pmxqr
zf*bCzW{`B+nHm00`JKe?aLF2xxZwT~n{CuN;-!J+<>27R{RDTh2i;X+3>2A5K&--K
zF`Iw^5!8u6J1>QKq0lHDZoGNMuX|lS;7Ae*X5{YXyZ5;<ZuqkQ>8%T;d^+1<(C`V6
zWVpUN=F>Aeos4>gFlkpd6pSkfi&5fI&osGu2<~E<M2uqMZ9H*-?*QA~zKWx-@57v8
zKbSVC+FJ7Z;PWlac+%^GHVxIuYN+b>B@SL(w>ac2mG2r7W(g<RdsA#ER*LFXc$)q#
zorEouU`^s(>k@EYn*66hU|?kGms85fERFsk!>vjoAS~^Pn%YVhaAkQ_bJc0s;0-y!
z2VcHEi~#j;-$A(_DN~NTq!AY1*APdjz>hMYWP`aX?EHF-7n4uSHz)_KCy%{5qoI>M
z1mlr%67jzYE!=-j&-R{`>ge?W*J{+b^-z7VJx<J&V=}o~ywZzDBQP<+FmEsd9XPsZ
z<&b8e)&%ZMRLC@UzKd>;E6ef)AblK3kK}w#E}1+9xdMa6ke+SaA7H@>7TPx~@U-ze
z^U>l0x$8NWA1eKnHPSO3ALek1`er(c+gS^BQlvZPb0Nt$>uuBf((>jFW|Mam;9h+N
zxLm(wNk~xebvxUrPQwzZB}OUEUl^Tf!tk&;iNpQO0o0y_g{_=L&*+%Mjc!LQTBQy5
zQ^N)2p6!w4U{-7hqhz>w@geX9;)`!-`Xn5N#~YZ_bLfJ3g7s{)kA~^oSU}rdWCv9C
zTl+cgRrzmUn5AR79EKjYfN%$VMvC5?IowN!CXDujX&-3K)N<$GQHBs?dQt<E+uSfL
z`sls|m^KBJc`R5iuA8Suo0D4xYO$O5&djc7%V|l}B%Rcj))But_$BO(yPJAYrX#4T
zT6Lx)Sl*(I$OSF-ty=9lgHdFI99f?)j*c6fU7J(&BJ|eKGvm)yRmwg5mU3*-P6dmW
z^i^zy*%tdm(oizN^sKFlVbZ;c9P7{v+AkKZz$njD`11D#+LVQ}Qo*%*3~*JvS7!px
zEk<0HS<K^gbayj4O7QV2A4=ku?|aE8`kh7uCS~vWpjlAHujdKZ*ces$KPY%kwuH#%
z?)=bGPz1QvXlaP764Qu%zIl)B%s}4E7fE!Cdae#1$=t!m3!^tTR>>efPZm2Bd#7*{
zT=kyy=%`TT@z-Q=G;yE;^;9?CUfc)wLChxYJR@Usm=9k>g*UI&P`zWQ)E4qcvOHhi
z!!jwE$*&(sujQQI`|%Hx!wAC(m%tt%7Hf!GWk?UEKVCkKu&8JI;miBLrx^_igW<0W
z-DoqYU;;)jo_;rz;K_OAH*e_2o0GNst6U(M(7>V#A*9l&@D>Go@gR??%U7@1b<5SU
zb7w!fZfAFja8-@G5?o@Pv~6Ob)-FuO%L#oSuG<-o47zXwB`*&lK*C1Ce79Q|T2IP>
z@1&1agj7#Ad0IY%dHEW$SEY>VQ3~udHyQZu$>wF9+qTfT+0R({xVordW!_hQWiVS+
zK`0n2Ql=2Ll-{qHM|`4H^zvdtx2em~Yeh_4))o$q2b_gWIlc*$foP9s#auKBQG=PW
zv+GFH@^@VJHaw|=%Y*#fHzP;SgzFeJ!nnKmA559dyvBeu#%NvkIVo;q4AWO{!P*}c
zi=RGasXad1nz_SG41eGEHD+lc;^FfE6h=x`R`%4n9qDc6Ens`=t4$-(?ctTowb%fm
z15>YCQf)*E_#x;j-LiaHP|15nE??Ri7I5p<y%<5g#zRL3jXWP4g34tiJX+AL>vM^D
zzU_camQm+*&+Ra%;Uu3}xFL3wjpT~#&d4)XT%1z{m#48d!cW-jvDWhbKQ2rkuCvu1
zg=o)Ne7-rk*m>syPI#y_6Vt2Oz}sQA(Wf>-KH*e>l)&L6$=dUXf0-!5EIz1#p}uv=
zwhL!xr~PAs()C*whwOQBYEJmx7qdY#=5&;dg}wJTZ&|iOCqqs9U45)JD+Bi+WG|B<
z>bEp9@u(l4t|sWVWhrv7xtqDbU{-a~=eSb|Fc0KBRtxS#qN_52QL$UG?;y;gjev1j
zzTY<u?)A&@E*Kyt@^;>ASlpDo4~%}ZD<9b0j%Y2V3smtw<40~Jo%hv*k-rviMG`S&
zJor<8w598!6+VZA$Hdd0#w@+O9|D7<qrpCW)KgE@b!=G?p*?V&X=Nf)U1Y6k^KhXm
zzT54|$>td=2#jk*S#=>E*%SU^iZrziMV!z#a)K)*pU$u2y2Z%mZ!p=F>B>vd30jxE
zVqRH~nRe`OfBWdr+j(WIzGi`gN3(14{aXKy=x!i&G7AxWvoz00kqc)i?kU7bYh#HN
zQO89tE;Nk$MXKlO&JfsydKN%qRE!eT+uF>L#i{HocEeaiqtN#fzpp{Ue2zjGrV24=
z;AWANkhZ~HT}~WVrOO7(JUG|s`0(pVzZDU%p5gfzbfA^z{&ZghYa}Ib5PbHW^%!i@
zqD4Pla=L#^c^F5I;nWO0+F}5<En~=g^o)JU;W(8KXMK4QML0q@p;gpl?@?S-1Y;a!
zNkoVG>XsC%Zs|x349ng!<(ia1PHMXNYyFqfmoE+6&-cZ~nCKCs*L1m7XOB3scr!uV
zFlW+w#_{p<y$70LV0DNrFNAFN^ANgkn4=Rmx_x2duv5|viY0v7)m8Yu#UPuj@<4m=
zx$JY7cujf|Ie8h_+TBM_!?DYPpBKpb;V)Qg4ma2cyp=m`Hgx6gGwNYnlD5m@!j`BY
z7>+|Mds%@ZA99%5MVu;sWxvjghx|%}fkMPrd%!GiGI$#c@!!mi?K^gy93gxzu6`+A
z3pzo!^T^g$VHk1(Y*42v#eN<ln~ktRwCIz#mQ2&oWl0ol9c%Ne<Lr_P6y<)qQyHuK
zn{WEDeWDqh?|Qr;Qhs;=B3XgKXZp1v4^(cc#ohF<pREoccu_WI_%(3~CNw2n>xbk`
z#9?;Vp+jrMZbaU%&2F<=VKSIG9E~r1n6$@c&a1OSiZt7MEByo^G&thOYBN%6f4+83
zkCEG-Wh?e>>Gn?anFx6{x-l=7(}@MV2m$Sxx!;+5h2bk#6#OCLs|jxNlH754{sEUo
ztFcYm!KOHZthFX$XJX?F8lolh7rVRAiYV+=Q0XfWtlI}%x+FB(w_Jy~d%x`MB(>U`
z<%oyVaNzQER)1iUCOhO{W)dL`;+BGC^W|PVn_pmXc5@Ddb!s`@J4FRy92UBGH*y?<
z46|X+_4Si^<R!n><a!u}OnW>P9@Fd!)5zw+n9{CA&}+uQToQ&gjVMYk!>ov^-bh_=
z`ZXsn@5jm&BUqz>k=rzI8ZJWf(_|lNh_R6f0RZDb?0LlOyc$4}AMDa5g4100nFlh#
z8f>Vyx3~REh_&}=A-l-vgtTzSm11W{$2~B6i~EMI!k5?fVeMA^0nKzakQC)uO@}k3
znGGvx#-TF4<(T$M%}!G`0`aUuzOTi2`^XQ9;76jUS)Pxzi9GihXwrC2dr0@&j7Q>j
zh%V1bguqPP^LsF$Yt)mR%JDlFEv_ft{@wniPBp8=OT;R6IAlYzj&Gr(eLFSPTRvAO
z|Abrhv{`|dHN=(nW{gJ_4%uizxVv?0nr6YGW*lTX)PvfkkPo~NiA4u8BNU+4YsX@;
z{Z_;|cwt&+?W28YqVW|KN0@VDWXD>;Z#9)+e;(QU+sSl8h9Qy|q~$;r3~5+{dJm79
zv^ixn`vom03z?9lcIHavnJ0PTkM7THQSJBPa}?qg+{T9D-2V{7uiz$Tyj`rW97#H(
zf^hkZR+k4LgU|N`jqI-aE+@4iLO_+t?hDU7?w7u?qIMKtnlDJ)4uUJTzgC8a-}U4Y
zC=E?lQ|yg{GvfpC*-mTkK5{@vvfpcdp+9Kq?pPU|GC){<cBny29Nh@&+rHNn?j+@6
z*l;Vus56mu&5nuZGmxbOf+EERJ!fa{eFVV+9Dg&*V%s(O26~n0$?lC~%l8gTICye7
zGYp)T4G{h<_B@QZ-Egr!^nzF7cCc$l_FwHTatV2fwHR--pLY{qyf}cnS{q^!;l<6t
zQM8AmM5+gPd4HQ9{qR5?hD0LT3emW*Vfzf9PJHoZN%~4TB<`Dc=LY*<wg~qrdZ;_O
zFe^t>=yU4X9E<>i+&c3?opQ8tf;HThs52ELZ!&e&V)^>O{bk&xBhA+eq*DN-us@$`
zZH5(h+{<rW)k?XW^3^%(xiO!^PMjKfws2=DTv!^=UmXOqNGJXD3=SNC+@AV%w#E3q
zvPHf6$iREVl15<6*F~#77NMlM1S|*K?LPX9X)h!A+7|L%urr-$%X~WDJa;4Uwz-O<
zIf&S+NQ?16P4;h@rdkpq4qJx({19ECGBG&R3F&U-jv_^nUD_=}TlRg0IdW!#<*nxt
zYjmI~@~YTkA>`9r!OPtv)`nuPqOc!kr*CiEC{pxt&t&5*Lr=~*i9YggpNooUiC939
zzB{lZ1aY{ubARmlkKME+0>;d4yU)r)>%kk2=S)|L2CFlOU5}UXk_SXV#H(7)c2d%<
zgCPlm;v-Bb<2$Opuk{*jdmzUg>K9By2Y!Pr6$T>?)JL5|012ya<!_I|O!Ck`F_a(V
z|9N&TkR`I7Q0R16p#W~~My~Y(6-Pj?7+>9o#9EVG`r=nj0}5c4@jE2vjG&S7^s11L
zvcXKUx7A;(Gcz5zZZ)n=AZPXuNoJ;}>XJmvzMg$H_BGG$2?%3CkcR%?mj&3q(uX^e
ze4-5i(b~bVDsj?GH=r`2CU?jAjn_CB8(?$aWiMZUyVVW{W9YP-BF6b{O9UluhoPwi
zMnAs|mI48p6xgvO0LEc2)?Z7V?x~m^Nj5t$Y1-L({&N9?GMzdx_Qtg&7Td)cRNYG<
zlb#c^)2&xf*oP3tH^eAcPV82Wyn>)gui3=GHA<w|XF5XGoZUd-E)`jjAP#e}atr%7
zHtjred->c3!d~3$a0hdav`3i`ukd67hNSoLadW1T!JV;I>|<rt=2w7+G;^Nx;#NCm
zme0tgehpR4K06Ih!9wc!as9i8%_jlkDj|-otB~8tsZ?*boYmCKW%BoNtogFcW_wj`
z_THJ42bfd(i_*bopM7EpIHH~dRBSP6X*~}&>~cL;R)^#UFwGK$igVA+m19+YGm8|&
zJ+vBY-AFkm4x|-;1KQ7+F2|d|_!mUy5{0jDL`WjC3zm=Q^WotQTL~tu-ZDuL?_3*G
zi4ef%a>VU|;QsV>Rnr|XXKTaQ31P~Q8(jnAQXe8|3ZuR76@2pRbM6YrC<9mw&hump
z=Ot<3?L-deBPh5TnE)ujo}8&@MNT$w_U;4$%L#j+Yy|z62zj2_X;lRj3L=?&FZK9V
znjMFWw0V-2c}cb3nvSu8e&EgV8z*^U&m-9C#hG5OAdR3r05;{@U`LlOJ%7>ycam}l
z6(MIZGDWZ%!1@51NP_vU7Hx`$d}|$zyV2#k_|Ko8+S^0q^kRu-tTpq4YUX-i&_V}X
zhAL!o*HFSAe3>$grL2^^?0rItB<fWU1Jh79vhm&2+S7n{-nFUh<Dd}AexNddpe*5a
z2+(sCN0>*t>s2H`@F0}a0Nk3S{Jv|k^)np%v5oUj2iE8D#>5(MW(0PE00#iMWMRQc
zAe;_=H-u|6Yl@1B9Huf4++am#+;kY`QAw1510Q6=fZqh@?&8I-ya2cXh|vUD4}yeD
zE5l&f&TzRH$c5xdIS@1s79PJjGX}B+<K9e{Z|Iho3704Hy^}d<gRci0ek)yc|5X5$
zo_Y*)d{h$>Zb~#C#Zgb{HO6imV|x24@nA-bx4&)uay)daXmOU4cJsh=kGuTm&!6v_
zJh}ts?1uCQLtI*iy{4Tfgz^aT-&N?`yKyl3Msg7#NU#}2lIZs2+aK2q7VbmrfS!Qh
za$a-Za5*RfoO=UjZM32@&33bqui~-iZSW5_Uc8=b-EgFv2RMD9TLVJPEgxMzL6Eb;
z*I^-$5Qyl<GrIAcBQ^pEy(N9fa*0^F6?G9}5}v$)tI^j>AG*oFM(Y7m?a>0HSqW9G
zKRXSDi4ER*Agpg!q^iRK!L}s4?tXdyL6iMv+Hvvywa?k|n3O^wsakg^`g+uzx0C%9
zKAxeI3EPNhe2Is*cgJqT4Ccq_j<?5c*;`M?y&XyHJ?|JmK6$!hu&($lc|Vwf;^Jzc
zNSdT`dqAnNBOrDB4^=Q<6USlpVH`S8WVk=#GRv_m?<8C=+j=h>aH<d{$ohdnFr_zf
zJCdXAuN4{3*`*XRJFL9+koCmJ^&v0Wh)+KZ0963D`xsYn@B&_VS=#ADFiUp<-dVeI
zBNCX(P&dWhVpTO0wVJwJW#<oRxCgh|&=9r(Lz=y3nAe;Hc_fUD#)rDK!{yN90O;od
zHuwJhK>&<%(5#|2lkQcX{4O{KCm?;W4=*yrlJ+TufNrcmge1)%M5>ub_SXV}zeAh&
zLF)p9jv#bK*mryQIM!R7fSHzq6k16bB-#-WJXi8zOPY9V3376P{H-NFL9BQ8=r`d1
zlNi?vZe!)aT)w9sJbZmXGX9WmYsOXZ_JGW%quT-RnF%Q@DpJpS`*7LZQk86h0u1jv
z8At`reUd(9+V|G}Jm8)nPN4HN+dYmco-CU`D7AUZ3pQqE<}s$3(R9)i5pNJM<CVw_
zhL0}wMz`@*i?Ucx0DK*;eCQ!$>7Hk~)P%{4sJPxB%yyt9VdyxU-3<B&MmYhErWx*M
zm)<GQ^G$5-?H28yR)93${m!1d^wn=q3Zze&4Slvh4`JIz&Z`qoGRGkQbFGp)9}n92
ze14tSLeGwmgJ&@&ltk~9*zAy)O{VVLS1mUORw8KP;I55?Ir1@?V8-L#EUIR$*%o>k
z&NwK(HAH(PT%H7*7ZZdVnVm-Rfb%hG1mo{q;^w_y(?o6OGM>l0TPfs{dl7`yUwV2I
zx647MClLa^_QlDi&yL0)xe9l2GCrHVu|zag4SX5Dg;{A?hF&q(M@{$iMytV-Ob?*J
zxqLak-E|M&0U6`IDiUYLPzG>us}~E?+%VJxa)g&`#&hG)XuA4p0OpvsjzWdt0=C1M
z$vqI0_xpFVEv<oR6M2`wYjhizxE<i1TM{AU#e^bF_Dz9$C4k0K)QhoIfjF8TbZ%k)
zbNW$k^&Axk(;eW+zB9@a{FGOWIlGK!iQ_ylRyO%!1_XV_3w^{%)qn`}RmPmVF#%pg
zd~twO&%>W#$$N=JWQ+{rxq0gF{+p9sg(0e=VKSvqs$;mg5~fg<ya$%ko?=<VlOPES
zy+GdnD)MS*f4qE12_!<j<xI52hLEAn1a(26q#P*6WD8<}TQ-NS@9tGz;8q^JbLWo9
zol}V8j^cdU5Up6M>km-adS#X-2)q^>SjBVtqat)8=-Xp&?3kUt6^ah*JR@w}C6VUk
z`^IE3F1>xTtKoNXejg~7z3voC-h9Xf6-TL;yZ63|M3b^K+4n;7-*LGNP8aeJzpj`^
z;e82&=vUEB7C3#s!=}o%MA_v$(c}q0sj5d?0Vq5$GY`OrjM{G_brH%+-D|Or_M0F<
z`(lInSSxR3N3PA>brhC=sDE-Y*<|bk3V_?o4!u7li33L13cVYwp#=L|SQgbJLWIo%
zH*?k^Wu4Yb0P*pb?R+5KFgi5sE-&uHfo_Tqh<|$QdgK+~${Nll0JvDR5bs#BrOXR6
zA>C;J8CWMRE=}KVXzKREha-^1B=0@OZE)A<nJM5twLBT%fWK^BS{$yduQmh(0zBE{
z&|Lhtm}Zr`VA4x|E3MFaRv}06xKutfgAr`&PqAb@S~wJY{>gfV4HvINzyo>*VPMB)
zX@xJHN3P|}d?6=V9B#j=xR94W&Q`nM3Hkaqmv${(!h74+Gt_5_PSkB%OGCX@(r2yD
zQ`$Wq_qLwr7(C*>fBzb%nqx-}oI1I&M3ntJ&2O0r?WVcM885M%W#4tN*R)ZmtwH7*
zVMN#?+k%8|*QgVnXH0K3k-3X05f>xLDk$7B7hw>axg8k*vpu;w)MY$vdzzN>__bbS
zCgq!{y6TB8{aI6cnlYcW+jHigD4)j8-%%8|$+ja1cJAd-jIM_v&#^(pkR?i&eb~+}
zy;jgNj@%oh-y(oonh$N}0UXqU8WV8iyX1s+pv>B0l<%BryFW9KzQzDb#%@gYE-+w&
zF-LBCIKgJx${8lTYRqCy;nH~vWDKvbZ(G*v@TOgX6$Zi59vP^ty*ZOJAJ_8{_fmYD
z3y5@5Go3{+IpFncg)3Y-+F^=O+bs&ebrdk59)YXfaMMrRPI_Xh?@O5TbaQ7qj+M0r
z${2Ipx6<angk2x*m$W10+^Zb=xJ7)bL<3q6{Jp|Y@>R?GOkKv~G`Mdh-5p3jGUEra
z9@*Z^vS0;0Wa7F-OB#GSKkxZ0`Qsy~w=mYbIBedtHAgPlY&hJuy{o76T(gZJbrZ}#
zId!k%MY4{YYU)Rz3z;;|at;+l`o^SgDb7E8en5k?L>TPImi0c-m_R`lRV{pc`ei(i
zTP=KRYcXWti7dil$U#C)v|e1!C)$h}B<pV^hxaBI%nmnY4{aKvVQ9t0s}TWag8}a!
zbQ`|FfF*t0QYFk!9%)YQ&i`GFFgM<sYeSGn+07L&ke9}?69f#^$-Q2Bz|4lBhH?~X
z#yw1ii9D<^jas-{>#!vTL!4{@e;pnX^Z~6tKips7uqA6~y;^*0E=0B)?;qly{S2i{
zfTv!Mc!|#6UZ^Q16=O%65-c)2W!@gqT$vd(b)X*;Ix-al)t+@`p)es~tzke9yT&Uf
z5@BvjUTrZOPX$>`*!&?OB1a7X^5JqG3&`bQQkM-MU$suUQRk=T1nknC`pER5#yGn|
znX@iBWUqnCK}-A@q$YKn!tdmQIoCm)jbkt`tPg?%0W#NkTXuK5&3I<W<kir2r;Lw~
z60c1Vre|BaCr+>h7<4JBCILh~PP0HK;&Q;&#pAI-OjU~ud3GqU*vEK+E<<4y=B{WL
zIPe#{k+)lG4zSn}MwRK9&@~Af;}GXePX#^v^1imz{dtmMZE}A(btS7%Qd0*akW^G_
zMbJbqL>ER*zde3qs=`NO8WSqxr7)h^Kh`1V4pVZEy%O|z`*yc&D46vgGB<gldJbwJ
zd5!wneg;{C&=_1TFFL8xoTObql5L9L=vB;#09vj0pwva=ph&&h9Dpc&s(CRtQ>O}@
z`=|C4&rb)jap#kLrInO!LdFE=6}UekD;`v2H|Ig9Gs}gJ3z*HWFx>wv5shvTFjY@)
z_azKjM=69AxnV?RIIXP+Jh^YT*3^G`X1Yb3auA8cJUbgTGcwG1YJwxowfjA4U~Y(-
zzAi4fbvo(!_N+2bu53e``~tT2=|(L&TOOXHM<<=Qg2}{wLQI}+q0?P#1+tKl9cenq
zy@06rSQloF?>O^>iQ7NmWN&~!=0V7<bfyFq$s7(oojflu@|udp*`VJ32Q!0p)AEA(
z4Ok4WMk@u)+JNMq^ly1I#K~-p2i=Ihz`e3&bj0OR8g)gGN*LgNVrBHzZ`l@x<Su8$
zAk5Mr$Tp-#Nwy2vA{S;-tLc`hj9Zg4b!LE@oiJ=CKmqwy+C$^WXE2ER>9-22Q0PO6
z(;(l#yC}uZTYh<OJUPiVZ5euj&F*f6_cqi1D&jyWgV*tefjW5<%m`n=l5;10+aJO%
z;;lnP#xlA+l;7N|4$L#^Y;{AaXWWrAi7W`d@HP<5F)mnEA&N`YJNwa~vo{Au-8088
zetN?$lcvXn(rZjvN7r}gFwX|NJP0Jz(hYiFA0!Z`5I#$E4#0eMiOrfNuc)zPw6L3;
zwssiT&6!Sw22SSEUfiU18~3TNB?;KEO7%GC5gK&b#Re@cpUp?RrZZG;Kvn>|?2f%r
z$H-0Wa8oR<7QdBd&{hVCVPEgqg9i~1+!`G`e^|FZ(tL?{X}JHY8VU#I765sE6x+2(
zt)oYaP4tP;ZUdgT45Bp1gWPIM;`eKYq0-JE76de|eX%@VrLYFHmusRjJWx!d$|_V4
zw@;-*5W0shBn(!j={AUV*{}%DO!K#Ka&nH&HW7NsV(qrmmE(CLyiK=a)pJF=?g2p3
z=f#G|%ccwEp2|0Tj$iA;L%9SjvSw4fKI}DK<pFWFDVJFwHc)7~tx02CIaRmUp9eEl
ziiMaq-D<qLr?fkC+qsLE)n66Uzkk1PxN)=)6j60<;5=ZOu|XGXNW|%CRBnG@_QCu3
z`0TNpiMKCHltUFWvrzzN*{yhRxyz(P?lV459)?(N;u4&EzK1Wo-Oi%<h2hOild$nT
z!ZFou2(p8hvAfoJ7$R;eS8kEseseL<W(<&T0;~1F#}2X*88nYs$mHv;tIo_QSVt{8
zZFXeA#OGwX25#UplSXV3nssz``pJ5Wrg*hs-*2|~=0#<F0D^UTR}%J|0hrGOKM(fb
z&P{a63jm4wXvI)8$y&}&z}3yLIqDiAUTuf<%)n9^=cR<aZxI-B-fwmUytK5mtgP+P
znFL{rv6^@_+kxl94KcTXTE%Rb6x)UMWp9tV4`YJ$z9ybQ3@>CwFf9s2OvjQ^!C1?q
z)FNOPeu5ZiXcE}CF(Aug{0lg3EIOB*QLC4byL}NC9IlSe)EmG#7}Uq!aHri8xN$b|
z?6bq?T;6is!j~Q<o68yEGyC^G&6>-nnL{lO9BejByW}nvcatB9R3n*TEY_yk2;**t
zV2hnYnvp@l%xys9TLzizNv-Egnv*ZITXr{m36s&Ni@!fG3Hj=88NEJyxLgvx<!q4J
z!06{f#n@~8QCF(Zv9&t$nA3};7__Nx7E7=3(``<&0{h(%cPkshg<@8lLo@hu3D#s1
zX0ex2@c(2ys4(QE5zivF7zgs@$wQX--QNk3A`k>F2#nT}e84)v!=Ysa|8ade3jdX?
zc0BHC=uJRwVcL&=9k_jf@Z>HI%zTmA*p_W^2_ZJ-9GOtaBm!{0z>MKJ)Q-i{EHgbF
zo)GXvVu*BfyD>@C_@ty<13$908F%o=t6?6@LQqA;bYpdBE;`V|BMnT(fSuaxj*Vgy
z4C3C~78hdD_E?t((?1pecucWts0kxH<F{w1hmB2cp!6wP>m0Teu#pY3LNm;))*8WV
ztfv#P+m_oH_e+D*%X&1}rJlPz$g=AL7@NMVJu@9odYJ_c=Et&kK+ji~BZ6>9#KPdv
zd=N)pVY9PttZe|E&rK9CcpSNUfafrAJUc<v*0vl{G`qPrWh)^?^+Wr_vQYgR!j>Js
z69A9Kk0j4(SavRedr#40!hO(cfd?Nt^vHU$unIFzz=@a-)r-$mx1<~Hus#l1jUlx7
zA?%4!PQRIYtZC<Gs9nWrl%Gwu7<Uyiy_YkiaCY=tYM^D;q)G2s?sSz1xb)hHpm=xh
z`7gfw<$Na<MA;D|L1iDp4Yp*Y`5+ko`941$LH(OZctcMbfBRM|kW%>GDG#28w9wBc
z_)sO;axw=>%rE!w!62?P0B2eD2k<8A$ZKRhnk;hBgITQGw8j^|gw<Q%0xX-<EYw&v
zBNf;I^I|whJ{IJnbN79$%A|4Cd8v41TwX45`|4)hb=tjU?8e5sco=xza;rfX&&GA=
zP>86!j$Y$9EjU?oj-jXmH%{60%75ZRU6`91W0dEddd?O)?J^&k=a#mRy{vB305cuI
zrHb!e_0#+B1lgq>*ST}x(d<AiDIItD!fYcb`{C1qA0E^Rv#p+8wDtGc*pzTN=Kj4Z
z-VpWv_NW682F2*l>v}lexHHXruUgf912lv1H>I4N?&H~clPz4<>hOL+sMD@xTRkPu
zD9)9&4FmCQpRdx_lm}|TX5BTxK{O^i0PC8$w2Y7~FWRKrk-wBFwF|L+XU9sx_|bX!
zc?(U81?^n*R3kZsYx1pTO=|FyU{(}L0-~V)ozeOkqH{ZyRX>=|55T=uM&wzOsmEu>
zWMh;n3(We8*|4*kzdZDA*JsRA=VK!+bJu>e;1JK#ju$%$PF+dKmtncTYkA2G4o63x
zXXSG5!vc==R6#H5Z%u(I_iHN-ukKY=jRA%2->Z1!TDa_QbBFk*ZJ}+*bhM(vXMExP
zs;$}Unc?huO<azNPUMrGwYo#t@Jz<&eR$x+uLOyPrku?1UC`v{<71U~!X^m<167+Z
zp%*MDBvx$zur)%?uNCSSXwss;xLwAC$g4LdBTz6V_qa=faI&G$6|a7yc}~wF`rykv
z43bSy&FpD06by?9$~9(Ht{vi9+1Az%faj8qR+xb+_%w2<(zn4B-x=;_@GY0Hp7OA#
z?yp}jF7$KS;B7OZoo*MIoNC(N_$kvyN$DF*!xVb(B?C$%>g~U)CP2Y;Zm2=4HYD>o
z!+0ilb@ct&?>*;7wmw;p-Nr5W;a=s`+2JBEqLrx(xz_B9FJT}1l@H6a`U>bvKcSs9
z4P3xsF*+&Zfj(B~O5a_RlX09<kD%5EVl%#J^v15Z_U{U`?*3S-Pp8Qa0JJ$%41g)e
zb=E^H<q9Fn^0E=V<*0brbB5_s?b5+FV4vt4rglx+&5rn}Y{2XdU9~+Gq|~v{=Vw|e
z_m<PbP)bVrw;9dhM!8J$+kF)lb`r45S9Oj%*6TH)k99mHXPRCJeKh>?Wxa8CJG})?
z*i85Ep+~H|cfLZrB;{hw$@{^^ouj(AuyEa@RFmEeFi&^2!v=rQoE?!Vf;)^deCdL3
z1Jg+cyV;v#R_v;8cM5q_48+gt$=Xk$+f}MN4^oLYuWE$SPo3$-5VdALXstJ~Yv?Sn
zX}8w9IH0pem{`V{tDxY@)c=DtwaGfw&@Lt2<Ld`9Spi{Th{Z4tEuxluZ`Oa#4x+1T
z*H}>SBzx7^8_5SwpW4R6#N<K%aH1UwdHp$R5cL4EY1ZJyCX|$*ffc82gQy|E(>%!%
z|DB~T(x9W(0Z!q<ESO_Csd~!5*Tche*bXYM!O)&Fe+8D`9|5gB;?uIZ9jyp%xrvwu
zuqbqoT&hB~Cf=)b$`O16Y@QkDq!t@*p6&&Lb~WCD&mmv$wvH@kV;h+Q25N%!@AWOL
zLG?{`IR}`w4`Y`4vC~r;b`V3_cRc{;Wmkf^7(C~3?gQN+$NBk2A#!x#x6~l1ns1PA
zt`SDv`J8yC;m6W7fLbeSm$kBT3)g0vK8Ma>#`;{P*o)jyNUh37I?lmX*&!!=6$9<j
zk>1Z^Gz<DS;qzcrvC#NQ=t+6O47<~jAY&JG)h2&%_KG*m+^+L8%o9?5)a7`XwI$2!
znz*6yR8R2bfN%ZZ&<YA&hS&i7QCL_a{>zuk{JQlgcrqP1<n}Uk*txqi^_V;N*VR=^
zwf0SRV<BSWR3&Rz2A4uFtMfU-74v2PX$~5qJ;BSWJ_@-Wr66XL2gQ&yJ$(VbX`;gx
zJqYU5(;hbaS0dp9wEE&F;Dh_~TMXSIz0+fEBr|Oq3J3j+x{)kZbqjh@PX=-EsEeXH
zxGKCS)hp({vo}K>!u^^0Yy9M$miz{-6QntECOihbgnc--nluR<r-sd_$9lHoL3$!c
zEa+!QFMCO-?*pVC683;sXrQm4jvfVEA-dCbGF`VJ>D^chBr;X_AWWVyFw-{8|MlxB
zNU92QpY7$>g{a$Y^oIa9T?O+3^B&7QvghQLv&ZtX9~L-NLqw2gG0uAKIXzNoNl=Gs
zt3ZN!E+uUx4m#a3<Z6SQ8dP~Y1v&Fqpd0iGc-l#0f&9>g7^^D8-O3A-P`l}R^85ts
zUh5IqY)>Rhd}COm+Y7JDxOV>7fZ$^QUUH_Szl^5i@DXcrqpv#x5_~DRNK6x$SK@s1
zT6-q(#ZsAv%vth@j?I{2*HKs#LS9n=XfO<`i@<Qb!Xs#iy+v)x1NRP;>5ca*jl)z_
zt*5`#RYTOC1Thxmyw49-Nxwa&IOL-<<O-dy!;J}I3dlpU+p;kkM0uH<#UajMjOCz$
zZvLL7r%$UEuz+Re@-`KTK;71UJRVRTYQvtGgvP?VKZJz25eAvt00QzY<$)p;lCVp&
zj@jef%-lzetHqGK0lKc9LOBBsy8!XGwuT29#Q|gWo8sm0r9`NPlO~&XoII(phJFG&
zqu@ov34>LOcc}&`q0albhp&B$<-vmI*gfvK-(8{8tCug|hF+K>Lr&!<n1IV8`PmWB
z!T@SO>`Do&dT=Anbv!5^Ea*OJ$;x`M*`9{z=-4)Gg{zBT0DIZ{ZN4g?Wz)nKfRzt-
zxSRt0z+12O$4{YAYYb%S9$Wc%OS(KByEvMDje7mO4PLgQKaPnfFi5Xl5lUkv7mt2T
zU#Q@lUDICvJ5BbT*mOAH_b&;6I=r|yHm{$8v54>nLGTF1W{K!%Uq{HbIovxv)|O>f
znc6F$-$h3Zz>oOso=|7x31rjyaZqF(@g>XJ-t|*Z=mwNP?vfCrku}y3*S?TB>%(`U
ztE<aI->T%*@9UN>BS6uFdXX#3E}35zfR<-fKmbPS=D~qgPtss%{o66K4a!rdb5II<
zuD4LYfa94EB!KUwX_u)E<SAD8&RyjB_P^6;REs-lz^Hft+|(v%z=v)+ol9OEPU!0&
zqsAyeKVieVrF$5XnPI-o0{|tC|M14h=bbVu{hBn;J27LeH-s>K^OG8DE{`+pu0f$#
zF3OIu#`nT(RjF<+jiqA#_zXA?x4I}3-|7R{cO4faJp?zW6I5W>6ki!86D5QhfjE;o
z(-8)2pFhfE)b;t}{hVQgwk+D5020BXNOtuB-kpX<(@(vaoZJ0X`k6~#!J(~j9b;e-
zKGH?MHf7(CJlF{MG#q1XP5@A(SnAhLw`9*nKnJ5vP(B1ZvKc|5mPQu&5aqV--EPd5
zx3uVNR4H)(3QF3LIF)j-YWNAmy`o=N_*zybEpu_e-;S7fHPiu1GC$VJFpk6Wbqp*4
zCVF5@3Ia%@$}~fqfb_Us=ci0+Lmg;p{!shfiy%}yk_OSSI+_S!SOYaG2T~FHP(tVh
zw-*`A?uW5E;+x^{QiHh(;AA^S53#fVVGEYRc8f*<!l}vC?Q?_fJBibPK{1@pLP2r#
zh0V#^`FfqccD)|zSFH&xY5Me*B5GNOKL2_M%e^PwVelUb5PPvfdNtn<qr}DG1Ta$d
zz}Th!DwYC%k^42wu)=zVF<4==CEdm4hn0g9NbsguO)r)+v_oD;iso4ujEJMYFn3sX
z{=(+~tFcTS_`GiYt*h2+07qHVemVIb59}b}`Aaeh7wf(Cqgm8N7aC}xaL;NjRN42T
zPHWP=>}Bm>YK@fNCt~8ADfg_M&SXQg-xyI*-u}@i<8cqSfAlDJoVB>77%Q(Rn1Pd1
zl*GO{mZj0n;~nkv0;lze>SbxvF+QbZQQkq(?<UAtaeUsIU2#H!1$UBoBAcq&cV!jc
zy{33Z#;s7k?7CuN@};#+&qNch`$Zq;xFz$!+dpVrK+-~{J}=^+FkUR7`?@yoo_85C
z@~D>grPIw3V-VxNe*M~{K^dTKIgl4fP%pFw0r{TmnRk6QuC<{Dr@X-e*B@vjAYg41
zpcLjbs@P+)<AX(^?o~z`<yb~-UaP{PMa>bl559gU6n8nco|nC>_#x}}H@lqK8pp$<
zLomArTc3VPi|pZ$RSZ9NZC7jJ^Q+F3Me?34J|3cr=7}sc<rU(Hh*qhTgHuZze9DWk
zlfUK>6>VwY_-*dd`Qq!^(MjH$joM^=rmA`)#?lQ)!aND^GK8sw!5zpFUz_zodPZFd
zHiX~0Iv8S;Pi^VuqY-!Br7MNSwsV&4)OI>4=`(vSb7SVlYaY-u<5dvZ$RMX!*ty$Y
z>jCeTuVwIHS;da0Zc{Z$7+xYHWmk;ua*7K%$hJ*(;KfF5e4`)2c8)tdi#sSBe$2U)
zs<6I2R|EiIf4d$u1YK6Polv0L^kTEDpMX%(5+ow0-!(I!Y*E&3A>foc$C$0p^VZyo
zQBWG+McGhD#bR*l)7Ga^(gp`Y$~CVKKM#-YGTNN~A!YZGhrJYX8GX87b-x+A8%;O6
zu7T;gb7^?At-8sM@MD=934wSqImNW|qc)5MpBlHZT=OWrYu|N*lauSWU{kymAtEr^
zY^c6dG!G9E*vq1zo2QL#D@!EPG1_j;ji}8sE4RO;kz*MjIoBBdw!ucyCq(vY_>pAv
z5y9UCHKX77y%;dr-LM!O6@7XijnQ!~+D*3*+C^>hpSa3;6VKy-J8O3a_H-vH>R6B$
z<qRuq<GM-+EKE!^L)Zqwc3S$fy5)i<0U82(NU0!i8&0RAO#4zwvQnV>MP12#b4GI}
zP>O)pr!lY_Ypo4w|KFw19q8>i2!tsWf566uLy(FbED@rGb^a|uVm8)yb2k|X2w9z(
zVJ+#!SRd>+BhK?k=Ke3sQB_Uw6CTt9tbR+Mp7jF2C^S{0)joU&lhotDRK9D5T8C3R
z%^<v?n;$3de_Ey9Mo+P9D3`*qYj1C$-uk@)-7A$l|48ViE~)0a;yzdE)1Hhp&*FeF
zW!KkTo<G9CMPMp049<JE_*j>>`MO87jKQbt?1GxOy1a;}GS5i6>N_f5g+sIuMZ1y%
zZN{GEZp{pPPEOps1#j(`<!6#$#JgnU2M<m!+EdvNo3SBtvTxpO2cF?a{`Yy-18jmc
zrdel1uYn@fdiv1$;z8Yy@2~V@xq<WKk*B|phkAlzyVf8CLiusfymb{JQ?!I~7<bQ^
ztRJ*`w7cOzaMX}gKs2{%HqLqSdhdzo*>l$vPtoSvx(c|433>LZ_%0JoMK*0jBQfZ0
z>r<Y{ymvQcJt<`F6_&aon>PLcR|oRloY%i4Z+!5DE9meV`~A>bS4ze?s3hN$JY}*Y
z$TsQ7yJjdAFy<Cd6)`8QYZV4q^y+w=;?d#fh(&@;-6LHAMrcQ$RFqs{o_4g!@yZXU
zK-VaoQ;{{_4VtfU#qLb1_e1zOA297LbEj5*PZ@sze|JNc_K5W5K+)^)cG(CG9^T+8
zFS+RWE&zG9ODnX*P1E?Q0S+`1Skdv3W{m;VcyCgKXfb#-Vg6_KLCb*do}$4+%<C2#
zAdey)+MDV4jIJNV#&)6j8UAj3#<F2%JNCmcKffx6_y)!AhB^FL@t1uvd~imJRW7{y
z{3^xTP%Pjvgfw@OC@zBXJ0TE0i&(z{-GuRURdiM!gr3_3Q*EtGfD4d;qa~6{OX))9
zjA%Doa<$q8Sw|~vo&mErd+vwX8#oaS^wy4cS}}WN*LzibwH_b~BR21Y_s9uwRNvJH
z;TJ^+zW<I014VUC-MtqG@t+)ia4}jj28r9d8a<ES*auvI1a6HlyvuZXCQzQdL;@cL
zJ&F{E(8%!9ArwSXo_TD=A;9ys9=L`fga|)o^r`qxQUPTb;Gw|rBQk;rA*k^3?<+((
zDN#F-fT_e1ACW#gc2CuPgCen$l);@Pq7ZduRiK!db#(;b7@cDfNZI~xkzy^p`3&U*
z;pPj&P&ToElwysn=FaN|2j;6)z(ulKyG6<ey*HY(8w}(A-y*}d6Za}g%H0|dALgmh
zc>=E{y8smcNP}R(_Y^nm09w$13JDrPy9Y*R3DH4&Oz-duZnFux9vjbA)PC6PsJU|3
z9eG)_!X5}^lA8>pj~%li0f!pQX0oHAUu4{lK6c=wtYTh_#(0N!)F!80WDP^!W<*X4
zr<YlTu*psyJU2x~mt_^$r6?PI%>EO+Y*kg}(VbM?0GPl@fJ((%po8`@h^>ZQD)+l)
z$Vnx~kFp<n1i0F^gAPR!tE2w?9PI4u>gafMcTwts;WT7OZg9Gmv`Xqqq)0y#sNg3s
z**PYnRg?|%MFRAduNPvF?i()zFGAI8)B9$qk?aBagieRGr&_Wt6lPej*M-Y{YO`3F
z>e)T;9j8M+``H@`@ZiJI@Yy92p%Sv+C3X&Th8LS>MttC1-W~t~d`=q!6bp@m_!r3Q
z!z{w{vrQUu($XVe*n^=d5sE`%#$6q|Sj25D<t37l2)sQxT7&FkGzUc~xy-!xKY>W(
zk#}Sv@o*BHfCkzTCwe4e0SvPfop<KRVy%r5bYRh0;0m74^(?*#`GII5pL@X$IZQPC
zl%R>hfrAJ6b<&ynb$5^TcSG`~Chby`EG;9l%cu@=bR(YQH-Aaxym{>w>zRcR7|=^E
z9bOsOI}r^qs#Nz4cz^wHE6c5v`-~Hx{_wah@Inz_5BCb036M#ZH%|k*XmffZg3_eA
zjgja&LnLpqNRRrD5Br*EF)qK$aGn!fLscN+^*h)Bn{_Sf-o(#5<_g$GwLKH#Pn;Ky
zCQ)&}F?MiEpx<fJzKc}`34mI4up_cvlzcO><B5?4paRu2>MAL}U<JHdko7hf55gNZ
zs>%0UpcX_JN#5dn$|_IN`5wTLFWn3q!sTq?T`}QD_nV@wJ%<7m$Dv2)$wI8f;G3G%
zVMS50ZMS`n!~i3Ps1g}<B{>*>P7un3lJHu{S-VK3xFI%_IRNMWZP1TlBT<gPAuc-n
zki8jMKvU|Kd^%^PU&Qy``-mXG3Ou2*50d{o+;c!6Z!V#$@Dmi7@Acvi3xB_lfrRrX
z4?%Q~q#;cv5G96A##+x^E}J!nlHF#s=1=S%Vfrhkg$lW~GSbqUn@m77pvtw0IGXaX
z`9M>PocDy=?%~zKY@~R#QtqibYt1}nY1-~%o)<JSD8{P_HQDGlC-D~jLOfN)>!v$w
z=9YHW%)eo0+cjd-oN1~TFBq+!y9ltF)-QNUz9JNY-`gkVw`L|!H7_Ms3v6hbAE-UJ
z37-q?2^F(aP+4{?U7dAOeMA6}SAO0QfQq+ZrYp~M?j47BGhk03MSa$xfV#f;^%L6;
zuR9BG1=Ya=0CijQ4Q2fZL!IXZublZ$GeTLrv`pIh*H3(0#E<THwbProyUjK&=#2?9
zYx&j^Y#5v2E$o0{uD+^(T`{ZGfa)>y1oA^IOS?>D0bAHpI{s7-3-6_^o0VcbKEGdw
z00b7Z7`wmu2S|!=U_%YOmtf!e8E#LR!Z`EkxWcUz2V1$!$J^ES8@`8^6NH$9L>B_S
zj;-dz)^`u;H~=UKKDVB!M6KOQ=V8VzS^v}A<Wa!EwW>UsPelfKnjh=?0U07sdU`FP
zy$(K<XGhpsq6X~^svyLwhF7smd{I${fi4rBhJ6pM7N*>OqQ%JP>{X%pLlItk-Y;1D
zL5uJ>s|oh{sSR$O5S*@?0(9({vTJroB`o-~U5)n#n2mh&Rb8H)2q&#N^nmk3+60_7
zP6Q;i(*RyaDk|}<ER!Av)SmNZ%(aLrfNz}SR|~tnKSTklle$e%uh=wX4mPx6A%bx!
zW3Pbi7@+G1@0-GlD9WdHF)^iUmplB#{gKCZY=SCzwywdQk*`<RSr$Vq<&b?f*LoJp
zonyW?L5sD8S{5b>=A3nYc_PYR)arguiRP|pbC<odjUS(6(+xEu?a>yYfH1?GoDbCZ
z0y2oS5J5+TS}k&|#wPrkRW&7ewuEsaoK9Z+2E3Uw-B%sNXM>*sW<eHWFf*|232km|
zjIx_j+)bq%{4gN4;Hg+oPt5XaN#%hKK*|82sGyM5M<`3yeYxD1Yq)36(SuTJB?X{g
z4Xar&1OU3K22$Lr3+I6*3(LY7Xtha#^kNh7(jSbTaZR$$ox-y@2j=y5LW2)vM)xHA
zfaUwdtmkvAk_8P^vrYTc;PpLGA*)URe4vY@*&js&NFBiotgDm%;8$8)g8Vit^m?^H
zQEl347{D_(7LlQm=FnR7d&!v90O{5Oz<;lT&9t5kG^mf%1XKGhRnPrZP3Eoic&LGp
z7H6%xU7BbaSWi6s`U;$$%N{-x;8k}~t{9r=jUsllg`Okdg~AS)L@290p{2LY#gNf<
zi-({RovYUb-40`4ArgzMga#|5a04x)h=f89$3;Me-Nb#Fj9EJylr<dZ8+mc7J$VQ1
z2gstK1b|o~#k%#z#>Rk(RRo9(gHZi~a2`1Z72}6(i?d)Zpu0-kdgg_=-5cn-A$*Zh
zgSZ%8f=@q3Hfo3t)2Z6`3#x~R;D5gU=P2+(AAtE=Su>%4vqE@uA`R|?YINFF3LbB<
zX1T286z<%_-DXPjQXqfJphf5wBN{gWtj#0)TG|(3y#e|yOVCB4O^!LO)?1Fln)WMU
zH{P~df5nJ;x4u2IY*sBT)qaL`H`g|?yU`zcTgy=pN)vA<bmZ}H$c6RDPkedyLduVh
zmF3EYa*@OGfjNWWeoHa)YES|6^tEmsr#X^AwT=m%mGN5j<^7b`66>k_?|S-C^>9!z
z4^v4o=NAb}eC-W5^Msb9Y@7?8t8<Qer>gy9y8QbzC2~-ykwbn7G*XAU)Y2w?p#MBE
zbs2h5ob?U57W$bN?O^eD9cxZ}qcVXG3HUbIVeev=e_tzw`e7^$Z%)oNcn#r>pI_v)
z2u~bxs7>5dp<DLDYeeu3zrL!he<?uO#eRUQs41Jhox=NlIr~~EgEjEf-TG9poXHMY
z{V^S^BE|e;5!fAN5qK=~ustAfYma154gh;vIRIr(@Q;)O80DJmfMUG=3C)XF;kG2@
z%#D;?tz){KLJdSg2t%QbeQnWw8d$l9BOd7jn1>EjnX7Z8InG`ViI3|j&#0w+<p+Mm
z5BUY#LpF6=i{%vn3siqc!V)>y*CuXQWw-uI0ek3~gYVlZuU8fUzWmiOD13ltv=s?c
z$hC=^LVaWrI5)Bkg{0{ef&qIGf&q6Z1d}j03k~Oq4eWx{%M682xtLKQDC(~S_$eMg
zNdOf2J3jwQ#_ZN#p(8EG*st4VWm%Dc|0?rVmVwPsxCIZOx}^r}1^Q{RfW!j7DC9>r
ztAL1!1e`hl(!I=$@IN3&@F`%Fr^WxwdHy7=t!pX558D71AZ)gOEQ81ji2Mqh|A{}a
zO^PtJAbRmn{8?7t+g2A4*8UGj9Eb|6g2??3#raE#ThVt4<v|>m3A=n`I*MddBn$3|
zS*E;!77OSjn2{cio<Hc1(HnRZVp)Sg%|ZBkS$+tm8~tBcmY)&_pF!SL%G{WL|KN%}
zL7sywLD5qx#Bjm#h>`-^DSDb;l;vVZQHY(YD~gEx8}bcntyl`!@`@~h&7o{#*~I<J
zBVRE(if|bb>yAUoD|)#8uZ|2`Sw8UJm<%aqI1+`hAh-WuGX59O(^*SRb@^nu+_GVZ
zBZ8u>P>v$pKgf@80NF0s?qA6NHx}+s^7f}Q|11504N&N+zw8j+bxaMcwg(g*QV8;k
zJy^v<d-sZeG8v$T%LapDp;u&uVkZC3ycfkPZQ_QV{Tl`lk#WSQ{!4*drDOR@UjHO;
zt90ydL~hwHtT_6A5V=2jDa%JcP28VY1h)7Orxs`g{!twNinBvpCZec_sy;L;{U?Dt
z>;nzq6p>qY>niQwKmVV#fiE8HX}bJ7?fJwH1}W4p1tfUc^pA_vY}>c@)Hf=@0p)k_
zrq|n_$J^Ok+h^gn{L&Q_)u*T{oe7Af{P>cPeFH@N`E|!*xi5d^Z??uDw(iH5y$XZ`
zKmG);Ae19l{c*n%@M6{12dIs#`r`|zr>^>=MC^a-(L^@<*ivTMR0YjEwn5-f?Th<2
z)pxD=p!6=~<c%Y{w}*smbK0=`>89P?QwpJLBwx8i-F$m==gYml`h~jA!YWUgqxO&A
z-LUbJmrqq<e2?fMd6QSB0o8m#xHO?m(<<w>{;ULo$jUCy|9aa0{Bbq!|3i)cP~(R-
z|C_=faXHi5X>Q9EeX0i9EBbHuN~BUxr5xF{lc-{_W@6U{`1|Y3@OM-t0pl`91Iycm
zI~=b285pvhkO+v1mh(!dzoKvN2QQsA^u0-^&!8?AS+{N-I6Vit8YAjRf&9u~PQ}#0
znMP_VgOxDkqp-a$&F)wTQX#-be(XU8JR0#E%W>7_M(PHsRBHH*5(E~kyPtt<0n8SB
zUqP~xCuXpr6G(0cf7rtGu5-<G<{v9uM*^?gEU@pD6(|pYpIVhqQo$T9WDm?MF-YfB
zBlTWxSd_BjavYg|yN$BSljS4}P6PWuo`USObtMOfe{7(IolyQ3*~u3p>gZMZF&s16
zg>qU-#JfV26;F!<28r~M0?}LGH;TW8Jrt}%{$eG(h!g-yeiVkS!$Qaklm}3Dx<W4?
z075U=gWxhfBnrYh*PK#?<55;ba%g0YEXpdhD>-H78aNK}6v)}BcCUErzvhdP`*Mi?
zx10gy03i@Y{MR{SIgAAE@C}PB^X`cmFd4~<PN}XCWmU$KXn>Fn2t$cIfGw-eE47kO
zCej1dex8@I48q?(^KRcb(1`L7_{4I^Mj;uHo$?S0mEa+W2*F=0C+hFM!G|bP^jExt
ze@D0vyoKE^%eo9jmK$hSWEohvxpB)f_r6OIC<I6@BKRxhpfL72Gf-uP9QF>d%*sCe
zRwCh*!`gRKQZL}`UhaQXDMZH*a=>>3pfnWTA)18{ow9;N>ecVL+5b6P{z44@fh{Wr
z!fN;*PxD_e5PzdGKco6TvW||Tn5%IbiuItVE=52P0Rqii-bcRNGGqZn1#bSdAbYtf
zcIa<~jH0;|;4a!_#Xul}h9E>>=nzQ{{Vt~Vx{xyeai@P4G74)c=4{1G{TGDnX*xaC
zD)s*>K?LOFUmLR@D)n=(ptt{?N>L=8Vsw8h%pcDJ{%4gb1O8Fea>cxVcQLDY*TJ!z
z-2>Hru^?N#Rum6WJNQMhLX^`|Xa;0=priPQ?+7E}gFrw2N?}qVouL?JSZJB*@BqqA
zR}2Vjm16uUrg=qz5C=k$AOz}zGyj8u$d8m+`5EfJ;>%A1@h3O)gERjPWA}qIMYq-i
z!Iz!PpPVZA!aukwidCn$b_$MEux{0PPYglo0lN9uF7=PR>#)4AVygbpJp4ao>>y}D
zL_~sqh~>1P&+j{*%<gO`iln>9tD+2nR@%4SX4FS%hoiR`xwDS%+(|rdu*N8d=5#wX
z{I|yh5xeCt`<I(Gp%49%89x2>kN;+TxV+qde)vC`#eX%k_>bfO1^$1oI1-l!Suza9
znNLVCDKd9B49F_#XXTjXZ;2A{g5Rjr6u{=-EhFA_9EL2Sn<ame9{$_WSK|@=?UD5`
zmAxplS(qia&9FVQ0%T_PJ454lx=~UM!O`1dg<6U9tuML*9zR|Kq5VoC*AZSS_udkT
zJOx)UKnQJD`-J-$E8+b2>LudMA?8^f1X2AtMS>jS|Ee?ocS62)Vesmd5athA5+&yU
zo6zHm{$B^UKQq3+4qyKq6#iZ2@qaI{SkA*%Wx)T(S=7(7<eL2z?(oL|*C^+pU?-XF
zyTq<UpMSx+zk?o@S6CH%Kx~NwjsR{z-iKs){{o!&+hqMOBHUFV%D*}cr1}3=g8N^|
zJs{*?!4y^$lcKkjC>l|2_(r1W?-+_9l#D66prp|daxUjUl&s;8+5D<l{D0O&d`D3z
zs1r2;%=`#m+lBubclgtJSK|(_(!au4{wBfwZ(&CNj63|qjA)1dnlpbA!`0{(<Q<ei
zd_`23i@vLa_5B%vfpl;c`%-$lwSIR|5lmkb#eY4>ckjBa?{y`0@xdhmMo~>I)2NfU
zxCl4(ji0tkB8A^v$TY>}%(ag-skgwDK+`eGWVnk2$E{AZJXZj>*R&^U6%Vgx$ip>T
z73o4C`wx3yFtZ54Wv7S58<av@<MUa9&#5T5H;?^-tL@Y|uI%)5xF>FI?_MQnz)Vd{
zJYkiJ3@mBQG)>bdoU<Wpi;?`N1k2ub0&WkQ>nfy^k&$`6r7J4dAMUj?#Fc2z&8U7&
z9j+{OAD2n*$TaO&T_}iSq@Y)Sf@=_qwt_2$3=Pbi7dmAPFBPmOVRUt);T<`18m0Cx
z@R~D<hmH6+Ci)ly+fvQHadvM{&j7e~Vvj(5wKs3*+^CJ9P5IcEI99j8Nfw)%GE?Q`
zG-zffqi$<Vx+c#R7wfO4BoG`Kc|`MLVWA|mpa33kuEGWKiX`oSRKK5K+ULXvb#w(u
z_??W5OOtLbNgqbKi`|s2%Z_u(UCA-&oifaOFgzXct;OxQTGl`rnM1RZLG09AYi9RC
z&dopQKLyf){P!n@`XV?F-J5tEB`GNh0<!y?He8L%1QTuY@D>gA*HcsNtlOZ4d4z4c
zPs}^94@f;cFfV{ocBa%y{{+6QGAq=odA3WAhZ*U!f2)Q&KCL_QZO)}mxgTyheEE1b
zp3MN^+!FSpeD$;5@*r6iQ^f|FwzmFh{nWou;h*@z)`B7zm!?V)xNyfB1Q+g!tf@~4
z3)Ak;9$)Hp5%i5Gc068}dhmklzAKNQKa`h+<?wp!2)x5C8)%D5CBHA!SejCDQofZO
zF0NS#?3vR2H=bEE(#*%#md_?;ecg^!=7tilX=!;qSe!r%T3VW8(9`RmMG4o|B<fC-
z$H&L#;60jRqvT1C`S}$$>KW*_xugCKIp58#CVWjn!DO_Mu<fAKDS`NG%gO6xjUi{A
zB>}s|1cJL?Mz7~BS5}hkiE;yjcI(k{PY=0)5mEO=em9t5)|*h3(s#+dEIvnhA@QFX
zzoEP1W22s78*1|Nr@9EYFw4~V_++?0ZczRa%uA{xyYCnsy<mW~n@>&1gu6n{<T1<1
z-P+#R`Za@__;0Ajn^ZmKyg8mTa5aJFnY_L%vn@?<bBc%DMbaXlj*=?jF<fmjKA3r;
z#KXcOh|pY(_I*%U@SFh!4X+~&2mT2_k~p+$S6xDT3J+KAbf_DCOKCY=wNK8fb*vUA
z94X|)l9yWS^1g-3@oX+F`~TQ`6L6^aFMgcsmNwi{sf1Qip;9EvsI($k%1%njo-O-~
zRw<=YWSt@;F|u!CER~%QvhVx88#8A9=QE?_s;<7z@819O|3A;q)04sH^Ip$6?{m)k
zoYzr(^Ga52v_Qyw(eJRs%ivah3ya|j<B8rXl&A`>uEGYgeq-9|WKuAnTT%~tL@4O#
zHHRzfvgPHYh;8Sxd~sw&Jf2JOH(26PT*5%ILr!{9(p0TE)nJjX)8qhGUb34_@4AK4
zzNAS)ZKrHK)Qq`(rM<nKo!!g1U4CiuZ|r1;k3nF%LuPuy)rN-z%BOHo_NXx-?Uc*U
zh1Ar=QcJvzOteIA7MPi6`9=33)lS-eWB;D%QY`Z;mQIi+@WO34YSz+TXcRS)dzi@~
zUOBAh-rc*?PT{a&EX^^swH{k`k_J)-?<W4vVC$02mn>aMbxuW-r&1jq#cz{U*Ncsu
z$a+eth`@+GS2Z@ODeLU5vn&t*d2st`JKz_{so87RhGm|eU+QEcz@$AD(~VTLya@UT
zNEbP3N7mEM3f07AT@?`CU)|X&=d;H?S5@cC8J~wTmyK#xnEVF&7v5cEql&BK5ufnb
zIpb9teC%RM2?j*k3?=}W=i=RS05o5_QFp?3-}vGaH^Z>#?PYTp85Cc{d;GV1l6T75
z+G2WHQ#*`|J1-6<);SKA2-kO_(B#Wh-LR002BB5eX~Ym{7plapc)aM$88#V>IR{2c
zBI}R-<g(=6C<|wu4jxgnJ{KLnW~>Tek_8-2?*r7+O&i@#pB0h%t?@lPb>hSctT$2A
z)Im=gr>rg}t}NIz)XWz9cBHIKTzKX~P*4zv4ZXQLGZ4x2gKN=cB@|98?Ubj0aD!rR
z=U-Ve@||*edjEZi?>^yX9*e1wg6!XHNJG<h(*8ab$e#OSZ(JhJiK(2G4i19)7lr}c
z_T=$RadCa_Z|Su7Pa4*S08opn8k&U!1t}eV35@t+sO%TOk0i)-?GylTsp*?rvOtKu
zQxmx1cQvHRnaNB#emE4a?7(Y_=&g0`N5Y>s<-iG!lLn<7`7H}WCb@-UQpeGi00*nI
zZdkItai~m2oS|}T+?1*g@$jJ)FQUvW+$-qeAEHMOW_rLOjeKo*JVR>4Py9gz4o*!e
zW*rNGczqu^Hx~%&zhpR*dKC`@`zk65rbc_7c1J|}r1UC0e7d23MtopRY2}3BU2n*F
zDuowaxqbaY#(56JiEeb;b>ZldFTK+Bet31)X=b(r^k5JAL|!B$g&s^%^^=;Z@>#(M
zxKbr$*deGa$+u;BsJ!303-Pp`=vObu$7Q+i?}g>(^Eyp7pcNGrWebWw3GFwl`wZQN
z1#0+)5Q0FG!erB&l-w&Qe+fdxp!#Ml_o+t9T!kK`cM3h{-zmJ_(Y3moOEDKdwD47|
z9ro;rJ5&AI<V?bAr~21|v`^a??9yF&F7d+FTwzA>vFq1)IxWmvSdrA<iWj8oMiC<m
zD%9^r=Ig~o=AXS75pzaNE`i>}c9-KlJziugAi=f;*j{DrSYV#op8%sXz-m9mqx7eS
zntjlNNfs9_yqRF0KDA=7ds&TCj$_J=gp%RFz~`F!xqNMWHexM&HhHv$S-Q(P;}|B6
zlA{G{V2G%$NO!xE_UFNZnoptY1)^brcGRLyTeG0OVNwx-$N&KfK+h6wcIIGb;6}br
z+N_>Ml0}lHXNW0hd>nlpkw)xOVY~~Iy7%sVLm9X~Q&v{yAyp&#jH9pTfZvc1zXX-+
z+czU%Tfa}Rd=LRF{Nqh!_><AY@;;$*%Ax#L=&n7lY12MAp5)#`{s`l=6ex<{OCOQh
z)Jyv|Udzr$enSXOf@>bK>j~Rk^Ev`M0>6FNe44G#FnyIcLL9usn_s}^Ol)sHnBEtz
z+z*1R$WJ%FL03gbYIpgdfS@32v__axG26b2zI$K{g~S&HuM)OHZ`&(c_q-lGjPx77
zx2g;_aJ%NJq-M{KM}D~h$IP;jTx~Ai0*EyD{CYjFA8qzuzMGt3^<?bI&f(#CcED!D
zcCe?avKVyCt3$XtAmu|V5RkQMQZO7Z4L5b`D{MrnRC`tOi7#kNQ2s;=Wxck7wm~5d
zT1Rcrmw^V7y*e~}X$$j*s8gAQC42H{uKd(>$PdoeN34Bh%2o<i3g5m10qtR+0t^Bs
zefX%dy!`m2SvS<9ncgU;zvUfcf@!Tko&cz1;fo1SP94)RZ4_`)4E5=WVg6oNicVL-
zU}QHFUaNQ%yeeihh3Qw7Y01XY2(jt5g6}c2O}5<0ixHx*dA1WllZdc<W-?$nVEFYr
zLb(fbg@g!cdihWCz=ZqsNku^IG^l{Bpp@D7y*+lRnC#&=z%R6Yd(WsB8k7(OTXm{+
z6spwOb$x8O$Dt6SBW>IpcyqtB)o5kkOh2ed?C5c*q-EtCe=5YE$A~ssd7sh4-ipW=
zHP5`xxv0|Vifo~=VN?j6->ziy6>S!MB<?{^@jpOI@sEbliA?UDjQje%8a9%t%^R$)
zWt?NmT3QHz@nG%bDRc82X^v>)D!;`zD8Ej9iYGW%zVU%|d*_C;{O`u07<`nKGsOTj
zH5Sf5!j)zjDhHUcoQs8~!ccf$JOQ7QvZq-UD>ho9r5mZt-|0%DU^(Ru^2duRG=6#7
zW3je^_yWkLUBlPdkVe-_>V{4L>9iGtBMJd!VfAc>SQmt1>8$+aJHky2_H9$c!PzlG
zpyW=Cm3vGl-|nyVW7p}4Ow=i8z#;&CyyPCKa-QV)|A>i3azU#kI#vxLiN=QIp;F{4
zMI-GtLijNf@Ua6x^>_(tk7Y1H9WDNDjCIh<w#tr<$Qku)W1Ok1T>uKIB3F|Dt^u>R
zS=IJxVDWgxfv1+}SN!EPIxl9^J{nJ>=x}at#Q+|J;-f;6y)FU6Rlm+)!S3~_lNkVJ
zLjVDKcBw-EL)O{$UDb8vb>&~bgMfOr?BTJ>DATFQND`J3XH6VQ!8>ZEq!eG8=7CO=
zM^I{Mc6xB2pvu`xq*8lQj6p?vJhVjI)~p0pajJFc3bc*wB=Nu7Mt#T@x9TWtNVal3
zq}8HfYKpA@iDY0M6mMfY5`iDd=UIVvB2Az!3A^n=HLEhK{YUgPCTdlR#n_#1%YDSB
zkieH#+)*U7m25}j?Zr!1OZawAHdy4`VTiEV-cCLZ=K-=${yZt}MzkGKk`xR+fI`kw
zEm<ppBAST2#wlgUo$OWz=YkXMc!~jy^d@q8CWj@5eftjkwbq3)-E<>S_M{o!%sL3+
zo+Bb=CUR5NfT(xLd}cA#=)=^o?SD4;Wgx?LWKRz}!D%FiL-H+e^3+8)6Z-_v4Qb56
zGYtT^``>k52>t^wSK(ML9&3fF?N5eo^j1*~CeBitZd8EQV2&2vobex>0q4$vKkBTK
z(qn(WHKJxFV8KXXoT{Qk*_k*ZViHFHDb!#Z58jwH=l&z1zUv$uILQNm7L1{$I#W!L
zBL(X)-5nN_*a!YBAGEo4eG5axOf#!vtJ^dWP9dMKJ$vK?@uicJ)K2UZg_De9vp-PP
zX}{zj(w%w8zimp9yOa0ms}f4=h#vj-JYEX<n6py+@KQeC#RzC48rtA@`CL;Ym1<@N
z#kJ&}DSa;vJF=sD^7yd14KZPIKK|X4A|g>N(R3ZS>2@Be?k1H6?#{3w3C>4_V);*1
zxV}+l6qSryj5?#N9Dv`a;IHP@J!#h1u>KtAu|tHu`d+JwTRs`X<Jtj+1|%brg7tKY
zsoWnM-)fyd9ap<abosj@Ew45+Y3|r}=ef)){-b|t++A{q|BsUw)EIZXQrlq6xvfg+
zv*qFq{N7uR9{R(l$X$4uXWRPRyc4<hBYIM}1CFQA+yMqoBmRR+Aidh`&};;NA)u2A
zZ5iY}c`~qn<W_wAG0!xkvX^f7$-*N?Cbvu{V~7RC3<`8O!!BNHdQAAk)lZ*tKDK37
z4wI^BU#h9m6B`@LjyGz{1^n-hw6ruCRAR!!WtT8I%Az%U3O(@r_&sM*`tb&N3G&ck
zHyuanOz;Z)c0Rt&o{A3*c1@w7qbJapjg1ex&wZ~enVs(ej4{xm$4v(3+~U@<Md(e0
zcr_=RmZsz0Qib)xm#iOiT&gUmTQDh$>*+ZCV|DNE8Ne`O4XE2ojb&v!x7s&n-q?p9
z&Nx9~T3aGHY4X#zw2X}}jF<(brR^*=hT2k_^Oi!<kYE_$M56o{yJ{5K`RA`)xS%~k
zL-ugfNp&W5Y!T8*G?O=%7^;ak=bLr|2vLE&rNX_2^mu{#o3r8e4g;Ux-pEUJTgqX?
zj-SYV<a|O03OUz7cQG*w+|Vrqip}<)z894;9lK>;<_)$7r$%bLyr#)yvJ5U@ui<&<
z(tT}tD%k>@vhtLyLWf!7O!Ngk24urC!c0F#*P;8?VOOK+5!bV4@ZK-|)Y6>{N4yoV
z6qTSVR&I%;`1p<eY!X-do(W>=jBzs^p}3+qb2sLE($)S5s0ZpidVgBX>g4puG02=A
zk>Njh>Qph*Z+=Wj3p#)1abL?IrqQn<%+xTvq^PKW#NdqTU{ZI%4@=>)>+U_CR0@w2
z6oRwH&eYkD1}c17Ug$426DkA_-*b;wyoj-McEC=2eBmpI`S@_-;V7--m3?y-)MeM%
zDDuqsU<F~kawHwpr$`oX3dw$24t^NY*@m_T$78$>uD6!TmXk67x5LaE)x|-0W`A$+
zq551he|KeSsG+wpf@fb(IxMi%)C*)26=ilAR|+Ri@mh8Al7-HK??w#IrloNJSHKfz
z)ycOJONXQVDBndlqVK>E!lR>yrd4})Y(G1m$rd0<Qq@MkMG2V?K~*}lVR;ersI(o!
zSRhlM56qNq1)eVQX7ofX)r2~BNQw@@@Z9esJzL|i5BFDP4>f16TUR{Z<6&%;u6Fui
z<3*`xCt{f<20PKYbKAD#Dx2PD#Bb7Oa~f%TB#3$EyU+ZH^IRHmhue>asRg9>_c)-P
zh&Oc~+c`-0?uF^#Yd%LNYQ(&hP;}dcCV#e;MCjL+i2d*ihvJ-?yVz@bdQLS=aghK$
z6M<d!Mpr^XtYhwL8i@C1Zr!@o?@?NB3l@VM8jTKGvr!wf%99UOM=bK*$?q`}N4MX<
zh$87V08xS;HlARU+7F1JC6gc6eE0SG6gTz&)kH=$JgVXJ^7h`keR<gkPw{?0(&)5>
zLNu?!yPc+fpi8^<r{3aD)#zgX&N7NJ;9=IC+54!RLI@)Vc9v9iw71)+Z(6ykzWlRV
zRklGO3S-u3+Yo2iDs4A+nQjYcN_CN)Nn2&}%cnLM`x6O4L5>H=cl`bR!7%=mZ7aA#
z6N)sq>;jezOoJl5u2xo72huvB=0t6c@50=<B8l-+CAa*TJjJ*wy@yB|td32Mz9Qo)
zT2fFzD#ud7I?Z)dZ7n}Xzi>N$v!{nE9B-<rSq_M4)y?&{SvV^x>MZ_0%yo^FlbUy9
zn~-3jL*M6XfoWD3+iyeKw~bvbGj|eos2ShwFa>P|P?eW9lGg$$D~C}YWjR7gsH9O0
zS1?GCYZG&~f&ohN+67S;9WpaJrCxm<9EvuRfUZ5&>o2x7pfP&JKs&`+1YbfONHty|
ziP6*3BY9f))+d-n_K>9p(_$^p%_ZB1-Kzzr$F)UC%39>lJk+Ftsp0IZDCehbkv+X7
zQQD-5n~Pfy5VGrS8}VlKzHLGZ>W1BEG75oMEOj`Cry<`jE<r?4TO%+apuU;QOTf`|
z18NRqJS^S5L$uDO0JCJt5-8S0IC=7tfZ9-5FOHA`yy2~jw)o&4#&^p9wA%hnr4g5k
z973Z8^ap3QLc>qI(x0`x<+0Z<Gr&53s;vCfUVzgk8Fdyba66AZV#Tf7d-?R43P&dL
zT<UkfI&`i`8$V$HhIpzirbZOAQ8U5m^y$+N+w=e=A{DMV@7_Hle$~mT>ti5_YF|ql
zB#?R{aCI=v0ICP@iH>q?!CV?y8yOOU9`6x7>H%=W_uSSB2R5W8rg%^_PWkWO1Fl(a
zDSI<BGuRec@8Tb1yF1{R6Na+-u;jLZK&SiE7p}Z%!ynAyI>srLvfbBDC{kDVk)_8_
zfi7h-&-0kXoK^fw2@XH`q9~wu&}_6+tQD?ay*h#$=xl3)%dM1`ZPm67>gt+~aa8gx
zs6+5R&AtTi2iro&50Z4Vz@=FA#vbsAF?0Foob<(&P?>iBl_nhE?b*ys-))>37+<x+
zYo?hkBI0#}F;%yQI@!^2Z-z+Z<m@#vx@y{>l?<1bAg2wA{iO`c-sSLM&nk`d1sx%n
zQF#(gh*0c7IB<@rr3n-Um%v$BN&k^Sv70W4XaD+YR*9kiD6KN9oX~&Vuzgl{p#S*S
z)4wOho%~zeq)f<{pi0`)L1lCiO@H-Y>DI;FGM=R2WF{@Deaf&#Yq*p#l;y9>HG4^2
zE@nm|rY>j_PFh_ZRVmMa#xtoUG01Ul@C+-u$iv03D+5jR&%#DX9eAz)W&1o>yC}FD
zN>jm$7szow$D}+iiyIC(_*@8$_k8*krm<7Y<<`=SIL2vLuP34RPcCH`lD|~{=W{=X
z^_7froEdst(unr3l;d1;%oP{|{DA=$8BRD?n~b>KrtppSz?Wu_B~DqK5mzeDz}o9#
z*c!|P?I0>o^cBX{g0weaon0X2$<&(NCHug+8ixIEm=MKHEs55~E>i}28`7eTP0_5+
zL|n2{vJW7qpv$d-tOYRD@`dyh7T$+QS2bHT=I_S|vw<dPO4jepWU7my{!dyzp)BwX
zQFNgN7%X&*gj|a<(wCzALmcCOFcWB>0XjgI0%i$|!P*PUhb#uNF@{Vu#oI;M8={(#
z1!bTOlD-%)NOk&RFfdH<#?C(BVt4^*gr??0BTG4GjW8f<3frQ`wab*YShN*HHuVoV
zPGmh9kR2^QD#{3(5y}GlLQ^|n!?G4U6ZyL9z%a<J!$NS;H+84h>?6=P11uzMt>ACv
z3$Lg#BYnX8v_5Dw!r-4WyJwRKIm2@Q(@f}dof>~&mW5)2Ms{k!Z~l?|kh#q+s2u0)
zuKsN;&eeR~RhTYpCo)c;AH+_fttPBU4YCr*hWs;aK|eGem|YJ>dJ{n6v?h?%Xw(Ph
zq^&t^#TW?<rSt`%aqH|jX}kLCJ##S(rZHmqLfE>W?pghxbS7f$o!xZW2#{@s?Lhd9
z)&s(dO#9!EfY^Z*VI>h(q>cGkyGG+ig#H*9oIKTMCkw3iZC}58;+MpQaN^IoowlL0
ziGQJ-p;E@#VT%8@Kv+tnPr6O}3pQHnO&16-4D&s`ymW5=8>4;q1b_t4S>QghTr{yX
zhw?)2BXj{0=Wl}VXO#C7xi!1#uv~Lz`W?3e&Cmt`tMEOKd^gi?-2N+h0NaL$6c8wg
z{QQoYVdC`lK378{-;6joS!OARCN92Xk!K>a5(D8BI^F!h?X;!H(5t6YXGYdVgipW7
zt$$}GPS!M5ge9j@AHry`7V|UFzciCsiGwT!%n}IN)#gVg$_SU8hef_I(XUTUI&C8I
z_HQRqC=esUhp?mnt~j8bE<oRK+98q^IR#E>&r%<VieE7Mx0%qNAUhlrZN@Uw&Uf0`
zOgl?{PJZ8=CG#@UPq^d<1!4c2UvtU-EDv0C2a)6Udxp>bh%xRFq~`w#Y-Gr^Y)FVD
zbeE|hZEY6CUb%R2=C~;1SQuu(NG}(iKatay{v}<be{tsY-?YY?FaK+;k>2&4|F@OZ
z|JyCYe{T`|-$nla^&;=de>z}1kqnX+&S0j`?N34xYYjf@t&V#0h6gNF>Jkt$aH7NS
z81#z?dqRdfUxW4LX;|J9%#;!?7GhZ563hff9083IyM<D9vX<r{JMit;0cXz_Rey|4
zfM1@I%C?a42mSL$4Hb09S!L!|C+k-=TI#^B6wtXw!#6AK*V@h(IZm_E;4zviGDnk!
z8GSMI=4#SEnaNEL)_GKEx+?y^m_Pr^OyIDZU)$Ml1G`MKOJIw?8G7I8pWigpAGCn4
zmJ6I~h~@G>*rPw{+6=$a+x|ywhep8j%!D>q+Sy7TMeGi+2)`(yzaT$kZuABH>PKMu
zl|}^U5VKm@tRn(hO`1XjjU(p!Jhb(5L+?9h$6P1EP%jr<7pA*S=;Qo_=Js{Zeo&=p
z+R}I45FlQf<=qrLKzBmWMnJRD5VZ^@fzS>74bkUQ5MrNeTf<{N(NSqMOdm#usE-ZZ
zs%8FOi~a5iSgoHE*Eh9pF1OP+^gFdKWibyI>=kX8bKN$-76|j3KR<PL{0rs}BXGhu
zlj3KTN7FHX(?Rtc_UJeB=-;(RY0Li2ef1AU!#6AKE4Tm3t@DqT_5U`LtYgeHR{S^3
zpZ{Sdz#PAFb%1z8{6@$f=~oBUuUO<eCi*w+(f=cHaDcW;s|67!Eh1UL<I?mAJiCYq
z2BP8@%>La>XisF|Y^~WT_B2l~IA=lLAl@TjSwt?)aff{8xB2Qv_!T2T3-ihYnych*
zXW}>7`o*#O5ApzU8A6c5cSntjoOL{jJhJw9b97)<+;Hy(#NXXtWvBE^1WjD_OhnrM
z){Y&$IUqlAwse>27@Mk+uZXv^HC^jwczFD5C7TUB27>-8>Ej4Ref|3X*!2N@T>sY)
z{JK93|G^#2f3?WxT#QXk)$k7QDKolp@%xB-=^brY13U>|yE6J4sqaMz!sG#4s&m)J
z>z!C^yiG5&F2<RXfKt2AnrzeCfF3PsoOYx~!!Yod)9<I4>b!L+v7>`*fX46Wi0v6^
zU)o44POQbYJrJiv2AvNf(4S*?xb90=`$Tp7$fWH2e4E6X%-gI2QyqE(X``A@hGfzf
z!ikVCCFB@^FQ{hyfPi}Qy5>qP)@ZmTrnh7iKRA;EH{&w_^!OZ?f!T|)I{VJV17r%s
zp^7$4guM>0wr|}$)5P3petD{_aVM=chOX7f7&54_wvDA{o*Gh<7<>0`_FBue>WKf^
zoZA<UEjYBny8FmomnWxXQd|~o+Hhw{(5551wm)C^l6AwjJ4!46T%dS*<-(WS7Cvm)
zwDsYncYMs*bh6TejcqsmJ&kd_9`;<4op}wdGi~UB8GA;Cq_Hwtr2roENM2uTI#N$2
zz;Qq<Eb8^irzcNfN=IiWQ0O^3*OB}k@N_Yjs%WP68KD248tLKi6OTJ*%gg%k(LFz>
zUwzycUd+ZJ=<mn5`fOZRWccc{Z_kUB%vOorqd>o9e17~<7CeR`ndm@sl9P4N9EbK;
z*oB-4M}7KoJUlTA%Sok75}<1+%x3}5&P+CXPne93mY{}n=?`x3oXz1d!;g1BNHwHS
z6tOH#ZMjPbk*w{jP4qstkp7_0EcnZn9InK$11PQNbH(bq6$z%C<&w1~pNI53meHTv
zbAmp|4W>wiDv#pn*HECqYoNO0Fx#22xt-o!jo(*6Z05rjO7%t<Rlc!EG4OkTuW|4Z
zJ@`qQcNTlJ&Y`ymnK5PRMr`WK`qV@Uv4kVUF~PppG|_JEltt^UFT5VjhIm8&_v7m&
zrZnH4pS)z9VBa5qOJWawkhxnBd^AX0kqMBnpQ=!0h9VCRP;x!utaGwy?T6MJHvRr!
z$~X@5@&L+T(y^v`I{Cr+YRwdDnRIvDrXx!nr|Ip`a?+8H&5^R+tcE2sOID{P!S2{L
zIvc2{==ay`(}=ShbK^TO);l@21D7^6Q5(#w@E}0g;KSpsqpCQHQF$nj?EslL**Lky
z`4fw%7PiMqYU<p{VycX0!kgYkst<-7d_3KyD!`VKQ_f89m5%NP<yp+3mf$lX@hGce
z{)Z?`55cba7KfJQiOd2aLSOz4TvCT`#8Qs*iQAk}gy9dE2Fn8Z#71;2G}Q?Ow9!r1
z&Yv*8<Gxcp!KC}{ssgA{5$-U)J>IzLl`3vxv&SdnvX^?J9r94Vq9MUFVxpg?w=CrF
z<UjyaTr-9u>eFN0)rsb~EhnfOyK!4)R0T97)-6|*=-*R@Kk5BoXl!;15jZ$x*jcqO
z<I(oK-nTg#4s(}J4`)=7Z@)`)w_etM4{rJ_=}?NWM}1HqVdRemBc0(N-yPwYG3+eH
zN;=S=q2-UkN0+qc7Oe~Iw~OvRC(XmgHKzd&wfLTbG<M&l=G(m<>#KWTKpk)7B^qcx
zd}!mEthZI*&Xb2iW|^(C^!{a$bCu7ZKYG*oe30}t@(c;psy_GSeXgwiWC)X3|KWvO
zfne%bHXF6~;M|AgBQ^H1RegNNWm`unS1^`1@1ne3%-L++b0qPK1`BA!)K(v)67=o{
z<Yehy13)Us`c~mwWK-ac2^7*zhZy%D<EZ5651dqW5Db%bR;q3**q#V-DMo4e!0t(o
zoqi9MUZJo${#1bPiJr>H@uBQ^DAlG-KnECAaOaz<yIGY;@^y!GeG0FKtzc(wmDF{i
zQU_(#lTIbsNIy{V`!kVszx=US71IZ)2-HQ}$($JYoNOb_x_<D<HpMTvelWINt_KF;
zHS2A9H>e-V%G?Q3w}FTS>A}J&<uo-E)faDTUN_P2Ct|LXdKIdSJH=Xem)%;ry3hv*
zpP5I_=YS0vLR{E6#d#6ZVS)PPBhm<%;P;3RKV*Enm;iYsK`-*Yd#yE*?EZhm9bCi&
z&2CqE8BN2Mqh+f95-U^B@N@LWut1*5NX-ICMu0E)0R;4WB#sRd>!s2n&Sl;U;Q0e`
zzTw|b5N2j!_WttX#-!Qg6ZN11Ox<5FC+Q!M-Ducwvj8nwA9^ZuA>*aNc{0Fk5e)jD
z@fw=jt})-;isfmrF26spaQ?^j3!jzdb-LbKqTOnSjQbxE0{=SI(rT*<;>3wKqV3SV
z^*1`u<K4Ym%10Ba?C}joV$E#j4my6_H_(i)o&N;F4i@THaGu`*BFqH|`H^HKf{d#k
zEMNZKQ|8E#6iydKI#N1}=;z>l8~lNz9yvwZLn)!;-+5eO>fGD58%3OENe%T3?kTe*
z{3|@7<~B$Mlau<KHd;(g`km(MuGR|VG5bAww6&h)8xYMe(E9#5g`8MhcudaXbl(5g
zXi|EYY~?YL8aXa$%QP>1-p%`P?&@60B*K88?%&nJ48el4kQKiGsg(X~H;|mp&+G@)
z-LCw`<7<w~xnrTST_a_p>dB2FIB~2r=kH_}RK}Y6_##03s!2nN%}l5~DoU4H&xU%V
zg`bWvCumO;iONDSc1thL`;FRLJK{(QVY(+aBb{tn8gp?eVd{Kl%X_S#+AfG6TPu7?
z&Ucqy-77^LeyYU<>VciX>9pf8c!`S44t53Wp_TIvC&Q=q?S)Z@Qq)K=xjKlq)pay|
zu~aB+Lt}M15XaXQvqca|cl`wS*WV3Rd7`g6@^Y!>XnT6Zg`6D84fAgO!v({`saVG-
z^UUiUoduDqwF!|Dyz{<wC8eXt&t|+rwY{KT6;65y{`p_gfYNgIpk}lcjH^QMsN~Lh
zUwLQ`IY~)P*GEThWu3u0tSVVLFRAI5-CcF>@YcPLte*;K)@){)m$7^vJzImJ&P=v1
z4d?}lXm2z>rwf0$e=m<iNwqX6(P93_o7lMaN4LI?pm1v>yFsjI@s0Thg?VK=$W1)d
zKi#V<G}-iR?|;E989XVJ>E2U2Vy}?;6Ze_boy%~M+P!;or~-D1*ok$T8Gp=!Azv!+
zvZ+lppYBf-NVPSuPc*N+#w5<lO%{m8JGMTiP$D&<rV+LN^%-43?UW&i2=?E7LCK`=
znPRYxV|Pd=X{?+0j-3l=6J$jH$*1`oLTC)Ez-mFK{->%s&6$kz!5_P1cnMJ@JT2Tn
zA*+hgZY>bvNH#=}i@rk1{tTY==1qVBFyP%=nq+Zfyq92=I{)dt@yOkkl<bc(IS)Ki
z73c53E8dte1CK|XEdtG^e#g8g?V*CYWML*(-gTm8S4P_OM3+d<0}k%;P#Ci78D6+u
zUm(;#z@(-k#g_i@l#|}JZr5`KmpU0xn(H^sKcY_;JmL$x+J?xZCHL-;@&pk4)_+z+
zWGW&xcLaN8cXfphuzi30w^lNK4|n%P8A+*$b1Q}(fH15vf7^pZR**(Cm3tOM7C|a^
z`8eJeix)m1Ru|WOY?Ar=6|(p#THYdUpx81=5ed*wSFB)7L4Ch*_cfV+q6$cN`ARx!
zq~zQdt-FHvL+YZyKX~p3*mFAEmh^GbaE>{2;ePk7wG}c2_=n(3DgG2ocaaU~I-V1s
zp|+#%z%p1N_;>EtnCUMB52P5@T*yBEe!#}yHyT$CeHSVzk_P5|-C9It`8n?ItZsLx
z9;#^;i-k$ocxE?Z8e2R5rh%Dp!MJ_^aen~?az_jmSgWFDDg*wenARbHV>J8cd;RfH
zy*a^@U+LwQ*Phu>fU`3z1nRH`LZwdL(fK8i^d<68cD}42zg<sWbCD=joFgjs{Cms1
z0&yyNRAS%!lKuqNX%DH+szhvj=t^@|a<b{8i}_DZ>eRM->vlYvU$hiL5W7WBvmV|d
z-l}$qte4X+d($dy)!fm1V<?ACE)e4M63`s$4&y2E7rU5s)Asy@9Q0VJ`^5q&O#&rV
zJ2fG2zr{9N08z?pgnZ$>#+7j|7XThs57<2<@+i~M`|G3(mHME(BxD~I1RtB!oC8Sm
zBOJUmtKCSRAit|?XMpWMVxwTenSvc>`x@Hf>%o@hKC5$uJaz42!D^PZBj+Q|mR^MP
z!?i<q@7#H%5#Iyqg6}@#8W7LL1Nk75piu7tqfgOYT^%QuE=?PG$rI@~$rmc`h!Qjl
zo*y>o&++kYC{y_oECZ)eLTjsSXJC81eJAEI2cl#+sx_PT*PNM|mPE?!JyuWIv{`m5
z^R|h>l5U3gd7iEuJD=6YbqA|!+D{HBkf&M`=c1ml{s6A`*A@aby)&IOa6;^mNqx$Z
z0rgse#JVSGuggL<ON3vEH>>Y4t511S7#abl-C#Nt!DV-~yfh3|cg;V&YW5LIHLeou
zt&EJB7gxvS+~;&><>suaQ2$L2cu*qVJKul<B~SLXQpn?+m1tz+i^a{49rfDoQ?J@v
zDS#nfJbao1`|!%;&(W%TG^E?}pVlOp3fm1WMMm*(qtMmvvTz-|tL)S5tEM~(*Pzf^
z=g+6jzry;dg9Kg0=%)91VRsfRY2(=+-Kefx_JGT;M0~KWZEHn@YOU}R;Z$2{&5OgS
zj}NBidkKi!^pFlZyv1}D>%tH`b$mCB#8>H1`V+C`Ghr<)BI#2tRwpnDE*DYbDw8*u
zZ9CfMAI6_~a4eO8kjPhW$!4p9y4qFhS|uQ_ePW&+z2D;~-Q3oRrK%>H1w76q1Ue@u
zBnY8v6Xf?9sVI=N6{M>7b$3o@V<5}B2x`hMitY%N=eh6E7$7=7$kSgcID*vJr`C7i
zr)udA!;&eaRR^arHQ1<|H*ehIzyK)KSojXZnz%<{>>}wTvm>zbvK^ns2U5)qT3aUj
z+#+-dIyI;yz0B*$KGcVvd~5{r$a4ncftq-$Ls>7<66Plb;W+E=tk-9dqSnoij`hTw
z*1mR}EC#Zcy!_EB`!<W4+uC6uhEu*{U?10ePG}9s6bK6FO1-{Va0z@f!VY867Om$W
zQ;OFL^MZ^Ji+o9tvbQt={Re8;f(s1t;dIf&b`D*;7u47ifF?@K@l?3lU6<?(9%{tT
z^+vrF78ymp(ytN!nopoFllfl#mF#g29lL`y3${4N=WuYR6ECT&sHo(vPusL<6O;jd
zC~yyDeOa|U)FX774Asm9w5BXpO9}8^(s#MY#!WsHI2ZQlM=6ux)7yZzO8=O~?f51H
zy131T*(a>$3c^l;WP;Vp*0CSN?}n^D+ljtzgvdX>I2>pnstQKfDbWLp!0|<XRRP47
zRGZ$%6WJHeAH76mg6tV%my1H06Y34>^Gdl4A5w>3$VOkk%Z-%X9zqghJr$dSIR-ax
z-VClj64a%>I!M`<!MBC?15Q2G^4*;mzzhXW)<NhQhhrWC;+{`SAl<7!r!r!4s5F=q
zOZE}Br6|F$+$-yzAH;)Ky7N}W<8DROmhMqrVeoh*UyTmDWdzv1uSBGyxZW7NsDbld
z%py%WNXFtD&hLh*Hr^PdT+4=gC!g4DSZT!`Af7ncmvz|MdepixWw1WAQ;1-)Khom>
z@JN9W+k-oIa`mmxSH3Y+*SV@y_hF>HAY`VnM9R<hwy6H}SRP}ggjt=Y9zlcV{M*pU
z2c9<Mqx2fb$@;`YyG2XmjgOTGN=S^IZdVbiRgKiJ8SiOVr)2CiK3f^CEH0ev25>_9
zL`^8O>o>u~gH9R%(ONQed&`#fZ<N|!xL0aHo++^T?MWlL@AHcbaRVopSj^-|fzZ<A
zIvH<4E{W7#fAJ9bD*Ps(>UMpcVgFimUu7icYgL?=PuDx(s@NeS(5Zb2RzxpLyxLMz
z$Nu6KmwgJpD>*+z(60=>I7eMfs?ujJZ1na6uU-H%$3hX)>D_<;hrFHOH1gp-N+n8*
z$Y#YZ=_uuMi_IJ)MU4j>P@hw6H4}DHTzS$LM|Vt*_ip9=;MzNP)U`zc;=*l~?EST<
zj?7!vLS#7trXM&%z2unTY;OTlA2hps`SRKX(*uznGt<37Wizk9e|D+0g;V4W&$;Ro
zC=InJdg9d=H!Z{U=B`p_<f`IhzI>b^dp|S_^``@;ug`o6A2b;3gewg!%KennRP$K*
zbRnP$hLFUd@bc69J9tC~iWrQ!_g+5tW1}y@dHmw3-{sO4MFoLnX_V>cCyBkAH_P_d
zCQ3S@FoO-IfXZ>rirpN)ug6+ySCmS)Qa8Fd#U>)wqTRxRlj%(|Qu#ZPP8iZ(9M+2h
zA8poUiL8_H<YkeJ-!)Vyj8|%i8vOj$c~?5rzGeQiSbo`>kD{YkfH-WkM!ke{aPea1
zSlf2kRF>#IEQX@{EOPGr0qN1p!1|_I(op>SyY-qOCqsDU*Bw13YFf^l;!MgpU~ATh
zckB$5OdanokHt_8bU;$c3Q1CXB5vGf<?=sUrpqH?*JM>P4s%m2LzP8phCSfofqgww
z7Pi&-5?JJrR;M0WbNK1@ryh6jy5HGG$zXavklYzXC5Ls5SLxcq=a;FkVbAT1^qxZn
zcz#@^g3q3fYjoe5)xXY2b_2*DeM&i0^u`Xjf(o^_R<$AC*qfmsP|DfzC-Z5`%ZAi{
ztTGpOWt27`iWV8z>Ve<220Ss2gLv(?=aZGI&c<Ha2OHx*?Yh#Wk^2qxEqGM~GIq?w
z8&wpxb;2bRp0*>opr`ru7d&uVyY^f)RKZ>xHVZZO1@ucg`p8i)iSf1}U*jmSz#IqE
zb<Ur60>_1tmuk3@y~oZ^T#_k5?f7ZX^cD`@3q((cGvLOK$;hlnp*IL)W9^2??o#?K
z?<QlMtAUJ{M)yT&i@TQ3rTfo;PB8`He6CO!zQ}Y3nb=7U*@=_$QQ}sACF}5<f3Sj$
zbK>Ms`ZkfPCV4JZAo2E$iZN1--`XnMVL*yWG_75Eu4ufcg5}5ILoRjmC(mqfY!%$P
zmo`yzD6hbADDylt4B4I~$_)_;(<PatAh@|&Xct0&jBy{bKQO_?v5lZgE%}fS2ji1c
zQLpW1wh!@mF4veV3(nOh4ZA8UusTlkc`fBoPn(*_D@Y{`h)c$thD!9q*=%8a;*bwL
zlzuty@d8Fmb|6&AQLT(9ZOt2RM;>t1GOH221bVYB%>hjN%Y;vXJAEi?snRJ{(nSB$
zr5tPFp2yxkvCh)-*il|*fL&#5=rpexE0qhC_=Bvs<OFs4&kOBBX+v5rPla?OP!eb9
z7nt;dQXq20S|5m;DTs-!2lH4^ORu|Zd0xQ|@YHu6;T2HZC7MAZBcKbHq}n2IQYBsN
zMxS1BzkHhW&V5<*$YS<}#U69%qWJR1=VtkN%^BBbESr}E2W|57o}47;cH2KXQ68qi
zIwVDH%WD-L-6(8uDzA$@DA<qLuw6MUo#$Np4Q3P%_uO^#9|o&e`4lVSP=nzSM-*Xr
zYD4Es?#Zna7vY8BMW_mu6uP=|OKu8Q=mLrG@GSR~LVj0eq^#YWIV^Aap)K!ivV^E&
zxDv6gJpe<o59ILq6joBc7GFsHc=##%3f7Aqhk+jhN?j#DPuvjc1ji2J<3&(qo8`xu
z-tlOhskbfg?5`tQ32HTW6d5a@?Sgs2wOxMKMM@`@=V2ZjKq#F#%2P(bL?1e;nP|4?
zq0F5JX58EnlAI67jZ`icaXsRoxtdh?X1(d?2h&ygo+B?;N-N}_4PiijP-0jGKfpKB
zGl45V%UKa5uzo|v9z~x#M~prA+D{Ne4l%}^u~%C1{Ew~^)vs4dQ#I_B0#{0}0k0I6
zDXoxTQXM3RJ_R>k!qWi0ptDTliJ^kadr_0m8zjy;ifke4uF=sW8s`mPaZ*~WH|>a?
z(&o#yHdt1$V<+~L8~ied2E%qHLB~x`gMmGJSdZ&1fB_W;UTB_rdFp2O=`tE5(7^%g
zJ!DFNkYY=wgumab^-3i@NxG+JuJ<5_arRabOUjcC)K0zf5TC{WYOjdQr$IVhjL45g
z0MuNxOdHbJgdIEySNk;<FLvES6ax#csLM(R<E%DRsSBsMdwSy8NZ;nSb8xR_QOGK~
zb#-Qe+w293z94U<HG4jlHQQtj&T+!?X+YBZC#a&rp3-2=qvj(D{!==<D8S)jVx(-j
z_4dMr>xmSw_b!g%pKd?1?8Y2CEBq`{QjE3z7v!(_y|`Hzw4vJyX9bhpJ|?&ElIe@_
zYxum}RLB*7^6<(opVUmAnp9E2Juxaral<(|gL&l6KaX$Hm%=<+0DuA|U5=m)X3qug
z5t>Dh+o}D3xDF#xaD@q15rM!S0}C;iK}PR!xQ{7(Rz7us15F`ffFG1Y<rXz~40Q$H
zXL15;IOaohG1>mb(Y$(<3%5E1%DAQNjf@rSt$uq?Q1?-QN#TG&jq=1_D#~ZDZ~@h2
z;b0iTT1^pb>pw5=nFDC$FpKYZv3ukT$e^*I&KGn&7VH}>OZFF5dGb>WD*hRv?V;@v
znUxAspLc?cOg(rGkk0bD7Esy?IIxw+1VG1k*`HD9qkMf*InK0}H$!!>jwt&IHwSR>
z3s_u8!55@ITnJDt9bHS*DaJBS{$}x3(Pj=6!t;NE&O6*-1CDmstJMMy5vd0mfN|D|
zyul9Cl4hpIrRo(lqtBVac>1cTEDdaBZ!eG-bn(Vjm8a)|*L?wCl=c3C$)+Pa`UU*Z
zd`iFlvEQNqS+gvmSLlC)9=x5_8<je6T<WtE#4`*mWON$JjI-9Q$7!clQf9<YO_p~Q
zk#<S1ZwmIf^yeURH;8B#+^I=}#gYFVPa&Z+Kt8Wt9k;#V4rX*9n3kc=<^<3XQRE@{
z)lyfA{e2i)7<e!eJM2)hC4rKEIE^=Fr~a1|4L6ih#B;G4@ve0&d-mw(-s?_JM1lW8
zos^*-8@k{Qlhbtfu}qTs^h7^bQ-lY1J;#4X&BywyM|&4=_+y=-)d^@W(VNx+y6OAt
zy(dUx_6G{EVxUh0cM(TB{~XL&yX(0wY}oPrEC0jsj1N-f@-r`$hu*LItLy++^Kf>R
zr@w6sIuZ+xTF-WmfCJ+*FeEW6sYRDoyfAmcvHo`=`4nm<jyEq*lEBORp|j=)TxM!R
zw9QB-6+^p79qKzih@Y{hq}@1RV>H+>)9d5Z_s(q)b`TJdw%VZ-3240Vbi>_MJ8ovq
z0eaJ!g+WwLNt*>-9ntXp0(xqf&9}q5PNoZF68hlyKLd{H`(5p?dAWYo>TM6Hk7bwW
z*lNGfG{+Muq(MN}exrjw33iB|sQJiLYb3;ffU)Uujfn(K(yTknacSd~y@UA4iI*n^
z<*`y0m2bj%ocB6I>=oQ@9ZY#liC{yK2Mc{X!{q&T*<1F(P3mJx_1g=LcB4g)f{)@w
zth<^3L`Z5<tI;6P9>CR7t$RL2$@_I(Dq2?+p_=MMe3d$o(mIGC?CH<E<jQI)DWybg
zx?yOjTZPj~F6$`jEgtgB@!;Xu;I_+tWU6WAbcsayj^IJKqCT}|E=cXaCNmT4d%~lL
zhemtqlJPk?;c)q;lVaViGT7tIuiQ~<PLLR7!%EmYxokOP()YP(Mhkp*xm}L1gzFam
zXJ?9OWz>v_j1{|f>eaES<8LqIco|i^4s}+gg0nH-s{||Q=-{#?3kJ75jVJP$huK<Z
zbBJ`x`p<NL80A|_E*=Iq%!|XCB@gt$vt(Qvq+Y`@4j1!}wRPW~%dBHKaZ@q)_JYAM
z#}>C!Fdm$aB#g&kP>o{uS!_ipp~tn+b4h_`!)8zvAKdCrx0hIdtKAbp`9v^oB{Uf%
z`Duu$?Y1@xK7!Jlfs@R9nrOwo%BJgpH7INGIcGx-SG}l39!JoD>A=i%blH(NV3yrH
zZx&=f${Waa0+a%<g#so7X7W693|bR+BFFHSEO5<TpJW0Xnm2fmOJa(+OHBarF)=*6
zfcUW3I(fHG(GmaJ+N+AqliNcd768oM71NRjTY`_W$dc}&fdKkuyecBRnU8N;wr=k(
zJJh+@OY*?c7Of4S2+|`<zgfiiv9eKc|6GYa@i<($QGtY}JlnmynsS6w#EEBv+fbWl
z{q;%VN}R4x@E}ATURpEPJh^UKpK=BoxN~RQKxVrIn8o+*+&MmON$D!xe0^9OagS_S
zj#x~CIl$N&M;l2+qPn*;=i1E7<X&CsqR6^;+P3x0;q4b8kaE?kGrSdk#LdUjSHHxG
zqsh`W?nm>_1Rou5nt2KS7}!2=(tTN<uYr_pF2w=0u4m0!AkeKlyj|T-WJWF7GDHQ}
z?y+v)&R4n`VW;}tS1^X0e0sp5<@B!fy;5ie7a{rM``7$n#+@ULm9rh*A!-2;4)#X&
z<qx<tfx>-w_wJVX+*li~OnSUl7yP~~AG|-jH!bYq<lJYpMr5uncuw5OPW9R!#KYS}
zEr#D?hChuwpCQ@qHTtxxLYx?U%p><2jvMwFC^OzzbwR6a)f;u6QqhV2T5*#ftlp-C
zTC)?5Xgj#ykH0(&a*V0b$*S!M|J%`yB8?b*3Ish6sOa}pN}>aD&7WXe7vJmcOiBPk
z(%kMi`N6a?-Eg!c9((~B5ZZ!EpYDS|USgYPH4sPugw#z9UpRK(A=0444D1-d*RoP}
zHun=LwXi^$ejaizLnWD1XAwtHfR4(KNZ_GnA+|Li!$9zzz}z_IRrM}s@WBJKz@A&=
zXU`sqjrId9;L?XO0BVbdW$wNpJpEy^fs_4*6<Yr>&M;WT)fE7xWaZpTu0e5&?wob|
zUWZOkh&HSlc=50$_YvL$MN#wp_^wwh!38|Bi10P7A!SaL@hkJ{kKLE`7F^;tm$X^e
zNRivYn<13^vgy%gM=MK4bAn_~2_}8pMaIrR$6~lAk>q&1UOa(d)yW#J^B&V1sVJ6=
zz@LHPf^g0n-2>KL58cx(CI@u0=Q=eWGJ$&??m6`!Ot%=PH{OH0yOQ5O<~EcpkuB35
zzC4lyw;uJWs7F1L<itkqfb4?(K*GtE?R)YU$Q%=8``|4_)K9b~43r+@{LzKPjA(si
zBDFCOxy&rd-*dHpIz&H@S=q}1qL33z)H}P|)WJ=LfRifWT~U|h3f0x+63w*toZR#T
z)?{PzxdhAwz_}}%NdYh9(%2ssXPZiw?-Sa^zk%^ukjtR<p6A~|>_nO0fzW~Syc=+F
zR!6`zst7fEaB*5CR0<m^ndG_SAo`-9fyR#d`uZu2@#o)j=q1eYeC9Gjz!~^IG$Dj(
zwQ3B8I#Fwuy7l62(MIwh#1FSHw)4BW4c}PEA~(K}QfuxZbx*b?&d?OZA87qW4LEQ)
zoJrADNieAf-@j}153(Ze^6=&SJUs~2C>b~p_bJ`>odJF^fry6^)E<YU_BkL%mGIY4
z4r6y9x*o1k^`FSt5!*AXN!sATk)OQhFwwW%>jzfl+#jtA?E=XgI7^mD)sjAC7lIR#
zlyQxVTUlq9(WfC5T%*UD-Ft1x;$^7dEr;MbF?KR_Asbq%UTAuxpd{P!2iF%nU%H>A
z<!5lRL!02=y|<FzsmK13zJA_ILf45~iD0TL;T=&xXj(A|_M?})3#V${UXT?*SHbO_
zJZaT<ZxwI_Tfy87jE+<v9S0orfIf)3DP)m#7_T@O{$MXuGU_{V>n(V>PK*!C$O_>f
zI(pd-H5J2UhQSk=JNLQ#v~}ZbdT(13S|FC6rD|@o+}1-)<_nhUIQ2ZbL?II)^uD9`
zCZvM%hB^#Wd<+~Q+<m{QfaO@%^0ix+b34abTy-vZ!p~lYMysg!Ffm`fIXri_*o+#|
zPP|qw@-wH{nmKw03X3{UjHw7U*bdYUgS#TxG=MZ=Dg-xnNA+$3@97sL#+g)a>UZ@3
z=UAHR9LJyiX}F7vWA^jp%SjP#Zu3--a=&wD|9B+nY-CVt>!ki*B3z~%up3I`5>GIx
z8Ef_3w?p-f+DNs*Xh)%s7>JCwu-@m7vT_}0^VSt1H+Q<jZU%43>cz2%kJ^~kO&zXL
zh*Au8s5Co%d9Lq$#MiV|BvkTqT4!Ktolo>I1XUe9>UEOT;(TM_c+7j)u1Jmelduer
zeUhXoc9lGWb|ryHuU@^frVc}7HP?Z--eu;w&kr({qeuM3c=Y?r4;jxPmxs*S5bhUe
z;MZFPm%=x%frakP?-?$0xL6BBTz+AUD=XLIQa%t)Lvm)hB`B(p>D`y$g9mO>rxbqF
zpz8|X&cc02n0HNV5r^*jrsA6^pcHruXeQ{JrKD3Sy@hW8TsiXT4%Pgw!)M=2Z88?!
zvU_*BaaD1+RHS8lKAhYoZBnk4bN3S$u9bWU+Prh8WkBvY!m78HtUB2dV+==*j~4ar
zD&3be!K)wR!21N3ItuDZe7kq!!lqxr)n=sH+qS~Nt3R&nlN1|cNEb<iq?kINkK?^K
z)R{<)5dDGFi9WXivchh#GuJww0(UcpJX~9z!iPmqE7p|w>>tMWrF6cCpDTXE?{W}O
z^{quoN!U97)7DKH4{o#cdKW&0cr#6fUW2u8-nvV@)=IWq%7OOxVLNer|JVo4&2Sy7
z<a|i3Ft-ms`eU1$qwoo*Q*!(uzs1JtP%%U+)FhZH?@82w%PO%*9=?suO--3Xe@<lp
zC3^4l!9H7q9YCQ++M0z3<TQD|v*DERvJhDegqxTUh(Bo_SB<qgb<NsC)S&QL8|9`y
z+yi<QzOSm%#XP#j(*uOQQ#ako#ly7h<uKe;`tIeQxF@taN;A=+I_3z-G%hJ}1>z-`
z@rzKgzg+PecYSiv59IY%10?fQtmQKQ@Zo_425v_uHKI4%!`#V(d(g|^eitTB<mTqS
zIPBWjNZ9;YVPo7}4MgvaDsBMYf50tSIJ5YDolJLG$c{6w)nYH*hZPoufEgf0@H_X{
zB_D>v2&8N&II3qzlMwt!1<H3_W^d8;abvbC0SRRREdg%SjQT!5vu&^!7r6?y+Mt~O
zRKv>?Z$0@u6uibTyfg1#Qnz=TuMy_$C=xw(pIZ_z;j45U``cA@)=EIgMhOZ9VUFp}
zq=|uwh+}28Ha(xxflEtEi$EDuOT8MI|8&Pz0TbfLdv!|b1MYfiBlT3iC3iD~C%_HU
z;Rlj;eqfrLA#mvE+PANQd8QA4xRQ8B>vECr(-Kr8_@!eo&!o1Ud94C!Un7QEuZx+<
z`++~!qR=Lg<rA))lX|301CCY*=7|8wp7Y@WR2a#Wbg;hnz^nm+K;Zftxd|%MCAWF9
zP4?HC#rt1-SMJ@E&KTBx-0O9b1udQ84Ce+8Rlf^nTvvP_gu{KeWCOoqP*UBWO?J#s
zaU1?$)NW+;De}h9XfKX)$*L4yX|wYXJ2kN4q3bd2FR2-v`~n-621o7GLJ|#_+G-S4
zR}0>-*}(U}_T-aoP!<zPGqG?We__pm^bNQ&uY<g&T1Ykt6lxlW>w@}i^9FDIrGX}y
z%EDa-k<xV+R0#YZ^G4miXc=nOM$TCw3~`j_Hg8ng$lZ7E>};3K8{qrW81g~jHtZyF
zfeGe(I<X@+&+`Im`to!q787&+eG3m-E!nW#17cm_y1f=I*{;XkhL95T`V@3&kjF6(
z?#fq+GdrTDG}8;UJtUM$*#De&xNF0838&KKbiuI%Ai9ESeKE4Ql%7F|GfnRf<-E5V
z51}lBX;cac#18?KHQ~+OgF7MG3;YS2yRmo(&J|34d9EzyQvW*f2^NLF=AEU!K6&ab
z@zBKdcoj?FhI5~`&K8G*lpl!on+>_5Ro|r*=grLpH5ZrEV*X-{?<>tjuN>wzHjw`Y
zv9*~8U~eHt0=Yn=)gCoWqF1c{!?D9B?2zDIiw65nv(H0#gWq_iHr=@#&OQzZWM<{6
zak%m=RGOCc^RsUD1fTG{y>X>4RPpnh7`Q+B>02=KF9X^2KY$YlVz@2pZEhu)X+ugM
z<eMOKNtvzs3O`t~KX3lyYlTx6xw{^X&PZ%I<>NbDssphmzR^JkdFN+UCchSWZHj{!
zhmGE@ypVPFIn6$jZvh05(sFWW%<{AOoT>tap(x(X4U8xw$ozrxmmBGY7J14Ev8O-Z
zpX)4~=_RHXZeVrcEZdjB)uP<-{auXnPfXaN`l9zG>$|l4Je!I}^OWunkpF~SD#JZJ
zPBGXBtOQO`vyVaVgFPfGV~|LIt4?XkVOJR;I|`hU5K{+5y(+e=Y}1Tsy8+pSoe+#?
zJu_ZaZ`X<fTW$a@ED%Xs4!LS?w21G)?Jl_z2{=ZoMT@F2t8hZFC@r#KuhzHxnGKpx
zcBt}JG&e&=>OPB>R6kLiyqvr3-vfh@;94Xk;^P&h1TYe2o4k0Dwt{5EK6r7m5qITx
zGO@}-<@dwA{Y0&~mBMd?9S=d-j1(~aB2>cL0)Hb)R<AYa9_5Lp<y2R&N76=wV!lM9
z|03r2^EQpKO1k;<_bjVl#7&DpZmJiU0k6+kiT?Q;VbGl5RFe}EgTTpKoPPjSeGL;&
zviWIv{QW9j9$y$gI85vRmL&Eel4%*e(DUCC&`b*bM0d2eMFWNW-uPM9f8K4?(WY-)
z>C}JwH&&G4`$2yxd;Gk)j7|hdATk5J$S;fzNn`;UqnFJKK^PjO75%RN7c397gw$V$
zgNtEa5eNNT>HS4uHLWnN%NsHKk)dL!0tCU!NA5`l?%6s)pgLn9`8b{8kawG<aj>4V
zlY&ed0={Nt0x&sU{Ct~{%U_nDX?B<Mf%mQlC>6<pUbXEZO80~6i$hfq74}{SpP~@B
zKGmwTWPqD4Eoh<S->0vAD<k`GsWYiRhBAZYS0BklGB0(#kUZVfp>pA1(bp%K29p-g
z5mX?5=hj)2HjB1w=Kzf*-`g2+LqP;cwB2ynfWR-L)wlkyBsfuL?m`%2X9-s5IK;<n
zh0L99DcBf<KYozH-?l`|urb{fj#ME#H#cS6o{J|<>!xlv!g2Lep$cRPsdUJ&+gy+o
zj7b9ZsSq>np5SzO1djE9{YbPQ+q|#-S<+Ws4yn`o8>^GJouWTiM)penF~9g)1LcXO
zVNpf{@DZn97@x&a@^8?|vWa?Ou;E7R_QH1_gcHY;JY^oG=5BbN8?tHB`JH#KbG*wu
za`2$ym0p#a>pPwYy|En#I;|)p)4O2RkylHDHf>vUw@T?vkjg>EBfHi9di+fK(4~NS
z9}^MLhCh$9cFyQaQ7QR8{*4mlof1Sdi(ey3vqmo8O!;eZh63Kf##_MgeJ=uq{ZFV%
zSvXeKr#ak?`&T@}zI$f`m6AF`=72eUL2sd9+0gKd*QbH@EzHlqd&NsEcuGKjIf+{+
z{c|L%uXw{v=S=FSFG57^rc9FI$EIiI>BOw12>9<m4eyUv&-(fT!u%~qGC#3Eyoph1
z5HUO~(6aefCkwa4&<9Re4<1Zaut%|4bW<`UnzshnHaLw}5Q_aK$NH+-z(qMcQJbTM
zCWi3&e>@;1CFn~X@;r6Gq9l+9k04gj(4^+TL4Sn}c3F}%n7(+`<*kn&uW%gelCo%#
zG_H!uD=B%MLGbfQZiR{f4qY#pizO%flJ>CQ?Zr_8s*o0Gur_nBvcE#8oz_{qwtKvn
zhs#;86u{4$vt0#r$V!(w9vq%p1D7vJqyZ|}3`g9km$kI?D_&pJS?%(7h~zJrs5E42
z1cppH*_{WYx;I(I$SD8wTT#qt5r^xz-pFVyH-`M^$&<DgwkuhSBdsj2+Uw2>Riijf
z?Br2WR?ZjX^%XI-q|SKm*;BToKRlg^R&GppzKhD!w!J@|k4*6==+oao$%3pq0}`Jq
z%5B;5&SSk;+MauES(l3axj81OfqY`g4FmPH_<i3XX$c4D0Mv6~^O{^>7VGxz4~-ji
z<-^tu4`MJlBkYxq!tNu1MlJ6g)As*P8xSO@T$7Ux7N7p5fkjz|Px*-?58c=Yp+Vw~
zVYbZkVcwliIN>=UFkH|Kk=&5ou&mgMb=}_QL#;)=<IjKRZL4Cjh$(|?@T+Y>p>~`u
zNp>cQ^7H2{P_XOpsZSICHUm0X?JREEHzM2vD6J}Z2c<{mx4}bIQdqY<Bg}cMbnODR
zwLCnN*=)LwH|B+353vWHSWUf@WK<DzPRdkA=je@vIV&YPH+zt)IscpwQ)*iggtFar
z+9f9|E}6?FVe;8#fk#b5l~}`uuP9CjXz1NPlBVHNkMtM0%-Qen=`I;zg4@uzCdCL<
zh5VxW)~$bhh4N)Wz}S8iGf%o8=yE4HJ{od~h(z^+=%yKxuyMs}^`Q)`^Tb>4Me}Z5
zPy@b1h#WuUW5CTu5l%F7cFJTzk@|GgWaeG=6*k>v$wO6H{k4R}6k)7QrOyI3r_mx&
z2e)}wdn-sO`^>Pc6r0}l%S$9Ke`LUl0-ydf$FOeOx;42;SugAqL2Of?eNzTyK}lVR
ztbx$-D?NZog9u5kpL5A`nOoFJ-i6D$Y-J>@Ah2UeU@|z5bT-YqIc{(ttThxq1Xn}{
zHU%bw4+FZ4)?@meE;+QV55g3?%ftSpUW<)EpFDYTr(FkJ!8rcL`mEd2_Ih(K#=Pae
z=xPp18ql#f7TYd4FF~xdEtjV%!iZ~(aZ)kdQ&rf7EDwfPFop^Kb3KTYdoSisU01k;
zcf|4xg%uSy*ca{|Uy5nygA{J>MeimiJ4f;-909l#7Ff>DazqwBZjd<4FA(Wx_Q#Mt
zf*cJHT)XuBL{1xa6gj5AHo^7R=xFS9avUTY?;Np@H7McM*+jc`Wsun{&~HsgC;H1c
z!JNt~I6zHb4OwG78B#{4ilZtcJ%mQ=M_)%o4XvOD?Qd8|N9x)j65wQ+a~*dpvHq&^
zIjmLRXTN)xhV}c5Zi+6Vw_OV7CGz#QyTi|X=ak0hl<{GE`?ZWtxqzsIJ1g*Fs<Vh+
z`GeyP`s07G%#OnkQn)Izc$V>??KD0HUa3vEFQBnkXn3>(U|X74=oh@wdl9HeNKP=x
z&#OKIHzn8EZik2SLykL3aVxmzA|cytE4ua#L^*+jvVfBg-Q;o^K=a_s<yl2o$*o(2
zwK`TsPZ{l_i~L0WE%Wn2oNK}9ff*e@o&jK$le06u6bZ5**C4yroJYA`A^vhH`O~K-
z$GcHbtFOPFzA7}(IOL`EACo%+wc84sohRjAy>=xI=M-h%*5=?)Ju19B96awwkEU?U
z!qm{P<=~Yo(#Dl<wo!%=+OA2R?96v^cXuC49ayIcce%OO(1OWr=gnD=#u3!=*)g32
z`p_kl^uSXG={RS$l6#?!p=n!gGjd$J4Rk_n($ReSJ9IEU_#8AgoHO`jYVq!k8`1CX
zpqz%^H?rlRQuMM@TzLX3A&ULDJvu6>WzXO&p3;}jo@+sDz1GplkCC;8u?DC($H|Es
zL_Rp~D8n9X28Brk3>D47DRSotHO#F30Nl%t#mwnBf2+8ofwy2lLGfzPPsDAeiy#qf
zrs;;VDozJn@{=JmVf~F*pTP8sGt&fEZEFfSSvl+!gua(39t%0jI?zcW@!2cWTcu;G
z!{=2)^|reEz|*R=@$jKTPA5-V!bNdSBC6Y6IOXbKU}Zx8Vq`UVcsRs9q2PX^03`Xo
zIs2tj2zMwjrp=(}f$up~z8AlN-VLol1blS4vD=`y@c9w;V?55Eps-P21^w^ewmi+E
zb-?}J(vy1?A?=I}rPUrX_lfPGtEz&#cDo1fW6$15wc(U1`Zz#V(y9K+W#^GR<{U0t
zlT}CA!ve1MfM<N*$lP{VySp{fq*~$aglBi|-E-%522CLqPiG0xo@c2$+n_PM-uC}t
z?@i#T+P{WzjT$a0l_r!7Nu?1H>ZT;2$&@KdLJl%dhek;el3S-JNkoLqbLNDQG4nhh
z^E|w3?|pC##l6q{z0dQ1|4;6x+d20B4r~6dwZ3bW=xUZm80_5r;(D%Uo~D0v3pWim
z-1wAd4;!`=@|t#29n#F_@l6r};6W7WIUzEIa|`z51OYf&jE#5-jx-K>cqATIJfq)M
z(il(qzOJUFQ<PK=2rWJgZ+ZPG7m6Q<TjZ1}7+ZgAG)v;5zpH-yl4!7~Mgdy`AF;(w
z0B`YW!`@7|I`bGTFN+~3_Zuu41%40-Ne8bdbd@&(<9CD_gVh8b-{xdfD*yvJ+_3cp
zyvpD8t(xY@-o|`6Vf&2-GzPobA+`P+B<6I~x{R$ljcw(qGf5g@A1Rcl_8j^$waEM3
zsr`zUabQRyy}L>ZV?xr$t-d5#v<rf%>p;drdXkCh;IpIrGI)HxEoWxN*TFPC(!TGI
zng^;t_T}@|pKcf1Gxi=jRQows9El;chT5A{*~q55a1IQ21w&_CpJB>$PyuNybYp|3
zKukfA0<~y%Y+|pyB!L-JmubPYCQ0J_c@=OZ@d9^2z?kp)cPphaCHgi>Ap2anq;{LK
zA+Px*gdoARM*mFkQMP93VE?HBON7gpXK~?5UfmnZhO-jfb}mrdh&_z9&Vhd(J(%w+
zyW0D^M6Et85aS1FiJiklsGN0bhJ@M#?E+Wc<nHkC@n^W$UoIx@7i!jL$VrWjH%?j=
zR#0Hzq9G#~|Apat??J@t|L|IU!Sj)2XIhahU2Hd8pM<-N(#~CXV@KsqFsOId=qi%P
zb7D<J;F6FrJeVDAR-F9UO|tcPUA-C*_2>2rP$yysw~>9pG?9DuK8xu3XRF1!n+(M9
zc&SycUH3PS%G4A+<5a5ZnI<8gDu1~fDLQl$i2{ic@+oT$L;9msO~-wz$joL^9jgYc
zHI~(egO`uS=EQ6ff1$6f1EF9eiN~ArRthRYrUIDb-B3py`7@zdmcl79Br)s0Fun5&
z=bj6%8G3$PkYMk80@o+{W;_d9@q%YVOZM--Zf{y@aNPDtO_;8c>A3o2AyLB}U?9n1
zf^x9Na$v=cpM|60V0gk+FU|^IwiTV9LP!Lwr<eF5D(cai*p0ma1`4;{fE0#3oiDY1
zx>m4|wCqz7W&0KdHxI#s$i67t3_&`da6PbXn*f`dOW$;Df32&IiUofFh(~gPc`?9w
zz2y&3_<and&5zp;`Q}5dE=htLJUy7}YgHGwtJj0I(x1CWDJGFrX>meb%Q!g&`SepJ
z_$fp#?owCFF#QD2$@fQ1pL?Gd=${h;8+se{l=m_k0N%hPF>mrti4wv#{@^}VZ0*4@
zh!gMVjMqth;&XezOIp#yq$7)P;93cBeCOl##UD+@!J-Z{_umSaulgp;-;v0bB&aR^
zmrkh5w0-YusIoPq8@SAK?VA^cKs}rqtE;OYT9&=!HE^Ht?DZTEGh6e$TaFlCrf};j
zQX4y8RiE+R)%2U_G(%+ra-DzOz-8m!qFAa5Jw^TOB`o+d`M|4hrjHrWSa^l7e)KZ<
zVAEgoD}($27YRH&485MCk4@XYSnSwHEZyj3gr;Bf!=*#+r@bb6xt=`}6mpzIA)W{V
z2d?#JS>o5cOaa|~1+LYe9*PQA%d(&OFr0ZHDdt>1q4Hy>9d%|k6+Lg*XKg}ST2GV!
zX$g)b49Esu<Cp0tC_Aw}?&{osyZtf<?K2+TZgBK9Ho~4=HBQn&y=@}{P!MzXa8bLn
z^!kef7qp5$$AG(0Ri73?+Ds#iR^r>v;i6AQ#4ev+laLz&YCclCx2ulcDEyRZi}5Vh
z)C?x#c@0beewKiCmjpN>T+B-Vr_6hhwH$meu0#55C`rSOtIbvKPT1Dbjo$q0$tsAL
zayXA|h#h_Vc+;UrTN>1WoT5l26eqNCLr+6?bbCqy*o=^p_Bm&?2X3DhJXmcB8eFIg
zD!(yJx3{*Mg6oFearfuz*-%=H#d@ebBP+|aBPkqq{ZObc7tAn4>=qSxzepTvqF~mV
zmlVsSXwmZ;j@}HODpgQb0=awTaYRr%Dxi=+vJ%D>^$veV%aWHb<^B7u69;?jp;Bj2
zc(@sLT5O;ZF$;=scJ>sE`H46sJKP-gc=xL9(1Kz6lFpyaeM_HS2vLVa@X>D)ge|eW
zuV3DT^T(&Lbvt&q{!zLZt^Q2mW@NS9r1<JkUi_Q=7K2Z71lT-1J>^YmYr$e3f#}f1
z3PF<sKO!8AIU#uN0~vN6?t-B7I_xM$>tly6R<LFHQ+qvCcWRcVd^)Cgb_u?BAP;&y
ztU6`JaZu~vo1?1cWg?=YP=|FrsRDldC;vZlPrh4XeFppm&>``w9o-sQpPrVYE5rSf
zh;IoF^`>JZ&IGQHS*fQWOnw(GOIwT%6tCk@v>Y9H4>y;O0OLX{+@>;q$aA@y8b8wA
zTu?B#(~oCEbAGf^K=K>77%w7X<B-sm3$gFp2NzAA+JEITdi@11oA#c1f^RAigifu3
zeTsAXs;JArY*2e;aP6qOjYi%YDPQAK(FJq)C3!C9<RV!|?qA>9goUMo2eO@a?wLM<
zvI_<MwBoG|IOe1kWh+k1?QErf{+w^Hk0{NO4TXtXxpBkaIOgNk^?@W|7Uo3hYX|p%
z+2ZT#HeB}$C&Xk!!kzJyj8PB{t(W|BU$&pZPsn-9;oR~I&+bJ3i`R}8!ySB(P+?=s
zw<}2qNCj#;2?z?N6F1KJ>IZW$(U0Gk&+|H$aNJRpnS})kkyQhqq;syH>*cLH0?_;#
z63zPBEby;*gVXa)GqXyGit<^&!5O^YEF8^ju_C|l@=nxLw~B*kf1ergBjQ3X3SxY*
zx3cZr+mrHej0Go0{6;nXR<Q9n{K8u7Er&-SzhSUmu$<ed4X$?gypTfbkHZmRdZ#|5
z-N|@!kU2a*?LLbrhFQkDBmQ4$7*O%c4h}O~^#PwD4HmJp#V>r@lcmtDVyCqfL`+if
zo=gM*kP7#xAMC%YSsEtX=WJj%muHcJY^O6iw)_`r)1yV>kt+H9c?*yO5$e~11`tR~
z%o#2003E*&1Z3S4Cq?G${$+|8#Md7MEgb=;Xwy|h6~aPDy(<43`@R8_ku?~Z=`5zG
z;GEh$B(8q`ZBTyGivnvmK-Pm42$&SxA$2AET|dilk2vtacmuot%0|-tk!A^SC;=V<
z!NW7N@d3yW5?TB!?dw2NOlg!dkM<!GkuGu`ncZ_eA89br;OSR=QYSmtSR4Gh!_NN-
zXjJl%iLHN*60nVs?x54!`2I`XEaxep6MHQ+cXF=WboOv=^6jM;E3%AH@4bpSVj?ea
zZZ}~=o%g8j0l~NI{+3+QnS+A*<LA%0Q#)nsnA(Q>)6{O+{{g=ib{L~LD)YY+V*w2k
zK>ioh|AN!>e~Al&fYm?aW2^07%R19L%ogob1z{46o{@{tric4tbl$x)fFrU15iBea
zD?E<xAD3^YSfx4U?7dLX1aENk-5G*BdV=41P>XcH{t>&uH_l#@?(9Z4qnw-^I1Z57
zw;$cx;`9A;hvkut6zwkGNH*y&V&O;+_Sxkw+RJ!K@7-j7YA_iQ(;I6;JX_DOzi|Wr
zb9A!*)a>VCa(h#0?jF{VbTr3auz2?j%bLs#o8IT`zo8f6<Y8yVLg%p3Q00l)_y6(s
zRFmlnc!SxqG$<%W9j7)>H-}E}{bT=M>}hYA=8Gj@kup*QW^_yj$^#P|>E*c(X_bve
zz^0yd{^g#IBTX~gabpL&E&0DnJ)3zdLn2G4;E=^U3%}`MjrLeWZ0mW@R;4s0a*BX{
z-oI|^IODv7oQX(%+~6RBvNEY%{P=liA~|g)ZB0*bLPXau*nNVF_G*FDL{D9?Sj1@z
z5()7-fC)OCi%^(i{V!2RCN!-5<w1JMoPrImmPS*`%9dL*3g<F=20XTL60I-z{7QuH
zl_B0qJtfUwWPN5;pwB8PzZLHC3JaERl|Hs?{Zrdd3+}C3vQMan$!PVqtvk{c&I-j_
z$qJplyKd1<g$JJ~xdgAh3|QZL<W|DrmGx$dYKoTnybkq;x0Vy}ZdqLn3$}z&_>kJ2
zmysMQm+9ls>llc6ToJc~N9VB40x9STxPYV>Ajdc_8GQl)5BkIcnen6Ir(&S^I|gwY
z)o96WR5RRKsZ+7gmuq4SUoKD=m+3zhBbnF880Zw#@>b9l2Ac)Ygv^3vUL#}ZRNSms
zNVZn=WPAA&?h`f4d<Sbn55FdRVm`ZRmqXexBCEX;yZ}<Y5nRGjVH96*L!lj0K^MX=
zd>*6i4F3JykY+(wfvf|$219~702YBNEQw+-DJ|>Qn5huEhWc-2TPwgBBTJYJs-Po?
z0O+Fu4$3`ow^n{Co{7Chkot|R!)g&7x{e8Wz)F!h;06i69$RTGLL)#SBxeEe=+b5q
z_9#&fytNXY1y%v>gv+FEjPViWyi@i|MrUNZfr;aJD?03OUt(#dks8nnKuc;fvY6Qy
zD}(TIJn!Y;1+ec~k@p)yn;&Xm5wjt0j?F+vMj-b^bU@+&E{dz}2V61})VA8hN8~LH
z(~#aw<93T#u{xlbLb_Gs%8w;K5^r&^)z<^pRMW$M6!u0|J_+hRq@*GW4=M3XiK(Z|
zCx|l*{;ay+S6<Fw&)7GF*{SWS2r0my)a#OxY|u=>q}<{aM4T053fq5pzm-mVOfNW^
zQlKb@q)9dAx>;1&q*zm#8ADZETGk``;BGWr`i<?$Pxh&oI?#s0V#4DwDXWTC0v7s^
zHf>a1`k~iFpu;$6UDUe|ADjr>v0Tc3ikv%V%?no`?S_B&NE2^Snr68P21S{>L}E2x
z5&WP_>o0k8yw-_LV%M%+M~J1PB_)N<5$}T3B%VHtDS<#iv4{nO@$0@wVodAw`EP_=
zN#RSl&o3W}E7d5k=iIR)jL9h2u=QcrXb@eR1+?^%lnO~I#X@N)H6jzrvM~6|1gNM(
zg45gb8qQt2d<Qmju$%149yxgZ>{tmu9i1R7yC(asaBoxI+Y`2YDgiQy(y1AbGydB5
z?n9Hu^!491>em@~${vd(Bqa^gt%6jHOHO@;6>-9Oo?F7h!y%hspqjDo;*f;13vtuo
zn}y+HVu^Gn?eUaQApA*&5thD%@D>V=#l9b!YBp>8(jUBjw0rYVy~U;Ah@j-Q_G)`0
zZd^&j7qf-}e@MmT8?6$Iu^gjZMA>WKzXxu)@|$$#wnWLHV@%}UY`y4)kI-)Vk+%C%
ziQ-U>Ak<m)c}_T@?-d1g-swiIH=h<&is@IXX(^(;lKkjlKQ937Q3^-k3{ba2U>tJB
zYZoo;&dPY4C*GSIu9vbYEbrmNJILI`oB){{wO>_5hBcg6`8ReverU^>@nJfo{_K1&
zNWXKp#lRcurFHf7!ck8wv-n$~AP(=9<fjcdce6gfLpoUy?TKj5fnlC^a~Aq5xjo<v
zR!(SY37@{TKZsF5>X6K8Vq&tIv$DP}c_|IUXNw^@rQr*YH<hmMh$TW%*$8bjxPD5f
z7z%h4f5rQv+aK~@g`w7`s5_@Y{j1&LZUQf;W}x7$*3bOaON5RcJ0_BLWvI_mr%W)<
zQ|yOekB<`nU-@QxILjr^p3ShNy8^W>T?RNfW%jaFnTb8`BvcTwNbh(!nPQoUn*hI$
zMRN?qW}zLAw-=90t6(0afj!{vWyJT#=9*?f+E;kHJ0yS&6$P>F>5{ffvFE<gZ!1*J
zAt*QoH3^i<ccVz1NB~;pSXCvQWmX;oAwSV)Q8Kb7yQ<uF?rd6>Ep}wE;Zt{Md-7w=
z577+}x^;0>8Vg8yxILyAuB0rAnGV9ext=4vwgoMb`i)UI1$D{1wAg-$QZpj|YX4@z
zUK!m%Tz`p}qHz-N!xn*I-P8Dy)}Q4M`!ygPBkZqow?Cq14Co^ryT&#Nu3E>arZ>}#
zNf@-qtnG*xg<3AT`k~q};W5P=hE4j~EBtdH@!^4XC}4K$44CD3Ul88Y!`S$vvFw3S
zkDJN0KVRU2rTrQ56m`EoE%#{OdBO7tZAUBA#bv0O&fW57xI0SpoN{%u6Rgv(1jqzS
z@8bCR;>$|ly@&mVkSDN`I8BUD#C!)@hKYa>U)b3~@1*y|>ClOUn?caEShLhnPC1dT
zBuGv8@^muMaVSk39tg_O%^o5WrCMtt*yqxz38juXs#d}%Y@epnfBzSYK7)tG=PWy_
z=+5JlwhQ;Bs03(mLMhP%_Sg~+5s8!L@c~zat@T4ugimDMQP7o(?Sc%LT50)6!xQf5
zkkhfd)<rH@!z7WeCHKABsp%_gS5=76w}Xj*CD~(D5-rawmqx6EIZ_yjcKx?O7%=3_
z0Gv5@0MNi!|3V*Wdm$GNj-QTp#KPCdMuemvC8;<&r`(>-fmXIeRxYhe(oPC-@psH}
zJLrkAv2uw<kV+JIxosLBR8&RUgqD~6Rh*6@!k!)aP%_b%cVr7&pD<5>VncUzr*GQj
zhuDpuA%}xq%tm?++4$v!>ApQ^ks{*c`l#OyicQFGn7(Trk<vsvdZI@!R8q(k)Bou1
zPOUy@2YHh4LQ>N9FJaTD)k?I9)sTqi4t62%=$gB0mReed!VEb%F56?Sl*dHR`U1p|
z9g5SbHtye%9n4_fehXgsC|CZV-wy6bF0Y<WT6??+aEtSmceH$cCOxe}vxD%`_l<l|
zssBI~-}-e1%VxabcY{0ov$~T|!6tswt1n`zp(Z<1%oQIgqx-s{(ti$TS^BNW^j<aQ
zDSf}6O>$$E+Yk8Qk-0SDmejo+se;}IwvCFs3XxQOVEn-Hyc<&LOZtZze3zsJ4}XRV
zV6|#J7x^@OZttS4SkBK@uO?d}_M>la5sjC+PV(|n8ZZ3OLwyI~=QCPmOqL7&Ve;1v
zpZ)OF;;l6}cg1K^8$ZZ<V`&_EF)%P-<M$I*ySJ@(MRX-?8QYKJvc_&_X;dpM{^WnU
zWMFyEfU>b$E?RtS=R3EvlshE1jVfDQ^9htx$S%nFGPL+`^Vm1EW@o(Q@P3W!ZY^!B
zo38$FlYGZ`cEqZQ@*<5>5R|^vZ`)XNHzFtDs-A?Ngq&muPO{Fm&J~80%*RPGur9h=
zE_P9h!8R;EC`<fwdBF`Rn(yA6?{BF|VbtvXwBDLp75i~z9Is)@($lg3L%lV!vVxqb
zXUK|*KE!PnNHf=`eclBgiUoNzUJR_wpp-*)(l}-CXYgymOa!wRX2k$hnDgx0(!|Py
z+SLV!x*v?W1{q7V3f%C=DzfL|*RJE%`>~%an(^|sj)TS<^SX{Bh1jOlO}#1*Ve@UE
zfpr+$umRXba?aLi*cQj<Mpm)FD;uYh<w_58^7RdLTV#Z2$qCCx1a``JmVbPI-j$K}
z>0$K81m7@T>+eTOx|b^>vgRrENSR@eS2KmVD=9AuzBc{K7d!xkAC8bp3j@O2)Z5en
zCMJLVarChvE##5eh2<lnuJ4fTLtY{I!NLHE{UbPj=g#5pZ@ytaUB|~O)_Mpip6nU^
z4tSkfkny=aj!91-V!7b!!3B8<SHX<@B2}v#82&!|{l~8)<N|OKu2N`xCKP^BHb@L!
zzvrddyO_FLix^c0OHFsy4!k-39Q&!a9<To>VgdCZ2i_pWJ*8~$e#TOvI>0CXXwRY#
z)@fk>-JP6f9KHyoDSlk%K6h#GCQQeP-)5oIH98b|LVmC!KKeO6)b%t#^K2dVGw_B~
zD_Qb9Bj;OF{B;CIQ)(rD6Xtd}%vgFCrC;Sh=*6z=)48_@t<*fRSJ?G_ibKU>yOc&F
zA878D?Nk8*0)fM{$D+G|?LR%KLd4#xOVHZ~I0Y(}TPD7fc**lu<A0V@+)s3E_v>l$
z5d{jOq+wXA+j!x<-PQjrN}<H!%LkfJ<>#P<k{;A~S?yw%!to!b+P}E9uO$XQ+hf80
zisehi15rwy|0n^$ZHV1fmk@0SxZHpBtb22rY4h#{CI3-r0A{RBSpLhy2LP9hrq?gS
zq<FI~inja<>{&GQU$>w?M+Cd~e*l(L^*P;Ct$HzwMhj|CPASps2qJHP{aZr&q7RH`
z`OoCTbqGYTVj^~26cvF`TKf8AqzzQ76q45*>{iPK0-|nlYlwgwA}cHYqap}u1s+Qb
zW|hG6S|~yHiWpFK(Pjnz;f*5Jhu<MMJ%dPqLb+=cjNpi~prqB=Y^MPi&l^M9$`O%!
zgTs5HK-pVT_j+G6zC8>zK~=x|mP<La2jM}zG_4r?Cgr?M{O5=LLM5Q_4dA;K8eDW#
zt=blbMzlSkS(O73n1>*`y{!=u!6>RorA3iip(~DZJEY#H@`1G&wlr{b9<sF+UaW(e
zXh$EXr46XBH=+&gr$Y@@ic%paxzL!^x*wsQ>3bsFOe^1$ZSek#39Hp~6FBS#qQd&l
z8-IMoXUVAh5kE5M2`@m6wQK{IJqjD*W|*a3di8dP^2MHl{T?O-E(OqgCm-WP^-Dv`
zif4QLN{V|bPDOZyzoe{*FC6(W%Msp+ii#2{*v(nKZk9RV8tCfML~F)p?|N^-89`M4
zqs%VDx-%zuF$-+ZCVNJj8MazxZbVNhyhh3B5k_c$%J+%2)C?sAJEDwY3`h1lVoapu
z-WfKkD<AR3Ij2$~-Hrzj(nl(3<Z?Rbp~=R}hKF}3X;#!tHy|^SyC4q^UhZo%x?mQ*
z@qnk0Jg&8PU1PspJ-$dt`XyIuXne`E^I@FHij^z3=U9O&$hA2Aqhj=ts+&+`{Y;9{
zORyV+q|IA{C1-gF7he19!NCfs;wREtYuHw@UN;;;XEmQ4Js)rOI>8ZrrCGL9Qd08Z
z;A`I-ni`zaa1a#~6wI!D_>kA5ny<IHAhb9gx?b_laGCT68b`Ss*3`o_ckzJ}NEeQV
zo7_-6^a%b)tu)-_Z`0uHt8uDc7&+Zy=X}YdyElSR7^0>3m6I3TQx7;|`pWpgzE-Ub
z5E>IhuY4FzKS?O@=zAT3w*qT>o+|10wFwHz>mo-mv+%&c-Fb>$VAQ^sV5Dl94*Zx;
z+T}yprHKS>&)Qhs?;ax6Leg6fc=7}nTnf`s45ovAn3hkbW4JvWq3lwz8zo_^^aW4}
z%|S@I1x~U*6n(;s6IJoS-y){(Dod+$@T5L^{NnY4n$J}aY$3thnmsBXazgQyEYRTe
zoTS12)alIBUjhex_edL%8M(Tp)hRS<*RExbXy|Ut@@)Fktgp?tcMTL{qw1NyFM&SL
z#PHrbQKi$l@1Ss}TZ6mS&k*rENw}Ic`b9~QlUR2pMPPY9E?mfM<nZ&Xjj;wtANz9Y
z#u~)2@H+p{xX>6%4Q^)Xcex@Tx!#ZDFF4e0%*Mth40&TZ;OV50+Fa3e_W2JDy|3Ez
zM|)*HT^5Jk=~PA?R2WQk7(vesP<JHcseEmj88N(_&JE6Ag+{h$z~^_)<oYwOcn?Pq
z{(@r8gMI?|;X<JTizaW!^syeJ)?$36e+irsp{&*t%TRFALMZ1q0fW6D)*5gsU+y|~
zD~x=9(+Sc+!~dsOw${3Jl$>1!jE6M(@5OfJyyH|&D0N5Uen^oKa4my79E(sOp##KE
z4$j@U<1|!{+E^J(uC_a`;0C`0{%~w3_~>rxU8x^=@rt)p4$vpsww~m(r@}ny)M(%h
zyuyp<U|eKSt6kulC4eUk+6}5l-?~HD$94UzV0Iq2)+M(RP8cLm!p*gQM1yM6!5y6C
z&ir$O>?qgIa6LIFrwix?&nlc))T0q5QR;WQ)IW>}7mJ}XiK>E^2OGV+@ZK`JLez63
zZbZs~Bb`)i3H8XMy9C6dK^xS$6`0$YD-1MtZiq?X`29aVc=7y|Pk0XIkGl`u`t{Ih
z@uy3-PxN4u<4Vg{P?BG@er_l}VfpYuoZ)a_tnZC<pkcV#=GA+`a?_y*t;QS|0T4ff
zZ6Cvoh1`AW>+9jZe~Sv<!5Hc7m!j0qnEltm#<UwTbFX{h0n5>Ey4A^lFpj2LEf|4|
z^f_>gH2Lx7+Pb>@n5e7_kV8YQ8F0*jDjyc#UoY5x@(5IZkA-HO!vt-=*l(M`KL;-2
zO}dSC^tR9;3^+5e*m4*=sGGU{p!r_4$=F43>ACDO*;gAI{oG(~Qe5n5&>X2zP@x9I
z__U;#!++TC_3N!wf5QC{h6C4rfaeYz_p(wU1_}2i!PbLn;VA^Dpiwy%rktR3qo~t#
z5bt!L8JCS-<-c5tM%tD(qv01|M>JN0I)uNA$rs_OUfK&37lB_rJ<1JaH@KS^`3FTJ
zEZyca!m{IVBcrRcv#B7qC*<b}0(1loR0SXqW9nUdU-211-#1aHVh)`}4(MDSgFB&c
zqv&c)sxnlT-oj_tO)GH(VOmGWl+2NG@Hg}JjWKeyRBk)I^wp4(K>JZ5bP5~mbm;@9
z;SX4$)W^0PWl=@pO2Z{Aot-$>y_Gg=%Z9->sAQ~cn8W|YAKz0^7R{&G63QF&E8ePz
zgK!lZS1aeG@LK~XS%!U84yHX#BT0i>*V`Jj7OU2M^B(GrCe97Ntrsp{+)@Hc1Y#l*
zZllnKF$wO&?B3b*9n3=5g!9uQhHYa5UpZp=ZeERzJb{Larbo|~kj44_elUr|hgOSR
zNHT+zLJbiy87-7kQ7<uhun8og&tp;~5Q(xP@(PvGXjB(|i`*^AJ4p(z#Ub>AMe))}
zaoouuHyTaEiY*~21LTs6lu!Og{xsP(1^#gaEENgDcvHf-lzSjBsR4n|@qj%-!srty
z-9mhpat|7|Mw8N|HY1W92?PUdCquYM%oYi*&9+}^c@!cbWl2v!rZq{J!+Mc1wuBAm
zBQZK;3=&L!FaIdCv1BAz$rPoqC5#l5M%M~M{db0&v5)~Q;`h3c;-mtaBK8}+M3ItT
z5U4?%lAMk}+mpihXefAQvH%jYC1sA!z#R%M1pAOb@N3+l6aW-L8U;xw__dg$coad>
zUX);?ObSAj2rwEkcEXa9Ad>)%CL1I75~h*aFHrY8GMm9xxP@5W18UPCj5iA*{?16j
zZg9<ff{bA%5rgJWP#%H&ss>gFz=L)2KPwXe3<{J$rsWEtsUC<OwUA=Ee1O<D#+$(Z
zu?D2zIp77*`&&Akk#K@#5|Gg$gq}(17f?EHz9}hW1|O0dgREk@Nl3>1?X1|vD<D8L
z0Rt6fNCF1P)wvKeAq*!4Gpu(~7@`R*NW2(0K?6v4x{{fhqJd^kkO_D46pgtM^F^DU
zVtR_ke_Rob?-wO?6k>pQ1m;tD%AGaRM$<36PE%<}!x-OwhZw45R?t>T91NwO3hPO6
zJ)tO5i*ZaXSybHSP%QlM=)Dw#XCUneqeaZI5h#nMTD-xJRtw@=%5|`B(ZBU+7Z}kt
z7jSwxC%Lu$<Q{S%l{r9vSt{gJnz)kl%-1@c3UsF&Wg%5GKz@0Yd$k7lKJBvHq7t!P
zmkwsd)}XIoe<5iAq`V013q&1}G6S$LUK7H;M%wsWwEC2fv>7ebDK$_k0JS&q#|l@K
z4#kU_mbNd+3Ma<o&lqjuNn|N+K#`MufuNOC@`Cg&Fd)b+vPTd<I(8|X^d+)jte64m
zix~^XRO-YS|Jtk*BO<!?|H`cY5E?UZFb$)#0EldFNK)88!NJ_Kl1=ijVSdJ}WRv`B
znE!v5-mr=i7|iA)3}zu6GLDv~>Lr$sRM`eW|FJD0^=@u~hOIsw{koP_aL;cFtEVo7
z5Wi`6!=SUjyc~z7z*xV%;I7>YP4*1ac`LrTB-~4?*8=(!HfzU5$GZuDHk@3B-$FG`
zg%aXs9UB2rjbQI?IZ7F=t-0V;<SUTBz`-YY?;h1ypN3v>EDh>Q%7XO7*R0g?hYt%D
zAcl+mQPA@hoQGdw-XpNBDNZGLaoCLepmy%wYI9C{4P)tIk8E(rX{t8Y6Uv7@A&;&F
z`P#pO;!p?Xivl@jZ-kv3bPT-CXavInnF37{Jrq#<?fC(=^|nxN`mv=`yPm}cTvP{N
zfK2Ag!?_7VP70B|dC)<?a%`x4WgQKNgSO?+eFAPU%h54gE%RdkX01=u@{og5864P8
zYoGEFNm;}ofaL2<Nw7@)lILkTHY(d+cXezu`uN}#i?(nzzv$i<H-?Vi_~G?N{|k27
zLaY(Y*f~;$<mh5L2+ZnL0mZ(uL9CieuJS>3&U;7Cq^FmUC3E|wX9_01ciq&?8H~+M
zvdGEyOtO`kFKI#c&eLE{3CO#V*kI|&!tyq#m!?BaaL97>JjwbYB~(T%Z-+AeWOn0G
z_dHK<MZoO4cBHPX&iZUbY?B?b?l-dRTV&}?m46E=&I}9$Gn%)%Ea=Y<9mP(On1-0P
z5gDzR{+ila)mDx^6O(Y|;pe>#*@ra8R>k&369+{Bq8=WE4cxd>8;FFFT)~lO^u&s(
z7UQTzU|A!Fc=UEEl5&jv;*b1eh^``~dX_H69-Ct1p@Q>PJTIynurtdW0stDNi^=mu
zr2%b$(pZ7zP-uZ#gkCulA<{<Myt;Y8@HW@Cgz;2Ts5s8;@TxSiCXE(yKvw_N+1Xd!
zQ9ZiH4B8${OWl=@BJALh+ZgxVB*jh=v^<oz-uK9bGG;I@Hb)FBF+^Vi$X^puzu=gD
zWc3tS28lv&(ta$v{5mX+bbv(I7<m#SEqz+jnG4=WwsQ=Ri)a)a>wbTH^gJwKA_28B
zlp1ANOliECvi#LL+`1|BRKPQ8HK}j@)_WaJZo@p9kv3z4{%-xEtsT@;4)&;JiXsbQ
z`j=n%q_DY}@6~Ux<kl{$p^FVWbmQ@6I%pzxWvDj;*%66fJbuQG4aHv7;_jdsYww$D
zIrTdOpv35AT)&i5bX*rmDL&;dJcubmN^aJtRUaH~w;oGT(m>Hmaw)*Eei{_zsi09n
z-z%ay-^%jJ@ZMrgt{XDmh#53~N<(ZS<EUz6kBI9Cv$HYP)xf$}hNdt~H#W1lz_=~K
z=N_V%H~3;~glOr?4!aaS%%O|m4{TPH27&4>^IxXtGfai|@Wa((KI9YXo`|pk;)BX2
zS=uYCa9<$hm$dq+b0gjr|C?Gr%bbHt6Jtl4V-Z4_@W;qI;8e2R%fasR`Ow?I3j8<e
zpvKNCC}Vzsqq72P_rc<@RQf-;Zy}2&_bHK+9mZ}b$sz}3MAT1{6PPXmM36R1^2{I#
z1`~;s2O3oCd>6a?DHBSPK{X-pr`4W;G(cN2!ayecH8)4#s-m(Ut~;<uMnnANj$ByH
zjOoZy+>zB{(gI1WLFff7?><hfGh_k!3}kIU=)e!%Dtg$&Hi(b~IkSU&mX3vJ&scY^
z#G@^T53U{S8pyWCd!X;_l@UgvOuF=d5C;ln@b!ZO1+kJ!7hipUKy7D)K_n5Xj(R&=
zK;8)QR6H-X&<VnhSZ;5GiBlgc<eUuiDmNGSOMkpEGQK(F&=15iF5_k);AMqxjt{Mg
z{`ir!s|~(Lb`t{iiQIl<+7vzpBIf4DoH|yIAUG=unhrz!EW4UP@>?%M&fO&3=U&?Q
zv!#odHk05#ovGmR0O{C_DFG7x=833LK{F!;|Js*=czIhEmSS=G-NHkPr-vF}(!NVt
z!_Ff2;t@m2`J4bdB8|iKKZ=C=HJ0su^d(h@g{o-(qZ=C^pY`0d_E6s8DkJ-tD~}?y
ze^|J+4e9Zh+ZVVUP_S<{7cVeSa%;w)E?LaIO_vt_ju+FRRrIwbRIr9E$NlL@*X)=G
z-;~_FK|IwNS9?p;(dRKUe9vq{cye(N=l=RN6#unf`Fh2=UK?M6m+XONXeGAn^|g!_
zI<%cC{1(twPw>%3J@E}LE{qprM~|U1oa$65wKWhS%@$%5wp@fEj5)mc)Ee$>A~4^n
z)w;BAg~ym483jsyXknD)UVX2UwI<0_T}nO*4`Ge$sA5<WtT@fqfHR+kwC=*PjyRUI
zMJ~+Jn?<TvnMEAKmfK+4xJ_tR;cua9K5^phcC+Bh@trC9idG^2zD@WPT@5><+#}-0
zg-q|rtV_EFAkp%y%v6&B{*h|ibH2fCZ20q^_bkvCDp;9K0@{)ynBipvQa}0d`i@6*
zL~aH<AmDOj7AZ;Wh2ooSRY@yd&2acQ{;}cP5V9O-^5Go$phsIG#2$J%ULW30eXU~j
zuyBOw=})DWx|nL&DB})j7`)IdV6<41Q)oaV0fjy@XMVJd3yKK@MaC75H5e*%vAL+B
z8A-SuJ&jGK^a{=VZxIl}8VNzBQSYsB>H`1?d9IBVPHbD7JuuQz7*OrVB6`~-AjY_3
zA!!35ioRw6dnVyE0Z9bR2u{pnVfo${y;T;+${6R06%5>~$6!~%ia*=ZwklpJ`?T}U
z-3k#;$6@}Y2H=C@W*ZO;<sKS&M|myy0s)TD6P&wJY+3fOigMCJ=#)CJGq#IF4kd|3
zNo^|wB99-m6dmemPC@pM>=}qg0Z<C-T?L3=<bq)fNXL;G$8?1IoP1P_`M^zw5nLq%
z21hbs^B6IguZ!dq>J9x2_05u!lEx>w3lL?&whJpdqGJLO)z}V1%?z}&P>&Ys<UZ&$
zT1kv(=0aC?jyL1Eo^}-NjRL>g+e>m6d7XRvGBv^mngJq70#;2wyy<IyRzW`H7elZr
zUwsKHzN)5%z#Z~`xF8rdlJ$9cH8u1tdw(G72$aZI>dRJu4Ixy9a2Lh_X?Z6}9H2w1
zeh8Wg_>`N#ehXX2$`X~W4OU)RO(PI?q8-Qdbxi2o+G{;cdCHMC(9ii^2Y0}+JEpKV
zhin{EMUk3YG^2&SrvwpUUlP3nS$D><AOwzq453?;o`mj8K^M$0t8ujTv#Y8~+SUc@
z$Byj$s5o+qrhQ(cu%28`LX-}zU;rWxT59!kJZuEwL?`(S!yLlt7=zk8(xIi?a(1A<
z+WwE%i~S=2@GLA!yJXsHAiDyeLeEW^51p7uJ1`_Bwy`RgtiRG{gb}UHc`rb!L$vBk
zle&$!_*pq+J4>FjJ0+NX?LP+X`h;6pp|afOJ0;CBAuXI2-j8bUur+9=*R6xYAJXm9
z*4CDsr+{nCon%q*b{iuDygT7~fuHII|A;-Y=4aNHfK%*9?kDAuW%%8XF=)B#vl?<Q
zzp|QaNxHPPOlIjkBi*^TO};`GnZC}>h!YwEP%F3}&XLR1U9KeD4?b?q!03JATUwe)
zZ;{&X7ZGL?SCR-Fie$q(@!7+dHem#Vkv55#y{~uGWyyWMzCL~(bd}4DFmPKOl&<@2
zpYREt)f~`|5IU$#;qA9hHH2>%FP7Y87C~t6ZVfBa>^ya1XP8(c!TI7rlg`}^3W?C8
ziRsaaGdgmA;GuwP)1>&reX$pJ)ZQ;jCjJ(4uXZF{FA3W=%YsXU*6=lYY=aR6EKKhO
zJfLNLWZS~TU|ex)l2j^KhZ~p&Sc4F04*Nu9kPhv(a2sFWO{9HCW{X|)S5$2+E?9WI
zz2Ys2Ou}p~M%qXSn6OyMC4W}e9ov0>zv@@$uEb^K-ogTxMb;+g)m7QrGDSXc-YGfB
zf{VE;h1t@$pR=IEImsiWQ5|bp0|P(Y)4l!PmG{XwtDFiZ>|mE=h+8hukx1;_i!KXv
z7{by(->`Uc8AT3JM2evb7Wv;F=nLe0Ds~N1B&Z@HZFbuKgz5k%hjbQT?_w$%79Eq;
zLaw4=4x7?8*00ZYAHbCqQZ(hat|Ffq-P<WQzlaYzzWToY)c%yhb$_zmi0aeQMBS7g
zs%mvzXlMmHAt_iyP%)QZeFlP3WSnq>io{yN)du#i5AbUEW45uws-kv8+DI!|Fj<`#
zZg@-gwuQAVy6^s8KQ16@M(%ABDhE_53Ed>+J!7;9h{3x59ETTdx)b1q$~DX`A%Bbv
z%p^pD;etE>Lf_n}XdK<BvGA+;3M(&qeKLa9Ukkl%6z!jh<Z-(H4Yn3@cZ?h!>8AZ6
zllbD052~8BVdl>R!V_S|n6zEwvVR(`r!YGJo{)w<fW1nlwn#ht)&Kq^Q#R;?$z<wg
zjwwm7L7|4o$O$9x_rM0^paZu+0vlAUHtkMb>HpSV05X?!-7I4sf|$j~3Bk#4aez*P
z;+!cE0w0Il<66KRsbppTHOI~Zh9{W0cKc-^t*S^G9e{F*faxwD$`T9}n+l4JZL9)+
zR0L5?5o3v=ICqr|LtKvgH4hx-4&^|9*?0sBa(6bJMkMPy>!am1K9Ai9i1<RxNj@U$
zNxLokt|b4UfOe_-Cr4`uB=m2!%Ho<j>zE;Jhy}B=VR*O#x{p)!0~y+Wy6?btLjSWN
zJa|o5hM#4RB!rN*8WL>KOnJ_b79^}5*~`Dj8i_>^Zo+2MrF|}`T**FliBmservy|^
zV=sl-Y%LWkkKzCI5&`8T=AW1-soDEIgJ9JI0l0eW2R-gDL-Se|V{W+4N=MF}_#a}4
z0_zsH7hM}0;g~S=N)q>=2AOq7s?wW3myC;uej6MUAD?DX;JpR<MXL45d?`@mn>#A>
zh^*`((8|U@iv%u<I^JE_UPX$X{bEO7T*ueCPKxL-n^WO771onqfNAjH3s?-O;7z3Q
z#Fe(}UElFc&hJqjWmM)XKInf;<Dx<uF()%vzmWqP!J{XS36ZpHl%sDd5P=|3QR~ns
z_CjCws$liHb&n0%pOLW8Do>|ykSlO=3;9cT>shIJUX#(24sF0_Z)iP$og^ZcWr@8s
z$p4gzPfq{(q4NC1HpB1#u`yL_aSJOX6ye*#w=aA^wmp@hwn&;uB85Q}O=@XiVv^W`
zX~lWoSUwf3(<=OwpC?12l+UzOTa0&cul_^`>&nzj;7I}1J7E=;BuaP#S|;{i!8)2J
z%E|u~v?T^dA!A$>Sb`5bQ=$x-1#}OMZ+ueyOqGPBPb8orT6dCQ$ze3Pf?vMj0XmAm
zZ&z$znVG6-K+shFc#r{oayQ1Q5s+Ta<Hrh=+6os*QS>Fs=(?^AUk>L606LR9&Rsq{
zFy8W;92lL57k=T&r4ASdJxZHd*K@kAi`#rRwCu&i!3JI4YC@im5>>_GE8HaG7BLYD
zPjF|y6O9Drs#(xGP8{)K=G?Tb@*9RZHN5n`AM1hCXH)J?BvK=tXQ!^8{L{D+5NgM{
zn{=Ht6S|ps`2F|jlO)Lfj$6^+Jt3VzSBDT%Gc${~f9R^RU0I4Fktmrm|HVj{KG30E
z5NY#pUEgD0LRiJ^eK>B?^rXSjGtn<_jnR*dRcRum->#G8*6mDV3o8ZPi$H*+;bwgm
z`*PfDSZGLV*1ArCTer&Q&OcZ_$_dz<bAcSA3ykVXSek2@Bpb3sXfV>|4h>E>b=$`S
z(X3)kdgBB>Z#XT-RZ#}vN^>a0bt=C+2NJ?oFiFL~0M!(;jI`?CveblRK&ct5b#$hG
z0Gqj985e6PQI9K*Ul7}*tQjX;?6mz)pn#qTkZsMqKr-}_=yIIuezi_AYj#5G&V}P~
zjY}r?=6YpPgaDTQ5iR^PC`<_Ec`T3&XY)hTU9sKcLgl~4m2)GO42}QeDeqs0xdYfy
zZ139~S+bUV_d2I`b3zf`9e0AFs3Re5R!zW!c{8U<Voa@&UYo=|xzZvt8K#^?F)1_a
zROGLaJwTd<j9_NfH6$x%iY%K|{IP3gybzqUKtT$UO%}GCo&E(k(dL5q#^R51^>Cl-
zu7W*-33#EUeT@s;MNt_!+m3B6(hW)0jJsC>K;v2)gIo@d_+@35ZE>4?i|);`D7IbE
z=(heI{jP|Fn+?Kxcl<Q+cEVZgE2q^z86*rGI42D4+wAyoy(_n(_?T738QK!OE~=N0
zj?mD&_VsO=gyl4Ny-H8>-Ym<#@7lM+_yy>K$b>114isOBI9*!kS-<ScYI~lly^|oE
zD(0VNs;U({^d;~#`#St?v?)4}T&yB<V-oytDNHQxfRWm)&>;VE>XH!wiK)&>5Y0-V
zurM-wi$o@gJ!KC{Nnwm#d(^0WM3baNrV!&Km?n5@`~Ze<3p1DsIHpJGjlQx*H|MvA
zDQp5D!c@wv82i^@Y$GEV;!D_frRklEqRhpEye)fT!Wvx7+GJu!i2}PP^xHfZGl>nR
z6}T$sW~cSTE1u!pi>5rE3_{;snm;Dg%HL97%YKSM4;U^GnGD&4LQ1%JQQrCd76RyB
zv&rOAzy(b<8(7QGWK7rID-$H|;J3JcL?wvL++O)!C|^v!F(*xJDbNE%JN~Xk0|!Pp
zIBt;ThBmz%ZX036m@1r$)W%&w6HJVdFDAfeAxv^W0K}%UA97H|fN1!^k+_LRhl2d1
zI)iZ>)N$@LW?CWTEbNE021oN)xkEQkc|mBen-;8vU9-!Q_;WasYaknOzcP!QjyL>n
zd95Pu){G&^SyJd^xXtPE;es0hwJhFixt#R$^uj0nyDD_0iM1~FP%ukwfAnHA{SqP5
z(Apt4t-zq5@a3wfp+DvZf#Ik45!1d_Gvh)kdedR()EuFj9`VGFOV9*ju7lz0h(Gd@
zE$<2=UZmeHWU2}7@F#c2dqMs!aZLEZMkN%UXB#g<{4~yL^xROWS#gL~v}q3|Y7zns
z%rSB$nXeyHt*{uOLIu;tdvJt=J5w2A<H6er1kXAoxq}~F)G9#5;u7!=%f9=%z)Z*h
zPIXkMjSsXj2Vt~N#Lr05<!+dbTQjR2k2^FmKd`VlTUafoJ10oVqDS>q>}N)c&o}Is
zWkGNd&gvrCJxw%lVRYBipXCWb)!*QDR{%J75n7&$kY_ybz>~1d7y^JJ4Q~(J8F)9+
ztmJaCNbHOo9=zM^Hqf%&UTL`MN1$rDQ2n1TTwFpSN{P&4l9vQ*JS&h$Uy`dL(pTHW
ziUL((-H%XTjfkKZ1W{K<J<Eq8b-5CPwFp~CEBtOu_#L^_MhO}GE-J~7eliVN3`u>w
zgsy2@;}HKqO-rhxxFysLLt?gRK6~as2g&6)$+{4g-G?6@I5+xrgw1Ui8ZQ1b$LKLV
z-_AgnT(E}z`LMbjNopqtIQpO{7c&$cdUKq>U1q$~YE4u`73+Q<=!<EXFS{MugDFP#
z-cpth#kGP<4tzM0h@-FU#W5;I{OPMi8|b8kh8q4p#1MMB`i!t+N5ht6&L)%k5l6zZ
z%VzaimWuMh8qQt{YP^O=YSK9i!lJ=UCWYAM!1W~95dNBw{UB~baHZd)OmeR)YMgw`
zo7?J)Nx!Mj&+>^}e;`;N#~#yVq{vv;b~EoH%dxL=p*S7uT1*s=KS5?YPaT&`0wqf5
z(?VE#li+A4g+^%v3-{_XoUu_2$(t?dleO#mJER|tm9;?QObAfh?s?eEdMQ}ryo~W_
ziahTNNN&mgBr#v5SbXtk^#bT9)E}mI;=ZB+CDo=?s_%lYWvqy#eBQ=t$NNDIylV$C
zc^t5K?9|0f|1^0p&nXuZQVz5E1J)%0gD5Sw6n)!HM7*Hh$}g`F%LV`(4y$o0&|VVy
z`IaW!<?)e<*6!PyTbfuBeqmAAOI-2iM}KchbTR$1^d?u!ua7<>osF0a#G-~f)6Lq#
z*_7hMNLp+9vS$E&5^q9@Vn=#@EJ9X08r@&@C6#>SZels(9*j`t7T@I7qI)&#nj1u*
zMxjCd1gh+j6A1QU6{q|NP)vYVepP$aNUxjVn7QQTkuEp4e?Y)0?;{&MIbcwxk;u-k
zV-t=}<|MI}pezB0Xm>@^#>2|P?$8vLRYSAD>)ahV?P!u|^f%x$gQF0q=5g!%Z29;v
zUzGmA!lx5Hm~hO-2J42847FT91?Mah8NMgC2Bsc{r)~y(j*=fP+O;sb>WL#^@!tjv
zNrRG~k01ukYj{KNr<U}Rv$wey#z)#LqAs*-_+Gup%b@vLN2YGW_fddqve{TFrdiJM
zLYOLPg|fUjZA<o&YE~6n#9>;Zn*xygyHbfs7FU%>YY)i|&H4O$Aqcmk>u0$leKMy8
zXe5%sWB8Wfb;)3C?|R+-_V^N^IK7qtVK^NB-KnOdfplC!Q*)+5bjGKt2jywF!7%mr
zxCLfldNl-bYjlgJ^td4;ZnXPSx4tL0R&g{sfk!yudhzFDK2l=)qwiDc!5j0iBEa6{
za+<UgZxi|<>SVAQr6u(?*4vDzT*yfj^S#t_lY*_kGu<9LV9nz|bAy%KBY~;-2!sAg
zld15S8xS)r1xV`fAY3nvV4p_wb2{F~J=$c<vbd2|Bkuz_?;xhhoE$dg0U_^j8vwbh
zh2b@dVD0jUpjmQaVbrOTVQxMn3n2lF&S~J>#R%7wcR{tE@To9^PxDwKQ2Y$BK5^zI
z(JWYMGpKhu;ZLf9sT-R7^p9B!LQAWMh7G?g7gjU1W+*=N6{nK;KsmIOi~T@kV+c`f
zptE*@I_I{;XB^Sc6*>rUchnIiD0OZ_GtLdF+^dHt7Ctxak*lm<0cw7HHUonQ%{?C|
zsyj7_q<=13Ad`_PzL$EEA?BtwprKh08BA|A%VQ4x8zW62q`A>I!*ZYUJ_@C|$(1~Y
zG4uU@C>1BT=5)E(9~*CWOG>X>f}9wyekOP+jlG6k%I#K2(k1E;`3FZ7bfGrgy0p3P
z%Qy873Bz7fm!av#b8i^jCsz+V2a)qD@<axo*Dw>-{seza(yW6nqz)5-7w1RpQ<cip
zwJpmLJ3}yB3E-O-jiz1OKS_(zS9UDI2Cnd$0=(wMI*u7vCz);Psz{&aHOxGe-<+<D
z|FB>oVeTD~mDfnXxVI&8;SN4N<j4b*@XrpK6QD15yJFu;pUlLip^xwYAkjbKceID{
z+?)^0@Mp|K&s^T2c@osDLT`$wm|IlMu4(?Z74h$A=s86m${mUZb2#p%S|3yQf`0nP
zT7D`j{~u#9l+)&xWOG6nHR+<7$igYJimFNf!Y^2AUc_5;rtY`;0`-Z{IzRcBWR4gz
zi+Vy(0A#>Qeahf$UJ!R{(Jz*Zjc4=dChfY|*yyfSIzp?F$Y}>cf7VP7`SeK~&f(nO
zWLy<-^w2^RzGk7>iC4h={w3y_N6sGZP(<~RFLLT9WYAj>%%9WmJEytw|4%~ym09^W
zCG4EgosUs7r(~R;H8Uqq&Pgl(qx1!+gWs8Y^Og0p9Hysp)!gT+S<d!T*>Qg7+&NX-
zoUQ=~c>h*HA->6fjiFVqE13rj4%EAuv;<hc7mu#5BWQ;;ec<qm8T<+nR+GVQZ@qef
zQMjcC)%!9f=3%i+TIy55H}UC|<5PBvk)C80QP<d!Uc$*o8(dvn3ivCbL}S-*SC(0S
zC8e~VdAYsa9486T4D*>QM5{DHd32b~nu;%YNIV0!+Lte9*?WaNon(qc7}qQFa-F@C
zvt6G7cX!^D8kr>9(RXF?`45<o+}+b@aB{V?A9ts$Lc}rWjiV!{c0}W#F*RHbjJbDj
zJDd1R9q_g(8I>M_f{P4C`um56hZpO+zur1FI`rOrr01ld-1zFJ3`YLtd{ku)?ncH7
znw2Ys+foJAmKaxSOZh7XB#ae&btWq5Wk?2$i7n*U%|20Y*^#tKP?>!6St)mZQ>sHp
zqLeMUmS&wg@k_2iMo^#d7}3qVGgNd>_}&GtHgL84c)WS6vi{p|R3=Gc%)KI!owZPe
zwD!riJE3E0taVwIo$trCLF}g=ZWV|fWlOf>5A~=K71Wx0q&XNKu{_Ru_td&*0s;0=
zCT|lIw2Z$k*p}$VWzzX^c|&5N4efB;glp+n^UX_6zLPqa{a5R)gqpO_wr)mafFkX8
zn<QJh!RIslpA*&yi7n?vk*D7j1*zdAHZP?Sz<1Rd@QtYX$#y9d%c85pZD_?eKpQGv
zLF9hsR4aS>r}<FB`O|XYog)b}#9)3cnJ}HUJ)>#%NA_=z7d*UA^{WEDf0C6Z#L84|
zBTojJeMNsddOS7kExPyf?fq(d#?J=L_Z5bnL{G=?bf#{v3b%nW(iRAHTh9JA*(T?;
zpNFQE_tcCYY=S#|Mj``U{;_zK+m<&~K)gPixM(^ZP!qf~FFEJFVx=%HYh!0;r@+t+
z$@8PbQ2X}KEy9NVSKAAev_G%-cR=RR_mv~7^~r&ky2*_qN4|c4b9@a-VwaZvJ8;U`
zeBfF)_3nhH>_#jdJ1%&ZJdYETH)}Os`3pQxVtrmN=F>Pgi;(@mg4WL}Ul}P%RNa#D
zGZzFX@wD@3{##Q3rCzaq@<!O60q*WH_sIw|T3lSn8=#UD`RvXdyfIImmI#4wGwX+y
zpL6fCjL2L`u@k94>Q&#b{Ur&FV`pAUjl~CAGBG1idRY0v$M4i&C96aX&=Qno=H)<K
zi@&9%`5{6I^rr3m4W43=oQ>9Ib<e|>mn0Sj)L&XXzEUy{n_vX=ck|*m_TfdVEC+L=
zJQ#7HR$%Ok_&dJKetX<P&_5r`i(U5d#sj>rD#rD%3eM{$Leqm}J{3G3uiIg2`!4|}
zSVbnX_wB_kx<9;vXosC>p(c~viK&;RX2JBlaNBCiAsypZ;f61aFgU)GnpgRYBtQpR
zy7>P7Fgjo8s27+)E=u*ly(@N&Zti!)UClKYa(BcniVfQHTdF}6(!7MWf_87q^N!&!
z4^}5!(Eo1DV7eCwlcz(Y`QL$kaSwjzr(D&*y>o9f#j}q=3d{amgdICKUBT`h&I0xc
zLY<e6jx4KnQQDNnDX*VBz|Pyy>>v9y4l*&M-WdE__}wl%y5Ci419w~CTKev{_GO|K
ztYBl~{<Za6El5f3T3MiAB=(RxLbv)okk9uwY#zDx*VDPnpDteiTPAb3g=3;Oe8yFD
z*11Z0OWC!XIGprvIiBfliBLj!p=_Ys4w2K5(2H<m7H5g!Pcd&koWEjp6}?fM)s@^v
z;kA)>tfyTWokng#o{*N<7!Uw;AC?zyD*K%;AKpZABpnckPKbY)vBho6zDid!*7t4S
zp#;O$GF@nLftDeeZbePiAd~fDrY(i+wNs9tZ7q6>zAnb86n`E%Z`APSxOE~Q?$|Nx
zBKxmj1R;EP&a*ROeCF&40nGzQ`u3d0qwV{9SJRufkJHil*Rw(n`d^2;df|#6S1#c|
zeVH*^q5Fwy3h}*C@5L!o_)<EmvuLWffNcjwJ>a7M_LG0ugg!{PPqm_Y3x~AdTH=<m
z{*<AGJn@ljFE$VtM;dA|f-7FTY)9GQGV`vmy_x&UmP`2<9^JrZ+!A0>^{V=C{mb>q
zhuea($~rPT?$KR?IA{RDEQ+=zmrd}|312yGzId0*i+JLbY}Hcu;vt9fe!znzOBB{b
zu`I9NlJ3m0nYg7KZ-%E}<1y{#ta?>Z5ibn25(0+IqV!(!5BX7ux4*H*Zze8AYJzOA
zE%Z}JH?Hq*RvO#+NUuaY_8|hVt9J>|z`~S5x>OFg(WOFDUAaVd)1j|6Ms`Le`?gup
zn)j8eRlT}wXeb2(_?N$d9FmqorTe;ekh7$=uzE|}a>&&2GG;UG&W?t=>xJcE#`Qc}
z#k&{0tblaGCl!LWtCN{nt(u@sEETNz*>2h_vAepr)n4kVM4dP?P_sb!W%UBei*`nK
zdUYk6<CY`IogLzzF1!zc;ClKqF@TbXh|9f^(S5xuw<Is8-lQ)SrFW$3+l4PYDT~KS
zL+V|JEMWh-RV?t9w0|4CB`)4-TyLDkU$8{ywy29h5`;mzgDyqri!6+4Z%@btl3Km}
zJ{wQ>V2>q=^1aC$t~}a@{-w@?z{G<<nD!mB2;Ys4s(Z`Zzd+jP1*3*9UmTWGM_<dZ
z269v;y$b_`P@yG~?qY)>BMxN44Rn!W<*s_c=M<brM*E5IHX=xM8iF+i5$DzT)X5uy
zAjKGQCrpm_&7K;eS84)Z9}Ub+EAj1g=^NXJ$;>?65eBgMdA@x>%@DSpA}H`I;<%D|
z$30lOa>6AXlYw59&2e*qgA@Btd+mKG_=7amoH;kBSE$S{f9vtYR`y@N^!d8-0a-_n
zRx|L7&{o`g?}leGu#Ra76s6r_3iU<8ZGL1pODooOy_tCF%u=zww$kV($1s~5x1g*&
z`b)UTCfj)M`V&;55FLcCF~_Zhx>_QoZ~UV=I~QB;VIOcLoGLXR?igCf(bl}(eR53C
zZ=cn3qp&f#%Vf^Sm>;H7o&e)tXPz%vxxVnsGiLQ62GfT4#m35+FNH5OvhESpY0w{J
zU(&!DGjOfzt$n8NS!G7IHIqBcv3&UOVa_!<2okej!c`Hpxytpj=o==nd1g@J`F>Sa
z@t4;TFwjHM9*YTzNau__2x73iF+<E_P!d|=Jma)%2Io!d&%SmB{?6MwVp9xrA;T9!
z(Nf;$-x;<R)AQ;#IG8xO_514SMgH)2y8wBikeG7lM&VXYf4Td97jLUWyA(r1!yBtL
z^FHK-X#Lbz@V@<O^cK{agytciOnWo}48Em#7r@0K{Y$<zXHx9wn}2xI$|d^x`mW!7
z3A*L9<R_gTrTOsswHer!fq@!{nE~Rjis|_n&Np#5cQUF&W9geWRip_Y@_eMVp}R~A
zol&R~g_GjXkdGR9g9W>{^F>O56w9q#_*ae6NL18)s?AbEhD`<jZUh414Ttk<C^eYp
zH{vgMGUJ6Tbj?lu?tEH5h<(R($eV=JSryBT&G{dpLrZ;5xTNznXxjQg+F!iIRp#x<
z9#65&_4P*Ot#ol767W!{#ZmtDxJ5LFYo4IbLREv!2YI}J8Ur>R^eT@w3;jdEdvR`z
z#ZY0VR%G#KT<OlI_1<Ih6kT0igoEm9A{$fuxS-lmb8)cdeg8u0Jyy`K<AKGEpN7He
z&@Z#4T7K!CeEX99D)tEh_T?LUc!s-hlF-NGaKmd$e(Dqk&rOFO@>AD)HlJ>7TJlLP
z%XRz7N2ds5Jby_moEIt;YWY!Su9w!3t=7|4df-P!O<`$}Nmt!ZhA>F#&h@_VzMab{
z(y;ZZ(*>iSr9mQ2CkF<?E45fd)pI=|UAi1cyc!BEZ=y14pfxI#W3>FaNP1+DGRZae
z%g-}*{On^xw|t<(=10Rm6_zndXSecb_1xqDHitr`WVm?`S25V(Rme{+__#T;_~K)Q
zi_jlslUec~+6TfVebyh+yk-#@?PQU<trdpL%*gnt6gwJ{Q3J>_gcfV?DiGT>wSvLk
zkt-R{(AGNi`IaG{f-v1`t)F~{p)jqU8QTFaCq4t-t3P|irKBF$Wf0b_eVX&}!@=@p
zpRZ@Dz_%@nG=RDiex7N||EMP}9<9ZAb~87wpV<29HJ1}~9T_6hcm-1nglQQQ(x72A
z;QnDLClJ;hYTg|tM8@0(?+?eVdxJGo1YR?&0`%P|I`W9~0t<2E%C#es7DJ=DKdW_m
zSjF&Phf>a1e1K~;(#pz}qMXrJ5k@mEwH$j2=0NhJArPKp18$;V<a~d93o4{VHLdWg
zS|$B~InlXzOVw#Ds&1~;HH*W~E8Pn_FIVyb`Ele`qBLDF%$t~0Tq9L06+BeDrs}lC
zI(Yb|5Ap*C970;iNGL&t@vOzso&|Z=0_B{MpNy+S{^&k^pTOrW(eL|QagsPmE+-a|
zhXs!|_Oa>EUU*;T1mg5&+Pw}tKJ)&)k>`79_@A2W=sH{X=G*8S&bJYg;qb$vZN4#Y
z^bIng|GR-pFR-y|HVbe$S!~RsyogL1{JaHLRx71?pvX)jbZ6*JLRy)bDX|g#O4f00
zHiPA`!(v=jTO{#R=^M_dPdmzVvkuqq`cF;?4nQ#T!@wO4`m!r?Gd(fPMgHa+1@<9p
zhObA^eO?>;?sBda12T8aBaN}|AHRP5T99MDI|KR0$J3oujD)n9M@@@9(0B>FE<R(3
z0vv@WIe>2hh<AxRylzX!d8UGuRV-qUf%TqUTQ(BIXxS8BT_Z$Su50rD>>PN0efnux
zb`I5mUC-cy)K~amTkG$P{QGogXj!@G)L*Fv_&!BnQbAsN$1E}^;U7DCej&GRYj8CX
zQm**s^~5l6sHm$$;>Fbea`yC=nQ1fY(9+4DeHw9`X5)UIGTr)b&$VWIG~{OO5LlDQ
z@yn=Tbi^JkC%CK<p96vCspEPm_=4%IsHk`(R_EKk5qLU*nr~N`F5=wJ@pzpf7%PUy
zA+=$kz|_rP-C&2ORb24T&ef-|dOA=jCd7!D{gCu#L2ePpQU6<u>}~7mI=i}J2b<C_
zZqct!mZ078*`sD<v*T_1Gkp7$;F{7-XCE;e##bJ0W3HNPo`q0^%jdDj=Nswb3R)rk
z*&>0_+mU*RSZzD@)5uMtrLUeaH;!%MA8cvsP-`u=Kd<}UnWNm^DA7rLlg2sl$G`<0
zq~05>6Dx4|laEv#Z?fJX%R3udfAI>|GW<|5EUYnyrNhp+NY&-iT1ZlYcage(up6|Z
zU|O#oGmSh{cf=Nf){PQr1ou6FAJmp;(y=R+H=?$e8I_OC`*rooFNpgjfkyDJhj%hx
z4jhfzE-*HHVEsL)r36!(1nhxgO0W>v3I+iv!!=o7^89&IV)Q$BKe)h7BIy_#P-NNa
zu?~Ct-7C<7k|Ic*L8xzpR$RPlCX7`s#A-Vtl`lO!;1*Skw7+7<+606ubic3Mv!l4O
zQk=i%DtL>~GKEnxE>I*3t4>)|=1_Ah9ngT)KSW9lvfDsm7aOSD!r%QJf*y|eL?;OW
zb!+R$p8U<bDd)EeW-g1tAArN6z*y!!ous7Zl`B>l8vD8^;WKZ7V+U4%R5(Nk46U^Z
zq-9b81pvwFW1dz?cZq!!Cx+OqPG(_HZcxjbvZ*Mj{@L#4U9+!ZaIB@&+DNjtn7hgH
z(b_+cGQHWJ`piA&s_n9cIc84F9<05y<Mt_Qs~sn{i#<Ge?C9z14(pE{7kjYTru~%(
zb@krPk5|Qx2VP2b)_u(1TX*k}n{RNNlIsw@?D@U>q6tf`hf%06s)$)m@!%sUwUd$B
za6==iP*F42v*(cF^_tX5#bteB6x8_(w)if}S*r7??K1t|vIPg_H|;r^u;kQc3i##G
zZ(^@_*^4b`ErxpST7x?~J@bs;Kr1@qZ6(k-N`_MB6ZNUhaq$sQvfP;RyN-Go1r^;^
zEeiN0#Xf17uyB!Xe*!mWO8X`Ms2uM>mI(dE;Tn;9bYCoJt)Sd6WwX_1n4C|B@{th*
zmCOB=-*pz`ERD0ErC6n>s%i;`!6Ptrx(laG-aS8s$g|ZfU?9{<>Xa|8xJ(}x&kHkA
zHpBX>>5nG7t6X?c9)^QoQY5B*x!3i!6sdC*Dx;uU8sVJBD9GW{I5NZ?y|@j!`}i(`
zwR=!iTrP#)F;R=Ihrz4&S3cpQfL~higq!u91WWR>9NQjwo^?(>-EAY+dMK5`bYkku
z=Ulinbl0>oEKg)}`}T3HS}8(5VsJb3^z?ME3Spba+o9SnS5;N>jLU_fyIOj>X!Afn
z^w|A4n))-uqWZbfTsvougI>IEEou@UZM@Ljl`8G`GClQ_z45mg3o}8J^Ofpz4{%hy
zG5F=nXW6({JU6demNN9FIX#4aW+#qZ5*HU2Nn^<hA7TZlyYl`9{><MTwh#GO?oG_~
z2b&k}B?VC1hhNWTTTSEa>WZ&P703#KLDZ`+)BlEkmWSX+K3r!!>_8xx)qUg4>bd?i
zRQuy7kB_uvt%+gz;;mYMMnJwg;L_+5^#&rvrae@2uPYaNP+7s~TiJnFPed9(ttztu
zKcZ&WR%hp0H$1Q4?%lhqR)fV~v2n^EKv6S@Kk%Z|v)WoQ?KLAl5^jp%wK4AWa8I@z
z?eDk)JkZ@-z*eQCdhHsnbTIgR0Z_(g-A~jgVF7W^;0KYu1RA*)+}aE``-I?_2Nit-
zu4QzN3>mFk_lig7avyXatj`$ovKTVV8@uTv#mOj$PYr5{S$?z_hD13OAU=aw*Xs5(
z-E-wPJLp|5eez^7V4qkT**Kt24xib!cG{%Bz3e`fW|MheVng<J;I0bwSzZE$1ZhwV
zJPu6x<SSe<<kc$nfVGb_N3UsYY}Ae|waojV3OxzCqr$ZpKH&m{pj-`z1&n|MLgWE(
z8c3n*JHDoi+A|7XlnFB#=wu1HR!hYw@LAfQyUe7%bvbSFw^#p%z3&dkx_$pwiXy2L
zMH!JuRzlgMNOpznh>*SaC?%0(uR=-L%E%_!J3I3>lD+r-o$s5T=k|P_&-e5E@j1S~
z<M<vN{!sUIU*o*a^L5Vay12QQiYT0Zm@Q5aNc&@Q2!a&The1RUfDma%#1!j4q5#4L
zxT6TZVY0{;KrKpU8p9h?XlX{nmq0tx5P>0Jp#}to5Fl}hV%kDT=m$Z_!)W@Wq3RU}
z(m#T9KwdN(BclM2ZTA7R_}$<J8B8CZY--0sH^ABko9!Vm*ZKEI3dDo0-|FFT%b5i1
zQf?px8lzzaf6@Yt;4my8D}soC|7b+)<m`>b#yV{Y`vdcYDg0fNJy^&VZO;>dbYQjn
z2r>de096q{hX#M;MUXaR2%mqu)E%IXY}>y=FOMY}dVyHmgTUBmU`C@m*1xRvzp{|O
ztQ7mdv5@U;{Vy&grfQo(LDM1H3J{7FQP9Zl{~-hj)*z~IyTP6vB8!mxz1MaNg!l+K
zw@rioSnNM3{ul1Tv>p&U04>(rl7x%KNkCil91t<-`E#OZg^rB|W3*1*5`8RD)vdjU
zF8=94rkfDxK@1T+?BBxr7l1|cIF=r@Au>fQXaxKXR6_GN8ZpqKjGpN49uQK1wSpn;
z|C#&HcFp!AWP{MCg2>pvmG^%G2=BfAxg3F%qA}%v0)+obNVdT7-vR;%^Ix~Dlp8H+
z-%V^sU}XroA`4drm-W88aGg#NCj)|c@faS8c*M2FPjHiLdtcBN<`M6-<%o}mF`jnX
z{qz4k1-_}~+GMAsRL`PU>Y{N83oBh#exXd7e|%%1<MBXMP<7a1y+1t>JT~<@g5S6%
zbl~aJFz++mOH;kPb;sVr3pW%hd3L>DA9aCQ*OjK%RT-d;Q<7sj+KPL4&1Pe*qM<>5
zbM5woqOVt^6WPu6n`CNtz|6QvpLIvpwBt6FA?25Fp4cA+&aKME^M(!Y#f*#?_*)X?
z-5=TQJLIQL9m(d)5llmE_?4yY;nR~?M^e~$?B+hMt!3re<@p2#w$RMQE2cgua?4YY
z`E)^7f!FfYac}Z2elWzULn!8k8yV;-ra}okxHvf%;hu4VY?swV<5EkF!no2Gr%t7v
zgsW-S#!}Y;7z#Sm^i(x^10xrCoYuZC&z|@6tW(P}<GbMvj!#gaS`Ii9ifs|j>y~go
z^Q4@7pn`nh{>=gB&79k^rrb-j9U0<}3&1^l+SkW+{d$?J=oK52x{0o*D91&~sZa6p
zjLNw-8sD27rV9N{`$Z_OOhp;&o9_@9RcN33wz@XErZd_5;>C;3PI=WFtKqKI8f(-U
z7PFsI6@KTdL)rWNq<=9frqGo?BOq%r=vtY7d8Py2-m(2zzCm>UFfQw{Ct^qF1wVCW
zTk@7i+!H5S(>cN}czS(P&wQaX%kQJq8D6Wgp0~pG@W{~ySg+ua-Ie*oWm6q3&aa$l
z;`GW`a7-eY#R^V3y5bNxj3(yu+Akc~_-NqIaal&taX)XZezIlaN(;fwI+qgIg&5JV
zEQ&?@M-+5u3|Yjae$ps%+$1~XXDF0FjHP>ooZsn_`JlC6Q0sds^};W`)gkIUUawem
z>c}HlGz>h68As$(Pn{Z;uFIY1@~jGhW9~&^G{gGC4RzYoF(@2`f(AE{v&_ayes|VK
zPz4dlB(8uvbx|#sg_ASGuHgPx&2*tdP1N^-6MFX+X}$4DE2Ozv=%>|2XUHP)<=|%}
zEXw)ut{mz<wf1aBf>W0=V}>#ZiTdJVh9E)hdY_%AqPkk;i-Bo1zolb)npg!XE|j4v
zwH+AA&}dFhdpjM>ZEc0kJMQb%;C9D+qO*ejHZ9J$gx6#Pgow#L-Az1#J%8XNt$rrZ
zU~BzrjzlT4?e~HWh&0o&1S;baE)wcpH8)ejkyKafp3|-(u6w;)*i$%ZHL-Tn&dzRW
z@{H1F&0)XuC&HE8vp{|vmIwMGD>eCqhC4DAo8w8;UMRNIzCl%p$P^GnJ4F+C)x5hN
z@qSFN^^?YAg@~T8>q$*`tAf#{%#B3K$S7%#XEPqFj5mz`?+K#4zC74Wy7=tWsZ$ir
zJfq(HNli=1H*YdI0z39hJ{da&r2+Et3m<h(Y7Vs|j!x#zTCIJcs|jUSE^v}d__BgH
zT#EgX_sb5KBeQyk%<4Nl6MlOkZM67#5iI{z<wPgFV)ooD$Uw^n6D|u~WAr{9@+)L%
zL%7AQf{R+uRm|-1`3ekbJzQB|?1L=NolQX~UKLJKxw@FVxmN0h67eD>Pth1^OBKCd
z!dS;vd`%_*Uc8t1dQrQt*pyXq%J-b7FZ<zao9Q%=qp+2E21DJ~ElH|Ap`xl#2!(83
z)7($7lHMZwG7xm92S=xy(?{cRF3K|$e`Ps!$dAB}T?IG@jV|YHd@qtk6S!4yP|4$C
zy^jQ0h6sns`mf5tl^^-ALm-0GT16~c<&l+@W_27*jTL^PW@cu`IHVMTk=9#tHA@<_
zyj+vi3+TuM*FV?*%8@2Me%wrN)6@)UJCrkXAex&-TjnD<;XV1=XYdB+TkT2F+*G8+
zBEcciV)95MrS?YqJ>S<oVkyA%CuuvL+-1ai^$@%*5Zt5Fe!RTAj^F~1PU(UW9cW|u
zIUu?CMA#8tVW(tZVDRbpPfJS!x3k5(4RT8oEmwwS2F?Ooa|Qs&d=G|ON%;0l=giCw
zRss)=*ep#=06bBP9<D3Dk!;=zUT_&&xKGl$YNTkSxZxl2vlMB-@%y<?+E+U&4R(>X
z_~9OxJ8Tw1JtO=rt`sYlPG1*M6vt9Du9)<Q)?ICC%eAFw{tz$<==GeatmX~(Dm;wv
zBiFX*R}EPZqcV?9Z8bbipAL8S+L9H4=f1X}EbAEfrdY8B;1-enJ4xz#!uN>{WN(ef
zFaTf|y50hzzL8z9k)4Mi%BJs|H)m#~>}P9Q6%v&ek_*x&;aVt;d!Lo4aNH{XM*%4L
zr7#+|DKOc&9Po^AvU!<xb8Yq^kn!C(v6^H0EXo-*TP@u6f|Q*{$y8fQ>%~ddu?}N=
zqpoc7nFG?{cV2T@T5POc1sW!;%S4L@)e=&4C(fp17_QR(RCook0}lO`i{)OgU%&3`
z>_n25L*Mg3SPGqS+B1wHq1n+xhz%Lb{57*$apv>`q}>S#AJUf@_Y`cbq#8U~ktr}u
zRM&HT5<8eB{YpqMZeN?PU*n8z^OC+o!G>KTa05*%u-xOM^J^s3xFwgI)`A;9dh@O-
z^mf$%rV-pt0DY-(=_D-~CN523;_{6%=gtL%awhja=%8@Em@r5UsZS0qzbuR4!ElF|
zFtxnyn9m8I6Kx?5F=!WqOL$YKrczv7JbvKvFi>J7ckB&!f#=ej`^d@@F4{u`WuvYl
zN_3Iof>Br2>?(z`N|x35w-V2ft`kc=g}gadjs=dZEnI2mQ;s$nYmO+ie+^{xDL8i#
z;&%`pCT3ju8D@+J3dw10>8<&o6i(RdEk%pp_by?q&cmZakPkF*WVk|sX@7>hmkh*O
zHkLoCfTr<{7xgxxnv4fLu7g-Z49ea4M3ZrO0V%KD6EqmINZ=G{6=@NWoTMFn@t%OT
z;$S#Um53hjwWNFVUaZ*raMxF}vwy+N&5tYAFF{eoHCNSJ;kUK52?mWUKjX4l9(X+x
zvB$2jxR;B+C-$}|x7C<TSX5FtZ%*s-1&ANHXLcxYREm(|&65?Z1aKUg-ZKDs-h78F
zIt>N&^oNSqEKhT#i6T=qDt{%&o*xAX+8n@$nVAOC&F0bM#Y4l8Q9u3hJF;(*`;mRC
z5@C^OIVDCW;Y-9L2512*2s;wLAL~|ep{xDFHQ9W#fgo1fwYJn*p@k4&w#8##XX9cr
z&&b{cz9J@xNE|Fu%dvVJNRjF^HzaMDQw+!#&nLCHv91+m@cu1uSe&{Z_9Fl#IaCOk
za70$aFdJZ4F6S8pZi=NrSGjZE0y2LwT3@M;TBX{&&^~#~X@%W!<@nPFpmGf^fDy)`
z$np5^f$HpItArK8X#gQ^YbWoD{s=zDi&`1qqs0e<9mXgxUnnWjIeU}-$3odpA7)jq
zGptjw)A~Ql7e^PC_&0M4M*2bcRJMuw5h3c`9INZi)(>Rsr=O_W6+ZTkSL33lUe7>z
zXho$_op6|MCXuW2Ssd?B46~M<>^64GJ)u)qds`O)_z!y#fY+w}`~9cC^2KhUIm3AJ
zq*^p&sHw=@uyLK{gua?D4}a{E^Bk(-RQK+6#XYy2EUGFfJiR(uGGF(akpDFknKm_n
zMO~!R{aXbMIN=^k4znpK{Kw3@e0{}3LPF%2dG+PPft2svDV*1+0Ia`9TEB!i*{XB%
zN9p8Oi@j^h=Ep-pS$~SfLin@oz~Ae{3nzJ&x=5WLGpnxH9!gHJ7}lZ6d3uWNgB?{*
z)-{K{!z=UgnyX7h=U=Y2HYPDDbF>t!R@FtEM`kw{Bfehx9<(!$+&syibc5wv-;%h&
zjEoyNqOvm4MTRJ9E^@xy)!fbZ#;9(p$)Y;GDusVQGcoHyDKRi!y!Z<}J*0}KgaA#1
ztnlIW!zm>o+)uwOIezT4!}mj+@t>hmQ&PTpXY+05GD&I{Ul{@|@X9O22gj}{+G$iB
zA)R81mh5Y%MLqW@QD+ajXL9j@h+QL0;?x8uBs@?20~$3Lc4mGlr9c7Q0|K-^=gE(@
zWWh!<DAd|sja+~uIP(b(IEPOk@fMc|<G8vOqd)U=dY%FoZsiw2cx*^EXeH4Z)U?g7
z)Vi67JSs2}LR=X50C05bCgkt|krc>zvzfsvAW$uEaun)*V#(cv7?)HyBS*KGD@Gk{
zdS>zMFg=H1RGI#|iB|Q=Q?vSDeqPbaI5MW3ds@a`&e&c(!$_NlN%oEf&st2p^QIF_
z=>)M+;7;mP)I}_g5iD1nHzzVFNiKMK`t?Nnd>$}X?v$wN{{8X)gg$0ndG^%pqPHbO
zF0uX!f=7@9%Z;EUC`9dOOkP|DBPw%R`E^nQ#HwGV7^3o3*UJP~7bkuHn1>?lQd}s%
zb8RZ;&sOy^$5(gN)I!5L;vSv=I{^VUwL`{v&740aGi_&a{s}m!*e_1FFQnSfk0fh4
zWp(A?3tQm}yB(VG5MbnWuvt_b&s|VzT~gJZ8#+p*9TpfU|4f?F>m=^_OqF089hqLF
zsOm^YI2d`6aaYYB@YxFk4|8@0g>g<sxnH<qJL3-2x$G8|G+D4I|5g!9^e{Zcs0#G@
zdlQdFq1@0L!SUq*r_DUrq3>Dd=aUaKH3TpWhBw!CW~KI9zXjj-iF5W!(KGAOtB1MF
zf06uUBH(uN*bA!_II-3$$IgOfEhl~tl5rqnQ{RSA1?uHtCSa}oqs<~9Uo%RJ=mNj7
z-?RkD+9U;shpRn9vG4P01h*}`S!xOeGcxSpcxo^MbudbsrZT3}r~Wb_fCxT&PI>vj
zl6fWJXNQwkHo&wiFdN84h~pcT1h(OY%kd{P@7><xR^e8W0(|^;7T!<1<u+9%J4SMa
z(^pIc%H3wcjAX`Z5)rG22C(6qpX$w9Ql-<y7Jm4}a<r3+=a&!lX6qHYCADou^BGJT
zY&pm`4i77TzWwo~Y2Noo*g4eQYiv4Blps)EgE6D!K8|=b;7R_t<ANBQ2K)ZrkIRah
zc|}GJ8I&X5o>CEf=d?0DSx+8L#((ecu5Jq!eX5Q(j1LMjj<0;Y<6Md?M}tfZ>SRlD
z+M@&dtw|@Z&a8j^7*XqulDoiAm2KsiKEhpKJ3D6gAp{{wLR6Bv(^XyY+7R6ako@le
z`Iy|g?l1Rbu<MG4-=S@9#srn+=+^zWD#1^C+B637k<qDOkf68~$`dCtwh6(S0E7@j
zUbuC)-smfdK<w@)t`T^;IR@+yv>HGn99v3?W`_?#1GmVIzFTjaXdd;tqH3|eA_poU
zH&;)M<OV$%XA(hdTe`kNQ!lIaBvZl)k`cpPdU1~V$M@GgiRDlWU>nHa=|r3l5z8ja
ziB2)FDbNgX@(+XK#!fTw-BbKQ(bZU9lOC5Vs+ZSV2?)+`{TSU#uU+ALor>KaF^PB{
zX!QQl@qPtTy6U_HNi_RKqwYL=VGf~`ZQ>ULkD3*CI2F0EgE)Nu@iN=Pv+l<e#CHzE
z1HXCOgb>*23m%gUW}$7D9*08JlHqsfrA{AiW6WK*az^$WVkV*tKZ4Y2V>^t$F8vC7
ztV8G<(0sXFeRVKm2G}53WqAPHW6SaObD1hiO6{r6%~`Vy*f?v}4ezG~`=M~H6m+!~
zHI2)Pmj>YejZa`eU(X$Mb@8Z3o<-b6Hz3eJWtIpZ+MVDy0gSoLmXC^ectA8z3Micx
z(JuJx#79rCxu^i(3#3c-j(mV>uX;SI_OuF1`)YRmKz{pI;N{%e#D{p-+Emf-Q3@y0
z4a6UWeD+vZPE<;Y1c%VtHp#BpfkkaS$w}Qkn`2FaunINLNVeU)5H$tbOGMLc+l$72
z9)W4kiZ$gs=IHXL>(s7MN|7XjQJA7U)IKi<bbHCak$LRaC)~qhNgYK@sw-`&xO&Ya
z196)lfo2fJ8)-|fiYZc<a!J)0RLwF|;}GKcJ)ac_|G9HIpz`w&?@v~t|MY*J2Rv&I
z^Wb0BGFnRt17B@CyAY>zbZ>0bClm3shvVPhI=<oex_l#8$Gh>1Vc19KyAWwi)AI{0
z9s&neAK{jxno{efpTK~jk$`AhR#xKK?kRA<VS{i)RlUs7AqBLo{x<GuoG^WNnkw)}
zggKnV121}@qJ{Zlv$U&%axcH)awDW5O;l1y-iWIu<koL_0W*lbUDhLQHNKKB8P+1H
zxeg_$_SO^r<s$A>^JMX0fWEo`M7f^Z;VStO(BjI-!`Ilvi0}j-`ZM}NNxwt+D?TQs
z*UFHIpjs#xX-^kV8-1XqWur%FdFw1SHE(-q{<}V_Hx#yDipV*8m6rPyg%yv3pKHCj
zp$ZGXzFr!KEW8?YQDfs~GNa@aKLjCx6<ni|O0RBvxxkjcU31?{n2v()KxKeApoHb9
zeCcEZubl;c)*KKckLwlQ<0LAjH%`;n^NfmLj;?(yO|A*!tQ2uSxV_Gk;NZ@UcX0PP
z@`0%j9=AzTvo7v-qy)T@?Z||{En3XTO$zwLmw$>{2^Nklsb)j^SBS}Rb3-!u@cVDN
zG4_k0oD1L*IW0}8Ig&d--C}yg&-c;M13%MVEDey?tp)@%&#f){r5qiR-d=WT28#HO
zct8R;0~bI`yv!yY-;PIMyH$`fZFxsGDtGv_fnc~FTK0j;(QK{ptNr!%d+YSeUK-YW
z%z=`ccbFO}NQfFykdmqeCbJ%2*?>^w#!~-HZSCd8s0=_YwS1K{dut8we4x;jck)^c
zn0*j#L1GcdV*$Jmnrv>=$S113wDDtBxuOns&uL5cU%!3KLw|&jUit3`*<1t%h-5|9
zk*v>p_z^bl6P4_=xljdM-}sc2TyQwp<%gaeyg#zJRPVI>1AO7b<ej$_73ITlvuvj4
zmZtAFG&DTqKT3La#u16$TnE=#U@sCkcDG)tVjgVzWpe{Pk-V?}nIG}+GoGxKcW2L!
zE~(lLeJ`X6%D1ljcw5ldaaK~$YoHCDj}E@03jyNz&5;+B{v%T|P-;TK_CRa=yb#2!
z{UBz4`z<hKaR-DS;tGmDfNr7(FIkJJf_P9Niy|+$g9ud4$FuF-76uN9FJLEQ(d&17
zk)+~JDcIcTfZ@S5q6%`FifoF<7Mp8T&wzY^LLLVj8=G_lAF-c!?wXlYvnV+WlqMS|
zBe@4C=x%@Es`SNSxVI<=2_iH=97{k2Lhp-XZM=QsT_C`s-V=ypAC(kLVc7=ZvXiv&
zqFaVvq?4t{y=`c4s{p4Ei&A@k*X?revGI-!5Rm>d?}J$uHXm_{J0MZebT<qh-}Y!e
z7s_F{GNysDpX>_0<7he^!lcU8`rzzwCz32A%ovRLr3ye8@t@c303Aq|WrIMK0|o5B
zf<X)jh>7p}EhfPRU|3+?YAN;77k1K`gpjRMP&kjRP<b>1Llwy7MTkTter752B0Cc=
z4kaqQ_CFw=HnWIY)SNE||H{nVeADuDQ?fb*!Uw}({Zcv^b?3eYV4d!sXiF`F5HhIO
zo)`vlQB@>#=~jW^n;Z?^))5*hU~6FAP=p2IfvgyIz&*!|F9@^Jf^dAJb92*53FVw2
zqPDp*YL$>Q1M1v3pB}NCpp*k9RkEOb#d1W~knh=CLv$mu4Q9wTtTioZWTIUL1ei^^
zMpD2xVv&$XhoLV*5On0l*KJWH@Plm59)U`wyp{M@6m3us<%x=RC1+{3RJV(LicF%E
zI}#!+CJ-IPnNu}rOPN+L69nu<0^k*hF=?7&jH(@(9N$=91wRpXd+E$S*zLr-pD%;K
zXf^Hu%s2K!wmzu?LOIL?G$ov<<9C=Vz|JmMmv8!%?*j_PVss&5%2WpGb*g@kIz3ge
z*>$kHv>DX_W#Im34K&UD4;G5IfDJed?nCipLzD=Cw^bKs_t0?kqhfB;K9dIz9$e#v
zLlkpx{O_l(?&?C~=1SP*ZP1}XELWTIx(J1JFJ7CeoUPqW`X{>^#`(-q2nALq@OIHN
zLNBW>C|v9i$3AE^`)dH|*W+9J7si#zhO7Z$TwB#fN%;<EoAml4TwRQCwK;@m*e+d)
zEBIU&&dY=b;6A%1Z7`0(g(64C@Hrl@G@p93jdWzR%3#g8_@_f)7FG)hx#^rac*~M(
zJR8=$*dTBg@#`RzlC1Uuwp{-^sue78LJC|EKVj$zH116@_r&wk7dBQ=8EQS&>3Ib2
zXE4QyAn0#FW%TPK3UD4FaVJq3&c<vU8iaG~DhVmSlf?W@6$lv#Z?+t!=qBSktW@9>
z@W=9$mA^Diaccpzf+f18Qf)9Hqs0W_HZE$GZIxQ21*#tFRf^Q>E2mSGgIVuCTw_{a
z8Rd;qwT46$SQISF%jWSxy)y&iYA(j(NE)zzFrIG}-nF4dr1d!>t!@*jf;NaPHlrH+
z&P%^SbI`#n7#x()i)2Or6{u7*r}WE>de;}9!ES;YeQDJNzMrigC2LEvx?CPk!IEY@
z{wzWMygcfae;AK#?jP$EZu}@6yWvz@T@4vl8uT{CIfG@F&q}x*Bu1yru!+#Ii{<^{
zG#wpi7wiZ&Rvf-k!K*HzWHn=8TnKf;s1U=p)1%yDHOV9)F78S96`X^D8>oIzU!zl}
z9DYZ31rn~Uj<Z4`7!t*aYED*<xJHg4p#P4x5MWVv{JuUG4{Mvwv5=RSPaDm*;hYi1
z<8oS)k(Y0DA5s3)(@!$i0T1L73fsMdbY&3G%kM5kCp6!IG6fahjH-q-6$4`)E#g|r
zNATqdd1u$|RMH5EkxT}Ojjw?*SQBm)fXgwmAl2YvNl8m@f)8P%k_OgU`E`KiT&5!m
zAPr__N+NgEoi;j2Cb=O?u5rUYN3YQi+1xB75V0Yja962+wXIvE{d;-dv`F16yDW#S
z7@r+<2BnCkuM1+x$n41w#YC)$CE(TVXf06Q%`BDC{LJA(3*=Vk-wCL6<=SQ;MT4b+
zvufqWR#5Cpbu>e#&g#zr$-<yuPtKW*(>UC5vRe`iHQnlLOMAwjhMOJ6pZWg*UqKNh
zejDR!UV9#L5?(;{GzN%ImU`};HlOUSxhG97YmHA*>nvM<GA>{|@H|sINa_zXoWJa_
z^qmw{jZf@=<jTA{<CCqVrv8DvGPLU)<y&vhu@>cFBIPkiV{n4Vt=e&SKMxOO$l)KA
zwE&OW=25`_<*j$ok`TrNYZOT`qLF`oy5F0>X-TTKvNGE-=Nh7Nv!E39S?)Mk#+PR>
zEwMi~*%D)u-5aQg-NuOV@Z?MS-%(K<SZb8c$)h_$2=7n6+1yx6d^o5Qk^2KRk5t=6
z*AzeR*8<b;eJh|a*Xn3;NByrKf-dDEde~W6SyaZWo+Q$oe>WzQdPaxFLCr;(^?r+e
z)%6VucZoV9;jmSg=U&!D(CIh?cIV%9@@?{LKHV%ubUvdY1lpAlD6|78i~$#wFK*fp
z$`Bf%WCD<10O%sry)al8Vek{v(h~sPiX=6l$+qOWhzvly9?}=}PF9N}^CCP<E09ao
z&sBj+h}^SB%ppPw1rj8(MsV1~;1FunWzTxlfb!C531n~yWbjd*<<=K3=15T$+d-Or
zmo8lrIxOnpA@#RZ#$54zBqOMs^Ynvc7!rG0^XSUXlkXoOzPR+GYR~8SI{twuQE&H{
zbmsiKO*it_p+vPQn8OvffB{LB-hk0<)S#Mls3XJL%eWtEARr9{HXSwuH6(=!;21**
zrW=4@G-Q#t5~SNYNJJZ|Ql776eg<n$G)H!^+-F_wu>1H$U2y%6KK-I4W>%X8sUx0g
z6kJ!sj=GK{ND%1qXo7tnl!HnLr3#4%zU#nU2Y75`NS&JCnO)g`Lvkd`h_>rU1N!<>
zLGytxRlHyZp*#ItnbZe08yl<PLD-rDP;L$JxDVgX$Ke|Of`B5}N+cKQeOjRsL%>U~
zF=sIhnZgKq3NwQd3J{__I-u|*I>=uIad!7Y#lsT{%QH|s{GXOz`aw143#6|Ae@hmW
zK-VTu3qe|UQAprWO$Zy?yGmlF*AYwFDp_Vp5U=})ztsDT@p<mtGYY9>RzpB`f5gZ~
zJkO@?a-T`j_#mh`L3myo0^h7_F{6VK{qxG<22k8_$gdfqRr;A))L?017`k92jNpTz
z+Z5{u`P7w#@iQIz05$Q`Lb!wz(?S9XKL_hP={_&d4$^3!gZyDtMTKp)8x$eajy)K0
z-k`68<9p=nP;+9lw)6;5Ww#K34TQP9kiV#>rw2ZXAJ8Nm43^d@u>SH+Af+b%VK(v*
zq<VFxrzR}uzD=Tn+VGwAW1vZ;rUs9F!#99n*3okukdUTH(&^6i4o}o7@rZt&H3aqU
z`YYd?U(60lR{{pb%7dnP7zIW$sHpl2+MSv8Aa!5akk}Oe2&8sD{tW_c6{2kWKwgp%
z#sC)a{-S(0-$5*a$pT7*NSPrhcqw^raWNE2i8QssR!C3yLhZw*0V{J}Q(x^H-+d}`
zu7n@%c!z?{LwjLJOr%L9jgZ)RMedFTk_NT;1!>k^QB`m@b+$___8>{m?YzWRP0qIQ
zO`T@ay9P1}Dz{|h0+a_NY%D1%N=~8*hR?aIFwosSBdZQ&6LISw4NL2|Cv(>g#K4ev
zzBs_7vcwK9T3_rCWc8u=0ntY6hy##0g<4>{_thDKLyzLV7^6UH(!sUZSzxh^$vY|X
zy(@<k-g$~LJh^o&W8e7Qd0aw{3*pW&1wbj0<E@c`*c80k5qSC?pYM01%0)*<zu_e-
zL-ikZsN%J^N^S|zT3=B1K^P?kF|&A1tJx>&mzUciaTLH!IV3^~+moi3_j`M`(!qa6
znf}Yp8?S*f(tIJxn`H$S!xr<RM&P$zFENznYicCqHxPp)xRQ$F%AAL-DL9`&0qdq<
zQbU;+QdecBh4SVnhA+fM-fIS95CVQIz*g}J%uREIRp7zJCowt|WG1Ze_7}1dCEVr^
zQpHzar;UN#yf(W$xJ)%WV$`)U3x{bS#w61WB{fCf*>I59XlL;Z4(-3fUg_|{xGy-#
zq}KRgtT8b&Cl2=41+%gjQt04#6^HY7uw811H!e|kOgERvg;Y1xmcf>XHT8``1zV!2
z&+_tY+(tUYk&F9dm*!g|8XKu2JnYoO|4<6DrqgXB5f{|v^=E$(AcTDaxfiz#MzDPQ
zv$2rubdlCP<S>y`0{r!-|Bo)hihMMI;~FtPZ~GIHa&i<Ukg4USx1KBV{0Xrc2)T_M
zxxlK<zn+#;Y8mLS61UPC%2|hG(7#BR)JO3T{QBZ?w+;$QDnw$=&xd_NT$JAFe%$qy
z`GajXpd(kH3J=Up?4ywDJJJ0`BJ^@M8Ys6j3J@ccut1EyoxCT%eJY0Ikm$>?<c=a7
z<;Q;<iuq1%SNoO^4E|0CW75XX&5tJYiz<(UPoq3<Y&G1l+!P!2+3rFhv-+)d5?zOt
zVTZN@tZ$u&ZUaV~J$^g^4`l<DT$*4DL}o*!z_X8K9mb2C+tK9h;*@Q>dxva7!OOuo
z)Zyj1c!(cEs5qhMK9ka%Sh4Va)egWTo(>rb&@P}K>S;>yg`BR!f;-M`m?<eKLEH+$
zNlk;cF)^9!51Xb&7$7H=W$@&#Y{9{J@SE#8!KQ>G+f+F(flr|D5gZ@vqlnZUzi-oV
zEO&FVlb#hl!L7IfR7Bl-hc;NZswK7}2fEcQ$=-HsPDVx(A#4Ratx=FLOke3wP+Jce
zJT*f7BU@oJrsmF4wj=k$BB%<@3uIJ}n^q?vy=<~L@vX8lWzg;z_*_s%Dmy>;7G}e3
zW#H#^M0kYL2ZXmLR->pJA7v>Ka!D32&3y_irL#_s>S@Uke((<p+A5LS;xY&VvectX
zB)002wkkp|Kd`lJd}nwPAO%8v`5c6|6csmyKhDOZLUMHL_tvpAmBD#2IH=x0bFSc~
zc=FQ(hte0ZRGVKQ5rmAL{T$3XI8=2s(UOrq3*js{WYxMvSr}0^={KoG?G>^jsRdRF
z-NnwC<eZo4&g;nTLCpf{ye1sckBS*u0=b80<8H71iUz)h177B%EtY8WA@aB5{OCEl
zv}>!%Y%3fUX$pmcRaIf+DO6*PA6xSsx!=v}gCPWUqtcLZKD#0J;K8q!M0I<+w|ro|
z<U+Ps^49$5x2LzMJo%vnR*6pvu8zoolXxx4j-oFeoFVT3ce9K_#t3SVUO6|wQXsic
z2oc=_(~(X@kkWM%fs!w>&6Ri=C@4dGd@mWn!Z>gKaCXa)_mTcMhlY?l2O!O*rKQ!B
z>Upp-5z=>pa)(LhYJ*v+p#pfWUX%jZ2mGZ)q&N+8c?Tr6tHic;C3nhcNkja*&(pC}
zk5vZ}6xjgN@|`!6-n}D&WJGtG-a5!OJWZ0cqHY8UiT*M|Tlu%)m6aO?zTeb)Ywmr&
z$<xZan!Xu&#kA9lte`@~onk(02~gm&5)ti-<<_AepESIXB%V-L_WF6BE=i{i6$Fu*
zz(7G4(*URZB~($`Gt!I?Mn=rwFBWR*`;ZKWT9-M8Z}`qlND!lU6iUL@PO34nv6+LO
z7tA<IMzyEOxDP3wJ8*_8ZBQN63??=qx&K!90LS7l(Gkty1gAt|(pk(1HB`g$<z{FU
zg>p@v67eQC9dF++O<g<{!@$Ot3YiHg3+hPM4}w~OHYi8QKuz%Le&>eNE%{ZmrA*VP
z^OB_eurpvsf*CCMNg)E#-I1bz1cieUgUpI4LsiTq%<B1C1C`7Wn0*LVFeE)~E-Xb9
zQs8WU>|u78E_nwm04E+d9`A>ml<5R0AA;j_(qVAv<j-jcfu3T2Xk(HHwZ`8IoC}}^
z%w{}23D#Oyp?B)y7|8PG3gc^j=Sz1rc=@rW`8S8Np@<3WjUcaAH%2(8L4I`&E?h9$
zTr4ow*3p6Ek+eTX_Hlbb%zNMJ>mEYa&F(DokM~H98;C>A&Rf%w?mT7i@?RmPa5%g-
zf|%aqSlW+6jA*hoxelBXTtcTsh<WVmy7@g-BAlleEFd0z_g{4pY}ONFJq2}8g}F3K
zf16e+cxrQj9bB5l%Ni?l!zXe#q1r4Rskd{*4JxmE7#hJ}hZHORR!)d2*cl6s{Cw4N
zWC%$;vFk4+b3d7zn*-{@6FDiPt;}}aE~MbN<wf3`hik$wU24#8b4=53g;%*`!MfgD
zU(^IEDC08}3-3W3KeXB@=RzNonU4p>UM|vh0OukH6$BY>ynrGrU~VY#P|J6yNKmwe
z*xWdXC<N#1zdd0!AFMsRG4+h%OJ|;a5z+o82;EsjeT_lKLJfrSY5_9=<Hh_+ksSku
zMRL!#&K-FD?p;5UW~Po&hb<5QIATGgc*<eqh>O5FQt5?whkHCF-65{C`}4m`fu2ph
z$!^!wxH=w}K}kxgJ;(MOz43}WP4)PRJz3Ov$MHh1kLxnX_$#Uke(m#mt8~YT{`k}8
zhuT5=$%>ECwaeU7{9^f8`C`tor-3GlS&4K<2960GOA^Zt89rnl`gPUq)z`4<!Ro<?
z+^d7OwGPhC&V&EE%iA4omjb&K*rmWO1$HU$zeIuhVQJ?To55N}sM$o$)zwF-6BCou
zU#Ltje(me~&F{STuV3UmaO7w>?F=`^WP5sgpSSb+{9tslY0uY*+}ziFeUHUeRae>}
z;|^B-CS%uXv%==!-93NK-{qE=*mKr_q|D48EAxJ&oJLJ>8SWQDW#`$jH%8gRr5#42
z-387Ox1g|Qv|#`VUZy>4p2;2!r9IF2!aG^n>BeMthTIyvn$z<6LLG5boP`-&B6m{(
zWpjTpN!6H1wf&=X-5zWRj&_U&DjThoa&oCTJ1Y6ndJz$gJ{59K=cK;>=>GmcCxo@(
z;M09AKqaU#Eb}!~N4f6&P-WDP=&g)|M(@!wJd7{-+YLt#ih9{a+Ak_Y$Es2qR2&@s
zfq_a~gGo90v)_fIBAiy9zj(3#gtM3WhFL`or?p=DT)Rxn%;a}iun+a5)d!631WsMQ
zz6{|s=f3A<{r#hAx0E%0e4<iRB$JA$P4sQ0dz0$8`jJ6_DT1}X3@JIieV(ZgA5L9@
z52YwrsPXIoMqf|h;=N=)-(w7SLr=0q8jFc(%;!~$QmpIb*&QV5ASWTo(W*Erii7cu
zJDqFG$#S1KS)pKS#3`h==$V?qYOZbS8Dfk^CSo3*_}SJ4>1Xqve+&;*>0&mY%yYIc
zY=)zZ2`$p97)`O#n~j}wK0rfwSzci=%wpI~cscLj!D;&i19J(Ch5`c?<96o7-Pcr9
z#(b#N@1VF(^JJdq;JD*7nWiVrO^MMovLTR3RmDeIN=xfvOun^EznAR*CmgW6$iJzN
zyAPw`9VJ;=C%9vfdV#i~Q53oaGNrDSMy!vW_=stvt9-DXeIPm}NkdaSI52RhY=PJQ
zOMrT7oLm)~&P#4mj2?R~4GauihMN;LEj1ya-TZm|DT&h^r`E^&jjxbkw2Qn6&J>lV
z@mXHSYXOst^z;U;Ns=Mg6P4AS7YxFMF@yjs4v8CT-WTq*CjzqA+Fq5*v0h(sW-dSu
zMOdE2=v3eYDW8LGv9#1h#gv$bK`+VI^DZ&mx~=@0@6G5HjCL*;(yR<004eQO+%I$n
z2L<^B-%YOeP*4jQxQN*-PVUyNAD{29PUSs+@?@#A(^W3rbYrcxvf8-ey(zGg4(2rP
zKve62=-SU>gg=_I;h0w7sf};={AQS<AYgd~(#W@NX)bkNgPMjotyVaPRwa?7&Oba<
z&-`XD#^^GVrCzIouoyU3xjaMSMX`A!<n!$h68`*{jU1si$g#M)xG7dz?N=ax&WHCj
zxH>K1%z85O`R!<$m~A;qn$HObl$BIeOs+Cqx<u?BvJBF9ju4~AI~h;$Q7@J%o~P@x
zG3#81Qp?AULT`h|JNN^5Fq+A9;o*C$m`mPV>Mkz#u~x9#_+l909}*bOCp3o9>}vq4
zUc(a5kIS;}@aZ3_S-ntnK*u&t=zN#i9HU)NWqEC_$8(FkxrW(*V26>W&mvw)2Wy90
zlQX_z5ZQ$x))=J*jF705evyeOv+1)Z$)<kDRk*SdE?D~W$K+hz`ocH&h}U=XR`SHp
zD1UfapEa9t^Cn(2!Jmz=RKy@!uHTs{3u}OqHzvIYaJg>CLF2snx_6i&pOg#!z1yT@
zaY@MpYE!WUSL*fB-6=R!)Rd|-`|}WHkG9;_M$eyHz%!6dcv1`(d3=3+PdiwFmn^iS
z!@v&XNfOwK(DuI0ETR*q(oK593UP1W@|~SAquJ3Rqj(V0OV|zd^@n=bMJy+s^rvEV
z!k6FLU{WQ1s5+!yGE9vupOaOq{LRrphnWh|LeakdQSvLFlyfmA__cYynvMSuD(t{h
zO*j-w>QtU-%71|GlGPYpAtv)FbS0@O3%{MJc^(Y-b^3vj(L`Z!v0Y*Kbt|jDdzfSE
zve-<WfgO<yz07Ah@}q=2g4<et>KEXm$8n5--H$QI=MoVOSC-#30abJpkK00*BzLxm
zu=V-QW;%7h*a4{=jNie5IuZw~`H{uP8L;kn9?Vf=E*Z8r9_7FKS+8-jivJ2My3jt5
z(Ox+iV@QEJ5Wo^OFrafVhO^1C+ToS8wRqj%zZj0bd*`5mx!U>%pg>>^^CRa@^X2j{
zl@}3uAMBnQi8n9CR2J9XR#$7Y`Xq(0sq;8HOV#EBD9mm?{P9UpP3`4L%<0kHx#J%Z
z7zol<pQ7CkwX+*$o91=AWNWSc-kW)cFq_7}QMNRMQy~*K{N}tgIovley-XqeHuw#9
z#(^Uk>v30FU34ZUA%R`4&sj8#w%%A9+&sQE-GL&f;l*f2S=nXTYJJwpOtZgq{tWjQ
z!i^NWdHZ*mc>b<)q<QlCm?WeSxt4UE9UceKxQ>g_$F(f5PT)ecQk6>B4TCI<8Dp_S
z9RYumjov#!*A=KzJBA7WGSrZ6K<pK~1=><?=WCi6eF_j<(k%Yu;ZF;20v3@cKE`J;
zf|uBW<lb|#0yCvxZq36$IEQ%|7(&xnvpMd>&d*-Q-i|n=9;%DVH#9YU6p7KF3#rcH
z#Fz}7im~QIr{b+0J;s2=qn2=4w!o=jy8qcVOnulS1NR&j+%@j|T(%*_htZ!vjb&f4
z0$5`tW0D4_jaUba&s<2y3vnlz)k9BS){WG{{2&yRfO^oRWP(bT3S46G$B%%df{Mzk
z(Dt);Lv=yz4SV}y()Zz`xw)f--NjYRo7ZUn_zo790OKpaRA#>XiRFfC>nmbeS-2&b
z>uTBy*-?BagT8vvvI@*a#q0;e$gpKB`<^jeSSF6C;RG0`Jk{64KV3F0JA<vYqqoBu
zoo2T`ar%U0N$i}&R1Rug^wVWH3=-ehxA!x~1hLY;@-Du|wI_l)IRAqfAIRX3b!1R8
zF)cXgou#Lr{_z=GbLWE3N4!41_M@2Eo97}C#`hwL%j-6|&YipZBMhE|@Vkrgk&9B#
zeCr^NI1x7PLeS2!{QZwxjWkjH2(#H-xBd%>^jmT%E=r~uIXcS9=KrO?T5e;)1=?TL
zA@n=vu+-z~j2}n~w;xA@*~{!8pQHj92#LzA_lz|`YX7n*?W*p>fwW+_fF(zS$?P(X
zZ+L9sRQwm%WXDCX<2ydXmk6HH{dD;#UWq~53~FbW+3fn=-piw-6FJX3F|BfUAy7NZ
z{S=!YA9F%P-I?8}Ge*MrEZFBeP)XMu^eS9NTwGktGxgPR|Ndu1sI@Ru!bVk1;poQd
z<OymF0v~w_68XkhpwpxX<*?M7jo)oG^t~I-%0Cs-<|1O9p0<Qc1(OKIHz@IVmwMK<
z<BqJaE-nvCZ<>~PShpnAz`gY^+EO<)px)lp^fN0-HoO4uu>%6>ELJRx%*={{E7bt5
z_-$umc1)Wt&1_)iW)Ll?Gov3<k9K@@=(641;su3JcP|1))$F@27}F83fb(%SH*cB^
z*byZc@G0#0tVt>T;q`%=M9le1ujeqn<f6;*pn8!xofWL4Y%9YLLpwf`S5f(%Hk=-G
zC-fxSuT>DhotRkwxl%Y0mZ+FI`mGhJw)3ojDZnqG!PNat7j<e+I}g6epp~$(SeknK
zhTECpYL-S-pkAKc{L+N=BwTFX7bnL6#s5_2Fh<|~bC&{tp+L-&O<d7rDN#IaBQ0|a
zGi?n+<S%2*hj^FRIWJSPP$GZw@Zd4uu{F`AWWFn{`B3}51s?NlOHB*(|A-lC=xF0H
zE9hxi=u%$3#KMBdd_!ALN7sV#68lwlJZ2$13v(H5Ght&x6JsN7BMZtac+A4a2F7N$
zO*HO9kJq%V^zLhmnrYbLF-vG0=|C^6EF3IXDUlyMW)VFD3vDw>W)TAo3+-##_l>o*
z@%Z@gwx?>-K32E~Q3EcoHsePRV0uY^S?BIQcPX$-fn5shQec+?yA;@^z%B)LDX>d{
zT?*_{V3z{B6!^bE0hcycY;;VRa<f%LMsD|Kmjb&K*rmWO1$HU0OMzVq>{4Kt0=pF0
zrNAx)b}6t+f&c3i&?7vIF6r90?mUje{eL6CvAc?03hYu~mjb&K*rmWO1$HU$ze$0q
zePr(==>E1$loO*&^mhp-C&yL==RYXn99L8~@8CG(y!i2Zlp{$h&Gnl<6T4K~qYbl@
zI&MwUkI!p`n5GR&pRE{9=~`&R^>}}l{!3V~VdMOY=53k-E<J>l2k#jVYfQEcO&iSh
z(9iNd*Al=sw(47cRHo>{(%)^+d&b+5p4M20nVjd`U90lFj~zrFi3-pCT;5<#b&aa<
zH{)427%E}wQkG2UyfQwpL8H~*)?1hS!Fy4iV(s3I@Ua3O2ZyG^ZMB2A(oQ@M^Aet9
z8u2{>+>bvId+_!xpWg3l^ZLNG;JF8<hFH1f{9@}|r_aVwH{QH38*u_ZAnxh1yMc~x
zhp~9nh%V`XtexVyX7ag%&bbmy{?Tz~7BBEIdNidqS3Gf_e(s@Pbs^(Y938iKfZ8EC
zeL0*C&6oSuI7j5!WkjPNd}86a8{<h6=XaJf;`zEU&8VACu8#GSsU*F%Bc_FU^wI8a
zZ`*>a7;c^<5EHIAioe(R=DoPLD_t_OefImP)sJYFXTIt_(M&dWU-$>=SVGw0gSN5{
zWJi|jnfSjxJrqLi;F>icV)|X9Okg>1$l=Y(A+d`sF{nn5y;rV@F6T+6WctY^at743
zgvDiC_ShHAd)qWcgx>82!(sMvtPZg@+{^ux$IWqdz0@n;7bfo!JU#HiIH}tC(0$Lt
znvtWH<L=$(=VMQKN<3Te8oo6Ux9|Occ%yAg<R^DIxvS;Z(@(|u8U%=>r+DaITo9;{
zbr&=g(O@i&aU>2ndd2SoV`?e=#aE2)b5z);$etPM_Ghv%Y4|s$dORXinHHs0!b+hV
z?4Bx6u?us18Qx{uQfOeo_mE_g`p^-|{A+s-m9^biT-}3f*DHI<E_y~vq@Y_)m~21m
z-qvFGy?2f**1T7%^C1}UXOYJ0-is<V6A`}l+-tz{)Ijq=n-~#sm&E4P-oPVmy_5$v
zc}#-WJ4jcHDs65*+EdeRNjD=H`@ktyLRYuMw#<vPZRE_?!kD@8wU8?_N!3A2GA=5I
zO4<1zpUlc5XI!*-fqU%5D9Xf&I*jnDuJy*mFa0u3Wy>?F>Q<MN=j;kHn^E<9^FIst
zwhDhBG_E!{@bKo5^9_QR7`x-m7@p|4JTVxi7}u(vtYOMCv9}U$^q?wG4z{S=r+IDe
z?TG+wi>q38YklK2ZKNd}LUYHBu2D?Y+<KFj<1#8OJHp&wY2PA6%JIdpK0l1X@!R}_
zOZkz*2FJP?&c<tnT&H+Rv!2(^nq-l&M6Oxwe=l>$w`bO_cz{@1H%~h`m3}-f?Hk?c
znAN&Bf}96FzaeYo7&3i+Kt*E0Bs>|PKdf%n+fHf1?CzY(xS9&lUX-CsjpOmk)dRB;
zd*1&dNUhn(cgm?G=j#>Jms$&#x+izP+c_tE>0H-|XJ2bKjPW>Qengso$>wU|yp-9=
zlBMB5$l8!^+toklaALsn*7=9_<uqS~+^p@e6M5g}hgMB>i#FI<S}hEx(UQHp)Oovc
zrM;$Qd8kNiP}MGMh|!f6&bAd5WD>Madb4wPhf@>U*ZCXUZ^(^h@#&Vg8@5uHaaY)M
zlzHW>PL`+f;s_THxVf4HN%YOz;XQb>mRhU*t#sK=+~TZtf!K=2bez)lSiSJ$(MC_0
zT*HnpTae31?j?2Wr<42@9j`I;^7e^MpPyB{vwCKk+4Z#4=d}#ZcU)viyOZSC9w6@P
z8IOlocr0r!Q$u7p^fkvh*Vh~t2^Y9U$|xOVwLWmGYn#i3I#oq|4;hH{US7(2m`Rt&
zcS7M<od;WvogcyEi4PV+r@sje;qk7wwtYI(>sujomfS7xd0Y15ssK8A8LanRr?{H)
zuj4G8Zg`qo`@$~gkdQD9&TWaBn+iprT<~c=Xv>8W2DPXU&77Jy{Dvxfn;5R=`@xM}
zviT{K?huu#Fn-w-o4jdW^EJNvHx?|fqLT9KXr2t_%Ox6pxby6(Y{P`+?YEz0hz_yd
ztT~-m$#b(&9NP)M-$bRFK~7D#ndC#!KojQ#7TvY1G1daPalOn2`TKj5?bofw%#3<j
z=!moFsqP(4XnbQ9AKLps`<nZTLp?hYk(lfaMem2d4t2y+6=uXzeU40gGR*g^=vPrY
zKkFM)!~C!tm-2qSe?)TY+9#G<9k*_qUZP{yK6<(L{PQrZa@D)C0hOatu`MOUx}k@Z
zXM*A@*aLd3(jF^baIKJh@-|cPbMs;AudN2OY$KLxOiHe`RnZz(r9GDj+5CT+W-Dbr
zqEUD0<_ZmwCCZs6eQj!Z{d0Nb`B)E?pgeZL6RNsixXt>+XK4+*lFJ*bEU&Z<=!oH*
zP;n8<aH&$f{q=;($vDfBTT`vC?9aT*@`#d)Zjmomee<~l+aZR(zmZNoPB%PGa$s)Z
zfN-AQk9CS_k<C4|+{vncr#i3vt?ImV34QqCAE?em1w~MuN1R%ddz6Pl4($1gKO{jh
zRxH*<D@Df8SWDyRBBrESBuq=sU7I|)7k}y%<MO%wQ$%~M9`WYq#bhkREE<`9QOdQe
zikYMFVfD^y_97`(J6`(4%_pNWm)Ys-jcZbYvJ&gF&Qknq%NuJWS{2kI&JLcacu&Fz
zR75z<+k)&Rn=BsgJT7O~^U2aE_1BK1FUNiv4<>JBW;$5UOpYun7#H1Ju{e2@w1o0h
z{>5loB^ic<vIE+O0-Nru9r1BDW7d59NyaW7@8m}JPj;STc$*B-xs#so9^b0uOV7B9
znPWju&*hECt>0pfRv)j-{}C9tpq4&wA{=1-MrrWO#;x&#METJjp%gW$J_8~4QuF;q
zk&gpUKS}p>eWZ77pMzJ^o4B4D;>1+hGfXLcOq?XivxllFSSFHR5YMJ^rQ6FZRq^5-
z;D7PqC|hX$r+ulSvaURg*|`p16Z2WkZl5iov!cU3X8KZ7=c!Vw<VRFbwNkJ&lawZk
zr!*aFB%982qp*QvDNnC5B}?>JRo$9JYj>;@XL()XT0dDeibFLb<9cIXRvcRnzun$M
zzN4xv-(m&oJ-MnfuCB$~H*pv~^a}oR%dx(xW$(~6f+=AuaXD4az;eo3Q{uq(LP`5#
z)ld7m^)<)k@x2aS`W8%7am;^qWQ^U0&nN#v4vs?Y$eE4HwDtJ$E0!S&E`d)(HjGKn
zK94B)ao$jFz2;JYq_MPd8msHIrE#&o$%ohuM3FD9`yDtF#Z3A6p1CQrk1N{*>&?A;
zKF29yv6;CKt4ADIjorWRoS1+0<bXdHOXWSyICtM*91eH918V(whYOw5=g88oM1A$P
zk4v;tRv~eTq|d{tKl;>`@_<*=auh|le6SyjM%nT33*YXG2!&KL2L><ydh(p|@%YgP
ziP;%=21f%9Ntk|f&>`q-x3pcLJ1aoVWbo@UmEPXB?Gr1Vy+4>PYjRzxDjAF~u_2-1
zT-rx<P_Q6PsgvnIK*MQ5{2x}i8TdF%>^Tiiy>rdiOZHAmrY`AIo>rJhjCM4*YFpF9
z@+lc>we;<mWcP#OPe@E|zGD&OcZfbJH|4DuQE&4Ni;3tCSCHJZ%i6LBEY?28kn+XH
z*^g%?z1|yeh=Rf1QKk)Rjhst+Rk(6ypU~QQdNSVl1o%8<&mhgqw)@(R)ZqlWeg1i8
z9<$P3I{4(n0l@&Q1j%D89&a_-{4ytf-k-2n!8>;DOKvV>%l-9Zk0v4?tFzRoo$&km
zBuc+I<ino8*bAPcR%Q=UQ`JJps`qD59ptwC+&x#t)OA@>@43q7N6}(V-k$i+zV@^;
zDODbiqp&)v^q5j&E%BjTz-Z-4k;hwW{MBZ;BR9n*&T*WzGk(MQ^*i6)%EJYYjXh@>
zjONy}+oQiL_0r0dO8zp+TFcFDuahDrykz%I&@AL3_Q9NL^;%Dc;|piy@>uEDa-DvN
zYqh`6Y`AsE<WyeeTh{w$ZC(WJ>#Wi&7C(;XMRN2UQSMkFHLch&TcWdG4F_3{XiSN|
zE7x!B{<N{rae{>Uotq?=u}cm63!dk7A<5LogjpXS_NQr(uWw92UCv2o^7xE>3yjz=
z)MtAmD+5$Dl=st|vEeIv*)XBCton#Y`iIF4N*4~&*{tKu(<^n_x_cc;+5CTfL>)Xi
z^|gtme>h^8Epvi-!KTS7ES@-^pswneWwUA4>Ch!H&g!?sPrngsp2@OZm<fLU`N_32
z)otN>g0CF<6w8vGF<{_iPKcMOxBnwq@F{W=?f#ikH)q3V0ZGO#k6A}X9~G5kJ^0iW
z5N14aC#y*IMy=@Ehe1XcMWb)NP<;Jtr0}tsrXcw-;vC}Q8r}YJ&7n9ATh}v2>74c2
zaS9?Td-{k2YaEkj?~G?VoSswr;hL@Jr1s5UL+2}NkwmaHOM0fF$)&?3M9U#o4Hw0S
z$BrGU`RJLdTV4AJH%sU8`P01)9B<!zeBhWC5n5mte(0ys>BnD^&s6hNL`&RsIWEjv
z=AYPg`9;`Ld;3C>NkeB3=Vo`mB73!4n5JP~E`Nd^kA<kYkkCQwyLH#Nj>P**a;cM-
zUnSbduvaLW>{$tg?qy{v7U!o81iiSk&yLWzD6!>iSe?fC;6wj5$ff*=ZRx&W&UIR+
zKIPq@<S@LCWwNh-^jOV{qh5j<1pPX#8F4W+F%*{`oSn=%Mir@wiZL`c!Oysd`kKH%
zabnh%iq(I4(ePlhmHb&YPyAnB`L4_CjbrP$a^;E^;cEYc)`2(KJ>M<X5>3i*3~&@#
z{6oYW2#(Lp)F0=Svi$LqNAAGeXX_M8Y#-GAP8V`s#c(74(P=|g%+rQfk&}o2?6Bc2
zQ6V_MD5+ttkDO6tg<)^~arTgnh2`pHwEF!AXAisVE#w9sO=%>Z$*mVUGuC7(sdM^h
z`os7Z50$-p&dLZB+6fBY5m*h##Ct)~<l%q3{p|idCvv}!*a!~CP&{^h*GUsa`Hs@#
zz~d*rvM(t29mt?@vd?O5A9W~vf_472qggmg{jzOc#X{F)n-0HZZMV8JqnIzv|CisA
zm`yLi2<HvkK3c=B?0IK7?5B;ZB_!&>9`7^Y?&D(iT<t1ibKg0>)iE~x%wKp9*i8wX
zmj{A=z5mrc_L$nO?S|d;>(`&_Us;NP|7tOrB24v}dyH+|NBXiN><q&+CE-k@%`046
z8J=?2K0)&1gfo<c9=Nuza7$cybO~nv{0I8upMukWwC15B|A~A7-5S2I^&O-c(kc3j
zNT<*Y-6?vQH{#GE{DuDDcaKO@7!BPN-Q&MDKzDbB2qubd2owE1^!5<&&);+VPWo$m
z9l!fQCi;6$7~jYE4!RdzDKsIP^_+QbcD{#~eb(bXi*XRPd}6<-|ARDXUjK)|-J%wg
z2F@);X#Vh}G}5r+%Q>TlmvHnw4hQd}HC!k5;Tx&cDEsN^FR?f53Vo<d2}npt!u;MZ
zHY=zr7Ks`X*r&$VbLQM*4fA?Wa@yv8rkf%?(af@71ciSc;mTU1gVXmxTt_DvKK`Lu
zS3Z6q$~Kv{Rt$tde*f5n|L&HB5^YlobBDHB;?Er#Z20T8ZgbCX8UrU-J|RaQ9!bpT
zZeI<vn;)@MJd{0=J*s9qaN-Y^*=7Rv_Slc%O6Q&6!g1XS-*Y?o1$xqt*^f`$^jv?j
z%bH>*qx>u+9SKlSd!C5vyt#^+#hf#2Ee{tt4?H6+EvdJo#}Zfi=^hHWB6l+^dt<a<
z!xK2_H#_|%E1I!?vkaQ`{#-o5%sWSdRsWTL(L3~=_0g7CZ3)UYdt&h4w&)~k{0)2l
z78y+T-r|7Y!inY#G;<(iM>7Ok24OGg&^xrXsx2n}QyRCvwk;7DqKgRd_J$z4`^WFL
zkZ<wbpJV^-{ExA>C%Vlnf5id}p4}#{`8Na^V0OO&1nB_X>lVUnOFA+UOaS>;PQ%!m
ztu5Oo1$tZmoDM=3WVkf9-*fxjFDz<%;akXseE)Z+7~kIpxUDH}Ng<l<+Y3k2d22m0
zk>+~v*OoN=UODu*y>g@x`uEQU7~R1z-(OE{|Hn>v-l7G3b;qp!WT>sZ_;Wtn&9>Hp
z-q63&_g^+(Yv20{CE{aZhPtd#s_;kvTspEaiBez34Gaj_f1V%iv?3tS0g;BA&!^C|
z#sJ)VN^1?{-S(5?aP!*ela!Q{mHE-2hzRxh`*7h1H@qqVkM+><ahyM|Qyt6-SLkn3
zyshUqV&5~cANyPe522v4XN(Hg7x3(Pe1!7s7n8=|4)DXBUku5%^vxgSEIhf;AkKW<
zVYD-A)^V|m<TTO7YQd)c%+HEeHG9LkM7X*hrh(@Tx6}=z3Y_PvigN&m+H@k`z}uvs
zG>YJs|Dxo@;l__KG46jT7c}K&?j^vBZ;sCnpApJjFLPcKecF}nytz?EZh!G?Z=*E7
zYHu$g-_pIML;>d$8wqoHi`}D5vA37Hj#7(k(d7@0CVwI4ywVsUl%i_YrXAqe!u(xj
z&gbjQ%*@beNlA&%lA7auGd4x0*+At~&Ukw|el5oqM<Uy^#U!O4hoYPd-$tge?%%(^
zKKTHrQRhj=b-+!$f|EQIkDPCkE_9$aI8M(l^?L&lZp-qPkiMu(Ja={oE?tZBAQQXR
zdgcw>H#0X+CG7S@-J$0Z?gjfX{p392m&48R3fd2Lh!BDaQ*?0MQLahvw?QyMdY@4+
z(bLBo{0Qf@2dWCST&Fm8k(qh2Gs}Dl9zk?S<@$M%j;;qi=|9CNSR`+AW7TK)-Fm)8
zk=u|byyYC%*IPeHjBWoLaewPLiKvOp{+CBAfVW=cXKGdWPI&^;v=ulzpCI%$Qkn)l
zC!f<vRL+=~zW~?P*9V@Q1HCG@BsoglEYLd#&!*Jexmz^+1MbjE><x|oR$~a9WF+-)
zLg46HkpWF!PXg^1@-?_@N0(qUazAc&aleT~_t+1{VYqUxqkeJ^h0{{+&?r2EfM4(<
zUB5Lh0&cz$>@9}}j40;H;5F&DwgKq|Z4FXBG-AWueI=gvM+?RLXxjvRAM>rxH8QDY
zKMBp5YY?}d=se}954REFZZu3OdoppeIjz}{`_qW^35rb{L-%l93?6+@KtQ@?iAV9I
zR2T>2S_>>hb$<4mvk{lwobGmG*5ps0SX|WRg5EI~teHLCQ^zwaVKG}67XcTSY4n!&
z2$Z>kUCKj8U+W03Fu@)3o-9}NTN1F#a;LuG*L&@cclz>H_?a8^d8yYn-zng!&M%KI
zbbwy|b@C4UXryw+gVnNF#awXtgjn)7_QL#?5e6_O-{Id=DiYNoFz(83e%;g=_J&=B
zC~svX{_}j#P<@ns8|mQ<jf!)5Ln(S1%3lng*R@5xMBWFt9QsbR^8eWT4zQ@QE#0;a
zw25lPfPjS&6eJkQ(zLCZF_1)(BA|eRl98a6f{KEQ3Q81AAOb3gNV3EP5>%onQ8G%-
z@%^VxRZ$>!Ucb5X=FWUG{rS-ab;90jue8_x*IFw-YQW0xbsMioc{}b{c+W&jOUsnY
zyYFq^mNx5o$?4v0@A`$ln*Z4#Rp!jY`Qvh1fXhHP8d;upL_4QF&#ev_E^KXWUAPiz
zN3Ge<skdXQGMwBeq#A{UGT8hQ*Ee;0`|jJDyn>UbC4PVWJ`c6;UU!{3b?TtS8O4C`
z9i?d&1z^vg1S$;v9qQaEC&dr+)Qcu$IoGd)))Dm3OLDX~O!;-mdKG?uJ`QJ>G>E)z
zX=!QBo$)2>P2zt%{emidYW`f@0@9L>?|4+?cYgMPuX5Vpxou-W_*BvhWsZ^%CCQ}<
zE$<din}Ci9o6!QO`^CKg2N6frUsBTEj8ozDh^r1bPCCY%a<8J;zY4vq7k$yOEmG6f
z)$M;fYu2m{m(sJc6fIsryM_qs#T{EbQYm&>y%&o;ty};3iLsIr-J9#d?c0l*%g{Cr
zoh-o+i?>`~<ZKFdQE+JcE!h|Uy*m8UaGQQR-rt_TS8zXuU^Sl0y0KWHPaO{KSZd~t
zmBwW+&wP7xvHsJYEnV%Ug8XaDMoK+z;06x?`_t4ZdU|6gOu297Xk*iHex~S^_|K%F
zCv_~t&{6kEp;Fc5<B|s(ixbf`>Z0>}-aK@FOGrqVbY!Ga))W<8w#-k;yckV&&^m|u
zd#J1PB$69AEZvPcKmCbqK3VpcX^5F-l2{?m?FG|1vy~}}??(7QEN_LlooRZ>nFYy|
zu`M+@FxIjTcOQvSaQZ-u_pqMIxRF%j%%`$+JA;TA<Ir$$BjXya6q3M0&t)4+j@=vU
zp&a)!T?EpNU{Hgc6w^x}T|oTAEg#}mu?3Jdb3Bn8<8W_={UGuKgGS^|i}=Ge_Iwy`
z8Rp!umXk45_NbVC9OG5Y!IN`gEGu~#<0TB6!n|x~0hq2AQ!H8Z{d)@_D+I@kgJ2AT
z4>1VwLdN@7r?Z{CAr10R5xi3SCfiG8@iAV-7z;sv2II2$i%tJw7m$~cSF+44hqY&t
zjK$^TG82y-mHG@?v;0t!xQsPr(vzS9<K@F8f>>+jU|G~)UB+B}a`{*jV)*_nmSFC>
z*ux>7&v;#)B%2k7MoUs=h*y~FZEZL6ZO~jfn?*-ZoqN+GlOPKR*y=+fCJs|k<omrJ
zd<Qkj2YF39<#vo4l;q!JKG4?<qX);*nA2g<Bq9LK=FR1=@g;1}_j#Kh9kwN+25?my
z)GuypfrGSl{in%BxqMGuAMV?evh@2!sp7K_xzMLm<9%LKyUh%bl;tuzqIL;Y5ZnAE
zfr4$_O5J@yj2Hm>2aAj>y8Z7OGWeH+2f#l}#vnPD#h4sY%G&u@iD$6Il2p137lPzo
z(!8@6j(iIHCqs!aVk5+T2NpB$9uhAm+vD#f=aaq(llgHAQ@wCp#^gt0L4T4F=4EUl
zB3GZu%R?f|5w~E<+Gs-LF_C2rSkJ@b85feunrV!NLVXwy>ua`>{C5#t#+x07WAQOD
zRZ{X9w8e<9PK^1p=|AiOg3zohNtzDPi$PXm`I*ZwlDI!npQU;jV<3s~_u^g5A+V`I
zPU0t%!Crk1-utI41Zz1NFXo^U)6SDBj=>xTpU`#++l?ptUr!oNu-CL3+2{}uZ`)Xm
z3gX8b;_lU5UTKtGv*+Te5COI=Bi}^#(a_RzsP?ygI&hcfASt~+I$}eL>(p-}wsXAI
z@PXmm&5>ycqJGbh6dWFr1a)e4qXbciJMXWzYCq_3+zky4&k(+RzN+rNee*}Yl~-L?
zcdfyj3x0lnlaliv6w=1Dz1-(7YdtA|-hSZKFlN<vP}9`3gKPWOg!IhJ2Gq>8mM8W*
z7Me$nMAW;pn29-^=-v!{?A{klnJ~Q%8%Lq7K0D@8id_c#%c8JpV_4=PrlD??qqyFE
zpO)6s8aS9>oE1>&ewoXSBB=Y?fNLt>3~t|*#@uXfWxb$Pe;&Pb$yK{l?>ak4yVAwp
z+S9p3mn``$=`qXz_S0Nzw4YnmqW>M~*K-J&WYv0MF}Di(3(qniOnjT~nu~m^!>~U)
zmmiFscz$b^VqjZ+I_Yn4ti7hCT6HUJ@~mZNBNTaxtjk|GxB8Z-e%L*8+l>{K_tf$p
z8N7XLr&@ORVptqU=-Mf<yUO=byS@eW;8f9Vo+$#hs}#CwLOg|@N8*(z`5M@^ZJq0s
zTEs8f9*RaL97D7ww=XW^LAc@j>m448Hf`>lC7~;@mqN{`AGdk49;Mj1_s8X~hJ@D4
z^78Umjm`tz)if@O#{|{Z7ca3RJqM}Y)u$kRwEUH?c~YFRCpCGnzj4c!Efj$ry1G=I
z6VaG3`o}G&Y`(InIq0@lDZMV*xxc3-1Z`8h4_JF#<avNF5LMJeNzUf`#;@N|ihVz^
z!Eg8)m#v!9##f9ccQT=Io1VIw2J5Ro+-EBN&{xsL*-XK%#Hlm)@oIs-lq*{t=qII`
zr3Jjkj(__)gQkRC$ZvrrC>Foz^P2qG)=oR)lJ3{Lr?s!w(aYidk^PcpZ*O0=yS+Z&
zb7VxQ%C6~_8})p|iZr%gM2caQjUBpE7I7uU>Nia(+g)Vy<?u)~nq5R|t<GxvaBH3P
z-TU|NpT9>0#JOv)^RO>uC^P7@H=k*As%L{3!P;p7#_e^v2!c#>8-JQp`9sd+H75_~
zMymv_E%PZ(Jnnt<s(^G*LPFnN-y@UWVna2RKKSzu^3rJ4Sfg<oZQ~;UcnMeNuRwt*
zTG(wretMDA#X}03epOFJNZMw7d3;~Xd$jqejV*RC#m3K5zU7^SOZWSly=ZB$sr>0r
zdfew2rQ!1v;b!j>TzS}T=IgmA#C>@~iZZEr3B|3ywL!6~A-+);omFNoD!8-78`mO0
zxTbcLYr{pv#dYJ3RU%d)50~xb%a=EMl~|gXGz0i)Wc1UYi0?S8Eon@LnGu#H;n3Sm
zD@2TyseKtD|93VWRj2>sOgd|r9K#{3-D_HHC4IGNUV8W4(yjVhQ=t6Sm6XQd8_xu_
z2kLXX>w6GPm`s^QSM+EK9q5Qnc50$?bxdX|>K*dDJd|SlYJcnU{5{70F*ZN6;Gs}A
z!7fM-bbb5QQMb0r$Kb9?;E_f$5KX*ssd>iD)z)bA%=7-ug*hp-a{9thG-zpdwAQow
zxW0p%{cN9>V#n4D*`q%?zL-}lHJ4tt%9N>o)tcMiI5^PAaouoVi?F8OO6k5wo*(*i
zvneY2<9E}SGz1-;P4tSy_C95M0zr0?nbf`?<}ST1y9;fTTIXWyd8fG$)$%ITu)Vk4
zpSjE6aAOivn5T<F>&J2i{#~98b=6Fg&DewbCmk8qb67hna&-79CiByuSZ*Kt9?1Al
z{LOY2NpFDo$?{j&0>Iwf4P7(ZqlS2p&^YXWOecgbdHBFO1!G@Buq{J`7QY?3b7VZ{
zXK!X`7sq@$(?=z(F2f_lLjIG7EF5GYs6FH@GDyXk4SN;v!E|?TXX7ADrmkc2dXk)+
zQ@`uY)!SZgn<<syC)2t^Q0`d8I~<$1MrOmjvXuz?m68|b3&PYMG@a=P9PAEs>nT8>
ztH!zqRouzbarTAyAh}Ln+ODaz##$DY=zo5((J3ye!-Z`jg8fSz{Kd7R&$5G9?4>%|
zst<b>xYkL_Y41gTDQ>-kXo_yGmZ<gA%)aBZv6M;9l4k3k+IxY-gobmH<PqwWo68M0
z`193&wakLa!~ll$H=)Ai6?k(Hj=%Y$v=b&iw#3=@=K)}k4Bu0v?{f9DydWjoWbtRE
zDHO#%*(YBFXy?zLpDuOi%hSWY1h63<Scct~+elxd7<~<546ZDU8rSG^UHj7m3+iL*
zdUea`)(vshZ}Mmz+jAa7Ch_g1c%TQp&DLX9FK2NYeV(V!QVlE9u9^1@`a5-eE}Q%V
zmc}{60dDDDQT6aHIE0%wZ%(>?`#@XsTfw(^gw40PWVJ}gf<C^vO&^U>zx&^xG?=U%
zE-!uP(;~TzVYfUgoO!OiqFrHxtZ+uw@Pt>j%=_cs*dzU0SMQKc1kU2_f?DDQXf&k%
z!>6LgQdG~e`OISPHj%!ff$L&>(>JO54eI6;pF7`L;_d;5k7L|32aYHT37Z{i9!yQ{
z-A0$2{dul#gzz_q6~`3j<a=GNPkk0}f&F7^dZh3|QZ+EbeayuklBEyDh&~QUJ;To>
zFT&b$XQrX6SmkUz$D=*B;9-y0bJ4BZz%kw5KR<~F!6S7OQ{tr**!wKDRAgt;PR<p_
zL4&OFtWm!O??1GBH46JD%1b--v`u;pJ-^T5Yi3$n9nNb!O&CgI$os(f>1{2gIN_g@
zn1?ImOmgmDxiZ~*N+o+kLXO2Wk&Cu)Uz;g?O7hkv&_ehtd%iw0)URm2jZk;WY%Lp`
zHJ_XHA8S26unA|6H_?45u^<OwWYLIc<z6bl$b(|0&s^oZ3QZ&XD})sf^=w<JSmST;
zaw3|K-`9xL4cTyUSJHAghSdz>R^agQ4Zr!Kqz~^rxV+R(&}AQ)HI>o{Iv<>x{}+zx
zAa4;NG}dUEYh4DaU4my9_I0<W5_e?Z-S*9%`svm>E@G{s;mn?yJo_<~VN2&xpj4#H
z3Vu)g;0TWMD(Fn=OtcGdi#9I_Af|}(t%Hb8g|k#ETml?vcx{(m>Ue9a;99@$c2Pp|
z1;Ojx_3lO+W!xmiHqQ!&Pg}H8ur%9#+>$tI0Nd=BbL2u6^ObL&qT{ifM08b#Z(~vX
z>;^l@TI!U{al3l+0_}9<n1j4ODOs?usL|QGaJApF<^4VFTYkLhOHu3_%X-KYP$&>o
z`#4#!fo+XhR$)##Jt%pgJ+jPHOnNY`w;|s4N8uSeHUonW#6SAhCjAc;sb2i64*s50
zzFnsBDs&{yOH4inuP4wRwyF}ubz}?!iiq(K1fxlXn<g!lUPMi;e94eg=7Y(%`{+{5
za?uz}To9q45nNBccZ7QD6Z5)uEmv~Rs7%bVM}#pVr^#Aq4kl)*OAfut+Rd?UC#NyO
zoaJ#-B6Jgb{D}?lh=D>Vf_j-VX1tj7){!)1&S)5&^nY~dmn1p4XkXl}(4K=ct4B(-
z3Fjg+2t*78jYY1SkMM{HCGTM0bV)=^%&GTBEY7ab%1|Ilo<^%fTtP?gD${ndmRDXg
zUcqNcP>nz@j^!(fAuuUGE|R{hon#FT&o5?n!&OWBd2&;l^><i!%t+dpCH-y6u%R;^
zVC2E2h@LJr);CjQZ$k=)Xb+KCnL)$8`ym9iOSZpWZ*cU-@7c4*D($Mb_bBRSN^>d9
z_H?~_1SywL)5lRJa1DCvId_{kA4Fep98LMJFx;FLd*JJ{qa?MJ-_q*yBLihDt4j7~
zW79qe@I$ly1cxU4|Dx`_8;<7~HdDXK&H~1WVBooPml4wfc&VjSt}(%~B~|$Ck06a?
z=YI50Z`I#mFR7?Fh+}-uycb-8jxQJ#qH8tMadiHO)C~IQ)q>f{2E0wg0$%2V{zKg2
zsoCepR<+P2Zd5=hJ{Fr7C&cw`V+0UZ(rHq~b_?stmALzXhAG*bQJl`4IV105L;e2Y
zR?4H?8Xp6XP`zYl!}tLhM23RH_2llB@Iw+AKFh+djo35d%%8kK|E@g|OZm2|@XTyo
z&9V#2&+zm3xD)CLz=7jUZi;>TOgvM(bsd&cj2?4Am_OW83DiMzb^p(dhp)`04R(1B
z`V<!BVBoI|TXS}++_5k?Wes!?mcZ_+LRaKqPh_Z*l3G{DG-~?qlr<(fH5XQ9@-E;1
zYA~;3@_kM(jQpIppMpkP<_~S?rHID=3bqN#p1>eEOCZ9>4^tc?gqJBs42I?qWcCw=
z@iWPkr-%i}i_}xZ-WsboSf`PUer4Pqe)-y#8Kra@0wDb`6ywJ7f-o%kCnZMny?j?s
zeI&8FRx2?t4;-isJZ5P*dD5hfKVLdop4U&ft)VEs$9BHir`wyFcAfcIz_gBAAWA|Q
z9?KfUl!lu_#ybzy(Gi=Hf>(1?1BSN~QJl%!znSAe+bwCSoo{?y*aPwBHNhJ$@-6lV
z*N7|zJOq39ky;2&0y_6i?=<ih*TFfg?&ng(bc(R`o65iD!W${mJhS)L#Ryt*=sEEH
zRtZX`pUlca+Q}B9tN<N<Wq6O`n$91`;$&nLiEL}phXWF<AieM9={m(Gf7{}7+>OS5
zN8ju$$RT#XIEj9Jer9g+(G3<YPr&{_Aw1MgbU=GX&ShkN=6wqMPR)uHK)|fcar2zB
zQmC-EF?rDW;t@iZ4aG>SgJ?r=rKj7M2n!1rl!9x2Uip?%?7f4FIzOzV%zbm_V6a<q
zSGoJ@SpPeB?${RF8^3#wIBL)ioNDjY`1IoB*x18FWP@CI_D`QODo*n~ZAx&KPH8Vg
z3(k*`voWjzzQT6^*<4<zydelt(Z!1wBTz?CHZn9si#VMuU?I@((`_pMb$|c4&WVTu
z!T;oZHg_`dn@=j$5y$u;MFd(n#m;qQmi$UbJL`c9t&)>Ez5$<kvukovn9BzmWi2oF
z<hI!drJDCPCU)vJ&HCNh7QwPnp9uA55Z%lDZOthV?Fb+;T}@znVcRch@81L2b=k{n
zH?UC;n)HtzJ-R4KL(i%G#TZeIh$+cm5Z+l-ypT3h3ZR9i$RhTTQY7WjR<q9O`(E?J
znLqjuW#3(V^uehZ0?oh*`DQ68@cK<WZQ#(^prfwd@zCIuSGq;|H&4Y(4$3lkF8u&y
z;JdNnWLrHrJ?`%Al-fsQ;)4^DkA8czXuwF)&fQazcViPAzZ;VKUMeY|5#c3^^syBA
z9iM%G9TOACexMPVmd;)$_6MJdQ~_D5gV9`(fXmb=TBKcF`f+TQY<+mez@39uN<NB%
z-C1Q$V;54m1=l7#R<>^VDkH$-r>Wjiffc&E3bM^~iXh=kS9MN3UBCF|szcL%yaHft
z(n@=M04<IqXv5?A!*gxV!D9-%8zrSDjTaT^DOXZ>23zo(0|Q3Yjrb@p_ugj~^I0AN
zq3?+2OByKft~&JT_N)GaXGgz*i;jGK*8D>8-V~$ubHV|?&C)IJTA3+rFgVbkT6DYS
z_MIN8HvD4gd3GV84#y)BI|?nZMOvFW3Q@0Lzb+IYsl3{(ps8fwM+p#g@o0Qob;Fo2
zJX{{#4YtYgCb?lDQJ1;;`$`k+TVIZW@I34Zoq#~Pho8KCi;_zNTwrC@&wvTqz6rv&
zYR}qCO=1x-%zV_MZ^37=Xi(?lVR2iL&tiH>@=vK#xaF1x1_B@>E^RYazTCNgn^cii
z_CCcF#INA~Xqw6wB-mTJl@M6rS9!GMGDVxqC+=<XOmV(~O`_~ms({a@Q?{{9hs8A$
zv0uBGI`K>Xp-=E~VX^yU-86$<LvhUY7n8rwWt{0Zhr8zU6XRT!nL0LiWjo*f=6s}8
z#Psvy%6!=%pi<ss({!F&J+dwT{8rEfG4ZO($~!j4h`Gx?x&XQtDZuNVjIq3SpN)N`
zm(aGyv|zMp<`<*@7<H}L(M`7VtEHE9?Ys0tZ$0J$HfM=bmIo_SfGjk#I28>SCgiC-
zS9>XNS}(_NC5mLuwf#EHBRo@V4b~NrxK91;+sSy_Jc@`bH32H8{#6Xv<eiDk1t@fH
zU*F?}hweGmhffgRcrVb|q3P+Ba8XO4kF-(XV^Gf*lDTN#W-|GQcp$K20Jb6Yzeh~@
zQ_0{!C#b>I_;Acws$J<xpt}|a?LTzrP}tMzAgW{8nc2bolhylG5e%HFyxmW_IY_Y+
z4n$VOV}xoQ2`K${eU&>l2K#%v>089(D&7gpRA%oL?QKe-;3WV{-nKc0hHf(tZ?({M
z+O_$CUh-PFoDZ8z5s_$U;=dLcD7cq#Y7=7=kYagmp}i~Y(VC~!tv}>c%e##~Tv@!O
z+~#})U%URiJc~cj5<1K|3^HU<fx`^1$A=>%>Ux5YIep|4YcA@jRvff9F|JZp?3u=R
zM^L@{bK(#{D4&LAa{DJq326$S<8OsMj!&xDr*Y%J<jIq(aja%ulEmu(zfAGDyHjo;
z^~aZ3bLkdfNE!Yu(=i~oBKjX-5E_65&E762=985+AhgowOQ~uy)U#!E=EZAVkpAG_
zkF#gaRBKk1k&!8jI?yjnp7JoX0pU4a;>yma!3=;a+>IJl03D6x=9jZETXh#LDQ$7U
zu`}MIUP(byON-C1&dq>fd|SuuIA35_s?$h5fb-|dG{Sb!=NvEpUW7w|64_%=*fS%(
zczJoX0(1D9PE!K3cFJ4%=2tw|F7AoaSCYO99!xD<gP9F8dHOdUn(~tetjAYEr*-dD
zM$hv0zSgWrINar>1O2^kU(x6{{*V?rvCq%|=vM9?c~0v=e!fvkCm?Oj)HeD~hrIAT
zYI<{GfHl@g?on&Ibi%DMp>?}}yr!lmZ@<_<XXk;FW8qc>Ub{AvTa$pf4S9Q{j`LEm
zgK<;#vVnn69i+~II6D%1eXge_xgy-uDSYUzF1<{8lO|MKc`Rfj9~ii%__(;ZO<#YW
zo~%O?b{9(02D7y-ln?=#uVC}y#F&q#xqdlE<DzBM;E+Xng>);$P8v!AJ9Q5@S%5>A
z<<nfu3Zj7aXD{K0GH>Fym7i&HwhddqI=S^U!6Vy18O%%0lJ6lfabr~XYV2Y`;tqp0
z!(2^zG>#TS4!+@T)G$JG-id+(2tB@eMN`7U!2dRDn@o=?ow`IqBJz|kq3l`z7+b*`
zwRG%9{#yAI5)6y~Bx|jlW7oUO3k|viL0D;Q4E8)Rf?iFi8VJ5NiY4xYXok%3#mH~1
zb;M@z<Fo-BC-lnh{`S*`+n<X^QBm3`-DjOsSA1g<zd~P#=h{(rkk7H}gKIz4_`BXf
zr5<;?got@U^P8qeZv`y39y+v?SXcH=gOq@GGz44D3;n<@^5MYP!4~i2MFD}1{&dE+
zfjwV*$o@#9Z&-8i!@SF<312Rk(p~o;@>Fab+}AUdP#}f(Id(7}IeeSEv=KYBZGO*R
z*b;D6ctu`;frMv(7z$Vy5mE^tTZxkn*Qw|-eYCs_fe{zxu4UNkb7*5aotxeDazbwd
zjwwuiBWeXx*if7}N+h(lhE8M*;Xv;}=KNU#JM$xSdad8U?H%DG>rq@2p@fA49gkTr
zvV$&1YwazI;D8LjE_^O^3ijnXQTp^t^VexbAG~bW7J(Trz2`UG4`;2?glhthdHu&v
zl6<(O^rVaOa|DKPVAOqyv5grLF8qp(=<tmS>HZ8K0I`o9u?Ht<16uneZS#|U1gBH@
zO_0#Go6vE1zEKd@Sex;_)7-vUlpmL3h%+I`*uerGP>)@S@O4t-Ag(Riw^>}B{0+Pj
zWplg<@N<tJ_P;T1(-*R{Kn?+^R0xivUu}aI&eT`>lX?O2<{0t?8W)_DPb;l;9GhR0
zH_d6FQxzO;H%U9<ER6jgxJ_N&$$g8@#2@NvDq~+sxE^O#dinakXE;RvfDAv8aX3@6
zX~T9USs#W8KWm#xQuqJ*b9fAlT*QCoy-X5v&V`|DSvrQv{r?Z#!TBi0=l}A<!&$vz
zf31q1{Ha+^*;^zM+sZGm21Et2GJ-k;XKpYpPO$gyvQA(Ac3gD?$ujJpSlRHn=~6=K
z{atSr0f^#UPOO}MhIL=_+qc;?;Q+e@48~>?>mLz)bv1pKnATVCPlYk2tl%&6bzn<+
zNF%*9Q%)j=Kl_QX%COgsu;%28{?S|56Bv5*@WGRU&mJx7!+*xoZZbC~u$Dj;!LP)w
z=_QS;UH^$u#;$nZ0LPY1)Nl_jY+;LY8baMAl6(m3h@)h}5ynw6-0^aHot;HrSMzbx
zEy~Uvb{VVloAj;apV>8+Kl2aXa$`55^7c(ZRQcN5UU(wm@zZjf^?|)N=~Tr5IKltS
z(0WeNb77T$bmHq@jD<CJVnmh1!#_VZRCo8-{$>9M#1(j?4pnFRhZa#?B(&yICRBe4
zE!u{32;#yfoJno}5rANeJ7LA~R6^O`>Bv<$@>QCc#@h>Xo;*~a;U6`AQ2M)Y93nGz
zQL<FWZx^M?qq6s;5}BfxxI|k48@=n!_-r5b)By>k*{=aAMFOc?84^qX8!CulrijH_
zE&5^H5^!aCgtzx)ND-l?vAeX&_bNh)WF3&o<@)iBA5usVEcCbfw26~}%jqKA5;HU1
z%X!Fa5g}@J9Om1e=Ms&^o>lfJKroVX6?nhQ=A>GXUDTiOOj(<*q0t4|HJ#?!F1`E?
z)Iy#Rh1di(+mCqggARQQt?7ZSze6~!I!jmQYkL)5lmWuxC&!I*$oM>(Q5d0qh>`qv
zl;9CF{eSv3oN**F;+FsBw4!Pv5)g6`@%I7BE-fV1*h9?a%TDv`iob^qW#0XrAqbub
z*nT~#+@&h)l6_fzUDV<cEFym)Q@%Bw;9w{O5d`6fS?JtG0&8F|IY6Cq_f$JrkJBCY
z`MMpjJ0cJe=+v<%Z-k$(pJCsGti1ZmpF~Qp;^C#9E`8siT#*C1p5HQU>Al-F1|A9>
znj@S{>Mbss%RL)|Q=ftEv$H6DaWb@MHx5$It`U-xlZ)ZE{Ip3J?EIf$Fh(#7@#}wv
z2RDkavUS<N>qFgr!V3w=T_T+}%Cq$ki)c=K_~0W{A*X6#`fE{ZW+pICfQIxJMxlQ<
z2H}nH@bDdu-QC@$BFH!#;ps7cHQhF*?awJTbvdC4318a}{QQEIx3Q}{L!IJrC_;j9
z@HUL-J~BlB-rnL8-ga{|@x74^V91{oB$l}iZDyGR$2R&lwq%GMEo#$OQc`NVRupfG
zLjl>SGDOTr4XMAQ#XJT=@OK&_EC`eGtu|2w#Du4D`A39wiA2^*5i45}ze&_E6m)D-
z+Brq*(f;w2yZ&4>%i0`qUpGN#hS$KNoVtSrPTG%3Z!wMz`Fv{p<Uc#dM`qVV@Njkb
zmh@HO)GxBiayu(IbOra|51jlH2WwM2szY_w5QqF6u2c9ms`gW}If8~-L|`|#w-E5)
zWS7BBC94pBMnq*jGN02#f!AgnQ4b^H{lcA1x&r4KGvykWiHmRVYrs_20vvSkkJS=?
z{7m!vT~b2;T>AbcVq*PylQaLCJ#VVV3kLXh7#|RpWW{Zq+$#%;IpR?Afu+I?+XrSi
zwTuu>p}AEn<$IBk64HYWpHF!na6!Zr0g_u>$`^PqUlq1Wo=BCKr#vqi$KlB8PqqKh
zFQmE$@EYiYPUR6EDe-H%-zm$|8j+A;b0v31nsS9W!&7P$v&&NGJWWkN&gIXXhmb1)
z{<YZu`=60@Do1o}i>%$O-cKCmO&2VI$60S>P6w;1tuezsoc>5kVAk(fMqxTrC16x0
z-pa=b2hhjniG!txInVHjD|6*PHOLtjXYJu)f<8YnR!&|pB~wiFk_GhnM`147b#a}W
z;eq6)k#!+C&!rNhBWpJXU18#CDV*U&=DvY`H5S<PS9-rvH7#0x55HJ3uw%4&ZfZ>G
zZD`Ne(^EQySoIuPs~T(;yav9FMm+VIvu8(Y$F)>?>jfY+eBZu(8U6D{d1!k2D0h#h
zD%f(%`GM2iun6(r+-Nh!=fAlB^|uYp<LMk?%ew1-mUaHGw3LJdh9_|$DAt*^^dU|{
zN2o^#kG$-7;a2N%zTxXeOu{q*XIliAl;a<kQgW;UvB4bSI}l%_D0KNZI#$nKD%Sjp
z0WTYpUBZo2&Y+d_H`wluUp{8y&^HX>{}%~1Qy+2|)_?Gq5tAh-GU_tNgdAR8gHv#n
zA{q81dK7`3;Wn-fza6rhK-qKB6rWuIPQNhC%Bql`hS~!^Y5INcvn4!I_Q;Adk~>G-
zOEn})K8>PQhq>;|h5r|MtfLL@Z)^P#yh8r^FX9-^4LwUN{%5H-)~rWx;m^PR&mzlc
z9R9Boz~Q=Qgev;K{ra!s@Z8Ao@b5Tg+>2yK0FDXBxN#PBl&ORl{lMO>F!~S>ABX*S
zY~P;gv)lkhR7TSXF^_=^>j2ozdhz(TUd%j-5ge>9hQcPd0qUCH52O^*=^{$)R={mX
zEO5L;aZ^1$E)@|O#m%G_<`96VU<td>s_JTC=lWsQn?}3^zv@2>efCUxdI`WJ4y7lq
ziE6l(9L|p@0;GQw<sD@+0|2G3^_c3OJ)dJeCT?GmvfNW~KzgL<fZ*kMxK5cfU90gV
z7xlhb`mjPwKhclADhwcKzzA-&6Dd|{t}|!OT%XS;MpRt-`C=<`99>RpLqQ$Y*Y8lE
z&^2Y2xHil4=JX|MivgkdvvRy!9`I!v%Ub|`N2MK<&f~m@`Prv|I2l#d#lhtRs?5O;
zK^7GNX;XTlRpguhdMgrNeqOZWZ4lDL7Do&=IyD|w+<535zvb7XZxN^mgrEO#Ix@EW
zbB=7q*bwdlaKJ=LXy+Uy<d$tIdjq)AqN4_<2ImB>b|g+kjHhx{=Y|&fyz&Zu_OPr}
zU@6e0D47`Vsv=!>y31@U0mqLg;P{?vgt~0;B%h0~QaAm0vkKAkqXtPxIlX6LFg=xL
zhMDhbftdowLoQDANF|;(7%N<9`V2nRIsPXoG9~SI$DnX0kwAT`#i8T-o{aaI7X%dt
zf>h*yI;>L5du;e#y03cnj2Sto`*fvk8-R`LZeK6~AKz+;00t~fnaEWsd-pL!IiHU*
zYvxS*+$|^&u}{#cZQ24uRTk<-nVWyjGJ$S@VsGWyN#2JH=yu_<Pk`Wih3|d8RL6k{
zrcpRn^Wj$6yK_9k)V4#_S%EI**HDYiNzUK{;Bz+L3#OT+4|d-r3ZiX1gf%lqgrRA3
z0*W6^col?XVtoU^Y_2SH47k0u@s77HpV?LfKWKY4cc*Jx=c*CsS}p^vfqKZgTw3Cz
z|0jX<Byw^fJzKi@lA7P}(?VY1!2in=<1Fd$GK3_?zrPEvJF1yNlnYuTgmOgOD=jOp
zX$&AoVxszXfjo;Svr`(`S<W3#ki`3X-_%&3FEmpKl(9gEDpIgd8D^uJ*6r}Jg^pK%
zVHwY#o#fOjot{2ZLwj0y8KkHfz?9IUEBYTNe6au#CYVawDcy0L%EfQl{CYN*DRCle
zh(|@~pEBfG7JyOd^_9=2zd2xR{C1zwDcaU3-&6iG1#EUDxu(?4MIlKbJ+^G#TyAmQ
z<B;-00<om3+a&wY;lPn2N2HG-h80uQ6HFFW9WQZmQ0=15{5l*JNzU9v6z4i;YKIlB
zT2za?A>%xYjG5^s<$MDz?<F!eWh8zcuRH}6RFStN;yB|qienMAu-07BwJ6=DowD5i
zNlJ8xy0f$M{RBkR!d&FOqW+jg1$5(kGMo=@lr*lAMMxAs+n}shc0&CY7H!ljRd<gA
zND8zO<yNd?cT1Y<cz}F(M=49?lQ$JLs7@D;tU*Nt9tgLQ@PQdtPBI3l|J}FRf8|wQ
zBP1aqAz+EP%E)SU;d5uA#FWeAn}LBYr6(X>#>n3vs~(}9I${n{d8>fdL^*z0t5E<p
zeYc=ApnDlsi81Njx}`x-=x#NpK$W~m0Z3(;_A?8hD-ffYm`v?6fL0&_v$%(8l}-21
z>xm%KL(!#ja>s%7P3=E7^0@||fXqR_g}?t#$#9^M5o?<H_0`D1KGof4go9d~6KbKx
z*B`g}dX|oS(8`r0u^6Le8>Jz~o4uD{T`-KegwMb(M#FF*7<V!gFBy+xm?F+}P3C2!
z<^FpmqZoBe2+ES985#=nA%sO@x-X>V8(IMTeb|VJQvk*_rigd4?_qHZNr0i8vY#$s
zdiO(_woK-pIe02$bNyg`aiVg*Rh5fC<UXrHrH|X?d~8C)u0PF_ytDgo@ToP+H=C;7
z_;~Bv^RtVWs9Bq82QS&RbGrAl@mucPh(7gO`Sh3tbN{@wYJt?m`G+=q(%!Tp+1NVL
zKCe|y*tlA{*N>;#E_uLb^|64&Cs|LDduv-$la?YAaC6{`7-)qDs~uY^Hua2)SS%*C
z?O_?p`kh>R(lpMlIan(Ng@4*6YQO&sgeM=7>QKw`RS}STxh1IM_6Kgieed3b)dKwd
zTkv_2u<(P`=_RX5k4@*{!R1JD+=f5Ddixe{D0?}+XIzES3HjeHagAx2_m1|9{qOTA
zzcT*PGVc%SIObnYF)v{Jg>}ItE+y(B3ddJN_Yz;R?#0s>4`)4%@$j5TR9%pm5gHnb
ziEP=o@0M1IMuhe)l~S#i*4A9U%*@QY_#B~^9Fb>n<cPS_uldK?>(pXgk-SsG^Y?pK
zeO(y#;6bEba=@ia7Ui$bUA=ns{P{^e<JuP>K?T73VD6PGSEBS|%KTs6>@(pN`1QnM
zjEIXuMSi{?Puhi@d<|ibA0NZnH&Voo>oiySnxjBsbaeF9fs>ZnZ~!5UaL{(vM<JrQ
zaKnYO?(W%pBv-6x(GeXtZk)w$FOE*hyG37c+jZZQSFc{hHhw}Gzd3G4J(q}xOkce;
z%n)e=+;c_d+$noAY4T*?;5(j=$Kgnvb^WZwamrt!_N{{(svApSxDbwy5sya=%lwCu
zmQ`E1$4GlzmR+TJI!E%t#*G{8`o(eZv7Gyx?hZAz_Eh1(gVwu@sjGGwUHD{SfYK%K
zGOh^k2T}?qECN2A`?N3b-*4&dZYoOwXsK3IIumKHqRQ*d%^Lx@`vJa!cEhFpy(NR<
ziY@K!4<jPh=s5yQzb0#)^f-NqHTjHbmG#LVT-$6U;M#g_>B;2e<n4P3AEfi%DJupl
zG9>rbl8s(lS~3F^fHl?ok=@-|ovmSUsjaOIs?PKpyt-9hUMElp;d9hJn!t~gmoK-@
zN_Ogf%BvHjvVQ74t?NK4&HsgK`7XB!?w9ynm0(9g)p-@26V$iBZCG{V*$VEI?jyws
z-5-?&YX@(<)HQswWU7?9w)Qf2Nm<##1f^ZOu5{G$1Eq&#t&av4r-7{07JY)GhPR&<
zx9z`lRiiLJC+EK4xrweFmzNlMc}kBfUB#N1K#61P#bt|js&<6AnDDOoef6w4bCy<(
zwGMdtbjyw9n^O`Gmn7*tcmXdx*I`b8Bodn+W*}ZwyIlHILhyjzV2}5mHOSIB_981Y
zQ(dtTr#%Tbude&0{dE<&YJT%%WcC!7=+ys_AwA_@YDR|tszX=&rrX)uBk?gIQ~uy>
z1|9Eiy%Bs(Z^w>Hch)RlzFf-}nQnVT3|<&vwU4PLTh(5<V-(y`VW~_JTpR9;(|?g4
z)i@ri`uz{5M{(R!XU$(dMHyvqc~=}btNy2#u*y^O{@p3hL|3m~-7<maM?gF{`NfMD
zvaLavR~TG%JubMrxVTuW+i9@BGjD7DlgE#1qzW{S9&JWNmX|3hG#U*lLRLbH>N1wr
zDFeZf<Lum2s>D4eXZDx65Bk$~rC9uG&qW+-Z57Yb2TxQ-w1h4rvWGib*52UC(4*~7
zU`%So%9VycDk~~#qOVpPzDn?SUY98CGLW!Lqy6N{{kNU!AI9d)k4{LCU%YsI`ljjA
zr?&+br_cXIZu?HN2kU6-H%?9Y_?J$q=z0{8Jy_Cn&vjplzUUSGKmIt;qaVzkio2(0
z{-mswly8~-a<XETV+V1_wjueJF5#4GO~c7Td|lRSmuVH2Mz1!;AN;sLMMb5(p#|hU
z<K~v%<2aUkOo4lb^}(p$Lib4*B1t7C4JQx>4Gr%&)d%vqPK9DR(&EK_8<OSEs@p+F
zWScLI9J?ppMy!8AwVYE=w4cT}F0KcHa}^v^i+74kELkmO-%^38L4MD&9<RE8e{gs)
zoZv5s1EKp0*86u-_?E8RaOhbqPCz2^c*LtJ{G*V1a=WDTz??aA)~;RKaz_IglDd0H
z+taA#4+sc=d~5Yj-Wl66G3zB@lt^Kb8CUuni<ft_R0Q4s5*8*BpRc%jN_B<OLPAnH
zpITOLx1V;_d5e80Q3tkGN;(?=U!2h|GN&p+|81hpS!wj{#v4F5cWG;fJKyy7uEs`_
z-$%a7W8E(u`z`14E&J1i*Tb?hZRVcn66fBB3+q4|%N8a#eR*Wy<|ugH`8^VxOvJ|i
z`fH7p^fnYZ;#I6%wm?KAEg=KmwG2`^#*Cc+Cn_pWQ4jL?&QwxfUf#K{Ss6><&!SN#
zjw~yRTzWbcpya4ov<(d1`~#JFF0V^;ejmE~o{89NAC+XfjJR$>0c!2enSk^s)7G$0
z+1Zt1Vs(cL1%9=-t5r|3>80S{1$!c(m27uF9xS#qHKjSqpnjO7lK#AXZ%<E8ouVmU
zR3zB^-R#WsOHbcdqlpSAI7B~v3O#1|TTqe@Qw7cgI`x{1$4ZyqrhJps7G<p+_S)!&
z6%mnlsGH6ucu))}!t0aiA0bBy^70kg?|G(AzYw`ww`<k@SHBS=)Jp0Ffdkl*Y0<a(
zNYanCt4vLivH!4<p>LqYQ<DcOtE=myXILJ4_!)&?v0~e|ZOZ@>*Kkh~0CV0FdQacq
zOh*0NZYw#?XWn}DOjGyEoN3d_o<Uy9{8-v0EG!HFh&#Eu;J`aZNTqyyC8K1_q=HBD
zVDUuV4_Caqj3TAcMs%EA%SsqeAH4WM9b9b%?mcywDI9-VSw+i-9Iek2d$aGE1Cx@+
zVZ}b0m}3iaz3^RUdJ-xVxA6+xe*U}%+Xye(Eg!L~*RSWrshXLECuHmx%sMg%WTA;u
zj*Qzc7glPL`kttmvQ^Xv^IXZDGPNa$&QMWblAPkzorBfIVlL0_s~0dlp)`N~{1sEq
zU&*#_epCkyoF{)`6%<GDr6qU81%8c^ppoM&D8h;OK_|K-cgkWGgTBwmxRzq4t?g$&
zwtdV;!l&XDn7?r0y6c~jP;X*l60T<h{A^3cTF|~~>Z`&x9&;bWO!E@(x@P&?3-Jp%
zq-EDbngc{oi+<YG6&6W>`SdS)6YL(TpTyCxho`58;&W8SJ~3u&w2G6H6YT9fR9uW+
zaU2Z+pk*i31!qd3KDmV3=Qzvin_JhT7L~F6+w(i>LpIBd)Ays3tns}D-os;c&$)6{
zr-@}eEq^eIp-<sEK>7|p<UZbBw^*bD$(rSgV^AQXsk!;1v-p6Dk&%$ocx7p;nh*81
zOP4L%s<RHtvacr6rT;-!8&XVyUDy5j=L8CMnW9$gk-2Wa2)x}$inh(7^UBvvZET+M
za0)hDhxcCQ=T=u&5;+ty4imY{8yn3fDe9`K_gx&<;I!HNvMlt^jtQi=MR_AxG41h3
z7acxQK$a28EfE#iD&Kp*Pu_VD@;+;s=8kHf+hZ11&?cK@(r&Ia&b%NX^Xc}TCAE-k
zt=8%5pbgG2J)Z9imMqy=TvMISL4~7%?=mtJP_qQS1rHAol3r?me|8-{CW33IRE1Je
z?)$5LKOsuGV|3{$($S#Zn7{8y{<Bv`DVdnmA%zX0!kIigmbIVml$Y=C8Am*}tsNDq
z-w)21IWz8ShxwPMKhPV+@kKgfCk1gcW)fG-P453;Zs<x>IMFu!rieP-T+1xt8voo$
zOhCWx$Q{^rtw}T#NZ+5eOO?3et99J~vI9cr5TGUeha>TF7c6tZ6Y`0+{?B0%)sJ*v
zUAnvATBis0w8?FXNIeQ(N8H{L^zCUMpef%s6TdwZY|-Dg?V`h@gPS%@jGs>YsliR&
zt~zsW+~?$;+{gk?;!kx&@Gd<BWUl!5XsWB<EqQ$RE>KdvsKJ5o$$1YC$U(AW_c1lK
zYnMyJXJw_QBa5yAIhPL|8^0u8;##kGc02s%E!(yw0)v|9#U&~t@=9p#T1QQ*(1Le$
zbte8lP&?~<K|9i~#l^*Kd>v-#$%u+>af~b|C_q&*#I~dxY(IX%NGxXV-_TPy=!lYT
z<-!YGrp}dBJ-{P&dzD$ij=tTm6k5k)p0WU)5a)qa5}OYeMu&QOW+A|6C6w|f@@`}F
z6kG-ze+bvs)*=pqxZr7RHQI!!D6XvtfD8O*{&d=Y1EKAHT{rjR@(FS~7P&9oeb=f!
zZ=0iG3>VODNahG)SnoTCIfNem!;`9zb0_C8QSZIz$c67Jac0MkwIFZ|<Ztl6b%c6)
zdwU0w+O^XgW!3ADHAi)7+v#;3A*-2V?&awTFSziv5S*PKXAiK?NwdH!SISSfppN~-
zDN`J4?|4&z0P>pdgh11451#hBPwY@Cr(%Ndpw7$G#Hk=iK@lP_l~$@-4N<@jsjO<P
zj`=W>)}d$5o&`fH=p*&rZoZgU#wYW^_DGk9t9wz3Uua+sAD_vUu&s6u4k*}Nc%5rK
z?+p9DbiQCkFIwrSspZ1*U`Dq}#>N;4>gkop#vg&ju#MmJSTV-XQ&3TSe`sbzYHI4o
zSu^5du&=nTwRW%%D-|d;XuV(W-5M0ovGDg5Uf^Rkh+;a@&z~j_^qBff+?l4Z#=1^*
z%e^P<?rG%srDZ87F#j-5Cd^0c{d!bJulT&qG#X(z1d}u6pP~AfX<i;w?aiAv<Ncqz
zuYYx0M<?s$naLO1zQCoCwQVpnq?}E+>uTC3Tuq0krHDL>p5$QcaDt6rWgh+zbkg}f
z5mc<Y=g?RwZ`W4y@y$j4y>YZ?MP=Sp+V+~mpy7*K_EE^YL~8ieK}S^f10w78>pAvQ
zzd%9deEyEO)pDdIJ$v>H8ysA>SXj96OKcYYt7u6mE!FSu`-CDcKWrx=(&U4@BnX^=
zfx&tcY-q1!tlHn;q_Nv}`9#6hCRd-kf>v{CqY>QOlU|TJ^Za4m1iPbKw_bR6ZsNp=
zSiNhpv9YO@AqZf5nG#iK^7LSO!Zh9-QA1+trz-g_9Au=eYC{nG%<0o-9T(YTAUaNU
z={stjvUZ@$U|MDLU<m@FC|0*YS=r6a4NH}Tv;dR$h&#n?rlKq*c2#kGI*|!2n>95H
zkOTU90<!)56<yXYSyF@DrE0RyLyZot8(jX`gt(np*BZ3wSO4FZ4fa<GDJ_3Evv$K%
zqJLBFmeOSk-xrIDK0wt86Ya2eN4HHMc)aRPoH&t|w%N;Q0e{&1oQiU_{?Pc}Xt+de
zWj;K@Ls8iv{=Fj|zySXB=PL7AyuA0Fl|H5?TyzQ)HYg56wy@+OBJr;+TgCVL(P_IN
zJ;t9_z-}kXsD5|QNTmhu&$V=k3D^Po=tc_75?^~ey8=JfV0l`CHmY3b3zW0nzQ#1?
z{zsgISmimk)$Fv6xpCu$b(J@A`*J=NB0u4F)m?t6BbqYSfs*@9@l6o(4~-7!%BwT#
zg>$&7IN2q+XFO%DY(=^}FCU+l=@VdFeZE_55#AaS5|aBsS7$MWd(Mi#lABg7UAojn
z!Pl)PK8@zlz+k11_E9(uiV3Lu)YaSj{yNG!e$cZdG_10kqho@zvwdCYJ=E71d-yOs
zeCy%9{?3xUE3E^uL%jKPjd`(I!rPp^gU~YS)xMJ*9UTw!BDE*8LO%3ci#{ATQw0g3
zit6eIa)chOsy?t<O--$(X9K-r{*6vEs;Ww=;ei7+(QO?a9Vmo?f_Y$$zhVjcd|)hQ
zNNefoi9A+(?mKKyC-L38cQZxRb8a?nwW`b66w%$){H>?kkbY`>kLm8flIh3+t<EXR
z%2HA2HE<RSJRAV*ZJgG$OZ8@fK0fz#zniq28vC|VssG-+fhwuw%VNeEHw`O06)dYR
zY^k6_enV9@J?=j2lUJ=I+_na+R9YjC9wM9NA`+04DC@oXX8L_ZpCgi=q8N4Bn`zUg
z_3Z8h=5V9ay+7>ip6MigM7~vS0}53<H44SSf|diX$G4VJMCPrufJsz;s-5hdcunqO
zMf7(CUoPZ1gg4wJLh0CjxOVpJ2^^Yp(4EeEdpD)uQ1O?YDIg#ZUx`Q>uDfZle}7x+
zdzj)j4Xn7!=YqbXwz?c0rxO=GAS?XqSM7K!ZwJMo0y|N?Ns@L=`diL@If^a`OE=v!
zMEbko7BAa<K3yos`u&)FQ%mK0A*s$oduzUZ<KgA?xp~v1Q@-YO@dV6Uz@ICl#4q9-
z((f?I$+N`;WR4wx(Eu8MxxYqO?hP(%tHTECCrshhH>9r8s9(70aFkQ8O<X3__nx>M
z>`*IreS7O^yPcY&Xd5atqkq}A^Y%NCHXpKG0g&pINl8h;LhU?>?e-rgGE4$gs+Pm2
zGv~Uxh_t|x+`YB+=``7_M~}oPH~r>6exGvS*Rf+OCnY8%gllhh>OYyf%i}!_cJuLy
zTL^g5`&)v3pEKw4-EI8+k5P{(*H6asm9C@@Cqa!DZ3gEdizQ${ARg?_BeWGevt-#p
z+hf!j&5LAZwVWFNz*&Xmv13zgG!Wy^zUArZd2LImxj@aA*huUnon4A4_B)UwT7p`s
zv9Y$;=30s5WoKviO%<}e#(dWwiPdzRTKT5m5p<}z1qIR_IM`Q>+I0I33~~=W=}9s-
zuA(CYhAL;XwNDc1rqHe})!3`AEYQJ*?~*B6-gsTxlHs#_os+`zy;Vv(Cs-dTMmFWQ
z3i&0(@#J5BUEr76?iB4_M{((*b@#4HrD4V9Q8vTH-obDCy<lg>LjNsWbK^QeNdN0x
zF_QO@Jmda61F+o{pL7<ic&B!*<Y474q;@80r@<hIwd&)Pp#E!uvY=CRQcIA?;>Ct2
z=>*Ha;@oM(dFR{zju%uYZ9x+b@x8=p7K$$Ph??hx-!K!LJNHMl{J|C0F0C0-mC7h%
zy3wB!@o^`tF$!A5Jw<6WttVGyQ}wX>_)<ZKj~Qdv7MQW%`S!$vPpHt-NzZz!dXUVB
zknAtN{6P)B=F^7?T|1u2Ta=y{xGABR6dv)pBJp@;It{Ugx_bx+ZE*OGqcpDvg^vxl
z`2WCa<>X%$RCNA&xJ4iFkmv3hQ0d@56N9&v^uE46A0HnGrHY^QQ5<LVTzuWIBQ|YO
zsTcXW907T|9oxjQW$H22eSF~D^C~Eqo;2EE|6tbmnQP1n^esIv@-F`~q8hI5thx6~
zoN=^Lm%VJ;_TV@x>WrNGzLzcqJqbXG1ZiOkf-t%RvnUiw{~Muu(#T3ju$owyn{v=E
zg`fJ$_`{;3S9x(cJ0rom-G^)aq~q<m8iq8)L2HvGw=1iw=fUq-rV(S5!RP-b*gL1O
zroP_v(~I0zeGI+w9x|L_qOZV72%SxlTSYm)Raj!{;ex0#7({D6L9o-2UD)9^hTUhp
z_n8ZW9<pNs<n15dxemVv@y3MI6xl!5t*c*7nY3W`{n!Dbvfdion+{?12nuM&9?^N2
z@mIyp!28)Kw{UrJ(K1SAprW2r<zLv8cJr-Rv4U`2c2T$!U3Pre?sf+TR=>9L1OlNr
zZMrOm%&K)%Vh3UWVR}1#*-Hepw^5hEG=6JrEV;djr?t8p>zUx(8q@@!c&P7;>C^Y`
zy{D$C8f}`tT_+dz6$hhtbk>RXNP+{gBUvC4cDUpFXM-dWjVv16i`{|K7uxhg&mm&>
z;K2hNn!dh?+9)4Ur{YI8%&wd;<;|NnVS95SUhw^|U%h(2h)SgnAmKT6O>jN(keMr@
zb>-T%#nl}xFSjhG5z$2=7y9hyn4T9mptR4MEg6)7!ga`^mA?;*{A)J9x%11+%|ySS
zip$DdS46&>hf>-r;v{!Xc0f(TFl&W~0IoBL3;4brgs0#h88UDi@PEMSw;xZcfmuwC
zt;3l}<EK0CQT-!yRv2mp(9naZ^%|4sRo8FWfar7E7@hqA@`h<wu@i;ZoWJoj!RLcx
z=68raaX2Eg{`3sr_y%l(>WVC{pPho!6N5vGk6hw{3*FmR8&Wb6QRrnNBG@zguPl*T
zju3Fnx@Ctx2EPVnOm)UGg_s8?dGKvk%cKMoR(*pf)juFWoFCyJHD{Bx^)}Ab8Gos(
z#mY!a7gQDV96AEUB$3ec=R>7+w_n5Z)|g-5ub>YO96R?N!3{a9njNOXW?!G-q#wh?
zDG_{;Fp82|%%N-}^b12+`1&ki%CF#w7?CAQq^;|6U<Uj8`>~k(vu7LEhR*FAds)r~
z)s+yBz~N=MR*cAk1?8w@(z#FZY}6Cr2~ZS3Tr+Cks#R^ecC`sR5JbOv6<gEAlhnQk
zVZ-kJeC8LU3_V4>;@|pDa|H++Ve@+7I%13ZV)6aPtI?BN<`K6rR^gXlI>X_|3&aO+
z`A^)06R+p^Z$iIdbMCap(K^a9TDzL>H<=_aKS|>MYe!k-WBcscEo7HVu95k>4cvdw
zmlLFW88i@DeQ073$Grac<Ns3Ne<|?46aWQEClv0Ja~(~?NlX4~dS<T})-!vB%!=i!
z|9;PGJJt4+T`s=`)hx)(jVWsQmTkTF(R0NwM-o-9Z+W)(*Iz|MR`abNe_rRqufY>$
zs4m#?@ch~HDi6iNXKT;?eCN~no#Ph0pk4J<id(Yr(r>Hz=D$-om)vR*U>wjrVfwFe
z={HjA`)98Rydj|2Dr21Kp6=J(vG-rnF&i&k`ll21BmtC5-h>95fe*OR<dA{yiYYEZ
zIsC5GeJFf%!%NE?+sr(RA^^Plm;kS$!~z1~wX+_Zm)DVHrHUDwiU2*N7=`3b=1yof
zm0loWLna`TP#y;2g$d$8+5mxDVt4-<!rgFGG-tq?hLvY$UdBS(u+Um;v<>y|0)9p!
zf&lhhLbck;>aEHE3jK2kAQtYQb3R0yJB*n*FUC6=gh*qzh+~rnz#<k}iG>Usx{U!%
zB>_t$AxRW2`vQE$Nzi|?Ir-n~{PT}HUF%z@PP93R>XC}|7AO|@BX?R2+HN?YFvGAG
zj_0}R+nSDjHuPS7UKb_PEUVJX*kC*?_!4S(2p~KB>tl{LCVG=8+2lM?aloSoMPB*#
zP>NCQaTqG@HLd*{oq?|1dJqpC`CTI&59@T{(SpA}VlV(~v0!Kfr?Qb^44m33e)f~t
z>%?A<5nKtR8|z{gTQTbuu)uA2PYuzujtwzmuO6HESvWJ+s8~-V@(~zg8yYQ}BZkJz
zd>?r437t6@a*)ishvW&DkS`|_9moN5KARW_>z!<79qOUY8Yg4HIG@EFBzycUqTm|#
zP}rno%#QUa2H5VO!wO>Ym}^N6AJbxvogpS91wWXf?h>mn0~3viZ-#UWYn(%?NUp-L
zzYiznunyXcsWQ|a!A#5{k+4egE7pTpD~fAaNJEbG9(7G*E3hhzZ<rHhzlS+~&i9a$
z=3K>*3S$`Zk@19sYlh^5#h&C`hm;YEpV{MHLRH<>W;m{=ltx)lPe&VN#~UJNvbxaY
zrvHvfJPu8}icGL$th${y`Rd`R`8xFQH<4vH#J%bDd~XVUT8deY13fs#wf@Izam{q2
zkcfPXzjVKAZfZwvnbH9T>0=ceRW1Q#e*0?pvR|EWXnDE4^z5{SX}j*!$bVb9C&%W&
zjDQ2J`cwC&JVs+d(`N%xn?i3g8U{K#qwq^5(M=GAUs}3R_ysKmhw=Uher0JI)`}1W
zVgo^0%)(|LhAJTWcZmF~B_!96(YTzsV5H_4(lU(aF_=M?=nIEWBSf4DlVve6hYSs|
z04Hy9-p&cU9mazkqQIObdt3~eC&&FW=d$ILlQs<Yvt+e)XS|=F1W86Fr?S~(h=bTm
zz_K}HqAhDyKe6S1xLRN|Yh|w^Ysr3A10(eWOPJV-hCLHfn3-%qQj~*rhBPzCNK%-^
zF-$HPdJyx1jERH>W~~NG+=tkOv5x<cVt;!D@C4?Hk!p!p!-x9_W@OEkG0-8%Gkfa9
z1>`^Gx-(~Di9+R!kDTnwpa=(l|Nk(A^C-jm{E~xX&W}Sc{_PbQqWMU2WZq%vb#h*e
zbz;(KgpMCtLQ;&`Jz+OvNGv)rm}ls97~#-NIM#tlRTf(?T~2~?Sj)v^s-e{<bvJP%
z%W*S{I*b@NB6jGc#%hQ{K5>V9fdy(JXaItM>=>fANSHDZ#jQ6@RuR40$%(Pl*Cnon
zaNla7hYP6<K2%49Rtwcb{QLxdx-LW;DPO=Q?Dvqn%Z1$qnGZZPmlplUv*kU3sPVMb
z&ZWxZ(96NWL3)LRgt<W_vMxAyjGPQFohYfD>4($RJ!PE^YRAzHUV7QWg$wP*G3#-#
zNroOF4^Z?3Jq1=8y+Cl<(AZexEe+`olia3K<qO)^vvm_=I2H8U=#PjaPXpd5i@I<y
zy%pseQCkdg)|dN;DrrNugOp3jT6x}^;tC4=C?8Fr1e?5<GkWiw@8#xOwM1ZzZG$#R
z4$cvwN6fglo`Qldj24N{z7ZKej24LlMnFjegGn)ahSV0=N7mHY+ouZp+1YZ3WW*^G
zoTt1!U}SUz>-4?g5x^Lp$+Nbhrh19pw~JcInNvt_k<C%`=MQ891+`Jf_IuCDN9O+0
zk5QyW>XCo5Pa=@3A&ka?y+Xe}M4{owRDrdQT~eXDo{1pSu=k=#yu$I0SB(x>xRt`z
zQmL!8Q|Kp~BjOwCX5Z*R|0e)D)=1l4%;B>MI@`kMp^5_dlXRCXd@NdY^ciXn>$^R6
zt+$%W<s!)+qZ@T_p`Ae0v!1rn$r1^_HzuJH92ZyYCO3XSVK;2p@cr&ur=D@fzG$=f
zFf1Gyb31(k5z~F=U;!A&tf;FifZz*9YlVx!8aT?7m)rcDz<(e`UvAphTh4vo1O?^o
zKJYAB8Yj;?Z;4P6>L>%B|ElCU!sYNuUZE4?ueTv|SIXa~rXHG+isJ;4zV}fx`tNS{
zFI*A^uu~f2<}LsMP$)DCnRvu;7?}L>Sa6)d?0H0fBbs4ALoS}HK@_t;prVH&>IzDm
zBZO1}1U}C-LHpx?d7DyE$SxXXJ`0p5>p9fkG4KeKwKnk<i67{xr>qYGDq?*eQHIBD
z*BuKr02isd{I^AJ^628Dt~$7*FR~)Z`-4cr#|G@qqg?sl@om2`=x-cgFXTK`*v3s3
zXRqI+jBt{k=eyc3r+WIduZz~BTrof>6u~+)Vhp=>*wSfd{_Jb@?rZ$&-HEQ0j}m(Y
z2g(GeAHN^k^gfL!icR!(Tou=Yj&c0Su7mZ-1EMG8atu5bW+g924@13WGK^-B`Lnle
zqEuB?xl(cgxM#{WOT7`lnn_a~(RpgW#+=d0vg9-o2j?E$MW^TsPgAUktSYv}V1Lkm
ze4HpG1ZL-ci&Fu$<{hzLkG((YRLNN|hp9tDzSB$Rcsj+F0!}G%+z(k;v8MUr|DaaQ
z?$Z9?&!0b+`5@^m#98zxY8OX3w?D8*KO9>$t529^y4V9jZ=n-h$hw%@JF^V%Ih3mb
z;P;UV_XMIU@KaVJNJ3|mww!)iPsIoo6PGvZ9RCpmXiFMSc4mufcOk`L0C|J%MQx}9
zozLn7$?&I03*cixX(&ln9PmD&oCka+_~@QZEVI!h$MerR4wj}}6D69)PMZEb?;+AR
z2)NBRt&(9?ni+~{p&HPpfR<UT!od?7%I+&-l&c}M4&xu=B7SBgNVXm)zhyd&EQ3aj
zfTf#>uh=?-a0vc6tMs?pK#t~_F-FYcJTMl4Jy6p24Vx-SXx3$<+U8{3;kU5VEV*4H
zOal8K)1H#c!u*K{g&2jKS5p7*xE?g68M>IY#+v#5$I_Nw49-XM+ZLS5j%0fTG8%x!
zF`bbL1)9atO@L%BDC?c7^#zX>j^t6&=CH@c`0$!$T}>ydu|}M>LE|dM)^=z_JLEM7
zsR>8|pB}$lu{%|TwVv#m3?~X9)8zJmMNJO&8G0)7#hgUBE<g-+_yC7EhpjqTW{@=G
zBRGwG!SJuhm1m8H@qAoB@F(k0jHUkP_6J0`j_FO4Zy$R8kUuqA#E}^Bu+Yu0fsV$6
zoYWl>ydlz#&Ts#oRbtF!IOnrT#%3YL>sYg844%D&LnARcNYzfTYeu@Y!n|LRRBQ3B
zEWob(9MZ0T5=qT__kzSwbbCgkcv4sS*{v%?+3zHtw|NsqU31N4lhiAVNgcx>hj}|G
z3T-<MD0HEc8<~AOv}={uRHVwUycJ{i`3k9zSa+j*=xMtqqJ!M@XRd<w71NmPJB+kL
zYdVY%hqLR@bcfiQp$13Lo^t_*uQrRQ|8az8uO9oQL-S(21z!d2dge6kAdx4mITzi2
zjq(lzq!>G*iAk^Om+?>dlK0M)ws_%}vwpQ+3ejY4VLiRnFrV+KtJee6|CN1A^;2|7
zYFD>4l{Or(6<OxH(cS{kjgR(k?mUkO3Od2{;2Qy~+atDX+j(Kl9C=M`?e7zJ9a{~o
zH5@aBzccsAL>{5P@@Pbgz_?K7!nr4bSa?fR0ESbytxCr6O(!6}dt-{w*Mlc{FX~<7
zV>o4P6P`G|^ppgk91lVf;fa1<O5f>E&Hr*TSjbd*mzuP<$5pQC3}55aJt=R)Umg!w
zSV<?kf+n<x#IzEzoU4o&&(IG>w1=qv3wa_Jouy&eYJ_Rg7%R)5D){{aAA&EeUxw~v
zP@m1d3`u7h6gK;j3p13X%+`SfJF%9Jcm{*nn5J#mMQm;(uO=TzG8S{O*)tiM4EaPn
zhy5rP^NrA<>;VsrhAjh39m-;2@@bqQN!FByi@^|~*jkH&7l)1f-&h8s`z`TY^<h-T
zgZ_h8hQr{o2~4&XB<R3gK2Ez&QhV5x+3Uw9GE-Tw7Wm(M6vq;9@C0!S>mTQ8j=BmT
zt}u#a4!e%Q9K)3si*%zZiNCurB*9pZBH8wD%7^48z==Mft?j9MjB;skP4bC$T-<YI
zpS`IRXuzIPwoLhK$t^p{H0oF2h=}qyUe(X@__5)Hf?)EK<gWG~Upj@|jsuYB*_njQ
zuS3ZGl`aFuofKnz9ioTRhRYYYupvNGtktJX@1jdZ_B|Cw13uV<nj8SgEZz*XDh{^V
zHTwGc^7}SX)^?pB6@&Fo3<>k}j0Gn9FU!wBJu%C|>6_SDyt_32(ZoQ7<7-Iast<lI
z9*-7AO?E`fi&fqkI3;COAzfNZ2Mo0zZ?QCcIa^GN>`O_1-l31`BYTCA47-ZDRhEMK
z>!~02tYJ#J=HNANZ)0QQSQ#z)PFcT4>ezRN>Y}bs5-OhGltQXx85-X)+key-(vrJg
zOH7Yn_j+*4Jc^Jz5kK)QFs?(v6m+*DR+{GvV0!fpiGoDF!HC|IV9*P9xo9Xzv2!*r
zf!Zx`8FU)rR`3Mz1Q-NtN!9GUjHj9wM%NNfE4oTJe~IF(dCa)5fgS`vX6HZ&Q6I(P
zO+zvu&0+c2_vDA~X+^l?dr^GnUa;vfCnB|!Uf7>Bapa##a<D}RqNsn&-U3!klrr1T
z>GE8H{KNj~r?A`&w&Ec^43_~?+J=cQA?5lsbkI7XgN`>A<9-Xs$oCZfi3(mHiT*-J
zV`u>QS4*q#oYRN~zr>3+RRaNA-sLrwExhChH|&n-6UylbXh?-I+F%Y10TUPU<|0bV
zpt~<P*<!#iAfOmMcXV?-Mp`O*MT<UCZT5T3O0c<rrZ$M^h|d{b#TFFXYEK!cQC2+P
zuN8GtPYB&c*@}t$5LISpf#a4(`R)jLjYi3_&80XGo$dQb|5x<f{pt3Tc7$DjG>51l
z`zqE3rMHncr1QjR+v>O8*RCN^-Ko2!5+|uzXwV5SW90o9ge1}E@X;LiE;lX!6LUdT
zh&!4)*ZI&8S&=+k7>#t5*o&g5mLUBB<pD;ka12;va&J+i>&i9Y-zDbNh&2A0H#eIG
z^a2v$4~rgpabgThCk$H-vr;rUT5({bv;jm*)}r)}aj{Y1;b!QJviyVE4?5C%I_mQ!
z1a1TbtVMnwaZ-({uvw|8wMbp;>O{<7MAJZ$Q>T8VVQ;<mM|AefP5g?^jyY>2?Cg<E
zX4fat71mJ-UB-`If!^NUvt6<Q76;b$<zCvIpK2e<7rap5ZM&z~ztk@pjfrc~;5^b@
zrNI9WFdvc>)6>(#n_RO*<TZEgn$A-K?PgtZ0y5W>2Q7-w7*mOG7!{053E6r_kmf6G
zUF~PN{m799bQw!H_%Rs$5Y24c!-{?&Npp&BrNM;nm?8QrEmz<*vy?{7q6er^cHe;L
zU}lkyyglfG5o^i+m<#yvGJZ-Yx~bpaIW?@PWD3BZGOqMLP6a}(y*kI^SxZNoQdffj
z`>GphZxWE1TjqnzG7X~*!xqJ=3n3!2->7C#jUk2iZ4`Lp=cOaZX_~XSemd=%4%g~u
z8sOQK<y)>V3;j-aKYxDzr)8s8@!DuFd8rh6Bgn}#aIxP?Ir91O;t7e+#A?l5MAaNO
zG3{qR&BQ*6?jU(d;|X3JYXT)YbD+wHC_24InrO8LgAi6uH-BDx*P;TOY1O%Y*VVyf
zb4UdlQcB>^v!-h_rOiwdCwB9MHuQQ&OZ~K5QLzgRD_q9{TCdx_4yD|e2*|)SuIu}%
zu9gQ|xFlVLD3GcCmWCc&Po$RCPV&$o{PYdC^|f^E%FdjY@P(Z?=pZR7`hNTWVed`g
za$fhfamy51Bq5ZiwM<1bQZ&hwDMcvKpioo_X|9K|P#RWch@uoyX^<w3nk1B>NofvA
z^E~%n_wzK@-e;ZjKWFdrf8Vt~>tj>T^BeBrdw+-PdtFPZQt=Zy)k0iMah>1X$}VXK
zzI;-B%~siPmEtR`g?UV?f*1fE{>0@hgiTc~%ZJd37%D$y(%{C+4h|{dgS~R-3sx05
zj{|J*J4E=0%MZk5jsqs>scCa5nB10hHzoLStbw{G`$l@s--f{@&l>4Dhi42WCf}Em
zNrgf6tWLK^tIu_d=W01jP(2LRuLLQE#e0uvv>W;!TF4${w8Ik+?k1>+<fAd{-unxY
zYlu0M4^u{}swc2LUAV;6MUiyulffa1fBnf9p*xbS_a0sNck@rOeEyks45>BF9!bJ7
z(lD2qUGRWLy(M_KabkgAq_8X^_XYtX`8HBPZx;E+v;mZLHxlP>FFArNdcU^rO?^V2
z&dHt$YwO2x20G+rZ_o<+{BU@B>vqf^O$aZL|6U@+9PMQ+k%F?~tp|HvLxzhjkl*wm
z6HN#&Dy}LbkpCz@#lRx#%)*3v4lzNZZ<rF7LKxz0uSH>ElAMBK+s1It(vGiH(L~)&
zy+m5|;K73_TTh-kbv5bC&!j%g!^oo#M?|JIVE75XmLjJ>!F@}7A7n!9?4>P#(p2Wh
z9)zVBy`KI<-laP&bB}0Z>epL3$AjBV5@-Ae4!$xNKmoPQBBwy>H7H+))Mi-NC{Z<X
zF0M6gh$J%Bz|dc~Z~^jOi!&~eO#GHzGuzelSIu>psiyeZRjtCYQv{DZ{9jcA&9WVu
z!DZ-Rk#k64d$Kj!{hzm6VPuwWR6`dIdR@z4xy(md0lLd8c3A=)$5=1o?zIJ7<yV)O
z3TR<%mIartIF3%2Mqw;v!h=^9V9mf3sJs;3*(i@<cB-}1FPjoS$CX+!E-NP|GP=30
zF1KFSf9V=IsUr2u0M!2spQ@qWS`r%2e~h(jFmtkzSXHlA#RSGX$K9j++8%_HEF$0$
zPRxKKFP`L<h!60D(VriO_y#T}iirMiyu7^b4aC=CZd8jr4hlo#8dj_0wFP15{<p8a
zIy^)}!)2U!>((tIzfyLEA=|bTt*;9SUkD-HiZ#86c`i~`#sN6&$Y5pY$WgwaJw_+d
zy0bt0Ap~+q`4bz~!<#L&em*nk+8(=6YlH4Hsm({dLL0L1I*G8Uw3!;Y+SL3^aUenh
zP$|y5lXPu>04-k!1a;Mg(~-v`#*KasAQ*DBqx`C}S2kdWkNf%ED1j|G^l{*%81=k+
z`Iy3i_1~Xf4R6rIWaFsFlzb9%1}8H;BcmA3K*VfO3@gTz{_L3C;LFCrD3}nD!G;>@
zgKAc}(G>-J_DL`U`0H%_l?6H=8$1caTeSdLbNaKATdXb}zU1(r48P6r**5&>WV(if
zouO}$C;Hu(cnw&|c;E+6!-e8#7Fu)d?Cd}RbPizx{33!Xv-;)N!Hdx26FjObgSCk$
zYE%AwaW>SRR4TfZ^l=Q##0^HRWJD0C-GNK)P`}a)sJq+!_~er`aJ~#rK{mx5S$%j^
zY)h%AWzj}v<lqVH5|0GQAf#)TTeexTI+1?0gws9Z?O&R|lucIcL32z@Pt!ZNFBHkO
zyu7uw5hZlE$%qLMSS+58psB+biI_t4VRN&^;6w$V;s)(po(+BI0G^Zag)EHVd%Q<N
zPN1XNNcstSW@OQEPI5QxrT8q^g_<(Utc2MANShg6i7rzTkR{A-E=B7!2fPd}_7k8(
zod^6L%9Q(Ey#zI0lZKI5-;W}Ai${O^nZA2$aqp*5pO$T%#<ou5HPr4C?zMv@>-8#L
zxnjhK{uhF<jP9G&y?ib@$j7O8AFikYG*{N!s-PM3WTLon$^wOL`lt4Z)`nhjp)C;Y
zENcJ_ZTdB9civKDI^1#cqyfZ<UjS&x620}%tMdl4_9bG#wH0whX}D~FMnrXq;G{Q*
zY1oS!tr+<Tn`lDyd@&(=uLAT4u|!^##0Gqm@g5Cd?}=SpU6FtR*wLotqR3Qi(O%?q
z|86O}&~wpja4)c@-<_#4AA_e&r0CV}IQ=LH``K75-siE~>cU21V8l-KH*)kde^w|=
zj{98`=}~-L?Hbv$2{#*w$U?HUv@VV>a#b<O=c1&_>a}qlWZ?S|9VohNIQwOh+jDg7
z)=lNx#?oIy;@Q$h(_oV={tXVkJ53rMx$N6rQ%PJGwXF?m0()%4vmZTvj54Itra97Y
zt)ZuBlM#Y6__O0^FM7V17>34T^lFN={Y)vnNE<@Och)7H8AkI&*DWYsGJiFbM8m5o
zHuv32w_PWjP`=o@zeTqHX8zsNFgXBiS@v{HJ^LQ62@Y!*)s=>yksN>?XmI$fQ>!;<
zEZ}QXN#}#CZbKcqFo&!Y{eU2UCY)$2MH*3Ko#ABF<&o&e!s@(ggwE;7qr*bl8b)Aj
zjkGg!1{zc=9NOd1>}xO60osEhQ#QQ$1#HBOL2+<(-NI+hP_d|p(mQ<kuudXcQAY<_
z#55q{t4shVSIVl|t88bV_ANyZeP_qxs_sF%L$^M(KV9BgqgV9nVoi5uw}?`?xjiyD
zMjJs2K_Y|hwOyWIL>MWDI`~a|O$v6^)*}j>pAKM1!fpaT%z_=$eG<?xh)sN<B2`sC
zVH$kYiH7M>vl;-eJu@#nLmOi6yZD|x!QY}U{xk$-=mXPgUujV7YDsogdF!zm(%)$0
zpS!9P8X`?W^B}w_V=^z_(a@^G=^W$6SFH&LG^(NKPt!pF3;9g|B2X8u0}JQ$;ej1`
zqay3$XjPr%F#S8U>Gv;Q<o?@z3|qpx-7taEhivT7U!FKKXb@Ua*x5NMN@&yTfD?nH
z1}BkJUfnxn>4v`$KAhgAO%i#f{7K@9pWCJ9FI)(Qu6A(N5UV`qKiAz9_!$+TIW~sA
z<+Ns#>|fx-&iK}bia&VO9>72n6FufX_;6sTL7%Gni*`uPQZD90S)Ibh5|jA_`aRN?
zn&VwGTB-&k#^(M!^+Pc#&W71eApG#tPsu`r7Zl$|LjjK&I%FUkIiDU!I3iw6+|<VM
z8~pqB?76}5B}i3r0yhe&B};%I8%v}(*zXY4($B3Y5<$XR2Vtsw<;j%ki|Fv4;XOsR
zRb^8t6z-R}g#r}w;>^s~tT7)`RFL;MoiuK^5a4{JfvU^wl+UNFM1-1pu>@{X>MGDu
zj8p>}n)wuK(A>%Nd@_nLDv7$NT~h~ZGIkO$C$2n{V@N0Hybjm-e15W{{_Dk>SB$7w
zu@i0ABVr=p^vk!iK(8%l!wgnsgZ;+sI61~vRf%m2F@{dBq+L3c&zl|E>j?28NSigJ
z1tuE4g-oxhmeAX!h!#gf>}9QCJ0K_HR~xyi0GHwQz62&q{g}KNu?TUKx*wpz3x|E8
zm)k~Rha9oa$H1s+zv_}kD1&MAblTK&jG6~b1rj*`^00j-W@?@#gzXQNeEk^|k9hkU
zIjW0S8mDZ${DCrNHIjd-C7R*m6+V2%)8H_CK->l{wFo>tU8%S9RkV7FEm+dNRaS=T
zd|qqW0}i)1-Td+Iw#t`bN}Vzk*zMQ&0$B6>#&ybPP`AL`N?@sQnF)v-eBFOR+GR<0
zi@uP_m5z1mX^(xJ>B-~ALc6pY*@hO4;zi_QkpDCCt^gFP0z0HHJR#MZxxEx}^f_G4
z<lbs)Od7zgs5?JrXrBE)*)3jYHAUm1$`RE4VkG-JKS0xWR~eD>q)%|}#e_{m>v-=O
zrs=nC-$p;F8t9-U`x+f9N>&wMOJB&?=mXaX#23aO=|PU^(&h0}<6D`qyhJ(*xnhEW
zxi-~AKuD~kx1QR6`-&m-iD@Id+q40vsBeRlZ8Al7=rQ34D<)Wzr(=m0gjJ&u8?oU^
z36lhbp|dv58Ilzl^6mz=1ok++q%hhPO<>9YYtAeEvcu8A@Ips;H_nv(<)6^%wqw3N
zp(pK|?Jy%03>#Q3OHB2_Sw8bv5;=XJG1No+I=QYC)3Ewm0!uOL<mp#~Vvt(n?D>sE
z6=aX7J0?M{%kqS{YyMrFe0$+GfbxdR{qYOuY9A}B$+U|KJnz2}hBNq$serJs@N-dP
zVBRpQ@~^Ftp|L4?2nvf=Br%Ki3nyc0jqMW#YTFBsm4>HQ?RA}dgeg$*xfSzxfi*6Z
zcrGl-97>C2)?2ZvUN1aoSd~;g>(>O92G8~zhri+P0g2Yd{C=kIAnt~>Qf0dKmPsZ}
z=uJD%nn-}9{)7aWZzBFKf0sd7cE|WNHzIKR=Tq}<S~;Va<pj_FPd`BOP_LUcptQ5(
zE@F_MszW=Wsq~yl1HuN%-AwrWbVzW*idMBruXgMo3foYGIjX;w<?|iAfx+3kj0vbi
znoKzTbcB195QT_`wwiNexPxLw2$SKLgmP<ekA>AGAeXkv#veO;_)UysC_2uW*vcfD
zT}uH>u;2BS{bF&`^}`#4a)$x%v@gkL_x5oAXZFp%Y101ccK^Tl@rF0k_+#}y{K-U9
z{I46URym}8)Y({iI<v3bzI_|X9oKA%o$3YzVZLc~ndRi9Rhu$!_6(#d4rSwzBPCB0
z5~xFbL91>si9ufs34VD3WIg8wBlUH6IDa#4`_fOIy$hLC46?Q#nK_`Mh47W5r<4Ka
z6hSlXkjEg>he-`u(DuT`uKD&GzuqJ@0$AS5mjEcc{0n`Hj>udM@4A1}5*>47{o{ZB
z>!y2nOfVP!-NQ0F9B!V9yuFn-4S+Z1e1CEvVYil&{rU6fhdh~qDT8|=aBG$d5xpJ$
z3H|TE4=*Cq0Ma=*b9nD;-2XuQRQ|~%)8D7BTe}viMrGNn<1cf0J*u6V)(L{hwU52W
zpA2mKNUpt#oZ92v(k8W}(;!)S$N(t*_zD`5iA4OAJ1}w?UHnuS^&F*mOAECm@QHwZ
z-DTXvVA0aWPth{~_kj(*2fkAgui*tzQ4o6X2vF4={|cnIPM0?AK3${4=};pa-<b|p
zN54*2<X`R`IGOO!Ompf{urm*zBu#3&O$m8>1xo=qa?*)o(q3%FKTGq6-C>W|hD653
zI->M|%IT`5XW_$w6zREJSIdtV$Z%APeM3P_>Ch3R0K6OeRpjJejflxyohfz)X`vH@
zSviD4t@6rUJ9gMD+K);h^(RrYfocw$Tsq^QqgH{Mcza2IZwXn2Pb+G{f`lZt8pyw_
zQsj;vHO*$sNaowTg#9u&n?~O3jrIQKQb_N>IH8zy10-?m-CAWRjg5Dto?ev))@&=2
z1UL|LI_KO7S;IIzG7i4g(B@MsG7%6ladXPNw?X$UMno!)_0Y+$z?CDl%waKF8Z}v5
zMD5k3WC?llz^=PV`s2C(ggNdrp%^j^%ZO44L2qFyhHA&Cokll&t~Kc%l9D2VGaf3C
z0?ZqwvoF6!#+|pCV1;;pLCnqEeRe<y6}&nBSO`_#v!<qIt|=T)kTr`}r);6N`m6q-
zaW=X(yi{`hVmA?(c2C^QutwoZ!poD>=G7I)ingUC5cZqA+z+D@-L!Kp;R-8zw0e1T
zTY<JJf-}_5_>M(3Kxo0j(>Y%TG*ITe6Kb69HSTV&ckEwT>NJg$DWTwL5ywO+E#}h@
zu#G4-)t)xf@MkUvsE$MV_FO=}$OLDq-9Zq8g9V6m1%NL|K~YiBNRpQD8}pVlaQOri
zK%a*H=+A#sIP`zXA@KiU^v0Y2fAEtdHWO@_|3#7Q|65ZzWLp2Dzx=;7EE8rZCNN6>
z-Co#48*l<+I$>OLV&w#%{8t1hUSH~Ffbw|L*xUb#WcQ6ksI&T7g7R&wj9x-8NaO^p
zs(l6^=VL7UqHqE%AqgA;%O0~j)x^!vj{pBdqcOn}8uQ-&<e&eF#WsNp{l951{F@$c
zjEnLQ|2Y^*kDSE6XYGGgK6*qnF=SN!Ug7@=Te|2g=uE>AYM~{4`;CQY5w}FRH-SZT
zEkK`<8VZRh0_m|SoI3SQ!%;kZV$s{ZX|^|RK&hnQ@Bw`>h?Gp&j;n~+sx~i3b!R-1
z5m36Ap6UWd8F63}^G%ZCQLp8jSjYBYH6riXS*$gF$X_%H0|dMV<Nz_PXpWxXpia2y
z`{>b%RdEQ2e!lqyp-x!<81*}<{evdRjR*+M%9y&P4)u9Jx|UwttM9zn;qKa#vnE)c
zPj+m*HfMG~J8IbnH+%!+W=?7gK+kK}4adbpD_L#cAr+Nk;L$Qo3H((@asU4PHAlUn
z(KgXI*XRxbUIjxC8kPlvVs)achC~4SCTEiHur>4H)<bqNKc2WHI-yl`-UN@k$_((E
zvrl92hzjE8k$fq8ljg$7(R|zo&GC=l20}_<E3+2ByNy5lQ7!LR*>a=AHNHDKeoqZe
zTQ*#f^S;3Wb>SyRUYvSvzYE~)n%BwP%d1cU*jM`6{oL!tjmUUv8szE&IPUf0s&#WU
z$+qv>pPj3(?{M>#{meNrPRdC>_YIF4Y<L1Fm`#x@Sf!JpVtYZIsMvO-?W%cAt7d9s
zpp{|j*}Y;WR#v(%u4+t6EzW;EHMNplFTQPs`D~;+`Yn21^C*S$An)ROST7b3O~}%|
z{C4k*is7lmD>GBg^FhHL?iT`tiF6q?-qX{Q$k;wg3Q&AL*=fl^yeH!=BZ8e3Zvs}2
zQQCn`nD3;R;=zL_g6`mhiTK{NowNNER4s(szFHr0e4my!Q`&(;QsSi8qU?uT8ARiV
zDz1x1&!h0Tqx;f6qROeR%A?r&28FCYmZQPrhlWHASaLa@g^&tH$u>{0EYu(`kngI>
zaTdt-UKWC`2O9@#jScQ{v46(RyLT7s5P8`BNvXcKe#BP$`CHkaMlMNU&rFZIs(np&
z`rGvS+ipMxHdjecrsMTgmh(Whs-kj_mij%3c@0>E6!w*tU7tKuE~E?XS4=|kDIo-4
zB<=aR`XnV}ze`ruF_D0FzH3T$9|=D16@1nDBukCV9Tih@C-2~^g`uZeo|Pdpv|-kV
zgGkR=pZo+s{m~;yLgs%`#7(ENeS)|f*ZcQ?)N?#5!!VT?7V6pz$k>APZJM@7{=<23
zSFP<k+o}ESYO__1iZ1RqequF^g~v^;aoh3PB;%3r8tMS6XRTvlKVjvw+Bn22X<uTV
zzDO)eyu5sT#V|1QGd>Ci6hqf%-_sa1$pF(zAU=|u_~Sa%-yTyn0Nr+@+x(iz9{w4D
zb6Wwya<x5LuSe}x0;eH%_WNA{5g@zQj;xz?O2~3H(U$o3s7|HEYWdaAPg{p{Y>S{{
ztG3}xZ0ReoBrI=IzjnIyu!@Sx(-_zz=-5JUTKL*>RkK<ldZrw))_^CGrUK%sn(iuY
z_RKY8+UAgYJ{|c7@0adcB%BCBv5HVrv_KqO`E&53-GUol@3S|iU1Nm%v9}r(&beFk
z#$b6{FTkG}yc#WDSQSpt)S)VqlHR_qM0QPMhy9XaRzi9tC?K?khsQ3Kbv-|S{hMF#
zSo>sUnHZq%&`di`$-U_|a_?Qh+*r>C0Z$9y;f;8JJ7qv9KTB^Zpp5kF6TCXwMQING
zOn!Yi@WyOyxmc4gSXu^UDZ+(N=^~316K_v<DHE{8qj>({Y`LmW>sG*FQh4|PtPGGv
z^3jq`^g*OO1aTYHwxNUL6#?9I|DmQBZCDFgQtn6_`pJh+4b&+7RE~c`gYtc7W&pf6
zWm!&^EI<SJ<<M)1<qOdJnTCHt^Y@01grS&g=mUe7AAC!#*@H-A+AGtaap=>7I8ypR
z)6ysdAEUd!gH%^|q=98Je0|ynX%`q|>7hT((8VX@v82V5<l?lCk<~DKMdIuZw5E?D
zr=wjLk2v%QgWn%~guxFC=7EM+(7;$`@r(F{+G|6y<J-2h9PIFlyaveqrQ10B%Sv>v
zu$)a4e}p)92NG;VhmgC6hu*%M$H*ycIa@D3IKlmC*+HnltO*2NxGT#_1uQ}b)g;LH
z2;)1r9Bcg5Ubm~xfgt@)4Jt#lNj^l@plTA@0E9TprEC~hQf^`pxf(r_(E_m>1BZXo
ze?Fi5lm2sC`p0NPH~ja(ub`Q1;49-VwOa>SPub)!l-w?ORr*1C5073Dn&!^+a?P4G
zwyH_c;~{Sdw}=vd6@4h=<<V`!V>!D%|9O)8PDq&lwf|NxgCjAnqbl8I-huu5v%vmg
zeK8X967VEGPpcHsnf;FMw?nzum@vH2Kzx}ho^trOI|m}YBU`GP0cnCzxnt-4-j>eq
z9t5b0wip}LYk=fK!6p6OJO6SA^cuxHSz^y2S<L4d<$v{O@YNfcOLAvMpnc;(>Sqw3
zEWNhCiD-*GOH&K8TO#DQkLecfRW|6ieT4WNI)P5#KA6{A#zd{kf38vZHX5alcp#0v
zi`$)~DW_?br<%!r(Pu*DXh+TnIG*zeDenmJ3{)!0MV@gTJbHAKdh&Js`|O{rg^rL{
zDMRtU5c0bD{OUU#*gK&(6bsIr1m<WUA<Ih8H3`|EyWBa-M66vyIt66^eRW3hZlDb!
zXTHn>u}(0K+7_(`>@p-IWO;ot#9mk=n4U8}4c}CDQ88(|)@=ViapOvxh>GCq_|ltd
zXbL;T^RD11xp^(bW<KjLoz57NsMLhE%v(-Q%*9=qTbXdj{aB@-%(@Wyu@3&gtoW%!
z(S@V0;aYxH-z^0dU;TEpDomHoYsgibt@kV@F>y=urRt~mo+T~e+oG)~qyy8)=AQ)!
zIdkA+|G+4}L>qtd1mwJeB;cX`b=!4$*GZp#tdE8Fe0v8D@TD7nla!Q<3B%kIiKfN!
z$}IpNUVT@C^DD51w<SSpZQH(X)WJI3tXS+Co~Q{bM5gh1n*O|!cYHSML!oC~ry=wE
z{EzuOeg#MSbA$JJs@bQzZ(JwBA@VIKK>Os~jk~$S%y9B0EpWYc>tc39*)(w=ddplP
zJmxBxv;DbIc7J_;pWaE0S!iHk=s`Ed>k?AZ%a7hwJ9+m(&X|EY-HK|}4U4$$rwnIO
zkLM_%4aBIgzqZKi;GvU!sg3OHpLWOu`0n>VdFuAP4QtQ1Om|RZxq0Mn#Hsy9O-`@o
z+bkWec&b$Q?f#j26K=*tuTgiJw`|#xMU3~q>hD^);N|tJiLP_4=6|d|Qk(PQL5ak7
z9S6<&3!6=e&UNAo52ABz)^463apXni%~u~j@Cyh8XNhS&c`vft`y}J+**S0dA<eVj
z4T=K+5M<F{za>~2($sX`<e;QsjjK0Z>sDS~K1UVyX?M5Di4!ZLC*8AscTp<@LamX7
z&~ke1$Psi)PG3Jb4Dy*3_tvjp|AuXEs97B_%?p^Bv$QDd<xjCc&I)N?O{G!;dg35C
zrupP2niTEWo<XkWBINR*;Z;>tb?@F1>zOvD7i_!yiGZmjB(s>v&9vz^GrkmKZ3aQ)
zsCVl`<|H*=Q0P2(o7&%9Qt+<(1SQmBwr%Iy3NHKG0Z*Uq%mwk9uhbqBl^yR37Ynbq
zfojl^!0IWQ$xdz1sZd~g7#0?WKt_Gvc?gtM_z1sXe3iI*@IH}lPq|GVS3wvkAOK3l
zeC#^g=N*eO@P*jzj%Lrhw43lk29H~egTn)kYMSR`^U{9Qe#o8BEGfgCsOe#Gxc6lT
zzm<80oybQbkeZjL5F2fd&4g|iLl3=Ie3Semk6l|uWR|1T?~5w<yTjei&B-~U)SoSI
ztN|zW56h23Lx&a!S?(Cl2gZq!k7MSX0Eo<vO{^X<%Xt;%rAq_)&cH`#ayV}y<r`31
zijBS2@RvxN!8tkvImu_m2;trlA>EZrVyT&X@1DE4;~kNsjpjV2lO}o`O5kac2}#hk
zglcT%o*-y==ad90MeMt;*U{A#*YuX#DSqxnZjS~YJ4ly>hu>D7R9AX_={+Jxdv>Ml
z9NVaXprD8tsOcK7V!F3_ox?}1&#?V`{F4RGuIA5C3OO;eV!G@mYF`i5SY>(zxV@sL
zqjSV%9e@+Y-hO_&dv#GO+`03*j;C3}(&$Og$_K7qCod=W+l@9*ed$X>I<{f&;BrFq
zg-je%N+dyq28LmRR#!}S+-TPJOvLi-k!{dd)pbJ;kQyq-q?0&Q%c)3GRTyhiTL1!H
z&_`7={6#V2>!$Tbpy>UDQ%vkwaj&L%2}|nn_05C(qZ|BGUmq4nennqJ*zI!H50MG$
zAQ!*sOg7iz>1f;idR_t2h_@p`pGKU<5^|}@eV7L6@v>DovO<m_wk&@_1T*grUav6L
zEaI@Sdamx5wOm|pI*xq06h_o?88uVpZ^`yco8trR>DF-lvMq+ctY1(#@MtH{EztDg
zTsYI^G7-C<+=HWw_Ss7FZn$!w-!d+n7Vs7I?M1?oO3<-^rjxm8Ct6R^+gJ0Oqy1za
z*+seov$a+;gFQg))O**mxKR(s&Bu3}+RM$)FRZotTjkLg+1UqQ&#wEFnF-aY)$f|_
zYCb^Q=A)J3RG@F0ei+tF)>;!jU<I~9VKOFsorFZO1yRslXELzB4y&q4LTopnE#$?E
zr9#ohjb6`A6dGNcyl`P9Py;8Plj8`DFBv?FN=k)QU-y>BGarqBKYfz*l4#^%`tA^2
z!O8h<-I6<e%H?={@qyQG1S6eaPB9MprD0C>cP#+7UnCfr29K!3qFvcUVE&6g-reCB
zj=}+YFD5;+<HK1TBwj_*_c2XkF7(G%tV91z-mFyu0)JKr9NkP=e}rbHkUQAUExF&{
zq2LW;-|V)vuO7L&E?PW&d)nfd4#kUQiPJPGo9}se@ayfIFVBiK^zH<GU~ikDXTqI-
zkwa1>U(<k@sKvr*5_sU`bi8o0#byH)70UAIMqE5RH&3>qWoN_s^&;^)YHFV5FJvJ_
zr#mGJvE!%E4eHL`Ki!ol-4>9i4?)-&ZEc9yhFRIz+kX|YyY0k~esi((=}=>&LbBEn
z>OX9q>L8p#(7Fle-N40l{kfIf`+GurxL?2$g~Pgf309jghejIZ6f2LSr<*MtPo^e^
z1qpTDyFX7;dyqz7YdkkyT$XFMqIrz+6qb+j#(LE|JMZ;J!*8(iOSERoA6@DIrKp27
zXeLaiPCoohk$LW1&)#D_9#D6FJ*zGlP!Z|CEO@9nl-qpDYpAVXE_|>l0pX7A5(gWb
z#iEAKdU6m;f1Xw<ajH7=1oX>;|Ky?^3B-|3yWxX0>r-#Z4P#)!I&Ui~fv^ph!fxJ|
z6+ArePf%i@M%tCCskB_$VZN`6@}#pTL*~d3$#s7nXv5#OZ~XaDWKIRdt>}*(6BQK&
zcZvw!V#`AZn$5y$z1mfC{mbu6<St;?kiYqP)YciV6TXgL@C56nmG84@JNUTG)2Cb}
z-X5I9Vlp&&rV*oGavl8_=;+YHhC~lME5FK#&H5I;ety9B?Ag0_uZ+xsl=pm>em<u^
zlS9%aB{F*dLQxDp5wvB2Q4k^qPR`Ef!Fve_^4B!QTRTXy8k7dD;pPsCZY=@$e+rYB
zzO{Wzi@^rI?bGm#uU^4fW4QaU4<HI<Rc=X`YBqNL+UON4R!p5f{lJ3wuU<st$Tdos
zilzaB`O^_;ng+x!m_HxB?`|GZoqPd5E3p~)h`>d?3~WrFQ?Cen(z0dhaDU3Fko61n
z^#?l<cn;3={>ea-bekmKekxGfk#atL>(;I52IyZ<O$(%erOGey5eO+;`g4HnSVXkh
zE{Rpe3Be^<L=$<MWXWg}{P3YvtYT<X)IHi=C&m9E)`5=Mw4A=8dP~g^VY>8DGkRU%
z;>H1u0{5cg_UzfSvD(lGOhxmFo4dO^JUhrZb#<Ap^Wa+uUkEC-o-;)D#kB`)tdtU*
zqdpggcAvXQBLZv{ev{0lO2IE*nzo)fckVgY52PnR;Av|+tKA|Oj?lJW9zA(tJ9*(0
zZixc5pHRoW9tQm|xkHB*Tw97)Fm`oyn$w*FhAI9=7N>ZK*>Z*0Eab+JBuj2r3|7`h
z0s+~MUchg)y1lLtI)kZGbqd4N)z<R-${}&Ay9!y%Gm;x22flOX8>FiEqR&9Y4rwWP
zVEt&sbA?9k$y4B?-P?%#?p^lHPCQUT!))=yg>0G(vJ7_g@?bn_`NDul*PV71J^g%-
z0}BzG-amFtFTVG%(j#hrU-hgvfk-T#F#6>7#cEmN8^=XZM)dyKnaIfhrVU;bg{!{0
zy88O{8G=0I?5<-ZeQxSWHm2(od(gaai(9ALXy%dF^Lv(p1Oho3I4=s4PZ1D;brJL`
zu~9Xo@#6j63-5R%s*a*lVRP~>hdmge20OrP_V1HUIoJl_MrhRr>uy>^<XI15`yxKj
z)8B%OAKc2XRiB9Cr=5=W2I_Sx6w#^=+;RvCs)tS-nt7MyT^I9eYeR@|cX1SS<so>7
zw`Oy^R8+$<e|~mVCeEGDX+j&%S6)A-03X5gEc2dl&-{fUpV3E{U1e%JV^txM20mw`
zOGWz-LPNnmL$AretkqVXczSMsT6Io%Hd2l&*G47?0H1n>5rNMqwG_hk0#9>}40#uG
ziu*cEha$Ruu9ue=^m8>KR~!%?9-bu@3H>`wQ#Md=Y~7j^wy!8<nfk}TwijAi%hpvG
zg6#8`DbqzZPF}X|$&)7=JjAqe7V2b}=)=cD+xNL9#xqn00T%`rXJ^<_&7Rf8#l>E8
zQUbI64slo9CNLyB*=JMm2)Dl{KpxS|#3YT1%zXgDliYH<n5(?&*2&*bC#<LSG0ppS
z?0Ag^8GgJYm^%66d)BU4vCYUt1rqb;C^b3$ciN6@LBl5wjRq_*G(_5DfBXKO)9rNf
zm+pH{>eT1;z2y#iDjw6{*L`YwEjV%Jt}_hV`aA<C6G#l0H#RnuC4nE~VSIdiT~ij0
z_B0CCb?6V)=g*&4-+JygpV*S~7Rl{73-5wtEoyfNcWcNY#t5|+>DJh%P@!}7Xhndj
zTNFhkyVd`IA|BfCw=aup&JfMv*J^*O!r5F3QUmT6-_fV7uCBgq$%3RvWc3TrM$7cx
zB<}dHH{1HJF>w%aZb$^DrliDnRr{ld?2yvg{NT&pEEp=Wy9_It8Mdi)cHU8XpB#Gl
z5!+`=WcaIB>N`-{4jeo7#YdJy$Sq&I<RPP}`hi$f@S~zODX(*yoioEb+BnC!nlG^w
zM|bx7R*+e`vWH?w{r!94J>IyYRmih!*_K`_ue-WzA0%a8C-Ua&G<!Z(Z2jU<#PT*L
zCx=ApC)aUsaNKG3d{zvv_D2C2Z%K!2RPP=;a$a8KMFJ&<ZDY!*@+xA?KYW*RdcZD}
zk(TC^4lO)oIXSh=W^|q9oCPV!l5}fiij1<!uFm;l(AQ(a=~IV&FPG%$z3x^9OUY5?
zT2Ig4tY2e(s#(7vQT*wfls0X^p=K(~CnjTQ8MkRk_`|+?O=>zi&x&M^#F~bkWxZfv
zx@qIaBeAbt=_BUOVAfJ~y#3&vRnWH&XBRgOJ@)~|wR-6xI3ur3NW(*vd1r`s@tHGc
zvR{MTCv2U3J4zqEr+>nfZje;34xqv$>9nLz`{kEkLVEGOKBDiwm5FK9F4PG(krQ7D
zD4ka3$C8qg4<ATs7YQynr$!qIvFNM1BRi!G{Oi<_a19Zk66hNgbVtEuR+O|eIiu<_
z2=j)37SEQ?V9jv(^Uq-Qlv&Z8`+RT?glIA!vTIg@l<hga>SiTK46@yix_0HtQ%$N)
z(ANv;WGO{2X!Of!!x3Md5*WqGZY;udG<uqwE(Ja6!Cg03+$ninWlmXdc^Gr0d-5G_
z@hPo()A`C73|zRm@mi+9V=q4@W<cpwOW*g=*&VSb-z}^tzRJYG$CjF!$|hTcCMl|4
z{d`ETzo=xpzwyG3rGd`Q^E7WawY3?x2)KW1kTAVNn~;Qrgz8g}SiOeFR5lLsXf|m8
zan1TF9~_w>U%{Q?&aPkRQi{7?Qfl_}Zub#|o-P=JAcstvM8To6t9Ikcl_?@jPkemX
zj3f}Ze9mthlG8=6K0KPEYmZg;UQ+*h-+c8e<fxXetowGGA?P&YhC(sprpjGK=@tb`
zrca%^ZQC{oj3zdTxq_T4Qx>Pri9JPb(guB9dGXFfYBckpNoHI6kLiTD0TQ>Ugu=|4
z0Kkj7p??B*4i+XT$p!l&!5$@gmqGWl#R^|Wv^Opzdi6dodq@<_*NWT?v3TbhH+*j=
za-+qi(QqQgQ7tHETg+*SEPUf#bs?rDoYJ)O{hk2RCTQyJgRm%`J?O{%Z}#7$tiR_+
z-e$O{d7a0^b?>%qS86gL0h{(XJCCF4_8C7htu$gz<#z7eSydbx-FT-Oz2=2_@x0+y
zY3O_YxDo~hb4t_s+}!-$+Tno{S597@Bb#uBu|pAJN%Uc@Sh>=-nacECidoj##U-bG
z4vI0pp`qE8`e@%un}ZnI9TL73uTbC34K>Nco`2u$DW{|qt&uv1O6`Y-%x12N{}l?m
z9+Ri64^*MO%+Vbb_q8W&&+-)^Z6{?tJr;>Vz^FllEANwD-5n=>mjV27NcR8In1tjV
zpK{b256Q-4mQ~lZ;mm|Y!1LSmG8?{KcAmGeE+tf-(a%KA;A0ikG(U(_GOQ1vDr;`u
zgW^?=bo5g+!Okg+5TwoBX39-?^A9)Fee!X9BR&0&tKWrsl-rDXn9dxRdx0eH#ruFi
ztDTD6cH&QE8(D+%zTmTb^LWdaEtmjojvTS7uU~U;qx0{0Wo|JXf|szYlf(+FGePgY
zlJ<6QtQ~%ZI(K*v_|QQeboz@4YEUP;D=SYpew8Z9CS8WnJvpFl`BG)A{cCn!Uc~87
zJKCU#Jx%lW0nSdGD5n#Nad_tEci={9$7de(s_YvZ+yt0*FT3Zf7l^$rJ|F-Zom^Q0
zmpuFn!tHY(ph11mEH~EZgNsRUaIjo&0M4Sj%xJyqKA`}yah*^L<chjBoi9=<!T|G@
zSY5n$@ywLM<VWz`{w&D1FbPJJKhC1HYtO@J!Ir;ZnXEHKV&Gos2p+O44=5>Ve`!Y3
zN%jMldGqeAVe0GY_h4LO-jXHF=a~O!m&*xBqY{gVMVl+yz~i9camrWJ=9Df4oz6F7
z#Sx29@~zHfi<@@@OX-!Bl_3m*qu}CEUQwa(MbXIig_jrIvu^SXK<M+7SC=F1;iik#
zH|s8w{L^WcZ&FT)+a#Q4Z!)aW>PZxtg4~*Lk%3Fi84%@6w@=~nEfM76+L`yNK+F{Z
zo*A2f+9if?LZf!V`tu7-(dt^{ZL1&qoK3f0_9=}~9;^n)a-7K|EF$M`p*QWk%=Ya>
z_V$$?Clhj1zAQ*GsY&xf#oiLMI6o=5%xr9hS4`yQzT6z!T8|8ykx6E@yfVDQ3LIax
z8(5sAEhu}ClHY#YK$|lE_kC@-;krsi-9lfUPeT#xY25i`%a<pG%N#%caE4D*@fXv6
z&-}<HYg3>bs=i*j!><ZNzNgN{i`54QSJG3aQ_L#V-it+3c5Rz9laUb}1F^NuJ@{yG
zsEM30*qm|>+1l8g)6FEt_=Toym_Pa`CE)r*PnNekZfV&EViHa-uYzwNA9zvwwm?i#
zUfx^c#r8*8fBrewNhO8V=+i<mqdNC48}$UjKmPP-6>=@hENyGnp#c*<=;`>{+yFPt
zCN*Rjq7DPmp^AR(ix<N-)gZ$MuK7E<+g97Iml;qCc7iPQ#)vztDz=Mjp!E*I!(D>W
zhdC6=YioI(|LFOSt_C+TCR=|0xOvLYVrplKaz6I)L8o|nRPu_oJ5gA^$hzZ<lpaXu
z47;CiV48(i`4Ic_>h38FFEtovUA}xd+ymq!Y>gMQJ;d7bHGqzwK$#a$teaOIUU-&G
zM86bF1HP6r4-fethASRcdxaJ?K??i#-$A)CM|jb^=E}^>OoYr(qrFa3<~7WyDTCzg
z$&<cBnfK1MI+$445{K)*s&}t4YX6!*NTY%}T~O7pbKvc0=2uXb0FT1!rG&f)DKHeH
zS}VI(!TID3d&U0a%gbgkRMfO>rjLo756EKxhKX9oa}U0P_S>7QNixkoHa7OFo~~xm
zgAdkLJ2zEtl9rZ^7BgLz4*wPBfu*7K)txuMqbs1Y_+4}9hxnM!W$?MO1NNGm$B2e|
z?zowr%0VS=O|*5eWv^+#-D|MCXaJc%|B=MhLpdnLWAP%vEq#L~kUc3H^(u%h3!{Hy
z;lgl~jO^@wm6omT;GO$Y1068j3W4U`wGj#dM^`2)oIQK?Q{VxFF_7PYO0ObIwA<^Y
zN^)23&t}`eb2OAoOiXOUhVYFs1}aYj12bq@f%&TPIy4bqJ>J~(yYZGdtN+>m^jEB3
zpIvxBZ0&sVsX)eR(UrVXwDFomTl$kcKl>py5jKD8tbcZy|IyzG{OF%#6?^;ZL=u-9
ziPq%bBEJ8l<Nx=?{~m$AN8o>U1pbMLui(GR!X1#_GK*hc=J*Be-_{CkTsKO@cb&jS
zfwg~G#FuK2+iKBIeO?-L#9uqlu=beteZzv&AC4A=xh!(oJj>2gZTkVw>vJz^i%mPc
zVE^2i&hk4FIDR|)ZVk)jbF&<*de+Zane_YpOTG)KlV&>0&fWjpTpPh3aX}T`?(Hw*
zE+3nj_D9;Zh;9y1aed{@L035)3x8AStMC7-2#>!A1o_J!d*t#6D2w|3Fe4802C1H(
zov3y6dmJc`=CGKrlHXoJE!pFgnE3N<M|Tx`qQ-lVg5m%Cq0suqX}OARb{k=y?ei2z
z8klv(A4L2qi}B~<;JG|g86SG{PAz6;ifRi4pAj8+v~V)eG*tyR23x=6M*aYRA+6ty
zR|AJrRS?}q2SY(2W(M#!p4#>FFVQ}89l)f(SZ#jNXfhukSZFosJDz{2{XX!K`T{~9
z;c1+?C3tQ!z)X3?VMetBu>Da#qk)w|$WvdYUwq(xCx2I*foHU3P&UJNSqaYdEw05g
z6*tC<4Lrn6+);4`P(uc__?!yC*m5x@;By1d6&OpPzj6z$@HxV#ito`?T82Kwg{yz3
z0p!Wk)2?+B?>qpa!xvxS>0gqOq@lscV~H7iF-rFVkEw!AWdM^~*nIvFTEG6_p+iV}
zt`;*w=!LR<YHM?N;}0>eJTR!!fq;0!oTx;9BZMMGh?}&xpoMws)n8Zh^2+pm(GNjk
zKF?nvDET}1xb{5|dx@OtXOAd*P@%tU8wB$mA!8CArP^Dqfb5mB8)0HEK-oahgH7UB
zP~f?ls!K@``a0*HY+dtBOmFdS7_foCGY-z_ZHtJv@6CZjOzn=KcBkuEzCaEVd2WI{
z>I;Oz+SRL9cYK(AO&iUmjjF2%jT!R1j0<@diJKZ28XC6L!lk$EFWp)D$wY3%`XU1r
za2Am=>R?VWeXXkfVOgq^ua=0!cjW}B&yxV1sr>^9P{3D3PcB)~`{E-QYf*+1?<#Jr
zNabkO)T@}_dUZ%{o16$l;XKY)HkXn8hNKUFqhlVj`mQ=~c*VMGkSWM&2k&#(FY5Mj
z5fcs8<BkjMebB}7_f19La58m!B;5Uu!td|zvRG(zNICTRJlao4bz5A09+>)N8HJ6R
zw1h3C3E}3aZ5Ek=81|@h&{XF5dM!eHTZ<6aF76!a_A=r|2w<}`fW*j~VKL#jA5u^V
zLO27lulSy?{4BC^-l&x#1GfB5ipN*hKKEKAn_BGIHt1;0+{g_uyxv6P;jtOwRJ-j$
zI%aF4lvkmLyn&IkPbxKHQO#ni2AfBQ^?9<#hPD)?XKEDe!g43j@C)~$)Fr&5#1&Ae
z-o7(_9XAe(A^Pz(9k)mtwzYrU!=FNRYJaENx3;ehK+J6=`B<5il8LgGO)gbVK%DK(
z`Q-Q+sa)v|3LQbsQf90f;3&O;LJca;l4$#86WvT~8<GwZBCBmBKCJH%?7DnFL04C|
zyguwV)}URp{QdnIgU(o3#NJsJw$fA}=8ejqEazBa=C4+rVN+jRI_>#fo-2p1==tg*
z97l@BPQ<Y2Vjft!Mfw6)?zzq0d*^B@k7bD~(5#QEf21PWQvLDUQLsQq$NM2L6bq=h
z<>vNtiVDDc2eQFO^$A$CH|p4Nm<5@hV!+b3mk8lOD<${!e30Rkxp&jg$0v+I59G{q
zsuED2>xG$>s96bXe(3Pw>Aorb2U=c!f1urR2|{15RmBB5nS{V*Ugd-whOepVQ8R&<
zV4sU`=1|0{maF2Xn^MeBB}CuDQVk-)A}x6e1JtN6eQ%$){Su)yrj^Sb#hm?oI9>q9
z*Qy-|#xINf;@0{S$yNE_9XxlgyA{b-i7acRf9kINP;FVfmBXSX>)LMREY8Rj6R87A
z78bpL#=qR6V#{)Fw%3;{ECv`n&hjPOUe1Ko%8Zf=tR*6V=&|V2^g<je@+~PbLfUcW
z`MrHpz+1b+(OpRUu-}8?>P?B=HvoT=w42!`Ol=nK&#~;S<dodiQi^R?SCmxQyFE7k
zX*=t!cGlKDuVU4M#i}hgb+M2EId_T&h8G;a+1bm{CZMryM}1!QLBQu6C>v3hJt%%3
zk}RVwkieloB!$VFbd+32UZfS{eabl%9sS8n4*kbOP9@yRO4_}9_kmUeoG~LzC)tHJ
z4+5qEcIZV~NP9MencU4{HwYhFIv^&vzBRZ(YF>XZa9U(ZH+d2}PlH~QBcegAP+Ln2
zw51H{hD9dd?+XbD@tVL=O10cXUP-vs6%89Z9Mcd8!>av>BL_F9p=oKyQei$}H`fS8
zJlhx>;4F;naMckhLmBZ^J(sdHUOOyxEAYW-7d6I8KL*t{AO$<VEdd`^xyOTaBoh~J
z4G9fxsbWCIdwE~NySHx-bWiZ2;RdNojkc9m7e?r9-<}fpZFkJUhhW#(E(>b=&ynXe
z%8h_Cjg|F~$ZE$-Qs~BW++!={HGhySX%#U8A=;AIl0Cq?w<zP(NH9XK7Uh^CP{^ZG
zVX=i-2NMVZWnz{0Dm};m7cV)whY!2_D3EVu?qIswR`YqCc}K$&ipTx?u~@Gw;;(Ly
z7(c_o3sAuit)<xQ%LTn~d3@H5w{PBXtyweci+0<ctk)Bblo`pv05pya9+)$iSQwj6
zCITP^Ba9hA_)-kK20VyA8BXFj2Z04`hFh#?MD93%&<-<%v&Qd+0njv)NJn0e`GDu3
zH-j^hCJ{W~Bn|mHG@pp(SYS~wO3`jP=#iMNra3rpHo!nW(k~*83zoi7?ZA!Ulh7!Q
ziRVJ|NZ@eMd<`t}0q@3^ftT(Bjc~$<OF~o78ggzZp>!7r3w+Q?`E%44M=W~srugE6
z5eEWBWcZ83Eh65{sDY+ASYx-31MNz~n*t#1fn$d5aa{&JdxVZ#=AHy{*%3b(z^{*-
z5ZZkQhKGEV{<MP__5s8@%{ilAh~|!xT}GS^dcT$DAHR_~9^s=7zmc*T{Zho){2!Pf
z^4_GuV#LG_c!(w=@A!c}4m7uIU_}f%gTpIgq-#kQ^k}n34ADr_0!;jGOq;Uk4}d>I
z@Ol_a+R&Oxk;RpXwFZs{y3>j4;div{JZ7-Rejt<XWe<FG)bxbXe0Z9ZO}}2#*rAV|
z|KGBrY4J&l^?!OA>F@Gp)EDVPLV||D`Tr}kL7P>)<nWY@@}Q;h`w?C*iK+(sY%tI_
z$J(H($8;w1kwz!m_i%@2QDJcrn6n{*QV93mtq+Ae=|a7r19|Y>=TSKTMfsrZoQRTi
zqW@7PrJs*?>e^6#t!lG+6m0y9>FNW8Xre4<DnN0c|2k|$=(6JdO<qMgakCc*cxspm
z9NWq9JLy`$Y9J2Gj}vddqlaIrvq#E9vZ|TkYcqpIpXUdtu5paqk+e4%u>naiMF#P2
z6ZyA^9A)YL+7X83L|ajRn@D;F<!=*7&W^uiA}6v+QA3F-_;_@OiyU?XryR{p4JMt4
zN7PulItS_w&w)HdLC$imr1I9SfPd$pvX_fc5RjOHf}?n#;q(v9*O%-}<hYNV4q#UV
z^wT7bo`^_2#x;C=XRD=Wu%e`D0w_=QJvnGIRi?RUpE-kv7~^F1B>M#g#e>3s*&U&u
z322uv)RjWQMdI71B<cM?3d?^08w(1Zlg2w^9`EA|-35+s;R!h<E^^|{pNGu^zQm%@
zcstXZ_wNZR5{<^dxXq=w-@m`qqT5r-x?Iv|{1b>tL7B6iH2!*3IGY6r2J(sUnVx|B
zIhzdfMkXzvG~c^2iwjrri-|WzgAopig!U>kDA^-uA}PQTPGsrKH9aq18aI{$K)Qn(
zbmxopFQ!>ZRz>srD|L4P752R2aNseMmMNeX@T!jm&6#z@_~+Sc0uSHl%Q<PCSHQOd
zF%p(+O{j$_hZPl#V<dmLZ@l339e6#oJ+L<m4rOD;drS#8`_t9-?b~OU@LK|D^M|_K
zSsMlpK#!`9g9*B-u`h;XWgduHbhSL8yueIV34;Dj`>h>J(TH=KmaS@2>v7;CqI_Qb
zJ*;N>;FI6=fNe9gxDLRosogGZRawhT>mU-3=DZFT&Z}p(Z=W`P#(Z1V{!~LHy<90S
zG}*e$MzSM;Y2NdNZSMgewwjID_-4<~Z8z`SNzFu4g^{x~X`(gF2PA5OQPM^9Y9IG_
zCFNI3yLdx$fg(b04<V%$ajY@`lrH2uX8gYWE{OnJTp?ak6`?5)5nlmQBG?2a!?81G
z6S&etG7_DC9le_6L7UDQ5Um^Nw|=U6iD=6vE&{?lY1EaClDmNII@Kh7)#IyMIuN&S
zueQ~p`L5cP@+F>1-P*;<`>KJ;cJt#Bi(3aeJ!S5&<}%dIw9t*6&sk@kE2FQ&dnLTz
z)WA!e^a!tts=uclxOjt3zBh+3G}$_|cU^%rHWeWD<nMe(LRNGI0{s)-psPtXVok|;
zm4agc%xSmh8NRyo3)~^#EL)uT;k5X!q{RFTj8<<Xpdc(X`@ZnAXsT+QvLyDGAHWWI
z=`o4y?&E%m1edq;KB!*67La6*bVDgpE9VP!BU*oa(kxH5X>DzlmMoQ}cFFeyiG$^m
zq#Zh<zh&HKqc$c~#ycbPChf?>#|I&moA8;knZ@#*0ZeD6o;hC5mPpFwmLKjS_XTo7
zFX`!)tEiR0Ay9M!AyD!-D)whdJT<kY6l4qjpW>_fg&I-hKgfM?Wn79>i&;8+IIo~4
zc?E^<#Dnv&y;%8%nSbXjQS%P{mPGh*N|93<8FV`;-ZQEr9t^l%vFRSe)}(7pj+Rf6
z7ELXQ1p9jx8?s3Edw!rDFKfAG`n{aCewAX??Cu+HyKk6M_t*2UspqezC23;fpGpWa
zEnKi*yJR+a;>-1TczNyP1d(R!Wcy_{MNGuB683|Z;20GZ?p@9#0Wd~mwB73x5LPh9
z=j`oCeJe<0kSK=_9023<n5Cu0(5SXZK$@zXu;UMo-@6{x#WZImQx|yLy0yTf8zKBX
z)3Y6mZhTfebO<rW*zZQ0HGr7@O<LONG}VUQj#NpBZ+G0e&<hf0ffR#`xNx6yJ1p?(
z&U$>A7T^yY2RS>KFwDYzZqWeSZ)mLm<eV3DAf}4HohUa-hss1(+oYsZlI_8@{W`aX
zP6nJgEe9Gr546^AXPCWYK6Q4&4!sR*3;KBCT62}WQ-58$X?>&PH#sDddESC_zf#rT
z>)A7~^OER&;B+A#Ov0S<1oUX6dIE~|B}<jMqVIh|Cf2Y+A2TrM^DXrD?IX<uEb-a^
zW5D`gdj$sz9O@^&(#XK;K$GQKqzs>ToQ*8IstKt*P#PRVGKQp>vRSBkz?>Fm7p7Ex
z{kq*T?d{uHV#mVH95U`2ovR_nj8rmcK{hQ?G2~*(&COlax3V<rYh|Sv1dDcAUP0P4
zSv9z?3SCXO!@b0QQXrvyhwkOet?ucv29~;sfZ48}$H&=)&SXQg?Q_kiq!hQcCwKR-
zTJ1-k4`9J`OF>3*Tn<+=$e<Vo0fz(WL1ad`SFiR3P$tE3?t?I(Q!qtP^Lv~njnd78
zDF-CkX!AN8pbZzRCdf<todBX&xBgB5{ha{%I{`#XYN7r-lxO`nCV-?_S4mwo%ZsP>
zJN!rhzEZ)5ivEH!YE^^unXm&2<=bXXi#L`zpr~k3b|}E#zxlwU@NG=f&zeac$Vcm|
zH*4-5r-H<#*Cn-n{`?vH8>!vc8cV17`}&@4S4{#p!=@(;Q*K*aAxspTjsRCJr)!pX
zyx&qE?Ji*`3<*yL4mfPLvQ0n#&p%mLgl+G?;u{xNyyx7qK&gYqPO-nO=IPp_oA=w-
zKSbwJGttUvEJYv!3mP4qzti$S<|Gx=(rMk)X)Qs`{LX}2(ckI9vFGdfP-JXgc5p5J
zcc%;gJ7QtAXjTd5e&T9%*St89N_?`BHJ%?|6VSwXtn(bO>l=%|?{gJ+i0CZKKoT+P
z7Z{t7>F0%8mY6i4`N1K)ohFA^e*wKb<v&P!)vKlCs2^E(b6-Yb7gXOSRs_8B9HX%z
znEo1Lee?qmFhj+UIACBSLF+TJ=)-fx;aft3CK{Zq10@batVbm@7Nln%{graj{MifI
z`0{5jGIqPyL9plw(V-mWom6R2e&tn(=?G=X1>3t!$oPx@NRY!OlAStTvitLgxR#H5
z#t+TXT#}&GYjG4SR74><2trV3s9Qjf-f%~nBSrSpax;O0hYz<!_fr*)o!cH73pvT+
zlCWnWB0mSLhhA*!;qqH^$BvJ|38p7)KV`teJ;vJ{x45>Lnh5NzyCcc>>PiMZavXLC
zuE?hnuZ}=q=ziVE3dAs0XChmvzuQD0Mhmb5<2tx-AWfN;%JO;KjBhd(==nVitbwSx
z62h~9?t+%_Lv{LlLPL~Advu9-fG&|LAH5VF-Dd~_Yul#+_VL#@HPHtLk^PB0IE|2n
z0m}IWpdl7VC;VtIIzJnG>au-5Av$*LFjF1n3={Y0ic605@%~gQZ<X#AdayHctPrGs
z@CGh^d!eYPsM`dCnQTqO%FAoT&y{c6L|Zf?$8{ewQy=KgjACL5o56&?zl~_Wu#iTC
zIJ(~F<G=tLAK&l?|F%Liq#F#t$xosqx=T)qeb_L++Ak7ZR!~fM+B)vf(HeX^{6B&r
z@^<pc)}n?r9Z()$nqCjM{^o&a#A+l9a_};}=y!7X)=t0#Sc|#Y*=kTVrhDRd=%o2%
zne3a!bQPVrp8;ynx`H?Q1Fk;gHIE?OfL0@mrza<v$vX^00hGhy*riveb+6aZmg9)6
z@O!M7F`y3~w{71*k$z15pO&~A>F8%RiaX!E`+4Pp32!tIJ`NAie`7WDic69;pjzD)
zYnF|(9Xt7%y1GdEdQ4T<P@gFdlF}%k7eEw`k)tFt6K&IB0HFZ-qgNP?uGzLo%cya6
zad0p1-Bsq8fgWnKOo0I-{4>Kf7a^n*RUjNJ!zC<igywF*5;(hlkh_i?p+6UrhQ5Z9
zFi&ucbX@T0J_bX?^1bU$T+3thZ%X!^<7{C)qN+-T?$I_r?Of}!B=8NzYqOx`ogHb$
z<nb_!k15c*3c(WdMY{-M_Zth5Vkk=DIK$b(sjH>sO3Of2QhueOz_OMjfshK?LqIyF
zF)p+NuLyQdR~yoJDOCMOZQ~wTKg3jkuSX#oWGvoUmO$;?kMz)jp(R#!UjJ3TmkhM@
zBTqv>Xy|%c{*k?9iA8-!V})$;IQT``qjohdyf9H03oS(Y<5m-g<Zwo^Uwxeo1E542
zApyr;HdQg)ymf1P3E*IP)2~0C;K^)_1q||(jkZVkhP>n3Q|ovP*-&6DlBd7V$e^s(
zl6yg#q4%}8LL?qII1Pz!K;PV;G-u1pPIR?u;|3;b$5yc-`Hq!8k~~}ieG{B+J+%Z)
z%;+$VJi_F_zHzN7Az^egeUBa&@6;BknKO44bc_Rih+Gs~#Fs=gJ!INd`@8a1w~R|p
z%82`;oia-UxwXt}CQ9F%>yMA@-|yr%4pEV~`fCE7o`?wFtHXZAwP-0yq&Rx)(@ja6
z*zE#jQe1jEpk#l85`MNQY67gzzST_Ax2AluKniVjkCVTj-x7;GFNo@<M$09VeQZ}4
z&OYAv^AT=s_u=25U_84$%m@ut8c2!}`Oe2w<lnNQ7u3pE&kEA?=4Y!G+I2i@@~^D#
z;r7lGHwLc}5@I*v*9QX!w!MD`_nK$R=`~F+mqKfRV5wq7vdKzzluy<&7Ja4ngtWaF
zTB>unJDC<_GK-T=H=wWN{bEC3BJY%2*Ds2$Ta-W>woy;-AxPW+e*kXKnQ8CSO-R1l
zT1%Z*?!hI1h{Gk~_1>c$nvOOih0@BvhY!neaua?N<ohx<es?#HV90Z?@`yUZl;I}K
zZ@M{Crir7vF1j8Vr_&!F(Z-GB9d}1Vn<wM66nIS9d?Z?ZKi%Uxq-qz$OK}%9I@^;o
z#id^kH2??75%Av7rffI9bUF`25*9}btI=p#{{2nF9R^n3qw}dJUsXzG)6$I+O>dU_
znsy0dg_{MuB)#ylwaV8lB=4d?Trbn_6~Q?xyvRR$LA=qnY~)pAR;_;a<jJZSKj2RR
z5x(7_`4zd!7=!?p(o+1}&xoF{+*$6|Xa#8P7Vb?i#q(kIv5mLCdsrk+OEvz2LcqSh
ztqTY~Y_B<!`v!^FoK91gxZ*43u`^Zm67$^t3=%H;Dbr!f($u=@;~Q(u`xkRh-cjBF
zjbPcPc(oT4l5&i<wLr1Ond1!9hq9J`|KQ;<evxh|FbR%|>grv{N?{L_itgFu)~bfJ
z+B5BIg(R!P5p*!M1Vhtzf_3>Z!}`a87nEl~K_A|}wWYiOtKJh!cc8zb`5XO<BLLqV
zV?bY7*M3cqBWN$z&wl~!^^T@5O<_5am#qN;(R*FzCfKi~32jYkWJ5V_1oF+%@)c}4
z{1$IwfF>{%f7VfSw6tb490B(-me$$$-Zm6Mum}6Aq$QUKswboM(h{!IN<eZAThzFE
zW_e#j9#7{>q+nw-_^FD~aaLu9?)nKV<nI>Z9q8bvLb6Y`uXR;l6Mz5OOHaY%FvV2B
z{8w5c@jViWaY|ula!_{wAyEOksjCW;s9gN~lFVa~Ry<b>TGUX8K%weg{H(HRtj8MM
zN_w7nn+Sx0SOO8Nk**h>9YwVVk>gAc+|=Mlk1BDbh}L!xWmD064Bjq6C*R)e&};Dm
zLkR|Sa{IoPac2H^WwA)q<QVxPe>->0s5<+HQc2fgwM5|FJM&X<geap=^iz9Wd+dvh
zP27!6AZ+P^K^3i`hN18Tgenh+Ay7xKmk^V<Joc%?@)M%Lq`)F=3~mk=bE!52>4eJ`
zU$NmJJ?;{ouSGq9DH<K;z$=A(cVo|-x6sfw+4CR_NX;93QCybgNXDj(b8SApvqq)q
z#`ku})_!T?>;pfIBtXyJoGON&Pfp-1S0*%^TNH1<*O4Ob(Eh`PQ`W(!8_w_KabeP*
zd=&Evd*Slh-h2(muyT8?avw>Aar)!Kxbg%fu#L^YZ~4R6gVLvUqu;a<RM}SJ-5XXW
z38aV{+ve~qa9dtMufCY6>7F<q0#l(St7+aQ@am#F!JcCqVZ<G0r>4jVK%>J03EhJA
z;{!E=MztzpWy9OCk{#M#5;{b&qVMT$4CCKt2c;*&D2_$99wJnr#p#NC;O61MA={3c
zYr^=r!HGfIe#wFl^`IIF?XiIvt#MDXo_Ph^_#s6Yiz!luz;@$gejc7*OPqUFBR7Jo
z>Gk+U7FFN{SuuVj_?fO!07p$`wuFwf3_Kr@CLw)ePkD(D1m;2&+IlvQ@iiguyc{*h
zMMhZead9N}6{~#;^kP$3kOi-(Xf!qo2z)zndR%DHkZ8X9AqIs}#qkqXl~5Jd1w_6@
z!MTN@5ZyF=GU&?>rN%$zSw9q*MC7Q@MjyJ(3YXl6u31IK&9uvntkph>X7GlsP(e*K
zZRH(ym&<VdU+00w2f;m{@1D?=8=Z;QUuz;z2YwzESS6HTDbOp1Lm8y_e3{(yxVQ(U
zG|zrSrRWMy<bgnwp1TJO3^9upMW_0wG@{?YqhWr5fezyez*k(-tRc<z8Z>59JlU0#
zmhTxzUF}&55UNy!_vl(*Khf(XaU39rrcuWSkdJ6c=%DqsQ3DO-28d`Vy+S`};lZd(
zMtMSmoA~~z4@rrx*~$K!DRHeqNHgK!Q#<caPd|zQ*)@B!6O<fjQvo6!P6=alKaBqh
zB3x{NRhF~6;Q!ESP?l@=msOKvxlw*H6QTNX8NYz<&_LzJmL=y>m!_?B%?FX3=*8;!
z;*H3nteR*+S~lKLwF_@Qq@aAOL;BF3B@*VwVnuAR9kiiGKug=9fc*t8Hv;3Ru^F)j
zsogEuJ}I*Nyu5#mza}daTs)!TV!(U@I+*z)ORcJY5Qg9J$(<-;eN?|;7z9qIIO?Y?
zP}nBxA#i;1L?3K`?_wWev};#dqeN_AU>)f3?H=@b^mlqpXqu8gaJY+^G&M@}qI1T<
zqc2smkGYPv!bZl9%3_P6I?8W&*Y@5DGXH8^vb^QC(pj4aW)PLn%gaj|Ke|o~Xr4+i
zwj|`^ywvje{_glx*`90E4u<gwpOz-Ij_shj-RN|MG*UGA)Re^cUuK|){CxJp%ZE^7
zEbuTR6-Rm+g+M8RWUVd6y7drAw5~@XTM`21FtQkHAoi|FCMMqKB)0_b99B^apyhWu
zI+&O5P#gcC-V8||a{Vo`{RT%_EUM)_Y)iJaRPmws7Q8rk{9%%1{csAy9E1z?p%u>%
zoB=1l>J!cUAU33>XGtg0kDm!o!CztlE~T?JQ~6(3I+ygdMU3A{?(CFb@w`EJY!u>2
z=Z{;*hizDe22g50b@C)4MX|n)in3J`+K7fKs&v<DfN5aFBzBQe8C4n;c}%O)4gqOM
z=h$E}bW{nIp41xo{G-3<Nkia_+tc(94c8s%U}MgEqg&Yq3Y+{h2srpCTEZ|`Bh%PJ
zQ#uS78ux*J0x_a>x(!raaX4Ztj=;pyu8pM1e;UC3UxmBUEWjv8-oFTZyl`?om&tJ>
zdaFu2kfuGIea8tvM88?`AY=^u_w-m)PTQr6n6K;*G6@X_+^=iT{kqEC)2;Ne2y@<V
zZ)MQKcO^GPbhnC(j7)U@5-{Rt3ZjXzREiE+-IqJZ;L=$f7AU4*xh~z=zhudhWgQ%H
z)tdmCD6jVhLc|o+6m$|9ANpBc9W&pTx3Qi9)Ywx<9NFh|fuU2r^~9RSpbfwS4TO>e
z9!o280P%`ShXhrQR_UNaV*g$lqxbV*qrwO~Gj<COVb|$ffsU50r=_un%OArabQ%zJ
zBu<^SF@c91iSD2sZ1hCmAP+vhYha+Q6J{As2dxf6>joXz+=KoI?YARg>2M3^Z34rM
z1Ah&-N`Hp`p!MP}?c9M@(|<$Bp56+>0#F!j?9miEqD6!C3Nf9$Q-`}a23{Gb@gSaZ
zaIi;%S7^uei2XwARv1XGk8F;?BA^|jV}_obrvGVZ|9D&XU!BH*Mwn5J3Inab!za&3
zhk7vCJg$RgWQ}X!LxWxexsnk&lSWq9hFg_rDf%O0PbZaGba(ncIIsQ*4xgMz7Hfx^
zkWV8LMQYnWAOMoIq%K6iJ`HLUygUME&N$wuVB7WO`2855bl=du>aV@!<;#}{Lj!$%
zp|>&iOBx%%*c!;I)l!v}YHr%W^_Fnm9f7P7!yI-bx5s{b%7*_sc(vv?p#7$$ZrA7L
z<+bqs0VY^S%Pyp!y}nQH{lkaTG7pW?(MI(YTE-X)c?|ltMs8Zw_7jYTdwN_VR-)}T
zwOr}t`L9__T-@9@;@Y7@Fxz2VkYyZP5elxv>pH;LrFMUToW-^e>3BmeEs<{;$o0il
z2}0u<G1=(&>6P_TpKBCTuBHNSM+wuI4@<#6+VTY<6gmpQ@GR~41DaRzr!eWGs+v<N
zJ(Y}}*JHwgO|CFi@6d=MdEE;jF3ZeLO7swE6#a;B4LxnNUMJe3ptYUnGtZ)C_mwh8
z?o2+KzYg=uxZz9aY5Argmi-`X-l~Tb{w~UABxb6PqvSZjOD;c%4n%5ghguYInt{{Y
z)w6^JOGPPCzf7=1^Hjxa`z#G4X}n4v$OwVHt;z{)u<Ivy>xbx<e67m6tj~`{y8V+!
zp8*mp_78}TMU|Nu7E>bCc@wHgr%Hki4qVVJj_H&e#79i(*$T{p$&=$pISy3scISEs
zu%p=kIAZqJ6k1LgB-=lyZDbZsz`xMyjXoE!-VXG+?C4%$)|AG{Qe*^N?5}JaYO);J
zFzP*FE4D~&E;CXszi@}=$UF-LFi|OWgs%fr?Vi3jetv#?+_XQ3oUHAOqIbw3F=;K~
zM|xwz9)Z(|dtr%q`xkj-<@+76q$LtH>z+$H^fqrofu^c(ii2}M4`|VhTHeq#&PnRF
zdWJZ1vk^sxN}wkLMXn+WFfCL1P?byRFIcyHc?D#X9OJb?xZ23dvA6F9DVvy{ICoAE
z!wHn>BZJdK*~n+%s<_|Zdu_i_-tNAuZ#fu>*O;Ct)ohX`Ku7PF=!C5<jCMO4<%^6q
zqL3ijey<OU&S+^pRoK0!C-z&;PTee@&?`*J{3$k()AyfEoEbG!<r?~d%lmzzSd&rT
zE9m&SpW!sZMCK}x?{ZKpY!`5Q>D3?u*5t!F*acij75vA5(&tFoEz6v}c<qY5*NA+o
z*ia2(>5o4yEexlv$_X5vu=Zp2zkt1I&GXc9qs8{|&q29u>3_Gw@s`GIH#au%XOAA?
z_!;AIB+NN504KFyD!SqSu=f^VRc&qC=%SHMNdXB7K@e%_mJmc*1Oe%87byag3Wy>l
zDJdWzB_V<$2%?~bbVwu8C3T*Ky0?43`}@B4|Ihi)b)D<Hi@k9&=U8LT5p#?u?;Cv>
z%zla6Pf@gTnNFi+7OIy&5Hgydr2`+t-%S#j4E9^b;I;&S<oG?T(ddA^FXsMl!*AWX
zH9H7anBi`>FTBXKhYug#GZBKG|80lUrW*wCtvz8c2z79nH-*=!y|p!r;W%#M_Eng>
z;V&d!mh^v+pg-@){SR-zW0d#HM2>Ie7TXJ*)uO5qS_w;lkPHtH2!#d9{1Tw;o;oz7
z1Pd|jz<-a`jEM)NxFE=6cDbWWef;8HVo-*9ve1cr7PTY+&hI187D&}U<=?ltIJmn$
zNABkYNL~g09lwOF9~-bHXFjmm93LeV9$E&~yAHHwc<X&8!r5<mPb_)wt+XV2*Cv3&
zarP`E{RW1D&k|1~P0Yg<-qkE8;=_!(x|Dwy2pv?4z6m<RRfuN=>rC3D(q}-W+rwIj
zK6C!O58Q!s59^o3<>%I%U4k-p0ZzQ5{Bub7ze_7P4*{k2n{(q6>WTbp2l$P7U@rlK
zv}HrWQ3U;8w$E#|dRMPrh0qBwA3nH$AIwOV6H2%r48b1rF<TY52|mtdfwc9v8S`K1
zl7Fg+{~)9OCw20_j|Taf4Ez^y;YX&;Ut*1qWbl8`?0?PA|8uF6|4+n1q1J#SkpV;p
z)PJ3_BtNu@|HH}J|DZ9TRGOa=8-Eiu|4+-PN232D>mLsQzke4=>zPnWgY|C-_CLG*
z=j)Do<RcC3rzPb`RRC4%Z`t!Gkrc}PX;b-A8IV9}W~f>8x2^{{`1kbof9SCvH~t^u
zoBnq-h9BxE=rMmR9sicC>K`<QUtRR4TJdYiL<w?rEO7m|P_;muY;$jC4IeM?@_4Ed
z0|H61&&LM9gI|_f2Z)s|4v&z&b1Pe25PemDit{TVmB-@LgW01i2n1#07z(iYznBf#
zfVfg;oWCapLk97{ejgtIoq#<BOef!QQS6u~<OGy+oc;0=K*65}m-7BrCd8=qo(JLV
zsSIENE#KAByCFdyz=#5tGLEA-zVFio33K2>R^8jm3?(|T3aQ9|@WShD3Xf9M)j+lJ
z!>c@Tg6+fZ+Z$3eNGX~n_AmR_C-<+{rA2MU#csus{v7u|l*%8&`KQ)-Bw_ztqyBY9
zSs#H*{m3UdiiY}K<vg+p{0EN#_3Hjb*+)q{|G{JUUqJx)*O&f3Ld72mIlqTO{_52K
z-w74n?3W?F<$q2l*7(-c3~YnOfI(k#>V9(adF1LOkh`E(|35(Ij<yJmwx*-Z;J#D;
zWo9tD!T2{6dT*u)&~XDd+4P6BhHH(tAbAnE6qQCLtvl1VKb-i7s_z1W7HJ0WEO6BR
z2)6t0l6@~+28TKK3*d-h*uM$T?Xx}rdNwd7@=x_b=oPsOFGLzb$W%h~a{#K!Y!L_Y
zC0K&~rtqZF-+@55vXQ?9!f~PqwEq%wfli!rX*_xkTxs&XgC(YNmx=KMCSETI8eY28
z_^*kCHG2P+*@JI*5MtII*Zt2Kv)5{>z=WZU%J5b>XuTHy1L^Fb6@sEX?TVMz{JTdW
z3Pt^FqnOvXz2?n@b7?zjnK-;Pvzd{DPkDjS1QiVh0k<n5yuKNCq_j0XucRKJ{FomO
zzJ5T#lOOENm<$#BE!vsjH{4!6eXJq%ssF~4#ofqbhqWol>BRAXtcLGT8f-^RE^K3^
zKsffe+!XJ@1n<C>9@&AC=+=<it3&DWDy?nCCnuz=<pK_70vL?r+}?|_`y#5A+Q}T-
z6B)O%dHwSJ0@$dGZ<yYYX=-16{=u&p{vNMM1$^*vIkvsg&HD0;5B{6a#s!Zxgdco=
zdRTLZvD{&=nvn4%AdGD%*o^kE?iGhiYBw!qc}ne$rv?*0fc)7{iw7BGj`r&rAHP|R
z%Fd4ApRB=#<iNu<sl&BPIAPB|W&<#reA~icl@tD`rzJpUzhyK4-b5QW0O%?lKCC6C
z<{J}0h1s#r1GlxA%E-Z`TxB5Fi@Z<ee19+Y4-?BHdWO2G_$x*PQ1;RF=EDRd+sY%<
zHDA6R7O<6_kSMFJ-kw~(wk;?^1!9a?;&<Q<Z5Odc0wZ}uL&>#OB~U_tVyx}OLP~L6
z!*gT*ovBxg!vL<St);~esmZ9`d$CwKc)Q9@c*yw`&~dNSFZnrv7QeWAng$|nt6ppr
zK}1d&1e`nUxS(RTmYa?UY7R&tEBv_R$cvRbPU~O$-ER1xfE$m-bNPvFi-{s&Fh={Y
zXQ4d9hY;$5A`mKHvf~GvC@_X{`3~KV9`QV6<N!;`9x<#7gYk^4zSWicv=FSdKRmiW
zjLL366_AoTgxD*~ZEG44=WDwMDorhmQU_n8Za-Ol#duJ{BYct#g2Z5OARf!}Fdap(
z`YKc<!NEavcf4YMCfct3DXVXAaCF>q7n!8xz!nU*UDMK|<+9*WC1~R@FQ$}z=h3E6
z6&wTH@vosz>_9l}Fb)FB1aOQu0RH6gLm{hdRRE6*yWa;U*U4yZpqyq9h3xUTFOYHU
zli)?X_RVw^1q+ADM+SQnH|7$rH<zBFAuT5_D(~1T6%g@$<{ke)-#qd)9I1#oK9rP{
zD}7lS2Z_00dp-hw`w7wcdm->8f&qhc9Vj?Hf_=0_|J@s43uE{@N+-JvSffeTEm1}v
z^WNgCbP}g{d3o8SIRB=CRQRB*JB3u&J|BhmWwu0}I3JHQ^*7Xmx0g`CeNHm}65O|v
zZuD13EpZzlF4THm|6f>b6u~z3OL0QEjOTGfh{5zyhwJytD)X;UYycYixBa2)C-_>i
z`u!i^YZM3WAPE1|kU&`r!|8FsqjEw8$e7?NT3o+QK}m_lle0Mk(5t^YX!7pv5S{xS
zYt8*%8bkr8p?}lvq0O)Z97*dxRhr*{CQ+%meIwx8K?$FK$e%x(ZA=2+JkqGbLGAt}
z@B=;vV8i~2kM`4>4$c1Q+XGj+jCkJR(|Vui=$*Dh5o3!h6{pWERyOw*6x8Un*VvOL
z<<;7fM)Z7SyFn8Dx?=tw*~99d?7ZCql@hKV66RZ-^3Pt&U5yAM3xCa%q%wJ1Ibfy>
zxi0?R=@f9ipBsJq`i8{e=a$}-o%f07(o#~qrgtSwR+*kjJd<b*nv_{(Y7HtR(LZ@p
z9QVID`Ah;CBd*MViHxB{KaClKBg1gL(h7dSt^lX6bLgKek-ZiIGy=$&{#5vZo>*l<
zlAtOsk!2#jCEjZP0e(O$f-m6e;-W76y%x;Vn5YY_K}^K_gXiEU2&!f!zJ<D@F7BjE
z3{K)A>PH4n-ot6c3q)N&J^s5*)DIJ!d>eEv<R$6?l4Meb52}cRCq<Pu`1yRncEG#r
zoh)_3$LAaq$375Wbt}4jc|-UHGus5>|E_mAj?^Whp`FHj*>yS?BG}KJl8%K4`T2tk
zh>;_}RcQzd3#*9$W?s}6VDr4*1xXI^zrFxVBmXG`s$UUEl4~I-egI3K7#y@!rk8kV
zlQIa@L!wX0+@wT@5<dpuSHYL|A8x`BKw({6DD-X<E<nHi=xsDhpQCGycxZpQ2K}9|
z2(`n%r!k>~cqX9QN*>BNJM&@O`nUviQ)#Ec$w5ti-ygzu;uYwqs9J8q-Qj8eeXMSd
zkKnif92o{GDykDFP5{9!B>g0$UkB!~M<On=6eB1RO8ka&N6_2&Of0PE7Ij(KCWz36
z#3P;WGg<|-#WSGJU#Ya81L6@n1@Hx+!lWNP>ezlM1DyxqPHFWJL~A<^o)$-lkUM+!
zEUg9sA>m@a+_h^=g2Wi2Rm&WPv#6{QugJH+J++*aYn@@DsCYY1?$Swcd$`Ru^}H{8
zTrhqQm|J(ZmPX&;IXNvsq8y(J-w=cW0|mr@zA!lTV7I3FJ_LkU6<wwuA7AJeCzyQQ
zlx2O18$eAQBHjWc(dVP8fr<p8Zfy(hFt4SPqgrr=lbzkFHRj~1Pa^PBIzcJ{*wD=!
zx7ODM_IH7i<<rNH)TgOYxf`wrUx8Btm{d`g3b4vU-yAHM9Nr_|+1?iS-R@kSL(zmF
z!(}^OS-$5Ik2ua2gM)uHc+hB#f@ht)VHRkL%2i7R-eqeG8ufq!Lk)|YtN$xOJ+SbV
z0wi{h5=2M8*Kg#2EJH^C#>_7VAO@>VwlT1GKwv$4052vEPO`*4*iy_Gge15kgc@sr
zG}e|u(Zy{1QaML4Y@e7X6r-q3xl+NT@xhhUPNU;;dFLE_8K}m1S2f79P0wUQS3-oh
z#fGB1yuOf#+jP`iSJkNhF4(?JUtbFYFKXiaJosp;Jj?16E>4N=z;-k*XSQhSt*rgD
zU#?r6zQrI!sjEo_Md@oGni)Y*h>SSy9hQklNC>2r8@@V=>(B2O+<pS|sQeEcA!ZR@
z;<QGBm`6c=2P6qJhc%L{4()=J{oXJcIFfgsoCKCLKj3IpOI26t$a*dE1SW(?)6Uw=
z6y;EFj#;u@YNkT0*5%6}F!2O*`FMMSS*N5`eD?LxT>z9cD&G;+V@2xMuRqRL2PLJj
zM=IR}Xkt<hKRh~bRg}mf^n7U+0O`6z<GBsYzHdRK{iQXIvqwX8yS=>~B)zBkmcD*|
zgioO<1MuLJ7A!e--~2(&h49GZi8zl$9%}VC_b|SJRFLHPO~{JreCDm8qGC35Ef+|I
z;2i==vqIx~F?Mz(^XHkyRhg0kK>qcqVQH``I!P(qMFj@P854P@GBh-_{}}aVpSS|P
zgaWgd#SVnLs|zmyS5oJdG!ZY<@Du=r4s)as8=FE__Fj@A_ZJ%B`%yq+)c##tIEB>B
zHTUA9?zck1!orf0#@u=n6BFfI$ed}@5A)5OsDW|LjG0W#gyl`RZkp3@Ht~`I836%-
zwih!OKe3q0(G5^+Lk0R?URxkiT9@D7J%mu}a2L6^DC8S8oj!1sY>YN95k3D9l@N#W
zJ^^{qhS)l!c!dN9Q`Bsr2@)rZqK1juu_^dijuyCwM?tqBK=q!tQG@^RSCfc$K^Ie0
z^KGjOVDct$RRUJ^0M8T+3!q(JAJRmJEQ3=I>0w3HWkj())omK3YTCZ->)V1E-z^?B
zQ3!9YX(4#=+gu5WiM`}u8keZY_WgQ(zyi{l-Qta(WGjcI&vIzYIy^ia!r)`8UPBnY
z79#{)71dDK3|1}zd`$fQMT&wUe8X8-Asijg2olR3O)+X!e^CxRK`E(GQb7%v&dAPi
zH%=Wwy$WEDL%s>~R7FmYKrk`2ut+yRe1kWbjrW`sa6lOZ0TNc5LMtpHNr{R1lHRbv
zNJ&ZczqHp7w`CA=(}j`d>x5}@{Q-)lQs%FfBiAgEER>7!ai`SzD_~fz4~>cB=jWT{
zJiw0vXXcg4ory?ZpspK%cwtz}D8IKX-nxa3th`-Yy9;cjdC1Qnj@~|Lz^=ak@W93d
z!vgh=J8LqyPZ+7_ywdRl@|@MJTb}Ku<}FEv5u-#=bo7%z_y(0fH01+U-)Xt4FMJJ*
z7c-4J)jM8g&|kVi!4q{aHumJgIpD~4`Rn6)Eg4^Kj!>xfRE3>^SvUv#E6|_wIH{l3
zrOFgp6dwadbaf@zbj1Xt$;@w{TCSq6pRgB{JXNfO-3xOZXG(_sJ!-wvQB#XD@|W`6
z*36oM*T>rSxyOR;8^YW8$Bwypc|D{{2A;_N05zmxH)KiYA@5Va!jH}jxrAB)3t|8z
zQmh;%vk1qp82~9YlWBOcGz%C;u+`-E!`n?QOr4jWehu8?Wkf)Gx(_ph)^!qKVYD25
zm0)3^HiniG#3m+kYJ43kH4l}8g(W0t^~WYq%mowS)vH(9<v?Fn-pef^arFRKe?*+3
zBo}>uf8W9bV28G09raF4NKU>o`S#*ORrp*9`6;#ZX;a)>3I+vyIUww7brOa#|I!H?
z&3O9H6$*cCb5n?N5gJld5$p9BXcxU`&Qec70ty85$9&4sSH5m#MYU6iB0x^e*kqhV
zQ98+b9pp}%V$2+eig$VROI~W9gfLp}KvczH;<H14EU-OA9Ikc_li6srQmN7|MF!6l
zZ_8XVAE;7@he+vcD~ylZ+c%Ihnfg7NYb9}Rs4m@?ga}zPov=rL8R-W}NnGrEK)l_K
zJ+J!_kfBj~!Ry!Z272%kp5fr=JRE{9Zu3EaqNXoIq6Ded3Hf*Q(GyRkB!xPp=tI|j
zMQMPrQd|zP|6G!RY5QC6p0Nj#+?<bq_Ph^<o}N%h0qD+BBL0*mDJdC68O86tZhiUc
z2H-Mta?W{P>dsaMy2z9m==Tm?P7uu=P(8y9anw=ic!m9*2e#Es5qJ-|-@~f=Pz^Qt
z<r&<+9Zi1V@M-`%e-8TQ@-j*G3#U=Zm0dHyD~F}w?0)UGlwPrsY2mdn@I_Q{<^ZMI
z1}-ppt!97(6RBiwZm!T(Z1C(2pl3^#+xBqTGGM!+_SKTmyyFIGZ{NOUyA1m{{r#Vt
zfhpi8#cah@OwP?A=es!-k0~C*#Vt8UJUum)AMmge(%5`r#qyEV?axt*vL%Q}u^yRz
z3j*e-X~(jf?sg=5e1&@h%=`U-AwGBw(g43K)<+$qnYl`^M3guQ2q4WM#KV$DfiU{S
zdEF1{1o_C%Z^BU(3IW~vL%p2`&cAoSRCZ1dNNez#e+5v#xw$zIYM;aY1d9dS2qny9
zdgmj>iJ*ds*e9@BXFHGES{Z2d{QCmq|J{-|wb)Lc9EY$Kz^}g(6$O-UMwXVR%uzI=
zhyg=L?V2P8TTHakAqtKiatGPD<&~7^c4c7=eu(kl(IfRK%{R1=I??@@4`K~^0%Aab
zhxf8}{~HJ}oaO4^C?Y?7n}9~v6e^S+Ogr4OO3WxM5>+kYyqqI$8bY|lCLrKG_A;=S
zVILxXSc&~1+(=7=c3?j@O7$Zu>|D0JOf8ZBJ24qGHAmVFK*5q7WIc7taF^5?aJts$
zpjsnAVg^yy*8?im@W9Pt-LTl#crBetMZw`Ioh6>1ld{;eo@&g?%R_AvKX~7+WdOlK
zR<Zvm@ZdptY#f}(!>i9M?;L!AO(q90337tOBaoU_hn(hCQ;iUM?_1Ehwhk}k#tECr
zMxYJf7=s8W7GmhAF_+dL=%{-Vj3E1TwCo)0E={?Bm1?bZO6JkNmjjPs`Hg*I2=M3B
zK`ukS@tIyXkeipLr!6-v(kjHQEg`4~^c7&?B|>U0?tn$AH7MGRf!o2pJqUz7V26dA
zipNuL(~b(o;PgCrR%7_St&JH}Eyg+VgoK38`Ip)4s4#v}S;CZKs@(_KAznI_nHaXx
z<vx>u6H@+lH9k>e-@X-FNztIl)THNJ76&Ro9srtBes=cU);DlBc(0^jK?0eUg_#-T
z5^tc~+uQR^Hn}Rk)%M`D(p`E|G?_!5&nXbR$P`8z79M^<S-BcaSu786nrAqU?{#%`
z#m2^hbl(2%UHNl%00<-7V2Y4&pM(4qF{M}2;?_5?XeVH~1+I&U831sf2lK6%Ta3f%
z>Z*Lt2@z-Mz3#9K65gUqP7lvqC94Nc$Ar{q{Qw4vv?kyxwA*wT15agh?%~6Ry%5w`
zYT|`An=VMbOBUAHi)=N4QHi5ExX*J%cQ04p22_xz4F>}tQ$7vYe==oYN_krM>&NVN
zi1ibU&w=Kjt6zuvs~@`VtXhE08LXQ!!@Vy+or07vkh7BafkD&~CVmkmh$~fAh~^2;
zcL`@<_ERnC`RzhFDd?r3YCWuA2Sm@$-GKmspRb|4E9A3jFX2ZL*xr2r0b6?KnWE2G
zcQP%C-`M}|VPY~$D`=Z1K`Ph@LIKz?!#rLbP>Wc!QB=^CQ3L5C&%LCR`@$vYCr+L|
zZ3iR|6ZHvx>Zd?}Ro0H=aJ`e;02<utoht`t<quk02Sfr6>Z+=$Xsd%E+7bs7bMO1Y
z(|I=#yTu%bM4q*tGW@=~8!2J~9r&e?;W-$hfnhTz6m(()BKs38y;GU0g<Q{`J%jXe
zp7v}Hc>Ow3tAU3dS|ybs_v$ub5f6(69}RG3ZhrtzK|(fH`(O;L?1L#d&E6Z3m8eY&
z>%Vq#=ou-g0NCwy3becYcg^M)7#W>*+);8$s3R<z-rlYgua7N$fe?H;TBI`!Q+%>?
z0V;9Sim<_XTy`;9g9PmykoCSlpwkb$y$f+a1-eBA!;EJS;9)8@H?lox+CRM4rUO}`
z{*4v;NHzT;$dxwkAm)36q9(*e9C1U^X+(TOMISZ0U$VsFwA`W7n{~ZZ=-Ok&WAdV|
z^ASc-K9qKJEcRqDdhVK?Dpz6M)iltjGk|Zi!d)7yON>hSOm10(=>efqy*rx<mg3H1
z(sS>Em`N9q_(g^lAGhPxwSnc`@3EVJ{0o%2_6XUThbPbrnPI8k^PG>a^6wMAP8FwS
zU^syADeDFmP?{B{BGX3sE!wQB_8>j!(yWqEp*X}<@uCzOpjoygjztj&s{}CE0}d>7
z;Z%Sm*yN|$9-M2Y*>ErW{8`__0YyLW3;H`~?4Vt}04i}%$pCfp%gl_GXW0;pz2NFK
z;0srK%GO<UrEhdJ05*Tf=wmKJdV3e_D8#U5%$X=hvaNs!-W|jZGR(A0t<P4mzW5U*
za=%o7CjfP=H5{~a2)eR>_qXFBW4-5Ml6MSrzu$f>41FGZi+_;_u*RnzgbCZ(N}*<~
z>Z*ukZc);m+&x_AG%8ZklVZo|>BS)-yn;Fy7FNTBo8N14?vZ3j2eWjQ!q$q3g%#eY
zXE*&o&-l{mf@;(}C%m7M0>fdk;|ebW#j6`&lw4n52fqp^S&;<0z#5Z(m22JI>=6=`
z8}^|88L&ivt&tSt%*Zw{x26RgAUAX*cP8WTD7g|20>HEcUMyb8shF4;Sh%&H=&C#H
zz{b;(D$b#gKa5&MZ+=X}VJnS4o-tN6`t$?YX|-6#?py)8UNwo69RvQnurvYghTg-i
z_)*COEb`J9gaib2UQ4M4D8}=RhN-ttV8W1MhZ54|SQbM{YE!BVouqH|z#=T0R5hx(
zxSI2H6diaHNp4$lXY2EDpT&>XFi80u)%O-cT2%;x`eQypBr0OGGtL9{1%rGowth9=
zPNA*Qmv{m(+h;G~WOO=}LYU%+@0z;D7xm<|^u5^rI^;8h<ooyU!^YbGHPleP38+jH
zE7a7~w0Y|(Fzl=t%5S6A!{xk58PZ1r0}zv^Qpv4XJOra4Knr@EvcxJ_NXn_IF6XF~
zdgk>%vwE^K1%vjT>LY;LYQIW<G)f*`Rp>d`(xQ~*0wUjcY}ltv@Yd4bPz6k!tZ)h`
zo`fjYd}D^Zu6juIZhKlTFU_~xq-}M0r$`zD$bj@d;5WClJlU_>Q<W`U9RqI$p5Fsl
zaTE59uI$Q+0}s4JY2wi{ip(^^UTrfPbB2jYTIgF>R_JCUJ|158LNe?*1*O#>Ip<7S
zC}~~1cz0kMrQRpvfW4vURMOKI-`n1kUVggk=H_<V)nzbSm8$=sqR;r*UPRk-<T3;Z
zcE8}$JfYLBp~-5rW1X$?pxC&+U|0^N5{|g`X-e!1SE3U>_Gf#TkdW6^2Cq?SYU;?y
z2pp)Zsy-7X7IGZ=JWx?SQDVuCam9bp@m^XQ?}t;A?<!f@P@BEDF47rt-3AsF&g=rL
zhOC?%Kr1}9@)%B+!9Y5f4)o~{foy$wS9AV=2V5JyaX3Bemx_%<+s`f7qwuZT%p>fz
zuJ3{rU%*~#IAZtmIkKcx-Gf%ho`Y^_xVG=>C6CBd=E#|<QH?VX79qIOR_>^G0n^23
zIO*+lE#v%$pn=*i5b9}M35kzo@9#;vOkkqen8!*xf<Va<+T2!866g}SqIGo$8w<;Z
zu>h9IvK3JABaD=Rve<ZdT#A+T^+A=}#=MS4HSl+Dh(7_O8QSD=5n*BmNnicl1TebA
z-QmH2_pE3tD_O|l<EIN7N%GI?%-8m39g3wfcD{cvsOSQ|XHaA9hWb7nh*w*EV7T+8
zyFf>^(g?Ku_b1EtrXL**_4cI9o!#68R^5`)(o!%nd;{wQAkO45i1YJnnY+-PI8o#`
z3318oZ$d_rjjt*{p=-s3d3sS@y&|BxdjU;h#^amD#MD%FZ1tmL9A2a9bia3yv~B%W
zD&Nt9Gh(KY2n4Jcjh)fM1FIVh+`3q_vtWyjr>oM<(?sEo!GG$?%6#I)NyC+|)L@0Y
zuG&8MDDMCg^p6Cl)*#souZw-C9SNYh-vJT`>{hVSdjoWpfTMHHBb>;(^+K3^-`CRF
z>}u~FSJl*v5x)kqgNLLY*jrW&SQMi;K(qs~S7@dYOrheYjL)7@IJ5?N>ZX?hD)`ft
zdB_Wmh>nh)4cbg2U7wLze45)Mp7)6UEov8@lp(iEzNjzey`J~*p<d~?W5thfDz-{X
zd5DwzOw120`RDzUVUmiaWsiivTs3&#5u`xm19rTksYWm(4kqG&V&?EZ=(cwHI7~;o
z?!1iQJCFhZ;<&o+H&5WkZt<9I1Ym=3)NtGcGi${n9*UtH&}6%eS9%L{>|jEJPyqZ<
zuDT3`ni;ad8ShPX6U4pT$B#dsGeW60N(&#eBi~!)Cm(vvb~ak9j)9`I_%0|*@lILB
zA+~Nh%jEPlsJIpI9>Ko#X*Q!(q`AgH#YO7;5pOSjEDi*B*;bbM0c#L+@g}gH`v43A
zY*6#ey7jxB7p5D*0-NNi@*LRKAys&)!E14#Cv{wlToa7A?vi!@7c>8;?dmDwQ>?6c
z2In)j-@VuE=?}m--I)PacK}vC3weAVc2p0-4CA*nP;Lm+mg>M@RBjzR04Q+ZnVD@;
zEDDN)o6(=XUQMrYm<AI-kK`4{h|Dzo`t~nj4?pb`5=(ANOSXe;+F~_*_W5_pg#GiI
zvvUoll=@~sMy{Z{vAoO|84UK}vnbEXkG0mrtrhIG=UG_fYU`7elkFKvNiUfpJMB*n
zgLG#$EHs=1DiLtNsT^8z$1f~G5=gZg0|h{{oMt4%$3Gn?Zf%bhFHYLgeZDU;T^<}L
zs8xmx$p>g<K%qzL#V+i&GNG_B7GE`Tx)det=SdDcce6@X`U=ZlTnh@eJ9hE17O;^8
zT^S+Uk9{9BuP&XVmYA6vN6vn6eQS#<RP4s`HJ2qRl;q9E!@~ni?jU5KhOSOE>rKn`
z9X%|R{|JFdcb}H=2sw366(pq1MvwuXFo**mvjK9s>aamDuf%a!Y*5i2+y<*mV8<z=
zmR{ThrC(1tzPzsQW+0MJNJy7q9)i}n0>!CGNYag;8BBUgkdl(l=Q@M7WwtT8j3>|r
za^=*M>e^a6;p6$biw_JWQF;t&tgINsT_%DDKD(YKc%Uzq<gmR1tGUQo;s~Zw%(v~w
zrl)iKc7^P~(!`~m#}xg{tQqgl8?&(Pty!Rn?|<`B{06A~$UGR;htdob8mQnH`_3vS
zlSQ-3A2!#QmpL2{37<-5g+;OOW^-Su1RI={AT1*-cb@t1wls5@A5GMh2RN_{=blmW
z^YbgJ5Dz%;1p)tE<$Fln4<cVnXVRUU?+JiKq|Bic@Xa`hSJyvjJOzA$PD_Bf*AqUM
zuazCqdXK0h$n5zUVjTeZup8~ANx%sKjp0e!YPHK3P;<DG!^Mk7cICDq1t?-L;MM6)
z>CBW#{2mb720G9{6i*hgVZP$+;}ZuyHdr8{<r(%C51rmILmdBGdpY9L+r+rInhyg$
zf=UHTVhC_4=3Rxr_B(_@r&ce*!qoF6tdoL{&Q8qlQk1E$?;fCYD`;yE+Jh6LGYg)^
z&DFJ=P)dPzV{MImPN&YUWlBbdyRXOVJE#uOdSS4(oswBe18Gcq@CFiTkSg{h_wbEw
zdLRVbcfaq;yJQaUIQW_8aZZqtF=H!Lk}?KcH2oS!2L?42l>(#ZW#1{lC|;VnHx@(;
za{j5_nt4w~Se101@`5!y?GtceHp~;YDa_#Szb{q>z3~B;4u2#aGm(IGr}S{AR!$uZ
z(UtS}1Cg56y@iPuL}XvMjhkJNe8AWaYE1moU5IC>2YG(ptL)SXG4~SB&@kp*CwnYf
zyXFYW_(uUCT|-Z<qU6UD<lw?4r>8d>;y;CY!Cr<VHE{_Ei}Xt;$dg_dcz{l4))4?0
z`la6l_sv7_DBtOpu%+RIfY~?VbG9Y<;B7D@x`9v_kjcSJWIc*K{8=0&rvfZfZy6o<
z*u0DwC?NB~bd$$4L;ftNluG-(?`|KdGRfd;5v%Lf&exg(>bo=R;9WgZzir+J<uo6b
zn7(@VVu>%`;Mp`-Z$Yx>rUO%pwxn$(wx<dMg}j#LZk_jr?VzHD{Bv;H0ZLa%r(lC}
zB!r}V{1!fr)D(fW50dCjdlO;%5)mLy0;d>CLJm288`r|v*w~UTVxSkbPLje_@8|2=
zog9^C+nqPQu#g-dZ@D=4<7r;9$P_M^nVZACDM(040IeTv9d<s})#9XR5Ft?pxqTob
z@EDZI2UAHjwq%fzlV@dRfrVB<LuK3Tf`$e$A>q|hvt~%BL&v}X#EQc=&CPGyj}d+i
zuql37SXlVO=5@tgO}iPUt%4LtBE@SOXR<?2B0I5<EE@=vQyCTl1UOgWH~a|$g+Lr>
zst82OR3iZS?5Fc>hQKS-9mH!;7JO9hr9QBK+SukPIkszppxjR=ALW+;rc%2qyQ8%f
z!58e!qAC~j<6VO4el!?yQ)>YdVK{HQE{@j#M9SN`6|i&I@!pS*R~$aNG3w_3Z-1O6
z4skxGc?JI?w>*jh`9Iw9|8Eb;fA0669{5iW{6E+O|24O~sPJzqoSKEBmAf?)KmS>w
zBcgXEJYGc`dv^<0CSFB*Q+Eq_3v;LI7I>19cy8{l7N!n(HwKCfhMk70NdrEZWK^}?
z>>!s~mwC+T?Bzftksd0ra@lPw{c2aOO8G~{j{|4FfADIhvE?+V{9<JKGP8Bu*JV4I
z#(``xllA+QPT7qE_x<_Msw5?Gt6RBTH7~-iAJ3P4rajZ;LFpHjeT^!>=upfdT+(<&
zK79MCa8%CI+Jgy|#}|w$DTziN&5Ug16jYh}*)8wS)c5-{a91&oaM9T0oQpL`JoX8T
zPlNf}>s*?4Jdyy!N(OcDgn2+?q{uaVw=35d-mn!eeCB!V67J?TB|%g}=gXibn)1D7
zxa<2!l#d*`^@ofnYC*3G&b_jC=<}e&+wOKK6eP8}kcp0Q-;DpE^-c1Eh#OZJlz6r0
zs28o@IrQIE)+ADTb^^z@@v0bhvvVd_%PLvuE1TMuo_l$X>7L6^Sm+hzKI<v#Bo5Zl
zFE}!nT3PfE4X2hFEh{(Lu;(jwG^9vZ)wuLVG9O4>h->h2=-025ZIL>g5)d%Z%|rD3
zOt&3%oA^O#QRnOPM&FwSLdz=flqnC}%g@uA4IvtFM5e><yqBiMLF4DVMZ#nfDLsdI
ze>!Q3OD_8%8Z+`P+JrbS^7yDn{hi~rCC|rJB#1)4JKMfgO)aR17{r?NvY&SlO~RvE
zH0=~RM<i0kiDUHsOK~%&&kfZFC&lii1g2dY9*vWaY857xod3G4zd6--a~R{rd8vLh
zJ_ncbMIz}C+E=IP_w<R)>pjljFk+Cg`GEYIGc1C5w)&DgciKQH|MDATpjc+bCVEhX
zNcEEziMxE#i?n&FEKhMy2KBv)^@?=3YO-XD%jKIKZYHVh;aqh+wN&<o^RoUcMs$qj
zgqvu+<C%i&e3YZ7R$9*CBz*~FJ1#0RQ%Wd*!}lh>;UM{7yn4{-ho44#hWOvE)U>;m
z<b)u|DO5#GUk$%-C!&lSzs1~k_ghU|tsUPGwXfqiS&*H&>foLz?$dgvNY@aR_U$C?
zXRp<RvMTAQeA{zI)~UzHO``2hjUL}>T$EpIU%cUpV|8|#KqL86nA`M1Sm?yiG4fu)
z+e~&&Nh}UU?N<j0gWWFfHwKx^m6FnsjwX!_+;e{vDxvi8WJzZmcfs33t!J6aZx3_!
zd@lrcdicp%iF|F(%0J_GQ<BwM%*!qZ!E=3#x$z{jK;foJ8WS5I9vQo(fCT;!Q&zv<
zWyPS7mEu?ZJn4j_)!Jwp5r|fe96^Z@4jUFrySraf)_o0%brCA3l8)8v8s|p2yI$;C
z$^FFl(Y;T%Lj6oa|Ft)nX@0sJcOJN!YwD}gy^+u)5~<|OUJOfdl%AG&fQ-C3Wy!Z+
zln<d5aSK7S6k<X5Zjqs>c!*NkTYvMD4J_Y3?w9svmzI7Cx5@|6guOqbNzJ2fPGx(5
z_{c;OSmHR&L_DV0^g!SRR^4X!aa<$++j9jBmAR%j!k%Sna;tJuRdb64X+QB%dba$U
zpWr}0Npb&7(wn8L2A27~+$yb^3K^bx4c^*BZblNvRyT~rc@d@8Ce?CvT{uRVl_%8(
zq7*lnP4cPZSPw`Z>!04pSt7!Zj9$xJt>$2Aj5asHco}m~<9;k>yR>?S7OunHgL+S`
z*G3V3S96F0@1=_*1V4Z4{c@(7IB%MDxaedtkFMk~EZ*bSM4n-sEo!)7)bWG{jp$vR
zuWrXs@9<kwGs!R0^H~B1Guu>#zSoYud!DCtz|VQ5tQ@zpAG1Wdq75TbC(q&Ci#mmX
z{1j#vG}>!aaY?JR$2FKYLhdbReofiPeB^SW^`bPX9808zvc0p}*LK?2=8LsSV_Z5T
z*Q<7X<L@GCDXhliznz#JQ;`s_4!+Px)>M1!{a#N(u?a#(JGUA!DNe{7qk62rdDYWv
z+@DQ3qeO*Y<7E!(lbOKw46OD@W&$H0A(^pa2Cs2BGP6D<HnoxM6OoaUg#*$_VFc5`
zDNkH0Ijn5nV??sHi&R9hS~PEWYN7d(SBypbY53pmO_vN7Rv{A_mE-&#ysE`sc$0fC
zY;}fEk6-8RY6SkNaeguU<=d&HabvSaapx7$-x0~n4RG_1M{l-Al+a(}-@+6`bG5L)
zAc{3K?cZ_6S~wBi__!_pQ#Pc&Gg?z3w@Awu64wanWYfNF&Qq=%d}=eHN$82go6DM5
z+9iR?#R+80HF&L3MHkc?KlRJbM{EYJ(x+(6ZWFq+XEj;g*vhy?6(z@lXvxtEr#pB?
zOExANICGwqp+RAD>-?GBwY$kJ@g7DsVTB>95pzcG#H&v6ncu=`j9uO^Z`+;>qr&Pp
zvRgR4BfXF)DcYXB$$D6oU?4Yqj!Vz3#;`>k>4nDDeXg2tz}Zx4UvuN|%gwpC8i7d5
z7jAD;TUMhaUq)3l5(|s!Y|hjfNy>ZRu-J0tUVRZj%jwpU&iTmhK;WI#C}K^<=H>V_
zoA!gzYY#o8N!aiA%@XK2d31l4h#kE3T9^3pmR@h|K5qE8GuxXwuX9bmPYC&`u<$0W
zDV5kCP^#gK#=W<aK5=f{KkX`G#zdNfu<>(@5X#YpzEE>H2SHDw^R^1u1yjwUxxpv)
z&h<Ax$wRy;Pgtn-K5ePDK-k*(HcR+HE(u0zrSIeN)VlkxuB3KoSNm$sh+A-}TG0u8
zIL7(4C620<gB~w<{T0i#jJOG!@&wMzz_Z#UC#~k4xF=+{Dtrgte~o3c?tbCaCsl0X
zhZOAR%5WGXl}9%!vi5OS&v^H;tS&6Qow5^q<5UfDV`-_U>rm~bJp;nOvg3ZWjnG=E
zzbuL7z3O+|bL_#g{D&B~o<0b|zUOu#CX#^3Kcf5ao>SUG0#1pb3lhA;>6|8;;2XA#
zS5^I5MK77(>+Y2k5XkAI>}WRVzD0a5ChmN)%dk%hcg4YAvkvyvHK(U0V<b9!&5_UL
z96LO}erMZ?ZrBgpYj|-pKsIGc*yTmQzE$ztiH8Lix8I7ipDWS6mGojs^X*W$v&{hx
zc~z6qKj?NseEk0^`J5Hy7yNf6pB{aK=s_V8|523WvzRoV5ovun9qpcgLXzX<EzV=;
z133;S#pqA;S1areo}bh#d%PhwsLn7zwMHVk<|4Az(b+xd-R<MOV(qdrb}Z{-oV=p=
z>Tx;^d1Gq%VC;-4eV?bkjmOxZA*nAN&hPL>Hb0}%uySM2I-D6z`!Mm+M7-AfV7sb6
zxmkla71NA^1NRsqOQ-o1=KVwks_y(8`8VPO^)^==>IWt3>z0aJA|tn`2)ebRk+LLh
z>ERgnl)n|N;-BNqB<W>mz-Jvt$Fy~FTUqeCboX8R<A*nOn7a}mVzH{23ACG7Q#4`X
zj*<Dx27XjJ$F$x^{{B`V>wE_dJNk5CdVPU*G2PW3mU9{#q4)Sia%%5OEfo$Y;~YDt
z&#!A}^WaW5^64~b_KUX7^Bar9AzoIbGtFl_+!*t#f;Jjzt9Fo>-&oGeSx!|Hzo+%D
zJIMH?+PJmMZkp#Mu=erR<ICNn9|}e1-IKE<>W~yMd?`cksn&|>j26Crog+|v?K-JD
zl5&o$sUVhPfd==`$v&d2O=mOd_WCn>*BxHZEA_ryXiR+N)05k877xf{`g2d#;+7S!
z1T=nq=Ua`n^wQ!*;)eLm`lqEkH$=n2vTliHvEIE1Ar;HtSwEc>5n3u!Rc5wH3X{K&
zjHK5-V|j0v&9bBDjC(fi9hMMvLG$pZTLv|^c<oZIoOV-WP;TfY%;|G!_1KAOzW3Tb
zb9K4$<1YCHB<_RDtyluoPDA%8=bb!l?$nEQY$Ymk#~(XWOZkl=nIGdoHF9w*Y+ByU
z=a}Efope@QT;aqI)?y4BeorD!IcP0XoO$ND@qW%xSn;Esbp1!cos`;Lvj-Rt@8OcF
zoyLic32QdVajwR<8k{gw8+`Lx{N$~gfg-m`LB*Dr?`Nu0WU5<5YBcV=<d{Sg3K!AQ
zndh?z)Z-l)x_XBHt%DBQ_nJjXde@f0g3kvg+vc*2;(?<2LF=Xy1oZC@)AiI1vI7!!
zDP$A$J62SmH&)I;1bm4}zS-63>TK&ZHq2arcezxPam?#eWku@QfIC48<^3lV2Avsh
z-bpXGpiH1=5r|&tV5nbVwC!`f!oya&&x$4rzhp_jx8>7hPrl`?K!cVmV+naQMTirH
zSJK<rr&b)4$5NH4*TRjyNIk^(AR0+PAM!eoKJw-;^VsomnnmVo!J*`fy=-sU-QO^A
zC(att^G3aGyn{A~!~I%Nm`sCYGaP46T_@kr^843ewCu~v-(o+X>(32sGLyN*$&l<y
zfL-7A@)+e<q#q_WgZ@NIwx@Y=z_rZV$<6t+w=ib3B@Eo%FP?KCMt<W$P^S=Vzk4*h
zhxM{K{CeFh)(f2b7K7647xv2Y=!$wp_*QHULkO{jHj`f&57w7t4;q@=74q|i8yK0{
zshdqw7FXOquN9(KROVfGPOrmuCg-A(h`eD6t6YH1K-+9M3;uB*-8+f>&%@QP6eg8(
zzZcW6-g|JzgZ}MqU598%8PYgPh#XgdW+!@Gg@%DnZoY%SPxW}tr~YR3^RJ#i>(S30
z7!xxLzCiXGjcX{MxDt=-swk_NSMZ2fhsUKLQd{~gt=a4Inw58ovI-*^oV_iUh{aUd
zucbY>b2{$YmM#r@lZ?UaMayI81zEQ)bTr$jF|n}7BtPi0S)8xbNnCnhZlmX3sblh{
zsNJq}`d!=?dROmb+4_VF$}?v^tvBBbX6?EedhOiW7Mr~E1GB)lw|7qNkC6n<-HakZ
zTd?gq|Mj}0$l0}x`KxzcbyqBO_eRrkD|>W_yuEREk3yzQ<W_Du%i~JN^KI0(5R%W)
znmRv_z8Fs^H?Hs)hHkB%Xcmt(%J1ALZKYY(LEvcZz8N`9(ALCx!VDLidb&OfJ19<G
z(-Eh*#Ho<(VjFv_7xqIpIg284&s2H>(Q^b&7iP!##A>TAvK||!6hU|9$|6QPNAv2-
z&1hlKEqM`=E<5ru>6?)Z_J$^6-D3z`UwigYqncXRM(WK>_LU~tJnkF5Uj!$`-L1)d
zAIDxe>#)Zt9~{%Vl16cP7*}#n*uxb+O4h=~LbjgxMO{Y{h4Ix5b~35gbmo<2Gas>b
zh!xT6Xz7?R#J|v5DJsUYXDZ+NWaEpxa9Csi%#lU<RcaG<>TUJsMKP4Du2SWO{o3gD
z7p`{gOn0>{2j2^Czh68?67DxA%YOUblk)hRqJ`2OtRrY@gx?T3PWBq~ADXJrUTyHZ
z8a)%g_5={MC^pjc+ru9Fd~aIyyAs&5J{4Y<@R~KksFySH>rH+NAv(W1k1|enZhS2u
zx<^A{__W9bNAnEAfY$Rk#Tn;N;cLCz!vl)-pF)ydY0>O2={L~Ke$ae%{4g$sDO!$M
z%wH}}Vfz-}qWVkA-bwUN_@$0vVyO||@8NsjAld5Y30m*l+*DdAHK`;i7(9We7FjW3
zx6P^RO_2Rm>24>>jX<w%J_VJF^nx!9hl&rorfxoc$EJD3C-F_Sat@uHJ4c};*WR;N
zzBMCr2btCrsgxyoKDwIR9p0uchkD_OHC&%+gkC!+jGxDg7pdqotZ#qwbfVa=u6i3s
zfmML_AGdXpf7sRq`2Ou}ebK<cX;6{GzwhZ-7N&f%1TwuMu=~>8<IYA(vytS~<>mLn
zv{>andPZgLFFumsaIrBcqom^;68}7O;@i9TzHVP#gX^D%)>qY=ys3ZaB(la#vD3Zv
z=3Mkn>X>Ac(Ty+pUcEC3M3T_~eml3v@+!DraxhD_c1GSEsTZS(zKA2o<(zg&fEtf+
zOGDBZb7Ge`{$PH)f+nTVhCMf{Zz=lC(jj%?PFHPLmlI>pZ3%mwud~AC=ZS4@Hnj@x
zviEcIK6H6xbb&?5pJ!Fg(t|Abw&wi{#!2HITD<Q#grc*f&Yf$sn!ZqSHAcN7y}*?E
zAqLqdj|JCLvd?@C&EZj-9C&#o##;vO>uZf4y%>(O6ZqygM~cArd2LQ$Dl6W`{Q9}D
z!6rrG4tjhj10jbQ+M3E-hj+u$lgAzeios})6qhylUQKFVbnv~@R8K-b`|Q-ql1p;C
zj<kI)E#txRJlOL=Sy`vL56YImjSSL}F@}+fo1X7%ysfhM75j185>J^OKAUpP16vz^
zo_3rB*K1qNW6W#3td=IBt_2TXUo9fWkz@RFu%bbX2o4R4>qIwOiEihAF^4C0{=he&
z)XFec39C0W*s9SBYf}*eUw<5Nj6>ie#eRL2#b-+Ghc3GL6KuJbgOBD&NMDCEXov+<
z?_8BrIYzm<vfAkK%F890YkEf5(zoQ4UlV%Y&d!6BZe;!`;pOPOhS16AYaC%sR(G!n
zE{^G`d!^41`rQr2WjwEBAHp@kZ`@`W-=oVRAB>yg;=_9T^6|7MS3cc9p1qUSZKB)B
zpVsF71nI2sIMC<utC08fId?he!VVLZV%c>bEEoy(7P6F(^G{d28mb}VLL<IK(RC)T
z=yT|4lV$CeS4Ng)Q47<~+paW0XdcW}R_JJ?mrD9*WX5b2FSMjE>II)K#+}LLPSLVq
zS+9FH&DA=#@f_RfL>ceb3Wu~W?C9j555FK9LcUJR8S(5+2c<}S3EDe;H<@TnNoM7_
zpGo}<)*{}>iMjGqgZHeBRa-XN`i8%XiCbF@n^QbyL|Zn&MYQg#fAdmp8gH!86roDu
zF{vB<cy_*g6}PE;&hXf_02>3=ByM!w3YjN8uXMnh%uP0B^(zmVBpQ_0-lkcKbe#}8
zX17@TS-g0HKb%*~=6vEei?;>Ret5b##YEWL9Uc$o1#Yz;Uz(M?+@V<EHcV|QaQfr*
zvPyZsg$0fUX;Mv+P`Oqf)7OIE^#%!WnO|48_qA;n36bQ8S)6$~@8{Tm2bb|IZMFE5
z<$=2iwuj+FPZXmRhnN&uGQ5jDw0hn?=I3>fq@49c(~H4&P^RpV7P~HDP2gOzA%QNa
zig@Aui8Yzaa*@078_u~5qfr`qnJ3=0CoFqvr%&`=6;Z;(;&w;cdpap4Vz;&gwu{)Q
z)f+3nyrjWB8R30ed3>sk(DH&7jpmiG+w_5_JFwSuum>D{>h%TtcobyO2aA0l2}e#A
zd>FCM*<He#WzO~De&2pQ_4*y|DP;NE@q#Tv69L2eFMG7_aIFLGS?TciAYQ)HO<+w<
zMt%yue38`QhTdIm0loD*rs-S)d}b$4Cdm=%sfTMMjH?@)#}JnrxSPE=N$XP1bJs=Q
zH-P8j*a?}JoK26j@;`HU>);@7rpG_$o_GIT8pnhy*+PE?vqhD|{v?jRdXw%$jE@?0
zB?b+KZ)|LjbMpoof5z1Qc;N|$nT|X4YXQNJkpfQ}H^gW4BKHg_k9`{)Ll~YGxmkFk
zU@oFv8$W^7AhbQDCn>%yk<!T@>sFxJbmRb@nk84K)yLO9mJg33v0|V4+s>al#vXQi
z)7?7P06ow;Ak)z1R7qFiiLbY<XxJ06FUFo(eOe~_9xEMr_f35O8YQ~z^M6{-&;IU#
z`)Qv(EBNo*ryuF7x`CGVVpnfj*VeZ2M$p&NPpls7>Yu{JBv?f|nH@~kVk49XzXXgD
zV=4!h+2>v$LVwx(qGWfm*v2)kb8@)<>62%XRZ-IyF1mJjl+|1ezIeYoSmi-s+oT(p
z$oTbA?W7#W(ZdSEtNu5t^j#G`NgW~^`1r40_4hrT{xq>LA01hx^*ZV?v$jnu&9*WZ
zeHRU3X3MDXlXJJR)b*Q0CO>k=3a~^)Uayzf^;ERJ8-SruPA^#+r4{fX6zeYUnJP8*
zh6<~IFNYl0@0ZXkaI??^ky{p=lSNjed4+MatCU@@2%&LW>=^cMEGxf7Md0<#w1GQb
zHgVm2aHD6W(&x}(=B-NYw?)5nW2b@*hnc><)R^(q#hS{MT#GI$e~-(8JuhY*+{oAL
z`gEwMDb(l0lS-6z%yDPlJn<fj>>H@G(20xV9eZcd-zrav^(Ev!>rA56`jYJ1?oB%9
zdty9$9|F${U0j$_GtI{i>3CdHdn_cY7C-*>LXhi>|F9ER!)JwWuX}ym^aedj>EGqB
zevrQ|W-NPdPVk$<ph$+wV~@d+ng?p<Yb!iTw_5xbi0eh@%YA&zlQTD4-ZRP9U*oE;
z+r+HgN@L*4yGOhfeO1Rj+U@T7+82cE$<h>S$G?!*uGLE7Hu$gJOCckTTCuk_9L%X1
ze6&k=F<0dEW=8!J)rLAerI5p8E^az{S0=LGWhJ{_m;a{VXU!NhulCq(`$K%p&0%qh
z;l9n0cH{WUL7m-VuKk^B{WmeA@<q~ai4@1UYrnUU{7#n=9^5z`RXQHQqz&8F3r!gf
zX@bkFAJ~I!3X+?$2UgpJ7JH8iGK#JxxaHt}m<)U-_EvN?wo87FMTxTx{S&hVyMo$M
zLQu3=85LRoX)?Dw<>tMCPgR%YXO=lTFU+)&7YQ9V<>gOxj3}YbY~iUfqthKMu-?Oy
zcNscM*oZiw`Ea&NgNBRM)G(ld*Oby1`MBJ>E0kW^fyvx+%7gJ1-P`^yLGh&PVh&^B
z-&a1T8PSZ59qX}_r`2%%y76+e&0=k=XKb0_rXB&4{`a`DnyQdjry4d44h7b8C*?z2
zqXX|{%kK0O;@(Z#aVW5np63?Epiz3RAhWyo;1t^n(J1*xstnj|Jj{vIwk3se8XR{u
z7Dig}bTHetl_d7XoFAl=t<7B4kU2(f|5%ee<YXHkFTpmU(!F<+sbOi?SWKl;xmgPE
zSsA1zhlkwA(leMlMJQQ|9b3n0jmYzvgF=PbBh;`gS+Boj;aL-%;<|l}BJ6#ZH#uIT
zoQpQi3(dCOQh})MP~f3bwKo#UB!fsFG77z(n(75~owJIZcM7V<&lrxA(AZA+l~!RD
z+&rBdraZXe6B$#a65vdLg?pT^p+kQUm%_QgYUyo9aQ(&wMO<_8HRP!-_A3VQ8F=(G
z=U=9svwK-X9-+3(jCq!5bz*SiH1qRGc_t>Shq7MgM98%4@yVDi-s#$RVX`<uBX_AE
zNF*=M3SMelmwsrsqV*>FMHHF;H|!SGvR34jR|0RjoQtIJ;hqhN?dPn!M!O}Y3mx*b
zg5pMJ&1755YpG6zG&ZFdxT=ritQ4l*XT@tf&!3q1aA-1P7HLE;ibdI8&95d@-6lQb
zS$$mUB3IudjQ3{u^e(s68DC|m?DUYiBuKvO>d`zYzaG0~()#?B-~~0I2F27H!7C}$
zaUFU7#4$Uc5CPnaXR&USn`bmHeVgZ#8_OF!?s4K2wbzR$Pn=!!s{E`&B9ohh4%}aB
zde1(OrnNicDi7QcQDdr?2h|)ZOokRcdnDI~dpx8VA6+CUisE8k-hCza1nFme=X#w`
zc_tm7k;Ef|*!!=PqZ4%+%dIUvOzv9gdI`NRbo7%TyIWs5dWW!T6q#G2BOv_F7;Pt#
z=`6>$FqO0y&&GS}ZOC&H4+WbekWK^%mRXpiwC4rBEmkHGC!b2~q&MCzK{Vc+8c$8@
zEh*_(cL`XC=HuPi&BVu)YmDw!dbObK%yLp^^pVt7OY9+nEAlkoKkXhM_W!TOZ;0^m
z;qfY4*jQP+Go9r}{UK}P?xtzsD(B?j?Br<S=*}d9$NOt15*8H|62#+`bFz1Gz3gmi
zZo$MWZ{cZUZlR?l`=|3%UQYX8T&IVh562KBDvB2r5oi#BqYOWY!*PVP!VMcs1VUXM
z!3F+bTm)LE4FVI6P(eDhM;>AHbo5_maB$NxejQ_$pspTH;LKi-lQYrO(onphrf^g>
z&Sz6+XM1!e1j515-4z~;N#DSb38x={j-Wx{L#_Cx=5Ee1nwn~;8c=0X|Nix}*?aT^
zh)+EKeEqMb|MnODbqhChXaXw4$mzP9xdR;EKp?PE%$;4`5eOU#I4|Pm?u<G{Jp+X+
zJRlrjLmgZFIEHHo3bP-_VMk@Owd9~o3LF!eTbo{o<3c#*Hv8?o)o;f?o&icgJxUgi
z7OpntOsKxlbak?{v9~z7@Z;D2aQU~A|J|X3y$7`U=)(xV3Eb>1tHQq=(DvZ#rt+xc
zGjLpJZmMt@j#=S&z|%(e;?Ls)XLngGI6j3ypfh{e>Bzt_%m{R4OIJl5IEH%AU9CNo
ze;p@zTWjmV`QvcD$<geh#?SA5Hg}VQzC<v?`2%Z<3#k5OKp-%fY}_ws!!gu@A?fO*
zg(?rvfMI#vLS6xmx#2j%&Pf&Z3{WRVrJLtv)OpnNeD=1MLp>+z{5xAy6-_vXxrs4h
zVXuTL5BI{@b9UGK`3#s0j`kN(_l0|5%3HV{wHNBaWN^3ER)+I%Urc*<SM8tQ#f-7E
zQM~Z8e3`YYGU`0)y93VlM{^XO4GZ1XLkrb*xEIzLQ&$BgIEFg0G%Xx;Q2mKI_C)BS
znIbF@P6#vjZH{n6>?4>ES_mbCEd1+?aD^+D2pfbwoUwqb7I4M_{`ysh=jeNG2s=1$
zgYZ1M<_5ntk4pad=C86$2rD@2h?qSpYmSg($VPZTP1cC7aMt>dI++l1@RtXiv-&UU
z`SC1sKkISAq{8IGJO}@u$5g|Vz!bq0MKEDXVM=34V#>o=QOxsy)U65M`PCxS^UT3r
ze$?XvceFSvrweUwgf_Y(OyT!9lyE}0{ZYeTdL@R|gX7ODw_HtZ%s)gg|JGMF@J+iv
zO7Q%pM>YPZKF6BGn#JmY-~Cuaf7HQ*HHb9?KmE`*f40Tx&+9*X88uQiFrrR>DfjoW
z_*F^<T55mvO$X=&8)&n|Z*AuJy^a>&I_f{~)`MAX^ogLw*U~YD_Rq7ZUa$zgxOowA
zi~G;Be0_Y2d_8;_e2@7SFyb&OFzPX$W7J^0Loi{y#Aw85z<7i46r=W!^8ei&`eW=7
zs3%9wE7WtNX4kJ4{@&LxV~J#m=!jT|<cS!G*of5rD8)oXMRb-(nTQ>((ETxPf6i~F
z-&+p-VgPsh)j$7yFVwqY^G8iE255vZ18n}h<6q{04m`09!VBtjh5km(H%9~#%KrZ5
z{cI5?1Lj%G3xAv4BB(k4t8Q!sY<X-M1QRwFwkY;lY!%e;uXh=n9j=MOZ-w8V4d%xm
zCH~S<3wMhfsQm;X=j4oZJ=*DEqZek9fvwMi>4KyA8Ez(1dwZs%OKwcA7H$@<o)*{7
zAW-`z>U#*pn%2>Nh<5zLpXc1w5eSKIu=aKSdCt%SfhcrCAgBcXJjVfRC1n}{Q7vTd
z;p+LLKG>)`5C}KRbNoktqY=o$`ZaQRxCZMO&OHQTKlbo&H}mjtKMU#~f?#U<!<PtB
zT+9+2Tnsd31Ue}i1}WO%YXodHXsGmM`24IGfrgHOiG_`WdkhaB3RI9F(9tk3&@nNv
zurM*9v>#kYV3J~yojfapO@7G~huMXKKQJ}}mqoU;ky5K`nf08R>#bvW$Em1kXxUD&
zb8vDA2nq>{h>FR{D<~={Ur@QMt)r_4%`?AlVQFP;W9#Pb;pue)=^Ydtayv9EJmOy5
z{Ri<66B3g$vmR&X<mTlUlszl2sI01fUeomIb@Q8+*0*gRKXv!?_I>Uj7@znuIW;{q
zJ2$_wy0*TtxwXCX9n~&01jdhM{n4_&wTl$mg^r1dfr*1^7aF=3s^O%VSSQb7lgV7d
zF?Atl<`2ZBkd4hKZ9K+uPHUOc%(V;eIIF-o+X|{_M=ksJ8g}b{t7U&Q?9X-$!e)Se
z^Z{HX3@i)`3@k_=!@|PG#YTN_@o<kmfW~w50k9PK5TQQAe|!$15;Uj-6B83Mk#R`y
zkKvR2FMke)L5y@c96%6YpuuFqAVo+a_V+R)I1vB&|Nql{fN|LQkK>3D>%_QCN}aVG
zr>=YX7U3k*EvYf>wB-_CtV4v{nZ!dxV9X(6#beLi{{!PQzeB|0<y~3ZfMe^6o>D{|
z9NkWmJ=}!O=~rH;eASbC|Lyri!-atH7XLGch|CJ}L&UAbfF0(WQtA8W4L%BbSbcO#
zuh*Hs_TW0kLV421r_+}wg!jnUH>0-YGK}`5w=aJ=?L=-X#xK~?zrNy<JFh@OHfq^;
z?Hfe4oi(UAa|N-lcZk3&+ppPkSuwv-v!4dQQcgv7e1`QS{}+4j8P(MLt&0W$=|!X?
zL@5H&lqyAnA_5`?3%yGXEud5l1d-lB1wo}q6@*AfTIh&~bPy0iKtzE+LJ5Hs?sDHd
z#@+kueaAWfJ?<U%!})!9$AAG@S!=#?&H2n{mN%)w?#zOnA=44rHcscudf8XQvo+@a
zoBs85<fjI_W^}+hDSD7VQ7d(*e*NBSIx8FA<f8&T^H6kvEykVU=iW;057YqHJQtCL
zG)q_`*5SANw;3HgM@h(=Nk+zZch)_tjcb}=ZgUCr@4w_{uL)fVaIDz_-7(<(pDnta
z3U3?TI99t2OFWFA?PJ;eQvB}sZ*&joE(`V9)V0@5`nt4Ftfj6m-g%LAo-<j?Rrx2l
z8+``3)abYa*JU{d)s#Z0=UPy;$fZ2T9XAXsdXxVcbaY1n6#n-AdNw+P2tnFsWCN-n
zJ;z03rCa9WAgx*Nw5b(XdHY-aCOY-FfoH29BuE7gDi76z^RD%}M$eV`My&mtWd2I$
z-<I`JUR8-_<MOUmdu;cKF2vwZtLFi~H8km^MdhxcWgttnjyo`ZN21Cf%#rwdB;*zZ
z>qKIs#oDVJgXqs4gO>jgNV}aA>Yv*T%CyCXLRcV0mOBKpEVAno;wt$r<!n8vsVwwp
zre}1){>NOBv5y$`#k}tL{In}rqx3SW=qI2tVxga9FCnZxx^Hs6lmX|@0SC2k{f?4-
zkf)X_z$1}dDf?Ms^3+|+9XJ8HG=O^sD8yHSV%iFrlT<UKH|eL-tW-Z(;;XgnCNkWB
zD=s+y%Un{f@+9bm&-;0*Fm2GWUhy><N?{w-m7B9IdAUFO#Qx?OiyR#37yLCT^LGET
z(T?%=jl4&CVB;}hZ|~WUtHLCmDhe`<J3&E`6O~+j!oK}A*eYA2`U}z>^?A%gXA}=z
z-0U*)l*5)IMN3H8JuN@e_!BDTytug8H{d!CUou4+UGtaoLQ6oFUSmE2hb+x%^ygh7
zF!gS<i!aYROyF?7{jVR%Z%HQJ#kCo;>#+xRsU5B9G4kH|*Vo-iH3e<s25+3jx}2PA
zC(J!4<>c*?*75YxPc6N7i)gG3$Ds52EAXSEamXH)5fl*(YpKruFdtrL)E0Y$kwScy
zw7%!Rt|2QSj+$A`rkvWQ-8zDhxz-)OT%ej3oJf=@IPq~40mW9`-mhQKu&8fsHJvYv
z4gHz!d)cR=MuG`-#$pwzd<?>yLJo~k$;Y6LxcFnxUX={}k>~&BE`<Jg?7ud_haQ05
zTObDy9FX53w#T5+Y4rO<+Jz3r(Cw9h9#q=n`~--H=!I=7LgL$Nl1+6JukXK@*U4{(
z1J%Puf<nG+mcNxKY6BF$4O+r6C?1KA&p!rbl+zy>wzbty;H1&$9EB?1jFK#CL2*N2
zj;~(_fG9c>3bNcOBeN->QjS4OrHv6%cq8uL=y_eKNvw?S{f}8}HG=f1f;WLHo~Z8W
z%L2C45<K@Dilh%k84;v9k{$;`ebrLXaSCD@*sKs)FSoQaFM<PTgO^d$N+jul&2!+*
z=LKF7(uAt1#wr4RZ$PmI&-{r|t2x;PqLr)vht(fUFNx(+=6c<JPF$#tE0hotl6;w>
zI@uvZF>LWvoj+s9dkHin5TZ=KI^P)`Ij+lf7i9Le_d+Iws|pF{BFT8*JeKV(xMnrF
z{=EAge5+<FYH@nCVBsKQEWG@w@2`t%UKLA$OpK|H6#kiIvG-315bQCi+p{E~43?|E
zp(LE2Ws<a!pfFjR`XRw5c&*@3YAHGh;(?A91CzM$-rZ<69ZkP;&kSl)q<uiU3(c4S
zm6ggtS$Ex>&LuiMa7@bxuH+cRc#)bB^j3F+n2TTcp5yqrXhsgIdJ|w$?^>*JWOnL9
z@k5=Fi!0L3pcNvS=udsuAx4|Puy4|&C|7c#l`lo@eDh&@Em99^JIyHRdP2zRjB<KU
zHIp|1jzz|Fs8BQrT;0bYp-$;>gg#8_Lma5>R?opB!@SJvq`)gV21p$Y-Vh?R(ZNbB
zLAc-}j&AVVOlSK~=;yz1{By6|uxl^n*S*wFLVxtYHncf#lfVEf3fyoIOXCRT-022$
zcm;TFD?{C%-!#Sb7%ECiOI*0ThnnR&MT#@PXUFYB_(D^0@~Urz=e&nn6I5zNo&+iz
zOSnk0Rv7mo)xr3?m_wtosK0M{9AvL%<6wrEOL|g8u>^ecIfN#Do^h&y%D*!=%Fl}O
zAA5erXCSh=Q&ze_vw*AL_a2Rvk?cpCf}B#zXQTBlp0>Thbn}74CvRI=HKJ6~P+_5u
z9G*(AIDS{>gr-P7!!%QJ2jm#kWsi2FUO_w|Jcp)P<9j)yLO#`xZ!1VFc5-Q%N*0es
z^~!hz*`?D-CLV4Rd8o}Z#ItCEBOt@XMb7r-@qIRe!uKZgq<bIRo+5Hi6~5`0w39is
z9xY@@-V#Uq>0BZQ(E1?tO)>jmkbUUfw{J>3Hl}3@e_Fo#DqAN0VTTK5hJUP6uAv{w
zNPYZBPxN5Zn`{&6LTn@1(NxDbQ&Xz`s3cvnHB}DF_jpqt6p8`WE2Ct3>C?YFy#IVB
zA2$zZ4EIa~BWZeQNxIWiOo<0!)V^c!bC^Aw#r?q9wC&VYA&cB!=0Q%M<_fCyv%0BO
z<SP+F;1g)~$}6TFtiQ)$MZG3|@bgbLUZWcnI(FF}z#J|$4I_BXzimD0i(sJ54H5*1
z#(+tQO`DUg52%2(#Ow)w%a-bdxVK*Q`)D^W1Ue^~ynSaXofqMh2s%1xYD;5#76dy6
zvG@cJGj?>z)LJ3bC3gs|U=M%yg?Sit%}K>!fXt8Vx}^gn2rSn-)}e7q1wDnN#$Pa>
zDy4TZn}Z#3n)Pjs>G_Hl7tF1Lqe#~x-^^4|E&zA&iS|WI*Gd6lA5>0aGfM{zy`NW|
zZXI!LZrHtnAr>1^J5jKSV~{)gJ7xvbXNB~#r}LrMFB@(7R4-4~w>wxgrEYbeZAvdG
z{*w0~%sYK)4gOVYlxeo5qpH+X*nF$CYQfN@3^qFW=uv<QL!zMb-5sONP;+yJ%utZ!
zQd1v<bOGMKo$PpO25Y&gk3HGuBALLqM@0Lyw_A}^4E??*F47$|nKrL5wslx^v;%kC
zOu0TuVIiM$fZ5z1X2?^X+N`&>=bR>M%QrntP7Q9E_i<jZde&9#NMokrk$U)$&97)a
zmnM$tP#W>iuVyEWs?vO&tdHGg<$8qAYm-enbdg*s6m!CB8rwO%%B}2W!`|>K`&?W$
zs5!-=%pSwGrbOLeZWEy91m5tKPkLvbPtN>N+SDY6heC+06N+t+_tw&}ZQ;e&&bLln
ze~tb0?3`&|k;YfEOLSu$o<p2gqEkn#zcVEa#bYfPxNlr#Dp#X?L|hy5MY~`~7cAcB
zoDlBOKL$lVyY(rRx4Yuc(Q?0%!erpiq^Z28HlxFeD%}y`-^^H9#|7FDnu8*-4381u
zZ=9Ly|3C~mxwYVTzxSg~B;$jNA#}qCk4j{h5z-5sDvXTn5RUllaLPBN&;R%1K$`gU
z^ygbO>v!jG%$P9TU0!4^deZ{Wf7r?{zH}042*~vZa!`Qu<QU|_fIc#o3I{wLLmw%q
zki<sR#G7ut_Q}AIN~EVyzgX!<<af3Dt@gFgS&brF^L|A_WGUWcD`b}i(yetKd>Rqd
z;GM$1Wmc#rs%`ojnlYk#=hOHUyI<!S>D;oYYj1?9z0~JOm=agSl;g=jOk%E7;e?`%
zwfw#a&+<K3_(}GRIFO%(g*A6N%MH88hZoXt4cK9oK1eO3Z&=&nsYfiP$L8e@6DTIP
z`(#T@|C0|HvP_efzn~QEE#fkiiDEsGGZ@82J#TJ@7qhD7ix(@t@#%;aM4bf?4*e2z
zKA*UOAJ{}<=S8*p`*g0Zv`46CLnq!QcJZ>#n+{JXeU(^vX<(<C%*=SLj_WIimEs0S
z=QG7)Y=u;WCw7}SaxTbzYDm=P5S|t)bQ#DHE8zHAZn)q(obmaAk@ROBduKS=0<eTO
zembHnjNa}1^SyPjG@si))g44)^(8M~;koCv+DRjy8XQL6F-j1Y4=6VWxOpd#7D$Hv
z(1==#_BjTLP(n%4C6v>^8Sya0C4ym>gM{(+OJ~osd5jIsNdXCio+TXOZefNDTHKst
z<QZVIc8)>M6-nmDAg88d(0GFPZWE2o1ONP*E<avrK+cFkt1V=p=Kv5Vrb|_M;cDAa
z@7PU&*>8_osXV~*g&_yC5r9GX2<-cx>-BWIA-FL>?WB%Du|WhIgdhp>u4fq;&zT{T
z-Hef{diEpIIcYFHLAp7}sf%95Y$&t$CHMjdoW>Tuym4tcMeCcr__&i);>2e>t~jxb
zgh)c|Jx$vh+I`}VzN(0HhR#lOBJ0_YL5*CAZHDFg74u#KTJ4?FO1=WdiNBVVi&n8I
zMcmmZ&$@NLgx=hy0HGx#12siQq|FP%SN}RQXXUo?dYEqi7tN=X-}F0iSO|sp7(~n=
zsS(^KuyEYVLiwze0A7Ka)3b`N-&++|)!W6y++`^QZE3rROuB(<s6NyhgfaQHrz6{5
ziV4wpyl@6_Pq*sCkNyq#1rD{UHIMYLog4WZ*UxercP+V`N9RLHvM8J}9!h*i7>QOx
zMW0$6mX6!p+T4a$;|ywl+}M%K+TgePV$PL&d*9A)>C79XDdvQ9=Ef4TXDBBDb{@Z#
zxe2{%^G)0oggg(aJ%~qorHBwSW`f|6TB<hLy@L-XWA;BJ`2CLGecyV{JwxKqwNnsY
zYJj{Ct~=R4=&>S67kOdoUq(MROr71<#hj48^X^7A4>RYvkYso6XlzH@%+fK4avDOr
zE46|=A`-x~tO3Vyfh3BW#|nhi+ktyhI&XT$uC}=**Vg0I$eEB!^|!_}+4(vjn1A&5
z56}b0VT2${<q#p-d5U>6=@OYc^p4tlmn<WRDX|}mKODmBbcpz$`W!8J*Pr__`uCm>
zuyg<P0omoRbIC<IF^J*Qrzl=!W5%kc*lpLjHXQ%-=bajfdM{f7Y7|f~$2#cddq7PO
zS_n)@w*5)P8i_@*bdQ!(ZhVnWm3#PYbRLooPv87v=bUgM?`6&MF^D9K+_YTUONSEp
zMJ<*u25{!Uf<o<5ha>|8k(!``0<J8GlOstMk`$qZFaV#>)rJMPAQL@OAC+O|C2ZLe
zs;-9lRxA8Sf6id(biZ7i^#v0HZOC%RWeJ@q#zsN`8g(6y=>x|GH)%yYU4{x;*<BT#
zgt3!LPnGm&-S}4YHTx^P&6wl)pg8)&I!NYSB1mrM8I28iCC+C6Cc#DbEx@BFR_;gq
z=h#~dNs@G=<m3T4Z|CrP#~?$?KYd6kWE~hs2Yl^GQW(B;rpK`k72Oc-(v`7bXUx}^
zAL}-CPUZ9LE4JvTUFTY=x8yLts1@j?k2cBf(=%PV5_K6ZMjfnf5`^5b=exXgM-sr5
zLJ8nl)%@?rAhXf^9C6y@jGDccg0i~vFQI~0rgHE0&pBIKNIY#m5e6ba@CFMacL7)K
zw#~gB;$zTQx%~Q*?`W#}01S5evhG$1a!KtFR~pd+gc4>-^AZCV4wBS&il{*QD@I)9
z-a9u+=LF7vZp`(O<b7~5m>GL5;8UCpmDDzdP$A3EhI0F|BYhc!9ea@Mm<5|6m%v(U
zTVwxmBY%2h)yUR255GbmgGdrY;QOQA*&u`nz6V{7kw@1ZWC?vAqNcZqzT>mj_Lt^-
zS7MdLma$Owf{G&SZ6f=#UxPZ{z=JkS*6$OB8?zt5oFkX}d;+DiokePh#^%-SA8!yw
zALZ_1ND|rIRPANU0f-0!Mu_yicNZVjrsOMm41!Q@sl~kx8sZ;RW=PcIs?1uaL;Z1x
zgz%$XkfaL^g#@O7EDQq#_EsFsPO4a6*|f-gai{%*vcpLk1wY)XWS5G@mk*rdYq0A}
z`QMRzfT4oM?59|KeH!5A5uoatoI>dT`bo6k0C@$Cw}D67NXP0tT7sl3ibc<^JDfLp
z&0)NJNIy~*pEu`|iWtvSK27~TF^+bP;HK!IN$<&LLqnGNtAE)W+TjyNg4e)TW@cHx
z{Xjm}8z2;LMFY`aPJUzd0C+BKKumQsLf-!Sk2%lBLG8mI>+5aZd5kaJ_;LSDaxe-`
z?KlR(ZNak=Hz(TUux!j~nE1yON6&_aPm8`)3L466Io*67i|cnUPrazIkcu=D*+tTr
zV~#;mhRYt*ryl?spkNXSM3E;#$=Ge2^b%%H-?UaE_p~kS4`WZiThp^U6VhP@Z9N6n
z0^a35@WW1WX{2@%_%z_@d?`=LdR90%`|7=AC%(l_;V2OIRoj{-;Es&W1ZOi5s?&%F
zVCUZ=yzuaTZD=G0m@e-!;kTt&SQBotiE(PJ%?p1g-4YF!T6*{e@wppR7P{E5g=B^-
z)gbHQlB?u#x-x(;>8En;-`=Ko{Ih&bb7!?_LH)-u=rP^XA3NR{iU}S<1XEjd44nzL
zcUGVzpon+2BoC)yktLokpCYc%8=N_Cp|-A3V<~xsmyS3SPtkuzGU++W2mDWPmhkNq
zU$xbR-ke#E_7)+@jGE^$=90^3dNn9XMgZ4JzE8P6jZh+=4mI`J6P1lmU9Ivh{j~Ec
z<4jYC0FHQ>z17$V&-~L5aoVvKQ{U9{)?+z;lR%)yqGG|q)t}zk&soPI;BWeG8GRdw
z+&octdNkAfu%8-@_EV$AD^V=*=;fwoC_mNqYZQ}Ug{cRVn@VRL&tLcSxRfBks%sG_
zb5%k+zJRMx#|^-OG*%p0>CeUrCOH@JcqI00;OX|^((Evp7;sp}I85q%H2*n$?TVj~
zERvQR+5bxX(@An6u22(^FgV+~E~tgj@b-;^Wc_Ps>qOAUIad_lxZvU?NwTxM+XJR-
zi8FgZe+Tg~HA|N<V%XyGYXsj|@q}S~){n(o2QwN-FWc~6{JrCCF7RBEOy%1jYl)zp
zw0RTa8lGztSvYI!FxwR0=2q1|4m-MjH@=1^6YcrwIlD`o|Apqe-#TLiE!-&W_>ln&
zNp2e_Qp_cWkQ~c4#IKHpRprhNnbb#FP2$gmTh(WxoWAs`Yvp<OAS_D^$o9xCOSsPI
z6&0!JByd7?ebY0|du61Dnjzij1>=~?tbs7W)Z5-iFQb?hoKmNOW^lp#;^Lr+iXyuE
zoFMv0Mi5dL%!OPk%da2V1c-Gpa4fQMCPvJYB@HKF;+|#m;?&?cmPfR4=F*cJm2b{y
zO2DW`cuf4Bl3*zEHo-Dh>D}aBio>FIrAfJM;G?U`v4;14UmcEk2;3HplZ0HB-YlvL
zU0nVy`lOSmtNi=70{AHtRe$^I1E{RAjvYw5u-H@&vP+~xd_7N|niB*)xn8G=!N+<(
znz!NCob?D1wyWTz+#Js@1O*mw36atO%P3EZ!@Cd{2$nr*@fyLy_@*A^#mf2lS=o*k
zrLQJzx}SS!7q?t0z0FI%%l%2M(}ZjS+zc9Kg&Ba*>4->2Q$+pkH?!Dm>DKhPzI3ly
z9F|>PYP%3WaQV)7hX<)=NmP0{jl!3C&}l=twqi2X=2%wpQ(n1tUeat()nj(-W_}g&
z{%8+J{iXT&{u%?G=<9lX=J%aIOdz^ZGp5}hydjMhNj5}CliI!U&xv2gqZ`NV4ZNQO
z1-Rz=m6V0`GQ7>?OJ#AQd!{bIQ1&(TvButYW+!G%rgo$@;EC)YoUp7s1(NCVXD9PX
zHEjilZ!kTPJULl~;soO@-XDW$?@>SCks3|H`JMT3!YLCu(k=MKH2LHKskfWU4~+LE
zN13W-t}HLySTn?zHdM}XE6V#+dnAAS%*E?7l<~vWQr|AFSLu)&VFCc>p%E(29v$U~
zA_^F*4upMNSA-%I49i-)L|DM}zxKe|tAG1C-#>m^k+Jvrp=KDLl^(xjPpkUWS$?<Y
z{QwUHSnFCEcyud5enZQCF#$UtI=61?sahp&-WZAa>e}F%yMKpC-u3Xs31h+YZ9O#R
z3H-JJ!4W4>MH;ehGRekABuEmk&ghyi{CE#~@#Mn|bMUrDIcoic)w?|Ds6ZaWvlqVf
zT=)`t_E{H$856dtTZts^7o*WhePT^&eP<}$o5F#O>h8IX;T0Zh{=s9g@GCt3%RHHq
zb>@9CPq*4m+9`@Z@Mxb&`M2STkHIlp6m9JEodD68tc<ccd4mwHJ@1BGYX9OVCj4d8
zD?lRf=tW@yn8K^9#Ob9TumIELoK$;FathdiPvl;2@Ig*!-tuFrNS&fOnKAtopq#Fx
zvC1uj=L~x_9GP^LZEfaB-HTIOn@Tv#LDL(Tq1*35Bi2oY-!7H%rE7dCiGWX{;*s14
z34CRj)<7>r%!{Suw-zr)Kv016x%=EF%Fooba+eoLy@~;4${qHS9KD%<l)_SNwz?L(
zx3?cnc6uP|nHOVQ?I#%MPNz_Mx)$fue*RxR@qF1k$|V41efi>^INv9nvbsNdFYS_Q
z^YVQ|(aH8vgqX(JD|!W75;Mr9;=e%%PHkl)pDCh`#_Y`|a$QjrSJ=_9GuWB?%Xnl~
zVRFx06Z8`vap~zYdd`K$I!u-?c2FU$7FVONC0S~(7Oi1T{9dk*pe@yulvD7Qk{|3#
zpiriLLmJB%z(9};$>&1_{{&;zxCE@amN{-b1ZBb+1gC*P8Ae!H;WUGhyWTjCNBZMB
zrtdJJA-((=m_NC1cx>(FzqS_T3f(U%qMuaG<NzUR$l!?Y5p-0P4vR*-jyB5}MZcIg
z&2n5S-B{{F=7~Ei{!BRxue8BlZ=zVEF0A3g0qRSdc!T8Aafu+wQQ@ss)ZUm@(%&Vv
zg|&Rd@A5XipIKR#Q@P;8*8&<ht-pJh2*BE?Hx3l>u`MYr$)m;^?0bf|UwI=u#_l8E
z8%c!^BbqmC9r1U=uN{L{(kQ&rJ&yF%(wTA`x~B9!WAYx_V5`uwEc1KmXOc=74wygv
z(G%T{M#7*ZIZK>AJM9YTT&ZF<HeVeoX=4q`BD(f<8I?{o4QevQ&TVSAI=Rq)L&L!H
zjw|qZq{p@hZ3x^@7X52X_gvkdY;}S6K8+iXcs2aLEQd}WhRGxkszO?Q_~we;(4}Aw
zQiuUQF0OurHPmlRUbRYWPR_(6w4X=jtX@U|krjj*$|uPSbeppEL+aQN&@52POeDO@
zEjH**^W6GrLzt>Ovf051^SxkS6Eq3P;L0&*FdraAoT(;<1`Y(WXT;p%NopDe>>W}*
zkMtuo*>b8@os&G3%@Vo$jpTAh=C(?GUDKDMSXVrfB!Qvi>>-KZN%RZzLl&|*<r*H^
z2PVn#mqGfeMza(Mei{ugTFf#owW=SMiu!Twy&p>P{7W~8cqV%^;XD1G5jI39a1&_S
zcrY0MQI`ky>;``7+gc+((Z>Akhs3tDbNjv%x5^5fHjCHtI0Ewlk5vi*Jk|l>{5&=d
zH-t6xwk8U}Glvz_n8nkV>@LhN&#iNZic8(=an6bDIid&1L`EHIl#haQO${kEi#Ywe
zrjPgW`Re%iXRH<PK7&AlAb&=L0PtUDsFmmk=s8<rc)l@SBl)=%Vfty9Z&=Z}@doWM
z5kwN-<TV)1CB?kJ{o#it74c&bw+)E8Hm24t(c7Yy79Ys=xq(CFUX=E0+gDaP_w<$H
zQm;FAS=?pTeMy-DfTQX$XjQyJ<rq{g281)?^{6ovNm8KK6xxre^~BRERc_JPr%|fS
zY;W>wTMR-3N9^QouhaeJrRzL_)<iDlB6l(z0i!#^bqZiDY|NPUfUMwuOq%|0_B+$@
z_`j0H@<T7>0?A=r_0ZGA(N+R9g$53d>zR|Y+F({{5Iw&&K3V^v6_o4wQW?{&GU7$c
z@cdh8LVx%UP|rz{_q}n5CzL*POYPE}SDvVeRq=rBTiu;7$*w=$KMeYi29BI<zC{RQ
zyg*zV61x%IGSQ~*3=e0>c=O(Rg)7bM{u}Nv2JWzv9bwcuv>O)}Z3fIDi8omN^5-!v
zN49OOf&cChJWv2;4Eot~fl|#RvFgvF$F&^S!<C4}o1sY3Z#2N{LECXny3+VSo;q8@
zD^-2CXzlN<4JrGDb5Yk!>fvvApjjVm&w|I{2?)b?BuGz7S(1)%aF*8X6z{)yKaPc$
z|F>(#NNxFfbZ#0+((mni(k^j;VA&hybbfKfXc8s;V$XP=?bQ@_{`N%}Gx46S<7t`(
zpt|}Hc>9$IK9WAReI^DWg;@9NYiQ_u+ur-+>7B502R?n_m5^+)1?m(YM2(`xAzj+J
zh*wE&y3q%B)@h0ahKwKHHpw?FG{r7nH@hbCYsZ;=-UMWw^Csp8pB_X6xm1GQ$$&Ec
z)oo7XAD5XuX99Q`8l`)Rq+v#SMi97fD$q0MnC0j9(^%4|ZI)m1OK&wsrMA3NC&Tes
zKHe-XDqlzkMh+zz_KR`i#n32E_0iua?638hw6xSG_XOh3!j@Bt>=Pf(`Z1kH2h%Q4
zG=QQCt~fCXlJjz(OED?yi?{68P@Wr7cJdjS%v>VQqv?81%77QD&10k{AeRa{_*)3n
zju__^TU+n<sxwoWiz}vR>w@dPu|BpxUYcjH#(i1Z6Ymfu4FIgqeG;<V%E8N#9^n>V
zw0<)g|J|ET5hv-dYt>8maJRRN{IxdYxA!WVnWRJ{0m=mn-L-kv*}<DXpvyY-OKP~G
zcrNyBXojnyw$ytS>yeL3TXK1EU(N416fVO_G8o(pA<6I)9^4ImW~s%ADlf8a$GuIV
zu1<<PwG2&qH^9b@-xBd~;A93>0XGvwS|RWZEN91~-M4ubN1kAR?`^y$n}ogjqDpSn
zEK0o}{bn=OAU(@0c-u^m@m2n`sU<OnnjUePqTfu-3c{J(CLrTi^R^blc@9kTJS|qB
zmaxe-JTpTL!$b8B?6br}B#otr%tFd0de%?TWGObM23Al!whC{8?|iO9dRS#!yBbT~
zH80IFxJEWMX9hj6uGs~Ocd%r3ir-j=rV<WLmNf_nt862zYfM|*iXu+a7R)2Yd`}JJ
zJ%5@bPWP$?{&MWI8CP+H44{eUAa0N|In|Uixn+^tu0u-1@UiBG#V7Rw4=e`Avf^~}
z^obM~QWbu81xz_jPs|+MYLi_qd>@u>P(E)rAJ>$0r$)K{;oPEamOq_LSN5Z)FN2&q
zeaVSEG}f17kI*O3I6XpK=ktJNM!h`m?_L(}Hyqg1YAiIbbk`>DfdcH{J^8cx#1GUq
znywWhfcil-0pK4*R6tL>%f$v)Ynxoet*os--x}mi9zfIv2RnWHVa7BIFbz1)jsE<a
z#gt{LGXGAX?-SF>vd@~SbL)#woSTEGtGzC{dr*?(NSZ2&Kcb%|hR0Qnn<BAn3x?3v
zcyOMI!en-UpIEHs3{zp3&}X#+06)fc@R3TEuE4QP$-az>G0SYm^3@NRU^_nrzksoU
zmtCC?tWIcI?!Jg%r2W9)P4g3g=)Z&IPm|bXy#3(`YXx7)jyeD$oxg<ipxZ6TQW)To
zHDfs^dEsj@{gD@Hb%0njSwO-8Zt?zvjwaSnxTHNlUX??z9#KxFi^h1&2{Ck`f2eIP
z0;E$^A`Jr;s)y;&<MXk+M<-F`WwRydi9KMpY`JyE?zQo>3z)R!boH(*c4l_nJm3RC
zz*g$$7~=JNl?b~l6NVaTiV`niN1^F<>Tmq?<#$K4vaA(dEo67H?pyfOsURZHU?}`6
zG$V8mhAb!rcu!-wZRmGt+O0LL1~n4MrH|%zTm`P%Nl_!E5LB?zPXjTrG`=Z>WeBX=
zSatM|x(dda*L`+UyO?>^Qd&qtXt`sGny3Roh2A6HCAnZRys#!M-74pz9=_qXKb7DC
z3{PLnO5SA*bpG8b4Ehc6c05#rCnJf{p#Ug1-L?F=P2pal8EhvVgZB3U5Aj<d=xCJ;
z{UcnBCaX~_N9#po_Z3X;;7dRGdVBkdnCk2fxhNa*-YJmZ+Pqd#E}01uLVZP%B+|Ok
z3CAFIRi#9K%1wMk4OUi-?d`d$Y|Du)dG=+=7ETud=GWa@_VbUDllOV)$Uv2s8Pm=_
z_Q@+69kR;_rHHT=|I?VAr!+Z_h&vUJFlO(cP58}AtyX#7x#iD+G;2821orS8^{JZU
z+)WWkk|VQ)4DIZ1poqdFcWl<#-eQ*X_vFn!bLsf3@89&?w5?;O2Kse509nS+Fpi9S
zMjb+wTO`w<WncL1d%o}|?_o0phcEYHR8$?MzA@2_$_B(L_GX<08v#4&qvKI(g<!#3
zMy7OgK&=SOXsbur>eK3V0}@ZaGc_lBV8j}T*_6{mv|FuMOKKG67<5w#)vYuqLJX(2
zpgr?hN61iwGrqqn=)<C;Ei4xr>!UiK;D_5gFkTHacyZ0xw`w`a{G^#{M)8let!(N6
zHAY8cS&2Df5X=GmFs!&?Ftw?=W|E}+U2N``fXbKPLEb|9+sv2G`g_vjS=X#cC4ew)
zQBm_<ArWW8Pj{;i2VE(q56n6W&hVVSo&<8LwHx}RuYD>T;VNE{Hp9h$>=K8#U|2tx
zqW&DAlUDWEL{095X1t#9jc;7JlAwEt8b&7O*D2wkWafwo0mvCZtY|JfQdtkrf}M$*
zLRBTEprLKZ#0}{SCr{r0Hg=Y^%&Vmq{RMq}77*BNhz~s84B4%Q?;!@iZ8MbFp%7zf
zTg>^F<;R!{nFXeP&2imJm7!k+5(IL{9(2?(h*`%PKY>MeLpWd^!rrFJM?|8Wpt)3r
zpL^<MvE|#Y>9l3N(i@WOM>Bf{WZjR*NHoW*KwaUnG@7x2<2NRpCi!H<_ZiyhMrvV%
zw8BA}t~jagKKHtodK&C}$s=b!C-;OOb8-a+JCybe>6e1v=`+8BN7FWw$~TVUHiz(2
zI_8BK(j)7MK<z8)I|KI`v%als?)~m7s_RiMDo^kGb>8qoXY#XL5JehLA<qCn+@}f6
zI^ep${EdkVY7sh@zoLq%TiYbBI3^LZ2E)RzWHt2~lYE8hYBv*a<i!&#SLG|LPO#WT
zUi}4}V4r|tfp~9!uOqe+H2SuqJ9yAjoA7<nn`^hxjT5Vrn~L&kLc&n-rE4sPW?j}+
zMSiA3FS{T$VAdj14%rU~awYNkIKIM>rTkN*7mI(ks>+}bBz6|g*L_M|{GIJ;@%>RN
z`<+Oq$DjzEhyft&3cQU=g)}(=ygw~XS8eikwo9nO*7=y1+tpjEVgZT#FCApgB?$Sy
zQwF)+H#=##sRUH1W+7{kQFF3u_`pSpPtj3QQ>ZOzhO}*c5d(UU(SC!7YZTELq%W|C
z=d+U#SFx?A*d|>`e1F9sg)dTV2U>wRZaLB6DqL-C?`o)9z}sdnL8s#IJ3(Z3WY@K0
zke97A9?}VYRu{^8#YoU&Mv))6=jyzx8q#FUaO$Fvj>KT93*ARg`W5X4ylEMQuk7YT
zhnSW1p9t`Tj&I2yBC(OXn7>BtRqUL3>g7&@-txoFpFa;!>y7A_z8*w*5}7O+F$v~C
z7?zSe>`8B(i|joY&rYg|vKh6tHSJ2YE`<1qNv3b#5$Zyy$jQAFsc{_Fz%hlb$U!2}
z+$h=^h567@FxUEB*tP!iS@^}Ox>pRBD(X}_+!rc%GyN~=i47S1o9mXRsYHx}xg(?z
zxAAZsQ?<81Q-?@xjd+h$N<rSYFt_locY%gC+~{LP<%lhjtds&kfd-L-#n#C0BW1|n
zNflH_)K@e~jw!P35kZ9LSdTvj^<X&H9i*oOdM!@BHdP$=o`1@t?=P%>h80)O4{E%~
zB34fartm1yZjAu;>0`(HV^C4$-ZEMny_D_1LdXa5KeH+(__ee^4*j0T^^Z$SB&^qi
zYiEB{=zOxcbgO*n;?FZqMAVotNDg4UJW;HUp&Ls~z4<3$0XgsnG{-tI__L;h4lz?w
zpPpJPH!D-XwFadzC#<w9-6EiSF>wM-vV6o=w=ysOj7weVux*`3xc#$s=On7>&N`hz
z8oZ<sY15F8-~nV5Qn;8}3MASjKQ$Y+KEI<b?8jMHYB!v(>HM@-(YHjt=ZpCRzGQdh
zXKK7){Qb1~#|qlCQwcPtKBdHTpWWNp6-r-=U<&Rc5t};!XN6XLK|o%;3*Ppm#7I{V
zajBujT20WDRuuw*sNT7p%H{!wv3_GB$IYS50E;yOJeLGf93K)JbeW?8hk7OfTlN#y
zbgx%#X=<rU58v3aB!*F+qCMerk5-f;Mlp@GcGfUo9m~70qj;9^_soZh^P7WdwKmlt
zSEs6`)%zkzsO{|(h0<ocL#v0!X%qRD@X3Gy;VgIegs6)0pFlYbK$p${=oC=z%p!sc
zt#cZAWadqTEKB3AlI5=w_#p1c_&MC$gM@~S*Y;|Aa_zcL%H6U^5?9ak6?!9#M41;!
zntX?Vd0}`=(j<gnvwM;8mKYucS2gMO+<|dJDoc>B@_N_#pf{N{4ZWU1kw2e4Pb5M}
zz$v35E0`29iE&^*xgx#X38mTWW{y^pL%u-Pq_|DMV!TFAQJh?2a^lNrXZB`)F89Aa
zgQ0+$p)1~YkgJ&TWa-Ia?V@{9;9XHFdaRmZ3@9Y16E17inpw`r*^*`clB-d{0Iu)W
zg^-XdIexf(O%Gp{yN&uDVel)}{h=PoDn*rYsSho#Hfzy!9g%zT@pnh&v$1(8OfeA`
zDE!USK~Kzg_=(%5=!vcKK})LTHLn)hZ+S)T4?Nx-hGGS9Xuzr}lky9JWT61!iV~p3
zCp=d(Wfc{qY8&bsb6$L%{%H4HZMe4(q;$Yy&X}PG0Ee`t{5h5GB(f6e9x;W4dIFA9
zZ8tHPfbR=^eb{DZ8d_6G=VBRNrGmR9yDm5Myc1ocMBx?ao^gOQjegJpFig<-+@o6C
zAF!qkRpykB!%0svD{9Y~9omUSTaM<PCkZVnokp;?Q1i+fy^$=#@2cZ?12TSH_2m_p
zu@-tN=4xyHB0a*5;hEzIDjFd`Mj*RX+rp}@E;+J7SMnLXc6olyjr;1~ZZoQtc$+zy
z^I`3~e+b`*{;vz~3%K6vUILVvPZzw-FOZwvi4k)2JjyK6Q5WtsX*E8>m32wSo3}=m
zwku!!Sa?T?B!L<L$7ex!+YHDWp;^Qy7%syyJItKz*I#$a7G{UfJS>)6bA2cuUHh!*
zaTG|tcb1?*;e``7NPR^=NL*OyuFAJemXmj?7Kz{bX*=s2ot13{V=ghVp&=Y~C?<yR
z5I`TssL>P?vZ{^OF^HF@z`R13Nl?i8rY-Z+8eR9n<U_4w=KBj1CgO$R#TC(@OY+}A
zlu<l?5mTolNp6_3D4!F!p`9kUaVk4|^(t4qOa5j|!f&p*ciL(6?=KG{@A4f=09z~u
zC{j331rgYY5sctGyaVXettQ(im9qBErgs-5UNzqA6}L0U<o(3D5hLFUfDe)qjg^BO
zPGK6ul7a}{SOZJ~YVYUH+4d<_@^7V*0cN+{i!ROd1wr=TG!=Wd3b-bKk^;c^q3CJ-
z4w_NKKThq0vkl`uMbf|WY?Gp4r^5SMZm!LmA^?ERV2Ljwz9j2}*7aHX3l`0^UEA;*
zpxN}n;5__d-8X`c-lG!L9HaA_nkc-efg_7bf~8@0rvvlEoA})IX<Jd<FY28cxGHIQ
zz5T5vmoQ0>RV876NFgQ~!rCT*H^eqR#n*K<{+c(y?>9$r4=73SI9R(%2xVFe6+Z-t
zyjr5`10l5!&^tK*l~z-sL_KFi%auF)*-DZnWT2O-ObF)?OsISoWVZ=`!RF~rT4>*P
z3|`%no@+x(yKiau^@naajxK<Z5H#KnvyWy(E)c-G6P3hAM8JqV4<k7qgB~;j&cFEL
z%qp0|55;QqK@)i>;t&oUk*d&oSdjzSl6d3oBP0>2y(#ASniz_4r=~OotJJR)qNdX#
zANv&*Ay`R|ioJ-n-!b1waHFW@#X!%}SkDj3mZ{gBgzT-jzv!J!7W1T^u1qJz9D~HU
z5|EohVL)AlL(9=BA_yK3vXsv~XLEf5s{S@Oqd(DlVUC}*qA1gNiEcU)k`E?{b5UN5
zB%q1RwLk<Ei~0@ZY|%?;5T*!4iZUsVz;mz+eY!uwGQapIIADIT5|(YjqVi$W+QmD5
z^b&pMNh3b;iXQkYnsPdccGqA9QkH<;E4SPma~zT0<iOusk$#4TWB5r!eKh)+JqXv2
zH10#26k>LeuZML0Lt%xwA>sazR7;LsP6MQVART2m)RA@%qCrj3ile2QI0E5Ump?@|
z{1Fx%O}iYc?6XI1x4iup?=RVwy6Ac1@k51GN@W6mt;;bv9oc2GO3Hvl?vp5z2;=a}
z(;_A<!XcT&yYB{xMd0TjYgk2sSuSrlHbCkvJCQ6A<BrVx=;*~u3LK5)PjOpCbR~7I
z@mo<?IteL}il2UWKO3=o(nnI`bizsy*QuE_{tX-ZaeEO3)}dRU=E6*>)TJ77?1DI2
zl*79i{I8-ag<kTurdJe6=7CP2bGS~z{+5A20d$rS6n?sg#+%)*d$J>^#F^v6QEHn1
z0CSVfbe~&&&dY1*WDO@jOV;RM2shGagd`)4^DczqE3we-p%HfL4I`)4PZL#&(c7mv
zYHu7IXrFCL*o}oqAUP>K%uG47ktPBHupLo#!z*x(i<5!Q%cdUpQ&)AdA8QlcnqE_9
z=D8h%`JNDWb|cu4UC=;iV#}!doPB>G7Rmjgn5T4Z;B!;XM>=T^*m=FRT@=9Glcdu&
zh0(PUF~}8&81!Tz;dHOR{OQ+^T12%*<jysueADUoFo)T36jxN>Tk|V{?~bJPsiPYP
zddS-O9HyLaS1@dcwXp2d>;{i`pD9~^U3qEgBm`!1;e6#@z=ihpt3vs{Yp;9h(bQ~^
zsd?&|L{W0394GVHpQon(Qqq+jDib<NWP#91!?tE4l+K-y(m&Jj1HUp2S`*jxiaNxf
z4qoYKMhz>)LK?Plhh$MtGl~!?1K+sa+bAra<a=eG&7?F-@MUlxyT)`YHjkm7o8u?f
zi_2z(VO%L_PN2j?Y5qyq-1Y!8(SZ*=Ru5&i=|@1Ytyi?L9&LH+q0Y}Y8>PdCXt?Cs
zA4VeQ)XRq6MuJPhctiA|%I{+ka5NlJrjl^540)<UR|KdhFiaW5L<yc4-{=-TNcNNL
zNDTYckUYN;Jpz|7Wg8H`pw{bWbs4M#Q8bA7%h+F_u&u!3@@s2DpA68%CJ`zXDR8sw
z&D^)e)0Sek@c@9<D#H1kIpegFh(#_Hq47|hJ}~QbV!5pQ89RhK@QJgxTJbmWB=d%A
zHmbXV(Z(zCT|yr}lzCrdEe1UiplIz9`GD8o6@m`{_GZU74B2-K`kq8%$-#?s-9ua>
zRV~$z?g^fcT`1Ga`=l-8;amQmR{$!JR~&QqTGT+K5ed=>Zpz0UkS}`@k3k#}iiivC
z1bvvvPum5rEP;;>wKDU<ugjkq_$MVZ&3#>1Xqk~(fwDzlFhI4u5R6|;3IAsCB>u+B
zi-7_bb${mE%Ac=3Nw&+^+VRh!ci(i4l(2Z92k}9>gfkKEjnhQdg|j!~c$A`p6t2$w
zy8hfO!-06Y2&n!{PZGMy@{zmQaX^g()QK@n9D`mYl16CFp+L}ga~xfOdFptwO_~g$
z$O4$X8_F2xt8z8EAj>YS>9kS5ap`jq$GFkgfivt&{)`b|K=8T|tOz%}BW`sWBN#f+
z`Ir_{@S_J6-@ryyT%Z1?uQp!q?2)CGynUuxB;zyCl;!ug8R&oG#dSRKUuHu7Uu~7G
ze$+2q-BcI~wVrNMD;7mse5}iRVyfxJt@e{}&NcF8SCo(-Bl}fQo_Eo^NQe%^X(STO
zi(rBD!qdQ_RYem(pTv`U_SaTS6P~4LjjO-JM!$5YOL#?TeN}Pcr1k;b5%_=J!T3Mx
zaQx@*|Iad{JD&XiUP*wA+5f9b_s{kIlRW8<r~dOkRx>RttS+P`Ff`s9^!tUqU+-g<
zQS(iic_neX@>>MJ#eF`3p>^Cl2I)Np?SM%}YCzjAi`_8@vvv#;hG5SE!T4{fN5>#!
zKE+!s5%m3UlLi2Z3to;;X-Gm5jfVqCi?W0q93F%EYkn7Sr2TV=e`N8GUi|Zp{IA&y
zE-w<S+z|tCZq$}vwkdhHF_&fuZx8#8`vV+G9Q5&(v<LY_fXDJcITV~|gD<PUfB$`*
zNU*ja_Dcatxn!iD{`9F+BA0j*k4XgYh%_F9P(avT*Z>MZ<I@2eY>(&DF$g_EQ$==P
zMFH8sn`<<n7M?VHg_o`dXeK;~J_cQt%O_KSQZ7S^r0_9_%218m-aQ7TxQywEfd9G1
zKl1oTKmK`N{(JTVd<~)7+S)#9<2{%crn4qe{9pjYsr<;k58(EC91jI-X|3oTn>|*L
zT$hgb2QOmB`L?L(qtFMAO3azH<zKySgD!7495`x*?$U0wZj;>rN+2sz`9Fd(&;JO>
zK0KD8cli&X?0*Dg1Mec$|GC8fiEsar#s7`<|B=N%viKXs{TKE^w`{aH)Auwq%R+G*
zep=<ypk#q$=Hur!4H?+&Ll6wTkK>|6$~i=`$^iV%@$`QVLQ$He{A18rHssMEluEZT
z_Mak8d@k_OzaRf^FWJn?DZ)1xVi?|>7)aC8l>Tm+L%B#A#?P-P#1PBy7%Sm=Vb|`(
zwTn<cf{~>^qoPC#-cUc;|HeZO(Bp8vE`%W|f5|aALPS@Q6jz~`QNohRb-7Qzcj$a@
zMe&6#<1XNEN1jl=8JNx6yoQq54XsBibC@2{Uisf+HL2#D#YHhemnANAo*IqxJVE{M
zj79o)pAl>S_qZ2z?r%T)B`Db$$S2RB%;+3M2>LYrV9fp&WKSLS5XC_%JO-U>g&!S&
zk3`Iva{ms60?NVuwdG*zK`6rl8SUN;^t0fj0g46xrLdjv?d`oDL*HxN+(LVyJC}Xr
zo+lrjd3?U?z18nI-^H4WTaD8e7YB8Anif)JUewhPSy@!+9w#nEh#;VN!@d!qTCCe1
zPd4*~8mnAsep|KH#332_gU^pg{j4iTr;Dc0=gZsW04+p7@kERwncBe1DYji5(u6aQ
zgR2SwssZL;Ag>agIEm{ye5WG9V}I>kO|T@Wn4l*%WDtr3>cf<<8Uu1EH%74il~UCH
z7kpb1nRA}3{XZ)%D~m@7K1+SH&KQuVNd?MHA#p7~7e!`tEb7+PYr`H5xWU!Ndl!0)
zc~pV3s|u6oyMzk3%73G-kOBy{*tNc4zP79Q4ZCMHwJD88E<ZA_FJu`Nvq`${)4gzg
zsC=O;G2#+M9S==bO`{`K+L&tF&OFF`V&s}=pT>JNsr6R*gJc;C$yU^f$SdlJha6}>
z^lYSe(KRupL4`V;`1smpVO?@qhr=b>wEDq9;?V$fQa;FABphneoVxXGOD#+2ofBbD
zB)L*T@#<a9t7n8kO$K~~GMp%3WjQ+rVd;$bUh;_kHG^>bFl+dPIHO0*Qr|U^ys&xM
ze*T?H%iS835GJ#LeqCEhB-HQ@!4$~v$g4f^PC=R^NVn7)(?b~^I#p_=-sc`lK1#D$
zm)n651xFE<*!;PW<*c}sw|ZV3XX|U^kb$DbJU`TWeKtiq6$aP!j##Pw|1CP53qb;6
zG4T!LObE@y;i&@>KSMK^1{xl=DOv3&B3TcI2J&~*Fvxi&DsRJ{FJg6N<WP7UjmxGb
z+JJukZ94y#7fJ!|*q<Mp&y6q#swul456tb4up5nesiWWlet2c02k`x|0)J5>DK^eG
zOUe5{b5L9?g%0~xzh76W-Co;pd}Y_q!!3mQ6RYlh_OHfgK?Rw&K?R&yiNAD&fJtr?
zJ$!%uiW19@hF=*K=d$M96Rp_e1*^9e-3&qL0WXCFC89V2|IXe+3|2&Z&#!BIjvYer
z+J$~lD=FE`4KQf<bh`b=P0^Lq^9lA*u_A9co}J|%<fYqN;OnI5klrq#5+k_d1{_!?
zb!Xpwb-knAkN5RVm1M82$f)1Ha%H2d0Rv3%cz~zX&Pk0V^?l9IOttp)RrEW1t?l%$
z0U=8<LN+hmemmb6lFEP>4=DjlM)H!R&7WC~S9koxt79{!fEi8~w?5EwF1JnGrIyf`
zJZJJ`CiaoMeQuCv_7YS4o_?nXGjCjoj~+<KiN9>sl*iI9L%-mOdZj~8cA<?xcn}5x
zF$m4&$lRAhhrxt}F17OJ2mTE33m39n4(_K96bPZpeuJtZeHiXGrX{w}>b8Jo1&CfE
z;_$}Nx*YN%$}H*zWBi`oj2*5<=%X#@4QJUO#~h#o-QCw^c_vlIkq}3$##0W28-}-^
zm;T8{(b71Y?YpA-SM7N(rU+=?n3hyI6Ul`}fC+Ng?Owsv=bejE6AgfuIr-A3`rX?V
zj)})b{I`fd7=xz{!DJVN3_g2vSUOr)hr*Q{de5e5>h8F$Z7{s<hn#`re6nl0z$xC0
z8+<Phg?`fNtn6OTi#+L1DOBmt<VCi<j?{)Wf8zkz{5MEv`gKPx<Y#)|6-0R=_vmc)
z|4W}a^Z$#_r1X#iz!a4=-0ZSr9FCTT9Z9tKMe7Zp6p(CNXL4@58o-lA%>Lt8_VRF_
zj07x6Jw87sLYw4vsS0s<>g*@^%Mkah`{eIGV{%-MPP%O9oteRz<fnpx-fzfk|M$bW
z+pmC<i^p@RmQIX~kV5XF3;9QWa(l&~f(yxio0qg854=Uv67sZxip)D+cpdn1%GEup
zyve~zZFHsbTDt!`!JW%reI1MbMPK%xeIMNhFp~0@>@mm(LVUcxz5|YN+yzFZG=j-j
z9`A$ruxMV$PYs|D(_sY|De84_;eTY4!m9qAd;VXYdI-*OY<!zC(5;m0$d2?96;nS2
zJAjpXEn2N&4_|R=<(!30NRJsj2Ig%MR7p>klqeF5!r3^j{9f6@3He^>N5*h<G51^H
ztAV{p;;3aSZ4APGZy88}=JST~xp0Mrc#Y3`G^~Ewby{m&GXokp19K_O)EqU1N5o9h
zCw!AJ;S7(c%2I#gh`-d&E0d>RG)i0%IWd1WIbZ5G{eIFCs;AO18pGPAf<OAMh!Y#;
z33f<(cnoT54%i&8yLdEHe#ZdBcgRLC#7?IXpfpw`4=kAHnqMGH-nSa=%x><UI{o!)
zO@~&|&2qsIA=FF-WfZ*gTp#F{Rif4vaG6reaX@NUi#F16X(r;dXO=GCl+HB;)#+?#
zlgWc_!%JDc&NudBO6Gdf2SRKP^NIeXeX<Y&OmZplY<acYW*Fp?>2;v|=OI)wTbX@O
z;Fa00%Y!lnh-RUmna~ADLs>V@a`~VWd5UDBcU7@Ei4XV2)pby<GC>D00$$hIljh}B
z`Vd&Mdc*|egcQl6n54h_B1?>);1Pp+e!D&HUX#_W(lJCsx^^jBZTVsuCs;zT>nvD@
zq_T|ZwpSuf;5*RdD_^H0&JZd}-+r(Te~T+1oA%T`J9CRyMrrl(+(khRcky;P;tXCD
z0t^rGf<)^wv}P<fOgrr;3hRP_)_G6&*Rai7&XH;xx^&dSpQLv3^?@aHRD{s@b{l=s
zb;-&7!4LLQ#sXm+e%K;b(Z*uu^qO;T5K(O*#1+EJwI0lz2UaO%(^rgU%8$V*z`3PX
z6BMn!B&Ghk(kZq81;(EUpb+dN^JT8M9g6ZN{K5+@Ha#aG-ED9Y(QGa5<K}voj_b`&
z*XP5Cr$F|7ji6o!)QN|r=<;oxZupE>|L~y6oyJ^!>uJ%Q+3O!$pRWIEckDd|H7RZG
z7lpFI1z}=RN#^aIZ1b>;%A`SQxt%^qS9jgpLa$j*`a^|Q&`#i32Pu3=kG;o$gL2c8
zE4BmK7&!Om5>a0>uxyU)OT`h1NMz*uZ9Q<<xUX4Xr${^~k0c6ttH4WlSV-1D4iXRm
zSz8T129=z${1YDG88M96K{%d4ZoK|Gf64sF@n4@V3cROchyMXyv>k)6C`wY*F~}Zs
zsPLE(#J8KB2Kng?%;jiwJd*YPJ@@b7M?Kx=e;GR)0M8RgxB($~0Osy>x8vz|z_1(q
zLGO2DUC>GJ1-95(`sd$40EwNGupR=wXK#taRj)h9LFs7UbOoOa=cff3Ijq3770&rr
z*dhHtFS;ZDB%p3nKXnLd^dcj*+E=698sID+rxk8esf|mA>&$DfOg_?AUYHq1*4{3w
zj2MU1^B0k?hUT8+4%NT@alDFthmq1qmUg{zsTIeOT5$6dFrExb(Jc6r$A_ooU{i#Z
z&i5EM#*QyQT6vjIF)e|ra)CMtv<So*{!EOSl(K?O(3K?+GfuXiOh71XR-sOLG+O()
zzAR1eUaGuowE{{Hl}=tto7JL>0?_hNI3@{96opbN_4pz1aQC65?akd1gbFFhm-A^U
zDUn*`usEetZ+lT}IZzKBsPmz<X39iK-wMr%1lX(ZQ0&0dO$t{@ucE@rQ<s*^aM9Jw
zom(6`w}*uk?RKAa{sw_q_c8MuSV(+_6hh>gIsX4(@4W+>*tfOeK~xl#*bp1Uj)KUx
zV4($dBO)R;KtYJgRulwOM36R!1w|1XqF90j5fNemQQC-f-AX44NN7rjBtS@;;akB%
zDEHiR&$;K^``vf{vE9jJm}F+wZ>?uN>sf08623H)k4>`redSn%5KFIssrNLdUfc2H
zddukJSB@K`Qc4QAyV-ue4jMV*%IkHf6wf&2&9reHlsU`!k<!j{HF(3gcjL2XK9ZS+
zc%}=LJ=d`21PiU)0^gE{I{rZ>CED6`PT*e7V4rbpt{OM`+#K0cGxk`Q4x8FMF^&~p
zNj^;|dw)U4yHFG?=G=eCvu?4Bi4Ag1RV%y@G5nIr$VbbbZvJiUr0Qw7f>2B3%z%cu
z9c;DC6Q1+Jw}}?z7Tu363i(i(e5YFWu6*Z7J||zE!#NwHJL!~XL`6<OiF%59scY4R
zyH)I2x>JjzzrILFSX-hWbLCGn-|d5ZkB*&WGKv3ObBrE&yLTBTqfM?tkGnI@MYVrY
z(`X%@5q!<{;F@Reo@F%5I`iCT-Nfu^M@C91s+Pj+Lh958>R4TAH}wq7KTmtD;-ncI
zQazmi!LHenvn(NIPvC3%r7pzYsl@e-?sA^PvJ~ct5MSQ=G-8R{f$DP`%&zN?aXI>4
z%C}|coBOl=xb_BmqFJhyO_*%*Xtynuc!BDd{S)Q}50Nu(UNmmL(8Q%p@u?R`{GaUz
zyx{3tNjY6UF^L@*;4-m2NhQVAaK2vz#bu1a0vLk0(NJf|5sOp!ouFOLkL?znQJY_N
z?BS*41_Mxq>%JLe&%#W3jVz2C$SZQLGAU6=YpbKliFBqDnRzpO8&)4pIAeCT^!~~D
zL+iI)EIYYs{2z!%tKSjoB%+)$;YjP5<hf6S^?8)aNuTPAUyRhfJLBnmxkVFSZ=Ahz
z-Mv7y;SL|Dksa*4H-v|YtoQ}D9Ukmgnlr&{>7geZQ{;lYrXI>TA^RxO5rM}*(4Q|6
zHd<|dTe{^Lx~RNmR>+)`DVOy}hRY28L+P}S+CPwVOLeIGVKq`9?q|eM1W1&77OPX_
zvxF(&`awBfp5E>J$O8>U|2Oa{qD&l`BJ851GeVTlpnF?SPu@dBRD12_x6iY&Klm)e
z=H?%zhBk?>Hjh8<6P0)=PNu!;S#zRJ==O#!jb`o+F_TX*1f|z2^+QZLYmmj>b9|Q5
zPEh3L6JtFWu$Zau7e(QL`!4<2_~!E0wpkpKZO{$BXY>vKNTS#0Z#r5fq1`dkTRPt#
zEhXW1lH7e{`@lMnAUCIU5`R=F%6J8}!4e<s;(tIZM!Kd;A&QxDm~Cl6WfwBG)cQ+f
z@Pt8vkHcmRop&Pf(!H24ACo(a6MXfi`M8^tgtDUf#fei`)mwNcs;6vmsG^=RkBTUp
zseG;efYHS9;jESg9H~#s2M<zh@@OUPn;h1h92QaklHqz3ZI}}>{_<L^?C4;6h5mt!
zGG{%v;P(iUr+ndo7}{A2cK(9ci6L&IEq0VS1z7H1;X4nheKw784NZ0Ten)H9gi(kZ
zWWs>{#A!*C`o=vT!sU+Jhe}=3I4l!dHVPCVC+$;d?XodYFCEi`oI*SSM~$i%s-U?I
zOy2xLl&_TRm$!uCcV|Y@Q(n~lI_|gzi?9cqOB<LrzbAQYzduCgR@NNZje%D$9Zyk=
z^dQF2<<MM}0zza!$UmF9f9(8CbAQ(x^k^a?=Ob0(L$=;k#q^~O^E~G1LY>H4W?0mE
z)3Nd9x#b~8OMa6!it|8*{<cB2pe}SIX_VWRRHc#m1;4W{T-krGo_+bflkFhOw22*B
zHpm~HI>z3v-UT#{J8ckJE|wKqVCz*LyqvI1Va%5tr>R#`J7@hdqyW!fV^FE5WJ7x6
z<a>>nrZn%woA~+ekK?W{nD7>PyS~;<>L6LWC8O27LmUag3gm2!S>`8>I2;zH33N~F
zvU$>Q%c?xZe!TfI<Ra>^=#guUblbSB=#YO*)+Ei;g<5k~JhS5qG`cWC&1oVHn1}6x
zH739;GP+*qpyb2sT2X*2ll#64(O2`D(x(cv;vXSOb;sc0lmnEZqUC<+l;KCtX*ISg
zCOv)e@{~jB%5Tg!QR8oBYrQ?YvSj?VU2%xBb(C<Xh)k=9OBtn*YVkm6O&prAsW3Rc
ztgO1zp>_POnL5uOT!?L#=6UE{F|+bEUivw9iIs2N+tUY=eb+l?zO>)%qr02!+I&g4
zf~SxcA9zxGOVx`#k^47aU$9F!r2XbB>-(Mu7m**$GGEcp&>r8^%Z=W%zhz0|{yomu
zY)?%(+cE2W$T>-y8ZXs`ou)XDKF4L~&M5GlSIL}kGii(YE0}0SZW7jsvcq;hO1Lw`
zG+Sqzk5))0Q_f8_C0R~L70u8bb=r^1*Dj?!ElGE%HO$FTz8abITKlDmhPOJYZYk#h
zYyZo<tdsJ4PDu~a9Icn70iO8S1dJt95$Nx-rdYJH!gzv!Fj!=$eeE~?<ph`g6Xz~4
z5pJhj-Sb&|Bw)%~7IAxMVCbimho3B>vmBc5Z`&2%B|mz|rluj@`#cBbwFi#&{){Qb
zc)4ecTKBeL#RPkg>(}=EL5)c}SXU7_k~;(2X!pQu!f!haU56wd%2~#|v{A;X^ZH0y
zhR<?!8Icmtltl=vOj2~m7A!H!em>d5a$8lfz01Z5#}P!<G!OF!khhs7J%3U7q_!Y-
z;>`G%_<(ybKu7!Vnri!*k$)_DI&1WrVHJ!ZG*wqL7)pxvl{C?~?I`<kN5fp{EVPp5
zJqXRyVC4|QpJ2Qq_YX;l6D}Iq+fPm%sWN5^Vrsg{=P}H+dNU~OA7krv!+J-#?Ml$j
za&agiero`eb%SE-x~FRlG>9J9*`SdHbsJ}mi8;SuYvoGiDdhYl+1GS=G)tEvXrQ!D
z2qoR7yoe?7xxoa_iFIG71zl{k{b=jyE=0Ed-CXHEfkOWqAAcNYP!zT~@AHNVml;cE
z^0!hym-vZ<MqP;KGM;&^_Z0Y{4mrY1(2U5ZZIrnIh~AFe)~t@Rn3V?Hqrk%JLiZSR
z6d_-K^<Vrk3@q8x;Gv!K-1zA&+xG`-3zh?n==n9)A3-**ZqS8LW#Gp?S%<VynwnP=
zUcm1sPZ7-}J{D4HOQ=F<7fxE_|F8M*s!J$o*_E|LcT=vtJCb9Z_ml`y=yi?S$>%7A
zfJ_saS>|K(lx*?akI<YLtTPooMV7qo4<V!b@A+_Ufy-FE3C}O1pPX29qsR#uyMrWb
zgweV0E!P)b^4EFrF7?sEr<8X$C4P#>PzZz5jRHHdusyl|##_GG;6cOl%QIfw$Re5!
zpkeO%F4jO;1x8?HM2u9W@xum{b+J!EFASSo=6#hWdAXi6P&s4bA*}|v)7{Bo|6SMa
zZ{Cs>N0N+#ZaB<QNm?85R5#H)%&Z*b{?k4f;ZVf65lWgUVQ_&GhNxNqjj@aW^&f_5
z>OPs-W_-1h#6IQ_m?K};QCf~>(aFLHQ99zum$==3^ECT@sGzG*T4o7lDn5ION8&BV
z&5Hww*hcvbo)KXUlT2vD2df}NzKS+B$}zmyA!L}fp4yvt5i_+)^OK@qsmr68?bMDx
z(YiH0+KYR6DhU@xUw_X>JLhkHS}i1-jp1XvX||n-FEeZX&-L0{AZNmtmk~mj%j)PE
z|JkAb_80v#pMH)>GDg(*FKU@IDd4ohqpBMc*W6>xGa}ps2y&Mvn*SPK)W^lmd;idG
zL(16vy*9bSKGD8f2U&!ui%zL^A?3tXshZ9EEgl%Zv|MpfF!2-l>v{?6Ov2L1oUv5F
zH|$Twihs|-T9UM3&1c02e`Y=nI7^t;o5oc#`6E`hJK=+KWzM5L4!=+M36GkM9HU9P
z+wl@RTi?+CZ~RnVWuG9%-&$+lXM@c%cT<Agb&e9^kgmfJ=73P#AK|z4vdI6Mb^kL)
z*ZXf>xz-OyuXUJHvO$VdI!REKr1J7B{LI1Rj?Y*S;+FaExmy3xeSpqQF>sh4vL;GS
zyL4WG&aySBeE;?b!p08y&u(!omR?lBMwuL@gsa~qv4-6DjF<X`{*4a{JvC2n)7H&i
zw8WwIb?P9Jen&LtHFE?cAxnUNSVOFeS3y*3v8mh3jjC>bp5be-<!i;`%cK^@EBW@>
z$HLE0YSe+Nm4Y-`A8}*+6W1yY&T#Zl%Xzs<Hv;_HA7MRsj2<e2EK~_JyshM~JN`RO
z^gp`j-_hFm`x^>}zlptig#hT`Ze{F5BvU&5+CoX{UW+#e{0FdL!fSv9mVu9mRnPbR
zXRz`=L7o4Vk0WO0yBKcx^yJ&&6>l>#1?CUTH&vSO@L&Rd4&bDTC429kH;vje*k$O_
z9L@tt-{rIwj2*f*jyTnB{ZE1Lf8{6rCn!67L2$&=y_SY&tRJi?^KzrFi%MEdtP}*|
z&eU2xrG*kU+&8@bH=(}2dDwMFytWy9yE*SI>hb=7w;SCcx*VoAnRX#Al=kU*WL*ib
z*~i8Hs}X*$>|%q1`j+)hF*jUCp3g{~$p~{db!em<P9>7=|IdIc4dca|FZs@fxD-0(
z8RcD2d(`GRrE0gbyha`%ap*g9u^Kg}4vJuSr1c!II+{odjY8al|8cM3e$@9#$b?%L
zWS^I=8m2vZm2B~2cURGgX0-7#S~nxQd)noIQy-)4)2SR%AQ2yk4~YzueD_G&=SMkT
zI*!E@op40UDS~0qGZN5I;1rkE;!p|>^{?vRWd&v88tXtJnM9n}76i_!SDq!QQ<-OA
zv2IhbkCxK}G$O7GIfh*nA6?>2Z?)SqkjRiBUrok7Mi!l;@&tZ(2XREo*#fR$Sf-Gf
z|GaTa&+G0i`Bz6>bmUtayJMM<R{YwcArb1_<SpoHe4q4)&G5$o4BQ39hZU(LM4(w~
z8i{y}IB6Gzj0c|J=YR5X;b_)k*!d{soBi0A?XCQwcGQ+k(FbIA-)fk-(=m|pB%&?h
zZSrXnZ^lcd{n%CNvzr6}K9oGR3kk-Awvc+8@6j7_&>H5V1lD1Vpbf|kaE(oIizFye
zFpIY7LRKx2`t#>OO0(f>VARhMw9W;-M_J^50?&h#hB2mJQE*B~rr<V(NGg8X=)A29
z*%A&BExuUs*V&2BQa%E{BWp7gZ>`_%((_tRL9_`R`E2-pd(j8ypgi7vVihF(I~e?s
z+TQm*O5Av!WkJ4=*3EgQ>XeFcCivAc1;ax}ZYjQewc|l=WmEzRvJa(3yjC1B56PGE
zJP5r`YAc2CohyB&uZ|?@HCiJjgCBMJx-BS9UF0(je$EHcBk8ey=xCX6CiPMD`*iGZ
zX#-4pcvaXN9mxNf^D1)}CNQiFC*qfJ<nzq6DpCGWnxGESY<(cIdWCX#tKRPsn_rio
z7f$Gash3mUz4CAF9hiSwP4{)xzyv(ljr{R$L=;^2Oo9bQdp8=*GOKQ&TznEz%%Ie$
z9_k=xnHl%KA%$*R2+AHfQ?$@r-OAt;rJ73>)Z+SxjUA06Zrh0t5H*~l7Ex-%{fAal
z_(ih4WMHaJ>%q9aA+v#-RXp3Fp2i^+qK&|eXd-ESudt+B-=TlBlgv?}3!%VqR~D&~
zT(jcWs!rR5*T@~sio4CqUv$o=G?w%~pfq(Ae0eNZ2$a>FD7Tx$DwH*Zh&3N*Ym8LR
z&#T%vS;A|X>Sz%wxmZUwrG1Qfx2;JJ)UGY*wEkT#ESs(FuT82rH$`oArqb-EERn_8
zOAa5J-jlBH&R~U|`M79thkOQzJ6XKpKu@c#>_Xl^rk|0#9Lm90V3h>3N{M6_ldq$^
z4PhSf{w`!hYK1aBJ_^_<X;|mbSnB2}<{pD!fma+rtEnHY#!KNkDY`cv?Z2j3d=8Zh
z3vXCBJgQ}y13P<C-Mds@(xY&>*smB=(QCT7^G}mWU#%a&L)ILsaypg*WCU9WHzDz+
z_XuP&HMF{c+{xe<xl1@&p9S_*&iGedh=;(;%1+{SZ&NJ-Q`H56%~`-uwK4}wZE)(O
zL?OHEdQmOrF6#VYUr>g**s*Qy?G7YdDp-{yIWlrxU)ghWE2YLf5HBQl!^->h+Gn}{
zj~%tthokSA_zRd(7Wq7S*;^{76NF9h;3vDLM@aODbSfKHz@5p0G11&!EW?Ul2RKHl
z*-CB+#O2@%YM}105sjy`YYq_WB;Bc_J-OisbO(1@r1n0y*d<iHEYKiHxaxwQ*K-HK
z;|wV^v#^d2=#A*Z@s#msmSOkuT)g9J@5|eh)WS06xe}{#aiCgPE|Kzg&)<9}=4vk`
zv$+$9&ZWPgAnW)Zt|j~|a!ArOb-Sk?cN#fXGT-YL*V-k0srh7@m-oE{^$GZUr%0$N
zAytol)EV2=d*w{Ep6Wn!6?FS2*}6vveF<AebgG3?0a$`=B>66Mq}|W^hmP+;GNUkI
zCf|IF73Pmt7c=o{QRJE4m-QiF1}a&Y$}iOT<zs!8Xyou^%e)LY@>eWMYDK3Ub?@_K
zvwQzxe4s3v*@SiE2>ib`K4OK7yx)V5>_Q%ej)VtVt}>@|<YBirMhe13C!^8E`>5vK
z-Vs&Olcq_c^@&#{AVc@nEiFT{9%C)x#7WgNk}g(ns(xR=+)u#zWL;^o_g6pM!$7a!
z@$BoUoC$FP8a}v)J`T+?ArP$i@c`R=Ry6#$mO3^Yzz=)T@xvzgVAFTu0Aq9+#A{{3
zSLM5rp#u#!_@8j8B36IjXx``~8c!pB$4)al+8(2G?K$hdH&NWfd{B-PZ4}xxN`^z<
ztZ4sEcLa5x1^Es6^hW!Va$1<-f>ZW0)V>rRR4~`8cyxHNPwMhIduV1o8UBXWzI-3b
zTX+JNUnpq<+g5NB?YzmCty1~+y1EPTbH92>4LjGqbK_-KyU}tOYw?a+(QI9C#&^X}
zU`#nr7MUXUd1lISoWUEOXKFiVJinh-ns+I=e&q6L7goxD@A8lpUw#Qc`WIjwFZCB-
z{l5%Y&D0zoB)&4*cSa6hTu|(z%8=k)c1%s4#rl1b_F}gL{R?Ze&CCj?4_RB`JK~_e
zechcEJdf}l+AnyRzDjnCL*X>-8gi>F^7=8F1-<C;%i12xuj@=Fq+TV2xQTY%bGY<=
zWW#T^upE=N)Vl|s?s)%f*FCp$Mum?4*G89G>n}MuH+%ZVE62-y=14+Z2`>#9#6vWT
z0$B}YG<E$_h(QS+fqZ#$9Dr-?DacEHIq-qfk|f)fZ73tg<*cQ9PT5njguVX(M|q7P
zRux0m1?h4~Y?Tuq5aAmIL6Oyfn{#D!ekLhT_A@yoD6T~tiWBYETOsL`DS-@PmN;YD
z(H+uA15PygA4uu{E6!WD3Xei-Zg>Ov&7BL#ICM2tj!{O9rHihpe})H_r4OUD53{c<
z=}u;dU=cu_S&b<aE>b{gziof?X=0xZJuJar4)-sH;Q!wYLGzl%>%m2Z-Z4@r0Md(5
znT6=|?Hb<*3Vsi@?LvmSVVhYw=S2IJZamo*ua^GoMf}B#Km9A5F3i~yG>v6q0RX1*
z31tr^n0pAVNQSiMe6CF5CTkK1h;|9px{$op=U_f+Cnnk(E20Puy+7)zFd%0cS*N6Z
zp><1<<BZssC%1ixxV-=Rj0Lr4=Dr`jXsyyiADz?wcfOHdl)ZPlU2Z9#v%aD#G;~mG
z423USikomW0BMs&ryPiI{$zyMBXa3NX2zYNG|hB{q0u2?zOC;TM724Gr<^0Yb?qi}
zL*M|b^v%;;xV55jwZsM^V;DH?4O)?VkEFT2k_6Ao_8h)U0|oM>+mIsy|2qvgAYRkk
zLI+3@TPQWA)K2iZtHfC^RAy#8?n2^eg3+OJGbE2Ge*Ln8_Iek{Yw{qw`pFviJETUt
zI)8*EZbJClG0tK&%3TOTZ8SYHz(&2P9Y3(W(lbf8y`=!FA=^WyaYd|;Is6^S4%7|&
zuSVZ#o;!JY70lx_65RUnypneU!N<-`6ZVC;r{$lum^W+4$>(d9HlM#B=X^o+YYxv1
zZR!GAVO8f@w6zG@lxOA2lx{Uo!&m^=#s2zKD-f8>ND5(ipHLzTDG;8Ls;I&au-EJ*
zQt4<4#IG779&lw;&2Y|XY*u7DlUoTrcL2b_q;<yvJ@&S^b7_cmg!2>rLhHfX8CjR6
zJ5()o`FiovX>A)!o`31P)c25gF~(Vs!GVS2r@_}ur~BErsFyGZ!{F|pxcekoFj#}t
z+?_q6M#L!44!Ts(gvfrHU_)#Ib&*>w_W{qGrC`FM`fBeAvsYj^RVFQMSoX}w_57-=
zyty{>^IkS(ICH88JMWMYX;Zi;z<!T3tg``v)#ZKC&5u>9sQoVp(5@sB4wt6Rnyrot
zUcmW=AVNF;ydt%`F^T(je*AOZlz7hK?S+WxPu|+a`U!B=nT3>Au-_3O4AH3tlyV{{
z*rMlU@1(Ow-V*zcVG1=Z!A(4;QE#F}u;`>C)))z2z9z!-ixIJE8MUJb2K&t1`sf$^
zErvo-Nx-Is1Ph;aA&UV6_3+vI2Esr&e0lICE|lv)$$R{I)3bQw-b>OdkpZk=Q4%zF
zQ1FlhsFV7#%|)m_n0h90(mqL&atW2&{bTvbm7DYkFA^jemtG{Mm~Z5hxi9g@B4?I#
zn)O65x>HNMW^O*QIls8Upm0}+x%t>dyVg6e8B_re<8GzSMpLts7lLs_RR)R%$HT37
z%z(r)2uL@e(omrk30LaiztAIthhcHx$t3NnZZ=T3U4Zdb0(r-lWl#k2iSYM6!BkNY
z^$t8VqyveIil~n_d&Hi#h`QafOfzF?=&_9p-_)MJv&h(hs_AKpKPJX`$`-ZL5Z(*J
zVkf8R`3D|upMUvF73apfM{~B{cVN!TdGFsw>TfyXr4%|=PqDGer2mISUzQQFLs$x%
z3muNsX;Z57uQhBvJ3@bd%;KjTH!LvBFj*qiX&{<eLKn4>JM&>AlXbxot5qe$q;Gm-
z<P6J5PEp>n-d`||i|$gveoWx_l74BQwBRY@NMvbap2qXjdOo%Xr}}ys+n{fC9-}#M
z(q8I31}D#&2G8R(qWGXqJM*#jXibDT|M0ZYF^frS6y=6>K8iFjhtPC8NBHeUkA?k3
z{{O7XOEPTodvgfg3b?N5ZL`RwD7#apTTaJv6+hzo${jWn_q0UoXK8T0dKn0T1<uZK
zYbBD^u=dN7`4AwuPM-@1qt;ZKaN-dRpLZc~ZxWsBsDFj)!#fg{YcDt?GOHX$Y?Z=E
zzcg*451Z&$J&y^Gh{mvN17cf;HBMGIGY89aik@HVHY0Z7lBa7oy_@jLtYq!-EEz@p
z!)Sjm9!h!{X$dj;Vw9M@pD9*Irlx{;68J=gM58Xm`$-qVpOdt#qP6)3%cCkTEr9$+
zcVb@bdn=Qzi}{Pn4Guh2zcTpB(zdBBWsgVy7Jlm}bXUps0|9?8Ea+4hjmz6<QbKBG
zNS%3}7dza?I$Uae2iJy|E7O!D&sOj6r-8-sBfw*~tGfjg_;x5k8Y5n@`r-!+^z+kP
zCtqqz?OQA`d3vEpt)f|d+q0yCBYc;k9_?{!&mT%WR59`4#g;GAw0(XfEs<&wGko$A
z31lM_1An2Cfr^2LdZ35cl?MLGEeGOapWyFF0`3K<p8cq>^gE&y{Vtki!E8X=$C)Jp
zQkqWT0gr4EIp9{L_9rXb5g<`YQMn`L*M(F;4NWjCvYS1Y>KdTK`d@%Us!-nCK+(<4
z(=cTEk8@p!;etrEBetJXagV}rDu=}G*=2QnRy;jDzB>vCwW{DTnVD~#1bntK5v<{9
zL$&p^gr&2b2eF;13#oAaH~RT~-HPaekbqaZ5W;NsB5`nT+IRxmhE69deovj_x)8@!
zK3N3jz1!>(K-C4RRkA{q+X8$)Waub&_I|QpQC0G^@4Gi_5d-q$UC1G*Ms4eO1_eC5
zF60$l!6)vBN;+NSwVDt>X^sA?kdiy(gFUe-nB0o4dmm-|>t!D<nhsoaA+rUw_-g+c
zsAw-^qP8P+q?%+Lf7}uVH+l+P05wg)n8#at(fz%OCBPj6$ea{`CXm~fNS*EbSWBy<
zT9H3njD!CKV!h|hMwu}`Nryd^>U<Aq{6vKsM@{<(HG5T}HOd)7s^~o89Pv63AC@b#
z@aGXEE74YoG>!_X(}3^;xss$~6u<V~3Uul^2+$~tg8TRpf$pNu^n54YJgO*=fV=W^
zG~_F-)SFAh=PT6-&les+7<og+u$v42#Hv=l69z(jgnkeV7Q!<R_@^X@^%arN1%2t7
zs3`>IfVJ;t=va|XVl{SoSR=Qz3t6Lvcr6wl1)C`3mDWSasy@CXS)U$_IWqa0v?*Ro
znD{bF^Q;)e{ITbVGdb+I%BE=**;R_OojVfFm<KVUJYl(AZe-OG8OCIIFw=}s5;Jg|
z$db4CzMC%;LKxa!_v#tk=g&=?bjtS6-=G;8v0|{hnSy9Rs}HSxb2~{iDo|w4p2-ad
zEA|&Z&0v=Y4VO(E>O;9t<ykSMD0N|VWZo1qulS`O+n}QLY^;ZsOlrAqUaaDp4$&*M
ziUfme292>>r(Q`%mj<=_HJuvyg2=xkwKmv!)apah!iaG%$ZrPilkR-G?a$4Lzt`NI
zNm+N{Wy=vS!<IcRal4;1t*pMQA;Vn|0b=}S!N#Fh5(TB6?Pf2}KBSh~Z8FxLJ$h)i
zZbh!l{Qs9rO-~@t$t)uSR)1X76VefHIidvRIG1Yy^LNe(pLIHU2mMhO;y~fAXjKHJ
zE<KL8yqdZICK7!SYM|LEo|<6QTyRDAixbx*C2;M)27Gyt9q($UKu2^GVvgT~^^h!D
z1NQ6($!q=%)eQU3*glM}j*Y>Zc0oLDf4Ghvf~Yo9KG_l*Xm0>@*pUTJujyC%ZGv8p
zzbojrP<YC%qF}|<L^=);az7(A2o~G8PQvH{VC3BdIG>n0>B@KFk_J8ou=D{mdnel*
zKevo3Ftw-NE)vXy=6NV@Urwo_3nw^Y?T_K98y0olN`dv@H2;rD5OVv^BKa#){uL=D
zTN?h~ij@6gB7HX9f+r5ljLIDRd<ccOrmv@{+-%Z?D9f`)5N|sn$95a#(b;zkBHux7
zvV^^KDoh0Y*WmvQBfq{A1aj5H*%C_`_bh{eJ*!YE%X14akS$kQ<$A6;Dsj_e{gD<K
zp=Zu|Xg+>iwj4oth#VrH0<6*p9&l}|_eOOg$HAppZRY_sGkZ%0%AH(@f+KBfKw2o3
zdqqG{-F;yOpbr4Wg5=lu>BK|TR`%Q=R`Km8yb~L-n9GWKG0PO3ji$)QTs<H83|iU`
z6(`HlpHp_Cd`<EQaP^7?UC7#Rp`_9+g&@gzc)6!2A$FV1A{{{79`sJ|{m>fMOQO-k
zsSNr_OiJi5cAJ%99OML%u#D56=2Yq}!l>TTN-s3uQ^kDk(1F(LO*Q>VL!OdQxb0{K
z%uzF0tZSHVC6CB|yE6@3bW)}m2d|W`cj@(hyi~w1SXV{kbpU2|>q1njRO8qtB6omO
ziD=NF@jZByN`o>^vlBb8qBBDa8_ZMr_D*rZ2ra{)y)lMV&u1-8*V~GdZ6Q57N#u1Q
zZ8+esior*ugj!|?V8C<>3FJqNsvMEFGB*(V5}TI+$vQLo#T#l|aI;JmNzu)DWuem%
zZcJgUs@8trPZ>{aLiVglQIs2=5=v6QW8l1M6I~%Mn7|Ymj`Dm>-PA8UhYlJb?79Vb
zxHbW0raXkzNTimaM4$yi7wQ8>^x_4;lLh4ftA#BxKhJpKoriL$-ZRlpsQ{nq*9TQE
zLClNMr#YhA6vzz93n}ePXX>1OoAk(9ZjFS2(`fY5npOz(8i!+x`fLEeS%$JqnEVBr
ze)u^-Fi}AFQ|%j>b7fMy^UnbSE~XXPZa@rw2KId-oywj61+6!YLcC-1VAhi)S1!&F
zC^ck2DGUwJAr6otWQzPFi3YT86=<J&So9wC>aV-~OZ*2KnD2Ga{>K}d4AwT?9)-)5
zYit3Ji#rK-d<1S6J>cej(s#s6CHyB{yrfsPb09Ty-d}O}uZ(r0%wHMn&y4loABU^4
z324n@w;UqXB9OS6ZS7@!f^AEz3qKERiw5`@aFMlzi1}yg(?u`53za6S3?DJ{-Ucnw
zvv(^*#-8?6NPto(Xw5em@6ra&^LCvyH@9AJ+C;FVyUkfyvLQlV)R8S*@147QjdazV
z*N_RlZ^uNYi;2lLTyQG)&PJmZN0^|^LRUdqs;4b|^OO-^hlG^q$;`S6P0h98efRf@
z6b|#;&5NvNp0~JjwwgM@&f<&jKJzQ{6<ik?PYl{~-1P0UnbQ^`!Vm;s`olXR(acmP
zM+S-u#~>ja1A6J+4SoF;(~eX=vJym)H~D6>9o_>L7)EJ)fvWV=VN2C_G8$B$vz^Lb
z3V?do&=uhNN3GR_*O=zedYKMHNH-Z$XRQE1dw;hC{L~TCkGvi^J0zR?t4RD2W#}8U
zeT;Rm<U75Ig+By}M~QjRWMGMQ26Dtb7vp>MmcOb_<nv_KMwGt-^1>8|iT?C6k<Cor
zT#D$HI4#^M&RC&pyA?4H^;Cx3FT8>&99hp4Y*9RhyUV;q`6}L{Alv;2RAbn}T$yM6
zLh)aj^8fivIsac7p`};l7$sqAE`OF*xuv96%Z{Y6E;KOPvQVZGp}}r#2K+xK1xgj4
zHbC)WHT`V}LOi5YDi&e27omqJShNqspsoP=?*MO0Wz)aj@kek-6nq0pO!lzGrC!{z
zfB}14n0fw0aeiYyHa@b2G#<_Nanm9)wE}VF*L1(SZlu!GdJg5-lBKZ5+BBMAM-e-~
z>o(hoSYKVS4EGf0A4Va`3%_&7uP}Qme-0Ir5Zr$>S~r@J8r!MRY!4;%_u@+PvsEN~
z&}81x?%Y9_i6>w{jgs-R76!!mSM)Qg%}`)XAPl;@Og{xZK<nQ7^L)R{2qKJ~iB&Rq
z8o(@l8%7RuG6>#DCn++-v~nIPJ!DOsYs7Ya>BMWBN|y}T8(?+tkK0qGjk!g)-~@`x
zj>7KhKGymH(v#U6WlH6DGYDN+<WRQqgk>(SnV?YGUWW7Gr`J6G#OiGBV3DmD?H|Tw
zu4GSj)jU<Wl)dDV#bEQGxAP9{xZ-27d~kIv;uR5vW^JzS?kI|iMl>vabPO7X{E;gq
z$xHXN3y4AS&o+YGp%cO+Id4-}=)LM`)aq3tR&^3focD$iP3WKEHWwymWFht?kRQ^|
zyCTu$ZG=8#Fkei*iS^jO-t=huq@PPxva3C3cOmL17ivj+qu=?+RwSBsA>=%iu@ExG
z$*uO1b?$xA*MEZ7xGRMIN}>*KMk9QSM47uMD^$()UMlk9ISVP<c!#SqCwoVD6SEY&
zO}K~OR_bQikgyNe=G%VQVWT5=D8o}mtgFz4!zEQY3EONTG1)4`O_yg=#HBB<Zixc3
z^Berg%Mf$Lr7sLBUpEO6u$CNC#Ek?G78aY0G%SW;Xk91$LkJEztFJHboC(|eHS2wP
z$nS?!ZKB+)%eh~aWoK`TwtApfX*74|6X~6o?wn-QhIMoIUYQ^r60_e-NZURo!9UIG
zc!!g#;Yi2PY^MUW$+`gLhIy{g8*)S!Bq5g~fyShlI^h#zbIeymU14&JlK_7Sr2aWQ
z@_M3oVcNctUK&Si2+v>|L)K?pT*OgB<Qu67RueB;0ZZcw{Qmr`$ZY%KWGOV;h35=;
z*&Ij`!(jN2p>PXal5|f9{7;i|h*Q?m)_A{7dPAVZIsx{}aT`Y@JEHBE?f8e2@8joC
zNE(Mj(zuKwFw{WYTGMHKl{<^!P{sGV(>|JcsAxN!bwo8qILeM9*fRSVOvL5V#SG~N
z`_*uYI`EHSd9n&U7OkLR9A^tMeyo8k*w8M;K7ut;8|*YMQ#<G#8j6T~XRG)b!bUH_
z(d=$+B`(C#eIi?#JY@w@u!AL$UdqH3o%G1`_8(zdvMmk_wioIcsD+>hS)}!}%=K?0
z(2;}{K^eW9=6sev{@FV);w>1^C~-T)#c2P2&F~+z0ntqN2;`9nU)B6Z$+vT*C;wD=
zFw-gQMkj(zn2NiPBshfbJg?h>vQH4QBuKw!K0-ffnSgzmvgN%Jz}RGf_PJC@X;KUa
zb<1O(zmIG?39ue#t5_4d>$VouTWshQx*iYyXtqfT#+_7vwLjEExRg>8VuME}+QD{!
z2UA~UybtOiH^V`cb*iBo4@@wBt^L<`h-xXem(E|V?nmV+6#$ZRXPjZgd4HgC4pQZ?
zCLQs&c7)0T{Y@1GGtV%&zcrw6``tcxENce*HM_HF8Gc@qCI-HX!MtpRBf-21cx1d8
z48pimgR`gKwk1++aTl<fqih-cq$7NzvpMuj_1OT2$^=3k2b-&gD}qI!&6+>fnKyjF
zA-EvaLX3MG@mwKxq||#9mT@3|zLE+olf8{_O``1E;<4Y+ijWH0c_o!MIifAyOuZgC
z>d4mQ^~^G-th|~Jc!nqd(YJKB6`klH!^Dn{&pmHoub7vjZfh<xrfe#EuvepDVY$z5
z`+oHGh=n-f|5W2%!O}KGznI{|#6`qkbU{!5Rd@4)*oJ;KvVh#^M`hb*;m_mz{6Ax$
zUTD*SHX_bH(W`p~pvqqb>>v}d|NqtE()}y{wS9upUumT$2;&pr7M|YrBmAhRG=x#Y
zgJ||sdkS~lQ?cJ?J^UWUE7%62>oO;>CTF;8`5qxW@8tfQzVG^x7YC5#5NisN>}d$D
z=L2;j<1-UnR?SCUX_S$3meT$@vU+m&`3cW)W5GW5h}CwV?#d9yM|o;d2av~ouZ#}}
z9FqG1jPJ}#GpSTaY;aB^cQnERsGsk`_VT(bMA7VzZYPMOBY<6IUr1ID9h4Awn!?#E
z#{RqPN%-7#m5uqyBhwW#Gf0C{7WrK;_E>Kwb9ZRNZ~6#osM#qx6|Ea7*0E>NPwOu8
zE?<U?^j6^ozS+c^T(HGza>c3ze<Zq<m|U`){rewsuNtl4(vs8)DRnnLqZOVsk)01^
z!z22iRNgU1z$L$zgQJtK0m{&n>8Xk*J0{N%tz|KfFVL2@-J_J`HRR@oxXW5==Il)H
zz5Zr-&U$}D$nk&SeonE59s2gkd88dFviok#ze?`WX7Rr&@yOdktY)SQq-#_FMjGp)
zf2nK4Q9tZghSWXhTmo+Z57r~G>+f~tOMlH&g1&-ZuX}VLN39`h{^|tZ;dB!F+Dd$e
zRJ!Q%MyVYN{#dm@UGR<4KGUkXcau2cJpN~1$3e3mQNEA`itrWL5)I<NnmuK{wTi~4
z5-TaP-XTollr8mYtN=`D$ZxQQOy1JOO?7v-Grp>Q5BX6-D-g~_v6P84HC0IX{CN+P
zi}7f%Ni*I!h+z|gAoWi7-10-83F*8ZRufl`an<)@!p+H3>0tF0p;V4#J5#6-!^9(~
zavP=k6y%+YdKK(*WeO;9PNLf;0m&gw#LcM48Vf-q7fkp{1Us(4xC<$8iS<0EH^iOs
zMxDO^W+XTL0bFwXa}|z72i=FtvGSe@Ya$zhcl{q*rSiRjDc~HUj>ekQo#E8XeHd83
zO}SD)6{kbwE*}ix{3Z>b74p53yPFA>+(9VgLk@A-SWii-AGuLfN9n_WfcFMF7klq3
zA-P>hW6nlc_pWsR$Y!dz%zCTX50u94e#Dg|#M4S$zz~F7co2-+<HiQ?mjJ{ESHZe&
zjzm8tctD*vs<A>v*hvgehj3GB1`hscC-BFXxiXEvo}ts5djM)4#s?m6!OBs5aF4Fz
z7v#1)4EJ*oNzlZ6_e`~imDo6RcAtJkl1VbQ$tw7q_t&^0a7KqBB#DVoIh7yrowSOD
ziB6cn)RY?4d(p0vfWb$$&15v1WnE23ix+--e6P1i#xJxiQA+^8Bhgmhy+*}>LbMY=
z<_QRXSq)^aoH$gQdg@alfTjQhsi|=x31#+!Hby0R1`zzxl@w7BWhfYdEd+-A>y0^$
zxw>Gi*;`mn_~|39Af)rr3H&=1c&(o!i47wv8Mt2zcs`}rDvf653v<yeg;ctbB>qmF
zeS@%mXu$fz0ObR)vh74Fg>ytKZt(!KDDe8=d87X}#Iw1Y&R0x?Q`<EK!>1O3-nY*N
zIMku<)Y&VzAnFZN;f5b;4e;j<xEu-{>Ug*wSx!tCnU`Ttzdeg^^DyH@((Br$*yD@e
z2*N%jh7Mi$VZ`rspLVWHMGmf1x|cke4(Dh?O<)HSX`^gG36LwJ4vhl|!c7p-+SCat
z7=r0fsf<Gb1uN}=a+#_+v5=~=Nsnv_5<y1-YHHfxn0?HsqTR_n{B*LW$efq1exW&e
zmVIlKeprcx&eMI<{NC@k*jU@S@ZsU{E;;K&JLs@Ld2@_o+P;X47Ea&AHD0ltXw%4$
zsXn1c4vR4h|0z}3gr5VEYJ&GwsbGCObaw--t%Vu3TK5L(Fe$`s`)C(s#kxTOKaE&&
zfB`It9>_A#lK9!X1zNtJodCK0*283hg6LEf5e`){4f`}E2Mo$bz2kBGKwCrQ+<z{Q
zZM{YR5s;lq^mKi9oB!Gl+`qsIfR!Yz3YzGelN9<1H3FaTs_{?zg5=@g6ZWzcaKCm8
zR|TIig_{Ci`x?6wN7ae57n1GiVUz=8{t72D%HTjQaJ}(cJfivq-Amywr~0AXDbFES
z@l8}oVsLHX4^Q!2VZ|n>;}hpf3&s^vstM>{6pxa*y+iU%cVLz8p*sF=PfU<L%e%og
zs$`^Kqw0y-JU>@wdxwY}u}`SBC;wdUAI6uy_!~*1lF2HgyddF2I!r`}ds`rHZl>r`
zjtu(A&IDI>7aa;EeWCIk$)EiQUqLv5=#V9M7pcHQ^){dtq2=>gUC4#RF)XL_Wch1`
zF}F9UIa;~~jdV13xz4}ZUNc#gL64*C%i%Ajea;~gU2<?md+L~OHcF<18Os#;GY6qf
zt3dL;jqY#X>{QZ9=sO*!aHsF53g(-q{}0I!|JN3EN-%M`dPNk8NbDfNZQ?yn2B6?8
zPEkMXLpy38q4iq1GU@qUNE90JsKF2|#tI9MAm&VrwE=)zNgRGI4ijjWCl8`yXvHl6
zZm~{61t+xKB?<Ay#C7ZL;Procn#JBQ*SmG}k(%FLhwp!$Gx^=6FIEO4t{(77*>U2$
z@8W`K#jc`Rcfd{Ejnjv4f06>_ExS$IQPyQQW#aS~1?Zj{-G#8XqHhf=Xo4w6sG^Ch
zyO8vaGS%Ut@nm9y0f3CBM}gG67g?nubR}d*FBGnAm|3vl>35O^k*_1e(Q6`j%ZOA>
zSK?}>HEMmFMI+YHPM}ST!_oTnN2Hq2mo(9>&I^ELt4e|BnobW^ZRkg^Tt&?LGAHyg
zbVob{tEos<L_c|md7VZ*d3fkgUoj)ssYD&R9MYVG6J03yyvvDuePjhW+)3b4!&wwz
zi63#Z#lGfU-%b0ZYn>F)tmZ0`aM%~{&fml-T%Qd-vY_Z04ybx|CR~bV=w^@zyRg3Y
z=T{KkVR6JxF2Do4TQexa(Hb;7oI1h01bs&T!pnqAXXuM=8Dv$r3{n`*wZf9~3B)@$
z{`3v!*<po~_g>C>KO*Mv%wxgYk@E(V!#vffPTn6-jw2Rhj{COL@JY^kWHrzv<?(Jn
zJ;IUxLz6^i#dyy_b4wv<s1W}yg&pwppD$?N-MpjF6djRamHHTw4!bOQdeMSdgVRK|
zjeSJTt<bq?rxt9G;?gtXaZgx+&Nq!nr>u1fJG4qYMi>+}&s@CY-Lo}MF4cb9`({!(
zQhP;@96BCAS|8IQ_cbf22z@|#y=}`Z_Qf{pC&evm3mr7HOs~9KWU>CmBTq+_?TZ#k
z$xO<b;WI^o68hh+ASwD+)Tg9x@Qzmps8<H2I*k#nDLPVHv@6%f*W^W}^3|Mg?`FI?
zFj|+TeW8d|E7HF5hT6V?PGRlFM2^YV@d&KulkN;q<|NUGpiF8`7~=gJfOWR^5THC7
z+F-S-!VD=WEBIY@8$p~cswGC6L)RCh24<TIs@a*<7XsO>+~sU#|8}dfR<f$((c3>S
zS@cnPLDtMGZaX87pIsPeC%Qm?CFX4A45@Y!PK8b*<5*9f?=bYB+kIlabPAKZq71Fq
z`QS;R_S^D($AMZf?E9C&zB32|HDJ*D7s6t?>)KqXDZpBu3D_EFk9GN!_C4vIE2&Pz
z7bb@SJabwUe6(hVdD{0fIM=Tk@@(Z5RU|=20VIXjG<nI%xj2T^)QY)j>v)v7oJm#4
zm9oCgG9iqX`Cv_yFyIAtxO~&7XF{%D^8zkps(=cYTTdxK2w7(Kdp9A01;Yi;q-rqo
zX)j5lTXBI{BOh&F@6u+4$hTlG$f8^7shDuoXS6QGTHd-*2-TqMw<0ZYwqJsY^DwAu
zcmtUY)3AOrNE@Q<dcXf(2kIF_Xde{)Ye1pTZ?TeNvLfRp8$5r}INb@orFh1phI*$B
zu@_o~Aldy<1?!)^3_03K<EUnWu4|!qy4%m!Jv9WyU<tO#e2PRD2vqnb8a)^ul0=l=
z)bt-2m!d?c$I%`VYP=LHP&D3#SX&DesD?7I9^T6tO|E@6eFtHsv>02_O_QctxOJMf
z#D!S;na1XvlEs8^I2W!ZT@n5<uwbMe`EDz5&49OO#J!;jN5|0d393~gLQ8G}kkghG
zCioc(bK;yl`};^ev0weGa`i8$T+w;TTe(LmlSBrO?H1#A3QSl@cv`;v{0U3$lpcCh
z>tk3kE*<es*<=lIf7Ev20>A+Jejw^}qVS5ytRmX;R($+6lr^*s+}K8$_`HGkb0_(D
zt9Z`hMJ~8+&#Fx~jNVay@N8{>TDYIx!V<YXzfZlh`~z|{NMeBG|G7$;G-nGTE3R>&
zZ)x25CE&qKKX|*E*TK>PtOgCyl&0;mkk&?qh+4iY4WPtT5lZ76>Vz39qYAqDGDE6^
zm@lK$RJVd}of6(%QUjgxUnRBvCAIxT2Ig8##?MU70U_==)n>{kMvDlN{AbeJ0YV;;
zV($_AJ+CC82DtusB9F=)k&m|T`QW*F0Ms{HoXmi{R}KPBnm5GoPaZ%*<XtNgBT-f>
zvEHBx1AY<1CprA2?3!Ig<CuhT4#f9ocfcR6ZMJ8x!R)b$JHTc;x?}kHPKv<5UXN_s
zYq-V#>JHNZo_qh#_svMBVs=jarKB@hN2{%vr)IR(Q*)`!P$!s^!d4)Xn!xy@H3_fp
z#jO@`zLS02t(3?d;e?NKppwkX#N6<eLF_pE0_eu)lq4(t5bJ0I?KwxPqE$E*o+E$f
zTQZscs^6yXyS_*UV4r=<6F6r&Z7f>vY@VUPbxp_kivxXNa9x!NzQB%`@x99TUm1Qh
zP&X<)ZlJ!^??u7YTmnwz78IA=>w%g!QuL%!_jxm!9A~k}pKb-kLPHz+xjKZtfu{`a
z<mMYHmkbFyT%fke?XA8jXL}-G&0V-URFSJgNo(+@XwBC|FHlFkha}R59@1EEQYu4i
zu*_t3h%mt%O@1`wY6{$qzoR8um7Ms~(xzuB&U2)u5i%kChGakd3R+W5kip_NQ)&vO
zxC&`Gyt&Hh6dBL8Z0gA)FnEUN?qNkTOMd2W7}a87exbzm?DltMbLX$}U5nU0jh;y;
znXo;GQzlSZuh3wCP>PX-DPc9QMf$KX0%3LJXWM7jNw*vI-|;2!5q(*kW^$f9DRBj>
zHqCl!uEP?1SM=}COG-b!P`~0ZzcD7UB{62_)QD;wK0QlTz~D}REHT#LcuGg1$=C25
zp!j@uQEH}CgMIkV&Rn{0ED6rw&4?+&oAcVqV`Hvy5)YV7(FwbBX8DKQ>KU|MfHahN
zL&DeQTh=Mfet%vucRLM_4<*I+nte(?w-@rXvqa8|KiiUzS2^S_v^g`MeHri##k7Gs
zi~_pY6U@ols~@>Jr)q9^ZSNeupE3Drz%j+=(@&Q@+OuUUN3quJ%`~~!GkkP??j_F^
ztA4z5&~}7P!Nol;HIFHkmCP4R9MOib;XNN3_Emgz83XRgrTL;UanJxCpAl6FzrH>`
zJ*Yfrcjw0vU&{x3<coeAxn-T<d%4h&wjl_&p}VaAtH)Ra2A}<cp#5w6s&R<`$E+!x
zTloIK!uP!CnJL_qI9ozt^@=Gmxixf<<aE}AlHTGg<l9dLf#if%RY~bpDCs0xai7dF
z&{hPajsSgeBRd1L!TN3czB0@3`;?jz)RrP(GbrsYQJw}=NAElu$AR*KhPyS=1e(Rk
zGruqRSa8LB`DDl-KAj3f>z+}^Gx{-z-%Ib%!ZBAtbPZodq>I~G?998HEBkdaBdT{n
zcNSo=4EHWY>(Zm$<H^I(tfKoUqdAT^Z-EGG`&B@xR^GiORQr&AU02pd*0C(;nx?}D
ziAhUPe7?Vrrb!8n*q~aafS+nX+f`VTJ6ro~K!D!)1X?VQH?)p6Fv$m3%*FbdK8;zA
zq%*~KmA#cs!%xSO#k`FmNd}sCf=qz9`D72y6S`M(VB%{ZRa6Jtbf5}pDCr!eW)?6f
zZ&Tsr@0=m{sh8-k9R5^Y$of8=djg2%y*n=?`7`cEsX53C1XgAN82f#-I_WAqv0>tn
zi22%~S&V9!5HvUyK8kG9#MXlnhQ`t0!Kk8?-{2Sx2yWC{O4VP$t<rJlwkw)g@({nC
zGtEU%G?_&%DZ3<86|I^X7`z@%jW<`ip2jn9apG4|T4V1MW4>4tZh^vPd%j&cbOJ4L
zLR(=NRv6kXI6VV`)8a_0{R-So#~5Qytv`PI4aYl3^tu&x!hsF)J$l)Dh#O6r!rn;3
zXZ!l-R*Gw*Jqj++&-^bKdTuf%Yt5>$x>u98Vr5S6e6%RzdsO3AIsWn?#btAqzv<`(
zr5^*H?GUDdh%UmpT>&GZvcmCmOEH0|^N)7nA#(?N?qofpwEsf0-I=XSDazVt!e0)I
zop7i)gsCLYq7Q^+q5R)hW6q#4`YO>8j{#I)&`wwYQBrdF3L@XPsB52Q-&Y%!I|S{^
zQzqTyo`Y(=kUfRSPy|Q2?hpDTAQ4~Sr_q{v5sBd5?Lr<3L->pOZ9@2^nN*GzC~Dp=
zLj@}$A^-3F9u5OW+|8!{5`1~C=vVdi(VZ$bmA-{rMtTsJ)@0x54GZGaaacIbnh5WC
zH^W(uhd_Mzg6^XWhtyL9i@qgKr^jJ?G5#`Yi7JueZbF=Y9ZEX0Pb~Fnq4=Sk-4saP
znsi`s*1fkA>(Inyt~oQ=RA`l>66arrlKjM;Se2VV3~z-`Y9rx0bHsk*Xe{qO74j06
z$)3IVt&W4CU<$)^a!dqmv7w|F_7YWD^GRQgK#&$0`wBD;C%2o+94)NI%EwEkZrc3A
zmZ)<Mz+?QpMj`t8hTZ!<;)D^^-dm*<<&}ffJEw3SKG(#*&eXl<Ges<l`!$dAf2^ka
z56x3~VrevW-0<0F*59YhTRt?F64OMfQzYZ<=v$wqA_yPTN@<;eptd|iU!CW;^Uk|t
z)8n?88ooPm(Jk6WxFMA|o(GfI3`BFKYt>6`f<UcG9ryU$O?Zg}<{dKhQa9Ul%~7qV
zwO8L;=@n)--K?s7&FopjQVM^YsQsHvlv}N6F}>+4kY7wA5tI#JK60(MKXHaqc^kD=
z7t9e*L31TCC9^(aFM{!LUg{i_Q1PMsrVU`S7okYax5C~sPZZxF#j}RQ`985>bu_;#
zH~wSUF_VLLx2|2|{A$Q1M(kl)p99?%s0sk6o_4`+VRbj_QLq;%Zlr1wAF1ue?N~Ml
z1s)?Jte>O~3%f}>NtSW+Rw{ARtWLSM$j!Z&AHD0TR^ruHf9%{dwf4|R1M7~|zDFWl
z>*#eSh)E3YVL-P|VFWDUbckCc8-**-yk!84;k>FINps@*O`*xxz0bN|QB1GAJu@Xf
z*mu6;600HM_EXHhd@a2h+%|je;!8;vOq4^u^|^8gs?~1Z4%B3yujqTuQv4G_2!(y-
z*3j$K)2Ji7Np3Gw4GOdkDk7%{sie{oTEo81H!J_5H)u)9#EnSfouGFEo@nPbJQ%ed
zr8d)rckH3zx3MeCCuWLdpJKi5E#gh}?n}1H<Fk1Qm9{CZisjoIloqf|FGL*LWv_UZ
zw6w}5?&Pzz8N>KnpJf)mHsRS)Kffo=CI|B=U5LCM`6DimkM8qh`d$%lbE02@(kNKV
z4tz^H;Z0#YWK+(2j<Jcf3zj`MJ56}*dB)wXV~0&ekgC^c|F3g*O`-7Tmk$CamI&_7
z23vyrm~@nmLWTMpMdfWzi;mF!N*tn^ud<B?XJbbWXH;Iw{7|%Q&|GQuiGE6edU6JJ
zu(u5FN}AyG>niG?dcmFWV{`Z|dzi^)B||L-%lXI-aI-twDNQ%QjL0kiA8#+gDxrU-
z_5RtG)`~ne*5bGBUub8wwJOwfQ#TjXj7`zDOwFjiKW|nWZ=*~_YbKq)03Cxi*%U#t
z_S}NwPMmi^kE2LOvqh&>u`YQ(!(4Elv!iUi@V()ZKdrtzb$>C|*tXA=M+n!(#=uWo
zhz*e`<j%2TA>9Y>m_9qziDt6`E8>oOuH0WmAMQ3e)u`x+D}UK<En}5K0;W`4HkkV-
zNpTc{Y7(;n+6S6KTKO7&)2=7$h_kjhjbeN5M$Kzl<;Uxscj=$ccwV!)?A_^Xp=r>R
zE?jQstZmF{Tpzf*cG`+oq60x!XmlzxksuM>0xiq)8fb#4@`7_N!ezuqv(80tbf@kQ
z_?ocbps1o+Sd{O!y3#476=_hIOg|x#w+XZQVwt^v{>7G&EF)j`tt+Vxli=}hZTt1n
z-+Q~Kz}}ZzY?R&?(3ItSpA{rJRDW?YP4Lfno0*!gk=C{}tLpsv3;9|~gYMm(Z?cja
z+S@<H(?^Ns^Ry~QSXsYMwU&KWP{D{&%HYo^8<HO5<9C*`VXLomzRzN`(#2B-{7En*
zQz8=J1ONMkg83U%FwdI|nkgQ}dz{&Kz4y5Z$S2_RidbaApXvwUQ&*i(iN|HCsBdqZ
zq{b8U2fWg?Nd+-6$LZ0#5&|k?wr@YVcdv}>R9^+1wHmd(Pd~v)4xWg-?7o{#TF8We
zii9a@>J~;Wg=$xhz7IQfJ%g-FnZA3C&OT|GzQ2*u?=+tDS#Sn@R#xb!K%Ny<3UERg
z4+VWL2pOv)pR+p8Bq~-bE~+(WnV)&0z0$#UM*8AUf4)%B8hL8>1D``5RX?D;U68xE
zZ=xD3n97g0T^{hrsT1OL!kW%4$$PU+HS50A1?Q}(8;1tTzsAFN!8k9dhPB=&#-sZ`
zp}CqAA-p|>?8>vGl^i9v%LNhwN~K}19i`4n%)K&G=HB+hrSAhg59FDfI6DujGdx(_
zkarPF$dOiN!1L}o{$)hLQ;;tnK|3L~%6?3Ib-GHAr8f=Pd1^&Y60#N0*FWDf6#cm@
z<-oQa$G9oa0x~B8e(L%4dQ%SfD0=IhoDsM%C8Of6vnVcibh(;D?4P|(-P;{F-Zhk(
z3(%&Px)~H0jOqZ?*js~@>Ek`4{#wbh0MTOmC)>BbvpqfUy}aQtm2EatD=t~@?-!Az
z`OJIRJyP$*)9;Rx-edjCGcNdSdvt`Z94~NHp0-95Z%#(rdRCQ@63uB&)wTC4woKC-
zmQfXX>q`D>YVX7UodLswV4xgT&*8uTjR5jHZ+G%MdObCRER#Hncey|>j;(xRzRTR0
z+q$Y+^1(;dqHHFv-6GTSs6M2sZ|wQIqw%SM>POfEY@$ewr^A*u4YXtj`#y28ue)pH
z65^G9a#QW@g9g&d@&iQT#$Jg&oe7wxSJd2M>79eU_z;$6ev~N}llARXP~K@#<h-nz
z-<tMGb4>cv0Qt)2yvK+pu$gBTB$`&WN`JBoiEYWz{N|8hz`10z{-R#xc&W^l+lnbK
zY@4c&xhg*6IX|3xv3j{Th24M<E~@<+s?1wHMhX`ZA2l7a;;4QWfA~B)aIrcXEqMrw
zYY;JL-{jb93HL?11ySnlpQ~QS-B{*#Vc#}Q&6*A~1eCh#@d(H-s*<OppR8*^K2MjT
z*Hx#Hhv*LBg<B-C70zf&hcoRgP6lYCah%j=1r%NPU7-;;ko}1pgnJ3FsdTItV2l%~
zvs^2JeLEUW#_=rQyqxS9vGdQ|9r}sdd#>J{cG*B>ZGE{9rJMfhMN?q+J{VM}H+3PA
zb_y2J_gQVmuM(1izJ;t|zOK1+Zn2X4WkU^<KClpC3)Zh)y|F{R&u->UthS2I6imkL
zA#(@I>BRga4x5);8vAWjX_3|1or~7i9NZ_(fAU@LP0bsQ-^Oeg8B#{I|4Fk2ZyC|w
zKs&fOO)H=x?w5uN$GCus#EA-*rPpPMr^$M%@>W!M+-%Gl&YP3D{^sVGYZ;Yy%SOH|
zkYDusXXUTE?^D(xgu|o13+?im^a2u3d9dJUaD?YN_Rtqza<uSm`_0Nd+n%niT65`B
z!omublPl&VgD2{R)J=<cs5nN;)`xrysB9gw?bBzs(^3xTcn{yQRf|u0Ub)n!biGk;
z5{(s~EDh_{2o6BVqyy$QU-{5Gmx>%X@GurVJ6U?th|&6@<uy`x1BB4tvB9w(d*0eq
ziSAL$0~9;`{H1_sXCG4Mqe1HZLr*rgpFWXChA(&?oOU)*zDkQUI`hDi*c(r5x2NgT
zD^#XNELS@b80Jpu6Ud^&*Tax#X9Lzpdc~K;DNC0M79{SnZ`}IjYuS{e%t?oC-F~u2
z+}5(R@bAL%qE<!|Q#9@cvDW-7WYW7R0;3^bBl=o=WI>`rB_<`(%6D9%KA2}$YVb7v
zX7kw9Qab|{<lg+~{IN)y6*{8OtX3jqZR#D`{3OKv!4m19>IS36l@HPvK3{cu^PoQ}
zrn-d6t?CokBs0z5>Y{R*wMYUP1Ly!IvmW#cfY4JLC$V#&#EzjOHC3G@zINWw@Lf3~
zqwmpm<4{en4uOADqGhLAG96b{bkQ30W|Zc_oKA?mMpYxOt6h`Ui7#_hg8*2sr{tmU
zC>huO$eCfZ*l_CZ*^d62vdC#5GXA-~1L-Q0QoriBbo4}Eg^|<glD2($4c8d-!xM(h
zX`8HkW;)*PbbhGLLPiKrotWameE<r)Z&cBsXW-e~@kn;CPx55Znm`jJ3M=+uqQFAa
zGAX0VX8W)OLx(kNwXNSl>0?l#q^umCLu<(u!tVyklFbC<48HVqIUYf&I|g|Q@&x58
zVSi{ieRky@EZB9yv^ztax-;_pA);efUvMZ&$KyM!R$H+tsXESP!OaRKfxb&3&(=&^
zz~6h>_2^rlq1&$bAno+i_5wq;$y@q~I~rb^CgZ^~Y<8IQQFX+Y1=Tersk8;NQ^+bU
z%LNRmUG+&v|K8wFhf;{>7L<qZ6Hv>+<*8%M+k>jrPtjz&|3CJ=JP@kB?SBx8m<o|?
zD%wa|NFk#INkYXGF>ON9Bw4b|Q9{X5B&9N1B<qxYAF`#AHB*$aW@pAaW;uRe-6_=b
zKF@tW_x(Qa?|tt-Wac!^oO8a{_qsmU=h~L`;9Ag2TvnEv)IQl25KXW@f75^UD;ZB!
zjQp>@?xz|~cGGO_0_~x5Gto9#4<a5%yKk$p>+d@ic*xS68NBQ0>%uSHb3wIBE8*uN
zPdJ!`qaJ7q@>4$m5nUeVkBTS8IK8g<R8D$(BUbM6In8EGl=m-~h|$@VA56>H7aoV%
zNnb{lDKsq)yQ-v!u&FZ*e6Y~~r|wo*y4LD^Qn0~Ne_#;T|Ei}WL#Yntb?2c;zi=b=
zK*-^q@|{NEde^T#X|&L_JpQil@>bbk+u6qDs}1_8gai-x(U}xUXU;0BZxiuS&P*R}
z!N4M&FEZL6WN*ISz#Hh}iH?9~rAv|?T_*v$Xw)??E1WUVTz7Zvm|*qp4*cOy?bpV$
zf=~C~WNv`piQgaKcUP^Esn1U0ilBe><BWf57}+zjcwqD_`+h>AWjabwsbt}cBhBN0
zZ^#0%KY*El>NoLlD9Gb`QH^W!?yI-v=h>he&{d``IluB>^x{0Cnc{4UKQy}37R}s8
zIxx$NaFiCKgMyy|QvHcS>3J_|NNH!<Yx-FUh3``DNuSgY#+&4q0~P+S(J?@UH&BwX
ztb4>h5zzZ2chzJT8o66^);>>cEkw`N5bVTD(WI|Eq;D)K)vs&5rp)g$b|!208(~aY
zXH?*Z9!*sFwtvoxKx!jns6x{C29zII5oth14O^z-LHK{>T6g#6EoMpA<g9o`U-YKH
zWPPW2O4TRx!!bBHo8xjucdWK@Yj5oX-x&2TeIrAP+Ry?8<}HGYM5R~J-C{T2jCs@Q
zlV)>!b<|4H=h~JH8&wC5_Sr|Mk~?P;D~XZrYv`NRCEXF~LMJ8=b7?tnB~!G#-6PH_
z>`Uv`&&h$gfLRIoNuPp(BGPGtZv{mtScgbP&qTy#N%%LU<jXc(2Y_>T@#!K>H7Alo
zVD>XE8tJE#G2%)|DU?zL)aZWt37U{TZD(PN0%Z7pSzb0%IcAd%PCZ8a)&^no*hu%-
zd8J#4NwcK}aNEFj>5dStQAa{eaZ833#C0u`Q|uZKA~!9*dCO|+@uCQ~O5Ro^Yx7v2
zE=c5kOQLgw+kG_*72@!<gl`p4XjTw55t;$hej*}n5~vN7)KzTcMiAT<0T5V*+!N-h
zq0kBTZsCBC`$^i<2hp~wqY>nlw&TZBRb-#b#s*v;<S}&ezc3OtTRpz-p-fVPk9SvH
zkb(+KRO+Oqt~J#(huG^`6VA5pUFhi)nA!C!@C67@{<tPITbKlxm?n2D%NX3wOd*>M
zOV2<ryXoN5T8f3heL;x=vd`j$$Hf!RK6I%DhB-*}g}%SPu|R#V^w#6{nNnisjBj0@
z+b!S%&Dt(Rk3+-eTzjf7*s1OHlb2B2axBeTmw*0-XS)Jisj&!QFcx;YBd;?J`55v@
z&0+w&4m_8-&{@HnJIMeX|9yXm;EL9UngDOkxsCsqv$}HS;CFUA#9EVf-@R7IsVj*(
zO*oR9nw~fKC0Vz@TJ73W_G&Kw8Nldy=fB#<?icN3;{$9B?ir=-UMd%#DDmZ$7$PvA
zQI{*skq@%56`nyAyGfgYw_puIC1!@8fz3@k$m}X2duSofp{dtTqlT&#%SrTbbo1g`
zbjK6yJ&2}EG13%s)xNvU`$^r<#@jc7H-`jjt6e%Wz*-JEWer~+!oe%EyH|5vnt4jV
zc+5?EIDy!<oYgRaxR%K90~0iUoK*>$b~Uznk36@a=vt}B@h;IX6qY_}+NBIUixlLA
z7uxNFUA4y3ClEq9v%dJI9S^n3=A<-wgQepA`zOSP;ZVYV+n4|IyDlUvm5r=!Ike>9
zL)m^=M+Vn_?#b=n*?f=wB~Uwftvu&BGk{2)vSU|s1=Ol<b8~i6D8h+hwk}zPehr-P
zKW%aT`SY1B70RiPtjA2x4C<=M@E#F(5&<`1bhQxnDtL*!h;EyzR(_e}`62g(2>tJ&
zY0~0cYskwrbHzbW{fEGqkNlhBFNvHcaw!T*60g?w_kC_v*kYKmip!=#9b^7qZ=V@|
zAM}@Ad1rLZsxWQ+t%u!<`8Mh{6jVHg_I&zxb{E&q>O`dqbcH2u)7j0zu2!{3Q_5t>
z_HJE@ZNi@Rm6vWBAHcocy|rxHrIEz&e`Ax+-(0iGPKc?raXf=-tD4IhhShMQ^Ix}&
zzjqnRu$?MlM)gl{V(WJ;OfoR)Dz^IbRM_v_B@PB4CK}a?S)^`h*w%C(bk*$zO496b
z)w&Uh7wUq34|lMNylxmoDu>S86@Q@;o^=OxKB19~hbTG@n@WrwZanF@8QAncX_KTR
zY;=>DmtbkBxa>GzNZ>)iCX@C*Q%C$>;hJfjZMP3gl+?&1A*F|JWNzJ3fZEgZ&urkV
z$N#4q-+Ate&v@pPiSoV&z<@$l$_~&!wqPUiwwc;BDf><VoRQ|grQbpxA1aTo7h5U!
z_E7qp;Pv+1+_s=0iO8$d;oR+g!|6uq$53avgd&6m^BEZIAJ?-y(y31ko)rk6+ZeMm
zB;#ho2l0++aAXqX{9AW1a|QnyymGe?t6j9qLH5voq)8v>$(PvghiFYzL^d{Bym1gJ
z5qY(AkwcBNPt1Qz!z4elCY#z7zEP9mF@J0$+p%<vGs|=}mtQWmeWdtP&Et~A+G{uN
zJ7>U5HT#dL=mh?4JcEsO{h}^!_C>CnSISEgA^*3o;w+FnLw>T>TqvVxiCbF^X&T*s
zkh9ri6XSoBirHyciCcI3gElQ_rHFr9L|j+te^$w$^{MQ>2g8_+IhKa^J@;Ll<H~pM
zP-XuJe?`iU&bSoho%^(5o-BY?|4mINU%#u;w2-SQ#X8o$X6f<xZ@rJ8iC=UCf8^JZ
zqeqPPU3(%gSZK?|Ic_`cy63$+=dqgW;iH_#@Ztv~HWvkkk?_hOx~UT&S&$Huj;4nr
zMqK4zqnmAj-&45_#FzmcIq7hDN8>=yd29kf@&XP~Bog!YN&bE&f6K|=d*@G61(|4m
zdMsMVBl=DHdXIaj&Nn2JoS%24WxWO^J{iPC^mmGa7CKfd;`W{N*S*nfAFQp4+Joq`
zCa#{XoMtFD&utcdpx5to4L4i4alP;S`y{gz{@>4Jl17|q1pj^}e?OCz|C&CDM+x@C
zv`DE_o~&FT`Lx>k0-|G!bZRc>B7uqMiO}%f^?p^2ba~YJzHzTh4_-?=MJ6CzMMT;`
zOBCQxbxKbl#;6btu*@}pGLCcusqF=T=dR?9tFo93TI{31mj)&f5bsw$SwRhA@RHap
z6iXlBn)~-jCOP)MpUK~H^7r2PQ}oIHmz~%{4Os3E>tu8{cWi0oD1pS&zqN1t#h44?
zRyv4ZE#-$8;|6Swo~pQac#t#jijxB8fhe}7D}ow6S+Mv&;Afj#Glo@+?gLW;EOIS&
zQsdy`gi#RIwIW(Xa^|SC__BQZNIjpRPCJWs3kljzc&8I(1Qf(?jzlVd>)<*W>HJUr
za8WeG!Nbj>jC=;wc;rY7fu#&mtQ<)_*`T(7s0nK6&eJ+r0nU`sHy~)O;30wC>M{gt
zXAex~^gO!lx%Gd-LQGzRMigk?>#0OxjQRp85g{Iy21f7(&ppr(rxI91&eY$c%D?Jv
z{ufD{d<(7t+k!40@M;vLl8XpF^$*jldzD4J4kd4<?oD4%?&F{QMk2@dmUWr-0cOFc
zQ=kEpyVetWNMShSH=J46mGvMVbK~`HMB&DHBG)=K4uoh*T)+VNN`i_>*RYZ~mNd~)
z)u*x8{LPrYV-FAWyDT`@@YcuJf6v<&7f<?}-?XgtQc%XEc+vGn8|huGq)R4^oOL-8
zcso0y;~JjFe8qEoy6YE8-0{+v=>5bm@aFQSI!}$hM|oWI?T_#c^%X(GidWH#SUSg|
z={t=n_7@+O+#F9!>yuDle``HwLPqRup-q<_6y5Y|m!TIkx8rLy(BvF6Q>}Y6LXfKB
zbIR6c<l{rLySS5E5k*pB?jJtI-rK!n31w}ZChEyB=1XE6O>Z=ieaqeWA@p*TtdQeI
z9kR&KlZh=on>Mz2(dK<ewEcAX_(ScOp}S4y934Zab(984xi6~D;cY>2oZi82x6=qp
zE4f*)Iy*JMi1mE1bY*$|jV0?ZmLoh`weS~pnn(@NrS%B<HB^y2YVK(j>$CChTW%XE
z8J4tOI9H+eSc;ynR<~EEOww-z{ZVDqw@yRZI;+Enb@6yOHc(x{epG+o9PUMXZ)@FJ
z_O7lmFnF~r55k`Zf$(s|uESSTQvA`Jy(u0-n$>1S5u0ec^0dRA#vB!ETXlD3rLK8z
z?6dVp?sFn!J#;J_6Im<i7h7<WfCgKn{8?O$w8)x2se~4xS^2C{lP`YZbMvd8qUUrx
z6Xu|yuELw(Pgp*soitR7!Q~4A28AZflcbcICyrSuS8vSwv@%S>?fIfhwN0ElpFmn2
zyo~PLl78tz&{b)Xn*{y3WkU<L6qdZ+qbqUghPaA}Nr8;aWo-d>dFFojBSr8im2w@P
z=dyj}`<w&MuIF+($SI_T#s~MOJ$Mp+Jl4tbxn{%7+efz&pF)->9~II(4Q?0QOVy}a
zOjxmO&V!~KdUM^<H}(e9hFTgQRavv_s_Z!}goo(#p)~*EYhI$e+yvrIH2XOiIo3mt
zdxaN)-X4HIK)oY9wFrz43uI^9_?*sm3t|hzeTpl@uZCWoZGdx2VV_!#FDc)6%c<V&
z|I%+vL7o!*4Q*31)OWMFf|E`-#MCJyNjAhgE!dWub=@`{yL$EA6NeA6A7XsYbH3fi
z?IBO=XxWRMo40`exU@a=tkN4D=IS#G$PNodGM0`=+phCS)|AfHavx%@?WTZjLN(B&
zVT%KdQq7$$O!DSe?!>j%zlEwQ*Dfr1B9<Q_Yqd`M9JhzSB!+JX_20;PiB^z`1Uef?
z?Q9~y9Is~eh&!g{g<8rk{-S2va7D*0`bO!Ojm2U&pT9G7w3M5>6!lCPQ%FHNDh81_
z)MZ+OQ25%sbtyESfGE>K`;)it@hkNoQaZ-EDz<5BpwfQ6_KbYV@vuu#>=crXfCRkw
z)V)?Uv=3gg_E1@r4b5~#QPQp}-p*F~?P5VW<v#6Kj_qmV8qV^g?`%ph#rZW#(hC)P
zo0&Quw_A%g<&#adlh4Enn_fAwR``+J%EQK^H)383%bCp?LZz1WiO|fNN=tB@)_5*l
zAHS4E_0E+c?K|)dj<Nys`7EvLjmKY8{WD88ZUT)#7C$1X=4@|LY^d2yZ6hpHfL9#G
z_S}9Uao1f<;L7q(j{^GUoIg_($aC=AJeg~%;*3qB$LK>g%pL9<yo&qI-d0V$Uv3p+
zwrqt=ra#YT>3*l(@$s4OHezIBWVt%ScW5qA=g`9<@$X4OIlA%PuGrpUYTyz2o03Ai
zk(WXNoAl0%C+#lm57TUH=E(o_YB%qa=7ENtM)XTeGaD?gJKyaP`kKb>I@=QMi~Xn4
z(!x!GMFmBUH(tWDtKC}sPVB_+ev-_YJaHFXGqSIl>Xt7pa^3Rw-DdKEWe8pMa-JtA
z!*{6W1O&SYdLcNli3h)SVx8>9`a6C&1_%OBn>mir59(NLE9EQT88+IGW2Ozx3n1Rl
zxd?LL%*WVClf?m=lHOCzynoMc5h|{*!k;9hq}ae!3~|7#Y*F>5+EHAME*Elhb5;mR
zyK0MzwOSRe+;zM?urs4u-xcb+M7y`2E<?2d&ZUK9=HkZXwN^fbd(U56`LSK?!RQuC
zyPGmMLuH&eK<1`Vl?6nNMT(oTwotoS+Ie<5MFW<?;r`{4=Lji#RI<)^8zKDD3cQ!_
zhm<LIPChXF7S>k!XXXiaHxdn1XyEJGo}1er`s|3lV}f}hj??txNiE#zr}9pPo7M(o
zpV)0*h)B4B{Ln4mnqbfp9SG)5kv^ulkXpYyL|#+3oS$S?EV*L2-rcsF-Z|T=mFCP1
zd+YQ;kn7y;C4)ipU3QN4(+QCctR1=Zpc74Z9fd_rR8<dYcM0zgc$IuR;KMn5&)q}N
zVu_^)e4r6l3}22;;SF#%YauOQ3DQ4W3O=CmzdS9~=EX0<CM^j)=>E|20xparv`{W_
zE=uhPz7mx}xdvGepxo!u?cH$p@)jwk9_62NI0eMteNI;IOH-&0USjP2l69Y3)*pe7
zH^SZ_EK;V_4~=3j*R)t&Q$K;Dsk~h7()!-+WJ+LF)6kvOtBrKE);h)c)Qcr{zEWl=
zH!BL;NN=X*g|rHA9@zXK!L-o%Y|$})hoz36)YkLPaoQ@(QAzHAiVf<a))7)^pK}v2
zpD@=ZVs)5i^5rj3%C$L>>h|jX*uM3pse4~#^n{G#uVO&qBr-_6@jBRjM`=Nqa|E8T
zG(jN}u2Ev{+N2lTk4Y7I<_yN}HavEWf3R=MR?F6<sC^&YfjC_DdIQ;kgx{Cj#O1AS
z&PkMAa_Egf5mb~v$oqm=<w{3_h;bT=Mjw3BoL)-cv~b`0UU~Dbl#B=UdN$J9_1WpA
zOLi#?UgL1AXq4NiL%H>P_RrL3IYtzHNo^cYL0eETB@>b0`f4Ub{N{ty{@lv+uR(<B
ztu14PJ&JyW`8m2&2GiR7^t;_BPPOT58E*=KIppH-BI9PwsCwi9YEZ923yJ&mzB*W*
zIx-v~;<tHA=puUd{4oK^9nGRr^LLO=+0tsTbj8Q?QkZXvV{^bFEPAJpEmp|Z=?y_n
z?W)WoJIBrv-IGTc^FFv%@^%t>%c=S9%jhebAs&6v&}S>2XOdb8Y3BG*Ll39Lg9XDz
z7PQGuYhQF&!hUrU4Uo4JMm|p>;Pj!-9>ztQL9fVNx$a&}3=2UED?QNU&@AzW-+->|
zxH2b`c_FwoZ&N!B6)K#GyT$H%+~}!spXQDka?vyM-LoY(xTO6kdbqZz08Q(~w4!^>
zsN`1X09d>`oxrCdr|5O2Z$x*|m;A%By{Fa}?JB<ZPU_s_G(#r2?>*j)?mK#jo=L4~
z*_S|58O6}H-42Xc<`iaB<*0>wksWL{<|bpG?yx~~huGkbyqXlm#nKYgM?&k$?B+u$
zlmJqS;QjD9VMcdA`{BN0saHO8<JPVXkcBipe?YYP(~RJT(X?>h%lJ8mdYr4eH)lDr
zQ=C<7E(y~L=X2^}f;ls|eOGzalI`r7>%qLap_X#gVRnqWUNlmmGkD&k%HgBTl@4Q9
z{EeOZn-+hTv3s5Pwm;DU<}-fUK(U7Ar#LX5ty8{ubQ~dJRINO&v)$9B#S?At-V5>i
zJaTE^$anyA<H8tx14UBV1JzOeNVh5UM*q(9XCF)HEK!imVA;mtys_z4-YND}azTVz
z<v@<wA_-f4(>^I>H8-Qd*z|rvhjG)w9RK-STJrqZz|pZv;2(9RUSKFkHJq-mBd*ST
zue?Om_Q+lizSxbs12#Hc)_t={=ypDWA)OPAp_8etJLrTbEJ?8Oyoz0GH9QhtTs|5k
z{ZS01W))g-t4ZMFtCEdcuBq;inR@WsE?PEI0(6IeU*VJE*zCN>ChO|_gWk9e$HQ=M
zIm?FR{co+vGbt11=+!{ecT&kMilDQRpiBCW<nZe86^e=>;`+kDC+|KETcXP`p6+#j
zq(T|YE)miza)qO)g++~$q)fNZ^4s6fadnT^<qIy&+FhgWJ%^X;g+q^tV}%bky|l<<
zk%2F~p62YUWWKnIJ9mVAxokA<o|D)O#0$knse>1nWu4@-?L}QC@w&@XH|1}}b5bL%
ze0XdU*PTS@IXbLVd9!j$S-v6KThpSJqetlvKuFiIm&9>Dr~%22$aT%}o#y>(0-Rio
z92aPS$Y5YbWqt1J8UDYfwC%r$i@#5s0iWl6iKQAU`a0uR(Up-wDND-|=$w7Zt1W}r
z#~WT8B-r$&T)9Qq-SaX}!+lrCXk;2)h<d$n@Qn1fLFEMV<LZUN%O2gD@3`<*)|PFp
z*uoJN^Xd!bZo6ugww+eKXHaX`o}Q`AOql;{zov&@u>7DKJns!PZH@YgmVs*l=j(5W
zG=B-(QFRujW^QpnsahNPd7MuYIo|zw2n<6@GjYmtyf5zIug9GlZ!2u{?B%1S^|8E>
zT%7S$AbDy0rb4_ZihAIxWvH`qn%_x(#)ju>2+a1Y#2VM}IqpD*ur_nBBk>=Pfdsv-
z8hmNn#b9>&TP8a6K?Ei&-I=ZQPR%J&5CW=AghfbRnN(-BP{$)WwKT`V6(m61<ip%x
z$<pPS20bMHoE7mdYOUcPtixQ*?|rO)`apu;Wv^bEWJieDIMUrKoGnK4y)vvQMK?Q0
zb-tERTo)E^mYZ5+6{E|uF*<ooyMdvm%uOtUexhkrp9hssp1|+EfQAns($!yASlK4X
z$B*V6Tl;~}TW8g(GxrX)r*ZC%oPQ7cNGTa5pxgyoNPZNF;asbtdf7^So9Iu4BB>+g
zk)i22^N$;!5_EpH%#2r~B>}G4)Ky1GBj(3A?pTZr-F_S~WZHatodH`(-Xe@@FltVV
z2&3)51wkm54$@EktkL5XzxTV7lS~ds8-@nS%7_@xjm<nYey1OW#(&Zf#9jqx;-vw%
zMAwDFcr*5cvk#X9^Vs?{Ecf>h-Lg%m)-e)UCwd`3Tvtr)nQZVmgp{uGI7o$@(|W)`
zr8xI=KC0LD^XY}wI%@{+f_4|Eo2_hlHw*d!=0!Sj7vy6~4osPw!hZ*@aEQwfR$z@5
z*;yW4ejqn&z^>u`jgz2ywSc~c3KgQbG*E}=TrV4<d_JY5i!MT}$jS_nK1mGRcupAE
zY#0~fe*?Lbe0CsSNq+}ZFP&-Lo|==iU2xuz<*pS%T!g!xl9*Ov0YL;`gW`ABZy}(1
zpgGDo5f@T%^4)c*TZ<}_wihheby@U-XP{>*-*wLgm|nIp2i>@Z;y-~98$s5Oy+kc^
zM0%5*EY{Yf)LkYQz8zeW_-KQ9Oz5E^F@&tcOH{_iVfF*MZ3}v#OLl&}&Q;dohh@#V
z7L4U%;uoHTrJXCXTM=rYiCpSPjBq!ja+R2l^*J{qgX)}b&ui}E(zo5(6s)V&y1UB!
z>51(N{S&P$lDIrwx(LK)e}uC8&roe>6%tOL-;ipOKph(mqt<E>SxcT3+FeZdS`kOO
zE9US~h1Z`k{yv1JW|+(d#hSJiz9?i$7}LsIUE-MN)q$(+lAqQ(Z}64k%erUTkrF^>
zC)1dPuoO+Fz(B&fQ9jfzB~vLbX2o5fT`T8oe3H?B;sS;%Q-l`862dlv%|;zNL?<Tj
z1dkO}M}^*drMLchjqj~&-L^G1giH7C$^aHo?`sRJAsKjHnJwoL_anK+Mb$lqCFR{O
z?!FGn&t&xsH|A|;hrkGzH%+@~88pXsZwgB`E4b0j!$<ezM=?rK#2KSwdB?__k$@Y<
z($}ysLDYtV;DL|PiAirZl=SYd8)(1p{PNnC0rTAGd|Z8Zcb&vN#d_|O>7gR6Cu?Nn
zUKPq6^zvU=otH+>rN$Ma#MQZC+%;&&pC^;}%9LD&?7sN-2OcV$!*gK#@<FP;yLmIo
z8<lSL(oP)B;lf|sC`D#%v!1_LR6Kb1I=iD53q&^D=SDd??-4$?1%cpYEvF-i^aG>!
z45a9-EpS*%g{WD*F8647jHIgnjWS%?@~XA>JdIu?9SG%$7ztx4S46MsOQ*f7-_Jdq
z9ll_9*)=Nz6|R`Jr?w5yDL|G0!k}53l;Cbi4@m_0+L*>)bm?|CZTo@(iKOJx#ZCNY
zEsvzV=P))l(%ee7+Cj&)R!HMNVi%{Pz2Ozu`7UvVn2i>P_2qDB_uONIZ+t?C^*Wx#
zn{<sC7;-Q~n3fhcb=YLZAnXClg=HsiE<Mh3D%iSi>89F&r7icEOH8pJ@oz0SeD}}U
zQ+NJVwhuB?p@=Bj+@920pXXi&FbEpB7jSg64K~HuZ#WlSP=B&V>9+WSb7O(`&#t(s
z`Rd)0yJ>CB9K@<^ihFB@%{SL*S0B|glbCzr@?6kuqUG9-<*R!|G&w!jRwy}D>e{b!
z+7;D+c=6KzKKGoEULM<G4PJUplT}bkE*49dlDe8Ax`V%KZLffru4?4DlNi5MP$>QQ
z+-{C0at5Zu;vrMP-I~6Y#(#*`(HqyIpz&rejcbXbl7jIOGoP}fFG4Z%W8W5U4Y(nd
zeMd-h9%xmu{(0po`jV*As+U27leA>c!&|xDvU~TBze1Wv9b7E6F)reA9))hRGDfj5
zi0@6J6;o^Y-r~%64jp$El-(a1OU5ql!X|Z;Ag^0!RI6WPnz#g)9(-OYrhn|A%`)%E
zkaNdZHeTHv{Mbd}V~KEvkr!j(S0{w~_xS1Hv#9-aH&{~~$9CHZ!n=Ir_#h<oic24K
z6taefIZTHqJs^&WU(9IuiyRxXR?=mS6)s~1H@eV#?uXo^61TgAu;b4*$S&Wyu-Pv1
z{-W-W1qDV&40=B^Rkl;tQOxR6%g<tgvM(B%=P0$`QtqvAu#9<cU%=bLchZiuE=3?1
zcs~Sfz+YOJp4pOs-sjb<81R^?U%M#VTyMn{m6tBBJ)bQ;x}$06W?habEIiit0E8#{
z8$g%59z}S0UIPdS$bz(<ZA>eu$_<qHzH|lgi>t)gJXDv4VkHX|ZDXNM7A*=N$;!y8
zT{I{4@Q3V;H-(SN_3enu?^rTm9xzEj0Eg%eA*Hl;$b-f}&yBjBDjfg-%ezRN+1o4S
z9bD<{VwUxIiyU(8g_*+z{s`9lZ|6hZy_s4petJ30?MV9de9|IWYLL&_c&*z}Cm(4K
z?HlLc{cwpxuB6Nw%Q>sdl>0$(?qV0`rqaSb`Q`y1m2HXPzV7dM@7k;TUP#JQOBO0U
z<}Jx#aaRlBB3cZfhPxBZB<7d{HLOL+f-m{<V!b=EHB0k0?J-+cxc5n-sK226E2!B3
z|B*6m(MO_20NScWzAtMvXf8lMKuL3Ei)Q0ktKvy?Kc=W=>iaVn6ja}s=ZFVJ;q2x+
zW46!Paj^;!Y<T_Uz)>^nCM(X=XJ32Ea^@YOM>Ewa?ppNrCgk->TcZyJO-WIG;)@3C
z(+;e9-Rx<R(I(Ca{B0WVR1e%g@_Vephcvh@o$U;@rFM#RfDZlB_&U&2awZ>#)h55Q
zPzSg*kqLvr7g3WvB*4U+-}!^|!K>omVntykI>qoR%ix|4{bDTp!P9CP9VhI{E$^!^
z=O4=_iH=|62y_Z(3cb`2sZtkc_E0Q<CA@m;PX*zmQgberEiB344szOdXhb&Hm8_+O
zaKGDX;DbutM{G$jI6!qS(evQnm2oFG>bmvqFrG%4s`-iD1nDPVK*zXQ>vGl)F)sQN
z;z!#GC>Aj+9=8w!7Bk)j**OS@-PU_puX*;U33tn)<vjX@GUwVHiHXm?#canNntj2B
zqB2`aLHG?72|8}P{)Kar5A9Kxcyd_7hJOLYSrW_?t@=ZHak?J;*L38sI;*2g>dsMY
zjI=bIu7zf7>99rin0q@|fN;ePcW9&*>QvB_2Ib3hM;~-j#=tCwAe%;@7Ca{N+d=5@
zSBE#>Z1PwPSKD6mfB0BdAbvxB#W`D`udcZ58^$raa}5fiK>^ST&IQCNmuL7PK#1?#
ze+AKXQU<g^kFB`#f70P+jY8Ghx5EnGHdfy%>EGbbFmYzd)H^?hD8>yW)^$)hLk5{b
zm9~pK*8`j(U6U&TGe9+ApIQ}ZAQn{8{kun_q^vp39r79L?n&x%gkBF8{K6Yu$p>mt
ze)}ro0h$%#9yE4uoUdgy7jo*0(ck_%;q@OncpD|qm836u1~orlQY9wA^@Z&O0tEt0
zd8Buq8<mm#Hcnwl4LUjTR$K}dgR$7D6~!D&fL`ryWe;>6zVMHh_8+YHKRs+`9U)LP
zvZgK2VfmFK0r9-hIh*%C1{+rLayd6Y*vKCfso8$Q$>??mqP041l1z<uRbh{q?*k?5
z%Or@5ZqVj;VeG*_``WDQT?}+1-t{-L^6Q%SG-s#<Zf!}ZYXlOzv|Nu38tmPT6qNas
zmgbi$+Zj#OxW3pw-0!Y?3E#Oxc5UoUhGZNFaV*b)`lE-0Iob(RYXq{CzU{cPpiUe5
z<zcB+mo6}Ek?9`y7m{AT+vj?A0wF%k*Xq7vTKjY`aTg_F<CI>1YvVh?_A4A-fhx+b
zj<aS(U+nm{SHRe~-<HDcmLU)77YFB*H+cwK6Y4eJd})$AD{=9RtJTP4Ip=TtXk;t8
z?I5Z@h%NXR@Aep|dE3f5HB8Xkl`dXsW|_k+$)s;vDY5ksXc!(1_Ip;H6ao-2-|k3f
zN*!cs{lTLc(9nk8#@YFWnQqQnUtWQV!MKr_M<B~&KLu<hO|h(sD@p;Xz1xv@ULRC>
z`*Eu*gX1uIYT+<YOVIc{MYaR>i|zf$`RFMH`0Tq*Oke$n569{3p+<ywcu1w>9W{Bk
zCtZF1n8?|*$>~$etnomHM}{Lnqq|v3@AEdBKcsY~PBcSjb*2!xI@9Zf%=cI#mmKAM
zUdo5G2cq3PFwXTE0!{uIb0d*!zu)Sgul|3SmK5wu*F9Y~8-q5vy)q2=bQBq7_M>aM
zrsR|u{8LH)rTG8)pd8&<e=P^H3?9MtM9i(aP<$od61P6hdHuExo!rVA47Qe&g#z<7
z1q0vWXs1j`#J{w|OTFxtB~^=yZAvdcQC1<#*CYR6+Hh@<jRh5}Du>qPI@+(d>r39U
zrPELTgM7LAcChkpL0~%CWl-GryB_;JQ=p=vw+r7`awxOlQ~u~EXC`GzR^w_}AX7~m
zcpU?%_9K6E8%?c%@r;dVaJ?|x_SMy$%~w%wb!n$<p%K~yA_x5dVcBgVQXM~FnRfl+
z!KIvL5qEp#f~L0eM_a<6NCqRQO@8ISZNR5PfP@D^1z8$V=V%6BRvUlX{G4|ySB4U!
zTYU{MuK?1M(cP=LuH;fn!t9)t=1Rt&IcS-gkKU8;U1dAEMzgLYLOqWm)rbE`dH(~Q
z{!b5^1Q4za-@9&>NqH;4>2pMdcNGcj_2G(f-fr|rIrIQwkNQ0l_D{d>ABZVNOYze1
z!7A-1^Yd)@>@)1zUvv@b9ws)n!dk`?2uJkrGEM2j-yg{LOZqRs<UfTL|KKFn-O@tU
zb|xQ6zo1k_ZnS5oM@z|lw(CKkFNAX+|0M>UGw@!sx!fQ_-o`yYao_H<E_232G+s0d
z$Qa}>)iqA6tHht`AViAfK7?MGQo5mGA(I?^up!ZO(1h9)(KKbL{yF^Ic0f1aSatj(
zgRC1Rk2{6=Dq|-{tMoxV!BFS&#NWmvO#Gp)@_wti(y0`d*IB3$&6o>jL_q?;kZtZ@
zgUp>szm}hJH@WOUpF4{nI1dSX9SQ{+U{Fy9i~!;VqtD~_e4R8j<!-LgG2LTS=c$HU
z48?#P{fnd<lV_jyW@d`(G{_qF{yLV<@MVv{7dRY1tMeOCrA$h7QqO@OufOIxekZ9D
zzYj>vH2pC+(VB~YKltMSGTRYi&wVeK9Rt16?)h!(cIejulv7z%BI(i4t4-j;PUtVT
zi_(S%#(7R-`%78npjptY%ofZKlqOO%?Ts);(YKFse00%+L4rUZ0WLQWwa;5Lfe0fp
zttJr37)wp=nSjeVwZJ6~Mg#1Scmni8RpNL%b&`fM00tu`5Er0MNANwXSbkpYD$r_>
z{4<{>5njr8fQEDHp7%IRhdfhvCmgxx+?q9AV`QML^q4|mR}!4q=n`U=smVj0hjWth
zjYX0SB<)9tz;-q`lr(OvM%z%9*wpXKg)D6s)THg?u6<FGox*h5VtkZGSGF~?+&vlq
zAej~vTR@x$^BRCIot3=9v=#wXkYl2Uo9Fm}M_aY6Zq?{Y0O>XQ7*GJDgrz2FECYU5
zRV-*E1@IHg+|{6fa8+78n_kdlW=33e`oPZ3*-vF2H|~AiO*Xn5&wuptmZjjd+-nWy
z2M9m5jVX}!xofIQVqNKnD9R6hc+YlpXLv^?TLMTR7&(NO0!yBP1}<f#KH_*p%B|>x
z6*JfRfAcWF?>k~L>pBM5A#y29g&hU-ac~}b%y%eKK)?ask&ZHAd7rFH10MM@v!G}2
zjwI_Df0+3AC4t5S5EXuu={nQcNgM=^H?)F%0oQR*4vqmQkS8!YDes=m<Oj`Mtyo?o
zD8?7kJ1~57FZdh$;|atO^7t>OkD><Uh)i&<hkRfVR0g1tjG@m4?74=#3up6;|M}4!
zzv#&I?+>b*@$|vc{jFrbyTbq9DA{3V&xr7A*U3h<2$m{48*1xzw;Qg3hCkb3my^lZ
zbR6q}K|o$O4!)NV*Jg^4uYu=+>_;mqlI<^!0Yio@j#&_>I<Owv6Y4!$O#*ZG5O7c`
zG(Ha`LFd`3rZ|3o`NMEFPzPAv4;d=(Wni{uxOv6fY}Oc<&a}w#=^#DqQa`B?V$Z=W
zG$(=FF{C3rG)c;NZt<I4Z_*G3Yc7Dur(ol2(|QO0Jjm_qEr0Lb8(@$}GBOXlj&DzY
z4~A?Y><@lRLva=L2FO8t5(_Ug7jeiF1xmpN*u4Oj_l1K9OgI>Kv||iKMS)9;B2Rl~
z$fNu%b|gjegIen->{eFX)kI1$n%0AZX|Qq8Ox1J13<QaS*v+hN@o);;9U4G(Xo`z>
zu)NP|JRavOOR)Q{04t}bqHvh&c%xfKv5XG3*7=cImjf``5?mcrYsj;Cnwm3aAAomY
z&?8l(2Zl2V?rF&TxJlz(Kqawt;7sCB<orFepKM=0y3H6n23)8elj3a;EIqJ_Sp>#{
zNWtZ^N%OQ3LM9+DEUM~Oe8;jsol<#Nah%_bZ2(_BO}<8ITLsz%V|$D*{NkXd<jA+j
zDg6R~9t|>_xeC54*M0!Z?Lz_9@JI--qax?OiWvG=O*8%z+OCb|Mw4j{dYQ0kZ!NsS
z<=TbSBTk_T>1GBRqrMWQ{l+xU!e>%0L_pp8B(rZroexVverC@Aya}reMjq$_xznQq
z+=^9=9`qr}KI39={Pam3mNKC?8HCsb(JWCcL<@(ZXOYNGN--n>bJJz?D;*(Wm*O*a
z>IT;7^%s0zcd?kK5WbvSJ`m?`If_iRM^o#dqzbn17SMp*+yjhaw;l2tO$8-KmP|C5
zdG(MNz9itTrirTjlE6HLe`ESc+qd*ZX?rl#`jXy(9?dsd^~($B0=&C-`Gdfys3Yve
zH%4iG-SeDoQG^T6OC`K|-8AIYy(ZMVi88**VjdW4!vm!5Nsn2|RC*}FcnU{5TEqkF
zTw04B00PM$PG)(pLpQ>qVWvM`{s2^y^5JM9Jb?&`A$@{VUz3fS4d{$}?1~;uG1KV-
z$BjSh!8=s2BVeNs%N(4UGm?<225%1!W4@R`Yyu<de&B0su0^v&Jlm~eTm<+Vj_5m+
zLu$Xj`LP=W-@L>Qj!huWfOfa1d*cGYJj7f8R?+P6Y&nGdWg3PBX#v?Sg|FD=Ch`|!
zX>3up1B$UQQt<e<FB5Du7#HDX0u-_ne8<IU^yHMkGWGjY9l91g_L2TnU$gJX`kTLz
zC&P|~&LFR%IPohBo1l4}rc!;^ZPLi6^m;C~-Ftf{<Jg_(0Q*6I_slWkJ#=#$WeKSh
zLnXnS=D<-hN8r<IY6gh4m^K_s{a&jIn-2zBFkg=Z>m}$TVZKeY>-+15*{aZdy5)(i
z48N7~HW$R?Gxlqrz4{I*X~^r3pE=agwujgNW-&&WQ1n$M5c+lR2Ne(^;L_(nZOL?A
zxI{pHa#SmEJ<)(J1B%@U`RLL0HdqEm3IXZVPXXz_AvKT?!D;3MgKHtsN7BDWtK?=m
zz<)z(;AGhrQ3i!35T1%7v00)l?*tflgVT>j3c7yVbjU+9tx8N6We^FJVxsOOxsdr)
z<)9{hQ6l@FFfuGEFdD%6y$Yi?z<G`a?E-jROb>2yjQTE;4`pO17FumPxaX^7)c8G<
zn&B`Oa86}=3ecm78Zc<q*6Df9tOGv~-QP6GUqOT5RiOecbX(I>U?o=s!QKk)Qdk!N
z4Q3tF<i4N*=$e~xfIc8L;MiiR6_}mPisRxbI0hSVT~*gLQG4OPm2ffXd;<v0R(o!V
z{f2F~@xKHXPGAxk-$?Y!r0|1%qy*>(ZFol(u+Y;35%SeN)8DS<5}z{B{{xlqqvSNv
z{delOFN#fGeAaVy*-B5jDhmK5`AeHb*}}QJV3e5<3lzP2w}Mg1eimL6h)dUCC5|p7
z%suEOjm|yU2OY#<@{2%3wp(4kmEaShmYrFXv#?cQlSOXIrNon$-#Omax;A%z+nT1c
zmP<yKd#Z|0KGTL8bQ=o91j)*<BN(Lpolg1L|C&yj0Ix2$l;sx(cSD^!e&y5E;oE)L
zZaM&w#k1ca8KZ$epO0KgYE!_Cpc(VxELH+1jlkl75{!R!5aCce<=mA#01gim%V;Ao
z=Ej8@>;r%2L1dpdV9zmr{goB|RayG~axI_3TSQ1PC?H<0SMgGpYHeufbmVIOyxFBV
z_r=}4Je$OW3+dOjKZ=w=h3D~aL2%$biONkX@zjWWlm}QgpEmX#;F+(AgPQoE02K8T
zAsP6_BRCcZjuwfG?TllD#ydIN&DnccbN{@aSLMJR{ekggM~;(WjfcQ_0CAurb)>d6
z_VCznd{*s`movC}`Piu;_K7B0L_VmrmeLQIB3{;#+GHFj5aw9msw(eE?8Xi*%mU_{
z2)H`&sF@yv9kA{5Ye}Dj#>0!IgMUr2TeAvLL>d~QS40#uhE;4o?KQ%Fim-OFn9i9W
z!CfJ>55OP6c%=6q4LOd8`jJ{0ii|JlVHxwjnRGhB;rl5De@w=kL8Z?m3jn{!^W_B`
z4Jd4)W4!7#=>How2_#UBz{u&j=-zoJz&-0>c`=RX*@p}{%#l|}ZKtb9J);d^_=keS
z`%1bVteqeMlk7;t$18bj(6iVdM$b>P+Q8Imr~oy-76!iQ3$HyuyJqyf{fQ*fL{*J~
zr@u^%Fjj7568Qb)kK3BTfFC!Z_(E+8$STsXkfs=*!q@cZ&RjMpc(RMcHZrr~YrJPV
zMf5BF^*qq)oJdA5<@F2sv)FM2&~4`T#tnH^d?OX6^3#wrX(0O;5PgtOqTxQJF-XIa
zf^t&;amTbEcB1oi8v^oG3>HAIELvf%$txSqn(-D<GRbgHQl;sj%E}cy_8ce_NE1C8
z-UifjBl<N+Jel(7CV$HnNAh$pa0Ak-N;Cm(WNSZi4A9ZMFG<qqvk5xQ2F1h$EDhg+
z#jcO62a7I$(Tfgf8(OBj@u0S5Nv~s{h0X*5sqmiEdH(~{bcVn>op*wr0(gav9B05Q
zET%AY1>7?=Q2+^32f?)WYvsN$PZ?kV;?n6hAn6)WtL{b<D%6;pK@|wNZwg1*9Ia9K
z6j+t;Hg3ter|aiQB>R)w^xaf?f%mpzths{iuZo3lm4ek4V8}X?7nW>}eU;Dgfy-EK
zGzzT#Z<$eV--~~}Mfqo7^WxtC|91lMEYTt>2<+cp(ntj$B1_(*+g3P$HIf*wQGnhz
zLS%5!436k-!r-FJujE_TZBq9Dx~CpuBPw|Z2PO%3>?1T)KWKcdv>cGcQ$4_kOt2lw
zBJF~j?x6+`sT+gId*x&})U=Yf$9!_x?Yanq$Xci;96VdF^*6!W89^YurQ=sS{}c2o
zmed~&Uy;8$BSb8zHA{fFr^v4OJq_QC*3aj&?Zgf)PG=C<bKVn$pjQSkf>am#i({T!
zyQgr<i5aFTdO8L6HB1Ce{GwX+uQpVbYlSKzwV!KOqtx%OmTkpZ$Zuk*f%HAXVQ9by
zOcq5%X_>?Cz`%|=EVG&PGN`w2C+q>}$vg=#&U+{b<^%d`Ikpf7>Riybl*d<K6H(Ot
z8Ew7=6oJ-IzV(^4gJMSmm-X5XYW#lhg$a%J14s-o3^cmJ2F2LelDCuyyyLS2<0uxu
zKgUhL|2<wO-@^A*7~v@8it)LqG2kF>7YQ;0#OoC#^V;MNF8$j;1^l}PY8>!i{_I%R
zHYzWMI_5hP#8*z#jj?%iA*t9&;U%Jd^9`fqSMO{)(DaLnmjHUW4|u1hR}d{0!Jhw!
z1TwUn+w#DgJ+cQ{%WE0XPXJ;12GddSwPFf9-xq+?NF)Nd!G2I#K2YsP^-yd+Q@Uzi
zFF5AAFC8tNu+GS7ZI+0a_q%gqukFL>`#kwNC({gNOag!&OrtP1*hVD!0<uoonli!!
zn)vZzPmQ~OEE+<jiYte;h81Sqx5O5?YRY_Q)B*}I?Yqsxl%Pe>IP_eWI{kVJYK{ec
zQHLQxD<}A4|6-PQ(yc9Ph2Ctbcxnrt!|YqEd@Vmz0ZZ#OL5x+R5Pc(xbvl=;8PiwQ
zMDar}R8sV<gXd({w2{|wXMNg6zj;DdDEN``s%St6C$El4G&+%STS?5=iK|m!C(wR>
zho67VmBnp_Xk)z09#(hi;bcH2XP_y-CZN1XX$02&vA_vLPYW>Hy32@R^4;hL19+cX
zDaiE#BJ<=1`OPf^CLLc64W7dal6tpA15q9!E+n31hm+cx@}!^-Zb0Kzfu0*??#I~~
zc^WP%Ev5eMD;{AFtt31?Jyz$`J!+IaH}dXHt_^MXIYJt9T48A<b2mr|*a`H1{@l2i
zXlv5AB55J1_gF1pub=Lrg(H~Gvp)10blAW(h_(`n4FWYpsiM~g)>>AN=&o}g34FKZ
zQ~c)Y@<PR>?c8`{avk)%g7p9mM6IzFJxp=XM5T*rtfneCZ+w!suIhxN)BJ>InoD-8
zU*<ojV{huZDlkvlhKyVbudmvDvVL>-dg=4Xv0_LWuL?A6LlJ6pVK%tE@*@cdNS~ay
zp2`hc@<;nC=_hSH_CR}n6Jad2F)V?OHDaC-M_*Mc-)C>5cF`p4roqa}D~s!|+@C*a
zJMKdSf%GN<Y4C_6k|{c($d1-T2~H-LU^+qN;t5!f*N@bD7{H<GI8#<Xu?hu7K(cNl
zMW7GK0Pshdq;N3ElVD()>-oPdlK+e1vA@91w<|UTm0E11^5(O-OsLxF&JQgrh^2Ev
z4!^D3T_s`bv-<Gd6?&_;jBt`x0qq<GSt~MDXAQu-<&&dtyNE1d9ax`P2#s#k0vjXD
zn4YdFO##(7^33f}l@=MIYT#hGAi52<9?%c#fYe+Xf@sDqptSxo7L}vdf)c~O#&YFo
zwIpbJm>tZ{q#++TI>_iLMVL1SIIAt%_ulaF2M#PB`6Fd9K(p^OxFB-HHS}<1T<dSx
zl6n2-AQsjhUq>3;OR}0kFogl5R=JGm`!#`ts}ICvvSA%C@B)OqLZ3jLZ~W+>nvk7M
zYJ)&z_nN5=?4Lor`WtX(+#;jfmO`HZM_{teeGStPUyoe~&v!rmUY(=4oPTHT^Q6ml
zyKav*UR<@(+C-cCafEXIDlY!&JP)AIL)=dSe+*lQqR#{3T6Teh(O{wgwRWQbs3*5!
zCc`VMKzK!LGQ2{B^%8>$C{!clbpTPXr)=XJD726W6xhS4uh$N&>r{PG9lK%w*{gH7
zO15{TBJf08?4U5%J#GvWbK9Pr(_}G_$uc*7ra$q2*kk5xH(&`9Czly`W*_E4onyrE
za5kWdE{6i|5NI8U$$?}`kBle(=0t<|c7Z0TExn60s%S%E?$49_s+%IQgJ#1Nrf3R<
z9fKA8)0}8>zT)^gk|5N3vI;2nB)LJ;5uzt{&>SxbF1GCP1fuKYL5^xX(E5Y+*af6s
zMc^Ji#CVcPKOL&%z-r^mP=n_!03>=I3~Z19b}PAxkfg)9ALfbfl_fE_40#92J6IkS
zFcVTuFu|c)`=POh@%N_@^0-3M;9}oS3S%DdV7{kmSaS)FJc$Ac_cKy{5-Ek0L0+JY
zFM>;6ctmTaQ;ukv?#NEQp5}U+tgr)~-_p70o@7W43}BwwZIfd%>+ieYoj5B6y?83w
zu+u_xS~iC9J31EL9t~%y42o14a#+~3AjkPXLHW?|LW{8J?=;1x_&+Cj4Y2TL&{)vm
zWRj;ah&3mWn5E;GY1xV*sq9NJO(+F$k9JTA$owBc_bakTNv)b<si!Jnji#?`l{Ki-
zK*vlV?4f~F07t_ccGRB8@YB|MnG`Za)WoN7&;^(_V8>Uu48VB`q~T!k@$-3)f8mRB
zyE*s1pFphQruM*v^$;9nN)w+ww>O0YJ9-_&=$SMLxI_v-iajTZONgg_F_hODL5!b3
z93zaYLO|$qpFv%DU-I_;e9TkBiAH7xHUhQyJb(fRo^H2|opk`5YPM;{3SvG%R0GQR
zI#plxbw(Im6N&Wwy3oTYcwWsjn=F+gl7wwmdFjV9O28j%)VE)=|K*kw8R-{MeEd?e
zPntLlPP6F5lnKNMY&I<jE)oFB*`|`W%a^4b)C(Tu*R%!-NNYeh$3vDx#%i!BoAQ_w
z2!Ha>I2c#12~q3NN#x)17qCFXme{_>i?ht>2DBaIBk6Rz7A<mAO>;oMv0b#+8;rv9
zFDLi5#dJl4vTDfPIxHbI7_9-^0CK1=?7Izt5C0}v-!l<rWVuWR`GdwqFl?|1E4_d4
zepBR~$P_sZ5XWKWZPZAg)K7`H9K>HHRmc7ED5B#T*fo<6bR+<zC5)0r%Fx3LF>(<P
zrm@zyeyb_}RlTHbF?J9c1ZM@Yx#OaygM?>Xp5J+>-<3t(0?}6ovF?ol@4ojW5JBB_
zakR4}8+6kn^dRUx28fqnMYe1fu{i7NWi!=UD6G4s`M|@ct%Adp-A`bx+|4EsAzSDI
z!0(!@1h4)=^#9EY_bb+uVcjs`vjOk?3W_mDncC-3%rphNW+NMP9_}^t0u5w+@}#~d
zFd={G%m2#j|Lx-`XxzZ~)(VgrmI%)E<OR8SWYo8h4R+er79?Y&257L`s`NZhzJ3j0
zeam#I1h#iPdibHdZ(?h65bG9U9?-+4g<!8O@<d1Fy1_fhC}SW8S|?5Nasjtx8w7Hp
zI+rKS(vbhBGrUouXzG3v;9HQuCtn7f>PFIL+u?gK9|m~d0P}%0or4~BBY!FX%3*nw
z0?HAT7z9(Ef@hrz{^{F8KcCkMEc7L)3|9m7TGT+JQSMv+w0LI^7C~yuVr>O-8v<}a
z^{cot+qygssLYWL%<GAjLBOTBX~?U~JOD(Z*nJ1kj=2@Ew*{94`=>*i+#V`S2{Ro>
zroIc`8i!Qeijbf~eeL#)OW2K=TIdB1c#oIRJzqM&u2~t0tRf1*jR1<MZ)?#v(<D`@
zYptle>%1*KLDI=*bMTTSVu@!t>UVE>#VZwugr!kbF4&s_!V8o*G^|T319%WUXsiyj
z<*@{C;XwauL?Y`^bQdTe$q;qdn*gl}ClKQ+bFoc?z5r@aOGlBNN`8cJRcf~WKK3f>
zwj4c6&$-8K%<Kc5&l>L*J5Y}2B7a0ZN3$Lvy@87x_!#UF2SeVzm~N1ddEoyb`OZ2D
z#Lf#W`>Sr$sNke}Fm~6%o&}h2JMcK#fOpCZ>^Va&Z+RpqAYss~P&5yaS}WkESKsNW
z=&95Y0)GsqwX=OL46xS(dTt9E1!oC~pn>d8GpI{ed9d9>=O;|y$v4~Br?x2fQIavN
z`=W(JCUWvSw^?(N`T+7-G1j627<U4EtIT4PML83=T4f&e03f)6(LiuFV*CLVJPa42
z2QR?9z%KyIY>h{```w?9A>8{~iQ`f^;MR^wkl2#%LGs+J1K?DXfEW$6Wq=2w*o|fG
z`v!=A<*e63FNmM%r9hGMB!eb@JuR=92BXSM5>TzZnsw0V1BCnGubh1hin@1z&Z38b
zR8bQYLvmyshn;<Fn1f+6`Qhn5)8fS$e$x_Df7G*lx8Wv@$zy9ls9=)hx9UFxVhI;G
zu1!A|c+Hasq<)z#6azH6z9~(Gbp_xIe)j3aUmmvRikLE=bxH%{y!8M<08>_}b>zRj
zzKN?GznNu>=GG9xvUd6ETsMuV>u>zDeXY-m<xchc+UL74Fa;zK+c5xCsV~f{3lLII
z7p}nz*7rILwAllId$8V091Ng~kqQ9UblyJ%HC-UE&IZvho&{30tECVqsV5-t<Tws#
zA@C3JH6+nfp!q9`jwuc4+F(so@q3H1^4lpFw6d=4?56?)^d43&F+<&G0<n^#X@DsS
z`szn~$wU2Sg_H9XfRNsI2U~?mSOj%BU`munor_81n}R$+s#1#qd);Yz$zRU)Gk1={
zU@zXM@v>dJz>d<Ba>rt=@f6CSLj-`pL~}Gy(kF=5I}wRP%||vKaW|lP=g8;jItFYz
zDA=C3fa8dit8)Q*d~G8zSI=P}yixX`<|iiJ#?`wdN~!YYTYv4jrIGQJUh@O6N5e1?
zL<s9EN*h(HClJ;o$u5S~)e(8UL%4%F!YUy;;>b0EBF3IvU05~XMU>NTf@DvN1YjS>
zP%Lu<q=)*D70qI%cj=6-1M&$9lQ2U7^Yg2;=&%J`(_!9E5Y4d<N_H+nNq3{9SzITe
z7X5u+>OD@C%les>J&@Y#wJh|`W#q$1<mYtlfO-`%OWB2;TALG3qi7Qdhp8|*pu{JG
zO+V_;Y(o>jqZiNhsfy!sKSN|-xhn(q$7M~=r=7Vo%_d9yloiKLz^4EX2I}y}sVKM?
zdi%=~8p5D_DM<R*0DQ->sn?p{gKAP?(?aMN36Rm`2Ab1COl2pKdC%QKq{G+Y%pmq$
zvfK3Nx~63Dx1gL<4q;FZIi>^qg7g1~tpC<Q2ta*~hk{|xj%X+E<lm^WW5a&vcsF6F
zpAv;X53;UU1}z4HHfa9IW*hGlhe;^{R^`mKgWE|bOzl2D_itUbq$>n_2TM)H%R-f8
zknJ#uUPLuf)umIbWW>H63sq#}fv@rNvZ8+aLl>Z3^=`J%y_@c|iJh#Z>O2(6#oJ;U
zNNEc1%g4apZESoyCMqh;T_YQ;E0&#|#QFB^McD{l^$_gBuPXrL0Cc4sS76}S*Xwa`
zEzwuLA5C8^%DT-~fs6Z8KtFR#At0z{W`-RZnUy=1!ZBRY*8p(wfmYnpdlQJKJ*USP
z=Vi@`g;a7_q=M8TJz^n&xkL$Kul38hoom}9$_m5!qDMCZcbIWl1aX0oie}x%6e%+K
zK{ZIHag*`@W!4cT#82SpE3^IqWC*+E6Gn!BfIbN^<PBs2eH9$jie?I&Ml%kqxH#g9
zEGCT0MQj2)unxc#VDAVZ2z(Ns4=9og8loUftQrL0EXPXx4Q!@o75oisW+I)x2-rME
zJ?{tz0?pCU68L30N5gN`O9h6?70BE=%;9$9;PW}2E`9Z+tv!7MABnI&dDIU`u|>#a
zMB19`1F5+^i@+ATj${dfLjhH@Z8J%+GZI;g{K-O(uIl~B)Fkx@G@}>UI^(6hwukM?
zOuw{ilTt%b{tMZp#0~zKp)o8GOVuA3Uxi+P?p<~Qgt(Ov_)Am7$NxWs$6s2}$kk4m
zg^`q*3Cwd)*>hX;sL*p@1be7Ye}gGaG#%C*UN2Iw1E9yM-NNbkyq^#K_R)9rbbyNN
zsS<FWJ3wMl<yf)*L<FM`*}#PNITWErdH}-fTI)#j@Oa3Cpt5(Mm@W6h1cJRHFi#P<
zHEK-+)@>XI^Z^16<z!x$ZB5=hfZtA0McKZ?B(^;8Y9vMr0Ft^*8)HL?@U_4^TVfRY
z3~GEU4h4}1fzH3v^Cb%*{~0=k9^ZoXg+?WkK#1^ij$AHz6v$qPH{BWd9NCI&vDI7*
zp;~k+-0@qb`-3CX-+bi<M1J&PBFmd&K;2t6ZVa7(ICdI4J--j*I323Yu6FxgBM++G
zCTX;(#z{Zbrt$R+pxZD(>{|FZ@G!o={*j%krZ^9JLt!U_)}L`ni7L}8eV!qvSMwx}
zW+;k4rNh{fw@|0H=T^UONj+-Q%6a<QKnT?k%0dsT{;1vH8ghpDj)99Lc%$`kAeS+j
z_}Y{W2(im|pjWG!4QkAQAjbC$-u_Lz>ez|EH8r6N0k>;$C`gwmc0Yi%#gm~yd3UgE
zFZe-rUe)ouHNQb|AgxU^V;PfV?o0r&8Zj#=BC~r|+JrS%7h!fWkB1}E2}nsTy`c+m
zI9*xwd0If2W`l!^3a+zb9*8dVX=r>4@hWyyG!xufT)r)ZH0ubUp`U^Kql|AI;DpHK
z6>zXVcHK`9A`&S@2Me>}KnXM|UJHRh`B3*T>SxY`Gq0TO*<ypX0kO%f)`e!>Yw?7W
zq0n8Te_S7yzG{GVzs?tUrmntlNG<{dRfOmgt6(`cc)~pEA=*I#HH0n-`%JwzX1#j@
zr@~JkM<ToK!HXc^98^*sK}{t{SS}^y%>c3|WLu&yu`4{Z$UY@OB;k|8?uD*twX#ho
z9R{#(w|Ou@4O9z2Iel@^hio@s3=%rIM&NHbhn(3E=p`!M!|niQ)tby11f535umkAH
z9ywQM1yB$_S_XlfX+@SN@M;pvS<ikdV&jdbV4o$vnaYgMs0HP~A|GAS`^`MC3`1U#
zZ+<fGLzf@16^vRlG|n?wKmQ7&rOACU`YO6j(f(`nv{K8}eCB(^iK%GKj3mdtnVop1
zD0KXK=qG-Km<}~X0!()v1R1<QsDLdp33x3oQ3>!Kf)~`gDH`h9>)0{Ui2_327DHEo
z3ss#bvk;|Af1+3Z7l&N>&>zVCETB_gZ~>%T??|)sh-QkjDgMytQd=~0pQpy!Ng*Mo
zlTMMwRj~E)aDl4<?4L=zYWph?((*N(pX~~i4Gt>XN(ThNF@6-ABW`dp$%52$k21Iq
zzZCkg19&}}rx9$<DFK|G9y#gN{b-W#i+JGwWADuaq3YZJ@sXlzlkCJON~KaEWg9I*
zDwR-KCP}s=Nyu<4S+W+Pw2YEc(x@zDnaNIN%Nj8ZBI}H03^Q~3y}36_&%HeNvwXh4
z&*zVtImekZbKc8!y{_$b{SqbtviN5o&4N%MJ7za9Av1{6zKOM|?swqb5M>FRVe+H1
z>1<WMEVh2}Swu5N8ndD&KSlBXT_wVApim2qI)X4$lIm8ZAzaxoak%Dmrn^<{K6kw+
z*WjV0Wv!xPmX-3O3b52cY7iWN;epyWH3Cs}n6uwnXi^Sfy|wg0AnWmgXo_$@kOn~9
zfxH2r1|FDzya8@kAX<;H(NqS3-d&pyyu3|^z~y`J1@Wx?^hjz|5fOzxjT5Bfi&b3%
z@3>l#_4Q>BYVMQ?Pc#bD`Y0Q<CubY~-A++5ansLX7o;Fm?&Gcom#Rgw<lf?g`iJJ9
z>7B_~9vBXVMr=@Q#QvTl+>}{8HQehS)d@b(*{;IE(xbm4^y3Lzf7UHT2h8MJ43NOA
zkmXSC^2Wl936VhuP)+uM27wj28_8NEPQUz<+FFojAwT$|e2kwiu4WO5T|LwMU>4aB
z994=DX>3MiO)&5F1f%;Kfm`I$@qY3YU@4dfRB}t^qnTdz8sdDQGG#a)%>czUAE?L|
zq|ApiKIjJ;dtmes)Sm*D>At7+6|;}3XFQj`0hWsyzX6$Q4&kI<Got}#q><ZEu-t#;
z{+WrQO$Pv{|IPNMfU5f63Kte^8pMmXQCJ&z6j}|wPG>Ctdhna3V1G5c7<oSDVtX!#
zZJ}zQ#;su4i4Qx>E>*T=zx9jvGb<X_-Uasph13teWmcu~%w$dblh7(q`(xC|GP>bd
z1Ms2Ct1^@D^?3UFAf}(UAWNCZ;;JOB|7jr(IyDc&+D^V`+A2`(yV0arnU*yO(BbN=
zcR0rKbUaH~o*e&ESP`TSF9BPIco3{5Pw-!J45j<W(W8G=p67z7kV#ZyDi$dCsD4Th
zTyhWvmjCJmzG&JL0BS?%z-?^liDPl)a^x`<obutCill7<*WIVcv9Q1~@=dmWCKv#6
znLvVo08BU!jYZibsH6pe1;1+ntU<^xKq^=xjQvl(`mdnB(kMvnyVuK3{suJ=>*`!(
ziT#*9dsT$|Fo`Y){56iIaB&NXCFz7=GwBs!3S3-fLmy5F4%|<_JccVFf7a;(Zr{8k
zrZ`GOA8E7<BI(1ReYIK&RJ96ZJ}A(uIV6PuSX(-j`W>Mj#2O=Z9y$2H{_x&WebJ@<
zvO|)bomyhE!3tQ9bJ9kBbNBsHgt25+YDm#<KxSbcy~b>6!kCh|Q06~g<i}3+hq;iz
zaVC=tHK_qnRAWRH^2z}*O-?Y^H|o#>l3;Q4KBI^DaE&2BfMMj=bb+nqJv9)(v%;hC
zT6FNCH3s=anpiDy?6E3>wjjSWW*jvLZX&1ANt~i*0%@n~(ET1E-$Fc#pWLAkgrP~*
zf?al{F}M+=p<-sIp(b#FI1GjvM+zp<w)8>3gXUi0sj?u91frNW3?8r!Yw0M~;2uz(
zKcPHgz2KY@+h|q<vFkE$pkD<su`w4DV35r_0d7?a04)fyM|FY`AwGo$ReuiWdV?=;
z1_OerBTk2cc21*Aj2&D;bn8U1g2k)gT70PdJV0}ajcPsZ*Y(O{D1(n@!!;Nh_a|t~
zx&pfg<KzvJZqF0bO;~<oIF^`;7HWCTbkb<Q+<*YtX_-4`dr%Qdj7~BVKNKmIVk7uE
z{pfxfH6g?9Fh$7<>6Lt{ja{u?rIca#fDW0+3Q8xk#2i60R5Ya?bam}v3F;Cl7y!tp
z^UeeDtT=S<Lu~+z7$jI3Q`IWS###gwn!q1$V@Fnledjd{GJx<WDNf<)oK(5Oqx7tf
zuWz!y-r=o1k(;tOE#UT%GuB1VQVX*0cCH-n$5ikF6D?l@VAn6>5JPMT8jAqkUMv4!
z0aE{#GI;hvqP^4Pc4RPdeAs3>y3#oM4O5L71hRf@K2-$H16d+~X$bcE3C#^nL;rGL
z_ZGO6+leLMh(VfsnU23kFQR6#B0!nb+Gc{(&(LH+Wa|Gv!7BVAx*U)ociv%WC=h7^
z12if=HtYwDoX7?0(vqbB1WjiGavv)V>gvw*0vR|TfS14uAK(AYpoJ4YWLioZXA-B@
zs!qP};hMmF4i;eS!Kx!{V7bfCBjB?k1hJx_u1**zyQZV)t~#*pF@@_1Z&!)>%g{6*
zKbus9oQX1+aGVH0R?X-_NLbyWqg9~fV)M6Vi>#+7TlQ{p6xHDo*`_0I{b1weWOmu;
zK8H#WVF8lNUSUcKcnjXCDZpE`P}%SS(if+)nyv~~2{YVV5Gr(p>o8*)VE%nBW0<E2
z*H|~v_MkBnKiU?_<*rg<u`&8+&A4!`Kubo(b4zB*&Lo5Q9WjBD$0#}m`w=X?-$Ge$
zZ%^yN#)JPW$YGeR3!OzoW8!>B_S4qLAUY5yg$Rqm>;@!3y)Q9>jm@MC5;KIf2pZT8
zsJ)KkLw>dy7Q~feK&(;_#Dl!=0Ot%1{Ffo$0Tj~%05Zc*s<6CUpx)vaSF%ydptAZz
zsBth04m*jMvRP10iZS<|%ToaSJtwE@(bqIimlwvP;aF`8{rRycA60{99DGQgEW!!^
z@^b@4d}4Us2?VItm>W|v`2e)(fa$b0Z8w;j1Q_O%UNU>B_eLN}j!|(e&?k?zJ}P%A
zkCR-*j4==c!{TrRTy$H$Dj9*#fTpVyTu!1`9(|~(L*rB<FM$20wOw?4HOQc;0@)Uy
z+!BzXNq{pI5PBb}i~5d8g`Ot&BJi1zW!2<LO&M^<P+-9Iez-!9CwyiFp}Q{7Bf<D>
z28L7OCIH#&r#*!%fWiIjp|ZwKkhs1W*m{X6qpHLM==Qq;U|S_p<%?7+##FuUStgl!
zj8^8$ToczhJy7cNY<U;x>3d)<*s=g4$(_}%05|~HL0tbKP*kK^v@OO*c)Q_AVv*O9
zuz=vzQyY1F(0N*t8at1WhSs0E!qW)mvtW?UJ<STB1@>q!vE(f>3@0|$szCFgk|H#@
z-AmqoM;O&Wuis_Yw#nmPr%yQV+au3;AX|~`MxTxn+pxv&@8?LmBe6LY4th$Y@Bx_=
zc9?UB7_BlPF{0EEu~*Dm+B1Z?0jS;<Rx+{o3ivgM0V2KFJKnnlCj?7_bk99_2O_`g
z<p!tc<B5)fse(s@u9_BgIIoaKD+)v@R)M89!|jsU4UDVBsPWIlnA65zh$9GpI7ek<
zA=Q}_&dd8@j)E}jgA<t4{hfx>>_^sQ(p$%I=$qC`qo3o;QUS9ZVP?YtK!yK86o(M{
zI-N;y9nczZPL~vZbDeW6$QJ&KXKfB`kNoViU_mzNz@yvvgEdRy%I`43C&if)ED%7j
zg~D)4VHtOqLJ}WScS;lIn$Xg!w<6zZon5>~b*6QA%h1Cx_6Sv1d>Hz#OV))m0rc+z
z1WN!`i=Y1rQhH!Q>C@WuQShWGWMEUw&w9#*7vx6uvNwTDYBccQN`kbR;w(67&x`|i
zL3tJMRn0G)vXHT8Whg`C`B+sDFQehWUTE-<hijk_jJFKbRQMFwQU_u>l?O=kp)Nvb
z95<i{_PRkkG?9U1seEZHkC}g}nzOPE+;~nYldHx2GYBCd>jWlX{fGx(ufj>w`Ar?~
z8IkZZ1`t!FVBf!lfDzw(iH*`d16b#{;RqWSraKqk1XKe;w21-W2U^?D(1w~6^nm#g
zr%dHth;|X!Ju`6Gd73vdnxBpbntc#x_I;)%g+K}y5+}E!LO{Kx)HDLH&iDmgXOJic
z(4>k61}*`HsZ0jGSezBd3c%`h7%YW9X5W9>x#@iEu9wlpD*UO%T`4?g_t-dI<tgGV
zS}%QtT?=)waW>x{9vHp>Eff(z9pVe}RBE7br*g(_Cc<l_T_Rc2Z{-pRLaXqOo$xr~
z(wNEJ1;Qg|P~MONx;~}rLo3`(OfnkAK0LV;LH@LB<AIuc&)GNc$`{ofV>|te<9qt`
zze*=qyRZ@Q>}={E=T;}wWgL_!Y(q6*4av*S1Rbm%_?q{|HqJ9R;pour3AJ7ChA|3I
z;RX03dQ7AqO=0VuE<@r4l_<ZGSj&IyN4E6PY-$I)vLX%ZZt3dDHldwE%77s3VBLax
z6`n!u?A4r33_}o?#A<`QB~LVn$q{k;Sy}W9YH3GMiPL9)GLb1Cnv;Sb@bpp{#MotR
zab##WjnH~tU^A<pNM+D({8a67fo=dF;4UH2c9=rFckJm`2ptq0^Oi_Bao_@gXEodF
zDdRt7>E?q*5L=L(P*V@^`9GRarm(ygp8Dz1h5kYO4>G`%W-{1+<RB#cfUj!Co5tDl
z9g*e6jGQJ}cqy+ty;+2)OT35-HwX+*JDRZ~J(48`3T1xEjQN3!@JA_NfXiZjR#)S#
z#wS={Nkmn&D=?7XplKGmD=LHjK$2|g9G-$PECESuBWAyM6ME!gn<kipZt!EaHdb6h
zm;e~?3S|94KxH$ybeQo3cRY|7XGoJffJYbXP{GJaWyD346tU|q6U^SrND}}V${ou4
z8#Fbbau<3i$mEm)+dC<$F&bnkn0gl^s0<pm`|R_mi|l@4)>y7QUOKQI{1T(~NX9kj
zTlQB_vs@3gKROu@_|)T+=E~2yn;u7LluT)vS_xH6y`Vp%0iFlL@Uu0cEm+l8VQKlR
z;X}BIKBCfe!mx}KP<0p=07nXs-(;H#p1zd#kdB=!-TBPNq9|{YoGft5ZkWJ*Cen)U
zbXFBP(Y5gEVh)M>&b`=7sFAGw_^v>NJN;pVOQSM900Y9v`ReGdjv$urFc`kWoCEiM
z)|xYJo>|^Yke2xy=i)*q&n&8y{|fDjGg?4%Jyw;j(vgQQA}${qa!k#TkIe`jpxnN8
z;}}u;A_v>oToqU%h-L|LDF{FX`5?7W$Btm+zawUE?|Hk{4~)+Nbe90-$+uK6=Ekm*
z`j{tZ%3&vAcc|22DP4s;%=JOG`Op*LtW7!Z16Aa+KVsisxyb8n_4AURkIWqw_)mY{
zS3dyeKv^wxpdy6k(7!Zl%%g`vcquN3>*8M!&yyxqaS~oiB$kS)Nm)9=WPw{0zh(i1
z*OHVQGyp4Ch{_EYOrWQP+<84G=AQvNdp<r->3+Y!cUs5RNfd^<p1D`Sg(oAYlcBvV
z{HZqZ{wHA|+gLey1{3tuAxxTy)3r)FQ;~4)IFcoIh5n2H+&g~L&Zr7oXuJnhN2yhF
z76ZLGjnn(|X8fgKAGe^>dYBO?esDMK2H<eY!yd$(g9~vZJFq%n+lqj2TTSd38&!Em
z+m|u`QY{x_J%9#GBPY#fl=1-BTCz5@0C4qrbhq;_CCx1;PN=H_RtGq!6bOW|mk#7B
z46F8G=51mrZi)_6Bpka5fi;-X_pc$!p{mJM7INSwz<Ky>qY%9uv*4s4+XV0Q&-j1K
zg~~o+I70@q1}ebOHa3H&Gpg`*n=bhVEQb35)<6%O$N<7IC7wC-ldTF=QhNe!#CZb9
zQQlfGc1Bn|j3d}nyU<;oK;R5WJel#eM*W|!6|Bbph_Cev#0LzZB5xq~nmqNV_eA}4
zRA#<Y^`~n7W?<Bo`vR0}4K%oO0bw&f-D=uj7Y)DXggXNeSo4`hpNn~kVaAgJ(ft!g
z!2Ac{^P5>@7J#>14NPzcOg7l>8VE1tnTvON*Fx#9;mYgodYjZ0obXw}r*quwp4jr7
z{rr$!;>hcqkF*l`-exAy%YJbzQXp!4_!(Fvm+ROJVJTA}Cu&q)0_%1XO$W^bM_P5@
zYk83{XE|{Dri{=+5UkxmZO8VZ#|{pF`gH@~`hBb_cTJ+if{PLtCiVGTi+RleK8&k4
zR>p<vL)upMRc@zN@-&W$VLsUu3{S59j^F{y^7tq8Sd=Pa;Fp-*FR%Y5_PT$96o?&J
z;w?^~;tC_h=xX}0qD~pDQX}tMpPat2?ELC82M5?zUG$I^h^U`b`HtZ0qyx~p3j-#;
z{St{8iAe(${J;jzSMk==<<@!X?&Z^`J<tQ}y*@PE4OQwi#WkJp#SR>OT~}Ch8o2mq
zE~Z;D#IeUHP<~{nCZgVzS8QBCLi^F0P;FP4-H%H=&O}#OEtS9Cd0yz1%UzQy3}px%
zK^hYR>x=UsfX(9!HV~*n1Eu>0R)eZd3W;<{@DpWf{=v^IP!$>Qpzb-uip|xb*n92K
zQ5{Ob7d|as-Ky(&mA{m0c~w^+@>V6{sHzKQ`+~bR4lDsZ&%YxU-du;N7TKA_T7m`U
znQtvLqT&7Sh!$#@1LkE_@uodA6Cl;|%M`|ZQ@`C`9d|5tU-tErx1`ozUlLvZ1+$HB
zyKmSiW;-<R85#!w-Yv^Rv9Kr>$IEr>RG=9oOQ|ie9|0KA@4gWyr6L!OuMg<8yM#&_
zu%j9j;a(3HA_IlR1H-bj{J++B<$7#>bX*i;cKhMT-QD_@1{E=8Qn&H<p^h(*B(Yb%
zgT&;!11#Zw+9J>v00Il&nRyw&8+;H%QFVfK66b$C@zlBm<W+(yT#6Ialyz6>IP05W
zFRt#U$y2?$R^ecAU2wSS##6UK3W}q<)vrHkCY)PvPka*>=(~fKegwAXc_ucL!(6}c
z+x^>lccH-yorU9!$%HooznT*~3atXYys)bn2jNmkP*Z{aq)@ynAcI#*R0-m1vNPmP
zc`9S2@%nw`v5lL#EK8P2-*?B~A^xog^(ES4ni|Qt5=>Y&J~DTGvTgOZthAH+Lfu?t
zUD$c#C9I7_M9X$eW}8%@{^T-kw4rp;*yyz#m|$-a>b|6>y$W`TeX%Cd8%?5>0@XY|
z=`7_z2(LQCCUbv$>dvpE5!IYuOG|;kbs>L86cNL`*HasOlY6g!n<_4g7krx&yVEbY
z-rM!KDEp13c<S)NHF9$nv*Hit8E--tWAI=xbQh{y0byrj#wNal=zV$m*BaV}Om05(
z<14lm6sWo=KlT_blQ(o6gB@!_4JslE({uLMjS0SL824oK3Y(T2?-HsgeDC|t$EW4^
znMD^o<2(NB1mN9Rw~42TK;$nasJJ_kQ6b8b!j6i144b=|Sr_%Tjt;K%y_~Y!hxv)8
zJ4kHGr1zj8+oa5MC|}KcC#G+rvAU0d;{Rf>F+`kp>T=EsML4oCJ^=?3pd&9vt*&Eg
zc3rWkglKI<jKi;7k_pISpo<oF&?%2n?C(o0zOXBAW-okm`i!P&(eF_V$Zx3l&~aWV
z2V|PLsbhL}l%88fmeYGmLg~s;x=CB}oaz`35>4<sVibf*nZPf6gcX2hGIxDPK<KIZ
zYuX5aSu6n(MlXj*Rd5EX2^e5iMW~5U078$09+rRMv-T&x9WA!<To%9DdMO+7`W8=W
z@x|F4qHGax0S@*CUP&U3Wsgn+00|e29tP6+<jZvlyu@i!vWJXhOnyf+JOc1`2V*a4
zf;9bC+j)OYWN<-~bblCz##2^=AX?_$x91z}AKmuF!>{F}-TGTsbFl3rlN2}`az|$K
z=JTLu|Ayy`9F^D364^T~?>_jlA!(*+=WjQuVhiYeLs<XCoxg*w;_>)|d*tAads9!g
zc;W2Yw(wVyFLy?soE&8+z+B5fPL&8S(uQNFb=8y~=&BsYkb$`*x}88jLjQaM=0zhA
z%8%&D(5o3NI=0na0iFc15uedDE0>oY97KTJNN+XlCHndf177zy!)jcKxJ8mi{lLqQ
zl+Wo0WOc84o|a8BHvfF2(gPm~;$eSB2>TN@!KD~MtTV-}L(W$TD7J$-`NGXNRVtM0
zLIsq+Tu=AbUwS-kkz&p&=hBmtg+G+!K%^{?73d%REV@D`X~2PMNZxmi^1-0mY0Ui*
zB`#p#Qs`@E*VXc0Oq2O)gaWcwX)Md9;ejJ?!7W^l&uVt+{TH-1l%X@=N!`<_QWT07
zb~iX&)A7oG|La%QIoVM=)tgj?q&IHf&Mf3a2Vfh2pgX2<N+;;%HBVLB_hKjp$OzT*
zSx@>@#GeFxG|pQ7zQgIPoc#_fUz&5}67P%}Effxw-Z*fAiXl&w4D&PhPzZ?+3m@AP
ztwv(|hIGG5X+PSsz5BQsGF~zA7Bw=-9l*x?9h&NDOAZ<qel*ZhI4cK=3;V~XGL$fr
zT&94Pg9Q{ih(5IgAtRNjb_NKgC)I70Xj)rNWB@5mr3B(&9-u^qj+-0o^4x?pNh91l
zog4FLtErjni{+w-O&LHZ$i2ZgL<&M>dNXvEie}viZ(F7#D?oD#b2mMMOqA%o5*s9i
zOT|>UEUNO@u%EwDYg#Fct<BB))ALWme{P}4(6cD`PIMp^)EMgQjXU^hZOm6whqq^L
z1^MZg)Y~4DI__#&>>UaaPmCsTocnoW9jK^q<{q*Huc4Nyx#5wV>|j05ZEqqxZr5!Q
z?qc`bw#{SM3_pJ&W-wDEa6!$(cw;sLDGYJuF}eyh?f|tX1V836#?bdnzPEAcQS9m3
zy=8BV1K7mWM?Vq9Qi%YSmYqZ;LcME!&|)}vSvmFdy<7GKsjg^{+cu5NR2Dw_bnl@b
z>A^(kEU~GR99duv6*uFq0oi{%ql%|beD9)$K0Q%0bEN|Tg}ZNQE$<4uQOZFW1&ME{
z1*^UHN08@-V2*H;zKS8mJWThXOpvnzGFtcOM2(hYs~&H<wfT{IgVBnv_a}um5O;KO
z3>SNS{vqtGy9QKPD}8cxT;x#-V;z9uUC=jWZ-U9(p0-5`apRO?2QXBesrrSe*TY4?
zZ6zM~ATbl)x$M%D*X~1)Yz0qzS$1aBtU&ByVs}BNL6X<EAO3)Ugu8#R?YvScrtNM(
z=J*0ZsSI(@)s9tLQ3Hl@ojPOfEF89;rBRCB7eC<q*edp}CCm-~3@{qT0q5#mxtsV8
z#5VMU^Clfb_o&7gwp~utD~tngsp74=6`SwxZxMB7q;0!4%EvUy{<V+PF$6S6LmU3n
z8XDE~d!IFlVq(%*kM57^Mz)_%>0ro<rtWX!HQYF;ImE>;!OZc#+XyPKu|yVl^d(CM
z$qGvV9mJ+-%SS0&hbg#qYY*K&Kjr?N7yupeyj16gFcI{W9}Gk0@L-#^Prb!@h2rO{
zyB~Z<7~60&3i5We-gW-;6nK$<C+Y9TI5}co+}`xX!d`y`PbXg!I_AV|uu$~SM+0f`
z%JYN!5!t=XE^+7h?xn=dzTnF&mSjOTYS19WSclc24|WbBi`saIpDfdB({0bmio|W(
z`}O$O_}#~3k1l$1{ZyCBefh2*pj}`Q{rP>_PHKKa+euHP;>d%I5m>ooTgtc56V2~Z
z0LZ^M6tF$h$kI7%>?^_fPjBn&>9|=wT64R*TS=P~bqVchf^z2V<Yx2zGI_V}K2nrV
zJM0L!;(TvZc&8j|0ALa96?aG#!R(r*b~>#PiW4L#x#LO+l69AkUmCQqwGhk(5Z^~v
zCT!R!z^B`QDkWdMP)(bff$(f<X&tocgwl^~R*M@HPZ}S+wA~=&{tD{-*H3#5EFaom
z&We7n5|#US#~HqS!FLInoQNCa_*%eSJ<3TY4Tw<VO1>imbIm(9p$w~|O3mt>oKBs<
zuQ{>UL_jG-axV}2sS!5y3B1Oi9-h2fnmsFoun{XuZ|T4XTgcU2I&rB&wpA>)Vd%tK
zP1onuFY0X%qi&a6EDU4Gk8J}cM?9cv5UA?(#*;Nnj56xgHCN^3+`q)@x#iR=-Q#Dt
zT`H044WNgH3PEGPv+$FS37ND!fI<fIya|m*Kmrge4hD)BD63H3*?`}bgEo!h9ucjf
zrcrWGKJJ=|#U|?62Z<Hd+hg3$q}aL2MtQs*Xx6no>^OPo5X8k8JqCt2=BNIWKa}VT
z#VqT`i?P2K_j#1Iot{L=#qN??U8iIRb;waNA0(b+#rNs7xVXNyjJ_m)RpO9txl|xw
z;;$^|A9xK2QFIN)O6F0DN`XPB_p%0sv1<xU2Wu;*Gu7t0RGAj_k6+Jjd3t!S|6-0|
zjN6?4aL5vaKL-(iq)s0(*G%iT8uVF%mCRMSbxM~TnA*=ySeUAs>pC{5xdq(P{kl9q
zEeNrGu)?e5Z|j%k#FzNut&C0R{MX60q?S}NZV;z9c+~680O`zfD=0e0d#Gh=1S@42
zEbbqS637ERNQ(4TuYKx{m75^6Xrj@tP5Qg%6t^&S7q0W#$j>bua>1|=R;Y%6a*rK)
zfACNNdFX_A^Jz-VePh?cZL8!saj|);7uZlwe#bV$N21Red`>JSO6mA>`VFVQL8q<G
z(~r#jqJBb!ciF)^GSTS1N|%WGaX@hJ7d%uyKwDF+b`poZ#Y(zcJBh1LMHn<Pt4=rN
zUiM!1fa9k0WfC`K7afb)A<Gj{*Y_*gIp?1@cBQL(&?CSso&W~t^ONh?oqjVblV>Xd
zpe440bdOnMa^2SD1}=S96$4_~7eDk-&6C%QZ#f)zHeQ&^fNN3%xhv%53%jURlE>4H
zy4!t&&Y|9?{ixHR*77;VH)xtX6Sltc$6Y~wK*E}w6e!kshWb2&@4`ENTuIVHA7D@0
z*a6GGUxY;=cIeP9;{bcdONhh`3tbFBJYn4fPgM0BGf@Bn9H`ID5Ob>dI|8dfH+sKr
zk!Wgr!2UhrC8>4m5UBSL;L2KHoHr;iI-r6VN3?)H;dMITcN!0c=@?2PV>;U$cbqt#
zfy4#6(i_gc9I&pVVXxDGbzOwHp<9s6V9wR(S@%h~d3XtW*467~-v)fE5TW?2)yM(<
z$x}V&Yr>|;x92>8+24C$+^pddgl3JK$+`qCXx8W)ng6Y;V%xZV7E_=5T>8RR+H-i$
zp#R~A|IQrE`N_PR`MTDb!Te>bei~+@fpbYNva!@0sYeO2@J-UJas7L<@K@$ODWsuc
z(aEJpx4Ex5hU_-1(Gn6bC~nGW+4r$j_U7U!h|A!=^0V$|8rwd<w%8s&;d7X+H1))P
zYvs(Hd_xI$@m1{m9&csJ^5%M8dYb$P>u&C+=Wp*o2?tN?B>zUw?7_o|kx2(ZHG+|c
z<0hL^qXki0L_e&PLxmszd(Vw}{q8hlXE+u=bjor3+o%5_GUERGO09>wGN=5xV>sEX
zoe#{DA%A7{{_~P+eYusjz%`kyaM4Sv5He#a#||$6V=w0j#$M&XD?ev?-E4_6EpO-S
z<-a6;et#LvmGYWu@;l7*c=!|f0*<jy`WoB*ttB@1-t#rcpMIiaH7)sEqU7)+63K@V
zXrI5i?*BzU%CI8lcG%uiVF^3A(@&wyPCu{tFDbp51xqc<n9;OPA7^E~d3S$Jw9~=&
zd}!79KhomAHwR0}fubU3*1g1=twY#Qo=#_ar>=>)TX`mSX=~Nd_sh`NgwFo0`&(!E
zT@ZF}&{E|<bdDxNe?igz>ozF|?up-0pcUb1i~|tD2Cy;ZZ?DM;ZyH~-OO9HsG6#(O
z`=HC*Cwh81Bf~dz7uP<6#m{`4>Bq(nFVRVRc_TD0_gqE!8a7$@7;YBe_|J(4jdRay
zPOMm3j|pVc;_NT}pQ9IfEOvz-z<hbvGWmDa;=f8fw7su9NI1ivnV)XG>z!nf<c5mB
zgKXz6kH3bejn(AW-PIGywJm(W#fFMl|LJq{p_Z?{vgdDGFNQNt|D7$&@$A{Uy=+GN
z5GxT&RDbRp_WcJ$&ksXJao0dSDIyExpWX!3*it?k;vtiKA&9Aw|1B>gr~u3;n8!>M
zL*vSHrm}?ubq~DfGHDzTP;O5^UVKQ{1M(qq|F^v4g75>L8JY^{9Z8d$t}Fo~g4yXl
z3h&DqW>|x&zgQp5@c%7i@$<~TQU|_BX<ijL4|uw(mpJ^CN&dZYVRk=$Cmq=8(%jJ0
zqydYqzb2*9TMjtd7m;8B@Zf7*K~UnS_4U8(-{YJ|wx4wK`PpQneGq+m37khP^hHYX
zeu@s<9dxYM1HELm#m7yKWNoZGw1`KOTcD(kP~U&-SBBeP$lLzJD_cK_i})2_)^lju
zmGQoLyGE#kD2VP-0XYd4!1cVxeqsZK6oML&$OLi)xVax>AUr^A)!a^uf9u%fRx9%y
zUD482Iea4W61T%08{6)XgGcM7RQ3cK8%nS86_p<0MO<+e>PA!Yrb&!z_JDjZ-sd1Z
zJ$PkpWU4K^s`gaFCGK8j4b5KRkgp9!@nQ(s6;Z1;yj!lrnToLc6GJk87}<t#yik|9
z)6@`1QFN%5uY$$F+-H-<TR|RsV*;pDpR<n$;rP3`9w5=oe1ZicQZ@QoD+B)4n9Ltv
z6B+U_iZ1a^_y)<nvq)HcvA_x$hyGiG%!1tO_pKHcUD<pP;aX=R*?|nUV5j>!ZtC&P
z-D8O_=2-l=<f~fb#w@K4(VKEBIr3lpM6CN)m#t<JWrFFWM4F-nfweD$KOrRroZTEC
zT5SO`DSH=X^v=-3eE$I<-@j)csb`TzN^pTVt_%&N9u<DcigqUJ9T1jC9g>YWA#_7`
zo3_NmQ)O>Azu)3gy+UI@L4}nKi{P%I5Xzn(*5y_h7KAB#p60B0h0reOdBL-BFk&FZ
z6aoPN+k@yMtecL69sS#`J^6I$8gE~;{~Fe<Ew4~}l%8#m)DdJG&=jW^QhG|Hb~Jjg
zbayajxCN^#A!%XUPCIm$o(ecnAsTSUxupK0yE^uX)NKsMbhL-b4J9!**1_xPk=q<d
z{I1Jf#gd$ZZ24QQHo;@z8~y8c;B!GesVcRnI1+)~mR<ck*XbE_wt}qx!1@XPhEkhr
z{)R<o1uxu9>!$4NZ^#oSZbr915pFbz7b*1AUt#>!>uC3C>Gq&AVVjqJt3Y@;g6M5?
zW4l}kVO1%mu&_;*aXCvV)k!4ANzL@bHk*sG9ox4riq276+FO!s5JF7H<P-fyifyRK
zBIn~6Gr8N6kL=bIe^5BP6*w9V&Uz(``sY6<@iKYn@)WDx=~R&*ro5$-SBL!iM+bGv
z3^Fsz#H>B!N@zCbpUdBszgt9<ht6Um(17}NIX#8kGl<+;m+SVT4=Kx`FOn0|w&+1<
z(e_t~id%QJ9ueHkwiCfRK{sL?zzWf;09CaQR!6e!iMU6&jnYL^Ge^n&kKPkbp^6Tt
zNm9+@A_*IrXu755HOJ0-hK%h7{N~*9Y+VOc?p@gLb30(A!p5Cg<Q1kL>nXewt|V-B
zLV`e{klHugawa;tP_;@ZEY{Cf#09k_=GbK&u3hK%#~NOD2CV5`t@>1nj^TXI4~tgE
z6^l5R^P62y^<2Ehxzc?zC<my&zlH@Uvv6GWKn$<F)JB0XEoC(ZYf@8-i}+RVF1L=@
zmg?Wi%{66MXdy?L2uoKpSV1>A9AxxqI1Z~KJS3$emY*+~zj(RF;^q4sd?;D(hymV$
zbO>3fkLEjJ@bLK}u}`-A&0T|qdW!t9mt~`hx&%rPLR67~lM`qjSc7TCy6<Qc<;bmn
zdz^dm^{%3)8;a?15y$<dy4l7PkhSs5Z>%(?4jCVe6U^mgtfQlk#_bO+_GxoaEYNj1
z%^iMQ@|dim+NzFVE%!=DJB6A;yOvE4RT%8HPGl-)A4C_aX5Mjl{JBk6#VblJ+3cWd
zM9GeiS6yY2#flMLII=>M741qLlMU^k4)GR0!EDM|lll0?S1L%Zc~!m&jjs5xUc&i0
zqEnJIkkMWhLPBZ^oS@5-yFZG=$3`m_+sF!t0bP6f={3pE$A)&F(CS$#Xp=VSUj#}q
zY*Q~eL<)e5<$_JgvRiK_)@^=a^5M<fU2hf{8QT`7y|YpjsysAsIuA!%Pwyln!!cYv
zPGVWN$p*aUnK+NFT#wtAAXh$rhA@_nImdTG-|ZcO!8f~R@K3%+TF{NYBeohaEP+?O
z5pQyRy43jA6IL1^cI<BiTa%X#@WQGbr>P0ll!VD!lRzeLGGA}aVtGcgc$3*HY#+|;
zvGq$%_wT)qw4lS$yb7?KIa4C50ZmSO8Cn%vJUAqyX;B<4FJH16x>0CpcGsr5vqo8C
zf9by3^W))1tutS}4Akq1Aq+V0O}ZDEc32Nz5!rI6q_ab$L?IBJ!51m9G@X%NtL(J#
zB8B;pQ}V!drEk%_5AzIYgY<Z;1}NC%ALiAk-7_EyR)32<E%5eH)*YncTFsH{(`bh6
zI`%$;K#~BHpXyQwQR1+gbrOc34pg0&bMOGmQ%Yrt>DBOHG4>0hmHRI8sr8M3^d2H#
zs<!}k1Es34|5Q%8r;^_1D;gi*k&e>W`#%=;TRhNO;@)@-ZHxAALsIvpwvz<t5gKa7
z)dNsEes!!^fwukS4rYy5__fUsjvUWffjH`px5k%YLePBaPzXdh;)z~Lj}mZ6)U$GQ
z@mzX3Md@<pS?h3Wy||H9;hPGCcK~)Z>me+Fz4)f?9K6#073PcGE(`L`O^<z^pWmP8
zzk1!O;2k}h2_|BlxE2+f3*Ga@!HztaVh7Dc2e~-)%f_ST8xMX_x*?+z)B9dDO`Si!
z9c{1Xv<$8!z8(&9C0%V^6B*!Yusrmja#HQ!r>3+UAFW@C9?7$`yPVdwwY7m@!@8@f
z;a!5^ghjud-27O5n7`FjHa_oTwpn!fnvee5w7&}OjM%n+9cnX&hinwvlryE1oK;vA
z8ZYne-_h`vBg0za-1XA8ve$Ap-q01+E$?GrSJrjL$Uu;3MR#n+iP}lAQ;pN(FK4|d
zmZe=U7tE1b=Pz>k_TggD!_W2HES7j1x+{`N1?CF0S_*;4(D>Nky^{Lu@xA9KyzUJq
zFb)>J@?UvCaGgT;j^-_*DrZ;c$<rDsO6V1UZUcQ)b8*fMaz?|ck<H?or~#n<&fc*b
zIAH0c#5T_E-3q-%H14~qj)^k0Y1U4v`n<<RbQGIw#d_{IoFg<IflEH#*RXx+ew?#p
z>%A`y8x{BFcSfm*8SgXmd$iJThw@kPiFZ{27SeP`p->*j`*}}7I>*w7A&U(;KLkXr
zEiu-XUOrj5PfxwE3`IEv2{@oBLGHC37BAJzBH8Ur<S#q5hjf*RD@)$g=YAvg!j^sP
zEMad0(K7OLN|AU4OngQ^@8RSsf9<@DbFz7|-Pwynm!Bu_yi{lN#h5p8Q`JgOLWK%J
z1ZgUqoYP+Yy!-9vyB-lKj5qlMXD&X|04oR0yu!K<^R;15Q-%k+DV$mkkU*<$bl+51
zoSVwlCR3rdV~J%SZlH!yGCP`{x$P_+kSm1PY;oL7?uT^t!aJ3?ho-?YcBYpiE}m$O
zOnJM}*=N~1#3_>^kU%($=o7J!q#_4-JB3v&Waz#gYO<A@))^7Tqg*}L*W6W+-2U)1
zaqkRln$u717Y@bWd1WVQ&OxIlIHoelHgzTT=YZmKyjH>@!bNHiGu(|t`vB+~HhD?R
zMXNqL|0?{OzFs{m9}@Q#$L^v6?vTi3{V0b<^{q16$<29r#X0ZFS6w4E?(==;Dt&XK
zs*9gu1HH6FN`k@}=!z7j6UHwMqVyvx-rkG(vbt;C<YC3FvObIa^mZ&Fs%ow)ovDgp
zxVY<+FJXFNDf%5G?R-LE_pS>q(ic4usDaU~8izXj9G<cAGFtVE?b^|wOAA+68B&Ly
za({8`v3|5!NI!B&OyXk|YpW951jb+8s)H0PC0fh0?of-({iM8iTg!nOeYP9yLap5&
zliOGDZA+8^3EeO5zs%j4@a?`31-{)8F0Qb)P%ks^Q!6@0a4A1$J0dk6A^-KDA+3mY
zFCE8c?*6c|*fMpK!kOBTO{rz~hRBD9S2`>09}wK~rTv^3B2yRtxS#r(CQRSj9vR%G
zMvo@rP<G25(rn`j+TJ%)g*~lfzv`W|<JrvhIv`@pGEP7D%B2XF5#58TmX8r+u0lrS
z3MG5J*1lEt&@H^Zxn`xVn6;6NUEdYYMt=ul7J8t#gJqqrj}r;Kz1n5vtLA32FIoZa
zaupHVUTAb~mM{E{aPm6)p?!E@GzqIncXj0ReM}EwrCPMrtEZc8ks@ExB3@E|p7XL^
zUt!O2&|}aR6bu-2l===Zkx50HH+A}2Wk-2Qc1MjI;2k+sA-36U)c;*C0`%`@iBN1J
z)YpI(@HBWm<xmR8%OrEJ@^;5UnP+J`&RDh{=a4M+c$~!0XY$qH)sH;sgV%?=hd8oz
zo^7oDA{Q+iX2icNQb2L}g_ln10%0aSxPg0AiFfgsL02+*DOT+5b*v`!Udk8KW_^j0
zRXJI~2X=3~wL$P&jORINm-N%50^C(^3Hm*TEAvPl)7&;b;!cO>+w-3qj7U6LM^zuK
z)mGn@(3Y>8FKrX%$3`RiK<A*(J!FI6hZKB}`eA{iZ=ua?YH!wNi0~Xjgm-Hn{pj5n
zB|65gJjTMRFfKAzQlM+_l75TLS+%Fq8*}BvEl#xNi<fRFl}L_S{hn=|m8@^|mc^Iw
zPORHlm#0`=I&V8sw11n}IQIvB$)eN``pPL8Ic04_HL;pseKZ<~vII_dR3SROaZrKk
zP(+fDQT+%D<}#%2J^GTf^r4S|y8NxWuwdggM>W5>ZG88VgLpD2c88V6<5M?ExD^F_
z5M5DTOT2e{vp`YZ2Cb=87WAhK^#HofW4jd@NmapCv953Gx=!ufvaev<#{|XAobQj7
zeJtWa^g_17i`+3ZOL`|I3+$)G58whqBR;-iLO=8DR<nXvyTgQ69}?ESo~Y2;nsJk3
zabtggw=zAkz*~hCtEuGf*@+tD(J3}x(RU^ztHDC=<CIx{tg)$^m8i9O<+V}d&70QF
z@vQKS>?}e?Agxk6p8M(v8P}aAI%3A0Z=|~pFCP0jZQ>UW-#_!}rQb0;iLfJ%4<cb7
zd5eGGnW_0td1h9<$?_X&Gu2OSIDXbxXP=m{v8c3v9e+OlUF32*&;>W*H5u1rrE_Ah
z)M~2N)rO;YUxh9^mZmV6Zn|H$-NWg=11+voq{N8>%R|W{hB7aoH#zh+M_<{&Huvj_
z@|97WkMaBLJ!8`-l|j`RAjhH7Z3fqq4Vq*7^<zs5Rh8d-8e_ZQdnW4Oebe$A*Y`h&
zzPt?aUQi&dnzJ`6NQqamfPku1-<qi1)nvE5y4m!}#A)9}(oBTVYYQW)VW&+Y+Ansy
zcwAAQ**D>-yOBaU+#W+J*}NaOF22a`bLB1<)qw6~@YxU!G6v@8*+k&W7-OR?@wNp`
zPT+|G41<pd*Wf`>L*N)I2HorG#1ytiGb1fXN$3J7Qc9>s-BId}B{@%gpU2p@;pcnr
zNUkiY5;xx`wf)Pj%4&DXmmo#-nO(ya;kL)4*06j}p6j|AN3}PQ9^lK-pF>M6_*u6=
z`uB8F+tp0}{jrj}hLu{64z;-E6`fBJD~wrV<t(1-IC0VZcnSj9-|7r2(7oCXP%Sz(
zosr1n4wS;24<ZdUUp%a{Z*E(^?$Azoe^5+^Jq4P{OJ`365v1Uc&6~qJNy7aima?cb
ziYH3>_;hZIg>F71nsz!kq?lnM2eM^RZ|apW+}MpdnYxupmQQmYz6+C4EM9qLY3&-n
zA=J7`=VI>wcZCj<LQIgRgS$YX_~iiEK4HC0n-W9x1TV6eCAuI_AH;%;?j;f+U67Ij
z7vh-@fls-HcpmwAXyV;Ba;sI5UJ^;0@BF3PM_`grLF9d*(DP<Pi#bpVyLAUUAz(}d
zGIcspE1BA?C+ga#rmTtK=PfpEUeCtUP@xb%pjgk*bmRDiV$3H30C7v0qEz9$?}%ml
za@xJb8AdjvQePcJE%Z=3TUW`CW_o5Gyb^Xqd$Hc)r7ugx1B?zYdgki`f&`qqv{smW
zwHXWrE&zU}0==bOp*c&dc5S^<9%9Yomy&NL6R(Mt&~~Wpy>Wm#gkHBHq4Pr3w|m;^
z%guMKN&V=cq_|#3hb<;1kSk4nxz1v7#3ZtalqEMXnn>sCED=Xvdr<#^8SCJ<UD-7!
z){D5&Nn+jBFiK_9;cdG#kigDc%%t0q4LF$c_3BpSez){GtdqA>gP34qsI}SNfhX6`
zoV^SRCYKqu5u(!p`dZTDpih=iK-gZe=TEzIW23bB3)OAO<`0O??3&{PGwcQPeRF?w
z{9kczO08$?Wo|V9g*zPBBke|E8r`ia`quXwV|yz6@3-9B9nJdWFw#joQ<qWItWWMK
zXq79vpnIvxUtC_}UEKAF!-tMGg^iu#KIgX$8;7qZf<(NGtU~-}+87-w%52QK(p~IB
zc<r~;H{{f?tZnN0Qvy9<L+#g>N=wBQ!UaTbWjqISM;$CpFR5#Bk0T59+q*3-@6I+|
zQL<z`Q!O;t=Q1A~A_T$k4xuW9K)~NK?3W!L8%>qBP}Oco6fjuCt$!z+n&K$8MO%6$
zO;&uQ>c{c|IjINi;4~5Vb0df|6<+avj~i3Xg1MDZAXoOm*g<4RPcZ(Rro8Qx#?`6V
zr*?vwQ&~+l3Kk8v`MY|yE%Uo6(=B`If{s061ZuABxST@uY)1z(*Ca^CnP`X5Dz|Mr
z6JK9+|0H<&)g0P`v0`Mx1=bz39nN<JQ}Gja6?MNplH2`t0j>OrMd8zy=W7jZ2`9@W
zYnOaJ=42Rzyb*Q^_lav1Da;h4@9F%|`HWtgpM40AZE-XEa-yg<v&Ki<Jf{h<)z4WA
zwT8|~3yLOB3<{w9;AM0heFB9qe0j8C<Jy;(Vr0#`lyd^Z)?IGC5xBc3(0PQ5N@|HL
zF~K1rJKR-*%FA@mNhMP#JtJ2uE>`xf)rQS#w<ZvtuIeUzDuKD{J0dyk>>j%7=%v?O
zqL*7|z4qNi=DBk3rO!6c`;UogM7yd<4Fg~FunypL*`y`aN`_kEXjf`-m#Bm(MXo4*
z)|S3xFukn4xGOql@^sWTMU)&*MF9H_xiBgz4OytLg09x_;&mrp%rx(QYUhXHtl%?8
zBQlebH$SGnkF&Y)4*l-f1pe^@IxpoYsffLJi@S(+q>=IHl7c2JQN6pvh|TOh_nKOV
zEkKWJZ{W1f-;Xuf9!gnVS{VCrk9(+bR^VY|n_sKcMI?TBg({2Zg;i18Qi@HuRc27h
zNe&!-&7IrQri|k+t{!sV=x5}ZI<zx^rz%iNI5Ho_%T(xCn#d1XeT$8Ko?z7zvc^R2
zOPcIdw0Zr!-hI=>`hVgdCjmRhw85k44uB21A8K@i^(}n0mSDR(t0u*ViGy))$jPbg
za}6KF#Xo~X=3j$#W4K_qI0XZSIakjQOF!7R30OWq8N*;W=8x9#S8a(q#c6DWj|VJ1
z_Y90#k09GkIy6735|1<`@S0`PNzN|rNk`hDvMW|@%@UGYF)kIJCWAW}mVi8_NOL`1
z|IrB&#(i)w;9k1?d2!@+c!}+!HoZ2%{e0DrUh}?Tj~{pXjA~~BHA7fxOAxP@9#PI=
zHAg(kYxvrO%5ic}TqZUcBkP)ge7p{SE7zQH#=Fvc33YQ~2U3v!{Ih-YvB%uEyz{-*
zmWw2lR`LY(EfcDQ=I+x3WSacw9(Y;Bg!e|M$rVjY^%CodY}rF|q(%m(lepa7iXRUj
zs~?id$OvL+PNBEi9pA9I$Jgkhfw45-*Sv|s`&<+!e?lXu#*Of|e83_YZ0{M^{?gRr
zlgns%_lDu~2AiTQ5LkbP7gmlcQiS)vU{B!B^2{?WFiA-cUcGx%-|1!I&3%bcA7Y{<
z3zl!Yu#RrSIO+Y-VhiQq<F{nbR=L6}MThY%1Lxb1cSX3ef8@vF@^*XdzAfvD=t91g
zFUtV*I_Sk%!)M#G$jC*uS&bt*qx8C727QcaNzu*UbrEU?U*E}4W4bhY?_ioyH<kKh
z*FQPV+*Vzmm{A{J(LuQ1$@M^J;_BqcC!vv%Yw4O|0obSQRpHGF`3fy=<cJ4ks;uV9
zh8Nwhb8>W>wbP?N8vC7C<ac?6n1R^jD^-3>TKvB7NAA!ef1T=cr)tk_GC8xIR`YyI
z2_YnSQ8H~2QAv}{c1kB0eU21)QZvtrBgxWrWvFuAM}4Jj3eBf?T)JQs=*h1oj?2gz
zRBK1A#BRUSKAc~%3u{e=#MQSu7eA3dusS$9$J!^~js4V-sBQYE_`K`sLKGS%0JBtm
zJq7+2Gn5`GsP7jfPrJP=*6T_3WJHtZtyCl@Q;Kdxfv3z3@-<h~D(RQM68AJcPL6YM
zJ=}ht8!nUfG<kG*39pGksPU&^rKe5L-O|drh1vsd?2po*IE)0*_R(J_z#p>O<XQI#
zs^wscE5^j{kWRLJ45ytay0%ho#gdNWd+`!!&AX&yq}iuh%uK=v62K)NwAdI{rF_NV
zABpjMWj60@-qecCyl_I1WCu#Ivly1lvy`mCWlwrhuY2Boiz{yaiho+@tVB!5%gEj<
zw$HjH0{^!4VseO^l*8C?LH268L!r1%5dX)Nq!TU8oa(X5_r2`FnP&vGR|TT^2$DJ*
zy%$p&?j=mNZ%nmuT&W;!p>?B<1AF<)LAQS{B08=RMC@V%loXACEFF=0rEG4v>nD1&
z??0e9UJ%{#rw3WPcs<PLeHQ3^;rXspx#yhggC%>}d#2=FVEzgW?2c!#fO-!}oO6`G
zqxy$`PYf-vSapx*db5*JpHa0h=vsdaMTxKn+2zI1@BqX#LRi*JQ?$=I^Pm4asMieb
z_gl3!OX`H0o5zh#&D*U9K(Vxo$#=wkc~J3g^&LS@X6Fm}vf$)`6aST~dcRWQf*pNS
zgzq0ef9D<tMD6pRJ5hW(Mm1yMmn+g)=Sz5Z*9Wha4~{)vF1gHPv3JbHwzjrUi!(E>
zSq4fA@e6K9mPD}o$b-7AdC$M#1bbX=Z8~e0k;34^x5suK*&y#@@VMPRh$&;S5-@HU
zJz>!!neqT}{o#Fp*1GCYh#TWd2POXeAmPrwrKP2LcP8!S#;tO#dk;TTtU@@Oi!f#4
ziA*Yn!2tG*kj|VVEzwKHJ#)_NS66xdk}K`0oR4}F2)aK*6g-PF7FKD|Y;lq-jlD=X
z&Drd=Z&&_{#|AtO&O0OYFFeoN6}3eN);PG}2me>5biq=4W{X{q|MK$qCTzjZRcH2q
zkk4UGHSP1nz>MUk#COl$U)U9;?uTv%t(|7?rT_fy(!eUbP*uNnUd&J;I7*#otg1pw
zRB+zp{L$q7yZ>&Obv1W)T4-Q;kq$ld@y>}>t(Dy~y&%5gK`#gs7zXaB;v+zj%sxK+
zbH9k$=N4SU`2agR2$%8!mauD#hiH+_)pEDHI(brSy}iF_X7mXn=0@XEzND0TueLL3
zoXX41Ocpan@R^(3t6-7Wv6tJXR?&5op-D1vf`*1SM9x0&+0y&_e|-@N^D8V5_WtCE
zc4ZBI%c^}bUo`v)8UA;5z#o0yi>dHpuVDFk!5v@TMs2;(r)h69XD=_@-T&?dawxJ;
z{L2fOh%bsNTC!o$Kx-Z5;1#N_=y1H;%6vO`)iJE&QP@g|WHeVOX{|J$2ukNYN+TVu
z8*6dM$w7|`9OTlI#SEz}GNG)_{C#c7cXV|Br*RGw;_JZf>!-x5Or)>yzN<&m_vP*c
z`70>~3@Yed4g6hmR^`GK2n6WT#vpXBkllBLw+tw(YKl<0CcN(EYH7A(JP1xL#GonA
zce91WA1Pc-1@8XnHxUvShOV&5o&l*d$Jl(0)}+0B^Cn`*)4l6H>aC3QX53KJHZ<2a
zkiECE9+Z5AfEw=^!Hb@Ej3|7VHUHfF<A3KhFGa3m_o#|lrniOG1?}bl5|gdszmGuX
zN^NyI#*BYi_?waPcTU^XZEDq$*%$qf1s^k->bUySpT*{}QFr|!;U<7ZEQSCUasG~2
zl)!iKU-;{r<^>Y@+qu<3C6_&Tdl=ud^WCX0%hjH)&bfk+?@^Ef-1A-_V~{v@w6P|m
z9agMyl_0KYGutJe?$q5Lx9gL#BsaU3l$-W>s9<LbsA&`e_~Yew1ohbbW453_b<HIZ
z4=mc=C+=ymQDt|fKbk(iYu<MKQ%dB|D4Kurn!^AVtw03mp~o74E}2#zlj>Vj3%>4j
zEZ8>Ap_`AQk1NiFj{kA}@YIs>BJV=6bmpii&-GWUI^n4VzAGVZ5U<0Lw>rK?H{1PX
zJviYhVE4?MG`3=5Q2H`u>&C8H$2*4FO`F5k*>LyDmJE2F{Ugb{-I0Cw+n6VBYEzS9
zDo@-OJS$fhwas4PuZmtVk)SaNkE8BNEw){|xn33c`bsX9J|o`v`#O5=U+>=+X_Pg&
zl*N<0tw>J_F)<1D{xfY$t0hHo7fb<7m)E2a;_`Qdhfg`v;mHE5y5PkB5b&B$i3~PI
zrAr*p1}G%LYvM4c%AeX%|Eit&Pu*uyUEzgx*2gUFTn@8x(K3yt({u4Jlw|)}gOW6j
zSt?~hzNr0zS&sgOR~mU`Pt7Rqs=ul0yaeV3K*x5pNs1T*?h4X(zZuqZ-A77C=;_a%
zhkuHF9uOJKq1VirPH_E=T7KR9yc2duKyzjh2@{a8|M3`*?>(XbJk%IY!0!2c|C}y_
z>vc}<2Rw~xPQuVrV3m3hufJHG1I;>^n-~8Q_-s)iUyeYmOsNmW(5He}<M1Um7vSf=
zo>a;peR|bz_}lxrIWaF%3UJhM6$Y;1?a97ekhG36gx<1j*}0E`5C7UDG3O~%wy*!E
zuF)|w1^mBwb!W5RIjqPW`?dEo0+jmC@X7rePcm@=-tg7A(9n}np`uv_f^0=UU;4MH
z=cd1ltp)hiRwsN%T!MOC!S~29L;P7aAhW)ae?eBgM11MgJv5xxkk&tkCGD4|n|>8~
z3LFe&M@g*VZzo)Td#@$JJOWY{*MO$K4$KOdm>PN_yj%Y}!mM#JzGv>-&!)+GDeqKB
z91#xcB~2|$;4y57x;g%6hHiP$FR7IO8+s7WjH@!}ulF6%hN^t<G<BMu8>WCo```ss
zX586_zWzMEp%L9B?>vqwGd?(L$(?K#<uuCTyl*g~#1?I?13Hw`FjGA&ZSQsTMA|s(
z;TT@Pe#-j(lhbJ@+RXNdZeJd?VyBpi+)9c(59oUg*%^I%)`VmP6~cmS46A2Q*?!Or
zU1ju{5d95kfs9Lk(AAfTqAP$5>~HwbGPHDh1r_Stg!9XF^s;kKs<LrYlr_Fl&_a&9
z$b7E|s8n~)&P<he0=~HG-XQ)DR>5@pP}~TzZiX&S^+)|fJ7I03?+91)7$8@jLv%u(
zpVlh>U+m)hPw*Sh6f69nZ2Ujj_<ynyWaexCwf^k?v}_bg1-S!d#-u*z1`wXr0Bp!R
zjibw|IY$*b@spCdz`(PVAhK50zi%{}cLKCNIS=J)LS0?yBnBsFwzh#FgU>>@<3Pui
zlTIvu(91zc3za@cEoME2%FsXs0`=Dq6o1tuXCHdJ0=oAj#qSI>^0V)VY~0jxfMh%)
zCxuK{bzo=$5YDsjIrHdQ&siUkQ?v-6DNitBQR9P>3+~8<e*l%81P?kUV4MrQ;H%K|
zMSw&%<%eexJ;^^84~&P;>QTPo{yk8B$iPf>;ii_R+?+uU-+#b4^v7L@{vHm@K0+#_
zUOiem$eB<@X+YmlVU9YLRWCkvjIft9T0xZ2`Spxpvx)&B$)#`Wfu8`wP1i?BlD6!b
zRGX<g(8QfdE;49s_Urq(5F|yINF7Hq`k8%s$3c8IG8rOadbC88@snI;P##NubH;CP
z=Yq@MK%qm-Z(z}9ONPSa(pIPu#GR_ckEUeLp<|wtdSH45w24<%n6QG+f+;-&6y(_?
zcmVoX&9P>F4d3Mn+=zTB+y`x)R*1j6|E+tmWZYF^_b~h!aA($kHj(I1nDhkI$Ph#M
z@-c+T5;C3+hr(|oSxXx|%|t?ool^=^U(sVSS}Tix{vyd-+{gyxgbo}8?mq&RC8JwW
zQ<BbeVd^x5eES@F3fQ3fTOIhalq1JgA(m&V<xDN>ua9;zUV#~~3E&qvChoWQH`$<K
z)W{0V)K}ui#+~M<5OkL;5$uZFFD}j<))vYzGvL9L3hW0^^K_fBRkq&YAQlymDAurB
zKTVN38w1E^fq<#+5U~wnEjHmdCxU1y(iI*BX3}b(9U=2;WPhdAG`ePb5r~9E$f*I+
zd!Mbd?>PPbir4s5$Uv^?#3llsB_D1;ARaM~3-#{K0$$N1g!8dqxVl6D=hd%W_gn}f
zU<!f5_C+`|;1U@3cz1tz_XLW90zWOOan{jcj+t&PVC*G{Ay8)niPgzW0iCIf(bRq?
zm<k&e&CO9U1kPT3B&&njCmv5?VWDAUkCwQ24>RDrX5u7Qd8(reygLhZgc)#VdL|N9
zT;R}4So|<)WGx7NI!=nKUnBP1@!sdNI@J*g?*`_TAv4^@LjbDU&J4Hjy|F(#>Ft?Q
zZ0gs&coEPxZ&i@d-%^C4@#lbg(h9TGAN8Q4^)Ll`;6wUA!lZcv+4>Ol-($GI%fPtH
ztpQG*CNwa*YL2lncFj(;F}-5L^dCoo6;KJ*`z%2C!v&8T*??y}1=h**?tKG(lo8Zw
zz*>5klYjO-r{9}isrmv^9Pze5+K%JiT7ufpnuWI%pAumPq{6`6vP&5xNRJ4BWDbS#
zXW2hGJMF9jec!-8<ndZ!7Ydk>X3^k-LKmNc#`j1dCtm1*B^mPeXDj&!v<UwO>G;*D
zMUfmA32L51s>cE0Uwlq)b3j9O?5q7ZN>|yQY<63J&8W0ZdndcP$yZM1_9ugAP)6Na
z3bUVtsPQb2{OAXQwB<y1C%8nHI2P4M6z&B@S9C{%W(W?-iDYts&~2v@RA}#A3wqjU
ze1Y1P!Dxx$V|x>jq+vgz_*lgrnkz*ioUq;f{ijl+2GUWKJ<)xzB0noHR$6eK(Io%j
zmB<65<~XK;fe(7@Xu1NE+klEMuL^}mY*1{({+=S-lvzDB90C1*?0tD$jBVThBt+Cm
zDwP@(A?Zemq>1h%$z-Xa)F8B}kcgHmMUfUtA~cbNmJubiOlcE0m6oA>)4ojGG|h7P
z9b+BU{k+ffx!>pgJn!$l|Ijp->6+_0uk$>P@9{mp$8m0il4@<Zq>|0WRJsa2KbUDr
z#vpAmHYBDbxry}71@^UGS5<8r5p>-up)H)=VRj;BMnrAn<eE8`BLfnuab`z4EzRjz
z)M%iA!aWm(nsn$UU{2I7{kbc{iC?y8m_;l^lIyyUo@8q?JGam+$qlkbwyJ73Jl4lX
zoV?&;xHKuO@?nnP>vztzna(h{SMBkGTVX&r)5AYw0*i$@CP|DSpE1~_6-e7HSmaD;
z)z28&XjFc8JlusQj2L|d3DwZS-JfNe_yqbLtk~eKRv9Z8^=k6t`<TDI46mK<e~Xs^
zeb;96g9aj?JzuZh0-wz|o$C-=RV&@{%=O0cyljhy0jsNM>I1zTttY#|GP%GMu0KHK
zIC|lY7g658gFAKT$!E;(odnpsq!3}&m?go&2(gzs`IM1BA0TW~JAan@*ivK0N;J_6
zZxpvf4~%*Y6}X~sGhbzotv0uf&!-&`+PYK5NDAF>6?*g|n=#bJ;E}1STxU{?{msPS
zpMFm3{CmxOpQX7NsQOyvUykOMBds-5LSHmn+yt5?p5Ws1y;j^$AaxlLQID!Y*bo<=
zq)r+cb}HrA2rlt=R#up`&v%FSy1Z3h!hie4f9Myh#gERAGgKXj5)t{~4QWPk?Plll
zqt%MDcFztF3v!hoy2ZT!aJwkYliZ{uZQ=$F>{1|w>q2N*g8Y04Df{l)ZL_R;qd5Ix
z$YP}+4BiaGAu>O*&CwEVDP~?1BGRV4GP2a@{m>1zy^UT*+>{32lfN6A%`Xtf6e`lL
zjWMB#LX3;hzk%H+i{{J0#M^v;A{~W&fNw~_gDGLj_u%O=U_k<CCCFpzZr#47Myc*e
z4R|5uo4hA>X4+9SZ?_ZPHG(6i&abZ@xUg$q2|*||#?yvafjlbW+`^v$uvLa1taH}E
zt#$20^_^xQA2#-W#^|dbODP|9AQ`rlBhv}wku@qBXg13a^{1Tg>@H#|6>)A}z3~~d
zmkY8KJ!7G6-h%z{kDsuCEfZIw>si4t-CM`oF|itX^^n>Vi=I_JpIG~o$@vLZf1b4y
ze=G;hh>qX<ql4@0E29c}?Au8FlBp~*cVan39C<}RG1Q8O;~eu^)A0@Mt+V1mX%7nb
z7=b&{<aKN;@=BYAK377$mwfs{mm%YIUHLco>qdmRD426&*cVXw0eJof%i{VRskZp~
zJNSO5T3OgiFGQ8A_=1x6;tQl@uWhX3TrbFSVvY1Eq2cQ8RwQe8Kz^aCnIgc3Yb(MY
zWG$2cn=c#^OPLAu_`(r74b3V>mJH3SplWC@@hpL+*xHlb6`1qxakO7Ned=REPlzwQ
z&>eEja`heK^^P~iEd{`dP2%3Awkh!}V<wIsg{=wZW6-HmMVwhU#ucZGhr4j>BYvF4
z_=b9FayWM)*e8w@j~*;XvNutipV{CTOFts*M&+I!#2PBySZ+frkr`cIV>;;jEQ9WQ
zUF+=h3H?UL-g;?Hj|`)KWyg{25v`oFSoQ@Td@4u?gtaQfg+BkqdRD~(KQ<OBKh4M7
zdo8mk_0IwX=*BX&u;crkkNX5Ny4-aKstwrFnMG5!AOic@;WUg-!0fXl^^3g#z(xTm
zDHB6<;g8OK<(K(5`u(Q0yde0kAs6%y*T?SBE!^FUZzcwF%h3JL{D1!bGvrvSIu&_S
z#|2OGG464q#!FKFFeI^Izd*aL`<TROd89tB6xSKY5=&$}Q<Xpp21!GIAZ;7H)_jjy
z)}F#Q@5J>4b4B3`n)&v08IwHW=jR>maUI9Oq70kme1)$C1m>bx&drH9#zgRQ#W570
zIOyjx>d<{m0##%~s<y)Wgb$Y%q@6L4JMvy&SJ*E*auy5qlJkNqpVX6bNVtwj;l8kz
zXvw75Q}U;??G6N-D!%(5<w?=2PO_)6cQO}3heP&;XP+B#cO_fTl@$+eCEX;YI&p3h
z0y86=&>IPH&1M*GO&Q+F>0|V^n|lmB9zGD3nvmNR%#^TY8ZusUZ>4W*%#g~_R6*~v
zd?{`Y)`u0JEb+C!Ah|kcm8<Nv0}Kf!&N78Q4K<|XklE5|2zRN!*SfWg8+~<wFi$Im
z3Z9zE&nt6CXedvVT-}&<U2K;8golqXZo+g5bJu6g42TeIVCUmi2)Zmol>tj3)Sn5_
z+&<W!S~z{O*INAOPi1;@!?c{691I!f=?j9Q8$~vDR66LHG+yno)|Zv_HQhgjMJqIc
zJwh(?f}naVd#WnxRVKh5fZPtw8FEl{@8V2u6fxOkE(3Kj``e+96X=@43Tf)W4%0@;
z?Veb}xE(<tb?1c_Lk<&UnWcu@yBw!|_v~5cV&XL<8|M3V>}oQ0*r#L{hgq|xp7;k~
z@Fa8|O!E@hoElEWv2gS2hDWQkt_d<K5cs~|WdG?krVI2<Rvd71x1)Cg2n5oE`GVo}
zkya47CX+cpN8b(Sng!;ivp#tn%n5ha)J%Ro<>1owLeJC()ZI>Y0R^BmiQb9kK-3g$
zhM}+lAQpNsfugVyh3|@4K@6Zh@f^J?yc%YVvnDMZbLemX4L<yz@ZaNfd-`AIzk5_G
zX{bRsOhwpR<1CIfQJvoHk;jH)bm0mv832Q^59kGbGv&UT@zl>4$KDYfM+iP`3I-;6
znR25r8(~0~FnE`RFc#S2SSuPc=YIWo$BsP!fdCdB#Se<@8cKrHq0VQ_GvIUMPh*zk
zQ8*sDDC;vulk&sk7EgL>k8TLZK{WFw!RW`YzdFbl6k(*GtgZ(!`^2|l-E<X0bb|(I
z2o}bm!p?_1F+kbLQdk|}hQuEJS{3n&R|V7u+=}!ABn>y9CU9Xit7DPT%JJ{xQR9CA
zvG?=MPXb8jARo_ZlFu|mEn#~NGWymc3?q^48i(Q`Kb>gx8M8sXmc~bd77H!^_GLo+
zIdBU;4)hmj!6V=K9BaZ`EACc!M3$Yqj(;J?ne9+CH48k5mk*kC70;TLLU-|K8hh+N
zaF6Rz?y7v?mWMQ35h*Nm6+~Ye0!CBEcyely0?-HcrVczTKQw@9w|>TWP;Z$=P56bK
z&E?PKdo2f^7E<`XGP~WJb%3edD8n##>v=*4P@xFcAn7nsrbLO(|DNXm{`&M2%;AZ+
zp^wzC{ifB_evA*ArqMqZi^>1G*-+DtUP>A9U}N2Lhase0XO79YIOebJCB);mg`676
z%pq8euAebWcu55U=XBL)Oga)aiYhm`foHOPKy2b-77zZVpc>ap8HmDW*HsbFt!@R#
zP@e;A-*<`jCTs{dj)zjNq8_4*tXB0SG2|gG96U`sPfPWDOX@#D5~X&n12@w)6eau_
zlT6}-*PX9p{MLc6b}Vpx>!#5y1)W%i8#TBAnq=N?<STV1d04ZjgIOtM;<(-5^o)kp
z`wxyerY10u8bobA0Xt%xb%bwpCIW~abpTHF9O|q`V5WzZB3-!qNZUhZP$tX`Z+G~D
zLR%^W3t^}U#<Lhw1wYg+{2RCM);zu$cOO&@>Y!;DXVDjM{No_c(Y66#Zx6oNgfzSZ
z7Vu9c4Bu&=3@^oZR%${#YTdI7p(T$!-{eeybo1<y2Y?BT%wW5ovk>x3<Ju*Td)P)W
z@>}y8!3wqB5ur-BN(j1&_4YjEWoo*S&Jd(+e&QEyUg+~U{UQGLL4W(8{}~@dkzw|+
z3GNz9@dq}V*udinrfBQ7N9NZeviBa?Iemvn%Nyx;>)WOjoTej8SHCO*OOj&%l?2`R
zk%eB`HFq-S1H*mZeIh_$r(E)I57?Lxn{f^aei{=i#`V-O7rB>IySt@4J3k|&&2)Oy
zkU+%tmA#~?#sU`Dv(#q05^5&{tl~x$=kie@WGQ}B5XkCebb~p0(3e_j^chp?lHz_*
zXVPg}mg>kdXcg3Tpae|O<0buOLliGkzlG-<NG%mcf5x;Q*eUSdvkZiSM@ciNU8)dX
z-6nd`sACR6`hKHlA+G<JCE%<_Az(K}NNZ>%<j4A)o8eygE=4L!&{(MVZ7av42xTLh
z38qf?rXFOt3IY%epCAqF){nHLjo_@IyD{r*f6XH&*5%ao3jxt^3NBd;vZyGpQRzAw
zyNJhhV8#lU97!XX_0S;s#)#<x2jHdWPQiCRL6(e>LkYAG(>jzN_c>KRWl;wc)wNk%
zy%Uxy?5lM|^ol^<H%lwfW(k)6NA}U=9g~Evj|Bl&zeI_#fTe9Cc4S&Ll7$S4S%&k!
z)0tiNjc_}GE{62CIWipF&(l-_o+(J6vI|Ctqr`!`pTNhS;UP<I^6)L0<B|OZNn$v1
z++;|Gwx*v1u1d0wWgz1IoU{$CMEad;*Ftjcq7zrqH*-(agp4$X9XxVxrVB*dycCsw
zf@|Ray0~gbnDo0DlG@e9f;JD_zvfqi=R1wG)Lo96633t0tRQ@qDO9|>HRnQX$J(+)
zE6Zpo%bolk=6rwU7k?a;sQSU8E)qQd%`5$2e!e?AmkU2qpKXFOzn|#atBA)mbVCKC
z!(jeJq%=B>Ooc!}idAkg?%`sgb>A^@yzMV3&YhrK<#UB!1B~%1lK=L8fA0PIQoC_%
zdpq1DPvs60HX&WsHeTOKB;LElJUp=K-MItV#sQO82BkER=>0V$^ujwp$Y&lSeJ?Ov
z?H{Sb&i+<Z-aHPnjcdSO(^nBt4Rcf+z}H1P=rw2>7)8xtPm$W5k@|FnO7q#M2tA{$
zQ|^~<77MjUrT_{u3?8P&9ISZy0#ly2yaVFG-D*Fu6windO<+$#)(vh^Coks@%Rl%5
zwHi$pILEJWwkV@EE5O3FAn~HA&A;ONpW&Q3Ow1qJma*~*<y64@mo4$bJ4XBgMgM&B
z(@b`yFZhg6MHpZ`4<){SR>#MtMor*nOw%gs(@lUWDnV$=c?>V6<-3@JpEXrZ4Mp|V
zqiulZI?w+a)WUMXz9usllQ_Y=9^a3E87%1~-B4$NLz<-gHNFPkb^?yIu5;kfmpj&m
z;LiIDQs@LI*s>I1#JH~^9ff`-I@IRWb}B>Fmdf6qIVWnu3*1!XRe3Z*>rO=j%v$4a
z)5AG|W*S`2Dx@tB8fhSobt9T&Sr4^3u7)}BRmu>p8~moyjwYvjH$G$Tb=r+cdI^tl
zeJ0H*og@sk`2}YUq!2>&f!2C;U+^v$KyGtm<~-zsB}8Z&C;}AEF^8Z*V*ob<-`t!I
zS%7^o!o@O~&XicH)$HgSjGJ_s&9|>uZ2h49z(IL2XUb$R{Hfj_-jMUDkIb26XhNq2
zsMm=q<*;SXy{GgfOiVa6KiPY#*}B{*@22J--BZWWN+m+t)fr-x`om;i+LaLZQ`k*;
z?ZiCg2vq#<tppzRhG*8e1$TX7J9OZsJ(t<yu#Q^Nws#js%az`^mT0L@m@gDAuzAs%
zs-TQE!LBBrCS2{xud|jvh~)V#Eb>aelM5pN>cVj(bgCS$xj+iM<;2b8i!aCM=rDuq
z&LN49uBkod9wZh~=GPDt8VP(8&JC(g0~wJT1PU9`4KJ3ox~qaCht+8zJ0jc$-Fwyg
zggEDV@b!4s6pA7LyQt}Q-_liU`CY9ApC0JfY5FsagebfS)vkvq93+LFqGMbv1)?hl
z{K%SKIk>F6R5V1Fv<x6X4O<;4%y!p=Db4`%5?ztLaU55SA@X*6ZO-Iugt_GhzT^FM
z<$U{(R2%AuDpicywXY1|&ArWi(4(>w|G^sFO3ngmzz)Z9dQJ%)b0`XXD;=zbLh5;Z
zmjVET=m=gd#poj#<sJf4>FxwV<&tUPM`g@xiT%&lt?J&1zuJXL&}%#jAn4=*LT$Q_
ze4su+x-~Iem^|k`0sHT73X$9yTT~5s+0UE=iA&`TFs%TH$e_+PQP{ROs8o2CDr>Qs
z6gnxESw<sEX-&QFiYw57)}QItv39K5;0EqS*7AcZj1(76F;t8>HJeyfCKj@i^lqI;
zEL(<K7l1J4+T+<<z1FBQ9h@qCP(O%)>5@f|tp#u@lE5}biih#&gpW{i7%>8yCb4dy
zvgV8^uthxxXVMcOwAF`cUCaC4G*a?L+)p|5;Yguo!CR7_xb5xJ#a(@$eEU`Z<&~GB
z6HgNxxt@2C?3L7}fkqPd)GnmW8H{%fe;$_iJHH_QKVKU4KT|;T?@YD+d#T!Fyi@PU
znxYv9EfH4b-6UGx5W}w<rn}ObMl5RxY{}qNO=hN-aF6zEAXyCPWbV0m{NQb6zr?17
zV9pgXz`74+IiM++>4lHUa6>oHaNrM%^$wV3X7I?5V9-0X&3`;9TKPX++=L6m)3;NB
zPRD}Y0rExBwbWJigEvrN@LURDPk?@(h#z#LwY>ep1M>{fqzPbv>Uh$|4EWK#nRzU8
zEC|q<bX*(STVIX?<4LJ<;?7PfBI*K14W-H~E~KAIF6!X6pf%Kf7wuInTPn-(Ddn<#
zC#5PC-5Q%;X9PycsG_`I1tT%WzsD8YS7Ztxg)U5I$TKYrhCI_{V<7`J80f-0cd&Vn
zGK$c1mV%btF^7P)(`QS<=Zaj#(M(5VFvM=;SY|Z$A_Bo0OdgU3o@JnpaqZ0`k1UB?
z38$)v{*uNb6Qel0;jSMv;n^Z~cpw;04Phi0`du(;O1cdQs|g%U`<(K=43EofLHL4V
zh&i8i;92@W8dpZ%``cLq=llPov&J)(dO6{8x@aEP@H57imJ>FN<A@KVhC#O>J%z>q
zFWn~C@S2Bj*aU_;n+k@jOQ!;1+;ZPdvPJW$C!28G^D;$fX%zM%53MT?DWFqt0f;BV
z>gjx)1G3dfs!HcLM?sPI9o~O`<rfF@5=B1<<ey>;;crk&_`A$#SJ?Mjim|~7t(e-Z
z0_7pBsd1JIfl?uJ1ds%i@eX2~zjG)5#HP*v)?zV1Z)RGb?k1)x{q=C;vV`c&<g2O1
z<|ATno0ku~tM17W)eh@JI9lP^xWTQ<sPsd4X1N=_I~xQ<+3{e~w(>@t@-hO$bpjGF
zgsFgD$fKdt!R*O^z}2umn585|Vy?t-?5K(mEtTUcsg1R_py{{)JvAsfW5M9Ag?-z+
z%nLOlvvTZ@O>?Tu@K{a?D<2jDk$;w!99XPd1!!nBEoYF2-31sY^Qa>$gT3m81{w+6
zWp>Q`%y-OGxR|4KSv1q#0Ibhj((Erq1wz%N6bdI)70?x$Hw8a9#lE4EeUXs@X6Yas
z50`#h2t6a>zV@pQ+X=0y>&G*(Xl{Z6@Q-_lxe^R8)-_+Ox))5s&F}Gp2lL$5fBm?P
zXI)*tbL3nw))sIs11rkK&TecK8{O)hiyP{U1?#rdrOyK+QSA<)Wk_fPo;__B(!PO+
zwJ6BJjV#s{2V<GyhhABLK@3k1bIv7dLv$YH|0=<Q&kg@7<^QQn;z{$qE0&5UOTw!8
znKd;zxgBQSk0Rq=D<1eNMR=8zb^X)Js5QYO5o4G`Ok(W-st2j68bS+Ci}&NrPfAJf
zF@&T56RPmsh_oe5?hE*WZZF|qBkNv!?>-_2(E}}f0eaR(@Y?re34d&zSW6g@5MoEe
zgi4(MgC0=PKj^k(>ipEpWrOn!g!=OE?6t<Af8CAr`a*9TCXRtJ!}^38T0%k2HiIBW
zYQt3u?qchINR-EFytcQv0mTT8jQrCFShSW_j#_N@BtZI7knGJqTL;bO5M9_%*N?+`
z0~?rYFky@C`$$ndbvOq(hiH-M@5#wh+*uZG&>&YFfvzAqn>~w3gWFtsIkO?c&odUH
zk?7mY4={b|&mJ)e-+=MSp15^rghD!seyohVonVWbq#kJ5V)&-BccSf+YJ##<$ZqIO
z-cu0rM{WpV2TAm6I;a|<8>^0}u$;ITE1l3Xil1^1p1DYhbCs)s=5>Q-w<G3U!Hhx!
zdxl2gg=B%jqA?Ue<_m5b(%j+hFj$HV);N&n7YmG5KBZsHu&<Zmgp>U6j1@MJPu(p6
zdmyFYIX8*9s_bdifB;SE)qCqk9RX2F8It70xjB3W-`Ua%OJE@-T)6}P0RV|x7I|m@
zz{cHBW1*liIN&sP3|cT2AlV|>_$GaBtlH##p>l~)Z@N7HXc`eGNzy8*bVA+8nglGG
z4h3^cr!g#`OX}}NP9YwIK<?=ABLq$^&)NXA$|-ag1A3sb6P}k9DDR2Qcd4DE4m$|?
z#uz8caV{CLCkb<qzL-|;<DW4?MRyDxVhKoVp<C;N8R@pEHhP^c)@!;+K4HVX9Px{&
z;nM_gQQf_0P-+yG#`^9T{%>Ta1&-KOo@qAwx=g^pSli(SI#s#!x&~4xREvGa<Rce7
zr!t4!qL{k*GKqor3ici7bh16Ubopf0{@f*(RuIfa#8H_Y3~wNKE>e*Cmrl@r27jB{
z)vb$qz>|cAoJ-s+w7HJA4?VpCTL~`m@C@UCL}k>HBRbGa8Rwb358OA>6K~xI17-d!
z<wbe?{;P{sTMoF^9yc|e^l1G=??nRm(L4ss*QaEduR(wl8i6Y_m`4d;grPWcBdE>w
znR3VnHv+)yPD_QcbcdHRI$D&a-c>g`gN~6HB6k4@RwC3R!zoy<Mm4Q4Z1kaGp?9^w
zqx_d;Qe$9IXTxiDFgleDIr;)HfJq2Eh05as^iRCF{#U*sb14OA3@cJvUoe_0VhuQ}
zSSXJtj+S6u20r8+@Uul1kfBfefM_TpT^$<ds1EKP?+<42{o~6k{O78(|E;yAf}hA$
zKB$ooL^h9Z`JsNKbL#8__7SA;kv*Ov{)k5q)9-y{dF%W1IoE`Is9ny_fCqDnV0vbb
zJ@oaX<KP$C$2y3*n%`*)BDA6l_`&CKO)_6~-s9jG(oxR}YX250JZJVS03^lpD1Kjs
z-reAw(@_1-7y&>#VRw<XH~viT%sOrwwHblR!V6~lXuvbd`7hzU;5g{T<2D;X3g1B~
zVg?vU?t-99)u;(+z2{c-*KoGcUE&p17{#}@Mc><<I0IYpOlGjx8|mMj4i0u7R~1Nw
zrYLN?mFas`N0+Iw7j{OA1w4}TYO`O<7TS(BLU%ZOQqgd?4=fjzrd}3G;)=_;({ylU
z6ozOFHxCR##T6Tw%d4kQhbw9QrIwv+z)^)Sc}1O%YQoU~sPUR)Agia4+cf>`16I)D
zFlSg4!EEw3?bz`zOz;1}yae`b9dofC$J=ujSDnfgen(wwKV**z!&0A1K8H5~Ke*uu
z<+a(@*<{X015;#PPvD+6Bcc~2`iwjh*cQM_C(xD9Toq-2O8(4mGrH-2Z{A`wVeuuq
zoJWQ1GN>3RH^^MY2rsRZjYXU=f;6lY`)yyS>!P?~&!sTucJ`6Ez9~5F;*fL}M5V+p
zHdsmhHZbIo@Z)YTNrE^Aq%dEQhA2boat5j_8Y?0Q{%#nZ3bT4al_}Wn6T^_YHm7Bo
z499ZBi;SO?WG-{1SW@*v>Ne9OqmOsWYdp?dm$di!+yjY6L)P@_;F4w$bIJ5REL27=
za}7ngV1fFLaH9`kJW};|ArN*3QxKkW>jivoIS_-2wCC<(+~`bc4w7CmSTxx-BH>*A
zMZ44$>a48$ue*~c^~YQajeIWHsUu(dOBFQ8sQ6`f;|ZyLXzM`b5RU2q6?q8VIYyf$
zSjw-f+hfOHBE3k^|9db{oDOE=M+zkF7M;rRA#jg7H7nnVnvh<ai|dyHYYM?w+x(1W
zzrOcc{ynYI&Drr8BM*qA7x#%LrC1DR;hQqyM@)gAjjxY_Av~t($MpeM$(1095v!mK
z1sPcb+(b6Wv!S#Jaxgkfzyu2j*;IyfDRuCk<^XFD3p%j=og7)Pa<{-3ZMKTSTK%t=
zFF*(aAbS}M$3<2+TKY8v`Gu(|7Rvd8AQ<E>`(aP8#UIW<_}{I`e4JUz4fO2I$Fpas
zlew`vILR*=;SPO{_*xXNy#SU;82HL{p60YM$1t7|-!v#c5_3eSqyJ0VTKt;|w-d~F
zTUa`>)eae5d$-m)#th9X$3x-gX3U#ADflE80#tau5AL$q!7KWV!d^xku_LQQtkIVz
zpy`6=ELzWScd4zjEql3L`!U}YmAB4dSf|36x?fp~Bd*ct7QyT~#+quRecfAkJZa34
zuOA$*c+9@5cT6od4<C*zJ*PaQ+LJ~WlAgfg@}GgU&7kEZe<g6nB~?gc>6ckwKR!-A
zBeR%Gr8c)}tp^?T3j_@*AW1=3dy7s$4K=)|UHBLvU-r1PsJGXFd2fE=It|TXk<hRL
z9r-}^UNGoyT-hi!90J7xk~-}@`6-rjGc}mnIf4vsaT&0{Nc2#j;5au`bEs^Qk4Rg#
zrNnsPV`>Mk4+_S<%Ne#N+Wq?etF8IJ)aOHB1i-|k-++7~yd>WPPVV+@+9wsy1SES4
zzFC|e+YaG#o2ko);nyi<?@2@YmE9qWkTh*G1YY|y7gJKG_S6xHOn7V`&cz^YNu*id
zn7qFk5&;q@-?H&Ya|>+tFaT2NTb+#$#{nZcI4MIk1-%#j04%Q=NVv9vj`1*>L+)Sg
z4v|2T6hMcPt3!S29#7b!$g74v9GYtZp~FV=VxinJSk4kGpiU;10WPEzfOci7A3zmi
zi;L$q_<D7Gvpvjut~bERk}?5S_}_5hZ@BPpg9~j1U2B@G=%o!s`E}=)?^+nM%Filh
z^P^39k~)^n5=#Z<%C9{Ug{|1EJTMK!Z4Z%>)ewSbwSz7Q2U55lkcjX@g#Zrcw+Rk1
z{5daqojE$lgIky4>6##bBkf1ul9Q*ClQqtA3=(RXCqyP&%xtNhIOC5}C-DinA+Mk1
zFV~Up`zCD^;L8sEr<WJd#%@3P*9De>`BYmEedG7Yp`!DezTqw^x5`#A4lX=bxn{<V
z``MMR*Uy=A#Ga0zlB8UOB{Ss25rxp>BBh2<c>^euMpdrc$1i@F!-7(n&zNbdjd-re
z=~Pq*5O!Y4=mRt{02bXb_Agq`<M)!KYdR|f-^FNoE=$o1X|cXuK5<p6jkV&Lkge-F
z>m%5HFOmMemaD+Jy{rPaUPa6)`vM-uCXt559vb|LGyb>bl}3XbORLcc=(s>z-6(A1
zv3Lvs0^~HBT9Ifl8Pw>yDZq_JajmHLWn>|!p{D&puegk?>mxGSm{1Li_m}^T{xsNQ
z?^T&qPwwwF_e%==pzCy`FY1t1Xv#*y>3V8IJ!EzLxD(ZSIT(L0y!CxbK91>%@16&x
zTp|!cVdhiZxYuA0+(7ayRMXWe%W_%T;~NI$GF@Afb{B|~7rwU*pS}jS0=D&4Nc<^2
zxqwvQ%V>n}%2S=$X8i5aOltTm3-ABXKl5VEgbZj|7ZJ`*`ixl#>0D^?uoB4B6ZjUJ
zF-qtOX>LU$62N`9;IHwT?EQH2v*=9p4S$7xTp)i|45Q4S_p;C6uilRm-TzH)Ey|zx
zO5-cu2bQMaLDlf_#cdTS<|<2St(QmF?okxD=Ilkn+;sM`S3qZyCm_WQKtt5$V}Vdt
zZ;KCRZsvh1U0-nAKC|)6hxmrYbS18+3=W+R=oj`e8Ymsh2XTRx@D3}ob-BY)diLmr
zFQZnk;<b_Bh51Kc6ot9e2gi8s{0hY?>r(1C{&f?Ot_|<0gS#mH_T5gEy{Mkk89Za2
zEs3?1n5Xu2@!>eLpv2zu1|&@w@FPX}3xO>>2?A;=WY3a#4OA+_=%4-rOB;u9UFJYc
zuB#`oMxFy1WDxi1H7l5*0<{N`FqZC+6l+&36KJAJBFzt7i_jdiU|zwThow)SHM%)A
z0Mf<kJb}xWu+}on7;=dDFtZ#++Y>^Ow>HR7>wTnsEfDnYKVuS*2lqP+kP2#9*>UK#
zf`8P?3BrlN+4L%q!Gjs@Mq`oOpg`Z3hHzu?S^3B<(mg!gh%h3<K|ri@4H<O7c*fBF
zzgE2l@5n4x)9{sZvn)OSJT~y)rk<XrGk-W-nyq|k(<_B_D?|<<FL%IWouh{5%+BWJ
za3^O-T!y_}VQb5_BgG)kv=aKVyVL@0h(d1X=mP2uz(Jze6AV@MdWa{RU9fB+Tga0x
z$(&C~Me1)0;*x{zXf|yH@KIvYCxMY@!@>G%e!$c~b2Lw)R|+f<%QJK`x|!^Qbghg<
z+P54R=o*`fg52ro0#8W#j;`FKh>b~D=}3IWjFH$foGMjO!`@sc_G}kCkGcf)Ije{v
z4Kp~Fj%Qc;p*MQ$n_rJUWFy$>B!|4}Zp5;ufb^lQCDHT+Wz+%SGRNehT<D+1k=ink
z@eJIr{18c!ScxQ2YM2mQ?4Jzu0O)-hzv_M581`^^F5ElRE~&fl<!>UGzL~Y$2pG&7
z#D>D2)Q)GZXz|jX$Ft#-^573n`I$4B`{>w4so->5Eqc2tE9J^cHL;OIjV1H<=_{!$
z#HM)epj9KHaE4q@V6#8C8S<i^GJ#=)qoLH52M)-H2o6Y=1Bq|nX6`j=+hUMB)(awm
zo{}XxxECaPHTj7Ycal1ZgK_GkxHFITDP)nP(s3Grp13{L)w|4VQ+#e*OTRkvuC2m8
z!s@E|14EP+KuWOQl;5fvac`uU2U=}QDV>XmSD#U-elzi4W`V=G>SmFGiD$ZXaCS6)
zx{I<Lwr_~dA`h;%L72;ZV86Y@AQFzF=`q7LCy%8OF#!yV&H5nObti#$#Kh_y7&Vto
z%Scvfk1p;kRZU2dU2{EbVM^GcfUtV6H<z>P2Xv-?qglVW!vBCWpnppv2EiPvozuuF
z?0M=?uQiB)j?xC&NyVgY@HGpw`=ro}E)c~gYN8%rlU_Z%rCNB#wF)hm=EN0uAYqBs
zJZVnoguYa65R{@Vtjv`ETDJ5|Ya5(HpuU$qrw2b6*^-H)j6L-AqfauyQxx8GpA6Xr
z@=T=NnV7FWR=D&v``5@#K$@?67D1@J2+T{AE=iO&j&SK~zGj@J4HO_+#<q9(`f*(b
z3b%3->atI_gTXSCb0?9;5iYf1C|j9I;FfP>kr+5|Loh%C<7M+UavvehpVB=*`1uCX
z_*RsN&UQkT5W3a8pl*mmWi8YkNdz;4W58&K1~b9w?Arvt5sM@S@m&A`lyG#=V!Z@&
zyD&;rw~59H$DgM%mUr+(Cn~4boh~S#j4VUMkS;6mkAsiR`e7HJPSp=i=++xqz~|S2
zN1c@CeuK2zGY9EEUJHgAeCTJ)re%u+hc&<CTS|Y(qG&6>WM10;Z1)A5+gA^L6jlpt
ziwb3#+~UKECT6D+j@OvA)ckQQ$Ld(y>mw#tSF9H*lfk6$iGy%_q9NkITM+ZBvf=5K
z{7ysY+MsL*X4)c^Sk!-E8c2c%KsqGAoz5?sFtVT4yuH_9{=<L}V=oLxd87^OPdGOU
ztlk}aDifyxK8SP@&8m~kRC3qp>%DlqXpU>_HkyOV;F^U&%5SbO{5|yDg_Dbp>8w_?
zE-n^o>*lEd_yk>0*!~maj_$O!OJ1!n)yq?cX@`?^*P}`4mO3u0c)iAKs<p^kg-vni
zO|Bi>7PP5A?zHh_?$3_FB$y&OyFA|@gWKkVFZAez{+<cXsfWoQk6EZ3uVWJK$ZiOP
zN1lm_M@H%oMPV)Pq|2f*IShbOcP>bw(|Qr+TQ|I)mR<v1hG~|QA33#rV^Va7%90$-
zJjHnd%dC*AIte+8=iF~Em?aRyYnkvDn{@1VP0d56Fx?t_`;g$Fm=d+5BUdx+brrK-
z*FO&UUHUnOQ6Ojk3TcSPAEdGtT}J9NhoLm1Sg2y=k1oa%%_{5z=$k<BM+afUArC1X
z4X!WyiFNE{u07~ke!V%2E0HtH3f<HSd4F{!G&I9=HLdexcLQX*r@HpmW6y91^&VU7
zSnFz2SNnh3A2K6<b&S630+W5d+t)~pD17Bb9X(6P{SNpiWUgx`nWb6rrDwQJ(a0+D
z$09V{rHCtMPZIkZi-U#z&&A>ap4swBNpyL?KA(c&3M0=KlLMWO6vvG?%vavhb>@+d
z$|GkVe@-g}vY7`-BMSIrRqn;3pE1f%0koDUnsoxvEW(7Iu<FDFYW+QQX7ai2dgiGU
zNtp(?%0;IGPWn>kf|!<uW1dnyLv6bWDf@cpM_TTtbe4M9X^JbQK`KalseXOhMOSx;
z1>Wl(WK`2|j@Nz{ex@Vkc<h*{J^f+k2PPoeg-(5~Xr>k(JSDHb;ERM#geRec#XT9Y
zQa(g6AnKI`F0Bsz(#@9G4DEGU5D;S*&^5T?7pWXxGg}!-A}Z*9;62rbFu_iHd2#w$
z?TcLUeOatZi^hW2d2?dU2uB^AJG)?j(z-PhWjZwBBwS?~v-%dM)z_G{>XqpShAcm7
ztE{`M;7MwA+7%z4AU%vHz_Xt+GkwO1+n+Jok3l#hf=X}(kQ_s_7)+6R9R~}p1$HKu
zAu#|l-MNqmgW>*F+ll8h)23l%D|K({geq*Tbv}~iQ(8;CMHpqJP1sY}bD24L+N-&@
z;#2BochEH7-Kk8dKGfJ_8nOMtFwuGDk+sG`C0|uhd%}Y=6*;QR9ab@xoXvgn+jPwa
z&fT}SZ5<GKzHOman7h^Kp#sa5Z!s>(L?2|d<%H`UWjx&8O4QfUD{znNxU)7<=_EH8
ze)}4bayi&<EY;F)s4BT{8<F?I!hEY(G<KHHqAJq#6>FLWyLBYR|7Wu6$yYTV&vor?
zc3C%N!+P#wXixZ4IZJG%c8~PwHmc?ndg5~GNi#`p%WEs^)H)9yp-<hcwchZs!s^0&
z5Bq_%ah{Wbwqk0en~mg@$_=xd7b(qgBjWgcEuvLN2T+oX<IXYy;49d|BQeIhOC8)c
zBzv<-XcP9_w(RMdGEXPUPnqA*L0xO#gRKfehj;b~z3e7zyVc{^!x6drR_*c3BE=A`
zu0L}74QCo?r_i}y>vaH5O@hcTG?)rPBS$h!sJO<>Y$sG0fT9u(0bE}Lt3{8P3&B~G
zJGjJ<*!nv5^!7T|4!Kj@(?TziktLq6gdV-S1&eBVu~5?g7e$u;{dF7Ny&oy45sl#w
zV`Ul2YfEO|D8n8b17euIU!uq+COrIw3VkH#o<~9ho_^6N_#yrK@eu!Zt&8CWjCPoB
zA!kW2x<#14UW8)@N%XmRMB_NuW|0}j_~AK=1RFbe(p_#N7rwc<f&(!Z=>^a?>pVqi
z-PEq4KJf<X%c74fvsmO@WX3W-hliGS`-SXJ#a6ewEK)vi@+xaj^BEdcBB+dAt~M|&
zA91hpP@EnxJnTAjrH=f7X-Ga}=S+Pkg$HdLW@6h(+GC6X<tE+{ALyx6oJ{bQ>COI(
znN8Z&E;DrR_Ug#|-42spE3g0bMCo<4<Lc@aD{KS@iN8qT@lR5)BSYaR(}_-Ly)kBa
z;l?f=L6$W7t_<H=L1M$bCab3$kX%w9r}AO~cA{7%^_J5Zvshj)XJ}mEIm=zULfzgk
zEi;qe6RtI|Iq&Y(-42&;)dVlNenBzvo|op5@1p2$uj+%UBh$>e1k~G(c7fUqs|7>q
zG&w%bL7h26-)dFmllYtN?)bHZ{+E~u|6lh{ioaz|2;ZKvYyRGtJ*&)DXH7i8Di*p{
zVxiBOpGqC5z|%O3Ak;P;Mo$;8Kt+ymt|;Y3?9;d!Vue=@plFvr;66oKUss{r&zPZ(
z-aq*xhqiFvj@PH$uQgU~dbv!XNz4BeNTG{3Bj!jQ70qxjg({acI-#qm$wIav_M)1D
z{!VJ2qF8l~`uVdtGH+L!d!dKVf1G~yreHyahdyi+n$XJ`3Lb*0226gLW4ueCriTLY
z{b8s0WnSXuMVPD76K-W`t}=O~qo9y*6i26c&Vq<Tn_h<sCs3LGLs+Js6Ve)wP?!*?
z@|~+6dQ=jr!blt0)d`H7X&p!Jpa0{AM*6V#ypo0VGVK{9R_~WB5HCFNbjC&NYnX|F
zGc3+!VH_2-gbQE@-RQCC1zx*ix{1SQOfK<0@=_T**g7jN@-q0xl^F|#1B%XI?f{Iz
z&Ow+<o`NxlNAXYp`Ep3sO;tUsGI+{eAzC|Jq9kOk<fF5dcVo65*tNM!biKk6>1^$k
zH^}JkP^ML_Pt&yf&};N&o_llC?0ij=pX2qHuWFi019d`eyQ!H(tIwcepoY;3S@Yd$
zOv@?L6JF@fI=J~B<I~eKp@s3}(GTLZ-=EB_QwT@7?i_xQkgfrSB+dz)AN)`M?GYB)
zFI_Cb{6nQZJT#~J()lyY^pEbF{=jKQzQP5IO?tC&OX`pAf=Bh|Q)q`GJ+FCN^1WTJ
zbO@ex=P<`2>RvjXl{s)!O&&uZkW!SZG@&@%bpa{)n2`o*z#5qSx@=O))0lXhTH%r*
z5qXkmo<#23`%}DMyvv)VbftfeckG%x4+xq0CThbF8SAQ$>zw1<h?`>b;%-H8LOQ`^
z-}LS|i$jtgxa`0GeD$`h!mN1YpB{<7G>R|k{Io<GE%b&EF(&@^a;F$!`Nm#?@2P8x
zUV8T{FUL$5b(H|WB|O|f;EMNx%ddY}^%--X0*SJpmIX)muZ5oRFxVt$mJLXG^h`IY
z(a54s?^c&JPg9-X>RH(+y{+-8%8DAr!16Q1jZ=glN3NgaW9i2~P$9VKmikLJ9SrE&
zdpfcJ+7#k1xJ%GGx6zzbQ;<n*W`WH`*Ox6j-nKA8*K0-3E5kWQL+r7rMDI&7ONrMy
zJlh07j+kXC(omI__T15QQ`shGZvxYV@Q;`262|ogoZo2`j!a^3{jJF_86Q*g)@9z9
zxII6IdAUS)@$lq~X{eZa-AE^{6P?Ak{JpdqhqFl{F3N>Na{5`BxEC*8T#kvbJ^too
zh{2p5>k|if?jz(FVPV|ZWm|@CkjVIqiA+MN4SdF15GGlJB?mC*9ghYoN5UNGlZcIG
zlCrer%WhYwVJF+1KFH1NUD8M+_)3Q!3v)>HpMB6kNa)h0#NqL;aj~33@WOXrEq*7W
zQzlZA`=D7*wEUfw#odA0C&<ssu$<b*X86#Hx^6_ZM=JN-yKyS6s`KQ+*NUZ_$88<H
zMsHh^t!D2zV(=E@me-TGKRpuv*!*e1zN=e8eadbQJ&ZM9ASV)*w0l+@R@K+E-lu48
z(48bL@zZ2+ExFqqz4tpSKf2y|UAMy<Zmyz0qIv%TV%77Ln5;Q7R~FO$8!h`dZo$b$
z27gv-@lo}<bLq;_ZKc&pVTxBxx0$EqZ(1uIExfr|&?oN%iEl<n{~SvJL-t{q=wYc`
zHq_UY{4@S@J6eSw_ypbpSUlFy1M3IZ4?7)Xt4;efarMru!{kP}j-Mvlgl#doXPRN$
zu{NVDV^M<aR*yx3*uVCTgXAjxHEg-B7hK-^WAgtx#xZ<%()P;*pZt0A{F8k#;y@nh
z2fs!|xTgzg3dJ14-U!l#<8m6NkAv2)$UbQon)}#m=V}ANMSEulG_6)VnHW6&nY5^5
z7Oe%i{Hd8U^r9Eu$;8lyuVrr)4YN-T4IXKY=%oIfqNHW6&OGQ5;Hg>b`DT^os%Y$a
zwf3IwNXB~XhmY|JwtKu4SM2kiK~=JR7KH_8JlZM|w>gO3h_$`ST6W=HrLCsN182F6
zo5~Xx?ddr@f8Ry}#rk6g>ScU)9sFbTjD=M7T==aX{~WRZB~vc+`}=#PTD@z^)89U;
zA~zJ>A5Q2PlQ6g6UE-*&d?iB3MEHj1VOHn|V<CY*|AN;fGn>l%tCgzfk6%==%eyOn
zN~3r|u3OpORRxod3F!t5?4gbJs~NiEkNz4-rV-uChFqzeB`-g4L=^ux;Y9xjFh!>l
zPmXymmb>lQ7R-gvuw}rrFsZH5n;B>wQq`TS%TOt<l6~pi0$w9Vzi>F?Ij_Uc&j+P5
zM+=pw>s4HIXm~@OV)MjymE-&DLyQ?I$=kIJSNu-gpHx_g5fA;xzWUb*JI=cP7vNqm
zXKlR)k?`of3x0}C_*G4_7az0L&#urr{CLKm4FNG$({)}MuRIeO7W|hTA6)gn;5tP~
zxt?Nl$!Mp$1#zyrdS%<+zhCW8(ehH-v{r(h^t<%dSG30g0(+=qp04Nc`m`5@2Z(z8
z3R8{|R`<T=2c}-*EHV8tr}&czvf9Hbr*=x8dSi#Y0^RrG!x}~V8CQE=N;iwe*66%f
zmW$S!UT=#3XU{gym2B6mS3Os}IaIeEE2^(}V8!Eh)9f(6x!&S){RfMFEVy5ve*F-6
zy6d#u(^o2s&b!Or?g$^2b$@d)L0h`T!(Za$rj9?3L<xjuVU9m!AvvV8w3A8Z26T4L
znFFg8ZB0R+RjgdX=WzGna(g9qbHwQKNqxlJ8HFCPt|#UlRg-+SQu9f?tI;Rontf&d
zm=jOvX$|tXPR#Xd0J3oRX5W$IxIN9G?Q%OF(}_72tNQD<m8MT?V4J7ymfUsd;+9kc
zh09Y89zUpKa>Z-C(A<9INTx&|c?0)>yR>R9S?EJ``>IvLA?^9HT<;x<i$WyjcW8Fc
z-xlPY-IziCxl{#OBf@=9IeV#rjtHu^+O*E9d#XzZ+8-G%4LbcKZSf(cvIw9A{)HJf
z9yl`%3T+_yu_i~0Qf|AdZYZU6Us%4!+_qR(?SS{)xI4bm6N9k2K)(JNh)<!NL8UFL
z-<&#CkH->@u1yXxSQzfu(^IhYkG!dhp&<mtvVVSUj<c|jqgZ;dSF*=x*&4x@GRrQz
zK5TGY8b2@f`uZK0<Yvv7Qa|u2G4{upNg!K_mvyPmY_FNU%yr`MY$zX?hsS10%rszn
zG(7o?Il>DwLdV)R;9Y;BbiV1KBTpJT!U-O@oJ=A5V5)eVJhnuwaM^7)+lY#nv&}Xs
zX1m_J<NfM`^p5PTSBb4ZF6&pQIM(O=(+hH(C1ft+1ffv08EKcT{EU*r4k@cSanGV2
z`7Y{FkRDlJVk~z`>o<i^i-JFUfMcSt7b>a^8ZW1HPECxV3;FqM+=Y2zlDOK~t-#{s
zk8>!0DheQ_)OjP;r1KF<z~;=dsSJvFz>JvYc5T_x+=WvrdQPSoqX*7^6cJMrc2$_u
zrz0Ob&PckEM49phn=^{z)92kWZ+~%F*Tr7$?Ba<n3eTq}orC@;%FpJ=>z%~^2_OS9
zVswCqAB~wi{>)R4^T#v-T+X_g1QGoP5izw3horCvmd=xWyISCPUzN0J^#acV>IEWt
z_<-_1#_RCdNAUp4{{q-=Gc$R5v4JX*W8cs%nX@H%<#Wl3iuUOO=Rcjb!Bo55bSCn_
zA!40QnuF$AmG<X~syA;Q)4n3$_m@yzcHg96U-#W|$(D=hwPEowk<fp`>2T9Cu~XOT
z9&M6!?3&&romI3M^4dQ={l9ZCYR&9cu{^>uZ|uAGvC7PBe%|!uFK4cf-ga)MUY3bq
z6h^>|{3ml{oBEwy*rX~KTz?U-lZn!KC!>h0-<Q8lPi5uR1X#s5r$&fWkZsp-9xE((
ztQJ4{*aneYQO!P*hG!8q1@S@bFD5t^WNYf7L8$a5#hV=y<|<0*y*%4*xixb4yQ5{0
zE1_0dqwWb*Uh>}ZaqNwv>-Ro9NMr8Yr(aCe`!l;TCqv>-z~yAxIUsFikj`jio`u9g
z8^%oA3g24O>3J6qo^}p4D<4}X_*{jMGfbqtx6q%zB(qxZCLJ@#;AuBI_zW5Uvk9m;
zj!bjCx^@?5M?;3{o>iOQOe1|PI9B32_tB2I4$rgREx#!YO7by)nIBg0qhW;v_l3;$
zvAVM^xeDg|abWe@<Tw(+rW29_V@`3^eIhrJBhEJMd_g_uE*)*BsaN0^Wqv)9vq7x9
z?u~%fgE_5Q{X!B$xF4owk)UrmXHC6_M*`V`?s?&X`jx2;0?!}$T1^*gJ+Q5JM~ak^
z@Nbh{c?!3`=nDTuxF)-Y-g|~)<XK6%pf&G~9WDN$uUpy_>njff7OBpEvRBkc%k;0<
zb<=qFsr38`Mr})JmZyAbq_kL5$hu`Ua&nuNyv|-RMax@LX3*&eMB33r&*<2ccrc)Q
z?$A1_Fs6~St-i=FA~NHoM?7<-ZdvbwgDdt}*cKb9on3@Udm*_WgE=sWF%=qv-JFf-
zLp7h8!<FYRIwm?*``X|`gl}Pu&+oj&XH!MhZnUYdQ_W&cy7eh8{?c2~8xxP^cN~ik
zx-a}l%Ue|Mw9?Nf%FiBQPJs}{M_aZd{7Huy`Aj<g)T!g$OG}OJnHMUCta-CGZQjz{
zeuFg&AO0m<`jt451LDLTU;h@ng_>IPX^&YuJ!ggE_JpUu8|N(9v|7ts?<AG)UiqWC
z7H3y%gGX-B(WCV^(Kk!uch@ARoj7!8jbg}k$O@z!C!Hft8B+H1NBmGZ<G5!I?-QJF
zo~hOB<~NU)5`6<rb$<8*Af>TI)%zlEos6sOpji=u*5Ti*Sn8{B!}(Og?5gvx>RT0$
z>d3kD<{)PoG-n+$Hhs0}swFF9%&kEQ^qe?I017cjBk;bfGXSEn<F87j6PC{N)|KTp
ztH?a4Jo@ULh&f#{0h{ME$%Z^L`GCD4PTWlI{(6yotq)Dlj5{b#j`d&$y!A6&`0sYy
zC;P9hh;iX%W@tl^pbg$0&4I-m>yat+^ci~<dAj!TnI6o$AEJ+~TC?MloB#imPc5^F
zoJp4p3XcEKo+(xmpP*_LV6lGE+2wi19&RnaE}?U2*;YZp0J0gs=_Ga^pO)rke}CHT
zmG_O0xn@Ig$E7io@n0eM$H@;U@5f?Y=hHSmr+-R4tG4U-4ccy{Cx%l@Yl&MVtj{mY
z-*?7)y~2dFpH}zRkZ2s-{FgwtJ@Rr*{k}WMOEoRr9pr#CTFOxEVtT%9?XGGWF+br{
zmEUDO=TI8uhxz$0FX#An&O)Y6#G*1kFRgVgGh%TcP18;tMvAm(5L^6&-XqSNc>bXg
z8yteZmVB}5Htu{>WaEdr)qY)i@%8qxb1zK4w&W}}G;w8tnm`9K8vK2D2B{oXH8%p*
zf~l)V-u%eg%}xK&xr7GWsmn@oiTfoLLv*ASTmHw!@vxn?JbfeO;J$kuwUDN8P)lB3
zsk`p%3Vp}GfcQ<XHwd1&6eT+8B<@eF&NvA5FTvZvksgQS8!DRlFJD$D+a&DYb!V4%
z!OTT3@7g^R_QzmEF){p>nUq<)tS9do&<gu^y)$U0c>YyPxjtbAzUdwCOI3I@!;=52
z_y-_Ln8b&tg1aGV(>ARe?gj8h93Od42xmZCPgZtf4?`vXU6~RL;Ju@tF`)Hf4Y`dU
zgyCZB?ll?v;vQ$qH+mk}@>~7jnmC!3yADq2lqb5XuSE|SfskhfGp=bZ&I`qqqbjaJ
zQ`JS*<`m8eU2p5N_Cm}i4B@wj^!F2v<pfj+fpqg?rBe60`e5enOtHt$-p6Z~s)UDn
zJTl6}#M~I((wW>lyA>uZneWb$lz}GulPrp;<)4DLN55u^qf#zJrT9)>v;58jg?l=A
zgs%d-@%A6!lST{Dol)U4yy_2n5t@AhVI7H}x^~>kzZF~a>WM{Vp4x?%PwpfMQYL=e
z!TraMwBPdAzq}`=LHCK>)M}?srl8k)Nn$EuIYCr05G>&Kf#l%;rRHI*eXK2GLB#6{
z<zpOinI+cp&m@k0Egm10Oc&;V&k2FFU>Z`mkdEj2HbaSuHKn35!R4T}`)$>|K@B^b
zr_Nd}RCdTqi@<;&J=B4@8|3zxz}nqCN_<hO5VYEA*B#^YUXx5_V?@0Kig&dAQXs`Q
z96ydzvWt9h{b6-t$cG(p%J;kDhJzm5yA@k-%xQbYf@f<M>iEwIrM2Xrikil26!BFf
z_3NvrXdA9S+!pClIt4B2cPkJ3KL}h)kRmxsCQS><bj+)}Y1M1BZgt!hUIEcR{6{RL
zrs{JdtaF3eLXdrq5hGTZ%qq?~Dz1A}=8Y`f#Oom5*hX-GfKD|4(a2k9{$1b!_s&RR
zS($Vvs2pzUKisfs#?9`)4FRlr6aO!A>F-TJ$9q-gO$vRO6k&8KntfUyz2(Te{c77v
zkLtK}<?%Z<T@!xi{|9ExY{di4I#T1EV)U{6J{;#_Wi0nJL|#!c90EK~ZR8cw=*M-l
z#k1tzdN$9PlHDxx;LG#~3`;is)R*=~Z%fm?Ba0?{>G?Ize8c&W=?1DJOCY_6bV*^+
zE3{s@9ZyS>*$b_or1($voEj)$0QdX)zrNr8U{n}Hev0^)IL2}%^l~+^<CE!&$4@fW
z%XNvp^FKDxe~w_MkB<B#_5!3ZiKEJ^=Ym@(Q6}l`nD4(xVb{L(b8LQl7>yD5AVB!F
z8~wMUw+Z~OR3UA-&zRSH;5=Q6<3Z<r?=-jRA>4Yq3!^;c{GK`F-{$ug&c?4=`QB^U
ze>FMa{{{M5yCDy?@ElL_59b?vQJW4$D+l2atjME~J>*W7;mSlzS+?J1nZj->&xiFZ
z$oI8&I3zg0+;Et;N#WgzJLja=(HBiP&A;s6cl_sZ+LC`?Fq9=sTA!6Ja{;|{H#3+r
zFH&X2Z}U@4g=tbVf7^6tyK~?l8@Gy<zuDWEVM<*A8RPqA=Uy?58gST`pTxJ*F5kp9
z_9uM8;0_nwJl*u(YYCHJz@>VwaNUmeXIQQ)b(e{V-#=hjqIdJ)bWsicmX*57i!L7_
z{$VRRMagOXAhs)py@Bb|*d2RbOQcVEs|Bu6=uC^`i@VC_4A4XCS1eK#d_LvK-NYlW
zPdpRsl+%1!wYjwW_!~vBs*kys@}s>T$ei286d+x<oNd(rO{7}=RhA+rA#tR)a&Emi
zW9=z}f(NcAJq)6-Wiq42<QLP#`#(-IH^x)dvsjhAk!fD9)WuGyY$sM;y+TVBahW(X
zr{&Noyla2X%DG~?`6Y#Wl?NTEQbUNeyTvnZy60S1p9Y+dzJE%m;mVj@301QXcU8>t
zUcBt!iRIqcgY>4gENtY=JaSjC2HyahJnx$bRBbg-*#k~6fmqy`^lGzeZMpXieM=u4
zjn}uIP+K@#K)!k7#Q3O5vWCq80tNftQ&*|iRbFHya>*%1`0d=g?jo@OiHvU)+IP>s
zMNd06OXJr31DH()64UA%nCdKIRA&U?!k{5mtWCqf?Ow+beP;(P>&1%O?;qGVG}%l2
z{<jSsMz3(*`z0}4Zcz9^?U)DE`nI-qfr?MIG_G#alRk7}{{*i!iZ_K*EnWJo8iM`M
z<xHXNOy|Z2))zOIn|*laa8ohHAyMI(uymN$bSvR%>xoT5u&c%j#1>ao>0q80p>zgs
z-Mw{F*WR+Z7W%t`mOrfzuVT1r<>Avh%(hI(6(KKLK(*}bS7v--+SZfK$zz%N4R}A6
zmZtV~(af&xa}76rvU>AIVP^lDo_7ZMlaD|}RgqW+_O7ej&hqGnt_16wM<cB^<OIL;
z!fUyJEFM}cL-i3kKz=@RA6vcSYNQ0cz=!mzY^c4#&@aN{k8@U{k40j_&hDEvbH?*a
z6C`g$arW{G8mzGwwdI-4>`fh4YgOr#fKTQY8kXK_XA%;pEpZ69Q^LIH6`d;JqkJho
zxtF<s>y7ocRHVg>^s3`#yi>f?|3{L!d01iZz*=%X1*8jGkGDB`Vr<PKmmiJoWKd*-
z(*?7S1PBbS5$c$P@eJ6uhk9nnV2C=YO{ToCMja*F9%(I)xv+=)2My{Y#<7@Rj-4Xp
zQ3DYXbc(A>x_%$l)NslNjmwVIb>*cLy6%*bYlDRi#5IQmBZlngs&hQw;)S&rxEq(E
z%B}-j9h|nkkKdQ{H2K-#k8^!(56x7Yi(P)<iuatyd$HkK#X^PTq~UILu%wkjO+`WE
zJz-tx8j?3d+bd0HB|N-frMI80yYlhPB&CT-S%Mgknm+ry?t1LGT9v-!Ei)ako72<I
zUz)N~Z<?BwkWV;fFN>B%ezHOlN!2gd)MWfHX*Fg#W>W4h3^4_UMhBV^N-_5JNyLP#
zeVahA-Y>j(|HIOykHv%XS6>tF)xj-yUs8_LN0X#DlVUh}^r9d&a}QPG%xjYU6%~u0
z<|)54T=L-==J8Bz$Bo$2Ad6#o7US(b6mre3)Fo?NIG*67?`HAz@(2ByntMk4PcApy
zC0cXCZ%2{iP=D}Cia1(hN}lSjY)_ACs11%`I_sCJL_ZDQc}2EHcfb39gl)rz$4@PG
zKAbUo-?hDDz|vrgE8{s1nH5@sO}C^_G17T?*`lh$qMr;*a0h~C77pAFq|40?NxR~^
zcPmD_;JbzW{z|z-wZ~Q}nKRX6ZD4Mnj4g<F1;KRc{L;;-mmD|szxE~WNi?uF379!6
zNik%G&rr7~iSs+29s=!XGUeNe1%Es(IezH!j6u<>!>Ph`7N53y_{Crh8*gDzVL^Iy
z&I)(a7kxdqngnC?85sm!^G#tQ=~l<%q<2}VB&}EZbmg#M>d`FxQO*oz${o%TrggOY
zrsBp`NgbPB+8^56?Q~Rp*42dy(mB{G^^V(R1y@scJoA>lubA?Dz2c6=`liy;<9}N(
zf0o!n;D}3hxREBh>(Cz@t6-VFkonC`dcRm|YqLLwyY~Cd{@E)99NGEL7)iWc@OfN{
z)J5&-o<)`lN$(H4t6EL$=RU2ixl{94T(#G-*1D!;>nv5m6&A5$lBYaJ1U8kLbl=^c
znbU}P%TM(&FXtFMT&aEgVUlktAz4a%#qH^RPA5%`qO^6kbnUx|@z9NBs-nWX!Wl*P
znZr30yT`Y$o65Rh%Mr;!a?_7(bo1xVqx|s(TiMi<n(ZXlE<b~t;J)DCJsVc%`Gk~&
z-^hza>!q8u4<x;~S8uVgKyTz}Dh9n4efR$3joBqxbA2wYZ(AQF^p6}nQKG5$eCAZ9
zMMEKHcT=~ei0qcs`ofjRx&zJzPPWuv-C!)K9ac&6b>G8{)RuKW3c}tvj{egd+x%YE
z*{@>k-yxU&_GEL<%1JqoNQ!$WW14%Wua3f=_Z0E`AioSYM6~>1qdEPFttDm^G0w?q
zev}(!q0+(U*Xl3*k#o+pGD@JZn$ThK_HD+k<kawEOMlY|R19jhyR+!cawRWJa|9<N
zrSv!#2X*zawq{20rlzKm@?yFZ7bWG*u|JS)BBmpZsmClQ*`TGYl>7_qwH&=Nl|2QE
zN>r^(<<>{6l+>rb&5ptbYOU>C%ecc$bl0QU)lJk|??4}DsJ(cs;sh<!Tya?A{UVXj
z3^^Z%|Hs~&$5YkqkK-H1EMt`MD5OGC#>g>bN-|}LILYkbn1>8Ugfdf6QWQ}cPUe&x
zvna`sG4qs=b5y6p8Tu~W=XvgP@AtX)e!suJ?&tOW?Xu5%ul3$*ulL$(ui-uJyJnGe
zB{pk~wPIstql+sQ&ef!(OID2@*tGxrBB7e^g2(0WRYm5LhfCpdt&(1$-mb8x&;4r>
zILjS=tT8^aU&*h+#x2I{bVqN;w-;gf?v7>f0o7bRczhUwvine4;{3*yXvHe(<d=Kp
zJ^Wo30w|+%VkE}1n#D0-S^aNjJDCOKxDBxp&)jTsWYwVN{z-J_W>>PwVZO#LiS*1b
zK;gPnJ0s-AtFFntQi&s0n`a-8k4H+8Fzy72x#)Yv!W;Z3Jj!MhIdwGGP>A4qYhcok
zS4y}ic2BsY;I#q{?5O3OX=>A93ipkUXv7D&%?DxY#>97-<}9fK#~#xkj$xWlJCt7!
z@5Bj4@{!DoKjXOM8FQ`$r`oX}gm<R}NGNnm-cGX$RvRrHM5yq$gF#afw=j`TO~B&5
zv6Ob>W**qlDy2Kp<D!N8zp1Bl*WC*lzw<0cW!pPd!+1AcY#lp`LttD?`&`V4?|5MA
zHG7k9euMLn#DjzhiOs>)wfrk~=+1jGEleZx?@Hav!tSnHJiFgveu=EWzF5I~P=6-W
zeSu;U={lF6g2bDl={<Ac+>r+OuiAIph)HfQp4A2_^}c!`QZ10QIGTYAzj@;E8-7S6
z>K1;_d}2J#gDbJFNdw<nTj}m`WNqQ%!1#r(*#1fpZf%b8D-DLK%;Hdl-apTAkt$@9
z==T`zZ}MWKP+|m$*@v+!opR=WP>v0qqnzew!oHq5o9H1=R=*bW==nVtqPmfQ1;@fg
zk;LIG<dZJf6298&4+zOb{)FBfi|w3FH@%5lit?aG2!F=ab**(dF?X2m39EZE=N>QT
znohpd;xk_#=A0Re*0(i}jn(2PfimolHFSMij@3XGf)Tn^NF}xL<h~ncN8~QOjLGw5
zJ)AJ!?9S0J-c|{-*+-Z@Q$HNd3h;~@%=$lioXGt;*3`St;$#b8MKpPX<z65#%E?)4
zD;oc(;5r!0PWJdZ#?cglP}>DXefZr~&%gME?Xfv}-mW)1XioK_e&*;mEAuSV_iU6{
zk}iRnNToigIceI3;nZ*{S{WCyOCe_JMWIl_u}N7@{_cldtuk%=y|cI;MPH{4(DpUp
zam`$CFXHKq0LA3T`!-wLcqgSZuG01O3^0&vmb%t(715)^2$-S|`%LH+#g`7eV_%A+
zOl@znDzPryxqaK&^U`?I^K>a&-l9bC+M?p|_HBqd;+=W2Ixe+IuD=$|T4u?*zE&9T
z=E=sx!zeqe$`u>!-&Gz}j%3}Dy<x9Y90)>RXKcFCcXiExwb^*0hxPt9KB1MNH}yg@
z5)~HTm%cD8$S{DY6aRYQv(s7rYZgEC1ngXABHnnmxS9ZW)YY(axv1w`#$Rzn@?z$x
zqGjRRDS4u$*)1KB^5o<D4ah-}L1{{CrF=0^Gju^Oh7mA0Vjp<zk(gDv8i`=PX=3cC
z?}te4slN$DRUhzQO+O8l&YqMeQc2yjhMF>CVLwdpUS5^Sz4Cej{x^@RPQq{Za+E#P
zYe6a$sfyow#rz&om9x%1$CX@nUmUN}Ual5|xhBSZ+wOBl1;6yx9oXxF49<mptyv9i
zhsI~P%7RweePm2)2W2x1(C)mtLTU6xUi=0#jM1)PPU(r5nUeblYbvQ9HO2dPzm1DA
z9bI3P7JD>1`Tf9u8dATm6050okuvyF{UORO{P31iy4dM-R#c=hXHLs5+43Gi%X^hy
z<KEt8*tN}$%mb?doE!Qo4#K#iua(aD9BnG`e{}Wv(3eq@h@)T4GLX>z8y}plEMp8M
z-C{IzsJAHR=j26beN!9+_hk8eix0KJ=f&pcv>p^WDJHS9L7$)*n|Rt7PB2<7Y)||2
zE=`fh-9vMOk$Ur}m5c7#d)-W4&L+swf&v&VHnVJxXjwXzkOeT^6%+*ENghIcduT^W
zC*CC>)8NF<_;9>*{3C)}$*h9{Z%^BSgDx-cNZ&lYnOSQgudEVqI^_VnycykRB>w8o
zN+0nFgP*2u)&GPRqoT&}RfBW<i7|>O)1;BQ?et`c>+#f%$5shODwP&jEEH1j8nCEg
zV>E2Y+R-18uytp`NXInv=|%cgYeU+&Vi#Ju|3ZA{_xlccdFY^^cTBUH$O7Ch&HZ7h
zIoy?+^lM@~dA-Fb?(1gCu*_%bE=gLfbn|{zF_JERhAjo0pZ~<G;QIL&;aGlOpkd$1
zlMdk$u)&BcxdzwE%<3r;<3qe}4NfJ@sqO!WS$~p>5u%D#M07Om2Bz?7kxEa(bUUn3
zn4-_`YhS(P$`Df#nMvW7c=}BI2^0&WC<@MO@Oe%t;_s)=t?$>%eRIAh_R9gw3=0=d
z1Wz8%C;^#=VH4Eck7?1@ln8rcgjPR|ugI0nb=|tL&bnKm_+(p3&iTr<g!*;dB}|GV
zd-Y3vVykJIP<kRoG+Z)8*0uJSv3Hipt+aEu!qpZElBI8SD}6_<0Y~DRBza;OK7!=j
zS}wKtVJ$X-Oil^^5l9zfFLGF1azED#R>^GHc81A+xA)!I|H>@)#p7ynt~AL;>JTt(
zazX<jP}PWrpiD<F|F$mt-s0)^uivvQW>iT$?!MS~g3ueq4E85()O7F0@OcvLXruzc
zVY;nP<S{+!jVk4}Z%D<*?!95JLoUdOcAVE2-1}8a2w5LpQs6mVwUOdHa_9Kvy7zVx
znGIi3+tGW^e0LqcFi)~dKm8fw+<V++0O5vC9JZUQn#8z=7pNX?6?_!#v{y6p*!*~-
z!CC5ciqSlh&-1<%;9W>gMe5ZL*ft88#jR!AQ<12=xb)kH_jyCM71~rN>ELYYzId2_
zC;ijM&1R|)-Fr4D>Yr!cf4YX>&c|D%lQws3U!0a`<|8X^r6paZo}sxHy1YX(y4iU~
z4Ii0)#pYo=O~=}3LPA_@x<J{Ni^xoo*t0}NxR8<g8HMiKpX#Efa4G{$_X@b)PLmFV
zJPg7xkiO13K0S3}IiXQoHiw(<YeBYT3}J-cIF$K4RnHC#jJH3}ZOo=$)f8UI#|=|V
zusS~FId>!tFJ&E%UwvDBs&Liu0l}H5iRO*e$E4&lQJjPE%bhBxTQ^Z*jpT?{Ey1uK
z#RH?lw9QJoB;Pke(`+ETz{}n}b7$a!n#!bs<vICeZ%E*PN8??49bE7&cFiRnqch%C
zl63>47wL$mi8Z(02kknpt3UN3xJ<d_g^SLW<D|K||G-5;C?MSOw={8SxbN>wsRxsq
zx=CchkJjmG+|_jr*ZEy4=0o0S-{Edd^?Z`Pk%O4t^wJw{HXRl5`(Ygc`MVJATqGNe
zXk#b=OXG#FbCFD}a@OE3n@znhPTin}=F!|IyS8gr9lHq<vwc@-vSG(9@bFY0&9O&b
zp0&;DE~!I{lBo^jpPKnKUug@%7#<HO`5+tzylZ<81UFO;?Ms_uHCuE^vf@nPdH@rm
zyuy!+QNSBxR3H8hXU>!=&MZFhc|b>H-!0iJCTocpYwJ{DM0-(i#k>(URa2dGtfg@e
z{_Fg3VG-$M^0oD;zPHcouBU%Y(mpniJ#d`+ouN7M46SdvIvvO&+{@{g+S<yGJSy{w
zcNwZRuJApwU}5;Wn<dG1-TZ#TrQRD_aP5EIJ=rY&===(A!;uc!w5G#9r-%<kYF`Zc
z38~@v);a41;sa^j_jygpR@-ro`v|Ku@K}t8DHEnudY&;%sxD9|b!{)-xM{ub_dbpy
zzfWOKX({|M%umB|+J)S&QcgZ0724g@b}E`M301FzWX&FgsA;z1PGV;6*l|8!Zbk5g
z^_isr&OGaZdYO*Q7{Wy7QwLN3rQ_HSCUStsYfQM?DIhS#Ou9knGash#c=dm&Hbx$q
z3!87T3Jt&DYRGuzz6Z?R7qzj$L_)LyUH7^>GkvUM<q4TZI?wD|8`Q1BC2zbr&U^m$
z=366;TxjJ#4Ma)dXO0>j>m0Z}^z2mUQuDh#k{tKrB%@F*O?1)yQ!whontMnqyu+)-
zWRD!3@wo{z!@}pn?~fT(9g?Ek`^7x_oz-(}G&Zg=m@F{WLEt3MVVk)`CnJ?dlq@zP
z(7X3X$Rq|fT+`~VIqrJL*6?M6l^q1cGBYJY2?_c1-%^FXdgs;>>ltiMsomT%H(Gj;
z_DJWsW!jgfhfT{QRLSRD)3$uJmCH+*nzC=DdwoMWRPsews$4?UuV8Ob7|4Pq@OF0j
zq<pbRbSU9?yO3LFW!jwl{&xeXe0tu9{g|w6Frz;Ze)ud4Irl%xjt~k*5R1~pX(AnB
z`_u>GkFII<Nz&j<PFkaU$%E{#X8l;_dhHL(+|{RWcxlKczniQYo?i<?)}&J;RX5i(
z+o(Cjv$%$%8)Sof8+qZ5V;jh`z&khwe|Cf47WEUV7Q0?#--8XDohV(Exyrm`oi+Q#
zQsRBlo49G__am9KiM9LcBr;xl_CVHD@YYZ+@w3uau1vUl3@eLhtnP8RE7pKux@KST
z>Aw1h=zY=eFx<Y;!jq`0ufE5&5R=th++@#d`|$)Av6#1Jh44Ev2jce-&pabp%~5sd
zO41^Y1LynTUZ-D=-y8R~<w+Rk+2?+-!!JJBuh1(V7E4F`mt4S|@!B8Ghgi5v!qQq)
zrnmR4)|Z*`c}(LoNoZ^Qk9$PSy0OuE<m5|YIN|1Gy@cvJ|IB<lh-6P=1^lc7AE${z
zO+bZuKeiy!fov2e+yWP(sN;{f3*_LvXwuiEY8wx+Jfp5;I$Q3$Y4ob-S|^mg!N1I`
zh^sbj&L_EJ5|(E-h|3g7!B_Xct|Q#Y(%Um}?m0_u2GI^5mmgVEAAoq)SjOmnA*p+H
zFEz&X>x+&Xm3U<rgcH|BrEEpeDH*I6;xgE(^0MeQUOf78uh_HxavD}PgsA+iTdqAT
zL&7lASY1tnk&P8ZHM#@GpFzpon4P{Zw4P_mc?DB;Dx0lGCyrNLlPgj(lVK%k6$oRT
zk!kl82Xk6}LVPrh&nIwK-#_d|?aI$l8ak~#A)9)rd%o<xN1vn+Ie!*7=GTzM<62K(
zu2A?cy$H!Heqj3$Rw;GeE!3^s-A%HZFUtf9IrmSZLH@0%nL&74I_Z#ma+#7Fx9<F_
zJ;8CB(5)%;@HE6s$l$^662mw%)KdPq1knYbzs}%ZJoORTH#@!ilL^jbSdPi9UOjHz
zmE-Qv(T})eRrQD4WLcOE$Y)4T+gflyJf8mjm5$X_jI*QtDckwrj*t2*=PoZ9m^+$~
z=}3-W+u+sA2|`p3ORr6b&%%2zUR%C%)!?IsR|Qug+gCNrB~!p%<E}MPhL5w)egBa{
zQ7Squm28s4Rn6Mt$AMi^W)-wnKaV%Gu?R#ys1p#dvx=|FDQ?al&h{@l+94EgIG?KX
z@Sq4(05Lmh34HoDoRgik(pN+ZeD^og*lDis^4DmRRgJho#X^=`by%8FbFSWXTiGoA
z=N$706^7goB=kI$c`oNl5b~O)pxX)G2H$!f*}IXiyJzP!4!Lp()2GmhAJxd6B5);s
zLkf7<(1Zy550NXmdGhE$SR4>ta2(5Td%hrMptJ=fv9t~Zf0IyOcyS~%ssC}9bDPTM
z9nSnfuOlhKO<yH8tmM8t5$0yH&CG>n!5mRdN(9~z@NuW{4|q{0&RJJ^nQ(vXZxI>z
zxci1ES7t*4ZI2}o7v77k#WN6X@b%49T?;}Ku{034!4XEOcP!OoGLK6<b2EHjy~d#>
zYkXYc^Jp{N;+IPsx__~C-$`d)l3p7GBG?ha2(6)eN9b$)gR*<N&E`dZbFW1E@IiYT
z&$<_@o5Vw1f00aI1aYVnml{O};zjuZWy?^)L>EQ~NQ(B+fV}AHhx5@bL{piyi6SG!
zNO<2;)JIIIW9`O$IebYXC2+RfxK-?MwLMkZI==kN#mE@K^JUn@C!$tY7!6_ET8dB~
zbJ*Rpgp#Y@Fnc_CWryv0i}v^*%Zt4s?oUdYN$X0Y%#+Nz34Am*1%hy=opJcMO?*(l
z%b@(*d=JBdT7@qnE-!l1U6gP>WIoCzyu-cetD3x#enUY2Znay|$yx46mc2{HLRmmj
zX?LF#?7YT(Bka>(X0>BKS}@gJano&L{GRehur7Sl(;uDtUf$-QADuO)+s7UC-bk|2
zO<ZjK-!vp0@;Ge_zUhtR#;pOVYpq4oZR~MUH-42SFx|$*O;M%2r}faL>dA=lMl7Od
z>sEbu#Qfyo-jA9!bc={eBqevF4QP*S0zT1csdDOZVisQc5n&Ys&s(U$RT2*5)8&~2
zWITG*^=1CHEJsy!_1=8U!A6<OHm-P^S?rF{X8aVKgrXKw%c=3~x<st>8qIe_=*(8c
z!o-it)UHS2%<If#Y0$?L%8N<UM#B}2!t+sk509e+B;;#_16a)zy(d`J1s%=@4y7Gr
z7@}!@zRZ+~H|PqTI_Tn{5M;k_(@=}kHNNnaAWb@it?(sJ8fpnmIzYTrx&o#rlbY?^
zsy58i=?ivDFiasObHV-bG08-w>w#neiYX~I(Akr$73qv4pcB0hoJUlxmp8r65!PJE
z(mp`9cDC~(>-bFG$n<5-_(GDiUL}4bab4osV9=<=)Hq85@4-SkD=i|!>Z+E|(4U4n
z%+J3FO!fPw@&Et*<&5tSO_TrP_F(w=7g0a{i|hM;(54g!><=?K=Fnp~#1@h(*_w38
zK7B3K&OQVet0w#=Yx#_fXuJ2Z-R#LK=NKAn5L#T*+cN5gA8OVkjH*V0!iPV%na8T%
z4Zm?KOYr826XatVdePj|pD*XHgN+O0g>>C#FLO;{6!EV6nXg;-&^n_A&}Jd^cOJi%
zP<!_&=lR|OuBq*VSXt5p0jWb0#^3Eg2t*nd+rv4=jDjv%)|TX4P2g8m^>lSj;cJ$=
z(E(jfe$n~pW4x81Bpwp&wW-Ec4mpwa^ZiBNp6?{(oml={ltc?`Vxe%7c6<Xe8Vm(v
zQjlVl(C~st&FYug18t_27xV`>j_Y|cM6XC(KW|myFY$6$i9d(U18M<gCLhCtyOh7n
zA-W$?<=*b#BA);?43#ar+CT6rQ!&2Hv+w!!V|KQWuE}S!2gV*`*cN)&WU}K)#uJ4Y
zFPYK2K@glb_6hkgadx<Tp3ynkyxz)_<Kps<03ho+7Lqt_bQra@Rh>US(vFxgp2w_h
z;nC-Bn7kyV&fN<=`|4GAk-7V$@z0ihe&?R_96Co=xA}%#`QsDAHn#3n^(&hQwfTGI
zNeZcPP}^hGcP!sZm9|XCU{Y&Qs)diK&oe}x&YzPY<-RjHjIwm!RhE01T4Ajvq{sL6
zhW_;>pIxzg6FT)#-+?Pw8T?loJJ*cqZj!5}SoEMHW8?G6S9t2UXDF888Jq6iQr6S4
zOuTxxj7}OCe8Y|)W5EJY5to8vZImLI8jROX9T@c42szYuOZJ+^==JYOCa<H0-ZPh&
zBV%#(XKTx&2lMN5+Tdf-ts`ICibzopqb^RI?!SeqwSPHqn&$?8icz(6qMIY@@{g=3
z7N3i0!(%YB)(1ShFCKo}3|*$v;@jEFTA`9B)b+0)DM=H!Q>QN)n}gQtvs{zeG^Vj5
zFqo3)x0(X^rVUxc`h@u0?CU)WmL;E=1uZy_F-AIoIiHWTN{%K;n*x@B<s5IAW6xZ_
zccj`yVtTYdYlmK=J35~`UbOI%;m3Dl8R@rO7`6?b?#ZpVQ_|=^J*DeCkSCt4r_(j6
zAV`rR07*EG^nDIDiY9uH{K;w*<zcrE(b6bo9@`@iC%(8R?Ts-`zJ2v&+Lk-|$e6wQ
z9@Ql!tR|j|!ZRl?O7hWgCZDDljL1iZm$e@`Vm)A6EHLnzAw1AD=*FwI2htITwDaO6
zcfF+}KN!i>!<=*!AQ*HaNKzx~QnT221W%+rBB_Zr?0u$>Qs-uN<{lnFX9KwfW3RQ;
z6Z)r(_?S93A0P}7GbJ`Q*2rhe?Jw`D-Dl14AUi4pw~DWBt4hH9{I%d=ZxpQwHG?m@
zggB}dAiB>+E9T>q<I`I8%9c{?@;@AvIZHn2Yh#z*5KmI?G|5FzA$NUyKuGQ!&rZZA
zHeabJKi1k(*NX|lF89iB=ysmI<L4HfvHZgWvhFe<>(hELKz=xL*R&+MA5%4*revJX
zp&h=fsnjXD)uhRp2R=T`pCp{@$tj_p=^%RZ$r~muu1`HlgpC;ZUPVkg;5Qv2MCc~3
z5owau=kN1|hVNnNl(`l-{pOD7@oB-1h+TiFP54(DhX1W;n4f<Uli`2zyn#&snj~--
zW;Q&!$5D2m%!finwH;JiSES4STE?Gy&U~mQd&zc)v#hmgsko~2i2>cM1YnlK<*Wry
z&97o!Ph?Kq=`lh-(v>3Ob1sl{J^c&!zYbh7j(3pc#>8y(15O&`D`P-o*}z|3ZNo7|
z>JbLpPY<jjc)l}ly5%fomU%l2Mw_Z-FBZ#DXC8;o;r2J3jFcm#<CBT8gyl8ljFiGB
z43CCUk%<#JuFgYrAxepEr4CW+lQF-cH{pY^MeNdGd`OW6?m=-0G*O2a=!6=wk0wKk
zfPGji+8e`{*^nM!RKPihQ$@^_aeG<N+MDDfM7EAjt;h$bZ_N`w-NY>D+}cJsIuNk4
z28lGa$U_7~D|)wpiF$h-+Iv8HI^j%Wf~zRI-r>@iFe$&V0<Ijy%<a`;?3y~hk9_>5
z!e=*Aq&oTR>f1gFyDG6G%9K#%3B1Xr@3_Sl|61Uc@9`6=K)Bv``xEL4L|(yEkE|aj
zIMY~fm3{{6SP)3-&`h~L82twvjf0D)Mq@f0@M6Z&cd`W67a=uM1tAr=mmKE}&}2h^
zqZeraJQP@O5Cf^18mtvl4y0PW;s@dWY;vz_Lg?nxRHbf7q?zB!_AIipeM8R_dXV`c
zriI&m$bJ^vPWCk2hFz{rZa;Y-B9s?6IVcqkuU```v6}(~pKC%~_;_SRaVyeE!+=~d
zzQ%n6#L_lwBVdOxqc%k68$>G-YYB<VCR;t7*DJ8;Ntw|eh2-vjzHWV*WP;);=0fhb
z&}0&&lNy6tJcO)44t9XeMW)?*=5Gu;yS~KefqdpQ+Q~>ZLP=}HBczJs;DWIM=B4PG
zZT5if+Kq<_M3~KD357S5c%>HK*9^ao*hSlanrJY~kz@C=s8QpDcd}%?Lj1|u`-jfI
zTBOIJWwwBReeAU`n~v#fm33_bJB|C~a4x({PNXqbN%Dl_#||lqOqxn<Z^I>pR^zm-
z0xr<jGdIO}hNI;u=ScNrx3F>1k@3ZE+fmoicP#817$;}WeB;>{D=9%TmKT-zak|Ne
znyJa&MMlux-Df2^OX|q!w|2M>RStdL<I)#+W7<2--1LW)f!FrwU)(wb95+*&zlpjJ
zyn9sk0G%F_@%X4^ve?S-3*=fM{_=Qlh(3+CUs{6tGFi1ako58fk_pp-LM)7EfX$Ti
ztsL6jS~e;j^#*QXSQ9uR8oe{SK~>S#f+yzV&yf+rC`5fu>*Suc&<xMZx8CJE)GKtj
z@YQG6#}|)!x@5<Um_FQP+)Pova!Vp<JCwFNog$!1OEyhJY;ygC4x<KfGmqg_(#g^4
zVa7V7T71$t3JVY8uA4xGOq`s&f>yky2J3^xy-l^@XvyQKYXLT~g*0|J&}j|31ZSrW
zII@q_fQ`S$Yq<@V!SY#`=h@<P;rpKFPR`uwn@Z;>*xbud1BJVH=8rPa*o?@!ZAC4F
zoH?#|ieWC_nPEqlUb?Fvuky}JtW@53`Q}VqW-}HChMw&4SNRk^T<7q8@t_;$V&Al_
zgb=b`kFY<ry?#22?y$Y8BtzPVtwq3qE?E}@%Rp7?!$`HVNR67|Ceg_K`0Cr!W(jjB
zZ!HJAmE1YrCsGS7R=fw_u*AUfm;$#=zzv$$#i>^{rF6wY^mC7;WSwQgx-d)M_^wD7
zNMlVGy9fXE4J$4R$(a<Lq|85=48OCjtVN`+N6ZvTpM_VNc9^asz9TNyPk;z<O=Ha9
zo5)lineSUuCwr3mh^i!mqVtjN_>d>1k<x^eAE7Up%1f>-hDoP|)W4`_mnd?L9KG|_
zwSOpU=GbeNRh-B4c<CHaILVDtA^K6X8~LA1klawueS=X{ZR?X$<_8e9igL2b#Qr>^
zuj(3UkNI_>3c4RS3hzN=`e>h_%AL(L?9-@xT0?jtVv8>-I&UO1l98nQkitkhIg7k%
z2TO=B-8ELmbm_sFuP+6qZ$DmHdi#a6EZB(}@%TU3L;P3o-xz-Wg{S%d(Q_~8Odhxi
z#7z2H?UR*UgO0m*$*6m7!}g^Lo^B)iUE^C%Fj>$)myDWY;<DRrZ48~OU!&rY^XWtk
zMGnhi8+Naz9Oz$0Jns%W6>oes_;8Eu(=UR+DMJBQ7k2b%-vw$7#%E&>ZQM~}V5th9
z6K^o4sOo#P#g}9M)YQHY#(H$&k0vszUKa&3;lfw*@HugmLo@PR=Sil(61+IR(DsmV
zg!(}X775tG`YFgI%jl$$WL_4k@Ru0?+llO)Vw6C-Q|Y*F6zYJAo(4+RC+&_k4Bk`f
zL4&`KJn8GBfheBv%n;!dEjz>}f~ksO8v!cizX=O9FH;|2JU8MI)lA7nq{O-2-pP9j
zwf<7&joHI}f#gr~UzMMo)n8<0-aDUXGoK%)VJuDAT|#4{D**e!(*s`$NLKj!dwq91
zJQj1R&n!;tH9HT>u~82cGdg-IuqPAw9Jp%$9Pp#04+R~w`AJQ})db}`=%de*glj%Z
z+&*OTsX4)VyxrAM^#GQlPD0{0W~CEzTJt$5>dv#}l4=<fqpi<MUpLc#Ip_K@hR#iL
zhAxjz3kZtzfXNG^7&WPe+`T5RDQ(wvvL<zRl6Ngb?fZ47OrJCsY&~ij;JBd2E`1I^
zQW4EfI(xqO(wFszW7%%bSw;#jS>Xp_G!kiImSk>VXkiXiew#(XDk<|Z@Vau7){gjg
z&If~MHUswYUf=Uc&cHpd0qxq4s~EfyP_^B0<eY$@-o$)L4+C~Auxup7D@Gj?_Y)G)
z6o`z(f9Olm<U#e5k`1a~MzH!yBzfq3>=Fs6iEuX@&B%K2dhc>YV7hgMOmqG&8a(pI
zz=8Lg!eOYhKrH94?qGx}w(Ia@fA{(EK-H<Ik9r<HxpX|9;aSCS4BK)onuTPEY)Jd+
z(C=L<e9gbg_^wBBP?t&BXTz@AAVo8nU)!!mW5yS<<^`=!V|H_CG1E6OgX#GpraOsZ
zvJF;1+hJ8$oJ<}84Z7L<G}k1_v~&ApG43JKo2+;18*8AC%FWgN=p?P42?whL0~iDF
zI?xJ)8zw4eKLBsNxWwG(QWXDsS#MclQqXyci%JpIJ<r5x`*B)9ui%F7f%k}0m2sxE
zj;E{%vgYX^&~UYe#LNaOiJ^_y#D~{Rb)TgQyuoC;I?>zP>n*)BB=1i-T(Rute5G6R
zzDdx3HdXp(vFv~0HQir1RQ;bkLjFHmcZQ#1Y_7g8fk6STPCh#j`jQvhF=d6L;zz}I
zKn)Ex`E%iZuHy2xrkA{2orBoqErTxw{Yo+NadLNMleh773GxtEP(6B-P2SMe)7>LT
zT;bTxg^p)XpqXocF51Tr?d$3rB(BOPf5z3<eFs}j>8R>4HhEpNH#)%5&&e4);FN2K
zr?abtfexFzzNdGPYXC^nJ$K608SUcArm4yHTOEIn{+xlhPwN}$Lv(ZyWC;F2KW8B=
zy)aKV2r@B&4nq*c0?{RULNEXU6i5jCK@fcb{qHHvECmd|!?0&NnLp>4KN;!jBF!vJ
z^^MNx{VJMy)XC4!n_e7(e0+lfK*hvu&)bPJk3sa1AjA$zWp{E8^wTyoJF`>54(`tH
zKOV&4Up0UR<^DSVAF==GirvLE&>4`x0@BegfzCbv4uc@ZJZHavAP8dK4boLZgZy^j
zojU9e02KuA#U0rFZ!pM%c3=7%eC-#`ISXBYlLuf<XAdVA0G9z+_R^oy-TwsttquSK
zlw;uP>l)zcEWXngW&vn7PjA;>8Gk$dhwMLL{yWgeI~dUX>wtqZM<B}582latv|n{`
zI<*5I0dSeKlb$7jB>_Ah;%R;MH<;!Zq+<c#1K@@44@OyO0~quOdPBDWeJcQia_9p*
zf(?I%b1!?GvjOP>AidT1(pl5r?j3aw)CFw`?FZ>J4_Bj|_7#I5264|IqjLZT<uGUh
zJJub1PzMG#7uQpI0G0)C5(;g+QwLBcLtS8q<xcueeMT>P=<d{WC;hFL(`hpRgTBcy
z=jv^+gAbm?K=BJQ`>hTzF<<YqJI@8rf}L^={6#M)2PPKean2B=gXhA$g96U|b}uZ$
z%~Rj#H~cD(0K=X1oqNaqynppkP&Y>UfMAOq+QG9Jk2nSB82}iR$!O;4Yqis!J8%eO
zP3Hu;LTKm`I6FhW5ET-KEFc3&2mJa$0U*Z>@`SuWiYv%<1u3rJ>USKuU-tz<D3I<6
zh5X731ZT5fn18$TcU*DE9i;jKrO_Q+XGmA<5fltc@_^Prs>dH?ibJ~KDj1}>|BG_|
zR+p9E%0a{Sz>dO>gFkiH8Q4jfDohO$haq5EFiqGgkg5h#|D$X(aL?})?bK%lJmqiY
z_=6|9{=%~cw6_2ngCHkxo&^|aDDaOG{-ISegn|$LJtsB*>FL~;j{j3zd4fApe_+V{
zLyMaJr#5F?VEn}R0i4GeKmJjMIO7E4NAPb9w9VgXLH|Afw^rVHQ#-AU{s+8&eiy%E
zX#+~Vf8FT=TEP?0?D{8~<^EhoS7hVfzd!8*?D>U3j;=^I-wdI@r|z_ZYvNhrStwTa
z@2N*ej;<g5aP+~^qNA$}HyCOeUNJN<ykK|>i8C}av@qZpIvKDGFaN;*cYWxOcMt8<
ze5YUS)O)9Q{Z8SZ+Zyy(P908RP6^IaoN!Jl&NF{tiF5AZRNyq^lm<D%f4sN9_iyn(
zQx4kYJb2pg?eo{q+WA&I|0oH(1G-(H2YCMdiT}_CtU!%Dp-@m}0BG->{^kpX1KdA<
zdB0Hv6N4$hjQ**2tM2sq-^*swV>-p84T&=yW>RBPU^=}6|NbpANrOB!aMt^C-9Z2N
z1LGek1s)i~cIFcZ1Zxcs_%+jmiC#rq8%%w!;zqvCM`XpFyuHPLWd(`{xCXiggt)pK
zfp+H2o%<l@tHrPRkWQfQ?`c6M5Onej82jG;J<To{g31CRXpiFG(++~Ml0P4Uo+~>C
z2Za2sJTOuBLQtUF@ngTP={R)2_%-$O=T|U}F=s&#HS_0B^247$sfD2Yj}X-C{j(Y3
zVSznkW?`V)57G0`G4Rm+Y=^*9L&r$>>-eo;h>o5C#>m9X!pg=D5Nf$0dO8LMdKd#E
zBMb(xqd-0c<6-38r=ZQmXYRzj-+%Y9E13^iBy=iU_$~VJlE*Iv#Imvp?Aa?QBy~Xg
z;Gx5cO3Es#YA1A0>FFC78lAR0XJu^z$a8jab#wRd^a>0L4hanlzZ`cpJ|Xei^`xvD
zH*aO<+`g0hu(0S+amnK+PphhHYU}EsH@s+VYwzgndfnYKFgWyKcw}^Je0FYrVR32s
z)5>SU=GSlEi9bkN+dFj8K@5K*3!MK(*+0?61L&fM!5Co7J9N>}hwc#017qB$z{IO<
z&g|sRxBu7`mfbp;4=P(&C4j;!|D}L_HUUY+St-H}X}>7@KO-#m{}W|@5cYSvCctDs
z|Lb6&r)OYfU|?WmVgv^h3)9ZQ!p8FJVEgUh_;qmZ99(}KKS3dMpbQuc#ti;*v$L{u
z|6d<JC&4n(|K~Ww!9WK(69W%~fd2PL+W(L4E&eBS0sj>TkRksW0UaM~Us<wpx%bv$
z|GWLbOha<(GUaJIVd@1NjlWTv0Po0WiI%scoW?87R*q;Op1k@{lXpwXAx`2-YX9q^
zLae@p_9}9$WNnhL(Tb@lY2C#&hr4>)K&%j+6YgcJm~GP!N!Me7oNr0)dTrI*#s>Yc
zYkfg5P4KRzD{}NA`XLpp1U4_|S7njLl?5iVUw4TR%g1RKaXwsI#+*P%=Cr|#`J@~~
zN7d~AS58rox2N&vOvll4`D>H-oVEdq^vlFELya~VrIh+U?(B%L!o+ap+0CT2u#Y{X
z@X%ceaEs`!6T_8rrep52(YQI`mG37_&E9)2H@s-bEY9_GF?qLXxKMqZFUzRAfGfO-
zfJmr$^I<Sp&Te2}ow%OhIiIHceAch5Iq!a6p#3e<EZK?Hf)Li!pbbZhk|t+LdxhX!
zE9Rc8haDYE_`8)=cFDdp7Y|mm+@#D)SI7`G|KR5R&edaUs{#+Hm0228T(=weOi=Mn
zI5Z!wKP7yy%&;q3wY-N0cvv&{5Sv-)f|yrx80z5yt`+bl7O?#Y4XG0ta73QTCgPvC
zXX&#f6+UrOg(7lG9<AmcXb^t$&`V%D_LR;``<|?L={(-6Gb33*qQqIdk&L0!2*>xe
zZon?rUk|#QEMwgelEx!j8h@mxLT5XpWCbaWX;A?jx#{cERp?j^C!i}kBIMIC{OWN_
zuy4@V!abOO+v2??!-wL_1rT8>24T3P#rao%htmq5th~@GMS6|+pSOPkz0pV(q7Q?r
zxz*NfO6R91N@ozM7*DRT2^+wjq@h%m?ykNRoJVar?p3xwTczp1tIy1ko>LA*xtDn5
zx>b712PHp8mv<kQjMck!>_lph;p@`xg!DB*JVO_}+H`Hp9shD}_ps%vz;^ulpwd;W
zqL@Ql!kOb;g726Sa#8pGW)SE!O3H7uySM83!nfQrG~{l*E(f7oeB5u$>XJT}<L_;r
zi`mRX0tw0p7w}rNVrRzu4Rk}Z9eqsL9u@0h>rF?;(;3=qIpLBQA1A3v-Nr*-Ow0#r
z*SQ>wB>R<wqqO&ntJU9;Vl&k`&6h$R>C3MRrQ{-kh>y%)i>4m9Oz8p4KHyoR8$o&m
zI=hKFhzw(_1#hVNU3<=EHhq19Y)3)im)GNGFk>UFc6~&ytM@^yy024F*Y;LyM2`gB
zoy(2P=DFxIdRLcT_X~`U{2PtG+u&HeYN{HPb7>m#m8h+|M!(rJ%cHF-=qAn?wp4D@
z1{~#D$Zs7pO+tQO_5_Nh!{6w>r&DC%beQ*ViZL@T2#SMRBxc6I=gHhjrR!_*`=ozD
zI@kp=Q32E)gF@Cx$4BbU-IMPOq3t#0n(A`vH*xu)Zz&n=He+o|NW0yFtrn)Sgw9b!
zaRT0~KL%XkWT2Dr{zLg>`8SzB;CoUB#FwAH_&NHjzqQPkF=>J<NCUjfXN*Pgy^ZYi
z()qge_YW=)eLGZA7vU*d*C{VI!FDP0nwC)}>pt6EJ;FaEXslc0^B<#!;fxdZgu{Bw
zDUnL5&xU#VSw-I3M5-9Z+$%k{EYr*@6Hf;o9P%3rk}e_AboRj$;27$j-Fq+fx~RK-
z>F+-3`!9Z(%nMAhUW`^nMuX4#2x8_zGTG7~yb#kWqV&8LesI!+xFIE%&8vZxa9eLH
zxswu#9o_!Rj{jF5!=KX*3C700^<$R2KUAAOC%?7lJ7W9$-JcN9u<ey^pAs373R85j
z9yH5RXvm%lTXc22vyWBWt!{M-2xMa=K&Npdff4c|3+`LF%=VeTxfy!4p<y~(eqt-=
zm9BM7ZDZ8Si`iiVmbm!V<4kJnT}lnew>Zjf37Uob%*OV{ESh@I4x;7$xlaF6;eWxz
zW!TzBjiIrH=@R2fTOhcZL8^0aTJ`L`3kD}19~-~$XuVX|UjUOP-SW-p7D9$QN|<<(
z)ISg6y<ykQD<U~b?>hThw;#=UZa2+ECWm7d2hSy{QU@jIyHZ-W#H$VpWAKLQTPNlD
zFr&Cq%!*HW^z|L30G#tQ$svCiZZ?R-+x3=Ee`S5LLgIShfd@s!j)r(OPq}VSXVKGl
zcGJHtfTF9rBm+9D+JPqsS64Ndm^KUumTn^_bZHiCz~*5zqGft}*S#q4MZoS-pW$ZR
zNy0S2`L2-(iscO1Z~V+aeP2XTS3J+c_pFilE=wK@y3wU3ergS3QT7C!n%D?*oD0T)
zIsqAZ4Ofe%@Jn}`K(~iC!Dz2vh(t>hI8+XFI+zzk^vlg<m2>8!1gR@JM_qk-4AspZ
zb7Bk-K3vy@@>@;Eoab!P70>5RSX2(U#16ZthWW}n`0T%B`c&gV;&W^>jwIFBnLlPp
zjQkMkMBtmRzc%0BXc_R#)>F7l9xEiHBlzW>(D0OtrA#;~OV^17E+Y~X(Yuj?nW?+a
zH<+IG9VL8sT~o^QQMJ7s(;}rt_?eH7&ALX~YX-t<5*a}RdKa<jD~Ose#nA#3<G})a
zU%TP{@O`%I>?ck33FdU4XAN7@Vpr6=dzfA3;xQ~3OxyZ#PKfP8`CIzKh%Gi8GXgc8
z`V(Rnr&ibGkINJ7h7$_C;HJaAQ^NLlxs_1ix6j^p7&6N`kf!L#Vo~7|xsYAJwQ6Ad
z;aeZqipmROeXwZuz2X+B-rn{y0ad4X<8Fz{nZu7I?>F0gB@={cY^WVi18}}2k~5h*
z%(JjIHE&|_uH?%h--qc5_w}An`uMO)rz;o<I=o{Y)dZ2FVrwGLHsaIQf|arp71C*Z
zuPoY(+%0QqJu6acx4Yc!{e?L{FNoCw^BR8qT^$fet<;sk%YIP_3!ieBbU53L-juD7
zs;R#Q|018gNw+E?cDtFmb0oi3n!+znG|igm%&*Ch?@H3xJJ4D07<3}!^|0l8hZXTH
zPSc)ZSJ!utWoEt!H5VuoAa~<D<R>D|kX(yC+LyG+1%DTm^I5Di+$c=grHQ&dU<3?R
zrL^|OYPYnAFX6nuQ<)72XA>WXq1u%*r&lhQH`di#Z>IM5a`c$@C;IgmDp6k2+QsM+
zXOR^CBNMabrH$EBdp|T<mS}B8{fHg!A*$Z1Ogj%bgBZ!5LW#N5!bUFgk;ubMqzChU
zKV*c<Yg1EX1yHNqbPCP!&mH;ij4pf0D6`U;0t1Rf912Z3Y`X3?5-mZ^YgAxPq#WPf
z%WI7CxwLO9>>yex*1~fS0&#P997W<$bVKw9MCEcDdVIs^Juq_KuBa?-n;koH0JQP*
z1a6n-EM)^_@84o;4(z!SV?SK|4U;lmrLy*9#({6<dtZALyFooZDlde7!1$9?6BAEj
z_)Od}R+q?^l(-{C9Z<9yX@Us4SuB>;lQm9CoKMY}sFA*D@Zu+AI#}T?czsrE<1~x2
zJB@qbiN=f{iijoZP>a98$8j^0Iqj2&-+I~*Or+9EoR-3IuCK1j94nm>Fwj5GoxwQ_
z{N7P$)GDy#D@0cQNT)G`&CePnDBZ>F@p6<Hbg!)?AGvqTz9Tesc+*H~u35S3aH+4>
zg7ab44u2e8cWeO-qgdBTZux&cVSi<GPlBK6e2p3NYb4&XkC!%bxelkCL+Cn{e3lgI
zJm+8)DHD)6QF_6?^hNXehnYocz#GtsC*PNo$ig(%A7CIn2pgar#DBrf-ri3Vl0D@a
zy}z1Qn6WlvLX)6yS592}$MHT1!BVh>u<zTFLhiz>DFp*pgCDjJk%9h9_X7%;rz!)$
z8bdD$0z}myeC5-TFa&C27exp!x_DJnXlSl~%H@9Vftz<H64EXhp4C6+zk<{Qqo`4|
z8U;hmixwnB%%!*3zceh97!qo;P5a#M`|OBia7{s_Fu#MK@X!mJMJ#1^J<a0K3~qaN
zwvQ@%nHGcdKwr(pbfIG%kC99s`!KZIMLuiZ=;;W_Z>+ru)^lnf;P?Vg38EV53ka@?
zSzDi#Pg3cH6FWZcLOm_g^(ubg(Rt{aW5{*3>sApm<xvcCXdX{7F8oyK+%)%Dxu919
z0}&HXObuu1Qo1GJ51%NJ0HH8f(i31n)QJEhYPa;=UuqiMPP}^5h&z<D*&km{v88Px
z56%SfRBlJb0dcr535@A%eZ-47&D%jVYb-%%J*b_{Z?2QWG;Dr|n^%CC9i#Dv#<~Lb
z7X!fOMQ6GIB5AR*mV}vo<9S4gMYiAde8r-StFPO;vpg=BCQWnt?8wZ-_te_PU1T^(
zb9lqNjy6f~trPLI(=n7d!bi7!58tV3mgny2j=p8U(Y=Jj8zW*7_{SJmlRb7rK9q7)
z*QRL5!m~E^@S9Ru?iJJfthLkyg>d{hFWfapR#G9T_5Pt{()1HyL?TUSWc^#jK)2!>
zfjdRd?5|nLeOPQCC@Wd8(9XE84n?WWqSN7c8{p^aY#U90xz*&T$COaRM`G*Lo72^`
znKqdVn&mG%Z+_zty_xs5rQ*iy24!~_@ecJqO>!bqf!Yvxa2=)DCbxP1QSk$qg$R4R
zeWbmE4j1gy(=hMjM@KPQK0xg^|CvdAnX)p1XJBf4Qb6@#`l%27JlsxJ=gtEj8Sq+*
z1*+RKh55CiYc^!#=)UR|O(sGXO6<gYc{!}2%L(fev;;cy(u->*C(Hbw)^NPD%42dt
z&arF38@CVC7lq~MR}wg{gbO?<_5yD2P=V~sNgi%ygx4vmhkLY{2Wn+DyJ)P81U^eb
z{aUUD;_2H#M4VD-?-BZC>9vYXhtA_8Y=bXiE;9mlCw2%#ciMuHRLFWOn5vQWDB$$x
zUE!ndvR)sRYvtZ+^G<tnzs!BQ)}AG@mU1!sB5QLqESJ`_x@AU9p>cmg>;)sWT}R^e
z^^T>S*>Y$55!Wj*F1OYyMzuu5*jN`*Z;S2SRT8HKH|`?X)7VuArpEByk^B$4ACRI~
zte1v+*yZ+%zJK0xduAwf%J7k&_N6fx|97kxH5oKeNwgecr2P$Yj7+@O|FA#OGCaUr
z_WRxksAK4%b%C5ZEUm)kqSbljRFdxAcb};h*kySdyEcB>6p!qXPs9i}9uYPj)-c%B
z`)c96XyW<_V=dy~7m=-(2PR&g+>bmBQJbhGjhH$4cue(jI!z+XXg>Ws{!3<HxW_<2
zmV%=3E!o5bu@9SjD10i!LQ*B!oW$38`6twoeAoEyD?W?Z;n8t>xvFami4neRXU|5A
z*b@r7LXo=>Gr1>zwA*j)TE4EqAWH3o7}oj5C3+x{i)y~NHccr!Tr?w_b}V_99oV}m
zLB7!=KcV{i4*9WpVlTeFZSQv(8F(V6qJ8<rZWrn2#_Gwj^5;cdH}_fA9F!Ex8%YB$
zevp*iNMhO0MlE`bq4{fAXp8b!4ds!9@!rL}BidhmzsrBSMD5W*gl0UtJ)8axiMQ(I
zp_q^ZR~&^?DRm#fDDhtTcuk*TW?Z_@2dWCmv~`^83Ps^jn0o85XUU3?T#a>hHX#PA
zC(N!Da5;5<30I_}tau>t<{J|cL?=?F&&<a7ppK7MZ;w*PtHsn;sT*U-KF7~ru#@#O
zNfF@{wT*pwa}!Yr=OCpU5E2u<VFgGbvuMR~uVMBHqnt0@jIFE^{<O_!N~Evg+pWPb
z-3i3Qo12bgF`h`vpEGyf{akoh_>KB?7p=>)x#PzSo?$C-z`pCNBdvX!-M3;v*7;dO
za-i6C#s$LTZALTZX9Zl+RG=g@Q`p2uy@f{10?&~q=^Y5};+QkUO+)KG6C+4LX-xOk
zy_S#9o>*9Nyo;&Wh!4VDpT^rEl1|8KYU0y!j@&Bj75b>nHu|Y9!FKwVOPm?gY>P-r
zTxRU?4&Z2s%=}3iS6ffn&1FTpMPO1Mx>S9`oPg3+d6y~~%G2d3>`vMjUU>MGf>4F>
zg3b4nTJSh{^#+CCa?Sv#EONym>MCM#<pwUu6}_mM5Efi>^JY{T?pja~S$M;3R+=#W
zIRFthmwjl?XvSYM0P5)pC@aa(7U9;52K(9;&uC5(#*$<?i%t`$Ydwk49hx8TmC|Cq
z>l7|hbui<*=DGfgK2rs%UKIFnU>``yPa}at0}AV7RG?A^EmgELT{0!WPMtOPpxSCX
zy^`7jFdL<e;QJ{2cjpjSG-W83gPQV>>gpeo!j;q=7)+{b?L8}k_Xe}f)EoIiK98S;
zm#t~fhB4MX!Io3p!Sdow*zy_7b=qEh$4CN2<F&8Tv+99wlXJ4X%Hz|l*oB^NFy${S
zB1v1NM0HZ=Ly7^o<@SY9QqrWntfCL5U8q`j{}-F=9;2y`nY8vLmF2j4(besm0<M<)
zG2rBO8!@|wpqWFq81E8mm|c~A@$BWZn6?8?WW6h?sWas30dL1RUrLM%2Nb`p-}jU@
z0LNRWrx8sFDl=Vi?)Z|p156PI<z&4j!+m~4rU+YD-(Se2m_01~evZD$l91DC$`OX6
zv0<a|ox~osAUVs<Ri()a<t|y?qm9o~u27DsJYlGu!<BwJl`5b*v|k*8jzXf)ZaR_=
zc^9U|SW}ubJr9Eey>AtF*yZ6z8p9!%GwKCYc)S!`qSC%hTAA%}b+LFF@6vj#ZVfw)
ztMBXF5T*4Yt9`2p>a&gSwJXj$+Ug2lXq8TpYF`yz6pjtKe^Q2kA-=DTYt6@7VcAeZ
zk|-blz%tV5tsV}h{I~Tw?VOKdcTF`eKfw+mNHV6BJIL#ZHK|9K6$>-w$Cws`=7}(1
z5R*i`C)*?yc{+D0$~Q`Xdn|XH2Nr#SlV4Ns^*gQu&oLAAL<;`LTD)@!$z5kqf*_q9
z{M@ADRbJKZx|+IfZT|qv<$d=~7olRww%KM(;lMGr0p(7`9l!y)XZaJ-m%A+Ai5_L2
zk?$}~eHJ}{;R#N)QW8Ca5^h*`eNpGI?*SMRl57`as#;q$y02@gUAp|9dHFyNP~~Pf
zCZwNE?wsNr<h)XdEVw_;U)!&Uf87>vGu`sQDvIHjte=5e#o;$JHYIW}?aef|oQ4e9
zhUAA*uHqil4tHgYAIjdME2?@?b*?T4eP%t=Xv>H>lPojWH&IPQ;LS6Gamhj=sRFuk
z@hwBQd4_zd&Zk9{5{}c@BFJvEchfZ^aTGbAQadp9c^sEmt9ImUp;BgD9a-{ym9dau
zQeVRYk1~h0;e{I+z<5j-k4gFlY=kTg2-t)pdP|wu^DR-HM?c{&h-WDirEFb7G;F?t
z)g)~2Cv-E3)JJ<BmcFCKh;so0(7of`{$&10nJkO*3HOc#e#!|v2#=x134=-DFL`#%
z*Q4L#;2P{?CCZT@+Tlp8TeP-I)(7QgTkdauzOZGDkJXYLmCW*cDwxiO!uRR?T7MVg
zYL{vsPt$sZI4%oZaXok~S;k%<c7OE6tZA=;!Z-DdS0hxrULQN|Ok?*au$#XiT?ZPG
zrvmsZ2WuK?OSH~P<`P5ebl!Y?IH77e5pj9sOGz^_{xZA;5!b{^Le7$40^^+d9P-V(
zq*@%l_K;|5QnKq`Kx|JuW)8kyO|}F&mZ*84$~aA&!QXuErFjg$^axe>nP9KYe)o=3
z*pQ(Q;9*~@WOEczbZ3lX>LDjmz|xau1m;e8GMnf0_I^8ZHH}RUe?x~<O_mHJC>b(1
zk}NvW+_q<S?@5SKR=z{LF>ZJ&v%cE>F+V>4f4dvnUL~$maT+Jd`)pTeg70fy11~|7
zzI^^@ciDKka=9wA3m0x~y>_>FA~e%9EOnDEzmF<Ew}&E-GcR^)PimuRu_%K+PWxny
z4C<lhY-YmZA+chU;bG>(wlisg$U=-D?xK7Qj$c!`g=|F88&*0u6x2}WyIh<7c`PU)
z(S)_`kaITu8+Mkj2c)3I>#L5*|F$DOW*|KMzwO$6b7*0dqb!NNG*Xu&WN9<3gyy{9
z<tY4Gr+$y6onoC<X$DR)o{mFn&2R~qk0l-2SiehF2S$Tj`1<&0KI4c|FPhlULc@ua
zVg2!mwHrF!vL0?Sn#&odMim(M1)SBwj=@RNXErZHcOcmoZe}}`oXOjk`q;xf0~K&B
zwGpooxYo^|(uT&xY6+$K$)+T%$P1&Ynp5@X+>NxW7%yc$+{%X*o1SAol@bpT+}9iC
z>^B6k>*2GcAoYl_wY0QLX~PJG(OzOqbf4^<W*+KV@cgUj(_kvo!YoQVvI9S4`D|eQ
zb9&@kr@l5c4;b}(Fep>fp^;>Y0>KV5wo0_Ze~dmvaINhlMs<A4Gz!~=SB*fa+ZrCz
zaPbKZICs?P*_G#{+QlD~Jp#t2BqLZrn*e*V$3uK~rdi7T>wQI<cMav&kr5R69ZOKx
zjnU~j%?D(iDbkD2152OTlhHM#Li1{STd9|Rn{PzQcf(G9E^BaiuDu;vvC)x_H_sof
zZ)q7ZBjr0q?007^ZVP<ut9hmTQYM18IRcDpxHZ7GVN@r*iXs%6Q1gF6Uay3Njj0xd
zp!T6%PX^IGO01f;4;$-L&lr-ltRxQG^0IxS&S4j%iLXguTl$80m0Ix)oixqjvmsZ;
zGH~&}{cGN-^BZ#R67Q@G0_%G=Z|-S}eBKP%d|&^YWFQjwZe!dJ!Ujn0@Z=ps^0yt(
zNWZ`2lM@gW9zL3pbUTL3X^&Ysik(QJoy55yQxSEz)}Ii^>Uii{{cO4Eg#mfAM2=a6
zD7}dj@1pbg(*a2fn3UQ?=Kb35)wE#*-e&dIQN0-ikJd)B|I8h2WaO(S$hT(9YnXNF
z9D*YZ1-O!hDoHNaop1vOD(j5)GDxkDd|AImM}8~0uUTZ)=o2ga4ic}MDr{jGEl-NS
zU4FIq;rQb&E=8U7(d7^PBeZR7u9E3kA&t$4?72FAnc|*oQ8O^(_eSNKklf+c_te4t
zEBL1hH#{W*?6W0eN3oyOiB8n}8q#EW3Va}%Kg^0YJp8gke&Y@-0EskCR}M8hdpzcG
zgSxp@fPWU<Jf<#Zy>A9S`V-Qv0=Bwnp0GFesYQd|u07(pBqbHyA9!GI7w3$X-=lk1
z*p8OH*@~FWPsibHrpHwb#WYApg-6n7-#q+kuT-0z;|Z&LGNZFy=_%#da_gFRu*OIN
z341!32pB7|6;ol-oo>eE0th|_Pc&0Ni79>|U?U#$;s(2oT-v6BNZ!bI85p-+i)eWn
zisR_xQ>W;xtn}$0i$Kf}8$^rL6a(LdDZ(><>?bs*f#CQF`C!>J-t)o7k7W1C)*h6&
z@u*c&-RFO?_ugSm_1n5=01-us^iGhfR0RZ;5>NpV6)f};l@0<T9YR7!x`2Ry5~K)9
zi6Bi{=wL;P^cE1QkpwA$6wlAS&OYZ`yPR*Yb?$x6J<t9A@PWq}X6F30F~&RI@s7&7
zJ4xne&OFc`B5&ic9Lshm3AAVP7#6tR%N4<v<^Z!Sle*>a5soa$C0`<QC_w2b0z(lY
z{DO2Dj~iGKTe@GL+YOV7yk7N8XGT2JWv($U1+GvS`VAGVs#O*PKCO<>`>JV4R$k3x
z{p#e3#j%Vm)z!%RRDb@V`u3IxDWe#69s)X>OMxP{ZsO`Jjrkn*jb=b5ln^E2F9Lh7
z=DL$|3~zQrbA3r8#fmCTEkU_p57Ith5Bovm?M|7!-OTp5Cts-H(At_|SrBf0;P^4i
zXAd>5Tg*M;rm?zVH?xNUD)Vag7v$0n?0M82278F&NR-%lFr;ZPN4E5XlFrv&FNIH@
zFP>e$aY6FNHExIcv~Hu@_oJT>UW93D5is~;V2?OcU)^&eV%nxTS(!S9-hVhyH`3_W
z^7hu<A_ML6e#yy8vUmZB`w~aK+<Z%I0VX>iTMbK8{UotdM<@^eo>9taFrk1~D9VxS
z`30%ATfr`uFCUUGuhiGM{q|rT%ATDhyVMDffifdb6MPn6v6%AMn2M2}Nz=gvqlXUl
z#toxKd6O^t37<;&IHSK#e--&?t@RW&_V*XraaEB*gXUXc-D*TTJ=DRc^$%PdALU8<
zGTupl_qJzTz}jtY33C}(H&EXPKZ39%q4FtL2p;bm1~m1YMHb(@X2wYhoDe&J_q|tE
zd2Qr?Kc-tL0ZzC9ose5F?$MZyguhOSwg2j|E->tdcC;=_c+PQHs$=h)we`e?nB04|
zjf@0Ed<L0sZX;smAjz4K+R8#vsR?E3{4z0PZPmTh|KPTIj6rHt7Q1BZZkozm9i8TF
z^PhBNIocO(d9Xn4Q-X=8XxfQDZR_Di(;+`=2XVQE*8XFGl?t{!;~bHk2TdK!%veqy
zXIR_drhwBvq94IE@B=71{uo}(Vau!V4uX!WYAnwu=_(HPmax}Iy5wg%9J}?v!h9LC
z&PVohh1api`q`>j4~c)#^v2c18?9XuuC49jOe^#8XwG8h_wc>jeHfly&tgS_gSkFM
zO-9O<`y2Gg;oK*~BVFhDuRf$}=)7h$LAL*l(W9ba_P|d`pYwbR>*|lNf~A+bn^jvs
zJFlfb|3kn-CRNI1!T5Ckc<X7gR}LQhS}s7hM2iwYLXpo<REJxSQw$niR40WExKBYw
z`2v&~7AM=H!Unb3Z$L~wcFCr}*eg(8?4;w)c;|UrGxl-y_LYKV|0`8LB4$Pw^QU)_
z1EHni9^X$cxxJe6WnxT*Q+P+`MPG&xVL+Iv!=>1}#5L!_nvx(o(<osrPGPt@xjgQk
z|5CnB@(ZjV%o!Uc3MH`<+&i#GG|ze-i`1)5lDP-@Io&Ypt+G(|?u3(ecH11C4m7&4
z9}i3d<4J0&-!tvCogBid`vmK^856Q!#Zzyi8iykFeG`uIq6eD&Cl_SyKo)k$pkA$@
zRg^X~%d$?BqG)ADDW8hi{Cw|7u22o6nTt`<fnLBuSou+BC6hbR3x|qsRiUU8*@5j4
zXp<R4TyT{tkA*Z}>D=`($V<CO^3}@LLut+u3}6Clt<2P7ggpVdug_~Wl68OR0{St0
z<z}ft`&QETo5^nlR&=27zYvfA0^(5(n|qp7FBrRiolqT;rMPy-=cwej^_b1TFQN3M
z3=?766M$;Lu($kI5dZ&v5Rae1l)2^{%m}Gm3%EYEGU2OiGc|RkT|9>WR&$z$D37VG
zWFCZv@`X0R`#SI@jky+g)&63VN2Q`4`Y`;gZrIRX=CVcj@N3<s(Wl#|_Z`7z^(Mu=
zBysJWZyv09nR8gvv25I>x>2NJ$KTR(Q9u~)S2xz#Sj7a8(_;t_dq=#Wc#^t_B1j%&
zm-bQj9_34^aAWtK50hV0B31qI4=d)NA-<EzC#^p$x+fCNd%Q^Uv2CA};c-xot=(Zw
zEjO%7W3mye=;q9Ea(31g`gzIgyUB}?2gWzy4ybl$EBmr28#Oy?zT6*I+Yg<Y)*NLy
zxLq@}ls3YDYx|JsS>gT6rLa@`UCU__2I4Vh2af9NGL@>*?QN6Ukh6B|z?J=gVP7>f
z#9^SWN8>8D2Bb-zxBG5-yL|e|&7$M$m80C5$>bM`Cd$k!MYVDUvRji<7s>`omhGT*
z?zu0k8t``X1>4U4$2bh+yXf9wzFW`pCU}s4`$acuaC1jT1fdj3LC25xU+|b4kMf=h
z@+>ULEO52g9h2P^z<};Sb9JbrwXA$Nr7&&5@xHr}-RCO4;y;h{W{YNUGA70PDfwbj
zHpKIW>E;5^1QYmV(@qs}rV!UZwhIomvBs-0mo_~-JUz|$JgQn2r)g^z*~HCP(Y<0S
z_-WhZY_dCY3R*o<%MQ%ACz|={%Ru9<Bd3dk%hh(=RXHEGxx5(at9^2OvEbe%b9tdf
zS=1%e+%qim7AzW7t<*{KE%O$cjaxr9x=}2~l(TSCFWZQN+L!2!A<Jrf8*Ksk*~7j{
z$*6<PmW2AQlZW$1<)xCP+Z0#_lXiYeqms{_f4zU3a20|C0aH65{Cs<ZxB!SK2jKf=
z8RQ`Hb>z@2JJt~vlaA0Oz6*Kz-N)<2%IlMSe7kx^hrZ<TaI?HSc0hbb-qf1NPCJx*
zh}2JHf$l@}0H?;)hL6J2G|vz?xq}_j7Fr8Vun0u8kErohc3TckZ;RBMs~<b6!nLTW
z6MZ0+AHp@vFoMt^TcFx+!ra|+u}5kBg3K<W0?AHsf}aQ56=GmvGpm!$<J|1%_YzDb
zb1~CNCDL&yU<%0u4qf?zx|ee3nY;e5p|qhwg@<R(>DwX|ish$V<@DJupOpMY|E}S5
zD0vP@qhdf2Ic_7WCWRpo*1PgzIY;DQ0O{q%&tvm({-yn6)rPAdgYFtird<s*#wU^8
zY3;K$GV3eA8#4i5f{2^-3B?)*52#sE)~&jmQkCySKba~2(5$=L=Hbuc4`f-(;Vd*s
z5^R1qT-7U1ln!fabbd77I>9w7GO|+wW17m(?Wb8r&n}LA1|=OgpcM7D&u-6l%<h}i
zZp~6Sp!3otuLAkz4La7wbBwOi@ATeyV|Mm1wIHom8IwQlnUtJ;H#i-xjA}m*Ta-q!
z0lf$sFKEsd{om<v^nVbp{=e~e_Wz_=AH!Y*IH@k8-mL`Y(z>bEpkIK8`+n$JJ7&p<
zcG!zxuwc-+F|p8PCkaGm5W2Vu@lf`gF&dwYFE_oZ3zBTAbI{9lNj;+cQj8&uPNEp4
zS3^f>q0PV`&6Z?#<ls3dUVv~>M5!~UC*N!9dmPHAWc7lQ_ubW#nsJZch~F2K4s7e-
zMA*+md+<kg5~!`MPz0PnQE+8zDht&7<hl2$WmtMp?qRiegJfyDHB=e*S|y%VsI7_M
zCB)*H2xWGp>kBUR7Vah)h8en<Ba7NDp5pxmacy_T5?{VgW8ds0{M<Q1EvE6)hV3{9
z{Ze8@#hPCemJ)p1)he8e?TlZ%^YRK25_;TtDG_O?i}k)!6S{s>2SQ6g%{|ZK!4k99
z?+|b557{e&R>LB+fi!>D`ja#7h6aw=!;?VH6FNS+V@WO2RA2C5SdfV>7;QGspKkZ@
z>@{hvd9<?2vEI?^;mujyD=YG;=8*&BDGv`--gd%jcH-)tP-Yuzn?eKm2ge<Jy=ybV
z=OWk+X}RyUkb22Sk;Cw+fSw?vI<5Eh%v!4m=Ou#P9q6Qzhgf#KR?^q(soj(JA%5n4
zbh-9nT@2eAG?5@OIA{ixp`<e)6TGL^dYhC*0Aw~g0RIUhWJyVd7m5P&F5P}&724Ew
z6X$8jA!m4CQ@?VoPyhCqtarp=!tYqJh&CzMlO+?g1rLwcG<kpXbL^F`(uaD(J~)4H
z5X~);(9&H77^=PT7}tVVmvK@R?JK;F-p1OtH%JuYekb_77X~vBc|q#q1(<UTeOVnR
zPpj$G1i`tmP&r^ck)#lxa%5jHviP}m{dhA^rR(mE(houYUMwRQB)<_Q2OOS~N4)?<
zn@x%?3hdm`_@b%e?khFiED(e$nYoChApJgO-A{ux-pxLr7xhx)*66ZvA_a=;f2hs*
zmBK#1raG)@>px<x(Eu%a^s(5&x%A9gma|L;9?0zte}y@qCd+nMNrf-Lovq051d0OU
z^2_QZ&xsHOT2bV}HC&#>rnlrJoo{8*^I;78MsMHNzp2ms@xwIc5Ko?=aP&pcY3yzw
zgw0S(ZZCwIqUm>ss7_-B(dJRrxGVeq*b36CAD_=3QgD~bSP6R+1Z|w`4v8tlpxVu_
z_tun1rol_m*0`*g(+aOwvmUtqT=s5x%kM2`t1;D=ti5%g1*R`_7z6lvw~gt*iUrV$
z2Vws`zE%juv}$)^NN?b3jLUk3mx9@i{6|d$Hf8UppydFFxg*+<Ux5O42sNCnMLI~c
z<w59`090#rH}l89l%d;@=E+F|@kiH1aF0GDTQJ(-6$JrEM?17cD1bA#<jIim6>d2t
z4^PVlR+bE{OB)$!c%8klC*Px%W1w^YqsT#@=$^w{zaS)8gZ1kRfa!GQi}neGWF2jo
zXdFG4_I~N=Olr$?<f)7+@{78eG76hqKVvnJw|QExlWf6+AEUl*JsM~P$UsDix#Z{b
zSIsYl%i}9O6}+~{?<Atn%b*1!bmhWMTAt^fL+8rtXtz;c{7dL}WXFs8xK)PK)X_>e
zR9<_$#wfwsn*2SJ_Fbc?O#$cP&=hp#fq*3cDZE4-W1u?a=t?*N-Cg`csU~u-+lR*U
zIS`GL!Eb{{5C$j`zv?oL195J!E6dEv;fvvcdUdzNV+tNBp#=sS5qocfybCg2zC4W_
zLbZz`1c)Ix^*E#?NoC%vbAuz$Zqlyyb!%<&SJkp6xtmgvAGjaIImMau&EqU@2|r5I
z_ij75LE$ApNv1GXV=sIehANF2FdhqB+Br`$NZIF;M{QYP0Ks++hz7htrvS~@uwY*l
zOSzEQC`IO#Vu8F#xWcEh3z9eG#pYDwErk^yv|A`zl)?_t00Z6t-#&-sqi9l}A*=~M
zZvd@Twhp{6IX4%}K5GzQ?|STht78wZYO>6o%+xcw5n}@R?7wNWnV@^0o+jTfh>hl1
z;4Y-Vhd`VM{HK%%w1B27Ign(~Bg#e)#bR94zAV$*p6@hjY^+J_^u>$0&L<Vx#y=SM
zWRk`{Z-mg?e?i<8&UuY>(wI$gDard*LU2~nt4$HN2a!h>r@7qA^?s}iEUvU?xX#v9
zEY9Myz=(4b(Cb7TIR<^6u%0<_G3AJ-=?(4iS5qgTU8xKL>sKV!-~|9mHHD3jT9H=B
zT9a$Y!-I;aL9$tVPjMZ84;PZqlwe)(+V%pscZ?Hkby18Zm5*l=L*X>qu_YzpU?QZ+
z1$0jet6Ss2%14#iFJ9EUYE^vf_nq-u?6b}=*OV1YXVUm4(5FgAt|%F^bT#9hu*+qh
zw6g4a)XB*r>D}}E`mNtpQkZWWAd9L=2Z$)`CSa{FQT&Gn*E$4u6Fntc<Ac7}CQhzJ
z^?ONNJ<=;Kd%DZh;tW&?rl^Oq+};C@93!AYMxflFHN$FX9&K*^JEX|M$s?8((%#%w
z6|Xo4N}hCy8w)rc`V{S`VHsvB#xRD(jiO&;s9XT;icA79NV@w6ni8!GPPoz^Hp84d
z&j%W*zlhCmS5?EG9o~L7_Cdo#?^1qaknpyoY#KMOCcQyB4MJ`H1@Xd;cN)}_+=)Io
z(~Wqc%W(0TwOg-SBDPcY$A+gb3d%*<+}0j(`hg#Lt0Wj`y+ixGCP*JHy(Xffz7y0V
zR9{BsH`b!B4Jlki%T3+=wl|s{)hDEGFG(l(hLeu$NETX#Rp(_9W=)2%4rfvk5-;7@
zMwKOVqT+X}x&`YgQ5NqHT@8!wpq<53ti97_L<oW@2UA+o2{eK<c|~eI9psF#^!`3|
zRYE{I=L$sKP(nrczDjr-&zdj9{HF{!;WvoLXE+q;gi>P7r}KA?1U{n}z0l2L_^8-D
zX0!8tQcdd`(@pr0BT5D-M2*L~p`ak&#sMtPp+GMif>Txg^lC6IymWTVW@V?o-hcnG
z6aVd&Y)+}POW$0@J-SX^yI)ithO)?@?4a7cC;`RO?yw`RLY7nIIWcIq;#YR(*{sg|
zxN>sx)YjD9+~j8G!J>xWV?>(~JvhV^uIWg)WVE1h`xx1=C%e>ONjWhn$(@6FeM-1%
zHTG5M4MsC+2FR|aq@Zied$s^xdT&gzLVn?G@}4i%P_t}W-^Kge#gXRFoI&xj$BsAN
z)606?ZDZKL#AzRG1yY8bvnB-3mjb}tULpk?;5)9*Y@?79^h$RrR?_?<jBpVKsBfIu
z_0wSU0qWaTvV+UHrFfVlI$l(iF!Ic&%&jpdI)JsdxJXH>YqJ{hQuumIfc@BU$x$HH
zVStT+b7~79R7sI{h!P#a_0fu6ej%>66U<&WRHa#r%(^;QN1cDqejrTWg#p)3I-yP}
za&Jd%_jjW9&Aw7q&40cFOPh{#f|%Zes>en(+k=KkbZh%R!abm|U5!aGS$=-b0x>H&
zSBJWV^wrp`6*iNvqAvK-k;vfh%E6+xBa~2+)lmDM!*VMqlJxqiDOx9-rD9RQ@ZQGP
z*f(x`Vxx(ucH36bK!BQ?uP@@BvE%w<{9rt8W;D5)w`r4?tMm4K!|y3QdOy&X+|F?B
z<=vA<=$@v+FM@5VO;MWnI*9DWvT3R;h=^~TACdRK@XiB;Wh3h+MsS;RapD=K!x^_b
z#T<?X3qmMbU{+06R|q~kv!mLAq*lM`)?r_&d1sU4-UT<_tW#It<R$9WgfR?1WxK>f
z)f>glQn(oaNsh?Z1?ObP!^nyo;Lt7n9KU;g$fKBd7v)jcDYc!DR`^llR&x-Gjn~zW
z-R^xgFw%(<{MWhA6Pi2=vyoTJ0&~9d9(db0SStj`FxPjN>Ju6LwTvH2h_vOimtyDM
z<Y^7oD2+FrCkWZfuC^W@y`83kplrC>bEjREhBc<C%7#vbj5eU!<(p50;=IP*w5OD|
zd3+yvF?{-H#c=oO;RLQwQB^b1F7w5uD3k+wk`MG*BnA)7WAO&8qmj*09PrFlE@$!a
z*y$ZDweEMBC)+?U^Vt5`V)31tUl0RPVZx6DAS&jX$4!_naNWO_<iYyvS+o6N)|SRY
z3rRlXraIho6ks^pHBpOFGs@tP^Nxc^DE)7S<Tbk8I57{Th9ja#Z+i0Nfh8#*1%Ef@
zcv34jt++bzURSH#rLJGzUEn9viI~0EE2CTLI?#{JsLgDUd$+$1?j%Udk=y%b4D1mT
zg{O?hs|)S{Hg^_9@&yW?vfhMQ-;h}kLG;%KA+LR!@|;d2j{J)~a>9BU?_E}kgsqV~
zQ2W~FsOP{_fLD5#K}x4}Seer@q!3bMcLWdV-8_bLd?~8g8;1hRw>MS+qj_?0)j!zw
zYj>i%m!{K=QI4L0aQ*{t!2)i;3@Cvfy%7}q1pqq>!3m?`s=#06Sp5lKNfF7*byryY
zP#z%r`2)A?kMqff4B_tp*^_n{#PrhcFHmF7w}9WB01C(Il)9k_usRIB0OqK4#I58K
z;d6S$hy00&AKsor<5hfF^_OCI%VSP3Gwfd1F%fu{yB!Mxthp3Hgf=z3W`iwoS7%uB
zbeVKTY2Qfjv+}wCI#C_d#GA5b%tY8uhL|tK;+x1&+LzfwFxNPgJ8B$j)HI6stUa`y
zX?+O!HQ-KZI(a+f7UTDbo0;;j+L#;>5P>l5UIG(^V{+#Pp}VCiu*cTD$K0*(v${Ik
zzTSVMT0<}X;>~5pyRc^Wg$<A)X*V9<*g=t0=9HKyGK83HT!p#$Nby@&v6RsDk-E&!
z$&zm$H~$D<N{M-(5;@XjD?ww<Cfm51y~HKS?fB)9yK^r(K99WN(!B1cc9%;`V97N%
zoMOAS1bCZ5JEy4)falC$OD#p7Egz+ws#m-3n#VDukn`h*#$hMs58S*x#wYInRNr&|
zeo24#aSW3fv<gkQc950Kg^-t^Av|LjL+;BFhiHd^hkJsDh^F2rLTMSr$U3l6RpyB=
zQpl1duXb%1p$@bsbML4(-%+++-MKmTLH>b_*3?b?tV4+d>rbWU4oxMhq=xT-6?b3{
zoOD_3sQRn~#GM5wSSh@bK5VVx!%ACCW!7g|MlTOU=UI%XG(TwYG!PGiJ#PfkS+E@7
z5G%w+)oQ6Pq=ac7lXsNgwuP(^-=7moI`_l!Y3TWB1DYP{0*aO5QG_rA4BLL_vif1a
zJM%(E+?O2^7cckoBSMZ|0~(%(G~)@VNueD}!Uo=%5V3@#6`pf#{Q5NFr@^qLIyDnd
z$AA0el^ZWlh>1MBeMEK?fEOEyu=-yR3RjO@N;Kd!1$fV=Fil?Ga7(wIvK{Xb`JiDj
zJK=a;%D$!W{-ojOE~k58NE~fqMrBCzFj)p+HhS79s@}3$tJLDWgXZba&yQH?gas%_
zzJgFlpOCI#AlU&@g{A;s^Yx2ozaRxt7&J{hWwIaeJ__CVkn)6#-f)^Q=~~aOX-toh
z&_isEmxsq<NZ|A2yUr~Qz%2SUF=dGw9%;tWe&<*<C`up<?FHu#r-HU5$;h1Wbax^t
za<aF~!p1_g?@^yXA+SenoOBXr^4Viy_u%m`UX%xX@*<Ez<^6gs757ZWI!m{<c4~gQ
z<jq|rCC^m$4NI0A+)O_T!tzx#)R+6iU~|c{)mbqL_{kpSH9O@dQ&I1?&18|L`q0z~
z`5WTs3vwLP1-LMB5?ci_psa+*ohMm4_0-(0rWWMLnn{YK<Lg+wxxZi>3~OGiW9K4u
z_U`*~h`)IoU2fALZ0;qxwx`xDkAD#BY`M{FS)*^oe_A@>bMH;3W6E@S{@D%C?wPPN
zn16WlU`<eSPg#U^q)8SA6w);N`<*$Uv4&?Q3(}@w!W6nn+AvIb1o!~_leuS9`wy*T
zehLT>iF(867=OfG=5bX}E~WOTq(sR33k>Ly4U#m%PIUkjSfW`;GQBk3uLzlnK(kd^
zk5Q=I`w#)gWTV)oc$YyOM~+r{)D)diYRrR^5SSEpF5d>#ehtBE%zzi=0S9viX1RY}
zCh;CkgaRX~cd_cL<qUNSI{tX6e4<m+-$ZjgKQAgu4Ac=)c;J9ix>$~IBkbb&2q@iF
zB4b5MbP*qP{88=K>5T2DK|}7I17<!#a?<>Aj1C?oMxU+Gbz|ZzDAFiR?*-(3H|PQ5
z0uplVf_tD>mVV>Vjk(WOGvC!RLkCn7)9870dPX$7pF)rpKwa?;5k_Oq0v*{?x@LJP
zmfGk-mjQ)oTs`B+h`>~FOGeZ+t25z;kWT%~mi2EXX+42*w8QFTfk2}Vp+=n(vGFfb
z4rrIXy!uY$y_Jwl|Cg7dU-vXXGA{;QhoW$4EI5_TLm2|kFZAT`4)Yp0y{)gw+3dXj
z+E878C^T2e{eFu+g~y{U=;5`EG6}G#W0BJ^VFeV93#9otZK3AP^b#-oc|HxRb@ynt
zOLh703>*!*siMsCKoG#%;PVNAC|pVe?A|9Kun(BGY+;@I@R)U9clU&==MF?jy$!Nk
zOo(FejX7&C|0z?`fe0TFK}pZnlu)>S?v#++LvSe3nekODx2E+1Q`4|t$-q6uTG;DZ
z$%ka8Oi6zqv2>011)1iieYU;(F9=AiQhc*i0V!|6`0b$7*Yb)VGw(NLBEFB8B-wPT
zK4A1sQ}(ZZec-4rH{HMhygxQV`wVg{*qFlYe3j%%<gzA;=FTPgbmx2w_De4GlBzBK
zp2boxYFMwV-+J%Zc@JMnd*RV7s1D<QLTu?p)$hh=8Z9V|Gt3*rAdVGHujO=a7r&Yu
zw*ci%A3xSVI>ph=`KF5bFpx2k$<Jp8;N#pw^roRU7}>E0lfNK_(a0(Dg<T5APaL`{
zB?f+kf-0e0BD|=;8I^syJ7XdKwcPo)D?4?<_g`IE_nm`(6MqGfFP+;-1S1&q=Gjm4
z@3OAFX&@$72?QZ;9VNw_rS)0lg;8HgG!V|ECys6(gQo!&;s!}(q=IZ<<Ux#`+sJ^X
zY<=*nZ>-K3;r^;He>TCP{Jd0`<uRj2l?^vORQVA5e?M>jZ2fFos3<fR&fCguO9J-b
znz8ti>6^x_5&JgxV=M~~YJ{C*GM!pTTfepnp`HehlNY#qnE+dNTzI17nR#TE^N6kV
zdu62SVnc&|D&LfKQDx}k(a%tV9*P<63V+j^plKRF7q1o4rC7I)S}ZrqBp$o7F6j(^
zl|w2tB^i=KQ0<o~sA4xPk8coyaW3g>UtDE{W7W`l;l0D66NCO_z2VOX)zkEmH;5$+
z`y!+Zgt89U0i=)%^oZsWPmdZnyY<6kEDttMNiAlcKb0+&nqweF_cj?KY{K*t`wHL(
z&Npd2Whm;g%hU}}lYW)LPJQZ8Gl@x@k%qG?zLq8n=DO!*HITLYPl``=AKOE8efHjs
zMk=nN7`2rFW029!IPUPAR%o-NrqO|4kVyuk#&ZE_@?Wo@-8L!D!W$GoO?CzHIc*w{
zHSWP{G3|x8P{l=gP4oOKeyqIncH*u(+#x~9hdHj4FviNZ3q2~n#H6(VA2ZrfqP|d-
zgFZ9~JbwgA0!nMc3bnSZxmVL%JilB{=P&G%|6I<WE$-BB=L1iJ*BMhd-@=?zV$aq0
zV59An8@|?fv8&#gwpcw?VAsu^R({u_^3Ft>vdT-j#U7n^2hf={G%n;Ij9>yC3pfv&
ztLQGIM4Put6TUyZJR#I%l$|ys-d$(yrk0CK6Hc!YA)lkq_jo}uBaM@hsCHf7Kmd1{
z<9q?ZI%Dfg%bJ0Qx0jcC=KMvnEp=v$Qgd5aKTJhAGDeIULOlo;g)?m^XDJ1o+IXKX
zkfh%ZQe^71F3YukX0?Q<*3R|v5PAPW17bNN>15_<@#^KhDoJuob{VY;U^&)WP8}m3
zu96b-cs6f0G~~ShKKOmu*2|?!ol}yJIoTe0*P|%8c8K;4RgI3tI?ax6LL>P$^(t)K
zW?dRaUoc-(4=}HK_`;Nl>Agk9U7NWK+b;Sp^ZAi_vO01c^dG|{38Dc3>4RYtu_OZX
zq$7#3$xY7N;iHJhwvYzm3Wj6uO1OTIx$VctbUS?AaFX~^J3EJ-cMxbgA&80Gu;3Er
z3-OaX8giU<@pJx`<a^Ru<QEE97zgmg-EQ=F{myXKjwSJ6!KjsDA*ysmFeC9nO<i1w
zGdD(L`nANO`~bv)aastmAlnrB#sm!2bJ|4fiJJ6|Rwfen2vX`v(D_?K+j70h`g6+G
z9yTY}g<toZ7%+VWoAcT1>ymD-!;~8>)H0aseL`RzVcPhP{rxqmrHNYBj_*6`FJ+Va
zew2wCv94PjGda$%2q!5R$JA;`>_idNlNWA^ym*Nlu8qIq_jYJQChdyoQ-v-Y^DR@t
zTF-=f7n&|Gq?abzU{s;erN+Da#ZcgVM&3!%HltxX(<5D`NA64LNsS#7K;Oc96Vi+i
z*x6KOVOG*|KM7rTl&{;>^?&?`?TRwy{oDg>bI4-|?KqG}Ed%=nVD}M0o!*3AI{=c;
zqnLE%PS^NMOxcZHXh~?Yn6(lqm6UAK-)ooLbL+0Aqe_#qx`LG=DH^W=-*jcU+7~U4
zys@*Lx^zEP*E7aYm@FINz6TJ;2>q1_bPDVqiU(nkN$;_~m2`PC$oL}K<ZaUR!?d3V
zAEh{+t|h`2P!w*J9rDm9AX{q-O$VBHn16n;Zf$L2IXPv}z{8gljk|TKQa#8UKRXSF
zu#YK?1psg!tXpWz&1GXhQOx|tQA3fQ+S0CK`_zR4dIyfRdPyJsRP&lKJG;-e{T2hU
zp2iH6pic-A;8Y*NEG?rJb(M#*iE7tDG9w&_X#56FiJt^(3#K*YTN~hriiLNIr)@2J
z?ld~3`x5mBk|HxdAPUprU#1~QZQ!K-*6VIQ0fNaS^l!cHeIO(aWO$`75SnN$Uf3<1
zj~xJ+82w2>AfStbzsZX6g~e(M3{y;BT9YaVzZ-eTz7JU$WPIIvLC%0);{eNQ@xwOx
zgAepVxxfk~6R>g30%TFj8A5-eZ$irvi>kV2>jXJ+kb>tOh0yz}lJTbIH%ud0!XN1_
zyD+SSiuW*BHi{&ZLMdwMfo~6WL-%jE@4H}@(n0GCfTRFi#stk9IMkiTFuDb2w|O_O
z!do0tJZ)`xNv1)8cYHC)Ui<dq$=Q=DIkcE%hP|v5+RbyQbvM}lg@1{-;|Mnw#K2_M
zqB{Dt89?0;juW5~kNc@r)GR;HDT<onmA!*edN<ybG+C8*?r-7tD1HSDxE{foiV8Fa
zL4PkiA5O4TVuk~kvf3~Ct)qZ1VKh!&N%7JFe1waqf-(wUPK@7^+ZiFr5!1Ut)lmQ8
zz=aMP!{~Mjo3&lwo#Nqj5ELcwkIV#qEUuJDDx^Po_<{AJh&Q|$&4GlH;PaSpzy4Tl
z)mpMs+`N#X{EtE=cm8kuQ%MxX5|+fXW{~8^bOdz@r*m(Pc>>HH#UAKCvLcNoME@u(
zneJlm7sy;SYN%(wAPkO*xx{aRR0YiUTa+Wzx7fK{7}qw82f;Pusp&n}w0WS~YQiwo
zImAQs%3a!Lh9zBj>tx9r`JCr3nPwY54@{>q%aIWXr-32@dj~4&=CWXG0on=p2F62}
zO3&v`<WEN~YnCBl@j8<SEr6gWgovI<xe>Wg56XV75SF5a-7<HOVqaA64r~^#ALI%}
zyEt%x<NE&SD2iai&a}$cdYELnFm_DAxd^^$<-5s_OE-xZxXZMv;D7-~T!i0m<V4ee
zaG^_y_;X>%Ks`lK;<4*~;3FHgr=Az&fA*itv{ZC3JF%I0%k1786$H8nTf@}Oei%6b
zJ6yXFHKTGPGx+0~c-$@3x~%KZzStN%MGQ1B2?)j9BJ_l5_V4f!N3uYhbpw)Rxog&m
z2lssCv=kRy7ZrW`x{FFXOFagb^713jh>Z@|sS%x0=U`0kd1|9eMz-p@)eWr;nHQoh
zDi)Ma$t=S<u(@c)PI4%!T}9KlJTO3y@H9`AAo^V61G&j{qb4_o|GJ;hBvg6OUyu2N
z^i=pyxoCaSR1D%0wHZ5Cq-;fztR~n4tkEMt)1w?7nlsNYdcg7CRUe<4Deg9h2uM?|
z5bgR~owKtfao<fFz=cWiQBqQj7OFdV&<cNezBvJ;xp1G9K>3UbjC6?-V87GP8W^t1
zdigu&pjgA}h%@UyxDbvup1gx6j<?5}=zeegeWqOm@XlL8{Q_v?+5gb)&?w^69q&R(
z-d(}UY(IPMAN7~~3;PA-)CQi?Upu=3M<QpjReD)?zd4~36qpZN_5$|E9;SHeQz~oM
zgD{7hr<IosWXn?`t75OHIMBDE!2$4*8oup*#KYW(<i|7pY~4_mORvJMU%2Yh^VEUp
zAI#ODz>C9X7TE1MG>mW!9ZfqjqAf;RYrDP@%ogc6YM!N3wx%eNq{X8okR5l-^cJs^
z7{rhstE$aQ8v?ItJchcIf>Y{Zk1@}aB>Z@UC_7eaId!>?(A%5gS~}qwe<}7h;!`Bu
zWeg8fx3TDNKEhl<Oh2wJo3OgnVaK9ewAs58lEA8Mb|E9}$Ib@*9f7^f<Gm*Q`XBSd
z*hjEqDd(v%ca6D72|LPhr(`GE%lfL_J{s)zh70(qX&BGK#F)s_Wt$o_W`JOwP$2|R
z0|@oB#|p>=WFP7cUo6a#J+cnYS`&iCE9eJVkxj93^?38aS7p^0(<B}qiJN`b(|>Xr
ze^bZzgM&#a0Xz4+m79o()?~u0*@@QYb~@Q!{$SMjwmMPwT(*&ofsnpt)rZB_>!aLs
zRM!7gc<DbAWcvT!|I_^%`Y%bDF?@$la*ocggTBLwm8}P6r(J34L$~Ai+t=G%zwxF^
zzUhkBof*`rmaSx)*MX$Nvth57C_J#v0Hi<*!5AMf|58EI$jf)bxxC+rcP7Niw2L$0
zOjxqTnEK~F26{S=PV_W0X?KBXor`<|aeu^-sFbXE*fVFCoLE2M`^NR><xlT|^xhO)
zULtSGskcCa*n3dF^SEV%$qx02lt{!?zGQ^nqWV}3{@mHihHBcz=Wve<gl{i+Pgx60
zlm}uWuyd&>MoPc}JfT&dTB+&7P@MGMZgPC;ay08sap@Jw#=K+6_DKPS9PO@Z5tmOw
zpKAjSe1tW!N0fCK2Pkfiqt5jneT~K2-5)_7(hc%~E5+GNw4jwXr20u+Z>XXsRB%w}
z$dThC{0jI5i|6r@{I@RZ@R;hf0g1xPke2_hjr70jWBLE|HKy-gF*H`hJ;1;SG$fx#
zSbpK^K(dfb8Anoap!7tcE2k*!u)R7~@sPe{&XwxQLrIXk)u)l!|0=Nb*U$g7y%g%#
z@PBF7{NF5&_y31yXZVlWZ&otRD=bbY#~rA<`Ofn@?pDn`lhmru%7+%>H_AT$E`{!+
zpQW_|MyyUSWD`od1Q3z^!#2Mlvmi4d2*Zg1pLSWw=NANprnrm7LqO^&DiUPK^E3LX
zG!(Ir#>tAJMVP~O_dxfBs^xsvl)rBA*I4|uF8<mh|F!di-Hqg03ZDh7eyEM#t&{Su
z&7K-3KHKXt?D4iMwgcEG?Jjz!6Eua!*yYnVd(Er8d^xtdL$tL0<e3DKvQK@)_4aL>
zM9$7-!fZTrQwZc6G2Xu*&ubw-BIyx`r?)uY{(@lpX{x9WT?}Yje|ZT|ZeXMlEp9rH
zb2@wwWEyql&}0f|_htYH@xLI{1C^+apT8hU_5(UX(7*2S*LeK3KK|M-|J?ZjJ&(|C
zYHImn<^C=&NOMW3=x#5BP1(n`I|#@Y!uR;BX-(Kot1V`TT)U=wx!X>wbaTYWe&AiW
z(xJ5K(vNP}AZOO>cHw6Of70}uHprmLsUR~<`R|=Fm%n$&%7c$H*#DhV_HP}s-hwc-
zzi#n2|Msu3_@BQ1Ut{stSo{v+{)6*EyW~qzng^G0rkUb~7njPZcar&%X~F4MwW+v`
zJ%}rI8_!M)le7C<qW{;(zXuaAXGs9P6+433-!rDttquIU$dm9pQ}aJ$(*K#?Po$km
z64XBsec&m5Z;H-YnKAQh%1P2E!sLQN^iBz3)<UpG(4k{y>7=nI@sjy{Mt}kn3@#+z
z*MGnY2@d6HM;MUMbMPpnkhUTz_LX94@!>S~Gu`rCAEf<X70LcEY}ZtRa0c>>x>`NU
zt1TX1JLjV^G21E9CBHmSm2ApZR2c1dMnblY<4c$eJ@sD&x&O(ZqAh>)dojmIroSMk
zjLCOEox&)_gw9Tgc!74oe|Gx^Y)cLE0K-ap{tI%f$!mWXx-VqHl=Dvkxpvmh#1dRF
z(6(5^M*8pfQp`Y6=;l~UOPBM9v1&)hz%Jv>Gxy}u6ZeIKrAuB~EKhjMRK2=VH)3}3
zo#tl!bn@}XZ>n~f53AAz$Il^!5XJ<9?tWHKm2XQRn|K%-s%SMlt5~XMl?<HX@#It!
zb6{<=KP%98W}_59Geit0@(YTo89JY2-QFrg6b|;Ucn+WjS35m<m8ke(eCJ+)iV&ym
z`GP8cNk|b<NA!bUAgV)YLJ6ndE0?6-kLxLuqV_!I`5}=u;lkYW?bR7&@d*CNWS><=
z@4U0rHz;RVY~!~Xp;1k<H>+yZK|Z~XUZ)4UraKKeRfFsjpC>T1g9?q(WsDZdn`n(&
z>i)#jtV>w4iL|Ots=H)Am3DDD^HR|fNr!E^#|{saWlQ3brzmO!<3!aII#QX{RSoOW
zyJ;bp9O7+LxOEenu9V(QJZ>i0grN`9Qj6bX#d>1L!`utciz>ZSc!L)oTv~trCNZeh
z?i6iAZFf3;zt?zJ-p^eq)YzyY`N!uUr!xf#ZWG@LC6-Aj>fT_}6&8fl>+ujv@FK*8
z`K)M!`5l7$oO9IoQG{dpxIxIysLoEN+D{`%+4VEFlLhvTH>xlKvuXJZs~So|fd<!!
zAio<Se>%iH31t)~(^zfDV0_@g?Xq*pw~l^D+)uGumD_~v@P9#=<IocU^O>;=&ve{c
z#cHbLP`)BXoKvT}?yZZoDZE?N*=MHS|C{{yKUi?djv#?pOnePB8bC9$3%5fNMrnE@
zfV{jx$!wtyW8VAFi{3mvi<(rTa@THoAQl(;_XIbv_$*qy73AALB~^}5Ai#UjG3cWR
zQw#~v?RzJ-`&ljxxT#^hygdoZm+pFP59Hq$xg>Rj?de?NcIzo1t4N{4J-g7Ot<++x
zVKBJx)6>~8;LuxU?OQA#4aFe&Y1bh6Y?<-jH3b%+Y!n?r4|+lA@Ko*h)K|wcC!FFf
zSbnSYD>@oL9(iX7@JmFn`ffq4>mlB~LXM%|)TQG-V7P4p%TE^<ujhE{)xPCwnYt{p
zkSrZ%8xbS)lr>U}_Z>Ig)-+EWMU(Vw4ik?&8ry5fJghxl@X_JAMi0TmC0UZC`c-Pp
zwwBgfd+qEbdgm^IRy`rrpV0j=^=z`Ghlir4*!gCz@4W)%qQoq2y6qO8_av2GQBG1E
zj*Q|a$(Tl33|6*&BdFn0zXCOilRwHC*yfvuZh#U@CYRB?<3rmh?ruj|q^-nPUKdZU
zcWL^v52JeHvLBwYsLwmxbDUw?1@lCcfoxCfhjAkG_@WVK=fiR`KJ57ur`u1LHr%~`
zz)My((|-5XquzW0Ovy6n>)t(kw3%t{NML2N_q+m3Cmyk<zrQMnI*Bof&}V$OWix7n
zuM&7|4SC82T2W-r9)LM{*w2q9Yr=`KJCy{=u3+t_jr12QSr}RhYlCfjM9-ov_sJwa
z4gC>G72z;;ECNcD!)<i&FQ&K6NDb8jD|0a8US+|v1=gY9Lf$JoQ;hy!_n>5Z#BoB_
z`X`wvZA}V$V&F}y`mZ+zt*!mN-b~5qNlqp@JmTZvPSxkh*c13hdt+gfJ1G>>lk{Ar
zCyg7`oExTL-0+zdVg&>u{!t*AVHM7f>SF*~5aWW{qO;om4{he?{}G!>=_GkWDJo0&
z@p*VGo|b~^k2m{H>k1v_lWbmPx>Kj?&6%>3wE{260J;PyaFWyz(9uW@lH;ig#F?*R
zZ{^RxoHB2b$G$~p+wUK=U(*pD#T%iMp*8iLu<@R;y@_j20KqJHBH8>lV;$`I(L!1D
zzNg$)5hPzWk*<v%dk%KjT_`0kPlJ?AzV1fQ^a3(ATU2?Y-G%BX(8}S_{Q~~YGao(R
zh1_5x{cruFTbpIZeh0Lad$670?bS_aH2f#`FGwAftQEWs;lW|KVc*m}fC?C3rCe~^
zmHoG9Qc%S|VakD2SKYIu-VUWiI19>6L{yE#b=UQU+l<8`Ztn@(xok1lA(;WaU^IS`
zs7eZ%Q=&-B2xj5Wp}S5zACm8q@iFvL6Lq>0y6D>lq%6&wXag{ooAZFcfaVV5v1bno
za2p(Vu3dcl^Y&8Rl1UNZcjr(VsM)6#e0I`)XW<QrKsZ)qt|z|#zSK9ZVXnt@5?Vs^
zlVXWzsbz-kggH!S89Zv1xmks<Kc<Km{lw{Sm-65jq`AR+eeli6{n1j8GUM4hLNvgQ
zq!5j1%u3EUDCc=kUsrjLO0PRCrcTKtA3s*Lo-4du${!$r8GS|h0^Ll%0P19wsBiMw
zuTo3#%`)V3wEospqew27Ol_X8n&%Z%N3x9TjqY|BoXYIFqrV+pJkj~6H^6ESy>p+m
zO%^~vN%qAqjZc0y8~EKzbK6y3d0;G=rOfh<?}^FxGw+V)BN_z0r5R7dYD+rs=JUI4
zC=Qa5j;>;50uTPFgTuSiWpSF|5OBN5k}xR`WM*+>HRKSCUW(*gM7l8lI8&6D=p2nt
zzt$3av)<y$ivdLKBaIhFs!L}|*q{>p?P5?(P+mIQVXL$=L}<m5wLXp@g^90TJS(>h
zeTL5`U+t`p6u!cJTxs#l#7Ta2C-D}!ol$}+46uIOU{Tr!no{R$M{aK_3Ti_^*`141
zuIu`hJ7K5SwCSkNzmZzV7klTh5lDf-jb?_zi;|#}u-sP4kS~bU6IaMAQde~6QPr`h
zh=}HZodsgXQs?Z1GfpXK-9waim3J1e;B{1LF;3CqeM0hgtv1nVV{W>AJ`4inK<C+G
zHz~?*3Db|y9nrZBY_r};M1!UHJ)oLO$Nu!2L;5E~I3Q0f5!K#+Ql5Y$ZJsr*+CG9>
z`Q3Ua+M&uT;f2@7FTQRHU;W+!2W-cBrS<K?KxQv~S5c`1(-xN_ldh>{3GZa&HoGMq
zoV2eA<T4+;Z!EBYy$y}AlOhCk+B)~zDc6T+S@&kmdL8X67V$8Fp2)V&Q0$ioLxsKE
z(18XGdYE*#2|a}5kwjq66u9a3o|Dy4?}!Kh)>e9fN>~o_l~AM$^3&`l0xpbN1J%|m
z7ftu!e^cuF2lt-n-ru<wpez%Ip(IrNg4oXPDFBgfo}XDMuy5|IVE@DSPq;5n{L%fW
zq3io6&KkPCSfW0R<P46x+zvQb0bq3zcDu$<Z~P8IWsk&+Go+6}0EnF(w;BMwX={$h
zSFYO08Pm}|YxAcIqEq}X*)4c!C>*>0$_916Z^j9=5{J1)ecQ^f-h~P~*RmMpSnGB8
z^@ze{Dz$EIZ}rg96QkD*WwN86P}SE;$^a|2hPRNc8<=zOXyAp5uLmm_HW?{(WElso
zQ%!i*<owHTzlXt$DQELn@^}ceY+RC{l60qGUCiJ#tcm*&2au?&$T^<>TU7OqSET46
zQqlr8PWuFLC-q>{!8nA%dIg5Vxz5tlA>+lPj=8ck77LI^fij76DdXoTfNB`w6FQp!
z-4QXSmg(@qyh5Em%x$dyEJmo1{5;sgUy$Od6?QXUHEXO-iq8A$V0|^qo2tGVDP6F@
zvY`NaHD-+K4PU3Q7k4QtEQH%PriF?urfptf-MscmK+)!BWZN<X!n{2@xrT#11X{d8
zk*0*Crs>rg&KBE?x}1DD)}cwl1*e5GZOWhgI2xWFZqNyv7}~i@jE(R(-On|%qAfHc
z9*vx}51{Xr^eGj+o$!si%JrW6okS^<7zEJ`DsS`&(phBNz?2T?K~D!Ld$Y6KF7HlS
z5AmTVexB=Sf;4lqxTD?VAz?g!J)r~#jiJdPr%~K~Fg`2k!WSgFAJ#Rs83CON!^tlX
zry8@As=vN*@{+)U7?qE#LGc+yo}t(f6?!8PCrT_R%3r<~)_%?Tglo-PV1CZC8@0DN
z$g{Z_SEDTujd(fNpERM=t~41i4=VDBlG?%}m5Xnhnl4S~)uewki?F1R@aGgT7TBvk
z$DsqYL{4FuG@<jT@Ziyeimfq)(;J6-vN8ki>sP(6>RFY1SRHnm|DBjG3!R7@;O;xY
zg;zoSDMO%A&$*8=q31Wd47}wY41RLMdv2;X)i|cS#iW9ga<7SiHtjPc5B2uh$eak$
zhc23G;Jt-}Yi3XMIQ;xS&_(<>*j6a%kl7|oEhEU`*0IchSA7#l9g>@N<vpIo$YzdL
z=8`8H`~4Jo9ePSfw8|zW84-Yk#mw-x5L1%;__tbQ+XD>cV_(AjI1Cim15Hf~wHSS^
zqN#ZqcYI#CDL+(@pAUMG+PZ2WnCYb5M(<9*nv&KC*qs#8SD$&~3BInM%V;)=)-hV|
zfOy2}1^=dpW{)QeqvQ{+SUj4FI?wF@3E7DCg$n?k6JFnqhpoq7XX%k(IDxU1kvA;b
z&&6NMvnlhxHIce~xgbU1u*WAjXe#EEO@09S<S0J#C~_5*<~39EE;ngs%Lo<r?uleB
zE%Su@7WS`JC9m1u=$O1&g;JjRA(<oHE|jLnQo?k=K{O^z;hzY)>9&<(vES%kfHoC1
zuzg5Ha35ip5~e!4zY`16@jRG+IsBl%zWkr@4ST6Cfl;eq=*f|Jh;rJ95p<-zTnN3p
zw{^na{eD%CeXhf#p?!114es!;issaMrmguZe6v=*<?78fvtW-J!RXQb$tO6y9OK<@
z5EZ0!*co&rj7^@#iBKfWc79N<pr*K_eq3uyTl^`x$?lwy9b1LD_MbGC`$q!6?TtUW
zc0ueNn7w$^Bs(>)B{+=v-?()l8G&Bi?0c+}sL=+H8=MH!9Qg+W+uv$K?3vmy+Q=*$
z)_HOWR4#E&+D)%z@zU>qJ|uqdL}YVnK}~U(ah^(3gpOEPusJ)KnIb>ZEJRqiNs3$$
zy6G_we`sFua!NubbLE~9KX(b?hvFvP*qH<La_b=*>@I@E_}1c=D|Mp*{-{;y9PYGB
z*eU-5m^r<BCQNaNo74gtJA!BUUJW|Via4lP%byc?)XHYcE6LjRY{Uu3QkB?fCaN<S
z`5wJ>uAX+}G?Zq5DF=-MluR5$$Y(q)57K3xe!!GJ#R?c8QSP1SEoM-&i{$|I6%YjQ
zqgYqSvZybsvn2Up)ShT-?0_mPHc!0eHL0R-d53G&s<`OJ<m#;b#g-7uLPn;iebUS(
zDUVab+eOMkXdD<e)EC(y8pVLdQzh)mDS!R=MIHK5eh6C_-9#AF?#}#d1o{vn2;m3k
zC#f5L-N^M^%S||w<3Dkd!Kgk2a`5CexuNBJ7IxOao1LO8gF_by$?2{IE5x)9Ue5H^
zUHsb`_7I0%t*giY<Pdr@7)_6w0nE?KR@f`DMRmODd>qsLUc%{}yOIu!L)1Zi1DuYi
zJ$suM5EBp+JrUW~C?EW){)r-Q2c+Z5QXt)ZsOFEJjo|IymJrM;oAXtId8Dt$19I2C
zcH|xV^3|udGwtr&=IW3~s1W7SfNu;xx%w(|{_%+LcLd2W`@8(CuW@!8l-Z<>=9{A_
zr+-0!>iSyKFNhRIC!fCc0da|v6fnE<0rb>S2o?G_S)jB33|OkoK`nvZgB_%tiS33R
z_D#@O+Yo7a{iZJ3qf__$Y+D6)?mLZ+INb^E%scfEAG->&1O<wot8ZuJ>$EBo)vrf2
zT_4VBoSItL_1NIPBcW9d>R7fINFh4uX12(yYU8!4wh_x6G51>|u6Xv=x!(=bzDo?i
zr;^W-__`WXqBL*Lzi}*cz41ix4w-Q)SJJK!abFo)DrtW9-z~IPxjJ7e9bJE@uDLq;
zWI9I+#BEC^<|O=aO>xux#KwTOW#jcd+Tufq6S!I4+?Jc>4ap-T&~@@M<z4aZ(xyB{
z(|1}HVH!ERvuuHK?X7HNI7R#%YfS9Up5`R_^+dPFl3{<p%;U0_x0-dvr;$qR%W9iN
z1lPKOz9=5YXa@RoM|Jv6gOfOj7fWE8ONsZ>AED^ZUFF@U87LcSW}e(jYw~mDKdxv@
zw#3*LgsJ!@30(pr!Ut3G$J$?vSylFV;0rD8B!%)EWxT%5h;%{F4{W6zMUG=dYeIv2
zSTA?1p5=89d6Mbk0I%t~zdTpQvLlW)aVs+8J!xYYz}W2FuRfb<!sNC4gazFbb_NuG
zi~eq&xo>OSuFW~5YV@vJFvQwoKFi(DWX|&ltxqh(ya>c?W-ZlkhRc_RYWXD^YZ{XZ
zas+N^KG0uqmnd^kel2-a|IplM2C7q=0t&zi+;M0MmnDk$YJ2rKLmvu<M$)4OP80fR
zhs&{~ufd;Mnt2}?xVsB>vdD69Kuk@qhrMEgfbhNepF`}mi~rurS&EAOf;>71JbpXi
zOfJ*cF8|90`8Dt#m_0302FuX0i6lo&qT(%_u8a_Lu!Z?8h6mNB4clLZZSm%_KZCug
zW#8M$qLI7^g=7trUl6n_Y6B>GFmJs-&hQaH=>Ow#iFDS>zIKV4>?d3FYBGD~#b*7$
z@P1PQ@d!OI2Uj71Yb(J|uT~qKO>JSh$G-Uj<?IIiiC3Z$W^s0-^TvD3>f|y?89gBW
z|J@h;;R!HT3jXZsm}_wM1+ki3mpQaT6}s{SdVC_5LN@vZL8z0=Yjl67P0;(RwXr&)
z&eUh6sQw!4ZYL@l`o~Ybxc(;>-5Q;W)_D}9oU|ZGakN|2u2w}?>#WZ{+HOw@r<tzh
zs8ga9urn1<0EzVJ!PY?4q>7Fx^q>B^KR)Og*Wg^aa}!rggV74FRo1$LO+Cg@eSFvV
zy8%tdgemb4V`{MD4-y&LgMfU<O5<61Y{HZcVy6Ew<bQJca-(NKm+ZuQ!#PM&|B79f
zx$o@Wp$Xty0S$;R(U729IoHZK=)p^)H}pIAZ=wD;fK7oInl|`upZ!s}KN-qDzT3Z>
zgnNy4FWR#`q>lxs3VCnOXa8XYL$txmhzh1qec($LkLkYu@%yj;;d8-`+<$U0^4jp>
zw%W((G_VY`uS%H4ih4X=$;&OgL+K6<#ZVC4wEbU@bE^pJT|6>{dYqx{k8A!<w#1)Y
zE~(6V+3Ig~Wm5l|gnjsGAjYublMvMwu%<$^Ac5FF&L8(5Ce|)n4-$m_XRls^H*43>
zeIIlctU04Rel2^YQ4RX>7W|L%_r*^B7sS+q_7io?=IZ~(SAoZ@b`5?S^$;(8_h4XB
z3N%CI?{@Z15)(}u1o@ThR5`Xk+t~lhLqQRtRNqif8~Qo#vKw?}s>oT0>OLZuM98AF
z01N>}&1sT_$aQN=Sxp<4PpTem4zmu=xS=0VUmRk4-%j=yBxM=92Ef%XNsvJ2|N6l4
zNCsakJ0UK2_GxmHy**WKP_BZ2-Y2K*8vM*>XFBIuP*i61{?J@R5{(=E$MG!#1ijwv
zUaBwZyIb-9-uPAxZl5;wuX!TpE3MYvTf*JHj=w+)1$oC-A&T6@Y4*JXD`-R1)-em;
zKiLNV%MCQJVpUXf&&=nn@SJ|{{tdP4ml8Z7hPRUQd%$jx-Gzdh&Rz<3yAK$g^VNP{
zd%qdatu73G*#BuA{$wls$>kCerJIK&)zTR~V?W%3_VTQl)&ml?m<DWrMG6ww`(K~3
zKiNlHgSCd%zh88VqYC!EMY<6R3O)L+pc;nh=2Kp{U9U2z$+qx2sK&H+6ZJpe4P!rw
z`ljH$4c$Lf9uN5m`@<X@r+prVElt45IBZ9Y@}FJwT5mCS&fYPsa~9Z~6!>rKy?H#8
z?b|;-B1%P@6lvO3(zHoYF4Y~O(#DjfX;&n7p(N&#79o_9HjR>0Vp=R&GRc<Q*_pD9
zE!$XT#2B+&zQ=Um%%Ioj^E|)L=lOk~=lS0M)Li45Yv#Jn^E{6C@jl+iadzbn@m`3)
zw*V|~s>quj$p257^WoP1-qU_}%@N&eP}loytG>;9@;U9b$ceQ+6v2oX#6a|2y6I@A
z(#<B?XOFd(gBI|A&NUBS!6;|n(_Mc|xp_iPbg#VH_w(at17;^X@Y<1}f2Z~PUZkbI
z-~0HAWlb_COgpK@QPz{N4WMl8z}LeRJzN;O`TmJG$5~y=ULl_cxtO!w0AJ?*>A$Vo
z{<U@bs<3<c)Ah%fnYFmZeaK50GO!ltC!oBAO?|kKe|cPYuv(_=XYuELDu{pBICj)J
z<<g|jjcx>3Ek!`ts`kn^J;PBS3QpOaBltKR_8F~a8tnOEGm^BgnwTHd@Ngy!`}-_`
z90++QOVvrbg@5Q)cRvqnDV%j<al{d|&3;$${u%|$c)*|aBb~c%{=C$W#(&!)9qI}r
z0an__nHb&TP5p}hzEyI?QPT09<B@jt&X>mj%DTMT<!-UvLStct!V|E6L!2dfD)<jo
zN)L@=Q9A?@{}ZeBPyQUC7yLZSgL89<fx)2zH<*R*7sYkwb;r@VJMgdx@KlNW43Mxm
z8b!dNtoV=qF+5;%$<w@`Bf53bU%!IvWnX}G#HEz)^b#Zr=LS<!1d(43hsY7x*z1(1
zcmEY|XIr-KsfTjR-Hl_fWv0z3J~e3E!iasi9T8ybM!wYfR|d|dE5!W7)t67*oXb`5
z^$WeW7Vo}pe?;cYR|q+}Ymn}EWD^$qM-`e?--qiipoQSh5&vz|lp}J@M`P~DHf^!L
z*d?-WT7+KNW?DP)W3m(^9wiXLKU`Y>@Q=H1H0%87n$>QvLJHr{PNc}g-7vcIcs@8*
zT#@tSgHd)1=6H^^r+hb(#Kl!WN6u2E@R}I8&;E0f!7p4%2w|7}^s+5ViFnOdr<yag
z2RRAOd0Ov?J?MMFkAG<qcD0S%yHaz_bfqJ4m&UuyyrdMhyQ`yDP85>N1t+!1>L<_i
z%iiD7b{V8j#Q-cZKsYioT=M9dw1-H95Jgc0H@}a8jf7IHdzpg|are_Hos%7Uh`r_`
z9iaQ(20yF?P|t?}>CU)MNWB!m5o_~T%+rd!H~>F%GnlviJZn(_t$v|B)_sf{fo*`K
zxfQ+lUhXg6!!<Ph#n@_`%)#^%14;=(`PUM!J(l%fa#1@3__I_=ycP;?{0p<Qpoh5b
zM-@7o3P)|1?eKx}ixzwj5_LMZL^2c``B&yo)zwDeVvBTtK|>CjS4jp^{lWmLb;BYb
zjoUzA2Hc@BNR8H}zGNX4ph!6GRM%gXph^j0Xe5+>P;Fl7R20xk0OHc$=F6(}CmYw0
zS-hg-IXzybp#GSGz5oY&j&?nIev*U4!GLB_dnKOROTpeDU2wI0w5|c=RYTLRS3k*&
zpM=I6No-%|Hm?syKskAkn2G-EAzz(r)`!Cvp!8)B6{~dEd9sNDbQ3)JEbP)T={uq!
zkB<L{GTV_#d~U)brkgsJQYS1BG)afREd($Fc;JfvC^3#l;WBi5H{9yo7q<o;PQo+v
zllm4OkP46F)1tm?#MbNB6q1Ca%wZSKzaEDZrUV!2MVU8XZeBjz!srxL&bXDvUq<UN
z60aU0`Cc7pFdL0BR!W8R7XVwHMLMP5LO>K~KF_`arldk1mT06u^;Z8w&ga@-8x9FG
zqg^GCDG?&Z>#<qIXw&RB8YGE>0d<Y^DR9O60<aWC%+pRG9*=E;`(wfrZJICfNFCA`
zom0G;7`)LDE&<qvH}1;4yWG~nJUd4A6RnH0SK%re>IG;J?m&N1*Swz8N%#VDKdTc)
z$`{+zOJlync1wBn-;zb5y5O>XAFc}dRBC41MCnw5rhYDcxH<aDZl-YUS%0pFM!Uzb
zE8Hb7i02#ZLc~ie$Fz3@sSo!a{{t=$T>5nwXg&;wPTIgZiW^uo!9`P{lS946<LI*=
z`*5A(5HRM~qkTN&f0&L(Kkci-q~Qj9x)KK+;xd0u1O{j<t}D??O@%VV3DzmVS<Td-
zzNGL*g(}RJ$X|@3!Ikv;aPKIeUu!r>;EMh=*(O*9l!P|YfwsneKK_}Co`{xSctc9=
zx`ACJD@))4;|am&8@eWBs6H9cvBU`Q1HBj7Bo3b&RZa4)TFf%PS(B%*^~fXJ;M+^r
za^f{L3>{V&gnzDhaeJoYW$AUeXlZQ5R`z}DGCBr&GR%WDHjkczk0A+y8CMycr7b9L
z*F2XmIPwqdGd>3KMUFue>r8$I&pbH7BsAyXbv2!72P>A0y&T{B=CjM!Wu7_@8d8F&
zxeV@nngGD=B!pRK3>_@HdGqu~+A}Fkdz4v)F!uz-N25`&`l&E(QMiNsV^2I@Lr8G$
zHf35_pLsZuA57Yf#Yklt{X_<)H-(4Xw=NeiXf3o`1r`2o0qcb%G@!t!Pm!2<REV3)
zDUbVb77Y$aT_-AXONkDn;p-B3gzo60KQKfcmW7`H2+e+ra?+O6Ij#t;KpQj^Bwg#k
z)U!6d4A*ktH3~PKE-B{j!jSoJ4YYW>>=Fs*XC~Z*ZO|4?n2_xQb81|Y*k_eP4&nMd
zFFp`yI_U#vDDEku)#L5pK1BFd>GbnvMyDb?qi7@1OiTDH+4zh836k#TZ)dba|8HmX
zw=??N8TtGdI-{K9w9W}4)AU@B;jAZ)R=>yeQ~KCJ4<v+EM!-xu(jE9yoe|-}z~rOv
z?{j=1oyupcVwq~-tBns+LDgSM9quhiAvD%jw09CVAEjpo_2FWfUB({r<#eP}1kaPv
zQt+67iY>uh?qh?I$EAxr?@#Yr^pSL$*+MCOm)N#;UeA)0=fY#s{(m|TiaIZockjpm
zraEcchih304Hc?my)5t1czy`>PV+#1e`^#u(k}u!r>aTty^dWZrq!250U7QBD3JV-
zw*56R6~(lARdC4f0PmTriCxM9X@!2@FA-6Fk9^LG5Rd_F!`(a!9cwLi9g(~U!KITZ
zGYr*vG$6B6J&ShAyz0ZbLXsjV{wyF16UClz%-MV3j2W*ZAlpXA64WuzMc;P~_9=f=
z0R5t!$Uk!;Beusi+dI3TcODlMb!+R&qKei23+mS2wsHLUY<l2kocWYs_KF%!zb@iV
zP~O+<uDqTY6?abN?9HkK>NTWuquzY<YdrYKW?lM^3;OsXJt}MoIZs$;wUI5I_Tq`u
zl@j~6S%t=5H`+KJvU_nnxRmD?)DnE<{9c)Y&7pORm_c*arD_R25WCj!wV`Kuya4`v
z<%l+whAQs=YviRRBBhFpn7%U7U1Fy&fEQMFIrPd~qVFyA7J~QnLFro9(YG<f-}6`_
zYE=C*dtd$r&Ri1K4RdzGOJm8S(6Pq-CnXJlFYMByAnPKuehy4Z>s|1f$x=zRG%n9e
zn8&n=a^9x_`287LfyI6vm|?P#u~Ja{Q#9*J*iZOcI&WZ-n!@Gq%g_{BZX#uhkWYdW
zjK4=g1ViYd_tJOOGH7;PC~Q*eJ^*8mF<Lsn0DfaC#IJPdhDp-h+9k%%Khrg3r-e1$
zp$bJAl$I!372sU&=yLJv0ZcOPi$p#pULu4V2t`|yB?F~*jL~QgLkI!Kx5zi4Bi<2|
z(5x;}_aA7JK~mY^7z)=Qg1O@SMznDvV>BQ_ST|0n4zAGwLQY;CmWce&+8G77RGi-E
zkBCm%;yXtd=>>{%1h1ISgSRE86at`slSSKqbM1v|H~gQ%@2J6-^x+!4Fg;qG{?uag
zJi#0uO~8W7(FJz%H&-oGStvVgw9I+`7~E$VQ5{tNSBVEZ%a1g$G6dmigg5?Wh$Xgx
zh;-b;<^DDEa8XH=Ua4<jn2a<g#<ipMayBpn=1NSGCbtz3nJmNVi0iBsub}nwkqTQ}
zWHi8bi#kY)PuQzOl{gqs(o4a^=p-2S;Ye}-3AMZ|1E5H!3+-m1U6b}W4MdOeP~|sf
zW0!+arYIRsp0!ucSMWULmvsBmXZ?i0X+p1Z1DNo-Ua(RS$<7uqx+~B{xU_+r60_QN
zi(;K1n_dp$bM!{&m=twb7uVslc7iXCu>~#^Y{Nz@f%-!$S3I%5M+hzuAV>@7Emb=?
zh>aewu8&+u*!l=8QM*&AfT&x8_R5OLqff+5pyr6UiqsQMK}%v05#wq-_!u@qfNBGx
z10;k$$upT?=rpcIO9Wlp<tFq8i}1fJ1Fkg_tuXx#7PD5dcND&7Qm=Xc`m#8~(+hK!
z&y&#&J5k-fBfl;oRN-bCo?YjN^<X(xaRaQ#AH?I2_Ehe*Fje||!e)*1%*KF2d#32j
zcs{OmVz$BrH6z_%55;xRfihDFK}UUxzyNl3_5v$VcvmCS?i=kGt(QQXKx;iv3_UNN
z8nf@=gy4KUVQqh+uvcIyE2%;%>9JWaJYyL;WG#?d8=>2A5`Q1<w`aJh_l`n)s_=Z4
z&~BmxV=Fd7{hrRwH@e^bd8}?mO0|oN7Iva99iBhk-v8KT8>m>ADGc72J<bV{;B<<>
zRCtz(i3VqE(r^by-}(zzh%+CUD_=~o^@$Gr071rq(uQzwOO@KfEC2yZoBHRQq@Rnn
zkZ^&xj1r`Nj^j^1S|U$-fqpV*gh}z^g?JfAIyy@U4=5&3m?~!l7u!%~AW#<3X6tmZ
z^&DDvHmoe-PgL;h&P)G<#FO^qruE_Gu7Hv7b){f8-=nLQgm}RLKrqu_$THr;Q#7ig
zJ%IKAyy-%iDD~x-D(I!x(7No1Ke9;qjjdp}<};vaYbp>LVHO_iJpY0Ke-nJbKmET%
z@b%_FQ(cXuq}aj~CHLEBy-qXte$8;aXmhQa{@sR?0_zZY3$mI_=;<Xi0lNSc>L-mF
zQ+^`-#~bQr9UM=CV_mFeNh!nEbj@!#x4xv!cL`_z?p50iTye|YBYC6VzoMNlzhZKH
z+7|!yqkcCqjYSdIiB<4`;(@fzM!GENrtFKHE8ONMZkFu0o8jCZi<mcW=voVAUeWz#
zzPi@GVbgDB<2+u3;^fxr;}d_13wSv{HBNtK3;s;mi1xe&)*=QkoV*oqg)}&j+Q?Q#
z8MkBH*z^2p0t2>Rqi?rSY^AQ%E6=j-_9xooGv-)c`R#hh=csq}(iQn~O!}3mC86(c
zG+wUTHEr6FnKpAu-)iZqHfxRX@~<*5KfQj`W+WWd#Si<MXRzl$urL&y{bM^C7Z3PP
zNgtno=~a&OiMuMRm{(-*rD0D&S-bC;g;Hg)-{K$5aRGB)=d6g)4>Ub(I-l|<Ko)bL
z;0(a#{FU$*Gf*3AD7~WtdQ=ix27LDRk=AFi4Rrht5fAez%3b0>43x6KX1gLhF2WlP
zydv7G3^0v)QABczdBt%ws5G>|hoMdLLu!#;`Rp#z_pKndyZnf!jd>ogngee2C1n$k
zotOLlm*oxg6HQxq7XKMC0RtX;yO;%+BtZ9?MLtG9nKOiS0zaI!1p6873spm%6Y;>$
z-;F*q0Ex9Vk^0dn)`jN6#fvcgU`Syb3YcQdyl#vn_nE*L1G5ssCAtB&Ara&+U)HV%
zv~fOw3nZ{cAye{jfwYRq)FrtqV){hVUNFQ~o6hj?0*~mFH(;VYR{QkhXZU}WKy)2x
ztQK9A9wkAF{u$6w^BlMfM^=_lFQ%xf?tQ36S9yA7nN?jj(?>X(n%lu$XejVzj02+4
z42VY9q1i-8ebI-@gYvE(HZAi#m2<MVd!Zg8SnF?+5MpWlN@m(r&#^hF)2Pzalfrwn
z`sx%l^y_)B<poTAC8!A->ZmymL&I1P#DZ3R51y#HU<-`0p9a)r(3^tDXQ1{^3*3My
z7nUzcg^1Cq1@M>#1*+*{Uw%w_Vt+)KWIEbChx!7wsAa~mf-4ev5dhA0OuPNMSAKS*
zFjZTQi(Q?h)0PZi2O>Z<^im(Lv_h!NqpvszvBS-uBj#M>Vk)%HD?p<hwJGdg1la9H
zeWd|7=(rOm-^u$}9dt4PVDKFMqzyqSAh0{3V|Dt+s>KarC;Uy6j<jgpj*j>U*U=U_
zVvjwRIh*#50y-LIhVuq$3#D^H0w4@OLCfWVFvh+TCeOq)VY4VjRLKxrSd3H!_e#V$
zu9+dQ9{G^U0H!=L$aEY0Aa|<pG_dGqpMS<1zv^7^&*Ij<ow&xp#Q2*S{nv<5(Y9A3
zSHQF+ex&p`QyY(vZaMQ{q7Qk`Nvi8?t8X7pUA<b7KPo^tr8`w{T2w{AmIm63C<@=B
ze{dk=x_Of&C%j~c`MOqPYtU>v?gePeuMi1Y(9`*Hx9JtABCS3xWg+@C?J2muGbOC=
z?pZ88wneu#Sa?mx0rRrg$16Z6BqdzajCN|eoW)YlK*6N#7jy8@^%krSp=T?tYiG&5
zpQ}gXWFw}32JJ31P8Y|k3&b{PApcjAN9^~49*dZ?&MEJB0YVw`O?DEm2bT58K@zf3
zFdv`Z<ArW`p^QXGVb48}ZO9h_I9|W9V*;FFZVExtA_u6h(o`&<bxqpF#tSYV<`FtW
z$1dtOHUqgoGcU0hln5w@zrjdE(fC=tDxW14SR*>T;qC`iR!;jQqt%#!pgG~Vh&%%)
z5r#Q?S>C2}%E0KlB8?Jpb<;j&_zE}=EZJozvs|t?UWFzv$03#+Q-OX!yC&=lmOKKU
zL4<D0o6yb?iM(tPYN*v1tX)d)&v=POPa-I_=U&7%$O+Riv%d&E@>$re2C3i1g<nBZ
zp<)Hp0u@0!0K{A9rw~>VmZC)xhUw!2gW4ExiGl~pX;i)<oXT5gmu^>1J4)?cP4FUd
zW)wlt>6$+h9qP&cCS*s5T(Cqm`hSOz{Y86;|8mFHhihd__To5je;+Xks^Uu5DV|fE
zEX9o3ow9{d;qa-T{);dZaO9><GKVg1!kq+&_E({U<#qJa2nf-m#7w#DLq9-pCJsH3
zh8alb1XErE?T^wk!{5u7(;h&!XGonRi^;FiCNpo#f!DjGy!%fWTpzGAGSNYB3DU<e
z(54L@+HPP*@!R`w2|~M|=*SF|-L1%{lqOgV@RH?4NL?BrcHpR(6XUw_xU6DSR5u}!
za2?$?lti_P9VSv)ARQnkIr^paz;XCtK-gjLABqjqLXi~Z{-0v_GMBePvv~d&!R1q0
z<>J>X#nuM`nXEs6nN0<250`^T0S}&qE{6qq4ft!d;7hdc9oRP*sgiF*)fz(Yc_@iT
z`C4ao<A@_+mrl$VO9&8-qMx@?_Mz4B2)qWQYmWZ(3=tFToC2dtI)yc1&e=puU~hWx
z?E#JY)7xMNMAQnU8)8Ram?gn>j2MuN(E)-ly9a_9#V_#oLW;mMWpW=bDGJqHBnYFH
z<PCi*R56TpnAE!l!J%J&L!08G73+cpw`o`Lpc|m^h?-|9g91=Fb`EAjN>_wz5J{+U
z+hc89N*&i72og263HU{AXzUS3_0niZycnGa7?Xi2D?+<;ZE@VwutXJ`^T1Z$uuG$H
z#=HTjZenitkvs<jKQ|JSw^cE`kbHs{!MUVj_Q1#9zHs-l<@Q@%AL&im^-v=ZsY#|C
z4d`IfxN`VD-1Jbz=wsB$Tob;D+Z1j|Z3N4~lVpY!S{^@1D~u{!z}e`>SbDUvVfw-M
z(U5Qg?|xm4B6JD_)l|Rxl*l!Cr6HY@@a<ZTCH*>@Q2E3T!R2VZFi{Twh6PXJU|?KT
z10-l%=OG-g_$>0T7~K26;Ci1)%l8Ko5{l;orD7$G1comuK<YIdFg9lJCQ`zM;*x67
z-qDJpg-h49R`!;3p<-t*N#vGFQoIAFxGG<UkE)2$R<o>?98)AcD9qY>F5uvL@GQUw
z#BjRA(RNua(X5q-5#RZ=bs@b{*6xQz95C)wKkYC2U&2pXl4rsg0dhKp1RdNilW&>l
zgjpv}n>x1I)hw^o&lGc~bVl2x$7@d^%dy=npHO93$L?Lsn!Su@xMxqKh5~FnWoV4K
z&}y9A4ubruV;=-@;q;#qrCE_bXBX6`*$5(kT*Txp$Hrk3xptF3?<lR?{b}8~@hXCH
zsKYH_`(j@2VP(n&lLV(zZgHEV>}Jdo-qagury>;aLWCJvR|Pq~&-Wdt9BDSIXg8U@
z0C$*$Q(GWiE6?fp7Kv>eQ$rVeLFPg@Beqs+s3Si%g4PBPab8cYHh?|~b}M_|ihW=B
z03Wm6xnN-rDMWw?Sx&qL+zL@JiG%?x@}_jN!_hWZ;l#hCIRAB0oL|)a{~Mil*N<98
z7UIK@OObUO9nW>3>iN!i9;$0~>B?xg%&Qy5Q*^X5wP)O6ejy)#NFi2RaD(uz5zA=q
z!*O6!RR^v;tqkVna~ew_rw@0H7>9vTkM(Mks|)bgQ)E$m+(??(&F58Dqi3#)dp39L
zwNnYR%%@+-_YpKv!FC-s72EYtMTRA$G4BgGnDf4u*5J5`86ELIUUW$%$O$}_fJ=rH
zj~E^d7%&yrUz9EGE&Iir`qL}>la~_3hoD%Ls6>>V*MG{O9IKLXbaTHLL8P9<RkBF}
z_l;O%ceHsT75JR2Vx%3l_<roCH~?I^v;%N90pkos2`fnIO2UKzVoW_2P8lLhJzYv;
z|4`;L1<dDc2SzLp=?YcQ8!)^5dDd>w5oq(kt@1do;#67$p$czdFA}Wg%N?)H>o#X}
zK+-FnZZCc<&YTUx0+#m_1#Z9{Tj6;UTF^85CidZiX?z=~93&pGa*%)t+odG2fOXrD
zVl+pQ^;9b5nhM>s_+h>DdDad<7QqBcL_%<m@qIY95I7%eeDA&I<{c6l*nvlbXoF7h
zDcUIy>bLe-L2G2EKaGeCdB;P~Uw8wpuNO%5;SNB&_qggE;!o<dz-B@$3*=n0Rcsac
z6Rq<!83$>65P=prSmCn>y_)EFr1d}v+7)eq>n!63)bt4MoXsHcr#dl&3S}wNsqZj6
zn(gJbQ<PiV3T~n8nRJaPZ>tA!r{@JU+o1?;-B*Ni!rj)8;676IY4!2VG!Ef7jlVMm
zTz?X+I)u{s`%d)h6n<|O>2(PgJ}3iJ;Z4t**oSyVu8z<@bc8sC3ln4j0f@EI8jo0c
zf=gL^KdWvhy=mZm?tu4sHX<5^eU@DS5Y_pc4x6V)9*MJW@O|O{d2sM7o&t>Ff$`H{
zhtRl>1L*V<t)6rrOA7c7^qsQ6_XkQpC3q$lok(PGU%n870FNbHar&0NNdg26tYcM)
z;MjRk*gSB`jLQ<~TdNL92!K;+kSqaGYIYuUJXy+51o$OYveC08Xk-29oBG#1{I4qM
z?9IiA-CI(=wH>|c7?9Jiri8<UQ=#20!|o)5v6>4lb19q=#k8)C#2>o)Bi-+^9kDGP
z)#^ewm*75}Nm<GY>RY`fXm*IG*q#~yE35hZcDYRFp^o_Ph<bpT`5z#0`bBqwf1`75
zN6PMFn}b__^x+P6fqvY9q(B_zg!QIV!;wQx^WF)=lclF?BwXgN!M8qHz5)+TlNu6v
z!JI{4cS2b(5v|eYxy4faX2-|20jD%x%0#Xhe_Cf!_cQcWtV3@P0%5Bkx`r}2nN<6g
zJfE95)j6r&PwRZQndM!5oYiU^u0qO+AejGf8Bk4a5S5OEP$^@(To<>LXzr@?(JJfJ
z-O1f@{vnZfsbFiXAKr_LjD)tF*Ak=Z>CyMnjiXlj&63W(Ry{?1L`ua#6gY03Dz@P^
zqr*JF2d$q?5#^3vwhenLRN-X56O3!6)sJ~51QeYgIZ@)^nV4f7?5o2EhxoXYTyd4Z
zHPL6F=0r@|4N^$iH&fU(U~NIkC82+tR3?h7NgdQ5(GeLmGQ9~T7zuVHZm+Mf8*?t^
z#&?tn@x1MF_<o*2?65&e-XG!P4ix;=&~jIBhho&K2Q(M=W708f8(ySC&;4HcTDv&$
zYcaNP1Q-->7X%d-f#rqENvNj=Jg6=~LL(Ke{gKq!18pw>w%N8m164)_i7iwXI^uZ~
zqB)ZKDUUTjQ)UB|CCor^!HLJ3a>+nG@~osEA28DR`+YdpE~Iy)D2sE}0y|^YYe<Tu
z41G%<Z}?Aqa|iN|!hYO2`=K=T&($%Q%r2H<eFuY%AMT|^3b5)>Qm17XgRdOJ5I!d<
zY?rHKu(sv(u3m_=j&T>%QNOkb;DerL(Ckzr=x1r!(Wpkw+Aj5XwafpIYL`^<YBHTN
zT2H}+he~s4n@g!$Q)hY}4b0R%bLnMdg8p(-)x!!j&2x%^wFn0|qDSY`89fbvg9D#Q
zGdq;gY=j*l^oS)0$Cn6l>Ftb5G+WAL+H@zv46bRa&c4@&+>h6|oA)xd`&S-ZX5c#k
zm$A|>$?1jR)DLUBs2`1S8m)Q0lujE|3ev{uG~T`vcE~VX;Gwr-Sb})poPCviEtqv~
zI)F4d6j1e)dT3w;XeWt02e9pP01sP&>a<!J!%Jvl(c#Hq0hsHd8nE3s^Qr_aj*h*r
z!(zQHT^o7jBwkMdFJBx1^%uv)X3+a9X7r~!{h253{i*u;Gj%QscN~f0!q<wDK3wr}
z5$h-+jD<THJ-jT2`SceBG#(oLc?)w1k=HAxWb#;lsuP45Kf?@(3K^oNMC}ajpk#)a
zBq{$A^S1z>2Wn)ZB#LJS&gk}&P<btS7$T88{+TS7>MI=Hq3Hht26&Dhr1t74jzH-d
zbS!WyNt0MidrcIC$X!5~wM>wZ&4<!!ZWK^FhM5RwS~1e~h<PEJUmGt7!cPS`nGMj~
zg<jFJ9>`&|`T?cWKma3n;0~?pH|HK}ktvs{hjpP{XUZR&0myd_HL+<R)-?qc;`Q)&
zeo0s@rFT@sPwSvN{Jmq+_Xom=Btq|q7Gw{t3$z&G*J937l81k8R;i6`SVikDLAqw?
ztEOlSD4gkLV&(GyH#o>)TS}CS4unTCoS{K>v;`g014!YHhB)hK@;Nj^?3PqkzjHPp
z;Nf8s!hI1@s0YXsPr=FG`Dx6pKf`Za+uFU@i*jKDw&C5qwq)_^fe9s97Rns#$p=7@
zLXM=rjvUetMzSi@?!BZh4^=oQf=QAR=lV!-Zx;oSj$lYM!VH7XPG*ZLB{=$lXCv1D
z?@rFc)|2GX%2HZL5P1i^rE$nR`Emt}f5f^c<$Y7ewwZpwb0*+<n#_yb@1k~gwQ@i$
zQn8>M{TY^|QvyA}r3zdKqYfzmWO!6gObhD4r*-r$?eHOSmNwIPYOy%CHGrSwDp1yT
z2-|rMLf<k6toSE=&t0G_xGmJiGTOAEKP55kmQ>GK!KAm?HhEM22nM>E$<d)rM+e8@
z#W?fIJb?!U;>;&O-eNbtU2OBjk#^+6=ziWEXGRzINr64a#XVCN{G~e`EX9f%B-0ug
z`SM?huo%O|TKZFW_uICC5Ia&qf^!$cM)PQIU;`}Co`6oRAc>OI{E-aE%Y54Wv;H8V
ztSQ+PPMh^p^M9{V3?wDxSn*<+mUw07ClkQn75!Ar$~Q%ZNK~xCK72(`O{xqy$^&!*
zv)?ZneC)AIr1}|@kyc8VR8udIzS?gNL>>hAZF%_3>gvzYuX<_F?U0zR#1YL!b(l#&
z#k_L&!hEP#7f82}z6M|hs5gT%%Z|ibUY4?m8jBu8dZ%IP=-10pXj6_Y_l1W%e?@Jh
z;4b5XDUZu0V8O9&Hl(9&Q2ugLmkW-FG-DpvJ5H$LHRjL+HvK9+p8SQ>`8!F0*18U8
z@22ihc~Bz`sa8F&J&Z8d$48Nd)f3Uy02@T)t-8$Ws6CocyKZNZVbM;eiO~}3QA?NE
zNV2q>|Ay!WXS3}3>$>y`3M&?apI5MzMZ**{5$<=~nOMmx)^n&p6ZJ&9c8NNpT=X@?
z*6h0*X=?AZIQnIT$^COt2VK5<L)u}L@tarGk+5aR1pR3uC;8J+mbsX;Tf6V#(Urx$
zGENo#)NStq<GzQnYs-KI8YixHWlH+?fyvAisAF-BO|;U4`*~aeMI@z;z7M78C&|7-
zySvF?qO)O{K~)hhN*n?O3obe_dhoqI5VQ<Pxy7@YxU0sLE4YsV8MebJ)>#kof&tky
zCZ&6_l&@H@t%U9dcjK2Ql9ufjBao9FnEq{ihlwZ=3(|%9(Fd?DTJCG{38gCswOx4O
zDxu~6Y5wGXwQczSHdjuzC4Zt!b|QGv-q5b;&9X~8!3$61-qy9bA~&inrHXN2<h$zE
z^OFc%myVA#-h^0^V6?B`m~_WIZVu;`@Qri81$0s)OE#!R4cQ%krLe5pbZy|An>y`t
z#$+kMh^J+dfep|11Nyak3#pEz3H0cO%?@RH-M99ut@;qa@6n$DOG;6LO%#q&F%@F>
z1sY9^Z(f|lk`#o?-_A0RsxcWiTh)3|t?+1kpy6q@(ZrEyr!F>3qSy-bsrdx%{8k6`
ztgXjp-$q#e`?`AG#o_<Rmuo%gH%u8&Y++7upAse$XqosTe(rOeS;vHL))l14)G?nP
zFJ1m(bM98$>D|QZ0yl0)jl;L{O)So%T0hwi^PBqu8?-FXO_?(-zU*|WaO1UuDTn)T
zUj72W9gkGa`fwla&@w&rdLO4OW2v`&XBB0USLLsrWK&&fT<T87%08|+{O;Nld&k+C
z(>&I|b}kw2uiYoEKeu+-wMDuMc+&bxc7J!t`0v>zgIJ<vr>NlKKw_ImNLN}m>DZTs
zvJJ|uOG{VGGE={H{mEFi{*(pO3$&U#8hC^zgA$ZAp8&fY&(LL2u}k(41C0O>se)@5
znyR1~(;NuXn;8?a>28zRr8keyyZrREZ2ZdKx1>I~Cim8A+4LEk$hQAOV3iucs%0jG
zwvo&H#>f;9eD~YnVm7L?XOgw}D}=$%p*Ja|Cp%PLNvwR|h2<2}8p}-i_4FIC-q<xA
z-#R{8$cEQODKO1dqW;K#a;9_fR@Ir6w;5l3a9w2@)eeK~kKJPy+Sbj=$(*5Cuy&q(
z^_Oi`_9uB|8FSCy@d>l~{91Ou!sd~0PU++A3W!>x4#mH}H|olSsTaClIN=6FhJQ=#
zfmr`9m)b`z{Z?u9+|K2m{_1(_S4vs<&<18i7@q(zj&dC7xT%l#0Z32jU9-*vOXc8s
zi&AD%<H2J*B6Xg24HjN(MY_z~phCnJgARTx1fbD=J;nl1(d`npnR_IE0zlh5wl0>g
zY1fA{uq0-H<l`Gf8L7|3ApXtrM!Vie*69odfH*gs;j~J!!TThRttd0cb%uDi4S_8?
zqHx6jXmuUmp+;M9*E9>^3G<PzxJBH+lydG?P#f)H=V3G_MWLoo%3Rn|<0)YjC@I#l
zh(ZrYGLDZ0I;kt82H&fm0(Rt;DE@go#bcESFpDAzjed20^pXL|KCbS`pAfCTh1tlz
z5>~1M73HPknmeZf2^bxcutW6t*4meTq1<<Z!TPi3`3P}W5+z794}dmtrddLipI|E?
z*Iyv5EP>4yVb8%S_xI`=s0@@)<x8sd#S!Ss<AYj~{JLG%|A*r8|Aa{YKbm>@x3MmI
zH7ttYj<O%E^>^qnRj{gr)$N{EgY+&XsG?tM>|uV`N7l69G};*41yFBwq)~WrB}jRc
zNfU_WG}d-f@9La9zU*;O1T68;n@x?U9k5J_R8$dcN_l;^t=(QQ?xoS>*RKr?jqH8i
zyh@dsaNTewJza|v-wLUOIiz$Ne|$GpIL$$S283YaMzfii4{{ZFeEYMI=2!qVQJ1*g
z5R&23QSK_B%<6*bx*#lP%t+oy<wkP0aaE}<l${AB-MW#K@xAdx$F)DQ->@FH2X5HY
zW+uPf1*hPdGL8B+1<+vj12<VnfKn!+&5p#(l~S$fTbdv^&kN64cAUgBQ$v5_QJ^6K
zO~g^?U}q2;w3k3pKw1wL(z;&gVBXzG@2E<VuFy%!KqC9|cPi+=Q!3~*IuWVA;kF6v
zwhe%yLSpor1jHNPxjAJvP>pXRfxp(sm-|5ykvrv-7#qQT)7D1d)N~p#|5)l(W?Np_
z3T$fx#}Q_V8vJ8uK~Mkxzuut-FPJ<AGa-X@KFEQ8Vl`?$^8+yXkw9kAz{tNo5C4<}
zvA*Nn^Y~JETGR}JvQ_!!g00GzoDB^(m<8Q7IX!C5_4V~Cy>*z%q&m8AG77G%x~!uQ
z=TFQhy&?*3(WaqaDG)3*{Z34ak!r=W@+mWDO<~*s5c<(k5c=Kb*3@dE8R+Zy)Gy~V
zRDyD;&#hWq^X@yJTamZ+{*w8dD(7B$T^70^yBeoMOx5CaWZ-+3D|#|GlS@I3|H2i=
z^?8p}2MBJXAcFat8jdzwIX5UZ1D}i5zvaqdnd2a=F|G?$Y~sU6T@T6QH@uBB9C!)2
z!hRO)Z}n+Elk#7yK2`b=(T3JiCb=LSrxGpdXj{7LlF)lWftp`#bvP^~o>=a6EpmU)
zw1yO4YEU2UB%ybenkDw}HlC-3Bg2YI+u-I3?joQT5vYPNh5Z?+d8D@vEih!`v2ma#
z=YDq)PL|>Tr91g1->^ldb6bdr<9fmMPbr5qktdp+#ekWjZ3=G4j6DzcE4*dKp|%==
z<3Q3o_~T>2aoCfTwoCnjU!z`Q4#|NPOR+=FpKUDxrf(#RiKUgHSbzmvWvE0oMv64N
zi4XPYFL2p0={DLWM&}V^I`B<B`e1FoM%Ee+2WB4fm`Czf{zk!6F5w9mNlno!CfIi?
zat&je>&9EeeBQs{(9XRs*~{WKp-60xTw5Kia^+M!XMQunTVCs-2mbF$jqncIM0kN#
zOVb~@5rj|3a&9E{;eGO~ZQ<yu=PP&q5vdjS+x-L9Uw+KcA^s>JokUahK%f@G;LEiD
z%Jo2w{SEKr{zLGt<bPLE?cWUVVrd5zdsQ%?u)lhV2ya*J)09#nG$66#27|Mj5epe(
z%_iwu(RS*3x!iaNNKUa-u@pBW5OIGYO{IThOaY2CS5F$H=XoPt36XUwI1ciSftfU>
z1@NcFP=$;9X^-7aLDozXbrB7LTKRu~({N!{_)G_H71~FFr`}?&cW9X<iPStlC+7-#
zZ?%J7%K}x`usPEgU%IxmUNQV^P&GcbQE)F9iZrgE%^#~gL`1LGO#un`wOwflcbqf8
z#Fc1THU=Db{&U!uMm7<C?L9%;V-F-^>k9Lm=`&PV9akf+Xq@Y|Uy|Fo`|_F0H>Oi>
zczPOrP?w(j;B2IlNvef;n0i{C>>~a2j7QC%Dpw!5a9ZOL23<fJD{02K9LWS@gvyc^
z+w^i75!{l<aq2sLI8{b7kL<o-wDxzY&@m6eKmOHFiEdY`J0>UT>cg@*Gk2%Xp=;$>
z?b&Cs;QXCyw%eriVx+eDEV6SZK3sX=`X5rfN&4dpNXl;ego$pttgzZ`?^*4pUU|>s
zyjEYT@Z0in-e-+Qo9lcsXW2XY12pZkXyd2fQcMNY(H<=v`L3ksXx}*gDy)b{Y#Y1Q
z>`(-ENAhYX%VYD7+A(6!r8usZp0WG<nbw|MyQ+b_&WQ6vg-AqEq&&2}6RxhQKj0K+
zH9^+d-w;=#oI}*d>1mNjql7#2aItukjWbme2Ll<{lgAMD4@CGo+6-3<6$z#KOOW#W
zqd>VM5_LvqnOqYLDk9n_*ca;{oKmVwB!)bOdZ9zm-b$;AE>VP(FSksTPLR+OpEeuN
z8RyICLS}ZL@omsaiF79k?{-cYvvwUzxAtT>tBNF*@}|t`Oa~{Z_?nb}Yi@P!=(J^i
zU-?<<j!83Z>#@0L_H3>%tfgFPAMOd2G#VW1pfPeKTte*D&Ea^k2pOcY&6pypkR8EA
zh{-FoddlK2j1D-;ShUt%(5U`&8L%D2kb`kg;2WUXF<36zT^EfuF)hI3RtUq8`ZCl8
z;j!sRmvfYx;eShARK+l?8r!4XsstXAkV)dGLmk$1B~}|wX&CzC`lM8HA(ETFo&^}?
z9s>(9)yGdDMhE~02j*4d`BR}Ik%oWr$DtV2NlJ(ZTBdUn<bvEvosYGMU*nYetDs)`
z+G#|7VH~}d{dQW`iLU6I`bE?z2SNE&<N1ZlK}a_T@XtVl;in~C?>E%#xOZ&-QG+=$
zhaQgHUtv*UvD*Lp*b@f&dbgzZ{l$Ka%lf;=#(%EI#-h!S1JA*(pN{c3h3-c}NY&*f
zk#9)_lQggjQs)Vsc;V&6NS6-p2VcYvQO*c8>S^UK3Tulvhe~!ftWaI`e80Zl{U&La
zKMa8)&>z*Il+IH<pj8}K1Tl=6n-K{Jkks`^eO(6>DXRVnR|7rtN`$~s)Z1`Dlyh*(
zPoEcA`+`s8%z+W4)SeYy1y#qjII0h995GCLP0V!W*kt7OuAq598B;AuNPrr^++UD^
zE`Gz{DvtV#!YKM%@=$@+N+}NGqeJg<ia*-T4@H|#0G@V_XWhrPMAT7)<R(1NQpwE-
z#Nw%Z7jSu-ScXFHL{seAJE|A6U9mh1GbVT<ty>Gxt|J*ZH{@rn!P>U^`yBt6pv_|(
zo{~P*KR944gPr&@FdfZ@1(^jpp4TW4tz}M9d`-?5MeL&MS!`qmw$lCWMo{in!Na<6
z3iU&m7H^m99<|>f#)N5i^<-3~kC*4B8>e-yMQ?Kct0{6U8{fJH)@~Xwk6+FEAb#9H
zjsLsZ<*%Dvmi`S`{{bKQUjtatZ`slUGWWEiaIDG!LG*SyGs~Z5PxBy;V?Rx(S$DdK
z<ZfvA{$XKts{P4=X$MaGZ;z=C`c!>gL6=xhY^{OS0-YQ4gkD_?;R0jRGF{B14=3s)
zt3?nlaK#C?K*R_41k6ukPCA7<28>X)(p~n=I%*ojA9;ghsCIa<T9#&n8$B^xT|C==
zm8Pa(<e76Pd#CPtP)k#`YolBubsoj5(U=`S!%_g`wgAYH^@zR0G;Ug2l@?1gz^l7l
zxcr+y^~tA$v5B`QFWL9|o)=4}&6IkWEYB^+rUazt@l<0`fg;FkYMCvj<w+O>-cqn2
zgPs8J_u+baAR8cYfuHU|x8{CxJ`%;fxV>02)BD+K|JwGw+jTCb*`Jv+=QS0uOwBP-
zBcVr0K<<119&1So7*bC%ntdiol!FKp1g5BkDs&9}PEG`8X{oc{dCtiq=W(Zb?gmXQ
z!7q*^z$V}?Zd^#nUagiuO-&R=34@Z)v3hFGwuaP1+0`ydxu<sf`3uI&<!dK?9{bo(
zb=kxO{_ES>xHF-OZ(A~`qS`qg5WFk`*OvC}Jm#|)&y_w0H8hdL^}=x24MFH&RTN{7
z&&sbPF)Gp=9LB#%+BLWE-EaC9OJy5o9)94OFIPuvy9g|TW?cslBMB#|xzS0+<gZBQ
z=_*i`QNIIyv5fx3Ns8wOTvsjt0k3o_kNp}ViUw+ttpM)X)7ef(@p25!cZMc=6SpWP
zYI)99{%@^*GyT-sm#5EMd@4RvF3Oz(ggB8;Knpg(x8WQRZVB#ulu`|eRf4|h5l-HN
zuw;Kg<peQdZ&ed7GvKL<S6WZl_9CtGPKj|3<`~SqvLj)t#(^vA^Je00aiZ?2|IyNc
z`PXx1{&DP&b^7xu9J_$tdT1h&O}&Q#kiS~Bic4b6R-C{cV|LT9_zB&<t?ZcRim4Ar
z?+sRXxzzY#Q7qXE0P7*F2<g(RLK*Kwzna|bey<KYhX^}u=bLYgP2H(GLxUBV#Yq>2
zhrf^aH=VIh>D>4U-%~%$wZZlhJ1AGtdIBPlg9>wy-~6vDUm(xI7-*-{jQKEE+;u@8
zJDXabBT?Qk{O_TPej1=XtHah}E77kPU!wD<K9Pgp2vzhE<?#59oNJI$JT80ylvEaI
z&I(Fn*=oPQ|Nm_(7~j$MQ+pSZmjjkSU?$~PQ@!lk312lZW70KRt0~k@uGXk26Q!ek
zkj}&Sk+jx@Q9$bN?q(~wC$SB&Ef=c+^=Z+Biyn&7QRdtU6Th(2YU&9gZ(MwhcKN^4
zz9HXvubb*Y>a->UarZSl3_gB%w-jHEJ&9AyRjejWw<9aE$30Ev-M(>}(<56|RrzT7
ztkka-N9MX7maju<XkaD!aJ`*ub!<dA+PR5*m|P4&!YR^tS}Or^29Jmy`EqUX{>X8>
zw7^<;AJec!<nhF!B?4Xhh-pU?cgNk>T9P|HJ^xeVlas19ik-8mpG>(H$RwocWC^6c
zM8%%I)Nnm1H2XDHfOIo!I^b?nmj1#-@}_amIXj+b-LLuny`H)*9aAlQ5@<9vDe3*2
zDMt@w&wUVhZa(c^BIfB7aE&HMo+envwR!CJ8+*5_ngCNOobRI^_wXP~7_wdOpn>$_
zmrJ*~gep2dPS(Iws2m83AN$H-xGLD0m+7X+POB!SQzn+XY)sYD^4(oZknvMWEVIho
z^V=1}-Y1)OT;6$ZrBt(ss2*JPSqgU}t_U57)c<aoG)SO2St|D!t#dN)W#@UsAF1pW
zwWM@AhpU)ITpZgxa`Fya`lqYGMK4~wY{%jBaMB(cJbVFKzt92e>7q7JYbk<T9m+77
z5B=2Ahsr?PhA?R1dZi7RE$McIo5=gIjo#fcft;$hG?^_M?)W?p@7or(cgfRpA=}rV
z+;HXe=4o^5is`9JJ#v?6^&i&SfH*pchFy&dRrC^ZHGBt59+(JqJkT8bE0Lcf*&q%>
zFnJQT4V)g9nuvl%*89J=M%G>cHuvU2ZX>906Xmem8KjBCa&$DU{xcb(4SM?p;3!`c
z({JI375Ml7!7Y*u`t?49AR@FwPN^0qqxFxubFlPEbr648ZI?@fo-=|;J(!9ABZvbr
zKo2p@pNYC*@z9^OV2PAhje`3MHvu(S0=$z?sraiGA~lck?NAm)QBRqiT{X3oJe%44
z=v&LZr=hPtu}qdO_qIBbZrs@Z=)`q7+n#mmoHSu@%xlPvU&2Pa{ek9$)YeVw!#!0m
z^;qZnXG=x4AHumWyw@GIt=zbXJaeN~r67CkGGC2X`;7K1TCzodY@pm+`5x7Bhu4)x
z_DSgNQx6UymvLr)iK72sE~)u<HLB9vDtrIg@q1T%qxL?16wz|K!|s&KrPjcO26XeR
zgQI1W)3!ltP8(I`r#92HkY&=157sn|#(H9LsE%0=zHKX%b(?SimA-aGh4IwlN>h%F
z@PQNm?FpKKQx;A-aJn<Z(oi3+H$^amy)Ec*)5Do8Gr#TtuSctRZp%+<I83ttJb%;&
zT|=v`P{oGAlyIRBt-fh3tYDHdiD#rmeh(l`2g@aOZSA6E1NeB+v{sXKzB<@5mT5ZK
z#qvX(%E`$6dqdy(y13*l+^~MGzuDF1Di4Ih@L_cBb~^+!IAkMaZ)dzy!F;t8Jt%KE
zk*MPfTSRf?w|45hb5C!?x1XDP@Q<vAPXf2@f@sEB_<+`QzZr$O&)zN7dtsQs1iO6#
z8}FtT&qdK{CMvrptA>9+J@MX$lbwg(N$Fqz5%OVy+Iix*KHOL)e+^ooT?obq0+VOb
zy$otRS{+L3JQYmgj4cA%#+nSvaq!)>Bp+eDt?o!yvrjgxLq_gtDZ80@V_IW{)ssW=
z2{OOS8RIZ}yEccf7cm2xH_DmC1bKfIZ6~cJ1hpaX=JQB^fdyuu3+I}#96$6Z)5()r
zTmA2@ia4%$IcMX$Y4%4K9rk`>`Z>b@=qEw>4pM<xaBDe#b!BD73G(#gK66*+;sXS(
zFIyOI39WTcD}0@~&QCYXU&&U`2}Itv>EmybhNVTY8{PQJc-dHVZ1aq9#mN&6oqF7)
zsk^5vu473>1?}#_^$+}7PoG;W+OciuGhwT9eg~QputX!4#&f`K%~Nkv`x$hKn(U|q
zW4G3kzc*2&F=?*Do;Pe+uj+9+dE1Mgn)sdztD3Rq(Ed-Kjn;ZDUr>t43FO!_UZ1ZW
z=lDUzx#7{XKYElW86_tNe$EIfmpN^e9je&4B;^}-6Oc6r*yUJC7YNV%X*(d~8}#JP
z@Q_rf8n}1XGqw*mra5Ko>$n1~qj7fST9x0eVz#L)7c|!IAC-YawI&n0)(^X(;S<Q6
ze-`}&d>z7{7!x3z2H1<(6<phKnIe~b_MFW#m!wp4o$!pyFVAdWTkh>&H8Eq()H&aG
zX5lTqBZGe#j=UcjJ_M;xvIhStw*xaGDbTxus~MMRaw+(hFVS3WxYSnuN7l5^is>2i
zwFz@szpt>Wq_#5nYpkp?q`0S^67IAlq_J-XbZnRV@tSnn1zK1VY667UTmyb2P)jsP
zoP)v#E3*UV!I&GpcMP0<78)_eI*8UTjdj_}gSuAf5W%Y}yQ-n=OZ>SVTj#gshONo2
z_}2Iyhpq~0W`H-Zf9o;^Mv{aDWswp|tX$gLVBsWpLMMtXqBfI8=;4~~g*<#4?zXPT
zo_b=wf9>d9F2|r8BTT_H61NFQUOp^3br#NW39hDp=)+C5i+?zeP4MUvZrnvfm!J1O
z8+_xH>>f{Dsj^o?hY?36k-s3F4)_?N(9?+~RA@+<No_uqh99M;(6ny`dxrNTm*tsh
z+e&ME)@`U-SH2{0dAU-j>%v3VT<MC#Mpd+Qo2WG>lWI&A1q$2)_O+ym$6Xf}aviqb
zn8@qYwQ)K&zJ13mnF|jl>Xw*tZAjm8Fz=M$UO*p8)<kbTCBbyIbMy_HT2q;?E#sS3
zc^^9XrhRr**{n<6OARJVWm({8Q2NP8Q{>^@AkMmIf?`WzVge?(%t=?dW?^ulRyb)V
z?Oo;f@^ahpUskm4{1Q9<_Ui3MAF4;#eHk)KoX8WohAhhuDKlf(#7oKJ4urRq>I&}P
zIXimN7Xv?*(VSaH<@~RBXcM#0u~E_b_p`_|47{-;I^pm3@9rF>SsL4&=YyBmT6HP4
zj;2TrPSTAvIY6J`@2>HOCFh(lq=Py(9<LCS8Cbg|X?*YE#+yCww;p5S-^#Vtgx~QL
z=oZcWv7*H45L8RrLC*4bVN$I9m_5R~ZPCf=BEQ+K%BeNky<$&p@B|It<K7#ZHjkB(
z@qag`l%}M?GbDit2~KUR9?9BtmBp_@kEs&wf7Z5?A)j0kQZB306@3VsM(3OctnUzx
znf~mn@PqNhpvX3Ed$iQKx*ZSiE*^<)eYluybHV97xgp1bebO(y%3zZF`%1ymEqC<|
z>d(j=9&h04IquwA<SnDM!(q9pBDl1<eit=xy<-*D^PzQIQVk&$o#Eoyn4w2tt-JF4
zMU~7XtF7-oVHTr`=8j)&hC7wNT{c3Xj*dl|+*uviI8AU#iOmAt+5j2bDZH3OqEDCm
z23<>|x$n-~H65{Z@yHI``SQ%nti>)$5MB&g4@m)XZZo;MHA(BpMQP-*uZ#AVw9bo7
z*&Khc;albm^=w3Agrg6R#NCEXMf3SJ41o$@F}4%>@hmG1^I|A~Q7L3V02Zw}<iUVZ
z8)|2Zy2qVOO*9Mf4}0PA*5GYSalw<Adv;A>{x!<ej@`YgAZuWJ^O_I%*GnBWUP8qb
zu)(R}Fe#OHJ9mJ6Nj42@N|b$H9jRK0q`h9N1D<=K{q*~17vT(~$`W#2i3^8+WpT1@
zZ640{UP<_sAkhR)3}W5plwDe}brvpANy+=rg@+@_L^pk53bux8la(@;)o|b8vGAsG
zafjwC#^FCOt4V8*zv~TCmtTI06qYiTyM|?$das+}Em&5)71>s@cxtlp1P4V`rA3?P
z-pIdE__mwi!N6&S7&Fa_u^G+r%E3ize;AkNtbOy^=1|^Fqxl}lG72WI#ou5Quv~4z
zYvy49*LK7z#+$}0#(2R&u4OGjMPQav&~+@PZR_I46yF8A&uy;P-CTdf*CY8yky49A
zR%;S}_3`7A2x(uIn=ZZa`CTSbZRSd^aY$tn<f^!`^i=a@Ty*<8%Shi@NuL9q<dweK
ztWhP>Y(<+x4q4s-Lx$Bk4IK;whOBFxTM>URnNCvDlXtU#O8w<(nq9@OV&BO&6z9x3
zx4!bn=*frtS5xnTA*^NKfsQQ))bjx%t-4s=+d6{CN6cg*tm`g$;3VS0zR%Tj<NNlW
zy0(aAC>?v6GH7CY=(!57qjsh~cvZc5E##@}PlZu2_tjJyg!P^k<!|nUUs-Arb8W<h
zS?a6n9<WKIVKedx3omk_4=2)!mn6@t&QY6vzI<BLk{N2>ZkKHcuRU?I^Yso&I5_^n
zU#m+Q`cMijvm+SIalG7sXJsvM%9}2oo#MEdy=-IbD`wE@nHMjn$`Ef1d2Zd~R1H~-
z%gO6(YD`BpFKsH@dU11pd1{Eyk%Ju`jwRme54)s9eJuZt^}~&=?fc^v+r(S?ea)=9
zMkLknlg{5gI^H5G!ED+hqr8<;StLmCcV`Km&;~oUB8LyLT_{^yc)ppzGg2bASlg~z
z=)7FZa}9C7(V=WPs&OYe)lHqNRoCW<PIk#kouU&Rd3wU(clS@7negm48BaNRT(ds@
zO9sf!rlVPu%mDs)=YaZHUhvRG2wR%veWYF_D_HMXyR7=Y+8XD`H75u%mR=M)ZH9NJ
z>F2Kv9>{%L4c^S5xMj#g#R-yzW#LJMSywSdff{`+q2|4!CF_V^``yRkACGeiPkys~
z^XBe|{f&)_o-`J|LkBxVeY_v-9~DN^T-~%;hpNwBa?U{}aveUnk}1!dI~vxnwM@(8
zX6_C=m)iB>Ui=4&hegBlb5Sh}#TaH}+lJ#y)Ps6%2kzaIl;l`)C-9hbZ{F}8WszMy
zLrf)+Xv2!H=1^+XHhws>PyV%+ehGk*mM<pfcIHgj>7TadQ`xdSTfp24s7>_E!n=r#
z+P0cd2QQ1m=Xb~HJ60%moG;PM`<BEx4Vj4wx79^c4Kf_P8u6?}I6(r?=R-Q76_T!z
z0Zef(yvT~;=D<pf?b;rm_(iYw<ANWrSI@a{)8`kc4@)I=?i3t>z?~ckoL%Zq*jJ+{
zGRq^~9iM|v&s@i><a^9*M<a#mJcorN>(;1$KNyA>54#qs+dWKvAOt%xDCjY(u%!S>
zYMCKh;I{A?prpEug-=6TTkoF@I@$U9__Q^-d+*%LbH!oDnWZVf{5f+Sfr6PI0F=ZH
zV!O6r1-11%=yVt+0!kmwCrS-pP5Q7^+WZaO3H2&-4#)1VC9e80pT+(%M$XXc;foIw
z^zq)5SVZuEm`>o2sQ?ngtXfMr-Q{9+#0T1BCNX6;Th7CY-)6G?#?yBXw|=j`k@$AJ
z;e5za48_Mpw-K<i4hbKlOnMrFXSUAm0IAR*fE89<dLfj#Z)$;H!lC2gT<-(c8!EMB
z5-iVXj{n_A{m`;GGR`mk2%^B%S(W-WJ}qjQY~g9?hZGn@%rUZ!P_EuCP=ydz8iI$i
zAYSY<@)z{2Ur_z+u$tUEwaeF}U8}d@mn77pml0XP5h2*%(TL{>XMC&O>KHzveX7&`
zKwj{~>VTlrP=590z55NmG^EM$p^5ps7KCT0MSeMCxiu%xVo|E|Je>Kk{*lxvD&~>4
zY6stW{q8hzp&>5NaM5N6adjO_1LKj{#YmGHUf6|x3~19;<jEOAg9bb+R0@#zz^2fN
zoRk<99aGNA1xGdBs!uB~eT&4j_J0i8CP(E-eDsM*h{seNCmT!#{RoZr2LDYb#0$?}
z5)i-%;BJ@tMgQAK6T_4k%69+NF9WCOzWNYoC>y*tCB{=&cl*oy{mKn<JzYet8p~Be
zyb!ny1unC*4|n`4ny-WKn}7<0I%i9pU9V<dh%i8^4FwBP$=@ll?aB^e#DPAXK76Dj
z3{!d<T<Wzs@W9avGNF^RQ&%(fr(ae|kUV2DI^;VJ?ShCQ)Cuh==)=jAB<>#cT;h&=
zPq8A2)A9>DTEEk-tZv;o|4^FjTz}BRSf&G;-oKek($>?Q(yA^)t;f0LQWZvWG$p@5
zzXAGbpOpL^vx-|sWpV!c4basiUSd!NjpYc54Pg@b=a~kD?cXiu?l#!+bepRBHKE1|
z{<bvi0^AA1P$1gdvSQHB&VKmY?W2LK(~T=9J581v2g$X6wMC#DgtKvqZFAUTswc3l
zsTU3uRcb|^PgpYUd%~Ue3w2{+ysX@RJd9;~0){l89ow`sQsL@k&J@!Cp&IEWc{xit
zjdZo3xh@Nv5<{n$7@C}YSFy!z*2oX3YQ}Ik6Vo`>!Uu%K^WFnRld-J5_e?@=0R;lq
zbjn+_EsE`18=RRp)@4k3{&9_`wuQNGRnHsvZ=Vw%xFhzOU%*|tlR>?2*`}EBSzpk>
zEPIAFZ=`pPu5q}?kP%Fxmbrv|Od0vA$k}q@myzkZX_?l>%VYxAn9qrD{ama!f}B?t
zF3gR1=C-!TCNVjv;kE%Iq5g*N3jF=&bD<zFH{3n7#-Vqq;xq`rzdJxAq8ZsOJ<Ovx
zfJf=3UW`tba;T1+Dp+^kDgNf(5mmb?7MM+(@St5vpIHW8*Z4o;bwSfPOl^*u;K!#Z
zLPSaN##^z2FD`ZtG2Z;x;2q^m=hb&+)fYTgE2Rx(m2iTXU+I9bKsOCmWHrNc386iC
zApnHs*4uI|$K3Y%eR)~#?(9j2DM3CkB)5lnh}}|x4*Jky!tPx|AB5?1w*_3SNUqIh
z)|v2XUsXTV^~cLHOXAO+Y4|S4HXd8=aMD!$9u7OHF|55rJ2{BI1j|+NOMa`J_GQ`H
zx7S8GK3pMep*|9f=T4}JI+~NNqU^DfscE?Fu6q2#bv1RjS0wB;Qon8h9KzILA|oxK
ziXAGQBJC2e=RLif=EV~cL|^7*q!&66=0bdMCUttCk=8Ky^h~;wIka|>_9RNThLl>V
z{;@3HGp0l4Yk89^^J7n(kLBl(sIO~}WE$!28L>v`^RN-X&JNQnkwb^r{dTwWEU)Sg
z-|lgcsmQ1r`?{pv{^fzVD@O50bzO2MU)>)NWHKE@d%aJ8wq~QyI<1@)`O4h{VFW3H
zHV8~7(&XGUx$9VcICFQ$)D-#Gg%0(XAq750cA1f1bN1TlW1+a{_$Kr>{Gg8o48>}9
zNaK<EG_o@KHRCzjMATw^M~%)>#*h_lXH0m-Qfu^lc_e>j!o%91gpkSn+IA5k<e2CB
zC=APCyuI>lo+hhfOF&N6CdUVjLaHrJFp_I{qKBd+P(D6Es<E`IsrcbtyE`B9Vo?LV
zIc(HprSN2Xo*CQUJEAh38#pFp!_=qOs()|#L!NSI-}$F*a@>)RY)kKD-emE&#l0_T
z^mm!RwI=NLl(GJ6T3$*iIiyo$$#O0<>tr^;+f6Q_MJ0sad$YnMIoZ!+dHRH0T${cs
zjTItT3OK4KjmBLWFOrN{q*=30K0oCCy!qkcg6P(+3}VtfRqhj?O46(?(N8XoyXcv2
z14Q`H`EB+{uo%MkC}j&)`(+4?HMP@sWKDS_bHvl?$tE8S&utrSpV;poBZXp<XjkxE
zdfc;6umo#KV2vO46F@{W2!!n(GBa@9eYpR_-kZlmxxfG8V~HZBvXyNrp+nLpr5GtB
zNgGqPX%iAEYqnbnB@#ku87-nL(_-JI>}__Ykg;z=W~{T`zAv3Hbvp0!KJU-@EWh8Q
zKPEHNJ@?H0y06!DJ+J4rEuba5V)$f4Dm-b=mwmlDWxu&wc0r*0J%ujo%P)m`z1Fl6
z*Z;xKyxoEHC-HAJu!%FLMZS*KxS^qY$u#=6)dS|{dTYWJke~O}sulnEWuqx_#h^z!
z-$UW>I2vI`9sY~@dABLweK+&qONSlu*F+bc*@TF#J(?D!nUyM12+b`Jph*>r-&3N*
zTu9!KZ&n6HItx=*_#WHfbkRWi;lkScc))UA*sS5=4MI-<So{b&zB5^fx}S7B%QZ%R
zHgC==6tvE4l-RPzx#9V0S7p)cD05@IWgkVdMk=jta8*>BLyc~zQQ<I{fo?PW-uwpX
zI(&yURLAcIv3VX&A_@SQ;p3bS{oRkL%`Zs1_V%s*OZ~%Q!A}$g{Tux@3n)8t*xcWQ
z6OL@-*%Jc2nmUpEzl0t{j`l4F0lHa_8~6rd^5d+`EiB5PnI-co80_D3UV1OTGDky~
zTI{Kf`&uf(j$C^?ASk|BQPb08+mnyNIe!Y&E`WhyF2{d{;vRDR_LW?#0aE<d2Syp6
zRgh-6uEz9*p1$g(H-)VuVJRm$A<Gmz!_6-6?*n?c{&M2yPujC5huRz8-j}{}sZw&P
zRCoDe#I;w9hI~;jK=T%=Btn%t2^+!Av;};W`QAyyy>MS}?}~{&x`;Dq%GFb-ks8$x
z1X?U1zE>g|Q*F76dc)|Gz@GE_w_G4w`iOG9J;f0F%qCmt^osgv9dj<Lk_6fH|ME=e
zHt{-i&m-(@h^j`~r!9?SEpH5Bjb0sbf)&DOy6H%IpF}r_W#kSrLgi)sJz`H}=ys6g
zl6WFRkcBS^f8)~t`p?a{KZ(#>J~gx8%0qae!*<wHmku<Mh$ZOaQ)&mx<frEax9f)_
z+bHt(&`t7fcI4JMHh9R0>CB;D;a`f=_OVZV;I|3I@w-qQt1^@rC(O@zT78a-$?ads
z14M&!F&Y92{LhQy+-)nDoF8{sdPgWhaVy61ho$~I3-rS!CW+sZD!ACc<a39zRFbsF
z9Uj~lqJu8N*|R0g^@*|c^>r(CUpMa@8W1}6-OaZz_^(`J_{%CMmBG|D_mH%Hu`_$P
z02~MIDCf$jGJ+AU(CnSt0C7m(Uw*kbYCg7X-Od`_fFs5#AHi@IbF=P9=O<|je7eIE
zHfQBhHMvGpFaB4);k#RCwoy{3>#60cuq7i8Y>B*MT6x0CDgTxA`9TJ%wyGp@BX3R1
zIG=d-V~bM;i(b8byw0F`CwY`Fdx%5Yrgp&(H~9~8{FgiVoY_hL^ERs@c2TcWU(6S&
zgSr3FeRSb|A4bv!*YhItR;HAJ=K6mA6sKy?asNHpOsygQT%>mS(<hnTE2KN`y<!lJ
z{)V(?j2bQPF^a(44fprst1&iC6sxY*me#Bc#CK;$TSu&s`4nhSF|Tss6zBf@XI2bj
zCe%5;(XE^sn2F4aIOD!RZ&UI=eqWq5R$p~Zu*j=@vF?Q!-9T|Of0k^N9<wNhs2*vd
zM-2$SvP5w!;Fj&v{z3U`xYQ`r8S9XL(_^Cey}_HZ4}B%B|AdfzXD;fhN4Ml_jSPxt
ztQ(Z;QrybcPJJ4T=onK)wzOakokdE;%3rPM@#DY$l+(!kU3Cf^@gLPp=k0BI8v8#E
zg}p!Q-lMp<R8ynD{a@?kpPb}h>Z;8$%G>gu7hLPcggmubW>C5GTI>G)Mxm}Ta&sXy
z;0WL1JCwB_CCRpb&ocbQuTXmsE7ab1?yy9d*G_5aJ4fbwV$A=>M*7jN{-9Bka@>RP
z4ywnp!#k9MR&YDl_9{Pr;J4Y~IIxDRONv(q=W5fhRHC%)l{szQlnMPy)$^TH{Ak&3
zND6hDxqf)K)vJ5uHr|e|&70f*QN<Mb@X&{Z#>EOs?+<0Zi&%5Ee~vv!MHDMP6+5SE
za5Qt@)}ts#9+7*Z!mk(@?B7_<92NE5_bQEID@^wfJHF=a=(g|_X#NMc=ud>9)76M?
zc${2$r9&`X|NR~h*;ftFKP%U3Mt#=#Xf+->X?^l~r-P!*+Gc?s4x9a#SnUUy`U@*%
zzkPp|ld7!vl@9JrQNjlOgbS%T|CkB4<O_;#w!6CKZdBfftVV-A++Iz4PjGfzA;W*|
z(G?bvp8=vZ{>QB>h9T5d=k`RZbpAr#(~)(aXAhJ93s~_dLeP_@snBC|-poyJ1i@pc
z_}E7~*(+>5>D|_d<nW^(!y~+K6ZYBvpd2=Yw)mgdjE~RRdNj^dDB{dH=Q6PdV1fS?
zWdECDV6&q-HLdB2JTURAtJ5N!?@Vy#n62aq&ZBgVl{D8rEEU&ZwQkQDV`jSfKdXB}
z7($&nuIUf*yxLY9tmx?8ytw@(h;#h=zR6!*s8x-^#fHVGIS+LYT)V|123jFcb*Sc<
z{K=4x2f;mghWLb_RmTe=L^hr0_1f;xKd<79&q|*A5Aq(u;&%yb9uf>AryL|RH;~PS
zgqSomEfz86sSFxw*a7dRN)r(Dt-m{R0SSOG)t;C{5d4AL6Nkk7Itk*}n*5TJUwh}D
zQU%#)LE2Nah)=@1%r!o@Po8UfNpOGGmyt6yvOWvHh*qI0Y^84_pS0&hh{5%?vk`h4
zs9lIbTl`88c>GQ|eW%>~G_8V%zt?a-3N254@2`{mpzwdK2`%=Qoc!85Q()&m&?m_#
zVe*r9nXRNp3Jatk*SNbNdN#?X=Yt7Sn0w@VT0uMN6>DWb#IG5o`(L>GR_ZZmI^!uO
z)`g;^4T1z3Sx^T{fvAAdtOc}koEJz9cn$dTD*mKut9>A?eGdt~Ff@sP_yOmv&r;a}
zG585=7K&wr@XY^pl3#1`OHO|6oqtRf>@V!a9%{nQ`M6rnU}Mjw7OqlAGV|*wJk-zT
zLg6>lK%i!+AjF^>v$@FCNfo}Je5#zA1Q#3t4x}f7GS#t7;Qzow)2j``DnSnh!If6|
zR@*5_@G;zY(ImnKZv_UwsJ92P0tX4?W~kS}s>@1>x*gu`MjHEq9^Dv+RQp!II6WTf
zf9o#~NlOweZVqYeGpNEM#$j+%J=Vfq3FKT*goW3R<1RQy?O}zwlg2^ezpjdp6n3-6
z2&|Vo#02@M528UR{Qtm>nEDM`@nCY0-v$z6JP1gM82+#<FoM_lZi9}wRlp*0r}zF&
zbkzQ_2hD$z#3{EsHDTM)B}4u#5)@)FF0k=_X3c<_nE#=d8!5Xp7kmf|dHGH%&;Ew(
z8@&U}*Ucvj3aY$yAJHO<Vxi#C?4rJ$yUCdAZ+9Y!*3A>U(yMhKQd`Oe1LW%kR6@9d
zmCm!KN|b3leu{my5i@x7{$W9n1!tPx2bzTJdjI_ViNJH~mvvkS%bMbMXoll0w7w3)
z1=AMphCC@>2M4_CZ+s60N_YeN8y8C5^f!_kXciQDcX54#pVr`m0v_772fj^>m0_c*
zm(hz@`bQIJ+f7Jk&p#*)r)OjgN^M<pV-0so*3)Q_^%w3IhX;4b(Mp)xeCxE(#5^=}
zOaJ&iVTyX-Nqf7oPxsAlIi1*yD3)35{jvG!?VUPHNUM^xQIAG3-Kj}b!|`kE8{Q`O
zp~7*ZN}d;G#45{B?rE80#>B=&8~0e~1sezmMmaE}cAD}YnLuasl!eK7FRIDoZ%1*R
z(h+pnz7NVM4S&5dH$8M8>)CLb!iQJam#jJe0pZi3>w8|mmC%$>)`*~8K@}^b=AKfw
zbxii&6uoc5-qH@2GnHE&%FteIY1l3DMml&5{Yia1s6k83Hm4_$b^cf^_S#mdv*Sj4
zc;_tI9j$v~Sw%z3wTP7~_z)p{2!xL-cC~K}DJ=xeJ&@)jqFrNNd~ZE<M}c0<<0nTJ
zcPides4}n|c$&Dz;tn@rg^#|KYbr~DcD~(78ssx9Qu{2qg|Nt0@I@*0o_1AYi?%@W
z!e^G3n-h3@5=FVFsLSwr&=O2w!*(jF-T0!5ka3YI^8_KS_K|B&+U4u>niXQCyq+z(
zP}j<>4|1jF!^>#y?U@%`!Y<2-h2xAGmW?dfR8;zQmx0uU>yqlGrmyAXF6s$+D>L`Q
zpGd+-D5R_KJdbS(^?3&pujX?*E2*SMB}aV8xceyf*i$#_XWC8S(MLApA4ArtKy|`A
ztvMb}g&Og7n3$3kd3Rf{8_xC0TsII}7iDd7ME$pAmsgwtJ^Li4x()u#2S46h2_&c_
zu%AsLK9Pw@<-TIj`vdStXkd(7hrk4~K-S0g&zWqmFt$)qb5fD-N~qki3C=Hr1M9fD
z(?Gyvy3z2z{TK7PU<2(Pb$uH&7~WRNO~V~x>K75Do08oYXy)f!wa>(^TzTvG;X~~E
zn80(~?=|Q6C{ugdcVp)kETBCs>xy#R@J^q(^7I0t^Fpz#rDL-8t9@Q-%jW8Ok1$vD
zlfX8i7;96pC87J$E#0k53+7jCcj{_<4^>yKT3GsM@vF!cHmmi{%<&PL!tfwLz6{wY
z)d8};z+^U=?XAQ@dJT&#>6%^;WxYbYdy9S3C4H}i>t)^RN*0GdtJv#mtu%KjDp3?u
zL_)f%h7q{7%5{XHeCrBUr&0Mr<IRfBo`}9JxZ%s84M$m*7q8!ZZNq+nuB=zm^q33r
z>@<R%kQ6L_@^;4-bfB+v-Jv(}c2qO@;ukwE1-RQ7buA9d`w-Z5>FBObp3$6O+V<AW
zGN<4cX<Cu$KpRuvC%U6}{VSrG-pkWTqGp$luM&Nrq;S||{Q8r(qDtnxBdGMUK{2X%
zYgwrix2-Ra)1aV?Rn2yV$Sxh<Cf60A^98JJ8%^kMDIwXV>(<92*y0Ze8hP7VRhw#e
zQaW)9RbctUSaS4psaxJ#gf1;_eh@myd+v1cHNJyq=E+^rkYubMKS~?1W9oRX^)DH8
zjMhlM^TFnc`7(LA>=3@svR~YGCMRcCti!B$vVx~KR!4iuRxU#<(zl)<l4p?I?};5K
zp<J_|y(=wp7%LPBU2k}r{$gj*ml*AqHm+CA<vaP8v<)?F-$%Q^G`GX@dkaKI(tc~{
zZ?G@bJOAZmMn<e@goLozv6c&%t}QpjD;6Ig-A|A^T_EY<)P@{vqj<fN6}xI3eXEUl
zU>U+->j%C^Ct`Ip@<JoLg#8g**wlm5GbaAdU#=mX5Y%bu#P|ZL1njGnE8*{KG-x9I
zoeeGs#ox(04?0a-POx8qK01aV{r%hAi|P3n|JGkIic?W#2tj0nY7<Wh#04wbqZ-Zh
z;(1y;T=Mht<V9pX^&}T}*c2=5IM#KoH>=;s6B@igy}h3%N3jCVrImEHc+2uSo4}&o
z=dLJx>e_O5e3P|9xLkOYoEsNe8?~?63ZgtkN}97aQTkbW1rGYfL)N0PAs?jA;L>)f
z=bR4MhX~1d9k4_&@{L-rC|EGxO{~qd&&=cAUIZ$t$T-NeD?k5BRHB8EYl>x&6Svu?
z6S_{P9xJ;QtzQ+Idwl2FB1Fn{<i~#HjuhkeglnMn4dGLo2ce_;KJv;|>-jItOQhwO
z8{X;+56IhAvw?SR%zL+w!aQeoE*TD+@3DRS3k?_7#L~&9g&l9b<ti#+s-ba6uTOM;
zX!*<N(2r+)$+r$c;!;ZyzSs6)7lYd3H2zR;^LD}lP!IgcTKFziu<(>jr@x>Wo3JG6
zp!a=07pEAms3N7*xu`7`zE!9+(iO;d2<1JOcGk=3tg=;_na_vMdE7#h6`x;j9n4Uv
ziCALdUC6pKXGI9YH+diSCT@`$sc~c+bFsGF=E~ONPE_^6<sKdN4kyyCRkx1ZT)A?e
zf$l1|q`=0-slDZD47D~@Q9Id<l!C|(A?^bk@1~d)xjPmg4RK!T+Pq~AKd;+nQLZXt
z4^(2@2z87R$_CwA@vm@m?e49N(SCWc8%n#v8@KiB))4IAnzHoW<yquNy6<I7JGKxR
zCfRZo?7kz^u#0&@i7aiZ#AT-zsky6OJl}Ryrr0lU`038QM~?~)4{q9Q-LVw4=c6|e
zhl~F2AO|s!&!s7;Y!l9#x?;(pcS6Nb@vC9}=lE(*8WP0lGgwsG@VmCmG90&+_vU)F
zjXTn^?lv0Q$?7%cW|l44p)!1h%eAsaX`Mdl>loR;vxlaR<r-5BB6M;+1`Siv&<sS1
zXZiOK@tY6wlPM|xJ&15?NBcw(Sv43pKhJ=|VA@)qs@Qq_WT*ZndTS)irIh3=MmKLm
zH6jmC!Uj~@33E>EX@Hd}W25)Pf;Vo8T13m8KOrQo(<UJ^Ux#qgo?45gsXnBY!2(NM
z+d>y%(c1&<u_E?v?{G?6F3T-)aP2KMIAOt<_Yt&#>%|RxpuF;4Mw4%Y_>3f?66K#|
z6FP7imcHYAecZ%{UytsyqJk;wm~MTRXmEp66BOWvQ7(`QJ$2}_k4dq1SUIsT-`k&w
zVIin7Wd~ZF+oawJ8q@S#74ou~E)it~>$|9^DA8=E8|=Y{pziw))f+S7VQ3z-Yg2wi
zY1a|-XkGDZG<5*efgUiY5IfvMVafhXoPd^+s{iT1F@r_juMV#mIJu^HN6D26nKKVF
z_A-fs^}b%Tpz%YrY)WnWo)oJ3IEJbjeeK>dx0rp^uDVXob0f?ryyVViI<M8%Sv;&$
zP@9H0Usj6xgzHeqZ99}k3MHfo--+do*>@A991k8%zw~*I)2daWD<G}U9}%4))O~Q%
zct$M$MPJ@SWcTX+jXAFDG<S8o3!>Db`P>GW2=1&oL5lu$M2EA?H7p<6Nc#uWVfGVm
z!vv&IZ^XO@Rij6k3eFRkLQLGgw2FV0b9kHj{!6MeEMW4uiDV1SPjhA_u2wsLgpQEf
zSEEMP-{$Ah?uRz6_eZ=vhg^DXj2_BdxA2K?6G>Xl2h~&az@Rni`j_qJ93RT)FHw=s
zV%a}&3czOC1f-p%5MSThQZ<yfWs#J<k=dY(+7_>U!%s86;Cf727v_b`-_%|ZJk^2G
zCPi@Elj6cqi*GvB*nnS|U9Yx8!ro#xm%!6?J44sGT{L*NUL^Vzf+3rifT0m79Xd4J
zBbGGSc>X1xby_|t&o3Scll`<9wZ$f?@<yxBr}ENux}NFY51EGW`#x$eQ>vsutiCcZ
z&$Z2ckzLN^`3D1>)*g#-de8l4L^<S!e1Yj3QLX_kG;KSD*sdCwjuQ6B)Oi_OLzh=o
zjg&MJjW}`ZVayT(E_$Z_ov}(asxeE%u-FrhrxX>pNE5QXJ}Yml=k@eXHV}v?%h_4G
zHGr3&=eaZ4)U`4Yn^{)uv&c9IUPE;c+F&W(Hz$9LeeunB(rvfJ*AdTETVxKNUzT%%
z+kOCb5fnZsQ`W!Q=*vxsvkBz0OI>{eVd(0tp#Dx_)0<a&(E-|4bzI~P{{e*boG8T8
zgP|rQ`(D0hUJ!@(uMKc=36d_<1Q9_K%Bsfvx8M2y+Oqb46CY3EY5qbinCTnj?h8tR
zkzr{|-=xsE2h~<uhp~?}JwJ%E8%(=&1Gkf0SfJ&-BXT@0gC;_`S~PrGR&!V_#q!wJ
zBGF|JZp?RGcq3<%W(T%tOx?1^<%8Fbx(%AA)NUKsIdo-a>oHU2C+^qw364-6_JZfV
zqh$QH^?3WxmC$pI(UEQ4F*?<bs4bRO2R78`AwSavq>=Rg&m&-hYlf*?j_VyspAf@2
z>B;t@`w|NuENy(sAIHO;Je6We4_8T+Kv51{wvKXF%LqOZ!dUz4HypF;GQQT6&g%_y
z2x}u3JI?phQIIOvSL<8Wc|L-j`JRc6x_b{3lj+XhP_e}=P8b5JO^k(msqmt~W}&`M
zLV8)Al_$ts4Hp0tDOb>xnZ{&-V4i%k$Ci%RkG3(MmbX7OKE5j@=&{={L%Jt&F&*jc
zAIn}$4Z1X{Dnm0rNO8ZCQqmBUY@VN9Z1co`Z(YL6-@1(VYRiRV5wzp2ii18B0cD(^
z_X4VKD4w?UZKaKUigNOJ!O>M81p@RH6;Izj)Rn=#Gj9HE=o6`Q9Eb82Y9|Dfq(<{?
ziW^r{8QCQ?7m1~heTa+7)SrLM<fO2B;xco7t@aeSc70z1DFgrNiK~t{GHTl~#E4m2
z^lD@F24$-lit)H5^<E5B$0-a#vGkF_TOId3Obf2x`SOM70olD#VJqasOy)k#K1sj%
z1%>u)))K~E256$tc#c$`OHr~p`>x~tB@uk~flbRpLZUWl>eso(AsZxILM06rD<!Uo
zID?QeP@_kI2A>@VoYhP6PrU+lsh>|RwAKG@_}1rVo~ZB+wz8KM?Ev#U4Zi~lG$US1
zo0}$h6RvbldKIC<8ZUORKC=8ke$0?V)1B)lHY*goru{~NicmaSs6#ZK!lwAZ=Cn+S
zMF{zv>`2)Y_-pIVh$7qeCPju^M{XxN4kd3e(qS5AGA+B(^ImKdo;PB>Ltcajcgs&2
z(}91D6Z5S_33?l~<4|OXSItSxgHZDF*6Q@l#Z@o1y<V{6qQpnPYknO9SN#@X2Ea@f
znn^ne^o?9RhHRWDL@jhh1`yq>R@J68TqG8~A6}CBV6Ekos6)kz5i6VvQCa6l*>`F7
z?dXLbxvv`aFS8Ecf76z4#aKQe>GCKh<4m!GJm|HKT<VIy=WR~mDK(oIbZ<k3HMmF5
zYa8S-vftbqVW8Wwv)c0U@ofu3Qf;hW@c4Q3;qZz730?QUL$#sR2~I|WQzK(36XP+I
zI$b<#Nn(-1`AmQLB*Lx5&Y#rzLm2e>Na~ioFWHYkLkuuw44D?gv<cWMbyVW=(B&@a
z=2h-%gJcA9Zd>=Hh0@qBsmvl+hN}PCSjx6VIm#g|dqdI_`CEZI6nNJ?%KCEL1;djq
zMh#<$VB5fEqf8v4;Zyh`CW>p~qi&ZQu6b4)bR*ZG^SA4wWxID|0gGt(b==E{96axh
zJ-3CoMZQUK4S8?rhyLfc-iE!(W|2o*3bwH$VT8xK)}7QWs%uvOiM1jpqQ%@N(BQ<U
z#iZhUr}rH#I6C2u1l%x|_8SWmMrnE-G4v@R^~Jljr2{(~hPv*!7hc&kWSO7v%Biux
zzd`D@YU7*}nNeaLCu-%C%8Qf^`iCs6DafGZQ<92Ml3RJ6cxzFQJ$p$Ic(cJ{#GyOn
z%e6yqc=-;{FCL^Ad0Vy-0#KPYg$|NvE)T(y78xQ-({{eNgk;3d)ec9j7Kp9AGY94B
zzDpF;NFn%H%V|hF?ZEhLV;NdUI~>zqDPi7dFlRjWiL^$@^*2r#%d1!2_S^UT#epcE
zdt)(7waNs=!AxpJ<Ni6Lxv>j&zPVy!tj_bK^RaysbP|vyfG}v&BcyomrA4L!d~HG%
zEWQxkPu;fQwbYB3W#X-Zj@A|#^;{+<`?Nt{6&;#WN2RRqC#-lnIslf(&i6<v!mP78
zY^3CraohWe==EmA;sIAjU(>z`<GoxAF{ZVZT?6)|Y8du_^};tN!j~T7I~ifyuylRh
z(9-tX%q3>nPrmQ1xPtcc?rJz!o*RgaQYpquv~|7cXe{t<02l-f+zU7%!48||9z2?l
ze%*M2ydhe0!I_C`cO2!zwaY7(+{)-|<HA>Ks_w2GwcJ>%S98SBTx#y|i*t`J-r0Ud
zXZgwjF>P+YRh1justnGqcH0r(gm_*Ua%T>2<N%+2i55Stw$&yqEgy@e$;ezzlh6_D
zTQwl$Z=ex(<^(2K5sIQ6o7>OTO3cFaSlwr;c-zu8Qw0xEdj^u)RkYsirt&ON-JoJ(
zVIKJA$nz-7{HO0rHiur9$-OC}4SLSaS@W#wByCCjDUFMY<99P&av$Ez^PWAhpI(l%
zj6WzYvo7i0#R3w|PT`4aQJBEHR2!!5=xy=r3g@1i3*OurnRtm6@58?6DMenj(yG~d
zo@we4QFicI)ncQg_wAMi#6_MtrqFVEW5h!bsZXV%S^NAM3%@!c|CWEw;nSn`)4X79
zNhh|~b`ahbASQ<)p>iH0(5BZG8s#z@{p$Tp{$xJTf0JWl-a)viCGRou`g$K)!26K5
zOzJj|NOrPg(~9Mr7q&UX-C5NC>GkV<7RCdgnd;jpt4Zb!=^q@iK-m|M%yX65Z>{uR
zG(ygDU@-Lk;hPy29ZL}ihHo%}Ius%+O3Q9fLGSTzQw@DcF{)dXYiTHdNxjhHtzV+}
z5uMhN@P<4ISaf3WE(lM2X#(YsWL0>1K@$iFtN>{}noJw0+A9O>6HgGoxQu^Vfa=py
ztzw}P?5wsDC5mIma<X#j7V)MZ{+PQiT=a<2pia`Oo+U$;p;H6|aERXF(#i%#e5efc
z-1z8p*-!vjD&m~X-(ULB!;{&y*t*fzDvwy_^0)cVANkV%c|O!TfT_z8q<x@zS!7;)
zMOd_g5*Fx~tQ#GF;(^}C9=hPp`%9ejrR9FJ=3V(l?F)LYmbiy|Ygy5ta@$a#x@Kx@
zkaq?Dt+QK$TwWAxc_~tMG(eim>Xt6TL!tyg4R1H9>64?*l$iDnRst_CiW}ZsQM<HY
z{VwxmMY|uRN`we2mqTsFzMn{=R)Yk}JwRKvD-Y(Z<Y_mi9Ux`6vn6t!Sc=I6S};>W
zJH7t2i;BiO%RI?z@lFo&-7(vEb<S5KBKBS_96Dld+iJs|o*2hHHJPNqf|kJCO7hmF
zb+sa|R@v|S_`3B){GjBbp|cqW6yLV_S!H!fGOqnTKz7RP{HKqJ%0R;K>P)sf)S2EZ
z)&shqbMSR55kK&8SUqBe6{wgAPi4ZNpx*d6VV$X8`!^rRKwk~t_gD#d1C6xzGRye3
zKJENd_T9%da{6vqg-!L<m~#)6Ur5lea9wkYWr`GPiB)eEYV%Qj4NLhCG@cA|l1a}y
z|7Kxn)|@am%|l}=B0Pz@x(M%E9mavE^ga0Y6ypOF_fkV2!5vvQ^W(4DM#u2A$W_lz
z4Zz7h>duCKyf2Qy2)aZeB#5>bQY~g!J#5E^ii0*y4$h)6(T4XMwH=R`&S_t?oX@C8
z?o6jEJ~i=M%$Dxl76djFh1o#}^Icn+qVL5YQpEf6(5|ggj}B|u2`(VHOLJlORsTnN
zkyDTUdph!GBPr0O4QEJp`*gM3u0&;R>aj<XEd!jbK)B+%H#AlU^{Qyg8UQ^oez%u2
zQIDd8*|h+*;A7e)$^}A?-`=*h^A)YZJ3UB`c`v|xVUpcu@I~`!*J&CA>x_f8SE_7Y
zFo-FMrf={p-`>N!Fpg(LeJdN0TnR?)4e5cB<fL4xA}c_OT@RaF&|VBpe+fw5`St~l
z`~K{0F*(Y;2*!!bg!d$=rx$WVP=%m;MYc9y3TC9XBk2eWz=xS6Yal=*@sih`Q72}9
z*vz+{3;wgWD{EOQ@t>`9ofIf9q@~r-Z&S!x$eIo0^oz)~H{zUby*Otu%2SiE!HF(l
z4&|F^frfC`gCU|Ce9{kyKD=;nOEqEWDf-(FNqM`H2l+4Gtj?s|i(HL!cWXJ~vC)Am
zYs)oZk0jSj5OrQ^sZW-G1GJ^}S#pi;4<XV)s8BP(Q=L6#eq%2m>R-KSa;&yU<HW??
zS8avm{Ru)9^UC5+nS54<bv6>l>d~Xiv_;hje7%g*)H1Gc|I|I*{Dfy>ppD44CHGVh
z2JO`PQ4{Etb7>s+`cFjQySMjO7J3?$q&HV~^Ih1!TI!;~P>Id-&$cW89fM>g!`RN)
zo<leO=840Qfuo-kc7D`a;)hm$l9J;5+HB*^s>}0Jx*aRJ`5$W{#-2Dp9o?!EL^|<6
zJBeUF3iWBr+RR+kcfWF0e*ffc+m9Z`204b0Bj3yJHd<QbE}Q3)#4!v_81*mj=}+(e
zzP7<l*VllZ<na6KG=o$RIo;a*3Vv@(b+$FX8f9=<41K$&CjC~cX6LFQ0DJK5#!pLv
zNw)5vynC>AtKy@UcPq}({Z^dr5Nh_(on8J?|49T2bWR}h!%;zhwyXYR56UQBaADdb
zGkxwXB)qNI<LNkwkQ^20@RpxMs6ykFAPABA)=yvrC1vN9*?41Jg>cs;F8^!lYM!oV
zrZh@eFM$l`$*NGq`JR6%xY-Xp)`KY)QHtvlwdXdza%K4Ix)z9lCdp^vUf6}JCsRwZ
zc3F0`n6S>4mG7MW&i3t{kuFe!!GbKN#zCKa`WO3yb0^I)#Lw$nbLdlZ!cDVQtyc33
zX?|-tWVZA^*ok-7_o>2{l!g7)0H@@46$2`)e&6b#^z7&F%2j*IJUWF@oev*-dfzGP
zne+86{^WyBd8+YERz^;;!_S}ds8pf?+_gaXAdC)?!@>FloaiSFFUkgU&Z_vq(*A4R
z`)9KIp-odkb7KqT{Ozuj`_<2uC5Sm@e1SUo!Dg^r5hpA+{mXwUrayhNu9_wHs9yqS
zbNjtXqw(8j>KI=#W_ufsJ}+!S(r>enVIgR_Ihs*@K45RiqUK?)0j02MQT{|G42x$m
z!q~i~e<rg(m+ym4^ZMj=nGdy`4_*D<v14&%f?36kCeH;sINTG>prIM_{<eyL67ONW
ztDxXoBX*5)kG%OkF3pnFDQzkWg=2&FM+6f2S2Yi>|A;aCN!zbP9^jS!4P!EEdeROQ
zdK%iDv%;<S+_^xg`A7)EB&Yc*tV<`5v*3eV;J|--i_b0>#$`yL599va+#pZ+&LXLM
zxLpLZ9AHizCJ`;@@+v;r=WRlA#<}7YdPwee?+=w|ZDmkH`n%W;<rz;RmMxn^SP<#4
zJ+^;gr8uO5`;n|jjdg9RW7B)mJZ*%d)^Y!P%3G#S`$YxaIS+zIp^9g{|Nc37vM3IZ
zHkxhU&gVu?KEJv1f~B={fdsIuc3|*hC@a9_AOaIcOVE3fLrhIo5X#T6;g?yc>ow<m
z0+@5EdiLuOq$*v+S?pin2aN7x8cH{>_GI3Tx3ZcaxdOFJD$njEI)Oc2E<h$L!WUQa
zyqmt(zuw_L-}g$V?upc#PR+ZVSD1T{^`TPkqpAD!ZfK(U!#e!zV%X%$-&DXNwlIIS
ztDxP<n{QK=ylO0xS>G%PDnx%*M)(7iGL{EOlgS5JQ^bXT^VG8{-}G8M&>bJz4&@cD
zxMI<jn^|R&(?mo!Gy<cM2Q~O@nn6&BGy5QOW!!x9G_d-Ux0CKyu@*k)w_khbZGly@
z!x4?1I@GS<T;q8@@9<9{GsK7l<x6bxLH^e8i6?~VduYnSk)`_W{w`1ejrITpNp{1d
zZtkY?pU!md<J9JO_#u${RKY_oe-D2ka!?mX9gMmTzfbD&(`(>K_1TtuHB87Z+<3(<
z)Z7w1^Q!iI;3kkAeXp|ZZ{-1v{szqZc?fABx(WZb#h8_+6e+^IzK!G2G4r~`vrJW@
zTaS#lr!;iO16%Z2om~bW^V<ad#hU-wt+3faaSVH6VIT@03$QG*M>yYc=fhbV2cSEK
zoftiAP>Ah*Z*i^4m+lKZQK4@`GB@$}nM_|oDjp1NydGM~c5&)C_*ckwdLID{%>4)!
zsX11gl1h9}T6<=jqt<~7*JgFb;FsCfJb&Y)JoOuHHpnQwUBP@Z|CH7R?G%B81gKn#
z2xo(Rtkm^2aSEFJ55`9zQgBG15Ufb>9KbromVW;dal05dEb26Yp_|TeqD8_wp>Y2s
zLh03vqbpR7?&NX?w$}6+`5cEIGVO%dRuejZbAZQeX$i0na37`=aK%sFLv;6kbKR94
z-@)2BVGXKmf#)30LD$M}2Pl03VbNt$SVljNvEZqRJ7CDJhvM0;#_YL!`G#iEe0seH
z9@YZd&Y3|}p7A`TJZ)GWr_f^`pu5LRTR1WLbS46qCZ6S8@^wmEzqk3spl=vxmX{8M
z{3d~f?Xfhz>dVGggTAPUv!ma_jbbyCgOfFys|ocapg~?%@ySe+dz*bJj9R;Q6G07r
zTa3k_r+~UTPq?0{kn6<pVkY}peFKU{$H7_XQE;{@0P`URsJ5+n_BGe_Yrg9XN(>7q
zza}(0HdpCKSlXIr0+4StG&PvvzT}YoMJk2`>4KFngD=_g+?>i$PR#_H_Y3+nsv=v7
z?5zkFbg-*gy*zgb;3COkjNVDa`}$p(5I7AHH8lkL*3+3(4X*4&s9bS3!w{TYpeW_T
zc6ghHO%0Yk*HmUEdCvv@DadQc3;--LmeC1~E<3D|$-0yZkv4GMosOZJyQ7DBUIKL$
zVJhzcw0pU5C7ffRz4*rTB2N#$P4&z{%?JO#8_KHV!dQa4pI!zW_~kE@u^ca^Qe?kA
zc!0&6?V6g2f_1_5k->`gDsU{RQxR!cJgAq%n3v!Y2J5`=k)R*&)&z29>kD{*xLEhg
z;4I3D)ld}T=oE)I^)T(k4y%3uvR^g~?$j*437Wnn4t&yJa;g_MECvd9JU~*HwuCn)
zAhCZyMYApi0avlx{_gD7dS^`M#sV{k8Ud$I1J?;aG85bmjiQ-jkUA6s%ikENs7~bv
zsqTfKbEyz|tQrVcFmO<e7eM7!9WW2LfZ)-b?_MHGzX}3rJ2<TvvB8+li&<z{0D$IF
zWqdM@F*i<lIR}*`L;z>fo<wBf-ac1$Dn<=2Kz~sjPTn=cwGFa>w8^TYhoFJ<VvJK;
z7>x>Wz1JkdI(7MsU<+_qS8y7ONrbsQ$mFI@BFsRi)ZgzGBDJ;<dyd?XYsktO{55j*
z=Ob5t(t0h`z63jr41;sR*mIKNIbS96_)V?t_j>)8niu`0=6|XAUuyoBnn%1h<!loQ
zT5W#|<K*B3pt4TVSS2I43(Q<dVEF>135-@^+<eOeu#WnaoxQs^iIKp%<Oof9Klf)r
zraLsu`wWa(0^Dy@raQ2zhRBI{*!9ZH5wg!Y;rq|k{7>4qa4Pm+YJQ*gFEsxP&HqC4
z!T%_lPdifMDs?m3W7+D^JJJ`bI<n)0Zx^Ac9nSCs{LVtxQ`q0y)y{6K``Q7U(-*Eg
zm!+R0d<5<s6UXd>O8wTYOn4(zBB#5wv3I9jRj~C}uTU&A6-ogvG#)rVDZuT{hBJt8
zeg~We#wUC$GqX&%RM17NN2i>1ZS*kwV(D{$QN=H#$S7sRT5J72R&>VOXs(|`)B(11
z?6tjcrW!OR39-+Z%{oQGgr9L!eR+y-Mo%xHA^UbEjfI1A)JNh3>ZYB`>?csUI?(UH
zS-c6ntvW)d)MMc90Eh&-NKgCmU;yJ3*F_-X#>C!XIGR3dFV2~@0+=yl`dr|^67^tL
zof+q6BM0xJ>^KB#N27)eS>#1C2WTGO1AH+bJo_AQ#!TM+jNYmFW0~uJw?SYYdK9CR
zb?AEb2*z)nrzz&66V%gU%mnUp04R|=>``0d%f-GsX1vg2{Xv|5T}R#N4)d#SpO)$$
zS<KU`Z^rTi#c)Yr%201yKWSnT0dBuGs!EmPEZbI`+tDPyKeSg5Nyhr2DfLjA4O>*k
z9G*IOEQ!IN8*R|>^4>|+7134e*Y+D|8^j5?fV*bbOBxmhM2csc_W|IhgBz1-&-!3v
z`?Gfu&QqTd1rITfSTnH9*1$4<3p~&A44j0LiBc{QgK|A0oRa`wTFK*H0#V11Ok6{@
zS$`e|5F~pi5!r0}s2PV`D1SY?RSm`HCR6|ycQ+6hA$>R-Kf>WF-y_0WpeeEsbk|i&
zg*i^phojz^fzrYE*dhE#YC;LYmk1b&8oy19=_>3QPN_l0(U#Od1T@~K-dUh>hv4C3
zm)6BBB5TVeTt(NLLwTffjEH6C?DHAiAr*^cc~=4)gJJ;KV~eJapxI}jzTcaEf6Xxa
z<Y5SdTAx82{J}EnY{8;QL=1rmqG>NN*4lHvk8irCzxGIi(GS4dAD=|%Y~XtTHNDEb
z4Gks}e+lcch~f-qRsdS;+q|q(@#f!gO|v_@4&5p8*ar|DRRKpEx&e7N8;<;1l-{Ma
zoVATYD0B+SNho#_H~ZXj$S%UXwE8gbdXclsOS2ZOQ=M~R^U)Gl*AeU>4vYnT*-m0G
zYN63o4Mf{2sP*9_g4=~|W)BH_ggLLSy_tesQpsX54}z%cYuy6w=Jz9S+A0@*x!&2X
zy!!;8*gp8Ks@6u;nQN~e98?`;+BFEi^GFa}&LVfO9!v@G(>pJlF2NzqzK4AqQPKsg
zS(-F_f(P_kOPNre)^WdY<h*#zfbrUiz|)?01n8)#`Q{;@)ZRY{0*eBJxx3ndZz=O{
zu0TL9pu!|fH+o<T^<Dc!v1;gk52&<d6`{dpXyEpnysNj|6K)1<%+S;`^1ZW{>lbJC
z|3S`7F|T$sMb(6;wj;vOQsJ%F8}U`=L!PZ37rZn7y6`&fN>5rCEW3;%2nSRyf=Ic>
zumS#A9$+>>>SHNxoV$)Nx*apJMH`idqEL~3*wNJ<kubXv8a6tQ8jPo&BDPDsuDk>W
zOL+6k!{Lv~>XPL<yUqgr*gO14rF*Vr%+NzIdKj<EW-+=u_M|s(Xd<`J=7FRGS+MJB
z#^aDtP@5{%hRrubpeSHrrTn0AFRJmCRNpPEt(jT%L-N6v&~A)=R(RBn=D4L7KQ3ip
zDx(HSFB4cXq#*S8nm&keKn4Nrh>Zb$5rLdEiCBN|)Oh?9Wshfo;KBQP$*<e}djhaH
zCEuCtfM<J^v2A%4O}k(-L9=#p+P7Kdal#tEqXF;V17<U<eKz=}#9E*lF9I!jY7)`E
zVG_Y;O%Iw;eKHVOm4;@v1cSK#JW$<{znkb<4`a4NWi^!Itd677(NobLZL!4ZV1zS>
zuw8e&4R=f;)*f=@L<wN1)U<x?_Xd#-R3t=1_bW%ICKQ76sI#5n2Yzc^IiLW(A2+-Z
zu%aNw+aEAPmOFd~_GM8KyKgEU@tMTBunOH0LvN6uIl!~@F3^HZz?%bFHYv*ZJCyKs
zB`1H<{^GwG7a(KMjwrg?XDnP8jb<#y7oK8&VrM~}L|2F@a0tlSF~Yn=1ioL~i8c=W
zV7wv37FVWw;u{H##^djZ1$?aCP!PlT(&*s+#>Tfb*0F7!7<RvnBS<2>=%-bksHVAp
zNQY8+dF&nc1M;34{OY5tcveO%zWQY8*b<&#7QkgeQ65F`B^4A58^LM=1b%{jzE{V9
zESlo>r0uqlq%wLQ*a`NS@Vl{VrOoI1uG#U3tsRGq0Hu(q@d$poeYF=R5zo92qG&7i
z2UnD-JhAJ6o0sD+CxSAY%FC$qp3*R+Zv{@EK&~${+CIf2)xlczZi<}kr_-siD>Ih0
zoq2f26%XQ2ZLK6Q0Giijh$T^S&{qZ2cHi;l0~d^21ob&%N;hyZwp0Aa+iA=WRB($h
zB|6jRsKVV8d-09#O3`cO7CS2LZ03{pPrD;q&!~OkrzUkB-2+UN2JD8mf#|9@Vf4VU
z8p7yP4fx6!nm+gQRKkx7mtMvIp;IuC4Nyc?(jtx*iL+ZLwU02o5J2wKCsUD{c#I2)
zK6fgzeQ?UOA{=S6p)pI$Y|JceW;<m*hROoLPXPhD26=}*9JMVGd|Hi@!w>=I`w_Kd
z+hb|;tL!n@wT_lK)W}9L_VS~qF}``MIwKeC%5hH+Ig(-noyPXk1mj1(#ZwN%3?WlC
zPN5lFo;0crBgtvvHeI0IK4bQ|?Mzh1c^kl1Abdym2Xo-T>G)wS08&zsY>>D#-M(gv
z>OANjiTx53SR}njRWHO20|<J05;1;iAQ6PcOR!Y71lt+KSeQDVr78xMjbg{%L%n)_
zn}erz_Dp1q95EX(^V{46W&^`O08_(Q9}Xq~ZimVSPJ<<i9{XfQ)kwt#u`aq)J|Zv{
zBy03r*M4zwNkNw*1{T?@;@`iv0sNq6uBu;D+2d(k=DQI6rmsduf?1boIHrqlm8>`|
zrkyn3PGWrC@kgg7#@+fG)~=uH@I`jRoUj!E`yyKi$2Sa0!cx8R;RlsADGDb9!>6@2
zLvQS9K%$Qe11%VTu8O}JOQAS5>cbZ`++o3X{4t{MM%qY6(10;j!jd{lF?LTNYdECo
z%@s$;jX02aRS!gOC3Xw>lu^L&@Gb2z><|qnILJX8B@7T9*93|sDhM0;0(n0+4*p43
zZhOlNg9YY*c0vT;G(GnM|1?HXx)44<oAC|+lX5W<ew9h8&|qJ~m!W8iHmp0MxbZm!
z@LA%3D?s7SK&5R#Gy3F#&Z}hkflrVh-QaKrk%3UB0=oHBPttu#CF@cdGgMp5l|&b`
z@Bx>kygYqu_W8`+R{t)vM+P@#HtaWZ&_<T#&w_gDe+>Wa<rACs`y6#A5zASmv1UlA
zD5adxiM<WyhIMH#p7AP@9zf-$Bz6ZY)Y>%l3itwZXd-NQ1DAU-$0)%ZRS}@-cYaNu
z&2F59ylI?~y%f%%-^?+is)FF7z@P!2QYS~L$6FPwUoDV2DY`%}(6bw;{H%a4=vFji
zF5ZyyRowhjQ=&XLil%c}`kVlJU-zt7Q!g~kox^xD6&6IaPlwf~{x*U0TtYIhuz*%!
zfep5d2X2RyisR;EC7?HGY69CYU6s9{o>f5{EomR3E5hI>Jrf7E{edLmRwF`rpz<ni
z*aqT4lRJtD46AzTU7uT2XL$Eet0LRmBW4m|*yyLV9)FU37u{LF+6*Ko$Pj>zo8(ga
zPXrrdx+*~!8aP@5+D<>#K)_ejRKKb`416-Ne$Wt5uGlf)`QLiF;rH<C42_U}#oV)p
z(S&78gfXi4<fm_&+r35180<kfy@Y7jrkd?KW(<Z%y8I3s=;#LZy4u+wu`_Vy_9cJ?
zo&&sp4tilma5O*_3h-XD8LDHcCr4nk;#WgXe*BZ6R({G;ZqU@jlp~4d8JQDHa0pKw
z?Es8c0H=fMFe;q6DaSYg1GXOBDF?O<xaIW%GlJ0s)aiBkASPHcjpCGC0-gI1l&{L^
zZH|R*fwzql#@?Vu7h;s|abk%ZQ@@dG@+Ykl{%@d%OsF-EFnl%!H$ET8`lzM1F4hsa
zY1^RA4psEP5^%SAa}c!Pf(AB7jI*Z!@j508EZ##Ea(gV(e0(X$n#8wb*_S<ua7bnx
zGT0VKICD!oiFK>&l@o(bRfJ>IypLmTyv>2a+C&oq7<j6_sHPhqV$1Upl)e3#1$Jc(
z_*Q_Ojn$F{6ZGPtzJso8k$z}+Va`{x(90d_@ChoH*vBXVwijeG%)S7V#HOXwvKH!;
zvj;cy0C1hfc71QJ5ud48md|#$XDrn;2QLQfY#sUwWY#rn|GwiEnrhJ<Mf<>)+jBBX
zieNE*Oj<eDK%kCDFv%?P6RHy_oq-;mn%(*>;*q(*X$%;&v4OFmSyPQDnhwn;sJ5WY
z9Hx?4s`%~NE5}SpL+Zm~#YBhM=T*}sP<-9MF3;p&n{%=bGsxD>ij!QhaJoe38iGH8
z0iX4s_HtQB1QR+Ks*G;*sGQ~C_q+qb`74*xKr?3hlrbODiSm|Cm%!d_hH}t%5zd*M
zHrW)gB^`_oI0rqVwtr?fqtz!7t&gE$0kHSJq_YN+hOZGY=pWwk)4IkEhdMi1T$6~S
z3xSer$wFVl4x7PCp^rsRClP%X=`FtHgyD0(fU0^15=5$<EKp^tbQNd;df;e1;2g{l
z!;Q?awh7iv_Yzg+ViyuyxUoR`k0Y!TdD`9rJ+0LNdaV2~TpF0M*V=^6%s#@nsvUv3
zzd$+;nM-KBiyJ<(ANnZHq?ciz*VDiUe=xgo+_P!W$-A0>-*H_A$99RB&7)GDHmw$8
zE?P^#|K9}kXCikhP$|po!vNZ=(mi2X!1-#XAdU}wUwJ-??Tupd0&fe=x>^mE|26oC
z(a~8EAc00LD#&WWsK_2FiWP8zp*s8+lsM{Cv85&v+cpe=B`r5K2JDd%qHtQqX3Z<`
zwj@|rM<C)^R^lwzmF{9p7igM9<U!%dPiKev`mLtnV4dw#Oq*H-QMVnU&qD$Z?mcnK
zEGCP_4f6ul$rIfJPHhJaC@?(}4`J6?5AnY`Lzjy)gC=!(t`8E{#$XQv&8sb~JH7A0
zUnyUpo2~=6b`~9V%(S~8*>%33aXHAXaYIiSA&$~=VS`CTD0*NH7|sa>P`MEvqa~7%
zK`k4~%UC<VSzypG>W|JlK7{;(39jR!-t|%+=mSV?qOB6ZIxKvn%xb;W2c`z}#^>vh
zZ(*P~<JVU4j~W0DT#pRr5JuGW@Tai9t1~?7!@l3y&5W_iu?)i45NZs_6GxuxOk$q$
zw2z*;dd6`_bUd0w93maqo$v;JUhw$V?CpppFhbFAmI|QxX7f~M+z%qy+dp{amg-=S
z|G|HE;WC?1UjQ%g^%1_HN5p{siedS?kS0)cQ2aD4TpX@>_9+gheI<_saEqo>J(b`m
zU<XLy7`>#5#95K&@Xa}W_L&3tlkG^uNT704!Oe>c&OSIgMCS({5)qh`DqV}OYl!?=
z`R*@c`^(r)`^B@e`u_`Kn>@6D{E~)wZMBMOd1pbhrLCvt*y!4d_y=|l_w5#J;$8G8
zRe0{EUR{K3H8uiE8FmW6jVjZQgA#|;1p4YQ+IjU-V}{sNZlPZcmKCEddmB{quLoNd
zQ;3^z#rIfaZ@4n#nRc|HgS6We|5i*Gm|5;+IC!$jJoZqENd)gv`3G%cpFYkP9<U$D
z_+`(3*>lOMOr2$xVhFMBds9X@zl_BXz=&9%>vyk3xk9H%tAEp;qwYW%>Vt+<Vw-U(
znZ!>MEPlAVX4%WMsAWvCJGAr1n|r*ga)-Sf%i_IXg0k`^Ca^YD{b-7L0yM1VLZIx$
z(Kl-v!mr0s45fmlamYH{r=syFT0QkzPdhqLOU@&RX>OE3v3h=98hauwvq@v|ansMd
zYo8aE7x8x7yTyZwt{=w_=!3*!bny^#K{}qj!~rYzUqlGi%uNpssDR;oi{IBSv;h`8
za|vJsT@H+pd@80<)HABaIHj+u{6*G@h<-KK&70@PznJ?)DQG+_2<k;YhFVWxStbZ$
zT_bKYD{~|F1Z*Qo-zjaUL-%2~FPoM$?pZ&yw~A}leFt148Ao=2$Sr6xNACA@N&rz}
zCQkrywgyUedh+-{=kEYxHU#?t%hm&-KMQf-a${!Q`WdI0%=$wnK<e$Er3@faW|_+w
zI03juLwH+`6EloeGRs|ZW&>EHb0i9ga$V&Z_zZF2j9K!5BOM^)3rBg#4B6SR;X5qx
z)YF~iKM0W{K-PQeR{+94`Lmk$|K6P7e=`Q~zv{&|qHV;4ri8fsXgDBn_-;bVH<l9&
zVR`7ab{d+Sr9)M`*iAdvcqvx)q3a1Es`0Vvrq^=Id~h}nxzF#M=%^!h#|*Rq@C;5-
z$qpJNu~+UgZrlO`*P+;nxmZn=z0woGB*C)g;2?&4$|S-+5_mP~V?hHw%$<jPt6F5J
z)*r$Gk%57M<2zrq_~0v;+s+AlCyoz(8oD!H0%?vhs)6s;0JsXnb|<!b4K$vQ@C}|O
zkT~nu4?g8Ae2Ehx&C&w)PiN_Z86S7!XeEtO2a#jHazh+>2{VGn?e!yBk5K`P1Zpi*
zK&Z&UiL;s5hk#v2bo9=I!}KN*7ziZoR{|gEA(RCCDZC9wB<vi!ZS3*yCEsrMR?U`m
zyR~q<m>~c-@(*0I<EDZMQ#n1r8<}+?`K{Yj>0Y3%D8y8AGbyEU<R&1*zRNyPYaz~Q
z$J5}eaCR7bZWaI9uL^vY7dU4!!xw-q4d5DPid&)ps|2G{(PVJ3>rFsHopHN9D?;@$
zVSE*cS2A{uHDDB}a<G9!F_tOr3bExq5wQK#k}#d8qWF<^>^p$JE(4L&>V|Zxj88UG
z4|<aZS3qMOsZcN3jCSo8^7w^3{#oQ<k?|}qcmH_x67D;TG*8|2u4dZUOJ)QyH=fF5
zs0=WCYWm%j<TJw`T<3Gt>s;D85r+hX)pHPzo^YxO8vyysK24+e?%S}qO>r1Qa?FV(
z3@qkVFw^L}*Wgp=*8AAubDlA5{{i%<iT5_F!xRO3p8as>4RCYaz^5K0weLbOG2?PT
zhvC}Y*_X`#-<p9BP8!6141@KcE2Qyd_E^Tc#sVeK&F4K0P*Do*6cM;!E6}3xgV-g+
zPdEv5X9dg+5zDSXqxp$E9U9|9pOWA`f^9^0P#Opm%BHug4U%4BSa&3f@Jyr~*p`~M
z+=M}iDPE&p9q<}B_{N*ByoN4_!TQibr($7%A&y^%38~^I$FM~7YTBW>-a8lEcUsr8
z+obH=gJaNA`%y3El<SXkV=Gg!t*Hr5Y{4pJma+|Z=0=x(k^0PkqH1UL&RqQR1L-#|
zuA9ftf4{^VQ0Ri|+x6K(o|!Pe@g$-Iz@nFIVC5@weWPr!de_=NOCOEq<@E|7j%@ky
zSmNl>wkzG)6h;43#}jteG23hJetvWEr0u~7gU*Qa3Vv!+B=t)0_(i7tHJS21Snir~
z*xV<ze~UxY)}oM(O;6Jk9zx~&o7nvbipqiUd+gW^mV1Wu>sXMqM6`$jZ}n}#q8Ep5
zK1?-TBs^DcY4|~?Eg5Z1V5~d~wH8cFB0`95l_gGeC9So9IIg%4#1v^Egz&BRttTz?
zMX`3VBh^0Jw^gL&e8D+^Z+^F~XcgDVIns|mMJ;9}CoDZ4%QC@dc`JgAG^fLfDXl^n
zHA?r^)Sly$3;g3bz$As_ffJvc+MMKeC8j$pAUKIoTylOAapVIU^i-?jFE$y^9M}dt
z25d2bxnK-<3{F^4Vk7A}KPwahwEV#K2EwRyKEnF|Fr)mRNa6sNR%r1+#L6q^(d?uS
z$v*~Cv}Nm=o1iz)6sdTWtqy?0qR8=UU;+)G<E7wIA+Xzb$;OPI`T7o}2QFeb!8ha;
z2W7s(tEo*a83W(YM0lk8ra`aO2%Y<Bz}>eT#j)|&h+i=2YntXSnDl=n-~ysbynUC{
z%icBJTCwv%T9s3L_?^MMx7+p`^sdcS)O?9e#!{a70)$?z4MbhYfk4Ya^d&j=z}jxm
zSk11HA=)9d00>~ld4S!*mqB7#x>5Tz^itYxt8CY)OL4VVkI>E(cV0z#&U;a;Dr(bG
z|6W@jjcKNPrU1IWa<(&#5_z`yj6KKmS-AaOoTj0JGlnyBki*X-G){kv<;2ZbnofY5
zDSkEyG7pE!3F>Tj)LXwbGj0QB2cTL~2+T6N<l7WW5m*N#ZV#b4ZMlc80KKvTJ>{dU
zw-|J>0xIqL^{;Wv56-i>NZ&<k%sJvXE;I3hUMm0vW;&>tRM4>SwgiylKP*<gm&?iy
zv`7<dhF(FV3$0=}UuiF{o5~93%!p`L&a47IYL4p|@Z`3MaurYcVwks+h)fCAZDHWy
z9W_L<g$ok#QIiNqRr*Gds$Lk4V#pwT#i~I#eVoJ+6@m9?0hi)+4g?3|uP6@^Xe$+2
zcYFYj@pl3Lh+!A)<!>ZBN3nqSAVL_>nNqds-X_>na4|Hzp0x*H5J58_I??r16NK2f
zO-UtvJ;4t*Mg`Gq)l;;JKY}Q2)DmK*?+5Jg8LTj2KobC;-ZA__ydyi7(Aiob1AX)Y
zK&6_j4FuAb0MaZL5(bXeO{H1vqQ;IpHLAjfLsR~&E1LaVMBYh1ttOfgFfIqEl`z2t
z$yNz2Sb@)gMW8n~SU_CypLswg$1(hsl?Z&_)k5^Ju{SvH5gDNJcc?(?)7fi1^kIXc
zi-Z9lkUZZT5GV9>T2Q7-y6vyd)Oq`D>ie3D@!RlMI%g)H!Z#Zha^=kJoRL&lSOip%
zJC9<?79V<8zTg5uZUoT?B%f!Zb>Z~)EueZ2_>-VW4`DhT0pj3pK-?)F1OdNw7@3t-
z&Pu;jGmauaL<wgK24R-f`90SIhzS>Gsh>GV4dB;Ii?7)hWJwO{#zd1NH{+{U&NKNx
zSlgX8`g>o$!%;J62DUFyhTiO>hOq-(k$~H1wJ)Fnk(bvmMUOA3Lyz7kUx-8g5qV(G
z(2U{Tf~BPKP3BPV?Nc;QKY>4@4k7>@Lo87Vc+a{K_;oA<UyVavP8xLj=nv}=u0Z3<
z$iPczDfkVaPB@07OTvJ3`52S}^}h2&&>jQThkRL}@?(Of+PHfOfxbS>k2E&YZp>bN
zidO2Une>++1I$4B3sQ%AiVb{$i}p}?JXIae?L#xI2=e5yhwwbG9?N5jNsOUKpi06M
z!4~~0Tq&tI-&gCJc5Uli5QM9_%f6ckRVEWAil6|*{-hiy)+NAcK81R9BAHwEO5Z*V
z4Fygj1S~na6iO8+gLI8Vs2NUwOEhUG&=^7NO0q@DjJ$uBglW#v#$;y8@Q(kO+>r`=
zu^);}Zg7<WJ@r*_ns$f6sX$TX-vdnP46`$vnj|_^Q)2Wr5``GtdKiA%#O{W5vSJ7G
zW}nZns+_G#$lqsY&QP<$?>KWXlR`QBm>`rnJC)eybLMBp2_Ks>eL19gK>L8)OBYC^
zGbsh6UNOTa^he@^-8cyxrO%wb@whB-#bNXn*oDx*;d-F)qR7cV@;jqkEaF$y-7^T%
zAQWnun?cm}8gNq|rr5V@{G8-L_{cDq17Una<ADo-=c}8429hRprn9uClG;o_7E!i+
z-d|*oq;YknPoRp1=xc_8u^@4tz>rDz-i$rV0>&d!m<F`dabbXf?>Ft`FS-HDlusEz
z_uK$JlYm<Nm=FFL+(kQA0POe=Nh8{*v%asf!~0p*ASKE-37q0JR+!j~eTHG(AS{7C
z;s6)Wdj?aKhQJKz4;%*9oYDp{*mcn8V>6Znz9sDAOW4S~5GYS3NCWBHoyIj9Kb^iX
zn<|oYFEywLJuHIyqJ}UbLY*pS$54#9$-zX9=d;oQ(Ip;&o97QmgV>@Ja3vfa?WX+Q
zsYeqS<r8q9p$=cPYXNI|JsGD9A0&Y;*k-7QP_M$>b~Ih!%_tURULKsT=bFh8NfOQk
z_aodF4A?@09xXE>KAVoI7RAwzrD_fdSdW=(%--)C#LW=z=W1c?7=TsuS!Cf%j&5%=
z2UO*$a-?D=-Q{S<0>*{bf>tRKKwka8dN0pk{rzjdwTV-HJ}24`1HDs#H#pOt?edgr
z`4MRN?87R9VS>Y_H8R-&tI+?Cy)O@ha&P+|Ns5^4yHSWJ%93OmEkY_vsDw&FLdX)s
zh%8x)Xw@i+vJBbEHufcjvPO(GJ2RFs%*@^It8*&U`<!#0=REK4kN5n+xMyznJ@<FH
z*3b31W_H7<(%?M$Uzi!j9cY5S-ux~KAdXox41y+7MPhOAZ1g$gJRc;1Zu4K1t{2HK
zikfLbz~-?cgV@G7?(9VjU2^Y>KaJ;HX=a#-nsrhX(KRKg-M|ec0XH<h#YY-?kDfC3
z76sU;=ZlLfx=(Qop9QkI_#ixK`v|BS5qAeo9&-r~phJm$7@GGGV!?z9rDJYK-3<1_
z7V;p#=jAa%F%`gkqBC#gxsL9CM3#n%0a7{W6l7()aEag*HjP3O&ZQPye-@Z14GRp|
zAhV~>S51<BZ*cw+iM+8fo31AN4n%e!{8miJ^-q3jDOTlI`KA9wGJ=2Xl7EqG`fGLx
z{WOv^LoOle6(f1B9OeP_h)r24>riSoDkWCVVmq=@9i5)Oc~0-h6o=7}20ZOH5$$j&
zcryqiU7JP|4QCRmoFR~Iz%?Ki-htpur}1Dj%=@-?Ihwyn#=WEOGbYX*SIPe@<5{hx
z$GC%wi&%!d*#75I<)MDn*cPE5){S{Ef}n>3S#DXVKoL%3w<Rss4g5pf_Jd9Mzd_rU
zrc@5<03lAY?8ygZUE7G2w-PS8_VDZ4Cf^P|&ieRv^trWfZfa%2uFTd@wSirsN#ubl
zV$Hw6bj&i)J+9Q%Abxl!4&8kTO*OP4C-=h39C)bG<S5H-9QW`ek-Wgtg@VX^mY1V;
zYmux3+P3UEeN%o!6NXk%ShQAtPJ<PF0+rzX?Px3H0<jf2<C2Qf(m-haK}2T~dER_p
z83{wT?MDygrosE}P-AdG2%s`>3&z`ESM637t(dbKe?r4?{bfa4fUW^n(vWKt9bo_%
zFbKz=z_g3jWOtnTl*tU9!jpxkJvV37XZ+bx%AVsH13s>}Y^6C#0^C@Ti!cu3qP(YY
zqid4g%^{o%g0>|F$1pn6+tC+TZtjc72rRYJ>uILC^6<w(nd`^J*y{M$CtP9cxO@g*
z0f@&zX`(J9Fh>F!_i;c6Fzi}zRQvph07Nd&IA5>B4KofxR|eLk_Uwxh68D|8TescL
zlK&)TPrvm{XtJ=@CLg4!bt(uw>~%v<-me3u@i$7me<DJ^IX86wM1=nTM1(#>bin)I
z6?!moZHb4hFUt72wJW>3u5N5x=Icn_b|ahHuMI`)Hy-C0V(_+Y{e;Mkpx!oLjc$7Y
z-07iO&>sfAQVY$tj_o5R_YQ_ZYHilSvc!PrDZ?*Mbmg0Ln(A!TjNJNF-+aI+N>4<P
zC9*!4pxg>x1@-hcxG);DMo(48!9V_Tvqg{{@a=9;fPe#)2SnS5c`#b)_o!6SIof@7
z3HP$z$>KtlN^j*n-FHT&dndSWZPMPO9nNVN%Fyj;+*tHmfl(G+mCjdmvtAU<!t1cy
zcL;{kcXE={AE2b6@6H4E7sW$)4#3A<S!M(}-GB4H>;W*Fr4juJ{aG4tpL_#6bJBe$
zom^5u<kY;Rwn%@C46wVu;hGI;KPareyXP!R3W%0f7JWYhexecAkFW*cs>R?am(%m%
zWtiu$=x)T9|3#jXe{98nZ7T}XZb72v)Qz-Q_3c!x4_obKnLDK>4}4^4u;94yY)acJ
zLFtW{m~#tmm+d&_y7|-vgb;3Y5(F_mv<=Y%WKz!nk_|U>umL$45&@}bQB@N`qHf(O
zMj8?5(kk0&(9nqjy4Av;>!eVP2;mu9Z8`AnT8CqAsgK;d@cJB&T7Wy&FL@j=Y2-s%
zU#ZL6q7#7A8B7Ls-t8cA0ixnHHN)uHo+R{8lzH~Tk4PDg6WNVH1MGo0&HW-`L9BIr
zu>kcO@$WzQr@ssTge^yQOhQNe%JH+4iRfW5L&{b34|LSZPmj|PpL!)z1kqKP$_UB^
zmx%ssqT%v4vgnA0i`JIC_j}Bg9WoO`b05Vl>~1L%#pWWd>*sf3KGs78F7-4aTXQy0
zs*HarOMQs(L5^)g&X}|^5^Et{^Kv-IMwkQQ<^1}OAYiOlj3BWg7PgK0kq3YohV5kp
zkPj>%&e0Gy`9;9YxGZ{QU=4;+Ud&{F0s#?Fp|#*Z3u$+Nny`V=6G*Zo4>g9S1XYZK
z48}#{vfFgOu=czvC=BQtUE&?ujRh<e$bbTsMPB(tE$EP6H;BV1;%HKjiZ)?$y(Phd
zzKH_)^7Lo9AhB<-m}>{2m(}S0Ofn-#gM%pV!9MJEtR5MJxWqBVNYZv2^95r-nW5E=
zqh6mt;}v61?f&+c2E#uAx5f4f-*x6c0XI-U{r@B2cG<fNIdv4p20{V$K>5eZNcSvI
z{76LLH->aaP&lq&Xnc+u2so(1B<SJiWp)SBLKf~ojlXR%iAm}mp?y#VP4^4Trs21E
zw)cp483u$g^=SSMATlpSasHk9xk!{2i0p5>f}t?sfl+8k8vLCb?jOweGR*h?7Vpb6
zT7bHrY(c0~Q!Ls@euCfQbyCdJ<-Rg+^(GN@Ci=$9qkwtr@OMVG0s)BMoOBTc4<anS
zz`U&%`v4{rl&bWudCL&ZFg<uC_pZj{%~+x9-0v%-3j8<Z!%k}S9Dk&D5eA{v|H(go
z1(g#oQmlG-sP_V^92O`I$eEno<oiV*$(kTpbC%cv70rOOJirxi{u=rJm3dW!r{&}N
z(F6qT#wjf*^c6PfYl=XkL4p+3xR$RHlf*Gji89YQbww_0Z`<uNtDm<m1XE7{lgzVL
z3?ZDhIcVH{Ms3gS1-5}(rzNO62gkv^_r1lD89`R5dWMcd>%H4K@mMN>8XMr*D)m0k
z>5W#?IUEzP8FP^rWfz3OX~inCkQTPN{-af|Y$zm2UEN4M+4oK-+tldr!HLbW@!@jE
z<oyNC9XC_IO*fJCdrs&7e4djr#c9`@=b~*F^6Hr?hgB)JLz6~!bQS5*QoP08@jb@H
z0Yp6?#ho8q_NADZ%-0G!VWhM_y&R7=y(*<#X*-331e{4vKv?OsIhFn;Nb7(cb$3yK
zf#>DDJJx88c^&pfKzQ`j6=9H8!Fs8HXy2wZly?Uvn_G)&EC<42LLlQM+Pa6bXqq>;
zKY$8nCTU{9KI#hCY_vrLd72FTf^)1H)R-cvmGAhdOMP9C_3iS|;Y5-%xPuV*rFE*c
z+=cYZ^eSvYg#;E>R1|@cc3le!@#uzDklD0e{i95)HGQq^1Tqjmbr#PFfGvRSPvzob
zVw4P?V6za2vsEr)Q}*S=!2=i5jK8F=`E_#DN4=Nki|c<%u7%2J#^4&hMMBtDxHH)K
zE#6?LGLRtC+t7^-w1)k{^81Q+s6foi!F};0&K98JUH~LRa3Qigf*ON{fgElL;Ep)$
z@|o)=`!EhZL;<(E8Cy^JoSCfk{m&%2Hvm!TsB6)1Ay1a2YSTrHmp6AR9W@gr)>jIP
z65z#XL3doKQleh-Oxw0cT`v2bLR;5fAHQ|-h+$AaKh+(uaeksD;fAi>`2vHD;pT;X
zkRG$u33f)^FWA=!(LqBgt_Su7tVTEnQ$fMTigR&x1wyG8y+z3m-3}W@4Iad59dA(R
zNWQ0SD0f(uX-b`wLEvtb4{QeDU>0jHz|TKC0l%y(T<yE%#QBXP6#nrm`&}9)3~&1$
zddLP<W9zUTA$`~i2S04Lu0y~AWhi4LrWNbN%_~05FA=nnHscVwMr}t+5s<Mfrp5>9
ze(BQv&l8g3M#r6fHHPO~YdrAd*%W0m`V_9gp}z%ce)dV%PNr7-&HT;R2TG>b9N1HW
zy^XWPv?uq?l_Mg&*$L=S)svg!<45E^ZL7#PIaMhrp7m1pouE|N`Z$8tLGr#{4E6RT
z4_LJp`XD?I7zb|y5j4cdc*w&RC<Vk9m|&U&s9Jp8trv=Wxw){{JaF8#W#|3z=mgBA
zp#6~3*v;-Om(p6zoLd#iedd?lO`t>cm1vT(2{7?OF2cy(ix3|G4D61>OgYX0mM)e?
zuw@JvGr)@_2gv@uNq|chNCc<9AV>PB|1xL&i#miP7&cQ~23-@on9smQu!H2Yr*Wmr
zN?09?TvwRCXthN7l3f7GZyX8aR;qiLr`uRF08cXzMxnBHG!$_PhNS?3wW7wLHPd@#
zdRqZt0ZSd`<uKvsSxOvw_>v(}{O9yG@FGhnU|-f~60~SH^7Ixpb>Qd(yrT*_DC>(H
zJNyAXV_5-1A<aR+r38uYe>H+2$&Q#)jw}+1fYHL>rcS%SDLk{Xw8Y=}x+tj>mMT&2
zx4VyUJ{-Bz@7UojJzmcw`sCB(vDaIgELw0v<3-!3`^bAJ8>qWlyah*D;?2xP?}rt9
z*n0G!nXFt7srUS4^)sv*@7x)Gq4+b!<oN{Hat+!V*K^SO3BZe6`;<zt9UP=vM0f?9
zafrVvbl`P*(`=Y}ORuQSg(un<UhQ{zQ@r*~q}*EdfPC3@T~Y=uu7G7xX0a2nAqnXp
zuGE%r6|8D`Uv4w-xN^|{?v&m!Yx-GC%>zo<JpVq<7jpVOYGK}7=rVinCi5B9S^%kV
zHK#_Hdq8D4A#Z8aIkZp*O1l+*DnvKurRSN35}ue31&`||EDrTcjy}3gM_Nq(2>joq
zocK5RomQPhf{mD=7}ayl+ND0DmITio9K&7gpkS38-etNf*$whlpa6Ci>V(crZ6dqj
za`3rUGm4WgnE+Nd&kWL8#enri1F`cEfXxPgT^v*-3|u_h@p;<etGVMk6h-PeTHF<M
z;GB+IT>I!L@#r@feNOALWW0J<uJ^I(jwt6FM4{h@qTVI_k`WQ0#qtC)96!TFn1mQX
z`nY;_9G#gsrbD#tF|LeQPX;xa_rq_<JV?-$f1A_lRI`J9kJ!G8S(5O#eSi_zI~Wl8
z-iCH9#vG}ah&*AaJ-y5<1{x-Idiwc!K{}q>FGn4$d+d`dA}D-7PZ2D!vR9A_y1}>%
zM_x1^L|exgIOzzF{3$}HV~s#^vZI6^eSvB3?n>@m>#{4s<T*>yHBo<U8`ap7Y0$J#
z>xk`V9N7&4>~|)z+ZWg(7PjAm_RtT~;6ctd+p_u~fz0grgciaLN%!EXi%_pyqr>p_
zWYTVOJTboa6SRx)=y7~mVy&gEJ@>?j_5hyA&LR4&W9ZE1BEpkj{`(N_V?JpY`6&T&
zniwA6Yqw7;W}9aje|UubHv5EI{uPb;^<P;VtX(B|D{@zNYgd*&V?R(PCA+*;RTvH%
zl)cr{ki8%QEW%-PXdR9?BARnpuWZPRWvuTD%x;vl?*K4Kp5{`T<9jjlYZDi_cdhBC
z1FO&(i4{n8?|R&Nb>0&f`Q=8s;17%=iT!!E&Ybmfx)Aclaosk5<W9rO9v6zP0_sjj
zsw;qjVFS?wfZfr;wlPW4PjmF}jPVhpit(IO@xE;q1`Us@_BC&vlsMraZgc4J!L$<2
z;<y|}*sXdGy%1cwRi&OF%7Ew)n&qXpS69f{u%PhTg@R~HwtDH^fk)|KZK^(cYNIn#
z!Jtd-&o_*xGEfAl1_b=rBvGZuiw1=+<c3&%NXUv*dCuR-5L>D%asTs%<Ix7WOio_X
zll&bo!E?*R&xQb|t~hw5-cPfNTse)TRlzXwKGFkPI3$TOC+Y<#Vwf-Tt8TZ3gwj7e
z<1s#VpwiPS`eD+E)0}0Y?0Zy+=1S;aF1I^lSc))8pS**(6U{@;)PMi^Wz?bkOlBJw
zvB;}Y{_U<=k~;}b&Sc|v0GV1w><Kb#R#nyMrt|!|Vn=DqSjIT%$Rl7A9iEJE+#udW
zRlmN_Rfi;X6_Lq~gtETD66{B?=!;en)<5EXO;mhJ`;?oSRHGVmq08!4Q|0>?%2Cmb
zqP*2J*kr0P2oM7+;4Aex9EPU=SbM!}p!vKSr1&@iH=asMwxs3qlueFJ?nY+KBkZP8
zv1M;RzUp)RG+xg)YKGT@U71@@(Yv}e{qU*-RU!`bhyW=JiSr9=n#f3_f@)eLaZb9{
zw&7!CHdp!XZ<dRL1y=K_(WGT5Oz3&gRun?dCN1MU+X>>Wz%sxDAtwhxKe2+2USID9
zqdAZB<_xW+5$AK-3U+_QyqlhL9GKrF;M#{(PRk315m<nD(xXl->iZ~spDCuSg_bN(
z$?r~%Mfhq3=tPZUKB6yDE~w^AQLzeRUtlM5UD1KN2DyfB>oaN!hsb?Od4%F+Wu>2P
z9X5HiF+FlNZi|@X7|UFTy>obUIY=t5q(}9U!!7-q5o24vzy^EalzG5xlyXp#-eJk}
zL8TaIkthm>p<OJ0;<MrjvlTj}POy(DyjQx%EYdz-qE?^n^rH`9-0G$Oja+)q49Pea
z?3i*L`h!Rx4F_e*t{nEPE6T+Z;~N|yJQ&D_d4k=3q~Aw1efVn+LpAzxZaq)IjbhrQ
zsNWtH2+One551niDZoTaHCbBu4Fks8S-meX<OVdp9?C|v8Gxlyj+mPUC!PSU-5VMP
zkyBvO=fDU&UHYZLBk!h39E@qVIr;3y-2Tz2B0aT*D=E08L-urOd&H%Km#jYVZtT;6
ziSQFvl5R}Te<T0@5BcWYapU5Ih;p1j5sT6_s<^Y2LsHl==7O#_@3jWXLj|I>lxx>J
zl3NE)qZfCWQ{<+m0(J6e2*`DGw2lreWx^+emCD*no=6R7Myak<jmZ6~%w`bdS`<F&
zadH(g{saiC`n$033+;rY_lMa}B!r(ib8XY57p2{(H;&U+unnF|S!m<nRp4-?s#%aG
zdDFt6EvW6e<QqIe*(|~ay%^>Ze#OL(6A50qN_$-Sl=kN9TSUrfJ;oLS;^SXk<bRNC
zy@O+pM7>HpqdFUzM;rIPuSu1|?)g5CX;bZ+`KC3@5|O3cTVbIS)u$a>itK-uz#o@2
zj`X&puAv1$vM5DzNsoh&yWg?*VvSk08LG7oZqjdzZDp%dSNS@5-a1`Xj%|bYX@z);
zx+hnhx{w(6^ik#uyjih$LG*dH({)oKmqkQ&w(!9odA5$20%`XJIN7Xy<Z+OzHfII4
z@)bO=84tP}lhnwMpFnTTc9M~9N!H`3WIw~q!rUu)H#@fKD~S8OzP{a|^C2uTS8>jg
zlWu-BW^qD~#b#XeMmVW3FHUHq$yaHsV+5q_+7Xtq#p`BuaRC7TFM697n`di{c@p>a
zDV2vHx!v{j-}k)qJDB;>GCwa;S+z}RLS(DVPHGt=(jV3Oi|?Bb(pIJBnT%ssiTk?e
zLQ#9L?#Vz8<TNWVkIQYebrB1fcx4^T&B6r3Y^<qK+l77b%Sro{B1}NpQUp=Qh>T+$
zgwdYE(NK6^l|R;H%S1xt$R7I|H-$)%uyXOcK7E&^wtLx5EG(fBMDqP~mprmLWfk=T
z0bPz;=bmfal*PC8%3-t8g`RT-(FxXeY3#;ppTU^941Ba#&v7k;X~M3`h9=O`;lbJa
zx^EMtt<^ngL3kQ=`~sq9I!7p<=2+njd_#N8`~YDTcwo57LI}ug2R1SXDU2LXu&U~!
zBS%7<IZfF?m*PJ>(D4lFd2{%)9x3LwGCk9Soqly=By4>au=gCO9=CF;o|CUe#1%fo
z&?K;}?L#PekW1bH_31PA(W6UtsjG@|c^Q<wDN9nyr2hg-m|d9bo@5*mMv&1x@b%`W
zI|K^h2h!6YCfwVnm?bT6TWpoF(?fQ;s*CMQ(e52f1=vqgcbVvl{~_s<*@I!D9S3g@
z4ciWCzB*(AjH5dOj-d-J=1WEMG2Aki#Wf(L^g~@9V_|iYCj}{gfdMl+^+cBbijRd@
z9V2Fze9UMXKYXlxgzatP{vNl>*r;mJXIt9qqPI0u-e&Ii@$bPiUK>7XCJfA;EF>`c
z8V;*~$~3O1A>11uUWGraJMJuh#ASoibXe+Wf{xvw=If^{`RLllXI{29e{er)tfToO
zsf!H6Xyif@dWrrx2i&-Q`yPIQ)dC%|sG$DUO>j?#IZW_n@`srj)VQWRTzS)smheZ$
zvC!NO#Z1f`j4by%!|Pfy*`BdXIVTdAG-YL3HhFPF&et$mzkR*t8=8Nh;~32yo*uk!
z+WGqart{}T0_P9For7zat`w6cjb35PhUv{C4hQ+YZ+WpU%8Wx7`G2bs`;Swg^YWwu
z+I{L(iUwj$r~9V4L%F8xT@%U`vt^|fiSBR4*9qRE1S(fuB!5JAKfoY_be}!#AsuZB
z`B)7CKOQtx+LE@h`iuiuz;(k;Jw-Mk#+M;CsOlySy&TTlA5~Cm=e9M9O}yY1%Zr9M
zq<rW=AlINeFk^hZP?w1C(AjbCs^bmZ9mjBauUEKVhrLsPv-S&zZzkpjMTn{o0z;PA
z4OD)N{`<Q87w*2>j&0!{*>_1q`|(~r>Z1GAcUDLloBx+|8j_T}s4c0M<U*G|G!uS^
zaN@9)&8pOardzssobBZ=O8fYR-=V;o`$?l@E}CiI_&oI)OOTo9F$bb<&w=bdl`k;%
z5q{N-QHZDMil?KUi`ezkoeJX20sh;4HrDWohO?l6Z2Reoi!vYhTvU~azFZo>8UQ?Y
zBYH~Bi@C49t-d$$2GR=H2VY?8uh8}#Ff}RpFbemWGSyDsj?0tR9N2X|rWEr)>Xn+j
zxfdA5#j4oVo#c6uZNgw0>7VleKt?WhKUs~!h_WC^ROqd9uRqx|Ba!}IQN^UDMWDva
zKv0hHOz;_f6M7%*h1OAY0J5)gkTyw+c?JBb5^^qufoe&_Rmk$V*J?G*h-I`Ynsl^r
z+twKE;VOT(>4*PSkN%}!q*RQ{sINe<de>ptAujU~eid6`nb#i%GTkQ<Y+X!Lf(D}A
zIj^-8KR#ai5OZ+83KFUU87~ld2VsR_5VOXe0fE9_87m8VfQlj9dke~p=nycATW>zR
z!DQG>IK5Gv$L-jG&z`QgU+bqA=?C&Eub!^(>iEGr*1*^&@zYpKP^6(M45#{fs!kr!
zAw-cy*mg@+a=y5U?sKuIcbE$w_U-j^c(dqxbpk*77qf*}8LptvX~Iwvz1X(7-hfQX
z4tMEp+Ox;;VGd;r*#qNcg-yO~70P#$v^K<Sqm<k{?)JIJAp%;LPn<eEf9g&kUdYiC
z$oGzmVE;A#o+t+~07%fg*qk3FLF6cfwlvMEF{I4c8NHcu37)!ev&vz&cF!K4Rc%1!
z4i>tg5y-RkAvktuJ2}3}8_|@dJu`k`w9emNW<Y1}nBjc$nKP_f=WJ&pe{Z+;D~kcB
zQ;YhKcdGDogT(8P6pB84<{vLi02=Gu>#LU=6hb@Zs)W}FTJT|$t>c!<%#{jnR!Hh(
zZ>HgW@u??FZtC*keeJ-@Uh1ga0Z!&ON_N=Dl}?e4t}4ay236aKVcjIPAnctq45^25
z2xk$}GkB}8pb&&8Ix~?nV}e#fwq0O_EO$#MknmLREnVaT){v~wnr$ev9YI6S_Slj1
zwz=BNTNb>dOn2~3-Y?XIxfLPOBWpg(ia9wj&k4Bhc$9s^`ANO_+n4LQ>0-l~r)w4r
z7t0WTf@sp~{!n*GOQrkozT4n$-YM|7e)55qsdA&I6VY39jE=5*vPS%j4vIdiqvbWz
zvW9f4mMX-J4TCVCimE7R+Qo{Wyu_azOj`;os&5?lV&HUqgTnCEiSxGx!Zm6_MeWrz
zFE7K3(#vMqJS%*3R8>l-t-EM#&18iEzu+Kb24r0kC?#t)I1YqfSp8zdoa4(p*XOK(
zAj)<JIE~<8&%M_WKreOrw3gZ}U)}sqHphzY*w}9l;;oI;r6x*bRDjg+5}nWUqnYY$
zI)JtMZW#d8`?Zjj?hmV~n}6<{m-IA%uk}1JC7AVcF&n1x=@TE|i(3I%IX%ilBj9?p
zNQ>k~>KSM}GsNW-3`?ZNCIZ(r#e5DFM*6F(NbfqW!^svO?tgq0a)Z=6PP$VjDmoZc
zgIL(en{_lB%*S>#b&&ubB}+DZBe@E*<Z_(oCHFCwjAKGCIr-?a`<#B_Bh4(mgD+?g
z?*V1$tTUL?2FOgX3Fl`+AhJ0j`L;d-a%>z4d7T5Z!b6pW>VR$>Y>}n(pi9$^Y5)^)
z$=UG~72t_uPT(ei?qGVY2{Y6ckZe<RB_^|87g_Cnd)RK9gxz#$<jx&2`bUub8JaMI
zrPqsCawDk$PCC71m-|;=Ys;wRg{1c>Y}Zlt`tgiHayyKy|KOqj%3Mgborq`DwQj!M
zzrkmNR*%t8+*<MM?1ur5IB$m2Su)NynJ*dM|8Oq0Mc<d+w1+>LP%H_H@ZR$G&GMEo
z^%8B%oFKM_F>}K6XE`~qsxa(2xz7d|5wVB=zKIyNwBkK0$d*P7D7|*m6LB%~Z@cu$
zrMIEh!QnFOk@!v42R^%~BZz-{@vOWsSNzM&pqSvb4Yz#vf@k<U_rpGx?7|{cP>foq
zYZ718hi3pRAXShPlbL?CM~8R4<!z3L%9KCcRLl0v)8Ll&Pf@FN_|(7^sV_VvSH+(=
zp~7uzT_YiOn@?gYVd)0@;|;s)+AH_t$&{2%ZYAHH$Lku7Z;)@L7yLVi#yI#x%Np6$
zTLjobU-&qZ4a_Pl56FraTzXd{PA4&weEdI>L(3lP(WR939c;dGA=Rew^y+?G7gow%
z#J{_pkKa1kuF=tRP1OFLAq<K>7Ys`lkG`FDKBh~_qA7Ycriv5)_pQ>RiVGG~s32;(
zbFOuzm*eQ6zj3BdebYXy{>SO(gT*p4?@?>w!rJC6S3?(o>+}a=f7x)aTt&;Jy8Fva
zw!8b;#w5gDL#<l>#<?(vtb8NEuI0eLeP}OCt1mcP6J{Lq;e#t)VV&reT1ko7q-Cn>
zKU8!^+(s9}*h_e|DvwK+R2>TaJ9mZ_=l0_KjhbN|JsR{#e;r_w7}{)2`Rpu~eqh2Q
zMqh`joviRzcE|JhTwlqzM-L>0HEX~GDaV`s%0K_4;{WAbpVF09HbJEy7;oGfhiSRD
z((?E#N_gd_<W23qcCd`UGPOw(hIsS%Z(RJ#7bq&m$myn~uaeYuxwvlD=D$%`G<0-y
z3|8N;=xn`s4fEz1un%KZ^i&`GZaw`gb7!T-`h)Fg<SDJZ>V`;|?|PZzg77^R^2_Wh
zV5a@cIb1#*E56ohHWL$*$TK%Ye|v~{JN1$q7jlB4YQ!0lPfh-s$-8ofmszx=mzgfd
z-9;_+@|b+rc<t)_TZK@Jw8;tHtPKh02?vuH5&n9bu;aGOJm)4I1mp!H_1X0XbpFnj
zoAEO9{E7Q(-)z48^uQXA?!Qx1C>kE!E5h_*JFFYFrRnJJ)pEaJ?tT(Hjky6DOu=#i
zV(B)BzY6=P@dwYd2E!Jn{;#}DA^gp!!OqD+P}DqT0Sl6u0C69*&lNvAft=YH2X}f>
zv=6jQD)_(hQUJgrSZYWrpgF_OZ}VUPBVw-XJ`EknpQHeP`ALkAdf5M!vG{%FU#hAB
z@*Nf>sPHk^BMi0;i<nc!FTbFL#VV75GcUr7UYOUT(T%=OuJe`v0HRfRsL0?8Ov3|4
zxc`l!?r(X0mY-n#LA!*FP7DBIM|9#CkDBT7RAGJPZJB$>D6hvlNh+&P7$42uT6bjC
zdUa-w%C4g3(c|CP4FAe2-6)<3_ZevFOGuK(?8u5;!_@<-DEpP{0Ao7{Q0}`pCq=*q
zBc=p$fLsB1!pGULtEt~Ai~p8a<GY>IOJw=8E{b7u5m%UPA6Qur1Rp-uEUdIIz{o&k
z<27E9DHfQA3)cXWPzW&sQo?-z&Bc-e^4H@Yn<G-Kp^c4aTCOk;DXOUtaR+~HF}%+Q
z6Jw9uC{?^ci!l{u18DaC%J3|nOE-yOLHC;iW+(u&v~7^ChxozV=iz5NLG4m&93T@~
zwvUR!un!Ba0>GE)91Vmd-|K1!`u{Cs^P9kD;4~KPokWlYin(v$-TTV9`B!tWi`tIf
z9jBHS+<auQs@y~7FwCXNSf~dcWJXWEX18tdTEV`v*orl)pH+TViP)N}(ZhRNLU2t<
zTIKgjbEdzrY}KCvo-H2*`~oQutbH!*N#Vsr-<k(tkpSjPP;36R+U}QBg#Vp=q?+5y
zQ;7+{FlDR3b%;;~Ey}T1=Mc9*>V#PMNv>Pka+(59&Q!mad9&TQfnDujkrFKr;=$ZN
zz!W`eB@5mel!hvLo@0FX5~f)?n6`fFc=%Yd37Q2^T^>drrQNnG+BGV7Bk{wP8!RJH
ze&V#d+g~E~DZJPjp~Xozrp`|;BMepw?`riHbhkC4xCN;yz)7LZ4!gA1p7B5Qj@SQz
zV`a0GyDBPC_@4QiMF=jH8U2*HwF%lnj*zp(v$?Eu;d|;BXwBAXu??CD+v?Y}3tIp%
z$GeGx6%jC$TwX&;fx`>*g?GKWk1Z3iw-mZuN~Bg@;Jo}WeSn}m+EU1klR<VTa<>}a
z=PC2lWjFflb!=dhNO$1*P?@#EfDy|MU@F^NyCs5)HdYbJ%DTiTuDJ@S4m{BgDu5Z<
z%1Nwe=gw78`6_FND)aP$aaraiINzxXD<Zty@r3zliF-l`HsTevvKMxMfS}$5uefQy
zk`z1(bv;>{V6iug$P-AFKI`DsBfTZzuvWERPEIwS<yDDFl9lPln&O(`RXAz%1@mwu
z><VTBpnw@1hwo@Aa7!D3i>=Y+$q(*Y^*E$_=gWBc9eX;Da>~#t!)PbThLl4nE^@v5
zg0T-uOQ<W6|7w_(f|JQ<JE4OKZ;H+!%JtKQh^BWUindab<g@BG?D}F2C_DAoOqr$W
z`VT9`UOwn^&tFh>t1=4iLG`6Qhm@hZA{hrbN{h^;GK|?kMJAQ)uICDk@wMi0Mr@Bh
z?yAMK$Npf9!A(cd*v_j{mnhIPS>jo{YSWzx9>*HC(>GH+SBpE=xyz_DFg8CDr{&@<
zV3^1O<}4S5w{m>ysD7^}o|;-w&bIsE2Fq}{RKHGUrUipCGYP_6XqJi|JK0!2(D1|L
z9h4IHgs>7{Nr_O&>J3*{Z#ZD<L&z=0`FnGc(eN@|B<o4NCn>A=K3KE0_m7w9$g{<`
ziba<9b5z2(h&*Gb=8)?lHR@^FBRi`|J7(Q`v&^e+_Lo1GsvzG9Kj9}lKsOr)Z@f<(
zrlnK0da*$m&H_e?BpG?^&cTohpDtVZQf=pR%whM0j*H2wZ0rfraIZsaCKHoMH}c3K
zvf~4m@l?gU!^ra8IS*`~eeBX!@`_YRI(>L|c;&9Q*Ih)D_$pvtm|od73zA0@l@95b
zg=VQpG~bplp7Sj2GZ7$AUe;_xBHz_+5pcW->l4C{Wp~#H<KgNYC&|*i18;fm$3)3j
zSc!4)0bT3*{D#oS;}d&NY7DOBv`U}%E5{M^tW&SpMsR?WC4x+P#dh3_Z<0wfu6^}-
z&#P62M%HEN#TN2hbw|d|6=FzR$bG%=Fmt9s2fo~Uy?QLBIhd<En4Wbrzy(uYz>Gwq
zFR`A~bt{HZSieyo|Akk$8QJg)Y=<7@EC?aAVvTQ7rM|%KCeqSDAMnvuur*o8z+Tvx
z@31h3SP(FNZ5%)aSjL;3xirrR8cP!WJ8Ka4TPzg%D?d{m$S@>}EF@t{707Kt_NHfq
z)W=kePl&3URYXZkS8hVzDm#1np;bfQdquT_RR<dFXTy$J=Df@Z&>4&_8gtzLG&Uk#
zUkCdxqT@(qUk^{EYydKwH9}x*7A31u(P66-f%=wF=+I4t;i#b}g?glM@_m#Vfd^un
zWYHzX>h*Fq49A?~c%6{@04~2-eJbx9l431MKcW|a=b*9?oy*XKJ1F%g0fP^R>g^?L
zuY%<%ti)h)JuHZi{xWaf0Vh_KktvXx$FZh*bD*RM^<|@H^0Pb@bUu2h)k0G}Rhb9h
zmW`S{)?jdNy@9kw`gOsH2U5H79OQ5{6{ChRbQX3~3}2}x;FqGl=L@?b^Z4kAe0JC|
zcdRA0+B_J^nuQKV6OMW!*ODVSoa1#Y?3_K<o=aA6&ADJ1Mr`Ic)F^xP4(9ET+C+N-
zaiE-DHC=)PFTOPYWV6StS9#krpA`Fp@qU{mHwNt*RF5;}>%(*?k(|k%X@`3XohxkB
z<83AGsJa?Wn{GY)N#T~LR`k#t-gH&A``yTkDh}(QI$Y*ts0;pjyLg1Zi{6Hi!-`KE
z$3L{C-+F7A!F#mutc`1W|BlX<A}iWMbv5rwb4G}F_>{~u)k(Ha6S4b+Z}UzAm-(%q
zoaSdv<#4%!k_eeKSH&Xf790tEy}4!eA@`--{d!tnugSI)xOB7Xwb+gPt+%wfwQEM`
zC9C_-8|raVEy#A=7+xD;dZJO*eb?Ny3Ng~n8qR!SNk1Ofd-@f;`YAeYW(?j2?()6(
zQd3z{BcTXKQF{xjGl(yq#ipF}iXD%m94>q5CwPccQg&ch`*vQX3!4h1Nv#A0Bs<_O
zKwej0oqwx0yXDN348J-8pjE9eJg^x%bk;|KZkFD=6P*dVhhJAUk2JAr*L+(4ai10M
zG^)UYmSQ)V&ovv4NjfpovUB0l9Y>+g*iW`w<@cBLMJn+b9XRcqAn3bG@iYHialOBp
z2pKI~CXHeFxGyei&Dtlys|^@y{UbM58fl7bn6EpaquN@HARIw+*dhsm?u|WW87ikE
z=r2}EyE=3S_gC{P3f<OaekGh{O+Ry?XnzatY{bXpa{hM^?giQYs)LL44SOrcB-13D
z3r-WRDRJvFROzmno3=6&RjN**%VYzKM2JvQes@F4!0V3>uZAa6UX_fUcY2`)Rt}Qt
zL3;$TcA?G@CdUQ{j2gCRj!x~Uk%iDZZc01aOt`v^$5+?hLQEn=ceTB6Q?445C_~d(
zW0<MTPssFT59%;aOoBvhOs<4Go$QQAe!bPvXI(MujB&X+L2nW^!eb^xgpafIaVwdL
zlCND=5vx<T%nmo2W*WRH{!mG1=aX}|{ohFnzL6FDnOAze`Z8<O7(-PAg#a$N(fmcC
z3k}dR#fdmb4Rgbj0LoT#o0q_HYt={jub6({&Q;AVLGyd_qxKMuaNQB>MiI8HsyjqA
zliCXlEAoqLHr~Ls9=KNQB654{Zf9Tl7IIakumFJ(a6<Evi)OEkBXlF)y^f9kw5eZm
zUSEEPn9nL-on5PNyVWJD04N2JjXJyQ_Fgd`f`rKr;3WIFvVlF9J4Bp3VTiHm9coAV
zMr>cu3bQ+PD{Q)vAFImPEewbg&zV2j4O%A1aOp-&@Cm%Fr|nRnn=|)QwdlbI3G+5m
zEqhdA3O*?Am+Lt6UYG6`Jy++z+oW!GR=If5)tC3AJ<2FhjvIc&MS$-0gz?L6F4M^g
z_|S>W=UmIi*bPg)50n0U*nm_{i_OBYnz}#ft2mpwP4;|a@V3<I*oj{KNkK=2gJYcA
zKXqT?gXL&rpN$eTN!;We-4Q`uD&(kM48mrEZMyZH(ylk{L~c*Zn9n+=Y}U&#W%`G2
zU&rX{UbhxTGbCRns+5>>Qa8fG3%HWJGBxj3KXD7|ZhtSR&1Y#SYBS>D+3IJD%SDb=
z^w2D`bTK?3_cl2TzHDzl{Yk^$UE*E1T$);+jC2`*@p)aS?VcQ)eu|POyV$W_dqxhX
zrJ8j$t7e&O7w)~HfxDublAqD6E4%LmjumYUs`B)Ds;-Su;ZMtDw)I`J$cyw68i<@a
z#4>f{9iPnUX}{tj7>o%<;~`jutBMmj%1F=_!ja@P8Ba~UYP#*pL|>%uI)AqF#2TTB
ztIwWNbg8WGv8qQCN1!diZ_sP<v|em&_#_b}7HY`0E`mdT!{rPIRgO^OLCjbzQK0y~
z`M66jaxIGQ^-YvIF*f;=NxQB<<;MKnphJ7N-j(9K5$$<N#5wC6z7%uaTYwx(aiJb<
zqMBOY4}Z|(`P%+Niy?k}?y=non>AJC;<`$-OGK<fed$OzAM_>k$H88`peF=upsGH{
zvDavsE|piCvw7AZfrSld9((IO63IJ5uQ)?PDN&rLf&}yph+))BCb#iSR%?L-zuC#o
z68<WwDuJZPO>gKVEyS)hY+ro^>p;7Qa(<4|CbM+oct_><W|?c*gvwKEbrqAd^Q*fi
z-p8na_EBrWi4`%rBg&9jt>dyp+j6{s=<c@=X8}bxHsMqL+9y7Gs?v9xLW7LNkEsv4
zZ7t4NgFE#!W|zg)XJ>9zGRt%L!1^P-7`%54n<0p9<CesFGxBqasz2H4nGJjP)A}Hb
z7?)Q~{b!W7A1IZ38z(Qr_~v-^+j1t@5ZZcjmAg6VEV+-63-;6M$53feEB1cNoNmsA
z9j8lQ?hWPMbc9>;X1r`?XZGzit6N9?y%ovvrQS-k7<C1A&pyQXdaVjm_L1}1xh-Zo
zZx>FF#u%BnS@2q#*4>zf-@a|>c%K%QotImb9YCtnyw807q^OIsu@;{Z<13MV{nazS
z7vgjNg;y`#p2>Nb4Q_TE4*5u%{ST3uUy#WDcRb3CuX24Sx=eJFT25Rr(mKFrWW+1t
z*ThzWEsoe=V<txk&+NstS!kWyFT9E9b-m@-!<QlJj;G6xXPF$-?!M~q$d+`ckEhaM
z4Qf515EnvqwKqQUI$u}O*1F&`dyQbE%yBlK{pYP(g|mrTW4(6}Sytm)diC04Ms;JV
z%62Qh`Y=Ox`P%u&!;egAZrwciILdV$><uSJdIRH7ZlD5-d}$G)QFTYWW`CQ_&W3iA
z#JO|VR*6tyT$yHuM1wx7GNf<JPX0UPg{OzP7am4%<uhNMP|9O@yJNKzn~%psCZZnM
zLGPmh6let8(F^fk*%(L+-qz?6O=zyvoWhEO-eUm{IADerjT~}upmJYCQX|apPm!hQ
z2w|@Gs&>K;&h{4aAGkU*JejipfskNjJ-_Jz;hmrE)-|{bWoQWdzp!apD7tqwp)<6^
zbNx+iwPTu6_{Z29<j0UIGd9{?Dkuj2I8C~q<99GdXwRfV=dp<nm%?)UWWKU!aSKQO
z0=qdU(-X-s_-LmiBun<{)<bk?S@lK0PuLR5@@sio-haAknRi=mi{ueyX+I6bI@B53
z3#v%oTwswf^tN3ltPjsU%5zo>abEsp6)UUOJ-!f`BfRP7f`Th3#u9)m6Y;89!JHYj
zH9tqY?&;a*`A>>NMdd35&#!G1_nkmU);U&q`@75b7?+s`s@uAAl<~Xzi;ZyWY}*ze
zti$O<UmfoZKX({~09O(KATvTTREDKK0WszGID7cViMit8-cE~hou_zBR{JaWjzV~@
z(uhY~A@-*yR<A+G?$sXeLjz|bfU4Dp5Tt6-5>++NELh^g?9H~xY@u7<@=o^tn0)h^
zwp%AISD1e&0!ADIm6ym35NhiV<ac}VQw*)9g+JT!n&}{vJ2y&C=XmBE_6R+yxmsuS
z+KeiGe?$FMFRuB(1YwMOG}w(lTaAZ+5C9uhmfX=T+n%e@xVc%O5GMXCL+I6f{0+WJ
z(k_+#w+;~}kdjhyeV6NpV>MMbnC=lzeQT>AzeP)nE;>4ZDP47g)@pv(JiHvAD={`5
zPiF0_<VW6k+?+;@v9;T&=#n4fh1=>NAh{!yP}im}w?_>Q{JhmvvQ4kv8me@&szvXp
zTUHaw!P}vQk25~R^7Q_(#GB_YxPl6cbp~BUQCYzHVra9~?G=!PSzo@$mVV{dRuR**
z-Ev8$k8$nv>a$}Ce>HCRHFo#^BfKfRg|eTzLoWn$=0P3pHVjo8=uFnNeAF5<_|ET9
zN9^7x+6UXIKGON7?D}@y-oesNiSo<ZSL*%vrPYe>+?>-ta;z<M<`VNIUpdqrYy%G9
zn6q=supa>vGrBydHP>UH!lx+HckQ*{Gm~Q0sawu)42Dj0-&`vq99;&L;+Pe&YpA=L
zAQ5t9Q-}MVUarxLZfk1>@=Vw(8MaVWLJE9bS?OSa#=<+8C>x9h@t#TFys(%kqO{p=
z&6ao$y;aP*55kDab`sk)MFdG={8RP6mKX4O9pDG2^FSY4L7u71!}|?pW_K>ctcU<b
znU7}<!+QpUu$R=Otryg;FT^~z;mldcZF?_k)?!_<XHagP?`_cmu``#oF2bhJ?TtOI
z$wbd?WDr$6PUMcUW-zHvPVW5u=JH3N3NCxo8qM(-zTTqCv<FBV%r$nZ{0G!V;z3<F
zvwLPKsV32^?0HAZW&`V@Q`JI^3?Gj>7zDy^g`UBDV48+=Q#r}|`fB@LkgH1aj$lz8
zZl^z;EN{$t@567J-v-;^3rMHL$&94HsNT784umhXj%=k{L|_fu5M|K1Ipa#Sm}$Ra
zen6<CYx}K$z2yOpQ%ppBM?|GD29CDDTrX0}$nu<5FhOT!7iipx5qoVRC8Kh84(92i
zYCNJ8P_U&ZVvUW>6L&40toK*QHI})V*S7Vp_lV8#k@I}?45xy#sh?NVyM~<90%UL7
zG=w{Q(H(b49`6enltL9E*ehOiWigCrRX11kM@7${i<FZ`NUVS7Prpkdl!#A<m&vk|
zReI7g`>=c_g^yDEYA17p&L0cUNrK;goBHOC)vaP=@$otAv&Uo>!ZCa~eZ_Wn9?b|t
zqiKfHHVs~#hm$ZF`oY+?&PlUO4F+!zv@SV_GTs?N*i=;(^LC$mh*55UKD^7fQ`iZP
zon+rlTc2i8-nEus6=snWSb1s<1}pyHo`?zMMB4Qu?pu8g?NTR{<JQ*)2y;i2AXuof
zJ!|9H&=$inF)48tgTdm)5}(q=7NSg>e`EJp%(ncMAI2ioGS>!+B-?^sDx>IDH%Qm)
zng#=Eps#+Oe-bx-2MRv5uyeWN19SQ-_Q<z$QEn7x)OLp;KzHKeS-ZCOVi(8RlvB|Z
zie1Tc6I=cEJ7*Y;46*mWz|P0&8PO8Z*4wm5zC?xfB7}e&XP=&8>}Q_0BOR&suyAV^
z5YmWte&=aCu1q;gR_ZBN-$K?+IEgoRA0PLR&62j~hwp?KtP{F)x;PKAHY8-SyrRE9
z>+liLO$BPEXsvY-f3wJxVm_@sJfX1V^B^MM!99K*-*~i`OC0=}B=&BBDdoI(oi_tf
zCcX#GNly8AvHkcn=DXe{UYl!plJJ7-14q_z)uET~(>Qp#^z=Sx-McyOt>`uvBq`O4
zuP35oAIXjw5tId(w%jV7O&)Ka5Y5gGq^K_-<!nw!$qZgIbkZ{tVf|b<SN4dB;NVx(
z`USQ&-0$pTT9Mw)!LgkgCRaZ=PuC1cP1@^ii+Tq``BA(8#GI!b>v#EL5nHZjp-HK6
za#GNyz0<l58S%Fd#7EXfM+uc~kh?5NwxXQ!ervX!aQNBlUe8X6GLLe7Y{!^=_lf>+
z7y7qsC`{qrt9$Q>xxo72ucfQA0hI%CHOk<{&fH%3D(l?VDP_<KGb8YAbVssw$sQ+k
zJGgxxMTP3z>b;A4nz*gX549!n1XZq~IX=7j{=1%{M}16>x#q6VPkrE;n!1st&gYMM
z-d!KoE?Xko;no}exR)rVF4&SbkeQ#a-L9Dx_14Ju<SJiRc0N76`Io!d%+v252upBB
zAMw*_xOArR(l+DsJ4x?TwpSJf2dzpXt->j&(^)TQ1tBltBTlIoTC90`)<rNxvGA>~
zf}Cvoxm{N-8wPl?Y4BsRbH`P>5rU|l54tBy-t9qI_M-V!cRE%iN*~%3l$USmQ{qN{
z=4hmx?ip6^W-=FnWbSXiR&@&j8jhaG3gOiC4U{I`TNmS%m^UBZrhYdS&PWv|8xo)e
zQ@s*(_C^KWnwR{ZCMSCD*t+O<+cQJeBA&(x`V1__dLc$1CKaByrMRWnFmrYL-#Qqn
zMzEa<Bpo1U#zD2YUDC8iMZ0Uj6u+C{e}F&L^)`%jviydi1Uo~|iTzlC^!7a>(IWH<
z9jA>$iv&POK5(@Ww43l5gH7OL^U7&gZr|34%DH?}9&ZDxA#*8bL5*MT_`1X)MCM@e
z@STeG&)DZ>jtZo>!tA{Ld<QH$!m+PAosxpxgl%UgOY=67ZOizz0@>asKRwye&Zru*
z;XuY9#xy&yyFLKPS|p^k)q6FeB{pupduytdouI6UnZ~WAH7M6lhu!|YjOeT^XjVrD
z=oVVDkPEO>kLu;=uK$^&_$Ob)j~r;}?DeFC<wbzw<&-^V3NAS|2MG<)4=zYMLu~KN
zA$Kf|2Go0S{IVm^5PQYvi4{FwPF8ID?1kzz#6qJ$ov8jA2CfF(9y?dYsn0oU%QAI%
zkqcZvhphdK{Q^@ph9KbiRp_q~;(EQ}y`w9R|Li$Ms<LUkZ%aS%Hg)IT6Q*03c!n-j
zP6mi}E>L2g0r;<v0}g5XmG7-Q{*&i8X!XR}8?#lxF4+6%@BxcmIPqPNsYhpy-X+mX
zRxDy?`k;k4R*&FMUtrgrL7m4VX;A8C0f+(t>Tksx^sN8xS<gp%yd3+k8hzqCWWPz9
zAm4~ogq=BRf+2(hmY?Yg^yC=mYk*jG1asP*7Lj0^W0yGUy8R+|%AOsk_rI*v9y<6+
zbS625L?Yc3&rZ=@zXO&O^w>~i2qs8p%0m^skE0UJDHPyNaA_?&(hw~C9Gf`8^c-!+
z_2%)L8%IyC@=~Hdj_lzl4ANGmA}7nyJ*K#2|8npVe@nud8>=#R6tB{r(gB5E`iyWe
z<*k9pAubz0rzi@b`^~wzxmo^p<L6GC)5{BbeBA00tT*pQ!4_=C_LJkNmsjZ$_JGvK
zapdx6JNy1$Kl|Bc-DGBgag1GVcweou`40MV^>~r)R=pJq?pF)2=n+q=n9-Ho^+|H&
zI>NBI`RB_=88l@2gM0t+=kaA8wHuO|BHiiVEpEBz7YLZ3ZRs-GP=%dJ+J3thvSH3y
zMuW=+bH!}%UipikAJYQcR=-l^a>ts_<+mQveWEte)&KD~UH<GVR@kpb!WQvuIf8*Q
z73^XsvZSx+ZLDQ!bdIZs)b#(&UHE5qvqKel#B)Z{m-|;~vmbC5f@bCaK$iWHtXuIp
zeDCsX+#Tx9MyvwpU53bx)03~M<TH6jrG4<S1vt^45_93y{U5N{S<kC>`f@2S!91+W
z14_9*W8`0*g2QlsMM|V*A@;6u+Q$ujKpcUdalxGqEVhyr!VK#ZSM!N>N07NBmMtF8
z2k$Q~m!Fp##ohe$ejVXYQaZ5-HNvdB`#%3SrS6sOWMi{e=Wd#;<6Yl#elGy{co`td
zb~o8uM8Z-EHk=P-`>J?Y{x+SmQa2gY`_EQ{U9i2(c7v{<nxMTXy?~xudD>4BgptRq
zXPo|cVQy2l+RpZgiF(@p#=Vg`Je2U_XzJS3KcxU}7ktPn+!H-H*7N#2@0t_M)xv(4
z74;xCEGxP}su7}UL0lbQV9QmWp_mzWt39IDw@thA^2-BykiUi?SOjK*fS$-Okw%W7
zN}t>HLF~j6{U9F({p~Oz7=8AYW9}9v{%@T+u3L_*zvYW0j#&N0W#e+pgjpaYXksuo
zT&IT;DYt4E8Wt5Lq+ynAycOFBm=tk9H=3VjTlenOR82?z+X%jyx{-)w|MRoq@$*&w
z$Z+wNWgd-s6)tdj!h=riJir4Te@Qk!eeJEY$@Q~sHH%gX%z5TN-WY?+#!OF`r*-{-
zv-Ya#mrZ_qE?DuCTx5@6Gbm;Y{sP0d0esktmB+BMfA1WWH*d>;qo7gG$2{<a7^;WQ
zcF8nf2WH!x$$|PG_6aaWR(yef_lCdv;;p3%eFr|IzLLEAsq4Hu-*8mky%l=)&l+}r
zyt~qu``u}x#oTV(yE*nbJvOg%_9=NK;&x2abp~c!XQm2Lk7*R>BYNm5Hr}>`YV~A9
z$y>bOm9L^TzP3chFX7^D?lnHD6Mt$){=#s3;Z7Cr<%^fsv?=6KJ~O0IRWt${luWHB
zYbm1VdvS9lBqSBF^Z+!G{*OMLN6Dt`?GVph+j;qXs4%+w__AC5hgVMvN33%X=+P~)
z`7AXnHKYOfo?gKJ&Tqf$%VH$_XRCwQJFah~Uthyptp?5hp*8F1&@;a3@JW|>g^mkN
z?_HVWnfOW<*6qj_{`lu)YS;sfhBQl?358J`OXnCXf4&>DVxjj#{y?azs<RUr>b&c1
zKHH6EH}oM7#IbVA!iRQFB@k?K-dA7evh=54wR?j`wIbFeu<U;EuI+!)i~32JJ?HWd
z<zY81aS(KVH`jlh4)mv`mmvP3AoZfkrtRmfvFF_N=CgLCR?>gLS~cZV=V+bK;U*Pb
z=*mC8FM{gv#DQg0s^S#GYQYWXBe1kt(8TobmZE6p(OXZnNj--1$IeGE3S<7*-T9R<
zy2Ls9*=*MZmego~G(+xPjT+`n!pO-Nu)(C%K|)Km0H-_G$s=p!v~Hp&UpOoi?|+vi
zb)v;6mkyx<R0N}t<FvuA#OliTRv!N;5xyW%hM$5pqDR{iG?-DL<&a~+ZW+vaKh_xH
z7QXt8Fx7DBV_%WC$S?m}6#T{CTQIRyBm3!`K^1^e`lu6p9(EeUISf;Fe+R2IzQ^@%
zv4du<$%)WjfD1p}I)A_J7w^G75B&~tleVvgH1Z-^WNAzpOaAuR@eThbH36`}-x}|{
zFmyY-Son{nx#-c>>aUpu`nhieF^5cb%5HYYLh2x=;JnZ({WZ+O0D#j1OIlSMbpb6W
z%pU=$BL`31{+1A;Q^QhG70|*L*zOUO$=tYi1ZeyM0)r8Ead`7VQ@s#x8umR1ufrU_
zrKW!{-KFgQ^d8rY2ZNB=utw+`PFvZVu+V7pm90)ij~o30;N7p{Zl-O{cUW@p9Xze1
zfFRFfpjXHlaJ8kBa^C<v^e>los_sLHe}T#VcUw2U_%S%IlhD-?hZb}JT%V~dbs2Hm
zw<z}&2n2C{rXFj_^+)$X<_rDknT;oR=Tfhyq45fIvERb>CT_?n`phq|v`8ayI}nMf
zX03<vbToKx|DN~fY9L!ddo5^{X&&F(3pUXin}XTTu4+OLZF(tfzU1I5Inm;X(F4b$
z-z)+uf<wNxthoqCa|kdxb^}v#^O8m`4q%UQ2M*yE)+M?}Kzm<$@yr{b)WABk_6`H)
z_kFh@a208Z_h`_~`5WlpsNZ}__8nOJRa0Q5LLkj~+~_N4F>v*JCSpm4=N7|@LzpQp
zklAtvSbhy26o_MN-1ysm7_@z|l-z06@8D9F+9gCxpR&Mnf{CVpmAAX>`}TwHUDDz?
z%?)rtOtBzm--E>!?87vzw*g&k)(6pp2MC_YTXqPm1VaN`N}XiPW?B>os#2t}DVT+|
zajct`U|d;n;7Y)tc!hbBA?*O#NN+$+NKB<1{Jzb{sXM?<QbJCm$Ff1!=$;5N5g3A=
zfEX!l)AxnV)X|D;X$O0w0NnyE14M-h$heph0dE$peDUji{Yu~1WM1WstBTLb6g{U|
z(}(}o$qpm2P`Pm5@_P?@d!QUffb+i4ogHZMolf~{-Obn98}lCIY%Wq0%6{90uIdL3
z#NXp$uXJ&M`#R5<kDh)vNc9HOU@9=*z;N+f#lli<6C<G?{p}<{<PS7`U;N3#FoDbA
zXX_V&2a=x?5~<!m1c9eXid%2`qmNzkLL&9LJ$f`Ay&K?Ee-{M;Za>&VqI2EAaO=Y7
zo<c4+zMmnSk%+EVZ-rh<K)zA~2Fn}V{VwMJ4Gv{kDh3Yrlo=jkSX|DP%MW1>$}v-%
z4s*uyMhp(*C}?jOH1Y)|^=HI9Hca-Y=%#i(N`C_blk`4{^fAEcVS}8EI&3Geuer6Q
z4#p>8GpxZrgrDLBa?c|RWdK$_C)nq<)OS|azj_zueC6Ao&$R?YWU4jF1NonbUUINR
zEG(7(hEn+;&|ZC@70OG5u1nIpRiKHpa0;g7^{J5p5ZL|B68m)?!CbNw%0sG(h=tF%
zI<O)S0kOrR1--^jFIn1;S|N2nXX1s9by43&A>p(@pD&O_{)tO<*TVND^i2h{2-#cX
zM<#)|`8f1r`8>qAd)Z_SP|84?n#2%fvkAHYUv11`n)~DL@2f%JKAeiG2dfLQ02ZS)
zk~oc|T}BU|xbj_DeRS7_%s~7K1HChib}Rze+dDXl6^OSQTWT;|Hj@I6G=v+y3sCZ5
zZrLU7$q@1h7<&-h-m5_Oc2SZUn}A(1hnR>MJRQB17PJm%v8vfB+7<NYi%(5(;5wrb
z3qW+fZT_+CKQ$ryIi7g|xzyl)w&gR>wIZ?d!?MtIY1Z$%em9y4T!@OBtp*}@?@Jel
z`E_8g6xo`T;=s<Ar8yxNHl)OY-m%7@cPy-KQ3fpjq~>Gh={^Esp(D+nGlA}B1kYqD
z7PtQ3w^9DinP{wL3J*Z{Z$^H)2un@_(|@XO1UbWYk_)@!fb;u+y#g{)Rtt1*X9PT)
zih(8(An7}Uo+Y4XIF<}+E&z}*@0)R;t!^v;CvQBd9zuOyv{;$+69wV+q;Hmslk7at
zXE3^b^I)<x!F8$l*-vOGVC8*#C{qq-FW3c>h<;nCrR0CiMH1lBH-Fm|&e`wXG0@(9
ziwjY)?z=c!GmuEVCi?}}6)_JM)+ux~nCt(?-n+*`)pq~mBbAUQNkW+pN|97Z4jG-O
z<j{dBVk(s+sU#A%l5|!HcZyLGB_=tQ)8tGi<S<1Vr$We#7-MGk?z^V@#;ESk^E}_z
z^Lc*X&+pS8HHX)>_nv)S*IMuOUhlOo6b2+>=g=Wyw_*3+6|q=`PLFH_Nxu;p7)#-{
z!R%UNH0F5hSz?RE;qSX1map-$8~GXM@O>`0!Vs27baa66Q9CQ_9W2G92fZ4+j%C2P
zW+3k^KEkSO!Vioi^veub&%v!h9w0kC5qlOc-4sfk>M{#1TRbTnz316-=3)QyeDLmp
zA)a*Xw3!^nyUBO%WfU@cEclQ|*r5)9VzoA+nIgi4b&6MIQPFYX8fd=`hu#*O+kG~j
zAm=N^AvalJ5o5zV5)>IrBoF*S@1$?CO_V%xWcKLRr`r4mQg>PBN!Vm1h6zjV0qm?$
z)Z}NF!7S!`0-gZ9(_^}{9Zk4>+#sYJ>5K{zvfWS@Dte+AT+gIItDk8h?(>5B0I_N6
zNMVM(&)~-PMNQEAvpJbJr9<(vlMqOX2X9abH*pF^YX7j48d}%Wk<-v&_&qAerE4As
zyA%u)-57JP^t3=-$|{fqHysLVfGB8YWh6T^z1+aGUO9LZ1IBpqR`;j(krEwnJTRhv
z0Mj<OEIxY!$|eLObmC{V(8Sn0IFAnMhTZX7>~1kI4xWSHIxDH4DLvBQp%X9hKL-r?
zQ<r)e{Ym*~I})+Wx}L#5xq$c~Y@lY&&;edW5wOI0fZ+fl#D`dz!wvaR(IXb<XshSn
z!N9j)^tMWLE+qHhgRqPsbWD7V@TWLYzwc}Pzr{DH^&sqaPCSd4M<%#xArhSYtrZVt
zERqhp?upxOn<pRrV&c_|L<_ge&V8q7Q=uR4Z6fUS!C*^)n6Bce$X6O)gL)b1-RKSL
z-~U7zrcXZP`-~!!Gsp8q^IU3-C#^6%=jrZ9LRQF*9NLqoXV)g%`ZdluX8CL}u6dbC
z4Q+8puP7#-3%dETZLod8k$%|0qs~&SQvfJun!5JKn-mgzq%FYrysSC)^LOQ9I)f|D
z_87TmOqFjE(q?sT`27|UXgKz#<b`yUF)&VmthNX&@uv8U;eAW}bi9QeUt;{QM;rog
zb;7;N(|XG|%qpZK9B^jA5Y80uLYhBm^wS6Eo0}Lk=U&(-DD|f(X`X1%kfD2}h}wa9
zI<Xww!X*q0F_s+({MKKVa=suIx@k_%GE5Pv^-M;s(s{4=0f)8ghTm%q8SJOPG#9@F
z^ahVI22SS>^IH>ss~8p%IJ}_3PSZYiW8g44tKX3n8#M+1e5N5h?PFVTq%loY4t4>1
z8ibCcpV%&Up4QCO;C>?Z&NiA>C;(=|ZqL1R*c}>8>w=z_$aaKog6(}^l?kLT2F@cf
zi2rKK(1Vc@j(`sCC-jI{8eS7PI-%GimPzF!FTjZhfn!v$hBXrc8!4q`slgBFfgdtk
z5KEO-(^}@ZVFOX)KMyYD;r_-D2)ANJ$B@1^>xGbVU_P-&hVVt(t$Hvtijv6a<_9Di
zMhmt>zG}l2`Y`mQ6onn0`w!l5_$c3B`MR6G@QQJw5`!i(eHIY5zEIgr4B1&T92^N*
zr|Jq~i<&FO!i*oslUBo%NTBhG2NZCbO~`;aUa%DC`gU^XAl6bQ3Z|gu_O2o0Gs5N7
z2s^`}L%iey;7=6<pnImBtXqy8TmifbSh;|sFb~2#{jrt(1lWXbG%1EI^vQCpwqPC{
z;}ZCOi8+rUn0#`+tCu(+V&?-nhw?SJ!-KJ{;ME0M{;w263ouxXBYLfl)wn=wp>Qg(
zXEtHj+5MQKGj|uugAc}C0eLG4*1JNA{IUxt%lpckft23&CUQU1xQCxmpY2Bn&@052
zsvH6e+srO|Zz)E%2a6P>Na?4sZJN~5NBFRzK^o)ILFsToIzA9Pzz-)P3BXcNJ)s*U
zZr5OpN^)w{c<SC~(Xrj`Smf{h!5KYxct(lq1QGGm@6VC&?H3(SgniV1(qplLFbNrG
zE>7?PNK}VM<;ck)oMI5{iNlr<K{+6gbqT~Sz{YXEaxJ7*)YlK1hZ9?#b`aSMAt-3$
z?Z=@(zhK*bGjP8dxZe!i|1TK0!k--svcY~K^b$duEqEgP$=sRo0})Z9Xg<W<rU^^y
z*qb=s?$cM+JX>6S?B<PY6W>-H)J#7tU+hh4u=BaWa>F)sH>jZAo?+i`rnP>=BfVr<
zj$U5PK1bG+8&6*wYs_4pgnoXOo?U*h^>R(6edXTpz2mONv}}0!^3W*1b;^-`q_22r
zd!UEFmUM2=o_PYCC}wfFSg($j+2}G!wC(Nv;<%<e@wH*Vj?p?GP}1D8@6U*TDC%;a
zl-5_)mqzRBCiDqK{DXX;^M&D-)63e4H9A$5^LvnouJ`Z#iG~FM>L)4|Z2^{E5&|Wm
zU~C8}xMN^vf20BFX1^c|an{42))ZTj;mQl*7Xl28fk>!c2Q9{S4dYrhVKy+)Mu;fH
z6;=~lX2Sykwy|1*_5dj{ZiO)Y>w{rpLYNM;72*kp!^(fySojychJL%U-^rx^iOD3@
z_{9>@%DtmvdFpgN3*og?N2p*ctr^|t`D>u({}!w~L$1(OVV{@DrS60SN4uOirl+4<
zZR>DG@rLxoJ(>ZQxF~!b;c-1u20=iM+2XF&OCAOXqC94tB%dNM>ECcER1^;{*U9`f
zzUO^QDFkgZi-z=z*i3IP*6BCxwR`Wo>`y06Ovy1ix&ilSrbG67d}guzb&>w=pVL^{
z4H@LF3%mvG!70tf?Q=iY1$FXR?+F79KB@_fuH(D~cH2vVr;!{Z_j1=;X#cRR41E-Z
zce@jN!)3#<{-v7eo!ZZnBXyUCyuT*({LpcErBp#fwKX`|bg&pR^oiW|IY7vi@%^BW
zLdV_f!-4-0^Vu4In_0#`UCg^=B<s!?klgNL0u<h?)V|^OLyWA>h#Xi4+}Ln_Z0Mk+
z{N>;uf69L8{G2DV94kAR=oBJ7^M&?LnL5Ph$KOFptQlxC|I!}oL<sX{+9LfyJz63j
z1;ENqBJ{mNZX_Pd&;Z+9vKFChV|$<04$%}NQ(y_g*|DKxdNm~BPa!1)EZ52m)mCrd
zOJ`Jmx4IH-&}&7DCB_f<LTn&%$5Bm%KShoIo&gdL`VPz1wIMyDGPtFJXaUzWNEQlS
z>)<@6@w{{V_(^GkjbUbrVoiu@LDYD%98yYT#iD+x8tBAo{%d+~VN>^j3I@0IXbQND
zI|{TLwK%oTvI~?RM?{3!(Of{|+E=q8e)heFn$*2$1hFNDuM6~1pegKj@<&IJ^E;ZO
z#efEs*>~=I1AAHC)5Brsv^56C)gyEgwl}W2%M@FwL+-2ohP(0LPZGuY!&5Acj%7Dv
zspo`tnHW1~aM%VT!_-3hV7O>B?dYGJ=D>IVI-E@v=;H_XL3j@rsd5r9Gq8W8#9y?9
z9+Y7IydgoyJtl=)M}gdf%_mo4a}NTv^lLx0!7$ZY$eljw5GFc7tudGks^V<Mh;zZI
zBUL1IJBt7~jVremU4z>t%x02>9qrJ>r~~@u%A7zRJTkq4giPANsI5r0hH##z?~*>Y
zf=%lyh1>y2HBy-9{O*B8zLo#nw*SA|yGllEUR{x;;f6F#x2`mPEOY1Pg?Dd0Oe~%{
z=6yhEr{=3eG-i?@=TotoNynrTXq*hDYh55_<^iMhC?*Iyqs9#PBgSZOwlc(#wmY8A
zP()h8;M@5cLvh~9C+J;`GhksR)lH!Ez0M&}(~*WKTF>Dy3VS>(nqE7c_Zo5(bWp($
zEc7ZvFND#!hP5<eN3RSfo<2U<+r1`q@ihb8<J|YPC%IRR2uWT#`~Y>@hXexT_e$3M
z$>3~q3SSuOij1#Zl`8$%{3EfX`qhc`wCwC{33oo6#Jx6ZRjDc$lUrk-SxH^6@@5=c
z?6s~~Y<ac9_n{LKfczGN{J$p#A)Yf>)2M-Ow=v<*F`QL-6rUoVMd?y}lHgq*<8j>A
zVPV%(#iYvF%hw!c{5VfJDwu>L7-wn#@k<B!^>f<1Hc{>YX7?pb@v~Qb#F?wjK4KfQ
z!i{F7r&UFTJ1Hhg%a%G^en$HGg7pG3B^Tg%Q-LZ|3(YKhCx@9^JR`8B30(CKf6hli
zqnai$^r9XWGPOoZw&cTnL!eJ4aKQXJ2==wTT}R3f=_llO4=n0FPi8M_r*WsJ;=FYs
zZV>Bjg`-MR8yUPY0@-Q2HjCON%G+tj7iIZ|HAv#cYMm#}4cc=k+4K-=(ImBbMh`A7
zy1ndEmKvsH`w*$ngJr{9O2%}I2(i7S5<)V$E6{`0-K55WMVea^Zm=t_sk`ia<HrN}
zDPw9&q9f|v$mDX7-`W?!?!C?srv61G|4-1q{IijF&Pxtt+U9yNWw0Cr@ZvFDWO@2K
ziX5@!H8u*N7oS6Vv*SdY_1LcODlyaUd<t*sQ5rh`a@OALo9v?y$n%)3SkZJqGV2|p
z@NL1Hz>{5~qwbKGzhqxUYnkm1_=Y?X9{ndT{F@X0%d?Mg`9m`czjdUf=6?IHIN@Kk
zhWT%H_4qL2LBfD`JsHajC32@y^Y)|f&?iU>-5udcYy~cN7fl=h?2jHA9P~l3EGv+f
zW({Jed@LN8f#aOdp`(E}!U`#qZkUf_&P`b{eZ^Px_@?tbTlNe{(NLSn=vgm1b}>B)
z4~fT=K8*q1O$^iFRG_i<k+S!Ueip4L4mN9}ksnWf{EDaMDY5QW$;Z?U$<u;ipLNO0
zpK?x7BWVuZ1KLC!WYz_#_PnN?fIlB{S|_bVG7bgJOgW(A;SX`cx%9+#T934Vh`%pL
zeZ`DSU*ZmOhIN~Er1L|9&Xl2)VER3z_zWbuKuuaR$EBUZYK#NOuPv;jl7V#Sj~bn)
z&ScKO1+))yui~V4*P;Zjr5Ir_hiV(H!v&DYWqX0FaWH!clH<QR{B92G#wRQlXm|FB
zrm!5j@vi+h2PB%D(}v&wNn-(9B<cnsLC_6|)gV{79!@8Q8G^4%#G^vG7<~<K<OKsw
zhBlD#j}vwB$UXhgHu)Qly}vUKc&kD(3zb9N2wdqz_7e?Rq_~&d_Y!F{_R~Et0NNei
zDp*c4TEXwQmYAN0bbKM!6JGo2)(Iym<2|5`7KQN6jxZ2@e<*8202j#IOdJ>^Pzre1
zV8^TmOvv$+AHOmpnD~~^dXD(Q2tPUt5Yu%#p;%);9hpe%+*ytECfalpcasNni8$o#
zYNRba5#Jof52c+YuoppAi=>V_?&4F-<Y621o5H{uyJCSZ<O!tr2%grGpEDnMn-5B@
zwqsQy%#RR;5a2Fzjc6_D?KHNA6OFeiXBrh!lMPYi9@}09PcDUl-X~A66su=2&C3Rs
z%_nwB`tX~nZ$I<l4^JIKN^{pi#>BR}b-Hh{<!^Ty`2Vguoe;b|IV+s!ZMn81>A02h
zQMT+=G$IZudr0fgM*@5;eczMibJRW6n!7OnM*P(M8A~IkUkXmhj%j&RAm99A;#s58
z<<2b{94BI}7<PVPHfS^C1?5;c6lF!kyeHrd_I}M{QQ1@OE_rTLcI87V|2e&_xbrs4
zttlwToqW!Qcqxl?zawoChp;VTMs`N1XN}dq$;(K}V<ldS^nz?dw+%;4He865M({^N
zIyDG@Tl)!8kvrS5uxvn+=11vcf%(+fkK??C*0FA5a){o#5=)TAnU(mDM?#uqyBNtN
z{w=wDDi7}j=4AOkQSRKT1sai9<C|Jz>#JvHijQ_|w<0E|;s^S|Fpevenoqq9N&rl!
z4N>49kKTtF)I5B+!2Ol+xEW`-yD!g+e&W7*jg;R59B#s?tY}QZf_)NrgbP%pp$NfL
z{fM{EEuhX^rTl2^S-sFrIR4rP)hv7`o~2k!o0z4-8cUw;d4@+mA)itqlKtfP%as=T
zhs;`<k5$}qe@0D@!wqYlezWfXm#q7n|2Qx3Zw3oBD9HJoIbpDGrV@DNX!VY!@LxcM
z^ZS7LLe+`WbLQ~z=oQ29cHYaXqATxKx6<FVWM^&e-s;*iKS<kl>H;r!>U8Kx#zO^o
zWKTNSc6WHE5X-lPKx`RD3e=Z|%D`uRqF#jFtK<e+7nQe~Q{2=F-BhTYQ{uh#4y4B^
zPcLq*z3sX8#>QuRE-orjxw2xiw<8Mow~rzQtyd@kTKI`8#@>wn+@BU&aZ`~rf4t_B
zS)u!0OvC9a2b1Ufz9#gLv1!nY5;22IoxY1k(^_bJAU=}(0YbkCjKTv%`RN|1G@1h;
zh#+IQ%+sNGGvA~tx{7QPW^TJHl(piK`5q0*e3I_lY`63LrPP`r&fIkVd9n=Bkps9W
zwN12l)1VaQH^RB&_Z#8-+Yru)57RZ9#m;H(J~t=Kf|(X`c<SM+8M-bd{jIl)cR5o*
zS9{$XLd)Bl`HcS0%>X_v#Xh6MRW#%)G-gi_1e$X_PhjYFg!LI!;1F0AR-9MN&8s;l
z6SmqqrF*37Y)n~tGE{crbJi)*WTllYUu*o9PZ^xiLq#{$Wu#S&lmYReeg!|(4qDHY
zC)^-(wDlTiq4z_1)`LXhed252)=IJos5dCEYNihz<ia_Dtt>|sLA9pUEe(fQTL4k4
z>I#i>)bSOxUKw*1fgiA&tI-oRUUg6%H&k$A6=-`EDWU109)#yYsaPR-+4F>9-Sxgv
z-RFemWk>B*Af6snUJr*DPr-T5fJfRNjkyQfhEjS)g9P4-*eawEZ?l&Z(wAAi<_v4E
zEk*2|Ko>3f9(jiJO52ACe=?d{B}nZECl^5BRSq{b#=Q^h$co7#1%tdeyozuN{}KQN
zSJ65NJHhwIH%ai2Kb1l%_mQ`T`Ou1)R4`eDP3<&+>LOx6H*aDp6_s}(%aUqoNd*2y
zh}vkKD{E-Ids4lvjK=(wY{=348S_|*6}kTiW}FM0>u7ta1Isi5$!w)c;E27-#~`&D
zz*`i!0wYI6=U@BITU-$E7A!^JEg%cOem7M<3SUFz%%$<+;mbhQV?FF5(Z`)bj-WBk
zkO3KNE<!H~0UO%UyWa12osf~DQ2zh_zE0>rSPn{cx>9@?8Luf-(sfD?QH{3wD-{rh
zFQZLHa#v=_HBw~uV~W9zU9a+;bi$^bS#f{NgdJC-7l>Uy9PzZs8ILKMH3V_z_LDi9
z1pa=SW{w(Cx{(!xLYr@No>#*@(Z2j<q<Fbnb8DwdlZC94!nE-5)`l~ZAMe{u=^FEH
z^%2+l)I*DBOfBBe528NthE+XxOtUL*nmVo5EY*8CO!`b%M`HkhT+D)qb}t1MQb&_2
z&jXW?1j(KJaR?w^5n&~jixpV0(>o!ApG_{H@h0}cN^&Jnrq?o(#Q2vGhzB}X306`z
z&U>?fGJeOH0&DlRp(sKxJ4ft&8sDtI({Ez%Zxlas<qoh^urO_J2P36w-ZG>V5ols|
zqB^i9CIzzr_Fsb#9af1HSeo@pVeCWSzX854pX-WUpQF+RFlr7oqS8SxDuiLzm*YHB
zeN&KJZCcAF4utV`y%dy6zShUBfbOU@!0ebdxuPC#5Oy}s_xzv@v8va>RxYLWRTFz>
zZh0VFJR2p&kOh{ciueVw`RM$BMdtgg!dm+ht%s6%ONG_o2|~qELkYoACw}BqNb1fI
z&MZc9Xs<IUQ&1i`8h#tDc~JYU<^~Xqgh1h68EA8&k$AAHR7W9=@Da4J?_Q;VZowUk
zT<{HN<$`@hW^Ok-_Ce^wkna#GigF*R6d$gN*&`$ew<_WY6-?n|N0dZqf1nF##mmJO
z3&<fm*V8~Qy+<;L5l5_QI2Nc+F)w<cTudK+1A$KHgJ*x^K?tyL7%`F90Y6sLYUt^R
zeg!-22%cm(&C${4_TyJ=qHW#(WfsG(wG#}gib?~XH!cXB(wECW0uzPK5(WE8f~Wg2
z3a?Imf)S9?KySddNr8ZP%e&$mpYwtm`}u!(6lwsc%s?m9`Wa3W@KaxioOyEmP*evi
z?1H(t|9d`!KUoqp^@Ulriw*y!sx>~x@h-3>;6h=|8a$akGCDzmb1bI`Jw<DY;k>{K
zP`LD}Q#sQ?T;@gGRW_jQ3JZz;1)f065_&_nZ?A75rDy*KIT+u1@Js9WMB!rwljv%L
za)s@EM-3r>zw!rEvc!!_gf%+TISAOi@-@4ybA&$)YrC3-LBvRaF*N3589#%Ed%uPm
zxIB0ye=5*QXf5QMSZ2^in3;0?3ttiXVSe0j<(^Q**24$GQg*+EoZ)gU`WhiA`w-`*
zC<GwfM?j0ALzDu9%C)u&_Cy~Nvg2ewMFMYoCA*j}CQ$abyN~!hS>)39c`8)Cn>Z`e
z2ag$%$lD;BZ8UAoV5VO{wOu4LEhz)K%T2LdHi4^}GlQD`XYtDsuQ1fVlh|wJy~2pF
zA1gq5H2LN*n<Q<3{R^#TS^32C1n8`crH>T|7P+x4sSsrKKjbpvoAr2_O|ASJjAs^H
z4od?IgXEf{eLYVbXgY6@2Rvd^AjquAGS1Fq?+1H)isKGOv5N@xF$p*l3eXQU4}v%?
za5KbdH`D~0hkZ2AJV?_zcUB2B59?XsI28W$6x>7!g{6t*Xi>os-mw~}14Bp$Lr83L
zM7^EJJbT*e-yFxkL0R}8EICf;bsdnfMY6Yr#6}|}o9fWdIL!G6G~k@WNdV7iQbc6S
zR}p(}cbtsE|LpV0hd3742FW4@dzmfLc5^?+%TMBGe@_Q@-_4&4S>z>OvD|P(Ah16h
z-{tz&A2X(%L)bGr$lN6#a^_Q0X}j<P3K;m~w}A;r+Y2uo=N?2y_~aaQ*vjvg_gFA=
zlu1at8>vW}?NLY@&<yfp^nGozKo{=k6!|TW|A+2fqqFh5zi%u<1A7_hy!95f`k3iA
zoB&&EYXA(wvG$|;1oIv#iDBtrne9ZF;d46rqA_uJ9<#y<7~Jl=Pz~&bLnjYfHQ{Qo
z`@&$rNHIqZLd7`bJ(l{MZq-ENa0Ah@j_nCU&UHj;#a5_u#*i@{&8D%w%JDk*9rszT
zj8rZW@&Z2<OkWo%uJ|jBVm(w11~Q)^ATTWQnJlP_HjCBxK(N7e8EnHIKocEo<&*HP
ztl{@Vq8?#EM90XS%TT^=!VBTdO8h0^))_dpQ!El6Sp$irV(heYws6)ZYQio-<eR||
zhf#%8#8E`AB|ZT}OuV3eldsUQPuSn9)(cgN4LuvNPw?&JGHeA|0;#W|MTGcnq<?Px
z;hS=?3dUm%{w<0hv2zx&*Oo5!->PqTk1#DEiNke8FVwnXRn$PWpRoXF9OXIz7|zis
zBC}N>nKMn6+2vEn8-tXZvck}FTq+hg##<j+eE;O(skvV0hH_s+&$0%5rxP>YnrRgi
z@Uk|Woisj=D|h2qv4{Ss+sZ*=Y3mFu#o7a82SlLQcRHsFc7CKx053G2!M|!p$q%aQ
zZ#KDLIFm!p%ZLn%Qjbi2ux;ChUCB3gOK&&7`gD`yjHOc>t{m?&BHX`jedfV~@Jo4d
zII(4-F^=UTza^2`gs(y!c>cR3k!qR+k;=@VQ{Fl{7rb2`n%rr(Tz@5{CpsfA_)E|g
znNoIw-T;~11b%*JHLY_#@E5ns#eBZuq;z?jfGdoXzu}&p2*PFsah%HDQLyvc6ioCz
z!jk2)Apz|Qq2pR&O5`1r8SUS}%1mmSAT3CNN?k~mN>u_Hu<~B_KNha&@>Kc8bVF(E
zC2gwjEYH1)PJPQ&`;Dd5-|t;(`#NHW%jbuu8+Y$A-0AP%wH0Bq7?&Bn;;=<6?iF~P
z$s$~+Veo3@){(jWL0o#X2H(J=IUOD4tq*O?HIbe*zS-^ELi*V(?MB)}vg(}Iu`%1~
z*C(BdQ-b{uE!WN1&0OLcB|Zen2g}+M7Y^0B<C?*nY-McT5L=2R<=ne5gjN-Y+O5<K
zms%VhO3>3GO!}da-A)I>sGmySkBZm~5{&7fVjbH~`$GIG3*B78`2I&3{e*$@!`%~v
zs@b3J7pilgCn~oE(D9`*LB0GIRqQ^*5Q>7n2e0FXZ1Gs3a(3Y|_n`yC9y*N>f^6yS
z9uTXAct%+ei){byPw@ZA>h|9-2=}kTpx$n_601AVcQWkvnhg5`u={nqj@@*Re#o*J
zt-0u$@wuhq@^XRKWkRPp0znd0L7s%<7Z`%Rq5Nr+1+Nd{^ir*#zY$^APY&(R@J+be
zuE)N9&BQtWo$ZIB$IUOdJTd#M=rntV^sPyn@86V*jFhqbuEPJ%RpH~o9CGGU_@{C}
zc$}Gl=&SOhsnumaeiem}ru|ov6%WdPg~a>}(sG(P;JM~bXPE-=q66$l+IIjI$mrS0
zm<+*bCDN-8!hj=!sNiS$3W#5R0AIkwXuMU!){`&NIxRp`vrMhA_7|LsAZlp};@=R<
zb>&GoBV4QH0@)XTBly%A9Ce7D&0K+Ai4}w`mDo6##g0ro5QZcyq}YC^J%wq4nG*ew
z&K+-Py@^mR5R?kV_Rt9=$5}7Medw*9#P5ze!0rC0JL-_tu=nJi1A6Ibzye~g9`{ZT
z34KCqdFu{Cy%lLYD#9TNNai}2Iu--Vc+dfEY&*zGRCAR7ry?PqYoiYThD9-vi<T36
z(^NUva_Tr{u&g$-$}rkdJgR;^M;SQ9dm4hNyqrlw{eZiim?|iyscw`sO@kfWo;}i?
zf7|K#+s(NX7p`bdz&UNS&dERV4QJScFvR%tc(UDmUg<ZSH-#A`ks`>p)q0wKvGdu?
zrOVmHp3eD5uO=qo4!`+cG`hLvPlz)VGD^K7qsBl{ZJz=$DrvafWhn0K;HzM1t!=cL
zr#H(ue2SosGw~*f@DBQ3K%H+ub|DC^Eol6POAruqf`3AQ2K0bKb)>`kLel0&3l9G#
z_7twSJ=NV13|UYqt%K4xS>(cSwWus2gmi{k=7ZgsQ7EC?l-~ipvVcAl---1?N3-Dr
zYP3e6nFn4TeE%+_;~I3jNc?_-V*klD2*07te+g}Pj*W8s8-c~HT%KTo=yAex-f@7W
zn7flzhvoFb0=evGI9I^qbID9selT@3FjAr$fGc%RsLlK78{F@PssCR#OdX&>+7Tk%
zE2KP1zdN;c5Ba(~7gqTX-yId{_4o%#p2V+gm<7SXOfmGd33kD&0n<fFojE53Wi9cL
zbU9Kk77BT)kpkOl=EompxJHBt0$o8r;pdDG4;qk+0p^DW_BiPT{8t;o{D}LR`yn6p
z%_aA2;_rcMkuO~Pb}o5BxyUat2^E-B#(84TZo(J{{u4k;xusWxCF_Oc5(XAv!2Trp
zgB0M+la7~aKSCEHLIoR?Ei?5g{jVIYFAu$ijND+qzu#4_-+j`5JH_)6sk?xg33hmW
z6cdOAlDCkaf5WBA@o!Fm5#4EuM<@6ZB?Jjxo+~CHE@)tbKsLP|GCH7331Kt%lVvb-
zD8j;WGZ5@N_1rwsF7n$T%osVxV9#_0dc8JBm70p|A`Zx7D#+Ua0lU7R^8j$~4~Zmi
zNAh0p?XU>+IdB=W(j1sO>?r)jjvrMi^t1GI-&f?GL*xmx&gHQn#OjNZfKmrCKa6pf
z0TOkDyLpc1s}nk1axmTd1qpUN3U<BJUek!+!k-IQn9pg2D`bdap@rm0q&h^L*7Dp}
zSCHU@1s9V<D!7g#wNtu|v(lT&@+op0&6P}@^VAeGL$Y|ght!yqn-iaRElIxmd1V}5
z?dy`91~dz4xx<-_71XOh=4`vA7-)|2t$iyfy_SIWzCjdPLroy=LV^fWvwT=h#rct%
zEEg>4o~gr$*|zUZGnSMpZMpgxzn))8O(!-`h&@MaTv7i%JSuBMNSXW_<B0wf80SCL
zA7g;l+~1|x0Hqxt4Rqwr8W24mmNtz`RJKmP;ULaFX0$?zu3xmc-AJi<P$xqzBXB(X
z^8ndPYZLjB=d*Ph=v|o&`12GQEh5gzP?}bm*mqoc>zBDtF1^v5W-)(`iOS5yoN}>d
zq|kLh)o?K^&DXbKV&b<;c`#&YEpeP;EMG4WG$V&#=XyMFZ!OQVT)<TXq+{nGY>ody
zfE84F(z6aLmUKmBT*!D!FyKGd5<w{u<Tg&7f<ue!k++(l=7QdcELoof9<+p}Lw2`d
zxw9Y>OfS^_Ni<_bm<)Lzk)Ly6Ij=eCfM*Uf6|o%eb#T|O$WzP3Qvac3HwL{$1Pw2&
z!bjLm_%rc!JQIPa<`Vp_u`z;QZ=|G7KER&^>vxeLt3dXG^;=AqA=MGb(OO;_>Lc_I
z-<L0ld;}^y#1xYPfrEPkg@<#ZaLrRQw|$VTq%s;?PgSeG|D#u3TT#Mu=cpFKVLLIl
zHC(~)4Z@h~H%~Z?X0B&{M|Y{<`<wS59sRrc@V~kFFeK4tbE7(@0$~xx-yblXX`@a2
zhI6G2=;37$dQQYQ+}*2|VqeH`a4i2Tnb(CBCGj^NKHfkX!=V((Y%4s<RSG%bGi}qN
zc|EqP-#keBV5EO&E%npboNDwkvAL|-1U;))++&05qjHXuF3{Rmz9DvZRw4P%v+ST-
z@|3HvC*WS{D~Wlcy^icU*2l7f?$ZoW?-R!zr%;wAyC%G>VP12%7?`qs{=vgNt>OGn
zV5aw#4J@U+E5owf%h2i0Wa<CO$v&S<mI=1OzX6B}Z5pZ?nKO%;?mG|c!WHi+upaNh
zf{xiN(9V4(i#j;@YV(Qb#%|-UiV^A4IZO(rCwPy7h1MY`Wi=x}Z>d>P_;!_z74G8Q
zvEj;Nrk^7$3F_EFYa~<?Tf<?CC1*^V{}ro(t{{iW6KS0qZh$dKe&u3)vkfoB*{-a7
z?l;SQ=1I&r(euwu0jMv^ICHmeGIL;=FT0*8TX6_i%A<#HZ#ME2U-+zg87a5sZT^CB
zlSRT3VTlelh^}zf^XaX<-m^wDWAW6P>%>ly)x8lkE=!JQzzJT*Or$m5theddSlOKO
zbw<Q&EBx%DErG4N7b!)4hW>YQAnFPX5;PdOEwdSNSWn`sF>&9wkk7=HdYe@GPHXp-
zp;az)SB|JTo+w@6eKvgam8>4QQGSMZ1)<wvFv!1mCHR+jC0II6`RR=vr}w;Xm=FSS
z$2dW<$AFIU$gH@dWRQ(OH7cq$5YGX`H%Gq^QiM|l@(^e(U<2Eg3CNR3@K!!l4!x9+
z?h;ZM)LHt#<j8H){l`Y{@Gk--9DfG&iGe!!CXr?$Z(0JR7uSG#iTJ;hg8K=ylJfid
z#kxR$H%Bg)z@FPq<Su~@+t4>;Bp@YA)88|g78^$+DUxLK7_2aQzsx+bX@2LCm1}Um
zFE)}#A^F9E=B(xCVE7ugZ~5V)NPqoMOPVI4IHuA@l*bfny;qlOm8Sf0@L+6OkK>iK
z4S9#mNtd1LvVQLpF|wrnyLy-S?`T&O?xL4(R;3-Ww`V!gqH@NGo990At`<&x-&y>~
zQe68}hDR5Hr}y?l@ll3!Z<u)G=!<%%mM?D;trcLpna1C6?nsEQHIi#eqw|3Qq1u^#
z7mRjQ6P+~P(JeD|)L4q^jf|!)k*Q=-Pa+M`;rD_QZN1fMv=mGFr1W}<Ekh;GK3(VF
z;TNxw1e-mdBX7c>mH1I5Ful|ZR8%=U$A)?z*+m5m0G$*H>DV<hBc;VGTWo#MMWj<R
z4ryOS7q#pE2_fTr@*9Z*i!>CGw~a2K19ypU{LBl?Lwa`DF9k{>uMV2gLz9py9I1}R
zv?LEKdr`(4>jdC@p}_0?qDhXkFKmjU&4=HYga=Dg5!%Nx`w1eKixp}C=JbhRxndid
z8+1@_PB3`F@SMfeUr5NbM7qeyf&4J=GDJ23GE2AT6ZdrwfWHA5<c=y@@9lQP5ij@*
zt)Pg2Jlj`p@~vP%&D(6`ZSH;e>E&V<e!{)<f@*eIp2mPC{xYMlKMH9-0BBy&{Fo3;
zPBSSaoQsrTAtzHPV>y|m@&mh>z9E_`b|QzJW!9JJh0T|F5IWOHMgE!SBkC1eW0au*
zQn407fC(BzPM6OZo(EC_%bCcVloEu>gWt9kYh!fkupFDH7ijyW^w+S+7gH0Q%}*${
zIy=`JfBH13R9bbz5&iiydY3iPwrf?pc1a$@bHVAC2l%QZxyxg&;e$M63XWxYDMdZ^
zY6x0zIrTY`@8Mfb=+z72PibfHFBuWCBUMM&uR2Ky=F(NI0ymx)^esE{1${tlb0=`r
zD`^7?G&5{0|Ck+_?NM$x3%a`nBqV%@kF#Gfz5eCKy7+;GZAk$qrI(D7iO8NO+V()L
z6n0v?O>{)K^YtuHHlxw{p|3t)q^jGcw+??ce0#sn5#^1>iFJ|d##wDF-ef)o=jATD
zx94r`_R<5khc7lS*R+Zg9kG?jliHk1jerZB^%!TV(UnK*mRYqvx!$xan0P!p9-PjI
zs~IUJtvOEQYpHa01AC$Z3n{0LuU0dIVpgT~ifCKm>QJSHnvZ^(9;xwEdDHm#`CrDh
zd}r|gH!1NNyjL~h#W^}gt}-0DZF+fcg<jnNUHSCaX?sYspW-m9l6}DS6nD1YTQ{3l
zV5e%<FhO~CY@?A<81Pbqntf^kv9234_b*`db;ly@%2r(3Vcu%&b>;hYi#S*B1}(ne
zaKb~km--w(;;=0utQZhNPeY0+MM1oY?m;c7-<8oqU!;hztOQHyAh0_<F2rOC>e)fz
z>eb}T{@)rCbdYhhmW=O>?1D{zvnz=&b*{+~K}y}fH;J7a1i=GDZ|it~q)Q9ES3*FM
zU4{kNi`t_<eFe=TuvZG7QX6US>e#1h`1*0%oxCn3&;n}}Z*H>}Z=U<9aoxI!XEv>v
z?O)!Ma~!OJABP2NX?6jPvka1R&8c4mcLPX$ABEe;?8ibZ$WtxFATY=`^7aI8aJ7ld
zpJtD+YWu(8>g>7wn=?5k_xF<~+11x`WY&f|#keV7E1ek<QC~Ob#yRC{M=a*Tz^2d+
zP%@dWSb-Gan*Ekyg{kboPSJ6}><H%9V1L=PsC))G1-7DM`S`?xMjZu)BD@^3gk_fS
zI6R{8w43Z^^pEPZdiUb5Ag#HyqF^YNV()_1gZJQgKPgO`<DkK%bHgt~Kez5csG{?h
zWjQwXaN?hBE<1cr{cBz5le2l)fqjxvv1wD?Y^!?+oic!9-!RyOOz=pHNpxsqXTIjc
z=LnsNM7>4>3zl00?NnbKZM<iEvCXj?58j5z)NLqOC+|KX>Y~zvXPB1A`wDvt5bBBZ
z^x*{S-XPtey_4wV6PmlvP$quh;zc|YU)WR3&zcz=D@dLt<7a+27i67Lp8xx*IrW{D
z{OB&XPig62F*yz;7vxAv8ojvt0&$G(voxBNcJAFZ59;i*OO~e5l^j;GM(ak-ANS0m
zQoc-^8N}D|=-LRtC=R%{gFs`HREh8sqSd4}F=n9M?c`UaLgK*u58&q?6@4f8$iBWW
z7^%fwV_1sK#Q_F#&4kGpP{qxmjR$A&q^{fM*_zJcvy3h;dX68j=S6ykOom7^){}<i
zD+t&?+#&aVu>e}jGXqB;%|d9{Ss(*DzThR1Qa{f!tXK^}rCp2&`0vEsPsF`i7mYJd
zg@4NjgcANo;XhgOBulYh)R>ZgRX>80{jrG#=k?Fsv5W}vKXLa|Y3e~VoYvBiGaY$*
z0w!fehY={0f2s;$yZ%^<>Sy?~xPY35$sncI`ECFs>qP;A?Shm9sJw};gn@ZE;{e3z
zfZweMlVzelxjZllrE8cb=(!9F?BWwa^L>y^t90|jrmw;ZQGk3-bU`~gzDz%6dKpG&
z7Y00GN}Fiob0}3N?)`>y1Qr)>DCt{*$<VPG^>((k`?-m?r|tb9qj~(!wq?)SmKNXQ
z((koX(&?QQ_ryorT)v0x)yyc&bS&r5Tj>S(fklh}2758|@|Jz&t!qTcswH+&csd+^
zo1_XY)MATs_WC9@&sAF=ja_$Z)+x7VrPJtD9yPQ*nV}yH9^Za8mv+nb6tTy~0Dh{&
z*bp$ahv{{M6pt>1w~7M<@`WryBT-gz`fSroY1>t7$@#}rLK80a7`;BO60d+mt-(^(
zYq{*KQs5i>2(5Cu&~v^-ZyQb(rklzMOO&!#SFJ7W@=Hxi@2MqS*L^8%Y`*gH)1qZj
z_*8+;RJdbkx;E0MfS`wLhPN}EK>L-7qCM*;xkZ$OfW7^m^$^|%3ASN`hG>cahM&Mj
zWw2%SVBM3Do<jD8u++k6YR2$3eWX62Es}<7AioSND%6CO7DERtcQA+a?$AC+vBDep
zeZNQK+du3l_AtWHB5YIY)>%7ZwlB3^_H6VqZn@YW6?P{4S?RQ|HAFgp9uOGT{g_uV
z_Zuz%=XNsTL+$4xc8y}4?b~qqs3$9wHyJmNPYgV>Y1xcZ;~uJOG*a?xJK7TVaFE7l
z)XX_Y;}0;!y9udYVnb(CdzGW%*>8?UVoLiNa(oNdh0*u+DVf|H6IrQ`KJe@!IsUQH
z(`kk7&S%RxQIg@beEjK9xl6A2d<Z!n`%SQ2pH@a(``OMk|HHVN1)WDxa-M>EUt*a2
z`8V@BXNKx~zFu-wZT`4e(E!{bwMdQ{YqJR&lO--zNEENNh-vTN{OZ6X`!1QW=`If~
zuG<~5Fn6v&8fCMAqDjJNfYr$0Z2B85P?>I^=o=nDOm;RweRf_Bu3USnyN_h9*OZ&c
zfpzDb>da5cM=Q(Dv75Z8M0uLhnapm^8ds$CAwuPVSsj<q&F>59gLC1&I#Wcw7iT{W
z?fLr5S1Iy%{oQNpv@(x*9Mj#qcmB0?Ey1PQVth450I}C1O9zr3@vhjc>{XbwEn}dA
z(1A@x^3+az!!36}kvu=$3$EgxlZfN$+gP3^#zCgn8uwf)Juq|1V^hD;K@;r)p5eyM
zHjwy8YqaOj2T0-5NaM@ef~viP)u!VVYq}>}XKiD0@MWR|7!09X>;WAki6()xQ9NBi
zB4EwnuQ}ixIh{#7M|~aNWHD}Ecxs30?Z_{A@*h>R<}qd^?>Hvklo|9lKlm4Cb8f??
zF?9lCp4>EI`kuSD&O@?pLrvPmyf?XXB?~^vdd%Lj$3jeWRwm_dPWd$xe}*QW*gY#f
zxUoNQRLP;WZxWILq*x_5HhSbO*X!C52%%2kw)JBNQsTF_1h*4X^_xiNTzn-ta@#mD
z)Ybj9NZ<^e&8u93bgL#lG2@X8`e!{U^ZSNV)WX-s=F*AMS<`pA-F^GczVx1!4sZ0m
ztqzxW<v(|{ySo9`sw;N+y2#slPuFMZ8CIbI#~aW|&1RJw&eYVH8Oog!RXcd4P=3?>
z?58eKyL+5fFxl=r3RmTWA~;s7bcmbnM?7qz@MLq}*}w=?u<BKa3~O*1^}46JX6xQS
zvs1bbS0`9qtroEsyRxF>4xEQ3)o0t@w|w)n+A$wLR;DDIExkGVTIuXt$@-5cy|hpc
z4z|A_JfI`4@rwza7;VN3^XuYk#<w4fl&(nV?YQUn-e%2C3ETIw-nT;+Wp2MOzun)z
zBoQ_n|7#k``bv%ZqNo_ZvTXxz(UC(-Z$T5gD@a~82cYw6UOvK61(enV)*|2pGZs77
ze+Ig}SpmsWK#CKZz{69<QVjo)b0Sx1cBCR@B4^c_>bYqFE=?!m$2sn6+z@ifG+uU1
zY2M|r=3^Yb;eNb|T9wcb6f#w!f#v+=4MF2SM406JPOLZ@pJsgVmDA4glO`isnbB7I
z=DPB^4|x`W1o1BmZ=b{no`)7(esDO()^T(*FQlEoJpbVvu52r$@0f`WMd|05mASR8
z1$Nuas>1?04{GZ+`J)G}ZY!$PyVn)}Bw)twlacrQ<~bu0n6y)}$ut6<ua`?#ITqsF
z(KUT3g7=)&x+a}jUwm3?+tHir6t*7tkT)gmyz%SSwdXsnb#-TqI^z5q!(ru-^60}9
zb&EoK$x2tP3qF%R9*8kp^<u)866NtZo7XP(zr0qL>OH~;A4h&P7x&Vc8u^N*XA`>&
z#x%Zl3<znys+H+})LqxwTKvk27U?(HYC{7#!k(xaW-DK`H#I%xP<6eOn4rTu08bz!
z9pyoYP|f(8#dt0<e|trrniOe6az^sCMRU?aWj*(7^6fnPZv67IRIOi30c!0&on!2N
z%dnM>mQgJ%|K10}4?7AEDD$UjbKEJe?(H_)_8L8xh}4+7&XX4vk6YVx$yipJB(h}!
z4kuzm5gM@Ji8@)tRQOkvyWcT3EQX2b3p`q%$FCoQTJFQJNNwKqM=wOK5|;a>W;ZD|
z?=2S@z1dX?OY(4-+P*z9+fgOp=yH=nxx1#W(RF6WXPNZBQ!g*cbvt=YPE>h3{j_fx
zC8hf`&G`@Q7u3=1q*^_WUBa8gapO*2`5-QFpvT@ka^vg6RlEEQ{jgf%*N=bQQ7_^`
z8R&tfLqkxn%)eX1lWA#X`~h2pLBp~Mb&i;K0<D_?tqdS|9Ci*uYio=~7g$F_Yes)J
zU`?CwgZCUwliB-ZUIg6yxWRFHkimKXp$8%BTf8DO#Wuvz_-oU?BJqYube<?)_z6~^
zX-MnyIt*gRzR%r5ZhRy#*r-q;FA%1;VmIKhc4YW|?S$v<h+&sdu~C+0>+hs+Ats>5
zE=)N>EFb(~LN?>h&3unz<)w#uW^TRyR8%+bBG+;(IAsl~TZgWWcRQ$p?)p-+_{~$K
zEBY3C`On+z_yigV9y^su?s7pMwi`-R^hqgMOjtAi>(Z!M5*>$LK9G5Se8P%Vx0-Sy
zg>MhH39I{;$Hq9t2t>2`V5ea5aC>S0#Q2Qnn%_G(H^leV=o7n+J)bLH>`=D;h47{O
zp2ot8*;G1PZz<>9G8GQ(2ZtRaG$tK}U%I=Fh#SN;Ju8}R7gh6jLv*B<Gy<3i*-)b?
zS#yMTbANQyOw~p6_16g7mvxt(V$--mULT@H3AY2yto@W4F{OFa({*>;E$^p)D!acN
z;MU~7zreJZzI&G*I=S581g?Et^U!96PO#6Y?IzjAq6ZF<W*l8srfs{LW5v9%>Dm^G
zz>{y8G1}!#!U@zNv2u8(e(mlO;hA{ave&=qz2s4pvFB3~{_NaqwX&7`$0g-)N64={
zi^dr^0E6(Juglpe_3}|{e)oi3I(tU9Tfb`n8;{sB87I9><4WT>LAkYa<B@rm{c>L3
znlE#WtH-ohJNV(`4J2xr!LY25Mmxn~uYJ7bv1dCI`sVGriS9BvuJH0a?}!erVcI0k
za{3D4)r$KS=xMi~#$0-OA$>Jqo>)naQf-Uv+g11oQMB#u+h{Kx_XGa#Np8Yt5gw9X
z0nzAx0b5MVNp92-VeGR}4~=e|uF&(kbg!-L@y?mo6!+}4%bUDs)}ps#PMy4kU;h!P
z;2+^)zn;{4_OZ0wa>YXfx&~i#e>B*JZ}@YV{i~~HRHu+fw=I>cd6hB0tG8`d=+gG<
zM=hWG<i@!g$S^B6W>GAw48$9GH{)+8Y}*sIA^o`W)z_aq_O9Ia$F+4T%Jk^JF<@e0
zzwr9I@ZePGYo0Pcrk~c@nsT#vn`5%+IP2PNGKzLhl1YE;*l}g;v&811`7*-OE%T9U
zbj@47{$u0!`DJA#>78Mf=ypZFxO<Chj-vP6w9N1i#J~^J)D$Ob0wf6Hj5$J4!t^~o
z*c+KKK|$V*vGx1eUJ;+$&rP8`DtWwaTJTs`?P$fr($l83TvjVLRO&e7y`5=%fu*zh
z(!)*zQ<v$zi{3lDzBET(f~T|4#Cy`bQBtlDv#_+AJ}b>;p1JU#UfFqKC}8Rq>DYRs
zH)_+P4aPTE6MyvWe-{eC4}!BX>kHC-r^DG|$_&g#U$c3;N;J#jQbn@ygY`?tZebpN
zll1fGeipXE!j2PW0+Z?s#7%`ElYvPj&^YHmYh=x7&<~Py3b&%4KI+t4w{6Z+_K~eC
zYsxhzZM$<ev1-A(1xE5J;h%-3F6m?W!j{U0<OzNSFAG{4zVPRXmJolOV*6j@lBJH;
zz3?$HXKDm?9qih@ZD+Xpl6vLKt7nE(gos#g5*a<szl3TmY@da1JPL16l|^87Y^71Q
zRP&{y$oBP%k2_M#PHo%mRvq!qbg|*xYK<SG8n3{=@n471>=l^{Z^n|_6+HZVO>L)J
z8ZYc>s%%_)aP;cylkUWkB-W`tnKtRL*&9PSzH*g@VVRB~9(~ZI^6^cATXBtEx6lKy
zC$jRJ*ynhQKzBLC<HEy4ZCepH*|1=TN3U`Y725Vcyl^t5^;Y><b_}z!Q9YtFt-_ng
zY}_(iV5V9`(dS7IIBw5syWbJh4G4Gk6L*|cPrxc;^L2tpBS+@--+y}{a`KJ|dGY-D
zteBu;N+8#ykUC+-T*^I8fbzAe3z;e_##2QYAH(czmKaZEuJMu^{>`E<ex450+i<lH
z@>Mb<Cd0qiOqILndHR&liMgt0rtVPEw#Hq?8_@sWO6~9^c&n<YH1?l?1lFVP@eYt!
zKSt10pqqzjgb~9y9#tw^$)ukNW#nhfrNq>x<`a@@>y#(Ds=esEIPd)-5df6JXvX$Z
zRDHza10MH2hP`}kzWAa;%!qU2HY?PLUUfVGblN?DRL$;)q^D?s!o}L6?1op8n`lpF
zyg2L@yF%-YtcY=|`+`F_QI)7d560g>$zQC9HUnaer$yco3!Q*0jl28aY>!)K^ZdD*
z+bq#5i4Kxa9Fr3Be{F&0R`jiMi!;H$iwZi{eM4saBJqY#5f4A_twkGoT9bBYylf9`
z@uFY%-o?3a&-XPA--o~8J1_0#rplvdPAZ=&bBefB6+f!$tAxh*hFg!C?r=0&n(NWK
z2BMBEa$Z_;d-$<v<?_37nX8|>++DobaGQIs+ExEIe#@siub$*v>{5nm0;YfS4fjvs
zP4kqsvpbLF6MM@5Kbi`GjlTteBal;HT=)&W)nUZFelFJ+)>Gy;-<R*O6oPcrd;Z5V
zoZohvmCf*OX+9aW9+xkSe*Q^NY@A}ke_Hn)6l<{GeZx`ITzUO~B;#p+Gn%HneL~?4
zs$r$c$j86iVOHEdYiHo}%TB_7mnha0c;Nsh<|(<ALv5P6n^u{nukcE1$`eh!%!fvE
zd*W|BF}PZY^RmvouV`B#_pULlMQq93W=6JyFrx24PEzExG-`u#k9e<V`h>g=agcqV
zar_L()Wcq$-1x$`J>uBa8O6E{EPGg)LX3<sz-!a1m`!l_cgF~$+1wjG3%?-AMh0ms
zQdFKFIdQo!OlYJ#j!Yl>MNRU;lC}+nYju|?n)|2<8P8E+!cY9+{ZC&%Ii;x{#AK&#
z?3Jw$8fu}aL`-9iyYSK%c0H*=U8uI%EI%FYemgzHfAOw&_P3gAQuqC^g;t{JW+qzQ
z*BiF6<UBPU*J^-UzC2F!?469^tBruVxMuALLf(ftmurZVIj*L3scVn(=<LJqQ+A$R
zFu6qW;33V6j-z5KePz67WL56X3Vx=fgjebmiLq(^2&(@wy$;$iX+vJ<y#-C=%P>`%
zBF`~Z3T&3%M&da))Rj+tDqramt#G<)|4n|WEM~pZR>HGA*huNcT~FI*N`$JNHClu*
zZ8k1w(p<F*E_&nkUVNmwK4ad_D{Gf6pP(91b~0z`hwY{w9t}jig~69H<)E0l<jkZO
znu}MiQ@Lhg(@XuEaQqj@^MFZX_vlr3OdI38XWqGY)!@$Y1)+J;2fpHPSMNwZdV-TG
za6I|S<L}1e7e?<V_x-mR;suA6XD?dg9-2ev&yL>T@1h`j_{2`Bf<mJPId>J&M`f}w
z(SIUz=zZ#Azi?a2NNW_g><E54VycOrso?82xZ+C>FZFLgWz4R95lg#~@Nnko7o-2s
zkh#9`siLT==Q4TOtN)VQ`4#72ok~B$mpl=(Z67ITLd8O@daV){m)(_(p6V|XGFDCA
zv;H3YW7^5!(!`EmTX@~+ghrh_sg<?c?lHvzuNNI!+HU=%J(5*pP7%<2X5VQ(pYJr^
z+COPNU}#LXjM=KRLTpx+e>0`aqk;0yqtJ%v$0tYc*Gp(U+aR~-#jcr)@1&fWM!Y#|
zT8pEfCp8XyxDa9Ojx>~=;l^}wLu277eXzIFH`S_i3hh~^eEG=hFD0gn4X&B5dqDXS
z+4bT7W<E70c<VQuP@nbHD4BWY+}N<V>&`dVcFwdtUb{qod{ThK$F*ZTMEy%12L03E
z>Sw|#gCfhZcotq*A5oOqI9<^yTz_lyWz!eq96g)1hg_1pKT}RcVwt>lrEqva{j4BW
z-(2v#Eob}wK;=Gr;3<BFl}UFzfmD>wa#@uhDPREa8K3>m08V9e&)ipSUU}L0kTWD{
z{sxQuYEpn9`4`a41f@Tjzcglj*>~noS7Gad@64Zx5c9XcD3D2MoNzB$o-fk)v4`D!
zFvOW*n8YLP6m|jwH`%Aq_$)=<?D~D7_DkPr^aZl?ER@A8q@|~(%u_^evHHFmFOCe<
z68U`WgBZv0-iJb;SdL6So4)k%f~%%VHx6DiZab3rFizs}m2op#6mD)|hUM40SyHTP
zvSgz^%hA;{wtdlRn-gkWwB+^YZI|_f?Xp7y-0pA7jVoxo@2>FtcJh^J2XPJVpB@U2
zgP#HIr!dq`4jP9^dTw}6kl;A3tGM*m>W_EZid{bilv-k2?Vjy!)-7@dvg&Uhy98N?
zGYJv~dn9CdrVA0hqkK#@FDl&2*n5&+;=eO^hU25wliLuXlcZ}LRm_a*x@sp^a&cy<
zK{IXr2o4^1SC|$2C*j5j4$kBsa&Tj*f!&vyUmBdfT`Py$biH<~Nm}TBDr`yY+paq2
zN2Ni^4gH^Te9$PGldFVp=@grVG0k=h`?m%aTr(R#+oGrRV%_My8m$K2ghdYW-S+j3
zaW{)1r@zT^*m+jny>6{W(`9nsx#?BAtgP#<J{gN^xEk3j;vtN0{yEwXt2IDw*B$<b
zgF4=Sj=TDSTMWy%asK&pa<b8f{Ym+AS3>d8xH4I>p@>5jt?~Cb?CgASx=8wX@zPcm
z!|Nh}LMWqsf-uq~3He#!;T{1tqJ_!Hza^=o(Uc?jw_UK-1z`ZATwBz`ZI(~uytI36
z3SBB!w=sk;B6S>x$ATQjV<b?kTi&zGJhQsX{Q~Wgi4x`GjVTuDG4DS(8!v4f*AN_8
zCGsV`_UU=HXHMxs-t;rumYCx=xos&LK>_ypH<+1gocA4HVW`~J=CZ)s!f`_MXn@}d
zm-VktQq6hw;)$sLDU)n>iqJhojaUR%I#z$fy@>dWi0}RzOFaT43IpiB<GEFyR6Tb(
z7L10Aev(sM?;S1ns4MaZ(C0Gti>>5|MQ!8fl<F?izdXVd{{Z^d3kUr|!zCK__du+3
zWJiEF^<Deacb>ewVc&zb?N=6WE*Eqy5c~Fvtt|%I3R2I}8tyQ9_I|@nM83#;!^PUJ
z=k2{6`9U%+T=I{aeK+MvBG(&Qb}Q+Xi?z1$W$>7pi75o8k-#)XJNM3wN>5Eoxwy({
z(yFPg`{!*8?o{;}>aQ@^kxKA70L3<-mD<Ds-NFul-WBJ_5QpZ@g`9kyJB_EHf%IN}
zO*5Z+M)7Q|e8uidMe<umjq4v9otBdF_+p`>gwg5N1N$dC(nt9%fTHKY2313^q%*#i
zj2;(;EV0wN2I)O-)33MtYO$8f<OkEIO?;j|!@p{x*LfU27}CM3k<xjrGW1Ln2)1@p
zYq$x{dmMMgEQq>XzqRFT)eM!Yrib63+V_4Bf}Rh5{S%|`w?XL(2?`2EeI|0v=J<kK
zAXkajBFEZ{W$t5erRsglm&D&U5eYdn&ft`kb?T#}r11+9n$sDVKG=+Yt>-8jj!PFh
z;oylywxkjV#@0iiHJ{ikL)tgnKmo^Bg*!DuO6pl^Sk}>QR8iaabG5#mzD3f48tV$%
zFkX??me<P<seT;l@hA{c{OX@9X8<zLknp-H*~maocO$*)NuATJGLjzWSlQNjei_v>
z`s_5(u5err^<6i%RxaOyH|3}+I_rG4Z}{1(vx*f?ug1+2#fg-Qm=AT<7&X}IOVA((
zN{BgX9{ejHJ$hRvhF!Iz(-}$K{P6C40#BqCY18gCtB#2IJdQeJU4{9|lH1P=pJ>Av
zj&>eDZuX_IvRihDwBd%kmkG}qwXcDg1f)OaK0!Kl!IMY5AV*h^d6kgR;%oMPzQOU=
zbj$qNIZ`-pe;y7gjpFM;Y<+@(E_^~w+COi`soFWGYPHX6wZ)B!<BAlT4)r`6GpHsJ
zXgi^pICmYfrQ8704R*aWRO1w&!CWn3%NqmG4v;e_Sjal^@$I&C^{aPeoZEXTDa3tz
z?6k}H$+mU2wX0{Oo|}2mQR2L;u}|P+5rO_%_~hT;dtc>wfb!@rAM}#cc}VXA1@4DC
z;o}bO+P84_HV0pcN8@8Z<3SXk3~EBSbg%;?t&+|_zHoINz1@zKmXewrRGJ(xBS^PS
zAwYgPQ?>Ay!z?`XvK)OL40tJz4|zZ;f=Allc%e-X+?;BCUcPWU>^wPn?$XRV$>XP8
zoTQl>C3*dCW{&R!B-aXo9=`A(9&JasHhsL}Enkjy8M10jUu=y%+8{5z^<?MTc+eyM
zYDG2{^QYjkRTN2DN1Zij!MmLMmG0L&vE>TQx>87EG}B+M@pepHGH+Di8CuKsH?)p7
zcytP+{&6_m4hfrI8WFxTtcqtKtxICQz-Zb;RBnuKwD%-fm6{D2Q29ybuAA;Pr#j#g
zbK?ou>CY@Ol#MuL_x9(_GQKcvzw8zqZYmBoIL5jEYMgv6kX(pOTP}F&U_~NOg&0il
zkh0EroSeO4dZ+Yz)mtZL1>(33<zoKWb3)Gnh9s?XdnK_qLaj|<_ePWP&Xea!PsBZl
z#)-TYvH0ae8C*2@Z@=j0KV*$xT|%0P#n7-prCIE3&Q#x5UNNx&-isUPJ=)8?JU2Cd
zqPWf6n*Y$>Y0u87Yet`5xEmKhE$^28jKAx7VoP;&V`qZnt;3NHEAxZ$Xa-0_7<-(d
zxM#3^SZh|5_o`*rR=rqK=yH&xLVNiIQNSJaXZf@J(OZTKI42s*x@JGH^`zzHXok2P
zt@j%DW^CG%1^3hJ;%<%V$Xcd6c6{Q>$MkkHUpH7E=11EO`r@2yA{QTy?O;<CgR(?(
z4_y@L)fM|Z#;38tXOoX>uGiSqBOYc*Z0&a4KBrwIm9JVNRlk@hA_EAE&{W*DktNE$
z!*a|Gde>2KOwTMRpmRqgIjycP#Wro_9<-zGB`>QqFQC1LvYt`Nmjjcwz?A1$<$a=&
zp1RHQWxQ0*hMi{`h)wpo|A)Qzj%w=Nwm|6}sZyl~p*QIuCG;Xv6zL^O5s<2M2?XiV
zI|5RqD@{7mJJO^{??ga)LX9TGw|{5cbMHC#oHxe(^WO0u|72&3O-9Df_pNWOx#nDx
zTWJv5iI5%IXk_?+tc6p(#gYW@QhJ&pc`L?$M|~P%)mNUfemEK-IoP>oArtxwSF$eZ
z<wbDx@k`{&W=rBl5=qE>7j6BlRIu-W>GJ$L!}Mq`iZc!g^EG2H(PH}|%!*a03i#pA
zvf`48dad5+9?tM&)t{>6|9MdSJ+2e)EgU<c-B4y*rU@TWtXtJ*uI!2#?6mW!Og6DF
z77MF%iE;(|J>u+pM65pzmmq`rT*%3zo6SjvfcjsTB36%@QcgrcWGq^AI)%xFw?c0b
zLZ^|gSI<#;2tdRPD7uMe$q)Ql%=Hosjw2QtsH8~=5qJVhrqU*I1YugPYu$g%a^Hvs
z+Q&P5*mOM=rh=yEW4dY#%WNYRF-Bn)KCcLaTcx;^T78uT1v;t=)tG@AJ`ItQnWyTK
zs3JJ%e*(j*I3n4Z)3%!yl6+2?*aX2%2d`O{H7eZUtfl6o{mkyMq@KdXPubcNapSh{
z0`*85mdp`x{h&+Pu{o|-Dki$nr=_#i^!KTt_mPs&#mP2_M`7yx4nfMpQZ{o3iO;<R
zuW3}+EfDJ4^Kr=jt^5Qew|dO-@JDXa>g@2Q&Vh%I&ELg4Q79X~`J&@SvE6CcEe_s%
z*J8Yk;@&2X)XQv{X_}f2UmT-O=mvw#_SgO?G~PZFI~GFQ2H<b<X89H`8?v{+1MLUP
zGDz3$Q-;4@Yr^IVzbfdi!%yOA`wu_;;xoF*?xMwj8&D}ou)!?lb!-z#y7HIUDU&{k
zGN68R-=me3HepAVyrY(4z=+7~NC3$<v~fFxM0dNJ;~=T>AfeJ$eNU$%LMx8>3GQtb
z{FlS@lPgJZNPHJ5hJV%<5}g$`HgJEanlPQ}$FMxjyC10!wFTSWADQeI<i>8pHgh0n
z?ts|eY~we9$&U?$Q8mmwZF?A?D*gP;HXLJR(C+?lQHAwJMk}8u7N^&WFhcZ8Z;#ze
z6}HzXAuO5KC@Jzfnm%Q!-#z7*%jK0}Pqi%EAY(J!6}e~9Ft4b8_0inWuA(mZUF6vU
zDLuEBLc>`IZ<hqpa7YZrFZ)JnEGb5r8x>eMv4z%@v@y5*lC!(McFQ78^VN|Kq+^ur
ziDPW&E#*-ZIkLZRqmj6O1;QTi%yO}zk<stfJr{RT`(Ee8M8NCLv}d*C4t_F@5)jk=
zD-hOSb~+g%PU0#8!p*Y2*3F9Sw6!M3AH~v(9;5hqn}lR7rZ2jw_hoZ#wY<n`3;964
zvZ~6zcBf9K)P~95lpsurAOBpemtFa$7VH}6YsiF=L{b*?i{Y%Sj<`D5&WVbibTs4*
zw8?|;@ayY2!Tc2JG*ba)@B$!|z5%mfln}0086AyCnH0DH0~?=BW8~9klAqtwWJ)FI
z&`MOjNqcol)tObh(M5yskHE4Ayyl=lDd>l5O&?Pvv2}2}IXl>MrxciiaxfjJ=xv=p
zIKWA?eA<JhL-`eWB9i)63Yrh@=fEob=(It-cf3r-N$E9$ch}+kvcSx@?k2|JL&GKb
z*86;+#3f%!PosAhE`EsyOCC0Q9$qEliPD6*OrWh-<ao-!2^0Z1YKwrQv<Du+kCm02
zhp#?t)z(`@l4zYaoZdS+7eCu<u~l3>e*$ywzv)D4hfLa1v@0C(%*|NXgM9}Unk$wr
ze;K{GRBrnHMukHz1V9HlssW0%VKf+#x@tiwWHWQZGW?{0U|Z<jcbOlQ=LBA_K71j5
zmHHqxB&l}nbP<^tiMGH_+0r%q5`NyD`BO{IL?BRfidglHILCt<)&KfQ|MBPjkLz2m
zLm?v5P~t5m$WT64*2t`nm4(L%)2kJ(#{>Lu;l;khA8#PJ>qHAMcgS*@ej6%pbA!Pg
zBZbHqm?=wn?w%Sb%D^bpxmM)Q&Fw1A$w2J()c5KX#cxpkvMgjQ1QfM4%BHs6w-j?=
zv|17oqHr&`wbtwAePd-X>n2ryp7N5KAPT0R5(<~l^GMZ!G1>u#90I&l4nG|THXi<c
zeRPV%E;9j##$v*_>}Nt9Usg!^3E3MnzXh|$p+su0V!tk-Q@h2wl<E+`D0#8X_4a4z
zkG?`A$IpjZ`$zn7`4$9t!MKnLseeg%o-mr-v#^{9^x2kss+F>MVe%-|aEdIyL*Xh5
zNjVvb#R)*0)td#Sl-Q_JWs={^+*gd?ycJ@a%+VJLR#2wKsNQtLoZYF6ksJL<OfL8R
zq8W;nN*+NhmxnfsO5Rypj9QpDn5<fX!uKI&PzJ>0ac5Zo_S-4kGtS4|s?U|aFDQtA
z%OXcDLPJ$g&H#V2;Qh-(2s?J{_ug@WJwjb;s*5;)POoWNB>%wa`{i4eG^2do?6*-e
znIaXHF=r-k1a-V#5zoOIPLEx;VT@3%4%BTVy#FN%46MK9+cEVkHb*u)G&Vo(jtBY(
zo}}`FXnHs~{HFg~;%97;LEoPIsTh=xgO_JLnV=|Q4^JX??IQcWlS&SzHn(v(*CoY!
zl6b)Tc8Qz_H`Go`mHB`YC*%cop}Qq9YwK8MZwt`i+dQ#7?Ct5C7YQ5jbcpE}AWt3v
zC5Ali=Pyhq^*+NULS?^u`X)u^GtOPg>ksZ+W!Nju$Vi;&;YrpF5$S>!c7YCO8Unmk
z&~v&Zaj1s!EiTJ%c(EMV+G^+6_Qi1{D)af63_yGm{Bis!=Y|di6ttkq-hwN_+3t&+
z`#7jWxY)2~e25%s8TNc~sm#ho;mPb>Mwaz2ui_*uYN*pSeN&W;4yybl8J`x4&cE2~
z&_;4xmEWW^t#J&1S(?zW7@(DX=^EMC_9bkM^GjW(fjVVfu@g=PLH)hQ-Q2f<%6IV|
z00F8Z%dg%b^Y+h0_vY;rtS?DLCgAl+s%PcweO)m(Z5Ly~##d&@K7+!hMn8X=`~#$|
z+7id3lndTa<+`S}md@s=t8bb)<PjF!KJsST%%%jsK#Sja;oBdadeds}>X3<oTc;k2
z9dRYSM20raOOY_-_uCi$Tqa6qKhpa`_DMo_6G!N5_^R`!!>60dr3Fld#XBgvw!KAo
zQs>5?7X;E`*0GM#1$m!kGHViq>J?;NR1W1c)FFq%n>Gkvyc!`ah!GZT{8bY1Mo`<A
zPU9#t9H&Q|n=}5Bs9Or=R`BKz4h5DtWF*UyvA>#4>N3}Kc4DxfPGoLO6eHCjV^Q~w
z{+Suyi})$xed7LLt!r~s;8rliIDOR?;$kW}!EG-F`pm6x=cz-gJ~SBp9OH;;+Jw+S
zrBE*s#>R2_NXCTP(l`0$&$4AsqnKKLO@F<uH>&&e3$A{!3nB}vYZeYT4FlppDc@g_
zXnj=eFcsQ2KoLdPH-l<0M%27}>-a2<UczQPZImZaXZa&zF6$Q_Ejq%i3Y<$Zlp1Wa
zoZ|wF^2A~Q2;;1)>j-7xf+XwjVUaE$bX^J79!weN!)|{U><8s8epLC$G&E5hac7~2
zT?hapJ;cKN>QIN<r=dPJ+I>PGm^+JoORjT`8vOO7=r;q|HUJjz`dT{$i7KE`ltw~&
zmbC|{NWMLvm5$M<V8zo}WUJTYBnU;e3ii-S$bp_t5&Qllgka4$NEKr#)|V9hdvCyy
zIx2|r)>^sGL(i0Y2Dt@A4}V7^LuSQ=AnUz2=-sU(wx6=fjL5{?TeT(YZs*lCj8M;=
z&%R&(;5deIPlY<;pQmkMHd^?O?-rI1+{g|s`}f1BUTBug3Uk%#u5GKC*u<xF<T53a
z%dcIB5ppX{_M9%`!>OV}WJl1FAu|H5f!pNo*!LdH3&UxKU&khhjaO9UTF_j5*0`Ho
zEMT95j-6TP2WY)QxAQ)SYfed#8^tn5cv)FlmHXL;Rurci&UfGIVtSDUYw|p9RF*;c
z&1fL_`YVm?-Tk@J(Z&()G>8IMXAI8ZpN2l&!Rw)>g5H#MJA-3hGbTY5&`KEm<}mm$
zme=sUUS()4LxGz*^^TQL_Y0|r)~6NSFXL(yVMD)bkL%Fli>gTD!F7>(RQ8shg*|rh
zvsJ+HI@6D*xzm~SPYaXr1$saX@ii~``tbkYkcwgG<F@mwq0ss1#;Dj0@Y#4_T?>46
zN9<RDCAu8mmF-!Dp+#b|UVSKTo${65YY5I!D+vk@QyRqa#gZUH65s%Y>%1<5xQNnM
zux^Sy!F$;X`o!vr`(9$t_0|B5jV4qGsQvfmYJA0(qIH{E?Z(Ou7_X-H>Y1TOA+K1_
z!2Gnon<f66S7oo8iY~Ba$Rr2F52Ur<ScU8Bh`uqs##er0uMVudBk#t)tPo!8<s(lv
zr#aih+yx^ZxnLxexso_skcV2}CMwF{)Kz{p+2De}Xs7I?EW$SA?p{tY+JocxbXno*
zHNtUe<C$np>KHS+hPSgxYqEAUg6iXC?RPSqU;S4Q)Qu-v4Jr?$Et%V2T`W(syyowh
zzT=~@9{SM2>G_U3JcIJ9LVD_xw2R;nNXT%!kJ$3$RTv_#V8mV;)1oHc_fc=eFZ4bI
zY0Pgk;3W7*7~`+&+k4#%9k%N@D4XpKR*hcXsf)_7*0(s}QbrA0_X6a-h^VSx--cQN
z<sdN7tr4;mJz$!tb1KuMlo!Ka8e93W{;Pok)AmzCz~pTZQE39e2MJIJf*}jZ*sm-%
z?S9c|{otw9n3Z|i3Vki9aHZWQ-dCMpUzn5*5C0;(s@zmxB+<Mu4w+i#nQ7`|JN8@F
zes)J^-+NbJbx`D~f}aPQp;G(8D449n5bbLT<v~=}_SHteR?ga*;CR@axUBTp$kUhd
z?fJM#d9>Z=lJDaRm1j}U!a}lcvH*2y1F{)yj-X2PfuJ-)I0HPv?w=Jhc%B5OD(dll
zi<0f}v3qdxnE8zy4x0VH=USgJy4#*q0btEGBuA_-lT@|ni2sDQH`g#Nm7m8C1o7j5
zJM%k%#6z$!Wl|Kfa6g0*iD=Ezo0Vp|<fiQCGR#M63aiQAq3k6|Py1ZyKyhz36s#mI
zwgNLpm7@(%_`P7#lfBOf`8lqpFWE&Tzdxp=%L^WU;d<0}i|I$^kP2k}6Q%?O$a*MI
z+*j>4`HJFOA~KYG59-80GFLxor8r*5%dzPyX3B%kbQKp47-KGpXB3q_y_iN^nKpcr
zn-u@>Y<NqBBOie5hG0p6&dzX%iUxY`Cg)PmkPKztTl?e1>kC9_|Ag|cT3xgS(adPy
zy;xEb>#w-v`conILZ)C1(Srq4Q2Cn@Up-Gol-ba$>YChIRTh=`1;PxOu_n4u8~<p6
z(B6!%rk4H~T5K<IuNbZC4CXo~DaMIbC-Av>b@k|v>tA`3a5F%bRDRz_7K|fRZOGI*
zdx<tn7+%NSY}LZ~eJA`9FBGiBUjym+UGjS)?1J*ekCU7~IEIa_-Ul-SHHN#c75|hG
zlVx3B-M2{xddDx>N=rXUrA|o3z!T^lvzKDXaDJqE37vtjQf@lS<~l5j0_d(nzW+!Z
zqEsXN?)whT>OmUZ?G!$4jP5Ee=@C|0It#P9SOF#o0wL2LajO)uH~4sxA>&h1-TM_e
z-3Key+FtaMLO4uAw_CrF4&9;*al)>4Q!0`pihF^GgW*E}MqJU_vv(KPVl_xz?o<T^
zI@sB2D=m%4^?bttOeYQ@z#^o<3Wv;bP=jBHF+iDa@;kKcOBPYKL!>)s_s(|Bb_ZU_
zdS}OE`}li)+X8_LVAo@mP^@szWpX3b0y2642waMQ^s5q;_T+i6x`{nyS%I@CucI3C
z_WDUxt_n0Xd$C^*=XbOr0)nBh6FV+=1Pu1KopV%|sFL|59>od>hTaDJh`+x5TOE>}
z$<Z`C>>0MMFr#=EZHam?_Hg=gfyT<!@|p5uwjRl^&=GrI{hqdj#JF>J58NAZZQVx>
z3E{<|s>Gb(Q{gJN{tlx3^;(<J?aHyt3%?3M*cOOI`&`apEViu*>Re<JtEL;nJ$|tn
zMLlg#CwxpU$-kcld9pycJsA~(i<~bnK2#Q)s&2A6r5YEN`c<w!>CW#EJe|KGA)u5v
zor8mNS7wIws@M2D+y0^~xbL?N*WOKFJu)34oSgWP9rDc>hS(!;!dlwz*4p2sb~~?`
z!JETtey&T;nrS0z!#zxAO<EEj3=PJ8`ub@gL!sE$XJpw-z8^=aW&3onn;1%uP`XaU
zxcb~}Y_h3SD+ydMPWDqLE9f}Vo}y@=E=)lb94T#eGcE_XZf6~fZLb)9wh8#{a>*(e
z!+rgz@tXDlKsOzeO||9?yepw>N)YPe`Ui&@3&u`^nG`h<{a3DYB8oy=NkrW#9}j2y
zvZ$oc*<?d(4y<}YBz^p#yWX4xBC@KeUbHIq8wUea39)yTGAGgj*@7^SHn(|n61d=7
zSu;FY5&dp}>$`r37N5xMu7%46LJv*ALC~RFt9QQsLiz4{-Mc&4cFiq9X&bx#-~>Q1
z7509AY88Vs)1+tzH(WTjmE{N^4%=eF|M`jvI7x1pioU?`57G`p$fu+$)7o5blxMak
zEA?o<fU|3g{$I~O@SlI<XZ9bf057Vte7MrZKgW95%|CgBoMus{)bPA^0+=u6SL6f3
zN1boke$u$V7uBIEoJCD{UYH3AMK0j!`4dmwXoh5)0ZW?!Y`=@-8*oC6AUua;A-ACR
zK2>78Epd}A9U$DSq}T2uZ@N!~sF})+>i$Mo7f7rT#WJDPx4N>?H=YB_7re_$fvFl%
z0^D-89b-}tfRg;~)-jI{8~($o>6V&2SIW;G+0=y(v2wKuUAM8r{yAa;@02B=?#R-~
z0SLn)j_||jK$+&qkV8R+;PHsmcZ;8N-Ah0BpRoK3IxZ-%bFUXP#`?hiaW5TfF20;~
zvczkOl*^CKg;cmGLV)B?cNIJ18*J9=cbzFxWJB&6=dQ|?xap}f!uz!(oxuK!mX*Be
zcGuIteyJBMEKu?pDc6bm7Yoa)l+UPljdwHwO7i^-vMmFJw%@UbioR%u^+Dt<^mc6z
z726tAW<mP#X^<y9SCCM&_m>!zkw!_!Y;&FVFD8~Pn;eUy&R9+h=o8dAJgEC^KIg2A
zbi6-JAcJ2jw}+|ub=PtAXef>3$YU=R;&~2ApBrOcl=qg6nDyjUMt+7p<+rV<bNCyE
zhl#^p(kuU_R{@<wJHSTj{x)sGA3OJ;5`7zJe$-&VK=VTYc2Q(`=^0;et9#2Ga}P<?
zpL9%hRC{~~Yq+T#J;&5$Tlr^oDXobo^dCL~>_&jiLI&INkH)e9kO_eDz@7ny)|De@
zaBiDcnzHRLnVTc~m-{Ll#28O;i~{hmBBA_9Y(*a>-v9%#;dueB^3>r=XEX3o_7(^*
z--MK`^B{?`x4={{Lip!Psm)zvW^KM5sgTq2+awx(t`$OEBB*-7jeOzYEjnaa2knY{
z&d?PF=`r6*%j-Qr1MH)CZ&-`0|3@WIF=oMK<Ap^!sE~Z{{Nid9*Ie=1^UN-r0pxJl
zr<1eR(PXl0#VL|+t5bNeUWrCFHL8dZUS$3U<|HhwzK2;HH!i#-;YnLQeB250Dd&c#
z#;Lc93UL~$C{h+6+2`l_ti*-7iUCn$FlVa^IvXFze6H9)YslO7;-40qPz(y_$bmKn
zVOOMGKpU6JrqeDUR}0AhlKNk_M0cD5eXJLjhOWDS<@98H+3^?VA@BHjKm^^lfzTKG
zB?FY^pXaWzZ*B+UZ=Wvf^{kVi7PdCt<a0K5hnpc4bw#STbTp$+AI7m{QxxeRgS74u
zzyfOJ<V$aILI{A-9ts#)Fcqu?yu~?jbY0+rmR$lb5U?tC%h=|=KG{<eY&oVvO9u#D
z#Y_tyH!bbXNO#`N(N}oVgg@ExyjQRX4s{59@Z(OXN&1gus^B>>pJPe=_V2!d_APZy
zQMw+W^fxRTu@*}soBT$)lP@qpX%uE}VO?+wqqE?)9pFQLV+Z<d`k0<6WeE|q1&ZN#
zXl4OJvtWFXW$(MFPzA`nM{^k7NCrJ~H_VrsqQ|$E99#?x*&NQb0y!5}f!RR-Jqa9%
zrJ2Q0ZRXydZA_3R8>HrWE37BkE~BR)&hzelsxyB5Kb~V{?EocOr`JRklopC}pj&R1
zhp|8ylq0#fhGP_oOI-E04)grxszX*&n3tJ;aO%bkq&U6uFzgb4k2WISb4tSUL-lW(
zpe_{{1{A5+GGBk5^nm4FjLx&GiV@Btdx9a90@q^}BZ(dRuN;LOL|BRd#!aX2rwFGv
z!k8FFs+TW$S+-B0Psb!TJ?s}L1QY3T2ca}u-6%fDKmpNWSJoC)(sFaO6DZ0wY-n_Q
znKwC<JU#A_iLeUJdnX(-mk(GVj<Q_jc^b}%0y}DhdPCLeCH;(5hkp>S&Z?x=()V}I
zLVLd|y=KAL1AA6XMkw=RC{X>cNM<hB-0eHkI0Q-dTE3|M!NJo3Sd?9^7i%6O>#H`P
zPg4@7w0LCn+zJh|OncHjm?tPJfGQK8PuhiGOcc@omK)qupHOe^sGgq5x99rdr<W=-
z1z9RpKi?GywZ<4d+^^u7cn$B-R7`vBHu+v8iD*@};93ULx=P2<o5+tZcE!Yj$iaA_
zj*($wdo$8X7S%^(_&XV()*KJ2hgEi5u#MZlgKb?9Nn5hvSh4_?;U0F<H$|Tt#yV1|
zXRRNiHl(U>D30H}2XvJxWH{_u|JoS-hbANkCw6U}u^nl6)Cni4q8?Fx*Ss7G@Z=Pa
zowVQGFo+zZ%>2;;s89Vv91Z3D#|C{S=~h3lnk6=WRLzUWx!_(X6!Yy*1kLT_)qHSK
zio^)KbbpmVU-uQCyEvh$ii&{ROZ5<vUyY_kp|`NrpLEgpq2^r(<;W_%UA<C0+_pqo
zEpqL<$qsM&Qs4C8FyE3?xsx?eW*F0r0KNpyPfLYBB#Q3=-2p}Sy(kT<xKLI$$=#lg
z3t^#4ye_+&JlJz8N+@pJlX+I-RwOvCLuI8&#O}z>;?>G1SLUl2$gW$vq^$5T8S!l6
zzu=~3*_fxO@**e$3ic7IjpWJL!Dg2-X-qGIUKtR~2tVj`NR{q!V*b^7XQ9kgHU<hV
zLN%y$BNk$+fwu3%?^{7Z??BADT*=SUv24SfXYz#<98p;CRER(Tqb4eRE7QL0F!S&w
z?MTabd_KE8UZxT9Gk*^)KK@AUH}Epu5%w;y!^5DQGadXIIo~pFSLLK>$Vz!>OSKS!
zhSh&~c)%jRU*BGH>8>*z^o(j)cPYTdO1)rbLQG)F)XP_w<BEw+JHDdl^nyNV4d7;i
zo_Ee}5+BuWdA|7_Rl1EnHr*S<%FO%6O?Uvdp3k+Pj;DNZT4V|}&wOq_R~L$-T>6iD
z{XeepHz6niF-)%h@J(bDEO%i6ObYvv5g>`}Y4gXXX<Ks~^B@$kER0(ve|aqeT$<0a
zYrVc#wGF|Q>{_Q;W9M9l`R%P2(uZBNQ{7uYL{Hc%03Sn^7)<9wJl67^mf+^s$7Akm
zEpj8n9}B4_P8X+!H>s9k@P)q4M-j4FEpVeo?H(7A?R<Yz{H~uhMQ*RXZI^%&8y{&B
z9Swd?MQgq~`q8&m&&l2`;_SqO)(OXd;E{awX%eK~F9qp8c;9nik^~iX*wdvn6INx}
zl#{Lab&9%D_!%4&`XHL(){IHGD9$<L7IrX%2g#S{Q+o$H4OAi?AzXQc;Y%(<R9w^6
z)T$}|)T26&M=RQ6SeLRmLxV-oasjsIy#=@(gj=1O6lQ)oN+0rNSS{rpRfRsVe;QZ)
zZPqmX&*0{51OVP-VfD-4*uHu2#Np{DgxIp)rawkXtY24|4bW}?WF*h_u)|}>bF`^q
z!iHGZuI7T0|I44w_#OWZ6+T~+yh=sLlp+rw-Q@jf?5;!d7T~_uf0bNXI(!0uCrnY_
z(7;px<L?l9ZB~cUQv*&!tG+hbC@7Eyz#J^BEPl271r|6Z*l)CmwD_&U`Pj?4JZ}pB
z;5Y!InJ`{+jT@}{_vaSEp`OKrR{C}Wj#tGVs)9rbxzG6Qg(^i~|D}y67+y0}nNcP|
zouCIP;>twhps8<3aK}G4I){S?PI{mC+lfy=CxDu4eO5XceC5(yU~Fpcd1-YRr_z&j
z0ZUZ`;8>01-8xa2-NswsiPyNfgPp13{jnm5<A^c7E1Ra4C0e-CnuHFhN9pb2IIe;<
zT@LzGb*d3Y;oPjuhPWy5WAVr(vszK5?$e}yDp<kZ6CCV}|DWD}f58;>mlK}4i1~!z
zZMN6@!i{S(q@o^9$y;2pJi<r8p#<}#0g&U{2!CVLDPr_+tE*sJ3(?`fKt1B^)jV`K
z5k(6b;pd5t@~Sd8-5R<TfP>;X0OlZMVd(h=C0%l)ve>CmcZ+S@^4;89zyF92Bch_X
zA+fcOzMC5qNqlkGV~Y!LayiHrQDVI;KChprmDen88|d<7`rD7SvjZYYLrcu@G%)5A
zavLLArd|}_eHA@_QGI{KKf~R{)tf;4geP$k{yjs>Jk?7clyd<L%7NAM4Q}9}&NY#K
z?Z`uGq;ymkI9Ddun{6s;-!G+ZY-6-_PqEksm;&#(Im^+Xgc5LytCD_Btq$*G^bJ?W
z$yiqiOayQG&GKJz&v$*hDTA5-Ln?s%H2i!0gJ2)Hu2!T-_+{Jn2{(bDk~eonjrUc|
zO|~K*T4ia!rG@bv1#Bof&spC%)$HnTn`jGl-b)HuvGJ<v(a&S}=SSUjd;_o;%d%W5
zV3O2p^iT-P3=i#K*&y*!#Eu-1**Pmgd?i~I3MBMVzL4tSO(;eza4Tskyq_{8N}wrF
z^pZ0Eo`&~$ANL8nv>EYqWN@~Lu&{7rCAqtpJ|L45hU+|wl_pCWW{K1=lEYQOxz$(w
zZ`pDGNqX32JicT`L=SSjDPZzYWw8yfuJTW{-%-7Db<Gdn)V<SvJrR;yJ=neySCU~#
zWV5ZJsutno_uBr>ZYxJ@0XdWdnbs@2j)xrl_+{3!x%pAE9!15g)B((~kJ2p??z?Xu
ztD+pR0|l%_2o%c69l4GPASq9Eo5((otlM0PI>+vAWU+?Xqr?Ys9I^fQSUN}|2V2)Y
z3+QjH@=EE>^d4w^Nfw=c?|w$Sh9P>p+puy%nFhm)fb|;Un*+GlJt<R$K>t)V7T320
zth*VZxOtx?Lve8UaaHhVe*Vo}{vRWDEhRv7z3$s+=#Ds6bc9pNKKkzF=`lY7Yk<dU
z>DsVsZtRXxlc~va<_8Q0Wq57SD&IjFZV!h`i{YZ?XODAC%W|#pn|0O<9_y^+)x_f(
zAAYAvR=J<RZ902-%sXa1<7Ab~rp9!_nQ_Zhr_|_Y&u?ng+OZAF;DP}#!no7(5h9S*
z(uKvorW(Zkiuv&fr=y)5p`-J31LlFY<%v|m;w-#7T$r^NMy0Iu;=W<{Vr$~s5oRpT
z`Q6Re5<1cQXZGS0KO%2m0JIad9E#)W^^~Za;{BNe%b^NWeZ41byW1z4p!oRYn9{^K
z=EF?kB`MsC+y9Wb{GIK_@vfjr43j_E1OC0wfz4pK5uWowUAl{&>%BWpJY&9Vr$9>o
z5$k)NG$^*zk7a-iP2ynefCw3lR=`sGnWCwd6i=Jnvw+ZLv&WaUf6Y4Il~+hLzVSb)
z)h&S&{N2I!=~hyNGV6Xp_dG_wR?WRSD;uO^mns=~INqR9d}#XtX@8~cNek72#TO7^
z?0pe?L$VM0&RhdJ&_R7F%mLSBGmEU)vpU>Zk1>`FfE8%pKEb&Ckb@cML(J@Li^aL&
zFz`Be<XWk9`(}^>sHE;gHxp3v%;HnrE5qo<j42;VNp-FDHs30l-ZlrdH<GCB=Kqe?
zV}M*ZEFCtS)mMO~l08f8HaOfBTQ>g7PCP0qye}E(2&&(_0kFHlUtN<+$_slx_?NJ4
zA_e}n6BmV4Wqj%V1Zl$V0ttEq4?Wx{wqB`>{FKn1?_k|zJZ9T8To%%uyMz#~teZE#
z^%1ENX(EVKc+-gw%H>KmqDkCgcH6Ovy*@xG3`6SNk08{iU4j@K<k~{NySItxbu&_p
zufEk^;avnpuDXmxMnqhDhDqTqLoC_L3TGFDn9g*q%t$4~)k``;J=-vYKHs(C;E7PA
z+{xjA&<QN6E#zYfMjla<RpW$yF!`V^YR{`b5ZM#Qt8L%P@(bJpu1lZBsMMoABKL-_
zVRVYBWp1uaCtPR3K!LYfRd*V^VwHiz<&;{one?UNpLSU-8)^?(+Xul_#Px?yGB9L3
zijc{yOXg{B^Ln6>z3G5wkwjLQ|1$i@1PM8QgyufzZv58t0aCyQjyjU!2k9IWzG*kj
zH(NdTKtGh@Q4^JU?|%4=AHOez0ZN35Mw<Ym4*`!44GLurYrPKv4jwuh6zbfg(HU=B
zK^*xW<S9%hp0lnuyz|$Zty3uk46`eppPv}rrb#ZN2S%nKgwB&g`u^bD3B?JS!E)ok
zW`!oP77J5=bHc{a4HAEIzP9o}SpxO(6I3nwahP0D*o~7(vxD8>>_mGj^RwO_NHq==
z=rj;~`)-i531f!}&FMs21IaB<>NZ}85A8rQ`~6K_kImm46_pT#F`KmwMjR2{A1rzO
zd*f}>PyMF)bI@Cg$68*Uftk+f!LHwxg+qoXwhdz@k!hP7@g7}wP(%CXw)&oo1)c#q
z_+soA%U1ZXbyH4p{Uv*=Ey$67U=c<)91if9SgbLcNcT{@feFSmS7)=6`VUU4Uv&i%
zVoNfYttfDmM-W%3ZMWW(IXkZ`7&JvK%07%S;)UHY9sDnX^xuzwzrPayJDc5qjGq7h
zHUB@Y_5Y>M1OMN73jQzsuK&a5iNOIPaq{4{=Tg>HFUk>Zin6mxI!ds#^6L(hXB|u3
zdnm*<;3h#&9w++*|BD$|g=*uskb=%!<8hE~{en;6{6YVtFooQ}w2V~d^!xYG63Ln&
z)Eft{3yy*CO=z8@{<%HJkO0FuoS*Wwj}bc*JmYEP*OFcFm0SMX&#YFaLaO!O{I$ZU
z6(mFr#fr!s1T#Q&%B(mjmUMmWjGM}{BBE}~$+<W>#?$qPqyd?KT-#9TYNQD>2!X>k
zVb;)H_x%}KtKcQuxWy3a{r$azl31*F7YXJT>MsK-LhbLquy_bN#y_wSs@%|?In!@g
zXRS3uA*tztAF|K=*21LxC3ibz`AZ734>yId%>o!rw_O3HO~^F3-f6(bK`aXA%lE21
zN2?Dns$})v(cEo)D`9C~8Y7lT?i~i$6|xj{>EFZx6Ej`G%jKKOX-H<cp1a@Niotv}
zoX|e*QHzNS#nZj(R{%p+!Y_JRSJ3s>uZ9Bn0x+pr4-Px}h48dU({D#5Z0GqKban=o
zJaQ^sJ}{p2dh&5p;0bQ?*%-R+8t{6kZSHL7Gz*f4=RJxQPYB29uaNsja#1P$OCJru
zv%S(y97_steCUIMZEjS~H+?}@V<*b0oYyxsrYbC&bmM7SZ+G1(!<eFmv|KNrhmr{)
z&avcFn}+l#M`iYqSzE%6*1Ar_&9aMsE&gXR_1sF*2Vq3Cni;xEdW9FY3y?5iIImQ9
zd^c$a57JP3xp|#;*8S8^U@}8EMse}&Wvu?!;O|q!<&Pj?-7N-9)giM5EkNn9#MifQ
z-d}<WE((6S@?bi{waKb|<^fIG?Reb=zUb$+q<hz?>m=?^6X%x*jQT#%{8dHu;Jn6F
zq5F#*ZQ>fYqM-Fwv^<d^yIG4ZEXz}K()dW|Ob<&AhQZ21#+4Zg48Nq#Yat_ZGT)Ah
z8<!stGe5p1K?t=$%wEx<4$&bfK0_cWHQuHTupQl=c7w`4y-EsU=o{4RaPlmmj$|vg
z(^>ttl$`X=0spu5C&rwb?DCG68|&(BGw(SwHPwa}#hEceNWgNV8+Ezez36%6jCzTd
z$H>e-p9|r0m7y^gEm}Eoie(Dbi+dN*p7_hjQ*_Bnfl+Q(x>gxMg`wFNV?((s+M^#}
zv=_vnfmH*dqNX$EWehW4@dLdLebPGnKkx+!sJ@Q`QI6xHKP;qZ!jx<oklI6F6wksj
z&sJ?Dm<DPEj_o21m`ZV%9y-fRVWeTU*B1Gq_u?qwzSaX>I-;So4`3beX1SS}Ddh9s
zKzpuyK54Qu+Ezxe|9<B?Ic@^4mns|qV^XHb#@V)E7K#RO9NYul{YT4d%Nr^!GR8au
zV%N4Zx5|HNsn+g}T`2*~Uy+b?2<62GWZcklW;7zY=S^d^L|<?71k9&)Z$j);eds~9
z*K^<Gy=!M2(_wA27IqXy2Q-dz4a-BOVf7n{(t1f0s)3YUmCr)@^t<e7IF{#c$Fjz`
z+~QV9v0+R9I7XyG^>Z{9c^b;WBn3+Xl28F;kb2zF6&BQBkx%O%$Vf6I6yv=ymdz%)
z!8{oBuUv@#&W-rr{vP({-}tosUp_Z#qX2IU4g%O#V92Pg@Jy9E=Ek+3U;0=IH}kye
zcFvPW-y1V`&A;EO>g(DqtN%|W7?r&WM?yh8@o_XUK*rz9Ww}n04vNgOMyb2JEav*^
zy{i{#1EPk7{+d<?&{hv2Z7j(t;`L#FHxX15IXm!R<`7JCNqqJ^Yd59J&1N>lP(E{~
zOyp*>B5=E#t4j$gf=WWfU4<d{jv$*n;y+<Din=BG&pgAMo!Nc_OXKc0gPYuw;isA_
z?l5yOr6OSw<}bNNcNCV*rp+^|5$&#wAJ1SF(C6s}qf7ac^h&qz)^@iXc(aQ7#bfM1
zmsrr9U-#Dc#e9=~!LNR%!+xk`{07_FAZxd@qp|W(0VKH3lb%6eVW8R5ZH8|n>S1)0
zBOAG<U}b0kk5@n;RTjY5KP;;~Qb0_iCDX+Zo*P0G&<2O&lNfTjtHEGHWR*+#=0*|`
z&^Y2eFMLq|wB2gJj%ninK;jK~18Z0~mPFcPNi!+|90<7QSYWdnXgBrF*y{O~tuKZS
z+925V__9-(k9Tn74aHexMcOGWv^~)}z&)B!ZL|%_^z_R6CPk67X59wWq*?YM@FkhZ
z*G4~Fc%mFn26y74j7*mj6Z0`VB7c73B3LiA(|4<_+5xnS;ctT1O>qNCyg~i|)(o#?
z3v4-JEw@|;KKMx&*m_lTU+Ne1nW8qjX8DLZ4REQ}=||Rnhb@|2*{5BZph(M6RA!k9
zNwhD#yFOOwGK$BIOw)~MVL?$+8pOra<2Y-vn4z1{?i~S0BV=|EkPD|+P7z<=1N1)P
z=#ZHL;v=jQlnhzk7gP$7wVnN;rwwaoJ2KClQ9nv662Bra+bPG;`d__iLQM8>yaiKY
zxgK0;Z&74fwwH7$-gk=wwTMS5ZRHDCbnf7FdkX!s1yY`v0JFi32H9g(BsrE^X+DQz
zSmaJem^A3V?dl*8MkrCXY2u5Wc%NS4uR<!|S#PGZ)69p60L_&a^z(pawuR-Li{HU-
zJ+mKMeIZ=ie0V`~Hw?s$(i3A70yH1)Hz~^G!)P$}TRthu9;7Cs)i>z$z3%$~qr;?k
z+J;-=Bg1PG?H9iv{H>o4=IGvRDmZh^2lxtgqho(?^p_)p_t5P7^KFo$V#Mp^2|q0?
z?KB_vO?#YNnK!CE4MGI#bp`)ePy{CN+Gk{8bDPL+aBtGOf6OaW0`1Z4t<xJ>RTYlt
z0>l$EnAH{BlGV4y*zce0@;c*NR*`10jrF-Z<!W2$=&<mTF40^r;aR-Qn+)#Q-~L#7
zz=lK}8wYUY&#3<32)fR8Z&q;B^TmZI1n8-unh>$euC*Ki)Xgidek)3AZ#<=9<OwDT
z!p9TLD0<&hH22yHAd0c%99tYz0d^eZ*cn^$g+^c$uJc!sQ-|5=Er-@jPTat|tvPF(
z8I$Wt6oqF@_>DM$PD2HY_*gPsw0eI@FEVSJDiWiUL-%mr)?orSVl{$xb7j9St$pla
zcuG%gHxPO<mpcnEblpSq`J6s!Pr}Cf_x+HmUl+(fnZJFIiYsU(2f|OB`vvsLNXw6z
zfizI=rU)uu1yyXE?_z^;AsXIo7)5Qnx~bS$?&oaNe&qSlYeLIEMiNT!p2+*RKJX99
zZ#i!wlzG(I{j_o<;!_QXY8{CA(k`WNy|JW8?0Fo2+rv^jEF$_PN*MUZwZQD(GAdVi
zfNfMUp8*HZA~8G$UxJ^u05n;ft|ek%Po&5!z$cFs&@QOSD>;;Qi3QXN;rFo;%7=`<
z_HQGqE|1v>;7jytX=@?pE^&k|W{*2ge^1?%_)2ov?YyyExow7_&k2^j^18|FxczYj
z_1v}9!`Jmj)l_Ny5iht&N<>)ZYWltIxq@P1#cg$*8r<t{3@tw-X>n3VHv25G_ET^H
z_Df(f_!nJp$oqv9C?QI{2t$Zc+JZz_5<~?VQZ7{y*?o9;-p<VTu3~?8`~<bfJmk9I
z{6DNC{@2!T`2Wr}>i_iF3oIt*?iKK6(oxgL{T%JV++4J4-t&2@N^zIx3tWzoefNnT
z<F$f<w~44MfA@9xZ?_!XARzFhD;P$kmc%?Dudy27IR~|l1ZYI+MfeK#ntwWB?(O+n
zNHtu$_-WGmrV-|T%7|UI<(}CEe8wZSm!#x8?)7?5aILT3ozv6A4J=(ruK!x%OVk~B
z_#TlJjuF@tqq0C5_TJYli!3%5&<j%Fe@wzn@Z(qm$01c$Nf-1!)wTF<<Uq3fnOSA7
z3`x%_m(_z^d1W;fU4`8fvrP#&Mk}4M98*vw4|hdy*@{c@9Xhru0kS)=`cSfk|KdF<
zzrOhZb2*I!HxR{@0Q#eeiM70_rk6a`9hvi!-sqq3bD7TuT06wVOz`)B<|!QDk7DS{
zv1GV~Kt1?i=A;mq1e(v}L2pyR{;VaXEPBGof*{LG!JAzd!2U;5Aiz)udSxo;J=9#`
z;8sCw7i$AjCfoB~Yb<NyD)&19{hvKirpp74I&!?V7zGprak|A94S<*w7zO*SYLI;L
z%3|N=%C9|mCr=z#Lvf#jHgVtMssOYg%|IZiu;0WsqDPLzl=&=&l^PT1W8IqYo2HJR
zrMM@O)V8?p0rxm~F4-Lfiq?csqYNy{>`so0mVgFns;;<0Y9Mc@Vl<ZB7)=e3O&&D?
z;t)0<OG>E#Mw+H$2f!Ck++he}BS5$N;<5l`50G}Xu`$;XFodonz$^vsiUQxH!+s_O
zXagu)CIs^$Q~?$a0Cvg@P$*(<GG3X+bs80?-OwIH>cJiBthPFQ=T&2nlg?st>W8mP
zdo|ukrZqx61(aA0sOSvulrn38s{x`p`=x@7u4|d)pbPo8gNm7YjZdYcMIY@XBk@1i
z%!iWgHF=Vtj3HkV&uymNnpk6A*6Zaum-!6qR~_gKH~B~z5qR}ecIXkd2|p_A`xHt~
zr9zC?MGPFzMX-@rjQ(#R1Nt3MU)wG89_a}zGx!G88p?3!w^CVm5yTCxnQ=kt?Jk>s
z$mJQUGP+`85~WRm|J{hVj|1^gU)Ep_XlZsw=Y&+c-af9HsEUsDvEY6x9n>)TnG?&^
zt>V+kp)+;N(63P!ZrEh2VTu%flvGrR*PzI~c}j}H@4xDzLntRWZQU@8!BQNb;aOQ}
z?zZcox#9(D_MY3n83E<%`}}*s_rKBh{!6$2f6+a-|IW<e|8j)<|LAwe|FcBq=;7e)
z^UCp=`(Ia{b}z^zWW?n;?{WThsi;UM_R!J8$;X*fLQ4J~=imOw#I#)8d>mhKifOq$
z^KsO0wD)vyBvV!<`;U(aSU$GUOIvBB>6)yan#gg8;9-ZFY*$+b9ou+*>PW2qK|Q=<
zM6b1-!{w;_Tv7KJW{AJWNopx+uVC<k>Pw7WG42I<&-I&LyA~1#@Ovf8g5dKfBbuW#
zIvR18J8D{SMw-G#TXvl`9k!oT^13blTud0Vm(VrIJrziLN>O=%8a}nHD(FyzyuD}s
zp!b{EGkReN216Dx8N&O^f3kZRET**-mu1)my9&5Dsx?I1Ad?K)|LkXEBZN8WoUUFn
z5a6+d5w<j+MZC-qohQD>Eu?p+tV_zUt7Dz<7rUl~_qViCtw7s@Dgn>WJ4S}N%enok
z{`x7#EidBE>bvGCM{DZKzJ9WXTeVCIKH0wleaf{ndt&i++kM_~yp@0<ZS8K{<1Ui$
zWS-6Nrr^%W`$sm7j2;fh!m9^os&Qf_`mdLe&Xl!DPYufR6axK@ewS7FHK;->#nsYw
zdq;IQd$qn<utfD~C_klr@RSFX_Sg<MVP{So13I@=ao$$17|3es-$f6H!w&?7g4UGf
zAV_{&<({IPa=%oI+nNsv$coP?#xw4@MNzR*Gs)|Zta-DG`NkyxvIFKUBcJLq+AzT{
zmd!U?$<ng?p6yja%QbS*sfFwAo`OT)c0fy8<}Y&0a~)Pp`6+UZD~7%7eEBWZc1ViN
ztf~}@HOSb1lRY;1o-y({HtV?SMDN9llUYv5;0OK&qCqBUH-15AQS_9YfE2ZTbr{p(
zFLMU@?}H4i?)ScKx}@KBdUhwRx6PF~%B54x@{VxJ4kb*+Z^mb7KU0fviBehL+S08#
zgRqE#BmLUMqN5lq`tuiISG(p5yF&3a)UGh9)Kejc<&M8<7Thitzsrcrkt_8liAcPD
zv%qLkJF~(HG9F*j*f|cq)p*ARkA3K&=0~u4oDu;e`7LtptyP~#@#H%)0EztBW@AY$
z&-Po>9>YYQXH^ao4B$??C^-a^-)(V8Iw4-)5mf8FRl&OHDYj4hRA1&-<|9S8>saIQ
zvJ-dq+&zq*DB68oNuNM5nyPCFN06+rflA>!n%*;kC{`7d8oVL=8ECMeHm}aUUD2gQ
zAaeL@=9Q#|Leyn%wHnE+#&h8sqqlv5%lO}lj6_W7F2&zAgvD&N?i2@dDOw~ZAA$TV
zAxo1>a(C&@;4=!(n!GH#-{+7fwn%lrcu#hIB~;%&xF5?gLxALZgXETu09ktVK3&ml
z>dv#|!HwG@vC_ZHU$dhm&MKQ<r5o6Iu@wnvM-=_B*+LqAQVj9j(JU~6H#{lGzvV@O
zK}xOVD-}p@_qWOC<&jXpA*RnDkD2p1ti^dE%$OmbOY5)HdZ(EU^A_`SlJPL59>3$*
z5b+mKhg%X44fvT$h2z!mve<fsg0?lsw!NxYdptlVFQ9mc`HIcNfq+C;^`0iIQN-f<
z^;EXes@9&x#x3>z`CesWEgeq>;X)^)?w9z^W)g35X+M*CQCx6~H8e*ScWMY0;<ROi
zy2a_+8=%S4X+Au<=k3WHX>H5+H8&m3KzmHoYhc!<%Eis5CLLZ_SQ|UmCST}A-Yh#<
zCW~Omz09cD?cjSjpe8u@IJ<CqA;>1LNFkAYYZfN}v%v0a+k;2X&>h-SRiM43KQ9I3
zv0*#h?%b{39r`%p;*O)%pkszDE&(nE0WSN=-V3550pB=>_#`=~npk^hgC{|1MHTDt
z&}bXabC*LBjXxR<FVgQvJuxehx%(+<{IUjJIy>(A^rgOZy1K+myv){gyu3m|fE(g4
z$7G?Kl2@`i(0CXlE4ZM2AL@KSkot6AwH2eM84w~IRatZ_`H4IHv)JmYM;SBSJ$Is*
z$<M_U1XWM^5=A@iyOIxkHgr`|5kR(lNyf^=HiDmk!;S5~85cyvY1Erj1NIc<eY{%C
zb{4zA*}f6J`r>r1h$YRfPkp^`v(LWNO#*i&e~Rfd+UuWuSIBXyOlU1~iGAVP^Tn6U
zR>Wl^{dMf%nuiQ&PK><q`+Kmg&SJLG8s0m!n(qrSJMp@=PE>0ZPg#p>H3Z;}o7X}y
zv~I2~m8xvn%OrM%iBh{SUlHT5e9hCgb8_Ms3)t3wCvoy(F*6ni{gh9gZp;1K>C;-s
zV@Kt|*P-n8U&(EDv~)$qKD)Cxwl@#+t5^D){FrI*RqAYKS5+tx7+x}9B<B~sT_^WJ
ze<zqqd$hz>#xbwEc!e%=h)XJ90VdNv=&mmYL-pAj&O4n>5-F65fJE(xv_nGVB<+|Z
z3-fy(k%cnP>lHI*Il-`xx%cfz?%}9#KVM?GGu52N7pU8kg7?jx9LZy&O~(8g#JQ>*
z6?B`hSE>v<cRO0d;m%+{>JUCBhLP|lBra5{Cj^8O$#liZR>J?u!|$*O1LsSU_kV<A
zxAn1s@|5}Xl@=3){7-O6ApX}oSkRrlOORM~O7H4HT&;sIOc8rswd-BkJ@_M+a_6`P
z-S6ZenQ)XK#A}QNXLsj0G+SkjT4a8As<jxqHttsF=|6g|e0jL%sT}-$VSZc-?z?c&
zxAN6<T}DOwXd@a`fy3;yvr@Lq9FdbT92`a2Jv^-C4??jiD_a|ZV%P3U%H1xeK_wbZ
z2N^SGXAiJEd;9J-Ou@C&_aSSL=YU@jI^erZV)GfELFK_holE8j<WCy+>4o;)@Wb?^
z!+_*^{)cPBx9?9F-ob<ljW3-BeDrr~tu*-<+NuM4m=%*G@bfzhyO0XrhmAQJQDq~+
zAB;rXo2-1^VoHs@Y*A-&F=#K7JHpm$iIBXLGtzl87Riq>`1UvX7mk~`IOGm2L2GxK
zH|{!w?YlelZ&kkI3ZoOfM5nNSHyA<Ez7hhpe5bWiJ2lQ}+xVzbuLJ6Gx(?vEE2+A9
zU1u&qgeMPpCsRizl7?VJct|ZcNn67>{1-pc@<t%;uV#jh(7(jvEC}i(st_c6B#+iU
ztp139zH!Da5JSK+LXf#)b_Z`>)66>MLD#}Bg??yhj-ST&PS<*TGRul}KTgv$oq@2y
z_i`p{{=@lRK|$lMwnMJQXkvb6@|SyQ&^{I4wW8>`vk@=kPhjE|V@UlvZR62=c&D1G
z4e{uqER?YLBcH#de<p-1&gM1um$jDeov`t^_YW*+(d<+HLn~S0UZtTyHoUACZ%NH)
zRl4`L-ACnvW26Y$-S*4u;2w7m@Jpt|PwYY<v(grGIdPojTQw9h@bMbQhjVV)%E~t$
zASO;<AG^($+=Lhv4m)jj8JALtF2RF*XX{Gu>yNkBYp!3H_^O=Mj=Q#5yT>t1862YW
zB!eFu9aUB?j4Rx%(do=}<FYj{5d6DbP(u1&b3t)&N$LM-E|_Pk?+urs=tP_SWC-Ma
zt`kdNdiC8P&EtOFk`Mg2^tdrzWbEi?ztFV7`6)R~iLm^>L;1)XY7*KWe1tPg22DbU
zHMHAybyYk^TYkhiU+d5&y(EQR;Deghq5Fl^d#}{wod;!7At+R9*Nx8~<m{_Ko`gSb
zug!1T9O4HzZZN;DkT1U2eT=QmgjK=XzR-<a7~Cbz5NP!=BNHSpCcJ#45pR?es>8jD
zX^6j<qJ78`pIh{O{yo{nFNd6FnYO;#>V0>o{KWZ2N@opr@F9CkNO`&yeL=!nammYl
z{oo2RMZ{E2mw)-QVviP0--P<Lpse=hPs>iT>FVRug;d70niimsAm)f8HKixj{w|*m
zNItk&T`;z=Ew3}=xhVYka)1Be=e11R@*Sb$hL<(%ad6)cH@2R=pIe@*KQ~{mVZ5s`
z^dVj<j&@Kd^{cb7SeG!##MY$6LyeX}i3dpjYrG226W2^laAO=~6D&u{ZrVHeV!QwH
zdle@AhsA4APcHdK5~|}T4+D=fh2w0(`kGc`$4vShFt{_WV+ISYvnNp|_W_COg8?s9
zcE;#YHyor$j(gKw>k4D@ABI8WX<WNS<r5~<+P1bM3rARUD^XQmfyMD%V%4`g(75J2
z0ewjs9=ZGlhof`c81FwN^1@GOc<9#~{7iQDr*7*ppoc;|kMOC5E5cK#s6diEy^a{!
zi}x+++{_$gSwG60t3*4$eIU}RHNwx+rmQ0RZgkz@83gljyB|F)l_$0Rg+K>tVWaal
zR&Rjmj!(tsAV%NR!RiHiNaC&gtBoPvj8cSTbn(wv#iK+&AooI;^qq=do-582dUGdm
z6_F5emv*FRWKs`_N0TnZ2y^1o{(uCQ+GpQ2p;03&6{FCZ)ZF>$z{v}`YxLXdo5m(8
zRa`#%-IR;s#9N(n3-f4M62DXevoIYVCvB}%8g>iw4=``y4>6~~tw{{{(blmwC$(a^
z`9cpQ4o1Ead^)Ev!{9?(kZ|HB=v&CVw-nI<b}l)`Q>ndYB?`p_pSU*DwV8k9@v5&Y
z>c4#|s3_cD3d_8zX4FY}vfKMDfiIciHfRJiWv&(Q?WHGMU#P8T^YUfwq>I)2@rb<1
z{!pZP9ePQjLArhG;+7M&*1h+x?TptRU3zO2T*VJDu!C<K<tJ~G7QR!8NlNW5;>tJL
z_pLwmNVzpGLbO{Z5fX8RYpmLNlSZ<#=zm6Qy2$+Oc$&*REFlM(#$;jhHDyloslrsk
zMaXn(HdiuFl+xYc`>EJ32~9Vn2hq(pFU~j$pQLffhmo8(?T4QxA2O$n>ri#6od~YG
zPRd`tY`7LQK4BXUgOb+e42o*e68-c!-?ljxh1zR`V#%gK=>_48cW^F$(0)qexvQ2O
zb%4{}<M#Uv9^3jNhtMJH;SJ-VI3XCXBquO1?vhtryfl<QZ7C(?vd16S+p+*Z_;v4`
zB)iV*Xv6QEx5$%(j#)lm$Vv_W7BJsDePhP+|B(0Q@lb#7-}s9e`xZioktNwuvTvg#
zTi!%=k*t$t>=YTP?6Opp779^evX^b_N{Xx@#!?~M3>uSJ`W=1VpZELo`P|>{egE$J
zuiyRny)e$4*Lj_Du5+%}S*~+k&ntm}F4X-P@BUv7U`PLD06Tjr$V);QU^!j%@(5#B
zwh6xw_B+Njz}?4-UD+Px3AVaNHIE!&S2poN`S^xO997x<asm|=YULGThz<xs2YLmD
zNofAHF+QrIc2q-^UD*)rj}Ea3a`ylQob-x7d3aeHpZI&PeA4jj|FBd3_376Pw8!j}
z=_!bT0fJ1x5A<smGBAolc|(we1tbsf{8%A|L=*%EA&>+T0e=w0Sitzt7#7w7*gwPY
zm%BHA&9N++8XCG;SzDemJ#F+mYnD&$K|%hE5)c#+7#0F5CgI@dB*8KcF+#!+2gsGf
z-6J&UxRurE-5hq)?*9Mxi!}PX2GEewe|`VoQ~$>|4o|O84?qGdh(~*ddIW%Q6a+Cn
z@CXVCgCLf@AYL;vENC~pTZg?Npn@QLemCs%k1)6g?Y;1iaMJHIXRQrE8ek^`aeMf>
zdxCHo2rFLrmw2Cl3IC%GAO*<B*elR01mz*I+ZI+KXm6Ci*Y6wu`1oJk{+E>hvtWRK
zIH39Wg#e$Np?)^z;Qt{&`*lzEle^)=AYA6*Ze#<((jYt$fwKGK&oDhG?1VK49|RpP
z8SZC$9E1T!Fq(LWoU#RBkPl;sZ@9@n!+Do|&)S1{K@i^_c;OGrKcD^N5o!q95|RY*
zbYCyi-S!oSAeaOy%=9b>gM486A!zH}^q>wfZ%?n2Mj)&R!YO`e^W8dtJYjXA5jMN=
zyY=~W+1GHlp1bjV7v0UQKp5~QY|hKycsD&L3q}nJv--0RaPdI@KX%InWx-E+h5n`&
z<O3HE^F3<<;z7A^|FDp=e?AM(^hTXB{WE=)Z-~im{O+?8LH@sa6x5B0F(ll2mv&GV
z(_!}zBV!N-c`{jf1={Ym=WaLxvSV<EydX4m0epHufe;OnfUF^7=mhv51ciV*-Vh4%
z2QgmYt`~^$0^k0bM(Ou+p^zVlM?n$4?}dU-tKTXA@#H_#N<cm!HV|6;oz?>~6fcIt
zK~BEV3W)XnD^Cf?5PS;<aX$Y|KL4o8(x3UD;X?2u@MGXl2Ywo^2iJsaK@xBz+yJf*
zKM7*B;5vWhZ3UkBCq=vUSpuc}BcEVUqSx<qc7XOaKw}u>4nAi=3N#e@R}TNyDw!hT
zhyK13AL54c=+7YhOIxA9lYW1tQ2MtPwftYTInx)WMJC|ujA@+d^Iv&LFikRj20!DV
zZT_YO{rCMpTY0yqP@qTA|Ca7Q`{JLejsr^le?J)jS^))U_WBo^mHsu4o|vY6f0s4_
zf8{d7*%RX(m?`pi>~1S~CH_JB1BzGtJNC%fk<}w3M;;$3I<gGA1*?TM!(PK0V11AT
ztQFP<!@;^?SXkp<>HlXu^jF_QyEWhCtKE9<a@Rj8{MWVy9Ls%zTa;Uh`y@AlTZa4e
zU#TRxg}9G$n{dm5JEDK}+u!^x@voGFc5wuy{j+`k>#}y=71Uojfj(g10vv$)yTpIv
z0b5XG6ch>a3<2%E%Wr|uC6M-C-@HF5f{VkC!cG5!yES)t{-1d>8!?|`J`PDR%QI^+
zA7wV%4gd39W|jr_w7{p)zt#=#$6qP_jZ&{Luc+Pm1TsViT?+X<(}Rg#L*h7?`n)7e
z13eBaO1S&`OZ>hSDiPup>J<{<<#`y|oi}%%gP=|8-}51ZVE^B7VHOaiw+6<(fxqLN
z!Xc<E6oQ0Q|BgEZ#!7*F2zsUN5gropkNm(y{T_lsy^pE<{?5R80*qhNzkY3kaf~G!
zf@oR4eo>zM`b8@Q`G1C>R{vkE5HBnIB?~KzK@wu*Wq|QA{OW|jRKvi;@ca5RV~Bwf
z24`YsVP#|I010aMK#UA97$Y3U!~}<f)K|cL2+qsIci`x8X8tqoERw-{Rjy_|W|cZo
z*(PB9ksy8SLP$ItyP(iMVG)^wvWMj4Rn^orG_|x1Pa2&vHZe7`IcsZY56JWI^z!!c
zMO_RH3y+A5x^(&4^&1I^Ny#bMw{G9b$-R3o?@3`%@zavhXV0suYk-6FSFan|J370%
zd)~b59UK}S86EpHJ~2Bt|K;nqg~g@s#Es3ZAEa&a&z)Vm7$Dd`$O51LpzJ^B;stau
z!r?GD%Pw6EjFG#9^TL@99A)M^eul+8m|s%mD(l`8S&u8**rbkG69g`Vd}J4tR-Kh0
z?vnPKvj2C4#s42s_7`D)(=`bu1IFJMjFAz>1cSktn3=%E%*wobv9h!NzS#e~IDcQ<
zyBE)2*DsI>1IPmohqHj+JsfNtd;YhtUsGTi8T@Mk;)F2(W`glTNQg%L|LMWl|LrO)
z|62e5PMrmog+`Eh_Z5<FS=iacbA6GBN6wthC%lYd0={z(-vijMz^IBCP7MR_Xw_GM
z@}Ldn#6=9-)@lTj`~9I|wEl=9S5Sd-i`k03;fJ-?b2wAXY)>peDtJ;0bP3|`VSY-+
zAq!|aj)Q99K7;6jX4~)Q?|8zNYw`xM-)!bIsC=@dG_v7i7pei-W?li!e^FaSt!I3T
z7+mt*Xb&IWyY2iM^v3e2W5Z!*hz5YB;8nB+YB15mxq+6iuX+((?svqgc{=sE%l;`B
zW0oC?bE0nV_mVziiAvj`m7p31%GE#7{+PwKlc{P?T`TTCyEvMYi>~)SX1PB^tcT-<
zH~h=ieBNhl56>hIiH$r?Tp(2w*1Ol4Tzv<~7_(FpQ|pcIz7?jGuTWkS$*^a=`MC?i
zdckg$TmPrW^9%`eS;Fw;e3Eot?UU~^@mKEl@4Fw$dG+f;gozeR;I&)d7=ZNpdW?i4
zsV?Goj-%a%lzt<-3FJQa%R2}hH<f=E+_`KPQRrfp#VqxJ@r&nZaV)UJZ$l%$+@`AS
zLWGRbZ&|Ud78g&I(`q7fj>&X2^$JS^uqU}i+l1rdb7l6un3w(3JeGDOMW_t7#l8bH
zX8zOO?!Wr3K<FW@(#m7i$f2Zm0&-<}R>_CL(r$^F{kg8+$!Q;n^J}mmTIidzzEoea
zczpH1HX}s)P{65!<R|-`X$era94SAH=8ssEe57&Pf3_;6D?x|Y6=*-7xblJmr}~qX
zD4tQ*<`+`2s;C%^=jP>2rl@e$T!V`$xd5q^@lY#L<3sSI8U^t`073|?t*t656*aZo
z0~jcxk4*PT%HLe+fq2So27kO4a!LR8QH|ru3C#~<ewVDRN-Ee+1o?eKOxDv+Z_ka<
z;tu07vHksINo8s_PZFLaRu8;e*cR|iI)O$|`C)-J8kjTv@q%(iUrY32ZRaz-Y7b4O
zH}Lsd=jx`KVWq<6mk$pId=t$!xvPEThn-8j{+*u-KzB!sc1K@^K8oN5kWt74dy(+w
zwbU1CD+|_-Or(zP6IYbGDQ6ncrzsIU-I7PkjTI&*&fr%lGw2(1#q8;GZXJ0cPM>RE
z)MN?^+<lnVd;R9+I|wsqyWzj;nBBN{BWle;hRL#{*wroamw@MEd_`vO(`JMu_-~Fp
zb9?AUID;RznR8opC(O6E@RPlQOPr?{8>(GDgv))2KgN4?uKn)Z?VQgY@hoDCuq~Aj
z8$(ugH%aYbuj}syno6Eu>=<zjVcoIKBUp|{&gLh@a?$%+n8|K)Z<W4|P_;^A%HKIP
znlfZL0@L?WlZ9tzwy@Fv(BB%1%*ne_lUu)8T-ziS`c?sMBwqGDj%D;Y>hg4I#mUXZ
zdeYAI;JQ)JHMVx$L1ugmJr{pElYRn!4jN>jNPFH9v!#l?j+seDY@DP^0BzL{1U=i>
z08tzEC!mq{#k|ps$L+b=BUF)?KPCm%WK3??9gA~``-3^yu58`^hp5AP*w6Wf`rBSI
z$=3?G?{U6nAz*tDWTjl{{r-5&3JSb(CV?+j^2xHf)O(TbSPBzv7xf+C^Ow?%TLQ8Q
zvFb;}l7I?Sd|cGEj##bH7%Z2spvY%IZN0lk#*ImScb}M^Q#oC=%CeAIDSSZv*`(X+
z;s4Wt;{S`!{damlu%AAlQS2~pw_P!xkN{C~U=;T(g)ZO16tzCz(}B&noS%&J5tCnc
zAf`M#AzfcRbpPz_N!|QrboCPxlYPPgP4<h*RJ9(ez%m_RD<okEN%^~qYQV^*si|gY
zRThT>FLppyagO6YNn;VNzP=tH4al~H>^Dj%97@=9pl<!NE>?zMvS$%DsV_4;dr1H6
z%N&l%eXyJR%s`He6&+nUz@e}`V&ax11{Q@iA<DF*UXHK~P)o-pDv4{%=0wU}^vc+O
zJ0h6=Ruem_OqT6gIe=KXRd6v@mRLSgS5Aa=1{Z1YO)?HROvp`<96W={zJ6x8Ln@WA
zHE0h!&x38zqehZ36iuq?aLXa8af7ew<bLD5N1-vHNEO)E$=2&}gZezykX2u&d^VD&
z9E0W|AM#oCnLB64Gp^CL^kgyObmdz7^yt@usqNU0S4yu2d^@sm@%c;`3)4+Es^Hk1
z__Jstauygm`xXb6pz;itl|}M%%u|<>m4>TsmL>Z~EEI&^e2NQ4`rxjMBgi}izjhpl
zu2$fIYerS+vi{KK)GK`OvH2Ho<-GOpwazdYAZ|4DuNH%$>If|>{Gt9bDUa~oZ-VQ~
zv=t@1{6VmJjc1X@jumfNQHXB;k$G8nXr4p?+Qd&<#OXtL&K0^0Rqx()m80<+AN)D)
zi`GC*{7llGjKU84RWds&Sp10SS<H==V^j?y53utUZj~L3F+|CfB|=T7JGMiOA7!5;
zhw0rj!sy}&#z^7i7Is>3j0Yih=ait+Xl~GuVg7BmC7V*?wypGUHaA}j1LQC#`UIk0
zs0SO5sNGJWb477)v?I7K2K%n7SbE<wvsmpgJ}fOOC9k=K9p~XCCmIoQ6MrBDqHeA#
zsy-B%@O#&oe5^_|I!wh>%0rg@xoH<h9YL_h|1>E9=4#6To@J|Yd3%hMM~*J1+JiO5
zffy~qB-2POO>kpkK#(09^zqhy|DL#tRyo-M00_|?U_)nTq6E@Mki2U79Q4j<erG)v
zvkOwM{hU!1F;At9m8QBVS8l=%4_fOoYKi8<M_JNZ0J2)!Ih+?wFD9CJ%QC}}(8(Pi
z`MPFsT}f)Xl}Fn`x@aK2^N>%tOD2PK%FlHYAMMo`NdgD_>i&XKrn#G64gTODDt)TA
zO|^O1bT#H4@8bvE(k_R7Iu1OBr>#li0(Fm4g6Unzn)(|*5YV}(i4Px?`J60Drj{B$
z1*q7kEODauWnXz&tx^elmv!@UsDaq_iXX)(%7fHIcA={duG~y7Upkhm=WL+@AXgrg
zhDYI{8WrrJPFQAd$k|&P`Kv%fooUZlEE9S3-HdFjg}5?bj7jr`-Iw>AxoqCs?iNkF
z$A#_kzS)MKe?3u9VVKiSE2rqizC$qLykF>9w6HG@q6$0B1JMVgm+#|hm0C7AE}&eG
z*1wP0Yx`j>v?~@)pLk0YBAJ3AB_U%%t|s_7s^P|#$cJ2+R-||1@xYfZlfuve>9qAT
zBbj?+&Zj^-%ofgcj%(p4puO)O@t&!r?NF6NjJos&u@T`D<UKWsqAi?1=GsFM#I&8(
zMG=MOYTT-EtI7o(kI8l4Q2yo0Pj0NdZAsLsX{yW2KWry&>lhJFJ`wj|to*kxsn_&g
zaeW7+m~-K!WDcuL=-snP)zQX&&sVjZr|_gA6Iv@4HS`Pe#(l!i<GUO%7tb;HQ_szr
ztoc{W4c9cg+ST7&Yn7<aEH3JObm5X;=F9@R54Z#yZ)hoh>MLTq)>uAe>`{UmcpDlT
zd<>ql&)s^%WF^Yh7M>ji+0WE>A<6RS?)5Y`-mzKx6~kHPCJ*Ukfh`iwzq#3ge9SoT
zQ_3{McD==`d8VcoyOw6qZm*~(hpB9o1Fk5iv+v=LR7O^69M5r&QgjsSL(*<WG)(%t
zPdQv`t8k;U(g+v>LgY#xPQas{t0Ia{TKe|tyhi!ceBGRv?IooK#5W7W^|}mkJn2+h
z;(a>D0m8A<xpT&ySM+}H@Hk;74i{#37&q0Y=zsGXf?g%>y)ysWVE?V*u_fgdEy|TB
zB&l)ea1-*GqijOcm7)^|8;4HbpM8DpfJIlKMxWJDhL0`2@3<YrMt0a2v!#qB6YSu7
zP90|{RinO)Iq@+7=Yc27+da@_6zMPokQdiZzrMM*{rS?)T(`2)aM;GMg`%$xlk2=H
z!wxXI1jATKi#+`uj*BWahrS^sST{D&T}FyzUYiO$+xb#Aj_JaYNQU<@J})qBCYXze
zn<AKm7LnLDuDk(}T|tYN!|0Nuqi;@Eez%@HHD(UCo||SZe9(Z-4{79-oMFZogLeIl
zc`HPY1|~Fc+>WWt6)@AmyU5{>$s8mtg2mc>|13PYSat;ajUCpnxT)Rk@I&W-!vLo7
z5kLr*qxw-CFl}}iuf|CPe@u9-U%KF$)nhd=9gF*xS^fHFUJs6N2Da{(%`1sN@j!&u
zNxOwXDf7gRxG{&}Q}SdU4;^-LRQw^zH)n&o!pxbK2nE{NIqu11JLM7=BA>BZJNy1v
z7qSZ3^<KyBs!syG!|Bcj3v?r|ow?ygcXU~n9Luo%H%scCHPW0V3)OMx-rIN%+Cf_v
zg1AG4K$3XTsn<IIOLQD;_+dm02}f(DAEJN2*ZCyq3s)5ScoC*@Pc@8kjI*r?4qp1?
zs-pRG^Q+thB&(-|6(7O)t&`;ttM;K+<}Amh+W4dVS~Jb1BkJ9JDlb)>VS_i1K8?w}
zCGk~h><d!PTPpZxBb~jY&~y$tnPEZLvi#cetk`$-`KJ#<x>XXI>uF<Kv$F{HMSUhh
z-L08^4kwv}tj7}_J}5>`#h})!yqE)T*Dk3tP=7&GfZ-LN7dwbx)mJqHc6`G%Tplug
z5%pvCc4r*imb_iyuO!E%5}bp$ze%b%a&LvEmoK(x(SXW+0CEDjwr$CvFq74NV-Qb1
z1mNmDYwIrse)jpcsmAHlUUuXiugxKsW4w*NL>{tIU1G@u)sq}Ow@>nH<8YW(!M#?f
zVbY~CBX~VSEpZSRj`2n$HYm5_$Jb}K<5d;T*2H%hRijQ|-kCp}bFA=^Z8SaBc<cL>
zLofoi9Z9gZoYWu%ke_+Y<WH=dUa_&n9GU{)BW)fSfwGMHRrRU~Nuwh!>MCMxsB^us
z?ldu!NZozxlDgAcdQk4=*g|-u`sS&S><*R0^x<vJBHmTEL<GS_Ow<y?5~E0DyAeCZ
zMK8W6)AQj%&Y=g-^jbaP+@I&4VOv@_sfH~CODca)2a?~jbIf%gIa0jf9$kH?W{`O8
zQPI!y(Rt8MGzpAm@lTv$goy#$ZEkTpCtc-;U1jm3Yovg|amRB<Cj#aZRK(|IEjbIG
z7ta1c;=K0ltuBE7WU3vm`9XYNmm7+5kDKfA50lAt7p)|j+Ez5bJVhJ`&D+G2rE=S8
zI&=0tNYNM+F)qNynh@Tk93cG*LQ+qwCEgEzC-_zco?^iBBIi2;cF8pkDTwpHkmUgg
z4NFfB9*#sklmz}|o`0WTvCF%CrnyYTmHCiT;ObZDwqqK-W!!@csFR>ko#>%pO<03x
zt+C{cT4=?HHtDfd4>z68RTsI{wZW2a$*Wg0zW-;jw5KekvZOfnMvN7$pveF%5l@R^
zZ@NYwZZ3<`?iCJ%1%$T@H<jx<X5LJbx%2P}_aWF9%Lzj|dn(ZoAkyr`&$z8PK6#3b
zIqdW{99?~-VQ%A_`T=?eL!YbJ44q_jzLx`f4ii@gM#}_y<oNWs{-GG4dPmf_7Esq;
z>ij;SSi6hCy{UT1z#`CKagF_8NN<o4aE5&ncjK~1lwl7}WKF5Fp=o=$z}c)d_S-b;
zg@(HamMPtNdib@&It$yb4o@D*#)t74OUU<j$oED`Tx)|{v20|_;8Ul?$v;4s7!xwO
z@Ep$lsHfyxo!Yt|J3^qmgEv3yD>^k{qT_i=m<7&xvzBH?9_n@#296{2oO{Goa2e(w
zRn37>$(om@!(5u@18!420Rx$2aP}xbZC(U8uVI*Y48hI<F^fy|ID*oSOkG*1PTj}r
zj6a#1brQGIT@807!*s_MBdLOgbEisX@$vfH<baaCfv$NmC&x9?BboZak1i+iD?f)$
z{({c#WtfRQ1b8khcI+2agYQ)6XQX$D@42GiJ{D>mOfo?$PwWXg_&R$z{9#p#rX34>
zzRx4&9TB-I3MllW<;-OCAVh0!jOi+KbRXs#d3rC@H{Nj2bSu5;gy~O1of}+n3~yLQ
z!CumBVx1&R10du+dOYo}62zLzZfR=jU~AwyS@gYH=iHkWUJ-|Kv3-WDCmHrt9GA+G
zrX@cIE4I_P^J1Z=8$X8UCk0P`a_c0x99K_#(G=1pe<v(l-9f(BMB>L@hGl$}>}v3g
z^2GA&Jh{4^sGeJn^3S}ubGky@S?ht9*-Kbo!fC`|i^2$236)HwE@096{0*~MfXMLa
zBEYgSeKCh1AKbpVg-isTXr+_E3cR&EqO{Tv-**};mytXvB)BIzaK+2_n{B4P#PQA2
z;MZ5XO|bR2|H!5kwq-~X1{-JtGMKe4{K%zWb+h|#q&`~fg5Nc?kvG)JUEf>{_V5bT
z^-3^^D)e@_IWLKOlyMF4?Rv`SBR1k~93Mc?JvZ;1Tr=s$fvSs0m{89S4kbMc<aG2o
zsQ?i1#Kz+*7`)vUW#@#bN%}HC6y>!*Prgj|!_A}tr!F-}0s2SK^l3j{csAT>+flIa
z6f?YJ9~mkw>7GC5&@_Tr_bDylaT<#~glW5|rl*ZgikJ{1WfeXDIoCaAHh4PgvG7D)
zXi@$99>Pm+gwu`>ooy5Jza-!T*n9+8j9713TSN3{Q*DTN?3*y9Eb>VxkFDUsiPNRR
zZKvgI&)b@a;ynC*UL#*Cnurpmb4dAk&62ww*pPcpKFm2<BQz4Q`t+5q$9m10^Vq>S
zWU-q_)VmHOkFM;CsQGSt;?m`V$!M98slF19{b$l{U41M2q!eB$0hu!5U@-vflSz_>
z1-Kr#B>P89v=m*5ohvsgzTKQ($((kpk(jXvl^I>u@rns5<7Vr*+as`^PjJde7RgVH
zWhLv(*5w~F6iN$6`7};M<>1YYg84Dk3l{t|D|-P7BYh0W>sH1nT8puKM3(fzL8|++
zn))5ml&tN&i8bT%NrPR!i|zYPNU(N=u<fsuhP~Cbr0m*xI{~Y`S5m~D?$ra~{T%JY
zRIo;V>M|jbxs`oI%0DwZ*6PEo;dv;8fp&$Kh4jl{h#hm|L3BM5O0H@4Gxxk^Kj;;-
zG2S;XZyP7>sdD~jNPy_e;|#a2e$mEK_r4=BjH$kB7IGvfiml$DS~7L=`(AGBQM7($
zZ*1cgUse>)bH?o^hG4NSwuk|i!7~Ms;(N&GXL{RLspr&di=E6)CLLBhz&G`&_ha7+
zi5P=x3s&=?u(T`##7Fch2XV7CzuK*R)pD++Q`l`q-N-$DdcW}c;o|-~XjW;))p%1`
z_Q!V#PjQ@E%9F?F>?q2)WvBG*m5&9+k9`G0Sd*%1t9%d0eq&RjKWEo%40158P7}K(
zB}*0Tnd5%(4A^By6B5Snj-&F-zkgGe<#YKi#OM6<>)M+fymHCiPs?siU~xwmAV<&a
zTI<Cq5t<k|5Il}X=uGA%@9w3iAxCp_PJHMY240xScZx7=Cw@Wx8RNOmxvg$>`PV<y
zz3>SMTXk@9?ozdvPm{_NMQ-T;T(~DjX)IQeUcjv%j?Q+2;!O3RJr9VLvLFV2|M0a*
zWyfrsmOT<sm)aKQHC6L)e8Wck!Oqwqt{&U2$1*Rp@{9;G837Ip^91jOmoHU8x^J*y
zpIoCz{6l^0A2psg@?544fMHYw)05bZ-dP610va<tP9INe#Gw0mV&B+lV-muYRFHg8
z+s@ZNJCDCRVD4w`e_KdMMn6P&`1z?<%C*T}kPmpP%*i(ZM;DWOijC&cET}|-kJtgr
z0QQN+yeLoi*Jj4gZiUxg?ak4YdN8(#??!*(m{IPsOn(_Wh~W)Sw^b86>?itq)vKZI
z+=0hKCjwF&zQ$Fxx9h%5Hqg3oc?{stl2&LL)V-eP&~x31Qbw8Y+=Ido!ox-Vb3F1d
zLez^@fQt0kDNNONCx<0FeT2SW<bBz7)d5$><L==yEe0sXin2RO#gjuip4PF}$G^mG
zlJ9p<jHT)`e#h^jGv6Nhk?!1)E0UYAv8TLicr;xpQ~rzc`vV4sdvy4;unWs{_GpSX
zSl`x+U85=igXY2M?-RJhTCKx>6sl#_)lsAes?0@%Q~F<j;Z^56ZsKw)Q-NemXB*PZ
zMO50izm}1?iyD&EPmWZ%N(LT&f4-zW(KV!;y@;z-Pb`nQ`0>dQU7_c2wtqRV(unY&
zmH$<TBHtFDRxNae!j3#GCB~5w7?nfmdI3Bv#)@%IPwPk(-Yu-n=!-aZr2tX=?qCoc
z6D=0C?v{-MUL=2-j_t}P7TR$;*0nShD;8j$dxsJna~XIVxJXW&L8mJ#$`I=AA8yF(
z;mR{QBFMitQRUjLIISP0xIo0D2k@Bzwi54oj8gm*`V%hqMQEo5)n==8I2FY!P=fW3
zZ>psA25%(um&c?6mIrdz+j-<3dLy|6*uRS4PU2=NZ~(W%i~io6$aA`W6F3hOIzbi?
z4lS7qNLV$l%9pp(%aJlbJk$0rb}w71%WKKi{ygqxw4Y#>s32`!>KU$|R>bQj7yB7g
zqc^tlmZA-2I^Hegq^JZ(iJ`5qhmnP@pX=ezy4MfOWV=L;r+}>-Sq9skL3YMeMcgQM
z>tFB_hh=SvW?J^S*E6h4ibc>K4jzfVJ4v7dMvZTbIX)Vz7;}kyiJ}-Kali2eKRmlN
zF<I@L*QkQ;=g_{O)$>;lf*m4oqyS$d<A_7^3vC=3sYublaxd+c$Ew;VoF>YvXg9~U
zt<556bDPORlkc{%Ote>+nWsTi7UEeP6P;zrDyp#O<L#7<OC9e}?y9fG8M}CS^xEKF
zhD;|6bnb^J<MSfYAJhAA)t3ghG&}PmJJ4KaOWk-Kiuy=Vq&scNkwc?VgzgQ{Q;|aY
zFtP$EeUz>pJ3K8VI5u!RII_f2@8Q(4#S6T}y+E@GLv^t|Z{ugUJ_yHp!J#XuEC4TR
zre3?^L5p^6RYQEYfsFaVssYK{r||0dDzkjlzNG~5^E=;KuF=3eJ5zEQ3s6n>P?=~?
zRehV3=N%W+?Q$!-NiW|=6=BB@mw%Nhcz8F!ntKd$>!SJq-V7kaf5f>XcjWSao*G5|
zWXJI%XP)A~^1+8E4bR_%0De<Li4wy<@YUwb)r#u&b==ubx&dBV8b=IUP6&^MWVfD*
zn_(*8`QDqdk#SXA>oxfy!LrqjR~a+E+?gl%u+%aR;&+)W%1-0f3^5k?zQwSC-W4Jp
zj3)ZgTZP=-*l#vdMCjww>a>S+!DT#q3s;n5lLZmcqy5?9JeI$o^JtRb5xaeFE<N>r
zb9VL=$5E-T$d5e#p2b%zl~-ey`&uY^9(U8}#;bhWnn!x<?Wdq5-a_ZpC->+YTzp?b
z_VRvAheE{Vch41=?vX7E&51VMB<uIH%V#w-tVFp4ziYbN+*&a7eELoG%sch7SKl3J
zQ`^y^Fw>{=WLK{geiuiK)qJ8N=cgF=KH3-*R0;_9?>z9jZrc2L;sIB6MXM_6`^Joc
zIT&M^3J+X_?ow;6iStC06}Z4o)?!pd44<?`j=A36tqNX&9=1n&nmL0-b69ySl`O4*
zE58~{n>koP;8So_+Z~o{<ic5_uJT$^<%`;fZ4LF#caEe+W<q7BxWDcn8^%=SccS@Y
zTxbttn1~A9)<%bG=fu<wS4`~dJabR@u2`w&nI1FhC-8!Rrhdyu#sH}W$1i9}x^^4A
z0s;rognW8yC1w$AfOv@D1$<MSzlY8lJJ}*n4kj|MjkgGry`4v*qoZXs25%3E_&Hen
zFNEcuQhnPV7kfeI;43x^3Qd;ZUR4k!FU-hZ^CaY7tKewPZu2{@yU6g<!R0$%bz#5I
zjWhce*tP8eBF(-SSP>O0lj5sf$sBG;E~v}(saso3HxlfeJyV1arct!zY<^(B=k0T^
zOVEjq8a?#HDk+Ka`SUagnt`<0A$n9SD_GN50xW~|>%SnO@>ssuaVM*s3G>o&bYO+z
zEsNorb<Mb2RX}M)5jx`CdH(j#to~E<q^s~i7KzO~dL*FnRqXyDC0%EL8kgDnqV!^f
zQ}cttI1ODpYiCowTc71#-$K~9J$4hL%lZ9YvfRt29v@m0*K=BEb5g4MqdeK}oRfmz
zG5LC{<%L{$2<v+&<{`x#NOg40H8Sus*p*AOTSWa@60VZ#+^pY7!z&gp@q9mF+Tt6j
z_aU|~%^tgGZ8UrIfW^HNo)xuD<Zv7clpTCxm*=dxqN?+x@|4D&(b3*90}nlYOzfS{
zm^OYo&mv2Qz6N<cI`H7A2!+l|(b^CDVz~FH%jbQxuPWX9b`G-G^{>l7u{QMfe1df|
zf|N7LKWpD5rj74F?&0(wEf4Hd3`fdEnp7&@cqDh5ar%rmXThC7^$c*#c@dE$E=9JX
z=tTjXXykQQ%^&$ZC=H>UbQLxazHXJ`OhU&5?h%eGw-7&1dw}z&am<Z%90_YIj}W!w
z!{(?LT`sSES5y`kk=dE8x%Po}Q(;?*)H^amWLYsHD!A1$xeqrfsrAC%a;Tck#GZ;n
z9-;%}@HR(%Ek8WFJK<5r?5CzsVs)qX$orOjw4aAWUENSNM?{Jtnt@9fdm#Vf3+S=A
z)&yg{0+F8-2Y9EI%0KuEn)Qu29UraC)x9{Ru+*UUzR2YL4acr0ssqoEhZj7xBm&<G
zEaC|#EfX&*b>jKx@~+NB4aJWu2mQZ#y))eV#I)i>{*JGm;&%HEzgr!^no-q5cH)_-
z;j^oDGI>L127VlpF=n{xs8qyqZD}}qfaS~6`(V7dHdt#JcRAukG1=-OMkHLcdvK0G
zt9mc<;)8(Mbj8%*-Tkl5#vQ$re@GP8ht1cGCQi~hL}!{M7v}ntw;IcZI_ws=`Hu_U
zDz@M%W7jyEvnPA1J4s7f?hW*{VGN+~z}?BwPssO4*NL*}Km5+V&TrS2^OL0h)UdBg
ziWxyn?@6}NyY`5TU22ay(2!v6cX!mm%lhU`n1TZ5p5rs|ni2yn3p`Z*e$rAd=@sD{
zc6MFV(HRx1O=vKZ>7B^g<K|?#k$U%I^3gk(oRgnh1+7`RY8?Sx%CwSLAU4Hw>B`P!
z{3$lA7>#894#cMIO4A06mHTvVhq~qF6U_F3wil%}=myMA9ovpDFI={}uX8ELZSpp3
z>1C`i7HofJKznYSY(an>EjX~)F;#+MZvnVBUND}AlEMi>omgShy$xeYVZ#lLz8<`%
zJiX6H6wDfnDQsd$0QM*TVHB<dAy7hlg!6La>Y2+Ibanv~(PuZ)y}5CCSYlmu?e%L?
z-KWMPWek}`Sw%BG<6a)=UkM;QfMt-gyZWhI%f0$S^e(V)TtO2NR<y@xAht(^hBI~5
zPsE%l9WO8An|gK0>!w`_qfEG+OLgT$pUwbFn@}#E;Nqx+oR#K%$LVlb=}BtGL8BAL
zG9(f*cLF4Ewc&168q~e(b3eAb(=|vK+Vd^n0OcR%T|@j<>LnJ<HN(<BFQjaInmsBt
z%WyXB*?AL%p)IMde8QRhPxWnWV^-vR_h?BUwx=DTrGffa%P(Xh`C6mLHhU<ZF&KcM
zc_Mb~vPKvBy83p`C>(XfQt1ou4ZfXIowXoe_<Hi|B{PoM7MJs2rMaU;Yr%YZPgex(
zE9ge!QkX@MC5@?B$Rs4`2pCt2vc{(H_5IX)xUPH(<GWwb<Kh@8z*;81pln$p3%wik
z;8wOaM0>F-=$)~g%qJ}qdR8#AlVaN4UAo8Tn`6~;Er%YawC!b|w0J+H+$S}4$H+xB
zjg{#{H2{@nr+R@-dPDX3I8QDlklM}NxToY^*QV%jiHr(A_UOqHFW~AcHJ%E1pY`T~
ziR>HQbM9!0-R@R2P*-f8$*);(va=eM;s3`Mbyv{@LupMtKAVeQTj><ux9>4$Op-+u
zgG+6JGWmH(dv{%JLaW00gBO$<MV2k5@{{LhQ87n}#%-=rrt5dyeI*_-)xJs%PJZiK
z%q8h{$r!G^hx;Q1h>+kLK=^T(YzrhlV3Zq7_S>wdbNCQ$ozNE~DEBCsz_ptqdpfp(
z+#KJgsxo!FX{Ym}nb7!$%j`5T6UJXkqj%xh<1~STHD|EIMDq=>bYW^ssC$>YfRCzj
z<jN0<JUI}#ms~ZmK+&JuF)3<dChym)eOM+fEUu#7_*#D7kxvGnW1Kc9bdFS+5~_5(
zmU}8wSIwCy&o`x~g1_t;L|7CFk+9A5|6N`ki=68xAL?Imo|#!Ljj?>$oOSH_ko)Ul
z4TQmHrAdDBdA&to+eiYt|M-Vg*B0`}r}Tc{MZq<ltC;>Qs^P*#=HaFKy1eW8if3G%
z_aBswk&w^2qsYc|IW{MDTwj7#s;folHwSh?HA%Ht!M86u#6GJO@yRcBPlrm)&6$iz
zy6<_Twea*<#y5a+`GyQ;vUjBGw2Ccs3Z6A!U8#(9@cbj^H+<8M>q<USA8d_7>w7nD
z3w6Z2YK827tQsJ?@hi=9h>5~e!1>^m^9*;oU@=j}q4vNPKS}2XN!{;(ZybwE=#EF2
zMVGSNhLNqx`>75?NFN8?Xo`~&OG1-naaZZNWnH8FcBHz&-nR$#94kQh*m}fU&UeM`
z$Fv;-x>J&_8o+xM8g%56e>jbu>P@zs#ou(}ti|8ZS8Qmi{zx_CGYplv{(`GNCtI=c
z%#<{A)Id}|r0C`r137t4nJg!}8cI1GV}K(Ft(ec`SAUvVlM9!rv5)K)o3x8ge2w&|
zwOgqWh0jNeE5N#@bwFDWQx6U<5bIMtUN;YxjeU6Spt!|lWGj?|7f8|VeXiwg013-f
z-f}B(V{X|`?{y<s^00k>#CyLd=kie^6P|SaR#e8B>Qt+Tw(#KY$99awp4yZf8Sx*Y
zM~%~wGo|^JZr5h-zaw^#<G&rjeB5f}cs5SUvA?N_)MA2yKyFC#Ap!jAg^Bt&{R2yN
z4VLL!I$llo)rxf!HwxVMXG{Ol_wlW$3i@0hH^Cm-gS1-yc@9)M&~HM#7EAzw+GE#N
zzq}j>PbBUiwcC*ICOW>`KiDSvM9@gS@i@B)Jjmo5?z#NG`La6nB<26g@4p8BOQqKT
zMuQ~~Hv<HgFnx940&W2K7)qxjq7pkM6daaWm21Tgt_=>?lr=(mzIRmc?Z^5r(zAYp
zpTT}U*<}f`pUoiB*Af=Brgn7V;v+G0hoT<mhx!|rq}!I-KO1#njNG`J0t#bH2||fY
znAzp)4MR<a?&vG<tOw5==Xo-$&OX?43BKnNbIT=KHO`BNhdzd2lO`B_?OnP|zbDu9
zu~yLf9Xd=1WeR=ikf&C#NPP`rGvJv^?BGcOasw<`_JIb1GkiU<USF0F##ilZtXJN(
zdR^yJV{Q76#}n}<%xlmO_$+hEoFx#0=;RpVCuC$tLrJQxX+)0pAL)K-U#B_~+4c%9
zdS>-kl^(?9Wss!<A3h^*l6r{totMrZobERn#>(E_GX24EcVtig`VkZ>$;R1@pAO!?
zHbW%3Sua)qc=T%?yAdN3^F6SuwzlhGb7%C`GnYzT1q?;zBXh;4Xd?uO7EepWcr^2n
zjstYz>)U6((+?BjS)cu!(#-6%#E+b`IwAUP!<}=|9CEz(;KpYG1EeTsrWm)8Wy$ml
z8sp){EP0G=xf2m{mht+$WDP6wHKNd23!#n)x17MhFQ(EaP2+;ny`2^KV^yWCx>;^O
zrfroGpD(P7qC}C6yT!Q);yA3Y`oJRNxf31c4GlGE9bu~ysJZmQb15O?fh-4c5p;Q~
z22c#;Nd(Jf_qi^QB63czpIl!Q-zgav#jI>r9#Qv!OoQ_(Tu@9Ekp3oPW(rya8;G=)
z8}9SY&VJ8S$40WJ=PhuK1t$X%{GC7FvCU#n?48+?Y!M@aX&Zi)j45?+z32vu@dCR{
zvoyjdKZe7r(n0zToy0Qh&Js#rRo0^4R~#!@nFI#icvRG;)3w%?z4<-*>?7Y~-q#jQ
zBs`4D@-)_wd3MaP|K-e@!lT4KTN~HMb7=A*{OTAn)%Z05(QbLoUYi?RTIk%o`Y@gL
zby)POeN^g`9u7{zny8N}H!B2y-r(Wnc_LrWTy7H1d!28(KYDg?YxzFK{L+J7RZ62)
z;mxzxAFSLo%FMBfShq4@x|=_0VNbe2%Z$~e8or|Cgs+<aLBu3|eY7@xg>T#9k+0pn
zr9Eo6iNFf4gomiN%w9|RiJ`L<QrO73B;T45x*XMsw`U&9=dAP~;>??BjE_UEqo=8i
zwe3>?sZ24oWrZ#{R)T4Gb(X?O4gA=mrM!x!$QcD+deKDut}$wNI-WF2pR$ep7{J@}
z=+@PHk_>k%(RV(+vEnJ(JydEtfb>G{zgI!spH~vM?)grcbmil#+Ue*Tp$m3B6gf$T
zNmvTigIrG7oJUakVWjMVwI;c_$ImWh8kJ7EOeWT+o~cx+37MF7&Iw{T)Rr51^-lQt
z)&NRM2c7*6#V0D-GSPsT*m^5?j>+JW-=~Xq-ltqS)GF<?^xCV^wxD1abVvTUA*qbk
zMAvtS38s~)CO0X}iXMZa{Yu5(eyytakaTc#d~^Ek-n5>Ws)&g5A3j^LjAID)tK7@o
z`IXb@b2Jsfjj(`di{X+tS~n-YPe;4IilBY%^vK(?Bun>asA2_UyXoSD)$&0L%q+*0
zv1MZt;?Xgs;oRUr@dT|gmd9<vZ`8JXI+2Mj0_5TuJ@jmSS^(4ZjX4fe#fl3osEsfC
zdJ(f>nx5|44vgCNo3~?`=%4Wfi~Qsr0BFayR4=YcIQNs$Q7J&WrMkH;<KR(@4+C)e
zsno+KXT^3vTE4FcwsR5twTDzVTtHq0Gs@W^U9DMTk>ck3Bvr0`H8G_WhU<8r31N5}
z?z7s;G&t>zPod)x!Zj#e7$JXl?+!Dzv}C*(H?+09e81GOW%I;fhCDuFE>pcNhm)04
z{}Ff~9CW3wt}(%|Q<=CqKV+<taaig$Y9}hwMg2jbq2gx0c8=p=PdmAdoU?ZR)yHDu
z_T#C7^K=tj2cEJI97YYADy&<6lBJ(snAM=g0ns?H+dmHSZKbM_(}~AsWv_<e6X*gX
zG4_$D3RBgACG}%?)4kr~!#YK*U+raurG)2NM!>la5*uYhvL<`X;`gHJwe`#03p)hf
zKm4MM4u)U7FDGry6z0CzDgrGc{oH=4q0=xV*(jhgYO!hmWu3Zbo^G_B`V0E;1I!SM
zLg71KDV9qj6*!6-)qbExRPKk8`5D5~*8zTh0iqVVo9{eSjQ5@?P+VI%@w`+z8xqF$
zVaZY%?Kq&)$f>HF5=1p4#8%GAsc}3!P@Zc)w5G^8C*8p9A;fyWeeK*Na2^Q`B`H9M
z)rw_P8Fb4*AlprkQl`Hqqk(L90{1k9ehE2KfMn8FiRve96LQmH5Hot1a|23+1aj_~
zmxF5+W%Ea-j+r(@75a9!z4K~0^A@`p7>9g-BP#**i`y7*NI+u40eHbBw*lZE^r!42
zA#m(xn@dkJqk@Co_ztQV`883a2hWWL_<vF^M6c$X_oBr*^ZvLW9-G<Ncf;;feR_K~
zvw^0x{TfePK!qZeZT6kcn@4OjL;myGT6<p<mdOh{T7q+JmSQLth|Sv2xG(o&R6r?R
z>-1aYSE32H_S|W0P?pD1Ea%|73f>RHRa{+n{4Hu>=%M63#A=qC9;VH`Wgpdye4ns6
zJC)mkX8TT8d{s3P_~liq>C0ERs51pt<&u+yci;rJ)`KlHT4tD@Uk9>ky&ca(&aiwn
zSZjgo7~S)E;@s}Rulz}vbe4(0$J+$kIAG$i4>ObWNM~@XUzBP=4vr4Va`UTmig=9T
z-!IJkXtPHjZ_U)sr#-{R%+Ryt=+=*3^`I~SO;K6mbIS=dNyYsB_2}1X{O(U@zKeME
zjc?}c6P5fnVcn-&Y<Iw++vCu&Gk{5%Olb8_hH$g{@@HbK$<LVkRhKS5q4uh0kGj{d
z^~>@xJW?NE?xzY;dK%7Nn@#KxbuPYj<~siSSP*Q}h(BqjE`pR<HiLO}<;4;<1;eH`
z3YyCZ8^?72;EShGI;O+l;%MZFUK2Hi8&Z_s(x2i5DP#)i674F^7d;U=uM#_eud8x#
zL<Q*DTcdW8*seTd{h2bk@;0N&sRHsmUta(9tY|8BOC!?53*)7JKbq0S$>r3*JI)jl
zjr8u(Q#NhZ+^o9128dK7;bgxUDl(r%H<IxAa|!9ut%`J;H?lk=zhz%{ZkBK!|JM1(
zQkrl2S8rf6FaiMG5;2iW;v-Lm65o-GYZ{0`*S}p7K3|iJm5Cok!CLaKco)6Xg~9j=
zG}-spgg}1ffEVMhV%m=DYop2uJ+sE`LIj{tTq{6P>n=ulM0wUJ27Sl~J1N+-U+8Y#
z=PxRCtk6b=a8&<k;uux%RENXI3cl|j#=4wbA7F)h@2M_31b7RtCwm0v2WnzVF*Ega
z0bDZyIAj$y686)RvL5!AaetujeNTttrJdAM*o+p=*lr{umL>Kxo;x5M%dxCYx?lAD
zvhBA*?8H)i$db8QC!5|Ip%%yV9AU9S6bENjwt+6s;*0t1HUumEBa-dM`t=r7&K&JN
z!*{RJU`pkEp1K;m3kcOfhVGya`B*zuhM$QPq1ev+mjL4%KmMW($y<wI=z6wXFPQH=
zQf%RT_7r`f)?@G2`J1cO>LE`oo<Bb_Q_6011*lVUH_>ij7Vq&;_g@4r?|rw-mc0aU
zkjp}6D+XE_lWo%?_q+AkaJSzokFOpa4omIhN)BrOoLqiz0@6wGfe=Se<ZZd_r%&hC
z_pb`gzHJeuN|TMFXP<@NQP0)b2}s`>lds~c;QMA(=?uGdPxRaIH{nvM3?#Re<(YM|
zOm@uWN+7%Qq4;KW;<%?=m7W)OHy?Z(V14r8$BH69M=OaKbw`snYg2)*NP=_!PZ=Z^
zW<?E*`b8tF2!aI=cp%Qua#zswgE7{=MdWK(=ljR~ILZc=a*ReMl#hmBRnt;Dy<`j+
z&gS2(2l6MT>trt?Pz3C-Oe7(=sNfL4;@dm>*Evx&_eE2eB-6`D3E}m|@<jwT$U9E|
z6D<MrWk4Tn8b37y5l4aj$em5&m4JffM?$n5eW3K$IT8Nh7EnDDDAG{R4Qg)g<ddhq
z{XN=$J_5J@f?P!rJ4hYC-9w1+lcmc*&W}jYu1^Y65YxPWNpf&4$00y}NT6w5{H@CB
zK|+OzP9kt*vZxQ7{&5@rhq}ZS+)NU3!$u#B)}JJS2nQU(6|9Rne?CW+?NranQXP6<
za;_-&Nls+MjrixKUx1QmIJP5$EY;tJPsMSz9H5?TCf}O5651ZI-Ei(f#Ky%5uaWrE
z%Q}J%6LYGZ24AjbGeNQBu}x*-5Zy-SDJW%ZR+)C-C3fiv?vZ604$KDM^9gnjk#nc+
z&FIChx0}wnaeSEQ)IK+y7nCslq02jSO3`~yR@IS+sT=ag+nD3@bm$zIS!W6!icuog
zoXM#kR09C!cEar~k2zI(@=I$jE@LjNf4W6NkSV~!u<V>Vc1E*SFJf+JWuoTUi;GG4
z3Bj@p51N9fo}c<~*0k34)P4a56YNlb1|A+GML9{89c$qu52!dSALThbcZ9R9{%9uQ
zv+;wNkFQvSJ2;>HbQs&5oV8>Jvx5O<CeQWgNO_ak?6W41nfpIaXVV@H1{j+(wSLOw
zRC<#)7*GIv^)!fKbFW+h&lr{*zObV33p)7*Gc#`wpl~OMZsTKZz$d#j`G-d_y`S_E
zksK@5(By1uo_^UPd!i^=r|_J~XjOp$*}p0)z9?H^-tvF*l~5zCQ~5C@mSKz+MHGmb
z^7rP+=}Rn+ej9LVUsl!j$_kH2?Q?nYUP|(wu&EVfEBj56;U@(!bO|9A`<NO-Q0^?{
zEvy>ZlBrIy(iFNJFkGeCUuvF|BtY8}qfVSoBqCeI#mULDFK)aZ(lzp6+TzAd-Gc>4
zRm;gZh#!ZG8%5Qx22+ki*^{tjqf3MZxBC?2nX6sXw@R;1lOjut6nIApgUn3s3EN~P
zup7_Gc1$7Z>{%3cy|Ip85HneOx!YC9G|xOrqAJ&Q=nsdSZ2U@D{T{{ZYo7X!vWiwm
zf$j_%1H^VoPR78ei_aW7xz?|^R>jwTRPg$b=t;8E#Q5HD>35O~Vvsa}4cm()A1I+l
z#drhf1-pXd0rYY=U5Tnp!y)~UP*mYu{;IBXt1B~sIr+Jt-GogW>1c4IkDp@Y6NfdF
zc{P;ad~HJtp)8$w<prHJ04=!)&YISeeoE(`Fo^aN+{&l1TCse<)hTy@`KJ!M`UL5v
z%yF^Agz({3`MZLNiM8Osq47$g+<KG5?Xm<pcG&hA11sT&_qTjn5d>rOxOs;lX?0#W
z94mb8d4}fa=_@y<CR^UD3d-?XKE9JgCt`0R1Mr`~Iw*mBoO0Tn;!J<Pb8Z^v7hY;y
zR`tn{^rDgLWIq1IM1is@--AjoJ4i=i26{iWgdm*lpO|;5wd6#(RhMl~D^5r$af;kB
zTJU&Qz2_d(uXFk{&)mAfetW7w=uFAVHU&u)CJ0@l^OFrv6GmUWShoILlWE*S@^d!#
zG3ImWb&WLG?CH=FZd-OMMH2pC=A!V&C`!>|0u#SR2PqKyNeQ%cOd#TBEHt*O{OqR&
zG_eBSKM?V`9`??=vGHcRV|-N?Q{}B*ho<**r=Qup&0h~(?O%?h^6M*$`2K>bEEA~w
z^S>aDi!J=4O$y-{51AIMmtI;`k4(o@mnteK>Am>d_CP58O^4P#2$78^SRDp#X7o8^
zuaYvrtGQ%0w)NI64neTRj3>_LR1#lw;A`C4Xv~yO!8(#)rVGaV#F%}A{V2bww9P(V
z$3FkEyXVUUvg&69`o|Ci%k>FAQW^Q{oF(rfrp7q_+dvIX$1=f>%a9_nZgKO{=(+5|
zw?~iGqzOAn2DI8qcxXaY;BF#5<`>j!5#s?+^{F<$pkgk!L7!x*R#8+3aJhcH={VFd
z_@mKurD9f3I{w?S{VB1}JMDoN;$sAxdD-i@na9%##QGE}3wE(Z?DhT51SPX=^M<;r
zOiP`8DW6S6x?|dU-CNv;)M#7-ESmvf@4QU(>R-vCTmYWKAGRErPPI5Z$3ZU2vAiR4
zh0`>gslUq9XF+q#P<gNj)a}xucK)gxB{FJ~R6w+7)X$$?s!($nwY=xifKx`77+{uf
zG`zNX#A*JcVjgHrQ{-bTN9jiw5hD8RWP1l9=d{dF(aVvclU1*p^5j-+-Z~x>T<jB!
zM!{VOCr^sMLKa{Mr~Ah#)?_w{%2Fly#p(NIMEleVuaO@*(RcF{k~HQNqNI%(hKmlB
zor#lQn@Rf?r{6%v&qm|-#|RO(<F`iekytXuc4*ek@o>3xvhkdQ{pJBcWeZt@qDu`Y
z7tdi6aN<o!@%l>#ABug;%yyj`*M1vt!1?Zg(|c}`XO_c@K#5Z(tz2#+G>krQxU5X&
zU!HgCK&A}>6QasNOWkZ&{bz^sOYR8mgs)t`=Q?_DE=oqRv$G)U)zj{b0+vM|s`e%c
zECio;A_>7Sz}Zqj%rdcr1e!YDEVf@{G-Gx>%@vW@F}AZ}ZRsa^QT)JbrM;*3G}7|s
z+?34s%UyY#<j-F-zaXYeXHTaDpzL7n4FCvz_R8Gd+b3lLj@i^lWHAp8PH<*-ZnOzA
zhd{7>U-h|TJAoSN`f3TWK{*3iCssavZMH5f+c~<EA~V%@WPg`OfH%IWuKvrKkFB(z
z!s}13AbMP^z~VGBSZsE<ae}vn(|to%3SHvNC8KfrQ<ZPW@%L>i`I*-!9gSmo>(4cx
zq)bc4!yV?+!N!ik)}R&QhF}HHfO0YZ->p(HWJ!(Gt!3<WJg+WeHBbFg!)NXLC!_Q&
zepH<((~I@9<|z^WuCNtN>K{<U@1^(R_^CRx>DmkSBjH?p-xU1JHC^ZCF74x#4n8G3
zm2tixj~<Qj;TazyEs}X=6TrbA582Ibh-y@HQA5PFaJJmo;#K_j?b90p-yDq|v}+wM
z$Ts$L-@_+$N<_uvLL6cw9HZ<ef<5qCvp9ID2G4HRaN}WaimJIJF}~IKNf8%j2%9s|
z-*1^)p@nrMf+G_&i9jwdvLH;(*0(F<bCR!FSysP~bT7Fmq(Ky<yk~s#_`K$u&)>f?
zJE&ytyR!QV;c0Pd>v2MnZlW#fa7>a&nKD;#X67GZZ@j<x<k%`=Lp4sE%}mXDo%+&>
zTba%QwzAUZU^`Ds16$mp?6E1_OJ#8W)sCSsc7gM+=9mmHb;gnXh@x{o?L4rhd@icO
zb8^MT(78)b6HO~0?KnJ9uR6SK!rNP6@_UotPfL!`!;)=CR%EZ){=G}NyPNzX<G(if
zpYCMKF*ayaL%uKqX`61NjU%ZUfua@MhJZ0!E(WZhG^hg4Vx^Vier(PvlTAJHDd$m<
zZsXG?)$E=W``j-pQ{Ahd`o;WqL<@Kpw&@)A>Gd;MN{bZza`=^Cig}bZr5i~wQyr@L
z)R!w5%Rv||jAQG(e0hB8bktUghT04`b#X*q#|!8mGAATaRSAK$b043kM*~0QBH`}=
zjz4w%<)v8k`23aoB=sp^R&g@{XRps56`Z)TIfG;+WB>5|6y<DQiK}c(DwLye23v4`
z)P(c)HoScE1?<vtf%^HnILaQ+tQ$-io;on1njT59+})))FZ*7wgEb^8ecB%;?ISEq
zD+eCH<}fQh6#LkAyzm0KpCWZUdd`xg5qU@KwncQPK+#RHGUb&q6-(9$WK<k&B*Y_=
zQAYt-ivTa`bMNJvWM}a4FMQ@M4Y!Y}@7NWG5gXH=6P;R4G(0lC>*?~tTICR|7dKNW
zE}z(guSML@7x{nKd-Hgx`*(kQM99))-;EHmgd`Oi5|V1#t(l5K(p198n6V`L9-=IT
zl%=sNWU{wdvy*KIVVFT<n8oka{W<6JyU*vo&-s1M{W!n#_@49akE!=#9%kmfysqnY
zUC--zJuft<SeKRa^O{$bC2_y??9r79iMzZB@q4aS$>6gh!AQ8QnC+|vn1F%do92{>
zgBoUsu+t)^;HnG?GlVdn4ini^tKLLEgYD;=v46NQchUCJ*XP7#Kcgb_lBUGA^8$Kp
z(l#SCyDtaqubdMjrDV?6sxT8-Drnz^L!RyUyg5mkp8b7Ex8F52F7GABHHkZW9S#h1
zVPC<qxB)7o2wJ0<AsU3$$5sol#*TGmSrA+nom^vu-e;5opLLj_LyIGSo(PFq1(Bbu
zaBt0SPoDz2&m6YpkpK&j91h+SaXZw;cs!Q74f_gnq~h&6GBoVw0?}ttxm!~q`an-<
zb=e+~3ua(K2IL?Jv@(`xz#%gp!NcKHo^DXw#UwJt`^R$`0%PNAO&t#n^=UCa6y-X}
znFr&w-8UU86E1e&DiCzG$W|{dZ*&p&H<oi2q4yiXTKgG7T&%PFlQA0<P$*6|L4p!d
z9k9o#Fbo`w83IVpYP>B421ex7AtTednQt+U#4$-U6xq2?E&uSHtAw6?)%kW7l5QVT
zwGu2Po11zBGxg-0?jQl@r|z<$B*ToBLJoSr;oN}ZnX6T+DeHG|efH(1b8V7JR6)eM
zFC2IC<2jEtgQ-`~K;EArKFIi4So5Xn5jnG@PP$ISaDTm#ApcaS<(hP9N#K3K9RbJ?
zI2%?Rg8+Y)!|JS6py)LbL=m~OL;Ebf(e8Dg<qy4gqpn>MN{W^{0!t~aPjX+hfY-uG
zyc?<qaaS`q7@D-;oJ-%&IiT;J)H_pB9ZY)t+%iD%u0o6IL$jd~2^<fY%L^PjVi?Za
zmPRSPK!FC-?7TCH;Jv}StJHpY=%mk;mjjAVvRrH1lFQ#XWMO$K#tj;kZ*$Fy1ro*+
zaK*_FYASnb4!=a_xsXlfsvdof`<nBtB9h4yM?yvdb^r<rhSHjoQnJvn35SSW8qShD
z5*_7!$2Hd|_GqQ1+SlUS{<mUV)UTzKv_z(>Em{MD#j~+WL<=vHc9g}=;7B&_U4n$p
zc{F)=@)zHU(XzNB^RDcu?D^K#AnAu-8lWqJraVJJko(r^S%=@1_F`4gmCaD^m-C3{
z1D9^#uGCrxP}|gxEle9Lc_*5xzs`@X4&_>vpo6IC;}Zt3wIX%GhNKl}SvnfCE|1%R
zTet$;sf7am4ImHYP?VdH0}0J{fpMfJxM7vZUp-8+nDLIU6#IFlAC43T`qbWV%2#T8
zAJl}&0-hcB(+f{z)Wf8|X9yYqFMM`+>5|2>kc-9X%@&TL#jjT_x_I{(OR!52{o$8h
zf)XRkp+?gK_Fj{wF7mA-Lh;i6FDp$OPf)Q$ZD`oe$HUr(w`7!q-<N4;!_SEll;Pvq
z@FJyrx`l}|#chJ83DzhhIJU52tFNw>H@tI1GR!(Jn<6+Y@qsVbO>5{4t)&hW$_T^r
z<F#R84e+pxZLe!i7JmQaUby6`afqchc>MN@>k(6L-;W>Kx<{*NAik0PJ5UV>iCp~*
z4KU;(3%oIYH7~nUY<~OQ>7nL6^we#A$eKVTeX_6i88Zsw=uBQ>Z5x|_hDt`d>*hW*
z(0=*173|b6^e-zo99O2dK&N|Ry9`A#sOhwZi74B7o0{5+?tK}21Fn^VBg5Yp6Fx0R
zs62=twriZr<IDW0VDYwNM3Y`hmt`Cvpaw$fGI+7_-}O?Du2qTjrM(;&Jwok2v#Yx`
zIz~88LOg4V(_tLkPA=>gh88UY?1h%hXBcxLDgvvXf#@o8Nm?Wi2hO|kEuJ0yRIjWd
zZS(-H;~cC${52luXaFKFjwj5>nqA*&b0$b8P3O!5VHbK{Ohtdw;4gEYJs)#}k@i<i
z#Je{Rcm)%}t&C_<0y&JqMccL~CV!Mopl2q0dQR%<r0nH*qQ0+7sKKk06wZVEMlgO{
zooNl6+)@XtxE?i*3%<W6V}hyK;B`p1yy->S7GuG$HTb(t^=wrp(#1*FKg;e#<L9>%
z5>P|uSlia==NP*0$b=Zw%^%2cSq)0i$lRgs*$Oe+i~&^qQNL14LH&cV7HQwkSgPes
z?U`hrL&K5DO%8%x7{s02uipjd`Yx`auK5VndED%7etjkN;Fc;234I&uE!=WojiNx6
z7;q1cGHyY^;5h|QY8aTOLRmZ_bTf1a^&0c3E>se&cNHCDQxjy??Hp%PShi-N7-iVq
zu<wrPPNTbeBz2xCSR)~9Y#ZEdX`qx^4LljZ0R;)Y3#cD5$7fCvD!RY*EbLL&m5@Zp
zk004)e!9-aI_uWS&xv*L7x6S@P`MITpbGVr!6WZ00DMLL{-gdw7_0qalK1%UIC<OY
zC3;jD^K~8XOkukzXzlEJmW>Qu0A8JgRL*?O*yT+(V)#<L3U?*peGM*l*|rw?-KUaD
zY!>&tK8t$vMn!nHV_8~50J6}K6Zl#eY!d;Re%RQ6U9{Z8eMO@D$*ZN<;6?H=0PZRD
zYg<jh-ot^DsK7IXmRru!%!#0L*L!H~A+g$r_|VtaKel6!-V;muy(F&i6Hflpx2ASj
zvajx`rr=pT%5+cHbC`XgHDjEiOuk<7f*)a3V;9Oh^f58cu>W>(@{wG=q|OwmT+!Kj
ze?ySZ1!CszZ!ZLkS~KGUx$$;G7IKvK+)bR+Y0fijA@jMu__@Ex7DpF$$ZfTa6#TR|
z)r*<T(qVl75!C4zexp$+1sYxh%N~8gMd446^18Qvx%cJ?`SCVM3qi%br}*B+J{YKr
z;pks&`h(=mR^c=M<KBa9cK<2+;rDV;|E0(H*0Cr68P)E`enD0!`xjzNGL-_zLF3G?
zhln>$-YzjL$Bv%=D19e#?*nw>P44=aaS`#9LTt*P-&)us9KcLMVP%9m{5B95%m9+P
zgG<$^l*7I0Z#15V6XkAUvRwCdwMmB+?LGD+d|W4QMxAYpnT7)tno?X0OL`VhIoDK-
zMl>h&6*c7<<M`rCHNME%C<R(~);q=pT(f^@Pg3j5=$0`-6!jK3ck9}@;MEu7o~Le$
za|%B;a3!%T;NsZ>a*>iWv1YJ@EC~aI=Ap#q*BIdqk>2j51iC7f*e9K_Xt(5|09$yj
zLd}Fmf&v?od=?i}t4z}`WQfps^D4u2QKF6Ncu(#t@oJWqdUCDdU!JEpC2}Qg*?qd8
z#cN!c3DA0u*UWUxNwON%GvOSi>51#cCL|h<D|hI-uv_=t<1oka=nD#}Q=GA{z8wuX
z#>b6!rkT(otdHP><&O<cj*qb$-zl`M`pCbSqh9r(sOm0$$D8ZXdR(!y;qTY*B`^O=
zpY#7kGI8viJ^#Hn^ncO4{qJ0p?3=xR_8gayF|}%ajBR9PB0SnaSF5rY04nbs&y*y_
zIax2dtA9&XQW?*O%x&kETQxY%Rd@#3URehp?x4;wa~Ox6$*A$(!osyi3FWs{y%<Xm
zS9g)u8M%yzEqNSTJNd+u(IbyzUp|QJ_XTXFN`?q8Rnby2lpyXIr9ey$IMLPOM+KGk
z%X>@wOHRrZ<}~t6Eku1l2VE9X%q!wrx+XG3U^mo3n^65}Hkk}-vO(m%w!1Q{^ACLb
zR(9^*Cm5{A6;*dGdi&nJ+`8)#bW0Fo8M8FpaFV&qB*N7WsM0$k-R@yoO>cITv-Rtp
za6&Y9#*REAvq9&lk2th={iXe}pUxlwOA;iK<0Uxmd#<(0W9WQKN@01b)RQ*(c3%Bg
z9F}V#gJ0T5p0LMjWNSpH%`c$_+^FrHgqyIC{?w;^XJ*qi$tx|}3v<oBO)_Vd^dIYC
zMo&~*Lk69$sai<c_#2KtC&<8EaKj4JN3<Zi1rqxaS$r$Qm9z4hf{dMQP4sR|a_H{I
zxf!+M6CwArN}Cs#WnN}Qu`W<VLE9dQu7ke1L|h81aeYo(GwSLm*(RpdXxSHN?f(2x
z9p_m3*o@G88JBpsg=ZTCRwJ^W;7)g@tu%sUOAR-uf1=y{UI)*tt2cE%h(2GvZgo=6
zbw!Jp&!v^18rb;SlC;aMx0VwOp;ciP5}Zcth=_V__f=RZ^7^~(_)#xGJ6pHdqml+?
z;dRg5DIfH9*MokA?Kn=1UJh&UXBfiMyN?93{&H@zXob~<o`+I>K$`EO8LagfOhF!o
zV3(GBSBzkNx*{vM2N={ODrl_q2AWPS=yjKBG?Xl^s_ws45Xg=EoMa+c|FZa%SIlSA
z(^(#|hthG%3^lqLc%bHlTbCTqg985b#|r=s*-}4INUW){K3>hYBi_D^zcId5BnUo8
zJnz6Z75E-jD{M91oJGF~$Tmf55v$fXw|O-r49#Wp*qHgsQF>>VLDCL;V`=ub3tJnl
z1cKxPMUuBB|7e-yYV7MKT*paauF~9Ind%RrEzkDv^X<8iq~x;T^tR%$T42Fjo`EZY
z2ju+^57kUgil$BSBPU={s~I8jAGyNu_YNQDmp_&xx~%kUS3$(hkGoYBuIxT59V@&N
z7tX?A*y;DNqwunjz}1ciV_V#xSkXdDzK+fxY8CZ5&8==(;MMu;u||Q(D<A1wH}jrx
z=wc%<!jueB;rP;^&Lk`f?)ovlyLnWwyF4O_Kf9t-2U_&K<95OacOCbAK=h-2rMOZQ
z3@1I!45$&IpQMRR?5WJb_FAo_#)Mz*n@;^=Nix+v$eHe5*eiTI{<sLH_t8STvEpPp
zb`sx?q$wdt{FHOl1R9J8lVsTU5PLn{;f?3MlRt`-?SJMzSz4ZG<)oT-YTbS+awWRH
zMvR&a0?{CbZ68Y)LD#@GrOTf5ZK784yZYo?NVXT-AHm_I1q^qd_m3*+IH<yXSqyRz
z!hH0cz=N>?J(hJ&tR1u67(^}QV)CS&+de(LRMV%T-e<X!m!PFNY1R9=6zz`L$>{fw
zjurKj(w<=l;848o!Gf$V1LVX^w2a-UsLw6*&J*5Rx7KS?yo%ULK72YKzrPgXS7(Y9
zV0{6~f5<2#1b!1lPW+kZA050d)Qep%8+!*E9Znu!J5ip*zak$txV5o$;o0)1{hDW@
z$EeP|@bUb$3~2*3KT5MwISq$Dsa_Qia?$JhZs<v$E!-sTx0PP13NcvD>8%ZyJGn?2
z!d4&Qm53&=WOgtUYfu!Ei8i7uOTNqO3Dsum;lW<Ru9w4hA0L0?i+VhD<-_DbX`tIQ
z1lB^W5Vjvy3D|vc6HMjzgwlbMPSiaqfTuo4ZmOh`&H0k!KeQG%8>C!~I_hh5Y{K;v
zZ(6=>x(YKF1Xj4B<j8#%xEo%@v_vx9`mx;;y2|rQ*-j<Kcck|hRI(Z0PS%c{<36j^
znhASXW~Dnm1}gbPt_Q0)Y4j-)YSrDVPiokrJ3F@hS?A=-0|r4K4syj^0)m@XO~}fb
zXqFyEjhR-C2#0O)Jm2k!<Fg<BI;J+%;HA|4^zjQeC38q~d%UQ4y_(fIx-jGZ8zulb
zoS{6J<Gnc+p8Zh^<v5KC5JDMh&&2_P>UySnojK$HWcwqwAJ;^FE9lBFUxBt^O$={k
zpTSmdk=)1{^Op^dFKXfxS?BMtHl|UkMrZs@1(vt-g}arITUubk7;$FA>R1fCtarV~
zlkO*f_}uAZm&1oFlxuJIC0PjHcv*s_s#qvI{H)!4n~fSQLemD>IdO&|B1P@%Uc~NX
zND{iR;xt*CZx7q1k0mNENB3=0uUPZm`v|V~;bmMhpYXNgc&W*yAv+e&IXWLL=4cPW
z#VrlqQVZ=>UcP87Q*GD!NzX#q`y#o{K<DPxg!Uu39Kp8>AY3F(61;F;R)YBitBQ_k
z=r8#`^dfn}@A_Qd+(@g3vBCbkHFwSC0}Be+e)i2_u7Nl;nU+G~CE^#x@U#dbGIWW7
z$TKG9f3{xqlnQy9B4??lY~Ap|f=6N?pj#bre*-~%Nbw4r1bl@UqF!{3-tG(w3e>;+
z(vhwYbuN3#KJJQ)8vvtm2PjBMT%9O<s<#GgfR0gXtS74*5T?&_f`U8F4H^YFD0Szu
znU7&M4;Th#-`n^_Lv>=H%o<$w%@y3_<TQQ6(BY}od?Hu`Q>cYOx;5q&4J0vJ*=ads
zp{CSUJqKQLUH{hLyumu57OK?jvk)EmZBjQ5a(PJqFz9~+{8|MwV<BOpP|cj&{6s9&
zP!64oqq;}tL~1oY*!?C=LEh}_l%h+{iKuG6{izV<mgfZidw4EzR?ao<+2#Oy6*{Tb
zaPHJ|mwWmTiXKj!d3#>lFWkCkQo`oNsLsQ>53-YV5$s6D4#4y{)dThc*ieWuf)8{0
z%S9882-sjB?ftW<u!)Zso_u01?{of~q-Ar&U(X*~&}Hfq((H>1*F*Bwn4-uRr-wS4
zmrzlvZbdIRGVU)=^!7`6d!7R<EnLM~Gi*qI(u!30g4LA4fw8(XK735HZ?{wV+oaYG
z1FGYhRJX89=J#l|yV^U(6O}oNT4t>lF446xmzht`HCa&Xh*3+8PJO2F1Lk+`cq?9%
zhaPAv(NA%Le7bbd-bt<vE3hY-ejPxoEvFvqMk%wz*5Kc(udv!TA+e%M)l=(8$^_`(
z$)mFv{YxDItX-IkOn_Wnf9FATCY2W0*w5Q?ly>YJvNbKS^j1kKI$E5qdM(~;AT!`I
zUy-ywwA2M(2j|9ijgikHhl9@a)ELwChF!<_7+mhVE20;_@4R<i=KP4gU58?cKh^`q
zlXrXjw)OnEjB)~LjSi)?m;+R#jE<5;!>p1&Nk(O#R=X2lKRP{nK&<`eKB@h!M_NSU
z8icv}{OuSoI^UQ=2reT?(9ZtQ<%ijphcCAibsi;ci#?>Z5qQ8LIwX~aVu^z1Yk_k|
zM97N0qn#NS2aR;9v{yv6URL|t%5IW^s3b-}G_C%owPm&A%wu8GiygaevoW>LAfLjE
z5?G^gb1cm-xSJPYgBzIrghBUxUK$47C)({>?&M}@Oh%oXh9-TT4}{1Q%t2yD*bN%E
zk>w#x_(9)`G!QiK|8QAtgEx9h=2ENPWs99u7d`1%ru?+vd~hnF%IvOAr&a7%27qq@
zpJ4ks+bB9zw52A%`}r4bo|s0Q-k<OHa`;t!D@k1snyjF(<v`Kx7e;!q(BV|I^(7hr
zVYVKlk;u5yl-zpgP(s69^ploUaX0r}8kymvbCVZVL>0^4h<=_by9K*{QqU=Hd>ij8
znUHAk^aES*A-7oCuXC<s4cpbPg)~&%J9MvKlsZ>EXj)U$dhIgd{q=ZzGUuWS`&wW)
zV;y|Ln>A3COG1%bE@m|EkF`ZN)a&gSz5%~>HrC$sc4XfgPj*E!Zjb3bF17>(w!mmE
z>MXMcV@cL;s?vL}>7S~IRR%TdU7^AE+#ZyOoh;$lv6Is>47;4aBd^*fh;xfS{EP~|
zA*_ZUT-5XKvLN7WXs7l)sctgScTwFp{pP(?qyuuGZy;l0-MRg~o(a+lTQ^oIGDF&*
zy4DaAHgpK>NW=xLQ0cl>!F#TnFJ~w2e3T-OcB(%qB~+er46c11!jyPU&>^6Z!+1O9
zlX3(UBTqAVZcapm)$Td!i@s)_s%D)VTgSHSsM>vU-?Jo@I`(h7N_g@yOUm_2H1+9s
zxiP{Bin7U<TO*jd?0g%=;&`2JMEz|VA7u`vgwV|kA=s~7$S8QRGO3%=-3W(if?~?j
zP{XOf;byJEdfQQT?O}VVYHP+!Wt!Y3#M~3jZXsN}^{lWX+EZZ`LED{7+Dfqm+GPCC
zX!8@YIkJfb4z&3GQ%N@N&#KP3YotxJRx4dRZCM{8=Nx$t6rhg2#Mh6oQzOB{31JQ1
zw-?JO`IeRu$k>@<vNflV@G_^w^3@H2M#2rJ701V_J}Ih2Cd+)Bn6#jf&Y$WOR@CFP
z5m3E-h^%;Yai}FkHze4pR%=)xC%eSw`r^=Z+u4H#39VuBVse)k6m=mu;|uE|^kWPQ
z3X*h~Zbw5>)oB+BMrT<v<bwPH+Ss}7-1t#PX`Vvr(Ny_@JdQ94xwx6*Z(&sjnd(6M
zLGULd6eos3)NE6imH6`d$Bsy>jqRusdJ-Ul>5*GvJdlro8fnbdx%)E@Yn8BHYBe8^
z(n4BnyN8~HakYQu+Bn(jzj5+#i$dQ{BUovXu}GP~uu-)_*b$Z03*zj{uiTa`Sv(f>
z<E$<~m?1<M!q77l&9AnRm5Q2&&lD%Mk6!BcD%obkxA^cV{RH0yuS!`vfgog_tSCh(
zZc>x)yR0UeL7cJd*O03%dvq_J7xefJQ!aA)>$!nKvI?w$5SlLklX}goH%=}0%R&!N
zM2-#jU2S(&u&egP9B!*A-aV``5NK+4@$1J@ajy=4)(E^jxfw`UbmI7M*E0sI&j%qW
znqfmqJ=c=HNWBkac?J$1OR<)VI&0%J>m_2aolszTQ?k`&RBxy*=5y?aeEO(X(<2B$
z?>6V*Vk6jL(7aZ*KbVeS2zJ)0&dyf!d-J$^>YE;<D`i2ppx=HR{($~C)%@wyHu>5p
zd>xLa-2BRieh~W=E=F*YN&&Hm@ynB*o;l4^@w-ap-R@?t?j1OGBK?K#f)4VX@c)UL
z$N#6C6WeCrAMpzO7G^$fs^3r>H}HfAW9lIs=nh+d_S=4ZjkyYnwt9{5ja%vda0B>w
zgzSzl`#J{un;iCqgWadxvqRuBYTxIzw(`U>*%K%HM7|}02K!rtV?cG1&(&}*GXtxL
zcC(8<=&o6F_*K47mTGiGJLThoqo>um0e@es2`+}92epOfO?GH#gl{$6b0ZGzZP5?h
zP5Hier)Jl^BD^}Y^|U&)*B$xdj(%th=Fqyzy9Y}G{mUp^V7dTJjEI6TbST%;h)d$=
z=4SUjqZ&%n`m=?Xs}ZU`GpV9ZM~d}*+}@NPL?*jjg&0_|xW7*xzVR8jrhG@)??m!q
z_8K+RY+}02iC3O^g6GbhU)ya0`Dn)Wfw_&brlm03BD;5<xKi~m6YD7o)ox*ncl~yu
ziOM}Wem@(o2+Rymd;p<T<}D>H!DhZzuiEeW^w2f=V;bM0<SQPU9vB)u>l2xM?L@F$
zT)=t`T^r4U(v)IIzLR5t8IlqU3<)BB=at^0Cq5`Vw1Azek#l++>0u!~Y$BWxv?l2f
z&#q$m5S->M;lV&BuOpPTg<PhF%9_iyuiRnWsMSc<-D>m>64fauQ?*^~hc{7=G)7lp
zfHs?!$VIH8%MdZ!?3nsJ-&xJ2*LKgdZy#BBpERjiZk5&w(Om~SGshdUa<riit1CkQ
z?v9Xjoo+eRGewC0FkWzXY))sJlC{TYb9Zj-gt>92#LEx~x&^BhJeL-rA$ATOL@j1M
zD%eHY6%uwWeQ|N9Pr&Dnl*8VS-tKptStTw`iI_$g<7<<PCHcBB+i-P`uic&Ako8C=
z<S<grl$)QC>$-i5+@RuJ*Q$^U*A(}bF(M-+f4t*j@ob|AO+ZO0#IZnl`Maj{;PGBp
zVi$hA*hSvd)nw4Vm7eV(Oul9lpc?P_e!JBiY{+VIW&uWoKG<k#X+jCn5Y1H1c2w!$
z(8%DN?St1Jym)nkvy%?;URKX9Q@9eLA_15IaOap%Bs*4_aiENXde5Gl8?whbB(Q!|
zAI$L)Muq#{yRiK2Oyoo9$Q{QMl_~x2GnSvAc5Fh1$5@Y(su6%p6@<BxgN72VqRJ2C
z1T|DKr_)fmC=SK$fx?!d>`2)>gW1la@sh;$(*eYEMIf#O#$dq7)+Ai>n9x#lSmv$a
zT~nbX>8oodQS#RZ#7TSVEFb|r<6Dfl29RW%4j~5@;P?9?p4sZ;xsC2BeR%gRcV`{S
z!?3S~Yp=W(7yT^>sfBGvR*I3>0aWN6gp~#jeYXAOX$ltI#AQ!$QKn<<N~z8z&E(uF
zqL+u<&c|13Df37)_G>LiH{2aI@M+18w(DgB@k*|Jl&^*F=}V}LlRCrpCSs5!m2Oo&
zO#3t6DRc-!D2qy?>HCt;8PRkp9*s#3+bM`ZQva#o9~Mbp`j$UI<43a6uPWFGXiJ&3
ziYI=}3_RkG>878?_94q+NC(MCI~u%z^@*kMg)@)(r2$*95!^7;=6>k7;<5Jh6m^j-
zxYoM$3i1Mjt%6dK02bt9N=GmQmvuu@G(VGO*StT&xrdZf{$}}v-nmB1+eEv|Hb-7H
zYrT<CLsCaTsPKrUJb`#HPR1{MBc9Z2+Jx*Vw+y|YSiJj{&7je1-W1urT{R_+pO5t<
zNvIDWyEM8&e1{+|0+@h<LK%Ha>Erc<OWkl_r)+IwmX~pkXtY-7@e7}n(n*X{P*s$e
zSmWb6&Gli7d=$@bYs=uJEJcb~k!+Jgn}R9~QG1%(x1ARH{^Z5`jsEXQ5(r?G>L@|1
z#-$y&?^X91pw1Ggiu+hV%V)LdFERC8dpCIL=YdNt!T_RCv3?`$H<zjlACI}8*d76@
z@MeAL-L2^@jrSi>UhBLiepyH{)1L}ohSTH`AvZn!P{g&yLTioDD#Bor>upAB+(>!j
z+Cjtknma;sXWxynxbwo47aAtHr%rzw-BZmWYs`fx{-8@)TWFf3!N*hK<x4SG9*m1~
zDcL!^COG7RfR0?WgreUeazOl5?KVXZcfPZypkHr5Sn>1g)69fIbrz`F*V!2=(0J=R
z3u|NY((ZmW=GxI9w9q56Wp#_5k-yjM0etUR=vXCgA&Wd&=Pj%AmKK-g79J-dSKPIf
zNvM5$D?F}_w{(kLWNoj5w-%&-T8tq$`9A#sV0f8fjxx&)Tyvku=%0&c=_>>lf5yFR
zRZO|kp&?Ty*Nn&yh`-<unZj62<RgIK_U)Zt@E05oc=8+4I3=W3K~~X(etJLTfxr^7
z>o5-fc@wg`mw{kC`@w{CvLV^PnV*4AHDQ&Q`*2JH8>_KIGK?W2#5!R`M0J~QLb{t0
zA<nou4TGK8XgjJZP+iX8bM+)j=<*+@9-ZIU=A*9C>UU8*D~82>NsJ+!-yn$qw5+Qg
zq4Rm>QijPWcihdzgW<luk3*dfr#vddc=DN>Ai2S>suSD^!Lkyx!|s4+ZQoBa*!2>P
zB~D*bnm?21Ua+|EdfIi+uxqD4L2-v-5TDiE#Nqe?P69)?p9&hiCt<-T`hLvzK6l0)
zB3xp`)&#v_!aH!KGW=TNm1t{A^<<$Fp@I9p9$6d-w7Si8htA8mPoN>da&CJ|%gX!S
z@w$b~azDAVOvi^4ZQ2<t_bB^`kxuOPZh!4aHV5rZ2=r4oVKxfj89_e)BTJHL#*8%F
z`jx@POu@*Is~fd<_KsKj7M|QIWBP3&%Ah!*bKS1dM_MLjrqEBD83vjSJAw(!4Fow8
z8VmAh&6rd!(lD95(Uw&?VUp=8nbx)m33<J`RxOr(liwZoc+@#@yI#S!JC%{n$_7|Z
zT%8)jxPZY&V^X}Sx|FAl2xm{-9}kxjozm+Xg$@A~W|vBrYe=d>8+LJcXxLUc(M|Bk
zgP9vDXEN-07$D6E3>_S0ea2P5>+!?p4HXF8=*)Sy{1)h^FP~k~kLFpEC5iPWlFpHh
z7MXl8Lt#)-3`oE<$>s~|C|1gZW{5GoTo}_O6M*uRP9MEmus^q}3EP8l?XHD7V03$D
zX!Fi5F45FZM3B6*E{5$Y54-XW>BhbiRPQbou|rHT0QrbOK86gjC4`9)ktEkirxj0v
z%$&#I@QH@4Su5CHKly~-Yj5r$9k+=9TE2}XLi8)^)BSr@TP&0kQAGXSx+o2cVboN6
z@xBMQ6;_6{JQ&wh*xTg6b>SIoG@QFTxsvKaJ6C{CKK@0x;R@<TMDB3OqQ+xKUV{ls
z?StsVl2aFJ4*`hD6*m+Da>_b=7|oi%@r~v_&g*xh4~^2@`bG@pH+8Bg+x>P0Jub_2
zn;M(6-$W-X6Z|pUPzU?5R+^by1IwW~Hp#-s?fGt3pj76&ZzJVV^EYl}+P57@j2m${
z@%ZaEFZa{v_cUG>PdOdQI7}M`Fw}TUU49)gO|PLU&`q)r@5Gm+eJqsPN6vole*46N
zP2FJi)qEX6VfIx`ldEH7i;uk0n~<iwYFJXD9DWbDMs?iBd~`CaMG82>?>~SB>AS#9
z2y5GJ1L~rT)dV(2KyK<*2qJYq?DQw&HN3$;eqxnvr96g_YlzC<gv==;J4J8@0fmzg
zF3(H?Ri?s2dM!Aa21Tq`pI|!~A(=evYM6A-ElevjNvo>5gs<Z3WlRISF#Q0tmf(bA
zr|m&Wex_)e_r)G@df0NjI(+SPT5)2e-O=Ww*A1?p4$oZ@kpL}nL3D{7xbcE9>0;)R
zFz$Oh4;eYsO+<tO_d`?l>j>Y6=Nw@eSCb7kT)J0PI0|*^%DUsHgsq7w^({JOn-Fzp
ze}shkAp%Va9$F6CNx)MuuJN5j30qs_%wd(k=oVgeb3LDb)KhKPE=T-w?yW)Na{-BU
zc;KJ_8Pq|GnFM#ob6|%LX$17F@}F|MP|eLldRta0>M^VzE!^SEdKGgP7AO=8bpm$x
znAMA1bv0CAlxwms_BY}L81D3=n0w5kS_y3D%s4EpM&QoG@Q9>ZQv`zW#x8VW=|rJ=
zywtHfk25~qoVs~h2dHi#>oO`fAqyFkxNWABGH~L<pxmLFBQ_d8)M)Cc%RemKb3J3V
z3~py@({lni7!ni6;mJuDNrLY39psYm>*9sCSYiWZ3g-wlMcx{&>`B${w%bVVl#^Gx
zJz{r7ElCvucm7GoV?Yq=r(Q8|KcINf-xJv9Hk1~!ur^NG(wm~8ixkNED$o`DuHc(M
z%(LqwF&DVX(W$sko(QnwU;6p20h89IZ6iWs2qo0q7c?^k&)4WviyGLz!SiLw@_oA<
zQ-xdfTlZGW`&;hK3y9~2ti=pM>kq*LfQs!>aJp>Q&S-I=hE(^)^waB3Z>lRTBf|Vz
z6OJt+Ug1^1$;8sk%_!ar*C?`w9mGdgBaW4;*(gdk-ErN!w?ww}a~1sd)IHr5I76tL
zx=z!6&47~o>tn*xRf@<V5g0M&?T<KL?X?+pu(gA#_D-Z;wEi5LD9~Bt{=SGqAVv>-
z|6|ry0Gys7r_WG5X*&ytY5k3|+UY~QgJmzolE}8%^@u$OpR30s_JxKD=v}dNc2yB)
zpC!-~lS7wu#|9cKimJa_8TWcxid#xg>!`e)9N0I76(J`xcC0j9Ovdh|6;EyxIoY0k
zCC9qC5mTp`^Cpq(yT#oK9*iww)g#%<af2(=OXLiH%t^69hOs>2<=h=7nCje**4i^I
zQqLo^-2=_ODN0*SYch7U;F~+>z`E9U8NmoWU`i5p3$D%sJ5z_ihk|#dkkt+Xodp=p
z2_TYj$jOaAJ?@&j^A?r*?2D8Fm(2T+OdiHAaK+Qij=&70z$RpIKj9}ce{xiF&1xO=
zl6OsxL0wZ@7@}jAr8U703O+AJV~yn6Tr;yN70}z1-5oDGY*`_0>nsYKk;`>_V;ETe
zezy%az;*|ogS+JUFp;Z8O2zP)u<K>H@5_k$VlkIbsZA|y$K0m01Q9@iSh!Z3hMWMV
z*Q9|u(OW+2UN(O%Ew?P}mSK}jgO8T|q0osfnpKcdPICy;8Em5<GK){7bg(7}8)tm%
zf%+Z73S59M81%4YxQ`b8$&LVF9FJd~vl%yeGc<QsJvlV%fR5dzt+6leek>!~UuK^j
zqxr;;E^I=~5{U56XMv26JEoYdd7AcZ5`bdx5Q0mT`FKZY4SNnPp7g9I)6Z~q$aLm=
zo3mocDQf3Ss{|qK%Mk3{`~9T5*rALPN7@Nm55|T(L%ke&{QR5%Z;i-E<$@0#krK1+
z{0#;6;940bh@0JiGUdvXNoR=q@U7N%witn~6D0<f`^`xf+%cak5QpvA4mW1Csd&2^
zjr(=Fu+PQ-KNhlZg>FF0G69|S)keiVhC9tTHjK+kETe89O32apM?9gY1YX2h9ptg#
zfF=MKAh4V^)3Vq{0Kt`W6s{?9s=&ji<OACemqHC^2JXK0JW_?#U5OPUgE~=sw;l6^
z`-2Q<>(I?&sh1Ih<;hfZ^vIyN7FyvQSNr%YH6^pXYHMnZg!1Ht>TyjTnBWbZAmgBH
zB5M0=e8=g^xA8~M)YomTHi(zz+iJRxjiyG%H)V)2+<^ICCgUOu_0SG5PFlm+H&Kga
zAK2OWOw8_8ewDPa^EKL0VRsPjc^^u^jhDlhZ?J&dE&kL3v=cUhv3D_Qw7xN*`;wO~
zOzXXq+mZD>eRuEsZiT8#k<>E-A9IHY&-VJ5czrQj{<N*xZ574h{TbNfCt~Q*1-R9*
zm#4Xg0Vo=>fYGHZVC1`3X?-fW<i51~b;2rp$BtB<EoOh~W1aAI1^F09wzspEst3bl
zG;P>Y_1muM-n=U^S+1_LcpiIUqnxe)a)CSs*d&YfjR4n^+ie+$i-l}L<efp!3HJ@c
zhU%!lVG+3P4{!8e1f6&pLTv($8q*n>M_R%Bl?vaL1uJkZGYr87L}dvx#1PiOeR}Q7
zywEXoZ^^VCM??A<^NCAvlE=8)QYfLwJkmh>h55`SD4*?E^W%>mFSt<^FBcn)gJs&)
zTHxcc4%%fD*n*$)1ntQhOSsQdO5%LFc{=gOsl(iF-mh``8~TG$9w5N2-(V&&u1%<^
z;mX_&Il+b{PqS-M+YBVO$gx(t)?;Hfz9SETZB7Y$&<+q1E*c8;Yl@&)5??<%>U?-F
z(w)so>@!pL4}WniBr}|F-fjIHtsn0;R^;4*A5xe=glJ5Kn9!a*Yuo;)m=V{+`AYSp
zaQJ=V(C$5?o@j4qGw3U02D<No-{=^|6O&0k1&v>3M`hxV-NR`KlrQ!S*#xBPh(w-{
z-X1yk^*RBQ4`l6>z}aZh&caJVg4I!lj3~FJ<DN_t1^#JdUYSb6OYLFx<L_55_pW(D
zaE1ht$7l~0@0p2{U`oQ;+2j*~3-heF94Rwp*lMc>3E6o82WJUaCQO<$L}Tpw_XV$%
zb!?GH{a&>o>5!W0i<?^T$NiY!fc!SGZVPD2{`LCfJ%DcZ|5FI0y*I!ET?v91HX<s-
zwzb>tJ8O5h=DqwB_fKJ24>!01%_6HxLU^WLshZu1H5X?4sWi?}$4tjUv0b?0nQ#LQ
zif=>gd{V`z&b#=bBOUy0Wl_1-TRJ<0)Wg#CAj5X-x8<y6z3V}1gVjPJ*}3r|YY(|2
z$n06c;G6iNnG0`{=nIRvcWdq($K-e(P^Ihbmf(<F%@}m1+L84Z631oCCuaP051XGa
z>iKa{*JfAGsDzSD&O>pYSAi;4*9Xq~pMRRruEq6b6=D^IxyDRq@v%B>6>XsarQjK&
zg^vytSH_?0vL8-rUQ!||S!9g_yixNXkl+$oC_BwHWBw3CwTZ}LRETHj4{vf*U%}Pe
z-?vh;JT%@YWNPPU-dUgrEOO&J4+|abs?p81nU$3(3OdofkCr6mX6Mo;zB7zdL?H@t
z;q)Phw(X<ST)~)|%oG+Uwrl?mYsO`A;_I5eAoq74HMdUjrVTs34T!&X#770@de|7B
zR0gB%f-hJ-o14$Km{5Cn!fg1<(8}4W_^$%pDz9@wN|U(W$P>(7X7Z)+8h0`@vj%Ct
z4Yq!@i8<dX1^fFdiq*ADBv+438i}00&xSmVK{KPlh7mJ#YU8stt#aag&yP!na3!e$
z<@7I8Ch?<o2HVrGZm~E$5g6|^sSH2aWhCoLj+lV@8*Zh6It#u*<6s#c>e{AOcWX_^
zKGAh*?+5Sp#Z8D#@l9t)Is~5%Q0qEF+%;&_#*db826DWttY1_7UDINUL51Kc(r1HL
zyxiGd%DV=6!8EC%^QXC*$2ftWBYY4se!ZG2yhdpJ@r!=<<7Ik7kG_nx@xJ<izUbvb
zjFpX%RP*icHPj%#4w}UC%o}c_T)+?CL+-Ho>bllPAJsh6H@83BWv%(s{9Ey;PjwIP
zavSs;n?lgQ-5~rkwt-7V@m@l!jZcGR$<2!Jy}$9G(=Ts&$+IV}MaZId&c^<lym6V9
z!(L7c<<6Q56H(CSQL#P)38F8mVs?m`)|}zXxfat_Q5E$(Mep~&(g(zNCJ9>&*)Y49
zX*JqD<6}XoVhn|i!_3LCHm;%9x7NPa+-dCimtXL&L$~j>P^W#SA7)4}Gs}*Z!6H~9
z<rI;3#f~V&C(qwjYDA5mY&?-*Ud?{nimhnxcC|s5OlW%ys9n~n60VMIpBI>0nvg2A
zZk+hSOER0byKQddXk|1xxW@if0GSDfS!AKzj+J&A=wIPXdErV?2TP!<K+8msPt{x!
zl9zc8a<yC~6PnSoP)SzAc!CmLlB~ogBmy*U7$Yl2aUB^0@X6s0x*o$9yTr7jBYmgw
znLj6y(m?A2z+9BnD0<(sK;Jj<Q!N88Lt~Q9H6Y6n^(TGDXCjvNEbNQf;rFB?y(!D`
z5nZFz%A)8~yxix;%x_SRNv>G7^%S}|>m!m4=NdC$GJ#@f5crj{{Bl9Ww5yyUXyN7~
zDC0nWBy{wAiL;`@;%N)HEgr~QnC;Am@ayowc`Kw4<`DUTqeG;CYz|I6x$=yYyP@TY
z)}r`r9L;&h6osPIdg0DXfrb*yr!2u)Ff|G0<MYy0a$A*F2e#|oXXfi^yX4WU$2upz
z$z9=sI4di?OZwZ4++|#3279e2w$o4$v4w%`Om37^w9+0`<Ts#x+arJRe4JB7{BAa$
zg#$Nt96Km750sanKU-e|u{&tWmNXCYjO8V%hALG9ucXlB;|ex1*_fJwWo>7#t2SYZ
zPg7@qjPM?%7{>TaED2g0=Y78)d6K`au*Y(r(9uc0c(d{R-7?1Fg|;f}E4imZfN`GH
zxbg6fXTtdK6ZFX7XuHhW9~zH~Ke8RI?zzw5$(||R<j3z$--&HY0>UzgKv7xt;!^VU
zH&@)~2P9pbs@!};UDDmMW9_T6UaB@<Y7lQffX@bxTns3}u8~9P2wb;VY-kS?HzXU<
zc69Lbi}G(@-w)5}v$`ca)H@Ds<BP`Hu$^su&2I~WJgEuu1<HgS=&F<H1mp%V!$VWC
zXICZtWZUR`iet2r-`d$vle-1^pH)y|_gWaf0~WW`jPr+a0q_V8HJ^xyU`bR^_|vL&
zhRdE1-&A@*bc_v7#5bSfBo(k<=H%eBY9&D7<0-J>{)P;JI^q*3mtnE53-OrPh3eF0
z?zB_J%Q6oT(W6%hcw72%ID|~#XSLzB8tkM5+0fP|#<u$<v|#1J#%p`9{a&n_cYS=5
zKlr4MczjddEv_K#dGLp(DB8S!)PM{Wco?qPL-VQopQm4HN}IAY5PW@CPa$wSH-~Jd
zzZo}Yl`6RZxHI7XpT;&FdidJ&7-P4uv(WxYH_lgH8twN3s?2U54vO?5*0#`0$Uvd@
zJrY{^5GWHmbT@eIb=MIWE?eVE8S3uX>H6YILXbi+Wc-dpsp)T1o-r)=H9~;GB%ZOa
zh}^-jkJ;d1R;;ivth!Bz?5Ou3VRCp4H%AzvkLmicnF_OkMj-%cfUq`&lJD?_f(*xq
zS4fUbrtHz&_v%}b>&RQ@_ydoREH-|7+ESBV;P#xhi_XVtg_lzCurNaz+Fs%Qs~yQN
z$1F8|^w>yQgymkh4&w-Pyr3w!t;r8%3fT#Xl>IRdV~1bs2!NL&sbh@Y2tE)?0t9E9
z*kJ7#>xqxgj9)_AGH!Cwr-HPSO@G|()6Qw=58ME71qSSs84J~*h2fE;9IbyMgLVw8
z3>i9N0N#yKXJnMGW4#&oPi`lT9`|ERN;@hWGV%nD;30fGV1>`n@q?&w)1?+}f-7$Z
zABq;iIg*+?jN;sH=NE~~3Tvv|fytPA$6C`}-3qL%1Ed!KC-`}qeL1h^VO$IyW`H*l
zsYy3x4f}Y`_^2;Dy@XIXAtBBP@JPSUcVDbABRFF_ON-`!;=~}x%-pbjt{)sx-r~2T
z*|vaIkIC3AM)y=3lsT*Cz~gp@|7_<xG8gqyFBmnt#e%-G7jF>IVRSUm9yffPj-Qxm
z_&Kv_8%BpK2rD@&ik0pvZ2wkK8hSM61Ee}`#UK*$b<uMFL8a<c6?UNK4wfbd!KF=U
z>=><X!p+a~X4wAzt?9hgel3cocX$=^3XHx!tEk`{zk@tL(`0SK(M4&R=VGi7>@qi|
zB{<@u&QhlrMe07BV?VYt&M(`n<35w^xxe9EN<Wwkau{=r@{_%QuB_h;55vgFWFIme
zDIfV3GpJDfxZ~z4k?gLcN6%`jz1@E39<gdE3dx2QVs*)il#>z78c>!Xzvw55E_WfT
z+rzbD?!Z(S(yyTSVEyq)*Q+&Ck^biRE+gr+2bkTNXrcNly+F)vH>!+S1J!H8FRFGe
z`C~<Sq^Te(E@k+Jid0I3Y<5DMk5vn7Fl#)SwDVdvfSV)@ey$#o+7QK(LHzI@l0oiM
z)ej!7`*xKG`z-`SYderkA<lI<u6-Nl?@Qm_T@)^GQ~qYvx@z?nnL$PO@tK25$WTJD
zUs}t8EB!9Ugc)BxXba;*Piw!T-LMI3tUa~v-fL4;YV@T^0ViAp;X2Ozga?_n>(U?`
z>^~7`ONUF>C+lUk$%fBlSP8Ffj>dmAIurCb{4HB-RaRt1VsbPuCu9R@MG*^O7|I`_
z=DNKITl2X-?G%EoOB$|;BuPKLCqmmwKbx_YH3B08KOBPNVr=Of8V#7Rbc(ptL*kA<
zs4<@MdP!br&xJUl9a&q9Qq83M=lX$c<xZM2^&MF>Y<x`It<qMbJ=mK%fKq=RReAiX
zOPcLLJxo;f0X}ZIyEk$dQs@E<1!mknk|{%U?NCdoZld*)r`z&Fui)+99_})|5Uw$%
zeI~vwuyy#n3fDkfi?nvZ8I_|(pf?=C?hlcH;5BI@lygUE3gqM_q)5$?PZ|b#S9Ckg
z+-vW>cMpW%-tEg#KzfPHc`kR5Ea+N;*q#l&hcz-poY19NmhczP13zRe5oM9i=a-8F
z#U2-FXKvk4!?a4}X+SUlCgDh-X}<o>;-*-Uj42R<bQK!jSG}_)$nA+p%2fQ#+s1F-
zIOC3lNwGh~_g6j3`F`Ik?1imM)!{O)i=sPPP$&*n!2{32j>yYLi(D`Q>K7u|e{Kf-
zzc>E;fBF0H&CWmOUcWknRQ@|VhW;yli+!`}kIyl0<=FRKRj=NA%h97EagXyw&dxAB
z@yvVh4UJj$M}6gfN4ig=HNFk_@8Rq{Z2qPIPdlcITVDeGfSR=P=>U$3EyyrsT?jUk
zf|Fx`wBO3uCPW%Z&H}C+$?^rbEl3=qOd$ZVuQvuom<}A{4z51tcVBW*SD397IAu=!
zCgkuYgz_`fg-^)!ZbD`VtOJAz>z>m<T~bk!zyKhAix>mI?GJ(gzzSqpX~a!Pu;ZCu
zKmDJ+;r%vq8<AC>tIL+jA)g1l@KVvx8=wB+#u_%Tj3tx5XfnC`WPUxr|5>YcXny1E
zWj^L6q>3<cNiP(&xNw%;O{xA{>HCAthfQy9TkD@wE2W?|s!{p4<B*1bc=E8{TXMk-
z6?W&}S+g+icZOX2GLs)f)_?a1ATjwjhTQzze?RVb#%%sOLspXvhYbGu<>0@5w=cek
zD!8F{Kk~G_<y$E)vtAkD^!m4|L*z9^@HpU(;bxWm*@^r}W#Bk}bCt@!zx&cZ_C5Mv
z>v%DybZ2ZAOuSb323E9^%oPeR$?ADl?G`|_vso)UDSnPVo2?)fa{|n`Shz#Cz_`+0
zP0IKlr+}hIxyH=`QI~{;q*zkyr~Ic4JVJE}9B~ddAjcB~yIn^he0~%NVwyZG!i}eo
zEua1=0d%?iHx=rXo|b;r2-UD_dBoKl6;jc%_5vET95TV%I=Vfh@1P<xSJ7vQlu9`K
zwxZjw42~rH`<B7fV)6bTYB{T#m6ve!hpD7Z2%*{gG8h)A9i903_g<lrFnxX#;tx0h
zKN^2+-v6^g7I)21-MdCB{}PCO>C=re>pzpky=y3<lKn<{9`ufV9&_SHVayKP2L~KL
zTLV?tejly<a`69SHZ_qR^nKRiy0RXk=9&F%F8d9u3@i5u_zx@i1y&i`!%?~3It74i
zfslLm@1L*@|5{AnlV36Se>&#=|24_EKk_I>_o?8AyB(6nhS`y+rRP_d*13kjh#Ey8
zYHVu%TTvPMX9I}xlHv1(xcO&b2SZ()%jwvdM{dmGR_!{uxQJjN>&I2(H|q=4e?L6`
zS9?;}-d@>+$VFLx7h^qe+g_;-pAvZGCl5|8>Pja-lTZJZ7kk1IzQ@X$IQGM9>>?|%
zS$Pck@^wcOVUAQ=$W&nu|EnNHAs!0c8<Zx-3Um{#r~zr|PBeTRxv?#i&#~vr+4L9j
z{~=%m2in8Wk;@WfbEfYYZuWK`n2b2gIt~SesRdEO#$h#I7SAs5of#as{&#Dv5PrZQ
zsyj@W%A{qm>Ht?I>kDD`t4+u*u{PXy<qeGhFi-9`LITsUqt905IpHF_J@!v7jlFUc
zl5&p4z6mMq7q~=Ve99ohnCWl+`mC(G|JKv1$zaJFeJ!EzDNQ&7lunF=*&_bpgJZi?
z`Ct>$YtOXW=ok8bk2|?U$IwjIvT1tD{K>z>gMeCh6Q*F`9@+!g)mY^2tM6mWP!OHH
zwg9^N;ctIa&H7U%s$XXCM?C8<n?d18$swtCM-J=}yQ_6GGOkl-$bauFz5|rd0~64?
z)_8`RT~JWWIuDMTilF(xgK)E{_PFtGK+Wys0+hf1uRz;DpMHYXb>E!C<(KK;YABu1
z&~P-GsAsdMwtG6|(+=Z-CteF{?QZ&1H`;m*20~FP#Lrc+(*kt$jc*p4dr%!E1l1Nt
z%2&F@@t<zsKANr|H!N5YmvK`c31rJn2%>cDYB&QxTp<ZGJyuvw3yU94KM5l*<G=-U
z?Rx`^H5Xv|_}-Y6Z03&V#WlM?K~NSAZE(=<UfBKw7OEb<v_RVVsAbxttd{f6e1P4)
ze{`B`fR)$^Oq>k}=FW${9%EfYfLN--{%?giVZfUF&%O@+gKdgL6VQHcj*YNov5cQC
zAu}A(^d~;!pTZSi57*=7USso_Zef2vbYYJ`)nLnpJp3kqy!cBFoU6d~-`M7#?^u5#
zc=gLv|LaV3n7qUYezu_pHafwIpQ5)rU9eGLwWLB=c3_LL-GrPw(-Xf5Da5^cIlE<~
z(u}eD7It>yi`_j`ebUTgOd<B%Io5H_RKP_<tlEKfrgRb*Ida4A(d_eZru8P|5Ms7;
zT$Ax+sR2pf0#?214<4Yd;7?da5x|HQUKAt0=MPu&Ya;_hh<|D$`x6<lU)J%5^#u8=
z8`&@G_#YkS{YlgMFYEYa9sjw6=znbz^e2SsU)J%<I(}IP5Z8wN8Rg}db^Ncjj@-)f
zH&q@oRTAQSv4>S>sQW=~ZrEld@#td4zZ{$Y>w7ZFe+Sl9p(XY|38_?s`ugylO2Sy%
zUF5(}E`zQMxqikL_qW5d>7#D02Kf(EXzlzElfQtMu>&p6!kg$I))7#Zd5>Eau%2HF
z>8mg>2FaQa==XqcaA)!XSnxlDB!4nu|MW$wQD)`M%n|zDr2hTT|3Jl004`T=6mIMO
zpGG2nA{Y0Y(*W2@M{Z1IY;fl%B2IzIsUMzs7s2GX@Cl*<x_SSUi}VYM{sSKKAEz$o
zBcwjdMQgu(JkItt^&K+?EZdG!)_Jxkrmi^0Io!&^CL|Fyu#v%_!@n6xlwC}w8UpCb
zuw(tq@BY@ohEh6-3o>M*A<Zaa;j7gnXFMZP+~aNzX<f*=G>FoE<S3iCK=_xhL1w_o
zXAdY<(yol7bB{#pTBbv5_OUs$WHrc0R<Mi*iw>H>Cm-P&nre9gx}WtLH|90QxJ!_F
znF$?)(a@E3xD8>?4*DVh1EK+61Pa*2n~)>XZ^0m?VI1WX<JNP3xD#%II0iV?vuFC@
zOVudWL5DiP<Zc2ZTq20N$pIL>3FbKjp(@<xtlNtRjU3v^m%a3*W3>+8-_OylSlj5E
z5M|wGp!ksv{>ff4f{9tO%2_Y6Ui~Yi(YdwyRut{*S7z6b?26g3V2zFum@VT)F5?kW
z7knaPwt|aw;t61y5?1NEyG%tkidpwEx}4u5*Y(C85~y1%Sr_I+z5)CWH#LTp1gfB-
zW9wEg|6ar3(w(H|KY<0~c4*#w;p-cL+kT_OT=C|at}lb1N{Z?@1Y*tQ`d2`4jn|GC
z!`yzTgT=Q8NBf350vE(h#bL}qd#5v?xP7t0PuKeE1NMgPm@sbkP9KhZ#RF9200l0%
zkYdgXk#PnC4#eP$yKi_@*a?4lH|8Hcrv|9;fBZJJ4gK;Yi?U0Cu1`NIrg~=SXn(wV
zl_MT;)@?l)|9!N9Wh5!a3<3nc)w7)(Wxt@wU&XXPh5`Q(vi$FRFrF%B#sd;z{yBgN
z8N9S_K)N+Lje;j~X5`WuY%0ntDBJVFZt~Jz@lJE{UDfBRZ+P+`5H^3cweAyOLrJS-
z-b2nH+r3-H*7uBoy1{}|AAGgM7V;TbLKDS<72*f1h0_YyM-0s#DF&bQ?uV1F{j?yp
zc7U*%|J6UM&7*xTbd!~1Mi@fphJLJWB(4rs;HMEDYA4l5>7JL`r;6W0_|hRw7#Spr
z0rtKGX`0?9<cd2m*W=2W(9oY7;Ln#D2}x64Vm$2&*@QUU4`C&|!(2cD5^WtEH#J=b
z-&pB^ucpni6Tjp72t7TXiiCOO+>Jw!O7|`Tk6lKUVOi3M6glTDJ!>BMs0XMJL`)M_
z6PTj!L6dj_ZakTlaqZCG%7qCV>NMbumFVPt)_dTeKfOCrL-+?+_O~<oc~(;_P&580
z|5JGQYC9gdLp-w04|g2)C%)<LAq6XK0Y1g{3^%YJcr{ZEoVEy4e*pJ^{Ro)i<(|O`
zw}mAAh0bO!%~6=`0ELSYJo6i$XjZ<s{~v-OBk{lDl77Z0BEx6|^Fl{2uJDfT-wz>o
z+<(#=PW@Kf`1>9C*XQ-m{Qo0LpJhxX{P+l86|hoG|9u?v=w}&-?a$`m{7NnU=u`eV
zzrpgT1+!g0deLHl-nLRPn6ew+zv}n`_u<qqt4ebPKchdn#7IVkf*Hws-C!Fw8W5jd
z|2jR%k3i99o$weRV+NOJFjdlP<?xj(W!%STj1*Y56{X<IggUtze=e5a77P_km0Dw+
zax>Y4;17d;EJkZp@#VPj48pR9EsW_i2irKc%0{{OH#3b#{=omx{eRfI@_4BB{y$Qw
zq-h~VOq&!*g|y(Pt|V8|rif`1a-~AVm_w4XO;IV8Q7R%PA&F+lmL$8el(A-=%!n~&
zIeuT=#}b}<?{k0mxzF?5JAd>#V>)N%%=di1pYLaRzuzAjl<AOzp+D$zj9+$qeox#8
z8;xn!<Ilwb-zUh8Z3Cc%rf`$_eRC{B5;5qlBk?_KflKG&HRP3_g~u*S$|Lj@sB1fU
z30x5$4M?zx)UVj#eo^pF7N!eTNa>IU=IFsq$}OK!ix;IYh*Acl6hZlMOhhu?UN2ay
zg0(uBVHipy{Mi`=K`~UYR>y<$2=>(>H+^u84j%twYgJI16|B{Nb*<VnX1wbG5;?R>
z=}L`1S>4R>lE1WTbL+<igp)l`2YoBzX3>u#OY@dbxFO-h-2}+-#Dj3>r2#%^d?LE9
zCm3PdF~X+p-A5aFGV-AmYA(le>nLyB*V~Qpo3X`|%y#`qt(iET^L5TZZr6Ce#X`#t
zQad((7^F8k*hP*ia-5!l6vbpO;K5fzeDMS!rNOA<UpUUdrfMa|)ua1FK~xQ}f`vNA
zNTVZS_cd1k$R_-j=yc>ciZ2*MPBbwJz`W{hhF_!Vt2;3<>+XI-;*JI`{+7Rm+NGR$
zo^-f95N=r<yH9gOJp3)n6CO{?OArAIB4ERzotgu2xA1neuQ9lFmTA6NVdp!$3Xh~R
zZbZMoX}nhZ`P!VBlP=5(6lzfGf_bVWI*68aB+Wo|tdGKu>V>MR5xodd`T+SWHwxk*
zhTEhq3ux(Lirmd`kq#ARXxSkO1hzG{N)+)~l>(@eUt#9HdZ6S25cC;4kV0&?kt2<V
zzSEEL2}h)iSr<T{>>CSeZ8#clpR<>^5J#D-iaW@@#}wYm&Ux%i%i4G|#oDgc?5I`!
z>*PQOy>*jn?&qRvuMxLm?@6#o9U$i@3rWEpn?Zt7g5(o9Wc)`Z*r{J<VM|_n^W0UH
zxsw4ow&hXS)Zeqb|FWQ!ag-<omP24UhAl_0BpqbPmX@$!P(;%Lt6uvOS%&nIVq@30
zj0iG%{41<}6lbCugFci(bBpEzLDd{WHjz8ik-(p1Qy;}OW^L$45kR+3*1~oSXBstY
znD=Dd-FXBQx5Y$}^cOH^g-ze(QHxkmB^yaO8&7jYQ0E`o00y&Zf8Pw6V8!~3HQ+-$
zoBsgirg$8cAi;_itk@y%{6Di|Id94vTQ<&{NjBb@Akw5UYj&d)%2wi7M=<tlRzIpU
zNVjqorP;S1MO)^sShQa&*Rda^fkpaM-2&SYX9~Z|YQpOgXB*hqT-;qSM<JkTySR|p
zZ;Z+Rs_}S<V8sva!{Z*4DKgIO>5wQ-T0H>g>ZJDgwpix5tgy@PtDCqKg|Fs-_a>Ii
zRfY?l0H7@$N2$}s7#U$#fnf3!K9ANdQwlS%%53m|W6s)7^`<l`i|_zAW3yTphPSS{
z<288sNf)qS0Z-1$G>#8|aVMmuAHFiM+jaDIZEIJ<HNp-Qhxkhf82(d+Jl%$?$);wT
z=z8JQo?;E7+JdKwz0$K-YE`VReZWts6YB$G;Kmdj_@=lZK-4SFwGz%9Ob;p2`%zbD
zUB`%k@!FyE6x5J^=wVJ8T`zanS~xgdqs!nu%zbX4^gNLd>ewMblefJ9q~nU6!qC2t
z{uox_cpkl57S|NcRUK&8++hep`NBmcZ>vGOElkH&G$R^#Re;v>d<wTHVkL~LC~*Zm
zXPz|}0ue9-h5N}0!>n`4F3L|d0%abu0B4%OX!hXka|&-IR+P^<Kl6I!=k;#+Usr5c
zFVW|UJp-p}gT`1X0P!ob6EW15JqRAO6L$v;es?5t21u!h!x?5sF|iwJgYLpS9bBCb
z$_?|oPLkj#a;NYPv4aNGHGOC^d$|6mTfKL1)rm>H=4X<>uA>Wg{&a}1$UH?crhxu!
zT@%<5Jw@=nPbfSAfKPr)`VQ5^&1g53T?o>bsr@J*Auw_O#_xpymJ84Q9bldb1J|V3
zO$EKRlR2cG*k!}7!)7RvnVb^q{BP1ClYXN$9(kT@Tm<}9i;5)6BNr5!_Ixh_@D2G4
z^$TkGgKJ^Dqi_?ZIvg<oTwf$(F2q&C&!c{Eq89<sSS|>qkN9HvA(ZLEtEok~(h%9H
zUf0X_3PTd#<J<uJjLyU$*#^h@QTlSRMAja#^!?f3{UcUTj6FtRY<|hukm~_$+G7C7
z1Ui5{n{kg9`Ymfw8wPio#Xudx5ir0FdJm=o6?<Hfso@|Aq3i>jN%noBHi_X$aJnko
zIEiUE0H@sa3dNTlw9UVA6v`V-CwS^O+5+sF{Cmn5-xlb(nmPTbefB-r_Cp54ueBGL
z4hRR7+!-h}jL%FMtWo}xqvo;iNCgoi*Z;TF4%{yQ1dO-P2j3Gc_<ych@bC971PlJ3
zf)alJf)}6)1{e2uN5O&@Eco+369)e(JG_9yK6arKn<%}8O}d~3=O^d}YON3a^D9}?
zev|{U7xiuh^7E_ncKj01P&1d7q|3V_2P$KiR>G{4yfvhcplAI<3I1*6=WF^=bAERG
zhceW{0>}Rk%`yo5fPaG@AV3LxXBM6;201sPkx}V2IK~jraPuft@vJz0n3gopjcI%#
zRkEJcL6<bJSz~9%l3wxR{EO3|7l6jIO?Ok`nh1wxtX6C$u}_hzA`ov+5T$<F)Q`Fr
zrCc3^?7WkM>8yZTC%wkq(Z&F)lHrIigSs^^yt%D3p1&?n-Ixxt09(^RQuNfle$?Bt
zF2H~4p^SOI6#o87>{1}X@X%a=oyZhvknB*eSDua2ErSz#q_L8wRXWPoCXp-V*36$k
z_gJ#$(zy*wG<vcXzn&w10!?H=RGfunZOQ<&wHG~1iQyvuB<>0jUT5<FTj*H=raYNI
zpmdp$mw;?w?o+t+cw!0i4(>WbK~p-EM0#S6{qM1c{)DQ`M=WC()VHdf$Py}{1W_{*
zdnI||Xhc6Tg75Co+1M?_pbzHmM&~1V_eo`)>}3R=J9)(bZPQ5`6JGio$;5wEZ9@Sx
z;up<S24<9v!5+kaBz8H}tnWwR!2;i|)ZSP%#L?tBWe~dn2~YyA=tLud)zr*=tWIyJ
z34<tr^nMM7FPB8R969)f0ypN5EtdQaH|9RbUqVCqOE)f8f!&Zv69WKmN&tQx!?7Z+
zT(zKMa?-8rz>8}S=}Y4R6&JmWTj1GODAF+*{b-)n4fYjKLuuvih6_Jo5YeYFI}5|8
zAtuW6#X#D+_Ek<l%5Rqlz2*-%*CAAlpA1>Ym}YTgZ@~308(+L8zqqsyDR^#Hl&cUu
z{j$(}k;n}|;NqmBG1v#M#^i7_`lh`bE+&ntm;5z}%>v-apn3Ur9PM@q?AiZL@%LY}
zXLJmx-E^JABw)B?8BplXA^Ap+c@Gr7BL{^a;@BV)1DwxGaEH9bfa3ehwhp{3{1)^m
zC32@$!Y$ol=vL1WrX<G94%w1~=|ndwD*V%W$%4_E&pBfVB&efBzp)HFx?DWe93nB3
zI@~8BD``L^YQsSJcBnY`-75H%>XUG}>UWI_qFhFBN4^Q_hzJA|hO#dLyYfE{DiPR~
zF%XphoLvzBLcX)q0)ORy7px|*EB{4yMSwdI*p&e-sJ{RR_CH`(z94g}?sQ%u&w4KQ
z?4!-xSzofxkn1MHCMe`aBBgKIM>rL_`ZTIO)rOf;>wQUMPo$NWJTw1NaFogVgY-u<
z@;Zfi!tT0T&<%$uJXOn%Ccq6|0>IL5i97?83tgfS%P1no2lK`KRr*njVWujau?_zL
z+r3-0fCutPnS~;;Mdmsnu{58%0WR8J0(HrQDyZ&w9~D#wi#sdlxJ)E3V*oTO$(nKK
zj_%EmGSO=vByTZY;N=KijyXA$pd7$f|7Cf(0_2Ur<OxjPZ%X|XYt4Xj%@`2h+o6E3
z5Rh8)dP!|F?*>g6lp>vgp(>7S5iWy%BK2(`$->l+SdO6tE(g04Uy1FqH`f5tgD)xF
zp$dHgLs=l`B<Ms8cb$6G%=cHvaCEG(*3>K_!w~9IH?v2w?-2Nl6G#ayBPZxz7xODX
z1dV*oC_zjoi0MY=rAOwV1%~L~Nj?x@kOYQE;1LNtqMwDAhnY`*It%J=iZY6I&63@~
zwyt$;aUe-K=j|vYMR={6ThrCF=xS!8*XgI5(-pS4ZH~P}0Yul@a8SKH)NRL=TuJG6
z%SYkwfVMvEY=A}4Q3gOGwu~ge73uO;QQT?WyLk{03@ZYHVZ26wQY@m3dV4QC=1S}b
z0DI(1kMo&kc9%2`)<O$(+~hrYdmgjhTci9>zfgJldCj6l>bCOIFYa2#ioq(5oZEm)
zUDyuf%3mi?_%rI=p&tjS+6vn#<q@TO9hYL;_p~h+X5_FMfy?ulnZspaIGgRT%c$kp
zBzfLF^#F(lBngW}UHXOzcUt2>Cf!o%3i<#m+=tTP<PAj5ZQJV=hLRW@(O=<p+C=cL
z12S|29s^gHpxY1SjBd`*wPTRq!-eaec(<C<DO^z-h-0LQ+x*|sE#ZsuSP41o8GJdd
z%K)<MM{(x>dVKd3o4OI`voV5E`+iaO?ImR-wUgv+^F^7x9K#E?G326JX}K-JbyY*d
zq+1K~LcHlBNT)a49oJsp#{K>s&>*61@Z@fSC$Y;IhX$<MPawawBb=m8+Rokpn#nt*
z0SSDW5~w?BBdW)+S`p~j=MZ~(C6S;k>>6qb@r5z(E|mbae^KVk`;d<fL6^`~M7BJU
zcPBO(2<#i=AlFQcs~nk@yg3Dk0re1o);(?jP`nsL`o^J2@|VtZ{Y^nmq_+?o^c~^_
zYf2+odCGi@J(hH%PXx$R03n0G=5V+oc>u@q#Y*@qG6;5~?15@SWbpDZ^+j+$>i$RL
zrh9lmPEJ=RIm9e6V8_I?zw#mnN0^wlYB}bY`iifa?0Q!vWoFdcHGwYIv@wsnje#tN
zsB10E7cUTlYs;(qQNH2G++Hw~IV(az+okjvpz<K8EmJ1(1{l{GHn<fXl?*If4YW9^
zZvpZduNe^HV|j_fu(1WW@In#C-8x93)`OQAw&_5=rRA}+4d@W_g2ppDkiaG}OxnIk
z+hDKAE(ozUGg@M>^V99hfC-W4K)u>&D4r!i05-O!o<aED$W>EnZemX`Jr^{ZGH4+5
zc0k=1YRCq|izW#N)*KiwrZGQ&0pv53UgLDIr`Yz05x6rFSEdeFVpA8k&rlLa>%uSF
z&DMUD1=P2m=1pVG_2f|3FnbaGq)eDQUyZ<DdDC4ZaB$GcG&X?RFWbPdL*7&sAlD7C
zGmXK(&7m~rXn_r_!t@~c>MOuz4Dl|;>6rCR8-rm;>)Q~6MpApwe0diX`_<14E%cN*
zR8B~MyHfRF4{*7`w&2tzHK_4HABhhd$ml$Sx{lM>a{Z`?%AG#TNo9nu54pZ9@MHq5
zxnK*eAliXG<6E&aK#D6di5Fy836uvixiEi%6#Lxpu$1uw;D($AL;mHBImm^?0a=kT
zYJEe7NAY_r+Fn=1^+l$ENh()Wf~0M9>H@#w+|WQ34a}Ei_qm(<Q7Lsa{z@12EAp84
zz|S7MU`(E_J->Za4DGJDhXuG3K_Sd$zFQXHATW*JJ9anIZwxdfYbfAweZhuWPr5af
zj1y)>AXA&cZM4V26(wiD75i_n?%}{Qn)nrbjw2Q7uzm?1nBorZ&Y?jWcya=$SD}bV
zJ=8ZXw;S7ZU|-76%VD9$(Ezqz`CBOY2tXwJ>$gfG7j0%Ft%Mq9J#hofnL_^~g+jlh
zBg()v!a<#+Qe(?{TF*J!e&j~BJPjKeTaZO*qmyI!%Y}Ug!EwLGx{|B-cY&u>4O@}8
zA}tio#?NFq(hG_u)VGdSrHG_+l#zhsF!U9xLCj$+8wQxeW*~^&AqTve@W5oLb~0E3
zZ~N*f+^Jy7wKRsI??a~y`!?DwfGJ=!9?tFQvURiIo`rKuoE!LDG=E_dDQF1iCrqc!
zfZsv9eZVtFg*f_l*jdKa<OThx>o7YS$#I6eZ`P0(4;#fW$}gC(SMV<iiiDw;4h`0r
zCOy<l2^J{bb@r_(@Pt`&%i-<^V5MIeT)5;iKwH8C_D{YQ+|wKhw{D_l0rzZlGaq-Y
zRl$54zJc!UpSuNyiJlv5ucbHGaVxRUz_4#Q8SEN*P|;6l-feYcD%{H2^T0>3=NPY5
ztDYZ2X$-*wxo(y`u(~ZXzma6dkt7`+l|=X$k<Db60JY*7alj<T4D<j)av){{GXlw!
zIR@mQOxRJwi+{99W1kyaq;=Pl85*qwJTQQ{k@!OR3)=~3v&fQHk;c3Se*4e`WBkRh
zE@RxvF!Wyjb4ugq)bsr)Ab`}WvG1#l+Xk)$oSX0b6&bjpV&~}3!!9BJ33vqH+}u~~
zM3(ADxK-kjTLgA4oDK>ajh{d}NU2j$w@)#-WN2I_fT#D@gvtL+VP^sxcT-^F{?az?
z7T+VWV1_fyvMujSHHtaVvfnbxY04@!(aZGvtA4VbqTMx9edbi*4YkyAA|pXd3@+Ni
z$U$Z(f!42h2srp%heLw-*MY}e^3y)%Jrt}r`Yg<-tKfAGyF|KggCJ>ZEtM}$nhw|2
zQ-5x&JL(<dua-I}SU<!#0592I<xeKuXZL1Q`SYuPx>QrWCn>}l?Ap^7Z#bcqW~{gJ
z#0v9ejrSGK(6jKu_*_aCa7#f#p<@LGxlqRC11pXR0=;#&Ar3+lz88j5l)*DN<vEw3
zMrDWxcAd@QLcYJ#a}qL3y4a5@JH$8UQ0EN4E)<#jtuMvrfd<vz?G*$2#c%nCKJdLb
z+>=&X7do*Y6{`pKUZ3P|=~~R+RXR#wqXjnlFYSkNRSZE`J+YM1abFp^bCl%Q-jNR>
zKhD|Rk9vNQ!m;%ss|;BY^!<L+aatc>8+!wHUYr2z*-1W<3i<-%0zuPocq31U9|Six
zARvaS*bVU~MfY;PB9J_=CH}mY36n;P=>0`xH2<bhKY?i!m_~tV{D+ezFpb|kk>9d2
z(BHMMnZPs(OryXw3QXfKiLd`n;gAB;C@_rz(<m^FzhoN!rT|TWX%v`7foT+&#$Pgx
z0<h(823rc?KtR~{MSMT%(HsFBXjCtu|2jC3pbYafEoe|JdDQW@bTj*#_G}b*jRLPx
z;5GhcuW<_hn$p8p{D(Xt@dF=UUB<gD@&828w(0%CsBnvQtGW>)L%&L$bzz3ftWYu3
z-Z5&S7L!XnM7X<i`cV@A4ppkKEVes3e~am5M|~wi()?sUhh3Uh!U{e?8)5#Wl`vx&
z#0(#P88)TmiAn?jP!C-nV!CI+EVHF?`O_<>FxNY7a@GjgSUAaKki7I;kMvY|B&xZM
zQvNbz=%xGm7Xy%>2=xIu(jnD<@!zO0KX9H)HsDPx<YGCEM9!4x8;0Mtm-|LB_=|Ut
zd{<eO=d|L$vvpz}2WltD>@;wA#V;pDE?r#3skp9m_q}JVW{WW4K@2}cxsL}BQ<PRj
z)osLUq1`@7EGtTL7Zm}d4|dUr*M*aI@H_zm7+@%J?TaWKho1MNih9+!=n>~o8wz;h
zUgvm@Mbw2288^2E=<;xf3PqZG1~*5yMD7M~{R^ZB%R(2y-H!P<eXK411KedxNigPn
zc4PT6_S%@)g95k4W`KEv=oSpHUUn4pqjcrYs!?$e=w9(P17^(zu3}NHHf9TcjHWy%
zZ#0K;hyWlm7T-=9kLxqEy<;}xf2ps81HhdSd3WtHI=P~o`8K!MTtHT-u7~)3AvwrR
zeQjC2LHyh}+(^>par*Rj3o(ZpqmI{tm*f*o$Fk)+d3VbJti{~L8wq-i!kPU{>UsDx
z0A718{gry1zJ0H5d0=%oUOWkMaI@(OA$A$zs7N6EpaRpk){ZwD02MnTbi|F@a{7wL
zlamLe#<4ug;gJ~&awm^r`wXpQiSJuY^f2fH{F+IC#1%-l?nfP#+0>_r!fhHFEh`=x
z0DX4x!$2D(MkjJp6K94U^B(x!gBO@zW7s1*Iz0f+anobCb+6;+o>q(q)ELXgAQ`VL
zksG?&vQNg~+d1}Q=spgELWDxB6GTAlnfe5-+=At=%fim5R))a5Tgx5+D%VuFyIKNg
zIXoDK_-vr>6AFG0=?SGf#H7p457V_j70}j><m&?>=W)_Oz8|HrSwjQ>?)Sj0s)>!X
zc~E1Fb2X9?5{z8W)s}S}{4H2AF^<LyiSU4EpdAa@g(pg)8CG;627A(oooz%8Ruknm
zYw+aXdK|`J3h<c!Z(h)3bzi7an;Fb^PsH$NmhnCqcjwd>^v%?;08?2>w;%O`8mw}C
za3-OQKSD_nyz2!ba0T%Z++t77DC<SjTY;%iNvE;oTd|x4p!bTvP=2Fn2@K`Po;(6W
z`S&uEe=&5Xz)%jd$A?`7fkzq9(T@U205=NaA%@#DJTd~0Qs7Z85BqHlpTB@mq(}iV
zfk0P|Kt33;O*Y^bYV1H3`?PJzW>@xMPgrHIUUcVNr%Ls_&QL9Lz+t#$4R`MVIvgS(
zrg-b9k4!V`CC>-a93Vq;zna?g2!($NXS`RXaO6O|;R$pa8aG09pQW(;*&tv^uZFv1
z@Oz51G0_;tS21Pe3Sdms(kXp22<-Ehz^DH31ZJG0@w~CuyFt$_y9{&{C$E9;P2fq}
z)n2R1j%PfMhFKnwcV53y+n)OR(v&!&`lck=TR9srKKocn9a{l2-7vF&J2R2QpKps5
z9}2jIu&0q4V%HuGpd@w_e*@fW0Tw;lA1)Nh80;k|!J>9FgRrZbs{>a+Iv~cUsxg(u
zrVdaz{lUn`%qgQE)rZkg2eFnS#9wgZEp<%h0OdAy6bs&_eM}3?!MGbUj?jmoIVG5O
zOG{WVPz=(d^`o>ek!66At&LsZG9t*t@UH*@KWCyEgFY1Tk0bQK9iy@|toC61FEpog
z3kSL|23uL4{V1ml3g<u%({c1$AdTWUn0Rv0O)uzEB!kHN*%v^zd}KtuLw$$?jLE>1
zT9`YdmB`sAgK~)&qw>+-S};myX=qxDj7bGxPk=PWC<f?Y`qE94J0H&d;EYDdj{&pZ
z+9sF08qPIgCh@)1l=&+HUt6-)f^5fK28Qou0Aw4AsX@5I;Pw(5?6@;dVmW){Q0L1A
ze+{04&w;y6hED89!F$1{^|dorQ4ZlTNYOcLJf*{z5sI_}IUPn|KkAf0OtvKbalyOQ
z*!T9=#8Z+^Wv^Zn&@IG~)WO@q8F}3>M-JO(3h}&kc^VWDwm#U`kGcUK=UWFM&eq>^
zj(H^)!dHgq=D=6x($JD=*oqUMYwF>Iew6i{d*m&ra)MdT@u9arWue?`2G{Q|9sjdg
zi@yNqRphf1Pdud)2j&;a)(|xFO(Ny(;Da6FLa<xsC2SA<2~fE>*VAV`?1kCjGodb3
z9+0w*6Ey(__+}Z?z%7P56HJQ|rV5g&1k^LOnHp!Q3i2D50kLqc4Tf`2NstI0sdger
z1phIKU@IP2fI+J4Ss-a_4z!Glo4++C^u>0aAWneV+cUA8LnC)Ow<I7u)dwyKEP*;O
zk#I{8cD8Xf%rGSOZBSAHnnK>l6;&bb)?YHYga7xRHNAfUP_4l9ey3wP>^li+aDp1#
zu*TGXX$|i0YkL0zpk0CK6_{QD{NaBU{_roGUXcHr*qsytS8d6vH+v(qj!(Sy=GL3$
zuZI=u+PDN{DtkM-)kSxrl@tLf2}4&acZgvTHz4lY1Tb;XGc~xcw;z=g%(Lmc(2u(4
ziv>bb_o#dXW`-lPm0;w86Sf!l)Y<M(1ntki@^O2#Vzm5Jhdc8xX}{ILOYDD%y?8cP
zJso{1{>RUHn2ZqX6UPGel70*{++$WYlzbb^JPjNrs+|q_`wGh7m@{faFHvV~dkN)t
zcG{)l6NY8b2Sw?B%dy9Q6SA2_>B)t<XBjJmjs)L*Lp})oPxk#bG0{miHt}1%Q+>&o
zw;wjH(2QR9LKWR73dBIuDBW|kG51WodE&>?T-^Bio;b}^dt1fK11G1c#Q8MyvABJP
z?DO*OaLc_KkX_G$*`M&c*#u;p%pQac()5M-M+-&TBX}A?X;AmhPMC3&!aD`>@84gc
z1CsQyL`Ih(D3KLydJS^-K47Z8g6bMWg+PYR$rWVxCHhefj(w6kDEw7?4S96}+oKNS
zX}<1RysZ|^q{Qw^;3wL9tBd}p&fQG=pm_#+2k6ApL3(eg3d<QJK_^kaCXu{H$iUiF
zcq7}0SBR`dHdxNuD)I_aBIPKiPYS;buJDS1Tk=k_ACC|)^=5U0<mBzzPVSs;5`TsZ
z+l5@me+@N0cHV?!@)bdnDb!s{c1VwOw5BcTiVdAq3FPOKSm2jE!}|g)+gK~%`XL<Y
zXjR+TZLd1LL2^~g8E9hfhdCRT3%3|n2O&G}<X}20;MPg6ad)r>$Kn8obfvjY8N@Dt
zU6g<;I?+gAH8pb|tJ525!XOGD$yI~l%O#O6M-KM85pGfbMJd}@;;-~juoBK4TtMRh
z{Cb#^+$f+E#E<s1G`?3Nk-r>_*21CDN`z~PlrDP{4F9~K{R}7?73X|Qc>rMFpVkLU
z>wO8Ro~5fPbC-ie>Env%hK*wgFT)Dpt(_Hl$WHq-xTi-Q3DP$=V0t1usQ#3W4Nu_K
zA8)inalyR1&F(;Qtj&(U>}Kkn1`*y3k~Cal4@z*wx0eflh#XXSrenU#?0hS0Fl~?@
zOKNrbZq8tx=!2;;Ec6SSP=Kj&sFTtm^aNi1vs(Jp0RZX;E|Mk{+!Zi_Hw5OFFb865
z4j6$tBH!1=5(#?bl$ef_1xtB8qX5ZztpY@mi&Mtm3>055h@5C*R6ok3-e&kUs=j)l
z2iM(invO&1;9n8y9(kTnyOa~plMc5B!Yzwq_i2vUQkD;_N8$fT%(khN2+xZSwg4<=
z>Qk4BLZGK<2O#)h2QWavtI10MgxMREZb%^a-mP*VF%gKL&vP{7Z~)$PtD5g{CjJb*
z6z<vzSrWJ?GkBBBfKoI+5=_lY_yP<I^$Ix$&|yPc{&*>RH9={15_bgvooDm3z*duh
zDNiO4C|za)kDwy=DcpKIu>|=gd^hU6++f{pw9zInB#jX=FBHif^t>d-3Y3!-NTaMf
z$lW?n+OcgXHwOOPoB?I}qJ`$ARPYI7qJPStz564-sE73KB%G`B70b~>d=JBoxfBMl
zN`}wxi5p>~G0l2lYZsJev1^9Yi~qUOtXudfYyC~CT?x(&!A$|noiqz>wJ#LOD*Xqm
zWdIXD2lN5C*UNWTQ*M&MZ30s4=@8tQOoJ<gcOyms7m+Z60RsaUPK45U2?s27oZUdH
zS%+9wGUfuuqNgB@a>;+9P8y>oI8@j5_!7!}(ho>R8QlC4ImWXRUglFi{MYcSUwFSb
zNVoL9|D(VOQSO0f*9$YB4(|*9xxJu_&l~iU29Hs}P<ssA4x~Pd?Dn@kD&VwZIW2Iv
z>3_7lgmFxFUV9|3+FAecnQH-M+sG?Hz17{u<%2il*qdG{Ies>AWk~CO*YoDfEV!x>
z6Yo#-3%D-StIJvz-H+0u_Ys)k+&Q2WPdvxVVN<_?O<PnnXJD^53A3dE%tca{w^A81
z5Kw>}r1mG!{ssmA2vvlTc-*AVP&T2%z!-6>{)%b9zV97<CWi4W6$bmvx(Lt)Egoh%
z2=P|~XwFH>*DthNARaC$>W<k$+kMIU{&L?G`G7Xv7<dgl>Kex$%MRTJWyv;5LhpqR
zHG{L9u5v70hG}5G^4g0}T1}Vs-SWI3w7^kZh}j8Bq>V?)p!PPHKfPSZ*7^HzXs|Jd
z2xuaM;ax4LY(UZ^u^(!_flE$8eT#_kIehn3lzXp7;sga1>Je>;hTR1Luv7pn{j;e`
z|8lM_3*_qfAm(Ux15_C?{XX4q9i#?&RjqQOu~4Oy>kdK;`k>PVju<DTOm-=Au5TU2
z<EmBFqq2)<o#*U0VS2ts?nGBw?i6i#31%}-o(#B!_X!}RN=v4&UvZ*YDZtA%tp$yz
zOpFm<38n$ck;n*THzSucf%`P!^Uoc|F81a?-D}#gY{>z$dgz|GP&xd@wSmZ)*Q3l6
zz4NVM*&G^xiWTNOfol%|OM4bb_P2sYYI7no@r%k+&omrRySKXyS>J8&E@@6`R7`W&
z+Zo$MBsi|v8xX_5t`CC)^81G&e(ViS%+OMu^dIae`o}X2tR7!`vfX#FY21F4XU@XH
z$&M0F_`XqaK??ple6)=SHkL;yiy{+Z#j6OgLAC0#R)K9xi_Y0K*=+bsu{ryot}h?m
zpltWGEir~2W_!RE)*a531c4Iq9B8(*%BK{`DubFoQFcIaj)r}`cE$hv%#F*UW@vq&
z0T%1vod^Z%oE*Fx%s=i#d%nZ|h{3xt8alHXBd)d78LsU|Ww`0qt6xs&>)B*8wf17h
zskT6j=9>wkhfnV`z=8RTmSXGmqqdlXwm)zkm{8FbkS$mLj$x&1rj7l4zr-(hx54do
ziVSmi=r}-cTnNM|T$*ZIlI}vzWv28N!ssRJC<;%*rcRze)g=e43XtXUA?3I%7iKL%
zWYri!$i>&k5O5oy`JK@?U&pNRLGhn4bm`5Eq-;dBP(#wPPHcaT8@wyAq~WICgXIKA
zh0LdE>rRUvZ@9aUa2ZN6egnRwW<3>$cB@m`-PW)rHROZd(U64!g>k#z&5!pCP{Ce@
z0RWH}OaOhRSZiAVs`H@4Nu&hRbsC-pb!-BzLwJb$05G?%0Xr3H(JzNOd7y<9hQTCV
zBzDWSV|o#gR-?$YgrO@4Y-j2lL(c3T0?0}LJ}2nG`rCf2@f_|EoMDmHY9x<5h1u-g
zAz$(n3Ac6aO3CS@#kY=zuGKpfu%pqq$<{5Wfz-E}c!|bR!S<u*`6%3mLHr8#5dICZ
z%h4Qo-Y<iZJFzy6E}RCIH~_~)F?9Kvl<rS}Y`psW^lUlxdv|3xxX$8&2ar3R9}0B{
zIr8Y=k}a(8f!qmK1Fo3R-d9b{NEpm)2#phVQ>$okXr4bb2kO{TPU*Ht#O)b22W}Z$
z7dZDvh7<n^#~AU}#A%L&O>1M&gNYU6YHU1c>1js?dEmo*NkEN0M9oYbi=0*!Vk!+<
zQb!ti0F5{Xjeqeea3#+gFBy%SzAn@)Ke{6#=_zG&Ke6F48K=?g*i7pX!dIwUvIF>>
z^3n`0@}>|r`K$#(^4xQ5ke5BS46L9>fmFj^j@1){8m|wAPGkCou`i72oT!bNIW%Hb
zl<(UivZKa!6$7DJaX%^seiGTH2bWVyOC5wx4brErl?EBrloh~w$w}tJt@0ctkd_Qu
z22%9Oe64p5Mmhk3e@}j#euyA%Gp4WDh}lBwu<u97LQ-(Y=5nZ8;v2=gq8ZV^tAaFz
z=To>v5i7B4&{Y;12oklyvrI98D@1l8hCgqhyP&>R<wTZH2_<Mm*eJmwqycce8oCP2
zr_BJ3ERRvM%Er7mGA13xnl`r^&8&71Y8%~bmJ|W~MC#i>l7&GtNRFWdE@x<JW5&Xw
z-EM<vkbB`ImbELmA7v0{ob52QaJqXtEWjMuPz!N5VE5H~L|Vp92g7ZXO=B%6rF1{u
z(L?;DbnA7Y=DkBxgpft*Q!(_w^z_s#^H=@c9x@Sd3#=JS5OOLoVUG&5F_J@51OU)S
zcp;8IGt|HLRkSfPigd(>?v|BAcj`wiPD&$kXC{#Na~11DxTa$r5AlE85&jB3O>1FE
zDp;_h*c_94slmIfWM#QOUi(4Y$!DFljIqY?4%8fTkT@K(u{04X6%jpR&Or0xFHrZ?
z=(;u(AJ{5ye;nx5d0+VekaGY`Gv;16Bbz)ME)0GO)7UUOis$se9XbmA9$Sr#kP<)F
zVqs&kV!@R;?mM^FB$FBu3LaU5RE17ti!>x7n3FX!rH-P}Ha6DN%R6WKJ-ybXU6a;0
zBSENJ+O47IF*UsznWo0T-385ErGe3*RX}V|$7+B?0Or`B9PaGEfXLOYP6r2KS@e81
z1oqNXVDo_w|LmN3`|-ek+*dyce%!NxNVV5Xf&ue1EWysqpZcnR&-Vll8tIYNWSld@
zjxHR=_B3XY3FjZ)n{4WQ*X-3A`!g>uq)crQZ&np*AG93eMmu#Z4_{vU{<KJP*yLaH
zN(McKf9k)SZ7_VDuvsp1o0>d|Xc93r?+3XgzeNh_kESviOE=Iig-&M^V2h~s-d9cc
zvTj6~c$GMAc=|xWFF!eO8S2FWUe4Xa=VIUapqLQb5_1XDJ&$RLpud1->EH?7#m|Ea
zk7W_r5OAXF+p7U5YfeAv(i!#<&_jhV7p_G*zj?;<bLzbLlQqEe@#IpFRj;3q_i=O$
zHAX~QsAQ2Ee{weH@;GO6{qosUYnu0ko<Fwtljp<$?gP79O~FAqkDSDJiCh8R&Uzlm
zNQCA1LJSxzudJee)bs{M2I46T(kSVC8gSIjD|JxLM4k#E1I?N445o}oFU&K;Fg`%j
zp_VoE1w70ruoxWa*T?<9k|35n`)EL{8bKoPp{!&3bI>BjgPLfhy#O76pT!o*sgX-&
z?u?0azEu9XWm}BQ&6qRRsIJ0=)#9fl&Q9LB@oq>X{Kt;bxI@0~C0mXGcwBYmj~EnV
zTgyE~j;4{ZO<ePErOf*6FK$d->37h0`HZ3*tVtJ|eWbI}8}X2eKtx`<0egeXL3hY;
z*BaWKad@@=@m$&|-D;ZrnH&j2WS%PvJjSqXU>Eb~#2!5Wzg@DM1B~H)sf;l64e)i|
zloJ@v#FIp}Y&V>N#=nQUr<6eq&Gt4>j%hX&MSu~CCtk$qKV-*mD^6Tms%rXD*|xIt
zV5x!a;=K6^&I{N21@x$cYSnSLl9r9ED8n5gS-V^h35d5>D$9^-Uwqr*==&`JsN7cU
z7EtR}1W$bUTY$rPNtS;LZkZJY*Sb>6z?66g$SfJOcr<^4njsJE0*$!Ao*o@>p)Lw<
zb#Djok+uEz^rM7pUQ6%h-_;U1NtxbFxJY$z>|-ET&MWmK9BsKWvAFZ&kN39sRr|0M
zmJ2t*BS#H0@=G|NB-e-ieynuA{~B`S19vUs?{C+hQO!Ec7G*?dzj7j^E_vKGv#kny
zuQ4^rRspRg@73k)*|o_Htw!(OC6d<*$3Y$Wu8wMroSl!4->ce|_0=Oez!0+sB>sNw
zClW^9__xPR0*geG3)W5H5)d$@^@e|c@7TRY+OnG(lq)l8odS63b=YZa&z+ROjP!Dn
zbwrezxP_0w?HL)t(85M4u+}4<$G4~c{~sve_7`-NF_i1vSR}VP&2`B9#Qb#)ddX`)
zsfJa35?_m&u{tv>!%-slB>W8OD0sZmcGc}C(Cw>h@O<(E@^wQp#mJ243zt>9nR0fC
zWMW4?t6!ZEzdl-|ZS33J^*Awhd~}6P=;{*E=T{{^EiAde#!OY)wEE`qDKa|`HTIH@
zj(Oq(wrS+BlcIpF{Ix~&{cyU-gGPBLXda={8Noh0i3iq6t&g4GDLOk&a((RNkZ5Ve
zTGO=pP_v9p<-h#VzdEIj6DSoaESq9AMFW#Oz0&gfcUH}QcbXS!{51dUrY^<SsO7>L
z<RwTRO-O46yJR=B(^oZri>>M2HS?<!UtB>4&EE2Ib6Vdncw_w^@DRqEvKy?q>bU98
z*+*O%K4R;5&My5ZBeUHH@=N9$ynC>1N>HTqt~2J2|Nn<z8ppp#IicTvJ<&jM&wgB&
z9c{sMYDr>ZU1y&t@IQYp-Qx0VNtG%rSB8kMDrLU2lYZ&3(uDKn^`9Q5TF>0*AZIwy
zb)WK_pY5Y>s(!RH=|FL!Jf%Fg`Sr`=uaEChd+)#GYPh@Tnlw6gTj=A!(dOfim`tnU
zxavGHs@4K4jHq&VJF5_<rkRx$qExcdS1vF9t;x=(Q|;YdjN|?-hF7!>?x2E2j`mYZ
zJyZK>9U`;4e8aJAA@s-E@r$;-yDOYdIsul@0lNyAb@b?gR;kB^5EJPpj$F+HmX~ME
zk3**M)Y$4z*yh!B`H%9Sx;#`jJiW?N<)&50>O1x&mo_{lR&eV?GH8BU%2WsV^35d4
z{WsSsYAmigCOUO`RPs{CdGbCiEPHt{?-Fhn5X$PMaTcfne{mv+90NV;VeTXnLfFB3
z40L(0YJhV|Y<VGq{0L%>iIi@x*6Z)=`Zy1UR7UvvkV#vB|9}xDFS3@EEYi@6z~<W2
zNlz+yI?-v{+7~V2s7AL2uLU7riND8M|N6e@#};APuq7Ea(9Enl|8<BQbKPyNYv(_k
zCESTQb$Y=neEWja8IR6){_3_h&d&=&-=WlbP)@)dG&<O&^iQZH(K|z5KK<U5Qu<MA
zBL*6UTE;FOJ(;QDxLpI=FwUI0*qM~=;>e>it97iTq#t|XBJC4wW%@LzaUi?<$8Y~4
z8N_(QG?J*Zd=`Z6zNfy+U`bYeAvWABev@0k)}^Kkx^5k{4@m3)oAAHtT73WW2i&aD
z`)-V_Fl#;ku7-MSYt7;k)fhcH)s$PO!jy3D9~@Y){i0{-Dd8or-Q<A#3W3Cjg`GQQ
zHtxt1c0(rTMEI&F(c7QzEUHnpI%u@(h_0}g1aP(f4STV<a%Fh!kI?iJ?Cl|^?)Rgn
zbaTSWrJ%&5+pM@Lo=&%^r?N20KYlfb|NnM%%nrmLcmv5iN*ekS*k(2C>)cd~eD@Zc
zJ@go`p(;xjy$eN8w?Uip{lT8tkDApULb&XxskqxFfA?pbCqA^bPXe7*H3}6T4^?d4
z2s}6s?hNGp4-MQUdz%qrO$-+Q29B5WFR3EmcoM&CgJfsGHk;O-(TyB;iR4g5nw($V
zyIzYN;)jqYIeVwcO0jL8SikbLB8$5wuQWPUd3^utt9h%M7rZ#Y488I1nNq+x!2;)G
z277D0p2oh~Pa6^GK3j`~6E|DlllNBCtP531`~2GH_xP*hF8p8fUb1nHTs593`-(+f
zG<tSxQdd@;!CJ9pnYs=|HxiTPX=l29&e^YNQvm;)oY-0zII(sapO$1vVZ2q_*>s%Q
z;VMUVVCzGrQ*|Cw3_q<u_Lc;SLI2+{;@8(Y$%NV45zWwT2=@7sYFFPb-KwS@FwZRf
zmDTm9`iZBbuNB&x-y5b;0&tKnONM{TBZsx^Ne&-+EkDXKNBs=O{vB~gGER=I#~{@{
zz7uV2J}EBsf|zw(V6EZl6LR<8{y4X_*Z*AUuF14}Vg0CrR9bIg_=iHcxyFv$vR_VI
zD5=>B`r4_^?_T~j0n^ue2jNIu(LvevRiSy@{C<?%G@O%(`0l0AcB^(>&^l7OUrb!+
zv(|DDntsh9FQhP+-G9QqDSB_qo|RW@o_<W95LCHDcUpMGe)Q<*+|@!{s2&~_qC3E&
zqkZZN5UI!m%QGtE*B$TkZq(U{S!qpS<nR_F7~-yE_m7FWK}IRAxk($n7V0fhO7ie1
zv~M>_F#272nZE%2e<*1Tsv^QcoupD@%X(VRIof{YMz%Z+8yQ=WMQNjxV?ce)XAn~J
zuO4GaX!xx;nk|W3u$lWJRkI(p$Qc@dK88&{R$9D!z|o!d3(~;QDya4Wwrf5p%DB?t
zmeb{E-kfS=2O4C8I_6S1nu@N-G&%5@#ECHVbuZ#A2C{?cnv}*cHW5j;0p+OTf0mIi
zC<l;o%6{m>)p353a8f_&_8vqSBr~Y9%qYBfM4-MA`qF0zG(HI7{;kKef7M><zedGw
z&n9c~uti$JJT(eCx_*<Q|5n2yi=^4<*6|<UgttmmyIb#4Ro(AiO9T4Du9<L`L{JOm
z-LvQ*whB{4kN;3(k|h)S=0$vK$hq*Pg%2fDet1{@;pkD^^Y#nIOG0n75g9%;z;J;S
zH<l4L!`!D0s*?czd1xF$sng{xPNIA*hp4>e!od$-5!gC*aPw1`#5MwMO-v`RhtSiH
znucm>i48|)nbP{&G3|&r{6^8WAEkMW55IBW;G!4_NI&~~5tkfJKgtUu__BXKj1eGl
zxX@p}RT8;qGb6>-cJdcd#pvBF2HEcygs!{G*@qq){#_!<I*ud9a>Vcp<rSHYcy1#{
zA9c3d_`4@J;!$facr8&n;?WI1Yr&U+>Hpm@+OO!PwkgIl?KmlR(eRGER}6lce**RS
zcE6h~#WPg`XC7}Loa`g9-1tv%^SP!qHYQFN65O`x!eY7;<7Kpen0TS(RivR@W8-<t
z4LM*xZWcVu@5cAfou)nCrJ>A6-U+tEK{Q3W;>*3F3zqJNr*$8p&d5J`82gaDzZYsB
zFmOS`X77J}-8BN@oj$TvYYKo;yLR$tNP)!M1y?wCGZW%ppaKxYBR3+~5XqHQX4Bz1
z?rTw#XC=Y9K6MKYUD2~wStNaKMQcK{+CA=CBnQ8iEnRPhJw#Q@6WgY_t<Y4#bm~^c
zsS;gIVe`^RADV;e&7FEe4bO60c`u+BjL>DrjTlcFmBq=4xpKMI=3nA(vXU}ln@j(>
zwZcBYWP1O%*gzPuFl3{Vf<rpaixDb*svUvG$PM3C3e_$7nx9l1hptExwR>TyWqI45
zsQi&z^O0H#3--Z(z7PJX?!;%~PISyq&YsAYub{941F~(7m)&NO{BNF3$`dO|C+c%E
z#RuLw6GPL`h9^<=RsvsVr04NxyNoGzQ~Oa-$sAHo6f|NHk#13}FjWQV#OFFJ|HsZ$
z-BPtH8hQ&#wm2|n&GJ1OmCkD>1f?FEABJ}NmIi2v;caqqvXdy@?r~z@=^1B*r+>{d
z5ID7ern|mLYy=%qLG(a~=z+t>f}|7zHy)q9w|f8uB-B5SSPIDtkTM=(yDSlvepGA~
z7P-5S^h!b6*VQ%BPts^*(TlaQ_pdixyXqW<z5vd=4VbOgS^#29`eX<iMTC+$k+$?0
z`r?@;nOCznDlFeOpS56b=^*Cl&pZm<hy|K|b9RASI}djYV=se3ye_S8zLq>peH;uT
z?^TMus_xs+3(Gc{Z8D5o`BTh;B<0@JguNd&T263ZH3+QwlaE5|O&H#z3V@t5pme{p
z>Gc6X2TSXI)Jd*W<x#FKR{dz_hc&%%p?CMa``xkrZ_Npecb+061rZe*(96v*<a7Z{
z+i?kQ_}(uGbS?Ii*Dri(BR;KNOjc>Vraj9F%#5ke!5o?e@KTEttb`X#`$mc#B&z?b
zN8!<m46&|Rp$dgKbX|;o7P36WR7P>T$Cp*h@IIH<p5Ay<Q$ly?Rgq<{q<B^A@W>jZ
z0wY~gD}Tw-k!@3Zu(&<T_*tCjc_(R^o7)>Ar%&tQrqhT@YlZUhxs*gqW&r_{(CMSf
zKlNE(dFl(o8P^DFu>~g=>V%dZKIfKFStY+R=gjUyH>A%s^e9kPk*6OxXbm%VW3Ie9
zac6=_m5KAY^+$Zvrd`WfU~)7{Az(rz3bk<>UngwJsZ+Kc^hc>F*N$A6XuZ^F)4o~T
z4qnkfi7cgl#(3l9@uk?rg3F0?(^~J#N@Q8wrlM0OAF8@+j)#8c=m<S`3V&w%aLqKM
zU8{GSd8o0lbSV0e6)~QYX||>8{aOFnEVU!0^TXF1cjbMQ>Y`PM?no+uEU+SZ)4T%t
zJ)&2(Bl7C;3*`}%4>S!`i=V+e%`Rea8}>r+-gIR-a&)B48ntY1lBt4@OmS3P*T==%
zBF<&++7az7z5Q@o83&CGrZILs-iRTPC$Wq3dYxXKOImI$v;AO<+%DHziS?Wi{^qa=
z|L6_&ryMCuk(Ed#c@-5tiJrt;&o+uuBVJT;<SHrdO7>B8DE%DRCZ4Qwt&HL%^&<V*
z1wEy}X|uY6>9$PQ%$xZSkq=-$GX22zYUoFO6!E{j@my%%?2jqaugh(<Jrx!8W%WM$
zF5?{dVyHBG5fjZi$abys`Xm<cBIe8nGm@i2Z`JvmPQCVp$LYJoPftSOa{#nbs4thh
zgJ;eU!?OGtNoj_ay{3u9q(r3^q*GFDQ{=qgzwg2`n=YNuUhM539+<Y@iMI%1>>+S4
zJoSDQ-WEVzvF4J|?7Vt<4!waXoQavjlhW8nGC6opA?s8{hUJN84`=DwE5`YWzdySi
zyGC8IHpokh5{rOo<J7^AvgQQSm(<#B?0h42u|^+jmD%Z4ef557>zsuXwrr2YilEcO
z(8aJ=R!m0Sr&XBQ?}DW=&plJ>xaL0V%MXWlyh9l!khiBTI47fc^0|YIy&hBR%8`1_
z$*sx`pLS09VjJPtwA5$cbU&@I^28w&^&mL(-*uGuN}^+#&;d5Ykjz9jFh0GjYF`&^
zCwATM%CgE>`GA!jSJ%EMJMq&@U5V~}cwlnU4b|@Su=VQ*Q(77ND(xHJ<t@B2KR2LU
zUV6RA&GUO?AMTnga<QAnUV-()DTT9>b~4L7Cp}KGoYb+V!0xA@mBzEK%-sGU-QD)-
z+O?;hNzW*jG19E7MICAn`4195Z94So=7F7|RW92utbdafg2}ruL$BJ;lWm-L7Pke&
za~CSCcCmdkFw-3jHfcZe(8>?jm#<P2vXK(=c=Itkf1#0oRAgU6UV%h+L`Mc!-wKo9
z<h81^E`Fo>Wru@vWej&Z)f_z`^3%1u6ONsoH{E^fiamw1g}OiCytoJOMH^}9F`Kj*
zaK>vb&0{%;TJqeGeW{*jb#Lxn`>N@z{>G<=uRgl6^y)<n$)*bc9~Xg0I|!L(!RXyV
z*0Olv8?JiUJIo>J^~?R7P6hi8&E*rY;w>;^7sg9=eo%cPmEh1hPeuVsY;%#S`XSP`
zLHf~wcNz0-eWu#23_mQB5+KAS)Ou&dNH%z9XiC>_^PZJR*u8Czsfvh=>*ZSy<JQdk
zFb@|YeD-R^{4AGynKM6amWz2eb@A;T8=t+kIC5c1_hQOgxb{}ko+|c|dJ!-En@Poq
zg+-nfc{BYf&{Duk`ZS33{}YZg%?_no*r+v?T(0ww(mhtRZ<`*Zt=J`6ht9-Q_p&Y(
z_CPYc^~_5_5mskTs#NJn^u#<pHBCX!E9>r~8~b(4vYv0(y1|j>>a#C315OG`!!5c}
z@%7u{Rbk$FlWyrdgl?HB0x51s*J1Bqo+HWJV+VLMixQMQY9<!b9rP_<-R{vmx@&2v
zeBg<rNBkf8o>ZvUxN&$#;DQD2LM+L2yXi=El_jC$3$l)#w5=?4S=^8ENzVf$Za-Z9
z!w=%|7i+bq_T|VXvJ!D}TrH%WEYEbxv?=hw%4(=%JbCa*nfnKPl@e>RYK$amwk*gA
z-V)oo_`>D|!o9Yr^PO2F$s~6zN#~^OqSbl{nhITC7Hcoe7hhTACp+VikY_tK1UoGe
zAWx`S{U{-8cot5j!GA7Izby0dq}0SjAKSPcjKFs<R-sm2B;2{?Swp_}jU3!-qTG>D
zuc*J&>5fV0hfML^S^FLzJT5QecV@MG4cVT(uO6<BDydWUvc*e)ZLOO!!-`{(aIUB=
z(9<>YC&SRAF&dlZ?$3+5u|;^g%JkVnJ;0uY;mbL?mlIb|tn6mDn$!faQf!wz3s~xq
z{8jwR(VNd?Qq*TegnnvopT=%rVfj%7+12!8X<8eK;5rGQD3r>7k@jWp4I|ce=1Stm
zc{HJ-D~ie6PfTp+aQWSi(<nI);rGbegU(-fvrX%5H`V!t8vC4j`p_r6_o;<NP}4`5
zxBCu1yqa@Aos^IDq@ZCtqDSZ~SuuvZr^dRQB-LOQgEJbt4;?aC)YNpwA>}gIMOKHo
zhVah;$opiCS<(3NB7};pBQw7EhUYQ03|5QnE;q);>Ds&~`0UqLdn(NvF_DeQiTD!W
z&dCg!dq@7V(c&_bsyXMb@v~ko*C<RSUe`L4+wdC4tb-~ukXdlmu`D>jn70%H-sHv5
zG*FN@;TB|h&Fh6UQq;T-K&M%)P0BUxIhx9J3C2-w10|!V))vYPblz`tUW{73_~M6(
z%d1z!SzN43a{N?%KTM25AWNn&Bhr;+oUw*ml|J7OH_tic8U10=i7RmJP82B=s=ZgE
z@C@X(PH4{Ao^->!IU~uq+;|H5eT-bj$H%(&^@2sn==-D?o^ah2+?-nM4Ci&O<)ux9
z4%M3`r&uhfirGxdv`{$sE?DuUf4*OXd<@RAo>-A&R;1w+#oNNai!A)eQ-4LB>e@cZ
zlRo$C+xG{aU#MN&7&tTBYiWl<Dhj=xpH!*g#O$p#d}AP&+Wz&48oku!tLpx)0@Uo8
zn<id5p_PY?lwg9y1h#T*MxC*5585Ym15>GhpyBpJamh(1>$t<)q!0Yyb<b1c&0M87
z{!!wy(G^1sia#To5$Y4Y&s)xarGaL`+0a}K&AQi;2g-5!s+W}=6>nJtDrz_9`OJ*Z
zX?mQcy=~*-fYtW(*o0FtTvIdNezvZ9mcOKKJaehkEiD<J{L&0HYL#t3<=cB{*wsI5
zu&51nCDgd|`8IfW)B#e*oqG8UI}H40D<5Ecrm=Wfw>up~EA9#Tq&t?kn{}Ju$m)Bf
z-uQsk+MB?dS0~O_PMs~!ZaroZ!6wqzdP;bRp)0d#E1QmAJ*Cn7nWspN7RfSEWCvwP
z<#-=WUS}9}@T1AoI;VMl%;zH0rZvXe`}s9Qk|g;d_(^S;_+Ae?o^&Mp0d{Vl>l1za
zswR(^%DLN*_`aJ~vUB_7#g!kmSZlJk3nO!pV)M!DIHoADPty}@^!e(C7{^HxrN*t(
z@lH1sw=U6Ma&ybufN;4BjP@M0H~3GSP>%?F9(fJH{CIv_^=7MpZOyOLXQdudDGRo?
zC)DlNR8((J?{KwfD8S?}J?jWQUl3X6Cha-q2Xj7ewNE_vd3#3o*5`Exz9_s~{Wd40
zLAA4jm>H@D;A-{QD&u;C=1fyJIp6e0Q?zm|C}phVx9j^ZJeZp{$)|B9`V^m_nvdPi
zzfHy*wDHMv<lp=3UK>5-#EqtlhdDpw5l(XU-xfE@klH$JN`v2|eMatKr%}RT;+)@Y
zp~ef@e(62M<^q3A0>Ohk(WYijA<xzlZ=6M)TqF8+_7a;@Mh$vWbv1h}_Rfx$iK4A?
zTZ7$9CE!l6F|{xubXrilxw=>8j(QuL$Z0WIPur%<-u8ap*{Rc~Eb?fyoQQHoE%Xc6
zOnOJFpmlJX>5j~TEb{C{yj5KV^8y#$Iv#41-_n+GK|1e7u*@B`YF+LEe4+PNP{M>E
zx^N`?W?|g<3373e^=gN%ABk3<;_dPXt7F$!2Pk$JdP0*@ff-G|%2rIf86)x1Ze9I(
zQ(W(!8`Fy4eVmYqLZK7_P!rrWPf(P}>bDRX;1q5li#zwD_8yrWR4*zV(O2M3v~qlz
zph#r>$Owe1ND=r2$cK$Ycd{0n(qNI}ccRP|XCzjt_fRulSLXi9(D|M#A3KRX`@#5f
z@TU0e)vj0q7Eq|UTbxypS<pPibfx&)Pwft;g%-P|IK?_!Eb-o&G2P+tQ<UbOO!C4K
z4f#5{VU|?s15vSMRW`|=E(;y?5{=7cN%XY?yym+?$0=H~*mtsBE}nqcdTPn$EB!<Q
zYSebOxf0gg!$(=$0euUsU(M=gFBniJF~B7tsn&Nt!7;?7&uLDM?6l~VcMS^K3dg@{
zFWF>|g`L&Y5gCSvS3<#L=biaKsroOLxKJaJTNRzVH1y!D!1p;fkLooRiWm&?O#g(V
z$Zg(XcEkhRTsHg!r^}SP(S+Q7H1CI^ilt|FPAp!zBmH=wTAJnjzLyKiNf|Jv1rkpc
zeZbqy$Xiq5nCJNN%JJ&Y+YHeQJoJ}4SBM{;D&)~v-g_AbJz&Rgp)jbIV`lfGek`qX
zb-lH8zTV=iN3J6>ISacskP7(ftQG7SM!9RgN-dP_zcoX4KIQsb2kspshr^$GZ)04Z
z*g0;ma9^A0S12ON-Qq^fgc!RFFG?AEO$j<!{2(!Cx7DT30aHHTtbKT^1?Qh%z<3o}
zPubhvL~_SGr+AS?+N6SgI}=TblI-e(HOgu!yeqe5JlAgXA|5^}boAhD!v1U&)|VI`
z5W$2RBt6IqCeGQmGid20Ci=@lZsseP``X+;yYJzg>^aWT8~hm?njfLgVKI0uwkgYr
zaWBFz^~mG4jsE`hs!C-R?#%XF#qhT&yvn0O+qX5&DVRq(QmkeRnHT?*n7_<1Ij(e3
zK>gHRO~-xq&G19tpD5LYEX8eL5VHuAPnrueV<M_mV+_{41fj*NyAQ7=Z8DoTA%YsY
zWFc?ULnNCh!<&D&MwES?*=|&flg-LJ*J=6rkm(OOUvw7UK0^(ff7x>B@~|1046G#+
z%@gl5Eyc<Ha5H4vt@dcWDIYKGeZGDEfdHX{CaOR6qh?<#2Se0MtK_oIFrbc}utOEj
zj*m(c{dYYLx%tE~KkK89-u(HCX6egBsZduQd~VaC@Vkv&|2T>|<6vd})kM>O;rXC;
zBFfvxKGR?~#|9B&5H3IM@wa*Ne2Sx~>A@eKAKa03eQCAxwvT@6g6=+%@khA~^7qC)
zN;D%X>}W<qUCooQ7V<`>M5vc*y({6l^|ep$7IV$A8?g#&KhDjI%(j0^(*AQZI)8BX
z2;D!9T*UwC42e2@a-&;W=8KzDBYh1_`x4##4*jU}o8JQA-!1%$-K44|EU(~9TD)Zj
zZJt3LG~Gr_Hm2U}_<Xrk{Y~)49NVci5xoWS%jY|}JFUyIu*30Vq1wC3TntYITP0sh
zAoMuBW+ax8s;6fyOull;;=pXz^K{<xhdWvxP1zuJkG5i>Fh}2Md(p#d7Z&>k?nDVE
ztBJ*YdekjyZ?X2!7QXDd84<D<j~mx2r_;Q#322^#`kO-Mf(Gw^RX90DL&lRCH)aqY
zW=kZxZL3<UZX6grOZH%Z)7rI9<d*1EW%BsU_&ViGWwxAd4kYZ%t@HL<)kB=h4%Cy?
zKW{IT?DK9rQf7I6fo0w`lMTCamI+PQoAyX(A*NJje@be?Gmlui2z_Vc=(p+OZhC1X
zA$|nKN=AfrhAml-_LFkgPm1^0W016zo<e@vXB+;u`N*;4w~Y&rZZvLCPD58j)Khrk
zBHIDx){h^JUv#q2TIyr*y=89-!*1NyRXFZtfu8hK+H<8<-0X@~r>CqE;)_*(tTx}0
zG&k@e6H;hNzs{dhTUy3z!#VJ8kwMWUU4@*^-Hgu<)u5E&r0txhOOfxuh3E4VwaUaz
z4~I^ju|zp>sek2;vrft#p9x$YYx6byTe|cmfB~(WLTz|ew5*h_8f{hM`Xsh~|NoKq
zCh$<cZNu=zjL1%w5TlT?CrgA;vL%tS%OvaA#=b;G3L$GnX;C36OIfpyT?>&l#8?ut
z&7d*N(s%0rzyJ4rKkxlK@AE$2?|r}T`Mnp$G3R++$8{d(d0gjmoX28S&c<6VXdipz
zs7#uYjSDNZErXt_N+@`nemPmgEr8!};IT|)%8TcoH~Y}le&a6*XO!ilub#e*>cKYk
zeA%1<S_ivK17{G+;WMV#6a5x+6MGJ!172tbU7;=|hgbSOaMu`do}<T^<xYs?Fq6u`
z%09phFojkP?HM?E$hmoCy#BF3_SbQJtt9>*`u>`_WzE4JuW9c^L-(iNm8#}=D~|X)
z%fFVoq)#lcr#(TsDZ;5|hz24KbunJ{TM+BPnm6~CAA28SWuLG~i-wqP!WeNd>hU7_
zXPv{eTDlr_4|z}N$7h~V+C&{(!MLZ7<b#TFHM~W3_-w+pE<;Zy_P4$#_p>)ebNw(!
zeR)t`kaV-V_hN_qOc&c?b&2+m!6<I>DxIa3qcm1gOAoZ+bW}lM7=1!R!8v1FGKKBq
z6LqGE<nJvP%ha=O3kivuUKytHDv~mYbL}{Wu<9vHTf^x&+rh})s;@>J9W^p0M=mCC
zpO8$>3S>{KV#^45>oBDYA#9?65?@6t{UcDg{QhcsT;P<L%+N!$-GM63v6{i8v(^u9
zN?6@auXwAkrK(S1h-5%T<I6je#9kLGCe>dSbQ~^m{djgW^rh(K^apZhj@;GDIB|7J
zhgkZFYwY;~)<cneXs&s6q+=m_PM1LWB0}+f*%R^$@~NLG=yvDR)2=BU#|VZwn-EK5
zRdaSNdUh9hWrhl<GjI;A`8s>x(Mut3nwP?#a3&^Fxj_%whRw7iS+E|ML|mD4o7aMS
zLmaxG{Ait$SA&8cJKbvftb(}fosbcMC|n7?0yqa3!A9pm^*%&9Ys<HvRkk=Xstp@+
zxJUPkHb3gu8Bt+aRk8kNMg<2sQz<ohkvzjXD&(r~)2rvn*9(=ZBJPVBP?)Z3Id!ZV
zDLbgN<Lddjw5%8Z5EAp-iHsdj$ghqG`<BTFzgfh#UmM1-GiKePy{g9$bHI{6COd4i
z6$M{CnA)y$DD_Nl(9)TA^7!+8a?ire=e)gNcAd&;ii&!L@{a1p5l&-xBiYG=z(}jM
zr+rVcX<oPLr+jxGFV$Q#Spy26!`4b)hpSAVTzbh*IL{~GiYB1Y-m~`Nw>7(w2|cY|
z+D3}L8)`FJGPMK8(5Mt?Odk0KG4a!01gDO?n#@UbBj8$kc5l6SfzrL`{MN*c$ISm&
za(V&Rk+tP1Ii6_VeUV*_sna=R2}B$f_yQ=7<R)hoqZ>8)dq1MJ*H2~k_|+G3dF=~4
z@%TZj(S`NvebGoikJ1mR<b$cp=$V|PDjjJu8}UQ<L2^2QIPT~b&D7rgP?;?p{>`s{
z%=G5Bz?iAtM_r<+ZQ!ud5_xmtI2ldBM5+@FTNl1*etIxftU3Jw61JIGrI?u8tg{?T
zTj}7tUI*<<0se6%-VOv&j4#Kmd_Uy+a9TCg(p@KU_j{_$Y}7%r7kxJmnkd`aJ%btb
znl`(sib}Bib~&QP$34y@!q#Q0{<%8O>2>c<oI}Z1Vj;+@8|TKAh!@6Q#*9vnkrc=y
z)4L;0CwsWxGc_iJ9zV`vw{*F~{=BR}ija%gsn_3Lnq<V!UY<m}Kvh*QZK5g0Q9X_O
z30g<E$VPJJKd$U+U2T71C#!P8%v_4;sq=_Knle*(4t39M(&KtseiVcK$LSYTybxK_
zvad3rbMyA{nX&QLOK!S)Vr!B&ASGseC7F$)=1{sc2Rh4{oJw$epFMVhI8wjn)5%?w
z#jQ~8t-CL3`Ak~DlK{<L=GDKV4gMbxCI8DM*8hv=B7Xe~hYJ5q*DSCtV6GUm&;HbC
zK}SRxt|3L6iEmrBa?6Q)mwy`=xiBdM2Q;GOr@tTvl~tO?=<pnh$~{Kcp{vrgw5dOZ
zX?)=Jd%)dF3KbmVmk5W_A1-^qGC<<FW<Jy#!I*E!Co6`p)YTx`TFoB?V3}ZqQeb*D
z`L+q$qD}yY$^8>xLVLC>yR=G{LQ(;yYq-TZAo=W1ZjW>Z&j%hI(!HhxX;&AK+X{Wr
z+Izs+wnAafo6TtFO2oWH*mYAu$>4YWC0#&kzwE5>CSyHz5`^O<%Y-~7pqho_TQEtF
zMf)Q~LUTt<^G2)hHnM3Puu4?_l(l~%dTMPc6qc#&_46nxotC5W3mOe1IoFY`{SQ?R
zD9HJgY}Ca`$_Xb|I6ZOih%(SfCQCaX**A?lfOW%;;W|;T8`H3ExY3@b+OCkth>Ua3
zFWNs-9WXzhaoO_zOycy{BW{T{v*L$GVG83<#O@PCNG70JN+RtDj`@W3Y3X(p`=XB6
zkbr7!MS6zn9sL6za#XI^9UGl5VtuW_Pug7SnR<jG?V%+Xl+m85$|B3dM1Qi4NcD@E
z+@p5~YGfYk?@BqSTwiwAa&*`CeLp4EualB#wNVoEub*wBdQfkAZm7<S4=7y9vKVk-
zWP6tWj@`2xBbh!rwi^n~G&Wd%;$)zA^(QfjWJ<nP2-ZYem0Tk|Cc_F^=V$Cq*t`mA
zYicsB1{8}+`nEl+j1w;3vdoP>0$tXcVTKU<!vabDfSP#v&}MC1KIJe~tDmYpC8BYQ
zk-q-^?%nOQVfEB~%E#b9K6v@QNEwUjcT^=<?vM{@Mt!0yk#AGZP@OB@66m-2UwwSC
zaO360`z2Gc5|8AImoK%{y}GrPD|4*p!0-}iF{VA_s}xzPwAska4F1N?VlHb!{`mLz
z@<jvhTE4r#b*YVWkF)u17?2ysq&DZP(E!fj`jA3cjN%P2z5x@DO#<aYUn3vcZPW+j
zF!a;aMtDo5O`UOqJHK~XMv@22h+W*TlTUnQ*esy(h5`%S7SvDLE9`=IC82uhFd7pG
zs$fO_1*v?jS#~8xrsu5G3}3pHLzx9rB#;h~C7Bb08Ujftt9pyRbA?WIyPvD7NhPOs
zEe6Dk-6|toap4&d-jXJ9{DSTj(tCl8f#5Ib#|a-@L@x&xi6a<t3(lBz>N`UAjyhSP
zsx$`cbsCU4fv08+68jgxiF1Ca3WL2u4#DB|Xs6JfEysI0{B5XyIuokbG73ijTIEAv
zqq7<k0!cUSY}u(N<iEDB$%)lNi@#}TUWl#8JUCaQN+IeZdVte4J~QJ+8c?k9LB-;$
z5b@%!{XfH>UOfL{-+S$78wK<9-mf-B5-5w5@De<owV#MEC*KVxB!Po<|6>ge#M#-Z
zmn~{rt|sx6gO4&aJ66_edSI8UeRaT?ACtIa1RXm^IZUoLdQ-jrvy`Jlby<ZWbA98d
zTRLB`yi`r5Dq_M3m4PIflyZ<?k#5oQC&?d)42@$>gzK$U!WWa2q^w-3Df}6M14Xi0
zc|Uob>JlM(T$FGan?Bi|u-S^`)Y7aA!gHWWqmlbAmDzXfiEt)AOzw)vFb&!G!vYJ@
z6H9vJ^%C10Xf;X<?YP}~46!^$KIf)K)`X#Yr6`W_u^-VtN`oW<dEd0$S9|m%=gf>M
znHRVA2nFWz9|%-02ZWVE5`dx<JnD;%`Vck3pCTVNe1?!x>{*2xLykshjm6%bJF<j#
z4{p(Hr25QeY0My=BM5oe$uh)-T_hMY)396TQQDE`2-Y1)@<Gh<`6B97GB9RL<)KOt
z&z~Ol>u_l1Q&*76=nLmVj!X#ERNozlJELH8U!XCTh@uMQOfEg%lNPB)FubXX5FFp1
z($`0q4J#IyYP_&688dAw8p?e#&nf4*`NQ5_sY}ij?x+DA6Vg-LerB#iJM_e0R$Eo}
ztt&-Ug1iqBRuuD$yB=GQh~M<hR_cz?)-9E!)#dg-2R-(j{^|3N0)g)c_N#0ll<!3N
zC5qrES7pQ;b&uX;{4!60r=tw*8Pem?d8^7s;<X1$!cx@r=lF{5xv!3<ZcDre-58uK
zp59e|$mH6yvly~!INF{)yxdG^H6RH2PkYf|2GxCC0di**X9rg^#=LKxe7b&mDghER
ze!{h&jC+hH%Tx`~${U$Vl5>`O$`{7`hnX!g<r{OWtEE1@xXITOJOsCEig7x~NKGPV
z5>fgT!fimVSUlR^I2iGv`$599Om2MGi{z@gS@*rtA91d@Sd<XfCv|j)YYa8xcwI4$
zD)hvdeVhMbbJP9hUEa^Cc%dHlvuD=M^2vToyZ-%BG@@~kx=nRb0F}2>WL8M;ZuyuW
z4qTrvO}hZSG*udt^`zm_oD04}n#x}~ZP@<B>_J{6GGaEx^7~=2A*RnhO%jyFMoVqd
zZCym;#-C2O{&(qF{O`nz{q-+Q5&xU6X#SE77B!&sK5L!uB8f(rRSy)G=g54Qynl4D
zvRSD3^dZ=Blg}HEcn54fs?2TdOO2LiY|^kz<_vZ!=XCd8$%)faEp})PFG8zjt&gQS
zXM{=I?M;cby^|Jow+b@nH-IJ?O!WU!v57na<_K0=wiN9-(mmPi6oFRwmKYj`r&#O5
z7`{E0I$(bZyHKxxhyB`jmiLs(d<p{<n?#BRR{Y2pKaQ4%6|~IP9<1T3KU1CQ!F_si
zokAwPJA349+0XMR&*6Jvij$o{gCXAWFi|mCi))Ika?!bC4_mUMSHKU&gI4am#$v0A
zQFDQB4{5mlRAJb%r(6WHxkV%Q7*&aw(~7u(8x4)p8sJ?$MVozPZ&>hdws`K&yUy=H
zBK5DYl~IhvVJJ&*!!e|L5`JWk<o<aB&olA>$rtw0V5rh&++R}B@8xUVk#BpSTwpI)
zd54;4=;$P!r9Du=BarPP$arCyu;(V3_39miw(b2Ifts&e1ulN8B0Mes6o@+6%bZAc
zqdkjar}u<h0rl!o<ZYjNtNKDJ@pE2Wc>VAV*<*|>-y|5~9fGIpx#(i&cfvq_K3542
zd_}>=GyT;M6e98<8AY^WJB~_d<n61s=6v{x^!}x-UfAo3XyzjccVjXkMnebIbtD<A
z7e@@~H#cu!3C6DAS%t9oxT#s>8!j5Z2#U%ufZP(Uj#LDF1*fHggX)XY&2%Ahs?XIT
zuWCOVTl?ZiYxhfh^K72oIy&|;=2MgFO_)}iY(0=VA!&E8^QmEmU{=UPQ9Uavup=mv
zJAqW{Tsn_rtye7Yyy8N<Xg6*@Pf9n16JCeSCJ!lnm~T>kEgr+b$FLd%#)kb2*Q|t?
zV>kk@gkIM|<WA`1=6P5Ns0Re}pYY<dP#?T|0{JMY!khusvq|OSqFcmLG-+B=zn~W!
zgIHbFIWbAHHX*k|N4{SxV!DSJE!>k3gpyv;ygPgA=!6~Z@p%=UGoNDnKi*9$=~H_n
z8K^6IHh&7m?xp_vhh1a3`_)k2BYgd1opR}^q>Fdbhu;*hJY?PR_k;9N9A)ez`OqFm
zn$G-!wDgO=o*WHHLJ;4-7?+h}GV~V^c0-ICtR@mm$p>?mdahEms0>Nxh`~>*iKQvu
zE?QgirsDO4;CA*JS}Q4P3X^MPMnA4te?f6au~U`EuU>>jSdkN^$f}E}tQW6VpP?9s
zq)f~lnsCqDEAj9d?Wga(`LkCw|3d@b|H!!a{||Y{U;n~_^}qY*`F~;kVZTOLU42~w
zg92Qge0CtrWlvURHDzUSCGj0lON&+ULb#u+xT1~OWlvY<AXdfm!Iy)6rx^P<xx2C|
zTBBWpJj9P2Q&M78G;&3|djyFeQQrA-5*-w1?i!$n@$tj>y7~r*A7fQK<Lc|aQ(Ebm
z`Y}*M595sqIPd4=3?6XGH3aSKYGHViRnY+L9poASlJqW|a&^YIxU%Z#u>M)cUtfMr
zLfoefj13?L1_&|&KhUo!NLN1$?FK=nrjQIcYG;8M;?WQsK%fXD4E`X9@d@MKQkYqu
z!2SlqU+-l8nr5Ci*3(0oTbLObpV9w~n)!>9pPx6QI0X6l1_gkMiQCxPi8GHtjF1q-
z2C%X@IS2ZkFgHK5gJGxa&i{YDNJGDC0QJlN>->K%{cpapxwr;80}@z3I>sf?*$2R3
z5XAJ**)Jdnf|&V0`mxX;za4m|4txQif&jj>1H1nLgFJ}u@*nWc-(@aX=z%g10nF*_
z;p76~A^^)>{)cq;e}Mm}11JIT7`pnp2B4kAciO@{0ON-CcKx04$H)I9`yWdFN1%^)
zFrfMOg#e%IfnMiL!2d&l_8Trvr*>ck02ets>7NI%B!EXl&{k*vgz0`kCoKSc5Onx~
zU@yxP00tbvXyg`PU<qJ=hcUn-*ywL?_7#r{)*xK~q__B9K5O>py<eOI^*~!f2S7UA
z!_|1FeZ?RMCXNm=z5rl=2c{E%vDhgO>Hu?daXqCEU^xIMdSOg<>Hu)UsscmK@1*b4
z=gSojy`6gQq<48boi+zB;7!=HtGD4!dGIV4)i22W&pN=xe7(=^JQq9*e#$lQH@yH4
zTr9}rf)Pju&xLyj1zh;^UU-@t+Q9hF@?{<YMmy;{_m28`|K?FpHzvk_V2d5v!LyhY
zoC5R>0Ss_5nY;R0?zHC)90FM}I6<xu2D%JBogrU{28lx!kRfyu{P%+bK#m)PhP*+F
zE68;PDX!q#-^$4UzAq5+0_kWd<ab^m_%#1r@{c?JR#qHx2dTc${O_{Pke*mR6bvwV
zKuaLi<1d`zkRJFJ4AR{H9Xx;3W#Lae7`Px@34Rp(9fzNRYr~JhH6U>~60Qr^fu91Y
z8t~(P;Wh{N{EecW`YeE_{DH?GJkj-cIV(VWGoUdDasr=IpacdA{0qZBwMv?B@S(rw
z!~~$w&ONDwe`qT-xYO&e67v7lqGtcIHfQ?AG|w~$K1Z0o{)I!FX^iPB_!$9h^LJV>
zf6xE3m3Mjy4SE#wPv!n?U;M4q2|%g$?>l`!E1&_*uKz%@{6FI8jI7`L_oofQ-#YZO
zcSgGTrV0N&b*B|v<Ij@LLNRiGPgNRLT2>lVda9JCv<SNmtAsVcYGAdnE=U~K1Z#%j
zU>&ehSlwUc|05pytM8$mn(y${PQ7=y>u(hPV_O4`<vhtL!YRRdiW9*p#d+qhQsSI~
zoJTl~IHf_3$Y1^Tcm5XtN6JCF*n+41t$qIWvv%GU^k0}jA24tL4nY6?iT}g{mY~LH
zC=}of0PVfQZ@y4CDEp6Z-k%h~#o$Na#{Y)9kL~dM-*E5J-*sx&2}pdG%r1>xM|Pdw
zf&ca{?~(?28sJm^AL|DA<F69`M5$|#YuL_w0_kD=!UKNK^kAY_7e4`}K38#LUuOk5
zaVKwY@!wg2;sLIKt^pygE(*}jyt#891g%^Ao(~xWdj6gkWC}srD`4#F{d<~SFa#9^
zLXe=!-_s6(v64Rrg5Ih*2M2`wfe%d7-4GP$c2xQIcLw&8VEmf+^=loBW6T*4L`(nm
zi}LK(FIq0Z{}qCoyni)8JS_0n%q%d50}vw*1B{2^R~rPT8U`kY-`AgrAqGYmoM{&`
z%WhUSP@s|<Vq}297~wD`CO90Fz6$aoI1dx={v#)L?K$Vfe88Vi`C9r@7KxK@n)xk0
z6C{sb4v5*!Dj>L5NLcEi^r6EtDynMg$22tcPU#yM8X2EHf5Fnq8j$Dg;_Bw^f%Xgx
z3JwVk3%_#xMr>UC&4k2^+js6}X5G7={VX>x|9L^-i<f2P6_r)hZ)<8>THD$?I^Vtj
z(APgOI5hlaWOQnJ=G*M|x%q`3#5Ev)ByEznws+`afM9=+1wQ|v?BCGE1L$Ie!(njd
z9l978Lw5-0fivwtvWxe`Ic6vSJqMJpvGAQtfBL3*x5QBk0{`WJ&#VHHDpOL#9nyYN
z_Mamx=D$SQUxfXgt}!qfF#f(^jEpcQ7!1a=iwRu2Sa$7PEUYZQFV;UV_TLxh&c*fD
z^$Q?j065@qI5YU=X4}oi{l9$u8VAcr|6ijJJB$G^6O0E!LjPktwf_tE6@N*j{+rSc
z`*);+uyhQ9%(FL_eB0FOd<@4o@fhUfg&e}`NG1^eP2oN`%1hgEl1>c*(L~Fyg6Jvd
zDJMO#yEm3YkeuCz1~58<avXk7>>E!nS?hgTshP$ZW2ZW!fHd}5C6MIA+{bPmh(<o4
z9k=aM4R-ItJZZ4}ap|rLY_TG{uk`!*X>}^EH0c&u@2LY-mwbLk7PH4wQ(3ihWP|8m
z@I#-Q7w^`5X$5*`cEq+;!5*Sn1K~{>t(NLfbhfXh<>;t*Vv4<#>>4JLUOMa>cQ9n$
zrr5`8efUw(Whnl}(r?MHf`M}FPu<&HqMHd+)#r{S4_<f<WoBWjy^osh3lQyOi*<v4
z+nCAzTH47q#zopq9wE+=$_c9-t4xj_y=3g^BqL*swI3cO#&6!DT}KmOFFxdC%?W{s
z*>8TEcmG$+6By#?(u9F4IV8#K%4a{MVy@on+4~@n{o3qYh>-@2zXsJcocarzJxaon
zROa#9Corgf`QMUC{;Zc?*$3l=O^bgBY+pGYlIw6feV4>T#&0e|`BA_Wtr>&-c8978
z!k7*Y({GzIFX!iv7t<<2GmlEO*MAU_Oun9bc-}JZgxGYE^#|<h9%?po3z8yOgxg@<
zF8j|3aAXOcq-9!hlqxxp)B<!*7pLUiDa<Wq*r}~moi=vsP@Gq-DbZBNgypr`lIhcH
z`!^XO+NUS%$B}!;UgsKp6wL;U6)<~3<_|npzvDesme?M5oY?Mb4a_56QQ%Z>vOL8l
z?E1`HQmG0$QvIb#alJ7*SS3r>Q#mUs$cOPz6H@(?|ClNT@jtlh<6l`>mXpt`Z@e#G
zJSFnjc<%w3l%-CHtN481&u&47<nNzV*{+^E_G#EF+`_z|WcNuwukVPlYWkVY>EWID
zfoY{ZJ>&z5)C{hhc#>!}2%Ty9gllXk;<)lpQ;dG^wDHeZl&d-#BA%6PFL=wHk1@T2
z&s5r%*H;Y4=Qg~4q~P;iB*W;QrqWL<hZvo^TRUnLqO`j@QuHANr<Uv#GR|5!xM3yf
zmFm)*#bYChBYVZ<4yPP8_USq%?my9(P0NZBBF9hSmnf5%Sh`%s#6?tVc7Wa2%2yR>
zLj3n0-TH7N<;q>eX=t<d-*c7saZt5EP%lAxsC0RQ{LSa(FmK-J?pu?B;(JmCU!WeL
z2<Px4=ciH2PK243#yw<L{{s7o`Py>FPr--3#T@0iHr;Y>`cCH8));2ddDw>Xr?r0b
zsuWU7P)+qcU*iLp=357C16Z~#vk7J+p;I|Gqd4e2jl0OG>G$%pgH(+Isp5}zb;b<o
zwh3Rl3XSGooZBdk_^sN?IvkpreYGO1dOg3gULx?lEL>l#s5_c@=q38fL{iDA_4#Vj
z_6`54A<#9JRvmt)c^NuAx9mw<KyyKz473X_*<v?TN^f8%6A){s=;FX|zZF5xFw{j<
z?#N|#&Q;tzvxicjwq$7zQiUVWj`6QZ8QrNm8to8$c9*|Z(W>`P5u4SZt(n^DJFZd*
z*K;}Vv)3>aN;?r``7G*#o*3*B8oY8wzOR;Y$kJKV`=KqR6eb+~j)^Y9{V(SPs5oR9
zV%dvWO1cgPx9G6ztx+07k)<3S0>WPfG_~(3jTn-=?maWUsC=eunRzbljnICz7h|ZJ
z|4%$K{8t91|I9rM|Bk2>w&e~Q#R~g=(-HdxX%C_d4B=iR(q({8+|`-R*3#4~ISEL2
zQJGa6V&bEdlGWw?4=&sp)5>YUls_{v+AHK!Z#}O_RqdqmFVYi!K{v64n>pb0pb)lg
zSYKb!zbuWzffw5+JwMZS?*T($j;^jwAPvYcgRIvIC~We-RZ^Z;MM)8ixaV<WI#L5u
zhjcEy&SZPD7nZX3G{DGM(%PO0&Izp%qqohlu&`1iqEutjl@K!@)nr_}yqNk_X6Rv0
z*VKJ?Lj39PkChH7lBGMB_9K>VKk<x`CKeA?6$5K=|2%cxF~(k-(Zk~;8yCN#*{{rZ
zNrh6Dx-Gt!xJrSS!%#Aoa*V1n(0GVySnHuOw$G672sA7hstlVQYq}BLr^96dnRm6x
zWFWbUu^2A$A@^l)j$y?$qTamlY(C`7o0XV}q1h+nn^B*y7GCrDt~BTQax#dSDFqxW
z4^NA|h#(??9a)P<zJCEaTW?WOI49F2X)!^5pe&^*!8>H`NnpxzTrkodcS8(8<|25t
z;MlY@eD`11FH4s8hSta5;{A`#yn27wP3L~oB!ey@wzg+E9}HDWwDjOdI*X)i!Vj-e
zj&Bp@l;Glr{w5VJdFtEdJVki{T0KfL(oWC}i9+(DJ!=%B_v6`@=u%Ye`!|%2#H@Yt
zW_ut~0oCtel5}Adve~Dc)>^{sMZ`>DV;hfB)rnl-08FS!x-U`>Emaf`)t_nI3^aV4
zaf%$IeP17|g(nyyg%%rGY59@Pgs5!;0lT3rzka=(JE(>8g@(-=$=}bX)Ceu;BJAj+
zh-$&k(ilYLW*nU(jB~98!Qtuev8rt5cKfvHa;u?&q_l+0v5nFZE*^5cJ|QdqCz3xb
zWm!(;k?^S3$GU{0Wg-zl%El7T(yT9y+p%g0f(3rds37JSv=|?c+<3FNIZVnXM-)@7
z!5U*fFz^^-8my!Vtc~^xu$KCLzP-=8GrFYdu=Ep+CtMvq=jp6W6kqxvl1DX%jovn~
z$6lNH^ks<}FMD)J<a0?w`SEtj)fAXQpM@5qhDZ*4i1}6{5=_+>ajrD&$Oz(Xvs7C`
z8)r;tO-0|Tyu?Hkm!_#?UT;j>A@^X1GzQ7UtyK~)?d>oLIMPLdHcXu0Y<S!ElZ`0<
zrP?y-{OkH_k@tC?KJ1WmIJ9Nk3&I4i9KiW%9ijNs+mRL3u|E;e#jw#&pA>oRObf;r
zYQOj>TPH5CWA<iTeO<2n2KF&M<q9}y+FbIY*o8Th>d6jtmA<8v<l=>+N!s?N%HcWg
z4-13C@KA+v>7h1Q+J}G(x7Tu(#{kE04@WVPhdxe9H<^kl@<tjptXX~QX3yH~w&`SA
z&vQb^D*L--@THp3Cnb8BEwo~acGO1%BhKxWwrL~ld>=Zu&BPaTFk<lmu2Q~no$WHv
z(5~)|<g@&=64)LEr;olT3X+V$kP??Vdbq;>CAv0tL-<pcR1?yz?u74ahcO{&zvQje
zbAxGok(Uyo?Omq!bhhik!0vXpcStu=WAmXhn@BauHDVpY-Op`&3{9K6bkwnvB7ki^
zqlG34PFJAHam$KNTAu<fTeNqv;<MPL_l@xy74=nVISN)XmbM`=<de}4hl{B)pcXas
z4`Mnt@{t#V3(0KeY0$?PV=6;+JuYu6*A4KbJR@3DDH^C#x#7OxXYlPdSkH?L-qefJ
zMl0SW(*qR^j#kwvD^23nY592{9$yakN}HU+bO9~Ok=n-M=N`hAD|N-=hRy}(-uHol
z{zu`7dz~!SjF!SIE#Vnqko9DBJCZDe=~%sm;u)T@UecS|RqrgBz`sGlc{eoJkdGSr
zeo35Q*sL~vJHu4pXw}#NYW$XZYJj?%vfmMHccB~pSb1=%!uBHP5aqaBb-=BZklHbC
zr*WI>%_S&03ypx)C4??@;rN}aIZDFlq=oNqFR2$l&(X?!-BM7fOMEvsP_4xf&6P~G
zBtD?C?I#>PlQnJFc2)Z)7nfb>s6uW=t6_b0qRw~Me&}ri-_@BK-F>$Qh8Gl<G$>cY
zkfgePg?i)*Tj{v^t9d66*7cuyFjaGXziE4}dYAbThR=;2A31GA2e(=0Gb9WT#96_)
z4Nfo@s#0G^p8V{Ca|UWARu8opg<JK2QT6pRH7R^8FBi6_I~3&yg4PC1<vflvInJms
zYy%r(FpL#6%FsXJIH(fSm{>u9s^QU&B2wtCm2uw-ZLhVWnJz1ZGIU3}zrr>fVLcHk
z!q~V*;iz|xJU*fAe)Cs?=m&;|-ko{#!(z-}*aU7dJ;9RuuojaOP{)2?a@S6D^_Z{N
z_k!dI5Rnj$+cuWE3T8TZJ306%nT@1DFkN}zosK8xOAnTQXNC31t!sAJ{5-zjrWafH
z_<J<+Fx87<gKf6Ly4H;$_Cy9(dL;|2m_Jn&J#PBIEWJnPTut8~yKmDz>Fk1-lMjVy
zZM551v?5p3AZk|-J~3PBX}^M<t=vx$-s$t`tGn3K;~`%wD_icg-3AWP0Wzt}l~dhE
z+mU6+_U_|W*WBaqt#)_Un4#G0mR+^69T7$8hnWYgznf9HS4h)l%v9U{`|o4eXa_AF
z2x2xR{5Qq&3~IJnA=(Jo<idzp5{}kDKSck8uX2ym5h}@bcO{Hx8PpE3jqF|#=nMbi
zsC;Z|efIDuB&`j?`VC^eR>_AE%hrKcr_Dwrn)ifxHKmzIhE${6--MT(+YN6RdLEf|
zTYOf2_#5)Dn}q*X9i6o`*LWH^mTF4aFq>_Bk?%3|^2?`wtupartG9+Xrlt_A^Eym~
zs@s!2Y<5y{=}$*n-O-Hfa(+$Mc(D85uY{{GP=Q7qC+L&xs6GUXj*1@8bQ`GPaF*%{
zsUEhrI%k7gaC3maJuo5>XYa@P{eZG9=SE<1@qD8O?JZEDLym^F+cl<Ac9GRQA`#C%
z`QWNuDyzeNzq)^4S7o<rDN_2#bN-OSQJy*vA{SY{D!yQp>Ozi~-h1Fe-9V7Wllx6j
z?U=)xRR7gf)%ZSKFxCwbU#r-HA6cE+j8T!jP!ZFrUyeSB{b=%H+P1`1y3Y7$-R&P&
z4*><*79_#KY)qZxLw?~pnKQa-eD%B;_R#p)giW(EmcJ;qdReV(^nkvSgPO7^3VrcI
zloL%9EnM}WbV1E-CD||QN>na9RBhcLAfr_|K6zl1J&$J@6^|e|h>DnDnIq+hyJMrq
zIq3PGQk|bJXC8X^Lc7TY&iQrbMQLLrJ5{fdU`E~3*^1ob(l+e4mmDhg<UU=kzoL(L
z{c+ybrHE{33-en!^qF0x5YcC|85O;K3Mj9%7sU*%kbL?^Y%d-e^_huN7Mq?jV}J58
zcj_#O{rdO!T3+A{tQ@HLN$jr53Pd|aPq%vqNo6^SydfD|mNdLJAod1kuj9!QSuM2V
z)7G6xkw`Q#+UL9lA-G=ANAeehq@GcYe-Qjp;Jq?DQJ3pg<`0I_1rQon0OyP)%Q!wI
zGIwU;;YjqO0|<8V%O5jKR@ryXH54g3?m8szyF4q|d{q5I5og~V`V?qXJ33IooUFjJ
zRG6`c%{5^~>a};557eK@QWL)2zQ&wm#-m*^vTrM2(nXr`rXW8m7DOX?Qm+e^h-XAf
zQye1{8j8X+KM484e1aPX>Wg)3(^76p-F<YG^APNt*{B|!HHoMPwC4Emlc*)zXU|I`
z73|&zW6H1APOp7e+fQ$0=yE(gNhj%F`oIRggo&wPQ*#K`KwEG`=TM{<VT7oD-KVOj
z(7xLzU$dRTslI$c*VI>ceued5zz0A5Pz2!=F7}FWm|iDNctyUgwtjQ*iT&xOsP7Xj
zmuv6sU!-(oYvWfGj?Zm6+B|zK9TUWBC@%A%RpvvO`1NMEIrCcTBtB_Ej0~JxGZ_+6
z+wpAnJ?Q7;+ijX!pIZZ<?1L$v_U0Lk8Xb2r5MqY2r&Q8Tllwayg~;Jk+V-7d%D7aM
z&nhP23MDJ9$84r=xjf`F*5=ohN&sh%{M3ed;JhFR8-pcSnIfie@y<$U&CsN!x$>mF
zz=igytn^d3rH*pABN?VOJReFG$elJQn8L^CaFTrrx_aAZMD1)>NROqe`#w9I!Y^(0
zp8|H{d<>IOhrqj>9yJWCfbeayJ&g2rG4886EyIC^{v;!e;wZP@!J3T4;74VR$E=v)
zGhNP!ABo6i5kR3AEpsxp6CqL&JFKP1)}g>L`22pLM~q&d@kVmlN#iZO<FOpk4DXnS
zz+MtHx=IqFWdp-2ZLStaabm?4)Gbvtur+WT%llD&{NlSM9$}ke(Y<;srx^B@oRG+r
zq$RurE4DMZOQL~i>OKeO-1MLLf@&i;oKTB@RUgnUb2lhh%|_;fk@!zOhDCgt^s@h?
z;^^Yk47t38s0K_|ywg0l&y<MSYdmy4{TkL4cLt$gnj69*uAGL{0xVjcvvxX47;I`i
zfi|-7t7!!J;AYAOG9GZE`SDbiNcDlolqR~u1G~OrsRO431-KJ@ms~x*Tc+uVpI9&S
zuesJ?R9cPuH)slB8+s%muz}VkgIVkH&n)^il-0i@_0d=r_@S<ej8!dme0R;?*)>qh
zHBL7y*Ucei<^b+->UF@ktBFI8cN6d6c;)2lFV5H}RE(k6&{ZLEqiPxcfut9{?6&Tw
zWFejs(UF)E1~=5a^t2%LrVdbkdhI$#Pq;$&!cE@#1#wm&`RSiQ)8G2}%BA*p^R}$1
zi>O|>b!edE0jHd4oBBb-s(axRF1z8VL)d0dRc%eo&5%)1QhMIYt?7>8(|u>sp9+mu
z1?E+EcM@K^A?&u@>ATlK{~rK+0GkORixR8NDl3T2%__AaPd(zs6@|YDX77G-`Q({G
zq2@D(EiYLbiQt^Qwyu+}=Z%Jm(Agy1U8l(H56_c3Pd&;!QXx3#v;6$6mh)=Giv94x
zXk<Q0IP7C9l1oeaRoG02HF4p}!32!d;CNR7+rD$RZeM#Z{j3oFMjSE*_I8oyb}B+E
z{RN?MH?1Et(Gs=9x39*Qd{3EKN}E7ch)<dZN)0U@ca8Kb;@sVNuakc@hhUeQAe<8)
z#X>$lRh4s8Pw-YS+P!WxEE8{{@4p9IK4-dzW^T<-VWbZOdEL@5MPojSm&lx)+edYJ
zQBl238ke@bKe}Rg>1JQM$9&7)lj1Dx0lW9Tk%Yb1GNTv(uC)Upgx*RJJ3E&52lueG
z5R<?f`MJZWc-ltBRSEC3j41O@Q+k)600!DsS~}7zl_6>v#f4~pESONy;AP@+-MY`!
zZ*8P&M#eH)%tiUqR)CMl>k|xju6@%irSg3wF$}AGY!I|1$BQn<Qq7p!_IO;W>y)#2
zVQqNcHKr&phU*+E1xqkp6`jX&0akw!D#lI5ywKjfM!l$3nQwRc)J+As{k-E}K78(a
zB_64Jea?I)5OynF7x5Xh%tl;q$|*;+ESpUiv<ab>)byQVCiV%fD&+Uv#jr>+F2@*4
zvp&5?cn-oiDvlkcv!W>%7wwWamp(r+eCi<(z;d&!vdm+@^!MHJ^p~t!b$&L6<+nty
zOGr}%I;S~by`WwoM-bvh?v0?cO@4e=k>+*yA;@e0e0JqsCSI}N-j_ub&>V!*`N<KJ
zSuHv2k@AFkMm7YOtv)7g*MeKe!t;P5S(zt4bq+ipU?AVk!#1D%1$n2AWZ7pmp{jCj
ze5rcn9uTx_V`ty4Vl8t^B25IjalFvV^qKxGX7f;2;35o1-yKV_r#jPK`b0^X5`BMs
znypvfK0QLq81$)1Y7TN8uXr@Fc3$k^_HZAry0k@`c}8&Q1rcU6xO`(cTi||h@j@A-
z^{zDNi(?pRPk$HdXZ4qLTvw>TMj=KR+Zo@1*<J*~0@^Nmv<{wDhsE@8MZL4q#Kr~5
zD<gTsHtlbGwIBJo-^9zr`;MTXlum%qz)OR-ij@hjkUMy*Ovtf-qw~p~`T8>$W^}yn
z=h9v?AJ)<N>@XLnn$w0aZU<Lh`;d7|;^FW-z60}xZBnt_Ecta*AC@OL*-}+h!Ak@f
zT+~)w-2b%yqz|y9i7snt(R!btt8w|tFz_rvTB4;=`CKkyraR&#^wT~%`2`<@2a9-T
zI_G#oRL?R%MRL?QwrsPF&5V^kNZ%*iU35dG*U|QbQ?OK{E?Tan=&pSJSpRVsiztf|
z-=fxmgOAbSBpt>dcy9WxJA*%y?OU^iv*Omci`xf=lI7E6zA1L^*VW@b&Z|*6w@7D=
zptyncZN>0)s{9n{M&HDbQCxhbhQirg)%2<=iezt@iLg*&Pt7+THTDxm4!6@}Nrv>@
z{aRUwH`XmRQd0NO{n9!Kp~~0D&>`5?YuYm{J<0`7T%}rkair(xXZ>{9PK6BbVjlTH
zp+0l(Yc_cvjqXhvm=f7-8CqhbEyXu1lhXc*;usbs$~iG{Tq5^gZe?m$$kD4$5ak~a
z`oXagqG7A33><JDw`DxMBcGUS!EIYq(v;1ar>mFlQv4&Ykeex<<fKVVvZ5S_1p7dt
zHm#E*TVF|F4`00U_1n=JKMQfbh%ski9R;=$-CX*KUa~!LSGqzpZ~0YbsJ7bW#mG|p
zbxy1+-}>OcmawNdG6}Fekh|W=CU?^7$OXXNtT65rZn6Xi$4<G@yG@8(XR6nOQ{x3s
zlKF)K3&wromJQ2tWURC^C3F!lG~M!@iWaJ}8?!XOj-d4S5l+X;N?H`U1nb<&<3Syc
z`iia49$tD+(F8Lc&xYa6usB=s{*9nVp}CG<tKly?R)I~0L+D5%*vgTmN;^`?_SmwJ
z*nCvaoR=6ZeM2P8tjnpIVQEY>g!ZUUDdOH3feIKkrY`ctP?TI`I60gm7bgCo?$sW6
zMpJx(>P6QfS&y%QU4F}#t{wzCL=+%yg^VNi&&)Nmr6wUoe2ZN*8=tCZp0pb+E~BLk
zZ(5uVp-rzR2#$T+EM=m-#ZEr=qcRgO;F##l3+7?DouBU{u7$ULL_4Xx9bs(e;nHq~
zyXrBW)II(nKoOrElKhn3g)0y5+c?&i9omZFIK9w;KTc5_%nNm*E!eWD*9p_zkdrw`
zK^+)bmXth1*NhsNkPsN|J>ef(V5a?Oe9`n3-t@lj=}|p3QSSFKlN_IfqFmw7)g<P`
zot1)S$-_p?%Cg#+4qYjegJr!3?ik?J@MWiS(0do+#4c@rZ@f+e^Xz27mC`osO>Qa^
z?YW9az2c1RoSIeEn-0?J?y$Vl5ryJesV9%_`B-ocV{dz^_2N&1Aab8^PRQ-UIa>xp
z$Sqdf9^~Y699TZMbKS!4sYejKNP$H0fuHztlU?O<YSvYp8FpGet{Un}dW|QAh66I1
z45BBQo^bv6khqq5O-!SP{D@%IgyK=e&MX4QX^#rcqM<zwV|f|3IFAJw^8eUiSi|fH
zkq$->z35GXsCU-u4HRMe$b=g05nW&r&)UclW?OGcM09F?HN8~2r-%K>O#vmVz3;-G
zyS=+GWx#esVix(C>z`4~mQs1tr#P<%VrFnDcCl}BEGr&sv$hyOH+gdHlb+qDt8?&v
z3*h7VoD2ns$?TjfFx@Ac<(d%BcaSW)rxq`$tDB2(2>dwabpH07p3C#^D(60`UAXp9
zsabVfgR+Z0kuAM^HTQ=YVz}Z96*)7`$oF`yPe9%$*t>0iP1S_S%lQ3{YI5df)b6^}
z-f0+Pkun!tnC?(%at!B!A<J@rovi7QuqgiK208LZO9!x4<mue~n7e`9UnG--%S_(P
zoCAugz&4wJ6$Bnwm2bA1?I!0=>9-eGlHR<keAHZ9ZGTrODKrf#GT@xu2R!hX<+Ne;
zL^{wOMKTd(J1q1SDyKzN6-q|;ww=2#bWgPK*tyQr)Gy!#0Zsj$mkgryIkWwO79=Y-
zF-rhAh#}<Ao8DmOF}jFH2p+&U`8nKl_NcK&8L~fd*UCtv0NKrcFd`yCO1<w+zp$5$
zh4)-gmVwIqmguO<f(PI3#-cG~nayQc5%S!m^mP|P&h-+uhKy#fOIq^`TQ&|q@G5ir
z1Y^(bonzIs?)nAojfCY<!7?eP%#qB7y6J$vQk}H1Q6C#;W$%(Gbnq5M^YHnfr9ZOw
zI#tCTj|dw&^vwL`O~#imZ$Z!`@a_xI!=hNgn#SxG<gt1K1o<wG;*A=yGtV3~DJ;hL
zmdM>U9jI767Ja)6D6PmrN^WhJ?rf#^7|?HCgZnazuV>Rk0gZ2?_Vvqa*%KbLr!~DQ
z^bE0Uc-R-Mu4QFmZ_Io9>*1Q)i1VnYC{g-huis0S`x(>|{VQVHc5}^k@?~d>6Ra-U
z$$A}?sWxAn%Yp~6bVHGkC?-Ivqh+F=ik~c93ZL2_>Qs_&Z#XVac@5USW#$me@e-oV
zzmw<=uzY>`se{H^gLCJO#S)$cy-H-W?e~@LduEkwe{4zR__H?#>fA#^ABJ_EwRNyj
zcfVqr_t3fKnOk+#$*U2*2Zw|ybRLSvKG-)sz9SA__tIvSJN9lLWVY&Am4c$q(_3-~
z77Yke=Fpxg>vmC1d@GWh-Fv9mw@WS<c{tSQja=;G!*>`b&bhHax$CQz3XVBF5jVvo
z$mc2AVZf~e@`mHFpE+D;b-~PJ<=xJ_9m@G>gw|2qV;uSLT+9sZA<mn|Ha*;`6jWCn
zB4Wl{nyHp|rMU8AUQu*NT3g1kl~1%3*-Z)3hrvlAGw@a+i>hRD8mO07{Q!H<rgC~R
z${-qf=(o2pTOAEAJgXDoaq85U`aoiNo91A5V-Cj4S-h&MKZ7kKQ4hnwp;fv+$MY5R
z)I?*Hv09e6hZGHXr-{0!4>(Eih&&S$p~%rO-!HpRtKFSv)E#Tv{!FF!1yW(oMMK>8
zJ^wtOaLO$HihLWMmoDRIpI4jz^i7}ltm{WTzGucICv$*NtlVbH_8wH#FGxg18`*~6
zMGc->wvx*3KiB*7kdz_AHCy>S=IaXs5xvaco<9KN#r3{Qv*;@!uky*}o><{vk&eD;
z0<ElD>eVOy(-S4*efRdgy%2rmZq6YQSXXI|Rs?a3&L%S1aA0n_Ct;(mSg_S<esj+W
zf!q0}97U|^M>4rH#yf6mC?0+X&DIVB15>yYIpP`l0qF)&I{By9g_@ie&BI;?s9WmR
zWj7-S5fj`A=e4gtCYLU>gzc}5Gx54NWaDa)k^+;JW#>LI8FNg$mw65ZN9!RiWRczy
zzL!p|irCttqcjP%`cfZ8Gr3WA#%oFUJ|`Txi_JXswMoE&g`?6I(4|N#hyr3$JcpL_
zbn2gC(~_akT|a=>v_*bGm+{Ts<97m4SviE$dqLZa0AGndQ{zWBLriiPtsWc?_d|`{
zfi1j_5-J7TpGnZ3Yo{6!U`GoMEVhm1p{U#R4_F>Bo(GbGfsf%*A!EMU;hRANwRIlO
zJO(aqmqMOQ8H&oTV@beKM$DryTq}aVfc6;YisI;;&JnP8029$ylrdjcG#(aTRbF}H
zx<rS;aHy2tE)f=y)UUYLN<B+Hgom(Haz=X(m1FUPjv&1qEF70Gzyg-`6a&Qe=)hp6
z_Uh5dbA=<tMZDv0!(CIX5*ej}tsKhVjCLLGWo{M(uD2a*<&jg8JRjL@6y%>JwI0+z
zc{Ei#E^XUK99J2PDpLoJxu$<^b|kBlu(X#O9zKdcP1^hSn5z|-HdG8qZe31X`!aPz
zVv6CytrwS!WcxQH+H(l!fIH3R=3#Skj#I<|_ubE10}Fk1t`%QSNAfmB46k=mTq3b#
zSIWt#;VbIxm>X)FnL}{2l9~KB;1qZ>vpjuHCbwp6HvBZ(RHMTsu+rStpfzH@b<^cQ
z`wF0F9I{gg@&Q9?I`Srxqy)y5y!7F5e02}?KCV57!uatQ^fW(G0<e}5Fi4grGSfRi
z4{l^=LbO*qg5F89u^iGOp>qj4H72Ur(XMr3rXfl#%dG!VVlyA>m}z&vVwc4DU3~|Y
zTP#c`%elJntW;OfN$;rcpJ&Lq1X7C$igWyMO=aS74&fo8r_P<}Vox}_3Ju47y3^lX
zHj;j)b<qh!vD(>+`f7>JFnKiyjJ1`cQ@ziAQ*#uF(~~@=&3oQ+wuw&RdEbpaXOti;
z=U-?El*unanmg-iBU%~G8@!;@FtT($aAY$xg^oNzG;DU1FkZdu<RSi;sq$@-f5Ll@
ze2xRI;f8QcZqCmXAVPw#0pW)N85qW00;61Cg4cRAoz0za`=pKlL9tWT2(DQl+S$4Z
z<mUKh73J|0_1kTqPYaHGy245WGhs~lEqXhSHTu{uD8U{qF)_To%<b690xI8PJAE&?
z7<gTy$dG-JeB`pxIf~BowozXFF7iI@%11?#LSo8lbu}`3mA>eHjkH^%(Akot3aFAX
z8cs<}?G@+3T;3HL_<!T>Lzw0Xl1dxs|Es(@96Alm@Ozf*CnpyRBh6knq#wP}?^H9O
zj?f)?W0aF{2?Wtw$CJSBKjsnDv61}wIlbo>ROy(^kxzdSR(siV*T6z`RrZY>xpR*8
z`wmJ*ip!+mmD|m9B`Py&L`R%fsHH*ZF=?RPuDDrQD)9bQtLRtdJYJcFj)_2t>1m_k
z15VuUH0GWkP5u50BE2Jnnd~F!2CZZRlZa>WS(Pth>AUpU{vGdx?W(-H#3xI`!0Hcc
zcLZA_-!?(kKbLh89r&e&X~bx*0WB|h+<uZXSs<UNY*V@as@DPg+5=iYeBasT8PRQ(
zc8M&cqXv*oiu<TG{YZBktq6*pK66~XS$=!r#YHXseO9EZz7OyBb02+zaJO`hF`H?R
z+J|jE3UsFqII4q?6d2Hv3*NyrR;nA>Yzm)(Vz0!<>d4jBmw%?}@#+Oi-FU^(lbIn`
zcWzt~I-)Bg6Ofm(!9Y%!RwN&mUJj(3iPXgr1eZ)EbIQMrt{e`Qs<0025FN9Mh_699
zS6VHVh`?te#AIRZ6URYY4^R*G%@M1UoNF5TiiSVc*vM^g=vxYA;`tM`KD^X$(}jej
z-rPnNpmsIxqkljV%(!;{c+B&lGxN$3A`_l;<91l;x$-3QN0xB^j;B_P#LmjZ*wmO$
z5krQ_$VuQY19g4s!AD{%Ip(_(_VY#^+lvuergh3Oqy`h@2e2W@hxqZ!mq&r)>HP~;
zwPwj1$6f2K%jK#@W1l$f%aA<V_4&PsGUj3zC&3!xMw%~f0b|Y>U$0RyV5-R^pgDYf
z`P=K>;CSM`A*(f+4x;VHeSOU$&jj>k>Q1m4!TpTB<6i#P-3b0i{!#x==l!dQEIUKt
zzt&*!#Z3Z%B}_*xY!9ioj%b!lM})<<j>_6BvM5%H9$e`gs3@v~vOVr9<6DmQc+%57
zcC<xdThDe_g6ws^59whB3tLe=GV1wQ)WjyQ)BcgpTDWBMLd#eEcC0>%v)(5+(wHC^
zUyq$yyiwa<ujhoh3QvFd!ghu$)%?Ok?r=DF_^!rqS~<>@i;F&t*eyxWpZ&0Kg?|5V
z{pU&ni;tKfL9{XStyP9v!YncCw_BHMI=+=F5y%Zn$<hzi5$wUM@zpxggdpB>dqeHw
z_T?MLztmMG|9m<cbJC;&^N`mpv&dc?(T7QhG<-%zw$>ITX&HxPYMxC7-rL99;&<~2
z%)6xblocMtWv7xQeILCbuai27)@|XJ4o>tK4U|gX*)aaec5je7XH^N!LOO4c+CvBL
zU$Y(()1V#2Pu81i7><pUiu~c*URl}xsG%+5+PUyTM}9rwnb0h;aoQjOqQ%hSvCa)#
zq!VOUog16ye$W+&@bs@<cDGJjX^1JEGCwKuea(q|%mlK%|1kC|zb;Y)JDHDLOE+Ws
z1r2j?Vi%l;H=KxwX|ovM=vCdEe4QwG!Bnty6qV`g`^{L=sD4C1@<Uq*{%BcYlU6zk
z$h6H<V{(MF(3CK;VTTweK@3;wq1HRkc=2SbNo{S#t=6DrarAU@?#0A_5ntwmxDdJw
zRUIgXa>aw?veR_Ca~?VKgV*7%FkS;G2f6eN7A_H|-Zb4y${bK+8Ib-0|J+X+1!{@3
z##pBrdwZ`JD#L>r6Emhb+b5@d;=Jv@-nC3;jsGyYaiCF@45n@P1v0kK#?cc6i*yCM
z%+t3BU%VI;mL>gkY}$xL7HtKTuCnxbuUQ-mS&;+=-57M(y504rM!uX*9oE6`Qr(q#
zqj8VI(p?OXOT9R1+w*#IMfP!gm*siKr_&hnA^h?%G0CuofM_whZmr2#T9|9!u>2^Q
zHaj44%{naUStlDiVMWB<k&^`~29*gW&k%V#r?YP2+*WxfdLpLgHx?gIOu`?2P@&Xm
z<fdG>@o*_cKP}TdWYt`k>0Zu|sWmB<mKJr4s`r+b8N6(AmWaJM`*>yID(|N0V-Kqt
zGi&rfJ%I)O1|Fc+ICVX73rpXfOW94%B6(B{(hpPZcsgfFdF|yNhMap>j&--mvUM?*
zvao!fseh7UY{>#$wtWMp;pHg`JJt7dqlV%#hH_ZnKm1ia@rU}5)tMO55PjS->a!0|
z=i}Sg?jK;d_XczK^E-2{yq!a(=KV-l<i7hQ)P30n(W@>W6-ie=zpb2zs1Uqt)k!&g
zfME=lNOcB&Y}aQH)IBg#M(;}f;pwL@!qfB%#~jAutCP;XQLYFWov_dJV>r~D6?pA#
z@TDdnN@6RW^)AIdEW#{aml)r4+kcu#_p#R(Pb)VAM>f?rRvOwZWw$mUe+Nu!&WIkV
zh*nS6v5EAj6{#fD%T9?Lg(AEP#oo`BRXZQBv9*17<~`r7&d9QmkV~Jwnlq1J3D(P;
ziyb*{CX%OV$^vUaJ`tt^1@AOcMt@91IK2&_&9*sbZ<vuKds0<O1)@6WVua=5K2z)z
z+qj`w-A%;fqw)h;{=Q;y8pF&_n}yz~Zg#dI;~V+O`IFk1sp?xkOcSxwY{qgWmzmLP
z-}Zh$Oa&ctaoV(D)U;l|6U9XTiYJ)nBy8s-Htt@i7CT0`xW#BI?<3h*-cXf#@CeqO
zVf{&_d?)W=^WFO;WxDcU+n%Mfouu4>C*)-?qg?3M(wH(7&Tq)Mslu_hBC?Rea0Bl?
zDg<xFeN|nW0H?h%iF7<dr~<78Bjil+ZSN{AEEvhh^=~XLJ}9(pTtC^DDuYj*PE%{n
zWM^U5c?@m{23@J8Wk}F#QzWj>^c$+DDoETxZ-=Ehs6F)6lUwi6%(PW-u{yk#dBMuN
z{AgtKK0H-mhHivw#Z&fz!>B%E*;TVI()2TPQ|h#6AR6z@Zae|-Hc?f{$;6{m($|9U
zadiH{Nb68^iLpxWg4$8MF`wJW!0|kmS!-z_38Cr6L2#~vEDbwPvLHK8;rY<jnmWZ!
zxvl)&kG?5l{NdLg9G0|T3UZon5{BlHUZ^cq%q=WQIt-|cnyy=aTcvW((Dhf7enCHf
zf*E37FnD{GVzwY$f}^NXt$Qm(4*!%lIY)S2<KyM!BVwwx{?S?4knh|Rxs|1pFAF6z
zAfeK(QnJMV!`^#GH5o<Uf<cguNbf``q5{%I1&LTFA_hUFDNz9dF#;+OBqT`hRk{=f
zL}{UefJC}T?;S!wl%5ENXo_?Fy|-q~yf?Gfo3h?pv*!KF3i-Z#-@WIav-dvx#JAyr
zq>)!!Ey|OAm4vBWI-$dpeY7midURWbm#ENq$We?lt8M%4EbxB^q$CY!SS>hq)xc{x
z2C&^&lp6CR)dOI=Gx&E=Og|VQ4|dQ%6E#TLC#A)pp@j3;yF;q^Bx>4?ijnQ|(q;8|
zEz?F+zH3)&uXD?d9^8*7fw0$jsw$9Qyut#KfWn0bo)2q?A^!IG-%2@RfcCqU#*Da1
z2gGhrJ6(qQk*wRjc*w&*3MGGs?EEP%!(XN&^LmyyCZVY>^tMrbY+Dlhg)<7)+x&sZ
zauqt)Qm@{Z94^A><K?8Y!X^2oCnap*fY;_IgkS({*6M*@$v|s@t90%#lgk49fuUg9
z!Z~m)$2AP^$g<|5JC?7oy3Vl2?fYo9oH%qV5pf>ddcQ@SewCU<+FhDYYxm$<XR0(*
zjXzmwh&HWgz@u*DS(M4m=BKccxIQ0iVK5S&pLcJERqeDb@>ApC4I{PYu=a_=b2E4U
z5x*xdvnnK-hz!3X-3kOP4&qosM2_Cb-k=oSoa*KOG!fxmXXo?AO-NdTJ!iMuVA1km
zo8UQuAUjL<o-Ja~W6Kd3i1<>K$i?s(4~nK)R*3&c9ijX02<wv0ebc+C;!<*}GnRe&
zg}0B|bU7Z^x&apHK@&#(+fn?IcrF2LN&Ub+sJ-^$E&c21q>20W+k;AiEIFr#*azvN
zwC={6!As%oQuc+(H$oQICp}qruLwmD>U<~(r37q4K*<^|3d^N40Up;CT;Rd1V<qoU
zdZuHmfeh-*R}&rOP<h(d_rGQHqNp^AALAvM3^Ma<Srap~SXXt&)(xp|ZRz$Wf-B$y
z=kKW5&7SxwyK;!ry|VgmH>IL+d%C`k&RFNuS^fv^*xfN2>gA1+)Q#<$FtTd3JjAIl
zZ~+<(BV8Vpar4dP(Dx<H%@t8{lFMVQz9{ii+&5hBAL8)lfBbp*jr^G}gIl3kU98Ao
zh!b=sjUq^$e@5=57}qqC#X?s7B<|Hj;uM1>+*n(310ELi>a(&6=AGe|u6YV^F9T~a
z=q0w*(BPa~8L4~8xJ`@%1jV%?G@Y(OH%F9Hor>o`{PWAA&C+7&b#p74b)1l&@e-)P
zYvf6~s8PGkaJk_6z+|VL!)u(x*9`3q8{|WYok&No+$U#n@3Dk>rU?EMiKA^%>WYN>
zd3npR$BOq~E3Y4JSMmE3ZG?+&;l*^ppcoF!+~OgmH;!lH93`t@{l%@-eB8`h{nIrw
zoer+^U&LB$V^bw$^4)lNtCBA0^ZzJW-fx9+GKXPYzji`ew0Tp{^<C<1h+$PN>vPiA
z75EO-e!|k_*?tmp8#u&-)5I_xxfOpQ-HX#DtuTRFEKBEyje61ChvS9j_BV}~L$!{=
z-<IRHEKfguYhGNePI%9KB>>1N4>dDFu|G2S>C%tD=FLJRv<Q;`k;}p4Dg>&G+13Tg
ztS$pC{Hq?d>8)d9&!hYJB0byYBFm1=K=h(qAW)5qnR^In=0a}$;Fj1@Pm2^?fokl(
z^uasjbeirTWbEGL$tu2b!BvY&d)DL(sa3---tyWk6vXBR0oVg3p<h%2?9PSe8sC9G
zH1D8VT-f#LSP#<ja`td}fxE565%g(WlU7Spk#8`P{orp!7$0_12SojX@hvFH8~`5B
z8;rC~59TqnrE38-7-yel=+09*w3d2hd`3;<DNZ{k%E?*r0?W<Z^m>3l9Sos5lYtPh
z9Ue}B^3ef_U%uLF{V^3)lO+|sCKp>q3G=QuK3PEGf;<c~m|}!sSB4Bg(Kz)9Kpc72
z<0-qwo5;M495F_!0TBIl%y<tZ1L>hif$r%v&rhE^1W(d?{v+DYsk2$Hz_#4T?UZ(^
z7wxI>+qD+}=ZA)!tB-iDEOREyPi|y8)dp!WD$=|o+oQR4j8tx-7cQR1|HA<IaUZ($
zU&ti_@Pr81Pb&iut*7Jwgab&pyd4?)xw}-Qj?<}$+N1qNcMH7UruzDX1{J?w0U}Xv
zTzfoKez0{h8qeEulz#aWHJK3btj%Y?@$PG%pO0po$AhkI=!x3Q5Vd(Pth`J*2*FS%
zch$(FOe?*2;3|_#YK)^5xV5+V9C!>KoCfQ;q}$mrqEQCnJZ7iOl!)LNnCUoocP!I0
zY+<1DVZyx1!^4SH>OS+KCk<QK1JCO*d9X|Nv_%M7m0WWpwR%KH)u`>ZL|e-nUd`^@
z_cf0;un%^ok||Isji!TABA&((&eWdwA&zd&)O;v;6tOrXTKeF1v)6pF(ZEg9+FM4_
zA}l7j(ZTpdHncqLGF@r1MUXnAX|tihe~PHiTUW1<K$<gtjUH~`kZ9-q@Y`l`cXkQR
z4Niv(SVE?Q#&}t?%+iNuM?%)`Ye|fp5u~w6^XI8FUezy|Bgj0~hIgJUyTT{(_$P5x
z@9&$+e<7E@U<u3Ce<3R~WW@Aj>mtcaf%?m_kU2;hhDKs!8k_BJnZ>>^Pns!+)XTqX
zGEtRxf%>>AF{mI(c^UqHaZBh>cIZM_3db1MnI;7wrd;9769z{%CRT^++BUS$IVXDi
zMEBh(>6e$wkTA7?+)`RqVfjr13%ZDe!M&lQNopPM1@f!L_Y|w6EY66%K#o<N8GLUR
z5h21jj6O|X2q(in%gO@Ro|4dyqxx4I5AGep&u6eA<*QF9+Q=HR$zDO#Z+X$wQPvb3
z^@<<qJ0gn)yYaGfA^CmC0>$@zfwI7OzUNhw3<;~mFm7Wov}PX0<W8h<pPy|13t^|8
z+vsu-GtD$Z9jQul7`<+jnzXoCT7Ot2WZTKWR!PM|188SFumHA;urqFlPP-62B5U0X
zYc*Z1CqzSj`Oi}AW~PO`6?)~q%Rm%JT)3||>d_)P3jGi`kp2UXhdegAn5uL&1|H@P
zgP`(>xm)`7pB>mq?2*Opw`Z(cDH>kBF77IoZ*8{S%&H+__Q5~pp=t^V0VPaMq=(!O
zz%^}0{8q@lc){OUbT5~|X~8jouT$#;=T9AO>n+S#jptF3327j?tT%6FX4~t@=yau6
zTD?j5erebVZr1%97c3;Qv-Vs*LrKOS(`N0Wl&xh6Z=A&4;`lRj3juNSvn^k?L{A97
z-=st^fsGOjxi|&#pfIW-?V1_Qp4tEB?gHN3`@L~#)zl?QNfY1Y+{KcaJT+~>*Okt<
zA$rOLNdNpAlyq}&X4$S5&g+J#F5O>HnUPoJl}x_!-SI>9;S9*2-nBV?;?4zWYr4oY
zLeb_v4MvwBi3Kx-sF$vhCQ3>+Eaz$xj9Vz~_GT`|f_J_;_+HrUZa*v0x`B8PBVA|b
zqY0r^<eBIv;otl{mC1vYFh(r)2{aA^0plE@aqc=miG}QcE$I^Sy!T;KQ(T*EP*vx_
z%H*#$&HYTf-$??Wz8!rzxZz6|GEkFr1@^4)FuKq(n3M3RMQEZ~*&FMq*n)FDA5+z>
z*goH>?dBQLFWb}h1QPqD{j4|ynzTr=I0YjV8Sp5*q{M@*xpsAOuLltbCEdbKhcBmA
zl1thbYY?prc3OvM9Yr+Z4))>2NoybL2_aJjD{%qazQ?8M_ew<)>j@%*lTZ?TXU3gU
zO8rKJ3;e*=7zeEm)iCtnVeWjFXp%eTaefnblk$5sG-_fbY~+xiZy#|y1EB*;@gVeH
z$XD}!UbJ`VR=|yu4>95rNk3bFYHxp0AJS|HX&m|0WV%_tbY3B7RZBVwQ`})q;Mmk6
zSuHDt;0bRQl*#o`bPn8)7MYJ(9bu|h_stsXsuJLO;!$&^l3nQ5ulHN-kLob^hB$VS
zpmyFMI}dIq(jK7A8QCpI7oyEi5qYQusqhrZ0A5q?gM(G3F5k~=Us4+>0Qcti<6Q0*
zg64~wrR0&#n+$T7*2;BkCg2&4jd(SWq6^rK(8iBeIlN}S%9nv+nsyQmpI~bIfJz#0
zQ>|^tybFq>1r_6?m#Z3@Gf!+;_1GR0{n01t@5XkAbosJu11t|qx;8jXv!rs-G}kJr
zCD*d9lC7i5oyULa`KM<pN9YojQ3}Q^V+F@cZv>v)Cd8};8Z=TDm;4u{(PHHNpuO=$
zUmO*CYjg==d#X$!(wJyty?Ye=$`-0FFcR~o77}q`c-dx{Y`x#HY?;-BB!~Iwb3MqT
z_US`*{f8({i8duqju@GI=0l7#um*o2b!v2>jb%hTEM^3>5LJ%A^^+V7KAg%eN)h|x
zy&00>Fma5CQdH^a$V+T^*A<_~@xz6FZkGZQ!M9E@l2-}fmXPQTauEeQ>Y%HbLCJ~u
zrJWcDXn6bNpG`}+yVN7uqaRg;uN`h;WD*goX3{4D-b6eWs#*RnqsHWpr6JvJv$iz?
zce<qjV*0DgibyT1dY?r0k&zkRq>i7h66{YQtm5Ac4q-Zg40UI#h}@_a537?YTL@mN
zd!A(P-$7HH?^BoVbVNQ}Y_6+c*><_5AgcUvssX|b#EAS@U<b)&JAxPN7T)_m_2oT^
z?D-URuT53Cwj1`hYUU;c)7qORGk1#5yp38=2x7BYjs=As%dHEaA&0<Nst-V1y!gNG
zO2ty;bffn+a3PBV`Uk4{>(?6R&ShOj8JPd7x>$N1<8H}cB(bi%=T8|N(peN{e#HyX
z^_F7KeYYO><`Z01b~iiYKqUH!^D1~5NzBLJ%gbc?V_o>CM=3w3{7Yeg_~WM{ZjaKh
z6g<+EbjV*SGd6DsTHn9+6S-=8<#pTHQ+Y|oPWKNB${R^)nmh=Ej(cO(5R$l~|Di1Q
z8m(F6zS<a?off5SCPxnXZ2Y!>4?BuW9U2^jr<I?@*^&We;z1!%%L>d%vrF}D%DKGM
zU<-J4zwiB`M`F5UQCk0jFK_Oh`7*cujon5wK|J8!7FuKk>g)uH`68rRPP1VnBumx!
z3KJ5pKmYP@)g|?o3htTi#hVGyi65gYEDotLc|a+vU<S&0Mhqx%3z8=1@fB)-|7ye1
z4s-(k>k~R29G!twcd``GrH!9;Etik3Tud$hIeORcUASpw&L5k%r>jown+SX@H~Ehu
ze~=N0K98eXQ7nM%>!9#j+P@;d!1#YH@~2XgP8?{|t%kU=k|_I3t8;jY&J}QJWyC0G
zW=q3@{7IKC@&Tiu7Wiv-NsVgim`l6o=8KqKFsbHtqFJY{Y|eLWO%2Mp|B=e$|Gv-U
z$zs+MaI_YA<_qrtFPa(3lGX(yUDY0~nd(au#qf|O@&ma#Uc8u|zlPe2($ygV*2NYU
zvM6G3+>8`K*CsuwB@Vxf_1{8FOM3SM7e)F`g&)RadO0AAa(W)LRm6qitqr(QUg4X&
z1Q;h3cinXgWp7rAuWX9QKSAU5GUpvW!zS>xvEs`LsM6Aa{Cr2&&3cht--|Rhe{@_s
z)|UqVSEbo+xIVbe+8C)iCH-8%MM8p6hFk3=VmDoA)|j?MiSN`wn!KSu5zf;DOObhH
z?*B}rAWo)KZF5o+&N%}^1v18;Iwl;@QwH54`&+%l{u9kggvFo|7oy)-+vMp#w+o+>
zn_`Q}b}bhhbBxoS?)<XUJkI(RPpFhV8Q#5E3k@}pL;=RSrk0P#xUMFZ^P~5ej#PN!
zy}yuvx9atDUh6IB7B)Zr5VIXBpl{I7m9cbI)#w~%RpdJCB%RC%#}jg(BBxqZI%zjB
zlf2*Wgl(?hw!1s`miRN^LIr9^O@iyDfNsCE?R@izdoNGduZwME<ZZMZXQVKXqx{>?
zc@N-<*Cl1fPEVwIe(LP_sYy!g6nF7C9~9)ux{GCU0#rr;v_>()Rq=c2)|+6;^=~Cv
zkX%-n?=vG_(5perI#1UEg9&<<!sGWq<|ivW-h9H_uhhPm&8{{EU;&b|p{JrfBWw)`
z37q|y59o_^A3u@!BOh-P{kD}x)#T&Oj8!+)oDu=!R25!B&VocMcZUiBGW``S3P#}`
z1>`O|g&{skD5482681X>!wmGO@qT2bR>|)ngqG3Aj`hj6N1v1mx|kNIRMvL5iU%4V
zaS@?SUV!#Z(uFvfmVC?cf5?C+PP>Ev6j3YKk1LUA*k4E`KzjC)?8s2661)`=lf%jQ
zh;}3{N}~7?L)ywE=e-`_$F!SD>`f)zzhtT>n@M(cjtS=J9&z$Q0L@S9Jp)O)5jBGp
z{P~YdJC=LBMY*|V(!uS`$YYn5)Z?+G(Vs@xo|YuBn{<J(_isSmpDR91|3|F(Q+0{#
z`BLU?VBqlFqhz_)THVgl)m1?+1P{MNgu_@c;%GSdSRr$$MV_qNi5G<z{g}}<^F<+B
zy=%jKPk1=r7fOvgauJ$Q-Ij{nHifl7t9;v=rm+ul+30H2(89ZGRt~7A=DIhknnJhA
z-<rKtcq-q0GR$aZUINQaKl?AB!)IYku55DkEi(Vh=A&NAaGpmziq&^!XUzTXzn@ab
z%6Dt&PpfTk$j5Nk5%fEhJUKSRg7Ac7Y-L(VbNyKJ`S++|SJI{R##eKRbA@l}Vi??s
zTZkB-4)EZoYmkh$GV)Q-C5PxDD%OlN9~X=Cax1!!aJgPhd9Knk@JT|qigQL)cTA4*
zwha(iyjiS=ck>{qmzk_|wlw4M9Z1Cbv(9JUe3f4D>ZV>YpK314-t6lOmJY*(!rb6g
zrSDsC(!_on^Zcjkam;ZrmzLl6{RaH))ZIte`z@vdlztVH%~iu=zA0B!%1aWOA~^OW
zXdr9q$)bb3R-uCbBWVp<mM+Ktmct&#Zr%ss)KnhV0K`LuWZ;57mE83dY>v!icg!);
z+_Ovhj3mbo3X?q2Ve@4{el3rjN{;n^4(>$fgUF8k7loH{+n~~Gxq|v23S0i%xoi3+
z{B~tdm#L#@W%-`zFwZGN304VWAnfk@x4=N$p;_%2^s^dOMdaIHw8GufBfFh{vOE%q
z8vKz*(`Pl#AIPl**OwV!!L5tpm0*MdSjDjtn&~AMvilNuC$vLGaB*|jPESR-c=qVL
zWTZ`T0a<WX;tOw)yZTH6wYwD%W%$_*LQAArJ1jDntGwB~Z0#$uY{y&m98-Dvs^`0h
z(JLQ66V4qxrQSJ})WNz2Tm!;m_ImkW8gP>Y8Vo;d$Q~8jIP`RNrfUMV;;9GO7l@%P
zPqe&Y#G)NtNIOif#U=g-$rz+gQJB8Q`~S3pohF6;^9qi_*6421Xx^A%1Cd-x4z+zL
z)^5YLxutGYJC}FLtzK|`c5OTP>(A)pAxX3L9qYxsc|G!`9|z~vXw@`X`Wd{(RCsGH
z4@PcHH}mp-qsT<|`>BPCl*t>4V7x%QaIu7V{tCMT0X$9)%mHBNnhVZCGsYXVG0`I$
zqmm0Bu5nG>Cd~$IxbtpXE_`iMQkA|C0@HE{RhgYj!aC}M%!@6H5!0;rt)*~j>r$tc
zaS-&@*t?auZ>oGXE<bL@KcZ*<-`OI*<B1?DT*5!ejT6O_BIz7du2b<P3oHU--=kL7
zr5-HH-b*6t`MXBwf5^ySKg)Lk%7<;e(hMVYw?gY4#uBihFHYqyG1S_9&gs;4zRNyf
zC^*-Qd)nE?(s)U_GWB7vtR@P#aR{I6F=NH#`c1Q<>wF^N<2@e#LPW``l7r{h&yD`5
z6SK>m@<_TIP;DltcQ(N^``ZmO<>Hl7%M2?N43XC9AUKSMdlk*C39e7v-uH0!6KZ|-
zc(kkhe&*Q&jiwTMwl)VikAOAm0a#+7J-AGN!q2hElm}G9)CPs0$t^-NLWNVD8LxHt
zB~iK$Q1Q0S!A7GliI>W1_DvOH4My9wy{;U+@Kkq8g?k0sfsb6|g1PTZRa2TkBm+30
zAbxz)<Co0U@7M8lqu<6hPpRlgNW$d^7rBfrT5WCepP2Wiw8GvcQI!C5B`i-78mocD
zypRX@ipq;uljqRZr^O_n^LaV>+Uuq~Yc%8?zB-h_@>tNu#m$n1#J>rmI@?w)qnxhj
zOS?e#C;OBsrsDkdZx7q`l?A+@Y*pEApDMTXc-3%R_=ICkcKb_2nE^Y9wXT>>Jb%u4
zLw#1!+GogJiQ23WI|-rNBohGc$xmunuRuS;K#-`z(S?3?TT+dQpmW!EW<QMB<VOti
z_YX`MJf)juMth#h5p~TeQ2NQrVKdgfQ%z(XlUs@A?Pi762CW$cx)SMO)jK}8b+dg0
z&rDBBqQRtRW!l9e-qfKC{v#EZZGi@$o(n>+zc^F|7PVULEmDVPdp=@;`qo{1Ysi>;
z)>Njqt@3T4$N@)JR*0wapA6ipFU5zE#?)ed0U4A<Jl}-{elmYlGqhkKi-XLU6YKN*
z_r3AQ*W?KOlBR+Rn%8+hCWK72#<NZCb^ceDGyiXorTOpiebB$7|9vgJaD9uh05WPh
zjTu4IDFv3HFXbJ3NeU(~=FSlxnR`|l)M6HH_DFlhXojFV9&@(6PmE3?mtivg(>RF#
z@{vGaVF6|m0j<He;<!Lw@EwTE9b6l)XPh6;X;6I|MLhBZo$sbS+%Fwjp=pv8MbIk#
zuEMg&$i@N*O*J;2Dg6USw(6`z!Mjo?Dmse|vAl^_R7Z~39t*M=YI96{>3k>b&X)2}
z?x@Tqc*S_B%czdMD^6uQ@ogqoqEnQ~)cus<m$xm?9Ep*nigke{WQL!@cg>`9m7|6K
zh>VZ!B-2h(h!fJe+x9!I^3ct<$0>RIF@Ru0kSwvmElO0qGP($rr?@^!$3wJ31?SCq
zKS|llO!r7%)X3WmrxcFV11Bs>yL|}441m_NmH#vKCMi?AldZ^|S#E#+LgH}PT8D{S
zCv-lgM>^KV-I71K!k+Nq+vS%gyqq`}>LnV4*#rKuHeKJGu*hoYRc6=N!?#_i(il?F
z_!M`z;bELEN5YS&&-=Kl_y1qFHTxfR|E&MTjQamk_whfk8+H8u$unU6KOS!T6A|B{
zJVECoDG_0D`a0_M;{Z_k<anbhCDF-d8>#Xw^Vo4h31s~c=aD^q3y!iIh=KZ6*z6!>
zol!_X??Un*jF*+|cStCGY#c|MJ##~fl;;-FqYo6bsUPJPPeaY8C%g}dne+#0rFyyu
z4@JRDEdnp@9V<^vdwFfRJAeY1_IsLDfmP-*WrZEQE1R)jP{H>^6pAZ2cAQ04@T>+}
zs7sVUs%;+KhNK_!y#J{T^Ja+O#O~1(+Jd199I>rdafdWDIdy(V)677MWyn-@$4VA4
zQt=N4`Hy`f((J=(n#T)bU2Mva!timNxrd7{7)`spe#NHF6DS>s`FaBZRFa^OBvfHJ
zpS!gvEuu=y$Yr&eQd#|S13Y>k*v$6Br$+|nvsjZ<3smE>H+DRx+$jS?_{Y%j$;{Ui
zH-6-(k#@Tel@%F%TW0(?rkAdZUbxm|1DSSmK4~gt8)!gyi<g19VrS(kuc*N^Qv{|5
zQTZg-jlKSjyo|kFbKD7ZTEvO;qTCkorSKQ|)nHJO%=`RU<}I=)XxrPOX`vqM5O*S*
z-QH67FAPs^*`;JRtKTVAKhfKxf_1D;H^O(_!=^m#=H`Mxn?-&nBWW(w-40M}DPyPg
zu8ks}4|4x-^QCMA(>9uZTbt{;?W*(ey7u8u26dF1Z7I5sx3Duu9F8-}@J{pgM0gt~
zat|7Tc=%~FX~9R(-p)PYvZQ`ZRO?$L`HSv}Hqftd2+NMvEo4srOGCKFwYQ))BUZ~5
zyUbzK20z6Ql=-e|q3Y>>A<MH6%+8Mg?gi+Cj>s-H1nsfx5xiLa2t}imjw7Wy3?wTX
zn<k%>25};KQ!fd&y|4V>6W@EqBL7*!xg4w#U72PCUZ^qt$z8{rfWUv4z6s!v18qxX
z#O6kut4+LzlkW8MbtLtP1jD9@Hyv13f<8lAgsll(`Lx?WY*Voxy=Q}U-%v(CQ5?4E
zhKvu6(ns_4QxD%UlxFR}b+E%)AXqL)B<*0@ukK}zj)_tHL#!0~0oDD(6}=ghndPLm
z|Jbe6W3HP{AM4VUgGxVgPu&kXBNupnrg?c;G<%s3u>_6X%MDNJ;fTULKYx`^&ZJQE
z=dm}6rO}UjPMnm#f5K8aL3lSYiit(D(q3Q|U^U@EdxIg12as9T)QGCNg^hE4qCOU!
zDrTiVLvPYmOD}!!lYa8J_zjy5CK@eF&fO{_>`ZGdLt|lXJxQZo3xcDy(Xo65b=6w@
z6>Ec@$zPCKNNr&H(Y9MzFZv~q--7Y7S%hXz6<a!0Ux*pE-p`DWdN{F~IbybTMdvJg
z4zg@q_-fKs5%l=0%>hG&<s8g1ZU8|&2H)Z%TTzm!P$E>4erJq0?u~?XSgnzIL~2gI
zK`vL<rdT_jEWZBx&PvQ~TwAjkB@HB^!F0O`rVgB@is{UeHTUnN)bqLdm6%EnRNlFW
z#Yzhp9K9JBTQzw0IOjbv$XN*E)muC_+7|R!wmLBn{}@HXTgbQ5mhIiOb#*hZycYGf
z<V@S)uP)4PI?bVAZ(=W_H@UD_F}aof1~UcY$LX9c%^%iBEPaoYvA-VM+f5s~=BxhX
zcXNhM1xwYJuQ!uUS3?3?uV4h2Bf$9&5sQGp9)rw@FAw!=kjIsBd)W1l=5&XH`PKbv
zwW)l&a*@*qJNh=?{QP=a?MB=p#bq2uDA~`I)>jT7cO9#xVsYjwjY*&ueW>S-TJ}4@
zxyAXh-shxzJcfPsVEerxjxAMT>mFD=yb~;$!;BO#c=XazKhcdTH*Azev0Vu}J8m%i
ze%8Jx{Tpv=`pW$;%V(v5Z_^A|3oSyJNoYM#_a!bdls@CDr>cfLo=X8dH6*RGo<cI_
zO-uUHSJ|bX@gVlH{{@pJx9dFFC3ZQ-8ATwm!d9fkXq#dm`BYL<NHm*t`z+e=w|5Ji
zstmoPPnXuS7<#5@B&>5<s`uqVKh;?45EcQGPvi*M!%CyBZ+Yw?eI}%4O-Bn72Hp%U
zzdxfN{N*f1;$2|4Y2Asa{~pKGMJqG1YvEDQ1Ku}By|KJ^X6F`_XWD&^jlNEQ$8yXV
z(lwAID&D4SZABBNzi40pki&(a8+~=W(A0Y}cC!|%dJBTD@i=vR3V-r<o=U4R<P7A{
zE0$l*BL6|?$}m2FwqZ3iPyK}cL0=Itk>2=yyW_j&L<QzeFXo?BkCPW}1YQyNd5Aa4
zy^7S`4HZU<Gotqv<6$-9zsJ030dnW9EKKf2&6p~+cuu663O{;Zg`pfbl@IIH=<;Nt
zgo;o#Ky^<1P8X4)OpI=0j?yLZ!x(X@tnIh3{#BC{rJr#VTq<?@zM8LK%3t0mrtu0p
zU&To+@66aUxvglt)cDI|cvtsqSa%EmxYEzthB8g|eP4A=g?(?6TJ^OaA50#&Si~0U
zSqkCUQrm)+ZOBS6vM?u6@$HjUYcucCmI5BGPpr@PJu}ol{j~Y1(MC{dDa*gUIkYp#
zQ`4v!WF8`Ja}h_4CL$ts=<s4gVo9&fwzpLH$BZLp%1SovUrf0rre2P!z+e1<Q^LqT
zk;_1@5Kq*NYt}uHYf9!1ti5}2_)DwnshS?en8c~A`-xLzge10A6t*(n3=TlY*nMW!
z-X93V?=4;)iFN0X1sarAs{%$k)HZ~!Z~56aAdX)vo}bZ-EqJ_(U0z<L%^Ntpw%$kq
zi(tiLbDCy@enJIB%t2OaAxWq+vrpH7htxWGFtm8ORj8XH_2W&rmK>MVqq7{(q<`s8
zHh{QR$H?7GUMf>ICUs?rMHn1G6=5mJn8FzKj*t@#+46EmmMaRbh1X)6cu!|S7zf_s
z`JUrAz+JgCBMXcH_9|qq+-`OKt?P5WkczOS8y|0K1Vq`4Elb$GThIz?{UW<e6T!^q
z9tKK}>to;?fX;*)!g<jaBeyTHMMI}2sGr}gL@xE*%KFOqsqNC6s&0ERP&W`$I()@1
zyvrXOrHkNczT!c;y*e}4wc`<c(!JsxTkeaWOXHJLzTQ?qrG>5A?}E<gEn9Duy<>Lf
zvZ1ZL2(u=l6DOQ%Kc@B#>QfwVWV%P@G1lUgpK2T?q$sgfbpNp4yh~F@-($SC>NF+W
z6JvKeoF=X$O&NQ6`6}F&<3H11rI+CZ`Fi*49j7Dx7=cr1w1)s%?Y^F_6RX4&+lQ?+
z-DeJf>3gC(O)I|<<R#F-Q%qyhCwB&4G8NIc834I@_~{wZWvjZ>_RfYKTlJv!jNO$g
zGvDK>6_+auPC6$UP367p<*kqo<ga$cwZb?t!;2(K#BA`5v1UV>?yTD)FP#H<qAqTG
z?dbD|GB@Y%*bgdH1!A6ga2I<X;`&{(o?DCG+NbeTyNv-VQbY5QMM168zDmXxSSWiX
zwJoeJoDm!NcTQ@28<Q51aE37U-hN!<L6sPi@u4+%3998U9X_aCZPd>JqSmWau7q>y
ze}c~F$AxDyJ(!~4{hDHt@Mu}FPt+R(anMM2lKO$DJkA{WSU<WYFDi)^5Y29TY-85s
z_&r_t%I!f#PZoy84a93$MKW_Cah<6)f_;1&I{gQI8b6KH_EFUzy*6;C+p8#7bvf2*
zl|OZEBM2ggHwJ|r{tW0@%AGX^Bxdzfs`U~ukB~eMXqkA0pL^8rmsHe^MmTRY*E+c=
zrrze_{{G~W=)i&czG^y9R|XC7=`n?llI1V@uNW#(PV%s59C7(N%g9kR!yZD@vGogq
zFywn&AD{ugZ}kutXY_88bz6awU`F3gtpm#;Z>F0lb)jCp{dj$D-^(5iI9uWE3x`IV
z?_0B3E7LqM3(WJNyI2j*0Qt~E843MMWG$<osYIXXEx7@+lzHL9?FnTDZHf<*_r2a6
z40f&qJ<--gkIh_iE^#snPk09vXG&3%yClntmhQ?k1)-H|zG6>%T|6iOMx4Z-OAn7X
z2}G51KoXSCc~H^N*1LL4(aEUV0A-?V+5>~p52zQyHAyR<(=`v;``d`C-$|tSN$=0h
z^MlsXRvqS%88kq=gQDoB)a?7Ne**APHT}=5%=~Qh+Ul(-RhthTxb9AU*_b>K2+NP2
zoi#r_5Mab~s;{k$1;(VDH{-OWGSYJ(5JA%)%0Cg5o8<bKe=3-nw9ed`(#%*!d|Ph#
zH{eI*R-#=oyD+brw6A;k1Lt-nKJ(Itlg=TmwBkKrbk*JE{;$2cDzz}*wfeVYQT&tT
zp_Erk6B9T6o=o{`-$>75Holmw#k69%2-LGQCHi^NV5%t1VszBH<xIQIKoK*k_0zJo
zO-saa)=h3@@`T#V_vP-kMe6Y-#_N_-8=ebv1F~vQv{8R{^P$KO#;a~CW|GPuP4UAv
zm1oZ?DX=0cZ$hkBCbvq6SXoRDOqMQ0yG!)}=z|4Gv|7$p_SA65Pt(UD-P`4FLbD!R
zb-XTqP}b=x7h~gLQ7e9?X&DRp8K}p_qFcnMO15GAAqMO;>lvcGiOjY^$J^29)Qf*1
zEzJ)$YQr8_#jUn1GH@87c5*evo%)irmCx@TP}qMy)!1xS|IX~Qv4=Ke(pWC}nZWbA
z7bmrIg<LbSk=jxPp8zm}D&KO6`KAcGuZ-3MBuwNt>}r4iLrsFI{Q+bs#jSxPTMNs5
zkJiJV77%<2NYP2<?{?d;&H60KKXS*n5GBw5TO4oy=Vz`6|9b{3e<%KH<Pdj&QG#8W
zG|<3KWf7qaUAP0y;o!e+&z^Gh13($-cEJ8T$q9t%!wAu`hexyrrzbldw4*@mtMwif
z==IR<-S4kWsgb>QEkNX33fO~B@Fu|Xkk`#XlaY&2K)Kt;okglOpZ`$emwz&@Zh+jg
z>F8~3H5KS@y~IHmd`9V}`jQ;lJ75P5PCZIQ`I=6ufv?v7bp6L~pJ<<<0$ue%jo*&E
ziI>0hhq9?Z;5mia0^PG7*q|H%su<A&Lf0Zc%qH%Lqq@3~rxsL?t?K<KyVnFi>GwTT
z)ahcSo}YU|^;txk>jQ|sHIs8~`TV0^VA@yWD7!`^Z5QlQB8-++`<-~?7}vPH)&u%I
z@ubVSho;v5EOzch+Ev(p^2S!Hj4b`mkZ*F2+<qFMh$1SD6$bok_rWuAy>Tz`)iNK+
z*~zvWExJts4_9ZL<xEt+#md!%T{$zeVCffA;CwCAKJn%6LYfAO$^UP{z5nuJP_CrJ
zCS8JvJ9>Zo^0hDWVW!aQ%}1QlW1g8x&t4Kv4&IjxgcUS0{qRm3X0TA;(l!{uJV2^Z
z_Q+o^x>JAHutU9rrgQMZCrIqj5t+tA%D;Sxx?791<8;vE(H*mmQE;6h78seRJoo>h
zwo`8Zw!rh^=I7L9wOZ@!K8Vh5zyq^2kd)#Kw3x%W0x%?8(rvZ-`q&CS?hB#x>EgN;
z*D;%Cy~apRjpTKLQ_4Mv1kIG$2i{9JaImsM1yd>+uSylkis6yYIosPa69Rr-QVyCu
zzDO?@W|gZ`3cAD9(0O^gszfKA3)|{gj&y1u>26&jMQ$~(xCdNt`+cYT$h5*!x5n^W
z&I+0}^q3gQU!ORb++1X#CH}1p;$jf2_ETq0C}G@<IE*7yy2@Q~yEJ{Lk5=$ZnB;8x
z@??_t=R?-(&>8FH@0(B}f5@NC?oJOOD!hxrKHyP3Ju^SOZWmJi#fL{Hv>^2?&pnlr
z8u|Ot$0YzS4ReW)K(Jzz=x1uk9-mo@io#FXga@@P=z%FR!X8on&u{(wb|WTCI_B`z
z6eaTH=iHxJ9*6%zW*3?1sZDSo<q1aLFGTSZtz&D?6b839GFG!aiagj9MyJZUX9{9u
zi}imDRS>FD1}t6@a}<EJ4;aD#BiU?WyB39Zx+3$Q%qp%#Y)L=ZzZ5I?a7ujZRI4fE
z<rv|>1&%2M$#zi40SJ$(CZgZiX%)LKC{~9({m41g>ha8AqMJigPMw4HaSNf289>yF
zZLtEN$~PDz4cf`<CTp|hO-G48y~RZtjxD>@T6fjbiW-SN&yE~Tzu!XMK%g*lEo`GO
zBy{TQjy+ZJeLL}f(VbX-Q-6!Q9=YaPvv)3uL3WOhp5#RjyzyT_h0}%dDO9SSKgsF>
zRfqhnBh}#$86LDXc|G)(Y3j(t&#(MR^Z7Xs<ZT5sq>TE+Q|9u5E(W4UX%?6XL{0qG
zSrWpY3M*xPWy+7R7gI*sF?D}J+h_Wb=dLQ44CG{}h~&f6f7|RLZqZrl$aTqJL8g-j
zL+RN3N0Oo?c~twZ)448VTZOd^Kd<Rpb)Y|{*x$3g_@PU^K}H!tnFlBQE2`2GJcK~P
zZGIz~>vsNy9IiEsxTR2e;)Ct<g>s$@S<T_*D#u=vu~Z3_S(Cd9yTng$;x;H!*~sLD
zkK_p+f0)!G2UhaIenxE#`@Cvjoi3lSxs+B)qWnn(iKTsBUJH&di=@jqK07-)4|yj>
z#CppvEuu5H&cNeT*8rD=&|22J&wnP@5L<x5I@U@KW_Iix#;!HKpo4QZ=p?qMlv={<
z*4ts|x{d$gq1^-_wE_)fE|q#6(EkXjIxs@~i<E(AzyTXgsEqdIba%XXMK0I+NF3SB
zFw$Fz`U#`T!NVVW2Y3+oJIZWS7aH-?scxS1zQp<3j{UO+NzGnD>z1DunViLuN}KJ=
zoGTVz7fv;?$r^INE5GQF_cuG2sW3t&tac|J!;N-zsV2EZHHU`Z63{vlC!r8<j`T9=
zfkwZ=GbFF&b^f_W5N6WG?^Q-}nF<qd#;x`S@>HI-p|Y0v;_Rn$h8&06g*L}T4(uJ!
zy%6a0;|y+mF=DYEyO~c~ZuOPb`bbU8caKVxI8r&hlZS8l_#`T^m8bfEeN4-^gReSd
za#f5jxcoWi4A2)Dp)WIP_1%iHWb~|(n0oR-mA%;aeF_=(2UTTij&#9GUMAfNgsh;g
zmrCHk?%DI`2=11{8E-xVD!YW#-d_k`C+hW!@DPC=#PE46toJYE#5f(!eDjL|8Dc@O
zfIGhdTj|6cV`yU;`nJ}KDI_ReM2L9}B%!0me<7otDG(QItE&Fd0+c=FB=G#r<#qEW
zO6c%irCi?7?)OtU-WPCNJU^bvdRL4tT+%KH2cn<{gAp6W#!?1p9!Tut?Xyw-{^=1;
z=QCc_puKsGFCjR=RaNjve5kAh^*j>DeD6$d=_|fRVTh}Dk8RvYL6&ZBman=^8w?*6
zD6Je+2<EkZnlhVI%8sWCPf|de@iH{jgLWEyXaY(1BElr*?JlAIT;iF!UmxY1azD<-
zOeIa|T11fc+{NwrAZt$!FB%X11)d5A%ens}HNSXmu)LeZM4BJ1(eh(QT^UL^@~R-~
zf>y$z(LjwD76*;L5dN>D_#d&LnhyR2^gxmfBSwtD!8tk$BLgi%YU<E9I!>tfFEiJa
zx$<o@R=+ZN=(l}`pR`QI_p$&DMkHu;8w|xW{=i9j{0SiKSBuZ&*qSA=cGxlNmoDYG
zNoMzh`N-va`%PjwkNJ?$^aYocL%OBkyy|0Il=Ly)*j8n_VJV%L${_nvbjYte;4a=e
zzruD>oN`(_gw6pEVApzAX9PvQAG19>Gi#@m;x2eG<ol!JH*)WA(~p1+6*0ZQ?8Vl>
z+HkYR?R9XSxV#Pbl5YO5BfYLUmy2yklEk)4k}feFrg^;aGm-pTfW1DbMl#;~?ZM1g
zQVTbP)e57BWnOxCOXn;+C_P;?+=&@OyN$N+JD_#Ozf(6{-rc1tUyI)I&A%O~SQ~l&
z8^WD+H@FQc6@6Gt;U(e~o@9aux5Gz@f%zM5%TBxAc$xKQ)3evw59aS;#slP%$DJFV
zBOJLzfGFJ73@$py{QBZ~;{nrSDIP?<6FMHMrn4R^1C`n#p7Ogh>d)xT$65R3!0*D*
zS*Tbfl2lJ|rCODu(yopOx8L`86kRl1wXK@&$fLhxrg0XPQg!`y%Q*mWxM63)0esS`
z2c_EJ*}hSc1fGCL6DSXzgAHOH0V~%l3Xq<4w8VTjE@c+!fTba39{4cKKnv5;BF#u^
zh89vCJ7uAyffD!o{1nEgZ}YXW8;>64-RVD*k~r^hEq(5r57GklnaacDuB9R9=c%&*
z1R->{mb4PHb=w<*+$AS)PQ0laCJ~egQo(1W=e14S*6F4PC0c^Qtb1yg9xRefdw8V(
zLOP3^ps6WGaHqg;RKfO?pwgJ#QXm+v{Q?e{pMw5Em|Q3HDcdsEOPE4|BP&nB;hCqQ
z7GDkbar*z|f9$dB*2dF|3_MEyLe`ZKLn7F-KtPC(l4GO-j-%|HZZ}q@T>&FDAvi#S
zOQy}b8KgdUk5tb~Rc~yn;;oyzhi-?J<(y%(;GM9n)Keaky<|1xiG+(zVcl1oqV_Ga
zD^p_ZFLzyjsQ=I+s%S?<0yJm^(<BaK38jnDm5d!>?Aib~2{AKDghznzLp>QV7^RJ4
zpQjr(k@Ww#j_w_2E7NJM8BJOdwjpM;b!*l9g{Zg$!X;GB;i<=95w)P10lWpnX8&am
zVH?xJb&S%8PT2!@x0@xGy_IL}3&rmhJ()JNdYRIS0|5$15gpVSsW2pt4KsUA_2t+e
z-*xv}O`P0YA8YDGpUvvU2@kq3Kg3%?gM<?Joxr|JXO1KG+zjODwQ9`UlO0$AI+Au7
z{hU$JB7qtDPJl)>3wSNf&PytHM#J$9_7R&q*UD6qq)fchbH6@bd2FEtJf9G)xpjXb
zo4Lzat}DwjFk)D6(M<D2Th(96R27eVzf6%u-U;gZo&_4T!k27xi6zAB@-lQwg66sg
z+6iC}#mk;B#inZHt?<nizN&7lsZE~_*-9QgBB$&*Z+~Ao^&|x5@=pv#hroYd|3Jt7
z;>UryI%vXuCOdp{pRjdcJVR9nA&@^OFdX`+^qWBZn}_r9w>WB1nb@!1aIoUv2L$Yc
zEv-T2BJ#)Mt0+b9s7CVM<*4i1s+fuCn>A^26DJ%qg}e3oG@IlC%`}T&CUHXc<EQ!C
z&cR*+57nK}9NFQcapFSlnaGYDi{DNSP4#BckpX?lCfo22xZ~hv5~#*TWZz9^vg~;W
z@%g=IlUik41?f&NH%-kd*}mRJnCHrKon07RXq57ss!>koCr!4+N97!^AcaSx#as@(
zV*j9NJL_O)&wp}!Df6~XZ$yf~PzCaH1)D&;F8Ka*<{W_SZjf@mQ@p81%Vw#Q9kLoZ
zGd$BZ@5E9`b_H$lQ)l0*B*C>KA_R2ro4L3h7iax}rz)gH?C2~`wVPHn%~>0cdz*=y
zNv~=h|F}G*y@C-TrO^-Xw%<;}Xi_VexkStd((V`9bakLx)e0L@Nd5<q*04}a1+xvo
zT8o|DrQ9Xu2BOWyrs;-q@b~LpPSBI<zxrBkbW6RBDL@7peN&LOUR9$X?#6Wu(!lQ8
zSG)klkAe9-7!z#kRlFlE8X-};$N_vS*+9r97X``Nm}h=XnA7*X{-ms^){na=i??(o
zU!Xyte)VqI&js-04<Z=vnA{7qR0O4vdUa$_RNk#mS5js(3#BQXp%7&Gg1O~*Y4$)%
z1)~w>{udGttHO4xg-=%6+Rve(6R{u16ygT^n%v|vLcdk@?NpeqAd`-Us65pQbwu33
zZ&}l%=!PV-M=$C-;yKsP9h|fORyT$#6NyU9PMV%qPk_<*WtZ13NW(UhY1f&(h?>}m
zvU2*7QB^c-`PiaYvCW*Vz3qAxmyLvyOStchK!N=8ms8taq{1+)CYF#p)J>LRa^EHA
zcP|{f^$|Z39&t)lf0*_~A=xr&{OSuopG~sA@`DE4)beFgT0EZcJ`*1Yj>lLX5GO)s
zSRaX<x*z+d*4}N?VUjxpk~oNBI%yo{&iiuXTSAkA>bM$=kPQsq%<YNcQIC(o4q+6t
z*6g;$XRg^jN-FF8e9hb6zQb}jrv4QR>!}(RhQ#FV2EaCruacDLBF4*Vq1wkC>qbje
zy~#*k#QIX08O>HuM<z)5AZwil;=_U#qZ#J3$c{SX0qr3x8;x(J3aOlafn3qG3vG9d
zNE(;wKV%BoxyAGP<BeoQEa)?VO1UEhK}gMWFaT)2vX%3DRFUz7x9Pps18Q}PcGuoF
zRrrK@3g?srUyv;VZ?=2PMU~`n#nMvj+~y{n+a@4XjQRT@{v3les)F46bb9n!jYWdp
z6={u^pSdJMEnqB&C@jey%Zhk}=WdZfU#g1_z=wOgH*tNDIIeT%#(4bO?V1_*hZ_s{
zJD5YLu2CYi4RMhAv$MXP$S2wM+ph=eT5rl}ux`o$>=jnx>7vF<r}J_oNNQUeOX?GU
zx`WlM;7tt?81j(}`o`JF&7&VJ^BEsP12RDTnAtEJ#oy-wlGnS(ZiemNB+;MR%Ds;G
z6<wj4dIT2zrRjk<)?1v?AD~E{Tyn@DHxc%unZl1~x1oY-qjm)`nnxSty_=_gMLC;t
zG*15Xe~TI=x^%$8Wck3(4p1ze_){u-{&fW=LAwc=$`1MdSgXfdCF8TS>@);g2b$2p
zr+%WuC?ILWm0mWgB+LCS7g^i^T+_5z81MJo94qY%rlPaKuRkLf<T(=D(@F3UZA3Ve
zZ-~yb^rbmgxqagFpMYXt8)UvsvOy8Irhm>O4>4*MoP-GZg%@Ans_VX3`4Qi6wo9%y
zvhZ2%{>>o{0-i3=0}2j<GiVP|N_uC~6&5MX_dkj&WN<feU%a+5nL#C_a3*K=lB(ru
zH*vDF%n;^PvS70|)WcZI6P%lip>ZwIH}+rrx6PLOKO1ZI|Fgft{!ack;cRz&0qW(3
z7euoVJ;Lq!M(x*_Cq~ttmt-KnM&^h8;RrH{X{-w8Uiom+$Sc8EnB||FGg~Vo2g8pU
z##Vlh(pM$>w<m0*)-7m#N}9Pi$k$&JTV!)!Xi!KcGDjCOYtQO=#QKMC8<3A%Z>Er3
zI&L#hkp_c}-jq%^B+Y!k)sRZt+%9_B>~$4g=zZoSP4|QZo8(^Zv<t<aq`R3ykTG8R
z9;kEP_-4h}ud_O~ien2B$7~D3#JN8N9k+fsbu;kh>*N7-j*&fxbu8MMk;CL=4%sQ#
z@q=2NdxjddFjZNfWIlXnHnnT#81a~C{`|`Z<-jQk4w20o3y$x`VIa_?ASyk=y(514
zl42)HA9$`Eq~t$SZII8?D9Q8Mq`lnc#0{Mnx;)&hQ(*f;R;D8O+Nd@)Rm$Dobwd1T
zB)f=wEaaBOIf#bcD+`WL^kYT_lN~dB`mhcC9x0`~c_JA3sYmVL3QzW|<Hwgt&KLcT
zL*33B;!<m%R7KdP^_%sL64NE+PuGlQM`m^{8<XY)Mvs>lhF7O@G|1tN-skaV^B4}%
z)$*sQ{_S=FEh&X-<kHg<b(JdWmn8R0mM@6he8GY^k47;-$ajgKx&EiuhFZULbL`h$
z1K2UCDW#l|l}kwr_WFl%9vm>WSPDw=SyqCX4_}aVBSkOq2O1ot3U)8y+Evfi;P~D8
zm0O?e3*AX^ThaXDJFxv1qE-3W1(E~7<p6B9)d1_vAN%K5cN85#sHvG$6aUn?ouOYR
zczvr^{{s(afsc~nG!K~AGjr2|qic~Jc;>;T;e>}x98t|eg!Ffl$g4HFGp|M#`*}Wm
zLEZLoB__zmODg+MjvJ_w%7d42+#3d5<XgDe=ZM3$b8h<+v<0<u6YHm=T=%=aZhRDv
z{n{G#lv97w@CpP4GFf<FE(oM99z1tZ%7j(0EIFCc{xAN73<VUg?s$(Sb_<!dtlQpk
zmNTqTchF=vRT^r(wG_+WwIDX3FG2M8sQWR)$nMJZXWx(SuWO8bo1y!^y~+eIJXyvc
zG+;q1GP0XB{0NJ|nPPPLKj#_Ci~Srk<xlqKn!P#(|ML|BTTug=rYh84+IhMJBd^A!
z1{%#2sU?eis&w>F$a?#+UNv^XyyIH3aTBYjHA{u&A?0b;JpO@r5Vu=T;vXy?+7MXZ
zS&}NV=~x=!*)m$S_cXrXXnkRMdY^Sp0Feg<2IQmMO{$0V_3pE0ymKR~fF&?os(wk3
z_oT5Tq&V+6<U#H6JpSD7&3cjo+WRjgELB$GFC-d>;|&pY3)sQjDcJJtAWfI<kJ({Z
z(-8ivxV(QMkqYGHkpMbUChM-{gWg)=*A_ZXhRPsWwIgcaZRY-j@6kJ_Hnrmq2V@QA
zbmp7AqN(;-n^t^HI?~&b_l=)znInPacLq(I*@Ixgy2VdjTJoT);`xr{2Y|yFsaQ)F
zG<EkAlyM-v61se|%0)qb+rsq7foF&(=tGP!*hAR#hBZP6eU233=nx|yTZmOjtH0rd
zG%&l?SCPcU)>UkxAQY!O4s+QFGLT@rW(xiQlWCxSe(&8Rxf<06F~gsF8Re_?X$ucb
zhL*k^xz7P{Q9AZ1^*@~zKd~{ntSzFLAp=490XkwRtwU15T4O<hPoMJbl-%u`iB5G%
zCs?>Q&pbYCa#mymxZCjmYcCAK9HuInQJ;~%o86UaZ#=2*lN#~!s=TdC0lK;Lr-qBq
z1KUW2*O@<l&GTF)8^rr9?FiZ!7O%aCG3V<q8#B`uy1dMrWJD-AA!8_BW_O%*x5xq<
zyf>L0f5IBPlL@m~sQKxI0U66*s_B(IESH<cUa)zy=81O(@F8hOG5x8)yaOI2D$80~
zO?udH-<@_w($%Ta-A~jt$GsroPE-E-lU;Y)#Rtyd3cxED1FoC<r0`Ze#}g(C>e(fC
z1Pj7$VY>HS?YFtlv+H`yQOQA-!E;=^aTr?`%Z_qBJE}5CYRPz$ykrkLmt=+jfehv(
zs40vo*3-=U7dA2+<BkRFTYg<WA;|Zpj+~%rYVZl{xScL+Ad~~-+c19PrTA#3L>-wg
zyGd)dCX3im?*q{?)W4S0b)9{yl=U7v8?SXAo*zcYfL2bn=L)nEv-mj-D*auFCStdm
zGJkSrUpM?I69SK0c!0;*(SAljNO(SGKlY&hQF5>?b$@B`P(X4w=13%=WehXv!+iYI
z&p+*pU*`O?Z%QY`<)yvP{!$Y~8MiIylYq+!-K}6|BlGmzoV%UbD`xtF<xh3xgAQ@B
z$>s$bak4j_1kWD}I>&-6FrDYZ%DqkKC;VN6PA|K&fACQqc=58)$n$(~j1RG;n|g@^
zTu?tF`0K-f`<ugPyN@PPOI)~SpEqM>bnvL#yZgz(@|6&RmqYb`8W)OY!kqCh<(F}E
z?F!N${Z9NJZbscM6U_uXw(NrMG=6z@AG?m9p)KkJumI`{G-H6mT`7DsgIwa!Kt@EN
z#p@*JmorYOX@B@u;Wq!#CFxB1#qEx7ue+ObO5NX56=}T8K3Fvc2aPn4p=t_GJ{U}U
zzi6iVYs^;4G_vTSO(a{8<1Gb2uFe3DE0Cj*7};M0C@ai)@FlDoL0P1qfb)VWj{v9O
z5}R&WWM=vKeGf?P|A}4R_N&|4Euf_j_zkcd1cERCLcy6qC|ZxC)(PXC1Csbtp>9M~
z(H(X(`>9l$ecmna!IK-gDPv()Uye++)+0!!_^Efx7GjksFiP?8Q)K=)j@Ni9EwO-}
zgEs|{O_ORxa#`~D3~*9@PSAldUKCdIX?8)+S??PrYc-|URCzgeL`&N1^{reGeYh}E
zXjf8A3lFt$X#*kfV^ke0x26d?JlPV@C6G;6j-1o1U<>rq`2mjD@AZJxCV=S);UUdt
z{E>O2$BnO32Nc34Em|5HYmBTC)9+Hgz^2n)<Pau@7Y;MhG3<aS8K@@rwmFDPVq<Hm
zg<%xNlkIt@xQDU-3wu`r59QjvU#S#nD%ok2LaCHOmW&o8m6SCxNo9*kD6+gF`&L?%
zvP2@=q%2uS)+k%DFJs@uSclpEPv`sRsMA?G-#OoR&UfbgetMbn&bvI%bKTGL+}C|y
zcc^04ySo16-tyh%erL|G8O(rzE<6HKW(%ghh-{ssH-i11mnWWLGu~z6G1q#d?!_AE
zrcH|YfWsX8LqHF2<4jSgvH?*zp1hQB-&~D@EcC&+=4r}9@AnnD{81|q(msZ}HzG1u
zGg+bI?v9^yF3nu|fTW>FI!P4Pe^`(iI_0Zlo_X%LXp>q0+Z2_}EeS!a$!SXaOlrMK
zE%MfpI&t2(mD#MOP7YmSYy+8FWzIR<mS$!7u&uG!$|AZ$P~hB(bJOL{>pnTx<OH8m
zt8wX^MR3!{<|BF_b$q7Y^AM#iZzklH$@~2Xatg=eHuVoG;^Bu$M$>Ts59VZPN?h({
zJ@U!=Z0%?LoRJkQqF&oJxE|^WXvxqzxuJ*8^R@7$JU;CPu;P<nM&3&a%jE~LFY??C
z54UY7uzq={?R?R@aie!(_B*8B5V`k25c~dEO=w*@Is|AlAp9mAr6~`MiMAbmrjF_a
zr4AceS(AOHZM9xfeqPdN-e@gilD>Zyr#jS)T?JYO;g2%gD?ELlP?sg2*2eil&w}Nk
z9s9+}Sx=N{>#GNw+A7-ba-H?0>Lx<w&Lv2O4U~sC<8Ws3&ik8dBX6{sRXkZW=Z_y1
zh60$24|%KED(*EOm=pP-a_}M`jzj<X^)^8KOo8`*Cpw<~-G48;;P<p2<F~3S1CfXE
z-m>ydsZM_9n2#}ZEuNXv>ajN10wU!c;i(UB|274_O7cb?92s~xOE}5E7kR@?n+5^J
zL}DlKBN?eH@S?_Un$e&MJPqPQk3~_}u%f9BARHYmb^R3~;7QHj2b|-TFxd*u4<sIv
zfM<bgAwZ0%C9#&_C9v{Ku{3Be4I+Hno1r=cYH3g(j=Bfeq4VK5(46}x+&TnrP-|dU
z9rNY_M2@ctk3OU6izd!_Ql_Xj3{6B&_3UJmB?%9uJSw3+t|@lFW-VyZRI=X8NNylS
zG%$iSd_n7iwaphTTO8>k2`^|}<pnJ(%Hsnye$}zag`JvUCc<dEpkvp<3!Wf3mVW2&
zZS5fsY%B;K=EX_P<T}3_L1(4!Dc-4zUfAX_fwP=y6p$!CeAknXvA>gLX|r=s&&~}x
zpLyF(VOjgij7fG+>!iWtmbW}=?%HRT*7hUbW9`%^7L}xUa=Ep3IAIn=vBC5y@O*(e
zx*`2bPai3{0=1=#l|M|1qH4Nz61+#1bhh<0`KB|Pg~cJPPDfHYrV;}|c-9*O7`Rhh
zMHm)NZglhex0!O-KeXcY6(1|x1$y6KzwJ66%T;7Qas@1h#<w1z4%;xeeKpfA?B;*Y
z8s_i%)`|MNv_DvK43}>o-NIx$t6}JQATLQNeK#fp2333CA7@<nVszC0o2XsM*#(#-
zF1rIAQa+AHJ6aD6hujN2zp+ZSM^UJR>vdCqne&a+=sN~$DIhS#G)J{aMn2S`3nkt*
zINr9IlY~;^z16L0c<bV_H+uujCu63AK;5C$T^JIRD!Is+lgLR`#vnifml8-YR^H}p
zTzr3SGTA2TaFmaEl=(AC-=t1EIci78yZl@f)3Jmu$=i<vg@RumD=n2wtv?ns2{s~M
zteF>fz0L>i0nahR@7b`3Dy!7liYD&o5`K7DTAEkGX7Mz8SzRu$7wgU8c(`B4MKZNq
z@A%8y9nGrAl5YNS_9%57Ap5XjH}#Bs6r8<_+F6hKx7L6|S5C;)T~b~DR?e(&E=b(0
z{A3NT8wo-hPTvMp0Q==1Tz#l-6)#CLFG7za0;!pSlqK=e3;&8`bOFUke)CQ=o&!gi
zLXd{YgE=~&a|2^xJi1STU)k50l3opzGp(i??9xY@ose_d$ltAo!s|v=1bvT2kv19D
zwbP(mHlTt>CAC!NcV}ZT)PG~CrY{D(jdd+FC=>UtUc#ORAwe_ND}n9Wso&{TViIdZ
zo@yW<$nJeK$OVY^Z%D2mMI>Id^Q7*d0D_Y6<Z1=36cFsP_ILcSH-o+Tn`8tjxqh{N
z4v=&61<R5we#q+gT<p_#*fXM>-h%RL(?QRNtd&T?pyI&DCdjb*U&$Q5C8>p>6O|H7
zz8T01H^`uSMaOV?xN%uKn3^vcJ|}HHbKqByeX;}|O>yGvg(L9?k+Yj7#O7wjC{k>5
z@{vBGQ%2;0Po8V$`YO148)gbaSzM=R-j5pRNS{lmx|%Gi^u~1y!7Zoyh9U@3<QtVF
z-_w-&cD+^{N`nMQ=(+Cjxn=Qp)s6vAh!d7_S&gz-vkdxFPUu@AHw$s`UqlJ}g|gj%
zNO{7oeM=+o_%T*TDT3hi)L38v4r%9>_kxTHo1Nrs9QGS}W`Fi1<DHhZg;tM@K}g|r
zEBh=MA_(Cz424ZKkjw*5hoCyTsep%!G_?Y8xoXrJ1!Uw}?LL?+7ycSId2~B!%oGH!
zn1Bes+e9g<cXSDr6-7D<IeV$2K?9DB!+_|m-u((Vp#kA%ihY@jaMXP7Ea@noM)Z#E
z$39M|zuk0y9h`z}nQ2W`mY$42?Ee{S!gvXOlL-$2T%1dgX+e%}51*R_`BtJUe|O}E
z%qXCU6EkR#E&L$KY=Vjs2GY>}<e})X@VVyj8SUASUM#9>9G>gPSGb6R^rYW1qJMHu
z52;{CZ{2z@ezHM>ASsKrWWL@KFmn7(bFnB3c-Uq+z{Rh~bc&|qfTcT&oH2@6x#9O-
zaDK(3#cyeX2MKM|oF@@0WWFOlMo#%xK-&*gn#T9bS%3(D0r);Pp9U=gW@)5@29cIE
z!TmkhGP_aAbYLX8a#F7!%#s^(&jVAwi3UCGgq1(2D>^5@q3c1h5r$K9z)mJo6M>F?
z$eTnbChGn=8l(l3y^lh=^N6-tCj^@Q0S&S;51;q*BXqu;_nr;l#}QyQN>TyW^2zvN
z-a&p#f#<IQa88@i*=47|RHFf2)*S}X$SEb{N611o(n<i|{$6aq(I<Q5j-}YDQI=^>
zRFaOsrLR9#N_jp4dNjZc8XNrFqm&9jw0Ae|2CdJ6hYyo?Vp~QVjHpL%tAVp~W&j}L
zKLy2pvUp$rECgXX6WKS0nhQ@KE|_oS?`2h1X63^}mT)V&*E|GZC9o8fBaeb*1{{T{
zX?>Jh>V=H>N+LO&nkh9A!3+k%ys`eC`Agb)Y_6`v3+@&}k(oX%dIE5EPUc31sBzA4
zD&wc_hl+m20{!Y5CkZHlQx(17)__+&BcmXPLC2HQ(9h`gd#{&R&ifG!x;f>pr(TFu
z3xGG)RD^NmcVHW_n(_qKeZ#Jq`uaT$+KG0jE&(>8xd7<MuTBA$`7JsV`3n>9co+~z
zb&Bf8?V@nZ34rt<u?KK{DLn8hyn1S_IyxTNcj%K@)LO8ZeIbh82gY71t`;}RroiEN
z^z*`1pXu{_+yA*C8S8eB&N4NDH{1vYvf}BuYgfr`Bk)iV+`l2|I3UH3Gkr@wi4MXA
zV5EcatIY=;g#AK)Gzh<ul}!gB9fUu>>!*Y8j|bs5LbmB3q=S$S!vExQwCbmM6Lb*L
zK}ZMTe-eb%bP&=VGvDJ>p)W`O_~qz(Itb|?q=WFk>mB_@p^g9Ma#a85Q`5|IO%>_S
znoK2EiAPmBh`(O*@{!1PqWbN;((ZZ*5Lq#kKRSmbO9M7h4@2GVB6CVK1jHPZI<8U>
zR6f8HCZ^LMPGzz%o(0zrkM&VmEpWt4Xuu@NoXR%pj-NyktwF7iG(1@!ITr4?9n{_f
z0<Ymji8P=<+|T#|5IgHZK!g5C8nmCCJVC8Nj$eXTqlRT+8Wa_RbDD&C$$PSD6A~{v
zVj}cg%H>qn?if6-vNh@5qDX;@FL$;xl*(zhE064?LBdP~HfjNQoH^-O#)8i*i0F*$
zG#3rJeXa(?<cc{_M8}G8);Pn!S;?a)YW@uVy^DO(&%lCCVSXdcgic}pQz%U2PAcGf
z#?8Y3+k+he%#w1nGopT!X#@mZ31xJu%~5M`lh?G{7erz74bvbRw5h8DNAb%<QST|9
z!hK0~CSV%@=cZax0F$_FV9HjXRFGyoy9ebJKFtsKvg&e$b^M>W;!i*54E*yF@<X9;
z8f54&I)L_&Acc}vz@;xh2*WgIE)^)I+phg~O(1$e(YH8OdO*=18c<aGS6Wfi!MY&e
zhaUa+jcAL1cJ$w0iV~%Rl@8Wl4u|?@gY_>(ozlTd2P++{KOa~Bm!eqdV5NhV4%VLs
z>tBkBrGu3YRytUJ9;|;U%9ajRI#}so{dutdrKnvxSm|J;gZ1aZ`hiZ}(0Nw6aKqn2
zxPi{I(t{%CDl2~zl@&VAN(U<)tbY@*(s@=oSm|K>n}C(hv;HzzACuNd1Uw*#nauE{
zJvNUivW>%MW30vV6FnRKG8m;#SPdv3wbV6mX=I=qc0lqLsG<(a<q$Mr3!)m;!$Ob-
zF@n^D&e%mG?5ZS8K0LY06iMVq&>(i@X*>`P?7#3HH70|kZgJUO{gbO`TIP@LueI!Y
zUL(;eIo^AE*oXI3_pRs4_h@c>1gT6%FsUETi)Y`fy!Os#!7+wcsBV4K6i^9#0kpdi
z3sl3G{7g{uuYJyc6+GoP<PB}#6Tcx_&Hgz-;%k|xU;F#=S8`~g{9j2lt!4229K!v*
z#|1KnRzF57S23;r-V1#`e_n@4bmhEM+VQ3Q-_gSB1W77z=`un(#Vwk87RN$np+SP_
zo*<b-5X7NJ-NV_SOvPr`uagA&TQ<0N+8yw9xckZ&4WitzKuun*?~%JoB`H@s8_#W<
z1WHRmXmJot4+KzatWZ%v)eEPEs<`<n|Ktlnv~vq3nq0OLA5!-lQ;L$DGZAd?n<F&n
zRcd9`4>9uKJ(&_rzfeQdnEXPd=?8f)#J7q_{P!}C{_E~(E8-CKh;T<b#jhR+sac~}
zYPUGcfFz;iL?Cl#1k#?UEmcefcm6+$mVGLG7<?wI)(+34A}CTV^`&>e{-}!S<3-#M
zdeBTJ2t9?JlHi|LJ6nity#+$2^BwrPBY*3WNxuU%m5%y|oscLufJ?f8jAu12)FzjW
z2I*N-2SJ^`)3tqMXCMft&pTVS#O$+l+;<-Nz7|?5L1)XSCU+udSt@A|!zTqW=xp`x
zR9V!<fc&lvg{Y4urxh27au+Ue+kZBczv-ye>pvp~^yium7tE`$-C^+jYzBVAk;#kt
zD>{d*h$dC)A!UEuFu&LYI}zQEp5@O*f#eikH40zqtQtQJ%B`M6KcM<Lt%AFQ-{FS3
zag!}-{c}Dd*xJB}C!Y-KLbWwrc5DPtQaptp#Lq2{V3Ge(_%O+>5I@d_`nb({^7Gp>
z4Dl=wThrVtOWik(n<_<3uGaoR%z*K)r2<SxIFK@?Bb<)#KN8_|ih1D<fIf%+tJLEU
zqS-$)x6|kFue6H);&V8im!>29*HOI?@pOd$y%0_(v*`$@Bm5Wk<A3{oIGq8fBb<)#
z|7C>JX>&Tl=?MQ{MmU{orz4z>@c(6m(-WcT2&W_be;MKQ>}fi}=?MQ{MmU`hrz4z>
z@c(6m)A?{Z!s!VAUq(2c52quXj`06wgwy$OI>PA)|6fM<pT~!**SYZblu~MDHKvns
zO@^~-c!Rd7S#OfQb1ixum5Y3>p3L2jS<Amm>4i_}W4B=~YYXwF5<&9B{*Ra6scp4t
z_L+^R^2SU*17)Y$X;6FCF&flu22o}Rr(W(&t5RdF8B0=rdz%z9jxE3tEyJ5)sGe>#
z$n{gTn_ICo$YX#K2lv~dh}&y$vpg=S%Z?eSOn8gj5C4)u%HNkK1ped_jFZ^u1}$`>
zb>YM7$bqJ-=(&$~iF?t}A?Zb<Ug4lFmEkD=5PppWX2f?$Y*j<^0pVMVdCx9wogTR^
zcyeN=v_ROk`rgp_T(WC_Y=Iqm2>q`cCVGJF-z&iO&!X<5`As(~pMuCLlpJ6~`m;g7
zy3;nJlnoxad$7qg2m{Z`h^?hT_58qe9D0lmm@hHR_z|@jE_EX-(xBwXl36|Ng1a+F
zS5)_F>INFrP;Ep`jPEFeK4)D|RsKqWEM4?V8bl5b^Z-S$#QQ<9J6(3Oud3#`$jSKp
zf91B^E<!!|zFGl>#Ek`C=*{KBz9QR#`dw^e&D3I0c#o11f$;o%bI~DGgE{qBcpeQJ
zHABu}GO)LQ@=*k$r4g>fefVIz@n89_^N-$w9N!*3H%o&WL9#Uy**Auo3r`;|_#pv#
z(dP&Hj^oO#e0azbZbkQ+hh)>Bz*3Ok83oG>I0{qK`Y5&33mNg1L~=MaQ)(iD`NKCV
z-h8VD*4Lj&f0ew>9`JSAD^&C|<pH5z`<!0J_Kz=P`)Bd?ba?*@@pO3qDy%=lo(}In
z9^QWzX-|hY9o|4><ui1D%aIQ6KOWwH7GqC`Hyz$|c!Qnif498-pGDWx;Z27(9p3+S
zc>h^k{XYZVm<HVg&b9JPO!exhd!d)jyF=!>rh&)scHEdUr4TpSYOR2D#?zqJ3b64*
z(IH>yNetTqfm>Z?)21C%Hw$!09q{vsr$Gp2N+{gV(iuoz`T{s2Gs%_&B-u|UhT;SS
z-yVk)MIf4RL_6)_AsVEBUJJKN1k<2R)!+D@Mx+bMj0SadpkHDs436qF$fXWV4pt`1
ziB1C#VT70uYK-SCP%IoGobZ9DVUB{GBa9_9XvJ=r&+e?~eVbMc=c?!i#ZcvU_b>Sk
zBHlIz)`c7Bk#@tS364M(4XZ|Zw(Z*8dG6wC&)cv4ZQJ%uLA!d@Z>irwTEAsU|BZ7P
z-p+W|rO6$jLF?@SL4=xHmS9UN7}8s}o(A0nUdJ&cW%2Q)+rC3E@(Y>Hd2Kt}!XG8?
z6!;W>@AWKA@Rxx)Va}5TnU?C@9^k)qP*LEr`3~o&UEs(j8njWc7(cx-c-Dt3HYP?6
zXhQ<^O8*#Yz$r=<<q59)hFvrD^?Mq$6YWl20;sU&0^mGeokD}E-=gmj@T3u>;%s<-
zdNmVO#qljQSCs}GEu;p<M*2{6XpqeQPfp&{Ngx!cN717}`y_6innRG?6RC+{ScO0@
zJ2A7%!)G`$=KAH1f_C3U?W{*Bd~UXN$uaO7`Fd)GFx7QS&&e8GH_{xq+HZsFuwQoh
zBz=OS{aGOe{|rME*l?11EdDieW(`mwwJmoHM4xuf4<ENRSO$begQ*XbOc;BStAR3M
zK^$EF0Wb{Q38$LAGBS()-;a;zkDoRHHcW>I(;$Oz(uq@3TgLH?QPiVG9bj-?gGW@%
zKg535zIy>k8I#7s=bFQ3v}Z$lv8b+bc&;B`k%?YTgAy}nkS+Wm$!vm(0&>1d{^b5e
zslO9cD}SPEX@qV2N*3E?rwLGR{LGVl&rviT2LN;yIb)QxA<KaV&2j+CazzH}{GN*}
zJXh3Oi3Du=)W{~ox_Q-0s9-^}{aSM~7Q(+MS^LrR7fl2?CV&RvP3FYj(ICZKFVwhd
zFM4VmJ!dIA>7=+;d$45o{U_VLP~9@Yr@jjA-iDdNP!?A$37;3St=~6q3+5Yl%71~j
z<(9Be8YDnM&;3x#=?C#-2HKw$Z$O7X9sZwWZRzm;r4_<|9sYENKjhzF`02C1`utk+
z)0&e$`+p@^|5u;==@9^Q_|q5v|JcPpJwkvEe>(jCvGD(6BM9CQ)VwEgjCO*B1~PTq
z&gG<4ZEYLr@OI6*F}OBN+^L_@z=tSET_-&jL%rb-T$zi2?}i7}(G|`N_aCaojcq^@
zA`m(KSl2H<i~9O25UK5sE%rL_wlDnZGjKImKgTWi&Ux7hYb|Nm?p%$q4m<@BXQpR>
zPc4$5MzN~FP5ST<J9A5Z8g@J%Sc`Z?2JSi`1c8%U?2GKN_@SH)aOT}+BKm@IHj07c
zX;{zs;1|YvlreBai6L1*4Wsjb8JK*rrIzv()!k0r05@ErK{QBN;=D7c@iJ~8&kI77
z-$v1(fpePiq$6<YYw|(hDINhX=m0ZlZ1A(}JSngQJ#L6%#SLEo-mc~l;^nUck6q;X
zK6|LJNhm|LSsCnNxaE`MxL#s>uagyt87{506Qp(mKN1K@zVE0E$}Ft}c^haDXx)Dy
z+1$Jk9^Ww-J~f4+dYE(k;0XS~861iI%G((K7QPTTD-q}kd}ISOD0LI|#C%JAc3b||
zBRvH9E82CwxXxYA`O;b<j4Qt*svEKT*9eA8WJ)mkW*{%zAcO7|9mD0}#%1Sg^o@tf
z{x3y2V7C)Nz`W2bL6n+TLxZ+V@r;iKhL2;y*WiX(QR5Maz7wC@Hc=PUdO}WB!=ak-
zT4=&76U7GIr+_q|wgIHt7;FY!aG9&rTL;j5+6^=)9Y+r3^nr&0fdBN#@%oar^M!<d
zntVVCe2pAGLxWbs!zRg8t=h4kdD8B`oN!b!3|>oEvptaH4G(1Y<|=<}J9aZnAiMw#
zYFLXJ8iXk;>b6<f1yYsxPf~MZfGe;3?0gL#rb*4OsY!dxFfe~Y$)H#pH3v2QBy~{Z
zb1QxCk)hu1r#1HU#lBsT&xYaD=T803Ba_(I#6)uaC?fHqohNnw1P#iACs!+QrF;h)
zxR8YXvEW|)>IH@PAghHZeb4Xg8BtDe0e>_d^n7TZw15o1qh=UGC+Zh$!OVf_Aiz(r
z?!u6mRLMoooJ3BlG6n&rEG2*?<!#O<OW@HIC(d3t5`PdmyJ<pfZdQyU#Wp7&=_5L2
zL>~B|kiw75M4h5}KWdyKeJ-8qYO<)(8`muax18!5iXcdlZ&Z?ePg8!xMe#yj;;tIV
zE4=B1rCe5{EY|#2G7s&KjFzCAiC-+w_ve}k$APh!PHcW90P{=Ze2Q2n%G;<4uUR8s
zees;Dsa#%SM&4@CE=)*N&@OTsuHzZ*0%A@ijRp~yT%|$9{fsRp*p(nt^eN>&sz0(B
z1R1d7X%KG|mFM9!FuIYx7!Y}(4U;Wh_7?k$jS|f;jc!z*Z8$1}J}6I`iTc(Bli2dm
zoVhE=>a|I`5uR;gzPcY06Q5++t!c7*;~^eb{(xP1#h^@I*xlB7`}bR_3(_kK&QFu`
z69B!k;2#=pLAPZ8F-z8hPDPS$RyKA~^ElBJZMpo|S2#1&?E3JRB@~zKxX>I&*?A@e
z8G#xDu{g8auhmn0uj1y^Eznro+2u57F|buZeTi^Gx&sZ0=BS;Ung7n`h@D?L(rXeQ
z|B2z9-#Go#YZCvvaU4lEhZKrdShbvPwpWi^z3Rp@0b5GvPKx=%NEaF82$*~${6{Gh
z7GM&NrPjl9%fLnnFNr6cWekciKK|m9z%R~@5Btdl;yH^rP9gjqMdOK5MF@MRr$bsJ
z7Bame&2QnH)To03;Ra^RB8)vS(QvfUlX{&ehRT$K94o-TL{Jt<sF4xZ=0Nz+ESk#Y
zFl`8)OX(S?K0c~U-Z3)_BI^|Vfscc+EjKpOMP3`M6xLDh!~Lt{;C6?aE{Yd?Vb2p?
zwnMd2>lec@ICb=iYK3*81gYr@#MK1zV(na6g2%(yqjC5IRZ+z&)$3OkCvRVMNnVpZ
zUzx<}{O-h?WPf3eC2x=fDrIQ!R5FOALmmf(YH<}V@<t%I>r8m#!y`MrZE{q|@47fg
zn}>Aag^x`#D#AH8_tT*FujYQs^UFCLqdbylGKE*9Vs4~`*b>L6FYWm$g%|oCtRP%$
zIvA=j%jq)_xUiX=z;?wE3*t)ml6HVCo1g~R$IL{1*-~f?bp@m-4Jtr5ze7zft3{3N
z1<$mgq3UZvA?%7UPD&sRGDa;U2CEP*_(z+m#G0gEGH8C;_A)l4N%5Yz#+br+ZvH_G
zC-n_j%;EgWC*z0(2JfdfOYmRl)r<E~ty2^)E4uH_$8h_7uKQ2A1MC*RcuH)XNCe_8
z@alQuE+FR?n}rkCO~MPFacI!z>p^i%GMco<ab>MD_W*Kv(&W%Y2W8E;-jGRA^_u0;
z_*V~PZcU`K4{ID}Dg&{n3&sy>!9Av?6xTA0<Gvg{zuu-veI4n;ug5q$wEb;Hk#@>Q
zKdi4!!O51|TQ+H|b5vvLK`?b+nOkSx+#i7m{z!w)!i^!*gXX1^p+`z(Q~Z7V7_Irj
z<e$cW(GA~t#osgT<)VtwFNI$t*V#$sHwk;~CJpSyFU_Ckxu$IMW{B@P3&wDU+TL9s
z{6*^y?*lcY2SB88I3?(L{~v%pxrZH6!!u(i+ta<b%W1Y-$jq^j;#IbWKF!=Ec3V{N
zIdhA#NOpaueG`IADhF;T=E6u9N%9C2)u|Ui%6;7xHyR{A0ps^4i$t*EWWnQw+v7$b
zD~CqVpvr9gOjW{R5k^c!72MZ`A@!eFp0aLw9N47&$tJ(yboD+H)WoP1Wm0sF57a!7
zf}EXsNS#|7cE!DT8wdo>bw69J!0SB!2!Zpr*ST|iDSrMDqCVeVYtveW^v~B6`FuTg
zZ$-$$JB800BSxA0^3&P**Or@q<+JJv{G*<DHG3h_S`0U2*<u}*>TNCQwf+2>eVZ*t
zg1C&Ndy~H58d2bpzv3QFoC!vbGNb>ECw-8@XZ4$(^f$E5lZlkAG^ix|6%De}+V-;z
z^fOKI8*cDhn*ZmnzTbl7Q8RF;IDOS*?GN|8Df%?L61Sbz-(<U7@q@S=Zu{Zqj3529
zg?=Ju@nduU%((oSf%d(l{*X{r*1w*%=-)xfAM`X|zN^X6$7e3Iou)zAn;jR9<=@?b
zqOn;Hs~;D?SL($tcW3G?u{5vzG4MBH1!{hXfB78qw&07mY}#d_vnM?m_3uIBi&2Ck
zi~>A=X*s^TLoZZi>HRSL{^$-326pUwpXOA^#SqnP_{NS~3fJHETrZnpY|CK&F(-m`
zB4zu|PnO{i>fp_bn>Zcp;nBxBDc@^N|L~6Zat!_4SYD?_I-(kdba9qmwQtKCo5m|a
zc`lbl$ISB_lmi@PPM?QUEOrH|Nt@50ey0~z_w$_bMc?7HP!&NVT@BJ|&x;9Nvgwcd
zYS<407Aet*iMqdz25Esc-qCYfxnQbXZw|-9{X#BVi4Up!jVVP*&Y6h$(C~T36;pyK
zejJes=nWOHa`ws4m>ivDY64128$ZSBp8kQ`L(u-LpwLgGWjqItFohrukq2{hz;gt~
z!g#>E@GJW|Q_{hbFxqOW!7hEY*$FwPjr^|N@xq-muNzenc&*VW65#gRLCSy)c*RO;
zsgBLOD@tAQr|zj=_>9g;e1i!2XLAz28atT~YOEG~B2{O@ndsA!dVBH~Uzw|7x?ioY
zOx}==o2Wz8lSOFI4XqjKwURj+<PKe))K;fK%MwS)Q-M=7h#A$vo<oDelBu{31-=~i
z4{*LH@Eh$lqQLoqL`4kTmjGjl$C<<wZINR>U_V|7^2N|&QPkCX_}L-k5OpoX+rRnE
zw6?)rgUFfs*?O@2|FL1-5SAo_t*NpVV>h@Zk?U_EG3<@RpG!X=xTTMA<MyS8xFc9;
z5C?^X0l^lByM~IvHWuH6$%j*max?7PL${EW(JS_5Tl*^<T;VY+tskaZT2m%nA>*fA
z9loF;BUT#t3@+${1>MFN9GUrJ7ktraP)~o-f-YFlXD-s$L}M3r_`+VJ-;DTWGlCcz
zebf}-VP9zRU+LPHMSUm9TVoacuHx8EKci^ib)f_2uVE4v|4eGJuX7bD1?!GF=gZHl
zK}ZUANKo!n5%OTKA-_1@yySvdpPVCq7gy!AY5a=fHi@yeyItE*t4<Bo-9#7h;fB`8
z#umn5(*-ZBV0bOKw|~*yf1>&Oix~EIKgr<41~Rv~A|8S{hTY*r)%H>BxlCn6M8NXB
zF6-_v6spr_xc}zd{bQQ^f%&BX`*<H<+|6-|iKH%%DAEqoOvxw=wtu2cPsL$}=kUd*
zD_!ePyx`u^e5@s=el5eO4Sgl@UtD9&lUiH)=9h|FC;@I)P0|@cc`Lo6xP)C>=Z-V2
z{)-rhzr!*UxtsbPJ-%NKFj!{>;QQEo5P%HQ@<%#o5NTNx+~0%ke+U9o^o>hs&`Pi|
zU8-ZC*L>P+8oNk!`?yzmfS%dyz-VWCI|u6WEP}97{bacThTT+L*~oUqa9ldKA)AN5
zd6`GreU9kcomO0Nnj-lF+pc_sEhs#WN8Y?JK$aNpI&vcBgtg`7XR=8LGllC9Oy2vM
z<)!H40piVHT1GC2vI?9ZAfgr2t#(ZX?Z|==bZ4LUww@2W@rVSq-5d{(d=!4Lt={;)
zx}7P5v|-+EhBmv;K7v=_t|*R4bdz=75CN61b(Kfotwr_1`rHN~JjTCd=E#s+R)Dnu
z*oFMK`2O7s+=4ISx?JmoT<f^Fx{H0D1Uz75-}^%}7)<*{M#|?<=ihq7$`Zi?Z+uvc
zn_m0@HyLtx^04`q%sZF3b<ViG5t!c1Kr8}OJlILylqwRNBfdvXvcyrs<L;$#NUnFk
zn|-B&ko8vKLCKMJ&CK~c`h}VSfA;TWFhpJR;5Uh5HB_9f7=?3*^J7N@h#c6~jv#SP
zVksK~a(D)di-}v>noBORee_->ziS}mx!0}85*~L=2wDm;Ml5pS|Ait6KVi}TsB_V7
z;V(zY(Xz$p>9qvm@q0MC!4T6XM{S{TS)HJc7e}tIl@?CY<L-6Y{tK>_-|M3J)$eH+
zhmRe{%<clk0;Y-qReX%2c3NLiZq@0+s?)_Uid!}F&hP2FX%~U0s3dMGA&0y}PIG@i
zO@})BX>d%Z)mhC--h8sUPgZ@uU%27Pd6(Mf#$=*>Bc0;6j@i%WVJl!S6o0O)<7r}7
zJb5wYL0?MM<FV&7$h#uPp1HiI%s=g*qU|kc#TAC0U}G(TxMyY*t|VSy%vO*Y+Y`-o
z2}t+dq84H&32R;T_uH|Vx4hxT+l0kfCktoDRWbE>79u#E8{rpDso{l4USecP=T%oN
zyN(YVB%fLA#XZVfy5d%Rl3JOc?uVTWH_vGgsJY7>jkc@IpHudo+`J<5O1kNik(WD~
zig|CW5$b5@@K81?a<+5Eqc}PUD+A3*Tgh@ahH}LTPPLIO&HFqgt4g+1@?=zSDBEbg
zCMA1Z$lIdU3LG1boZz3nj&S$vrSF$NsoLGFC%y6XNRW<^=B;A;f;}6ZOc-&WTjW1=
zOycb{4xChT4_j0(zuwN*>Uq^jNaw*Ln45lElrp+(0yMSd&iQ;aM`^y_op>c*UsUgf
z+wm8#AeAMB=DUwEyjc;z{Ad+IFcES#CZocq+bK`7<>j}SuG>>X<?EO6l^r-`bH#M0
zUA2$r6}#5yWTgH5kFO1Ro^emXjkl+75dAw5Ub#Y)FlX6jy_kUPZQP-lZEH75rFC7<
ziDfvzewUv)&O@nZ?X${fj+Vsw7;1$Khv!o}>%428UiUtV-bi^-ZIR#HRCGbGQ*E1$
zBKHKOAA4=x9&8}$tlnEOKOBFyHA&I#<*DSrqV#Hs1OueYtFW>Ldn@~(<?L6N1Z?zM
z+aj*wl(4^wX|VQp3coCLzkc`c&PS*%;b~}JIkwJ2xjc>>ADZ$fly2$iet7J^X}O$F
zv8;r+XvF$`_IoN6jw{u(Pjil(B((Ll_qlCL30$T~dEABPRHcMRGzuWrhwWcgaWb=5
z=gd0pvS;l<C~4Rnaa{|)2zMUIVj4V7!4=8iz0cz6#TtSgd>glI_hs$tJr}r117M_`
zb2kIY%JSVbL6+~HrtqNR(c`B8f0=;#C^r7A|JQ_I{_y{g1@w>?@MEFO0=<x_oWAla
zoGT3kliy!-6f;^pPDyKPjck3vCN^r&X589)Y^|8uojL8nCC=-}tc3KDC9#g?#DY%P
zb=`6|y;`zL+%EaYamgMvYAKv-;*VDlFjuvXM#w8188&<bIUR^0icy($f~PBEi8BRQ
zvlovaXz@DUFJR4s3$rgPIrvc4W8b|+G<9{QzsB$ntXoA%bli0Qsma`sS7K`7j_}PZ
z+qt{t&n0I*ZP!;SQJhUnhp6|Xp2WUTdR?6-;C6Fg|Go&OuPi<T?HdA@-wXCnSzxB&
zO-4m?ux~q%ax;$K74N82duQtmGo$iT<qaPz%g1N0uN<!&S9)a>bV;6weft2_9!f<8
z7N05!+#Mcp%HwvT`R$|I%FDw-w`EuMmG+9K+!|yqx@zQB!Wggwa^fe6Q_HR~<;Clg
z)v>K`z6ya+XT@9qcBd(GN{n?01AKooCbzHe=6C~_`c#p#35e%d;b={RqV5xr6cW%p
z>j`xl9c#|d#9k*ak0h#|t`{O~q6S+&FMoD=DVzQtVevgz@4m2ZZ5~qK&-qV7&pxmi
z<#SE>HO4a}rqJIxe^k9@GF-I6r^v`Yd3XGpw2d~r2lljB-;{ohJ(l1P@2e)Cj6~a6
z7fRm34}?4t!&onl+Pv`uYr+!;w<-Qu#&DuV*9f9B@cM3)mptED8_MOz?SfhY+2>!`
zg_{aGS~5#6=e>?CNBi8nW_gPSE!jwP#IMZX7dOa}tdbawWAIHZd32tIY+)AnZdvz5
z%NU6PNn|p*DA1#(%`}`reixo-R3J6}U}dQ7#qIvNquEAI9jeIVUWE)9-rU%WlyKa6
zu@#Q1Y9)|7gV&1c{Twj~OZ9Rtki40*mY(-!k=8F{eOK|$>cdJA#=)}B$#(z2BWjDY
zprat=x#!5<1V_s;t?SX|)lKob{g`Jh`PEIQESOgupRK-Q9bAgS8~Ht|uYl7$5=o~A
zkt-)na3=hmN;NhTwd#paYC?Imvc`oN^e&E*a}6e{E;C7xiqN~9Kj2=*kawrM+TfVr
z`~zUc-ZyO-nNEcp<r{Gq#T8!~hpt4`8OsJ9F?!^}D%Lo3obq^z6{&$~VmVS{)BGf8
zb&S9C-MbxIt_C-6&35y%!9E_`bGyu2u;L1|Ym$A_e745-B(3~A=PJrQ<=Xhb$n=Pr
z&ITjvGsY}IOV#!T$)~pCA)LwRBIoK;$FOn4?QP>lDXNbxtSn+Ts<*KAb8l>18|315
zRphPlmKJTC_UkqoiZA9Y(bI{b-Od#7F_+go<k{7nbf49G?-pdV6bhVc;(qZiVWWOa
zTq1NC9y{rO;DFloYfAq1X-7|y^sXmt;45>CeKMi1eB$xx%_UdS>S$}SM9tJu;f_jo
zn;dIYlh!p&KPg!c-pYW*y5h3@dG)1K28!R)z|JMXMQVZaQdz_YovgczbV{r^ZyB3z
zknA}h=<u<rC6<9r?ym4+r}CeWoXBrwyL_)@{L7vnhsHMyXi4#e#aQu&hCagVo_N=}
zC(>t!ZfmO-b6iA_N*-Ux>#^_vNA|8}Zd(+iotqkg{m~k(bK%FbJU(u`0~K+1Twr*f
z*Z!#8t<6OyL}Z{8iz0&F?Winr;2lxxz<MQ+Rg9mWP7N)b7GEA2&;ecW%Opp;oH?|3
zY~M!_#;RZ3zhP0e&RfWPs>lw}j@#=T6^)HL%dc>}dR}wk&ZeRdsyOR|F}#GHT&v2Z
zIg94=W`9V1`VU%?Kk|Ox6oCB8j-IbFXn*E88IlHgxHq1Oi`sbG1^ReY^Th`n%Hds-
zq1Nml&a|f-OWHMc!hH4((Ym9O9Z1A#qD){f6@qLh>x`!(YkQw|O$skZQ*o2ZIS;nA
z;`jXZwf5Gt?Y1mx)n+*CK8Xqdv7|@DxTzKE_|$ws8LDu<Y&H8%3m!YUatU1ns@;>D
zN(QaNU7qkg9Zv7A^CmNEob?kJDA37`L3?gR-*w%v;+`<X&=N9BYKIHa{QA_2OtC_>
zCdR|J^K&=VOqRyfM}+O<RjQLMynV-Dtty}MCH0vD7F}G}f^djv+=V8y$9InU$Q|es
zTia%*+JW>C)^N>Vkz(b|^3d+Ar@Th6ET`i{Zf6F>8qKbNx`Hxscq<FaNZ~sD%l!x~
z``g&-cMzW2qWX8)2yVbhq1wIi<Cbj_;6>ull=UfCX3AsO5aoqVRs);LwPa1oeX_%|
zK?DBaV;Rp0lHSj51XtEzj7Ym*GB7Lic?k4)?~E^w$i$WHn?VydTqQqD_lJkgf<f6X
z3lWeU-yQ>}T)>TCJ1*QEbLdF%bF?KGTzQ}9c(1Ukai>n79igmL^w7;>JJQFptDBs>
zNfP9hRj@J2wT+v2^zmZrCwgx7m!gnnxxsxIkNU#H_Fr);i_q+k#32PU)tXePEZp6L
zo3OjZ+^^Ry3)~=FZ8+2~Id~$N?CxCbT#aYObzC8@B!m|%i{7C>6yo^ISmdmSXhZV0
z_>WuyTi>{6bf3E#fw&ln<G{5Cptw`@&GT|4h%8;pdt~a}Fw|BH{<k|HSXh<@*gbEd
z_I514I4OP|4R!>t59LaxzeYA5DAI1SElsZ@Yj7&m_@vdZL5nUAJ=DE6O;=aP=1h(V
z<0!LeJAx?WnXj+hff<hO$QoUh{lYv<VjH%VyD_v*HdWp1!P!HL8ja58Jb2ia3nFqB
zeac~t*E48ri_dvPsE94%Dt_%^_<_%O@ip!7$Hu4<tQVXX3<&MUiT1XaLJzMJc#C8{
zhB{oEI22i_n))Foto_-^J11fpEqa1yj7IAR1BY(|XIDR_p{CPzw0N6!CwjWq*EI5U
zV&QUgp-?_+$AtIq7Ef%gR$g_}KW|l`r1&lcWWu{LA6+kKMbI?{jfB*_QYTj0485|v
zVa(cd2%9p;|D+QaTE5W{WZJpDaJ)!3nr$GOJ$Cgqg89g5fg3w?H(W7Xl)`ZLLe>NJ
zCs%I1$X^K+*TU*VkU_jSq}G|0c($(-9)3(<Ge{?ddm5Ujq`$f=7&5#+x=q%8WX+l&
zN%@1e=Z(gbp^p$Ud#!fg44Ex{IwhMD=gjgJ<r}R`D7kT7b=#T9Wpy3rSJc%TYWJX}
zHQgAl_X^&Oi&k_N8K|d0DJVZ0WQp}hEtX$u=V8GCHgRpjgr%WZ;oifn0S__Hj>?S-
zK6|II^~{?47dj;bav)g4G<IOTHRp!*&b{x9wym-i7nu@$<dcE+YTE+dbEln5K|Utl
zQJff<bb!E9eYB9XzH&`t`4!xb*)#seZq_zdyZxSYdlVxg!SL7to1?towZTQmmn9|A
zy_tl1mjk=6N;$Sxny!d!vP-%jWMFRU^UV6KC+~|gU6;MqRA%k5v*007jGo2F#c$@|
zKzRfRiC(_)t^*U>Mvk6ySqd#<s}~ozO&ny}Z{Ck=WShR8OnK_O$PUx_M3qnGZka?<
zJcfJr%m<5r(hn;SX?Gr69JIP#SYlEu=4rzak8Rzpxc%vpSF7dN3cEx%o_MBxVlQ(L
ztMe-OwMc4Kr8_FcxCGI7cCJ%7t;W*Qa@&amdBKOWC3o!@$<*ogy4|8A%M4M5f5zDU
z;rre{@ji^-k~{qq@An^Qn?!qa{;4O^CpmOJu8unEr=Dl!mU_qIWaZ*j${@|d8<!$R
zUK3S}AxNDd){r*YwcE++eAvDzTgLIFn$fg<@1|(Ut7%!G><y0&zFE}Akl!wL15K@l
z$3WspLlXl`h~JOlBpxokyD{s-dLL?w;HrqU$7k|&8oe2g79j6mPnmoI)?Qg%!DKeE
zE{rdEGim?8DSUt<W2eIn$;{Gws@|K%rMQe!Lpp988)3T}w)7U)j+3vJ%u>Xr!9403
zO=Rmz&(CpahleTc=vC&qn+Ua!G7oX4-8w6hRR8v75o=2Qnxm5M_d{z%4b%C~xHx%s
zsWlxaz@<u&*>IhKMQWbUkDeLr7%`}zmYwC%ll8JlKb@~{XVoJ=b!KVz`Ar%@QEzeW
zzTu5v(k`P4kT!#f351TM&5UplWescT!L0Ym)&|ui$hoX3{k2%*;l7TTB6Cfd-~i)n
zT<tK+QVs&Xxx{>0%kZ_~V%1(Fm*KMI5Zmmnv6&*66K)R*=ay6LK(8ZOl8R8v9JMcD
zEn@XnZd}c=N{!{w{&c;SrR`VtH!PN{RXT%c3v61nO&)f^uMwyXXy%raNZ9sz(yQ?>
zk_PdYI~xJJ!G`N@%{B)jGa9(L1EvNCNc-W}`)zR`JTvbUJ}<CQ>h-lMYGd+)sS&*s
z0ehC*ySv6dSa6l9c;q3U>Z~_iTMlp7C39;CKX!0*Rtm!dI<QuZdb8;`le=@CVhxqq
zmq0cnqjwN4?3Dr~HuX44$V_{tvHe?c3?mntr-9GH^-bs4tPHAX)4fAaoWVI4^9mW7
zLzULJHRF9KL?EA0QzAJ+;sQ^qy&LArlP_r>^|f5l?U?Ldhc3-n(TcKlj=xm!f_<)T
zW@J_ZKcY^uaTKN8wDIPj^j{MBs!s03KBa7R+bbpfozJ}wvyo4%F1DYOr7};9X{mW3
zUq)k#_<1(iVm;F1?Q>9duyN~MkM8Z#v!@Si-g{3{q60mqf@dK<>_QZd`k>OZ3mx^%
zno___C>b=8Kl}bh*l@(&#a3Q|`*)4sJ3R$T@4Z!3HD}Secrj9E+eo!hz-|W3)l1nm
zRzlDj=H!0`e=?AX&aX3VLBJimYS&bG{>R&D!NqKaaOOzT>*c7KLO&D_#gCHEQ@frL
z?7Uqy(yvX-T_W=SAX7thi#8=9Lv+-Ke4XNg>)DJ>LDv9QY3|mk`ccdDDAQcQY7BXA
zqQZ6?%852Tz5Ct*Dl7U=C~iE{apzVM1n)pT=PZDm<BM=CQxxs#lSKC@yOaBl8g}4P
zbH}MwGrky><?;@^x1Sf6arQB;&pm!wtra)+1a5zUoIK?y4X7)4_9S|novI!Y%oB_6
z5xY9|rO;!wGd{Vjc5xk!4`01f*`S1ZS<crm9+mCS_2hXjepl@JD|?+-3C2_wmJXPc
zD34)sen(wFV4Vq+AeJZIZfoAZB;h4%R;N#$xwr5Gdl8o<t*$E2GiVx36c!%VCsa<W
z5sU*;QQl1+zP2%TNp)3Q2ClLl1|GQ7drC`!RP8jiFJk?SYq+ov0mo`G3DS66!>I4Y
zkHIx3Fl%Q8CL3&tjs;UlkvTTBF&IQ(&1qcei(5inizj+^d0=B(>~0smII#QL`KgwA
zWaB+6QG1h8LGco~=*DQ4Hw^5#=g$DHQjFX20ug+D2@CNpL9IQ_P0UA3j=vMJ)=Vbt
z;cDrOYFsK$+rZ%@eb)!KJZjInA9O$bI07-mfeXVox|zeT&p3<lGj6Cl4W#?g`h%N5
z#*-Qt(NpzUVhNQ6H8y&D%vXoJevBi_sq5|JxWzWHPM1@Pmw}p0kVS8h?(<`s91>!F
zYaYCrR3i#;4nH6;b!e;;raYG4Qy`HZC@-*dxrHsr3cRFSE|L+VdedGv=|WREch4f$
zh6`IFkcm^|IqJJjcq$7c;b=;jMPcrGl5&-tA#Pk!bdDiAJGExG50`uQOj??M^gZQ0
z8V7kF+@FS4gpj#GTdJGXhlBkJa4`lTQza%X_+<QMB&Rz(*rV$(tdkKAg*S_LHXIUR
zoXP!8A(S5$oqw_Kd@~6B?>$qCEXt{kj-F&ScVIrfpDcdp@rS@mms!u5w5=V(c^at5
z?Tu<2@R!0%rV;Yd!l<2!+<kTF{a`m>pxZde@m?4In!h|d3c(l0&vlLcqMMdU#EMX)
zLRGfx#f#af7(LT_!&kXR3=bVoPx=5|NCoMysI92>XYk8j9D%VTAdrCXNZ2lCH5S%I
zQj&6Urc6#6ZjH8@32Ci1K0T=}v{a927^KmLr!aZQ?<PalxiY2?B-3TATkv}MQl%JM
zvzU=99o9jm8=jx;y42LlrHvDhf2PcE*JGKSE5gXYaIC%^n~!&El8o-C^l;`+uiBkx
z(5spie~9J&I(rSl=l+d%uWoT)ufPQvT~|Q*gs0V$*Iw<IDH**{PC`-3ugpD7Pa=No
zY%1&2N;SXyJS7%y?9DYAq}#If+>++!>B1Xu`l$0rSIJMArr<%_N8JV1=e{3V9OCEg
zy79wym5>29c-0zd4OELJe;n)y!bW3;yT-7ECB-npnZJK-UB<-4EZ0WIhIIeL?t860
z=guL;EcXdMmY8IP90`QL=_E=xY)mmLDZp`(4H2Fa{D<))T4SXc>zzUTJyLFqPcTeu
z=GwSUM*AjQPzIkx&TWyz5gB9QaS12kuFC89rOulNIhxW7af~KmTXJ6dzVUE6sg|A{
zv?raT!C!nQ%d6Yhi;tPRgG5^t<qq5*Xs7(iaHGj)E^-g4>rRqZfqEe7sbH$-2fo`|
z@fk#O5{CK~#Yr}gag-ribu~y}SDUq6ZM0bPeq1v6jd$N|)~(CjE?T|3T=0}>4|^3{
zfTWh8CY7@VfvHdxqe|!_Z{RL~>2W4cULb{Fe4WZ`cU7To_$noAx<EBGEyK55$$MIr
z&+Kgm_nr3fCvLAJ5M}^ii+Ft<9A0fq&Gf$Dp}D1hkXu*n_`Y_x0i8_L4f3{*KKtFQ
zJa3QK+gqDs_%?Y@i9|_M0S#yvg@2AAM=MWz&JlrV-3b;ms*8DroGjS%E8lQLKO-xQ
zrMsp03l=ujhfa?idtc0jvCORz2*JqMD2(<VdW^S{y}!slV&&8N{9MlD+D&!CDGstu
zN|%oW?`$r<ow4`op|v|J975Va;y1Y;yx@JZMJCZ|Y!TYSVD-fIY%9W9G*Nl-q3Wem
z_8VSp3BDL7E$v?@$Txa-XIswO>D`pDMEC-_kyHS)@(WU%8ZfC3j0C)GMI(4H_nmtD
zmbmj~D#avT%gq`y?(p=c3Yvs-WFP1tN+39hmO`FKbMI%R#z~TAt?rE%IF3wMwAO?z
z-7E7y;{N+a8pGe*&i)5>UGoBvf0cVN(i)l1Tb{GFxqRNx;?r-Kvu4cVlDotOb_jg>
zRZfms?8tdbV;d6z@!jHk1b+CBSxm|FqRsis0%A%R4Q<XJJa2?Kcb-{Zp85YTW7kx~
z^){$0jgX6etXlRcj+;kdxvN~X(94(8#hQ(qv0LQrtX)J76>LSGa^*=JpE8Z$-6&F;
z)N*-ImC0@%iskcK=OQgs4yk#rH4yp2Mzi>QSa<xC<=tM<TRl797MkVgQ`Vi|iX+v}
zjiQ{}6IN7h^i>UGWc8e!oalNxH(Q_OdGNW&(atQUTQB>2YK$Rg^1-nZIJ>0u?Fxt4
zsaZ-%%v1?1-!(VT*B679Z)U;oJWFszt=qNoQd(z#ra_^csP+)Yt61f|vB@8084F)*
zh-=kRn#+nS?)A%~^>+lkjV-C=6lx&GtyZ%*f^Q-8?^&aOaM3$Rad~?b9!@)`m%*P;
z%GY^6;Ly^@r@X;)=xK~CPB86)WJl=7eI35{y;!(feO>j=brb9G0TPSSm+?FB&0Kq;
zg2xmvvbIbu#lE*AbsiPBV|(+YpgJYa^w7ZR8!v<<Z*TC9?p8HG*mWM(omkJ!d15<a
zZ}zj^*a2$!gOA!HkDq1j74dZ29DPIyDJv#=ukuA%*=`OM*C*Gi1UC<U$da-w-+StA
z^MhL3{cI7M^<8%TtRhdZxL)7Yyf5^P^I$+lPF?NL!PSO`I2#f_^lO|Aj8pGOs(wP!
zR&*%KVSE&`>a{HMI^l%8tM}Dy^)#|&Z!76hv}E00_pF=FlkCW4VCagEObin^vnlRz
zx$}OW<)(e>o5JH}Z2J(aW_B=sOeETK?3Fmh+P?2mOf|)^3AM>i#h_5}L5FISb3A`F
zpZ@CN3Teh8TWq-k8AZ5*3w{04Ql<M}R~cEKy|Vue=eynj-4Y}2#J%1J&{w;9Oe<Ze
z0Z)(DDye<cUW!;{zx5DX>VWg5IkswCpvQ%NOvc14#Y$3i^;ADGgELie+GECTpUZwf
z#<}*V)!Xvbriht&mtJHkuTs1-`;n*f4$nZ-Nt2dEk-plh+z0P&ecQP)n5uv5oc*M(
z`FjNu<ZXRB)i;Nn&TPim)iXX*x$Umy<Z#?t=hS_$%N297@RRB&6+banYjnU^^0rct
z%fy|M=tH-xv-htLedLM`DC4q<yA(4b{J2|k&lSl<D=RLxY;Qvlk5ca5yr~J<Uc(p(
zB}&@Xr>CL@<ldnfWA?FDG@Kba%4#fnY{sLVA=_`_&WSs1x=B|ud02}XAG&<dcK2nF
z!(pt}ooU+Rm8W8P@jWM__Mk|b2Kzy4*N=_@f~?O)d#(o-alL$fqW6Be7UHG`n@h$A
zYBS%YfVh7u_mGdBGJipAIkCRE8b0yF%lggY&E3#Ft2;$T#$8V9s*29XFUzgov#zw_
z@HLH($h$@fsg^;;-9oQSO&*>*c`RkR<l0cI?pz<s*;Nu*$?ME5EV<5Zd8s4Ry?^`j
zZoU%*huZo&-Q2c%jW^bD5ua9<SIC67x3BUviduRw&VZD8*}f{$c9}-A#Zg0)u&Lg)
zQ?nFH+2yNFE-kFod(@}$$cnjapSa4t?I+z0If+!I=IjR?isu5frdF1>_ZyrSPHd}-
z6O<!xE<QEVkUZ(KY2RenTCO@w=ELO%dyZ_;IU@8vZq(LcW8m1**1FaeHd6(u(m@C(
zB*8vWl90G)gM0f4Vj;=k;eioV)ZsDrsnQ$+j!RDx9P$f?pxiOULkfCxT4N+8{&B3#
z$@MBJn%cFj4K7C$wq_#Y{m*XS>G$zn;%OEYv5Ldj-<{lZNWXAz*TqcKQNrOYmX@>c
z(w8~Kua}sM#zwq<_T-Qg<mnxlyjgSAjxBrwlcEBPjnA*ka82}+pFFtS2NT;Hv1hsy
z!Fs#&KIap2pX*Q4`uB{^J=QGO>@P^Dpezr6%5_(7hxB!wQmzUnk@Dl?+gmyX^IsiW
zxheaAbUaU9rYL_~$pt-qJM9w}tU6A{_bIvzL;>H>6W%S_clI5sF=}FGyYXuEah~JL
znTw8MA1}IIC^08-KX+WCkL_f~mXpF%jfQKDBbrW9E6e<nOhjYnva4JZd)+IaTZb%t
zJ0Tu^5jDml^1f8Q?=okkMPC;G)-l&OhZ-aAlNI-l23-w%p3}gh;Zl#kDYyHvcV)so
z_YE`e@>cjWxE{+n07)->*P9mioL%_*sl`{0X>GDuPw7eaztX{Tp*B7#Hxi#$W`D7R
zWf}Btf6A*eld8k_-t<=sbPhb_U(Ox8d`eL%6k>1jW<D_R?Cs6WgM;@v_P<+ZD9orD
zB!08Oe%0fA!4W~b4^@{HhF<Iz3ctcDED?Q><A#99p7q$1jCGT_OfU4=%eH*zzbJe!
z$^q}>4}0|<&C`sRo9pJy&cCv}>-d5EnQ~{9jIycL3s3F%CDse=vpQ*aZ^qiMG>7JE
zf^E{eRkK99Rj^+(=hjf8=Y8LoVdh<W2*r=_&oKo>9~(5o94=~Xe{m*ACwH@HU<`Ew
zQ$^3V1flbW_gHd|Mfh~3Idd$(dqssO#?&=hpZWaj@w~wqrV1`!gI;;!4uz-KSiZE{
zrxH%qQB8V2-1*lY9oy&rX6La`HH9E6wKcuoi610%*AL&^2DQ|s9dp*Rk%clG5r<5q
z10?ltN2o6Hl3%pDH9*-VWalvy<T_p2SetraJlju~%Rs>8SoH(v;EPB0@{K7kpBCBE
zg?q1Nw(F6VxBc=<yykCpd0v-C)7o5m1l18Z0UvYSeL3Cd+?x$FCZvK^-1n5%654R_
zp{ttDkz<xZIVLP(P3<SWAD!5~CyJ`+@#ws&vX!F3b=_shJaxTd=hoSNL~LZ#bC;Cp
z(pYumMXsTQYh@4=m0TXm>ET9{dxtd;e#3l2iZ3rsioyFeHc3oR={jnbr%Q3U|Fn-V
z_hKa<@xD^Mb6YoFC2r_F?sXwKy7STM!F2~!?J_&J9%B(@64=_AAfk<6805V7;-TCn
zlij;$!VY8IK{p?{_?7PUNqw((<Hep^Y@rj#_sidm@iMHpM!mejhL+|!w~{xMOGB;g
z7=b0`jZkgYfV(8bVaB+=g4^&3%yVBIjd*Dl$-_mU?Vk-XQr{TACT<fZQeHNO=qqwN
z7SFXeTUz+`z=zPXHE)+oQ(hiQ2s<yH!!DrB$MEi<^pRG<j>X+awltmRxv)sOc0@d<
zR)@Vx`>sZ;q^;gb!S>u$?-UbN*{6zP`cuuHHzFQzdd9c(HSuZEptORr9(;aCZjVm4
za^gGt#&S(Fu4Oln7n9Cn+d~cYanfq71=Y@WB+v2(w-UG>ZNJ2|{_N6wR4DXL3hv&$
zXBeq$_KUl3(}XcOSL%8MIV$ExgdS|&z6m#3w|Rg>u%~>)5gS_Ge`7T#v19jw)trX?
zrSk$}YG=((+E_4)9TV8OW5=iU++}MU0cok7^Q*m`dv-|u*419lsiT+{$yGU{*!G<n
z$H`nPRq~d#Z@k14q3E_}$$K98Jpmz^N3%aREqza&%YEEAayVZ9ZP23wi_D}q8n<P0
z2OT)MtW1Q{Py5{6s7IPU<K<;tNFJ42>rEFcZdj9`zTUxfPf-TBQ`1I;=gHYg7UqPY
za`9y~$?q__x;1`>%TBF`T6Nd9{J9H<Z}zUT6L)$Zos>K%oTIBN86AAbUubJpzD?;H
zn~EsK7dPBaPe;D=JW<ScFK<|=>=c(};yWRhCIP{{RBM&qseRIjmns^Yly&kB+}m?V
zxcyAf<8?1qtFAVpaxD?k-L9LwFQWAHDT4^My#hlP)j_V~tL|uNw3Ya6^136e*ZpGr
z>?4-fTPpvLvUg|}MQOS)w{6?DZQHhSwr$(Cakg#Swr$(&GwFzq`0n8Ce^7(0T2E%J
z%w)(}D<ISYmzoeK1_OhU&>7mxnJPr4x#x_+u^zU5z^)<~H4}s>?V#}lvU<>KG7cC<
zVMLP_;YM2wq9G*4$`g}G6tWb|F{=;XF=OPQ#$c)=YJjTIKuiL9DAa?oNU4YHE+B~S
z8rWFKk`PqOmz7K>`RIowMDHyMib>g-iriPIN|3Cl-<gzTTnbBki-?>g(9gzdXbKj^
zsO9(|%0op5V&a&>R4xQ<(V}-^z`{X)8~5)s;L=hQy`A*0B5HrYK7vBW;1a=%OJ%^l
z%tsA=Bmzh2gHr6%LQP82?2AcntO}Sl$tMQ9HWB4+S%_r9&CctJkLbS=*`w1s`%57$
z5JfT{nkp{62F-&|b|4NGr_5yLU|(0JSAh(YBPh|Uy8Nl{r)(-$M10lJ_5Jnke1E0c
zo8PSH)wU~EI=<fWH2?4*hKo~n0ddr_D~3y$e~)<ELlS8y^LR|cSw)(IsK4QY?``<)
zse=}{`<(D}Y^M(dN#2Tgn$nj4PDOcS8w;^oOAb<$j@6v+_iWb8!(U^!6j8C$#iO*`
z`EePemGyFySBZsdo_8Vo3}m*F#=e@^Ht{C<+|r}BQ!{gUNd}y{>s!IEm(8p1t4dzT
z+Mg00s7&MBGGHNDSPs#We*&9Fz=N>CjbK}}G0+)whlj%L-s)W+aA4t{+0K1;bbqJO
zh@WdyaK^fT)u6Bh5otKErs+I&&u0mlQRRb!2%<Y+3$*L9OO~MxV(Vp>G_&cO%D0J(
zXya-vw(Oqw8XTO1fRq~cmc1FvI3`EP$K(1UI;1yx)!%u2mBVSV$XL|x#tUdp4wB6*
z{5Wsha62P#I1RID#>GE1wRg~x!6Lq~UED{eDVB9E8<$}jnBgHn(HwlVl-h(7af|vy
zcY)zbHaK0?XxF;ohG!wVg|pdOYUAIH4PW;P3}6enZ+{(=x{lAN-pu?%htla`MXVeH
z&hZQ|g$xiY3=nJM4-D5bF3VfdBKTFqLfRTblyOh7)9yZ_d{WbRh{gjvOVoIUJjB~K
z53)D#bY179Vu8s!2ubjCg~{SP3t_u?MCb+P(gGx{b%BliI)}!&5w^{HbXnZQS;JUx
z+J^Y$E(0uEs(7i_2_L(XVkmDggJe(mN`OjEkOEd$OWJfe#1!n)`M%O`bK7k2LNda8
zbqL;2>)EWeqgP%B%CHFFuPvkm7p)c9-MHfG8bDvy{rf(c_w2#^eMzrI6>9w&1jV`c
zcA#f(JR1`6D6=<0cpSM_L%AnY@pp{(Hnhte`o?SwZtX9~l^IZHDGS2j+ZBRG?6v!A
zwXcqgqjP>Kp8JA=d&ByPk&C-W`_K}(tf6i)wTp|Tb24RO)WzG&pzQn4{lP=6=(BhG
zheqv0;z`p-fv0XaNByyI6Nsdo4iwkB5~R}uS~Bb8r>NsbC_wpO;BS3}9_=nI>ekEA
zYm|IzrB+^^?zRpMee>VaK#e>mpGs`g6x~|eq*c8g_b#$idr+4ROo$@_^@_!2-r?+v
zyS;oG1>D@pg=L55$<5V=+wqDXF6|tja=VM=;a+pm+TLDo?~h)+5L-(LQmSTZ#Um^c
zJlbq8A+;UFgH~irzDcJDO}|w1l#B-R8$)Yf3S)MOO!H&1zGWpe$MLgZ<buhnMxJGp
zQsq58SH0B`><#YetRb47X<>+X4}<Sw;xnT_HI{l`rn<sj7rJcjjQaJS#s)^@vU!A#
zuDR)qUZ8WFvMzH`4+GR#6G_Tq<FRfW&R^j6N<B+E*OD6W%l#m>x~T?~Ac5*)L&wJn
zv3)=-h{hwsN4?yZ5Cew#7F);d#>wIzs0nwqzj$i*gBr^~gnVW`UT^0{ECefdz>cpM
zY+&)+*Dp$?|3YMJgZ!*(8RhF~+J23&+T4m}cFV9EePY*o>og$1i*R%6FoL5W*)p7*
zOI_c%>Hhw{0&>?4{@mkazhIVUi$}g`&Jmcd(}oadkBP2x8(1eVpLig{?P0hYS|_aU
z0i0x}29m~4^x&jDr~Y=mz8f>HWw#Of8$8@*wCg_vn2nkJ|0BTvZ#jj5ne~4a*lo;B
z$E}uUUH$%6g=)0^7y?d9uA~abizYc7&BgQRoO=PC8#-FLcCz+FH?HqjygdM&#6l8j
zkIATTUJi~|Yyzw*V5~3qht|p@JAKAXFiJ^O_5Nil1gh|2A{DXzWhQrgC^;g>j&KxM
z`k%+atk&NS4&A%((B7_r+s(<<UY(EQ;a&Ya{qCPuVb0&t;dlPsoFCWc#n#$ljE8u{
zf&?H$8fZ!5AZMW*mHq)@8h4pQ6~fG!h7L|s$u^n_R>!kHuiHm$YpkN%P3F8!1FG8A
zuYYeooTRrclcVY_2r!bO!&x-yg6q<WH-&2jWD3(YyTTbNfF!3Zqd;H~Vwe#bq(lbN
zNih&J0Qvzhz^Tq??8K7ecgVQqW8ud|gQ4T@1H>AgN%kG&PZ+3GWfoZODe`cjD4a=<
zG#k02XPMFBNfO{S4M>;{Nlv_t0>Y<7KuIl)nu-9K6)BL#kPt5ZaG!eZd((H{l5#EE
z+FY9I5!vf*86kud7e<A8NASCK%x`v0W}zME)KPQ^i6CmldUVor57B4^UlQhoD1wUc
zLDeV`DaQu7iC!Doo0czcN)Gu}L{&H`YBuRA>)cLfG<q0A)p65^1Ab%+w4o0tJQrlY
z*&m_b8G!6JU=ZX^4@d8#^P)1})AWiPiOCtC=+P*T$8z3(I!1Mqw2O@lObFyzmPAk_
z7T+r*w_9u4&p;?U;y`z3eY|b7Hl~J?L<r=9u4%X*{xC{y8JnEYN?-<<fNBf0A5sz-
z$0|?V8=txpMcM4$OKY>HimnJ=x96>ch)S;4J!#TA`fFm&TpCAVE8;0nvw6544$)?N
z-*V~lDrozJM1})lN2DW5B3CK#_Zg0R=_PKj0=3ZcZ?&i~5tKDa>-cW8;@YMT$8Z<m
zZg+${e6ZMP`)R&;KKEHFTa0E7EhXznSx_k^c{PbHO~Ovb*g6zQi!4n71JyWflEZW?
zjvJ`jDmw4B1rf+<>!igt*$Ca)I2Cm&&=CO~bj$;sZd@O2>bh~5${6@ySQZ<wMM_?6
z9e`sXs@1HYA2%=v7E1@4Ki3l)9Zl3tG<Mmz7`u<nZK%BGlF3qB99P$Hb<c#wI7>+5
z70Joa{>j2E@Frk<6Qo)#^O1pqqcrKYKJx|6?HJ34iQ7iq22&TaYK&@Oi62O~V(Snd
z!`|LnYHgcg+c*jkbXQj2uBy{PKx;qouE|=cGr&ntC9R3hnsVFCq+qC=*>Cdfq~5N1
zZq&pn4PzeUguP`GZRzKwruCM-1Fv~Vng0g<%R}wB_?nPfvn92Qfi43E1?WN>P`@;x
zqcZ88Hkjx8$Kol68tAiy9ve2erdDa$*DI$Jt6`?HYu*Q!sXq-);~piGm&J1Oq8dIh
zBRS?R|I7c^IYa>a2y8DALV&;$aCDJ@9v*@bZzOQo8}e9>iD!6V)+<~w{JF~|coHk|
z&9~<+Sg0JGqMqG<PaGJ+mBoLL7?=h`R@V$*#bM(z!hLgq4lEqd2=B3goTeNE=A^pm
zj5G$kkP4$K<Ix!W6=kZK7Hwus6~g~*;izN=s&{Z@F|J#osKK^&@*aXlR`8?G14aW}
zM7sk@Y1uY?`%HUOjySY68vGE&hd%@$je<WNK3HjMlusR|GSYD+aQA!8#u{`<e|;_x
zHpboVve3pRO?3y=>TPH*VP$5p4)NJV^8S;sTqktK1R)KmqjJ&{y>cldOUVCaK2p?b
zX+zybbH&Za$h4UfNSDu|HK;|uwg3yEYkB|yyacLb*9@1Z$>seJT3%Nt$D8l`88uia
z$Xo~MuY0_WJd92m#2s>aKS#O{h%8=}oUc!L@=b{x<Wyk1ALUmB2p*Oo-8MFGm?Cq4
zr3mEf2td6g(TVNL57*8jE3iHnM<42zxHA9j_v-qq<qm@yB|yoc!>Uf3*e&esV6Qjp
zcsU(Ih<Ft`ED@<)Zo$NBEIlateqJ!)F5=Mwlw_&W@3V6O1+3rC4JKFuWC9H)_7U4`
zM@NQ|-|wTH#BbDJcyWpda`v#_?loIy84HDP*h#@>zo~Sx^hJ$s3I>WJz{d*+lDVUf
z?DYP~qRDNsB9q<2(u%BVI)1k04SglOtKmlT?3S+e<iPel9%3-E5<iLa%dXGCi>fR!
z{1c#`O2|SD`1PB+d7ZSdkxfItPmL8TyM95H?{{aT2mbb@%VlqtZ(n)Xymm^s5o$41
zM8?4xkM>%sRfy>&t+sjVxetW7*@7*$F}tnC3{0=yY^JmIDBnLLxY<?MJdP9(j^`4u
zgS7ZD{2YF9nLIn7%90i+pOyI;s77v}g=TYBx|=%}gYF<}DK4w;h<&|O<}S-<()^sj
zqXV?hGvzQe`DUsNh4GR(fd!e2iDD-e^$i8UvJf|TLU;R}A0*Bgpn3Y<4zDnS6?#6`
z%zaaX+ZNVQBnexxHn%U_ZOIwf(MbC_p9caKYFPX?n9DTkFAuI;TFWkO)zYn2LGL3r
z%`U0OF+NW9y^}h%uer-)qvf!Eu7`P8r`_vm#I@jC@*J|_6JJlo7m3JcD7H5<&nrAn
z6&#TCP2J_?OG1XI%HD(w&@tRLYV2&8uv5-u?d-htc;RB{GGZ{3MSP>Htf0a{9wVeb
zdDwY)>VZzeiqR08qmaZY-H(A~oE5<FtR(0)Tis}`wYVa!*S=YuMF<yz2v7Ry<iGGK
z1{UuB1raQa{{a!qoNWIqL|kjy{L9jjes1&&;+XbK--&w4+HvNJEwNtbPK;jtgN$*G
zbZe?bsBP3Y-l7YCyM^J}smC9rvjK(yPvQ0cupGgJG0)nIv9!xrJbdg8pZBx#n-Gml
z*0f2<r9~)2OraEH$jG6OYi24sx%+y4O~Xf@@UqED&Y)E7XXyRBZEr5B-@=*eEU4H!
zawJ8T*!gh+9%ShQeRt;dj`nZzUSD^r@Ai6le4Oal-n}2k)$08?AAsP^<sV4!$7fl&
z3nHJmn1<HT$jhnci*ehrg-sQ&&;r~hNHYp?;yc!4HP~cfxs71TV~`FooY2N2k~wc}
zcu`pz6jhXpi?z)VDWl2^nG{AWO*v5z8TMlaQI0(uD&dfWHElWt2!uwv-C!RfBs6Mk
zS9GGu&|}E{`lLTDwTN3R-&9x|CLf<yXs$tpJ(@MwSesX3d(+Wh8g%z{)=6u(uGEe#
zBwc%MH=*gfsP_*UtnX+lNWy%1x^$6r0{&!|M2*lx6?L53+v_eFF0sCS<~r5yh#jQR
zvOBct>h2-P(ew-pa??{mBQtx`T!diPJ$0<BK^NOV?d@EN_T?7c_#8LtAV7ler;I-5
zz(&^C6DPG;X?u$tG^`WAuBla5^R2U$0njUgvKI&rw1@^}HVF)PDvyBHOx{36<XxSR
zEuhU$h-$tQ2j7SAnw)p7jUZEW2E=v3Vu*3D5+c!9gogE2Phzp?q3XVmgN0eNK2qbL
z+BhRd0h8}^f<&Xc>aw!CG9LfR_`?U1&vJxGD8LT^=0yX~#{o+gkApi9oQdG{!C;K^
zb<(I+8gWH}?te7~uH6)0yUd0wjhx#S3)Ymc1)IWLna(;pr!hRoGh7FsL-cq%T49ct
zH7><o!;OvcaDW6Hsr5+}0thn|geM|xpDY1~Ft9xLa*evX1jBE#MMHTvQe9j(q`PSY
zz~%%dxmSGK?He4g?;v`Lx1!I0lD3*?>--Ma*2>%0qsu-zOn`c3<p*Na%@~l1F!r^k
zRHGa6S6_ED^khFB>B1n6m?qvwhez)#xjq1}j4{pE%>p-F=~%;Tl2y?=32E}_<RL*4
zj<bXyhVZjd(*GGNBb$OzZ{yZY*KIKnxl1qr2hk~ARsSmvbn?(o*7w_SZ{{QtH@%hD
z?#0t0J=!2LlGTR-2>XVB6`;2FxEXKG3md;i<e_UFiMq!i4VVGWK^9T6hY>x9c^SVp
z@PHKa6=$P+m~;&ks!Dk2FySu1L7ZbaCX$K-@Wf?SuB(R3V~~%n($cmSdf!2{#Rqyb
zuGrPfzUj&MXl6k-!8K*ql7;nhjFjco<LWVrNyVfg;xTI#rK&G5*BEwc|3(r8Iezmf
zjJ%LB$zK<~$;FDn%d=<?q`Iubt^P_wuB&)xk@g|$X=!x{M||`b_NdCNS)a+IGOB^9
zb0J4@O22s)<cDg3y=($9DuVn75ihWATTM4A*es%#@fG5hIDvVC8UDrj*6YP&$1ng5
z7wx1v5*tScBq1ILo#Vk4ddsTQp&UNqOdNTKteA)^?H@=VL^<9(ka8`4P0aAOfYkma
zS;dJr`|50$x$%6yc}6zx=Hh+3HYlwEc=fkfiI*o7pHT;orAr3!1vUW?8jHftqJjT;
zSD_tD!LHK;z-0iwliVu?Jk~A4;h01MPA?Lnzy;73n2ZPgKDKmEF$eJfgw#Pyx@B6r
z?eBq~48?)epmlsZwueOIz^~~AwiNi-EK0+<4PkC;zETMa7s~fMQp*KzM~Q{_#SC=>
zPj=Ij<XZW-4PzFy^!&PFhDpRsWJHEZEI=DJ114nB`nqybIXqlHjZ;6LvNJM{y12}V
zbaj*aPRotdZ8cq9`myfYGPuRDbph`Dy+8V%-xoqGMW^^^rCpK%7;ouzTU0D94Mq)_
z<WY`uC_4P@6jH6CLmO@+-&EhS+AZ=2%CkA$<Tx8=pY2K2rpp4C;q8-iGyl9ZnFl<x
z41NpP8_nLWKJy24BgMzg`rb75%e%q-xBkG%+dG!O=iqC$q{sOa%HU2j-BbGHeI7uw
z2O&S_N1p#YBY4umw&$s9@^|?0=lfi3a?iG<GxkeVG114g=wCaWVI?SZ^w-Hr6-*6p
zWWk(Ij3Nq9HP3D>S7+S);LbR1apL$_l-<!;?Az#fQqSFvw|t%usSFm8MH$mhTV(vi
zTm~wO^rmBTb!3h81#uOdF6jzU8Z~0VBwEdf!@u#pTt;rXGqg%Ruk-T<DC<^A2U#V#
zow>GVAT3_D>SHuPnMCk0fgWeYFBes+``pQig0d0QdA{T7DyEw$j|`;A7LO%2bu*Pz
zRkFG%uBfRQ?9>z=ku~edk1hKBg3B7}h7s<V1Eoe=KPG_*3N)<*1yFxCdtio7O|41u
zNw+O@lLJCYsnAaA-S8|s2Nqu;2h72&`LB98w`^~qgF>K&7*Y<+j_Twy5sO&A(jo~D
z??vJe&=z#anPnYoZmf`?fXFEzmcj<{2&+VyQ=gtZ`2NBKhn7$!1Xoz+$6RjwF_Njd
zI~2VwdH0>`y<mpqz2`NhKzM*dO*og+Y#@)4OBY9+nFeSbnGndZ_Us^NfrYl22p?m0
zPbQ#p3wc@RltsInY32hl8MT^imue#+a|%!<C0B4BjkD~Q4Zvik>(9gj5Qmwuf<T>k
z_5~Ra#%s1JKYwrQTpyL2thKRtc;XucQ3J*X=mxujcb-_ke#A7!+-66CowsiCJu{(t
zZ<8wtYy*&nOKmGk;Y6q=lxnc%fIqau3~b2V)ZW$f24|Xd+pgYnYlYLis4Oq)Mjd?3
z5rnQj^HUBsPV5FQsJmX{gnIn@DBI((?fBY9`2*(tic97Gq<?AOMv(?5lC}oxVh^Ql
zaj*NfMDiL{8#!}`07Gx>W@E!^gduTTlf@t)k##Wvzqu=xC=-IaEdv@4k`9VTMfUX<
ze@5n|MmPS?qTPP4{Y$z=GGyDj$zdCgF}_Fy8&L+PkqH33D+{Vzwh)OVx_sj6fu$cg
z*7kT=pt~CqK_Tju?+D92)j{QyOIwQlv4Qyr43EUvkF7q#(H<*RJmC^Ccu{7UAsHq7
z6tk@b-dKFR8&3QDny!m3=zuH$5N!qojg})WI-*WSKFXnxd4MQ}?ZJlqe)ykLgHwUf
z`p75&vx6HY0OP-8upoVJ<my*$1OG{Dtp7o4>};I>E3L(7N;_?GAoZN8Tfl1ztFJ@g
zc^jT4%VtNLC&^MBZfsOc;ZQ)ODwIf6OK86I?t*myCN>-lfg{_oeB3{Shtsb>_q^R*
z<h9o7^`h(0vn;kxp|`q0r_eXBnDb1dwN?(dib$q5Mpe?Bj+bWr>h*TEC1T3=cz64{
zJ-)o0@ULq8dVU`tPH(cKi)L=^c>l1o_56OJLicX#`1w3OU)wPp5Vv@LeBFE;;-mX*
z|NMOTRFrI%VjgP3Exd0l*KFsjXM^b3mrMdhdK02i8MmWF$t24ZdpdN6K(=Q6dHd^@
z_0R8O1Dopl39U$GhLKy>Q-3KdeB5*M(xPj8Tyq&}y3C@Xt*$L>MJtQ5xcSfB0@nC}
zdg<`?z+g~dWw|RWx=8I78DgL_M!<8HVO|wni@p`H9})~{nOk+0Q@M?4Z4Iv7uDd{^
zDJ-3UfnX2^J6Q6^A#v}^4qZLcd>ppfd5{c-+tv4=q8Db-I!-A;q~R#~bXbQ*p}$3l
zTQXkPzC?<!>n}<Y)7rUw?mhELHm(K=jVk%1Q)&PQ(QN+EzI^t!lt>P$jKHcj<eJu(
z#idaFf%E0|PN3Kr4RU3%Ry~Z|3Am<tZ-azun4ti;ztpkK0`5jiWm?3{Pb}^>uJ5(L
zVk{VrP5um8ud=Z4GIoc=p1#ps!_G>RK=T#ezp5De0fQ>fy8%87Em#Zw0A^bzP9ciS
z0}?zV4!5~u4ou0$i$+!F^+JnT$bQ=>F(y9p9JiAgtoEVX>bk4I@BA6RFWa95b}g+R
zjd`h@I7s!j!TuN<%jF_mwkBJa5a`=+%7?@hu3_b^*6n)6%I2)<<i%)xt?58Cu9js5
zJDJ-+W>f<x>*v7!)@uu;)sI0T<?ZI~Yw$Vl6-hew69)fW7}@8Jm+LI3Xr!T>KfIHg
zVp4X~L2*3b;4VIEMSSXOM%rIO&K*Kv{!8%Nxd@P|s9t~$uG_wCLkX!2U_6O+rkl@$
z&PHr<FsTs;YqXVrZLItHZI)*l`*4Y5Ge9r|R@V^#>dg}Z_p&CU&xxS-tQd$~2@(ss
zrE&M%OfUNjxvlQJ?%sUYv;}0N>H+o66JXFZ7AP?GhHY1OwC0p!_LboE@))cobKLKg
zMuUVm9f13hqFG(&3N&(~lAO4@<6HfG=d|+;y~@~&4^8b};4LIlswxVIg6+!7`Ds22
zp!{KBk%`@XW?cbJEBhZ`mJw33^C!zq+YG2ke;clI$~`-+VMdmPmvLY~i-ix_VJ7iq
z8lSa78jg6}>@i~~hb*R*`b$;#K0k1M`Q3X1X$~4Sw1&c`&A~!r&B5ew>ta?!r~E@U
z$ge<PVdv~Fb03y)TZ#Y+3~jWKlMm3GJjsYJB!|ia8;MBkc0A|CNj%}p7C8R4i3*@g
z6_4A^jBzoht;re^0Sa6IuOEkFY7^-Ux<Q2~@J>&%+BCBa&SPXM&#R1Kz&_tnb(V(Q
z%3W79B>qi=h#ozYZbC<dk`tWm*M^(ENCRhjPh>kHp#g{G0pg5!Y;DO=18ja--Uz-B
zZf-@zAy}GLXk;d+Qkn@S;%STMGnF(9^1w;eb&I<`so%K4Qn3fA(}VyS<!@SnW|%{W
z%tK*`7OM#Bo1<WlA!KqU2e8C*3kT~LE!ju`rT$B#!e^ZZF&Q|gq@*_3VsX7pnt?Qo
zROU$omnF>a!V<ugRx}mww&9BoFyXbvM<YC5QWC&d61awJPuo@`gE={rT=Pb`D03!K
zEF*D%ET0kqoyRAEJk%$1ekjP*h3B&w5h$YuwuOA}&#3b8M_Tr^MR%HK1D7D@EQgwm
z#c;fgARFk9MS+oUvPb;O-!^3)dzQdxduqb-Y8hl9M<5m!L#&3rm9RalpS_9pd#3|(
zR&!HIW1+K%QtJc`<?0>(ADK=uR_+=T@V;~a$9xQOj^a`u0`bzhvlWJCUPBjFQZ2wT
z!1X-2aFh=oa08r~alK@kNlI1@R8-EoWY^>UQi2Can=)Mr4G3jzbwP8`1og9mb;0jB
z<3W?g*Qa`duFx7{N1|2&pU;a3v}mt{<t>5>EjKxzfL635rz6fFe==+Y8yf^Yo$;M4
zeMNY;07o@sYEe-j8(<92V#H2&HKX!t;Z+`meEWR$IdHU~6p;l1z@T(;Y-R(0XF6OL
zaH3YY33G(ld8n}=XRLu{t4slF9{VsBc-i3AZbsvd9|hW1dBCu7W(FDE4K&ZxmeNo+
z9vU6r-uP+fkFD!`+KwD*FZD&ts|B~UCYV71w`B?Sf{=$ExNvAt^F?X5)hSBpbNlNM
z)=$`gzhY1>CpXl8jieCmMBZXy45a_U^-2?=xGb+72P3<;;&2Kh){P^kM7oO^(DC=H
zD{#vlciqiGFru3k<uLbwQ)WX#kjE0zQb;NPa<5S)=>St+9xnE$BalDHB;g;0*oqFS
z9(_RUUa;h$X&&&B5FcE(TvnK6ns2+D?PT9$7yR>f14aKZ#NaNJ^10Ox4(%z_LZ`K<
zW~(XY;R@x=lH2|jrs3@jZCex7x)P#3*v<jpwv}$~bCExkiwvP~DjA?fv}DNQ#p?Ej
z_XdbAX_V<G1)MP{Kv&-ZUVcikdRch7hO*t_c{jlyr1facbC2$1&7!Okk#X!Z46wYu
zgneQY8P@5!nC(@*^L8u+lA>G4M&)EoWxs}sadOV2h?vzkIz}cTqAG+I+=sWXLW_da
z2wus36jy>%X&tp?a)wM;R)+L&hn%`+48|nm=b<zp4zM_f$0jlnY7Lgs!~%eP-W$Sq
zDF~!9W8Wu!vXcumd2C|#@+X3@;!x%%4cerV`ATwnmvElu?GFe?VU2&tGdjX7yndb=
z?Mjj@PDPTS>z^kf9X;%<+GFu|gVYlm(^M)Vjkb(|t+fF)BW;7gFM+2X#!zCwfi@rc
zf~b@ls+e5#3f{v?rz6dGcWVRg5z-mt4aY0Zj!XLSXzR`H1hD95avXUTEQ6eNJK#|2
z%*5@spDpeTP3Fi1s&VaWQ4HR=f_TcZy|xQdJns8lr69{UN8jSz;*WaALQz9Hus~A3
z2ap-FR<s}!#d}M@rw3tatU~bh9#MjaicWia@bTl*0uC=BooqB}VF&@C`_@V3F%>O0
zXTBNqD$VS114@%VL~irejAjB&<wX~s!V&It+wf-F!#^#a5=5o4{-{l3N5w&qN^KO~
zJ~Sarx||rUKM1tQJ3vALI(kA!A>KnU<z!!2^dKG|PBADK_98d?#C=A!x5|V`A_xuL
zoSh6ywBK|@-!4bl>XoY$$=;^u$(EjW%$PB&@vF<ow3-bFSH?0K7+evot>geZdIE3d
zf(q4Lo%sOy#+k!)rKd~}1Q=^RVRFZlf&XD&Z3vY%|3LPn(|`ju^YryK^C!XcSvMt|
zp8d@8cdpWZ;78CK0etEZ9SR%GBE+`~<$;7>`Wf(@IaV$affS?<Ks;)@d;Nlg=vkF`
zal#rOxFx!{x+U5|9n?&XuL}@kd%23kr<_Ys`rZpmUHSzQGsek;EAj>%N_{CJ0cZEG
zZ-_p)cWAzEvj<NawrFfrs!s;inOT8KoDRSUMX}VlGChllWpR>geRK~Ej3|Ve^!c|@
z#H1p8q4jhHpz*S-x_2g5UZ}W_))XazBl%<JR`v!`K4ypWZ{oh!zbEW>8!-X*KM|Sp
z{~|IY2j~B6GQZZ+cG?g_@jX>PLo};;)+4<x9+$b`u)L5v;F8hD;in)FU=&jESC%jN
zzUHme)m4)KyW*H}{xC9F=k==DeUD{uErE00dGf`3iPd*Hks#r{;2ew?<Po=rH>Atn
zUu(Z|3Ital)Q{6grm@=ejwZPm5bHTVN=rK(T#O2k2EAd91gp8T|NZ&~*YN(_sp|E9
zmFnfJU74Ym)9>cR?D;-a@#A|xemc4e>Zl8giG=;u$s>>oIaI<2U08gDH&5X6LpE!@
zH>%(tigHPYm6*d9oqU$6ogTn$WMzSNNDlE_rNIkv>h6$eEOzgmI$*(m>+&nc6|7fZ
z3}c8`ij;<YU|jtO9x)Z+;6vgYcY_)NUJJE1(DAZ4LOel;)({BaI4L05le5_rpOS%~
z5${7hPapykGbD~*0c8p-rjLxlukW*QwHNWxvzF4jL6%@rv*x%!Qrv0$vkV%%mtO*9
z*&a@;Bqey^b*v4Exo+=>`Luq`G?8kEMurHo1)?>&bfqy{Xo5OzNA!5i(i4I9ah*od
zmAi`u$*gFrGb$Q;RG{BHs(XxoW>cQrB$h0|IX`jsVgY8*{I9U%+JS%%t8?(~vq2Rr
zFsC6cVr{(bPfUi3J|Ga3k7%%|766ooV{*llvtvjP9uRolk~Ue=U%YA1&J9ZK!37RE
z;DJytj3VX;Blz5h3VHQL+oJ|KNNS=0N_XdwQA0qC>U|c+MLv2zR*X><V>W;GQ6{6@
zy*72$^2OYt5RpJ%T&<w}y#kxBM1-_vJ}~@N<Anh-oixSQxZwa_gjW?cB5G<H`1g?J
z%gCE;e05Qzv(u3^35MqA<F13^W1=}{Oor4M0>2?u@Nc46h?2kwfgJ&qBO|QJ9Hd=I
z2JAkJA|8u?qx2G;xA7SAl7rPMZg9!6Zp9Lw>UH%D2}dhR?S;0kh7+$K9(Ct<LlLUN
zi;3)b12obm{rs6z7Vn;=CRF=1OW0L7kMotl64EBIhszS92tyE|4l;__S(wLWAG}88
zUC{IM>Llk~x(myS75lz-(I(z)Deywx79<&-Fi_&+0?n2oB&gVk<)gx~ucqoz71hI6
zQn{HPKyIh%r;boF0&S!J;L9KI@Oz-MQITWvMt6)dy;4IM6?6%A6+0$9D{AO5<Fm(2
z%@3Q2$<zZS39@{SIV0%0Kjh(1%8bek2b0?{34CzES4M653%nFJr!Ai%Gq-MSsD+Qf
z2Q9BlZ8l-(@A(Nao1Y7KWpe4Q!3Vy4i98kxw+?M=PAXksk=t_JvVL&`^D@1gBa~aa
zRTJ&iSNArbR|Pj^u+^pp>73$v28OP{5CF`M1O}gg0k1v*ed@{DX}0We_IuiZDE&W4
z2c;C+#aZt>wxydG7<}9YT>vk5wbb8K%KcD6#Z+N^)WwEdNZ{}%?+cXIm{W-0z3r)0
z-+~puKYmkXmA>i*rIYp{`7Qx`M81^w4YD|*RU7pdNd;sb4QsU<=A;mY0*2&pUzsm-
zrn(SF0LTVxff~i)eJ}+UV8wI{p<R{m=GemWD%-6Il1^)3A^|XsaBi!t`K<SO90NJY
z5eEEz-e&7eP_Xx(CKB^hkuX+J>a0>@pR4yq))?C&y({3d&xkUqXo^Q@=_BWbzPZnf
z@|6^w>nHkk;2s*8>(H^)SdE#!T4+9uv#L)IX6;n*JUZxcnisN$37N+XlC28-(+8Ga
z6lIY)&<sq0gHm^#R`-p;C^S8_Pn4Y*BGFM_l3+!lX5Fl?7lT|n>5@DVzf@Q=Z094g
z`ZZdMYQrn9*vi7<C7%3BC7uGn{%1>1Oyv=>f6(M4O9jT01{4o9D*3f$?ZAT(bboor
zyz-#-S@avr7oV4hv%GS-a9@6|QqCt4z;F&xJ`h~QB;{$D;l;!R#E<|WkLA!flXh-j
zm*;s3z>-0xgi+>DAWPsisnJgaxj}zW{j*dr$Mc#hPcsA4IY~~HUXtvV1uB8(?t-8i
zda-W*O<Zfh70TrA4Qf%a7a&ktv8j7n+#$+pQ2=rsi$pGS+}%w+J<)?fT<`-bylqya
zb0l?8zC;IJ9c=S^xdI5Yw@@K@v*SM67RE&~({$fN@qNGf<5!>7PPDsJkGF^N*c)Xm
zseD;vmkgJ6{J>B>D<e@`3m-wb4<yF^<EL<2l}ZkM5u`LvGIY&6Z58g{BNwTVHGXpI
zr5$>mUGEoh)&$?CYD4q>A@DR#BnxPzc1lK`1^fr#Hu)VPy7l0l1fw^yRMH)S3?kHf
z{q6}?p@pThSKq*iE2FBAR=_>iIu>p?#e`XZw6Cg&HYE3Sq^FIibD<^d@tB6@0iXCe
z{8F)jUvhN*Yo?DWvdj*-@ypFH1g@nlM(rE_yAXdIP3NZ{qGk5{OK&4NA1r_<Y6-qE
zoU$+~2E|JBGivo))Eh&fBF`{3uG!@y!<R34=a{FJ!`6c;uKBp?!TjL%mVV*9)~x1;
za~;uG6-ITQFLv_L&y4aXLA2WX^EaST?huu%#t*RRL}L8Ew40Ur|7tfA0|WE_THe~$
z)VAAXL-C!hohyPFg1JShhpscUoVUHS@x|OIR4^J=q?V2)A!iP~ZoR!D(o{JjnOw11
zlcrHR+8@2e@jX@OpQjp_Uo5gi7blm8SV<7mQo^LABO)X2Cmqlci86g+Gee<d$%2kJ
zJf+X;)|fg`*?SvJ`{!EP!`sd7@pSrFr0i1U`5F1r<nekpdhxygI)XG8p41_rA)&E`
zbrPD&C%&pTC`Lsk!unmQJe#XzkIQlC;OO8Ys-uaO7{V9`7V!rsjTS$>Bwm~?_P>`D
zfe=+`ESo8eDq4*b#k`YsP6nV(#8mUQ0!>DaH7Uel?p)U~!uh@N{pX6RY7tHFqo`2b
z=@|qn9X?_5`8;3rvWTb0i5g>wO4md~1}8rQL?M@_I1Tz6p(Meu!g+KwMB65_xx`yt
zX{rVyhgFo9L{k4W6+kA|*7%eaCep{&#j21Vg!058VFCde7OJM1v22+|i<?_%qt)V(
zWj>vgA3fQ00ernGj?6-iku6^gJ}i`5g}!B+V-xwc1@LVIW(@11&0qGkmgU|2QZ!kC
zE03KM9EN~?;~}D>wR%p+0F`6c>-TFgDq_bw(1ek(e)|o)$4*Zin1IA3a;Mt5>Y0Ov
zf8l5fgihw&HZ!wX!B+kJvIhIHGjxCNUuvks4{+%X)yD!#|HA_6u^wCP{eJeAN)RUF
zlAx)=uIXb=p&Wr!-FdJ7?!p-Rl{X~rIA>fx>K^^^HeW_582Cidk~MuQEIU%x>uOkk
z8Dt}eA+8fOZjT*ditgsq23CkU{XIf9d|7+;m5U)O5F3vc_@OL?nn*QPU34hQaCPWm
zIxushPCoF0$fjT}tJY$frG6|DNeZz@S2-R|sX%|GQDs<7t_rxHnnum=*Zxbky~<o<
zcs<67e>&1d3ok1)(r;=a#jokZ6zppWsQDyeC{(eq`KfdTW^>6ybrvP6({;J;t!_4p
z=$`9$m@>D34s#C`i!{R39Z(m^%O`l6jUnaH9abC*4+KJCA8&@~wR396P)0h5>gsdk
z`$fR}6X@!qO{;T7T%MUnEAu=50t;wMR{qjCCjj(!<f@-)2Y@w<<`2&w1@(W&*VjfE
zi0XCOGGN}8PBKe)Bm^1X2@@NamHSI4ruQtfp@{%6`PDUD+i2*#6$>(0h2<aoAw<0#
z639Aoc$MbdP04hK&zL{ay`@nSS{bqC@+{{7oFgCX2I{Qie_|J9TgtP;Mt8O6kil+~
zlV{WU1{)r1Vg)*65R9`<ACSYo|2hEXaA;Yi0cWhQ5RW*vj8cf#&}(`-5eE_ygwu~y
z_Bf?L08GVOD}T!{X#1~?!Gb{$@><n;QL|838c4CFmG};Y0nYwoxQ1*R<8&?nmnp=2
z26?U<WI$NZJ-kLOdn;c+q;a;je@?Eh-%<d|(re*5K2DhiQUdeAH+8I=hBB0eG>N7k
zlF>Zn4HOb!FWh<H7XF;XR{|xG!oJ`4?KgYbL64C&kh>g65`jhv6N#^thOQcd{Qfw4
zNYgV1oqLr{NK55>p}YYqt{0b)<MfWtPUn_OI)|>Zty)K2kFMJDV`s9-<=b4<i}O4b
zu7)Un6S%GO)3q!=8IT?R)P0}7Em9;H3IaJW8dz*HSdg5^Ea=>I4tL^Mh+xFog4cz)
zyK3b;M*R?a(+(M@X+xU7Fo1<E(clgh!ERW8B<G@*S{*GNtx*at!_`fCbaqj8rzFG*
zMH4y#c#2h5X%$$L91AY^rNOqFvJ4b(q{YMOoa+z4T0qIYi`MD6W&-PIjZHk4UsmWQ
zV$3XcI0P~m9N5~_L`g3Wo-;})mT;_s%?opA%>a&+FjVHH{MT%E+EgjvvAANH?lggq
zfM{~L<&{PZ!7yhO7e^yV4ajFm0h;P}xi4?d<|~Mkg(2$A>5f6jB~tzmG)^#f(A}1P
zEBuWlue&@X3XuKJ0PnjyJ}N9pLS1+*rjNA&_VrlG8-cS%80m5$_KSnQY3JU0yBHs)
z5S9f%NzlhZU02d*GAJM^4r!P9^P56sd=Uy_x;^I&AM!y<L|Ey#!6AVHlC6+oW^go9
zX@u)*UxYQ~D*&HY0&BvV7^f-`b2WYlc@&I9K=5~%QkT1m2WP$01O3g{{W9j#WA*~U
z6`wgP$mCs5-y@UoUlAGQTDS`2%q@ML1=Y;7?zT5>*+$*Z;K<Iv5k3!>KXz2LvOyK)
zy%5XFO-f=3ZOVxLhTG8h1v~|a@?+mS9vYZ~1+f8+-VE*v9RRrbU9>8$pdHVmeW;lV
z41_c5bqglNwTL$pPMH{e1p(UuosR6FRQzNElq-gHCT^|ls*p+dvOAd@Augnl(Q9on
z*R@k~t4A+%TYj|}p;E{SUIp-mIYxRrqhIk-Q}IYVEVn&!7;D?&aILtn0j>{`xN4o5
z{c-Z$>WipV(|l%7=HbvtGWB`=vW&ZkQD#n|NjGc-f$&HtVD{#Q;`tWx1lVxuZ_4ZM
zEduqumk==S`FnseLg}M!4zl%>*vlyEtx6$AhboL{w*f!bctkr#J=gww3|Jq_o%fVD
zfPU4?t4SlgdTW;<pFbsE<NL_F(BG5!8^{udNTerL^}ky!n0`3eIsD2{FQ;}~-UJL8
zD8KP*?Wbst0TC~IetbW#&JIuDQWT(c{|N~k{}&P%IT)G#S4cS4`u7LMhS2k=ekU=j
z0g(&hen#kWwQ|jJPm-y(foG8rnLbe{wBY)5+esh*nJ>YMLv2|~9)wW$RtJ9-7UFQn
z*w-&_&L2Cgr|X0^L=kI&5?u-y1=LR*+zv5fc_%&w$@^>dHCbsDXl$N7L_vl)W&x)A
z^Y!8C{ruxC0t$yIdP<DAfA_Kte%9q-{;A*VV~o|B)wB7kZf8q3%HFG=aXXn;OXv5U
zkIplQc2Iel&H){FH9<_35J}wlW0QgYH=&NEktJ{G-Po+>Hi{)H3)lUV1VTAJ4j73c
zfJj5wAQUAUb2n({9_EC3Tu_gAgJm`Xiv7SP4{?{J^2?V~Kf`qchBh_jy-M5CQi;5T
zmGUNSm`Ho+q%qx9j4rBTCbJok+h0dGB^`u^Ok!?L;Cevo7AwO>8K?ad4F-`&zz;>y
z^7(Z=mV|}?gHcq1Ir-mw@s$xjqtI&dK4<ZKA*lQ{#Yr`!f3YgEE*9J8)5p=LN&nKW
z4zriAlpy?Z7fl3HqS_xWA!2+vT-gjRK|&G%LjeT9VG)!iC>A%Vr>hW1L==?3{B?|a
zFnN#Ca3M;nh*Rl0R?1f$Nha~~C&k)oIVu&A6hKqrl5GJ7Z}ws0ZL^pn=7q}sV##$S
z<eR%u@4bAix%bj6leVHu@A(2AM*JW)tfxxf!|M>?xdGRn(FAh7en_$AD1dBFO2}CU
zvWc&d3eFGXuFL{>?}%{a5C*q53W0QnVhcOmz5>ZNlfm5=XE0-g@;R9wgZ)6WRB^Nw
zLLVP=+5<=6L+RQQ4^(C(9KuNM0UH-5I#l-G)mW)C<&!+&DnGlapiBBXtja|@R$j1X
zV&!KE06Sb{^0?rAy!J#fCix=$1R4;Os0X2THIS2a>}-u<pNECbqGXLXNqG5ipjI7@
z@9{1eOf`iKw3&o6G}8!(85T^J<y(ng<q>2C7Q%%VMO2c(V#FsLq;oCjFp}^e?v~8^
zJRkK$G+>dE^pj&-b|-8)BLPnYpc8-qgU4c|8O#DTIAO{rQB})0S)^f}V9aZOA*qQb
z?|&^UqnDYq#>Gl)XPGuJ#d3RTh5Sm*EMYTU9P2Sj3geRY;RE_W#EG`FTr3Tnz`K)N
z=KNw09(ia2TyZqq7%p`1vfw>_o@zp4E3l~4bUbtrhLO}4Tp%CO_MPly>Ga1mZOIm<
zSoWxAz+FTerJ0>5&Yp74=cK6UA%UoeCvgWchtMdH4WM;3XFUA-6BuSH5mxPLi7kjD
z?yHpjg746`vn*ZD=BOf<@iGyySCzucIh@3NFbBr|V4=OxD;Ut`_24^+`$DV~BMWz8
zC|+JZX}h0TSj*3Q<*F@9Vn(jA*mzdKqa!;GfDv%my+gv8DD&*8Og_v&+H**{$A993
z^fX-L_q%<;?S85FXZrwXs8&>AkAb8_j&{K8o~`JP;>zS};%UcR(rI*|yr+(zRw*Zv
zi8())^jR)Z0yI+eNLb>f4-5=66fsAV78*xGf?+5HL#qtC2z*`uKB>5`vW?y^m8S(t
z+T}hZnC)IzJ`;Fw<`%ChvNDQpHh`K^Nl7!EEud61&nW;5Re945t<*boXrwiC%@O8;
z@Y9a3Dk9e>l2Agx(PboabBrP~Au?P(8?I}WT)WjXtjC!}2%G9g6<+TQkNbL^*4<u)
zTm8cZaa&{&u^`7b52Bv7kiIn>c=|9JmqtXS^GnJrcujjU9IjI?3fKyB3;0K<Ia(QM
z)*Ov_4O-nGwDG-sVTOJCy}UAGbbfI0`b!hV_A-BuTgbL9kn{&$&vEJe$y(KCShA<6
zhZ?RHPOoX0{8UqlA|7T5irLnEv}CHWs0%GXmoUW(S-mr_ueii+E6)hSbW1H-MLRK|
z=AdR`LR^_hY2R4sWPF>ab4}Mv1tMw&Y1~%jW1cWJnBQE9=7L1uxqmu}yc!EL_GG<?
z(z@I}BYE6M&HOKJcUS_G0#3Y!5MyF0y__5IQ6XXW4NQ<kgnf&a*)bR(7rw*9K&z;Y
z->&#je<(pRF0eSdJ2AeTv8V>@4P42-S}7K^P#)z2C7gOl#Y|$0Tl-Yv!ptl>?S*v|
z=4BwYb%EwA>3ojTrmo`b2OEy^5IZIP`QOQj0xe*%CFhKrxP#ilN!yI|ttBJt0dFTo
z<9hYW;MuMX)SAAv_+aJP6lWpZE9-#1=`1O`{TxP^{s0aVR5zgpbI6jY&7*MEbiRY*
zKCJC*kE=}pM3ln6=rjHfAjSMQgPcF*h4l1@hk6NVX&sWwlKYjyI)PW~d^nW${Pyfc
zc_<K&wK=Z|w7<)@QmT)>b-rBdSc~PSZm|%Mh|$-|GAX<KQ(0VMb{Y_H=DgdZfN(L!
zJL#LSM<%+g0mSFG)|$k(yRsX->W7Vf02`hR5C5gIZ2y&tGcy1G@C8fTU5=pce5%_^
z0<Ua&@T09i5wVL#<y%-eE0RNN722>RQzRwpzOKGbB?(6~WP?aX63qn;4l<k1WS_DV
z>+qnpy`4be^Zwd<J)cJrf+~h1^+#=n1&#+=1qCY(UzfB|V+ybC?*359QbG)%h_w-f
zAcaG7-P;BH#<_HAejhXF)k2tee0}`=1HZiH+$m7zn#J(3<u+K?Fd(?3ldn<-aS!zC
zAPYfR*|P4=*!O!5Z5|>hrevu)GKPUmqVbF=9y}+LGgBc9N_&|<%}QA<l)7}K*YWZf
z?`yNpF3Z8%HSPsf?O8>m{UNvj@35)<vG~6EvDjxhZZHNP{H4A&Q#YP1qChTso82Op
zS`Jc4z&?_jTamDfLNWH4;X@H+uo}a;+ejp(Rlb@0m|Cu?&>HNQ2gW7-@)@0U%%)`-
z<J3#AaLYBT0~Sj}bzzj)6s;u?_8{}Kx~;iA!fNoW;PN|PZhxO`!2D<6hna&KB%V~%
z@Y1&P+Vy&~-lCP(-S)yo`*{^)<qlatYr6(a^BV!=+N@v{L<s{=j1Wo41Afs((=PLQ
ztqX*5hNiz~jjFzObmR|uF<jVl6258x6Q7H4`j-CX1Hy~LFWLf>uS9sEAI-v>UA&oF
z?B(q8p|SQX@?DL{_<<1<r!tnMY1c(w&IMbqUdfaL+{|H4ns3-A)>rR+a_JNx&!c>L
zU9Fe>4x8g{FDTPc4|5P}4D+Q5X_c)K5us?fb-EEjihpx+C?m&6c?6_cwik}!B98_x
z{-4?l+oz+|3mgCVeVnj0cI6fysC`H!9lp=n%Ri@<uK_-b*8>kOzzN%46|a*hX!%`x
z5Q*#5yTVM)+UTw|I8D!<ROp2K5)<&PdD#4pg!B>&m>F5teBZ@8@l`u8&(hoK)qdr9
z@T!|V)o4ob4uLSx1w1Uj_mDu%r}pqlx@ZaU8v-HCZobYKT|x|1)aX-vr{o@gkTFJu
zd!dy=Nl0M6emp8X`ktV~_>3+C*A3t1nzQzY9+n6sg;6EnlT|tJIhDEYtmz2D>qF?`
z!JGZypJUQ48#NoNcg)U_$-yq*WoZ(&g(V>LJsUOvk3B&m>JO(Mg0O-ZAXEITS88G;
zaBavmDclUdmRPOQ`tKa|y|nx0Hc`RIL!d2e(A(IaZ7oLFe-A_-&KN#1VJ_6vj{*8Y
z6Hz3%USO7yCQK#HI$m8byQRAz>tOHxXvet_jPf07lgM2yd-|MJK5JJ@<YGZtYr_gZ
z;*L1yJp0#U`KA{k#TgOof%btzLq?*S?)jw(z;FgKx={x*%Ds_{EcRdv9FkGB-ku)s
zV%L_3fnosB4Qm1@!4Zft+7;Rf1WKdjjM8@v|Gb|=gN_L?7-b@m(@GE}3ma6nx82tD
z`JxdLDF+b4Q5p80`{~am0LUfE{2N0_6CvBd5CgN<upZqi`V3kqi8z<4vmk0v%#tIm
zC%kGI%D6rOipFDO&|ED{e#a&NTjj=w?^r?!-ofV$y$qM0pV@v(i5W3t@4roiAs#U&
ztFHupRIhah66o)|9uZA2@T_xjG?bwU(uSQ3=!-ZdzFyd<4g$HSX<B8jlglQ~k&YSv
zArqIS>4;31z_L(}f%E6qiwfXqO!bX(pW&R6gy}*;{&pQgu^$jP8C@Q0<VPz7jk2bI
zx;@dwX(*eV?e$snm~^3oQDx*%OPsPeMMlH5A5M#triBPba{Qa^)f`rkNt#C7A8kq$
zpce;l{_AV(i$))H3Ai>)^|!gMSSZokc^Jd#%X_Myu0Hy5JG)Zlpj*mRG#r|!Oe)Ho
z;Yv*<>!FfB7FA&;yO-dDg-AuPIRx0k0_#_YCIrMmfTS|p;Wwe#k5~K>I(Fn_F!u!i
z;i*2W>UOs6++$uv!4id&7GB&dDLgS6I*GIy6oe^GhJS>yRO%7=Z=6hZIw#Y!4thl0
z7p##t<4oc{oAhQ*#*2p_ka%WgMhP<rb-e$HEQ@oyyMm_=6nV?vXCs#oV?vd|^{&(7
zmH3)$jD<G~N;gshhoRnL?{b^PSCLg#61i8*L=uAwI^e7Bpt`UzisNEtiEQH&dtkyt
z2ii}ps=!8J&IEgkxUf11St0oaNI?pC381OAdwjre>#g^?*nCoevCkR&;-&I@<TYk1
zQ-wnmXUfW?`6Ea3UxAO=7ORAc>Vo%N<fBYymp624d^;#6E5S{4mi*qv{79fFa%jvi
z;)1){6mJ|L)`qvFi%8*~zEM<;;DPpGA1J`HSOMd247oRsT%N(^5MCtJ>yNn+Xl$I>
zaE6bT&6t35r{z{%1cf;HZL@TJDcn%uWjbRYYi{j&BY%|RTZt1tv~e12@W&}qYeX~P
zLNFkl-NP2#)#cSz^;vaBbU{OrzhLXOM)f;-+dgqou-;ARv@i@KmB!l}#}ReQ?X*9>
z&+5N%1Kw_G>0sS^U9hLqcdY}hNw_F&x$fI2L>lX14Bj)~j{rK@aA82VJi^Vwh8sKg
zZQo3EslhcAi8w;iz%1}po8v=JmNrJMa%1O}CT)^%PZF6#aWt=9LeZ?@Mh9*tZ;J{0
zN5~cw2>%ab@6aXM)<xNdZQDL!+qP}nwrv|HY}-Cz+qUf&FIyI}a0}o554+8|)*7RC
z_U9X%9_1_Y6MA6}AS3jxEb`e&CK$?Da~8c@X@$IHuCTCVev)kyX?qHR6gT?>QwSzr
z<l9)E-C2o4I+HwNh?<Bi*tft50mGa9l7Y7<AcvS+5ZxI7hmpk)D~8N#J`i=o>#i5N
z;Whyr>omDnjx+)ZM2omgL$cO!Bl88d+M1lK`+kFE+lw6}`Wzk8%zFmy-wU(LAC8HP
zh3-Q{iX?fX1JubM8nTJ+?GmVIlA!=*bGF<tWY&W~(>Fbph$Qm=+*JOn2F*zSKi8mt
zAN<`mSdn~RYIliUED76?*=;;!*GXi&){9&gdq_O;N!O1>QE(F)e%B>ma141O%8tJ}
zY-iWUBL4n_v_Uu6Fkty}9>RtAu{~M1Jf5^7#t}=}VvUXBr{WSaWy*8bh}{vSNl9*v
zd3Zc#i!DPW#G;hugU~)d`(w3@X5>kfYRqVX(j}4$ufvUbd9r$6yo38@e$p65cS(7E
zevA^v+#h#KQdREv_mrkRd>&8lPfds)cW1J2VD`d;@m>ce!HyKAh*7~r>kZ-zLg`I&
zRsIQ6WT_)Z##C)F3zdv|{Hn9RwmsvE*5t^0Ea-dG@yUfF??&m1E^U=j`6Bh~6$-~3
zx-<&sOalpp8uH1ZJw27u(uw$5p|ppFw#~%?`-JL>uIa|{sZu8USq_yU{&R4h0e$zs
z`|m_fn7DKkPSoZ4-k4=->LE%}J13P7u}zNiXPmWkP@n8NG<BbjN;=lXTUFXNK6@wf
zw$pTVK;L8zsy3-Es3e2tik}413vvc*z80Vg*$1x<%eoPxx(KSFn0PYngsgkvm~7-A
zXi8U4i@}XxYH1Yxm~sE7i6xA(XZYtnl`2vZSKXgAC7c340E~n7RNhZXRN1i5Lh7;P
zOr^<7CcIf<;i4C;Lpt2l$P2`^+3S;?9az?7VdqX>C{et<c-Kaz3|OWwkrj!p)v~kL
znzalE3^u};oNTqv9*1tJrM2!Gu~C>F+ai&9{dNbBCkxky-4bCWI;$$8NVbnz<YE7G
ziVKS-|C(IBy^gHoIWYtqMJZN&;`C|G&V1Jsf(fc@l&Bmju}=Tkcg=zvke@zT8lo^l
zDmg`mBb^#~Xq*I6%CbEm7&rpBX9uc06f4K^Z$GCD>>Ot}8rLxS5XQw(A6S;CT3qY;
zW-~DNo>oj_!SGUWfS~|kgnZcRqsFN*+$^MPcD;9E;p-xKkkKlQ@hg238ImfFKd@(t
zkBbYs1_EtsBUf9Oji!sLXgml2>y<0JFwL1;*Y;tD4Iq0)*rqbTO4B((rJR{in?;5?
zXE5Z(Gk|UDvGv9@UC@CcJym<ko8>>-zHunt)dksu?16Wp^(TP}vV-ypywGw}YNewf
z29+4I3QvNPiaV~nj*aQo8D*FF@=YccAI1_(;|k6pWtZ?WNkl+w6eN}J8zhyrIcbx^
z3@W_zavpQZaxN{)XNUBPeEIPuzy{aJ=RA%fV(xu9u2()6@Rbj)*W{=xHhQ1J0u?|C
zunNJ#pYGs1)hxq~VqvvH{Uw-Fpyhu}Ikv5rKph!m!eo9-9yEPYeF}@-2Ak-4?bsZO
zp<;r81}8{C7%S*&tQ~`-qG}+s%cF+cVEM5sN0?{;*Apr&6%WoD22*RMEN%)2RWpC2
z(J^mRYa2kJR7lZE_X;^b>E>SwCT3fs7Z&a<=^lyN2=fj2PA><^hT(Mhul{P&(gPTU
zEWl%6r02_wEU>&aIbRt<YtFVztr(hT+vd6G0wqjgc4G{MkOFy+q^=$q1c9v~;=y2h
z%(scj^K;wF>(2y8jBwFA?u=YL=sk;j^*+t%cS?Kth_cTP5hE>6*4e_OW6L68i8DL|
zC@cjl5FV6YFvfP!UMZ$k7K!Mn@yQ7tGc%{4r%hD(|33HmB$1X_gjj5Sj^;N`v;<A_
zz&#9667}C1_MyFM6}?r+K7>xS!_NuA6?SUm%-sEuVpVwP+LY%R#%lyx)LGKqayn)$
zdv%GePk@V><0~fms;;A-jT(rm3l_J*6F=w6s?4wb>!aT=fR_qd9zJ0a|GZ%p??8TU
zyNEakgjF2fVB}SwV-OG8c3pw|rrt$t#gxkbnGU_FGMTsq;BAATovF<_n%6^Mu$nuw
z$4pnybq|uT(6+Utc3~YI(MYMTZGgH^bmKcd&sK4<xUPxE+N%yABCwJ9+=5+J$y)Ht
z39Rm}J~-epk}!bVtAM2I?%Q2j>nH<LfdYB`85q2s3m9LnPnZZ&jmHbiDO!s;kCF-D
z=z?@<SJ5y6`<yqmYQ#x+KIO0;x*`Tw@-58mV5))z%vo{iuw<>j_s&5ZjOAR|`pU5o
z5_Z~R&r(0Wc5j#7zogO$gwEdM69zZ7KM9lO-zNg^(+QF6O6zv;3lswgp(GP&|Cc(N
z{-RbAubRcQ67IBrg7?8_{UTd(2Mq=`TSvDxHT*^)-H+6v)SET2{hDkZ1OJ^rN`$pL
z=s;L`PLQ@tc|2>f3Kj9ntz~k-%>IWlR0-&n^P6g2Esu8M&C9KN()fTTVnhomruFb+
zySLqxpB|PMESJyIH`!LN%t)ZV|L)qm9Ml_Wvz|bS4G%j#i%QQ}$E>;^#$6e8+pLwZ
zo%&B5s7L#C7A%=R9F@Lly=*OTX1#`|7BhdqsL^AZd7qFdFaU$J-}tLm>)_pJB${WR
z)!#jzktiEufNX5PDmO|Z1jdJX;?%5H82QtvKn=mBYhxJ#%;Zb`D8CO^|JJMI3zra!
z$SdXTTYbdKJ&g`<KPw5ML6EvTC+Ui<@`3_EE4AvtfR~YBHb~!eCN00S$nr2Hlh!hh
zu(KM@QLNG+$C#6hp|>P>j};J{wWscl^JV*Us1g&lCmwrJFmaogh~6J*Aso+=0(jm9
z`LH~TIv>tW-!1-Il@u=Y4t)nYojkEtm7GuenOaUd&R~C?9Fat1{Zy~ny~q$SbV7y%
z!(GC3^BfT~vED7W=)LarhmrivLNMyOpE4y!q}+Z7vi{LVUVww0e#~1Op{l9?{tML1
z&ibtgCz<H2^XDCb)-W%yY)gu!Z)LK~a8jHPu5N)jw!*B-nP6aK^^4RqPOTjSsyLxu
zlx1+gzL1A6S4r7V0t>&z9oZ5rSuncZo*+T($-s;n!5T(0;P>iYk2koD-gco6<|b;<
z=Fhu}Ucn@=z;2ZATl`u3qVxY+ci8@;u+6~n3uF5~-~Ry7Ro9B!XhZeQ(<^8lx0$>p
zQdHbP5f3m<0Btyv{Y(V`UMsO-X*fJ(%(Za)*m#K#OQalj8YAJ?qt=}{;!?{5+wceD
z)A@9Axcg)&w1`hsc0i?O4Ui9rMk)9$N?hy;QGo0jv!+Fc)^5{nclmHc724k&Z_zb!
zq}924yj?p!TiiW7tR&oy2`9>z`$rWcF<8Lq$d9HcG9N2~E!cI#Uaxm}pI56zMbV+d
ztj_y`=EIpBBPS_0=r{M$UkFOD=Ydcw1P@b{xzx44K9wuNaWVp4aEIVi-p-pJHLw9l
z1%bpX4G@6u27x4s2$M4WO-s9i;8B0)qU8=8eptyC!Sk682ML{~7jap93V&6`s}WCx
zJJDnBLD3)l|4|Ba<He|nvc=z{)-}H<g?7sa!|I=BbKF3W<oIyX{%i*#e%wD3&AQDh
zJV$r^3O~VJwmli>K0s+&anRbOBf;V#1Svat*0mArVO7)59ZN1wwM)iOvDuHRA2)79
zIu}*x1o}(Uk%P2eoK0>CI_J-$ieI_*nC(A5q}rL@oio{1<>BHS_0umOsZR4g&CA2x
z>yWbU*|D2c>7CJxiDzgcKlr}m729URVpQc%7}}o`*7U5g-<L$s8Ktb724;V$-uo4r
zmx^l|RcjNT(rEu|bWJ^d35@s)V^x+4V;1_8D8)sE{ax3^m2W`j%~u;X=w>MEcntFD
zoszA81DJQeLUfmFSba$zmMH8afcV?PUddmVOtHEO-0EfmXggHMZ0)uVdrnRF;272D
z0tU6VZDfcp76&PUf<4#8u~7L+R7XLiFd*jE;OIk#PAU$?lk9i&OvuVW?7)adQ)xyG
zne$i2$a^mO9*L_O5I9<qOB-$JoXOcNuw>7$Pn_-Q;1-dQkO)g^u5(`Ft=)_QYfuNi
zfJ3d}1hx>^968K0C+mD|GYbD?*W6qB?wG+4;}A1z+>zbfe;&%9KFBU4w<uW%F|{k4
zO^I&*%9~$RptbX&3MiCCi5$Wmil=7x-*QH&TGzT@#z@F!d^F0Cbrf`E&a~#qP~RC^
zYp1EYW7DF(WJJl3f?8*gB?R~d5Tmz7Q)k7-wI>xvB&<NXyCOLN8;biMmm4WYsFjo@
zTv3WLl)7MOnI{-<kEk7^B~&XMS{7e64c&t>m)5?J8zwIFcvwR6*u(aX%UmSxlHD3d
zZ{$G|$JPEmB^wT-g8CD&X`>UU>!Rk0Fj)Cm0`a-O5WK;<&DPF7u&kBijk!2~h>W(=
z0y>E)0wob(mmW}V`u_gmOIw-?c~`Wf@<S%+9-cd$1|%mNddz&pJ0GS<!`T<g=j1zD
z>9{OT^W|@chDt1s9!cBY#`kxSG-vE_D8n+BeHdpK5*&2N!Ntb0rKf*p_u4TnzN;vj
zxvpD2rOTmJd+CIA=C5Pex41X}e_m<L!xPuS>7XGTO^fqTAowUoChnZRC(=Tb$G|OF
zd_hb|#Ng!iYA^SMHmTODD56=Pmc}Ep?`y<T2Z^h4Sv#!$<GM5=_i9Qnd+KUqRUNUR
zw<7$AO0A{aLe4+|!~lRQ%W0Jp%%HKpJ+bw&qqC6v<O(Sl^}2{?SDR_;S8NAYf?Quh
zF9Y(7@}RNPmp;|%dJr8!Btbyrm1sQgTpWLf$-P3#7Iu>To2eP(M=+SbIkStDxEZ3j
z^<8Ru+a95<D89hTb4smC)6ZXukQm26`c(Uwc5IE0w7sUd)LqP2&xKNp5#5wwv3q~t
zHWAs`t|F*Cw7Rkmu*%xjlg;R7pSgW^Qu&!OohN(>A!@!$)$i)RY@1N)Wa0?ymr`)(
z4Tl{6QX`ubabNV=XXxa?u-eq_SJ-3^Y0F@F>-qC;v?&RONVhxGS6Xc*v1(D|l!Kb1
z#m1nY8j5o@7HvC8RQ9$=>Zhk&W8OAn>!2@$)q^GojG2x^(8zkIoHG_U;t6$s5dNJ8
z{*Lf32&OFN`cM{a#bGw7M!Q~Dt&ulI2@ViVz#v#r!0<Y9PHg&=qy8jX7y<IuSnaeG
zK71V6z3Q&MC@8pFM@Cn*g+JA{P;_va-Nh#)4J9L7Jd;zyW*?)p&ZURpyV1ec*s-h9
z`AXr!(Q#ca!DrPlibFCAIG4WOy~CTPAOyW!0_9fSr49oYffsCDi3kb#6d)#;T!Y!O
zMy?;&9$EZ@JiK=a4M#F&RB>%3@Zc3uaZ$r9_+|g>X2D4`MjWYJVZ15g--R4<p8!Xe
zey;dCHzIt?h>5h6gt~rEeY3?3-i##o4J{eoCA~vkoSeFy#H<$guQ|#|>3-9Jxopkt
z@6lBr)~4mIjg~(|Jw#rPga(7M9Y<Tu6FiHfLb@ShP>Gfu)8OJzk2nIDdW+^5AgI%y
z<ANjjN}cveCU^g}8n&m5O`J!l9T>Oz=)s@-B(>P14@>ibg(SFmm!uxirq(cTlS)7%
z?j~uEbegP+(<&eRV<D)8?~T63BW0Ez-jn92O<qb|2WqgEQMao&4sD#q4Y1M9%k{1E
zYS!<MZ67@v+HU|A+C!cHoZ$ae;$>m^pTVe8zX66{<Vw$l+GIk-eV=<0_ho<m(U4Ys
zPIF4PI<Gx*@{oHd4m{<=c<oLt;4Nir5rxB2E<FJjfA5EtxVJ<5uYxSjmy^fyTeDJR
zT=CD<{-a^R<$(D_8*9|U)4ozs2+z0o`^`x$s2sSlX5A1*_FlN|_x*nBLK6I_l!bc{
z3YRX{M4{d<-Kg37=cD|``+moWQ223wWk(NZl#|ov;X6mq{q8#D=k9*+BMv`}W-UvN
z2N#2VKi~+mccO1hafDi;`h@(aXJRI+in5~N*-TH_>xcNym%6pv+c%K1Z{~_j{S=n`
zo-7&(nyGz4IE_@ht%z@ZMdg%BQrOkUSG_oqwHuchj^=c-;#Ch3ebe*|dzwiPob^(Z
zQ@u0`a%2&rG)ehRV)Z$x&Ted~AT5Zjbg=!g;-vZp-gj-EJNKa$eeSBrK$8sj5+?fk
z)(E=P;5<8?Uv6FP-2`oF_8oiMPzDM&{ovIsN?<uN=nahQfkbbMbS2?J(Z8Z2Ov`j0
z*vEP2M4y}lnC#*Gvp|8-NDJ-y<j7_*<0H@IS4t%^+l=yFU9;)!>Zfa_UIyyxD*Q=n
z*n`<VMSW=zuTufyiQqNf)oT_<yCT_YBQs+pGeNH9B6v+DzWLNMKC=nJpT7ruretUV
z(_{>Hsg7rmN7>UFk)O>?()>@cksiJJ7DRnudJvM%ntw<*!|OqZXH0aNMLZ<XSf=4O
z<KWjsl5Xt24x_PYIpP?geoRhaf2!3_3<~ixE5~cK^>p9pmRh}ks{h#yTn@xNJ$b0t
z&Q$uieSNb>)UbJMF$rD5Yp7)pGd@?bsd*sIYwd!!S5w}vOTq5=^F3*SMn}%sd{K5z
zT?~+l<Op|Ggw};jq;STxcpTr8tG0+}SV!?W_3l1YKKNEW?IEaknun`APWpAWkMl^V
z;<s6@-E$hJIFS`SR@Fd{#_W^XJ5duEhVwqc%uyF^MysUdZaK~*1plFSAu2@9yL+%d
zq^kUK#Z@tBE7tr_!M$n>1+mY|`cq~#{wFZoNL)S`VKsI~KT@TIidDqPD;h|0U7iwN
zRZUf1(4N?1nU<~**17BXS|qP^?Zwcm`zcB0j&uh_O{Mwy!J7cEq-dPWtD*h?gw<%!
zxdJ!~Sm-f?zBVs}Q{}C`O0c+VyeH%@c*^3NK&#g`Rz@|vgfTz`EKxrO<F?(0w@x;j
zh?g$i%vLzwN60+VFt7U$w&S+2$eO2){P0U-$^8NChS>%V5@&!Ss>*g^ldFy|Wo1?!
zpsY44X#LH}Msy(W+#J;lzUK4tHNjdaErp;ecpMyllQfaA%-;@#r&w8o?GY$rR}X-^
zMzj*;DXI-?L3!htP9Q$52j`PRbZ1F;{ZR0B_dv@=nN<|Q+O=bogB7}hC;#B=lJ}ga
z-`Gp`xaojG!jtuSrB3nRCm_(6Y~oPyDb3g<poDgW=<Xy#N&4^+K$ySCk>Bz^1)+DJ
zrCF0<2m@@(RGv<A3LQbfOS`g|HnnUr^&5vyf9)3`7L~~*zrS!8d@{@`VJ+B-o3^sm
zv`(NbVBl{fFrk6T(}N)QfG8nThYz&AvuVskk>LP>VF83UZ$1?zz#Bx#`LIFTmC=g(
zqb2fL5+wP;Ph%Yn+(~E|oGt|rVdE>B8j0p(3!;CY5jIRj6YsWw8G-PIMW@Hv^CJP-
zhJ{COL+7z8%)72|wnuP|H&qR$Jh+#;4lXNvobrlp(GFovwetB@e#c+gEe`;&(G(H~
zKzC^Gm%CuZHngeybPUk;KpBwj*76%S&LMz+5<9zUiJ^t1|I~g<jFYX4v!mzZQ(1~s
z{>rSPKmr)83EIg<p<LdZVT_2NB$W&rTdZ>@!KXx*9U6Yz_d(VOCGT?0M<QCa5qHD=
zDyF}C%943WdI1n~QPAp8E7_{=UVLfr3c*~Z|KoHrtCJb$il7Z(8f$!N_r83Bj-Ghz
zugpQr=1Txzv^!#`>KZoO!q#M6%MmSrDwrTLw+Fa!G!yv29_YL_7!Ha9K?DsQfx_-d
z6|e6A@~oBog5hn*m{|H-?FNSu6cLUh(q7uLcrwTh(4|kw+LO;!?4Ury`U#;kZZ1`Z
z(qe875H}YOSdFkIrEc!JqX3^Q8PWTUctJ*O?G62B!hk0l*CeJQ<6}!%QOpY9G)+W8
zCB<~7Vai3;U$(se*<fvOQ%{S<_n1D!h#(Kh6SANPa+YD^iZuirOZ?AwS7v0!4SZW=
z67#mUNx^sP4HV7aGXaepM5NVCIDD*<X#&#ds!p)hlgt)wnmt03NOoMZF8IP|a0+os
zhx_GvzVyf4Xh)<j<P~@sdX?T?j?PXExx}3yx(Mn_qG6ea++&|*CMZe@+|urZ@GT8N
zrtNoT3#+rNHx8UGlroT}2)wG`82rEiVkbZA21n7Qq!!g7q%{`hp)?hZLJpT(h1M+#
z9cwqZ^WB9KiU(?HuMJ&8fW9g#_>!2LVAdIK8Thg<?yjm3{X#mCwZrT12k(_hUht+a
zOV3#GQ5Z9LNQmA3q%*Renv1gSe8~0}*$L%n5DSA266MDwZoEB*RUx@(`;@s{v!0kK
z)|4oT&o0W$%Gr(3Q9|{m^_4<#bcD4&oil10A!J{&rGj;R3Uoyq2$#<f?Of4g0N1Rs
z1#nWNP(1AsBnC^nq|-$=4E|gDmS*^u+0!_DXifa5Mxw_jnNFWi(cQ!&<`YPg+r`lA
zFzbs{(penycSrX;SBhA$qE@FqXx(|z7K05ZmUA60s}Qw)mkmx6k+tG2mt|wX@IW~!
z&KLNgLL|AFI3NTX5AGk+p&#4}-oIebK>C9n_>M)2g%slHr#lRCrgD*(h~SBGjdTMH
zJUh?UKlG4G!F0hN{+DqYR$891-Z*xTqOUVUXvy%-j0BliIsUCKVH>?OhWL`2pWDO$
zD_W5ozz8Z-bBW%$Do2d5nBa&@`E5QsJcDVNx$-$mDQ&uf_?M(4ne^3XaAl~~8$8`0
zs|cdoE7JVn(F^F1+<1l;SJq4cP?5>3)@0k$Y{>BfWCd9W@Xp8(*0>!Jx9(jz7i4`#
z_i9-Y5tT?-p`3~AK9(iejfXy{sk@fz+0iD-r&65Tc*!+l56CivFA1tfC-&e*MrW52
zQ`tbXQNSiI1ec6ts?<MO8b7hL-#5$w$7({pd%8d)ez2QnG?u-$@t$1eipriqIOO!l
zKhen@y*_h0wMe#knNfLFt&ia6Hc(fve~pVcj4R(-&~l49dgrh4rsKh}Ux&(iPvYfh
z9N!T|uIK9Km)XAtEjKaDlmgBr)md*0I7`$3_V>bGS6_cH(~D11`dz?RSsN-VD?M7O
z-I+J!Bq$epFB>r;wPWc}8eI`hgE2e&Qq0AY&wl{h!4Zu9bDjK;!Yd;q$NyZKxzdzy
z+89UtT_*|HDu|}Devy^%9+X^1%5U9~lUY~5;(|!liSdA^W8<}QGl42Xh%M%n3XTKI
z3kwSy&VCr#+GOl+A7iyqNqXA|)<{9oMT{ESB+_JoQo-%wXcvcr(G0?KXWHW2USGa<
z&s%zOsChKo-`^W&*--9ozTY2jXQjh@z8-@bd9dgZ(2$MnwAmC=&VhqNV_k&lN+C@N
zxtD(_cs4GYq+2y|Qc_N2>Mt46pd3W8wk-$<nE_Q}S44;qRv?8EDdZjzp)b!M5c$bb
z?i`7rf&&&<M<AW&kO0t;G^&JymDCm4bUL&~LCC?%Wp7Qi?Ik9YB&Wd=YbPFEPh#v9
za*=rcw0xKMj0f68Lct^jmG>b}cSJynN+TPtO~Vd^`rBa6r^l=?EJBJY-wCxSgPlU`
z$>k%>h5#5nmvBWz(Uhg@kL6K3i=V+zl7a`mNND6K3wS5R5?(|sIjIJcSkXA)<|q-2
zdq$UNx#v3qerg8Z%s%VOU^&bmXU)`u2B4Wt9BL`Uab3vA5<+(-rw(uFrr_8!&?`)U
ztldBgqz_<cBV_rvgDAktplK1fTNTbR-<kchPP%DmdZg0{Hck6hRPm^PV6JQ$pO!s=
zNz(AqdUaFD(KSuMSJQsAO0lbDdlQR3qpBV>hbaWkGFk4TnBdk5t7mIAlF1QAJPUae
zzAA35ny3Kt6x^zM_Vh}rt@*rCX=(3iTeH6YB&HUgYIvF&hqBabQ`yMkn`e}BtF-=C
z{`GrJ-y>s{m3%sRt)@P?eMsMknv$uJv8+~VoSKrF-OsN67rC39Z9)8=!0la3wIuEB
zdj5nv3w291nIeV)gfw($O&Bv~bS9fWILmmOxI5|u(o$*ji`yfcn&A*ABSRr)G6h;&
ztRu>uqTru-T_xB$0_>nne&E&G4(Lwj+Tu4y?Ps(4GNt?ZIv>r=r%Tbwa@tkTK1cOK
z%G1#K<c3%v43H~Qw0o)jJ~N0FD6Uj>>4pHzCB1S?jIUvN`yTLAG*GsE<z123%lHBV
z>*@uIaJ%K@9q0HsK}AXu47e)3;aVIakn5>Jv4X8U>YeKVpV$R_G{`ahiGqB1Y7X|E
z`&vGKAx{Wzkr%y|-pfvp(hKn(3Buz)Ua=xzB5&(6!-g2gj%sI5VXtTSX?R<K{arK?
za%%yf+IJL{V;m~KAyOxnHV`x!4g@9+(tq?eG&^PE3OQZ6+1@-B4HR<&9`H;c%2Y{9
z&xeEL*;RYQ!ha0om27d(md6Exw^&%mK9c-GWsTMS%*$j!7YBa(*}u`I;Y*=kve0@I
zy|wtLvLFVdcDcU25aBLC_ZiujPp~XWCM6|G;26#`$QT*fZw6|clNCu}9VuR29!kpU
zsUY2bn(8$dnKHfn_n0Rq`L-vx`Wb&{P0fLgIsiV8k(W#Hnk-CMLIO2c3viM=$hZfT
zT!5;j&EjR@6tY}7o|aYjK-;L|sq}_4sUKV)V+=Q4qo-`Z*g%UY{%5^brUT4WeNey6
zCM!ut!FEBW%OS98R^@4nQ7l3O`nj#a?k1JI#Kyz0?kFZHe2L|oZVEpP+v4lzv!2c`
z`|6@yt6J5e+VUx>nG-V?!)<iSy9c_j1=|9fC6on8Fj&;aj0<MIiOe3EXltwXp@R0|
zNaoaLZ%Re;E$iI2b87csoX@mataJRf*}SdfDctY*ic@RJViYNJPT&tEJ9P>dLoq`E
zyl;Sv;SN<Kp{!WRSf#8JAlc~w$l@wPK+6UB0>$A51^O{zd>FV9zPKCOYO-9$wI}km
zoSP#N(B}e)8NAeY@EPs)6YR@f?gj#!3mgPP9bZE_TyBopN|y6c0jMpQ{P0srKVi>y
zQ732Z4H%+SfoH?ERXfyMAh9b#!0aa^qvDZ`u3kuDV@(&{$JsUSs#pOB@p4?IznQiG
zf5(JL%Y<ZB#FY?yZ%`@P$%DpXPetR>677#CGR0;sf!9OL>eAlU&b*`2>xTb}iNOe6
z=l=5DoXs}|OOlPFC{hOv6^mlabc<nL>F}qTeGXmgnrYHcrOvsyhbAE_NT)%*8XAN4
z=kQKT^g7)*7Hsne#s#I^tM%+&S0J13AJ0_`yIbv#j`H`9p)UAbRD)<ctWfX!9KV%V
zE!4FT+%tk*?v16THSuWCg&2Ytd!=z^B3R9C7y&m2C7>|>OBV_!<WLuRZFO;Yb?Xx7
zV1~ELPpoHk%SN=&CKkZsIfl<0-9ACq_Zcg93L*6hg~5jdgqYx)L~@jY9r<VRaMAU2
z`92t&wNoL4n0o|RK8U$ffVtU01lZ(AxWa<CA@fCuLc(9|5AeH62@7D%_eCNC6q<e@
zHRW58tu(g_8yVhC%$M=li;L0zn82C31LG~sbr)*3O87L>J{droX^(TAxE%6CEgqb)
zFMWK(pNJ<1pbT*^gHc^Yf**?2iad-z1`Jik>fHg32AhMn0_6@AuL_$$Ko9|WGA_^@
zOEinltq7xckzNR&7}@kfwX1%q0?xfJc^9kTUKnOxrcIVbn%9?G#8#?6MxY|;K}_@%
zfteK-OGbaNfmnJF@5Lb#BuOye^2EcvB?VFBauC9f#`PaU0F~Q{jN+I;W4Yhy03_LH
zuJGmDJdNz`LBy1k@kT?e^@RJ+^!%Z5-vi1C51w>RJ)lMd?9|&d+MTGGS3l(Lb|f&U
zTrMzy$vu~W9(1=T1e#rmi`EV4yc$UuAmV-$5>9Es5-CN%F@4_?67K{E4F!HZO4oZF
zkgIuGt=1B84ge<Ox@57SBI|y1O@9A?MSSHb>>t529rNLk8(&;H8vCG_8V$|hJJ-_i
z9{jVO<8e$j@ux2Qp1@ysODY;KDliG{rQ+>sqmanTzKr<SFD=-9i1>wISC{J?0}gUO
zJ@|fwP5(UepN%r#o~k07g-pf$6E_t)CVMGT^BPsIZd&<U1-_Ga{}P%9Q0(7PEeuwQ
z_f_t8Ojq%Kjxp@soa}9J<Ed6*a{&F#GF!q(M{0qDH@|jNs_Mz_$9z17;_B-KRL1^g
z*#3u{lKubH8U0_I0t3hYj@BE-uyVp~z3cQH_PMA>B?Ne>$(p9{9mb4H9+&QA$AcFa
zB+B%ncN3$uB>&-oz6|W*8f&2TDq+F`)TVy^;=J9Z+U>t6y!mY&*X~nLs3(jwt80TP
zV3GWyTeCr;TyA-zMR-5zUhfW5c839I(5U5*^7rC%f9&5E_K$|GX_4_n5woU}lEe7A
zygv+mzaN15>3(>S1%pfN{CIOr9NOHSd_%!Y&FJhOKKDPjI@`s(9`^mgfp8p<G8&1J
zQlE@%Y6InPb-16FCx++j#DG@s`9bp0(Sex3mkQa{Rm9NniXnGT0uX+!%TSuQ+IU|V
z<%-uXD2RO~fIo>4bTI_zhn$rA1mLAb2`@R}8kuINsm<q0|6F%P6kuroB_E0Gb-(fr
z`GDN$(<%a>)k&fSz+Cr~e;~2}d>zb@ly->sBh{l<TiZH+fL3^md2_Tu{2T$j#S);)
ziu0zt=QVnad$>Q>vmI&jeE>BMaRdcpW*uqPzL(_OyUQ)|5!CLCb|s>Oi?Xt^kMRWN
zckdT;voplRpN&Dzn?&=Nx3uq4ON|2k+${+X!9(9AyQjzZ%s1Pr)UN#7Yt8u;&|z=!
zlvT^j11<VU_Bs8tOW^Hqw*9$7^YadiqH4ts+34MCSmXix_GBZ%b6!|M$&|Q5-(6Wv
z>kKXir4|`VR0oTKQm19bl61)q*(kYivEp}oX`HWU>EyPiE!T>^l4TZ2ZYcUUD%|a}
z*)11I+>mUt8$Gv_LZ#M;G}t#@-!=TCb92OiE9go0_HwT31{T*S@77(u+rxC<Gi$c{
zjKNMy4OGY_+bg~|9vmf9RtNgNi;n3~YtI;uW1k2!PAr^gSCpqjKW;d(U|mHZ)Guld
zdeHYBs|TTgx0xo8DB7-%1hiHV)0G;uvg{?Ke2x}Saf!7_!&0L}T<IAD-YL5v>(Z3C
zG*-ogcy0tu+4ntlZ5dRGYiz}Z?9a;T1?7jpq%TfhZHt?jgjND<42|tfY8`ssCkG`M
zCAo;W3FB9tSz&p8kLI^G)?y*hZjlqb8YZhp5>a$dOpGR?tZRT4zDjaKC`M?L*DA9z
zRjXo=|B+eJGkOsp=D{hii6(uFNcDx|f|mrKL3~=XBI8kAU5t!b!z#0o@eVapp|h4{
zQ+}t`s^!{oa|=#Vq=iarOTtgRwE$_5sAIAk^ijgD|7qbLa7Yh|f(g|lWI-eV7wL}3
zT4#kA>P<7-;G|y7$g&w=z(DH=VUSgUW?-^q&R`AD(J3PXV&kjzPqn#aosH%G5cH0l
z9?6@ok;RP{A#*S|xrn&DdReoYwN2<_Z+p==Nk@uH1h37A7W3L}E8Ul=AE+0tpho_L
zFwa)71^2JI4IB#rZn9~*unFK+Uz)#CmR5NqOi@lj8z6+V32gx~LNY)22*5B@hN3Bg
zX%PKn+>>oFA{_gM&P||VO8C6CRao;Jlp=(8La<#p>?(iodv03-2MhwESih7~#<pri
zb{HErx#96us>pUn2D~V{6B6()AD3Sd#NfJ2FsTkpcTG5bf|G5X3jTeFJ!kNbCL0?I
zBJq8(>WU{^J)52DSDRwM7gL>IX8okA(F%e4z)o)7AW_-8H&If%&~PnI_nC@SEaInT
zT1rJ6`O<Z>W{Yw)33^8H%r%pRcAdH+;+sV>9^+Gc&h-OdOgOOh>=-X{=)Vu<i`&)$
zRBc-kNhA4?iKh;bLgc=qiTKd+m3Ox-T=&FFZafHllSyA)&b@MERD-R@HbAl0`5gjb
zJtGSAy>;!ZSd!p8^{cMlt9B^@HKOfDzq8v%#b#kK7tw|7h0vPNEVLpUDoYkGDV&c?
zpb$HJSP5ESN~2WN#ZFt+V|kgDmVWuvRw-c2ZLCXvd&}>OOp2w1B!_pmjrmtWSwzBm
z-t{TPkMX6evka(p{$i_*YM(U>W_3tl<7}ld`o-$2CSlgUfOPOH)g2lYOSOZPRN##j
z+cyw*Sjh@ck~gD#mM<WEyA5xlzZqa)9<pu*HSDeHwZ(;CC$eLDxngBD@0g?2^?aSO
zc__hx25M~buUQY6EC%Oi1Evx97Mu~Eh^d?sv9|WsV*)RD?=V2ewd9NIZ%DiNVjwC}
zLD+|s94(^UH&;lf1f#IN`9)1cp!M5GNwpJ_`r8VP3#4-q65>SApt0O)oR(VN_`Dtf
z4bfEtKr4w&X_j6t^L+ry#9)aXTKIaA>JTzP%RjsOi6hqJS<n^C+LKU-@3H%b)E2Tk
z=Pm07Bt_z+4?D`lV9Ug#J>>}CO{yz$X5xs0Y`Hg)mYu_l(o;D`U9hF%x7ur)Z32gX
zdP;k?9u(mSX-jIQJf`>Azx?=UAt4`hBPAT$E7ex9@ZNW2iOM40lzLi?=*5^lBou5a
zZEw77{M@;hUalwJuenlyjp#d=-&U8^TBQ5{kK`111|#}xd-;7n+VfE2=Dshr%Kbhr
z$Lw5^S~Dh&ndDpwOL?qL-p8u0xQs6)GCc1mAKy-Tx#&<|Ya~F1eL~+qE_Fv`1ygxZ
z>~>c0Md|pSPVk-a_9VXB+$es4n%DSf6HuzReTSri==~OrD4W7Rq6O)E%tnC^xIe()
zl^0MYkMmS-K9&D-1^ADf6b3d%hW|BMv!uTDiyuMooB1^=jj2y2B9+Iv;?DY@;)bF~
zMRyko^sh$KD5olF*zUuA?6@5!G>~xR$#h;363XD-__}ow-^3<x7#u$J!zTrq_1dQ<
zM4LKbB#nVa7!rv{XT%#b5|;{~%!S8}Fq-zqjoQh{j`&5hpY&JmY=1slCWq_aA79O#
zTHjwZ>IpBJ5R;fNf>EYM?Pd^i>dVp>Y9L8Ugc|XXv-xoPUe@H1gX6=qhnoZ=%ZEA0
zf}AkJDKYBcr9V-m^5-E<3sE&hFJ@}ve15AZSI2S(KG6sHF%HOdrx>{b5TcWSP7Ej1
zXG@nIVkT(ICpa?ih^V|wf`@^TZ151adS%z|FpNnS^@0Z%c;hL(Ua21)?XF^^L;nD>
zPSmzKc12CyMrm4ijmBli8jS>=5OY=C^ELPP1o&-!?%nIXsD?y?;@Bl8GGrTpXtLvG
z`UpB5ej|mITRwJGJd{&yuN-amnWJ?mfQ>|b0PDXXZJi#O$h{voaW^W4!c2w%n(*dF
zbDM0JG7H8#PGZXJ^B(l#U*Q7>H`^||cvQNHLdKX}0Ahnn7AfC4!&hkC`O5Oc)llS%
zEBVogW0}z#+i&#4vTFqX8xxEhDToeKq}i(K=&QfWqg_5~8mmO|qgkm0GwaCS0|7QC
zC(B-q4?*ArZXfx1-YnB;Gk{=w)Zb{&t#}&W8^?uHSLZnOZy&L!9@l8T{8J~G2ZR~H
zCL#SdSqUxlAHKK7N-<R=EH{7xfSuqFDF&q*P|f{cdM3n0;#@G#C4OcWyy@`+1*aHg
znWaWjYJeKlO=qbqYc{2rfeb;XNc|cq9icML$T%w%*0)uo8DVg*?0FDf)~N2WIT&ts
zx{5RKsGJYr<Nl`1HB9Il<=?ClD|Dq;b1vO&7S&x)g;jGu=_9zU4bw{VMtN;1w92#-
zyQoz0lJnjv0Yy%Jmqt^a33yAi1qd`+3)O~_(e+|vOi)*uYR3>==mjAy*FtSnNlLEQ
z>Z_B^@yJNXg1&0yA2Iv#i=z&O?VQ&23}i-m5AYI%<cyUuuh1pmxYDdux%ua3zaBcZ
z(_VzdEz)BVhk~n<QQFYG5b3rp%Umk}VWS797avS0AUiw|ZwiY<mjfMRAu;1^#JwaJ
z{OtEx!|5Oci_ySQv0B^-D6yMfTche<Y%x@i2Bi?!LwF>~^Y!j6GuBN88s_BmRsid{
zkk@0{DclVPbS=+n>X@aELC-VqeuP&H40xr&3f!X;?LPU$4tumg;^m>8gzy6Ur`iJ>
zdo%rS9fx_|vxsjwE2%ZTMC0Nce#}*S_wn9;E1CH9dd|*=ua>&vj}9MyN}BBT76&mx
zBjNcCLkx%KUHs$kPkARHmnHBHQ&}oTtJUG_?dk@8GNp7J^Rpnt?M<|JF)<NVz<YoF
zu&teHg9k$=_I#L87uB`hXkfg9`G#3)s71_2a<2r2F6~`w2u!V?o1trCA3W$jDjIjM
zDjohD+~1Aa5oz5E)UDQBC~UTsW-XSdR<ngNYcAkta9MxyWC1>TCPZ}H;<-A(^D%&B
z^K9PqHe-RZRgCz0`y6f^>U;qCWVpWpZf?f5S+SSKQC)4DrL)DVr_oxglJB<glx|O(
z&*@Tr%Yf2ZvWY@?xD36tJ84ccz|3x;?BTApH8mJbmT>xTiG>J-icjQqo<dV2F9=GM
z$V;q*h|ON%OpF!{mLXAxNSP`v-x47VR#{OvPSX~HsSbgfQaSdad-ZNIsJ>AeAZ|`%
zym9o<3<x(3%W%}XBy{!%U?qIh<DVjytyd_qTh_s9rXzeTNdcu*<>13N@%+q(QKCQw
zI9X`&&tr4~CUw?1$GD=SNknSCD#%jPo?1NezaJY{I3O8^c=Z6F1`x=9Mw#Ms<rq2|
z8Klvjx>CnpG)My=NKuQ*Xx^=g*+(PK74~RdLz$HWiTL7OHa?){oG?BAfaa?<^!X!e
z-s&}jM5+#Hp;?x?O9B)8Wk3fvm{ETjOF4L}@~6ck=*!O@oCPTm<Lyr6&_i}?5_%Bh
zeS%99ExIA3qxLT0e6?sCEgVG)8ek92Kp-)qHwiUI#6cB9&p|M$OXTEzY!b#BBn-|<
zHeM_e8)iTp<1*tF{bN;BxX26C46azIbM^TO)Y_=s_c}>-*-8hwaTl95Em6fi0}`c7
zT`ph0!YvD(;lt!R*1IN|10Fn7(fH}m>QXx|m&t8a+EpsOaIP)&I{7?jI(D6(5u1N%
zN+4HSP^%8~U;Zw-%ss%uA3C4?5CPBdVxM#<|1}CZ{_8Z%&P4yej>03&4aeV6=<}j>
zn_wve5``VnbQYb|Cc#<0ur8^G6BkyGkx@-Tg<?aVB+=Uq4U`NK+F~L#=Dd!c0D|r{
zXgC9CgB1;o#l7Ldz3Fk=xEnyG$$$#81F8+GNFf*k6I!^-juH(g8`{hLidkkG0Z9;%
zTvmwYdGCGGy7#?DV3-)1og_el3Y&M+>y6dz$<s9W;Rgdiu<7;YX@?MH|GK{v;8X8*
zySUBG;qBs#!eU65TXXo!3_8~?5gwgPzt)~`u-~d2{Fq+HzwX)rWR?Pkpr?gGfWSo_
zOWd{YFdAu3q{su2Ml5%cH|_7t*}1te_EG_Gj@R!&wMzvBQIN1^`rdQTO<9Pds=haL
zwMG`E;57C_sw?S2Z1QL7pD?w@p_Cj93wn37*G#eC<>KI&mb?5%?6sELersR^Vw0Vo
z21p^YpAL>lKOnFQ;04=>iCxmFbT>E8djLW|*ZgJnX2RO&qLE{EO{S#`Z+UX{V8}aG
zxfLtV;ZGtg@S)sw(Upr<Xl-cs5TazH96M9#CgPE@ydY+rMBl1<j^l4X)=>-x6^=6I
zYMEm=?vQQqnXoB)FABHitFMl^?TZkJT+m0m!cDVap=xq8P}zFTnwSY7y1Twio_4gs
z13{F402rZHXp$=X?yAI@kypxBRhwN?`fQ_Wqh|H`YZB;q8+A?7?{tdDMp4u7FxvNd
z6LUKwXu5ckYhj_jzgs*e4o@afbX7fh=nk@C9IYnmG_i|}Wg0NNN<no+-Xx_sdV9QE
zY=%ef+j(aEGj74@Aw*hzF#_kyr%v16{W1L{RH@&Z0+bEO3Y&xa!>q1ya4o6lRnz_D
z9m9uRJ%5dBy5ic)AE}~`D34TWLszTBJZ-9xp}O`j^yix>N5vz>jLi*+9~H}|L+lS=
z<Pp#~UBkp>71g*cR4hCWoY5XZNp@~&O4x*Jwd};2``5bwPxp~iZH;5phK7dccLzP%
z+aj;n6g`;hhoT+gBOstGi#d73JOX`8nc1yzBv}}XE;QyeO>W#h(;z<%>hquea=`J3
z{mqpH%~sFzW?6vWocHle4JHsL!}zxRxM_*-l+hI5ttN|Xi{vVIi)5RxFBj$XcS9?B
zlsBvHRLvRs^{@$*avzd50HPLnmEoXTFkC{OG_F&I4qB3wGw?t?t^~R4siA9X3TjK)
zIWdw4*1m@S!SIn%*n?pJMUK%_)hi;L+6k-b7W>a!58oe2v9jY~>!v!m)@&7siS&o{
zwTil5`cRD4^E?CRl;Y^Fz0JLaB^5>-R;Mc0^>q4(%MThXw)bE@2Ux&k8*#$wEAb}9
zUyHkmkP#<AnC5g6e@Zlv%t%%dgCmR@j#O^BcstDOp$0bo<wPH2P>%O`s+?f|=&b9;
z8}BGBq)d8lYJ#A`@#gf09te;*B>LFY%^c$0D+mCQlv|?L#u@rthNXLew}KUt!iJrq
z-YA$?#+hAPzaqiSIsKR=o(g_M#L?Wbr^IDN47VrJbR?Rllu0kAD4hhOUmH|^_^gbB
zA{>uL|NGJ=MNC#>mxy}YnNT3q9ZwJjYEZ%ZC^<96p41_7@C8X*g-GcLeBU4dELlI|
zysw-1(|jTUtk4aHOt^HmTW<|g+c=CSQ+a>X;y~JgdeaE`WbKxI<}eEmNp`z?d%TY%
zv1r}jzrhqoB&C%+Xh|e<GJB+Wh*wpaMA0;$??_QSQb{o+%dKu<L#*KKvMK+1j_Vw8
zUd;cdQa2KsZ&BM149$OOA?wffk_!tnE}&}4B2Lgp)M#{B^_F-7Xn*)*=M<wyy{AKe
zXRN`|FuE;dd(A|y{T(p16I)m-V`BNB7jsp}`@G6Bw3V+p+gF7RPV!DIt_k>nQnl&5
z(`oXKLp-5RUw;8ZOb0B&rsq9&kKX&%9(3rL)ZfT76=pn}kwjnC#a_Q-5EFOFZKGFr
zS7w>kU{s90tFjX69=F;YR<mQWLtC`P+;hZCt6drtuERXd|06D}));p|Por##M#_K^
zgM=vLx|-eJOD6t~Knx)^YuBnLZCGoQJ$#s0YzLV2)KaJMWYDN42*ts+7_1)jy-dkU
zG5yPJwbVfgqbYcG$-mH2ej2T2p>BAMMRi7=9Bm!!Vys?2G0s|Q)voLzj(^o(1!X&V
zc<t5Q->HiFq;o^lvojmB>u9VSGD2E8+rbKGsiJ1+lCl}ko1^T9nV5yKEu6XQc;wK?
z<28D28OS@RzB@6Uu#4d**NftJlaxus2o@`_yQH<c25x^=P;X()eOo}ghNahIvA1JP
zd#QYw^wq8(g8#}N@bv_Ks@1{jew!yahuGG@+U{lMM?sGn@Sw9c)H?i%{m7AZR&6^#
z$?bibf0F4<->j8peKjOn5d&$*t$8XQp6+PkoBxL`*GW-N^=JR5HK5aQ<_EkQx%}ln
z=k9+TK^Ym@|7Uu2in>(NK?`EfP3<g=@)C>;fOmLGQulKSYs$T<2YGvyQ@(KY5H&s&
zV65rq#X^iNf4-_~VrB*6tUC+VmY&~~wbQ0?R&UqG-r1HLzObs@c@wQkyNBm?nWNI7
zN#@b;aru_FM6#qxhY$5Qwa3py=}aE}Pd%z2$<EK!LEMgACAY`>>)FW_U*7loR+#qi
z#-T_iDrxq<>sWE0#0?{_JMla<^6Vt5VPRxb)o=+yr}E*`v4uaVT<NqmzDWY<k7o36
zsyLjCpZfcVCEnaa7iy<*=8yD2%iDU4m?hgswfN*Boodo<@fo_zMM;M6@s1SJLnL*(
zv(L<l^kBw{Ngd)1(0SqV{2y4?@knRzUfI2T80yN%B1|nym=4VHqu7l|s=3A9nlfV`
zGSy~D-(2drbtNO8C=mhFl?U`l_)Z=}&WrZ?EoRQE#-?qZ<{7ILG{wZ{V`ds3sT&Gm
zL>}X&vdv<~i52M4v2Bkp^Eo&&wPDXWyrl_|dD%iM3Hlx$7h;LC9)eG6Y_yy)#nE<5
z@w$~TQwDCk#X8Ly=KW1DUu>+4P6amZ$F68_GZ?l&Q71dB3}f=$P%g#9MF4nvGpokO
z)UA^Yo&(TmFPQJameF#OTexU4$C$c83$Pm<GEJNo#}zp8+r`)rF`mFCxf>n}$^}j3
zq96EvVafYBNm2yvRAEsqq>O8}+#_bNu{~x?p2W*F4`!&+JCV2?Jb2&df54CIBiyz+
z=`Phy?cg}oL>G6*YHFQJJ^$tzl@g-<W$Nh$ajLeGIKo5p!p``op}X9#YVgDl0mz%X
zaSIPQvfnum_V$;pwuhoTY6~x;4Bo#>y*SlE@W7dKlXtzc#TF4LyFSq{0%-GYz78>O
zzk^Zrx;9~G5{W@b<(BpoFYAKi&xgbNr`Sl|K>}b8pBy9hsRm9KJnY1{l5@AbuTUoy
z9$)n}4H|A)W%9*3b~%aCUtfPB){?Vh?X9e~pT&|xZ#xraH8bmzzuP<&yo?kDAhYQb
z&}aJ8@~|&u;hM`+E%s$;k5i-S-=x4LJo{C31=r0SO<9^9sl(r%hiZQKa(WMwjriAh
z>~;2e)vK)xFVO;Alvh+>%L95vgH<uOYdt&<yLf?v*hCA!zHnlD1wv2+dg&7Z>eRe7
zo_(Ad8tNlP_%?guW##5^2(Dm@sp@A|O=`SF90Mb%YtHYu|5d`(z9r=3bU$Q|iL<|O
z#d=*eZNYs*yJGYp|K{UXPXzDHL-ecJJy7c4aZq0K;8-=$)%egVRX=e{xtZ@e{QTz4
z-#kx7V&q0dG3S_?*gRp0uLGOq)b7Ui++zAl{FTk|@WEkz<I16=0Qh`ikL_Y@=<M6>
zi)wmbDU#v?VD}u6OQw`lu%Dc-n6h)JEC036Eo4k1y&H9A54HJM5~Bm~&+HNI@@2|h
z8%Mtb^u%cgm%HykwTL4Y7GrO5^WiaNfL}@z&lfsDQS4n3)H`y)jte&xu;%(xqXRqf
zVap>7{NA#GlUcNSBSk5~F6|9uax+m!J=n>jZH2DGugy%0(0>L+4GH9CurtNC6^Xd%
zO1Hz+p#TgxjnW_gar4?;fc@;O$)t_?s}|ae9!qgM6~rF*@7!(LuLY0SR`3Hoe@Soi
zzQF4V`k&lD*4;BPY%V!}ZF}bdw}>VnPnt^(P#r^d!-Pv;fiBKNg+GNk`i+(C&#(yx
zMh`nY8Ea-*gpON8xysO@eQi0{YX<UIA06&q>2?|2;%xT_IV1C0u|#r@>qrA5L+F@J
z#<eBbF7aExBdY_=u13lW)M{7ZMaMs(TSv2lp*n%{YBHkUc>ctR=&$`!S4h)QEsJ@0
zP7xe&@3=qwyPK^{(a+cJr=dz)oq_r@z+jWd`WP+Nd*4jlGlSif@NO7s)RKYnqivfq
zCO{sm7$w|6QPf8d@lZ*cJo-KCc^$y8+!rIm?o*Fy3|@DO$RL&K{(welsd|W|Gi(cQ
zraPgzlPa2OFv#ie%BjflN4pM6wS<4~by7deX0T#~gDRytN<x%20atH_0^Rt19CX2@
zlQtLC!5t8409FY6{UG6EQpt#WRN!_H7;6#IfbTYe2U+{i>(^4mrSg<B;09BiroBC~
z-*!LegY@tM*q;N-a{(J+#{QkR_kt>#NhKC|M9C>g^9-_bAdwl9z%qKn&rm69EtVV3
zlo%<aP`G7}*1e=h;P%ZIMfKJL75Cgr>c57fHgf;)no5mWyL9(*rHuSjw7Q9?iC3{W
z{^J_fV};hE$Y-x_m;4hT@uP<4D0)YnOm$-M{CVN_4<eKA9!ufKqNc6n_0d`<Aax~|
z<FJO2K^g{#`#Moc$XsV(?|L>n3i~xidCcekqwF1mG~uFc+r*c)ZJU+0ZQH1{ZQIUD
zRob>~+qUhzbs}!>{t@rI?r!hKjJ4Jrqb(#C&blPVsV901p*eDkL0WDh@5VpUpIGH}
zGQ2dB8-MNv)mIB^iaC3YRYAlEk!w&Puoo<9$#xfcp)Et``l}gDsI9X=YDG6jzwvwg
ziQI-(fZIEWePExL#-LX$pY16ZXm(!JIjB?4jqmD;6PR_N8>SH`=D0WSPA2-W$4s<%
z!T9{GhCTVIrsiYybA>8Qp+BRJbAT<-=7+lxg`D|{wEr@f$Vgd8)$)2F$EMwPA21zk
zuC-Sc{lLwffPqd>GOUQ2&jUM_LNKZjG6)FZ=+Wu%tMmK`yLfAn16&G6!f_y>x_t5i
zN!j>gf1p#-Q8TPiLLduq3c@<8_WYsZ=6oWjzdk((6&QXU$H}Al8dGqASVyUR02{0V
z4cdiFZGZ8Qh0M5KeOHKDvmEx0b)y%mO8XduKI&uks+SLhAys)V1zUhml2%I<u7^It
zJI**jvbD27Lvz2~=m3e!z=k|+dw4aT!kZ3Ln~9fU7OAFWGb@Oyku3F(uPBW5Va?4q
zdqs}G+uC)*Jp3LFq?D9m%dyI_e3oXX>&&TG!{ui_aIvfPhN#Vw&W(kHuS1U=*Zqvg
z;j17b)hTtzXM`=XJ48Jl34RSd0l&mAwknOfMZ$Uo%|-93cY+v&WmY9FnG|S=@TmX!
zIy>9)15(me(fp4D#s04e%SdQv^c$Lo2b$sk?(u&y{oiXy!vF7ntZAaG-3B9U=*=4%
z*LZKGM6=1if#aJ+SOJ!nTbc_5otja8v&o>`?DvOcaUl6>AE9VhA^v>fsXyiVWVd?5
zTV_cqp=hc}>H1$*;Xw0X$AEI|uVEg^0{*4Ax6HYHsbMTl$i3e(1sxyv=1rcl`yvY2
z#iWs<Vs~HX4p?^jG>|pEqR~t`eBG&v(+*FyK7mp9mQd4M56?$r-J|%a`dko<e}O%^
z=~vMta3%IhDWZr$bLwy6o`u~URTMnk8V3+_O)IQ8e_Q1hWesEHNBnd4genO#Psis`
zX>!OewI#wwp=<p5%X4n0IB6NmUVc>w1z~wkDffVCclgit?Z)Z*H;_E}Xqvfo(Tqoo
zcBGl_3N@J2g8LU6TQVjNr;tj!7;$rNeg<%X@n3IH_ED^TFcM-aAoVsq`Ptvu&;eKY
z))j}EL(8o<hdYRyQ)a~2;0ii$mhH=8K9A$D@0IKF@!8-u@cGqk0N%yJR+Be^970ah
z@Dnh}bfCfQcFWe}<b!4Wz`r}H6*yKGz#v}!7rUf=uP54IzUEd-fgfU)%<o$hSk>6P
zs)+hQF3@=8%dT$c$ylo4AGz#E`m!t&5__){DqBxg?WC_h^Wc|@jc)zdFODzt2t><T
zFtIy9wdmiaVzu=ywN!ssuO#-|f;VaZL8Hsp_}3-xAOG`9{ReTDnUUpxZC&*FpHR*Q
z-#t@vOLtz*Hjt*bvS?PExTPX%^PV~uK)2A$>?s;|{q+3~2C0zRvE@}K5f2bVe7&7>
zbXy<WN|V2zdMPQ9=l7*wk~<QrSiH`O<xaY?R*Y<%bdwL;B8aHl-<v7XUm@$@f0okv
z<9$`)ktxo;vQxN-=KTRr@%_4%sY^93V_qs3S?}-SSQ2>5ilPaesXhDov9aW36IqyZ
zt4$u;-cU2G`}MLf>3Pe5Kz%36{|YJ@`z66rV(NW|fh<>zm)Xu;C#Ck~SzYlE^PEqR
zYPQumIJF1aXMt8FCrZCGtwkeaJ9^(2;}la7h-3-QJ;~cteDuc3SIT^rW=K58o-wZF
zemBapZ2SQVOO@CjtLlMa9gKcXr7$z{JN3Cd|M;qZt&fpUBEx3Wt})(HGOV;Ch=AwQ
z*~!U6L5gDE&`u@u(@{i}D`O@H+q>9^H%U(C^uc{Y-;12{#m4Xc&kaV7mvO6H=J!(E
zqPK(bRr~me@NQ1EHk~|sYK77v8_C2Pql=f%tlJ#dd3VmoFu#%)-wKZgXO65GL26`^
zs8y-Dn*&kqyu8y^dq&s?iJfF_55|h1gpLI@>^|1wux|#;t?%RKrR9U=Fc`DL=h4(*
zR?=AE9YZ%#p{j(Bsx_f{3R#z%vSSU4bX`JGpVp~u5_Dm*5TRRG4?2hGD9+`){fe6x
zvzxy>yqfT+|9yM}Ylu;-noSLN*x#xZP20+87DbEUfbttXZ*?!nF1niaJUp2JhE{U6
zTnhf~;DxqO5eRl}cw!tPU~Hy6n}pya+(O3v=08DU;dom>4Lcs`^*bhNh16#J5OTUx
zUy!lCumDSIG1Y-qRX;eJ=FG9UxUg>fgdNNhar$Q1RR>b+vA^~qp+8K?%ealg0U0Ec
z8wB3MQ`opO6?X+5P{azG&vMy?TBV7~xjdL`OQQDpI;p_&*=NYn^lP4&*E$zat0_|K
zqup~g&}6KXJ!>ZZ7M;q7%@gB77P_1CUIe`s0Pv-ra}V29u5{bm{=PCQj;Fsy!5w38
zK-kKW1rnUsxD+caBYh6ul!-+-ancJq7NzGKCz*zN<TX#;#bu}<6w~{bQw#Re$pW*&
z6~?r-f=c9&@y6kLGOpo&mS+zB+=3>#(9i26j4ATs<5f?d^9;9AsS2H~^(;cf?J16<
zHIQfCUKZP*&BDfpFyJwAL{kvei`uXo!ZY(JR3_lis8TzRLv?Xx$Amo`pek;Y+bbqd
z*}DMWX}(Lvl<T*96tMNxYEBfygfO61(7xGN9&_c9#Mm<Q+Fra3tCfjbVzH!+4Y(LT
zu}`AF?gFQ8EaRa-(puImT3`?(hEn<;gb445HyaA$7tF#5hZn>c1-e-PLhVub7D$*9
zhdV&aBl9e4TrZR&qnW}_Lw&WOM{J<l>GZRrw$&JcQgIN<5rnkGLkonAcI2CBEX)w=
zxEKIBNLbMc4;wu_!<c!Sujj)zaf@I3S<5b)dsQa)N!;`{vNm)@Maet!iUl=N$(H32
zb|5^Ly(#?$ZKmKA*Y+AzNjo2@4|6{hi`fxCFe<i~m$nFxttyd?U$7W{#iGihq8C4b
zxO&;k#545>ns!KM1gxPSzi`DMtMia{p;rX_L&gLCAR^v^Q5XSGIpNfq8Z7sUZs6UR
zg=na($Mc)+Evm|p`3bSO5`gTka{sLw4zWyH)w}~p7nM??r^SlLPW%HH!@8ndVAHo1
zkoexdRN5473@vx9y$G^#fq8ngBFTg^G^pR(*Mz^k;pyu+$$mtxWJ6|*GX~8DZ@pnU
zW@l{G!VdizO!-*1uS5?y<!-LsRne&8ap~E_5%PvOL##M{Xg%ZKt!w{oi%0|=@DR`u
zLKQLxOS=2#(a<^4TfWaZ+?=M;H=FOk`|7-yonh6V`)lMUOKKdSNNsY9+WNoSf4wPz
zDjWaU>d46oI6mJMJIfHXB1q_iu{uTr!xt>5a^@`$s$<|mxMMNSr#e_crz{=9DV56r
z=kNYLH15w_(x_yLg^?a5N1er2p#8@+?w$pV4<DG)QNxxUm#<YL2nZv9fX20CJ|RvA
z;=2(AB<^T$d(n7Ycp792=qA(o*V)&-7dK_>mi0LRbfZ*B;4GocLL5=z_ro&4Gd+iu
zHg6?3XuUNhGhQicHklp5DpaISJ`DHCCfuFT;udMMKl1I~QF<raLpnw}lE*|$AQlkI
z+=C&DfK+L~Rb6qN9<xzEN~w6SNl&EP9C7H;sy?m<l(F<7#lbD)unZc~`kWj%pYN`Y
zjKGQJtODytxM5{1VKzhexJ}0RRM|YB0a(~2#eQPs{&wcoe^+67pu0@~%|@Tb+_QeK
zfeo7VoCvOI5%|cno%$&D0FcwzE)96Po?j*HaIAw8gUQWgMa<oDq=EOlH0x?&Y5Cqw
z=qJ|JLALG&U>>c_d0ueR=PO%THvL`Qe!4`8l+p=(ECxqV`gq+!cR8}XWbf_Z`U*-U
zFEgCZ%nF}zFr5mB0}e`q_QuLm4yg5U`5||4`T_*@l=VJ*ZE)SN#xWheEcC5VZn_z(
z#n5P+h&bS`_OPgo8c6oMs=?EE1{(+OEkc9vO-EL47vAqVZV=Va$Un?|R~inb<+nc9
z##NmzRyjh*<rUieE8Tl0cXxha<cGN;x#%}zX#PFuV5HZ4biuSzMgnSCS(^O(KJNqy
zTiz%C6(s(}(Elf!GP5xMucv!Wnq!WeY}nlwRPK@?|E{7npa@ktSF7EJ%cw?BUZ$P{
z&WlOSJQK1C@un7L!a_(#iz{2lIgtRqpC3~5mu!F6_!;yn8FIQ3i5mvdJ4ct<GbowK
zV1$asmp3kb=-WC!Tfd*yXvAzlhNB4sr3mj77=E5Uuin#k7}Ry<bR-<xl)$ETzMpNs
zY`;Dpr0E`>F4SxEhnMW^aK~Y=Gu)rY>j6*YBVGHMnA!knySNTrpH{d|84j~x$K8Ca
zhe>5Y*y$B79nY1m9fGWLM>4>S^oxOqTPeSKNwF<~ul0VPtG<5yrvm#`nf1j1k+mPE
zYln@-wY|;{`@Ut|p5`x`E`|;2Je!++hXae8^;$0{o~OY+4i4Pc<W!^~o81R=+5qY_
zwjO?6<|KkX*ET0^a)uZ2(JO?DW~iv*tLCMAn>-6an{S@&kXT}a;Em)GWnCN40^^n8
z1&NCZ-L5tm^ZEJOlQH$Fue)6+Kr;n)>N@hY%5Ep@ZL2nBsjK$e%Z75>RbJ*XGz253
z@ur{!LN#bBeqR5P9;u(`;gKQEa}oRuRMFJk>~dK3b7Y@a(-m8vbF98pB^uO#s30xr
zEo@1qF^`|qCvtZ4nQjP;`D8s@MPKIi5xDIxEVcl#^%kqH&T}eA`yR(Sfacatd1`oL
zfaQ1Qd9V{%1T(q)^b!KFo6xb$fnqwP3F8?=xy6F+4*1OZH7M)=TL-k2N1GAn7m+{3
zDM24Lr>W>n8F7eNMsjxVdcX!)fznYY1$C5#p>?K!W=u|XG5GL@g~3lIbSZv587>_G
z;dquuUsBry@f-vA#{kRUuP<s_(a61{_f{Xs*psWUKR#_S{y983BU!hHkC95)Gl6Z7
zv;ywcVwi4?T1KgsN~5DTtdU4bwRJqyXdgUAgN_B#0~GBti~_282|cz0&1sXoFZ4iC
z3|#c$OS`+Pf8H%vpt4kuh(u}@=P2R-DDXvBAsBb6_}RMZ1|4F|@I3hbRA0vH@vF<x
z21JWeBd67l3i}H=ZjD6kIq2uC|C3*;+X3?g!EL?Wb#Hi#jb#Mx^_L3uTO{+Lx9Wd*
zsYP?Xp(Ys5khCyMBCb{*7;<mI&c&(TY^W7ZXKj{A<wn)DKhYz9x)*-8ZL8f$ahh;)
zQ|nnsvaTrJ)vnWrUUh#qY`c9RZgMBr>H4_w*beL|40`JxHA=u(iA+;bM2=B~$lrmZ
zV3jvAEU;&gs@==P2c9W3@heb!3gj?J1shHZ{tn!oP=h+1kz+eI_Q)SC>30-_I<Ou`
zAk*m+#v~3;bTCCPLrBCD7{ww_W6RkGvL$ZH+<w5HQ%7E<wHxSQuT^$oWQ=AcFV|Ei
zgq9<X!g?f6#y;=b*psz>FW|zP&AB|c*gt4q4>?Nf_)VE?JRM7iTkdM4M}t%*4CAH|
zk0q!G3&hU`d~{;0k9bZ_q%RsrokeC;A%tQ!yTAlv7eVfaN%TuGQ#2D-K}3slZ-R4u
z&=0zz9&ibYChZ$U(`?WMxe~OV+9}i@Hc5DW8op2(-#sjvCohdm2(2)v&Pdjdi10je
z1buZ&K(fSY5{=B;{vv(KfwVqZ$vQ&RY2_In7gg_YTUZ}lB;KimJR6%{-RV`$5hfYj
z2hyHghexnw-M1PppYKI$UQeNerhx|2Pf?0^<|_CKh#+21XE_@}9b6Tp?IvW)PK1xH
zjgRnkL`aP)`^Tw{JwCxp!GRTmY<K8~@an8Z^_dsQx-2`z(J@}ymL*y5(NX74PBEi2
zPx1s`11l!7-A}&bY5qFc-QV_J?*||P9+XA&$rd6C-*v9b2h3b8D}7gCp`_Q5(2|d}
zPJgd!kUvcHy3JP$zYLyp=v$8qGg!J8Wb~!GMi?3|JX4*#6OF|?bltex#I&?itVwst
zHat2;KL5EHY~&5GnAEfyZVD)S2mbr5Z#~tSsMCX1i_K*8)TNg2)A4)5u=LNHxX28&
z6l;N<(7}|Ws<DX-DrJdK#>lv3KR-*T#4nQGGMYSMafApZ>elI$jrakv0LF9nZat%4
z!YOsik8z~A!88!`i<j#zD%Q6Q11C(UnVJ?oHz5$lF--YX3dXpo%<!tGZySuye^vLE
zI%)3h7W(;|GKdIB4I31GaflBU1i9FzwwDAdsEDl~qdi<(7Lfm2Yy@vpfjo~{9v`eA
zNLN-?I$l_l!SttlBoZY9iodnE)BL}A82Qwg)eu8818^>Hfzaoe7GcQa5u2WAp<79%
zC=Uk1A>RZgnpE)BlTmSFLWtVqea18?6Vqj@aQ_};cLdto4rABmae`@3G$9QYelq;(
zi_+pg$S(NNLc`2kjdo(9O9CUOJ}tn+3!)=~JL(vr?9U?0#2y=@N`2f};u(t4U&hF#
zAND1PC7PwPkG)Pc&H5Vph-2N@pOwV}A^+SLe8H6R`H^YReZJo3A2=B-^e_y?jNveY
z{_{@@5?$ThGIzq&GzngDVW)P<)927{A1kT1Vc~$BMN1iIiIHs2jP&(-Ro&55^%2M9
zGwk5>Lb`5z8+PWA8DTVbx&K0*$-lT{yFtszE=EKuYhM7l^iE47vzfre+dLLm2*A{N
zcjjC0;qoUOg2`L>8Zl!Zq)P;pHi`;4wYkY+mn-x={(XXyEXAbzi94DLh!6ylI%GCS
zZm4@lpI$_XCaRJCA~m))ne?U@ehsd(lrw;Sz`3VUw%@BM20&0XaZEs+^?)7%8d+tC
zdoZ8ip6mT|>2|~?yfxX)PvaW_iV`#Fsf-_}aooJ--!CHjkK}pu-Q})%P5KL5;+Axe
zE6Ll&LIsNL>g{|e(lSHPMD2krnoM%*QBzrhfh&L2q2;E3?(;;Or0)E)+9rm7j_Z6V
zNEr3xZ%l90VLy);#>2|XJ{%0zTWk`$FXrOdzGu+ilV72p$iO2!wf`9TnvR*MUnj-A
z47@d4PrMQkK4yQsm3(thE{)LpF4>h&rH2ZCn|X*Zr<6b?^)BtJ{}$1?^e3Uqa>E>w
zv(Y42535D0-9RE%3zOiBwRWFAQOkO6x|>1DdC_&kc$Wxc(QECB3KN{S>C%+7UgF;y
z_<;zaK1cu0r1f7=V3z+qV`oY8UnyD){a-11S3*_tYCBEkh_GZ>Sx%NS1L@@O6(}J-
z92RO8K&kom^x!$LUSL5{UB(G~dgRjc<%PiEHERF-64f#h#eM_Hmhn$2kU@9DLeBC(
z7D1B<muH7Q4x`h1Es4(8r?icut9yO+v`B{3o1Ly7s?M+bz57h)PIrQw7HMwZhrRoS
z`xEvX4jiQu5JRFpd+}^#Lk!(r``l<knLRl3LET^rk=C(PZjFu9IIvlEPerwtp}oXY
zQN^luq7RiQ?BOLNxP|M6s;@`kx%af<Z8YC<9gH@Dikq*7t>7|}c-}@2&8C?HG|`3B
zj!wE|UtOoon+xG6S{N*`;q<<xIsxpQ>vR4kRc*DAde2*sdx}2kfB2NEmg}Hb<<JJN
znL}eu0v!H%)QaZH3Ywn$F48F3)z0aS&6>5P3!@ADuWvm2`Dlrxe>j>*ni>M3K><P1
zT7g2b%-d+I&|pYv_iHywQd9?SK*8EC!A<*`g>i<0U{e4|F;razYND7)kZF%$C<-`;
zb~Nc6-*u$0DnZKz^L(LjCN4?2ht3BMdH~^&qd)>ldAn$?GOoQz?Rd4NU)v5tP2aZP
zjl8Oa6$-jlA8}h8l_M%s(Lp4&Y!Iim0}UUU$6tGjgt`1qfaGWt2?w`=BXkpeuqLZW
zKyWcu8X=Wb7jiCJpNCwT{yHS%Hha?8ZZq<>i4Y~|h<I>uFf$=>{Jc|H?Hw{rN^NlQ
zq<L?rBIE=V%3u^FHs|s5x|p=DgswjE{KVUdIy|`u5jlksLso<vKL_372(P!}fiM~@
zzs>m4a8IgD9d{AwxPWwI%uJbH8}89AqILaQ_3HAHaI2wSm9j6YyeX!z2<DC7Tva0+
z-12Fc`k!uWP+4g~y@af|$flT1MCh(tLTC6VXoobXqgXdqp*(n{s^WY3a1Bcl2aHf9
zj1q|~`pUfN2Z<Yxc_5m9Qi9~3VMmk<&7(bJw;@KQot3aH2OMPB=<N@h>p;EF0~-t!
zT~dPJ4+|l~JzdSE4(6Tdd7)$hig{zFVABAweu};UEtb<!h`7(UL_}CveU?h}AIkM*
zvnCX=e<nF5-eSijW{gopsXvDFTVIfBqnW4SjVP4r0qpb`OAc-)yFjq*urTrrS7!R^
z7X`m@_)iN#W|Caz+H#I1?D_aUnqZXS4x*#kUA9yC!VQDS=O=!m8=lt51TgDo)g|o?
zO%QAPHrrzIxqyOwwvvHMT*PdKHc_Y4ISBIED$W|ZZhjk0OVTz@<SNs*@KqPIh`yYi
zYPiWzfL4O-f<ccGOMFx~GfV#lIb5mX3W>G|i!N{~VWwgUclZS%X3WN+RQ~tnie8dv
zwyf`|Mi0dBUKKGFeHnmZISMOm*ae7c%@8-|44)o6hKhCvoUt#D=l}Z?#9@t&l~U%Q
zwnl!G7eRh8mK3cKa)*Tqti1(%Z?_61y$z$fiD73*u(MIBHgR@;dhGR+WSDUd6j0}O
zYo~jYkSXGUz?x@qZz=a`26k8`A9ZN(c3;@!%aNeD#|Afr^wo%g87P$KBr<crewhi}
z$o%!+IFFDO;6Y_F-gzBuX)rC)L}gOc_&X|`9YnaM84B`**J8zwZA`UjYqDn)z2$X7
zu>l|av1cYml>?F$Bv~f<ou0PwX^S&Gb#rda&4NJ5LuK&7c{}FzR0(9W{#Q1Dh<Vv6
zYfJKm5&i+ZUqx87B53QY<8Fvnr@W4g)?4>Yx(eY<*$J8-8D16cud!$Df@$I&Dm8?p
z33cKV`Xu73l*%Q&=y>R5n~Mysj+TxWFIOBH2?dda>zetG5R)+0Tk5>C8aoX=d4|0=
zmNtru;pf}XMCU$^u&Ht{Zs=)UD7Z!=1mw8cXER5#LD8JOPeDjphe0voJfXHIf;;ub
z)S_9Lk;Lfg$(g@zSBt@_c2CpNeO!U0WG`YxYwu2&vqECIHqwLl<#8b^38PY_@{~^?
zdH7as8saD|L9d4O4QW&YU}3D(Rxm8!{3;Wt)`O{idviYp&O@a0n&&tx+k74badW$Z
z|8`{265U2AVXp5<m{S%r4-%XnA%4ISI+-=B5l|dJ0uvpDI%hS@5YmKAMP+uHZw>&`
zj>Go5OkPJxRB&^xOe{3))tYDy-H6?C#cu=bqiguZjsDu(Y4ruYwF~A-8Hf<T#O-T-
zY;Q&;KP&Br%6(l##ieEogTwIC*7NQ@9Vewl3gD416Tx<^am66u-?N5GVtZkh=$5xV
zoA%7MJGN=qi1G`aer&A~!}8}(Q7qL7U)~vTf7Q3)X;6@jZoz+BO8hg-3WcqNE&=sh
zP5>Je3!|ULn3avlx{!!R%od}0Z{ft|1-kdZ1SISpBxq2~M1GL)n455C08R`Ui%||}
z6@_=tyJ-yJZT$XI98vJjRKL=@1+4qM7uRH0Z=JHH-ZJ*CDRUAym>N_hntJzuIYT1L
zu!ETs++#zWsGqvX0m|RS^WQ;AC2evgZeC%Hs^+L+=ZvAPb8j5xJ*cwUJ<*tbZ^96f
zA7s{X7mGI6s`pyaK{Av78xxyk{Ezh20B4}*<1&GGwUOGD&6cU|tLJ1(K9%9k(_4BF
zd`q1?(<esKA^tu|$q&P&csI%Yn%#x#BK2mntTsRHJs$Q~tk=V|7(%!(h=tc<Kz=0K
zE%Imik1)r3%*$P-K{Q5B;A+-0P%6A-Xf?P7{eI2{n_s@6ECmy+RWK*PFr#VRahjuB
zW4g-OQ1hpJ9UZ$%*K<;H!!h~x@r<Z)dnygAc$($sA$2S_MNk0T<pz2{sBndCYmE82
zzPYY&z>A)7?;NO_t4`ZFACMq}mc0w^HN5h_7VE~!wWiv|eVq=$1LL<wn3(_B`fi11
z$J7fS{9q6G?c^7UYe@CyiDr){W}Y`xytVxg|4MF3mkg_uWVzkp$NY*qh>xA3>mSXD
zJ0jXnw;xBJA)*<p31owQfgxs|mnJLZzX{;ep<Z&|*dgcG31guv?|z3o35pvd>G6Eh
zwn4)*(Oi9S*foOb)!tcyo}!2mTRTa_&HT*ofhU_emoQs4<G}FqcK&G%#FHR>zEhXx
ziU&?#gax@KN)v`_IebW;Wb_=0T`{0W-XI*`3=mb{{r?ri{&lDR-+>o1+y5F?e5JYd
z4|NFd_oAOK?lElkS{8R!#L_M2n8riO>$Wq`H%jBF9;2out<^C-wYB3*2s}>Xaj;rl
z!|t?zkpv<M1PCDT{}rc(6!-1=xH2?X(V`{FIf0TP5;Y+~mM{l(ur3nC{Eljsf%{YM
z{&QZn`?k%NKvmodPVT#c;K%>raCiLqs)CJAGZiZ@+kZg7|K0d%>EP>{ef|$`c-28*
z@q8C#>@g?_HGQvwjWRW2i{@i<_SsglO&ePs3kW|-2v0WBX(cjkLz;7RIy|NR2Z0Ep
zA(ttXTwFW7KP8x1a_s!8AdtrS$pnocOL_8$sdP#S%0fZi89GZlrqt-92|G0CdP`5d
z!%c{idgS4r>bJiwJ{l_rwyAc+C{-)~;n)2;G=i1z>u8Sl#4RBZi>X|#g}vFUUsW$j
z2VOqdk2%1RBH5{?Fx~3w>&~*@<MX?N0#n3XCd&GEe8ENJ2IqtPjh_e4hP++^#xH9S
z!?Jn)!P-xnP+qS{L#KFz*qBgNKkR3s0rVL*T93;nU3i-pM>gO}8JTDP`UodTyHtd2
z+BwVLyTxaZ!#|$STd&D;&8Lwuc-AZY?!rmJnWH(NZDEMVKWANXdX(yy%^Up8>d|^^
zl~)S!B~WD=<&)AK*nK<2bT(hRa49Ctf5@LtlP6ha6-z&Q99l~=U6s+EW;SmdlDe#b
z1%oRbQptacFtyf(U55;$QGJ?dGV`&iKkOx-3=)Fm_L>Kj&oSXsksCgon@ja;6Eoa5
za+#h^->o{T?eDR}>hZFTebc$geY^YfcWge3ed5S|(gzi8*LIYPXEKRa6Jn+U4JF}&
zMm0gAaN=7vv{&xZnZ`HMVkgQ8+Ru&HRcwMTzJQ2o>lCtHXUa!n!k?Bfv~zZa=8_wY
z8~zEje=WAOvIHx}O$Z}?T1kV&sjGsiCP~~SQ^|TRFPUqrUo_JtY><&+@;i9fwxO8y
zw2|<EXdwWDXEJ)lQCzQ4L1HAbR4qOt{<|PX(9K6C*N`3AsD_RNL`~jwSy3yv1`!y@
zGH5|#0h^V<dbaS|m|i2-<v_dUJivvZg{i4cqaNQ96Ca(9>7zk4K`0;Rh}_XdAVy*f
zSvg)GezOPx&N)AbG-|>iZ94*FoFqN(4|}=GadSk>mBrL|l$hE5$yv@OX{a#M7E1kN
zqG{R)(77BEI;be2S=)Hy)2_@q9;o_!uHMTv_ZoYB{GngFu3|;eTaIic0nrHQuwKNs
zvq#mMT#6JDJ(}thzvL?LQ@R&cO_&oiOsG)x^$xP0Y{G+}K%{_{m|%Q33~G~2%OG)V
zBk}!!sIKbiJFmYJ1)yla;8hOaID?<o=e;#BrY;3upT81v`?Wuih<B><I80ekXLNV#
zH!;sxc~UqxF@6I|^^?7bO3LPC8nDEK_W`xMEfv$xyhIoX;egq#XV6}3o|BpE2{OJ5
zL4@%mJXpypOgDZ#4w{|^9#tJ{-FPo}UZqg<5gYB5T{sF7t^lNmblmRN4+VqZJ~jzE
zSDek2>u_C5_N#Il!VOxeTWhAvTMiqCcnNFu=ffQJpkwIr0+sw@pdPYO2{e(jvVE*o
z{Or~FFfa(Z(U;8@aW8`$idVI^h&inH^2RAg9yVHFq+fZj-0P+=Hw|7Vy#U_2^`1pw
z7qCZUa&ead7nV-j{YNmoJCY|kNLC9w9jaq}>{1RhR4v;sUi3Rx^{|SuMSduoFwLd8
zVaarm{L&9}x2rKQM*$9j6ee_K_|rZV@=b(ToXd8xrs(hiiB<^gCP9i^d6Q43*1+%&
zbbTyQY^d8^I%sK31uyT^<})`D9gGzbuuiRNr1;$lQ-o+`14bD`(g>02ydCn{3Y$3u
zOzIqQ2SsgKQ$!?k{IM?tj~zuGZvLLKPwtLdcpcY(9k8lZ-%_^fR5Rmpi4sQIJOVaP
zrR>pLp1c`wpPzb-{)u_ArM=N-3I^MgJHScm;NkR$y^|<81S}3^c@Q{aJ_e*iB6G@1
zV3(sPX{4m2<S>mg;)+>>P$XYx*_X`*n}W1pSBPm~;fN{|9(0>QvP2}3>*Qh|i5Jnq
zm(+;e2K^}z($&GU?<N!|I}!i|8@_m$2wIN5-RNx%o)-{zY)PB2zOh4_Sij1VNY24&
zPbm!K>Rf7k`|9tKm(>B;HAr=u6AKwcYCo392GR?jwzi;aI(Zn3MUKjKu8{wi%)LD)
z2gvSCD<4I!NP&9ha1`PQb81q#D1DQnnPQV1xHaSA$%eD+Va=I5j9-JmERt+CS~j-R
zS3{~=lh`aZH9q6yhC&47YSWrT41RB4Q&Y(lEMTrt#k4n!v+J%2AfsEMBnw@_M{Ae_
zl*H7vXKqtt<KsB`Cr>9O18cAL@klRr3{R|2ev|jpTV4?_7ikLZ2mR#LD?C@J*fn~Y
zHQ;tca%oZx@#4}Yq{;u_Hz?|>4}R8W<*W6F@Et9xa|j5-UL?aQybgMUwmL{F_%zb6
zJ~%;nI~a!<`8?G`MTakqDLLAAuo<TtzYz#b<%}=yE{T_?ka0h(w=BoUi{p4q3~+fY
zNh-SUTE@2Ol<(ReXc!-ie_@+t#|*s<5mojh8%%uF3+)gX2YZTZRsj*Eo?-Kv4OEp~
zdHb-YL_PDjkpqe^(m6;eH|PRWIk%ybot8|ntscJ+q)v$cwtUBi|1J9iAP3mdLMn}g
z4xWt*ag!<@w#fUW1q+oFXccs^K`uz9{T#h%zzCl0705s38T$rKBQ78##BtEc0SLdI
z717%)u+G-*);&TJNy6ueEc>1bF6y9y>$l5jgY2-9=vVB51Qlpk+K}WPxU8g!Xmc-p
zFWrZ|^Uq=k^edbvvI2!@42jHd5b<6&2bMy;tMacSBR?BatjdyMqvGyRJ&k#7I&H7B
z=~vI9{#@1i+XgYV<+%)Ul0X)(o^hQO?|{JU2aeGR7w<X}aJWpZFm)#5ydV-6-o(iI
zLRq?4W6(4#>s|7njpDnh{ZoET_;AMmOiKR&g=S&=|3UG?mQC7de!AAr*KR|BA?og%
zKHP0|CI92?yl>-F>$EGiiPnW&hD24<($kILP6(zD-)PH*Lf|=z2tshjciSD{?aDr3
zzSH&k{64f|XQ4L<twdRDw$j2<ZKgz-2*cT-H@$7DgD6>jtaP*%_w&9d6?OUkL7?~Z
z@r0o6ME-D@wzBhe|9-V`<M$2q)AN3I!|{4DV^_1I_j3>3<Nolq^9c#$*YQ2FVh2D-
zfR=PD^Cu5eX0!>=JItBXl!N|jm_AcV5|1q;%G9w9bZgdgeR1_ZS8$}+Q~sHEHSr5R
z>+869eXq4^=%jfwC9Ax~6_0VGY;?O3zu;ZPw)x_m?AxpK11%Q7H_ja<V8Y(v=%D}f
z2g2m_v|Jud2Kb3Z$`O4wTiR&%%pw)Y+Cq&z%2j!fLuVRQbB)$P@A&ody}pJEmz#XI
zt)qavqTN&T7mT=lf;R9jYhdzTZZj~GtmFjTp3ul6)Fz48TJ&)+j?@+X+#U+`zjprj
zVZ~r`s0C&(`W?k35BfBe>ia`2<iA?IvSF?$iBrMZiJ@X%oM(H{n+K~FA{>p#j<*wV
zoMN!Ns`y;6V?-dz7^cu}>`tF>kfXE9hJUD2D&6||@^sgXldEWhsO~QxbA3(YKJ%rg
zC-BEftj=-Y-vsvE8+I)=>U&&Y<(Cfa*A;N}0%foskj~P##N;(?y{t<jnD~AT)V%o<
zzZbTz#;?lt3I`1+4cq$cY%A(_ykG=15FojUlA{BvRO6y8h<}N;lUf(%ObMdI2<|Mx
zZcGl?u<bHV(E=Htv=tI&cnkgOwH2bW?Z^mxbD3r`i|t|vj~P?K*?GV|F^i30A$fv;
zOw&@K31XTzaFP`urWGOk>n27b7v^17YOq2a4w2PafTDA)8M^{ZgQK`k292Qm*8qbZ
zSHUI%;jp*|57OWo<XY!WiH0EPzxD%72*r?CmeT&Z4;he9W?2NmNg@f+bbHsM(aRKx
z3^)?^W%2h|aHRX^+~Y-sI!PJaJtP9lOGl@NV|IDs8nN2j7%^rE3j{tw_g5mXwRctY
zCD0pT3q&LbU{V(Atw-yvAIoOlBSEmE@!4(Q6><<Rk$p3CBu_M1W@b(aqW5#j(zJ{$
zvPs;d4lIeFF>eOFaLy=i00k(_5!nQhR#?7VxcrF@CZ>S>^yH(ZRE7y$9|InYtmBU_
ziC`MalS&TU?(_VuLC0U|YM>VIE7OoOvp5oRb<m@68FRc(IINI~E_7)8&<L?@Fn!QB
zMGN3hlo<yiJxd$Wj1PFO+r?i&%%W`$LZuAp6h)Bkl3vH%^9;67ih}HqGV*`D3J!)v
z#GG}CWO9|*y-CbUph0J13r&ME<}~3RN*h~c?!(O6CpV&nMP$;1olL%u>dZLVjqxG3
zQ%z(qGTH`#;9p~-gf4tNXo*PxNr$8a2`bvf=7f*+fj-Q$MtyUzB#&qWMsg)@fvmLi
z#k4A&4Wa>i3J)hD*??W}l@yuh0F8q~Vr!b|quzy?RKYIkHvnE-6Co?CZhk6Y6;`wk
z-EX=zW$flm>Qx3ii%Y&M$3S*3&rdBO|2B(NZ5;Dft+&CU)cQzEf*oGwMzKyvf+U0e
zens^3jSHSaA)EAPqL!0t-QGilj9uHGY-OhZw%;cP$+Sk!U@hd2X0hcwKv|A25=j6H
z5J%%O0{zY6lT3gP%PfmOAm1r&v5bfnHG!_oL$E?9R8S=t26&HY{3NWhwU`>>`x_cX
zm*|q9TT?zA^ry0RPjzi~*DN0W(A!NGEebX}VR}!|e@(FZkaGR{+%-(M*qgAoxigbH
zyWD4AWB*{|9tn~*3z0A3A9wG!Z0J_evA;fA^Bqu9$V$*f1@Vm5t4z4{hO5uY&3JdW
z&mfjF9V3Uru?3vI%|?~10TSou&H2wOdrS2hcu=`cr9I%?S(zQU(Mi=6%^7dPb1sOd
zxb0Wirj`{;BcKESEtu82EB4{+qu9qnXMv%`>jZ3H*90F0<|iyQ9K$tiMs&j;w_2uI
zAD8UdVR;~8K0bW`WKbth?>!Xt2z%UC2{Jqai1S25LYpvP%|5rk=R~L&<P2Adw6ybB
zEx;$YS*{qTy6I$UFAD{$K9<>}7R~rETG2e#5cv7yv<Yq?ejCBkQ<I2(4#)g)`5kQ4
zS>l!j67zTNCX)4M@m9&{Q9L(lE}cFc658mJa6a22j<RO}30aN=L(t(wVny;yBpa0{
z5d?cK#dgF;a&<p&f)#zPUeTC&<@{l;YLV=PMOCF{k`UgSes<VK1k(9glXA42AiT7&
zGl5(lsQ!cIYATg=|F1d0l6%3;7fI(lx%g6mfxJMglyuTUDP0|Bfcz1;t+Yg@wSm8!
zqUY(aL)JAb%0Wm_TmH3kfN7miN?wCDh03a#H7zw$tTX}n2=Sq?Ba~;UccOG}H@;)m
z#~|`1)bW_KH8Qhe^$WgzR9PjjU(^Lzj#YTeyLA^}z0?x+&5w!OUm%AyEw6(7U&EHg
zeXju-^EEl*W1P;jsVBk_J^el`l`(SqSL_8hcm#e32L=hhe67_-IaDA0k?~!C;5iRu
z@(8)rUJ1eU-vO3F(&oEYR7_oUf{G{2O)iLfm(JNQu+8;nei7ui#&zIkLbrg=kdTyE
zY1q2e>4S43(Te7fea^0c({gcnDdgg^(kkUI#XXt|(P#)f6KU=++U60At@5oJo9A&r
z@d#s!hWZhFnER}`e(Ds7+}6sqPUc&XbEu_9fF<WSK?7%}ernvrb;mdfHLq$B1Z77%
z4hLr~$b4g2!(F98KB1*Tz$&p7@C=Pt_0_!x7w-PTDPk^s(VW{+H;xp^IV8JghkyO3
zUD8i;*>lG{2Au2?e`AN2H&ZRxK-Rg%0~wjlN9L51)kWlT8d|tbo4$%-+j8U7=8ov<
zQdAyClxoS=i`q(3mEUkoh5ke775`jTUUPQ^KfEe}656tgI5iZyftcM3RR(gsaJ5K*
z7mj`p)ukXg6=|zYS^x_e{&q1sF6*B#cA0_?{F`G`$DvDc?UMr2zrccIIBIq7lKp5l
z%2gIwD8;$}hshMU_;48g^KVZw)nf+laBE$zCH!lzUAjd2ZZYH>aqiNM^YJsBDNjt;
zlNpA)q8i7&4$-FPH04i^xKXtu@z;Km4~?tOkkpNHV2J11u7MU87PadkK~SPv!BmcH
zMXd?6-;iLg0M_<=v7+Fzyyg(UGA~e9{0{74mtVC-hE;Sc+v}HH5LQ7;G%5_~HD3*b
z5Nc0Ho_v{JvuXum!d<%vBg4Ec;62YI$(&Ob-}C6mwwFg28xen6Ue<+$qP-vS4LZR=
z{ZgOAZ~KDCydXPtX%Gm%&fQOj=^|I~zvn%aGMj#)4JA=3(CtB8jr)VT6GUcIn_v`R
zqlgQT*$DQVoh*4imRczNFY6h;#N*+dp@!RK+w{B~M_}Z(iT2gJI_wX}{dI;2?;g5c
z1MufRY`sd#80^k}?drVKC|I~6UaKG3YX2Jej8WITxeyrbmK<tOh2!CL((=cK$7rZ+
zI>JmzxFh4kM9mqgVy4cYT(n-NZjkD}_q;j+SC*!UlilLj>%&1PXmDJ(#<62swVPM-
z_qn1B8ME1_PQVHtHmn6{US0n}j}cy5Rx0p-CgWq8w+dY$PBEX*eY01lzrM?z_*UT?
z)aQ2Mo>bw$tZAeoAVGzb2v5Um5`uDuVGQ%C^j3$>p;0L;8d=y+Tt56ja&bd_qy%O7
zZ3o?6odu`Om#WnQ%4B8HPh=`Q;`ju4&VSjZ>i{2;Wn-=w^p+F@y&^13A0FzmK+u26
zZ0m=Q?QVfYaN&9raaKN*pW8y6mV^M_fvr#^zY;_JdB^25e%J23e%Jn4AGgB5Zu@vN
zEPF@6ZYzgZ6UA`puFJD>*Z$fcyE5eTKMyzm_hJ9v!;gLE^%rN$Ti3)xKBW>$*!KFT
znUDS1Dz}!3HUT~Qz}}L*5eh2pw&IQfCZ<jsoN{ic|1L;Wm(@3O8%2L}xP^$O$38wb
zx5R@ae0u;_hfc|ib6H26`y6+agPUXcs>+#iOIk<=Scd#YUdc)nP37L(`}>d|aK?r1
ze^D;kng6R_%=EwKdTeRf{Ika)eRuZ=5ttfG-OAe8C=+(+Rj+_%UgQLP)N%sqhnhmp
zhf)r$y+wb#%sC3_JeoAwSI}*cOhg|{IlRq|Gl)ED;oNVhrf$^OYqo!*nb1a=;#T5=
zpg|<WVy!6FQU`u0OgP^yoX6E{L+8X4N2o9j{erA(=zbr_x>48Pp-W}n&!4xhi#Edd
z{kaZXLa69qj}i&1u-ozdNU`erK#A<!lZ$|XOQqySr`8w`F4C0Yu5=A8Y#Eq#Uc5vn
zhMP*|^U-gHd%Ag`ubDDv&+)Z$WUqB)Zaxw}Xm>3-bEGkBpkenoB6T+1cxrKW96Puh
zGaWbLcvEsLZX>-~W<{A!sHH@5H5x>QS^!A@4K|AyQX0Kgzf})e9}9JYo>rj-R0C%y
znf6&sc~!qkzGC{37pxBKH0r{Yb4NJRDng}_CFH*5h&#=*LOv+JUY4<t3zD!G@y~D&
zV0e?uDI|uIr|X>l)ua}!#hn8EGq{;5-Nz}AE<|F;?9|Z0B5|-B884ShOs6lz<%4y0
zs8#sn2Xq!*B*`V_$z35oh-&SFHm^CH{BJ&t!6IOlp2`d3{OE_^d>1Vp#oE4HL`GQo
zZ!4@>c);O<Z^?(yWTqqbbD(emMJb<`x-MA7kngKT5qjVc9seuLk5~aIiUr9cf$0xO
z&<bA<r37fl!M5`OY=H<eUn5@3b3F0p;B&w*-?4560^6Ixi!tZrcxb5Oc>(Ma{`2EP
zF^P30oUGH?=s3fLkEw)VjSa3v4kPfl!LO^ZPAW=MH2N^j6#k_8-i_qsUAaUm`wS}y
z9ewg<re=p!nP|oin$`1c_iC1RSF@II$pqaN?a!GjwaFZ~<xg%lmIkCOM*77N{uy!F
zu;N#YW3N9kLatQ%q^J$|!Of3T`jO2IohSXWWKdBH&bYR9Fr_)`Yz4b8XnHrHJ5Kcc
zbRrhe@4SvK=N8cf*jn$^Tk{|XE%Z42wxFopn925=ydnz7%@^Z~6HgADf1Uf8UI}AB
zK(&fqP)Aw;CWTx&Cq7U_EsG#2wg?+|&lR<_GJ2}w@4zJvRU`G~n?l08S>;!{3*`vH
z3%A2atVTApmc6etp^f$%_?eFB^I*$=UKLNr)$Aq|riSyLFf$M?9oPs=P5}#^(43O-
zNPY+qlv#9lFiO}>p7tH*DkcbI^Eov2f>ys-Z5jYTnfW3IM;G<fVF%gL{7;CqWd~41
zH2BM5zbR=R=2RX#2-?B2y;yKdZ%gzpxNq^~Cf55^c$*NqTtF<45+*JqR4(8!+@LO7
zM*4@luORM`A%|$WlN$ERT>QkRb>$#-TC)jBkZgg9>IqIEep8b~_=fX;2=iVsWCh4{
zGJWKi)ETCAEE!CYG%br~>ZiAFxxV37GqBzoIL5&ZR2BeD)W5*fCpMfb1lFpE7J6!~
z6#L7V%{mmtDk3>H?br4p4t0ARK=%o6MvCv@WQI|Ii{qcF>!DfzjczSZ)lD2`;WZg-
z8(7#cWI{B8j)&&)5B!OB4&5*Wz}vcPCl6mcJo5vb;lx&9hu;oB@%r;GpNGJ-52}JX
zkXNDsYUB(&pL8x6@`Np1o>{FMCJ03!E~-cRGOE|>%^V4NT_kbdIw1KHN4lMjwZ%j}
z4J@7Z?@;}&iynQJe2!X2!@qTyk~gxa;dqz%ozEW^%Z?HzE~?3#x&KHkX)(e1Z3+tR
z_YxQrNTx!KP%HNY#n(pz{dQOKuigwRO{PkKm`zpTn%n{qi{t*0aft>+@r0b4)|=Ra
z7(_h|wAn}*hU!(CfGQ|5Afe5-mG5Rz(sycY=k}$l%4qV#oxOGt0CmV_?b~bKCIH8o
z5II}5{G!oUw}o#DgO?3@Bo>fok6(Zg^5$%mp4rs14>D0?lZ#s}3xK#A6(`vut8JLc
zk>^=nmlfIL{^2O_|2Jd>8ykNNYveM~gOHMVz7^M6N{6f8vClS2todlxRu-Wm4SlYx
zwBeW^r?uRgCS8DpwxJ^S`crG}4(M@6&}c`!Z{==>#|Rv7h?CQ5yRO3Wu6MZM1X$?3
zVh~*n(yPhK^Zb4MHC}ZzlS3HkJxb@R`Fa2{1G2!YIZOg$50=I(&eL{4FTFi`+h}$t
z+wVFUxIEm_eK2poXt#jOwrna4i2GxX-?D&g$B=D~OOR!AeE{dXj=R71w2w^0+?OmT
z;pZtT-B@H}aMMDu>iyaZ(ID%4iZ`=H#>f0hX>g}neJ6&EkEi*3oaE<=qcb{<zQC*~
z^PtT6wFZ$R(!s1bjb33dEp_|tmHF)1e2#9g#>%(Sm_2ir(yvS9B}wDqAf8(hV_H7q
z<|qZ{?+J@m>}s0kC9XrqjK%eNDO<}|^cM?xj|76m+RpzMGL~rHQOgGS4BVE_Lr6up
z-qz>S@u7i&El9gozp5B`);oAdo#*r}J}K!YmItykxqIu|LmKs2P827omCzV;vtdGk
zXi^UMOEuk|TuLBbugj^Swn#*kGu`4}n(s#@E&rvNi2_(}>L=S<c8L24?A{`JQ-!<^
z>6V9)I5QT3!Wj18?T;JuiFBuOHvuk)%Tsx3iT6T?qMZ$}eWiJMQp%d=j!)q<{_l0k
z!4jilM7e8HInV3cO{KU!Euqd{Jhy_c3U$&Prf1E`*T<s08}%SpOBTK@0-mn30BS1}
zTa`m89OyEQf&vC6VvdF`d40p79UX?}oN9s(=<+&grkW2&?Gu7ZC_k^Ct<{3TQ-d_h
zEb7N~+4UP)W$)8<aoQ(2Is1!F7)+D-xF&+{A22y$uqct|og=d6{g<^rDUGc5d+<Hs
zN%Y+UF+V^=&DECwH40_<ua?9=6!-s@JvOAF8F#>j?0fUia9bL{7M3i6mrg3c#yv(Z
zNZ!E${%nX0Xq3@jpWBmg#Q8(`cBO5C=F(I`F&f}ecS@{T8KtV9;3XdCrN8}Y?DcR7
z)5pSKtcg3g{k$7eNAWlPPQr7e>tz^$1ifJ+Cc5qC&F$y&{3xv@OTh7GL;YvCPo3fJ
z!_zBE<lCIV&TrNC>oSTmyCItNfH@R}ZiI4lvH(X;$Uc=UAbI<J1Fy?Xw(I?Ghn}^a
zZ#$Ab?MpC5Xf({8sOGr)>CsaRQCMB@Q5l9mT>aohqg(6S3_*&6p;>|l)ha1`jBM{9
zl!lhS%L3r4pj<iMmYwP3+wRqJy;^PQMaPj9^MH`(``>kyp2Dug?f}LzSU;ZQ#e{K{
zZ(|9N^xY=y16v@DARDnkez-!s{#hgc-+O9+fham>gEYFi!856`WV*)&cI+m)sog&1
zqr9#M&1ds~U#wQ?wVKH{{tsN$8@TTpY&&<K%=twC8!N{z)JRI<d<QT&%J+XJwKEh9
z48e3pEK+Hxo;KuFrz#N0`|y$p*;4@?cl~E!_3LTjHMTFjaqu7eS+=DCW#W+DUok^k
zq!%l2T+Goz9VUBHckpE>`&KXhvkv_Q0>m4jS`r6cd}VS=*q58a=#U||{0->nh}8nt
ziwfHdQjaNwH@$Kya+nB+MKp7DAZX2l%^vd@_ME1fbYzW8pT;HhH*@sS$gMbIjt_Xy
z0>vlBg&JVdWa3#PdZc71@x-T#IjkFL5XJ>JoCx<$N3XMo#FSW$COqj(aK@?NqYo}#
zj2kG|Cx~dmB?fNtfTdl7k~{67E0KMxmTquR8Wh>``epzz<0wQ^MdO`nVqeod>)&B+
zgYP`@2>)z_+}(E7_J5+u!F7P(AZeVUj)v};7JKOm#LQ;CM25dgz0?+#h;ZW9ud*x_
zBoN-wUGdOXOF<9OGhTlh_K+GTT=~87{a@MibHwa5L3(B~{+}uoendDu`apIlMYPCg
z_t!fY-v05LXhCs){44CfaC5pj<x2cU4-s8Kx!rHa5Hdo~H%|LDzj<L|csWzbz>ZSp
zc*2jL+tosIc!igv>nN&*Ewx)K^={PduUNZzdTd|X|80mF4S)gboBG@2h%)`<U=;{T
z>+0u^C+oQE^zUASZ(ki(ZK34LI(z^w$+TTkf80bfX?x`2c#5U2=7ZpJY(%C8{F#(<
zd+-SzVw<uEow5L^rX3WLkX$|C^_>|vP39C*%syxMvu5Rizs7yJ92Sk|)yWxzV5mt-
zb01ht3RV{&>@L91!x1|t<pqXby$tw>Gr>_5dj{@Q70}`_%d9q51m13g(l5o(DL#AX
zjc~px#^%(TFd^mST(~0C_M8hu3HAe}YS&A7Y+YcUND^-I9+ot<Z@F*WOWF(>W$2P7
zVtL7+O3zi-S#h{w1vFVLwq;_(5_-!fHm@BE2o^za*qf!Xp1L$a90nOOiaJ*EQ4Ek@
zsPSTTuqUD!B(#TrA=DHuBKb*9(Tu1E88!ksMWp&ih4iI~7+OJ-&IHn%EvH1xkmb6&
z)tA=JVU1yfv2r_?AO?msaC(LGegmj#kCjf(7Lkm4^+h#~BJtz@4`c5bBw81AX_jr<
zwr$&W%eHOXwr$&W%eHOX#@z0R>6nOqBj!8j=lOkhuAQ0dS**emedZ*A0s7lf!}63l
zVrLV$pwJ(MylP(BMu#$7X#fS_gM^aTAFW*QB5j`gAb@;xx6YQIjs#4_Gb2Kq5S^<h
z{g5;#wy*@UoO9cnqBb}PFg)=G_R0ylN%@G7SeVEwU5)c1pz*5*=Ft@|c-|)TXfwpK
z+Ie6@?SKOHlm;;a_ratH_Rx^as4Cra62UTkSBA)5)$M?4Od51v4tC$LOu7>VeF=`i
z-)FW7vga%Tk@y&#xBzY)B#G&0g=-mtbgX~1WYOTsgQ`!v^6F&)9)OE=a<}EV3Ao7y
zZsKV`mrxtmzF0|5-r%53je;oDEvNX*bgIBnsw}`!yu}@XYum=#IEi%Fuiw@8aKHjV
zHURg1_^=X2l_4TwI;B^`)2aQ!96&)W^oVwo-3<~$W<{&oIDl4IV>4UV{qr*WOXXUq
zs~p|!1;eDrw&II7@v;r1HIA~d2b6z7BO!kZl#x$;gK(i_?9wMrMNv^Cgg<iw-Nx7g
z7)dS${S}6OxL30Zw%e6LhAL<af8+hFjX(~C)W~s`j}`Dxb<f{$rJxWsX+=(c(P7GO
zStSDyTcnPO0FM&HmR_FH^B)VZU7Xr1O*`bbcY3GFY&qzie^#!ABi<<*EC`ny0BxPX
zTJaXpI(19)ipdsOTAhL>{6#hPCIpKr>iF7oH(L(VTD`igr>vIK4@@jp1N-AE#@k>8
zKwB&C_ziHso_ahc`XYqp1z6$jU7c`@^#9C=4%q|+DO#5#C?rj@W3i^vRY_w9jC@k)
z+h%Q_a;U=LKXZKScHqYYOFb`kkhmeRAK0nGNGQEkFlf)C%uA_pF0vBqi1W*=%L<Xx
z2L_igmH?KC?PZel^8Fq?Q3#;SiQ{k9F1<A(ef%oDx>w=Z=oSy>;x$%AdFD4r0<$>V
z8NIyJ1!?;CDYds%T(*h4q3mHfeG_4%$#Z`HGlFJ&)6`o03n9kfsWQV0r1TWb5V&YO
z2jZav0iecINJ9&zff;=Nd}<8xZw2&>n$?nuc|5&HH1^>wKk-aGZITANV@6i_=*<ZO
zKk6A(3wR(r7(TLmLc&sDN#E<&U99C5wc}Ga20=+nj~Po2ZEMrmYg^oDFg8xQVnkti
zwPikWK72|6jxL6CPvxI2=F!gy@ho{5$Y~icRSBi@r?>dGDZ_kRt1W$SGbh9(Nek+H
zmb?tGltWd?tU!oPeu_}ZGYALGJc1k-?(d30@E7#)cyT1f@HQA}4oa$z|8aV6xztTn
z)zHWPwnO|Hr#_+RoP_K|LeAVY<fROq{G)em=L^4s-h})=RnC7vB^X&a{>QNKYYi>C
zEOi9m*E){4*j7GHHM)sbrM3$jHOr=o-Gy?l)`%F8P#`(Bnt$1Kr`q-~Y!=~3u2nC?
zed)}P(^KqDU<Br|5T2ZGMm!wegd9)~XN>SMvxu+pD0pN-h4(NyaG>PiW{3RndJ0Ik
zTVlYmoQU~DThCX^i--4*NkNE7#$%v-7?N=?al5_Pf3fp;c<+YozdnX9&2Fx)v1=R!
zaKh076IQMjW5uRM1<y2~4@Hez<eQw4=v7h8q^ZYdYS?f-WTck((43Z@SH~y0Yd!h!
zym1r&<Pk)(jf5zX_yz%?=UH&gIbTV(@Coqt7!b<C=ic{jy8@j9h%gF;7%&Lc9lOd2
zIUO)dSVZ(!AKi151{hvDV2?#vY8nd@fZ~5{3g29i?VUQMDmmm2%%!IfAI8u1%)p7`
z;tRMOxk|vJI$k0as+tHn=W+Cfto6wRfc<4cLcHD0Sq-DLmMl6o>78<^x-Rs*)J;Vz
z6JJG;z@re3e1X8L2peh>Tf^e2Tai%ai+sx4vVKsI|1eJq3qgf@`j1em$1d_jESPJg
z^nJx!zN(jhYsQumgcvLMVN6%EsD(%oBaGx-7zU7@vfW>|FKe4J&1qPutGWuJOPn~s
zOj*)!a$yKwSbDehfuXHnv=1@qOF)ETBLCSpX_Yc<-7?Fk#4zpXsm!Onx*IK<%lYc$
zW?kQK6!i?<e3@a2E4}v;T3gy1*nn4h#p9Za!Z<L4NmEX_Vv)^&+uv~Q7$j+C&0A(4
zDpBQ#I5^&F;XtqN1A`%fc8jnnW}_f#Ko@#6$j{kO$86RP$Ex9zZLA;$L|;f>Qo26l
z-w5t4XY%?oRz9vhDB;aSAe$FWUbD)Y=@3;-&EGiB7Ex}Y!VR?Cf4E8R@LFng4f?=V
zDdS7++_oZbj4D@v#4s8FMxeu4q^OHsVUiABx3mNX7czlg-MOe@U7zqQ;r&Ba^#PWq
z%2DI7!laK>W5tew?H(}%6K;8Tffa2Q&pwF98j;lA*;w2WT@}J-G(!|#+XGzjNXN2+
zInt2nU>ybJtfmvxTpQV_A%%B92)q-5W+L9A0AESkt<xgV5N73Qw}=Sy0p!(d2ixno
z)=!PM_BA*^6yW6N-n$?;SLl|axzWUuB?|eZ4hKAROq5Gk%@GH6rc<#PZ)T$UC!~@s
za=9`(4HmoK9*>rA4K=uK+1R`Ee!#XMD8JuVn{rGKRlJr+?YNWZkw%Xcn^_#wuku!E
zc}kY-LTET5lSFYl(^~G&W5}g;rP9ll5Nbn*gE$jMN<Hv6*U)HJqJP1>ZRy@8JU`xx
z!J^G99|?=rTN|z~2I};F7%&&+8=Ec}Bzp|PR8WDF4osdMKWa>VGzENC5H!l`Hh4`d
zeYBzE>RjSgBdhQB)x=>>X)d|zg@@A;F#%CuAlS5jd7*-u&2gFCjHgT<*jNwPO3=o^
z4?85yHXiwH7+e*PAIDN*5S%66Ga(3Sy@}L3)`B170l0DSmA*)KF!&R%#C?_Q*I4n{
z2HOM_S-oDdd%o;@rM$6lJ-qNdEl}4g*nRn^<zeua-ryhaUV3JV2KeUha>Ola)>v3p
zu8Z@@93873vrYG*-f@0R!wYaX%o;b@MA72J)Kyn9$%l{&MZ}{j%U8^G?BySwPGQVh
z<!1vVRi8j6Cist<aQ?ffL)IDc&lbGbmyWVgesI_mD%sl^;Gb)Eeg7BYz$*eFy*xiS
zvZ=zO(K84zb9u6OI4W&!SYL!Iuh$+zF3}UxlNgU<80Nq4nUFDC<vR*4%Wt8q5t0A?
zs>C<p#OXsRzbU5AobT7R%NaMQxDzwj3d$hmPw$nlQIX@NsWY?7I!&$Jh1%(@*VpD;
zCp;^wwUH~sGT{68Jb+d1Kjc8jPvBwt)y{?t=RWo@BU0p{v)AD^hr27Lpn5;(#YNl2
zR?bK}S0r9N_qMosap1pzQv7v~|5X}TnEvmr>3=yC|6@$RwZ{K}Z`Y_;F#<sN8*e(7
zT42j_EwZ>Qkq;*oNJt!{6QL$F9-4mK>gWq*6dz?>wVyB11GTqb^&2*DKqH!+Ae^4<
z4WI3;OG^=vKM;)<GiP%3I)kL)?GIp<%&a|kkoWW}NF4k;ea_bD;=_i@AiTC6va5SM
zV@<Xd3HEK{>%q&3{pIr-|9(l<b{CPDGdDfrTh4(C2Nmq5&==zPc4T#`wdC4yTcR%t
zf(J)gLJH&Ry+y2y>Vwwr7(Wo0^r6i6JDXGR^c~AiaV!P*yNYsb61xb>7QG|(##u_h
z*__r-<|b5^`<OOWxVeunJU6XWsA~Unu1^8!z+Zh>ICCMRiW@YLx=_vvT>^t(bM*_!
zy$h7UTpa8XEmh7i3vO;2m=gIID`aN9R~{#wvUHnf5^|TTB?>YX$zKmwR8GvfBO~D1
z)VaHjM|EyE73qmS)L9MqSyXg+YXyJRtgX`(5bX&B;`+OTC`jg2Rf^PUeKHZbqu9NU
zrM0u15&shNbd5W8_3&e^PCb16(fk@2hBMQFubK4unaR_UHvW!o>2Uu}+uN2@rk9%Z
z2;Rui(SNk!t+?u4Gxj{~M6(-B9;PmBospZKz7$7a(^suUc^{}N@z+)gCkX8t<^#*i
zczFG@JKLmXr-?&v#t0fYDx&i6{-~HZYdA`jN3L@~<#>*ncmAZ$1ZLkfwb0G+ccc_I
z%Dg4qTzpX%eLQw@=G)h9)OZ29D8}gIxTAAl!TDqNcw(cs7VcGin=|wB_~Ef&hf9k;
z--+f_!^!6M2x}HB?X}A#H}vJoIruZa1Fxq9&j7Y4IaZ!HE(lW`dhW$6!K@4lqOq6@
zDg%jN)iOnpC0PE1P7N5VlWIDVi+j^KhYAHfV!+m+b*@^zZG<(|%3OPyks&7v(B6Gb
z4*|L@_p%{ib4t}fBe_+D10GmwfBa^EJRsH?jCaafA^BW$@UFs{TtktL2{-&n4V6CC
z<ImK;n1?;aX$9&Au=M)EN-iycfMLYrC8=psOYIM6m}E>Xb>|_wJxF@}B?H7)Asf$_
z`|4^{(Dw*<Fn3jdlS7rwfH3S@o_yQLL^K_?UB(Rz{zS(HZEToma9isxg2wili325Q
z*Zy~>D%vU=4c7XK?q(`z%xr?tcytr3y3C6>x8QmbWziTXtFH5hZta{6kD3)U0u*EA
zD}X;$jgqwrmBI~-oj)A0llxITa&oh(xmuu3s)cDZ3JSyM;W%HsNVG?Q!C@55aEgAa
z2CZh^3JUCM4ysy;A%)kVkb%&F3v?1|hXfK{f?;b+8k>S^pmI}Gp@V>*=@gr!O}vR~
zt{D_4G%Mpcq@|{<g)JC}?gnFRt%8GRg?!BmP<;Vir|56ItYv6a7Jo=Nx?7R8xzc+7
z^2h`cZ$u)07(x<GZD`*NpHO6^Bv<fBY>qqaCZ*r0W9%W#q3+|PLU%1oYFAv$d5Xo<
zn3Eu25C`O6`}fm<Cu4(W>(9O|MPjE(9dd~&YUZ*p>g5}o;#6tVPX{F2Xtgs;NL@p)
zO^KLeQ#0CPs`rwd*k2H=V55rIx4dm9;7pI~VFh*V!V{LQhBgro^c^yt#m-f2+gyst
zqX2<k!Xd3z7KpefhId^QWU7zXb4TmcMZD|wt>>`5Z#{kjBt%=PT9;Ab32t2!c=-n5
zjq^aNG;|RON%&De14kA2#4>j~tVrY9k`A@ZZ*hTiX4hhdt)xWQ2+`BZr=%3gUu5<K
zHuKwn0t@+Ti<P%d{3r<nBN?q*q7IfwRagP{d;FQTk?@yBv_o*RR6A>0Zlsw4yb<A^
zTz663Rh~|FQB=Ez0LMLTh%wK4-VH0`SzE+d&VbnT#0q*F*^_hMN^77S4u&lzyz`&?
zd(m^sN7FOthM!iV)$8RPR#ry{&nxG3&emyf_i#TyeqDyUd36H_OHF9_)?b8?`a`~m
z?;FBB#O#6|mXD(zxF?TyzxxW|5F&02YE1JlV}#!&vZ^))im4(-nP(3>%mELuJxqWP
z6B3cxo_X+_I12;7k8&j{LbE`j>{YWhc&TRmoZYy5fX^X;4Mn2BcohaV0Z7rynma&x
zf~PGGqH`L7f69!|;+D50SIf6GnlNBib%)3Y+c32<ZbB$a`>8$SRWzPiB8213I4&zd
zO|bQmUbzWE!2XwVKzsC#fZW$E`<;#v2L}7&fiie^q>s$pS@tU)v0**ZRlqv~^`xOZ
zh0~U($cJe@PQ_$nDBC(o0?o`)l^wB6k|68!Pu(YYDlH+!7qF~iY2<%aM7ICCA~G@k
zYZU+Q=l>t-IIXcA_YaHhxBFMfj=YF1PVa-DmewA&@%m&xaR+m3(9~gXqM0_ErX#W2
zYcuTU8rP9RV_|*9Gzs*<P?6Y?gBfS$%Ty*vRv^ev_uI(h`9(8Q01@?(L%kLl5j-Ip
zGDWdo8?>ze#oM**YyX<H$dVzZBoKlsAR_B|`|r+g^ViRzbw4o{l%#1CMd|)OlAq?k
z*PCD3xV^ih_kAYL_e)<xSCJlslxklQMJDf=jN|qEP+HCBZ}-P6YLJaZ9}ppg2~9i|
zA(335wVTN~QLMf%)?GS@gF=?R&+~2C1P2SV5tT&0f5X|;6iLJ)2yJ^^<}AFbRpvTT
zDa#i=-_2v=#&PGHwn|p_cZgVXji$Q`^XW<?U=ref0t{L-azBaV;;?U3!mu6>`Qn9E
z52G6~6(LA^qnkV4U+K{Zd`6*ugYFz+B<gynLI}_#IyZuSN2@+9W2~D?Olje!iKf;{
z>|#{a=i%7I`s-3w&r|VNvc<5Gy;sYaA>W-MUoNFN2XWWincUGx5<?=CHSpDJ)sQmE
zw+Um_jsaTgwI;ESKb^}GQ(8@rmKG|gaWze_3{f>@4#ugf)J;SbpleqYpdrX|kZN1I
z?UpHN*xmMfq@BJ4_ocOPpC`>|-u2|#N@&tskEK`9IqV!CAxgUJw7pjX#fCR={obO6
z>`m2Q)!kE(z+<s?8-6Qq^<mj>!nqCgxuiV71cVr{ez~GZ(aq1N8ffym>7!8K8Mf1<
zl}?v;JB8-`4MN!oEJ8j~Mt@xWNC#J#6cIUKd%1P8;{~jzbXWUuTZ7NgfV>2g0{kec
zG32Is`Bl&%tctR5R66KkknG7Zckmwu9#<vSPfAy20M^-NogJh+D<w#}`8hGKEh6s_
zVS$P5(!;$50Vo$^EUHmGkb1?D@Ze#@+Mp!mK-G7fTV~idU3He@C|01zoAYDMYSeJ2
zk?(x_`#4{Q0zg3&W<aI<^i<Z<+|HtBJnyWZ(HHmXsqWwNqlCqcew1V6W8++2ZJk@6
znOAamGrDR}0P|is&j%x<;}OE!e!Bh^_Sxzkn{3AKB7^+a8<Qkr@EBx95h#aKyc;DV
zlEPSnB!Kwpled|Ryt^|QNT$JwPJ|~n%rd#a&by{h1h{|h&tN~Swn5tJpaEUVcYHms
z^X%w4jd_Z564p}B<2x$y3Q6EG<z0YU@Tpd2{j&e12faL$pRnIW93F@B<xcSYF{~i;
z6s=A>sNemYyYz4NwTlQXaODn)Q?IWm0|F8TiXOa(0vdXCd*~m>!t&(S5Kw(0L>0{#
zYI8Tw;(^ZHI8jYIZCXLYUJ}RDiRsYT;A5uAK;l31j^Q9zEoGFVZXjq4n#crU)-d|F
zCBXms%IHN_SLL`1*}$X#U`#SEACs<|GOl3)7GU6=qE*vFk`U84OsFk?eM@|*(dFV#
zm9VMUMe@l!Gt2Xh#>QL&S(|PEXJyk}om-pzom+XbJ>Pnb8BHkiCF4()!PxbQM55Hn
zG=~KiSS>9Prx^WY%}ZD+`(s2-tV4%<Fba~vzPQQr+WKOE2=~O_bbir0FdR7yX?YbS
zhX>0P5HQnR|BMo$N4}US2<5cp0*VJC)cdUu{USGRALzN-4cXql@xH)P&^PBWcYiVx
zrTKQa`=q>UPaiULC?*%@h38IAL_`br{gg<I^nxYCQ+o<yrCfrU4l(xuPd}4zVmegZ
zXrL=2WU=E2E#|FsMTmUMM)O$i59vXnM;CsWq%>m*q=BR0ZDzQ$ZUb|vL<(HyD@6Rq
z@HoPQE|Lrhl7#ru(nQ*9KiXi4?bC#yoV1VD?k;Um#`c_OQ9uH<v7Q5SwUPmGfD1II
zQVZ~YgQtq+<1DbzF5&r;sBZ0yI`vS%=~WJd_HPn^*;z%m_Fw5`P@CCoWiHh;Mj!#q
zHYe8fVEgFf?17_$N*beC>8{XV>awcIWoIcOibV0L9`Cf_#vNqA1>BmcRdiTzjTlBg
z-9WatlpdmT#Or75>tj~4Ua?U_-FM5gL<~k097;)Ly;j<pe@hTteY1@9*frhNwLnu4
zoUY9W)vdz6Cx{8ojR;W!n2uf#f$wZE2;NTsZ6mL?L_{?$;~QW7`jAwJiDUJgS7<gZ
zZ*_uWfkV|)Iw2aKV(#B1)LJ!0n~xIUWHJRilJ0eXMu%>ya0aqy)z1`it6VgWjQLWS
z!92wyu$+9IMwmY~0Wr0O=rCv(bVGXISZ6qCN>)so>hgbj-KpJ$Q1-qN7tw<^ih77d
z3*CuK(5vp|$Vxba2nGo`iulTc|7DT5r-@%)+&>06@CMnD0uszF_MZ3y*K+xL2GW+(
z8S91S98;3`r{$V$4i3n8y?NVxaP_}Q6>PSNN9*iPx1e5V@1lcMd0_Jprjd+eWa^!k
z&{;2rHQ4jjVqSD#avjNrb;<s9+D$xb<5P}2QoiH&Z>$w$`c>Xj1<Ob!e_-u`Ez{bS
z5t6*>w?t<Pi*KpE$f2WA`ld~x<yDL0H*48^&DwZ-?mYCK!)X%-98Q(`PVwDctJS?~
z2)n*jWy9s9Z?l_`{l>Ypd^tyuG{CFQ>TtVsXl+N*!QNIPG!=|=le5)Z-7eutv)E53
zVnw^ru$U>;46is$nMRw4N4@d>mBr3hlssskoH!1$*zg7hFm}9j^QZBt+S~g)FVZ#g
zv!j;hNam15=d_>Nnto+Ho`{!B6J1a8&&PGp69nc!&JTN3!LA+~5=M&--0jxG?Ygf8
z{(DCrAXxJ2cF2;5S1jBfpyGb!<#go=hq^q7ZXN60w0}pR;$c=(8@2H&i6>+8dw+*G
zx#E6!w%lTDoxqO&Vh%W^X5$96Ilqfl)ap7Ux$qn^FOTb54^Mw6tc#0PoIVf;lZFxE
zjC(q=mzMJ_KMHrlyFZ6<DmHMw&>#)=OSEpc9lP^>oS&EcCN%z|NLe9@B2TVGg~0L)
zFW>P){U1_@jrsq}w^s46HzA;xH?&lCwt=FTBVc4;_+Mhk(aD*Bla1s5@UI!!SpUbc
z;vP-KxFt3O-(9_Zad!vnxKSfD@?v0T(25iU@siduuXa4G$o#f{<Dav?x9mHhY+dW|
z(m+af86<}p_w2hIMrBuH9LPwV$UQ$<(ddeVSp?Q=qQZP<uxvwAB4%sksDYAv7e$6h
zmuDEVuFd}OG7fk}>nJVBL>WY)A=cfmFcCTT{Yg8>64eHM0ddP5;)j?k>x4wLEBA*T
zAn9f>@zu2*7<o;h!*kTSIwaa$f6yV>ddhVN;6Vd9Wn%5~a0epguF1ZJ%Gr>3obK=9
zS<bX;`r|=hCG~4KTFCn2odCt4Kq@V%Omcv~5U(ndVH8<uKq_YRnDmc8&=tsX$ZjBu
z3k!_TO#YZ*+N1NBz_&DdV8}QZpE!!OvJsLY3$FyD8Z&@}AWtxomC+4})FPrYwU_Lp
zhN?JQ2k;UOL#8z>siYQDhQZH2dE~_FV<gq_fh2~~<im~zYeOhSzpL;93@8ME@aCWv
z8>%OZ<DD>t4}GL%JYjkt7bfmdE0vfhBD=^Eq)poUC*)@=APz|ss8C|p!yT&fE8&jE
zv#cUz1A<qAgkHd33InR;gVjnYLSm)UtyQjzGW0HBAyxqcQ07)@vb`xovJw@mOCu0B
zcnWX@H^t_gi!AmFNICP^dPq<X0PpK6zJaps_rghN(oE2)=)N>a89Rn^p{b?{hv36I
z)<6p89BCq_`_{4&YC33d9UK|Y^RS|qi^h5ST4e72z9AO_eXo0pxIEr}e@+e;iL2gQ
zyEL(plb3P^QQuX;aXgNth@Z5o`A<RH)Fmr*rwGf+zB!gu^_gJ<a?1K517)+)+p4Xe
zPPTL~Oh=WOm3p*%HM8j*xc|FMk>Y9hXwYSnhg*0w^)4P4kJOv%(%t^Sx2sEY%UUkm
z`E+^zxL!V7CU+{JBi=KUOtW`(iJDh{DWS6EwI-0{j;GL)#*P&xH747vuX{ypWmBO=
z(5i<GvBcMBSX@MpQr8%g4PyC#lbCduD7p-Dh8X<VI}3C0b$>(1l|bH$0LEBV8knR`
z8iv05Nwzv%BU%uU89n^NWuFAuevE1E0#DQe;hy}*9qm~o0}O4EQ2h!#aM&9FEew#v
zin2h>e%<uWXsFW_CeG(57P&s=$Jf-=Y5yK-4aov%rvvzw1<)=-ZsvS1I+WODj($7I
zclvyK8ft}(t*2$A>9m!KCUdlEb9Hjla$1LNHuH2ESv%kF?-D>QqG!A{BxRuxnhgUR
z<_=Og3a(ZnM9*WKUyVW2`|<U3*U8gSv8TCQ-Kw1$`?Sa7=H%(e>t5dT$KbZS?BmVF
zbk?i1SI@^QA<>4cHkJ}ag$@Ny<Fo&FR^-{dwSzcz^>#46Ui0T4%I2ial0$oktRfjD
z@wOCq?~G!W9+{7oP+n(Og13I&6z3<EifC}TEr86{<i6g$nwy8x+EcGH9Gmag)M_IS
z%s2~iVlo!M#W$bvHRWuZ+?&F+6B@#p&X&ATt$$yHu^X=79k92;pPx#2jnB32Tmjd?
zt-x>R)sNl7mq)fas+nNRH&c}c{LEi^qSskqFO}ld?%lVzI8txN8@9oS%5uRBKOfki
zuXnfQ-#qU-@+da=K!oQtc-v?UpVdRtts^#Evi}&*%050`EKd+n!smYXXx8joTNLrI
zc<~C)2q$(H4b3FxdM3)+$}~*<`cmQY&?CF+n#Z;{pIMdl=w8)8)~LRk^E2Rez?G&!
z%Ftt%t}-uUQsO7Jc```IRp(kMNw)T(zZ;S!%Ps8ETK!s6t~E=4SR^25m5ok?(0;fb
z3O7fSnCs6!_5iIb*o3am-qV@JULd)t>v#h2wRR58XLn8BFHYbtVzpbO%Bo*mjJK`M
zw<^u<Sm`lBtRviZ{evEjm}h_O+iG^>$0gy{i`!72qqAM{{5>QT(VJu&<)YT*5;%e+
zC?zClh5zt*VvG7}Px^5c<`y0Y+YK<pj5EZjhqs}&AU#+{KCv<Lf(P_u`&eu%K!R&4
z_z<<@DRzJ;zP<${Y&3~b{tN!J@WhSW<eX1|tdRYEDWJCu$IY90y6LYldKnzrD|fcH
zCCo4Fmu0`?n*%^WZC+)@E4}flWM6DOdbl!QtB~oRHPt)H`B7o~RBXhN{d5v^)h;Q{
zKWHE?9|H542R<^Gz_k~SytfC~vZy3PoobU>p~BgFa2A#@i?99R@PQvuTQ>h6#g3ho
z;s3t(GcvLNk2~Sh>e5dCVkrMN#*Gv$ntj}2F{8O*+nFh*s<MY77e+*gsmz;#<X9a3
z`&tJnUxcZ4vArmk9;m*L4sGur3QA-S4AGhK-ItM*w~R!D5Drz@yg^KxkP_wJj8rr_
z6%T!FbTLC^$m{!LdNYr{l$k8TjnVIS^Z06e{nqTZde71K<>VSZiW0W_^--@(Eh<lL
z7W5Vy4GFnIe|M#JuUmtKa0XR`(O`}^oQSRg1|yNBPbJtqq5or(MAm?OxTm{3EB3Bc
z)Wll;)U{GwwIx=Gyat*&#KlE5ty7K<VSvayYO7XVw!)04=kjW4^<+nbCMyZUN?`TV
zf_km5`WCgT2DkKNq_~|^&IwAW%cxHkDFWDhlQ<ilv>Ls@?8doOb~*xAwX23366OkR
zZl1k#37cKwp?pmH?{+nET=L$NOPP|3t8?Oo>PAM57*#`U%k!80J71ARFpl&{^(lL&
zj3)Lq`e4CqvB+WTR=!_mrXG*tr8VO_%evPHW5Hg=tbLylib;~#OiJ)GU?GsxdC@}t
z^HEwE*Lj}=!!bdaEJW%S6)eePEH{e?pn$f+DDM8}8v2Bxc3Md<exdg{?KUZr{mu`_
z(hWoW)aGzv^m1MHM;flkWCqa?b&TAoSnv0$nA|KTY7a3ONY&h}L^fJmDGmm{9;7k{
z#GvhWRUPN2nfR~m2HsfWe%5c`a8FYZ1SoYNuuBMvJ$L{qz+B0GX0W<GHa1=+2Y_=e
z6}F-P9;)fKOiKb8qlXRg{+$w+fHI~&-<G)+Dg4k(uGx|rhqe0LYGm`KuSs^AgSFwN
zPMab)SL(xD175AH%BDkutJG&aa1dRx#6Bqk=_WV72i{j0oBz`j&-*O!6}9F555Th~
z$e=hMkhd8KG*&+5VR}NJr+wDaDjUDmCQEyF(;NW*1W4aL^sR@tIvx~G-!|_3Am7)#
zJBH!z>vak%jP>6Iz1Hz4rp0scg^&Ysu)3tWIonqgBj`OKULfmyLyP?FV_#GuU`8=2
zHFw%t8%jc4?fk0z=yKN#hX=BH9<+Z4Eu=cOx(n*`X+iBR+;hOW3=C@<+tY}oirbF1
z)#>;|vjLVTLhr*GnuEhbSwunbv53d1WbNXUuN39I+?L`Np-hGbAdZ^?JRo0k-!y6K
z=dCbx1E}fZ#+cMU>)@FQ=)&;c&IBU-r?@Ad_YsP41jh++v9pUdS0n&5f3tHpe!PRp
zME6>W#RDTPo8)%AQrV`JjhUWnlx0t_h%0NnVKn&RT5)}~tdrVH5HkLxLII{py72s>
z?Qf%=lXI2Ue3S+HWSq#9#-Renn1P62&L}vDi+(9>`sW+akV)4P=r;-ew5<$WMeohs
zt3}x(S4C3l9D{C&=u?=o3sKG=xT`LkEptBja3)B3Re?i@LZrlIl+hsy=>1KGS1~p>
zEs9dU0qPk*NGX5BXkCqFZ8|mbR`d=w*FEKbfDpL|trB$H7u(+Omt}-Y>>LwMOdVbT
z_qII$&!CHY#Ol#s^qDMjW^NFzk$uqPF7n5?k)A>Pae63A<_q}egD~bvU;Jf;C|H-o
z`$Z1g$eQcx-M<V_njVRES1N#_PPg|uEfHi}X0q<MN6Ry$!85*}enNg__xvbL4q%2d
zh6P(z-8ZKxz>~~6;;XnJuPU>V?cZH{^;qywQMI71->}9vG=Tqkp8f}AfRU4f{eRj=
ztx=bb+iFMb{Zw<d#nR85EW?97s>j}pIzl4g1_WC3(sy&xjC?+8E^hGh^NFl3)Vx@6
z$0a6DP-#$~jZ;0!It;b*^%%74<>&8t-}fP8BQjFL==opqg3d*cff+=bJh*K5NG5B8
zcJh8%t=4?jpsP$CmoKV%F4u#9uyVhaHa-^AddR=`d;VkvqE^W83ior!8&L0t_iYu^
zOb2)XPx;|qDLf0ew6WgJdpH8r!V?jW)aw%;G7Q?KuIaGDwJ)v7t9lI`1CNU=8F|^8
zeKEIsx#3wa!=-@}x)0#BMxKuAX?<<y;0N!*iRbk|>6&vUX;_sqg)qVAsBz+4eVMyn
z5l<;6_*;M0I`ID3jQIB9miQFl@_vo9v|^8i6AUxg`XnpG2&<|Mf`(%GZS>y0q~LA=
zM)0W#V~V0WX`~o$p^#XeXXaaG@}$7ITzAo?cfol(W5Io{>*8p%a>J?`S#i0UDi_DX
zVGy;|y1bcDT>DGMd@Hdo-zF4&Iy!pMx1y&u5veg7i1st(6xJ6vs>ze$ckFLtj^6Ma
zJsVoB!nM5SN_FSN$YoaMz&#k>=QC7w^@5WPRMEqf>mjc9Hkb7yUiMz`y^g_(wSx(+
zBT>Q>E-@E)jPy-VJkh<)lNiN;TSdYHGj2t*);n&mt)go?^|amWg*O&YgO3F`wO|`j
zU_gQPK#(R~(c}Puh`7VPBS5QBsv4vta7tNh0-sPaaL2gw#hp5ZdnABUANrf0z8oFp
zOgu|-2)rRwGl+dypJvsl$WXEtgsJz#&m{eg6V_=V0jQij?FVE2<SKWMGiEl4G2jZE
z_^hq%nuUPc8_L@Nt`xob9$uDpuE)N!x<nx33eGtadLQ!`@+0J|Y&W7S$Sgt*k;c$h
zC_h@i38UM1TH)7ZL}UcDujslTf7FKN$DQT)w{!517A+=G8na<7I%7;Y|Mcq0uK&_@
zLww4#v5h2MFI?Y%U}}2AViJnIhWqwy1-r?)XN*T#h%O)Rxm+9Dn-jwDSL|Rl4vor0
zb7JiM`eU7oRgd3mA{JZA8t}=VA5(?^+C!LS@&jdrXA>Z{5u%1!*b%uIFrGJ{DXMsL
zT_>2yH*H%0EjWaqd9lxWw<T->Ib<?kITrZ+OEc)W0ETcm#$F3|QdB!3yKzp&6r_sH
z^e5Y^UUn0COUS(+6!S>yIRud2Y<_{fIKlqhA6kVjn4C$bQer>JJSM?ZQ(i=m2%FU`
ziCI^M{M#!g!oVXzEOM}IjOb4wxFrIvWNz1E<puzy#tVNqwOWTj=+SPj5FhVKI|_i^
zgWP2Z4JP%EI$A+*s!|_O7Sjct3DdZ7Z7OF`FRroa>L>nC76@lu;E&U^xx!PiGRjvw
z9(l#MEsUQLV;n(aIE$YU7U&vJoR37YWU^oN=d4U3%Zqa+ffuAVvwz);bQA>PO*Ji(
z0^*$+;JN4Wh96~DZCtDGa-{}XKzePk!iZy@Vmdqauu{VC(*8w?#G~ovjKp91EucVo
zqL4r~3dKL83kLJux*_@FczT>mk0AcyAYYk{+5jYw1-Hv7)E6@nzumPkFl9V&2(keH
zp!cdv(HSzqxByieH`5dk-h^jNAj)7N)6r+!)Iy$)T;GG4CM403S;d_O2f{UPd?<eb
z^;>}Y+!Tje-u#5V?WlDX=e#dK3*W7J(xt=&ApMyjQF|mR(7dSBpwRcZLmK#|E{~CY
zJ<p`$hm<PW4|9n+EcYD)0`s6osK>iy5F6Ycu+LbGb-K_u;gEer&YxQ%QS28_JJs0~
zgvRFfgvwB`w29G0tF=)Y9H|WP9i&jpyDz0kP&FJ4Fd-wfgh`baZzjjx9@p+9^dLv^
zmZyqmNSP<JFOmPj{fXvMRB>6ajJJeA{a=K7PGSV=EvPIF**AwneOJ34rL}q2`gJ$;
zkD{H|W=knVQkRn3Y>yEIjI$*v70m0KxHk>5-A%x<V|JDO*3)#hap^fm6VbuohORR$
zcMn;*IomTeJ=G!9L{7Xq?R^eJfgUfXA(PylUUUnD1lyc+VPSyNILlsUeHx}KT;muw
z#`>GEHRA0=7yKE=Js2@YeWUT75x3%C2O5f{Wfca2C4qj{;pgA;Z*p^a2nw$i3|Dav
zdUU+t5H<vSYKkJX%x2fXyl5)Zjg}G<z_8*`zf~1gt0c+8N1<8gmL%}E8Rn*|`-yc>
z#7K>_V5K+4K?=5o(S=X9={sFPjgP*wX5=U+u*M#cspaTNe>|I>8189(nzKL5J9fnX
z3h(-g|8@`c0o6W2kUXgm_;`ha4*x+3k@vbq`7m*~AI_l{(MlljrVB62Xc^BA^TbYb
zfKynAej^g*hVNGn0D*DuX0@m5V<5wNLqOybyR&b#@w=EA4DRxVS}G#0j)l(adMv}|
z{R0%c{()RFR{J4>cKy%ApUzX}pr-cO5)_`V<s~2kosQ&ubsf7??29W;UcKdvsJ9f^
z!Kwcz6dM?f`&6?qTbHN*1C79mB+(o?r-+v4x`tN&+Cx<&sWIAYv8ezY8&B~qWX3-Q
z`otKG`#oB(;udbw)W*!O$qL?G-s*`Cig--=0~*3%NIYqs8vLKJX$HnPUE-e-!<D1%
zA@F^DI{Eu+c==x)MRvCT^4oCyzma+W|HIX7P3fe6p5T9?;B2fSy52o#tp=yNA;-}Y
zzv3CE+v-z5G8@rIaw&2t`gFJKo&ZuIpXG2Ur7~MWieT<+bXAW%+{}%c=kqfP-%i*2
zHB_H>Q9_!^jJ=JS%q&nkxFdbL+7zFO3_(=QiO}KY<9jmi_kHoQiW8N?>-odZZnx+A
z8U6L?AuWwx$0GCl=O34`N}m1q?FQCtzgU}qhNRl!^hhRMkiH(>C^TUO%l1B69Z(_`
zVznuY7Dg8re)(pGM!3;cr}-q-)TV`2kVZ`X<<iIFj=xr#{icBiPA$@G^`?Q7Zq-&~
z-&cKQ^SZ&MF2P!ByXL!j*wJGPk6rYV+!Ei?t%>jEPZ)Kg+3`d;oFm+VVSr=1rp2YG
zGTu#8Yo%L^vaHW#tuBYPsGFPWr5`0<ny<4~*QL(-2tUtK*z8Z;MekO}=73J>Wv6E9
zy`yi`je%C3fYu6KCWvc)6K8p*<;B%QE$!<KsG$q=A3A<R88_cd)sE^+1G8b<<5iy4
zdK~Nhr3Vm$DJ~sI637gPVFZ(o#nnmo6S)^1W!CFtpuA=svsT@q9~1g2b^+vbhx%v_
zJ7ZJ2erZZKL;B0oXR415SKocz?i%<_!zPbzqG10|QC&HViN~#~Y>pJa;wKg~$G?>t
zh*g-UD3T*v6Sl<JCNa+l?G03WJdJQ)+8#~_L_%I~6%l*kW9}DfH=t5y(fUJ<^(h{}
zMcS2g&cB)U4g`#fKzPnP^bsnOj-koL_wl#a@%Dw<z#!r7DEycuoCD6qgvd|Yfe~=}
z7&$4^y*o8H!Ty@oJOQnqyl}ZjKR#P&x_<D(C(PY0xRARV!2!PE;N_w%-rBZT3YS2i
zhla|uJ>s~LxYd5<v_#r*2c{Y{U-B~{nLZuXgq>cmkUkD~@tgtWfy?yViFgsR<*NVa
z?*1M#340^MNSGmy!N2w)Ap7^pw`Dlti{3{9Fho@8s9Bq~V|Lpt)-13Eh1f>=7KWwz
zOLjp5P=@lGgz;J=;Kziu9oR2v&B|^YMe0O=+~F@!7|UoWK|-0=Q5SPVnp?PSQ4jWi
zBBGUs%45lY5>*Ui>)`_J<ygURA+ob|306)wj^BG2ZGF)iAXd#nz>f04xQNW8pr+$7
z;(?1v(Tx*u?(e7g(5I9Q#+t<<-6dW~3?v|R96`$EuEqURpEOt{StR&Lkdlaz799eO
zcf35ZoDCu{4ukl#`(lwzQJ^?dT229?C7@YTy3a)(z(iZ)8e$8r+gUg)>?^MxK5Ub7
zlHPN`&ULV<sYJrdgxS%IAMBsIg}eCUP`ApbYEuuTA)}c6-`rn(7B2nk8~e@^34vlq
zyrjJ)v8ctRo&+B1mPq`e3iT|WQ8dOpV3A4)u*|z9^ZWx9;{`G%4i(`q7Z&50xr&2{
zO%KStomd;ZIInWa$cH6h;eIc^W(W?63dCj%*41u+SmQYKcg%V$&Ws_d5-LmcnHfj~
z$dRqFxC$N!FK{SC;Oy)pNvBw2asU7o)<o=NjLLXiauPC7g0p+g$yTz?ze8}3^J)Cc
z8#2W3Al8SC18{uni6e6kRq$kCzVvgy)jb!6YeM!HFtC=ImmAlHl9zd}Yrt2dWIK9p
zBr=~7xf=)J-QCOCvciX-sjmR9CWS$hx1w+glLp9k6^ESkysvyj)ihR2Ovjo~A>2h9
zg<CjUbsN2dqqR-8@qI%-EBV;6#DDW>0k_9^1k~RqO%R);{L2H}I1mH0f#P0>unhMj
z_6p5M#;%;UdO1)w7tot)f7ag`8!T6g81YO4mO<nMDni3=A7}C)`2m9-cZid0`N*M&
zE7z>$NDG(9OD&Sg<O3{RAxnV`lY=eFXTo^?N?FSg1kTz`jS-4D_MuGZXXXVO5-1|D
z<SGRs7Yo4x05>B?XQ{_hIZ!#YXD4v9Y^8fCgQVy-7Y-x&t8Xhh^mXB+g2`Fy0=U8~
zLGK*KZ(c$Bkr;rUF|*QeVZ6mG-un&D7#g2)Kt&%MQ-FzW?n_WRBBu3!OEN$uHTNYx
zb@|3V$c3Dc3&kx28-4(#^*{Vh3@V?@j;Gg%MzV-`aRC)NFb_QHRXPrJ77pF-Dk>?g
zSDHh+xoP2hTiT;<b~xK@bynTVvurExVqK1o6%eC!uewSP=Ec}xz$L|aX{mf@+i{yg
zdMVx!2+?7~aB%fWS~D;&GQbJr1a<W7CgZ2mPy-<D!S#v8UepOB^tNNW5zB_8(ve~$
zZ))p2Jd|A-!fpE7OIRCS<@%$~i}lvJ;{O5TSNn+YL`!>?i8m!exOHr9g)G3>WB-;9
z=7P1PaVA(1kq{1<@v7@~sirSc+`{%@PtYE+=MqGQwjk*G1~`y;>*<J(Hslcigg1;7
zp3>hX_BI;S1^}&X{giECQAtHO(e&oI(giN7@%OWxMROE^8p4cOHCM(8V*)aU%+l$K
zfTO#OQs@&b`m(>-l>2#6xSmZ-$8awux!+7nEfxV}YkDJ<e=wL1QHw5VA*C=o&W#$;
zVSnD9Y2Ds7qOT$yoytp#MvTjNi6dXH01jz)%BrnK85+hwF?#+t<E4Bizu@454&_~h
zd<VfFl>jQpP_~Sa#rT_KOGGzLY1&^Om%d*?_optiUqDg}G(v+X?$}z{E56l&qbvwX
zt>T3q7-%(UJ#0y2sa5R`QUKP;{w^n1-bd;xj|kM57{Mgb_N9=;;<uk+_SagTSX!~T
z0mh9LaB?emM|V$B#3*06H^|<p#KTnH)ZTTgCSStFr$&a0D5&Y5)GW&8T?4`oF(k_T
z@o3=mXeXUsqRRc)$U#HRDs?^J!>Megbu$x0nlPGDQ5tbFc?er7hJk8LHYt(}l-n+Q
z%nF_+04h?>a-L4_Bl^it@mf0~ueS)p$M!R~_520|L)M=-(I%Z@xH1e)h}Sp&x(VL?
zXhAVYOq6|!y-tXtV_dBMQW7ww%Yj+GUNk0LNCBBm45a`&a(G$@k~XE$;`ylJRi&xx
zA-IIbBL(a~m!M6OsLM@GKtVGQ;|h9;^tCF+PjH6a0&mTK%dp1e%z7^bHFqk`B_V<D
z;XufWLD?XAZ=X?>N>Azq97RwatY<zo5RPU@23F=n^<tp`rzar}i-ZRubEUktpiC%z
zdNw`_92ssNAytx^3KaB#_iOGH<0&iZb}LkQN4*TuGkntX7I|X02y)f4etWTWi@jx9
z$#%NKWaKig;)?Hkf%uSC3%`+F?5GkEW{mZjZPhEMNxgg%PC1>EgIX~a#y)ryd3w<H
zGZ7q23ScBhA|{fMAtR4ax0QgN8RjH}Y!k$c&KgL^`GY6bZsBi0LLvm*C)0!s0|4#8
zi>$oNjSD9qXscJ9iQ^u?v}+_FIEQ%n=zOzjegb(WO;0Tz?IN68k~Mx_8m>=glkM`|
zIKl+QD}u5Gj|d9i3;O3q^Yu$1130S%|C|omEm@^{ix^}%$8Qr671#HrNN4g4{B>Kb
zCCYSTWq;!)s?g9#zTtzp)O(L*!h1_3)O2IydE0vo9GAljX6c?hg?DK*#tyDF%Inu?
z`Sl&o`*((FxMwp5?72*l>a)8FpY8a9ZXQyGg@S~{?WSpX!4I>qlTFjD8SRPqDrk^0
zw+4#87rDD`Q3KvXCuByO)OG{sb+~|P;{p8DDX)+#7THC((=Ci~%a_taY0@bM;t_WB
z#%VlcbopV@(I#4thV2G=?B}!m=HugJ_%9%t3gdtF*x3HV*u}*3Ki}SO(U^AoC#U~@
z*6kI5cUlp@?V)e9fxBf{FRsdNu6dnFzW^vEPoYOHMJ6M={l3F703#WeuVwXWi`X+@
zz=S^DV~2}xhKuv-ejVQ&3|EPWW}eHK&To?`X(W?j9!g?u*A%%q6gE!k-TeF<r0j9&
zBWE^Mj|qeye(c+^|8{!4zdUG`^8BIf>Xu{1`+Yu!ud~zpeS}T4+u8a$RGNI_Z>zJb
z8H1LSnqt|LK^Hhpu%8GIj1Uob<L;7Og(sj%aM(V4T*H@L<T;4y7hiAr*4I4P!r62z
z?^@}ta9_lY@8++~>>RqcXR!Hu3m4Q7t(KZTgfIJ8VB2;>8zb8#wxg=UmkKMx-~~44
zu#j9<?O(R|B>Sv@`>()OeEdMo#%T@a*SUGEb>ZRBhafP@4l>#+3=n$Z_*U~esO~$h
z)(D<trXpx9+HBs`Gf$)XP$h)sNtw@4N7R<+OQF5>Ped@AS<vB_4H(0D2a<9LwxX>U
zD<tg6pbDw?W}EaO)c2f55}%Hz1(Py7@iAGRn@lK}0*y$E@4WJm1|*sVh%CkDQLyt1
z&yZ%LCM%D%=Dtj7D4IJvl|IgU?b1Rf${fF2S1DscDr<Vh>Y>fhYv_hkC>&pGWr(SU
z9y6DGWSB|i@hIPS&9ZzdC0fk!D}eE(s&n`p9HfdEG;2$P&TE@L4C9>IwwBsbLJ=6-
z*YhuJ2YhYf(1{{04tdQSlj8##$;4b*`_DDn3OhC3>ArypnbOD}Rp{!y(CaFMe$`_T
z69L9j=yRhJ>&H8Ped$3UkPggpgN|*T5JQ6}MbgAAiaO-udq)dot2a)%BrqHJ$REjE
zJVD~C?oQFoj^SQdEXv%R*$~fciuu`Qm2GJ;%pmI_D5VB82UEwBHc)J>Ar<zEiu=h$
zVvLqnd;(r3#7_I(mgi93CY&3&S+DrAssI$Hs=&3-dR%UJ{)=m<cH_dk&~XVVR#nY3
z(nI|2g)*L^$_PBD6%CO3q+n9XnSuhpwA5>{!ndU5Q@lPD=vUuy^AS~67aUY*>D({J
z3peCb;oVp$^yjHB2^FWyVD44DgezJ5h-Fn8@nPKqPz}YhgELiwMLj_*e|!z)2#h8=
z#O<}Qs@MJ2lcDuz(EJ;|_M%@gVM^*Lig<}NHw@fRy-$pHv3&o*9A&zvg8(7tB1HVM
z_3g++r9J=(f_r2oNcyrWDnG5k$9eHepz_$%qFgK!1K5Ji>j|}=Y+)jd8yKMw=ue?|
zpSPzrGVU0FuFxEWDzDb=Yj6s}=lU}3W{d7N<Nh=4hKJ@R2XH#$mTGHu?3PPTc2f*y
zHA7zZM8e+V4ho2J<U+}MM4DBPTIA@)t>OiA=f9f*@TI$MxcR54ofpo<sGs_eK71WI
z&2T%6kpEzGDT4cPVfmd01Q>5{FD@!ZPspSN>5w{}Fi0CnkxRXy60j~%GQ3@ImO(R7
zLFsKAH~NO_B3<?pU5f@P4PBi#8N5mw7J{7|uSC6at%8QT5>+VTNe7R^`7{=itDj4A
zPah!W$044QWC*=G7P)>~P`4xxeCa<=Vi3(UpeWK}Vq{3n%@R3AntDNRhj$w4J-%^d
zaL*<n#q0LH0FRu3wBvVh$=;{2em>}EDCI=4fLjcPLOy9FjAXSqozmdtb-6Su_{_Mw
zl!;>{im{-*sKo1a(-wkYz6Xy@*2>Yi@?rB51>n#GsxdHV4=qSjRAe4Ha038~iZ>jG
z+;I*I^|51etWfQYq=cEIF&wl;<}ta#@cYB-<Q}v@3QRjEzY`bf(`vRY8MzS?V(xJl
zbB9U{4wKFwtiD1SppxN4F3LlVm*$E_as?gtcmHzX!1)Ifc*QU&0yXQJI?510faKN<
zEFqvfwg6iFj=Hv2iBc9oV(rocBj|4_ABL)tt`Atb>Po;xKnD3*IQoDp;7vGVcs|1R
z;#=WJGaAKYyW1FLn(@0>k~lA!7XgS{@kWd~aV7|ZiJsEBP1=R8;b5Jk_dlqzrw&G7
z!I_wQ`7xm4Sx}5fy2Btq50ZVS)ju?XuE_sD$O=F&xOfr{>*xYk7GYsdfLpK_lxbvi
z<N-*)y>rj8s(S@u2%9$on8Azj^R8aXa3ICBk2HP}K$cA8<{PfptfhmDydDTX|3FNx
zop%~k3pjHDvMi3edB_jRc~fPc;Nhqh%Vj>QaD2n>IlbDc4Uz$3p^|d$zayC<X8Kg8
zg;<tiYHcr93jOD)6S7@U`4vC_B^7MVOZ3hzXgku2$>NuIZ&(yiGII0Mh}>1?x0tE0
zUHAkwRgE|IgFrqkzmk9`R7N2^agApFy@RUi$K1HwXD<r!vE21KWQ}jnrfk>j=l9FH
zL@R9S+e(I^dx|eWcBfPNl(nZ28u_3U^C{p+FcV8A%X+kwv;{U~UGjevB@yna0Il^k
zi)*P!*2=B%68;Er%oTPHpKtazzocgy4Ky|j=%)*9HRiTvkYwcMN=pd~rE0iFhW7Cy
zo!3+VqSt&}-0clnoJCFP+aJ9wEb0m{i8RWT85Y~QUOTT%tyd@p7CPQ++&Vv8;b>C8
z+U6<2WrJwbx)uxC*(3-rru-vLY^Y6-j8<+gx6I}I4YJI<IfTi68riuwv|KXM<KXzs
zJdOqGD_JO;W_$1L0e{rg1Nt#?x0`XMTV%=Phu=KgTH3__>$p7G6{SZnW&&8e&J7DW
zt}b?bGmr@A0_25AIZYspUuo`ie_N-T@Pkla4seFV01;Goik73Wv%rsz)OiSmq*Z!W
z7o}ZKLwa<g)0FqUt`9xlOcC;97U=jlAs}yuzf-OctCHh-0UlLXF=kWrRvY8zN^$AY
zI1gP#_ZyDVc9>HVI8c<^`={cR4BR=TEYmu_8#eT?;Ja`hwcVSGXv)ee8%gy&R%L%r
z^Gz8-<v5rL7hdr2%iga$d*<!*m)yKtd?ptG?HAaMvtjAKCXMX><zZxH{2wR3ENj@<
z{oCC5`3E$QW72ObCXqVmg!_LOd#5JRny5>+Y}?vp+q-Prwr$(CZQHhO+qUcM?sG0q
zM86UJ<qud_nRCs|ImR>2T8hr*j5RL-u^05@h)2+>&_z=cj0ayfU615rh{<dYh^Ie>
ziDHko9pP_fP^7Edg{#>LZA}Z&L!+Q15%#%yShyOsb`Nr7%SNl!4RAMU$>1<$+hwl<
z|53`{9y<f?m-d%_e)c+B@ppclg1T2lYj>lA4S#?3uS4(0uR_s7b2gPP>;>B4S({~M
zVV8;!S?I|T{Ri7wyi^=7Tk>B&YdF)QXGCG96e87YB*_iY9y#{8n-08+j%b~P5j1Gw
zAzy=)#V@U%CVR*|m6-H{`yy^BRs2o>)=cp32LaOGMehxyF_r%)o*<;UXCwCYY0ce0
zyzSFw+=%xbL<)kVX=mnp3yUj>{(T9p!^sU2X!3l1&Atx6#19%sbF}x?So&*8@Zn*r
ziX+0l$u~J6slGj!tU5v)EIpDNm%p8J)D0#cTGosC?<RL$nEFaE1RXv)!1F7wEUYSc
zJSs}&s_I$g&rMm+b5w!tg=Vmh@O%_Mkgr~y1J~>-$xi$V2@YmFx4*MEBuG+dl_xTJ
z=t)qN3%dbqz_RB@@3ulD*URjTwT2|CQm!hGdfbq*>{05gg#cz-`S-%PmOI_UoA9us
zs)hBNzuh6A>UjDm-;Js2J8%p(Y<AETsz6lVF?;6mjAwcDbuSKii)0WV;x-;ib)MVh
z*u$l_=dJ>sMOO1w3UXpn$b>7~5%&P(`%eW6OQ}CzNf-aw??EAXCHDl6f(-u9=n-CJ
z^L8~8JfGg!F@c@4$}7F;@mIUvV97n#AG#lWC9bELFv6T(>Jn+6rPjik%)=w21$TbF
zSC!wx&1V$czkG_alZigrGsc9KWhsywJhlY|GM7p>JV4FQX1TU1P|ZQ_x-*vb<%?Ix
zhRGX^56HIhJLoA9QO7$a)7Jpjx1eCEe7PmgmaF{vs~4M;^}2t}I+fZCAu?JcPmwO*
zc4#fAVp31IU!d@NDHB-RpG1|Xq<LsM-<y`+izb7LXN?0T186|qMFB}!E6$3FIS8#z
zk7uI1$Ff)UQeqJ=n$Rt?-U8I8pu-La_2AG<p*eCo{1>O~m#>f7ZgNeGyM|1Ncoi;C
zuEmtnN@ZlXyTU4%Ij4zrIN=ToK?RpxLcN(SZf)AiuRitw@&}Dmq1j#Q=XR{)z=J$~
zTKG|^JFTH&L6Tlo4blL9^V>>NLt}=uVj*st#unW2Z47Zg8K)VoXeOxpRjSuiVT1ov
z*D8eq_A7f3PA4&4tEU^%4Tz$mic>{Qa@$%IiYCJ)(M~Y$OoxK&@JC<oYwz?4?TUul
zz13XPbim;XhC^(g=k3^s!E;oYir8ye_$D-^9g;X}XOJoIPxC9XaqVt&*mfDxb5zI&
z`5=XHi4dkKG9g0-V6b>6F=0nRYv!;~j7S<RC1Y`1=7k&u$uX6wQf&pH207C-R<1Dj
zMFMJnq~g5QQLakF*yUSR72CsNM>Y^^uPS>$Ve6uiQ_SA2j{q4?kzz}{v#{^;EjuD^
z>RTFSpHxyuv+*aP(Uh#bHu-tQG{7diCkjt|fP@xWF)d+aaOH#EsaQ13H;2*Ic+0RY
zu0CU>6DK(XfCqq(svLF5IHFQbRGe-;2vFwXfeRZm>E<b!52S#>7gij7mA^s<VSvw-
zp^sG!)tJatT7ZVB2ZJBKSW%~GABYjLAkTG3T@LH5V_I2Wf;-oK-vJ-UC;pOx{ME0W
z)G~mEoSPm)A>L-#BiklFm+hs-)<(`VFAGkQfpyE7Z~<{73xQ!sjEB%KY#tNa?2e6c
zdA=PmGNrkrnSn$#N!+X49|;)Sc+Z(rE!k#?v+JaOBZQ>5td=AD-Zl}~+dS?T#x4*L
zlmyVkG>u3Hc*G<=b2VcVuFw^3&dTHSUt*fV3wrr1WLu+WK(#WUyV{_H&VHu%q6}>w
za1;X|K^(+RsOuW&5LossZwpWATOMA-Nk};<Gnk>=k@j7F=BXC{jWfvsZ=Cm4Tyg{c
z&ws~%y8ZrX0sr^<9}J|_B^{4i5dW$E?$9bEfDgry!0}lv*X1c&3@shRRx~Lh2GjM2
zMz7+Qs=h84`GCEm$SIB&R1UJjL<{H7Mng>k!8-lGcD29kz3-P>6+r%lSc>+0VE?<t
zF%l_N=ygN#1St@w;;;JtTxg0s2Su4n6c7WErd)jI?tD9Ye?J~aP+>s{)2dPy?6a-j
z+S=Iqd;x~;e&5~DrDlZs_I9_kP}7wmA4ma-kVoqWfK5pZA`KM5t)N~#O8-C?chO$p
z-X+<uT{|v{4fIe=6nGBJTmMUwn_sF~#fd+OLmNdK&7kqYf2gvwqlc80*41gS3+!&9
z&Rn!gW#=w=E-AO`xvA{@GJ^whh*?k&MtajQcXiLkS0sZp%d^ux0*4iiv#t9qm%Xex
zi|88u(Ox%twGlrI*^)r%dSlr|paLS{#7rs?8&V@kOe3d|PjWRjM}u19x|Wg*5sv(0
z+;d$Q{Aa<Mbl{MLyk+FYI(0=)rQfJc<lflLY=f2;O{JGl;kf({E2U_uKBrh;LwsF|
zGtfgA^6)`>&`1FO4evomy5Q7|2uFB<S<g~#(PO}IE_@!oQYD60AOoV#G;B$jgd2gO
za2uaasdxOf{Aj7+ymNb_`l72%)7jQni)irzz2$|p{uNA&`?1ybfj9{J;7}`xSZyF`
zEW{~{;ps*>WzO925Hm(lz<gHzhG%dEWIA#ENxy}}D_hZF6nGR;yx$)HeSx+W9yQ<m
z2wKO)JXg^t_xM;M0jV{5S>h5QhCJ{`rC*H=NCP4GwR-g}3Q`TSoTEz;(vzh>K-+Ze
zF~k-}_RU^Kc$G~2exPqEQ&~k(0j-E5#7Hcz*8?(9$dm!gK{pp1u#-%%xz?k(>ozBN
zb-0R*VPUL&W@iz?Bmte~d(4h;wctLi^u@)%zhY_nRLOrnM++t>!u9tpB;I)U>=2})
zqLq?pKDCCxR6<Qz4+Fkg^M}9gltyh40q)Tu01ovY0Wj3OG*Cdi$W3fbf!{k|b29eI
z=N`i6OfgWPzbhbp@bi7~mN32+aYZYqgaZFGIFe;jIG8jU3)iq43k|GE@XbZ)3)VLg
znhHgNtc~z^Hz}d7UPj8I|1215B!JLDj5FscpcpL=hsH9m^%ad9g1B9cC#Z<Pm9R5U
z?!6q4O{vTp@KMFdfF)S0Ep3<W#)4ryPa4#jTJcCsa%B7I1FB(-z6{wx<fz3-g1DZd
zD$p=!^RogWbV=-KxE3v&HKh@gPE{A2zZ<hM^@xLq==qX55qI!kS>J*BxuRg$t$q|}
z7%)%JKokw+%vX}-28>FX@f+~vFMA2LK$uBQ-@#D#@vGErMPM_!;^w@>Wi%2`q1z{p
zf;<RV>hBAl3Db<x6$djxC@W+QH|Na~_!1x8UNL$`TG9SUr>>iUw@t!4e3_Ci+g_~E
z39_`+gPDTlG6P=af&8$h#*F(Zk<NGzYi7Sh<EZtN1ON7#09`ao@pW615?DhHUx%Xt
zhq3Xj(12)-B#^wpdjbD;1&|SbZK=|*O7{<WNn`H8+fe^znRdVbd&bovWcl&Qgw!Qv
zxHg9!V$gU1*TocLKBB(Y$}-JY=i|im#+>D;6@#q+noSOv2BpptXn1m?mRjz|TS9@6
z8bpIZW0ui3eS2@hiMr4sPa{t>8GBD?SYMv5mzea$yWE}>U}_`6EMjbYp6*{CMw3-b
zv+4Y5gJ2Wf0aKUpA{~Zqd(^Fi1Fxiqs7VhC=Gr?+%dxPsv3Ag-a-VZF{)476O-fhK
zk<wy-n9+fL+H&Kpa&w4<z+=!R{{euteJuDry8597N+CH}MK@}JPr1cf*>e>kc%EYN
z*;KauuvE;@!nAT&){U)oUEV=up$ne%i4jX^9i?^sG=fss1ee+26$C6lID1#KWkYWa
z<!e$PaqGy|;2JHRgIXPakb2>P!swZQvboc1!e|GCI=kA2g>BVI=knQgEmtM34qp<|
zfVZ5_Pe2-SfIF2F(eRWkpPB8JECj^2h-%h&62@r;4||c=c|i86)8CIXO#&S`^{)`y
z3NS#Jb2b)`<3|*HH5ZiS4g8o+gLu`GbK*$`2pIC{a-vwyBk$wr#C2|UMnu!pVhHcI
zMjaX~NmxkkZ|Jc)wF-{<==xKOP{=!$sGvs9(2w=A8@cz{w$NHm^i<^D*kJz%FkZLR
zRy@6(apLA_Zlb1F-W*c-!7WlxZnC<HauFb1#h5pHAZ-y5MsP*d`e6&zLE^$#YOURB
zF}TGIuDtgHw@_RO&fsw}GnL1x+$maGL73+f2(;nI^kuW^B6z9!<C(wply?@~WcEBm
z%=_`S)M_cm2Va3Cn0N-<E^a{7h8~lekpoxjPXUxPy=`xX!o&Kl1)d&K5BANSIKiJ?
z&5g98UgsB4;S?P;|G9evo<t%D3hv~Xe2FuxuskyLaPR|hm>5bvxgHEo+IeBz`m3^r
zA+JzTXM?j{V#Y+K>Ju#J+MPKG7M;PKzi`4G<9@kjYHpd#yvdZAG4NDPi(Z#H9GB(6
z*H};12R;26J)X?iX2HsC)MOLFYMy!^CNKLhBS{?{PWlTKB=(NwB+F|^oGFUk96x!=
zD4FM=EfDargE5}7;pV*fg)Id%5hA(BXwAv9tHa36qDlz6lKiZ&tmODkq@9fl^993s
zPGVb7F(P-|Gy7bc_SAo<-2H*Oc*KwvN6=y{Z4IJ81@Hx>ei1`e{)}XbSFmrzJiPm4
z4psS{Du*6xR(}9L^y^OsmJq($pAC_m@w<;w?p(8o-b7a!QUfCH4J#<<_LF*K>N!G{
z^i1&F{Jy9?z)YY)n34z&O9y^B$gQy5cK1*p^~F@UhZ~33lu$wSqnq}}nSPTE*XZF6
z{z_ya=5S(T2hzh8Weoyqg;yrMsGxXOM+l{8csawRpR-jUQi3K*cFE0ksnNW!SB$21
z9PI_z+sqXfHoJ2kKM?rSvdi9oCD<3ZMBmfxzj`C~|8hidu>Fq}yxkb<F`KM+-n{~`
z8hIve5kwI@#KH@}*8T<1PlV@zR+E-3oLb^L(<hf)gZZ?zibcX3=;5PNb~ZC{WUr92
zJ?;DGzpAcmbUW<v5R*mpob6o1l#)(2Y4)t`n`E|%$@QDNH$Ay>39Z=u%~H29$JZiQ
zIXxDe`S=|NLW>3Kqt6FFJ8Z9EUkA}D;fAz{n0Wo^lm2M3LIp6BdtTOzDaUpOuw+Qa
z#|zg1j8fjRzd*k3oJpOAhrisSdxD$8*Nf$ysO}p*ug(lf%g8jZ!rm2exqe;*aTzZm
z_d^YV&Z#j)FL*Jpn~!^&V8;+ymGRDUNh?z7!dA$@BK}}^wTlh^XozYpq{gk-Brszx
zH;@zXAkBa5l5He4*k6x%kS@%N^Uh<fnS|16BWo<on`uK0u@OUDoE^j~zX4b&0{TA1
zBZ3hm7;SJM5m5jl&I4+sn@C|?tRG_>+vzg3#i{Vx&ha8B1w_6xpOyyTU`2U|#<*<W
zE`iP4)aPdxIyDUyd#_`AsIX)+2aeZM=|OCfXUzG~`_Ym6+v1=hgA7$niThpMT3XHx
zH^2Q)FSxgP<`sN*XuAMd$8PEtUGTh`%gkEnLIME7{;F*>KSOTm3Bec<-6)8q6?mcj
zrDE(jO4J0~Amur?V*1CJiS)Dl+53gyQr`fQ66H(0Zd`w1-RSuRGPNmbi^ArEum_$j
zz3}P<3}-E^aO*(bD%M3vWD1?YBo&guK~rAlNf<+zdF^MIFvz1|52W1vtgijK6O!!7
z|3!N%k>9?Vj0ov8ymUPlj!+W{lObDtAQ5CWY>x`ppCb38Do_dt5Ugqdlrp4;V%}O|
zOGHVukU=prILYtr%Sr<$%Eteid3|Q`HqO9AMD8P2*-of269H`g`#M-wA}IbjU4{S?
zrP>76CBzai34R;8-)f4|)jPKn+fy;d>qgmbr|2csp5f4`H-mwRWD8_JoTQ*=fv%ds
zb?WGlt4&d4PtYoMI1Y9LO9mqE`IqX!eL@!usK~2No;!5AtRoe+{q71uUxL-e92ifS
zbZTWoxRj2_hQD{PeLdI<*MvpQV;~BFp-X=$Rj_Dbq1T#dy|4eF?sT3ScT1h~WcozQ
zc9K@e;AUa8a91DV$nIpw3eE?F5cLa=bw|9sWC;)%cE6U8bT_CFFwYCWL4a7iA9m6i
z6C`9_KtYR6=5MxwmPh&2euhE_J8tnQ|A!t0;naW<)Nq3zcGU+n<|3n^u%91JKNCIA
zq%<`Wh#durj>E!~E+NI>93)fEoYa~Z_kJnW%|Y4`tvU`LSOn>j{Pw&saHMz&6Co;l
zo8bn>08|Im63G^N08?^dD<0BZON+%3z~F5LDRZKVF8M(~oW6s&NL+nS2T=mZEuf~6
zonm@f6M{$Wt<6stW;9805H^u4fi@GUxvlDM*LO6T=h~2^W`FlL2Hz#IX*xfy$~YYx
zuOeaV37)yW-GDc$!-c{+`U^b_5+<7e$T>?^L%7_Oxh0|<@Rm4zmr;8J$S#F-&AiYp
z0J4qWG<8X4Fa`CbBf4`{pN4#(uC{$klsLzGM4Jg|V4WFC<<x|VWlW#aSh9=B@XUl5
zR+`E<P?*gaYzUdw4gJ_@Hw7a8sgzFYF~Z{${mGnnc(UT*Xw}f-3w_B$<-ci71o!s{
zZ_+eqF_c8$1pX!~9=EX{l?GxSLQWtk@sMLw=%T2%nwJE1nQ47qiqjsgU1G6MEI}N3
zN{T?GXxIlO6XsIEOCe%&3pslm(16+dmF>0)19$M3N{kN`?@6nc+JokM`?t%nWr(@-
zR3(O!^yac$ppIf>faF$&Z(<z}63B-@3$&!k-Vt=2Y|BY-kjywDfNj`ROhy|5{KI$f
za_w*lJre_zq?ENx^`JvUikqUrxHD#gGvM@odk>y+rCAgS<6VZa-6_SJ<Kw^Mf3R4C
z?l116z^aV<npoHPZldbPIduXQD0uWf5oXD=uhTpZboNAnzRw%?#&i_-J9n9bUL#;Y
z9TQANe0jHWuk?bq4Qu)Y1SqZNuS`&*_a5-OhF!TnIgsu5IQpV>Z5~DkC|PBI&j$4s
z{<P#yjkLT5v$wYuD>hoLcI<Mp<<LchivnA3@VEKz)2%;(OA~{d6Seo?#Ll9I3>{(}
z+YCA(QUwqB$zm$%oF!iYxR)cMK;8o|(tehNS)E`lM56FD6hbG#=OfZ4KWdTTHdc<P
zm7EB9esx~KT#P;}bfs8Gv=0!3H5FeInckLfn8#@Mn_3g^tpK=fdEpT3(W);7JhtM<
z+Yh~CI(FF=h<*D3pXuTu-<*8PkB6nxg`g))rtqUspd5T3KFIlGTkBmu;}dFAFOg(P
z1TN2)Zz(DA{CO)iLJF>b@>Il{GF{F<om%=v;3dgxgS@WV1m8sBAvBB!b6PLJ_g`;_
zZFHGa5$q!WgOs9vTux-se8NFMiJ)I=j{Q&{oN7d~S0xhRF1P~#lD>Pv`F8{C97UMa
zH$10f5)nqi13@V{JQoXZ-<r`IY7eQI(~-*U7vooYAfdm<UIVccqGKjP98ulsb<9sD
z2+rNRD)zoc*rQKiOSPN#<#c`~>wFx9jO6Ekb~V~j>6wvy0nyKWD*m$-`2Ujtq>OD$
zoy-Ur>Dm9sP^5oZKDPfj`9C?j`E=myeHkPa@d-88fScXc8$jDH3ci!HK*I(|`3*<O
zN7B&;KcBE1g6c{KZ>-n*aIeE&3>;qUS*fN&g`Hud=%%|>Gd*vt;e}bp8W(2H*u$I(
z;U?Vj^A%AwXyOC2yj4DL2b%ql<BW*VDUm2aZOXPip5Kob*T?S)ls#cAT)LFg=#{rS
z2UBdk-%TIJ|NPyiI6XeEt{J}G_V?eN(4KvmWMWL*0rrrO+(>X4V`*X!X3hSBo0@S2
zkHYFOpf4ex26f$5^%O7c&bnTVYn==w{W~pO=<5YEu9k{GOYF*#x3=9Nfk1Eqm3W)d
zni(5O?BsUJs?5fn{pGU4s;F$u?#8u-_K{XVSyU<fmwtnUXyOHDZ9;TXyh@n{W?Wfx
zEt?$<rt88qHTlC6hJiVQV7{o%y9+W_B6@dZKwRjN8*7^m&1&xA-p%XF!*b2iw{ngt
zr|*Ey38PZt;bpc`%UkQ#dG{?0Yn{$v`xKBkW$o6XK4%d$aL}ceAGH`@jsFshW45~W
zs0^xK**DQMt)(yR>B<bWValTKo3AC9aO5f>FQn+lnS@{)?vtE5*Kfd>ge!;mW$DFD
z>RGns((z8SqH$U81_kE>v|}~=ZFv3YyKKW3e66s34>AiXHJhR!6|(E8Wf9ItU|y})
znVed+q_X@(CF#8&9cMOG@TQbj>8r3h{v}@<cY=8et2>4sl!xY}f#pNGRZWevkFz%1
z5jGABZhmhp@_JfTCkLoskJIWE^t0BTM8U9o9&m{A2IhFnnC7Mr&vBuYZzp#Aot&>e
z2nRBrdY~3LQn8TLv8&_kRy*v=Er;M$`@I=a$KF>3hqvs!sIiFPeEK&HFWc)G{0sNx
z9MQa<X5x0FX*SlMNR=Wn*NPxz8DW6$D$JzR%*GC*WoG}B#Ws?eL8$qXBj~{@yNUm`
zn=kHZn9VNunB%CPVPf5N4O8+7VjmPun;fKqBuF2&=wv~nchG9B-3osMY+&JGbA|ZA
zKDDe{#{^CQ2Vf^Acf7Ci9$>On?ciFO&O5yKu9e6eonPK}GQ(U_xxzH94$1MRf+U^-
z5e@gE12y51a2b9}4H`+vLrW&f%d>Al5qZeKq9D^ic>~Pd+G%3_#}>mZt6F~m)#9Rs
z+f*FKv;31RMp+6#L!*zryn#DeLin+zGcxAe+5LV%+g8;Se|BUvtY2Q3m??}pJ!&PB
zi3{Hn^s<qBQZ3=EjA{OQomU4c;SI=?m4NRn03Y?RMDHZQECxk9$-lu#ihHzRA9GmA
zl7<i#ktnw+HGm@C^Sb9W!d7gsG&VPvhB*#{4l3Z~NegbDqX56}<Oq~0HMU;e{XR?@
z<U#MCmKR9NzGo?m&c_w3S8C7w0Om@)0V6L@T-(*ev-#Z*x>v&>QvS+gRVgn&y8CrU
z+>t*3WN2q$T23plofqk#`+9@MUf`@wenr79<pN=zH&D^60e`|W3}B_b`m7ex#Jo5L
z0*}9F8o|Y;*0EU|zW;{G%8MLMsJp<F=bL&yIT6TFVS2X@NWdV?KCHwX;&PUgT{f2+
z!}fJQRw$_OC@Yd)QY>5Gf}I!=%Gkes0WP@?#$eD%)XZ9REpnI6;9Q9V6#-0nR-GVs
z>&iQaYI>zlmWh4DTv2*g)pc6zP^!MCROcFrt0n;8drF%i%daBs<$Zqr^&g>t19+Zn
zztLJOZ()4Ak12;&eJsk{uK(`sG!~8T4M|SDuJkGboHOSsHVLO9(!VY3&t%N$(Op~5
z+Zt!g$pnRN%-#nuut(A~sj@cIIb-&|Z<(JFBK&v>XGkBa<$-7H+v*{gY^&gc_#!t-
zeZi3xRk`Zg`DS!K-d};(!c$#-hnWK-Fgsr$L=+0C2uxxkKY*hi@4g+I$m<eZ^Xdkp
zW%bBl5n>5>Mh=T2sDY3arZ`b*mN_q|BSDgVk;yreFUgU+x*qP=h%CE}5)w*~<Uk7U
zzmJa2(<T%c2as~dS`^~S2y@@;B~FHsXspn#al!-(-R6_?nW4zd3v_&Rv0FSaRz&E1
z9}N`Ws6=s;MUJ3)C6<aeBc%b#mahsf?fuRv1gY%L_NuT=Qw&OSgLvrwVqE#1{=Py2
zMH0LB+2!pi5lZu<LAY`Bha)1>fP+_Gc$n%d_9xMn_2d`o4d4ntC#sT3#o|+6T&OCl
zzFKP@;E%PC_2|Ac?$lV{HP`0=H*e_!`?p%J{aGurt5)|IBi6~SpTbFOvG$*Csnck*
zyFN=)hZj4E36TN?(W%%A+LaOdr|B^~*NZH^x^W9qQL+4@-a}vws-vcw5_K`UY*7=a
z+Jd#bi}_JEjVAm{x*tw9;E@GzowYsb6A_qU?wlhS&7W4)?gsi%z?r@VLwrJ9JPY{p
zC?eg=?oH5s88O{dCyU;aRET50XyDGBA%TUmc&;W@s&qh-SZlUoQ%1^0izhq$c56wt
zP$x!=oYAB1Y?2WO)J+mW!I5C1oB3=4A!&sOH*WLIR$(qZJ(%3NGetBf1g~`VZHr<q
z4hDrBYtfWs{;gX(AGAb0Q?~Nbe1A_Q-AIkU`}FGD$p_3DL!#6-K}Dd)#QMD)Y1vcf
z0|L&xZK`#AIa>yb<`wf{s|L{Z`M`t%B%<G_ofju)lr7QgA-cR{n@i}c!kH7-<?Da+
z+bZhyOqRa``1}!@k0vTLJGxKjU7gFA%&UCedlt}02ClRC)Z$Z*GMVGUXrpFP`owMX
zl^@1tNqF-E-Hufu^`A}0f2kW;|L1f4W_3-+%{J73*!nrmV-*oQ@Ob8Hs|_PIZ@Y~u
zlaIaUD>)<@N6L7Ogot93mfmh~_B^3E#$z{A-5_ZD8g}+KI3dwVL|kv*dmE4Y3*bT*
zqwr;|*b>)2`JxgQ0>xO;=!Kc}b3uudRK@3Swz)q)_aC2=R5fRAAA_D(U7t_pABRtm
zuD?V3qn1{_Svpu8KQ|u_E+4U~_(f+DaS?LG{D5?P5+`Wdb9u-lD|W>q-z^$3IUi5&
z#Y(rO=)GF>2$G^!#31=FSi+F27R<W3ZSMJ~Rk9*e5D8{H6%F&-!@W8c$NeItdO4bJ
z@<8DQ>9f~6^js2sjvDN<{XH0Kc%<PnFdnrttC`9zk=ziAY^$xQsu>ISEn~M67SVyN
zYf9@Ur*!LG6qUJU1!5B97?He?g2DejlSA{TiENz7E7+@+EFZMxN_x+8M{dt)m>p57
zK-^rp_)`(=ZziEdGP-o`WNz;C;W9Yf2aa|vp--^6p^Gw)k@8Siud-I+(1y$Rrmcg6
z?&Wx7^GOuB%bY>yx>x4Ps&7ARcCzuKW6#>MtxnZAYM|n5a#SmpfTLQjm@j(ua3qT!
zRtk>Wu`u0v)^0|RLrlKa``?YG)gP3+-=2=xLMli|FlIh@h$61Gmy)(?+yzc&QEXA)
zVUrxz^@gek_l}Qt%9I(xeLwu*?vWC=Kmkk=k@dCfUOe7&WL=%3ogKoRBIS;mjH(K7
zHZ}DDf_5-6AMYQ;-UP^a>hNF{^$yPD+=b;69&dL6kBDn(OPv8_?(t`~qtdsdpPKh_
zU(hJQl(NTTPFVR;5_!t^Yy(Se+J@yk=Ky3ID-@N&$~Y+^mCwvUK$y{6-~G3~bP3fP
zzxTcVab9h_7hF%@A(Pd8#&h;VJ9xO9=3O9Sk2$Ct)13lL;U?$25I|IRV_2lp*EktB
zgFtX^w(QwPX~jdmc+N>J_=iH<%oHO{+eN+rKcXifbSoA2_+lSQOx4t>s74azarVl~
zE*P9eP9lnyOLeriB-b5vJ<|n^e9-{)0v0T}nTx8A1lGPP5u@?6iw*p-Dgv;BW=RjI
za2F5;N<tuMp@%rO*Ss`7ZDF9yk6kM<YDL<?NZW);P<h61F@#K5C?MeSaM&$a=z}CA
zX+!=!TiYEl*72A|;hUqC{v(6=`t6)%Z8^ZzeYSTT1;Y7>tcLAKq2~nqcI%U4&dPB@
zPA9stazcR%n>Z>Cn~J6~?JAQQB$AD@5ZOj$j$e2g+D4+vJ&o#n$%sbhmA$n~0BarU
z23%SsIQ0~MFsRE1o0~;a>dHc42N<@F*Zp~Ok0BZ*!GOY>o$(%@hf<IYKFMx?vOy3-
zJ$;|?5g$Nkfjv+4pybBCEi%W_V7$VOP>Og8N5e>>>YTms-38?XNZf(3NR)v`>36J!
zcnjO%us0_J^mO|}eSiM}jvmMJv~m;XI)s=80L?_5uAZOk2!KCrJ%r-If|KG}ySi&K
z4I0NTIY>)Mp)EIvD7<VcyjrZiMjV(N$5JBu3S6o!)Sd4zWcbEt=6ZC{4nvT{Nh(~%
z+_hToRYUa0IJWu#htjXmrMl*_cT0!42YE_#)X-fJO9nXX<prkS4smA=)^D!3=ic)C
zZH3qMMulfV1d(k#7w0uef(8&DJq+n39I5AXHiQr7)MG!S=5UG_ZT1tc5Iv*$ABaA~
z6?ESI&?MILhSL9?9B5mv18X&&Q3sab1cZz{=Ek%De{PE(lE6-s)?hg~Z^1*M#2BK~
zIzW7<<@;@=$pK)onLQ(nB1wA2Xs`N9jMD|>oaWznka}B~VECJ)*0vJ$H_yp|va6kt
zQGV<hnADDA#-%4HU60eZZ8n1N7y~>CL{)!?iZ)OKyq+nwBDfQdT_}sxE&a$<uoVLw
zgTPES6ag{I-VaiUILaMq4W{r9$X~lppdA-Br6E<BkDMdBeD!6`5C+(ruY{U^;mtW?
z3*b}<Al&8#tb(Jpo#_Keko4RI+NYDBV99B)qHG|>1?p1Y$%XlGgu%a9%ly;?CoQPG
zIpfa=FF-C~6y&(?_O{Kq;8pF~ENSOWLO7M)11^H6AY!-YV4;L9AzI_xng4uP-rKu(
zXlH}@9P6m9rdJ4dHpwRpr*S_l0U<m}d0E7iTDG*XD$d9^x5_OcTsYuo!O~!^i$s25
zp<X)&5eD<XXs*XC7B{kY{ge5^ur#&IQYDqB=5$dy)BQ+gKThZ|<3cz(AOIcgQRgKJ
zN~qwlv`5y(`zLpwNVb=9f^8&^jGd0TS;!|GW;hrhlJhiX8sz}XjI+ba9E0WaaX}Yq
z?p5C*t@9EFWfAA3vE0B>uqefpN(RprTQDQpA=I#^WJC|nr+JK!TQmTSM3m_9B8MYX
z(4JcEuAH*%`mk%pa>ynGWc58iHC?zX!=hA|=>Tg%{2~2ib_aqa7aex?D{Gb;wb_Rk
zA$UQ#5S)YB*T*_$)@4Ksl(MFSE|{<;<7qY1@q;N0wBiHOQWtSyQX06mMfKJQ3c9)^
z*&h?r#n1H2`vB;kL@~24t)>PE=41P;B`<(`#I46DlpIx|k(%aVCxpqjxs?IRKl*e;
zA=9aKk&0sguW+**NgEWsTlB*vV|aCkPz9C!(3yfs>1s3S27opUbJ8Qs*4yVnJX@*G
zA=;`K$Kb~sPRC%l1Q#KV-yCcLSx_T**G9kFuWyh&TRB_?adsTB?uH?<TgXeCy2Z)0
z-$*Mx&Aic=S*k))mcAlB2H^^Sv05jizz`rTocQuQRk4C}X(6cqMiB?e_EQmEAbf<y
zu@K<WaPN~#6PSY%n@`mBH2iDCrR;~)wrQN-$O71BgE^_^?eZ(sMuxIA4-+N{GEQJ5
z=&cPlwhK>Z@(!kN!pvMmF#@%{`hqq#?jv*PT>%^gq9XiH&v>)=ouVKJ45#v4?x}Y*
zMVby6#(%^Qb3j*RHu(W(8Gob%K|mXgXORNjS9aa(I7U+eWk3x*>&=jf26}>%u5)~i
z#FSa!;?_Bx*;k_%m3xAUj_`OvpzRi>XD3?F>2>m4Z#8S4U6U`Z7QE<XAyl&kO>WIj
z3yyGBl7j>?ZV<97mGUQ2<g#4JUrLAl^22qpeoY;*Kym}Bpupl>qu^?petVp5z44GT
z0VK6cy_wh;J%a8ezb-+Z@D{#F3RhPD#5p|nQ%0)`rX|H+3AS@(xsqg=x9xT&RYO??
zn6sVH#Ck*yVwHL$JJH!IM-2Yi^Uyk^0?i{jn(^SWymX+Y==n#j<>?iwO~wz+CIFON
zPkZ-6_7LWgB@*x=E-<{vH9V5fUICx8UTxyB(@14^R?|^=XQcb8#P))d?PTPh5a^uh
zH5w*NDr5hEfYMqSx#V)amrzY6J+zkB6Zh~cY>;Q5Oe_>}>2e{lx}!xn@>?N&GuA<~
z*(Mqc<$JrJ%Az~3fP?<(%s<5IX@#&v{N2ahcwLPA1>O+~Px!Asl;gil>+~#)|5J_7
zGRAn!rpU`p?VK^aRbR!DL<3`5V4_<(Ks~S~ADlQ`(pXBGTv3Q*t7ngo&a$U)d{dE2
z%)D@Ow>O#YwW`<i%dl%E`fk^}g>ibcao5IHtx^{wr-@ePGO6x2mO0I>&+gaTG<N^>
zQMz1!dOQHq@5QHUt<O%^k6UX>h$y9rt0nTI+2ajMj@Rq{;K}02VP&rU=fyX?)6d2A
z=5LqpB}CWW&_Y12b6!&lq@;D!)6FJVyvQ`-{Fm$4MNd@(T<i4TpVv?z^dI|qSmoV%
zP#P)~pgZh8xD0SgD2ft9s0c#SfGWK;pr_m(+bz<OW1T#glQnd%Lf<{GR?$zb(u=O%
znb~rQOz$9vNOF(Kb``7Q6H#KSnSbJb*_WKvZvjOqAu*|A{;`sqK*A>!h0(p;pRmM~
zodlkb2H|9llJ<9HrMDL!ZyQ@I?ADD%2MXle?Nt24MytTx_S}clrOC8L7!tMqq#`Sx
z8BHX}6johCwt#(eqqt?8dWWcL&%3&<z-k&1jAn8OBw@#7w)`t3D%yM?Hzh$atl$L-
zO<@JyL!I(_B4ZwKE|!zmbE(qNamA+;F+Ug`J|et5cWaeUHXnSiR;EUVO8LGn)R+01
z1hBGqZ#nE}$S%!#q~dYA^ZBG5ez@lY?*)s&qkZi7^%!V@nU;4}=~zQe?Tw5ME2}Oe
zf)`uuGfcI;SvP^KT#Odd=$$1$(T?fp&ozCCNL`iQW0W55w}v1yvFoP~>^S#Ll)uy!
z4|B}#=|3L{#UNM6{xGo~QnPsXkZQxI@1H%>VItzTk+Z`#XMv{E*-S&w)S+;PQ|;@J
zGn2o@{Zr;o*lb_-ayM`wtWD1tD2hJG(&|>^F8yqNe{#cwo(x+oTZ&E5T-0y6F4dNJ
z0HTOl`eDhJ+GWMr0+e8(OHVmVU(2Ss5OM8!70J2rzTw!>3wgub1}vb`r4<2u?;J83
z6<N)`3yQD*?yx!gM-IxJ-htq7j{;BCmR4+KcjM9ksO8GoX(`Jd-8nN|!8<MQ{q~&Q
zGZy}B7lHZFauq^pg9tbaCkQvu%EytMm7dM{q-)@n5P9*Xu7pz6jOgtkf7qiM(?T-t
z+@6b>XNXM}--C~F!u*0%P7z8fsCpkg+Z_*jSE~b=I%wAgNt`#Nf4@q}OP8WY-eRk>
zuy+R-*FAW~bFhvs{U*E|tiT{~P2=*?T=A2Pd%wVg$LEW&G9tn^1N?v>lEM|&EXnuF
z2h>6v<rBylIwe*&mr_}igJ7x-3{v9rN%<Fa@a2?mC&PhAYAg}KF(--Ck_%N$xB~7V
zy0!$n&BJQ#+ehMdX;Vf32qIo$j@$A$!9Kn$K@t_ZaqtbU@R0v>14kvLFkEv*&`h5M
z@$eu-WagqkO5=>5JD2MiM5{GAuRT8)1YVQglEl0lzx;bHlo}xg6zM1!<at|r$BT{&
zhsFR(##2JA3|vm(E*<|bg-M-@r0JDFcvW8}_;GFsAi!#q&y}y9W|n2L$N{uNk4UL^
z6p7kt&@n3bgB9F8_O>{Viky@ZbpRORuCCts`}MknI{3r#8mD?+(x)2iR}%r=7;~e(
zB#1t+>L=lC|DI$P{Ipq|TQosCiNN@y5)^@9D2T<eDbIhii_c#yAsDbl)s4YqJQBsC
z+>WEipLx`_kviC9Jiu^fYO;8xIDyYgA4iqRRtyX@(Y)}Kej9aML6^B#WRfvniUt*|
zX0T^(-GMZv=6f<*RtelG6=u?7_bH7T6}8tgrWWQ1SI;Wsq4Zx7QKvpqohVX(GZ8~$
zyr;{;Ux<;SmNcXsTJ*v!W!840#U*=G07+eUI<)&7ZgPxSl>eeaQ(h_V!-QUET3HAw
zUz^eo>yHmNEemisP@*RG4Y9g7DVa2?U(Tf5XY~dR^tTH)kErw<$)jJ>JoQw>Sk)n3
zfR)`IBAwgPHXAq2qA$E=kAAa)14cM`g~7IFx5F*hOt;?)8l<-yFyAqW&Hgi$cm|7x
z0NmZ-ic$)XdZNva7plm|kn#r`vAD~PT(Fk*aE^qtW;)2)Fiq*+Dr{j4<b|8rO?$C9
zS{+JZcFssu!k&OC4}r`APS9{dFdaP-&wRxo%{=oQtL$^#vQa4z7=uj50v8?O&w%hG
zQL~YX!l08hGy;{^)K2PA2)K^TZ(nPr*N!LO?}OMhd1*+LsPf{Mc<K?|&87HT72{0D
z8vdK{85T|l)wEwyT(co=T6+elGm(u6|4qt2U|iQd)!h4IJObJTNUZfiuQLA=eLnKp
zArh(KM}W(O9yN`}YGq{7<2p?L_UoTI{VDIz&>84GIr>GE{?&caz0+!NcrIHtXmy(O
zK-1-O08n8Lr~nE>^XYtWiN8r;rzs#w1amER)sYUm9V>AsAiTi#*ZwPt<V4HKpNHv5
z^5YOB1ENF5CI==9cl<d>^-tTrWf3721udubyj~ymH;$qh^X*?+Z-@~8czy2GgV<S<
zHa<sv*Y2!SlQ&AnvO;wnNCs1u=%hru!}b<^f)kF4gJJ{>7=J<(ayz+8dDy0>Ls+67
zjR(}k5FpkVK&41Rub%?TUp_ntq<qNnAe&sxnU{2<Sb;>a6BVHC%qp@%3dOT>dIh3I
z<2vL?@vOlkJkq)|M(OegMf)CMr)BA5c(MiYkg;DI=Ua7A6Z>U<;Q+@EF}sh{0jB8~
z+<^Jcdm-|!b&bSH(Wf6g9uI!2_0Yxtv#VhHKY62!ER6r-=;>{B$=J;?#GcvOIT8Ty
zy0&klMtILUbW2lUw?G2mBfVSyKhI)O6b{r0OgUd4O3uoPMJh}#4Ll$m^&i^Wv9%SE
z2F;q`%bk^MY^uCn$pf`sH_*yeHa99AYU+wd18d9EF3*IE!Zr_=%``$UC(oU(y<Z<U
zAFY|aUw12K??i@<`p|1kH4Gw$MO6KxRYEu2pbAw++<YnR%s$SJe&fAOnTO5PjdIV<
zorTlIO^w?mwbarz+1fQezW$l-=6Gwoy!Z7s8JEosM1_H^hopL%Cbyj~5X?`C&Q_d6
zgA6m5F{?91wPm+1IA-H@wVoFRwNo25c`^0)l_l;4BZjUxFXoWBh&?c)P;_+(?Q%@-
z5SgiW(OWhhgWv2*g4@k?6-(=K_(1ZgIciv2aW(?mEUvG(&G3gNvP$TtA`Vt;`|L{2
zT-@_l*JG35IwDpF6^7lFn-%-+%YEFr7_2&{Dij87wQ#S}ZAD_%4#%Wa?j3oYnUW0B
zW99*=jgXetj@lK=SSS@;1>wBT*P?_=tWDS{)LK+H_!T9+b|YFVtaL?M=|EB<m<GQr
zVzTJ}Zqg(eFcolI>wIxW!2AygA>OUm{mf41{Q8bIcHiNd?A;O*1NC!(Je7o2i{)1!
zrdFJwRvAcZB(%J-%3Mp*VHGawuA;gt>60y^7xm!1EE+~^pB8yLSnf)A{~$7YBgS2n
z_Add*l+_N!(*c++0wl{4P;&5LD5_^#(j8y-#=_*64#p{+5`;tT#7YzniW|dS#Qlln
zXe;IYkMlzM!Hp>?UP=n|vwAj6mq_i7SiFq;y3eT4Ku1&&B!bS9oB$Mo?6+$ioBIKk
z!0i6Gjg_y5G#s)@i4^MDBgT}izC@H1IU35HBm-PI@;TFOwQRXz7eq#wt{cxS*qheQ
zRfT6Tnuz$3RRO9=Oe)DCEu<>m+_2Q9=GSY9RoNP53Os{>WzwOA<MkV5&a1eyBbC$i
zCDKGELKX0Xk}KuOAOsJRI=Qr?VdoYVW6;*AmVM4m^y|~jZ0e$-w*ztPn^B@iUjb1?
zAdY+NiB1JmT4Zw>n$*nW+&r<)D9RVZ&{Uay*qm5M$FxN3O`G?2nP+7jZzTX=(24+o
zJ3Ty6(oqE83v7KZxL(|P7e`nFP*ARiHn{<-cNpX{1hD&|e{_a-L+lLEQ?{)kG>2G<
z#CzMgL3;()R#4YslHZBc!~Eb4o>nLqlE8v%_&GEx-DMia=y|FTX&cy<ZP<(vXGuZr
zY@ub!s=>94*M3KNnox3VJ8jMg>34SzW#)=Y=6cFw_GmOOa9$F#%$jV-R7=@GpgAPP
zu)QogM^i#7ey-SWcCO80wVP$!2&cven(W*4^Z27hitf^~v>YPWHNXR!CZAhpB`ka7
zpit%QdlF=$#n>oNIpLN~bX)begD%z*-tt<@0TC<ovl39$0$g4~IQ!5wGRV8>#)wul
zpu}LyJsU9Tk3LU0!KR_rD)?)O^v_E!fWeH;IfMmWYc@k`+#%|VXHlb(55&c?9Pq3N
zf9^~mTn6cB%FZ0mTKrGp`gBE&%H~|-2MS4@C)zntAzG!<6^%&gi<r&;@qo{|Gx9o%
zdoBQlIt&qI({xN;vldq}&u$S%C$6lWcrj`=eaiZ{-FfGrjDrxL#bIqyf~e=Eq2bq=
z3pIybxcA?1)FmKVjI(r!g}ez*JWB5ZA^Dzee~kM=vUJu+m|Tw{LXXnT{s_$`sG8MF
z`5UAIIWq<xtw*YVN8W=y;M$X>DYgNAxjIZ(`WuTO;%Fnn|23ATrIVD!4uqU!G?qRi
zMt50kZz7$JPi>j!n`VvI4fhi{A`z+{X3?P$1ns%|N?+=4D^h)mPpYA7X^!*I_j=!M
znT=?aYk%1#pFM~jL!QlrUYZck>($vP9eXw(VM5Xd?m5{=x?`ts&_a70<v6<*h~&d)
zfwAKi^<)2w$M7qQ=y;Z5$=<?vK-(olk;<iKd?G;Sk?A$EdGg@A*Gj&7@HvLA$1QC^
zcMbdxB-cx%t2YXC$yDQjG&1c2MJZ5U8|fY(4kRL9T@13=5O*lJLYDkvTkQO@;(xTe
zj=syud4dmmHa_$9a&ekjIjcJURovQ6sXEb*ELqJOS_$79zKj+!SIzvkY?$TNcSsG@
zt^WOeVXHN)&WUn$w3h;fQ}N`Yzb3it2jr};g5=SqKC?*MV4W4&AlKgS%4U`JP#5b5
zmHMJJ*@Y6<2$&iCp;G+o)(^99T{OXzry15fRvjL3hhWo_yPt_e_mmCO`$Zc9Uyy6h
zqF=q*sD+1Q+J*xH4%dE#4kgJb_1EPOsT|IkO}p!6vlQY5pHiqe)2}=x_>jBDeV^b{
z8)gm~prHF6))zFU!cH*`U!>$4JeXm2scWly4<~bPZGb5B+Vw%t*O*v)^F$8N>8{~s
z=v+>LXV0E8=bhx3E2^F#;WwkMKW3kf{D+b5?>XvB-atRegl!PiqjH%XQOZ&ev`QW}
zVBb%#K`^kOI-q0=dtd}_)jo7N-bcmzE>w0Dfk_O3C5faP^x-Gg&QMHZdFy$7PqioQ
zOewvGE-s+t9>=O9;a5KPTl$v*(qbs7RY)Tg9~)fe4Z;I79qf5u8Xx;S@-J4u_}f+e
z02auRbo~|}F1)i_(PBPRx5=THk+1^|&;}|CoAmPQGieFL8W+sI9q*jlk@B0;DPRc~
z^_7e!>vL&eri^WEBWD|RUkr#UN=K+2H}VD%Lw*2Hgr!PEY69~4=_R2S^t>X?5RWRf
zJCt`e(ui<6zuo>^1iA%4Sc{E1c`DSKT~tdgq(<uK-F5;XZox?11jBKy^5YX7W2}-|
zE@m+IY2!(O)SH5JNI6b2X=xj2f%#Vt5y3s6#bO|fhN*b`lxS!te4h;Ll~8MVVd#aV
zHwOF?UozWs&t*#!m@T+JVj+>q-3YNNMLd7iR)H!Cr9H&#eGPc`D)Xdis9ZDp+Y{Ul
zENge=eVqbLia_FpvDoBW*<Qiltmid(+bXusD>(dK=6qZ;qmn{}Y$p_71hGMPfpWBg
z(LZ!h%^+I~<pC&@BsqWa$`x0lX{cGP&LNXV7H6AsHtsvbvc6gQ@nw3Y>z)d%US{j@
z-Q;DBqcQHxQ^|<V;8)W%1$eHDG{0!_a@|XrFRF+wU_t5*ln}ok!*q~f#mIK&*FQVV
z$I6{cf}X}Bm?r`uSV6j*EPuvBJfFSK+;wS*2q7Vqck#%g(z4wvc#AC-zId~M{8Srz
z5lX^N;BwO;G3pROibikV!cXUHgNQn@W6KGYjQ(n2QCT-;$X$t!{d;vWlI%kMmDp2s
zHB|V)$zMQc%<u)q=a2FEkA9Jzp5g!O7g^~4zpNF4|GQJoI&s=zlK{c*)(0xat%Q_-
zP#EWW#ePFDLEOpGlK?6HY)s&f9kb7802+r)N$Iy~Z*Xt{n7otP^~CZtu53I3k*tGK
zPd>9K<}BnfCZ**Qvg6EZ?<78(Yg9xfa|$ETS~aiH%!fy>H|6W>rF4RP)`?N~xI@I`
z=U+FM*w|yq_bzWB$Im;OkJiw^*HA&Tx&p*=WOx#RBOX4)%yl(>4D>SYkD7+~j#k>N
z<h{ajWprih>paTu$c3T=namyF;;6%HD=cNf1-&Z8Le@f9*1^OZgLW?Xjs$I2u*T$D
zs3!4Mi>`_t5W69W;&oKQX|erdDU6v{(5}9Y*X<o+(|)JwEA(kKP>Vh1JeiGVe&n~=
z@eUHElXQ8uj#ON=W~T?}vKE%an@q{b*<jJ*&Sk{0nWrD0NO(6|g#;4FzE{$vQi>G*
zw{+RV3{*A$=0sEDXKrCJ%n%MUqh*0D%&;sl__KT<1CK~!-9X*f=9&|Tqsuib#?l(F
zRnR?U|EJmlUD5L|?4QtXJ>q=Oy31C%{=>5#Cw2JS%&$L;fQ}%RKCZbA>zK;3p9kn}
z<u=m@i_M_2UE6#1&|2^@C&Z_nVSzqY<+16mW<BLgXI;D5`k^%=W6%l^c%x_gg<=SS
z41cOfx=TPuxe4D->1Z5uYkq)8iXmG6b36T)@$r8p{`~Ltf8S2m8vk$1VekJ^_v=l^
zq445A%t?BNGPCE2p^?uK!)s}Fs#r?$Nb2nVykh>#I7~2hf4xa);@bxoKZ?Vd8MQSL
zTRdXOAOB=?dh+soJ;`)UA(29!DH6NEOOPTNP$R0AxltLPH7^;Tu66aj4I$>G!$9T!
zwp(K!wn^NjZYwgc-^RsSMNvKv?{z0@7(%LSdr)K);D{=Qgl?1+PV!%GrT+{qsEK?)
zB$cPsZXbDPOGjgAp{~3}D;a7*+T5~frFcMPZ4u0Ccgj5NC>x(93XHkMpS0MW-m0<}
zNw=}@XRO^M(bD6{AIiW>6o`{D9XqNh%e2@9@HSb{R%UzeDSLTSPlA8{2}x(BvsLSu
zvb1tJhSIIGbyhoh+pc;fg4@>LZ7AcnYMmopGv8{_IL1EOtwlnAoiZ3i6IGRU-}50p
z_-Ejwao9baV~UrL_u+1np6A}Bw@{2ron!6R^d!m#i=cL92O7WpqS)<(eZkpwjGmSb
zrtw#{+Gx3Wy4GRyaS>d+U9kaI7<4g-Yhy?6B95pRqL*xv8~f3nKz(p{N*P&veK)}_
zB4}i`y*p>sxs)pIl@sSM#^lCyQHE!mDuHK%PWgZ>amGP31uzug@S$v}8E2ba`&#*X
z>g+GXy?y-YkxG|&K=EpkKvV(Q!IxeAMR7^tX8|P9hf95UG9S4oS|s%nEB^X(=y@kF
zax`b$@y`55B6B~BtT&$en~D>#P5&se_}A{U#14B4UecGm(lDHO?y!a9P7!|VW6QYn
zx!)Q>*<g(CP9%*3N-xIOhC^O2C-9#41i11^(c6w>?U>o;DX21h*@IoZyC#O^V!tSG
zN@5_59}3-d_k5FPb#0+aJJ8lLZikee-pfiBsDx(X?Y`b~RXkpc{y>V#pv@+_&7v#N
zKNk9y|Lp*GjJFUtc-2PzEo^c?laG$qAbJZQZ#DiWXdtIba36q-@^4L~anY71@0ct0
zDMN*Ed#GDFl2AjKTlaS4)3`O9`%|Xjd{uq3i>8`u@HAOM8<m^NYcy3sW_1MktZV55
z=?u&Ru*v0&@_hRpaveUJnS&Kf1G{L4l*53PPzR?i_YV2>{8S`?ArZz|fTm{vMFbVU
zqC6|&^97hDP@hM_C;J7ug!@0i*vCp8o7r^Wdy`^tj<e1vA=!wxXzM8C36o@0Jk8@y
z2@r^|TFM^suJYz*fsoiS7+az}?x4hV^X^Oa%Z-}ht>FJd*+1}z)<s*pXxX-1waT__
z+qP}nwr$(CvC6h>pS5q!mz$IOCTHJ2FlWX{=9q2t-dlUtlv_sX#x?UfR)fkQvZ@nE
zKEmSE8%A<G&{l;l(ca-)>0e0Ov{l0zBX%AT3QY*YlfmykdfMVnILHu=!wvWR&Gn3f
z)p;EZh9xI>B<Dehh$pzohPy*#jA`=|;86)UPAGlmL3Hr2AUfVbfMAn=Z2^KnY=6s!
z;244Cn1ox~{laa=)*4;q?q>|do-E7(ta5~B+u@e#G3jG}4D(JjjKBsX@Zcigq<Z*l
z;Ix{Dq^tzN1WEF=lYp7c>s7JA#DNr8HRuhC1ssAP8)Cck@dDzJZ5h+K=C>L>emX#g
zS%uzr;G*NCq(J<w@_@KQG-BaXa|ak!vmwOU=E!=;LEyJS`$QrB>Y@VO2HR_X3InRb
zIEhB$*NRU-F(L_Axy46y{2rN`hv`(k{Ar^DP%lLYw}MdBeiq~uU)%>KJYYC5in#&0
zUMS>Wi5E?kSwI1NF90`4fj2+}6<ba&?(oOpH9k1>{_gJ}#Toq7j|TT4T=LuD&kmoy
zM|;%(A0tG6(B4^n|Dh13eS7#}w}Q(cB5)87?8n8#|J;`d*}*bS6u62DNz!xkWr{?a
zxwgO1s+I)uC5KDGVJ-M;pqox8HUhVW*Zp*fnE%i++1t)1?5-q0<-|D;hQTOAR!cD;
zPSlhjy~W+AF4V3(MA)@b<vvL6kE_lC)hGm$Hz(&n*AF2q^>lxf-D9DzmDob^n|mN5
z5K^WUDMhKWM2UM311be}{8%W4EgAo=sjg^vCH~YK56I-}WiFpLuvvK6=nE4#&?@8Y
zsZkV(23r>oi1V@r=GsKZYe0OGNLf=5ZEL6m88MUnLat7!{;I#=@kgE<)V|e%(AjH0
z8wg@;e@pk)e81m(&LCiwcKL}1z9|>l0;=o<X?5v}{|5k3G`nHQUL=odnuek$BWex-
zrTh-dY_CGG(n27HkvSK$aAVq$C@J!Nw)m|XzTskvP8Q`1{SW#GHx>P+5#Uqiy1HC<
z)yqsl^g2zEMf04$<K3_kEP`E-z}SqL=pe%UwS5S>z=H{4n-UdrWZS^sf)ilcZA%q*
zb(bu4os_UbSM^@O*9wy1Ta4roEIX#*5yJ&CO4X-eQNUKi39JUZNPxBqd7ti}ASrve
z65xhpSh5fH969GpM!$b3wSjaGW;2A45*(PfPIcc5wS=7O+n34);~~I0r*Q>s1EfWX
zDET_mB50KdlLeFLl*2tBQd?~8zt;Ac;**p_fMg%o@MXr_-xZxALr|fRsv)6}u<vE=
z<e0Kzhxf7s>z(%gck0PM1oLe$j4_Bm^i&b53_M-&{`3%i-m<q?C}x_0V+Ry*&zdT>
zxnd)Rbi&#6wv&|k2P=!ZN>fsT>Hd-y`AHUO0g1>jgd`|s@!}8a**~=zab%oLYK(d4
zYo*fOIfQzqSv-pv)G;!BF;-UMqt%N0o`ChCj&TIS<1pY~R*Fi|5b>jha_$<IS5X3@
z2*iJdV^+%ii4Hr3UP&Ba1wD)X5;IGW`a4P0&>KyMbVrosV1P@%EMQ2^^Nnw;;>CGP
znCm-Re^FVP_AO7QbYN>lrw?Gk5k;~P^~NMex;^zpU~@P0?v0raMPP69zf_R;%y_*r
zY$7gcS(8`NJk#Af!Xo&lhlW|;gFJ;P9x36G>7MED`A!WK07gnEb^DdSfFylWNBmHO
ze*L5~A}EaFlUHEW2Oo{Zd%bl)I-kveST%1PaBl@l-6zQnQhO+ss7v9$P~N8N^~ntZ
zZ$?+6Ty~=E6eArwII1@idA9Ci9=0+xcTV?1Gu5sQb(?tv9bNOOBY+YtMAXr!V!68O
z8!dkB5o)!+aumV1`d)vvA$Wcs(01*YdOtBfRZ)G@RDxhl@M;gZ9j<^Ds0&~w%;HS?
zty5;fvcGF7+DS3fn#xNe60hKC;la&*<K^VJ;>Rn){lqN>af>tL>m`-J#?k9=d+9F8
z+^F=VFWe^!EU3x$M|LT1hDI#EAYQGpX7KRb;cE|TD7m|;gxzmz=}%)xNr{o#Zy%pb
z!fd+TEQDilroC@3xNY;l)0RlBlJ+6)m#uZUsgHYxBtSi6{7%SyR-w=>56FLod|YYI
zKu0&Yh`Z}kYqp5v_nvtut1)*cLaCl6I<W-HzN=yG=B6M@<39yO?YG+qjv%}$@fB>Z
zElT-(aw30rz@JYUJn;;t-<W}bM}@pab@0mKC_o5QP4z}5OFn5X37@ECvu*d$rNs*?
z$`A=lL5RvV!HoU7ydJTy?mLSAOPezMhw+q=o%w&=aBtDvbUYkI^jTHgOQ_V0Wl)33
zJ{q{Z;>>Eyp3&6uQ15CbKuSbN0+I3W{1-Pqmw_D~z;CZs*$uI$Ps_uzT_*QRE@PX>
z{Jwwne7w$P2P)^7NL89Rbs1bG3?-sSm^5YbU_lnl`TP8EyW3r2v2!WPB*4CuKK%0W
z^DiU5&--!Td)Aoq5R)TMDrJ_I_s2aqV*R-5m9N(!Z%3#5*WF^H*9rgUJ}Kji`uPCH
zxPZjOTn^cCq%c!H;ZK?s^_$Xms_o=UtNBoto#yJ4qlxbMRp*Y!eZ|$>RauE(JhjpN
zmKIx}Lh5|7y-Wor5#*skS6?wBZk8$nq43$$o~OPTW&);)Z=y0;HhE@;>}4fn)%}m9
zGE#*J<L$rJ-&$m({sA&{<xk+V3R5S~Bt#04CPG*7c5+c!;Eetlr|NDgby^Hx!Uj*>
z*FZA-rb<h><{G$JeQ1mot@rJiO$)i|&Dc%$g@#9v7%Np)X(gLza)E@tPfVFk8^vbp
z#+H8jNz-R#kBeQ@N9~%5h!*QoC_m&WbMC!?=j$2+qVlj&<ji7?*6pT94{aHm2};z(
z04ttcc{U}HBU+qC2&oescsW*^`yD`c?j1lW%S6UUWlR${F~uGM7INy&-x|LF++<x$
zwd2AyHWPM1tEL|qk=Y32X}QJ@+U*HDw2};YV^`ppk#{y-xYmLoo3$642C5Ykt(&j$
zL5;Zg0{Z91t<<+Bi|y49PkI~q%um(=rdoCb{vh!uVook=7O4VrT=>pD8nz`W#MCHZ
zb11$QJ!f`)=GrFra4u|yuAMs7xEoY<`J0<bQf&o#POlz29WQ$vi~NfF9}?54LYl2C
zHhTuGTQ!>&os=83mYti;!VwL1qhaBnAO_%9FlP6QK|zbc1`(=LnGOoK5FuBNE-F<o
zD-d_SkV9#Nk_Y$JI^exSU0R7sb?#@Cg*6?4M$ZQTiY=<;3(P<lejd=?>;-qrBJYfW
z+RBldq`HlEv##CuHn@CJxFsH7`Yt`4FYbz0{X9iN?H=vR=y7c^hH&+yfKzka)MVk~
zHI~(23w%dfyVscQ4hRx{cP-5he1JohmU}&%{PFQXlik4W+s}*a*=nXev3<O(?~lst
z6^NOL6{6HzL*u4}Ty~c0dx9vr1nro|yxd26i3e%0KfkO*N}@S}55}>3)A<Zla*|*^
z{_+A)pb3fuVws2~Jv`qoYFlzPYNx?em~kEKfx!m2nTqvbc<>v5Y8Nd+yQS)OYVto3
z@i~N!_W|8YD`iBNQYRD=kK7r=nGyMUN{y$H#zeZ0D$dj}UBJhYo&d%cvs*!)SvPZ&
z<pjx92^Mim;4;!nO>DPhyidT$!2nJoj`7f&_kS4}YjyZ9qwH;h9NES3HJ7cm*dV4<
z!jH7jJJXg6Yi_*LOqQFwEb-14wMgbbC75ms+;&wD)2A>|b{B5b8r*+Wiai7WC({DG
z{tZR;iN#}R%7UA^R?86p$|)0wQy-RI>U<^?_QN}KiaE!}x|6yZrf$V+N&w7WskP{2
zn!NxM+EF_i95mhkw!XSH=Gd8J!OWLnL!Hq$IJXKzVbT>xkRQE#zKkW7TH3=600w+M
ze}X%VI2>}^f50&Y7<mz1%)L<K2Luve_Po7}NwQI!M{Jn57S>M1QE-kP(;hlJ`@x}#
zzCdQL{)8O~&vh;j42AVhHw7{CGyi3UA3*Fz4=pr5oelZaos33ZdVmLrPfmn;dxYW!
zELj)}H>TnKZ#`R6dlTxTsWoUA2=gP_RVKVYnsM_?E)njLSoThHiJaPIMGSdl(!_vK
zzblZnFRVJ|)p-}g6zBA(4@wAQ;yt&s2jovSD+{v10wN+R6{k5)7{y!Iq`fj0;E|NV
zMG74eO6&%ZcBP9yei(QN4M4QDJXK-i*Kg<08dyTH)6PHsFJMFGiyXYp>j=3Z<Xt6Q
zHHn8mD+W86gl)9>&Lz9E6&XJ-plX6vltTM&udbd4);3ie%>d%}O{X&<t0I(-c?p;)
z)&!)_1Kf!h^<Z+CYSn^0{Dwn$YYq%<UonMMmb+G1C-K|80S^B<is?f-XP@2VccX6q
zJ-k;79brd$FcASgn=r+EdfT(5%I@YDmP9r<T59G)#GHV%lY`1`O1&4YVwQXsUeutU
z@FJQd!2-!s{7w1~ZBbmW9lRE%F$y*Kq-0#O8maJ;1n2|IQP9L)N0&sb&&S;mlUPp&
z_vd$YQUviLr|OJJD)ow4DJ}>2$EC1PQGJz8hIZcfV~602)qe6x1Y~mLry={4V}J7)
z2zfmXFYs<v3k~@!YW4x$&HcGr9l5_9RMh}JiJt3u`Ed{f8^X^HW5p{EzFQCSb2U>c
z)HMHMH$KM7oGw<Z1Q&iKYl9oWInu9~5}V%Nhy4m`(_W$a0$_v;A1?!k9m*@dSXoHh
z3dLYv_&nI6)}9}3H#vmyxgP1YtB-vqb^CDa9HcG^7=JM(l&Rx0l)5dmuCs=7xI<n&
zwQK&?=R>n8a0uQGjfL%LVQS{gQlrOMP33gAy95+1rGDlm*x_?iaHl^Rku&FgbPjuT
z9q9#`T0i;>hcQvJf^Q%HLR#}Hcx};YqYXA~3?#$lTSTW?<LciG6K1=UvqTzfUZ~Wt
zgy%{fVP<NZoNQ<z-R)!GFfZT5M-oiu6P`fVL^`&-5J(Bfg<r15Ob8S>V#Z?AkIt!(
zD-q%oJcXGsgXM$_pw0!}^)busbkXB#*BR7aeamN0xQ)5ZdmnT*57^{FtP9TNxu!6T
z<FTHGS1&s$t7ZmrXwhnDkh2cBupE&i{x};oRz$;<w&c1LU!jf;2>jJw%bRX|r$&LI
zmI~CD1^veAX(a79V<d>91-`M9T0OZ1D!sfSBm6r=gR|x=+C0m(6Y>Q=EjSgcK93kU
z3bFLXuOn7YVxD$$Het8a2?R(`U!%^iSH)&|ze3;a>bC<_J}vCMv$YkmR#vx^p5|$1
znftOc`O`Y`GjG(02L21s5~0dj%GZxPu?(FOhL;GRm|4V|=5!`7wupJWe6lh`Uie5>
z1>(hWKvnzV<d$IUYrPGx)LrQYi^)h1H3}rvXw}UMGWqB|hgG#~x!b&-NSw1B3Rj?L
zm|(M=!phCLVK`dY>P8oeWXIF*(jk9*tLSGN1&cWk4y-NT!EA4|4tA)&nY-jK%2#te
zkFtN76M!AY5PA%@V^^nCr+U2PBFDIgqnqmC_-?DS)viqx)OYK4GeV5H6Cc-w-sQ&)
z+q-(RF%nI`AR@%~+G4eKwR$|eg`dT&V0P9UaQEqX8wTF+$7*t`urVBj$6<Gr$p)I`
zn{d>o)J}#x2{I7|cktveVpHhIT>S8ZSQk`Vi<8~?GziVud}MaW#7$MHmud!74NvB0
z4AP-|+9sfZZ&drDZ2d~_lmj;3OD}jz)oMGQF+O8pxvHy!Wo=1k=i~RXe-qY`z{@S<
z9xsf@Tp7}^GI{4TI!Srd?Rp#DU0{LvxsWneB?kVv1cr97s6$s#=qAMkFD}{SL7T6P
znh8%j1!#0a{7Y>qK4#%;9sh3Z%r|64b7_K8Mi0qYnhb3?k$tJTZ923s#qGxtPW!r_
z|0p8LEfq6mm@#i2`T#;f9OqLAY7BGh0`DW{S&{d%1FkSNG$zD$aqku2jMP8L%vMxO
z8w1GIEhZDMOLW8)V;Pt!!j%Y8W;q{VB_xp_yy(~G!o48m!agYX<LD;B)I<_wrYw3|
zB~#Q6Y7iOo^A{k9hcfv;wKm89Z(D(po`vE6k2<i^<|v}ys$PH?J>uN!cfv|egPC<4
zS4-lC^%sE`Hv*)Dxr9F;zlI~(_hwfZumBE-f3a~{lB{0v;lAx_o7bnq8Y)~Z5KcFz
z2QQb$L!v%1hA~OXz!jrDBZ4;3v<STx$SI*xc;SSy_&=le;}V}~-JGsWsq)_4SQEPK
z-D5BU60g_i$NFR6vQpjL*5kHyq5IpF^@GF9>K{4|tH*sH142O2kx(ZS^qf5`4gZ7%
z`awX*bbg>EeuNkZwgGF<%UwAbc%!{RnKwYAO!VU$vCeuUT+ykLEb6+;lC-SjhuNNK
z+DP`k2=to~B1&Smf1NK)G?sv7-yG_LW@*}6EJul(P__j}CPD-4_B@4sVN&5{W6~G`
zAcsmH30v%AauSpZRtVt+2#iEwj%R>1Jls~7V)Deq_hLD*0di2T=)S{RbeMq$cKT4_
z7!UY#TsL|q>iAHz=M~FN_-X;|rJ4lqbAhphy-vYDWAliS^eh9-EeO?_h&Z#ghkrFL
zPuF<ZM0ot2cHpc&>yt9PG=-?LTx{zn(e4}N;0h4HNM$KRl$R}Wc->>KZ1u1t)%ocS
zGvg)kVqr@DJ?IJ8v|Y}T9Ydm(Qo6|-m8OlvB8!B0oV-Ej4KOMb3v`dm5eH;%Q$&r2
zi2o`cAevyLLo38<U1GQ$Msuq7)(9NC9O3hgVM-4AS~Z~;c`w`uw^<c7%d_vpLhO4D
zLGF9#?%3g?Q0PuZOKYfi9=vZAT@zX^EVgr?7l(a#1&JPWaVS=e0l+S2RumAMN^?Oc
zkdk9q6s%QGBxv2m5V$hGMQk8G<dNLP$>h3$xm(iK@YPYNj;?Uo6LXc1wrQ)U1{Y<V
zR3xET&DIsYOKgyY1xdnQ1AbS40oxrmctX^f$RPK#`BbA*e3C?}sAMR+gp06sa#KWM
zmVz<A&(khfE*vajB~_A?JS`=kR;Yr$3nwWRfTe;<`F6C32N46&=!-|_-n&Z-OPG+-
z!>jjdzX-!P1ujo1t(;Hq&!g2JebkYCj<$8bD}9bo<LOXg+zG#h6IUFH6b$Q3+!Dk}
z7)6A0Fy_UyJGf9Ig3s{SgUYKhn-?e6huZ3`DM~o)*%^-CfQpBWbw@5ACon<E>vGbn
z=hEov<xb&r6u$OLoZ@6kU$@+M75b-(>niXYPX?~{<6^*Zgir*LbO9vYIki8oxe|bv
zF8nOuOlksn;AwyS{KzLiDxp^9^zn!&MFTrzcBVH!3S}VSnQ;(`soWM$Sfh*%VjDUQ
zn(O*k6s4_zQ^d2#;(s(R-T3UpX_3eX_k{MsnA|@16EF4J=Z5(3r5mZBNyX#AHW)u~
zPQ}6>nopwk!?*BdaP;g>Y(SCR6n;$@Q!v8J@(9+y&ml_eW7rp36x06^Dd(oBZynj|
z%>A6bZR3uDd|L4P_ir3<3c2HMLNRcppSlAdRAPn^hTYr>c2~t^`w}heRxa(rGGWqR
z(u`!X#3Paoe=p(S(W}9^zIfe0He$F01AY9t8XerKF}`+jf8%f3Ss>UgK+&02?wfb<
zsCE7%*A;PVj+kVl7coKwg)hQ)VBqW{jhV-~xUJSYqt>V{3^8fpbxn?_K^f$le5nbp
zF>2~i7j7hzXkUq(@;)q6S5Qd9MvhsVO!n!3Ep2RaD=rgqe*L<_t0`@uMdRV3z^&`t
z_nhXp0_-=hBkodc0ORfs{%NGSz>E8`?_D-=;8#n*o5LSzbFI47+HpY<;N-vSS`m3l
zpk8B-y0AQ?9Kz{cOTXc2%keA=BsA$wNzxv%$QCQHfmFBdvnHaCV~iLpq60)SRRiI=
z=z<Aa8c_QdczxZDIeJ0%w(Z(qYE|)>T&m)WG!cN}k>l&k1PaBm46X~YGqaC<wBQ12
zgNa6nNk{mG{AN%Iw1q`8xIpKJ1ko~;nMCD3v}qV`YVnLh5GTK}!))0xxOve6bh=pT
zR5yL}`*)}|(QhQLQ&J*p$u?cn^;9*yEO%*<{*lxG-)Gnk6tK~-N*FEk<_rY|0xe9(
z3lM=mc0mGbkXJ=LjO`cPuPvPrdJ=rqHvogI>{KuPJFj!@0FaCl&UkL8b7N4(UYJ9Z
zrgF`p^{iY0$!e~(x8o`|Aik;r*tSVh<wu)XIGq#~Q$vh4Zn(PNH4P<Tt}_&-!VwNE
zy%o#-VGX^R-VC+rgC{gF4M~QFfG+l$EX%>TH-2p+!8!NfV(K#!uvYxYPDlL4*>#B~
z0qY%Zr<EP!`IT#+ojfu5Uh}=)`KZ>X1C0d+yh&wNt%b$j9q~}C6Dsp5Ijc&p#WVW(
z5np3!;l~6VSX_U;c3G)PX|53ogW_$f&eW>u-lUNV84M3l9&841=j4@5zqM56zsacj
zx9BUpRGJHKW3m1LVpWF>prvA<hMTbRz|F;16=an>Q~@z<B|Kw)|1)CGKvA~~y|!kn
z0mUk>x_0I^b0#%(H@mh=pSZ^~5OgEwrnAk?4EM*ktZmcL+QswMvuhp-bIIHNr-^RV
zU%8vmaiPd_+_?Gt7Os^r3sr;;vP2_nxaWof{q?#>X7B;$&w}86iCV>q{aUVRxp1R`
z*e_p{bjfn%-LKjnz;j&7?|H$sUvJ+xoolMo!lgsx1MueWLHtW!{{eyT=f<y~$MtW}
z&!=H-&Br(RA|A;~(8n+Mo>|<*ua}^sPcz*%G%VQVH;vD(ncGbL$Gf5H1O9e;xB7pH
z3da9nKQpqj{;x&Fh^E&6aB2QIxA(M;Yhy-200LS!of<jTYr3+mSsi5Y6S?P+{uOl+
zVujxmb6?tCheA*W<P)!*<@BTxK)`|qruRNYOD9wWlknsI0iZu#!jwruF-BAzeKR0Z
z#+4B>Cy%^SyYq@G{?+DwUhVw!|2~B_P?H8)K{yoU{$8(NKBC*j%5@YYCnOvyRh%6i
zcjd)3IXn8ke4L{HEUizi-p<<WGX3tSsj0pp!d}2Q8w4vxz%v^}#0y*IikCu4$vE~7
zSGGMGx=cS6Ltrtk=i!i1O2Eh*M!5L`p-qKPr(@GWju{4$r@r&ZTpqiy*=kb^MVtVa
zl)5Y*nY4_CD)Cg%t&CihroXp!L&Z3H^jK`{WO+AGtkh@Zdph#cdEai5)p=WK&I#_C
z^gP6v>V|@*zO~m|x?;Ef5hq=6=M0&A&g)Kym}n2a2s*u>Tz@{FiIdBIKKMFYQYecj
zNZua+(H13HI!?Gof6-^mp|63MQf{P7oGQ}Gt2D7EkWA<2V|ybG$wPVWhUA^o$TZKc
zKpHq#5`s(_#s8+Zdk8RMH1$MoC)!9~`TRw_@^ZgEN;YU4+?#-eIXF#IL9Pj{?c-KI
zvZu>NoZwtWd!{iw+Shugt(I~S{)sd7-7Y_kQA&hs`MJEWD0F_fn2VzquauZ{rWe0x
zX{RWa>A{lp`@JYi^ZSK(e^N-i^<|sL%Q2QN0GY*``Fc&7^UNP~>77gyQL0|?lGDZl
z$@YzW6g4A#x%p)9B?)Z`nxTy(^ZN*>j`Y6ggQ+5y{-HByp{+~xuqRGIqIkYN65Gyc
zbaD$s&K-HD<ABZhmjW0!Z}p4+TYj<3sOO!Z=$j%Ti^s58GZl)I5A--%+#tloB*uyE
zw(6pyEc5O@f?sc4|Mwe?Ndvm~w~|zd*{2D%Qf%Uo>*$-Ic)Bi$_C@N~d-3tR=K7^%
zp>6;kxH=56GmMWyu00_b>Ly{CFSaBzc?L&3dTGM3$$ausWPcZDa4M^@R@Pb!Y#gok
z?M$#4Bkc5rUv0J;h+3CfUvbtwa83|d!{dok9XqZj3GTQQVxt+?d3O@ghXa#=%qr}Q
z38tX(56By7%5``YqW=~XmRy3@QNQIMBgHP7vOU!H>_W^-;41Oz!nHF>E*L7CkLd$U
zw=h)(dyb!UmE``X0bSlWJUvO}QLnKvxHF&^lhBX33&U%k>dLHna-1>#s)}zX)!MAa
z&W+^D4mm*{cs0n?Y~%L59PZ!;8wLL$dLvKn(OwQbBW&Eb2D;p_HCFezW8X2^0s6yZ
zxQsl^!F~4y9DB0Zk=aTYr6VS6^Jp+zV^pXF)2t<K>tYU5WI3!yGHcU@a5g5xwCF?>
zC@(H8NBhZ;%<3r9Ypo71(^w{2|9<aV5r!GI%R5KxT-hZ0m%?hnI;BBY?@a!~&e2lU
z5+iR6o}k5~FV<?O2sWI_#Xgq|Gk)NXzrk21%Rl@ScKwWpR=9eYL3D)TJaBsi7DQXY
z1P_5Q4@``7i}x8Us9)nvA1AFZuw~WF6t*gFD>A*5k5M8GcxH>$X}YLQWiY`Mz)8NQ
zn6-_A+!JG3tR0k<hADJkWzWs!_59bnlNY7BTkQn>S!Q#VigsA)op*FKoL)IfOh7$E
zDmU=ub1h^UC-=quFEx2v)-GV_|7QB&roDed$|MpUAc}a$I0=D)W2>@F09?s(*>w-~
z|L$Z8Psty#$D8-hz~V#jMr=*@t24%kAz;gG@vQ~Jv`vuH5-|`tufAX`3p;Rh@^KN<
z9+<ZYaBqS6ZS)EBYZBeF%7gV%B@mW6?FzCw&24(DyS~$l8s+Y?>xC7iLM%LU28Ucc
z6yPpT+jKfGM&3<V%^x~4<?{YIpD;RNPe_}wH^uw=I!Fi^)_68FKn(TXM^r%+tVS8N
z!+$k}NnMr0r3EiIq3b-KSl)1zKH_LtoajZ$VY9jV9GIQ<es0WVpf%030<6w)Qv+*5
zU96<ZD7;(Ki)20)o*x099vrVfqPv$8j;ZTo<ZEvFBGa$GgtNfG#Mh~n8+}@a+q2Ll
zF74PKtiFb@GFKz@J|DMgyJ0rNi?B6&dW`wnL0j7-a*Kc=T0rsr)z_J(lM++3(^i?@
zh$Mx>r;|2qzZazCsFIZ8o17zrrJY(?xLf4Ntf5xzozo@%OvC|oEG5%5787q^ZXz^M
zE?2!Kk2K}^I~n%Tbz#c-UO=NCy5=~^M?i|b$CAq{ul0>#1DcpnK!oOVDd1ymb_s8-
z$Wqm)9)<(K;cai;Une?#opO2Oa!Dd{ms2C`#?wj>*cw&XGB~q;<9cb?A%a5KOxF`c
z3r^%EauOgt1+pQvPL$xA-ij48MK!xYRb^|fuuZ2!E3{G1fbdl(g)9H4F2@=*IJgWb
zS>H{aVlh3pwgXTPX=g44=O10RbhXW5uQu*A>+!OUE`CALiUR4L+;ycK;}Fy(GFn(S
ziEZ5O12xcr@@C0=`7^Hw)s}&kK41i8Y#4`Z0OP0s85kDy7qB8)!|3!t8)DFU`1#NG
zK1SPLIL4JM$lPJD(YF1$cIe-pV8~i=@TgR-x7t+3%Gzrs3haAuv_p)jy}$F4T>jF3
z)a6B!G0D5pCm|>UT4lHo%YZ+IoQ17;v`60qf}VGf9e0;j(5-v#8g6?`q^bDV_G^S}
zAoF`VH;0i3dHJ`ehk^SpPR^9||N4)^3(H+vs%{i70iAIIc;N$@OS4h=>2UFzDEKyL
zYRw<)TEkf*cSicgD`|z_=p!zAb*U}NT}57kNNb&=C2oRTs()0Bp8?idv6UoWK(7su
zb@qc>#~1PqRCd@wVmoTSm~Giqaq-7)IG?2l93WPjXj$evU!9>Kc2|jgv0?>`s$_dM
z_F}6b_D8vfQ~+lS-=2{8Jeo7m%i&DmQHiZ!L>wZDh_KE-Lfay)-?T9x3$fU=hGegF
zF2thlZ|cp$Tc;8qk*eukxw<u`rB91$?+0Y<cd<Us#O4^oZ%5Qzjby?y+DF1f2RHm~
z5hX+h&_{(~iwPK{@(i=nL@{@180#S-*n=sJR8~pL?W{k<$spx&uc?mD31W&ZTd!KD
z?7C8W_|NTK3%2kb6H9sUnl5=_?GH4EfK(hZfK^myvmVQwHHGx4cr4tHFvo8=uWrw)
zjp*j1WZhma9;}7G?-w?j#Q(IH>B0KIf*b&pUPHU43(DkvCTjWrtzdhCn4PiLZKNS*
z8)+G<Pov@78Awws=GYnOx1}z8F+kKBRntKNxnnoUh{)dLo9DRb(+13RmW+NUx|?wY
zu$^{MV$CWw0rZGEt*U~X6EgvzfBZZz3<vc5;=`c?WG|}!cJ_GsA(k;cYyF{&mbKl^
zCmoOL@3B$ZB=YcjvU=SWdBNvlw37wZ$`C&I65*7_ZhvB$^gfVqF?~#8EkJp1O}9-+
z33$;4+dh4FSxQW0KDlD^F!%0*%<kP_B;E?$&^v53Q)Tw(P;b-zZCDP_H^xsKDU12;
zvEK%Py{x3T$z1Cl`g8TI7m-%I^k1@-=|4b1Obq{L#`gbxCbH)LK_}PemLTF-r}a7{
zY07H+b5EEBEEZ_iE|L%_u95rwd@dS<GEhR!wT>)M4)_GUt;aT^7Dv_lu|IP8iB7i*
zX{(mHu3TF6IY*UJ8!d~uYU<Lb^G&6IsdQy(qnO*>_xZ_pv-|rtG^2@?yXE&00mkQ(
zf4VJ)zq5Fj$0itlv>f(Yr~7+-J(lIS`+I!MRWll9KGZG|0PYvMf)=5B*RT4kU6YbX
zeQep1iaBK?d6X#fmu^>sVMB|a)*!cTnDTh~!P-Q4Hn_Fw6PNnOIjN$U{VMh|t7X!p
z(JicTefExhLyUY#nD#J+pCV&^0=&&lM9j-YE*9+A*6%8^*4ol@m)p)(Sn;_pbiLIn
z01}aG3*C*pvR<RZi3UzYv3VSyo-w@S(c7fmfFpg%$q9<6;G-)vH|25=<RK)8NsmUR
zS?69}^|+2!_1uz%<Z$w>42iRSkfkT6fM~?3Dm@{Q-P;L(hY){*%<XMF0=KQC!)5`s
z=5oVB2}UbQo1tRK-L5|5fYsx<!wGyNMmUs7oS}``s~yqffHlU?_TPk}bu9Qt&yD$c
z>H4)m*en%3$>g8K<3NdAWbgnl^>L==G(T1r+nJ?=e0pqgo;mxNTy2@hS`bw@l3oN;
zlHW7yJ!rVL`~6|owTJl`bGvR%DUDPa_1~*BHZH@IY^3*nfppG0nX;>agr1>86eRFn
z5agJrqJ5Dvjx7zDu<*PRY19iAsaO{^Rc3G7SY~(1-gb@!uRL@FhUN@wPqzUzPto8B
zjIDL|h4FJRt!BT%)f0qLY)`Z%>KNI0%;7)EXK%r+)%a9mT*@-v;zdIZy@R!inHkQF
z>ohgA8Ab-}7W{tC3$UJfJDf?3$ktkeA^xAT+-v%Dd{1Br%w)(xV{PUk|9}$P3-y&+
zjOo)%;zX<mCpVMZsVj_IHr*#jm6qE#Q8IHv2#`YR^8WQ^py@EvH7rT_Vtqemw(Ul7
z#oS5y!9Kdq%+c=s3e6K^9;0DFAxB3!haw11RI7vUfw3<J8f6lX^Qb00<3TLblDKwm
zZXw|9anl<Ug$d8odsl74V~__jPFH7a?OPfUgacaB9Q4wOac+#H=`57&+#@jKf3opv
z*SJIe0&7^4wV4+kZWW?CO-bLOrj}3^AGnc90#2lNtCoWtxjP#;u3GD~;ArX75g+ip
zHVZ;^AY}`np|uawS9tyn%w=h#Dd$__S#au$!LkcVo%Bl71A>$j%PIz74wb=6E$SR6
z(B-F$e(THGREbU`Uqz;ARg<0AM<F$p>o^*XuQZ_4ymw&-HFP3VT&x6OT@5^Tzf>@E
zAdPc?=pxG5KgRKMpLFCgKzFRIPsAfKSQ@4cSc$yi5>8eOf}herbQn<=m2M{|HnyWX
z)T47(&)r-Op9YUNu`wS|yh~lnwAJI!c3yUB$AaCCwr4RYefREVh9;uSVX#o?{@i91
zn*@*?aA((D8Y}ED2WpU#Ko+R*m#1JOR4&<K4sH4bBZBXL1@5thAEBuJ*qY0+5WwaF
za0=t!FA)>hCN?GcdZG!Kqz=;8M7+7qGJTa_QL2rY*}a&968Ai3?|c$u#=}TNi|x{J
zAP?&fY|OpKzNpHKbCDO<F#6awOM>M8`I|d;+GwSbXY{EEaRD1bzzA@Qj~)Cp$?xh#
z*{MaXmK@eMSv;}Qk|p%+e-2KO8<-1>iYmqL`ua1YRk)T!5`Phw9B1yi*<Nq7?ach_
z4&oGeEE<L^Vt+z0RRJcBB}v*Zw}}J@qN?d1kZ7~++JE}b*Q7t`1<tLI<$DL3tjn!T
z`{hg7kPF?%!|rL=127I!z6QF*UJvqPOW-By-2*vQfOnaj2CbMPXwQV);abaXRM>C_
z24fX85LntS-noPK1|?!4wFJuJYca`*5W6hI^yBJbs4TD+4Mxnff7$T75~iRbmYjsO
z{>$(0IN8%E-x!S0ND;4|IC<aUQvMVw8{l^9*;qGR5*<@h{IKYME&4LLu0$Z_+pt+6
z{XmJMvSQXXd;f>2REY`7L4eHS0f6TpLPgc!Z^w+1jxE329P@sUCYJvxC|Wi%NAEI+
z@Bshi{@u#{bu~|IkjefAH;UQnfWGx|xD+4<8UY;WT7p0WR7myR#vy;v2l+ATYUeMG
zPR~8B-Cz!V`6lBuN)Pb9R6J0vp9^I!ut?PY`&y_HFXyV9ohNApiAkm@B+3{mVm(|3
zuNvT7F9n>JS{NlhkTCj1^Q78RMqDNUoN`=eQ#@pTE5(`5DvY#I!6`tLnh(p_KcY|J
z&2`xG-f8a<mYxh}<IX6v&=E%yEe#5ZVocOiB>_8J0v2<(!>6sd7!x7!Y3~rSAS1QA
zvx|myIK02%GbU5xdh7;?j({n!wjm9SqdS!52VPHlCBU>+*}N=P7=Ir7`FjJ^h1G&3
zX#r9jBS__I*FnOf;Yye-PR^k8z4}`ZQ%-gy=hxXjH1~esVm|Lpe;X@r=+I?G$4Dmg
zvx~~`%&N&r$ud9HA&dDJ={QJR%;f~fsvu`_e7#092y}}!m>Mu)+wn=od8SSW=1u#*
za^aD2a3Gk!L-=!#Y^wSU%G6)u?YhT3TXyJfMqmJhnu6RSVHA~{@;NhOe3@YopJ%o2
z!W<C82}rdxXvTXQ)C|qq@rU_P_3H8IltAnG_`mwo&V_{|k<x<8wa3OHUe_2yn6_+g
z)1Wh0u>RhS2bar6|8VZThZy~&fm0f7nHUIf6G!9z2UtuQ#a!GwoU?{NjtdcASAhAC
zjt;<xuAQ4KKIA!{D^AU4n&RU7x^glu*M$`@hVaSf3x&N(Sfp5>DXp-$O*-qXLQg-6
z%458SVe-&LO5`Gf-nc$Y`9wD5(bpW^#b|V7Agp+f_unxp674S|zH;WK!2}SihbTZO
zTzen>7|P=WM*RD|o*>RreR-bbiXeiT(iHmMiD4QB(9^F`uJZEwrwt@&Fudu2R7{w4
z9j12T-)loLE8Kc>No#BMmeI>5Xjh<f40l1R2L<m3^RG?Z@6JXZNaK-}Hs;D*0j+uL
z3;AKC6|fyGZQqWDz%TO1My!50=V#`o&b^Ag5{GuI?8hx%-~yti_bX=*8ktXvPnHk_
z{&NPjfRSoGJ@PDZ{w{<%p1WkP9x%(kQRH@_D=;vo_fbWLl(6Avjo}+?)jSH1^RuOB
z0RV#4E5wU=9S2Sqd+b+1`&Mq{Sw<=;j}FLRR5=sg3?xr^RKT#5MO;ql21OG}gbxU4
zO~$-;WAD-96iJTeT9MHEIDFsLMLnJ?2h->+5>e~5`IGn1Bva03J$LKGN~O$e-_kE(
z>%!0#4_m+L+;$rw*@2KVK0&o*p1o@e*=}$g4sf<cr|r|eo-sbS6;iWJae3mR%8#%b
z?=e?W{34zQo4rqpX^9?i>|!D+lO$kJ{(>2B>o;>kOXh^sfE#;T!_a*cl(2?6k&+yP
zMyR*ZySw><1G54CgEgbWVZ;|45WL=OH-7#~Tl-OV3gJ$oqxT)-ug+sa7&wyT@!n%V
zJN5XSWn`oRNglh`*WL~Bf6B@nG4D^2xFv?C=sx>j$)OSOmHeX$YwC_92m@I2#GT$n
zebBkEvKlnHUFh*j;juftG=5u<!GN_Nl9GLCs5yggjXqDn=1EKcZW)!9Xj3qe0ayr+
z{qpKoFt?jn9@GGWo9B|f%8j|+Hv*`?T^~rn>DNBCjKYJk28|nGoLt-NmObKIUCOv@
zh00BwdfES+ar4j8OM$iXBu$GGaSS6^k%4ai9?|4`L6~168t_ZobFwt+-yyi%wcIX+
zC@j1<=JI77s^|z62FI5l*ntai#p?xD06d97=)Y^4MkPwvqe~^4q#4#2fay9y$L;Sb
zi2gAN)BD$u0zaX@;0XISoqtDyJ*0o8Iq2TbN8cMbtRNH~KlX10@{7M>U<3p1r|swP
zTPy*ejnG$%q;ub-;F{$5M8*=|DKIz9EIwW%Q=G(wNYdE>i)`;^4%yd20WAeT*q^5*
z?vi=+b?mY4p~R$9(%vK-vsiyAiet4xUcbd?c0plmW@@skD9k!e-hAx&fq&!P0b7LX
z|9!Bne%aRny*+;SdU#Y@QvY@BVEzwh9|IH1|JsVRt*-TtUyAatT8{8QV#gpAi7uhV
zluxxtE0kms@R<x6Xt)#$AmwP%;lOR(=OadP9UhtU5eekQV3K$`c#D$@-N58r|Jd+s
z)%JV>TqGMVxDQ#zWnXU3jts$c<NP^WGsSMZ>*DA47_e908uJFDKnFl(x%#=gqvJMb
zw}BgHl=e}+y${rtmdDrPbK*)@%a?Om5;b)>w{3Zoqx*JPX}B{eiW@Dsd$?s0J2J1)
z=G;WJ?9ujIBEnDY8ZJaJ7|f3MXZ&>fyrjl57YgnZc<hl!*yjRe5wrPcOJ(X96Eehu
zE*M{yC4t4sX!-#01O%WxAG!-6RCv#7_|pMXYb_u|T%Gt11ZvRUAiwW<2iBMimq?#6
z7*Q6Mll+5(ipGNe+0`g@rpX<|p;6f;nQziCswf+?14Fvl52)~1lZj1IV(afmnyZ-p
zm9EJ9_t7a0zX6~jNxeWs6p^nbRJ%&`EvuPN6$&SLKcPT$)GhGcdBBz-Cmz0>RpThl
z$z~@gBG5YmR<oXb&7|5o)uUmK5xIbz7Sv$tsSSZaYhLj0Zirk`58apMTEOdK{k^j*
z$PlkoIt?iRg=lX*$5&HRO8xc7VSBie#-!mxWsQ4yIQ~_lSu4LgHhjK}WzZ(Q?9y9$
zJEo|`PFj%~VG-VgHx~;Ru9FzGs-#KHcc16@15u2apFu1<QZZYBs-qr=w0wX$J+^I;
zV0p=74|StACl=`fIX;<5NjK-(wJzeO`tw+(L!04Qjlr7`4S&@mB0P7t{U#$<R7qWw
zioz%6CLD?(`Q3YhwhK9v#&F4%)xKV-gDRN`kzU_Dh<b4ET1E5l-iNTYs%S-gn;WLd
zgJ)vwGvS;nr-`!hP!LDy1J)HWp5v!2&YT7zse0P|nJpC1rL828@H-aUcA1u?a~X(%
z-9n=SiJ<v<zKc<cUMWGtCqcp7Qw3M<xCP&>07>l+lN$hs2u(8lgs|U)Ms}>2DG&bH
z#8lMlY1v1UIr45w5+m&-2{vsxtlwR!1!kH=t_*XKJ3z!DeXJM!1{ygG+9cFi5P+O|
zIjT=7IF{(vj*=Y-)j)KIq(JEQgwT(9>MiMC>FfB8d3B|O4mA&oXUTVjw2#Y$=_e<d
z++$P1q~&kbEGHf^Ozj8v<mnQNYr@WDdfl^W?_mI2sr!slfFggdPhmv+BZGZ;lJIkx
z{R7%Ip2wCw0LvRTdWNu^WXI|{d01|oe1FFt#LKo|ul1-h<ceSJ%m#F`4<SNpvLwa0
zZ;)uLc6r0-To?^S-ZMHI&A0h;Egv{%(nS!6WMP4R^m{v8l5V&N2{IYC7lhEAf>!uz
zCo{MgxIyl5cil(KV2Eh*J^T|)VIBwONuvFD1eCpk(ZVz}bd*LVWgb^)V~feiHe8;a
z9b8IyxCgr@V%1;8;|OQ7h2={qt$Lumw%xv7+-yB#qi;<kjNWlkJ{?s9O<+?>88D0`
z(QO?<oD$aF9`7&Z9F>}55k}=~)nxBXj2XNA133b9Q3*d^vq#E<>b$ZA;gK!UrARQ_
zQ0m^m1ggT2f4>oB)DSl`IUf>td;O`#=g^tLv37Aibv1g38rVxgv?+pG5vEP@*Cpe>
zamxKwE+{C0fJl;Ta#WTxXCHgQ3L^dv%tEN2Nh+{|m5IgxBdS7iC7$N_o4`r}J!cF3
zYRc5?OFK%hIqTc!l&c_d&E&Z}&aQ2(sM=+n_v;xmBb<}+sR}jh<#0uC-kqTBkqZ;m
z2Tcr>&=99%^>^N0&~s{9?z(Mvz3}HTh)C;xOut=t|9Lk5S4s~(+y8C5{(mnCm8dJk
zZL%W$f6it{#(Gn6NfOe%kNIMm1TO2P8`}9W8U1nAp8qk*O6=&I+W&r=vKLcx?BvVn
zSzCz1^<})XGoi%c5#fC^dpNOqe9PW(PioYb#NMnZ6cVRcjw+#U*A#xTGf<l>Wh`?0
ze(rzFeCF-`el0nZp~m^W+k3qp&V0hNJ$($#yecJ)_HE1ZZGT@rcQIp9%Ag)f4&}<4
zRf5Q7+QSDmZHbo@KK`B%dRm;FXsLa>v%Sv=_qTE*DJ0U<?#VgOs|!zUgB(Yw((^?H
zOJL49|6McZd4JYYh}EWfAP25H1L-MbB9p<RNMOvtbd>2aKdmaH>K)2^zrowFc{Eph
zbbNkaLiL1go(VRN5Vshq#i{duZLP-J0%G$cV`uX0J|iwec$!2?OoDsys5U9(i2LOz
z`UT&dncDuDN7d0#qcK0YY|QfOEsPJOrPm9>$4ooQN&wMD+U63e<1NZnBPN3>;Mk(X
z2mak9MW?>p_O;@=LprdtS#31zYg>jINRTk9`H&;0Y2AFUq{!W9cz@07&e>y}HHPD(
zB;*o9s(*auf}E7S=iSuTVI$t0J0}EHBG~g8Zg0EE0xZUq5<z25T~=#D&m1uPNes&o
z$`SjK@@0XNW-{Vz(NVa*SpH&v**}oCRdd3q*J8e;L-^x+tU5E`_DC%BO!s)YwavfC
zgu}xP76c65S6o;&?63RMVHIg2=7@sOKiV2b<8i)Oeu^3mD1?D#VgkP55l3PD$oBX*
z0<I%GSieJYF-Z+8_4vJ>I$8(*1#q+tj`#?U68f|A+FA){b(<=uoQ^8EAK|mOIz2EA
zq&r2>qH}Y=wNFL`D{b49Q!0_9H>CZ$XylZ+*VpSiwb&*A$EK7lqK?1N`ODbCSVQ?~
zo<+&PP|Nv=vp=*hJSLfzO+=%GL}w_Iy4H>PfH6;}qa2puQv3w-u!E>uoP?fT%7~ll
zt@xmjRHCEw)z&DE#G^4dYG?3--3-uKJE6VytDY~C8IUdTAzKo2*QPq?LTr+?8ai;h
z*S6njoAfTjkSqz0n*fBzs;*X3a~0t14NF{Y@Z>1H7eZR(>f0*swYk)$#%&icb*R>4
z8CQXT;kw<AKF{<`QD||@&<jE;DiscV%8AL}5-If}I-j-xC<Q^79_oygmfy|-z-`2@
zjf!`zmC8*7ee@F4(789XO=+5r;yQeC=cj%&3@29`@S@=KkERv7q9QbzvgF+@E3IU{
zwOQ08pDXrJne-zvGCwECxEO#3edHlxeKYf2GF(pBC}O7H+TXAme6T;ff-n=^;9f2f
z`0|{L@SYC+hel85$rvpDVZ=d9z;?5KWV)X`FdYbLi1gV)W5YAG7S)yCtXT&Oi%M$1
zh%U8<n^>~$lT4iZ^or~CEVPfcbBO|?fR0nbOd;oBHSU#JU~zd5;EI5mjqc?Hq{&k&
zW~|gayiTEJYN~`Tn5EJAe~;Y|6&W**_o&K;>4$8H=Pa+t&tVy$GMH+Y7{*(@%OhzD
zo|@$;gT3M@_`_-Ec!*ItxWUur#!o%+Yp@uJuQLh+bAdD#%n;onZ%}|~$CD)zp1|)G
zkWZ5zC_cP4t>cskg<%wetnqd=<(3Ky{XVZ0Tlr*l-%gAS7f<3QKOzG}WF0H&dmEVS
zd^HM@z%q*H6fQ!LT<*&HmoKD_&!Ty-Xd#KD0z?9*fACg;=ioPKI&Nhgnr$RP&-VgE
z8pzvB+=?k|nXu@pj$i5vw~CDL16b_c6{vp+_QhtCJ|c3h<mMX#d(BQiY_ZhWZYDM;
zanyOmlLi5>TL{}jf*>sn^qAv`^3PhHbd`C*pP0M_2HX>}_)oTvm>@3ByfZB><L&u3
zJWAkbzuc0y_bkk@;7dX0JpnWYAOsz){{g%o3|tuyiNl8mo6t{!XLy<zP*vOa0iy!)
z6m8;+-qz7$9jrk(a$5F)J%vSE`;VKP2o;B#$9|?>>NjIBsg=#a)<Zy|#`q+BRJfKk
zEZ!<+iAZAR0mzy;?hS6ES31Nj-Xd7;uDAm(#i|3=t<r*ZD(QV3_-R-qWP97gl`%$<
zcTc;;Qosa#&CdcJK84PJ%;n<IqdpKRp_z#?y586e_<m)r0}<c*_g0`q5vSiQE}$a)
zlmtZvV`%k{DVP@XG*08FpQD<Lah$BDDJMY$%MHx1keq?GyQD`s|EB^9>6mI9|3^Q<
zkP&ldleQuR$MWNHnW0=F)RJ*cER>>8PqDDT;?gdB&IgZcnQR1P3>$n(S+r?*HOsER
zAYyt{RX3)b)_GRZ<pe?;lPK*e4Dy#6jVMWe`2pHe&)E%(nal0<(PCs?bb`r@_%$ub
zuMZ&wCYvQ4%~qPu3-|&q_#Ee1`UKKNAXoMpI(SeO?6@|CFf=ASo1HedBtd(yW$g5N
zKmvjVg$rMrQjV3`&IjMkyvN#Y690nQPrpt5^`U0_`MUZRA+2bGy*BUA&WuxWF~qzh
zYs>Fo0cv>7jno5;r$d5kJOZ-0|82qmdt27Gq)va&=BwZI499%W6mK7V5CNToz|poH
z8t>^VBj5v$O<r%u4t^>|O!gIoJ&!_Ip5EvyIZpA8&?t~`FH-ewbFRCizQnNNj86kj
zGrg`Y^e_j+_drQt;u^0LaL}9S@fzp#9AVQrf41>*v%Ip_lm@icygur!FbU@>iQkFw
zqj^aWhIWlEq)##7+SOjC&}CJ`#lQD4P_93BI6Is1Bhu>y-uj)dgrlPVzXZp>SYdhs
zJ3~t-Zf+>L|C_YM@PDk2|BILxl^F);L3)JH{hq#qQP**n4sF^rg@H&}xB(HI;no8p
zk?&vgdMMf5HtudTT73hZb&s?T%dR7T>PUHi?m#6$GBA81(Ykmq_P<?A-|HAqtvgp)
zMmxR;Mt{$s7<Y;&10VwnnKY%B>@Vw1_Z7zL)?Kd%4UiHAhdwL)2~C`L@xVWrs<e9m
zXp1O@|8sr+kvqche^PAz7wcn^7gwW~w?Ad<7XSt@A_v|7kC(9hKQ>X>!`_5|PR`I$
z$=L>qPL_ayp8kJyo1>F60S7Dd|96ww82;A=(-{pJJ8U+Do=Y`^Gg6N5DZ((l5X63o
zJu-f15LcjP7H|3M7~;RFKo+nhzrA+kT=SMDRy~UTcn8z9yR$I6-uB{~;~JPIMSoI|
ze@=_XJv;^`Igtp-&6*yPQ=kPrtWdJZ`@$nTAS4_T<9PI0<E&vs-^2Y<mGcXSEB0uD
z^;M#&fvQlZu+>>Lg3HINbxlN2Mrbfdw5mk{jS}*?IYn@ep!vtM2olv$qP?*TNU<7h
z2-_bhgd#-(4{(N;LF%M6VwF;(g8=HnHDCbvCB((wN}3u0Jd6{cBlbp2136|3S@*^4
zWl0>W=2JlwrA_u6?-`pMNAJe9>@C&xGYBF|(?iVEii@+I8iOGKo86<>tRorWAsPO~
zB0B4CNvgL&A6QO@D>R!1yY-7eN)B#lqd_72)=xB^RHyINg&~qbiZNENDK3QSal$_!
z6OBz522mFqMr9ke!$Swtc_Z|^MF&x|13>P0$onHiAa$P(iW5mBQA?79IJ!Q-MKQ_R
z=kF#VBp|S}hl>B1Kf)w@${%sMM^1Lv`vUp1BSnyCT$aX90<r|Y&oWSmjzqAp!)iD^
zAZ*GI`hQXOPF;enU9e`_wpD4{wr$&XrES}`ZQHhO+j;7Y9(~pO;_JKh3s%IOZ_IdL
zUOB=c(NP{mrv_x{^Y@S%R>D}^(ZWQad7)JkNb6PlMZ*0ip+<&);SzR41Pj2+vxlmH
z4B@f!=&{Ce+mK+jUlc)0++iAsIQHR84<Z1T44rGl-~33B10aH}g-QRh7)QZ*AI{Fq
zmA$`ynpKxW$6a_TcBaf^c&t(sUBQOEIV)r@kH2gkH0ab&b$7lM^km3tPj{Ev{nW{0
z_6<QupuK1;uAuLhD-HLnIa@1D9V#mLem}GzL&@LPx2LCvfzLonmgnOuWFahqEOq<k
z`k>N1(~IJVLdlYF$X$;iB`e?~-;Xu6%bcoCuEQ&MS9Y<Ug?^1DpSDdr*u2uib`vMB
z0*cBKNSDTC^Q1I7dD&Sv2yZ(5S1ivnx%^oZQ;0WUqhygaFaFgJlyR(&CIC!Ff3Kk2
zEoAx@`6YXN-UmSqWG*Kren`9aK0(xE;;p%S6pNjeh}C_n^>ebw&DYEJk#el{O}u#2
zf__ilYKRls7*JmID_>12Q}Jk4*)%K~=4Erz@UCF$-ZEapF{f3ScPfi8WmSm(POe}L
zrkFh7O+N8xT0x^2gDlR&j%T@WZX`FU(#)QVVUv(C0guWKz?yQ|tk)WC1YU22UT-=5
z*#t2BSc?B5%=FEi8!g9|rFVmrb!G1#e<o-H739>ar7`U?J9VloY({nnTnu#UY%(zt
z_ht*R?&uoMUB7m<E)Rh5L`rC#LP-xnPOym;w)B@TH*Vf!i_f|8t|qohJrr1=9HpWD
z;8BP1R)@J}P04=Yy#Y&j<@ApS4DXP2@TC2lb@Qytp0sCd4&DEkzTfe@?v$U+i(*?s
z*p?=zIqAZkHB)Hw{X$pxYn@&Tpzz|^%Rz-Mg4Ic(yK+a(?v6i#cF13<4KF&yA|QHE
zK=ubvBS;7|akW4Ey0Oz2tE|c=cjl)$CN?I*|KhD7ea38A<N5rl`^)pb*rJElx4L3;
z^1dN`>dDJNM`xNYHJLm5Hs#rnqo*fRcX|2Rxjk#9W>hWFw4j%{m5pxyFBLnFX(j7f
z__h!78TI+w8_J4XtFKquMjFTq-!w6!79;6({8u}C?=@jKg%u`%D-bT+X^e-t^ZX@_
zAUwg~Q;U)px_Ohf)HMiF!f@R7FU+c2pRJv8DDM{W#5PF&8l6Bwv21`m=a4efRh#Eq
zKHd6FbH-L3Z)DDMjMX(V?rV=u#R6R`n?5BxJZu`Zm7L>d?8uRNJd>vF8pYpfeq8jD
zS>@-=uS0D@e!3m+oLOv)N|w@8%fi1Kcy2y62pK(Q+4NYqAc_#0w#070HY*lZT)N;M
zh1uF1i8JQ_Y?QGiGj)=<S++Hvg+%CgftB+Lv0*LzayFrv{8yhNA=~_-!r?mbdGTI1
z`ST@VPHewjT%&`zoydzRYbX2lCXK(t2TI|+va@$0ZgOnl-*-RbRK63rDR3L&5}Q}5
ztJd^XsUNyOJre?*D6Y*5T|1Jw9Kwl&G+Z@8%j0@rxYUyD(Jid_`XRbYxmI_7HDc|^
z8au~jSkh%|d+16h8To!2zC4(0Mk4Q1C76D=^`dGoz--5(crs^&lh@LzOQv6^6*4_>
zm^TS98;M4!JF(3vEOw~x;O=~o@AT>K4D94Wm?QdC4qBVMd&M?iy*GMFZ+4FaeRlRu
z-y&MGjJ;<Qtsl+}hJOLY4;R1xLm}B&|0AKzz)t@^v%Q<uwPLrJQG9M|$Fv$P51(n!
z3tPmS{DOk`o&OeunUdvXm!%sg8Iw=ep5owhB~BG}il`)q;_SxEZs$(_UykQLkfGV{
zt!<9>uf8y`BdJJ5VWv${gE)FfF=?!A#aXO6>ea<;W^M_|Jbw6|=14ERi>F)iSkma>
z4AQ+TwN`<y+o#a>8eXdIZ=Q;A+PHBDHpB>t{Xx9FXEEj^)S4B%=e$R9i388vFZ<fW
z>s!1xTa>Vi)-T_SuHc(`hUm8r^L#HyczD10r%ikPU^_26=gkL|0zhr&cJghVa8}R;
zbrRBmFsOMlMX8FsgbB6EyJ9V%x-}kCu;IrR-WBC|Qq1nk^~iyvc3)|>Vu}=^MRocF
zD$U#Gi_msHd{a0{;sZzgDPBO01m@~SGB9H6$_M20&)ix!Xr)TeApF-u#}?aAjrCxf
z-c6KwYYW*xY|M7?Q(i?vRA~q?UFfJ+SX36_a{BL_bVTOc2t_#OlFj<=ADtna_Essx
z(0W@exq_Hi_;w|zL1TlI2Hw3H3BGbB(g?%$IZvar05$OvqMD9rL>ZZHgr|P&?Y2V9
z#hm%A00NyFCOd&sX!{7ZghHLULMqnIhVI5=nlh@Ouv88i7JmA~5+u}p1tL@>Nf{FU
zDipw*1uytdQkm@!fJ?YPAbb4LHmh{P9n2a7m~FI@!XrftP1`aMEA2#$_Q7bumt9ZR
zoh9eIvLqHxglqP@F~@wn)JyE=ayzG(syuVM(4Z;|QW9lqaS>aQ!be;=j<qy#0i1eQ
zofE9SHUdj(mlu`_0H;7ps6+3EG2F(h2l7~|X)r;~O8EOZ$5EHu(ly=T);u=>+mau+
zl2BBs>5u+yS(78!Q!UHJvf|&(#Qqp;UI{Vz4yr>yxG2T}xy#k5a{Up~bnG^<*CB@I
z2_$AE?8*JqqaD!4vR0Cy&eNkh-2x=R5tD~&*OT=-Niu;Z0%PuVhj;Qj(Ai95i6ooE
z0__zI=bD*bCGH6P(q{jF-9T3UD2W_Nb5_TtbJR3Pj?arM{YrHX^m+{=j=cNle@KrL
zj^{M5E{H&*AnPp+wN#fj7RuXSRb5UTbe8{CE6P_!@s(V>*yHGkaG?5Wh*qKbi5TSb
zTrp<$TS)T*A=628?MS5OEO$hX9nK+2r(Rh*sWZebP|ucFtXk3jYgv-xeRI9`ldqa%
zV&l-+SsKizxSfT@LhBld=~-S1^T@(&$!(f`3=bAFc**s^nW1z`5MT*a)>8lS)uI51
zmcmtnApU{GHA_JLJ3uvO9{Z`anJWcdn((EVyK>#41^*YBXo(+_B8jWzk@|>hqWD1X
zb0d3-pQsQ|3UVB#__&4ZgW;9~wUg*O>umol0$~PcrX>0mTx?vbMymo!>BBh71VIxi
zK)!T9sJo$dpS^VuCwv<CjL{fE`j1M#y526H&RQ0D*d737&)fk-r|!og)VUhG8V;Ze
z<1s`l01IeS(8;Qw54@|A9c~u%O5l#)C5i<&&*EGOMX3Ks*T|fNyzva--saKPQpjoi
zQCdYgop+*@Y>RgLBfCxMqa(-VOiK&hsQ&NjY)&wB?N#vic#2+1-u5Iq1TwG1*ZQ)M
zd6+i<W&r?M5${VaJ~!@S5+*-ZB&xyW?3w})G5y~kKG&p-L7NixO$Ps?z{I1*SCtG}
zGbviN@RJJ^(>3xzZ;=^Bgo=zY)U2?k7=9Ff`+qAkd|-i2DPl<xZ$JgYYeR%^4QBmC
z5nv<?*dZW9X_MCD`g<?kh1hF2aKX$qs4s-Qs*w7=-Y0YrO2$Kwi-0tcuafT`|IH*n
z%dKg`{bxcH0bYX<gDSt^#1krzy2Wr|T?b3r^py2=ibtya0EZ{==)}LAYPG`FLcPE=
zZr<&BmdspUoB3AMf0VkKU*;#vCw*fNaU0-3fu9MA4ozf(j3K78j+k<_wJm!?6z&s}
zaJ1}Mw4*BIP)t~g+rgA+{h_xEman9x57}_PH*JZDx^lgWOeK1S*5RFB#|&Rd`57XG
zo>Mxuk3k5FZ}G}DnIOnF;ad!tI?=AdYzK)l`ADigJ=H$M;na#I2GSio>iu#y<1lmu
zvI5iRLLkvmf1L3Nz+_tA?^L|nC^l<`JEgyCy^HFofaU3;p#c3;bH3El-cBEYCv4h$
zF)+$DNdh1>1;DZBa--DtKIdLgPDs_!ooF4-MVCRrY;V;p$vr5TZcIh!D3qPIiHe6I
zkBkPp-}-xtayu80In*JjD{|L_i-5~BurPla&Ez@vX;^mDRwg?fM&IwA-ibmKf$wxg
zDfajdb9Z;}z2uwpWBU)FPB&j$d3gWGPIqiGk6tV>7}URpHNZhi!f}U`UVZf#VuXG=
z0r^lu{m$hnb*DXlPbdBYwEX;A%b3_2J2^X=7})&Zum3x|GqSV&7c_{8>3=?fYS#FF
zh}k{8f}x*8g@{xML?kb+&_@lOB5<2UYZ$BWn}Y%M=j!YlNKXmO-~Z@@@(su2a2@A_
z_`(Zbn7mAnQ!vQjZUQ)8PbayPY`eZi^-w4Z3FG?hkZmx+L=stH`t5K%LEL|msEG&<
z?+!jrb7yOJ#BynCd8vs=5AHrvkB^_1VO@E3eMd&79pg#&-<Lu4zRcchZ}~{S?0<#h
z)Wz9%Y;qwC$hyP3g$LQf*h_;ZsKeAH8A%*I?4o2&1K*TEF|Fy^w{sMc2K{z<A|4!7
zo~$05SGHTcUtP1=o}v(db}RACiXy`maa2bEVKJn{E7S?d`Qys;+Z(wys2c3s<++!B
zS!i5Yd{mchOAA+)I_9%A{=pDiaa$^2efIA_Rq{1NRDswbXA|w+v%rYZh=FO^4m&Ew
z@8j67S~><-WcdvLB|@@iq^``rzR*Ui=-2RkDer5<)FCAooIx&c@hUOZ`ON~Nk+Z)r
z$99ijL!_4gHzJq@G_ZvIuz;U>WNf&;J(Z7&!=aRRTsGb8CVkwjGR%}yV6h@Go!#S%
zDG((ZRxuLyqwDt=0O_%CG*m=-sT>emM;)%oc&#6o(Lz~O8W(j^Q{BoKkQ%!e5~!gQ
zlo3Ym`$D@mX9liQjkCxZG!R8PZYv<<(%TRE|0;_UEgQ2yn7X={RTC%cGrUj@w9r1l
zg;K}SywR&MYIkf7?rm5*>%C#VcAR~??;LzEa|Mc~xMP^Yu*CbZ$|yiIg`v4xmzK0v
zYT7l+!dUNMK^I%KY*w7(JeeQ(XrOsQ?*`CS>kE2Xo9`IwnwhyIrs^~;n5djk>2&#d
zO9m24gc%z$(`4Kecl`}|j_s_l`|gt>?K5Hrcw*bOF$9f|GQx1|6QfojXbqZV9Iyyj
zh*doGO>?tSVlyr7?4sOcx*fa5M2M&8-_tVR1&Z<eUBqsXMwn;-Od^A;CZ5~fn|pHM
zUww`ONI;?(ef*>Rbv&{edOlO9@~pG%7MSlI(#ceeIgTc{<hJ{cv2;EUuv&6AqI~8`
zusC72TWzp=$W<COcGax5bHuOfVMYKZ()j|%4#UaUqi;CDhymS(L~H`iAf_;>Pt$Vu
z67jdD_45GU`_j)Z#fk4&P%#@F8jBM>5K!j>bD9YH(~%X#A+IdRqw(q?dvt?&X0mc5
z9&-Cqe78m4)XZ~vDxEjYc4Qb}R;x6n7Ojv&+!zXT%p7SW+jI4H4`r5CH~M8jE=*uu
zt4@tk2pX(uMFf<=d@+^kRGsdO)+h+bQd`saz;oLKB|f0G^+Y;YMT7fbO@ClbXj{+-
z6kuJ|x^Q;h4mmu>Tn=CXB~}mHis*O`511nMCJ^jsxqS<+?&NyD%e%?`s(bnM?=i=w
zBwL}~h~Zs6qP*cfH>fd`hAy`y&aub9dkS0dVgG8`3ulf|f6!3<MlRjKtGk)GV%NFn
zET+k^|D21?n{-u0zuW^~oL?ylQ$X7(apfkNTVNN4`kPCZlZ;;1#=Gv>z!v<HICwv`
zPTB}Ql&^4DUOV*2i&@<B<Ww?Q>ia>o)X>gU_3|03-n5^|Jx};BTjlzfZYF7K5reR6
zpUXjqGE5J&KfCOrUwIPwP5SLB=-c`GY2{>Pt`CN^Hg}`H3{P>=B5aos2@o!&#DIz&
z2Qw@zz2iEE`JS2?+Fdrp)rnoyjYsbQEVp&#%6;Xe=las44RLDJ{=RB*pDqcharhsF
z;}-uIwtbMOM#n=H-&PTs{xl13ei^SezmQsvHy?|hN(Q{GPZ<{3yF$0<o%h0hRoY4l
zwG%rhoQ_`5B~3snj(vX68?;jq1s<IDdGjJ<HZFTn9TeYiz+VaYl0NsZX3<aSV5T(<
zD3CE=(x6FjTw&f%;k#UeLqJ2Q<1PT&H0wJU8yq|-!d3JnnbmW&*`bb<*t0u?hNUG?
zR7Bkm!oo}<vF!UhyoYyxa*n>t&y~TtVz%NHJsZ-A`k;8`7loI0Eg^!LHwn?SEp}i~
z%6DxdX(I{7XHC_BD3S_GV9XEbb&yH%DW+GbczKVzk3R36<fwLrrHa1_dv<lbuOrW#
zttr)!M%UVd_vpD@-B*w!Lh$EDHvTuWHUspbJV%c856Qbj97N>$BRF19i_DO6s&>0r
z*LcoKqRkqY4S{ov&}8DCeRujW$`0bpM|iWuuzXkA<uEjzf09aQUhZDVEQrSk^uPeO
zurgP9_mMgqeP!-K-;lE|`&w%_nT`$;M?s<4rpYBxuooOk<xLWsTHGX+c<Yt^z0Dh$
zIft{bI&V?jc~^aaX*DagADZ7S2AS=nir)o#{}B(VD-H=Q`Tlb0efYF<-anl4w-ML3
zp+wCuH@oeybovwzdZG8@8qB2lq>|1Ly>@~|P#(HWwfrlJl*dMCm|${l4u>O2%nZJ{
zkr{Y>zFG0gCoShF^x83Aefl<zEBt0T^A`uSG^Z`^t4Sr<2_ZF5?@8SlGZ%O75I03{
z@v5(6XXmfM$M0~jstN<bWh48@Qje*^*o(%<3%w(ruChsJIln(JycY1wKjf-D5y&v{
zi%!A5=`};^pgT`57|J0dGTUrVeH&VY1%>r@7Wi;n^0;;1SG|2rKNFxy+V@6Q>H2YB
zG7a3JR;5j|w~m@jXyXIiQW7KHk7vGhMtrTqR(I}uao|vCj;|kIFgMa9?@!i9hLF4p
z+SQ;5>$t0~-K_4dR7#@~kh?lcQRDz$;>CdgUyuAgkc;A7`k)C}tOArIKX1C8;P`!l
z1cL;~+QL4gpU%c9?j^LgVq8nL^(0y$d&7oSW~A;-tJv3dJ6SlPn;htV+!79+dn0#m
z$D4=4_xZ=8l;k)B>f&%o*C^RTg|i&mC^KZS3>ZgEE)N1Y2di#b0P$N*P?FW#YoKS(
zu{PdOU(4zK-g%tY?x~eo><il$4BeJ%LNvImwT<m3m-UAB8i{esp;59H4+?&Z67aWQ
zT}s+k#`KZt3!hkABJ8l7?uMDpE9Dk$Z@U0VUuY;Oxd0))<%Vw*9cSroJn@j7Pu!A?
zjKb_sZg|l-J+z%0aWI3}`gaLk#Ql8y)Q{H1;KTGUfCS-G;(wVf_Wvr3q-X!1_xa5j
zla7b15B~^_s&M(0(ii}c0a;n3;bx3lo$Jb?<v5JxC^CX*t0?7hdAPT$cEH<l;u?qd
zcrDKIhs(EQ?-#biYPWPfI(pi@Qi&@6=1tSeXVYY3s$|osQp@HVwO*0(30hyh-|vs?
zz2U12`Gjz703_ebPW+vprk?kA#*_##i5Xg><omPKSbQI^SkFs*TZFf^o+XQ?7cp(`
z*IM77r+xf81aWP9L&`6`UaBEU2vu?t+-Y`%=wee5T>^_d()EMyi&k}exTDSum_nM)
zj`N$i94!^d0aJ_*17VCCf;)9M+$}tXlVDT!Igl?elCFd{ea-V_4{&ZTK3H+VqxqUz
znMnQR+ZP$F4JkwxZ#U3PMj^Dbt}C3p!f$)X$(|g_Kr)gWz9b=|E99GJ>@V=1Z<|>z
z?T;Z;yAcP*KRKH;@=pi_Ebt|bvUqg;(m~90>QF=;h*vA|st<j9e%QoKgUi)~ERsl5
zOZ=L9a1`=*sAp<1tu?m?G8R&T9{Q;IhNyWCn{qVy7&eP7{v~GGQ)w$1kQxMwD!b9J
z80(Se?FQ!T_F3useNKH6`m<b9wKITxcBcJpS@BXzB}m~%w)}~y4hdyv_4R#q!@+9V
zz#rY9T!DEQ^abk2eymd5ngbNd{Yla(VTWtwG&`9cSJq11Q_0fEDH6>?dr!YWD(S`2
zq!-l7kj(RICDxh0X&FF}KGNmO_N|<Ke7(Ky0!o^IKHWp`x=4lZ7^hCiqzV3>2gSG^
zvYNs+i<m*Y4mnBYSsL<UY8VBr-P;E|_=tDEZ8nq}j5>cekun-;uo)qA{}k#?I|AnD
z696G-!}tA_PCx*+KvinjE(UV$S4uKEc?As`tbT>NYd<T`1=}KyauaF~1c8i4_dYu-
z!s)wG(t=CnbZ{i)*3n1VbI7S57!BHWIAdm1Ka>?ZyE6Kvwr(?xC@{6YQsx}aGFKN*
zsW8T*J|J!mKno&m!h8cwWAJAtf*$x@p`d$)EMQq_qID~#?68@<!Heb9RCeWD1>YgU
zj1@~lP>Nu1YdGI*@g0&*!b=K@%*H$n-s>~yXK^<JOP7m(I?jHVZtq{)*T@&zn3FbD
zhVX?=VkD=@v%t3r7dN9*23i%8`0~Ju-4~sBc#>zHwzbPma~gCSTgdVc?8e(fbqrBB
zA`swG*_G{tGl=nR;p?WJ)$ut@#zdD7h(u{3ObhzhFW+Y0xyM3<0bkL7n8^6RCTg@1
zT|oa%eR~BMFA+u9-UpejhQ>3#+xd(O)X8Z8v-J3G4gQvGP|Sizk*fcfN*{?no{G+I
zAcHgmS~|}*j=m`=w0Htg(OV<$EJb%BX2!qKv?+U2Eq>mIBxfyYLcYm4Vf`g->y6P=
zTEI4=@O702vqeSPxmUnz_V$(Ijxj3)=;BCSs@KPeu7v_|Zv2&_AG05%fJuH{51T`}
z$noKss9a@lXNL?QPA`E#ATmlIn?x|b{rwhO=A*{_A_wE}5(C&6=VEhUO4Ip8xMA70
zU?+}Z3i81`2UQGuiKQ3BeX`k3sP)V9mtXI6KNqBbcGKaznn0+UHV*|8grx+$3G&Ug
z#-&#^u@YU_FrUf<<%xc3S-BZyTV3s>bWJh(ws+HcU3TY5i^h~zOUH=#Qf^rl;HY`3
ztKM$bi-%*d3cL+Si7aM@rp<j7VK9^TP3iwhx!__ivjbI@uMA8P0cWj9M6Tedt9PpJ
z$%6W-fZOizl%e6ZcdL1JCThnD|B{%<W_R~0e@?>LnV9pK$@EF~*JQPYg>tnz)^d@l
z?$Er;G|<ZPg0%tL!ZIMoz~a;B17x|yGzPVYpRCJ#o`$sMwUR9=AQT4o0SMO1KuENt
zT~;7g;6EKLe;iETYeyqKG9YK;c6Y+vX!iM)KFv?Qw1{dnZucQByPY!F5CT1DfNZfI
z@z&zSBx+!D=G8utw}%)tU7=@mR2h5yDpCt5@L$hPOQ*{#PKF2B6*7*IT8d1csdkHQ
z_AlQo9lrC}Xgrsu{l3(mO`*2(m!FtL*g?3(x=L##!LGZhoM5X3;I)-$H6LcAQ*v*o
z8%4a-7G=3)KOe{pLBJZ*2%*2xM=DyO5$f#Y6E<4avc#HhXC&4Ck^l)7frq3m8=vi;
zqSv;Wyw_Kh8*lD3_F(fnsDqdWa&!Fys|2aC=Kdw8W5^nL&6Y;bCR=+-X*2_ZUR!IF
z7bnB#l1mqrZY+|~6IQ3Mku!yW6beUk>{CJ65AY}{qUlq9jx<-}nBKGHGToC0Vl<#$
zfcm~aS~`=Goat*sq3Aq9<aH_zKLe5HB%0c9uX3=<*f=;Mw0Jm*E`%}XG;9C{2yv3o
z2}v4IJ4)y3V(;mU^YcvIaimx#SkKIykOusdg*6q+MLtsK#6?lSVt>LEGD1vvf++;h
z*j8{%Xc1<zFgIGXj|DXKOk{ttZ=~#d$hvTrm?UBK6R-zN&Xa8`5y5Rh<QzHqlr)$Z
zETV+pvv;%)oDkyJm}`EE7>KZOMxR1uX<=}ivO->B=bj9wL?c|Ci)&DND+{5#=?K6F
zxSV~e?UVxjQMDY9*;FHWl#bqcUg261oywK+ab$&6C1h3o+`s=-GwXxZ?tU*MHNq^7
z?}XPwRG?S(7B{i27W_Whkxw|l+T#xJEwkSA2T~Z{8*WGi@K2%hdoa<21kAA~ObV}-
zRlrcs#rwXor42iyGO)yZ4aEC`da#74#wgrsPP0b-UI6U}q4@Aj$y(;~urJ`#<Sfa5
zc@B>M@Ei=R4F5k;`x^g}p$I;^dW1!Ql)$WmN@59MYs{VfF9y(89b-ZXXc07<g2v-<
zM?9KOy>4TeoC&L$z#1qb*Wh?qaAqM#udw5RCVhdcVsxcqa<ZUufl0;jQ32;f@(TLS
zLv5`6&Ujt0Dg(sWGSWkXyEA&(+S;;W^>DGXdfC~%crr!%`M2*H2A@X<2dj}%Re?CF
zJ5h?5czRibWs~!t=X7$i%@?y3u!p2Tg<C}%L#k(CiP=3cq!Fdk8Tv#<DMg5qy@?Aj
zL%=f&!UD<QMB>CuKv_tT$egexRu~LA$1BvxDep20H1#=b)8a)2+f8dVW{oz1%kL>i
zof(p~`kRWyy80T*Y8o0-^$c1jGOU-5wGCx;Q*~>5L6f9fOqcuw7^%5avPEz(Y}EcN
z4yL&)D>>r9?=merN$+bdNTslxZawONmr_U`f1B_rXMV&Jzvuj4gg~#{CW(Zxi0f$M
z8!m~)u+q&_V50we>0^4nOrqe>Te6S5!zacMt7#}1<4ZQ=O{^$)(oP^9GQfn$qB1EK
z)HKxa9qdp>#&WEfYq+<SO*Zs|*w+Pqj&+U~GPk%s>Ju4~{@Hj;ZB<F#jv_@oWldm>
zXn|VcN==^Yu6LMnPWs6HG~$cU1u-lwPOb9{Xx4F+PX_}(e2bh2O2sHNF;O^HPh4WE
zoV@NB@Q4c5`TUj(SL63IJH1yz{|%SV8fFZR&}WpeUl?!w$*M^b3CE*Ek`qHFG?@6i
zV#5BeC!3O}9p#90gyKdum%n!Ck^Q^hI?Kd1LayGI|Fb(E4y>1y`)-GFj*adco9e|r
zn=S^7?j8%aroFUj$-9;7IXYf(K;)0a&yM<<zm33x5~ad`V;}^6JWc5~N(;qA8EL6%
zd&<Fj5*!%ewSua)q|&4g#6CqcGDJkpxd|F};*#{2brj1`@(I$(+_^sQw$h9KdP@j5
zHe^&w1Tj)MK(+T!o6~p>a-;CDu7c~zJ|Ektbda5@sn0cJy8kU6!r>-`vxAimU3cy-
z8vv+GwhIkvy2XNlJrCdBED#@&@l`+)6V@t^XK%mBD5%(#OUr=9Dx9D`t*h_hxb82^
zNcb;aP(9g7t)~FY?53JkXQ06WK*)W)o8JopUVe!#euF*{vzcbuzQGqCJDNm_<#)A@
z_e$mMNmi}>C2b?WE@maTzjoyI1)kx%0sZfz4vy~LxJ-IPLSPo5(wNw=P#U-=VnPbf
z4+qV}my(~og?8h;zS0if20DZ`yZ{!fNKEJ$iOPhb(3p#9{)+jK?wHCdhz_UrYFI?l
z!OEqJ&OZ87jLX`|lS86EU7Fq{9$(b`<#WXyXJ=Q&E0Z;D4I_Bl_d8B+z`CdEpbG!(
zo-5-lTeb?N!O(QcR7Dll#RMmzk@Uhov|<A4JqI6acmm;i7{$7)rR#W$qBs*0BoI{I
z7?=y)+?2^qGqZ$u*x8k8+yFfx1=)^uC*E<9f!3Eq)A0Zl&ydLWbqKHg0CaARUyFfm
zn{Twl3X`@-pek|uWZ38raDycltH!*CVNv;4$Mt+>2zKfbf-5191on>WG{gb@Ws8ie
zW{=>-#V$WgX{H>VP^fpKztEobuVe_Yrp%8>O5oZ~WI9~;l$*Rir(b%&Zf+Sgd_(6S
z7c!Mlr&^+S`SK1d!(DWx=Of#S9~@Tgov$Iiys!60A>br@ey#QJojh}tM)vTK5CmR$
z-gruHf^s!@#8|?hL_#wY4u7WJmpSWLvcZe-X|Sw&e7tUN;YycAi-#5(OU*^GTm1J-
zfcNcK|6%`G;O+onCNz+zgn{v!Zfs${Y;L#%h#Ih<)YNy*$_~Sj^lJkD&=#I;jrEna
z5B23$)Z#VbJ;ZXpB!oWFlP4dbWvi}$w<~T}I(*0;Fa1!td!7vGm-(EkSmsCqREHJd
z19ro@DMA*U)RA3d>c1N`@NioB4P#7n+lH~7qvA})03P}AZ~OuNL3~W4<|+L(K7jt-
zk%kz67-v(o0-pq5<SJEcE|=3^_L0D0fzA*(&0D5Y+E+ZOun6t-ZXFmpyd~ho2Sq=N
zi7#=syXj?9`mHpT+tBDdTpXc~RYPdn@~?WWj7zJYIrVn>?cIth?^~5632pc@o;^+W
zwm1HdiRbftUJS1d|EW*^ak=nMpP2sFV%KK$amP*8r<>Zn>ZPb)1b}{$9GnH|gp$P$
zi>jOiXRsE6aAezKL$T?i-5u~QG0CFrA(xT`|GvRZHG;bq-Mwz^cL$e;GZFIcf4UUb
zwLFp-Upf#5n*X3`bz6;JVSMlN{8)(2Egn19h7?7Hlg3_U#>eB`?eSw;k3xB8GJqz9
z3i^YOUOS^lWc|G8q^(_r=5WTxGrHZl>*M`>fldH|NCrs3@*7~C;{gd>z}(W1sy!q<
zl;0X;mmpd;5KQ5`ZlV@xf&$WGkx|iaqFMl$5Gp2}kme5?_`??$jm$wPHGo*Sxf;@(
z-IJ(6o*q&7$;2fx{ZPf+h6K`H@z6u=Ky$yS(i+?5VQr!=N1j5=K+TSP=dzXh`+VwN
zOf4yBm_+dB#~l!<3ND&<Q)U^(G#aYPrPSeh91&be5HN_*zULO<`j1ae+>dlN2PXxg
zpJByMuKC7M*xQ6a{%n|2=nOoQ&$v)YPzL1#<!mbJm2snOi?_i<&9Keps!I-Rs8Evg
zrs?KF!G2GLOkpP>YA$!rq*f)aiGVGuaLXbIgKc9oratld1#^hB>|c%0*qV4HPtlxh
zg1YgR&fDZQjBL3Hw-xymAkN?>iVXUA#|uhp!2<`hwF3z58xcSyi-U3XdUDuq6$W?l
zAO;ix1+#XGE7}LU25vF;!p0Q&Ji%>3XFt%`?+AOgwy<wJ#6P_oFd&|aU_a6nwd)>-
zAF$-FXd=|^2bl#OErhLp(epa7bFP-KNY-HU9YO(HXK01MCxT8G*D`?^A2r5|0#20)
z%3Z87nJOgXl@DDV;}bMF0tie<doaxv=UM9mk>H^sKO0MB2gx_0vTPj`{&JDztQz|e
zQ$)){(oCRr<bumjB<uiZiZ_{71XnIOx7WIFdr?So1dvn#*9~X}_>7r)XeVE`=h&s`
zj|n2x2--$#5gmiwQmT;v{#FpeTc21TaqkHm^dW|4ce2<08j|lK&8#yrj1RfCoI?62
zz>$HosK*chvD{vkgyF+>;f7O9Xs<~r{5}#GYATS{T&WxYVBgs@Ou-^Ug6Vjr0Mg=1
z1Q<`i099B40vauCQ;jKWLhD#C?-aI#pwjIlUG-1LQTqybvlU1lG|(#I#Z-09%hwCA
z1?eOL1MV1wBy<f6#}KFf*`7qw;>^@BlbEjR9q0KO$Q36;B}o>3h&OP}ux^m458*QJ
z@etJlrfOJ7X3^lOo*yr1UywR*c`Hh1EXaR^p$b(d7<}h63P%XuP)@m@nuJ=pOhFPO
z!o%b^l2t;D6G=G;6B(TQ%q3fDJ<y$F-lg(gcs7CT#Je*qW+96%^B<VBh;P_0@2Crl
zfNCHBrk2=TmJxu~w4!3t-%0!@V=x+)3ZzbR?3(3Cyga=-62t>KmY1t%Al=zCTDj<5
z^A1e`?*^KxcNQ==G;{{HG&ha+_S0ZSlV}oNuIC!7L_K`fC?YVxwE1ajtD%6st;=$`
zKlyjO$bz&D0t0nAvAyu$NS;J1=5dGmlS=)tWew4y7#3ZU)i$qg2}MH#c9j6ttXcI5
zkM%)HCr0{Z*XAs`QRoG&2ihtX2Z%GdQa8EQtr|-r8%%5$&trnPFG}x-5Rh!G8DI~A
zyMuI3&bB8w1u4@?+lD3c%Z&h4yw&wyG<FcvvC&fQ0u}}ZJF`oixT-X@JDgpJXe)a%
zL1Zb>z~wEuT}Vz*sw-z=SXY^=j5b^FH%wGxK=~aS1{_4rn{1r>%fLWT`imfY%EG3+
zmPE4QtI<BXx_MY0;XIk!ZxQT)B+v)b?oRZyMD*AL3Oq2DAOB^;$O>19wT0VC4bhzT
zbx>>kOo{+_{CQF8mF;U8c2167Ln%4Bj`!fsa4b9ngbN^#V0lD?J=Kd9h|gIjkcZtR
zHLiDX7d89L{*fJ+m>qB@^L>e4eY9AE*nspeOmn=qaCb{$rN(qaXJ>C_+0q=BWC~xI
za(*@+Uti6_D7QLXkiY)1>dk<_;1|NrW?xenVh8krLGxGQ+QCj7K9}Xok{rYf!GnmT
z)yERcU-TF;a<qbf_@YTgZLfTmbt%%q@RK}%I;iAs&FtoLqwU5D)8;-u@V}Dolvm$7
z`9Vc|JJ{bK3BERaK595~z4N!fpgUaM4E{r&IOv)Gzd3hChJWOr|B*krts#}T)q>b_
zsdiVgP6-PV0D|9BO+!&n<)FPn%Dcv)FFplB1jj-857+Yb!8`{GH(u<TNX|jrtRjT~
z(t)kl%fQVk%>A)lS$nJ7{f+R*NU4z?PTjF7CV@82Y{Z>BuY>#DB>|OEmsvgab`tt}
zYunrV`I4Dg9q08p%Gd4b`?%V9yW=CCc>bfzNXtZU+gjS-9I5MYfH55P2o%B{Q7Ftf
zE<$`5Fl>RjovlA8C$<d~*tC`9Qmtu4Y3z1k8U1M5SK9qlUE{FTY-gX&vrtjNl`T`~
z>E%f;69_%?Eu(GS@u4_B;4s=_m2uPG1g`y-_u+bJ3@#v{?H3}Y+P4@9B?ZpPrQ;4%
z=75PBWoB~2zdz49lRm}>a!f{W{cfBEj4I3!zBmq--uD1-V5QAan%+Q!Y5=O=>;hr)
zY{!6IrCF>2kur22DLjg5C*L>Hge8VKt*qn3(aetaR=2$<=kSm`YKINYUp;1kzyv5K
z_kAO-JX@hgW|Lj9b*T?W7CcypWj}}X>}1!0NB>IaCMJ|Cr`gQB1opIEdqq?SPKQXQ
z03}#^dm~eA#i^@qEe46}uh(`2{1xd?vtqWr(&P`PO(m7ETe(xFQy5sa+X9^sjRmh=
zF3_uiJgEy`hUZXMcs*oVn?X3X20N1BISuiUK(Y>W|Etm%`JTyg=rMA4RI!f=JqRsH
z6h#MJwK)UE)U^ueBSXCe;CeAQD~$VE(H=$hsv~D)BoUozpu#i!Tj=pHTPYTHdN5$a
z-^*7X@-+d|!qyqd(!UsBgwuDhnK~dIM)bs~T=5slLv?GGqw4}bHPLn)A;`&>lUn>S
zXnv4IcisL9O<2veDalXF-q@{ff;E;(woAM!j{Wc;BRm0z_ZZZ)VAm>l*^4L<o5ef-
zDaDpIy#>#{X7ex!Gb9pkfYZdmHNnk*_h_5>NIvZ|*lB)+{Zih8J>#NnpmRkC)MK#;
z#l|B;wjW$QrPw`8iR~ttwnx*1+$<9<Hxh>DY(gUlenO9?A_@ei62KcxI*H+234%lG
zcuHFq;iL)%z>NNc1BGd?^l!jJccw;u9MYEtF1HWkW?H8E;lB3bc93J16PU3cQ8atO
zJuJJ5+qG5XU}6KF{a)mUOKvH1343TGF$apNvI(r24rb|yWc#x+cl*$(|FvvU@FnNM
zgrNMB@N3YT>TGlIK!ob@t7)HjWb*Q%49;lg-|M67SfMTdx<2i&kJBX4Omg<`_U;J>
zmmBWwDhp8VF-&m$!x^vDEBEFXQG9dH+t<fzJaL+xO+;MIGAGwxVjIg&Zb!%wgx^ZC
z5GYQ)Gc<@I{;|s!c>m1D4xHo+v`oMOi$En_%JP+i%vE%5RNfNGA88z{b!Y`a4sNqw
z^f{U(gYQ2HRx8Er_~=z*EOGoO@GDX2PH3YXj$IYnu(?c&>3rX2FtH8{=unCnRP60f
zY%^$HmHTk1=(Ogf+rTmP?l^bDkq~mhvEl0x(`G0L7|>yM5<PogF0~sMVG2lRDzlIW
z1#U*X7G6xBtyoZ!)Q$=>*GS(m9c=MZXdlh?2;nAmqaMqNW$ccFWN%I>k#w&>@g&qU
z<s6>|1GpJnyOQjKTsl}*Gcx8s(n__VR}|lE;u@Hd9D!OFiltr=^&HRYeV=POtR8l)
zV#|tyoEXxTpd^P6lxq>~z~)tBAdER<UwU;0XRjuj+`|ZrFoqBur?=|1&g{_|e=>=n
z8}n5>`+cNR1S~iXJ)}TNrW*USXVT1uv}Yo~Q&CWl=WCWL(Wn_oktn&#__V=dPOF&T
zUjaiFicpNIrV^*uf+5nCdd5f7QjEwLa%f6t28ZiY`;Iw~4QKAqEVlS2nV`8J`<zP4
z&n8HtW6@6_+kpCx8T_RC*E_$llFrdLEw$S~WEE};>5A-HB;q{^#Unr_?p)`t^(j}y
z+u)Hi9cw#pfch^<FcPE2g{RuLYv!c5<7_iJkl4OZDFw}>@vc#4L+pV}xQ7?Qx%DpA
zYYwyrl8fc7rHr4wS3BTj|8~l;&J{Kd?lf!n6#QtzQjHL)*0sesB{(~smKn%o%mm;~
zDa{4O!+YE*x6OnyYZ@gluQ|4>rRH_;Mzx=TM|NY0C(_K20TM_EIA+{ZAYGL=jEF7b
z3g$iR8<)D4FuWj=y{4|bq9s8Yj1()J9fn}WyG9gr6@=BUp;c4x97mC%+<bWlM)I5;
z_3BPF!4_zLWLI=#L!!EX#^IR*;#oN!3V&YwYmes=4Tn&SGK<EI=coE$UwJ=qzVV=h
zEVjCl8!4ceYOmTEk2y|;>jdXgHFou&g9hGSaY}HX6dC25opdRiAp>K#JKd>Xa#!>2
zd@0N!K$d){Cv45{d8N5)MD(jE@BdA0UDAlrkyWLJh3k2{;2Z3K({)2%T&vEviG)H;
zQR#34JE;Ns3Ye)x68&N9uZGacEsWm&5D%MZ3ZK#di7yr(V~d}+qLfA3-Yv2z{Irjv
zz8fuy`VeA{@eT5OYSep~2ebO(?-3mGUGnURHmBl&H~D!GAH&d!UeK#IGK5_H*uk44
z#3ul>Lmm_O$zgl@QX%00_mxbCDTi%%gJizGM+P`mP|F3w!8WrTIFI0^okMqe{+sp>
zlwkB>Yf)O0SBRN;LJY~rY+cPU{iiDtF(|N%*mOLO9wYO5a)vjD1rT{-f(rX<?;y_<
zoV7cLs*(D>Bm;6b9BUmgt9`z=`kOVeQ5wxaC(Py^qT6EIz>g{!?Ea*GEz{j;2>oFn
z!}_sEFcsdm>jir3vs3LCAZud;;6MGye{fJ3m|58WXFsB56T2vi@ROsrZ^w?bzR6Z=
zV@{=m2(C<1>?ZIHpwe8=#sNJF*D~l|@vzBtslioG_gWxN9OTTonV6k!cIsW$`cU?2
z{qX4J=tvrjo~Z6{Je`^&uS<ZM&v<uI{yLHxA^V&0>*zRQ-sKhrnpGF(ANR79cv@O}
zA~##ggeR{jY|IIb@~o%V<^B$1@pvkCe{9GLTkQLR)SL75vssDW3-qTfO&;7b$b98I
z<33KfUNCyPSOC#Nl})8(ja5U#Pq)A9cRm7Gfe8%}kI9&wiprQC|JfrkGxRMvNhG^?
z<a<R*C*`lMuE{VIAKP-0h);XPQwD)5QYbe>*L|brU$$@OpDXd2(#P@7a2A&5G%}EU
z7`%D{|0CUo9Q6))z^vFVj48AcU6ZckqgW#WPB!@gkh#Awzd-)dz!U9}_%lW5{+&to
zfW&_!%!lT0uwg}+qRTJh6EP!S*8<Y$+x#L+eY4_kMfawSg^45k*0Uh^7ar6^Pk~)_
zGSdr|cF`hx*<x?$vUVs|e)o44Mp6!-1~_hNVD`Zj+%v;NPJNv{DBkdY1tOX}z>^-)
zh;a*lEH*ZblvZZ&cBCbcK>l$3cs?C-PC}wI(JCeGdT{ka!XG*SY#q8X&4>dOfc32t
z-um9qW-q&FbQJ^I90KaD9e!RGidk6FY+C^#CIwy8)%Oxxh&3b24mDKb04m7Ch+rV#
z@hfNtPH?PnwM;5jxg9T{e)~suU=5|23z-m1Y@@#0g7gCck{=LSd284{qWDv51Gb3p
zSUkwV>1vqCyZ}bB;%u<C?jI@$7uW~Yx;*2a1=(QqL+Gd$E)-b1(iE0EtEM$Qo9Jbr
zemNjTB4B1B&AO>w&<emcm#UF2izJe`FO^K@Jr->aEw#RuQR$%G34q<R&#+b;)Je_8
z*&la&<&H%vWO9&2u;z2Y+2deL6`uxrKe~|-hv)riM?N{6ay?0}QDrH|_Les_F3jDg
zXyUs^06fw6@KTCNNP2hG+2cxlYdYzD(fS%I#y1v17)V46Lre1E6xaM(%cae7vGl6x
zON?ns<@QeOSG$I@@>1?#FiN3Cd+}TI0Ce4~k(p=TYH@7q+}OjvY`;Kh7Kmuw0X^@-
zfezJA)vrOkGJHIODR8s{hzyT7qSvNlB^`I>m2&phhZBNIbFjJqrWl1d?#$XJRQAC#
zn7JwySS$5^NR&$Ks*LX#l8}9OX=fF$pvh`YdI0fVUm3v-kz7%H*>BIyklNmhwfcsX
z(1<scxFr589zBO`2Gp!`R;iqag$RCZm(qM&xSc8#dUQ;^wk7w_F}9C<cA4BZp)K2w
zqtD{fKpV9H+zW=eD!w1y(GZ6OJZMrsnBz-nU<|HD)Ht+=R#p=wSQ9o6Jsz^g;<k6O
zz%kPbxS9mlF<H^JK}cj*<jNuh&4L+Kv_|Nph6Mf<xCA*bB^`IsI;_jsS#ts*5riYL
zWy6-vyCmrr8|X91-xCtm*{Y!m`T_KI>06vNYb6h)C)ExS8WCxyQFPN<<LM|N{eDi;
zC}isOPd!2kuj#nusBKPg+JX7wn8t^qSwVATGwZNknx)x^7~bc*A9nW8WmAZ#T*|41
zTH{Y130e}qPd=ygx>sh^4)<OT&7JB;C$#G<P@l?8Z7S?6(@YM11}}B_S}gD?v~ekN
zQy~$ltCsr_lWFRGvasS?Gql%uJ55eWu`17{lnAFq^NYtOP%KV<+}aM&0ALy=?iCoB
z$7=|Qvb%}&cIUp-=E#BKjr)UDwhmi)S)f(a$96AMxpePY`4!Opd?$Hf+Bds>x3Z$d
zb@5Hd+Xc@SlVAkdS?uimI?M+DW@R(Kf|P|!;<2i8`$<@p_HDTEOn6l`;zAZ3do44l
zyK2#7rE6hvap3={ehFp<bml0-v%#JDahCxOd%(B$&R+aq8o=;hM^UUC|8tCJTmAoi
z(#vo4@XyE>&mLFGxB`=?WnM3Uw@{d4v}uedS4&#aK)JTNmB*P#Sn(wGrh-xLGG)(t
zKOwNFgEhaqxXP?-Bugcn4$L}&lnSVn=}RPtK@u!vZ_@<FN8|pJl|^?LM7yKg<@GbT
ze2IQ4g|;~Sk1X58{(F5foj4g5d;V`GX9s-Y`I{8vI*&Pw8(6GYXG}5vFht`?wJO+J
zVP940L8D`}n-#kL?}#dx*|unVf2o9a@+0qQ#Ra}#evUA4`wJZo^F%<@JfT!o9E*A|
zsSafJEwYeDk2Hxk$+h&rfVm)ygZee}W1>3kAxa#Lp|@bsQi*MOK~hATO7>w2W8dJ7
zV9{WBRklyj;#?Cl2E}DvWurO$v*|gU;$!Is(Cm~UT<I+&$b>?ZpSd8R+ImaVe}ev|
zQ)N|A&h##*rct%TW~stTr8*qDV`Cf~4US7aom?`J!@_yH<6IQYMMa`J#fqDne((>T
zWv^(OW@c&4D^+y4U+<g?9WSb5>=YcAF9g3r-~;3?`DHbw@HuDQIAr}xC$^KeQeTP9
zpCnt@iIR=-jT#fD!{)j48Du!Iw(`#Mc2)Ao^DgCU*(Tk4qwqQF&+>J+iYHHqUB!*E
z<}8faE0cUy$r=TJK|2yWfwO0Y*vdjYa8Jr^1<mHQO;*F|B3n5NDT~Y2F<TcWcrW6x
zwSxdb^WC$kr@D1(<@x051k;j%S2uR_Ns4fb7xz?=5mDvO1+c&!C3QSxJh@ns37hRQ
zS2e<g+|2QDYRaSB?4@9}bAg$<D6Zwq7qqx80cw75fILCh0*%~(zOb)#X|9Jvo)~nH
zxH(Dx&DG5Q=FqiMw}fJPglO>0tJD7dpyv4kG**;X=-4|<F5a1z9X95;@CGts23f&L
zBxL4IVi6jr1Y^Xj9K@7RTQ{k1&A?vFcIVEGqB}|DWc%rQfD_>zD9C2{JOOhh0y(Ax
zv%XHCow49Y<B?)~B=3lX8Y4^j#0ugWK)=<tiQ1JHY&zl1iHQ*sF7xhP8b-^?QRu|r
z3VQGxk`UsA)4(DIiE>nSLWu+pi`VFe6*Gp*_l7jOpi8{!v=L|kI`%2z`v$`M2i`qw
zwy-7eT;l7d$f9pNt|AQQX?c=OxVoK=JG>ZWY_UDXcw{s|kfBA@k1boHY%7YX*W@;b
z=;SO^fMEIuVqS8=Ko0XH{NK0kvdj}!IN3qJheB86vDqw!gpn!8FYQ8hZ)-z`wa@re
zD-@##;(Z<=V>Ct40{0Ocglivx0dmknIf>HrC8)Y&O~Jjp-~{`F=qXm!y7Lw-3(Z6L
zKq*1rsCK@fBw=nht=_feVrF~=#MJ@?0Kzlj_E>tJwuWI(ih7(t{P4rPbugh%T$Dh3
zQ8dtt+aB-0l4j0&U6qK#ciY^xv4PBcb$y>Rjp*tHHm@8Eq4<A2cN$l+F6-w&+Z#2$
zWNZuL0RCHQS$IMXs1?i*ybuCzhPP(dg8X`u9<N-eK2>UDN?QVdB@I)EZb(&XF5S`Z
zLq@&3>_VU!Hr*v#QbfZ0p(zAf*;PDIo`chDu7+4D1l-Z%u%)^H|F{nYc@=4EF%I(6
z`PkKyKbOpS8NeS0$`odq%IiM`^k;pRs7^PO=WXb8I%o;5-5Zic#>zs|LyfwOI}ZG6
zCS^Mt@u29R0fHYxl2JgHI<^+9PrL+mr(~JvTWa--%`A1}S4o~HcZLebN-BGlv<Jdl
z;*Be!aN(o_zd~RK+~*d*Gr#M2=;JMRKNj>nSl#rp%TUjIB8@A7!Eyd-o^~5SdA(1T
z(bmlTaAQ;V0SX{otBWFKfQT%E!@=3cm1@Kl#~mi*#rL0FmC~Uu>_Uf&P?FDu5Z#24
z{<*!go%+aF5c-GHCM7<!R`tr1fXMhmJtL6@mwcvIb`(kik)j}4U?0NNcAfMXwu){t
zi#MCb8*_Ytxr0t2kw`JV8@pp&Bizg87G*uIuv5u9<1rMw3?BSf{D3}pA_(-2@38_&
zq?q;o$ya{uxj<I*!>cK+TXjw9^{`xLp6gV9ay}IGU2o6Y1*%zK09<4XPMyO45(wAc
zs`B%@Ui}Wy5&~&3xhfVT4J%gTKp97dV}BHS^#Tx+lp-Y0qakaKJhNo(cXu(42J*%W
z<wM%GbtmZUv>WpHjW}tuj@1(htLCfmau=$$dEo7`lZH4X(r0z%By<!<4LYCnn`>&E
z%nyn7bD*0vn-Qxh+82;HMuPZ1)$YG09#+QxJ-~vo8M7$*#G_|W4$rR0ot|lTCyWTK
zoTSJa27VT}nnSGWz5cPVZL{p-IXi_(){?1%Cb<kHG7&X;%6jS}o&B#u%Pi84E%G@X
zMOOO24}8Y-GOX|3gum=5Tjr$8&7?1t&!$H!mjNrbFKrew5(9xjUM`2vr~B`R>A6F0
z2N99fEPUDL%hzzFnS0CM9~8FJOcV0$(b~K7aroU{pL-+(|KR}5{C__h?D_{VT|9UR
z6!5WolZ7i3q;hQSO-2f+t$(;BSBd53w=}gj(l+D2n=J1lyT>lUshMSdh!X+`7xZFp
zL!scgP)LANB_K1mFB5Udw#%vL6&CmVl(#Q!REavtZr-LtiMIKg&<oedkrI%DuoJXf
zI*rEpl?k#ePU@l#G62MmH1j<w={&ZhwQk*|c1Q9^qoGnDF;STJf1}XJlT~2cI<Zzk
zQtJ&>-ILs0s~fmM@4<$=4fUE_%I)3KnI+1vU;zdLRG$ITNkJuf=+SKTU}1b@B6O>x
z?(_jpVuRp~2KM_Pb}fa^vBM48B{L{%UElQCCjf$&I?8G(?S%!rU$rSN4etK*{2$8x
zsXGvEjT%H_+qP}nwr$(CQL$~?NyWBpTNS(WopCO@N52=l|HPWln!IJ<KCSnEu-X61
z{<^J4NZ7LcicfKDSzGxCCg1fvR;hZv#|z%esshxuLszym!)+|J)@|G3B(%?Gac{7C
znlZo~cY_T=0#?WTvJJQ6`?%d~uRV>O<8PgVpB%l=KiH2%D`+1;5Cs7q$I#%+Hm9;G
z>8V-1%xEEi^8q&;KHW+e^?o6N?b-zl2T~`05Fd<p`Wy40trIxT3x9HHdNp-aNCDsv
z6m@FPye%$isDtnch|9(R;)=-;<J=sM%cCxCp$Vx9aE2hb!w8f}N@8bAaU#osK$0Y|
zp-Tx$l8{*Xn5Sv-8P(tsdN3Z?17d6WwoCbRgv5~-6KRBqnT|9_wJb_Bx+w)Lxp^x3
zKrVeD_B56Nq9HXjPy{LbbhI~fCY|v}b1jYHUIxcHHF#}!&vy~{1}=xJxGYI{m)EK6
z41B*2d|zs6RLUWk!bia;zC0Q_f*qT2DTQ)>f_x2og2W;K_jZZcPMBkFgZQFk@&}Bx
zPjff&<!t%pkL=e>c&0#{OYq1}GKi>~j&J;PjIW&*=?@am7Lfb|KTG8=mMc9*3oN5#
zS_aXO4#AHTAh#IUjE&>b@?}@HVK-K%ASkc!<m+@*RVyDKLPo&@aMU{3AexE?Mb&bg
zs!=4u#p<99jzQF`hDkvwkEkcULNaXbTe;B{9Z1`2asc<xS1gDFmw<zWw>I-{TcNW>
z__!4jnd})HMjznSVHw=m78U{NorzBXXX?hXBVs475UN`NXYwOSw;)165A_DfJ8?x+
zU#;C7NPT<l^IG?HPD}75$$%%0ai9odoZrr^`<EzRc`aJkw6~<)?zW=&U}})u8^DXX
zhzgaZX))<z{4QWW!w>6ujx+v)Dd}PZ!lF|I*B$CN#J7sjcJL^grjOQPRjs8}f+sf?
zCgHa0nsz<-)G5Pji>1Wg<*HV8t;O#-H_ujMj#b#QVds4Hm#2yYqJ<z^D)Jg*-QDy)
zwltnpHuVQge8;pXU-R0ky6Iul*VGSv_r^Ut@3HGHdDl184&7I?4Vr)Y+D!SX8q{A&
zT~~S25)}+qQGfkgPq&h)$+9hXre+lI@@P&f_z<Swr$Tov1|ckM{46ss=Fg|e*#Ih+
zDcJyOkIjcmT-TAw8&0P^TEEK70A!iZ2jci|KZpZ|sMhv4X_ynb%x#f)Vb{6?8Um(l
zgB|q&FjR;HkMh<sAN|s(Mt?iu8cN;vs&7I8QUgP9;e5S^#TAtKNd6AP!=DpZGD39;
z>6x%}goZaA0+A!TkBO<g!F#k-M-eNA9F#}T(pjk_@ke41S35{u7Ri(w+2nI}gdj~j
zh)a2n1bAB2o!rE@(Ye(%MAa7pAt(tI{;C78moVn)4##eUEJz+HhxuR{4l}(cPj_d4
zR*TYI1_damfjhtUmQ>n5V%O2}NSkT<mx4S<s9tdWS7mWIl`6))MgiZ*pLz`M5ePx^
zDJ^nJ+k0lkB(Q!8NIk5cj#m3~S{wp!`z%gzO+E5LaC~yU4iq20gFavl%3T@8JqP2u
z7cc{-!CetDaj*FBH-rkv`#%nL^55Hx);&Sn-$1HAemwt$R80SorTzcJ&1p#6Z;B!O
zAFWWWlXdlDk8T_UGM>$G9>`W%SO8uUzPi0J;@`f6N)CaApSL+@6O$x$BQdtceDs*^
z%*<id=`U>#2;Y<E%bk~#1Dy+sIT^#!?}JH!e744bcWIypVlR}!G%QDC>+$-&Jc-Uf
z>;HN~l9(CJ|Ml<R^7!oo{_Kn~D-XW)NDJWe*ZmrWILjePLK97b7Wm@nJJah{S+ucy
zcjoNQG)Tm8AVji}PJ}=yC36NO+SAlEbx|N6Z1Dn(8j;4P0q4h)2CX|A6zB(2Fn-P5
zoW*HSd>Rp)lE4f_+~0@{N#u}PJ813LK4ASw8xJ9-3=_PIIEhJ|R~gRfc7gq(9OzN;
z5#jL!6dB-C35imHvfS<Ubx9Kr_}#;zdi7RHwss3_AHCztzpXjChfI0x?{*laX!i|p
z)>(0g5IZogUMIuPadnnY$qwz>oyg>Xj*<~+vTgEOUyl|h-2pB!mTir%N_4~?_C(sY
zSAf~$JC~{*8n^`9U_Xv9a(DD;;D_=Ksb{*M-N%_pg|m<hGr;eXkoEI3!*E7?{o9Pl
zW0!}*efZ!mi+aSf?+p##Iylr)rucRG5Yg!a3*tm~Vte8zU%t2Xd0f6%BYY_E>twLf
zO^(Zf6r<A#VxwBLarN)7OEM!B9X1u|s<z&)eLp5Tk3J}@es5<PXG>gXDAjwxDl~C%
z;J_r@dZx?`$9DWXmB_?lZBAd3yyMRVQ3cvnE--*8Z}P7e8?HI(-34Xva@6EMwFHCm
zX8y0dCOYXb3IYjM{LK~^Y6cdP9MPN+)q+PprjA#c1~TF-L(OxC#2O>;V~mi}Ns-xq
zDdEN?VYBma3J7=!Bpm7`NDhs&=x%Lnv+NGZCOM{7$%$ZVw9g#o*>J0y>T4VhX-_AX
z9mUsc{=w}si1UJih$<(Kna!)>GxXr(NbB=Re*R2U)s)!t%5Xy);w5pJ;^53xNs5Ol
z&*J&WOp(a@R<C;B&B<K%Fvrj&G3InS$P?0&`gz$#|2>}oop2c>ki4C?ioPDOb1*Mk
zYpkh!?zEwE^vtQl)~I*<mV<&|*F3nC9Vjkn_+jUfR+b7jEANqcMb-lBU_YuNM9U>I
zv!vk_jne}0i$a?ms~G9K^~Qv@OHXxoLd!i!hH_h<gCPa6rQNX7R_;$_@oH%+pT@I#
z3USgir;*g*MS!`1qqqw#3!if_Tz`q3fJg6g=$YNSGIBi}cz|Sui;94`0c4bv{f4I-
z?-?Rh`aIQfa)Wc-Ba3v0bHb#iFy;lHn7nM%R(!Urs{LC@A=}kRMc{6)4hC2x#YOSu
zppS}ma}YTI1tK=F_WIW?nuLbE-P01g3yd1NFi;y5>8PEty|SZ(ztyJAcO7J)hh+dQ
z71^WWsxexv<EMub`W+y&(SJqZRut*^Yb@nWW+zz=kamkzi7Lla-yIdAz~}>V^>Cx^
zedC&p;U!nOruDX!G}$t`t!b8(1acoCn0h4W;i-hQ?)SX%ciyn1jWubAH-<w{6kSiT
zO;RscHO!J-_Z3){_(Dcx%Ynu`DUeuh4}}KZ7#?Z6OPuPS;C6sCR95=2-uT0onKDea
zB7lq?bP&O=?HvXFz4rc_I7@<Rn|5mw9X>?Bh3s<u?th@40L2i<PN=5qFXz9GB{kbq
zDM+7NY7K{_+F_+wHR`U#j;`@(Xf)YC^2HAKl@4bqj_3|VA_}ags%X(qQ#}p@%<qCE
z|KO{mY%!VwB4^5Mb?P^Scr20?&q$BMBBf<;w(Id57|joBH<gsM+ua;ajoBmgf}R#a
z(Wxk%3Vh*nw$te11_R`!BR#dieObvlfC`$w*TTQ~U?J2I8fXfiDvNtA(}p>Ob;&vR
zw+qWS+Wj7##yBH&k*5Dg_p^|7`}J)ebq40r%iosEH&9~q4$}4s{Ln=P^Pd>N@&ERV
zFtV_-{jd7WwT86wuigIthXGb(LJ;u)fzQci2P8RUj>xWF65aunYX>m+RFb4}T)*TT
z5cFK}Rk>W@mKWmvmafmoHC;YQXH??Z`N8PZ!M<jySQJISNOMn&65_o~Jjvm_7jkzj
za{}#;#?Rwj44GA9jFVV-F~Lxc*XPyz+xz(CanwDD9I0k)yTpP1E|32k%IL~$uEF87
zSBCk0QQDgK=f&{bL!KdVxdzEjc}RJ-BOb0_SJj!aa$T>>YSPoeY$&f7(maK%q|iRG
z^8gYe>eIlVvY<gP%y|`rVT+}VGpGB>G%ZfhTqRV=0F`1u&rC!@=b=Y?kkC045C%|w
z9?|nAM1=Rdn0Xr_%1E{31eB5N{G1yXTd>6^=5+WcXHg2AMeoNtZ|gX<3=+%~=aQWH
zWf|Q^H_}9s75*)K2M%#Boug1>NNeQ8=nho!!FoIPQZ6X{35dyMm>!E^iv4;kGAa6;
z<)6JW%PuYTUUHuMQnEhI&%isU(it%tI{U*TfbiFT^D8&6^8yftK%)5Zud8mp205@T
z^V|<~^1cW^>6kwW*Vg|A1ld~a8HkkX_P1<Yn3F-t<R>L<t~*In9J|)hOGgt7FU#mQ
zE3J0ZrcvE?FKYwgGRU`prO1Vj3Bd?Zq4c5r2D{Ia8v7gLJIf)tJ)kCqVSG-ZM`g^!
zOhyKxy!2s@tthHu&n5ZKxJ{*?8NcKU4x_*=3{=^`v?t<FY+LHEZJ0k{uCFO8dtT_|
zIQKbyC=Hbg@19nmag(UreB+aO{j}+=AttGvIjdVc*A3d}+5HN;hwQiA$gbvP<B`H}
zgp3{Jt^nu2l(WnTBBBeCpQ}++l3>^*#$1+^&q^@}2qz4qYvM|+8<yR24OV!&tVlH1
zjUYKmC>5Dh6y!i&qbLO%adeb}N5tK_Af}pfw)RcH>tu|%IQmEG)-%D-EpXEFZMwYL
zn6YSQb8fHp)OwlqGNFNsMxP)`bhq9UCLX^`W{9}d{1F*55UmzFS~d&#@o=)T8u2^-
z8Sy9E9v(`g15XX)gys7Kz_1AdJA?xeF_lrccEg&l#@6yZt{vWwT*)-uD^uX4zB06A
zjF-jF`Ni%}z<(mBDCY|L9XN0AsKur1*uq0?(AI%j((F4BPn|3v*kpOB*m&jV!<9pf
zTJZ#MZ)FL7K<exSJ!8>DQa1DQmbT^;=o1i6M%oj!InXJxbJz#8WMuDw;g;d`VDT$I
z3%qRk@x0_~pH(^yIb-{%WgKv_1~3ZX+VwG012GEpS@hVp6%B0mGa*;z?`Z_dgr=so
zV=93Y#cUh6E`~LAw6`!%x9_9nF-xHCwE^Ff?|euYRO7Nf&-lFEWeBk2^V~n>H$2I?
zm-IkL_UoXVv4>;^7e1@&J-#I}9UlVCOp}EPsw!s}7}n|JOG3<$#!hzPM3V3(kbtkl
z0h~JlRL&2hi93XWD@p8E;lrTXc2iS}17Og(EH(qrtVx8qG}C+29}mFX##-_ThE!ZD
zyo+9RAK!%b6-p{Xp%3hZ<LF6Aa_ye%{Y}CVOqkAB05k6bdVK!N8K=wI5WC_Cnu&?i
z?rENhRESvi?5hDQfybB_nBrN=w<=ZYL^1GKp7FKAIf1oRlpt6;%_y}+Ae}-&c}P)0
z&mT4kGm;e5hVNF|<vHCBA|xyp`X(6(c=7zz)6)TG*rUJ*@)a}Bz|w+FcE#&l%qp|#
zz!#L=q7C)pF!Ty~cH1Wd#J3;DMAiHvX@8Az`E3+lF~0fMw|Zf#`8rqG0zV}b94$-(
zmy@ge;B{cnB=>QJ=YXaZ*s;R8<WUFoYJ5w!eXm^{CopxO?|9C;p2VvRlK8M6j7xVa
zQ&Q$xzk3RMj`#FnUTg|pCSv*xi(URRZcc)%fV7nMcc3Cb(6v|n^$E8w{qd(4_r#yu
z<mGL1znCa`@rSeVt@U<!kRIf;w^wzp^1o2wzZM%Anc4nl#b#M!*Z%*6uwL{DA5Ga1
z7C#A<RH9+eR$MFsYjh24;5YFsqFX`7l*q>Aq6$7X-BBdfm25`km|Q#&DWlx~a^v-W
zP~gx2<9u?w`f;WICFOuKLV>3G4PhA&?G*@c9QCt6`r$K0VqnE~^bN+P+3WrA_`VEU
z$Z+EQCc3AO&t}g*#y8_zj)ng)py&4>!h^CWfvwd<DB>v21p8^z1q<iZ;!(MB;(oFc
z)YHI8^kEDPu^=TVfIU3*#W439_P|82*y1E}X~6h-mZxyzLk2ry4F)q)FJX2#C6tlC
zkch}J;CmtwQ3_*sx+a`eBlE0LBejt@Fp!}z9)NtaB>@L1l5ByS@t4nQ81Q_KvUw`j
zsy>6CtWk+$5E<epYFKj<8_=Hh;Yu6q9l)HM=`O4Skj1OZMUP_)$#66*E?{;r%`DdA
z)@l%pyLqx@Xq;ZCYOUwm=RB;MY9)&oak^n~Gj{VLa6Rd4%|^f5;n3OiN<DP6Zi_5e
zQm3D65^Y`_HK)%bKvLB(HUTqs`La<asSf)ib}QT5gtB<eeQ(?hxSyD7UZGJ%Pdl0F
zo1J>k2b0<V*c?{NUJ{ozTYp{EK0RIJV~ghcaxE5k8@T!r<e2Mo=QzHs$&w+YPJoJ~
z)HdyyxO(NBtNB!=k$2oiiU0ed$0N5mzqZuh_Crr1XF*=EM>R7aF)#aJv?&_xg&0^;
z#zSO4A|3}6S*hO$$7u|hn9E~YWE~x8jy6gnFP+Q^UcKwI)AAq^)rhf>=&D@CBU@_>
z2|+w4CO2AzzSWgyLV4EB*lx;AuBZP?4a^{P=-|;Lwa<d18Lt#2wAJM1=OZ}5fJAgN
zp+Uh!SDOso<V08HY3m*U%uuoMXJyz7j(lyF$A~Ce2xTHS5*BTz#mR=kp8vkJQf7H7
zKd9iJ8TE7Z2r<gPxwJ`>xEqM^G5<Ece{2Wo?9m)!cz_0*Cq|59NbL*_PWY~c|G09v
z4GFsS97K-lCD@#=QP|7vxuS_Cqq11^H&(G#Pws*`RV%~oG=>ty+aMGPuP%BuHT4;*
z<Er<M^@5nwZ@NEAaVhwLCA=lRU``!WtIFn1tn^b);QEQ#ndlp*x!Qa$q;wo#ls^~5
zdI(VwU*!-|{;0_KqIRp5XkBV~cN%1w3DXj^MuRufFXAF=TBmf?${EH@@G5O%`Rbx=
zvdl+159j`vna8-+W?x(R6!rBvn)WR%`})>A45@$eq2H}v+xn)xSO}N)a8yzG&XcQK
zA9D;%r7ilR(FKruVMkNdEY`Sr!i^?I>NV%-mX4E;v3T+<ChC>W+ZD2s0n0V_*`TkZ
z1$}Z6g}{8}YRjytguQ0$$Os4C3#hA}GwEOc<%2toAr_RuETA|+)78h;D8;j$f_5HG
z`pw#}k`}8Nh1;ufwc)FLkGpAkY$iI34M-Rl_jB9!7PsKKhXxOX<jjnK(_0`X`804B
z8n}ZjM|3pStFDh6lZvrK>0Xt`Zf};t2wFVWXaQU3ueEt5s5=Nv9`;=lctLtc;SHMP
zUfs%#6&OWw^Sd)RSUwP+Tz+p6$^ewa;dZ6~WY(Wr`}5x~e-O~dNgiU`eO+>nryI}^
zVjtiEQnXAwJML|&lT+Vxtre}EACz;81PrWoIp2gD(fj}0=(NhcG`M!*KI?C!6O+As
zBfLM)vj?x#1dMJ|ubzV61}>fyR?7NvwL+rQbN!i}t@v&<E-Oc-6mX1gBB$G(00cY5
ziX@9)(wf6;;JhoOsI5X5r_-Sk3Yz?4uN5zQ5&qj{tFMI*D%6cuf2V&h<qQ6wZ}fj%
z?(CfZ^X2~kP}Kpm2t{)wkqxkqs5V%(T@_Je@hu{MU+Bal;vLcX?`Js9Vw#Fp_oVIo
z(O}wK=H2Y0crNO2uDY3?j67bib4PMA$33tapYyxE`|bdRS@)BQhsz=8poiU)T|TdI
zi!JRDpS1c2K*a8gx65ev2lTX4=G8J}yo84g{JC*?b^D#3frhtB^tJZPyzrf)1KQ`@
zpS!Ov_T7Fl14l7f3aNv+Ctvf=8nu~*ipkmSCnLY-eVIr>+!F^m3gbVs6#KcSBw5U?
zGr;&O;&6u~G40Bd=AO@or$JYjISp%(0!4srQRaA%g$Sze;1@@FHe#)&1i99xusHN?
zkWYbcU~9UjqC~MJ;O*5;Kw^aySazKwu;LjYjF5oy2J&jb=iLrUXm7Drpva1=>y`18
zf6PO3(l~VK9P>4$ZD^<zNvti;X&f^$`2creF)>(i3^jL%0<b-Y;=1>1I}u~I#tJjA
zhG2=-OsZUig{dvEWIdshnS>y?4j7eo)HLV+mhp9m@w)+Jog(kX+Adu;n(4N}hAn9M
zOhuu#qAQ`TU7VwN$x5Oqnwy<XX^3p46SlZBq5gW1+FgDNI<z-0vPF}aY>{i6M9pkV
zQ%!Tzy8c|tsernSF=TS<i{Q=?z@<>2w1*GD0v&xb-|*&vCR5?hH9DU>xxM0~YOnWR
zTrSt%NgQ>!i9$a%nw1J5GOP<G8cO^n=;5BGJeXuY&aOnwOemh(lnuOw;5T^hl^u>#
z{E1OUvI1r}*oerhgPo5noERcv@3#K58G>btYU~Ho*rE~iu3FjI=`=q-wrz9LR4dJ4
zWo6sYLPEjyF}$^D!e~q|&mIzuQI}029XXtU>J`;oYAkOOIT|q*A!Tb!F!NJv71hH3
z!*cN=6sTnIJDc^lUG7K?T#;qHriI(P;UCX+EK--zZX}s!2NQyUm~n9b^?T_~99ckf
z1qg^H@33dg_L@nF#Xm(gc8&!!+wFr92%PSL`K-oaL>L=dp?35@al%OAByuz`)9q(N
zHfuu<sL#iil8|YqU^hyqK|bK?Qm6W=oe2_e`HE;y=1c)4oj^?U?2#mBou($n63Zoy
z_QcR?g1{0q|67s00{Ngm5@K3r^FGCXg;+St*e?~b4o1N!Vk%-o=~D<eW#3DD$!WYb
zUWo9m-moE}s<M!D0P8mGZC|&wI4(jTF^pqjt%cXbfs$gboCME4$d_P17%qW+xWEYG
zf$cf6C}N#G7+9aKk-6N4%}C}8q%_y@IGje@BNr?Wh!5By=u;b_1s3bZMgX`i_mo=)
zxfj2JREl1O-ERFoEL`VIFu2Dpl~}8$XUg)$d~Br-d0*G4PSNeHP-&on6(~Vd9`6n1
z3BcF@Ury+7sEmyM{)Xp;Qc6!+woM(mhB76<Ar;qm5oYsUC=v3PjyT@6?WGb#!~BT(
zZIo<!xWn*{ETw_=A|}Sq{uFV0tQeilkE*J;g`KbESYi%Fj*IRzn)lVf2J%-%LxO;Y
z!&|ZOHsSzo84(_|zy{qyIM?pp#Q27fE0y%^j1vtKb^^t%ZMZp3`Ee$lpQXDJ1B%aJ
zFUbeo7#*v4ml&`2w{pyT&G5qLJo>=QSVey3mp|lr2I5(p-`!!^+tTEV;6c9P(qiyd
z<HKc|Od(d|KfM|YU-nu^m8--fGP&s&z%$}E+T8d1-h*Y(;^AQGDaL+ETpah?40uOA
z=4J~k0V+Z5Kh2@?Er=adzdq~{5|s>_(4Z)*WRadeWT>%=PLjPHel6w?{Yo>vXsVp+
zs^TJ7XPEEtVDF{roxsj@UOjeaGidX2R2~Njnr!0YZ|U_+?ZcFTD(y|F-3^-(9eAjN
z9bGl&0eCd}Zqx3|a*Bm-UleEJ9rxIOU5Q%z#lpX(^^mT}<n0JnUs_rYs#c^o6Uu|j
zsTaD_`s2l0HYN!ZOBM=gmRj~*3$=)>3SK+{MmrcPk7womC9M-H2X)uPL6wci5Q6Hu
zRewx7>T1RB%G7XAtdHiGp8&sHTAl?6bgf@_jSTAr1ZioN=Cd1V&=xZ*>uI%RUfj-&
zyfQK}JwW8pR~8en&uE%gjbB`nP^F(POuY62Ey<v(LWPcjg$*2;1suj8W$=j|Y{o>I
zR{|GoI;IlY?3py<R6JZs{H{M1bAZ<nYE>bZOK=kECZK}MjI-a3FifS8-U?Dgjh5Sz
zp5G?P{l>(+T5D}JQYKF%Gz3U$YC`*gqmUgRO^V}dA$SZd9Ad<vgoSyj;VrI9*gj3@
ztR%iPot@{Nsuyo??FqAVUlhbg@rFyor?Mw+5@x}awO!IX-T9<VQexGk@7E#AUQztK
zcem9tmgO=5w{7N)3b|Z}1ZCgRSt;~5{|i-rCqe$-J0YC^d*xNhZ-Mt=E2>|fenFen
z9HTmks4muH!FG-;a<e5C21Jz2FRmI}Q%l#<U9v`<w7tZiDx|XHuCojS2;V!6-RivN
z2#fdgMFiE|ez$jfKjeyg#y9aGO4j>AnIM&O7%El=*<ya8OvFgK^(soD`2D=z!0+1j
zdV2RF{`RN-TXm0n`C0S(IPs-9{nucWAZ6f9jz?ah>w-QXX}@#o)9Q7YW-l_)$Rk*A
zyHQ$;Rm9C>dFHan-?FU;ue$lj4tGp*@H|Gb7pNj5N~DnRTXM94XE=xtUmDnLgle@(
zvM=fWZUMHqZ?SxM=aNWiS=bL5yIpF_hRk%vfBp^dYtm2PFZvzq8k}bV{)1l-&dhDx
z4~{s=9LENEkj+f$BvL6~*6CCRjR;5E+DWl7e+`Bk5RPxU3~y^LYxKcHbK_Y)iWh3y
z-Ze)CE{w0o=};D<($34#VmQ)PZ_N1I&EHcPEA5O02gi|DAm>chF0^D<^0d?+-?kq|
ziR$uU@czAj*|kUBc6BYt?!8zf^AU1QzOCw|p5=!~Pcol1KsA8?>Y=UEog*MJ47sK<
z6cL8fWS4sdjO!kc%Ti(&234&Z@4vC&@uNNxE);DGsi+iBTvdwiqWnsF3OocqD};12
zdLRI}S9F2}0evYLVn!5k7(<Ffn77$LHxI`hY-_xi|C<!Jx(GYDA2UAwIMaUxSapA~
z#9@kRNH^~fGnWbQ-D>gLT?l2$d%9Q20>46aD63p>sk+YH=(L+vZ5q89&)K}$-(h$}
zm{bO|(`{Qhrk4=}`Q)k?z25G<A$yR#<7Zj%NdzCB56zN6sh$l|VXP3D%h)@=vF}AA
z0o5OB4{>bILg+&ZMI+`%2!+A}$E;T*iWD-&V>VqM-S8#K@)HfsK4oT1Va_$n<@k~P
z4KM^50H+F=Lz4kUCf&x=uCh}`7T*0!fsTWX)5xR|{eFm!)3L^(Wp6~U^ix%*OqX~F
zt&Il?05@DVs_NYKk5F+5^#9zUyklmIod$2oAzn%-!F-`VjPPPSSR;>7BZI{MMz`!q
zZ7sK3Q>MFq`w7{jnhz`p4X7_W`s7v83{+3{xbo7RY`CNOSZLw<Kh&ePMeDOx@==Am
z^Mr^{*-u2{8N37_9lbO<hMth5#?AI8wobzqa~KWnXH;CLg?ox!`tiE<2Wil;LA?Io
zec+%bPV8QkSQJ}cJnT=jHH{H}Hd6o7?JR-BuvF%RlJ>W}>BV|Z4x?6U+Q@^nX_nR&
z%+4w|;uC^4v$^2@i9R>m*KZ%#>!KbHi0ORl#Cg8GvW>xeZo_ee$>gmx2njc}n|4jB
z%T|CyhdjOyrYk?;f2e@8+KWlYbk`CAF_h$!+W{)9qP$*72=RjM-ooDB&oAt@PPO@K
zeZ$~~j!N<8y6`iY7Hg#^a7YEug?iRh!ejzRb@r5<x%4A(n!Y9<{x4Bz>Djmeoq2^r
z=CUFGRCI5H^lTyAIt}wuluJ3VJ}iskpoex0_!x4DU*IxKaaMN&%YMf1T4^rrbAP9>
zC%N`X)heotS#_-M(m0}o_jA^5P?#|j&iNq<R@Nl|>)UFmE5P)DR*J}JcvI9`Kdl9`
zF3wIzN-g5b9P0tQ=*n#gA<Cj88d&5SIy0$<$V2nB@CfeYg3HFx#A8DXz_pC|h%<y}
zPf*;YC!e?Wn$at%>sz~JTAi1J!+B{j@^^U7`qAqF90Su%6NeLXsrNLrpN2Ix3@9^<
zRUv>?b_dbX$~D{BVY9FH2DlC$c9vf;Q+(>J)cy&==dog%f+A2?%bJ;7JUps64_8!a
zqBGP{zCHk(-q9FJR;E&BkJ$>xT%6sUAkVBmI)Kfp^fAxSV7adVUjzp@w>V8&&v_;W
zgbq7(p_Z=R-|-k~4yAEtS9OX}aKlF)ln&7)9!Qsqq&BIPPV%zrGsMK?SB8uKqn*r#
zPx1CUdd!@@ws!4<kA>Z<lEVL;7Wa=F>yyGflbGd9gW1ag<}Tyh9{P$vu@$$GnVh=M
zzRvZ<HQ;`B*nF?2Aw*iFfc#m?ohJVfNI-L(B7I$`3#Eaieo3<of3qIfM@R{{0RrSA
zV>(v%z@{#Yv+`1e3&Sg3zF!gw(a25Tb}SDX=`0Oj{_GTQtLyUL@rxjuSMjSLEjHag
zdgXN@tu9f%7rkkc^z-u2m3w_|u}K}Jz78yVh~W8lyFxH#4nWF|z60b0+X*>oqFlW_
zKlO=9D4-!`HN7USG-Ghpz0SMdY*R+`>}c2wz(s%0ja`QF2l-4;_OJw(1VUt|y*mMn
z)YL=N{&zs-vqV>DD8}B|^N2bwd#dPP+Nx7YtDDKTO+KcT2Sd>UMFV<Uqa!F)YTka#
zwCJkr%F}GVluRQc$UDuS(d3>-t|x6jTL^Tc(KFRR4?4`vdxqT`&XF}^BqZ3QeD@t1
z=J(H`M?oP1rR9d@>!QGjVdilXv$nn$zgsNLbB0NPd$+qeKbLtCCkcV=vMO|K098af
zpj;!QF00TZrtY_VnjYVw)xOL+fqsxmE`5rkvfjo^w28(vLucY|^_mM(IJK?6cy{!8
zB)SbQP;1Vs=}gK)BaqV7C!>IJ?M~kIs73A+I)54Mu4&w6*ALK{Tm<8P-UI*P=3`=D
z{-51^IsgBm(%Cu$2YM|6BT=Q|;sYlEq+(0BRW^yOvJ;1V+ge(YR2an#-IrDOF+x*G
zYwze_!w9Uz!Tq}|`!J~@-DEL*{ci2<_x-XMaZ*asr1=ZJ116*l5=*keWwv+X8575s
z-Q`~Y*T~Xi6AaXtxu75<_KUAsd;U5;pAPAC31hwZE|D>o@2|@;ef~n5rw4EwAKmR^
z&Ch3bAMc+{`aJ$GaIyiUIwkw1fx?i+?0TMPBoDNbA=Ksjwe5K#yExqvuh<Jb)m4@0
znfMI8IRQy2X%DoJ6ckh+W0|S&uY?E`8LKQj*46Sy?aY7Ah@0BJOVeuXygjY3d~ub^
z;M<ood)~*#v25KtMjw6_Y`)8EBXn4Sf-q{b#8~6W$(DGgprk@4^A`XkzMZ|&9#Y1k
zaf<vsYGe_y4Hw%I^vbcmjT#-iS7E`-E6<PDGN%U3*oii*SMjXn5wUIu<)f9mgLZ$%
zS+v(^qz6F6vs4$Ky{&(#xaf@<9t;22x=P1?JYgBIQ8M0hYa3H?Kp=;beIPuL!MHVf
zo42ZQt#nr?iOGK--Q;*4ux7#3Mn`B>-qI?c+XJqRjBA3?vOm`1L306}0s567P5`Ev
z%vA?C_Ol6b2rZZzhlphX<L$$Hc*}b>_NWFOq3Qc^T=GNJH~aIQ#l3G2Yt#XYb-REu
zFkwzLBLdT;_I1HP;^(vPn(q``76I8dZA04~yloXbE#XHipQ{!eS^LP|2Wrl`ZStK5
z8$R@^{2YXv1taI-JC$490Y4)U6x*Ulp3zSe5fZG-Q+FD*@Lrb9Ze^iM``$o4<(|_A
z2$L~y92o%f?DWyfW^!6}!(;<pIx8>(pn1@64a(s;Sj&Y21P|vj@wD6DV4cQoP@S#3
zo<gplEW^whveRND{MpJ6=lEpq9NxL{LeB0ls~$3DW{xuK7n|%C_cocIRRJn0FOj1Z
z78}&CyGf-N4ud2jpmNPa(D@kDy=|*InXu*8x_Mf+-EI0S42)q_4i}N`yZ{AKw|*u3
zTqr>$%G38q&^^#c`KdzCd>{y0w%a*7SKf_btLy7VbN@jwGN(CJoqEGms9Nm74iXtN
zXHavm+%@47g8QD7nBoo$U++#Q7leV!N=VBIr0-!}Kug6L;Q{%?=fr!^ghPblh=5?*
zW588<idAHFjb7aM!#R0LvKCC}rt`z}1jAJj*nk{4>(H14856!`Z#!Lvvk)qJB|}C?
zHnl_=c*zb=i{7{<I(#p<sqyGHo>6c2NDw-CBy0*v19a;5vf$O9n<}lJ%U0e#Q;|le
z+O1P!Eg(+SKhpUXcYq;Q2vm0mNI*RD*K>x@j<yNLZ~Xj4;wUr+@xbMe6}|(N-ljlR
zR<m^;_2VA#3SW*Bv`6*+tCd*VDS6n9qk6orj0&dhd=xri+}s8{^sN%MOYaXy&WVsG
z=<7#w0UOY9AF|tAa)_ESVbQN^Sjsil^rFg4;iKr0$}y01(foPNfKWZp<SJALal$Rj
z&`_~C+9d6u?wP~jHCVeW;w^z+-6Q8I)QqSrlF|psh<lk39`BWtqy&h<NN${D(bn5n
z@4QmB4N0Rt!kpy+DDpP!OO;jMk5@K`w6*m{<B%a+KKpg8a$V#4Fa#uLJ}WHu7WQu}
zw}0WX-fA0;nak=qvD^U#dJqG6&;?SV$-fRk3u)V_jm@2HYVUjUIC~`^jbTPX^3qa8
zY<V%IaH|=8cUchMzn}J+dUo|V!6c&uff9o&vC^c4ardn_T#|0^9##6+Q12jk5t>`{
z)Zk@=l%Pts8_lY{oFMHqB}txK9%iseFtZQ4v8>ejrM0`&5t>)*#(|0Z^LrkOa^CJ(
zMmsMWEOpK6EB=m=x+~PLti0^_6YMrWIat)^Y<7``W852p6q<2VZ?_D>xz3-v@N9Cn
ziPn%ePwTOE#XEsHhc1<nh^pl(?+EgeuG|cXuQ@BEpftw8Fw>b+C<fpf`C<CPr<IMM
z$uH-nEiN2L1nmD)$%b~xO-%yY?%B7k<3TrjZb`@C#KAQuA|dB?DGDmxcwKjpJjrUZ
z#oA%M%19efJCI)X1S&yL+PYatSx>5p_K*W57+hFg+-vLvg|omaPmfiE(@354&1Xdb
zD`oO4>_ve*Marf=bOpaRPc6?!n6cE8u@AdOxKX)u)3#*ZDWf9*S>VNn&b|0v=-0G@
zB<@@9cH?pE>DC?4g{)woy617JTIf`ZNDYbRfx$XAD1eND<TLtfw^`JoR;?^4sW+ye
zaG)MR)JNtHT7qh`Dds&6NvckKrBgTPNANY3+OEq^I*yWV4?##DDl52P(-1w{{Vd~5
zT>>D(r0wLn(%Yy4=~;l$U|?2KB@0i*xZyX3DZSmv^!3@egSXw%rwV2H>30UyIha%V
zqF(&4rrGo6c)ty;WcWOwUT2;)B5&%4%_jDRJHJII_mzewg$b#iLeJ(V{AtjfmSiDM
zj=s&rCfu$~)g+T;9MoBS<dZQ_t^V{eudeETZjB&YP(CRvc+mSovY)k0qTMCY{YIPc
zrLPByLSTGujc<&$yG{a*>3^cyN+N=GS=OVxmhDKarbYnkXV!z)dfpNkE%YN4B2?=@
z!oYeQlJE+KzpIq6@uMuXzyD2t^#?sZu_XQcYExUy!+$MWK~InAU_1$!@9aup;9>A$
zS$~mX_vpWF2-g4dS#ogx&l%`x4coZQcEs=B%@ToLi$GN39x1M7(AMtFy;=#j?N<kU
zI?;lbvzB@<$EU_W-t9j-gyG4gqmBU}pf|yUP7-k8gXX!ITyVKT93H+;ZOhA5+*sy<
zVobv-^&22Mph^@-(3sUxd#mwfBJr0JQkXm6e>+PpJ$_D}xe)g+Mtk+eWy<eP-FLbB
zH&-b-d_TG^N%ptDg0p;Z{`L;uexzWw`$l)W!@tzw+wK1C{a(E&Q+E%VrA0~{QVQ%}
zAt+#CKBsaszuP4^$VRGMATWZIWDZ^AXCOm=*zaS3R#K>ISCprW6x?g(cYD)%F|UpY
zjwlT=MLF%Ry80V2zE;<8{g;`}YP8hhtI{c{((P+2n4B|qu&Y(L;yZn{OKR7^e$C3p
zW~0*T#q=R$+iwh9_tLB{IG)B(39M}QhVpf5Zt${B=V|cbf?pj44j-H)9&+3Q3nB$!
zBFNR(bVw)Q2MLQoR756N4m_*M->)Y2TMF`+z}$ga#7U<)o`>U%iOP(<LO?Yegy@(=
zcIzoEYcj1);NIm8-w3mAvowI-wU(`&cuT_)&KyQl0xw32>_kl@(V>sPG$hzn?Y2ri
z=VsJO<KQ_<$0cY6LO})@*jEf`#&Oe^GthoORsD1@pO@o<A1vSKZ#}XI-~&bGCn3Tx
zrljb<nusVBobcYS-9T3>MG@n~F`JHmFZLa84;V4D?<nrvN4Y>BjdWFitKg8qa49x_
z?JuW|lsNva*q>^)UCu;zyYXq$Lt!22n>?!TAq3+C$ZRw)9mK9pbrTS#O|>HjHgW-;
zQ-YZBL`^95SqT^+Rak1ZttFg6N_<EuFz)gAi3;$8gd47;44o;pn?i?EvAFiu$-&_+
z;lTI;wk0`y=E{Lh&!ij+TJBD8=GbMJiUGBuj{Z(ubokNK0J>yp($40?bhkAB<US3E
z!kk4_G7o&b%ZQ-Et+6&>z^`#n#>?pHY~_-7w25F}+<=<^%*zZ5*gFC{Qur=xzm>F5
z`Vl-3vlzNX=RXMT6$bWJYP=A?8<*wQnk`s0GAodqG6;AOEn4mVjY^@ju`B0gRWyBi
z1V0Cn1ZVSKjR2l{oRoZ*%DFsx*3j>0z)X#uFXTeAv++ZvWkU?7T1O+*=)oUD3wdr#
z;^VFns=2rgJ_HrpY^@(0Pv>m$p~TS~k|<(cyft`Q=E?5qaZn#G<hnK>0Cj2-8;YHs
zSmdzVD%U<h4g1#;npOJSPe}bdL1V}%k3|}F2(ITf@s@}Hj21wGuiYxN5x)v4P%=!_
zuj`iWud(QJU`itu=Ks;F%*ms{4`&-Fz7k!b>7#6sPH58Q%i~e_R{trP>yoQ`D6q-;
z2BV(ChLG6ntGOxYE~cf>=BG8q#s4n)jrw(+^x*T4x{&E|1t5aHV}!J$3@GQss+?29
zkXWA?qXhu@n|-ph&%ZGcF?c5vs8T#4xl=iSxxE7lL!bQ4%^AZNhx*N{c4aa3F~(=t
zAV9~=%6>JE@!1(8NojYjJdp47^DMt+Y`RrPnP#NiqEGn9c`uQ@;2s<E+tx}CKJL;s
zp`(VFjsM%em;49i7fS+g+C$+Jw4G5Vx$kUlo4w#YAf05ow74X|;y7=0jj~^W%WcC+
z7MdFQ4pW2&O70GxXi2|Gy*<z{kME9^?teWY7YO#uzIHzPgr@!0+(@5rOjDPml`SB4
za6>oG1vN`CI^hd{?+_dUgU7q)kb|%Ac=p<6r$2{ra+Tel@s<=6+D-fnJHsKgS&D+!
z!)X=RB%Z5qkW?&$O3vr%1umGH+0XSPIZW~vr$&B6d4$c(b=>cF)-%MJj11P{*HbZn
z;lG;$e=}}2vVRS;ZrrnTs#?B45CG1L>Ds5ZnIdlrOs~5->&|^Gr=Bg%sv;x^#jSO`
z`47?>C(BfBVWva}ctRvw=-og>L0q_3$DC129nEet$6E6PwY)B;2c18F6ug<uxQR^#
z>X8L!%cjpJczXJ*;7^mji@AfD{=<nrT(yO)X#+bL#wX=T%b?dDpFOWe=ObuvxWv`M
z&05?}G^7!#1)aa8G|f}Y9PElGv#cuE%pYasxRjxA;+cShA+QwOdy;5XC8p@yGv;HQ
zP%jcmT-jFyX=ZWlQrsd-$_02W;GBi4Jf(c;-P?x1VyM{4W*I9H+?kvQr>J8-A1~N-
zT70dxtfE*Jv5f%JH_md?#E7AUP{#;w?pN}$CcdkL#OP@HaQgULT|KqK>~-3My~`-N
zPlKbiRjD*~`BP|tRD^SVJai~N!bp=vWKfjMAQRq#CFj#fb`vQj^`F1YtjjUdN0O}P
z>SmaRjCpk|%^(i%GN@kF7Z~FrZ2BD3p|IC#i%Eah&Hst2MJxA_WLZZSigLQ_r?9b4
zeu}1#g_7S{+DYOpLPE1U<gIOFsRzH9bv~8`ypt<NH+38s)WsR(BP{b_d;3j>55iGv
zGV8UJ#&od>E_98C$w$F$KjS(I#kxXb?q2Uf$DCtsB{lgceAjor<4+wqr2Q9m|L345
zJLCTc?3T3OVngZqP(M?HHDDwmwHlw6Z9<jB8B%_XN1@;?92IJTX<p0te&_p_fF;TN
zCy7YV3rUoh&+l_bO40sO@%DIq_OgkbzHn|*0_|Z<d)LG8Dw2YX1!;%th3{2SNZr5I
z-}hh8uK{%ggq+4Se@KGY+x6b=>+_3hFX-J+N-Z#=;^$-evW?!+?(2dRu6Cu0<RAUc
z!DiDW`>oy&7_?zDvy~t%f?{F`p-(>EgOL<brdLo&h-gMzt-&q7n&uI84gJ4ORB7K9
z%Sx$bPPvN0URByZ?zW0AqTNQ{XyiBh#DAhXX)X#|A1YH<OB%7Tnz(RMR&^7q+>$|`
zI54<MX7?L_*J8+W^Mi^7ssJ~)_WGl}l;Juqwy^m=HV0<5CcmN>Pm(~HF&{99GZ1u?
zk?+QA5bxqc3?o_r-iLiMUI-16m42L7(N*u`uA)oDis&L|l%jc2NW!`cImoibB79@P
zOZ-+Ry^t>$9|A9R6^Z(nb|t(M!P=B=bS5NE-6jwnmk*In)bDDB#(f+fmAV4^VeC+W
z$g#ptG!I=JjOmdGC2OSHx1CwG$}{2sgnqQPC%9Pv+yR7plTvUcKP=19H=3iItBlVS
z%g?Um4A`o0`M1Vu!JUxy^Jaa7f&XA2N(Mz#fptUA>QH*@9%^i#!Q)dNX9+ETb(G8^
z@#(9>qjH>xr5^h=qr5UYy3p1@+YKu3S4a1df6Cd$gr}oJU^EFQ!DE|((r6E)xg*H{
zbMIL*Kh-&c9SlnuOF-d&`YocZBD;!43;V83fv*}MSnwX0h5Q4xgR;xxtkW?S7^MQ>
zV(f8TgE8y#jl7jugu{!h<04onWNBo7WjSR+29LHzJu7Mt+&9{sfPz8npW!--ooinc
z3lE)H2p%4cN#YP);}QoX1T?{&)<dyI4tKInqMiJU0>lg_+t@;du@)WXQ%Vlu<jQLk
zjOe6*YFlrux_c`>^ooKgFN8Getid&8!IN0&sHx<c{7R-`D7DgP(>gu3Uv<pJ!GwM$
zM+s@1^{HSd#fLE%5C1TRFQd_F$YQuDi?VSa>tUQG?<1E*;zkx|Uo<gV0b7s@DDM3C
zkG3<N1aAW(BnCNQG{gFHO?6jRX$>AJo|GCz2z5G$XkP|AQD73}-uQk}!e1Q_)F)}z
zvWXiPAbIoxqyc6mBbbyi4b>q({$|!cr7&)8m3(bV80QTqOJNpm%2`5X#LKiM1l5Fp
z2>lq9XIQ!a4GVh+-`T!$0o%yEp9${El01Z#Dw)(84TFnHzl1~>Ar6W@B$d?K^ye#c
z0VKjysY%6gEDpf(1yZizcA|7~S8r^w?LETRCL2SI9&`SI>Jp-Z79XRK^lSv^#)Jl@
zm~@?v^wFl#NdpGAtQ7qUWme-z?9$cDKQD*2B=t^7OHwoG#}8j3)Uj{6TOx7GwHgUm
z9+(WY>c&?Yz2x;1m@ot=xzcKJqivzGqMkOpO;Yu*kmoKBNReYamX|&VkR9L@Z|>gW
z%W;>4!HQtX5#JS`8FqwbgSEcfm=b*=<wbo95=d!7vd_}=cY)}vz{ZJbI6DNwO9IKv
z)`apxAtH!5p{0ytLSnuYOuGu?+D9X0lq4jbL8rYF=h~duNd>QNd|g0Yu4REm9y@~N
zyBtwV#&P2Q?em2mv$r$tR+THkow~S*#LEhYgtUxUv&m~Pa0vkWu|LKt+SvdMK*0v3
zjj>LjIN5p5pZ4Xw7SN@Jo_RtAYx=%d1<PWwy7B0ob)UXIUK+)h_+va6p8z7YV)G?h
z;ce)56`Y+xY<eQ`K8y>g1+=ekc(KRk>X5|6<!XeBwF+UtW))`J!EecyPbz0KbA#GE
zG2=87(ppfeR`IPeusyb&MRkecJ_Gfd-X5I-GUAdFHZ$cmc(`n-W?iwoIekOB2Xi3?
zS0W0q3_zElT?PIQnmN53gI>b;gKOtgbRd^A1~M6gf|$w~^Z8*?{3hGXmy*{biIzEN
zqz`p@ewqvx-;lupgfzQi5myAl73_8yd-x*Wi#auR{T-dy0qtkbtO`W@v^FL4m^YFt
z+c|uPZREXj^~}t^JAUmijOweb($e$#_``2<YqE*H@U5xf-iYJJQ;II!&B^k3*jt7V
zc^-SUd9Zl@6aHJ$Tfjd9McywYGQbO^Ra3I*FKg#f+(Y`#a^UjLHLo_UEt}rHlh^Le
zn1Do9IQp>oI{7eXnymZqMl&3U(%&nd+llc$@+dE;m;NKd4waRNrQ_#l{3(RodHV3Q
z-uu0jCiY)A$o^ltI5YeIJZwFzq5bO!M)8}in|mh#MwM1VA}v`PTw&D?vpr|xufc`d
zD8;qwR88D27XAK!^P~`0%3+a=h6SxOPCR|S-@>@C<+=Fsc>MZ2TZb&lNK24$*q>h2
zqxj2#6e8BZuJ2j*#X!oK0osuNy<YWpURUPt$1Y9e@8|Wrc~$o|J1bY{ForuEG6EE$
zeL`V)GKc8;{(jq(6a(8H6#x6#OhCU2Db5f$Vb6|+Wd9^~ykb3RE1HdMPJm6lpyi-m
zrrBeB+LT+CI95F_w_TsoxKQ`3ot*B55nTs7Br(}nXL&d_{hiset*Sz*8BD26#a_=C
z4&Olc75(r_z^|RFG;MK;KwJK?TfxT_X!`rp5;RBCvXBA!DB+OUFqDN(aJkfnXWFAv
z)9iRf8TK-k0O*|wgc~cGT`{S}+Noel{s6qrXkgAL!eegy{lYbygXo&0|HKf?o^dN6
z-l>+;(lETZjK$sseT;S?hyETMoApd-*V8*U32lsG(JM$$Dl&(#jJ{B@bc|G_*jm~9
zRJhXd0obT24c8B>o{g02nC1;Ln+IDX*y=dV=z|DZA^nM;tRovNdrVNcMLuxdhgD`F
zB7#<=EQhfR4J?Jd1u&YGawFs>QA4#%q^f2oSqpm>!1%D-c!`NJp-~RqvFIS~w!Smu
zmvx)uu{Z`)X(2)UGrSg4P?*xNyfwF$#}NEi67#(%77XoU<F0-TEA8Eyp*lR(6mO^$
zp+K;-_cA$wW@4m}>$l(KC4(b0dY=cAIWywGFi!C5)}%+(Lhg7f=@=Dhxz+kf3Us+p
zby|+j$=!h*d;2HqOXugXyf(qKt2R@p#-Sj!pHN0gs@%a7c&}29l?^-nH-yxKty(GO
zrUv)v)0R>)j%A8W6!m-><0(-AnA+YFn%t1=LhV{KIb$nyuSmA4?c5aeaavjKR(k}K
zM1uk;c>rOniwkj6;zHV6lj78+tv+>Bj+xFe5E?qE+EMj5LE&0bry^}Z)Y#D}xn&z(
zA&xNWnAr0-S#;Z+ZE5izvhS#dkYuVu`PgDz#Y&rv;JqKt>g3yeR}pUjVL^;1PRt*|
z=Csgjl!AzTp9@B~4mspH#hv%Fmr<?tHUc2o93ZyRw~#RwElzi0WRN?8(J9p66e8EB
zs(|QeabNSy6hd;I@proCdB9K4V+I1o=z;Q-%m^IUj@7_?1-;zcTW=2!GxlCrt_sWM
zXMPF0D(7aWo-{OhmDkx@+b5K|NYVH!_n>1PH-qj1#;=e0%8USFZ%r?z?k(V0`HM>l
zf~+8yPk?k^v9hZ@8gHH%L7YSq>5)UK$GEOOybt3(zX)(}+`fR1S5P~FQ_OG}wf)~A
z)jr{acznVo)N+gwmp|{~)Mul}*;p~#rBeJIPkMXji3o`DQ(b;OK1ydO-c71-5366@
zpFj~g`9Ek!d5`RILTIS~WQ>0<FX_*fynk-esW?F)Jfn0!L1zZtE~0&KQZF%`Rjly<
zaQTkfvm*DnulC)0X!j^#;(+<oIusNrP^FxNK@`3quz3xHPUg54BVC+5Yr-T2dJB@N
z$$ewd&=(`aN+ADgZHWXM6HLJ{T=ZA(pFI^=vdml+jf2V{R4*BIL4pzME1Z#)V=ud^
zgJ&BAB``#T0gnE$_w+B+qssqWe+?r;A|p#j8VLPZ6w|ZaS&?nQK%r0|>+1_(99VoT
z?KTp=)+T2g%6jJFfUz?O7y5(LdzC&fN!gq!G@4D&1pT=$#})GtN!QI*c|-hK>|>hv
zcC9_gKx-GA%w4{o+gx&nMtV+2O4ebYkR9Xn25_(nPLEnT;14L2YeMxOPc{eZ|6RIN
z@pLdHpqDqYQg*S0qL(9JWMF`z7qfJ7b|K*8V1}ZXF|{*yu^?b#;`pDX&1Ef3du(we
zztuVm0{Zb_ug=b@I4uakY>*w;dT5YV^tB-dm`t2W9qHm!@s*jn-MY+TT0AC7ClYgf
zl9k8Dd)ZZWar~Yko9uhiMo7%wNLivU(Sj<_5OhUQdm{=elO!i4!VDn_(X4@>U5lu8
z=D!ePLC=ZZgs0Y*MiFzMB#bE8e=>6ZFUHP+Ika`_(y{IA*xa#g+u5;gYsa>2+qP}n
zww-kD*WFcJbx&2Fwf@7zJH~tlL6w#<3yOmke^6YxQAx7+aPN(h`MRI?)F}d#59MBj
zcW=4U3_meAzAI)c8mO3XJC&v9T^`a+K?MNRnNeAPo<x)rM5!yC5I;EKNe;wO59cS$
zZA{A&RSb%3tf9aR928~n2}+$VN4)8@pF8voVa&}mbQcdqdwobLjDmTi>55x5t-qeW
zN+^uS1PO76(I>tgu0p9z_-=teBu{F6xe+E(d{V96$u10%J0d7$%EU=&7i>&RiBgVo
zCXBpv-QP}EDLv9ea=|bTu$(~-$-Pyf(Z&cV`vL>qH)K02Ea|(_CFJ-P{aZ){D`*8`
zB_#?Up@I@Wf2DPS%u;hvptZ0w83Be>H&e|SWzwqOd869ul-N?n#a65Q`oe+JK!!N5
z@hSCX(ip#?EWAq;G-tH)fIWOA&e3G$gGXY{j{TrWh(UiPZj`d%7q-F`1<3MqM&)?Q
zw?U<$;(#l#$VK&q59-D#S;^}n$`?ZgfrhLB>EMr2k>4oWE0LEZ33JLL0|ANEK*Wv(
z7c3nJC4%nt)FeFSB^?SBhNdNj*F=Yh*B_DHBN^l9ZB9$wSL=8KGfRU+&3>GnEv`Oq
zFH3Yw4SgP8PINF$((-aMXOHQ4c9_X8v|CQ6;SJ}jbEZizI$OAyOx8PdWXZl+{e}*&
z^KS6hyAHkuzn{zXp&zC%{z(R8<+pI6H<G^{HDwa9`3b@i*;S^kNBNK#QC{IwTF+wU
zjUe-PRsgrJPU~joW#!!itlCf1Ki~HtuZ7(~7quxju+=9Ln;;M7&fmI!O~-t*{Nlmx
zw8_%cTtG}C`&_l<GS6cD_I#@U@_M&sX;1wsG^=_)o>#Xq9F8A1=CLB1Hx4mN?YPuK
zOS=ZRW6Q{}1S(!ZhfX%KuRF6!&3p{Kj=r;J^I-Khc;7!4nazS<+b*Ahcj4EB(>;S|
zM^yAuBY*?M(4#;J`%QO;KMLiEFKi--u;8IjLnC#+!Q_q8je&)5yKbfdsKK;*Eu|`!
zr`Do&H!30!;uF;S6-M>uN{hXyqPwWmka#I^CL2yNV`(E?q`TN%q3j7YUWSpq$LVTT
zCc9fa^4rJ{OXa!470Nc#%#86wf8ISE{W5q}8*55$J&5s#sBUCbTZfY5s^gSzM7fIO
zNn5eXAP{9zf+G)Pw_=~;<sP+p8wtWA)alSi3q*ehi9gpPtDEC3M2y?u5W?610fv?f
zjXR~$?84E9c3J<;aLF5wf(B-M)XCH-ZLBb^DR<>MhSuy~cj!8H)M99^E{t`+C`sTl
zx*4<TPE2z)9b4ylm-s8uUhMs+W73lfoyQQGXXwpw0qG&cYWUVZ4+FE~BfZ0Dn{P$5
z<ccGcgcGB(Gp>x>Hi&UyJQq(Ec&!Vpem;G)v8Fv&rSW?(n!R&d@itx?FRKm@AIB)b
zGRSN;gSAJn!AdALh@IgAeLU=&F1^=4dgzte^9B4Ve<*}#l#kUCxr%*Kg<&=seW%iY
zHD1)YsD@DOVtz2`u0Jwc@|<ZaZlzLkJ_s}nMyDim&9(m%1l_F<tj{;sP)<{;XN*Zj
zq=7qBW{r4{HL+aa@|XQoOQ(P0t+V`|#}d{^EUY-a;dk&%Q-z{{-{9E&g4lw1AF-Rp
zHB$6)dw@{|B2Od$WETW66kUBAPVefOwKSyKZbGnbPeCW{_f|W{Vli5&1#b4BzOap0
zsFV{CFn`^=uq<PS%<&}}0SOM@NTbBBHo?rpPgJj(w%LZ}JlWYgs^@2Sojb>7?!0AI
z$FybSVU3Gr=@S1HS{kc59o9w}jb%;{eH;548JBze2K5dKF&hE&-Iq&GrE{5S-da59
zPyTjB{Bv!s;nw?2yYnXN=;fk|8RM>OmSxf$Yz{z|`BxTcc#`K?WWIGyHGy5%|1{aD
zDX`$d3PBlEfyqh+IxtEWJ6C*x=izv959Ry)<&f?8-QZ<^g1qOO$q$FAkIBV%@9gce
z)={s$;KRSiSrN{**>{j-&tx79n-4}UbX6t@5Lm7QwC`{zI)ZwerCi+^@ZeVSU0$3Q
z@ccf}$MvtKY;Bq)>gTZT@x3?jD47PkOc+ll1`PtYudY1psWwKA?&h|2tJ{U!!6Ul^
zHa703w}Z2Pr(J5y?h2!yDy|Q5J920_i->S+pm?&3q3BmK3am@mxjXvJIl&^YEVX#_
z>F1*^yo0ZUS>GrY2_H(`))e)-hsq9*Hc9incrs_PC^Z$YH@hms8%sgDD`$4#luUPi
z{{?^RK5|*<y^fJdiJDry?Wv7&x$(Uo4PFPsm0`49anc0k2pfH~{pMpevvH5#a^kcg
z>*qBW(f@YfUTHA6@@a|wmFg<0()a2reM|Ve++<6Vz^ScJIzpLh5%@PZ2qb|<G2_ub
zK`?&t1+iqUyA6JV{&^&#9g7*zA-2lV2idk*5x^7K<ijWj_f+&0N5Gtqo8v9+NpXF&
z>Lp>}x*34c_tua*33n8}WK9O#vkqsdxt-_|+NmomE+_BV9~by@`O16z?)|4gZ1;El
z<nj<Ic*J>$5vtsZfWzFXiYjxlJeo~$0*#nR1^n`D0pU)GKNpZkcOjx>RQL&t*wUT6
zgsxA6M-QmBY)xnTVf76yoF=K=KyaZd)rInYe?6;e*gQe8IpAv;=Q#2DL^?>W@$f{(
zSk9N2J<8c#&@?i*cwpd?9@>b9qVX*vlEyx*1(dv9M<UTzJ8u0<&jr5b2g5pu_9ugl
zJKWaUt7!7X|GW+?3ZtuQJDx59iySEXF78QoQX)LIbR2<3PgIsXiwO+Ig0RJ8A5`nz
zYCgC^<l>)dLB?#H-p(uo#MR327eR+DasL4wdTzP8=LKF>{-<nYcGiD7zhYpg|F__T
zMb-ZcPN-p*XCwgf8K6qVor}0XX(+WQCnMYUC6PE1B2*8nJ2$@DZg-U<G#zzb&~`_}
z*VS3sv~dZ|7Y@jG$<HRuN4KeGQV_En%OB;Ea2sHVvK`7BrEUvh<R8u^<L!2ptT~HC
z6hVvy2~aCzz1GQ=k(nBTG-8t>`9&77{}O!<kGI`gKl+Z`%$^+~jMn{i^40lzxjYZw
z4M0?ap=YzuLl}{;XX{I(UrB@%D<vGCIIB^4rk~0x@+&l*KGsgVKX1=7Gr7__yv)&M
zaig-4qzo-myVZrlo)29f2r@PZ>apXpVbY%4p*lsSek1d6TwggsrZVyG3kj2;RE*pk
z*~Cqo+T!>g@(@!td#-~?T_Y<Yx2$duaYa&N=Ts*BVo@b)<@U7KMMxNNO0S{;7;w1Z
zQXq}E@#~^-tdbLS(|6`{oboKOQ+)H~W)>?#5Tb1JhAzDurjQM~a51U_35<aNq%9IA
zi4gu+>%NXk_;_1s)ez(`+n0HE2?F>%k^n+tUo5K8IeD0B63Dy7Y1MX8F>H*c_AhWA
z51o&jiIqdK_%1ftZz~-MtW+fV5%A<W<yLE_Q|{CGqNQ+fKPs2nQPVJpBdsKBVWW>M
za*q@HChaq`E@!fIr{?-n=yK=rNNXss#)-}D^>cmq%cj8MZv~ZRZA*<*<$>ilQuqKV
z&iD#e0$R>_Or!)M%mE4Vvx)B6ap^AF*fz}r6nPQBB3AIb?Rtm^hy#HhdyR%8k_Z8Y
z^QFr-XIuqY4E~6a>p58rQ3Y7nn0fZrrmTQWD)RC(QXQ9|Lu+8sr@J+Dz%yu=z=*0Y
zl*7Ky)dk=#8rNx@6vU$T4Dm#v)R?ZB1eL0YM|&&35v7hfoeUhLro+s2W&Ad~XWZT$
z4kSVS%1tw^%H=>wbo^#iV*|#aktZe#MT+7pn2YL7YsEffVa?ykN8J8<@e)xGCcd=*
z$eRs;+HJh8;@Ii?V3DTKn!pTae9m{tLFUI)%j_T=eCZT(T`l+|<OZ|0R%o-i<@GM2
z_L)*5-ce1o(ukzQd4$49Wsi=`4965Ee6&Oq2*DCd2eY0J@3kV%PROw+Qxt~e`lSX?
zGE<)IMa1X6Jpn-O^eSBnO_4Oq_hss-S9+93QqAr&+JYDqehfqHeqW34?{)@xLIV83
zK}iT<3UuZgrSEHb&=$g{5wEk7!#l#yz>EQjxEUE%uMk7PcDLol8X?(+>0RW!yjmx8
z0KhmqZ{VHM(v=FGPod%Y%E&&tZHdYWdU${R;;)q#`0z2^zY0O^yOqf>cZd?VKuh=M
z(Fc!dHa?3oQ~r!AomAVjQ*|95>G+w;Mx(X=0viY!eUCRhg8pv7LhsjOxNqd~@J!h-
zCki8?E3Cey4AvaY%rUcBvDL@RWNy07fg!a+4B288HFl0t@;DNM;%@+6^=1%#f|?<U
zCnh6T7L|?z$WwBRFZh-N<eu@mlhn9|lZyZQu^wxAJjNiVR>HBHK@k+uX7;cNTEqMN
z!L$Fk^@_5w8*`QD{`>jXFhW{z%ir{DJ`6xAm~4v<psCSn9w5eMpFKdZ$Q@6mPZzMR
z9vu7vT<H-4`|XT8=nQQ2*U3d!C7%L3yvmiJE<_ff0c$^DLOVDt{=<waYVOR!;P+ep
z8HF6eumC>~80Zcg6im-E0``DQ4?k{zKZN^uPdnY~Cvcq^4#z*Nu0I>;KV0E3{Ah9i
z^ZjRWwRAMH$EOvs)N?cvGBU6+G=k#cfpT!PH`23$a$Q?g)%wqC<4?r<On!$jbvH@V
zN<T}3Fqq@zf8H9w){=0X;togc)2mLrw@GLV)tcfOPHUVY5X6bE!ABjYxr=5w@@D5p
z;p<2J$s*E55b|52jLXGN#Kfgblx8h4J0pZeP*O=H=3idJS6}Pzx4W{77_u^Xzun&6
zFN(os3)mWx!sH5jTUnDzZNo{K1~}o89M6%n;wDTR-fFkH(n&REUDu`5(UvvZo2=w*
zM1U9sQbD$v^m&{p#Yv@@%Bw!b4nGTerxOOJP$SM;R({fXOTy+@US^jFOkRj2geN)9
zwrR6u@{+fhy8909HfdTE4Njjlf<33`PCU2w?%Wll4w8d2wpNP7GN3~&BU-ldnKH<m
zJ(@*1X*+%O?F`n(H!k^aTRMLrh>epSrdaD8{nL7DoGt3Tu3YU8WPm~8Q*f%kpTn`c
zdtKDzRo5B1Dc5#twKJ*Jj(BXGV;n>vO|Tub7h$z+Hc4Qr*U_$aY`m>GvDYrqlV6Pc
zM@Cb#1p;<Lg(1fwSg`>gU&h)pW_OmC;G~x$D{=R<5aaDz4)==%dtem-Lf+N6v!uV$
z9>|CRhHh+Gshk>0xZVO(YIwq#Cc>!A?@XCr54*&{FckU#By5@Brhy{Ch|qme*%Fk&
zjEaPHK_M%83I>oXccRAk=ZueL{E)NKq9C?{?>jl!<N%4FG)W?mdj&f5=W&%4^C%+2
zOT9hOI}}w5zjTIX-gOW1nhk~vUr+Y$^YIc_&uJJZ?s2~cE5E+6--d^qUtBNqeX%4I
zxF??ak9Rt{PXU4Z!>HS<9AhQ+V1R&C*YhsKAT64yc0K0_CEwT$5fN>Qece6P33sua
z1=)<5GS=iL+~7Xu6cL?~Zq_%Jguy&=Qjgbt=T(*DO_A2?T!ks7lTZEzrmEl^XU$~I
zL|}~0h<4gEXT9NUR?>y*O^We5uCY!Ns~_EeasWPUUFgaxcS)7yi8o1a0lh{&2!xP&
z7T-L@FnX}{OxaZ`SSf4=u`6Ez*Tf-J2_Dm)KjPV1I)E&CY@`zi6xT!{!jUZzIjTjf
z8AQ8N;%acI;D7EtnBF2>UNJ-=TtBVl{v{rSXC(*!>!r-)|0s~PqhefZGiA^c$R4d>
z2InyfVG3_xC*fuC3#6gEDIAUN+-s}Q^X;RQRAe-F8gat!9aSO{s#}C^oEad_8`n^f
zwHvu64C(g)H^6uVh(bG7b<l~_Unkq&a624vdVJ?{lrC{WyJ_f6X3!d7S`R}>?;5uD
z01jXk#w9)H+_ru2nXW?{hf4<TcB$Z27>?;|XtA9KmUeGeJTq|}R5^1reG}_Hh#=N*
zQmS?b&=@G%xJ1V=>_OTBSr6Z8F$v!%=8*#GkdI0##TS12AW`|n6igMl#D15OBF{ro
zO$utLl3hmZ%Yu^`>-y@uBDIw|n4$J1$jcp5p}-*6Ug|Mj6aDh|HG7|9vl9wsYt}(y
zZl#Kg??Sa*Ql_4Va+Hb)vp%yoBM^)Db$hAiRd4ZHFpgKCHr0);-dCS@pLm>4FM+Im
zPU)G<%g}r6Kc7Vz=>8d(KIQ#?yIvRaRq{nYb;9hOZtmBe#xIoOp+zT+HRHo`MqhE(
zj_4AQ50p~*J1n^2C_jRN7^_C!Vda1&GT=L+AnkVtBO(1X9ruoeqyb_-f#0p&hyJmI
z{!>mVBjdlu19YmcNBtmNyq;B0<gJN<&|~v{JrKY;vqJvx0a|3m<!72MP<6E^CD)z&
zKRklbSM!#(Hra&rdiIRi9DaVmJzbog9rL?+%exN-WBxD&eqlNdP$>{DG>BoYl;8f*
zdKm-ryv8=!I!GCC0}#a_Vn7lR3+2nS+tV~v#fXP&hDg#xG`27A!K~L+7b_Q!fmWU=
z((!DbuTxLjpKs&uti1gKbo9M&3lQ=P>tUvThLjb9WftdEjep7#{l(_^`Z#(Y+ENH&
zLvZy_e%HG-N3Dq@t}8F1tTIsEk(Q-bSh+ZNnj}H^sg1ooX48@Jy3)bhR%%ht4eS)A
zoWlXgBA`x1r<;G8Y%1NG5!1o(T1e|kjbIEw<T2FhOBb=I2(3;>KLH=n*G^tv2Ac}e
zM&MpJj?_lA76ap(G6O5eG8QUx^Kb6q5+Cx<5ltPqJoS4`&c@<+vmbc`_*!B76+&h(
zh72?KH322vSXOc`y~KB7fXSjr*k93<ZKcvwHd(GPU6_U@GBl-fE`qwGA|bOvXcmT)
z@(5F=yslBMJb$!?)fT^5CT%jwT?Y)ql~Y@-{BE9({z&>AKtGwJay%yR+p@ZIG-7?E
z2Q`66pm5Rd3bW*lh4#)Hmm3h9sCSQ120+|kU}PBw;PnPDhle#@v{VL4aHhKGu}m&#
zcpBS38+K|05lu3B@Ne7#8c7}>Ig!I^BJ{8fgqpVX9Ruz1@TlMl!BGUF5JW9K9(m{>
zf9NTo8xk7~R^v#MGqGYg*7_gVC6ZmBCz-y;%Z5d@mUUwh^c&&&`i2K=a(016U~y#P
zkO%o{+;yWqdFs^elk1Mo+7Iwi!yR~A^Ssoezz0nFm7_&_6OG|$foR#Hn5Dw%1C)*}
z@)>q<5sRE5dg?ybO6Il-_v~iYR*OlT9AV|_C;l~HZMRsip6mWy5a}8gXVdg%>srd>
zPOV`A*c7;ofWX@lX}!jAU#IwTmUSNzQpMq91ynHXc$WPX3ZtVXEkKIJs0~Ur&TT`D
z71usjHpcY@Mw-l)iE>f2G3F*Gu#MK}C_m9q2SZhVYUxL}<!rqs7T~pT<FBT%WHgEg
zY6VlJ=2UgN_|ifwbvG#Zzf0WbQjA(O+~Li@m%>I~j56IyNW`tSr_EL6MVDh8lhPNv
za{YsS%xz!@ZBsM+GBX9=QyHu$6s+zgv@`4r=$Y;_Ji(_C0#+ieEsApd3@-KNiITs8
z(c1zd=K}0;Bf2L!*qEKXoQ}%ER}ZR3w~Sz95Eqn1<fU9|$VuHZAmQ#9@R>xA?D^tB
z%SUaF^De>>=T<>HfxR<Ug||i;+piRm;OG3{%ro$5G9^8lEJalncvEe`LiwH&Z|fcI
z_5u7LC>e~A`owYthMc|NOe{@*znDtMXw&?fj4!s}Ej^P_G?jD9E{f7kxOos)bMb<0
z9HS3d{p#)9B!3pEdOjZ!Q-9eTy%~xf;E`)tbd-M8McL&cbzKv^ooXMy(yWXop=*vt
z$3pSI0D6YlK1u$D`rJv?zok|#$K8;Ukee7}Pe`zxlpKujtC)ge<mR0)n9krk+Wwg@
z89N$b`v*U4y(G6VwXpaIaZq3ca%Orry+0gxrBt^cPUb<6CgwrA8WA}j9WmDk&45Bl
z9$}{#bL5PbVXLq*cpTp!oPBnO1WT(QN9Kjko8_R_Iqw_^QcM**=f@Ady*tD0lwDt5
zt42=|ZF%XYdW@Ox@en<BGewQ<_f8G&*1vYSC3HOYw%6%VPP_s2K4mcfsS^LZ3H<j-
zIn%$le`l)x@A#7+>0HBzN#8Xz4(UP9Lm#?;0)e-Y0<A&1diBKC>0;>j+cxKcT+_kh
z6{#0J#T3l(S5VQcl~8`Rwr9KN?U|@ZcnN;s9_1fvqTD<sN&%x^zgcImYe&Q-(8SA3
zKdzP!AJ@;ftI$#-l9#Xls(;jeye>v7NEI4Mn3nC>(^A(j(0AtW<B=N(Cm@%FP1KSJ
z6bKn|Nw|vK3FIzx%aBSy<=JDBL?C9FC_;fc%gj;NTx--XSGKO6XwxDIn|(x^wBvj0
z*E&U~-+aQ-d`y<~h95?URT}FzX*c7UN-%%au)<GHlO1IXM<9-Ac9;%mUKH*GK;r(U
z+WD?3eQIL5B-~|)Tn8Q3uMZ}PmNq%ler_s_U9#LP(=T9YuB?(%?ZurPq#$8OcL;P}
zk6vpM?8LoR-O)9((AefV22d8F;OSlfEvjGaYWShE8zWymUooNnedOHFA0my_1szq7
za$yqe#>0Q$Dx6Za0ZmfKO|Uf<<5zN~?jT-GS+q@Odf!rcvZ)uRTql*I@bYds5R?c-
zvQKbU_vEJ95QR=bje`KQ5NVO%?Z(Hersk&Ym23TOmf}sGC0p5W3=Z4e&}`TdC$OVI
zVVYz!rkM%<SRZ<x6_<+AZyHMMG^XJ`wnKF=piLRcW{%BH9qF1OasEl79+`UK*57$L
zhhEC6IhgQMws;tZ91DRacd#C0*XAC1O5RbnJ5!h<rZ5cBa>tp;+g^oxci*&}7{8s-
zJX^2b*Mo{psC^BTgPTt2ah1W4GtoDizmPAjVwI(wB#$*?Nq*Y(lCc8ZA--?L=cwGv
z#p18BxdGX+A$PKo(LIh<H7>rkoS&lC<Ta|5^E3RQmIAO4)X-r`m9-aU&&l9Rg&PVk
z)eY;hMVA<qv`LWI=hTAxTR!7?cp-!&gI_$(2|>1Hw6l_G@6V~hYs8ZrN2fd?o$4|&
zukEuQk=1H{1sH*uYcrd{x%sD)a&n$N&{M(Zc5egMZWdP}>_e@Q-B-_rrn}Z6q$*rG
zZRnX37}G^uM$7&a6V5K1`MYx-XD<EG-PHHb0M7tLQq6I#Zy}4tyi*(ZPOU5w_O;Kg
z8!uyu7L!+zWdLz1<&c?cMFQNhja3_NqyyjFYGiNcd0UUY4Iq&DhCyS$@m-!zwQ@Z>
zHHJY9H<Q~Fl(Zit<Tzw7YXs{-<(0-e<RFzIGX|hs1^nc#oewZxwn(xfD?cq5t(G`R
zP`^(1P{k<y1z{KC`BpPIcGMz@>($SQAxI;2qT@dZi+h%9LrG_pX?d}fX#*Q~z7-YC
z+A=ihTYha<#!inlwf)@2<BduW0R!vQv}8an7;t+x3VoYU6chdq#tMG{NnyTLMpWW3
z=SU$R13aV<hsxh<!@{3x!LsGt`M$$8tGR=1WYFYJ?(r}#kC%e3zhhd}k}2oik?df%
z*!1S(2p)Cy;<Zh4=c@LMLpKt?RDSeX5;91ml?CAPQbD8kaq6pzAVnE6A;=nW1I-TS
z;*k2<P7r8|aR-d8W&`JBimL<4BL^xKPJD0f|G<U_v}Z!kRbBoXGeXC%NSf{9>D*tw
z+PrmJnjLt2jS|ok?}wOFdlM2|hIJnv*f97eE&6b}L!$GmK##@obVHA_>*{vJSu}@|
zCILm0`ANJ5=}w`g1Ev_vb{|h_m>15~;#;}JC4-?@2G|EU08%`ShaRBrj!BB6^R>Q~
zt-b^%gjN`4)dlNyz?=9q;xuZ}Uv#nba2JgUB&zlhc>2tz{2SVVZz>o1J8%k0f+B8=
zF7ZukKCp|c47T@{UY8&MIe_U9#l(wWW^@tc8y}*m(g_j`(Ebe(BJ|H*|M+bGnQ`u`
zp{h{}5lFgU^yRLa*xrEvB_urVM%r3Kn`fxmAJ|+bN-aFw<qO{{Ag;aKDqwgx5=e}k
zaX$Z81XE7NJTrDalPFUTzl&JhIBv8*YN|%{Un`v*FV7djhvDA<3B`&a|JW;7{<~K&
z&@ugMO}<b~-2sdBKbm~Xu-;$*;6!F6m#x*X9i>bilFR*0K@&bwK}{BwgV>SvVRx~w
zwH#wU5sgi$gsJYYlb<VN))9Y$9{*-^?``dl50)rDKh#nw|4V|9Is8z?p?%_&D}Q*s
zjKcEmrtR(zl^gv&QlwEKnjM7U$Fugy>2=oR@<u6Fj6=&Gf<5ln{n=);jh)r|CKhp3
zQ#m!<kLY@_D0}Vw_Zip{1Q9(W;g4y<Q{57Sfm>|efHldT+&53!)Pg9Nl()IlX1Ye?
zyo9Q>i@={t=eFxu>qWYgs;gAgUl2niVm~bRynw<$h`l^rw>a>1r~ylxod)vJKJDu?
zcbn6Nu_=MgpL~TXaORXE5uSOX=^fTDN+U;R$8s)g7hoD@@Pk#yOpeDM1Y|1N7v*|i
zetg0C&FtH1Wx(f?lYT&6CG!&)&B@H$5hsQgnXFpks9O$|zRELQEJ#3|gtLVR;1+67
zz+@<t>zjVteDeYnIGHsvNl^kd1GysSA(x3F9)iX9@hC%lO=rsW+9k1#GW!Mi1Q~xe
z$&?5=okOW9l1m_g%kTM=wO?_fxhn1H=zLHXHYRR6u5+9>Ru{~66kNH%&<TzY3U->E
z?ey(fhDDsuGG<d@o_UUx>I4i6ZOfHs3ptWCA>{T{<8WJ^Nli-#!oDIWp_AHA>R%f=
zcz!Ow_1qVg(lGvPwuV3HB8-ODFyS$e(fPlFWDy#K8|Jz8sgI3rOjbU2M#Re{w*k#R
zvRRlaKX$U`O<y{iUO?d5JWhR#wQlUTl|HkO>+D?oVi-0zGaC^*#zDF()e1#$SfVf9
zePlx@_H(s=lW_z!jCJ#Eb3b26=<4R0y1jw}8M3FdGIRb3(~HFRf2=BmE5qA(^krd=
z!cb1?i%}CnC9(tw-vUMh9{~~ul)zT#wDDwHb%<&M(;7d`HTMpx2|BP&qawBz=js7M
z;#7W@j9HsB-4Ajmit59YjLYE!GkLqa6u#-T`>czz?Ts#4^vKTw-V)7;g-*I%5X>KE
z(=)K4OMI=KW>hm%mffaaQMu$}#G-d*@~FvS&#*mGl-d~X;B4x@Rk;pgtmMeAl9WD9
zT^nYKzTPJugGT!?mO1iK8;m6Oroxdw=sR2}1wx-)5Ty77J{!ZIp8N3v1G~KFT~OTJ
zkLFI%@^PW!0bsMMAiA(DAo$0&gj|x3V@#7bx%hO5-q9Wwz=|lgGSyDGnCU}ygsmkj
zo#LY}iq71hr<wpG)^bf#(2;<VqC^_@I$htQ<cS>Yykr=$@2(B;a~V>kG59fg!9M~m
z|D{{bJd<hh#5l(x$@mN!oH1f<MMu+m=t+x2?IartdKvoLlW4JKmihN6bqx;Xw<}r?
zI+{s<!RJbov?wV9aQe3WZ)z|!ItjGU6AA-QwP7><s(L>9-t?~`Oi78hHq0Z)(9WQ(
zZlQ*S0nre56L;=m4B?bTrc1@&;>1Wv#q~)-H@4N)!$ffaU9lYQbd-MBacq&ANb}-;
zf>;Qnukz=OoCdI!b8F4Qt|N+k_*MW$s~;(xib*A+W*OT)b%mG|*>!CBVcL9i7Us+W
zkBE`*_kN-PItyJdFpeV?=yI>FU<yc1;aVnKUTDj`53+q6&6ma|6|&qJ-%P!|U$s@M
zxQO*3itz22!C-x37fT)vHBi`0XMihg@?8tfZ4ZIgeht$cP*P;^5ry=JG`TtoQUnV!
zr8x(?k$qFG5W($Wkf2aJ#TvW6#wIP6rFaw|i+$r!skJE&t)I{l!MQP{(6G9LY(s@w
z+#>(vctDQ~x_7!w+5+K~%KEeRvu83uwhcgr6ZVCrvV)}L_lx>N<nmlZR%y9Qim8%=
zEWVxr<^65phsrB+=MPsKy8u&8n#9RGtE8<lv9}-FR1-*K$Fr+4oE-?1KgfZqT*M~O
z$<_D;ao*`So(B;8>LR<!Di-Uq59Cqfz`HjFEoGAfd)w|6?t4@qGLSx+#6-~N(%%X_
zMfxTI2P@$Brhjg@jflb#9D0BOQ>LZcIyAe^USyw691?yD!n2_c`?sTry$n6rK+6KU
zy$2>y4g!;dn^9LS`=0D?46kBAELU(9{|nmRoGECQchv(WRGBg<TRd>+LD;7CmyLnr
zuys50P?w7FdyfTI@p;$%TpO;>&WlQ+5LA6YVH=*(vPbh*x&3*T;5I-NL){Kt_BEc{
zP}v!lzpGWO>*3q9Q%jBVmKJhT`UL)3hwcmH(muTKJDunx3@47ZEic2#l;X<ijN~WS
zQU>i?5Ui6bG^e8Q6no-{AkX12G3rC6@}}QOmA4%gV+n`CyWU?fc&X3TM*jN>Y7dj(
zq328qF~aw;_PAMJM;o1JG0LZ&<9OM>?cG#)cuUNm+B~1CVgus7g<TSX(nh8`s2Wt9
zU%-Nq;dEl@taa$GbZF#OED!HI!_=^Gc=UTst$V-8HnJ+aK~sc~vO3JmmJ+HqecPP2
z9S!fZ$#>G<GeJb&nVex>WZYJwrNXJA)e+GdLzmI48>{1SK{5lTZ3y>ZIw(z!*TS~3
zda$1Ng0Gwb^KYU?*%MEJ3FR71yVC4Ty(<0AdFC3A5IZ=K@a(&!aX6Hpo1o4=+ZZ<6
zE+6PJaN7ka0mg{Bx0}fBpJx~r-2NC|HH)nO3Yp+ZU?5d|gI!q2x#|Y|^oWp12EwPA
zG<Z&lxv6W5z7#*=i2*NpA<HWFH2&{tf{~8ypEgfM`hRWOE>zWUNEbo*nI?4a6qq2>
zJ8Sv4yu{MkoJ{^SS7n_UY8qBr7C}JZuS<}AcbYs?Ll~AiU))BhCpm08bbShSQboJt
zk@8NGR!3t>cLWfJQQ0&-1l$RfLL}z0YPcSEg@}t4@t2f%zYSj;G{<*u%MhJT%gA_V
z|4sXRds7R`ZGwS>Ma5_yV+f0w3E=SFi6GIIDv=;0!gag@j26ovUwAWi<m1`N2#5_Q
zqOGPWYs-)Y7B2l>3?mjAoP)+7;);^5STc(uba{C}#KSWaV{P#ag$L8P7+Hf13d-Y0
z;9QlDBgO{G$7OVi%OyqxeMKGl)~qU{UVdcSA$UHgko45*C3<qo4`;Eeuj{cnM9NM?
zrK~=44D3R3?m?F%>_frhluEWbzPxP5UuNZCwV9wTXVEaynb`*xw-wjNjD}5DU|ua7
ztB+2w@8rXiOqVnh1f8@(Bqk$_Em_mHAd>#H-&8qYs!YZJ#r4WdpwkOLKwQ{hmO!9>
z@bp4$S$QN>Gjft!Y_nqZu-7FZk$6};BHoVpb*bBL2bM-1R38PQTD4mT<k#6Eetsn9
zBHX^&xH`{%d%x}y+e8t@)g8MpPZV8@oo7J5J7cGE)L?tO)X>(ates@tNON;?&qY=B
zgaelApNqgnOmDOTzf=_yTW_IasygC2T=kj4EQ6hfFKdxtA?ea^Bic7pUfuE6>q&1f
zlM{aCCfNWF@ht?b_Bj)97O;Xs<Wpx%10&1*7vwjpX>We#6%hR=hVD4>7?>4p3O<7~
z^1Q4@Pozw&&4k43x8)`)?+%&_gjnioSvs1IX6>#An4pzaXCJLuZO@?5*+lcQ;*o`g
zDy^gt#)rMe(sv%}$gudf_IX~ox2r?~B3kI~f&^ruwe-9l=G&k<bMp<*I$F)o@n2={
zXN6szQ8*T=NaG02x?`3_Np((Nww8+@4frW*Vp9u{$2nB9f~W9^kX!@Sy)-&cqSgn2
z59)nRzjy^5Xt8trZ%f+LCN0<>vZ;o$(PmY6#iXM@P?2kk+OnKq8}de8F|y@S@^Xs=
z7E0M%+oWZwuHtsVYn$jg0n42jVv7`x_If;1eB$5LowGX46B@OAS&oC6`cXTU?d?LJ
zL7A)Sr6dFue*yLrBJlw|V&`mQO*l`#Q@6CG_tk^@0$HV%SzVx>;NY#pv8XaRf{|fd
z18jRl?aI!>gXv<C)m{!gpeHx|BAuK$)8h%dUsDG=RrI2O%Lx%d<n7VPj!>F_SmJHx
zhvnth{aw2UD3+u*0UC(<w#$<3z}#rf(7~&=M8f73x0kcHE{cUyF?@nnnP{*~Z(|YZ
zc5k|8kVN*9J6&Xt7}iTd+?0=B5`yOoIjx`Qj14?%?B|asGy!<r?Q%47z+m6&h!FwQ
zm<Vub+^db}h--<a6Pmd4k()b1V5`=#TPQ&UFsY*Wdd9+BG`^e($x)_;{!cLR+@I`M
zn)Fv2$5OAjeUwc(V>{#l2+!)xH;6aSwbr|8&DdIW@9NDMHP(RST=s0NE>+H?vCN}c
z%%NF<DNe&V1$qSFtMhD}fCVB^+}5YAH6Gmn#f;5u$mcB_oAyrcPdEh`@_OE;7d3Z)
zVO%}qsP(*HSv}*40Sk*M$^LjPW{-nOsCb4)8i>H`A=N}%o|+k_j<;|FKOReQ9TZwA
z*W-(MdGGTplD*DOdiY|G#T>B^SeZrohj(OC1%!Q+tNV|3#OZ1GEbr<JTfA@Z%I%5y
ze=NKID7~=#Yw4v?)#^X={I9DYdVaj03bnX6N~$h!8;u#KrB<?Q_6EMLB|BOXx-6~u
zJmI(3@Q?O#Iba<|U@Mdv<ucsvaTT@Fep~Zw{V?wO@J0>A?F*7Wfi>M3<7PL25T$R!
zbx-;?;U<6l)~0$I0u7i2Nl}OmkRa{+@|`t%sQ6eEFtQfU-VtR0=G|Mz^W|k~X6RaW
za!=ngLFg+z`sng9+uQT)hSL({XNZpk(h_pT)+0qKq{2rOMyvo%!>37+f6aMv_2r8Y
z?GZpY9Nifos-;814lsx=K^Do#$7q2za#yCV=c??@iFtcO-2K}Y5PUOF`deewSh=(-
z@Gmnf*O7&j;&`7al3_C`At8z{QSjM8$EB09kvW=lI!$Dl$Z7g7b8ThG`m?dNvHOaS
zRfifezKwwVE5<0hc+_#O_Xv7vlQlELozMV+HOA@3)AGK7BCP^rd8f-di%WOwdVWzc
zIy6cm*AOJ!=TQWZ`iHi5$TFjJ)YWB2G!%6l>hlJxYnKe40gPI@GZ+8CgO$->(R;t3
z9UXj7V*Z5H4ghXR^<q`V)v<gZy$kR8;H_A4i=+z6$vmd$?1}vfF4@BiRo(VBg)C3`
zF1aJ=k;BlW&`K`rgUSL=yMmp~2?rAS&x)z+-<<N43pVp_GmTryc_8Xz<GvZ0Wyi{u
zEA?L&7LbM-eRi$mNh66Z(@B<A<M|=XIs&RUa0w3Jht}!};D;eV_{`I%$%TN9^B5?C
zU|15339MjioJdG|Z6ab~1_Sh`;lhvz%6f2aEhG}iR-mJYOja<$z~aB#ipWY8^0E}T
zOT4MhrL>QeF0Esw(T!8pzr4rJrp9cj?lni&7jjmalLN^2ylMfL0^a8v7sY|b?r1gx
z_qwPNkbpA+sLmeDL=mCn0+4ntTF2J{Yjd>3%Ds%nk-_AECP{X;jQ|3FAq-=A#LV(&
zoAUZSL6zB<g|N`Kby&d=P{BsO{w!w=;{;W5PM3(zZCJ>3%uHtrWSiQ(Jc#ehWsX7A
zSM`T1>8I2H67ql)x5_K(AZ`s6XH8V~Ea#X2CC;;pIon+;ZDkHq3@n*b7;2+K__%98
z=aUgXhBKHEe!lr6fj~>fR4O5%zhw*pl8RqI{xFsU04%f4RaANQRFV8<f61<~1k(k}
z%dT?fPnyp?G{$z}_x2ZAx6Hl1tPiA_RB5`mu)fTgB_k*D&URbP0s4o8bM1O_^rQ0W
za%+Culv^4nX{i_a^nk@AZezv4Qa`*;1t=qKrZAf-(<)=;dSS-S@LREV_*^(qhNb~q
z_6raK&Ua20b371b_kd5YJ8|HCa{hTG5=tRQASDktZkT}!b_DoaRXb|Cw^+)G_xY$v
z_5PFX4&g|t1S42Knwzm+8W(2@V1-K0JV@d><L(0M^$9knyMX*DcQg6s(rq1r4V?E2
zikyw;kBc!`JZal+)Fu+d2&AJcS2nf%Bl%iXlwbT-zsJFRG|9!t3q7J~Aj0&YGV&c#
z3P3ndPOfqkI~xAf^c8)9_jfMs$uSvv)e+3_vTulZ#yHt4XlH%cg5)i>+&h*bq9D_!
z@ai$RSDB}Kw(w?hj^OTWRc4()Y}|d=y>k0PH$B-*IrD=Q5+76n9}>4a&XHj4kfK?4
z@m>TjL?>jEi87~vF1{Z%(u8`=o4iZ#wk-9sdR0nFdCTB-F2RbzN<Ff1D7j<`Bo&Ip
zm{XK5m4f1NScyP_q;vQ^YwdveA5DEFzpv-mU$*wr`w1ZO9D|_`y3aAPw!GY6VS7J-
zsd9PwQ)PAZUUni<Vh(m$EG<1MEH!py`;4fn6a~g}t)7yeTz&(Zg|MRjQ|+ezr;(YN
zp7Gy0j|NpY<L6lryt}*m`r{b~1@MwA`yYIpdaxBGB<~me^&CUv6mLb%IbE=&zMrcK
zoRHq*{O5vsh?5Ezugtx>HxtboqD&w5E`ncI^N%CY<oTfnm6>#e1%w%t1Bw{ZX!FC>
z4us;=2~i_;zBzckoxX<W)su$q@0YtLwkCCW+_S>^H%p0?DCMqLi`91nrpVWT4JtzH
z1(ajiZ-pSlg@*D2Oo(ZfRvOLMnw1ndB~c1%EBoWQo#S8PcCe*hbzDC$B79jX(dGLv
zcHPjJ&Y<6ELOi1-vR!8n<XGVAysuT~h^XR$6%`I(rbLL21b&veFdzX{-;ZeN#@a$N
z6Ubz<jVU$UbXcyi70040C!2JR3g+^g4cA=b%`?gs$`weBiWp*LeM`COx8AhZ7BlJP
zOKQtC<NW2`19X{Q5_wB|6!f<#mGrr;!i01qxi+XPcb)uxaxuVJ13e3GqYS3VbgS;w
zCpK+RS|A)KP%=PPVpkA;(L3VNIZf+j@%mFMBi>Xag;f?`gU8RySVlVWP^9+FDAZ`t
z15R;)DP6L0z>5gmzz%Rutj)A%$<1?$cG8ZO@Rc@rfAfK+<43IgWM1Eu_h*_o_%iUR
zPUt6^nV6_mcOf0_x#J(Q-!&YPUd2U}Lz~s(C1{DKYc-k_ikH&5SE-x8_IGy8KG3mD
z0y13cq0!%^Y-?aL^hu=DKg`buklXhgg%UEc$Nq$kip1c;e_0N?fpfLfX(|k?OIrNR
zbt%cAT5`^f)NI`@Sl={<l*1{VTm#aO%7U_6YCn}R9vRX6BgOwhk&2xOTz+V?wp|2O
zV4FyR*t=y#13x7>uqU;NDGcrv*ZU1GM{2FPbAvSa1#EZXvWG!Cr9j(!CRWWRy6fQv
z<%SQAWC_L~_s%K8(F=elpzw**ha6+e^gTOaej>$fK8{P9e;3$6wqQ@yxX$}*7@768
z5qWI~B#te0$px}TS0bOjsc;T#7%E<)!}9~ODnc+#lgCoRTW>L425a3=0>=vEITyK5
z2mxKYNidKb5=9)rg@g4s+p;Azwg1wG*(8JEe!w`qbHshQjr=`Mx0Zr7#uvyIxOagC
zEwE;VZYtrT$Ws__>3R6vBYFtJOMw4agV||b<zhFx>pQs!@MqkfEtm#fBRE^OkYWl0
z1}5$@7^S5*(PHz@rQQ()<3Jf?nqOL?S-&oO;GD+<raUHtH61W^UlSR885Fp+ar+Rc
zsj_sY<NCM(Tsl&pR`%IGwNaM~-b0U_Y3`snVJ`01J4_xZB11MIfZHzjzG&aL=J>Ux
zE^@yl9DGb|i)ETLhPq~rm#Yp-E>B&H^;?ZnI^ZBKeTz^f+N1%6a3?aKVN!o2g};VX
zTlX}K>s-5QREr&U!$zVAyU-e@wB@C^;<L`CEJmP(7(tQiD^ES`GWEDP>0rqu1p5iP
zd3Le&(*7>#BRFzVTT$O^?`K^<b9>ciCfw5_k3&b7tt70C@u7Cc1hw96-!E^z6_3z8
z_ma1Gmi`37QgXj-j5Vvy4bT#^*!i~2!wTBT+_Zo<=N_S5nj+grX@@?A3vF3$#X8qo
z;hP&2y2GmSu3|7g#sUczIg<OO4^VF8S}OlALU6_qT=UX2i-e|`+}X9lQ=15Wr(~o0
zmzhf`smvfYpZ!+8BlagM%x%|CN~hn|@QUuUHcqFVs#U|ZH$yY8mOd0W1&du!{ayZI
zy~n9?ggn&<+7r@;SLjX9duUFs&TLwbZLPpRad;cX80bR!y8@BkX~MEGsU74)pdYKE
zmxMR~ADws{FM!K6&@qJm>yk`fP3I>IbLr^YE$6Er7uEHftsuEQ=}bg*V)?>OM@50}
z9lmt{UI~JAjii4vKc6aj;B+toAn^!Z{5>DjMHp>v;?o5WH@%+#gzx+ckoML2wUb48
z(_OIGcGc;XBBwDnY1f^zIfUxOA8jIOiqSb-4!>Rw&Qa&972V!pDF;6(dAVO{57In;
z!O7-K^HOE-0R3a@VE%9IhMw-<M=3U{YDDM%&x!nv!JW@CI-PMDkELpTa@pqQlw7HT
zMx+=%rjj+Z5U2R!_m)9aJb)hQCgvkaPwP9jcQ%d)OF*nkJzn2xd2<ypAZ&shu_kh9
z$T8)ZW^@4ktJjrz(+lIr?V`q%Ul!&w$YGjb%+%TB`*L>wd3qob7fW&)4Yq9UYjrao
zPnV$j%hxX*>;Ay&DIM>DYS!$(Umv?<wgD)N#^L%*H-X2oxAj{w!%D2vk8N-!I>yGB
z5w?URMj6P<s7yO?6?dtzl83&Xq-l!1S_))_-0z32!nny0x0tAaAFfE5U~yQSSn=|L
z%ZW@{!M!hZctxdXG|qpumi&);b6jT}5HsBIYaHJE-Kd4%zvX(_xK(xSBNEZ_OX*M0
zzIujEfZ&K(j|X9a@A$b8G*=~P!CAH}FW|xtp+)IgzSm1a_WL{wN#W3y`u?;Rz?<H3
z^Q6%_S-3yk4QxA6q)HnHH>OJ^WnJJ*m0W`gR(luyXfexe?Hmh2n6w+uM0LAzaUKBK
zgjKesNErdGxd{x_bJV52<Qux$v@Ka=s61%}rui(tq4VeO?IT%AezQ6y<m;wlIAXwm
zOyzyTX{i`l8o9Yapg5kG&UK9G5=s%%uT{}kM6Pa;X$8AhTh5al8LS>QRW|*u!$JvK
zU|)Htfku5;vUVR1jX!kWZ*Wp#lC#blkjYw-Q6{r$7yQ<nMk!{9;TF0tCqNLD>QUOk
z5*4MJ_UuveqD(k^uGrX6c51x<CdrX#*98kG_F^iJCl`zlo}XfI=U%#|Ie8ZB^e&)3
zD4GH*DfGCbni?W!864HSF@woJH)Kw>TbbUA)>{d0=uTNXby0j4$TZ;3cuKh89!q5N
z_fv=#1a!nfNm%CZ<rM$~nQh~K(LnK;$pbJHMe%V20@a-r1oV{7;Unj7EY}J0t23YH
z+SD2F87_&tVx;Q<M#s2Z$JY%;6SWE;EuPG%ylXd3Zg+8sZ-0YCXLp{Z1xb^3D?8EP
zza<xXl*B)i<$tJc;)c$0hsydw;Tkr}_Bpy~vdA-U!L(~=WscF<y_y=U>ua|1MHsDX
zSkVM&sh%V$2OcNLFbuXmvcjlBZk*K-_pgq1yUBW8Zs@!&P2q-(XVYq&WssCpv%D|d
z1-!Rz>Elw=4T=&-qsYKTykewGyjuQT4;1z*#e-4#g-Sm+l!GeKJNPFzkffRVPikTs
zFss$2o^|w9wbPn24t-x85oXLUfj&muV?FFK*XM;WkmnNI!tNSk0^&z4dRxFgiC-`M
zz|RG4uG+$X21bFGQr0X)kSed_=V~1lNMf~w*LFyY_-)Oj$;^5N)IDHmJvOGaT>F{;
zgyfBE1e7yHDp9L+yR4KNv<i_ZrLmOM#<;Nw5lm50OU%m-VvIF%Y5JzWlr6$sc5N2B
zx7{4o|FH`FzgOS?=c=2X>HoYv;Q#-v^OY+98`SY!y^|loPko*DV>ZHCBLr@;PR(9|
z^GBMG0yt+-hg*aFyyjfcvy_)lUpNmZk3AmxK|zv>GzG-6hU2M(qgS(!i_16=!Khn>
zFbPo!rp_5Ql3pX3l@^UFVI_5ZIypVwFXC;}GA343ZGF{!I6TgWFH1yqCKgFyLGZ9q
zjd#xa1$qW34v};)^aV?8C{)?kmG!(F?K7Z`5f6GPR*k!d>wIS3)2h#Hv|Eo_*&l~7
z(=ey+_<*$+ne_|dP_Omox8e<!|4JE0LKkSTntu##yO3=DVGCjGjOY3vWNrf}d6cD^
znd@PjTd#FhjCqH3S(;y0tKVgrHSun2?W!UgujieQ>j&dqb*<ol;n5vAj`WB@O(L}R
zOm)7=DP2zGeqMXY2h+9WTB$0buc)V)2v{)SwEuqHn0riT?9{GA^QMi;jIm9~Am4^%
zyID8iJuaNpwGl5WUGQ^^vPA37)G!2q&BtY!`Q>UFDl<Fb1W>mVrsk6u3phf#&%}tv
zx#l9=8Qfoq?bA<}^n;2DnSoT=c_kN^D&^37-JZEF4ksuzS~vicQIdlchW#?y9M*JY
zRk{TXcf-tMJR#(v9zL0!c$|%v_9IE2uCiI9Ktn%+yeVvgkYSbzXmV{yHYq*=lZvVx
z<B<5su$BrB2fj0lr|HU@3{7+QXxg)b)x|FbLK4GTBcnHAo4MBBQ`63)FkzZ^+c=di
zCSk0<<8@<<HLe-?p{R(b%Udyxtwc*NCUG7I%8nb(U<51dKUH1vmnCtTnQP)2q7TKQ
z?MlboPZ#XBg4Wj9O<{raIbfJyD{j2G+~V)kd6cvpH{XwazT|~E6QL-p;|Ajywt&F?
z-ex4_I`$fYUDA{m*7DPZoSa3i#+F+@nrRN^>9DB@!?6BES$76#ZPw<@8f?W8m)1bp
z-f*e8{o`%~Wz87VjBFX8)rXQ|8-cm_g{=rkosb%=EfNUn!hGQCG4#nyz7A2$&Map*
z#*WHTvO79(u23OVLYq8nVRitPF3<<Aiv1dxrqExU(90Sgg1Gli2d@M<>lx9LRNEdo
z!cd<q>yTL}l@X<?203hj>#s%f!iwOJ6I%r|_lPWpSBMk={|zSJr9aP0%(<)iBLH@V
z+~4v9<Rlr{-LR*W10JoyObCN@$nRnGa(#SoLE6JM=W;<G<U|B{Ys3uKWCb2^J!JdQ
z^7cMcaSs|yn<$86WhzRKu!xq5f`jn-d;{vTZJ_^C4PapU?@5G#{@*XLg+HOr2P|-1
z)75f|+6TYWiNXqSN6lCg#TQlJ+A+6{9dyfAH_PNDx!qT;qz}7k`F!dYtfN`QrQO7Q
zi2?k2z;ohF@knN6uj9gR@t&`@ys3?d9bzT&^WU`!>cr3s>5OnUZStpUDF-L3pT9Gr
zg|e>=^X6aXP1n=g7l?1W-+p4TiM%|2O-JUR$|`m`xjuc0R^}-rsHf$EP{g6Ak;FzM
zAYz6rlOn4F0!8EBe96K0<VfwG>B{Dx-wF<Cuqt_V+I6ttV$aHcMpElHJ8(<gtyqO}
z)!^1REz^e&Wt}_A4p{#R9eD<rt6gt1MqqXMLp9R}>$w#N(lILGLYJ^4a2WPr&3TZL
z+R{z2S9%-Pr`p^vU*E@PYT2HBv9;M}k~M%;4Pl9KwqME)jz(U47A3lDeW~X1+abT*
zmBtet1yL=d1M<THW_dxmR7^_zY{}Llv1DZr;K@;z2yvqcmtP`l$TW;AoR?`oH~k6x
z5AG-(TW1*w4fbU!slS8hp3S>7_fr$^_DDg*?iJJ&P_={FQ?!e;nR=L}NdePJi8XZ^
z-|_9rt?S8$V($W$26_&k$B}Sw(~V*sj#QQ^Iv8ghyDtVVgckoO@mK3t-vv7D_(#=A
z`T-D(^jp+1vYG`pw@u>F)`Y8@R*WFN)%GfwiEu>JwTlze+L>HC_?8G-okn|Ux;q#0
zZS7Dq=4(IPbQ~`H<4MS+ZQUaCeAyfX1+>B<AQ_&;+5mhN-uW2I7LF5|ixjV;*XMQ>
z&L>nC(I0Yp0z*CUNK6k-1OaRyt6!eBdsRsyTZw`Hi?MSE&tze{b!^*7$LyFL8*iL+
zY}>YN+qP|69ox43=M2tl559R_HLH5G*1dix)y);;wk2~N3v^Unv#sJHai;St%^D!-
zz<}AC4YtQ@J7_B*=J<e$g_X^|4t%<vFbar(JxJGBF<PpLQlQasG{L{+D0#Z+NaP5)
z)<xC)xL~U3DSGQT9MGmklu;6P^um_ZOj-TH&_M(Pwlp<B*71XUcwBc5N@>9`#MK0%
z44An6xbM_^<cB!+Nq+?)PTEUcd|OY0<BWkswGfx_IOW>g50j7#?k6xl&R!dM(OV&f
zqg`T%zx2wAx0?@jHS9Ro=+V@BZ6D}!LhDtzdAS9VcP3Wzp03K_fc?f91SKPp-iWz_
z533)&;Yg(dfeStI__<z%2!N4doq!BioHNtwc;`e1+=6S9{B{+^q+%eKQCh10tZsQM
z7444;Vp)kl<JWOP96uao<jlxEr~a{_?`+A6#`Q6ISV#_|7f3zvD`Q89)F24Cns;GH
zF7h1ENWipC*LGRRInPzwU*i$pk#Ygk)<rObyKU!Y?2I&bmr89d8RsO&fZjgP_0SN=
zV*rcV$pO0h<V`>B*-7G$&`i}-kV_1rtY=Xm76LIRrU%@JuEwVe9;~)o`ZZ#UN`W2#
z4Jv%dE$h%Avi|&4+50YxtU>fgr617oAd0v2f{|DcK@Nd+g4pm*aaC5s<Vc4o1LUXt
zT8O50!CCHz61|E=%Gm*wIJUtepme-Tm(3oA&Y*vNvpYF!sLXtASnqvGH%Sx`9NOa#
z_sAt-%dmieggJ&>aeq5Z7p&iq8j5|-tuH{sxk+^#<C;StO?ngp7w5T$EQNgQPn$;M
z7Qu1;6JQ9Ohr&JWO4nH19hR(=2g(%nUno>Sqax`GfnT>rK7>;Vg!?5H1^K6X-@R52
z;hREYo_bJ7HBCy$kZsCuAn1<4mi1JmYJMy-Vie5(4qh0koKh#(zdag;@VpmGbGWH+
z0NoI1nQlpd{MX@x0zK*?*%>%!E~|EEo^18!xXAk6#OGn39@~Br3{j_kI%X)5MH3hX
zn2i<%qu?>-F=b4yohpZt)uJB6$=uRqQoFQq5oN0-iA^7EN5)z>kO&#4A#taY#*DHX
za0?5xp&c8R_1)jTqA*GKa*7@|ub)A<UbF0`rM)WswCXRSj9<Y)MM7>IDLBYV0RpaA
zqu1xo3mY6%jad*Iqwj@UvuliQjLJvltf#6ZIIB_J5kp6e4>79GOR{U4ke)(mPcG;`
z?t=>{gqf>K6S-MNT#sCOQTQ%buaC9%&CN_$3@GX*<T42%NYxP3ySCIV=Ru9&LK@DM
zZW&hger9d21@boW-(b^AFPG0M{$!=;)T5P>RWUZ{)o+40ZnTJcvY&43bMS_>VVQ?v
zn2f@5w0WBJLcFg)h?f=v1@y1BSJ{>?qn@h~gJw6UD)O5nRu$ic!3g9L;!6KesC01O
zLLcfN4Q*`c=Xk-=ug8?1Cd0IL;Fo1l<ANgV&)eB~Y7*s;4D+O3qk%LM-C&&N4?|i1
z(MZWDe6}g_ZS}mrS$F&TlhPM#gc8%|pPfvwCt<H>^Y>6RM!I8OZhgMblw*0{`OgEN
zn1wy&<_pn`(|<UutWBG43r7g17*=fwP`SkI)`Ot}LSDw3JvIKZtMY7q0ssCQw8)$W
ztL6B-{62d9GN68(UqGR>&(oI;0^!2d%Y1vv=W4Ch_|2p^34g!rEqu=%$K+Fzclsp$
zpuV^L%qw%v!s}j_{PM12Hy=UBvwW2z5*vFboPX~1#TwSkFS?-=+eY|VOwt>>2iOwM
z=|%0z;TO+KIN~kL9Ufjn<9g$!!JGnn#v_B#on)t7^@Yn4ftc5u4*f2Wb<JgaW^<og
zNBR_JZM~I18L2xrrC>)O{}45Wdmf0r@~DP<O~DM<q*Vgh{(+ikd=Ef8X6e;r<0>U?
z$u#d81*@U1%|}FIR#Pm<6<!ERG&rU3tQ&bn#IkGC$!4_4!X;IE{mubq`1|#uj+#=^
z@2l=xO$=|9^Lu_#%{KNMnA>!_^FLdX{}@Y{7})=3T*6OUt<AwO@{bgSC>E2t;^&k*
z5y5SWf88IfB0x&UPedPyu}Z@DhjfHgEF|jVS;@u8K|DMjuL>B~4Vx{)%hSnAnU$p#
z!SrT#XLG3U)<d*`h(v92e@7oblq1v-P#WnULEH)_B1+1sDZ&W<?KX-9D?8oI<M#Z?
z9zsJfck^(t-%>(|z^CQu@}V7m%g98`626P!5e^7abL9Vd<2Wsn9a^QU;5_ZH5hgp%
zbdp-@)}osZ8*BEYDn_BLBkkmzykvghd&07GgDsDg#AKRILIqAtm;Fcu_p(+AtS8Op
z^47NZ);s+AQ*4o*!)7@Qc(?nF$K6yl<@x#QQq`HwhEl_KBhe6bm_+)x?hpqu0Iok4
z;rH{N3v9lTzai6Y{wF`xw&`GTdyU)}sgx@k_`Qh-KKJA4#Q=?qmxJBSf}cDKyebh`
zUL?3eaUi)}tO}$tK_%Sz0+GJ3J$56PD&fax6H(T;90>K#k>CaudV?j6(_wll4UY;x
zQ==GLf~4lUE<0p^kTa7=BH2Cad(Y7_GXp9hnhA^{jg$G<mb2T7On{Hri9E%1BWE_D
zH<0lsYkB9gVf1e#irm9N?{_B-VHVf^Eu%hhpeE}OC^z{}IFv7QNih2Aj)*R4b{v@h
zem)4^b;i<5%Z(oPZLM5af1AFFm0e)mEdDMo1SuSmZ-d;iwDM_tEiz)lTlcb)$;(i8
z@}@$dJ2yA@`+_4(cqU!okt$5K1o8grZ%732J-tO$cRTS3_szznvk8qZH=y#v+7E11
zQEA#=OnL}$^5ev`aj_iZo!=;=uqDbk-Y#D)ig>IY-W}&LH$zmh!BBSj{9MXJUV?s^
z0fGle90%K36c5nc2*RmRJJR?BXyaI8`Joyi$7aIBFu!NOkp&|i{*m@*3YyC0)GKL}
z)6B#-mBU%lAHWH=En5lex2lrHW(Wtr4Q(VAi~br@`o-YZ(F3}*f$t30UMHJ@s32}N
zbrc(vafG1L!~69%l`+zBSKPhWR0L!YJ84YSYw1b5SrDmU{mRNjjFsrZZe)!TyCQUU
zqYU2-L2ujD92Zw1lL1SLcs;&#DZD<Q3klM!wH4WN#9prKR-@th=_mT2#$J?JSeKdq
zVaMqT(Pt_N$145QD>9kaKWSxnSHD}ZpwYVox<M)^mR>f2gMO09d==H+AV#XU+EL=M
zO>9@`6w!~#VO0es+k2Dw58!`1d382^F8TR#(|H2-u|7L&Si*_~E6|9m<N-%suW5>u
zSBG*VWx#-Vuf9GsqXsdWZK4l+0E-orDCbuv64_S%U~JTR27ZMSFSt@tpz;0a2^ii%
zm0;U+3GCA^qm|ydvs$f?er<@641w6Y1m}k<_ZJ>LJw5ILK|jW;FH5y1b$y54(AbyT
zK6YOx(ZP}NrH{r)XJw#5piwqOGmJtJAU$iY-{6wR4}n=JL2KGHhjN_24GzO6Q5jYN
zFpVt^E3#oU$i1wEkI1yAhQyU|!5oC@6G^7Xn6*;z(VxYbL^34o#)ZmaB!c_JUPL&h
zIR5jkr@LlU5GYfM7EHoMELsw(%B67H-yjS>tK*ffHy<`??Th#@=FY_pe~=(oG;T?l
zPB0dJzSxLb$bI0x^u6n1bWP@W+P5D}Yzh+QLgp471DEYx^jYHY9}+$7x&?1>S3N{Z
z%M;O!c1;YOQc+~u8GTK}ruI?x>Y4+EA10aYWnyH}!FWL69z7o~mkq|<#Px5XZAXc{
zVt5saqbX4eNss9iA?t0NgjRfEVJ>%J=9w;yQmK3>4(DH<!YCPrhUHZ)=x5=O@ZPu0
zWWU$lfkp#~tMHpto3@>*vNlNqs9_pae?B+e59W<T4kIHRiVz3qACOe<BAx8S6N7KF
zL|}!H;kxKYh{>ZY{1TK&IRX1iIsg+HR+Io1pzotvf?*JKC9uufA-y}0_j=(m*f$eJ
zH%geVcYxLi4{fx>Fx>waxW{2&v>-5)V;yzH0Uq0gRDX4gj98r$zQ^<P!x{sX{8L`_
zGior8#g?zk;+mY<B&IIy{%Nbv_|g7lQLM5SF!4N5Awd7s`rWZ~MnAu%J-#h&b!S$U
zHQk6&soB9g=i6QxZ7cr?o#TU@Zn1G_%V&L9BHf+J_Ns%1BKx5F?~4MH;>RT=GBLII
zJSC^OA)`)vg4O&MhQqX}$R1IYw5sr|<J66PhJ_G$IDNFxNTeAwz*BX91GD5ktapm@
z_A77Q2$5rro&RRf9ZI!gJRkS^sQR4oG1dajMC%7NqJvCQW+N^7jl$*LHQ%=8#C!el
zGU?LwuF2w3ZI4Zjl~bs+cyZ?aBZ^V-d+u7UEI-d%xnEQ&swnotYB}e-l$`uG#I3_*
zAgj}85dN6q?6i80o!Or*aw$bO8~&}&L}2MBD+E5W(zvRRw3ZozYl_{mu|Hn#`qILh
z{kaZ<I_sw2vT2pEmJUr`w-R9m4;_>57$kqq+fshTM!*PArMp`0c^?UJz*CmaV;55d
z*OQvc1)efxvMLiWW-w!?(v7gU`mJLCT-3GuAgbbSI8-zZ$~8!fF0`~whYk@7RDhhc
z`y+w@w#YO3!cxN{_NVaN)zLxdyZ=LR3qX?1S~{C4Ch4;(;e79TKCJL#XN;3BRd1qW
z`?ud{=k|8GOz2_8EK;#(EHhx-z1HRBRA2aQ;`Y{dcVw5ZO|4wMc}Oc%G4&+6do%OW
zN|@Z)H<EO<xW3Im@=hVV>I0k;6QBQIjmA$j-+#C*SsDJ<1ODaD?&C*%)qSpVBi@7o
zhXh1;`4HO}olz{UsKM#D0f5Y>j4<HU<jptdy0p3W??{wTvqL#{8x;Y<jMyB~zrCz(
zbvU&>cD%FY<@7ErM6DZr_B83XFT6HAWUg*x*5SgehFN?V`Y72K^A`hU%j3qdCk%Ad
z-2Q&qS^YS$<ho~ZF&ZppzI1r@<b_Xs-=T`n$@+A6f7-$P;K=#@x`&pL&Efmn9Xw05
ztq*UYylHfP8abPHTby6@a><&c@R0^9J&d+3G}MF=3rYtkw%zm8Kb>FhCFEi<l;lv}
zt0eE)TprHb4hqdg;y#l)m&a#-2ctu~9{}2H@1oQOnD8E7`EZ~aC1qS7R=aKWugPa=
zA3^XAwJiBf18gx)phRON^5jN)*rmSwl3cjQv|oV4R{kuwBoZ<l3XmS=Uh1J_G|TMF
zqr2#&Q?o|UmfbS5fy-btYD<fpkoR^EUMDWiWMnU(O?Xf6TQ|rwb5txFWzdZ%q)239
zBjLAoZB#}|CtLl&4D;JnX9kX&#qOr2iB#&~v}#sVwTp@USGj;=;?@RW%DhstP^fwb
z9qJW^4SuQ`r=?+M^+@S(wTYwm0zTeC{9w@SzV68xJqC#9dLVz8hGmE*GP~mw2%k=!
z*SdhT5|j0zxO+h1ykASWPr~LC<bM5n?$&<oK1!o#58-s@WFx&&kns?2^Qsx@f=A-5
zy)Y~qB2cn#rTX|W65z>CTjR20P>SC9-b^XNQiihIT(OxH69pQId&g6DN&=cGdJ-~P
zpj16CmO4e6ws&>w&zt&ARX9GnKnH8=R7s;aDbHk1K0a(6hSLJU{@imPNhvVvpgKDe
z1uYP;L&tJ4;N=41i`>*xjUk!T6t&5L!jq;3#c~+$T>T6s%e+o*8B2K?AH?Z>!$;pi
zcnERbpX3f7ZQkL3?L(Eg^!t8*F%V%rzq#b45%zV|CP8h;sDw>k3N<?g(rJUV(D3M5
zM$mRi*3)^jl10PrF@e(^gq5v_&kDU-*QbXA5eIrG?hczN3=XyeI?!M2s=9lF$ksro
z*GVM=(w7Q}nVp0RrNd0&oO0XCO1Rg5_YS#~h~J;njwR$XN~>cyg-DKTHNs1e8(W#!
z%ZX75avb8k0mA;!*JTgTdlc6GiRVGAW3dR75P=vt*u7tHAXY9eZJ0)JuARfi>sJI+
z03q5Dq%4EAABNAiPS#8`6P;9LnB%ltZ8W+XfwgC#$%<WotF}p;7m68&j`p`hNV^$X
z`Z5YK$i_3qbijta-LDS%g4+cS_pKZwr6O2d|Ato*CtN|)A$k7=W<E~fgj*9=Fhf#Y
zKH03}oi-?~_NWu8ti=Mx-W&mi)Te0bVVzDLCDG})FrB8YGYUF%AxJP9!8V0c*s{!V
z^&yV89&j-3s8*|S1}dnp0bGGi-3BndZl|m6&9A1xYSRkjy)UCb5oAD03<5<oK!ef(
z<d5-p;V1OY^C)1wM;ne<#ct_f0wAdfmu{OY16j@1dngN&trSZpz6^pgF0;F2#K2&f
zd}Aooeb;C@+Y+fGrnsW0bZ^J(TCQC!T_Cc$(eQlBDAKEov20Ed{*)@v%6cjmCWiiQ
zzdmhmrltz^LM?y8J3~7$%^TRPP#|_KioM~pOCok0eSoA_Ku{rjxp=;R9YU4tQo7!*
zfZEkjuW6yhMO*~fTsK#^*k`<jYJ1#?DjRHRYj&NyOL3<^36wO>#3oRYzz|5o$B6om
zCGHy}j$m~xv@z4EAq9oNOe{B5xn;1P<`f(~de)fpJP`1G=y)LC517h%n=-yIPc9LP
z@i6v{!3%G+>!8p885HP}a@im%34%%m!lXhCA6d@R;*#Up(d=H>7hrV!*k=A%UD4H8
z)Mqor+n&~lXB3`j8ZEFN66vGuu?W7s5V|{C|L2g&{9jByR;K?M8}<MG{Tpm>-Pbj<
z^4YFtyuv(dikV!ejm+6gTNxRk1&8j7rbSVOQJ0Qu+uE_d&P_Fj>8>cU#h77%{6K-B
zh(K<(xHmhCC!S1P9#pOvhO`kzKO|nia8UfANHMNdKtP6^5v?(W0sj3>O9CKxrsnr=
z3O*f89nGA`qNCmWi5y#7@5hg!8q9?AgyjUCX>Cz+O6K{<b1WtIIQeq?h0p|aUH?&k
zv=xKHl@XqARdU)XeN87feg;n&_cE=zTDo=Q%rJhaWHY)h-5I{*a4NEpb===D*2De|
zD^+Hei&It>8XLJ~#T>2+sUKCB>MT%7mT-H0s-(22AVoQ#COw{i**Q!j&YBzdo0+q;
zd|^=3mtNPOdI>lu6k;bpcfX|3q#zE!p?bDb3g)#DA`=+c(UIx8Gk&%)&FSh`g?U-T
zO{KA2P1jZilB8Q{kT*xF&;&9^jh<l>#?1l(jop_7l6#cm_m;}Zh}O)ykHgMWX@D_z
z`+TyTE%Z%1v8?5zkH$-)22J&a-fTN{m6zXKcQ#&PiS?xyR8?y4-L&T&Uyil3aq5sA
z*GFs|Pp4;#Ce`h!ky}wqzvUta#zI?z?@RnW8SNd>Ep}u<lj>DT1q=dKPo8I3<EI%+
z1c9}TZ)RaIt>mjIpNc{u{Eolk!c+HPVru@>6T&J{r9yy!?+|L<hjV|JJ$AH<OzO(y
zhnq{$o8WN=#VF-=(v2^6Tgharuai3qa)0ciOrlflV-bJS`^b$Te-I=I0!bBvx$W#0
zyGH@OcPh#=f~YqCVpb@wU6e%qG4xybCNamzBx9tCmb=~7u~%B|u|O1G63<tI>n)lG
zi`CW_)=#xSuU6-a)#DKoR9l|FkO1?io;){KcNEb|s}~7}-F095wT(j(w+b3Zk`P{2
zW{n(phTb0uqeIM3{qm0Vrfe<i0xV31EM*ve<PU7M8yhKCkgF;dTW*8f;EHowZ8%Jd
z>2aIQ$Wgn^v5_RBHed2~&9@OV{2H=vr_H0Dd4GxYjk9vs3PrrbJt4pWVhtSnV#?jp
zCa@8@chi1<i%_tK!FCuCi~;R(pWkqaGESUHlHCimzzdgLZ04mjz)^!aff5eZqZ$HH
zHqhT2Ml`nsRQR9_xeqF2l+2VRzq`ANN@;x7Ot<yY4{kht?o+*nV8Rnq6S?!bx5|91
zprTyZRW{a6;I0D{3QtAgDGuUQd;NFFLot7j;o98&xxQA|?r;j@#4H)e(f4m@CR9ld
zUav%2JyiWDLa&)GEP<CrU;?zscM9{IN?=>T;iL<&uYHGqe63pEGB_HmIKqAWtkW-;
zx59RREER4TjnbI$GTd_k`k-w*B93-Qc`?_053$myd?kIWqf#eusAKF)FYA<QpaDb#
zQ@mGe>oA4YW{rJl4uB1iyx}Z#1=!SYV*?FhM-x45E-afepZ0T?Mzbl;fOZEqiQ3TI
z$Hn#ju7l5FXVVz*T%nkROQ+g*>YM~=r$uZ^m^mm9o4aYwKCioKr+)&R?WgJ$3*BOc
zE*?;tBIRkx=1|rD_7oy6oV6A0)bx*Vp2{>vyN)x&YlSQ6slr!(nNX5D&CESFWw5Q#
zUd#<QmqugXvnU9W6Ky<N8;fu9M1Cx#2@GN{%-!_hO~I|7oCBSCL)!EqxbzJrGRLDT
zSjjU;2n&%6r<akgR}0Tf%TXI#?57wKn&I$~Xu0V%_V5~iC&UEc+B{L<Nl2LOjE0EW
z9po84qw%z<*bmZ9EkT~c;_>?vYy#^fbGuxN78%wN!+She@Fcf&F78`*HAbDjWhi>|
zSkfw8&b|^(Q~eoRWo;qG-morZ;R?Ko2Bf4~pV*jh<}PpTIm#;y<-h2DDz%9YnYgWi
z6=n&yH!33i3DY@DhH`5=<66fvmCH{R$z+Ok2vF$NlFj0g1Z~ZTHZ2xJDeO(6@}dGd
z2y&CE1ZSL%TE40g>kP#?HLXkAn|wvhV7^v#naUCG46%zKT4;ZoP-f&hP56dDJAtqL
zrl%t)*!Zszg5|%mPB>Zq=Q1*-dY-V}jQs8G1O6UWF3UgwM)=ss9Bt8Y$x`j{cm0&~
z6-aS5%phFprS*?={MV(HZVHOT`9b8m#D@q_GSHLX=Jr)MWD78&wvJa*yZalpg1~US
z>TsVqYAMI8{G@APz#`WRdE~t7?dBQ1tPXT`RCAbked361*!JU|?96tx_7-ij!o8f4
zW!*nU#O*2!2{c}xfP)rx9}lnl4-3ssjnDTQdb_tbLZ7dFsgeEg1SRG;uzi#hZX|84
zI3{GHK66OV%YkmSP?kn3jeqFS)b&I`19f*Hr-nEUfZ~M(<GM&r#GIWQf{^XK*nOiI
zXkw-D#MDid#$0)FAqqv5wJ0mO)+UmQ+KO9;mL`t6?dPE^Dyv1UcujKMGujT)w-(L&
zv=@Gq6lFQd0p=N{(hRBmQc69pqy>`bS8%(XcHOeT8^{zv|5Y@usFYh|&Y$|DUrvS!
z^Y1vl=;kY%f+Ka$to1HgiM#T-nGqlAT4nl6&D5czS=m~HG~HY}p}Fe5-*HH54t-!*
zS}XaHXVM7bTCruID`O$*CRWR(_py4WrFC^{^=KyvFXL}RU<A=P+JZv<sH=$Adg^9%
zskHzwS^vCJOIKS{D|LD2#qvl@%~`PpzzLp)ZU9p@d-mc8r^h2|<KPllxH`L!Sbn#d
zUW&BF$|StLXsQ%gk_(j?9dPGJe56(PWDR?(ro@rI-VbI2a{WZ2iZ%8}@v+#DaFrM#
zU$q`UI=Z7(M<cnA7y?ct2~|Ey34QP0!Y)0x6s>|rxZN;hWF5NI(won{v)45=@GL#&
zl<d5Lxrt$-d#G<BxwfAA5pdbtyd2<5%#+slYFx+}(I!YGyPUh$YplrLA=MXWL<_-}
zrRnfe{DuI!i^yS%=ly{^60-li{o%c0%VrwWFPWF-3zc5GN|*x{XCj@y%#1z?Gw>{H
zR><P3Nm|^gX&bZH|B^g^J@^vM(A^eu8P<JugPlbP><6e7Cv<=tz)!=vRGm*I1Bwr%
z_ukLI!AHpq)l7M?4wL!!mQ8LXzbdO+WL(D#rYm=`!Xfmr@Tb<Qo%mTBB5w5xUNm4(
z6NqHT!)7Z*m{pjL3-7-DjA2Let^Sb`4Ambi)eXr$#10D>8U?{gw&j(<&R+E#Nx?iO
z)<jSnn5m4EbDt-%S~L6#ZnfkRfehSDfUWQaYrvnmWu&@qPni9^D`<tFJJs5G`tKmp
z<3TCKk0`a>*C^~CS8B|Z)y+yt#&V9^+2PsQdqqJLnsBz12USEN!rf+4L51qT7mF_n
z5rmS3Wo_Hr`!S3J8Da<?q+?oN%wJbi!L~eoX+g&<V)S10?6|VX@+SDBYk4x(C`)KF
zEGRVj-#9e&HzlhxZV_1KzPsi5={JXw0Yx_y=&sv=_V;y`SG-P%(ts<_{_CyGodZbW
zEU0X;BE?xwfdwDiQlJCCsnX`-a^Xcu&>srF{cku<*y(uJ@t2bCA!R~3Pd6sP+)1}W
z4=n$^#ibB~#GqnJ=m%>Hsm3V!5M7Y8#S5TE+_#(V%6r61)}UWXSXEMBsc7}KR$LCo
z+n>!hu3!(&8qZ`ymTJp{#Xz70R4BE3a-}G`x_*n^*uZoKu6$Wjpg8lI#05G$xov^%
zndS#$_Z%(+JV!<Rm<n~mr@zv@UsmN2WZ<>4$lT91qk{0*qaf*JG&Zg#nSWHr=0~ai
zako#<;Ebd=6%!@+eG4U}qy`uH&ON#xWo91|CmsbH_<J+@!Yd&gR~;Y~%aLclm&M_3
zS83;K>CK+`18@589HCe<V~DG&YnPnx!kaUOoHCX@02SE*QkxLeCrj*Nt5lD&=60t*
z!eaj};i)uE4aHaG9$o<#;;{ukS+Xcx_H)El<$p-w=#m^1T87Vhtr!=9C6rD{<xYuq
zHEOFWoVa^}g>3;xw#4|EvAtHubZIU;XHq*XzB&R;Ls)sWAE|%6hw9MqQhXh90*-^5
zSr^pfD+B>lWOJKJ!zgSqP<?%FI{y~iZKr{9J-ZvG>gA7tV@Oa+wK;*0dYZv8H1<^e
z!fG#{Jx(ArH1VY<E*|lq+XoKo0I1Vz0dS6ed1sRYtiK}z^r=S&cQ_mCtb?m-8d{cA
zP_+-_2_#9eAc&ZqltWNg7h9cUtH=r5#1fpz6+FD6uV11U?x8L3HdtMnEF`jQU{BQz
zh-I(U-z=6eGt#l%8Xb%K21pQ{{&CGkjx-1rifOI3O~AlM`AB;8uVB9~c_Ty1wIXY}
zB?E~JlIq^pWC8_ArUqm6_1ime&Pa!gZfg^GOyuaCXK`J-b+P};SZI9s1~5q}=t0vt
zmNyu1UVa+s5UiKF-;21)U@!vrzKC$iBnv91cvo+KFwO8x4Fl1Dz3Z$BUE=PlzIfx#
zd&TWDZ7+#f^H>hA4ZvUF*piVaI%>dwK5nFlpJ)uey85@=LfIJjZDH&a!rijCd1~?x
zBCPE>`1Ex2kdZz6*0drlgL^xK^;V61-@Z2GjV*1cWj>;BKtX!q16#Cr(s%4mXFIpE
z><^c3uMU3$GY9C>r2<ZBR1%JP(AYzQkcczMFZIAwzUeXbuG{{z)c==Nla1+rK1(jD
zZvJcw;e2Lv@>-~@015e}G3OK3oAXUw6V_GF|E}!A0qbj|bBkj5HZ9*?x%}htdyNw5
zR+~bCD8vFlU^k$EtR|r>Op0wfnWN*#VHv<)FKp(}_^uzxhp(ItM?}&6^dV!|=qUi~
zAJ6yU-s|3>ET)aDk5*@T&bz1UBxYbZP^)vv2j#yEnbNk====Lr&d4}bC|SacRR?Bz
zQad!nYrJfittqS`B0-H}>EiLXP^d^bL^j0`CrLR`L1eB75G;b|sZL7ggtJ7E>Oqlg
zL#fQN=01rA#_h90h6b&Ab`$uR%Eplir;N&?U&~hQ*fyR?o&dNa7EB(WH#ESR!)|tk
zqy-4^v;Vt1z*DXgRa2W{h%Lp$6b>aM8)@3Ha%tKsXRFh7r;HGTg79Zf<;aWV?7mJQ
zLda_5m~$k6S|Q}*CgydZKCP7T_AZTFl7`ynf7XAY;2^+}-%`ynszdFYfZTYgB(z5X
z`}HPEmf(5D3+y^0S@x_XXD~UV^9cQS+2na=KC3lqfk0TlWb<L>&wXZ&$0VVuulx_)
zoiWBX7ng*^N;%rPQ?Df!jJTdr!>1I-D%%3>2pXYH-9n$D=*wg2^?Y(Uh9Ju3<6Nq<
z5z`K5p$iqpme*F0o`@G6iy-2++~??!?-g1wtMHAtMYkbq^flF?Raf6kn<lG?xD9PY
z)ig{Vja2Tx`Al^IPlT4$jXG=C6;n3+`7mu-A1Q?w;{&;n{Wa@cyGh8pX-`3cGMO|g
zlRoKT-?4F0hOWB}l=*lJofqej5}T7DU(mEeU!gAWJUFAp(Q>|kMkVy1)i9iMFA}@T
zn*g};b2!-Bn0}4wk6v`Zm9i6PSv0fW7R;zQO%#Ed#-RtPwsMzm7Z^_gP16SA;+<yW
zpl-z2KZc2hd2@AHw#UC03f3tL?pCcb9ey!us-W+`#1oFqQ<5yY;3^3NW1pWxG5M8f
zgcQv+NT9dDP0OfL(z;p3EUch$RBPa?v>7JS)w7w{*&}Do^k`Fbz1lq4z@Z6(1Jj7G
zc)6Wij4dsg&75H<vJf4a7|eA7+RdsjroT8(b~#HkEL)h$%r$d2%aWIooP+bHWM-v_
zXoO=cv5bRKXf_xe_14r@>Psg@Ktme1Ej(nI{xPZ?zD=68#EBlTFjm%7l9gAQ75*ZB
zQ~}P*eyShM5i|wTaeL4)pnMjH933#Bi7y_n9`O|!dEM$hF)H|-a#sI0sB?o<t*Hla
z*93;yPQOCXfhthH$`3{^Vqa8!1>8w{Cl05Sm|pOY&}dC}k+sXP>>s(5Pm0GiIqCU)
zvaOXZP~lZGGWf4mLczPpAc8?k!gWhtvF<=5{7(i?lspqL-_IOBH5<4F?NO+RS0v*g
ztRdT-C20jcPnf~9Ub;hsWiEWjUpQ@aq;)WS?8$M%CRt&wF|d8oAE?!Sh7F?)PE;wa
z*l9BL#2cdd3~WUX^U8Z%eU2+wZ@)Pq46+ezutoMofp7K8=4H|k_{-VLr{E8WSpW>8
zDetz{{@dFX=b!581XH~1VXCcs=@*~Uh_CF8cSIS%U!}kJhi*s<k0GW&B$`Re@@1ok
zJ4&5Z9TEPFj4RnRm=lk}D>RLDHC@O%ATj-Z<w}naWpFQ`(3?dN{+MfzpAqv(kspEP
z=ffYYx&f&C@wqeHK;8DMB7w?LS}p$VV^FTeBWtTl9+j3Ryf)=zmp|f8vtXHICVQH^
zN{C|8QMtNPsnGHEMLb_{S5#hZ*tNJA<6QAOj0Uek6r(Aw^CzwHcHz+!J25g@WDBE+
z?3gq3OLzB0yft9_vx_jHllA8++Y+xaTSy}*O2Je*S*~QAK~l~gk4&RRs<w)Y1g{TJ
z=?f~K1z+#wMZH{OvM??xhcV@u6Dc=TNf1N2-acmF!}OKfPgWUhcKgIwYbkUET9BAL
zi*7Y>_Y>_fji8Cr9WXr;G}=2cXyFtld3*1ob8Z~@(pkBDymStbQ#Wo0s+g&KRS)1E
z37D{4A3u|#<B$Y?p*}dj-zVW)=ROc{q?+-lwrR+ps*}Xx=|`d?(2YpjA&o2{;L2fW
zpQuqwD1RnD+uXvz<36jWUVgUX?45~b>}q|U@n~Ui|HP^qj3d{b72D|Y+lg7`2Z&Kl
zeL=K7cJcgInPUA9S1J<|Bj^8m)GAW7a@cQ1{vitVc^7zLAlQ(GnlYn3Z`o%(q-Q%V
zoNC_r#p|Y_rYi}bSss3Qy3`SgNH%&&^n?<DUGT1?m|z~vaqZ1@H1y2OXm>AC*H0ER
zMIEvKD0+oUBF{<FYw~C6rt9x@j=o)A|JwdhC2usumK6xqc{~fL$T_m%&7LvrH!xa#
zrIm2=<l$rI>VD&2;Pd)e?ub$OE=)Op_3`l(8gR+k>hTWXns()P0t;idkVlOXE<{OI
zM<ydS5L`F)h8#a`i6&jdRhctY+um#|FG|UWra)pF=Ql6^fc6*8BUOA8KbVMsUAw15
z%405WxUH6&PhqcG$C#>-cHPX5phZBsuUA#E7NLh#S$bx%>|p+wY};TXXgeVb+TX;h
zdOQYCg`!=^3hMRa*2CuJDb<CEksfht`oYBH8!S&bvU!aEB{vX_KGf;VY*s=PSM94r
z-q#XAy-AebjB=kiX|0Zyb58ahRUp39M1Z|}OxRK`0~r(>@7X?B|F8yW*3~4KEUnTH
zOKgOMe-6whHBvX;r`xvC>RA6PdtMhzWmi}Ziv=E|6~HFnKWHUA!xYk*w>wp>-VBZd
zn=;@1h|sk%>!0J+4|{~=s#8Y#bnL7<2JwZYr-FWvu!L~i=O4zNI}Mh=O%kYcXMc?2
zoY)wf0-~8k8ZXF?gF98IFDao9O$7003(YGVdfM$HXFVTYVAJxh$YYCLy><}G-(m9z
zi+)KENkCN3bTtx;m8Hk_vg`f;$ftE49CcTRw?db#WI2jms0%_!pz@Rjx(<To3RGNG
zgy_@O4OUbuUwX_kGQ!YTQJtJosnY;C3(+oXP()+~NV`O6QM$!g?dor|HYKL6RtetG
zY>&!v%~p>OHJed7)2|V^mw<x>(W|+M$y%k=Ti(bFsd+67Wgy4>b|EiRf+Y1(5j(h0
z9Uf3|;J8vl7+Ho7243ND%aMr*&knjfR^!CK102uSp|2A2sF`GJxDh|jpd18+N*ck`
z+h$!}B`G(^05%;@l2`zv^&>RWI}6ent5gsdxWDpRI+8B|bQ<~dg&)#HJOuBZ(8}&R
z(b1=e{8X{0Z+M@c_Agq|-P#8A4o~^3nY_H-@Uo!D>xd@|Qf$Kl9(bJR^=W18U!8jK
zdk$cOh6Y48-L@i42p~_&us-Ng97EkgBMYG1MtL`&4Q{TvRX~3WvIR`CEsNN)(RZ4F
z>xv4>ct0V|W}?FDSnN=JY;`s;?3)~=9V8|;;0Zp5fz#>rC}z1N?jH(9C<dPMu^Sk6
zra1737;LveKhB==Xi87ps?TS&;a&pC$31nf)kSwfEO|<o=iqR#%|PY9#ml1T#+MGY
zNF0dd3s2c>JaC4Fi(%ANtkwLLYiU_$=3D23Pj~#;Us3$|jX_sms7HkvvKByNjNS}(
z@7CQ_s{7NYh6V5oz0jx72}t11-ow8H&IFx9pgb>k-%9&@uAGfDEG3s)l=rEjM9>r%
zF3X8($MYXpMXCL%jBMcnEOHhD|GZI;eynzSwNO5XQYRg3MspkP2ua~Epxh4XvmpeM
zcvzy&W&u8Uj{5)j<f4MArN|{5KQ^ihM!w!&u4J#=nfcx?D`S*%selpl@|x!6?ylgD
zmD79Q;BZWeop7wQ937sn`CyqE5i@~n1Pb(aDdZc?Sne4$?S=!x{U~V9*HEBt?^<aj
z@~GO@1-|xjsg_(`b-oD;5IhRymjpb(s=#+&keIMw1n)Je06)$PV_~i{-&~L!He@gN
z-qvyr99oOeglh0x^wF6o$FnioFh15q>_$o`QfmN6)+`6#9IreQDVq1x;Vwk363#Zp
z`8tY(PMWKV)tOZUr^q+m2GcVj%;ybjWYyfI1!(-yKnE-FG=nGe{$mF!+=58M6EbSg
zKRr{sWAd)Q4@_%(9e8h$Asc*}r<SKx;_czNEr*hbr!~B4^zRL(mJVrr`!S2EpY4TT
zq;f0pn5%M#l@`VTo<PlC>-7k@59)o5b|!z{WO5mrpnvJu1s(4?K`!Hu(Sg<3`91=E
z9DFU@-8Ci!prA1zL<jGu=Ue<;zSGC0>|$cauKEw?fGT^N80FiJ5}wqmf%>Ozgb6{$
zz&5?Kl)m%UK%`goeJQ$YI)sSu+~~^iud4^++p`tJenHta)+F6}dZLz6p{kCu0C4Rt
z&&NHH=ODeqVV5+qc+CK-!IkdiUBiI%Gb?I8Iz(Ed8wjIzdfK6>!0lXS*09TDO&N(_
zk9tGP(AdF7#X7Spb_{*xHy9j`Pm{*FG_C8u`W%G*@Ud=UM(?sl8(T93lH+J3^l!@H
z=hy<=ZTr7y5=_YY_FJ3?<4zWeU<tKjsmcEt+CHWT3ArnbSik!P`xK01{~qG_7{|)s
z2gO%4pX%Gvrm;T@{2QjS>$fEK;8k?n$eX|qerW_Sx&n}88a+|=wo5YE3K9;4*g$N}
z%;Nt5oFB%dDNl|R7kuo`d(3yE8tEC$RP+_jY=@T@g&?8t-CDT`Zg#BC(r&J)p)Rhw
z#J1UO%c0tHFKNOe?S!Ue<~0Osa==7C0(L>=$BV8q<rddFI)zQ{X7Qm2*-FR(tWxoO
zD734pBfkKzhb^b6a7a{9Hfmz<Urx_fgFZQxE?R5LxR6yIU0#*nv-_!|v^TO`$SB5W
zC=t6;Wzd*3XK3Qgo%G?<+gD~@G5s#T%xqKHNJ9GB#qgX@g!0-zT!j+fXZA3eE?gd5
zU9Spvyv-8gVx<J{L+K&)Om>_L4h8OA6xR|TbjFA6?ndW`BAX=In?NPq>;&~NA6~qu
z&A|-od4$&=7!^)*7UIe=wLgXu#f?<mp>)T4(in6lpk|QSy<ofvMAlD3?%u8Dnm3J5
z>AC)y6QHN+RvC*~VR(C7D|pO%a@}gfF_<96C}k$3b72IqDm&AB)%*M%b<6>BDgFkI
zeo_Yi&*ta9W^@jY|9M8=RF$?lU_<G8S2<akZa4E5PYk`FjfEagI6yMAqPI_l15uZD
zft8mEFLaKfe_wAIr5=m7G>IL9TN>>nANpM{IaIDvGgx3XSkOXIQzJV~55^v+Cn|-n
zwJj<sCQ*qUlW^jW+NDAD5|&ksI;Fzr^ZfakvCsr~IXf#+BvIM+d_Q{ld_+9PtV_>E
zqKKrxK??-kn{>@XmEoG9k4zP>fxK-3s2Q_ZztxHIXc;&x?{=2jUR>jbG;~(CcQ=VF
zy2!5l!b-`F6C+lDUJw+9R-pw=JLf5VDK3A{ve+f1Q@IklO7AYE@x17Aulh^m3;R^u
zU0~*sUWHI3<d;7%v_B26F+!)MVp-8u;f_{?yS&y-9@KLaSTH4+N(oxf1S1-QK|^lH
z#{RlEvc+Tot#bh^0-}u2Ou*H03(0dbFK-DZRoF)-vWSjrU+1-TY~m;q;we^-ug$_-
z28cK1Y+gTU%lW<Sy;se?Owp4Qd7Xz9q7kh)&E#6*>9e<-B*V`eAv~!RFIp90E&KIZ
zCaM7EK#o=!Fu);&W}RN8iO|N{p?Q76qDKW4NNJNrXVM02@l8tTt=ab)gm!!2SoX_j
zGS=D1D{v>9VrwJ4?Sp21qR-@o9<|h^*&aLcK&s-~seLK4hS&2`g;=-I0b|EvH1MCH
zFJibv=#}o};?Lx#<weQPavf1WcnS}03H6NayV88(;<1d$$)iGQtC6zPoWF*U*S)=d
zWeIn+8;gm2Ao+7<re_&VGA?8DX(8f!cdP)`8ZsTqGyBS1#ArCDkz~UA@*9{smhN)1
z!q8-*df<N69*Fv4BM*C@ZA6g7PXw1t?5z5eo+5IXO*PerujG};IH`eBKBIi8=s;Ni
zXKD9CF|LZhu{ZyR=W}w7Lt%s;2_wbv)!%kuha=gs9WDe(FE<aoC7H_+e!lojx_C`{
zIbn4?9W1iuy9P|Kq$>8y_Ta#ZX$nAlW0m`51M)S9-7D4c%p6dJ^><dLp}i0F(gu`F
z{Ad~{2D5IaGAJ4slA=H;+wpPWN~s<!bknk>YBo78r_emqu_sFY2JT)8<Rp=-Y*ulc
z*r*>3sEJ9NVQArycsi#^ntCiS8nZcxKab&GXUpujv1E}bg|Tac@#3Ecd(lFxK1U@S
zX1cNVZUTNU#e?RO@VP@LP)RRwsP1+scfrDSFU9tTx=JM0wKj?hvDV8JIqF2982ct#
z^e|oOiLdRVeupJm<WO3d-_I@WzMC$5onoOD-kWGzHWeGY_cVCkYv|e1?>)QEI`8d)
zNZ4e{9Q+)Oy%=x3VM3bx*S`|5d}WxxL-8whk12WZxJK4ju@zMzntNIbg1(OBr_yhH
zij6-<-yTWci6{)syK~)id~5uKpJJ`&4QJgY04>*e{Yt$HdCK9x8YGY33k~Mx`iom>
zxGsiw%;K-cJsTbT$0zh(jEb={pRjN!lHD&*PW4fnWON03EoSj=pmsA+uYja!v#cgO
zJa2e%=;Yl!dv2Xd)dAr_V6C7<|6JwOO7A~rn@O&sK3C8{am)Q`fmoZNSkRMWI`T6k
zjA!#U`A3o!J#Z0(TS=TmWDUd?M<2zG%gOA=mHY<u1*ef85f*Lh>A$2E-wFs%dK;F>
za@B8cX@Cp+6o;fhBO|#-w;?^oL!cX<gWooRmZ|?1=;8Ac_}%YY`cS^--)i5<o(y01
zLFT&T&#xBZ6^vS4pU6cBoxqt6u_Jo#sew)V+k6z-6TXRQN1`iIIjt!kD*eU5@ndUD
zYWTF?^45(zU5O~ARj_qU5tD|hRrySFk1dG^PNZ7MOh0*dL0=5{-jfs9Z7PcSqrvv)
zkI6*!HwjZ$)2=o}&o(4An!s|kFc;)rF|RvW7f)u@w~|g@yiwSDr!DvKO~C0)(+t`s
zBK)-&FOykr9F>$Jh`)^;B9Pd}?9U^VF;Ce`LNo{@5H%!W86Iosk(>h|`QsA<!MXIq
z$zJi3t?i?$BEis51FEP27aP}*n65xiBsbN~c<yh&C;bCiR<It5@dv|)MOMh8;%kqb
z`G_GLqd-k$T=$Fup3rSnM6k-qytjTt(T^zrnLJ-r9LZ@42RrdDMmvfzH#R*Qo$y@U
z-&o^8{&&7>lN!r=f!)w<f~-=@Z4FLfT@~N{2;Sb$*<!e=!BEgqvy4!eeaCVfpo^3R
zG&k-yxTyy=r10bdDk)^-OtT<GROw_REC7MXh&-Ke2?vl`{ocn}PDIvMcXp)Ey!aTj
zomqSk8o#E_>MJ4}azo#8Z+!D83WVpVbIol<44gOqqcCVM#^(<b*Rri<pfnOiDyt@*
zm8x^WKURPF+VxlBs(#xU1i@|VB&K<Sio-uZ<LDeb`fO>tUBmJt>s)?l@+uu7_+&J8
zQEKFE1pbC7+fb+>@8r9pfGqGljDE=i5i!RTBTj)6M#sSSrb4TD!|)Nv(!Wj84-;Rj
zWhqE@0D|4w_0sLK#L${R&cF02b0_%wU!|JuKXUsxe<-y7^PWQa|BG*T#G8M?`2vXp
zWXlaYt~kPIs1lsx@pPf%If})T3aQ_>UHZU{sP_RT^9#JbUfvx*B)I*fQwPgK2Yc$p
zgK`By<PoeAtnvzq<mhUZqsQVn7(QAb-gZ-I*joDvvzcbL^V{toR_;Bm=|f?1w36^q
zs6}clI)iDE#n<v)eDAi*8Qns9-7n`?Fsm6cEjBftk<hSejOI*vXs^j^xOZIzt1;?Y
zbmTs=9DzoG2`ojC4*98hT*roO8LnRC=o>S6#?(dRQ6%SMgna1-t+$!D@j_BgT#f{S
zC)@+sf1Mf{4l#>U>7F%kDbnjsC`e^<Dr!bQBM<tVo}@J_oiX5E=vSt{lXi!ic1e1%
zC4?n|#u(dv`t!gW<ttT;BnvyqX=CWlBOw!XWdKr|QR1Q86@sVdT08lX+rbxbI@HdS
z(>exY41we`1dE{9{q)~mxU2NiEf(kH{Nl}}5c{ByJg3nW+rLT~Or@Ugqg7e3&nX(~
zD@N>KS7*9MPr28b^hdN_(uK(^3lNIeG{IDu<PpV2*c5oF?w3>?>s(5@ci9>OQAS-Z
z(h(}$<yKK97VCp);huey!p%Zfr<Ut(wDR0bQzYq2eMMqKR1qUc6MtIpm7-eKj+-4i
z2+;b)L?{Kwb~Z5eD$oU8O+Eh1sFTtecGvT8RWPT@udIqK6R%^Qr+N9j2O^s0`sdp)
z+?=;#>w|aTP|5hh%gazTX+&O~SkxfGk<92}aU$${nf-yPg)R1Iq3vcMPZoKOq>h>N
znWPSxA}pZ8no1fJ?IN5+`K648A_V-my55LSnFt+y9+^!WPj6uz%u<enwk10u4nDmj
zpNl(4PTBTj`ghzbgP1rK?Sff@<8YWEC;r_hvy4Jnfx(;=PYHLqHY072%Bw%h2Bgmc
z<lPm{@f2aqlL1HeG`)?DyyjBHj<F7=GA0q1$CQlF2q`Ny-$V>g-RU_}W;jZEzlNRh
zVL}1cH+U=yP`pS<7Hff<Y*cZ-$1{ScDwUrq@FyA#;zl`db18>aKxn%`)MOccm5yT4
zUbgQRo3tJ~0vTT9)hyU@w?zuG<yl%vih<=ByGHsE6)SgV{K~oaK=-nR8Rf*k8%!aW
z?e2N4zA}&sTQl)-8)LO4G0dl{-Pa3zEa@=J5yid36#t!^l4Bp&+?$>+*gImZ)c^df
z|3g>F#KiDFL-B%Cr4rWJP=0s<J64=Y$Z)YhgsTf{xPOwUmzRbU#`~s$h4e{RP|9u?
z?=EV#tsMjM$?Fx1s6pU!=T1hJH{-|H;>J2~U-#bMzK#~x)yc}y(wMl~d;oXEbINo8
zE%IlpJH|zYjW4_J347AlR(S=J)6XgeYJ5xfx3TA}>*kh>$<YV96GRsG&-D&O*M~)o
zEQ+Smh{|5r!qRgP)iL3qiAZE7m9bCY-yVN9%Bjg9sI#0AAl24p7EUx=#2OG2E=>b2
z)rp5HtYfRthDit5)P{SyJD`~o`xB{0b6>ob4?(c}kdi1JMFdMDp1sA~l_!YahEOeC
z{Yikud~Q7QJ;&=&2p#|+GX@)st79<<HW%LCoqq6;BQ{(Qmk?yoD#<iKr+~z8BPWl)
zos!SZ>(kEXqVr!AW(Q`+8Tb`v>8g5n46=@U2NYY}Pmo)yZ(`g(X%()^t+zRyBb;1M
zJ5Qg`)`~6>i`i!Xe6wDF7bU+mVdz-LzI_4qS$p>b0}h}BjuqziAgZsA(WL`cNz?f6
z*t2S)c=%fTd=hEMpjpoMyM6GEl|Bcs&Py&NQ6<?+-2?j9?U#)muzNJWio))2<t1xd
z%L6M<p~dc2ovp>$w2DT&GdFhL{fcw1+-w*nMTr4_{>+s^S65mnfFWIQaJ@9KWQ+t+
za7Zr4qr26ec2Lw$V8A;ps*sLqV33-gUpAl5EsNT9jQw)IcFFD{^X~B0_sMfxc(3-G
z1^wci$KIZkJQwZZ3E_btyXL4f$h2gSW0q3jto_{Eh@*sD=Sgs%**kLj?HCu=+Gp?T
zIPyVcOmNIo5ZSN*;2nz}xvV+9q|_E>AkIW-VpN0L1q0CDaccm{X!&tK+w3vKRxm*Q
z>g6(!zIa=Jx#<t}llZL)zIGBq9~ewD1tS{|?`u<q2JK*6rwWnCfd>g$-sv&Md}a1F
zFlX?gDYCA7amcxql;BU2<Nun<OqhRn=6Gg?o%zPGg^V0vghi3A`Ua%m#$!bmZHs{l
z_TqiO$yhOlA5C+YLFlb3qqWiMpcUC|`RF^zZ@xDQJ86zxU2}$p&HFT<*C6a5bbeJ~
z0d&w?M)-uFsdKFC%FRs%$vm^U<zv@~{^%oi0TlPPAZ~akK;(~z0N9(*khNX`ydQ&K
z)=RaHq8Fx(Ywip=*Fr8Si|NRL(4_oYwWUmJyfpJ0SkQf74e;49dg^U(UoSn<ViE2n
zRhZQG4#?wJ_iC(&gS#Z>&alcoOF|~?tUK^*aatH83L?;lWGery&pQ`91zU*x4~#+l
zA5tCv#t(h_2Pjq{)F2Rx%A<!v$BPo`ANv{K$HEIqXQeXPo}wGC9J?Sbp&&}s1(gH4
zV0UX@%|{v@<tD8OyB<+HE=hIgug4Z_8%MC|t56TV)#LUZEP7wpRLPwBm{rgRB@r9X
zgx9n~fOOB@bj}aj@dAo2bVwcA$NE~FvzwnX?h7J@Bf3;X3xsx+qusyi-%_3N(Yt?Z
zit7U1h(t(_Kr6^BHW3umaLXkeOGD&@y`MHMQbymTcGAQKi`oSJki@1Zx)#`%FTK*a
zL{1Cdb|Y0GpCKYGg4$+Pmx#q+uQOAWsx6UDljpExz8}pvWZ=d^pM<IC{kq(N_#N5~
z@em+claUWOz;%i)dA5_FA!EH?KRsK@u-?&k_4%!DAA~bIWNuxC!)QDusiX|+RE;AB
zL$Vj<Cm|bztc0F(OvleGooAo+k%0dp$f?1J;nmK8N$n?qtL_g(#c!I|4gP<Wy;GB=
z;kK=twry3i(zb2ewyjE=m9Df$+O}=mwvC-D&N&e~V$F+rBE}E+u10Tf@2$0`B3+W(
zPq6?Q+(42XR#7rkf~BAP`v%F?Qc$;t@cP%7<?+%Z&4P9Tqz$0YQ&9ymn={jgrV$ek
zNVf?i`te{1=exsT{%ouWM1{J(Tnkp0`76eokVH7OT-f?B<<y%UxeD*yS3p14H=B6T
zO`B%G1i4d3S#PuIH}1+>m<<pp49h|ZcN-m=sn0E1*}j=GnQ_@~)G`ySXjar>I*6S$
z27N4_P%HcV-8Bml=s$4VWbHyi@B|vpF#Xp98A4P7XQ7%lvT2IL0IXW8DJgZPnBP{4
z=sMLO;P=8!^D0df2ia**AZh(41FCG1A^H|@nbzVYgJeVGvWv0|VX?V5bqKo!N9<=M
zv^<e75@#LxZlIgRM?z%z=Bn{i<q5O(zl|&vb@C}L!mnVY`L$1$$t~)FGz2`h-BAg>
zjl+E9DHv?jEKxlB3WLEQ8G*DE>*QwH&z(juFb82IDr5Yo!V9n7Chc)=(vVQnJJ#W)
z0jZ#go`tOM3-2if*1fVTy{^tBgGds1*T)c0ZV$4|h$7LS=0LZyuq7MBZ?q%Yl{mP`
z=J``to|g@^xFU{5*!}S{{&QN{S08SeAx2ZbE_bY`<qKpoXPg{p${d&>jx+Q2rdJWi
z!OJ&KLJxB2=c>mFDNP<@hRhBnf{#Dho-|&r$L^o&N?*j}6(cH+Ada`Ut@VmuH~Jgz
zYog`dI3|X>Z7Fky63$4Jje{q$_Q8bx{cqjGbkZ2#L`RtC_~(R<1^Q+YrlTLB+!TM?
zPF-m5NJ(&WGwGXv(?TBRSX*caHI=qD^!Vv8bc&Y;23pLZu=Ys*MZW3%P)<2SL_)yb
z?p08>Y#GhlG3xN=g|l>s6ak+j`HfUi0{hHE8HF)Q=<kBi;i~>N|2dh%PMz}`2H3Nn
zbHnFDn4DIDQNoE~it2(>fZ;gad{*QApnuZubegZGC}6J>u4y6OS@s8po@6RIm5-4q
z3<&QWH}y2zfuWp)!;Ai$kk_xh0~S?T<)YtrN%p^&GJHl!*Qkt~S=~wTJOH-B)P+)&
zURM-usnZB6{`>JxOB8b5Vk|u7M?_7hOqUiPo4mX^9=}ke;EBx)TgZI^K<El$g8Zvg
z`f$e|VE=(%GA5H&Aw^^xPQ&~WCse7Yb@k1hRay3*#18v^c*8Jqu>9BS#>pSCv%!M&
z6C$~Rr(bogNF|g!j0ZzW`OQ}IFB3}C(q=NcH?DW7W)*GF9#E9R^K?;zo&x5SP-$NN
zeqP;<4tY+5pqI^~oyp_Vs~IouDU@hX3!{Mol`X=E*_#jd2hLMY-cQTt{RTYZew3xJ
zK0E=8t+qbf@BZO^a9JI2tTI3xS1|AVcVKt#?DX(FYH+Z8JZe0Z<7w|~(f#G|Vsd=h
z>^N~{)QM=Mm47d@STTjdV}>Zhw$f{3&z5FyoIPrJCIPBwV$>uGi!FdRce+Z~2Cd4w
z(lkoH(Po*u)$Mc9-&XUnY~~UEtJJ-0#j>?x5!=n8JFYYE)2dBrQMJ9xdbw$ya{nTo
zG*S{KVF(yHEgS_Y#he5s5F*M9EgkU1kS&84^v8RJ3i_7kq~=)AdaC|@mkU4&`<ZND
z;YU`37Bzn()mQX3pA=#q2KXuTrEE4Wf^-J|<ucJZBiW;jp_r1_qs9d&nL1UrTx3@*
zM|fzygup7{vX%pxU(1`5JFlZ1Uaj0NE|EHHSZ$Syh#4HT@XuQ*5fjx@<3eVQD9Cg|
zd+fXDy{|bD@T<H!%rz%J(&VrxH7r*{<Y=_2)!(ZYMsazu76yp8?oa34)koYB2HZh2
zhHdVcAm!m(am-4~4T!khgazjwoMk=W!o-jM#_7uD_wXkKmk2tV$1UA#0AAnBmHN0F
zkE~j~YH-TJZLYgoT482I{kWP0;~xUAW>>Af&CfkRE!6}eNXuRET5fg*IP}T%)M1*z
zB)ig&!%W-+RDG#7GHL0T>NSRh$sA4**q+5Nqyl}eXG05&{-lnuJ<Rz8V}JZ6Q#a0`
zXfN!PBeXC(khZbvA$j>H7Z#^T<C~YQ-3&Q95bBHtfwUG{OB{v|%p6`iG><1aM)JR@
zLdYqQ$H5}M+b^g=dv-AlfL9+II~+layl{sbL_17&S=&h98l;lT8V>%3$k*DJNHvEz
zjnT8jx=>%Z&!lsKe-(I@O}(N_ur~vN9-w~Iv&8K!aQZuT&{q_nM*IBy{`F%&+5zc1
zb1T~8ux}y(|KXCJsj06^!Ez}PMuz7-b3B80^OD^nku-RS?xRhhM5of*ZL4ljt5hTj
z=n20v#eyIOo0<AYiKwot>%RPJ(DP`tzmm5e)Pu!t@+95>EhGv4Hse4LVn;3wo?EYm
zBMRzfSEkO)ok!uaS{PWyl25-;rQB}Jr`-^($;k~OiEsuKm1l3`mJ#CFc#nwyS<tu1
z(OHDY5Znosz|aW!kP4lI!{*)f#$8yQ{<`(-d~npBV!soMp?MxfJ}94)@skFbfghIg
zga+CR{l~A?QXM`2a4Ddis|*o9BB_XOFn&)O+fT6~VMAb0KX4>%v&Adc4EnnW`PT;Z
zI>J|k?D}I&1^hXCJM#rjnd>ijdj2nVE&gn<KpHfss104S=(v6!vuMM;9!Um*wOpmL
zX;3tl(SoFiU%4a9UsQ^M^(^_Tyn?_e$Png=4z&^tXeLX}rZ(chEbqdI{)ML9YHRq9
z2SkrKQPdhqNB^G91~#T2n{=>uI0e1}CjCJAohjmh3vw40dGL#)2?F_wTB{z8>@=><
z<DlLBepn$ik@d!z;K6rPvzAslr0tl8cp#U2Ctu+P|1j`0W@fQ6Lnyj(nFS1s>p=Da
z7Csj@pvz@dx3DBu1yU`n(v~e7r>{Jk?3)Jfb3fG`DXP|1Al$+5H`$_DuWG#6$KyQh
zMZZZ4ckHxzi**yBWc}Hhj>;6Fv8C%DOHt?OnI%7TIUyoMMWC{-U08<G2xNgGXm;;a
z>+$+YjlHsu1lx@7sg9z2*``@zX6Ql<nmAiSU@;aj<!2g8monP4gq2!)xsA0r#tn%&
zV+GgHn0WT2c5ER%v#JWIgDUsxYn-bw*l?SQ;ut3xI|uPBTnoGRSFuDycg<h=%noo9
z8=tfP4Fx~n8({TO<pyjTrLjaNe(Wnugn>~@3*UQ~_Ru*=tgd_`)jTmC-RCnUOtwY~
zKuwgC_hA4ycUC)b1w7BP@uIs+dU=0+Xf+mjd-iZKta1-HgyPqGsmPow8Wc4#RjG95
zbZe1zF`?|%sDg9hel%q9r5;~aX&Nc|Neq`7VxUB<yp}P}WPzigW#tyErb~Nzb_m=-
zLKAglZgYkEhjWIlFLB@B4CRruX>3kr9vQ}b_s>IvgGHEhjWLnpbtQxHwI0B+Li?il
z-S!So2;S2i)|K5BL2{|LH!&1il8Dp9d*;2jG1ZRIT$`HOUa=n-uCmT`Lrv=K-ff%>
z^O~YtowM;#>TPa)oerVJzxadcORJlbs5I$Te+tjd@fW|2t#)GzuhjPvKgN~49&md4
z;n}5!o?)UgBb+VLL@)tn|Jv%VMjmv{JEv?%BmSG?7p^21t<Bmc?@%v!*0kcl0j|In
zV@A27t+~*59vJ`6BltgD{C?a{|3B<bKWjt|l&)j-6NyC)DCofI&ehmU$uT#jS|#VS
zivSX0Xyb&LJ44O;^X4B0X41u^-D38NMF-AiOf~;Zj;-0(#Z~{}UH|ZU&$Jo#xDmm0
z%DE=iGzl`ZzDT_(=leV=-@Z*(xY4%v+nWQk{>{iUpKn)Z=j)$)58wOX$mzHj-7mRc
z1YLjHTE9O$!Ybd#huaBL-5K6;#_0*|ML9MVL?+^lY9OD-g_uu*<jMbZOa_jg_`eE*
z;eXN86hlzKim+)fF+7$|<F3ey=;fVn&Wukr7*%geso1Y|sMMV*RxXzmU9FTqK(H3s
za4ixZ{kzu9JvtlSL`6ua`z&vDfMB?_C?Tvx>sp#woGH-9kqtcKMiWEb<{C5C^wLo<
z0gifE9W}N;nP~`4VL9GlFKPX{WM4Q~6l*TqUMlO!R;oM}JPHs7k#7<MLsi+LmQ{B5
zyv?*13$t2CpqJ~&%*5-(%97n9ONe%a%57YOAcC<VhdL2K%0#u@2N3>r1#TL<L+Hll
z;IE{6c>%PIf5X8MsrD1yT#barhORp{z>a<6puv#YayFE$W@0~h>y?)+!@4Jxe($H5
z@(m}ljJ@#(-w@)YDZ8wksqm^S#T2dtEu+H^>I#&YbZPHPvXx+i2Fsq8>_@Mxu|oPi
zu3h4JBen%7WEN$GgGJ4uRd!6ROBE2_7mMuP4kuJOq%OBhHR1O!aJRaC*#Kk4nB!fa
zBwsaiL|hHJh>nRCSdr1vS~re_=`HU>?PV>5YEM##-2>G@*cMkN%-DjX*nxxCnyMK4
z=Q%cIEmGmE*xzqS-uWwW%Xb94sfX1zUzt1_DHE)bJ5aKEO@vx<y<kV7wrzWzaI3EM
z!&VdGNYUJqKiV{rZx#6brM=J?t!mHJ)J{jhY!fa0TX#{i9Ljj-R_4KnyAAg^HrWG9
z-eHtMJtB~SOOcIG9~|NsuBP1_`iG4U`K?s9??i=OJ9VVEGef(urcRe2mZWavz2Coh
zpjDXtK-D!DdP$`KBT=&f$2QALH>a8RY-V2!=HppLKF@~ZAV+K1*q{0w80mtX5(lgH
z4`}OD4w=*@gBe0YPt<(zXxRX3y5p68x|%*b?i!NvT!lcv9lqgwc_2;FVlRGg&c+nO
zE)6s^d{pIt!5du~Q3;OI@Xp#`xQt6Hn^$BwAIB(wktpxPdr9GC=?%*NtF#%8OX<?m
z<VOQGhj{`aNlAFYW94uc@(8CgJYnmoHEavsXK7Vh7;f<05SU#1w<)FVZWvMnr(v4{
zPelGw0{@W%-qHVPQ76Ub=C~=M!?-YFJ~opud$2wR4r3zJ&K@Bp<R?_ugDbf(FGG{~
zIT5Ghx!9no-SJ(ZZDL;#>L#LSr3!565+E!fnw#ODTvXWArfiJdv1&;R%DEabL7FG^
zDI1cvq$68_dxV;?xM+mf3jJr^@Jg*4S5Z(`&r^PaJlFT3OJpzoJonp$v+H+Qck$3=
z+!{A<cKhmGYf8pA8|(5_Ivn{t65hNjWRW8hVk;s=k{%7<YMSdRpqFCS=_=*wE!=R#
zU<Q&c5iL*-sakAB<M9UROM*RtX*C_8aF?u(pntu%H?{3lh*wXE;846z65v7#WHF|^
zfTR{5C7=aNQSJ{2fZdIt!(@dqiqW=n@E8$CIOB_qvveGehBtgX;tgoOzW#9HPlKic
zUs-s0Q6`32Ab_Mw;Q0ojR!nj(p~T(_MOQiBGKnksB`%yh^XzDj&4qK!>I=8l{$fVm
zhafYF#pV#Zj{Rb&#P>`STTkBi50PZc$b9;Cfs2>_+nPEgtp`71!CXz+9v>g1aMGKL
zTss(ICDr24>5?@GYj!Ed2A)tQiUgnmfuskS>J!)@_(i=I*V7r&9-3hDz%XX6M+l8Q
zdh(dq-j!e{*?E>XF^U5HbDXsY#&NBN>96kHzUE_M+9koMh00A7CmsjO`yh1A#V>R0
zh!eKkZ!U2O)4_pC8!FS1h~nsbZ%a(1I!Ky%cj7FIeT?v?&QHx#*OYeDM8K;QaoDdC
z3f<CW?g&fK0i!tzzS|B}HT5J~0Ox7v!H8(Le@=WWg9lxqX5VS;9iG=9%_BJ7%=;m7
zcj8zSmjBtG2S!|Yto%)0q&C~3n=D+3@5r2~F~uxF*auu;M%)pm^_bAUfP_)VeZFoX
zavLw?c*g4Zpyfdt@u(0V2AJrX0NXRqgU5;4*gq);$Wy>}ry41vLCFk~p}oL0m*}?l
ze<eIYVgu!Q<8DzaU-nb7VN+1<(?ye~n<So!UqJ<q7Q>?Dz@6bVr-|(bqgSp>DUUzL
z=qSljc}c4X3O{xZrMUiEMLQdOfy=~bBSM#XapATKoQKm{MAG0uSh9CC?6|`aJ<~#^
zJCfDI*ZDgl<Mp60XGLO`>>7Id@+V;1KI~69u9Xk=?_|_{@0Lu8KJ6`iJ>4G^5RlEh
zyM1riMu)@1`^Y3-o|Q??P>-TD!UI1Ke0NOBcQYZDU>f7_Hsf0Sn{wgK104g%U`p68
zYhcr#@04{5V67{@?z9SBhV|j+KL|-4a5NmL!m2A@ct=1lIA2}gR)}BGG&x(B?1B`W
z2n?@*@Jp37pZe0lngXjRC+}YCriHlBR_3J~D&xHWQz@-EJC3}O!=BOm74Z7V8mVE^
zmneu|EJ1%6Xn9ehnl6(H(W#B2ZLO<%63rqw1Tl6ED<v4;E<Ar+y;5D`*R6`7ARron
zIEKDe@v%xPCUKvV2!M_P)BVsoI_m5o_<g^u*y0hUgD1-j-;t(cFhebA^{y9k{RsXC
zeUc0|ASEG0thMhst8u1b__qm8s6B{iO*iaisQrl?3M0YX<h2{b{!}#zFR%<SBsrz^
z$FqgBpC`ui$4hzP%;LzRJ-!c;<=3#JUF-GaU4J)lc~@)cf7&(tn3(=|SrPkxUs!Rd
zA#K0$Q&#k2ZBoNtF?~r%%1&r<)XkN&wwISw7@@A56hkyGYHX)m+ukA$^Sa^!Pe4P6
zA5XwZ=--`SuP*9!6$q+NJT6b9pf7|%NwBC2S|mr}Qcyzuty(h{xSS_*aJs%IFR#at
z9^J%<jXK4`;Mm^ve!EIvPDm%0sxzyg<kY5;9$Fq{PES9req61W-ObbAh>2PKXK3|)
zdy0<s81>9Fk2Io1sn=&i4-%19*XN>?P>teR9!p0W6az|-(l{U=dX^D(=`f-Qw$i{4
zj-E{E%pe0YBT=7UZKSphTRIR6tyC0X-y$@OY~|$FwvZJ1z4K$BW1`(s5FOtHk&J1j
zQ_iiiSM<&yotb!H(Qp#<Kq;WHO!gha&zWRm$`_e4PMMALBN$ic1Do*#;g{OUgFfuj
zj$`I*%#KEbI<9OkZAO)uuAIbH6u8kdzO?EH{eFLY-IPInd8&|My(fc7?<4eY*k{kv
z!5cGsjnHFb+l34UB!lA{Ief2m_{wC`B@-d%x&$t%ug^<_!ba1*zGt;cM|D_5@hVj~
ztSe)Ly>ijQ6gIM3M%AN*^{H@9RXNzdA@Iuv?Gv^ab1N=h%61JR)_6zMdj;)(##r7T
zvFRSGK}ON{$7#qI`JYY1wKQRQxbUO+^M5VAAS(nFfwe>VGwzG{!=(WctT$tk{pFJZ
zgL=y^jpM*FrukqZqDvxyxnKxt36C(FH5S$xKv^18PyYS&mML-F%V~E1{H*}_mo$RQ
zfGQs~syh589K0N*N@umle!ic5GM>)_h0W7)$yHH3yUkKJ&>vj_U4r_Y=6My82^w~u
zjlMU+XKwWs`T&Dxxz=IyxA@v3n{M~GwqIi@Q@xzFvMx9L;}?*_egf%lp}*e1RcAK*
zvZ|j6A*98<ZCzv|u?l>960EcFzww5hJ&F`aI&9`)D#-FlJEWV*JAdCns4tbS_w{Ey
z(1|N0r+ixJvR*#1Ah@5iP5=C<*Iqk}i*EJJ<4(7(C98i?7h}a;kW;r|CPoU%WlMlu
zC`{=l_l~;8U<jjYMZY$g!PrdyG7KK{89R38VE`U`*%A>x_?IQf?=Z8Hs_t5gwT@lV
z6zV|VMCdttAY2*9xGV%@nPvAWIP+^tU^E<KoV}Y3J>go`dp;{?H*k1^vmg-(-TE-N
zSe9WsRo3o7z<DW?99X$BN;hTNAZfAE?`KuD?qhvs<FU@T807pnFOO6_sS*l`cR)Dw
zF5f6FFWI^T#gru%9SUEoHMXNiz}l(DDD-}z_rU8o3x(45QhZ~opFI<nu#6+MrDaw1
zd}-keMV&S_j>W;+OjA6sn_~Q0=1k6`SCa>Z1IdCzLuAcF(QP;9U_gpa6%)o}MS%7k
zR;GDpKlH+llEtX;=^eiYN2iy^U^2Hj{EBQB&1+jiPSi8|>VAlW2B`mfo=wzkx~yS^
z{pPMfkEtV8=tORy_a5u+zJ&h?MR_xT@OT_lni|CsNsV-XWQ-B7lPdxBa*(Fu(bnHm
z;PdOuq7XNYn?5qw8RGhf=+L(M3QvUq1)Av{7o==@wE)U8jiV2qNC4@WhTlTAu*>+^
z>*^oojI-rVEY2g&OyZFf6Gl3Q(Lv36oDEhM>7}m)$9bHl&y*?>>yV1*0O{Fqb=b>v
zOf<uUV(l7DjmC5aJTzkoiy<o%EWKz|D1sxrN~#!<TFfHa<_i9M*u0Kd=1~uE;NSAP
zg_?Xmj<cTrRpb(fO`|8EK@qQp$|W4sq4*rfoBVt&Tkk1BJm>)KE^Hcypfw(GykRq)
zzcZSW=D7+Obvnhpim_`S3z6?Om@z0qnh2NYXtSTt^=kJjBPew+aG*A>(YQKbx!Nnn
zcn+s=KJEMXP~1L$pUYlrgj1E9l9yRQ0!yNdh$r7jG9njxA%Zb!5bt&_igD}eum@z2
zk$&P)_8TB*#-_%ItAFLX<DfdBtnMzCj)WoTFwbNVDde!<1nc>k0)b7V<t=V-?cLl6
zXQVB+hNK3TI+K9+g**;)DKiXQ-zmL}`F>FG>kwrXk`xJ#W6&9z)WwA`!QiNn2t>69
z&dA?DFGe%u+tq=%?dF(eDo-XG<7hRx)M{l6%*NGupPg{kTU%SE2CQk#QdxZ%1gP-W
z6=EbE42|G=0dN8=Z!TX6fbwc>Fphnp@FX5tGOfDuO-on+Su4%G3abVkW8u?074}0a
zD!<E}L8hlV_ki|(!Wds|`+Ba7wQ>}9dc>4`krmDB<(#V5OK<(m_tGf=2mK(*a{5?Y
zVYJ)nxMuY~j>G@-?q~S#dDZ{3ddF=>{#m`J?bmd`0yC$nD{*SvUTM)u(^;P$)GPtx
z1(HhL=b8&tvXve*w+Gn8GjFJl6>-GUfkCn0#9USD_y3R2cRea~JPMONP+f7bjAyjq
zkX-7XGLdtiDJtdsgfAtN%*HYKkf-Z?_wjM>_o&CmWO+u$dGoUUmzQrCZ1vVo%SQ^O
z-uKPq<8$@a#$NYz_aEo?`Q>JTPrKL6DBOQ?^hwzJnh1A0zN$yzbQ}>_&_k$cw-O$!
zBBQ|-`KGW9P3{*_9z~9L82QnSIO6q}4>*N#^S`@JAvYEYRom?!ir&=+-NeJ#eCHn!
zi`Vz`nfF#sDy<yK+8Q$s9LrT9gEK&FKPH&%*f_SI9WKi}UHXDbC;~i(q_8@Bisx}D
z_NOx`8fUY>5kOR3%vHnc+w`qNxeQeQnrOQ4A*`P$YunrN-!DAg?WIN)Y*6hBi6}wE
z>x#L0?=H>rfW9&PhD|y4WI3^LzTH2D;S|m-$wvYD3kVj_bnFoc34fs|OR6V{s}ku`
z`nwl;S;?&UlQzPf!-V9#1)x#(+JMu=UVQ@dL9Y6KnC}5i0N<j2m{2*TRRV=!vPZ<D
zdU8sVDVD+S*p9t_ct#jt(;#i|2J^jJ{RJKUn|f-7`Hf+l3CnIyIhFhXKma~x2{&s^
z07`DiEC=!#FY~@ffc}mVUPi36+VJk&q;o{qX;h6ZUH)#RGsk%9H#wx%${UaGP40&#
zG9Nn92ULFi%lby?GrxtS97}u$sE#rH%YE;5BA_$<xC5-WOJ^EkC1_TteL}`q(}MX!
z01!X(H1`|MY30;JFi6nw^iH+##dwXW_7SZZtQIZc+RQ@{NXQdL^V7#8r2gQP0p!P(
z6wH|TkBupGI+o~q1fj&H!Fyy$JhWg^NYQ%N35?mq_}XC^Yp$&>_nIa4QX!+Pi@8O|
z#vT0S*nGPw%8<oB6_+8)m*?wAhR&C~G|-sH$&J4W(Fe$=$4%&-0B9`UR=+is(C&i-
zxoRG77_h8e-Lo0y@Pd&vC9ocFRIeqp%AetU!7CVfQyo(%sIp;KNuz)<zT}{BR@sZ$
zFqW6w>=tq>PTOa#x#GT?N=#(Z@|e#|vK>r_%di&H2N8#^;vI^_^6TQf)BjDKRT9Jl
z)jMG9dmeuW!!tgnG=OanP#7$^GAuP-hW*7WR&`@&MDUAKvBFyTwQCOuq>k~{#f223
zua((vv~*<;6?ew|_64-JwnqmL^YNCnrHB6$AgQ1PRKE!~FP%&K90d(ITJz1o3D`QC
zwZ%jQ$Q>4Lnq72WjVl!SUsIM_V2&f~FqGK2h_ljO^g2{F;Up*IjYGPQMP!ngdj69+
zc(A7?H%l%O9>Gav*Mmws=|$byW~y>;`-2z~r>-ej*Y~D_z#*Oo@W&~}6|@oP2`@~<
zHN=d!w}R6PDc=i>rO0>g?Lk#|m_@f!$rCKZgSO!Q#PWw`YEaO@Xa9j-@P@M^CEs&a
z>=U>_Y+4GvS6ay$zfXY;GrCM#tO*xXlv79!CK{w9#eemT#_evdCV;4<V<^kCW@&jQ
zj1Au`h?BUK&b244D~g;s2*{W{@vmq6Ba(F{?edaa=quunM5M6Wixv$ubH~-*JQZ9O
zcpevY#ef53otyAPq;3J(MQtT}i0q?uDe!pa<zkW})my{mPd%u5$<S0th$NYiCFV`#
zc#(rSV!8?m=mV!d|0;rUB4L!KHac@pZDF$NP6oU8Cex5Y#>Lx3XscG;1oj4g&P42l
zkFd?xIfVh(tfk1*&6hT97wFpx?$ZL}B&41-K{k>ZIeA~bqxZ2?tTAUVOzsfsGe%E?
zuuy4G9r92M06FkZhi?Hn>)^fjDm@2DIan~|>gmAtcIAf402DVsK<~x5x|Wl)rtqo+
zN{Bw+jc4{wMU4KoTDg-uc<d^lx4D&Wmaln16#|3!EL}pw@UMe{`AO>pCXbL(H!{W4
z#<jF^y1@EB-a;dQ$!oO-189(m;o~_<aPJXQk!!r%7TM4f4Zl*#h5}7PZjy!Rcs7Xj
zqClkQ4=SS4&~Z4aVX$>&M^sOYx`T%l{h!HS6NixN+7qAME%E%M+Yd6vJ%_=}P-nWy
zPrpmG6ZLiKpI%PIDtUXh8We-N0eY8lXMp^2QbSKK8|c|-wPB8Mgo<HCwD;+uu%o{q
zr!S@+;U!pD_Y;C(EXH?!Z>FL@XaSwmGudy+cK-FO?=Rmv9<t94kRk_vFHzBLCKWS4
zc;x5+*}Z2(sA|}^P%{*7_7M1UXXV}YI)*yZ4E%*KOi!osG8~T_KB{R651+v1OvUt>
z2qYkAkmQSoFo|qjN~$qUfa0}nt8``p9+t9A_)L#5)R;Ioot9_eN=tdZ_%;WY9VZ9u
zFtz(7?PQSD9)s>_;=Xty$m%0nscS+wFGs$OeC)MG>=Ogoz%QBboo*ZSAP;f`;ED0|
z&w&<!8`RZ5wQ~!*>?zhGtzGm=iO2<WitK9p)&hRKib)0-5iz?2{8=JT$wgNPbu%Ey
zBOJ6O^B#>rMIkg<6WuhSNR;j8U}ak#2zw1BlL9_~E`aMR?Xwxjc7n%+S<i*%%#Zk4
z7{}G-oqp^{?E@%56lPF|;u1&Guz#;o8>UB{Ns0z!=p^q_yfh9Ev&V-`IuEWwMSh_F
zec7k_-;#lell}jOE~<Dsm=e*;8(As4*uv1u5iv3_{15QV$=QX7lbs2M{{O|HF|skT
z{I`zeNv&;rTz0rW&-!`8Pofk`BxLY{0S%%{Ah-%F_%*apkrj_^xgolwR7;=F@05V7
ztaj3=SEd7Ew(D55(KvpQ*FIKpOuS`0or$L`d@{y-F1cAKSDomhny4Zy<V-MQx=~AG
zx&yR0!IE*-2>~>9wTZpoxbpC^LJjED-1!nuHGxXT<$K|aDJ8RtG*1nE7=OwC3K#}p
zfvAlR%qT<^1gbh9o~u?4g~cH)T=iPzo2WQV%)r7#8rh*RQ0_881eyzx3o2AmLs#Sr
zF>eIfOp!~63MG;;Taan1jSu%Q#27$anJUz-@rFB&i3!q_odO{#5)nf*lcq<|QvD<=
z3elO2&Q#XRkVYBM(Fl$f)Iz2bCi26{tVp2zn~y+*bYK=*m?+_IfMO}gh3WPzZQR5m
zl5KQ@dB`kPTt{m#g1IYmV{;pT;b^I%s6viZ29ktOfI~ykbjobB5=rO_p?gG`2!#W6
zi&ZCjl9hZUXduFf{th_6fPf|0Zo&vnNHL39v%)WCHY|eSNLk6}<(1{Znu!`=s|r@W
zfT=LT{Z+!z7AqmMj!GfXR%YU^^_N0BQIIHU0BDf?5(T%6EeyR%*__1nu_z)h&Bbwo
z(y=2(fD$PV<<C!-l%d|4sDQ=+5)cc{66GFVTNyZ9fD5GqvN^jf6_YMP5wzSdr#r?F
zDX;E}39MT~nUsR}GV}JXziQ<c8bpzLYdrt+dFf7yNmONbdpHl+UrnTlIW=f)aq#x8
z;wCEO%1gq12M&ZKqi!QPaFS%txof6~&R2-Mgn@Odo=Uk!IL5phajniKY`L2JI1hQR
zI-lgvIreb4GvqtZ7QFUxbhMK#O8cnr=H=J_!|Bb=ug{w9c76Q%KA#8}uf7^lDgMbL
z4J)`fL0B5g#M~<Nj~#YYj-SifwfRlcrLMTQ&TQ&jJz++Xsc~z$8rC02t_?AqBXoZ_
z&Q|$x>H1{4%BPxNADi3WI&5^F=($bDmguJOUhhC3o}fqk3h)si$v^a#7S20SoZ4Bp
zf5C6N8a~V@0=<ZvBKB<4dC}<LVKZ3z709#23BBytjg$DD&&o&QNkqcL2OV{l9lXrR
zsfeP6<Io!G-Ws}Dz~iY?{*+Z==~J}mAzTEXjA4c*zAYUU!fVVdQTWn?Oz|2lOXGN}
z^7`W4*kpA1^!$__YS*o*?!NMQ_g3`D`*yy!cmIm0%dY>@Ezh6*hfV$C?n2mHee>(?
z=5loJbG`rYYoEGN!!&YWAUI|J`)@H;8N^(JU1~8>>?c$Tl{XpW<0v@`x&4aR`9C(e
zp#w&(tw!~xyr-=)i?Z+sFVl>Cak0XfNmMaXCkwJ6(T<qV+OlRgf2Sv`4>6I0yk@5Z
zJb00VSk}5LrcG-UyQVlxpm;tYG<P7POM?4zij-TsC1Cb@1iKVSzO>z!^fx-@ID0mk
z#HMd7-`>|aMyBRZ9&q`eJ{8%pF@t)~+hZqBU{`1Qyk#4XExf;csIk3lJV@J9Sj<{P
zYpa*h1AG7<?q9~Swfb0gW=T87b;5P(Dd$W!ox(cX;^e0oJPKkRw%DRSKFw6yKsB~_
zX%P;S+skOYz+q0mWrz;V;>?ROyV=ZNFCOeI`vTD71c>g@%Jw76LCjmHOgjx%J1$!!
z!)$3YSd_2n4_y!=5%E$R;=e;~CCJu5O5w@RjV~tSy>Tvzg{%y2<eb<Gm!A7rPIqKU
zx_U&-GA-8;UuUBd_QQWE(Xk_JMcXzsW0G%@$9ui&acstAs&fMbqUizCw~omHx4etD
z=tPFFJ8Wq+goj3uOY7LB7}aBlS8$}DgT(6?VWE1CBZY&rk$qb*$tTc^yI3`!scu}I
zt}=ysmLqIMwNoyffs1Fh=bg=G8aV`F;!4K+2~zG+2V4ka&+h(&*9Sd5xm<j@s^pi4
zgMR>M9+KJEqWY1TF6o7%sZz~kG|wn%&Di=Z_MTO(eOS^@Pw5Y4L-qiq$4KZqIY9iZ
z;uL6Urn|)91medgh78_GzowgmXNFoTI{>J!R_A=m*Dg>`4Wj8uXShJIX?1Np86Wqb
z{Wm`2DX}Gv3xrdVO$K%2HrRNfH(S#Y>C-pq;(h~y&Wibvk9D<R1{M2L&&qiKdnf7#
zW+R76=Kr2w%nY3W_x$?*<`XB=e{HS*pC|9U*#EfAW@y*S9vSF6cfin)a_iy;OY!`J
zmI&5K>P$#Ie=KO-0Q0uT4P9ZP!Y88`ZDwBH&zCn`NBc|1>%rN?*V&+BEl+4^A#AVv
zWPlL;4DmkK&5`tVXLaJu%l^k=&~YF|kfqSDOe9iLi>AEapAWaMiwh5O&@1bySaZRV
z&)3!cd3X9_`t8gv9^ZiL?3~V5aB=xR-nUa$e0wJ214vT|SO*)ZQ{>3Zd8qbXA)8Hy
za|gMiR(=_k?C=Qery;$Hk#Vr~7Gp5Ohmmr`DD<JcLWgpKxZw(nv<0_x)sBO@Vi8bZ
zq*fhfscNT7EX8Gx>#FVDl9(CQwIZQJQRK(It}7_mAv7~i9cH@tQ7iP%?Qh5z<qg|5
zOiWgPF*rCV{Aedh<PHDqWfX0e)emuHL(+r-l-sK7jba2rsZEhQ=FJ<QuDo2I_HfmC
zGYEb8&SaFLVRuhC{Of<)*Yegmp+hk6hc=k}4(}NAC@;|OUX&OUp9%(PmD!<CXt5yW
zEO^5fNAW2_o%%V(Sr)1$>&2nq2Vd*^lCuBtM7;i_Ip~-?D$QRejN3dnGfsb)eI7V^
zVnW-oTYTd)Fa+sB6vluAQ8wyg9Lx(wFoeaAUFOi<_IvY?)<SGX<uTz(*l(nZa6M_q
zTl#BX<wiFzt7}d+nsTJ^BD%U_;u`p1FghAMXjC_OPQOCli)nNgr5VOC)k9|26>`HJ
zLCubIVO=jGDFzWFtWX1FlA)H1z+xg4t?je648Lcd{b}fsgt5#)DJUP(KZU|AS5Ct_
z4ey+8o>|IqYX*eD^f#>_4t6dH)CJu#Q8(U7*<bWT?8RO4>YfZ?+Op4d$;N_~FMZsp
zX`-pPW>>=5>)1*FYBwi7cgS3A1Ca$87&h=dk~~2e8xJfIb-;i58%O0gC^1g?d!B_+
zBBBQv3n{s0w0U~_0pb9W`DIJK7Xm3OO}tW6n4+$oWfC3zT)yOTF<<ZMF<Eh8&t2eg
zJ`J7BUqL3^%)my@^xDl!^D4d(A>w~1IJ7poByT2F?n4Ub6>BKglL0sz|Hfpi?Wx!W
z55<b=M)6!|yA`bcF$+-Ei)E8_c{?zZ3s{n*O0LZhOZ@9O$N>e*2#Zs3R|Ptcxm6|u
zeG<eaIoBBo2zl^7DQnPClFG?Z!7~_HrszYB-Cc<FRbmimQoXq4CT@TQ%-qPhqI!f&
z)c%Z#qn6h&*jXTh;ZG%Zo9uN3HcKGxxh|xM4#EdGRApcxyCRiEKrXVlnJW#+H|Q}6
z*Eh)T&RQqn*qRwbbAy7q0<}PmvtnQhcan{+rsY&3L4&%zVyP%E^(D=HqS45jiA%*I
z-m*FYm4dnpBEmZJU6^Zic41RufPR5tcH}zulDxYJGtp$Nnbv^uCB=zuJ`tm0d|?4m
z?RFBSBbFvB=|-C6gWwwE@zr{i0T&!X){WdfI;j&#ut`VL#XKm@OAlxl<^5-q|KA>U
z-?3UY`+?jUQiod!WI^w<C6eQ_R;>-^#<61LY<CV&thL9f1s3dA@~=FQaH2DTtcxh{
z!i_bJt`KIpmd7_I<2P00MbC?F>$yE-*Q{Mzbuq>PCZPiB{T+Qpm0aJ-r@{QZtC9~c
zZBWy7_cucjy^|XF-NK5r6M-e2s@<-(qB4~~2*uZV$$Lp_aBj6^fiSsm<Z!AW&=@-N
zhdi`TmRrLlcoXw~jo0k5dASp|yt>zuRh0|umx}%8#<cl#XzHt8*IH+y;A_U+p0bGs
zQS6d>qK*E!#VpocEKVG-wsv(v)H#5oBfK8*U)P#Jv88OB2V*t2N;S7{WJYV<<X}xX
zr3jKcfwL)AH({?O1g_3C&{OFGlWnDF>ZT4fH?J--ICQ+%86qb*tWf<8IUNxQ%VSHK
zmjJ3oVMLervl~~~XdS-rwmcI}&T1~JH!56ab+M>fxmja$8qPyp8}wMJ;MLCHO$=Qp
zgE)}3XIrL`TcBjDfXToDI!VJNI&to2DwIt@l{qwREg$|m^&Vp(P7p-si8*-IUaYu<
z0TVKx8ki*cOA}s#TlwS}Tbo5lVIUw1jMDYmF-<qO;aU)D+UFVSqx8~#H;42mKya)n
z%R0Zw*9r=DwFWhodrNY3_A-q%)>kiWJy&6%=U-n<BMxJ5)x1lbGnx7PC(O4$mp~X>
zS$|+fX)r<D)xLs_CxZGynKlZslZGyC;_N0CsDo2~rQtw=fSYXr#p+JzznH1coeoDR
zyADCmU(yN(-!h(u;q)4qAVsLizU`tc*nY-Y@g_w6@xucQUnI{Ed7a>z<1Iat;t9;<
z;%B>4%)q*Ah1!Wiwul5o1!mDPHZSS@vhqX2n01Q>!YdCNN;Lv5!*yvoK97?gZH@k0
z#oDTxjGQpdH79?JoE9aV+~(QJ<z1U1sQCDJ2K5YipYd;Vdty7-;KFffq<z-bsUKWn
zNuKD90`So{LEL7xnVnuN+PiPRypPMjpZ<VjlaRqC+u3_B?<1;06+OMLUB)Ekl24D1
ze<8?K_WrR2z4=$j$O}yT1zd51^!*=)`hSGoG5+@!a?SrI2K%mlqh@9G%j0IXFiR4(
z-SwPQSv!j$fxrT}i8Jpf2K$jF_wDUo3nZ#kx8Gt3z(m%e@ke}rsmX0|KXtqg>%2bp
zo{j~R?@TKC8{P350wAG>BN+;s$vEEf=HHR2T@{RX_`cZ8=wA(xX#IYBIXSs`xVbFv
zXGX#6E;{}YFLz|i84UPF@jDtezu#A<moM3$Y+nweFqg|fJGneOm3eV{8Ba`f@5b-S
z{^a{UT~*8teK8s}0ve|6U=b_gwR0cP#2H6wqnJLLYp*_kC@nxZ_V&Ag)=2U0Q;u{b
z?)X6!l=NuRJS$Y?BR!zXLyel0b%=RmdUx}nn|6!m3D7COK%>LsNXk{37eRG^(O^;?
zYa-WBMvS_Qb1ff^cp$-&;Ku^ZIG)LTQk?P`IbI7mV2Izu%K$?#ca*fv8g0D?g*$h9
z*#Y2xVO2Zy>&!I#8l~z>>~#?K>FB99XJ-Coe9^1hY8V-+%5tU+8{|CwMT#*b;j{*d
znYj5fO=z1_YEF*Z+WK4S6Fc^`N7&B&c2C=@3;bX1QI34mH61-dj17*b>~xdf*;QEy
z-RWcTNmYreZ?<}1gnFp6&!?%+0lq}9kcMx=o_@~B8$fdXnj2jJk_KMb@|o5cWAIm7
z)<MKhO)4|P&4oay!QGw_Im)*Xdz;otO7Ks2*43C*qwW&AtvXFz0vP$WHXzw`VGO4{
zC)w1u;2u8-wp|i$Tr||54Dv$S-+RflrAh?}5(2uR3<r=h*FBFfx%OzK%Kk*Q9K~b}
zq-`s{gjtx4d70lEb1J7i4f8-$<}xo-Lq+gb-K=s~F<#=D#VYEi2Lkg@;Gaeih@hN8
zN)<3=0fU7RjF2%ki4x$a6QFBO@-*92g7PBDtyr`i)s{B|#v#GhbIMyktrKMCD8~$T
zz>HKd4R->(Lu5t_0S=;v(Y)CP8PiqfWR;`|D4z-$go2O|W7zR^Os-_JAG+!f*ECz?
zqW)K(gZGpFPmI8M(-Gb1?(;P)V6Bay1)A+eSU%CafK%>nT8mn18A(ImPQ|uJe;H(t
zlwcN>l}8ol2<QrtlvG=3?fQNUtM`R;304t__XcYy!8qh%Pcn%Xi*r@=KTdfefzv7B
z!ZSiyT5hX_D95d{INf8#+Z$phIv-3vZz+JDsuZ-efyFMl|I{@*bif0JRaVKDPewUF
zEb8%<RV=7i@W^gld&2b=dM*z-&!21JUncB9NS>#vn2P8M@I23WGOmpqQu&&;-k^u@
z@KvVaGZUj>FaK6K->Ui>gaaZ7EnnOd5ise)<3f5&MZiMtiXwpzcMR5#U||lg8N0a2
z2Te7W%jh2rBGC}b#qeOkbG1Gy<qtcSDA|M$Dzu-bo!(Fmz$QcGAT|*?$qHL5H%k$a
z@M$Pj*hb<!NL=vGJ%!lnzVy{Uqq@v*)*h}5%Iu`6ay@L}TBe5Qr*j>hZ)cJh3mhDS
zc&NpXo0YF@_mU-6x4dLnPC>8{dAN{^$tIt*On7!SQoEkl85}fm%D?E<BhtPgaeY)h
zPP*{oW<xHdiBI8U+FQ-e2w$D;prlJFuhT%&^%Qr?$2u7~!CzANx7g??v=^Vz0mOKS
z$P!I`@z6AFMjh$ZNR{A5?W8hslVGy*9B~Aq@9<sH<dbL5+*AErnKA4p($%r_i*Aj^
zR&LW#l$29rGHq4`q8y*Xc_KlBBP1ZzhSKDcpGv1|B})Q?0+M8{4HaC$XXyJj*+NM?
z$T@FEd29<DCo6eA;dU@_+zY#w^s?NgqADFydZ+g5SVzK#Pqg#fWkCf<8%^Qv!@Hz;
z?EboBN}u-|1pJv)btsy2E*KgcL0-?264+A+f=yXbZSesQgN>9i8FWEkT6mwbH_}0z
zq}?U{H~4()@>cR9Xo*7wW12NfVTo4*Vim?P4Mbw}(3ATYWJSMmOD5R9<FRSyOL&nG
zLqh^FTksx27_?&~DVRAdtas0ah@7pX=B9cUcDRxR;#Kb01Tax80<a0&!q)X07Z~yq
zwF3}Mep;W6cay(lGRNk&y9`Dh(Ir1iYd#q4Pe<#fDhMei-@p>_dd2XCZK5<eN{2B;
zd}83+mTa{XYXgpZu*z#HryiJ!v*jEvY>I!z0-Ht2hpN{%vhmgHcbKpkSdTE%mmdKW
zh*Ss%)D^D@+&MPl$NK~Ysy|nkGimH~uIURxHfk3r^dAKW6Ugcu5^m%m{R5BA_duHg
z4@Mm(at>J7UF0N+9T)5vU(&;n`1S%v-b2woMFJwCu^=%qLg*$+L@Ru%!+pwWUC0#b
zMCHdZ)7PEYF?Yu4;ITp#8dawk-kH<VJOTziJe!7=MHGkbl@&5WE9Vctc)^x_H|Qvn
z?v1Z1O@~qo?-b}8usRZg9t`fKNR0;KQZK+eII82FBnj>!)j;<?y*3q|&r^bty#~{!
z4{OlV5kfHip{$!Zx0(y!N*9X6rLUD-zyCbHiYVu%sx=o;f+HwnNe+a$@K06{1oTNi
zmMa`U4<&XypM>O}2?t)=-*9ngg1tA-fYxYz(Bw$}v2qNsw#Mf)287L2{1e?J=8oB9
z+D<wNf^=_Y7%^g8>20~20yhjMDO&hCLKc$l!QdHNcG+|d#zEB_=0ua4ZLCeq3Hzb%
z=M@S(Xg^TgPHXA^AQl)I{sXX%iHVcxzur(|%cib3-F9^cirdvk?IaRDT4tGZUwJ#G
zH@<l}c<o5V52TT(qG(mlEq^`1F#w4=HP+$Vzd0fM6Pq!l_4-7Lm4}Pj)9Fpob@|eY
zCMELhO0sB+kwhfPKqdc9r_D#Kov0u8oBVpZv=+PXWg}<q(*zOOsJiazeto}KygX!8
zXGuxY(JK!gs$b7=zuunO`g*y#Y-fM@DN>E&-?=#C-|A>iqAxiPN#Hq@5}mKr6_xjG
z`|Q-H{<-FM8930yXpsIeX+@dPMU;>y*)cKbP{7~oB@`2!up7Cw^gbTE|6Np%fc1DH
z@JkWRK{=hP_Wg@yLUJPsluIL(Ov!I(7S*<Cn6+g`Avy^^1Z7D^g}}V|{f2n;35Sqx
ztx)xI)t=-)?uqEPcU6pujpEaxNE$M$l)dfy6Yxn?ZW8^pJ<aFlWqk$>P1<0I(Z=Pi
z-wQ2&zwe9hz57wExIQ3)#gj_YfiH^k?v_LXH?#^N*Ht>vwpzJ#GgXB1d%SXdEmrbl
z3yoE1WAbBJ*0E=_t!0_Ou-=wl6Ng=U(<{MtwP6#T?&;5Zes8TE96=4e)&2s=$6b#d
zisYzdy7|cOxv6-zbWY{{a)^cKzjy>YP`>Lf4vYp6jsPW?l4$0H1Doh}O78g-cW!02
z#?B~-`xYQ7sKL#w3yqej_Yg}&)6Tnlh+$n!zAD$WSSaoI@YIoA4QC=4!H5acyZbth
zlPa1Hw=u8MEE5;r<((-o)mxiO!9UVr#+TsTbB%VrVjVY#9E|X>_l?<p4=}nUB|W>8
zBF!RC^qSG^C(}+Jh#uzH8qd(Q#Y{~fOgA;$K~+omP{xX?9Ke>lVEPH)Yx+~ZSSTIo
z`S2Q{hh1>N*8~OciS3~+)C$j$Z`R>F6xdjm?Zf3pb}<kxEC2dVVyZU5!PcUE5nCRL
zL4jOgS^&Z``2*e)vaSN*mQiix{yQWJ&QiVVLAlpmuMot}MB%w)TT_8->shS`8^@qW
z^+_cno@6V{oJ3>$Ww-GKErCv8JjGe{1}_^qg@lolO>^p~T;s)M>Av!Q+DWu^D>M{y
zcO+2xB@bhr;;>}^M2W&kheKVp04tclBFdHSKQY%nO)WG%)(%tv$A1*0k~tJK?eW|Z
zWJE;aIZTh1Gr$F--wq`iZpteB%A%dYln$Z%*!{xdNVqEe11>0hy%k+R8M3vz9-)q@
zvBFo*H)GwA>bIoT(*zN(L`cHn$U5<OchJve=1YNKrjCH0%CH(|hRC|7nF8+Sa?yfz
z`Cx%!lyFQRpYWLG;{w8iWR~$^4g43I^PG#($xgdxTT7l_5Uoc<EMx0a9scOk`%Mff
zyGaGVr)sErSnhnBc65!Gj$hH3$VxLe>~^$-fbb7c0tF2)x4eVG8L!%2(%wc=z4###
zL#o>6@EmG47VnX^FS7e~;0k_&eZJ?fY35u8+6T5BFhxbRAMs_@Jd0~&d3D6I@3zLx
z6CwbMJ6zX4&hhz^4ZsLUb_&bd2B)dFgL~;{?7<uC^{*MRv+F_g$cN5dsAeOJuD&pg
zpn70v8$Xl~NlK)Nn%xhX4iFzhZ12}E8&oT^N=rDJuDC!255yzkO?q>6XD)Mw0fbb!
z%#wtFj3NZ$?!%`Y#1TxhL7I7l8r}XRl!9$2J9Xfp0|ru2^5$wzrzz=vFn`5HKIT`0
zWK)r_Rq`HpNe$UN@xq~8p5ko0BNih@q0`J*E2MMI63P{;To(rJq$?1rL0OUzgLHO2
zG0>I-PC>L@<F2)rNL`k(s}6RRey!&|2|{6|aDOhwLY~0hS)9^VG&s_ow=(Kwb<cmH
zGPHo1dhwnoTL$WUN8Di`{86bvwS-L?@iSO6x@5X#z3+E6p(W*3efkEsB&c$@7+WlU
z5oS)f_u<Jr*!Qf5*KUg*|09Q9vNqg}8&ukaxUQa7kl{Kq#C~XXdH%1=lJ5<z^P@z(
zagttE(_4rVps<J+r4@W3T`!hIvOq<PWkO(0>H1M+904@v91`ZFYDhT{D>mRnIcZ60
zN(0)b-qNSmG9<dYA#Mm1LTLlog3om54r0a%mm#z9nK|9C3@`F`tVw6|W>rzfie$xb
z6>41yI2<DPD(dV`!~-I74`{s7(?INdXZh{;N2M+QiJ>dJpqb{VxEEqwMFrdAITKvZ
zfLnbFa9j8zdIP!5;L7>ZhxjjkD+}#~@gd%_acCz$w{d5vt{6-vK!Z-Bkr!R`AM~Rd
zCLEo%GH@cH$a9Vo+r(f!KD|@g!*gIK6`W!n-b5HXtz^g@Ad^bSOEIL9=AjO?B?f}p
z%hm^M$N^Yur0)c^`sGjQDKz`Owk>60$ONALy`exu@@M$|?Xr3ADR%jl877|z>Wyx$
zpUd!9vP5#OT45!zqepq9gr|4BWE4~m7KRVP6Is@`*|wWbIOBKA6K@JVhS89I1q9;c
zl=sO@(;9o_^Y`SGOdGFg<D7eClp$2hI<Fuucc<`%;LQo>5LuKh3l^E|mztCy9Z6s1
zL(TW{u?48#nKF~UVJ{rA8ix!1lXfj}crEmv<D$){9cExLIJN@_EV&Yi@ZRgRG?}ru
zzyV2i5UQ<~&BqZDzWNa8MnkG5Tx&&?>&_V2^Xg9!@7xB&61gV*3Yt?5b2+LH4XR1A
zY8{dh_%viX&RTp3-Ei!=+V-Go%ypjT&k@fc1HBSP*sSHST%?sg9*@Ip%Tds#+P=j%
z#4nD{T3r?HTEX?$5pI=_!6o(VL7z%qs%4O{TaRjGS}6!i4f4Z6(bK~4%Gm(4_b}#d
z*DSq<7-9CU1z5B#f@OOao(d~ey#}`3$G5Ly-<M1#N1x%(yf{ftuEd6H#rI)_6L}K4
znC@%sNe>{|e&B^~ccv4eb%H2sy_3P*bgV~Cz@|}TM6hqatz~r8d)1>t88!t^11}Bj
zHsSLq-U5|cIYGD>Vhy*hH7*GpU24f}413uo`Gxc8WT&il`u}0<otgw|+hEPIZQHhO
z+paF#>auOyHo9!vwr#tn_Q}LV>^CO9U$Bl=L_T+BUinuC1Qx2)H89VVlSTBi781p1
zV`@b?oNO0D81c<&4&8smm;4K2Xwpssh8^+0R1_n_|DB{~VqjzcUsY7JhUWi46#2Z=
z?rN7+U}gZg=ieV%n|3BvL{%Dl=$5<V6F?<kJZRx1koWjtpMr^wH)PKTVnr8XdwXpe
zoZ@!QdVZXzPF1eN(j?`;822OFHNJ@K4=ExADPG#RZs+=-fGO6s$0pAD{#?I=F7d_+
z|E!dbjD$W7JoMmg6CWJ(;OE3^`MUe~{Pl?p9C_;43GaHpIc`~rjZce=*e8fWo&_P5
zv_R{ElH`2(c~@kLtrr+(WZ$^n>h{b5V@h=5J^AS3ztO}9SeGqU4A-sKub<qxkXvcH
zj`!d@$11n2x+vh=HP~TYYEgw3WPpI=HHH0sT0|?`Df<)MfSXJsm*;Z-sl4l12_$6l
zwT`d9QImOBL$IG2vZpbh!1@`+Lj`i=V3$Y|wQ&J&SW7H2E$x#GVEn14oH&-zjv;9+
z`^eY!-}Iu$$Y8t4oZf_~VbCQRH%90ay%N0)Nf$l+!`NtP_C3UxvKEILt>6s&8Nl(}
z9>_jtb&+ypVb(R2?ag>f4gc<3$*-31F%^dGD*LtIr)^p7M$+6jrq@!T97v}>dNS2U
zhyAZn9k$DJ7`Trp3bR!Bv@#<`6FUbL>s{V}RW&6Ma5?qq*C4Ps;b2L|ioj7zkZ@*3
zSe|lT0kFXh@-n^VfM$x%Z_C{^M|AADLUE^DF(v8c7{%4oBJZ!8V`p(G`qx!q&Pruj
z21Qip?AdG4zGRSvn+0rPSA|_1{?jyqO8>iE4!-m<+RvSMzDILR3=p0sUd9ckm4t~|
zUi(pxV{Z-2lF;367|E3tVR~8p&){4c<KD2&HVCz-Ha!h4+daLJdvX~zJP!e2`MCfD
zVl1<QR;r76I;gU{+>TQ~1klVK{7Y2T#v(={UH|dbmqHE=S1H2oRs7#?eg{4+?^FW)
z4e(-l)Oty9q5QlkLdhn(?#=Nz_TG#vdgHTg+GsjJ28m!xuF}eAP!J}N0!0vMxb=>h
zsBn^jp;cyYx;&a&7#sGQfj1Ix<>}n`2I5fCWq5*PH1n@I$7wuR?7ib={?TN!pvwRY
zSGnUNZ}ukk?f1HaP=lZJJx&&19;9rMlc@>zLk7liJ}B43hz#H~L%{~jLP@vb^p^#h
z%;ow}Yoy@lmTxQmaE+Lc-2o^~zjt9(+;!PlD4Ut7Eusb5XwrRGSszhEL0e!ooV;hX
zI)EdH5c*-DN!vhf_mW4wZ#q)(ibTOx9G$73H6{wva~5Rxc{aCkkvA1T+Ivy$@4BdD
z9LZb~4W~a&pA_V+F;CRwANkYD4<Pi`4=<<#dncN*S2^@`rEomYE__M$3w9n8Zalc%
zGyvB|yP%q|27w|kymqo1biLp^di?Ef@$&>%)28kL-+lqf);|==S|SjJ6;}JJ7sP1N
zp*{0E<$|0kST$sbgVu;zk8nZ)ji}Np8>5%NNn&f{t4QVsoh6e8L?x3TWUSaxj?xDL
zzL;FWMRPH1*|DNNo$JlTxZ=9ar+^r%+ox7er3i}wkp$5Yl<P`eGH;dEh~{G0nn)X1
z!tt<^Uj-xqV#Na<Kzf<CIK(GFoC*7tmU0;};>{Bd*3%e?c!z2U5DVF|iqM3Bi1i9j
z6VbM|4jn88LNqUv09xZ7j*Xlof{|E|0VJNvMpqQYnIe#DSYb0-EgWY4$Q*3y#4GiH
zz(!+r1n1^+GO)X~H*pt8>`%#OpR=&N<n@%UV!f|B8i%>{=VW!Rvs<tR!O&n)#SaZR
zkfAsNZCWrzcUH`B92^zMQef*)LB6WbB1ht?_@`B8vtH-b1`3ODSINQ%6TQBzVxf!g
zSn=*DAes_embq@Czf3hNNFzM?1_DzMI6E0#356huJb{PUN{#?RLtq)g1mOB(-||3O
zF@OO}yC~B)Xbnjn<P;8N%<&<=weT{a3rF^gPQfsOlb47@KS?U&FAY_Mb6;(*!$*Qw
zxk+X;P@4D&;q?Ygm|L#b&v#blF|#;kBqW3SowNf_roMI=5p|}Hg6~ox4~RUX+Mx9S
zLK#4m#djxl`aH{%Tfd9=a70iMeHEi}om_urRrot7T0oVdi`#IKI1`oGO=0MYsbVL8
zf6!Sk`($(e<=3G0!uvIO7r1@93q0b9`3^6x%M7n7g3{_7_5)U+Kzj<y1fdUpYRCi$
z%6PH~V?>L~%4%5==1_l#!NnYig&YxufcLq)?sR2m1zt66>^b`o8#h07%cX0?et3=r
z`1Y%_s0L22LIQg-pd6_aFKAWmf3)L&dPhPS(~B1OF<VCdn7$xvyQUZ&*wh&<6PXRi
ztEj+<8&7@#O_2Ab?E^uxFCv-H85>MJs}YY0hYgUp3i?fj`q_M#no!ji=O>-PWh`Q8
zs||1&1<%};8*qe%kmXzNU(s9HFeh*(`Jq<uu;A{R1$*%Tx9fg5w@_zZ*qdmgk=zv@
z@mf$H<)!NL*7T(GCZbEFBY3^bAnkn1G&t1%g;aGHQk|uQr*E+`O`C3tEULeF1U-cF
z9BX)$nIadkmO+qqy3yjrS0c}N>*-2&7eD$tzy4Gr>`;JQ?Yc`+Nf5Z4z}9w6)^B<}
zypb$8lN<t>e(%U82z6>twYZN+C2K9fBEa+|LZyn3r(!+D`>}V9wJ?p#vS70CZT>MM
zk#X&t@LA|a5Z2Z?V|Haf%sX5tzyb2MO`c*>*;O^rYyb1r`VWD<@9=S~$mV^|iq0n%
z5|ANdhD8(}%;W$K57Td~Z;IXrK2!Y%+_Xzq_&-I)fB8DB|9e=YOVh^gmrL;7-P4~7
z$JAUh#IAVh9g*m|z+&Mf>uLwTA>6kbF%?Pkmy~nuab*Waa?NqS6%dXCgh3od-z>+&
zR7?{5L6Gb9sFbBS@gjHlh80&q6eIbMy*Kynn!sU(L^aq-V;}xl>V9)sq@@<ifUo{9
z%)s^PwQCnx)_IsIOIC6`j~`c|CvH+AWk5(;<J4R)6I{6)Gp0<%<9=+LR)$xF2a=bI
zvqDSVf09y`?0H&E)PcXKT&|rm)l~QWg(iKm*<x!zVPWbKS|j>+Y;lOxM>F$=n8)cy
zfr(TLCI}&j8gKe(OBJ*x|AB^6@6|;;vLGZU>SRn@onH3oG{biE?7gH+`$vtmw5Cb*
z#Z(FY!C1GA<6*P2iQ9}Mnd?$7!}7Rg#vmAI<w%RXgk%6w9aCNdsyX~#efTw8dnak=
zbW5L_!*xmPfVJS^5Z{WqQnuq+iUJ}E^_M5aw6m3CSs;Setl#uod*H_7=kcQx6|&8;
z(%SzddbpdE9Vzaz7=F}F@s<wkWV!0^(QKNS=l8bCroOS}#+x@jo_VQW&57ORXh7$Y
z-%FqZ8b8}@Xg&fXz=8Sl2#>Au1R3&=n>*~^My04f%oWZi1JyC$7%u|YDz-pTbU}X5
zU|Ci%TplwrNpZy@>t<g-qq{nSW~Uw}8*L43*^2>dO9`6i$$FSyx}xm`n4C7wz)3F7
z%vf%>Gzz%$lrotSXW3*rsLOe&Wp1gVZX%j?ZOPBB!uIoB2rBIfKHgeXoyNz^t<hR7
zwzz77^t^1(FA$-cyAph0II*HqhU0&v)rl>EJmJJ)N<0UB=(X_&V|(7?B=(>x@J2&@
z?ZehdBg2>7Y7#}w?KH9Hw;#(xXPC!|+ur1bqiiJu!ouohizZq>d8n#Wg-&R9JnE`G
zHas?iA5O29`NBxyt#4Tc?L4g4i+)~{C>9aHht}IyeLY#<AfsAV%jf-mJ3+4&RTiXh
zw^AuC_icIcWSJ&W^fFRuHJ0L7IXBrX^&T4kT|T%o_CQ9ZzCVmIs7zEw;tsIQKZyv6
zISB3&Bn6UJs%}GnAa@^e5C>=&8D=wV2FH4o<N7D0uY={joE*ClY_&CCo$1|<nsBID
zG(#^6T(gZb$5O&dU!l~;LI@AkqdOQJlOk$UTJjg@+4fU$p|$*}`IpcNdt%(QK{-Rb
z-6%u6oy_Oww#^KEBH~KGWm$rOnk2(B?s0Er6KF5^M}Cl3e%gkWnWEtk8d5h^^kXZT
z8Qv{CWqgbe^+_HeM&MrO2Jc{%R{a8PAaSSN9Nl&~sM=Qd@=M3AE0~<Tcl`M|ypbmo
zDoe4y_0xaAn&+dag(wCOH~C?|gbkP5=7hIJ&d!dE@E6Q9RTx2sF7VSkSQ*T^?Zwmf
zZtr}@7T1re(PK`GzdQ7j$1=qW@mYXhvDR*3#c{Sn?59>akA|HCv?)je%;b&Blom2b
z77fSE=yxEXWEVg>y;%^kkB&YMnnYhRVg&;q`ATP*NZO-)Aa=qQLsh{Ei5>$8BkF99
zpH4*2B+$=0Tj@V(ZKT*%2#^BZ?_6ZnO7yBc2O<Nqe=zo>@K98iTF>oHDDDOf5K1kI
zfcoMJ9fY?StBY%(5VA6dif7#ck)ZAz7M5S=;u$qWU~q<^SW5s|c|F<b9xw6B+p`pI
z)^-dtHl$9<1*T67x{y`yLKag_Q6>S6sXs-Pgi6dX)?7~io?HAk6cto)xDOM|VhJA1
z+$fS<_{zzWxXb#`U&>^lldocChB7yuYItgwHkIbMRlWO#4SEQ$vf2z0n+u>K2vt5Y
zk#DVjB;<cHWPR}?KCETygh;4KkDZ@H*%V|@%M1dJavPQ|2^g2%Ps(8L2uNpB3ay-U
zM~M-{I4N<h5OCl7SGX0aL14D)5at2#dT7_S^tL1@futyk;iz9#?}Pv^v)x&6bA~U9
zu|T7!Nm+Zbo=eku=^5C{0-@Mlb-P~nF^`R?T&__r@K(GCXYrv=>>DWKbATEr(RcJq
z_6wcn>VZjpW(4fv%QhQV*xk`5+JxW@k%j4|h-IN8v_M1|B3m%T-i{IGTtyHAg({-#
z)<}JdrT~JKL?lHZSy^Ud6nAy+iXuqLr`op4B0}@Do9bI<<d3=C5a<!+CfWQGx8L=>
zeaN=9-+YvL@o&#VYy=@apa^PuJul0nwy)cU;cmq*XQ2Gul|Dtd*GxVpvW=TA;9ID$
zg<=*!mXr&+>|MNMIbF^#$|<F(YbW`~=JRF)7XZ2po)BXJCog$~#tsHG_vzLTma_1V
z^(K-%kY?q6`Rw_VK)!aK!cacSJgx%NMdD@{B8sX^4cCGZJlG0HRM7LL^OHd2ljyl}
z*y2J(ZT7<Czy(DV9HR7F*by?oX~!@;r0r27y8B}erbyknkDqldpaHt`3!Qpg_F(Vl
z=4NI(mKon%ztt!<Qa|7nMG~ndd~8lQG2RDFcj6gN2od=7y5m?%sYPxvuej3_6<Hop
zXj1MKMjG`xW||w=aCf048dG@a?Y{o^2K+kw<gEpsI4GGoHfR965&xi}+1|v^q>9E1
zIK(Xl5y*7kp(^%qmd5Y#xxqZQp(Q9!!b6N?q9fuml(lQJm+Gx={<t6DO?^(`1e7G2
zF<H7kgsT0uZicR|eRENk2MY*kUI5^-*qyT<N+_%kVl+z`u|z%~!5^rSeM=K3ym={Z
z1&k3T?&0k?D%Eg%az8Zzm{Tk#QO2g=>sy+?ryu~EYI!Q4n0<9)3?Z$jK?1%`8s9r^
z-sKR<D%swm%z;0lm+Yp!|DDP;Yvg&AMxC9|`sw`o+8Ymw+x;5Vm9M};QtwX`?96~o
zdcxO(w^k@hB}}AVn<Fl>=_(b<h2Q=7yOx!F9I%yPKS4)qb<ds~^Ca$^RTF!T)TXSa
zK}9UD2B3jwS=_)&C{&xz={S?UbGs9mLO9U*PzCJDc;dl43#`)2G$fbL-yC5<4&yEL
zb-i%32Uo=S_bLUMT=h`BSADBqT}x%4`9@u5R#KmMIbjpPtRrb(Km=&oC8_;%kK3AU
za7=eh4IvWsi}&qq@N0x`C#lbMg(;yq`t~LrkyK}DbvH@m=bVh2;Y`own+@4v39>(0
z=L4IOIg_mUZzZGe0Oj=d>#l`9xU<eIu6F)iFQ8~rOwHjm0XOip{bnpozHMMLL(2es
z2(FUH>NO(X>Q8*L5<fRUNL0)zOqfG)YwC>+5QTI!2262Tg9*~kr09HZCRHHF^eshh
z9QA5(8t~#Qee@TrF(b;Kc0dXccUKTnFd0>rFZf;oAgKn}6=8VmTr*Nm_~8uj#f%B#
zEWGYfZY%s>=z>FT^8Y0kS^mq{V`uxHSLCxbW&aPvs_PeG)g-DqgaC*@ke*4y9fM1K
zrzO{*%&e{51R<@VToh4sY3}po3QQmxV<Fy%4hSOX#iG~a1N%nodx%UZb_8YB{C{$|
z_<PJL12^(F+p>IMUMPtgOq}hW7?8PxeG;=_vjlis#f3XvONW|U^37xSjGE<^j{3G0
zV~Ki46*XH2zHIFk&fOKG?c2r&zB{`Am<L{pR%W~N24BiGVR0QGk)e?UJzD(bO(ok@
zJ*vkYFx+oO$Z5j5Dvb-Nw7W=Uwd2X3xW=v)VjU(8*p&HY$GHMg+hg2}2zk2tLPt9=
zmC@E>;@RoPmDFoCSoF(j>Yq5zqO?%TFPP*FHb8C1E(Ka~H99~U$paWW!hNp|V8^KF
zH&Z2Z+6f;Wv&$y=bXr}!)ZUzwVk}OJXZycKjrb0RT@z!IBBn82PE7nsq8`2{A9U3&
zY)UR+QZuh=<Sca$yIe%JO|C9Ew``faWaKVu-nk<v41J78SdG#^pBRKm87~;k`-YTv
zj7*XCJgHo0)rYuRWceozjV<l)e+(l<<1>x?&Rw$E@-T<5PTiOf+iO~uLdQsGSt!Ss
zg5HDP`7#F$MJS_%2`QsvLW#Na7~>Gb4L;zam$1}^2WC6^rywj5!(fNgR+**{LVjVX
z?_OXC!m3Q?a~#jQPw>O9i>|MHG2Zl)E{?I{!L<5-0P(3n*H<*$y$53_y1L}ehBfCd
z@oc+BUx`Qw5N-<apfLzaH%u%4<e0o}x4hQm*4)@l8-U1mRGN=8SFARk*w)TjHa1LK
zrg`04Ju82I!^6ZaZfA{C6;%*dv`(Ov<px4x8q=<Sa%N7AgsUPX@Z9?l=L7^YA<a8m
zX-e10l6s`lAVh!A?Tr;^fgUIRtpiM^w(44IqQL)iC?8b@F1wl@`CR3BA}gAYc?hNV
z`jw-=K$-#iN~;ydmk?1>@Gw%mfyz|`YV{(Wf5Ov-5)`luD9H?Y4ui82H0(R*jK>+!
zeB4PI913xb7t$o~xQ%N84Nsl3JbR)KdX<NKU9PdvhFq)TVKgD!Th+;PRh}RKoNIex
zgCCu{L2{bCrz4c8UmP<^mv3jDw}QvnB$?|-3G#NS1r{Bv@6`dWLW7tM)&EwzoeKDb
z9iNm?Y3oB3y%dis|5IwCO4M2_!A>Sm|5h%3zSH8bkg4?_yL#|Ywd#$$Lvi;qr}A^P
z>_lp2vq#}U8!TO{IVYneq!P@kpmDGmu@CWMelnK8%Z*eqv2zcD2#YER5wffJa@6f@
zE%uQCi=(DoCAIw<8lMeph}FdK#ES_TWC1bjgNp9p1ppR{5nt~#u9<RqLFDG;6~YL0
zIx<M*4#s5^vvdb>d3{N3Bz6hCZLVEn5A`9{@iG+X%}c#3s=3iz6pkv;bZ)G6_|JsQ
ztW&g0nl<P<m7vcL+)Kp$Fl`+PBa#F`zd`330cP~iDwDe8KW0EQM1R^7{+3`1pse-*
zPDp^1)Hix_7&ZYfx_2jfhG8|>pWUEnO`=dbwBT)*_*xcQh$^N!V#+-l5s}Y2%*AV5
zYh0`i!eZwYn;i|Zka2QWuv=4uV1EYhwD5jS@cdKRV8Cc#;G%Xl+os<!)$GUqHd*_?
zlExF>3g&GuHu?L_VQ}|1#WNn|_wTPdqXz&88IE$d!0W&OalcU8mI{%v;dEh8HsP#t
z9atKGOrKQ>hDXD{r<1hgz_R^_V*W3n;o$slutw^Nd_KK}N82m8?no33mlt+zK0r51
zPd=kdoaBPLEKAV?k}gR;yHQBvSShrA2T|DsDs++UV$_f~Fq!7FIC`@Z5M@)=cakS(
zWR#Vxmr@ELKIF1X<6&zrwPq(%1kD<Y?|?1f{M#TL^_IC)R;T<;^7w;rE+QQi4dklN
zXV%YDPJlinHfiQ+dGlU^A?#N6KPIex5(uJ9mf<kOhQh{G!p2GB?#+Pw<M%K!BwoK!
zXm=5~#QYWs0sz4pKlx1)gxO7KmD_!UvLYvHYu-Q}VrG})>~=)g7cO=wcvOFSSyq;l
zediQ8>^ayiz_w|0^G#;HNFbrY;jNAY&>|0E90g9<=Zbl@r89fz#U9n+ySl&1y={e7
zr^n0^LXI$^cH@n%k~l(E5}Lk>>O(b?u;qR3AwBs`W?MxtadD&HH9}GaarpCKAKpev
zWjaZaqmd>H+YM%{lR8(gEJ-8?Uyx;WfsaX85}=qF2pIVJiv&AZ<7DPIf|M2{-oCv~
zX<>#iC&%~tNfR7NKdCf<N62+nz#ERhXU${Sh?N@eY=lRQb$&1~(9D5srA9KgFm_jf
zf{UBU^6z6A==}}RW!N7oz)~{q!#-t4xPG6{7amG{84Q>P0HJp~O7z*cLxddneuJC!
z$K^i4BL?S|(@U*AdEhT6i8Tq38rGw{TrmIwX#@VODW7lK$24YRiEQKY7=eEvF^QO^
z60n80E0UAJRZTY0I#_%{TqYI(hbilJ0=e3leSC6p(?1lXK;S<vth}K|SnqJ9h_fe!
zBE1Wd-jVzP+1mIicx1@B7Av;7FKGQ>lL`+P0>f>nG>wO#_(iPXIrRo13u*fvc2PrO
zgc1?oE-9)4w*>#T16M2-N@4++mx4;jdCmyoD%wMO;2)BLwc$L$77?gAjR0_U5(}4Z
zRExgJAHY9fw3#fRo>z~vrMH3@=?ar4b_jeCt*M6S>c4YDgT&__^y@=-gek`TV>#wR
zFr)UGYfQ1q=C<*wPM9Aj^J-U3ZkW)8J7<^45cR}(lrRd-vOf!RY^JvITmxlt^Y`j6
zW7qeQdnRimQ$z9B^;Q?{$nojWFpvk^PNMpVhLu+M<uwm?2QAqN+PA{FS*a%3F0NE9
z=~AGI0A#<YdDh{>m|ub~IM4CzVxlAv^M`@=pGgPa#|2eQwYT#4WxDp+w32f~c$y=`
z$8nRor0`rkc9B|+Drw^Qqs*rdOXmv^ef^HT0UI524~)*9fzL;bHzZ=3hk42v&LZ!(
zCIgJr!HyHaQJUC^f(Q1q4CgUa$S&qFZ2>6L=)e0hQiqmJoKgodS5xI>9VTB0LiE&i
zvFXBcv)~VJ1ue19Vb#0A)L=>^9sg-yWNm+043fmR^ANPd&K?mJCSU9-q00y$@KtGZ
zWUmpBsjtlFWCJ8yv$5{Eegfp{sDqC0wdA=bfI3?FkV>W~0B^*HDneybk31<Zra^X>
zv$erE=|pwOV&4N;pFU~MGOa8Kl5ulFK_Cdv&w4544Z^i0#Z9^5X{n<`_Lx+@X_5_u
zr8g=|Agmf>gxqX%Hk!G#zdr8PESCMdL(#JmfM7375(p+hgY~R@oQd%mB9XTbkV`?L
z!qw(d!nPsDH%4%el?3AdR!PuA5U@-?nBw&E%r(cB?tGKo3VRt0$&%F`W7DvUy@cqx
z`UJp)QwPoT+MR3@C;<hG1WDdR_drT<FO%L%j^PfnjcIf>wz-FEYn;}HnBt2pPB_T6
zoY{oDn+9!Pl^(+N^k@CMALB9=BLSvK=@UuBSG)HWq4O)n%l#8=qz4d}|MXiK)N9$E
zhw6#P<|oGTd%yMT!k6zlW8l$|mQ*i?>kFv9H)ON5ZE;YMj||(UrZzxqWn}y9Sg*L;
zalrM5@U$4W=@+sZ=0_k-8cu0N7iUrX@V3?Q1Na?yr}ba*_}7~KztCq)>`ec2F}qlE
z$7!P#(Wj?JNS>A#nfO%_$)gdMoAYJ!L8Dv8EQ7BX$=XmTo=%d}<>&qSk6?r&$q2JI
zBO!tgK*!FtPsed%C}ytrdvNK=*%kc_eN%>}sZ;ul?TuLobETPBl8iZbhwk)ux-QGr
zUuDjjZmIIvZx_$EZy^ZgFEOS{5QwM=`b=5vj`LFutQ^*tM;@!?*v##xd`$hezmkE8
zBO_F=j|byV)IA}=kulBdoY^P=4$1J^M~_qLwyl2r`vY_JNIb}Ba!~pVNetUA?$ZK6
zL5R8Gm|;wtbx1iaRtWAsJ9yztR_AU6ob-(g%e2%#UL>tk7!d5F5cCl&LAr}q;|EHh
zx%;X5oT-t!7emZb!yf|=Rc|x;+*sde-EUM?8K#mw^lVQ+PeMJ{nQY#B7jVM`qXMk<
zvFQ?KWY89sY4Hg$Uvp2_c`iu??Mby-rLn03%2nRjXf-AJPAnSWB^b-O&bo~da}l0V
z#57qlF&q?!%Ut8D;oeb{vjS}%iT=)%X@^I#U`k{=j0zbMtx8g6>{HplQ^PxWnAyP{
zwy0X_;B~oUv}p~K4jQ-S;|nEc9o;Umk7LO<>QfDc_;WN;l#7yG&U18QG*J(KR#-g7
zq}=EIwYls!2#cH{CCU&<q<h0k)AcX5TjXmBP(>YK-0!$oi;MS?-L}Z2?s2C?Hu{jG
zok<$1vkk6|>sOoL7$^ib!2KC{+o*hbxFq<tLwORJ6DRATV?9K=cYvdH7_`xVnFf7Y
zSg{lpI}uT}m^D6$saWXs3xG-?P=nVBfQC~aUjcF2mH^-?*8u;)p2Q6^P)2MGWG#}Q
zPD+((*(UfNSnupIlcx@Z)1E)OFCLDgHSb%xQ8&-<?A|O>XxTSA#$NPg{|Lp-)KqNe
zZe?}JPW3nFtQ8%<=ifuj3q@Iiuus#)(2U%M8QoHb?=jLB$Y)!>hZuYa)K^{fCzclP
zLQXK4f~37(_8}IGw+HKU1*x6po^8F%)|xxRJ$O2dL6;Qe7Yc!e$1p)U>c><~>R{t=
zZXZtJyPSr8KC~$C(`2%Ky`t)>4!OI{M)rQkB|Q3FJ%}o^TkRNOY#Av6M@lo(u<;1u
zX93sKbI`-5f!1i)P6e3n3eD=$2h^r8x?(W6L({GCl=S8|EVw;(LPaJ&pnkQ=+<ZbU
z+A!<yz$4vX&R6peCdQ;}&L4nnj<nz&iLR0bQzYwl$R>E~RoG3vTx%03?0Y;d^Jz(k
zF(w?biNe{}A~_I2kgwQibfDS5*0_K5+x1Q*R6oX5^x6q63|Z}i^2L!v6~WW4R{G>E
zwTP>kbQg>6n)nMX*#lyHcHizkMDn^Tftppw)FZq}D}S`#zyT4`r^Ou^tPj*HNP?x}
zFONAQ!A4U(k`PFtD(1gs>Y!pk9#Asqu((`xaQD21NFo*(42ojqE;@T^`@Ov@6raUj
zQx00#56X_)pQ&oW@0zQzdULuhA$!5R)1$n1gYl8HDeyz5Lor1K_%ZQbJv}U~Ndsd5
z`^|-*k|{Rwa|Ja~HT2;5uO-;|t_L0Q;bM!dkBY!m#-1R5QKA}RxqTd7Yf0rI-l5s*
zb*N(bf(*t{uUX?Q#|G;AS@wWpd4=!1tHJ}4cV)Wgfvy1K)dSEim$r-*&T>gdB<-BN
zkn7{Z<DnWd3zne(aB_=uW{|GhL#(rith@)jLv0PK?)BK(Q{)75We@Ux(6v5Z$ipPa
ze!w2+J>dG{41PTFG$_dUF}cY#NwX4`T-M31UY2;^l@A12b_)pQ1bPtAr4S80bJ{EH
z@RO6L0~_HAeT^+Ai}`OIwzndBN9i@w?_o`d@BxV@A;VFV=afrH=TPn(X6$mow8{mf
z)b^^KYJxKS(~y#=!xX2=_<ep1NO0^;jrciS*aoHuGg*od72aM$lWj8=?^Rl@2z`dI
ztj|GpvQ7Jf$&=>lRXtN;&XPz^iKOe?S7^SU+gB~@Qh6OZAw%RN_8!mZuBvyVa*C+@
z3k9p7ghc^!seGiiwIXY$1)+w15^9P6?G{tbB_fPqj5CNiWQQou1$H-g=TG!uM&h1O
zNKyj&>F{?-Rg5fH^hGUqh;RM{9mSX6X&;;bfMCr8;lN7VJo~Af63Fh4p&42iT%#Q7
zjCLM~$FvVh<p{atUvh~`=kN%RNa5m+mZz}L)dqbRB8Dwq^Uz`ir-s>AqU&ENJ$o9F
zm33gV4Rt`<@d?NAjcTO8h-PPq)Ab94tHP2pEy^}Gk<*r-Ys{R|?Iob9=FOTQ`vY2p
zsc>1d3u!DsqITRgMs3T82zeNh-hUk-Bi<hT4iYZI^~4!}@_5fqups*d=zAOgd^1+A
zBxH#a9wbWzgw-X#m66GerBj;F=!wk(!a!D|-zzLU>z^TEY@|mVWX`yFUOY+XtsDB0
zwkHibS5dL67tC;4J0w5Bpvpw8e<`0KzE*gAP|M#w3g?iSsgX4Jq38`(TY9<{9ovr<
zcc<jOsUnc|iF|lW*pMP@3_3K1!VkatH;x*L2LCALFs;>pq~l2!I|aUe6NspTR&i0G
zz{|Zpfh{ChUPa`pGG4?&L$3&9N73O4jM=CHU7H(v^qK01@K7itJp?ehc+3}M1?LCt
zbR@YJfBth7;8R8m83m{Sj06qB1E(Zv)fuDM$j#bw-^34za5=HnW2F!ES&tjM>C~SB
zFmk8*r=Z^49q}*0bT~5)sCXcD?4bn(A|Wxo`G>efrvM+^N=QS{xd#p8a9p;Kd)W(m
zc$=XmcID38xls8ki)S&$t}^yn0xA__iTpo!4HQzvmUzU1-<ycjUdj|1?!}_&we4Ak
ztb5Bs87L_I2IE97`BbYX12mIKeDq`D2pHugLUQ=fW_Aa(YUkH1YhnkP66b<rnYr?e
z!Crf}v`rT?L!WMD0*mXuQZ78w==i@_^M5M|_iSns3N@Qsc9UZiaEku4{O7a1A_e8;
zA1%LP+WfG<Q=&31&M9>GD5)gpI=@^PtF*>QnH3(Kv!Xv`a#eWr$|nMf)nicltz;JV
z*=T6?HM3XLB4^zQSjRp@h5jfEV)rWy!Ye8)FQrY<Q2+!E5z)yy=5dKRyftJJ0a^F~
zV}L&_jXonK;5R$0g2zyY&a%vu{m2aqA%gshU`ViKFk}Ap7vGFXCkg6Qoe{Oi#D;AM
zmK%jBW&L<Hkkr*wzk&Hf*_EwcbS(dPYfan!4Y-rQDhV`+Dm*g(&P7p%8U>j;X+Q2Q
zHaJEp2J|&>HmF#>NJRhln&qohR2CqOTR#((BM>|I-2`(#+02sGG~lC!n?!n0yoGjk
zhbL<N(HH8IZkkYRHCw|kc7ESnlswQDB6e(sE+OHml15X!`p|s+&XU8=ujqz6PYa-;
z=!vZN`Bh2fzVVi$Rzb*NIU1;V7||*qwdDH;3#6wnnnf2<yXcwVAC5eYKe7QU8}D;x
zdCu_i^34@f3X##urBN$5kW1%|@@_58-@-C3D0cfZUpt)&n!C;(ABmNqbi6}S*f`&9
zRiCY{7`@$MnaZszB42PJUWy19NahFe$NSYXk>prFrQ$!jJ+HNUi{pjN5_(L;(h_%*
z5re6)oRh7upU<g3fG<BRHvgr1IsYrB$j-*_zs3}cH8&i8kHX$F>ZkD)4F>yC^LF>7
z4*SwNDG6&^W51f0#z6=XE1}`Q>x*rcfRwA@V5?ds7y$st;N@(gbkDsGV=xf@X8)Ea
zr<>c0{*{0@{f{L>7rp`#Ge!isGl5Y;xPK94J`uKzdLVxx^%l*P6#H8hA4ZHjRpD6s
z_(X8|ZmLW{c=>3!ZZ)JF$v`}kgLBP;rNqLh3b@&!`JYH|_o1#hvD~j-9(tFjAnB~*
zmzs+lm>=FRkkZp^(iD;-s2NA=3L^SYVlqSU#2Hx!vro%NVvX?OjY7el(~CGIC~}I&
z{@Ts-Jnw5CPHRFYITEARr8de+uV&|5oW@h{;)z_3FWrwFp7y1SVm8n$lET|+)y8=f
z%wr#1Q->mGWw!4!hP{&csTl??x)XB@-2;LwrfDMW6+;W%X@^}Q^<Z|0RYFzX!muPH
zv{Lu^M3dx|XkGejT@h!+Q&<dwcti&_M8wGSTUktRqbT`QFmk!cKuU4;_Ic_B4!d@_
zn+TeA{2245a;J;vsX}V#X1;#WVN$uTRxNZ>Q0b=7N%x;PDpE?GFG6epg$|6raiWQ0
zp*fz~hAZ_xOfK85vC_IcOC@CXwtKFeUh?Ieg4fq2FaniI>pQ>J&XaIVW#Ed8jq_2r
zT{Fy%Nk(z@N~<e9g=^|@7Sed*_q@^`(#{`D=+{`h{5HT4nCZ$gOF{6Tu%tjs#?$o{
zI4^H2Nm1k5(*}#>r@6ho&Er9}NMS7GoYBX+5}i6{797A&qu~uOStjdwym7<)mOed*
zewBNiXBiXT)sCGoSyIHzStQ=aa@zMa@!SG9?{W5Y@m%nPQXqCuE}4ZgtdhBsth3>u
zp)ESvqJ-}FUc~f6z?YIL+TnQ|$~_hK^?&A%Y3$OMRtVQk#OZKLj0ISkVUp*JZ&aJ>
z!1Y1M!{$E{9rlsjt;K-0kL>nVsmJ9+j2uNRO)|NOg$33G=3*%DGFS1VI4Hh>o&3D&
z-LNJNHbqf}4=31)$L3bxCJsUziGU^^ocd=K&m71C5y9{tIilp?ubq%dAX8Q^5`{)~
zk{=o&Zgep37^Yc!2f%s%APW<R>#?8pLqZu1Z`EelTPa9txudBa{p7>hd~)DD$RuSb
zbGB?!fD}?PlfSR?riov=54`Yk8C>@}b3~R6se_Axx_nzZ4X!>B85u*%5RwG_1VWr-
z&w-8akKP{-UL74#n%54a`%1b-PX5sn2H-TTAI|bykB5x|CGN(*qJ0f7<W9)8bKsyu
zJE@;}g*Z)rZ#U7v_UKHrEmdxA*uqT8_WFHQkC3ak&Z5jS#P+iP%V|5(-99h~ubWZR
zKix+UqwbXhYLx*#le_drzBFSP#W^j#emSWO5B$zql_Ei$2VHT7leTa7j(6+C71wbP
z-$?%pI_(^3rocpEE4hF*`UDLQTwWkwLi6A$Q$~e_yaP)Fo@N;1t^PNM9>LXixDyIf
z-ow7{Ky3&3Gq-cEpXQN-g=g=N6IMf~mG4s`EkWUT4E=jnt|-eyLw6k#CL&O`fJDM*
zUKP%xYoc8&&d5NNC;(WOUoFfr*1_ze|KA0T*PwmTvRRcllEi9oGC7#VQ~RD#<CIKC
z)8~#D%B60Jp=FmpD=y_TtpQ|qQtnw!3<8x8G+@Ky9|I;d03-zWh&}*<SqZZjS2usT
z0MSjo7(hFrhf^p+4NCgi*VL@qBv)J?O<1;6NWZhpOS3BqQ*6zfhdPt2dody4(9FGg
zRxZeV8HN%IDM7QASkeSVk+#g`17^J~#zh;e>NiT6PkB}<JI$@b^@9DWwg2+@Ga8il
zcV9MUD?7k#!djNORmr3Ajd&s~-)|nKtQ4XvvBwH>&-MN^k|JPqJ3xx4h0G{^CB6Mk
zDT+JR@YSTkB@mERPJc5{i|S!Tg|lc1F@9aCD1)eo7~0dS^#_E<FXPd%H53ZoRxe{P
zfWur+>#L6Vk&I%5K+S0GY|s=KG&DpA?31={b~DZpJpq4(IRb*g<{=DtfE54f9Z}Uo
z$)-Zq;mk74KeK-?7K86yH0siD=LNmO)&_?4&ndraNj&r9<H;wH4UlR3=a$*t$MB`G
zZA5w;xZwthc2PeenGB46_;hYULq=(c>wS9uyU>DX?^jFM*3t-Pe>abG2$7)(79?wQ
z$HqSjLohF-tNw}`orumeQJ_t4k$4*DTT=IT|FRrOX5R>>tC?<jvsK#l`fRBT`yIki
zgZH#x_}gfK2TOT>HGFC4lxn2RzOdAzI-6WiZn%RpQCJ8Ka<zaq!wDKxFWGm+Gl~oa
zB>}gO>K~~_yKtQWPcHUbW)a>&fN|D87Dbc|5iqKW^eD)@wcDcfgl%z&bTj^C(7gP<
zK4uBzJlLqxe;j@rHTJh4?hb)5m|Gc`p8*8h&ieqSK>-5Jpny#=HXB^5F+48-8U9iE
zyr4(3omxkc<IqKOMfHuvu9Vi`Q}o>84zqU!toOcq@MEr@4x_F-{b`SsH_1^efsM#7
z4hV|q1PYU%V>>ZH3~h^jwnu%90$Z|fGR7OuBa}84gSew_*9A?nugm<|fBZoFZ#Gwd
zd;fF*u4pgxpA~;}O@I+Wjbp(>mP9nBP!+dygTl1-Exj4ImRZ#TQ5?sB<n>l&+)8rR
z&xhL{urVx~omz+Yqu@(GhLjKar|`mNRgjWFGJN2dU~meDUt*^cZF=ZYavdJuYJOuP
z7b-G8wWFkQ3u8Z1ZuLY6u2DTv2pQIhListMkT?hlwQvl)^3H9q6YET$oopV6U@4|u
zFv&6)A_wjDUVNd95^Ap>#_2n4Mnf?ajf?EaRj#hUkftFM4B#n5b~!Oz_K#LL6}yjD
zphqU(TPSZxRMefq@1$3C%U3*(=*d`*jj-wpSG<5pc`f88tM>wbyROehW2hjgm-uI@
zJx0BodbRY>`?}4T^T@Q<QN2!cutFn}NO;Q9BpcIDOm|2D%@iPuaWyAu96hqLvrCU+
zO52D6K=U2z@6cmqBxY?zjfA*cmHx(uRWo~Sp8o~C!H@^6x1La6c}#~7IY&rh@a)th
zHu<?_iOHV91<}oJ09Zf25E~aj_@|ydvNykq93Nr?q6>`@xG`zqbjd-<0%I5W!J5Be
zGuGQ*XXv|to1Y1q4;>-phQ)Sav3lu-<?6<tV@d+3{`jmcK0~-4fl6RimJbxDoNW9Z
z!{{M9eHp^O=torLTdwX_(jSQgx-OJT#(@x!c=dX(e@e>zt`UIi(V7?NW^1iy1mQ8(
zecOTyhH#zEMzd+Equ#!c+zU|NRjhRiQ1O+lwGgC$S~OAK#@<-6-BkcwuyS;)H$(6g
zIHvDPjgZ?R0^P?n)B`%1LQ1x?*8K`Jm0Iyj&T}UV*(Sk37F5D)nkn`O5@?8hdclbz
zO6$m{aZ+yLubZxuD{Sql)Um!@KV{5)!&YYppOC}g0f&7Lnv!4Uq%pj<<ewH=;Dk|5
zp@)0fHlh?W_SAI0kZbOTUO~owv`H;WE(A6stL@xX&s0wiTZhySRq|6_$K2YwU5gz~
zdU$4fal%wX*?WGV#ajN^3axu?x5xJ1g)36=>nriAz!=5x7?MVPISKH}HKR=W#N5#%
z8FVPMR>qe!7&H5Jc~kW#&o>~?x+fU@vz_V8=KX?SbUq9FAIg}4^Zz@A@&5<%<Y4@N
zL7q$uO#gGiXGu%fPMaOc`?ug@^kJYkZ6AUUT_~{L-vR3CPc#O!#TpR2UlzWh!9-G=
zd0o51oMcNgq9h~@InTG6GvP_ci8DQt?ujd0PS=VN!4>|hJ)~GgR%b*@MuN@=MF=EB
zhEiHVrvyP~;jC34y<}Rxb|gXGimY1>TTrtv2yw}PB~uYIYFaFyAv0EVC=n<LqN5dS
zBHpHk9T`)Bk~qr9!Zk+}L%F^VMHsmkX+)--L?DQn2#{#(<`9Uyp?m4?q#ToQuzoF;
z$&)SzgK9Lc0jQ32NhXYq=;?>4WWjJLIfXUoXKricM`CN#DpSrIuRCeHVM-I9xrUKp
zXvC1XRw$tiz?}Mr%1F6>m@Pz~wS_TDZkH{EpqMQsJI$mHHW39HF-C!d4U!|@j}^pC
zd0<bcIL?BqPm-1`7p(Ld+UT$@pRGi$cx$MHl`=DkSzeaET+V3$zO;yQgAZVC{?Y1s
z_EFhEYvqAdVtu$lI)4qNCOM#~9%>9H3Td;P6Q49-kwxMXsj)hK*ElPa4kOYApQ9dS
zg=3O@==zVtYwT3Q6z|5J>11k#ut9(^A2j4cvE>Qejie^g@+ee!6XT@Lp8>&>gbj@4
zvd!T-F<CaXfQfv~76cL-qss<SfFg);PP(K?m9^)5IwzJZ!)csM*4;tPb;{m0$9*!w
z6vuzKnkq}jvdKF6&os6u8$cvMSc%nu#AvUtV{2<;w_P0q3gm;&M(m%DwZ7c2eEpP2
z9+w8s$4?h#<U3vqc_$BR<JsA5d!4~#{WZMKaApAqH6~1f$9EA9RmV!M!IB~P=noU*
z3&|b!1F<`~9eZbwSL0ooKW8`W%A0AOSvfE}hSg`*|C)Zk|2cN+-k#N?sUw$qfwD0C
z&FJmG`sVt*th<^JuqzP6J0O$@n!{$G0gOLo>NM$EFQ-v5<2k|BJ9c|G^`P4E{jqep
zlDN0=wm;-NM+QF=(gVv6v!kof!Qf3-kBn0v>dUMet0CGShNl?QB*d2jwOVM}KyLeb
z@p-n{x2;cmx4YzZyi{5{MT5<|F1c&IoedJ8x%cQfy&JQ1lQpu%T&EB$f5U;O6Rx{N
zJX3njk^5Qq&}I~HZ-~4Ju316dO#sQes(-V7z<aaRFl`?b_xo3&Gz=``j3Nvk;7C+T
z-Yc>d6TDbk2Hz=UY9(;4SMwQDRc54^ut}g?70nycz@=E6au_MD8df>l2QV}nzOyvc
zqp`gY&tXJcT*?!5G>F;nGPGz4H`dydQ%V5sq8N~M6R}HA;Y~kW>JnLcn7cc>E~{bS
z{hMUu!Tf%E1^;+EU$S&*{G4FW(a)h%PfX9H*YoktRKhDm4TDlcPos1PB#dNz(~2gj
z19y1+qvcTu)Za&rLS?c80i~XMf!?U1R!c#WXVi^_yQL8!VjrWwtYA>07mZR!AI-t6
z@?{*NRhsyIhvJQHe4^F~Ixp^@C)#}qxM{>KY*$)3fIRUvZi@51J1E3t3cm&6Aj8U1
zlA#eJMKd`^t-%(Chf}ZRHYbE>6H}{$e++6LFt&fSWOocSaQNov*L_*v=p2+NM0f!#
zD8#$6F)B!U%ZP)(MSrX=%MXPIn+B8k#)<-p%|5R4u$Lhv8X?Akde53_ItMWfhj(Q)
z=1U;jhEim9vE8ZlyX94eR(t1H<lQDPOD|`K_Mp$;t^C%ra56A;c&`pJ>ge9>x80-L
zl_jfQId$Y)Q=2kTxi(Sk`>)-{+Y3xNw<Uopl73L%m6%ph5uCXT`dwk&6P|+8-71V9
zhUEV0ORAy{#U)@a)40i5cUNO+!yO-kOS*Ji5#=K?+u-(XJp1Dy3gT)pjkJ*eK#rkt
zrpnvk@t7}l`U`R0%5!TDJO*tw8=FKVlJSYoehd%R3ZIR!b#EKas^fXv8{d8d14k#+
zM4X)@dkcZIO+D0vrkQE}O=Bd!4$&FKm}ag*T=iJiRyjz}GnduWg2Nc$UAg#7)<gvf
z|1Oo9O14JAg<_A&x1yTCQq~hYgV`Oa<#FVhtVv~2g|B8jv=_Jx*k8g_G*dffwy{fH
z2%TiK^{kl<YMpRQe4Y#m*Kx@YBxGkqnd}0e8EBCLFD-G<GUyw=fOsGP)R_(y|8nY5
zf=TL@5MT154DC}^MG4}J{O~GcVQ$$Np)D*5aGUbEgt1R~!EC8986X_j*bKX=92qJ{
zW*!nP&I+y=8AtJi_1T{`4p@<#qrbU7+rV>0VjbMHx_iSJXJ}=$9PEJLs!#xZbyRY^
z4H?D}>973kXt;f8Hg$sKiD96v`K71y8zqI<_zR`je5hj76>jk&BJW0Ctk?`(9RVqY
z6S+q9-Syygbd&BbfK*o^=?#GPhyniod+)B42f6e%=r(o=cdyRiMrb47?KZ|XsMt(g
z*~0Y4fFruQoyIj#A#C1kIoLV?_I1B0%fZcyd@dv1MU2}=F^=5V#OrZ&6t4iLr{5EM
zW#HS_sJFnVqZ$V+2g19vpmLeG>P6Z=80m=qBNqEX_@lp4p$||^6VF{50^U~>tY$@8
zMJ9Wfk%RJOmHVX(x!*S!K6qZDL4yWY3pN`lkT%G?pWesxE4ApdYnfwH&i*w4&hD%~
zZzmUHQO5Tc??hlvan|}bh?~#3jJN)IxtdF8wZ$i0pKRt;Tzxu(PyR1x3A!2;d^V|f
zKftp5TQUD3E|`ANYyV+3bF%%fXSZlgZKn+`IG^85?}Zv3W#)a+mTu0oF4+v(`RmK9
zYhWX4HY?3X3CD)BT75qP6V885OxYnQqDlmJfUf`}g+mm0-tMUQ7V>s_*`4Z+Ds@Dt
zXQ#)ugPJtEsyzF|=`lz-%ry~nQ-2@TAD0%HwU=r~kmF0|J{%CmVcs?IvfonBnL=tJ
zL?4wQJW49pn&-32(I3Be4g$p6sWZ7h6y|0hUA1Ric&X3kmKR;6`RyyT7vA0649jv3
zQhvJLV3ZQ7BqyOjeUVcFSLHS<8#dTnCIpazo#eNja$w{tkV442(g!wg{+*3;TIUYV
zxqu@zFZycRIxkdDv=^F@12rv(8tt0j`ZLZHCV<UEdG-^tKlXKHqSpWM&<T=v%;~|D
zCy6D7Ln?{bOW;QddJlV(Bdmy{K_sFmx~M>U%8xi5pO!|j6m9UzVZu5Ft@67W3|1a1
zw&@siS@Bcy8Lm$gw6dZfNm9!i5K~=?B2SO_UYlPvS!>F4Tu%SEKlQ$r%qWsUlOO!D
z7Q}Baz}+MJS<D^hM0Su{fCO@1v*JmsaXV+U0S8X@-D-9<-4uk|<WlFHSg=3ykhAg`
zmaGa23IQkeyDbB!i=hAlnFRkUVW=IK_E=Y)ts2JjxAJCc^QrOFdd)A-wQ+C)B|tp)
z$$Dv>;Rz_|NrGN(qWK}R<SC2cNubs?#kE@d2gy&q@S1tCYdg(RQ4P6W4(;L3kCY&C
zc+XsMP4Dw?dK3<k3dtYeP2l#+mpzB!dztGi=L6?L73cS_0oLb))5UevZ4(+u395GM
zN19VeGOr45Hzw|#>8ys$qUrae_j>Coj7?!#U}e8?;`ML9Y`CIC3qVtzq^aDi%kxWw
z{-i%dQ|8B{Th1FfS+q~4xCqbimVD<GT((KT<mDu0qC^%mPFHNYtIe!fs?{K)4U4kD
ze4@v@H2Ox`A`8j+{`f|4%bIlDTNpjN)itu3a~D>P+X$Ao(SZPP`e*u!fci!FEMZ(@
zMIgx$tCGARRa9w81QhfdR9)=icDIVPvP*00vvbOIU+vW8o?t}U=RS60fx_%ES<}9n
z%5I4*$8fQp>hr40xEMs7^DOTb?8!NAS0$OS7C*kG)#QS`dP23hei<m?EkC>It>EIb
zGo|f6o0A=3+>vA{_OGTZ+!yeFH?Gw<TbAc(8i+A?Xz^`$`OoJ=kfcEAV5bZy4xD{r
z>nFsamGK3#XpsD<OW)9snjw9lV+=F@J(mCggNQ?K1b=|py{x((z2_ko1(+~F=g3x|
z8>$`$6pP~3y??aKL5#3G+->p}&6%+fkrq;~KmiDMJU&49F+2~=v^ATQPLu?W6iIZ+
zJLm?jCXwyzUhf{Ap0!_^S1)8@pYg<<-XRYf1)>Kh`s`}y+Dz>3zXQKvPyZovVr{bU
z)K0eY4Qy@?Jrz|Pd9A!QnrM;5+>a<WxF!HquD#NI$qT*vI*(pKZguFo4Ro^CJc7N)
z;=s+iSi7YN*+b+R&N=M3l@K|AOd|O0u4WBwdt|RYnQm~t28r@k^{rANzfQJ=ED%kb
zh8q43?I;6^36*p}j4cTIsbryJ5cI8eL6{5FHVi5=)^#11s(2kSB;_T7D4o``Ril@P
z$UdABTl{RUug`SqWxUVX&f(x3Pp<d0E^b7iWq&87TdJ{&u_3;#pqrXljE!716Vi6W
zy?OxQQ}%KHjky;s{7H28L5ww_DjZ(K5);Ac3cU?q|6Z!O<i!K;A&u;cqxC@arl|bq
z^!1uB84T_QzX7wwiv>04V!uA`HfP#<z`bsLF@_iqYNq2Z4hMs&N~>&;*Ur=u?M7*9
zpgFBJoxY2#iS6wAtwJ`0Aena|LIrixw4ITDL;Nzt=Jyp>8|jbgaD)BxZwUd83%T=v
zb#S#?f0?$koqg*dEijPZmX*mU{nX>#hrm6A$k@QZqX$N1El>-FG5h|&+D_s{1n`sS
z7;p-K<YL1#OMC-UNQ;E%T@8>O41_Em|8|(7z^w(PI^|I27$txY!x)_aw4si}^#5V(
zonmwi*I>c6ZQHipt8Lr1ZM#?7wz1l_ZQHgreKN_+zSue0w}0~GzpSeFDfprL1_Qal
zGpr^aV$Eu*gbK0!?PhH$vD#WI`s=2f_V7<@{f*!qH@pXOGj$a8pzVcWI!Zp&{xmXd
z!ede+!Q|rt6Vg$W4fRcjVR+tO`CAB0R`K==yRIHIM#h(erx&RUq9VYLMnZJTaPg&9
zK*X<YEA;^KZ@cY+;m0w^PtGJQTWDK}Myp_K1SK0I+nU5)hPNZY5Ya*GZM*O&WZEB`
zVo7=OF_Zps>X*C&IByiGH2^6IK(X6lukLDwl|bdn6=f&6mZ3to=wWw2sZ%B&@u4|r
zDNpnymO{g@@e)G0Xb=uk`#3_>uoI~wJ3NHAFCx@(ypJm&zs(ua;PEv1?b4OruVwRL
z!fya%`|{Yr#TkVPUo|m<ee#b$c1=4gH*ImC>1lZ4+8tqW@9~!x1d_iK_nuC2L{R!U
z`26ryB$_~F2*g*JiHV%9GyoNS)B7a;5tWCB2>$-jsPked=_VmG7;IENgoe+;<p{lN
z?synQHQJzLK`{EU=TCg3L6cqd11RNuDJvl{oFt%_x2s)S;NPgvZ<^;N&qJC0!jPd}
zNYFv(CNKJ#OYt<-{2*#HF;QQ3Z9?&Ze<#6AP&Vfk-$1*a<K&@ReZ+4&!C{sF@YOkH
zFyEe`Qw0ZR{f0sH+ci9Lcf7Fc+ESOa-Qxp5Uo@1^zCYmiY<8F=m9-z2t!^^|dUp`<
zpNvr^yO^!G;iIo;N4L{9Xy4S?AAQ8W?d0zryY56;Z^RDD=JY+nve0LcQ%u+BLbz-r
z6&Z4#pUO(v*AgWPz`J*!l}8x8Wk%~p{JARx(?BbtLQ_lV1G#5NfcIT)yO{~d?RV1U
zT<_X_#?P6oBsCb@zCpMs>TOHo{~_X!L+>*%IHh9kBOk+HlIpS`IEq{nU&5!Va@2*X
zqx$`dj>1YuL`&%I`S^F}kZFx#19vrFVo<;{=|UhOo<ceflS!>O!ZwwfM~Lb|y%drf
z?vD)7a(NE~1gbEeIj4X8)al$!_zKe?*67XY0;=Gq(w=<!paO<PH(bPJ(xX#Byf`QB
zk-}_?zz)BAa29_hJOF~#S)@N;MDKAS!a%!eWN{!GQUl}x;vthqp*f^{tSDdNES%5F
z^lY|_nyHul`}viec(u;ka<7;HFtpj;w+&)X_{u@=+b>gM-=Qag5_%rik)qLrCvv>~
zRF`j|v%iQ#l6Idt%`_;quNTf*+85-)pXyozbMyl|-hAXQCtIN=0bAmvQYv}M1`mL!
z9fO3xSBIH{Dc8;lDAMl9yB4vxY^6;uhi$2)9?|~2vZieJbq@4`=eD{@Ngnqw=Jspt
z=IC5V9d8ps=l*?Hs`eF9)tu@Q1J^bK)|UJFj+IgI5j5<HPBz<B=SB6{IdzF3Ot?(C
zRy08rAJ!QR`Ys-pK!ETKvz+bS=Er(PQXxypPs{Ltf3mL6m(yc7J@yj1TAAG=o+)#@
zF-|x1*pG-o2$JmUokbe?*Bj9@IPDKemQMt^Qg@#L`DBwUrl+gap@tE5!Z(3uj&wQG
zd}J!|rR3A1gDsaO_}8*GKkS;Tv`#zu#iUk&?pK!&{waIw3X0N#M|skRFp&u6uW1DL
z)dIA;_EpaCcFDDO`l5?HBcO9-0nDLuHNkDMVU>T{B?ATxt>Q-H93)n?uf>7r^-A2?
z>U`(Nr+P2`Z|s@*zb2VDIR97noUNhrf7r9nFMBTJSzy>tC2`H2XOUUhaAuNS@8E9v
z=v|Gj7ESRc<*fd=KqsKkkj&nklr<$nDxe7G0D#+d|CcqG_@L&{+TG>h@|okr%Q#6w
zl)kws<2;EpMxJ37P0ie<A^ja$Q$b|+&_-J#CNwf9VG*U1$iDFN8?E)NJ={H0{<jFR
zy1_pJ%6m(@Tp#H7IrS)h&yi66P4*|jB2?>f-FXGljFFuon%tWRSze+6riR^WamY54
zt>wsSRrBl=KDDKifsgvWq@Lyb5bu6I1lv!xDe`{5Cg+~U5gJK%{;0xoy>2pPZyA)9
zX3PA@nrgxP#Y%-&9c%}&G8{WGl+gQcrv~RIp0Xau)gDVs1_g0CS|H!96!3<mMZq~-
zt%kD(t@d18RkgUQLbI*L1{+Nm#Z85$%|<8HS+^SMjW;EFjjd+d$x*`fTsK*J&3zzA
zDM=lz%!ceteFCN#(X{wjx>kP&SCC%N8E%lAvhh8mq2f@{e)iN1t~cDNCjgFQ9nD00
zNcmb&5I*c4;h)G`@%_@hIedm*AyUkVU0?{jnve&3JiNL$%EAkeR$mykoOBS{5f3{~
z`k$G_+tIc5)wxf@uIV|r?_CA|Fl@BQvi1G+gU5=xH=fJ*RC5uWrmg1O`=_m!@j^?7
z(^HpzXATb_izeU30nQwY*}e5O@|Zp&&{O6u!kNvIijDg7ubdm}bara2o~g<L&1;{}
z_Ts<aTant>B@5S}5_OoIjh^d<!a<Zne-glgjyQcB*08S*Tva?$S+b=vP7nD;1krr5
z?=ifcme#pm(3IoDJ&;1pC#hHIV=8WJ@g~9K;wNbx<9Im|JbDVZYIPX3VW*F)L59bM
zLld$vX(b+<!vDxz#}I}10lz+Ovc_1A@*rT$y1=Jf)G@DQknW-tcF|vXCx5&GYxw~Q
zq-=ME9O`n?cTcAt<{z|#h4gbTAC+EEKUS@f^d}a~4Xv%M8Ih2M%Av3${J>feCb0Da
zA1LT}iaSHRIZ**9J^3_LumS3W?-9ZLaoXAoLDar|q92gVVxn~Jl{7ecsTp#w1_3!d
zXB%-h@9neSnH2aVOIVImj8N{?GyLH*eqZSw9DISwv6SM!&2yyy?BhP^K0G4nQ`OwW
zpX6)y>sbVCumpd>!*>Sq!h}8Pw}8K~nIag{ReRWM>X_cS*kmQl>w@HTZ=iI{9p^F|
z8i6l1&LNc^<s%}X>!_m!6HLX9prbLoBkCBH2GIRCO}Fv1zO<k$YMzk>?0ey=7D$@F
zBDi?Q@TV}#R;y9ikjRd=wY!O1EGM|(Vck~M8HJ<x@s;R7++TBYq#bs8*e(q5sGhSm
zZIs2730mT?QPZxOmTNwMB!!Xm<Cff6&f@kl=@Jh9J8e19lk;qZnPaf$b6;h{HiLa+
zB#O#RrHe;YeaOLc)1EGvBLgqM$FRm+-<ihZ(aIR(Mw-4pX<B^~bmKgX8mbWyea?hW
zPb1o&iItCW|4FAcka*}+y#ASuThIc{XLW1&El^dv-Bl%|{;vebt9CNzAT4%@nV((Q
z_E8Hr4uao-w4z1IxxhQ09iT2C((Y@j?KrWwm#~v3FU;waQf|M5qaw|g4eVI@??(eu
zc?*PkYgUB+(kOXVf#GDSLM)0zRt8}0+}u^>mXe|d#n@q_(SVemi0%C9f!A~;$C99|
zZ+YqhD?Zi42M2ixzk4Rz1DPS}`#Lv(`1sA==C%x`gZvOldvJ|q7kmcDiv~OT<NHrG
z+OP3RY$10Zr)zq=N`q1LN)eWtU!3xmoL(iX8Pa<@EW!~mhZwv&PnvEy28V(s)rWhx
zTCc@VhBf^D349kQVbE3TyTtF77f7+4+V>7e{0nF;IZ}(mx}1NKV#J9!B{dLETkUuh
zN7wq{_H)N0L{b+6V;7qdtcudGE?WG-ffP^>uk9V_W4zTvv$@p^T3{#Mr5I2w7gXAf
za2UlL8EO88<Z|4t<BMF~IClSe|Djfe(6{G-?lUxv50Mb0(Hw>pN2B)A1Gb>42T{aK
z5s>^K9ExzrM+(b;0w;fXYi@qPy>gy!_Ls^`eO@*~_~YRos7MYo4InKRW<abs{g1`I
z*hdE?Y(4BRGSznLzb2iJ^<9xTfy6-P$DEd3oti29u82j@zH-{*jpinj?nd!!`R(HS
zllmkKvZF%u@g&z?Wm-y=A3Pb5c?Vpe->x)h%x~$vLGWc&N@S|Ds;#PknQLnD@H`$U
z=R@ne0mUpRH!oe^b|8+%CDS8Iq)$YuDrixsvM$!eZ4uiQ83I3|Smftm(C2pp;Qjbm
z;9*#9J%ufS=#SYk@u6rKhZJ?@(<ZZJ^;DP1ceev$N~uNSWls+W<Z)i8d70mVz{);B
zXabT>>S`Nq`g`up-D}n*>;}t-J;}G1LB2Or-!zvobTz@j%5&TpAKEb1Xswrj=~e;c
ze5?DM!7rAO6fdMG2E<!68NW@(6d+$U@NFLs7+oKBApi1WFzr$-)(v`D;4vWglcRxe
z=E3%$pSGBN@&SGK($W}zoGWq?_j&Z$3MbP-BsW49J|^2so1s0c*h9zq5n}clGzCRc
ze$Sx5+C)i7?0A<}a?8TFn1}LEpJlOdH~=$wpHjl3ygDjA*n8QTCsPf{5}`KrIgxzL
z**6=9HCY|Pr))5WJ};+v$c14xTt^bsvbss!?o@#Cgw9iOT8e|_RB|*d%4;3_ge|`F
zZgofy2)#7c0g%|oaUGarGY|WC#G#@80IdlI>Wm8O4CUE+o#YV}_Tb@UU&_-%e`84T
zV(vghs054FkO)?4h9e{s1q~TO(lQIP>i*G9qQGY5Ga9~oWO}VlcKMg$_rFuq3_R76
z1kpN2FmgIc(E41r^sC2=M0ta~)^aiaCzbdV9C;X5m+nF|oQ)Ch{m!eOm=>PB<*0u^
zw_i?MC!c4<!Di)|o}Xgf^GnBre|4|5^+tv5U%IOrs!qGWdRi2;SX0Vp_bB3V8p8FU
z`kpc6rI(}L;zS?FAh{`V*M07Sq!D|~RMf4W8Xiu*)zv+qaJiMwB7uApb1p!IZzvim
z*i2dVMJUYBMdyDH-M*=FVV$imi|&bYPs0gH699_-i!<I=9UT;#2N-5P3EciL(;Dk;
zTw-T4TIF(%d)u^&gdwGZ#CJ$#vsk-t&N-t10lW~QsE>Rv#3Jq`(#MR!tb`rBMK~;8
zVLD-tBK1QUCGC$Idws5gq%nL9j`!E&G6}Epgy6UY<+JY_vE=8y+x|K8N9#?E6^S!h
z**9u4Ld3d1s<Zo6ITYq%DZp%fi91nid$p9j4d>jp#??c=;+!09Jdopxo7(_5$Fv)r
z8pz^J`6_B+CT<M;>f-%({><2xk!n(J*cNOXxLE>q!dYLV;i{URID}!28^%2xg|MI{
zx@J>9YIVVfMHbw9$0lq8|E@IK&Vi<kN&qoHM!Zfuz~w9GsSgw6#%E&Tt1E<|ojAU9
zcc{DF&p7xG!&VcUm)cp!UKqD!-qwO`VD(h1oy{{pQC8CJ(G<ru-$p^uKiSJdcQdgX
z-b}fn1yhEzTQw%K_8en7@I%CG+`6{euTKLeyxjoTUP&w7*LY%hHJXgPV@QGQ0NM(d
z?+MUzzh@_4j$!LJXJ4#~@LWpKNLNnF*MPqYd}U*eqXcHm1Hjw<wb6LI9Uf2oL~~Ro
z|BW`Ya{QM^XAZ{ynKn16|GxzQGqv(lcqzYb&p-fiBK9>!oDOSpQp&aJ?8^W&g9zlv
z#e^i}pN}iiAeQn`%o~HNAOY{fyZ0xn*TSh$e4Vf3j-99_yOSGEHR35}jNrnh1MxTM
ziTHiS1I53pzr?oI8%GZ4$m`*;PBx>8CLxO|1=f36rnzS1M|Uf$C$)eb@x=#?0`Ov0
zqTlTxVz-;T;M58sc>V+i+(pa6Pb}*VFzFK!KLVk-P#vbH5_htmBaK;IL+WNgjc`dx
zF~;MvxLMX?qDV3iMe2sGJfn(2XQxQK{GCrL)kE<h?4GXWh>v>Joe20Ew7b@Z@mAl-
zuzhDEeFA@t&H@IuRXj%~n-uiAy+?v6Z85N%AkRk^{qL}>ek5x&n!_n?3Mv}W4H_8@
zRz;^ITvPVE?^bg8m=zU2&%(_K7d4x5CmmKpxxTaH&P&Plc*_`{oPK$3ONlZ=0;5>~
z{=#;sS<plkCXU8DC|v$*W(4%qZumgC?YOW>JpoG#0pj{gMs<Fq$eXkcZ&qV=>Jb<7
z6tjR(A1JL%a(ad)o%x`e-rJYLv%YXp7-Q7l%Jtzi%LWtegX04X@f##dMOs97i+(f*
zAVw;r{U2!A8pJFv^Lof1{eFceOqU~2{YUr>uML)u_pLV%P1C9>wv$aWobTx$kQK;6
zNSx^xXP6HV_o2xb;@fb1;MoCH-4`G_-A#dtI7+i;QhhzsHz$Sfbb>*X)9PV;E+bd)
zmoLiSEeJ;NI5l&@KNzZd-l*|a-KYT5;_l70_q}GVY62%oTrxc-i?=3|K3C~BiOQH9
zVYjHS5(uk?=3j>ulFwT0DmVfiW!PR?_7V;}os5(k2%^pq7w`h`+ccFcht$h+dB;xi
zf){jFIcJ%l=uj)L?==bGdpyKc73GxFr!N&W3;l~vMc!GRGguW_{G@Mv&%WumAbB8W
zSg(-)E~A;&x!5C4V>Q%zDIL5RMRZXsU!wiNxXYd-7UcFL`RQ=5?=O`1$}})s_b7eT
z?y$N6@S6OWePLE;P*=AN+bMcJ{0wa-vc>-N*!CMYkM;VG(0MVx-TuvPUxdu4cUSFQ
zj;GG)mzlsiJH`gfub|=uQ0y%xtpAd;FBe=zY3q!$m&{6QfK<@}3a+MMRy)=-=Ko+f
z&Ycd`m?;CcdN8S6*+0zR86{dvu9&$}NJlpQgw7H-0+An1n;+*tU<Lo6YDz%co%`<m
z4QSF%zmPXziDwi+uIQqt?aeoKfapn~s-G>WW;HP5Cm4}ZS`1@%Xhi5`nX}(O*Jxvw
zN%oPF4X9#g8)m?U2d~_HwH3k>Sn&^8cR&P6$bBKi6qm}E{0Ut397Tzr`9q^ctOhT!
z$6*~`M4KHy0Wcu_kTOnpb|S1=)KOH?f-Cqojym_4s>e5dJmD6bQ`KfLizJ=Fx`s?(
z!bG1Cwpmmf)b3xl$K4+op|6W7d^rwis=l1!YQ1&xvG(-ny4H`AQ4`ezRnBFodD};{
zaFneD3O_Hqhyl@1eceBd$uT`l%S+FuVL%Xw^cM$L2!au)c#!CbWHD0Qr;ApbB?S;c
zL9bSyASKIMwpv_iS6%Fhz|p@Lx7sj@6^!N4%@}+Ajd)(T2dp54=6=b3p_BticDTqn
zkXl|<IWz+iqLf?R*QP#DTF}yH6v|CMsK`8h^2SxCCk7j6+CA^Dew5;59~#bhI@EQz
zKDF|p4pYV10b3G<fPoNA0zmeUJtFvs_rDTq!H+A;3PMK61?`!ViCC;ySda|j<JdoT
zb~kq9l|z35dn~T2oHM;|%9UCAX_-!Iy!Y&OI`ApJHxBbbU^P`|jO4j`5a@l3K(Pj+
zX8i6QzV;8W1Qc(qQNr%Y^Qs1yTIg!4uWnl!bl91EO6FrO9O4g6uu)CNlv3a_Yqhua
zUE-|gTrNN37#Z-mN3sBwrDg8}fQ)x(VZ`frs&x=t9B^^S@}VcroW{&o`5A=#dETy7
zN2DP7pN~Zl)-r#zGb4wm#-9KSb>&)jyMDaq1P3*6IQIg>^Md>U$-;ivZolzsv0d{&
zR$w=4l~1r|;Ps1+qv2R(sO1aM{E+_A*(k_oNoP-gpvNn;@myZV8)gv-QVNlHaqNO|
z&;Do@Lfb{1FP}ZarZkv(x$9pbO>_jX7MsS)7Yc~N6oA<Lc>ic87aa;Z<)DY*s%G85
zd-zTh5B;PpJ@r*h;BKQl9dJFPf{0J=P&_GAK24F641$VTPuOwxk}9fJ72f+WEE__l
zC=9yhBV%rQF)T)a$rQ|fbRbXm^!&K{wY7h{_c#}L$v#EBl`=-A4XF=qpJRE+3k&hX
zxiwsLynW5X@n2EJp$2bC1inGJp?bAE%tVn&ba$NXLp`(WeZ%FeQ_1FIaQjp-<>_FD
z?^cHJ#1Op)TT10@hPLxmwk)<|jnu1=#sS#V%YmhFtm{JvviWaHzy6B}ur-%mV*!E7
zv=?^V*GUVc>dx!XJP%l>=LA7`Rs+O!G!l;wlhT+1T<F3<Iq~A8Xhw7raM@qJhKo!_
zso!#qhl<4@09a|!6zUQG*P&XThz&gwvTMJe{E2~58EKqp_i1-_122lnKxfgg>ksXt
zyQ0!Za2H-2xs&twj1v$<21_i~D>Vupe|vV$I71?+KwkDNK-Rz<M!4@))nV}1v1w*%
zbZ?WuKKdHqw!A;?l7&V&^_<<Uw&9q%NU>-1XlAqIJy~4{n-ehW5@tusgCrfie_Bnd
zVW*&GaBDHn_K0fHmN|((VqYL;Ox$3roy)yIl1Q0%go#w-HW;_SwrcPWDd5heTyG6W
z`LE$tcM8$714M9)#5gp_1}^t5wg4J_Pqk=3Rkd-Sa%!0%RhGBWAGaS(O#As>b!`Tz
zUwthV?F(A*Qc!R>CDWo~M<#xG*b)LFFA9l?`H`gV0Nj+C_(W0O30RF)!xq+3;WLyX
znm!O49M<3LG}65d+6W)b9*)Bpa#3-$2)3J`;)Hane$q8;p*#E;i7+t$nOmo(-Hz5R
zncT%jcjlxZ{B9bS-v$sMTSzCZG&;c<VzQ(@;nVEnB5mw?OF$&M%cnB#X@duTpUF<|
zO*IALp`=-NfUqEOIb>f43~POy2p8CiO`@I-UgbcOd#Z?-Wv!cT!{Kod#OPg4x+caJ
zZv9<0LP<N_d;VWXZ5Z20mdKUXZtP<SKH%2fD(anM4}lGdM+({VT!`+Vrd??5>Fd^V
zcj!^(KcMrqrNtxer1HnWz%dcUOeHx<tn3(97f5J+<pB;C?-xQQGq*R1r<u#MK~Qj@
zLy6gE$4T8J*@cJI)Jgb44&Mct5rCFDDOkdNvyOjtHKhHk9h}xuYMnFPa~o}OK(XDS
zQEs$shK96dwn`o2@DeH}>a<;2-Zh&gU|Hqc25(x&Ty4sVNO<;<dx=I~YkR$A@y7L*
z4}1{j*So49ZPI(5E%elXjDEhRx7zUPn3iUZND}^yF?+-3)m|6FNpL;iPQf5aU^h9$
z1FHLWe=_Jx;@p|S+rBc0L;#Eij+2%)Ox7&QY#Qr$zUQW*OZ(lHxdMq(5E|h==@H18
zY-PMXN2HXszYLCgN!tj~?F`OQh^(`B%y1U%XpIa-G5O)lGS!{M<Ogw;Ox<s)r(bN_
z5*&}?(avP6PzZq0MTF^9f>hK$!wt2w%JrTF&<YGoFWb<%suJ#Q+<9Q@ex(1oAi9<Z
zG@hjjCj?=d!rf28hvS-2>-j?e@m@{+Z={;zzXhlmSpVk$RgC(71gIhjHX)63jYk|V
zaaW9OrleKKKK$kD24U$n6CC%y-fgtS6C*EM72yOCtXVqba`_N)Y|PwU-@ryL*50S?
zfe8&u;!~8TmtU9z;*XgHQHZo>mNVCO#$0^Nsgx$sm3ND};a8&Yy*3VutJ}ZFN}m}w
zmPERIyxH{_0u!aQ*1&SUy?nYpd)Lg73oz!(f$%@Uf`2nz&}eDBek7XupU}mM28Y%#
zx-To3f@%E_Go#TGx7q}tD=v^^_|l+c7&6mvVWzUmhLtT0gzk<jYhEmR5M#)Zd(1&z
z3k1<H!YByKo|a$227ZTn(AdqbR4-xE&K|fe&V-IDnVmKdme{u2nbTR3gQv(0B1!jW
zvR$Q1WNG!-{KN?l*=wmK?_mv)^x(*Z9@@+#6tG}?83pP&J2@|e!G7Z58V%X72U-wE
zJAB_AH5)BADHAf;h3#u}kfI=gjltRmvaBmfP`qOyZg<9xy{Hfyakw=Q1?r%s-&s}}
zX_nxOl;<sUFC=YjGE0AdvY_J+L!j$AzLGZR=(`!FAI4ASWU$bY*wuX|K3X6MBeq;h
z#^QkyTrh_nAjAzTzqW(&zvEL&qLynbys!xq)(~Wa?-ay~g`jG-KR((a?Yd{QtTeIf
zyo{c`a&2A?y|%CVPi^fo*sh%|;QS%Nr1j|gk|&$1Z4g^Was&2vp_hOpXiB#?IZvB(
za+(Yhc`|zHGkJpCv%$9X6Dc7Pq)XY#`9hOPoQFgH5Sh-ZRbV7jAwqQ+k|LGPq*CJ%
z2Ri*RStkzk|LsFb)z4|~d@i&^MtbDk)LRG2!^a21>D_IwnWUp*(90CFn<K?9g#)+1
z;llB@lm<h5m;{}enJkR)2<XW4G>XClcZLEMjYFep^}f@O6sGZlVtp5*aAp6T$^h?p
z?F#G>MlYG(gPt$4-2nB+pVd0doa@*l*1uiRJ}`E1`9{J6>)*vNC@<BO*+rnu9VeX=
zXTsg)ti=uDyzv``QVVZrG_%a$Lwq9n$YMzK5!Dk}!eQty!%+PDMDY~)SgyvR(c!*m
zawf@h6qi{Mi;~A9bDKl9YoM+L@B~l2>{Of}tdqe;(>8$*7ZL$4JgXG{xvlj&fnC*(
ziy$-()WAuk<Ol=|<`NDW(G(c*h%&(2D5@T4$O8r>ogMZMUO*q2fP*QM0Y!SG0*}LN
zH!cytB$BAyDCQdsMa^ZJH63J@wGC1S+LUK<V`Vp5Jcfgli7y<;Y{HF)$RD^NyAX#T
za%WpP|EF)O8Whc$u9a!jwZ}ZfpBRoK{Zf2he@qG96wxOuD%kzBT(7Z}vakV#xP)k@
ztsC6thTmjPV8`ihhwcRoM7G15;4I-5ltm{Cf}&MN4XyUl|5ng=&`p{mn0V`riCs%8
z`560hzdV{0T_%UNM&>9z67XN$^MHR9_Qut@jE(@xg^iqU)d|0Q$L5I)KYZW--)g}&
zcgS)mD4a0wFMAyTl@v+B@%II1&DqSLqrUe-kNseH;M>Mshfv2`LkcC56uwsM0^b#-
zzG#bsgcxGm9B!UUR7t`055<)cF6IRvsIQHd{L_@>>4kWx)73)exk$|HTfz6PJdALn
zfCKuM5@EM_yyG)}1v@B#7jU6of*ekXs8o~^QnVCD0(p#AR_R`f@>n%D!BBmVbQ#Dp
zejrYfMz)M!6f(Sd^isW~7W^VS4-<@4U=X7!27Oi_;buzsdXYkH4D{ZKs7=^lpGkLx
z+nmxyXJ~})%+~kux=F7>kl@JGDk__@J;wz-fx`jLlLLml(=*`(01ZN&@YM@?W65Oq
zH3~pI^Nq+KvW%!f9gq)igGbf_g#sr7RNffzq%b^aPD2+H0mT>QPtFp+iQC?SkjgN-
zC~8S5Z4gy#xFk6$Sj@MT7J3-n_SW%L3&Z(^b&=H9XP$?hje%!ptX^l(y`P|T>=zK{
z-=uk^nG!>!9#QF78=!n5P#1=JiF|<A-?FF>A;L123N-ncZVn1BRGvK)pd}>6?=>_L
z)Vv3_#yiw35-2x$LrNJ44y>?mh+Wjw`(6M>2m)ZfDaRd(E$>k?3@`oPqHx5On4nIn
zL~bjoR!Y2q1AeNVMQzE;V@&UR6a_rWi)CA3-3{?VZ8G#9UYM(#sGP;(1s3}v3*e9m
z3Sj*{%MRE~TdzBSM6)&z72#4WF2pQiGF{&6dygW=ReFM@D>cm(t%}^nA`qu&;;B=_
z!1r8GRU0ickK22*M{~DhldevbGy7XLI5@?wS<MARJc?5gB$xh^?DFP2``fw20Z0HR
zeq}r`H5ITvFfrkM1B={4KH`Bk!zI3ykDBM)L#1mp;r7V7>k6Rl^_36}n5kqrC-_BJ
zUY-<R`ZM>c>QzB>GyAyc?7R8pEBP>b*I{4co)E)@YEk};+kAHTDIf8&9`13hACct}
z+^|y=4g%|nrv<=(i919<Qy<B(haU8@(r6@5u`L#+(UXx&zn$|N8xB;4giqulL2)G1
zF)@^yl{-&xGcW{IzJqzv+~`<8wWZ^@wbx@<oW1N&e`0ZLQ*XmvEJKlE?g6u2q`T>`
z`L48|x=du)v?3Tr%!n&Ja3z;8rhm+T-auVdK^(?EG&$ZXH%f6Et8=o{Zr=6qdX>Pp
z@{bL3-S5abPSYaKec6X|?}vviq#!xRSFHuhsJ#J!2@%mr(CBk4W1^lie}MCj5GVf|
zr(j_FkB7SdYxIQa|BIe5axnkT?$IfY4O?t67@rq4nAy};|KsU46j6We3^sv4FjuHX
zm(RS_!0N>m5DD3AUmti2DTk{qR~kYKVE5Dv-U-66Ea7RgR>|>-T}#FIS6$`g9hEkJ
zK)Y(kp?|wcFHHN%GfYSTN%3Vu35bew^Ro8EWf;_=St^H2{Nq%lox5EsN19Fw)9Zu6
zx}2%J1B4+2^kCxYx$R}zkFlhIN%L?;ML3+86LI49v#C|leXoPp?w!X-+=sl6Ch04}
zdT6P;vDEvCVckTs9;cOXeD3f*Dx3Ez2DpQ!DTDP0NpAZZb)5|Hq%;aBK8SLVQT~km
zmWWo6kc#o_|L#fMClR5FXpCmHhbAC=SnI_Q`K#SA>p4_AqtiwsQjuyYj3S81jNUwn
z#Cbd%(t@U7b4FI9<1ocq|BxHa;tWjyL~tbK<b{yn1&5oARIK(XzQqKuYQ{tCJ+tb=
zRfM+$qgN1VET`g-L<rr-kwEzakQpH+{2>&SN>K?~V3a~pRIL}cB{ioTGlb&_S1ABM
zH%zAJ9aX2^$O<B&PERB*B3dRV$#u`6F0#U*go-~A_#H*0F%hA3(k0|kwoQ>GO#gEq
z7k)x3NJ`$F3?bo74N`vFNm_mgJQL7Mo`a;Q0*UAk@}eY;>O>$<$yE>9D$XMoVu~bH
zkB{sC5C^=<Xo$~d(yq&DJ(0vIKME}_L9(h=w;HFfBFOXn>^kY>g0V_h`JDWAw6aq-
z&{&NlTM0awtDTdJf5qebIkPLyWrUj9%b<=tX$fpr3T?Oy7OUZu&=SzK5-yItiaHOu
zX9qk|kS?^|+j_FzI!$Vodj0xh$fZ%y#cQJLFD+1bKMg&C6Yg@L2av)3aCX;cUJjS+
z@$ebKWGAL3)wF!~ykH#+KCmzs>jCCdIOf>*dHVW=+rY0WO@H!hT4rq+TfLYYbANUA
zEQa$K57#ctie;Fyk8SB=UTp>PW=faNzq-xXSi{?UwgR;9`ZYzj8pU^Z80E;GkQcT|
z8~>gBdtB@hil28$+<b+h8$jj^Cim8_5rrcbt6~i!UksP*0G@i2I%}aDL*A1~?&ZZ4
z<ng=pCP#kU``P)E@Pn64o7ynsHk>6r;>pF&I&FD*Xl+Rs#2`Kg51}#!+fN6xHMy+r
z0?Hj!uTMoTFk2IN=Cs5)@{?S?OP}4_QKpmuJ#MKI;aqe@14^9{=LD+`x@oxXD=0jh
z4&4@&;RS~x9CQA5G<m3VuErhfM)}N>-@mmn?gzYOZ33n0L~0GJ6-ZqU8y#5YVbl~r
zj7ELLW?xp+He(K~G}o+BPn$8}j@;Tp;v6Wvh<dmPCOX*0!4}{pB}Cp@`+t6%-yCfY
z_&66s=Md8w^=o=Z{8bc)pT|$cRpX-e>_RYF$*^IUC)1Sq#Da+-$e_#MUrKaXI&aU)
zo%V3aHu^d^%FfwA{Oshv*YR1!!mbrG5%-~?K9kF_cYCd{hsry#de>K91AFU+ETVws
z6NAM~WlkRfO>dz3vZ<8r@Q{R4LKgX3o7qSeojE$Tzz>ZkXld~>)v8La5K-!P%vR{l
zm43HC;8gb=uxpz=?mlG~C+YXMcIeHC72I9l?rWfzdl&T9)IjcD9a?XEHV+v$<^yJ(
zos!dy!!AZ0r8c=0jzt^k6%<?bjQyB6)hoU^*Us$ts?P|3j)0lPWXzI#F~F*gl+GBx
zV_<i}BSG47{e?nKz;vz3+U7hvialbUmg#icrg_SXEVk2kxsi#6iI;bGPXp83<ApwJ
zS6i}Fbec>|q+x&GE%t+<eMgW<&_EXZ-@>6LjuzgL9UKT8#wI&&g4e%U_IO<_wc~_#
z0?>;2P<s%j$gzf*IMy|f)hPFL$_?t1<XD@!2u~fLQI5iW?Xd4FflGj~qg(T1k~+J)
z&c(C8lWPZp*c>R=V3NeGEf6jNUhuHEyWhgX=&d*0is%?-ZkLCm*8N?q{$C^haQ4+)
zY;^Z(HB>{}wN!15DyWSKFBOs{nOF%6lIlLHR304_K#}yZT3zkmY03^_rG*W08_B4R
zwVx)#fo_IRrxDG^yG!Lm;PZn!r|g_fz2orcCr*q`Y*{hK&<}GaLBP-=l`|>mCtBXT
zj)HnWD===YZx8XLT-p067Zy>h&5DDFa0M?Z*5ezslo@yDugvXqpO}a#T6#TI#sB1-
zY$zbx`0_}c{f%HVI_*GSvJKkny3Eo({eLu0LA=7c7RuUk4?B9h$YbE_(Z}p&(#LTq
z9tIw-lnwU9M$87*99e!;KXiN_=D06$Kabp&*K=R*<LQaJCs+7KF6#MRZ!EH4Po%D{
zt-Quk-Pkm{{eBftH_V%3Un-{F^h$-=?232#{gLsPw0zVSpIv*&j$V&<yQ`=#-E!)k
zXUSLz7>(#Z&ZY~dy(Q0tG0i!9n@P*|w;-6UtvBct7fbH{g3iqUJ36!e|LDxl_CKH5
zmNYc%u-Os*&+?lQ(BSkvSpb1R=A)2NV^a_xibs5LQ)hT163y7JHH}-iS9>PG>(oAq
z&zSjG(aARpb6QfvT$O<mdZeiLhxb4gu#XctL=Y&YOH7E7loKh#f;7HPS}P{lNR^Nf
z%CwpnIFMOy*OV|lwN-;+*aMLbGp|TwoQaNg9GyUoYf%6#+A5YGG^ACv42EG>lo-7{
zz!8i@e0RTSfE7Fmw&ymFWrQ{4+)7(15OQx8K%cZo05h+0mP#J2Oh3p`8sqQ00HFb>
zCLm0(TNyqSy?dTM^oap}YZ&!oJeX0iARQJ6<lkn@)hZx<;K0USAFx)8koP^)1E@HI
z-4JLdn9kc%TMHhg5W(y|nxP=#YJEwN4FWz%5W2ip#XdHx^m(F7CjpN|&@?LKOZ(oT
z&^oTMBxnbh-XU<`X#muMU4lHGV6-$#yw82;62^G}$UNmd5&{Bs$iPN4JY)BP#<|fM
z?mV7*;LTG4wE#%2Q58^o2U+6cv}YdqjPwGC4;p!<OQ%VDXN*<&q!lv(2h&+Z0SQy(
zSJ2>o|73B@JO%)$Mb_oe9ZxV@QA_}tgCK2xf7)3h(A~yOV*dPeRG|PEh$Iy}Rs{cZ
z1a{EVMzpRd@Lo-jARy;;e```yc?@~bQ|Az<-~VBgk8ML>FNVrv8Px7W3nL0h5FqqM
z$y>09ZUYkM3}RH<wKf9<lJLl5v4je6+$Q9IT|nQf9u^KY6SaU%+Bdt|u5Venyn?Ts
zp?-cQ_G+Rm-LO>AyM?~X=<@S>K96by`(C3LQScrf!R3?m_NG)0`1uL_TX{nL;xL1S
zpV*34X5!ruqwKD=cW~2chrOcuTBi46POW6)SM05JH#6m=Vt3Y2dSszZcTlr2YK2vP
z+Z?-&w4C^A`|Rz3%oz+DHEOzGmM00@8OgQ1YlNO|mm5`iP(x))u%|T)+wf}3SoT-;
z=?w*Qs=SrfWz*kpnGqw#M#WGSxSsRN2^pU_-$+r4u8<#H`RAc^`QJl`a|3SO_Rk-c
zxwQw76Iz5+WecRNuvVU!kKF7kJ}qs_3HrO#2Ex9mAM#dj%g4Vg9-9m9vPQAhOdk7>
zQ1|;LEZT0SJGwUR6;pC3X%p^nTB^0!K@SnOQWI*0^9OflaZ$+Vd6Aa#N@eWD=0XOY
zdH%kaOB&blZ+H|CRGP&T4*B>7FI$jxZu$@^dJvh>%5#@$3aedb_4P$uE%R!LSv%3<
z#_V@G`mZh^7rF;e6Z!_I9@Z9GO*kvhx6&@(WP#Q)g#1_5xX6#yhVBrSnML(cczOyr
z%a21$XHBjSZ=wCK7Asjf=V;Sm9ouJH#$B{G+}w|m@9c-;7t(t2WV&rUySw4g+lU1^
zCBhgA4S2Hi`H@*(Tvr#Tv6?Do*#QGY61r`C)*{$HWa-lF?`aC0+S;Rh#Q2`lneuAv
zU{i93zE3y#`}|GJi5#`%5i)ZbFfP^bcVbb}Lgwf%2g&R{Q#&*ZfpOAvN&dAgXkipj
zNlWbrIwk_~p&!I98^(V~ZFI_-8;fZ7gIYh#Je$6m9cFb`8gnqlv^}>52pdv_-$^3+
zHjX(?ym1|;-;Jl^aZ78LP>Np0jwV{=pJGJF8pEVXY0HaBlz&(YoVtr~_t<(WIYE%I
zI){O~;8w))9XtY496500%Kx~qh`%UAaV@Zws7rLc2AB@Xlw?i4&nvcssa$vB7d_E1
zlUrnY30paIs9h_`XDRl@%Z_C|_!u$Zv13O2$qDKl<$FlLU$Su>`cB%2oe24fK%1Rh
zg*H}I8V|ONN{4H$1msH<OoM(ak&U`A$yo_HNb1Zp3cBCbg0(kzvZ%G1$T3-M*>Dd_
zpN-ZyX{tQUIKth_=f0(XzIxqwyV+-T^4tTQ#@^}SLTg)zYA+!5NbogYp4RD$Xr(jp
zt9kGg%cSFd{GGhI&1!#Qb2?PA;nrN|t~iMHHRC!S&*SEofPwqua<=i&qh?-c3hToo
zfb<GlWJ<4C_7Z(b+mK!vDNfJ@*Fs9=ZF}r!eex|E$RuqlFQViA+j1YoEtE*Dxg%=(
z_t9KcS#W=CKt~a-jRd=mFY}Mys^&L$RiQ6XIB#Q#?`A_u&bg2!H}+!V@s=@Jc1vRg
zth7n0@&zaFAhISIyakXo-7@~^V9E4M)bitmCsu(lQq>iXe`5*liB2x+a_OW|jApuN
zzTGNQvCw;olgrFfG3L^ebL{%dnj02AcjU{N#IPybN{Mrp|8mTC>~Vv{a6$m^T;wYi
z^EM6+T{N$x$@jw>qQoKPMO*A}#}NNl=muaKb!75etJdp-+sb2XXQ7Sw`pp+`G`afS
ze}PsO*8l9_{l5y||Aba{#{bp$J*Bbbgv^fMH=~y?mPR>t_S?dBD&IZcC;)!?7?L``
znP_Fi-IOix{c&A!pyE7#HZAMVN5IBq!SA=OGp1-UKt+~Zla6xRQ;s2~iP`5q6(1+<
z7#1Abha5_25^pc9(bH-=yc>Nqx%h;L5U4@zAj+t=CtHbGi&N8_;y@SI;u0+t;skdn
z8Ro=bPwP;bD1aG~WVnfubOm$Jyid>vh+g{Djz#ebkVQpCRjWXo20$n-K^A<M1XeMO
z+hial*)t7j^0kryxu|*`$(&E^1%hH<Kfap<1jT8^Ai9`EY!aa83tpy*sXfw{t`c)$
zA7P&Z6t4(TG>dU`BCkU(00!QJIrDb_ywE-yfKFDcXjcy06cAL~@-Y-`ce?#qj}U3o
z14@W6SWivZ8AUO{D2rs&e^y9<bei%Pd8k-{XYyGrFkdJ9Twtnzp0P0eY2hfR^+Cr~
z68d-~feV|S-vgW)YpCM@!)H|swHf5K!ajH9lpr@NLFKVHg-4?iTG78*`gh^vN5yY|
zul(=%KvA$_80WzRifP}W`pv6^=yoBDXwSjxSXf`2&?t^haDe0TY?^__Zvf0S3Gb+P
z=b@tJdugb7(XG-{aO{f#v;DP`p!b@W3w-@$wTQo)fVEdxuK|E@(S2osi@&lAWI2*9
zaGa2UYXN~|RXD&$#wg1L!CJS@uwH(frK5YhZc7_<!Gf1H=z+o1h|QE#1ok{w{HHSx
z-WZZWpb!lJvYV28vjNBN3&Y^o#<*SY4WE36WOM#`Wx6(BsZJQVP2Mz!{f)i$iCgaF
z|H!YXaO2S4o!zY)HV6`q8VHkAdHtcJT}5j8XwJ9jc|!`roYI*GpRsr9$;%O6@{-!;
zo10A#-4W_q({QS{Ss8Ju*hI#ej*y94L|TChBQ5V%I!A@0R_r1b1^roUyX}f;0mpYH
zG4N2jO(}6>eo&vHV>4{A6x{jcyL6iEEjES#8PxL*UP`0296SULw3t0O$s|e=Hx3*V
zh26xUfu*L#-<%e%K2Bz{p6n0I7VfySqA-~x@`OYW7z(sLX{1my63_3)kJj{ADc+@u
zJ6&X7ynieIlnYKMJz>2!N9f>p@<rb6jxW?60B67Nwx~BC6hJh{NsMwepG7-cCCb41
zaFct5;WB1*yC+wC1-@#V;^Z3MyUqKc9@rtP#q{4DsL(}SRgMm|5_J^sO?safzpIbN
z{n%3QB^7W<3a1K3a8v_4tOko9J#68-U1k=9q%Xem&_R(Vc2}>1So6uJsSP%xrzF6;
z!{zwR6WSy^ym9D(h_UqUl@?$b?j{9qA|a`bB68o|JX?5%Myakv-kFl=%_@>+)(?z0
z&2FL_E>wAl%E$_XKM;|e79^zKzovaUqFOb@aBnoyTw}*FRk@rc(d>8A%DfW-PBM#F
zFz0{sMYM%+WSiDyRo<99Oy(cU01rK2z~fe*lcDnZT7xrLaYSykg6o__sUyD>6GG-9
z3<gT(igjQ~7Nk2bSgEaZggK|Xf7>gav^45##ajJy0ubzxhxS19vJiGMOIK|}L64t!
z_A|Z~;>N9E%7OTD)kwndH@Cl|xdv96iB{)YN;K0ONEBYFaJjKeX)C3Hk`eJV=!&Kn
z;Qj3xN!QV)lS@n~j98)Hr|Znd&~hnd$Ux|ST3k?3){+Y{rl*7}Dv5%@+K_I;mIfLK
zxnb!Rl69DM$zobD(kY_p;?EixV--0w(nO7X8-YPGlPJA|=X}k>7}o~FwlJ`NWt5;^
zNw>Nyz)3NwI`KH}E%vv7b5QXjuR$s%+jG7&ZRl)59q^^ApaoyNb}=Xh$>_GhwH3Je
z^O;H%1y+`@lj^XnOX`;XHUAiVQg?HNIL?ac*GVYIivu=L`(P*Wu#dgR)(xM%<MOT0
z_W9r`<Un<g;+C|~o?om$zBm{&N=3KXKmu-S-X0~Vzd(H~f_;=fW)9G82Z1*8JahFR
zZPcVT$Q1WGT@oO3d}N^DP77<Ud-C^=u!q^d#4~T!j+xxDx5WhT_$T>FvK7PDHWF=p
zo_R!Xuyq356+UGB&gjr~sfE>T4ov34>m5U~h9$GylWt$TA=WP=YI+7eaFJ@6UJD0`
z)avt%f+sg=JA7x*vgoKP6^Sj~GfDMdYQ03)$+xMZ>Fmg-)uYB%tK|Y!=(NgEs+WVD
zFFZ6~S0wU8=_Y*)sC$~lEyv%0wuK{BK{G*_mB!2#Oz#?av>zRa^^;d$-;|fxGQbHf
z3<BGuc9&zQtnWN$)Y2mRL;h#otjLUUEeo!+Ceyc*n*gt`J$m~RZ+J{rea(!XtD6|-
zBI`2|*PoJ~J~y9|x-dQDZ%_b1*6RNPu51kdY5)Acf-CcX>em1HRe4BL!|69{@~c~C
ziIqW_+esq}B6s8DQK?Wk#-QmBo>ry*skx$qpRAN>En03eC%8TncnH`!Ki-vhEX`N3
zB%7cnh6)qk(VK0Cz>R^x^F~6}#xqDmg$0|m?~A*a6axt<PbzbefG|hq0urWwa22I9
zw`yY?yPP^qaz_Q5Wrax~6*&g^C^FD_pDIO5A-=O5j0f{UC$;qIaSg{H&X!W;1qMR7
zsO8H6nJ9RHuB)M96~V~&iJJ7W;HtXk;ZW63lNI$V(E9&F9i0UTI39&3gZ^?sX~hV$
zYBK7-uS09)2z!75ui}U;u;i%{7;0k`@B5ALLuXKhfM9}QP{e4{z)<|H)3la86$aZT
z2G#ZsF(;s|jf-lMRsrO{2!ML++oq&bz@jJuz=+!;QDRbujHD)w>tdEphvFOn@Q+&O
z-ZdnR4a6D;8|?D$f-CDQIUOf31JH`<9SS0cpcM#plcpYA1X|V1XMyG`V7+dvCRsQo
z5dcKD0yar=WdKM5y#ZVexIYye1^q<y4j=}lBW>aWmSpP1e*+x!=NbnnWl`PLT!k_U
z0^3a)?i-eZhKX)y1uF43#iIm8S-0wArM(1FM5C|~0OKeX5lc{26(GEXh<Fn80_gXQ
zp$2drQ8{H6SerFVN5<ong5Rc>M^6K)AV5brO9I_#+C`fN$yR}Y3;N|GWL5S8dpTOS
zvM0ET8O(a1umEH;K_UoJy0^2X0Ze5L%>o_{&!3!L?*Y7D4i0Vla`dZ(F3f9<QE!|r
zC;N<Ecl(!fVivHU4fuo{zP-oLY^-d|)+wvf^%iv=p;IHj!0Hw_Rmy>O+~SZIOWpGD
z6ODZAFc(Uv$xnAi6UNmOaalt=Lk}N9SJ&pT>J%_FVQ#Ncgca$Hx~8iXWQ#AkCU4Kw
z-?C+8eNEhypJ!_Z%FHL`fefWDZ=u^L3Sa$(kQ&<6PMy|>ZeIej$z$1Fu-{UUG8d$}
zh>s05LnS22pgCC2*m_cp|9b;cre~iz&+33l$}K#NUDM&h+g22_+z&?jcb%0=<R-yr
zOR;XF8UK_MIinbf?jmcPxpOd2a87m3@P+@|<9u_x;~<3oPNGD2x{taQH(ff>kt;6F
zFGBVsf4u_lQf5)YgR;TMY2Jbw@#itNB#1QTASpcGF*j1Y3jQ`8<A(2`j~2GUG|r=4
zct3Q%MK;a34E#e`)t%R|GCXzp48eN5Nl>Bb&#j(ln&<nIp+Pjb+$mgD`2ozDwi4%)
z{EJD~dQqb<awctU&cZgCY?a2|pRQ;u*QQ0D>7U;_%LKI64T_5COQ}Lr#`U*+s+)tQ
z6nxB8a4L(Hi-v7V#qm8JVMSF#to#e43>>CTN44JkPYrUZp^pF@FfkI`J+nw@{xw4k
z6c6^R(Fur6uyfjuB&T=6s3>-ZS38$iha(>BG_e8*p@aSP33*V3U9!JX*I!Ah2KuOg
z2{*HyDsK~WxXRk`??Nfmdm05$GoY8_qIl-gH0>wJD<L>ZkSDE**+^S2^tc_~#a;Gu
z0PKvJum^6x3!y7MURGrO)VbqrmzKr=p5&qKzlzHi=(%(L7b==w!FV%xm6XDx)sK9#
z7zRa?Mdxgxe-fv~Y-k2D1n-j9!r0u>XWYMuGCC+nZcwwgA}BU;MoG}S_6^Wm$(vjA
zU5qoOKJ#z7&9kP1t%SFnQDfH6XNxm*`&NrT{vB?~0lTqWB#ozJRB&f8OdRFW2R+)W
zP4XBqNn}rmgn{QxWz{Kp?p8r#GzBUC06*<|y#HILj$a?7Ep;Xcj}CKb`Ls1+emkGa
zxr<**d2zmcCg2RwK|1+#*V-_0K}Z*iwjff0$*1npt8(2-UGganiduwxpv}Y1LU=%&
zX;R8?cEe%TA#j~@#}=7F0yV<Ke1eU~#2Y6CZ{5&Iyn^Uk3s<az2<7YS#4+02qn|D|
zivaaWIW^gws)5OXh|xQvbw#EAfIhG2GL}*H^7+Khi33mWVJDBLf%Ci2EImk8uE`t}
zUhM9pZU}i0l$1r%?b`miSYYvi*PU0Zs-$?PiabSk2ZimJFiI^w*IlL{DwPe^&ishv
zoY2_{6OcV^aJ+f-5}FAhA431GbPlaCa{m_04P_5|%jRp_xuWS=3DMlk;IoMKGF_c+
zpW>}qDOW0kh*~2~*8MA2EEYO#ac{66c`LP#wd6(IzWr3<+^9?E<flvGIMiy-6QkSx
z*kmVKnY{Pqrv9>Z`Ic|8BSnza6aoJI(n$6pU747|)oTHG`1vQ-`TM|F!CNy*MeUj$
zziW|a@nOqhY!}5Oo$7~&;A0?15!G}5ZxiN)UQBK34=DQ-=k<TVRd&|@9O3@I;wt-p
z;3~)eToibuDP{M2t?fQhyJX~CUGta$5R4qQh{~hXB(e3p6G>Q!)SOVH9do(Akc!RI
za5%|d0=5Xm`<L%(#(_Wf7w5VvM4`EY+JL1pxcA3nsI7;Py5<vCMIv9GA$buERLf#m
zkin&xl#i*((KiJ#S?LlI;-6vD-Hh{)ddx#~c^U#1DR_b_!CJeaBxVwaI9#&20;AT!
zx@y3onYzxMCPNa9xMn}rGN5WNWwa3i=i)v<PnAQ~@9x^IoXSv>NH0S$c4!8!xdRf*
zMmrcJKmf?Xs%9Wz)*U%P08k&zc2J)XW0F4M4W!n9e#in1@UCV)-(z$$7->BiTtYvz
zrRnVgBtI|;M<X!4H~{Pm2QZ{J1XAsxCUYu5K5mLI3CsghaKE^A(1xUDj8!-mNtl(t
zr%K@nZLsnZ(!POUfMLP`FD#PiFq9JIH7ybxNmK@4Xo5HxbWpN0hio94jXf(wGt(qL
zC@lm833P&ET__0cG@++_g;gvwidfAq45^@ExInq2Q@|_1Ef!E9lD*{r!`L}B=fXr^
zKK6-i+fQuUwtZsTc5-6dwr$(CZO&VB@vo_xxtOc2?hnwbcdxyF3(C79KxI!VNgzrj
zOp{I^u?-?vTiOH~ySg9ozq9*#Fe%H!1Y*`~0Xs~AJvb|fQS6dADaKt!0uqDLgfYXv
zv=So&*aE=Ds58xQq+w10e0>Ny7+YXf0yF{Hh<FTG@o>aIu3G1}z!EgyyauJZg&?OI
z+K1SEa;+<;EE1Pk5DgLq?1a&=!Z6T~SI&A-m7GQB$TQ1nCP05kRXL}>IOdFaiV5##
z1mJtz6?y&id4DZkIV*E|YjWr_<{g<kxlb>$cq{dKydR!ml(2r%D?4iR50CK~09{jq
zrh%#4UD5ZBs*~8izjRE#=r?qOQ<`afeUmosdtu%iY94~43<G6?ZKr<pImy<C^)~#6
z#n7pC3$#d?&t3|vV8Q0@?eMR*W#zK&9l)<#a25{RVP3*rRgNjCz>}DFn|UTu->hTw
z^)b!e-R+B>geWsK<TBy2)FMw@Z)SkHm|>T=_eG1JI75GR=8Zw~!nmOKhIal_t{084
z*g^kdXhyCc98v3>`7<__A-Z<y{iX86dF0QO6M@NQU|m|dSk0qxkF<pb7a^aH5D%{t
z(gX2IGZ{a6N{0^ztx~JWO@k9gi{9;#J@xmnj=s!?`o7JzBsOFGpbRwDNVx2IA^Xlv
z4#mj3B(*`d0h{j)T7IcHBg7Jh9_C6@@^eq~X|_<gG=5wDzfpsX78>&vD5<A^FHXUi
z`7PgdniPGO*=8*BF@p`*piNALUCjP<!KT@jnJz+&SymKB1ay^-OrwxF;fC}KdkbIg
zQOn#i=i61CjM~sW^5UYnwgqeN%(qlAAbb_%Ms*LyR(}(_n+Sl;CB2p+MEKam-x)*@
zk~e4TK?wYLw)h7%NVs9nY&KIaEZ*(>=lrer41PoQvLOFUI<)KRaSI}qtiNC?IP4!I
zQ9ZB^l(O3kPchu>-lqyB+q$Ffgc%OWFOhmJxibllXv`j&t~acCzZYAwxZhkFeIa}h
zN8LKky_=}OSCZWMm0Iz_HnN?NTZ(9Y_^DS3esVD#jxkK5@aTCBZe975cdi$w&QpZ}
zg5epbM{bnok}E*(+7~Zobv=$xl!6qdpP~J2$aQNzhbV;a-)H(=tVQj&0AAyAFh-10
z`j_?DXTK)YYSu~*w2RFaFaFZYrb_9X52F+MN+<o>HIc)IQ+BaD(lo|pR3zn95BEG%
zRJ%;JchU4cZT}euYLvh@etBE8*$W-ORH~k9u-xGQ$S6yLxusIMq~si;wYi5k&|L<%
z7}xm>*tK3cv_iT52X1Y&>6CU=YQ*gZ(~X7T){@JQUSR7MtRlM6-8xg#PfA{92H9Q>
zSWekuPUob%S6Rl^veVN`$Es6B{)HYkSIzI&Kx=6RuxYYlcv*(#LSFDhvmJdiusN`)
zvFL5zD|7BH%mA4^VN2c(O=)}fCKosFyn)yK?PbpIAh;{I+1hKw-$RsD7RO@|**jnY
zut#n&FmlLMPT@fl$tJ!PB%eZH+yCg3XBKUA?;YG;u6gJ#0^>Y8yP|PP;P{!k3GGRf
z{iVqH_v<l3dzLzk_I;<og%{n+>HHa9aDD-siH{S{Ny2en%D|{%p3x_l%X7WF%El7u
zpxv|jAHCa-$RKRJ(1E9rPU&=V72+~&V}9<$D&5!Sn|rS5Q61=?uK7p$>ND2!(^cC~
zOWM0(r#Y7XO>+3oc<n&E=!QN<MC!HSz+yTnXH_k9lVN#F-tZR{Uin*{!_s2p+K-_#
z8^^@A<~Z+l-D}y=RaP!-!JsY@9v;U0d_B^%&1)@)zFdQ7-O`P_B(ySha^?0eL+ejd
z0hCDc&Y`ldTy7j;lwCUhFUZKk=GFg0Y&n_!&vxtoU$JFl{r?U*Ue^3iY}pZeKGpA1
z-s00H+5a<N@oh7R#bUUo$HOKXXknY10@WN3em-XA7^$ozZ!Qvwc!RMV#?#+VhKZ)>
znuC#TX~f(^ew_x|M1_z`u#h7qh7%Y?v_gW(G7iYqkf!fZjE5v*f(b_!NC-*GDF?r0
zK*`9ILRDli^EkjK@^}!nBtx_V6pL}S=_QIoGL;1^TS412E-9rkkI}~dLr#+`lJxsC
zm$e9d|G6Lq@-N#`7fmY~LkI?pCqx0!PXH&QDZrP5(Q6f`DFK;+jR6g(yI=wnG{U?~
z8Yrl8lnQK1?PUepWD%J+p^fIMK@Xt0pwcP=!BQ#+!-@ykSF?`IJt3hhl!bnlb0Vp<
zb4r4xAybKhDaSGkXf~C%h(N`z^<@dQb%!Dmu}&}sLWz=%G4_c{zp8>B)0+54IegKA
zPuw@6C7=RpO$IcSb)$*!X<=+Jv_DoX3sAMrAX8bmkWzBUOZ~A@P*A7F^ar8uaCAbZ
z6o5yyXNtRIEGo-v6j6-fHmDEu7mP$2Mx?F=Q#5dl2R1AZ7CBVGrYwwS7Ln{Drz{k(
zj6t<mMDAsVPjYbT8)i)EM^!*J6>zqi*w%}-8Wx23TAFe=fdj!Ez@vmu64h-Z)v{m0
zFHpXOc@>hdOcQ}=OM^jzmVhTwfDkElh0g=p_3!wv>oQGhA0EnCSu+V@-ZE{#1!#nk
z;?KameJQU$TVOs*xEyGWpcF6}B9cg{;=;6M9xb=$b7YoI{6B=}(~PVxirKK%pm)b6
z_wjWmkMEDOcd}KOpL6NEO-^`L_&@N{xMCc%_GxdmH>YH1sJ{ezjD!20J|A6w^)(nD
zGPf7<Q?Gq@KFQ>{clYftOSeCIvsXsx+NLeqvX%h;pAA-^J$Fha998yhk9mRc@H_6h
zVK}uu*RA|M+9Mylva@dWH{RwnA;p3^@y^Hh^v36n1%I3bolajFI%!)C&+D9TPQ|uo
zjvym<dhOj$hI{W?aNTXx1j%~eG!;1$6mLE`JlnLn##$0fIlLdwN}+3cpsRh5n}EO6
z5i8gR01U_4RbxL=p|brZhNpzdR@H$EY#Y_P$U+`_<%V4*n2@$f4$WbJ2iS>|MVrU(
z6VCqtB`?Uy^$Wx^)rwOtbG3Z$|AAalePfo}UL=3KbbxZ7xTn$l<si*w`rYmY=imA*
z@5xeF?l&Q|Q_OO>c0K|vG0t-x)WA>Keb1TUe@iTYQ|bj!oHAY<%>{6>S}7ytMfBI-
zU%j|?N4wA5Fm0cJag=O5QL8|F6RvB8f>^Au+Pv3TL_MJ0dY`ZG4+fYc08OYTyX#V4
z;_p^?r$wwzNEqu4GZAkegPU$yWrx!*E)NuKLS`R9Do)se4?UL#pTF25rqmjEsgExX
z4ky9~U;c)v76Lo&vQih(hYg*Po1b#(FC3!-`>UuIj>LUCqq<i`G#oo`>GlddLzz;l
zN+lH!VwX-B+6{%%y*tQ#2KqqOwO$hl>Zz%Y7Aa?)Y7SX!=TY)_ko#pcPsrt8oZqzf
zD~+{LE@nH0W+9JmS7FV#beM9*AgqP=s`@ce4{J5Uw(X^Akd^V&Xx63WJubQe|2p+-
zn#Kp!SZ?l4TrkxmluBEDipIZjMk_(dr<Yq4yb~|!u+wCGvWCE&o3FPwGgV8G-)LLr
zNPC{I>@#a7#nlyiMF#w)@@I>Dv-!ww57pBV0qWp|w;9h|$<Lc04!b>`5=cFwAA;uf
zfUA;lv80Q;rAAl*_;(Wrywb~6TNhKYZPnk_x?ulzgL)awgF-*jNbG@j0x^di>v^KI
zmL~qghSkrr&FGz#Pj$So(~+2prz#^QQ`4YR7DYzArwi-TcIKrg_Y524vg~sYB!tN|
zA9V2vVv2&Lj3d<x?+vT`8I`?ShlNt+2C9QuUjOKW(^|~}oN-e%#hkG%V)c7d`e$lP
z=s&GvmwNJLSf4gvi=k^XgQq2SJJZI$v4=wJ*DwF3bZ}22R#n8m8z&cwr+Ki=6tdk4
zt#D3>d|>?U^s=8|$TI0fTOHWfnneC+rFL-LuXQ`vHve`k6cM7a%Asb;hZMMDz!HRI
z=qDvgoa-aI^n5>bPIZ6Qrk7H3CT!`qSdW-FyghY_tf_c<KV+EyAhz1%t_|%o-@dd?
zpsosT^-m**d7vo!(9;n;4&<7M@AEl(KQVA9=luNNLdU|y{C^6ajH#`;vjr0&JK%r5
zW|;|Dm;lWGQ&9p48ULraWMN|E1pIGH6ox^}(%RY7k&r>m+R)il#MIc%#1w{)55~#a
z(bUie#(l%P1vJ|3`hje#C1$b3X0ye{?RR6v^1pSH@4ubg-yJWzAHc2~ApPn{a=AiF
zwR5G>P02(glG<6Q1b)Qq0R9Dti`-MpRtODJ-5dpg1gQ)VHHYLfgw$3Z@UzFH7gDA{
z0m`pW4@u^Q8|+;;j=7qG!ZJ}^HVtmF{9&L~20Z?fO08#dBRc+s*<D<`@w0ACW$9Mo
zUO`j+Qpt;+S(}+$7}#CD6J&IO_Du{$$3O6!LRafR#Xteih?SVRip{VE!=D8I`dSL1
z$z%AZXT#<vbw>m>76vsU^-U$sP2?aNSc;XI**!DJgWn;%A#Cp?_4lXG?dOIA@*{#O
zdHdEUrb2#tKO;FfJh`+rH9(-Ke&fcc#)c$5YyZ9f&Zp!kjIE?7yz7Hu2$=#f0OqZJ
zeZMsfwkGDLm+yARKjtldmSS&l7Qfk6$HcbBhGrEO;EeTO=C10EzkNzTB-%76L|(Ak
z0HKjz0jrl1vzZpP7qdusdNqGPNvYAHw_PSxX&Mw9F8~IHP|I)d)RpXSW#JnCVVc&u
z_LnrkOPR!c@%PUcca?>SDT2bY^VhD=uS@IiOZrb7gZWcj9eyxZ^iy=S%rEKB<5yhq
ztNyhL&<s4A8Mayp(B!Y3sG%Vwf{-1TnV6gqml~Uz(2O4(xtOci?l1rL(DM5F6#m_9
z+(oY9&nSEAhd2fRfB-;Adq#sq$w9n)s)7^XHQPCBDl_9oYQv2Blz44(X<KYZYc{K^
z|A7IoXPGJ_)8(nMzd><@ZqHoR$I7jqftwdLPJdG%#;oW7MZt`Pk!KsQB0kaq9=1($
zlP*1c01ISuo3!ia9!p4tm2eRg?ij(07t?xDFSd>5<caXrlNNSJ>-qew``iX<VHWCt
zV@W5SMy8Cvx!lL=spJ!z-l|9LHQE$VFjygi9;nchONNms0>obt)dm(qJ=ER{g1w2u
z<e2R0->@h{?TfZB6V=sJ+~Hszwvh=>-96JkuMc0I4P1L(-d3+?DRx{JT{NC~8<}4T
zogc&HkO!@dC~#8^m%ax}wy|;Ou5j$rZ0}mUOZ(kj@6=$W{O4)4$3d4WL>LPTylRvj
z4Lcr^P#t=mwIym7s+OKDS$lqh6mmV0rbKH@?bFLn|3UO<7LSE<g8-Q4J_r^_s|CzA
z|6H*wR1m<NLJUy(fn&8}irlM%9Xvt5Gib$~O>LWdXe7jr$0~*~$`-fL`$;mDJ0d;1
z+lpcCh3>{*UpfHZ`Pu8Mqq_;2!+>vkhMw<xdMB==#HfXeK4Ak_gwa0TkT$m2RhsoQ
z@<DF{g4e0Om_I47b3N97V`WImTnRm_8J|k6GH2PJX5R$;ff!T`5qRVOlIv;9o9o}D
z7T)&2DK493u=!{mJRqNspG*pg$lq3d-7#n`cipv14u>ukr~hkI_RXhwiHZIif|=RD
zBo-|4aH{T)oYy1BjEyoA+C&px42KhUp-)yj=`B=*2b%es(>O$umUU6Oc<ulD92sF?
zs8c}_H%c~^;8%n6lAYN&10uMQ{VEc7fu;U#86HG!&9UY<fDq#F;AM}7Z1ZvP_2J*!
zW%@SGEf)qD1+S>3TdXhLk^N1?{FWr){o_$D<Si|9)i&Fu2Bc2n6OpVY4BpCLWKW3-
znNM92vb3@udwXG|HbkdIPT`(7$pc>+hZY?HM}k0dw0EYKK@JaHdrS@Yo1)`|C_8zs
zL65GbE``_zR>2k^On1;>*Q1}ln12<+3*kkG){`>*I_zu}H$LXZM3?yMby`k3KD!jq
zbru;s;Vx%Dl359SbU?woHO;xKY?GIBT@apHU1K+Aola&axP13-LGWqg%p+>V!8ZVs
zkdMCwPpce_(N@C{NcS$2&me6^jyy+u@Qg>d2>5uadbpD#^^fT-oAmeg_-o`~!9QU2
zGQZ2-=tf)-_9F$o7l(q&CXoc=70DOecKG6P4V;^UZ$^TEaw6-If#VYgobPyv$-)Re
z%Uk~~hJ_)Llai$JiT5@z!^Dw@o8Mp_xTA_IoYf?CqYJriIfKRoHT1ff$Fk>2<Y1*x
z>|DmObWb}npk>f~q8v6%_e9sK)Dk5dILvE4o}rkjk&++@>T7Ln-$X2kGWtsnGr^0$
zp(3s|443Kh6VPQSrQ?NLY!^_!9P;A&sf4iq@imUPI&Kl#y#cAdnsBm%SGn0#$59rE
z#{)ek%UC)imI!{?(GTGGxipsEm31fCH55q_ws%2DP>D$v^6AwIO2I+@%!#2)UySiG
ztslqvT@KSmi!@Z(ZM%yRni;$q7TaJ&;D4w+!UF#D61cL-gqTv-cUrCA&;w^XvUl(<
z;EcTvUM&A?EW-Yc7s(IdCVVoCGJxYXg7Cnb8P`eXvhjXcF+}JY|9&O6E_Ekb^pqgN
z^RZ8SLwxUz{Tu4Q$nO0Kofj)kk)1lq;EsV|>^i&}D(yynAb-gwtXhy}gj0M~0Z(m0
zK^J?~)c_w>F^qlZnOsLrkpa$IVyXb@ah}GZT*jLfo-eEbmS59LDupyxyA4bLDcAJp
zW%=K>969@cfzh(9+k`Adi#kr}8r^OCVO=(gdT7qE-!vc33=ON`!4`u)x|9EWc6{Jx
za93qx9pqfB!OBp$%7iTt%*o3JDIyAPQi<Z#ny}VdmSGG{r=Jx)bkzRtcWWxu2(8Xe
zwSvS_f7FrMolU5v{%dT8-!isB<UkXC8VfpvjKLxiXu*oWU)9fskvsE7X_<3EfLZ3I
zjB(Q);ttWG1-l!bG+xOdEOu@uV@OG2qH@@LC%CUcW%c^+1}K~=e^QW%J-XOJrc+nC
zjADmP>-pa1X}!-WCA*oV(My)3Y!h9EMe3eP{Ze>WaMOiO1Z|upB@k>TZaBj<hZ|#N
zy`(2Ps>L=>A2Lw44lh;WSNo)ypMrr09q}~vp>sgA*wRI4K@j$Yp*2@ddjYy?RjkA&
z>^=nFpg)U`;b|s@CD%LTWVE0q?yt-Ck0A&#AR$|KSu8-@*ilHU@|9&0M`D69HgEXQ
zlG-&69yZTBZy!jy!ovN$naqSeWSQ|4GqH`jttfM<zZN8LH%uENQ7sFCqn69_pBi)T
ziJe4pGD<H>-fb`A@YQptff_~iQ3Z8O$#8=mKx^U}L0cF7^Az>r#_RmaSI#dol8nXY
zE6HT$z@KYy1df$}duF#iX|6B@0S#qG+qZ-N*X1wwj}H`qT;k*vb6y?A4NXDyfZM8h
zs2PQ_?!ZpvyyJELh9;TEc13><-CNu~ak5MM^3a^dv_^M)qmw~I=*uK%w8F<sdMb0v
zZnicy&6^>5*ENTk27;vzqGhYGR%`e9gy0I|fP)<S1DN$@&6NvpGL=qvMUBR?Ly=H#
z)3SD8E{#B3giE4yw9CzHurZ@gHlp%JhetTdzUJa~s+N~6CLnU8wnjWM7Mr$DD9u^)
z$o`J*#)a5pHs*|m*e5BzqZs3}Ule=gRAptx)llH&>*Q+5*R(*ZxS0zPms(M700@T?
zLXm4RK}1Cmi^9(<u^=klX8{v!bKx>Ni&%00wA4^3L&rw09x-S$f!&im!1su?t24Ne
z(YvAl!zUUCzt)0rJK;YBI1`%%uV-NSwsK}pz=Xlr{BRu+hla)G#JC(Ijsnj5-M<mO
zCOA@Sfs*OkU_2>3&W}<6x#tjc6e!5G1Y6!oo9#@P(Zq21q#G;o?%3mlpS_ugie>y9
zVORa&>QmowI9$ka&m;VKBJ6IV8*&lV%KSC^Z8%|&!1LW*|4SLiS{@YU+M!vTZ*74U
z!M;hPDt&wcZqxR~H5#<}*wRrNYl^vOU-K3KVs<_UQi8WxizEn4%kxyMNZMFMz1*I<
zBrNTB{}!p#zg@?*qgcFmAxXnM=ybpk=gON%gDK~a3L(C5;-LvxI@(8Hl9H-epjJTL
z$Kjdyr)DOVvDM}A6oU=tJ>t#zOSBi-f%4z$lJ2|xyFwN%F|=60^f6G$cnBPQxujl#
z=_NRpNsw9q5-loHQ;d-T|1c4IHtK40;TjC3Kr5C^V<**XoFEH)2maeb0pE_wYXD0F
zR1t)D>3V+whMsD=$L>4`xbLyWAfINHp50#<GOev{f8*Q#_~3~*MOy!HvlD)o-ReM{
z*3*oyGnw~dbQeeWwxcwBak4(Ze)(XyJl(I^dJ3o|JUU7X`K;}t_+7+-#piA&mMj1o
z@&(xso}&`^y?MmvWwZbA`*q~v=lTDFL{gpeR`0j|JSt0{GUh{j57~QXarOHNT%z(&
zgg3~cWy%d4nM7P?c5b^RN1rmsoKvoHwvZ4{VfF|_mf52P#?B_pZH**cj^i+5F=!s`
z<Zt+~ZfWaS9}KW=h=JFKlB$0${n9n>vsTJs*++MzRXtRrXbe2Oz!t>e_>a1!RCkm*
z<S9ypd;)3<J)>g-D>$75PJY#aXRW1o1xAy@+<0hd$@421D(-HTXBRN-C?4k^PB=*w
z-<&%AD~goSZ!3KqgN_Z8K1YxrQ6}c$%$JM~)nvwKH0@sMRy}f{>!*~5byR1%N(kQZ
z1VFLgvyq<MH-OVjHyA4Hg5MK#Fb;~M1Q&GC(jF$A!lCIx#?Z=kX3%{_7pJ#+6~D$K
znyMqpMbxgYwL-vdE+-{XEbyJy(xN(MBHe`puzFbp{t~u^e%@ZF8!$Jt&)q%_@J)c!
zff;TXsj5F#?`zj?lxPmhl6l{6EeMloe$^nNr*b0AIgL228-(+$Nt#Ljg0~Z~y%;%Y
zcE_Vn2BldrBx}rm1-v6Km06=@*bl`Kb9bREX-pgE_82;-`>hFJtBj?fYw|-x*>B6u
zJ}7tX$-ry;_x50QPBkayU-&))^Kh?l+OZB1uU%ySEgJ+*#g?Q_oI!SFzLSUUV7f6$
zS>Cu=&w_AeihIx9Z#wH)U~ZW?-i-y05c0JoXxMl?+r(tiJZw@nJC&?eTA)f8hC*Io
zWN;WXfz_fZ7y@pkid?8J==|zt?7Xs<)}G?{iHS(@=IAc^?$;y6(w94O-47oXxDA1L
ze=8T-J>H$J9M3)<mqc4GCM@9fhLLj)NmF*g(XlY0(F>&z<;4Tnfo$DO6PV^9w!468
z3)&l?*BJO#7n@dB`#X&@>v-!c<E3rYYj2#WP252mpku|<%eRk)>yioIGTQf$6>>~s
zUEDaEnYw%2ACJ9lejCU%qktUQFv1a)%n#Bii70VJJo=;Q%Pf}dJ-m%E^1c1F&9SfZ
z)$|DPrsr`pqI+3lja%AH@~gCD!QdPt!6?kFUEC)O-ebcU!suw@A>7atR?y+V_CF4H
zrP2S7L8$+t*lKqdS6;yucU`sW5^f8^`eigK7-=C7{7uYgCb{oBSYJ+y5+=<rNv3#i
zGo5?o>{na!z@)ywq80av((aQqve$Z_A-3J%W1&YrenoFQLVftg;}|)p2;NUmI(Qib
zCmJHeyzYy4#MQX$74r#Sk=3B^^_hwbfwDhF1btZBpDgy$oO!pT4b6p)1K|0R!!nj$
zeEQS!Shi6O5Vz7CcD#)#bP=>cGIh)&7A=KUAyeiaHU?YO$AlPg-4ng142RFoX{qy@
zu01Ta>j8_MExeP?SfxS1^yUes*?_63J6<Dkmb!X@w-AT#bvC3ET^weB27iHNi%{l-
z%Fd${`?C)`NK~7x9250pUlFYmO=IlojWOs&gHP%vBU;LW9I>}fkuiDkq)~Z??jgb7
zS^_@PgPrk&$@93Gy5mwuX0uP?(({kw^TsxH>hby5SmMPcn_;dO-~64@G+VfxMHZRe
ztw;95YV@@O{Spx`{aC)3XQ@s+i~QUE0TW+f%PY(aQOv>K%M}YkE^u8jozcfn(toAw
zoI*LsziV_l9%KQvjmn8_NL)-lFjygK2@B743vzaM(`GGTH@MZNTRmcg$o|(|?l*?2
z7mO)BN{*8RXG=6hgo5SfiJSB|;hgV_#%T@223LcI(J{CYJvg0y{KenZ`&+c*ca!ID
ze0=lmG{4tC)WMqGYAeNx!Z}58hPQ1MatL^C7JOJ1?<@~Qo;>zZiu<!nJ7{wo(_Ys^
z=O)Aesj`y(3pnez|0RZ20%xSAOie}8SmFsOrQahoBnb@p`8p8x!eIK7hpBu16S)xi
z2Gl^p<H?|FvLa9I^3e3``teJDqIym<Qljnwh?MethKE7(=gE7Ql@urUH)2}<e>4|a
zs7X9`CjMhg3))gBgm4g%NM>Sp#rFfkou8K8Wr)&r+OvWUGw=ru!}d@aqPK?!w2WYj
z>vKvjrfpgagko02NYjuZowpeI>o3=`e&2u$Czj9u!Z~Yq079BgSM^p_SC!+jb><I2
z@M3|9v6F1e6HVT=x!m_{ne6<-P17)$DRr3(unlIQtzBxEGL=6bRFtpi3<;yQx3WOM
zE#MfHl#%!aFb`1i+|XNV+Xp2D)hPwa$&NA*#7*vE(l?K58lC?&Cz@ToZE<e&UrZ5!
zdBH9~OwIPM=_yF+bBCZ?KaZ<3M5&q&*y0{tsPWiTuM|%G7Hc5A=wQMXbx>Dq1Q=a|
zeBTS_CnHv1t+zW@%T~5bRkldZO*FH)Y0R=Psunac;1Lv1ubiysnHwlKUg3D`w}9Jd
z%)zu4!0(0Fk7ptZ)Zkeg#E=YY#SWrXV?r+Sw~o&_QiVE-s%#ME_`oIQ$gE;wC;H3W
zPDUBg=`TO!-R3N%{0unhn}59xWbU77Sa&T`z}WhYG*7w--mXUKI78h7u}wlTq?bXv
zO`%agfz7MwKuc2c>BysOHpoUpOBzsv&Jn>{t2Kr`t1z=iue`jR4ecr9;_shl=&$_!
z){HjNr08JX@fb<dmRJJ`<uBx-5Ql`}UtQ4z;>y>w<wUvlz|I4TRiTT@_%5bfsq{o0
z7Roem_hj*!zN7*vkbsJ+2r+s2=tpp#f)U=n^q~T^d8<(43m=8v+zW$-Ax(i!l-LDv
zHhPHcu2S&rWoL`wU8tMEv?GZ%y`d%zYD6yyW@j|rMyeq1P!Kja*ebi%D%yLYut02X
zu^r@LHX>)oC*Uw(snO4@u$WnA0j@fL#k29do{Q4x9Q<DUHN-m>`qWk2AF=*Wt|9e_
z(1vZp!u&J%Mv2yPQ?1n_=&-|A7=8L7Grr(TNn%{mjxBd|<E_K}97G(ef^}F6m3208
zUIT1T^4HA#UF=@$7sm%IS+QVfS(8X}Lu2yVkb1#s(^=x`iro0@<92ew1d1Hlf2AJ;
zwj{r9Ie2N7MOojvTw=}akYk|?fn2TrHHq4*H%p};EzG@#n<HvRtMO(_9P?uiKucu1
zG(Y>Ya^M7|UnMJoX)fnHS7#Z}kf>j!Xh>dL1qvGXq9!v(Th8*03Nmf%%voi`*veAQ
z7SY<QcW<DGFrizoI+Qz^UYF6MicL-EIMc9Pz6(k*<M;1tFiS318}g=5kEoU4A1S|@
zb0!46{Y(_&ny#26cz6~XO2ny-JMwa2`^Q$N1MaYeTJ_9Qe|-UAOfM0VqvX2Ne<AM;
z!rh$l8NokQ63BT8)AgrLKIQZ@W0})`%?+I!sSxXQeK;OX1uDzhzu<cOfg$6>?zcRC
zB}Dxz(V7FtTpVTk%c!Qqfqat`2`CVdx~0!Kj<4(&6+LHe2!m1r>uAUu`R9|SQf-h(
z$@i{ZAjb7$lOWLiXgz0%09>;$bRG;`4^BooIto9?l^Au}fsX3M4r$@hG$Z+Z)TOHu
zv`?x%0m7_XWSy}QywV?t6zRAca3moW6Zq&c102Y47$M7~VPxyzli~sYfLYzTqWV#2
zvi;p_$K4nq<Tb8geOvJ+4H?h?n(9`Phs`oNK#({NCXp7(JX~DW$gC~_>}6#4i(ezu
zj`<@yR?b+R_t0|(hv0l2IsC$R2)$cPMCX1%JG4QdLRX`F$)nyMRu1_-i5>{fDUB<(
z(*NdBzW_F?_}zdlO144uK`PB|m8om31{21ePHsiVl4sl6Awn@t^=dg`mXFeIVmMCc
z66Rah(n4A$G#uRU8%O#rKNML8gto3oCDJug;U2m9(4uy(hBDu>)Y7g@o+;Fi>se6B
zI*&QRjnlnU__chM34kBydpp#}&y#vg5KdVQ(xq7g__y4(t%!WE9?IfGFqS3;ChS;M
z$c{e`-$4eMV6XoGaV%mS8ziZ=NR*SXTZxc#qTV)1TKp&^Qd%UU%512L-Z!$`?TcvC
zZ(G!|mJ8&{@c|gtKh*1evFq{3M~y$m=$_i9w|Iplzl_R=7nh!EgT!W|-phJuVorxb
z{$>dD>)~L_nsWcPyVw7;z4X%oweUiB{KSJ<E+RzE)@_l;rvB3dqLZV{Ob)5D;wjn5
zv6L^RQU{JFLp0y={+*3W#=wf*#GE=9{2fIT*7s=?ARv0LZ<%4sEa>qzk7l;cbQP6l
zy$L|Q!VZi~lmu@?qyAHBa(-L5J+On=I6M0%A;z#??yGdnC!^M<zja*AZ9$xG!UElB
zR|dFNhA58?6UaLB+H0}MLQY2Ka0(W9_U$B0+))(S{;>m|cm4-WC&6dB$~x3=*#$SF
z8Z4lSRz;BjM>>b{!`Zc)G+U{EPI2eUnp9&PZhzAsTSEG7C3B2GYT`VFwn514fpib9
zHF&6@(5|hM59NVbMT*nu&+J`B!1u`W$Sv-k{013YmK6^ihzK3=P_cpDP+b7X-+EJO
zepiA=qYCAR4*lyUMZj)B^p;7x-tXM4`n<Li&&|Y37E7{w&~WTW9a(DZmINO1546r~
z25gCx)ZAPz3aUITG32<p_?d2Ej!h704o$H?YLkNHZ6m=2?Lic7l)Zb7Z|tPc?<VvI
z6L|^;$_vqGA<rXoU|Q=d2Y+9(syC+r^BbR9A&RK36ld63-TH5>+H*%5O`23*JL3c1
z*F|kn2C~F;2Kj4s`O?TO-HE>KTC4GxvkvPExpgZo+iqid2gGqURead$0b$)BBshrb
zdRPlr<sVXmU=J9Q2)LDV#y;yi>XB{~6a-UvWqEK9Avw)7#DYLIegyqe$X7!reKA=z
z!o1#Gj7OKl0cTFkLRSyhZDwCA$2gg45l?CjLgm^-q8Z!;X5<LRS&7giM?QfmEnnXK
z*2nAY3@?>!5pV!DC$V$Ztq7XCDoaV`EvZTykwD9I^%Jhk0SZv;;V_UdW5{-4p-Ec$
z8V+!y=lm*m1(Beh@?o#m0-P&Zz2nM8p>#=0zL^n5C!PT(pK}UD&etB{q)$Q8tg(M1
zJx#c<Za~REDxYVhd(yepNj0Pgcg)g*8TptU9aE}rKgzw<{EA)>zPW20V>G+PTjw&%
zaRtGtpHLgioGC>$IH~&-eybpzU1kn35KP7!v^O<G9{N<k7{L%HqlY*&7dR-_+V+0b
zwit5?k6mXkwUlSk3jd~b)(|L5sL-?;>bwM<F95brdG9p9tWbl+mzN*CM;Vbl?8gd|
zKVyhaCeZ^n8*J<c*Q?`5c#s*?uX56gP-z)mjYkd@>%jmn9^*YETYVCFcLH_O@(?zM
ze}JOah|gH|GXk%FRPfMBAq>FDLBZ_gozZu^D!LMDz>ILYRn8N?2%(?^-(}{tA&K@O
z95@UN4QQKi{cEI|tlIvE7s+=3ZX*kaN+6QWCbYOM9v*hl&uUMs=4o;{Zg9qL-E-#h
z`$f1=k8|9>!GYA#X|XOIAS2cpa8Ws4v2z3!YvtxPQopac&n>R=ojk%XwY7y!v1NeT
zh;!s5*qYmH`Tes2syZZxI70TTNwnu0*R4+hXhDtStO1OwOG%lMObJcVW<h~{#$QoC
zSVcCsN%}2gXrFD9@mm6Gyqm}MPu~IMR5-P@&;x?x;0?^~^8I~2(*OQfcc>}aFzSK+
zoExO{S5|th6Qy%eA=@bPubPrG?L&)S3Vm>!%RBHDPrE=dndkpZj<ABE2gONd#jFtR
z!vxZe08T^1&>|E1XI<@*DnAaJI1#a!zF{~s(NzXYJe%M=nTj%=0D$_oAgiy-O63JS
ze6e3g=_cSm56D<^McT%wGOI7()A(KP39NOcbMU0C_fchE3y@4i-ob+jY(6Y_XWNU?
z=o!LnG1%Z6zp90WylJ9>F=e<^42pJ85k2lD9NiD-t*N8xePB!lx?Th_%zJUpX{HyE
zwI=63*pul=K5bY%eD#x=zKKmHgNM=_M-(KA{o`A}z?od<VVsFB(#^%8U(N;1&S5oY
zamX@t*S>eimGDrr`V8QFP0SZZ<_u;8(0!;y8%n6Et)=m+MZRTjZu57DtbXa?l+;|h
zK02^ki55%k$&G!^7zoqjYZCIthzA?%vlvO9p0eV{1Z5u}waQqu-|Zl-k>AmAKx}U{
zY<jjg&+Be<kDV1WaL6uQZ%ve#|GvE8s9mm7$3$W`qf7nGeyVjNLxUGYx7nYobDTZt
zFV)?^HMxBW@<!p4&v%7&wGB6V$CA2T{(h_nZLiZ#e=akA;f)2%4xi^_&5O`-<lcK-
zuWV={*oN%oR9Ih$eG!^u?RPi%_xEbwj<AwM__I=O<px2J&{Hw#?abH5rZcaefOfV9
z)Q+}I)jc-P)bEzSG7($pL|6cSdZ}UZlyx44NGKaP;T}mPuIOya6$*OD{{-sP;x65L
ztiW76yXr0Ky;0Tt<%d@kBO|w@Qoga>@OSxbbk+YQ;mMlm=q5+*yN{?n(_)sZh%*T$
z3fhES2C97h;{)m*4UTyyi!_p5MgXg0w7TmYMSnd0Z532V*A>LEw!i|pw7;2Y&Pnb0
z9nP>mv6yVHy5(e}hf%UZ0o~-7qmAEFqG^FfBnuN?UAU=j3)@4dc)*^wkRwt=t<^j?
zyW<ooCG~Q0lq5;p3lT0~!+dMh*t7(F_B^r#F9gzzIx(`Ewo+0iH%};>zTm@cPo2pa
zeR6M+Xle4JqntvEAlI^du-8inQcM!4G-ynGfy1rc<{)e57ofCj@NPlf!G8_j+6II6
z!*hUZ60u6WTw#E#)Kj4t>vL=Fcv5jX`52-w!u^NIB%zF#X{gn-FD$Szeip}W3+IBT
zD5O2fY+r}*1)G5UoG|ZI2{^TiIG^`8eU;S{@ep?=fIchMl~|pL2U`28kzOKLWOq-w
z;bd=&BrWg8@CwnhPC^iF%XjUtn!5k8&-6<d5`2SiP8tV$kyjUzHNYuKD4ZNN%%C%S
zWOHu4l5NlZ{TWy^2;;AKfnZ4I9Dbi-<98~ruVk~H6oTZrqa*on1o0jQ!qiqcQswwc
zedAXf=sGw<K@e=|mrJ^{j#Pj&2x!yEp0o()?%;*A0*DsHf;f@+rqUPyU1;oh7qeOq
z&hTfmAE?umsov!jag+)2nz*|m6a>#)MeLwwU4rgLJmn`|P>&^DgH#*ihk(jlrX-X;
zP?PsP6bFMgg26tmGe%tvhzNKMA;9i|_mC-7X0<gLbs8i9T{E1W$h8O=L0O>rzmW@p
z_@z0kFaw7{v(Qooz3aFk;CYHf-gOU<Toe?35rz$XJft<MqeWk*e1MG-nh@;3+QYFv
z*lB9CKh%U_j-<<mO7mG?fg?bh<Gb)b$BmSclLPofU5<J~GNI)p7ggFpSQVEHuD)-=
zG3)zVJ(5TRPHv^;BGEo=q@z`Hn(Y?i&|))Uf|^%K4*yOKRYJx;`a~#1_Fi0P(do-n
zC-CL9Gt;<EU@kY6xK3ms6c9Vvo$f50@#wN#H!w}7N%HZ+FD5inMR1-eNL)LNAF-?P
zpMwy}H+jBI;>X@6>v_@%YuZ4udp=V$IQ#EbXpwX`W2D^cJZ$ws-=io$G34@~S~=ph
zoWG*a*+Xxd{>X{X>|Avn>Qga_mY-b{4UJHy-u%fCb|KAzbcQZZN*j|b&CnDpcLtO{
zY1XrRRlqc^BGM_LX4XrDF0LRrJ%Z^HPmKLk_<3qxu5UQTf>Z17Ln2hIgGc<=WANu#
zQ*{OPqOTvYpg1|t){I@hh_^=t^KvwoRtjrV#(y_H8CV)wmiJ|)$2L7(9-BFXP#yE5
zK?-Mb@{eQk?PdevnOW%~_|Ylei;8ATf5X8I^yUhGSCr$lyCYZn!*v0y8W_>gw~8Y-
z7FBl^JR@N_8=t}uoR_VGpC)Ttwr?;~QSv_sd{a-iQ%<!`Ul6F-R@*Ztl7o=5b$XqO
z22)a-46SG7y%|8D!CWPqd4u$Nc3x4Y{^*y+pSoUI`E`ROM&YE1NvMKE`uF4p`vM+X
zNFHoo!gGsM((N7c2jaVFq8g#Q`ckfJIj~_0U~EZM?9OErKYJQG(f)3nB=@!8={Y;1
z#OM{LhJPXj7z~EBX+GR)PSaFkQ}ItnxFvXsJ>2Xp-r++~0b1EmhW#t3j$U;(&I77>
z=*pKLIxDUFZp}%&A4yXow#s7g+f<U1QG5ozr);GGV=@$zuwUh5V)3N1e+N1?BD_Di
zYBH`hyO{1IkQ9$gc6;*5#_W&Y1pryHq|$#)K+}s}dj#tr09j*#jKviMgvObZo`d>&
zOtmj3O{xa>X2>Q_=6N%p2AtV^dg5<;I0*Vgdwn>~SDmWuXL^5S{z%a3r2M7I!64+M
zt~ahYr?@QfgA(@-XYxF2q>q!ESq|R?hrMsv*76hxqPBe_S$Ab>?(L^%r8m&7SlM^3
zeq<}l5VccY%B3k;IC|B@@Hit<uzRd=Uq$mQHfxz<)D|l*-<{i~V~@W?HHa-F>v=@~
z4ZjLAUrm36g`iTSLOp*(PUXJ2qUsC7Nwchip3P4P3<+W=|A^}6ZSf_18KlKP>~1re
ziQ&NH+}1SWj+b6F>H7FU4^ZJ^X>(j}#vNPqU)`vO{Tza8KlFkDUqDL6)7aZf!kDFT
zziX=0NOQ*rzJPS0r~7<`WA3w-?IU628WC6O1q5{I`X$sVLd*PAK|VdjnZhDF41mpa
zu$fxmuHp{Yp^qzT6-ri2)aX&VDHvCqe~4ygd1M&)tDo%YFJ!_gMvc)r=uxa1^g<oA
zBLBG@>(L$)T?%?TgLz=Mc<2;h@debQKT6ibJWf2k6M2}_9gVG-EXnNP&rZ(wEFC|g
z*B64-n6n9t&<I5A(0nRjshTKt+VhpLFkE+?P)KA5*)4Jz(aAI2{ovF$<B*KRY2Mo(
zIs@}-o!^w(_qg!e;}mdSB-;tUIlj+=h~x6&4A5OqWMag(KnG`QyBJ$h<=C$zI3_sj
zYYHVeGN4{By=!S?NpcKq^bzY*Q2h`j%#_a#JMt0WY#;9RX*c%GR}gsRQh>uraxr>w
zv6I{_W};rD9Ti4L3k8@qz>Gvk*1`7hIt11;b9vE(^Q6wWWQI>cKxn2g5W9l$zD&IP
zr&Hw9+w9Jq1Wn=?#N>&J_;R;n<i+kvJ+WF!?q3FRNM@yB?kC|%t$&qVK3;`R5S=wY
z4K0H9LgHs1v9qLb$1pUwR+nLqAd5sG?a;T+k6F3CMTxF_&9$XidUN;*NSL>TE^scH
z@3U5<U7PkU(V<v?q7Kgb`TsYHUzt)eMtg1{=6l(eQ0y?*zKXf3v_M;&dAn%67*;xc
z%|I7BFPSAu_vdf1nz94(Vw)N}uO8pcLZeAId=g5L_{O)-BxOYvAvE(EXCg0hdOW85
zcv27Jw~W^bZ(P8K7|>(BE^{sk=a&M(v5;e0(%|B%<_||y^C4pvB+2tR7Ku%LmwT%l
z5*@mrq~T#j?$sW8X{*_`6qW7`<7L)^gW2c(J|;MCVW?mnu&*{N1~=8^tTj&NI3eGo
zHC~&eImLnvqD|7quQxTTw5UkaP94_LNAiiz#wwRCGcfMEFN&<TPS4e%yCNU1h!HJk
z^o-Q6?IM=DZ6IZMx4i&%)4YTRQ_yaUEjw+82ngcOh^o2?QaKxbQHol3aMavxGjODD
z;Ia9dZGNQ7kG&IQ8GS<30TFBuMTH=g>dVs8-f!8tmcd|rF@9iZ|FcO2=IUI0mXO=3
zsU;3wA`0Cv@5DhAOcp2Dp)8=Ku_fWgu6<W7>}28F<=)=^&#~nDTkz-{d_uV@hIZX-
z%=~HL5KLzEKK<Q3>S2O)oZ*xBO3b56S(0uiEH8Se%%~h#H+*I<-%Swe1Ih6P4qYy1
zUTwzB8O@jn15)Hz)8I_^%Xn4NMFhl=Z;}6n8HG8C2?tOmC^|XK0R|eOV5H~<bMzHO
zuU~>X)oEh<M|vstCvE09!0wpv^01D5^cPpPYftr+`R@b9X8t2`qVR9t=1`lzTEWPH
zBR!zzX4fZ}*4PU=fz*%Ix>?}XjMRuZhb4In@=jeUZK+stfVNd7aC=u7;u1!0(3D*L
z+`4qa!9~YNEey6*v9K7$t77{6VMouQ`LGrh-}?wob`Jb070c2}fG~z$FGq|zRhsgB
zkF9^jx*laq&f?SzIwG5i#~^_R8QOQnW5{Xg{@OxAks=Y*gP5>$os-#kc8wrhoF-L)
zE(lH~X|vY$ixqy)mku<E=7SOq6z(2!J7iA6`j3C$M?O{4hD_mnK@=f<aV5$l1F?0Y
zJ|-d$R^?#6`z%e>LjZ~K*CD&tH!CqIRq$Ky=0(qo1O3R}Y6hWPRQ^SNbbwqpgH!Yq
zb|mLpwhTz?L)2t-K<A*AUxbvPB#9qEY^2jX8Ih{AR;{$T$2(124Pv2x0K((LtP7ly
z?gWn4gWb$V=z*YIj!~uuov*UEsA;a;F}g`yd2)oqYL_c}zar2RKeZh}W~q@M_3Tsu
z0;=~c+BFYxcXDHHDa(}=@iG8K%54D1$P=NxLs9mvv;JY0iqJoim=pq57@1(JAwA~s
zdqffdB90hG+40UvtM2l5RpAi>6g0a$SWwP2T0TLM8+EVJqEi%ArGQWG)>ScSsTFt%
z)!TR^l7Ag`l_d?@8WD<(xRdVNw`ej?Hg{fkz({2iL%JSg=+xorp;&6Z>xR9Y1>a9)
zi!GTZt%{&E|2=VaNBUNzRJ;gm<+S`F+fR%+K&q^el)$4oxk`Zen<`~wnoa5)GIgP8
zN$5rCIzL=!<&wNM_O2RI7n6ff>3Cgq!aJ?|V^&lDVIGca(mzCb6hXB`WT1EB7Bj8~
zQa49d(63g?i9}CONfJ8n=BKTO!Q-`1_nLN&GEA7K8vfWt3M9b!jC}2~rzxevrulYn
ziIfx1{P5s>-D_@?>TGZ!sMBts%>lj<=vTqRNz};t03-LKi%7HK_P*&RSRDWQX?!oX
z&yd$4v+IV~$M7rOS3jSV8OJH9wAeu296U$h8J6pf2!B)=XM8^igG=q)HJu$Np|Cb|
zSlaqo{ABf^8w9U^zLc|bd`wySoi>T=?I!2n*2Y@pFTRfh5IAhKrTNBPm#Tv{{lNA;
z=yNjKeMf_5X{`U9Tz_^=&%G;}CwXjKraj?ejAV0)66FXsa#|~!nNsduoY=u+36b;>
z`}T*$tLZ>%`AMB%p|EC~NkATWBY#%{k1VOSh0wC^H>X`+AxI^>vZPd(LDBRz_yDT^
zB@~j7PzZb)>wFmzdWFFpR*sW0gZwu56=vR%EhKMBr?hrQ_)#-F*-CXeP6Uw~TZR;e
zZbEde9(vT_M6+n}X*|moh_p>{LW{#qQ}&X|`WS8sQRE6}l?&4fi>>YSv2g0s?9x45
zdOi6(=Jad8`t|}9YKRabn-<=Kn#;WalRj&@re$QVSeKi;Cn31RR_xJEwPi?9)*H5;
zro-?lCB>_;cV!&@A+2q{Z^C%p#RIeaq;N2$<du*%>P<BnLWVeNB86@cJf3l&c(4NC
zSw1ze^I0kJ=f^3YpR?rFa=x&9<WH@nC5oqZ5Pfkuec#t>l+(Q@sM-L;`29ENp6PXA
zDG%Dn7TJnX*Zd~b;&)anW<C<zU$vSr=BSNGZQ%m0PXJz`R7#Bg_P$dK4P#6#h#<2a
zmKO58V-l~>)|G?w%pLPT@*EHTJ&lPu&Mpz@;M{$MmK{H{5xj$n#@r2$fWMY9N*N2*
zdH=vRbN14VVWMip&tY|@$C~`KAV1A)ZxhJ-H^Fl=^UaW*RvKKXb?W79V@jEVtJ-=H
zX+B`5$px>qB9fkX62jjAMOzDspd9`En=y%7?!h|=q?)!rcK8gPej1S;_*GqHKHK}K
z6NU6%4=ImbB<kkYRQDxlXDauEW|4jCP4SlOoejajsu2+SGU<2VRs_}fhpb%+chtG1
zfdIaK){=a>rD*7e&MH3qJGm|7cNRHo7v|C8vy}q}WC>no{19(`jr(a=+S+4!lt@A#
zoT?qD)~~(wTr9@lJOJ^p)_q?Z*q`P=g5V$V!T2T){1APOx)bI~{-aF{@x#*twx!Pb
z3Gl4`&bZW9O%)*LnNjHyd=cW`G`h8jx{vv8-MV@Vr2A1rMiccLu#&%OLuv>?bebCB
z<Dka<E=q9f;6c@nvBXdkt}O))=Ip#wwS5nhZ(leDzPYfd?k<ACiae_IiuXO2L_7iX
znco&i&MicY9VjB(f9Ix&(IDc+@h>4yLrGZ;tr<r+Ct(HAUD%1$@Fc5~T%p75o3Ufe
zjJ+}W7Ds%)0>FLe{hqQwsHk!s@iO76<PB@d=$-l3rTq{`BxLdsY6@R92isS~RL21R
zG!Pdb)TX%d1vo##r%*s}gH06FvH<0_N%#R0v2|igGN?no2ZJl{dPSVG2>Cjs9V}j3
zZ$w6WyeE&<E@(kt?o=X{8uZbe-T-{Lt}2lXY^m{UF@BR3m;1mu@9<sn|7JYO@_Zt|
zPe^J32%zIQD<%iL%uWL!Vho26R9wGMv09h`UOUERR5XOohoXeI8Yf8G=k~E3P-PF9
zf`h^;Amrjkguow{OADh7>)TZkV3+ulMkL6U3#X+ZnxZfv+kHFVvMlDk+sizDa4P=D
zE2^q6KNG;%@m9w8ql<p`@d%4Fb`^}P>9$chVl&Y-m?@S*>6^p|Pqqa8V&_|VSO*o+
z7+4Ii2Jrr=*`WHcQlKEz$rDfFMg!G}<y=RYxsJ7_$p@Ifb391kS>&dMxzw|MZd9Bn
zBY7uu^>OR$mnS7D5}=-jI7r%XAB;^3X#6L!`<)yZ!t?a@X2(p@$5os+xQ)kBkSD}{
zhf`Zk9|hArGhVOi2Qdt{FH5C14yCV4JQcHs|K!@}di_;-bG{~*EcHsb?i9S;h0{#Z
zr9LvHNBI*>^dy`|K&eWY@^QQhiElLRTA8J0y<o_GflbPfz-Q4&+mERIbB1vuoeJLi
zfs{q)@>ubTB$4f9xux>rXRW)BK&VBfbBJE^&Rkp%HLT>wG6k-p;MR5F0!S6}r-U5R
z)aPVXOs!`l-9nCu3dPn3y`~6T&6z=bsL3-6j5BXKlBt%ZiM|Ba57vHsR5#h8HwJ}<
zXYR9_{M0Heyy|=a%f^tT7KvHG%`^0oBtTSPLDj^`OkWR`@Ta~T%7&3DPWXUMd`HCH
zuWd*aNN)R5w99P#H1{ccP2;_{oh)MOC)Y>kg?=60PtoDV6d7DLz5p3&8Th=KGUuDQ
zPGF?xZ1q4(Z3|(Uj_3K*2CX66E8^HQ@h&D{-KDZInuTk5Hzt`~BJ__ccbGd3O8mr#
z&H_$h&RN;ATICD4-qY9XITY<*Kt$X+^{2}1tu%Z~KKxq3XzgD{*{|ej5L1h_fX7I?
zpM~++fC;DV4mjVXr2aFId=N@BEbVn*F$jh84PzkxI&(Q0F;R;+G6v*D+Z1D-dkLHk
z`9WODL;u()Fhf<73H44CqM@O9*-7h5H9%3gJL&ho07XE$zt_r=Tq|@!aG10H7Uj0Q
z>51(R#B^l+DdLciLBwsFbMT!AT5{Kh5Z3aD@$j@;<hci1SyQ{&7`qUK(KeQN84gq{
z<}dn#!a_+jJKTcT=!i=6wq7|@Cx-KttX)9VLqqPf&5T~j><5NzJrVDue|o32P>zTa
zMi_sKU7KGSWcN#;?jGfe$f`Z>+Bf-Aj1%?u_l4*8_{3ZIEZZBjwVBu!+qM*`gq~<G
z4_zEe;(<9%;NpOK=-Zs-sc(Kn7XH{<8Pb#2uV5AU*rmf3MwMI;EPVw_+O$24K|7}H
z-r8ST;v`^!;|-+9uov^M83IB{(d}Q_3+F{N_|w=7XH}HiBzPv>w)oEi(nr-z@}9i`
zT8Qqw32#<z=!U7D{EwfZBH6r*FZMiwYEUa5o}M;S4hJC^d)v2MMd3%~rq!80-9-bn
zo`sl2=zqOEjVo6IXa;1&^I0lToH_s6pAV1pZsvW_N(t<wT$Qn<jAm3!s~OWXgm-OO
zWm;>-^mOOwYCJW_j5}i|(#V*7;a7iIJVz~bvtdn@NLe8}C=eL2$xL(g)|@MQvh$~5
z8ivCR1p;UE8ST-#S5>G<ufLc2u;>+Tc?OPC4`w?T;GzoJ%eIl!?VO^YnfQ$g{qYP?
z<LbfCWZ%<2_?W>YeG+}9-KeY5JH6*4x6{f9hp-IO@Uo{><7*c4O|Cg2?2&KBzlY2+
z_xZivVayejTP$&SDYhaPUhdeDWYFpJNu-(|wolyegi1{zx{vVJzuYmlvRtYR%LEd;
ziUPo`79bEjUO@F;vvO0^I)8YmdacjQHnx=)WbtfaXZ{DCbB!-F8=?|AsLncd7)pO%
zu48Zc*9t^y4n0l5<CnF;g>Sp(8>vf_rB<l{Wr~6mg^{Scq`z-R2SvT|mGVvV=!dw_
zTn$!u9g$#``=`&4P|qAtbqSCZ8YQL(DL6>gJ0d+sTR}L-xqoL8k*Q1=Ri`Bb`4%?B
zw4$T399~0%`Gjm(>}M|TNRB3fP~-lBM`5q<<H|FdtS{YylxI`NJA~mTP~f5KxP&g^
z3vVQ*dGLlofdkGrbFoa#>^L?kVP7}b{wa?nN|yji;oY<-ZpWq)%Z&uA*#Q?ffAumE
zbn_ozR(P-WTz%EWCNG-mNOcM6WuD<cGbcI^Y&xX5Yz&7~%jA#aywUGuQf2h(VTWQ^
zRp};Se#J-4uU)4_hKj4bQLBD_l*&1E0t;myKXK6>@?6=%(v4kT)v`Ubz8}P&$gOSI
zQ7zhP*jcdaq|E7MY$|`J%k07-vEPh0I*bT8@$L}mYg<KFntol#9e-&G%rDXfMmGb(
zL&%1BTQ37k=(jrXBj6EI9LE3lu?`H&?~~aw^tgwAw%*$fTSzL$3d-S05jzfQj|rc-
zC`=}ygfgGahK4n2ExHU4gAJt7?i-<{7E}SdwOVF?Ej#gR_)SI4e@&f7v;wt?_CDE1
zyv63jSN+cLd0wCW>~sU~#ndT>0J`D-BQCQD9MoYo-nx;pHZ0Z@FLvR2LTWjLjkO+;
z%o&Pkpkvck6<wwNA^649`PvIBfzs56Jz&U{OmR<wkT|eQeLPBu0}N^YJJ?$~sZ1ZW
z?xb?8?Z<Map$ePeT99KT*JaA?KpBPHA`Cg=rLa7ZYZzCNonMS@33h<VX7r|I;D!Y$
z(T{ttn7DURlBK~xXXONT?;GDJR^~%g9hq{kO0A)y+^l14TxglZvG5t!oV&;bnrH=q
zLpAagz^GDBy``P&?Tf{g)_|S*Fm^1Pv*3&n)%fFC($}nFB8KC^CJ09~D?!wQ>n4%p
z`-9$@+;HZ9Ab@;0G>Pv3=f)2NC<$k(LI6O!6gN)|ks+^GzG)>!M34)df@iikjx`&P
zs@54aDThiV4pqdiEM*8|zwSkBCi!k7uRvoS@YPiN8bv1Aoj087CLOI1-}JdN|A`Y-
zu@vyBxyy;022o4j^M-@~Ly=?Yy8D37vqM50eu~?R*85)Kw7$0F#Cr*crJ_IwrSOuf
zd!n~c8}2)IyI*wzXH1mlcB_(BsWe_|6*y*347Wbtqbb77s%WkyBFkv;7yx5TOa^LQ
zG4`*tygbTjVn}D{r}YFROiK6v<*(Mf$}R@*&YnhofE00$E{e8kzQYev7#Pma%J2!K
zsE(Sp?d$h4IRqqp<RlXzvJQaTyl_rj&iC=-H9gHq;8>zu{#AO3Uu~fntQ;&_^J`4^
zNo8IBtrlNRYK>fnuT9nOQ0$~(Cz*Ek^=)CQaCP`MbPhM^F^ZuE{6U`H-beN(u_X=X
z=j$%)6KJv%eilsXCWpXxM$j<*E~!6XB7nm4x=_|rmF3Qgp04?Ku;NT*1MbM7;kUQO
z^;7PB*V=&kpsP02U(NCapK>^+elRjF@9u)yLB4hjf^RF|b`2ToAY=;%4voZz5qIrG
zfBOB$Qq4#vh0nPbks#4^fwuhRkMNoGmlm1(X`8%`l94A?_dJ6kArlkrb9_{Rz{LGj
z<>D+k3_>uD%_Q&H0>7Ng19!D^ut6_xS|ASkt&Czbxe5bN1G6Pz7YxXLcE419rSwIg
z3yzRX8rTm<iQzpO=NxVxAIe6KpId7kXg(vkc>RKhF>c4Li=@ZF6Pqbc%7nB|@RSgr
zc-gsN3(cXb({fXiM>Gvm+~vqRF2!5nh<Co<X_~eUhKtz5XY)~^Wi6F!49AwSgaKBC
z+H+|GVY&*<+1umo6q-laV@n!(<y}O;5i)RKzHPUjonqP~1WzZ(PN+-}mUt*~rw0yV
zb-N*JJ!)pXx)V@Fk5@kxq@3W+l92UIXOwX`&o0zDkx-ma?}i8G)G+N<xzxDp=B|Zp
za?HG*ad0+t01KK6HoevPo#f(4hGsNE46f}sZ$F@EaH~T!lh)xX<HKlFaMM0&46lZV
z^<yIogPqRk_B8eVb~{tM;jhL#SooDBM@%TP9&=<m8$VRH==z^WR%J`udZB$N0p0?H
zSZ%|1J8<ZS6T$Zs`sazWqJmZ@Bg6@%G?F%0?h~RuT1YlMkE7)lxC{pz)ykxgwwrBa
zwubehpNH609$@&faUkm{yDeP-1pWS(YPvFNO6kD><Yni8ON+d2aPxx)mF8PWTRvUO
zR}X%{#Aawt-l^B7TW!-c*(t^jsd(?E)5d{LE>FVhhu2LV&5*_9s)D8uWLZ;HgS5n;
z5_!#gsll7ZG=x?nMq`<hP&`|E8$3L05dNjN`*3?@smAeV1e+KVn?=7(Pktb#y8`Y8
z>QG|TNX%<2^o+cv(y}RM3VjLd;;t+T3dSB(x)0?Mnb|?Dbd8%=2C6lPP)Yrb+h#*C
z#OGz;o|g-7La0+5RKYGp^<Rm5XADH%ggM#qQJ4s59H9*?&^uPB-bja|3rUPf+Z!bS
zy#Q(yE=4dUN-Nmzy{#H2y<ZR%Jf1()O-%71+=^@uaYVj)4<ZFWtu3~&fr1q@wmBR}
z<T2DeW;7S^KOFOZ8h{NXIY?MX8L0iML9|(TEjucMkQm#vx5gS<BBmqQWK`Ylv90<#
z2Da1sqOgx_i%WyE*Jp^gq^=DjMSO18B(tt{kJ*<yDXt-*NOP&d0{gSj?2_WKb@afA
zaNX)RZC^y`!p=+wxVc2SW7NjQ#L|#`P)i-G43|r7!b-E!t*F7@7^&dPz!E!6oa^j1
zW5gspiE7s%!QS_6Hy}<2Bm(}8k#4C5P$&a;Q@yw8yl_9cPPd5`NTM#-?@r&SC|^AY
zHU(BbG#QL&$!KZfP?m%p@vpiNH_%H$7L3&&^9Z?_cAf@}_wC+?s(B~6vlZuok*Rf5
zBzj#a-QYXX^0~8}^!*M3h;r;Fx%;xS__{oK1{#&_6K)Kq+UHow4UM&_L<|ZfFRvF^
zgloc1SvPc=4tjWXFapqO3ObnIGQg!Ir^$Te-;i<2)#zGEGnK|po1j^{k9G*)&B7sw
z8o(4F15?ezQC(F5deC{xbPNfhX3g<y8D#TZ@jo+7(I$HbKX9xHP@vOEpl>g&hKMHC
zL(u(VVJ%y>HbuV%K)Zld?il`r^bO8`3{SDy=(yTp3aV*c1G92(A8Y?u?n0Th%og3G
zu9nVGCEV{v2y2g4(7kdCvjXJ~`$}SV6=x6!Nef$zFl-)RLNJTUER;`A$bynLOe_Kp
z0z$?-w9`%R4DhRfG+X=1%y25ju|cUU-So|RIcVBK4#R)vIEPpm!j;9Qv)3ub89ty0
zQnwngj3RELP>xjmFEZYNR~TSj(p|P~>sz*M+qP}nwr$(CZQHgn=gysEviJOfPCA|L
zepB^S>F#7(+Y-u4(jL-c=*yIu;dZM98Mete2Yt@f#8Y6G+}Di-maUk_K&K&W20Cj%
z^DIK4C{SK6uh2fh5lR^BNlBY4$embz>%+!ZYHHZ6ttvLr55x(hx{8QDBgPWn(l$Gf
z$F8F1tgt*8==BQETOjo0SSiIi^VNp>mDZdgQ5tk$$9C5d5G7R&wWLA?c`{P5OpFcS
zkXgTa3_27Jc$u`dbJ|xr!Tr0it&H=OJISelz`#}D7d!4HEBsGuu|2#bKG0n;2T$<q
z<^`6hVbbxoB4ap;2IQXUJb9E5)G3azZBO}G{iq!`7oQ0f8_N{rJ2fjE6i%KCvr~BB
z$+?JHW?@fWJTjFqK<<D+Uyiw&p4Do2X%Jsp)X%S-IrOvbOg`fiiONL&s4myd_U&U>
z)WyLsk!ucj_<x@Aglw`GVtZi;EBfovp<$YAEmDP~M5&WlSVTZ$CeTZ@_}?+w$srh)
zA9Vj<?<NXjop>N5JQ6TCCDd267=nx`)}Q6BMTg1niK{+lwJsI9fY9_Ad~(tRd+1s^
z7Um<%Gp(OTNbF3h{#IP0AKnbCf1;kA&vgP%2#Hv`<=*g&96_T~l$ofPu<1%~_N<m0
zF`75EOCmG{)VCz^O(kwrYr`{9zecgS)d~b0Ee*@KC!kZle0jc4Qq|Zn$Wf$RpyK)f
z7S)y3&gi1y$WP<2>K&P=Gh&1u_K?>!3J}Rhxf3gcpmyRXDpu-f>2(6EB`=ymBvwZI
zwNGAHp12nxtc8-LSYiz&F;_Qi7G;7oh_ec&weshj|Fj|{SGl#ry>;9pXBbvFbovCb
zYvo;}?;e3l%(R8kEjHp(24VVNG<}=tl8QQ{nw?yc?i{H0<dT{7z6=N{vQ`#JTJ7bt
zIApOqJFI)wx1O2kI<?`e8x!XNXyS6Xh40Ld_cP%L75>D89Dl=!G4l7TAW+%14vd`H
zOE5E?#aQ<Kf6(=e{EF{ys87^x(|Xxp|8EB!79gy5>BjSA-kP|qq5r^+yL##qvK5li
z5NuzSO6B(S1*#V`4(4E;oT`yvBfNWkn8Z+e4XE}F!!hM*Mc?RWS`08(#~^vjBWB_I
zuF%v=+%xB8&Rk@Ku|4>`9$w${blt;+F#YYXy}VKtGoAFygu3Gj3sAG&jsK=ofldf1
zh{~C)uSt80N>V<x%x=mq(%%*g1$el+ZI?E?U~Exg2<vHhLY&EXZ<Xq6j&+z<Py!vg
z^Qy6}v<%`W3Ss<KwdszpCOzlX33RDY1y!S$3glAMho04=6jG5gen2*EE77i-TF0nS
zz;yycE~cvNdGJe0L5c&T3Oit?>)8K3<++;lO_tljG}n@P04jq5(<kc$LGw!X#%&pV
zRskyp*D(PMA9^hbm+Jv)xx4?yp_W(rh9(ox^=g-t*q}!bwvRG1oFasuyfu&pill1t
zGb1RKc8X-$i2<uCmLxOYU}PmS28amT;|*%VcnygQG9TZGF*bnxCPz2aH(%pCTAyyq
zTyCT=>e;;1_RM5QDU&{l_@=r}v|a~6kiq0C*jO8>Hn+EK!GOr_av`!nX{c6*jSn5A
zp_Ozm*Ptya^nS3^`ve4|(;f-OA#gS&l75JldJ6#2P2*n!)uau1krxBVD?l->TKIkP
zpA43HzX+j=TJ^Uh;|pC<Zi1s0UWQa%gJte@(ajac?CBNn0u|`|#FKO%ub?&WuL|BD
z8~Df?eT#(YX7>(YgqW+-0HsYH@FGy{$3KS0OAsu=cJp0RBS)G>d6c*QJB1-Dq0gV@
zk(Q(!k&5515L=LISGQ`vF?6t7i2YN%W>8lvFQ?qfdq0g+EVu%7rO#4+=i0uos2}%%
zC>(8z9^L6ss2j8JWEk4yIscsSRfYc1ApuqNO-F6c*OMu4lQBp*X!+sFN~<U5XWh9u
z7+9bcOh6KJWF51O>dxLE3Q(}b%K^?i*e$TWHbGzo2TXIjs<-wo{}8t8)koS|8M-HX
z#gI3k$-&=opL9s-je+V*BO0_u(o(tVD-cW>$Fj`)wd${xH0;UcBgXzARzotIBEhv!
z!>CYs-&_YLnh!^7B6j|`#n2;lsrI})Qc}L#lV?W$h$KqW_`ZDP{D}n@q0La)&$Qo$
z&S!o60sZS?lwx<F^vVBJ1II98dS2*WtSr`~6ux5~ZOY$SPZAB6$hX|*C<($@JJLFV
z!UTRkircoTQJl;gTpqeM_zTmemvRoG!bNysWOLrJsb5<8SiNU!```i0IR47Wai1v1
zPO#vK4=wYXe)o!LZVzyUaEZ`)Bo)Q{Jix^voFBSbWN(HMyh6Tz7qPu*KhhUQ(`2>A
z_xY3o%Zo;}ptX|19C_&gYY90QL`$4)-1`tEW_xf!T$N7$WDj!4ePr%HV@`sM-6YT4
zBqoS@JA46!xRa!}rp4as&qiKt*^a}H!z9{aX_4ckuJhu(&ilH;+5jHLJT8zCl*Rq1
zWz;kd8_c?ZQ>X*EVec%5$Q&%jkhWM%x#Djvm%I+<w$U`ze}r-SXM`AuGu+((3n;XT
zK`@aNPEDZ9iTeWScjh-_@-G+GJC4L01}RrHJqFfgo80`)cPzsYlYcWK!dzeA?_=<G
z3#-ASFiwDk9Atzw6t>R4<>l^6^5r5^9!wY^TnTN?gAO-XnbV>n(vrc^F}d5r#556N
zLXB3f?aq>ReBu*Nm^n=8<Ess2;;T;Zx@kO+TK_CGtJ+`xIsFc_ERv&HS>p>N%2i=0
zWm`4zN7>c$$6XHk+>3@n!uUFe7D1!39Q_`4X6<i4nkb6bd%k1zIA(bf&&OO{=i`9^
z+1aPdDP&#mjeXI?D?%~z(ncY;plfuBUrpa$TF95-EKt#Hp2EMu^u(3Sy2*re6mmiM
zbYcE;MH9&UqtI=gf!_Fh#7BlZPpvu1PQ$1uNxRkfVnguShr_Lh#MP{?a*!+vvG#T`
zM|ps5OowRmUlMQDa7967ys_1aHJolY{}_~zB{7lGMN}g?YDuR$qx>{+uF$Wl=Io5+
z4Rht7!IjrPLL$&|8U{janV4azKjB9StX$bpis4^G=h2FZz48W&mn7&cK&oU$0&KA0
zeXlwT^pppq+9-&^VRNdJbfMLTRY--i@3BJ<BEw87SA4FCpPczO4)O@2Dtw5+jK&=6
ziystRC2%4?ks_>1`ZKcHz-MzkN5US(zM>xdd3ty8yN=o0P&Z;#8T;|Y0g2T`g|SB7
z{Kb1)d<yhv%NV3kzd1|W0`*)N=#P7uKbrQWSI7;u$@%L{4(^fi*7N<mcS5L!V;K)A
z-TVO6A~L?NCYcV8oAgEOh>!d1v}33AJh>;*MB8ymR_hao7h$}wSMrfQm0Ol_`U6oP
z8^Z_{SOoQdCu!J=atA7}o$Nl*;NE^wQtxUE<N7--byu>IhfQXHoKE#V{?}r~AI0KL
zR%S_YRG6@y9q>@iYC{9QrfPGDRaTx!@B-j^f`6F`qO%yddY9P{17zyVZkhEodrI3=
zjnE_~HeD{1n&p_(nLt{$M+KaGo9tF)`d{}!x*@i8D5X{RS+V!e<(E8cC4=E!NA?-!
z<+o)P8&c6t>6Kx$)B&6bsDhp=Y;>08XU^?C@)$PgS&mWQ1#cmn60U|rfto|wlx7&c
zD`Ehus5>O^p9m#}Qz*-(X4q8jaoRZx+NQtwvODn_DwAE=wST5iQ3@ls-@@Omvdf)n
zKZZlN%UEcP?J6e}oFQ|7gZUx!N4p(6D}b+)d#BVK_=;Uu<M!`}VuzQON}xg`+M!A9
zENyft5(1aFeVXQ}FvIpJ6NeAsIVkJElE4GN#h23ie;uD}=b(}E<-4#4<5gJ?nOg16
z*ya%!`igHiqS^b2feGJ@wJe|@n{8A9!6?S9V^RioD35p5`oA+_zXlF;&pvl;f-+_#
zipA_%ecu{WV`BYVgfJ`%+BoFfQrCRQVd1z;=5cwuLm?wcJhQQLdgw<amz0ppJ;e-k
zHDsdM&}Lr#rI3FqbX&qOY{XJ!BA6i~cOx(M@ib{!&+DSe(L~z~uA?tNo0x5d@%p7|
zoQyH31fbXyPSmM?uo>2as*u%lv(bmoZN>8Li+RCiBij)k?4*wE@R<l9Op2*ez4I=O
z0hTG>w8CVAEg4KqL>?gPOZaTNmW9T5UXIugF4B4n&3A<6sqlY0Qy_wQZEkd4>l5hk
zr|kUq$`3Ph<>MVw!=W8Q!xR3?cLnF!0pESKTY(;G?{5Fz@$dwNg}cn90KSwn6T^_K
z#rgI=;C1(>g+s&Rn`zP>W7JrpH7%*$bF&=r>Zk}Sf=&p(<bBO3NZ_MzVHh$+t;ZAO
z<C_v(^x|x$A69?~kvWTU!^=V%bl~_z>$@O}`n0H<$K^>J%)cHBj@WyTg1Q?60z?0M
z(Q|3H(4Dn*t$&UW4bCI9B-QwVSSm(R88d1QD*W`ZMyGaG8kBgPnZ@GWBA1_Q_p!(n
zFk|0*(C)oWtpmWXOi$mjoeCXJR^pB04!AZ)L1MMb{d6Ynvdq0!Fj^IdagQ^X6{KK!
z)BJK;WhEQ$!xNOmQsV$B3Ae$`I5p1t(J>xV^iMyBqLwRA-gk1-M{}jQJ!W!c^0oW|
zQP^>Q(lYn(k(}=pM*nK56bGpkx-L(ZYns26^D3uIe_w<F+dj3*ae9_3iUXh}+w9i6
z-4CjgENDta&9SYvPkGbP5KLdxFFBr#hzXBU6=a;&v`~2R!~jP<P1qatD)EwfxK13v
z4?0y96YIMC9(C2rR;qf>#F)rf{jQ?E`(~UjEFwt?0Yw}*x&EkX0kFX%^Z;49IRTjD
zWEy9iPsYCc0(AJ|5jztgHN=v>bep{>PmC}t2}YK4pJQig&r^+q9&r^ET`x|9chvEy
zJ!3^L!*IJF+uL5ldT)JF#!uFo<6%}npS|9@tEiDUc_m>X(mg8<yXxjwo58as{_Q^`
zlF3N#^(N+B=U1Ls{v0qj5YYf*a?i>i9Bsr@W7qZYSJjV&iA@Mz-~(eT69Qt0_<sfE
zBfJD~N_U=4j&AhU$?qX!ak<P}-k!OpO2%GHMbYv+XBNqantvUwIqQO{c-GwM`O%=p
z++P9^{ESuq5x&en=Wa(!XUms2YpWcz@{q>&tx9rlqP<AJcP3_QxHb>9TMD7j>R)@4
zKov1IF^B)dp*8_fFiO7t$U8c+PxDBywWo443;tR>eOJa9jT<(;aue)fx7uj19YX5N
zi$M_01P}W$Obb(8w1m-^AuL1n6tsP$Up$R6fj0ciKh0dOMygNad92DX=_7E>HN$I*
zb75$szx9SO<eeT@eg4Se@{g0R<hEs{;c-TV9*gfPD5jUfHY(*(*R3h~VKOwOFS&YR
z5@L;tP!pk<`mKPHihklPDPr|_al=#0*+?wvuwPc{l`8ma^qqBEEW8ct`XRYqoK2eb
zfjIDEy*j4D_mLb|6vn@yoG)vTRCbSW<K4(-uv)FKr`0+>0S8-z{<Ddg>2>)XYE#Fa
z&<3KdC3=8F5sGPy=<vG_#SqWctm-PO*&(%S<W?wnRnRbjmO;mc=#HoOGBP2;GqVo1
z#SbjxDG+}cETwhKpSjmIyjMf{PvYMci44fC7Lsj$Xy@r6Z_j3~74;1b8{#QV^$Z~*
zU6fUNeV?*u6g`-Z5gi!1idMH=fVT0UBH|JO$%%QAfzc0N-Lmu!t2w);#1FIQQ>R>@
z{2}=l2sM104Y!we1VbEv)?j!rY+fFHVN<-@DRi^r)8sy9--_aPSU}ktZVysfh)<Ga
zq`mR-6H<cVGBm~dr|9Cl;dJtBl>Q*!<@t9?3o42mr&M_RBObJnvxSj)0fv)n?sv<?
ze6(G^@#zO#V4#S+a~yAI;th83kwoL3O9SbPlB;|K-v;3l(<VuMmAfWQV;ZX#bru1R
zL@&4WJ~%%f7a-xz41|t5hqUBq<<iSjUr}j4MJ-rzR-iSn=&q=BWg<=jN(P)lBZ~qp
zx^*e7P3tZlKFO>y+L-5c5m~{CD`yIYa5M7%2|Jc?AHfdJTQUah*z_ED__XU}`RA+9
z>4=aG?o&rutV`#r(dI#{K!*0R%Cq+4!xAH`$Ecv^p=_y4L(8Z5+B0;&MZ?-GN`O>V
zs;+d1u84(xw1Ke!0y6?ZW^Vr-cS+8SgmdN=s3+kbW>pq19hy!I9(aP-+Q4;}&ora@
z5@WOUy58my`lDj@nQ<boe}~O<b}XU4&EEOyg3&woSWli0+Ivk&axc^E6;1dNODF{I
zlNxaryWjb+m6VUoA9gY(V^7_;-JD@qe|y(Y4W1fW5zL!5@KTP5%hg}(mJKn-M!TA&
zn$d_T!zJJW(W~Mr1J}$mX9oJBdq_hybS*M<|7VgbV#at2UwICwZ7&GAZxxt;?V1gM
zURN6O@}*{4Km8C4sL8~4@BOnXvgp_SM*7F>yDu?5hF=h^Ge+S318#D5z$C&a45Qzx
zXA`omXEj#Lt1cABjQM<J`*$l$%WYVM>vl;l3<rfskyB1K#fhi`SXS~BF;$?~Na6ms
zgD6x#vmkq81Qg7^pw&udO}*kVc~hzhDlK(y=t!3fruf)@mC*{>hK<M%IcG|j2DvUU
zI8^paFfNM}4?*9`Uk{8N*C4J;>c0JWuaSw4T^8>y4ce{!Fq=OV7MVK;a+w92Nevv|
zl~vgsMji@ME!R?ZEQ%p9Eg&3?F1B+?cQJ0Jt>)wjhn^d*-*n(tSpDhQqx|eeh_};z
z<73#f+cCFlx)b+B5Jq4=Yh<$b7)umyL;W821_AEHx;e!^dh43s+Dr?!<Iq#{LJ(Of
z(rq!j)O6U|shj9C2<Hw^y*5_vGWmm^0B!3lrS5Eb07{#_O-?UjP1z7eHM>KA-QqT*
z$g-P%RsrMk`z`c+%_gveL$#lA#)sRR^fU7|gtc^dlh{U(B<N!JkqF)oSkH_s9{E+F
zl2$Lec47n3OB8=~)1F&1hAEg7L-}w3(MATmi{+s2qR~`7>C&i0oQ|5qldHEgjZ4Sm
zW<!@?56s~8?^X8(-+^E997a9c<0eoMzc!N778NzXc`U3;SW93+S0PG9((F33Ji=NS
zPzrLu-k%8x(;$%K)aJ!|AsRvKH4Qr+s<*Gp_(@tus_pt^?@(Bly9Yv0WV?TO%Bs9@
z8X^GUQ{UHDe+8{X7vobY^e$-LR~JJF&H5xAegDC9M*JNbR)l7N`QCb+EoOh0*hd>P
zTzu7&X-hNIJt6rwJx=dZ^HmA-UGO7XGV8RbOo`QO|2yQmrbH8UCt64(Uv#EV)Kn}5
z^>ga+#=Go-gf~hKY~%D$UOeMG6jJWu#7z)c+B4wjv;?ikVZcZPD07yG*>CD|0Tp4O
zt>6ZclW03tlwXDT%ZlB0bcpveiG-qOa%5|FhrC-pV7~gLFkr!~q?4N3b{VbK6v>vJ
zr&n0Hi2K)XA$5XOD(V53xQG8*`?A>VLa-eaAxH0b=0;aWXqOJ9q!H3RI_vo;W;T$N
zjP3F6E0YcXs@yWt2QwjrEI>0|(g{ps!fuZcD6E$QiqG^_d9Ty!A7;YA*kjo>i!PO%
z973990UG7Hys~+tC-PK2@EU=H%phB5T;!g6s^QQ`evz^EYTk=bV3)|%E@w@{41DNi
z^3lMV#vIL}Goq9tbWaso+UC^(ZQil~pf<sK;n>($H$xp3K8C7o?YjU;(*8`J;`cTQ
zkX5!YgXu@S{3<`n$FRrxG4KXj*#hSwOmrjJLmM=1SmIktN(ECa-YOSPJg*L?YZ5%%
z14d#$4}HCn(G}~3iZ32ICm&i(8T-IFI#*b835{^|tf@{gG}nfxv7I`Er1Mx#hHuhn
z=tqP-tx%vGxr9i&-G0L;>oA-gTa74JPCy}lJkqskDvDJGNO?n41c8xuy1e#DV7w^e
zz}ru3@`mxf7PqGp=ets5l?}iyeoK(~lB^QQ-^=SG6nfG{kF0*w>vK}S{gMy)j#;m(
z%xSy0iTTzjq|=Pb_DGFg)kc{RkZHI3mVEkMnTv{hzFOJAU<@sJ2u(QS@~kq2OEGLG
z`8@x%t6~-xy@(xPkU2wXQ;n+2u9kI_P8s6>RUP01$Zd$do<C~6c5_7ClF*BfuwNy*
zaGmwtDw_B#kDq?Tkg<7{V6Lur+kAKt9I91r`wr8Fs<H111Kzlf14-P;R}y=EE|Ir&
zZ1%}@1fL=+6^PR9DDDs_bXeK5_*y7vz67*~jys8g2t<w5vs1QiO8a{byPTqk=-8iK
z=2CXva1PM$K|QcdjW&XV?A?xKZ`25ol|}11200nhiE=LnQNZUz_-jja{aJ#EmrYgl
zA|e|aoJRk_pH7LxEO>yQH&8Cp1vSeIQ{Y*Vi6ml0DkM~Go{eO$MOncZx^aQN`7|hY
z-HpM_URIIeq=aD6=12HEY@{V5HC`wLoo`63fF}?9J>82mx<vnoT=XDU>d3$NsQy2c
zy#j1tFb*gF!NZ{~Dn8G@>Jfa}cUsNFVml0*zK4|UjiwRBo0+8|TQ=8yxAqUK_}EB5
z@uWqBx`k1cU@j0Sk3%d2>)<zR_^s2B+?=&hxfDWA2obrKWM#V<b+)n{Q_O=|k7F;Q
zM|slF4^X09;n3g%Sxxjm%X{p_>nU+U22!6gw?Qg3GCinw`84w22J{PMc!i_Rt-H5)
zUBaxjVt`)GWecQBFnOX8$9TBu-eoc&q6rEHl3)FaacQRS`McKzl4@Pdd|~}emoi9v
zn}ypO5EiU@A&B+voH|0XAN5c1*GIJrxOS_l+-b%qyJ@|hlhWt_x#k`X3qgwKjR&i5
z6WVQNpCs$lg_qvO0v8eV^Z<JjT$DU!zFqIzdxG6XlK;fB*A3F+iODrzmS9paggYbw
zY=+k-I#6Cv%n0B8j#vtl%reL(!-?Hq$&;S$7?1+BPzHZpr>bE7DWC!_xq`KK*usIR
z^I{W@MSMV+01KbOrv8RdNO1&c;oX#67x0&)B!Q4J%>-1Xy`b`DZT*Hstzt1p<VF0?
ztujH-;q!Y2^^njyP*3AjjH=>2##F}4ke%`4-NLcamm$T$IF?o6CA4zw=J_{ISaJdW
zV@j>O|9CP|z;j6xl*O_U{}0kK1XsP^hhU3T4Ra4}%Pm-4YeYyj(?<q_#XtNv*#CVx
z#--VAlNf*-cOW&gb84bUiWVHp(nHGZd^KE*9q9(q1F7>}O(PZ|IXNRvZ8702Q4~;;
zg+{!Z#R}yJ%1;96o?y~;(7VU^3v6KF<c8XY%HiZ(c!?vi&2NqBLdBBdXWp8r{xjnV
zfn)B4YB}d?xj$MIS@chyY^|bfqc!GtSfQW`f+ms5*tQpjs{CVbx~on(G_|>s_2l%p
z6-$*4CE!3vj~Aw}`kzbZV+q2TwunN~xk$q6Hwr#r#}yF7-<Ha&Og^`ts~*$7T=P}N
z(O<b~ByzTs+B^qvEjf+kl^2i?6qheO5iUpOY4PQ5f8=^ATFO+fsax;CKjUeZG2W0%
zyQR=~k56oTK3>4WU9_ZgmC8F-j_HE{SZ$JQ{D@x6?QV*MkF!g)&KIN?G6`wfd(xib
zT6BaICc0o?)1HrIpq0Zc<oZEJ*sBg9I=~Y7bm(V0#xNga(fb{>6HOb%EQFuB)Xo_8
zE*8e|BPNy^40-41j`<Y|2o&oNl6X2juYQfox2SpHXxM~}YKb-FSUuA315!b}e|tMy
znZUf5<4-bKB9Jan|DQ>Js{sNM@Zhk&)%CIOWtt(qh5)#YW3KTJD=arFwwx=+Vwp{}
zK`6OUCdtMP)DMhx#Yq|t(hg4x1fpi{jDa)OTaBmLsbAkZZaWYc<j_VmaC<0Y=5D0d
zCWN1@p(bDYxtCC(V-5q*eN|1)8oXCtt>`^+Ws?D-Fd@I7eyvdUW3J1p<r(c(hX)%>
zWThyA=uCELj9nxw${<7*>4E5qMmx$8aD~W;3r`J9Fc*)fh%f>EFxyw->F;_jE{%@H
z=ukTD6QG@5{PuIA2N<b#hDD=XGz0eai`QTzW+eusc{99ZnP6gZ$>4lqujN0bS)V-W
zAd`U={!{c1|9o+qG)adEav^g_9-7o7N{_hb9J$iVI-`4y&_i5AI4<?OUiy`F1Fl{|
zQMt*Mu-cv7>2(!=3|9$g(O&Ch<ZhIwbc`+|OOpT1+Rd}h)14v3W398-(F6miocJ)m
zKc`AM%5FIMbA;o#EeFo*6%2tH6u3I|3xy^;x40hZI1oMD?N*A6L7dy}iVRgWM7Df>
zmse#P)j4Q#=kZ^Nl0k4As?x)|1X<HD9Lzti&9i1ew~(M?)s1`(8hD5&3ujvLNktz>
z)=PHepP&|Qo&^2z-^>$99V_|dfBLBSgnSllq}+i=aI^yn<c+%#5kf2G$Hn#r|IPh+
zqy(fDv7j>(d&zX&W0G{!IE2k0(s0gb$@o6^yZSFw)Wm~dv1U<P`B%{&xexlXCL3cl
z>@$n@#K>$(h}JdE@s*;zIx@{xk8#U)W(v$J#!o`U+vJh^$(gXlGb;O>JiG7M_LbpA
z#-w*dU58UMs+bO;1E0H&Y%Hqq-4-q>?iZYHXTg<y_IXSSdT&8-0y=eQg2HF9GPeiX
z4HzMPf>se((e3%9<%-!9R;c;00W24@6r766WHh4%qJBwl9-lG7&j%QW84EdYVdpq1
zR0Z4<g^VHydIR~Mv$H>qbnh1}I<$bUq_8gj^^6Dli+7|hm+uehEW5v27X#2^FOKfU
zH*o^+*K!lE3;-qSHD6~EN=IiwgwZ<I_2@Ga^^%sUa-_R|r*81>X0k16TC(a;Ex<U+
zPVj@&l2yO*Zd_%>?z@iV3kCKK;rY~?Ji=5|-7XR4`w&lxDj^Q4rWVR{hO6p;k#2Qo
z;{0uB@1Rf6)9Fw6XDKw>v7vD@0=fKM6ith07WHKgE~&B?o};P*6u(5|V{xXei!0)l
zs1z05o%aBbgBkT`qCg}m2|AGIrgQT~o=mA7%1V4lwP9q>@`&BXz%e6_<8vlnbVz_}
zaZm-%2_O@3k=(O|uzkO@F?b@W`V?vAW@<XVI20B(S_0U72=>G^oL~&*Xl@GF%s*|B
zl2V8(9FpaIN7=S!vO7#(kJ`L%D_?t_sL;4l*f2f2dXQ$Jh-{koa2}rmRCSW6uLkmr
zQcRVd)u30czXz~h^;AJs7b^H<meoO<6PRJ3Z94kx6~%C6q?W<CCyB#v;wb&LFSPLa
z8OgtVr^ldY(H7~`fMaNV;$mg-pAAKYQU(@69tzlR(VT_{9qeKpk<le3<ln$~XL`)f
z@6@pf$}q!XUihFs_<a%^{^MbtWyK11_L^wW0wrDC<|!uKsSL(lL~?J;ES|4Mp#G;5
zj$ggHmFq6oRD5Mc$uEcFR+5j8ia*ld^bVJ>Qqx>>SsjG-Ytv-1V#83On3Fc;hTq#5
zVi9Ru#&1`R>j#GcS@N|mq1*>G!4$M$;^R(G%z)#VMhp5Q4inxOI3NaHU6Ue=5z{no
zb=*X`Sxa`Sr6mH#szUF|h&<ztm7ajan#!n$wEm?2>+HsEWs2PxiInmA96Z2E>;#0L
zYWaGZzby0!c|MP{oAN}fiw8a-&n2fFj1VYiSZMxD*!Uawoa;ib0RI2{gOpsd${z1F
zTsreucTkEvQxY{;uMX<TJBmmy>n6tuGZg}K$++TtuJ`Fkc@$a39Wgn(H`n`yS+R!h
zrey(bA;bGqw4Q3XX%oG@d7UhTNFOLX<QyD3ZjjE6g2!W%yN;^T`M{!EwLCd#%_7~6
zi++npbnYl7_(VjJWc4f|XHg|wTOvQ?xBkh9A+2=DO0I7wjAuhO8v_ly#~eO7^*Lla
zBT$jER+di1HSZ2ly!3P6X?uCLG|kgVexPX<Tm;>U>Vn**`L}T_QvDk@Ppqd?6r}_>
z@y*njBL%YD(1pq@2gf#@PjT>oUgj4hj7U^{)3Yv_QW$0i#@Ub31|lFsyiTK%i09*~
z-Af>vByb%CuD!p-l(hW*a=_3a>q?wAe!GkX3;{cxfL8bIPYJ?$k0h;^2IDguE&%{F
z8&h3r2QqY5H|8@}i-I3S+%%}7NxNoKzT4L3XWanSsW3>H%wgchYGArm)&4S}u&pPE
zPWCcKI_Z3`MOJ#*K<f6!F<&ddx<&QjwUv!rW?hm|YkC~nhK)1HwDyhxrzyA%^u-_P
z>FeC~y33RhNx(^C5mG46%+9wGWaq5n=oO3%q6enD=QD$NeZB9$_G-4p#AzkQ)S5|q
z5A5}o&$psP6qSV3sYnPGb(4W0o0IIww(R({6WP96Kpa8SRg{7xLY$f<{`AH*)Z~hT
z9+UxjLI7aR4W~Cp@a<8In+BBS5A6hD3*t#f*5@z{4_!aU+|@@*oaxi9NIUug7Cd^^
zg;iU_1;_9or|MeE<YfW}s#>P~2Z}S`iW5tB$v@0g!?6Q<#|T<;MYez9mn^{6kA7bz
z=vjVvr+l$7ITc<ih(EoSe{N4e=)~WMoFYCA&HXJqPdj85P5(scfOd;%+U_OCEnFj#
z(B%34&1X&a0~sBOF!|=Ic~k@3ku`CG+{H<lsnYV%@jq5w@rPetsDNA4N}cs9qh{=w
z?{A(Y?j|EFTnsk^@%2xZ7034@T&J43$|l=~qE~`PH}GOS=R3NnvL5a%WH>r13?{+&
zJp?R+C*+lE2*u<zEwqFnIcNQ7qi&dfE#h8*;nGQyK{&OQIEsh6cp#YJ-KD8>?nzS+
z{VP3l0cdoj6qoayP`eW`+4)Rh%}1Mqk>=fZ$wng8_a+B(zJujX?fU>orT(wsIfxGC
z#Rb~dCH(W4li029N-+D)(;(vx`44MS0oHHHQk`PHJlP`Lz@Oc+v<noBs;H|FUCvH`
zV3`7EDf%u{RfzoV(-b^^ZBO{}98@gOoH6ioKHiCNcf|7TPyJR{>WXd-ZMio1g5qZ!
zRY`jW1`3h8?NxPPV7P5EAi_;HW0ybFz~u1f{iM4G>M?ERmZ$s=N@pK$DGl}@v`U!A
zZr{Qss>M!&W_8rqk~Y_$RhVZWQv<wwx`4|nDk3*|sgnrU<XeCkzT!Jcjkv~Gi}r5!
z$dAd>s~9PjFLR+gVbY9uX39fTmWx;_UB_Qan3O}k%#+$mG0%Lr^vjK)MwpMV(io~N
zQuDZ@YbnG{Wt>+XOx<958|49hERgCLNZV;gn@3K%CRyapWM$EgTSe`RkpNW6j~8bH
zqtVEoH<s$Khqn{Ad-MZ|{(KH+h(@JdBwDOxp7XH5_1(I2LAlx1AyGQw04L!4=~3*<
zhGV01Egv2s#)d5F;}pfphLo^#==XAho``yMnnnl?H_WVLK;f>38(A~tVKUnr2Ucq&
z9V60I1i!_i<}(2%4>-+PNmWc2i0pIY%(6xrQr`(N;6=mwyY;C)>Hk2(di}V)S#&#?
zIP)>DIQ&W5_9m4e8Yx!p?_q0N$ljR0a|JGB{uOtU(5@LcA4WnZ0O8h2PZ`_v3*7$+
zmd}t}d*-VuRauI~AGUl~15=)gj5|dL3%iBjn=i#f{dnbhTqi&i#$(-9<vB|(_W+U5
z+ic!c^eVf4XVQ-%l0WdNA!dHb|Kma==&80$7rMlL<?UHgHi{jNt%{WqJi<A*r^<^b
zB^mBcEeLv5N_iedLmlF_W~WHdnp;<hDiuS#U{#rniYuT~n*_Lwhb9YG8pmv>AnYUA
zL<y(X@PR+!*|F#HD&oufox1Q|o_-5l@9(Dn!}VkTy`uYNgZs71YCGSB>3#=%&Fh^d
z&bcj>wUY8$T<asqH~B}Y7;}&2;Jm8&;j|Glg3$8NcBy~tuAcaCnBwN9aY_5YPeGJm
zj`gM59|NF%J$xuM!B06d$sE?GCuC4TQ1mi~|K%J7Dv9AL1Q5J09Ne7_!jOl@xS2Oe
z$?HKtsjL&ZM%qm2VFszYj;r(N4^c_e+SBZ3+fv`#X_I>8fA=dPF3<j#@)d&Gn4|&8
zsa!pBrCeZ|pZf<PAiJM{m;ho_AXgoLXros3t$uLTahW$T$^OY&_}!T53oiTFMS}Yb
znE*Q<LH*?M6U&F6ZIlxVpC$aq11l=>Dn=&_q9Bg&Y#vYM$$hD_;*O9QFJ_Ch!J|6a
zavVn~)P9z9_zNeb(TvTgRu|;+xZZ>7o}SW~<TlQc8wf#jL$;tio|tL`QpK=H0-_Ln
zjuX^xA<l$YrVv+V8og4Xvxl{x)#ga)A#ap~vC>b+P++ja-dGr7!7g(X!i>AUWv7L^
z7;?Nb9o9_s8@!?YKWAe_pKZ+Op~9*dlAspf3liE+KxTVXCd^1?mtZW(w&lBDsu+&H
zlDz@osV%)Ab1=$S5RS<dmVVNqE8sN}+S#iSv1Z^xqvqQOnKF*|DEDiQU0?4eNR;zb
z4blP<Li*#gtS549ZFB};(^D6F|NVMZnG5`h_V&xz_w;61RfTuD<cy$Nzi;S6m|w_B
zLkN#sVyh@``gwEm7!KQ65v;B|RPB3whI1}egYIqegbJW@+{p?6=I!bzW!MFS@v30O
z=VfXB#!0zgI9;5sRxTwiz_{td9i%z!FD(-OVblfdLZ3&W1+u_Ejxq}zj!YV6^$3j>
zOZ^A^MCkkY&&5mAnce29R~E3k81kzs(;fyK)z#9$pAU0|50*?I3jD7KR*LH~Zxyh-
za0Zn`+!@!&1fkUdLWI7{CTK^B&BwG$>h4iaKT$@>Q;Jr_8W4QQ8Gj<=a>Kih?`<?o
zJ*WadlI*e1XWKe;65qsW5anq|99lVuA@{SZvhrcIgj#VUtB>R3o8!YN&5V=|Nimq9
zl)n06=Pq09RBBr0LX}w6JjMz-=cMyNT#KMUg&#gf*!nASAq}K;W(SRlHPN-7KAgRm
zSeJB@hsavJI0$?A-bakUM1iWaNkxlH#Vx&<Yvz>Bua1KaA`4jkcE1iCve6tbXM$W=
zU8ZPb^aISbXHHfz5VHT8CQ_PGPPYpjBk^K{`m=w?!@ZKXsDWqhEij}*WH~P1KuVY_
zW*Clc8&x6I(rduCFMYpe)<BiX=PIfQmF7~^ZDgJ+Jq<9F6fs@FOD6!%7ZhxU8Sg*6
z`~Ai@^IB=dF{&?ehbZTE0VXKwToVc4`m*1NF`f6PymqNTL59j$=986FK@22$6K#?9
z{FAh&#lHcY!vutB|M}Uqq$TAb_4z<Rhp4xCycT6>SH$gt4f*{?4y8^klmu^ThPZ<G
zvXgE-YhiE8=eNlqtgwA(8U4oD@WAlg*!a^?xNJCMH=1!#OedJM7dcjZmLa1O#OhX;
zd;@Y*t>-!0GKl^A8jHAvCcrM@iadTu<ha#+%#@fCvMtJec?>R#H%~TAn^D$9ls39`
zN(I3_dlws7PAczBpQ<Jw^YoZ8Kf%CY9Y>grB3?nb4MQhOVta;DZ1w$nM?9^{2zA;`
z&lE`*_`51|F=n%|<<+=kRG*KG9ij`O_|s}fp%;CE6#i^pcax&SUjOJ=`B;5`XPb2%
z50<#22>d;*|LrW&T}{F)Q}-Rj8Yf>=IQJ?jhsR#y?W>DRiFnf*ntUIvt&ohN(#VVZ
zhFEeu+zkItlYlqCf@~}l^pVFOAqqb8%o=eXVSGRDaZ|I0hmDz|*^Nu(hs(i%vzvS@
zklT*<KFyDs_DmH}vX^JAyDF&h!Wc((g#9xhzmevl!&U1#>eCNn+m#nANv<=GxJq|r
z2S#4Vk*eVg``1hB@m#IYDf=r4e*9l8Nq*ioL1}Det)sNPBPRcV1?HJ*TFv!rm~OAQ
z8nR=LltDBJkmA;gNamLMauYFUs6U_9UP@JUsD;&PUc~9L&9eGf-O^Gku(yeF)4@fd
z4WFH*)rDGf0m>viZJCxF%A|Ppa68k$2+M2ER4=HVB6HF)8s82Ekkfw|wv()ra#Gc)
z_AgDjGP<O1?CXVaS+=7kf+-#ChsSQjqt7DLCiG-^%XG#7985i5G7vMD9z~8q+3b^t
z+TcR%kj6$hqgs46QngTP=BRs$4df|gW$p7U0Mv7K)~2A3wLXL>;FKtn#2Lk8el6u=
zrhi4Cy>Dk4!o{03&7yGkFO3laLZpf-L@Lp%%vYvNsrDO|wZBeInkU27q`Jj$TWk5c
zw4?Gngj`P+1x1RYZsclP$)1QHkb?oX0m8~mOW}>uqR!OL{*Yw18{;VUA8%U+;fY3s
zW(DC@0MKS--}R4%!{YJgBgMXjl|h+IA-O*W)(KP;@lcC+1M_l-^`OeMqSUfMdC@0;
zFjU4Y(k)NrQtM8i0EC|P9#Yvd>q3HoZtYt)S>$82@2~1NLz~p`Z`#KBtnWQGvu*!4
zH~G&w+Vn1=q4SYNb?#Szhl7CDPJ{Z8C~Ca^d)|PQrg%oKdxHR~Mgw*#Bt}!=+70~k
zTjRRnU8?HMBa?aghKZbH6k7n1GIiwWg2zT04SKPnR`~Da*~#2gmy?U=q2!@Nm2{{U
z<$E^FDK^{n?k|~HK+Uf=1HRn5YlXZ3tq2pB@@CeMjp=9ORFMEq4~QRR?og5XF3P>@
zZ%ol<ToOx$rXZ%%yT{@X&HU4%PWDLsyZDMjlOr%C6yrY+Xm^IfTQZlT`(G;p49ON=
z;;s~AUC}s9x}Vgzgo4hZB7z?J%WT&Lcs0DMJWH*)aPV!FgP8u4E`G~_0ZNc!n!xXm
z$1)W5$V}9>H_tiI4l`nO*F;^5y>L5>vrQ96)+pJQzBw`|$*@vbPD+J4?NjMh&161S
zeTVH#Vb3)N%$Vv!8dgR>lhIUxe#^kiH{aJ(cEkrllE`YI2vBwW8Ah_C52=;53ftXC
z2%1C4MlCN&w&f;sUL-LtV`sy{2E@f1roC2Oj}8mQ{NBFzDd0$_W#p-@yl8H~vsDe?
z8u4FkY^%(qTnMy1o$IP%a1v=rulUxlRvfu4g1qsEhXF|d9iDF(sG>AR{CqS0jo%a1
zJLqlIDI(+1(`Y1z??Aw+ee-UG;d`cTy*SMNMM#-aUPkspaH_afDJ#9A`$Mrh0UI_9
z=gdyTJCvDQH7=aH!Td7x57cX-KHzlOcMSU(#SCBEg1aWvN^NUonccRQfoD=T!eAAj
zZ#7N=nkTwn0n?!<pUv<?)=3LA$-z7w$Q$hoeqj-$ajm(#b|qHR?^E>AR04x{SjNnY
z{V%+yda`!-V!mMSfmUS{%dZi2ZH7&ri9!W^{l%TWojj)tr1|aWzZOg3yAkvdre@pz
zZP@&};h1o4e=Pb+FHxl-jLOm$de*T|Qcc!>U6lvi6{!A|!6LyigTmWu6KnVGT_?ia
z>oddTz9OImTm`t?vBQZbbB2<G>C7q2>}7oc6X^GT%R))g-MR6kSEz=f!Gvh*_H=nZ
z&ia1!A)9EmsN;{Hw@2$cU^WH)1}ru=6t0tgK3^r_L<qCbqIDJ?JUX!Lv#y+cn~85B
zla1_}x#FJZ${lyiY+|lJIH$zEvm3IG%_g6>jmcK_a<JMupo*?Uw_%}kSbZ%!>&E$}
zxCo{d>!Sc=K$^b<J^ha~I+G2VFJxh$NDu=L^foc&A(}wcxP^)6m3Id(N3oQ_5QwIX
ze*bI@btKIbAwp|@XZlVWR>97Xd3Morefu6cK3zaevi1%x1O$!yEoH;ajpdJikhrUC
zYiat8Aa(kdIci93CLv*#pb>lwa)Q~k+`Oy9zs2#2uYB!`mE%iWH)i@&fhjTjw9i1H
z4<XOE;a<Y`i4`bY2F6%kTXw_J2IoOx?6Wn%nstO0zaN-pX*+V{UR@usq_h;2_sXSF
z==Y}=Rxlq_029=?Y&>OC%RSJd^yPH9wQf^LZ{-Ns81o72q!k<V^vW(5GGyBbPRzQO
zA9zeS<SQ|CHEhXRT@I;!MRsyKGNX$-un)=jZXyxHcsgNEQ~6jeb+wo;Ko$3F59QZ#
zHZOc3&{G&|5$q|&$B1sFO9U;m=ut|HvoT^Zr_^GU!Y<lb^j7V;y6L#pqXnnq`pX&I
z1wi)cS{Nj2T9z}>lyG2L`nKx}Z2}7R0J7GF;c`S-#H)nE1*W<m-I8kF8v2;npS*yS
z08YOWss@>_apHT}=fYHwN`OZ^X@(y`lJ(P8)6md93@#=jcvb>}=i)RytzSM9{p_)n
zMaF1f-aM@2^T5cI_57T|21o^}q;NzuEeabq4;|@daImpeFs9jTx?<~*6=}Rau1E!*
zVwVqHFi<9GUFI@tW<CsyKY-vcic?7;WUBy5#(L&5PDUehyQ-R2G219e1;g|XW58>Y
zT-IYs^`hd4AJk!w;9Pmq<hkyISVBX(Ndix2v-Z+sBIxn#=ETbMTd-Du){I6&pS`D=
zKTeABQYy|#UxAjGJ>|Zlm7Qk6r-hinPg?ycwXTvo)3|OCcm$mS<QZ%Co+i*5Q95cg
z2pn24D7T^J2e*>X1QWjnT4pgXw$|=)H(zG_pmXh#m@XyxEsO^Es+VO37udu6W9zX*
z)hPgn<~{Znt<Gv<OOXMK{GY2ig45MJTB!T_s1HMR+UU6Uxhu>#6mmV)EuLKF<Mn#^
z_Ryo%ZXI0PKQN`}X(%!UGk`<H&a(&Tb#uhVgvP4iBBu=>UO`7G{6q2wd_7aN)%nk6
zgrP~zRzu=8Zw;MKdqbKt%=2qdv5{kQ#Q-+3vUDA`a5tFQfW{2j-+x;iSfM%(*OrUC
z&c+iKgyL3}W3OWbhS<SqWXm*=9!r+qC>NKVVL;P=K8`8%Y{<!oz3kt#LRJ7Np?8JM
z$R^CkxKzRj0-+y`#1RmHvB4lpIb~9Jzv5)I%gR08<P{)I1KJ!2k3OEKAzK0#jrK*P
zm3pss5aY~lKF2sZojT0OIcsS$Q%9kgYwc`#l=u`-HW&z#Mn(&CbpR#68Z;|ttQj?K
zz<%5hd0Faq0|+K|+pNiS&%N%)q87Y~d9vI0d8+@UQL!vqaxGRzG)NW#8iyrpC}*>c
z*5pz=s3^9xP5?b9)Lm|rpA=VZkgeis$xXg)cVR4RlmoHx6%<(13xmZ~CL{wY|Fd#~
zZl<^GCVzPE&}Ghpy16)v(N$XXWl1Nq&(K&(C;pJ<tkf@c-{+|6IQ@<xP9?<)L4U$@
zlh<r=X7s<zTL<?7h&&POp20P13RU#_W#;iaIxq2{Jfs`-;>w&<?}`}8SZ3h{Vn||`
zrA5!r2APAfXT4~*HbuheDlswg0YUb3ZF%FT)TE2n)oA(PRBJ(?fN9$TjWj3LeT{O+
z4{E9rDMNq=)cO9SQt77y*my{2eBj;X-PBnIw|~Chx|-$7rz=8_UCd!hE=FtvZC%L;
z&#s1te{^mFPN6q{n{B5J(Y+HC#d7|aB@gqgwwx=(+uHKh$|FZ~9!@SS6lu$X=Ne}$
z%Qw}uj&8~rna|f@m}Wi6c)h<xyq;{3H4*RtPLvwd<N#8c2>(^6F3%7O>Ea*5rU3s?
zl$j!Nc8oZfZ0yh4?kt^AxEu;4@ML5ixIe-a+at_E%SJha7n*Wd;@O`o7qq@1c`l7p
z%f_)mTE8RgpLp&y^Bk3?6|n{Vv$-G`h6q1ZTSn%>q}bBZOg=&D(|2r=b7OVPPkgs`
z*KKK;_j%Payp7zT#{Nn_GH=z?NnhNbTi?i4_6A7=)f`I8=YT(Bqu(oLsv@Ct`9x|S
zSibn{q_{(dS-C|-a~k#_^9CZk!+o<<GD)9Z4^^Z-APv2%Lu*M#DpEfaC4c)yM}z!G
zRUEQShzZM!yatym0URLy>3Oh$dYa3ZfJA`gx(T~lC@=Z-ExeCbwUy7P$Hrl*C-sa*
z1gM<LgO}?4?0^)q6vcVh0Kvd8@^|($5NX6G-;5XK(Lc!lHsxs^H%G714#3tKawD;B
zslJ;Qf}|wJ*1a$8PLO!0b{SHhTZXcD7u-gK{ox=HIog=>%PppWxau{gSr=*!@3}cl
zbl)bd*uHdRT+DcLXcTv6Z`^{&m*#viGIDbiAaB-X$i6zQ@nyX+^L}9|b@UQ=p3AVS
zSLEwvd1dL3ZAamTjiG#n)JT1s|M1xue<%t9l?{E7PG50{pH8;~<=_qEdS&k-mpWXx
zqda%_vQw30+Nk@Vm058gRX!?gBgMLSyRkgZ(>^?Et<bC3qh<@=^mAR;)&1oZ#a^wA
zq#Xu-9e0U6w8f~ESLpC%C&=cId8)9GWx3gcK+UblDJBfdsJQlRboALtZdk~A+l=U_
zl_96d(2df#Nu@Mq^k_^}bj{k4HO7D|kUa?M0>qU%BKTSW6b_NKc^gMwYm#Cj$UUY5
z62K%)Aq3-Xc;nt^zdwAK`7VE9sLX}rWvn?hODtP-T0S@nhh(4*&r&(#!uXvnJ-+4x
ztV!202}I_i%9XsU=zSmZ1Xaf1wjCE+rx$Fd1@aCw1qC$Bp{MMiV_>JlQgBU1i9ny1
zQ!V2Y68GA2<8*sXRYgmRd3R{LwJ{0VLBIq(KlQx3${gLvE%7XKBTHC4(fTqS?WIpw
zf|hiU$}o$keat#0m%t=NW%>(yQ!aOiM=3p;ta{{|1Lr68&cxYI1*Stoo!^Llx}|iQ
zQWCp#RVY32<6*~k_$#iK>xE&3JP<h!0h%%QbMwli|Glp-lpe>ks<K95V?qTi^P<P2
zSc&wEWC?S=5nV4jtR1HG0TMb~Py#v8{2TpBWIzcMd^O=hzoK_*Glm{ZCIJL&BnCYP
zI=2>2&}4?_%N{}UkHEqEE8wH+_NOKtC`6?|=l=DHctX^~)ms#`{|X`c)yeRwlnNGD
zY}S{|Y=;7D>g&it3<C8jVWo%}VM}I>|GUHXHYg1xk`sYUUgVl_CTb(mhjDK4*-DR(
zN$f%`31EkU^4SmcMl3Cb1&}FUI+@~rM{~$jy6TJy0$A1&RN?(-t<h4?$6((_i=v06
z-KbVw)_JWc{)C6pZn9sMH-!Ck04w90`bqOyXFxRO^BpBoioLgGqDEY-xcc>wXcwb$
z(%~OM9zLk9x>kd|zHT#L|DlHbq=2aX+5Z(9#350FT8bI9wRbD^@o|W@><5OBv1*ep
zhuMnw&%}F8^890jx>xegtum~cw#W=l#caN)g@J-;ua@lzo5OJf5rda;KtL(bzvFL_
z0GB_Xil){ft^na3b#)t&cx4Y*itAM7ivs278a<KXGyXg2cGx2{AAf!5WR1pjGT$YB
ze$=+YvfDD=`PISa8QE4+-Gj&k%^HnIV6~o0#NDv^a9KfJ6O8;~^b1Ulk==A4p=bPI
z1jLJ66}rGuw}ehD{7b2~tE@Ex3!HKy+f7AG(jK{Wj7y}TRT7QL7xrCij32iF2ma26
zTW8Vy)8$5$#h8V2sDy@;rmB?5ou+p^g}=M|Jh_6pUZ*+y%eQ1cp4_j^dmeV($K+oH
zgge#YYl8{dk{r<KoOX$++{RspCpP`MfqPp}Y)GqLfv-)J!PjaMt3*`?x(j+bIV;p+
z4+B|Az1QfBP%$M-pkV@rhxBf(pn)8F8zND0UWOl>?pyCR=4?~Bi~rK7;RrNXOm05#
zz0FyT@;KxF`pRasQU0?K`1q6O+1shEN3bJ>u9X06cT!gHV}or2Sm+%ms|3Q@+prwJ
zuD@eYf(?=ul^pZ1c&*>Zk6>$G{u67l+q=Zz)1xQg<iYCXVRFC*OQCR;i=OJTdHYZC
z_T_ECpiNXo-5QU7-t*giJ=}#}tAe;`Zd^K~;~G%>!uVVb-6y%mmC;)u6$<%0v`8dZ
zeaJ@fG<fki><5G&?}DXJ@CniD5?psYuYT5i5)dvPC(G&hk&KAiw5%cWCKBI|0-|_E
zf-=6L@<eECy^SS!){gdHlogBT;8hb6I+8?g&(TWS%Z_Gn6hGS3+xh?M>Kvak0k}3^
zZLT}J&9-gZwr$&Yb!XRRV{@Cm*|uGqZN1N@cjkHigmY%jTxaI|u5&eXwvbx82R3e)
z9v}iKqE@p|jre?elgGs1c&hjic`;Y5IaAds>unMK=a7{U6@UQKZy`r(=|YsyQ|?8C
zwXm0aAuAOD_x1Y8`^te-U0E?GWy1wN_^Hhwy97kQ%D;0I3ko#IP3t&>WF9H>&C%C%
zYXk1Fx1YELJj{+iR|@<DhZind_HCy$ZZQ(2OY)gQj&%!j9!z-ePOl5;HL+Ih6`qc8
zE%sY|YV=zJOpKoJky$~ea>%F?>=Wl2dy$6l9~{tX_G-=t-(+UAt3awk7J$&CB8-#>
z%~$tpPqb`Kly2Q<hy`Y@xxwhfjOG)Geb?nm{``HGx;ytAC$*kAm4~fd{xg|yio0Zp
zuA-GUG;qFv`E932L2BnS-;RypJTK#kwh;PMi_7pVENee-w9&WRD$l_Jsk^X~HXZGl
zVN*QsT6?vQH5N{zRP72`%Ml;2Gb~Yt+a)UGbvm66r9+VoL$i`}&nMw*fg}Fc?C}$c
ziOGZ_JPAyKwiXsB%Yp&2z3~;&ek7G}uQF2a$bGzzY7@>xwWJrdJ;`cdrv&pR56IX*
zm?>X=5JAGG+^C{^_doFmcK?n@FqU$%mj4YM&1o>-wi;}O79@;nau8Ued@49REG+O+
zyJ^)FH5=e}-;o(f`SBK79UMUIdt%q2&nePBk18c~&VMHto!(6cHJ$wh4_tPFc@hYD
zF45MHupmje7NDKUX2A-PNA`k{--VtBTab3<0AVk{+V;_=hX}v$2L2FHW#O7Lczs)k
zAiWd4Yp;Jywps=z8U{74Ilv)RK@1|Fm;1~hK-A)RE0L9vePr@Xy+FM`&szISmYd-u
zdE|C!Dtri_CJdLCsk<=YjKcF6>!<Wr>?1|5&a>jGZzjE9hy5rM3VVz&s$bUtJlr4M
z(SrFJN#|LJkZ)^tLZYvGd9=+Sd0ii0^!EOXukdn7<5jFuDf24_Gt&B-+VwsjJVtvU
z)FK@{l7WMAw{Nw8T#v?7%Pa@Pk$u(zIq|UMo!zSKG+Fe*WE`R`Cwdx?P1PwJ_TQB>
z<4E3jkKg*p<NT{0UqA#?_y)^BM`3!pS^YYq>uSshy2zCG2BANg%iR-E{hRny8q*0d
z8<)1y^z(Pz_{u5_@nlE&Z^OH6rn-fx9LGG<xINW5D{-+?#m6+987V(Z;z$6|h^tZt
z&uOpQq=J$aFdx1B#Z@fdVo{~xU0?uGdv7H07H)=!m*jq^c=tOpIW}D(uv1t^Ji&6?
zCa2bwT~`B63bnfp>lZP)?alc9Js`dA-uQ!rmUSQ%mC)74^g;4;D8{z<;^o1vJVI75
zgh=X_{lK3b#RUq7CT~^<N|uL)WR4j(-5Ucln+E#xqSRd?D$AXwJc#_(OZJ)n28M;9
zm35+tg9oq|t#l<w7%NTAs6Kqw#g8@PZ=DtG%K+H!hY1zNjTP|4Oh@V&+LePhR50@1
z?4px8kS^^7Oki_1fk1^Uqof)9TC~x*1jWJ`-s~WTISpuhxQA9WgnEohy|A#SnqNCB
z(s7AC$=@!r1d_26wvOhO{GLrmxF~wp+>hm3Aeu`7#yA{S5Qzd#=D|G0GD^K;V%E55
zJdtDX+k4qDPRt9X*|-}KNBCHyt;b9(jHW8+bB09%_OPOj;qpsim{IRy=f7UUioK%d
z^grs1^p6{QPcLGkH=fh7e*fD}r@5QC@&l<3jygCkctcN72W3kmG!C{*P%}4%1j&c@
z6DP2(Rgk2)YT*ru4_$eR7Y44-%7;$J50Wy6LjPm=DaJ)hOI;H?R+Zi&&X;kzXYU30
z&U#`mR0)=>SCpltbW|FF**r%0q+6fzI(OK`l=9k)Mc176%;4naV4LZ>$YMZWq--}r
zdW0Z;t8^gSw>?fZ5FwUntW**V^V4Skj-1`o->eNYoWEVJOiD!%kFzLCuRmt+x{{9j
z!8|<9FBcgT+<oU6|43buzeh~Ra)d=mVr$Z-z_NPpfaf%({uCG`QsjMor^^QQytY^S
z&&-`X45J`kUMP~j4b<g$x`)jOOw-dr^q&Fb$hf4Pq2<a3$gCjVw}5%IH3+++Z1aad
zUGHZ>(*uoZ^6)zqOa1_0sOs@DZIsE3mYjObyEjNsu(LdSdUubXev$uJV|EE(8kAA^
zyF<fHPyI&5D0}Lp0iuD0S`SSTp2hs@?l(Rbh+^AH;}nC8=rC}s&+(vTlmBL7ft2=2
zKh!IM!RLeF>+ZR}{}F%Ma*CF4P=huH6u>OluI@}*2$hTqdi8)3DO@u?qLu{A(rwun
zVmG!Gl}wJ?2NjuXe3^XLMF9Q>hFTS%0a{YH7_B`Ir#F}*^WY}22=*3>`^gZSX*2$+
zM4o#5++TQBsG800G*e7ifVT@h3fF6pW9^-&bu;y+#KRyEqagpz{&K&fjawZ>QmM;x
z<J*$OJ@xFj_)7DigP)@^n4<p8_M0sXLHK+?xTIdjre9BeK5W+11XX}OoqBrdIg=5O
zXSjz_l#@`y@V<qppp4banKI+-9sq<xtok}Ln*XHW`%Vl{Q%^HcOYeZHCd+cW*T+VA
zY`^`MQ&>(<0?bwG$%;!5yWpjfSvNRH2suf!(o9Dl)u4zSzFmDU<jB0*b!`lix%{yD
zb<$IpUirv>$cVLJ*<<RVLZBb3r<lh`{-u>Rl-;eQ;f=ZyePMrhZMfH7KFv)tI5mD1
z{12&=twjagj=0V0ld_Z>U(eE^lhG5PFE2-uknOLKhrwwTY+$w9#E*>ggE$ti>e8o1
zT}*;})VhuZIXUys;S!B2fSQLdS5n|Rp5lLy1aZs9y=ZuI)1T21?l1u@j>7oezA8R@
z-gtRL>vN1uht(u*t-K_{1p1|R(8bbUPeNYOqn?VtnI~crXt2Oq2{OMOEn!Wx*12ND
z#rv|tEIG+Pu)hw-j8o2s+6?=^%o^O+BLIEa%M}ueg#-`9;WjaMc9!KQ<A3-mL-^q;
zlxeTP1mhO=sVy@lfGh=WoA`&n7dOh86!(rBD$9mddQZ758Y0H{sGT)CHHO_Tt^jOC
zlclQr2g#KoCTqJ#C!eH@QO~}_;bc1wi0;;irO+gfz%x`hUN(Euy0PILqn5<>c%iIa
z(@ULp^DdqnhAREsrlP#XfoNgSKSEbY&;;+Z#FlH%AEuVz|I|)@8v!&g6fa#jfp<UL
z+RVvH?tL@{BzL5Fp6tns<+<W+_G90SGR^qh*tD{mTQ0VW<Ul~vCT9(hK<!yo=^FXP
z(e|GS_KH$|rN?r*e$Sf1w&QA#dkkVn)}npj0~Q105(YBC>8;X^#0m;-73Edt$dN+s
zRO)KHix;^GQF~N(7~0g(kD#fE3-u42+2Joy;v(eVW{;XQ+;{Bb0b=2$&qLHk`SLFB
zoWD?TO!T2Y?4+DDE48AGPZC982w{PBmEKz@Tr$tPdU}D#R;D(sy=D^0>fvcsYrDp$
z2uv2W&)&#9a_|)-@837gznqN!nwsxTeo6FTK=yf(=#t$ya{IU-wXOjWp4nWCY5KTJ
z8fE_EXC#bI-L;!nGyfn=z7QFzyU=O3z3pEc7vt|k{!~Le^;J`?dLFApcP!sC>0URo
zpq{31ic8IP6Pu4aVU`&I2V)%Wmt|tAF8&PUY}fzYvz;v>^U}ML@pO(1SLP8Nhg13V
z<4kaz*DH?V`{P5<;T9h^GDy9J#DI@TUFRv5!X256QC#i|c5DHM&1L$Ft|4cVe~DV%
zZzEYklicqxX!_6Qa!}(?D3K&^${AMnq8omJ^|}k5=-*U=b?P7*qBRu^cS?P>p@zR(
z&Koa3TZ`0PAttCTz$#uPt>9;$=ts!h!>IP~6vrv|>CHG}C*I#o>a@0y>a$MFpuixZ
zy41c+@Wxc|6{Q5+t^e8P!a|otgYp<Fz}jIw!~r;^x4;rzn4c?B>it4LAh_vGR{ax+
zxJ&N56534gNCW3o++Q6UQyZVMqKSxajYgxS?kIAy`adRPo^*Y9988}{BYjLzhrPvq
zm9?i6+^#wud`RXvA#m#4Xf_7HjD$1xVa%W&Jbz?7C|ZhcK~AhbmC0YMXa;nq(-ZI<
zlE=2n)9E;L-zQVrgtspkeQ$;ZB-`;LSS;jWp=O2$CZn;0zXq<VU*H5Th@%amoRzs6
zf$9?{ufi?vC)Ga%Rtgy{61!lpwu0MWn#15LG%7Ow#C>?v36a!tL``zy&2iz4Qx~QE
zOKXGa`mhY6jULAS`i#D06(ORmZ!PJvRl#-5xb=zcyHqyg=Nq)wzYYf)YHAUQE@^-3
zc=9oYetp%XZ0yKBk;V67^r-x0apywU*PbL2HCPK-x>=3e*#z)O;w3Sr>0vwlQxBty
zq4~?FSP~79%AV{9E@0QvGC+oY@(s4F>mkgSj1zeU{_e5C&#)?1x*aa)MJqZ@P^|AM
zDhBHkH+eL{51!FZM<^n&J*w{XLU<0p<|Lghg-D_k#kGwqT(6;phpL4D<q}Xf*+?88
z#2cMH=Q<E0O#WB4Mlb;1HdhxhK*4D(xbvH~mi{fD;hqZS&3?^UD$UUPS=Aj>)Ex3X
zMUiAQUnGF%A9>T(59MC?#&l+F2-5xc_<jL}C$n*3!TwJ=#`eXj^Me#2Sz>x<^hbs*
z(UuBQSt~5?RhO%yYQV|6eI&6j`lC5L>I)nBO<RtbNT#2XO^<{%yHE{zic!)?Jd9ry
zK!z<fWw{w@#)u#Oin+fem@I(jm&?V$^RMox49tY@dy<b{QnlZESKjD+bzK~f{VSu!
zpR06CV}~I9aeihFeQglfbN8#}!$nc&+QkM$QFg)&oMyVoM)?BMUco&oaup&2*m0df
zpp=*%v3gWneoH1db#$XYsOU=5o)UNXz|(>c{&rF=*1RVpn`x9@!bwHDm!N8_FkF4R
z&3h;lAT*o}{I27nQ|myOKyuFZ$E7slC8!lw(jw}2Q4ZMh0+**DzTW|g<wNM)5`t#j
z=^6gON##~cQ$rx>aGUBbs4cD&O(mlD_|0(K6|(<T9lIKt#-&1bdD;P?@rJ%wskukZ
z2>J&7oMmh@ld7_qYff3jy?0mO|43q0GhtaeM+zb{E@!1L%iouG+-QgsmkRo*vLSoZ
zlk3>8Yq~!{Z+K5hdkhn2v8!t)ig>9Y@!JgVr{}^}Z>~c7%I92;Q6-r#_M*1b#4YO=
zuN48ac^4v55hWt`zEZ54q}-MsFBSy#dw>)l+K2J2c&_{OYWYUvyVelz*8)dR-Xs;7
z1Zi?-%kSf(>ZzPN?=VVa4y>Ke9XFAV*ilWY=T^2?X6T{Q{}6CMz5w}=FI++OzA9Do
zp)xx}Y%xk&3&gI;iX!Niy>#H`%hK(JU2IgJxwYTzDk6&I6GH7pK<+qAfUzX{FcIJY
z7jA;1{Yh_@lR2KihlmY6s<x&Et6bBneSG=GVOBIjE=;+DiOaiff$0zzQ&v@u1#w<3
zn5@T-WABTf2T3J?sYeXXfhla4=@^O%6#|tK6CJVbzoTSMpubaJ$ssi=ItefT88MOY
z^(=j3+G$ghiHgw5G1QEe&(|!~gobVA%#Grwbjz%HAXU_^_9Hipd5S*J{-C+9Sdd;a
zA6>w#ZMdL@+8Z*ae9-oe>Rm^ABVE4!hB|9RcPkp!;Hs|o#FJ;Y+`@Ra_nka08Mecp
z6BZoh9BUd@<%}G9mPD}cbit<gB0N4hSz-I)8j(MQ3mO3)Z-D4|LHxdo@Xnp+Bs1=R
z@-r*sae{h}jd11t`6JH?Ui_CdJtd*830$hDEi<)TO)&Wdk*AU%zK_4G8kD|O2WroD
zjEcg4+7$3|dyv}Lm%?fJeTb{kBT;ip(HN(Wz~QUr9I-*T<DN3q@hHPDo@&%gWGp-$
z$R2pIOF-L^@H{kUMl4j8Pu0bI$~8G?NT9GypN3ujI3{C3pw#+4uSUaEA<zGQ?hd@y
zENw5lzulL#wFQ|@V5gh=eNyzY9%8=;K_F^%Vdvut2T*2pbqn3QT~DBM%>57)1}qYn
zh7cOw6HZy%Qprc{%jY^zQ<+}wmDOy6H^VU7OIT{8;}XyLdMFE`K+xe0Y|8;G6v$y)
zR3cAd4@CMR^FJ%B*wh|O5od}a_TXAQ9YRnc&0cn`5YjiTn%flB5GfV0;U&Aa`pYBb
z3j0JjENS|X!?w7=;@8A_1Oln(7d#@GeJn3bmIsVOSMXkRc<AJal3QhG@%yU;&5=GA
z8VzEW|K`(-$}RaYVAIgbNa>i<Nqlo<a45omWb)IKamD*(c%EIynx|~AeuA`!yC^XU
zETaTjsFNFFhj8*I*Yw1N<30lP?^v0tAiZ%{g2xZ<c1b6Kr`^?eKM>y)ttL{}ZQq!>
zG{W{+C%{R+k91{d7b#(CrcARini(NKVkE_m%|g4(GQnKBrbsG??e&F?Lw}SG%nAR)
z=Z04ZBI!$gzBHh;$s5ey<dH~+O-1snC?(}havlt2vAF!Jju|$zt0Kh)kCye8dz<h`
zP`FrlwnfG}@i*3~Lb^^7>n|#bJ2EXa613BXoUA4;&nIP?Sfce(Efcqr&-Ey3%NJYl
z40gq?O)!~>Z10S!oF%uQI>Y0|*oxy^!BF;FdtUQBJUvbT()M^>$cCB5Gd#W9558gs
zW7xJ#kRK|p$=sCF6P!pH;c3Qi9EigJl(^-`y)fKMnO^x|mA&zFei~!>SY_B`*5cpg
zrA9+0PXu6z%y>@|YfaxCdJR|<Laxy*zQxsJZqZvOFIQz~J2zr<$cHDl6GUv0QvfM+
z`d@3K{d>rzZg)%M@bBle<i8yimi~X@bfDGH4M-$F(XiVnRK#4JsNwnzsj|4JKRH|>
z515=*IeO-t;J<*Y=N@X6VG5-S9ItEVROGaF$g14iQ}V9Tp&pYx9s2K0>lezZ!_NZk
z=$2xncwf|}Pslx^kW4CI20H{(P5g}Prl<ve^sd{H7Gh%W4?(gu3#jqTdeE_`6%YVC
zOrr_$hm-V)T@nPyxXzQ7Z$0KVtE9}F@-%LBITyh@AlSb{Atbpj@VKvDu0lfw+@eeC
zNR*b>7k<6AjtL}f_u2GxAGWnU2aY&Qj|S|gC<<D|%6a_Z@)$QT>3qGR8E1kO60@&n
z4ycj3vWU|9TvBc5RCQu=$VsT{qeLqc?_8s(93JFNKh7sza9f3;<U~);&Ik4(4pUEE
zMtGirsUgUB9z^MNyXm{Xc^JKday3+pwa{PDg~?ZxJUglQRzBF9XKeMTbAKz3IkUn*
zctflSkLl_*GsZ*jwTwvnK%|h<%H2K$R6ezhGMLRmjEm(}KK5Ya-OqIc4`1}_B812B
z)g!l*D(~A5@go1Y9Wcz7t@zC87K*MLrPys)K)y?+Uw1y}L>tAjLEA#NHSe1GsV?N)
z-ayCyrP-2`yT@E;8QWE`4V!lmSrZ6^uf=cz!9gHBd4(PkbI4@&B<95lBWkRoGUpLi
zbU;5b1MJZ1`?5$ID3YJxU@Dm`aDd96;WGV^t{Z@5>*!$O6b9}L=j!Har!i&FH<zf;
zsa(zg35?u~mxP$zWuqR=!lcLYl+H3S1@5FvQ~Fm*#E5OOT6Y~;@?5+5Ffh8Lw)5Z8
zay*Pn@ROCnJ0{5uUi7WwM>Bo<9{?Iq=Fi6gYfS_Dx(W_8yj+MFC+ghD{x>=|?K*|0
zu#!yTd|4G9i28_S@u$|GIpjqKutXPK&y7bfVArh2J~a*x?qX7hm>9LE0ZxWYbFbC4
zkj$hLx5@3>J4&+g?w3BHFj8N&&0m$**z8lgy{AzC|J!}lj97Z#t_&Ei&psZ-j|mFu
zNibpzNjgnR#5n~Cjz5VC2UoO~h?B|D)e@|&tFd(w5K?Kk9AKFtYq=q><02iep>>+G
zy#G;CyzDS$qu?QvoLGBx(T)c!9A%9ID;QJMzG4MC{KsX~(zeyl?Gf4PZa9!23-hp=
z6Mj)7W=eUh((UY#ELU%-nGvWvHJCBb!9>lI`2OZ8|1nt4u)CdK3w(3CCl{c;g)XWU
zq70DJ+3|WQ;$CQGdt7TbMpgzu-;MrH*mJG6Pm4D_8IX1@VTBSf^~~-9WjWH{!Z5mR
z(jSWFwvJr0aH&q=xP-*rR+_=NVe(`Cmb%-F+{_+5YHo*}sj1z$u=sh-a}(2uH$Oa!
z-K;Sd0vqX!;0M|Dh~Cz0IX&!ZetcOE&aumwZsx+vs2{9(q`>6@4k^0FCQ*SdR%lJ2
zu5K@OQ4g-8A-d&dw|=l8@4_a1E7)oj75$w&alzbpGKVo4#(eP~J9=tfqhZub@!MKs
ztFZBfeu8HA#HvU5t-!%GRr!wi;`Ev$t~}}D0IV4fg^E+z!_Fba0s>hM!7G#@(})G8
zi5(L;3*;2_h+kXz#sX_F88rj%@IKj2dZQf}vxb`6RyVRXF>)&eaK)(qMB+6%qWUOa
zWJU^I&#Nx7LSWST1{k_Tld<&%?k|$!%)pj301-;Z09(-QFtR*W;*|(`gp>zgXG7|k
zVvETBL0Ajta*8+kp+s^e{Ox<!TzSW+%1lZLtup2y)3{D8RWAW2l+raO_>k9+q3+t-
z11tI-5e5Mr9p;t{bmM)o1@-<{rTn-N1GeR7O;r)ab0OouVT4Tz7qh4z$f{c9=SKq3
z5fTWh&jRiS1xY2e;2Vk4%G)<RRW{~p{fNlhCx33tWiu5ygvZ8=#1(Ipa(*`%f$7!Y
z60V}d`ks=xH@ikN*S#!0373cT*+(^3R!}R_lf2fI&2+m!mJ8C<N{mNJG%UWFqcEcK
z$3r0yf>dH^lV*4~G<o|z9Uvl%5@DQH&;=Q{`7IibnWemjklcpD7yqVR4=XqqC-qMs
zb2lJoVf1R*@9_TA4nc?w@IesrF`f}y?s)Jl-nhGCtP%}e9FoEb$_iWgIICzVf=Q}o
z|4*9++0)5e*V(HF6(`dC8AHmi48z}kYSp*;fIrf!ojsKjnHvyYyhbL660gPFC{-xM
z7SmZn@5ZM)x3Ih|nPE>z-Az+X=`6fpy-&!Ji0Qu^$bom8=9{XGvV-2)C%N)k!+4{!
zI}ur98#vNCSK-%l1BH8o3lVLG^vFqiJ#F%Skp>afE)b3-9I6QxwBU0j+J<AC*sl&d
z{qh%o0KI0sW|^EOJ0D!)<&O4!`uk3cU>P`zeTaT@dRW!c0=5SEu9i2sbia9vDJO*k
zq`c1ZW_=?E8>!}kOjTV+D=mryN>%<?S>J=5rlAvbkR{b?aq`5Oqiw$ZK-5ng+r&I0
z&^E;2&V6C`lu*Zndx*CH{*E94LfS<13`YcK@5JBcP7xDjHs6Sag{Wz+k&>2KkbtZ5
zd*-~LyxaG6KiTdcbedyr2_(2zi30Kl!u(hH{7Z2jyX;HJqfh`b&St+3zWlLz*LP}{
z#UyFlQqY+VxZ6qhhE2KdaRojSp}~Lds1KiwATs)_42)uT7P{mXo-Z$Y*G|dA8Qws$
zFtqp{M^|q=NG|<y`QfCq%EfKjGFst-2kMI=bu|?lG!QJTlv<A_jHar7K@0^j5mo!P
zG@T=m{oK1=nUi$G>2tX@MMUm8H3+XYWyhOABw}a|#L#mn$*qt;^aw{=03gJ{6}Ss-
zQ`NtSngwHM=d{bi5G!ZXdr_FVI^m+jE+=Siq_U-(;8%%Y(BIVA?O9_9pPXBcliRjg
z)Lc*aBrRCH&^@Ug&3mOdz&K@Q|Ers&%g$g5okrwZI&UQ`Q-03-qmF31w8p$i5{2Xa
zFqzGRls0tcf6?oqh8z8|YF<MO5!XfC&y>MCkX`gXPg3MRvJ`(}f4E)P6KlXl^izib
zXbO`je~K->1s6&pQf9F-JlDq(Dw@WQ#0$fATTaF9C|gf=&v(|W^j4wD!+nKbrXE+7
z7bxB}iJwyQ>o3qOqsxm!!U_(Iv2>3pzh`Fo#LP5(*mBT=Znc0YAM(y@gs_1_UC|>S
z(DC!(f#@uf>q5Y2<zTzfwUdlu6M|G;1$yzrRok{thA4|m!57`9IRoQrInkvqN(?L*
zeN3affuWuF62nF}DsENYFTNq^bz23p0(qnLxCu8zNbD_a03?i4U>L3D2Nq6XaG+MB
zjZKAGmW@IADvq+<(C2E^I`p8{+o6i_?luzVdn{6e&F|{;Xxwczk)KaJ=Pl8zoS90x
z^(z<i7v19=wTJ$L_uNfc&DcvGb^HR{5KO0e{Is3*cbvGwDDK*&ZfqWvIQv5tjAHHG
ze4?^GW0QSo4oA(k_)+w4Z-tbzIrGp#g(dM*2SIBJDvIi~9F~r}(tKJhnwdE`<8wm+
z7<`xgKueYuuoIfT;rhtZ>2MoQm^P{7@6gdo|3WG?^2X+R{5EDE6r^zi4xdnNPGVPq
zp__bLbJqCf!`@Cm)SqQ0RgYAz&+msMp%y7zlyLo?7E&v(%WMxH+a$53I3EOAxL}9X
z@lP=>Slb-^N%q|HZtM@kVQy<8Ab}JHHI_=Ge{03?`-T_@xu&o`x94@krD2k2RT|jb
zm(-NNCLXapi(1>{)>BLm{aD)T{@Em{!GVrwY3nFyP>L*VXbjphzEI98L?@12+KU{L
z9E-@u8|}919GN=PkS1UCyeGEqxQWmZ3P^ykXj4Z}o;tpJRV!Y39M;DJ(+tGYXgwQ~
zqv#lt1hw9zlbj$1m*YQchM{7?@#vr{RSB>dpa@VAq%5mP+M0z9vSooX`nZo0M7zUD
z4z(lJ>3!0RV2QxbCrQJ~W6(xynudM14RUT)Qqn-!SP03Jbt%s>jgAO@eFhBM;iXbE
z6CA2_IM?2QDPvIzQ=*(t+`c?`rD9wVz{j)C3JR?@VQlsr&wbIiul757oxbPx`VuG(
z@zp{dRSDi|+fDg^XBYnnD>KcGbnR=-1XDaVRe=HB^ZZ0KjQ@{#X3P(<I8L!5F}mT~
zV{1tvN^Ip)B5pRzF=J8;^hAqKLivmwT(RjFrj(#)PzKh#?1sD@Mz;sOgdU+>ZSG4s
zv7<aWT+~!BG?Hiz(z@mVFUY1Y(^59W6XAu~+i;NWw3xc-s`-+aN`6GgM<poE)E;2l
zkfcTDE%ECFz;kw=RrC0HK@zg4Lb07l3<mA7CX0ClU)oTwWO=W>6o=*e<%X1`2MCA8
z{IfXWXQ_DgdV=WlpwU(ZsHk)&#^5~{v|TG7o1y%+x8@PNG3-jf=iMkJ`QFW_5$cY{
z@R{XLkbq$5Ga6`B{@Y+xg=zb0>E|DHO<|l-LM~i1##gKemCNdBK7kQ)Qac&^Q+`)v
z+2sKy^$f7|e??tp(+Xv7WOH<KWnpa!Wo~3|VrmLBF)}hBFd%PYY6?6&3NK7$ZfA68
zF(5ZHISMaKWo~D5Xfhx&I5#;m3NK7$ZfA68GaxVuFHB`_XLM*FGC49dI3OS(ARr1a
zMrmwxWpW@dMr>hpWkh9TZ)9Z(K0XR_baG{3Z3=kWe7FOQZq3#;+P3X}+O}<*yRqB0
zZQHhO+qTW!wvFBYd(O#szx$uu+^nQh6JyMpV^u0^6&aDD3Z0O#ouP@iovkw+6Fnn0
zK*i3*(a1!^&e%lJ(T+~o&e~YY8Sqa8h6`Y9VdM-jG%>TVg<<#?K;FQ{1i&C7r6jK?
zP4oYWGd3~(PpXuysU3jxKMck$_W#4;YU1c*VP^}V{HFjFz|q9O*v{74<3BKYJ7)_c
z69DCZhyfN(00V%tqk*xBje(;Tz|Iu#?}Y<YJe-_OY@7g6wnlc2_I8d2&L+kH3tND*
zxd}kc*5Y55it|6tf8e&p40etHyMH2%03$mWTW3cL6DRurt}9|^@8M`+X6_82{NE`v
zBNHR-zeeW&pR)7-A!}=Z^1lF10A&*=6Gzv7od3-)ZeU|!?eTBz{-?KpsRsB*w*OAG
za1yt0H!)VUa5ggkr*{{V|G1@`4XiDUglx^MO#pOEO#gAJ*&3TTT3gtfDB3w$IR8sj
zIwnTO|B9)aTNqi{nm9QD*!~kTu{Hj$nqvPnWNcw;_Ak|JjSU=){~yP{!-@tLw$7>^
z_9g(v|9!jvwweCdmNWQAT6cgpBRwM{(?5y-JpVJ(`LA*+&W?6gCci9<oz4H>jKacp
z?f@@3mVY{+V`XCna56Chm>Ag^eg1E_k&C0FiLLX$x79zb{EuyF@sBblChjIiFdM6O
zMm)im=~)@h^_Y31)1n&fMBrj9ZX?~}mf;TP7StL+hPta1#L_U2=5+C}!nse_CFHK2
zA2QzCof)EshBEOXL@-o!2SzX)6&`GWY9i!Z0mW{ZcXNm1@&v**H}k)OcaS457m(_V
z`gDykAdR%>uS(q*bNCYdyyp7iFn#y;VfTCfY-Iq!_%xX$jS?u3?(5-QzN<b)joh&k
z-(u?kP+-;}2|!gZr%DAR(iuqhkY~YB1o`*)AOid&1y8aMo*T^+LU@t2OfamAD*PtK
zIBv-hGkm>MeRlB%x^TPB#(QfvXVmX@QuSGMKCv@)q4aVF52Ha#L#W-9;?sOU60Q|a
zmnBa4YWH8X{N;H9o40F*m7)P^?qk3dDsgNu=swGoyjX60BxQUL9anN|wCWU{npJOK
zKvaj0#Iga4Dcq&8sU9M)*2t0m?qKHZQ$9#^$r>=a=KO9Wt@@#GZUCI}cc6ijUKD>t
z08XEf-;(8s_S}7zy9Mc(>=<=0(^QJf%C{7gDi=|Y$5w<MZukw^A02Kczvm{l)v*^u
z<}l~GZ*g7xzu#pxB_kEV()|U28zGAQdkF1ro2r^Gtk41XK*H=4zM1_KjnBO7A0rb<
zi5*Dpn+*M_TBZvP{WHE0Ai&F2*h3HG8j07I)7cmN<>-4aLsYey#!X-5xV%I<viY(Q
z(bK!DKm5xUH7|UwQnA5xX{O}tl;7xyeIp}3v4XQV`bZFIUR>+2Vzk%C#^a)y&^E=;
zG$O-Moy(<_`R{0y1I}l@Xsq#wrN-JR^?$%?b&82#VVtTXkH;m*k$<ic{x~*godtlj
zJMN1f)ibW~(+~$%SaGoBgdYmy<oDnzBwF>=`St;Sd&%&VcbpDqjtf*@o2O$>y&Hd!
z7|f@Uo(8{IAigCbcHC!PD}bw!zQjtap@Q2UQrH!WgHRRKLTYL6#CLTuE8-TdkP|-1
zWgdsq#|p+ohLMK>I5@hMrip{YYq<);|4Obuh^FOqtP~4vC?yD70#zHqp%!u7H{Y+$
zX;b#b9)mrQ3SZ0S`(Qe>$7JL?85hZa+<7QT@-uZHFS#J0GM*D!$>#n7a`KfP?&h58
zsn%1^c2dO8R@c>QHqVkbqiXm`=^%P-^teUiIKRMmCO$v59$KvxFl<<50A2K*K46&6
zP9~mP$LKs=q5<X1t;N4MmknX>Hc1Wb<a@<o(awM?)V=jv!C48U8?q}e`#kyWuuPF-
z(n<ClxdElYSR3x-^{^0u`AgV@h0E3bHu@ovmPZ49($)^SU!;jg%vC8<mGf~K!YGXs
z-R}u`7|2E0C$1^4#Mq5`gV1U-t;~Oe=TQ@{5`<EQng9^kOufJAkLr%_5KkKfJV`s&
zOH-*f28?ze9_&hHWz|6>mG)Olev%mms4nX`W>M;UhD3I*FYB34$^obor^&tCI_ikv
zI8R&mH(?g$vQ5c#bl*I=d^4ABSqx3X?ELPO@<lYsN?~8J>I^R9w7Z{nM)@L-_mwn(
zMOgE0*Aq%4H~7H_1W8J?hs<^eDZ;_YGZBO``?0oFw&~=2-8kkkDOS~uYj5d8?F-+_
zH0~miou5kHVSwXEly+x{P|Rhv*DV`?nc(y9POlDihOvEd+q4F)>BV(OqL9K}5c;qs
zFoqs6y}S;`WUfl>Ha>aFum_*y{(B>~Q~cm*9i-rnJ~&ANh<v=`nqeHwO`m>3sN>N|
z$CdF5-le6OJ@NJ7(%p*q)4Ht?l_;jzMk$-Bg9_P6vm+bNtiz#du?GBJC9KMcW<mF9
z8Pot?IwsB(YIg@>=wK8<sOxWuQUvsCH|Ng-2igYP>d9^gqZ<<0)~4HT5E{~HuDMLC
zUfl3Q-nJ;u3Of7#Ro(g*+3NxT+Hhaj%4DBcpPeJeb(4LZLme9->V!7x5Zc3;5$f<s
zqNv+tFr)=-2DWvzLReXIeN-No#d>&emXwRo^)vIWfW3+?EXf<Xvc=`(Y^@#lm>L95
z#SlI6;P^pf#t4vgMxy)mcc+2_I^Tpf^xT6&^)5=|T^I3gp(YC<eDTVdr3OeTU2e&X
zIb}>r+)n;yybDFmK0mR-D03=36q7x<I=4j2mDI-4%_62xe>cx<yvJ#Y+sdkEWfBT*
zQT1X(FY_foR6f*QSAyig>@&${V6BKOI>s3TvP{i}Q!x4Kud_^Iuq!siQ>h<|KdaHq
z3U&jAI%Rz=aS&+ST@Tg|Lplm9XH%HA&e(6G7BxUR$BVRG(7~H|YDH#DyoPgN&{#x*
zVY~jDKMFwxO0pZqMZvVaAuw6}8e;Y$Vi1OoWc9U_KF0`4&sNiOg(=xHIJVuFHy4I!
zob`;z+-85PqRz#32EE~hVIq%Ip$;;rIDz~#BjfoxE=Np>Ws+7yb5U`8J@Ho&jw$-8
z2IR~&?iPhBXtfP1vWpt%mvrxX=>B-8{ty{fW_bH1#hAqXe1VY{WJGeHzjd#iK{4wk
zAiS{vd%3UIt?&BvA$&SWuCU91!k)OTina>hU7zWjoJ@5Id#y#wwTEbBmE?7+26?m)
zJcf%@*5y_cLt`~#^+L|hfhFqOt_(0j={ja!%2@yH-2Syn5N5{Pj$?xLkk$v0b~~nd
z+ncT&XbB-4=WruW$hPCcUe{BZ!iA{nf>j+yDukz{-V*d&0eVcVom^t5eSaq$gX!ad
zn9|qHL!1cDn(nsT+H=F0Q8a#aJyLYkU<MY<^jky@hrsM!JCgU@fW9KqTM4OKdZ`~=
zq+w%DW&Laq24vrN&PEL{6V)QE)sFB`iE<JKSWhvK(vglNq6*3|`L72lx|qoqjX|+B
zz4pi{5*0yy8YZRD)z%aXF}MwRqgNB?XI{g*Q#cC47aOea=eYjy>My1{Qrj0-#s<wt
z4=}ZRYU49d84NkP7&ZhuGc9X5*$HMnY6B~N(%6TVfrL%Sw1w5ML&+<9zjVe#_c&p3
z=Kvd$O{h7poAaTVa&hYNy<d@U<azl&I5Q%uG)QO=UfAL5I^c3d`oeO5^A72f3*V!6
z6Qn6N&~3Nwh#&(&b?^zm_Nh47pa>b*<7SUS*HE>Na5BXz7s*Ht+HQ6Bi-a>|YH~~(
z%d;ZjTJ9Nx=4j9TrJh<XkwF5?y>cVcr7^0jR5%sWgH7l8d&DVVE>|3G>Cvw=Bo^P`
zxp3G9J6{M_4QB%sh{urRI2l1^g+4K*RF#pSRv|3r?R{y#YBObwtvwz2rOhWE!aZD{
zaj$~CYb{;17C-HNlt#;_!Z9clc_0bXVB^b)^8Z9LH3QAEz-vzg2}K0%py=67amY<4
z#Wmw7wmKJ}TMee!+?4ypAQ}c<4}5!4A8)C9LmHR`P=lk>>|)Uiuc^$u?`nbqzjCx;
zpDa~xv^9kgE$v;vv3IQ)3rTlUZNz1H%j5MO?800&=2-fq40vJJ^f7q#B9`9<r9Q#G
zkAjXKUb~L3stQt`?rW<L+gcM$-BG|1w!1JjS`F5dXr;W~F;eI*sz}=F@}x&FmlJ06
z&l-VZN_tKI^4<9Psim`<&;UMSI6h9D;6I>s2$~Z_*`SI|Ct-3h$M(tH>AOnCcbdku
zm+Ktc6%l3(UnE6TxC%xm;7Dk0<H+^k^`DHFEOP5m?Y|oCEiN{4;u>wEf>=kAt9|PK
zE4Ar4t4`^EjJlMp@>e^cgyjtnqL14dpYJSNb^q7FuTm=9p9O8zV{{Z$jhO>7zYhg`
zlWko+Z5JNi@vyL%gsQ%%!jGq(<uTq)=Ali@b4a4_=XL>ANrx=nSKH(U^TQ~&&1L2%
zl%pBWsHMj3nBZ5k-t1w|^E@MBnWu`ms+<K4U-CQWU&zSsjG)KUcEBvtzEOQ6%#Y;1
zAx^p&B;8bjDZf32n9x#)T!89rtl?KRTuuEN#h)BV+nSh4WX0aDRY>5zza2Dy7BGhm
zxv1+bh&L6O2}=Dn2Xbq;Ki0kjl95e?LeC5PC=0+6glUUaT}-!!Z#`?9X7#s}2*h(s
z50zvIx`J7p=0rfx8P8>dAo5@*na%wc<%ZbUi^XMif?LcFVxm1&Vm<N)7D`Z~vQ=op
z5kHG;@J~=<HIBXehQ(d{U;}7SAzNBxeHqNce|LJ6pW+8)a=;>r9IK%_Jrnto{)tYI
zho?H{%~p8yJbm+vIvjASk)$H&EMX@D|B|2^<4$H-tL>i6X<!#~U-%hLi|#8eW9wWA
zwf9dH@=g1c)w)OD!Jx?+r=2o6PkNP-PKhi=kdTIXi5n(c)La!M_P<S0lx`K><L`yp
zV~1PbX2$VrOf=`i=_TzC@)XJVb|AM2M?404FVxE?NEP~9{<nE8xidXEsiL|RM6`!b
z!GW$2bE<BW3yo~mQ~#u>r=x)D;C2TSsF_FPO&tMu5M0;^OSu2ap!v#Ty?OIl?%kJZ
ze_IU?Hzv6yqDvCUsTTA53tqwgchkYn7+9X8uzZBIboP1k%DX%mQls1KT@c0@N<vZA
z7-zBaXFxUy0*OuJ*gqdX8KVw9;%<xMey)D5oN!LQTKO5?7aqH@X8Y<IN9u3Np=!UH
zi|pi)i_y<F=x`8%ahnU`3!R)E5UsBvC@$>vz;Z;pm9MusRxbOf+c){gR8^bDb~oxe
zM7Mw%@Yr#XBHB|xzX=UlQn=r6tSZaI;nGi1NmP1U^Cvsq-m9y;O_ekZHsa<ZcaG8+
zd^cays9SapP0J*i-DFHyW$d3vPH{M@AlQXuc)gLKs8C2I+phy%X105;ao<6eQx=Ou
z@#HX=03(MajCg~qjqK|;;i<P0bSzdT_ynCYhUUiUn{4M%wcBU*)Ut}@UYkc7f?6WV
z<UmD=4i~6O#x{z_2rhUSdcro~Me!(zcAXy9upg$^@zOWn0>|f31SQQIl?e>q?!pyq
z1y|T|>=D`}Hv0#(BZuF=Va-L;#V-K@e+ssUQ#k+*P!yy3@Xo`*uboqkOMY;wj8u%P
zV>)~n!~3i5BzH_jr!*3Y;Mpe{E9&%EYw%`yNI$p61&%(yH0UvS7IrSZ>R?OS!bj?S
zCUO3_xn43a)R&JT+17QwK;u2m@=?k<hHz;ZC2sBhHR`J!-|t$gL$qg>Kb3M<!ya*%
z^u@_zcmQ$h9qvIG#lh{_V;DFIVn=1d^udz=tx?lWrUn6dkDDQNPE+8f%-Iwl89fOU
zLM>8fwA*zXJb!PuVh4T`-m&&tBN~Umo%fjjh$+7W(UXf#IU^g~Lxn^K7<T#|lJdxO
zAAZOjwm=&O)C!Co;j9N@rc1ItpdDX-ll}=c%^C!-_PzOKPk2%s+y3_HiPcy7NvS5E
z8>>WNAt0y2A!pAs09`<$zm0L$B@J{<yn3eGg<9Gia9rcvYXf1DDXA~}hMHu+^$n@W
z9K|dqS5;bKNpldCzxjfL5e1X<XaEiF;Y|9v(!Wr86Co3A1mVbe^JcrsC|=NP;^yb<
za{Xk<E6>^t&Mt8RrKxDmJh0e%^LSn}rktIA2*_Y-J+%bDatB;2OxXt*$|=G-M>$7|
zG9q|V`QRYlhT7h0fzwPDbW!cF0=ciUTtlW1etY@|O+vKxbQGN!ZxytM(CP`rvO;1m
zh7wRaz8(Hzzd+47n*E~L%2)3hp)6W)Uo$rJ(O^f|H~K*bBSA^WWt(bcTDjNFop{{K
zTbN?nVTDag*h}dJS+{~)SuYHtRsXsalzWSp$YZzlB7^|jhKp4cMGsa5d32UL3E14*
z=BA>pmZDRYvQGvAm_4H<_qlJb-{gfCGj@FK9_?av<^aIn0=1kM=3uNZsHCqQLJGF~
z<f~W2sxx6X4|0E3JZ-Plqbz(-tO|Q94hB$N6;<s*v37y}_@dBC2-S@2Ub@t#Rh#Ek
zTP0}bTAJEzWa=3!QQDS{4^b6s@odsFu&J5%jl6JMgElN`hFehk^Flb~O$Z>@4mPmH
zkz%dH;fhv_gXx}IWoSATL%AR-Z4%Eu21rk){~8*SfK2bpi()Nb({QKnp45{5#&_Fb
z{QSnB;pkXuywxQNYJbMH$#+8U*Nj}U3w8%K$p|l*s0p{v6)N-q(NbIuPAERyNE6$(
zPK*goU@CK_Ll)SoW)179WNsL(?d{Qy!J$4q!vAf0S+6<26veVZbZfZ#j4`*Euni1<
zMs-e(4itW_rM(DrLbK9xCf@TjKmi=7S{5~nrzO^_dKAS%o1lMsG<(hKAp)WZfJN0r
zGI{;v2Rdejq3-p%l!4ZNRVyFF=Y(zW#RLTs%LGZJZvi?ndc|H*D8Jtyo5mh#*kFve
z62~%i5t6}K4^zS)n6m5$Q4?LPAvAJq{PkL-;5r{Th+12>X&M<ZlN(O}>^~aUuSHcf
z=V+V&R`HFF_GkUjlUS?6?bz>O1LZ>NT~Xo2Wco|4Bz>7%6Sx@|4a4v#6}IlJTE0PX
z!8W6e=lV>^+<GV@W2U(pPUP-<Tr2gHB92rU<3j{aboF#k54e#vv$K4&c(ebFK7z_1
zQR3>XkSuMi<@PWY>A7t>ICgf6F=Qk3vN9QhiNsm+V37q=U)W+o>Svm%FSb-9Y^!k`
zY1A7NtJrHwmAzk1DpAZ`Jl_qRMQ6SF_uOX~`p`2h05|IBXJ>oqIhZfjDk6+CCpvK0
zxIkISH`bsqlULS`fk$gQCUBNintlp_<}tb)RxyZMnpBI#%TlL!u&4*+uGk`C{&FU)
zrf-dg3W%JuY;7uZ;KcDG$uG>|^tpB5KI#>6=P^n0bM<%Azy;2E5eB={D5(IEqhccQ
zoD)x?dNfy`TQ`oq7|oi+nL0RD6r|x_NQ7Kc_ME*7itG?C4x}cb`Bh0mG_pDNe6DZR
z6$N>d1+6TMZZ-u%wyRG^z6@v;HK(7!7ck(U(I}ofuOF!haJ8aV5V=l{DHt`S84>3n
zWz@&XLqxAqI!-wo+;o(lvpdMZ<v{F;$?PvQCDRpa1c?cFTkS#emm{S?#UHcYbI6R&
zvqE#;i;n!9m?tNc4`THKc5_hfsx%HGk#WU1)JNs?8i~ce$~SPRhPz~!F|lV!Z{j4Q
z1G8bI5LMa3qEQ%o;_;%en&hw~n;`ipa5A9gj(0J5Vx<!huXXtj#-Is{KZ1YTj^(hI
zi<vAdHKp-eCN6-33^+1Hv?w0!9oC4Ny1*OMlK5#tv5K6iv9}wY#%{d*HTigfkFFQa
z5c~pJLzl1;jZGvlYXV1gJZIZ>_A~D4*tf3k)@LUMVL{6>;tir+UG#5oszehPC5TQi
zYw=|qF_)rUpX+n-PF;)ux{7lM=7iUdi)5bw7hFs!aSu=hS;LRR(a!lbiM#XOAgvf^
zkBSj+!9a)99h>?RQyxa%;>_P%2RALc&H|MN^Gc#bE(Hy@p{ionowRwfVU%2^JkS*p
z#Mya2za(~?w7YvwDNc~Wy{VMuFC72MBWCAMp1@YKZ~9WNiE#rdH%nwCt!8-YMIQ(I
zxxRFVa5z#LJKiGHF8Ig4C@|xK2d~4Iv7~gDv&V8fAyqLtmt9DZJ!lwmkxNhXRnWj5
z5S%lV?ezA!)A0S?D6FzwJ&WhBxU`spWD^J4b7Z;!`3WNlL2|K1VrvxOqdX{ELe#?~
zBLWBCo?>DPuHv?jp*ffY;@E%`uvlZl>uAD~Y7TPoDPu?5z3maNcM8pC(~oL=iKe|)
z!%4Ox<2JoCgYkU^MkzTTLItaO=hO?yt6=L@+55|lD~gO{o>CcHP5Z-3z4F_>##yO*
zP8H=FUG*~nViX+MX`t*^w+X-xC&r5peo7iN8aX*Fbns=vDpDEP7>8TxpyhkFGb-ss
zY{t3(W%DBZ_r@LyGDvJId9$7{O~N8rwc|b%`%gE^9lj{r0(f!;elovWo@6umhXyCV
z8}VqUkWis4jG=BCd?l&TmG2+1av1*4#hlJf+_BRPE)Qs4h1Kqdmu<`7h810|iK~&8
zUA#nIE*+dTb9abr9Ipwvv;;4lk(3mLp*CEEj)EC#7TU>%dp08u{Vro<<JYe7QR!SW
z)#dPA##9imxshe7p-qfAMJaLNp98j-AcxD4Wp^#*XbgLJPc+dTbw`<mcd%Tvk%leK
zsVgz@=;i=I<&KTZA76ZPJbjV6edFIfHt)@tXtJYYef=F8ga{0<d|@gT_2)@cP($a|
zbNsAFdqECMP{1#tbtZ?BhFY%AlBX<O6N7Xa&5h7Z#c$mowsBvzZ&Lzv3}yxR=+WE6
z?=Uc}c9-VEgom)WupmI90xj)qeB(=7YG)T?5!29x7*vON7Xs8RIG1G_hNdSU4P*x+
z-90c5X)`whz%HQyT~@Hu`zuhll4@Kdhyyr?1L(QP(2+1KOURra&Ugdn1_vBr^lr}4
zp=wDzu-^Ov4+EgD7E@QazVipCAiqrmbalseQm1+s7Wh@9&-uzh%g><=xR4FO!^@YJ
zWjx1?;L!7(!(XUDF|9KwZ6Vt>h;S21cPc{b;>`O6T=O}b^4IGsEkd6>Fh{L@0oK2R
z8Lc_I&4^Tva#n5l<Nv1E_0;@=b(R*#)kCXuK#VjnR4TXnmH-E`m431eu;e%db1Wd-
zLzuDDy^76jLCa@-`{XvUGGeZXVqM=nQ-Yw`D2T%2aH^Abq5lo9Vt+@2UICAsww#{G
zf0Jc0vGa`cV3L2`6b6x=t<b*`I%f={Om@`&(?zGiZEhpuCh?XyP7NxairO+xYFhJr
z1h;<KhHJno;45S%!aD5NJ~mp#d;api{f1k95xqqDh;03O?eh|e;J{Utn5#PB&D`ce
z*b9MOS+WpzD&T>ihD|5NMt_7f3l5$FKE7~rk<c=7;-R7cc4=s7Ek?JaQigN*#t&Qr
z&tZ8$2<$_@-;*m;PsHAxcF?@g&M;qz7anCl9sPIU1}B0Q_SF7J4rv%&!7(zg!6L|<
zTBrd7U)==v{ufu;e&nq)jF=uju3@?e*Y&l%ly{r@XAa%ay-cAIyF~XhNSPmsqBY2U
zQTs2C7t$2xN#2@!(9NMk-%$=*(_U@cV-AO-24W^JL#nbc6y7lqFn8wSR$mN;pfwY3
z!1I_$quZ1e1|TC;y0&j<W`p7uKQ<I|JlRjM_3dPUhq-t8oGKV{GurWLq-r~OxnomR
z;wUi{IS53$;%0d=+!gZxYP`;~)03(SPXO{BGW=;WR};82aOZo2F+7L!pI<RFRt2|8
z56RVIG*?DRo6Wy&QdDS4a+Nd*)LGL)^TIkjG|H=bNG(4&H&Aj7j~W82stxd~d$Ryk
z_8fXOeIKs?F^~nbf!pV|VQP5GmeQN3k*LTWP(Dj|!~NWzn}y_BT)Yw~kxE+b3x#_p
zDgfWyEhi9-h~8p6Rq30RLy%;}83ls-?n_glUhvlk-l+o^<R>ViSyJ&paE_ORFm#I7
zuyEXR=SPM}kVbrR7L0kGCnx*keIH}yqUqzSGgE8nBz)ZF9lxM*lHVJ=_Z38kqDUby
zaU3)0{01GX*A%=H`Sxv=M{*??s}|=3ttHhtsP9c8F%<JB?6LOu12k@?On)oB(Kmkv
zIvHyGTROh{ku~9-iVG#{IxF)US+|TLck`@-%U}x4<)*C9DBvE5XeW?Vs<GaB&W`Rl
zjwS+;k)-eU_(kLvTc6nB45}vCFR&NE1CRKPh7?I;*v>VJQ@%!Z(u=+W8KF=3J#dSk
zW{kmAP+;*1OyPGcd*v`FP8$3oGx&+XF+zmx4^L^B*JHG~o`=E9Cz|};we|tTTfunn
zg1@T;lnQhRc|h?U-%T$U8VYTMb8idOj?=63_-?%b+a-_tktP6-`1>KqQE`$nw8Qhv
zG?|p)=u~rrHVH7+b%o;Ce3CB1Y)iE79um|FX2wm1bheH9x5ee+4dzYH9|_$-8h3O9
zb<o?cr8M-%s7x>PaiB)%Y?N7YZ{y(7wFck)lX0G1#~)4|WqvIvUjD!j0t2;NU|vj`
zH{X5Xr*1D{+BMF7!8^!kXSbw2mh0$bZb_h<qLS~N21~RV3QR0F&(kzFP9#6MiEfwg
zvaE`bCh#}8GBoWz3~~D7jh>SAJo^1u5A2~EcKljKNN5me@p?&u`)hY_nh|7bGGXIc
zdy|%`Os~e5mwWO-f0i@>t0p>JD~f~;59rg^Z;TRuRZU4_KJa94&eM1JUZb3(Y@w`a
zgTOuN*Wam5B>UG}PqEQ@9qIAQqcv+z;Sl#|=0+vEn-$8PGIDd&tq?j0#0-VWl_CCC
zTO*|J3eR%`%<l6Y=gJZM>`}o9N7MQZ`%_nA5iynnY<*=~*_Z`zR7&W;ORf;@9kvKy
z8tmDgElt@)-MBL3Emk~XieDA5`%f0U&V;DmHh;>ty?+@@59dB;8Dj0seB$ZSLFHTt
zkRIKeh8R@4`{w+a_Q`ZPf2!yxb4p~sHL*kbM8HAw@^w3T+fo%x#4@r5&fhfB!<+(z
zg{LEnKh~|F$R{-iHuRhs{1JaLBFnWJ(ge&eWr%akEF@v3J>opHqn=?M5+`e&g&Nep
zM@V!3sa(_cmN;aw5c3Jn?#enHts<=nnp2#;jvGm9wlDC%c4OOty>c_6?K6h`Q~K1P
ziUqgJHZZu-_Ii0VQI0cEgd%<Z-Ap(wx4K-@5`e@aY^|<3Kyiz*pBjd!FF<UD$sJM=
zsybV6{D4@*J~)mSX1m?2NK8dvxA0AicMRM<eGfbl6EjX!SHq%x+ijsoA)b%-GThh#
zn>iTs-ANEH34EfM9$;*|4;rFAP45op#v9|G&TYo_mfT3_<7{E6Y^ek>TfGX5JB9v*
zo>orY^#a1EW33c9(NQ}{?kI=*OA83@%cu-9102hd^<12`bQW(IiS;LE#~U%G^XfJ9
zs4C^^h5zM{M#U=Ew*j|sJfNtCR`m+A-As;{CBqrqmHHVs@FD*UM-bZgRUK5tdCEG&
zhW(hPR03IplblXrNBj8ql=f-XThPs$ZL4F2%IEvR*nEFpNQ*$J+!DW%TWUJHJRV)h
zc^_z=mcZGbHhEiv=2suR0cl3VH|8(3YcpCEA1rVe%4Spdsf1hZ_1>K&s0zc9?(oU^
zA^I*Mu3^E3sOrws-N6T+Aj$*E?TwO-k7{8;=|+gdK;9oWsMob<@_{Ab4!V+B%n@7T
zjOgByP4xtQ%@b+<-?4)7zYGWg&&lqjxGDo|7rgk|@YMtEQMrcGs;Xxqyu65$`H^En
zrMNRByT)kbmTx=QKo^DhZAa#+=T=?=<N0CeNo8#c_w&cn9DOB5dSQ2Pa`jn3hzZ3-
zB;7is+0s1VUrAideT=s^^-+;vbLHuz60tJF!=HYJp20LU379AJPjASxkA_-j4<ps<
z^1y62^C*?Ct1N2|2p8rY9_RuwuY3-fTGOdoy{UNM&lFXb1=a>s!^mo3d{ZBhMF<H=
zz7|379tgM}G4Bhmzos(ay-u8wW0l!lqnk7;fy*p=82fQLWPpvi&%G;*`Fw9gOjgDr
zsbWmW7ShKO0c}lS&#$5v&s*g1A<G~jrUblL(|o6k;60JPm$9`zK9y4a^_(sB6Z|8?
z-ynyQDz`d8tV?1TuG0(Q*Nur1^bm6q;7kRRI<)TkI|n2sn|CzSN;s)^efKmQl^mNo
zJAc`|HUf#aD4N9g1eARWR>DT-XVvhQ#mHoJ`D0*r%{fSqDkp)6FvceX6(f<v4e{B9
zET_6GS!bRPDK$=o=nzryrhHWOqV&14oo7Ov!SWoEGjSfSo`Q{1<_lpH32PQuB=|8f
z7DrGdy=79<Hvn|<8PHccW0h}gH{i3^x{lp0ai<Vu#Evt8+M@Frf^Ja$imCE<-zfZ6
z?YNe9{;bX)>cSwqxJO?2?0T#AJlzbu8qvbjasA~?j%5!IV)@s#6gb2v)7fRnnb&Sw
zXQXV%JTWkNPi`uhD6_RB=7Hslf1Q+f$gwN8i0O`(?IS1aG|6i`DJd2>yE`}Y3rSLf
ztCMeXdEmAgXrte{T|#=OPKf8mAyxQHW(#rS4qTbA_s;1r<V4S()!a6ywzCJ}!`cph
zPj2F{16g(b@jQ(l37ZN63tI1OcxH+G$vrI`ww_~a535eGvm{~T!Qh8HWb)YF6Qq^<
zrFqa`)&-hE8^{ZFy0x1^rpphFE%B9N^+;3`DrKczTOLJdiTj|COC9(XYT5NvNHr#%
z&Xy^6!gt`JSn85_#jnb(tn~p4Ip%bduSAG%|BC(gOW_c&CbzuXFe0_SO?s*zijePk
z?L(q9MK)OqBX#~|YnB;{UtXo;%5e$XF5^c$$4eyCdt>%Mc2$KwtQ`UzxaNUuseaPX
zAL&EwjU{n5EX>CofMnEg7<y9Z6hm#Wpr*Tp5SEjo!8#tBhN2MNhg3=u)Yd|17-0dY
z8L0Dz^Q;KP<{UZ8Rlh6!`S@kg$j(fQl0#&WP3wB@q7S6`i684XPH-Tq0gFUU8d*wx
z-@LQsXFwBWn6CN(=R00=v^UWiA7Z|=R<#<<AN;vi$8|BpAX$Ldv`29h@lCBmHWbcc
zmJm2!tGiVz|6p?y$K(QTKvWijX#q(}HA~ZQ*I)Dw)N)Sf&Aq1tdV*la+dh@X=>{&y
z2Kt!|8h82l%-Bjg9LG|8_2w2wdTepXtxwfkK`^tAkAs}{<%2^alxC)Ia*Fp_eqV~b
z*%@AGFN`QapU6YhhrQ`B6}P^YW}v}-OW+WCPG|Z0idqUL`(4%NvnX)ehjuw$&vHd^
z0IZ;BwJek)=g0&*jkx`7|K~%b5$trb23v9qdZD~+Rpq&Wx>l?8+5z$%J1-<`7|M|J
zT&Xdu2IPi$yR=^eka)Bhm4Sdq#~`m_QYi+d(M^_TI6ebKZ53&zDmi1!AH6iV)xTeF
z>3A3V-)w4N7Ec;-slm@6a|UITz3v$xakr0bAT{d%M?&c==ZZQp@Z8}-2O|x~-W1WZ
z*>>Jvz}vgMIq<4GsKUPjRFY9TT9>iikfh(KZBLUT*+(^>20t^Xk?9XWmqEN31!Oqj
zRNqF1Hs>no)*Lkj6Wx;<#Bw(ttqKM;DTZvbvgP&8{YADdLXAYtSdaq%?JXBOhDntn
zshN={ZY<3f{^P7}XEJnc^3O97+Q(-;i6w20iCf^LCpY&h))f%nw$Xf}yNn5sB}{nM
zFIFYJjGFnbD`wB?ebUs=Nt)$THMrsrr1lh7^y9ng5prBvCbb$yLyb70Pm3z{3RURo
zHe~4w_@Y<iRc>?Hvtm8ALd?02m|{+ORv@D$;LD?L)GT2`CZZ3=r93*=Vb;OC7m9g$
zf2pil7VlAv9+Vrb5vvB(2apA@l15wdN&SZVCP+-XXko;Uh%m`GsVSL~+zaENi>XGJ
zpcj&KWKbwTh(?kWBv8^i0$OJDDyKAvqCbGslH3M|+z0AQGo%Xsh}*Y`BE@@R8ypAx
zatOPeEFU{4wM)ds6~VARPRHE8H=U7ugN2tjpV4$DXGbaQusA>g7d&rOaW;xEO;j|r
zeA9R@NJ$L;X>k=8i(>4Uzm5@tA3%LOLMzZt<c2jWIxqi?x_JIepVjaX!KGR4%*Ajd
z648z8z9{9;D$A@hgn}~21TTC9`S@^A2ab(d5S4|-Lb+Xzg~qLRs+))T3C6h)rD7eQ
zsky>uX|*LG9Z1M8kC_=CsqpWbnwE)efAp=k42UF#la^feP<9-Yr!%&qB51ugGhj#>
z4qWLQS+ES+N)Fa-^ocy_1xCHECc>eYly9dR%B^!_hoKTG-0Xb77n5|D;vOz-0ICo9
zgQnCTDb2OKV1t57_YlEjK$I;reyKa2sms)zsF;mYCI)Vj3cx5`l3Cft#n$d-)kFS4
zVfuZ3WcuDe_;!?;rE7XO-V$dF_I_=4HdpLAr4J7<X<n1|m2_?-pn;d=?13{YkAa96
zdFuI<Y#1w&8RV|4oJ=~=nSW&ax5fOofZIpf8ON*9{Jyt&1yl}1brBM*^$}-KGU`K~
z7$hnaa5=`_y?Y5C#0sS!MPdgP-tU^<j4EMJCeMprBtk|fR`ESP{l7;gen>C5oCi(9
z4Wkg^sR-|gM=2qkU{6cGP=~JvZtqozWsH3ISC8VNmyfEMFmU6fh2PI!#`T+$Z5&xg
zT$Owa;W*Tl9Q>;;-hHrq{CuSTd!J4Yr7b-lN*}tiia}TraXqzpg)x>L_Aog}(P=Cy
zAASH_IzU7RHo$H}?hpvkxUo0;;gkDO8bqmA_!51c762G&x}p>B5bG?Ke$hrD>6>v3
z(HAiiRPKJv^M;Bbk{b$*BHh+z*sdiDHMsDS-d<kf(1TtE!_?m*p*t|3e0;93$#?R;
z=_Zb3uc`@2+<2OU8z$t%cnzWk&#gjm>Ds9SU9nu}G<M$HvK(#Wi9|t55!BH`9Avom
zJKpf%US<(PlX~y-`!iu#**4mfFGDd^b#@&%vU56Qs1P(H7Z~;Zug|9);^BOGc6d?J
z<PB?Rrk}Q`!Sx1#@?JswF}rQ4Rn$#jvDWQ!8fb8|TlgnBP(HsM=@*}fns_b{&RUfQ
zL)l`ZP(&R)&U?vvr0b7Gm5j`F)RLc>f<LJ(r1ELynQ(SmB|&m0JMaYLB`d;i?6Wgd
z&7Rv$4Dm*O8ouws)1u<nPvVh-X)dvyTsj5Gv6uZ()HRhgx?tAaC%G9xt6%IG-09xm
zRO4d<`VbbsJiQTb7ukfRy~s{AO|VmXAz=vnq~JzLFdr%nJNoCqFQDwx5zmH`7D>8P
z2wMj->bXEYjxqzSBPq~0#QsF2+HX5vYCSV{Ro~f!#2FCU;JB$vRp&>l4(U@@${$|7
zAYYofNGHx?u3TIr`h~8QV6Xw>LkJVhN(Gkg+m=g~B)Rq<8yGyDf4biwd~u=Fm+!5X
zNfChekA8W{)s(pw=3e1>ORBd16gB}JOGSRZqdeK%S9AL(JhNXouBpSU^yh<VR6F_L
zp*p@rGQ-8OmIFyvbOhavEBLoxr{w%lu}0>e5-w%1gPvT3t@b@`O|W4w(JnLec6fyy
zt(gIfne==xLhxTXOQunzRf%yj0wom|Nu@=5&I>$-@{j*LHBV?A-r{g3gWj||5bIM9
zQZw_+S<Y?V2KN0L9a>jzI~`G53|K*<5^siDi-Wq^3o}J(BzEAmgo(JE!ZbXW!Nd2B
z69?g#iXq!l5N5<NJo3A16KZZ$KcHgH2oUMvA%zS+tauIYMRzdJy)YpFpe^&ifk4;r
z<PM?$<Wo1gkNg^6Vv3Jn-P(|jp{f=RckYIhd$<1f!DoX7pvj{#h%hZ%xkb|%c!A|X
z(V@pHN!w9R!=W;~ouM1EQb9yep;Vu)2<jjgdU-{QL{!NjZaiKVeL8<l!)~r9xKn&2
zT${)^9|l9qGHqCkZ@N5xqC;LSvO)uM8JP}8HiGv0?8A(B$7TXB3p+h<a=`MbL=|jK
zcG`}quu3`tYUfp+K4^2GJD2w>-mA95VvcGZ27aGep0q>$&A%GZxcZ}MP}onSDW#>i
zsfNaDkY?f)l#Nzz<Q97Cc{q=$ha;t_aAY+jxP26*(2aYhubzSIsbn&G^3zlT$D3fx
z9finS2s(Kc;EG6x?)MGeZhHpHdB``j5dRWG4{-4=^8D)2cKN?=74XmzVFbUAu|jVU
ze|cMEfWC8XIDRftNTrKlLhP2o0U`F|6gi=<VsXx@7CtIuWGkTFh6U?G=<L7v!Bj<E
z<f}TYOj)`;jB8mF!r6yfl_W1*{W^R#x2`Ld8Y$X#zJtWo!H-r6j;hKeNm8sPW&lt1
z9*KlI!Rhcg1+R1nnJrG!bhEf6SG*Joo4=gEx1v4v+w5T1b23F4Q|c1$%px`@)dOPV
z`Qg+^VGyN;MK8ew<uB!weF#sjhee~mnl#2R+A<HN-Rn#%pFT)19!W5?AnE7p+<}k`
zqLgssoYGRQxgHAI=3t^-WtH*}?otvdt58^DAJL-FNy?hfb}d8M!=I4^tWeNgjQyB@
z&Wp`%mS{(J^;Ou8Z&f%n3~`t8_W{A`4~2zJPcmUk+%1PG*v&UPavvRfHmF}=VA7hQ
z@3ZW6yz`_owlDAGSQ}U)Y0O<YRce9<F#owYD|-StO5ecW$GII&38K6VTUoC0bmSj^
z#27W$_0)U%R42Sr59N>G`EmZKAq-<+1PqQ_m=0OuA9)5Dy;bGmey<H>)8iHQpJx}J
zbO-~-sTQ~3NDV8<01*Bd>K-foL@IARrWKO!hh-myvKfSf!er~J>W9xBPUWSa8dFxG
z{poJUxHL~74C?k15%)Y>;I;qGxusfsA=Qi3?{^;UUn+W`m)Vdp*$%aUMZE=ClP`17
z!eIhU1^oq$zuzhA4?oPojgEmm`onIJ_hm&yM4@<z)TAHGe1IUf?SudLLD?|c{44MG
z!2$lP*1^00x(l6F<R~jg^0woGJy&OJDe;yC=u{D^-iH25mf6?C!XyYGyv$UO+ayk!
z8TSn$av00S3}dllzY#(H?D#L{_`rmQn={SA<9g3>S?EU+F&_@f6Wj17e%c6YHc(Ue
zqxm|~W@Pzv>^JfC5L(N%BTcj1+sqG6X3iIwZtoKk-z!p~h_|l8JH7qhWiPgxfQ4rB
zbv2_AhCG;zlP~haPg*EM;|5lV0A0Csf>31j3(%-YptjFe1$2I6!?gFUPL7?<#jZwJ
zGU(^*Bs|)-FH9m|LwS2EUSnakMbX4+#AG`8T?DdZGr-)bmlz##_Y-f7ZuWXwZM(<g
zvH$tGe5ce`vbVUM{N_a!$#LS?a~&v^&)No|Rc4)>-fB?Sr*L;AJ2_RKAop2wr`V9E
zZX(x{;=2S(%|C|{6%Xl%DrRf2M*{h4i+ClcRj#a$Hskf+jcRkioml`SOiWU8WC2EG
z0@2Yo^hTCyl~KTR(mWEoY979o%bN0m2i_gO?UvuL^O@h49glk5A)Ply-X8@FmtE2!
zun^S0k@r6sR@1IrX_!~B2h6FDycm<ST~FJT598_-D95FrBYhkhD}_1C4CeY2Gqsy@
zYw7UMnZ>{#T5}Qph5QkvM8b3Po4B3Hf0Pxm#%JZB0W4#U(v}r1IB9OVUW;wBMkq<?
zqJrrHSG_nVEd4n&a|Wqon*EKx{HxI2<~Lh+au>rnvPVnwo!%R+vPv&T{YuO=^`0so
z(1<2zfg%%1Xl@rs!C6h0nxVSa?>SiMz|_jSuI&*5n3hlq&MHoluXK;P*P9KuNv*L4
z@m`Q?eGFh{^IAlW!o!bIu7$g;QiW(`oxlwC;2;dmRVEO%qeeU7<nR-vw`rw0FAsSe
zo)yd>G)Y^%?-vT@{qz}}RhtnvaJ>3}lV?=?hN7P24G8q!bcvs0iwF%|=Mj)WM~$g6
zpm|YRU$jG+AOKxeRh$FT0^hV?;=s*~Yfro0tiG8=%5WQzD(-F_T@QAJbR5PM)eyxF
zpz)4CBI6rwuG#K7H8@mA@8Zii<!W26v9;nF9?A#b(m5JOu4Af~DG9U??TxP4D8~vE
z<l&4amkShLo85z*D8tK?Y<Mktzoi{i^nAsdqP)C=eg0}l2igq^7HrDMP)qcimqN;~
z48Vtp0*^rOeLL5<;p)+UX;b4p^aV<QWni=OFvaT1%jN+C<L#-6*UzSjmFBvk2i>$m
z?$F(N(2e>9X)(I`yS2=SkZZL2zekXqBcxDG*JDnLyB17K5YdPD;yN;l2G8;=)=rb}
zN=_E>hYz}h%RhAtJ-cNqSd13_tX|0~2n6FCN;M&#D|Xmk(b^EaXq%>e*&#&nD#i+&
zdU4K5?Qx;*w#dljS5wzeEQCwUBuJQG7Q^2pr<7JR?*cBaI31FXnnSH;jMyFY42Q(P
zv)`d_{6`^vkKdB76R^z~Es;&STyXia|6TEm93NA9@ba>&{4xRhJ7q@c+OpcV41#TB
zLx<Z|jSwZz5{Xp3$kpI=YSB*3(P*oeVe~*$4tjnt(_6|puFQzBV`TVLXJ>0y#$q$V
zAop`*GN0Tk?;r*J=##gMZ|pm*hX#TcI<O!y)`Q$zP5YKpPdCk2B4=g#x5~WG>{!AB
z1hAOEL5m9(P#HMIzA==j4!y`*<{=P$4tR@D?PX(vKu@hzP=T@el%8X~k$^Qh;Ns@5
zULk^Ro(E>X5{x|}2Iyf#Yjjs&g`A4!sg(w<?n@FTHCF@2ROS+|7lF1M5zl^|YKqqo
zylHHEw~t0;%!ci(>V&{tX=n1l)p<UQtD?%1nDqghs@sa`XJF&bHb$yHheNSGB73y+
zubHBHz2Qp~TJpAg8GOdXlC|{AxM-|rL=z?6;qcd0$=5bq+O0KVT8KYQV8vkN{JriC
zZ3m}BzW2fxta7ahv~+kvd0wguZF{4(?Vi|>hYVY}3wn=#z{SH&J;BqkgBM$_$$9C1
zUn22HGZTMdCt6QISs1K9e5r!YEnZ}Q-x-soD3Q0q`ZH}SWdQml(;E8mVxv>=9YxMp
zZ`39rqKgu&GSM9b&ujF4l1>nCD9%wa{UbH1YSz5s*sDadu1r|zqYs_g<~Gs%qCNhH
zTN#E{h57-T%e@i>HXI!vNdi!q(lv>jPSTW;PV`41jYoV^DqpHo%vPeG)GD93iDHI`
zl^eq`q7PW2;c1g}iJZyj26r^|nZxk<F(V)5Ry}6$5BU@8;=!DTKWH~gnS~kg!iwR%
z9Yvq)%iYq_l*7DyKEodmA*GOu9`K|p^k>;R0Rv4i44Z9exS-hW-t6f8J|_S;!nQlu
zg&eRkysE#xwX?r*N`B#Ck%WJy=oy8Y+>1q2{sLbYd~x@PDzw1w3WeM!aE&~4OlYR*
z74E2u%_->jJk^an@4V!2x4SfG{Mls3VjjKKTBV2Qu!VRTR@eQ)$qW2!V|jrwCS6+M
zjqbY-Hf`bHz*WnDK>5!RDoVR6PgT*V#Rg8Ka=82;5Rvgsf|dK@yF+ILplskFI@dSE
zQ-VRkUQF@~7zRtv$?yLD0AD5ln*v8Vw;w?`pV3gi>zR75wA4)v7rXp52I_$la>(EA
zFP@UvjFY~N7>hnTEU!^1s5l;Zr?Gh%=UzB$dL?XJ>)b#fzTv;BqK+4D(Ugu^Dz`Hk
z$Mrt*4iL-UF!xikZdhlT)?&bYAH?`yi}!O2<+-$ckdcxu_y-@Yn6-L_h4mO^sT9u<
zg8Pe7ePT8_d=h~$#@^>>O_njES3|*&-O81ALS-=c#Jty+V?tXLV%wwT*ZKr<J5P!E
z?+B#w>b0BPj=F)9%g{Uoxzf%yCG7IiOc-sk3McW-MWHvY)kC!SzNSrA9KgPIhPhaf
zTwVw}xF!9&il<367oo~*B*M_novvO!3%AEi70VVXb3m+TaEaRB9pymyai-1CPKj{v
zR>9$hNc?{6gn#V9;{`!HE1p}+sP2$&DOFGIWhks?*8UtK7WZ8(=U7({x08E!n-?MT
zVVV)3``ya0;QlF%TgG~IVd+TBY2I831Lg#!62D$Ec)q<A8BXG-@DT914dpuP9IavU
zbN`bW7QHD*nGngvsW-i?4^zk|E=56PsQa6rbhPMxI=11#P*M8`7=aLzbFNOvLyHdH
zCw`0Z@tGDP(x+110Iei1L2<*oYe*&WI!PEu`ph0mY~=RUr++GAl^$e|tRwdE%M`_>
z)vCMvf@%Q!?g4r&g`J^Q`10!HjwrJ7uT^iz$shHwIfuSFBCiOzuX38JUJg~cKhZ5-
z0k*!Pl9bU0^osonq9aIockEx|1144iVfnN8Vj&6TQ4C5c9*l>)nMF>0&e2>1-ag&a
zs4A5?Y^SV1K<u1PHsf)~-x|ByM?!>~T4ajI+AB5?+{<#FyaZ8CMq8?a_piwx!k_CY
z`{H1GFCB^a0mNEeGWj%)U(vldXbIl5OmPMAWyjrm)`CO@cHq7(1C*<}C=~+?@dsUB
z_&$gF-<KNsH;x~YAzx8A!zo_R!M%)VhU;hSraz^Nk4NVmrUi-)bOi-;tf42I;6pPf
zOyr{IwS2Fgd9FIbcnO;P{h5uk7AH@NK>Ag$M^SQjE~TW|>v0)40i?vhqH_C0R=}Em
zkJ6HF=w0Y!2*>aBUxL1i{#W3zvo~HVX_40{)e+AKNg*B+fw?jymzS<$Ru((6&WPIU
zue78!?qptpqUd@$f~YANqQ){+0|f%6%j9Rl+5E4LXYrJr+dQ)6cNCRFdqso_Gxn(D
zq6=jVTdkOXyCqf=sQ6iCd5&xQ$^(6_O@cyGziuTfa>@JO-%cG~)PKD}f^8`O_7evE
zo6GNxy`}GNFy*Oq!q?_nx=fdvKtl{CbMq0mv6EhCdzz2!6uC{hR^^vbv#Q`GjmQd9
zNn=B}`C1l<V98S8T!8t|A~vWSIls6*T3)T)4q+{P&Iof1%~PQeEvO_Sf1=9$Ge$dC
ztZAYAs~u+J4jz|t(khaw=fTKr6OPW<<&fJyO=Wwu;XY8!<{SfUdbrBi>0uPyz?G@N
z{Ph!Y*|D$sxgEj2`}!erR2gZ=gia$hpkL9Sf_+%1`|$(dvhT-81j4b1+zPf1dAv)G
zIoP{G?<qDES)u4OcJzpy&+0o&Kj(%~5Le+G4t|~38`8L!?L?~6{bx>YZ{U6ENW*of
z8-C}iyoBNu)aYh%btNytBC;F0)P>?*xp99HK*|_Lt*A+eofP{3Jlt$Xlqi2c#AkU;
zMF{IY2yAzk)!n_xUf&~R`qF68xj#u<hl-+SNFXEHk4PM4NY9Ab9h#Vs4~h;OB09wL
zOH=>EHnousA_6;3mz$vT?@`Y{6R~SxS1+%O)<e)ZV5|f64<bL92#wK%gTAk3n-?fI
zrt%MnNmD-NV@9S(NzzoGC&s;x6P0k7R{vM5!P5o4Q%M*T4$p7!n+`b>7gDBYX)ixJ
z<zo-5PeqIEwmQ!O%RaBqu&>Am#4g%zr@$F3g(JQg*cbwMjAPStatjWJnHDkW8pb2W
z?Xs$5Q9bc-tmrS7gx<0w%wzwlScSgI?<JT!vWKwp<*LV;$x7N&bPnI&I@Bec@B<VB
zh4(JI?=yXwTXhs3S}u5<Z}dU-H}l}o%8~7G9nQGv(@eioFUf$YBnb|DNWKQVipH&H
zI`p>Az7!SiO7ayVQ|?Wep~;NKS}1l6kFws;L^N^2#VQ31c};cgT$RRxzLeC6ho#p4
z`>xMirBPGi$6>%~4FBGF>(oSn&3~ma=B^tW?1fEP)I%>M;;%xoX&x`5;AjN_k4y30
zqeV4eFW0+aWGCY}gXnt#Z}0ckT2*gH#M+>VKLT?S5gPLL|LNkKmMc*dAR61YZQHhO
z+qQRX+qP}noM<w!ZQbW{KccI;t5z@PhFP@Gq_XpmG+q3jPnt*?Zlm?55V|{M)s^GY
z3KZ-SK*pp+A+d6~k||BF^Gq$ia4T=$t}JA4Em&keL%?Pf9_kP@M~rBuP>8F$H0|MR
z2xfR`=0^E%=nR9j^bS2e&K~BGVCaC&9Xf|kY_p3MtW^9wY6PVlMRlr2%T?i|Ue0uW
z#T-O;%L{%72?x9gA3=Z7ROcrY$Kp}Dp?aD}zg!?1I*RC4`cw+|8zf&8$wVg2dFSQi
zurvB@W9wf_U_|4URM6`0yCMOBsoheOmKc}!JNz_jutes@+0G?n*B2s%slFG8Pv<uf
z`)REhbS0_ejS2>xI!iuJcIRYOQz57o{}ep)K@ElHn{E1f6Vr?gkP9K;iBE@fPWs~p
z({VIJ_=jC51tRmV*shrAi4nx<O?rMR_pbG%2SVzh+eGqV-z@x6(JC4tk}-gy(!>_+
z&#!&LOsG{0N!+g8*Vcl0t<|F&9itIx9n3cwotwNEh#%SInAD6Ft|HtITu`*F&qlVU
z=JIGdk3FWOJ37F1UH3;GOF#`@YJ~;I>He5wP5gWu9j&`;#y{mtB#TtM(_2t#I>xg@
zEPVIr`SfXEREwyxjx89wn(rw+4JK>}C|b*m8a+ohJ#J&woc>SaVvDKITyrMHJS#P!
z!T<)hjq1GuMVgm^{dqSBF|Op)<^92!Cb&OvPU{3t<aZ%qP?Z+5^FdKQ;stO1=L0|D
zZYWAB(l)N1e(6HYD1%)t9x<x?k6qM`B*O2;Mjwpe)Sv?@8>BJMb0m<MAj=GCl50S#
z;JJ{zs%93WeQZolP{xZ4tL7ZcefcNbIK$p*gA#XSlpyn<#(n=z4Fd0}L`nES!G-4q
z2Qupnhyu%U;(K4SZN5XPDlEx%x_eaOABnPN>;b1Vo*{dX_z1mLutjpB#*9vrO>VRg
z+BQDpw#o@i(su6p%qxUc6xJCOv(CjYPw;_^R-pECm|<`IP*!s8z4`CbJJ+q0hg(w0
zwEDaCa$U<u2h?gR%`;z$XFVGpu;r&8(SXMIRTq4hth?Uso*12fuB@EHw>ca2bRuFA
zuj%SrBvNfC+rAzc^DKBEjTrjiP+I)3D~%1Pn7JI(saP_mO2hk9PQ^WjTjs1>Omvuu
zDrCAjGm)vcJP?UG&HX%=`PXUi!ktY!23C3kFBXq1&Ox&?%tHx3bQ&w7(OJPwXaJ@M
z*7X~tetND#(h6?vTuo0yw4K^Evi=y{a;p=?PS&yv*}e}aNS$O_6*=CNCo2*S6HyVC
zR_|zCSph>YiYrSjLsaFWloFt2c6GBae&;^3<20hUo!1h6xith0==DTipTnQ)oSiz!
z;gCA_zxxXc)mnuJMVLo2OpvMNH#;;T2P$Y0`Q+9hy%bsZNZ+~{k?ueBXk`gI?k9ak
zRyvF}`1H@8WZ96S1P%Y(pvSOSF3(_fXGVXvBTMf*yRlr_Au$Dwv*7!TaqXRstr_yO
zUh4fMY0ZI-ShTwOFmM$kt$xs0swIPQH{)tqV>NYqfc4PNR4$QuNOsaCgw#ntZMhuA
zXFY7~he8rF-#i4pK5&Q85jurW&Ik5-pX2IqH;19C+jLk>BrW!9NKGdY=$M^1H22Wi
zXI}lrvfte+w_X$#bgEGOy9y+BY1@R?!_j=XVW6{8+N_BX49%qEgk=voqJrF2v3dxI
z6y~Hl@#%bz)#`i;l&EwXpSd>Q35BJx_2dvs^$a>kr=pjsYb%N%-aUf|xKh`e<#u?L
zYDZd|+XF0ZsAsGCIuFygVak@9;(1Dh<9k1GxU$(uyda3-|0)OF1ACZ9hOowj0$+Md
z5Sq>(vIp2&yfw<VEqrq$E$lX8XUyq8HK0fe)dSe2*4Bib8r4E9Q_=EJFri?b2+!pO
zbc8qGA=Rp>u=JCs&XK?_SCPgIbt3}?&oFO77x1u){vvoBl=|kI^#yw;sy@`-N`d~i
zj&pHE<?9X8Z_I8^C1BF_Cc#N70E&)O8UxT*I}~rjk&f`>#Dz2<Gayv<P}F)P!)RQb
zMklVNVo#$qrwpYaZj?Xd^MB?IsP=x9kPGc~IO3C9k2BoFuNWGS0y&t*Gm~CUc~cwN
zQz+-0m!*=&LC{?pvmwQXm9O3M#~9gV@@P~u?2j$7H#_jD7X7=y;>~L>gCJ!WSa8Fa
zO=^Q&qub{$ECR_&ywNli?}yw4BQ_A?Cgpf5_bs$c7VQ1pdmQ0GC6NHN<cYV~^@Y_T
z_IQE~_b68+m@L|9`nzlaog%Whfgx~Q7V5cZwNu*7O`&?~pq}MyFAl;|au~10Y>wH+
z<e0V0Nvl%rxIj!3DrLCW{RTC9`#ckBX4w>5iXh`gUuI}7=}tnt`pG@ST2-~`A9k8?
z2Moj?M=4s%hx?1)VV1L^zpj1(9ijJhYl;ji6oCCDo{Zc{>ms@GjMEGo!f5^Lfn!>J
zz1GMjlRON8duY)cdvHz<9MAo*@i#R|(M&Y)DN(1^!-*kfyJp}&`oUEoTD+WUCB@Hr
z27dqO=F~r#RhCmtf(g4S!``dmE#{8Liuf<+Lvz@-(uaY2)J&@1`YJ<>^NIM;6B^N>
zi4J*X66(b{tfM0p-64`_lv0MZW??8pqiD2s?&uF{J9-<kOo>8Q49g?-{YN?+F2`gp
zR*`9rtuGuY7Rm4HT+)MAYC<7Eqgt{~O3MXS7~{1h|7v?W`{sRaIkH9O#fK-`_ncV`
zuUBJ&TG2>2k06lPawAVtut1ydd=KqAh~m>6k~5Z@hEDxRfL1I>Z!RH^P4lmkVnK?I
z`P}s0l9@8AukIQ*JD{eTERV=@G%(<w!(SQ)tW4?z(?NvL_m?ghWnU|x-qbJp2gV(w
z?5b)hg!fc|d50)ojl|NGf#U4351;hR^Sb`A#97*pcZ2Ip#}EH4zjuql?!R0G)5Yh*
zG3U>|q?IOolQa=7Wyl3Y6543c#yXQ5EB`SWqM7n%ba@YXmdD*fv90foy67pkdIFB#
zjVz{4z(&sIAI9x9U<o<WSsttcp_WYZsj4^$IdUkxJsy0EWB@LMNOqh8PG&ae1N^xT
z#^}ycEU_?&(%YH6&PVHlDq9C<hZsq6G8|hD++~0-l$Pe0+38MID(-m&49svr7}>e}
zG7OcT^#^aCcWjP`7-<XNwv9~&^EIlR4<%(M$PI2L>vhbOIbPjHvC=s8O(EV}GlqSQ
zl&nQAas^p8V7!<BJYxmuR(nkGb6uu`bY9P+A3JRmb4+O$N&Rr?(9`IB33S`7LWY=x
z{7eDPowyVVWr<lJrCBn7WMGmOMc8qp?<cDXJhs$KRYk6=k<8Wci?lJs2XiVNH{Q_M
zq1GKGvUqK{OtA}sGzVi$&zfgb$0+`{W<Vd?eD1#v=YELnli_%nXD3HEE2hG|c*9Ep
zQC2dSZuyBW?J>IH7RbtwsO0DO;qP4_-;LF<;eRE0G3f~l2Px)~-Uy{asqIi`*b1qV
z`#?gDXlQAqQh$7&qtKp3QST9)O`{_w)>gw|#-wr?dkIT#;Klyi-MN4DuCKF|8nd~z
ziNs_*lS@tHy1rXLHA-7U;ibp9zYazY<{l83UxwF_*pwgV$;{S8Eo81jG@y)d`5|Hm
z<$kmcTnu5|e4xxPwes@C^$>fg_Zd5y3Ol!$Z;pSxuB?QSbr7|bCC6Yz+ZM79MFp<v
zYSE;OjqK?wkqVOaN01mw=SnlHZ*iil(aKhI=i8k%MepkOJ|ScWw<$95xMml&qpRl1
zm2F3C8>$~jaMs~tScVT)K|?KKf<HQH5HBFb#lnK`b<Or|9HV}@IduZHgTS`ZecQ?y
z)6>knwNz-2>hVlw&xL5%KSt;-9GP(SY=Zq;|E}lozegVJ=QG4E-*cri6W@Z%HF?#K
zMUEjtEUe#%ARV%UC$~YFykd>1Y2qMj!h$>dm(R8k#Ru}lg6TGR`rs0Xe&N;TqV{{m
zid;M2X&ceUWKcVHC++jps<*6u_o{1v3&#O)PXU;PP?12~V|q3AE__|-CSuU?gq5n6
zMkXC93`Th9FN%kW(Je3TUsiivZ`~j%71F#9H$LtClsupk8}h}=l^}ejBiHnFV<pIG
zan-<q470;FK_vY3j<gDvi5pxf$F0t6%Rn(^QG7~4{+#{o+=5fUx_DAYI+UUpNm79_
z^{vbayV)30yq7@40jku;)1HdZ$+3xils{dcR83t5Nnu5ab3njty51%?oFv$%z>L<?
zQvdWT=aI7d_jKXn$@jISqGQ|(Wju*+Auk;2X^E|J9DH*!aX);^$RQBoxhA!w!PSBp
z{Gl*O5ZmB<{&c1re9cu8!0V2QHT;&BvbW2$^-)H9edRPwf{zj4&{Ny)gD>X*$8EA)
z?G$=6Zano^s?B>HYt(C>6k<H|47FZVk3(Uu$6GGO;lIZ(3md=DU4bh>Daxd^a4{Q>
zrItMI%6={s&u0JHjy|7hJuN&yp7=aRu<^7PKV642&|k<WwL)^XR-CZOK>Xv|r0}&j
z`3hlGNEY&NZ5@^n2%B!f-(AeXy`XUS)Im)g2$+zT6$7optd$FnF9N*}?8ro(>4IJZ
z{j^a7<Vu6l0C#&aGWGvqhcxN=;Tn$J>8S>419Rzpv;f;#RN#3v&PcO)`{D+7wv<3m
zne>N70Luz2(c?(taw{_+fNqfxf7pW~7O++=t5WAk+UP6P*mKj!MVEfBO$qTZpD*EC
ziT>0TyTT^wx0w@z0g^!S8WQ!Gq_mLJ`Fn($jn-%=UOmCq^I&(_I}_6B96N$DISYPS
z)W(aEmtV`c=rw?PMB%dJGqQPd&DSJX9DxKiY&FjAB$FbIV_f$XpF)~ocv1G(F+K-`
z+oMbWgq9ATB{^I!?DnU?|CE#}-cHZRIfmK>%b(X+(NN1XOGP|dZ8f0L>^#{aEoz<6
z1Q$<vT&`@{%7y(G+C62Dhc}fYW0<fjq2V$I)j-J)qXF10d*}JzAGfB<6%;+h0(_*@
ze|d)OiD2@FJ%5gawtl<EQXC18WWaD4Cm{229^L1~YgNi4%Xd=H-7WB%=*`zM(<0IZ
z$du>)T6HS0_1nVfR{cwOn7gQ=aV=r1mnrbB2{@ja)ZEH@je2CxxDoF%B-aQHY1CHi
z<0ATZ3?{&KtX+*;`<AE_Hc3%jb~}-?An5k#I6RfT0^G$@@q7)L8^<h)eMZ<ErK;#q
zm*+3;*f<&3>hBYEIjhg(j+EV4l6gB`h~C4u^XyS*m7ib3F2i&-c|6b+7$WBFS`ZMH
z`c>ULehDeg(P{}RZ9D<W>*povKsts^kUYVd=gsu0jt=EkkgY4<>z0O!JUA*_-~1()
zC#j9ftCWm}=fL*I@f_<E1c8Ob^A+~AOu-u_T5$XAz8tn8g>@vhisuARjhm*>q(X2S
z^uGr&mplp<r+hb&_mjKXEs6-6j3S#A%hsqxf8WRMtO$y)kB~C*h~Df_5zip9fM`f7
z_i&Lhe&kHMe&zzV`S+3|q*d#RxOl<8aU<w}N?`mT2b*~31cP~;898?rP(@x23XDa%
zNAxU)jAmnOQwRYA$@I&#V2eM=%(_@R&`286t`ah}`d0}rlb_ybvf0iaT6EG~tlOZw
zzhMqkSSY%4+iUo2mSPn~P;9&1#QiK@Hb{Y1r$+%#M{&z2zb|<dvnZP<8DN4Qb4vey
z;~>5T#c-JWO%aKuo7aJ@Rl{ylvSSiCl}>z449YRzQj`_trs1=IBkuP`7n8e9@o>HL
z8CtC?%Y0DjC+@;G&vOca@?;sJJqDSYlCQv3N^m5;4aq-ARffH3f(t}}AX=MFt`8xc
z5VgE0aV$wRRY6cocw})b1%W-`G2$DjMmSeBO$sA1?U3knK6qLhycnc5=1FRftPa>0
z$qU2k*Au#;P1KB?GqflSjBbyOdu-dbZQHhO+qP}nwr$(C-ru}#X02({Cq<HPZ+vJ6
z^G)5o;!f~Q=3bwihy&v%Ojhm04nmJ-L(zyfaoxuqs_9=u3n*<8#pf7N1J-6|nl=$K
zn|?{x4~vl~>Oaw92NX^<eLW>foBe%bx#2!NiNx|1??Nxa(zf%39Za{>x`bp&GoaJQ
zK%Xq!*4P*t*tbq2E+%HUP19&}n~WPDeQ(lU)M-l>GaaDS49CE9Y?ElIJy~k10Gic#
z7jekIS3$GY7JuY-atTlJRjhYi6m{Mcl_@q+WeFv_Vx=Hn91Aqqya=&?dLUjFjuJn(
zUgRV+M7xNEy!n#xvdkp$n;4|%v&D;BJCcipc|gX;mQnSpi4P2GURX2KG36%Mz##4B
z8us{IW74B>O7yIiYxb&)wUfFa2uA-e6%9V_^0=)XI69uOPiwT6=o{>02mBplW{7ny
z1z*z|Szh-o_i;GmIbXV={nLL+u*IE6HbUrF*~CU<OMJ3u)i;!iO9>(q6~pdxQd*3h
z>F)C)?1#O8MK|Zxk=Eh+3EB9$kDAT_Jh1gx<+)?Y3#iOl8zz_)_EFrSa&asJtjSW`
zGl8nS*l>f&+ahZ(9<Z)XY*1}EeYHA>)k$a>j0_6j<z4(8MI~8g;UpqPZ-f#=QDA6P
z;zlE${Z_23Xn5^SK%domfP^%6G043QmNYIpzUNb(v9jmYO8%^BY$c4&sBwRTLg>Gd
z)+$-3+A0X|DOH6`8M_g(8kklBbki6#KZ}KnJnfpL3*MARW+-6qGgD7B0FU2-e(280
zTBf;f+v6k(U}8qc4GoFgPrNu-L5!q0HxsZHm5J;GB|zIa(heq)&Jioz+oVTJ5@3oo
zS|M5sLb_2=>PI>H!TeNHA^uRJBaHw?XhRwFDX`va$ieJ?lGC{n#^!BlYLTO{@C1xY
zYcBTfNO4?YeNGs`*~Wh|C$fV|SszdulStLB`do4Qm(dsh_0omeK|ShS0c?t-@~}&2
zVFie9uf$Z|@~}b{dm^;N&NC#ZU8R3K3h*O<s4%M-{AgDuR@Fv0TuSkN-W1%zFXA-o
zQh~^jhonmb=T|OjmTr=ENh^q1faQ_g6Su#qzSXA@e|nB|#esx$@&jX0^zN#Pp#E34
zZb`{M-jZABgf+_r^=%?{fF|xkH9|4vixB<^&OZE|bpGmfRV3L`vmQGCXgQdMc&@>u
z@590`VJ$|B(V(zoZW_|Xz#=}d4!^iOJqXX+j)Vr2#b>;X^+l%(xKdh_g(Rz+*KeK_
z+Si&UvbLUl69{$sNDJz&UGooC#2b;Nh_<Ph#Z5K{$TDRsO_EbDM6-1G6vZh1o~BiG
zbk-X_`u{YlT?#s%2D2o+Rid6vlETAmo@eP5K9eM@D=?lzjz7<a%|b>U59k36FR)t3
zR>euEE7{ke949RW{6NyU0@0mNwFOMlyWy*ho)fSdSbCsh<Q*SFk@{<l7<_S+odqcO
z9MD?qK2b#JewbJ1xl;Es3IU(9U&0zZ%Yi~6>*V%6s4g)0HM*lugBP+eC*}afL@FG^
z`1L~y$k^9iX41f-E#8DK-U?O?XD|+J0Qs-+12@9q@nC>#y+vdHSvMRd;B=eYbA3I$
zixMp;Sz2Pu@r(ZmIgDdiJKDP*RFgi;W}Q3KdHq%zPXDH`YCawi@8Lp=Ek8}F<vT3z
zu|!#UWFH}CWd%WdA1bF5g&I*DtG!SpiyEx^iwr6el8Z=B0sCPhvw`Zj&iM(!YAMs*
z54cs5=w3SUC#l}&0!joad0i+w7_<%4;qmn43>fl#*M2+i2n>za#asW`#TUp+w`o#b
z^M3}(nM@yThHXK<<LMMb&u?20;d?4HNjynT211f%L>DuCuxb*lRoy2$%H+FHqXITv
zEJ@5ulp*UWl<(XqH5H2~EN><KHgQO-pphopR;VIFxa3^Tp_!D?b`t7A^0YG~?1dS8
zxUJMt*|~lrYThoKse!)?>2074lY}2fFQy+`!O7gn@YkA2VMJakvsTOqUF`)CFVlZW
zPTaMWP+*<9SM`A)0qP%6-NgeXuz>hqXz@3YQ_pZcz|6T?AzzpbuVvd;W1fkO8ryF;
z%S~H>Z8vCaaC~j~(y&&fc%1VOV<vvsc6OL63CI{mjP8fJ<`w<<#6$Xbz&E}ZOe=tR
z4mniSQJsh}`lAFRD6EDkm<#2&NJPz-dLG1xMhW=b*h*{2ZME}iqt_-?=ia3LLue|i
z3-e!Llkpbn%x(dKhyAyfy_CiI@o1EmcWYIG$d(U-&QBcb%J08IkZe**mMTRUcHl#b
z?0u)N-VT!}4b$!)dXt-gNpLd6ly^xyTE+b?8vajT#zU_N9<C*nR<4D$RpXP^E^t+@
zKLl6ffC_y2muoo&97Omxx)EWwd}DwahBw|vbS!$k#PX`;2`rw1xU0r)%m8fxs&F!$
z9UU>+y{@9v!L%;|h`I&IbQ_N4Dp6mAy!;bb`_SW+`IA8y_Ue4HmfaH@nIL$-PY+yh
zuUv!(oer8LLR;|4j7|;4^1rzE62On5O&4cuO)gBp%R6asG1KNE9H`z?+m!DpL)+dv
z;=nV-;{YW>J}C2U1>e#H961!I3|Gr_!Kwbu=A>D*hSOz(hP7>DewPIGAvF{MoIsx6
z>Xk3e3{6Dm04da4fL_Jz6d(YGiRc>F6{q4_(4#<0r0@2^tGwFhDr4kf92ZeEy^+fS
z<)m&JC*!H$A4xfImSg+qTQ3?qZPFQGYeVi0Vz}j!$PuA~LzM@^^@K`Oi=EX1?}=()
z52hs}@Y@^mRSNm^g_#L;;qECgHc~|d-`2+4DIsCl4i<T;eaMx0vVHw+M$8@59s@ve
zK*G}O!(OQKdnNbzk+bcjW1tgf?Eb1`^yrA!*n_=3<=&*v6l#RUKd}ZCL^%o>J+P`&
zT%XV;@8)5ueUFOfu{938<{4;|bB?Qq5mG`V+|FU}pdPi}p%~W7mVV<+{>u2P`ALtf
ztQr`z9IPD`EG*f?<-&S5$V%_P%mfcN3XmW<uT(+=eIOpGw%H+B<^rQAR;1w;R|VZS
zV^g*|8?pr^eT%3vi6ai;_t2A(f)ge!T-b6J@j6E2H)9`(BKlo;xoVn>ld^_VfKQ)^
zLzD^9U<+DjyR-_z$ZR|T3(kaqfLyhuHV2~d7^Td89E^*br7j7QjbZzuK)S%Hd0bez
zq$eu_?@{`3_UP{-4ITq*;jOm8(-5G#J0Z)=1ysX9OO(kWw`S4m2u#K#Es}b*F$v#?
zS`nAPfG~(GCTGrbMA=;w#ygA(*n}@R+(2Smk-Nk67zur^;a|<(=hlP1$fBEHf(whC
zOi^xzaoFF>Y@-q|5U+=C+os>|%$x9Lg0NkIE43*^Ml+huc)hoPdvt%3FjMO0ZraJ1
zNn|zc=^$iXA2Y3<=LmG1;d8~?ujHbzbcP#zSJ5idA&qK>Bim<GUYLDAdt4zd?}ivf
zhI5afso*hxbCs-maK~}<+@i=5d-(^iFJ{f3?!Q}Ua{Q2L#EmUs+kN9b0o-_4dh^B6
z!kr3UVsS@1UZ&39{6%w3YAsku&8Ci{-I|hy(R_-emYCkC3e?ir`1WXQn@e4;);I<6
zJ1^@=uY`c!qb)v4k%^=Lcw<wctg2b63krp6B=N$G&=?CvlzADVCV9=&2&b?oqMjvc
zCdLCL{N$r24p9yVp{HKk%D#%1QTqp3Rbq<|ix}tqvIY>s6p8vm;899AM=|!je^NU7
za!ZI1!8NH{G7_Hi4nGJrmVIy(@Uln579fP)tR+C@D|HF=CN%iE#KP%i%d*qwvj*Vs
zjR2$~Eyh9}DtRv)sM)v&)Qpxa%qEX=!-jNiy%X+9y%eil+z`|0)>{S3Ti!Z4g<_Br
zjA&nZbUO-<dsRFsmY!!k8>fMcr8*)RYZN*oN05>S=8Kk#M?Up@5L*?ccIAreyUAaF
z)@1-$0LSOhFh)6FQd`Czw0ywFKZ`-5lv<3W5_oRvVCU|WYJX1QNH7gb6pgv@$nLXo
zG&<kPcpa+VT*cko7PXqC_pIPNHmvnGy8k*9QdCytwW1%mUDU&^CEV$C`Sf;2*wO0m
zy;gDY>QNIi3p00BD&Ezj%r>B^7lljo*Q19=!#iJ0{Il_nrPC-wxx3dOV73Q{27~tH
zw&~jm+k)>yD)7dMA8Qu|8xuWJ*nI<t#R9UzH&swY<2F>nW&fhY@9Nq6b!&4AJ=OWh
z-be?K+g4I9o?-I@VnT;GWDiE`+Mm-ef6F*zI_M;M>R+<U()8ef$t#-_@?Rk%gtX5H
zHoQz%mkc@paq6%4yyS9<oeJ`#srINt441*EjR##?a%k}Z0>=+PfzkQO?#=RaD;bq(
z$Po0d8q$lKsL242;+nG@M?Oy%qH({f<W6TR7gtv$IY##R?2U8XZpPsQ=XmKPEQn~5
z-+yra2vH00K~-nkXrR&Z|1oTY)t^U;?7`-7h`K@!Mz71jq1>s>5|6AfF{$VExcJDa
z7z`E&$b+C06^~uvOzGUrTSO$YFXl{>;^dx55Gc5v6`^s{xs?E61~&bL@Cc*!@G``K
zL)Cd~cKPg_8tdAL_!H98C-Z=Joww*nD2uGbFPFyUDn7I6jbbQH{KoF>gNDYsR@a@s
zo*M{u&~%8{WGPftBF@x{6V`QfW&sL&)Llvt3`k*KU|kpU_WoOgF?~$^4q_GV<F!h#
zp@JeKgWqx^?`&Ci_^7}8mrbx2j&8K6$OI^R#G#{<|F*-|h)*h-_g8IUw1XIQw_QML
z5U>euk(e$TkT7cG8|TN(_%ndCKJMb~cWSu=S@>VaCMqG4X*)kW)m3f}TP0gpz$dAH
z+7#&Z${&QGIi0J7v26D%{88p04ChS_^IvzO@gkbyecG0zn`wh@E>%g{g{-mtm-#U&
zf1}*Rl4n@w0=BaFZyisJKd9kHYN)CKM}UXxZfS^>xHRA_9R$1^b-<+1(|(oIrR2y)
zFy&bvB-7Y8WFndbJgCuK<CjXWcF@U43*Q7kWidVx(HLTk`za#Tn?8`~F|;XFd(Ot)
zZh@LP!`aJSD9qNb@xzn|;a5=^wCwv%YfH)AQUlX9^yso)IN!<FsC^78TK=vlZyhJl
zT(e55^bGVSl5Zz6!|co4OL};?w+NTLSE)`ZaIbt9kW?Wdg#+&{OQ<Jcd^J%(a_Q>l
zKBfZRp*}j-bWzV3OAv!K!w+yn%Cr;6=EjMO;q^V0vXUIyR;%N<t7xV|?d`qZ=d=`F
zNgRNA|2~@u2vk)(+&fLP!I0yW2CA%|Kx=R<PIE?MisbUaUowaI{>^QwNu5x_Tskyx
za^Io6t-$@QELeF^VtXY7NO$;xg<7M>XEs%(nacG?YMa1fgo6`Y;pnyf6b%N?>e?Mm
z84GD)#osoQPqe3O?<OWM6d|MRad49<_-sV1PD?=BO;{N@8(G{KJaV>+I^oH_X%o&P
zne)bYn3g5DxHyxqm6R2snY4bkOj7$*p+XDBI0FNNHfOaRGdb2KQQU9s<~F7;b4)3Z
z5X-7>RQ1m9ZNY%^hMde88vY!%WU9r`li+r+Xm<m{{Em-d8ZXVwxe9%FVRuXswN&Iw
z`V64h%$7tAT2B|d>6%k$sYWQ!SxmiHJ}(gc8&)UPc0cH1UBSa}w*|74y$8WQ$MI6p
z!t<nk4XlTKg=(9Mb}t=t)FKnY;aup<uisrt;@IMO2}0@+6b%sK6^i?UUir>Fc3aCE
z<u<Go<PT5gtb<G#PZh(8LlSQd*y_NEEzEl>Sm%TmS9YF-rqg|cZD|2Wdfymbjn-WZ
z^%U#nxL)V#vF`=w8tXNMW+@BagHSKB*m=Ihtm*!zy>Og%Q=oGZb=uTyr`U=t8kGtw
zuZ_7!I^I~QDO)Vazgq5xTjn@Z4VmUP1$nnR_SOz^poN`NRamPTSrtmlmP|ydPI$GA
zZg~DZiQHG#81{rQuGvu8UAVdP!`6681Xrs?>cu^?IoC@a*+UwDnHCyX1S<M)lbHX_
zJ_WRkSM-~>0_?8(JwXz_1B8fT1oB(le_&P*9kMQS1|N4kl;Fu|)&xnb|8aiTUy}hP
z$>+36!~WHB4nNUYy(N1trE_ncp3t862#{2L9>5SbI@5aGF=DqB06Yf=GdK_YAxE#G
zyn#fqjGW|LbOk@ls6gLWiUz1EK7d*rGwirENMKkGE<fnx6BlB@eAyI%o?xibn-fXY
zvMijApeW0F+iDHXTG`#no`+{lb;l8UpLhYOjJ8vtBl-BVW}m&#=kHtxya^p*Ri9d&
zt?xndv>wYuW@r|=<%+<x#P!70N_Y%lZNC$d!^<=ri*6<207oQz{b9_Qb8Dz11ee-M
z*7|~y^iT7bhH{@tshv(pRC`mcBh#c&y+>`64(dPdO4afv1hmbMW12h%!nmCA^B~-#
z)Y`n(*PJ9~p%)Z=iXBnICre__ho;_TA}krMvDjdHYQJ&?%iHHhg^NIp<86B1cUQZg
zseL})zMyld^Fv%_A^@R^+lP_yrHORwYfvO<??8++h6D8P1IiMbaL7dkV*!O={tMiZ
zQ-+9pUE$!Y;qIrTD?8bgest!XBG$K4DDk(d#h)NULmYmhi)w`+lSQ1I^h*mjo^~iw
zsbL`J?+FojW2Xd>0<7G4)F*^v<8bLG5EI_#!%0A~bd7%V6O{dKqgDq87m%|F77Bd6
z2)lLn{SdSm>r7w-X|45w!od*v;%Sh6XplcQ{=Ao5)|9XhK?%b_#mEpg+Ot>rq4k}t
z_&d(RDc3}h4n<5ShR2|xP@9-ae0CqR!(E^uDYNn&8|Rawf)a`fX_@3BPhmAwnou1N
zUEl<sUMzjfdIsrQ0@1cPZCaraqiCMh4-JD6i^#pj@Keh$2Ts#AF=GwET{O=#{8eeY
zJ*q{T9h++LKszS6J~0;diTR|~yTl@^DYqC%t2lagPI+$&Z?P^e^Be-FjoG7_1}`5g
z`kc*Lyc`w;TVRbFKmZ3)rNP<5?hVc_yW*RgzousqyzFaCmO7*7!(gItj?<jhvMq%8
zKdFCU;o`2jJX{4rts&WsF`65Ig&qxbWC@jN56#RJJxsqP>L^%H49Vlz@+|#s`q8zr
zS=TS&Uzp?P2e4d~TSzV}6MWv$FICZ#$ZQPfeH4z9Bk>#{Y_}pNP-#!YZhW$87#YI%
zwCSBq1`;paJ})6#;SllZI`F{KkHJBo#7*)w8<g~hiqg5w6_g_%pX}<eYZY(1+{t%I
zf(<^=$iSiMgea=F3)kzeiO7?!`8Nl&2W(XQ-*_7SUpn!pZy=wNBi&9=v_aR1Xq=1i
zEy0$b9Py)3ljZ&14Mqx)w$(lDmqQ``4E@un-=ci$_8X9!Kp)KS<C!ns#ZFePz&Bac
z!c_3&MAEsWEQ&s-V!W|omEvn}1OFHxIT|+<JXtBtl>ev>7XU@-6z4v|v{*bHpT<e@
zM!hvO`Pls=x-T2?JwwL$W{fLjlM!l{AMQ6l7}PaEH#Xjq2s}2)yCM8;03^-vHYYL>
z{ykSIcq(vi%=9lre)=V3ZFo`o*)11z{xDbB9d<jCMmkzPkK3Yge#4xVads_PDYaDh
z<hedTrs48zR^bVxvconaqxZ}C$8xVy-RxLcoGllO=9bp(J-J3Z>jZ*38i2ONcXsBv
zH~tr~5rWcM_mCV?O7!%)&3QVZa{`m>5fsAChIZa)x@$u%VX=dwP7`QxnFlNKP4!rH
zxVYK8Sp`4l7LIa)f^ygm)ZEqJz?pC?wQ89z=ay2S=9LV5xhH%x?`p)YB|?Qzv4A)6
z*O_KmS?ayR9-RKklj3noAt?cYCC&uJ@HIt4Z{i7>f~q7Y<t!iTL8@8PZwPYO5a@P9
z(t$`I4+u_kij51X&MQP}YUpcpHlJK1Lk!9+Pf*IF7;r}PgR@peaX~s<%V!@6;605^
z@=g70yRMe?3b<Z8kZ;@|#GhdvxC-n5S?r5+>>ay&A!)jko$bBG4$=5fD{cZQYxqei
z4pGoIw?vK$1<gC0y!hJ(e-VdDD<giuXdiDhH$a5wuocFXidKf?o8QXhD_F!!xVUFi
z$moAZ->UC^VxD&xlDG6D*_AqHRn8k}t&NC8{f4di)n>1I6G^b9^*fbDJ{Z}G)p*g*
zX3ZTRFqFBbW={QiZkX@@1p%q_nqLa9$sLr`_9T<G{g!bZ{HOX_dFaIQe3%Kv$hmph
z^b-!J$@GCE^Y=`-H03*OMN;eCiprk@P?v)kx7IDn_d6EDnq2|X_Vh~#16HRR+-S#s
z82sd`QY;~joQL+)iWz3&3Aw$|wJ)BbXhFakwPTZ9k>43d_n+r)nP8hyL_3bf3K%)6
zc)B+zdM+ic(=&A18O-r7)5_%lWCtdM%nwn`osqwum6%0JHLh~nu5bgyM<(3zN6!I(
z+7HJM!_;A8j!PT^f(YPD!^!bCGXt=ULZsiga-bK6Ei~)miOhP#=N>f-5&R&bknH9T
zy}%JvJFp^?U}o}Js4Wn5{i1RT%IQ*#ZMMP^x1eY#0&CF5>`*8i=$?q7bjkaN<CHXV
zi|pENSBWd;8a+(qYni)+`~>P@ic4hFS$E;wdgQ4W@e_3J5CmAjh|bmKp2dF?H<g6U
z7-|3`rJyvUgN;P4`|=EFAHenGKm&`N9ETqk^U~l}OzAf*5+TeoH|FE(^c;1Z(q<+{
zlTAd8ksCvjv7@}V;^^h|Cbe<jxeV-X<Y|sfeq>!=Q!{=^ofEF##1POR!6+-A%!mit
zU;rAdYnM#iMDNPmq+bAKJ6@l?ib#3{&IueE*aDF`HDyp6aPghhyVArhCksyu7t}yH
z>#hDMN2`kE4AdW)gS_Jyx%!U!@9dM2I60Fx;DXYv%JvDzFi?fTHI}R#>#<w6=$Ek5
zvws8ZEbt2~RVG%{&33-#NI(E(3#TAEqO=%012rCNA9#D3d>z|sY_B{#kK9LDh>-ni
zm=fvfZbD2|P#-d3PPufM2;Huj&N4J3LwOF;Zfp3=H9iWNNfW1_FYfLm0;t{a6V^}P
ztzl>*cwsUsaO?(iC;fg{Na$%iDjWgd^&Qc^pDa6`qT_{b#Bsq5?%1-;`vl-&$+?YJ
z*>JW1dFGqFX}r_ntKSm18e>_8ONL?NcYRsGMskJC5Kn24%YRGVi8jz^4C`$<kX+qB
zS%9qq%lh25h;uNyqb~@B0++oQ@K`-l>X>80pfY6tf_cM~)2_I{DxO#j(!ev8KjkdG
zfpxjQhh5^j0U1JADy61wgBpYTD>^-mt%m491N>f_dGF|UW7Z>(&w4;U*=hhBeonW*
z(Rigs?LWvg>w_w@{AT75@RoT%JBDT7w_EVY-CJn)f`Aq454BAvTd*kTE(2O!L^0Kr
z^x_vFqZY(iX4c?2YB_pdi=1kC5bV;EGS?{#*|dO4pArRpP{BdSogPb#PFS(s@R(r3
z6I@0+a@La2a}>b~OtOP=(NLzNyQdY)*u#VZS1cHj=6%3D!q6YZ7O~Wg_AZRxOb<Zj
zZv&s&B+12E&)h9mBo}&}P_aNuu&?{9_CL2CQ#XdLsw^M!CBV0Vfxh%2H%sP_S}%*P
zw|$q9iW~%}e$`h=HBM3!*2Av8Vd<U<xj6e*hN1|z?ck$8FGaFp8V6Quq#^rDAk|I<
z77L^TSRjNUCc&=WOpuUZ6U^Nl#T4Q{|K97HC+LIn*6NPRRv^XTd4;XPHKO165)-V^
zS@2oG&Cb~PxP|W+T?|r!7jAzid;;Zr8)3^EatKedUIhB+F#F~N2Jd7sIRT-F8~iw}
z?OY%qh(`KLNK&_xVhjX|#sMGS`B|}E!vw&Ug$)<~<V!K`x&PKh1)AD&LAGl*ZH`}I
z`WZ_a3QVB#NCVzXU7x>`LQI`!Fp!&Y@(Kvl`DGiVlETVTC7}as?spb9*;&3>5Czf*
z=BfTKX^w@?)s0{Jvj<Una++{t@0+SVB+WSuNbiYyA4l$UddzpNv$88eE*@<iC(s{K
z4p6`#2}E)=Ybw0FLT2RXrfc86x_xOIKI|Gi{9K6V;10aUAOnMl`EiVyGzS9Jr|SU}
z)$W1uvzNsG3KuU+Z9%>;yN;oLD|0pY0jk!!-3m+VROlB)KWsSnwH{jfL>nU(6+lOy
zYI6nCgSzS~<L3DnLQ(nqVo#<n+n>D=flwbJm&|x@7`zH}k$=ln-%5afY8lrlW;-5Z
zH>z3l1=q(3fAE!b{Do*L2Kmx@sEb8Ve`u^#f9;n{J(to^%dnTuIi69iM278sbf~ne
z)%M!cGO*paSsviGF(`8yh~&l=y&*3$Om84=ytb??5uS3(lXkH+Q^rdIAbEt!2xeke
zYH4$H_;EORtjr<8l3dM^k}+FF1@lr#Q>3qxm!%?)=<vD?&Fb_4b#)z;?VVs9(=XP%
z(l!9i7~gM0exCuy((vBfp6ZReg?p7t!i+<|tMdSkVo~<Vx+534DgpDV*+}>+UlXG)
zG7Ibe6&!A$uJ}%Zim{f5yXesZDjYu)=Yn2*y#CO^QI34(yx9~4kIGrGs1~)dZfG85
z>E9~a%LD!x0+RNvx#p^2NcL){tTgT5(jzvo&#epqG$H(6nu5<`%jd*^g%^;1fo+}8
zjxXnCCjVT|lRiM@ima`H!hUV~E8LlVW-O`rL2#(IV$%_G=0RKSR2wo)*Y{1-Fp|GH
z%IRQ|!+}qAIARi15J!JbG~po`-=3aTK8fTa(P-<`ZQS;_<i+DVdn(uM)eyrty#eDG
zdC^tu{@jJKoGp;({Ci<852xSpNM1|#KA<6VnCcQV@h~O(?}kxHHO&{dSqsFOR*5k*
z*zg}LczV{W46POKXIkGEugT=M>+W}VBz`YyCO<}W!vbx>==3DbYvCa94-iXig!X?F
zS2F*P;!1}9ucu_hXJY<e#Fb1ejO<MRXK^K{JjqNWO>`H`ZM}or+Q5zum>DpZTid{n
zjlJ8OfxVlXnY~-vTaMS^?_1xQNcVYo>|{y@BbS<*iZBXEMI&ZWct#JT{NnCxNJ2_P
z2z<Wj-K8-UauXXcCD#ViWIMpG0D?Q9DZUb#Kjb`+aajL%oxX^X7QVlRqJV^^Ov>-y
zVLIT<9}cOJnJxVO2iw;4%0VAfJ77+>z^f}jreER`P`3Z5>8mSi?q4azW|zNoz<*UT
zQ2%VeQ2;bDBc}YCh9s8$UF>^3JQ(>WXIH>YPGI9}>6;wv>zrsEe^Y>Xoc<9+i1~Wp
zpD9JB5Ag3WTeDH?JCORnu_F-r7Pl5B2Oz+ljNhi(^q!zHr18N;Q2r-AAy!mX)KI^k
zjbA*I3;;q_R)7^hglxb{ezW|fExvuf(=_H*&<*$BG~2(Otv+PnZ=wyq(7<dG0bG72
zllZ8L6TY4<pg-8DcJwWOX_0GcYkvSy)m6VBOjlEWGOTzU`HfwS<3~CL3=IwS-=Q*F
zNXC{{U`{Sz_P-mT&5R#0nFv2jM&J#NZz&coCL^+@KlpdLwnm2WS&^M3KnVShyuL3z
z)~mmDsL?<1E5Lap149FYAF}(rvp%uUKLy#Dk%?W*a!^L6hVMT!WR#XS0C%Rwhk8JC
z4t5Rz?rLlRRa9Eky+0W?7Z*pM&5rN2=FhT}zt$OnKXPCYPXHMN)wFAwkD!`GCclp5
zgJU?EaI3ZiL8<yKc-Lt45UyPR^X4FFnAGwKV}x63j3WdQr`w0fitTRQ6K~c{zlghx
z@sY52!ui!-7Lk40w__0Hgu)YVwOfAo24`+^DSQi;wr}L#P#kaWkQHMFL+kJWbNzl#
z#f$!2(!@)Idgeso^WKjCz6jy=sX-ye$|guG(1rw@wPCCwbKRQE&tTBLP&U4ULC((8
zfRtSy%2S4XxkAS;-zN9s@;$>m7z|(fZ*u>QtoFy<<erWS(Ra_&N1q~{3$u24)UXFj
zoYsojz`IjYyVmrfs3uocjIfvK&#6k7K^J54TbQ0V%z?4thuZBb%1BNykRibHILRCB
zp;cRkhed^BM&if;-&vWL(`30fhHzOPjc3eHv_vns7dTW55bPDWXo37Y+HE>F8bA5f
zKBsKvz-7u|4o6TL)R;Q2G0-(6N{m4M$Q}4E>@^0GXn}GC`j6KDm4N$BZxu9ueqz1B
zx7lCZU&TI$V@<-=4xx%jNrSq0-uyW^kGf{5+Y`EEPv!Q|0fZVu0JE@KFOzM-`5PH_
zO$IAOEZ+=+^8mn?^cDK=F!AHI8wzq4Vn{7Vk?x9bUnCNvN3Bf{bG#t?Bd$VtAX6`p
z0Ju-&2bXeE)`U21G;q150lmINtGzuPcA<1foV`=tr2jZ{%CSGDU4u+JOU_CG+*B%0
zS<1&VJW3eW5kHOedblq8a-6Nyvld(R&Z2y<C2{h_$I7Jvsg5AQ=w)azl9AjNJ(s0<
zQ~uyba3S+KLDKLfIBqeBlN0+S89wRm+XF^&>TZ{sS&`w4YpFSP?i`FOU(0tq8N2ts
zWmUwCKAEL)>nK4^LWBWb6VCXmj2eX~1ti6Z>52F6U6`kO`i+=yMyD4eAZn(t>yLb;
zBm{dScEL;P&~qP(b=wIZ6bM_E>*ElwyF%<J;QB3H2q`?c5DUYpc47^7vz(JG;37cO
zu=R2Et7Wr;4{jS^@+W*M+WHpJ=H1RcUTS1W)_B`CTf`7%((ApG&-^zU3mT|UW7jUL
zb4M{P6Z570A#SpRY}I(3eK*mSMLze8b>A7IEct|pe(9Xuf0=Vvf)*PR@!j2Yp;8_Q
zHQ6QojF`f*nld%;7nQfKE-a;Ke=}$h+p^nLzVqe!CsJMAzKJnQy#*eI9;Ff-qL98Y
z-kMe(Zb%0RE|<B|ZS(2T=Y=N^<p?(l&`CfTC-TK;pRa-PvG-KDo2-UZiiuapG^1;!
zhQ(x=z~<MMg$G2v^gN&!!A32@#AB5Toy&QnKEG~k#LHUs?vtf72KZfD-}ntXq*jQ4
z5><N$!*>Z4jdcD~*&Rdaxcvve1Q<eVz8;st<lT5{Jw(+Qy1%9rH@0Ca{-IuzK{ZYl
z4>W7qSHSgrhpWdLvwNDEqlvHD;u=jQLZO{?5Mv={JDqmmT#sZ=(}Nn9C1{&wrvwu%
zot&M>uUqP9#q*U-j6nC+D(>B~v^*pc`H?b<t;!B|_|!MCFvZXOvg00o8yzG(Sod3G
zEJ7Nmyqi(I-6e6AwkGWyO5#h4_rtnSCp=P=U)+>9-JrWl)<gipVo({ROhvq0^OY)R
zLg7AA@MI@dH0arV^sX0@N=4)g4VndYwMzma?a^bjMp+;-Sj5>y+6R^4?O~8VWN&a9
z@%T}ao;;B$44Xk%jm-0~68u_6G3IxL<^0Aa`7@Ipw4<fY@v6!B!4fVAwYvBSlJxTZ
zV1VvTRR9qMT=2>qZpLs?brC61;v1;7G)yf9)blRwaAIcPgny)~>e(``mSZ`UTnV)8
zE94)Md~PQT`Ia;#MsRFD`SqC-{(QA1nxr8Z*<;L8BNNP3)!+K23`Hnr;zmrGA{xbS
z+K`RgUAJLT@|$yHp(&&2A!VOWmbDW5+hlok+?d{YDBpm0>tY<-!%OFPl>UP!x+Lyy
zoMm6X_lC2>VywcG-wU0TWHM700@R3r-$mF|k#|;`F(!;^i++oC&odY3mc)C-bz?aw
zmk)*Q6GKzWvPnTK$2I^%6?b^OoU7hVi*CVr_1A>~fbKi0#=LUx*Z@r>NFs3=d|@w<
zg-A{Vqc_@pis>1e1~IftTLD{^nb}OyDcIh+M`lO3Fwk{xFGh!kf;RkF1@qsBv-YP9
zQP=LJ>?@&<vt^6c$uJgwma{!T<F_kK5_*;|D6ejzIoWGhnoZP?bthjM3>XCTJCvd4
z0ovH7d)u=baqNCaNzwMp+;iORI@J9HQ%wzBpr_H`EEE#B+^9jvkMhqji|iGZ&1gvM
zemcSu;l+-VA>h;T`K+B-FgdDLkp!bX%Xx@u7%_Y+__(Gnh`7xmz7*omgX$UFI&PS7
z@x{*GASf9Z@;ed`Hiz927->LUxs{-VNv|X;3Y8`mIBUo*2w6FGH)^^7J#~9U!5QmL
z1ox`x*&LgL?A1PI_lq_FrS1x=n=fxzqU9_=8?_-S2Ni)xoN=A=pswT|sfP8P0O&mg
zOO*u6*bi5m9!5T+WhGu!eKHkwQy-G&Z}FL*+VKHQP|^oErMhnEw=6wV*n}_1NVqk{
zccsFwEMry?#gE&#@>HtnQP`3r)(fUvPf-$CO_x$pb7@ruv#5&Vk-p)nGt++fdc<-Q
zxzu>NTk^dPIO?+c@AEg)gnRG9J^XiN=;h{dXqEc>KEpr2s2Z6Gpv=T*V|rW1x@pZ?
zn0Ge*6M4?dqcYk|O<KdF1|T((-`Y(9K7w27<T=|w2{D(mRyrvJw`U70<<iN!58Nlh
zjiKzm;ox#PA7W)9QH|kt8>enHXszPgLqR~EtWaW?hH9f|nl_X78#aL%i)XJF!Xnva
zFLs|M4U*RemLQ*JkJntw?9bby9;RRcA}O)h5&3R=Hgx~)Ym>x}R5_0bbwZ8ro9Ryd
zft7lI;JeXW%Lj;MNj9$uc)8*RKHo3(%-m>%RHn;9hPk?Hu4oJr?$iIE8Fd_lNW0fq
z;9GW^Z0W4U-n|O@z0fbdwF{x$cnYg<E);HG4gfVh$q|{<CzKa#`pvM(UqzlG;!Q0^
zWdoi;+QHw?^ijig*e)ULpPDbKnVKP8QRg@sR0@!w+@?S!IGtLBCPvdU+j40YI{=&|
zAo?OCU2SJC_M$_xs{7Mt?@**j1;AbUScE3Z_{WqisTn0Cvy2bgS_s2Mh#60Fc}N>D
z4mk-Xy!sjLa~yj<seGG(tyfYyEb4mli>%?OZ0&%J4en(YVs~*8+eL`$x7N|Bd|<{9
zn0en-tCzi#*vi}9dQN0kgoa02dDO6<8OVy%B{Zn{&K!Jo=(dt?EX#J30tGB5iC*QX
z{k5d!h(eX*c|v7o^RS_20=j69A=RI~$H317H_JanNg}u1<Pv8=$w>+<L)m!kRZ9t_
z&yJFpU?OGl{<uA1^J;lXc6J+U65$umZ;gD%#H-v(c2rgT{phhPsR3wl;zr20sppZf
z5^%ddn}Ump71LQfflpd16%VqXQj+U4ljQL2q=AmuJ><-Q4GSJqw#?!a)K};k`6T-s
z;2w^Io)vYIeJd(y3Y`L_Ci16F<YDQRP_OrC7WB8etb;fwmH=JTwGP13(<&+gVa=Yl
zHOWT@bc6&Sqgvvf_dwd<ViicVKI<tXvAEKtSH|axPLXK7zc?Lj;B2+U92>*d5~ehG
z)tc6z8I-HD@)JKB2po4GjPU}PP=b_phkyiX9ukBeLmS2NIiR}Z&O5I>?xMPa886eW
z&*{l$1w1UCM>44c$a~Y=T8#l#*(@{WnjaUSIJARiNp9*JGO}LxUk+5!IPfufJ#zic
zR}n2-0=Rj-GpGY7#&vV>pXaeIPZZkT^`Ro^Jrht4opdEVZi(xglf@e?{$Y0TXCksF
z;8?#;kg^*qzF@20y8<(==tsPrtGuz*2{-~aUKCb<pElTiK?O|{7};YmnDk3`s6X2&
z?a3C~<EvK8wIZ3Oz2w!e4_7y%_Z-K7903hfs)QVTy))eYeXL5X5hRs`sHZ{8Z{X=-
z_`qW`pm4VKeOYaeQ)*VG@q#vEku?S!UjeOjB9=-OvG%5;V+EQ?`T0XHgQqk9sIZ|W
zQ`o`df)K?|u4NMISwCVbcQeQ$YuI{>l1_<SA_$I{x_&~_hr*G^^@ln$plrihvpFyG
zRUiQUz^RNCozr84m_o+PFk;R9WfCA9Q#Aqcr8K3r44;!V<FX8b`@JSmC3l`cc<D}9
zn~f8^ApcApD9A_{Yz3B&cLXhCp{R^r8MElPO{BqAFFDzVd~VEOg07$Kt9Ca09fn^k
z6PLj++!x`g)Z+UT;#ZF&A#j?GtN)v&m*RYwK8Jf+8b#Sv;oYY1eFr&xT?wQ(UkjrX
zvYi|tK<`q>U0EVYSUrJF`e}3gQQ(C7e(Z*y_~h@XnaOx7p&@5&5Npm}NJq|$)3@fG
z`2W2xEyz~tBj=Y$lmM=&ls02ZcHh&&wmS_)VJUL9PzsmZW~L5Y-TCzsm$>t@&9#HA
z;gmzSR-6c#F_u&EMK?gG4iKeN^HqfU-YLfPsAdpNgI28`xi|7kx)|cbkPrMiej9MJ
z(32l}q<h5Zu0IK6t>IZQhet<0wrX6NXmmNLu6V8v>nMj5q1(=7grPMM8X_L8qUBC;
z**49y+PMQ-O1Njykp{yR*Np_lRg!H0WI5-!IOrW+n9#Q&%K#;T%9wB5ofj*X)rLj$
z^KrGA<1uf>BL-;FZxwwjyItr?HUHutjePl%%95%Ui+W{;yGGq~D{k%IKao!%mTVxr
z;1ns+2#St-SF{vKF;L<=S}E4PNy9tWi4Xin#{ZE=Y=r&=?08M)>z{wO8LO*)fRj>9
z=>d{}e8@$P)rML#h~uhdSvXAX6dIW)qrgrA8U6#UJ*MsUbjY35JtF$yPXRoRyz^9T
zGcwBy6E^br+I|IV6ah|ps~>U9CdNha{j)0~X9PUDZX?!OmL=5{6Vb*=bOsI$97p1W
zUZgw4a{7B9<K_;HNrAM`8CL02_rXLWt^z@`{kWSO_BEnslu9U$e=65J4Pcq{37tQ5
zJgvAi#P0zB+w%5oXD>Lr+Yu8t(V(RB=_3yF{*OnqBIu=>^ju4^aBAQ(Nln4gD^lxd
z)PB;YWQF7IOUlF=$NKk*f*9T!62k&syH%Po_QS6KiZcc`E-qk?MB{BfZ9tk?`ctnT
z+}}=bmy&$;v83--z;2{t7P$_jbf-$ruk_Q_4*fJj7-o^I*yt49jt)nAUnvpiT5Ycq
z#Y@A3X%fb9YA3ptrZExHn|DBn09nznku{2dFj>Io5G`vs`|NN{K*5GOsB#g0>ZJw~
zEz_-fjV2^R9O0-M`=&uFGAbp#5pNdKlh0$2-pT~lAqM_QxHu>=V+o+}n2Nr#1ION$
z-N-Agu~pImVO}&*P3aq&aOEzKSqI%6Hg?eN&X9G*bQk)CD4^YTf>M_gyeej=Z#wy2
zADJ$FSyJnCygYU4Fqc(n`9&|xCtGKG^T?RQHAagWOhrTzUTJYN$i{uiVvVFCL>G`G
zVaV|ja?d%4@w*jk84W&ZHnjh|lB(kd_<hQXdNX8<6S7z+7rF->2t;Dtp!xvMV1+RI
zFSUp-$<&@}BhKS$QtWcT8a8nVFd!NsIQ+He_eybJZGO<3LXebNUA+(aU`PJxlqCk-
zKQ1RGA}oxxVZjkecGwmW%o8>YvdBzOWO?dM^$|A6XO%Q{g@ro&aoJBUr|?Bh5laHh
zv)VC@GpSHx@OzH8ka?8O4d3b;PzQS((g;n_86TlMQ#29(q1Ij9&sMpGk5`*ws0t<#
z8+XAcTV@x7+H8GUmeDU3AVOeZ`23&lKzI^r*45kz_*d_9N*~S-h$%inZ(K7}7^w;t
zJ1K822N2%^C8(I$fBfL&MY(_?IH(1aXPPq^-R_F*tF9%byp#md47Kaa&W?I$DSpeT
z(vpH)JuM|2@S^`1IcHckqU1uib{mwBDykq3-X$^7)S_Rc5F;1;c=QJ16ywnA?(l+r
z(TQ{!2ul0I<P}e7TzkrltD1p(bQHF`ER@>{sZ~oOv6VXrsO}=?DP43&97L<Pf@Ywc
zA@3u>7}<&|Q0{H;aKfE>zD3)(>c31EUJx2ig`W9unHETrn9`NUXd!TXTa{jD@!f$X
zLNd{FB1b#H$kZw0+-}2i=Aqs3>|}lje0}GVWJZbhD^y5)^7V3{Ns}5h;EkTkI>X0w
zcJ?=LtPpF+rGRTfu#M?c$InBJFehFbU>boWr(=%Zj|MTb4%q_GW+$K$+NDCx0e))R
zh@efvGWVe2TctTjMeb9BxFZ_|FaP35C^hq3*|h%%gOs#o7Q)tsPy#ld)?%bjDc*)K
zS0cHzB2YTAq<bc!`TtRVv<4+1S5z;FK9X!%>AyA0KSL`Wg~U1J#9QX%bj9b*nK5oc
zsi`wqm0<Sk7%hZpo}=1*b6o8cVL^5s^wCvj9?(_6qqizZ#VSzLT_R{mjC-+GK8(u8
zqPx~FIVx9|eTKR@mIpdA+9=SBl4(ki$5)fVpOij}SKfrmOySjJq!75lmORZolWk{7
z0-0FG05rc39`dkW#reD+jA?S%oNmzZ|C8Rz1Sq(Zll{QZ-o$Ak_f~+3!6cisAODX$
zQFV`cLP%GE*v9~SN0;$(1aYeBQ%@x?i<<FDmZF_q`l~)*2|~neJ$^vsQ@SWd@%290
zAa-SPD3QorVp4RvcftlHE6EdfZZm+YN--HIE!CciuzbwWO<qknz^|Bv!MW@E$Ha6e
zTG&HvTcwEQl&CX`h+;eH47Q|%XB27*bGCa4HkKDu9Eq@f4cDt_W&aEu?mCUk1q(i!
z&R}3Vz3(M^(Q3xr`yQ0=Iy$XUnp(63CK_Y)kaG;=FF@6uY?Q+G4!exOArYwNs;L+&
z3wcdiADyy<F%UN=Kb1?@qjmRF4fGtH6p9;|y2ls9pMtR1LzyT-kYjznv(f6I!rg_<
z?ECSbDLqJ}7&w;F*gCT5t9dkM_@6P9Aw~(GP|&!GOA5ND9O#U5)x_=4+3>(#CBoz4
z5KSDtJeD4<yg~E=onqnz2P|#b=|`N=HL`HyrW0pPf0aLExYuS?6p2Y%co+&>&ktn6
zWChWO4>^OZT?W|6_2X@2cT3z%*v6?1+weGU>Jp>(i9QpdSa|iA%+nnC<~r24-9N$`
zGEhO@70TF!7R{Cy4T?8TSD?j;MLK)J47hb@e|EDwY8B+8b=Y+L#e%~L^FS>{NNW7F
zv8f_WzT)T63n<i|q`+dL|8g<}3XX?TSa5kU%noYlw1A(7`4pEAKF`?~5oyZ#wipkG
zAELP`DHF;4h!<2QKJpvS7)UQsq2F%e`*(S}4^!Bgz*&B{@DboRf?SNd|KWMz2}yNa
z1-87*hlSgf!4>!3_h6vvtb52wU_JB~y^E`afzd=hwaYEdl3QW8a57H6V13GvEO9^5
z?Z@FZoA?Mf6YA)eAaJy*9bsA|=a<EY7Y-tCt9B?vg9Sw<HYGE&(xQ_0*m_HtHP>;f
zQfR@v&c89|7I+~${pQdBGgam!!RJ*(CVnfFl>Iv8cR_vSvA3J$W40epCA4-`OXT4b
zZg=n%oJ9y2XSlnSl(h!PC&(-b2@p+1R@1Kp<_kGi>#rCHON<a=@t3)SV`B=)b1^^Z
z4i)}>0b*{gRVJ>zHd-WE9H1Ay;2`yNGiZ1t_a@+BMIGN9L55uR^DK4VIN1$-)f_g?
zCHechBc_PU10su752mDj29V(K@jQDp2cwB$0-8L3oNG(VDP6+NEmnu?!b!T`N|%Rw
z1=Ae|7EV<N>73h8;V?`%Wy|^LX4A=LH~+5SuHg2xN6kl!;dLgnAG1h7eoBiv#<u3y
z?D36cU8cUvUsFFr8I3fDho=1_c|wYVB)&Wv=c_9_B41sv9c4Z0vqaoG;|-{3Jk#oE
zA%EpGltg4+w&JTk+_8|nXRMWdeF3-DDO7Gm&~4)%qqt@tJO$tJzD{8x*F@&T0>#fD
zwxhR3agFACkM}1ck{K`LaQ%!E60g4wWoGmut)cJUEs)cDsqarpK|Y9w3PE8U2>0zq
z6jJ;k9~QFYTdNAPrtGhM^lv!b{GHrehWKI%fSuPIf!mGfRei=kH<-C}=H!5KC-9se
zE5qM<nk&{S+{0DuPuEc+5}$TpfKj`$E17I~N@{a8a9~oSpA8a#Dz<KiM6a76(3iK6
zIo9n*Dl8}xLI~8!vp6VpdvES7EKhJoD~)^BlPlN3ayH*l^-4K#jS-C1*cA$(H5Sk}
zN<&)GF;rF&MrSknF*o;3nm0YEU}DVgn%CinM2%;A9OEO_x{J+zyR%AWeF9h(WFvw9
zus%0^@a_kVUjC(yx|P;`zUMwZj_pL+43S#8^#Pf^!<pFZ+3<7OaalYu%z#kHiry){
zVWRe1#f*YdQYC<$S?7#-{?qOHvWW!(eWU4XY{gE`2(_7ur^;mUgRV%u`o0}nInJxS
zJa;01GlIz3)P+sWOlda`!`X|xv8gOAz<%3BKj&`>YjrRPRqgmD%CRqG{JYj27?Q$C
zSlrSKndatdlp$6&`TH49Bv)cg$&jF2`*?nXM?TH*N@|Hz;ic$sAYq722^~Bb+}Auw
zP3sA<=IDah9On3oQN#Nl?!8-C!^m6yJfD*tN#GPVo=uhKT9{$hy`4OjMGpE|M!^gX
zcj4Od4w)gCl_D*11BCp?aKox)va3G~lN>WsN`*9E_8LG5t@)2p5vH&Ng>R1Aug7bW
zlm-!gkFtxl@Jm-FKqW7N?i|tewe$;wVQjO7cV9kbQ_DG9ia@OjYS6C`Xi%%WOOfS_
z8%DGjq2`pI_rCwZjD=)A@aXJ0OKqz2Tr=*^g!<Iud0<)Ii_}gCq{(@!zMV?hHKngJ
zguL~gAQo6NMk>agBhWGbxQsBfJiR&<o8`xX$B1vubWHNT=zE%d$mK<A%CH_|EJvv^
zi9a8=v0=Ph(R)d#+Y$GDosF0l^I2Ji9-ARu+Ub5P*rcgB&KZ4!QS8GA=>fhZeYTeQ
zCZ+9psUR{v>GxXo7$7J-rSolzYPvVKQcC=8eS$Ix_AHu*d<%e+N$kd%TT4;0uSdLr
zbsW@3q*Hon1Y^q+{)+c)yg?JE(vlAIQX@C^wWW~F(C@QdU76M;#JI$3E+!OPlG^-h
zX$|=}vh(P+=PR=Blj8tC<T8lR5}9Uezd$oc9^Ya+1%Jnxl<HN^Ls{&kFEdXakV5ej
z<)D2}?XT5(c`>tw-t_zB27c_-)5@|Xj%z7TBE9CN_LlyKRIQG%G`M1PXA3DNOv9{#
zc{ZK@;_Lfi#`Rgc{p9A&Zjh9}el}&+^(aWEP#Oall3y3xLG?hksL%U~KMugmr|m_f
zJ}oxl-AE<zte#_cGln02Y@ZPZe3rXCNMT~d&o;$()h;gBNWse&O$rH5IKn%N4yC&;
zT>^vT#4balFY?Wjj>h=jou1Gd%gQ;N`Pef*j862M`3gDY)rXXoEm7zyw8U=uNuo(E
zxlJ6T@jo0KYUWK<rm?<$@r{rOH`-5xvne&Dc|j)2Az+Ad;ZQ)R%!h%H>ZHv;b#lUg
z*lFgb8e&1#o8-$S`Y6%e0)}lK(dK)`=|+;9dySM}r>8C$*I@><(IE(d?=(vny1>%o
zQvt3jyWkVJLJ4ZA5A@F`B^c64DVn)Zj$8d9WS1GC`d;vnJT)Ija`9>(A($4YUHcrj
zQ7=A;Z7$*-y4f>F2xvVR@V?iuoQqtbL;~nR($SU5A27D>n8^yr8Cser-QC0sWQ*$N
z{l-JZ1%3%se_9#Nr%t;a@0|G0H984>-?#p1@3#kU44qB(r!R+NhS90|<tY*OO*gdG
z#1m%i$iNic$=IFtbnMP5AKaOPmes}9B6m$e)nx4Err-%Y7Y(B-J&+E{!+KFd)8GI$
zZUMR|!B#?R9KK@3cxu_{dxA?p&bVr1a(S(*M<+;OmLEqH?w>F4gOpyhTUt$(6POH|
zu(P=IN8024P*1}HG;Hq2{h@kp%8wOkv11ImMBFNnWcjjw=RwqO@58x8UY(B>`2Zzg
z5f|l!a^6rd6KDcmtfo$4I_7HFZ#au)D9yJm2(_5ZFBkz5MN4#h!?=AT7w^UHR^sU+
z%=c8C$<-2OeSE%W6Y}NQ_K>+{5vP14Pp$t4PC&80=1?iG^|=R=xXC%cQRl*_1>rzc
zME%HEKq_5!^TwE~byD4@6Fb`SKrrV82I9-~r>xxW?bv#y&*#QPDM-gJ`{%HMaPQOV
zXo7-4e5uJomhBb&a*&HpwaLMXgq#=Q4jdzDr-c?97(wgj{GV7P*=eP5D*yVO!y{}z
zJb$%f0;J8FTx_-`+akEGN^kHJN4n$r=^binpU!0B%VU?eNG5q6s1SMlBPXLY`!^St
zmEnuQR_}=^8T#S^t8d8W{+4~};g@oVLov7*Aq@J4#m81SMMUCr*QraQRiSnP&UcNs
zjd}Cbop78jYE@+K7ha_e&O<M_)261b^mdL87E`Che<mHx)_kz%hZTtu1-qmi;T^hC
z4QC6^4t-u+qI6NJgHJv(BKHD_d#KFSyCm)H{X8c&%<_W{h*JPQtvTsPbk-ElXFo<!
zxL;Zw0f<m5N;#c))_UP;aPvt7I=s9a;hFa)4&O~w9-``mHrI-h=#brpN3&sPZTRg1
zp=t}!>Oktfzdp)0Zm>5?ULPBUHo(vNu#GiH*f}rce(JB*IIz$qe36r{kxvjW2@v+&
z>32AoqTvyn3$unnu>~@(Q)GOPrStjyjfvu^=;WhhVPXNh8n`;soJ{xqc>g<!TnBh;
zvFKTH$?3d3?mnjv3CcMgcPII3&%*D_WK+-P(xYZEUj8)}vkt6_)BDbkNv84d{kq;V
zbH+a0_;YGZSG%^>#<()DceBR61omp{`K(`S%Ti_;_;y=NR~E#XrHHuwl~IfnztjS9
z_TuP;e5`GM&c<f3MHE6lE^O~332|Y*q&Yun$I+UtXZ>8Mu#?WvV<5Wf;->s^OIe+V
zgGq-<L(_Yjg_qaF37NNQfTO^kUWhpWHKb?fK`HZC_jVO8zINY{Q~ad+&I>D@(Et6A
zq2pK24kKZo1;k;T_cIIGzyT&|m%W;NgrKTWYtVkf7k`%0@R*v&V!B;j5#o<E<T%_v
zsVH75S`Sqj7J&eDK)?!%MM)EXo=@_<%boMDrz$A#(fu96J=E8+d$Yeh5J^qYinY<<
zwi#*vEZsQ$t%0G^)z-X6lp6N4J&&GK8jvimv`CJg9BKMgS=yhNB-hL6<nV9=&s=>&
z$9w1xi3~oT%wS|YL}x?#^f?!mupZggb=89xm)B>iRa_cTDunu70xkj^Ly{*$6znD{
z9q=dHDKP?2;kCE?t@$tF%6K@RGQ?qaN;VPQvr&^|=xc#&S<_Xw<P7O9pX;X}ZH4in
z-{0mq7%)l#d#`@FVaTgpbvZ28KM9G{(0>*Y^1y@W?M=3~L?13f;C3egWTr%^zB6sj
zv9T00WWsO*kcHEH0q7|Lsp?uXYQ)u1m1#pejPoTklwqVj7C$8VJdM6TK%l7k@CZPQ
z@%A)lfAb}xuUBhi4(BrX1*xm;{sGqV$)wDD2@yzQ4Ed3qI`a(W{zqV`7>8KFa1R`X
zYC5g$jFv}KJY!%1frVweq|8o3mLbUX6z;{rNwh5Y#f+PB<`Q@8(_Zoq#D!?&0Z{}O
z*4Ny30J(G%&gU+<<it0jW(|nYcPO3gG`gtVzR%{Ytz~7Vgb#OV0oBoqAm^ZD{0e)F
zsZ^*42d<$324LnU`KHfvc^Rg5_GBA%4a1j1D!+luucU24<iGtq@*tasnl*~sjrL>|
zLjF>`q*7Sdj85vNaZfS-{$?(S!t479_~xH;raKbyYF%fWXo08UpiIH8HdRT;fV)S(
z+b#?OgL!Vc%ljK<5mo5r<{WSW3F>m&AX!EjhsprUax#VAPEy}lk^LqQZzP@Qbir)c
zdA|1Q-92|`yIi8L+^_;!#~3`R;g0Yzg#Vyw5+?2(Gn4weNlHzNhVh}XVsqve$B0X3
zyO^67+q4~yh+L+jJzT5Bef@Yh8A{z(G_3PR*$MZNwD4ma=8*$n%CmbikHK=CKF*ek
zmJagE4FJx4<7t!Pr`eO7*0`4DO#<l!S82eHO_AOW8De=xFvnyHVRSuBvUyRSLP5PX
zgS<nVcrZu@iQ+JH@>_J5o%Vmu9MYw~AY|y*HB{B~I4dS7bmrhnZKc96Pn^#m-|dPO
z1{!O>%y&gD#kjnw@TSV4gy;QxLX=br68X^gLt74w$E~YKyqJZr*@GMTdvpa{MCnWn
zw;uZ=C;9OLA2sR`+&BIjee&@8if#9SSrPST3~1cHQ|Qx24hk4e!V*e7?#chIGb2r~
zEC-t|ANP-&TVZ!qYtw!avSgxS`5rrED4EwgJBY>$lZ5f|IWs2*4u^52L42bgpT_Vl
zbhJ<q_V++r#N8hBUF-==Rfof72)UR`qNt<1)%zTWnyei}hKM2~&rPAI9!f+>;b-9_
z?eaGJRl<l5lIk_D76($B&JY}Y`BSV!7YBT(6*T>7^(ZI*D}T3!0OQ}OZE5fc7mngF
zToO$3xVF?qEdE9Q!WwKn<9yI&?+nb27~=UD4QKbsTdFXa2SamrjNCqgCK~ftd>OEb
zBy2<cV;t+9<-|R(xijYq5x-q{yn7pm2AG;J<i!nN9f1nwOsB&@4JPgA6Tjh7MTQ=B
z6eiam?$Ge4kmFj{s^hXT3Ul_lBzcZAuiEiSlEYYSYHc>YhZBKw6@V@07h=VGb20Ea
zK>=Hj47C=hh?QntIW>F1(QisEXG}op2H=Yr)vTj6U=9q#sS0VTeM(QGQGbwCo))&%
zU8(r^RjL2?&!D=Y5CU}R)awS!-E=Nb+W~z!rpNN({UC))JREuVj~=5&;H}}chq8<y
zBpyxkv9)2>Gl+?Dfp?S}4sr`W{G<_TZjo!4VUE(n;|Bey0$>+M4xl?2qQM7;H4hE)
zmQBhpRd0sT*B_W?)kTFQ_8zO{Ls&`GC5$xd`Q}@PvH+OZtqqqNBjl*otzuBm7p9rJ
zIT|v}tY|$ya9e#&*f%Odp|5A&)U$8=SrI{#%4~KoX*OG>s#tgag0S^+uXl)M5GO%R
z2d-Of1So7uat$cAMI^=qOF26m>}xYijI#dehpo?D604KJZriWfgWP1(M$^tK%_Ls#
z)G$JK7s|=;R{}cdv9cM^=j&@nH`Cu}Ul+KZif69+)=8QJ95mj}kDGs2b`XVT3*7w_
zC|U;Z;q66hBOPFTzICH-pQKdzb1@Mw*g5oZCLv8_b3A}Y8%7u5aadXL(rLdN@Y5Vi
zUkhc%sR7T{aTNNsCwZ1*P5pZ^J|wHYaos?nMYA#cFOC2w<WeaSo@c2hTL?_dj*=qt
zO;sdUu_c*ye{ai@r0PX{0`6}nEJ;YiC7Nq2^F4nRT_|&v@lOOFPMktQ-n|0O6j!e?
z4ken>7>T>qgM1#-@`L1q&?))@=Ddh7ZRiDZTYc3UZVT%X?svF;jmO@2V&Cd0{03q5
zwol}9%RlcBc7PDYHr><pXz>aN%5iC#g$uF9+y-`2349-umn%;*%4zVpwq2eGjtIEs
zxA~>Q9;T4eDg-&|3*25sZEj8E5hFfBydoE;#+R=Kc!x<H>uy2_KO8hpDvK+A$nhI5
z8WHW1sxUg2Ht45lH1kYy7JXLgBHnxc`OyWlFqpc}i<XcKj@-KDQFZD7#udOVU7_~7
z9;x>vflu*#zLmg-S`oH=R+p>}<KvFSzb@UWCY^|+0S>yGg7s}wC(+0!_p;)d!xnNq
z$8>w!j%ICLG7(G=5cj}c$S6|4d5v*4{zdmK3HzJ*gZ3%!y(u~AUaE?ey8C?gcm3m(
z_~&KmELd;^c_4yry+L^3FQjF9L1gTcrKg}ea=3!u8mt@{7nN3opkri|dCnaLc}Zfh
zKvN7h7JrO*qU(1|67b(=5O81X8jlKj-vgUwo_!VU9|o!{Cae`Ch#9PAu_O*P@mA&U
zenX}mT!9kd37h5KN$1-IHdmB!6HyjZN&Kzci8b;*%a|eZA13fd)r~P^bd>h3Cxtim
zqm()bj*W37^Jt`sI7q^$hb`#W^~&?%Ap-Gog|F46YZ{WgwgAr?R!8;;esUKGABj`i
zH!mPP7>~sxvUhCXli;0Qa(MSkUq7~j9OhQf->A#t)gX2xilUwFM7OHWR72UMrGY*|
zVteYt#~rYE12<O|NhO@&y(o48pLrp?Wtn+aAzFo+1^LTLC?Tw?D9=_VjOORMw;gJe
z63u=nO)uZ-jD}di7O=TO%<dz!Gn!K2v(syCg0wxhok-3^gta>x=)O%5VKBi&rauOG
zs3e3kk*v*`+>BMAXjfTr@?8rS4zaUi@qFPmwQm-$maa`U7J6ww>e(i4mB5#%kTTVc
z*QY<?G5b63kT_lT=%VAFJ7)tJ(Mal?i3@$9FXsDBpEOiwaVq&pNn`{%+w(@scDCu^
z_IgAhR2l!UNiQcR8#ihZ4B`f)v10i4v<Iw`nrEJ))rI`1&dAz^aM9ejLlbMyAG$1O
zv*>o(`%E;V>*bt6WO5mWCDRcHZ;x`aXTPuX9`SLu=dA6&Yg(Ks@h0G0pKKMxYOrd$
z(eA>9tO0eb-a-+de6jzmO?fkBQWS{7n#ADdUqj!SYHNy<Ju%`C6o-at+ysiqmW2|S
zIN!UkUti;hXYDGemn_$e0kD@?o8yhqj1~GhY~kWoX-T#aLFHOl`W<Iye=1Ia<z?|f
zFB-$fFWVgryZ9#3gZ^e70^_=0?QliYD)EtQq~7D^ebbd0AJ<vgls{?p&Pz`xuvk>_
z-1YZRLJKohQHu3m6-;>62&8q=3{Q*lzWOg4W3)L9S82`TCx-c!PCYn0O;odDA6w$7
zMZ!0ZNO>sZ{+@0=4$ioNu2bfq%=HBAsGt4p<c=kQP3IEkulmc_#6;G%Ee>75SY$sl
zh2THv-96t+nW*TXCx&+vl(@l&txaNaH_^2m^tG*fyB}hl$gDED;2^y<g<*OLCC~I&
zP?W!o_$OCZ!+$GSsY2SHd52X)Te6S4N5)sXA{140LCAyH!<412v5Hw`IV>6pec<=U
zUeratlGmM2zNHHh%l+)!k$IsNEnKHMp~y*vX);E_9`zd?<uL!avXeW_d>64q1genh
zjOe4D`7iL3pa!0d5}xK|5Kgu(L47~xn~FaR;a=@)K|Q3vo|!Kod^CY@Sw}p{$oc3$
z8p6e1^eG73cj|L(x`V5X4holsj-p(fTKnUV8iN6|g1IW-WD(T}nVpzzW4kdr0b_%8
z=8G98;YqeJ%O*Gjj&*FD=)3hW-49K?Ol&Ea>!G|&(nBgiMjPtI;+&igThDhtVq~PR
z9aVkG(G6y)nJj3TQWiadb#If*3ixjI)_M0?XW+6lA^2Ok!BO54G^3RZO%KW%@zsT@
zw!!c?(qA)dVMG%?fmYbCZTv_;=f$Bz7cdV#U`!6;Yi<yji7Qpwjz1Q{n6WnCLVt-C
z0Gtxm$&TKZHFVvj-7A`lZbFTkQMRejo<KA_HkKKzyx!zI65u!4jG<_qk^L$|&Rb3a
za8&ON*kYvYC6%qOXhvkz#ZzKPC~{-Bor94l{s~x&rs0ibXIMN4HQXRkQ0mq9ebdJH
zr9Z*ox^?FI2{+YFTAn3e^olPkyxky`q6HzI(s#i|S-bY;0}ge1B-Cw|2lWo5uB69C
z;J1Jqf0#h`WJg&;2x_!R22-0(jS+sg?b0?I7%FpIp;0Ke#x})ej<V`KLaQ-eK6RB}
zc8>buc)@4w_e?;iJ0QaL#uIlaIxEmQ>7ih9e+iC@`&K-RgdIl%Hg_7vso#01!JH>C
zVb`)@WGs`|@F1*S#}u>pZ75p<5hg<l8H(S>n|~(mxU=xq-l|_4O5m_{a^Llj1w6RE
zSVZiO>(p~wD%28Zk72)DB82WIg^pux(f#abDr9YD6y#d_=z7S-2Gk0Ye#wpcCZ-gO
zAgM*bPY5hmG`XVSO-hh)3@twJY@mllM7d^)z~ngc1d1Aba1I-`m%GBOvRlpM3x5ug
z-;pr=3Y%tJ+zRU6XH+77gyOp<$f;JAUuMtIFid{o2vPFPGj3#gC(!|0L~Z+&X73n&
z_{X5l+-_;;lvyKZIu4QzTMeiEWL?~3X((AaSYL8c+W<yQl5a$9Tp+<)Nq$ZP>oVhc
zpk{tG+T{<=QY+2t|9!IXKL&DX!<ldE73A-|6g&`!*ePjCLiXo`Vc$wRb4mNEPowey
z4|q0XZ0R6Rf2({g#>#Lgb5hGf(~SS%s%#c~zp9|!2xsGt`&<e{?OT%yl!Fy9a#G(L
zp($2F7*4B3x$Y)<ojh25rHJNRMQWPY^#DP(fp<y3OOLx+3fKjuDBAOMG0%o7-|xEp
z?DZy=%$n3y!|8PI2?B)$G1pm!=VOKwWUivq+S!*_A9Yw<mD<a(W_z<a(Q}TxNpZJV
z_S<{>9**ygd~Z&TG9*@hOk7H3|9ktg9+qs=!+#;m>Y26xI<g#)Bt+@1os+U<&mb<@
zKrPHeY5+i~p~Sge+ZDKPy2<PE=Jo}e<PLP$>j3g8k?gwnPIb6rXOD2DV-t%#=i`s=
z9lA8L{N1SrevR$vAB0GML3&tsh$=qIy8NO>AbnER9gRUNd+1Raj6tzK70OaJF2?X~
zItQaq(0M#P^2L1G=2sJ1wk=RPH><$c6L5@-{@<6R;!H!%vAnMk6Np!d*A}-L0}{_L
zvve|v%lJd05oNBr%vfu9NBF4Zf35NwKevk*FL#MVTAa!n{Z&d%6^xy?u_pSJ7WuX@
zc{ls-CbO4@9(--HDtgEzK4kxjeYb=ssl6Ebs}6+@T9T#fvyFp)0*0BIVQ|%a>xCDd
zeO>iC30C8U^h>UakaDHteklFM`XU<2X@!!u#vrp%v0NV-5En6fRU<`=G_ISN;nuHf
z|AlxIi-EWkQ=XYns$IRGYF!r6rOLS}s9elWcbSfZMVyc=o3rC_fG@!pEr*CjIcfFE
zUg0PBcM)+i*1+h`Z`kx0Y!waKD!n$R0<x6R_M8{fn=c+=+`dIC-uTN7dEWpaf}FR|
zwo<kLj-e@+L`Tb!)@Vef>1~*hZ|z|%33gs&vZ|x?kJRBPUDU`T-kYTTryI!7JVUNS
zWt|z)T)%JJ%Rk6a_<tAAu*$o1h-n<Ss=?$!fa7^Pm*_cvKYKcGwusbvkW~L*Xl5A&
zTqqN)fau^*YGdeB;~V^{GXXwb?Y$0HTAry;gk^6b3CO#iWAi&?@S_jCY!uSd41{vt
zrN*-ZZZ*xBZaquzA-V*4E{pv3)Eq=Q!UU6$sY>P9lRqb{qI+%!bSD<IB1)7&D7^uU
zB2_`1cgbYhZ{^~2L&y$c=QR@f)w#+suy+g)1Un4)PIo3yTxah3l+J}BHv>L)z8i7W
zHh3!!Xfx7ieMqI^FQf$Np*<lHVw>D`k)?(CJ#$A&=E>Dpos3^e^Ipm%fiPP>Z#186
zM-hITWho<#^GaJuCkBDiq@OS-I{SFZV!(b`rGqOe!<x`<&*@F^SW>XvFAO6(0_@7h
zQ+UJp)IWPxme<AZ`)SH|(fn9VF#lXMEqPBMmBp>qSVm6oH!0>>i2w&`tWwRM(oefs
zQfg_{%Z%)N^ga!Xa3`C8Mdm(`T+KSwKuacjZDYnL#Fgl)FC(;AGxNzr69r=bDfSUY
zg`rEAJ$Xev&~i_P=nXDP#zi}mVhKyQw>nW=BM&Emt3lRd?-j5FzDKT)@>g%4e+yl;
zsT7Kdo7fAQ5M&59qf*_RpFcdom9S-yJqrhr`G)|`884bY`Racy3@2lX^6CF*GONIb
zO){tMmVdmSpC5OeQOslB^+;*3;<R3DP8-ce=AJuD-C?~X8{KNPp)xr3z8;R6^&#ge
zXD245a@cg7doQa`oO$ft5Xo?@Kg+_%&(`RO{FWc_lk=Q<mIj?<OucHO??WH!f!|EY
z>VIETDr6wkP2{>F3k3n9u>+RR$z43un1cc9@vwwJIj}R~m$Ao_A;%x1mFHUBK7~%@
z!vpW1#OO~y;Q2~93dEM~=2`omK~;fC0<3!DtB!+<CSE)U9uE=uv`5%A1-tgIx>9;!
zHd9g=$0DZ~w^ncCYP0|q4&#ZUIxz*5dVXatlJjU-7m3L>4>*>=KkJ1s6{a)B`4SSA
zQz8-aJMe9~z0O1>0{4G^@>o7fr)uk*g<LcTApx;Q1;zCxuP>By4L@mZ*ynE0O>b4}
ztt#PI@r{G78~qpFX-v8lMm9>GM`EoDzi(|2Gt%d6EO?8WY#pDME2F#b(mXI;-XWqP
z7h&E7n|+>XGEPLw$dq<kNPDdoydJr4KLxe6V%!sJC2*_11oD#>X-KrN=@+&_JU|tE
zcAwcI+FY|0qB}dZDQM#Ezo6xribuk(e}V;+6eZ!JAgbxt_OXglYBWz6;P!Fv^G_a9
z{}T4spC(sHrvC=8?$YJpP|nmHACM7sz+xFaC^2%(XBQu0UN^|)COToyVG)I#O9-5z
zKbv*;LMwryCC1qTYAGekVTdpo#3a8QL+0NfPk?Orl;^y0xP95>T>Baa3HilN$r%--
zbE}jGpvEw3m|>+b*rc~QwB!Bmq4b;IA3$Twsq<ND3mnd3Z5P4Hu`7>v+QpaMrfcew
zSk+`7c9+~<3N=BpV5se5m|53#2(O;|Vm3gtJ{*35;;|{sMHV0yujyOYiVXz2QnGzP
zpXOb_qeltdMuve{6)p7AR`>e;_5AW<JfOw*iIHI(T{B~jPu>apOOjKRY6!R)79B)o
zUFv@y)dZthb0cG7>ps14A+ienBS|YF63i$jGr~@XGjoEAcJ%9XeAG`*9@mScbyoe*
zKlRiQ*=|I-b<X)W^0cK{F1-9Ss_Su*;j$6b;&D$S$mO${6e!;=3%tHMq^+UP)Hw9m
z-vRI1OhP0N^3TWMa+N{h*fCv8NQSwLhqDDrd&_4Ds^xfFK+rhhn}=-P4a;>Fb;VK4
zAA*D8HZHs&G$yE+lAts6JYDpKB}f|Fa|=k7^_|Yc7Oro85lmMs3|JPgQWmTo@VK>$
zK^fF?9@(7$tW);hv+G-RoVl#RFu8|6Vv)3gm?u_kpF++OgsG1Hv9_uUdkJ6cD%m-E
zZ_8nd+b%!scx5A;>QC}&!yeSO@wBk*#vSzO6HqU|@N0UwT&<)74&luq5b+logI;3Z
zfy@;RQzLRe5vVlhjGv5PC7(z*b(#-bT$N|`aSw7v7`~JeMMA~$;Sux7LgX3CTmd18
z@TUVg6ydB{m8<+C^7l%!ZNBQ<zc+rd^eAxQ*J`&rI1u2jiiWQn(~2*J{}==QuAHxs
z<oq~oZ<*&fU(S$PcdJHs5gcvXf`Bk57QmelJeq^<jP$$en~Z@eX-pgiWa*0S6uy<!
zD5JqGJ|j0=9Q?D5Hnxg~jhBA;$qdqPO*-NuV|Wrm;~#ViimaDEwcpy2QiQE%CNLid
z?{^jNx`#i{4@@fHTXeMbOJ#-0VXZ{m;{+9+Di)aot&>dnQMdS@Uh-z{yDg;!1j!C>
z<DUZF-l!5?!IlrXm{h~n`1QZj&TDRvf<EA>t<Lsbl~Wt(bpox2_O-89ii9{E;C+_&
z4IlgxZ%Rl}e9%{E>@1x#W0(o=qR0ZZhnP7uX$N4g;hV8MIVL8l{rE6!Oi$O#C9+Xc
z`RTh|NpqG1M<T#Cqsm#bG*N>UVy^u8h5E6aUAmeaRKs4oZx*$V*>f=yY-2W4@y#X{
zk5;S-XJbFeeb!W>87LN_6m?buyK`VtbF2ps=j0@HFuxD~%%I_AuC;#hjLm_h06pI#
z?&Vo*Z!T(m11O)^pQkBsX5CvF*|iY~$k|ZEdC8R9kTkwoz~vGPjAT7QhkP$-Bt7NB
z0>2kg>bFm4fyfS-e@%PT63irW>rCu3;V`QGYJ<e(@{IqzAqRm8u6+fMZlrzsXMq@c
zomoq<!F(u~<G!(C4g6CBM|XhtrU%-(C;*m79^pJ{ZLR<9gHN`*NHG287zh)GPcY+?
z<_a@JbaTFi;;uu51%)_0_H*jU3HHFwahPI~lR$MAe!eJaED4r{kZGDGXKdYPd40o^
z(uEJ-H}fzX;j&zI!_<9DGD}?z>!5Wc_VZLgaeS4uQF<jOZut+3TrL-XCAF_EYfCg(
zhRZ>J^32x5#q;&Mj&|h5Y9(F*kGT>T)EN~2OTR{OU(WsWfZsgF^jj{3q;lcH(+`jq
zU!YVi%QRh>g_L`UmB6Ev-qC2zq}~HY09EB3L>L>nTXkcO=0V<TO>H)N9lGLb?2^@0
zNx4j^f!Z^P{s~gd6vB3m`g;_h>!Cd-HHTLPxv1Z$?%wWjLr+aW)s@Cs%fyzkJa$H^
z@qo$l93$`(-wsWmj*SlvW$a#g`^WsNS<PMp<rE8B4l9-Y7PZk7dB8AGTJ{<BnT)ZH
z6;|%}h(eb#9lrAGF_{e6ch6+b=Bvm@F|yQBm$4aoa)vlrh@wWa6u)e5HuP24QGjko
zvKF@=#`CYUu;Y1Zv*gkox$fQ<x$k4>0mbE`!*71M2wuO`#5`0!a=_&^=hYfNp`vF~
zRuX7x);9xPl<hd#mEN~pfzOl8+D(Z!?zlIcQfOxF_tVJI5`ZV;&a`aF-MHsAf78KM
zira^6k{L4wUzK~qMN64La%z1_<w7lgi9XjN=vv%lZTiC%M1o3hwQZ2*y~D_YyEtb-
z!#bK9-EJ&(2>j<%Z1~Og2~znxr0JMzBM)$LAS5ggN=Kx!Fo2<ozewT~*6Q2E5nw6h
z?u?v=R>tTKwh8sw+kIdc(I&AP@-Egq)EP%I93xCPO|UI@yGhhD<UrerN!_8Jnbu45
zo+izm6$<l0;E+|ug~O@6sQd*59ub7@?k5X%Zn)K@VgCbfsy<`6_-E_&ZE4;o&wL(U
zGXVt&I0<0Lxi+E$A>#xYTt*f?HVsat1~!tDKz=Lw!y`du7|hY!>){t##{8|sUMM@O
zu$dMPQeWivnqDY%m2{+qvo6Lt9@Gv+o}Tgkn+Y6o&f|LEDRM)I$K4+`o7S<p-d_z>
zVQ6M?lQnvg0sK+*LYnkLq5a4-E7|FnPh{^o`ylP&@43uhWi&;6JO78ObJ`L`(Y0;a
zwr$(CZQHhO+qP}avTfV8tL{GcZGA$X(t6TIa-|L461#8yx?+fehDF`oJC$+P)XI&q
z=cnq?D)Z+c-~>6%r*}QlhK4u6CWTmWkfK8h{jyqqSUHFKlan@^Nta#rs%*$8KgaxR
zd*3zDSXnc2PaQ%tPqX}gyA}+OVmJ`W8EiP5+71e!I5O4}5-A+$Ng++)L%>1O7~<#x
zl_(VNL+?-0)qL=rbaK_5l%a3>`s4U(-PbdFdxu_HtU-esOHYXa%di2K(9`f+7DF4m
zRShbe=5ovq@F`_BKv)djyeLdd$lOb7B?tIKAm&M-@Q&|(iUnYN4=xd6!ql*^L+okk
z;?P}|e2dGm!p9&Nn#$C(H-w{G1K05i%%l4oW9rZiVVa8#`m)^51|9ciS;Es*G$PHM
z=tLGoESIKtUo{b0Li`76lr|!}AFs&2JID;$gKaIJE335KhwI3b-f)*5<gZcSOxFX&
zxHcd1-(>D0IjXL4dNV@L>f2!!z@Bz`3I`_7nAY;si~KvR#-)Q~UBdC(V=;QkxqZ}w
zuh5G@S`7BbLNjY7QhNWh+@AHDy5*a^b3lOrRRV)VH3r!7lpD3#R+VO2<aTFGkcTuQ
zKN1ACAU-DjRtU*V;q(`|pOGP`dghw_&g;}^_Nv7ifSZ%8tr`k4&PnHO6w+g9&RHeX
zSwZW9UjF<%mLQ5Fn8+e`qppI)Y!@PRsu?wtQj*=zp5&N6KLD8FT9<&6EXgDyAJl$A
zdVox<D|0Aym*}QU@>0@}N||k#iZQ8hnt8IF(0|f;y_DtdJZo8mIJkwdt||qg3d&)@
zLIP_ScgP2wyN}Bs=sQMW=DpoFK<!xS_~Q=sm*%SWAvW-BdGHAfo1}sDxyd+w^6#sb
z>4fA}5d=bDXS!KBfD2nb0X0W(6w$AMeeB1<(0dmQ=j;?Uz0?jgIi{IHueXXcX!eSr
z0j5rDpDv!~n~{h8sqD;9bfE&|Z9%k9B-N9jnZapvGuf;UH!<xev%$5S-Q2Ak4|1Il
zjGtjK9xtx@td0wS*swqf<v+TfgIPPSB5IK4K)~oljz#di0zsQyt3Ar><?y7hk{M#d
zBWeaLmI!)7%ZK;GFs8)qNjmRlC!r>7`;!DKNe+6pZmnga-jSSq%{(|l>MoG2o~SED
zhcS7Pmi*^5Us5$L5lj&H6(CkC<EynosSQ1axjg0CHiLorua8EA1!!PaKTOuG)`%rM
zKPG!WkHiK3U+w788*3wDdGn>!d~`9O|1=5RHgv@e--px!AybZR8H(FUA~1`?5so<C
z0S2UB?Ov`xU=;dVM<ph~liylCz$$``{u$^!KIYvGe#gEqzylC2SaVyGu<@>-6KIx<
zM!^rl?v+|xJ~DP}qVGg#0wZp}Bk=*0bq^u%Kn@yIFwu+7g#~vRInS_&f+yx6Nrp~*
zAH!23lNS2glH3<J3)RD?3*a!clmThQOO0b7r*}G+t_83stWlXhaaXv=+P#<`%4t7W
zjHLk*`nunCRvVStk5p$?LZt#+AKGaB_09=A^Ir`)Jxp%t@pFiYY{Y<}f_c5oJz$4b
zlo#?$myGIowJ0|P>F|L<GB@pJ?$BW1c;JaaUSxwKZpiv=C1YBy-bt?V7vTd5s0kPB
z=C%k@_>e_;dQ%F-;~31FC-_v0#BrCp_Nl7uq#%T|{i+@Cdpy$xJPYj}QCtw&FsSKZ
zlDiJj%2$YD5}z|^-*Ow&DUhPK0@yffjIG7wyve)JTR3E?j6L%7Fr$?uVQg-e3+oTA
zXY6TgDVDmy3fr-r*3>_5ms$~mPtG+xBaG-kIYA#<(_L2f;t*A?;jO&d4Yi<)t>%Zp
zmH<;l-eL5)4V851Xz0e7S)iUg$Wk)|27!BfMa|X$T$>P?<9XIYRgqWA8Kvz^e|L@d
zm2q}Py()MguR^Sm3cp#>(YB*1TQ0dGM^QnM^3V%SWvVX2qj20jD!E%o#&yIJsq2OQ
z%vZv;ho3)?g1>d7<8RB^8TZw}8FU=*XJ*BYiSEe)I5HfRq2^G&qIdJeH~gy}(Szs^
z?G?f|m7=_^$*Tv^D+?o(_^`3o1v{UpmcAAC;=auYRSB9T4&w8vnHAV(SHqn~R7`vQ
zHkY%F-n_+7Vl!_eF*Ki7#YmjpsGRej<gj$PzgJX8E~j)Ox7$NZ(|Es8Cf(o?aSRdM
z0>4j)ww;&}0)%z5pYY1g4M4P1hkaU`&)cyAVuDG1p4mA6&AeG``e={4WO)$@oY`H`
zAH5sS+>apdO$s=?>BOm}v;2f@Xq8LxcRHCfN#7=N{h{>UGg-HUv}hOoF3E<Q;MiEq
z_f?{PGtf9}*070bs{_AK?#JjdtBjMZ(`d5Jy-lhTVCpTPx#4iAm&UU{>Pg;OKj}6t
zcoc-v{>m6twXsXn+YP>cMXyJH=^rssgHnhKni+q*h*cuJw=zpq`sUI&RZ;<NgIA4u
zo^Z+=FsB6iCqmb%nh7pS4hMt_?mV2BwiCmCQnwuv{^=Wm;LzR;6%NBa3JUS(2*1VP
za-qp)ToZ<;=WXV2n0K7{s)yfuAFpAmsREKoDUsY)4ZY*;LjJ)~uH*S&OmBw$001)l
zAe~7fNa_vD>f@)QaQ}U!;i}X9&$~$6NyX8G`TgW`n6V0wIxz)QJ_Z=1G37-pbf{}8
z(*vC+!)r8c8OoX)FQH@Tm)#?j{C`w+ZsB(2`0Chy<y_uYF)3%pT53A{QJxmvKa9B}
z9KB@Q#gO{u=%ZKiY!geC`=GnC6^YGBg8Nqzvm!ESeMcmOC_*z(_O`Q545;qtyFglk
z;E5ZCTxU#02Z+v>T%ez$Gx(?;x@AQ#<QNVT)f*$!OnEtA3<=MIp)T~pzQ7y;ybd?W
zS#$I2^j$&qj|67%m07(a4YG6IMR${%E%`k&PkeF2Y@J|8lK}a$U}gHR4*LgOiS9XM
z?apnzXQFIni;iE5mBu@c9*cB|Z+SPdc<5c!@vv6tu$w4XfUeaqTUBdzzOA54MNFJc
z`n%#FcT^7QjdClC!$+35yyQg;juSTUipWinP!ei&+wT6e`H?1q=)D%@rN2O(of6p}
z&CRO-YVm%n%m`~fW-`Ho_b#{bdvh@l0|1<yXM5-l+N>Zyp88ECF_{ZTb8y3ytOTb~
zfcip2)*et-*uYh%z`)*ePr^R%2Sfnxg0)EK3DNr!TyM77;Z3}H)+;^D+&M0%8qPd8
zy>sd76iIk?zzc;UQ}r+Ei4L!3bt}(9zPaSeno2+($1yggjqfK~+lvyazwsTfxZ8QY
z?`8=Ge>B$_I^jrRnuUkmKr0pV3NgVXqEBtfL`07eDltO`wa@vb&y+}qW507_tTx(A
z=(|m~r=)u~ojlcZ&s<>YVAyVgN|s+og!Qi^>K<_J0&iCC-QZ|AJOOV$O8i?ysn(_Y
zA;?)2G%#j4WI_6OBbH|fge-5<$9T$kP?bLx$vo?Ei6BA{T9<Hmc2d`~NV1B%>~(~G
z>)Oq_*KCo^Hu<qPWTyhLPW}tZ$3&<WcQIvJZ2C3G#tZ6HPWS(iBvt%t%bkPGs&m)^
zK@er}y*a!#MAHv%<<o%Q9NM61UU7748gIWVOQzF<GA(XW4rWQ~x5j(5+{6JIdg=ua
z&%_U(YoOXSDKO$Ilk;gIQ!ocF<3+0UBuj-H6BIt0U0FQBd3`Hb2L>X^KRI>zqcjAi
z{xoH5@*D&vgnH5d(7XZM4oaTh7*SAFhR4zZzovD|Nc9SFPACFyy5%D0x!=sKR2LXM
zs~|V^tHla6nO9Gi@)Uq_K2JfjqwHCm^;v*%G?a|67fU1uP5=5Ihk6XR`?-VQ(^=@s
z)=LSszV1uJ{R?I6C_^Mk9{G5wA7z~+>CzT!(XXcgyMw%G(VoB4c!=<wy+60XO6EA7
z6Ky@$@yh%0{H>x2nxBwvN#=on!3X%nhy1!$iij>}?$`71pw&kAI(zMoTxA|**dW$j
zv$oQ`@<NC2z|J_=>@3;QKdAU%=ETT5Xl+c=dlylUQBw+f?CoJ$xQV0{@3Ct24dxnB
z0P83}n7CuYFL-}5F2oL1yCH1(d%V~7YmLv{c1BlHbIE6q=1y9b^gstMoPkux28qeO
z_OL$>==6FG@jN1}Pz~7LI<Dj5{Sc+5XB(>>Mbeo!(cqm7;IeD;^A&DJaK@@>gIe!L
zLN@v(7{J+s(O!!pI^_nUUpp#6%CO!OkvPS1fIS4Obn!&XlVkAQ0hf&lE}B)IqCDdm
z^q-%hWlZHYZcOS|X%L7Qy$><sMrJqxl89`;5X-%jgd&Rbq21XdS-E&dzl2BQ5~Q}&
z;?93<^o+O$l7xGf$(*M)C1;jG^dMSK%8!6nvR@}vzA^dBac(~nJ`B~t2yyUZtd%uC
zR8_2`-6H22h@h{;(}WHS3+xR&STCUEd|_gWwPT}#2;U9*dQ&YuWBs4f0n#9l!4}h|
z5`40tbn&wH%%GBi^rB=8i}ElBEh+6KVU3q;%&x=&y2bSTwP@=3@AHj76=@fzTy!ka
zf36i1Vu`GqouY>*QqNspaI{`Z79TGA8zOEYhMCdH&_S-XybQdgmU}mktxg>B1J}OZ
z%Y2za?W_skGtlVW^rgHRWLG<`+XLt9Jqh>n`AdrU>+@4`WLQI49PTm6o&gX(228FR
zaPVatQbK6*C#z%Ws>vY*C58Wuilksq*jwxEvd2@-oqRaADO`%O#cO4xvymnoApsD*
z%Sbf1qK%j@JSu!OsfT;pLAECDfHXK8(P1J;$K+6|hCL^f2hXWTT%VdHMIju%Lr;X^
zj0q<@7}BCXw!_Gh*irhPqs9)X0(3MGbnxF6OzI(^3Q=%~>R6-kRED+_@;A+tch~BG
zJH`~<8xvy%t2<CwCDD<q3()6vML4KH{n>IA`M);5LK{Zt%^)|%Qx1zUUC=d$_zkjR
zW&y?zA|NlgJlNQ-wFLc({~7hE7`Gkb5Cfm8rZ!*<!7CtJ#pk}ff*diKGKQT#BH3Eu
z!CDRRkf5ZDwr2GjylO^5$BysP4!$CO^L8^CUeS2ajSizq{;!8~ouVaNz?ycT)x}#r
z9^7$eV9uH<fnByC*yr23*zqJ-HR&QZdEQ=LuT_POM>J?C_s!_OkHj3Lc^?M<&vZxV
zbzo}J;c-I_U6b&!(&zZ}UaOtBu;VCQJs-Bv%TR}t$YQHz7PpZslk-~#>k=-cd61T#
zuahJc71j$%heGkqA<hib9SReAIflrBnYB8Wwq6j5WRUk&JNgPslESYaV~h&^?C$&C
ze*ar5WJ$K9E@Bo%l{<{4Ukh9oLy(ohR(UFXhIc=z<iX{o8w5_m%ITP>%X<`ahJaQw
z!SkB)uDH95_X6E?HhvU1^gu+mMdqd*!$jo4BsWS9RTBpU0^_mlN%_8z^h@ICPfaEA
zD0}qf9Lm@TgVTl2W12!?og5ctCeEBmPc9NAPrB)e6{6sg&;6n#B<dqJU(8ZQQTlzW
zy4?`zOASUcAIR<*olJ%Nf(;TaiMAO!!(N+=ji781lP09DZd0P)R$z5W;=Kc{W5Knw
zXqEg|hg3!H_s_8Pe*Y5ma_6oWf_CQ*D)=c@r^$}7do1WIc&IQ@#nuOS8GP&coNZa&
ziiJIGpRwWFPy;{}u_3}8<2rT48)@;j<zNs^PtyKVS&-;yJz(UrTP#w;6xML#3}X@X
zO|9E6)G4sRe5U8<B>?IGFy`@OIF&?#8ZzY4Gx;2W+q<X_Tyqomh`<KpIOvn}6mHYb
zAhy*o!KVsYgDvbP{g((_D~ndw-T{J5Ps!8Lb@Vaqzs8W=MMjxE7oh#AP}j_*Pi(}L
z%FE}tE~(Uh=W$$+js44;>*FZf`^hXBbQ!|9X2{FH6nZ_dNF&m`YmI&XdgS`EyUQ+%
zPH2NDW~OJ=S*5(=+#YE>O0oO%agLw7Z2qHLec}sjmXXn<wOh`Q<5N2m@)w-3C+wIs
zwnp9CB<3l1HDWHhK=8d~0BBpm8?t#q(od}v2m~iOF$MszQqz=-ZFHeP(<H<5I=JY>
zr6OmZ>*E({SKjfw_gIK2msBIL!hSFWlGGiJ+6)#PSdt%^HU@Et>1Im;TY&*lCf~j<
zPj|+KPn6H+WOL7|rJIl<u2QEBiP<lvEGZ8YOTXwm7K<DnmP&D_C9NyOUIyXy!17T>
zv06A(t!RED+pjGKwpe`0&{(FB?c0XBQBvwAU92k%y>aK{>iq*#A<^dOXGTZ4lqViL
z^W8?xV}lFcH|L<WgMW&3uP<n_YD7gj4x*GA)kdmmQHvVo+p<(Y(c;hR*sjCd{lM7`
zK=d5%LE5AgeM4k6N%*)IlQJ7d?O|lk1;rBo7fa;=o(rb&)w45DigcVsVvu9cHF%gy
zjz86nfm_Y9y5E3xRourHCnQ!kHO3lw%NL^>j>otOLOxY7hQF3udE1mp#SD?h=Y^3d
z=4(>XG@gK(-j?cD(eY~3bds&7SvH8}k-m0YJ3Z1|SHV}#Z1-A<w_7Oi_@dOc_@+Pg
zcRY}?$ySA>+ws^;+PM^4)q@{?EzjhhNs<VSp1kT&%OF?SsR|ZvTfHHqpf6TR`>j*@
z(w^zL6e=*zQJ)*?IQ<X+{0;ggs4ddwWa#FF9|U?TQyqc>mBbj)*`c@a&7?W%kVP*^
z^XX?ZY@Hg&W}w{3Nga(c1f2fKgw3F-PApT(AfMBw6|A2`Q$ts$*DeKM1r*(AdF@}F
z;iJk0zcL|VE1gjW%}ce@#=MlT?NJwbyg+g+rzEH7EZfeGE|Y`ItCkT8)B||zYuy9T
zslC*G(K|Aj4}Vqi)=b7ej7U5gmz&)LH;)|(r&moxqKAT?;9{*O!-x3+0Hx_eoU|$z
z`9b+u#((FF@7-lVzTxS#R(<FF$tCYQ<I(v$`F=wBQjYu8BluHg0D4ZsmLaMNn37tP
zci(M?KcKj3R;r<0v~$FHP_r#3M+5$Qm%NJ{$Q$TlpUy+@OTbL?dPeJ~?Adm343bl;
zb=|OE0il!?_|o-p4+5GVo4xc|^7Q|tqOrN+zqOZd>z6=7m|DxZbqV&i0x)%%NFoq<
z^#_hg!(&l8ZVHC?4qfLwX<ACcncPHXtE$}~^G`};++#~XA}Z;(cy2$ZfC|ZzSg+C`
zlJ)o9)0(FJ$CtW`xUN^FC4bBa88=$aHV~=wgtsHp?MO!wb?@?bFx(5IJY>gQPJzo!
zWysTQX&s-UGkOr4TZ0kLoOYKMicJ$&g`X!;!un<k$G@$QY-*t6jJ&P}*>~;R0v6j;
z^}@Nwzt(mQcS@mDpMPTeWr9RAZxRpyq>W!=)Z=A~rRdQ>hhCeL$1|4j^BFTi`T+uu
z(4h$MVsODV1=ZQ<qkae<s`h?{te!{|OI56*i{X1$0nW!fjU3HKlR-maWd;Mm*j7?T
zLESwypFm-=5OM5h@J;<A<CQ4i<e|#N-wHUeTot$3?A`9;Y$hVy`L$?zsKU#v4EAlK
zGA|2RJ**Qw$hCkkM6(`u2<z*zafn=2hWB&zUgAi3|Gmh-J6!vw`YxbH)+PEG&NPCv
z-?daetX0-;l#WJ}7qd}&L*^FC<Dnw*<HMv>!HPW<Kh)?=ZGQSA<aXz9ntCHbK%w?F
zdq1FoRg@g@A-ujTjvOMf1#*?mHjK!RDiBL&d++z5a;}Fj>pC)@QEBY|dj9kA+vV*t
zHW!JAHX-T5EYGFLnEI*mx4eWh%#OQ>>?T+#&L(D<@<{M?z@BhB#O1P+Q48Io<lR~u
z>FPm&Z@vs|<S<L#EM|JM?%`{$eIgHzDflO!{aQDC$^<MU0wfz=BD&M2vcnyl<dVGY
zWG>1D7Mn8XH6=|Zv0GCVqHAVON6|PkJH}ZkKn2#}m4Mw+DRYB2W3vAP=6U+5!zM^p
z+gja}U9EH#r#=Gyrz%fb$Dj3LN{{@pU{VY|8ZN|R4@t5#v9j*&X+Td1GX|<2Ssqfm
zRq`9w$iz14Nt3ZCL5AV?FRZ%IyU~1lJ-tm3L@a?Z&J1L{_w5p}E3^q^lJw)J?}Ze7
z`@9IGNQ!>Q8j?2AwULAN&Tw9LmtwEToFSNnSdsuG@dWWXhkT+Fvk2*ld(GwOU;MT1
zpJoz|%Fh58I>D5w1!wA&Jad=`rLmYX%eUM=9*8Q!Gd<Dl!9}MPP_y#hqA_Qd&!(g%
z*}&ynOmE2PJ(p%*Yku&{U-}1>5;T7d0cgOaS1I|3G72hSd$WwA3P(|M@42)KG&TML
zPseXJ*guml*-a|8aOG}Cm7^S>mVpt9wsPZqKF?DfW|CaI<X8B)$kyEdCfWRmn4V_;
zQ!TlTKp9o+KU?@%agT`>%PJd~@0)3D0AE=P#d(DFa{k#a)UZQ*iTMi`g!j6#xj8UN
zsKdedXh){soK&?}ES!7H4>Lj$4<ZJV2u4@mM2m!Y=s3iIL^8*g>?Pn7mL1+xeBLg8
zUgq`JvcLCK>VOS63h7y&y)qeS4_vncDwW<1_yxRb=8`uoSk2^JQtR?q5CD7ncuD@5
z!^xUE{mB75oGKjJY5q|r!B<U4_%72~yR7*re_#cIpHZP~%Fye(=kTEDObrYWZZB0O
zlt$rZu}v{iq(mY(fd_=E(oKYl(>)hwI#!plwDb_a<WuNa0<ROsgMJLj-1l*$FR9dK
z{S!%rzxN0_z*4Xz`?30O5*{#57QG)e21nDnJ+$*ev7n*pGD@Upx!*0d%K|F%B=n-T
z$%-nSKP^{0{?ty?5G8DM)nHE;66Lh(;4&OX9skd0KLj6Ao>%b^Mx1hDjMyk*Q{fKj
zTSv7j!@n>?=-*_+sqz<=ZqGiVL}1mt5jQQQRvgo1QBME{){u99vhEfwEqo{Dtq9?8
zPS#6<iqODE+?IR!b&~=H9G1(OhlxWTJvhIJ*Bhe6EU-0-O+@+!?Fv1dl|4;U$@*p{
z_F1W+6&=Y$FwgMcB_L!ESbi*=5?Z#UfnUp?h+qotA6drHy?$jjoTmoeCLCOoctCzA
zFbW}zxq5719MZUbQ`a=t0tIj1G1vnna9A~KM|+GS6vSZycj}p8jvvC^(S7AbbE`F(
zCp-T6EiU1+k%AKvmWzT_Y8o%ZpUtN8d%)9mj{Iyp1njF4wanvgHw!ggU@@Q!FdPhg
z==_GJSc1=$AkHgW7P`u#o5hJyk!kIFZB4zF4{3$TnqptC^FBsPq2MQ39>+?tj$P<R
zc9w}VqIRf}tS{FyaC(n{UMqB9r`eMiO___99`%%+1N=C*oIgm?Tqy!MAT1iJ)ds92
zj@1vj`R5T}shZJi?|8sd{*6Lia*@2u^Uv6jTlJ7SIJ5h7D1pouRmyIq*vgBpcduB+
zpwN`){)-E_F3Io2#Tgq(nX5%Ho~{@wxP46j5fDB^Epqs!#lt<hl3L_*IYdjKEC}9I
zmX3_AW(U)%38O^|@eU1hvKtjA?W$+waTRk9^G&|g7~3C*hpHL>Pl0(%n!Q2<ACb(C
zR$WL~{tj<;_s41%LxWu0M&`4C#vw)gF_SToQ1&x+s*9+#t}S>A4l+q2GKEG6`z#Wn
z{I7iA1(|qDl37Mn56kQb{Ld^k)Mz!0uMPvAXYH#5*c8G}@l<z2#=kDk`&`GXf6r~d
zYKz7i4XrG~b>5ZsYNLiT4{+8cddvD7F_t@0crnIu239+^kd8}Yj&N|GSX1lCkJj~j
zh?c5i7C|^K_@pl4c+DIvjD4y!y@I#lWo*KhSD^u32B}<V)*M^KBm1JSa(*c4eAje7
zxW!Dm6c7sK!F^qTJ@6=`e1x~O2?eWkO8jy-xWkY$c9IX<2;vaT10kPswrF+jb}ElY
z9h*(<f{FAUvyhusQc#D^0iJ6c$4w81-Mq!p143QwuzTz&J5mg^t$}|tq>1{ArBcEh
zOv~<?XLiK82eMl*b*<IMRU%06+{7=dgF;p1+?+24<PNn4<C6--Bk!23lWjyIT-awN
z0h*P}g(+DA|7+6DGH{Ugc-kPqrTrI46yjsw!Rs@sEKVt6@1_VpaNWQp4P=W`?-=$k
zIOwdk)2{{fLRUe5*S<yMP`-7u+T?EgzkK+oU)F9DSA`CuLB-)Vc5XXO#r9f6jxL7%
z<{%Z=K_74C>aIKH?~8e-qXA&+pX14%XikN|8=?e49;U>+p9nd<62gM*=K0>k?$gx6
zb}6Y62p|ZlyFECEYvKE~1Mk|a)id;dR=U%=hS4Uol;?w{|89{}j>ui70E}W0^O}|y
z4xw<ynUi~F_`VHzx!R}Wr^R$55umsj9qvejlcC)vd}!)8srrt9U8#KfAqwpSYIe$^
z7KZP^z>&1@9)^6#)4yiS3ODKibw1sppoOIy+0^DgHARk#{VHVKGfOw2M1qujlNF%S
z+&j*dc=bx3>`f|)A+91N>Z_4JQobadf8GQRLV4w%UNLP`0)Y^69$8yZ811z)!JkS&
z$2fF?URWe4I;%<YJ!1;oNF&J{t6n?Izdw1m(5)MCu8x#Gbz~l9Q|H?1d{_<eeoJvp
zS?IICVJZh;2QL{RX|$&>$(o5{D229EiHe}ymtrQ_kybv8&AHw~&!&*=b8*V%G{I@A
z?&@N8M}U=yA42IJRiL{}-#3DtF-%&qCgX5)hnlkR*inTGRH+Ye=h~dsMd5>G2)yUQ
zZoff)yrE)yDVTA5l`DkJfO2AmSy`*hxeoNzVE69h%4-l}+8)IwfC{vo3V6W41DLs*
z$C-e#<-u){EQbJOK%2kXFo)0%svXy6Ns}TCo+V8khQR?#+2C)Ocd}ksE#o^v|JeQ=
z7tJn**O5&ZaumsOy4>q<5k{6jiH*f*iIl<_hyS!{hzsni7+K(|j~_2j!u&%e7T7e~
z7n&-kKVL)Rdfk|%^QLBmquc2<?Ey;SGpx;)5BxpaTwy&@D^nQyMh%*#X<tc2Ytr-r
z^jjhUh;`AWglTbvuSpb_E-GS{2bf6Ponm*Cs5$PT<u!HF_5~E8%j7LT#lh8yn!^iG
zfYJb#0I3!~@QC=P!O7zNv|;RTT$W0+02O|>Ln#pHWS9Dg&P8q|Bk}!|b3+_VjB`Uh
z9ilQ6(SwdTnHzX-NBg_No!LG4_~0M3yKIssY$TgxA|XSFe7VTRG6s;2(`G?2Fb?I^
zUo%qws$JTKxrIyJ{TGD{hD%aw*}|&w5L%e}^WsI*(c^>82T~P=@Jtid!+AsU0v;S|
z7s*f|A+^wc5iMZr+b0{ugqj1;Z}Q@VtELiU91S+CoSo+?>*P?XK(|^>ZA2^QT48>q
z?<g`W+4!&O;IWd!;Z@g>1BIwud<7uhbFxwo2f|_BxbC7S%oVkkOlAXtz+jq=60hpD
zSUK=Sv>(Xna&uH`DucS8U?x*X8pjKtqFP^Vx8+Bi0gUf0B`+TAI?Tg&uDL4bDIU2T
z+pU3lFfqX>aqc#^mFL}(x|jgt7J-dJ=1h?U?ZSzZchI201JpuOMq5o)b;fL%o#R~y
zaN27gx#tr~ml;EPlocJ9G`Z{mEQQV8wha1g!RyZNcvF=9bvX`mLsL6kY(1{Yd^{{{
zluDZQM53a;ty}O>v)~xCFQ!0mf1U_eMP;wxX?=N3ln8W5xNPs$#-MN6@8R5#V3Nir
zl@tna1zqlmUCI<=tnlbh=@T4kS6R_7$<T1h4fih~gU3@Lq7bmN{{F0f2LTgt7-@s~
zLN!OvkAbJ9S}JG+csYMY0peUFYRhAp=Zg*!V(T*JUwEERnZ+OXx?3$h7KY-QaIAv@
z;;<x3*S4!JgM7IBqq}~Qnexp3=6Kgh1KR5};<BRZA+&U@K#^R|?)d{|d`o~c5#?K3
z=T&Pix8?m4r|H`%7=@=+d)F){eg<M^<nM_}Emm3&&gJPotnK5Z7L7*`F-!i`$j}&b
z96|~6?vRRh6tx6l*fy<lOd^KPvZP~`xL<p#P19w|(RgE@K-f@RGI<zXju}>4U|m&~
zIcVU$i{*8EK9cAn%FdZUlkrBucmUst6Ew;lfmRq#-%YjxOky_5oVt)E++qc0H43MV
z1?Bi|g@8Px>lU=%mvOq=>76)s$Kg8<u%{?Yo@LDBzuj!yrcXwL`LGZ(_Q$zhBhBA;
z>qO(<)@svH!}uGbp3mE3`B2%QpK}9(14#Z!9=>yLZrd)@N0m;YyCi7ku$`N2F(=k_
zIGvjr8SGmgiVMYo(*;ir&R;ba`lY3smadpZz%>!-x&5(W{)Aj-{5xXjPLf4Dh$nT5
zQQz}<@g|!8YP{@{QJJ_hQVW(R&w-Te%Z7J=d0;!FEN^rIyR2}s4XWE=#$IUkWw~(K
zOG>>K>$pn3auy)u@F7Em9-elA_+x_0tTpSy3fdeAHEw?)`+L@|FebraveD)D9IC2L
z>)39J9k&VST9-41%$&voRD$brz11SPYsUOO2_tVjCpM!TC0y~oaG9xGmJkCQw}_GV
z4lFRzT(iQ10Tg$6qRys5+?HW>_n8AN+E^Q^B7x9xI@gLPm+^{oJ6kc%P5ALn5OuGO
z3sXBFs-SVr%aWAS+1mzd7-~WoZO&GC)H`X?9dgz)HYS@BKB?wxo_*SfpOCO^SNe7K
zr8vWf3Bw+mfTP{5yswGiaPg7)Pa3~rH(KEjusMdU&|EDrdH1s3HA0EbP=P<F0i4(h
zTBHg|AY5v-!*NIXF3f>Q1mFT|DhTy8YSe_I^Id4#y$g$sd@oOPg(RICpk&@TdZQiY
zs=JD)h5&(+QoQ5A?gy3uN)R>GRshH@znR4V`-01Yb!}-?7tM-i+cozWztFIC43kQ=
z()Beyvk`s_tf?>Zt50?Ug`WKl<hVvrN%7Qa6H4UR#S?3iUfWHU*+E0y+u>R`q)Og8
zwg1?H{W7sMD|R#`#J@05`PqnPcwJT-38$n#L6!78cAoQt<)}b|k!>4=yQ!^0;MLn#
z?4ma7+oKbUma(9+$J~Q49wT2EX|Eru;M)clPkDa;uLwlygJ~3m$dY}MDRvIDdjsei
zNUh8Gp5It>6jzf3bY}MTRx0wbe0H=UJ<=-p0iv;!#oL{vnlox=jlAamR?DSO`EF`;
zEG*btE*_L!=O<}eolt6H4-f02xP-kd5|=*PVweO$0PCYMaQsz}^n6usw|$4e`rB+b
zH~eIy)T!lpev>;^ep9GbFn+0D9ti!hb?UuS09Mz>BmORo%P#Kt9^c5cHxUX~oZ$2r
z`ljbX^1|qS_>1;cg2fmYRdpttJXV7TSrK4dPi9Y6XGEn&1%LBS2`Ard@0H!wbr%Y+
z=vpwPL=Il)Q@uBaMwvqO?u>1@s}Ov?oF1CzlFv?oW)`~D&5Mz%wdk88<#Y|TA!i)0
zB9T0ofe?xw2m=$OEfhjUGSH`CQaV>TF$2fi$VPdoxLESG3B@h+S9sQStH~o2gSSRt
z1E^nqY#@CD&~xd2DA@+u_`^0iVdN?IACfi=m1zcwW0=QX?J0^B%!(w*7X*?$CI&pA
zYSpCtF+8d#$n+Z0Eb_Up5w-P2oYvZ?8aR`BPeE4>eJzZUh7Lntr{90cQF!tM8q80d
zMN55wMY{PpeQuV21=5jwdON;kMkYG%g{@rZb$EdLgqReJiu5BPctFsc7`VyR=~-6Z
z7J7ZL;mIc3GP8_(s2nvZ1s5^kY|kUh=ZRif26>CwFBGA5^p>TXEHm)8l3WAtd-|>`
zINMx$QjDwHHd~F2yMwgqw#+n1<&3F=n_PIBjzosTAFgNE2+wj{wSYDopKXB-m&6E(
zEbxw2ZyDe_jx<nPv~Z3-ao?R0*KIn&ThavgdaenfEi0=LLqmBeFp$vkd^#$0_bO($
zjy8nFnIn-l2FQV0Lf4aqZwR|siF^!XsL;B_EKUC6C1WckNDdB)X;Z#9W>`8^EdJUN
zl@at_v-z*#nrKL}2vy{Z3=4Qg%;2&-e*6SM&~j5hQ?c4%gTI>bW4npN%fh>2k%ap}
z406}k4LYPeZi18ok9ZpRF}K;>@q3=`T6gg`13(I>Dyb(rI5f_)eqhVB9S<le23gS&
zT-fBs{Scd!dx^*S4<b(G#%dZc%^dw>)!;K6FOF3Kgl4D}4~XTGe!yeYS!1-UsNeIl
zo$%_Z3F`}veETtKFc3)VlidNdtnKG|agk0Q56cw(jVAPjgkc{A%7KzsV_$8B%7Uo}
z=1bYCXI6CT4c%yA_vWW9<`wu&8}|KcNzLpre1eslxP~8Y251pce58n~<?4Le4?J@9
z;Vdg33UZRFug|hn{Sc`j4RHLiF1bVPYFKu+B*AZ;HSGG5^s6fOT<MRWH4@UDxk5FJ
zJUtP1k1fm#rL0pD=u>Vic>NfLHp_>dP~ETu);;8ptksOX?Rv{-r_%wiEHPSRYul<m
zCl$XbA&Acc&M+c?aV5%~zMCiT`K(6I3BIh$Y+*0O`c1zw8Q|PgJLgtXC5n>{V<fUx
zW9RH!$Pdk^m3KuvMq8(M>Gyp0`JUSpMn_`D$C0u8q#4$xGVzqnM+^uBbV8V$f;}-M
zbtWsWlycic9B{rm=JPmpM8m%qG~fUGaNIhmT*y>KN<d{ZX-}FSyS~qr<)Tdq*!R2n
zsSV3}i2m%Yu9h?EyO{&^WFx%9L6?9$(n4&YB6|dp)3?c9&5EgAq5?Ykb+b<<Xk-=r
zW5KqTi=#>LS|IZuZskhuw5)q1EXmjxN;*q6($$>f|H4E7VgK0^mGkjbQw6Wap?j14
zq@q`#l+~AK0RD+rQfXa^trO~WueSnCh*X`2veYoZl3;fCL1vbdw^6%FpV>br=}zHC
zaMZ7!Lno(t7ezE{OQzyLFYN70Z-{_mDBjxuKS|6kyfZ3_)XDi`L+L>^Jj2O_usJHl
z1;<b`JBH8=ri9ru>0hOGmXZ7-W)&}w#(QyiN>U={+#q;63j16I$S@}SG57!WoAM}n
zP2jo_Q)M+LvC%FU6_4|H6St&rAcogE2}%${9C-VQOW82F*JinVDUT(3d0cnRc*`3O
zYg$WIB!C5b4JbR-0c#80=wxT^B)3uT)Q5uwxDh^eW3J)sDYR_soV;bxal6rG(AMD2
zv@w(Se_85bZK3&ogwx}{55L0C<LSnsl#lGaBQtgGTGP~s<`Vz=$~EtAlMCQ2Lu{TE
zu#`GuN-I!_H>{T;RjMU6(#wZ&vY34mw1TnRkbNlPa7RC43BltU(EKTWI>6-yp-)pA
zAij~4+1->65@`ig6ZfsazgyS;j0~kFRZ}|hju{l+C}BDYShDMcrPf{4&T5ZpZ|?0;
z73bP!y5w!Cq2SSN@lau)na)@PXwCR-G>_25<GB-3WHy`_Xa=n&6%I%&z4(;!1islH
z;)N+wIXZKaGO*VeBJD2Z&+p#+MFXoFmlg}k`5|*uyim3+xGj=$VY3Dd_PXcJW5&sW
zAb{@R&4$nnVd%npd;?<)CwQZzsZU`nYG1g~5GM<;<k|fry`Y9t6G*n2)1o`O1a5s{
zg6JmR#z`&($Sp<GiP8N?st&PCR3}Xp=B&DFYy0ve!*1wPC;smlA(!SsMp2KgcV76L
zBB<M1S*u5=w5N#RAGA*?1x}_kG{wXnzsIedxQGNNd<63QzN07msPjzMmf1y6v>ezX
zH>@M$Zp5{x+ZEE4n(3Y%-oRkII1=D<DZIFEBNMik)$YZrD$2a{2XF6QGFA>p_QzZh
z^*`Tu=@Y=@j=_bx@JPC4XC-d}u+gRY2CSWFn&D;hTjC~=DUm^ZE>9Q*7%sI=kI$r)
z>~YK62tBk-I{Yn!Rk<I@JZV33vi{$5bCG&R@?{^R&;0~Aq#@4ltL?sIfGP^26eox^
zlb|vI(R<^K>mpur0p|M0LmM|;0DaLc*XS8_N5LLU1vE<$d*um*{dx!hEBUGX9eG=c
zWVh8-?{EB8qb@_UQPwm2nti4f*Ji(Ar{J_d{spG8t(Wph{<}@raJ!_M@2uc^zaK86
zC;CMN3ao8?B>>$Q@91Rhe26kvjkKI|;R%#Om9*>4>s4`8;TZaO)d9{wuK31Yb)atb
z&7>l|MH~~=YONqLGf^2&UQL))e@(wI?YyWbE_n=pB^@5UHccQ{gJEYI>DbwzbxpR=
zM8{wHYg@WN1w+5uOD5k_MB?4YHUD_Hc`dD5?~jGF72*u(Mhi}smen!RKsKfhH5m2R
z5t4E$I*sY@6a*!IPM6sRn`Sb(Q6jkg4G~ml1OhdgY|_!duqkL98DH$HgunYb8{7U9
zW~mS8#ISx8Lz2BWjbD)9`s9r%B2ikPc=>VvOw^)Ye~{U0ruJZ`GpQJcVDCpiH;J?u
zNPSPJig;#iPC$sZ9_UT$nOnI#iP`2;W8XsS4bZ;0s|(+l+RuK^Ng<ytC@w(fPAyRQ
zEU2lLx&{VMFEdVr7jt3VR>@TH1_58}_r*Nk5n*RuxX%r4F=C(oYcTtZJkZRhlDRp}
zgtaADHU`DU==#!VP%z#d#Jvuhm$XR!N`MspXxW3&=r*M=X^EgTxe?olHbfgrZQq`A
zs+63}-wV{~>~c8E_*z|rB2eu0rJ&6*wsG%%Wkl%q*i{_X9P02N5o)r|UFU<7O`DTX
z!vCRmnp>{q{2=Lrl7__yL3NW#fG-ZT>yE@6@Wb_=qn0~t>SKlpwpKt1zmzHEb(kBD
z@xRNSTUlQ|gmAZHY%8m3bBKgL1HFm5Y5jSjM0ubEToLG}M5v%$4LlQPI_S`NbPnG3
zcO`j^@=IsNMQ>zY?;f5+hiX~bdc%<j>%>wb8f@nfEZvc{D2DocUgOrf>7gu~OzUmC
zBJ!eRWgk=NF3YJC|4-4!rH-Jw8i8Q@F>+u<-sR^hvdj=#{9|!~O(@&dZwh7_{MTlx
zy4QTET9N@uF)}gUo}q5}!x=foFt*;%0Kd)YM<HgUHwDPdVH0m}B%nuSVBtS4E`U`?
z6p;@mjzZJB)<p)_?*)g(II5<zKLPXYncbpRs?|dOg1Fka8`C{TwlietkwPh-az;%G
zp?MH*(g9wIcUQyG;#y^<<PdaHO@f8>=_87;{tT#rfYAu~%}$8yC)u22A(q%V0SL!t
zW5p2+*KhL_OOY!y(GGr7My2mxIkHEFm>LR3EeJrXozDeNnk|w=?(x;MsM9Hno8EYt
zYP;3FW|;WOfdj<R4mtC3KyM`=8`mlT8$k9z+%o;rp?`j`2P<84ivG=6u6;>s9fVZ&
z05zCNrHSVe_XstP*yy}hsfOE5zCXf*&7pTi`U#Gn5ScMA1HGM^A9#MKNJ8rL1b63a
zfDaMYok&}?kSYmMe4Ud>;Puxd5NxP~IN$*Qj+*G1ZxcF|{%YLF2}fz?&2qWRw)%FR
z`6C(Q4iH;(F>?Ef1w`Y5Gw>4dxsG4X>E<Gx26C=B|Frp+j)P#jnN)154JU20I}K9?
zBezM)xnQMX$25fJjtrgx(YBYtuRol~y*fut60DJlQiSXf-0&{=MPQLcq>Vmq=1A=A
z4H?^Ofj%ECHG-^>m$B_%ELyVQ{)y2#zFIc?TWRP<O3`JLm;;AOg{X)46I(d?hZTG=
zDLjyZB_%x?rMi@(-zRV4vUdv0Fo|_s=t5=pV#j}BfO;cNVzqUO5F{~JN?8mGBLQJK
zK2*W@7~S0C$_h0WgsJS<-E3%7^(8LBE1ymDXPH~JuN8cCgp$f+ZkUxHd5%9<9JybY
zF7f{Ah?rW2=K`KVtW#(VDC5<rgB-ca@G5(Q4-S8T@Qfu9yLd_6Tn~fz$Bv&8UdB@x
z)ejr(!?}!ZSY-R$$0b#R)=8hLlE?>gqq9qDX&M(>R5IV%{{P=Gy2V8|$w^L@Z00N6
z3{|8=)6+6RF|$lo$w(-(0}d1J#A$o-kioq!jFP=k9#o$v9MUf#{<9)jEGr@@Yb>Gj
z4}`~^>MC7i{KD8i`~?QR%Y%wId%UkK<aymEcFXhMHcu?R1Ds&PQd}QIXi7wqJ`*Rw
zI>UMynQsnc{+{bk#${UyBVXp(NJeg3|1t=9YpKL}&Ldq^a~SCcwvgJWWn}@DF|FXg
zzlII0zO3$IrDK?iDET_YmPMtU;3<OhN6_h$?ohgf#Q7PQWqs{CO|@`ECB07^L~c6o
zj{DWMCfM%mfx3uyLKG|*!W?_1Bb@R5s@wp=zBcjDEX4<Nr&w{FH}V=9EU8Yb+5-}H
z58Uu+N$)tW4MF&MC4s4~tJM2z>oU$sOGJj)yX22A<TH9=6%^?IfZB%L-90&`P@zry
zT6|raP{ydhTBtjON+o}QWEX=PbITAow+BXkdKXsT>LK~j2M}%qVua-VEzd(StcC^9
zH*R$oOxiiHOj-1T#8pPN)%686J?)AHwSanjGF>4+Z$ELl;Kt9MXqdaULE$=`6T_=$
zm@r2eT2vgp*k7|uT@^ImK@`lUfSE8MQ0Jzbc7cSaf&*zM_?qZsmdk~82hfy0k>w<5
z8&EK}Na!$UVb9!*;#%D$uIh%Cf?~{gbx!)V?rcOHB}zO6wtCe2p3&E1d9v`!2?0Is
zL-W9a2LJ${$@3-GBs{zK_{eTgoEw%SdKyGVeZ#UJU51qdG*?|$XRiK6gZlv43e9LI
ze}5!zuZ)y7(kJ7#P7{PuY_WQ(=a)i`+F!78=tl)lz8VLzCOw3gr)Wl4Z~Ms_)B@#t
z`+Iq%LXJjfB<9A7^o<)s3S05Ry>qs@z>HUFS~hw2=c~}e=rVZ(JzSs?XjA==5KcKQ
zM$m!+OS4&Q;>9>E&!faE+PX0MzKZ8mG^OHn?Cr=Lohr3Kl+IPo>$_hE0;|D~qML-a
z48l3(3NL&#eOI{0LW}fQFo$cQs&@WYfft?-yOm_oTduJR|GuNnC)y_#U`UgKj#8b=
zD7(zcI)WRG_!+sK08R*ROGXmhXc0+es9l+1>DMUmyv@S-{Sj39JABuABte}zxrz&g
zyJcO=nH8gyU^!eCeEC^aX689G3{n1d<f>gE03#M<76Dz^Yr0C74HFh@d^*X<MaEpN
zQ4yQOU0*;1SuK*ZC{CCpVQoDKpn8dXc_)lgnXdifvlE*ymuTRo-#KTTYkX`{@`LTe
z^c-pY&ZBQlH+f`gcM>_v<zZ8&C7cnGZq2eo%<9SJ)sl*P%VnYeMgGw`7p@rjVZzX!
zHOq?A1S5t^hEVNs`}osKajIvgN$DYn#qx=*NmuhRnE}k|N6(-Uc9hPud9E9Dy2s@q
z&Y2tlSa0BM^MI4zHG8Aqj{}{(^0yYkJhYoXPx3wO%UFm1OVrhhGKu^=XCP*%Cl&af
z`w}ck4kM`OUAUN+&c7r~a+2n7tc0P?!@<xc?(dHMZ^ke<Q&#DDg9t{rP6%h){|ZVW
zl_KOq^|Pk7HD+69faH*q2T+0fH=-y&{>Xq2hoiC)DLXI^K^XJzP{yJtf<)%EU0c(4
zvLJ8*htO!Gtd9tll480FDSL&ow1nB^RWWXZo-U+&q(;j1;GBLVRdlciaVryV<Z+BP
zZ16QnV9Sf^Jx39AM`-wiXPW?{LfGJf-ouX29skvc=$=3%8gUf@+BY$9gX(}zu{&Bg
z?S-9%z2Q~@i>V7I;Ef8?e}5g`>9_(tb^i0O!>Rb)@L}6_dxGNQ1<j&7@!Su8BYm)Q
zmoZQp?_$l*<G>L&?^NYVujp*mtkk=~YP9oHA}dj-Q&-Gc5IzEl?a72wAzvf`%PD<j
z{_fX&dZdBXhK>`5mMS#4YzRD<SwN9ozrp7H=52mc7ZNDRI(y;s2VA!Fvl5bNCb0h<
zslAs??imoRElvb+HR@sE-$Ux_-J`t-C<(m-Vi=zwi5F5gJc@r=O>&s9YlG3WgUu0s
zT!>89;{<1ZLdxpP@4yBcY|7qkcF>0s+0y9SJ@zAWHP#V_^wyb7Ix51Do&?u9PD#|q
zB6JK-WcM17HyN}AGll&4R59OxvX8G!=c1D$ipBe9n#yvH2pWQU4&*{Y`#$d*yg|B$
zGuR60IU<l~LYL_-08;vR$t6ivD6nR?%V5@ZU{1~Rz3rXYI_$3*wu*@U<Io9t*mQTC
zW8s`QtbmV_AzpG76KyZ93_e3{8&9=@*)w43y_3mWu0(vW0D#d&di^fWPj$ijlK|GA
zCunCykhfN4<KQg}Y!P!7snojHwxY2Gt(>;&hf5)|aP1`_NidW{8ujNQHJ+dW{(&tu
z85c-uG%TEg+BmWf6hKC`X989uc(@5iu4zhfk~`fd2i%5PCi$A1#L=^fs<6T8`qs!f
zEyeJz*tf*>Xqq|a)y0jkGZAdvRPZyeWFIWyitB=E_1qb>`W!20*VE~9&k;7WgAzG0
z{=CP%>t6ab1RtN)UAB6zLXS{_Pv7@zi~fe-XOZ_fb|P8FJW%J|_v@&sM_UuEc?Kb3
z3$AdP@PIdIt&ujzVUgqwx4mKRWUsi@<6qoSvPPBWsgctPuEjs_jQp~`*|ZN)sV6hB
zy^}Uxl=cip@rPB2Q(kZGBWgq{Fwj?kx_BeVroucRf=;IOPnml1SCVx(pv7Se?jIn>
zv*0s7`xIB;Y&A?@_Q7ku6mE-R0ZF$JyJOKYtmezxxAE#m17-;=awy!nsGj~0iPTa8
z8NWBEwXAK?64lc-)JDY1D5Xxis)`BXQCxYWPNG5DF2>A~m5)v+ehSTw26=3hKBaTD
zOz!vm1T6pf6yVE`eZ?zP>h?dIzz-y4KUvu_owb>SjegYR)^_ZFO=U-nlF~xxSKrd<
zAivBsNVVUpnW@ge2qK_mU%kdhSv_5TB+k+?HtKP6-R4nh0q`&hX?Scg`U=jb?{KD+
z=9sa%IiNi*qxAkeghj@1h-n{23MXASop@7@G^zZ@H5AT5gZCpfdIUvawFCWt#n=8z
zhxP@!hiaZKm19l{<5vG_)#xH0=qR6(L%flQK8Yu9CcrNba8p7mWsVEyCWOX6*PPdO
zl8;S&8*;clYfYky8$hj<quwNnji|@V7I5llOj!YB?|!3&!t{}HDH7dnf=bp!lsMNL
z7f*p)dcf1>%agi<%nd;Y_o$SNR&KBYSvBm$c}GTCg6IO4hbdzu5Xw96t!&(jK3`Xj
zG^m7OFjWBgz=^8!>Wbz6TqegCGp6JqJN`9@Kj~QZm1e;H`U-cATq#0%vrHz%-)=^s
zjhr-q#uxiw9O1?ob-inDu7D<H>a6l{zcPcltaN%)^Puuy5KQt;U+<E1a{ass+GiTA
zh~Xu^AjuKcuk4k8<tfL?W|{IuJm0(2j%ZN9<~VpQXb0~TEz;t{4|QUnPKjytjvDUA
zdGeEL7>J2MSZFhl8VSTNpx$H3hZSQ$!tFap4F33>i9-BKdb(*AK69Z1^jzDgJNu`T
z_ujKqP4O0sE{sUzGqXh%yDZ4rJ&Qx7hl#{NyIeqtU7OZ6Rw<exEgLq${fP1g%PtLt
z6mb;M=@b6};oXA)+9%EM^vz#1DZE<ja*4R(Wx8mg1P@5;Mj)BKM>uAw;hCP?O*#kn
zZVq2<)E&El56`kUdAfYAm4}6|oE}uOQ+2k55g!dmmS%qRENAiZtOUp5sYG`5{X1|c
zonJgTP4|sU$WquaZcW}JMcz=Nl;EyDgte3vNto$l+&;tju7xqu+XS5WjzVz4SS!Tb
zQ7GaqnY1A%Di-qquQPCazmx9?08(sloYDp!<;vy1o|d)0U267@6U*7m$pO^>*U0&7
zM}Rb1O4l6%);wsSBcXl%KraV$Na=*RU}}Ishd@}r2q8i9?~(Rky_%_Qbasm%Gl!)i
zStfIadYB3jXmLj6p{4h9Kj{|0<NpQS2qO1ymazmwqUS9c5Qp7;swTj6x_@a8NhJY)
zVo_4r!|tm#$rPHW(0!pD3uGfuWiALCF0XLH*_Ta=k2E?2&?JUnr#Cid<<&{7(xvM~
zdm*HZ>de(UccTz-^Q^4i@UdHFlX(ZYe54hgG%hEBQD|5e4tt%Lw8S%`%blAubkJ2D
z*`>yO1J}{wRFZxo5uyyK`T`yvj-grjkj?k)n51i%am+?<Va2L^!BRMz1}-c}sYLsv
z^stOpJsso1nU#Y@9|XYQpOxBkg6lMt=d;`t%ioF8a+me)g~d|tp}vW}ek3K!T{*BQ
zgq%1AFlqK?{s5u>@X`hlYoQcUA*RxFd_~b944@6V-8&%o5cL-3qDfXGcDeK`dPZzh
zOUk4aJ^N=#bv16mlYQNy7oTRixkG!M^5qtsXkMw{+ilfB-@43JW2t)^#|oaSyov)}
zNHqtSCGbI8fW!%J%46|s!<)O|l=ke($d|$`FGznzXt_7z2OW%2Ho)_ZGflj(7-mA+
z1xqc_764Ol%92}CwBc72fc3;gyiMn#G8K|L1{pQWzPxur>sB4iSQA;`F=ScI`_0g`
z2|W^a4#8P}pHZXu4)G*voq7eZp)lHRhlB`uz=4#`e6Z1v=t(b-$=|IIX)|4G;1|=%
zc^bl$Bee^A7pmF@c-B<cGfreekE^i3mjA0vt6CA!Ul;6Ikp+a!sH{*bF`S^f0`L}0
z`8gq`vpyr^Hk>N&mljtn&}FtlRN*xSB4;`bRf=LrsJ?e~JYGP4O-I<l#a(Wiv<s3z
z=7QSn6u22e(%H(Cs4AXufi!C()C!eXAeqxP#1iU<BME_9e;JXXh5P?es@*&Y^$t&|
zAr}Ap`nOZT+pcDu>NnhbUc1aROrt%&{X7BnK4w?V|8YJ>pbxT07*UQt`V_NNbfwa8
zk!qFQ)^_mn6efrVNUF(4@*Kls%n1H`)$Ui)gHgS?xBPrb6JcGA97oPB=+KAesHEb_
zt!fjGZafy!IVFsfz(^Y@nCoSLNfV|WCJq{Wu#jEXTE@))aBK41ev`|pG=A$~P=Hjg
z>h$Lrir5p^oPmm0C@$3NX=vryB?loVqD9F(-5HK1;^F91iNJ0jjL70J#UD9)pGUFU
zGZINJ@HuH_Hz0`$$f$~{`_Ciy91>|u{6h%#7s7jCjwxW+IbK=e0B{yd?pS)1YwJzM
z2Dbnqk_4FuSLvAnJ8!4R`*sx%y=K~07K9=>-lE{KoPb6+sW5kpSaC+So(M{3EsmfK
zB!|jfG{*-;Q3m}7g4a}g%+m4g>cTZzqO#j#{)P%#Vxlu^@l-73T<jsFS>T78D}Ktv
zl#cgEUcH8-z<8~>K-@c-mWFZrNJ+2|=RpuZIR3}#eu&?6ZgH+R-vP__n-L-&(Dt!7
zB!HFo+Ots1xkqQ!krN09h5CA<*Zz9lBJv`Y0$s7}oB=O`VaeKBcPcXU*?MCF>S8Cq
z*E2?UJpdOJ$dEu@yEq5ED446;q2dSkGX0`G@AqIRTq+?3lg||c6G{Xnq6ctk6XZZ4
z_G9Ld2(f@Datdk3Z*12|9KC<KPfcd**X8rVxstuGT#{HPtl9?+F&UG;h2mBe_}`h4
zofhh;|B?ddG4;I&a^8OO#Ox|-;274p^NH*cL+jO=M%f&Or`=+WrPaPGQNT9w<>*tr
zoJAkhJ_i99@!wqha(<z=Psv*dC8zM}t5T1ZLzv7fD2wc7oR3<vBKT7eq&N;#{rhDZ
z$eJp!_A$dJ|3Pnu^7DcNLuR}IBe`G24VH{42dQg;@ef-88WFK5ZQi?JDv1w8U~PYr
zNb^CrD&u=)95V}LQJd>5Y}axFuCum4IJsow99D&aF%1Mz=i31C^^QGFJEVihBL;$E
zr_m0$?Mr+Sxiy|M$DTP15UXc06&e&$MNp`+_D>c_Erh5izT5jxH^i`fW1=TCaPTgc
z!y!}~Z9TE*guu1_`HtD~17p(s2f5$advB`}kMkV~V0c7*8Ye}v^s3J=2ZP};!C^Jf
zA-qju!Wg=ymptz=K)?f~#tT)0`ppNCiTNea|K!$8hG^86J1iL2?Y|X&`Dhcs&s})a
z?#e65&U^?J3i1Kr**ZPhFpj2rywScer<dmj;S%RT!*du@o6FAAB??rINRXWNhO0ZC
zB+@F7HI#A4!MpWzowyt%Ip~1?jZsB@2gX|_*T4Te)1$|oleNU4FST_{n0Fj;sk@=!
zvvN4)S)lDqP&(XcC{S=_=x7bt{(y|qsOH;)KQ^2wNBEwa9DAHE|5&wpU5!h__G#@0
z2!nuRSyo?JQys~?`iS`a?GeOlTIrL=%3;37Pz%l)EKem_x>J{))Vdx*QV4_T7V>M)
z<aS#w(Q{c3QO%;IWE*Kl2aJekf*TPklPpbay#K&e+ViPsh{-cFTxS}I93gb4bTilR
zk12+6(kR@=GwtQ9<sXp&iW&<Pzy)HLL}eg?GZlBhHLpp<^@bM-n9z=bcdf*4|L94~
zPh4~J4bdiE!Sn-%CHV#wZPUWV>)#(4_q>9_?hHGPZ5y0@nyI#lAvq5i0(`-wp!hdx
z#gOe{t$?U^Aw-9Q8;piBU9aY7jwFm3S7!%;Y^ka@^Qx0*jpVP}Gb>_47LrUodXaqI
zZ1XhD=~+6!-u4Udq8v!gw?5Yq0EtZ{sBo~c;=pLG=j}V5<<IW81J#hxG>B+Q99TS`
zK<UO2)p6M&Wh!%LhgO%WEb9#2TDjq}&T&E{`lsVQP&pZI+cSn%ug|!@xi^?#_b$fI
zaDTAK7ftl9inUcv(}32moKq7`=Z>h*^+ft(9C~>wP9}McBC;w+p=$C*8nf?)i1DXX
zOwilhbuYM*;B|Cs(vDVjYCFhPD1xN?iW#J+AQAJVqc#Ettf8oIDfFc}brA&Lxfx=s
zEU0kVe)BCREgaco6xod|Is59;!SmBtdhxzbxvj>ZR{~}m_XILm{?wK8%LG=K+);J|
z)MEqO;7)JDOiG(-3T19&b98cLVQmU!Ze(v_Y6>(lGBzMEAa7!73OqatFHB`_XLM*W
zAU88P3NK7$ZfA68G9WcDFg7^~FHB`_XLM*YATSCqOl59obZ8(oFf=zaARr(hAPO%=
zX>4?5av(28Y+-a|L}g=dWMv9IJ_>Vma%Ev{3V7Olw_|Xo-PSD}8y!2jW81cEtHV3C
z)v;~cw(X9sj_sslo@eh}?>XPM>(r@Lb=B0EW6iPFk9Cm~E2+{8o7(?0ld!jQp=V}b
z;sL1IyE>VeiQ1c*DLL8GtC(51+88?l{z<`c15AM?E`XnA7C<{V#(yCcjBU*TjI!zy
z;$j-K|97CNnfZTerR~h^0bKuSVe0DeKW*I1oScF7b^xk>7^neGX2z!Wb~c{>fhpL#
z08PvQRR8G=06GJV0WMC)re?OrPSyZ>bHKm14p8-Ub}_Sc21whP*gHAcI~lu}nF4@z
z02fO$fVv&<U!JPVKf!<CcBYK>P5}FVGEM*!dsjObC!m=#!+-ON+B<kU0WB<D0960S
zWMN`vqWjmE<^Ms;01&pZ0jT^7;0#bPb2f8w`zQF{{UwZTfi|B17Vm#%`<HBhe{}or
zQlPU0(8J7B3Fu;C`OoaGX8#FGyBOO5O@!?%Y|H>m{|TwvnVLD-0PW0_?45xw|B{rR
znd!fCYL-9~YdbS%X8`+uQf79h|HUZ&&p4()JBxp5W@l>bWct4Y|4u6z1MOVYJRSZw
z@4wFW-#+vI@#T&G5!M5s%f!IM#Qaa;KkI)i{r{p<b#bz{Hq!)}x>){yJBo<ddjP!Y
znYmd3^la>G0M365W+o0M-~T_{#MQ~k%+BTCm+GHU{>L{5{-cYTnTMGP-1>^W32(4f
zdRB%@Jyzc6l$cgKF{C)F`$+e=Rk-6BkVY%$r@;y(i45FdOZs?tk=zHI5(+o3H(8&p
z&J3}GpR(~G#BkJg`zCOl6`t&XYGRaJL8WfES4+pE@&uwbcgruKTj-Ieb7&1FBl^Y|
zutqwJXXS3pSpvy^K1(ABxW2o)u)CcA_A-D_e41>MRtXGP_to$=|79POR_<7dU$M<E
za8R}(NkCOEmudy{k5jPhA+LhN2#OD=S0iQ*?7k@xOm*DYzFAHjIKwM-!)PA99z@f-
zx5{IvWX=ZU&Vc8l64!=%AqCmmI<!kN)*R(@mOzSOq!zk<`y5aMLfDpVm86thb0V?O
z3kOx!L;mW<c$kQ&(b(T<po5K#b>&mfwlF8{Y`Ig~d6qHiuT(CzSfsnuZp#2d><1=&
zG}*jgOT9%lr;{er1~H<FSR0uGwTvq)1k#bhf|-of>h%mj={C^;0RA(2G2F}Hb$tL5
zjkF!}XDg7%tQw{#_cqDxJz{-)CYy;E5`~uhi>YmvTX(5UQ2B2>7M~?OP+f%iTjMU8
zhL^1Iwt1`Q1o$FNBNV{Pz<T9hQo7ieUn$NVVfdz9=4IUlTC9*caC85yju^^b@RUjr
ztntksGTb5J_gGneE-kS#8Kh>7L5C^T%E7^Ch%)O&#MX-}4I(>Mo(pQk$voxSa$bi#
zJYV*#Qs&c7SS(_Y3%)XCCwAjhPr5bJf8((JKt&O*nN*?gTc?AWcngjgs~?tDoJn>^
z_oA-ZjtuJ;iqfo)P#P;Mbdny_-R&r=be#OxpUa@E;_thKdMf&2PnzcH1<gn|WUHQZ
z>G3O1TYYpUU3n9$Jr3!NWE7pD<#3LE%1Q}BX;1|xBlnBm7xHmcfhPfR-s;^rOl&sI
z!)KWeHTd{WjP839$0J9u;emUo8NlJZhbR8gn{p!FjP(J1bP66{WGi#QHTkwt3t9Ut
zFg?&n)K<>=uudya@5l|HGz7m&&*CoLj{A-SzEU#gM8fIDUMxzdbbza)N7eP)m>&?$
z^l{(Z=a#zC+@#Bids4wUwIR*I_MWt-j`Xsy1*1bhqa+uBo<_rFMyc(c<ieI9j*DX&
ztpdSkL0*r3NtB|>tZd2eZ{gE%JEoKk(-{=zQ0uje)Te9jSIoqOot@RptaW67E?+Z@
z7ljsuU~`3Z9PScu^sHX9+nHbBtMB=u5cgC9SO_BMR%n<6{e@AP|Gl8>J#95zmigr}
z&)BDCts$)jZnMAL{1{-Fq6_>%uv0CX?b#r2?9gk&gt{3Q@A0Q*&#x)%H-N_lO6I^(
z;5HPja>+HEu7kRyD1q9i(#x1E*G-R@1|VcX#eAZe>62q)7h0La2eY1fJjp6e!o_Sw
zUo%m268v-i8@yZh!Rhsn!RuVJGwb7q`eAXkD4A;$j1kS*oyiK6{!Q||mB;c)3((!t
zU9xG3fs8#;??Vcw&U2_G3jX>~Yur~OMKzSDh+E0LMQwXHHt&lVrQ%aV*M-aA&LBWI
z^lBd!D>8Ms>c(K{-bqTUtU<gtByGQ9`sJrq5i$xMK8X>3rtzK{g#|wNPljR`%_4l2
z@{Du_NB$c!<qumbb>h=GzlUawRAsFUCQ1y9kDtXHN|%}Ul7%Va7SS~^4cUa8r%S_?
zeoJ#J;HzuFz)1F7Quq)Z?dYkSevl6dRoXY#p!c^4%3Hz|#J0hz?XR-sn*}s#kkamF
zsHoG&N58y#c?o1JFwK(%6pLwF2~Thau`B4>er{DJ$b$@pv%8V&fXWY7sr7yxbBJvb
zhDuKND*<{lIm))698<|;ASxWake|;HUyF)eD3K|Q@DKLuo-mB>p;K!&#TZI(MmKJ9
z9fbG{hx@*PMTI6w@7WXujbyaMbzb{+ndWfLwFzk-kA3jQu3#QpsV9f{1|8mIka%4*
zMGx;M)eI0{79Fg#Q{;k-p;Za-3Pw|WfjG!-{ct&?AB&a@Bf%;b%$Dode&-;&3BQzF
zmNlVguf=RW=eAdRE!5D*8o}n3M&J%f9%{TcD;1n|zB~*-i{2v(n=203I7JkJ5~Q?2
zlz)lZkmsxnKlps5Sb2q)NkZLiXiXPcne!<*LAQ3F=dL0V0DEumMs8MQw`2LrEnQ!Z
zD;-oan{DnSuTm%HR(sV>+yT@{M@K-*AGgwpvFSIUze^SDL^}ppwplM`hw@0E%%B0^
zK@6|FVi=zvmi`Ka5nT5)<a_>E{?Xmr0Y3RAbq&Q3?+lpPVPc@kSWNd{Dqq;fX@g}g
z*pLxybE29=0|!n_Af9Y}1)cC<C88TtNd~S41CVOqq`~$88Oy5x$v;?#5(Hh|YMcQ(
zzro?D1`fBeK}nAJx<bsxFWWiIl0e7RG$16i-Xxk!6s(m3n?vt{6O@<U3QfW2mfTcP
z^P2w(Kh@<FEH<;xYGWI`_g?`j-&FtX{(1=|)|fNfwuo6~uL2i2k-5<hdo6pPX0m+%
z7Q({St|!%=4(;B5X9CM9DfHH%$=8_XI}-Z4(hCF?xY)yBq(6%bS69ge@#YrLS#Koi
z5iNz@)GI!S=5sZq_D4-LZK<~OoL1W@H#i#{ojj@CotmUHTur6Ws*{>V60>Go04Ib_
z#Gb|30><1E`0u;Wz^01lmtRT7i%rK}u<Pn78`E2^qCq34W^foR?TQwV<RnBRanlFo
z*0SYh`vJ<wQc<73bWl|D;G0h6IjV5?RF8)6vF9Ot_7Z+}*Yxc8G8J=1pSMM2DR(UP
zn-|L*=k62YBR>xkQH&d96Job7aNjI#6T6EE1oZzgB&v^Po}_QRFURe?3u;NDJ!lY~
zaH2fG8l2BAwU4bG8Klj(_f%!qcmF*whaWV)a^hG5!3}eV2FTA@C0cC#k`s?5kip%_
zst@L04-0@3H1<@Dhoe-yJ4k+fCXUtX@Ub*Go5Zi6*Fj-iTsw4_G)j0XSilF}a74tv
zJ2Vcq|6|{1(wOq~PKNy<Fz~fO9rAHPUN_pS)aS=$X6>a!&}=DNMY{#1WuP<^H6To!
zqZK;~G7#h<;HQgTEOz6lBfvJ;rn}bMCe&9<>&RS%jL+vW7bxY>*BUJoV{YnxzDnr^
z&@RIRDK6c+zifap8gsS+WxT|d|2Y*pya%!+TuuQ*tX5^3TUlW*7t6C1hCeBlzWn+%
zl|vWT6It#9l7;h<E9kTO*&ejhQ0A)6{EC#eMX<cZ36P1!IZr$JcuL}&5XW?9p|{c-
zmd0y(_OITk|1Nh>cJqD9o34$Bjc<FlG#Fx#PqCD;Sc1@D|GJ^OD?3zIyQ>0*)>cRy
zAX}LAHbayeQbhS2JG}t!&^!zCdl4&T)D9&cA&;qN9-bI1@)kvidLAX6jB4l>{6Jlu
zPBf%eQ&P9cdIb1%X*~qK2c3Ke%D>M-OlAekyGV(wxn|$j*3AHuSXSExUR=UWQ8*YP
zYc@o=I1gz{Nh>DO@Ah!c@73qVyXmhfVRvg2T%a0oN^!Izka&bfm#urzKt<E+Q$)Ph
zfR|jv>8a?I!D|0DR~>q(M(NfO>95gTLW*kW;|!Bv?1LP*zvSxytkQeNjZkwp)rH+s
zYI${%j>j1U0&-3F;56Zd*ba`8XB_&_Ys=mJ=nU4$kY(m%ue`jwvhR;>97zi@i8&IC
znm!nHnDWxq%srtJhx6mI-izfxgES<#RPMwhe2$9JMAF8KX@PnRoRHc2od?j&{DgAr
zxt-`qNzY?@O4O#iY=aq|xkZ*_1J^ifs)@9U^D7!`xUXjKu=2ay&ft&JHYhtTDnEOl
zfLYfwYj?bS9(|i_yc_TVA3fUW?oIO<Y+R7WWvGMcN->&m#e)x`*t)8u`Hg_bACH&C
zPL)Sn{)-7tIW>L;*1X<9Ep}+XeRQmS9GN4#gW3;LHUjk9**bzDb4cl&Ck-0W)Bi5}
zxhrxgQMr+9)~eqMH_v^OSsmgw*iGWH$pLZ$W7n*NbM=1AJ%YZqkT;7asvs8?0M!nM
z8OOl`f1C3i-C+xww{gnVXk-#Y?8J(0U^x2$cdmu&QYlWb=6?T%XiI{C4A%`?G=;Bp
zEG$!GA2|(5AeOcNnS}9&*Qok~ajL}nk9Ta+Vx$fsczjl$-zZ}kfjkLxSRX^-{phw+
zcRi#sK45>OQaVMO(Ks|q==VI`36)uMB%G|JW&qHV8r_x<HKoquZVWE)L86sh=Q4rE
zOQ{_tS@=QRe^pG<Nl}!OA31rih?1?+%9yp><{L)|osPMt9>(xc>jljqCq3M1Z$98R
zd}noO>Z4RV@u!t5(BCLXb9zmfhEZZ((iIwcd<ca$asOV%DEB5JzXf(byr-iNVcKyB
z<u}P$9~)l)Z>d6Gi%4>lS8z7;>IHE5NEoDxEoKmmjAyA?=}C1=XmJ~VLuVi{PCL`|
zFI6x2JA4_+Sv8bAq!h6f;I`U?T3wMF)};}9VQhIzQoBx>0NG8$GI;Z&3q(Sa3wZZ}
zZE)L-ykwhNF?u15tgL5@fwvAC!C?7CG1{?5G5J<)4!01j(tzd3{3O|faWbw5o#Ciw
zJTp8Z?0KAZi(1tGNu56wYczt0i1hxa&daiH_oBM(qM5U}8cQMEo;@EYVomsGjN4#%
z9`|pXrS?U`D`0EF`(g<`WFT!F4s28d(H6+G>%5Em2G;<cxur-FhzxK_(|%Vwl`?fR
zWE}FvxmD|^$%dE!*oxj6<)^Yo^H!?ag@_aH`;|DISeY8_aVF9yy8A|HgT>3GBtwv9
zM=*GbOtB0eA^ok|i7H7i56`JVzS%!J+_xqPoe?4TSsJqgtA{yoPA6~Zyk9ydqojw<
z%(Ty^2q*VN_SD;$BMQ#tYBMWlxmSD&iL|CpUPuTe!<N{CRV&F?K4^NP9j-P2U-R9}
z?sM!J^bl{V=v>&@LbPaP@E)^Rjg!Ekz#Zx^iX!1N&UsI=MebqXc2vODCQ3J;D~vp&
zu^|3tyfgRz{fvV{x3xAUr31h27pqV86TvFMg;P%Oh0t#jN_r)d7<>^~(>$<NY0wZ-
zOx8SAO%$jbKxbP(3lk^sBG`9=P@b*)I+!6b>O<4tqZ-=3o}_zAj<N`?U2aLv_9Jqg
z6V0ullT!1@^DCLYWum3SSAZ|t7CP44Y|}Q)JWlV3(f2oIKxPV@n@G`PQ!XRHEcK@a
zJs8@Rd7FeBS68FmmBO5Ly$O6PRD)X)A63nMNiX`VMJ~exxkS$-VF?UO;7zkG^<>ey
zCT&`JG}10(%zkG3w`u8>zMtVT0THpW>AU&~0n)KQVT^WS?(AvoF)=3w+)dfI#z$`q
zluf*5>^EI)B#}~fuJ=VAS>|g@`KV{5J-<DEn3gjdsYf^JGQ!eX?Ro$~^0p|&-(65n
zStWg_^7d<UVc=G?X?6S@J;&*Y6izQ_KbOO4$#M?ISN<L6HJq4E>l;od5w(x*@Z5~?
zaWx@{<ZFG;__>$O(}6cd_0DmKew=Ro#`Sd;XRci|5L;T3A6diNg=cKSYs@C9mR1}q
zi^y{N8}4nKJ-ow?Vv-BOXUpLH&6ZVX3H0th7UlEjCK9>Dpfs-+7#UL7a-BjKA#*X-
zD2AgygGO*CiQb}h)f44NjylotFkj|4Ttw%=&~2I7chJ&e8C(`LNZ61h{@>A-KetpK
z03KEr1}-k@#3su>Fv50AP>>r~3~2pR$|7FA_Y&`Dtn)Up%&uWU<mdheYGiF0N$o_L
zsIf}DY77dd35&l53DW4l-OZTek7(jx&ggp)q5}zlHB=IKiGNGR^cN;nhT_8&Grh{e
zNnzpLf3wIrgXE5x*hv&D;33e-|1>&=l{aFi#^mU_{$xFry79N8ok9=9$Mum?X49Z(
z2SvxlF^A(NSSMra$un`7QT)Ys(oH6^in{ImhZ|=m@AmKcxzE~olk^*AP5naHl_emx
z^^c{Yu3vj|eK)-bC%Wf5wR$LBUWZII={hr<7_Q=0W;VTjGnm3`pdM5hLJHy%CeEWk
zhEei0BYX;sC^==pa?iVR=0<Dq$2HmAhkB(u(rQh*=t+$`+!D7ity150JbT2(-_Uu!
z9syoaAv&j2)k5x_qa)$I*AArnBg=m-eLp^PZd5<yK?`+4Y2(joxn+UlulAaOB?PZv
zsnDOsmm=|Q0`iuMB%ZrsXH=tJ2<+1tgfFi;QpxB<91Ti@k7TSf!NjswtAWfO^;a<k
zt6Lm<BaK3}MQLW6p$6HE<+A!=*@kihOe0Le;)|2HJ_XQ5msrlgrHhKfBo32N>cIUq
zw?zJIk)ODHgzdA=iBa6;VNYVL(7N!a;+1XBdt0vih@fOQGqm9iFvTGDy!GSL`AS33
zMs;G&R-(o3KWE;h$HDLweLF#kij_C1BYBeT)iL{9$%cd|al$7#&WJTHE?r137qjSX
zg(|HWHB${>ZZT*>Ep$q@x-PnTNXEnNcpr+Z=XlE-0>*4qlVg-Cs{4i467pY+YQ52v
zBE}r-G#wQhD1P9a9V;W9&05rm@`zO7^Cj!bLAWVC5o@_&Qu9V@WM|6WLTLXm`33W?
zoIsvh%Lgg8wETd{tak`I_EaaU#W?>B2!d+g;KTR*#V$98O}(ocg)4%lUh7024nA{h
zWfk74RO}^L>NlzG{VlzD8o}764{`&{%_h<MqL7W#(4Q<^kD!Nz4mp0J%~Jd<WBt8$
z>d|(cc$OqbZrP-<zxqzcPOj?^y~E$iEJ<FNWeL+dj+{J+6NORCA4qI7N5ehje_`s~
zL~xMJX;o%pPP9i~s?15BPG@rNV@%hA<~&f-==f`dSRsOw)xlSD1BP}O-j0!MxADjr
zJtoYQ5%j<*-S&dY7k4Dm-k34l#ByTRz^G{>zK4_#Fk*$1a;p3eyM;U_D}ndMN=?kx
zBaYb&lDT$1$N%{iYT5q)M?kp0t$Xi%m|cjPdDhsE*V^$b2v`*%HJ~g}9MZ!A+m~aT
zU(L10M&(y<OB*v%Og!wUOv0ite@JcILkLD9FPd5QD>R5yM@Nxv;GEx7r-tMqd_fHK
zy#_8nJ>2pH4Y}#{3K_~6^(rNqmfhfbW|8+a#!g;1yEC`P3bfiR0scU1eYv#!8veNQ
zZeS=OQYm%3U=S(B={?J!f>q94g}d&l<v7Q@{f-dP!*W)drvhY7_ynGLmx{0!(Z^u%
zBO3kFE@-oMR%$VRwc@@}vLc=IYQCjWCKA795gwX@&ZQJ?e96WQXAFh#M`unDj<D%0
ztHoWn+A2+Ftlse_pPzs45zQZz5Uk<jn6<3JF>^oYPQBj?+~{J#i6MrBhxEZE_2Ee^
z`x1IaKXm7-<aM5A|4^TKWzz>%qAPMDj>jTFJN`U=zYlu1WEsU%(np}kpp}*$)%AA4
z^kClbqG}loy(p&(Cr5$Y{rMdq7eu>1<67N4F8~3H1Is}>Yfr>g;^zO=*DuE5_jz@~
zX~FKpFpELeN1KCOzRSj^z~$Y)_jTvPO7s+#KyNe}V*U{l&s^sq5hH(i&Jo*6e%p`p
zH=ph$t8zV~9FS9O1;hen?1RGfqzpq-hFAvygF>^Pj$qPt$ktyuVK`;qu4`(mjjW@v
z!G(dFa@M^Ry%b)H-6>G^fR=pGV~-yL@nVHzWg+YAV=cM8N5p%_kqWl^hEOD!z2hj5
zEHn&UKS}j6A?=_Ju^0E9*Cl+`8Sm`fS{WXdev24S7_c5}km>V6F~8XQdCF&k2s%Wu
zsgWn%#Q)f%$g5N$7<^3O{x?^#+^9nKI>j!wksh_At8FM^V)+(cgeWaztgTrS!JeQi
zbby{-^Vj$<B~BI=7w5|njI17Bi>C%xiO%WGp2qp6x1~AZ&f9f#pceFWoQSeW4ZXG;
zDfs7%t#waY&k?jer67jd{EvFzYFGw#gtPt0WIwwuPTYEkjC4V%SaO(TGih;YzoK->
z5vBj=#W~5eYfU;$!?zzGnC+F+4T2s`dpUMvJgP~{Ml&R(OybweVccVrkjW9kQ|LYa
zMDK)-<jscOmXc8RpCV8T0z6rc+-j6_0Fm})K|!YGg4+=CQ?=@J7`HMAQQ5VW{!7wX
zd&Rs-yLaz~Y4;&|4D-_5T6LOeI+O5$tA+ER$h2PCi{I`!)?j28F>S4F<SpIqF3M@m
zr)rY>{l@sYq>eBa6`Y|=kV+)A_T!Ylj5W&1rYfeRT7eMN!rjTJsXn?6$g8>mEEa}?
zIL%lCkLQnyn~`PRIH(=Dr&mytlxp$E+8hR4FrNyC1zSW3;)vN@vmwSz)VNy~%vUAu
z=Q09PJ4}Q`3s*c#Z$y1_C&mZ8d5jxEmHLjiHO1kQ4$sT!Jp+0NeVutl8;4|rB2Ru8
zcevbiROY%lwY`f=NWQ#%I)O>wA-c8%$xeUI`v+>)R^G3{Z?x8UC%=C3hFs}i>ODT#
zP5u%2@m`Al;a(K$21Br?J!LV@g28uu8}MdqUyrwVIsM@%hKKx-6Kh-SWP%5lJCPv=
z3Ut@8&C1Y$?to%dgh(H1bzQHn1qMEj=JTpY&r;>NFA(A0V_ZrPI<hO&2l4Jbfvj&z
zu_*SA^*w2hny&EIVQp?u%(193=}vKNXVZ-kzPH&Q)=vq^PRqme!j&KeaZzTooR4BP
zqS9$H76j%O;<yGxZd41k$Z5Tq7nF^mPJ}~*8+>hdL$rq<Xs-{Y#MsJ;Ide1-E=6Hs
z6~W91u~?%c65uh{>h48bK~k#-1^lUf0>UuPMh&cJZ9S#!+}SK9FAOP!A)B#kI^Ud*
zOp+Nhgd)Nu6wXlc3cRUTBZ1%wTbw`t{Au1xfq7oI0l~<E-t;2e%PqPs&Gby$-?Jn-
zt<uLrhTD_+X7-&{SZb&Cx&JdL>3>%9j;8DnL6v(!%-u&EFb9z(@4oCQ#69@R&Kb4T
zfb~i6D|{%pI*xIteFt;Oyjxq2$P%#FIRyf;A`#B7x6<{NQYgXVb}gf-`AL*~bEVAB
z(RzKy0%<Huh?)iNSF35WQeK6HR-}+dKjw$Zxf8$ox$WD&fk+wrR#Et834P8gXJ)mG
zfu1Hghit20OTVH1QJ`XO8}V%tcUx*b-GGWhtK~Vo7sJum;3`@sOjN)UVk#oaR#UUm
zo#+|K_;4SK-nWkB+kE#-gz{T{uqY93DBs?dO5=FWozs%jJ9GnGQ>qQo87Sc_+pG7+
z2yyhW*GORGLNSw(Xp2STtV|lJe*%vQ^S!+5@spZ563@@+WW0RNOg+%oU-s8ZC93Cw
zUSzuKJ5QP8L2#a<?D^yvJ*MW?b({NDIN1)`v|@CE*A@`nfTdNjj<hdEHdg2FZSFnf
z1vTnExb`Cc$S_Xd4o@NS54ylU)r-+Cszk}lo%-rWd;u%l<tu@<db~i&fORXqYa13I
z00(pVx#ovg7Tq_nRI=;rx8zkNevS0td-Y~oe!r0bF&jRUjGz<eN^s{X4`8~pyy1IK
zN_n^8kP|#F=v_PBp9y{fQX9$?>@{PR`~COfKRR>ayiLCl0_Q8KMGMsL)^NPNbq$~2
z82Y1%bTgT54A_W{75-LB5{K?RPIYv)jE2F>R2NyM41w}rn$SRXf8Xo-id}cq(}}p!
zV4dC1TrI7Xt)0W>T#&ldiQUbk!WIkAFK-<h8TTVm$MR{xD;ydCGf9wK=-;o!67r5u
z-@?2i1H+t&av*%cJ~!XQ)ZxI_qdy={Gv$vVYCNT>Ey~NPr_xItNR&L1zsXAvB4&6o
zG8`cf^rs|b96CQXXJf*1AcZ6Pdeqg4ucRQCKRI2gXMZ5K^{wa8*DTir25n`}kK-j<
zR$cOJAaS_8^GNk@@2TrB7L9@x6+PY7oCR%YQjd5H=13M-3MKDGw}RDksGn{L?{+X%
z48)zx^aHaR=vW_Nxk~3=t=lnB-<`4G{hoXcRJ6L_Ry`(pn82%_>=}K3$nhH8AyhH0
z^V|~bgNyt?QIcNp7@h)oZ;(TrEHu6DV9%o!bk?Za2hnKyBf@*LX`fjQFysCGjSys6
zF(-obqN)0-Ao|M@T*tP1_fJa&Of0V-o}GJkBa<MOvB>0;fcsf-aO7)Do&-=%S@vaV
z>?;ykh|DVro=*KDFnp@dR|f9+2z|Eae(>UfHotd1Sn3E2PLYlXWc*Av9`UV#kHZft
zkuRGoBLAq1Qp2cFGM9Y3yx3?}i9up7a~{GIeGkzYoaUoPTXBs_D+OC`@-5Hqi;MB_
zz3b$<haSO{pWj9~&X&5lr<nvS!1!^UOlm~DNxZ=)*S#-H-yv)~uF*<z&2uGAC!$~-
zkK~tv*udJ#^YtznTgh^&={K>3KcQq~{B_of1ofC~h5PT@idLKP0iTgUB-iJ^k+#XR
zB2xkN>eO%C$f6mnUsmm5$IdZgx}dRqRB53NPY68$67HsH-ag6XlRu90T&$7gc<zfj
z{J>G}Q-3VkHh*;3(b7mrt};#|P`GcrR#7pxVrn@uqd(`!x8X}NZbaZt{!QMs+EeCA
zLMAaU*`?Skl?J&kD*4Q5utJ}v#KLxW(oDFEcHg536I~a%NH@2(q%88COHf_zFXK(b
zCbqzq|6^ESkNd?^EL$@(jsp&8WOp+tZ>Qz(v4liW3363vraiV^)z)m}BU!C$;apq4
zEff;!*R5+-_#FyMUS<}H{tKV$8bhpor+JzggYjw>V#H#Q?N<&hcU_6+E0L-!MmZZt
zZ;h5N30JHiM`<i+0M`_{Yd~~~70rwafx<#U#)UJ#cAW2QRa9@``|r-K%);Dmnku|4
z>AQZj(4h_=L2CQ=-FZ>7MbF(6z8XEc`tJeihdwq~rc`y+Cd#$kztOegeDs=iDJfg=
zyB-RgBg67rCsdp2C7E>dG3aCOY!zl_{M(N%x4gXbag1$dAp~K}&;{K~mFW*fuaBY{
z2bNa0EyX{>j2TgDI$H;ohVJv)tlyU;(fW*<c8mLp=hFw(4F4Xyh0Nk5{f%O*m;`U_
zn;G<J@q@UY%aWR(#s>(~cloPx_3VfA3*E<-xJbiF+uv|t4qZKT1?>3TI9BhLvjGjG
zt!D!)NS|`^s|gQ16LgSzd-xt7C_u|5UkLjC%(Xdz3kZ*)2fO)7Y9RA+Lk`Q%Oq&ds
zmmbmRluHs&`Qt;hZr4#aWCwDb<z7Y*Et$rNMtOfV^nSd%eX=ME3?t4*!9NLr`~szP
zLB!)Bb+5Y+$$s{1g8apgtL^?)dHw}Qg;t~-^p{Kb+lQzm`p7a>E=>_;4PYA$_tQ{;
z+Am<d^^TfLq5+*GJ%sLVK`xRka8<B$YdnOpSPqIu_%p3qf@vL{EN`4a_E}di)$S|@
zr=wQJ4J;5;GPM#}cS4Fphq2cyK~CI+W11Y9Bn0w0wv8;oT@@x~`Xv(N@O1ypl=|fT
zNlbu>hw(S^wy?@m!Zi9Cv9(vAI8v9bFhJV=x+Snl20n71nJ-MOU8hZ6RzoV4$mK*=
z#iEV_M$N-?a5`CY9_iS8O#I%f%b&tVO_Sg@{>VeS&l`@59^$rlw|_l<or<8zq*!+-
zhF+BWCVQJVg>B?LTJG3E?@m@)?B1DwUfg}HVw_Y3BOS>^XYg8?l<%^pBM-Z$nXeI3
zrpPo0ZvrZr**~1O=kjU+S84&Ny#FG)t)7Hca?T5O<V#{qw@d(JCr)70$M)yOtVKL1
z;N!wd0w`wU=Q~J41Zvz0>W`(BB{qgQ8nrmA@$3_Fh;8eaiejF^N9XV+lHB^1QM!n<
z9r@BP!RC$vO-4|SKQxvJmiDyrSCkQ;vYlYo@|K=i`_340Wy14cTjv8XV^P*IFNc@2
zg+V7a1xd4I`DRXh98e(|?~G7zYK$O-n^G<XJU1C8Ugc8BO<S{x2;y7v(3Xq3t)7tc
z`>}b}Z&i6!HM|(>{kfduB2uzjkB{vry5Nd(#T7CrDI!@z+Q26#M=hWOvnk4x@`O+r
znh6u0vw`3El~oSEodn~z_yP$#51W@1GV%Gd?x&OiSJG!8iiX6tdX1o1>8C=ied+?5
zNae(WiV%jIxYPqmEa^2vv&(qw!r%|JmMYY*JdTwp72<^=cPa}g(`nddkoA^*&{<l@
z{*<B;(pye<=mC7!yvmcnMk=alo#AL*=GK*TdK$2&L>fXH(4^bRAquavvplmxZTAcQ
za5Q-*J!MCS5z>9lUK4UB=SHz#ZpKP1Ua_2>^a=;BiJ}XzdFCmM6K!1uyAv6P8sN}U
zl({G?l`Cgjxt+Lj*t7~O$31eDZEu2FPI`TGpSd8ia9e&fcOpBc-LYjd_HA7|bi&v~
zONQ<qy?hTDVcm>w#v07lfeki!iW^5p@UyG?TH3&j&Lyng%A=Rl{htQcr*q5Rus3mk
ziL45Y^0rYz%~{(X%@-NbykW(yoQbgf?K7^MJFB5U6!qPs3*s}S9aXkr-@y_g%xgv5
z3eY%QrIFZnO*Rsep{|o<Kj!A%ZydbiIv7ggR?JZ$jn$2q;$%03L8}nYd7=(l1A#%;
zxe8oEQnVF<;*=u5v=vS^0_T@*LyDSsOms^IL>Ct5H^h<ffEbkr7`+_9!Pq==wo^)N
zz`YmcYGQD9-!46$slOE_knrmy?{B7>GW@=&10GBbX;I`2JGTTMb?<ZT2$_0y{C353
zP51X=qv!T?Vr`v4r4FtT_DIA0@e3^5J^72!`0qnCJ=bgh2WNC}hbcejIyBiV=o<RL
z-zqf8tO0t61$&fai(ZsALN?!{Gs7syV*>oIzi%(0b6421e`nJ=LZrJg;4}|$jDc^M
zeB$0N{O;#4Q8coQ5DG8efj(xIOee~Z6bh%eiP1?#O|MX)pMko)%u=>+cuYrl=A=KJ
ze4)g)J_Q7fh1nD?3h^tW216%DoqU&C>O$Lk0C0YIt=MtFa<&%CX%SeTlxHgEI=tnL
zq~#}7I~I?YW@e2lz&%rMW^(z+la>`fitZNt5doCq&mrTrP>HckU{#eB)(h){50|%?
z@ueH=ym?SZfWu$G%dof`dK#%$?0D&RTvg&#w`q8dVd|~xMAG;PJh8<OKCC(a!dPX|
zerH4XO$l*!9{_vx8DIQam6K3bHGO9*SZYB(Iv<0p->1qU`)74aX>2>J829_fkTvLD
zlE4lAeP{{Q?rVOZ8KGb7Pi2=F9z7L?AsvmwHcPyP;^Q#+OavJZ7kJtVYluz8iF^^Z
zE6~wUJ<Zx!f|rsI2QJva%$B^sZZco2pY&L1<r#Gopny-mRsMFrw89|mzq5tK6Y4tz
zsgDa&GC~Z=i1+3Y#(YkQ$BT`;0t)3zb1$ODn13#%!uSYrK#q^6-M<u$cjZXWp)K*b
ziF^(Qa?4o@Uk`6bn61IyoFF+{{Er3q9UoZ!`h8QaSUfX)C+p<$vXX3ImU_`EBF7P=
zn7ljl=_RUPe2Zjam6l*??nY$N=pFC6bI8;#w3RW<$|gV_?TCf9m(?Vq)Ul%-xuM@o
zc6((tLs%I7#G&|U`2KF&pm)kV<E{`+B*zJfq=U2ZM@J=_IA8<T?E?jn?61rIeNz!Y
z5M+3~8~noy;WHjuS^2!zS@q(xp}cRcJx~bc3xBt1tUL$)!tt=(<EjAAbL3opFv-#4
z+d{u&KkwvVwB@S-AyVMF5;@!icA#HpcG_*_MX!y8<n+KK3-3V__`dZdSF4MPj47DN
zLy}~uD`VG`uinT!6-iP&UcH>&Uyu|fL#m#sdDbl;0KVU(*s+pZ2)UIH^JT^l*>5c+
zTF{yS&ol~fyn$j<iwzL%Ixm#*LwUK`1D#S=!sNVMBj{IwCag?_rDM4U#jCL#xH0hK
z7W$$mix3Aj$hq#gAeT4c?KSZi3KQ$`8JXX49J6OcK)U+07=mS@1h-M|w+N?5e}OiZ
zrK*FIj&!A?+a6+J<elX=&Sj*Ij)!M_%wt?K4>W%E`|h2jphh%$r&?WOui=8L(n(s|
z7(g&3TLag2^QLXHk?A^2KvCFSb!jWk1(O_fJya$0<w!H0tXa)E*d60f=6%cvpo5xU
zU4by&ZVwmKj44If7J6f~7qXeahX_Ntfk0ZbQ!XN1O0#(LXE^$7RffX(WsRou<ag4G
zgvTwoiT1(EZTEz7nWkf+nDZ8aMvJX)sD#dq^#~UV{Ft0_QLh{CXEuPFnV^Y2WrNnL
z-5OLV<#Ov{Bc!c&_Zs#S0GqwcEk4@S=vo~<TE%UyEQ>x|<qv^f{N^Y4t|k#IoX08V
zgFhGOhf`PX9z2O{2`!Ig=eHBr=((`9p{*5*TXx{#S#+>dGsa189vE&K7-N5G*$zt^
zoP$Jg$kt&gw(R0BKT;NFGMH`W*{?Ih9Wt)7h9RNP%v1Rz60ii->Pkds4$iscZOQRN
z>{@<{O868_F=umA*d93D*!nVvOCHASHsT^tJ#6b=_4v4>x$gI%g?0Wd2tUc$oBqZ%
zW)q-EQDdl(yOR?r#+E<5*ub4oz}$o;>|8xq_`!*s##Vaq_$1z26=A9mnO4sLt>1wI
zzN<*|lXKu3M&i(J;Zry_7cy&DEf;D=h7WM^BS#oJ@CjS8XevFoYmQ@p04Wm470@%f
z3kN0tQ(Ar}`aZL+iw}(UUE(GQ^AANjKG)RjW@N>5sj~Hu9XlTpy6}2&UhgtBOrasN
zE}KY=Tsv2g+Mm8*PVHwTPGa{SCPPmw<__>cJ++hZPGd#mz?$k`GvOd1aez28{!ZS)
ztu|yG#V;FJMtDqqtvYEG;F9rA;2CsKt<@@j^bYLzvI^GWpSx&kypyI)tkcg1vIe}L
zz679+c_3IKKWRN~5QIBoQ1+RaUyiqNAl|q1bv4tjs+Fs=*ODLQMauJDO*zTuuCXU(
z&$m;l?`N;Q!&QdIZS5Dmi{TUBy#Ou<(T<q6($k&OjxQH~5&ad}S3$E8-w|UWGjlPu
zcie(AE2~}qT3nc8!o?NUo-f`6sX|-H^NKQ61P;MfASFJpu&&-Com+BxVhF}O3(yPC
zvIQ?Z>cn(Cu#K?)bklayE|WDBe3}*7R4q|?%*e#8gUkF=5X5~9Ej@_6Ytodm?o?PG
z`^b64;vD2b$m0<;{5Q|lOoSI0YImD4(5#EQt}aP_(Y!8<QI5=t>#1Ibd)o(-Hb$MK
z(<B&5UJI{DnNlBUtHNSej^5L1f1UY?f0Pu3N=OyYe}2~AOu!(5<=wF;0Tn!m>uRlo
zd(k#$`~;uc^p}KD!+5vtjff9>ivMH*vM18-jCyD+pdU-Y>Om8D^qQJ!x`1e+7q|Tf
z^&T(u{aZ#kqlI5QBP68(Pu~YWfr*tAq(f4ufu=;pCSOP_G{}x&rq?3zmZYLCncT1s
zY&FB%du?~63|;~qa@(cm8gQfPibi#i&<WpE9u8-A`U6MtKzp8sMtIan?R=Q$w>ofU
zM`FqIjsy2))Yo8%NNeD0U9#iGOXwV&kwk!Zbl^_J_%Hrv2+r)QnwE3+FV&jov0NBi
zQ(Q!n2vQ=1lB(zRi-V_<-A+;K9`e9!!7I_Ex>GL2o$PC-x1baO<Rq8hRS^>Aw&vyx
zR7ds{ZdSc9mo+^(-it{XIJs<&^D{_H15M~FI0VW#fhB7fSMC(GUoM%A%&YX5ff@xh
zDA#(PWmbBXif3<2Eu@{ng)=OVH#44fHp7Gyk@Ry#`fL}_>v2C)rZmx;I=QJc`Y*IB
zynnE{7YjbalcI?A%m+izbpd#jL1r%Ji+}bX0aI7fJ~wK#W(9JnWk(q|4e~e%&RBDq
zM8FqPpSolBZmQpFN1;ht^!{Q6ZlEJC8bQP54ijW_!2j-Pnv!XVOkZ$LuJbpfe7Aef
zhqnmLaC(-kn(<0YwQ!Eju0pa1yQcj{q|x*cT?+(sMdfPm_38Ah_ojn|Yi`S(RKX%8
zZqY&4mI~i>sUNJdKmcCkCo0JAx4xzDsReuoy)f9~RGlqTyAoaN+q7M_CZI0fRwOR5
z7V`HDTPB1N=*J4t%#^oj%1|>X*CcCe8P#H@aI@@_Z>6sq1LJ88<4|cRhHU+Xc>d_p
z9(n5!*{yQ=I8sz^eSozM*woSPEfz2yOU(^Jp#Cmf$}3$m%MDDiJAVbLSV0Vc2jNQ<
zc8z2&wdpj4OC0KsXOa$1Y3uJ3!|r(NUGw<dyHSIJ)++R;foT%ytpsqn1b8M@;R_ut
zd=C3BZLz5`XYN=b61pTqv^+T2i*S1`A7Z7y0C;JRss}wbVNnT|mK3urNpP1l5<cKr
zNy1Ed_98aw20nVHH97u%B+So#oaE7`kCm-s^Xb`0Q<qjnhm={~)Ciod6lJCOsHXdo
z<SC9M2FuLsmEhh{*)}T@HUnE`CDa>`KH(UZHq6pn#Sict<y&kIBbT!sQwhzhse&y^
zJ!IOqK?zkHIQEAJH#-(z>!ebLl=SfBxAHO<==pJxzG%-Ct{|E+vVO@cnZ;tlKH0d6
zN6PItGmA%cN4Kcq1v#5icKCn+ba(6I%F7<uz+Ti;-}V-m3GFT_N6?FD#XfXJOURGG
zBOPzS*b?DW?DCUVNG)T+ZF;4GLOsY)g`0sshsC3yhOJhR1fkmH3E$JZDP~v?*~>T0
z?5sotDrCubBx!#`a;OVwYz7Ra)7#neJNU+IYE%b9lFyV^e%V>Ax#cu9&ArDd=#`aw
zJ}W81n*VBZ4!#g0pw}B=fJ?1eQIvkLrkC9^cBvk+!>3kRK1LMieGk3T`Mu?DPV@w?
z*#;EcdU-%}jG-uPRjBaSVvj9#@U@1m{Q<Uo+TR3p`D!6Hlk1c*d;0ZG`_?UOm2l;y
zKjG!^g)s%52NSl(kC>`cL+%VA&CWQ@UIQ6jn1Xw|Jlg@$QRXtW(!5?C9dxMuY^_nd
z0JoBBHeo_^nX>GSx^UgRH;8E<ZGVSay@+wH|CvJF6W`8ZaG%Mu-k)6H8#f{dGL?jK
z@p2i}P9-)E4N=oLYr>Qd5tP&u&<0_48Omb`rIaQVH7BOptPTGlGHWtu#0l<FG|-t~
zKo5Z4w&e~%@`k#=S?-cru1vf+)iu1lMHRXCZ5L&1#=wjudi9)NL>S!6!rCv*qoypf
zf9&9O>AsHU%gAJ=sRjKSXXswh@VBi<4IdC6-a}^a%X#?;bB(kb8!A2!ITUJ$J)i>g
z0XA9wh>W0cY#><V8vuJoCSbe)azsoW@NvcIK@ug}Pbm`*qld?S1<bhZmYXSZ(4tDl
zw;#dl&3X?#9Q9f9XR;E5E_t7XbvN8gNN}MR%6IjYn58yRZT<y%2FULi|FzLw>ojf$
z{uYEH6;avmQv)QQRl+ZZ7yC!gFBoFROB-B^MfHsUMlhQ*w(N@H7UohoWd~Oi)-?F<
z>NI<T0wsLE#7<-P6_%*}@^K^t=>OP<l~KWgDQEscLPIXL($<^gDQCiig92$r<z6cb
z-azG4H5w5^bgyK@3Z(%uXF0GqoM%ea>)jKkQfx~-8f?y6o6Z)n8<4)vUZ2NNsIw;C
zugSf|XX(=z2sw&jP&unpNqf1n<`WsQlnD$igEvTKa$-B`;DI)3SMi|TMq$s5TG#u-
zCLp1)E&p4x<j@a!fd!T7Py4S{Dibco*GYHJM`$R{)A=>2(aR}&h8$$SBh3`gzP}i3
zU*ZAM;Fwn}Ydm5z<Q9F$Bf8Wur){FZCK9HF!mz%Oo!BkyX(FlPm;lHGjC;f5>Ft&T
zSJWv5$VLxC77^lhbFnm^zw;znmY+x;-)>Z(ydcxc*KOUSNrG;z{7F1YIEwdR^23vT
z4jxuM;2E~5RtGGA;a9Y(Wm7wYkT)@y?;BzZRxic(Xf@{<=qXWF=%dz<ADW;W;|LL$
z#FS-C{6bk!Yo7eO(Q)(gDvbb6C`9);HXqc4&bQnCKAUdM1xbhfd0}c+O&&#s6lanS
za0QuTlvLryY{nT;C~c9ZunE|tB)VM0>2=-VK5`DUUYog})9t*b6B{I4f9eM2`J0_}
z12Crci{e<8e`4Bh2q=u)@s)%U@dhfE(-))KoSh2Hp{pq|72`X#2e$rA`yueDY8wd<
zC?m#Mkdn(e`oK@ydBT2>=``To6&H!t-{<$eo89q4D8}Zrv=H?E{d#yXTek8bjD$t8
zSw;Kqv+q>h6k7^o<alZsfSbT`@u(7nuFVhjdq&u%K#D{O^cO_}r6R=okTG}>*q(eJ
zJ>S6=|MN$uTrE)<FLF=TIN(p5;g+Nydu^(pY%0SW)`1|rtBEYA=kq>h`SxuB%P7Lg
zgM_NEOG}-HZ{?Fr&mXX4iqsak0j_vBguu%1sRZukWC6hy#rl?x?s6-@yvR$zII_$0
zB(1I)un~p~b9O1_2-ujOlMkUzd@S^rA2>%rDPa&qQGI-S!4ZfBQ*S(6lj|~7#Jp`3
zut`D4_F?$8hl=`|G@eTL3!psoJJJaPI!t6wVlpCR?<<E@ZNLC!;B_pi&<J>{!r6{D
zadx^iI#oiCqmeM-dg|flF$z~gHiehS8PMiH-t3JWC;v!7FCdaes_3Y!4<fs`-b}0m
z)17-ySAL6(A!-nMCy>}XCR7*pkr?Jw*+^ZOlMLOYE|s<b<gF%%%h$P_3q_bL5+d)k
z4;^Pvx<dQ8z8H37(P8Fngu->{SVh2C4GWr6d0Y3ibG3OFoo!y>R+Y!U5Gf3Z?&2O>
zW(=mWG9pgOdBoibDHKK3hs=ta^t6Fdkg3F5MO4J#MkUESXuVNno=0QONm<Rh<lz)U
z5Sl{qu3q91LL3*+9XLtAy`e*;9Dj1xpDFSGyq+CT7}0Kj(YwnkR&=4lf3{^&ba%Aa
zYt(p2+Bdz7ia8_^^}wXdkUgiK)o5?Rp*ye4(EL=)nd}PrpsiT2JKj&!X2vV-xVp}!
zul2Jfi|V^ks;o`hyODz8PrP{~{IXi1JGu1!U@$(GL_u-?+MzWP6(%jS`K`6l@x*@k
z&m-QnWw<xj?y#9HQ<+E63&)HPyXkEsv~fghT#W}G5@lWu<-kGd>o%VR1ZFBKyGT9i
zhyKE~8>V^AO5Zss6!$NOPj8F^SIR`l9I4{4+hf<E#EQFGrLY0|Wyz-GIdCu^8%?ym
zYfOn%=PAV_^Q}Tk|7{pH!I1P~oh5bH4{qa^guZ3-8!tRW#^(&27w)eG#})<A0SL{o
zAGKy<_fGb|*xNoLQNpv7u!@HzSRolow$OQ%NISbQHq%C-9w(UH0*bXlBBDL5ez4T|
z#NQeVY<F;-uAGN(@k<|LH0p=nJ{jSIjNYHvp%DG9=f6M1I4}=;2F^=6sz<4+XU0|#
zDX`T=IAFEB(;E$%_=j~RUU*)j@`@ApS$iHfsn(X7t&>jQIkls3D9&VPCO#m%u`EXH
zAW;o%akG(GGWqyCA34$7nR;qH(q_p3IYEa%wfLf!gW)mr?(HOE^uQ$)4R7)Qo@k&&
zVF5=y)n<__DXw<3)WZ9;efA?9y&oeG<e6oiG=~VOj8K~o>{c?%vtJH?W$u+Kaz~!3
ztM~Xi9Y>z@C>YjNUy{+{GeIQq>34X6{g^*lECghtp}~s9A6O%*JPMd!`_Qt2Io7H;
zsb!>0m)l<y7Qbq~e3*$8U)-vn=*XH035!Hd15pmMF}j*)9n;F%1%|lu3;tBi2=+Q<
z0AP)!Wl!bip*3I0vmlrV&2}Rt!YG>kzAl=PHk*}Olrf;K#%dg;yk(PK&yGKLy)o%=
zN#*fY#>lt?a?9hD0;p<cV(!Kcm`b~C%*|?=$7UhOo_=+(jsX~6n{Gn_qvY^(|7hlA
zy$O-63@&JG%Yw&rggeHLcbS?7R@QC{TtUwspm?|*6>NF*U4>!tje2f{3@{d+K_rKG
zAk2VAIJlKo<|0e#Wu5tUz<Z=^PqrRSgB8=L!3S8iW^${B1k#%Z@&tR5C*ppqZmD;_
zKD4h|Z~GZzHS{!u?McH2e)&m&6Q%|YVqjY(#@^9GA0o6uPpaVkZbdECu1Z@<(4$!s
zx=afCJhH)d;f69AkQ0YcTa?B4j5ws;u93Lb>i|V&k=c`=Rj+j~KbKw;@+mn^CwRxv
zKVQ@jn3l)4JHr{00ORRmeOE`RmcH13o5*o*#iPx|#K1Ia=W^gZ_s_XtM<E0WEDXnH
z{I~>ro86@Jj%sa)N=hv5pdSDH?38jxgfqC<xk~$cf#N2*c=-5^IKe)>c&U5j{yye=
zJ=1lSh8e`^5bL-5Q0xx{;8i#{K{Y!$p|W5rq4SmK2P2!_8cOHXm_gL6KFAy;;@gbd
z|M1`4Tj-d4PBYcxEuF3w^@I@jr3iKhuB>wmLThYmuJ`5*k{EJRk&niC{-CTc=sgtY
zOsy)~K8n9NldCb96<{7{6sD2KWCJ%W9j5s%e01|DM&=jQWI>f7soPGpHdQu*o)&q-
z%we3P*v$+@EV?S-as;W0z^!<KDRlz_dq^83x7!wX``rVIqb#3ZY+6eBa4+gD88_7?
zHmzQ`w{eUW;!$r|Bc*}J@ZCSZ853EC(x8e+kS+c@zWvGfORCRS7X`~Kx)i?rK|X-Y
zh$+8;MG^4=nA<@dB??|$>N#*g9!#99Aa^i<UBp)7HC}}hy7mL|Dh<Nxm^~0<4kUs<
zH)@tvTei4DMT{{Q1r>T|JjBI>?7;JVI=@pNfmRN&Wh$N7CfVW5Y0k#CZ>C#Pon*(r
z7@Q!=B&8|l7MR<eU-)C>#C;Wf_3Q_^@99ry91mj@S78(g1)IPRC}=cV^KvNd`BP9_
zI(z(3Noo}CRbd;5Oc~>y;E%OR(X|ShEfe=iiEK4hjH9;k*blHrr_%XGE(EE#u(xtz
zGDjs>hGIIH*o;E_fjlk3eu{N<Za*j`WNFa1a}Os?mi=aFRH~X0E@h+uI7~jiM9a9?
z$!Cy~X0c4;Y1C{Y?U-nEgG$eE|Be}(YIC^#1NDAQl8R|^Ovi00dtXf#OPw)KvTEpz
zM8O`!zEAIy1y>nih2`5nbRW}VAUdymg>@gTXq}gDe{S#ktO85V0i4vk)fW8L<q^-}
zglH2il}OgxG(Kr@f?8+HNIasVF0MPTOVXQ;Ka@%S`PCBeXOfV5uPFQq^mW91c6^@N
zAU<+h2ayjc97jr+|6RKd)|QRrg{j_i*7RLTr>T#R3{g4GD^rtl&09v4VORgGf*}zn
zFr#xlxfK5B$^L6H>dGo!ps*2H4*b1qZ@9i50b!0IC7i|x<%N|<M>rHg@ad*^S5%;K
zy77E;hKLv9k1iOg^easHBgfx4B6^x;3TbNN?zZ;y|3%h01PKG^TDoo9wr$(CZQHhO
z+qP}H`)}Je->s>c#eK8QGK-U(<g4WFIrKQEXMv_wqYRaJi7^T9rKLAR`NK1(!D5Ye
zqlLg+AQMP@v1-6HtUr#M3{mGZ8i?Q36pMWhIE_sUfS~Q6K*;u;#pZapcEShcI}}o-
z>Njk(xTMQ{X_!GD|8fnA5VzEA-l+nJozaQhXw<`%F%C;%F+LO|WBDlQpC_4q#M}FY
zhtQyU{;|k`MC}>oGkR!yVw_WQ_Xh=JQ}~Z+?wJunC%?h${OzMmm!$QLA3~?21VA%b
zuM*XeBWzMqe-SF4U1-Hj8Xpx>3u6Qs`9~d}A9?=VERSsTxo>o}!7~J=#M$SxD%85{
zc#(iPPbewM?7Sx1+3Oe~iTkn0pRI5U+n=|LLw8YEh>JjV34~W6c9naT`OvRplYxNZ
z@L7y6y?j{gj{Y8|rlyvdb_rHPgY+&ZNHqS_2m6JAbvCEEplTL0T^}_jIb4~CmmDrp
zO67CQR8vC=*vVF)`-&Y~j=P#Kk{)O@?@#smQ=Lqj?SnjQtz$G4UP$nsQ<kzG*mMD+
zgQ_d3%vxUxnzCiCTMUIZK@S-!u301>BSlhU%qY&kC$;h+@!7k@!qDn&NgP40dO(%y
zqgCFvH>Pqs9)d);AZ}w2FoL5KT#h^Lmyb{~6omU^MpK^9t(!o!u1xu2Goxt!u=<EL
z*2E1IKyaEufE8O_VM!bVLs<u`V_ZH&MqOA_u=(U$mN~}tSWv%5QNFj5{oF^D6}j>c
z?)c1WlgzXj!v~SQJ=p2)kf5|ZFWZW`965@iuHN@t4tXU}vGG6qtmT~iX#<$Y;e=;6
z#@H(tk2o6&9(!v+Fu}OvT2zN;(bNRoBT=d4&@!pLm;%FJoVpQRez~7`!)4cccfxy8
zFxLRHFE;aEF+Z0XA3P{gQDn(^J)=6hKy=+_1n@t5Fr^Lax8dV4Yf)B~Jov||Cr5bt
zvhAc4a4-ii5g-333dT7geu{+X)Q1kwQ9V_wnGl~X_o7>8Wmp?wB4YvQSKC(^SDG%z
zR_Crsu6)f}q<aPP;Ag9fABhCW4tod?*YFJ^IwUfEad~t^`1H+1f7H(IE2U{L+a_C=
zq|)v6xPHvIWR9ToBACq8ZM#2?C6Op2YjrS<77nL>S7>yQWE#g8Tx@N4PcjZO5fDqf
z-58Eq&jVmV7n3-b1z9+=?Q0<X*Zd6dkxT1VthlK8^MheV!-EW#<RZ{MyoX*5s~+LG
zOL$@+vkXvyH`XZZW3k``a*gwZ;<VDkpOYUbUFgU$WCt>O@iGDGsN2vQqcrJ;H8OtZ
zOuuKwm-z|a=l$K`?Enwso)^f7fcUx}tS1`sBEPSffYW3k0Jk{}KTvUydWUmE326bx
zPK}3cV%wE3yaTSCN5UDF`b>`5Z%f)*{pdw~n09)|W;easn<YyYPON{BuF1Px+$n77
zGw``jol>F5fcK#|1t^jWPv_na$&R;^Q@(N$HFlE-P=fK;3bhZLXS<m2u70QThg&Oq
zDJs%-dCtNih=2{Uhu#*L?qt`Q$|eNCp+zH9lajGcT-HN+!mBOzxJwG}!lUAL{1l$z
zBu=);CL5s6+mR8+eC2^&5c?&b$g;D*sja7ntfes~8_NL$(50Bd^&=pIQa11`f^+hX
zNq6Xm92METC0>`g1kJ<336QAoZ;#MYP+$hO%mdCj9q}OMvC&q%NVgh>AgPe@NWC$D
zVSl3FhIY4`xC~w;COz=QeHaFPpuEOF6X!+B1VkM0TCss4RdEnglziYl>WKA@C%w68
zz3g>jc+5TmCJ`9ulua}LQT*tZm+j94>cLlRfgz8esm(W<O|XoUab4fK@jt!9+6jwg
zYGTl1WUeMlBHmgi{|~I9$&)_a+v{^8=82Gf6)_9um#W@Je;$MLI&KPgreZj;$fQ?A
z&#T<JBJ-t%o~vIl{u~mC4UbPj#n}FiK%c9Y+OQxd3M{j6VBsd-ho{Qc(1@>qqbQ)O
z$=!*8zBQnJk7fWJtGTbs`fMUUmi1C|u@!CA4X!kLCJbI~`!T;)1N1liF&Fh7q+l6C
zu>u<==(LKNCg0?aA}`umf+O=np2jKSV&-^8;|5&zWn<)&gf@m-*~&EJa5Y#gXB2?B
zOENfxmc?U3Hn(#1pmAN3E(YPX`A+K4bKsJ>z4QD2koaWE9ID`k7@LSBxGC{U2{2sH
zSK)y^Yq^3lb!3gf;c5w@j{aC;>89PZN~jjzZ`)|4P<@ae&#`Eru2z2qJh`DV^C5>Z
zKf-_Tjw-{HyQ=L8qL!JlE{)sBb?KK*e{Fh(T=^j}I?&p)reV3rvDyG96p)ba&Dtn5
z@Lze@s*Fr^oQn6167?@g$PgcxFR4agEAfzkNI#3eB=l0Ra+9prne}!`i4+j$qsmGK
zE0_Ae)ab*O8Z?s9C-=()Gi)tL+SHlDH%BWyR2cBjUPaqvoQyzt{$`V8-8TJOii8nu
zb7MWfwB(q7)7+HhhDxQ%7cikkRwOq1QACtfbWRu!BdJ0V5In_^X(n1BpVe*XuITWs
z-dOpW1?S1SUBn&pR=g{={o~&6Ksf(^4MneFh03rl_6Xi#*>V>>^l9>KcV1d<lAN+O
z$l2!kbWif412LGg_DT_o9s)n~DLA-6+mOJ~U8FUZDy@qLZGiG7Bo?ygHPmQ*`^}Tc
zb1|ly_Kd^{15G7&_dIZ=(wp4i!2I*KG${lm*Rr#E5GNmH<s##kB&8J`q<hJooYWY=
z=5&?F;OA#RIm3Yh$7kg`3Uiw&_BlQ(agL!4h1qiYV>g_yT$_M+Lki##JvK*608*?w
zOuMFi{-X~*>|?;^R8WZhRJ`HBfvzC#87(9bY3kn&FV&U_@GMM*p7LfjYQJb`aY?)T
z;R`RsjAV@_+ooKRSNQ(vYc#zbxEbIyVWCei-v6<xj8<2EpQp|tW<JW6V*X&Jy;UE`
zH7I__`pZi`0cg$F&B{&C>3x;)W;>rp{f?zS48ZJ!+K})jW(G9AiV3KZql_V{RMyPQ
zMtAOQ64jZ93R>fD1sm)^>Mm8>ew!O=dO2Pbh~J%frTrE(Vywwh?Kc-pe_Lc>I&$8;
zM{0l93G-D6zSShL1C*nN1H2eXqBT!&Z6cnpJ)=@cc>G2>$C27fOi_c*dDnnH+_}n<
z&Cd<;i#dcrGSh=sCFue=+gs*pfF%LyKG1Sl#E2S$Xo0sz?}rH*cH~J?Zoj`~hopAv
zf>7Grtoc1C73FTTIt60n+X^Lc3H=Ny%9$2GSjxOG(`Y6MYZFb>lKv`NX4A!iKxsf!
z{hG9JifcJ;G{N8RBne(llC;&RWIj=s+`fVch*!l1fJ1$Es6B9lIuc+qJa%iOy4!<k
zHjjINkb(H+FQU+qlQhP`Vl08_Na)ksq9yg_I>{n1HxKA1j$hyko$7{Q@f8lBe)Vd}
z)yM9e5&l8|AO}|a3pMDKg?g}-vFf58Y2^?e9@Ur}k7pj96_5&DtCx7zijkEZyB2Bd
z)G%v>PFb^O`en}FOPF2_8XrS8dd(G`tlYXpyt8FlUB+5p)tYVr8-&f(Hp>zE07x8E
z#ohapOn$mQ_k*0sEjDNSRC7=6%+`dA$IvOrIy>Xr4fXV_XH9_Kizw|zXxlyUn^srB
zkQ&Uq4z%yAm&Im<_%^O46L5{E`c>FF5&1py0G}ApB;73}Z(ba7P%m~2GS(3<aIc5Z
zHPfm84*rzT<SZw0C?!9Ez`2KjKMMMUtxR$UEeVyJ(~zlWkD5&Hqy$-mRrelOwysOq
zI)rIpt<P7GC*f;DxKk&Wtq=XJ*)$-`_Oy!%k3NcCum!DPpXE1T7s*(@=B5B}!__fb
zmzN7e_hfhMBX#p9S8n$%daSm;bi)&JqwiwhMf0(~9|Oh|tQdb5Y?vE9GoKJR=U-@B
z7d6`VJFJcz-y_l9=5p9W2mrd~L9USwnypYToN=S(Z<+@Lm;JuY&Ny72S|sD#%DSLL
zKjIZa(LbQo@X#?}rF-j8m^e=TAV$7KVWAll(d-uT;@FC-G(Xbf?)=n)4x2>Wp_heq
zgnbTQOGZwV;(f}S;Hh0v;gE^BRBlS_l)zTpB&YALm19|^&qDfr?JG3_<>9hyx9t)M
z>^1^L)y=6m2-%havk1+l*3bB0Ix-Al3WFOLRkH1J`s?LCBM41`;z;ARK4xNYq<nCh
zOf2ioBP13s9ZoKq6(&FJ^PGk_%6sATmoAoDN3V4AVv%kb%=>p5U$?GGhxQ-q?7qqe
zi+-yZ{SU6sNW&GpVIQ3-Z&LW#I5#kZ;2G)J2heR-anWEo`qI&mp-d%PQJ_&0HpNxr
z3U05|^?wUA(}&OJ4Sk4Jjk;0d`?IAn&#_uUToy6Qoq@lR1jQ95X9Iy}?0V03haP={
zDGn*NH%mJ|s}3-UtW)yQG2Q|<65_8j!oKPFxiP2Y%`95qiLtRU>I-(Q*9o|l`POZV
z<gZPtu(Wxc`k5bZW#aJ=;J}z+qYkpIj^6R<r$_Eafw=dS<VXoASD~DRcoYY1;H~$n
zL&YV~=}oNQ{psy*YTf=m{punrf-^W_F*(s(<13jM43A1mcvPMsH7P`17+D1CU}lsg
z`C$=gseNpAAbTVc+p5wfbW!;~{PdkK3XLi~R4=>SkwUx((!)<pYM2=m(MZVh%`kE0
zcL<+@(qvIe?l}Z}!J(+-HE(l?$M*%fZg|A*iratmI*69!T1u+a*niTT7X+<qxwyJ8
z_L$x40eOkzpXPL6ksvvCojpyiFqoXo$3z?C%eu73-@%<Gd>MIn?jV<akc?F#bOSH5
z_`~D9d9-vf#%iZRAhlg>Nr`%3d8el4(jSjR2g_Nc`UVy`Pu`gf-55^QDFT#0jVu5e
z)?*pUpg^P3nRu7&$~V)Hug!H>O-o_h5LTgu=VZ#`!<RiFv?QS`3TorqW{(2(_2e#N
z$rFxz>K>w4OmKe|AqTebn5p=ec&80&Jhe(HRwW$?Bm4)pBfspa7xZ5UM*uMVzSP;p
zjQPnwx)fTJ(gjv8J-$>2^rhcArqk!oxYB#=vy(RkhI|Hd^f+2C{4yHE2SPhO#W60G
zFbI$rqQ5;8I2NaFuG?VbkSQVklLvY3!Ow-1K%OiT6G#iKvG(5L=AaA0+WmSA_Z4ie
zoSYGO1~|%iK(@t=7i7(o3y?+6x>zP#kbzXA6=7hg;p_AzyXyic44^|MzMRyJVMQkO
zw@v!jsNs!z=mJRZ8c5L=9h;qisi2rBJ)u1qi*D&VSOW@4{?py<2TQvH7ha#9X#%yP
zfZ7zfiB*?}hN%63%yM4f2+#XPn(cjSUQB?Nq?~l`=GZ;y&l5tpp<`H#-e@yfmEds2
zh1+yjDj!&dYIKZ53YD}eQZH_J+mZpyQaK^WJe1gsMKk7-4o2u+IQmg!S9z8C-`Rel
zwAJjbUrSq~Q#w5nm~_)dOTx7Sabbf%Xk78@3<Gn;0Hb8`!P~{h^HY%Xex`bn&gGNj
zM93X%>pO0Yz>vc@khdM=@)PRB1`5IlGGa~%r5FEP;}VyHZVGT{4u)BYr>nlCh4CjB
zilD<q7W5UC7lrqb(wJL1ml4<s{D7e41ci6kHI0~EB(PU-&m5HfVZMfY%|4}u+&ib%
zP6B`wkY?0w*CF$U{u+I3*^Z>skQZi%?C+XP4A{G?S1+_byt9}|K({{Nb3N$1p;$(A
zzxuy(sTq#xyHX?^!uI>bi7E;?VeTLP)&wErz}3$!8&r}*zYR#f_|97B^FC;B%t&It
z$Dl5zwAT&wHsZS%Z<`%YfISl!DG#k-C0RvoRmhrA^o<3urxnHMlYr5Fb^(l3E^_`6
z(WT@Oo<p#%aG4#-w?%i~$LdAvWL1H!B-5sB?I4D=lu^^YFy?~FcZQ2?<}`EHl;7dV
z$G>-@z`<B3Z6LIVHX2qI+m0<2KX;riFE=GQwe&K<Gv}tL9VX&GEVoaJFG$<Kon7P6
z!{yOt2doC>L6j@B;@H7BTa2D@&)tev=A&|Ul8?>deJAjQ^D}Z&x5~Df&I;`%?2dwJ
zb`TqzenIwSmA!DIU}kTXcGhkzmrF>QXDHG)v#btz+=ueO7i=^|aU@&Uv$A<e-hkxn
zkzbVRNO13}Ib?b}O&8bT;QO27+mrDZsF**WZW?R&yF?ok+xUnJk|06Cn%d@Omrth4
z$(2y^JagQzA_uP7f<_<s??Y-e=f+jnAyb}2)W)n?NU}wYs0e8l8dgbix=m&58>3Ao
zyPMr0Y3m3uUHETBulyC+U)>Ex3)wKp^=QiV5(yf!$Q!!x``(3do_CGaGWvTQWJTeh
z2dP0AVDo}S<IDVN>NzPOa)pw$mD}-LDU5<rVY!a86>&`2b7nLSG0ms{Gn>P~lCWkV
zykumxCU$JPo4_XO?xadr#qs2H2y^p16J#IuFzrMZEz*c$l>eS@6%G>F<Zmz704~ET
z=6ZCpN;a565f?du`e!x8NClAGxf@EPLB2Ahz;|JpAxIC}2e;<+&Z5s=&;%F#TBKYW
z%ee%@p}Ng7R|YcsA^<f3`R~Nw7X%wZzKF?i;(CKBgW)qAc7!~U+@2~ygx=2Y;=jjG
z0!>F$_Pn|m5BrHp65w4C6$(q&`&MT~X{M)rp|#((Z7w7pp;;QK;L3-xEj{~YZKgMR
zVNP-rJr^_JiV~AHsSvPKg$TP-X}c|^Gh|RmD>k%UI{jcuVM*_q7uYYok^<IVSm!o2
z4&PJ#az9jYeX3qebAp~<7zE!i@AI?*gp&RKe931PcVV_5D`ES_aJ<4GU(Ss~dm5h;
zG%lAABG$O?mCHATz#9eO2vov<OI#850CEzQ;kORysp&(c-MK|Txui&hJcqMsO!N-(
z!>J1ky)cW`+J{a!PB`l>DBoyb(^7?D=nH>Ma?Yp_Y~{I*8b-bGr0vO2IhJ#>l=(!h
z2TPB?;e>8YENhu^SNZUIJp<MGz@cdrWW&hMaagYjQURH?dQ!7-Df2BAh@fK%1gNJX
z3NBGs%c*an)9_nDK@8vo7Xkl9{j+EVYO;LMQ-+!i1yCMa;{_QscJQJvUeEXA0IEFu
zC$%AN|I|n7<h8`yO<sIZJMGS+pt7IP7nA%~7mDPuQ7-OEn`@y9A6z(=ygO6^50pKM
z;$oxV5rGcTx96m5oZ3FPx%R$Vm*8^CKIOD71m`N7cK0%p^9wD0j4potAvK1~_UrIl
za|<A>D!vTH1EVCqhJ7!h{${^{;eusu>}$$vNX4W@8sf+o0ot0j;81gK4DlVEc;n*B
zY(jj2YrFGMCNSNER5N%OzzFlj#grH=(do`PgeWrL&w_DWSny>YMFQTv@C%319y|bl
zb(7Un(YC_JRzEsMonVA)Q%0SD71Ur#gmpx<*YYUa^qQ1>>~&EZ(x!mW<A;vUKU20U
zLjy2{Vy9qAXPfon#G<f`CX({M@h^ee4ZKQ28@7X*%@(Qr$q)nL^L`TmJ3z$0gwPct
zy%X3V?|TgT(Qb}dGoM=utIY@OBy8&Uoi^cJCji)GbqRBrb0x)Nw4!bF(>niqbtKC8
zuX`Ri;th94%cEQ3mBESfpu1{+f<1lgzT4-YB~`04hqRBpGW5=Gq<X^XF*6^C*v3aU
z9!N2*g%|I!QcRxV<ZIuV9sWPSlV4Nzz0w$^u9OXJ)N-D;obWMJFH>kqc6gdCDq<{8
z^1Var$<U2Z#HdR&LnVJ&ow;fDl;)WRHPSgCMQ#y?j%fo}c8ae0(LSDPsCeEn)3=Qr
z+te+HLTn1H-hfKM6A6f`A|CEpcYky}0>Mk*!}kedih@AgK%m4Ai@#L^Tbk5RjZ~S|
zg54`WqyQKT^0(CzHd<9z9$Nm3GG_{aK^WUVU?uhHwszg&FF{}vU2T#BP_`tYZ~NST
zzaFk97{%H~YDJB>S%J_T(_U%(ST3vq#Mm~_zP3Fp54@(F3RReTns#JuzHT?WW7h4d
z6(ysHO8+#BIXZm!RV%rQ<qn8wn?sc#F&LNq=YV>$f1KuE%TlZzX;$N>AUi-tv?Ou8
zuAt#)e9+xO^{NvD0&5}Kkn;vg`rsS2Tg-8NmWqi~4xysCav;8cUvx-juv248$^KF8
zD`0S2zIm}LZKzgv+_9(;?<w{~UML?NeuuOFH~x}*x7`a0fuNRtN5dMGhj4&^7p#GI
zmqTk@4Wb$&Fenz-waHOUVue_br$sqnU*f24E0ImqS)Ybm`{Jwzah$UqRg~D|<`xpA
z1oXxz9bz48p`T8(da4yj|I7U8#31#@UUup7eVV2`p8`QD%f}G;(i4fMyk!&fB7$n+
z#sVV+Nz|_A_lzYNH93j}E5(RC?-W5OY0UfFT_FYN#g?}3=bXjNB*&C$=rjG~50=|w
zh$Yxs+S|&Nu2iXrPoFXQN1~OhbzC5=lr#DkHklKt?onPS|I=TGx_y|I5FQd16A^oA
zCN+#k?hA818l*q0>EZnHEjelcHvOnHJFP9(yCsglTc~J|ltE6Qzy3UfctW|&)@AFv
zoCT+NO^u(5&!h<aCTZ23)7ZhIo8k;$NC#Xq4ucj?0kCu#&Z^scs!G~mi4Aa(GkA21
zYQ%LNITx}cDAJ`_AVk%6&0nPs<evX{(1hHQKgy?a4`GU1nV(yTuP4v;OBP6d3>VhQ
z{}cp5AHc(cg`CDAF~ow+l+98IUBajwtxLh#15*?*qMu=yK)=b`&~2bNMAJ2AO+`d~
z>*lsmJ=KI>+8J24p#1zvlMC6kvR^#Fp{)EhN)AFX@#0nZzGboHFU{y=Jit5UkZllc
z?lmtqfW6Hg9NppWgk{aW#(M$3j`-q@Ya5}g23$l_%B}ZffLTjGQ!1COktqERhXQ&u
zD;G7+1%(Rc^)~;29ad3Z$TL$ix!+%7nfhagE+8zrcND*qmNp7f%)sS5S34_=H`(J$
ze~<ZPGEwLH8@IF`o)Z%Qf`Bv1=f9cQ#bR!a<zU3iNz-r}Qg6{nU;M9sbGb`7oZn9>
zWwb4;XSQ$MKDA;K(e9j3C2QLk8|69QtiH*E2<vDbc<F3z0yg3eKmDh`2IESYO4I!L
z-BfcNw{1Q%{3-E*-NU5-=3qLySHKhZvS1~w|2E`~X+WI@GmsAlV-r=D0pfg^^T8#`
zE?V_3{E;KOMx=XsVs4${$CRu*SPglB>L^T$<v1m}2!8-?QTVMiOgc6GVDz37v=Bx!
zQ+J1Xt0VBDG1yJp(b303S*$Y4(u`#3-ZTK4%S+e|LmgQ@UmvCfp!=tcVIrA!WD8*{
zp6~%ef=HfNt2;DsI%F@I7@QU%FpNLvsY<4n{upFJe33;RjGIVAVr(3o`WFYnx#vp<
zOm$`L+rf|IWe%Y;52eNaee*IOBG1W3Z{126>yfLKYD3)c3gQBslkE}ru+c3SgbADK
zK4V@+OO*tOnar-0w4whm4D4QFf*=RppQr~I)KC(*!60kJp1AOti4KS{yXest6)15R
zMHW8bE#2{DrxUVysyk2A=CKyvCmPVSTa!DSk%yXTXMNM<B`YcCHpt!tqF*UUk=%_D
z;6bYe<ua>O4EqtlBdr&p6qh!88NDu}lBL*a`_?Bh$AXeQL$WXhkPOO&X92#1nwN@q
zxq8}zq(imp@3nj7mVRkgx5F)YPmE%Dv<s=$#tQ?yHh8Z`)zJgWG%BYs|Dq(YCo5RU
zJ%Ge73f#gqOlT(eYl?x{jW_v@cxrG&YRdF(>&O<3_!UWv(^~>F7*<#vJXkR-C@vdd
zeoEJjS2|e&Yz#%{tBBUmN#7F(FImqT!xe|*8Yj{aQURq=xLIsVOcW`R2u|Pu!{1lt
z2knSQyeac6LK(b<$7+0)L`=6Ua%JT7m(i|L?IS`;==U-T&g^|ewZbSjAmCr<oEd91
z0>wJVzQ8?rjEj46Kl4pY3b(IV>7XDU_fw6G5GPNNq`}^G;(~zA&dn<Xwv;YPS|`}i
z)v3u1#2aTts;*F>GLx{F`F&(_d<JFT6rR~+7)<L7?V9u+$UBa82Sy{|C8tHSw(j4W
zBbL$os^Wha96sV_L(zMdoT>)KFe%jr=P|JBX~YnwudCyitQygLlLuxI{S5U~tXv3A
z`qqR^tSn|3FCZmy4vegiFPiC9afA^xvaDN3QUc;v%Y%B5X5hA-Z-gvXk2_E;P7YNK
zn-N@;{n50Zvr7}a;`kf~sA`eylbJUSgELmADIY2-AwAotM-lyO0mrmk@UQZ-g_H5n
zYytbhqS&mtjnRv)-#pyvg)i+L9`$|47iu?sfyIL4njakAkqRwH36n!LbW;}sDJEH*
ze`NF_7FBngeEC$|UVK7Ix|9&GF=Y~Got$|xtn7Bp#EsK-tLvOa%?a+O&@E0+Q~O-~
ztJ>J@{3Pg+&;84DY4*w2o$ErN5vPV9Y!9kVno`T|bFfK32kF0?QDM5B!vn$s>KR@U
zRd-C!obS`EDrR<V7~(FB=ZV?*TAH5%hNSj@mTRl{Y}*K&tNRbW5g3@;V0NLXt~<r@
zuE1qvI`MN{_9>xsGOf>z0fsqb!dKUbnM9zgke)g<A!G~)8j^4t;E!0hjwtV1bZCg9
zC<}$-jFepgpPfB8HW0DB32?SA*{EDCKMwhEG0BmaF^!d++%h1OE9LAN9yDbX+9e>E
zNQ$SXP!`1fL2OTM)a)eR1A2^E3W+A9!w?HJyS90tWVx^OXzeKb0aQ1iy9?mzW0R2o
z)U-F&fYirU5hGtmR|&IAO0(pi^<5z`hCRs;N@QAer}3pmMqM<BmT6M6jHc#w00#pZ
zJTo}Sc<t@1iY!tcBe?KbO-yWn0t`UoKcDhSD%|dzx$(Yn-c`YP!?cs1NNY*VWb%Bg
z_2x~*tyysen8;(1)#Zmq`!gI}CNCnaF{2?H{=QRR3=vW`VeJJP$UF&3n@EYx?>67u
zhM!LSi+S;hq`jPT*A71b*z~AX>E!3?NSgon9X|2DI`p@}{JLvYe}Frr0Y^e7_f`VH
zzyxrESat?Hy&r?rZXd@ZF?C(dU;+>-%jX)rdN@@+%BS5uG9PXKNX}sj2*#CIer(hQ
z(NXMA=}wa=mwVD+@dg4}yB|AU7LrlT3vsL!w^8NV^7*85KLbKWkp%n%Ed#ux1dPAp
zsvMiMy&kvKj=z+kJ}M#UDDe&b@uFthp)GokfS%R)ms^?w<x~>%NhcX*j2T3ka!F1E
zx$2{1J)J<WCO(EyLQ<f!EeqtIW;7EjGs+q6K(|5I_tTOWERGXMw)pq}2nV`~$>Vjn
z>#V*w+SRuL)w1Bi9r|(pF?>4fuCyG~TVxUKA!e6mU|oCAnLX0!&Gt!1!xCn1wn-EB
zVmI{TJJELuo$T>~8DSL4V<n@WHT4VARh+Ff5~oGRR`eu<E3GAId|F~<HX+_t6Z`g+
zsY*Nan$=DFLVMq67X-Q6z#s%C69E2+`GtFPiF-?L&YGVqN~_8l`-^A3V26q8^Ndh^
zkw7LT@M0V&U5suu&3V20@36gdp8e9jwq%)OjO+&g{)GvL^e;%XUrrxvu}Mzw+mxwx
zx}ETVB_n_QE-Xm{LLlF%;yr?dB@02~!!$-|r_4CH2Bp44<D#uj70mDtIG7^q=X0Q7
zM@s`YhPP=J?Xu8C{Ib^()ph4S@;0mv0n@?|)Ewz24jlC8p2-T>>1BLTcFrQfhldCD
z=cJ)edsJzXZw;QGG`JAwJQbP^4|ly!h#$L4Q0Q^N_xYOwQH&NQ?9fy+$CKx0z^cC-
zME6Tb4?M&b8U+fzN5)(FdDixdn^VXcb18X*IVv{d?07t+j}bUIdi`h~r}k>ZV{M|m
zDvUY#9v8&!`!}ahE{vr2^l~brW*Y4`;OKrz<}h~zr&B<RZ?3HspEdtx_t#!}w-%r5
zg^VGRQ9{>3gs#rt{?H%8ywhVjlP5JiP?$8)$mj<h@GVt(nSszO+<#XrjfT`7hk|6P
zxEJ}&lWluc<j!@fy@Pgv+Jj6g0%qF?JXV|1{PrHi1XSc#(6eN~>(I|syGptJ>c0UN
zN^tm*<}hGQvn+T+h!qh1hL4XKQN>F~mOLb{P1|3OL0Lac@bJ$}K={9Nt4dRXvs8%H
zut3W&JYNKUDV_IGZ2-$0*UkJig?hGgW(9a#Q^jMYRANO-pHHhF%1M4mw`1hY@rvKa
z!e4o;lS98422H0Z{C@S{kxzpkj8B<Nx|7(6{lEn18-eaa({`Tbj|-^YpaE@YL_J&v
za?^D-RV>-ir|a1}d|jjhVgD(Edu-NliteMasJZ-mwoZ2#rG7y4{Z$(`yZtv0VQBzY
zR2Q_o*~;FN@-&zmO2|4WYX-W}+p<G5Ktx5E9k;q;78WQf{3pZw@b{p8iuh08THQ!E
zh6Qb5PZc#OXx;=$iS&5jubWn$%feRL8B#E=SJ)p0*@i{^*9FvRs%3Fw%-uWL@Hw6e
zf_Qx9Ysi|lpPZ+?{7_CYB~>b3$&J<h%zlM)J@4aBix!*w<qJV_cv2RBvu1xbBBwN{
z*9)=}jKBS#<LG${_JABHTmC~dOE(&(EG+2o!QBqce7~?Mq36s~`rv3=5PB-H9eHAg
zdLZ5p2+p;NWUVjUr{QSP@Y#b@*}kCE1%-?JWJS*Jx(B6hHXGHwg$W8W%A7<hh4|=@
z2&mK?t$*Q`v^16YfJO1EDwpr4P#bJT1O@tQ+$F~8zL^S=T6FV;|C3r{4|DlM6DT5G
zd&DS#PawGj^hSq9bd!q#&S^+HlHxQ1(v&GDx~}1L+RKlN?EpMpszA;~_<C^-;<}wL
zih47lVwcTmsqP8naSps^RT-`*RSH-$E=zQN(`88NHQcom$W7-=RAB<;VCDAFj@OCD
zKnti8lppl~F@+y!+W9yTPHlz356L~Cb3@1dtTmOr1%=L%rh9=lXR;u5nnDwvnjp#B
zB(p42QaZ^zB2f(Nn6P-d2P#vbICaIB;RbNx?wGwNLIs3c#|wyx*CkfrH-Qj^mN3H!
z7q8VFsY@M_$4hG(<}F*{gmyb7eRw<7{O-Ri^jfaMMH~p`wcDQQBGb_TC{&K#h6~Tp
zr2bP`M5r&Oi0G*U{kGYc<*M8la-?LzxE>L5rdW=+s4sbweK^?0#sh)s4}o~S|Ax#F
zShsyuSz~8M6yXLONe4_IM68d?g0NEzU$mui`5mJ9&O$Xk*pJ78C?pAHql}dLDNvC1
znOilIRYGXqZp-cHas9<Mv@XWsqRg$&Cyn8CNhp)$vtI42xYUukFl#_MXVm?1gh*ar
zUT7k#6?v<;v-k4+eBIUbZduD8zN3SiM@e_OmkN?y(h2FiVsoAem<^b~(&sr%*|*KV
zkAUO_M6HY@;QPX{BQ1&DiIvs+i+a}waow5@=3HK-0Kcqn_UP;%{bduyedEX)y2K>}
zBfbF}Sn@#B8^*CP_XaN-fY@!ISyvj2V4S?Cv4+j^4^~WUcMz?0A<i^U8^@yNbr2sc
z!bY@JjxSB~b#_Q+@`|zXA7I`uACS=gf$R+3L)GLqE~LMekkW(vq`M&6Wd0FHbZ()T
zH2>{Vpq6z_-R!i0xG86kdsiRsOa2BtXoO`Nc=FUVwQ+&Jbnf^8*MrxiSAlc@Jv0Gw
z>9Trj(v4W(P=LTb-XFGwt0JpKS_Cix)T<}IP+jRd7ZY}2+QYBervwjwkS!AyH!0bd
zl<`eclsP>$00{HCCxBfU<LR+KQ@|h*5ATpg32yQo#gg!{jxcqC%D(|5|I~&n_3?qG
zMrs^beNt$>gb50ZQhbDY??zMObeN66Kb;aHwyeaH2PNjrnLQY6*G37AGAOfr)|jZb
zj?%sW#CQb@2S2iF?JT6D8Do00V-pnIz@-G%2q}U_3{p969X3`zynXSm^*Z;-x2lnm
zm)@lrC6W3s6<At2+I!!g*6_5JC8a_Yb;Th?+;ndH82HZEaD^<d`?Tl(@2nQp;d4N|
z>XA^2L9HyQZ||0CaGEu+Aa8J-V4X|h-@;pmXMCd6dlpDAJKSA`saJRrZ4mEN)$>%|
zIw#T$EPm@IIQ+w7fw*bdm%b8w%@syfEY->|n!<_c(s}KJe|310F8Q-sOhdVX*WM7F
zV9^!B<A0ATU9+BXr?N6S-j3Uj2k~U>k{vdMR{;v&&|mfFLcieVo>LCnqZ(i|6Romh
zM^4t`eeBV;_i?_q9ablBqJNjn+xb67Hr?MNIz6=IKt1~GciK=uAi~;TjiG!=5p~zb
zwtBLXT&uF0+4paVZKT%6G8Sn6ur;Gie$L;8A!5_e#gPFWsG};WYRYsmZ9}XEwl6Yy
z+_3K=5~pC(@bA;%{#mFus_F03kuvpNzl9xFr(VDEy&h33TjzN>@ltn0Jn}QU`qaf~
zYDhm0N)kR4ktY-ogy~F6TDh9AUBnh$T(|vv;PblZfJTYpdQV3??4oLvVB_6H!H<!&
z$X+|&sWT3xOODZ<{l(;<iL3K(29jP-GE{D;Q{v3S<>!uYolKFN66^B01-jmy0MTkl
z;OR?~O(}4ZoCB5Sceha|_7|p9dM5pb6IIZcRhH?C8|aE2_uGel1rZxbD?+}xKVakz
zGI|SK_;frr{E^V5fP()rG{xJ_$6x}iNqTd%8`-+upo-=idF7cFGI3X;4#GR>$yyvF
z=x`}dyJi4g+M>UtAOkvWajilb&!?9I(cx6DVr2LswZ{0r53RD?Xak=cvXAdN1SE%W
zD+qMN2hbD{MX}i-HXA~1ZZQhiRU7Q@K0;<?%}tLAj?Y}SD!iMPl-!v0PfvS8lVL3~
zKr0*wndfyJ$FXMwku9#@(5=h^aDc#O>Fsr&b=(1>TfF_yNqCT0<R9F;I_9qIJLjdW
zTIQ~$&PCa!TNW=l)!lBPO@`7nyF`De0)ECnxAXe>ZDs6$`4a(7rtMC%J4)0X_t5g1
zx@r6GOIw*WG6}Q`Vao|iI{=-^3^gvp04<?X$`V;%!RPyPFFCLt$d2he=IyY;&hTt;
zt+B#&E0&>xhzHAr1EpKdyos4pRz1mP1sGM)BIp|d1e|6%JwPK1@{Mo{9FPsU8o?9M
z+O>Ot!Lt?&&#_+e;A+Fx40=T88O^AuT!$M)$jQS74fp@Ny^hSXxr`SFwh5|cu#vQs
z7_rc83%9G{RBujhL`|7yYQ9k68<-&c)TebIUd74>Bx3hOwg3pot;9c2=0=8nU|2i2
zo2^W~g9$lCCZj2fz8np^;9w4L{x(^WgdnF%+29B@TaW*3!NaMjysOHLlPFy%To;;l
z(WHh4k#?t8VV;clfmQl7PiNIjHmEL>^*!?sFI@c#S4rVva1(G*)aL9ti;q2Y*Jv3O
z^_#1MJg+QkLBTxLK;ZaA4}~Bpkq|0<N9_R3yW^rdYrHlLS~L^P=NW~}%h}J&fj~NF
zvH44XdMhgF!&r60t1Y7(V1K^VI4$(%>CaYKJ#YY+5=0?6%V6&M1X?ZB&#A|H4FI%n
zsLT{qDAq`R4ke*&mw^vr4I5NP=<Zd_M6q&UlH>VvhslCKy~=obu>Jh$q{0?u!?GB`
z#O`^(^zo#{S1OG)^}WpWL*|{#74SycN3>y~52ZT3-%pz;(e)b<Hmw|Y3O~dSD4Jqh
z-fHhRIdV88W>g*g_yEo{rU(xX{5wC{2TRo;&FBNchvfEidJ%rI>v@pk&(E07Z|1RM
z7)Wx#E`Nsw@OkB+VvwQb+V2}Brj;=Fk`z;Gp^?(j5n=rT*h27&hlA@%YmCr@hZT(f
zZVqywI6`h7*Nkz=WdlCFe9GV|w89rs3##%EA{rCMLptJRZnf^F$Zi|5)i4w>$mX%y
z!<C@@MUNt7@=a1DxR7^gd95kESo9r-L)JI5@b0-5Da3H1&)YkK$~N4ORjQ*pJ8vu}
zpZS0a0O@vzt_N6XglwsY8u>$5>moBWs$50yYPs5gUI!JLpNZsBax6V*dgeVo=Fg7<
zYrdJBil!p;Kt&2+>?}{-UytVFJ6G5JCVhpZTPL1rYtGhRbESmEN5$peg68g;pR8<V
zjGS9r13pioZv<&XK_z30L{0Q}6}B`xij%;)p%=1T0rgRGu!Xl7A!7%c49}3vGkPu8
zyvh}H;vp%=Ufj8VBh~3{!SZ~A+B8{5;)X21cV>zUBS#{Z_JqdvcO){TPI(U7!&fyX
z?+(kdN9NYYgl+^K-bf#=rjdiv6^@MCcb$9hNCR189ndwEzX$NPvUua7S~Qv|$xyG5
zmJt!pkYxbGvH`>K4e$HJk+9|>d>Sp@{|iq=r}%^A5*!F|Y2X56UyoqQ2M2x%8e{pV
zB{zI$u&c{$XVn7`6E+PPiYrN7Rv^|;>btgaP3AsZCh1rU+Qk9}NnH4PG$Uu}9Y)E}
zRLdEa@#*aN018NUhx~=E^H7g!Wqi%p=!znw<frVU65EF^M>+duH$^nX%?<U@CX1AW
z_-u*T%#`Q%Ph9*luXbruErUo1*_1#SsaXFc?*?(N{z!`Q>kNm%Tc{XaYRD#^0j&tG
zruShj_``)S*SvzuB1ubS!kP>5@ikfx1$JMKVA0{1{%rkUMvI=xYz=J#*R^*%XKuKk
zw$0t!k4-&{qs&h)KL*4bxCXKv$sR>--vOJ|Pz}w>ixf;Pe@d!3H5rV)>?!Y|t3sKZ
z^pA#sa}7dZLANcr+?apab)m}khLLUPQ}+zx4Qx^uScZ}PEu&Jf`7^MRn`L(;O20)!
zzk251l#XXv7$}%EezeCGY2^#T8F=JF&O}~FYfa(@VhKX=?4T^eNm9BJBnRkG98IKk
zM>JIUqksSd_R;Ncn{Y07%^QozU9#>4c3-X+DIe{UB1}6mCmnyLs62v%Pl^X?Cm}=v
zlWi`v^M)WE6;!m&)2tP2Y$Ad4tEemMh`B)|>`tx^gh5BH3<9F;C3b45zzTK_OYQC%
zNkgb|f);EDyG1rA<&kKgmq=;08S%XrE6^58#)C0*M`lCr!x0c_PYDEd#s>I*bG*Gf
zcIEURZF(KLy>15zRmPgY(cgx%U5=*$VpA^a$<^w6HIobYS6NJtk6Pe8_l3)N!3UoN
zrHkZugs;d|D%K61<Z2DRdzQ0TQ>M%xW^#^XA_ok4OjyHip8Mo4<uEL-ZCs(0a?B*}
zB{E}Fa!L3m6L31nEX%U}z*J-jblagb-M<hI(?ze~Lmm`o|Cg;U;f%c7&9?Ph-)auG
zxsUrjdsA4xRDRmS(%lwccaI#IC||ECLzRr3G9FiH7#B7Z+{^$ED9k6AJ6DTw@x8Cj
zB*24y%0osfhWy!|I0Z-ESJg#Zjs2k%dACK6zU0f07N@30i8dFUK9LCV8liC?(~D(*
zFOQg#$hH^aIGjdIDYv5cI<xN5VG|r$LKGLj;(3&{ViWkil_q51kh`s~?dLaexhxV$
zBs27AsvQGgSCZ5x*z0Y{HLN?}#EEP?O@jsvKYwz<0qneS(BeI!buD`zx1{V>%dk7W
z+l(7Jf0F~l2?=UQQ4)lJ!-mqA&;q+c3<`cr@CwT7Y)0Jhecs3`>ANWHB~GjYQbka4
zR0fbF!)m=8kqmw8IEFmq6T^3~zbw+5<V5!hi_)u+WIfAeIF1B6FaJ-gOs6Oy6cnQC
z1hZKL9g1qVtAvQf>8zcX(((*A-ssR^1id2mVyWavq1=vs-7aPIpi>;D!6ZQjMRA@)
zh;D^?bjMOQ+)^=Ec+_xWJPs9$r=7L)2(sGIrwpVXtP(u;@89Oo<5S{&8hdY<Fsa0$
zo4_Q6Em+>#X8<EZZ?P+YKXh`i`pPyaNn0ZcL`U4;nUay%&Dy_V;@1#IRY2B~EIDyc
z7h~quj@5A>gxxq&dqRPuaT4#e6aE{6Ngza|&bac@48T|$9Jd~~)~fTg6wPM>A8UyC
z5Nwnh#j@6r(+}YvzR-u@dN+JTY8_HkKiNdcqMidr;b%UFwH<Q^>RkVqq?jjg5qx41
z)YV}jvQTYB6y-{#b+PxK!VJ;sQL$5(0;lTykf(vseG|q~ydsQ=KpZdUUrDDDOY^h9
zY6GfDClK)3h{WnE0Sr9F(@p?LO1=Ub?Wgg+wZ_Ao4F|^^`Gl621cd^?DFhfc3Zz6U
zQ5sOt?P*5VkZLZ6Sn9(m0H$FXJCey*FI(2=A@`_O_tDO`RE3F}hY$y6@L9JqxcJZx
zE+lZjU%XklW9Keu(=f$R2W{L>JunnOYA+6y1wh-2(LO9c3nafV@BPX@Ne>wl+%HRm
zy7@zxM|%3&be+iEYRhos6h5+uU(i!0s*9|BXbyS0*!-;733Jq988vSUg_4!eC*^f7
zNpbdZiOoT?1n+A?Z}e&kSRfFPPg_<mK}fbxgxBV)o%q!vZCj$Bm(Ln!d)nq{Kp2P&
z3IfG1JuvPoqf4w^fAb6<g_X{hNRz|onpfFfE><l6YY9*z1Pc#jo`Ud(l*^2Ag1wEn
z;h>4ewB0<zfj1F1toUY)msM3}Q(RAc-H^bM<$FM-7KTZ?Ea(VlQiBCK-HTUGAR{Fn
z$en_aETD^soZ|Jy<$`RP#&5Om0#sZfAuwkFr0d0&S<E-#(_#><7SA6Lc2|P)&#)rx
zIJWhFH~To<Uz^i7dOzGd#_z!0f-t>C@rzoXkoRa*We^z=W>m3bLHW{=#>B-YT^lsq
zu))M;F{oT1kh-fvXMS4G%4gmuCv|pmk4n5C#&vWqb1ee21JS~c7aC>wPK^t~=7o>7
z4K3y>x^*EX?N+8c{xbA~eQnw7a+g0{bcudYb<_4?l|n8~I^@-lRzjDvnOI=+Y(gO3
zD&2jyAK@FmN5d5Gr^6i_ekTs8c9h)5?-$0mtoP0U`qIePr0xYMUsr#JxT*pVLhw%;
z-?aj>$QSg>9ZTx$T>PTFq@{}hMT<>42fujBqA90R`udS9o2&%7=63f8V1Z|}S1dLL
zy86*QFU0a`ojKo(w0tZgJi>Pim_O9lJDz~x#Y^>CE#jl48*V8Sz)z^C5%*UjNf2CG
zpg4j|qFpok*%_9zU@Lhga$dIyFI0T)t|a^Yc&mdFGbDA$io8bZjgU8Z#(*LG*;~Fb
zI~(hdf6#Lmpk4V&{4A+n-zK|07fuVE(*|#dwP`b={xEs~MRc|Sb*UUUj!9&-0<6K4
z6g7%i<1wn3TpP`|p%3^mRrAlD#CZ!`_EU%eXiL4+MR#B}g;=kAPVfhhv8LvVeVk2t
zCgM6oXQ>1LY%9pK!YIrkcQU<Msv6I4%S*fKlIKq6XF_eW^Dj7Aa@@kPD33wf&0$3U
z+#=p~23Wy%EffKu;6SRcng!B!%bO6MiN_6}TZgpKJn1+=u8Jg4e~-H+H9W?j#DF}C
zkw(|e7854`81cr{(bhI_#^f_`lSnmgQz#+pOoUrOiNHI{o&;ubY~HlkB!Vn|<?Rjp
zaxtRsrrwh|DM3s0$r&+;j(i|-ggYgU=m_B}Td;k#?Mhy9Mw^_ALGU$|Q5UOye4@ND
z;?p<rdow6{kIWazaVTjImME&ckN#?u*g)b`Q{ZrriKq_qMxG49GeU4=R~H@F%O=vN
zx>bxbA5z}10F6H)OV<7I2q+avBN#f=&0n^-o}RcghiVX^VI-`d?iL2f3LL>ELs1?e
z2chxV1xY2dx}_nBm#rB!xCm=VVd;}1TBxEqc_;2h(0>|nk(G&q5Y6q`)sr4a-L;bZ
z=}sn0iB90lMZM~Z=@sKvv}Zf28AU3(=!f0c_9~bRt!$v>u=w;NXB0^K=#MzL``c7L
zq48x<-OYn6rvqtC^KQ#B`3{o*fdLCEH2p6lSKIzZA>#wU<|y&0PxW}SLa0IYX>)Ya
z#*{v+xI_8_l&0N55gGZU`R@<>M)p>*b}FTq@@AhA*d!};5`Iy(74PbiY=HpM&Kxx1
zbM08K3i$3b_*_0Ol`Ey6oZ$J*ndWEGb<-mNf;dR(#Z&LH;p(*APTf3U&<KgNNVC*}
zd}gTgw2Ex3Mg#7|MgROx1d$0Z6Pp0CWrw1~V8UpJ<)9c!?_I_tvKULQB`3L!eHkT#
zD!;wGixefj5@@K8E5_WfMd`*#1Kw<*-_f5~grUNn2>5Bv)!Cd_8%&KeNhD{U@V2x?
zHMlyIF2J3L5n`uF!Yp-?8wJBdjhkqW-EK29_V7u{#2et7fv{|8Kcf+d@zs=ScbZ^;
zwU4Y&zQcOQzoHH+83p%_Pn4hv3BQVD(F=4ONqL5b-Efielq!frUi;fw@!yD4RwvL^
z&MA`e`_3!iCL)3b-Edl(!zTT4ihwol)N8n*!nwz!z$YLv>d?x~HZ;fA-piZ+%+j8S
zt3q!6xO_ISuk_6w9Cs=Fy-)K0dm-|3Y<8{sTo!yIUYcSJ{?t{kN>h29nvzisY#@Uz
zZy$#x<ST!j#86^g4c@R4>K2oEc|KPFqzlUU|3#Rbg*WLtCHGqxIL#`6gwQ;TEIUh?
zST4K)ztcduVI1GMUq%Ipug%gHyQA1N;vr}-PY<qDYS-~rU+sol(8bpB!(dB5Hq!lR
z#W*|@Y!GAM5i-H<Ox_RDc(}N=gjSx?i^%rFbB@E<2TervAJbYlvv^u8aMC-n37&tu
zD7+FoHAZXN&F2y*A$d8q3(nQ)%=Vj08k38L>!gCt6rFk23bkcTPF{n{+<JX#Azo`d
zh9UM(2do2&#-s9HJKa>PtIuSKbgHr81EABz`jdh>@wG@!UNT_A?8M3)CBI^zDj3c>
zIRu3SIqLs1hQoKonIq+%ZJ^DMA19*f21~U>bNSU31SAIT`&bm4%DYQnB@>iI-F+13
zWY=Ym2pwI<)%TZhhEifDZEmd5@9#`Z?7zlyf=v^<+M;>vop}DOJRG(>l86dWc@-6|
zi}9`HRCMA5;cDAfZWKjn>&ASUuk8s&lk@kh3WH|wrE}<xXx9FCAYV&;Tj{&GtO#nF
zn=ssuj->qRz5|zt@V^7Iu^Z=+J=CB9pB8x-xG_#uXRVH9>S8+1qDcYs?Ow@&w2Nr^
zNLFXSg~hVyR~b<+5-m#G+LS@#Q<-P5nrGf6640ux(cKw%DoANSB2bqjL~qU)^4=lq
z1XcvJsc%S#xzRc54MxwL(y&us4)z*|^X?H>@5@oXX1B%+&n#s>C?VFYg<Mu4Rll=4
z`5$lkGBk&I^!U4XJO?pp5gs<F<^bu(d`mh-nBVE5PP^a?XLmYy*?b5mggcjK`KKP&
z6Wx=JAKj$ff>cG->J%AlIn;7T{!UAC3?*qoy@X*C>?bwbKy-7L9PE**hR)rI7~k$v
z@n5y*CjefUb0Q+bHR&P2n5<{NRbmfX5YjD4%QLt88T1ak^zV=Wf(NQh0S;qE{;>O{
z3cb3Bs$&;n$wm+oP_XSC@%ebm+6=#H3+Q1?4)fJ!YxHKrpo1(lV+w=JSVZ3EQRhq)
z9FSF5%j2~Z?H3!J$dJw*52w;>6LKw_%ZHy8ggheE2k_Ike{sU4;%<Cvc6}#^7uf2_
zaj1g7(G7<%(fg1?KmW})ZwJ>XtF-p319IuK9JlQsGfB?J&~^H+5N5^UJRs?A*Zn%9
zJ<;deM&jarsns1jJBotUq0WLF2;Rd7N|=k#_Kw)AT8qJg+}2p)o&+!0%C2<ux8q8b
zz#H)qD9kHlLf1)FFXF(%x9lpyN%O2GK+U3=trz_k11zBgx>hs$g^%4h+e#238?vRI
zAlrv#Q>GtgTlfLBK~yI47RywLv^(MpcfL#*hLrUs3QIMoNG%K!GVH_K3(J}mKN<+n
z9tkUC4MyGBFLluA#=Le{Y%jFNa{Uii=b$4Dpe*UJ|FLb`wr$(CZQHhO+qP}np1mj8
z&F0*f=t@_AU4>rBwmi~_QY_WNgMaxmG=AS_KaFz{mY%v`{)Kx2lLFF4G1`NgoU+^W
zvA$--*on`W-}lVOy<~CCBhpkycR14$h(Z{552Mh!%AUNTs(S$?v!lm7kn6O2`RvFr
z1-RfMH!cnLNVxc!_NvB{h2z?jIB3YMUse3RvlX<+=gC&^F9ql>mlW+yfEiSwH#4Jv
zc*zuxrdwNn!skWZf^vSS$*w+dD;2|3@vDbWR`st=xh-MTP<{rb8JBe+hn;1hZ4#(7
z8n07yB~fJ;2nH_+t_+^@i9jeYdg(q14;r#!0)Wkxz5LNSQUmCvIZ5%Ug-Kw1cxfS#
zF}{Y@Gj8lUX$7CuMccSrX52)Y$u-9$U01Wl@l%(@Zo4SMOXFr-F|+XiXHarrfAos-
zt=tM*zx$2{udwbuPqf~fMjk)={>@O`C{M2GO^|K^kHFM``g<!eCs|c3pv&zKpBs#q
zED$rD?@R^vHZXPlohLL;bt>SWR|gzeUb4!2O_q&^*xfrLe%S|mq`%5E{Zqlr;?C30
zT4NrE7#iO2cXT%Y)MK0;&Q1|#)HLnAT6PKdb|m^?^EFiRaKtSjimcNF^(<HAEsnQe
zU&pVx<Lxqt77jP22~Xw~l?2|qm@(P+C$nX>c_ay;nqpgD95AY4)hn4LM7JUvl1LD-
zAtRj`azv_5&G(Z@v$$?*#l-Ll^q8zzM1IfH3GOOVXMRxy9?YpN!O|9BJ7pPjL&6=w
zxNwN!4ZuOF`-Y$40pTRL_|CT;iiLLK_|u4quQ3xFX{ikmL550kCnl}?KT96Id8`!r
zF($sOct&#o%PcDaGf2Kh7#T36IE!L8st~FCw@~P}cbAbmd8}g6!+_W<@gngQRTqk~
zEV?4xtFOYFV_q?@PyxD~N%G<<bG@C_CBe}6Ah`jMfI)=hb$vVvkiWfQM>Qi5aC4he
zkqw&v54#^KHBK`n|CA}js6&fj!zHbZHnnO|{`HZka_4sIk6I7^==GD7Pklr4P*rq)
z+wyO76w1v%VV;F#4+@%*BxB0!GgB{ME%&ypvpnMQu!#-*3;bHHOs|X--y7BbaPp7y
zrZsnw+`fcHfA&V-A7|pU$_6h$5ExH{NSJ@iLTYr}w~rBd)Z9)0Kfe3Pg0{m3L$@RD
zcWGnplz~Sf;<WMj!`-+)i|0=)Zw=Ng&<xPf0Z9Z83|iU#Tx7dy?Ed?|7u0>LfMGNJ
z-Fz2~N&U+}?2%u;I>f_-&U8h0ehv}_`0oP{Jw^IzR^gk;c?1Fe6DnU6ZWigM$9DdX
zYI05ue{F0Ys14CbSfNrZ%^RG$Dt#qa8~Op4N<!Xw*j_wk;F@{n%s^i>4{4}|t}Um0
zhYlE719yi9bX&bws$LWw3af4dI@xxnCz!pTeIm-I;o|JvO>&%0A8>SS>F$kK*+N;F
ziOiucOXiuEBXadK0U9_Y>?wz6rR`KoljKm7xF>cq#(V4_D=q1NGKNnZd9fI%@9+Nt
z(!dpw9eb5<xUhT$ETn)pQnCb%RBpAE>TPdm)~RyxIO`r!(P5^Z<Y<k11<5v!tWG+n
zyN~A)NxLtE@X=MW!NoT1`d;80%=X2VxzB+~g7v|H>;?hQA~77%M5~Z{T=G-lYs#(=
zv}lhLCZ|AYtUveTVG)NozOS}>lK?8oNzdiISR>xSX|mO^bE=)ZLEEOU2u3~KrE>!z
z5dWOrTvleTA&)Jjg5*sH7CO%D>PBI(@8t7qFK`Nm_tFZQS-LR~x$s5thLVgq5!F7U
zr9K2Vgy!yqR|he9w4eU2=i<`nXp9b}<30h}Sz2DPsmb0GYT_axux=q};YoFtZ%yJ6
z>S0~X#Z5w}P!BtGhpoCIvI=a!p4j)~Tci)SHRdzjT$eu^vu4JW;tz(gt*gb?C6i_4
z1a(*QQ|_oQk*b>d00!;*7>UvOiy4CC=@u~Fyc(yaD&0>~CZ-styOOItrEb4g<P!)6
z{DP)A%)zVAFxQ7}qJ!aZrnCp@UQ;vk3NU>TVquRi(k}-swL5BS<k8fKV-C=COm!^~
zdtxAZJaN@+k7t7u+)V2Dm*a#Ac)f~Wf7{Bj8S@G=c9NS=&6j2}D3-HoP3~&fh-%hO
z!VnxN@zF(8f7UmkB2z_y8=xMXxXpic@3}#kd4T5UrK&&cP+naLNJZ#bDq)x(v<2v-
zhXB+~hxw@LFeTV*P^M+9SF$opt+f;K!L&BLmy^^fg(?S?XS<)%&+~^3IXoQG?CLGp
z+!?OMYnsW<z~a!mQeiz6)LyDH%>wb2@dF+k*(PPKVj%s3OqqYNXm0VAMl#Q2oIsWb
zE0eJd)(8dpE0%jTOBojEruWkWiXgRJbY-|nOa+XE%KTV6cbYhQLRwxv4vqb&dENI+
z1Y(+OEIA7UjnD`?d+D1eMAb&f;)NPH?`)ds1%yM>o(GR_MT0mW?{qOEdlrgh?Iw{%
z!wd{v=O|4iL|eTu?aX!OW7h)(?NcGml#0Bc04^VEs$^pUTKtrJbJnYTO+X?NyU8uW
zl)C3~d?(Dg2;$Q-gRuHRw?goP3ME}Z#Du#AB|j^VQdFf)fMFGZ8{b$I6MEcLGMBrB
z-wq<7j193PL!H5WT5b8UddaBcFD(e=h%8<6z44DzLchcCd17bQMZeHy`liQ~{9|DS
zIFQz&V{7X#ZJJ3%2bRR(RPoAv9QT&yz8S$#(Fkpxy52D{Hl`P5L#|(!%2iAZJZDo+
znCfxF)(>~!l!D}8n?$a;)ft&$^4{7NsZXf_zu%)aN|sNA+cM+fvHeFOsskp=GqLtB
z0wY{J4jD4kEMxV&vs5cs6a+>Be`9ukws2WyjFw0yT`oC%S$|hNLk8IXJFDsG4HKnS
z%cI(XispX6-T|3U!Fju`y|y4~X5>?2Uo9W(;Ydn%6|PSMtI0N}0NpD~N0JFibEO;<
zMLK#)u7P_zqI3Mp#)4pZ_;fO>cEvd`;G_-?S1y7(A*@IWXRsQGZAGnZ=B=*66#nuy
zqS)2xrY^pC4<dNwm0M70G;5T4y{I&o-q`B`qNYZk>*LZs1<_|@uw2wd>S57vuCt)c
zDRBOuzYp7Yce!g7wY+}Xm<6nuhr{6B$F!CoRib0QleRgQ0TpZWRZ5yds&%-JMXu<s
z6?<e>TY!YO+=rMuISn5>;gz5shq0eqDo%afeCVhS7!lzklaSEyuxw&&1&|mT?_;De
z)q1<IQv;oq@VA2U?veZjVgo=G9`g2ti8%dSGZI5r7^O-8i}t!zsti8Q-hY$sYp`s%
zD0n0NP4BP?D>cnEm(@Y=LkI#xi2KphfuI8{bs`H!w$`J!JeA?2OcsNL=SQ3Lb=-KD
z(ToRIZ{e`^VY+Jw0zsQ1&FO3IKqhMFxWec`vPe_K*+7dZ2%HL-!nRG_9)rsX`cuM@
z`s_y2nDI~dh^MQ^BOB^UozfLryZiG6*Xmr=Z=&``u}l%XIwZ|)9TNyLK~kiSwxBNI
zO<S|X>&eDaEP0P-Ax1d0AALFB9R#OOy1ZpxiFbq&?a0TviDTx2HFabRpbscQ??~K?
zmB_*+ElU$SlN0NySAZCq(g!iMtAQycXk#~d^WbT<C#YGq0)3znN!(vMAK+XiGfz3Q
zyc&T?R?26_Iox~j=)Vbd^Ub2dxc{U{<64Z?N-u`CGx?a=44oKpM*h@wapU=e7@01?
zxC=#pX@)Ao;95Y)0g)r#07T4>@_&+3`}K~77H(9UDZs<eN}Z8ZCF+<T4dT<;%BH;C
z%SS({1!_!}u9KKn5ITpX7Jb8ix)hbAQbj#RlkO|=|2w@;{9lz)r5<}E`T{Gd1zkSf
z*~zOw;$-7%vLd-O=%QlMRlk&nZd1n@&7=@l{X~vw@Q_K^E=Z<qP_2(or9^){drSQU
z@5t~E&`n3p8lI{o)*GXtm7U5^J`?PR6+Co3a-~*K&}&Wl-kNw!yhqyU|LTC@45;(f
zJRAG+6-o2+DUd)R2`8;ji;uiF#|e`8=`jsmO1*}39u%QXg0g;qy*L}va>3ye9v@wL
z7lwB?2<ew2*p(6e*Lutq)MCQ`*LMUJYX~&Q1StAKHV92<LUig3LcWrMoUT5&DX#!}
zj~b$8y4cj;^=c41IS9Mp^(<;XrC2`hTKrpk!`H~JzovcUc%=TnCgj7DhSl3mJvD^7
z3(m8;>_uZTqX{C9GFMx%RZImg;k{SGy}hxlCnoo3j8`h(bgKT5?!If`#t#RC<mZP&
zQCMv<C<wZ`?qYtVKf-vwN6{5WDOb{y27gu#F=7A@7;#zAyOqk+J-*#5m%(ev?G<8d
zZsM}I7}q)BQHm?4E<+$n2y)j|rYBcw|Do9QzxB7PDCA+_P&c17RgQn8&QFR0vi8#;
zFC+8-yn0=|XkV}^Y!&^y1=S&uD)@Gca`@$;1CKS&ppX()NEBqV=PpJ}U_e<oA{5IH
zc(DON0$yPucvLAlqip*a{L{o>GYX0+w=N5%a{gpWeIMV@&w1qu`>nadMbCVaRbNHI
zHiQydrgNt0UR1vz@VQLH%EdTc*?iGY6BrlftpUFmt1Fr{$M?mR(nvy%H$@l1izH9R
ztGmwSFpC={{Eb5qai=F}F}6U1j{Cra9&YWNAN6;QaCJ&?;hd6~5d{32MdU21yPCbH
zr{%Am<eRnFtc#!Zu)>P{^zlcwdombbgyy-OQ1keTXl4)$DlWXF>{a*ba(`huRiJBA
zl(_KY1qo<DR%0z>yN1VEAIL%)nBgKd(%E{vtFThvGYr?1^=qGR34(jC*kMeiJvVr*
z{6yEbe@~`UDfG>+Vjl?_{|cpanF3mA;AY5^p_}{)2b4(ROt51F)dd$J5Y{cW@Ra9(
z+V+B=`&NPR*-mU4+qTC|RM`+w12$|P(zDlxqmywxS=8L}q>mmY6nl0xu5;)^_08pb
zEEEWk7P%X;#C9NaWUHCm%3Ko*#Qh=vl|<J?D!^zQaCk%4SekMb$%N4MZi1;v;CO%?
zC3RC4DWJCd7?Mpigxa%2{qEpH!Sn)|l(_CXa*HgAe_wh`D&v~0j0x0EY%-&qAr!8l
zXFt-Rk3(_jYbOnyA9>Qk)2-dmIwkudf31lnWeq7V3Sy}`vW@>P(;iVfVv}P-!$AHl
zX!b9?>qrTA$h5{X*ynIS;mQDUl;iTp`@QVq(k?!N)U&@UO^0rd)R7~ZO=^5+w{$BX
zZO!$T5v|llefPg@a@N~!Mz#`L<AwvGgW?#>X7zXRkmDpV@wzCeK)SZ}qb|ZOAM&Qx
zryVU$=gskYgq0^CQ?FkpZX|~dmk_9m^Os!O77|275v+nLNh+kKO|SIJMuV8uo=e6^
zqyGI~5gz0nD$r)d#d+gQ;eo1{C4vNqaAGvi8M;d_FeUPiLd%93;Iru}>u{>FlBJ!L
zwPs)OPMaOQ?RNP}R%`qTR+*sFC`U^nT-D-lEh@Ht$|=}Z=vdQHd<&{lC}t0*XL;r9
zj-VY2hpN<ut3?zi;cG`kYom#*Fhg2=PlV2u+{Be!zg+I~3=`)&R=!z$`snlC#9EuZ
zUA7l_!{D!fRJW&a;1aE=g@*sB{?ooKi$Tz~`In#Pn}E50U_DWIv97dsnIU|0JTDAi
z1fofcd!PJ#D^+x@bi*3wZtUq4R@wJ%Dy!*E{~)M2{b6`2$q6b8p+hweyPEZ=Li$Tj
zT&Qqu0rQSe)FG`41&%=1Ekhd@9Z46@UUaPp50oj(^nr90!Jc?Y7=3{xG&h7rw*)dQ
z9HtzZJND#`b6hKFi<u97ev(4&NhTQ{F4J51d??De{#yF?o<iln0ZT45>%B&q+!2hp
zT@XPG3KU0YMNxmH^;fjrt*RbIT`CBhS8RqU`Qie>wTLK4$L+~JGj3I@6e8w%#ehvg
z`Rt`^dyoOvlym=nK^65l+A(utvgc544y82Cg{E)}Q0a$-|KDoY+>s1Knrrc_Rgp8M
z1|mc_$Pg~=GMId=NWdOo+pJ=g*+vbHop1Y?HNq}J72Wlf?L1_;nQDvQ0+^EZR=Iro
zW^WDLyW{H9E-e!*fsZEHm`h9-&+jJMF;6$&O`rf?=O^yp!qKLBKoaV?MU{kA4&hLR
z0<`|Gj47qNoHZ)P`G4`8Fu<*3!-p+}94X@g^_uu<9%P=ZG#lM+*>PdIGh@aos6P$2
zti%-efwXeJ;191?G=XIU`c%RAwOx&L?@yHrf;8jThcY`*vN-08Cbg4B16@V6w{K<G
zcGun}8fD+d2C2jWO=(IzEJO*%%`AZ8REjd)wYPTqZTMIjmfoiOjR)G~lKtgWTy?~W
zrnRzD=!3%_=-ve7P)iPV7-1W=43#%W`1z{Bx`W=STEbdWChSc44&WNCi`$=b6Qwu*
zvBTAV<|b^@jD<|E{vWKUoa#WWP@19tgfzj|qN2xZ9=C2i{dZF3jg1W_c`qK^`%x%{
zeNJ=7=O7p!W-*s+8f}Mt&4dF47V>1$mT*53F{WB;!(ychpyRG7mQ+!{36P&EO~bvB
zf68Rbd~adLO5IJ_i&o%;H?{Z5{Tsi~&i^IKsc1=qq%6Sea_O4!L+ko}6Rrqd1Z}Au
zjK|YqvJ;rdMNt?CgY8lszOkNs$Q#zXXl?HbC9KTTay(5fHlk2@h$wpH?Rt}Mk|sEJ
zJJ@YQd;IYz!QaH6+h9SDFuzD(RBGB+qZzNC)Ush6?|&I4o?hVp(n1@SY8v$1=Db__
z>7Gq^+4QsUY&eGIuET_LRgRHu+Bk6q15FAQE&JE=<5Cr41OwMwunI!3`?<niVF;0O
z=-Ow(RET2mxHK~ko~~rWjCTvL3GTiS9NWsyJRp=Do<@%UEZNsH1V!6}KP5gMkP?xB
z)V1FeN<N->9{gUZ|7g0eZ>9$cT*8~=u~OK5=?=!6b#5tSOgTY2Nzw%Vc}&ll)j#^}
zB~w%8JcOT2vwe2w)7#L5BBV}0-E!K%F8G&ygJv4U-_eQqD3~D2u4%TgK;JEV#aTBn
zC6F=s)F9*;J2UJ?gwFUVGj$=c_7}INSQ3jvhWC1Fh~@SojJ;98{AQi*mf77KUm&D}
zC&`yB=X*1>jl>m-vv9%J={^7S(p(&Z8-QSu>Ly7wQ-9jo^iN7*VyhQgYa-1oBYclV
z;r(br?~aJ}AMCB=$jQ%+i+X8DEXJ<u>)TAKA=_9&vm6R7`iY-y?FI3{Wv?-G-RH2H
z*(%PCr+S+(mt-;hq}7din=1W^yF<7pCfkUngPLG(5P>Xc52?}BPClHFhH3tApb5%6
zest%OKI#Y(2yUW)x$BdXj8}3$bK7w|l}N#KX7*D^4OlHx<rEvBQ9;QS8J`M%dyJB+
zL`>|28Bzea%<I3`@}-FpeSJ9-R~9N*5&Z3$M#f6F*BkZiH_KhtZ0A1ku!Kv=wx&!B
z2@0a$?ZeFgWk8z0?W|odSg?%72bf=w$6WH&6rviBZhy!?s2r!D0XUm9<E7083(rMc
zkGvdsVc4Xb0>LDP+LlAr%#vFs!L=zo>A<Pw&ukg?MI%76J5fpln2Ij}{}=-|jRTMh
zj+MH+5hy%iR997_OdpB2GQc&=i_spfE1-IxZy)VZl=wU<vq}Be)Ky>h(M{|lik@+h
z)s|F=HsaVNtwPe}Pu=_Sl?fi&>|t8^?D$X11Ifa%FgYp}X8CyZ^xkR=n1yc6TbWVP
z%XEZs__fXcDHDdSnp<YI*&R>vaH(KDSMsqxf*q>@*Mrf`n$pEicvAYro`WEh9JOyl
z?Bh8CjO$ga-Uk|}^^01BR%hy5GnDBbvhN}+V>FLL=dEf6Vtc}1tq5h+y*cTaZHLsi
zJx`chd!itWSc)55>ehW<Rj0?zWuxPX@>B9J3(Yf`uO_6>o11i<#-0R9Kxu$viB@u{
zM01;OQ-(r6Cd`KAT_xwTa^>B`(j4X`dy$Zi6evUsjl)+^kKQIwXz_R#XwB2^6B;#r
zoXU-!LG(H3=Lbu`UdUH4qvxU1Jk1{k;fQ-YoeP+lS*X2Q<?kr`-Us7N?=mvK1r$zZ
z@g?wnie-}$Pb`Yu;c&-sdu-C_3=u^}5e<=zkN#nwYfznjufJSJ!Ew2K;j=9=LJI~%
z2=b0RH+n73O0_fy*>ptbLd19}Br-Ox5ARJ+<>H-jfyE6s)p9&sOa}&KUzo7Tn^u}w
za)FKWlC~CuuRa0kvqHA3Q?u6+0Er#1c|60!@9Uy-JmPueuuNiGM%-78`Gs7Wr^qim
zSLqu{??ue>ziyj*NCERje)Uk7hc3+6_;H*J0Ho?c;+Du;K3>hcthF&F(d+(7`soF%
z^yPJ9PK}~lQn{cmnY)Ze#n|pNn}j#&>&baruf0P5BT=@B=|724WwYiL$6f_mFQ^~?
z#!<HT|G!i6lujfh<g5Dg<&ai+&8C_9*QpceG*hYf3Sa7j>GZZvD+V^KM7G<TRPsso
zLjwoBx--5fr3+$Vr46bSZA30p5|J~5r&8^|`tNXBpw9@HtWWal^#iZ<)kU4AKV2iQ
zMb7k+ci=E>k`05FF@vkS{HKBAZ@W}%Dvbpw=={>^ongyTHN+UNwa`mgFX$=BbesWc
zCX}**m1Ou;i&55a4@QAig%n$IHxfs7w}rkEE;%2km|CglJbH}bpPr1P1T38c)(!sL
zFX_84FFD_`bbKoztF;-fhzWdSrKFyUAd;Q>DBp67YtB(;$`LMA+B^!ftSjV#f&;BP
z-e%<8G7e13dm3{!nl*$ka_DJqni#0Y+Wk{jDD+-DA^GNPuq7|^$?+jO#CBI61OMUZ
ziaJpC)d6TB70dx+0utpdavH?JhVrFvs-1RU)Ftcmi@GX8+s+5p{4eR?Fv*Ga=u%&N
zPJ*BZpdI@dK)|*~ZO4!&SYqUHgqSi=d>;srrQfOR0)crW%oc@&H0_k8^z=~N!yrY5
zEL%5CyRCBAL{AIw>S<X2zWlUH?Gu@6Q(=nG3u#xxti?pQ8=vO?XlGcdrT)H6AV9XJ
z8cYCG%MvfvO{RM0OO{Hl=JPR+6ng1SVm{;$(tE<RoMF`tv*N>;T+bTZ>bW;2_72<?
zHdu%F%^tI1u9C&<9GR*ilkC*J3YwJUq8JGG)}_I0l`MKyxl!iYlN-Wm0+Nn0C^2i1
zRde`qrCZcxuP!v<&YA+yT5scC0WA>)Er#6zU4{m6R|}0E&jnM(6Rtt1Q^@v3jyAYX
zg~F!_nxw%?HO~{bl3^%o0^eIN<*S7Mwetoon!X>yUFIsC>seR7I;MRL=)mUYrKSN6
zV&CnWx}Hi$%JztIMyqJ)mj_8+d_wP2IDn)-hUvw?ER)Yu)Lta|KZ$gBKxqbXRDiRV
zY1Y|y-{byx;2?n@8UcekSv^L_o)IPDW+P}5TrkxA-|Wr%n*9*-iYQ#zs7nEcnBZ*C
zCa?>C)PrDnI(IQj>5iW6m4CPEDIoqyJ_~3Lh(D_o`>IyM5-Lgs|6Mr~){HxH@PpYG
z=yx8x(L0k+g_{+%;Z733+393m3|S~_v_Ch?pdf{HUeP~cO-1*}D=K@b*x3)D(vvI1
zLQ!x{vJ`Z1Ybdh^R558~znlp!514s1y&c*X-^i!omb~DHZJmjn+zg$CDsMfF3^iON
z1kiVD>@XP9maj3RP09b4Idl(=XmZ*5WhTvfs+E^NA>a2?P8>{Z7JuNOzbPre2{3mp
zQ$t;qq2#TseCvL-U7+}P`;Wg|@UA?K{0mGTS#c3`+7nvPiT7|uT$b}WKkc)=BbXsf
zcUn4}@JGu0vW{iaTk_f`R%h*aKBHToH(!Q+QtIZZAO`xz_;qApw4T|Z=?8Q>iRHJ$
zWoVkF4rRPb5F9FJ8VhJWypG@FJm5C>0cWyhBTr*;e%1-p6^U@k;>z$XM(iRN+15Fk
zLOAJ>#)vN9JikBC=s385tmckdxmst#f+_-B#U7$DZuEm+w(HPv$5qp+TBsybl#cr6
zbn2CdEXwU&XOI#HAi-};sSKHoj$;a1&%s&U3S<})SK<ekGb7YjV}x^Mt#e*CjA7hE
z{ZpZYDI5N!m1uQoA({u%v}DI2NSg+c#k?Vn!9fV=OFu%SFNoNpV1x~xc0W7py{y|2
z<3nwdo_T8eW}eRQa3QCJY%)T8fg59|e{Dr%8eOQ&N%Rcngjb{OiCwymuI{h=?F42c
zj|UGx0q%U5v{hq}7r)Qh{ZaNH`CpC~!`Y(SE`3VEPDUjxuN&&y%spw~tRCrOkQxde
z)XxF1i;E1#qLLSN+jC&p12tTk1DgzU9h```++A3!_-@!Jb71RCXG=E+eP}4c^Q9ci
zMOGvOcGy2#@s8@<GLXo{Jb;jGxce2`95RAaj3Pdsavzv6Y4bxFKSiEx&W(G|9-BkY
z65k6qvQ)48Q}!|JG*agjfzDYLUeFB%`B(r*xs$xL!vfltv>cnV8XXYs0Zx^T(KFN$
zGZ>u6ob2!GM%SjmZz-b(3g!}DR2{a`Sd9!o?N3mO#2kF+k~8xQ)X-GZhMRANemVk*
zHOJ~K5t=LyrZ}Vh6QdhH+L$S4es}IiFN_ulnw>x7Vp?r9W8VZ=5(AA<7XULas$ROF
zC^cJ7pCsVaD94}}#~+@i^3jmZ3ttGd6vkS3drI*!!dtQ4BGIz4dn$NG@Z4Mfv|{*t
z9)9zdj_uHmxO1r@Dk6-9JwwKvY1Q<MMu3$I@_fs8dW|BJLR)sIAmWTD|L@4JuPv|A
z3mCD9>>a6%z-aQ32Mddm9HTfv04Vp?n)sg=Pih|J+NPz?vpzAm#^#^=$owASTM2-m
zmqCnWOB~M!KHZ0w9$tkOZ0?;1H~hW)T0cnF=-AYZYT+$C!O5O%KBS7CzB4+@rRQ&1
zKa1;>)s`HR1p>l66{z3qTGEW#Z|Q~3W+<S*oE@CmBSB{+-4pL#;9;3`=CLIGsHbrm
ziN)bS>i0*YWBcj&-F1RV0=MgsMK6*>HH(-{vSSk#@3}^Sj}zMz&zLcs)u0)uixCuA
zTky|0IOT2j_7(uQ<J3zVO=&24QPbmf;m4SQaVoIzGJ?p`FaO^G^-g6giu4743xPNf
z35l+u7PV5^sL~}Kx;|BnESq+wwE~@zr9q0eca|r}&FSun;DVn^RzC!+*CPQZt`kcy
z5yT2qZ{XXfk};_!=dOEFKsFn_2+!Uwf_|Z-Ni{^4&EoQwG6v+vF7c6B=cCCZ{Y*YK
z%MJeA%$yjop6N7%wN+p)$G8O6lt;76rng1Pl*EN{@jCv9{qe>uel1Wd7TJD~;#=G4
z%q$Q(d`LE{cv4*LN@Zw<w{3FR*KZ3mUcBBp_Pe8;eUo24eovl84rnK;(<sjZZ;<lO
z0De1CB?VYf5d$lXbuf99r4_n4<yICrQa362{dMmg^Cefel1+S8Y(L!v<y*|#QW_B!
zt60AMm*3D(%u%mmYC5YGIoa$)Emu*8V+?fXN+JxO+Y)G!j2wqCEIE}Khc-hp!WcJB
zo2G>7)|?zWpngU=F&eB^)@;pmg1%fLSTT`1K4K-Em{iRm7}U(nh|I8hF{#NUeDm7<
z-}b~Y8MNxebAm();8QXB-Mxgzt+b81lIL^gwuk(L1OnGXSioOWPSOe-`i*{iXE%n+
zBtVhPJZ>@_i8kS^!_o1ss6?O^K}E|^!E$6L%<>6cejm~4G)H*ZI)sbV3}z5O5vm*c
z5kby^S^Tc>!f!PG3;8yZt!sMN-aq)7q2B+I>!LCka~M`VkG60Qtb7f4{4Xu!E~ZU6
zrrFH86-=;dOM$YlLu{k<Nrm;3M>oT5Rj4M1_92iO8&CKHgbv!S!;D=aW6Y9lplS7A
zJLn@7sk`K@lqfv2;yz@zRs4#d6{Z#UH!RSLO9XzhA)dLado=67OK8U7y&_=~gw!{+
z(QiS_Gg&aEM0y!tZ*$5-tbFY#bXN5)t?aFqtInLCh|Yiar(O@<Zmyh{k9>u6@`ecw
zF$6fJ<A1&~P9x<UA6?rOFx~3&#yfLp&1^DkQ{<HqGZMD|XSl+Bz6=p-6O_l>tMbTL
zHdzKnVEN^0%!bCXYR`p|2(n$^fRg|!W|7nGosco(ee^tyazsMzc6v~0NA;k7gS%bM
z{aT38%FZcVhUCis>%rg{evN4i6qxUevWUqExbRLmZe~%us}b#`dH7gk*(EemL@#3=
zkvjSZzUy?1WYbZ7PpD-cbmh7ZA;B~3ILj=%=y1<Ef|8f{yD@PgVA)LxGEjrs1YXVt
zqA;D@(FPv_btR~KmxkbQuy>x?V@d@&%||YK_obQlm>=I)f6l5Lzt&f(2n1mZ`Ne=|
z6Go8AAQZX$A4cmprvQ5f+essXH5F$v#zQ`KXX^&iYn`9AjEA=C5zI@;ZLnkIwXz<>
zVo|za{KB@CIaMtAYolUXur#JAS09eBYs>reM|VYD&%wACS%%PDg|{?;x&PKgS@C=x
ztOHbnwpk^-CrdbOS-HN^3;n{WjOZ!LnGy7l>GEK<$MJ*w_I!=c@JV}Y_~HU;F@ZFE
zKok?K_}oX)<}VP|Z%yX2OQtICH?niR6w&r~nc{`A9DJhWnT{I<HXDDn%n^~S<4>m!
zQ^Qa1O^q|R?{m$tZI)k~e2B27Y4e;GqB`uHEqNTPP4=AF1ge=`vpm|3USS-3iC)IH
zok<y%>KV5T0!EFP%uqf>ox~Yal;B^{oFI!u%sM1e?hFu?!(AlxEF51ON7`!}_ozko
zEUORO9{&Xko`1k3mPrSkCcPY;V2R`9Hn=;mHt7;aMd=@XLNH*)J`TwR#Id|?u+mD!
zVsh)IhBDq*KVx`8ubw3w#hXQ~Wc0SC>mC}1sEp8IDgGjb@!4XW^hPThDmxbww$SA#
zD*TjE*r&y#`Dh+FH3;yhrAjIeqEsT6JF-%)4TVn+wJ4o&F;?}|9HgUMnvFJliFrro
z0$#!XA<z}SMZEJ_Nizl!#XgsyoKMvkgULj!+!UWA%5-I$0$UN{OyCAj<}W;km;Eo>
zBz-OccrMEJR*bS!$G%TUA+!NkPj9X<s>3@{u#yg`=m$Pt^S1v~##tBXJIQ;IeO&S6
z!W$!NbOu&UUbKZ}EH7%VN4`o+N8C`CSyn=ahAj;=U-oFn51jaB1gn4C(4ilZ$I-!L
z-*eJoTV1>jYA}arN(U1Dbg^~wQ{7+Z?d5iMIMWJ(gB0T~BXgD7Fj>wTFrW%oS;-*-
zk^g~Ms}m9OAnSSZ>r?x+(x0{Ib=P=hzeXE*!M>ZSC>}IAjPmVDclt3K5>uHF8p73)
zMpkp*g1uw+I@3BBx`=pP`bR{K{_4y}ILVN5g(&d5tc3L+6^F7)JGLfLgPN|}7IYQQ
zcz51Xo{Nn0#HRTc{awEt8Rbx|3^#iN6d@xhm0Gb`1N((<w-ijC;|zNyfUYF89mfcA
zFRU<n^E%u3?;*8-Xh!?(5jS`c5p9=B{WxVX3E9^QS->|QcOiEj?(BiL=v5YneDt~&
z3^@>?PKXuK%-{WfE(*I|c~%1dPu*(<+eoRI*SMXw3D0l_#28^nn0x&p3$^52x`g0$
z0ZN$R4NC4@P>`d%tkEc8_RK2YAdk6#JEMVxuBy|$!z)S#_H2u?D4WnlY-=Aslli&w
zAMKL97Q=cxIfbB+u#~n%2pC@lIccJZ@YUs*n6i-+spkBG|5W((mkmAq3}j`fDp9H%
zM3T{%ZEh1Bg&yuw;D+(f$<WX<@nj&Rk!u(D>^#euxu*jkb+(=j`_sVDbAY<aZCrid
zdnmS~t7tZ}2rzlpd)iBlSDdIj3S&Isf`M(|xJX43!0BALS@8?1oo7ljo*90;IYNdo
z2BOk@=YvXUE3}Kw>exL{*(na*y;M(@4s)ocdh8Q4Ii)S?wXh_9i3t%Mc3SAnL*j98
z9w>nI8<*{)gQDCmcGbLv34A)6>earO#AKY|V@OHFHIcZheqgNxn$6e~z{<ETUGS=2
zf#h$CcR<iUOU~RaF(aQh@S7HeTLop|C+hvBZdA{z7kzkCZrFDHcIxBR;?7csflh&E
zaqcR~G5G#`!t}P7oP<N?_}pgPmFUt<HdNCflJ6?YR9hC!W#}NcCrxSQrqxpP*ff0?
zA;$C=!$<o#a)3Hfs~rF(ntF_*l=G=X?HvVji-L0-VoR^nL}{PJs<YZNV4e)rQ^|y*
zz+S=JQF!wcIy}Priw^&2r~s$@i1a0yNd;!jwvf?Kb6oeXC{sut53oMuWaj?5#s|8u
z%kErgFq;xbH<5%cu(?<y-x@2T?7moHKi0m6TqROW?+iGdAruRSgk*K%05{LQQ5U`J
zHm?YL9uUzzmnneP!HOO{#Nc>RUw_gq6rIGK+!LU-JfDPHHee<qxTJO}ra#)xV$$37
znnAj`hndv4VEQJ`>`U!wOte)pgWG^jT&ExJ+*toq@Cp|m-?eqpeeI2&*y~$oZ8=+g
ztCT>l$~EL{hKI1d!=ULs=DtDmCfqf)9z92JsB|5=<GQC&)Z10JCGBs`2l9{dCR2$x
zd<9G|#b))}P<_?_##cEw8mHLP0;|8m8pUr8Hq!Opf&6gtr`cD?DkF9K8LH22Z^vIs
zgu#(}p<lD*cB(r1%p`67iVT0Ex(kz^B=J|HpsWbe<zGV0_C*klpDcY+`^V0tQl19Z
z8WxTPHV5-G7tjK|*0VB39;tqW;XXZar|z42PtR*7Qo2@=(koSwE<^~-8I}q77vBXu
zX2pLV*23UShXd2Aa+237q(`I2aH##<o08zFKXy}Y5J^ro!;kMWl{uQA{QWVr{rkyf
z!1cIJ)WQuodMMYk#+iG9FMJo~YW8VsrWwk^Ls$%+fF@psY2%q+y)eXiP9>r&pKMYP
zBK#^F|9Ak_U;S9Y2~-Bkxfj|4ZQHD>;`0zJpxb!YN}IPE-?HdhZRYQEhHL%X2V%r{
z<s6w}yI0SpnN+0_x*mbsZ|~a?ucfNXwb|;D%MZR4bd@^loHI=RT?iZU<smT*Vx@>f
zu+_9~a*xIu5=}|A`82BTFdu15c#aR)aeb<9Ol#xarOwFLIw_-`TTWfSJEc<`w3zhV
zUxC~HvgQOEbWxl8Q)1|#lh;FsJM*YN9i`)i<#WmjUcOnqa%*tIVAAHbszT;dij6ss
z{n^lgtP*v<uLEFPi^_N!NbHI?sD{zh-$SA7tYubhBGANzB@ze8oS~JN7drziK)>jg
z5_AgDYX*+xe&ywu-7Jc7uQS3aqhQLnI7Sy0@&>xj_!mMD6!ys@FnyUW`q%p5H)mTC
z*epNOH+QpSSfZVh|2Zj=H=f8v3nbb<?P6^m3^`E7b$csaQv;yQ{Aoh_3E;IMkLQj8
zc}*!%!e67~v!L#CmDR*JCPHJcGz*Bb`^V;W<=uL})!?8s)oOt3jl`=9vhVKHx7d}t
zjanJ$Q(=^a?E>9(YQQi0^6<>PT6fV*hO8s55g4_n_nk*%!~UD>Q+2HtHvF|QapF_!
zaw+oel_LDBE3&>)Jl6B9I2XB+A&9dH3!<#-ceaZs`kH?R@q2~z=rR?*BNX7YTY$#<
ztaacgjj<^Vq}cHM%?Xc3Ms3fn%$TsGwiA5%>YSk?iB-8ZY)UtvA!6)Gl}}L*2o3C=
zC-%GB-w=`}*dd60bABB0(nBFI!$q_!MMveX4g&k+u~&?7nx>>d4(<Td9ZI5sw;)7j
zfm|h4@&+5_D0cofMD0|x#x$axb&n#8MPK{6te5B`lq~prtHL7{;!qlHCDip=#47}f
zhc&%V_)hAT_QvQdnH_mvP?KGNS&}dx(9bO2$F+Mx9&MeN)18Qr0q;$gUcZEzKgCO6
zjwobn@C`$f<_v02-dB4ib1bgq4Qv!&(O#1hU3WAwWe5H07D8FI32&?i?GGps$57Iv
z^A1^R2BUdb9DUAo{oA$uS#mQUNnuXjf*1w{lMvUPvc1#cTa}4qRuk0Fp*E&or*2vG
zFlR2(cw><5$V)apJH__6vk@cQnh7KI9%d}7Ci}P-!y{}+bKZOBit^%PudDLi`VNsu
zXe$j?I7@q<Eun5R54Sb-bh=PWmWa;e19yof!)V(Ge1wBOE_H4M>*{|nusUuku##5;
z46jR`TC5i-V$i3frg|lMBklX=E4mbu>T1w`6qgU&l;EFcnF;Lcm(YKr4)v^rfC_*H
zK){IbxmyW>>yh6@Tj=l0wU|pFD(eSJ^K%vEXVps=+$S%Q?h`>%Q3#d-2g2pmmj1xH
zj#w~A;54c<1wvfQ(F@w@4g~9FS<j4(rE~`EwSpML6_K?*{wIW{-TlXRQj~B>RxIrV
z-?Ak7Donn%jID`^8rcQsKvH^a5QIf+(ocz1YP3M=pL~gjl$KJ!?|7H=yR2-Orn2+R
zf->n#Zyp2i>mdP<s5KK_UEJ^n2SYbwLyfvIGeMsUw4cI{(TYf&!I7V_8TuO2_7lh4
zH8r6rJk=hdN2NdCLLB-cP4sKCvE0bH3_Nitv!)D2EPXJ!dSfba68o8LUEv0Jd3ObN
zuPb%V3${~!R({NH*>a_gwkWSdi@+QtVaR77F<yNUuPDU0dz8fCecn=2tlApwf!lZf
zxXMtY;aEOQ`mWGsRm*+VBQ!J)NI!?5N@NrzoZE##8Nk|Sd5r~3r3gq9cQ{&6M;^b`
zV!r`cUX18)`HA%{KiNOnKhP6Iy~<@T7P#NDkpt!H;Yd{(uEyOm8<W+lofr3z*d?R4
ztV(h#gt!D)C!1f4n-Qc^ykgbo9fASqZN6>Rr=WT20keyd=i%ZPttONgcT-uGw>8kS
zYn{;M%5GWL{US0u{37t~AH#Vgo)hw*O;tK6mC?WL)?!EIo+XIgt)ua=Zgh80C8<rN
zD?vv|l4mL76Ie5~=A6<t0o!zVB@0zH*Q->_d#;2&tA|Z^2F5O2V25f+VD&GLYQ-3`
zpu_4n8z_MLT)APvCI$Ej{Ht)u5;r~w_N{XJ#Pc2a1Vr4V-u5of0hOmrywgh+x_u$b
z9a%FKTjgV@;g?eHjF>CGxFOld3;}3o7PU4(2hPe^_k$*e$`B6}aYjbC95!p;B6%Hm
zV@C|L3bNgF_iws*S7_sUlQ9njLOa)xb|<h2(3$w8{$GQMc*ocuuVBP8+{DLZm>W8g
zjpZl;pcRkMF$6C<!Vf{wdhCFbNRtrT|I9j%(DIS#;&;VD9e+CbZm7&x)7wrmr@%C>
z5h<^~uCAP3T%o!Z6F7%-*dx!pwHzT)B)H{c23Hxeg>}`^Z1dSO&JyeAF>bnZ5BYEZ
z;}Ue32+t<REqL)MhE!BkUU4^Di&0E%7>IwVz%8x{JuVNbh)ft>nDNsx#RF=9nU7fs
z>XanybO-Z^zmLJ^^=0T&`Eid7!zLSXg!0=WZ~@jGG{e3y*Fgg3ign8X{N#Z%xB~?L
z$q3#&4S=%6Ut*scd<9S}^cr!g!e>zYNaVABB*n&%QF%^c%rHQR(FZv-vFHrvF5JT$
zy|wx?AcYx31hfyGjRHB1p(TMUq(kn}F$gV@fwsCE$NLFY;{jTq!kRp9=l?knRSQEy
z0n=Q3Fr|SCNdGlgACWqz12a73O<)l#bOW~&q?8uqGUdU-Gb}7d=zwfRhg%$(Wemw6
zLw|1(haf3IzW8Mrr1V8A+pf&r9&7S&>{fQNUP}2#IvY7J>PNaYRreF&0Uofk<;L?9
z7(ohN6BlEmhGl8HSQA{YE`$dmIb&H`*nY$EKq1p`?zz;aA0fz2bsBWpl2sRiKknUE
zWqgui2oOVQ(;~9JsXvs9J2d@W5DMS~4tHbODHfWj2(Z5oc}SQ`jZ;7GLDLInoJR%H
z!?C<^F^Z<oV7gT|8zKS+KBgS*ULl5PY2RD!d^n*9HWW7$XZ-V&qzWe7#kxdUcVOwu
zTi@lhKh_vl6QV#wa68%B>P<E^-;C~d)#g9CakD#buS@Mp_i|Dz=eXA`gLCfsz7$ou
zoFV^4Q}U*M#rM#gkC^`P5{$0C)cK5}Y73xQ0MUs9Lx+l$?Ul$P)BLN_N6e4DtG1P@
z-_;mut4LPcN}~2WU)V7<G}~+3y1*YJXu6Hm)6h{_XfQ>E)d=IQi4lXuu{SWoj|%sv
zB0ZfQPVo=6Qa3sgffil+@2G*=B~kReq+$2-T|M}p3nO#_7a~VaI5*Cku?N{bwwRi-
z2fMzkx3yf=i{bxlcRI61ax9XMGXfhH{fLh?=i_Dm{G9%JuHzQ%)z{HiSw2VYKtBA8
zOF#>j8NfavFr}VLrht^p5g@SbYX`yRu@lQW9jVIA$ZG@jHS+O>m{m(=Tp8jP&GcGN
zG>4zRZ{<Kqn8u+W1ltQ*@!frKTNZ;GXwSUMUK=cGXHd?tT<y!_bntiQ_0%opAwDS!
zxPBO2&D$$F>4zl>IONnf2^*?6H?TrCw<hbfa0T@OuukQOy$0RXyL`EgqR8W2OxP(Q
zj{?@XC-HnX|Gv%^wM+y7%LJjvv98*^FLkz5#Xve!Bp-J<;#JGgNHY5To|?c3OJBt5
z=!a{G0kAPYihT@6@xul7mgrRz38TaV@I~l`bb@NW@j5qJxuoEem?>o`a{*dosVyw0
zJ51>a`+9)@UOF-Mu=%4nnIrS8s%PFuCubUqS~O}7L6l$RmVIUw@@_onjpn@4rgOo$
z+#4Y7eYbn!rHka{K2mz3moV3H%(GW?#Z0PSi;R=}p~L+T@uhn#IBU%J>T0lMdY-k@
zz^RW<t8%il*J(E_XPW@xI&{4wz+8HKsGX>xJH^2#B}MdVpnR!U=gD_)P}_VYNS)<q
zHon7;?!G)3m7&P27R^0-)}jRAFQ{hhPw>==)Qe>lnj)H4R`C4N)A5t4r{sb=<F2Si
zzH5c+{Ii2{YTg!zDm(`pikWt$gq~04i;bd)>I*eG%Y;VW#Re%Yt5R>`FB~r(Ch;HU
zSTBWco>er@$#og=#>rGzxn2ag@lEDy<ZB!;Y1Fpxy>}0)<HAM^IYM;?a^AZu6DmCe
zj8|WLFN{1m+)OoYGttXhtJ@40Tu3lL4vsI;V*YSO<Pj(R5qWd{TL+Fio$kP>7bGsa
zBk_zdS@Xh?PjlJA$-e0P0_=&K2!-@t>Io0_G)QByp^bUs1O*2Fl<d^PS1O;mUUU|F
z`EoL5c~hyi0`l%m3?wU(YWQ!Rx31-fmL@^lS&K9G6GUovO_Q7XGfzgBCJ9IrgFfRQ
z6lmj0Yccv}6gV>La+E}oJRp#HOUYLb(Qu*;=2t!SF(A+ck(|c8?D)E*0g%$CIj`95
zcn{rhIjVqB7quvIE};MUO?xpNRvCwaTEG0Uh=jCOrRaNTo|~ib6_c!INFPRg9#AaO
zA8hy6Vx_uwj0jHpK8cfyX!I9+oIIVB7ky+pxbddC1Ns~>x*V&1@8>mO(!<1)q`kTn
z<#`DOSck)o^KapvxnKnc2O}?NWfOj3F+9S^c$J^T9J}>SPX(a}^bIslFT5I5x>Sa1
zSOHxdHTjSZ3X;zC4l6GzGE_9f_9+{p>YXg49|&ditK%LutmqO$!e@nQa=257<`VQ0
zLvXl5W;B?#8xX&?Nga4zZ?#?0dyMp*<?1d-2{m34Aoz)j6!BMwiD+dkZ0etNGllUR
zo$-eZR$~f~p%f03pGMTsigS!%#|dRh*QT#1&A<x_YsX-rL}(hrcTqqVi!mm#GvdVe
zbarx)Bno{Ocku5%KS!_3x$NY#SlRIREQdD0_BKTk5WjyQzkX}Rb~fuIV4js>dSq~3
zDDVz8d<~ng2Z^6G$DeMg6v=+N(Q{oW@08Qj+4FCu_rkf9>9~&LSlVFb=Q5l)lYK6&
z{;)S>!!|wA#~+7eCP32uc(+UE-J%@hvM42Wla_l-i@*+K@KR}$s1Qc)J7HDcPCbVF
z$<HB}_Dt-rg~fq=6WbF5Vur!6F9@#Dd))uS3L#{21%Gz7M-w5hF2j^B#}x7mkq(X-
z<MSk7`X2TFL_gQ`cQ9>IZ(4JdId|j@a1{|qK{UUt{Pu42i90i8TS81*+1^Dk3A;3>
z3M<A8MC$LI$1N`h*x*w)J$7SNC9}T_l8ZM_lUOkhBpvxnp-c%&i$W&LbeIOhl59t3
z$Wk@wDf7j4l()eAT8;^bkt|~c3+8Emt~0V6V$vS1t5OUF@vg?h4SyJu5W(vX3NSwn
zEjOKrs1{5~tD$Nu`9h3?lpqT$luIM)e=wfI+|BN_J1_^}ijkEUttKMwJLTfd9Lu|D
zyiW(?w0r(){cSw#<Gh4D9NxOLQ@p)!m0Ns0bQe#$AHssscFIeAN{w!%gRVm8+*^~s
z9hM^h#kQgCmqlE;ntCD+(%*7GPk$=Cgj>^<=BG_$e8M;Vk1u;H5Lgcbu=|X-=BA&B
zmBV14QOcErw(3ekHgK#8=Ob|LV^TXGx0pEhF?D=E1svdB*a}j5Tq(4rvALl3DiIE;
z_8_1)tpYu{8_5NA?q)}!U?eH}FU&Zaqx+8H@e;#HqYIetqQ<~K_-A;T(DL_RN^t8;
z%K0}T4>*O3J8i1x&B(Dhx%3rv=&;5m((0)uoRWIX%9dU(g)4or337fOQq3)(+4XJ2
zE!kb!(iFZh4)sEbME8K4u4%{cZn^v|pZo{ODF2XK#^qpH(tyYUEWT_<XvleArVDK|
zhmI@^G0M`G9g<Ai%9x!q<dllJg`N?7nZ#qdr<33d$9XP$gm|wI+705&%fBB8i*9a0
z8FX1^BNSe@Xv^P$L-k)yP%Rrn8ZC3Ec508en~2Ryp%?(OL?=Wx&s%Y6mi|%o6yP#K
z`9M=#HvpBs<k!>b09<U)v$Cpa)m|>(CJcpXuA90<rycNGxEr1eN^qq&eS9wxRoe}C
zLPQd<G$tcSXv1TcN;JGyXm)x7_>WOV!6|#6qrlvQoj7xnT+>k;`Q~6NvXUYk3pCq~
zsFiG@922Szw$!hNeirs>85Z?VFOVA$@LPxX#m4-3$F)}kWvU<PVJrT%U!`oPEoo_!
zc&3z7k3c!fLv5*%XwS;(1Pzz(&7!KlWG&;$O-H1WG&h}nm+E7OSH&cC3D2E&CUcU_
zU-vwGFYQGXQ8V5P8zO9R-(^b9FY-30l&Lc&11xq>GAO3XJ#5G_fDvwe^r7dQw+P(n
zT67Gzfy<>d=ShB;9b$09=`(G&E4TPbamrrw#I~t&@~QgvHHeTciX+P0f$Riy65Mud
z-Ko;KkEma=3t71Uf(%|EZEIA_ypa67r<2)1m)1@xzV)ZvQ2=Vt1B^^;YSP|66Uy71
z-17_OPJhz_S;lp7@xIbrBG;=ZTloMC?6}QL3K=#EB0(Kn0z)o~Z19Md4HvZYNCV%_
zuA_N=IpSs4<YaREW&RT#L%cGTfj`}2n=gY5Hd;`izcXSfLw@=5)4nbzcyLn<Aqqz#
zh<Y|l5lp!ZED^P>cc~Z>WPpjsLS@G2g}F|)6+(BziIt|=7pTjPQOQdA!mU?q*inM}
z<n)nW15;#x6b?n&6T<n`nT6_yfT(4bWbeHle1lKFOysV#4;vi>mwvPJzS>KxzH9Lk
znUgA*#+_|Sr^S9^tqUC8dUk}eC8ZX!ex8Fh2U{Hl@&CXlCW|Z1;%11#ftknFGQoN0
zUxEfN#b`Dk;QTIzZ$Gd*h+CAfbEc5UB$qoI8Ye{tm}eR8U#4hS+1!m+m$g=Ut@o0?
zHrn~U!dSAo2S(Oza!brAZOuGBEg`9REU%je4(4!m#*mwlaNWqHK4NO%#;?f;@jqyL
zGeIFEelt1ppd66IrO-g|{+hhS>I_-H|0mM}Ec`%8VG842*o8O8K(wr8-MUMIB7#Yb
zhCPy-S#=Q`f5!o2IJ*8pgLHX|k<mry@KmHSyQw_ZNcen9ItB<m;J~58l?G)$lGAo0
z*1KSO8WwI@WwlxzA6i#h0UghzgX+;Eziq^A!i^`GjtwWs9P7Ir>B*8)&HD0*CiF8q
z0ZR|=!dN+by+V<b9m?C0bb<t&U5$%m21V-Zi}{;LDayiUN59aij3f4#7+^C1%j&`|
zw7$Sj2&(DTR@vVEi+!t*Q>RlR{jLe-PMjl)irdQNPUabr^iJI(c-PkIuhE`vsbWbm
zR^Q;i$yI4UZ*%`1L-7M1@FCu4pcui>E5$5q$Q!5xQrf{|W2!s(M`A#yjT4G%^DoE0
zt)Z@w43Q3fpSdOy0J@X8rha(=<T-WrvFezH43#F+s!f!RXCAaBU~_})*)gHS8M^0D
zoZC&w_e=1>YL39>5`0S@?6R>(vdw3;cZqWNOfEI<Gl0%&L8d1D=Zow0{MxImK}46>
zDG3$7EZ^qi(+RTGHm|jQe)=NbOPFVeA!?3XJEO)7h(Iw!XlTfYECVY<^Kb3bEMF)z
z9pyn6R%;60a9YaCv36pD?DtH8l{p76nH3L%5}#YCQpKYYG`y2g8rZ%b?y1H>p|{LZ
zo9&%oh%;Q|2yVweGs7T_&ysu)zH-{|!cCrC{uBofFU<5H3O~3s-3KCMUS#*goY>Ji
zmffNwHd?t=1M+)m`U+-_u+8rPD!3zs4sbL*E{VK!_0HQUPv3@cr^|R~+uZd9q#6mE
z{T+wr)Neb8R<{1nt-I`#Wm}f3Yd&C(IEPC(A*7Dx8~2r96G~xQwg2w9pq?+pqmJ+`
zmqdFvQ2oI<<83G{MRhkK{qWZ+FBzm1KIuP-DondPSkWe%_?uWrrtTCu<xb80;Wapw
zlvMk{Y3i}HiW;1~4^VyxO)~%z&IMWB9UG)Fcc-HO^7v=2%0=_28X&n6L*wx-5T^YU
zo3{oUot>%DBeiGGj1_$*h(ie$ej)&~IZ64b*7+C^c6KY+b-1Cl6-bQeie^1)d)1w*
z??1YF%NQpY$iyn68VLE<BwzLzRV>XrKMc-*jDd%&XEXVsD#4P&&t}D=ZOfwK6r}5}
z)ifyy9ekJmOP3dYMc8>zwZs~snz&%zncvhcOI#nFRWmb|;m<aR)fUIC2A!StBLtgh
zJ{I`_)z;eHf_ZbUDXEW24+N+^mwk4GJXJQt-nkcxrFrMrRLD*Ny@cTIBJ(f#s0QvP
zT$~Ix2j`6>`U#@PLWZ0=!Lc2<FCqjepbBMfWOH<KWnpa!Wo~3|VrmLBF)}zHFd%PY
zY6?6&3NK7$ZfA68F(5KEF)<1+Ol59obZ9alF)=VVGYT(EWo~D5Xfq%%3NK7$ZfA68
zATcsCHZ>q1ARr(LFGgu>bY*fNFGg%(bY(<kV{c?-3O+sxb98cLVQmU{+KrY4RFqv8
z@9FLmqy~`g?rx;JVHgG&Vt`@j?(R}LB}5vPZc&gFkdly)6p#|Bi*M@t-L>vrvlcVY
z{_VZbf1k7Wd0?c|*5j15b+rL2xkBNbyj(nD0C^2XV_qHr4-Y>V4-X#}Bcnb9?hO9h
zj>TvQhIv3-p<=&y$iu)u_`OX52)}pKaD@WYJe>i&`~Y4dF<xOY9v%Q650B{Ij;=5<
zfCA79VhhmV0;st{!5&zQ@~&>aFo?Ya{C=Ike*#!QtN>n7QDKfh!U3`_U>F1hgaR~x
za0jr<{fZ!<GeFN31OdZ+e@nq4;Q)ubiE(p#dwX*MT|Bs4VfIq28~|?!+yS5q_5j1Y
zz_x&&B?B~pF5tgP<HBMD=sQ3>{selic5rVX3=Ft8I72{SsK<SXC)5@U1Kh6;&{I_h
zXt{x*e>$uG>A(T_YjXg+T)h9J`z!iqA_(-4U?2$O>f#23`a+=g06T~?7@(!3&IR{@
za{z!)+n<g=XAjr=c%T;$;taI8clbj(5TGQh0|4Gv_*Z=%AQ;3A?!n~&asF8&_s=x<
z+pGw+m3MV<0Yl*)SU>eCKwx0d{l5Eh|8=-dP*-oL|KCq`5U8!)&oXR1-M9^)5O+_o
zs={AB_a>}=ZT4U|K#)h2PgqC@0Cop}eLxP}KNryVbp!tad4HPkYY6mrb9DpQ-IoCl
zgxG=azp(s0fL>q#9OelQ^#2w7{)ENL3$TTN-~b!2Jp_vNZ}xjL*zV8g`}2oEd;n%V
z_wC0E;Q9Ia=a<EOhuOM9oqhjx|Ko_cRh4w~<xScDD)}EMC+F$|@aN<c0C4h&3h@AV
z1q1~E!oqxjz~2*T10jD2`6XBtYUc_N{ZsG#M*XeX>#rtY`D;R00l(+cbiHp~Fo5OX
zPB-Tf<N@9P;r;*a_di1ZZ*BfF%m1^}|EoyJ)7kkC&hn?j|8SrS#M$>RkNej3gx`;W
zhU@(>K>zD%1pae&HNdtIPnZ8vRXFf|9Au&P&i`yA#6t<<1Gd$Mz(EdwcFdo4gP*hJ
z41t2RT|FQ_Zx;Y3FAvXu*!_TkobGQBkNdX#1A*`7=l3NQp&(b=pM%CHC<Fk)U_f82
z`(wU;5(N13-cO`0*yoQ%25@shUE%j3fcx?S0d}r1te<Df%f|=cwt)dbU<WwF&GvpK
zoWXYRpFV$K{{P@VPxN0d@C)~J`%kdoZ}2~%_kH>&{TU<&vHyn%{0IC?1Vw%!Hvi@l
z`ZdDk{_6cFK<F0$cl%{=-oL;fqJCYSm*=mNoWSt^D&zZ&*!)U({)>K5{hzRZbqoF;
z^{YT0!GF;INaX(={icKWf7mZV{u}s<lK;g0BIRd?*g3np!v3eQpUv?H`QN2P?<>2%
zQo!~wptCK+!_67!`)_UchvIhs?t%Vyd-?f(2mcbDU%;>B@{0Tp{v|>Jzmi{C<=4gU
zXXBry`!&<A2O<2I{eMq^|3BmxS-+Cs^ost6{32NJH}s3*-xT~R`L|@hb@#nA7|{0T
zjrv<_^9ufI{}u541AdeH^Qiu||2n!N|CoPUR`?(D{Y&CksXz03fqmfj_m>CckL&R#
z`RyhHdBW~Xf&cNgxp({f`H%k}z+fLR2y1@U6(k<v*cj3Ay<V2mn{#ta@&n_h(Gylq
z|9M!a=MOyeOxC)*uoc+3Z05T@f~9pumNS_J+8h79)@JnRwsf8LtAJ~(RNcwVcC5Jx
zvNz*pd$LV~)R<J9`Z8MqH|_z3p-#xHh}~+8Med#=c-kez-@ONveVPWV7vD$CZt85+
z2|dQXuAbsdG)OQHt(aykw#lC%r$>WxQe&_Y9r@r-pPhdoDjdI~RZC^Z3fxcRFZVZH
z<x9Gp@%vPx&*#xY@rc5doErIz==~&<zub0~+Jle&mF3H&({xF8Zg<MkX6O2$a3{|^
zW<fG79Blyaw^1{$TDf;}nfr3CJQ<_7?+slsj+DhiOTT#+?opoA$ZOa>MyEoKVlp?;
z_h;~`!F72D=s;5x6mC-vS}De1592o*ERW0)PT(KRClh1x)GW4;+rU%<6JnQ*-FjX1
zu@HTZK@wI&P{mWy;zo5cWtB&I<`kOl_xvfFaJjLkmv_zpZbHR;)OUNP+4vfR>#rp2
z3NMUg%Li9irkiLEs5gdYwDWz$OBpR-`GhiMtBB}#Nri;P%IKN3t04!QL3TgX1S0lC
z<}UrcGL$q<OTGtx|NM$RH>l<dYTB-Mj#tg+q3P&bQr?!m+krh*1OyxxHDjlTm`*53
zBr`jMg1n>(1&J}s<ZsMIj9lHE*DYBN4#i2Qyu;QcjBd1ymk2N`$=pIrasUx%i)7x#
zLm$^<(vd;gRm`fSlLr|;x`|4BUxhXdvMyVDs+JPUW=^8pyY4fhpFy)WlNlJV$ywF4
z>uUMPVqV)+SnZQrWJ3)(rhi-zQ-yarnoX0cG)^}eMrx!v+K7~oY6Z^)!Ou+PHgj0M
zZ$s9~?eaJ~LR-wVF_p_$dqciUqF7-aCa(JFPIhfE57RozO5vF+q0OF-sQO)~L{`=N
zKFTh61VhDXrkv;EZt!s&72(e%KtEgjx+WFwC-rcv6JPW~b2871bK#p;=_>2BzJ93!
z8+E&?dSKy8cdGi&m5Vk$AEhqNsXg%mJ5y?zcErZ^B%Lj0X!f5@x{AGe)f~NBrkEc8
z0<u_fV3sUvnKgcMjvF&%md%a7si-?D{`OdRCs8~DkN%Ai%V1U0V@hdxx=|@{dk5C8
z860I`>mjBnRHuseQ;$TL%!93?sC*V;T8Rb3n8aMl0QM->z!~i~Bhe&$OjQlGG+FL0
zpz9TebyYlv#+Tdun@Z<k)ho^zy(Sduf_a<>TIj}?_W}OpAleu^U#~;~p4zLvJqs_b
zXzAlm@uo!MvBR1Ss$XlF5zR%Hyx(2*M{3t1yCt98qPAJ@c0yk4qp)o7jcB4$Ao|&r
zmwc;BknI>wHIq&az8t;_%ntq_thvZK$O%7VEwg;CsLEowl%`>ZzXF)*e7&b{S83L@
z*G)O%_i4w!XbBreAV14j+ih|J`Yb<KHiTL}SVQvy>C$fV=E0@c2=s;iS9^Uq`*};d
zwg@efGNT&oNP;f@$Di!KG(kL%w&&8pQk3VHnyry3A2Yub)jXrUaIfT@)8iVMEn!dY
z;CUMxEU|V`Ve#_4S3&Kp3%cpW*YcuwIJtGSSCXAZh7Kn9OuD!JFX;G!7neG=nfEkd
zLe%dXl5G-;oXV`i$rq|7*D-eLZ^;TD^i(|@UE5>sk%^(;eM7p&-{OyVr1AJuN{H3@
zmDT)!GVqgrw=A*^;&9w3k7=cx|3+CcuI=n*CmNsHc~<R))D=Kj_X9@k*y9@Egf)~W
zACprsCg&dX>Jx~+t>4~%!EPpyB6q@uK$Y2z;Vuv)E?m^401!luG?GTryJ%<n$e&p`
zxqNJ<zIkx@{QSt0*+1c0o7M1rb5OAyM=mdqsRgzERpN|19|>4oQs#s3GW18C5LN(1
z;n@t$5{DvTXi>|LF!teg#<~+LJUQI4HenW4?5L|qAJWG{PcyFF`}%cmj)<K<i9M*r
zeD==T{b1Vn5K?$zrB&L6BD_QLS>rKXw_rVs$1b^@#IPa{M!Smlvgw|beRIXYs0vy+
znGP?lX1plfrnm^U%)-i3v5@f^rd`ft7KLygo>xbeM|Mu~)aWE;p@p?sOFk3{pO__m
zqI!*jRlZeWLxQk8@G9bnNN;Mqqc0N^OPjRHIztS1YhxlIxzlP<>Z{}*elVtGQuxpK
zetLqWgg|jkFNRe+$7{B9+kcpkZ-xt<^@s!)tXXg>Y;t9Ndy)r&Gg)QH4?Zeb&sVij
zn0xZ(c{OuMazyIb;kklxv>w`k&ZDUiH$=+nnTT$5A&I*$-|K_WkLVS4y6tBQcu7{t
zT&GdE*+oujL)WC1O!7O!kRBT6vDc1I?AE_HpNi~??dlb?d+^+tP@{KgAFnu!g{UG3
zjuzyTFtB|Js${RWRsp4Mc#+WOu}>h=ZP~9UAyYdOkc%@qZI89$GK3}Mc<s|;hP`H7
z)Tru1kxdhtjnecFj(p45mT8elu5i_8ExskCy4TUPRr+Sr{v};rvAiB+d#?Y9tKGzq
z{ew7Xa!)Z={vzKjYJ;OsB_EH>frY4Kz2d0*%;~GKcF$$ja-Vw5eXfbSHNB~G#MYE+
zv<NRJIpXqt4LLL7@4vzn^p8n+`-mQaSe$a_qsg_!^CascItcHv|8--7X+%8~Ki~wR
zCc;pU!>dkcY49`G5YXjOz&azzywv2>=L1&#VEOJZS<_DRrK$Y1B<rq?GHkjAA32g{
zg<>`&C>aF$1Td@<d=%as4IsLTDbB(dY5X-Jgc!u4v!iNrEWE!g7o6qF%T^a;GCNun
zOI}?)7KlMTrT(fY+b>zxFeWylp*0PBGsSLb{LXCsnGMG$Y8T)<TWcJ7+Nh9dwrO0w
z#c)}BM@)D?oi)wyxHJh_JlyM8X}iRGZVbl9`euTfHkM|VdxLWwS<9<I5P6WWSH?HE
zDn3jUV9~}fafkSE^6ljbqEAqiKf)m$=18gEI|);SGuI?8xXqB9UE9MWotLXY1;c$M
z^^8+Bx-kLfIC7B3Y?_QmL3`AA&;6~Ym~7??>|hd98!)7>N0dN2-IVm9fgg<vSaE2@
z1>K?mENlkaq}Q#dunA|Qy%4gFHX|Mi)En`T`m2H_X_Z9H2e15BhYq%o9ViRmsAexQ
zI76Bbw|$$^AIxJsx`Pp~NHoJKa^oPzh`|1$gK-ixzCwwcrmQsZ&|*$l$Y^Y`>PF-w
zZq@2X(u9Of3pDBl^7>v5WTle}v3ix_xwKhl6I`V%6}{lEq!6mNsSo3ND8w?ANleyH
zZz9^hDK{bd9U#Q|8<`^L@tSQ6yu>u7uOy7nUl=BB`5|W8{knk;+^SI!_VJ)@xXw!*
zMIZYa1yN?VIpIb~3WYB>!p_pkuA%#^tC626=yQ-@Al>#;eX_RJCe0R148^qSq#0~O
zx2Gw4V=gDBoEw;HBSo3fiIf`;6r-D`6JCg>%#bzc@T@1%5Cw8@m(XW47%fdW+jQWR
z6%yI$Qi$Xai%=LQNR#Son2Q@&^=_(X2M_LJJ6gETN7_@D48YL7j!&_b1$^)BaKy~9
z&%g;4FNCnO_}Z-!e|0cbK)^vbvJGv`YU4SgKiF!=@hr`^<w*6=UZ}KVwf`y;9fqW=
z_1Nz$6Dwq6tEUxHuf*~w<7_I8gI0IOE@l7atzw^fZtYz$1hJmptI8mIL|kn1&`jr8
z_rs>>9F-G-ZCBK#cTeW9#TSS-@hrt>VJcqj&YK1X+b5c~3oRo((lx?irS)<F+1I;S
zhtu!EygOt)J>N?C>klee2)Etsy*sj9+xJhue1S1Vv2Fw>{T8m_MjTZHSqNnzF1}zl
z=XikYW@;hws=L3W^@TUc*-G)@{@iuX>#-1JBnHaMkJBF-di!1y?Y_v7v@j7SV#`aK
zg@x<f0_co%mOezl2G6R`?GaaPHAHE8K0qI`wQDy5n!Biykn&dJY*zRqp6|%D`Jmu~
z32@G3cy%4L98teVl3_$%nqNE_mGcF8<2~^s%g1{&><AzLx3qltpz={VWpIb688g!$
z6C;SkZRfpW7#05Z#xqi`8mHhP)%4m`kbu3!>`wcGjYYZLwoB7J^KK+LQkv(Hf(~K%
zu$NKz6pE!q5u~=3#Dm_|eQ`+hkd{|2fD&=^Gd>)0dg;MkEDLL^2C^zs06DsdHVvZr
z&e*B<lRVZ?s&G+8(IUbJC%Lu}dP2d|6vm0tfl&$U>jmoXOe)pJEq!HpFNe4BXaRdH
z4*mv>53eV{jTA1=4>**#Z8x4!$4yhys|Y`o?sLS+GYkwn&*H<{a>mkXcD7+rG7%Y7
zCj7C|a3E06HHZt=!^1#mw5XJ5XiOuqk@0{3u1uB|kxMTI_nn<of@{yYdg+^Dp=|C?
zU%E5;2Z^;*YQeH=WsIFelVD++M9b>3ZQHhO+qP}nwv8^^wr$(ibWFrd%;H<j^8SK1
zo|~EH%##b189{>>#_F*(WQH_MdIf9VRY<dXl2W^Lmm5rTEtF5gG55}5%2vqEeBQTk
zKQ>IQrfy?o{VTRru7Bv>vFD*KBjuL5rCHiw_Ec6|$rR;T%=XqTP}QgYCU_uzsiUoS
zcv37;<b5wcnnX66011xzH>>WGtS-Bv$k;Hu6|Nx*vV1}^2UP3@<t!`yN__;=iD#mZ
zp5-mUqMfw4#>&Yj$f=Q;r|3-2_(XJLDz+;>K69=qg$hH8<BURQF59ddX)auIGjWPe
zfCp+jM`B%>fAA#7v5Z)gs=K+!ysdy0Tl&0P8<8mJhQ%o~VeOgW79j`0y`0qcHWP*m
zkJn4oOSD>B&*zYo*<XL6YMOHq^UP)Exbu!neeLiR|JwgC6}0|sZqOyF&~Q6ra^uFG
zx20RDNV)H9Z-{xR<w+-Nyo$^sOHcPXDrko6NcKYRJIiWt42!=um=Z~&5Wuv3WmtFO
zrnlW8b$G%NRUcWukK_Jcxh1)Ug{>(h`oyL=<Y*<$n9etm*HfffRAi`$*|e|-DvrL`
zi%h2V<#QlGx39?rC?TW|*@R+Cb8-<Zh+l6`h?TSw3ghr?6Y#mxK7o{7#zxBtm>p`x
z_pYxkDh@bOZx8%b9XvFnvXBPNE#y)oFS=<rf<iQ;@=xt>A7F9GlBNBLJ?z|rm>Z#?
zJPeE*evE54AaADg1dn%!#dWp?g45-le~zIX(qYzt&yu)ula;yPY=!O)_I)DiE5h0j
z)xjd>7RvN^QSo<~6S!2#1nUUB?jsktRMot@rk=sA>(gLry;r%Sg45^WfcsWMUO)9>
zp%F6EUt4TWvzU<>bIg2m5UCeT_!^gcCe;hn?<u^8y$@RG#1DH9r0&2&Xu_vPL`tTE
zOwfb`$%*3QHxV|I3b=XG2`mp;*Epf`Gj3>lpP5S_+2wBh;_eNbD%8D4%u<W@+nJ}@
zxLqQ!jkGisb3RoOwb@c-ExDs^IM3)B=fFHLHB=J6^s3oewq$_qP_tkIBTjgTTSLX3
zN*4M=M*u}^ZT?5qMqiusUTh3LBP(`x;eZ@Q-@&M2a?L`HPc80J3UBuL2^_S>xb7%z
zVb!b<_G>MtW*+e!LPutupWCKT72T)gM9#6e`DHm6p4SNjE=oJYV9JL6P6CYi$D(&?
zpV22(dV?$BYVALN7`A{#D0-XjzDhgwlB_wD3@MOX@M3qFK#t)gMMlQ@CI#hzLT`JT
zM|OvF_8{aPUS<y0;rT!w!-X~4^e{+9(2n<t#r&&nrN5YCW?P_sejI>1M^|%mw4t6_
ztfIy0^9MswGCJmQ9NPpT;`vagleih?_LtxwCU!jK_il_dn71zXo*|S(nEtO`lN#^d
zK%nOJB-jFQfFWDkp}w|XrT0iu>pVumeUUma9j+|ZVqmKgTl+a6GkF2@t~kA#@tJ=y
zW|IQzE;5#TtlrZJp-cgm_-Y35ina6vuk($FUU_I*JG5Dz0|Jc3r@XgO9uch6sS_KD
zkb}0JdkZkz0{(5apEk{pXz|T7(>gbHoX<Z!y_4GOL<_Nym@*CkQE}=n1vAV(!qe4J
zbaqCu9-}})Kn>pH*c^OWqqvKTmY8uI_mT5An1-b%3wj_E5x?zW;JTRl2~a!K&J@N?
z=FvinE%O0Jj`0as#KC%|jv)GD$Ajz1Uw~0Az8r`@y!ky}!aU+?R*5;+X<Sl#j!3ml
z2gO_;PsdYhq@z*TR%I+BZ)BCZtbJDH?=?4qO=Rsf!;F+S<>=lLA1dEH<4oPfcxuC2
zq9ZfLnq1@V&8Xc!DX|plJTR{784RiowtyH}or$HI?LyRHSIcZGUX<zTHi2yd93Dx`
z^8+og=ynMmW_eoZv=6Ma(qdPq_bKqQ<n_@Q8P85heJKsm02tIXZJ)j(A#uG$OVgyb
zuZXITSqsBTQaPj*YD0DCJ|G7aD&ec49%*;fcK#h%Efpq>7fh)GFJfs+kOBqQk^J}1
zp4gCU*|{W8J4(rv)g>DEVJ=EgidO*pN#yg4kKbuN%R^f2aAgURnujFafurob47cRu
z2y2_-od3!S3xyj4Pq;5ga;fg#cT;PKR)i3@OiF(vCP-f4QKp?bHrqDYawQKxmoNYQ
z(|(KTH*G2ks$Q#%ICLD(7XK^FD)Hlp%cIRzE;T7S|LyZb`99g>EdYYUq(`M2{YvXF
zt=m!J+}$O_jdv=Gcjh%92GekGl){&KyM3#s82yR~g>B^l3@t13_91zeLomN~!$G?4
z+GwAT@b!LTBT`sDdH(M0A0W8(usCO?BoL1ucUq>W<!B`0k9xTxP&br~i*qd2`dR8+
zKUg$9IOd<pf5uEy>K@?je>RmG+1}B$ine@BTvlGUug$CT=}?P$6nDp55c1GS3pjN#
zrMIN9eVBp3n!wddofio4Clwd!qG`5>kQ?95;i>LL<!Hw!J_GtgS^P{A+2uJM8j@OS
zyi?OzaOZ=$gQJHYtQN4#9=BhV9&FpeH{}LYfgNGL#^MjpH3VuyX*U)cz$HFCYL}0X
zMwbB(+dMC`Ir(%+S)-r#QzC07dVwQVkuGF1kpW(F73<@B2m!~H*TYeA$l&n629)=(
zP}c9ja>oIfUb%AyA4Y;xV+Yc!7ZQ!!>6ZP|-+KPAkD3vK7fMjNEM%?nK*uQkaHxQL
z<VCEw)rzAXyY<*!Gqkej80&DQ8(a7KE^*_z)T~<|_BP9YjKihG%}<8UaUL>+nfuiC
zq1rgPMnlxWL5JJ)zRN6JW?Yv@6Fb)TfLo^Z+<$7r6Hk-+6fO;Yh+HCaZAcpu$<;UO
z`??Q0#-7Y_CGHCA0Q8+_<ZX_gcX^Ff&4dCSOxdDyqg1C@A9e*$9vIT4pSC`!Y5ic{
zeU)4)s=r$_%GQ@@Kb(5sd+4=a%cAn1oV!^;RJ}|Y{XIC>9>xRL!HH|$vuZ~~0ay%j
zh8%-`Fz{1WPLTdcRlecJhE#muUuafEvECV)jBu2xzPMq2wft~_^@?#ZY3_Gfbf`vz
zgBZb4<$bg^_fXRk>D~$L5)fLN$RRLeUj;IXEyS66{Y~#pgV`yv1UmQFJn9x_>_av}
zu!sgJ2t6=F^Q4XnOMlR9I&oN0gs^shS#4oAlQAWO3vk+KJvQ5lDVN1kVNaQgM!7C>
z%y)mBHOU-Mj6AMNtRiDQ5md}oVCe~l^($}mafoHIS)b+i*o!Pb!K-J$XeG^(^yB}#
zbit|AHwR<XpC|T-{o+sb!{{`FdR<x-!iDU)m|7;pFpRbOB~puS)L83AF`IPI{lN7{
zu259brD`%L3FV6f*b!bp9pj8XA>9a14$RE@Lj;^fbm$<gicn-(mXLuc%I;w}rI570
zqa`|dE<`<g8B7L)tdFQRSR)TzZ{twDZ$b!rA3(G0oL`}_bMn<B9o}16?Cnbd3~gmc
z^^|SA?*rip7z#j^ABZ7WpSE~WescS!SbQvSqSWQ#Sxlnrc$6I4&X{|$h`%&OPks;q
z2<NZV^TXzbYs5uiN6oh6c9ND4k&){fJ0Bvaxf{+%ek=+f^W=(qEH&uCjJ%I6u|RB9
z?<S|<f<hNhoB6qX@-l`^(e0RQ%oGw@w9Bh`rtD4fWo7(G{P{nHSx9+A2IfG4dLx41
z7rx+dQrq{d_<`KND+6^}ahkPAahI|s-{_a$w54M4+%^BK6!C`|PALf}>Dkdsle0^H
zWy^ixG`I4?t~ZfSw_0t{Y~asbR<XNYQZ6&DC1)g>{qsYzu0!)TR^{h$?BHg7D_dT%
zPQ%DH=iIED%smnnIOLP`fwmnDM<HCI?&2VqZTo`@rl6t-_+U~SWq?sPthodjZp!@<
zt|9LsQ%4$_NG)>r6YVWDc@-y*%5wP9Kw!hS1H1>c46<590O#8I8^VcF%{Wbn4E7DB
zF%k?SrjVkVz~(b?O$ujP1%(6s6t3zI^sg>`L}^Muvhv*aX|!?u5^lxpX3wY!bq%?I
zb3n7+nz3rL?IRNbXvhb)+Z%rRbCiPq%F48xE!vq$Nrj5*fCG9ON~NLyj~CfVSn1ZH
zBS?rNoew~14(fRWv-{hF06W6*5XSU&iV+fU=c8~lf|O-GYi%i-O;#uY%V*s=fGurl
zeW=@9d6BuUqiQHny@`yFI(odS)Nh3-H5r#~t^sZx&YHVfOpJctyTmz$8OfWhYzYsz
z2JJu<1EMR?juX)ggtMDgPPGJzJI?0*hjzgjmGvm#B+s|cYKLKsEO=9_Ta1s-gy||G
zeH|@b;MLscQ?rvN^}CS~v5rV^l#sxJT^*n!W?bOIWMgm}Mi<`)?c9N_PH(szM}FP;
zOnnoZPCqb<y%~2igWI?4hUPgm2ix?(yjsn%kze~Psh`Q3naE1iC^E{3jfZGKmoOi_
z{~ys>rdF}&Bc5urKB9P_M8Anb38KWZ*6qiG7t)K7&Uf8Jb&QNq--PB=7U*5wwtgV+
zdSX=whxQT#st`sIPR+?1KMqH7^n#^It!kFiS-y<@`SuX23bL0^fjNav4HJucMMM>B
z>z&zQ45Vn2t;#3LbVa~9nN#J@=db4z50A`wQ?2f@2@<t)=45EDkKJOtZo?I-Avvw9
z`{7cmnOD^b1g1z*WZlnHtsLXYK6cHi`PSC#zT3Cvu$SPD7=o5B<=70P3$0UU(w}G$
z-1Zi3DQ9D7kY|me@u?RRKY^k}5t)>$MFcTcMJwuU=!yKo4ZKN#D&XhiC8PmuTe8)K
zsY9ai8dBRteC}9b63y`$L`#loCkMsJv7K%w3G-kB75J70FbS7RbT-$0Rbtf<_ak_c
zD2MKhn(ab7z)Rb7YR|a&L7wEvt<_j;uG_rrc@dEnU^LUo2Zg@!A`<ce9#{_{V|8IU
zw5PM-Cj7j~&eQUnukv&v|C)viq0#B2rd!vg&3YSoB@R)@(WH2D)&gX-s8X$`zj?f&
z?LMH)ZkDb+u6tKXupJNjrXwSqfDUiFig#>7qA6qLM3Fj5!JFN!R<|c-4G1`(<R?sm
z0hZn0ZZDM*K-4Iuxr$o67RhlZh(&pK{@$2S7h&_TF*p?8fpoOPoJ>6AD!$djuQ0!D
zk?16Q=q_zMi?Ukz|4dy;5s%Kt#a?>bo$X|HbSZqum<S<3y%&d0N8Ugq`Wc(lc-uJ`
z2e!#0!IS(=hd}S<dOj63s@6ls$M_S8<Jy_tEYB-RtTamqfDrrcJ*+<Z)Gy{l^%O}T
zwk9!C@$St;Jb#b>lqq?({A|@rCs8`}O+1A8iV~g%mHV7#A)ct5Y~TNpg=U1tq<w&a
z8>kyK>W}KT=h+qxpexUllDAv{`Q-aRa>OYkrL?r5yzBaXQFZMP?gU>3H?iL}*pu7}
zfQyO{C{w|nb-9^9IpFir?q_8>D;p{&O_N&D&=sUfCV~lE4~4@=@3u_6UYmzAQWs3^
z+Zmtn3%t5)h}A)+^0j(=h$wKqJ?&GRYAWF3j1LReG|vs0w&GlrueuIwJr%?h+ype%
zaV6@1K0EjBLNO}EFYEYIYNx!9qk=<OmdQ3r)v6#^Vfoob8SI&Pt3pTDuruA|UVrC<
zz#sDI36N)t?HZMU>)SyNAHX@M6-d`RNzTuy-%^vSMA(r6<q^d#O(>3(!T`SddAIe5
zsY@btUE7pc(&_IvwTCB0wo$5^TrHEK%_E~+d4B~OX8c6X>ErDJsvwc?{#;UGPvQ1h
ze6mRcE%B&yFkY^)S$~necO0&1u(&>-&K+z4-Xf&~b$DjVS5UOkm=b>{&A9kLCXIid
z{k<(Qw0D}SA_=i}QU0%4v2=nsY{yM1TN6O`9L9{NOxLG0_HY|a@|P>O!Un~7S-}}x
z>BY-P38%nWfzihMd~=)!z0K#ScBgdrR|iLfgKsaj&wB#+d(`1{589e<`8At^CQDqO
z%0l$Zu=Vs4czlkw61$+j5jZT@=Uk}>dLBoL$xs=nL~gZd-Hp=&(WB-Mii=51S+Kma
z9|hVV{X|(|uxUJwvkBe4nFs~<1h#^Cv0U#wwxTPaj-Vv_OUqWtCc)EcZJCloY2!^=
zB_r08!vf@Ezj~}MMc18B$6nDgByCI}>if>7)oAi!E1~E`=50mOcP5>Qn<^A=Rvof^
zdO(b((ADS}iBfhhR}jyZib@13Lef|SbtjZq$0aA_0J@E_MjpcTWhBwU5?g`I;bq)a
zWPQ<|n{)T^1czR?%O-GSDFJNQZl+)!MxP*PI5?gx^#dt@T_du84EZVz8@uOZ%9Z@=
z0CSrn`{>A1wFah~*u&^hzf-@1+9vA*UG=T2vPBV7V{C=w+!YdL1sR4s7`22(_@?aN
z6+3vhDQ}KuN6|%C>C+q!FbJBj!mI=}5oe)CV}?JJd~c&xu?Ny-u1~HC%UTH*iUy-t
zg*vOfsNvz)SlFnL9>>wI{01?K2f{OD2MdaeC8{B8kTvj4cYpyqv97ySUWoH9lmrrv
z?{D%QGiDEGM@j5XZ^ZO2Ulwdhoo-PA?<Z)bO|@IJyk@>EUt(FsazK@2p5kqKbXq1j
z3Zwxzxun?U(y)g)ZKQiHH*PcXtuc3T`kW7k)L`C`BdYKkL4tD819s_6eLPnaox9?V
zAF^)JHmetiM&<CBKDc5*BuPDv$Mm7B7x-zUMI6uB`Dd*wdNs%)rP`4$dJy0}{SQNQ
zP?M&}UX=fcTc)$9t$*NL9cWb|C*6yXo=vrKqAuctI_Ao`PN~cQpg_+>EtT&~L|0f&
zP+1BuPM`jvs@-<U5dK~R^>{>e9$P`9FDY|D)F1@1gXFd4G&@lwNck{ki_QEt1l20Z
zg(NMKq!_SuCJH8K1~Q-ZA}=vLcYS!V=)}UU1Eg^Cx=d-y`B(>(+qq_)z!9-%PPHea
zwS^NApj7V*)ktuv)oF)6P`MbtLV{MFw_YM9WcgG>Hyos(SZZaIM;F+U$OXn@E&lP7
zHq1aHJQO9Txo8D-a8o~Gom~FB?ma-_Fj;=lMmd+YImA8VP5Pr}N`hvCWuCyZ6LO+{
zN1{;!%)65ux?w%R*2XNV1^HN&S4HKt=TX=rXwhUqIF2etWT&T}sSbB_A?YTJiFZRN
zTL<NYJaZe@t^38AC9u}?^|ywV30CBZTO@nu7DZ>SY=yrfw#L~>rj;bK>8ZB52o&-I
zdL#;_ACQJ&K3s73sWzD#^hm3RG2v?}gc5Y{+F{0HevHNTthVQ8m7TEZ0*Tk4RLfAi
z-K+))aDxtkcc@Z+`ua`k2=i7PQKI`Fp_>M-qIpY5-^tHTPP8PTrV-8HCaVF3(Z`{D
zVd??wHAeb<3-DT&(hkv)jYGe|ybVmjTQ!}_{EA*ZwIh8ecW!WK*c^LAu#l6R>Hh99
z5wq5!qM6At17g-H)Wz`N=hg`<dt;Yg0meDr9CaY(q1vvM7k%eyqJlm&_Ob57W5~B*
za7%dgZ`(ixPpl!pQLVG}hlRQ`IDN^@`?N~86O-DxJzS%Uj*WyJE!ZwMe#x<d?OXu?
zEY;4XXYh^C&)<|YRbUgF_xX<<^~d6&)m-~?I-Cj&&@&kF4~QEqhr#yP9JCd?&~7xj
zyig3^SA!)jP#4~?HJ4=q^6?4me7N4il*tY&fda3{tb_`nXvF0HA%+LN+1A$x`MnLK
z9u(ZWu;9rR$G1@N(`C@qX!11es=GKBpkb<5N3okXMwSP2y^mD0t7{WJ(-Hj1hAPEa
zlR=l7Ydu6u+j8w8@Pa$j<`i@UQM=pd+kbS{=AuD#Im=ihRJGK<%f(J}K<uyGxisUH
zWD*d)B`7;Gttt1XJh6?ZM>^<qjEW>Z5;taBYsCFfw)Xd?%6<0MU6c{XcjNiiy-wKU
zdcO7%<%<2JI@OOv4ozCnwZ?fR(s;!`fO(MH2KaBO_a#>vpWp&F>&KYC5Mt764-vm;
zsM(q0|9T))UgsapEZ(FMc5x0Zn^K!Qf8EjJ{_ORVvZ{S3yT#Dm>B~Xak<*UH=^?6>
z)|zo&P<S!*NTz9f5)@Quq}}d?JWGy{iAAHVQ;B>Tdy<J!q+6bUIxo+ntE7d=BNdtY
zY&zmvtaqr%r6Tb*d4tndTEew!Z)f!Mii=_dW`rq-G!#Sc#Ur%yScByhMu$o1J|oln
z3B%fKY-UqEx^V;w-e+#nnr!e!snYtz!fC((T<j8DG*i+DH*P&u8!10mz;eB(v2u%2
zn3X!RN(|Imt>JnejF-9D1esH?zNYU9vx|?Xl4ZnSTTI1pTEZC!rBw77N<q-Jq#cai
zcd9R-=uUL%NDm~{b8T<O726!ySc8de-|LLhKxPf+^*GbSj~Y>3fO)@F=C53&`EVY%
z4WJ^lUkseTH>k{y^^EKau23R%LE1T-8kEC2K@loIn}0zR8>nA;<%vF~x4c!y8lMO;
znOw(3wa1fq;h(;PqMhy`{RN+xTP2}9eskBkzxrXYLRiSd3S-grYJQiqfnj%<G+nJ>
zzl9HT@W_j_zgEW-Y52SEz9aKgKmu>WDm9}2Y~_sj?2LxCzZrs3-RzvZhvMOuUF5X1
zeDpM)L#n~6MLyMNU?VfApMxbsz$<i9eMy4?&eD<q%jolCXq|q6X&4{tH_<Pe@4-g(
z_v2l_*7rr*;R&%}pD5lCw15e}TgXZ>T(j#6Bx|;{UGO*~Q+Vf(;w=M(yS@6Qk}IbF
zw)nb*F|t8j$)C*+|G>bAT?;~F*7e>x<hN<sxcEhujW7<!i>?%%8J*ZWt)`BwVoF#P
zDzHGiaG4vzB<{0i3Fc63<Ab=6Ok2Uvsxi@q3^A1g$ZPB|G?v5XeUeY+PfmyeT$InO
zS?9zoxpS`1>G7pV#fMYUYrU5KKh~e?K3zQKFl*~BQ#i<1?tzH+u?y9T&PLBmf*gI_
z4r$We!vnlZbsS-kYQx-R$A<G=bfw*8GCeIs=1L7lbdC30={TT<^^}805a{`2!$xCn
zF@Q*)nEc6%i7B;FNWtK+H3*_@Vnm?_L=r4+25Y-?djbz<=%iSiyCa`hxZyRa`bZ<|
z`O1t~5adi))b}J-hT6{Qu8k4qy0H&<5l9Q?*Hxa5r0VpOIAxI2W~mc5-d2aH#f0mp
z7y|8?W?p}4)sACYqP#A5&i`N~c<||rq*LHqF#8{JE~#?QDohyNnO`#yj4gD$qyy1M
z$=8HFcWiHi7+X4`6S$10O4}HQy_~olPRm3*Qm>m{QG4<U_4q0JGuQ0CA0k!D2{<AB
zh$xS!+mcTz&7BD?G7X_o)AoGi9a<;*3y0<rf=->AqTfO5?H4d%0VbNvnE`cCov;Bv
zyM2qh8LS5xm?t;VsnsImV&$%<b~AA0dQYkUMiRN_+P!^krhJCo1*9$ccyc^A`YLD_
z={S}%R8uKCjqgOZZij9vFc15_NL^UM(b(g`-T(_xccDbdzSdSIKQ7sL^w9A*Tos38
z-DG7s7>~iNUdQ-9c;|#zrg>16oA1RQ8jI&wlR@uu;iZ|30h7Ky=Nd$}Ulk-VC}QcQ
z?DswzJWYvd-_)X*7T1K1`<69IXVONbaN0I^se0H}P1ke`#=FHR>s^9myHEIW#q8QL
za4|~PcF~!3Ttp~OC?#qIq_b_olLz-7uXhMHC2l5H;S)5E&DGU;gjZOsnF!zNoxH2m
z;oqo)SUXnDVKo>{1-okn_+sc=I5#`zZBj{#{E{E~N+1Q|9RMAk(3;*+O+1hDKn2cm
z(J&Zbji#5`rE2TXgf~UhBGbU2-XTo=nUdMA2l*1b3Ru{aUOHDi#<d(zrauQG{d&c9
zM=;S7S`oIjFG4t}u7wtIV=y4J!SfW+P30@PqvG9O!r<UZsX6#kIBVwyMGP;OT?x=R
zL-oNPi%dfhVNDAw$rKf3vRn*z$0@1l>}8Nvc@NZIlT%LddeUFw5>|v5uY^#?g-tL0
z6DvKM9d&GJcO&al-nCSQ@nRYVWi#y2z6Iul*@|CG(oe1Vx-`6n>2a^v^^ySd;&s$X
z+idpn$5^-m@WmZXX~7})G)dA@T)|maoQMTF)q#AFi}`?Y42zKLHmZ{LIjT#;ehGv0
z0=__e<i5Mvv#H)WQ--^UGd>yEN28eUJ1yeH$I1g=km)%tnUkm~fYd#KfSGG9vCS1}
zOAr!cA}NEikXISP^5Ib8cPD_Wl3gXp?x$Si?<>bmsIac6dD1xleOCj37r#Z1i+TGw
z@K87Jvm$teh)J0x){&c#c4Fh|_nDqaiw6)~qcvZ=VCD2tRwc6ot5U>C28~}=mG5pl
zzJZO0K-{8pZg`XFw{!yB!p;cOGcZsp7Srt-=|h*`nBQYaT>3fft$y|vZ~jhK5VRe!
zgZ|0JXXnOeB2<_!Ze|+tM0s1DN$%&=FJ2B^OW%K*F)-2phZzF{GZX9oj1?I1SsB>a
z|L5!ft{B)DSUCQB#PI)FF}Q&%Cv7gU2I7kphl`Wp+n*6X?ReehFTn!^Ck7KQ2zZK}
zcT1?9;|ND811eN52t*>Il%V~)$xnB@Zhq^YeW<KRZl}1Jyi9LT*DqnXqIwYW0(k~t
zD<l1l-uVCRuKW{YW5FOG0DwZ?0{8_CU0sdPM^HXm^U}AGBLxTH?Z<!l$qVxJp;yYF
zNAv?YA=vpJ3#b5yK>#8o1BZwM`2F$W$?w_V0}?^VgLw2`0C0T$Nr7NRkQ%87;&;f?
zi;%!NXLjQNWFgQ1hzSTrKB;i>_rOH@@&7gV!J{1lI}Fr<{dE9EAt0b5?|;M~Wi7~{
z!6+#3_s-A9;ar6tiR5GdYIgz1p^rf7Lx}uyp38gnEsp`{)MGu(ntu&l0T_{?f4KEQ
z?EBw?6y*a{52Ant49u@3huM3PL;36A=oc5j%n1qZ_x5k}0(SZHWdQ@+2mPqr$i1i#
zpns_`q5}!E5%k;Jp>HAV0XT&KJt;T8Gw7o00-)Rafr4}vBCL7WgQK57wG0CJHp7CI
zQ=$X(twH`yg@xwpi$iDP!8mniD(qE4yJo5Ey+}f^v5{j3k@j53VT1<$Q(@fD+@0F%
z>_N!2-R`CWL)+fkvxI&N75hVCpZvgA7JPvRxgh**o<a-*2>Txx3LX#wlyd=GfI1+y
zY4)vQL4Kov=(IO5y?qiQi2DP(LYMPR0Y-ikyodzx<O4uo5G?J!@&W%gg$v;O^FRjz
zt_k26*dO?*jdm2k_S#&%7wPK){J~H4#Rs^%c&sC%q8_|LuvfeF8}!3z@c$c8&=CPU
zo0ss5qO2_J4(R<J3<BsKNXQQWK~4<|Kgj#`+ZfA>{!$kAJGSi4g%6-MT>CL>yR^&W
z^+f=9iWg!J_@g0>7EnTq-1lx*ss$wC4}XPr_B-{eNAb&-si*R;m+`ZUQ^D)(@{Mx(
z1@q>2pwE_G^@<0Q+5yM#FB8D8H}wByTY~=3>g1I7S`%vhjkVP8Q|BP>!)Zu90fu}8
z4)OamFjq-OThhe~4PgH+own;8PubHpurKEYg6Zrqtk>Y*^Bb<G3~JM_LtswL>J`#w
zR&`9*1NSfyYAF0L!=to^82FL&2_!58poao|AsxZxT|oqV3#?~s4)gprWaIb8i|x}_
z0$9uV284w`zF!uIf&v1;uDfBpV$}!k_2KvTzf?i;<Khqe*EvJ04St<`jf2I9Qy{mF
z>MaklxwZRTH310m6d)K{RDuXsE_bpq-*jjO>L%OW2xtD0{b*>%-xW__At+TdnKI>=
zCnxg~y==d5h7(`jXctnO|7LgFAI!bA#Bu6wy(7SDya**~)cIOE-fxZ|-X#KCq4;Z>
zU1nMlDV<nSu8Zw6g8m>&$l*1rOXSMf+iTg$N9y4ypHdNN5!!s-;-;ErVx`0+p6>f@
z9B$EoMX!srUiMCPo3xn}pQ1il&b%$HN^|~#bRaxS`B3uHAHWfbJ5rmDFB7-dd~3ar
z4uw(y2xhX6rBdny+Uk3f;%RNB__0y3ox0m{>(eln;;CKLdD+y_X)`>Ku|;uWoOK2N
z=i|U>Ks+G3NKsns7_C-pbM{1({*dK)k`$*vrjtqkq7IFcm!ZR4G}yvpR@ym^_S?HW
zrPScQmM%F-jZERB(EB;c5BRf+9WKm%Yo)hRi9%Jyy}Yr#<qJm;>SImY=>(QzkaP-!
zW@s^I;~$9ngji4uc$y+OSe!@yJcYCs1e5o5m%PXC?2^b^p?-Czc8uXr7X>)3DEe8W
zGV5`Y+S9WG7|~q`ZoE7~<>B60i}$JOL9=Dy#!U$MQ|+y(J;Gb(ga4i#l>kzI#;0W&
z0S8lTa2In}lc93~$)is>7RxS9sWj{Nz?m=n!v?PqMl5`-h?@MxdGpj8iw!AXD!Tf$
z->=qBY{I#EP@$<eKMh39HS_f$%-SkbH@~z#3&{R$OTOHpG9=5guQjL4zr!U(G0>0$
z`o;b_NA{sImpdbgZ}(bbX0*6p84Q>pD?-k26FK!o>%p*X#OjQ!XUCc<36k(A&HaVL
z%S&%8A=j2p>jbq?q!Z?`WSeB*WX4SdY_V{7v1hLUE2h)5dP>j2K@H34uP$AH8yf<g
z^1LUccEPyF71v8(Crxfc2OZVOvI$)BH(U?W%*nrJM6P}LrMU;}=qsObsNo7#0DP>r
zn?&o8V{Nu&o1%8<pZ`7W^gA9W929(x<K3%<BR2iDf$09)nM|{&v>pidOYA3k>`zYv
zO&En0z?=0%Xj|Hj%G#*2lO3>aTkBH4ITCnEUFG4(^SzXeI784qn*_8I&&}+ZA~c(4
zS2(KY>rsRhZ=FlEj^T-0UID5g{kVL3eGFm$#EOmgM0;aTg3=u9>>2d%8_QohQ<&z!
zAaUWA2aT5~@-A1pb%8Lr3zss~Db*3mFrAr<P~dq$AZ|0_s;VU||0(AO54C$0yvKH7
z<7;1C)mYA`^Lg!fA0jEE{N!$XfdU7H^`>y*+ldPBWuEXAq?>jM3fw$5H{0}4RfUiE
z4uBg}jA%N~mx;jb&nw|oeh17l8Y0ajYy-!+*Ca7W#sn=`nwW*%oMvBMTr(#NR~~Ks
z)^#_dM#9;6_5Dl8alOt<pcPzCVK%o!+b>zXqfg#5RvZ*=zKeU~+KcGAocpSMR`9?(
zIlV_J@1myB77ju#t_bjHCt_7FsZ`mx9v13}3DE6ajT*L@Uv~dD)3XeY9%>2f{$x;_
zbpq5il1R&KedDvjf*0P;4ev=lW<H=sUL#%jg$YX3w#!&Z9Y^-(+hT1znUhLbv|d|d
zcH7(opEN~IEFHSD>_W5c7VHJdnH`VE2nkuMwZytB;cR|ik7XtjwxI(|`{3@aUkb;U
z+HW&GGE0MX^NBO8mD~xE-QxnV&zDxuyOuus&(ES2sRtYw$KbIndIE~j4gBF9R8Y~g
zbquB7pMIGxg%v28v6BuMhWnnYSvI~Hgd#?fE$G&)UFaN7G*H=Hp&e5FNqP(efHaU5
zu#<3f2wwQa;&F>Yy!f6v){8%!J;CXgbE=0CjvM*T>TyN%1}^~t1dp>Wt%IO7`1E%V
z1$R3oQ?Q3x$|F*6x9gN-RyVXRHRqUolizW#Mcj65bLZA*Q1^~(Pr_c|Z;yO*spRci
zIir+NwpC?qpxwP0=Xh~sAJTJrR3LBDBTbo}M<!P>>jnXe^E3Vh<a@PIhK9zwA(w0#
zvqPYdH0N>`uZNYNwO$_bJssD0ZIu(Fsl%BjV#5RLdxwSA`sQH?m6yHE=h{nKNr!@e
zE2X^1wHajby%QUH)!E$6tNU_HF4<8H?z&1-!yAtpm*nM}0V^P#$`KcA4{N%X99j`$
zL7bC>mIIkW+xE=XB)%jfwjgA%p~yb0JhiH(eNEP2A{oGW1RFzM+o()DOyK1FAH^UY
zXn`V9uvG}b*><B-a+sxkiJ>a@wU&D9KK9zT(F`87qYU;24a_VxQXtbd`ZQ4=cVw^M
zUPcbfz9C61mcult0CFH!m5f|M0pHd_+*IL~itBD4$T+3{<)5z1gwo+2L$i?>@+K(_
z9kREJh|G0b;|YRk)vW)-Cw360Xz_h17kJBUEzh#cWEnNZJab1y{;0IMm*C(Ebgw1e
zqJ(w_nsUFH`YtX3oVG8E8lA{##g%!lR$U0Xw^%t;y(JzfE12Dqqm56tdx5(tR?{ro
z5G&TjCY%I&2DtfdfMLg4@}E;MT*a#(ilsjGb9=e1_W3FO%LT=#n5TwzpZC>?+~eOK
zCPlbpzF{Q0z^Uq*rCT=BwuUaWV9EUk-oQlxV+Q*~MwnbSet_!;2xt>@%wmjSxuBkC
zhD(HeM7Q%&rQUlq^bG|oD``^ZXLwyzRm$zQ#ZK&yn;FD82{P>>^RGALj@Ks}a0EAS
zGp|z}PRXr9Y=&#Dqu;Jh8kx#PXy9Q*S#)QtW=%=O*0>0pSjCo>^(-x=uDkz03gSZ7
zre{44pXKt4fusz529elufo#tq=G$eRtoe%r_W*JL<CE8B(4t*$j`?Jc1i{2!N>E7M
zQ9HHS{Zt#PnmY!%CrpRe=|f7spMM3GBt8dZe512IsCUvO34K*dKwH<OXA|s5T`H+k
zt2KR)+GU;oRLk}aNt&N4Isas8>sx12C=ahA@3Mg-H-lrsGSo^iEc4<a-Loz}JVIh`
zYDs+Fuq}qSD;!mu_hOh*DxEXYFfUqgmtf4b*eY!9*fUTVd)*}X1q2rfO50wfXY$N)
z9{qZM^5U5i5&1NB6spdVnLUUJnls3PG&CiUgDvfCFxh=B&^qy8lGUMz8D)7U&xcZ-
z3XJRo!z61Try_(lt$`3e9M>%T^_HeDR15CfmC|~*2@~{mJ}i3F-9{XF4GjjkPOn>s
z_0hcyoXD=|AH4h8$f&r|SkvBmbSs)QoG;#9v~eKFlYLkC6TIv$?L~<?`*t@mo7IE8
zwE;o#=fNYT;{`+|LawEmc|^7PEn65%4WIL;(YTTD8BpJHA6P6O)8}Rl;;evt4$1~g
zIJ+(HAZ<N!NJEGwyH$On{avTwB|C1Q<l02aR<Ba({^4g`il$VuHQnPKG>b5#cZJWH
z*Bs}znE1$qDHyR`hFoh;WyocO064Tq%{PvafId%3k^LWw@h*e+E!$_iM@KEvg@wpJ
zRG+z_8mE2DHQlu=Lsg>YNL4@a=CkW!lqbqp5ugTZS9qKY1vQTJ!^TE!Glv9J-2V1}
zyW9)W-NQg}V8(}1((c`%PQk&4!-7=6)80ykFBK-@R2yM!z4d%O2J$c<g{F@(iaAg2
zi1j-gpo@{>2InJor)@X$5c@=r-})W1sJ6+)$m}kdLdaj;!vp5dF@6jU>{_EeBS4Mz
zyD75A=W@G!*UER7S==S~lHtz5fY^-Ud5VZ+z&Y2>4kHtz6pvXCe;ojuZ>+z>_}Q9r
z7D;9KaLc|8^_CVHazD)o6=qdmqt=?=l@2Y{sRVkQKVo{8mB^d!TL-8~?3mAw#!rT`
zm~_$Td*J*KHBCxI^-r|;mM!<bI<Ae+>zdScA|F9^ymgl7Ejjn2z!ckQC5j5=VSCGK
zahLU84m{`J`h`-HkfV>myqHugfA}8=g~7(3aC$>`pL<XJv<cPVZCU6##gb3me+I|)
z_iLM&f$E-!r^|A+)W(>)pj~ZWPNrZ&YaYZjErl1i1V_Cc8D&)yE@{e`ErN0n)@8#`
zzi{ZhH75dc2n-DQF*DNB?3M+Ml5c(iQDIpZb~HO2Wd_cQG{r;C-<$4M=t19%=D2W4
zYxo+tO_4kJqe~$Bv0-w|_oI{nH|;V-<V@l#m13Hnlg;8kY3VJg1}rFLkg+iBTm%OG
z1DjGQPc0N2R1b&@)aMXV9@oV64w)$gp+|o*<iwWsoVRquNEP?JDT|z~3?#zSMqKkw
zxP(=!GC6&F5T$L?eC|aiCw0m@-OpoSN|PH1t(bngcgxqoU&QtI>!&&-1HDzO7=L5)
zpAjo-)fwZG6m(j!I~a=ixb@fwR9v1Keg_xj{n#3>WP0#mI47Zqc+It%a{Rn)ubw>E
zIy7=8NFldL{j*X#3o+LbBumrEA3j7fFI&?=$)s4cI<3O=N$(4%NLgn>wgIh>3N}@4
zGM<sZxv0D(->T_oc{mL-;j-z-?GKI?9%WBMk<mx9Rz%(==Mt|~zn>H@0#gj?ERVIh
zF02TYldI6!KZeCVspSU-)@gX5&+f$lIRkz`xX$mAi`$*jm;zXKMFOYijfGdlf3D}q
z>^|>Qdyu258kB?y#5Q)UUs=w-;<4p8Lh^b|gU~^37o+OA<)&;vONIS16X1*_Y@pRu
zNHKfKXk8xX7q>qqd7fNGxG!rgF~mx4K_Q<79v#WBHu1b|#fR6`Ntz3Q>VKh@lPU{A
z==mf%URi}xaw94@cx@&4JcQ^ZTnovG`d#f0#eP{~v{2oW%m@0<{gEQC{7HhtF*IqL
z8Jwx?r>fI>r+D|ouYt-Fd$<LIm74J3G;OG39|KjyvUuHkGw8Rpp9$Loy`kRM8^m8T
z%UYKYl1*N)HRrakH$${(e@6{#MizxAc7uQqQTO#9?q<4H<s@DHxRQ>gL2T%>q<LPY
zb#*XjD;s|ZNfJfXAlAqO91h9)Ev(pK_M7Q7f}C1V-6o)lY0tK;T{PoY$`a*D2Xk}3
z+A=UeJ8BF>jrJrkLyOM$x??RN;9jhn*77CTeN?77Et(lPPfo8ME+5+JK)dbS#YsbO
z#5#&j)9VmM#2}BD7zWT$nMtK5bT>62=rvy%lLft;+<&cVc38PXd3t20?4Pm0vRq|5
znT;VFYq+Qiz<xRN8Z$?%$nv`+Z%_2VM@<Q(@B~dCF$>Pvm{?SBo?hhkmo5%l#55xb
zFryd=93Er{b{n;6E@<f#{g{jA?G4SYz>!P-+`XeTNR*7ONJhR-GGHw$Gd`BTDQqt3
zR9dUOLdgnLy~jf9ko&~0h%(m4-bZPlqa!AJJ_QYN()zrjF5^IHeFGFpnPSwuOL3$m
zxzK!6Q^#`Vc9#n#lVsr|WDSR@1J+R#%X7`Vlzo!QO>PU_VYNBfJSzB`i9r#gNRS@*
ze0PVt+Y*0}4nkH)-WOC!RpfAQ`J36PPtX&f=4PRjoPE=FMRGcK%;%jShpwE6(I_3p
zd(jP@ry4ma#Zx9;=LT`K7cEi3OC}a~-8|T0!~{*3D^B}v-r6Njw?lWkq9U>#-v2v(
zghk)YvL|j3S}3a)ZI5Mu!R-lP0Ci}VGKbmJ2D01&BT2UFlwM)jJ*=iz^~xnBc6+D1
z@LrhmH|<_U776D;hOd2$0F_5QAf1+D1>?3=wfqde*DI|!I;&=R>5;Jgf3l0K>x!wW
zP1f0aJ88=tZ^H<@ugF7Otk#crymx#HyA*yZ@<<_9k9#_fT|9v({dSHR;Qqzi7PWl^
zOE^`9)sKaC6m-30seYPk%3iy2^C`h`i^&#KEd>QmZC5x!Nffd))cJ#FYTHW=baErA
z$+d8JXncS|o?gORCh?G}^#b@dZi$IGtBx`5BypsPVG@#|%$FS-S4CpH2S<jY<iuj`
zG+N1%{TEvAC<CF1Pw4fs`bU>boyxCZRt3Ih=bx@BdFhPewuVhaLt<9`7C{NVw<E0^
z&*^Qb+q-oMRi*mjiq8-Be@wb8qqMy4jZDS)QrD_qC$vcu@*ay-V$LP1L#W>17P*^=
zHC|ds4^!^{MqGOu1pYoqoE|7C@C-nh4QHS@w9~iG&00;o(H+_&qEqAEdQp+dxoz1q
zQ8l!unWF*sNIyRw70%V`5LTUq2K4*x@jg-U`c7jT5m(Lm2Xc>gw0>39E#-{_fkHqG
z*mEf%0ENlX7daE_5>t4Aa}O?U8OCs3+*-%N6u(;V<C=H(rNSRtxir!y5n8;QoGmIp
zkFpTF(3*-ju(;U`GMxLCn$RI;`1DxEja~9B?<l;h?b^Mc0FS4y)p?GNf)0M~n{qaf
zY1HOqbQqqXKVEHsS_+HihgJsx7h7J{89o=uoShq1AaDVP$u_!a@B5S#8WVOVrJN+P
z;}U|-a%qMPiPZr>qd=|;hfo$)Mc0~^Ls$UqM;YY^U~y6%jz#<TF7Q;M&mKW{MdYLn
z?3|*~knvp*O1vIcg_<Yn!EM*E^~ff?PLt{T2LiEuChC4B(`s86wh6|Z3Y^O48d^^q
zI*+DGGZC<j+Q&Dn{kiuSL%4rk7)|}))K=1sV%oJZuD$kIzAE-uQC>)ymG3*pcC%<5
zk)b&Zizqn_i6#N0@CvAv;aYySn_noc$z?zO&Tlv>Dx2Dxl2f)2a$S6z=rFsKCC7co
z7ZwDIJk?!VPdiMg;+EvZIyx(b`VpT~({%n4?+np{Rp*;!YiBevu5v~qQ=h`I%+?Ur
zp|nwx4iq-D%t9ThWEe5^Y}to4+6{Ss?po_Y?B-Saa`(30k<QZNJtD6y5jA;)Q6EKQ
z4q1DpVEFUS%MFrrMMiIRxZa&xs<OIr?=<n&ae8WI{t9vU;991yur#$&>xvjwy#9Tb
z+Y$}Z7kT(t2xjB8zE%#QU4x}j{9q=Sl)4-3OfsB@s`z{C?$c<TFehvqVC6C}7whw6
z4Ti9ZoZdr#MI+g<u!z@}B3qm${&k?y+g>)N1-MdR!j~4!C`(S01uz|W66MO0cv3=K
z`V&;E4VFIxc#w<axa6MHZRU23nP<RTX3g+AMlr9xo(3q8AYRrIlcUN9<C_qs(*pJp
zBBs=@-NKh!f`R_TX6yLV<v^TGESRKdri1wx9#gucQ9UHwm>5~%uqDOIs=QK3v||a!
zoq^-3?ZN%e`Z;}L_@B|vJ`1#S;Dq%wv}R)O<%{*wra96VTUWA5wJU4m%hcZ}R!4c~
zq+x5X2zTNi<zrc+JkiP6GltuYPzN+E;jQ6tw-x81)?%xR3UT$xj}nR@iq*)G6eHLB
z3d`3n<4G&W?n0RsFW~uL>WM@Xb_Qg2gDdSew(>q^Dh*CuiBsV3YSs;lGDqmTYl?O`
z5~5Onu0T~o-E2M0kObnCj{}=6=&w~CVRWZ?A=d@)xXbD$n1KMUt<e}wJtf{d=J83v
z_jgnH+0i^}ZSr~5g+v`WpAED#PVt^aHcJeCt1+jr=_0IzO6_chL-ZbT_kA4azbU<d
zrm<b`8X9P^8+$t#ywy+Rb<Uiy8N4kJ2ju6+;b&did2i=4dKgUZQ)>*rJ?sazv*P!z
zy;grK(&hpjW*)P`t~OkGok3hhc@Li#AjaP$iq!7>=Q`=-_<VeG#v9YQB%SvnsbgE~
zL*a~?5pda@eSBKi&EBXD)g0d!JRUU3-w__5opK-0yxR17!A3<T)h}lut=1aQjJ04I
z$@Q=s#w)#)hjuw8u!}HgECv_??a$MWC`>+_Kgh5hOj5JJ>`2jCQl7%|>J}ekcQ)MV
z&9&UIpH`(}uC9Yp@w5XYaX<NHLc700g-pD;%4vwqf{LzCD(s{OUfu?dH}85h%_D8O
zkTabx&L?vBumO;Ya@R5)O*W!w{Bwec25We&LcU$XVxy}{3g>&p_o=#5pcpG;4iZ+K
zZFjJ$OeR9TG@GlEbn2ELJ37DS9XbJ-iDjJVcqCNt$c5{;5oTbCw6J|ch1|(nP}I#J
z<8OV7h)mCc8cq?|e2b+zM$p(GV_xgjysfO7N=6@?xyyeq1CFp-zt3NJeyV|5W}0pA
zbcs_!U=rO)Yr^567Q^LfPQ2GE$muD(g)^naRcNgauR${u#J9*Y)_Ukq{lxY-`kugq
zo&p#zrdih4h5Ync=x!c2%))$(%^|81$>pQXt}zuvg5RZ*98@pHWTK8oYQNtp+jGa_
zXuTr)t2qJcs^c~o;GlwcS}4nQD#R@<uF<_``-^4GS|$_u^uYN(E0ly?>h>Rn9vfjp
zKl9p^DqQd`4t007*mx>&$2{aZYv9|flr(%OfO5|@JXqEwbC#;|n&=X!@Av!wcC@av
z|6|_B_@DE}|AnLd%Ntpl*#2YQ$jZR<|IQnoL6w)bS9o(IgZ%x4B$=GW-P|M~2I>3Y
zndq64{hJ7JB+GyhCDWXPiW4PA@$<zioV|;BUVC1<e}6T1TTN?rv%6m#U)MjgPdT!y
zLrH-oW9`AiIk7_#;ZZ>*0F<Aenu-DW^UMFGhJpj*7EmJXLO+g4_L)NP;^f$*k9Ytm
z|1f6*Z^{DX(6Vr_z~)|^z`-Jb1Bi?RiH!XC{2&m?KWOl=@qp!l-ubZrJOci#*kA&Q
z_LY4+-w5(%Xvn5nel&nv`ndk}O-;nVx6b~F;2?mYLHYnZ@Drd10qY3Rj^OEi{qhhK
zzj!2VCOY!uau@^zdU|>S@!*Q^vG%E_``~Ya3bX)55Qu?y0qj7&Fwk@09{fMjq~ZgR
z^ezE}zi%>ovgCU3KtO%$TLt<M&=KQzp&Wwn08=LbNHjYB0c0RA;MJeteQ<A=)&K*6
z1HZJdYA<zy@@u!o@aWL5cOd;=1@vbANJ}_C^zy4H!tVudfcWW$bogU9kfTRI-u&`z
z@xWkrolfxlI^z%k1JG})?&9=77ZHL`gatRfqR1aAU``X3^kOJ&Vc;O4f{9-jaww7h
z{tTlxbWf+I+xqr;|HO)+NXRE=PgW3Oc5(zv=u4AWg(V+oAcutSo2SqMfI|U_iG_-W
z0O4o=2jH#<A1XUbyGRdQ<X@998TjOVNH>u71JHnwft&*Le#C&{@o*sig6=|rKtD`(
za#6zS=z9>LMF6nyX5`UFw-@MaL+5cZ?cN|uaQZMq_HyU|U(cVf6R1LTV(e4fzjkl8
zpD2t<s@-*!gRf=>exu1L`@a4FfCv`;{^9KK_|VD7;Npg;0>9Ko5P?6-V0X9LH0em#
zN%v(sQykw_CV9Q+ecx%E?f}2krEsHjs6cx^S;wsW4*2Mg2tRyI-)<AXaR<GXUv|{L
zx1mej+uMF&?S7-bdGuqz7dO9X0E3!#$fAIr_R+BVzc9`IUkWw2LkJhw_qz%-$Ph&!
zB<W0{(?A1;f&2}8zT}EH7vK)e!G(Rae(aB<b}v#^H}dF1@Ug)^p4#;Poj1}00e2eU
z!V`z4`w8L`i`ta$uD(r%>OVcr@-FG?w;j^$M}$iO@I!!%<$*Xoqf24K<3}r$6Aj(6
zH3JF;(z8P)0rgE`gV+ZWiTmm_HWBkfJWW1tBBG`OKotB8CIdmde}z%Nz(0K3{1k!L
zvjGKIJg~XY!~9JCd?WJX`@++YmtnBK=>}WL2wzI1EZFD;#4u=lbw43VAp|rGfD4!N
zHbZtU8z*I7apW*Q%XJT2z-e<mrF6?WrkVDF=Ud$_y+`3dZYk_K_8i8uJNX3jq#)W)
z4Y3OC1gq#LR8A2-Ld}PnjA0ASOH}WjZJ7;7;JUY#o*G5cp~*ya6uTN%`FfeDNVL%6
zB>fvYXEI#W{L^ZMzQKKTdRXB??q4>r$}K)ItvefU%1I>pP*JFYg@U1xdu4PQ3O4%;
zu)c8M7LhOCz%W)sBB*B4b>Vl8c^yBpBvJMW1Zmz^8R@+<%3Vmr{1p88xT~U}s#CMG
zCGw%0mrbY@eJGj=d@C>>o-eMvm0)G6%j-)u{|X-EMBaiD{G!AVl`cAz!$X4+p^UnK
zXe|slRdL4FQ{z*)A{=hi-EVhNiNOANeh%xssdr9Qca-2eb7O7TSNKIAW4YUdIt_@x
zg@g0+m0k<}_ntb~wiUb?4nHll&^pkf0xC%t7A&A8>W8h$f+&@=F4aTB!p&QBl~BPn
z8l1ivddH5H+uf!=pY4r?zsIi)`tQ$HZSn;WjtBU#q!p&9fWXB2^E=^~gxq}k$bVCZ
z=zq&P-hy7qqc+geWBgqy4D0ibWw-~l#}Mn6xp7{~wq#s02hLo9pMmN{4@k6Ta8SZz
z<|H2eh#k>)Bb_h7owkrejw~pdWCsfSFUroTF%v*rvaxNO9VcIG8y(xWZFX$iwr#6p
z+qN}*US=NVJly*;_O7b6PV!z&9+>AIGTavR#-Q;?JRhJ%qHuV(UZtK(9+uv7ij<Ve
z<0$Ecw~^Nme?d};=?GJ>#Tjq5F0y*nN_b=vP?MXC=Zw4#=PMb~U7UPL7droPP%p@>
zrRwhKu5Mdu6n`4Y{k*ma`j!P3mE)~!<;3P}YE2ch4$-umQ3=+ECvX6_+HzE^MeE=<
zN#mrRK9H+RZ7wgZ8%Ba^=l6D#y#0Nax~k1~%jf<l^H(qmnvxT5W1{Q<=!8+sZI~Qh
ziVZ6Gkfmm9+i94+J;*7`VF?{<J14nWlI_%U>cdYK4EyF1H4P}~!49I7V<9fMw(+T6
z<aRA@H@?Zqg9p1!L-0FN(~<BrhjZldgNx}7<LPqo4f5agkF|sRxmRT2QI^yfZ)>a7
z--}i6@@hne)RD5R@H=Q`hO%N6h>a*B3x}5_Zx^E#D?ZmvI!Yk+zl5Ew&EHJ!(_q;r
zJ^+0VG`ViDvdST|PMYUcB3Asj*!I8hS}Ha9S2gdkl-fnhnaSGfSv+d^Vl-_LQ*O}@
ze5Uu}u9TP!^IZ5<MmR<J5^QV3{+=qS8a4hHuaJXRk!uQjo}OJyZ=_s?@aL?OQ59`t
zm$@gFM-MN`I^T{(t#~$*^Kb9xPp?{*3(#q9c5YlvqY+)(^i9*WbVmhPH6s!{tP2`n
zEl;g6R^w${2X^->1@;UZCmDKVq7*Z?%%ruzeA_3AKnVKnVDbd|gCnc9aE{{c!lCma
zbDBIR-cyJ=jkLOrdKq52v4P9uIoPIelQCU8()dPndL6a<?#Gw)v@MCv%!Ob=YD9Qb
zPoig9_peKb@uWqReMW6WT`^c{R$7c1N?xTWqtO3d$s7T7rgWSez(iS-XKSo1&)xTw
z@x(5oSnGi_+8mM_(3L<mi)RYF!Go(a_)V5s>dpJ`Y<nO@nFvlk2P;=9ruV)Cc0~qo
zUZWen$vsJv`|)nQlxG|FuU($1URLw%KD?W{tje~dVdkzcj8xj>pZ|*Z)7`U^;ZlQI
zENg;d!C|rfP9E%4Y|5Y7tRUeaY}Yob+N&pkv=<6%;+#b3)paVZ21`v|+Q3MgkC)gQ
zHHI%dotUVa>I){OUTsomD4nr0A5Q4UP6Ynfo>>eol@h8TM0|&VJpS(nud@!*xPt-i
z=CbRAa83q?w8u+lL)$eM18qfg8{PN`hC7WX%@C!*#!#0pl-?-v>Ec;$MeGlBW)0de
zrJE&YNqm;>fL3nYpAkn|y?Ymh;4M>HvZT-Rzg{7YYQ#+4cvGwk@A`GIxfdoRY|f*8
zS7`~)yFDbWX`3fT8@Ttc-+_`1ICj7tb#(RX7rQ9|@D)boPJ*1BnP!4cf+1X*YPI(~
z59%Hr03m2d{OS(%r0wZl%z7V44@icK3ikGG#AeOyP%~U!<XSBvRx=j^oM+}pQg#`J
zK@ZCuqmKlAh9bv8(*+mQUtJ1UPL}-1)y?lpjgGJ5Cyr}ZveJXddN1sfm1_>0qM<HI
z{Cn@*`F!5gctn@B#lO62PkZi*o0X?e{k!IDx<P-RDI}NuGI_)k9f~p>c6YN<oGg|a
z>TbXd4gsMD_Mps{1(j=S8YCYwi^a<fA#qyf^zx_CI}{%1)htTUZ@j*kzkJhJ=lnSF
zO-zCLwT_~lIp*b!7icwvFG=n0C7uy^*hdrF_B$Zr6;;)JwZ#dc^9g4uROX@w^#taR
z<1sT2wTG(703Lrt*RVKY>MI7Q-YZv5s=mZ~LYpPDryQr^=v%h{w%nVuB|=B#X#bY1
zWZrHIFXmXr1(swmNPmH%5rxxePoxhL*n^&oL9af=kK=~l?`aefGw=GvHm89;A&rXF
z!-sAKy@)lwpUczpyQ#u3gq`j^>^`Nr0~qZ7oL8@6X3gO~g?EO43gmMq1d!(dK|4>>
z)T-q64F0YI8z~#K+68^xXeGCDC(Jnbdg5z4kjU;P7z@Pd$j`RJE?p&QK8@#jDx8aa
ztl+wt4*y<@@%h@hwbsIjv$^DAiLIuw9wc#y>;@pt@_PSxmaV&M@7P&-hgH-Ys&C%U
z-?P5j1qWn(&AyvWuj@9a3(TalpmkRB@3Dy+?Tt3c{Ls=8B(dn3cHt5qd%<C(0-()j
z%o=V~XEK8vP(A0_eqKmZ$y*ra?n&E>j*fwpv%#Z`4tI<I0=dhWO}8II?o*G0FcM2c
zrFHet*_hfeCZ<>gQ&ym2QZ-lBW&$Nao+%SMb8gB?q0){o(IT#n=^5zEYF|I}NUE*c
z3o$WZ!X^=OSz>i^-`^H0=36tJKqhu|PFfQ{ClR%B7tv1}VXZ7B9wL8)Qy`1IWhg*Q
zw8qqpQ%9iPWU&t-vDfWO1@OpR2JZR!XpJAb(8T!GgvLDZ?D$~|+a0opTGw5`Eu!p{
z@8llIQrj(_9&s_{XE)@VP`n#=df0!OR>n7oVI2~t-DuGCAjZg7cswcDE$OIWWN5yK
z?u#9m&hWE(s+hMB3f|Y#t4Nj6^(l71>EZc|bDIj$^SVU{e&3@*Bg2_v4ri;u&fAoL
z?V_9Jf)3wmq!FHvA=%EiChttU4Koq?bL4aGYxQG*+M#CA_1p2CU8*n2Cr@o7+r>+<
z<y#_X{3GGMY$oStSM6}NDl{^NK91re6e)!Dd_NWmG0CeoL>aC@f!Z9Q9F?caZDPZx
z8O%~gR$VhD8XSRT4GjOZ_y(S58WwC^T!=7z@_sdUO?$Pwx(+E+rk0M<$CjY2n9Ot`
z`VLu$dqk6uLG3+QlzbL_mhcT=A1N-P(bR8NSg`&TpS6<aIj8W8-=5hkatQvrM>J*Y
zvmlA8*H17*fI&_3P=6}pCQ|;kzucvUMyHf)ZFvwsg4Qv4TIpN5M~xw4sr_c8D6KIi
zT9>Y}YNFSq)R(@ig<G9|i`g9L77y*NQwR<yJW>xFIzp~$DC!i6hvu_vf~_U^&~LR_
z-L}`lI?ECarq&!0-|ftZ=AX-;l8AQP<+HY)k^^vB(_N@+077c>EgA#CaZ+U)jd45g
z<zW}ouW^;M?Rs`;Y2Xf%bSFA~Rn(BzjNr1<#W=}s)WqDFnn>yR(5Af2X|K{hr*U8r
zg1metoVHb^BsE`2kS$he9wTAJ$xPM(dd-PN)o!W@qLb9iQ1D6T$vwu8pYM46rMDew
zn~SM`ZJ4;WjM;^{Q$EugFQxJ{t@{Sio^#oBLsKJF?a~-Rc6s!Byx+l-)%VB_o*in1
z)RdjSIR<B!Q`c_6goXbt>H=&abB8Qj6M3d1PEwh6OdW9Gm6+aK2y%U?eEemwVyqKr
zaAuJYb?T2cAWgW8a;5H*D4t{GDP<#@H{UfUY&o5iG!ZSH5jyl>{zaO}mfXF97LoYc
z{PpWLLwCSyzJ`qEyxB6|y12ls=codgH8sCC`0$p?uei0aJZ0@4-$M9`GTn{x1Fs-k
z`|k;%y;_BZa?(gP^Peq5fbf$<;@Y*&s%n>Yp$c^;DZM}U&W6uCOWZHG)h`On6d2uo
ziO({V(RDIq>WD!=oL)heDpugKyMN^kotn|26GP+j(DXiuieN*CqS|yJwx|H<1E>dM
zG(yC@p9G#s4fKRo2{fJaL3)our}-z|y;Ad&hH!jY!jhzgL;HdMd(HOXKkt0Q^#%h`
z?V#~Bz&Pt48sLKEcDk%O=5P9}E(b&s)|UBOL~7ArnMPFCw_p@}8eU6Ty>13=4XLMe
zx!ry@4#X-_;UoxoM@NVz4fYvr(iQnNKlxx4F}eMOdWIL0;Iylt#+12}+65v!Nnk4+
z)`(g7h+;(ZtB~d7zg+8@A+ZqXv;+$+EGoawNX&havcY+RuE`lkc_Fjg|5_?z3Stj<
zi^tHY@(2K4P<gzg_D2oyaYqR_Ekoq3H@azcQfci!)x=Zo3Rui2v+<PcMjRgLZYC+F
zgr+?-5aogdIuqt^#9ux_rAUXt=G7@<PpP{LqGsPq_T+&H*mn{;o+OuHgPjq8?^vg|
zd1<;)u5xV%W!3pg_x=Q>@CrZgX2L)-a*}M!+`=hwV@0y;jq5~4hp>P0usrK#`;^jG
zZ=}Z~Ys{*6$oit1LG>>`1-AD+4r<JnySxcl^HEWR8b};4H^o)37JoRJ3UN_ZH>hem
zn8~xvi(ApjrUYZse}mq*wtK$HjztE+NToo$J9n4^2F2SKkx+(l;?bTpuM@DMn_~(9
z4Vn4$qjBTIY**xH|LBe!Y%~Jk6Xg>-1T?^Qv?R{HA+Pg;O@0oju-+aUMn81Azum7y
zyB~DJ!yI$}OsYa0h$Tx!@FIG=#@e>ZRyXA_=zc{SIO*xqFt*8JxN|cBSDh(Z`zm9e
z{aG0}A1nJaSGDvqN>+aBq*2l=s8lvKX9mMBOCV^XS0L|MzthryZTUbRNu9$%mXTe*
zn*@f?4UIeX3ts-{8?XR|<iKCIxKY52Bmr!Sq`Q#FGSrs7-SD=L@8~T`K}eEFpffL9
zIauy01*WsO?jyxi>k;>KocO#N30-N#^7zbn{V2(IgLEipaEaJ8;~L_N=(>SGah5-t
zlsG*{5@h5kRc%T!+j)BgZMW<E&KY1~2fjq3qVKRY+!r@?wsdl8%PCu1O#kAMvwDJ_
zsYea&+Qq<5OOkPrRDoXmXguwVEv$4WN!=@t0#n~3)#+?-ZdC55hU?Mf{h~yLykn5A
z$A$f4QNJpK<FJ~*eFW7;q@(#M0!xJA{kp$pG%CQe3eWWQJmJ)VIctjii6h92trB|D
zu0WA7u_?6rMI@PT>{>~E*6K#StGmO397&x|V*^*Wp1n0~n%;3*s|r1QQ>7|FX>rI8
zaUmmS`oPX8zI`aV^H}xbCOQ;O1%Y_%5obGenr17$2SpJ{Z2qY1L8edX>)t{5F?x40
zyF&<uRlovlxtM~cgag*njjj6JnLDyCX)#lLi=Z8O3z_5niZaEYGp3_?XDe%yRVlh{
zaaMDB^tU^!DThQ=(IrT%T9o;Xf!$X2)RRv)Gen2^!F?kvL*p`Nx4`g(tgx*LbAP^5
zUb{oyuBI9{e%H)?WIQ+KP1Zb8B<*~efk0KYpA^&7%_L{Pi_V9W%&%&|V>@}pZtG(1
zTVp?(E$F?J$#VK-`Fuimp~#*FB{EX!aZbitWd(3ulVq>JEj<$*@np3Ysu+L4-ln;w
zl3~uE4Ct}BY&2wD-xu|i%Y0-U)sESVX|k3RoMF6CM5O{3e#_pVuv{hOcQPgQZ_t;`
z8W^Ivhi{uV0b(XO!!kfAl1b=vsf0U9(jKes;kCL-=fPf>NWcEZ*7mRI4KE^)PoD1|
z^hwvEXFQ2BsWc{Uo@*<dHg)>xqH*G{DO+c1drxO1iqxP)LcI4jY~uAc*ls9Vi#93(
zm1iU6Se8Jm1~!d-N>2U-yYOM!w>41gk9GcUrs|@W10rC%EhH;1CbXk>ZM?5L#S-5K
z>LTe@k;~7jPQ6ya_8nJ6i6Z16YY3CSHU_l8TiNUT+&t-gNNr2u0eQCVvsejTPgtGK
zrsVuAYbI|XNpqtCO7fgzOJ7*N|CMAD@T(J&smd<IF$}~TM~pyJ(L}j^F1e(Fvpkt#
zgAIK}#nNuVXY5W0#dyL?`s^|aGgjw@ZQa3lk8?hmSxGya{^a*o!KX+aAN#+{#mzZx
zR;46=eP3@>4bW!apL6UNG_FR{zC*=?*EEp)^n}6b?fAC0&s$h1`e{E(dz~Q;YKzkn
zKL2I?i(U&Fx+YxF1u@!t{If+lMGcY8UssroS36@(!qPMI`N}guP+C~!GKg|#hknY%
z@k_bN-(|`n*pG@Ox4g2AWUq(mF59em`{uIdLsfQ_r1rD(Y!L^%yLN~nr`tvYF45+b
z>gpcCeTXcF#3%0qSw}^xL-Ccgbw-^$;?GL90D=RyUhwp=+&D-_?u-O10JoDqw2GV;
z&d}K{PNcnM&Rl1bS$&#D&UI{Q4|`W?eL#s2*<D)?UFXdJzLZ7773K_TR;9U;G~vdO
zGIkKlILm(LD20)SSm&=!Oo#1Nx_sxt`ue!98}BFNK{=-|Mr5YGUHD&tl6HN_viJ%n
z`{!+vRqm|JGXfRAKZ5l$qrfVN;b8$X`zN;_2u;5#2p8p+U{Xy+W8L*e#ig<S^@B<)
z&YyqUxO;XTHn>zAM)ec#Hr;#!Mp!xw+Y7IbwyjEt0g{e7K9o;5gnP+|9gydw<lsa&
z5{V2Y3&e8=zMkXeO^=p;?!O|3z|OvqadDkS_o^PM8febn?-YSGHu0J2p$J1p@ext0
zc!YYO3o0v?i}wg!x$zOXaXI*NkvA<!d&T@VEWfC?6aAE47ApVD3S+cwRP+G5aBjWP
zbr3HAWP8+U*5(6qt&*m9)*t;8`ooV<G@{4)O-ri$3UckJ3$^DufvQh=xDy}Z9sP7(
z>ThCzy`1_L1m!YRcP-T$_R$RFC_>6fiY<htz1+Guy=a>6_Tr(q4~GI64hDLIDQ1&v
z4L;WvJ=-x^MS<GsCllF68b30lxkd1RkvFzxAFiduRjBxX7}d^|hy#D_Dl=WM=XrLk
z`zhqw#I6Qw?i1CoLJkR@kho!FtA0uAWO)Tox%Tf~vLe=3ymWdGiw*Ui%CG6_3Bi**
zQGl{-qHULtX2hAh!e32cICCS(F}1nIwwtb5RlVUOQ8jO0veYsnl+Ka!x7`9;G7*5(
zzZ5L~Q-dt`ocS3V!;<W5y&6jpCQ;r#o7vpx*QN*98Lk;cM7n(hrGq^1GVCB;UAr)t
z#fbz3S{CWG6r7?Y?od|%r)1ecqbEiA?Mf8|`)rH1g^pEe;J=3*4O=WCwJ(z^P6}p`
z21D;4jb->eF?bxF^0){`qqA-LaC%nEVxvEBAaZPv6t6mTRaZ_9U+zTtOc~aUw@%Mh
zdN%u&?R_%W1?iqf4OHbF#7LS2otUicnwJ@9KU5@6V>CQXq@*2M47ozxPDHz8<h3Hg
z_OfN>FUdLEEkze@<|4?(20iGsQX~m9vT@nm0#an^sN_~+*dMI@H=6>Ky342HR<$P2
z+l=U{{cq2{E9cRV%T9~Z2yoOkYGdBZfQ48Pe1NLu5g&byggU~jFwVPlcOmg)?RCDl
zi=9Y(_+99^hGc;CsG@s#<8HPt3aV8+Hd?~SVxS#-J|nM3rtNWE8stY>$A>lFIu|I4
zm4+0Pih&k``(Ds^^=BFAFdlU9oAP%MjBG1c+Ew^Sn$-c=4c#X_ZeY!I3Rv{z`Fs_n
zQ?<3e#VMerD>v>UlHIPLJMK0gk?7<3NKab{Bcl-C>eVDk*LAsrocxwp;w|-1Ik1wA
zl5I+pJ{i9-%b8P!|4cSyI%lRrazi7}@7`*+OES2g&TnGx(39UWb2K}r&e%h;hQY~@
zKe>ASmLB8D?V*m{*iDs7H~X{M&=!?)@O!qlzSqfldi}Eq*G(Zw^v47Isz>iVolNrI
z=uJQ{)Iubz%l``;Uw!T?VlED<=aZhe>q1-H#6i^qSjuB>xbo_gnTt2p?a`Q~S+KI1
zQ{dr2^^!OKLAw5iWT6U-t;pGSg`qMBl6kk3^_h)mdxwr;(qJP7?$OM(<T=BVokg6a
z<SbK@x>>Rv=CR_Kik-Z>{YlxVM}d$PzsHU8AJ9nlb2CB1dz;eJc0fuxzQ}zhjd0ot
zrBHvwmmf=`TpYv}kY5!ynVRA|G$iRa${=)J{NP00HJk^VKI*xX(~GrC%s|!mGMFGi
zG)P=7s#bKXSiOBYOGUqbN;+`BAXSnh@1Z9IF5RVrFx)cJLtZ97J=PTUt-Q*$BsQn0
zvA(%CZ`+mV26_^~*c98Tn}1-d{kfagK+kRW_yvM-;HZy_k95MF6eI|wK6RVW1r!1W
zSG(-Xt>xVRq>Oi=98J#qIl-YPKBF)&qsvryYtMS-IM)Aaz$Qfz?EVkCqBiVR%od=4
zG22qka^+Pt1<5hdz53m`Z$1hktJ*lz2o!Wi^#lj4t22|kWa5@g0_;%%OI}Pf%h@2u
zO&XMuCv+)Gx`yW&UEYDn8bHjicbYMBhU$!n(RP=T)Ln1s1)yV5c`A5}=Mnf-S+1GS
zo2O1BB*$(Cl*;qgb)DrZWM4Dk$U1dv`)d#A)!$R)OZ)qdQPB(;j@P!c_AN#suq6+|
zJb_O*Ei%t@m{|1pl(k5G;%marlE#c&TL6oqkM%$gqBP@?ACO)_8S;NkjsNXV|3BFl
z6X5^tQvapK0LK67+G1znWc=Tz#uBMi@F<`OiY(4<ZefQ&piu}cjBxNE2@J#_5%blI
zP*8z@`Fwxp)Ta47XK$mfm+hDDA1j}ojAmb_nZ|?5lS?x~1XWNYUs1|nP!V`w@w*@}
ze?=e@9a1DDATStMpa7vEJUj{oG`M?6iettgBHnzcwnaY<DpF#7GzcZ&LPp<?e*@J?
zSO_3!Pys_lA_r<BBqA^{sqd>l0#d-recK9@{xv`Xu)tyNC{9$Qn_qDXm}+Rp|M-eP
zw&Dap5{imhH)<RLbMS$|T!DK)I*k8FRiT`7$Z()oq<HyANjv<C_LFS|3fGE(fqZ;?
zAo}d`z+m&DF=2?iprY*{QA9aNMeIY+w~34bh?fv=(=z^Pp!)kTPoL4aq*=fD^vS^d
zf%O%{p*B%b_rTo(iGb4vfk!g9fPCDbK3f}S>>&C2ayx<j!32J!UMT;$wx~YU7+2?z
z_m2Pd2n6@&fy}7jK>n&NAPc+|yaMUBEy#;APN4>bLwW_}GuDBQ?5Q2;`)3po14R+P
zhUJu!qMXOvqumA%_$s1(rb9SQR@hUT1ZU!c#oUv9YvjV9K|2g-cjoTi%<u)1dhY`J
zvCSy7B|gZb$mw1Ome`ktVTvdn#lk|6e6t^e2m*@)2;{XDF#_!a`@eE^z<kTt?VljL
znSgx-Gr<LdY$9I+nZZFzAT44CzL6r_f_Vu8Q59j5QSI`>|7;=$2>?-+k;3c(IRp%Z
z|77EX3TOG<FindK`2ed5L_G@t?auD>s<L5ua1!q__WE`HoY&^n_tb~S6W`d4{4`Kg
z6MF*#fBsvi_Qp{m0liZ*J4}(B=34wC*7KWuidRu=Vgtv0>d+l$eAldg!~(Vbz(FGJ
z=C(f~VJ!&(ZT?`ivnrsS{(1BNo>lxozyDEsou&QB9Q)}cVr9U(Oilkt{rX`F+R9_>
z<wwAp^UP-i&5P`d0{-mPnI`THQ^ni|dvg4+tE7O5!Uhv<gT2&SfqGI7dKb!q%cBxs
z*|B_!&iu||#zI425^xvt<D&ucm9hGs48^7!4NEH>L?s>9qkuAX{PrmNwTZm<=rj~8
zG7O|ohH@z$Wx;Gp2?GUWU^)lA{~<jDBnZp_5&{jZe+mwqg)Fh3i=PYz42B=4KN^NK
zb#W5)Zt)on`vd#4)!z$mm<IGG47237A1RgzbcYcE@+&{eAOG*XA2`hZLzgfQk8hCg
z{-HV8@H?@ud7RMWoew<*NU(ia2eKR-aSo|4{xUc_9*;`*(J((30oP!orB8B1zdp2v
z@rz4}{(pU=s8QH4(&3YCn^x0Qm3BWWnMQ30>tz*WktIr8wk$~9RdO8oYBnT!gn4Hl
z`c-A`m2r`wF}s2~4EHl+bBuSSQiP8n4Dpz-)FIg*o(a@k1eYDR?9=p~u{)K)W=AHQ
z1lUJLzb=a`m(z>iG1_5MOk8Z<X#A*qyb-*5_j82eq-blowHi$*9^pYP`ysHP*9<Vc
zocIs9UPdH`YeKVfez`HwCy=5$NM`=?Vi6cuA_BkntXGQcvSo;+*|2vaHNLe#aD~o}
z6bjYdWxYb&c9dDn$xk?mvM%H>kE9$$9pIDk6J&Ecy87u;P*jLNkp`OPE~@}5?#9dK
zpS$1(O^3IgR((|I#TTX$HdcKe$)nmgqx9bgW0+5r-K>PET0v8KlWr3%9k2e^<sF*C
zQ!^69EZu#=h3|GvX~KYV+dEk={?xWUQ1LM^o^3TQBOp3-pnw<h+x7YnEH@D+ke6Jr
zWYwE9R)(1E9RD|k)nJ{l@G~85Z}<9EHG!L7RhGHP<Jp-t>+fpU-XN*@&g4a&064m~
zR&^=TJjKh=g2W~=uJ)a2U8;ie?H;b*%eY2gt2h1S#yx#88s!#)N^{=A%gNR1-`~;a
ze+i1~{veE#mPQvBCZLdOo!3G&E_5sAnip?V>d6cZI{jF(!HnD9Ey}B?MRnhuyL(h+
zkL_KzgeS09YRQ7Mw})S@Ir>dj!V>Z|mB2u%?IqckKC7jd^-*U^CJ(s2w-Unu{JzSd
zkt?j{`S(UhrrTIwf=P{RAWmv@l<cC{zR`o`@r@D_yE?xk)vr(ozb314LsQm3^^2El
zMY$}2=L;OlIqF2@mcx(Yng+o&uG`XRPSVyrln93_EJh+8=jg`<ROJb4=5+!o42Y69
z`%)nC^3rIGbW7t>&pmlKFZvVwN(Qi{g{rO7OI?n5mJrkbplDoRNLxRHqraIi*og#L
zVH(ahhWRWx!S8L6yUiP(s^uNw9n3bUQ-=O=39d7zscM+F_0yw=J2r@Rim7v#Ic;;`
zPC>HLFpY|=C}k!#+4(aspw61gnx$8<&$e<kTq!u5sd;z=;;w|BJfR7yu1N*JF!S=~
z_*SZi+3(CZ4Y~@V%QQN`F>*^lGy;$0-&W4FzcWxm^-0-nw8NVwQ1``;U`)bRCv+DS
zVmAIc?Utcnfl|klNb{syxLvYotvGgOHtJb%vE!rk&i_~CF>ra>Xg#~{etf<U+o#S1
zkh%JBx~tA+B9eQ<8;b}S$%HkO%~Im1FB}@r8<x8?3rg59{8ig~04SNz@|feuCPP#-
zIhvE5j>X1!e4fXXvV=Vaj@tfl@%hU5)ANVDQl(?7-qdG%W6SLbb5X}CrXH>jyd(is
zr&?*tT6#CL2kZu@9=yZsV43pDA9u_;b{oKE;JId$WtL7cn6eJixbD2;>7n^%zQk_{
z>0g+9LsM?Xip;|DHf_HXHFq3gjX4h)lh#dzx&Y8O3(NN7^-`M5mFYBJ5^CjyIlGDG
zJu@XLTDi{LMccRcI99kObg5br&z(34aWDE1D+`+O+oqvrrL<JO)YTm9Rx{4FF(wyY
zNg#xXz&PZ-Z_b_Lxc}?(?)^FJIZKYKGC{4^^?B$t`r^dVGhQOEbgxP36_c3L>}ccJ
z-`Dg#_i7d*<V38O0uPO-ieI|daK!y*TAYT6rN51UTq7jn#%!aiuSj)LJNSqC<N0jY
z!3|1W{46YA*D{ykK`+=E>GT;~AbqR2tK@>xu{K(pycM6-TeIBquk64$8s)$~0e}6$
zDBZ#X5Cv;Pfjt*~TiIOzq8ZcA@z};at;<g$$+(?!1FyCu&8AOkbOvUPr<uwDFwHxf
ze<p(<2iJCtk?o@U-*Yv(EMDslt({FSG|Q<`66rG-sV!4{4u2(vyX)+=QX3TKLkb2m
zSgs%Q*<e-&tTLD*DW**7dn}qWsG9LS3s({9xe1MkH`vj1#<K>0c{;L9w+l+T+g;07
zN$cQxb%y!@a*!wrWXV=7ZW;4-wacj&TMjB+a4Sd>XUR*dQE}b0e{@+x8EcXUt+}l>
z3|Q*%@;VzXh{6VSDD+3c%=W-^b;}Rh*=^VlrX5)wSF)9*Y%JxQE<aAQkb~#rj_n_(
zEKNo}WN<hi&4J|-ay5wlY=Dv!qnaU{Ci{U$1h>^D3uy7l<Q9CkU$>QrppvxG>S7+o
zPIQnh^BCHy`FuY<%pGsPkEod$tG)q-PiD)<ZKr`8XL@-56XH!>mXhI^cn{f2r9~*C
zah=t@hD`(J^!LIohnKE85a=G$(5?Gtd3GR6bsA&P$Ui!U*I4d$Q||h;jVr&($`=aK
z1Mj1h8;__*AiZ5{ke5V}b01zuELGE@LYnE|l#C~HYdyc`AqBQ*dXK;FGD%YHZ=nL?
zEKci6UgiOEtz}Ku-vP2VcQRU6Lu4y^pC*Ql8mFjO>{G>6tWN6&Z2;exW>-+&6z1pg
z_sJt_^&upYkCq$cJQ2t7!K%7v7d>L4@r8fNu$<C`MCVu8pCMJD!1>Q7ka7pHY+66h
zOv=`mzY+;)13b=YtvOOoRW%)LWSu1lj}a^y4)aGdm>e=wh}=?<RD!a-?*d|(<}M3)
zy#hbs1teSIH)DspZwBfFC!SB?N9Q8wD~muWZ;!i{SZ2}$jLNw7qVm}kP~?Gf@LtHA
z#!OArsM5}EuQZluF($A^s2|XH>PN53bkq3)5xT{vGKy})vmmM@ZzG992z;~d_`qfR
zf%v>opxHEqQhG;tl9XD%XdViaQ7m*F+%7-}^*muF$z2RLIDaoSe6K|tTf&T3cdm7y
zJ)Ivgc<BkZrK3w4uqFv}6MbkvGXw}TY-b-)w=@cKp7HkR|7=$fBN&PZzX`=K$RXYq
zpB!u)@IIm)6S3~hau?z<U;xpp*@oXO^_cqdSCrZznl!9!KKY5f>^FD52-0X!C>a}`
z&l#J=11U=0Gq_s1R~I{$GoMz0hPW!Wd1;dcu)ZoVSOQc(=v-(1IQvev&Kawhtz)zE
zjT>ie8%)typA9pF$f?<_&$QMjO4;eqZdygfpLr&Kc(*4qXQ4zNtldi~mT2+@$Y?C_
z4P08~To@|ZK|(eg-1}=159E~_S)976GjhLMpl-y~*&oyg90ngQf4R$#Q(4o094VSM
z&>c3tl_zaC4pbA&#uIMMTB@VnW)DiRSf@{&kZ^>EcvD3fe3MyS7787ACMEMZA?s+p
zCm^i6Ra(E-Hc<0FXSVJr71pJrdYjyY#jsQnMO^!s+PrpM#jNu_(k=#@^7{^|BHAAs
zr`B2#sv=oT8Qja&>T2=iXT_~hxp+^)Xr%eh8NC%II&RE^qAaR2jZ0`^dGT0ao7g{R
zy9>Z?D6`sK-2Sq)6XC8hiP9#Gscg_h_~^+W0BS#8B#9vPJK`cQ9Ip5y-WAmNMl$LG
zlq0Dr__LH?<qWa;yj|pzI|+i#(-HRzsx^pzbZ&mUn3(ub?l)15_Us!VO-x&}dK4|M
zTD6<8(d>h9%b*AT`wDrwrCZLU2YwH(tq*|NS;MkRk}=?Ap=M^#;!7t__Qrim?!`zx
zFMH=X`oO88wXz5N`3*ADy$G?ke-N9)|BQ5|i~s{1?t4HMTs1L4TMyh66AW_P|FUCu
z#as()u}2n(aE(l#1Hq8J`W99dI{_)S!0BSV$-k?(m@r5s33@?LtR`ul1^wt+)OF%Q
zaQgg0bB|#3W~f$vsQ1kDi~GJauIt5|l=wT^E-ad#qF|k_uK%qH<<SV-c8Hqgw=+GT
zQWe-xFy(%wrAP{l?+9D$GiifeQA*ALlGK6T4jF2Fk+EkZh&;Qc;bIOiMYa~l2DPdV
zkd`RFD<9J>I`rJ1v3!J2F@4|P<h+G5JaR6R2??_5XPu{ud&BGFnDACD@hCK(s0tGT
z=PiY0+u>yEI&u=)m4avYb?Zt<W|YU>urDu$dX~aHnN*yW4dmh9#xAon>k8Buq^^><
zOuuGMS=f7-3Cvl2qTFs)AZ^5tMeDin#iUwf#H93B!(?~dTscuw(iM{nI%+2!Fqd6T
z%Uq41sB!^G32Hmg^jw(U3P|;<oXL|o0J|S+Go_G)ET$sJZf0Ba-=i_L^|zkcU5LdR
z3hNSRn=BNJCgW|jjL@a@f+d>`A3fl?@HV7mR?EHWV$|n*cNcN}N``u42W!&{tLvgu
zZ3^Tn{1XLdQgJb@X1?5QIz`oJti6;FXYM;(lZrX&(zGwupwIn(d6^RO$r#*&OUdtj
zjxAjF0`$880ZhD@<u+T0<>+WNs>7fmcFH~EHv`J;0q2Cpkmu!Q|DGp2)Yrbxe}gM+
z87@+sQ>K(+xo^kOYZcIB#IcjC=YWx}%-jVp<2mN`n6$Be^`dZ}A}X>S+Zar-12yv-
zhCXgHVw_VV)6YrGWq)IR<BfFlD^AOeX(OJoYNI<lt)TwXrL*-04}sk`&C1LK837cf
zFmG0(lOo-RGa-CVFV*dSbc5whV6Y6ok__LxFL0NsFQ{55_R5Yd=Ooy=*+xT+@-02C
zFdjTQs^*n*OQy4tjJ*Fc?OyLmeA%pI`lyD+bJdtje=m*3M#Zc7>i_ny2`KiwTg?MB
zlG*YU!rR^<Rg+`LS_cQ3OU!Q$AhCnqNs1NVNt$}8%hpDW1eJAZP3A00oR9W2QV&^y
z*f=f(9UL9C=Mwf%Wrn&)IF%&rBx~f#d3gbOOdRku6&&Z<>MiJy=C)Ry{VCS3>Q5(e
zie5=;3WB{0HIh^aHLc(^ZPI0{6o7Rj2$V+G`1-_2Eq)evFK@upc<_0-RV3i{Uhc;E
zxroejin|q(dOt2z+nTgKAEs`VUTM#A;SGZl)1aM_WeB2=sj#x7<E}WckrWc6<REx-
zYqSL1jGS9(qdq_L8`N|A5Ylubt8m&QWwjNlTx5GSGB27w(=&ab0cUuJO0TK#DFaj}
zHq)*X$_+jiHkN%t4i!6O3IMm66-d4n%ES7jX=AmF!JMwi^e=)Sd?=O2Pg~d@;eu@@
zeL>h>djhNQb|uUop7gTnGbXo&;1`xlqSvv=K`pUbz_a(aDwBHG69{cV?ADS(lsfEB
zg!`*z%a=tAow-@@DN^ol(+#AEipL1KUdY^#oSsb&5iRe$hTU0>*pob+PV)^e3rV$T
zo!H5PjS$K69Ag&Cv8QQ!^o>EMVZJ7cELJ`DSVN`pUI)$ftNpd~=wBwigZ#8KBsvE@
ztm1)ZEicBf7%XdfKMizS`;^7*{4q$piN)(L0}ak57@3jx(YDH(l9Yf?(t*8EnZHNS
z@F8oZ`7^nL^n}Ye(AFhCZI8z=XRfGuyXWk>wn`SKnSpOQb$`Pt7Y^e!zp>PO)z;6R
zo7bOx+B;IqvOT%_C7;kvmpD147;TddEtz`kEMnb%AJ)PsQ1h%`Ym3s4;pe(dV_%V+
zP0To}E{59_)>V30@>_6gm)7`@(P;+i?Jjh$`F9qCKd!3KqPVJhcj<fJo<C6(+1M6;
z<yn}!S)~o2cwEpVPQ;Ry2ftcSnt~JxWpN5LkesufQSiT06Qa*5@?z=y)w=0;)R?z?
zY+hl_zvId-az*iZy+S~}GOPpth=`T8yZ8EHtSrXusdwY)7`*?oXKh!i!6n>8S@MWp
zwsxpOfN?glHIRQ)(<lIZZ9XrS_A}@FJCLS*$^*8ZVrId!vGWiim+E$@B_^|Nt-Wed
zLvY~&+U#1?%SO!#*(>HdtV@TlFQH7|;IlV^UL2`)pW=qqlM#nD8whQ$m8K``9t_Nf
zH$Yf+h^L?Qf(TCi2Q8=r3B0{Kw=3;)2pe%0u`i{2TCBbaVu@VP369<l$&q|-O#7xC
z(>}ipe+E}i2&oN^ELX@|G&bK{LOHzdXB}Ldua1D8RYdV@o~b>!od`X7=kA}jb8$MX
zs~NsKQbl-uEne@W<8y+Xz;0xW9=zQ#0;8#X&zq@#7ykUspJY$<y`4xb)m5?N6fS4J
zO2RPnb!t9k4`QFWX5Bt@;OxMNZ$)ab@cN}#6u+FC%I6Sgy1{jt4=Si^+6CC+^A_>c
zPD=t%(YFpSsMx1f0Q+UQ6!w%U|IWIy@|P-#kIF+jS8Q$k_$WU)uZ(*LOUS_3Nz8hP
zWC^uTso~elc~F$=Zcy0r%yz64-`&7mPU&}T@zP`&*=N@(p-Z2H#%8~DC7g<5azK!$
zwl$BxPpK01E3+ahN-K%`iZ4j*oS;9}v6#{~^!TXlWN|j(VyF>?(7maeEH&~u=33{n
z<%d@-bYWIn!9F?ADZMN5E%a6ScbBVcUH)E*+3*U!!da30K7M@Ud%*p%Pc}>fDZgb3
zCZXf?5_VG*yatYqYOnl?ur+Ghd=k;Uc`9Ux8@6FVudeUI{Fgf)3JTXI$20LZ=}`Rp
zFX58SIH$7f9#itNKqZVs8sJ(i^5L{|$C)Q6IVoMePI+eU-LJX=u*1TE!$i1}%gApt
zkp+f34w%wISISnK4Ni`BrE6q__Xr>@b~>Bo)q@c=qCSfLBI4QF&Dt*(G@(F)9q}EX
z>)jYL5>FosZ=CwrY%lMzPu$U7nYOBK1ott^JBT1pw^l7XK`s^9-JLmLIQ$lhG^(O`
zgwtY01c-g~A~rOQWlGr4$W&a-*Iei%B4KGY%>dIFZ{Gsi=rdvu!q~z&R?5p4+bE`>
zWO`WI`;Dxk*oS8EfM1|H4oSm%YSk*Gc*~-`x<Au~63Dip5EZX>*gTskH#w|jKYdY}
zeOyf5L>Ku2RIlQ7I_#{xv0^M#^=VT@6<yqv+d4qn&tB=6WKpbA7H^-{NZVn{by+$e
zaJwCiYRPUaK!VK0Dyr8&D9WMTLS#$7I-k9|RTr7F9NtS{NPMwq(R>bn`{2+Ce!mVh
zF&<2Xm0Hh+Fdw5fkCrVKECh9ixEwVu&eebsY>|avN9RB>e{+S98QS2v+y$Go%nJ<m
zm*HBSHmdY}f%?t%{TUT}!{;(m8A)OaQ6!!@WojK+v56Sygg)>ZArr!Hcnm*Ky~(nq
zNMvYYt}@%1V@zcx_{(7{yAYzCl>oW#C&bDR1n*KzQ$t#7ST?B{I#F}_(bTo^+N1$e
zAC6E#!pF~^>Guu@dOZ^DFmO*^y6lRws0=|Py*@a@_7LADjTwe1u;~ufAUFCwK9*Rw
z?6!J$O55^$7Zc@1k4vG4d!lNMA2H1heP`B2b5jqx?|r+IWqqqKIz-MOzN&auPa?fR
z_XTSLL&|9p&rpP(4M@IOg%t~fKz`r{(vN)8@(aX|d(jy&HjiZ7yg*z!F3nA2VF2i|
z{<&gM5~*|warx^!l{43ai+8QGg`R=8$cSWDOlP*;qRn)3RI7YJ(`@rc(Gu-J!}(&0
zn_^bw$fAT(XGaYazILaNk_I;>)6H;~oAk`9$6KYyL4oix$#q}PoUIGLCon06K783X
zmSF@F2NS)%S{1rZa%CS`(QSOdg=+@Tq;S^*h(NbqsrV-qz6zg5T?5qxdupS&ugB8n
ze{u=qk7yMuq)pgRT(F%>R+xj5mU@CqM3opS-Fk@mJmo7LjGq!OZnFFucwhA|MJ1S<
zEqojD(Ubor7DPNVF#F=oo1+84&|nz0c$cya@|KMM!sX0an~Es=7VyBCY>p$x77Z|>
zY{8lg8MDyMb&v)dIu|#RvNczpK74^0#?kuvd_dn#{Ed$L#F}^uc8-Z?S|ZfSbK8nP
zWT;aMVUe7j6DzqW%Oc1bh^BKHP(K(*c-_frgjta51-VFTbFiy-r~7{2>7>;tk}yEe
zuJvnN+udk0LE$4Js;IHWF~43!c#`E%5B))<Ir<Fzpx}!EwZ<V`uf~$w73E^~0X-bR
z9|jf!HdMjx&C`x+ec0Wq9#zcSXMe_y3G^4jhOs=y<RDrv!-skBiQ28qP@8+SBu5jI
zhR#A5aT7Z=+Pv%~sIJPaRiMQb)q<fT#;LB$wo-253}yQ;?l@8ow3smy@-2?Xk+qhT
z&g1U9cxyzqxu-ZEUs71Smwse~S%tPs^ITkgr!;GF-=SCGT?;cPvB);iq((f1NSB@-
z%`Q5+K`^nRF<o-A!4j^#LH>TQD3zzp-CEFbOV@Ddyt?lq{~_o?4SS_a%FfDTVa2i$
z4~BP<P7CTK4^T#quV?+QL4MP~lv47Fvl0*!uT>cGG?;5-)0k1T%U21BZFF5g2ytXh
z`NKT$D@zZ09kspe3&}0+{m7XBPyE6W*a#-0PWFgiEIe#aG4MoeoC7g)o8?Y<lbt?X
zZpdWJW0_Xa10GS)9zrTrkrQg_u=3n`j<G0G+;MjaY4M+EnWhl?6#-3k=wigR8X#FE
zkdu^QXYa;ES7OpZA6Pw~ZFX?JeJ~qPK2fZG_;v2J9I~d))&<s1Q9YV{`yxE9sUmh>
zVuE|hBf~J(Q@rKj{)bz^uR`z+%P%~8>W1~LCZ9v6(2J4wb-bIFT}{jgS2Q^pdm=ee
zX)fT1A~3Qh0d_*&Vf_3DSe!?f<-f+Ntp6=mWnullxL;->R!+cw9pkcba4`LkV^uY9
z<prPKQe*`rXkc`)sF*qU`FVK70W468x@ac-!eXtCBJg-Q>Uc#-3t$VlI0SI$aPB9M
zofE#B?vqWg?~}`)4<By6+#gQR1{^F{Do&!y{BtM{O#yWzGEjv%)&wAbfr3DyI4Dje
za}eQ9p&yCy08@n|NI2lsH#<*sG_<IZapivVtaN%1gunOB;82ku!Q{rl<R$_{{!mb4
zUkt)vQV@AC&w-pk&fvhBP#lXvcx8bChQApp%>eY1y)i(xP>2HRnwkket=t0R@DYK6
zefIwJSXXc^gC_sM{*L?y`#bVY3f7v%f)kyDjB@|{i~-i^0Y#!92gCCTWQVnmtq=C(
zU(l6@;IDTD^4MqlcPAP;0Th^GhB21k&?uO69XBcv#2!Qq3k>F^kJEhsKaUECp1n_9
z2{JD>c;E-?;*)6?@{<+Ezb@~#ckt`ziv|hqdjlITfM_#Lp92HJ4TNJrJ0Hl$1z!*V
zF^vc`i0_LXsKblI;a7+GAMWoc_9s0aoPruNn7_l<*HShl7`Si&FBq?WO)nSPJ30ir
zF>6~8%)m4}jJQ|aPlY`CKk%=;(kRTAZY3<<9{kIflWTB4KAvCm?ru05bI9N(Z+=FF
zpT@4pq2FXyp1D5<6j(x10))U8umaqW#;KY;DlN*p7WJoeROQYNX1D+do&%l0mqCoe
zX5{7y=p{IK-tQa*zn!0&*OO@Ne?wfLpgiDb1xl=}&zn;Yrp3Ei{8kS%IM6zfJ%36#
zpfAsFpN4*eNnl{N;cuI7uXZ7viC<zb(XYPjpLWR&%pjn*=t#i8U?M|;Kn4aDzx~Os
zcYkK7AoN%DRQ$i6mx8%Tfad)ohGoz6z`eejeXe`|^n<>d(ei>J)GGkJ>;Sbwga;9|
z5$^r4Px*Cz+MRx*AN>Hm`Yjh;pPj#JO@2v#@i+bB{?a@%@FLXanBzUWAag+U{VXiv
zy@%CwC_tu_ZunWU&gtVpChf}A{1A{#3FC(baAEcRMLOYEcq3*15@l4w!n5?h_5XU$
z0|E;c%=`Jfrw;DW?*rhmcmJ#o+?RNmO|tw)`rX6q6d581(no`dA|I{BazllDg*h-k
z1seVoV-%1;iV--_25Ni$1df41d#m+{kOU6)&Z8P)pnl^g5$5?#g;(oqN09og_p(D4
zD#WA!^)`hDLW28VN8fbSCU%z6+BZ@631&ac6C~c}2JtlUkdy#OF6H8Ohkg>KZtnef
z|1@L09_(zSJ?Y7$gD*m^9uH;kd(U2@z=^KzBKu*K+F?;_Dp&q_Mp&}}PVB7oJ-PNZ
zjn#(Iy5(;2f4k06Oo9jjjBeWj7@zDWkGDhQtFdZSrOlPHlA~dV(nmD;e|aF%*3G6H
z>yrw?Znh7286gTkecz4_MR&zZV^oXqIsR00%o*xUn#8{*O6G>Zpd!c0i6MwdhjIsp
zv2RB^y#J~#w%E#g_R;MM_DtoI{INVO)QYIPz-gV0$eClZ{MrDO+x-K_P8x(nTx}|I
zsd&*%){3%BhEku^?NLo0b<94F;f3h?MYdYpQ)!I<dw7UtS;cF8gZmMYEmd_CCls7*
z5sgYz`fmeHQN1HgL(K<s?RZS&Q?yborZ3=lNzI#XD}}!S@ZiAtm3J(*?La%ZY7-C%
zwVzu|Ho6T|g&KsG6B7pRS)oA}GVrAfkD^?}h$XnzDbzD^)6>O-xs@Pvl%s$uC#||C
zW>aoS%K3M_;D#j6pi}>A+&!)YqB7p6EYZt8CjmVtZ<NgC{Lxr?C~ip4^vh-LNu^f_
zJ%94^vn$k1hJWXUvO^bN+84F7p_RyR1skpQfSRjGp!qXF*Cp*0hGD5l%C4N_p|XAo
zJ@&<ji?}3;9?te<qPC)QHk|`O-LVS)QYGfYWTuuavTLwyM)geGUCwAdF$=I9{CO*B
z+`{p+8{mbf`DGbe*}2+S8%b=Ky0b%*L{hJ<mnO<0L{B(?0d_J&k>;7ZMTwJy@aLO{
z#~88TR(rZWRfQpOpG~2a=r*%Ng%yIr;D})$d$%eymUe+ejK$Gv;?ZE1bkMI#l~2BH
zl<*s!*1|0fqt()ix>f!lsnKrDp@$RTT$oP_D%FsbiiyRMj(M4bLyabIA2%r_>mG2a
z+S`iim+NYxFab<HU?U}{Rw38|p#t+qwOwA?y)>Sacpz#$Bxd<6$X~&HxIc7m%OM4u
zf5pd(Cs|y`hLo?Lbsc=FMZ`n}jM2A-dJlQ<OJ*E$yr9~WZWPs|Q!9YUt%wtfo33|G
zhD+Z@dv5Bk?BZ?l-UCiU0~Dqn(6lbt=XS7rF;H9O);6b{dj`Ig=0lL_B;Ejdao#O$
zos<T@o2{Cdecpzra%MnV1;LNRHxQ9Z>uk^2s*gWz#amW83nRsgQ;uAF9^6;P$}t3)
zEzr46QG*#g)_B&*0>c)S8k1`ZH8t8BIwOh0x6HU9qfz6Jse_0$mDozG+bUBRSxmnv
ziEyGr4i#6VZZugu(R=UChSBi8H}-*0*6yE2_P)8eLIO$8p&vPS$v9R^TkmdaK4|<x
zQlC^Nt`_T(M3(+tcX8lz<C5OJ<Vm__?`*OR8G32Vhg&+70gp5J()4A3xiW)Hux&DK
z6<QMO-jG>jSn=LBmzzZ3@70aQ$m?JSaRc7EHUMb~a!X(`^2&1EpgjXQf>qcE0dOoJ
zr>TjtYGslTi)i9LR<1bo^$!4?ZU*zB2MOECMTk0-tXcbJPnWr3btofdh<DihjbqE)
zE-P`goyjemyf8exrEjTOb|gyPzW~grB{H0K5xPLz;Q*w;<ff=&)Z~vW;MBuIhI&--
z_tG~A{A3bNdyz>nUjofw|8&VHfOiyXk09*`-8eZVkhQ$93H>QJLpdbqvRguoS6LyW
z{bkOqIR07T^GjgY{k87Op3-OAEbY-msehH~frrS;>0;{5sS0w3*^areT)>re5+(U5
z)9HFc=e|M}fyZqYBS7RdZxBH6e8M96yMyIS>Yj4n6K@tf7=A(LS9#W?+Ro-}rw;yW
z#ApLz1GhZ;76$1}ZX6fM+?)Sywi`0$;zKv-UH7?_glg2TASHDs<i&W>ozr0$7Xxn>
zjnA2S`Jz8zU)AQHy5F9f8#b*k6^h-{@nugL-<~H}c24peHM5B*TdI%IlK0Vvcc$HJ
zt#lKaqG$m&hso7;?Cu32P0@;)AYXa++)lznCc)sppI=+@Q=^Fvn*Rm>?v_s(fA1mU
z+kT~~gfvwqV#B|4mM%K8W>39nRP#a3dXG^?v~r)lkT`xGpZv_70s>^B=t6VI{KGPa
z5^Xmw&J%FLEIdp*U8Z22)}~jrOB`$zj#R=^@sx(iE195^A?PFQ!s;06Q{xQ{ssWBe
z{1_N75Q&>dDgX{3FZP9u8?GtmB@}jW?;AA~wj>x;lv}Nc2F#X+@+zM{kmeE+nO?4@
z7Czkyb(qaso~-=4(zW8^6dSM~(V1*wt5Z`la>}{CPCFXqQtrrR#X@3(uPZ)FoikwB
zYA}4hnwugCHb2L`MPC&;Yvl1>SH@%+dS%|(^le`XuMfYh8hrP-JZidwt-s5`Pv%#4
zgtR>gX00nYDoprM_*GiN38MQ;`m?`md!5=G6O)XU|8$c2KkQqT3d+_uJi;bR%P3R<
z-p2PK){KY-M@fCpcoe<Qk-uoxwNN*>;IJP=bXzTM?ndW-MU95x4{aZ(noAJtA1GZ}
ze8)I0$2StI>?f)(oiBS5f4P{{kQ#*1D;%EY<5k90O1v#RUNhk0@5}I(H)qSGn!ZYQ
z8?&pqkDFsQ76Y7`#fg?7*R<Yg%{a;TP@%uSDJ*csoniU^4`b)hBnq%@*|JaBwr$(C
zZQHhOSDmtL+qP}n=<~Y!_UH~eBQhfM59A>C`qs86a<Z+Ab*hCEgdi<)gJ$H;UyL^3
z*x3|CP)vlbav5X^7mh1J1@m&ASOz#2>?Q<GJ*?^Wdd>L=6<<%e3>b8ticlom7b&lk
z^Pr%i&;mD;9d=q}R#ZeJQ6#3Tfi9iIsFxXb;259+6>gBp=a%WN{5J;b^@STTfunJE
zTEd!Jwvp$hCIT7@5)m9gJMDT>8ql1oqJjM2mQG2_{5jwWtu3oE)8bZa8F)*H{=u{v
zR*t5QU^C}U_xwH+GIh%A8!~FO(bup2OAzx;q3cb*He)WW<jc5fut_zXdK|>^nC&5Z
zp77v!X;4iwO?5w#*lJ%n^-VHkSAumK^J<Xdm52DOt184D#wG8i%f`HDGsT=orxK!t
zb$eB3F`|}G<*5MnZ#F1_xLpgsQ9=~Pns)e`7V=yEqfy~a;0=DPgE9QZE4w9Q*ZIe^
zZEc=V9P|tR{d~PdtT$P4c4IR8sWv$;j8#_~RUJMY(ptv)g1O4OTa!=u+cXY9w7#>h
z&{p*l`l=s(N0F$dTQg(YimjsXcf<LT%fl|xRjhb~{}%_Yw~vvnx^$c0VI<_DC}G<S
zR#KPQF&ZwdawF(69CRwevDGBC?CtnGnTC~;b_;4Em8m7)n%sfeghWNfK>qjTd^3@=
zU{ed~YjTGPT}G>qplOGwBI4t)6GJ)^0kgL5?;1Sjvgj4H7bJY0;lzr4R}v#`<gATa
zpH~CjQ@c##%&M0hOzK}a(@&3gQrp!`QoDB0vwyZvHG4}n+UC3anky^t$-q9J5?<3l
zt*BOdkKO5Z^wihv<U6D^4v22fZLt6l=JAIow@C^BFm+9>+r;_j#25VCvq)6qG~t(p
zwH44}Sb@K`8|_+ww4WU92p(%4!%2OiEmsc=vGZTm6ZzYBzy(@7zO%%djNj_2j7T52
zR@^1a(sm7Ahi3=A&{uwF+tSwK8G|WnG{47@<_>3GUWC(#rJE(~2Wo=(0x3Q`r^S-B
z75BYIpSyc}RPQ5?DY{@I?#18@r`8mQ8}+(8xZWzVJW>xl3GjJKuNSbFfzz~1{U2AO
z5LRiQB1Q6FEnllD!rpATFE9)e?r2nO1IiuoA9CcDcJk$L&c&Ao$~K`7N9lI1j|m#;
zfxP;mOt()q@FMCOVA7za$=;Awv%-|Peyw(i@v`nk-bFvzc>+Ybh+R&NlhH$lmb~Yu
zq#rNMwgY2$9Ra%Vpa`B$FQ)}4`L$W(I<b=U>&GlIo@}*bg}buPY#Zyjk4cNcWA>-O
zDy&*)Zc;7=GXLn}DmbSqO9|MqQ@F!|x$?tF=Lai?l8#S%Dif^^-UMc?0rrVb`&k%~
z90y-c6(^+)H?;bHys3NiL+(%cm2FHpq8ll_<m!z+E<Bdz&k^uKBF|3r!E-kUBF}m-
z5c)Pz<3V#@No%00es>Cf9nbwc4e<Da@0GF_@<P;l@sRP$c2fib-d#2E&dfSWYKO;4
z*@gz#=pmR3rGL`M0i$lLfLU~S0Ym&q?L4mf<NE=3x4l#x(l$|HUw#Rn{@>!}N)d03
zWMhmx?H1p1Z#jPyI)F{qnec3HG#=f6Uk*NIM~{jPgHO!nxH-}J^F;Ic2ceE3!X4Qg
zF3mSExgAG)6*Lt}9|jIv>kbz6Py;J(Fa10MRD4sXJjVtXmh$g2*-+<{riiV|R}Wyh
zCN_w?(C1@`MJ<aDQ}k@gVET}7B}@-N_I9_4Ge2qnlgTc}vdfppN8?xTRgqZ+;U7Ai
zfWT|v7Z3TzFPw`8L4+~}Qz;Za^~iUNcbXy9lMS)m_OzT!8LRalHF3_a{jU=+ulFVu
zUO9*b!$y<Rw`jt715=^bSH4L&1@$;DUMm$@5}(fR?PRUt%CI@D`yJ+Wp(hi7TuWU9
zV6bi<PJ{(d%xcqk<K?iw15|nNk(JOi@D@-f<(P+Bw&oK>6NT7VW!Iy@-AGdAoRr1V
z@pE0L%0cp32a)7o0JV*J-OnQ%%?~-UiA=%zk5jtg4bq#tCWF+Mh6*WDOnHidId--=
z0+<4l+VzV(AeE)P{nx=LZY;6uapphgn>$jH2Je>X!}J%L)=1>1A%g^rBU=*uWc@wt
za2`)d2v$6&LIc;v%7hLk7#*8jTN;0HUJnEhPnbL`loCP`m*WVYo@ppqPG{45$)tA*
z-<Cr7M+)0yO(T|K*goHDGI}oXvQOD<uW)RO4T{ViQgaBfmWSvd@O_xGRgtL&qvd&S
zn0!>m^HqHdcA`bb0!`@xZ>}h4XJTgCF`_By>;E*omb_%IXcxw?A*>o6n{dKvw|f|R
zKiWpzR1kL}pDv~LH6|u`UlF~XW1W*lfs>6N9>}m@{%t0WO1k|Nng)`9vyI0~DYbjd
zI9$fmVmP;Ji&!ARY~v8FepE_(EJg>9VRX((yUMag3GT!;f~&<EcaWd^<D&sV_(|>v
z*)zcLQ<%YfI}iSDnDYA0=W4}!@kcMIB+c5GuJRL^ey$M9$vvINPI0q*E&odB)n>!o
zx{RcEF1}@^w99$p1T02OI%E7~r$4w~CumMuvVs0jz>bK2LX)l7T%_Fb#A`kZu0lm{
z70z}GUeWL?rIQ3hYX{wQn|M+rXge9VrL0LB_;;$c#FNYesguO5U98k5gB5(3F3%x7
zeI7-i=2hs`Y|J=U1>i2)ITOPO5sq>c+7_($N4Qk!&YHLNU+7vd1{LY6MU1~W17gAR
zMNCcvucXx}9PD#8B-3tKS*TdtRE_jjW_0ZDOB>nC1zxIcW{+3tt*!PS<xgf8;)G9-
zLF`-;HZMqN8_daw{S4VE>WLlnGp9tLXo+gUeAty{V>pL3#5ul9W7BeZ&(gmM;cUYl
z*}?&w*okyg9R*7VPLtM4XDbxjSkqH!Sa=t2(%PH|Y{iv~C&p!|AJf%xgx!cBwMhCw
z$wTLgs$LyAJ^LIR<h;py?N|NW6oWJM7d)tR(GCZB@d)D)<B?kmaaM(3zR%gpjhp2g
zBU$pggCL%GI{Uh1H*m8D*7@G?&pzbrWa+G;x9k4~LN%|oZRf)k^B!2xcx3}`I={7v
zj4<|bw+Z&E_gfmOi!xATn&xl9wpxl$r#TomZ33AXxw-nErc{lLr6Yx3(9T)|J-ued
zzc_cXrrMt0+}xl<JXPr$r6}t&QhQG<U|8yNTZXK6Izt8Dh=fmWPGh<;K%jHnuX{a<
zC@%~%J@uD4c&A5)dNp*}o-;E70{r6dB6){f(CsM#gVIlMSJyW+Q4lTvV*NpjVTce`
zGn$3!B>jGa?mp6T{azhx)KDgzY-p@^MMO;o>zb|vkAqB7h{{nc;@8v*Mkf|=Z&ulL
z*K1E$7KnI?95OAA;g~B&YA{UvrSW2@tzQZ_QxsrREUG*X*ct4@nV_ZW9`7uF?@Fcf
z-O@-gadXXGP5qjnCDf{Jm{UjpYE(VSY@LW1UA~X7Mxy3X{iD9@Sk3GELh#vlLq0ht
z&ZH{=5b8-G^ciJsEB$2-<vK@2uA(WJB$gBt9mH}!T{QQYgr;~M7EWv%Bv>^sss`*5
zUVg8{B}vJVilnKkH*$s@V^*j$$J(<OuVS0B=P6g5Y3oKfqOi(>6ke)=IqYyTNbx>u
zsX_di<Pl9cN@>{FfBPo@g~gfB8<u=9XT{+|P?KQ2X6P$;eF(`fW!F#e7VJaEQBZK#
z^=bt2cj=0#t@?I1lJ3VL>`{}x5?gYWwcT`A#l-q#T&)7X#MiLHf`{_OtQ+$F*)W*`
zO83jVBkFSiOOGpIjC^x;`Ccm}pC?^lZdl&(!q(beYuHy<uu;jOv)KN~C?mhlUj^MQ
z(aQSC%meL}(%xk_SHF~gIqeEAN}I8PTPcdwklO1j3z3b7A-<NANwRq_J_Ed}7@RZN
z-9<sd`aSolom!4J;nfU$3@xrKIx<wV85ks_c;@~T&oVbV^UROFOVk$KQDQVPD=_Dj
zR<-c(`m20Sp}!fq9+?o=BWC?wQ@>0@{v#0QcjGUb<kFFB*Ld*l!ZDQs=l4td<2Lr@
zNSk~6^!^;=G7WD-qSTj&S%*kT8K;72W`I6sdmI&ge+<v9moy^_(jc~tny_*^9!m~i
z5?$>UMV|hOsPa*p{n67l`T1T-%mNmUBv@TmrILN(9AgS@ecW%AUZE*Ia#Z0H-IIU@
z<;(7s+3yVOI^R@pJ>NrcF<hM>jIM@EqzEqPP@4(P8*$It!by#hd=T!x-(<6BwJsdl
zuCImW4JTY3nh;Uq#q4t;<&GfN**AgEGZ1uKv8rEX`YgPxMXYbKOqC<SW(3C!7fUdc
zF6u?8>`LC5O1bjP2Ib8rLbNjwBwv_de)J4wFPc}rgs>KE&4C&?yCqy2l3tgYj|@AK
zQR@ZZ>PO(<nURb;m_{B#)eZOO{lYqq+K~pl%mM4o%yrUR3rkpEulj>Y?Cs6Ty|K$M
z;;GSGs@o@?!HS|nbhF3as@l91zfSce%*{E-y+|+iSECyH(}J+G2tkB}H}1O*3@^vD
z(&zyyu}%a@R;^=f5bls=ty-QOLlCMW!XIZ4lUnwuSM`U}yY13QZv)c<hNLN?tX5uS
z71VgD(NKvd*snEF^{yQ^IPb@NMebolNOQsaSt_z^?>mW5`P*t#1RFCU(R-4!&Y}hJ
zCq+m~hlU|N8S36$v2sV7goQ7Li*2<0?WJyvB9lu5#hzU@ID|LNy2V#2F12-o?o!%k
z%`S=Sht}0{yNIg5oF1!>9@jaA1*r>2;>T99u0PvQXo_lGB}>*uw;4>w%e&8IU1L53
z9Q~5Yw!+eC>qX(PkhYql=iLnzAR#n4=iu|<8nY+}RVrarM4p+uJ~N87`R-rY0CcC`
zwsHg(@s~8^LMYgaC26^Ru${TLhLufkMGg(TMJ01VIxRLUowuX%guFfC_vEZ@sb?i6
zk#&ltpYOtD;h{dHd7d9P4=1C_?!wDRhcl*Kn`3y@e`ZFFl1f98oqC>fxaWa=Xsh9H
z(J?=HoH0be;~H9-V)iMV*gQ$_INk!c+B#+n8*1|S&XyEsw+DS2Adi|7^8Zmurbs`*
z9xoF&9?z152ai%kJEbgdnFc%8;u)Wq*$VzFmmO41XtWW7yJJN%re7+@7k<(=(+6Vg
zJ3LSa*%_=}xPLR{v$JK(R`D7GQK<{i<k-1cD_29P%eOqkwJl*?d?2UQATIu?QeL$b
zZQ$yR|G@Qd!Lx0Mf6|g(XBgLWiIDLi_#slq6ZEOprCE!S_$G0sz-naCc)-=RfL;QH
zaSfY%p^aDZ&;QqCjrG4JYb^BtFH2*@XJ=z!`EQzA4mMVn|L<h22~;`x8kLoLdz%FK
zpW@a{$_?g!J8ohA5Ee5pNCpK5CyTp@vx$pKwE`{N{k`>?@tS%6ZSJv~)|}n)((}6Y
z8f~@{?Kt(!Zg7>yibq5SIz2i*KmZ=Ix`+gFcYkaDx9Kr3p8M_5=JYhBMTOv}fMG(y
zzrktYfC2a8I6?8h=B1IK0Lwf&0NmdJy95op1`N5m0eEq7=lw7Q4GaN5^U>B|^T$Eu
zg9Q0;)1x2c+4RuY`VoHI|0h>N)lvnhEi0oR#dGtoqJjWx0Tl$u<ZmMyd%hRd%7^BM
zvxElg_xYs)i3%pdz9Ai%zP`Sm26cYk3mm{MQs)5J(QhN-M>_%x^a!{D=$!;L2h<Vp
zgE|~A207yj<oQv&6u92s;YWwKE3*gO48+HcgUyBKf`_?tXXBSs0nWb$3;Z&vx=#ZH
z-@9D_aBp}2#ksM&-3{oA=bPcz(kiROf<J%`We2$Gr^WYgNd!aN0-pf{5WMom0@C5m
zkNe@L!(Rh6e>A@%bHL}9Pz49@uDZRw;I=l<XKMpb17ACK3)$)&<Yv`CFeHO>aRC+B
z!%gpfR00^}Z;8`+J^o=*HG^2fI`~~($AzhH{M8A}DaC391i89_R8)D}JWvREkDCDm
z0f%_FyMu^&0OY>`h>xo#dcEEI=&{Y`!@sqAHu?$pv6W}@|2Bfm=da^AyN6rc1wI1>
z>NCjA=l}f$`Oz9UItE<}u-ON|iU&pye3yF?V;TFQdv|!CuY>E0K231%0?6I%^~<Bi
zZ8rfAX?y!I`Y~(UdT6NF(3HILgZg13CIf#4d~<q;0%Cs$@c`)J{t^V(9~Ij36Z3!t
z`ON{H<?E?{_af(isx`*ze5wuZ{>kN?+>Ppodb?HEoD&-f1fc$f-^vz?AAPwW{QQ-E
z<fHxdyZXU7{uR9YofaEk96pAdc)<Gk&F5FEi@M!?Bhus9kK0-e=CuVl@{@_-_p7cX
z9fQ3(|L85F#C~f7Zh!|*^J7D}`U~;`xXuGUXoK@xFrnuev<rO#6zZQV#M9S{hYUbQ
z2RHkxAE#w<`SoV*&i!x?2jZ!n`P-w2U<lp(*{$OpC<uTL1NJoRjeTE83~_hrqb<Z0
zKZ^$j+}4gS|5gaSSKr0&&x0KL?v+dE4Pd*yThU7l2+;b|OR^)@;t=#BgbZRI^eco6
zYTx-o^zZ=CM(|6(`yu#C1I4|;+vEES$hUL(f_GqK9_Lpj0<h(azovKdt5Q=<2J_N(
z<#+bB7UehYr;7?d25ub1@rz<fg?plmwBo2iB6U37&OymB+-C8VnzHDJyK><X6O}rk
znSs#&^huJ%I2Y3Urju~2Z5r@0<JVh{dhMc=DElbWSFI9wGi1tivc-O$Rv+lIF4xVB
zzLfpJs0HYD0~p#rBOU{BZ5iEulS|k(`xrsn?Un4q`<&{F<@@ZSWLBvtj&s@KoXE6-
zKk|>T$Yhu$wF=F~MVXNZnh|wB2x9a3v5KTOy9!XH3BX~x!%yv<+jUN%yRjUxE7#7O
zEetxq2%+Hkv%s-`2a?pgC0l{HUiBL`Fihvf8-p0q#KDej5taR>&E<`eqQ7k}?&H?H
zdgqOGD{2dcADwXlJuj;0BEeIS60Mt6)UZyO386cE3sbOe(U_)X`EIgiahojKjnDah
z4oFWVou(X<gmcc*U`)LdFrAmClz-yGt!STqr>c^(WN|%`J8+hIC7bf#Ke;aDYrZyS
zDN)DSvbLX<aPE;gPj|hE7KQ43@x=SR?c<byL_hmaA~=9<QJet!_nWGjn(=jvOi>(^
z?}&O=Ztaj!>aLk&jo6x4@zO{5nP^%t^}yCycO9su2Cj+cSN!&=kc!<Q0lNp&T(qr^
zKhgB0C8f?H#4IGl^^GV`+;kP>+CKLUZJYlvW|Y03y!te1Si|@R0zn48nj5dvZ$J&t
zvCGW585}T#EvhzKqw%5=xd!e7tnU4otT3?<Mo}xGOaY}UtVY%~scqM}R9cuMwHHnF
zL3-uTi~Z437xNEqpD#lFG4wA>k*yeCn}wWNCJQR<bhFy(s+v2qcg^hwhu+PMiPSMQ
zlbiT_AN7<nMN!F+=IK=0vZqI&_5DIWtnlBWOjQV`y|03og8<5n)2)5^y_Amod9iHw
z-<kLP2RSr}a|(6?t%1(j8!I-yuI1GEE(#wyn_PzA3Bh6k?1(O1#I)F((exm0q4Jty
z*DT|0B~^Niw`K7%f?l#TZB(8Y4wRnw6<cXApX53hr%A)a(`eKauWllvj6CbTa@<%i
zyk)x#>y53aavb&NbbH-<Gg^Z0LAYr4ZvVEfS~go4567$bfJSJ#EIV3>^j_sv<ns%S
z6Vyam9=wk)*f-5{?^dXo;%gDrKOWa~+F)BFv59qcBnJ<6*9OH-j?_pMOVZ|yIv^Zc
zyJD>7z(gFYJCW~p9}~PDD|?i#5!i_1-014eD$+;B7Co<Xi^R{{hqhSB8@F?vCt-r4
z^=}>#T<ecTxtT};O*2gN_3+7e8p~)VQ#KVdhZr*1+pIFbRjIJAZ^`EEATF>sMb8do
zu_-QVTx$2AfU|7;T8_qeYhoQQB+h?0tlwi(*>a!SY9Ys#QC<#m4~3M-G_7mDh_jz{
zHjIiX7+0y!Y*TlpeUmEew10eeUzuHNXr~CM-@ldHl8TU7>|Ep<Dk{%da-L%-lyU`d
z0V8rHN@n1Zh3nS!f*3Y?f~E_#i9y>cXW~)0_CD&LaqB8gUP)#NQ;NP9@D=JBs+1qf
z_^lB=LK}lEEjDf`s^N77Q8^k02?>&BUMZDS8fOtjDOMI-W;-7L(uHfEv*x}MM&IC3
ziihULjvSPDY}38W+u3B-zB<#!v#P@RL7?K#Z`?DpR42_0EY7(sa(t~n?A$jH{%gII
z@OIjLnF1PWc=I66Hf#hs1Vh+0jUdEci=L^W*RJqjv!j@40=Rdk*Z!`1`azUfAPTE)
z;2-NECHI3PXkyI5d|vu7J=U?5vMs&rqs*vjZ_Pv-Nw3+gt-#Borqjg%&K|CNAT<1Q
z_hJq3tip0>I7ws`;bFmQTN(Ly6*j~Tf`?#ia~*w)&@jQs9DY^3G`n;O8+USZu+#>x
zZD<4>*WrDRrcK3pR)vll>0?bww*}HXb+hj5!?Tc>cOA8EkLQ~RnBJ9yqueV>`q27I
z-}`V-Z9j~B4f?NBJX4v}$_#Q6dw`ymqh6Ly)HtjdUZ${NOC6_@yWQsmEo}@o3~c=V
zwNuXUhW8mH1&!6<rJVtu;w1ETbeUziR1l+o?=xA|2f^}*mJU{w)bB5cr^?4^t|A!i
zT9Qb7bl>6CcJ_16ZSn2voGH)ioLjd11FBEsy)re8Py^GBcyCc0N<#IGZwNfvVoQ~;
z+?lzGmI-tBSr5xX^O&J>1Guc1c|P)(z<qeDrtKtUA>HW?R^9#^Z@su?<cH5F9gmlv
zQ{^;$(d2&j<f&Ao9ieIr`2=2e_K5dU%G0FR4%U3@ou`5sLoo?P;yB1e1Thv0+9}(g
zJ|_j?>v<e(TLrq*-a?NlL+bLywuK=h(2w+-*LjDG2`=jK6Kx-=QXF0}&LsF}Os|SQ
zMuSsw&XId&%nDut`X9xPZFFM^ZX9H}k0!DEHX%yJcGzW<Dkc$L;*{$9I|8P|^J@ar
zqrP+SBY;$$B4*xl70Q1v_ntnb=>+*mP8oN#zBS`k?Z8s_;=8RP2s>YlzVLOIc{Q+H
zUDFrK$?#fbQO7Nm+<HQXn4;1w))34=sPEtU8?GPhL~`9gx~Xn{YUhLDmGPRrBRKE0
zhg??$yd?G6q{Tq&jO*aJjl4rA_$Y&I_eWRSYe|IBKs6gLH*^}D&nGpdr~;-`Ay1<*
zD|<8}6LZIUiQV$lqCh*M-e#~2GD~E>W9!4YYr#j4CD-EX>~#Gxe_nidpOf&2VrZ-M
z;ozvOtF5MOdVJ1I9d%biV)=aJ>Kl@i=?Pn+7t0hoDWdsSi61J`mG;{}r>6}t+jBtR
zn!JZ4OWeiX)@6y*UsEyyOO8;3`-~&o`#&3u>kVVldq?v-z>e!80+n6@L8Y?FGsaWg
z(VG%RS7K5yrPj+2Vip;7^UkmHj}=mbE4ryVd+$i0j0n3bLX!XZJ5}IJ+vQ8)1kTfq
zzi+t++gXnQ_l*jC%C0HV`}u-Ql#SK^IB$V+w(9KoX~5$b+gcUbnR~*ym5{Qu)97Pe
zYJp(jo~Zk3p+8IY20YXJSO-xU3xw5F(+iK~2KpYC(^VfYoVbQ}C=ZhJ<*`$kirYa?
zD1TR-aGg^2j!_}bqq>N4AG6U|fTpyPq;Mww;Jl>EL5+jn^c9#KkHpS3tT<0F1e_|d
z_%4=o9X^hT?TX0(>~qsufpt}WbTk_lPrkLohn3iHs_?!v5ILILXzbsFJ%YiziiCJA
zDTjF<?l0h}-)JF$C2`yFd}81wTy*{E^vi?9-Iz$1>?F%V5u?m%O>HIq$Cq#BaJlW-
zJ6e8@s6Ut~9+`)z3!+?d-P=k!-_f9X2hY9>X%^GG4H%DVF%HdHgF$am4V^aLYU_Sq
z*}h&BvP8S%G$Z(_eH#1;Bew85QlCq3w}j$&&4BzF`1!kIS@%F~NZu!OdjeVvq+kH*
zMjoX3Z5VA@D@W=@w}|Odp7SGB(*-hGMdvUB>YKbKwZjv0)t+8Rk=|BdzLOkyyOp_G
z3?$g}gqZ;&g?mGcnKwa%ZI9wJ=9k%{wv&j9?JRGG?$@ZSynn%?QH4Drv_K(TVU-xs
zT{%E}0gn-yO6S5d%LU`hsjr-|Qcd|yLR0ms<cEN_Tv?NhgygVW4i8R#zU=Lw^en?v
zphV2z-q~MpS2!OdE0R{L(g`2pNqdV%g@K6TXO*{S^W~nmNGVMfh6(<tyX@(HgkU4;
zqU{`<Zv|%r8~#D)e_9uWnZXsVhj(@A<B#3p5PpEmo~~Ku(&M7y-TQHlOOU0hf?DJq
zKs1hRMjaQ<7{EpOKwyf8X+*%|gr$rLyzwOKHF_{XYO>-~PTirT<+5ZkX+l&L&WIO_
z{#JiYRKFvKIvkjR(2RBnCD7Kvy3dQg!Wy4HYfXlrLXk&*kplT*v5@ar#lkU!&^AEq
z9O3!bYf+@|m1;%a+Jbq$GIGpN?ANAP-AsW4NczuCWYl+Dk31@l<8iPV_pNCq^&21K
z6@jC9B}xQHmpWHL90x9!W%l{yN=*P~-4pZYRn=rIJ2oj>AruI&AJo8rQD8??{39`}
zF1_2{5))X4xK11<tCav;;O9xf;EZ+;^?m`DiB#k#_UP8DH}@p^)xqedu}Ds-5^u;V
z^Xz)1hJ0{iejePzJ6%JdL03KDR=H~iY4<m2R|6~}m|ElbU~DJYYk=9V!5<g#)a=Ze
zD}_QSIqM!jN3Z<{ntrW#=+<sc<nY?vNL;$dl;-@ZPYMHC9ju-doJR5y=@@zr_)%{0
zw*Ub}X@&>&BbM#a3-EH@SJ4!Cxwy!6d{f{1nw~*6$OxhJ4lPZ4)G;gyY3>h2?JEIk
zNt>=PQY*prCUf$Dxaq!n!kKkr&rsGV%`L@2jX&N1G8`mTvk2x^gOI%2Vs|ymNH~>^
z;${h^Wp*O#X6O`r?cRjk57#IO1JOj_;6kJ%f|i4t?|uKU^`q9jBSwZ635F(={X%sK
zpoJSnta)eftO1OvhYq7P7Q@=D+|0)8f#g4xDhc3xTn}-<sA>oICF&tFNvAr7J-HTn
zP+XU*5l?mZmLtJRBGS&SPuxilJc(PU+Hh;JlxpUsI?4%DopzYMntVgk+s*Y&7=Z`$
z`ybCx8k)6X(&v&{P<r)9gxt%W-uiWjXeRMZW_-QLe47Q~QEWF(uk+1y?VZ=l15L}F
zYUIOEpdEca7weNK5F)4Ko6$!kga@9*tCAB?yE+E~yE7La^VHsPrp8#oEhsTN4Tjk8
zx97;%;PBW!ld89Q$uGb+O1p7pLJ!SO3!n`DN_DIV{A<)o&Dl=E^rk7wk#R1$@EWIH
zC#3?Vce!pXrZ9}{1BPMk;%)N~pkjHQgrRYeekCtI7WztqS@v{pOKcOec)+qE7UUzL
z5ER0|X-Uk;NolmkT(xS6#U7Xdk$b;ov`^`=PLV9>&LLQQMEG)1jHtkQK1{xgU58uF
zix57GOIw@mWESemp2g#;M40XuQF&3^@e4!_+jz7$B%fe?)00%wA#!;e%ep>NdLruA
zWix;9S4*j7G>HWDr0|{qc_}K~%f)6k<k;7#TK9>-Kx?i;HgEIQM5*YDH54Sa8uuXb
z_>9eqE$P`9WHs)0nl6c|RUje=-#aKoSiY_jG>7>Z^bAs81w8-$U7q4x_p)vaLF~=;
z{yBduXPrg-mQm6D&^SU4q3+`@Zao5_E9)KQ`t~(mech6a%H^YACWyl^cdU|>i!(sg
zFq}J~$=_o6<V-PfwxRi4)Fydv2Pnu1R-j^RTbV{YPT8XwoBsw|4U2A^$LjQMb3|}=
zbtDae(sF<1SMInAKuy0-q5(=%PfOpnEe#svB5AeTu3KYiW&IJEEWaB1X1?4v@xKVQ
zGc}O+;Y@+SF#X8_*BDuI#tlb;_;$$=#_k150oA3TRcRuQ`51RDx4^H&c2DYm9YEEH
zvDo#Pe!R>To9iCTMwmb>F2Ja!{gc6l`1g(33UIx-CvjZz&fJ_Vjtm?%Llh*_cjG~R
zOJECBhAk&59%})8@)%2pg*0Gn{9t;1)fb8yjS#=^k*$*=^3R69^3tFvDi#XO{9reb
z;4)dES-tlSaOVB`6M|)tEt)b_9)wok;JprJBX7ZH7qz;q!4GfhtjXN0SzIYuOd9sn
zS{a;5v=W_l2fwre3CLJTX>5t=1|BOkj0<|h>(Y?d1cf9x{M-F@$jMSnkjs@6SQvwC
z81p@!xUybv_x7;Yjy~JYVOvP;OJoNNF)qN)5RuZQUOYoUg+lj_KWP41w%Bw=-K{P}
z$}`R*GYC_3&W&OeHw3*3W%%-@1rvcYxy0F>W_!2m&G(at;LS5yiOPEjh>TUiC*zEK
z?)3D8I~2IX<)r%Rta|=F8{ls7(VE8ri$i&#Z~Ev+KEwujTRIIE%O0B<=e1<HjpvVk
z#H!9$9cS*!XNT6~3AqyY5`Venm;JHRxOQVaO;^})lhK*QD48<O>NdZsnpm`H!G=yj
ziK;ygowW-jU}CuVWM+Ai4>qy{z1rr^y+xDFZWh<NU+sGn{=RE-Cb3|Op0_mqjKTF(
zP@$){6;mR2<)hANtI$7Z=v6i9DMV(FJr&Q3{H}TMP({HVp5l*_GFcN;$k4@?cz5=a
zukh5vz8x$PJEHYjnUuZCl?<~nneJA5n5`elA;;H<NgnPfI7e%Uqp%t~HKghQ7`iUY
z7u4Dxr|KgFZBeX^?BI%=_UE#pulU;PFwYMooCSE)Exyw~#NUyCq-G1F4mVS!Tn?<D
z%`k9+eG1%abaEmC0W~)QZBDf3G0yYu-j&T{uaHaK!=xC&=fVWjxaI*CDv|X?m{2Jb
zMxNGHjGBZAfmIKBJr2h4UU_=HxzcFy);XvJt4~3rCR*M}Dn`w7;uFFlpu=_Lf{Ij5
zM>MT2XUbZTQRCt+Fi1?5vO6L+Bk3z!h#R8!G*ev+eSRwRM7)u-Xh|!vmBx=k7A1@%
z0!P&0Fc^l}j~p^7qzD?S!1@?YxqT}ri3niF8!{}tD_okNBJCV;*meaM9(z)@SN95;
z5SQolcC0dDpXcsf>H&<BAv(t`TmxZD)n=Bk#pJ8-f8zf5akwQ3K$Ar#&%V5pIzrcd
zx0ManJ?8DPKKxbc$7Jh9Uuuwjmt$$aF-#{AFDT%}*HVVwrtsr?w=7zfS?Jl2@ziT?
zSyOw)sIpsLvlLq6RW=_KH65;V=djXD4tRT#9ATv~_qOwDPFW)X*tN(V?L>GzZ_MO2
z01QRvEi>ZvQB>cy&SXzu!}cOVLa1LVp-EL~1gva{-b7raMtmk8?tz=luup?DQWEcC
z^{@)TV<x#5UKW-70Ux^}uTe1VS_@<dE-CvZK?&-xXSP{++Ig1ktx*yzVal@(T_Qzt
zDKnW~fDNhH4KCslII^X@EDlP#QbcT=wZdvYbwy0sg{C1pWG^|p>v-%kWz{*YTRKCm
zSSO~eXn(eZ|LlY(VF?ofj<J~PIe)0tnS7|X3fZ_`to7niKpc?gmSg5R8JNV%VO(B-
zQphiO-)WY)6qiMO<n>Rz6ZX-Pv}eqeiWl*@&NP;c#JN@>@8oa1jh+HeG)b;%b7;>X
zVnW&oO+9jocEpsu)Cx>b9y<!gBZ71e8O{bIIqe9|q$hysGy{>XM>E+$!Jo)3iK`Ez
zj@Dk$?)5$B<4UBj9`}0!7csv5s)z}@v*iT3!}mqn<?qm_q{F~htHBzY+)xRBeXJep
z?(AyjS#SEFb;!wkTN5?u_M-RZ+osU0SvaOu9A=ZJ7w!8c@_Ra+UC+;mZv2w3(eHLv
ze(`0*AnjTy19{AajEbekDoo+-@UA`<Z1q6Ye?^ch@bQk4`a&%7fKpN*K<!-lV<U!3
z`cin29IEg1!4Hn7@$fy86pVB5yiZ*?t1SL6|C2+~KK^rZF?{K=_o}?xI7L|?ttJTx
zncY*g3YLz44l(y)H!c;Q9yoFmJC=`?wF*UdZ}0dxhsbeZ+!5%)dPYn0NVj*2AU7;q
z5&1~bXRy;A7VT2*Y*#2lkBRy)eQYxFLF)@;xH;!i^aElYQ9eu9sB20U5w}i;{Xuc*
zTlXHkw-y&p9I@H3D345TXjuCk<eRQdA%u)h^7ksO(2S-^Of;W%?ov<k#+)0S{Jxwu
z!$qdw$4jmaM$W_`&@;<TE*(J?iB4-iir;`BKe+gQoi%|F@NoR0*dUG=I**YejGSx`
ztUfN3IP)pU6PoL<e>wSH#z4=JYZx7;hNeWASD%R2NU`>3ZKEP5%~oXnpAh$3$UV9b
z{U*Q{Qr<ckmn^LY_|p++=f#B}ytPNd(Pe@bUJ3favW*(d9sY>4<ze-kV$a+n5fqF1
zF0c*F_r*Lt>2Q(1F8ssN(0UJRDyAn5SYCdps<{UynLGMDk5(OQA-%WnD=JR?*OCq~
z)h@Ni4tY?#EHsa;&@Za7KPz2jNY!EpPSeXBFC!O_P-GFCmk&S(<N78Is4Mw-MHQnV
z13MQS0Fl{jRHk415R8IDo#vB45Vexy0#uW*sOAd|9XFhoMjjeh8rs^y>Hc<-E<{-S
z?%iKs>`EV6`s_vmCq4~Qwipekfng&_?{2}}P}N;&?ck=4^9rBc;d&($Ow|am5|ys&
z*S&%yC=0hUGNLv>5n6IE5_U=jHee!jYEySpV1#$GL%SQj-86ppEwf;l@0Q(oc3DB{
zJUU__BLcdheu|4=Yo-Ec{awhB4besMt!FpCQS5##;A-|Y4?o98kv0x}Dd$N!HnXMz
z>19kdLLQ($9}4tg@~56Q>_)CNYBz(G^ru}HnJ>4p|NJH;WmaUl&%V}DeQ=8jjiXHd
zfxG;<2`gB8XNjrFrKBZGkCQ&{yj$}GC28KWHqXV}P-I7Ia@0fA@bN!WEOfh>%Jy2w
z=-6zgDF?ID3@Hx#Ro?FGh96g*3%&xtu3y}@S-Jf^Ols_^2eVP0cZT$2&o|?)@GM{>
zE`R6@-lSt<f^t7#iROw2tq@MA4*H{8is%{f4ZCxXB(lpLFZE|f*8|JsQSKTP4zSUs
zk|P;!u-y!I?O*-NJq!h%F9k>i=a)=L!g1nOY@SIXIz&mSaj^VCPcyIT*yKt1iSb?E
zd`yQNvJ0;b?M!{Bzh|X0A8tR<k4ZGCMSip;98OH5y;~8;QjP2=0iJ_So|v%8=xpS^
zr7v+h0QJz1S8);#U2WG@KqGPJkjBxgJQLQGvR~fD`^Zydqz4Dn2obqez%bcz)*X#1
z^^_W%!(ZD3z)FL3h<sDU$r_hkX$v*AgAKnq%BcnWl&6<o?K*FJ7ZRXFVhXV|sB5GW
zB(?Lz&k*P7sNu_TpOS+I?+#W9jooq~4U*kGZks4ZWZOANt;<WflDX}=j;Sizgz|hT
z8I+h7uTYdC_6c)BcOu2z%G1cAX7~1Ep1&yfIYEegFM6;FU9Gr+bRMh}cNCfQ%~@E-
z)>tq!z;787eaO>C!a#XaE}l9p*ca2I1nYTF0hn^TDmif^CXQp*_?(@-+20%vo)D32
z+{`*F5&?=I!AyPa+xu5kz*H>7ji;g!9LQGv`c?}92ysZrBhGTgMeXAo<nJp5iDk|t
zw^`L`0*JPN$i`@I*k{=L*>T1Ih#yB5G}aU~I5SG_EbM}1%E~uE{}c+;!C4^5tMG%2
zey`>Sjhqe23gT<Zjw#xbf-j@+GMti;NqwhM!bAS+?OGbgri?EGZM$iR<zdS}$Pd%u
zRmQpyL@IG~(q`=}^<JrT91ZD*zMCA9K9SQWoxl8HSfkH(-je+l6d{>V3eK94Kl0t~
zV=9tzD6ao*ahDT0-rh7i9+Tpd;)keA-&2EL_uYoFTfzI}SoFQ>?9RJ$xJGIPpDyoW
zMIdqyzPmE|=#!Z$ZyEerkaw_1IpIfH1GszcX@``VnLS&M*qoFN-0sfqLQK@1FaCq3
zHfe@XFhKD#2=h(s$fc8LLP@qZGz^4$88mR6GP`_@bvM(V>3JIMsg{=vFKtgz;250>
zsT=U@hmnP2Mqb9HdepQ-le=osV^F`It4JY=XdN>UHoMzr7EkDVfm#8bk6X(tMl++Q
zm?VKHRE1%7Ud_Zf6267nus8V|VgMr&aVemND{PI^PBpq7>Zlwe*#@)6F^i?gfO;EL
zN5gqIT<3aZt&qI*vfk&CXi;lSO>wgmvHJ-@TimfLg21>SYlx$e*rz=f3Zy^c*k0V3
zJcG5CHwmrQ_~dfb*76J|Bpwbk4ab()a)AAgdP|n_a3#;$v#nIm<tf@@hcS<96gRMH
z+7J(x6X#Iq&V_Y`ohoI2FJ}2$L`Yh@mzWxPJ?mmF=G2fsHkXI=ltXh9(ljjhIV(Dz
zOK~yLOP*q@vpd4qtB_5|95D+<aBM-$UgZl;)yqh8(q!^Xw6|8p+5Is4+uRp2_uS63
z(nu0|QrDQ^m?SH<c!PFa+E+3WPlpB?xWV)=TuS}0YPM;=C~#J@N|mP9ze<lS^^^uw
zE<P++h6mR*_^kLgCp4Ix?OdMdkZn)!FFkDLyNS1etDtmuw3heVq8O6x^5)zHFa0&y
z*vd8d+nW}{%krK%{#NZJwov0uqp1M|#{(q#wD`03^M<t~Cl`!S+qfG;f8MPebhtwu
zcGucpFGIh69wMGyl*CGuOxmTo?}IlQNpo6YiM{!NIs%CB%&};#3)S@sPs4}(eeHzE
z0CjaoU<zyNBxa`wJw(T4vm_vI3$^rANyAFDx@}fBs}}=_UTt1aXJ+zAjUCXWmistm
zj<=21upU`Pi~Hr61zv|iX-lLj_mNr>W40qX?r_H-7&*?oW@w5d4%rV{bMFF%@;C!R
z*8{}rECyp+?#Xl{M0de3$Uo^ZWJC>VmkV_@D)Wt<g)OWoFp9i$N{eytz=n4mioEe=
zZZ+Hxg+RUB?nC#OrVtaxTP-fMQtuyD49obDmM7zxR}KR&UhZfK`+90vS<aAEpg$+;
zBK4Y4Oxq>dK@#O#BBAK|qwkd8yQ7P>F?&=Sf0lWzW>#YEF;*F$!+|<~mmsoP23<U)
z?74R1ehd$p0%+EtG=OetdZd(fhYfvB`dKO+K2tD47QvSGb^d@#D$z@<M69%2<dX8G
zTh?C1wbzF~W~Q4Aj^g2z{aL{>Zol*;N!$C{A?;d|){hn|Gk(Pbg{hzl;;VFUt2%nu
z+tk%4im~7;%2$FO=qGa?LH(OhBT{TdifPU;B}}ni#Ca_&LW$dUqsNwz%+@7}F|X-a
z!{RE79?{Z?J<1%u0#a@R4>J;;+U+PEnus_qSA}pQ)0<SL<D~A2az(Ab-QBxzExO<9
zHQg?as%};6fw6&{a??F-j(Hf{a>$Sl;`mK|afRhiq{frfi_z25a(_fqU5wx<%(@Pq
zn7D_zA2MQwTidO2feL-J{l+#Z+qKQkEOd{M*+D#oUwVJFa(G;L9@J9vl0*#fhm}tP
zImgEqQp>Z|WiP=_|F0tAxTIxGC@0DF9|52hMq`J#{0GxVobnedW=l}o>p|W*T@A0q
zd+?#!4QS7^wjtNyvG~i{<V}I39V$&(q2l7^a`X{LdGI#KL9{KWz@jGHqNWyy*D`Gw
zNtD)GtC8gXu*Q7W6*dTgr>2h4o`6MbtcO3k4<eF~*K`G;z(tCjON2a^-JrAcK-vrJ
zUI)FPq)~Of4;OMHF}6)+io&8EBJI7485S}s6jf+saX%S^Xa7E#|CN>ZmkY!wsY6$0
z;4W~@C<S7E9g(xHR@c?`-Nnu%r`A}{nM^mAF+(l5vJ%~}EZs@L%*0)GbOswcZ{hLx
z)cM6};{4LAt+g&#yI*WySEqzD=6znb!6NnDQ%DtfmA17;o**{A53@p-4k+K-TS0r5
zqpP{qogwz$DS<F}7&7t2-XdFtl0=)7`L7V`cL_+7G@9_zQt@)*2i~S85jxhv8g>!|
z-3mvs7|D}~QyJo~@#(&2gh%`!x*>AE_5&Z~VE|}_LjHbHH?!(Ye_^V^`iG*Vuv@&O
zdvndNmXtdZzhbgI;jhX5rQd;QaLxFEPxBGO+t7Kuxe~9Z4}?4H+n4F&AC2)?>G+fQ
zsV#%ZEia!1_f(|or5dL(cKMh_x~i(=75%Ng|2Q;U6?u8D?tG&9ECtOQ?4nC3-GU|2
z7QeTT&>hU>@yc$P@fL>^0TN3ll=3~r{TkZKc};pPm1Jtp<EXLD-UjKKF0^UHYAanJ
zk8wv-@)fG6d>ga2jg!t&G#Q)QVQuk?clSQD^k3;Tw*N+_aWFIfheu<?=ip%G_|NU1
zeE&a3d<F)F|0{lgq7${Sb~bUurxUd{a5fP(F|so@f#T(ba&mSwF|dJh--v1Z=Y9A^
zV~fS#-lmVg`JaVJkG^SK*eQf*QB<U00Ro&1YjK{4XaNEOl5F;Iz5DAo^O&78=&HxI
z^V)NJxibq!l@!2GTL8TTSq2t@?+*G7QUoy3F_Ibx1ONzVzzYaSfUT;E1P=uB1wCrO
z0_-^`ARoTeOP$;u5DP(EauCjsc_ByuK!ZomFOQB-PD(;fOM(voj|d<CH3@$~11j(E
z1lkVZ{0^Wg4*?%LSOF6P>K?SUg)rvy^8vAEe*yv_DH-|r<qA&j3D_+#6ASNOLo|SM
z7Q#LMuYsLDlrzXb-2XRtFWx~6UtjU-%h%V}J66}8x2!b_n+&cGF4!I%9dMuDPA?yN
zPi-84M~rSSulRny4oG@S;EoS@L|!d8uaFJiAEXc83?C%c`97cxpc`;(@&Cy5(cvG$
z>Syr4xc&L)!12e(-rU;wy8S?b1bn%;IJtzn^<(1!;D`R~K>+~+epqsPL->Kf^&uQQ
zOwJ*|gb{w@_}~cWi;)%e-!9<*T1JKfh#80Yt$@3|fc3-M^Rh#HRf_)T8N{zvK{+Ud
zdU^m7K(No$ekt?s@S7cVgWvtGYs*0_!681S9YF=Qwf#^I4ug-_K>_kS0a{jkSp)^!
z|MYM9L-?2Z7p13{;RC$E13bC8p8mS{O8~&UX}x2KuD~OL1PuVxj4}EX;H%+3{2_z5
zIfDkl@AB~{5cpO7QX4=(25IxpMgSno$ASoalyfGAasCz(ec8o2f~!O2Uyy^xzq@>X
zCDS4v)Xs;u@8dh^)ta))ud%MM8~!#M_oGTl&j0bp*Cmhus4Yqd;Fp(|hry?%g@Smu
z!v+?c%k%h}R6{y|0u}#Bp#3*5CAt182W$WJPn_=mJF*Z&m;nu9|0U#%1*iaMc7)ge
zYu54W`u=P5UQ7O~R`6>lYRRjAXU8*&2lk8CU5EyOeOC&a*v<=agnTBL&<OCZr{uJU
zS6B({4EXN;U8{<;AD#^+5{{KP-xWyc4Mga-hu?-dhIeKNAjsbN+n>p2zo_>J1}5MS
zCeMDh8Y<uk3Go#N%V9i|J4OQ}ruf(j6`uX~H@6bXA++NQla)Y$!N0Jz)d&E#VO3lL
z8J;*iTM{k**NzeR?HykpWC7H_iUDK~>UrdSS>1sFpm&Du3{C>5ul6;L7;<m*S1S=2
z;7;qC1~LNZ4ab1qu8i;QFI<C%-~PaW_|_)Y#jDuoZ)gHJwnsC58(G8ljo)D)(AZ-R
zcK0u6_%E1tcHi$naC~}eArm}|v&a`T>JQS-t;Sg8ftcpc2Es3K9Q70mOo-<%Szyso
z-kBdT1SHVgwr{ml{;OY<Q0#&p6dm5FUj@9-3An59*8wk3JJt`Y7%7k_>#bj)_*cLW
z8PLK4G{}9xuTcnycOL{i%=&3BG%7OSolrd2;(eZ`9|3>sZX7xL8p2O)7U<nzJQ}cz
zfbE%HNZVI6f&eMNBNze3ublpo!%i%1@95nvO#I({{3GYDN-pup`%ixU-j%+*)L(+a
z#<*6tho)UE;PIv2qun?_{5ia{=tg>Im(<u7^GIK(<$UqW(N3Po`j(xHc++(>x=8&j
zDoDrG{wcLQ&$6SvhipsYQT-lLS4Y<L)XOu=sofJFgCWZnQ3fE9W;><x;8y3e&vh?0
z{2=J$5gTo8GJ~Wy_U-uGx&DtH7O#@`E1PKd*1MUg*<(J=eJ<6+PXIdpV^9PoRvJQ$
z`Z6cH=L8a@NlwGHg;4U4p-s-!QQ<}BuYUzs@)N}iS;QB&fLSXP9)na=K_TBh90R2$
z8$8d3vqlo-<3P>nk)n)XW#fy`&AI~$YstcN$yN;u>E%>ihU=>H*h6Y|C6ga|rEn0R
zO&#YHVtW^1zUGe>p^-Bas4lg+3+r|$>E5TKrIxee_NNwO1`3ANZUXJ-GdwkDU}&#f
zrTPH+%x<*X=dwbT6?8N%@@Sih^g@Nf*SvD_Ivm8OrO;68rsIJmV7xpkh)e)XeWKda
zXm%zI07cAIMPQlcp3=CNyhL#BZ#_KMz{?u+rh+rQO>jLJH?dp+4cPNZaw0_<UDLa_
z+fshz6rT49Ytd3FF!N<R;MWbO(lnAeuuPP+9d>seN;(^o0*=0c9Lv(_bxXnSGs0%z
zBx-P*ij<-Z-kIV4OOIutKf?!1^>8E&3D*r8i2Fc)rI5&8m0{o%BQV(<{&e;hZWeQR
z<lR9VtXrVDX|{7IpVwOv2ruy4d|I^XB~zCY$_|;SMWy&=hOQ^6hHW5slI|9HZ*ti2
z(6-21$F>5Iao=P@Z8rIR1pHL1_(AF1bVdZ_wq9v){a{wFqP@zA=Y%d7gQXWz^|{!T
z7t|)JeT*YXSAXKT>}0nZdARx6+i?BHl9~HZ&@`X`HhwNtxWROm`;-*{;S(6=#U&u`
zc-LUfkYYIF(d|%6gZ^fLK=vF8TNTYw2g40-H~%q%rmpQ4S7@=DekkKfXnWVhadn&*
zY8CvbNQl+5!k>FbtR_^HIT%^xcYjCf4e43a_a|(syeD_|8(sM#sFp0U1sClq9@AXb
zu;y~eFu9ltV}u4?M#{kx%z6>TX`AJ;%*WIaU=K|GQCcz<^19No^DYb06S}XZt(ksC
zKaY7(!OK&{J#i_LEzI4m<Jp-4WvbX#SPMRfFT`wz?$MX0a{SPI_vQ}ff{m{AHi0gy
zQolIqi!6C`$q5j3=!{iDxwye&WDU~}JU-8ES3S&#%iRpY;Q6%Hx<u&|c|uS|sn)6F
zV@??;ipW}G-ga&jdY%q{E9EiV(B`bY99-Q3!~=~5Z-%IWQz=?(D<KIH=+D~i!3brC
zva3`Hns4BoCH}hQD;%cI6ErqYr&4v}+l{w1)nwMnd5ft1vP`F*fZ51yJo>CS!x!r5
zd}k<pIvfyPPcY%gJ^Pui=#a%a>A1Dgcu!6e4Lj=*ozQqhd4>L?vgJ?)S%Pxq5eY^v
z3I)0AnJ7=p)gI`E%u)+;C2y7W_=Wl)L8ZeS#tg*6kxds;_+__%ju=J=8yzO|hdE0p
zsA<$vbf1~QU92$m*-jGf9nO2VG*&Ckb!&>L>9X!3jq>W$LU`CTajXinuGE|dM+53n
zCU5`T%)-Ext$HDa&u%A8Z#LLrS~ThtaY|u=76W2Hh;<o~31#A$QTSRmxDiphC5uE`
zZy3lN2!`sgfb)r?Q0p!aGF;9t*N`N<KUQ^=s&|GS%3IRY^`rv}nax0~>Y&G)wowH&
zIj!|?GD`y+mg9Gg-(LvYqlp@C)Vhv&eZD)xkC4D>rXXQcqF~W40x#s}Q?HHtR>Itm
zhU755OJq|SK#b){{ah5_vWr5MiP_vTyvH_(p?cf`4LU{xY;E0M?5tl#8(<$C3%1dR
zL>i0FtEE(RpzM^6XT+U;RGjZzJ{zOKSf;#j<4(3MgkKn!@Adfs;YUu9AuXd5#q%XP
zjaFV)@)1b($bwQ2L&=6ADU+V~V(uiFs#A~hm_xgCBT_mmahw0e*g5@W0`&bq+um&3
zwrzK1xB1H0Y}=b{+qK!X*|uHJ-UmM%+y~EV{(_m4nfZS3er1vL925=Qqk0d+h%9Yv
zCz?A9g1P!Eb8^EE5(!P^)GHC1f>_!+UlK3C?`$0CUAfA|09EEaT7Dd-A#H8R%V4mN
zK@<n_EFY}r`X}V=_KOMd14WYq*<3awhUTvy_I*1qE19F+gxKXuet=)hA|PcT7^nTc
zUa(@~>l$X<?n<k%Y8n5UU+O*Nn;|}pz!uIHovTPd1xYGV`H{H}VdkYw1c#+4or5E>
zE~rQ22&b-<YH1OJ09B*@!s2T^Nr)*&SkBD)GH~~59=5o-9+xfVpr%eV>0j<>?7!T&
zDKW7=7s^E)JfO4=ZS<LmhSoYOk;UFCA92iqrse>YThJ>JBWC%!@*!FEJg5TV%eMS_
zqpug1D%{(ryF9#aY$2WAa7agvYCBHNW&m$^lQh)Az}^@nX<qq^e;5qV%KxGEYY1T<
zn&I7Z*JFQ30&$?tt1Np~!moHP@I#PBD%o6J{<^p{*R%4lQ>@uDgGrWTYPB<WN=_i%
zRx%7fif!Lb^u9_QtSqus2&t>Uu#h|k^(ru^yWsai5HttrV(5D9pFF8FD1H*eX?cSO
zb8?;|Bn3Jqixc0fwX^KIdTUSyu~vWo*zZ?Jfnh-tL8wbj@N}Y1qu(11=aueFgH?=J
zfm)uw{DXD$&DLqPE2J}?7*AQ;Zxq2tHN#dXSDM8i&n%5@rGIMax-|4@T${X$t>OWX
zh@W&5613qBgej!3HIy_>FIa%Y-_9=UqC3foaT>0UQi0<Fp?;`<f$MXyim<vCaa_9n
z^C!09fX6mV?c1rK^TqqiNWD0|tzL;=e0Zbme#yFp>EbVgqplMtrU+<ly0X9eG8bi6
z#uYJyW`?<GL+KDj6rHNh)KOgS#Alqyqd-1CX2MfD2-!T9v0-IIcUEJ?{S6(pLb4}v
zq<+cxIg2Y*W~}rbX9Lf|G>A5m!^#(C=HrgHyT^N}euRH{t&;;hIih>9QQv60nYggJ
z;bdok>0}v;*8+ltuVol#Dh^&{z9MJWDsPd<S<t2rk?S&b#x19GfH(_KZHeQ*vv3OB
zdks^+u)zfUUEWd+EC%i_aiQX$h2y^?iOaS5Nh9Pm9z<g0HddUiWZmMcovZODl-M~g
zgt<y0JZeg3wWt?P2qTC*AMHtzr068`#FQvh-<v)WHZ>ECL-YgPiXakv8Y6C%MmnDH
zZe1tQL}#F{?Q6Sk#`<DSUs|fnpnWD&kgRKMFAQ;dAm&2m?E1G19*UU}Oic#ZlqJAl
z=nGWmPD&&V9sNG#wJcP<!c{$YNy`lMk94%2N#;x`s%j9Tut-s)bt=CC{c>)N2PPJM
zM{!*S4VL>(tZA!DVm+grGaKxUgDrjuI~=R^ye}Z5y;ZRhB`-*o^DBSiWPk=ORXQzc
z_}l7FJqkQ6=ehj|3Kw&nh$x-arzmVYvqq*b>H%K;LF^h!+ru?j%6FF_P@ud{m^5;k
zyoo<5K_>NQKna`TsV__4U0(P-)r@VNzZuhEwmd@EW(0f^N$^WR$8&aNiCyP6y*pxY
zQUa_cPXEOXSpK>_t;vBadkX^|Mqub4jX8HJb1i5!Xe_jy?|>y=K;vvnOcgvEP${54
z6XJ>o>u)JSPHa&elJ(BGMx8<9Js9}}9LxDtn@Gd1u1y~d<R7~Q(S+W!-*lxRK9!-P
zQD_oIVlcgM2BkS=8*FE?DM?|Bv`*A&D#-CIvXCmL#xT?B`=^ADkIElI54n`)Uy(Ay
zFoD`ndG#yCHSPUofJ2PITSaw$qVkSZGak^u3q9yGkVUXK&ZnnJtT3;op654nB3vB@
zKkMO#y%Vc#rNY4MsZ!>1&bqR}@+?SxHm}?fuJ4F7`@9@1xlKY99cgy2Od_pLk3eub
zv__W-WJ1s{7c>##+t1**qQm0b_6btd+A{!{=k(ldt?Z(2Wi5|nnT_AqFn`MdQ#(!8
zP9nRVdgA9}SvHX43^mml<J>D^js)E7ttOS?l8e2*wo)SQ2+3Cdtm_IQd=UsmZ$Xen
zXyL=b!^ho$-{!1u(>|QkTYDX1RP(1a8sR>SgZU9?;()86q;In*Wzg9Aw0n`>&Lufh
z5*SeKR4C6tZ{M;U0VE#R1$)N7Ox1${auN>w+GP=Cn`$ZxoUuu&&ymbkQ*9UbIp$Ab
zMJbf5KfuYDfT7iy6O18##5teKI-ntrL1(F#hst%6YubjM)8<rz<Wt7rW1Cx|49}*P
zfycW)S&Q>QF96Vv8T<>IFfuqvzEn0TyII2ydKW2PkY@~|=O^vX5M{ngh1dLcdeX#}
z16a<w#Q}Hlv|+mqURo^ATf8DSr%6GCs=PYRe^{8L9eq6)dDzc%2#9?kH>JBob)bUt
zD6H(<_Dd^Xe!A+friSfz7b}o$yVJm`P7xE;oOm8!*9&*&N==T6+PC!VD+W<_!91AV
zKA6>GuGUiZ`tMz=0lW;99&}%OJX;1%|E$K`Bt_xlko~ReAz!$00Sx$-KH50@{ax)|
zZbaivnheXaNkV(R^q;0ivvd`Cl0g(Q*-G)WE`_Lvl%1G##WWW8$g$TA8WM_DYr4P|
z{2ZT_xZTf}Xd4swLn9jMMmct$OeEJM`6nu~(aLP&v8(Wl+19u4f_4XV)S?^w+(emy
zK?QRGPb^f_50vvI(|@%6JXV$9x$(!f)M$n6=yLd&1efp64zs~h`W1SG1~c|N(qp*8
zXMK#|9G(cf5L@d8*e8n<3uWzgS26YzFRME@y2zpwgWQ_q0%Xs@Emba<M9&XuO|lY_
zvq<;YfCaJ!q@`6RuH>H$hW*^vLdRWwFFIB;=}P(|tK5TgP6o<zk(HJ2aHY*WxZlRT
zOnQ2sH^p~rEp>#lS+SU1t20A=RMa_OUq?$GgDE0&WcFKBGq$n{M2GAZzHHsBMt;=-
zSa=f#&lcki=+ug|Ov<APzyL_+{Nkw<>jby^p=2MW0ps$qtW*d$kK0S>HRDj9<HkLR
z)KK+_s6?O~q8C;F<Z{dkF1X00p9;97TWZqlA_PA-1-;D@Q6q;smg1DvfpKX%Xc7;B
z8B}GJO+zhNC-hs&9y22rqwG@iG2%v+*rRUM`n4_6`t@)aooP<GGDMAHyUDfDguJDY
zVn<BdF>#;|Z6(~aO?~fceb2xiib-&c2u?&hj)mI4Tz5Xz!Kv<%2h`&n8nz#rWYdcC
zS?51w7hy@a8KHU^UpFniP+F#bp|{Pu!FaZAX7t}Q7KTu2>`fCfTmjt(7I41&@j&*J
z86yKV-JbNZB7VZYuy-^?r(bTyHl2^8v|?`aq_<R8aR_s%?h(6NyB#p?AggcMHB&HD
z=1tSed>^ghg1QpdBdW~a>l%jU%a^MHS@#pKdnCL$FpHmMszLIcP7r}8j{{Ny%DU?c
zX18{%86t1R$W-_CjSr6vPMU75*Xs#M>ki}jQ+EcsDz&d{i|NDiuecBgZp_Lv?P<-3
zrG=1X8DXX8l;+Q5Jp=Aq$b}#Ix-I7}&GJMa_QMZaFQSB)A!JWd1L6@%<CC0{Hks{t
zs9Ie<Pam5M-5w^5H2kdJ!*&Y(ebY~l4wNM_PKgaC)g7(tp)`_A2`A90i*4EDEjyu$
z%QfOvvsIXnYt^0+1=#{0^IixB+m{?j$g<)Zi$}O1-{&4b4$h4OF4KSJk(@!hRN#ER
zx@LMXf%O!X(#$B;#Z>XpP&wy&u@WK+ntqzaZJhd&y=1%Q?&P?_knPGu$JqZxKh{Ai
z_cXF8{E(2ppe)zOg$Z)&(%bmV>|r|G<lqY~f|CiMoFN3a_)EZ)uJhg;M1asNCYXX~
zpZiuegk?dA5}aKh@_mvLfAG$t_vkemF`_k=!6)AwrAO!>-Nc%UNbhMsXH3im>cls?
ztiGO*crD4Ft2D^WxkTi0MRkLs`qWVKqF&0xh`np=cd4VId5;AgD##V6I1TSfEQpWq
z(i>^2uvT=C5rsb6tB1%VL!pdJ8l3PJmb>ThM(ejvI22_ZtES!ZC}<4{NI6a<M_sh^
z&OBJU4Hk;UZT+TGnLBUCxP!ySH9V6_Evv^#UBwpq@%fv1{(h6cgntPTM73vgr(AIh
zKn52bi&;3HguLElO+m{ek9lOk!7t91qg(w^a<*ay=YPGYI)G^(dl%3pY3kWtvZ_?)
zu4Pvn2l{b(gBr$?cDR`-W(HGB-cI80Fx+C##@r?C_x0H~yQMC6E&x>k{>;F>J|tV3
z#CSo}S^Io;@f>VMSu5H2-hPcvKz!0+i&r%YkGEgmEbG5~LYLflR$?nzbG1iCt9R_n
zQea|fXT|%H!%5|>2%K8{69y@MJ<=w?(Hb&Iaj4%YW&;2Vr>yZ>pAJEF`U!Nel@+it
zrSlHVsMHnABPyQS!$3yRNm4o|@4zlB(N2u#a+wyB7#0{?@v{&#LL6NDQzx?Pc_f78
zgH#*#PNZptF;-@c?qEMWRTxLup|$I#9igB+{M+NUO0fg)K2tBD2<>-_wMu=jT^oLC
z^_K3_!n)u51x!tftWo<swgKDU$fxuAJZWW2X|s6%reZpdam#w{$JNtKX_0VYrHxP+
zd$i8eab4v-zF|Nq?}wwLr<srm8>-&-`-(4gqHz_iwxqrCyKZv{PO)>#6S6tGUvDTC
zz$_g&F*rh`1Hjp0mZv?>PRl87ReL#pGxk5LDvDFFHDnOB97Hyc-BO)O#*0C;PjYKh
z`WtraJ?J&urtM(jd%_=O?2QjAu3)A)!%}=J4X7LpN0R0jd3l?Wer>}kQG3yb{F}s)
z(Z+1KEC}4&@~kDY<wLG{{vKjHa_^wAm@ATBEH0S(9VhcCINkhfP@#Wjxp|)6p=jNU
ziQZL6M)~rcF^C|)pzfOOA^`CdY`fg6gEQZ|$AddyWu5Y}a)k^Gp3Y1WnGx<v2O9z^
z+H$gP0~$_H?=Bm4%85ml-*M+8N(DwXNdlQU?VLAaJF!!-b{kgtWy5{NxWg~z@!b$k
z0CygIT4a@q6;xJnkHd(P^LtT?>I9C%i72kI&ci_VnyLG7b*)k4lu+VxG8@lr?8=25
zUw)drmE4DHHc@`{{-}I&GHO&ko6WOJgy~U@5BI^cHY7Z{qEip%6T0`plw}0($vEeo
zrd^@Pq3hN7V;`sjdauvW)^~b9s@ze?qQz!3A_+wL{QLs|o9tXu!@e#C3l-&xzch$&
z)MRY4<OVhNd2LTi5!{QvhZmZ(+OYxd8ZfY>C|4pY*3dGGJw$f8=Ys;wf#Zrn63c&8
z9vPk&V%55g(WhQzf>_dv{uW+?#Y-xnho8#Dh&JK&ppK7ARS;MeN@-@|bh~tV%H(&1
zjWxouqi2$)E)@qSIw#&rTJa#K`_iUm7O)tgWZI?Sp+)E6O`1_&u(JDWpcQ}bubXwi
zr1|Yk(Fq<uu^}Ii0#of;Z71&U_lwG*y&K;5$m1Bw?{pb!pp>3GFeFF@Gi_}mbUs4v
zT8Yr^mv4U(?D{ZxeQ@39wz)P+y4(JYh#XaFW9^|?8TuWnAfQhVNHFrSSJRN}R}h_O
zpbCIS##;8K5y83A>C@G-DLa#7)0<+p#kouu6^+R$em7l`EchYt#?lk32qyh{sM{_t
zjYg5~DIB?jf`gVy`0KQlZM)5I+cCas$&&u&upEuxbD&=leh@k*dgOrv6h3|!<OsG;
zQgB6Y)1Yn4Dy2KVm<W@3i$v;-b9I{X<P-3sXhhpx^{ikHU6UG<^Y+|juBa=D!N~5{
z%JyiHzHbV$jLt)JpYEE4+k{%-0IASMVz(Rl>EU34>A$7>+^|J}@ZLCDTB_Q))bS2U
z>j7^vbIO;M+nq;X_{R<}^fw!OGmf)3A_x*T8nF<b^<-*nToxzCQ1zR{PJSc+-dr&v
zUDAB7$1lSDO_e?N;cRQOa41!wm?-drV77y38UjA8JqOwVX!^U>$0_}9#qWYey((12
zv>YTJlZ!{Ik19K|0vk`$J5iya<1_VXyixwYpR9}}>OuPU?2GH%4Y|bODzT%_`}ZXo
zXp#Hv96ps`<)Q-}0l>`?govX@=Fp%o+aF$7ZjUUq567#l$X8x*os=O&6RAeWV))hE
z_^Zcc%BPfb_uVosZbZ8PO_0JXo~n#Vl_RA{oT{s?*^23ljFrG#VXYul$nd*pwLg6G
zm!eYD6;P6)55t6UP0Q0&O-HJ0^E(}ZN~*Dq89#bfgH^8{mmf`8ZWq+XN@t>&kjztY
z=J?E{TM%vp%p*kEhZ+HbE1uucw2M)0Ih*Oh|9+2;e+$>sCEhVjH`z3vPXD&&hO%{m
z_~}5bCktIEq~uoAHf1@O5UrcH)7gx@!OH^*J)W*b_PvhKeeQP_$T()83h)G<@~7ZN
z=i{e=F2&jXr|s!XW27xFsO)KS3ze*#(;~p+rL(^nimaax#!u1Q_8;g`om_|iUIOHP
zMPS<(ZNlwxpm!s=P$f0EB2|BIopnjax$40Q-<+AhlgAZHxz+)ZPQCX^>wL%lIg3)p
zy0}roUnZRN1A2F48IRJI<q4C&(3Vaa;4Lm`2!_`oyWCGgfG3>7zdpcS3P5i}G(Gn$
zI?YS>16@-icqQwqTO!+imOwrbH7O47>w*&hVQ`uYy73`XMt+>TS1vomxE{MReq7ng
z*;UqdM|+<zc+N8z7teO*7=c&SKvAG&1V1$e3}{eYiK+(5z6-Mz2DG4z3_BMJq<FNi
zZq8SjT6cdq+{qWDC+v4!FPbMZ2Spc!;C4P^6+Xaj@s2r3-KGNkF#L7S-iywlvqhLL
zb4rJ~Uh^v2vvhTWb!|M6)f78!5U-!&Tp~tn5-(XGk>5b;m(8=RpYGrI1KAVM2Q^b`
z=@coZrKTe6$=@N8Fn23N=(>A&ZQ!gKxn(i7>0r_y5oO(UBv;b2kHW*$i)PQq&+O8z
z&}mi!37u<u5OJ85C;Cp2EGe~@i(|5*N$l8Twdaa8?K$SosXU+007KaP={x1MtktD<
zO*sT5j2(ZMLZ&v>0oe2XseAA10xm+1n6#$)o(#E(j$ETYzRbl}6}#b*9F93vJ#MU~
zRw;|$NeloHQ0(V`EAx!8e?U#*qVfG<KS8bt3bz^I!PYmAX#SNPtZ`cKj<$%)F)QMM
z(%nb%W6PBl2KlNWSwh>Hw=wJJq9bQfC!^E|uu#X+6&F+yvdN?vo#bREO5JHI=u;~@
zeG{SiFh;?GpW{BBL*I^2S@mZ6-Ce3Q)>~P?omer5FF?NOsfgCik@;DO!n%@x<kWN;
z`$Bo)JNN{#CTso^mpwZ1Lzz7Sx^ZB9$?)@8tH%r*jCId8ynGaKJ`Fw!HY@I-hokUp
z90BF%>8-}$oaJG-+O$`6=Va}OCMue(+?9a^H&_r_gtqw2PGY2HPnIbmLqDxxINNZF
zSJP(Bl_2j4v{!Oa-8@y;%~W{}-WcyMxw)U8Z|Dv-WQ;&hi<JthYdPF{=_6g6KSo&n
zc!hl^%Q~4(Z}KjzUOyLa{(ix?2j!|C)m{i1xc~h180ly#77L-GzOb9%*}<q?8PqjJ
z)9p4DHf^WAVDZT-^Nw~gPXY3Fv-cTJAF!_x>XC1u^jgW=Ra3B@b`ZSsLCO%VU`tLj
z(aNovph6ASfv`v|meZb#4et{55W(48#N9gOfIUay4ln^q-NOs_O3?PoX8%<hJ9a{|
zX7xM69Me%NLT|0O4ICu3-CTEB0hfkm`qI_i<`6lIWBl>f?X=<>M3}nb2^XZPyzH!4
zCEE&pfYL-&9o1V%g*>L0P6LL!B=S=Br=F!!=lSyblwIPn3UjUQSKSF+%(Rw&6H7*V
zG<KFnvsaO?0ICCk#Y&>|dsT>4S~}i==u;-y_=DJqPiGyh6YRT3&_hu(b3me73UxF1
zj+F}>g^^Luej|a6vnFR8vpyT2mn*PwCHWol?a(ZqJKxtA&R3`pofLr_k0X(v*vWQk
zMBOe*l}4QGp%P`Yp2B0KF75pvRL@f{YwwSz-{;%_)p`6GV6Fj!%~>9+%E`54l;?-o
zLO?(q{W%knr(St5WiXMS(-^r$)MZ~+ozy6ytBJ$j_+%RA*3g2_8_@NHN6`HK!;$>F
z$c+4mHBN=rJ&R{`Xs6}Ld6>cs41Zz8@B3`C?xu3pDzfRTEPn4r8LByP%WaEJ&L;0M
zs5-)jDS+GOq`hADQ8DcA6KGS`1l~@?2ZWju9s$c6i&+J-ZNn1I)Xs3ic}2B$=cIjx
z#GxRIef^FZpoL*mPfH42%54(BxYpp{R_-0Zz?v+t@P4KK*NP6!9K1#=Q@D`rG@s(#
zfvgC9&^%__6fxJf$8*S4)g^!i%{WUohM0*Z3^xNJdxtpIUXz(!PLS44<ym#^XK6D4
zTdLZ>nj!mdXKq7lw!k@=%oIKk8ciQ7kE22M>K}wtJMQYHEVWgJBaORXy~7>)U08Z+
zY!f*TI7NAm9`!iN*6a`J>*m1M3eP;@i^GFN>>0TyxLLV00SQDvZsByleJ*Xn$ZN$g
zw?1Fsg>U@~xQ;9tI9>>k7#SM$8V`D7is4}&)L(%{-Aan~l`oUI)2TG2VRFQ!F1cc!
z2d^d|j_KUFM<-a@;IKHayXr7um}OsmrYN6<vrJU+nOIvg7ZwUG1N{1m;ii)i3_s5b
z0z7SMi;;pYz%2{zA6|0hz%aSKokJ~3-efcuA9)cl%bdr_$2sUAper(5wA2lA=WmjE
z&-SGde*(H&-^3m#1Bq~E#v!{;%f_dGDoT{WvvQ8wQ361SLY0ZXHQgG_0Lb{FM`i>r
z^>{T|xi0BrW|d=oa3e@FXdh&<bmqRXxo6Xt7kdZ109G!ZM=68gDE^^T-D{<&NrGS@
z_Duwl>Ne)+RckjQWa1U+vLvVc`WMK+C;QW@zi<Z<#a317D0=-#P!b)^297d2$>Y+P
zz|DF&#(geFRpkDpOR%bcdo21;=kbYz+l30I+lQT7D;S~Kstv`J#nlpWumzTAVj0S=
zL__(`Tu(Yc$(A{ygq33~YoEJ<681b@N<4QM3$gTIptVu=y~O{>ATXAFEpOBOLObvJ
zrh}yqf?8q&?Hm*OQ6muNh2@IZn3&$%wc;#XAq^dLVI7^G;$eo?rkrM)MY}uy=p^_D
z6o8%e#N$vAmmY|w!qodS9%05lNS!nB8jcb6nyE^SJlKN=S`Btk6I)QWB*r;(2$-zv
ztx_gykPK0`l%irJ9|~+U?C7j)=V84ymE7aTBka+e{iUxKP#(c&Pjnc|<Cl!)d2?y4
z?DK&nmFV}ru`aS3jX_7+6fE&!5AXmL5WN4#3gcJK19Iynf<RVA29~+<PcD||xMA^4
zeHV@Dr5T`ZEzD^c;R4(-I&o$G(FNMpdW?AN@UN(N#p9cWDv|Fmyt9#&T$m~!031yS
z28efB_QjRLB<w|^d;uXko|CO!t5!)I?YvzE;UpGU+d*CA7K(@=0}odFLsw8U!$bLT
z`VfWQghB{;h0~5a7H78I{6cbW3ULp2d8#CvX|8U$u+?m3`13N37rrg>A;SHOECqb<
z%z^Wq`RpQu)VOi>*BcGld~k8Pj$PC`@jke|Ew-&j93{<Hl2_)x1ro+gqXR)_)gI9~
zcLhe=a1Ss=s1z}{uYlwhrB%-@($Y(jIz8=IT|uSD{^`SVn>Xgwk-A2$!hU!zcrYo}
zjiJG2<90rohAEL5zGCENTUM{cA?R0MB2z#w=5U_2b|Q<PAnIH>M^r$4tyCON;{8r-
zRW3TPSKqobr`ENpEENs~xQflbZPs(oTb8jSo!9b>sD!wt7NYx5*1wyc=!d>esgmnd
zxk^~}uYlfpg#!QjOO>P3Y;Fq$$E3N0->^()L!H%SD7E=_jbx`*6<ZoL^7c+@oo>Az
zf&=u{Vq3G$(M<aqpN$H-d)!I3&1Mv_D&6Ume!g5yGgu+96Tj}{A*nxKE$JEw8(;qV
z&fn=n5_lDeeZtS@r+gV>Y(VvT8Rab@edh?ydiDI;4q%pSxhd(WmgR^1-StZruj5#M
zL&U-P$<s@`ttTHTWRSn6N|A9`-MG(n&_kk+C&x9REi{C_b#5Z!dC@x#$G>AI<Kk+3
z+=ZwdXdsv^w|8B*KAY1QayRFIBOr`Sz>+Gmrr(9w_Tup!`1+>-gWIL+&#$(0lV0J{
zbkEh?W99D3B9#ZV8d<Ro{0$!1!L{~X&@yqQV`G%CbaEM)_Q0Me0jf9|QtVtef87Er
zeme6493;9<IiI$MmA$DP?&n`psZwHFX3&!I!GZ2NN#`j%Ton=f;s0dIHauVCmj(0%
zbKT4N4|un_;;*_Ta&D^KjY~Rncz^1O8@f))Oyv~2;}~KHg3AMs;&h$H)UW|xNTq19
zLA(3zp}5=2T9(@s%#L@UrD8a`=d*-Q+(STzuy5ZCPq%PQ>|A;6M`ov|gCO0i$IkeH
zg66xaz~z8fOZ-<ke<Y+6{`(G~o*|}2qzLIf(a(B>s#@vkY>MZ21LB<k=J~UeQUXxj
z_K^1CK_mhBhS^L064xh~jYe&(-D?!@*`oB2?BFVA`h6<TZjH0bQyQViwu=%qTPZc<
zq&-TCV^`;`bT0UO0xm=)m3Zzw^U9*g-*E|c=}hU~bf0MM6@M*x9K=WTc@n_3f%$Kb
z79B>XV3b*ydYXDaeMmVrJ4#eE_f!D(sLIsDBTnfV;7+e8_P%j@=|Y86!RRpFem5Y$
zQEyC|OT3eFm&Su~1x}|ZZvE(1??Ht+XZU$s4{|m-XPUkJuvU;fX!zQFy|z#!?rLjC
zdEal2YEO&8-IKJA@E)9{e#J5VWmize?G1{<G3vZ^?;OKrqGwuaYFHE>19_)wvzTd^
z*Ou)>JLFk~*msO-BI0lE#f!Gb7NlwN7U|3*%O>11gBME!oS+gCS5%AT7)K^}g{*)2
z?UE}CS?aN0ei%ILC-PL3sq_;6EA&(FiCeVjvT)sbCr>R*K3nRXwp>T7Zlz=4hNYd0
zL|c`KEuUyAj4m^8!cnV2k2yF<b;YX=+MX-77cb9(bSeC6A==DY^XB|RQHRw&8{3s{
ze#T<K&ttf-o4)R6@pN@B8FWpB?g}UKE)2&szU5|LUQvuM@>W>t-Pnvs6->ChQ^`h0
zV!e4GMT|g8E8I-_w&`XOL5Q<f2PAtYePq~MLU$0?Z@FEl>9TMS&e$IbiXL_Ybp;a>
z1dfPWuqzq7GsP6>IRN5cateI*BFT4E&$JF(hEfSncC06x69Ys9r9cG?cv<ub7<8No
zt&)76aCc&#2{ZPUlEWP`MR0%2lY;fL4tsLc7^ydO%=;#Rh@Ec`#Ms%4|LPLv`EM>^
zHtzpx2L2Bnj`_d3ggLo6*#FlZmg<n|>IR)97&zjx9Z;YWLMA8n5T}e|#umQ<j4jfC
z@Is&#b*gmmk&DZULKJDJC#On^@;=1ges6z%d9QT1=QL0Ay!+<0KKbQNO_&?e|3;AD
z>04Bh&i++LEDcpwVQWW&2oWF-DME}2!c+^3ANK2$1>9rG2!|d&B>B|=L<0dKdT@QQ
zkDFf+I|RbUlla#_fv_>`4`Z1hC@|DuU^3tE;RAm_milrXsg<=*%0fg3>OcS*7;Dhr
z?#?5-6CdZ;f1ukTF;Y`g556&x)ZK!NcTkwo%HT(M1~+anNsJ(7aUcT*9R<Io`>Bqj
z2lLH?gS)!AAo}gKMKDHWcCmzJ;bUE(aQo|+qp1+!K6sdwkuRX$1u}?#VO#AXjDEG?
z72-Pqj1UmQxrqNWbBc<)5f(MZA@1Rx{fCDuiwh<34+8puzXbR{K~d9Fe_Xrz7x+^K
zlKDdhk)R~w;YL3MkmdY}b%a1DtF0jmzaO~;@0UJNBE|`j*xm2XHR1-l_yzrOhL51G
z#R(C33;y*38NR=7j=PE&EB1w_V0MQ6#h@oBxT0WV!$=w*>!wkNgBuE}-=Q;Z_BshD
zQP36d=Z_WZNWOS`jG~O(6i<V8bp^Vh{tn)s1>(P+ZN&YXTUu9C6dxkE9o%>e#WFwd
zRKaJq&QNhT&*B3D)Xxe7_avMM3=70PcvlESi2g4+)Q~a15vbqy=Isoour>}%h?t`=
zbUHdd><b^?7J|!1_Xv?7)D>bA*ls%&0>r1!*QaGrDK<uk_{#_4H?T)ElQtALb=?0m
z_hnlU6Em4>R!LYKmztP?5fTFnD+xqaRu&BWD`Nx%;d=q<<XT&yhzySUvs3Sb^Rv=)
zMhIf(oh(+b&;Mv%Y*z;!{O(i1**2aVCeR=DySMy{?!e#lWrqG6d+fWB1c(f=mzH&%
z_U+%+Z|@j9Lj;oBM9AF*;;Q2DgZLZW2I-wa*8m6OYWFD*pn#9V1Yur;o;wLPU=m1N
zzwO4YNpOW2gxlwYa=e}V`jhLEh64g0mQC1rSLhnc2LtATNU*DixaQ>@JnHz62^)ba
z&wsf)DERkpAs{6wBPfI^vk?&pj1m>9lhf}!W<b1h`ZE(HcnHKlU#}oUcpeMLN%#PW
zA7Lf4GE9sR90o?M@yE#v;&3PEH#O*OpV7^<p?u%U_sUyZ%|C}Q4#dqa0UYW3Z1I=z
zx>Gf4Wm;d_?nL7~0qSk4@9th3o?5%#G;dJ`F8USi(=2UCK6}u!x$*w>hmM240O4(m
z-z~QX<)b}P?2h7g!Fa6jR^M1FwjwhUb>y(MIxB-of(2iHe9rh*<j|Q#;%n!<^4LG%
z>e=By6w<}Gz$xNyyt<e8Ou7_I?85BLdJzosv0=72yr!m-*y3UBV()aT%vb=5M89dA
z`dUGTk`Gq32^*A<zu)7FxnbrBpvO%qMqC4j;GV;fq2?lg$j{;}cvN>vI}nd{FRa3x
z=5{3`YGc5D5@lU(I$PnW*C%(D>1^(#s1#Q|pXjAytPm76gMtL-m+H9r-o72>RjyrI
zd4hiK*^jkOscpLs)LKU6dsAVc{VcCEl#60>vQ_T%WGCNHQhpM;OH?nhq|k}4`){M6
z{%o?bA>(P$!77xOwsE<VsZ*mbvs`&4yHde=fz~$XGr!tNgQp{DnVdb#p@{Me)hWhn
zxgK8u^po}{nrs;XH!<^ao&X1Tft=wLr8U7~{SXkx&xS3E09|9*&^8skh1LFqG0Hp{
zJ=}rJvMWK{w^igR-a<Lql0&hu#~GeSLprr<UrcnMoYL+0`Te~gG!4KwjY<$7?dy3p
zG)oM|*7li}!(zh9p2kFN??)p!+;gwuGfb`bFUaU?01r)VHo@7(*-GzsQ|+dc+qUVx
z&iHv3irG(B+7P4;!9{PR{NjoHMV*DcG}jsT;q2c#k{R@U6D!nu&(uc@yIghhznNUr
zb>jpS&%m94Woq(6uVdJVc|FqtoM55Tu{*^ZfhB5f0Hn3)=aU^TICsS$7OiFNH@9+M
zn*Nw*YETKRS`n{Ok?YM@ou4fZTVI6zL7jH1H1c?+l~pQg#pR8i8!N2gqs<uwFN}SW
zLFU*RmrXO5wY1J<?RT6t+hq~%5nrwhZo>V*-od^+<4QN1y@aLrk;3X46)auk;0^og
zEl`P#<~t>grBq{bN}$!r0wk!+336Cb%g$Ul>8CjuU2NwvXbUbno-C+em*|sOQyN_(
ziXEq%4!^NJm=_UFr<0-l+*fq$*F13fLt8r|#Ug;Vd>$Lx=1;<U8V2BW@E2G6kUdb8
z2{CNo4lvPNR>Pal@U+E;ovCS`Bi&JWYZX7gkK~`WY-C9mP9h-!4W-uBa*u)Q%yCNj
zD{swrRLptpN_;>JAt#x;N7jg0v17%QPt}gi-xLm??u9~$%~niE@$ttCcv8BiN}`B?
zVChfubIt)5^*sfNqyj9laA(+KuHuF=i+y-SgWdKjR_Wh&*wXU{^~Q>X7}^t+nB7T|
za#XqMLNsHcum-z?W1AUHxyjbMcE+@rA<;8KOPMM9#Z46@8GD}AjZXZJ5&S#<^w&c0
zHL2&(Fo}CDm1B%CbRS4uQVThT$JPWhzLLYq)p{9SF@ZUG0-{q|DMlF9+WsZ(amwYZ
zAAkq`I)md}PIewlwR$tgGT2xzlQjgUW-4o}+T#KWn?0qA`A6SVj$F7G<x^<@?&AHI
zpLb=AKLsP})n{4=n8K@r+{YJwYWA+kkR<G>UcNlTax}fq6D>oR%k&h~2XEHK5CsP5
zh8p_6nZcdl>Zg(!zW-#<%j@YpU%~KCT{jvq7v(cx_@4LTRmvxrwB+`U>M1VMoBWOj
z`NXs6n)fXP?Wf@I-FBrV665)B(+tmxo2}SmTaV>w=C~m1tKFa*Yi`oNiIoG*!1^Bu
z$~^s0Vxr~(kCv4%9;V|k|JR#g-*!eau7A$rRj=cdA6jxG@f6ZqGj;7b`7#=?D%7ji
zgcO(nGc12TU4QuVQ~{w@^~;HLm+fO}p5Io1^2<`-%M^zdgk?Js8&PB4hp#%MpqGs6
z9y3JmWoe_a0?mvP4PpBOV~RFGr|Wf7cE7N**<Frf=5){HuqOIA5P;E?hhXObmH5)b
zNzb~Mnwd7>>DEQ|w*?oi!s1%7ZN#AX$L_wb2uX9GNa4K7<VT5;)O%$*Nx|ysd975S
zr6Ood39o5$D!ng5hYhN}r-Y{GS&%(I>+*xbf?Q0r0+ZA5T3}XJ#}`w(-P{<}L^i)*
z;PpnwDJ+2JwcagBv)og%?8V^9SHZfp<Lg|<s(k)=uBXQ^@Pu7T?zy+Tl^aL0JId+H
zsDdq?AKf|EPrrl_@!rt*%jzcB04X9gZR5v@9QX{p;j2j6as+NgLr>6qS)v%EvWnMS
zf3wg1c+xUoQ(o=jqV!d(1x8oH!(5=5drGiS`@?Tx+x=Aj0q*XB-6HK(Zl`d;&k@4l
zA>pmNJ;sfy@!K|+WP5vsuK;*2t+fI^9`il7IMVb#sBUq~p;od@-!T@^O{!$g{VeBE
z<b}=*rHp!7qCYPH0y}S^GQ@j>0E?rqLZwngdM><!uAoLMj|bR$G2X0Z!5TBU&A_c3
z6Ku@!0&47)M1{+eUa_gzZiVd%+py-}xPB#F1r$bf)5ItsWmU|1h^YOmQQbknQU`CI
z-$zOsY-O}k!HZO3%p{NjjnKEg*8@zj*M=&GJuR*RjPnjV9QKxK3u6Ho=o!u&t#(cG
zmlrfuy&1C<9~Qk_Ku!saPuK*})9}Y?+FTt=&yUSu2mu3a<>7y)%Sft5QZ4}8g0!#v
z5g`VRu$3vE>p5<hil?wtalip8>WI=}WF%YKuMl0Kmt<P54`s2ml-ggdNM_R|A#w*)
zuX<g?QltH`j&(52?7z7qFe`QYF0EDR3W>G|j*?d4kd?WPC+J8g?WbWWZfWy|!R3T@
z%1*psWB136FZ!+~w=$U=vK9w;M2DS00jZo1yCk!AyG1x9Eo@<X&?-8<|J#|UhZ$cY
zVaa2?E>4)IzEO)jWAucgc64E*$KiF*iHu?yPO1wUv(UiRqTRa0Wv=GD_)I6<J*U5O
z!DrZw;mUbavH-NlCPtRkjc}})4mK*`f{LH8`aKC4(7Bi$paUkvZNJpp(ffmw)$W|3
z|6r{ZzAa4hZ&^(uWJ!@fLDCJ{`4fw!0>6PBLcu@M1WBHyC3!z$&&pvtiM(^)@A?5z
z(e#CV56v739G9ZTW1i)eMJ}%4n>pn|uD=wV(qDfewQO2mn&`MpEmY*A5L_bM<I6e(
z2zb$2#($TCDOZ>OtRep$;%1-WV=h2-i}i*tUi|K4Df}~z42^->ZR0(Yy7XZ@J*0dc
zzFKiqzH!)rp#M3b!a!?z;tvx`OCbf)z;<SteR3FJbpeT3O&DQv-Mmh>{YyI8)=0<I
zP}i@Bprvj6R9azXoXm4?^z@|&$LskmIwVT;mxtzs`X$#{)_8pA;^jhk(WyMdKu>#p
z{JS4B+#r)z9i`2ka9{lJ*Xh80^=o=$hAG>=6uO9RQ9sYPQ*ze{gFP*4UlcNx8x{}p
zRb8}`Hw)X(@G+Cs3@OT`@O}0HLIiR8H}W1_pFdVXXcT<?sD3uG$Ohd^J$+9_ikDvg
z8<(Py&s0YnN0)apnPT*{Yv_1C=3LrO>Sd1iE`s?C<!n+<LCmx;XqfA;^EM+yJ3Hd5
z%6${EpVqZpB3-)R8ff{U#ah$Jb-dh=(lcI#eCZp<*=bP({-j4y!X5>7xC?Re=xN)q
zI7+*J?s=Yn7z;v@*ehF`Rg+qcEnn36&IT$t!cLau*8^zMqDVGMZ)UWvVCowIXKVr1
z>Pkw8wZm~W$+qgA27CInuFj{wYb9&vHk@#G#uloCBrzS--(pSNRFrc^9##Y(<L)Y_
zqG0c@SVT{+=Uj@jIL@E}E&8^<3$u>sd3hXaOPZ<+WYd;@n!CjI_+1N=8LhsLd*IS_
zNXH1O(TT|>g%4E)k3R0dk&{3o{w_T;>PoYgJyIn%&Ef^Vo~3x&(qr@|JM??<vemxy
z7bBShwVS~G?b{<1uDYt*Hs|&nYKqNz{-nN>^cGO(>5^#z!z$u)smoWAlN5q%h5M64
z?PEq}b`KZMpTK$Pv=CqKlsxpykf2?IyPl(MuEQ|a&ISH`WA~1*9@+VweS{tJg){|H
z9N%}koY{dxrCE~??O?G%hE13ph1)nnI58oZnC?Od9bh`nWrp#v7eJ<I?!y&GnQADh
zA(myNc&JNSMVs!)RX@UkF9PRyx67{5;S>v`rnpB<oV<Hn6sP#hI!pOCMvc$uiJ(bZ
zvEM^+ds~#W?In?}p?|{!SzFUzqoIF%$6Bf9l5OD4eQ^YyzAg>Sm`Sdqj{v(B<rlnf
z|B5GDcWki7)Qzxr#%2%dk49tMTA+-!QsV_CK<h6lr{QO#nQ=4#97j!dO?+ifm%P)&
zHEbkTn#Xu2ph5LL?=ekHO~aQ$6B<iId5|fcBJ|dz5jKr#FAkHFxJm~z9rOg}V(L3H
z4I5S7zXB{SuF@8Q+O9=)&C=4S7T}agcC~dDQwRe9vOK*^9J%}A$eayxR(oFqw|_X1
zVaL`aElCE0llyaKbtLu(Da3t&BUEUToAVJ<iq9TPH>7Ko%HF1O6&cR(z;^hbhW)H4
zu6CSE&_AscyoSan7`z-=Cp5)A1~q^&AVvWwg$T!+N$n?RC)lrh?5*KyOY?bu>fcC&
zY~bq&;E*zfD1iQK3?<j9Gt9hUdp+0QF!l}XbY7BC_qq%{UeQgaT^}mYI4MdSwUE>M
zb8^yk?wS<f=AD=u2W5;MfpQG68)_m1&%0NI$SkR)GJ2+Z<n&LVe?KZNz(gBroqcR|
zWFk-lY>{SYNxS~VSk<OK;jOLB+OW%E1WDN+IO8zc*cviNi&j2<TUM!6lHS+C3FY5U
z1G<iqlzrZOlf>5wDD9FXTwJbXaX6jpmB)P~<?QjbeCqy<FsM?^xsY+=Cg{@$)}O-N
z!IXx`*A8KO++zqxSe<C)HLnmUF^biAb7NO!b3zq!{@Ej@8Ida3RY#=gL6AzAa;WQh
z(8KU>C%%gNZqFz=tRE+gXUz?eBi_~g>~?6J(ON+721t=R65t2$Frl+r{8$-TZ;v2(
zlyW-F&A2qFX40^?+07<-ljG+Lt<DuqbSh#<@W$Fu$F2+B*CV5*aVnhlt|3s!rgMnW
zPH>h_TR(AQ$&NE{?D3&Qno1^vmw@NmOup?|UV!Hn!Ji;9R#~J82WrF7`aD<<1KyIU
zW-?b{$XTcUJG1v;T$Q+~<VWKr9nN7Xs~{)kC3}OC{|aOr;%MGDOirmDYu_W*UubhP
zjz;<Zezy_#PCQ>-IbG~DJH|c!x}U9T%RXTJT&q>NWoI&D7f<_Q%^WF9&q7)AXv>Y0
z;*PXiVBx;SQG3(j@h|dxpImB{dkA>Y0p)))H6AV>$akKnl(vPwmmxJLIy-*UKMxr{
zPZte&GO95^ytk-W*2%atkMp+WdZuaE7z7-c)ByO23OUU44qh+VQcG2+^U14QaQxd<
z26DW1C@v|kR|o-@S8)O)S@zemgRjZGiHbD@dHfo)MfM%f^z}O}2Qu?wbpaY^r>E}D
z+TPOc>DJNptt(Fm%t5(MBM;SgZNJWSkxK{mb^$EKCHcZ*XZsQ8Fm6&DGG0Gai1EK&
z1GrAJnTi>-YhyHLmhCDO9y7*j{!Zw1{~>#sPIckl)N1g9EOL#Hvls}4yF$%jPYBLv
zAAc<gli-$18jYPvLt$Y-HD`-dghZ#woH$5ia9+E<YmgAMM3h3cs?wba%V3)uLMlIq
z?Zuu;{kb8gfw~!sTRPd<QM5@4{pDz*##YC2X4^HXOoR4XaKb3tUk*KtUZr{sl-Xf&
zDxPrIHy%mZd!m2~!#o8keFkMd&Sw{3i(Snk-JDAloj)^vgAZ03@PzD9c*Pfb+Gi&F
zR(5QKYsu~lJ_o+Bds(3L<YyWzr_Qwb2lZocVXR5<uF`J^XKzWa^plDu?z6LS7f^fj
zKL9sI%&$_zllIoU=q$W#N`ypdQBd6XWn5aBw_nyh5|1&u*LJ4lY%yvD&^ZpTUs}kw
z_!854dRliqFR5pjyS7XjOo~>V+0sqeaHqclIaJ0U0w6YVj{kbCO#G`6%$yh<=sJ7M
z^EJGHr#xpXSIK2H^loAMdQkLyo<uoJZHSkdvS~^KFADgRgQON2pCWi{NWvO|kI6P7
zdl(6@+Hmg*r|~B}kBY4{I#MVF_}EJ*jx9qRdpF*6bb!9m2ZuGk9;#{0r>t4s1-%dF
z&xX;sh57JSMaQ4YUHkFVKkEzp{KG1RDL`gK?nHSU4#IYdqW9aox@0?7`3jG4X~7(Y
z!0ZtBRcrAW8p)6Dh*z@A^@k?dPV|Fnquvio2jHTG9-6N4K;M3Ei-rsD$mhST%WRh~
zF<D7NKuJ_MiemQ|zOg0o!Ol5Di&=dOo5hie`(ZTiJe~@M5`yMm;2LI@Tij}M?V!U~
zU4uUve#eho!XGBKDQycMpQfy)1KHW=^L1`B$acmd`Bi^AUaJQzLC({dGm}?UY84&+
zMf69RrQRle0%<m3jiP38d#Js*2)?^UcJF8wfPcDX3{<R+WT=)3bT2zGDu8Y_Y<W!p
zkw9TLEl%H~#s|UOu=1A7fSE8R8tyInXk`x7hY_36`FvNDEu>(6emW-sb`i`pWs`@*
zRMs#$AFV=}@r$eF08(YMuEE_%AL4-4mH8nJnU^k1Dhc-r&1#k_@I!>RJzc(pl5=j<
z@V+d(8OU;cAk$~O(w}?Y{V1B*Yjkss^G!4zni6^ynEWkA$l}H2ih2BMLmqF6CCm>`
z(L0enE4-#7$%tRgcEkFJu7y_XqMUn7<`a82<w&edavL|CCXa;rk+ZrN4f)CuUE;@y
z=c9NOVMM3tc&UAk8BS`0&`Le3@)mL}2aS?is8h&6Tou})TvAlZ=6#b;;9i9-xn;2G
zMV4l+DG@+-71OXv&kM1Epx0Oh*V$0(6qze^qaTMi_R87^BdPQXzue5marvx<ry=FS
zs9h7oI~U}n)bDP!d)b0$*cvC?P}E(og8<&C7RUVro<{)=w|5beAblZ-SLqMo@N=Ds
zp_+&vPAXGgy^)qi#B)SmX8)Q@%EL)^E-2C4>43DRq{+S}y*@g`Gp4qd+qFue2-6ZH
zjfcH_j*7lO0jFS$R20K=zow-TouNUTM;7swM-i8PDanPb3D084)|vK*w7^E|`;-%*
z-GhQ&aUx2wiUN%dfS*hCsH*=)H?P3W*J{wlVJPIPUuYvl25L>uXs}O>+Rz#8qHOY0
zP3e*UlwwqSLO6kfcfVn5I7#c>@K6=)$`;=<$Lhm;+}Nc<lE;bz1TE?mJ2~-+d*Hl(
z8*GCH(wJ+i>^$tbp=5)m##%#kG*-ezO2Uf}sGk5BhMj<QT&Wehd>Vi<u+ZvgdBzeQ
zz*eV9Y)`?l<A-xI++5?M-E4-HOG%apsww%SaZ;s@0rSeIAO<v2%6Wn02yma^tMgm`
zvSJ#>NZ$4w_u4%|p*=ZYLuT3VgTX+h!Hu^phJt*Z0={a$7dO4&*5(P4wVJ#2=@I@l
z>EE+&2KsMi`u(7kaTnHn$-4R)`)8f>1SlOM&%~Q*-gQ2_ycSzP+z?xODV&&M!~wMx
z@>KA^nN}i&=mm2tE%JaX%qVFTvjG=z95_=^vAp?mmNWpM4|;NdghOT~mbM|+ihF+c
zTqb_bO?jTkzkBd!^kP$$vD`OT=aeaI8FAUI_EJq(L{S!ZacB|99%~7t$lWTDX`E_s
zY49W$o)MTsk9<_Eim7ARiTJP=_F&LJkw2AEjP4YCd_uKsFPU>Dj{q(5cEWe29uQqS
zkyzrik<`&B7y-_9+<%HKXoi|pOlvL^?5ErS0;iD9E+b=<xb|~@1vQb$#Del9fgU~y
zT9&)sZRcSpT{h+2v}5aAE;`Qt%awWIm|TRIPCW9%fm)i6yFnaIB7{jfD%IIds-fio
zX;hMIKCzhrifjK(K4WM8Z}J%n2lxN!(qScLW%++Vxc|Rohn0tg=YK1o0o7pnlfAJO
z#ab%K^;Ek>W9Zq$o*2BPoqm$AO#HhAmnZo~;`QYkLP$|b?1QNp8ZcC!3?jxgbUIw=
z1kvDpuiNCdFSo9_dbcM(W<00=vv|hgVyDk6$az7VN*0WO*-*<MjzeLy1UGfP9XvUr
zV0GF<;Y(;Lnk`0w9ykRH^ta4bAlhsseV{XtjvXLKmcJG_W*5QP>$MdiPLBF4cE@0#
z!f@)eK}x6IQTum6(8Pm5!K6d2NO1`B8K&ZpcAh@%sp^0DzXb|Y!W)Cl`qx#0z;Hz<
z7CM5WolphV|7L5G55GVLw_;a;tP2)qC2%xLy0z)Tt+U<E#NC!6C>TURg_!+D|NIU@
z<%72nh2TU;1mTQ<U<Lh41Yw8N4#r80XtqZqjSv8a8W1F{iVHt>jy#9}ju1MyYZY&8
zhsc{0O+u=UF^fDB{{sV&8%J{vWKtW1(1=BtG`I*PD1tbMssgtUBtt#Cd6{zv?2kNu
z%qYSFXt-oioeZ&P+}}@>B@n2j*q8EfXqiK4%l>I`%AsGNyy%WiO;8`iaL6%Xprkn?
zqH9K=u0jwp1?a2lC=uTa(EB8~@W1+v(nrNW5_sxxF!-@|B~)3`XLVuiu*AvyO|`_-
zDfTnS;r{4OOsGMQNd$WQtn&y+7J}$!`zI;s()%JlVhceu3P)}W9hpv*C?oBbhbMMn
zLU{1O+_>Ekar0vqIQEiw_Fc2IFcZ=YMoUT-^?lg~YA-7@OFIV}6$tHwl@epttQvR@
z77MeZs5n={3KN74`yIR7hJ^>P118q|+cx^QwN4CaA*_q8<s^gk=y3M;?9BJAtKXi<
z^ai!(!d=mn7!`gJfXtZfwUhq2Bxd?aj1e}O2?2HgG)&ZIHG9I02FCo?5r-AM1m@cH
z!4-7Un3fecSLfjR;a_jAF*f5q6NATNRkzdjmNbt%HO|cdDfTG3(dGSR?{jaHnY|)1
zlqhOD2vC85Kv+qQXm)M|>76g}j|FQ&dF>{5!DQZbGXrG5Q9SkC1x=0>4ncfE3hZ|L
z*w3q`hch1IqGF1|a-J5|Z<TE~@09Jq<iC=OH#I4#WKu{kJ@C~C{;^l&Ff_$4J&KG=
z5O!_uxX~|awzliB(-(p(r#DYF-|NYIF%I-QG)u8+K1VuZhi)Q>Q|{#tA-Pe;qkyf6
ztUJeq=Qbus_o0IAb<fooo}VSC%gBru9II=uTlV)Cdhd$a9tHy$*#h!8)w4k^QYE#>
z*ZJzn3ZS>v&b+>3?U=|hr9rZeTTBq#c%5h^McLgE+2E9h#dZQ|1&q{eWRloCil-~7
z^QugD@C!H*#oUkfjZ{^4BxCc_BO%Fk>r{~L>(4(TGGJtiCltj}9>(OWV-@|&KN)a`
zKL3ZYb7-%G>$-I+wpB4Jwsys~&7F#EI~CiuZQHhO+vcg(x!%Up_?qhvtZR-j$C!6E
z!!Q-n%E)0+*OZfKSHsKw10CP7gaHI;B}(i-y-pH7gMHh&O%7}avnou>@K8bqQ?JS3
z#|B)Vn3IhWgMGsoZ{8lE-1FpFt7Oelp7|Id`dc)VN$s-JKs_a)4~}IW6PX$-ES8BC
zF`s80*WTeznP}v$#|A5A-DIn~w)J}m4(jBI>)o4c<rn6+H0qwrCjBsajFXLik*yJ+
zdR(a)9Mbg717{QSmu<njQbFNnEiGXk*S5Ig>gO<Li@MS43JI_OsyQ7@BwR^E&M)N{
z?P{&R_n|kaWt&#@i2c4UeP&`t$9pt%sdq!VG|u&%MkYc|<PGS;^l-eAuC08XT!J<3
zfoyWO_OZO-obzUXS#D=T)Lw2aJ^Vf9!CMWk?#9fbcZ&8U>-LN!A=l#)p?SN759z7F
zf$?jYu5lm)eUh;3f%j5)kIAy|9)N=e#?vbe<2czxZ91uBq}wrjKG*q0QOyL8dP|e8
z5dss5NSgehX<%Li(szfu{%GbuiV48|(CRWiw`;TYeL&_a$pp)2s6`uYt;#jW87iNR
z_@zPrdd1*zF_@HIr%D#9rr0OmxWOp|ahv<@l-r{LqFFFqK_EJ!WKb7lBK}y(>Ukdf
z-l6qhq<w1fSzcuP3H>_E9hZ7G81YOKB%1N@#5fCqbwo^J*3nGvp7&zkY9MRQ_+GF_
z)`iQ8TV-hbtT{x)EYX6ro;prD+3s%>|NgJ!tH(6nPnisT(H+wOv3dJ`4T|ih={!CC
z;NYtrCYT?E7dWGVsM3>a=beI8TYM41<Av%Dach{;YH;uLOQIsm3dQWT3mf7bjEUrB
zU%5B;P@8s(arfuI90N}sPF)i&s4LfQ?%_{X6%7ha53mv3I47ZPZS(YF@Qz2Xw_JxQ
zs=G%2nfl1~fpH!&*7T7Ve4a3&<O8O3UNu0Mx|m!$IS29<XD<@|)?YWDhQ{IK6|Kc>
zWvhz3b?h12A^`iZ`2!vCyECrK7~yiIPbC@G-#_2Iw*`+M=vNiZdplH?)IRFVOQ&A}
zmo7pyc}JBb;L22Jt@OkC8O0l!cZ-A*uuZxI+}ULp;h^+-3@ev>mBd08$np!DL_de=
zg7%Z1MXxvGW*cSMCCBy8-y~GJ^VI(I&-+UIr~Q6U$9!&zI)UTct~h-*7wW?7+D%4?
z4b^Qiaq1WVUb3=g3+bW@X>_F$1?(c+>kPmeiytRogQY)>RpnvTTA$3*EK9*8IbrrE
zis-w^{Q7|EbiqfXadlt^k}mMlSwBj7xC=ktb%BWKG?Ky&`*c;!d~S`rC_Gx;BS_|$
zhYfh{$7mxoj|sJvj-Teg%Imq$)+|;1kahH8GGAf2QzXwd%1^2@HT|*^c|Vexd*FB;
zkLp$X7ukd<atFJm@a6{CNp>x^$evhU3nsm6Nz9^@(=~h8U`26i-cT8aCVu{LKD!ye
zhghwtC|w2UOkQqfUT_@3DJd@R8&z1{<6WM2!lu`O>26kVbPnGFUcG&ev3<3^xW{1G
zgy85KI%}QOO;mv>miAjscgqeKx>eYonX+1|#cOpu#w%{O_oZgZRH_kYlBdU#$}=HZ
zrSk`h7kbs<*z<H|B_Q~nErr!2eusqgjZ8ioPA(O|WREpVf3lSm+fG6b`fnkNzZ1%u
zw|263`#%;=_?0_NPC`<n3+Ew2V{Q_VRNmuT=vefz_PiYz>y_49YOVvU`;1$A9R_h9
zf$e1q969F{4X-Ao0i=@Vya2JLJVm5|OG*ok+J-8}XV<WBcSY*ctn7tH!%r1#@7t@>
zh`j;QYAuW(;I(fX#J8b+d487Z1!kk`9_GBRM+Y){w!h5}YHI9akBx|R&n(qg%+qbc
zx7(Y4QqsJw;J`&9Vyktcml(DWCz%MV>R?y3#JRD3OGnGQiQdJ;DG+9y*`_90UbQaf
zJ<rbl!8P<>PAZ>HNVG<IJ9vXoLAN0VG>|q4U>9ut9_4d(6Ef{S8xJ4@6MCf|YVp33
zqcT7{aJ%vM8;Ffdm_5Z=%+WO}!XAc_K`@4=*)}$hzHC@oAKvErb#XVgFm6!4pRU{7
z1gSxGv~4CX(4HwP6En6#La8RY#kzT)m#;Fl;jJbn&S!u|I)TY?v}V*GL3zXEGtkj2
z;y?I$azY|F*aSNaMfGqKq2<3J5Y?wWoyv8)_G{Xp#Rz=kHMlO{m`*k9T7iUoc3}Xs
zV`q^<1uCr&=xD{TuA4kj!evq;*3ka${|AHaNuSAM<la<BW>EnCe3aNqwDfB6<vB$y
z7RZEfzV~{ZBrCCLYpxvba=JE~pFrAx(+j)42M!a|E&?xkx_<Bqk5{yQlF@OExydW<
zLOn@Ty2m#$wU=2L-32(}CSoT@39V{`mZ}){eY-hJYMvM@Qpup@wj_6NhHIdD4%3=u
z;H}#_6~WN;ZZ5g-sLZpnAHBLuS}N#+4P4SDHE7Xf_wDjNug~}f5B|WqT};#X&-tGD
zzvg>pj{lkOSy|ct6Y<Q(!piY~o$tL=VE_rATTN(G7P&#gU4tpL<8fZw&vmTEko_ro
zNim%vsF9hjbb*GUv>Ib+hH444DAcRXw4+Z5#~II`QyyLSIu1H4_gdHA*I(P+$L;~I
zbQOiHDC!&q1eiiGe&XXLAec4j>PpzXN^l7N9_@O1fB;O)NDxEA5rv-`qI!B(c&MWM
zdQ6;*po$PP{kb4Y>O>$4;6%t6A$hRMFur3W{G*}HMZ(+@{uyAzT!w}qAs@F$q~M@L
z^0uHmK@a?T*uBFeLD(*|AhGYkgaQwrY8Q~G!4#xF^uL9%phi}pX(92UiDIrI{#aGm
z5FsA!N`MRu?Ckj>!k+$xpba6Yks3CM`TZIs`Q14NFU8**>$~z@_H{jI$d~0S)QSQl
z4$eP76a(<x2?c+cwEW>k1{vy~JIZElijK!f%rUq#>cs;Cu8YCAvc0Xl1}2m+w8!0J
z%Z!e5DhuNAl2-`x1KiGx9}TGoq=Nx9ilmtzvM07{xBv+184Dmp|1z{hC%Eb9WgH#>
zr}gdYnH2_)gO_`1kJkY6v{Sd=2*x9BuIt%@S+d0&O9uI4=NlMBuoERnXwnZ2f*0N!
zKimVR?qdH76prGJX!P+r4CN0cJOL3A81yi}TKtY@$7{BRpP9Zjkpa=k9-M?A@b(4n
z6E#%@sg83-{6k{fKy{=Oxn!T|W;pP~2Nm@GQ3NV$L0_)!Yd{R;M!ezOXCs;bC6e9$
zH71pNiNuyxbccZrblCnQ8Z`w87XL+U*5@CE86yZB`_(fI8;Mzr`P(-4#=QiDtPfPg
z*8}v5k@fY2!h3yRODvyJoOzI9{)Zy+&(!WgOKt^Qn+YbYyKnea((l~L?`J%5_<zyu
zm{y*j;FhLl`o(5+uez9bWExmt!z59qfnzivjChEcBjQ^ub>v9k0wZnwL|{>khcLc{
zOzNPm;5X!`f+Nqt2HZdv2qTaRj|<M)e-W<~9}P*6l?yp37+BxRuR3&)n$7H_I(U#S
zcm2<IGE74a6iZOAZ?YHH1~lBeWSK1ZU|5i<(U|A=l*RaI-S5nAB<F^Pnjicz=)zf^
z{~*!_!-&E#@(Buvug}T=lwi{Ei-TA}<5%6MQAMKIAL4-h6Glw$-*YrPEdKI#yv6xy
z$7}pGWfd20ln>_u_eFKYB~zblR7H_*^*y~>P*DEoQUVNpLj&%c*a70p#e@m`KR$cX
zFjdkDk{Q}FmV{9~WU#^@vBcS|>em9?Mzbg-D<`0)Bdi-77vBE3KoSJCc)BHWU7I@0
z*2XqDD|*bVy6alO*Ds06BT{-6Pi?ka4z$(&*?`kC3yU1b9Hg?%*F?it9p?+r%TAsI
zdyaEM-{Ko_`_i;}6lEH4GxW&z;u<vB%qSJ{*jCi?3a^5G+^uhwD5k^m$ahN&nC!QH
zNti5&Rv$_ZhoRH$OYLIkw`M3_X9o6cEEPn!IAZfQ;0IO+nT7LCP9p5_?s=cLr9d6Q
zO=xc|-2Tjf&ErV^X$b2V<fP7K;gnxCHd&VJA)`weVLgqu`K%~Y7SCX7t0z%--tm}H
zdHeJ68s-gamcdO2E18%9TxIzpJ`;6cEfR-vO^+o~xjwdv9M}ESy3*y!DWC{!w1^AS
zwS1jMV&6tmLT~6!vzJ^`__JN;Ik71gP|M)_X-Uq^K$*e*;7(3BpWP^XC%m)JcHDl6
zOkZ4SIZL0){Or~0*lvj@z8^Vj!;(CYnaLG^rUg}PMLsXLPAQXMatWhb(vXQ@q-z;4
zwheU0(=q6{|FCZb)|xkFHOfOR4tGjg8-Z7M8?>M+kfni<D|g&VPIRn&Lhc#sT~Nt|
z1+^6Ax{H1s-WB`}_)T{#hGo}HLTDn95f#t#{`)uS^J7Ihbh_d2Y5pGEd98+*z1?iO
zP2xSI_C=&GLX^u1;wgtqSEpkIl2dA#Ikn`P?M=Cw#d$0I{ht&)?}|%-PR^UQ%+y|$
z7i%$hwD#E!*;VFj_r4T^Zca29?9fALPu9(l)K6tZ?Kvz@;aOnnPY;DI5%<4x*IiWr
zX`=jwyJNHM0KU2I7X*Vs(L=~8nkCBs?w{r(&DZ4eZGEu=9ProPII~LIa`(#Dr#xrx
zI%@gWJ}b%_)6*TLRM&&EWRD+1kCB9m&#V=cKbBNLHQ(43J=-DkmfF+KB6pg;aFU|e
z-i8>q9Y)T>y0Qz$dK3FQQdM6?&iuyZlIrwFUwOlVVQ$mgtuwk+txcqdzEeEcUVpQJ
z8BCMt*~r6IC8)0CrgHHGv}yZfYb}KPGzOJc5MqMz-Q+3Vl7P~ma${~SY<fxojUU)$
zn=cJ3_}rLJg-hj`XK(`GRxXiDMLTDHERT3J$lZ^M@eCS&qE~MunCPQ&CRV4fI<+`L
zqzj_W^75*!(8zvXEFa5pRU6t+IY&@#NcnUg9tu4=x`skr<2f(d#a$=86mBpY`D@0$
z=1!|8c3g}s<0c}?C*>3YSyYuIc^ufWM6YpoJaX3^ECi!pz#>C4F}6g9t`EAb)hXz7
zK&!Va&xV`MvP%nmu<B92JGn(E1X7X{c1M$hZ-;wd(%ki<|0R*l59;?KEr<OSuh~}C
zGcZt&lhw@DA|Xq!@D6gSZ0X5**UE4uPbnwxV=BH>gb>Ls1-496EP<BqTc5hxUZ}MM
zfP1VmZ=BXSC#<Plwn5ygu{z$`OU;_zwD~}e@saJHiS{F|==?tWk4rT*pcX|IlqS5$
zb&`_!@Cx^4&*}NV`(~_|^4DJ?G%n#okGs}8RP~ivC&OPxlfIN{pUp(_Rs~OuGfdJd
zeZo=Z$Kfr}e2rQ%$~H$_Zt^CDZ3az7d3Ib^C>(R}9ynp8&;5LDypuOVQgp~A6s5p^
zirbEOIs1gRFY(T=ow?3(ARp92J4=!g`#Wz_z`LAXLjVO@kVa)eW5d1_&72&<iLJ#?
z^;9Rc`_)vEBY$${bpYWcH#@LnjopLMe2`7U^=!so{sK-X!7Q!Um!8%%?8i3L+>Cf{
z$j2Pwq$jC+^O!h)j$4VS{g*>_{F};>1A(N6bM?7u^UGUwL@<IG8)UpS%b;bWv?dHb
zCY|QMUVR)}u4g-Lw=uJ;x;nxAM@Y4Hn^I2)PPP7(-XG{{vC%CAV+)neh|V}Nqh6dr
zg;a1Uo4pkigFpZaX(^(omA;=Fw><+=IaOpLQP5z&$l7URtf+k@e(zX<%iAY7nEU3`
zdZ77e_P<2Cf=nv|;Ra$;sZ4VqFHGlZKuE3~u4P_ehzv%*fbmZn$KCmN?a(zAF2rUc
zTp3^SG!$q(w-aoN!S@g1ZYQ`bm5+r>399L5lm1lzU7$Upde5g$qt%(x&a;L*Un`Ef
z=0B)qwWjhe`eX6kA^Wsi@b)^m5z&qYc}gMZmxcRs=iW{?8@~aGC7n{jkO>WX1>T1=
zunHI?<z*PMoBPVU&H*GEgF2(l7+o`J%5h$)=|~!xCHj>Ig!+ETr_7Y=FMA>A%n^lt
zO+<474%o^hv)#K3{QBf<SI}T&o!$efKo@+qg4>_tEN9fMVsU?7tmkE@`sk8F&2{f0
z^Yfig`4avE&yar$(M%c$+rPDb!V0y0ge2i;(YqkF*=VpvLM*a;s6DA&<nRxSt1|y@
zwX8kfDG54gY#a7?hp3X7g+76>UVg|u4Uf=do>)`wp(Vl*iQR8WH<!_*??02qRkN^%
zh*4P;a%DF=O}oU^iCRZ8is1kQp1REvDqxI}b#3gXXK85eJ>{IQVY;U$wJncadC)(_
zAh@Wt2D_6<<6=a|-xr!E=+dW7LNO*F9jqC@{OPp|#(q(1^5{(K4<)MDC+X}YC&HqX
zf>^OExjEk0D*R^dD#r5_TpH7MLTGe4kvmYe2w`S|VcfYozVFO)Oef@=xvxzt7LWUu
zC?zomvT5z1Jby0}M!K&SXKAmN_=7l8q%p8&fUx6G$TElWyMS%=6lz(fqiac_5t#Jr
zFzg!sAf&-cX_s3@F}TzTRZuE5&Z{j`N%m6xX6=>$uaOd~#`(6-rcTNk(WkIuoPVsp
z;#%6Q98va_hb-&(o6g&9$cph;W9#g0>iyo*n<MQm)w=*;e#5={Rwf&&b?hmw++AT$
zUy{W?B@}a<HyXvu3@xlRqmkmIrc>$8!SqKAEpn!N`N?LEo~{r^`&E%ecg)FVu^;Q{
z8`H*{RQPn94^55?!i&{i$2K?v%X3{_AQX<Hr;mfe?m`&|p~?v^J=dF1afc0gkkiE6
z#qBDsCn-Bs*@i^I_6F9^1IKr3V*y&bI8wSpBwjd@72Gi~%+14Gj7cgjc@?l$@U(NR
z)$Yyw*{GNjv`r)y4g<4yg9fNxkIF$fX!%(gG|R1E3N_2mm9@g7>=4?dJBm<2O|Qn%
zOQgBtaUMe02<)zkD<@dh%DqT-&f;sc37P#>?RCo8F2uKk<~Pq(X04vGkVMt*p6#_?
z?L?f!YtmBHhcF2yh;fcddgrpS<%HfEXEUmtPP@Txld=DzE#-dsMLa4PBzy1N=#o)<
zHnqe<RlJ#s5Q@5^aH-8rV(O3l#^Gn<5>A8{tT|&p7q)P=HE6o({?2im%a0NY?dkID
zaL3TM;&`@i%WUMMvGo1q&R@noPQahpc>e?c^FnjI?d+z*;le5&yYUzFYe*EGeN%>g
z!Pj$Ga1p;yUbJ%OQMk|rZiNt)?B4{Z-P6G`Ww;-^{^_6<ZX9_$!G%V0KV7^ESE?X7
zv0u>RmD6Ka`qVf~$BijYRy@L=oyR*rW^u4qS1qXYf=RX%m!6geguz5i1`&2a5A{ls
z`{mZz>eSK-=nP76{M)les_hyWA^--0FI&|mK4W$+_Ltwx3Q8z$OyYpuQ8wS_R!ltY
zC-0X>@w)OWMN@EzZ2KVnCSe5g^YSlrk>X^f`d~k}3^W64|2Osc0=+n|lGahh&Rg`o
z5_rc<3^(8?h{BYM#pb2sczYNo-Qw#!P&aqpB}L~)2jYq+L}wSj`vh8M1eN9L=-!%8
zgAO{WxlB{a=@p$N(~opVF~;u=rt+CIG27h?h@oCANm0yYmt!E?r~BoM!S}^@%cqJ2
z0_m)kQTq?H+aOJ^+J<^_*-!yOLmG*}1n2lM$zsmAtrK@D0k$U<ZJVtWRm+ewoiD}^
z2|gL|hA^3fY{``sV|JEtBl<C9MCs9W7=8zY5$gue-RJt2Bo$@1OTTiD24Mgt+a%n9
z>pVXrRy<XU<rkWs<yE{qaZ4s}B8?*w-bj`<f4|j*W2&(t(L_Y2JTbed;B^)0^nrl^
z%^Si2ch)^>mXjHdic2(s77vny<04N=%Agg7jPa#ZO7*Y)%8*oYqSRv;T6LIq^+{El
z@V))A3j#NlT#j0}wUUQzXtn0{0+e6+RYr>*=M$$!IXol{rRjPE^aJZzZB>&$0z6{<
z$f5S?8U1U_j~d1?xY#@o(1hfp`SSuwLvNwYNYcZ2eUd^9{s~i#GNkJ@8Ev@4WrFyv
zr4mn1RWd4G@-9MmDfhyUt2W^NO8xP<7O={p^x@0KK}go(y_lJhJoRyGa_e6ipx_xU
zX{Q7OCL&TAs6FQ)PEnF%ayKS4ucNJ%M?M9kQmn#~b;#lHc+vXlu4ToPC{OSzf@c1v
z2JQS3E>*=>&LW@^+wGhXSw^=VHfQUIH&vVAeJuWav&`W++C(7xhH3m6*PLUGz5DkS
z#No5MgG|A7rj{EgpP`gD1=13@RF(!oE5;%fH-n?-&8j9*FTmC)ZSX|)2lic`?%cuD
zu$Zi9oeAw+mV%7cfWn0FlQ&9cYl4BX;cW<dJ^E8=4a<c$1f5CPn6NTXNE<^ftd4P>
zvG_t&`NYZO%U551N#_+-<bD*3JmXgJ6P~!NUI57^xjYG(wOtF6jot`)BUXVI+e&Mk
zz}3{IG)|B9A~DxoFe8Rg%;bl8cz^f%W*MqOnB*az*hJPeunba$nW5s3CaEMAlIQ#j
zFKe=3yV*O@{dPH)zl&wQb&lJC<0N8goimW=?vu>h>+SH9m{Xz2{dMCAWtvnGU+dT4
zO6=7}&*4kv_*|lwr&W9dZ>Cc10_=|8nL^}i&?XY9!%7!(DD4|B-S3(D&7fEbuqq8H
zHkFP|=Bs=4TT3m{VxT|Oa2q~D=zJmP)y@)Gaep(eW7(%st)+$#e8iY`sPq6^yo~-_
z-+ab=w^sJ8ZFaA*$GOw1<F<g@A!t2Lp{W(<2nlW7E2d@sU?pR5aO<SQ^Irx2+?i&$
z*<<V%m*~5o{Fh@^g*r|k{wkk#QshV6b4JIBB&1tEehCIwG{3%aG(7I;!MM$Z^(r?o
zo1yc>ZnBxIVr@9`98H;Yo<6~t&dK4ILGvPSLXuthQCihrLa-t036XK*GDjX!UFZm$
z-nlh;wG{IpH#7<@0E5*meBhZOzE8De5(`YDsTJdUr|4$t>R9)>W(SxVl8%odJ7Yf0
zO7>Nd1S{9LY~01Tpd!PhgA|qla@4_-k3H3Z@;F|+HU@h}Z`lL`tUJ*mB(z7YZRLle
zgCnAI7{Hn6%tr>!U0&_s9@*k1e*wDdRu9^j<LDPUYBZuZGd5GH4vhZ!WN9pE0gC4D
zx~9q)-{NIbi6fkcH-nA@CTijygb6ro6gO;jwk*FQX80ynA#2PlmRuD*zZDF%2JNG2
zAH{^nq&kk;LoID<Dcc(I%6F3%6q4zk#=ywchJ&$&S&QLHhLX>p+!B+cP|w}l4*ICC
z4ZJl)l-R=LKD=$ibM2>Bp_#15T7EJ7G*y$tVyKemF-0|@EKDMb>$Hn3U#at)P|*U=
zR-65FJo31^1Mv{{?E4iR-ygDIa<Nn}7qS{f*%-N`cz>^38SO{jabN!6>CIltS$HOH
zZ4{j?w+ZU=`X`3awpBe5Q(&z;zi=b+g5LV3B;wS^o{NM<>@WTU)q*48#JHfRjVZfw
z6cbf;rg?-NkEiV>@{<Pm<3|03(Y__blL?EpVZ}{`Dwp-UQde5o)oT7m^0V8E&mkRs
z!M@2k?iPW<DEebgT-r${CQG$+3scC2XZjFe9}bpt=a!J(UL<9?k@L#kB7JLOi@s~5
zL<qK7Nf|JH!w0V=WBH%nD(ioFt4zQD$6I9vu>b!e^8aQhSU8v%|F7PvliELT)$6$_
zAxtr@(-mE#Zej(7+@mR0CuwMc<!CnLC(HDo5ygi!Dzdo3n7A-X8)rI7(GtT^ttG|P
zxXO|5pPkp;*XQqD=dD$4M=n#X(_T|vQ#Mr=hJ{RsW1_BM|CB9p|ClL80)cODki2CL
zqrxk6U_f&~Eg3Xaj(iAQ;S>#xg)~?#qcCy$A3b>_#(V)^gd~`p#viPjNR^evMPMbl
z^5)_>^$7k^k0SarAwL2^XVNaP7;q4Fi1FZGe7lfS)e>baB!0<*Gz`gcWeI!^V?ocI
zK%)_TbF_o{+U`>sqN4Tqg9B9zU=t<$Jwq9Z7^0Oxq={ydLH}TPsgpY+lsw9@C2lJ;
zp=0;`_5JA1`R2f`o!7%T==T-i?jhP2zyU!M_-7!&FhY$IAxH@_@>&K~#1<HG#wA1d
zEAyx<=Hw7z=rRih3f`*>5-XF%fRw)}N>JM}R^)(U&;Fq0FZKgKOZ<S`=cm?ECijJ%
z`yhUEUq{BjDbyRaG)@8KH^dA53x(o7Aj_`|_74oU_bkROP7NQ3$hBjNi~&i>3SyuS
ziA;pbg#!l$fK82R5P+ftCqtF%W%>{R|Bgl)5;Wul@$Y3!ABGY>;xCS?kq*F;5GfQm
zCuclCUVX-DN6jl^ND89(`|#LdsS=~@8nOwQ`46Q{2c0BI&a_;c@L2MF9Q=n<oaAHJ
za{`+zj&6xiA+9q7!~^Xg<`$Gs?Ut{v_6s@m6|N`ho9BpPF61JdGOwaMFZ|6c=-&Zi
z#*krt#wcjy8nS{Fr*<@tDY6iFlDw*CFv@&|l5Ni@VQ81Qi7Vi}O)}hzL;A|#m+BAk
zp>H62-md#>nJhb|Y*ZE0*LtN^G%NW&rE8P7d736jl4w7(TkLp12J2sWR%TCb><J(d
zNluwyQ4chD4)YG9p*Sn>NoQ^mQn{9l8fBdQyWs?+jF;$R{^nXmFGj@|KUEo-|C8F`
zGvdohw5&j&x5p0Mo?^AFg0H*F<11M<8ZQ?dEO$&3$G0x62`4AN5X{sfr|*jgbvS1>
zrc?bJgCG3;xT9;B2pfLS_1&Ga#edd2bydSQRUGKhK6-mx{IIwvvalekiraOjdG#89
zbE17Abj9JJrlqYzC+msWNjS@}eO0e<v=N${kZ$h#`yQD871WTHR!qS1uC#8&3%vev
zz1r%xGjQGGU)Pu-U8QP{hGb_~dsV{B)`h)y*GlKxJ!)C6v^H4O9oo!7q=EcANDZ#B
zzG;*ds(l*G2i4`(Y|M9ApddROn{61#DN<fO<+3>tKWzTx`{HO<S(VUnCk8qHm)vyv
zZ=}FSML@LGytYYH!|c#Q@RjP+2?gF$TRaVlGD;`iY}E+Ma9zWt=oOFMM{2o>K3DN4
zkl_H!)z*4*h5fg;gj4c5G0FD+(_3kRLUiTrhpy-#d8XB)3%6{|H^IK^o<jwj2Eq&M
zb%<3gdYcpZtsmzsQHdqv+Ks!Yem;Y>uA4;ZDnwQnRvi0d^{`dRxcO+RiS4!_`1>;k
z{Ftzvn^zU?(A_}Fs2mJ8rDgj=V`B3o-He?S@x1u2g-$#A3KYt}=4yynsFfVY$;pjg
zf8;!Mn#irgy$0<V$_E0m(za2t*+^bAl39;WRbwxq@Y_DM>yG|hl4^~#lwPG8M8K6z
zUhDWF@B$Mo6Q|;7J}q|HGq9<CY)8KUJA~idCqv&w<s?cYNt9UL{I4~l#1ECLT;_0b
z9<1CVxVxDd>h`F@+FS_VG@H@A_9GT`qp*Ha1=fuCvABF8KgRDrjAz-}mFiYD>MF~X
zM|#m#;t4(hC8av`%2^-vx*y9|pAv{n0A`r=HeaTd9zDC7UdIe(jWZb&mU$!(>Jd16
z1OB&e4GM?)8xv5$WxJ>EekHI}y*R}NOlyvX!jG%d`vqQ#ETU<>bAx7fTyfb@PwVzJ
zM=Fzt@vZNtb;GN2ntOE_I3_b3!@38XfGwWJ`-a-ZLe%AWoODi{Q9tTg^nP(afj^M!
zj~HfPa)P57vxg4`Op$W}x<Ga5;CA{Xd^}h|r~t%shS8;3O`{pm_5yjnw)zhHWo+ZZ
zRX7K{LVu*NnISbekE*o4fzjI0$I$ZoMSMn>haT=ynY{;@eP|g7A4n@KA}#}jEp(3;
zGkSYpPck;`+>sbQe^R_w&UqpT-*Et3aT<?Kxan_9^ojQw>T#Pz?bmf=W5ciD-)r^n
zoy6NTj$&32#0s8mi>F1wgLC(t`y8;^9~{50c0n6itishcqaDjIaFhu;ufCShNQs(I
z0aie-LVlguGf%U(<h7@E4Q)JX3@ci_pYl{ut9eX#y023w?Zq?i36o8tqi#<AuWA5y
zo)HN7Oc05=4r!d`ZqB<8WFzaXoW@mKUW*d7*Tews^5@b{d>5gatI^buvV)ij>rjxc
zzJY4OgWT1H^4f~$VF}*UA3yOl>M)n}O=YHeZ*eeVlMNAL&nl<4G0wvn!;kCUI$W-M
z!4BI$ni$H$uOj$zw3#-44$L3VtP#eTY1*%zxu5P9>7^%6pSjY~sRefcETBFrJ=&9m
z(aNgP!%ZeIM-Cy>WY4{yOck%|RGI2AGkiL<wz<qtBlMP9jrtx*<K-XM6$|Nubjlx}
zs#CQpCLN-Zy3{o(L(xzB(QQyW=w{RTZ&X<rWP~ta!tWW#TV`5FEHEI+8UdA^6?hWa
zGTL*_Sy+x^JcPV`<22~5rzz)zM^7D?mxf<9#~U8B0}XO>0cCG+hXO&1Tey^br29r-
z_J%ag({RPCbQ8w|3n)o7*BLI5xdIYr>gNl8{h^%yogYBO*+IG1Z6P=L*u!$83L8p@
zVAN)8+-~^BE4fIsYa85=!+J*3)4v`()}VB|G#{@`FzizKQqF%sVzd7sU%vSDk(H%h
z;~DU~;nj;TyX^Bdm~L)*4h}Qn*u%QJh5WdRn%c8hh!FFsDXa@k74&wsGpHMdYuWOh
zMJ%&h@)h&`q?A^!jc;%B;keXXo2>fjbJW9NwAiBhO|l$k2XQf#9bKZ*X=4hbTO64~
zjVpBhLs8I!<Uwe}Q|$OqBvx%StR1aG5M3-d-q5ontzCX}l$uIvaYHaxA-{peNkZtY
zMB0l>RSn&!F42PpXL?o3+g)<&bd{1fSGG=ohJ!J`k$Ur_v~Pj96+WL3DdMq9{Feyd
z>eNaO@vm3y6xRm{>7CtR<;Gj791Hzu-x*2B!`Bqkezu;bw|Tlr#iEnQqu-osXOZ_^
zuU?UD8fI=Z1r)fg7Wx7F#xH1>dWy$m$|{W$P7oGKB0_xj1q*)`G`sn!S4$LC#T~t!
z`>mpr)zrA4Okf>wxibP=H-n%NYGY)|QVMN@#Pl?_>&T+~eb=3z1c{Yu)@i3^_+O+G
zic$dB$#U3c6REP&6#Vm&cV8TtVoyHH3TZr6^bEx_F)kkG(B&g)++y8Zj?re%p7=h;
zgvFbx-RB^gWyT+pupiGsIT*o*`&?bm)kTK|@T}ft#U>mXI6wb2T`WnAll&%E8?VvY
zNdHv5?avl0@tzfU&HpP+lU{SlNYV=|n=^C#(vfP>LN4rB**ahWd1r%tS{nXo?2&!t
zT5Dz^Il$C#s1uJC+e&=2WVE%jdQ;R?+kvp#HdV4=LT;9qQtkn9Ii96GMcODnGAEOs
zU4C3r-o71r_|3gR8n65B;_s?m{d1{BXVziKaFZV9*kAI0jqqBwGpBi_2od`ls?`DZ
zlCC~EatB6NcvY%}`g9#+GmG^@<pi%@L{8;2KnlC<Jy8A2(`&=CyEZ7jdNk#&--Gt;
zffr+~8N0X&rav~Uadi~TFAP@jTDp=%_3*5(>eJTS_=K8=|AJ=vo1V)U$ITO@EXLJ|
zGrBl=|Fba+!T;~|p1tcUVY~z;&ml9)1E1L7f{s(R0$NM785f!Mgj?44A%>jY!PSra
zbNa|(dbLnxEkVux>$j<Yoq}jY;%}%i+M_~@i4+PNj*tN5E_WsP;;xSq<g@v)>bT{z
zQ>ApH-LxzMgT$*M<V^z4B$ZPPSKcOt^BPjZZ^;Gy*704o<^zLVVhRJ5@b1T_bK&Rh
zbpwWO<d)NZXQLXr8%U)9HUED7?sXU7HU2#bDAAEvalF_?jZ(d$jEzE54?U#3A+sVC
z)l-f|Ui_kB5;7j@-R{^2jqF7r2iC>lM`8Zr$FLW#fX#9NpRz@S2ogcwFg%x-@OWbn
z?vtdL7b3?)Dfa}AFW-SYuLF8ZSIGIhP}|0QjrUe;%AXqbUp~6!uo>-d5{_mjB2^7U
zCMWa{t-s?7W>eFrav@9{B+1!iE8z2;F$vk2u`|jjk#HAQ5xH|*R#m0X(q%1bX+QB>
zjhIRLnh+ouu6cfCc7O?ax{Zo(Q4}{F#+oj;{0@|EOd_Q5U_RrUay1DKx817X^|<e8
z*HN>%3H%+E#H?9|WsY_gjhN+-4F>>qIZm%2a63&b-_Gs&WGDTWO=f|T19V?1G}>1A
z`tKFf11_tscMKQF=-7C4#=X&!60~VvH%@cPJAQRY<;}8xoUXk@R~8Xi-M6w6hgqkL
zkcHwf1>*yfDx~(cF!T-Df*SP(hWnl5o+4~lWri>eJ&wJV)@=rGUice2;<0cH{!pnQ
zn`|l6Wn*O7xRR?;{CWTQJzOG5w{JA~PNfAf)G2tN2<}wlPVwd(Eat(=%4E+t^1;M~
z)^ECbwdL76-#<Yz_0}~~WpF1sQ(d`CY@gJVf(dBsgHc<rXU%s#URLT<l41s3X4Zu|
zc03e_aw;|~LW<^tMqAs86Dd!vx*%lAG%yENrSMW^thLGLF76{B^L5ja4;}BW+}I})
z$DZ8x^N*+`@JZU#7SeJ;02~DsDa(2DPs%_3M0C%@hcl}gIJi;7RZ&=rXUnfpMP6RI
zaVwZ(8GQTjfDpp{r|f0_FWJk&_&>6jm4oH~zy4SDvaz!>{hwv8GK9MF`4Z$0GmHY|
zfJ)zh0r<^1)VY5Nc{E1Oi5^DBU{*y*Y7(N&LU91Axc{7ALytg#!Wm)9bC$z(>$A_(
z{Y2Ka&UM(ej)hf~mU|Fr7-IyTHiG{Uj4y=1oG-xuT0TPWDi{&+E<9@P-k#!6A2EOo
zvJ$knx9|rWE$U2=E65)h5cniWDRDR{&_1FV#=oX{mwWKFq_#caAmE@q`X&t2jM%=!
zxqu!_>tJCen9RdDeo{sd^#?-0#Ne$iyBc6AH3nwq)x;U-8-p7pkk1W*jVC}}5MlSw
zLvndx2ZDr@fF&=`cgg6R6^<Y9T>&}sT3xfV-RZ~(AVWvI`AC2MzLxIEv{NsrghNCG
z)1F}jVD_x`|9$5i>hS{LL<QzR^EOeEp}X=!hdKWmXdK?fD+~<S*w_HSoT2x_uE<1A
zh=GL=+X=Y9Q}6@0;8hdloWS#gn1}(;C7^$}1oz=j3m-z^?W#i++`l%)PWj?v`tgoO
z^n-}Aq920#qs^h&f(x+qpo9~z;6q~yQu0KwOd*GfaPEP?W6mfkfFX-7{KWMG%fW77
z*~OUxffV@D$0&~l1?c_OWyb>#b%wOHb=lA$&;sNlO+`_LKoK(#Cy-*g0sX>RP_3TT
zoW1?ZoRMkaAzXN)J&It7G2z{u5sq2J7Wg3k0yy)yr@WYsEBNk7?WSze_qyNmkNgBr
z`SGA7vt$Tu*}cC-{sM>;2z+9{u*ATih?R3fnn+7P-k<^;RfXF6_1wkr76|zH<pcJn
z?Dp%s@?uN@0t5GNKYlPmKX9bS@}RzfdpT>>`lLzkey0$CwPeD<`h9Bsuw}itRzWl5
zs@53M-tKw$JG#l&e7;c;7Nq~jgcBw_6bzXbb5B4-9j8wM{st5+4c@;!8zGE`h%97i
z2b$hDA8smm2A+0tbxPd#NA5c^6x7I1&nR(85WzOY0Q%2RabFN(9*+ZEcohB-WV7dM
z-T+F`C)*8OXmxFPWWhWXaaPphTMBCUUoEJR$VY<et9+#<Ti`P&{ya*tj4M8KfrTK%
zCRl<_$9Gx1-=&cRdXRRa1W`UuXVMk-EHBk80yMO9_%$FXypSdcoIYP~Hm`gSs)~Ir
zU_j73B7Xi9TP+C~34TfCjbpx4c`Vu_#+c)5Lc#SfCth75zy>~IAr4_LzmXpyMRDQI
zSc-1*bBX$c$9Px>BOKf`q0dI_T+1uLEGq7jG_%utOaFxQ<`GggPB%;zffnqPdirXL
zpx#cG7#lE&HMgqje0G}amS?TpO{lzfaZ+lmC9!=|E;MmYUy@bl_VOB!@%^)uSC!@Q
zCk;}hIw5;ZXSymu9V>TSI{&xjkVNY-{D|(k#&W^B=Hl58U9*S?+2f%wT_N_tbgKke
zyZo_mHQ$iq<v}O|U~=<_RSTvs-SAUX)m5kM%0l4~4dE%x_%(1s$UAG!YsPHCzEPk#
zZkqBD?IC@(1x#)X-U82iDBJxD3}DK$nsNroDIWW>DpVhfwr_V+#Hg`#@_a3{!mzSk
zrrJiZ+I(#2tc(}b<raJH7gW+2@VSP-OYJ47nslaI`E1&Z7kypXLWGD$sLa#d+v68X
zw`N!_axFVG$}qHPBC=SkQC;b80WkDQ<Ml!RwiLPtQ=l=d+U--UaoSDNtg})Q7jCy<
zzR~v3on9q65$iM5QmDvIN{MsHjT`^G*`*xi0VJ6W6-0L-D+@iCN>v_PW4;>ogL`1B
zUd@kDprv3${$(1YMnFFt$y+lr#N=8<UW;wYYA{}%!uZ!auYJ$(ToS*tI2{+N!o?Vp
z>QY<A2^jSI8iB1`T4PA2QlW{dRAf1@?7lgy*-Jx@n7qvG_8rfZha@xURpp7c#^}R1
zN=1_}lY1a;vujBj(<oxi<-2N+Gx|mIL<Be|CzNRzIV?S<`bj6X<Z^P|G@pUFu@d3A
zN`r&nHc$Lmv^}PbwxW;C-w4+fNV~vO@!0sLx|)()V|1iUv%EnER-2!aKdIP6x_V+`
zR#;eb1ZyK_-BH$>r`Rz?PsnjwBbUXAi5+ejF`Z$>&k&ge`tkgmIaHHojznYd#;^Rs
zT%oKZW}Qj|W1Th1t~$;nLzEG8Kkt(rby17AwV$v~#=~}CNxnr2cd)1F*KY~sxK2-(
zX|wC==f%4sibx2rW>XjRK>`+5*o5{qOFB5=4HhEE6Lwq5ULwFXkWIbHg!C5XB5k)r
zm)+iFH%a{&CY_D?tdJIL++r=WAnfN~T-9~8Fu1F#2h_qLp7mFwan~DCN~5>cJ4f$h
zWqsJSgv^J8^Pu4%ohxe_a~6-W#{EwV5*uu%1C1kTJgfrQT@*hCNt-`6r14d^8t8iK
z98qW=rzw}gMyBZ=`J^m2YRE+@(dN+=+>izpNJXe)sfsME6yClZTNF*?c<O@pFK<a`
zHsjQ<?ibrLV54|ptZ4zYMTo9YXRVd8brcE(2#pIVMNZRi@Lh-{In6Vaq*hL)a~lX(
zUO*wg813|Ay6xmNm-&F|&Nf2zr>yWtSC5&@l}!Jk#L$4Dp^T@?ctafn$3B)bFAlR?
zbvX2ft|U*@w_d0>NRJ}9em4Zat##b#f=9weF5tH{iE5GSd=2oKMq+D-UU?G#`sH^@
zWY^Med%Ey<=XLo#B=z_w7`(KQmIk+y=OJz)?1yKWK&9OTF4XM>eojUa=$x>HMsX2s
zKC6PYuFhfT2LAv=r?{Hbfq<A&ce(Ci*g(RZlH$ld59h~RkHLGk8ufa>zNQNI9aj;R
zGsL!<MneE2{vqUtWjOvPYmwmV7V*eSG_K<U(<BaYhJ{os7?xMyFmi`wLf|w<yi{AC
zj+=ysCUzidX2t0RC~P#&LWbiK3#r;thN&s~Bv3dK@HR&0ROqoeB$RmWpBbcz>g*;{
zZM|I8>$VXrsuro0(7hFJ;hy5o;pwrQZmmv629ixF*pb3Q)^6tbnGYZG$^J36CJ;6j
zH-#Zul@3PiCiGh2*^%7uaDgTLOW5e(by<ZG^NlD8D+(THHbQ_`+)fMOZPMFj%$UEY
z(?iW7^!$#XZ~TtG4`vC=+Vgu|e2I8D)H={`UcPxQ9j-hZ8LyB#QHOVXKXx8tOD(FF
zLQCz6;}Aw4`Ph^+aS~XNaj*T^F-mePFqGJI)I5^Dz@UYM@$AN@gm|~slghMR+%VwW
z(j8v*(u|<$>4laff~nB$LU)0JC0C!~X>|?4>KIMosyQO_n^^A>wzd^z>UaE9sne%A
zh3JcTXc!Bfq*_{i4heXIN-wcYHhnR3(l&EKmlT7j1-DTd_$j7I=u6evPm^+C#cYzL
zTQM#U`CXVTpT>leTdK~@j!TRF>r(^nRJla!hsJqXuE^V&g1hZ*>|D`dL4f4$R7cxv
zezvA-2VZ-xK1MNZ>)sp7lxmf+*uSsixjr3UE3({i9NC^jJWMk27~eJ$3C1_q_I4S2
z-J!HQC1&uoaF>@V@pN{!a0+%)*K(7qUz3!IDWf($V$|USYOcToRqa!pB47f@fN(KA
zWvsDI@)Y-B;C^=&<3+4xGkK2U6sA6x{Y>R05$LTXwz<ud=gdD)--(o_Z=46*ZA!DS
zwH)^iqm;x7g7bD}zmfN(;J-}bz939q<Y=|u*Jn*<ChTrS4t})kjnN#TGa20gAM>vg
z{tG9v_4f{y2YSYPPaxYZx+)09U34$l(D0|-e8*Pa{7dHmAgv{ckajyKkzq(VO(q@h
zD10S9yUgvhyc6y1BHX6b>s?!gUgW{)Jc~DEOUZcjv&$z2kW|X77N=MneN>Bcz0Q;C
z^&VN<LKNA$N63eo&NkObNF{Bv@O9!zt@K;1x=Oy|@bL9CNxxCTtr%eb95!|hSA}hC
z)y9E1C?=IrHNr#n<ubP@M&nA&g6GA_<R(+)Itwr%Hi(-Ks_4Ogkf%HBcn=s9)fs9U
zL!CBbbyC7;7G_ycK~M6e^m@PT%lJB~Ugnb7k6rbAm1f8I2WKVRI*NZ+5jaN&I-9{c
z&}*ginVZ1Isj4)7Y|%>u{Pc7uchz9|>)vTx9F6`-*DdS9skYEr!9{8b2siBmImX(_
z#>oeMfMg%An|_UnR%ZiDEFQKD?&k;5?BwLlNzX*^0+`%NGdShiM;@Xk@6(b={x+7|
za+;HbR-K8w!OcXdG4VAs*R8jQH|{5QI|3osPhhB(895BwTJ05dI_3?3^sZY}sC@$A
z54T)zi4#7g(A2=!OaR-8_B<cz)N7`CkPVC!7EYra0FKk9Xgznr4_oYyD3`MAr843^
zP<uJ&yZUsw?y*Fdtsi=|G34-OW$}FU5@hBlpA#D?Er9^u%f^_;^s1Hvy-O*m2mg0z
za%lGkOTfy^9BNsm%mZ=<$hq30*cO-7isQm}la+nEK=yBJjPr4#_TRADbJEyq8KzPT
zg`pRr5BH+&otl1%Mdz}V8b0Jv(S6K=^Z-U|=MFdRKBXrmz1LvI)y44ozt65nUPji$
zD|hMUjo{%#O=rwv(W6fKK9W3w3!JMoHsb)QXd3GAnZxV~4fCQX3$*u~fEFG(26D$H
z<4GEg>rE>|)@M0RYVM3W8{PT|_*7(wk4mf2uGxrdZp5Ckg$|E9epUIIJu_RTTH>mv
zaPceQNh}ui$?>I5HQ>Q$U^io@okE<VcexwIgDu))c_(_KsqbPlI7AAn*Jp@9POOrG
z=^0c)>3l-la;`KM6_j?zGqBuFN!z#H)T<(q`h85o=dGo+(a-9yansOP1)!hi&x=H}
zmAY`TURr=?shqa~xPd8&h#-a5AbM&3-{=yoAYr(7p`qFPo2TO)xfW9y-qd7G*Sp_W
z$B)(wHZr!JpZ!N;!$Q=154Z#!aQ?a`+yKFmt`6!ypr8m^6Nnc_u-IBSS|r)NoC{{o
z5TJjA`8MUMYKeAGaCPBfOK$y2*(AldBps?bOOq#g1FOm%d=_LHFyIb81QGDi4P;sU
z+1pw(`MxMU*8K~3^$*UXpXfA<lPv3F7!aR-J$LhhB!-nFB41DDE?CS}#pZ{#ks<!W
z=#-H@RIGkB-l$2Bky@N`hHXRi_TU)0G=id^)Nh-F6WG5i?^9+_CwAD=WN5EgS(a)&
zo7lYg@h5vhuwz38D<!^aW(DGkZ;RXxB~&Nes4<iQtro}8w?0a)u96!_h~`<Xzbh}m
zul5^FKKqE)+r9HW$NuV*oMj4;-9bz?{$WYt-CQBD92DVVdm;W5jAXJfoXg4l?i(7f
z3nV*HWYJgz%3kef>~oQ0-kzuI0wPiFMZOvWB6*j7;kJ3;OEl0Gq|T%tvFuV5lNZbF
zhj!wF6=g<Xke3sPCqzYUNn;@bqgb01SSuTDKjKVs-_0fU879u9l<CnT-6Bo<_~W89
zzyE<{=3&EiB=HZ`oHm$f6CE+{F#3oY{lrn(`zfv2odT7Qyp$O7;TzJ;-xJ3B)+^CW
z0mggrk%$j$34Op;{13W}hU2|bwS<CnJc*+@mzW%;1*x$@ebZzcAo~S*u_$ZRV|`6!
zt8WPbGTHAr9rtb*8aC0ZgWqPAyI*S}PYch9E+XZ)W&3vwP~I~>JQXqfaYU5f7dT_z
z3yOLSUg0HNfWanq2?1(u#)vAbGP|L5Lo*OteyJp-Z$X9F5Jm*DweH_$wZ%X2CW`;~
zN;$FZNbV5TSzlDP!xTgPt^n-P@OZcqNAnr((T#7e-c49C6U{GP{e?B^M0oMC+PmTR
z=ZH)orL9OaJ#T|(A8g308Xxswd+sHap^+`r<-J32X?wjZ>k}_H3@eu9W=Ag<NlUxp
zdwnBqUqLBa3D`fayZz_iDf+ikoUdD+=_gK^-XVtqmqB*hy*jxHkGdVm7Yw}c@HKU{
z^*c0r;16Sd{_OghN_Ant4IMgC@Zd8NJ7!IO;Mm54%3e9Z^K$Dh+Z_Dto(qk}My)&A
z^4?ZrUd7mG0mtVLO{k{&Y=y#E$Ce~eW98*{{XQ+>B^E(C+g8JtCp2z3uSwcnx%MZM
z=%M%DHX@W_BC?%rpvXFidpgA4gxWi*kvfnv6q)dxOCx<`!Xn@?D4HeYetO2M`W!t2
zR*QzTl=L>9V)9&EK&n?rmmtFPXuLSt>L|v&^Pb26b=m9}=f}a-Qy+}c)=4vkc3_<c
z#7ln>nn3w&;L0al_W<IXO+gu}`-{#rbrI@symW3WMngaC#cfl!jE~mcq@Pm!>bT+@
z2*w&OlHV^|%@z;B1QTt)%kcV36Of*|J9pFE#p1c*4(1I(?d8Sz?T`qTlA!7X$hoLp
zC+oyq8xq?hlKAjPVUe3~A_td|aLnx85-aBv?c2N&b#F7LosK%qc--hL9t=_6o<d_1
zs{06Bow#^sjR<Vy`?y%z$T;q{%5EO^TN0QG8wBGw((v7i9z{V6EA^2IZ@2&lJL?3N
zt!@jQSjv1l-HX#JUG)4SW%9X;MD?{3T^ZANK}M2?S5%kU=3P+^>**DtQ+)bfg4Y*z
z(<m0z9(AI`=a9p$5BmDW-oHI{BCfr5wq6G~;e$=o*NE0*icJe)zsksd%~4L;{)TDp
zNYiCnc;-5`-^2EERxIMSEWlu>gHv#u6ZJr45q9?C&~9#zdX}*k=%99)7-94N{dD1>
z9L|I@u~ZWzNsSbU8jVEolSFE~IqwCQ9_NA6o)p(+41@Klv>W@>U0Hc`{@;4En`i$Z
zlw|#=?%<qaJpV`j?c21-i4MwYd1`y9AyPBK$}??<kJL1S9A<%}f7)U&FZ!ylf*by!
z#=M0^*X3SVuIQcNXE8aV*KgU#C*Kv-->ttaI_V!j%+-d)Wq&81-6G-j`3RE1;$+5s
zsX{S%CVrj4UT)QU+K1{m%yWfrh%#(qM~~2yFSpq&xvt+Ih0GKsL-Ds8nEb*G&^-u<
zmYIVb{xyw}%l%R7sbkR#vA|P<nkQ0cxi+XH#FCF(I<-lqKi(4*edT)Y^|(No0*6wp
z%@S<i8Y0+)j_M<#IEt00I8~9>on21km_?2_eurx$eXOcJ0PN&XY)J2tug+4r*&R;{
zMU~+|5_N8Sn1AvH-Jj^0^x%C*y}XUII1zheJ`uC_4ujcTe9V8zYJ7zTI<ai8H`9?z
zYVcg<7S=aCFkG2p(LJV1+E8r|>KZt6l_KKr|9KJN6jKw=$4zDCcn=XiEDJhO9Ry6b
zlTxW5xfj4o98RjJ>>l=aoXh;IXq9%jAD88RXfbNZb2Y5v$dL14m;3E~(&wdW0|!>i
z(Oh#&`krtT$4BCW9`5bi5<mR^7(0g`QNlJ$SKYF8%eHOXwr$(CZQHhO+qP}@?U<P9
zUi^#Myb)jSG9x3u%#-Jw$L>O#DypRCy0Fz3M8JUMX1wu@LX04do#XoBlu~Cj__6Q?
zhf>{Pwg97eoo9}km?Ut^{!Ih^Lu}v^q1-lf5B}8x6|XmGi@?vw4UtinxZ}ZR{F#W+
z^Xpr{-#Ec*U2y{ux5_@VIykCK$Y6emxQ(eM_8?)?M>#MvfK@P-YiNW{9GKfZG%3B-
z<3w6KEL&ytq#B+#a>Mzv1FEU9ccI+j2d-Nl?D{{d(ro`pm1bsS`JW;iR(gj2xBB0z
zG#fn=!~eG`?F6ERWYfumFGeXA(RYT}g0LVa84fN+_Qwz!EedAs3?)uPAxVy-7~wpH
z<{Z|%2pD4g_s+HF(Y?p<XSK$1)3NvUaZ9$Fmb9{^#;*%x0;?oIzyP=}08iYHWeEEj
zEubDnZjcY&jtl^JU;s}aQAS2s6b%?i6h9dcEscn87y;G|+8)Sn7(lO%NKHyX3=0qV
z4}^aBM{M>SJS7!a6bK*QKLz?D6c`#*tuG(Vk9`U<h6qqcvI3u49-wZFw=FpFyLTo_
zKgO&I(eJc($sJJ58tO0NK4_hvdIA8l{EJPD9i=6-UntlFlw81Il^lRIDdpJrB^IA=
zPcZQBq9jzXzo0%qPH)%m?~EE&w0%$}r@kHv0BI4(6oOxU2q2C*HxO3<zJ3fm@XF{{
zhhWB9o){BL4Kt`AEd+2!TN}6rem1atj0zu_3M63T5J2@6LI9d7bS)kq4vZhb+zTBU
z7UrK5$l!Hd)@PnRoD(xJ(O+`R7*I8Kfw<KPw7&o`K?GS;dTQ=eHsRM8eLD!j+^vH-
zdV=PKb7<OSgj+a3eC&Ukl9d2b?1q>z0|5Et#h6;4nE;H#jWzP}WL)T~etT!=;Movi
z`ueo!S2<4TuCV4_z{4<AP%$TV5PnEONSUzApKYH%*J?EQv8dtco^sn(wfjdJ{B%?O
zdt8LRR0Z)~Ucg&<d&+pLepf;b=jeite=}kukpAR@)DOBmct8NIYJ$I+Lj9s58Gwnl
zeP?<~f6LDB<~)c|S}_6>0Soj0_SrQ-ZXuiG!;JD(1E#<Ov&{0oHC`eR?untb<oZeg
zg8T93{jx1Kh6q~6uLl28W&vL13NQGpv*Y8%{Kf-Ao8RZA8*;vp+U8YMM1=Sw{`&5;
z@=_TS{WZ4VX2+ig0|+Z59Et|^jIK)|>jR3y?gW{7_HeEO<l`rT40Hxu<J9k}JG3Wi
z?-52c0Kfy*w}ojB`Q;x1<a0zjq2>Dn5Tr5qyV+BNYz8R!B@7tY2Y@$)10$Ljx7RC&
zK=r}<@5?L?d1&Uw7MwEFZ!KN(D_S4<I_t+eus&d{*ZPWm0=84*H)JRnfX_3G5Y$tV
zYA+?lcesP+jnngv`7V_@NE$oyT^HDDC1$0p7eF2W9;Wq!)A9}o03PU_li135ZTP!W
z?aC3%>7$g?w_IP+{#*3QS4I=lv!Bkx=7kR6Z1?cjF!0yfu=Ez(6{HJ5U$^6u{_Fkg
z^LO&!@xOoNY5?E>>EZeK#H66{{I5gP>gra1jRdl>u<Ki|XE!6XhkVz4b->W3kU@#l
zu>|>x|5)u1caJ8`d0X8od(iKZQ&UNi5;Zx^zex&2mrpb>Q?!BjhFlj6&dDuXO%^Vg
zMwG5*g%M6z-Ant!@5OJPlAL4KH#cjKcxusi$Fe$xvJ~>PR#6zRym|hyM^4!JTh7Va
z>{~C04M*gWKDAnGgLb~nSV8*6P|EhC@Tz43*}fVOZ#RvTWPG-*Qo6OfbOy3(10MO-
zcm3uExPxWbrpYxJO-8*15HzyweRn-Qx}cJ-ed={LX^xoN*J4(gEa&0TPMxe%5BCoB
z0&6<xgS{F0w$)`TV`Y`2rf7fB;)I)zy|epCz}SU)QU4@;M{Iugm_=%^dqCz=z<QP}
z3dB|VNSP&pti@EWNeHkz)%Hl)2D-r)k&IiV^N1nhr=WyGpc!?M0NEzc6Q200#JSSl
z6`EH!x>)8@q{bknj~jy-WAUkGW31&5XrtT1+&aGbl;9%x1z9<kS~kIRL0TsSTpSk<
zF3p3}?44-vEByfWEz-Nrr;<QwkFoYv#~BxJ&1U*%$&^<8P)+JzqjTN8@`CY8)b$g6
z_wz(GX)c%%J%ILr_i^^p5JN|lCKJpYBN%o)lXq8%n)D3yh8UrZoqM+X3yJn9ta)7r
z8j#yUsn~1!9deD7#58qf6Y2hBgVKp|MUXvOgN&v`v}!uHoVf4hb)@~5!#kb0R<cPy
zyK0~-mwf4@YUF9gHOcs(h91m80tEY_#&zY6g?a%L(f)`9WP?X5$n$2yS7TaVqmUoN
zGDQE^5K#e8E_W=i_l=m2k}*-ujLWvQV=16uBMY(nmv*T_DUI$~4E>?;3Fdy+PU1D%
zziu;N^PWdCOr0mC9;)=8J9K#m_VwE~YsZi&KE3VI$=A_(Eqe9MTvgNcqaOM#6#Q}#
zif|KytW2K_3^1R)vsIY8r8-)4d9Q^92p1Ll^Mm)AJr*KVECD?e<SK^@KbO^5EJLCQ
zSPrPs80s_C-r4NZv7Jb9hW_{Xd=_qW^>sYaEqTKCPn<zg?Q(6|*DpovRiza5)450G
zM_m5Ho}|b40(I{5kKT(SRbeW(7S{MyCPvZ;T{Fs%6YFI0?9hWd&M9MJ%eaeF@dH*h
zq3wyhiTwm+)RXNuifVjO!wJ*qkLmlpc?nwn839p6W97gm6(+KY;c~h)#s=rq@(U(@
zhi$~9uA}<u1e>XwUZj9Q_Pt@;%5iViB1eW&0K(Zk%#!*v8rd*iWTdzFGj_mKJ;vys
z*J*AF{au>mPe2t=%k0yeNRk$gSp&jUElIH0bnvn^ciT6pGHx%X_nf5d*lID~HqjZb
z<m(CB42X$OL-xeN?Dd>W<0I|00aCfl3^uSxI5U#E;%!c0?Jg?RhU5N!V$Vx9U$fEJ
z_7{dU_js~=yz-%ojf&cblSu&YW<Tlxf=d?75{s!ANY}4ARi9!gu{)hpOw)I^OcS;A
z3KCAM#+S<V(nPJM;vsYUxA)^@uL6$R?(|HOxq;U0Z?;`0+8bcpk}9_XdOo0<jlCA^
z1{#*>5h>Q{-JY%C1<k08WUvtZw<?^ipjxcA-+3~p`S~i5SyA$5B8rj(!FErsoHy+0
z@ooOksefIb9%~|QtseGgOc?;-sdUdX-uyWsP_fSWbgkgwfW=H?85D76f!cF{r#_%U
zV2yNx`stG2{`g7Zxd47=v|NWxn2vcGYX7$ny<<#!YOto#Rl#1G(y(~;V9wf1QzZHE
zdTMFc+$&37eR>7BR%hP6!re4f{3<KXLfU;eo>{fY)N=vj`v5E{Zj8lV&;eL~K&W#`
z?^J7!xVgaI3lmc*7Pn05`T;BQitQ%LjjGo!94SNsYV1!kv+DUBkIKA0<<Uz-unN9+
zU*#yS&scbtgdH!+RRb9qkJj}wLmMJ(Ayi_l{c#^AJ8bHj;zrAQ;!F}{->82JtZ1?3
zxVjALj&zDi`xewx+Bf<{H?Wn34x45Px3LLEnCwF#liKc>H3cgNT=RF$2brWfMQNgp
zGulOQEmra5Niw2q$~{@ip}&H1h*gNu{_p(kYC<k}MH2KP=yA0GdZs%t$Db5l%K7Ln
z9{LZDS3v*v>hJe8B~X9o(3wk&_6-oA!(AOEF%7pA4usGs2^E7`*|F6(7+Q#GEeW0Q
z>{dPT6v;N#<;Aj4QQPMv=qZndklU#@-<`cORBG3M^*UB*ZzYs~aQK-oz<ku$&DQDM
z*?DNC--*t)HRiJ_U6TyWe7{mR0vPvLzk)i_d8DcQZbLxVSJwFMC^ZZ<FaECuyDYRy
zhYj{N56>H}^q}iz7%^Fpn-DUj>*XkL$gYqhl6$$O@pc{>U_#i>se!WhrBYmpayqev
z&(sm%sJ?cXM$?mNqe;eG9LUAkMH05io(T;c_uQ>aZI~YKlDwlq8D9L6js|y&Q90$}
zTnM+9X}O2zCCcdw&e0>oG^Ex0?G*9wl%dNs2I#g3tqI}-r83Suv;HG0b6h=zB&=mi
zoie4{xC6pWqz>Sr4DF^?N{DQ{tYCIes>5g#zFHt*p3KPQ7rQYLv-ZUGOy3)5B4f`k
zSxFAjl{HW3ikcx~nQ&ca>DFQRKlb{bzB}8r$Rq5-2)WyhXM8%eIvJ<#Py3@g`Rb8+
zZi?)cSyg?>`_F7f;)VpzhNMm&XPbnP;h$cHDIl-lC1fh(CdY2O$p0AOz=Xn#Zk@6Q
zS(N^bD?5y|7KgaLisDo8;xv{ucvy30ZKk)OSb(F6-M*9}-W-dyiM+ggE*Ix?9fn?@
zFNgjRNsd_dncxxJ$ZTsoErWZ-89z!sz#YyS9{BjE*F}DHgr2x`^5o*o?1lAWA2v<5
zgZcLB*=ouk4OKsTsWQ~HcT~CV+s*Tpr%%OSZ*q4Hu70~>ty~L}emkNAXkZp;x4^l~
zZB-17*p6+~n2eQ)-Xa}&9b#a9wL97&eNDi#38;^<c09^BZ^PBzFDclV;M96UfHxWF
zg)r;@>Nfu0of31?@C4kjsR%d&#_8J_p6^9?$y(O;V3?lIwzUCPMCBU%p%z(5#6Oci
zgMU~D4#wUTP|k!{9ssa-r<S*jqj>#<-?nkDiAp8Gcx4rqh?6@jSt|t~OYku8R3pZ+
zG)ElYzsM?hqK|*f^sRaZt#wz$TrdjHK}2vrZxh&U8q2-5A#o286F@#((o7{OUa_Xh
z%Td%Ox7a$ScYI)3|C;y&L&rEDm%opH(8s6bdk-XN5`vAw*cJ3)UF<A#Z{8stTYr#S
zbZgK(h3i#f0q@JQl-+tn?I<vL7vt!<(<=94+UtO&xH#J*Ud0uytt|3=DBQ!o*$Hyd
z014<IrG=z4f^26>W}W~h4Kv6teuh711%b=Tq+FoU%@f3?x}!(X+{%;kT<MgP&|$K(
zHW+#;hIo!O#qH1ST|1jmP+aI;2+9xZesOFG+xrzS62`!g>o8Ee^Z%|?z|+xYbuzEs
zL;7s>_SXDt?7M@U{t^8c&EvujhBfLBd@p@{v!RQ*s*xo#uQfi2X#FF=Z{j?Fi@InC
z2=6~5D>5;x`g(XZ&ITc+ty8?p-ZtX~H?T^`{Ib278svsDdas+hm@9anO{zGSrq<BS
z*2wH1t~g?02-n;eAZ+~PwxR7}ebFgK>Z^7_Rs~)WHRgdZALEXdm5-dw1m>n(DE1Et
zXX@sPkoMjqe{g0ZtTiiZvn#D>AEkk_BFngYAC>X93Ue831*3$)emRFa>(^x({&6rh
zc=&Ky5v=O(T6$N3`(H(>L?_F08_+Lzt_&s_bBx$jXU5V&FW<P_GZVJj){_8K66N&X
z4XbNvBN=3XT(oxc>h~Za%Y4ql*c{<zTN<g_p5?<ReNk|@)68GcO#R>zvc=<X=-1m0
zZ2e=#)<KzEcMl$;7&sM1jN}hkXV_l0?M3BL+3d@3>Yw^*#6EA=km%!K(NwjstSN8}
z^>rTId``xjlx7S!!eNuSYAwXIznFmqJ#AyOJos|9`#yP!X5$<`#@L>LCWy3N?yLQV
zX?E90UGJ?(la}}D%EpSHbOo-FjVPsecKSI>S^xZf2VNS%3h4rU^!`T#OX55*7Q|~4
zNYDyUy~-FuTx_hqZ7V|dgseNZLZ&|tlNBpQ(-&<IjfD@pONf{j_=UZ%$E;oPvHk(z
z>)h0zjG2Uwz=r3h5GJjkwD!2-4Aezn=RKh}!Qd*27jJKh*_GP{EHjRpIAUxn{ofyK
zfMlt{!s=#Y09}lqnywFiJliM;1G`%)WNrq(dJ0y_|1elWhy04AT?BzSW-jWhg9;}1
z4vj}WSGJ$^MgyVLyN4Afm+NqtvycoqLPj5GuCcm7Z!(L%pdg)fY9p9-6Lmq@*iHBl
z>txLR>5j@63lCR=F6yvv>#wH0*VKRMKdZrO+>6nvR%D$xlFnoiSz#}-3Rdn=|BmI3
z#8Lb5q9NYaps)K5F=^jqh!^R`aIWg~ba?9ZcN|Dcsk4gwTH)Wf&PD!d4#LPI#LrA9
zIl2t#_`-VUa%Uc+9F`q?Ox*P#eSJd2p-TCvP)We>OtG~C2L8KlFOUvfyO)6Os0&xQ
zz}#ESj3mt|W^~}7J0pc&BG3CSw-ELAME=6*GZbt#N;1kZeJlzoQK{Hjy~?P~$##1E
zR(LrqNl6a4<kciDYK_aC>amLH>+0MNWQn{Sg#MxhHpZ#-bg&&S0hd&&xP!Rz4XRdm
zAwDjST#h+^YQ1X|l|^shKo6bEozKNn%&uYdV}kf>_I?~R;p1&yK3!C&Kl#v=h0-JG
za3gZEamZYP-4LF|V=6l=uq2lAwcG=EDd@qWhr1e4QZcSZv*6%Mnas25*kxs%SEM%P
z{tJG~&c0S>7CdmJp24>5(D~t5?gN{66EiJyeF8vi_^-{ZfTDJVvd4uJ0_Fzcc^Ear
z<PsMb_tSY%{q4R>H9Y6ZHsi9T;ksV-hI{n)?ttRG2j)!1l<`O!8yKmPmNIi&V*$ml
z;rVlU;-)Hz>$Iqf>@`UuyZSJ_T1Bhx4NghXW;3;|W@yK~Gae!~pA!<SQeTi4-&tiJ
zuE&QlI>PSy#L}g8(JiTbr1-8vtNQpzOn&_O4jK)Ns}AHN^Y_T{ZG7<9O%dsC@*NbN
zT|L7k-Q;`1@s)~TpT1+mOKPZ&tP1_tWZCZ#@a{1GvK=%hGIdjCB|u_0wf`I{V`4`y
z1t>f7-sQ+#L^mZ!jcRNh6a7h7DXmjMqy_DDd!`JotN+*Y<-J<tf^1Kf_7#<(62da{
z6mv0}!s+T1S_b+KE`}56SYc{PRJz?#6W`#5sUYN7c2yefU}5vDtJ>TI_aQ`h#xyu|
z=_XYJ?kK@`)vWfDm}75|;$5r7Sncb)Xe8tlvA!h1vF0)>rJCTmVDs;xsAl<DjNDOV
zi6dTiysL3b6q-52x>n=vW6=nN;5cj!0dFhJN>!7sE~!Z$Q^IZqkpkHuu$#@d2g6>V
zTk>wsM62b{xAsAKDE&#r^>ft_{xeLbB3iAjRLsO!XmI(es19%9c^lN$eAuQ+JPev^
zk^zG}RJ%<*huL4ev|t%Px{n>R8s$e8NllbSj(s^5(zfjZgFC^)J;;W9o@e(Vn$Vb@
ztK%ib8I*Ndig9IbnIaz@11?d`#+%KX-UH5g6{dSqY&4XKQIKQ*o>vrt-Uz}+-?RzY
zSL5~4DQeM_JKT`I+uT&DAD&%HLIWl5<S8KMOfD`|qG&caRFl$#0@pU~I|KQ?BA?dO
z{LhFQ>1DB@JA<o#4HhE!hO@f2>4yUisi%5L;~of@M3toX>Eczp;c+q|Q(EHI2^K^b
zlz{f)^Z$aPZW2P0qWSeU$WDPnrCf#ND;zW_-p;mxPeQK~@=f8lHfB_X&*k5wK+!%n
zmF$&Qn7>2|vpt)vX8IqS&pQasSE+pmS6c(w>-gAU;maAgHd6STeek>Z>?@jDvbR;|
zk`cVBoL<Hx?y=^%?o1>aTt8tqht~0%tUxL)h1fY;9}0#fO#)7QyT3XeZ=NDVuQuZO
z)f{T$kW~m*%}Jw~KKJtk!=l*~dT<SaS8%@hc(V|g<5JNbPcQ#aY9B%}WWF@R*FK}#
zK`;2>vyGAaEvP181lFCld!1EM14&e`7-WjHkU{hA{z)|q{*C)a5x^ElXVY4M8rk|j
zIb1x=-NzcP)9K1}dO(jfhLf3b8Wn#K`8f<aszsMaP5EJo<+aCe{&sVbH^Zg=Z7CbW
zW0@+n%9SMX>RE}R^p-{BYQpeIoDQ_OGzX5Z8^h|fBRimFog>@bW68g(Y?&=hTLTwt
z4m1De*zfZCklvO(F3Wyoy83D<8Q^q?2zuyglo2>K9;rGg?I9_#y<E}=1{g}uW%8&%
zmhg-YnH{Nw3p5E%zpce$RoGutW$&3EF}G=0*WYZ$aQ+$^4O|2}@ns0oR7*BF%1FFN
zA!Ai`zpP%2a4zxuvtL;s!UNq!`8<8YphzPxOe`xJI>M;C=hZb^fJjH=X%m7xfXiGF
z^O@J2&wXrcx#$prs@-1nRv|8Dtsddr$%MxGc<jmqQfc7;Eqi+~^a-E)$USeQoUzh~
zd4?SWsOYog?x=jN9!E^Dq#e4KeRV}0E?n^BXVBnPzfQZg?lYAEw%;0XAJv=2_2O7n
zAeVnOUW2yz>VCV{UtTJV*MSN8VF88wnrFaN>CwN7+%qHVi5hzX3xhpOvl<r}PuttV
zB`s$ZF%L>mt<?kj{%y9oVDck*r|9_H)s~hM8<j|9X8QmWDElo4!exNcM*V|G;}Cv|
zED4rm#XhKlEQwOdOM}DeE|ZF!ui~7ywcqB6kSeTUiP=VP>6q-Xt3AY%>EX(!acCQO
z(yOt|5dzywzNm6lnfF?8o1}DR@-HHyt&xOdO2<{%yL<Qk37i@kn0BBr1mh43bG|f<
zWp2TG`DsL#<$1f>++Qf2P2nU!PmY)^&V7l$$Lv6ev5tV$M6g(Uu50+wal>ZDAALsp
z$3dccDM>+J8DZvxuxFNDP7zaHdEh?3K@al6-i0gaI@O@`V>rkzI7{m$%8wKVD3D!$
zQy*8uJivAWoC+>+EG!FtZ>*eZjB5`=*{q2of{_KYYyPUFs6M2o=Cv2<!gz+4tmm2T
zQl<XUef=<_PvTd8Ttcx@(tHqON5c&7{yZdTLNcDy^*%F~#~CT|<yCea`7oQ(*x_GB
zSO0185cYLyF|x-}Ks2CEe1fm?ITz$QrcFYeI&y~>Nut?H=#YhZe3fa7BIjA<S}J=w
z_3c_YzUenr1S#e-LkC;w4S<;9K5A@(7Ks8P;MpEjistw)5<2m=MlV=-29;%d#-m^G
z-(v)>0dzrRs*|>Li%HF)B32~;tn{O{5u99O3~^bSG4YP~kIHc-s&V(mBp)~T)oyEg
zZhckL`RaH^NF%1eqK-pk-DW%Le1A8ZH1X6UlW4a_)0OY0UVpE-2qB%qBSVXOd?1EB
zN7;5(qU=drUjhSKR=*%oJC2HtlYC|#bZ0)kN4AXzxo^Titw}dk*Ko(}LAZ{9&hfFy
zan_@{RPy1(eBto9%#Um6cl*L{!z6^-K}4QIn`D}K1!i@8{#^9lft>>c-Z*_~ZQtd@
zeczUaXxsa6a)sK#%W02iY2Dp~W8+xc&dvEP&31;tu+6|Yk7_X%rZi#VjKi?lSx5gM
zp0V|gN167%r4nfRzHYG=-1aBLHv!}gw;{cKjmiB9L0d<r;7WE2M?^0D=1*rTqDP+Q
z%2r1$>+TCMNt4ASfZ<LO*UL^bk_F~9QZ}Z<5{)QJALvTJTdf_fz+MA?9#S{$kKxSh
z_<OM@*yENum)@LP$bFOAw<V@VVKO-=)?9X3$zNtnzOLr!2QJPBs}<ZA9`KSw&RcAP
zcf{E>4cY|q`<LzKiUj1yqX2N~;!Hn(P}F0n$aDv7Th1DH&=DKmkuA9l6|s@yVgP;s
z>)U&2dF0uB>r>dm3vP5+IpWTb@M#?nH8Pya9wJG-Ez4*EgFWPEb=-r*ZmEGgOnmDv
zF@>{;cXWhzY-IsC6SEr6*@gj9t&B2*nD5VU*O1Z4W9o)*P=)J66LNyupuZx_yURno
ziyp=dNDS!=u+BVKn}iG0+ED}vdB&@1FK%P1pP=-(?r1ESjPErOYNCN7*+)kQMV9sj
ztqCve#gm*jIo6#I&!B&BpIROWtpC<VTmw!Xtk1Q1C~Yod21n+pYpECZOTUj0zn%}(
z0Ig4jfK(5_ejq^ZWAvAZQoS}$Q=E9sAgLoQKJ8K~xl-ctOi3abxB7ec*@MZfe6U1|
zq&&2vj+V91b?HTIefQsK&$8J`Y}48iGB9egvGc<ERr`OHjv#rornx(&ZAL*FZXRR{
z07CIMKL5*(%iNjy>gY4qW{E?%u3t-M>y0BDY*BrUoPua8{Ig9ybDt#g&att6Qc&HL
z5Hglytk3|k?i0K$r^kUB6O@nbg{;dI%I^D<a2}<6Am$IvbANJX<p&)=934|YU;UYw
z(V0P~$NCH!Ovp^>nbmGLa*kEl@kaYge>k7y6t6PPNy>o2WlU0g20$7+a;3_h>@na@
zy12Zb-7stNyJnu6B#gU{^;ax$P9z~LP%$xw4V|aOPYU_iVe*Ls;T8WXAWC9>Sa{OQ
zxa7?4x4_)oQ%)i1k<zW}gj8MT>S{k4LKPIqdRfgBAi(G&Sd=sUrZ_Ai3dPXj_dtY?
zp;$vl#o&2{QGNw;(F5m*9B2APN4}AKBC4qE#bcv23|yi_)kWHd$Cc|YD=X@X(vE@S
zpm=gjG2R?wqQi~?t#H|*X1!94>GP`m`%pYH#`mDX;4(ss58-~5VT}!SVj7SDBENn9
z3)t(B@lyo?KmF##L)n0^K-2tE)U;RNneKfLBj4%r6`K+}oV$9#1MWj_Xa2r)_Gv95
z&=V7V_`sQ#1C|Vn8|1Y660*pb({Z~@F-&5lKN}WT6RaJkhd1O2&RMt>MOZsru9mFv
zW^Q^nOj+~Y4hh@Kxb36kB<j3$2xr;8Q{4*RH&$eqSZIX2EbsPCT{P5nep>LtHA-vp
zZM2KZBcCXWwOi_89WGUHFyFBuqJRv1XpSaqHMw><iNG+@5>rLNA=EzXO%>XCSQ+Lv
zF=#zCis61h_!ZYnzj+j1Ik~sMTQX4&RN!2Zlj%S=dfM@t4g+hRXEKJ^n3DEE9^T|R
zRUyyQ%eN*?9glSKA#(t6*g%M^39#0{gL0<q0mrLKahie(W{U1K3A9<){K-=4)`Q~s
z8ysxBN#RA^3p+(;t92;Ay$ktAu?x>LY}<if$y`w+h++Nrg_<Mo<ghe$2P_qbv$6#m
zUvzGHs3t0Yl7`vz9(#k!Ka~IP$oHNV@5bENMDZ>=6eP^l4V1-|*R9Wx_yePATObZj
z5XK37O^F%?s4-U<B8q2Z;V-$TZ1QxyaCYM`eAv$#>_#Gem{%c&)EL*umJpHg;m@HP
zul6ZREQZh2Fh=lwOpeZ>mjH%oT!w@_|E#$<uo(FD!Rh@kfIH9n=6|vgF#jhW4g)<K
z>;KpY=-K|o*#7VGfBOjN+1cs-4-p5FR>a)O$=CspR>Vr*$ymtP(ALNpl7|P<(aFJB
z-x|_wqbb!`DQmN##(ytkPv-FE=BB2<+ka#`8`w1f%+=Mk6@F`*yG@ct{WGh3W#{x2
z8`>-<gZNgp@#IS+5s9*V0V;C?9f<hKD*HSgJ<Sck;h{pLMP^FsiGt4&)sj#Y015~R
z36qPxy^L^f4Bg0-(&PyG2+)p06X36O)~?_#cUoH8{y1O)pJv|SX90QN8W0|w6N|!Q
z9T)>pS=7%TVop+14@1JzSL3{D_AwN(ZyW*()7yP;hE}_u#K7d7AK4P#c>-tvbV4Lb
zPF7M%_`b1~zj%XqvjB2+j-IRm07@<`0ET(N@y+OcGthaUOTcB75tU_sh?M^>Co7es
z?;urntZi<x|HPumD>FNbbNHpz1!S-Rl<|YbDXA)dKB~f)eb($|!RF<?^!$|0bROo~
zx!W`AGgA}Wn~d<z0_cIZ{lU)W^PTufWr=V#_UkR~4$WZSSpQN6;4|L8vYH(nJ3BcU
z(YrY~n$kNpo6)oVPUT^SibLv~o|%Wd`#A@10{vM=L)Bb-b@_Kyev|iRO8}1eNcc2K
z_j2(dI(?}(x>_H&&$e&AZ1r!W|DE;XhwcpkDc&9Lgc=(AH!@{u5fNA&{hxIXAnRx=
z=^gCs=q&9lfDqsH9bf;Fq+djE;IIxhj-PV;-(Ja>Uz*3>1>vimT&dNK$<eXv-hbAr
zmErM6-0t_;QI`5f=RfT1oZpm$@Oq#Ppg;1rPo}ecExzkV!N^6#L(0o%$2(xNDLOkh
z-JC7ZGdDB7?!V)tR3xK7dj<ypbdHUHX*;1xTPsRSGb1P3%}8IlFX8?>pGQ`wXZ>%!
zCmRzhGt(=VzZ=thEhQ$t9JUt*vt>3=4t9VNk>9beoe)2^ne21u|5A+{0Dd<lC{ex>
z-^|p$Q&YZEx7K&}4Xuq}Y3S;lfZlv2eXnqCw4{Wpczx^hAZLfqv%$YaZYDB*<YPnc
z-gCQfP+z+1#n#hT09(Hdy_D~}#lPr)6}_+&&!6PX&5g}30BL`w_t25AvUb4&E`Dd(
zKWjA*X;xMsMOgSMer?WuKhf9ys<?j!AA7lYfPT%273v(D=o@}&nd1_H$FR*RfE`%s
z-&75MjZ<A%JaWDDH61PYK1;rg6@DGFs&-FbzHb1(+dBZBrmV1jn_qYmNK{(*NJ#-U
zdk6e(t>0+-Z@Yg9QTxa!O7qK!BtLykPH0)}HIX<K<Eq*Hg0q42j?T=E0B4_gz^SgT
z0BBEM(^)_|ewoPuL_jq;Pha8vx+5Fm`le<Oer9bP>;TY&f6{)0vu>AvNN4~G*LuSf
zA^OPfS?Ph$h<`~`01BUWhd}j`-ZR$#qZ)h{J-OR`kybrrihIwwBQJKJa)hsKJyc)b
zcZa@ZM0^)JWe$78G5gFOYz<62X!SqfyjEepz#0J1rhoizA*X)*ZxN|}qOES>4EFVH
z0h_)<M>}Dj;Jkb}CVlrhr7Hgonah7eF0(iOUHgoefw32=`4g-Ms`(eJhw@+Qp=Z+W
z<G;txe~<T0-rF8u?ROaxmzZBOkjGAr)!z7H?U+yAJMEZJ-M3uDil@_*pUXc6AD&jf
zr_ax*l^mQK>VUj~QoFj_zjpDg4lcm$mvr5`I8(PTuerb9vbVgDalhX8--H@MLo?fv
zIoH~R+?Sr1>Fa9gzfAtqr@RKzKhGrZmnR;)y1$dYHTD4TXRu8Ex*4sE``-rXTWDGb
zlsT;Gkis+5b{a9{dE`i7?MW=_F`?8c-O_*lPQjVI$d-<^CMF)tKT1Q_Q7;kw2Izmc
z>^xK(9E>Qbn~jdC?x4b{Yq?QL2J&;$Y5UCgHYEmqh>=jLR)$$~bTii^!{_n1P)%&_
zJfjUiJieAz5+QCctSn=ZwFao4kT{82D!mTJwqd(PCSvc+^y1DN5=46hr<$2SI|kB^
z8s%MkcFei-6G8#!qCT$I|9h&@6?QVX?TpEmA65#~1jssVW=<?E9aQ(l!B&T;Y>UP8
z!YS1wv2h6{7)2Tq2T)$d<{WR!7RCRlq{$eW<Bn-xtkNlJ3%g0JR(;CFeh?o#8zblY
zP+10^)5jgMu&UXaY~ngbvBm8lUYC>0Y=TNtJ+9E#qjr!wS)$61g_VGA0tCN6&p$#i
z>h_6Btl6>#S?IeYGkmn2$63#NdcmX?M7m4oa8rvX%aVQ&qw6*+<6o`3Mlzd3S$RE*
zbIT-Zqy>62Ec-%6g?2o<Z$0y5V$l}_3augFd{@4Cw|IScx?bW{&S2U;992Y1HCK7j
z*w1x0OPJdg4i#A1TTqJ$;tC^?9Azt)lg~h#F_wYXstYmUN|!vqkQR36$$P(A2YrMX
zDN2ZIP*i0E&*K>BKkF6)b#5ek$<ed$lsp>~<RuhL+44QH`p^pMIirW8I*`dcig{9?
zu7Qe%p_{;)-?ri5tlMSf7+vLMt}|31N_hGgODGj}HxI2*T*A)j{JVXjd{0O-OIYGF
zF~6vC>!atl;2)%`Y<SO#nd~1FKz8~sB7}eYuo|URYy07MIu8~>WP)K?>kz&-nUr&g
zB1dc-siUc{f<4G(Oq^7#4}+=PI2+Tw4l-SdOU$3Z>Vff_Uo151x%el->=2_oJ+EX?
zH_~A7T7Z9qlIMfDp<LtKuT#e_79NG#I8l_@KWVdvVW4GQYrN$dV#EY)ww+;)q}Alr
zbvcJD)UZ;^-D0^t+aFsRb}LL`ZYV{48i<T5hmAQXdt*==$LaFO<<8aR!~lW+K&}B}
z$R^|n)9{Mb*YZQ!Yd&FXur3pn@yHDQ`^A!ub_KGox{_Df<KLhf5X5VQ4P<PHq6WOg
z-jbHX;}e8(ZH&2Twu8~Vb)_D#Zzw)Z(z&6HM+M4<9XQXlPCy%nAo8<UGd0_d72%p@
z=>UQ19TksCgHp)JWh20tBJNpFV%8~pY{4AL9&1y43S<42(TUE;oq~;W9y7Ut)O|f7
zdjKmxM~ph#6aDa%{MAnP)urq4V}Yl)jl+Fx$HseW_jxr`Ms5~q0n974^A*BR*8&J*
z&Cj%;6Id8snjpFZ@8dFMWiMx9(ZLfssEEm-?!J<@bDn%l!8~up(|jNHAg7@p5%XQ6
za1)x`#fJY3PfljplD90sm>R6W_U=j)18+5mL%5hDLp2Te@*D?LG}y?B>)V4-k|A4I
zmw4PDdxgElY35HF1>)&k+Xb4XFRU~44m16+y4!?F$Jt#4Lc|u2p=GLi)?2@jiFo3s
zD3=Fm9_$ihn^I1FxKY2*B5f$pNR@Qxoajxz`I+<c98l74J}6`y8B%kMxu6G69n#Lo
zo`y47!ouWPVQ5C4PQ~ip-pS!GzfhlI!k2PGa@(G)LO2#Le5R#P;<R#u_%<T4xdXC1
z<$mYS@Q3fik3h6-q0V2fhi#xZ8b;4fmxGKl9%mAU>O5^)+&>7jnR`wG|7)a`4*|;F
zAI~GT3yozwqyJ6I?%cB+h;STVB@Rm1A#p`y!^UYwb9A@XGUJyWVWE&EL%@hRCY<Zb
zViM#i2BmxHsGo3$H~SB}cQi?=!ve_Q&}_**E;W&6|JQxetv>>d$VY1uC3cuDQO2gF
z8fhjh!U5|&G%csj{OsaiSfq%OTcipqSeNz=rC|!jjYdYJ@v?*<UFE?mBFf%WA}eq=
zPG!P}h=O=_<B^gRRw1URAF_Rocp#`!loj@S9O4x#@j*0HZe3(+PN5u9i&Z?!cY^xQ
zjCTls7R3Wn*D6!%!JAMkR;Pp07XuAxke$he^imS`4W{wWuCocQOmaho;{oC;Lt;<*
z*E>{+3w)l7b4ML%WjqIR5IB4tW}5y+&rM`=e5Q3247K>-KIpG<lZ(g{ytMKzcMp`6
zTgH0KeK%z8tOd@ORA$1s@*0$w6;!MdrF3^5rl1~_R(Z-OoUy$!6ATLJ8SjfdFs16!
zj@a0B%fHiv@MT1>?g%(V;W=oEK`En--r4M6afn#2jMa%@<yw<QB*FfxdK>(P{cMQj
znk0FLqHSxaW0~y`?Nu<!lQFyROVfS^8M@{&P?fJWEl53zHc>m?&39<Mvj~(riNPDz
zPL~2uK6d_vwcyKfFp<vO_s1p~LIufA>>5J?e@f>>eG8@1EG;yntrt%Tm8Z*9taG@J
zrb33CG(^;=^@!<=AK;WW`@|$Uw-bs(t?>vJ=uR7Cp$n+<>)IP_ukx?jXhuDbAcNW*
z7Lf$z>Uk6w;vkLw!8DSq@=4FhH9DawPo(PIph8k}ykfL9_9JJw5_LcAvIIg``C?{P
zaqMqW2eP`;2i-Xk&5>WWWtkiZrL3eeH4;ZbNNzBb42ae;fnFBQ!pc&@wnHTpDxJbg
zbWK1SKz1Gz%&jz~F>)`%buugO7~kJxp!xD<G^^#IZ~$L_Fon4Wy{(&mx<_*la4<X`
zy|uK>Er*^WOBuuxGaW~Apb{%X_C$f?Iv63M!>t}SWvLX34^m_;Miy(<)B*e5M-v}h
z4Of<BNbYj;IzkJC<tl^-!gSQrMgn}rU=xShHw;Hd1nBf#Nxsvl1h(}95gK`H=t*?S
zI`s{)vXdBDEj!^g4Imc89FJDmu8cs#gP;$CcYnnkO?ZDiPI}w2G<7O~(5uCizpA>+
zH3Tx%Zg+j8ZS*H>^M%B=kdfaKREm+&KOQMWXCbSF342rr7#m`k;BFbf{q0Jqb@V+2
z<W_?(@N^FuVsABdCL^W?^hc)oh=ZwAdH4)T5>ShI>OB`L?pz`i^_r{*2i8SOu!AR2
z{?%ApFZcp@nb=z+FtScqUfZ;UijAI#l0Wv$QdaU32VB^=@__ia^)-iHwXe^H>q7Q`
ztyNny<ZdcVWDXpzuCB|wz;5=EFBJL>mmIO$Sh(RNKSl6@<AyPFFcty-#dG-6OP}!1
zAes#^nFM^;UA?;b$^*-u0cd0shA+|80g`Dg9G+S%#)REOuP}N@<(fL5!KF5b#_3kL
zxJ;E23|cQT_C2i(Bc5?Y#HxSKQWjpBoi1EU({F}SLi_aEoEHGNXvg|T1!;I$Uy`9)
zKn_le$PF8k1Dcv9PPBwCDlyIRIazcuU}u47!NigL-dotL^z?o+3^FS|Jb}G}T+9;`
zWa@T_t^sAw9;;{3e8a~YvYDVu2797|f_TCj@{pJuY(`d@FW#z_i_NZc8B*;p_Se12
zqaL5;D9w*>tI`F_;c*X!@8H2vYa&QE6cPC+W4#Xpvi?=XNl55w-ab?S0}iEVi#JoI
zWl8^9#`H+5ZqOGhUTN};7=dU7w1^PR84n&+L&FMAe)^wlW4ViR5J=wq^-E_~5GSrl
zR#BH^jUnjph{`+SE}Lby+8o?#l}aGZGUApVLdg@3sGbOxDvIJsu!2UUHv4$cVceqz
z=R|8gO^-Vvxn@n2v%$}&l0Ctm1{~B%Os^<s1>S-(w~S`(U!Gmq22982CfUByrqyz{
zlAHHu0AKxZygozP!OQu8owFl}W>YM;rLk6{lqTKj9>(&mRdR0CLPX9lt?RW({iF<+
z7Xa@776jW)Jb4qgbKcnWVgN$dBt5Dtn%kiHhyh7)PWmrOwauWCp1Z{}0%R>DYy?Sn
zEtGZleQAXofNMz8SXT}Zk~<}in(A=r$)Q4`nYnDkyS4UeS<6@>CX3W&S`|cfqUl`n
z`WL^2yRQh3whl20Lmp~V9u|3@ekl8*bm*zYV2ZZp3l{wogI`p}mh#lfe#M`Z<grOK
z-AobGrP%^NfB+WeYBKHyKoePF(QY#+Ad$E1_HotuES+q1Nl6)uj)FX4@Rku>-mMa;
zp0|vts3gcm!^eNI=&ui*o6D9s9V0iI(hOF_r{qV`KN@H40FOQwE>SSfBh1g|TUJji
zH_QE0^5r<N%~JvO08Cl`Y6U1Q^WNYpTZ_?%nu&8dBx=$r*%Kc@MupG+zH4*_<EdV~
zP)~m8yfkJk?+7Pe#RiaVFvc84ERE?#$<@n5!kbn95!mxz#|R>c)KdFtd2;>OrArqi
zsH=C8IWlZ4nh!Yuib~T-gfJ+Bww?znz#wH}aVQMZIZ3?;>%E}D6iIZkE9;YCJjvg%
zZkpO!EFv5{+Bq*uxCuT=1>9ySyUU=Mr1>5j8%{_OUnqZ`T%d~L;f9aNaGB7*i^`+a
zO>mlmiI09ybg&SpvD_8_liP4yz^Zf*;5`Jh9_JOTP)qo?*?dB3!=>q2qWx}&aKgCY
zjm%KJf;M_50KRtx^z+e@8F=6Akd>Jk2gcyr5hH;|;v<!RV8kiWtDkVji3Ic1D4fr8
z7a3b7T?ESxg(~KLXXJ_02d<C4Dnk0R?k%pyuA@+pVh%d1Ow7%~4`$!9gaJ*Ug-=<|
zbwXt(jRa$Ca~?<t8&m<zJr5yU>eEnZ_k?(DxNd&P@4m2mQQK~H#tx0M55hv?Uvi=*
zo=L8J3C;LDfO6FlF*;6s>kI_LMuAJwaCQ9Q7{^Gjgfl+At|2!ZT#KMw<4@TNNZ5&X
z9Frc~v+r99RZXA0pfYyY{KU~)i2v8rk7KWd*@Ro91ez_=$?dyZ%c&%~+th}u_9oaB
z@DJ}~fx;2Nw?f_O1?5sHaHYYQ^;1v==mYD47om}6&>jClH#}@T>R)(6GktJlcjddy
zQlRw@f=GNCPC7Glu2%uI)0td&E!*FGwQ0|(C;ho!)9g{Bl{3XF<yO1vq`y_DCrB>4
zxntZIM(Q{wDG9?rhDzQfSOrTjK;*=A9NIHd!G&=tac_B#N(BS>N{~3BS*Oa5BL4F?
zH5CU7>(P!t^e0y#5?TRJd`|Rc^dfQX#``Q-aGZNYm?#{vfaE3@UkNwen1=X8`dk$w
zi&^JbgT(bpcMhk^_#;+3d+A)^+_OR-TH@NJ1j5<A%!clJNErbzlqY2(RSaCe7@(C-
z{)F*?Chh~p*X3IRy$w?U);QFvq~HgD+@qT0^pe!q5Gs4t$hfBc?0L^s6BC*YC5(mk
z@=kQt|3TVskh#Vhw@mQvBCDx7Wa+L-bF12`ywu`05Md?eZ!ifHKDP!xxe|R!!bT+>
zgZ|&K{U%6VQ`5lF<VqD)WYwi$m<x>&x#~F{v$8V~mV;4{n&e=VkUa3#);uJsBl`E_
zoxc8BJ+_F~Ls2Z<Er1=ecWxVQfkn;AWX0xW{B#}Dx)70z^?K{@26qdUXiLOU0L!rw
zKn(traUUs71P1s`3^ms$WEbYhjb_lAk6`6fXSxbiQ6S#;AJlThRQqU|H{dbpT4{_n
z#y^`=<N=W>xq|LXh+(biOGK@q@dC9}m6E+a`D)Yl8@#x={?pJqC<fMV7ztR@_;T|F
zKI)THOb1ueZvIMS*%mTY^BXw+h_iWOH85ib!<=@|OkD|4AeC_x25t|T6s@R}V4)<S
z{Gcb+6mm#7IM&89dNFN!701z`7*!nLXWKi!X^a}rR;LjNrzi;fI72;J1vcZk%(+1F
zsAR04gdQ55p+H4lG3J+KC{||BxB`|*$lcQg1>5&@a0Tapte*grMW7|qN+SMkho;AH
zPI6$4M{0P+>HUD9mXEt_vC%XS(uF>}ug5IMxBMk4fHIg@Hd6uJRL?YuLyqeO)2<jT
z_!5Xv?VTkDf-JUX)k-olrq)LCI_LTo%3LdB3`}6uQU=)CO|2ax9~i~E4il5W$&2b^
zaG1^#E`4(iV6|Vy90bsuG;AU5_VD$_Eu(Ja*C7q=&2M4iU~?CMOj2l{BRqTT=TLP1
zg^1*VTedCRO080W8Rv&439axC6NfiYyWi+Bn<7XfIMjESa^5@;!cFrrz@gFtPAq>N
zEO+ywfp2@NjE=o9ei{5)vSoiaRmRWH74G!0{e#Nvx5xg}On(;1F+Ei^OZ-!0Jp~0M
z+r!Bhtrq+ZVK(F4>su!^o`YcinSLHEW?KA=K9|FeU8lKiijk|eQ7W9*d<4cbJYHEV
zcEXjq^;1#K2Nfg6(V0<xyQQ9TIN6KEdi-3hzyxzGAXW9k@3y6hZ@^UsK-%F?@9Pe=
zET+nS_lY!0p`tvWh6<+mD%NUI_Pez_fs#ysIi%-|;DNp;`ty!2YQaE@TiXRi1|3h&
zmF42`Cqr%)BgUxA^Ug_MD~=<Q>*El;k34|HkvE8Qi0_<PB=%LL{3W4YXjFf~<Cd(6
zuk$C*Serl7M$yst4I2Wt9z^sEyd)yBZZ|UX<Obk0n@m1wJ4H6R9#BS5dmVtV4g1pL
zl0@Y=yEUT|UjqdkqHfLwlvd?f!TZ-#c9|}lu}xxiSVsoBh)zD2$UNvu5g#_Xt-y*k
z58Myd-F7p4_U!$OZK{PKUrY!EwWkKv8je{Tb5t&w&5yEv(k%2Rkf*!;@SlGc6x*t`
zOy=lzwX_Ix<wYAWd8|$zd78}LNXFxXd_M7E9RfFg*-dak)VAa!$Nr>En^@;N6=lb(
z*9D!B%S?VOjN0lL?a@jAB>VL4(PGVIOsa%$CTafB#tOa0h@}CuC!^h_cppBw1{9uG
zg06n00#b>Hs%PkGz>)-v4)6U1DgT>tl<_{=W~H<<y(aQy(Au!Vz2S2Qp1n^mStW^^
zpyz;7A7z9_85flqw`Di}3QjdvZq$)EV&s`Bh9@u933FAYqQA<ap6nYLx7W~~txMm^
z8H-<Db;Z@b^xnDPr-}*~{fH8M*dWnZxk?gM?_dxF*z-sMx30rFG;>4@afQPanE24z
zS4n%4yV$CjfHvr3M(#%&w?OBh4M|yQb1>l^e6=%^<HMdof)OPkz$n&>lY3E<6dN~@
z4YMDra3ZVQc(7AiUGEUfu7vl+QEwV~l6)&2$OGbCnSvFa8@U!a3}5ncLfN=m|6-UB
z3O4Is*`5ncSf&m;NU#8|w&h-KuNr)GG)M;b$Z1MaGe$4bPJ)%L7yPidXs5Pl;gawF
zB|Uu@K?f#*dF%lA{30l6yL2f*k*>Y!u$?Y=WVd|jVwpao6(D==w<`iTcr7v^!<lBx
zDDjlUJK8`y9WXTSGp7T*e<>{2zi(q*kq@=0b9OsdLodnMzHWDXG>YcJGSEekd<rI1
z7e!HpBtk}PdfBDH1+EGC@10d><6ty893NB^3eOsBckna(b1&K(SUB*KVW@P}sKe!$
zMr+thiPUa5S>BkRP%AJz@ccq-hiYoMV7YGO`$<oHF2c45jzcVr{;(EwSdozo3lewY
z^K~Se4AWtmZ8Wfbt00BbU5G^eN@$WqibInJY-!%lA@cPgL}Go26qo8YTLaqCCKFot
z@Ww?jf7KV5$c+EPESn2LhiS1Ss$`k>`c>C7bjt}b6%RX}@RA+u0j&R#j?8OOv2$g5
zYj;dR6K#F=NTXyu?^U8m5!4rv&C&qOlEP5;QF3jY&dRtSd_Q=bCGGYcCuC@2$<)$b
z7e{0I0srQw0dAiwssSz4rEH7}E3d(k#;VVg&G-WGV%)3QcXtZ9pQ)HnS8qe(413ey
z5iKd%(!%-VmY0&i1MlUTJ#I<O;?8_R$qD-kdg;-KCeyS8jXoSa^KyC_a(9{8Bp0Mm
z1h~UzYEie9J_=^%hT3nAqWmmP>AM@`Om@x9E7?VB=P9ZR`*JiqJ~ufXrFWdaJlr_&
z!`bV;5(lU``MP6d$B3DQIm(Uq$YP3xwRp|Q(^q~SU*&a1BXc;L@W@1l8zJ(rz^RHZ
z*U4;}Fzn1qr}S5F*+?9gKgjAl%6Iaz=#}qGyTBUob#RJEl4jF62&FmrIQTHjh(DOL
zr!Dpi<ztlp&gV#_@Esa6J5(e6l8;<|rtw=iTyAr6H53>8K>7-=bmz2T6_HYZ(3nNz
zNhX6XqI|L?zCTJu<-KQ=9Dxk`Axyz7#c9(<iR;sGK`e}xLC|WKjF$|&yADc85$<D!
zs7^#Pk9-)g;&3#dh8N6nV=Um1JLn%T>RZC_`)G~wfuV~LHBGgx^xVOJ^Q;37h0IN2
zIaKJ6rj>mJFv`k9xmkZ$sTn}2H&o_I#(BK2gMRl_QL%T{Y!hb&vAX`RO+iwvOE4F9
zdeu#+4M=$D&pmWhI7J94ZR=6c9$M_6|8&tpdMk&<Ge4q20j*5jE|YQg#Czx|Qum?r
zF%2!)y3t4dXE6>}Xgy57c|DiND;zZd-TXDvp#oB&g!K)MI&?wX@IA#t?7?DpZm-Gt
zcNmjWP`i{iHO~85<<QLVAD$k{J*3gg?eI>_xO7&bz?1i2$078swG)>85$ex{Lfj8d
z;I3GKfp#wIvp9c|03mj3F|obBhE@loxbV~LE<Dl<u94F3dzgPEqvc*F5jTAp$eI(E
zm`xJh3;_3AwIG&kftd}Y?xo*K<B}X@=bF_ibW(~|Hc#w8hal1c!>Hw{r03Xej;BHb
z7TR6XL5Iew8!$RW%aCgCwL*B@+cM6Dm4OE4zno5Gy3F;hxVRJ(V?<|NA|zEZffME5
z9(oy52dzkR6o+NJVpl^s`q2Nlr#UQ|;H3Mq_fB~KREy+AngRCk2qj(?HzPRRkPpkj
zMM?skt`ebzR}y+e#N)JE;&P+SzbcR^f<~@fa07|v-~+{}+AxHz!J_gI9r{s1z&^j2
z>-4lC>}sI4aflR$T_6TpqSTmR62d@_mP_#=FA@#`vN{GuP2*EGJi#IRK&6S#X$&8y
z#A{dU0nWX%zxGRtQE_Zgf%iZIm}Dp>zvhWKgl~Zs7+OXvFyh;b>N<Ic9`P8dEoP=<
zmundbE}J<P9;&dXqe4dVZ-D44UF0m6T9|8J#IR;<f2TlLRphJv7h&hLD+(7b*=5_d
zZDTLnwr$(CZQHhO+qSL#)5*E$<RpDL-(Y@~sxe5|`$Yz_VCKD19J3<3iu4fFr(79u
zY`1&)osGlA-%r+L^Fh_dSD0>q<ajR4rcD9ZhzY2E#xdYe%@xy7Uz46PPO1N6@*F}4
z7Mai{zI!8v8JRPJnOhC(MxZYLtSIN^Ur^FF9SEGQ8UJfbemUd))L1tAuR$5tv#hZi
z8Xs3Vq%te!LT~(uRsf15;fh&Xq0f<Rb`EW6&XY`Hi+C6Vl&#OH433LUuS}FGHB^g4
zGO=+c*e7Js;KU$gn0#AjKcQ)VsSx?2)bD_0z1p>gF^q|KUYy}EfKS-QpYqUz)tAD;
zY!~#YM|7@z@nde;&{U>0Y9wch+~#8Np1=@ydPjVBWGX}xRFA5FB-+>q46QW$0znOP
zd|FI)hov53_@1K<C#&@g!LT7;3xchoV^=yl($+4TRBK?n<TdU-^0dnl6dyy{Yizro
zSGF2$u7oX6XC%Ep!L7VM)&Nf^i!`M(P64BwPb9&d7`KH~Z9z}UF$G8Mo!+-eUH($f
z!=(HpalsD@o5(r@h|GuMr(zcfztL?t&2z#(<D_=(gG^w5zF33Ujv78MX)Y@#*rte*
zNvb?nm(#IwgN2u+?p|$KoO08oP-(=_#|Vkqs$^e_k0>J`_N~m@w2sf$W#kQ4@x0;5
z?zqGU4DK_ls8Q_7e<J-sTMRchS%r85<E@P6{<q_`Ru%|unYRdq<+x`SyHoJ*{bifD
znFRIXjE7|x8S3U&UrvU@NnWF*(ld705d7vecF1DH@kBYGHCZ{Q^#cn^t}kHX_<QU#
z8MD)lkUwSccJ3IsfsS64!!0+P-g_B3#gT4+MPHwViy8?-=?NE^Hle?Pg0pg4v}}3J
z_4?#s6>FuuchZC~5-=-%0dN#JfCi}rEe+c=CYknODDXY8DFq*85>V^&kU=Azusd_x
zC!kv1mkR4e8!=1gYBC`ZuSWNFPVX^Uy&P00+L}+Kk_{YEBKCCwqXl;&1sa6>HSg3l
zln(6s9RWY@bAtX}hH-KOkWbFApG=mhN&E?q;Q4z0HOG64PsArOS@slNE<-|R|EgKE
zy|T_|@svK`$1KQ{&J{{U%j`db^_{E@<d%W~Tq8n}8H{33P?2pmga<L6(tB)<B128V
z2$vcjzm)>J9mWmPTF$kf_Dxv1i1t5uvK?$zBfS`ggae^j;V#^VT;<uU<wf+nO?j`A
z$O$=>Jo>$iG<_o<Algh5roUhXV9ylcl1J|tXQqLvYTFfHfY*I}{y9i{nn9#=D$p2_
zQbj)<37yM)o2G#}Yg@xR5q8CENE%VR!{?_Pf;}z6U~1algoesFrD3zb;`vP-9}`92
z*7uzme9T2|d|Qpe=I$T>A@{8}63)m2B2{Oer$Kt+g(g*x-1TOaz`iX34x46af>zl}
zr1P;uU?_O<f<0)m8XFA&DLi$krXnAaPw{7iC>yDCO8i)r2MA9LkzEEAHj#Cmu8W8<
z7Y-eFlRyyXZ4LA7d0$h#YhKPP!~~u=F|9GXIo*oX(o$gM>+4gCBYQWW4SgdHv?F+|
z5H6;KQzq!6Mo-HXAA#+&cUvTzkp#?%3UvZ|-GG_H#*Nw8k0#JFGHu>w2hl#ip)C?<
zslQ;3XLBYS95fXGk(66U^uH}d@AI`XRvWz<#smbgslxa~x@uP?fc+@KGUFYv(1if8
zLOk5hf#FDsJA<6W5it$%ZBQ|&Y0ox%yGX@UBp<m7thdW)^g+><9${;vF!8TJBG@&~
zqCu<nMV<neFkDto8eo@6{7kDUpSMMe+`fBj^tJh?8)sXYc9NI4A*E^I180R@-#WHN
zSE@FFPm*}VGkTk-7B_`K@q%LNN(;RCbt<7@?iHGtc@;Yi*G-sKPgo2Z@`M`eI}#Pe
zmV?}R2ya<49f=L#<A(6~23h*NYn5$ts=D7oJoyHxlYA$ca<b=~e?`o<?P}grI(;N+
zvGIYzNLXIV6uXOEed>=+6yvFi^-^oB7s-7;G@Tt-=>9vNkBfKqTI3c=j#f_{b=PGp
zJgCPMW7P_x=KsEYFTj-Lo7t$s($Clp2uB`jF(rtrz<xMrH?<r_gu);d!lhhLVn+Y5
zFq(3n<#MGd+h&o8Hww_p7ku4szoVEj>h<v>-DyYXu8YLDn5)R<>SOsCj|wE+f!{cW
zdzP<n59%Eg&FHP2`X?uE(58s%BLgG<)}Q4J`!H4{Q{^(9@EGk4T1@lX8_c#5uCB14
zJj>Lwam1Lrc!<?FR-vemr&zvtHSMn1c(C=EfFkqCjg=J)8|(mtVjwE|;)ivReC5cf
zD#g2Zuo+vYDBqSG<viQSfK7{})GO<ar63K+zkYgcpRg7eTn7Zah*WD<o=s*|)t;hf
zdaNw^48%~={h>s1Er#MuV&cD_dx*$<T}d)7CqK#;b^#Y*P>HLK^M2`<JkkY>iRACc
zLfBM>uEZ^G?oLT_F}I9pX}EhTm91+^sif7gJ}eY?%Z$;Jj|Xv*Qe@K<<zQ<7l5OOS
z6*_ffK$JZOv$;gMgWMsEPZ8f?5)BL`FsOW}WM3zV(FL@j%N`|8dS62wXZ|+rowIt_
zP;*%@rqvVUj`HwW*!x^iAbr5DrBR~GMYMK{5tAD?b)Wr;2`Q}k-?vz*2qj*am*5Yv
z6mG*mfLr8|y7Pj4h~Ef9`9&Z3G3>#*C_3pb6*Q<y@~K?Hjrn76z$}5&AZ4keyLhm$
zxQ$)GyD<j(7W{jFJyq^0Cj8j(Ag%8ow@trYKxLuK;eK(`L6Df9WPM<*43tAV9id<?
zA^YlSc#t1F(2w7DsZPT4re6|kCWqg_V3q#lnBbzdga>P2m!?26k%Xw*96wjDE$-RE
z$PotD+R-{Jy$pRz0E5wCwVvkQw*rqgL8dPo>2>lGD~*4>`)>~Os0q|@;1x`FYy3?Y
z3=kM7;UH^o`|EgW5h%0xq!hq}I9Z;U({4ng%a)b$QAvKB_RxgTTl_3ix=z9UYAeMc
z??qD7a#~pmabr`dYyAkR(4C$H9Gy!*0|FN3MvHFSL*DaQfD&wgV$kKdOj#^cr~Y#(
z|L0`-?0%mCax}{P7Y|XvsP{=)+v*v%KGudU)RmvV=gM<j9LD~#7NMT>USuVqv|vol
zINty@>WoP8A~$IxRk2GmvcxaZDK}phA~(bRksK%JhZ~+-+j&nBv4_I?#+qkWH6vj(
zAKVyemE5SRkFfhk(-(j?42_q0_4f;U=)?_yjf?^R2%Ye~v6_!DVfvjlzh`MI5&BaT
zr0QdR8{kHjectZ}+EoOEC>h<dqXP9}U>$JA1>&6dMLE)0V(Scv08D>NJA}Ku|ILxt
zIgYj$fD+v<ooco^&?vc)l@6dIhA=DXOWK(pbo%9>-#q9+7`CWu(KAmxW%8Be+fOb7
zx{^KBOR<SLq83;z|2=Hmqu)}&dGuKWR+-JxspcK?uQyrF1=h?TYIJ7~oGgv|pfyU>
zy)xO_A$%fFtB)yk)F{kOuTT?3_6Oq@;}79qOPs$ysE@1njKo)e0LKm6a)sWvG2A%$
zr+LJVp^L{5M6Lg#&0B<Kt>-Kac-whGgxPIE$$TCq^ND%Jg5f~8{XMh}wTs-oOW)p%
zFhVrjXoC-ztB#;$%6R8;;}wyiASb>~*E-8mKwE~|1M^t!1wF5}AttT9hFU4YV|xjE
zM$9oRuv57KjB`8wI%*nQ>%HPQ)~z}m(-m|9qOM+QgYMVE3i;%0zuj?+^Tik#_+t&q
zE~Y_NcAF*K3u?k7ftO|)`4By@HHo^R`nHr;o>NMBinAd|#kj4Y&Ixb(d3CVYLV%1{
zv_@tp=j2)nEyQ|SbW^E4!b#1~+u=<`PtTNlDmsu@11B)kV_uT=VUdg`t*dMAx1!BT
zl~>`DU2OmQW$miR3w|G>?NR~Y)>T&w0!Q!CI^*YPc~kqtJVH3P=|H*a)nNH5?cPHR
z1xlt_ixU92_!G?Neus)IlYY1kxftvGQw*?8hOi>FvBHs6SqH^XeX7fsHU&3datBtQ
zT(S5>fPf6?WWh{{zgxUHPD7rFb{*rcKMF`0zwuUx*J=_p%CV&ODQraObGBZrVvxb1
z`H<O8)naf(wWc4Spv7y9|AonEQX6S8(*8Ia-58QfPC{qFBwbph*6P2vZOKr7)?7P@
z9ZcA~B>kys0sp8ssb_DiAWC>_7|aROIwGa>0X~~F!+p1pAw3hi?zT|38!7wy#zIpl
z^mfS*YdZa$Hhb<AV=phWQ+~Xq!MPA+CYZ67GJDyFFGU6a<EpXXi+sqE+zzgV?Yr)j
zK*&gIz+$Q2D=?S`WTL#dh|T*3p(sOo6PVho+RMf=l-z~Qt5$Esw{EW#aYxX0BJYb*
zOn$?mDA;^fFZ(=t8-}+f47ULdP<0(!^r$zm{5P|7f-R(#I^cIAytj_nrRsr5M6l6Y
zmnXS&-Wa1;W6FY7TU~7pYMk}%AG6LbS{iP&Q!hJ@h#u{`pi^y%#?5j3vfaMA+jkEp
z^_udBCjb`y%1u22pGTt%@J_wPD0S!Jp3Z4ISiof<Ls1g+B$WYpq-eOX0ItTiK(dbn
zUnqK&HQ`oK5$nyUYeg^z=bCptaUR$ujbtPhDv0hjusmlB7U#U?^X12iekWl;(9V9?
z$Oq|7fn5r>d$qFjf--ib1g@w)t!t;j=!AN?=+$Z$yrQ5~WU@1Zj4r$ze-9W?j)pdy
zj0)S#hX{7C3ZY~J$?jbyJz)~xyk>=xu=rc%ZESGI=p>-+c-4a~SJ1n*rsKgP#mzT*
zTX~7-O}uhO71BIwnM0F@(bohFzNEE3>pw-STbfS?rhIMuJmSX`GaW<lQ8)q~v}1|n
z1njGV+Al<)cJVh{H`{^c;du)K3^`8X-GGk9uqYX<{5MAH$^H)RN7Xh|`0rIZePWXj
zbQQO)yQK8?3MhtbQQSyyal6b><8oK&*dJHrlg*h>M+&Uk3dl00v{*X7zZuQJ!LGB*
z6+^}YxlM}9OH75fg*`}qYN$L9OLcF-<33r!SA0V*+;Kd`%^);!!Gr8tU!zpUF=u^o
zQ%|^{&I`OrJhlR}l`(0tUixBws%BG>CNo$i&-X}yXTozdZme44s*bE@-z^u8FYM}4
zIR(f%YM{Onmr=EpS~@_y*vKbri)&?8kkYDf`uR!kT@YA$+KI3_>k6Blk|4xl;COq3
z083@sk=-qa$I%a##*j*<o>dYitZ}i-Sjz!=14=FpN`$CfJM}hWoxamO4tH?4Z8Sb<
z$jNPK{=CcgiH}{@&SGBXW*4)0jH3Xrmb_`c>0&%fBs_i>%;8*S7A)uL3lqwDIV{-E
zc{`TyA~{|MEDv-RqG~cIv_vU{<Lai$vOZ%G1i&4z?16oh3i&g-Fd9sSonh2RPq7eO
z&(|Tnf4ldearn)yZpxM}U?CsVhxMd4Y~eC12-fOkIytQoVYptS8JL2(YM(D7O+xwf
zt>euUApxqQzXK<XZ|I++y+mp<``CKtXag}&-CvIIvxH7pX8Nds+{D9HjD_VW7YR)S
z1aX?pNW6%yz=B8PQg{tjXM>lwDV$HsF>O)zx2}Ozd0Rp-L?SwvK)KYx5;GiyT?Nq}
z<!ej~Mr~yGla}F7jlH{$r0e7zvN0#{9co;1VVSzihILI7tX3d+QP72FAElg+-84o}
zck1I;mJVa}qjShrq$~WTuKAk`L<FkbDPWaf%%nCMnvjp9D#jJwDbaUTFb?SJhk5fR
z1WEX+YwgU~T+k56P*%~46!t>F8;`_v!9Jp@c=$c@wnQ@(^jrqnye+iwBv6jUSlmj!
zlkbW6Cu#N<uxY<g&a{14eWAZD&iaAcY)l=(ob{5_M_^BtZyO6ic>oJ{R;AwuZmB|=
zgEd&j`5*!%H)o-U;>Yw-*_vI}%Wh1W$3r<2M`mZF{_+eCX7|u!4r?E-c)m7_SV(8G
za0sIAf<M&RuKGwt@=9$#(X#~SrMUjc9Zyo=N<9@YDTVV}cI63%N?DKF?`>!o;DU7v
zRJ}>51=$|_)@SA(+l6(vaP?#K2u`{Q#;l_T+V2Tbq<L#7K{9v9^+`4oxIRv!I*!6p
zi&+fnw=Js}@pD$bG?37y9Fj+%nP0Xzo46V8-$)nGo;7~jfz0c&>+Odh3P5$tjx{8<
zz@p**&AT7Jwww*iTukDTZ$B=JkK)0O@0M<}@{zloq_6z%NN>fW#loyG=K6($+LhLq
z*Gl!okL0B4b<=}q{{xFGKs7I@?j~6ax|`pkMk&{4pbrdiMc>n^tVU5`rs}taBprao
z(r2HGGycFo+V`L&q5THpfP<ba?NtaF1^UF&f2L<}d$QGbQmP8`ES1+Gl{p40G@T@k
z+M-+{xW;ax!N+KxrSLS}Jr3|B`Y7<*2DPQ22w^Trh8r<$Hw-mUV~P&iux9M6AX{R_
zH*AS)Y{JP(1Y(T%%AU$}N<NyIowL6Vz%L03FxzON-R2F!{M^6M!HqEo;ipZz(I0Vy
zOCIfp2OPJ-gi17pYJXei1fA6f#zLtA*I<wkI3ZzgkN00?+#+gXsNOH|u-nmw5n@(f
zeE;LO@xsocH(1oxHzKXm0Xj13DQ|;>W^IFeMi^aJrUoJ!GGQQEVYk4K124JDl;Wq4
zc*_X4oV0htEw4>D>Y9f+gLQ+2iPk9QMbBF0QUoi36nRgexI+3AJ*C^pS!uJTb;sNf
zzD?mvWvyH_<c}L)q<u^W?Tb#6x+)#qfwY>9w9QCRv3w1|Ud`;d@8_>S$BvyXnR;v-
zk#II_wfEOaq;9>n0S=V77tNI<uhtjhOv-Ku=PY|mV#|noCk6>4_H1hu0fy+GB56VQ
z+Y-S*6%(>0)WOL>YTodB<tf%Ys*?7Y9CpE4*2>3BOje$O-uSUefpS8H>cW~^Cqu5g
z>Dqe2E-WjDh*o1#_8IUzwFj)@Z0Q0GUeIE!**j@n414m=MPG=qQBHoN`to>IOuZr9
zYb5R{V9E-0gVvmF2}W?W@`oqmJ>;-^LIhi7N^{$;#Zl4Pn6e?V^9d(}2zO-H9UL9~
zo;DRaDcgf%+EEWY;a9MOaxn`*LZKT8!WA}7iMy^GoJC1yY4sw)oBT{L#w$aki*$@~
z2BCG3KMk(_=(uc8e0!b79HonwU6&?e`0-XQ9SZWA2y5pd1M+KVWzInIe?i*+gh7Wg
z$VNA1w&L%i+9^5vVg)3+M>5=#Jzah}d`bfnkZ8HV2w6xWBcL}w-_HW9-6Xoy5#6<T
z{62b#*QE|S@nk?G{QO?P16^&*KhX8pqXj9+VLA_ZUm{TmiX!Jp9vr(#4_`zoD`t0s
z^FN&Dvhc1x?aa=l<f3DAwMV7PWwO?YI?l2YAABV#dn3Psxkl&?6S;|*<wzD0Wy}9u
zJJKSx%7zN?&sS=FN!`<$-;u5=Y%vDI`BrW)204tnO&ZR+X=@#JmXjc6la%+(rpcB*
z#xR>H2+#%oJcuEQd^%B0b{WhucT+a9X)BRgl*a<rNRygp)2;z*c(py|?H))b&w1^V
znb=mFtJp*$ELzpp<Q+(DWPT@MlSn<K)PeYQrRIN3DJp#pMjsFHQc2~vxhS~svLY2{
z5<UvpvQk)+*V0-7-Lw~75iYJ`OIAVUNo$9Nx-Sto`_kjM$s%sbwi7><1jeMW-pwwq
zEGEk6tA&5%&yJevhpubGbM`RQ{p8@XrYJ((uKe1S@7Ud1y-1tvd#C2DfVVS9AwYK7
zy=fP*85k?5*Px!h@3LP@6a{&E<i}FSmkQU0G@;t4?Jha-mdM(;ZMMB3nE~C#O#0Xs
zQheF=d#g`~<1*e(#uzR)@m+B+0uu|R{g>Gitt`PCUX<0rRSOEPcChdBeUUW$o6&Al
z$5XnN1Rb-Az9q$YoQsOgbFq-0J3zuT<19eqKMQ);7QS#4wn*%BaDj!xF~0&%kJUuA
z7r&BxQ>Ffs^lyJYrAP-LtPGRrIRMnZB^Noo(;T2$f73*h7GuR114|!7i4ufKoxS!n
z0U28jeU!nX+d(>m%_Wha&wG;n3v%N>ce9K^*n%SSqHaMoht?0Fo0xy3%qX+dyJdF&
z<7t3!Y{h53AQ88-me9D8jsb5Y7rwm^JFATrHl6=eccP}muf}OQ<7DO3jY_&{WOy@s
zE*gzw9ZfIgReZ-nVO_^Ij<O&%%h8gOsDWmz0ja)o>uCX*%V@qL(rX7kGv{I$Fcr{e
zXEQ7+zZfjx7`(1Cu{NWpb299sbzb{AL4F?0*8tM>!@MI_BM8s^nh{141h*rA#{u*&
z0jYr!ZE!V<QA}oJyMIY`T3$w5GnT2h(M?e^#X~4-FMLnqRbkM?g42I#4c8N7ZGJ#a
zE60w9=G8^Wfi16N{r9|s1&@}yX-Ha$WL+p3>=kVk3YbQO>iD2z#?C&<H!DinKQ7Rr
z5WM`ge>vC~lkCXd%)FF-75FnKc>0d%5>iTBa7`n^b`wia0pc^p_H?y+hekrrX1G8F
zOvA+jfBUGin}J@DM{yF40lw06AzQ2kIihma<PZ-yT8Qmd{Fg_`WkmdvyCo)No(3+@
zl#R4w4WE?(m*-4niU=oI^cE@n3}K_`(gnP}r9r}87?C&g+%#{fxIe7Om;q~e)~-sg
z0x2AU_m|tX<6CdYzm1M8fEro&eWY^RNbuAOT%p3AndN3g5^SS1y{DIC(V3@Nx2aEM
z;gGgOEX76$bHxEW<|WtA2?lvw6qzPIXivv%40aDekLQ&J$X9rh28WEcM*6Lyv?GyK
zN<2I{;Wr-21@s6V6{b7MIUlt6Wi=R^TVR|L>MY!=+Zbcim+LA9g-8bT3jcrsO+GSi
z>IJFX+Ek{B2p+>CkfA4vL+EuI%<oJQe*4GStDCD-HRgzj^*YVp?nd328DbK1<*S(%
z>fUarw7nuHo>SFnY0k=yuD-WiEZ5N7SzWp^h`?jeq#i0t=-U_jp>>?$3yEx(e7|bQ
zC7feN26Qrh0qyVSq#}8bAwJ7N=S<3?ZP-<j!3O!6EgPnnCWcDZ@Oy=5(qP1P3$s5#
zR-#>H&UqkSc%JR|E?XPO#MKQ!Em<Y=L~^FIuATiM%`%|JHFRlfB-wtI5A*wnJzERh
z2eoM|Dl{(-Lu*YWn$gsQE}efLJo<<V>eXp&elQHu1yHT}=$wR2h#}#$IO9&1h?zlp
zgw3uZx7UN9JNAHPS;$lJPkiBMe{Tri#-j>aDbsL@Bp27j_EU7pfY7zY@bKDpL!#pd
zMMGmtP<1S45~OrWdlG@zMF0s7$Jb0)n<|WUw3PH#vP<IkO14s!oViNgyoRjO7dyPL
zo6?cdc5W_})RO$`CF$`4i00s4r)HR<{JVP4-MzG=kn?>-pjB!V6}s`THZZz-$QIgz
zgtaa^e%TTK8%PUU0j=e=Ag%}~E68JgLz}8OKG-x^g&<7>GyP{Dfd&RLp5WlX$yNSL
zQNt@efyOTs3`8j7u6;{7d-GJ~h5j$9vaUCactKPX7P1fx5=m9Fbl*yVus%`<R!v?@
zI@<&g2mmO>wJl<=Apky9Synxa&-Z;v#q+j29y{%s+%_$U@x14dJkD`Nd~agEoybH}
zvf(3=;jA;!{)WyLqn1o#0D+c}1<6S;3L!ePr0t2{-D*oYCE6zm&N%vklKMx6vpzul
zjR>#UKF|qghc|lHy#Dh0H3{qe^-AXK+YZ%PFccYkn8zqW&%gFyn#D>(DL4={eok$j
zeR|^OgOyNmIwf%KNg@=^vYW1YRu=zN$&|vg#|)1&u7dc$+EAIq&0eyRQTHY#^Fi`8
z9#4j!_?*3<%Hx|+Xn_(Cv3d3YTBlX3Z+n~{GLlN;wd%AOU0v>g9~qtacpp~(Orkr%
z4(%`H{s2GzRS*>~-K7XuirM^AD*I(Cn%QyDzhu+do7bke1DacaF<-qFmn{~WTE+;K
z#@(T*uD+Ngr^Sqqu-RK+{OSRX#?u0si*m<qp%Q;9<BW|TN-pY}fHaZ3j+KL0D4Ith
z=W5pD$F3N+f`74V+>gA|3=ve<^{^ZNv5X-nUW!E4@pn^{hdagm;$2H~L_^&1l|GDY
zJWiz`_l0cSX8-Ey`Edq6BE4NCLAhBQ(@u()<>l<ae$8-XNCzn56aTm;Bi(;LCJSU5
z;Zv7tBn*VQwgJ4!ma3GdB-0lYKtULqiW)}qA>BtnFG20IMkza5$|MjP8uvp8gkZxL
zE8wr1^iN(rQJhcEL&BA4@MtzJjmRSZ{aV@HJR&77$ovVFihxk#+u?T2!dx-TKiJi_
zRBA$y+VFfZUJo|A4Rx%$8SRw1eDr1iZBv6xP)Ih<ep$GQu@><OumSEFag76f+T5Bi
zU{o#A5s=UJ#lF=a@+--CM9fiV$$$liW+`D6LP8b@AV=<ukHxaEHAGws0i$%DJwP*h
zVaVcprnBI%8J$(S*N&FQv4N+5=-`?*nZ7wMa_}}j9oYtl*@&{Tq%%6VT|qjDdHXRe
zR7|GTV|yn7WXUWnTG-avM2a5RPMwjK>J$R5gbNoKH*CFPEv}1u;|e{oqU6C@fS$p~
zv(3JHcgk1&_X>VqoX^8y@Sgl$1z7|0*kraz6X~bPUc=+kZ0#hGaEiKtO*dZEd#X$<
z1`^-pCa5hjm_J{rFM$O<q@TT}24U?=Fb?g3mE&yn=;5U90wLRU05!82ECtc~HB6bz
z!kz?&h-@&KF{uI{%lm02gpTS8y+1PbOvwlaYIv=@FQFw(U)qIoTVO5%qX+DFG<9zm
z%3(2DC_TE?eCj%Rq=H&!3)#5@hl5bdee|Y)Dfx~RT@hutgRLMvH#-Fzmk;{aIyBR6
z6*dp7;!OQ*E7TolOlo%OL~FTJ3fBQyj!1W2I;IcG7x2iE7~9F%L#?v9p5#fSLikwp
z+dMV)4n5G-CM{3&a5dcH7v0ha2+KgW9(cTuSXn(;%E`C(&l012P_@A51ErR753#V?
zInneB_TCfvt_kzJ*Pwidw1UnMi|!9byIlTXFm1WgU1w`!Q)B|{tw}8)c$Ot?^~_)a
zH=WGNJLkbxA=0><p>ihOxiG9G&R_guS>0l&deP#|`O_%atWVE*H{GAZSg~lHwCWxB
zQ+wrxm2JI)(~RzmcPX|6N!wDJ<qgzg?=t$Yk_*?3X>b<2X02}X%J1mz8owS}QV-4M
z?A6v+PBzDXf4;K%C>K;bIj}ZFzxQh<{@AB|%23U214uRCQb;byt6@0Vu?kJYO`He7
zh$0R1+tA%JyDulCK^d>iG%?^RN6G@!`)wwj`^5aZMXrTC*Z(-TfWTeqAu8;|c}`!(
z6!188PC-YK>1k8yZlqNo_iCjxR%A<-px$@c!wd4hrP32|T6Bc|g)&+4>D>y=$p|p&
zkdUC43(W~IHhi;GQ^EbCv3cV~c^%8%%9cY{;}Iy0y`;p-PteCJ^fL13&+oqOzA-v0
zP{DtGfYLBcPyF0z7g!NirNW#`rNf)urV$nBDTSNuU*E@^N&k5Z8T9I{BN$j4`@0?V
zC`0wL$U66`D08darr}z+w?DpJJuw=j)3;rjmyjw_J@Cc*o|JUHU_omOcPE;x_0yDL
zv;)fLlxmEHV-pzqu{xwciyVjXxRT&~Qj6ZEI&09<K>s5R4yW@fE@07s4cvJg`_l;N
zK(-gJCA_KL-g6fR|E(cy{7jEtNJM+P#*xRX_WnH~NrR#S`<^^kM5*yAZ!FdqdJ?RI
z)a>2$EJ+IUlW_|aHG-R@%u@MOR!B;o{Gnt4iL$2lkh$~TEUd9^P8^%+qiF!6i-*>{
z`iLo4o=e1G&_r1Jt2^qZf!zNW7$Zm?A7}9h-uxwN4YgOvldyxnu+WROz@D3z#aSR8
z*&^YTCHPAVg~bH&?q3J&#b>`A-=+hHt4_C0<T9wAifkejQy}h;+NN|Ijv;Jn7{m$F
z{ex`LQ|c;bI`zf?Lx8h~`Y)-$V0hNq@QQ26)QV}cZXJ#{hJI(`d<#7nj|YC2P}mBQ
z>#ohBK-5*LWw_~bTm63gMWb10SzH=_gDO!qTtIT8cO0^Sho#f7R2g+_x7C=W^(yp-
zpN%|AxEaF|d#vIt(Y{H5S>_`rcnr>!=J7H-Rf^{@s}h5R-Eh#|08C^WWRcgJ%sLR<
zRajZwa#GDfd25k(aNEn%c@nzBYC1_hgrqy169D$_W2r`OJf;cGZ<Izb8b$F*qRs?b
zyO~7b_${`|nAbeHiDiNSWy-*nbqzGe^MmZ0ZM0}?!xe6G5d0p?C$RF|<l49M$FpXy
zm%<Rn6;3;F|3hz7>bP*-qQfjdrPFNvSMOMUD8U<L40S_A-A%EXD(vT%MEq!?ezthh
z`fp3qn5!NTi}t(Lzeb<m3r*_K3y)?+DGb3W%#J=Tt{Oa$>^LdN3@qz)Dsfq+WPx1E
za(0x19zg&tP^kwDYT;3&CnB`3_RjKXOwSTLm7S=;mEsG*M)<dTNOW}L)(*%%G$e=;
zu5w9Ob8DyOAbM7s_`K^xynKSzMzO0WKn-B0&lv@V528+|*cH9d&8U_rL_;Mk{V%8-
z`xz$p2%o5mAe<{kCzW8gB)M3+a!Qr<?GkjP2pmx{kjc4J#o0R@W8#dc4QfD4W>rNj
z|Jq?0Eq7Wo^Ff;pUpjv(_K%0x9G#3u`8s6N|8{BR*+sQliuOI<`;>^$fuNStf$y54
z1s99b{%yQekkQ_^fA{e0pj4i%3flUuGJ<~m^x}klq$*;<gZYvaXVvA3(Ta3GMjL7Y
zD3%-IP;Yghe;fpTUzerBeRFI(&xdTF;S(?tSfw+}&D~o33>Qjvln8ErC#~`4K<Fql
zR--}!wultPhN;;jQcJv4HjJ(k?BY!s&pBF*&9x{MHCB;s`Q-^a2Z^qUYyW*Jrn}T%
zbWp0UU>gj-q<coYoj5ej;8pV0EDQ9i%DZsz058eX;F9sbRJ_=x4|uxjzcT$N{32XX
zk=qvS?|@reSqIem%}rD=Oow_M(NVXcS`vS{T}d>8ZF-~XAth5@V@XfsMuu!Elf}@^
zL#c!Pl*nT>%G`wo*}skb7C~Lh*8aybH1XKMNk(XR9K>1F_6ySbd_pgv2|KM~NOMAN
z>R}MgrVJHl1G#RI;Ls-eeM&Oku`IjXKSi1RRbdc!WV!0@<BnAPkvc?s;pKAWSoLav
zN|+hc#-kK*QqJStggkE4WH8(iW{dWk{f5|>pR<q%ZrvY!%mVq=u~P3<%@STirOla3
z<`i~}<XB+a-49r|^*xypjd*RnA95DF8=D=?I)Yd6TF(Jceoh-$`+A6Q)@&H0Y8?s$
zLoMOGN_F!#`P-s|Fp1RaDK*6U)Wn0Zr&Q!?Gf2+>{qd%@eS^;PJ9uXtXhYK3+Z)$a
zOa`shTk3PAj<2Ct9|L5!ovuYrydKj#M&8{&CpIaDTy6T|X&>R77TG<CVq#)$3fV6G
zR$7|_B{TJ|V@&)R+ub~d7ViO5o_A(m7a09^4&g>XGHg)IpTo{R*L*c}avT=g)0U&7
zV6!k@dw!c2)GOljh<&5|kpUU=KxSS8uq9g+W|WMrkWBlq<B$;UVl>MDBu|7w4OA4R
zx%VQd2?SS<@TA-XHLygLd#)!bo1?vZbrHymqYj|V;t5Z$dWNq2KbLod(qX!*YS{CN
zy(CADrmDu0(ha*jB0}krpKx}wpeXlNr#uo@UKn`;@6jM$cubT~U4U#mm2~70P3b1*
z;u+-;$J}nq-A~O8rvtz^p?Af7aH1_%)PFs9dNWywV{(^cPwkQ;*WJal0eR&)R_C2J
zNKZe^uwz}`v4XYj6&lQEA+j}b5&wKOj|S8|K6?7NHD%6RR;W0GCT>|Dhr%B){OxLy
z!ihcC_<BIQ1|l^+-<%8w;Deg(INrw*k{N<+=1nTwqe?&DUeN5wbZG0^5ix#D?iPUF
zVP6b0jn3B4Egv;1)NIS<;M#p6toQ8X<xqAKacASE0qY0>l5>{Tql513|GO$MGyzc(
zY&1{ygjy=NB^z3JV^hJqX1J~dNtTi<t3@;NJlLI!AWcx?$ACPV0$eYMm!Xoe<B`&V
zarjYY#6(^U+bybKqxJM3Qq$GSRH_)isF;fT+5E2b{=rO_Vnu$RI|pQ+fxxE!RH||^
zW~02ZE}wAY)5<C4ImTc_Q+X3s?*U94?LyY85P~s3q)IAeX0KHXnjMF+|C{P@RO^j8
z6O3upr~0pAOp!)WYvJ6Oc+yZi>U;_i=>9#bqgT=pA*Lqw?~~Jw@M$r`PYadWAkxCe
zO!9UZi%%l%nKgIlM?weF_L)pepJhnw^wAx^dDtmQW)|wM*SR_%Y8(oMPe+I$V&sh~
zJs^OK=Pu~}&C0m!E=Lk<p2JF7x?Ty<W5m4?P}nYM?)QF+i_}`aut2-m=`Q2_%Y!q&
zHMZxcx5#O-Rs(0g4mUADA8-wGDOQ9=pJQytkpmY9?uwVn)~~_qlj3fn2u?7g7atHM
zWIO)iAuZv6rKxx<@&~diWrUb26f+Oa&Ipi!z4{TS$IK_0cusDIc9Ie^{xnYcKL>%8
z?_@sooGMmc&(zlJPrBraWe7O=h83RQT4=<>QQP8(cv#dqydo_Uo6fz1WbLAotLa&T
znq!|1gX~qA3$fwRDZm4j38ucE@_Sssy$OqC7h`(IWCo?mfgSy1y-bTJAkY9_ekC$r
z6%{<JjN1(bnSqZ^8L|x~=5U#NFe~7wt}#ltkZx*h-yD65BMlO;*&kh_t+@!!YKN}?
zwrze><fV3loMKa1A(I%PO<d~Ly3Z(+YqHdMGcbvl16&WeInS^`rW3U&j~$+&0{wq?
z41#M%P@$yg3@XICyh9oW9peoEjJMrd`TIB5b6hCHYLyFaU-G9fEHEE4HdEzYRF~C1
z#p(|aB$wr!?E(`j<PeJ0_TetuPh(mc!1ff3-Hzhb{*)46Z!t)@<+W{G0QfS8X|H3b
zbN&o6y#|q~BP{>Ql5cJYep|)u=1Dx%rT+A6H(R4z*JCt28yLCVR&eb|B^k_H0pP$f
zp{X((MH|eIHaR0v>7$A8OpO-lp|YZ$azPsWxXN+%=~nM|M0IGe7ScV$?wU^h3qJ6i
z$O56iFARnlqj-Yj8$XUGOV1oicgV{|Z&1z25|ex&4s@p8#t!v6LK?5k`M@}*3GZg)
zOFDm(cJ4sp5U6V8+P9)p*)`U17z#SEI-UaM9>BUyVxmzBK?hlB*Iw2NL1#^dkSQ^u
z>Tcf2T{h}9ef;=VXQ<2mi2;~!1K-~oxqH3DCqf2<AD^?p7aXlo7Tsn*V^x~FskQSm
z@Zs%Ol)csTysj0+FlcEBa`yYa0EGJfKAcc%)eblav0}S^7Hm!?9)O<e^0M5wM-0v^
zg!y;332rtgQ=Wb}y?#T7-N?^&EgB>kEgsuGG0HkvvyRP1yJi^jWB3fI>1Gb&^1S+Y
zR{Y1uK^2CB6lAS~w2%l;k2p9-B9v~VUbPr$XOoAS1oa3!F!bUpSrCAzJu*0rt<O<S
z>7)+hN<u$7C;q?EMfu^qMau<?i~39uF*H`5HFFVTKer3*5nrYu-{3%nmfV4sOCyx=
zGu5vEq__ZAG%?G$=n?xRe8-S{%<fo@^?YuN&R}KCw6XboMjv)Ce#oT%6(7g+zgg(m
zSvdcP7so`v$jHXP^gqx42Or18&dm0IJ#_znd|XS1i8A)~Ds6O^Rf1ssVf?=4_D+Gh
zKfpjB+JE#o`o+V=!VzecG(iVSx{(t1j^;DoH=o&#y*)c);nW+}w{54J=xB!da;pX@
z^}%VvMg}o<R|If*i2Uk8P*-PE{k35AuB{*)0BxLq1n^+MLCMe{*I*%>l0iEJf&jkk
z*#4OzTTUlVU<ef90|4<s9D>@pL8$;v^!`x8u+=vwAb^1B{eG}6=~dWWoa--~Fb>V3
zT!QI&QpNhRx^(hZ*lqk)_DxNH8W(acB;fxLtqs7R>|26>`eRc8MBq^2{HdmoHXuP6
z25<mR5wPofPytLq=R?ndotuqWS^yj~Ex9f=r*Y<>tva|l2ef~|qmolwV{x$mX$q=n
zfdMR-0FPgoTl{&fgmUm5wV48!Tk?<nJ$uN0E7cMe(H7X`(*YlTH-Q5{+@l>HLj5eg
z_SN@zih{lCna`D|(lLbf!2(FL+1knqW@qE!;o+>_*4pAQpj*L_fAzgL+vNYJ7E6!9
z-T&ADHH80=itZ{pVa~B-i}$0wKa&QSl!p#4qCS>Opt;r)2s5L~i^~16{X<`8Mzr-?
zZT?f^>PH3oh+|{ZU+Zg`yu6$Q0B2ZUfIWn8ZFZM+2JY(U0-FB)%*{iP8ubGV1}NUv
z(f)mw|8N^W`D^`Pca~UxnvK5p`R>R2^D)#Hb5pzXvwzClYW3&r6x7+#{%auu(Fbh_
z`xdzSJe$ss_4A=Zs)TA}(%ihBoEvm9Rtw;5IvPJXy?AB+n)5_joD>8AB)|tC83zX7
zi&7q4P=q?&zmMKTJ%aaHoHxte)~D0W{-d?x7E2E+n)^Fy4dl|){5jZJ+=);cL%h6(
zQ;7e~J2pcf{rjvYv;`mlIE3+c*Ic*%wRlyd_S??qzcoYp_RiLq1)#3)p&o{}1oiM8
zx_1EW3jx^H-X`?r`-Ax|3jsU?R~Mw+&Ncbvp~CP}Q?qXl1O)jz@fCf%oBxCGzqm_h
z;-4!gonF|5)`x5u+*c0i;GGJ%_{*dGs?%+}eSNLe2ZvblWBlB+&<8K?S>s3kqZWXF
z1gA`*njQhP{?lVDucBE-HmrbiX{!G~ulakVZV$+Fp+7LUf%o7`?4!%!FEakC-Z=9D
z|2*3Sst@{&|HN-F<J4T|6x7xM-1hbpiZ?_0g}!5c@|y?ZNJ()?E>6+-(_-*zGRC3R
zPn8$50zjw72e>{2dL0DdtMZRZL*EDXOl=|^eyWoO$i|6fXgfplPwL9T?}s`Kd^n>3
z{Q%6~=dbNUgBS<hZ~hFMfZh-O9YO)H-~1aw0kP-$u7d;u;=uhg2n5Kk_e<vikWKST
zpYz%I%lzxGc;N^BQ=94kGdb|H9qemO&4QkWD+F)x{zZEii-cOA0ll<<@J=N<-}w7}
ze<MHoGw;9eQ+kfGl#oa-gXe~a$M?_q!lVd)^Pnd{ppV=0Th{!(QvBown;Uzv`<jaZ
z@c+j{h=u{(!~lIDzZG42@939tI00o0CJ;MBmALj9DIpHMk#vJXUkuG|PH0FfKpVhW
zC#gUAC^-6SArNcw2~#lq+NY+aR7Fll%R0<H@JscJ!oh?@4BCmqHNZ*r6o8u3JpC03
zdIz)n<UM5}WmyDt9*rXo+3L14>JY<!kXN>ZrP54`PUw|9>z^5+eGSQT_At&(aZ1j;
zzmsblfn+$)g9J^pWO_F!A&R1q?b!+8RwZo>0n)xuj*!e8e9s}x)ij9ytZ<w5XiC8q
z_DZ}%t0y&M%*$#`dY%W8Y%19jUD&k7)+3_Hp~`3X@Duv)%ZihkOrD6aw+Hs5tZS_C
zqBuK#Hpt_utYi`mzC1XjV`eSGs(7<W0C+TJ!gjZw%=+j=!tN3`P@E)T&yg?8@Ho9j
zRh8Lt;J^{-0M|CW*PlUDkl6V+dPQI|Q2NN@WZastHZCA^10jLHOGcqkk0n%r%+nnM
zM{xxJ4>hAwzR{<&<|O*%6O2qJ7;GWSyD)n@nmxtSPMmi=<#Zq@D!EUw@_p-BlvPtK
z&_#iKmBihfZ0W>}xtQl7ay^O~Jx;f8G^zW0**834>p$uVxm$a4`7he-<k_=o`d3Y*
zb`f3>br42^zMbUW`0_<m*w(V4U_Nq)^;K>?#6&0Z-oOVXRo8;Kv(O|V<MLVwx6m3#
zy%k~smx$ct-nH&sprPvy=_E#&f4^o1M6Y``p_QBWCkXH!?pU^lAA=z`&TMALx8ixb
zc|j4H<S{%N;1R<6DHv(0Z+9mx9wr67m&B<v6we?VdREQ4a>gM9LAlVs#A{R-nAdp$
z*^1pDrA15c{_3*6$iVg<NV3yumaZV)64x5W|LpOCnU^Hv<g${0S`bmisIGA5;!jgA
zC9xOqt;x0aq)estTjxxg(7$)^^Ei&q?Mk4S3y=*B1BWxww%-MITchgIOw)XmmtcUO
zLDk`pD$^2l)W*fbyzm0pHlX%<*TMlA_NvW$+YI)X+rbs1Rd2Qhy@0vR@r&xDeHe-w
zH$wmisg(zuri#`)+DB9;rwZA~Z5mHUTAaZ<#=i}vGb2xcN$l>_dz@JBo|t{rs)5qc
zt=&Y2ri@_|&MK1q*kP(EO<j45>*k(i>^?qB;xW!PsmZis6ka3QAF2*CB_Cisf>C1Z
zQdLxTX0J^t9Pw`Ztvri#riy)TzQq&gizt}HynyIGpG1OMy^(s+8mWN9I4<pUzwV3Q
ztURYd#Yxd_?2Kl!!F7cZelup-|B|DiKQU}<!X{x>*sP|;zKQU5fgpQ4_9i#hMw>bB
zM4Ynn5GIq@S)PkKIIo)#_~lw_pCWnP8KnVxz<N->{7!_c6~g!i#gt`elN+%=QfwI5
z-|c0dJDXob4Fi2a0#mJVN~@y^AJ2`b2rr_GFkj$jByRApfe-rIb%0Jo5s4Bfl6*^q
zj<m>M9CPFEXIF3_LzH%%(tdeSGHaH@p-Ego-W8LL*MrA}p#LVTk0cQY!&n5qJ<C3>
zP>-l?xyywbcvREl_{jk&0?4yPj>F;#&Um!jmZ2QH>=U_X<uht$y&jk7q*Dtd*ZWaM
zL7D!nBEuJHpDgXn$e!&5S`=zAAQ`_8pW6cucE#V~B54;d302K{)UaK>78MH0Z?$X>
z6~-WiFc3AweEKeOQns2$)GuQqs~mW~vV=)y=3_3+i$LLD)80d{#HEdwwWL!PFq?!c
zglWl7#yp%EyY|2Ou}N>X+Qw+Z&osi4jgg=7A@y)g-1!uJA`>Yb$K$Bnxsvjso0kKL
zZj}!-wI{nU46Bi6deuALFQ0BW5<U$&4S#dJ<Itl{qlRyg7UE-{l+-T!(NV2&blP4i
zWOns;nFDh3o*IOdZWW^j%$1;OP#y<VZ$mBy@kOGAu1G=5S)YWvO#26uq-bQfdS{5n
zxJyvzVxCup6}3nw0$2n&!0%w6$>xtoT3`K&8{V}9?rv8Fl7zrL5ijzHfR|kLFfYwc
zg6nDk66h^UZ;|AC%4UD|DhQ5|m|6y9sHjZ@blr@y=hjHp^|xR|H!Y4p1!26@XjnLJ
z3sFh3G#%{DLY-EsH{Qo$*QZ<0)2F>~bnzt8nM4%K*C6-SFw0t<e00qZAJz?0jgNS{
z<lprvV2<{_Uq#qMm)ay4NTXtda(H<NJ4DFLNd8#-Y7<yH;GC9sOThGsDk@5K44(@w
z2RZKUF;evy8W_d}NxcqV*7+YEGjmpdh^Ehem+am&><(>%a^O2qqv=Rs&IiXIP3>O^
z8ZEjNx_ww^>;dt@cxP8(=^W)k;>8sXXO-D^=1cu=tb~P*?vEl2%4F%^n=k^t$j_gJ
zma<^@y7`QiPjJJ7Q=%P=ufq<}`V)7!c$#*w9^0aQ<3u|}ik#PRXhB8(J&?!!h8VPK
z$_ztykHS+tknqL64>Vx1oP1HIz5{bSM?eB|*$dgts@ae@Ni@@k3EEQJ!jB|5{9r4C
zBlZ}H-hj#mBd>e<H?kP(SvFC;L-2>RG-z6%YxoWPJ*K$UT&16iu7zUouyO1Y@X-%Z
z<NBMo!$@9Z|9W2RqY3RC@*LlMWVsKYua)?xi+OI!b2_W^Ih9}c_C@`tqJiCFkE+8H
z-b%Yae>%h>pz5OX4O?<nfODgEs=Ia(IJT8cr1Q<+c(~P?g;3P8PP@b_28uL<uiG53
zjfZTVmCAHfe|S9~zR4$C>2jV0nu}BnP5wyMC&>_x6o%0%{bng?q7~k+CM_!eY1hLX
z$}q3(td8lDFIe*p!rDwFCt_v*o@NQCnRJsl_FAQEk4?>SQ|{r1eBtX#XbxeLMck4`
z#<YwL>ygmj>nRGW>))Ql=o$ONF&jHn7R4d<C*;(@w8%<ch{=?zzwgyQdl=EtR9Wei
z<DC?4@;F8xdr6_`lf!=bB@{ZWyVg?jGvz>bo^bHshKz(9bWA=o{ppA)H5JuOh$z&y
z;)S3`Ne;I?sL_rUm1Sux%B=UJrzg#)OQ%_Wob&Se2?VkqF~S{D8XVj8k3hu`N<lM8
z>o?DqY-_%V9Q?u5SO)-HiX)Qw6T)XS(a{2}nRKlWkTi5xQxI^j<-fH+gM|HD)hBlB
zvW&UCQG7&KL_hk+w~ye>yFo1FdpRn(8Z6mTBkr{FRUV{s;W2FN2!GHsnD3S4e_e+1
z*(jUuxxXvKwvZs%-?p@lEU@V4#)I61g|nRDJ_A3`mn~WZy|}M3LnjQY&IUMq4#d(@
z{18e<MN8K;u`(7Y@AnCa=azS8*zAfLg=fPYmvwnP!rAbh?Dm(NHR39rM?RWydDdjU
zSDD8@qrpibMJ`U}?z1(isfb=RSGLx)5}ord!21g}$}#UR^kp5oYEkDsV^2=3A>hz&
z`nuI^ybHO;-q)>Dagw=jM8AGjR~64AH&rHlUnb<r4$TIY+gWs~9x;kkUGcPvfruyf
zh4}QJ=`PPYrA!?sI*pIeS);0#Pqcqkt36v?we;5q`M$<BJt%wPjN*MVm9&;o7}$Ey
z{(R&!{y(Fk$}^!aHXW-|#La=u`YAYeLFNX7oHk=n{j-t2rI+T+tOmIb%u>wS8h{8f
zO4Bm!6Y>z{9OnmpFqL7me$Fe8V=s2cF;}JBJo`4$N%`~yH<@XvFx^lBBw;qre({A9
zob8g6#0DvlS5!2LCP;d)N?CFl+*9F7nU5zxP*|~;e@?A(M3G>CQ>aPS4lD+w%o)Zx
z0T%qnE|A$#&*BCrn{72A?CBiN2kL`7!xR;hv$_o|WH-u?t|DI%gAr}~!;*<<OcnG*
zBihifrDQ{MSfk89bw4MAardvUtXwW7W$7bu2j!kQjR#&}6c~|^sFlsa7&#q{mFQt<
zm;<iHHk{m~+d94CQMu>G9Nn5jB5##~_qHHa;uoy*Dmh{P;*o`yKInsg5G4#L;Zsy?
z=5gDV6LX=;Z|?~rx4yR|8IdZe46zZWB08d56}bX_NO*x;7n4XSWuZ!JD=(kq{^I=S
zB%Op_0wwQHX;5VsQOyM)G-(q3H)c<c(T11x_VTp(>;h<H7;sYWiu^0>FTI-FU;`Ar
z%(bJZ34YU31(d0ZXVjf93d8=Z)-<3fkGc?J1*=#pNY<<0J@-YSjH<G+E0{)tip7&v
ziP;-_BtE3A(v?tCYtP~-5}M<SAa-WoYtdaBR2<V&EaM^ex|DhK2PV6!mvC}YX4B?l
ziTlaRxo#_kt`<&QZ<JR{_hXZ`+p3W{eu)l`4Y^iK?EZX2qDYzAxJM3NgRmrd;OWlX
zg5A<|1t?3M)?`_ZEW@^Ij{jgHfu74A>1dESQCdb1=pt_GbPHDqA5MVaQZ>>8K)N=o
zEm{vu-0|3qp1W2|40KQLC#quhKjp$%3Fndcc%bGRh&z9b))K}2-ZTaB<izmoaBBPh
zW?SJd&P}v`cSQ_z0kNUOOs({U78^EI(#x-?z!uL+Zj$MTY7wIybKNqm=4;a2h@_HX
zTfn?pF5Lppv^<M7k@j#!skYRv#_UW~XiZME1F#_m&W}+Br5MVq3kP~=hE;Q~7j9tM
z(yEZt6HL$8Sf^3%m8GFPj8JnjQwjwe6>LZ_BZ%cRcR{P9LChQDW$xXX;z#Ke)Xf!T
z;!VvAd(avn<6Ox0i_!mwv2zN}ENt6sY}>YN+qP}nb_X5Xwr!_l+x8nfo$0Bmzvesm
zs%8%N=~Meb)pf6Hohn$LzbQEES3=xe1p0Y7sshrCj424)nx$z_9MtY{AKw2C^`Vh4
zcrCPIsc=}#5VsL}HKr!F3^Hjy;>T_RCkLnWB$3tkqcyVzJ9W@!`OzAU9p<AAdh2jX
zP=3Q<>&5ldY5lxsQSyEIz61F868PcRU;H&$7>RV+S|?;`M_3B-$4w;`BKVCvD6{f@
zg<P=$kI-on59=-i0}IlORD5d~VP0;_LVn~hABG9n(ziWd%?Ng#w3jOBTIQ6)p+!as
zY7ZCS<5xIB<BdmW#w&bFsngq0u}-wHKx%;~FdwYF?M&?D)G{TLKemsh5-l$o6?A{P
zstrtkq5WRnE-Nyff|H!xR)9N(enGQnCvSiTnMOtFTcxy7r25@vD=N3?M=M|jf&4KR
zOs`>pg>m=WrT&eaqHL#%e8X?gnuTsJjFb+K3sLPliusD`1Q(5L_$^8h6sA<|1?#|5
z(!i?lYiP&E8Nb-;ig3@{J8XhI3Nt8tzFa2_f$$t!7Gmz;4d4gnMOZNHV+rZmlrr;N
z<tL6?Nl}~A4}C~Om^1=4kiN8|QQh$fF;;TJSF|ENdw6`d@^r2HFe&Fv|Hg|8$9<%*
znv&cynar@zsNW=z<xw4P)0gr7Av5NJ^vTsCae&%H_HU)DJxywxsV(x>+GD{s{;m}g
zVp~XiFWrXjf622Wx!)<_qRif&A_hs!>U6-Ch%g3tA@i1Wnz-ijspbCDo)$jM+?*3h
zpoA<`Ixq8KB!QJg3B0t%=r!(|y*_FYkCAxd&T}co-Ps~SD)%)x#2E=#Lwy<fg^gM7
zwGfgbh))}j&V4E|+0<AM>*;>bC{AYJp?|Iz&C7-j4K3p$pvrS>XklA4Ou*UFZ|pZE
zvYEO{3;DNOY?(w=PE#I`VS^SSs9=wuA9X^T&mVoR^{2giXKC5OjRq*=XhYkI`{3Xu
z>&mv)mK9)TjgJ`F4d0w7t8fsUq<;&*j0++_bC6#)QwXxVoEd%Q@mNdA(R@%``v>w3
zfhyK9he)r5k7%Wn1P%K3gh=i??0d+RLW~-3u%GZrl857t%8uxR5K6B|2(t>>^j(S{
zwh82H{moSg!+s|2EH=*4pV{#Yrnpek=#JG_<hqV~q|w-vIuE7)XIYXqUb2DWV!mr9
zORJXuVbH|8TaCNcxuVuI{LTpE)m7%YU-#Hl6vU}lt=QEIcP#ik)Vey<9EN?7+GB&R
z5z2Q&D1XhxYyzB9>i?>E19ON|?k)$N#0_Z7!ireU?!;SRh~exGV1GS%7vK(zMpp{S
z8cAWCC|V3C74hKmou%GgL2)OgQ-QbqZ2<A_rENF7W9G?E>&DR=uU0@n^!*i;%dh2m
z#{201bMlubSf+0gc67e9wm|x#)rYn6MjdA+AF%t>kGDVy>k)FGJiktDU<K24KTd9-
zsag^CDwb#e^(7;L|2egN1{?Db;nSDgS^ZpXyWoErDyhg8Uehuu!oMC<WBl-CZcy%U
zan*Oao|1ghYZw-a&i*!1vD>(XFTeyI@?Hn!=uHTrQG9N=q1I<9kJD5Bw{;VC5q(Br
zzw1@d{9%t&44(nmxCN;mY7!W;-bFR{t&4$8JI;Bv!a#oB`RSoL?q9L0(L2*%<qF7J
zqomF`cMGD_y9Li$Y}cqTUt@~>pxcWX{A7b6$uBUG9Q<naKF}QbK|8f%9VtDi+l@->
zdbyhPBIJs9(lE&7-UW~NBdmj{y=VI21hAM!(pPIMKp(y%6Z+`20n8^f5#D91aog|K
z{Z6ntG@fjD4tqU7r_)FY9%M^Dn-C-F?ahYCGm1yD^2bHUIFfV{Vtl)4;nvM1B}f|m
znyyRNj}g~{Vaf>7Wi-wy!9zY;F{{Y#J-<D(qN7$^Sh0j&y;n@c(Iv$uerb9X%f4M|
zo^XnE2bw5(HAg?pM2goh5gE@}HL~j%x4{A$FDemLb<FS1)T>X0LanTX`V#{M)4D?(
z6kc(L>YoQs5vrLpvLq%|p&c7x4Q_1_(Ww2)yafbo1B1YwiO5hEB-a)3aLNR$mrJ2#
zsb`#39p&CGoTYaN9PKFKW+^)&SmukN0n^4Dxkc@~-7B8B$7mh|f&Jkryi4DMau`TB
zp?0A6tgqE;)$ur@byI_N57AjA3Vg*X(pA_Sp?CZ7bE0hQer`iXV6J!mbo4(NPmAC3
zBs|bt-}+pq%8{s4Bn2MsQm4P=vZo^pB;T!xanwHD9>v`ycBlqm-0y`=m_D)x*VhH|
zU(TXjyfOSsK+cx!9v*i@*<{(at9W;w$Nle-a)sCoSs5sCF&^|(d^@jb=;iLWJ@Xx*
z(*!Q$M80-d$CRU%Mmw%R(wvkCSAI2Qm5_>1ta(ER@^t0+`*)aL9PunJ4F+Df_;W$|
z&7`N|Y?G@-t&6P6B!p7&QmOw^5%$FTUEe!*(1)I*wgGj~*{Jrn&xxaGMH8o!f})3E
zSUR$Iy8mmV>4dCVJe)8Cf7v79*JEmYAk$P?K=a>h$OsCz1If1x_+>d;!cn{MHGELC
zP_{af<u;?UMBMm3%ObW$O7XH$8aB2m9#O70ds!a_*&9AAP9@NWMOR6C1>s6&8-6>y
zfX#d^&;0s(QqG!we{?SM7A5aHavzj0^9{IN=WiCaw)4=myW7|)Gwc(i>QbxtpY#NZ
zWf4%(GBL#Nurw1?YOxH-Oi>*<laf)=W_t*6^%`|coMI32FFKN>p4ds3E4C>SW=^Q4
z^x{UQ{ohNV*bS6AVGbGx8T4(au--_%!omET?W!;z1==qPHf2>wf>U{#AdJh?ccU~G
zi*-m-#nyUz>zORx1QwHjBZ;2|#FiWH9MPpH{yBKq79Z`RHIp@m%Fyhuw9G-aQ=c5S
z0n>aC{k+gk9=bjLJ7&Wi?ui|Hve`f)x`=4>63WRF#c<R#+B?3F7D@N$>a}II*w895
zDyjkpfbQ+92SejCTbMT75LTky8TaPjzf}4sLxmHw!LZM<V|;D%wJZnP0z{B<2&(>a
zDF$u0LFuil50_sIu)6%Z9E`q6%}|vVt7Nte$>HHR%Lyp|Vq>V{x>zbBGq}F3O}Lkd
z_~00=N~@pQj(T`yP|wX1xlpR2AlgyCs5ADiHQZ&ryr6#}i)%M)w(CUOo(uTIddsA6
z(WhDfqVBh$et<}|eDx4hbn>yEXQ780&H~Ot@+P~f1=_}*a@4{%D932}==no9_NT6b
zWa&j%@zaXFKRfGUKC-D2Pokg7wTo)RVMLwKRXfjimu*~~ly=#T6%6ncdhQ8Q@s#K}
zas$>bHBtyBR2OHL7b8&}5))MQ+cV&OQ_)q{#Sjp&le{p=;h;;&DRtak!jP(Tq)whD
zYTE%NoI&p$o%qnRnc&}jOWEzdOZAmnlOfW;<;PjaZ1@wI*xX8BV=2mVHriSpMtQ<%
zQLIUeQ(Sa}<Ggp5R&i+E-xR7$u>)>kH*1oKepf_zrFC2aVWADR+kbNxwBtx^B570=
z2#8O=eom=WlIC}w1R+TK!;$k5em*y1+!NPW$%y}BRLU@Ug7*0WYoQD8kX2#KCa|x$
zEx<=dx*Ct3NPH!W-_)DVx@(}0WSvG%kQ6E8Rd{$<&-#WG<FS+C#$sM4-%XvKPU$!F
zTgD4w^Muut|1U@FG7iVDIw%R4BY_5H)bOhl6PJeHn>E<MiIIRf(egoqz2GE0XM?;e
z;Sa!L)Zm2<$JSUJIlqdq8wU4!skQQgm>St3=x&TLm_oMPE|!Y-w=H!0V-P1lJjDkC
zN?MN_J?N-@zCc~A!9BFuj!!w+maf;AQKFCC>o&`4{uD8^r6M)bAu74~)FWKv&=R)x
z%6-4`N)aBVByGE6DpFC7SoCr3OybJA(0fKBvu~q9MafUgH>kU4Rqm9v4TEVPTuK#y
zNj$!Q+TZyKRr}n;#_R{2+ZoJw6|S4u^Sm#6PTN$;FXoVkym;I)5(M5kmCvRBKFYcr
zidECw*4zwu;VHLM_!HFxVmbenRWQM5-*Xb(;A6>WV@})txw-aTO&VJI>4X0wf-{#D
z-T3e648D_dG1jE~pllc6Vaox;*|XkoP5{IM`?(vn@tVi^(KsnLpAimw3Q_ex0RXGR
z1$mpFJiY6>oSvw%W)MTu`r&mQ0~?+byV-g34Acr|!6*K^r%N3t7<+O2uN{Ky&T-B^
zZ^+wViTHrC>mH1p7c3!FNJ1vfu0B44&A1ehp({!XFE`E3a^YRrU5(jw#RNPP?<`uw
z_A78Co!`e9Uk~j4_c;e_%m@|GGCum3j)3UBFX4g{=bht41a_dh-}lnSyc4ZLK~W?U
ze_v_GW0?D2@35+qn+ZpLGc1maDzv)Z8DF}eMKnVUzn?-mV%E`5P71nPH(Y?N4Y37$
zr|plH`A|ivnm(uc`*if4HloIg4h<~$y)yJ7cmShpVbhgToF~sH9cWXt?NMq}j!32J
zR69l;$G=tCVunYR9zemM_)s06*S2Ju)B+ZQ#Xzi-Gc9$`(xb5e7Z#rMJbYR#=?tv#
zhdlqPBEQ8i`wiQC10a&}*vi@b8jNQ<{(9K1)mV@~?7d>O<Gm&`WC?%v{-Eq_Zu>fC
zhGMFq8p7?7j;6#>szszGkx|K-;%Jzg@N2uD_Ffe<gbd@h9%**ok8S-pJXg+7nVf;w
zNY<8JtboU*4Mf^X+@AhYw0Y<@HoFod{D(P>I``e~#OWVZAymKka9vcel?2W-i^MP)
zIZLEUD&L<Ut`>WDtx;#be%LKXv#{ofS4@nspYmE|B>WAi_Od+N`NTr=6e~R#*3%zo
zbvo#(7`gMy9nn)A`UDx7<kwdO=2zQeu76P1SaTs@BPTnB$H;A`VvlJ%SQ{^Hu>&M_
zd#9uG`B@yD(kCK$CNp}k)0v^;WE!;a7etX~MnRXmHTChD-4D<%tbgBynkwfbj4)z1
zk-L_44ZBW7(KEkw7V*gnR3*wrn|pJ#I{SGmWHx;C_}RAaihLF(&g!I6+w+Te^$8H_
z4GU24-o?4kMphZ&t9#V&bjdFu%qAr0-=co{J5F!#4xj2{QLg!Z`t74u*jkN*)5b(o
zP>1L8*s6q@{h7C<<)QpH+q>_A;n~E@5`{)keX&GwC|?XJ;Poa~idh)#C~e@)g+^X~
zUt^N3e8u%AorFeHG9m<YE3h2bL5hq%jJ=c7HSVZ7*)(|Es4&UG;YU<z!Qas>S5eBJ
zD1HPfugJS_i?S51pWRD>C3-2wTEgIV2|vOyfH_3!^&(WnYPQqD*O2gYFS-=9#SZ2@
z8Qdzh?~U8Z?<X8gbP;5*Xx&3()eL67e}2cB)U|8SWKk?md^~0ye;5nlfRT_xz08U&
zc0>$5dVJOUhET^$<fQ6_XUe@Fl|7ASQXoYb`$#j+e~Qh1O^h`Kj-VugWD*~@fIgqR
z@NLDQ>n28Z1&*tgLjv;VnIjuN{l}cST7EEusPtv2`p&u#wf9znvrxs<jlh-xec^nj
zRb!hVyzsF9XCB%EW<tG&SyJz#D|BC=7aP|o9B(hD)5Pm%1`89u^A_j=3Sc6kqA0b-
z!qF&QowqcwE+P&t=B`4IAxD<cYVv66yhcsMNAqUfu%r;Tz_xXZ-l=+rd;%_6oOd}<
z?}^ZlxhX6)vm|Db?MYBwCK#oM{r0I~lLU*%%=6$QEn8C<bEan!4Hi}`=r6QdGH&eM
zcTG659~L#yqRM87mgQbfTmwnCe50U{)UF!r^sSXq%T&l~3iz;#-idJ?C>5maPgYNd
z7<$avbJaxofrv)GCs=Y0w(H%lBkvjiB`-3<yVS7BTqi(K_A?FTH$uMiF4QTghcsv`
zOJf^p`uK=mZtC#^cOxa}ulYc=RlP`W6s>j2yB-(Sa{-{O(i|QW!n7-V$?HSUopYYH
zzP4>UsFWERhftc2$^YYF!(s^IQOIM%+GTZnI;VvqE^X^;?pgTx>|iEy8UMFZKBA9x
z4NqI4X*!}i>s$w?h+$EjRRgS;yD2E57+xnVt*Ctf8>XHKPX2Z@T^;(4atV}~=XFsC
zPwtCNU5w5f>$3)Ca4mcer<!U+t8aI3%<86RFw+&Uo-M0A!fo1b?W+C~lo^XLpzs@`
z=8-m$tB5JhIVs$4Z>jVf&GkG?K8RAr;-pw>{?1gy75eA}$ns0!G?>LS$}4JeC|d(+
zEgoqmjhA)v;FjXiF5-A#<@e)_^&Z9CWGj}oX$99|%lS0Qc~^u^f5c`*Hir~T<R`t#
zX~8E!gmr|yFRJxB150f1VzVUIxjPm2dFbhw+F&X(!W9I{v(wt#3vc59yR6}OH|E5z
zA%<~LenM2TVnar}@)fi@Bix(5ziLF!55g}UCNxB_Chu$w)I?Dtuep(-KF<|Nnk_nQ
z@8L!OO}o#a&b5n@r6sLH`6Ztuaw|`Qk*b*hCVB59HyJ_u?aCTVVZFN8TCKg-5mu|`
z=04ef?CKx@G*11x6vaMiiVLfs6QM-!kzP;IQD4(8Rngn5`@KE0EQ?4nl^zMnkkG$C
zX0eEQE-KsR58~DwKW<257hCk>BguJU&4>a1wOB(i9L%02a{o;1%dLY_`peMJWQfqD
z1tXVhSIrQ71eTjqU&_B$94&Y3x0)FB9+zDa6g1e2yg<R0MO;TR93nu==&+wV8S~L2
zDWG6?$nqPu$^^)8lxj_qot{|Dz@Txmj-NNulO~yIVo3OD#!t05tAg2-Q!-FX5pDx0
ze9X$EV!7b70ME_l7n+4P$<W`)dtKTO3{=vy!?L^#cj{>Fe9HD#tpV(Wlp4D^KE1ko
zEk^pX_LV3o<+Xw@|E!cz<97$#XI|Y1I*Of1gtK)3Y_TBk7Xyq*NJz|N65H#8=7g0m
zlUPieHuKpg+L31Lwg+aV?X`0h9AN)ZddR}TieFDbcIb$NWI{GA?tR1-9DP+IHLq!p
zpZQ0@M!J5aI2N!;qj!dd9@){)r0HkvkO&idHMVQ6gq^kG1szg*^J2e*4t<#7je_K2
z4z3`#JZ_M1?bnNHd^{PJOo>dwllbx|$!Ow>_m_ASF6XW2Lw4ypZOS9ANmM14C;noE
zxX-zY7B3@@=!{zCOWraqa>}bdRv>oByNQ-0fi0Z!jShN&&dEy3D}|_K_`3=?!0%c}
zqgv#m%z{<VBWsw&4HeTpX^M60t8T#l{O>y3?c?8s7w}a-&FOpM%}fOhn9Tvuz4r2J
z7Q$O5$ztR{uLsJt<6dOAT^ZBYO|I;No`h$-kgdyx`)4!@wQ@F8?I<2qv%z_IH_Pf9
z-J$2~^p)94YqZ$ACxR;X&c@67GO*ubqpa?J8Z%6@As^xdHdfb*nDaMNYK(VYOfgJK
z*IWu;C=iWX{(aT{f#01krAyA)Gleo%><`Z2uFFp!^E**%KjS|e=P=0nqAD9-D9250
z{a>wm0mJVMymA)stSpw1luHWHd<N#VQujv#8ELU=&Yyy{L!b$L;y~!9`XquiPqXeu
zjE)steYD*clOvfGdvjf1jeqnUyS^{og8uvl&yiXq2iC;<eMiFfbyT@8%L9Z+6T!<F
z`R_d8(~*%L|E9fq)aQ60DSBGu(9$k^lotVmoszlA$lM!$sE@^o+Nr`>Y@Q_{+`!>w
z<t^?qSC&XsdCpFXQ?wIJRs`mCs6f?IyBmj=dpp@#v7#ZL$)8y>Exqk42;J8ag=?%7
zVAn^S{UGW^i4?Gt)%4qwazm!!_=X!KOWWe#TL>BNNmXn?;vJz_ZvXDP(38EKhJ6bR
zoL6cuL};6xX)~@{fz{&AGg7{W&DNw2I?)I}S*^D%Y5+%irRLN_rP;Dh(Kr0CD=Gb5
z*Sd4Z*si4aNjGJ8!(=3>J*{nxNt|tSZn!A4u^IZ*5+yeZGqIBk-yi2*d?%wsAx3gu
z`QAlF?S3lBOj@Xu(z5O{`MHc;;{~f2r>hb%@LX<d*TvNjhh4yPk_L|>POgMoll@^g
z_iif>)_s3=UcE~bp#Ic8Vy3=JAO5vIpxIL+q7B7R43VDot%1YUZ+H&=v29F|PyTdw
zq~BZ*!yAAw(znJuozTyxLS-EI9Hi>cOI<L}gHT@{kCn0{+>$A*t9tJ<R3K7t@Tz!$
zbuuK3xpfjnzs9Kb*CtjI9!%StZBK039>nC(D+uE*hT-CM7)ojKY9wD<Q-HA{r4Pk-
zC!|V?^f<8uc8fW<lp8u!zs(l;_``Vjt^5)J+cF(PGWlfZ4LN!%PYbJecOsu7SH8iV
zU!&N}C!;xks73I@f{8k6&bWz1^hgQ(K+Hgbw6{jED*tz}kQEN+@P!RRz&J)^K9LJN
z()}n&al>;SJ%@mufAVuH;6sGFqOSH(!%8@2UFjFG@YhtE(96La!)te|JwamGg=^-F
z({7LLOz&3fAh8?qGBz&~h%(%vb#E=%CeywqOD03HJFZW@3P`HFWx9}10*D=XS1*yk
zP{+mzydj~fAckiZQ|##$uvC1Awb=GGo8(NJCc1HYgjho!k3&bSiWtm*M=)<V&812r
zY{{PyNJwZLV)(<~slR*PX~x*S!#Lie;=(4)zp%he`-FH0c`r$?P!=tmj!}z69AW_t
z9d#?$qZ1TYjoCinM(A14CC)>O-t725etf%qFP!B>Bg!xuyFi1j#q`<zaHjmQ#_Ot_
z7Y28S-V^?`Uddr`78mZYHB?WdRaDQ1I6Fqua3Sj8hZYChNbVMc8=h|oV)Vs4NSx=6
ztlRI)wc$6(zIS2H@XD$|R%DykpSrHY4QRbu6{jRlOgJF*hO5VECr8k{J5V+T2d9+%
zSfP`_-rDzaVNO$inek}26dLIQyIX4LLI$cR5-uFe07IKs17iFNiAohp_#Y_?WM_2S
zTxNKeSO2Jdd)t-)?G<mvL=jw)QLQhrR|PwY?O3)k$SqAVQvip`3{$phEED=dnj+EV
z3pA(z@CAoFTOXI0*KnqDFJlZ{Qp9SyUl_5e?&=+aNy1n$U2DgQpiL>;L70Sr=2jjv
zQqhPdnV0u7dHwhSNr1)hH~DknIDnZ?k)CNwG9Ot7rj9H60|=0wvR+@=d;F1P#}QUT
z@XH~=W0Jtfp#WBT1qNbpAU3_#MMQ$fuNYQ`K(K7Cv$4W~dDh85#r;qfPVT;hNUF<3
z5Tz!hn{eg220PY~2m9>QyO@qkVc`BHESRyz3ZAA>A~@s@padxlw-gKS$eJ<r+`|pk
zimX`0e!|5veSRT140+-17!JuP2>(;Ib=5NyHwB$et-d(HR<SQI=qd}Pqig=y|FO>a
zC+wm5cq^-IA}xg2bHpm;Cd(7|ZLpIp#pIlbXV)_N7-H&|gVhb?32f?+gK&2OdtsBR
zXG~dUyMuO!y+fZ$z8%-F>GU#AqUNrf1F28LTS~7wSd}>nG^oqFx~NvyCL=ff66MAe
zdq<)5p2kg^a$O`QuYiL;S#gqRHCHzvs^7tdly@%l=|wA67*gPjn=3^*Ou?iTa~W%X
z59q=20j_6J^BY|pF5Zpk?A)W51{e%6n&KMa8*O2%*Npkx2Rk_5<fR)L-?_s1D&`6S
zHAiGZ4MLnxBwK&fnz#es5GDYwI0{wo&QR%f0Nz=PWUf|8MMLq8w^^)9g7Otx0Yu~G
z-<Ak>5q$nqOS&t&U;@o=r5bNf)1e$Up)80xVCZSMEjWO>?fEoNpC>1pw%>@Vb!;;<
zv(DD5d%)#IHZ?_E^7tiW<RHo1%jB2Kjc8?zb?%X-OTE=8iemU!d&`7>f{C;h-=c-Q
zT<vFz3TCNKLdO$Znv0D4mE^0-vk60HteE_^K#*}wI~YQ+F^6~g_3m4V=@(tad7ORr
zCn30f5JA6}!=>uPs?X$P!uLMvz{EUY=uvQl%%oJX<k_MshoEc>%!afNQD0u>f{P|E
zS?Y;(p^t85avjTDFOd6QK{{R-s*!KhWHfo~8J_$Vq1zKrpRsk6a!S)`#rU}3gy4~$
zvb{Y=X7<2N3WS#so{zH@R#ycwB48w$!}_#*E_{j7gUI>XbhE%-DnXm0rCb#(OPlSH
zqG<CPN0xLbH$Y>=sF4%PS{8b~q15T@LBKXuGU`D@dbzblzVgN=A*X&opeE{ll_*=l
zLWu?Ua>cVxVwisfW3KWmi5S!-h4=;w^EPD3{TOi28mjuPtRK9|ZO8FRyEirHuQ1CW
zQFI`6U2l}kLt~$kfU@^1^eh}vDztYZuWdT{RmA0&YXL3RGG2YNsbQ+Bp!Yi<uBQux
zx_2JT7x$)GS5|0~=4i`f!nUrQ(LPaer?UA5ZF)tj=vq=2qj^!TXIe9Y1$ALx{DG+y
z<J(eo?{I<31L1T2&B=Cla~Xq2-&XZ!|ASkOE|CoeT<f`XL0x8kudt8?97&(4lix@D
zjX%i;?O-&UcAPW&G9B)V?~hBLPELiJNT<w{K`4M*ULRjy87=g&iAe0h`oc%KyTS<m
zCUnXQVRiIV*wf{$OC>3>S=97qGtq^`W_=ZC2_`iz<Nn#i`YwzCr_~sQaF2?De1=)t
zZWfB6uTiTlWcPGK322t|Pw#^lQjhG(cuJ^at_&>XAF<_;TWnj4VGJ)6r))=AvDWYO
z45jnfN3~C%z4fPQbzToNnLiUk`hvz(vZUzTtp{f?$U#<ZHBiFv;7gX_;qvm9v^{s(
z)0a!epT9Mx_tFCcA}EZ&FRd;Z5D6J&Hg`t{%^~u~dpS>`J7{HWpT}~~M~z3uJ<dDm
zpql+f+wpq+XV!blJStpnNt$@w%GX{3fmoNTdHyRo!uH?D5mq*~|3O7qh&VX@@7I5V
zBb@9U|A!kP`hSBXac<yhDfVj&#MH#_Axu=cqS8@BJMbjr2-85K(*x7<L<J-B@QdWc
zJM_dmMDUB$#2_IhU+13z_kYi7+-I{|^Sp0+>uvXV>%JRUY_IMogAOo|#JS)`X$K%v
zfUM5{SU~{+3l$&{B2h_AOjaR^Z;1Xir!@0N5GE>g$`}bGK1M=9x?H6|6#0f2G6qt&
z^bHC^654MJ3#tG^f<giV8T|ndBP9cRAi|`=Dj0-R3Lhqt64Bu9DCPyRu@6lG`uap4
zF#ZFLgoKHC@-7NF!6k&GB#Q)B04%a0gy-Pc0=5DAw4bH~HTg|Q%tdCeL?sCw5TL24
zDexQgU!K>Hc6bcjA)?Fyqz{D>;uIVN+%F4Z6@o_;m_8;s0W;qoO!DJ$K3J<%M+64u
zWZMv_BVsh68^a0*Ck*7^2zg#v0osB`+|WO883^nJ6vzolG(dR=a0+Y=gbW$_`&&3I
z%?VNxX&N7<8A40=h8T!m++83AHwF?YBsdUM_(kfCp70nXWHXe`0XR?v5*W?g4X9uX
z=BEzPIy#7_=(nF##Ww)wJutG<z8M6L4${*pXh2D7Xz%@xC^FL2R$uqy55FES);9X-
zdwnuG8bsTNKE(P);vNu$yGumH#7|r&JM!Og3wU8*u^<tULlY37C{&<B3+v&ZPY{1k
z$qB+U!bf}f-Mn}w$)FW`Y$V4D^y8O=c~oeZU?3$Z%c%CVK;FOJBxp!LL4|2lLEs}e
zFr;6KPr9rlm-UY~U&u<FLC|*^0U$v8fW7{`X$N$Oh~ev<AJ89GlK^sTaC&$5+kV-<
z?W!6Ww-9$|pm#8WeW}&Ie*+2aYpekQ0wWe6gn#~7?ki}RXt{)*>g^o2p6X-u|E2+r
z?|lwJzgg1BztuVs1I7QGZ=+FxwDkBxeFav3BcA+FJnSj|&FA^q!=~oh+WNV7{Q`dz
z{37~?dwd!~fm`}Q9R`BJ5eME>Lx91#d%W!#prf68fSVbuzWg{4qE$k<hX%D4Qpslb
zt38Y(bfW(VcET{pa=pE#LB;Sv1R+8F<|V$<fC(WX0a6U!at9IJ{#$`GFbDb1UqB5c
zX_>+~025F2#g76kObqk`&~J|X+Kt$o74#rx2I_wXI3j}q_vfzKdlQWio(Vd%yGal^
zXa_?0WpDP+24Y}<IEQMX%qk4h=f7=fd~oKcPR(~eIKT<wTkbIVjsLmg{`Ge5Cmez~
zr=9j;0qsnKy7zjqo6Zg%pKR$Wx0crpzv3kv*T^HEZlHO(&}E7%AI4g|yh94`G_UF+
zKw`SP6x@31@_6vMwT#>e+Nhe$x;UnjD9V+0KHaQ5)>u*J62#(aEo|0)(Ym=+Kd{_&
z3T0C5|AJnn`_&@78QDR#3;CvZytgAosh}|ahgHHML#?<Iy7zS`+=V|b9h`KQk{=`H
z89dS6jm~YX1+hVK0&;wv&%7N~&AN;q)}`NVu49GNK0Dfi7Kpo9nt^=FG5G^UpH4)S
z=zh!q8FIw9T1%I`*pWH96iHMjW1{GLmsNh%xWVmmNZdyE1-x^E?C;Ka6*vB<xv0@8
zUu$O9zk+}S=8@4>s(sV)IpL#Q&EMFaH#ykah6;4>nEP*d^^$6A#bMDlJm2I*U2_#H
zadYolIxT#zi|)7biY7}Jk1b@PjRNeBkVM<##`fbrcRRfmO!Jl$RiB>Hm@Q9xsBx+L
z{-NWKG)_k9D_2k&lT4|`=^3(Msk(IL<b;|9>>Nr{!H`oTo8sLlXc7JKoy(!K%;Z*)
zWd!G5YsmN_o&H&J5GNbO?(6H&-A*O~7y1Xj@9hC}D>!_8`Q0|Z?rVoi`W-#R<>ozr
z>}t^3EuLfVpan6=cR~<v{L|ub3*JSt;g)*42xp4E3%DsJ%b&>l?h`mSRfmxM#A)Nr
zT$()!>23;>0v`FAMhB$2)ZOaXKMeqD1ork79ktR1eT)_u)}Z~fK?s5Sa%k=OkZBJR
zQ46F^yhnPK{ijd!T_YtpDD<yf&REy>>wv_D40-QWF#E>R&+uC><#?Lb#yxt4;!<+0
z<>BhFfpA#-7|y!o=AeA)3C4-z1-}`~bH&TkdD5khgHm47|Lt#YQQ-Vm_9fapFHZ@<
z*uqqR>pDmm<4W$4D?hnIRYq9Ykt|80aOFRL+kT9(Xv4iaf?f&}5Im&`DR0$5j>Q~n
z+Z$O#7AI((sTMO2t_aC#j{$+tS_HtoTL00L!Y@WY!)g%E+?Oq#7EfFTTWdfY3U3??
zW=wcOm24v(@*JrH9E&F#Rl?S<LQ}%k4Sp_VI%iIFv||k}gWu!Fp(PY4w^rUZT>UdW
z`H5eQu+uVs@h{_Vr@qv3Ao%kyHgX}#merKF_S&QgUmb0J<=D64CZD6lnRC=q;-0fR
zc;s8os7fw#k|66PK7??6MHz#zBi;^6yHZpQv<r^m1GbQsd#1h3S$UtXa5b?>r2#P%
zE&*W8yauH)$n@WSOWr^5AJ6{r1cI5d@%m<3>Ekug%MN67r;5EZkd9HpW8>PeIywpJ
zWyrB@u%5XzP!FXMS4@sICYK8`v~-Tl;J<cA4J@xx7^~l>*RB{I1MErMPr1%+Ozia2
zxoRoQHb>kRNcWq;*jDlFx-~}^rw^~*4=Kz|ln^;PT7E6e_z-5K<d3)RW^)s>+jETk
zj5N?Bj4>%ZJTXBK#$B=*O>b&qkc9O3aY26XLQ^WL3s>SS%-qP3L<`FtB|#4|+Mz7V
zO{&wM&_c;MgQ)}rSZEx7Mw_CL#oO~ulPeYgQM3(u($B%ma70-xMVx6e`*h<{f&y<+
z_<kTmH7X4`5XZDhsl&rs^DFp$DngPI;5LQ;KS1Z1!MH5vM@8WB2m}naHV}xWTp?#M
zERq#T7ll{i*m$i%-c#wO@9uwe^FZn>adGljVHggdDup7|NA4fU?00Yrb5o9``!`Xb
ze^we!t{#ZzTN8E*dqtBr80VR%7sj|`r!Cdr;$FW0I{x_xMy^;&bi!vXc@T2buB2ge
zcZp(M!$;+n@ZfJ$fL2smVqXlKPZ#&%Q^)s7GnP5?6dza~^y<q&%|TY?x{B*jx;UNH
zgMyL^aF<6$!t@WF{&qr_tHJ@xYCA)obG9czN~V%6teNe&U#9R^VATETD=2fPNsEC*
z_8oiyS~^Ptl!klvsEyMeQ3|?&xU1Gj!`2U3H`rYz+|gj)Ca3;tMTo?bzD|z1Ao_!x
zj&2zynQ(?+8S~K=>n>1QOLGRdo@)geji+fYaNx47o+Zt#YaPQk8A48;Y7TvWN7Xn(
z`1LQ$ic&`UGtb?w6x6(TsbExRE=F9Vsw3^W8|Lye)Rs0w8_vga&&z?Gs{LZ*EXT|h
zHN8d0v1qk_GFH~x6%rSEbWd=uK0For%cFfob2pQAjkED1hNZ0M^mjdHj=>6EPtq&<
z8D*5;m>jk2quLw<=ch412RDXokyEoQc-Y+v`g3@sQ52%?!uKk>MI<I2`kNryURuX6
zNfwk$HiaOPohXJGAyst@ss(Pw9)C1t`K~v?txTj-oSnSropJEZ;edT)^?m5kux8Ya
z4u<&qkDS~I=4O38K(EHxs1pr~a+THt(y+~99-+Y*_mbD|Q7I*V!X<kB#e3*em`ukW
znc|XNGRp9BC%+aclW^=}25q~X-QigrH2a_%(wPST<(8*W=0&m2vL$b9351AsysP7z
z7jS&>BADy1<%c69o`&(bQ1FWEwXOC+f29mNaVN#X`c)cPHtSJh#v414>@fj{*;P8*
zF(~|l-1|CHJ73?54veS$6Anu?xy{->04KERYfEcrK)GcYNGda7taDf2=<!v+NNS)<
zEU}uk;f8eCSrY7OeA@@&g=3Ts)e?0`ARlq&HoIjAzex~upXtZ`D=lFonAMlge~$dR
z#9+bEK1{Yc*GUk)C`j%YWCDiX83gJe_&b^Mmc@QqS+TU{9Y-cmPqt9xvC~UX>>Q_M
z84kAHo@T26kgvH>%4&+FhF}f8?QC42cABY8z;AjvHtAdfxQC}?uBx19XE7=oq;1vN
zCtFVNe(cDUyXwu&XRUwY$~tmoTGT38&C<aWH!wuE^~YHnH(f|$^r+P5d>wEsCz%?%
zoVT)$xuD~%ydW|5-D9t2Af~fa>v>qOITAn~(L4XNSXyIJDJvVT{;OlzpexJe9iilX
z!rz2xWw)lyc)n}IX{5%iB1OCsbg*@23ize}74Lg2e4A568Q~lxmih)uv7=1*^M)l2
zs%o_XkJ^4mY$aQCvVRIm#VF#Dsz`-;9(eD*c3@t+_BdZ9$6b`Xq^)&U@#pW3CrAT0
zT;=JaX_#J+h3t2P38R%jVDE!Axk#F~>lh>uAN}y;NblWn<<Z(6k{hO-fZ012!NuKT
zaEsYJFUuwj?SqxNW!`!tAt+tp2i@ic81{~}rYFVAx!cQKv2NaAvKfc*j>#dR@1W8R
zNkN4e^BBRdFGZUj0L^kHS8paZVX_*U6=mB;>_OYMi)C3GqRWZn{=IsM?hx0l{0C#Q
zP7mcE)c1bI>hj6h*ev(gHsb4|!aMu|kv;1B=W#}MVQE-CuK)+NNuo|Ui$geyYrT3T
zoTR_YH0myahl2L~ct6?~10$|gssqPh-dc3i_@@^Bh!-6*NdsqGHwAhHY|6LY#mY)z
z<7eL0lmFs^c#Rmj^`7XLVzOcg`V^xElmI0lY#Aj9q3jN_7@D*Dsh`W`K&RD2-GLjz
z+xoJv0A7DY6TLi<LIgF#A5-kgL!D;P<O(r9&5pn!wZW;M+romT(R8z%2nMmHb~8pL
z0XyI_Kw@~<mNtsr_z@X$&bi3~?t5K`L{g2}=?t=44#kV}Z;6A68T(%lm?D0wNwz$N
zA**@4dj&8AT{?<7rL2xJMht<u9Y+<3Z>UNZc#?06v8tRL?4G5@$IW-sOI;6F#po18
z>}O`PK3*5a!<z9M5S~onSl4FiB98?niJhJOI9^!~WN?*5r4Vov1!j#+SdSeKd<EeI
z6=4{5lfz%HZCyUrR|uU`ee5ILO;-v%y2H*hBOU!JwQ!Y&)PkLPq#1`S8u=4KnSE-F
z&8voG%*#+!u7d!}`Te<#UwGN&soYF8F$i}WU0}Jr$PEHei%Lv<8<9<N5oGbGCnZ<I
z7-W;)*@CBtOmD>c>f|pJbyyQu5>;U=#qEXCMRaNB&C;LZsGnd_^~Y8t26v~=Ij!mb
zGq?)yKix6On{#DT*k2Z-GPWUtOlosgmu$-)ZV!<ZRf`Ap!P5Q7Qt6-A4492$Y0nZa
ze{iDYCc3+o8PL-=5jpeZayk=pxm=Y4FAjphC`h8EM_*3FCMMBp$!}k{HQZEwk~9Kz
z{|y4WH<6R&G^v%kKZ(@c$Z}jF#QEJ#D3O1n#pTy6URpfoVm-WCb7d%!*JSL(1<j+N
z(q$sxMGOMiEsj<cd@@edJ)e_|2@;@HxET9aw16NVr+d1$D8Kj6KJ1)EB3h&Dz^!G;
zrKze;SVE-Ku(xm1ikPoaEZcgLsGI=sa4!pQr)aT7<%p9IU%SV3@T$wye-grKlTTR%
zHla=s<F|Zs_Q*Nh7=szRebimXc!n)}?RTTYxLjH>QMSa8nf(FZF3kn#<VO>fxtq~b
zV0OK5i-tUC^HcCS<?2_{9QBNl_M+hmVZ(r|c&JtYU4gmODP-!{Ltzra-%-5P#&MsC
zq&8UNwA@#ARYC$dNj~%Flsg@@+PY;>VEmS<Uy|+|hU70VZpb6c6uM;vT7qNW8Gaq&
zF{U+1UF0lHMV|HJCxbv?$ZKqx7T-A)HqhfK?6+$np*zqKZQ@1$l;W)$nE;kIbM+|b
ztG0<(xta!i0gq=-@6ZzhgS$^ILq$f0AYd7yJUI^0n>a;pAIK-c?V|?Jdr>Qe91qcq
z_t=e?Mdt?q3BB7kUO(7`AG?r=bJFEzVZM^2oMVU^g7h!^-C`%rVE-wC9toqo7S_JB
zC6$2YeOmApm7>Np)p8)d(y&g|aFaQC8WWHr*e2?5NO75@5FA7fEe;0!m>EC-yB(({
z$N9xcF|KrFHS5F!!rWNDCI&Uyi6w08%DdNo*-VP02&XE1G3MxoX(IBq?mz34VSF38
zT7eDQCEp2t!l#7<v#Zvo%G*3K5AGin9=zsag3G8czK`GFE2>YoJ?u1kzM~PmimhFx
z9Buw*O{E-c7``F9X-2~(WXs#fu2|{kcKv5N_KZ^N(8fnjW->x+7PP50UM+$(TnFT|
zwmr+q`<v(g{=nGN(hw{j-UEUo<5>n^a3<@GItK=YX~d9AZ$hW>x;Wl2odQH^lAJra
zY2ljWa1cQomZimMi)N3*%;0qs<ZbSYi?SksyA@2IEB^o{Wz^+>o-X>mRmDNKt(QpQ
zMMMXi8Zu@koo(-saPA(LtclXdQlP_Qn~}Sp&o_??kf%<=VBmcwe21dR8Aa7kPX;vx
zZzZ{yxOPf)Yi|LltU-QGzUH$L*U70Uj-l~stim-!A3Wn{TSdOU<=fPNT*%o{wm|Vk
z@M{@ksbMnOC?Ct#AgO>?j32Fb2w&1L-N1}K#Ho1DbM=0G27P_iZ^&rGz{ORm(b|4!
zmHg;C7Jfy56nf7<nrX6oZ^e&HO(()|)-K#pI@W;2<3;Y$>Sd0XYka$3PgBZUqq=R3
zCBN4C%VS}{R0+_9?t_}tMByxwM^`2a-lulc`{zNlkl$pZaS!oUo<MiZX%QcH>nfuj
z)-pKM=lZOqAumKqHN^E;1;aGpOQx&bN6YSCpRaNbv7Jc%w)KczpABgugk_MMlKu}d
z|0i$KdsF9oG^e1<zuP6Gb9M+nk1lr)+&<NR>`y+DM_E+h0t=gzirVgH3ebz0RTz;U
zZ9Awd8@tiLzgpz24sYVzgGX*a6Q`=ysrO_G`muu<B)VeI%+`j=^rD_+UkFPT4&4^y
zBk@bLN6nLy_HqH$w<wnVNm0mfN1%2aB4T=y1mb0Mm!lJGj?60A{E`5Z62qiA{<alx
zGbn6MHO?dwcKREefzqrWlX?Pg!iUx#%pbdKFYTD`3N?L1)k9oa=EKp$$Nf$ANylO6
zw__MyD$LRhu1pp0CSdaJjt<GY`ac_Tek*MZq+_=1dXJ3Z(ML4mY&4Rz>mVz$a3E~i
z7{x6PDyAZuI37;gr4c!v8%nh2?zpPp>HDM6Ai~-L<0*jQ+o$_nNPOr0-F<@Vxq`FX
zqosN~5?%c#dDs`d2p%S<)$b9#WSwOQvW5k5SL~$I87AhM%=*!i{P6N)Y0ny2l{}V6
zv*XsXJzmce_RS1|-QiYO3_DhP+eVZ}(z;=(twj@Nn<3n*O3;qc*HZ{?6(=(ad6Qz?
zxeLgX?xP2KPs{~)v{RTGOUYHHqBap;N?B*wF0v{NIZHriI3SVjRgys4En9yUwtLt4
zdeQTME0GScq%uhag04lXP!ySOi~~S&a54HrDw>lu!pf7=YjEK_^)~%R3r+zpgyxJ}
zBFQo&2CjHsi4c>-J3fXU$c|&U`}{uZxY_HOL-x32S<{fk&0V@y9N@YYsu%T>NP4qQ
z#Nxn{t#x~Lz`rD%*nnr~Z8tjrubpb52P3@vrJ?EV5|>}B<C8Kqz`!${MJ$bF?jovv
z6pZ9-3YRZ8KK(S*POM^8EyoVE6)eajJnqwxSYo_5diP9Q@EPC?EwEevjlHZF?1u$d
zdcr6d;M$G+^J*V7kX~$6Yc=OI{PZ@*xUoW4wdZ3%e~+A~HbaSfU&239IcZqkC_fke
z-BKkYd5R-?!!vNC7cqLqQnow!^57%-8EYx%Dd<Yz+R~fV|Jo`f&tY}{X)1MA^&KW4
z-@gRR7^8{U{_S9|B^0X0zOk)*Jkrx*5v87S+`znGimMs7SiU^@mm}vs>+`i4l9Xsp
z<5TJiP`S}5Qc%ZQ?sfX^X`qY6*~HUn>fSw)SFt6(4$+9<T}VoR(h>oSY^PoFO0{Vv
z5M)8B@clukLRp|@pv8+OSRqY#1Z!Kc*XCs_R)2hQ%%k*OV9vt&3VO}0j3x=d-ogp^
z^(#jYIjCqqd~nf!3Fxy=fIe-ruL-C%hZlo0-JQ_7JnCVdPX{75`q%~`E4T!jjManW
zIq`lPw&?wARvYp5=^7hlANgNU456ILNF39O<lE-Kf8wW_*;R7hDFQ;1``zwqj6=*(
zsIo!iHW**R<@;y&1yG6(bM$=-?K85=o<g(MFh;te|D~yiAY^<E-_>xdaA|h5OOf|`
zdq|_4vt+@ec&L!1I)i8Cla7s7gol4)Q`+^SMvGLO)F(Dh1^=V0LI}+5-#lfbZNbvt
z^pj~$<P>?;Nvk`jYy*wb(s|%|XJSpA>z(C)`m=t9cX>q@?=MN<?x0}IbvX362!Zj1
z!p4+04;XvD0MO1C00o#EU!;sjacvyQd9l2?R$-;AK$(dK^E#v%JQ{d4i5$8vf4gBX
zYxiQFzW{g&z!^DV-ZOQ3qDs;OkMw#OR^5m7UBXpa68^sETgi@t;IMGKDz%k~y}-!^
zNtJTlG_*d&Fl8ZJe8yPP)n8&8cBhKPO99+$9Bur%*}{ciV<+r56i>Yvs8Dc=dTQ1-
zOrHu`vtMuiD)q!iX%+uMG4QYqk29afk4#MPrM27R#80+h)y>_XpCnNyb8+JIG7+(p
zkP!-{jcA#qM{!;(5NVF5|8<H>+TT85%IYQiu<>_zsm9?i&OA5lVuL^Re?$b`#1rk*
z?GPc)vUd3^fIYY|Who3=!v|_mEz2p&w$;(97BdRJ_^Wg|<t9BGRFTnkV?C(snyL!f
zWiObE4MJyyM`6oxba!9dreSx$PPzDeGu)4)i=*R0%l&D|%Fy&ioVaWi5O90@9G5)(
zWfUXAV@mH>LTTe>N*)Z{6SrA&xCF}>&Eef99a*z`9|QRNQw)bx_Al)G>3tMlc<9So
zuV#Y7eD^%BVqX6A)vHN-35Zh`;bsE12^kX51cgkYS+_8lXgGH!eAK^Uaa#q#5*#j{
z?(Uvm1R-4rlj8TRBC;8sC1iQw&x(SziD~tbzV)4d2D6@bOZlhY>ZzGZgH!L>Y}ow!
zGjQCmo7<<8^x<mnP4?1iiJtK17(Vqql*nT)i&@-3p9jBs;vB)PvX4~(`-~oTUjBO@
z8oRgg2XjMB)13jzO^>a4s;gb)4kHS&HqYgL+J|m}?l+dPD#DUWQ~BC{xoYEUPY9&h
z-u67Yk^wQHbdJ2NJLC{KEh^jm8+cp?#!_ht)~Mo*9xO<$X)Oj!FQ_p=#8;-S86PS4
zYcgO-Z%UOFUCmDPBv;BY<nEg&d8*hBSOQ*msb{Rxhb4*&mouZ`L~(l6cSO<k!_36V
zaguqIugMvx3gkRyJ#Lp>jH?aSCVP&?3o9OT0R)98^8DH)Mu8A%27_b3AK9qC7f)7v
zX9Wv%C=2~r8@$ILDc?~bzL*m<m-ue4qC9i$Iavny;mNz}Ez5H0x%zzuekw224QItl
zd7dMLAYZDcC5+B9&!j#?{^UTdl$g=3PP{yF?lHOLY9f9ky?nl`nhqCtJ%N()PaPem
z!8L9x_MGf^e?Oea;Mt6)HH2)ZWa+11YM!LTYl3fA-zo{$xq94b-1j%|8s~t)(2e0c
zOw5ms$b$Ir@*-^c3P4TFE{n-7?CBPjS$od4vy&c}aWYU}^93R^R~q=of?*aeceR$_
z$DNZFNphfybl=}n-AJ0$*PCnhqskyI!bnvUr@?02^ApRwzz5P&PU;qE$@DWW7Tzx5
zE{|JDm;tmqJartLRW^n5mD!u+II9BttBK?84@4WCrmx~wv<+@y__|h`tAgJ`m!e)n
zofoRRxJ(omP6ao+j;cpOqHDYvA60t+OLV`bgd4RFHi`wOS8~(ML2M?$%nUGk=*q4O
z-?#_SRyLZRRiY@YDoB82PqJ$+eR?{54UhH0P(LEIDtxijwx*1S{zT6}8uC*CHimge
z%Cx@5z-#e0b4P!%yOQ?q@78xRPRwr3IERfnzf%O#d?WhdWPknF7?|U~#lRfQ{}BVT
zuyC<({7)+UpE)oKD+fFG|7U?<O^2ylseTJN-j;`gU2#zrSLHf~g56e|{gwrJ<$Be6
zTW$UB>&wm7k;l(veITDHpRaLU=jyg^zB2&~iZ?lgu&J(SK}KyHDkv-|E25CNvi$DA
z<mABmwnmol#Ofs0-P$y;q%iR{5StTAdn+=t6UYP5a?ltMashHsD*C|C%nZCFFj8Bq
zt0xmDyDNy&&uBqmaq$o1lpryX=<CnZ*Nht*XDhJMUw>y{osCUxQ2U&}B437-G{C_?
zl&nMpBik@gNNLGyzkj1cD*c+O0vZYC5P}4y4eDtwOzlJ#+T4b+I)W+z&RFLGs`>k`
z{Isc+nd7INU2?2nD<mEWG^iMVe&Wj!XMP6C>NVjPff1xz1Ba*E*Czs#Cs+m#);iaB
zZg4XF!1mb0<i;BzNZ<`c+Yd^UYr{9L@%P#*rrhr8!k>kS!R=Y-yY{M{sP(`*Hx^c>
zkMg$ScSFvB)nmt!#@fV3LHh5jPuRz`D>Ea9BPb`3F79vDDI_Q+4$qcm7N;NGPsWVg
z(D#0}9Q+x3xu0La`~$#NHv2|);BIbTu#cFZ`4?ZB`9VLWzM#Ckw6B59_CIkS`FPX0
zc)=PJu}*X~fwnl@-?jEO4xr;ta~g>a%#Gj;Oh1u3>l?R%{tQBqfskU~b9sgDvO@#w
zt7DV+MsRGwrb(^tAG#<)-+|>UM?buYKZH_WKVtholn;M@P+$9P|8c{IKmBL#{i46P
zb+!NCXo3F7gM6LggZ_s2a{cq=I0&(Ns&V^Me)mIA3Gsy(|3ireO&QuBiqyjVO$)(F
z`WT2cgFiaG6Eya4svuzjgbG^Sulli8XJQKH*1*yZ{8#;PPZ$JCD<ktKTXAE0dt&i4
zztQz864uJ}Yp?PL{oL?SF68p^L1fYG=QHlZG3U$hBsXQ}=C8^pVeZ5_&d=PaAeD$n
zn*)#{1O|cz5GXF|)E?+-IwB|T{{llmyuXL{|Doc)KrH{YmN9U0Fn0rJ{cZbi=0DB<
zkLSNuI{ytu)W*ox_@BL^<YZuD{C9i(Ps2ZMBWDMPzpDOw;{P`NpX+}&IuPgvG=f|C
zV{614WRaGc?o@}JJ2WAp)&GkHd??tiv{AXJz^}%(*EcO}o4Zo~y*8M!X^Q8w-F@-Y
z?qgq)_F&`}mK9ZR3h*+Yz?+PwuK;y}vhy3=K>(9(DPsDWT9oJ2g7g-ICMUD%!Onz&
zqHf7M6ya1Sn*+?%6OA94OMxE-`J)|@k5OtvEM^x<69`&e;prJ=;_=fX)$h^ap~2o8
z$i97ix|~>83SI+jpX5`LBqw)$hx<mOPsmpnj`(s|Spqjgt_iwLMY*7Z2<a2>++z(t
ze=vg_weS2SH9wBa@k@PMJo?C|_PAya;2fzfJwD{VSbh{laNI^P%<nmbA_+bI6lek}
z=hqvZ4V}q_EIjaF82Z;s01uZ_L3l}%oVME6NN{E1QWB==W|PL4W4$6BCTBEY_|_2o
z{ov+%MR%pe#qRae(J!~$ew%Mr1q^f01L*MmlmYKTeKen4;o$o*U*(LBA%2yU(tRpU
zaKX)MV9=F*1NG!I(#IwHh;A~F?RWaK%nw0Zy6^KHTB#~cS|_t@$lB|d4e0=R=7P`V
zM1(WUm{0??j?2cPQ{H!-rXl#X+zIqD;3j1$r|$^mr;j6~2-}EaLDu*1qX@+mVM~Ou
zNY1Mrlpk`e@k+essUEBx%t%omM5MJ1H2zUW$D97fOo9$yzgBo}rMQsj5>0d)W(m1r
zeu>7I*@-DSnMTpmd(4`k`MELIXC-o!c)*^HS|yGZ%(R5Xa%z`}(#b!_37=wEBwmMc
zey1<UVuNM812L-(HW10r(@<lE`V4|0%oxp!4+(aF+3i3Ue;<n78YUuEIxkeK{+|E#
zoYrF5U1DStJ-La`3{mzCo__->6YEv~X)EfFuSoe3#5qxAwutv!q2D}tC<$^Bx+x4I
zE&hjURWn(|&wF0#$FyTNCGor;t6molaT2dnZb3X+TF7#RK>yLzA6${TSWa&jQ#)M9
zi|E1H3_Gqfmef`&jND=qyOh5EN}R>alHCsBWFp}S+x=hFx`wPj@a2yoL~ZY&^U12A
zg#}W7o)Vho2^1KhhoKQz*RfDAbo&S>{t=)T8CVxnACxC3GNMI$dwi(z5r>1`F+p7J
z%j@i9fU^{lac+x;z=$kS$IUw3W~H-EJ6Fh&`Jie)3C4SvdcY7*sI{O)W{+OtBP1gb
zl1_*SX~4s=YLX3y_earzZCdM>LP^l-%heT!_1ZI&R^^J#D&mwP)CydmP$~%i@g84%
zI)9b%>*pFOva8kwsN;u0u*H`ikH0{pAnCjaDORo~b9KnRNziJ-VdqgzzDt+Gp1-Q)
zAG!8yVGhbUD?dIEDQCVOXU~0xOr!#j1cq^OF%(}MqUY9%L^^pCa~#pZ-$DfvUyUIn
zuo)Q)WgpI+$}tjQ#eN%!w?XxkLp$Jq(0N(5bT+BbxwI!yS+GS~J<f`ca0ijEW{|o0
zcPFeVam%3GwCnTRMX{8ZNhQ|Zd(zJ)@Ikf{pZGk0P+jq!LCCsNARmD5wa5*PQHC|P
zI}EeyVm!~bD|#!b@H-ta0R78U&FtcD(+qZ>;jYO~c3APYk~arOw7M=Fg9*~6cb78Q
zHgF@T5SYyyLr3HY%TU>V+*^!z#9m_?z#K46Xq#_?x}Sw?sU(SW9=Dqt&1u*=c<?5+
zgO9-$NP0BfQ%bL$qOzqEkGq%LTj2b$<%k_I^MniIVJFl{uHQl~HWZE>=@#nm(|v6=
zf?cy_BSz1JmrEQHx$2jwxb{3-x24tPCYhnc5O$(gF@VkZv3lE5m058G-PaiyoVP%G
z#Cri9<E+Y?)`Ejp%<`lAAvzK=RvaBVb&?}GFVpZHYo!k&H+XQPWe~YWR%4jvy|EyS
zBV&?*0Gd)5kMHLk(G5c1P46&#IOwT;Uq_6JhZQCe{gtyQx78ZZ=+de^iHE;<Aasd=
zBz4}Nabu?`AfgMTac*RVzH{g@9OT5BLdxr?Mc50#MKegnNpXR0mh$d`%xG|9$D_e7
zn^xKF83h?QA8*3KAyEPPo3|yQt_n^{`Y{M{@?bqFC1}Ejyj4w>wu+ZGIcZs<ZY6WA
z()@!dF+-<^`|!QQRoW4Jf)&u@rwlmehk?DfDWwCzs1*`ccaOv!ty+<bEJBhW`wDVw
zTgoCPm-IF`l<J^@n6YFN4B=9oNoeaX9Z&JEg>-~LbfnS8O)frDX>xz-8ox^*GZwji
zkiXKf#1-P<d0=SN$v{f_ITkd%+H4s8ga4gkl`v>7oU{s`r&nC9+3H|3Lt}T6Yxq8L
zrqjgE;uPT{J2%S8TN9}^e(lrrkKbSq5ynVkZw$?R?z5IKIzb>`qpSiLJ+y>c6IFYl
zm(vuUM43flDpcwFNGJ?BiZH6TzvV`0I;#3ssAvg!HYv!y1mMBZhCRCQ0^<*ns*^N2
za{QzrK{I&*LQey3omX1Xq&7;gs?@JC4^yICxsJ*A=C&aOXxS1rmbm3;?;VH(K4O=5
zw~yD;Kcl2;l8KEp((=SBA|^{sS!6%|fL&vk2k#@+J>rg+Xi75n=otNwIGjax$1G85
z@^!;vg6i?O$wykOmbI3rdxr<ef@fPX7*}2>+8qqqQYr*Bnfg8^qwhdni!$3q&#<aP
z4$%q~oxE`B#LJwt1Me7_J)m|}i!Yl55DlVthX+BYalZ_V{i91LM>k9YW>q`s9N;RB
zBv*YMG&^ljQv7me1$?Y$b}4AjI#WM!bOoP>I4LS*KS&b$^nPzoxTS$I$!1%$IJTZY
zfS*b+hH`dVG0X57&W$f~x-`v_)q>@)ExlD3YKwu7ab+xRVNO?$m5%k}03Y;YR{{ii
z^CGXq^l=ehK9-siL|4?ni3bl`bheH+f1@V2mb8DbB^!A#dW%Lp(b@4@CZOzKz0RKa
zf>6Tes^oQA<tdr#r50sq3`<Ts>wl8rQi#Au^CldBGL-zrg8=Jn*KJ0xgj1c!T^Wug
zKb_&?KL*$deY*tjD?nd2MP#h^GF9S%6?LGO3-QB=3GD~vHY8alm(Dy3(LQE9Arv4Z
zO3P66Eow0tlwHp&Vr$m3GJqBai0}ljXSVc96mhq`Th>*Ge-Rp|?FTiCMrPITN}#E(
zEu>OY)#}NxRxk!}711a8B^qYGyzdb1LfjR}=V9r=W<5@Iu!%}OqMsXtuJ?MEvajFE
zI%XgDgWR|iXrSm5Kp^Gfw^-Rl%K)T@{t&JDwJ7HkQW<_jYoR<E@h1Mo^*zFnel}uE
z#Kh0iM@e(NPXS8>CmQNBuaBF8`+)dW!Ba`9<2Mst^u^wv0jl-=!M#<x*uV{VG?Sa5
zy@$0YHt|P<vC>ihA4$X0_J9I#PDH3=tnEyxm#GO`VK_%`O?C;<T;Akmb8Z&B3iud(
zpIjZ!&QBf{hHFglIZVJzF3QXTpB`ZC+}#T8$6P|*v=1acp;)JwFz5W|yt+y*>(}VE
zRXL@YUVk(MWI~+KkFERQ5HDR`sQlz!HF0;Eu&XEbHD*ceEkJv27<uMy+!SydbV9Ey
zHooiA7Z3B)j0aV2jy>I4Y+kkww3O^p!%-6F3~ka&d9D2Uw|$Q7s*dIp)O8_T-IcJv
z!ti`bz!Z5}Xa1ATj8*Vp8w7*=hopVdM8&p^KN|w)hyE)l;!z!fSu+zKh^=(<_D^-c
z-%qPtWhzJnC4&&nPz-Jh$Yf(c9&8VTwN#(s?tzf?F;f%h0PO7BS;76L%;)nO$`nQ#
zBst88YljY_eUxlnIP1xEp@3F)oYkNux-urGogWfHO$fbGh}f;;iw|Vp1ou6UX{fK=
zRawDHu>QmmyB9x&A0=PWU`Lpq%00@}ID%(Pw?|(zd|}q)&H`w|anRMcpz^auMO~<o
zVx$_$Z?H*7WakN@S?~y1(yv^nt1Q?m*Al$G%Y^p`lbG7E!8%)NLbi!x_tyzG+$2qZ
zBfx_$SM@JWBcqeU9)%}PeSlzzO3|FVKCn#+XsgZO|FqEO#TI+f^MN<bLe$%4M0lwX
zH4GCe7IHL-2%V)id}{%BjHG(Nm?Z?k&8$}W)#a-jDBpG;IP$7dbNpRXI-NDHCJKs3
z_$C`cWrK=1a3WrygX5Nc4x?E2qduA{+nJ&Wif#S)<KUB21@*Is7)yQ)mq?*utzRCa
zB5i(#iUKQNgCo8JYP&EAd`_sVE`zBGrAbCB)j9=D;zeESCr|zooIU4NzZWElwKcZ{
z!TiJ3jm?JS`KAEq09niiLU*a7XKVZ{;k7ZY&Xm=729)($AUW&YZ#xiAkW=e50`F2~
zI(*RLFz!s0nhnvxLHciIY!g|~vzc`fs<~}*6@+WUfxI87ZWQUgO2sE(t~H%<bdVn2
zzA<+yUd!Tk;_+v3Gc$A^yfL&qqM5<01R#k8g%a_b`@%QMf{Jyae<}lmKRC@*U-sY-
z0euAKG^0QE1PQ;--<m#(i6XUXT}kyYDe`x#;YZ|Hr(J5<b(ms;@Ss&YmAGT<s7$%B
ze}7+FNltT`G|VR%e{VXGTtSo_;UxV2ux5gZyPJ=)V#62w1^pb(_gh$b433mgR8W4P
z%h%wGvbwC+VFroYia3v;o~QzA$2Zg=@}-^NjK2?2Q;5KL46ELUZM+kuDROa3+pr3=
z@|OYZGu>{Swh}+@U|P>SXDL63{|y;>=%Cu47SrkuQpTajeRmZ&n-9Lp;i#DRQ(0bS
zf);}2kK|)Vk~|nh#(1BB{Rq)UE-21x+wZsI*7%S`Kl1&?c%E4ph3R#;uW{02DbF1F
zmfRu5zsN@P`=v2_)9c5Y-AS@g{F{teW83BklfJXj>ukiNb~a)0YYOLuWJn-AHm;I}
zXMsiCPJF8|(h4>82MoXLnp*h0K-A&Ncf2B_E;E5+VdoFio?%}GP80SkaYpV9T1&Xi
z`kZOU(rp*9uA1h>SF?c3!z8#767qok{gFPRe!jfl?|zb2qKq;1-rGh#mwz~TA$(Bv
zte><O7RLRo4xX$7?Ns&6Y*s<JiPzM<{eDWn)w9s$HLg-7=hO0qCmHn^^l33;fZPLl
zwhKU?AJ_wu!}^m9^WczNrdZ?uf+MnoGH9i565)Ba?inHyM_ZsN+uE~c4<U(!@XBuo
z-(p8;5Y}KRiFH)}`3{wDQb(BFOt)3+9=LDy%C!UR9Gaz7h+~ZtQM$H_;2^j6FP>cI
zsb%-|xz>An`phXBt2>@md+eF7s{;(*+W70oH!sA!H7qX`nx}_(EX*>j?rT?kEoy<A
zfs{uoH92W-(0X|7D~nlybspCrovODRU<^xHB_B}G_f(hJpDis;FQME@sky+C%&2Kx
zD|Lx_AD8n~D<n~VXm&&am}u_O6K(QicTq#j+2e(x|6mT>yvr^thhqc5=t=NMuQC9J
zl2~q8Q^hEiKd@QsRb+Nl_k|*h#!OrE*_8xJJ{S%f5>{eox54y0YDvfgATj8ZYr|Qp
zSBpTygEOHixr`j3N|Ok#WkI=W8hT(Gldz5PUksLt#Aooe)hfnNx&vQ}#%@g~S2>08
zN4f>s5g#Q@ucgKz#;N$ud&}`RP<>t9{K+$)d?%VK*C8#skNZpI^rStTsM+Glj=&vc
zm@d%}JWJq&>5Mil3tDg|l=`l_K(7@nL|!a%>`|r4Ou+Q}vdTKHu37OPRx;nzkwyH+
zg<HQ-w={8@AzU0ZJO+i9qD}m|Osn~|7?k<QRK>(dOmnque=5{7zIaL;nEu@V-bWVw
z-DlPMoqL)xp$(@nJg*Su7%~R$Y;?YHpx^k*!YD{EW`aPW3*1cUejydQFb36p=jg2A
zDtdXQcr7oBvtH-TR9GGrO$!=jGbwyi-MnzJ9M=?dijS^wDj)2y{woD1HQ<+Zt!gOE
z@KphT7$#8rPke&VRT|iLi46vZTwU)@^ox4XVuE$E5@(svh)b01{V}nk+q)dm%r+(6
zX}w4ZFwqb9JjpaI$O2LP-gYJl&lNU8rY2_zJT-M69^xA}>|m1bBJ|F>l5VekPuIQ^
zCRNa#Ib-yMH`8;r4p$}q)7ml%dX2sEPMhPaR(L`mdpaxQNLQ!P0K|#HDEJ6yHrC*a
zCi|tiz!0OlUq8tF)K_MN{7hl5-Lf}6m*QTl6Z=36Pqsg!cTN3?uU9UxFm<f)@<3sE
z!s@$1V?ivp+JZ73tdv+OtFfaN%dX0@ODw2-;=?1gtp*9T#KrMfdC)AD<$^jVzyFzc
zCK)a)n?ZA_3HDC!L@H_(YOcPQ&H;w>Y-S2d+$^-t)#T(HMu`s45pz(fc6R8eM$?MB
zktEH$dImLujz-w+vx{H#AHb;!%>E`G3R_=17mqZQ$fxL8y~SrE;C=*oB)}l|0gt3n
zctCd$118_XoNK+#R_=OAo!FroA&x@TijtEz4<C1sXG2LZ^^fsUeCD+AO4c)Ojk-0G
zdjOlnD*Ks<F*@1FE<b6-Z7b_T)g{-s@z{Hi?!Ct1q$Zu3sx$qsnRqcNBnp2Xu<s(7
z{FD@4STSTDGP~A}t0HG}aX(hiYZ;k>$?CzB!uLBEQj|%R2s{6|V6CT1^KDd1eg;u_
zvySy$ph_^O_k<9GbbSufMj8Uq_mPJKM9tMFAE50vh)dTHlHBG&%`A>;v$$fQL6DiA
ziT|3x;44ihL%!EY`iIE5@C7PURdm5gnw-CYzOCiX^{j?&(ojdP+woS*($~ePRK?;X
z{RQ8ECdP=`&v~)N^ib@gA*pm)4V-z3YG`Us<B#w%DK-Tj9bIx>xLKN^q+ol?30$F>
z0OisT2Q{FA{;y~ynUf|o>x=y=0FPlV*48>!i<vj#-iY(Yk~g?!6<8BwTtUetg-~1Y
z#v+i4s#5pG!X5!&>Gr4n%wizZa>?O%*EN|0+(L@AQ3Bsv7McZ=@g9<@7cTE+VHDD6
zbCd!>pPjpTc$%UZ!XVGdGb2RQwKIg_aU+?I%|92q-O>+d8VLt#sVTs)`qJjjW>I={
zp9mPI=4?gnU#1O<#bc?ylvl#Pr>_0BB~}INu2jUc%B)gHEammb3nt|cRZA0$BCGXD
z1oqgD9*f|^SsR5)#C-sHyU69;wAHg%>Zf(}&$!v6*I7)IzSffAd|hvI`Rqvp>hk*N
z2hH?emV8o>RAUfSXA)JMuAA%0UpHCNV8P=6UqO7-gMCXJW6gdzVb!HCA;EaoZSioH
z(v9KqoF4SGf_8-W+(FES*Z$5^9w7ieMU{bMTg=<ds@<&!td*+<zozSIePA=+@EWYZ
zfmgrAfqED=-(gN#8?=j#nzk0qyU7>hrM;>-(;~lugPCygjKErupb^Z4$gSSkZM?yq
zt}Iebv^}+-Ht3+!BT+Lz+>O-Q&Pr4FSbz1Kc=@gO7nr~mr@&~QjJ1+-buKau>x4c=
zmjJA$-JX)BOKHnEP|Y~H7&b+nSNQB8*P|d)n0q+=G;nsa8$@O{0bHS^IMXJ{_AHjJ
zpL}*NrXPGF+kRD_P<z1hr_6y?zgA~6|C4qj{c~IC5&=Gnl|MvRbV~&ikBQ7{^wpZN
zGuJPryPrQSXa}uMo{&pg9D}FMqvBIf&f^x&R1PBV=1&OM({}v|2HNzLFIXa&b}lMs
zt(p2MRbMqHp2du*8i$~^s@Xm5_erEBb%$SNg}-NF$E=@NW1x!rY9j;`z}!FIjlJu|
zKC+NjF<H4f^4-1GX)BsIu%d-U5?vLi?(8t1d7mZf$?tRz?$_)cNP^T)(`tTuQuYa7
zwHVbe934gE)XpT*SM5Ow9-q;|zRs&tA>QVHsc?YJ6KX6|4=hRF@Mg17Ew}n6S2^fo
z8a$r)VwOtiY;Bns!Ho({w2Al25a0DzA#14}i!I8W!h)_ERRPPg`&&hirvvr@RP;rk
zepK7Q@XV`N`jUwJ7a;lgC)TcufI}KmCR<wK(oi?GaUTRsXU7jcb5%y%RR`4fezx(I
zB)kJY%s4hd=YyP;6AMgoe$lq#yVp6_bm%#5QoNN#q#zLPI8$j=_(XcPBn;7A?{Yig
zlnL9ZJ>`g8J6l9JP%UzC3X(m>-8p|wmmDW=sBk*|g!4%*F*er21s1ZS{?f=T^Wfh=
z&eP1Ce8TyRnA)&>*o>RgMj&aL39_P-8-%YLGi6(vku5rG{VKAZ>v|KW!2~%2_};Q^
zC$FJzTW}*uz}PpRtD`$ggh1W72{tZS06p)HeP*~Evcj~igP}l+VBg{-NS1%05?uCL
zVV<KHHnG2l03sq+saUJw<`_uo0RHxj7eQ_`BW=cT?X193JKq@P7zCVJ`^B5M=rn%}
zwrWysQEfpcG@c48uOl(xomw1pZm4^Zk*!0iSXyb^TfekA#aMqy3XJOa1W#w$bj+6n
z;P3PAe#VOK5q0Fem(zJ%?NPvb%Y9z;pv%xPJj^*cI#p5%SR12RNiU1U3*GzdBN2A1
zA@LkpM=t3EtbU=KwwQ0Ss0;u;Ad9u1DMSh@dA=+C4ay+-%o9ehxkbzmtQ>n<afJ6<
z9cZ}Svh7tY@RZzW6tN5>OVRM*yFAvw9@fz(H@%oNA%F;Pkj7<YKz}d$o9iqzkZv&t
zgkH6ALx}f;Vi|lC@3E(y{ge%iMX_I-YGByDS>@3x?7-uswJ~AVw1xMh4}}6~oT+3T
zdNOm0ja#)T3`$E|L2R!A2B_m`sp9%m#;diuh)prYC7uygf~OJuw30vkYcvFXGZEws
zke)}A5$p6_ForP2x5^zg{i~M6Z}onZEA^Tg%~)GDGQUMnTt|LkYG3qh_?d!e(zC_M
zY^L>E6s;S31%c6G{S2#7cKg5wVWz1$jPh%=jKA7lD3PS>#fFg>tlDn(qvz64t5v8R
zbhJ#;<*~q@_Ik6WqBKp!y<u`ri5D4U67iTPd^~md@L;jeL#QC-!vPet+%BELLOqt<
zs4?;`cmb}5&loQ&l7Dx|gOuIsMkNs{t^df<zMuAwrtf_=ZGW~4<TX6}J3V)KWEi^M
z)evoR0BP~7DRp+mGXk=1>B^xy&9tD<q`Z%DFy3Og#TaNqT6*FF>t$l8y8ZzYM-!x)
z)t^u*d)cnl`fc+!6*_VWk`+RU04MlWt|OI=J4>#vpZGon^EzxQ16~;(zYjqJi7Tvl
z4TXw-pl*dN?H1B7tA&-+_F5GW?Q7E_YG9^h+&c6tWQw*6I#kB7uFD4r`ZH=OA;6ES
znrMyT2j|{4YcVo95YUzj!nK~A7B!#C;co8P5S~yHrJF@nJ8Zo6k?C+^!NLK`UFao`
zQ_e8=uKa2%xke<$g>Gx&_(&8s^5)3ccaEx{s?2*Q7{wE@%zQ3={d^b{@yggjrxV2_
ztRFotKrI|Q;57xm6NiG-NuQqT^T4p35H}S`j_}E>P3Ru!$oE629r`iCNY!$qykZn9
zYa1jsm@d~zm(AZYd&P#0-}|yo>XthTP}9nIR2IJxiyqQ&Kn8fHGp<IsC}CF;A*$X&
z1ro(N#5PMW@6C-gqh=@@072ka<R;uEEZaRjlqE4NHj4^4WGy(|m7B^9rHFUHHMMCU
z0=>%FGdOCTx_yI4(8Ch$1KV5sZM8){d@V~2)UaMt$$Ce<v&KiWS*a8kQYS$NYR=IA
zP@GqLy_an^arZ+(kYrWEID#`WE$E`s2M-=4<fzo9+?WYMnvyj^pxc_`?<iZgu*-@`
z#8?S7(R>I(deQ*St+jOW#%^+xe}JeslZrxs!WPc0M15D%9Yc>`Op+v`1heLEFTV1*
zcm?@tciZd9x6c_%(N|HYXJpa(Qwd^^a&Bm$mt{;OX_LUhh8zlb+8yN6had=%YA1v&
zxaaP9mGcc$oRe6B`;3)R?t%aG*x=K&9WanSe%H(fa?~KKiaP!-zw@p01HofFyY=0L
zh~s})rH{L^n?rLOz*7SYMazksT9SvW_d}7G6}|SFQ_FGywN$exefzabV6S|GDN)Jl
ze4I8h#Lh%Aq8x<5{^JgP)#hDgz80lL96GARrY+_)iYMTb|Iln$OxJocktbU4$R%%Q
zGqgG#4m8(C+1BFIaP4SvrL|micvFwEGWHj4m8brVExrg%YI7=Rq+3PpfPi2L-;Jx}
zso0)e9%X*t>ubYtYef+Rb;7TY-vt$U?`Jd{*Y!d_BO9H?#Zz{NyGOFw99xMtI*j&o
zd=D?orp`4cx~=c~At6zX?a@?l<6V3|)<{9<?jZxyt#PX~IYiZvNZliuvAGWj7ZjGE
z@o$==g22(MlJxUI^}9BbJ6hs9X{PZ8)dr9qq8kEp;L?+nN7R#oMc%$^?l(05I_-b+
z9L>%xU=_iou08y<3W%MNxHA>d6LqRBARR){Rv4S-FWxh5@GH*$F2k&&ecr+-y37A8
zI<2T<@0cW$xmFq8feX_SFrIQF3)4PnohsP#E^*?rw#I;7`o*qWbHJ)Mw2O(eWoiW_
z@B~FgI)+4Gwe}}x>FD^-f3bTN_~P-~xx;R<K}DLdcy*hN6{xa`M`{Nbs$rnEZWR`C
zlq;^w@hla}WH|gQx3Z%(@8ktGkqdVf-9%>DeUd}PsX)mR9(!0x^Y@&>l@KcY<f&Hy
z1A5gR5?*e$BLmUQ_%zoS3k5YFv83Fz#>qfG-#xo6iK^14#I^+z%-bzt_Cz@ilEIV^
z6PFyF@_L^{5NA8m1BLd+T=nTCx^H*imptprcrTVzsPD^1gvScphd`c^hGEa{bjE9o
zk7Q}9CTXB1aW(SeH6q*#1>qR%tiw;fR^DH|N-Ku)9e?5iPmjT0KBY_cKG%7>BW6Oj
zPnLuDqCyRG2#cr1Y(FbGbh-~!7$zP@A{*dc3{AbgDhn1J7B8*@BF!=_pbfE`f97D3
z1&_#u93k?K>rTP+cIEx5{>|Z@t?W5hH-nv#a@^N>_IAvbc^XIGs$(82O6S?@5L$Na
zKP6uFP$edl--=gMpX>L{?>7~a>_LHD0ITVGtzDe(03%%=^#?S(U<idh;d{tMIQU5h
zzM4t*Luedl-8xa=MqYEX{Avo<ZPt9ck#sUv8Xk2BUlj9NhRZhsH4l$cV!$qd#WT^h
zxTy1TL|n{GJv?HuvO+bW0t>9LDwj-Teua$3c;iq}CIc3F#~0L+SpCW@I$QFjQs-El
zv8lp87{M)FRh^$V=L$xIyA&t~dx6<IPB^FpD%%L>)1@JE`A1azT#Nxv!IYJL<$^ND
zPl|L|_yB`oN@_0Ar2Lh@<M#9lOJI==?kx`Xfo88IGn|11tf~ZhYbnt7NXShKg`bu~
z$fW3qV2nK;8!y7ATdplFPcg5)MB#DtG(bU&=Ab}ZZnI$Er?nTu%<zZs3|~~kRRcMS
z=al6sak`AbdCc4um=G9Y@L)G<t-=jyLHa5P$z3cQpUY%_J@@@Vn)yPvd15|%ws6lK
zeU!bcpi7y%VV5B+5e4~;81)eXY+uVpiLR%k?(PC~sJzmT)lZ3>Vf~4F%w<m6u1jM(
zUGQe@hzV;(gJa)qN=E%QhNcI^J6D~d^_rhe?d36rGzhT@sa(c5Y3|Dg6zE1YoKwB)
z)3`KVei4J{BOTtCT^&7w?qq55Fo%Sy7cbrg{iUcv1xkjp-MPd@!K`Dj>2|)0^@iMJ
zI=IluQn8wXZpV3Bi9b;?gH)$y<`%Dg_<p^4N|bBQ<)2it`m`QGjG~=-SNVKcT0gCH
z*g1zo8T^5F`9N|K;RCPhp`KF5<fD`eN7E9Df@F3{ca#>TwdB>kRdNkXjBbPHh8!l&
zeHJ<)2-s&Tk`+Z|Y)}vtc@e|_{e1Gmcr|V}QTqB*KqGP96tthiSJSEb-6`Zcb(+$h
z7Xx?YpsD!{g{IH;xHRKeNaHJU?n729onJ!;TF=@@i&6PE%do)(R%)e_Qx?p|_k6#F
ztL{cqmCI)z46mCW?j4w%h2pR-$5otoIQ-uei@8-m4=obo74N385sP}*hZ-m<&Z6LO
z*hKd(U-147St*-fLS0@c!~4k>`L&bG6?W{>P<3SsEKSk77U`mJ$pMQUhb83DENV|j
zd9hl<%D9UQ1wurJf-PJMY!FI$kBxRpe8(lixp`eJW>-%INs{vtIy&iz)3V`I(N)IP
zo?E=VJ{<nBv9)!Qh<!ULcpJ`=ju)>@<GCXdIOElkgC8hH)#~RLW>H*=l>yEA@!=I0
zbLPePN)Z>e_ak!RFnK2>wWH>VtU&Dr@}+q^^$Vo9w~Q<-%}tgqvLhC+-VMgHxJ!Q8
z)@R#ke!y@+Qgf`HPCxOdBz&tpf`LSjl7z&aoy1;E<I~1mh`Co&Fe`(@!Gb^|&*6rE
z6t&pBRcTbXE6A@b68W!Op_C*ThfmQEZ#Fbs+Y}`cLY;J~0yi|SggPQabF2)G2hlxn
z*0HcC90jkOu3AIWM+t0V;2k;hZ=J0-MV09)YYemAs$N;u4AgkAuGToRuizKIIJFgK
zzA0Pil3Lzkzy%s}%CLLy<mXq5+}F1lvaK?+D*<V_76-65NA1C+U#18(UN@t9;Gh=B
zRB+n60{Pe&f=Yz6rWNne=&FPcU4~Iky4!*`mZ)7C-iVM=#PlG8-%4}6#>B&9S`CdO
z<)2;S?+Me5jj}4OAo<Cn>-ml~;5^T2sqQXcufV@I=%WQDQ0UFaUZBxy+HdD9H$Giy
zp`tySw7mdPgVYv(A44@PUx<OyvRmNFP(SApf<H7Ew}-cx-ErKil8b!sNylE{8z#W0
zdzY5;y1>_AnD8;Hj*LR^u%C@Iv5wAUo*cfvzTkA)%O&3tFqsh=VP)ZBjwY{o$7Bic
z_830C#1erNDiOWyyzVA?J3%MHOz`AQOL;Qvg6!1;c|t{ycrxOL6j>#n2-abtrH`4r
zc(Cl1?k=bmj1xB-DO`FW1y6`I*<y@Y#nUOdV7Y3!<HZKQ&#7e2D~gEU*-0su5FwwQ
z9vIzqN}9~##HTPH?@%I}q_2pU><HR0P^pi~avJ-m)xd3UE3G7wOwXFzy{0(52p^q$
zjB0KobosW~jEdKa7;^O@P-;g@gOX1n6Zaj}aF6z#dvW~X$BSv`;E0BovbQVb!-UL+
z#j~glpKaQ*A7MgbsiFS_QOM$8RbAORJ)k?GV#D(o4YoWFARyf?94($#ZGc(+``1hz
zM`8fwdklz*4x$8%Q7C1T&(f?S>1bA<w8k8K$_|Nk-nMJ8d&8wmDoPyW+rrv+G6@O9
zd%<yRrL0=(c6{fi>ty>=V->8A`b+LMAZGtaM6#y7@;ZtVm}@Ow?Kpr+{-jxvBYtih
zN>-iqSl?sXPeDVOMGk{jR^L(r?;V`Ov*YOECAbjpy&OB>1HoEG$}<^fItgK)_uKD3
ztlA;E6LO1ErG;%Eb@$qKg~h7qD7eI4-Yr*t8oL*n=g%s6Cy^RL#h>IYMg|~n)P0dq
z5}5%M);gB*r=Zh`-ou&qQ_@-$Nl{@J!!*SyW$-qndcmr#gfOxb@|EHxC=LTM;@Mo=
z=dGN^>iGF8OlFX%&7VK2jv~z9g48KO4v%Y!_!mW;jDc{zL}5=Ahv!GKbU3ewnw}NB
z$dH%D{6}|;Pp_`YYN1a>9b;Ks#@Lq9bjAq*Ai-(Y2s)z90r`3=^txdsm_WCL53H*Y
zeG3?Ab8xA7W}=J1{ng(d9?~ll_aYxFy)fgB^T`=%v}jU9t3ukk8)^(EHNOPTK9ltp
zpJ%+;vB-Ixm&F(s{a2%wY?!$ZQ|@XhOP009xS{MOoM#kKL=Q+#i)9)kelfetN>iHf
zg}7?QpC3AfAg2SVByD=GAUE{vNoa0jkp`LB8Qs8&d!-XNc4KuCKb+8*da!>nY8)u^
zn&7@}JyWmzwDgU@=z<NY<boFa%GKiGMz9O&khMW{t6ejvZFXd45gueA_H2Gnk>F-c
z4{(l&2HvV#gZHqQ!b5FiF`FX(jy@Ash$JWY0)Vk7U|&r{?xr{Lb|>(>Gr=85Lui09
z0lpjAs0rm*F3G(csugTc;i7yAmbSvNvA6nez>8G6W$k{Kvw}FH3I$-PVni|^z>P0{
zR44sx!Rjlhf{21RM<*`BinJkkEyBMpz4fN08etcf4A;-%J6IOEgXKURVnXAf>z8pR
zOFf)Be-F6d3MWnU`8g`@M?cJ>LK;u-KB9oF-8^f}u4(v&kCX)SF3Vc`bXl~rfu{^U
z^s0Em$`b$7)(can!l{H9`gB*Sy}ytl>J}CWSx1Oq&o$4WR_dCyZ@kY|C>Es4@=elr
zv6{on;&m0r;o*MRo(Fsrx)!_)FR|RM#2*gGS7O$b(+h?eE^jqU73;UTXc_vmq<L@f
z`05)Ogkm&Un|IAyauvymi}2tVYl@iEfoDhgO3y)ZDlsEZ_psG>K09re(Y6gmyHcu>
zhSgf1GDXoX`q$}C&MqB)q#iVyd&dCqX6Hul*_FD5d6iwJwN`P6w}Xhh*HG2M@)q(g
z_^;(+rjAnmd3`xdi^;&K`uVN}Ew|kJ0WQ8i=Ay~98y;RwwFNTFmIzhMNLDI;#Pp`X
zok7`DkFK+fC~baS1<>-u1IW53LxGOV9OR9vpXHNw;#zeWB;rWlm5=6u2x(x$LI{D1
zrDnt?{g5%6UwsJ*{>FQts96kxvOm!pex!=Vw%l58H@}b->Tr{A9R>NEbGANv2NWZG
z112$%kyz@eNc>*G^-F|=mFg=Uwt4chagBq8dnks+9*87c8dLt!YzW6w3TCHDmCeO}
zLOfmqqU&Zr35J;QyA0u^;wTsDx1a?{*DHF)qVi^<QHAN{o2a&+6XGviB@sH~a$dNV
zq2H*CV{thqc^J9%VXgKQYC^8i?mkFipgzL`!`CAAHnjaoU;)#4huS-Mbu^?PE!H{i
zTXJtVx$_WKhr+^x&~6uNx^N`OzxG+Yj1dms8{OBpkk@_XOm&Y`S@g;r^N!uAMuS+z
zb}!}X<(wQDuHQ#6lD?3#`l|)cR1|;bYY-+rPS?_wo|fKzmv}nloFq|y)lSrRN0RV2
zB}7Wfc?-V>ldab7<=6ct^*&*@a^P3b&cNs0VPGjx=Y-e4?|L^~N?Wf~G}fX~-2!5B
z?|v7OcmunxVW?w3{7Zdc?IeSLLEOFo<)CU|J*4$V2UF})f5CX-SZppvDGQglL{lJf
z((m2!J^5lm9Rg|&s)eL2%CZi(P)xc!{@Qy~&dGr!YFvQSw-p9CzCfcdKy#i8?Wk~@
zQ^%$XLusM3t3?R8HZJ^YJ;UDD)s(5}ce3}(1{3db(bDmBRq&Pp+|5nK;qRu&)n0K>
z_&Kt$W~C|x;@abS7U`a3)A))3OjCL3zuiri5MX;yrKd<8Bpia8a%bub9Q!UxTHR{X
zwflDszb5Z5zoqO-?1LJ3s%AO1Qw9a4%31f7#2D$SnE`km%{Mn66_tO+##c*h9OH&`
zJl9r`6Lt$7y_2i=0LKi#?)$J1&VD~1!zs_A2x~<fY<2f&(4s*X!Pwla&=8-P@W=s3
zSEngg2Mfuy;2@svz7h1@am3%25hb|=&F*dBg}dZJ;i<V25+xOwV{>z@lQY4fiO?MY
z5q&+!)+1QEK6}@&@{Ad^hjKJ2N{rRgqBR)Hu6lJDcg%vm&~0hy;!^O+gs7yvMLF;K
zF!U!WDL0QJdw0g(bPY$oOn=rAs~ZS4C`!hZn)y@342nuTfNmmAgKbVPkgnFNR&F_t
zmu_<4I6n}PTADCgf7ZOu<Mzo;_%E9XEV(DB`17fT+}|=4D;uLcsA;Lr7X{VofYqE5
z=CSabIO>1>H06tUKKJJFi@7=r<$y2J8xpW?lFLqqL=>5{7Qc-GOx0aiCUSO>f%Z4X
zt}sWg<KeRhlxL43)d41oooQ<mdOL<9!d^{&=U~L0gHGmx^cO6?JPyuJ@(}q*w#|`1
zH`-6>w&sadFXNhbdfGUDB8(C0zC31jTkCMc2)7R9()GQf;^NUbg$hBIZj@X6v|bvg
z@U0B^-Nc%s{RQ_JQs&vjM7V$A*KcSzWbZ?rSz*^Q2FGT>6R4Gzzfhxdq3B(o&E=TP
zJvR_jQbywQv!Sk|Q6n=39(URjNoJg3w|ksek%WU-$O@Pqs77Lg?7H!`l;t#Fv6|fb
zwKmZ{DtQnV;UTx8A)<sf33Fl8An_e}iFM%B1UheNoi8s)u5Yv*B0R_@CAf@)_jzxc
z1?m_(x5OX<c2#YVoN02IHz>wG-ubr$dZVz#DPqt88_JMh*RMXlQugYgu@TD-U0nqw
zW=4~#`Tgf&BAUWggTwq>Cd0eayp}Ev99)wSV_<dm2SZ7-;#<Ny9Tb7TD!U@l%2gd(
z4#rOk7t7yrD~8qy?pAkNV0)2oEit{(UWKf000w>u_ldMnjoK@rX2#|5J0MlvLpeOf
zPIE<|KD5G{hZoR(Smi7|7<fg~RKSpvG0O}&eM2*AfIUW;l+Ay5HY2@BNHio!AJdLe
zYrI*9wuh_=I^7ACg;EQ`>f53MIJsdC(Bt`Z(RZoORSp%J1u&OJk|Neu0C@(FX@2Hk
zv6=RCEwGCHL72jcI{-}GM7xf%UpvNqBCdU_<^90dx0=qUn;6c4pd3KIk3k9BIF_@l
zlV6OD@D25&7cC;jAfSCItaukJL&l5>Q~DntkP(rDjmg{mXQIuX65w###9lT5cbwdz
zY6mjosFP=!gY$X}y!U2F&fzq(RP(E&t#x<o5qR#yfv?U#ovlu8E#;6Dvu>ZZt)!iK
z-gbULn5UzuTuOj=PX{R-A#G!9*_ark+(YGmTs&86kvyQ6tClu64!z*(opGR?A&0t_
z)Q60Gblx{#*w^Ie>UGcFV;2M-GO|yp2b<30>(I?2ILHtZ_n7S-7_rxXd$RbUlwm_C
zq9qryd~MUC?x2<gzd%}<@xp2n;@kV61#N@rO4WIbN$BS<=qvBLtC=9t3k=1Ilq<bI
zkqz9);GA^*o-9^_4F#7C2wac}H9S<;Vp-C~sUlCj$^w7R0W&uO*X;Q8M#tmvxYkAs
z5w!E3>nSlje&JXhp=aO%?dfUfjJ#}1iEQFr$YXx*=qg<pT|pKU!5PN*Q<wL7;*s^e
zpZMLk^(lAfo3HY1itUN`+P#!TjmKtgMwL2Y-?c!;E%}=^axp!8_=k6%P+x7xVfAPh
z9k=|XaT8})2CY?=&UOV&3!x-g!Tx;vNkK&nSvu(43(>4$m|Vw|c+WSG4ppw>y?gZ-
zCOnFHUSY_}TfJ~mT^rUv8^xVL2%i}>eZB6urCo{0;!KIV;=5TZ!2r;1Myvz8ZI#`m
z<0V}FxGsc7h%FhjjAZE~39OVAR&VNp@Q1fN*vUWnsFU7~nQb$ag`7B<EVXv*nCf;;
zzGJBJU&gJtD)5bIUz7s|J?MI}D43}2BH%8{P67FyPR>8zeo&o?A{1;RqpOcksk#o#
zQm2tLY3-zUDQZGgTcNwB8Mar3?_odniR8#x!A3>mS%OWxS6)n!PzGu0=DQGW$SP!w
zpcg9F*@*4NG2WsVC%qxH4y5Ftdy*bbl(}|U^;;YXyBL?@7XUi~BPfWtqFhHBLxIv4
z6X~%Gy=(_Yz6~OY34SA#1*F_$c4^4RzZ0mH%2_~MSD+zP<=b*M3JGNhujX2)&}7Wi
zp4b;3U}_WTe@5Phq_;H<25wC=_?CL?=BixYFRs$K>VmN9mUhw<zPFn8yaU@QPDOs&
zi07T#FJR?NtXSm_r_Ji%f;@|km~cH*z@+)xR{0^PBeygtbxH^Y_Op#R$t_y-YxTif
z{gB5o2vZ)mwaK-&1;=G}-NU4CLqiP~F@gY67w*RV^r>n&PnC2zFt*tM6Sc!h=kWR9
zo&F8rTanD5`<;OX<GI~-1MS%?Mz)27$5mLde#EZ_>fartdZ1KmAA|5%Wx<$!2xBIK
z>P^bm|GQo~=11U<P$li@X6q#ps!@z<Q}Ik5N#9@F0bKU3wgHuygt7@$tS6YITx1I>
zZ_Sdt2(!!sSiI=4ep5zj)ue*ud~<)`M>|hA!^f7kmT0n1zD7aFxfq|XgYYetoL>SC
zfmWZ^w0LIXs~MDo%fyTgxq-=IY22h%2DdMJdK^z@U`AK0ELZnx(~=EtB&8v<#*9&k
z9*T{#%2r<$8m*$irYv)f?7j;8n0OY65#L)^xX~$a51quY&D2v)dUJ&J<24Cd4;$*!
z&WJ>eAbGE!g|_mLp?VlG3t&7AJ}gzA?`95D*wW8f<L`X|2OoW(dHLq{(KtqqKSOa(
zS7emV?+~a{E!J9`y5>%dqL{09X9&3HtT{sIR}Qood3M8mG34k#6#4_2nz9(r9)v>Q
z!L?737YCkgIT@sew(7hw!)>Np`HZ`!lcoZNZPEq4S;<R$DdNPA=`K#3<AUun@{$5B
z^*4Y&+;=XVQeS0;9gs0+_~o`|qJ0c-(=ycSSBG-8zFCeXPl2U7%^D#%<`Z#pz34!i
zg}IWj`!tnOLAMvE$w5m(`y_|dlk+iK$Ylogj!ha!6o5uWW-M}2Nd&d>&*p*Wh8+Kl
z1AmrK*f7WpB}U}ACQ>ohR_$$f5U&31&mTkwvon7>k}`S#7u<IUP!q*HB>kpTZ_^IZ
z`)d1$mH2uCUhzv?YERd&hS48Ql`hG#q)dNq!q(yC2t$>tsrp&~C8)}>(I8VmK}Bhe
z-~RfPk|Z%lWfz%y)n;qUZw>$9h?lJ((^JF^P#68NlWXNTKV7^3+%E{Vy=3_!D#?td
z#MsEHldUK)tz|`tNoLB#tV^?r9;pH%iyBt75r33gwuB%s?GHT421V6CyJ1i!m!0-7
zqVg=&N}PIKJt0xkzDKMl(SNo(FPm?>rnR!omHWNqYsK&!6N!F6p9ov=dX=)qwKrgt
z!uZ+&fyDBG{kS^~tBD{S#BnmDAIfJ**akz^5~do<24|(Ks*9cX@-iFbjY+xMoKS`(
zrQGZ0L?>2o4%%}DKl=TQQ^VD$>j<$gR=sc)C(sgGHd^4&yXIRqcE7@!b7Y#y5Nll!
znFgPCf2b-xW)pL*flDpbn+!H@NfaMwYJDI4juV4m2KCZrSl{7q(<tVsLQ=3b^c3#O
zOpVL2VbW{|1IK4*c|H`kv0*)y7AMl(XL)8|%#{xM?u*5FD72luExL@aU%ItVZu>0e
z4@T?5j=<qRl7<vvkEZ?T@a98lI&xd+RbE#g#k-G~FOVxNE&|0ud%9eFN5S|<7|E<%
zF7WoHKXl4z7R&%4jA>Q$kAe1iQk3Kg+RzW76EWTfTuAsj<z?>Z9g}cI3^zdLikhXd
z`yw6V&Vu*7)XoXdw3k{hRHxa!0?KZsx;pBMIi{IfG483L`fb~wVqK8ip7ZY?KblgY
zN}O|YjE3wYM9lf0edycYI&I?&rK6xAa5iRWJ;!Qv{J44NqlpS;Iw=b~a)ZZF0UK_&
zoAF5|V{N%|lGb_VBgvQGejIcCB%?J0g>T~Q9hO}H>AVyEZd|)7*c63~$;ic&@f}V`
z$x8}|F-69<;<k7wqG$~RpQ^c!Z+^IV_46+JV<rG@DZ*qC8Xd0Zz%=lhWx8Ww*|sXx
zGh@Q$Ncm;Ud5N%+6{B3rRk*?ntR@!42kJ^g1f3re!|=O)3WLF}GkI4en%JsC-?~B-
zo1)-+(tez=e8iAF%AGh`FGL^IPu0tjMJuW3cg1f)*T}J``4n{_Iu?+LW^92Y)3gUz
z>&BkRRUYkBaYOvc=`@j#{QWbxAb}pM1BSyug0*pay>M3CG?dWHvt@*!VO0P!p*Zi_
z4_kWQeL-5H)I1m2B+;K`z4Q+r6j97!DxjJzH|6wP_AV5Jx%Lzw=;&qTg`GoNQrb}y
z6s$dbMd=5(%1k`|$?~4MC(b>*;yPHghqLd)MSO#jH9vVs!;e#$r)vMFp(Lc;8qHRh
zm>2d};X?T6flpo?$vYUl^RIT=)(f_LrUXi6Hw0XC5XXVTy1~4<bW@Ok%<N8R>HHTZ
z9r<bYU*PG?wVcS`Z$#t4WR<&b#@2Ir7wMee_eK_H<UaCwKMRO>1X)up-l38EGcU7g
z&C>i~8G43PfA~Toe)I3DdVl?K#)B+^i!+ZYG2p6~5cAR9HJ4KwKHY|VTPdFyp8eT2
z$BP_I8G0O`Yhrf}js|MZk`BhPTFy;eM+XP6pXqJ9`lk3^Ly=KidEy_fV}(aS>!lck
zbC!7cilX(Z##NdfY4@ageVV|$OaU7{sKdW3dY+2lin^hNz+~@!2)%ndC`wP2tI%aj
z#BW{aK)%54fZ;nM6G?iBn7d>O0cDE5_b8)ZmK!`h16TN#mgm)rKci(9T7d#as4r{`
z+U=uLxaV!BBEM3`FKT*VRz4saknWBK%Y_#{Rx1eCXGSAb@ZdV_PPSCV*<gRME!_Qr
zI$z%U=epWS4K+z?d|<K;R;33slz^POLVq6V4+Fg9A!o8=%;j+b<#j-m)ms%EdlI$8
zGp>lJ{Km-JB@6VWiuQ(_<7=jJcxy7w#y*<$oueOJZ?1tsoEcP2RIw+8>{K0ZmG%#p
z3K~IxzB*sT5FyVAy=UxPhRvqfTqz*~-4Xw|7`4coyDQ=#V%{U97ZcK^S6aQNV{#C0
z2H<61cv^DaKhkhscNr!cSxSSCAqWJ8PNTrw*cEp$>m4=(n((BCeDOdr&~`T?@%Ecd
zeOdXFEyzgd`qT=Hl2B}(O(Ko`%83RN0<OJ#Az>V>;~87W=n!pi-7D|1>SYLe6qnaa
zOyRfuih2g75>A7Mq6j-1TIRADxgTsf#rY~mGU>*eQ>T?!%<OGplvP(3+_6~ODgI$r
zw!E#AGz#R+y-)~?;03ui=MIv-2>4tO$57?ge6VG2L(9jQAJ2+&u0O4<TIMq+hV3!a
zLlC{y)r}25L}TFf0XBZhCEYlO*#Pu9JMSaecL~RCj4%9i5)mGiM>pd8EP+%VTO_24
zF5YvLl#yrVD3@n{QUmWSjLdWkOo<ip3%AJ*a~E1KH3f)B$ghMq1~o$KwlkC#U!O+^
zdsnzuicHFKE&|3Pqm*{D|1WDBl;ryufAD#ftF}!>k}8J0eMd#5`%r$(K>K+Zv&{s;
z7)*Ij4c(-b>#snawFM?dj1mT2MnPeKmH!JIg3e-I4`=RZW@Rq%YST)9DJ@IaTMSgB
z!9(nrv_dHkv6KrIyz(`>O58s!c*>(x1P50T1mQk!7fcz*{y*y_=ul5C0GQpD^^itB
z__xps0&z|nL~|Fp1bPBRXJVXgc74tewX!6LqD0s;5XZ!}60!nbEUR_uciW7Nfw8dr
zE^!FDqEq6(Dk<uM6RC&S(+ZS$|N9e~{bW3c5E^J&?)Z#xSKJj(&~>yuY?KLq)!!j-
zL&hMa^NR@hCAaYD&6}5g4{ma&7@=#XdK4NagEYi{6XPvN)w3H1CRs0COU`HaKBt%T
zkU>BFVZ0c+T~<PJ{Q~q{>!qo@cdJfKzOn3;3%edOpn0RhhEUb8jKCB7)3Ja^!VSxr
z{VM>)6hTLR-JM2Vp_Jd6XuB!Yonn`fy9nSF)MV$P&?N*N<8+SvTJUE_^7#u`_nn80
z;!77kR4nHNB4O)QUdA<w7Ahli-YqbuvI3T(8~`(9X#fu|JdfIPhP?b9sg4F)fzO9|
zSC#3had+~XSYKx?B~l4rTx-)#p4`^^EX~QcURjzm5ZSn$>T%ecs6AT2|4>nPFeVFD
zQk@KTqjajy<6JeX=E5tCjS51F)@*tXm3j8uZg5ahlrGZH;e2eOCm{XnmyDf5bY@YT
zaAVuHZL4A@l~ioowr$(CZ6_7mwr#v!|G!rE>cKbY;ho)eN9R6gKYJ^Wmur>L5;>ao
z=<PS+kqppAz&H^)V3*?JjQy+DmN*L-c|x_Ujb-|z=Rsdt6(os87TJb#5s1|EW+|%R
zr9%4+Eh?A1EDQ()7VBn7kPRYzr0tdT{e{Dfb9>r*g9lkxAIUguE&hg*az~>^+<}@x
zGP(ad2!RxE`1`?TC>D?YfojbW+z{-8p*2O9kcwXr@epKA;Wzji8`!O>6Btp(!`{<x
zQldf8FCK!RXvvLJ59Rpe7<&0MQ?YL;bB#TN_oZJWj}xmHbqr{1dayL_eo!$m8pJWe
z8=g8NuN9VgDIl}cp*%k6p_F|tzP>dpla7tcGIY#2vAD}n=#v7OOVn^owe|oe{1RiA
z*?%w})jDx8Y1)KvZ%Pk4afeS8Xn`>$){!TVKOBFw=Tg@^yRYQa2=~{^qQFI5iMjz<
zn%$6%dQFJQ`o5h&j5`y}Bi*Ozdc&SUU$`$Ze?C|}5;vLj@?uu1%X(1bI?hIb0!f||
zaAiBBBYI-md+|f5+9AzR;9p&M(SS$0qvNdCN2EHm=+h{TN}bcQR6x8yeg>w<HIB!-
zuKrL;U7lR3HkkzCpDA_6v|!Gg4OI7PYyR6>?^W610aL?qd|y~p*BmKotv;lUr%>~?
zcc}GIYrkvg{!2WK3$cQgu`K1)!IYUSLxVj~oxCKqFN6nHN|$#YTFKw`x2^u0g9|3j
z(U|YP%JYWPC(yCC@F-d99wm3rF>G&_Y7vw)w`|3YgHO9Yr_nppRDs=e@kMsCG7s*P
zoR@eu=bz2uoM!c}m_Vh^B#zhSYkMGO&oq6TaDaD7(okz{%%?|I-}xocVv??dJpp?-
zjDbApa@@+J+$E`t_lAWWA}^2H0$qz-ZGD8s${GG&Uzh{I*eS^wM=ER1C_8RRzS0B)
z6p0a|Nf>`-YIgcdNlWD!x~J=$%`d6Gb>TqYi3})hvHY>UZEn|r3aKuHl8;|KSvbZ$
zam6#1WAOTeA5B3Nz|^t0dLW}v%@hJ)t`BD`BDMPzJ1@Y_ZN>>F;)yI$R{LKNXC9?g
z<A|<jr`SDf!Qi3Gpl+M7d~tQMU!*%1V{9U4faO`mz%(j(Kxt?}Yln8L4Z8h1F`o)O
zk)RQ3@JIuj+v<K8wHJ{=hh~G4N5Nh+pH~k8m#UzFw+0iC+Rd><|3G~_Es)5Y|Jc-<
zE;a)?geO-(Z~hqTWZPet;Xq>&dv25XEo+I;ZTEaUR}Qqo*Gf^~MD})TY%Ns&9yj_l
zwoudbA>}-Z2=uEikHb~!MNz1Gx}N+^BE+7~Zu$XT*1$^;#>bzzbcYx&&8ii-p7)Bq
zGNR_Snp%H_6Q5O)@@xAPcxsknJ^!D1y^FwjNfeYwKM@GewiHOIYwcC-OjlvXS}M)a
z#yiySq`HYuIXgu+hvTWE;ZyEP?^1kCennP$w07HYeCX=2(D9ub%)jEa!vnl!v%?Qp
z$v)sG!6cD4yd7)%3_dxS#lU2m^f(0&PE3Q88L2=Xh(6lYCEI)>S7nZ2Eupb&uU2|~
zPVPJ-{$*oNJ+bPt2q{^$V_hu`iE|LWVL(hZj^IT=2kPH^yYF6a3S5CTxddjM3|*C(
zSd*!80}<cD@|cVMst5)`j$d=46zb0&Ldemmtq$zfmYB3s|4M@UVQhl$Ij_=w)9W7c
z5srHHG^c3KV|SeEw~gZd*j1IQSdWi;&GjV=e0lq*Eva$c<8W+qJ7%L=-G6r*V5j}p
zOBSmM3~IxSw!=`zVny%0%;%PPB33Rqszbj_T+Px@LnNnob@}iqD(?~p<F6!&9^NcF
z0(VQ<k!;l2t<!wu9foNfK0dCR2jl6u_v7Yz#WB5~jCGy=O_%GRAM#=k=W)iBifdzd
z=Fs|jfE8@L0J0t?7x!IFNU+LtPE8WE3x@uLSA5k%5@qhe)^wE}!n(c@4_jDJF;_u@
zoaiWTt5^<x6v3E<=az9}UEz{;M<O@Wt8`4*D(|}6&ugdYk@-!a?6&+mu{a-0N3d4l
zzB%h)V}HfV!%H1dmAEXp-zJ4{4!$|!r*8jzu(nl{^#1Ajr9?GO?7!jB(~0`vpFE1d
z-;Bq`go5INA`mb@9!QDIJ8KS!@AO?7gbv4c(ZQ!+2W;G(mcN)Une#XbBq;x-VxHNV
zVut_N*7(poJoekG-sPDoz>q<l^uKUSYe?<0i#c`PDa{|>PZWJ4&*sX`_SV!aVibMU
z+B>lYZ~h?TX_l%B*MrRIw8t$$`tadS?70^6uT{mu2qv`BlZyq*`?Q4Y${&}c#t7Db
z1MXgobb&|Gn02LO>u>spG%wbyAGFyn5A2SyH0@%ejy0s)^(ZJ{Q`2-GwlVc7+RJ2%
zObTm7>a%@8Q|omvWlMVi^#PIF5>v5pi<?~I44iD;W*sBLgyM~-!PbGR-SqGi(dS<K
zxu#6X#(G6-*k0Ab&(Dj#d`hqy^G~yzBHBbn3hq3@o3ER1#WoH`82z-Pt!w1zqxO-(
zT#T`KQ&XFHEcR(?V;H`*qDD@c9cc@n=c8MK!M_{)iPT8M-NlJ*E8{U|5Kc1QB9!)<
zIUzw;jT$)f(s$nMnt%M%YJ8h7jo;nRL=0aP9(&q}&nWp-?EVN3kdGPSf;bAnI!gOI
zz7KEv<I76mmAgM%4$T*(Kh+!4jr}3hohlrCp8-iTM4mXU1$G8iE5vlruWUEZMiV~|
zYBqW`&kO7p`V%2U`|gZ(#kgKgI^}n$t@9g^z|O=+g%I2x&AE4g+VB=I9e;a4)NZ}d
zX0?l?SQ#N_)D%F(Q&m!qG^2-rU%a0*|HAv-`0S;ZS;}pemSEJm7ZHe+h)~w`qB?^q
z4tL^&0ALn%mMyLKWYmZ!AN)%efaxlgfJlwhSW~dcP#OnuRP)28HYOtesIW#<*m!f0
zJa@~^8rW+^32pnN?>bxFF?&;q^*eZ@>~Px1DO;<P>;c|qrb`K{!5<C_J0Ka=_~&nF
z`PbQy#_!-NK=(1+0~EGy8u%vX_n*1(?s|;(c$X~c1~>NCu+hMcy$*&8{|f`j@n{(_
z91&OR2=A_|7`fb9T0apT3MQWFA?r@i`5%G>2d%%vV0VS#7*N@l>g>gpe*0C*$;j=$
z-ONM?9B;bgb7^pB<_y5h_Ad4UDk5mpP7cXjdjZoCqAQM>wQcLB%8|{{-fzyYGvY|B
zMcU+JzSpGBx&eQe8p;KV7(w-{3+OPVUv=)Meb9X1b;qFPIh2R@Lj8V^=h=9q;1pPN
zyTeH4x#jX3O{&wtrF?^uz*V4u?!cm>y<C2B#F65J%$l3>RT-j==>{x%T}4_|WFeL?
zd~S&LWUZ3s6}074uvgO)T)WTABZGR#{t43h6U<=49{e0*Dd<eZ*8cMH*&s)B&{{dh
zPLjoBioJr&)^<xVkFYl~HXwvya4NI8eDJ~SHzDb&o&5kNI1jjjjd`61?g)exh_xax
z|K0}a`8==TgJhZ>Y9BjtsK^gRc~GsK{`Y4k&|LRN|2|Nkt4!7Cdq7>~cA@jAunO}T
zR^$mY=jdiCGYF5TfJpS_v!go&pu7K-4qTB#v`_fl*1*hW>2639a)36V)8aIzPl%O$
zu@f;0E3jQp!6hV(TYr!u%8PSC{6ebN;>Bj!Jfj#>s=(GSSs&5s_rw?oC$ST}-OX(D
zKzV@nVWf8MvT4J0CauXVhqg^d8Er`^`HyajWN&nvj3$~zYUb$)p6zwM!>LEF2gw;A
zR&NIS^?@inBB32JQ3%86O1v7fJz{SW7Tt6J5Ru7Cp+@Eq!per4V-UnN2LhoN27c6!
zdyx+*OQ+?Zw4Af{rZZdQjurC!1qfK<q);kMz=Ewq{VoOJl^=A%T+pb>HWQ{2ScO$w
zh-Ktx5ULo&|Dk$gQl$#}W-eJ`YS%mgrZ0Ek+Sd)9#*(014nn-iJjes|N37DK>D-L7
z`*Q@wadIc;U;Mkg{e7&x?$T(;`lK6w_khapB8AZ@tZjMT0vg$)f18UzOtdUUK+iOh
zbPu+iGWs)Qbi>^$+FXO~7@O?T2cct*&gJVoKU_n7byab#a+*qQ`ZH$r0||i`C9BT8
zd4!sC!8ZavI?&`Qtk;_&^Lt-FXAqRpx3t9J0wy_e{L^`Flg!jHTLWEWctv1y0UX^&
zW}KjXGk9`_cBYwQJ#+HtdmX&`UU$8#*5dqIRh&}+BAO7;^q^t8YAOQ)BtH$V`;jSJ
zTO}#$4Rujrfg=pWz~v?eVtg;a<JS<gnh1q6+BDiS8p0?ECxZrE-(md;QUhPE<WP6s
zc?5|XZimM2kH%x{-p?E#mGE@ES(*ZqECA2)H`FW;g3aIdQb~~J?{D%gB~@c^({T(E
z*}0YqOeUDI_7Q3LT0Ua>8G3=Fl}I`|8>v-(CIYeIkRuwzwbXA_hcn*C`*3B3!DY!c
zK2X6hX{PD@@iuS=&y*i&7?EXec%c?oqpL+uMo=d@l)j+T)!&Y$#OEX|z!9<w)nj8>
zJk{vX%m^RP5W#zgK>||otq*c4lx!?VBUW~yp;FabZTk*xJQHPtT?-}Ic2IYsoV>=Z
zy_e|6%}u|=UF!Ga5Na^fTx@YWTk8Wx-?$ASC!vW3*ZkSa3}e$Hik%HMS(MS|Viy+V
zF5u?q!t}U1e)Vg&8%e<Xj~YMF_I9LqDsm%rOJ!#hrwQdzLvGBSrrJvWvGDZ!@(fji
zE+Atfmk#EZWFfv>(>#Y=0LUgCh{-xOYB-nXO~J|rA)~VPSE~oGqm*w!xG2RcyFB_M
z<p#IxBj)1{aks-)6-r2z6ie_9_XQ+*bf^6ZW&zzoDH89vrW5_u|JU1!m;bYm-u~+^
zy%eQ9^T`z7ztwj`aLqn7nxVdyS+04Of|fgi6hU{AQ&KjV*!;zg9Yfev7Q(E^#P48l
zXWU&ml3pTd!PlNdhiM+eqovAF)olce956~api@6Li$Th8*5g9!DEqC<DR!;KXJ}9K
z?;ICylr@RZfuY9?u{qI@oP#KmWU3u>wNS?;PTguDgz+vo4(0qFG$cRjEkvs~>(m_S
zXf{h^i(80kCI8BowPU|I%@7714Rp&haSK^~0&)_Fvs?RGR0fD%q@>WyV}#d{AYg(F
zG$fS@t*V1*D@=gM6!5pOu-lDu)5?u?QmS6U0k~3hLPGKP&6}&X0V*7sXrk|NrD_Rg
z!LgsFI`!>w+DzDM7cFk$oETg7TmU(C0Al4JP%o_Rpl4=os&YiNlG1~mXV0j`E%QDf
ztx=JBJ6}}rPR{yY@>%9n(XeCI4KWtTB#K<Q77oP=EK7B<pgOkzjU;U+EL|&^!xE|M
zVo{*^{cPKc(r$L-9xZUCiBT|P9=xd-*H}*aMm?a>cap}vle=jwgGLgmFf4~2mE(xC
z7W0>T=mC)z@y$D#APb-#EnmI+#1$Vul(_SO@<;8|W87by?=}FMCGgxLUmBdG>AO!s
z<pcsa690i4OxbG&o`29~=u+gvlZ#oxY6ehMfrG2n-M@QxL>t0yKdyTdVQU5796Cnn
zV*RhIMsuguutRMR$QjZgza<UYlxc|WJvUX<H}UmP&S5Kj5C9*krdky7=j@HPVK87c
z{sNT7J#L}Y*>c1ZjOLhsr%n<}i$mq*S*(rHFrnL=lB83$MA1@1@2@ZgAW2Nf>P+C<
zN>c)bfHd>Dfc8~2P5|%G{_Lkj#t)uD>-N3A=xCfG%AI0NspmevcU;}Y<*MV-mR4b2
zyHkn8^6<~^8)|G0koW8U?>wj8dj5`)E*ZwVG{n84)CBOv%Ta1^BTzJW>XEo?0a`uj
zhthj(F*xcK`M00vzqWAqs{}f{{Ve!xm3kOM&`5?x?dpt@`TPr4Wt1Sr?4vV>jg24H
z1BGN|wo2Ga7At7r=>)t}e^E+}pL1Pf>49-W9K%Q-t;3&U_&`D@As@^L2m;vU0*!u^
z*YDT{u==+Om{&U5I#Byk;s8Gw)z{3gWvNelQ*=E$=8IperhF5qCCPjEvl*+DY~$xR
zy=)6n9D4Foq3sTd1A5dbN?<`k#INFb_}v9CPs&e4<CDHSGDvmLpToYht+)AS%w=S%
z?9`G7Kk!{BB#}j<NTv2=k@w9_uWB!N)cJ9i)6PX&<=q^>{w90C5dl8FkC*w#C(D-d
zAUoV>aBbJ(M0yc$W8)YemEr#0l@egT`^O*^mch5E40>K)+c=V<igv1=t+r@U$Htnx
zi;2Q4q3+fZa|fcg9G&~m5|8$+TacF+BI^%)sYc>DqWZ_C4+W@Lv8)g6pcsD3GxC9t
z$<K_J%qz6^8dl$>#`vJ^;g#ladY`YUSQWEZhW<@@a#Nrlgj=GJW63l3T3S$*J1I&{
zw|R%x?kZ?!RZ@iuEd&I3>Hg!{h_SrLy>ZciB7RG49~~%2tGB^QlC)t=#h*obpBOIw
z<ePMgF%AFSp?mGkQC(STL|uQFka=2&xr?EJ%OCqz95&W3@5Wo^;jm0DfwrIjny*7D
z?S!zUiGTGfxZEi8AM>z3Kkg%}DX9#-fjs)RxgQ!3xhI>bfkG;)29@2$H$Pt$LkYmf
zQm~iSbDi0nv>!bKWyo@@?IHhu>@xkqwL58bCPTz^C+oBR484`j$Y+t@CO$V_ga_eP
z!)IMm+nYeQGGtO-|7J>{)=jCdb}R6t`FWhvWXEnGey^u68@Yn<{7UW1*H8dLNfEr?
zyxQ^u!p=Lu?US@P$+HKafb`9h9BXop$jnIAT@)Z_**H3mj1DfDYSHW(GyJ2whZ;{|
z<9s7CLqN0dk)*og^`opk6~BRNU7z1rX(QLQg)%s(@}ySu{ho+>&OiL#H#$VqUH+5O
z*0(aTf<SRt37(zvha2HL80|N&dyu;rQ=I*^$H1$y$X$Iy4J(tY<<5IG0nYjPJVK}i
z<4qz2j1rG7MEvL@p~8MWn*=B#7f21`Znp%1bYsp$T`BI~o6bSbV;r_FVw+PlrD8eR
zHg@cH#PGA5wZj2GtpSC;xdCLRK|DScs!H*#rc=8Q1mezoB16a~e9!!D5Qz%1r;J5a
zJ8^w(i+uBr>Be;0_{%t!fQ+dhH@)@!HU#}})k^0T33$*5M9=g>*_*;{D$E!}leMnb
zCl$`OD5b5HecQ=#rj<^xE6>cBVkbu}2i(frD_TbVG7O6hI7(_Q4qiGBQy|2eDdr(E
zU<0<>ATnw9vAhA{eA`FN#){GwXL4eHwt27BT;o&UBDH2F%Ih_o-+{+m|8DZADI>RE
zE#}<9Ud@Bv>d;1VxyblLtMc2`&n5J4&YmxF1GaB`N#46AdOAEjRd|yYoeYS>jVC_-
z0n!*?-6xDHda0FNMoNP-C6U?m!9Kz4a>~>cx~hZc27~c%GB@qq3|gzphccS?PlM>N
z*_<#`q~eY%87g{SS{WV@EA52zn-Dkk;Mv(9<Nn@f+toS1TOqgVE{^qJ=go?^$#z!`
z4lO1zZk<5uo`qUOmFRq(3ft$k%{5-OVYq*P?d<LTX74Fo^`+HClg$7JmK~WAQo&;p
z{qIXleq`U8_j_V<&?{b97(Kp@KgH5QyY{qVw}WzWc}Lo2^Mg_6*f<5q<v)6as^QNW
zRvYh3;kYJPp%o_wQetVprtmJGD9$_wBrmy_RVN$KW-19`&<hg8kC&h;Mx|3ZW_77e
zjcI6^Z=C{77Ie+6`R?|0?3Sv-^30RYRD7YxKQ2g&YqCAtekHo(TEjqntKW@Y@*6D4
zy1*kE^(8fxaJTmuxNJxI1iiHE8vX?<^rCecye#cWK?xQVJTBxWtP@~<M(Ww5>UqU`
zYl{&M^=O3gg)X??u=Ou4=d`od0$KP1af&Qe?Jp9I#rYDjhx}*|>{9I<AC)6Oc669H
z|5bKo^0pbzaO{62=X*0L#BUGJLWW4g?M8RA)(|g?|CRA_C#i0JhPO*XnXE|%)2&(0
zS)M#Ns90m9u0aWfJHlio38lFoFdX5aAXh@hxAuU?q1{P1G!=F7YlEW!A)5nF^r>8M
zNFu~egX9#!O&S=G)JeARR&FuJWQkgw-igRx++f`RCHelDQc`>vK^<fU5e6spk*n$@
zU`j0i=>6zDv?Amu*!eY$1gcN{spzJ;y2}7Q{#aMp8s9|6>U+^p@9JqwyN;%S(5&2W
zkf~GYwwg`y`tzl01a$M_&1U}x5sP4iMgI^UftiAlj=lJ8s{c#j1SzjutW{F`e3C-x
z>9D%StU#~xV1qeK8{*|GC51<an|qsM+KV&++)DNwb!iVQc<}4az_dvYtxEGRg<dIf
zg*yK43b8wNyGC=9%H1i-f01^@22}0?{l38jSP8n4XBDM+uV<B>KKYX+VAp)96K~4G
zasna;R6>Bnk#erPgZO2etur>;HdJBP;LKH}Etd2;L`FP0!e6#`E1Sl5_)7#brxY5q
z4;PFLI^;5$TMmHNja9#qijc<HbLuhp7~+xJI+-1QgFS+VH8R)zBd_2*DF50EYNx6I
zL*nbG(e%AQ;JOfglXT;DCAaI7w>40+)he&9Fu#S8cY%4(=${OBkA2Ls%Zu}Jq10$1
z4Vk8=s4K(HKbUs09?XD@E4BJeP_OXC6Xp|B-v1~$=lYM5b5^GRwwkjLasE@8|M%1X
zcARsuaWVgY#rglCbzS3XqMmHi#u4LOmM;?IEEVL;`7gV~G(5$`n#{I9PfSDvPa;|7
z&n6j`Tuj6)Nt%P}(|hXkdhh+Yx^o+saoD@kvC?s%{pzws2gmcV)#!?Zn~wq(y?1iD
zj|$1R%DT|t>d?>%`irgon25o(63Pm*pPwvxa*zZORFhN)sXmY$gbINPWCIw@zZD1=
z7^p8j9g6{0hydg{Gyo3|8R*4}06|{9)=$f~3&^?~Abvx-PXx*2|NJumdIl2E2{I7w
z6$x<37eIm%;sODFA;KgQ($yMMSCPT)udRN@%ZCF4HF%XfYu{}5w+6-`(jvk_I09M|
z(FaQFLjcl(T0?m1V&|IWPou-rhxZo=XwCm*%1@(5`oKK^5ei^ZL&Ea~zyS>aBQRMp
z9r!{(K>>lax;V6b>h$+)6Fig=QNcOh?t8{<C!GuJLABGvfr$a#G+j_peXCFo&g0zk
za^g87f)c&F=DH$jz3$R~58PPX{0?S^f`cUT*A3t8u=WSyMZ&u<y!Ofa%!Pvi{iYtm
zM*!jd>H=W^f;-cPZ*l@2+<#?z-RdiQycGZj)OcA@EscCCupPy%_E`cE?+Iag5or4c
zd}g<Xe{LZVF)`fUy8H&;6ao73_FKQ+D)2KR5L(LvZUF3H$9|GjUlZwL+WF(p5u~ZX
z2DU3e>jQQxrXKT9-vx>8{r>sQ!@B{T4uFbzKd%Tt-R+}bMaq_rc5}5(fL|bx-!3Z<
zeE)3}#NB5p@cNG*^GYod<}HBwA}?3GzfbyG?nU8>(N?2=sE1<9kk8Oh<Ci^oJm#D{
zWN^I;*qT<?mcnJ+H}@I>loN13z-#3XHkE%?UfeBVdI0J83ig-R*P!1oq55q>t3J#N
zh2w7V2hY*U%xu8p^~adnsV^)N#Ev4-9)O%>A$aZk@-9nC9^nA|ZvO;xAJiUpCv9k!
zk^~9~747Adu<?iZ0Pw1@hyx7j0WACO{`U3;{0i^`{P?f{AY(|@FbrS4?4FQU^t(P)
zKGXI&tuGORwrRgkulwx2ZXfeXAVGTkTibU_iCa;K<t?z!!=)>)c8I|jXIy=sb`;GW
zVt)P7)#omzC!#;-Qm<Hc5u;ML4u7*rdR<aDo6g)wdS<*%X0W!g^Y`j-_LarM_aKX^
zo+in@9O%Xv0XyFmXBPNO!kNUPP<V%1RSR`~ri-j?IFrZ(H$Z%u<pRmzdjOfpP$&rg
zX{}hkN24EZp(;*!7i`l=iZ?5>-v0LK;kOA($G1Y{-W@rtmsD4NFZ(i<6nEDeh_$k|
zV`k(R-8)`}Uelp;=AE&RW?;q2rO}?H5y%`b6*KgZK%u{yj^;~i$XOwVSp=2AO21Of
z68sE*Jp?gcHA_o0`u8OP!;$>E?-nbhDkcmN%z~BQ)TiLlb2KZl(pv7lOJQh*u9F(f
zP4WiKV!dswmYOS1RA)RKhZ11RjIPQCHeEzfyx;V&Tau-AORnSQMLkZ-x`f-;YnHSw
zi>A32e;*GKNNXv{l1o`kmQ5vgJTn(478>OhCK;Px@1Ll87<t4mWVR0x_%Al`*JZ)?
znwWHVA1Z@mDS1()9xk)17Hy4bk~zG5a-i{W>zFCAsv|}=NxC5UPu|}fgEd!jGv>x_
zO$fpKjt9F)xFQu{cqw5K@SV%nA=|=P3GW!{pQ1D9ja2L$8dp8l-~P<_(eQ|-qom6C
zaw}QQBL^Lm5A)wh{$Vghh4L@PY%X7?Ka`_94gZ_mEjx!%fsjfUB=?Ed5MPOpFjpj2
z#{CdBzvqD18nJW@_m%?>Dwkxti3%dFVNGtYDJ{@wP;PA)%;|JEgKWomD4O!h2Q%s(
zPWlr6o>Q-ce59-$-j{QF@oUB(G~_}5t%@yh!{qO>s-MH&p@VL#-U_9yOYpD8ciD85
z9x$pBpZ$}Old{4?ZlW$350xp<eu`&Q$l_mGC&i(aQbTQGfDx6ZT*zC0t`ydpuJ0**
z)Inyv!vpctq?7|v7X^qni-Ht`(6A}TTs^lei;<J?sn0y%Os177`h#LZ1K`vT9P<v^
zv#wFh137F6vbwvi=;2>o3qv#=+MG5dp}cg<hP<(x%OGqEwhXaZ55tb!KR?U8;D&^&
zOBE_&`?=^pBH4i*Z-kEq7I?x7y@Dk86U%kX)1=AgbNpvbG^(QFXWlr$qZw`U^BRq(
zc)1Q}NIEFC5taCUo>Jr3t&z&hQOGmiUk$hJR;=S&r5nARqGxJ27~~hSctP(kHnq_X
zaDeM&`DNkmksR@#zzNKO8)>|Qc*yQZ>>(fb0B&~UO6X6xQna||aagu)el{Mt(D8>x
z4ji228ClRzJUME=ZxqfOx{x&KG);}~2cJH=av+3PY^fGRItv&FBgd<oE8HtX+4V^D
zv)6XTaBQ@W+6Gu%gl$Vz(}oe=a@>Vk%OG!o;#;ml4S*Ox_hp~$g2Hue`vfsn+H|S+
zD(Fz%61cehA<2E?2xY^3anVY&1)g2g#kBUatf9NL4Xqh+i0w`5&7zJGJL{x+gv<)e
zShcs=uc&qA?Ahs%5BL1(d#4_EGA|9iA0Jr*aji4I-fWY%mjA9t1|!VZAN4^N$G)!T
zV*p%AlW5P<_9`{|n+k~d#WNK3*`e#lr{e_NDPj*apW~W9cSV&%a6UQ3fEv1LPk20~
z%zHb-d8O?#=VOaueQjPGcOyWt;8l4jkh3JJQbCRf2fXwtNDYhp$gvBLfwt1UO)sfi
z&k>IXvAUc0bydZR+u~_qR>5SLqts&uDM+w|jnBa&-JGIf;j*h4x7C$gwjJ|KLprz5
zeL^OYHOoa4k6tZ(b`RiV%OP`-sfDl`;jHOdGP$wlp7E72u_jQ~=lV?3NzD8_5MJh~
zlN2Hk`B~JZm;5&7NM%hBdqee0^@$lT_FXc=745yqU*gV_%7nO-r^>u1>JYEIlGR+r
z<YCi8f5p%;59}ezL$_7~*jw;Bd-5kEvITTW+LfRt+1ga>IT5BCIn+h5NTKpD(DmQ2
zUYsmX#Bg{R>r9U~9wVL;RWFtYCEh7_Y#7Ydi5z9$^Vmd80>pY6MT>{jwhr2L>4LHK
zO<_{B@qE7e^n=hiv4OK7+M>J+v}DsE&UEq+9~v)=CkwA<5)UPz*ik-kNcR?9@?JK8
zJ&X`cNu?DPRd1XJI)>0<(y^n!F-1&s{bfSDD$cb&CHbKRXfH$!laiR>r*5?rH4p9(
z?;5+9WHU<TNL%Xwv_(zvIR#{c_QT~8-46*N<wZcxA`jCCkA#d}LCo%p29e)286~d?
zr_X1Sev4#;z+{K9h%iD$r-Ber%HtTRLw~igX?N!ELC_Es_c@9^b`ocbFVb12eiM2#
zA^!FhXU+yc&qtsm+3cS`$qxO+cu08A?lW(T!P{ZD;DoWV@qlmm@bCvA-94%}JGzL)
zNh^dT=2P`jR;{f_nS71BNsUR>7%^Npb*>(J<A{Ici!??T9WK}~NF)l327W&&nO%%D
zgfY3(Hn)RFoGO>duDk*y5d6COAgvRrFr?IoQfhJvw4sj-MG^mFb~We8F5};EnyHTB
zv?QI4b!VPYqF2Nf0e54ptLud=W+8;rEa=`o3<1(!Lpq`l)iQJV0JeBH+})ILV}y>c
z^tF-$z_KkPbCu)uX^L-AwC>pq4<`eO4;*4HTJ>3))g98VTu-y^xLTGBE7wPAXD-0y
z2ik5xl@xc-(~4w-&d43Ff5ycDdy7|j(A+-;!}qmQr&^Z8oqNXrkXR8n`hsZ;S<}fU
zkVvjm5E+xS<UVRyswp~!-Q~E%LA7u>Et)hOSG`2ksby~z(fAtSRl3=IJCBkP1A4On
zJM-H^6U5OFi;NuANVdwY+^tee4YITKnr&uH^ZGJfG^*}b{nUV1tZE`tg@ukdZcA-q
zq&3g@I-011^TGQ>Q4j}f@i-D=mSu#WV^pEz1%kVO4Yb91K(f*}-9)TZM${D8#Qf?@
z-<A@k=pFMWe&s^fzjaW~{i2R&MXh8*F_c&~L8xUEp|k~pF&59@lM=&Wqk7Z<g1AXj
zwvp4pqzQ`Ap9}nD^UM@u#VUT4kIH;jj4^Am+v=FD$l!eRAk7Y-+xR$LxU~vmzUC)O
z1?O!T_|54xwtMH_Z}(USuD3GoV*!(L&vGEUK*`w<-ej<U5OGEzvD>^_bKKsoEy+CA
zd2O+p)=uuQcNTuzF*&@Hpwo;O_eiQ<*K`MA4!n5>dyGqghZXgXL52do??5)G;al}5
z6y1kuX_)S!6}b_e9aX>&Tt9Pd7!huWkRD@(aJefk@)MR$@y=BVW~gk~Qpz`ta!8$l
zH46gSCCGujl(tN9#zsq^vuG(bXXRy^!jEw}z&xy68oTL4o|o#;){-sbXPT&n11zJy
zN(V?f>T}0OvtZA*X=*p8n-ExGOfQJ5pJ|eg0TZWV17usKcArG@F;@ugVpXR5C{7ax
zm%&Z!$T{1zL;C{R^LOo3(`Ursc-zHrH5jbKb>?wmwam^AFyo`D_D+%CCNB=_k6|o9
zo-;%6ktH3@9Mc9;iK|f}O!vdN>tQGj7y^2I@pbJw{X;>P4r9{z?yrpx)1jx9Et#8P
zOiF_G(&}jXcbH~>GRU_1-}XncoIE^XHMQrfl@8|RSus?~!uK30jEf<dZSZRMA^*^2
zj7(u}RKy)M-gv^;T*_TGPMwJBUjJ>AD>|%y;SXg$YIuG%)-0?ATeQ1Wf^p{0i~1nX
zF`yA;SEN*uO{P-~`*LpRjPXVCcRF)=I^TCC?E=;5%NX;*-T%s*XbVSjD?w>Em;XL=
zkO>vd)L+a~uS02bEuXY@3~Bom%Q8coRji6({7}lH3`th-G83Q!4{NZs{pPu*lTm$j
zNUF!DEQ5*cAod^=n(-2oQe=ER5`zM-IoT3ly^m#TaH(`NLEF+vAokvPk5EmT+j2v0
z?$^PiaCd9!2+_CV!H#OKTF6}@#RBGOFw$Jrne60UnB28O@}vPx#8EuPtG*w9<jJIK
z^F%v>&k!*7k$<kClbAgRtzSLvbmLT)eQB9@K8s$=pTAhd^}hC>AzXCaZ$&;Kj8U4{
zO9i)u((-l#n=sv?Sb)f2Ey@(%O7%Wh@MoaMlE@X-DWf@Nj5hR-UY2(OqlsNaE}c!1
z6*aiv^fykZ6nEJ+Uyn||bWXR(!L*u3qQ_@C_!O{tveb#w;U%;YTr`6-CUgpOZy;Ar
zbPa*?n`E<k$;SXohUjE~)&S2yFh#+Gb{OP?J-OHNF59DXdB@S&$$4oaoE}LZrqSD{
zsEeZne^|NuH{>MYcawZ|$lkfNvFdPtRIG-A7afMM%2~bK&o%(VpNgr$YWd>>qL=LR
z!Bm?4OP<W*0V{+8AVMl|lka1@Hpw_e%bsVmLt-<?Irc8f>mYB8|2_cfmJz|<X<Ilp
z;sxwc@n+|wXF<$OfKhgdM@7h~;4>$dgGVY`N?>`{`9Qse-mG6<iG!o~bPrJJw}|#~
z9*kK?L+_Ik5E%#_f9$6nkWrnwkR_Pt{Jo2R%(=TgpQ4kI{KH&*kVieJoB7~-U5YAx
z;tO+&6Xe<QDOx2~<XL{4jw}~__X=Ie?V*Lcdd3(Y%1eSM=QcPu`FjLk$H}<_o7H=c
zK%wsUZ>wnT$WY7Z2h?j?S$!=ZlXy*gZQlq(6l39EM_Fa{65|1~xSO0xe!4{maHtzZ
zPT$YY+geAT<FFan4zf|}-uKmxF%mYGu@E>h=5U{RpM+PRHJL9*@pV$DkopO1JUu8y
zW6nOleF#a_bHlDH;Omhw-=sr2FhXyiZmD#_>{|yWj67OgE51ZS?Vu70NcE(7FAgZ^
z46vqGpd;$8%n4iqaNy_$ll|rIw~@G{wNe&05Pv-QPcon#@Z<^Aczh1<Nh|NuS}1V;
zy+YThYB|fu9Z%l0!Mw@28BlCpoqkmCSgL*S;Nj-LmZ)3+vr)*~gr_2G{jN2v`Ky2X
zvKq7@HRJXp9(BbLTx8pww0-ST@eSZl0;N#1phy$N^J1e?RI*T#;hM*tH|d{8=xJn_
z)w)PCw3DXHl$`UeU^;?b;d@WSLQlj3Xj>b3`;_u)7g>eX;O_fE^X(V$-$ON`yJV|P
zo8zPm$L}svX?24+b2(qwqwdy@o5CXIxYOoLj!->N+|YI#bn<6Y#(}~-a&AuVCCh8(
zed1>(&3Pf5HGOY$a(!Ads%|km^!KFbQijyAjQvhgsrlI2d1UsziFmhCwmAZ5SijOK
z9ZW@7@LoW=Uj>XcQw$7P8=WNLS(CNXo~nwZ9TP?U>|Vsa1!%=57?mGem-7hiRCPeX
z(L!5h=h@a2Eb^|Rs&jDamXl>>>Sj*7Fm=d!j>GFFvTaQnq~jw#Iv1jwpSYBH<}HR;
zDz)mzOmEY?5M_vcvy90=T}e|#x)L4kE`L%&jg1K{4nB0B_b-2)I&N@wH20{UlNJA+
zZrxEv&eAd+OQ;A9rG;&7Id&l*s(3Pz5I>jvngm6+6z^8hQbA;n5~lp*aJzN(vU`tf
z?fJyV0O_3Tq*UUl_w}yT8Xk&tIe+V%*c0)_eE&Utsd4o#Z-VgJrGulI`hrg-ek244
zIv-Lo@oMQXm0uN`4!FHhk<RoG^wq9ZYE6f7nv7sFev+1RMRW?obf|w9n3gktpBMZ1
zTXiolh*M3xm%^&<ih~dGj(z)Ujl((n9P&!fUbTjh?fz%O=C>OI-2`6nCad1c>(X(#
z4T1h3?8H2AgpnhUQ6t>Qygt&el`}4sDx*15SNtpGzM_V|G*5k8nW!CS+G+NbX0aUD
z<7XW?hiogqq30X)b8!X<1$$8oUg&6lGEJ9hu_ZSa6yw>&m6}WiLnXYrp>avNeux4&
z-tqeou3hd-v-}kYAZ$-KRRi;OYR3sFCjTwXt|()Q-EowNMRZf@@n^2e^XD0EkJj&N
zEp%A8Erv)@*EdQBZy9=uPn5b7<zd-sCh#vf%sMLky~4tzGlC38Si0GM)*hdfeORJ|
zG@a=xNmaZ%`ykBLd=XhN+E~#24ADd0Ec7O&oM0iCXMBusgC*NEWGns*eymM7#lW~j
z!{NNDOvs}Fz-N`{GUWkow(vH^`}1UX8Q=w&>^+f0Br>Al)E=f5uxB{#VlB(M{{dhZ
z$z=B(S9?OyTpCQR&YSFHVN=}AinshXaqwb1e6}^(LsDV<QotzFg?eER*B?NvC<{Do
z^}WiKRr=Si>g*r)Qlo#z$h@dpw$x`w<*mzJluQnj*V12~1NT*<c<|>D^D^;PsK3&X
zCsS$4|2FxywMnzw)K(=}FXMC&!qJN4|C$~$$?Vg<wohcDGPzT`koAz8r#B@~B>r5S
zoaIi8T)SnDlpSNi+oa~`dvPB%E=TuEuw*j)$~Ya3hvvi{E@v6|%3O8oNwN-YXK&EI
zXC_5lJ<{0__IL9;QuJCAni#Q={j#4$K!5zAQn!t6b&~;GHnSWwwd;lSN;os0&FI+f
zt-WqC*UR*oe0zZ3`FcAOj*umMT)NP0wRCrxw_X{ig^(?P@jHb$RJi&Wu$xqc>-84u
z%6wMrc^8%p4a(W<>NOJFR;na@OOOyrn~v9rMddV2I{bCe(g;CMiEJbC;@+-r(s7cM
zF=zeq(vf7H?axs6df%x||Mt~N4(9<^-pkoxN7$zBPERdmamZ*W;q>5yhtnr`rQK;8
zi?+_XEw_911*7k1G0Ll*G3C04w^GcV^NaZ>a<M<As&PnmlDd6MGo^Z~Um)DD^w9wE
zMi|h;BJU{e<t2QM(>EmpsZx4Uzu>c=L*LG1TwruAYtH+}M@Qo9p3XyivH?axm4=Tl
zBj4<|>&YQDv>WJ6?_rK87av(~M7rCQY7sHG@ma{;EjR0%m*MA9U7l}mrQlQCHVCCs
zzRo<Jv%u&wOnkMrxC(-aIPqkIwIMiDoj8*%47jmb9KR;x_RoPl`8s>|3P0&JHEdq~
zgY>FLU9IL__K5T-8ozNLJn{$6Gx~LfsYYMeUX9K{Xvt}5-NUCuKM=CC05L^4Rg;&y
zj&e+s9I%b3?j)z=r%{h11q)GgYgDL!^d&m8*&N8>-4!nCe0KTUk6X>V!-s}c<e1C-
zjKc!S8`-0TchMQ!5k#LrMosc(FNUAE&*A$1OSQvq6l%QTefSOxdV);b^`djz{&w6A
zhd%dHJ_PFv`uH0hcQq&GVSkzkuNKzJ0}zf=z}fLszNjdc$e0$SrB5m4$x$P&kSC^;
z76ksN6IE47I}1Y|JVGAyJ$&m&>S1O_S-imJ`ZZ1XXp8x9wp3-AQI8ih;}eWgu$G>R
zH=zdio-RWf5W^`qEKi1KuqRwsDbFhZjY+YQu6y}nRdx45p3W{B_}TuY-N@O>ae0$F
z@J-fEbW5b?^ojV(#nCE?TUwQYu^Y_UB$HH{nb`x#DxR}7QoWEm+ApfNZS{<w`s-6~
z#N{l?`u6mEm?S>DMm3x72Rkuii<E5n?`MNHs$SSyBkYh_-H%7b#+G2Xmk4ii?=)8I
zL|bL_2X%L|8Prj#WZ>lTWOS$aatpjLg6<+_+4-G$qLG`Sd^o6FSnNd#$mv&0-uNl#
z6Y1y?L-QOouu|?zCZ(GY`-OWVbSJzbH}u{1e<kxq?A!>6kCpBA{b3lV!cYn-y4>lt
zc7bVks15VXH2LJg{>{+}wmNP-&3pAfCf?-^8?})yB`#)ePcBtz)_k-vay3-WG|Z_e
zF`dpg)E>jbKUkwM?2D+Z6$JY#90%dfsQYmRz>HS>oAaGJ(|?zQ?dFY=ve>&n=Fr7F
zocJb%_dM7b(|$i|WLe3{j?1l@Q>9#+hVb8B+;#KAXjOpb5hS;|`ovh$ta~HLA2Txj
zvqzn%1Ny~Wy=*lYuR#(RUZD4B^)AP8F>Fe7!aB^wAIZU=&QF@~S8<6^%KNUektt~_
z1L_dCk!_>u20>4MARCCyeBIy6-XpzQ(l!w1x25?IXsR_{^1@TV6)dk(3?&z-nba3y
zcJKLWy|z}=u5ne>vBZi+`$-ys8zUsUw$eYrF9r|_@50O)%jNzmdW=uMybIcGIAV=H
zm2CT#kp(se7tjB2$Cgbs#JbCLf3C}u%l?Y|lR|4QS)ty0w4xwTp*D-F;xkc{1Z*Wl
zAS;&$V<bJ4U~O{bo$HZ$q^d$ha&3yc5@UrrtT;<J*DfyEh+)>Vr|I;VX}wuTdr%sr
z_seLQ#h;@_gAdJB)O{`Y)<yxzR|InC{hUb~^jZB~LtuvBiLQhJJPU3}P6B8M-jC^&
zSEHIFR;6R#8T{|+sfx@K$kB=6lu|l0xu~;djt>oCuF|pez9`hwFuN75_W0~QfdwnX
zx~&B!vX^O%2o2Q0-I;&m8JzH^c4@LO{e7JTLYZ`AF#O}c^N3O)B@IX8aRtf>$4`)<
zavsQmy=;?+2OT&qymJ-@ylh+mzSc{G($LfIZ6j0>t}F`tUBmB*%j*p)3{t1gwC)QG
zT4_s1!M$NmC2TkE8NI|hlWPr!IyP};e$Vf551WnI!~@um+1@5TgWaG23|%-jyEFQq
zXYWj)?BC&LeO*g9ZFITU`tcihu&yD?m9FoR6`?5?7-kL4t}}Gt>h9Mg@rLbuOms8q
zc)vts59?&19>46l^|`$>d=Hn7X>Rd6?A;z5e=RDU&+mDCA5GX*s~OnP484f}X`^Y{
zMXGG}V)8_fsH!#PF@P+-o00@aQfiePQC8LDUR?M>^~%PGx>1pVyz)$zfq?YUTQ-4O
zX{5ibPITn4a<Z7-CZ@mco2q|%PY9_p<2GDC$(i4BM(`_oSBARyA1b5<R~bauq1Eh)
zQ65v4KKB=ZFMUD>#dM$HqPw2R-3!z8&T8Yb-!n|DZV3w6gKtg?dT9|XnuqgvHj2@5
zOy^@~MQSl96h)rbcioG}&dqWJ2Y#NBl1Hg2)aHS^MU`SZGFNSq*(X(6ZZr4}1*u5k
z6lwh=bR?zw{N8+vk4jF<4seMau^D@!MqDN4X>Mo{6v$<AHGYWcq3xkCjNm`QCamo@
z_p&T8!KmqhvL6tAD+}o2c}g&+AgE<>dCaN*Y0A5miox#X3{4&Gig|0NqdU9JQ54st
z-Jh4O#b0|M6@%+Yqoh*GWH2(_be3(dHn(oqf~_tVp-<+O;qF40;4hS!$IP!;8w3BW
zz!j#8t{xe+E+`(tfS0dh#@WmkmLX<!fcrPcpDvWrFrPs+Zc>0kDAwW|kE964>nA@_
zPzzPi7;R-C@SY}k2%-HWE6+_?pizwVw_EID*6kko6ulC=K{BIBK*5bHPQ*`wcoOHm
zGEYnPLVl)%STa^P^3g>76mnJ(&%Oy;MLU{)Zm~T?L=9bQSX`<kg}ekk&;PBHUYZ_&
zJ-~%wej`_D&eApd+J<o*f(JjV1umVtm#fufu2x%Oa89Ub(0(KD<}5k-$#)AneUvQ$
zHv(~_NX3yPeMw*Uyg^g7QR`y-#ur)Z)-O!o=gF9;yxmpnTz<p)YcY;x{v<>z<e2T_
zX>8=|oAbmgxT59BVl-s}ZC};dPf|X5BvLr!;g@8x`x?~oSaZ?r>H_hU_Sa7u?Jq)7
z$qEneOsnI`f;zKzNTqCL_^@p!9l6|~fZiNJZzc<LQHK@=C8n&dSyB16MdvX>UjkkG
zMs%Y>6Um=Hf%y4K^U7^lZBkPv_Xk<Br-`_38T^`QW9OoN2xVBdkM{fWh_<7{1Pvb>
zHzFK5^u)l8FOHpW*=y;X-*AWLPp30g&RycTJ3j7gMiCIR-4ng~iFbwwr6vaT^*198
zZEuC>G&ryIgy@%>Quvn7t|LfpTp8ysvsq%Xf*Pf(G7)a&>NkYK<W6txM5nR>ItQt?
zGwARKwWh})lIx_odpjy#SeTIHR2&h#*w~52q)i5;Iek*BPeW3uOh+EC-&Cp{{yb4o
zQA+2`HIa#Ylz9VH<eF)QYe<_bO-u&c^oLm_Qv#CfH{DddzF1w>E@2e$qP0=#VmSQD
z?S=mGM4d{*<5x|wgAX0TKMC4Gw%eoUWovS9#q~yTqcck@3rk2F@pH)Jwt9#N;@!DY
zZJv-%Eu;Hrq!mxRcm(=X>XC<G(coS%pKO7ZpUqbp01qGv0V#;k)j!kZ)c(lvIqgdk
zwNn83Cz;2E$Q-S-VGI>9M8V@D$O+h*odVT7&?ab~z?T8p`v(194FDU7L#J%eRuC*F
zVqCqVSZXH)AX*B+O1KID^794<`Jc!P4yOM|W-$LxZ=QvSg_D{6e`@sq8_i(h<mO`e
zzvkz={yRV4YQu}Z$svD*y6H+&z1C`@@~0u^&s1cN_j7NX-7Z1fr<&jQVf6h~%Sx_Q
zE&J=nth+Csw2}$CI5KN~L<ia2KuSVNL`a;d$UiJX-wcwe3F0qnCCD`}ncsA_eR8tl
z%n6X1Relgx|8jPm=;T(k=+;(pa7;{aN?ICbY@#UK+1atRjm5FF@FQkWYHsd+9~~e#
zKHdGh?DSS1(@F*srLqt4=EPM<!pjMAwGRofin}~DEu~#EB}G?#JtZ}qoa|zZRw8L$
zjK(fmJcU1%1MEun*OtO{rIhSprSt@W;iQ=iL&yxJ3!omK4uU^b4WOP5IcH1_hlTbZ
z9+}@-`nctAQc!jp^V5bSi#?fnSz%!C%af|sk)+E3MyU1k+Jhl$aAj(7V*L#3*Xshm
z(-V%J>F+~+@$>N;NDg4AWhk9liLE8`Yodvg`GHhWqxHWO?9izObV~VS_>oTgm)Zi7
zpXzQIr}XM*`!@zchL^V%pYyaSPRLAG7-C(4>iRkWf!sV=0GXlbk&S_eJ}jnqt}HE(
z?CGm19qpgNuaTcJUVsYa1VD+ue|~=Ea(BDmujq#o(n<}}i5HfFOSQF+TTTaX(%S6m
zZ}f9oKF*~MB&DwLGh};m@|Gvy=5*?ZEBZK_QSdDTq7UI*|472j+U9Sh3fB1}Cnfy&
zuR_20A1m~5_kdlGpWC@aTUXaeAg@P1&V+vzdJz(ZhgtH&yFM|{tD5WXABB4WztZHF
zpyUssfw4BQpBC|f{<|6i21P1>NwZ5*(<?KJTB|ElGeI5b>gVNLuPZbtqoXOOCFos`
zA7DccO-c(p>GxQnrIEG4^R?(c1F+BFP5*Yg@gx68<8I2%Y|f0P`3Crw{os~;sd=fn
zFpl*t{YX&52sHsZ-XO*Ppq)t_42}OC5gQyI9r|%QT2q&D=U)5%LBs}t8-CZdfb=h~
zZzPG{+O5-7126!BzJL~%H#;U2gi@g<Kl!Lk+Wvv7Tj0;UP`|BG+nE_Zmv6Td0LK68
z(8qOlbd+^aZ*8#5B?~T5@Y4>p5b)YLOKaQJ=b?MEO*K?Y^|Uk9W2|Au6Ld^!nSAMD
zjxO;{zpn4MIG?;Z5Xrc4t;Cr~Y^Ry}stD~#$nz?OcZw~K$oXMSS*wSpJJh0II`vTX
zffh`T6neCCmK0j5e;|l)ZI9f*Iem-bO5W&Vh5x!g3qP{^C}IX~!Y#oDtLZ(v%BFcg
zI#2L)|2jRB+=V~t2Nv+L1)(Ds1lkmQj%FKKo?LtHop?fhSk8I;Y?2~zYEu=H2z^!2
zCha(mtpk6v1=}LKY0cE^QTOKTx@e;l3j?++T=&j=@8i!hw7rl+7t3MvAEVE9ig&8W
zfMgUxr0J;@+zfc$OECSXM|%BXt;?Z`(Jf9n<LvQb^IIj_f9q@)I55oEf;G6vAgZBY
zq1<4y5ej+Bk$-={y;Y@9Uiz2Gj+C%gC@=txMje50IC%=zJcOfDXY|D_{+oZ??us9-
zRZ1~rjiNzbGXLr$YKpk9u*4{vh5tMi&)F{{_ll8t)nLDW=3CyFq?_jtpHn_k<7LE{
z)5f33cx*D?9Z+o#2LZ~R?8N-u858tD{+IK0I;>R#9B?bQnn2Iq$zp~<X2Z0Pr6zhw
zNm;F+7%7T`7hj|;a7YqojI1a~403xIW!DD_r*Yg(a!<BZXIis>!Qom|I{PDp&sX_4
z+-e1>y~HBj4#7`Sp5y~K=+>}fHh#v)PPg=9GOHA1k{6@F-W6oey^!P;HS9NEyKRF_
zJr{lRw(oYG;!Cd~SL1eF5spbBXIoP2X<2Mn-RH3zQX;uOw_86}QkoD$wL4Kjck{nw
z;!xyENXTV)p!O-dUDL^+*qz{ppDIEA<}Eavk{YOSCa#C=pZfMyD9eAT5flhs6<SeQ
zsT4YA{4W4QK)k=2frAs|49Iq@w#eV#XVP{N3-h2-^$rf)58*Q+dU^<~Q?GFrsCyf4
zVQk1Gj44v5+QRUMXVZD)O6#o&$4u6*Y|ecS0)|-3<@W|P?A&CP8DM`~Pp(;v?ow}5
zTIG#%=V!q;)lzWn?{-J?Swv~|2y7s6Au>|ZI_mGv9G{qf+7L`!ZvlSn=#%BYC4BLc
zW?M%6L=={al(&B!bYba%G?JBY@J*zmP#Xx#YkQp)9@7W6KY^<J`Vd%+02^?t)%_6c
zF(|Od3$)QjOg`H8vu-Olm{RWAXpcmLG6>tHHV%A|hbUQk^Xr+>`Bu5eZw2vm6aQca
z`nzrn*%BTfHK4oz`z<&CXlH0!9s!we9lReeVbakuIa^0Wy(H4?zB__`spDks*WIv6
z0OAn;LwS2q1Ph*X&Y%iP+W=5~#W;O*v;7++(|~;%Z%R4q>i7W#RbkesHdNapox}R(
z>SM&MneYI4%(x!|&BP9dc2u4BN<sw1BD<z|l&DP&Z`31k_6Wj7X5u}{RWHKNdupLo
z6k~0&awq<U4_&|A&=*rhn-57wd?zwOlJGW*ybf0};SnW+4bPD|zqp5N(p#OMz2>b8
zXmmXC`BaBk#46qeK5*$$E9=97swPm}+HJ#0t7lU5MgjendQyMtXx|P(93J+b3cL-b
z-zSL`$J$9u|55XMMpll+pMUYk%FVu8m9MfgI3sR2725J_X^#LOp$kRB#gf=y^t%c~
zU-hoXzDKkcF~L(s8`4a%scA`Xc=mb3HcA4?BEm`+=Ph#&*DeL29Vw5`n*PU}Sl*+e
z^DjXJ>%}~_(3@W|?3B5K-sY(2o*O$(tTkoX@%L)|h!9<ddN$PFo*&CYiX1Q0v$0|F
zEHQ|yfDp3Bcr+rnvMqY{>~F1Y{he{eB(&lPq=V{BLZ+zkuyE^?!Vo>xI$?8BAz1+D
zDGM6IbEj9l9^l|Kr2caB_YkGJK}up@uR&y0qZi|rUa8M4=sQJpg9mWr=sjt2lr=O=
za#@^ozlO>LhTX#j3QGYP@I7zR{+9-tjs`}xeUsxxTWJ}MIZUhA&*{pUg2agY>e|>3
zE`dFP@(jdZ=$@XpI48_IZBiM9&JnEN4<W%5>Be7*_1HvuvWL!z63#s18h)M$MsT}!
zbRgxs-1BuBvF5NA?(=Qi!zY2bTH})^L&ldoKP@*NX-<%C@z7ebo@ymrgvmzr^8;$E
z+da(6@%~u0wu=&QDGPU)OL|@TcUL|W-K!&M<<?G$%=zN(XbIC$xM07LiPPIxE@5|*
z_i3Y$vWirLL<Gf*3WVl7U`7LK7x<s>nq2`3nW7lTwdmdf(oJlO^uE8C)x)+9Iu?;l
zqxn*|4Yn|9dYvZe_=MtM=tc(mMZM=088X~u;3a!H1hg<r7qt)-qwaqxfL`3$M!9kL
z4%+RnbN#(RG7q?KHwEN+KTqI0^5=W$rW`8H30Y{*GZ_DLHAse_U%9qHkwXY!%ahwa
z7f&Fd13(O%GRn{uk1xm9-^=T3#ThO3*;w@jw^nKc^b%Fh>3^1dz3aQ4EB*iyzsEep
zJASwRiJErNcLePuW7YNr0~}$1el6+Z64s*Vr5zlw{Jw0c0y9nA5jsB7n4Q1y9#NLt
z7m8N>xHKj;CaU)M2R{fkQm~6){GpdJwpm}eP;V5Ob>0h#y376Oq$zDp+>xN;wZ$eZ
z_JdqAjDz?qtjh)mN%`$fD}*>~rIO5OS?(>h(KrZ;A)q{Ub8@w8;kmSn?r>mZ9}O#8
z&1rnCz7#<f%ije(u-}5poM@RRn&1|G%KT0%g4{RdDnVXl^0betjUT%e@Vj~Sn2L^i
z81$s0H=8B!s9LzXS$)Ycs@NTUG?fUF1BWZ3pi<k~YbHN#ch1@aiN?leuiAXS#*?qe
zO0(3kMP=KA=i4x;g9>9BcT<eO^`{Ws*)ZoqWF@3~d)oA8dWgAJNjs3G;7Vu(a<uUV
zQJ7lka-yWZi0ud5SxYI8Y2suN3E?s`v3qzDV;Wr%XPoD^^_&ZIia?gr`Wh*R^!l7P
z9H8_!F#71!^}A+lV#zJVY3cb~LfcN>>#qKneUDQ*n4Hs7ntUeoV(Adnu_e=7L5ibc
z{JqXeK?+>=mLx@;yeOlf;PAg+9>!QeaIE6PFs=BJrANbO&RR?@d3pEjv`tf9w{y25
zE+<Nn0_>yj-~G-#A1j33?Xso-!8<g>OcyEF&dY_8(dGKT!RE>lR`v-}CY}ThX4Zdt
z!ile95eDMkqu!swEm*rcS*6fa<rn@|r3(i8*i;Yp%{n>!4g4KA4M4Kb!3~Vph{Z<D
z(?`*09{Rpo$~;bGOZtG<1~)ef&{N@&>0{5P+6AAHIPm#qMfa5JPR?=Lra4%yIE`$<
zSYV>zVS?ehK}uK&lUDlk5wR522*e(egCab#lbc#6Qtd;=I0iBBLUO>P0;e`bDO3&6
zNK57iMlN$>=1})K`A7c1@CEy=ga<Q~oncM<9r7pb;kPi*8@I2GD2tGhhvVF(toaD`
za*VGBQ_>(^2U3P;qs+qqCx*O(jIqShnuvF@bsh*QLO9BpQ$N)hcLRGVl4F==n#j3#
zWKmVg87|xME0W79!Rj+(FTg6*jZhkkw*pTN4BMFg1v8B_y-R}n&z5&ZJJtchVhs2_
zg~nge)<(v$@xToTVLuRCvnsH>1HV#@suA~rF(6f&x`c+BuX~!@ud<O#0@Y2WA2G*T
z$FH<IEw%N2;LbHIxa5M1T|{iZSz<wCeh4zf_-TT>_n(~R(J}2U-_U_&z2y19<&C+A
zzmtvKpg!2E)r?9=bq(wId8;Y3-&ewBtau&CZ2XyCM*7{vapGS0*ud0pLSnn9)+$=x
zuQoPj`_ot<lqdN6^AAJH82pQzE)UZp{iy)%BSmak5S|esEt`S-9>P!Vbl%KGu5dIa
z8)9`d9<9sKYeDefE;Qcfv<(u#)BQKJp($AS(2W~L$X|XrSSH5!A!OAPS1>+I2`2Bj
zptq?h<G%v-L5ccfI=M^=CiDra6Wb!b&)i!D-p~S+D+3u}+E;6?)e|;D5y*SnGnK%>
z8792L+eB0gOKF)Qzb`+#lW|(}oN#$Zucy>S_>&INJM!)k*gB_oNFq?dPC6x$uP=9>
zYW#Q?I8E)MyLSu-wFSR}sce(Bu|kkf4*|MBoZZe(ewFz}2H*ZJD!k@i3&wR@2jAZ4
zzm)5!iY6Nj!F@`?l4wt@S+b&~5Q#L9RHuM&UfUjAzEoH6Jo<zMIlAi2>$4dw_Jzq~
z?rw4Zr7SB0rW);)%O`!ex*X4KSo=64AONeS`YW@gqt!^>j!5_w7=wPl6|re2^!Rb$
zf;14`$I41Wg7AIK#d05PniI=FyCG)^X`_aOwgkpy3*uKPA|VId`q#LYq(q7Y)X~5X
z43R~6Dac}H-Jv+jd*QAiC<y*2HrY?y@*uCAGl1VmwJpv^1xXavsCd~}=}VCMcDZ=V
zd;n>zBo-r~H;c7HzZ9&|1J5~1DQ;H%Vz3Kbo&COD9*A@U3ZlgoQt;acS!)7I6|EH~
zbwgM5)%*=9xDP6pAQ<GTGSHPvHojWar$!{n%a%NRt89P<4zgk2gPm5NT223LaBIe(
z;-0*2Q-iQ*FV6^DzGtK2T)*HlC7)1_rrQskun#1n-%b>!m0*g!qd^iy%qD}BJO_2z
zb90OA>jw{j&Wu!o_tzF@kB~C$wRqVJIhkD<?JpR}=1%FBk??KAU^B|C)0qaLTa)6@
zC-#n#d_N@hQZYMTboETjWA$>C<xQ_PvassYE`uYx?XnxKWK+kuCiPik+%F~>ROc9-
z5b&wcb&xOAzDCH7Vy1xycril-@jXQ<gEV*A%bf)dx3__wpBx-a>mC^JYZI|=sT;**
zqRknC3|ds9VX$l&^HhCq75)KX^Zf{VU!2dSFCiGsdA>K@a#e@x<M^5396=pN!I{p8
zh^$fa0t%gV`dIdzYB(aO>xRkL7W~zr0MScB?E#8G3y#Z;+dIl9v&RJ|@Qb~!Qr1!e
z{UMr3z(K%aK+;cs(X?n%Tl|$_<fj!;F2z5HTa4L-KXq($QKQP&J}fpMyTUXxc(%ul
z)xV~r%GTHmcHD$u7Mz*McQa=LGKZ?N)!Yp#uTIPA_47whbNl9aSne}7W-gMOm6CS5
zfY}<(A<Xrm1CO?K&r4O(;#l&10~8FTm%R8;I0mxQ5beVd983G)c_8&O>2X+S8JMOL
z_)LiYNBs~4Yin>D8S+Nj6hZuN+NBuzYD~ioq|MPqbONb3Ei!`soX|qx?9M<%+dNNQ
z)v}`RGf5*gn3JAueD2qdD+TL+Zaz?!cF6;6s7AL9zet!xawuy>=LqNgvuo15G-?>M
zOE;DzDjV9xl5K6)q_V{xgmLhBk!xJAjITx*DW`v+^Q%$c?qsgN+1K#UKOJ1%4t~_i
zHSA?`-e|FY`6J8g8x;{t>62fx+OE}D;ow2_oHUHxswrxnT)ZpFxgL9~GfjofzpeX;
z{lFi#%-cW{eQsjHwm3#BFGj58?uH0gQL<KrG)R5-K&9OxCSzFLgAk~EV&6%@znS}k
z`})HuXnFw}Eg@iz34=EJ9*3<l{ID{y2#bbDaJA54F)$Ux(O^aX?r^=FvG|8_!%npr
zVu!uLG30dxjM;=SISPEEP7J-i_b4jE9k`QpzI(~~w=FQ@a~0!Gi%B2pHz@<P(!QSp
zunQ)x;~fu$w-7{)sj!j|g`n!T_Rt9#tTo?3ehg`~8z9LI6hP6oi}dX?5p6P81V9sa
zx@?N^kznj`D@vb)+YE|;$I)-8c0p1dfc@rlZ~v**ce=H(`ZbBl6A1<le3(h@kxLI4
z&)X&RG=a^<0X9jNc;0g_BKH5bMrME7w-K)4WGjRzyl#4Tkd;H?Ag+r#-?;XwAwM10
zt8o<EHfshObWO!#IW!|<e7pE*b=eak9;bX6(d3X3EtHCi4<lomT+XdkdeCXqVW9DY
zca-%3UQU$^SebA99Cbx~*brtvrjFRy0~sBK?8%k1vs;1wE`joLk@(`o8-n#bz(e$&
zaA}Z%t)`G7&8I(OdKl&TN_roR-pau=U9&lRy8?G^kF{KEhpc^SxJnJmQsw!@+;=UA
zh=zd5jT;UFCFe|7Y%q~U9<AbJj|^p5v1ECmPVz07Pl>--CmTH(E^nH*tTMMymlFX-
z;mL)Z1~!8>X(a+z?I%+nwarmn3<2L-4F4V&?-cj}CG0J4qyU)8xzTP>U)1KR67MO~
zg)zi&nMYzSUML}=Qm{h!UNXSJ#so7EZ(|Q@_Y`UWj8kW%C4*2d$T)8oIrN#VFF4K@
zj*&oRaN=8>F`h_4-+}zMOMN)vPh!hAd#s8-;AM367z1jI$8BD7IfGN*$}OlyN28~h
z9DOGPp-t8n1J9E4ef#id^{gVEDf}_cPw?69makGK<n}SvTA5hU-7{6S!=+T<IvZD#
z8X1Q+YV*bm1Jx*Bgr2y-D%zMRG8;EpkM*$JKaZ=V?|IuZeCW7HXm}{(bN%H7p@qKW
zy6JZ%Si|hY)2|69?f8X~d8v_M4!^VganZVesQo3h@krMDTkaU*1e$6=VLuoIL;;8H
zr@be)s1lGA&#3k8O68c@&Ex#pAUa-;AvfU55|1hznomCJJf@|tT}Cp}?rF{tlrqX-
zl#k~$)~Gf?$a#1k2eiCAnH*EGkqV|Qsme=D5W)U7I%;O^eT_g~@{XFzo?Zxhv!A1{
zESeGmk2Mh>v;rqj<llB2Qu%|SCA7XfN@f3>XxoR>Au1k^QrJLkGldsQ7hFDMf-|OJ
zEHy#bpS|YQl#VhbaH8xGnX{7`2YUWMMqkCN<r@vU*QY{2S)uU}!l>J!VOaa&y+QF;
zJv$q=uxku=46J+x`<%k6f>W6~4)v$B{t|Yp?OyAF8jbjKl0Ar#d2{ojk&|PSclBkz
zV*_M%kF^QP;nehrx25m(2alKlz`4w1jx6bX)`9F@=Nihi-rq!wUMyGmI1gMNH%LZd
z2Nh#>5Z2q(vkT6Xwtuk`igeq(G)1>kS0S=@&oeV2Bpt>HEo}e%So4(Q9Aeb@M4!a+
z)<a%|M`mMMXpc>banq@n(#X-U9o70r5-rI75vsH{-m=AQ<Nj94IQwkB%Pq{Xzs{oD
z1)4;*6Xp*?J5Zoj>kD)RI12gXl`)4AL5d4E^p#(3mz)S`l_CpnC9JR#pdet|sIx@3
z7;^y00?ISalyCY^m?sLi<!1P(t^&bA8xnkVh<9DfX<JBg-58r{O&f%u1Z4MtbB-}L
zM38uS%hWbY36wC&IjK+0E=Q<J`iu+-HEZ4GE<Sk1p2B4lo7HmVV9XSDGr^KCo26|)
z#%(AMP{2Fe?;&%^|1|AW`fk?~g@0Z~nn6ry7d!Vm3DhDkkJgp}*OunhlkaYCNYLhW
zn|bHX(h;RxWHj@&+FR`S8gcCIuo>~xxjK&XnbR#6C;e%z5L`1)zr(rdI`vgB7Z;@T
zlYutG7pw_lomzCfP0x?+%?b_7y<>%ogR{twBFkvrOFH<1vDC0$>6ZEGyX)m`KIT;{
zURcr8!GsL!X$AB@r!yxFBVVy5?J2v^PsjG&Oi<T2MD~UI3T$c#n1H(3&&b<NvBg{R
z1Q#ixh!BFldQKkjh(Kh%a9u2u+Cm-Y35xxqo`>q#%t{<kBK54RHzC1nVP+z&Q>56R
zL|y_-?~ztCt<3tJD~Iog6K!Q@J1T3RiPwL2&?c5*;!bYlKH3%d+imT)CW0)@vlnOI
zUc-h9oZ!r5OW`TS(gj;ehxzhE_O$x{5X27j?QSHKvInm1?nAx_vPd_@$SD6&xnWa&
zyfcBTf!(tg)iWK3Z@{8^1Y2h`P~u~!;iWec;(5VHVcF)}xS!}k%yZ^H7Y_93nCNHm
z<AG};I5Xo!Urm?5QEYh$qeCwoMK3yPwtvLFmQkaT@Q(!%B1`oi<Q`@;`fjcb+R4{L
zS{~~+G1}+Lp)g0ZWi2VZ{`7JCaSs}hvbLIs52VPNsMa<=^DKQ$DEJ03YuREHnkt@p
zCMdjbiL~LdZ|CvL+?l|(y2aJI;fe{dd@XTPCp@E=%@{OyqB3l<Z1(;0BvL+HUmrgu
z8`yT%?NbXVJBXFp`8kWJfe&4|I=0h%Zj)kJwEE;4|E;u`_a^z^HfB1=KZ|ivf*lQ7
zz@OA?;Fh~Q)C5#=-~U@<U;jB4m~o507~9>Zf#GoJ%A<@?t`jEbaM9kzcK_6Xk>i4C
z7QiN`aW7;=TTLoW^>P0~^Fd;~g1`7+e3$#QN9h-&)_gC{0@&uG_X2du=h!bh6EvP5
zw`FC$nlaa<#&4h!7~3WWv5ZgAWm01>OVBt&iln1}XT)q$G$kJ<=BBsA7SDj?usQa&
z+t8_~w&miHxRCp;*O{N<vK9sP8@edI4&|j`0&j6={;NDrL5fo5<?Sn%2~;jXrhr^K
znMdFu=%>K3q!!@`k={b=fsyn#Hz6%tCeU+i*vI1l1#Jv;%a7p453d6vd(Z4Hl_<qy
zr^jSxCcuWt!<<>WPuye6tG#A<LvUv_W!TsFymJShqog5U9@@w_*@gS2{FEI|e>4pb
zU}-uLf1@V6TK2KP=Wy2n-5NOrZ<Oc0rr|1>GY3^31#?>wCMU==N{eZ##W{5IS<o5}
zVm1nhXN?F>Zpb&k3_sp=$`vK(Ol3B@_d`Pai^m+dB8{Zc4L*QhIU$wFJ8W4Ru^2BW
z#`jQ`&bn5P)t8;o{>?B_zUOi7eQ6n!P*ql<5=t)*(`GIIk};N5fME9N`;mWF@`S$s
z&6H29-<+8XOEp%v!Xqq<!I8G%rP+QC6rJR@nBWjX74Q4CE4LA&E}%+Xw~dZwtEx^g
zs%A|>+jWDnI>m*?jm?O&r4WaqHuzbR3+xoEmt7wHt&W%lewdzy%Bd#s-J`2#jfZYn
z7VVn7M1i~-yh>)mmS`RUuIUutAnpt4X~fCM7fo7z+ChT-T@FQeZDY6UM@mx~mQ#)M
zV5c3hi%x#{m6e;kPO{*|{BrjU)Oc)dT+T@Sd-(}_HECI~n$D6pBavCb`B=mssVXT^
zv1Gr>jAG#7DQXLEv5v}$M*}kXZ^LXVluF_2&fj@4#+s3rr65hRmwImfC}C%HSN=!l
z_s62tEk!iFFssB44aL{ea<Aad>Wax=r~-1KXT4CNUv}1lWz^We{e@6*IBOnYojT;3
zBQE7<we_dMKBH$f*fGv`TrV(ex8FhQcq#IvX!z5~K$+oUx1w3|Vc~p{mtlqsySk$u
zES2xxO7#8T1}si_<nPi78Ks&&UNktP5>^?R<Db%>k5$Nho0;^zDQI}*U@YQ!@wV-1
zga^Y>t=+QBs5I9!3?pP(ij(K+iwxuLjIjikhz0)h1=N9V;ze>>K|yuZrme)Cft&$D
z46sW}X2s^ZNPwV3y0Hp+Ch8f8e-_r<sT@yOL&*pLZ5CWixFLmzh-7(D-#v>kGZbB?
zb>`EX^{*yctg(1i#k*Oign#cJ(C>Idj1uy?o6VcDF`QAJ<1?52)vcnhlx*md_lLzP
z?L*}fHyK&>6*6E4OK{UFTcb?UMxxo3FgZK0=78d=q2q(%nSoyH7QP{rnC&(x$$pe?
zWH$OsYe}P?`n$?&yRZ#1dB4O4O{^Wi0A920`&PmC(^G-CXI%M7k+alC%RBT^?BRl1
zmd@wWKR2xXL77J|R^q6BWCDZ34f65dLcT?)>HpgJD64@CZ|*<^Y0R2e1O6~zLDcrc
z8K*4uT6~@3I9S<|rO5dsC5*B8(MDV_XGEUaaNAud6+<ZRh9$tn+=}l0vld{(1Xbl3
z%nA?1(B+8!a%=MY$lK`N1fQeo>^+q6hD+<&^%0d*Z40XlFTslCOshyCXH?_#J&pNS
zxbLlZFxhGzU%XWd*7d{Vc&aK(s%A=B+BG=TVrCIBbwCwR40M;>J)OwIk6><}3ex@K
zAjaIU(V<sX8iTvuF)3Y__F1#KW;kL6B5oRN30YY^BI&@XmrTCoj;pD2-1{91>UZIk
z92z<UIPWC@gLZv`S8*Gw90z<we_q8ApMn{T!dn-pMuXTZp$qZ5`AG<)iFE|45Mp)i
z=#TrE$fHE3e|}RUO8Y@izR;^W`b%8TnuK6~y0L0hampQ|r=^_vV+^2}v&=)+Tz|_?
zkpnv~_Ob%*^}uN2;aK5~GG;8fjqTU&;ZybDOH_YKXrx%Mk;#0~PmVuY<;H>!2wz~O
zsTvTM231Wmz<Z<-zcM5wj605SizOMO5hSNN3}_khD|v0fj>17!%7F~{$R3gmiL|!<
zpr`2-M3cJ>gc8{^<Rcm*4rYI7s88+1`oCY4gBL6vi`OM?13l_-P(1qj3nwZs(Uk)F
z@#pZKAZKYsk#du<Lt;jlA#eDW(g6i@4ds^$StPz#uoVhL0g-^r${2hq|2V>n2=0Bp
z>!sK<>FF&eAMci`@x8D$8c`O6PDu{fSN9aYL>jC!g;HN_W~^FTcku^X(uV8xwX};8
ziZ?tQou!*T9l=w_#<7r(HlMP)S6x3la?BV+8+^xzV`O)>dQMH&>Y6Q`)d@%qvR+G3
z(FPy1Va;CFDzI(Au229j3cahD4tk6k-{Vpi0pCd<gvc!}1RoZOT`;BUUT@iCrwxur
zbBMjbG7C~2c`v_O<OuUzk|v%sn*>J|E6Zj1-n1UeIB{iX-9>VyO_Pu8BySX0$@)fs
z5Ic&Zrh(4EYVHi`E+%*$T*lEmxjxynSmX-kz=Ks-lT9<{=~$3Jz3TOsD=bU@97}vW
zQ^IrS-Mw(xjT3H`N@^m3naS~V>*?V%AFcu-<j~&PoV=-UUC+?yx--*^pFp_X{47*m
zxE{3Op!}DlbR*lMJ%C~>V=@6+UBsW4->)qePp|YmpvYh*aNrOYCupimpwH0rIE^R!
z=S-4#-?c$|qhp91LDoE@i<he9&TuFkb$rI+3I1sLr-;C|Da;Q%H#3-BAiecdABi2m
zVG&Vo3{nCJ8mjp0xb&3P*bvi?U*ckfGVPc@sDeX66jcqiY4=}26@hH^2;|@ZSegrV
z-t7+%y4y4pi6Tf9?fa(I+s~Si(dQ#9nu;i=PAiS9R~<?AaogMmgHI?O$rLbj6&3u|
ztzI`Uhh-Cm@H-j5@+6C|_R%&|<PWgJM9$LT7<ea|<i`pt%NQ>LmuEzEt*13)O&GX*
zD;DMmfhxv?;#v%N8?s6QG3;^2c(4LD1$3?#FjH=k6|q_rHI>cA*+w{**k01p{7M|^
zX*{bAr9CRrlw!Gf*WiI$T5>-;V+B5GDH0G2J%;cOv2*Kieg#`$F4^5Tt|6w^D~{Am
z<h7Lf?d|wJA+y=kq+k<CYj{ZfrL(W0$$qB@UPiL7LtPoV<sAQ~&?d@dY|hlTAP|2e
z6>tp{=*IX#Z;a6w^dpPG02n$clF@JQQbU~>doPR}ugLPr$)C}rZPkm%awuplo#~iR
zcETJjg`=M(e##XW_o`Vh(ETY=s7CLco6&6DpM4ux`DRS|qq(}2rIxR=?-y)y)U*~9
zq%he%ds3v~O0?4p`O-oh=-BS#t?}yM9{?x=s_Qcjj^jle<Vk_nZyGLgmfzHNl!|X>
zBaWUei>yVN2I^ZIA61vWr|iefMPV!Z|6Cpwwxj_qR(rx*0&$s(IpI>{LH)83{SYp5
zdk0tDWF5$fmIsph)4k1mBJp`nArt28c}^$a#FeO$nU}<B(8yTEE_glPS$D_|7)ZMf
z1&H;dZ!F5nw$6AhT<K*753|Vq4mpre<Ib%SmXoF84e2V&8tp#{*SP~yR8S3pF-5lb
zJ9Q0+^zN_qv|fS_M3hh)+;~dr*aUldtjU5_!f=JN;to9%F^?t(KgRPW*T5%JSZSvZ
znF>8wVT=fj42AA}J!jYmUL!-5tgNkSNq@wEjs+UgLYGH2(D;pDf~2n}Lm_9kqOr5-
z7ieEX`ewbP_7|-F5qT(E#vYvCZ35Bh_vm{b^4Z-z1;8xhdil`tRl7xDNt-Q~QcAc}
z_d6IBNg?uDJ|*kSH70&7jw1#KthH?2PbQE^=Xp~yn0l0rbF^nE#Iwd>kvEu!ko#2O
z%;-f;E9=HMo0wJ+X-Bz~YntWeB5G%wA4SV{eIN>W=W-NF2<;OpP*EO9vf)KgngUXJ
z4RiPjRQ5Rnx7EDeh++J>ZupwrYjLs(T}Vc1tl(ZCXQykCmb10D8v``hYuH6g5elGF
zCxGc6mN3h#!`U;-RgGL=r3qd`P(JX`OZj?P#oH*{{(1g|cJAX7Z6QLNl9-)7>tlDm
z+N1bQL|ZiWZRj#<;eo>nu`dW5%$CI-t+!L9_Mp`25W<EHdYOZzzZag&9{h+0)wA&6
z=h<PC_L-0^CDB1Ja=fJ~pPM(aqH(2x%{`Md)xxOfR$2TF3~^f3qLM<(Ct#87^*5;I
zaHDG6LMTkUswVAWnb!9Gofs^nHR2CYi<W=HQDQ}+FQHck!T3sVjTH}0-_unw7yIX@
zI(&GtbA7jzK)eP>&WxUqQe9-lQe>Se8DVy1%Iyn`d7;IuW9^$$JM8J&o;AUI!DDx$
zP2gyKNC|i`@dET<uq9NYajq|;hDl3N@X-kFSW99K92HgSdqhJ%7_&c=w4sfdWL@Dq
zpR-D*e@R)eDx;y^Pt<WvQ7p()%376vkXOr}HVJ8&3{mf9sHmq6h%SaQ!5B}~80?$+
z%IZrNh#H6zBZo;VIKF%_9l!a#AO+lsrI>6DMKJfm5MXiUSV6YLmCn)k>Vx&1p8Gq%
znrcDA49akxC;TDX5BG0DQ0p~YYNO;Q_U9XH$iF|DTOKrFnR;@AF8o8g5c0uMJUMd@
zANeqF42pdmTt4eJ^cF91wjTN%Yupjpw2(62Wc2=)AFMV#neuiSddL5U&Oqp?_~_JX
z%~>D`1kJp80-fVk-knhSdhRaCuswOYBVqBtVyx)9{Dvyghr;-#4>QGa-7cX|PKf-K
z>G_61jcsM}sk)1-_h;1s@n>t;uL|%20$ngYEk}<wjSSv*Zv03uptTm^13{d};!n!D
z%g=FlwOT?9kn=jhgS$euwi(AVl_9L6vVbOcZ-Up4>xy4n6(=UF%eGrb_!Dfp?w9l3
zJ3cui=<;D<f6A>e-XtN<KeZTD(c90jxpjd>sNd-y<Z=;Z59UIWeaEE0;`@2<JUo7j
zX}_rGekGmBj}mdA7Tr0li48kX@8-f^ydMLeXYOR&))%-&Q7JmYYiGZjeJApERx-7k
zTf#i54zchWr_Oc$_{aIgt^pX1Q?)&0r?cBHHcT_JsW_KXKhou^@*`nc*TcJBU(sm^
zXjmCH+rF9%Mx^`01fk*Mmo}yK+BsOJ9_A&CuQBo28JuFUIXw}ZN)O|E0f{1DHHwv7
zggKi^Y(-u}sYjCU-z$r*g(Oc4zsK_W0TnIuR#jfOyr6SB?nbqQ&Si;Lq^J$jm6Ju4
zBtm4g$JaKJd}|t{CfZ@rZvKD@PXGS!m`5K$=?RO9K1keqZ2mc~|0f$P=Ys-q&NLm<
z;yqwc`dX?&hd9diN$8!0F%bb=&PpS=f^O!GTU5$fw$tej@Y;?HYcgL<iJDYFXt)Pg
zWKcfuI)crh>J_=Yt)Jr8RI--@=csp`<acnLR{I$U2gm*V1Z9oNVxA=EeXPh1drvtR
zWB=A0_r{6&ez8%Rjdktr5=)LDDN{xV!G)kiBYkwN$s>~1slJy{cdD>iHS<HWr%P})
zrTH<$N1V!lB9KoT(jlVU-dy9ej2>0w;ty_$aD<2<b__$6UO$rk_8ab%F6?f-X<j&2
zvIsXltDYj>pU8cqEA+Jq6XQCLs~|SoLTB(}eQT=I^>Rr6CH+JFi+2O2uW$?ZlZip2
z&y=@z;CDiqk@O5i5b4OSY7~sPkGpIKqz?T){1<a}epdnSdz`_-{V)9ZJxJ&ItS4Og
zCJzIG^U_u`gk6Q<)d_4i1l~1cE$Jh29Z{MBmgA%BQZhw}>A~VBbtUA_Hk=-6_*wi_
zY?ZOrKw{btG{Px;jdedqhh6pg%2Lx<eOy*-uV4N{59EcJhN)D_`<_21jHgdojK(A>
z!HU$nsi-AjaH1kB!piHD%C@~zBFk@nqN>tLc2D-zwJ9dTS8k{d+UC0)r@JR^Qi^**
z&KXh3-f-E@{Tz1IFajNtiDUD)ZDi|(#IvZqZI2YviYrkoI0C5{IZ<~|LtIIMW`lrB
z4=%UW9rc7IkdL8%gpD2n1IG0qN&<Xz+f+L2fSr}k)JWij(+vJFBS>E{s%$E>gsyBG
zHEp$-J|y}rYnwR2KKqh;-j2&%Ll>uBaNOf=53BvjsB3GD%nWsd>Ma!L1tp)fu4kW2
z@oy3&jz8x7O2zu`i9RmL_`KXe0mi&MLBiD$<HVm4mIoMh95b4|lVw8X;KOHDf&Tn4
zuCQ7?Wux%U#3J;Kg8B+X3$ZvWifd>H@^kc^6AiJeU2A8pUI!``qra@}rt~g&VsFC*
z#@O3@V8t5nkdItxW4o$7yYdSeeH}LwklVnfOZu6GPPiEvyDW3rP6MDg{emP4-Y566
za9xk*B72b=^qTFUdpUXaRws#qT}o83SOq$$n)Xw070|p9bue*jv2qg%9_^ST`%N`Q
z$sv=MW6d|)e!0=4sjL0T&_}iI#@hVWhoW8n-G7l^wmbee*Td|Ll|V<`C(4<1%pi5I
zg(+2P%h-~drp}vd<QZt-ly$W4lK&KqHv^7K9|!}qDZNS>VEH~+u#wERBptf=@aIs{
z>1$rQY((R!h{5phr!mz5Qw{HKT$RR$&*q45b7cA88K!Y_2WY=g%O$@nSUqH)e1@jR
z{;$=I%Ek*VZ*`22qFxR`{jX_n;y+@K+EmvVg*T?AS_@1rS6VBuNqhiiGtM9;us`}X
z5iccjj3rD=^gZ7tD}s6+!s?%ZyS4cYEA*NeYGL5m>=~!K$``vQ%MKu{POpiLW&wyr
zC&zaD3(4d6br*pfyCG3+F;eW|<~M2yOwIn&jYfj-{TpHDB<SOVR<x#)rW{YA&Sz*S
zkQ;KFKcZs}RvmO&u#>NM>rS^y>tTh)fwa+cErGu5<(UnVE(?w(Qc~lgAWeB4v+$mQ
z$LkHI*@cDnI@P0f3dkxmiq0m&tt_dFvR)9JD_^(WHBrV5&c<QL4ZopXRX$`Xw2;b@
zg=ZeNnwXWc+R;AeHF;0VHI&Fp$=@^2Jk7YBY-yNXY_9N>oL$5x^A+&m*JO->o=*p*
zCi;AM;*ty(;T^xbFQj}rtX_$(58*wNNR9e)eWYpexBD9{a2c|ULAuYo4mkHgou2mn
zq->eCdwxmIzy!?M4q&!LlYva$tdnjH_7}V$5Gy?`eyPF>jd|lsW-P@Ke5AKr*B3pz
zTCMwmVb=`XJ>#htnFeF+Qfn9Eq3DUcFryjrTlJ4<A-Aywq~N)KM00uFxnnsJO6tOV
z>rw3jAv%X6o#Q-I-aQpo46A9G|5ibJme^l(F}`yR{)~`DB}F#eHc+B?<1wygJk7F4
z2`v~m1hc1<o)sb&q{8>My)#hj4Dm+#k&);_E4W#Ep!Z|lWL4x@AM$G?is&p4Vi@DF
zN^2u3{bh1jiupJynCnPOvyCP52P%erP3qTphlq0s(Em;0N=-K>2C|>kxdNo$lnjq_
zKzG+w&<q6E6(`vSaK>6)szkjcD|!b=(q@7HJwsoQK@Oc|(#K7F`r(^f$s+sI!R%Tu
z)P&a>+;h5-e}7ewqvR+{{)-u8Yv15Ue+O`GH?Sr!9)TB)YL75*1Q7|v0lRowsx1N#
z-)^2t{S8+jJoiH%N|>&rv1E<q7i(T}rzPp<=44S$p@%#28SpCQ<FSKr>D1B4zU`V7
z8Q#S5X$Oa%n=VfxwOHCOG3=$UWMd^}!_rb^ggWL7b*hJ*5Y+v{76<K>b7uq<&B+Q{
ziN9Jisi1nv$_&E?Y<NNm28x%j*4Yha{1K#dEx{SvRhswgSXxX6Me~w5pnNFTIxP}1
z<YbhdwhAke)Mwju<Jeq@Rqj3l!iLBxD9XDA9Do1FEz@O&%}RIf>Rlb8)=>lO!BBq2
z1H5Eu0!ZA)T6HfyPL)MbDZ@4B2xXuRuSY{!JO<#Lz1BRvLvXmqK%y4sl5YE%`)g0C
zWmujd03(pD*<B&Rt}u?Tx}Jn5=R8tmvRREq?n;BiM{pohmwPc>j3v@w<fg121&B1S
z$f-#Ba3pROYQLee`UJ(sZch-P%QkKVR`r>S-rkKYlki4u7GBk}c#Zfx|0Jmy^^S1F
zm2Dbtg8*9aRb*1be(L?*2mMmcPrF$8JP#rp7S>p{8ItpW1UztZv%(m2A(-Kj`WhwU
z%jCWlL>9;#)af7xj_G~Ryg_`~uK`C1IvPSi;mDy|_kM^Yj0AsQTCt&1OI?>N(@4JE
z(}Hqm_o|B}_4J%Zz!m_5nM!Wt{BugmcO`|9+`;)uTnZsQYV(9_DMtEwE^Jg~U-lOv
zT)>AnGM(HDPn~;(|CVsu<I9|3;?zrf`yDIgY$_-%IyeSt(Hv5CkAY&O^;(~B!?n(D
zUs5+H1XHa-rn%yf@U@C{*uKeKilI6S#L7_-&F&_Mpz}CblB3r#_gs^^AL~~V@y17x
z(OZ=mXIeaw*qgN_rmZDeXl9<&-9XfC27#4d*KxPfY!G^Rp@^tFpI+_gq6C75uVsAP
z*aooGa7@ic$87N~1|6=jmPe6(cziOM+G<di1o`=qsabRkqjiIy>&v4*o_vfc9b`vW
zabe<=M~zLtE@)6YzCm=+lDbXUGg4&GcFHqe(pS36ifp#cg822)#9xHA*|Vy5$B45s
zb0&0OAAEW~&XWj)FD%e_l{`paJn5xFkClR@4dgO_jT*W*T9759<>MSd*R6aTOn7?;
z4bF*AiTHYx*Z8<Hi;{j}=r*bVJHWCL%8)dlC-mvE%bVq5ckrDDsj&SG8YmbhGCSh2
zdYC9P&`WF&6V;9e+4yOsLiH6B=<fJb28w~1{-tPuYu@6_)JP?vIB$|EDg$#Hg24jH
zF<TvKZKpesG?W=^Qu9SaLvzaN3E=6s`So^FF~M(8-_At8WBZ|wg5M5@3h&eRZ96%F
zmQ>#A>v}?l`7)J)VNa&NnvQu-ri{Q*SQx3vK|hO0V<})ve}04k&*Smd*LPG%^O#Y4
zL^^_i-s!A!<WuU7P3AptIRL8(YqCXw(`8t!ON)R@39&_lR!e^6LCMlt39*B2S)3fj
z)dE1l<N&!f5nenUG7q<$Yd+TKSVp_>MQhA6S>lwPf|HLHvCZhL7}Jy{mqUcq*tc13
zVz*Qf=rA`A7J)z5U`W)Fy$t#L!QA|Qpq9%LGPZT-l{+$n>zRB#-ExaNNU#k-B?2*X
zD}|pFZ-T2Ee&<L$oQue^(3$606Q{cf_J_v9Ld$#Umpy&XwzDi;tSW|pDeo|V5sTnx
zbuha_toJku+#lx50|19xlf`K5>o!dfxT(v6wI_qRuqi60$!%qwbR;K19wH+MRE>E7
zo;*7{oiD-6)&{4fKYVLq0)BTLyTawj7U(@O(|8KT6Cd;zk2k6xnp#|2J+t7i_x-?@
zqTR>ukSg-sz$QmoeW_`#IjatgTdXOC8fIKn666&ct&K#AvyeL}W9|<DaJHmDSmV7M
z-p_uUT8|1=2;?4VJ^u>j@$0XZF2U8Og_D?GWSH`EqtG!)Aj?rn2gmFD28g(&ha@Lg
zB@-@+#CD)K52{#1QYPMVjl!v4ifVdN;kxq{m>KD`vrPG24YlOaCXa&t#NDOAR@K)%
zhk-%2?wMTeIqUzJT76elQFT*uoDI@)w@uIo>jIf9<2&2_Bu@-vZqRdS<l94U^bJk;
zRqyS%w=Cu<KcWA0w7&run70h@SQLxi)ALEjEA)KAs$i9Z(Ut5T9|;;k-4#mh0>{|r
zg_58GSiC?)6KL-4^sBPX>i`|MT0P<9zmakfu*he{*Q2S)K}QAO_N(-DC*k;@rjWFM
zy~zG>a4}9?5HhrvfT2z}ndG$9bF+EP)pYCFG5aW<o|oQ@`G}}Mol9Z^OPF^XrmgG@
zAK6!!E6VtZ9X1)6OML_Ijfjg_-&6Dohv&8nh8{z9;Ks&5@~FI-ilXbxPvGNgJG+l7
zM8=!v*w6kgWVu7*8$&L<!g?U1F4<DKNq-Z^H?vwaTgO|8A<^CZPDzu!Y7EMn-CLuT
zv4X%Nm|};43c}@OHPg*F^PPrq57m~`pG<1qx4CMq8D$V~8%(1qgq6$RJV1MAVq)Q*
z)lvLHAOzV8`-6b$tNG^nCJg!GGybB@LBwN7ys9|Kdh9`JbNRSIYDutubtCe&m3s11
zKe+Ri{ochkv>a>SZ~1k$2}hM(@4-0C1XFN1KUb1&<H+OaLHUK3LPCHj)-4LtQ}IC^
zKCGbT%JhMUeZ)QN>Y_8ccl*g^<wm|aEfyZ+jX*KMf}RxTcq?bZojdCH<nl_eQJ}e~
zqqts0sY_k;Gw7n^k~36tS27-6U09;1v?WUS15SSs5K1>bkWOllI)Q#EPE*<mzj{4-
z9gqmi>B_^%rYYR_yLckqjKb`m^_RRZHAz*O*r3OSmB(=n858E+k?*vR!PM2au%kP~
zlohMosbpcAuO%J!j9o&RoDk_f)({x}FP^RV7Ydf1GgCtu^{Q&@llt{fIRqBd<C|4q
zWXqI$)M9yQx2GuDU%%O*$d2S#d4j20#BQEZ@8^a)FX(RP7pKd4%BFarIsedC5p2iO
zgpu#s__Tb7D8`LKv*gMcH4o6t@I=vc(T>Gw5u;_omTGGKXh7+U@JU6senP0{L>85t
zBYZLS8U)BZDp5dWM^TWTNp+Zg%%D%vJ0m(g4LE?~hO?$ad$Q0(4fjeySbtHNBPAwW
z;-XW|cw5D!_LSd)cGES<(Z~KI>O6Uz|MiQ9D1k|Iah5ZN2fj8@7l$vQ?`H|nLA{qD
zd@Hsx7pFhDXN9^~zeFy3MsXkn$rOZ!)ff25$Jm54PNOb@vEgVvhY`V<P@;VuheWKF
zmvBf7c+U*Q%yPf>V)mVsi{f^nyAoi8$4^`N1F-IE0_Zk(*{GUa^oH)B0VpXzUU^js
zCf#>^)Y5=(24QxYwlH&4`nh*YEzbns$mQ-%1Bqk^RAXlU99&H9VjF;*=9%DUQR(j9
z%w80ek3rtLhX?ePq>vRjQPiiOPZ6zNH1L5&PQr)@zP7j8VzcYS#bOv%Z;U_^=sUY}
zm+|IR?`j&FN0qelM8*i06;Y%Vv$Z+!e}fTRu!7IJ)paxiJ;~mkJEl@E352ev;M^_2
zbc{z<^5k~c&Te3$ADWp9qW0rXk?OBnVAv^!tLkx$SM%Mq&t{(P?OYQ_8CM$k&uO(`
zW&E9B4-O-~vwl8)P%7Ube&+vPPp?9Uae@5BT{-N`LeNUk@{xM!br3k`FY@N>q|H<4
zcsJ|67hUEw&IiH~ZcN<U=2tURft>ija&ElG_m<{bwQ3W>vO6S&e2xzuq)A5B*!teB
zN~LDxOhI+;CV%>zWU50fQDgbZEw+(M^cgPKcon17)rV17Qvwk~^GxSB$$#iB!Kstj
ztU|=Q+D42YPF7Yjm#(<C04XD3g3OSQ4{Sr4^dpUIogVG5x1sl><t0PpBsjs&JyNb;
z8-av!tbB*v=QQDDG_6NCQq&Sg!KNVZ(^Ur@YDyH5mfJkmJ6~IFw(05w#B{>96Ndi?
z>TPqW!fx9j=4gU4^t+e7_6;2u*Omldlx_v}|4kk~<rs>*#SpU*CC9h4_5@uWh`ERz
ziOauSo2kbo+piQRkcKuA6m@>bF2)sjV%9(_4yjSy{CgjAGDUETLoV=@phuADtx?+$
zOOuYEZo)!c;LZ6oa-KXZ4-^UYXffO4{mjDAX^{z3Y$n<xO+}cnCXKZ-C#b$s|AClo
z_gViV&YCj-vb684wpxn2N(zb`g?UNJoZ5kCuomK%#`j;}Kty2)E2gw{|7@fM>)4L1
z`}?k}4?0qDDtB@#s4sT2K1m?=!UKBA@;KQfKx>iHL2@dpNB+<Ug<WBieUgDoy{ybq
z)L4(Q;6Tc^*Lu$o5g^R)c+%m#W~K-8?V6oBq9>Io|GfG#O8#Ag-VV08=xh;>WIgqK
z`#tN`v<rDD8)D@F1ek(7pY3tXK`o1w*=CB*hNew(wQOGVi+_47Jk}o;_XR+45ImM-
zw=<A9-JVHz;A}cDB?e0hLyUMnmB%xubuj{rrpu^{fF)K|Vq%`MtI4rp*Q0MrN+xm~
zV4W1$0b(FWM2~w5_{nDOUr}TnUmOS(u4=ohlq|?Cf1?DEIz@p^Q3mp>vgu%la%Zb9
zKNorw@(8Shcat!__u@VR_UJ_?DjIC+Tqg+C1gb7@Pc4bZ$Qr~wS)f2E*Fr5wu~aIz
z99qy>xr9mfL;v$11e2s?8hr+Nk~~4DTWETR_xHy}2D9dFHqQ)H@CAZMz4gWBv$C%T
zV@{<AGBNO6J*<d$IP&FB#}vR@3R}jpldgvY%nj&~(=JH&skb`1l1#0XGQ{B?rosdZ
zt)D=T*30de%&43-eOnCxXrL-?O&(m6qC{wH8VkpqB~qcqfToWWIXf;s>kwh=P*?Zc
z>v&~Tn#s}J4~gK=jBf=3CWbQ`lpll#r+s*K<wU>iMO}K-rw8J3<X{)FP~UfXE_?PK
z%ig7SA8k=^&S~-E?mA27nZB1<1!#qop3~1)MbS^N?~Sh;U*x%hMsNMXnk55@DpZxg
zlhXwcW(M2{AEbH3^#X0-cxCaHt_p;T(Xh5C`UvRb7Q`k_;8EKOmxoK|w&1?llgMW@
zk|4omyHDbJhc-HXL(u}eRQ0Jx^Hfcsh{O!ySyXNG)w{2x*XDF>Y-QsEP_An^VRPu#
zB%~A>jnXpC)u?b>?q!@E$M-jv!IkAmzH^-9Y1P2iNd)r}QN6Hf*B`7Be%w(DNmW%a
z=u*;sM@0bq;V6{k&ZmC*eFacAW*<zDsX<$6X;Zp*sSO4H{{liRz0+X;Fl=5u(L4r5
z8N|9B`~qffsAN}PR$doPPTQFkxXF$Ldn=mQ!d9CT)I}n`V$YPpy?DIC1ZBwons#zs
zCeI8p!3NlqI6#Q(;h>NMx`}kuSo~lVc4d$Vy1yZ-3rK2RVg(M42}7sCizDwUkmd!e
z`xOZiEJ;}XZG{m=#tcTKMqd-~nx#Po-ed3@JD2GSef_rbjZSOZbGaOiVXO-MfM}sE
zR%p7`lf9IV?mYFd^7d;KT+A4J^jKU$hpK_ax`;SP_#<mco+OWFJku^uEBi$>aID!^
zN5M2wW)w0tX{xulug7j=-_p^Xag!1O^N7gtn9j73y46xwB^aZSC8!JzCvyod-yOX3
zzyZc9b~fl`d3br<T`y)oirPDf#=8XW$SNhiT10$0@FM04i<n*EIFZiTE#Y7S*nON$
z9*9ci+{n5aNDG2zu@<0uP1gx_If%3^=az0+YEU+OX3mL7b^LSpato|%)1)-N%O7cG
zNC{FP3|R9<aBtNyi>sqX7809=#0?uLO%s2-q!stS{mx9ZFI-<ibI!h2y2Bv`sUKU7
ztto_VFgmiyib(j#*-o~f_0tlgg=+bF(TWvHfv`C*4WN!mu#6*YAe$VI4Co9v9^ftu
zHfg?cGWF=(#=>y-hck7~@GcHHQ%rx|kbdU%*gHE>9Xp%yV?U)KwUir5bb1i6W1bu?
zvNE#}Ov1TyUo`uJxgezGfGuGXfFq|jTLw|dE5qo>`ieM~OPw(4=(wvbb&{{8tM#9N
zR}XTU<t)Syr+zy`)%r#cG5Q40%dRX#1*b->$=+vHyhD0(R!r-(jCbOV5k{xSzt5GE
zsMuydJQKC^nfMt~I`Wu~yB(^aRGyF%ggSPIC`>^X!NohGf5^|6=}JR^T{9DblNbnW
zuPhqeaz-%0o3Hs&Cr}#NJjxxTa%KMwUrHhy%)%07gT1TZm=1h!@{>9<-TyO(^*?Ol
zaLsNMgIIxGqi3y_YL0ZA9~Jo;k#cgx1S2Q7_;)GsP`%C0*xxzpsn87Vx#${LP<G9}
zqc(?b%A#M(yC8Uq$G^*$9IJRL2lHl~eN9VO5Q^Mf*JKWFo>1(q7>ey8X#vXX#s??S
zN#6&ySem+3gVvrU%Pb4Cu)c3-K<}up@1a~CZF6B3U}svKCuB$YO^mgm^9f4rT^Z51
z0`ORXLhw98AJG-)s6S||HX9hE>~yA91LRFNRvl?a3^`EnV~DZQn+S(06nNDwcAShd
z45~gS4&}0rFkNFl{mIoluqdvf=OAG?&n+V&ebvbO*}e)H^bw61LC;nMHop`u3lrL8
zEzpLU02&;0zu!M$CtWQz66xepD;%NJ<Z@dQXH1Erp@f6rO%Z$HSL@f>#XA7@EvX7+
zoI%4J>QPd}kf3&SJq@p05Z1z)x^{9prRacD<|TC>1IzGZP4F+EE+JGBmDeC*&0$|J
zw$YYS&uNFp15gFI%0w?D&ox^{^~k47Pbb>r5-zQ|w8|Zt6}Wc`c6v(O6RZs$!K}AI
zVAg)!14_Q~xgKnUk*wSHrQ8KCy*jNao@jhi?Yy?!BEyMEiMv>>stA>l^lpZZ!NCi#
zV{cz~;f#7HX3qQX2d%}n9I0gUTK7v-vKRFu9%0-sXfa_}Y(7T{NF6@f-idLKM9+A8
zzzV42wQYb<q=*S$1SP~{fuXtzpUacdD(KOjYBr3WLy%~}mH^wfZQHhO+s13#wr$(C
zZFj%6ZF}CIh*?a;EGD8hcT@XYCr>6mi(b0EsAPWb>oLz&H$8J6?YH`s+>lRxngv2U
zGfVA*><bGd;_02Z1S>7F#5{7HF>B+wE3dscdG^dL8{=Php!fkYiw&DV9|~MiZS;91
z?r^cAc-)VUKAz!23>&?s#;6pxAfhEi_~&D_xWJr0SbLKm@?2B{?;>QlC)0dU+B6JH
zFmovZeF<dfayd(#RVP~zOB*-qvs?*KZK<irEzl!bb!w4q-~JPe#lU*@N{(Bk00H$F
zx~<&opLI-<)5L%?Vht1hL!vp*Cex{%@buIzMj0-6o;-(@&i4>|m`cfdmb|?9DEE{}
zaK=)ucEFH;^;I1~=JpLf;+RK^6Z&6Dw{|1y<E^ugw@!fSt2UXiNDap#o0wge*=uWO
z5P)C0Z5m>J_)op49_w-$-1Ne_$ClQ>TT8z_wb6m!QvuFU;%T<qjbA86e>{irL(AO#
zs&mqF9WeM}Cr$!z^Y<XS#0C)6!b^0NtLtc4g?OU@PH~}orFLUWt^7JCpU3LEa|B5M
z=Kz7?mWCgY(<x1;2_rm$V-n7@3b4`>zRwP&GLlq(StXR+ZwNI!&nU#X_v+k>&pIz=
z8@SRcRMg;NuEgb@11|AfCUmw!g&@x6X{(~A1gh~tx?-RCqhB7(o78!aNrEqD<3{|T
z$=nY3Vr(pOZa~aaxTdgJ&nHOl(C~d5V@t(2^v6P9GsQ>$VveZ)W9|wjqQ`xML7rkm
za$8k~j)Vk1BKHy>EQ!X2e|8yMk8_Att>jz9(8;x3MKza>@u#tJ#F8u^7cEd(S|W8F
zU7z?Lw)N~%gL!Bsio=D>JPO)sUk3kfmPBorLYF^fsZDr7K)A)6HyC^0cYN$6^nb3o
zv&N3t&$S%@brP8Y+SSBFYN*ZMWX-KTtwVWg?&#(ck5%+L*bplwlp-I%$!Xs0U}$@w
zlO>fV?o6-QPTqQ);593eX6lw(!fa=7XbB^gzhVtWf%NfK$F(%RT&+rHM0;cV&S0P5
zK5F4|0`XYscrxMD<Nav3-SYnv0s!?{5fL^?m}9FKr?iZ3;|&)T%S_KCICmme*jOcm
z@$x`9;k~IY_l2SDE+n=`T>gxL0!!NhR%?MJ-mbK;KzBBjY6^8_f@?mijzfPHi`l3t
zB<4wPt$y+kZrtu4W|vsey#H$u>F#Jmp!^o>u<wo3({REiEBw_snFj^o?4~52R)bVj
zU2ehlHE-lFnKA9R@QAlybXR7>R$kIas`|nIr<Udt&4dU@Eq9xHYK<NZvX;IN2XpsF
z@LQ;v+~ToT=A&4bDI!+c?{lTBVL#0_n}sRf5z&fD$Te0;CD+bVN|~%nv;N>lDti^a
z1aeQx@9aDNHq+TSJX+Gxo=g4F^+_wFDV`<hcUj8ol{36LfWV$y;og_nu!K>cbBooJ
z6qdzzEO*ym(!<|$`X(ef4$>}X<G(Ee9#?A2Ciz_-F@n@)UP4o?7?)gi)OM=xz?m(`
zXxNCUN;u$bFWH6c>EMO|QeFRLxy=rt$-ygoO5d_}zT8aw;@KC+eKk_q;(|bDhGQWg
zn9hnoia3x<m9q$9+|fRW#~TC?!C~DBWneL!yHEuHE>E@=aTsMyGc(`EQ@Q=AixX)3
z`>PpBJy|LC7LijXxEv>?K{mq-rNo|n)NJA}sg}cf;_6JC6@ldu(ya?j#Q%tm4Rp#I
zZGGxPZ4bxjL}}+w_pBV|+us>0+}}Fw8#jX%-vjmuw?W6LDDm^zMH~NqKjl4&qpOYw
zy`+wdK~D77b_svHUrNP*2!O2StGMbV#XQkDc9B*qEdQTOn!!`?tTue8jOPp)PU<I1
z{#Zel`EzWrt@L|{G{)oer4g}5SS>+b*BNziXT*KkeCKC;Ob<t?STo2f%tx!}qTA+Z
zu6BMO)M|4kB7%;3xxw`v3!*lPlyr!(v-LAwYi2s83|5)b9BPiQ2j9rD54daAiUB@P
z!+uT&n3N~#fKq$eIox3dYRitK4FMi7;Gs0%Y#A1Qq|W_(!)dVCJs1jW=+lBJph{he
zBGdb<hF>X_PsW;70NbKHsEcphraboU2T87-zVKzzSrH05@j205!gb5>BhgvG8am%#
z#J0edsMY4w1~|S|+S73ikA6(o*-{j^OE!}^@C1eY;UZ2}A~4sFtlrYEZAtRn8)$g0
z;2dd~P0pkm5EFZiD!0@wvITA8J6)Ax%~g+A{t6Q@Bt`nQNd9^V{gm43wA{X-$brb*
zb9e1|;0*M;SkXhe;Q(UU6HG$e37O>6LlhHpSu9rN6~ow|y4(H$&H@M8v6KpD<ni3f
z*ejDPpStOZX+j2MFitwbt8#vDk+zgw_$kUUF6V}K*W16)$pT`v96OJTg>Y_lu?!81
zqB3TWzMDj6EcXk|gR!nF9o0y@VA}4oMYlcN9ucJN{X%zERC1Fvd8!?*N;xiqDAn~o
zCAK5EYp)C7r;To$X@~it%9@0fg(_&HO>Di7qiT6b&FBbelIqaH-bc~0Snt}vTw;w?
zJVXB@->12#WemuDHlVR8T^<JEXi{`C#+7=aU96Zj&Hma=+k5@9B?nyC)bD`Jm!%69
z<_$aI5fUpt{m)o$%P7B>Lc=Wt2D<6#+`|=Vd7Px9)19aLc$PcQ77p%3Gt*<4a5yrX
z$4IG$b`J4f4`1L#QpUhgO>U1JR+>|Yea2C)*!{N|Izf;5^6j_=<0Z`4FSo5iJsfQ|
z90Tok;O6IrJ}nMr*(K!DIde(jZ8;qUjr4l#=M!ciXU6Q49txUWn9mxe(%FltkQE0w
zs7|OAmwTIpLS6J}TTtmsry4<oM?ia>Cq8+?a8>G5;njTQ{9KsSt>p_ZSaIdwd|vq@
z0FX5q6LV(TgE<&`Mt}T3KI=$+GOE9mKpiVFa<9z1vuPE#4%Te7irLHKGb5hR-L=j2
zh&_F-IIoQ}$*a3Hz5K_C&pIN5Ld4Pb*_)i(!1Xz@_E}<4ckLrRGV2nrmkvWp1}Y9C
z?{_z~|I&&+FDlMg!O+=1*r2t7Y6a-*ptq)>!TbA0O|Z933&iP9fm`V)-8_&;48~-s
zw9bf66v3k5Zt4!5z0^RF9jWBHY{Z$z=3%GLK2J6YT*5Mq1W2oKUE-sia@d@rUhXK(
z0j7ru^Rn*-z+>_xa%@Fio;iVzK&GZOMz-)GY-f<>Bf{3DUeL!?-%4ApHDyFMR#9iF
z^sv4L0_84XT~50KX2E(Uv`ne6u`7yf7eYUCSvdccMybll8AgJ0&YN^~fmci%l831s
zBKvE}%3)Z1M5^Y(&9ricZ0`v<@w)dD*qhfMKhaEH75Aj|@?hL0dGjJMTRp}%@?S%=
zUf`2_0>4B@DG<-HoT<V6Rs;PZh4cefaNi1tZE=DZvy6HDKeU<A56zkTB@yD2G~yf@
z6_$&k5On}q9{rnxC)vN}94`=c2#3^t`<?w9?_(ir*0*kp+&YZQGPy9(L-Ec^6Jc8^
zaaSx-U?ZMHiWMH(Y9HB-V-f|+U%MrHWMDKY-<23yzB_XPqcc<*&p-!3qQ<v=C*o04
z2|kcjn)#Fme33$CZe{j<4%d<i2Y{tkMABcHpUR@bGq5FSagN?iILp6Bb30p$kE?a_
zd>dh_j^|pp^!3&0G6Spw1;Kg28d%c@9G|tzNuBr_REt6<BlJ#UI-<i!BB2G0T`wM@
zO<Y;e&ScvdZ;p+xIU>mw3;`S0X1}JV3CvD*l+~yd6zQ!elYreR=yG6?SAN~Vfq$&B
zcv4r_9q2$`4D|}Wu;oMWJ-h^Z=uc&*jp15r9nQPvv^i>+GY7Z24rVQ0;HceV8+x8T
zfT$nOsP2N1QYjAR%bgJ)<&d;N)kLO#oW+@C_z3$ppS~}MGNuSNUzm?U?64B9$wtFT
zjJxLIa7(#rsqBslpb|^TWT+QGs6d;tyQG;+{)X=;YTJ^+Kp-j6k8~i-+U@3WC{@wu
zgU%%uiD=HvgvOpHy&vhsgUGQ}Y8FPQL$JP7rm}7Eb$6d5Vb`90RbO!Hpw5{FxG>!~
z`D{|*e*po}t?x9E%D@~qxC8kjmV>e=`n|It-bLw8e+ms;hN{1c$_p_A5BA_XurRLv
zerZ<bhWT-3&%QmHfB`SC=Jkafz%_2$<4pcM!g<R|0~9VATJx?ADjcJ(bLs>+0sZ1v
zh?*qvOz7$BGh^PH)zR&Tufz@kaNjm>NM$@R9o_hT_}+u>eey04$=@IgshrhMg2YDx
zU@TykOKV@-_h&NHk;3L|hd5|wQMWPNw>A(|1LvUUdU^<8)}_ZSfI6~RFu{pbaoOwm
z@JpCOoC5nB&f|K?y=r2Q98X@h!n4*eFVShCy=R!KbeS%=T_1!lk`kj~g7uzQRiG;=
zwcCBsQ?yAvw)vZM((EboE?V&9k`uawEd&3Ie<GOn9_b0#i-eB9?=Eys5U_U8D6aDV
ztGg!Ur?U%6a8)uk?Wa+M_;8aMiWW&A)-`VGyFF`ual0TJzi|bYK!P!K)#h5GgE~kt
z@LuzQ2;ZxtwQ>kJYUs+_{R4bSz$4`lF=T{B!WiX|Q+C+MdwHYU6W)Q`@DMC6lP;!h
zf4!nBKrO>6#u*TTYmrZ2O@SR$?gP5sOTKwK=Bg7;fah#W@w*ssx&+e9dh)N=%`_6w
zs^dOGD8WZ6{rNI8TN0~Otgo{h>=B0HDXKY(?~G83aqIWUwp?Qn(toYiX?$DG=e7wh
z=AxyD!~ux90W0&~%xp4`rF_!{c<`bOh&7+$OKDVKCiSViTzvPFW1;9^C3N5k(H7D~
ze;hy<_4{Aa#OcVA#!9hA>v7glltE3ei(^*`zvyuUT0PwtQ%ul0A~!3C{N$WZNq;H?
z(20m#b_fYviLB}Ywl7KX`C3;+U0&-n1Aj!(@m+J;|0trI=G<!-yOVK0)xUaC`G?+?
z_E6AdVeIVY@F8-$Ic=^mk*xzORGmaq0mfBEs|)wdmB}t6`fbaj4Uy=<inN`_cq?Xp
zitn1v)fa<F$fKSeBz_7=mbl!2d`y}Z5~?OYZ+4E>^PV@o;uALtfjP^6rjV`tKEVjK
zvN%rYE&(IvX(uIh;9QEZSxn<_ic=AjV->zn!1ew8NEbpPu4?ypGfr-CxZ+P5J{oYB
zv+FFaMXY(nXtKS>Uqv3dcoqtgZ8P$xirO>Q*y~zR{$r#YX=Kp7x^~&Xkd0@w2uXlV
z87`)?QmIR@LH}A-EIx+3;rvS?j_1oOW?gNeHHK<~0$bH8zh-#W(ktf_CIS$JEtpC=
zT?8VEJ_Q=bt}+31P?nb7WSR=fDE8?T10SvNj-x=ubLhVMdA@c#ol^{vmkcAETL8ZA
zwfvyod+bkHB}YV2yF?P&H14nB=5;KjsZm-jgfihdtT1J=3v2P6h<?VgHbiZr*Zv4f
z*bdEBzRbBPgh1ejc_!C`V%h1*(EUlomK&P&qNwW<wh#REcJlNci6bf$0e6y^(IOQ5
zVhgGztQurw&^)rpj<+yy8UkAjtlm5^^P2#f6xr<onj<-nm8dib%XLNxcj>P+I|q@M
zD?OG)KO#KVQILAhLS%4Ay<x_*#{Td6Tr`HgbL|SkFGx~&Yh+k_ZY4)@Q#BnVeCHyY
z6pqhEGKj?S;{&3FKg+!2Ec}V3ZGLo2-KeWqX$;eVFc1lrI&rk8zZ|#eWWDDg23p9S
z++uFFB!AdEhdPa!tP_$RO36BAqk7pPZJvQ=t;aQrq05^}Cw*Vb-7`jMN~Dcbegc>>
z@85Q3b1UQiJ(Yl%aro@Hy{=d@yvI}zW%;_+aGLiZGYKd?P`8SjM9x0yhQC}J*2M2D
z5K(c?8HTgF;@veIV#nGEb_CP`(`^k`71omqg~RnFfep5J6@mCyxfLRNZXE2$<iq4I
z*8UAy`~(+WmDL^lvwjyr`6GSxNb@Hi@IocdrT;3&#CiMDC7R+hZ|I*tE5aLUraSle
z)*%U$Q+}cAem>GkpH-1{Z+9`kx$#5=xT&KI&4`*G2?bwLm7xE1C=i)cr0d`@@i6v@
z@@>=%jmdkN6fiaeFU-I=EaN-ohK5N=24z8LTfTZ{+c@2Ood%*9FHPFw*)>(}Og?%H
zpSkXHGWI)EV2P~KnA|m8VKHH1gGT;|RT>Ul_S41*GcS!V$Gmj?T0KBPKh)6P#o2|?
z^DdDnnK{69O364|0EPue@EiYT&JiZ9m9S!(a_tCf7z@aNj{{*`T%UEi?hp$h!;BTu
zAXbmkBKh2viK1&U`U%>Hd1YtVPTtMT0#<n3!*mgE6&)QuR;OXSYl&X%zIU89Se90C
z0Bk0ZCldO4#)Zb{Q9+Wl>W~kPw%eJHm12v>mS!Drj|(bo4LBBoes>iW7Us2xh8r3N
z5_pZ+T;~$fjLLM>=6T<h4a&v&oR3f5<gKjJN(Ve{q@ED`mbYd1t_`hp=7Yn%tjGfB
zM{*jmh9`|6D|&L}iz*08JHn96sIheZat4GI;&2!WZzZ8&PJT{7JV}W(VGooas6lwQ
z-CXF5kdC;7>$~q4n1-8@M{}7<CdrjPuotu-4{9TC#(&Z1mHm`(N<GI8SL>2aO?^f@
z$xT>WP}Vdkf>{&pfeFvJhbN3cdCmE|@9fjq^i5CgL9%jtQO~Th8N*b*Jx!aZp^6Pm
zI9v)19uJENyPVE_1?gvT{k-q%MreJEq!T#;Wex{bQa7~<zX5EPKj1or<`!6qa=HQ4
zNFgdt))~NkRbTRcdTOvjv}HRDw;~Sf<(}-5;P4%zaJQ(>Hl@!g29un$dRWe~$<i(1
z1AKU&^H-^e&{L9Xon+Ck5N9L{B=GDkZvH*uuFP))D%LaqPB%Y4GbFFzI#ZhRTtIz8
z&`t%^o;1h)Whw_B-8Xv|-xCg8gH3{7B<AxHWGo==0?KdkCmu<|Qo+RRlJ~vFUDqy%
zlJtkyn>H4s7y4=P_4<ZSh5$kdUZsN#z%d-;ONS{f->a!Opry@fhBkP)7+nUqJEXoj
z>p{#u--vb!TNOWQ8JHS^j7^INKO*?gz<7cdwY4pRR}#?l!vS2TGc&&v;>uqCfvh29
zMj3awxo|Chlnm-9(7*lGNr5i_`*i}Kq)0PSn6~?*k6;q9v~GjNsiJ&F26{Q38Icq$
z_Ao#<dJC@bp7(f)-C}B@J)B&+BuPMYu%80ioGD!DacJe`%F?nCQsYjMvn;QuCvuF!
z1}Rbmtgf@heVb{$@3tS%?uw6V7p@?<iGy)OdV*PRGkNy|okiJk=uxOac@G#RkVkBi
zoSio+ginE&Wz9W=i3m(zmZs1xcB(YEi6NJnv5#m7HwM;)Zi#yc!~H?kL{!x9%NPFU
z-U)4~QvY22a>WtNy=vA(md>Xl<ij~l!FIx*1@nm}HBFPzaa)iDIozQ)PTJ!f!^I-_
z?41lfFcjFekX{yT6UJLVeQa+r>`KGX6serqr-5BGV}*aHYXK^VgK-Jx{rX&A{fDb4
zJ!?0aDd+aOUa-$GU5yvPnh>VQ(H&ofX)(XxxYrFie=;zz2|&Ri`twh!yx&7T((E4R
zAa$YkQ#@pM=LB5s${!S^vJ?B!jGM+0-ugF5*z-lKDY7-DZ<QXxf0!P@?E`@Bk%7#J
z+K_l9!gFPGOE^?5`Ks5nm}dBa%<txiSo2taSh;RmjOQ^*$vjQ1qvSaQN+SmNIuFUw
zgd!P~lwHvU`dHE^zT;mR;j-esU;?Xrel6IIgW_%5uz(_sPSp&)J(wQH;Z<E?knei(
zZ%zl=p$8fmQ(;v~kRoxgjVKvK8y~0Ig`G5<wynQOJNeX|d1Ia{OdHsujn$3x`G^8@
zBLDs8WHRaLO70C?(fw9+<3|0)?|{$f`(}b{=vv1Q9TlF9L0`N&xqfrH7$ARti@dWN
z(S&8utcA)(Pc#pmA`7y=wPo;ci=q<w5Sc0Xx`^-3C%^UM8ZH@Q43bMZC5V$hQwQrk
zLVW*PIsa%Ggj7>4O^!#S2W-S!bx=rtaBv6Wz~fkh@6cEi<4jrh>68-@*Kqc@3c4!=
zshgI6;%P%|vKjsz<HvGzUyMbdTBgO|5R`>N+0a(+>9BUofoO%uSgCjU96MHz)F#s7
zIB}AxuL;6Ji-Ry&w{J0aeAm$|!2tB<ObbXv5l%6r9)&|A5TOkGBoQ(K?Y1y|ntuGW
ztdM&qYt&9rV*&1+iWdBo7a(2FV2`V_%z99;tQoF<wkWO_65Y{AN+to_e1NOd?m?a`
zbYMJaV(C6>nr|pkBU`flRiyM}2${`c>VnmK6xz4;s(xM@I#gh$dB0;N+2K9^DV9Zg
zzlMY<;o0m|ui2X3&D&;!@HLEKAZ>duJ$2?Aqu}<9VjytO^<?C^z~1d?@ms{JA8%1}
z0GYz)Km+U*oro%VZZWtr-Hg{n&y{npJ=>h^DUxa9wh^@UIhA{n1Ku1CVFW`!8^sEQ
z#!H?>X6bZ<?UrO`iZ-<TN@|RXuxc?Vd>EJQETy|Gs*??Kw)SVJ;eau+G1ohjf7+*)
zK56yOKteAqYpV0_r2qEut*{I-)*a_}n*@k346>fC$@%(JcQ*W1c#fZT&q6Wo2KhTs
z%br41o-GJdUFS70Gakcr<rO7x83m-}%xAGWkdQ;kft)<ZbBPPJ_c@cJM_bwElDU7$
zt9fiznXGZlaUZqbO2vJx?|RU@m(Lo!14b9b*ji=iz=^xkR_Mk7J}OY>ALD+OOQn?=
z&x1^j<>FZW7Dl(MLGIaOSyl81-UUv~PKRui^-ME|k7^7S7=Rst&%|J}kWtgURZoWz
zQX-ieK1-EZv)2q$iY5Ynz`wP@1HnXi8g!x)F;IXg5AceQyJ#~236)`6FQJphOTS3f
z{W0Bac@7kJAmtWg6ZB&MTD&p3c57@@E;B-{1x|!VtSl+9`GR-C3V@9R<>N{|AMA%h
zW%%0)SZCU$&<-59i5O4%9u5g9%z4Z7;H6Lo6s7XIvB)4~(aR3!NJdPkZck?|Lg%=e
zJ+K68_ghYK%jbKoEPL9Pm%Kw+-zc2Wy_^B0?@p6`Mh8srNVwWl05uMWKNq%FLljOF
zi=sP@mk!LkAy=oBs!1nwb_`&{`vCGfooO8F+`+K{2`<ArFwwk#e-w#BA8bHQ(<rN%
z%whqzmZU4vvyQJ&es(g05}aff%``DPz5$=`z>=D*UEz%Rrr}6(f_+8}BU5M3VBkTh
z*;Ih`-?BWWAkY#blv0Qfj4ukHvt^F9r#6<wgLjD;B#|_vzN%shhM7AQ>N3L|iY;yD
z1;pQ6(_KjhP@hl{cpNODofSo*jW~8ZDylN<Rn=|4h|>sLUc-ZrB9dD#L}<)C<h~+-
z9GX{v8FV~t?UbDBz!k)2NqPWI%y<ujZFxzSS?Ll|L^J~#(s1j<x%N2>U&GQ3x7>tn
z!_x2&jY~61BgQUi!%EYJVYpmE2oV!UtPoh@5p2iF2@p#^JD#=7&QhnMoC|G1Elt>}
zxn5@|4#L|5WK9~tQEH4R%~(Wq`A*Bb<Y4F6Lh=$9dfPi*beSOq)u7t<u3+@%jL?Ya
zcq=@K7|Bx~?&0k{jVtWimlf48DDc~u-Y&Ctv<Ys(&?vIW2NWGYzwJ0O)e*t#vO>UM
zK&?czP6;&{Q5C<hAvBPWf<*_)UYxG=S?&iGvm<pz9??mTzq?Gx%zj%D`xgyMij*-Q
z3o8#)04tgnaUK)1nEH&Jwq5omteJA%wE~bC$OuSH8t5g4Ijv2&Z{)_pipg<YMjz1{
z7*CPgRpM!Ktx7yvYc>nzOwr(JJi|+qaUjJH`V1mZ$J+h$r^R-1ezq|fAo4S;`6+ib
zW|Y7&eY@Tx#}R*02>TXvwifc57enJ1=*-ub`!cQzd}NBdzU_oo&nWTq>x4Oy9{ESF
zBriwK|K(_u!7EP<2-pI2r@Gl&8%7siNFul1gyiE{ye}J#{oKHsk1y6B2aVsGC7Ls)
z+$V&{i>fDS`68(-qJi4GM*vSUet2jq^>#I-jiB*_T%fWO=5`S+j^N1QF7`?xd|``G
zy=o4i=}w$S2z#osok{IUhkf-hD@l_f9R=u$G4c0HWQYjjAx+0@#wAdD0=Zh23)2}W
zifybE7Uq!@UvErDqV&==_R$l=iwX*YU+YIXvp`j0vlr}jQn;m2cUSGx?zCm1RaS9N
z1OzEp1QR$KMAXsSAcVs%Dlaj$(Fun1MS9lZ^>N7XLO87NASg)zQE>8VdXO!-hCR3a
z>MprENDg5-lXQUpCWt?Y&q<qwFol`2?0QlMQN)FbiwQ7M-)??sqgDDZ%e)@xAUCOa
zt)C=~5}6(w$LD%4|Ao0@>Od~&Oth{j8AY%~E5UWSs;Mp+*&0ho{<WNA-W;+05^sdf
z8L;&Q*64cUiyv^Q=2N#N+<=dr=Nl40SU1MPm{EOJIZ43d&!BLUTITPRG!>V|9Ujhr
zE#Lt)4}k!6l|1BG)F^s+e(IU|#A3R3nLml#wcN#GP(41F&$4N?;<5dn0T@9E>d(|$
zj$EUEb6m%AvM{;3XyMH{Wtu1P_~kd6zb1L#0Kng@8)*y=ZEMyyOH~m}sjn+fnI6?u
zH)DQJwqWA{FT&YJ3H~h$A!n4i!e{|F=U|0z#V2&0d1G{3SHj%r`w;%=5@?D@oOeO~
z8)6tNOASJtoFMi)OWdzX3fy?k$h@=l0@JR;4lYs>*F>$9Am8f3&9fL9aqdSb3$Mrh
z2<T;cp`m6ngiMXWUaQ6x%AyiB9O`}a0;}n;yZ8`q8WTSGOL_YdzVSX4w69RAiuDYM
zPs_rbc6IGAvLl-93%$XYtFpyh%a1k(#j?=tDW?IK$W<D&k7yH;?W`F;aZge8mf?;+
zxL`X8#c4t5sQDBGSru*hc1!AVgmQhB-5204=OAYjR(PBNupj%aA0<}_UAa|CCh#KW
zHe(k^b2#U4mrLR(sfWl2wOP`NP!c~DfJHw3*>1bU^`Ga-Y=hg6Ctgp$g=_ud7`Bvb
z@p{4_rpLx(PGIY%-P=Lv3JU0v_w_&*Aeq%SS)#P;H6cSi$`Z-7J1%#Bl$uQaiX2$P
zFTpM<uTOCE-ADQ_1XWs%_pa@3%nuU#!vg3B8<8n4Jd&kcRE^&07caWw_C~w&k}PoI
z7Zc7A@+h~)51eHFr`*J~5u9%)euk6@4?otp^e3j>g79g~KL$w}97oOvbyRWByNzhR
zyNQ0STT}_-G_X%+e*uNfc-zSl{D#MG=qn4Cg^GjUOf#q7>!pWaz<JY+9B`&lii(@a
zu?>Id#~0#C1!PiNmy};DFp?~i-9%fIH@jA!^$Gw)BMfonLPU-KK+E%RnJdGt#Shfr
zO#cYf6Abt}Dawl<2K|=VRg|J|i$&YC>s<V4A_yJ7+#p?NdO0i%CO*rfa20>pfSNsq
zY)(SRv_~QwarA3|IUQAd1`(Gy#>yJ@h|)2Yeb1?)FgJPoQHm{ZaU7qM{6$_iQ>(F?
zhD*erJXOOGtZlbK+l8l369Nh``2$^NtV;U}Dx+@8F%_786z_>r&{2a*33(n_=)Rk+
z)7^`Yv{lD&mdXhZXU>VEoxVn)(jfo0irev>(jh!vIP_$00U51#jMp^A)*Sqsj4VB^
zgP7*Bl;Ya<Pf27oF#Ta=rJ&}&u_>7hj9QR=F%Z=FU#)1Je9mFeq__Z*q+P3lad}=&
zU(zy8A<fl4C8-gNjBL0C)}018nyS{*)rXC}C<yBuQP4J*Ft>KK()W47{m~)D;!#oW
zoGXo}bS&+2t0h8V1jjpmMi~wFPB}q8_SD!SqgAp_95Xi1dYdYsP?bhd>JLrq!MQ;D
zA6pI3V<br+TNv8)j9nM8HD*+Zw(3CVA#42MXn^VlGnSf?J~($*)4^R*5^Hwm?W)4@
z{z&>j%dZ}wJ|_OXRjU9g2(ViXcuZ7b{=8yeckEOgl(EOabg9mxRu#ZOEOA(-d)0u%
zdI4f>ij>r(U#5P_`0WCZ`0>U2Rw6-=4AnEa;urg@=p}f!=*}SWM=UTETc(Usx7lL0
za8k$?vIIw=eJ29Q)2-k<C80;UC@Rcx;#6Z1ddR@TxIK1HSJ+N&!X);^dD6CzCV@1;
zOfc}>(FLJo+JcP>#&GbhV_ER2^YQ*fia{I~6GfG&c9uumf8FLla0NjOv7}fftB39~
z{Fi`R_(rH>t(Qt)03tD^Ffd*w@k43M)LMk@Q1G(sMc1yC`qQENJhm&IB*HdlbDj(I
z4gPLVcgAB>>0>Jl7S4EEIIJ?~acAhPWXTBy?erPHQQ*W0Zw(PME<HA@Y8*OhR1hXZ
z2IKs6c3(3m5~$?u5WD}=#G;?i>1gd4D(oGS9U&$K%LW6lbsD7=R5?Ea?X?38%$VJh
zmvc$^a86+R<l}<JshhLV!AyPpp(|T`q}km0x0AD#l)x$TkZZgV{>BsmYl&?Tp+k{q
za5hl+I)>{yRZg4>2#vTGGNau9;EcEM1=vM2-qKoV>X}K1G~ZAy3p|^B`jaW<66;22
zu{x!*^ULU(2voSGkpGeodU3k*)Ds-TIdRGsJuc}<;0{Ph3sIXi6M}We?WDeYmCj)p
zV}@w*n8G)}tzqor(;+Co6#KS;<Z1O*5hd&mk}AMD?#*+>Nv6dDmtJVs?OUyTRSLn=
z>1@G3)d|@9CZ=$IeO;y9_6`tltaI_I%vBz2A$JNX2O!nV_qWu6;CF`{)B*|ynOf%Y
zZZ3=Dp88X3?rggH-jr)P=B$>O_PqB?S_f@VTC7YgJ$tfuzBqfzoxD@8+vHXZe?-)V
zX5HJ2_-;aBhw5f}g<)8Cur>k;b<~unUHhpPiV@TD?N1Mj+c;&kN(})|TFRx6$O0t5
z&8f(VB}ldW<K&oT3O>sE%_fIBs2m>rU>wF=U#bv6FKl&6zhv6wuh`$;d5ET8<0a+W
zl-y2r!{QR!f(0AA;?jg`*Y6uf>V5BsuH%A53)U*;%r1#-xqd$M^Y_~<i(74}M_G6{
zOdj@D>xByvh|K2D(u$13>}w%)iPbrL$hQN$ckl>kI372vWs|%2&_o&$i2OYCdKnpb
zAObMsiJKup-=T5kX1N`&br_vC?|(c5Y1Nv>rm>zxX>KoDgat;t!-j}jfM1Pom}>&X
z7aD4*>K&Wi@KP1wxY`Mq08CAEY?I{7zpC8yl#$oxbY+zyp3#b>{-YS@D;+e25H5Xr
zWlBUIk(fp(V|vBdj*nQw$;oZ(A<$nYcE#iz(wiD_CG(1O4Ck_sZpBK`dtX6+`t&Dq
zx1fvZcjVDjt&xZ~`e`J1+^ICqM+hGK0>kwU=y2?obK}{C|Fh%9^nZ5Tm>HnxWlZhN
zoh|-#+?ZI|IR5AJe|6lLS(q69za6(uWA_r>CR!Y~<%SNUO-b_ebvZ6qHJi;=+kdCY
zc=N{lyZg57m2ZvjbdB?KO{+<4r96|>YSv|Utz|s?DCG<!+6KnIkZm>AS>`Ev3jkGA
z)x$?x+J~4K+mcHZo7$&epBbA5Gb5f?VRvDAR7+zQaU50w6bmpoAQZsN3;-D&9i0YK
zV0?6SW@mbM8ByRpEhsWH^yOd1tsMYn>j(H9``p0H2$taYa|F}E(%2fb+4W2M-Ve7N
z1Oy02HUPkwS(-pVLQYRtNDW4SlB@~{0ay$8`k5B+K@Kd{EEK?!SvbE1I3CKt#u&c-
z>lMhr!r0d2Wo{Mc?AH;J1K<e|u2n#|sn1YAT{!=mAAz<X&BEUFWbf{kyMG$cz^Kmt
z`mGH-mJNW}T3DLA$Penxn!WX1U9k0q&5yVG$H;Hu$QB0&7v_4#1{ct;X%af>r%O&;
z*=Nkp@n4241dwgik2l5!Cg*2kv9G?5#ZOV(huIq3pLl-dK2HZsz&{})sHUWMb9c*c
z!sdGBM>RAlD5GV<PgMZA9z64G6I&Aq7w2d3>-awnw-)*S-$pMVZEfi`Khe#vsY`xq
z1}A3#Ev%&>CdS_x%d<E3wulT?nj!uoB12mnU<0GCw(aqW7k&ow{O>+2k@vZ2g>PD+
z-SN@Ci?9YDETN|HEe>B<NCNNw8gAde+2g;!NWXCqU;IS3zgg|Sx<$Xo2)})~r@wNv
z7bcdJRra7?>Ii%MHyHjbdwJ5MAAXOkZA?whpZxE?Zt|*oeNkWhixM0czqTR%Efe~j
zXYFU2oR}Ryu<4xPnH@keM>9CEvVfx>seWG1^j(oHi1Ry((^zlnOMWfL0%ODDZ@Ubp
z%FM#hmwuUZFZK`?wokj4ztl^8mF5b{NeO96CGWd8s$8T~eUBwm!?X9d5bu3Sd*xSV
z&>&yG=PIT)(LYZ2@Ok<9G1UD*u@O1A0%Ow?53c}B%z*70n%zI}9lxT(Px{uFKH^@?
z>i%TJyu`#y`ip*in7<R>d0_r8enp54Y|V@x=B~*&&owbWv73HaKLM~gSJ%IlUv=1f
zw=ciO-*p4vK)L~;_105tp*#U66_}Q>E$N&?NrRc+Ju-Vkf~};mlOgZDD{Q=c6$SsM
z=breif=@9^TYi7HQfF=W;Yljs#l0A5C_NJc)IHqwNK|~@$F+}eaah8BV7N$kmu6Lo
zd27*2+6Y<JG5v6F<ftevQ2Ye;<=yJnh4uNB-xE1bKMNE1Hi+D@OP9y;-H2EN^Ow=G
zbCSrjdH5A!Vq;+7pSOfwaQ7BX#`@#$&x}3Ea}Y_nI?p%wSulPAyq(OVd>F%x&s1ow
zrJedBE6Eu*URc0(Ap>&ye`<<Vs8O~j3dhq3t+&Mm8M34~f7{HHU|_c;1IHgd{;a<{
ztt!E(KJIyx`;M%&m~S%BWgpPc6HVT*KUz5)l!I0}=q+)JKfbyUwc1Br3d4~>ZX<-&
zG9E|)IR&~b{XN<>t~jQDZ*N^ZA1X~@jn^USdT_bOzm{N!SSlM!xiB~A?{v^6K>d&%
zUA7X)Ui4d-9Ysz{Iv>69G5<qICHEo3zN;n&@4n_&bFn29&-0VP-;GXTU$*f$l&v-Y
zHrvsg^>-U;AnB^nKu`O@90ZFNGTDc4qO+srffOn4@x*wlI@--<$i+fAC3HB0F&~r*
z<2usd`zZJvQz{i85Y;h#+gN5Vf&;_~Wbl=+^N#Li%-?bq8_0Zm$ZC{I)z^pI@xlhm
z@mDd1y41AlN0c=uyl5A>o5|q{ee>Eg2`-8I_RZvyIKKBqfF82I!sb@RzLEAFApot$
zu5N8nzYzCluIReO_G7??HDa)#hsrUTq&GJ5Ag5%q1?bZQIuxhBsLe7=SB?-{v^{}b
zaviHxR$lu>M990dzvwNf)YjZI(}eqSnLSJC?$(gkIQ-SV6wY6>UUK)+rRC9jyYs%k
zuhme?wDJ!1xA2OaLr^BjX2b$=J1$3Z=RG$MIHlO*&z?6#8rtGYh<U1WOpCyRt{=Fq
z98IBkCd&p=FgHimB&Dj3btu18y|1b7jl&*uKSc7*wR(edVscr?278tUGu}?YItBX-
z=Hd8^zjZm8>F(w^%qbh3^7ThUK>trpZ1kj>wfV^#$lDl>{g<aSzo~W<$1CqiR20P$
ziOtcce1n2zRhuvPD4m3^B%<{khoe$SZ+>=a+eTiI)}9)<!m)lHznXCWM|LGEpGQ~o
zWAz{JnU!90oNJGamSijWNjID3&_V%A@N0#$_g*Ps6xtpymzna%6#9pBA)Z|^eyo6-
zwP>Qu{(Yj)<P*c*&(C#@3ytJZfs@SlThS<tW>(qhRRjeZNEWVJm;$}ZO~4+mOSOA}
z>LTSMKYA#G{)QP#b%6@YOLe^kS^Vvv9dOAtvy9P<<T;Lv(lM#TFENslQ44|DvH44L
z_f_qyOttp}S9+Y8o&9i=bE*>{emxi@e|?<(JFbPVsvwU+<BBfnd0Ht`MQNP#t~-J=
z=_uJrU)C)}juw%j%55CT^xYj*Eoh~(CAh>GJEJue;*K0%m`vv6-uG6@oiF^DIdsyX
z7>;}ZxH!i;1zOwlJaR^=$i~>C@*RjWJua)yG``FO3C9E1)*(>Kp|mAegShzlSF(c-
zcvM<P^hSC`lkyM-{<$%4v`%Mv7Df`@s08&;Llg-E&mL&xgE|zx5Vc=EIEtmp@l4cs
z;*yoz#9Dk1jcwu(yKv%mPb8Sms_8tHdWc%CQjq=naa~kU-X$Ekue2ZRj2@80j1ax+
zWo2rwnIW8EOY1L`h2xhS2%vA~oJ?@GE*Q}J)WO$@7kVh+Am8d7U6Y0%Zi9!k_j_N2
z`*~KNKrgY@Jk=GK66x{qrMQ*s!#plTRcu)wZkHSr!q&rAD*Ol8bL-8j&dja~?GYLl
zpNYjYtN7g^K`}I_Grtym`KcCC68kd?YHE@rpHlMul%4@RWy;mX0*&U6Eqm(nr5KO-
z${{cdjc))XFroJ^4_-3}2DW)w-0kZ>up3>N942(*Q1+gm6g3jEaRWGE@B<ZZ4pm{j
z>Zg^gY%!XF*ke22JKmSLE>^g4a4)ewOY#uZm*?=uAM+Q2qF(5o{hZ<WSoruNd)VUM
zR{-*QK=6-p?<D|F5OH(aCEFf}PgCjtqQOC?e>F&C(|6&1Ela$oanxsw@*A3-2%X=j
z$5few8QQqvZuq6F7SoSE@~uehntV`*PJdId^Vn>;9_bvN3BZ2#dvR8Bv+nEhS#i-Q
z9a9OM&_m(#NS<kp`9=uqXQ<{u{;F6K(>%KviAN7W+)iMudPHB$6RltopLVtxu#$=m
z0?TyWCojiCh1q<imCPqb;i&1MDq*IQKVtQ!3zA%DI!h8wa+Ytzxq5=Qj?nf|=r-0H
ze&)9KjFVBq<-7i^H3Jb_y&DOYqu#7*ZYzuUf#5YC%(pYYRuO+ontI6I45TPRK@0~`
zGQv6GoG1Y(4Af(fu(z=^tQ>I{{qa5S4(I%*E>S^M`C-G`?^@a4vz$4*H=;`~Uff#8
zyJyg56R?2ZW!~eWy~a?Ilu+ytk~W(;L9)<aAj$rry=OcLWrfPn5VnrEuQ_P$>p7~E
zVhEocXaA_(*3i8>NM{_!`=udTs_=~yxI|MUhA49+EB1EUc``N)iHb{n94E;*(!)5x
z{#JvR24{~jBZ4_2pp*(nJhA6CAHzNRS!CwO8DnI!_vH%45vZ+nx?+PPQO$Hd!jFj9
zgJppYNY7K_FdIV~8RFV-J!Hj7m%hBUB#n_B!G;t<y|S0ciP?!)T|3x}-xHkyHRp{)
zRloBH!fh)CJJX#vl@L$MD#(w?PJ-lkQ^GM=4aFCnAczHYwy{4*PkX$w@H;4wgI;Sn
z#A*a%vL#+fJAt-aH5<e<o9v(5y_Zk`zX3TvGI}hq{WgigI6^=eNcY1@56x`HY<i#?
zqHI-|L_#?A1j|p1KBOy2l(i6=9cKx}GIlgEIA4JL$J-#V1bE385;X`=#RX^6f*57z
zl?~ymLgQ+}?hGgPBsGyHZfi;MBH(j;eq5ugs>0gWBTQ;@#u-kCs>&xZAGYi!q+H%t
z$m}R3|GixO3*CkPS?#R7a<wxj?+a3LdC5En^n$GJgEN4hS_me&{jJ}+BQ9sCD!^bS
zZoI<e!{=Ci{V1*Gzk~F^wblxY6n>+7MEmmK{IJU-H^nH&GoEer>T7AKC*%$pA`^PH
z)_2VhTS&Axu;yAq&T#h-liVM`Dy_OPvy+~$;ExD9LM~$Iyq^?f%V2}%JY$RGshHl9
zzvF{q?&$svhcC(W=%0c%l~}Fg?lI^vz+g2^dm5kD_F${-)G2(mD4&32;8=aCtjRAV
z{^;+}%m>h1t2qQs0~G6c|K)ND5ukOG;AGVe%(uj-B`E8Ibcg0C>zlx1IG?wc0}vFS
zj3x6j7O^QuV<i4Dp)nbZOvO#>UVw=|9eJco5Vg<RQFPhhEbu886)E_BVAS*KbBcv0
zUb<sWJm~RGRylZ)l0};+x&>sN3EMv(eY~FwH!+?y3X{cQy}Y7&!U7z;7>z?z6F*ig
zC$rrC42Of}WV)fZ20Ym9R|dNog{%DxVk@~?1&!t~%QKm1#ijDCD`!+)!U*oo{J;%r
z>#z^m#cN4f;*sRkkxpVHy+s@ElQuV(Y_p3nT!uF@%%ce7(0q`TG73`}PMvZ^r1DXO
zGXhmPXO1)>vp~MCPPS|5@<w_V?R;eW#fahYgB%n*DVQA>;o39Dw1z-J>Y(CuO(snx
zvuvxhms!Ud5h6Se)uBfik%`b!O*o%h6Ev+R;D)j(Sa#6BEGD~Y2M`QB6P|QCfjs<I
z#ST`harau0>C8b{`4+;>?D8Mq103)ke5tiIz(J(1l1S(bZ|nxl&)kc6;e%m2sEp5M
zEbdy~^g__JM)@7D(e~4#3X2z8BMqO60|IaI7}3(rxh#3-i5T#=%kT1U#t7L0R6skU
z)S2*sFW%-&Am{Gjw2p^X8}54EceO|{yt3!IJ^H<RCUzoRm`qgoo(*KKww}IJVHH*x
z@jva>R!*bAi1}wEK*Pnxws@ShMYZ>&n}cWhPVmwrFIaAN$;aI)h){R?6yEM92x1%u
zB&#r@?*k^8U$OoH^S5(L;>K`Xwce*dYeYvhp_~U-b(E0gq7PVlyW2@e^+g>UXoCNf
z{vtY3VQfZLg@kWAP`xt_)_r}?6;QtBO(7_jd)D80=OV0!c?ELeMgw{RPi;ZQ*uIgH
zhDr@s4dRUjO1HU1RW<~Vc^p59-}LnMGzIYkhEwIuV{aD=(?K^@euNbCncJ=12DKzZ
z*|!R3j<IdTM$%|U=Gx?qSs#)Si{SIme-+c<P+w@&>x<?vF&3{6iG5u`u=~!#mjv8O
z)ocT)HVz93wj3O~9&=n!>nVXptP!6ym@(u}_JC27{LO^%lZZ#cBNBJl!a*C397;L=
z17HEyCyB-<3OzF}cu#o-b!-*l9yy5z!bmSNayQZ*lP_O3!cN|7w?81G+W&TxJ9k73
z=RZ6inIZwFsN|~=Aks9N%3eL+G$6Q5w(p6#g?oGK2K+jA@m67EZGgJ*bBgu11eQxE
z{jedx1UE7!V{Qc&3=1+#%&hg?W{>z2-@pR}1jNJzIytSk^C{??Pg+mn>wWEzrkpR8
ztFsWBJ&xt^%57KC;6k@HN9PTZ@m2sW$XU*Ni#oX;1P6=cSIvRZ)$3c&&3>Y6qJ5m=
zR|Pa;640z)GPcL7UJ{4ntRZVjz9J7MOPuO`?}t$aV)O+vd(@sSjd@^=sJbwzkw1rw
z*{@K4OXM%MSEYXh?1k<$azvMkD#S*&9(K1)okD=ap~_(89+~Dh)*QKgQN*SmGt3hc
zE4%Vgm?`lTS#486VL(FK+0{=6rIANjxM#BR?o|tk{l=?B7P5j^;V6>clp!laOEG8J
z=KNTJ*Hf$17=$}|Cpf!(FNP*YxwpZ7jC!?0Z&%-a+#S^|;|EeKY8tMS%eO<`w7rKm
zLD7}0Ty^Si_acw%au#p<hl;`6*(ve8ln3<$CMakS11kHWwW#`jx+1Ftz&?i5se71P
zlJg2lGcvYxy)WZ74wmv-iX=80;y#<U1M5?p<@BT_DNJk`j;XPB^iC6N-^e>V8yU)^
z$GMxs`tP|JU1YHv-M9EZ`$3&T2>aYbj2r;%J{G`yZ;BFxAS)*Xl|d$}w=#J@g&(o^
zG~J1G<`a3^DQ@zn4~weJD@I>7#=H#7_^8Z%HHUo3tw*M<e~4@SLEUKB4p>SS-)q2U
zHgD^A@9v4yS%|66is6@B1do$U`Ucbwql-3Z@lCyVef~n+_MR8ZB!K5!_qvrF%})y%
zQ+efJ9MWgN&x{C`O1>R6?5hA*jEPBfIHR?rP_e#bwB>Wk_`yDrC{~T|m|~5kBzoXb
zK+`ke;ex?2hZ>>_{5Cm%>CEe(^m(&J-)S~k7^w;?qf`v%{<ZvWBZJpBkshFaN1Wcu
z<zqCY6mMq+Mq5l1I@kL=Un*A=dnTzz6~|O&H`_51$0lx1`_W#UIm&DBg6710!oXsg
zP@MBX#>2bxF(kovAIVo%=1Gx~RlRG-eOo+FYb;k^nc8E0!Ja+U=26RD_FV_uw53)g
ztHv9AZCDoCE%3c%B6bCM3wMt1WKhbi$_>mwh>s}1SSiedzza+nPt%b&2HpPT$t&s@
z&!5a2ps%5`$DyPk7x<H>|8E?hL7%@`zu_%!;E*FW^&Ld1-C|?yBglOUp}!){bW@uW
zgkpIDi@(MwM&W#JZwQ!fK6H0F%k8%t$rRPeOiDA0U<Fdtyzwp(d|XYJvkI^-o?u4q
zJOnsk9<F|+i@P=8f)2AkQB@-AI=$Zlf)8HqtUP}#rAF@M$H!@On8rXm)~g<eNfxWz
zeg7FypyWSoU!WbJvD9)X2a3EHv9_LJp)CA5H7hB3QEDLvhBe?xp|>rN+|at~c#B&z
zpG2B!lWQjKJXI8_haF{J<r1@nYZLe<zQGUVL{zK6^%``oJ(3C1BDDsFHmsOKY5=R?
zaTkNXrE}@(cCH?Pa=&LGZFYoK-9uY(sf?(y{v6}<zU(Os>rcB33-LstdT`gE)hkUb
zZG$#49BH6&l@p#rb`Fo#XJE0c^^=Ym5tTQZ>7lnq9`3g<wv{^sfmUU&*_Cl`s4W)S
zP`dD^jCz!mlrBVcX6bm4jf*I!axTncZp4t~(>#960#&>+upep#J#B^%tASy&bX2se
z;K`+2nluZ--soVIB5UnFB7A}m`nFGX@IS|ZI|p8<aZ$Ha_}q2A$^PQ=fVYkj=j|v@
zp*y!4sSMJW|Hb;Q8^!jcC)*=B&b#T-BPaf;P_Ckd>a((74I?%i*I>Al_2fp16|D~Y
z+k<t_@ka-?Rzf;9t{7WzI1rym9)*()$l$PWwRXI0@<BaUG_9F;dIT<DvidAK{!MxF
z;C%Yh88dfW3avQPU^BJv^g9MWUJW*v2>wD=bNh<H0^bXnrEa!XJI?#-(S|C3Pe)p5
z2!?LPOhLB$&phP9tDf^|ypgxEUFw1!1sh;@x<b8&G*1P^`3diMC6oLqDFBWJPOOYe
z`b_zsx2kDcqz=s1iS$n5oZ|`0lC<u68>PvJ*2XY3lIjy$Aq`uuh<>t|RzOt9Q%8pD
zE-K@p{W1xaVJrC1zlW8ov#+h$58o~oMqkTI4wn{BwLNC^6<S$`cw+A!<5Sh2(?3|a
z78U#a(KfpZ@+w8OY0r+0KRBx`cJk3ep}U8WP;B1P<%fE7>(IpSLesU1r)@I1gS*Dv
zJ%64s@E=EO@3Qrk|1M$hU5c`H6u@<b6wpCMA#-`$JV=(1=G|=?bJeCu{dpyzzA;OF
z<$afs@8;pqo%qh7n821n-T1=d<`0#~ZUI<&Eq9&Rh{qR_rlFH@#*4er`Ztfn{gzw1
zWlt`k_@N>dD?J#r69v~56%%*tKVwQ&(k#39YGOPy(Ok~DT7&K!WU4Z_hR3yUofWxB
ze~94^9Vk%bC*z2JP7VWawlrH$-U*q8T|+zF4|rot;r+bY(W&rpqs^zS`-gVEam3HW
zTwIn}^J{s0^W`nJCrT3oVjkj|MMjZ|>iT6zaj>g%th(xCF(%*bWo05)gneet{<Sj%
zqd9yizhsqsfV85WA&^!ScMySGUAFst9MKP}Vx^t4Eca7r673qkGU?+xS-k9AnuOX@
z>n?aCM@^Dk%7WY74<FB1^NRe5P-XGY!yGSOIXvBshK`}%hZqV~Z8drk+Tq3~^6OS*
zOVKj~7~hkRMm_{GJ$_>zNn-*r-$5;VwD_uNY@GuP)O=JNq1;vo-FV)gM<H1Wpo4vM
zu>Xe_-n)7{DE8pPO<*3Egq%Dl@fjHM_9T^_;9ann7qRh7Vr(DszAQ2<ol<SfC&CD=
z8NL(`+(2hIx$U0Kl%(v{;$2}`pUa@iilzyl{1*T;K+C_r?f$t?7!@3zvCdtE(^orT
z?R6sfCe_OHO89jI?5)Y%n60`peB@*q@@$i+uN4SKkX5bYL66O6HPNrUJa)NT&81;d
zVhIEYr#U|Sa1YV}M?QzYWiBg3C^kif5#NmW<5VLJ8^G=kf2^2LnQvJZ()g+Tp%kFD
zMex=MonjFZs^mW{6P=d_TDTj7PGy(-qt4mBhIpJw@J>ng3$w?z{a|}b`W#R`ze?2^
zLFG9fOA^nU?>^c`erdL{c<y}p$mG4A7Y`5J$-dE)vn*O-S$oF}n71Or>CsIjlB;nw
zG?(I;y;9W*JO)heyfqe9OfOnGj}GaIu?8{4@%T@f3MzOFeB9jD*jyYG>3)`n|JaH`
zcZYCzR?uIWNT!u7b`r9t;O~ztGiYyvy)d{PP*68e?Y4BIWzW1Zw`SdjmhEU*e{Py5
zwER@<=e!RAZO4{TwsL=v3sV!W?Qi927F;(fYh(MvYKmn&6}xw<yU~z@O<*aiAsk3a
zy)+L3)+{Gsdh2var1e4SgxHeoERZI$9HUdEtDv0uY*ZJ^$0-_@^{YMtiadD*4*8h8
zvrYwnuz)oto61>VEs^^K>n!ub8Ud}E@o;agKSiCU5H<E_{^E;{QTTEres3n!w`{I%
z&~4;w>ne+>A#>;Kh?(5)RfeOk(D43lQ*=xET{o$|<tIF!kkt#^n#jzLA!tSWUsTAd
zw;f)$!$8#cH3zPPOiqLlAEf8e-V2<gP{BM`uO)}nasi8x-g9C&^2drS*2}tp13T@5
z>dg>hP!e8_9^d7y@h%35jq4&CcG$pvu}XJGBS<At2;7U3#&vkvM{-)wlPd`7>|>lB
zlOLv`6r{23LZr<H`tB~-vMU4;JsHNgKb0r9+xgonaeGsrBz;~~uMn_t!EONoWb-^P
zwX+$417Lp78YVHC@NruVBoO2N=PF7ATzVuLOy)>qp&*w(aMDg00u1{OQ<qp;wOChn
z;6!Mb^tKb$sk*xq)EX>Ify7brRRP&X0vQ*rvTc$*hm^OsQ-t<O{#b>4`b;!VuMIax
zd}U$-es8|$u_QmMp}yoxR!(twmK{Sj!}STz6^ZYdw%$tRrePiT>=1G6R|lAaYQy}q
zd3wK#D)57=4YsG4@vgTG@p}l(L5Y*xqp6Y-B6$<r{9+KO$DoSWH*-mJ)QA%j#$0+!
zOGfgHdGdfMHM2^i&KN&<na;fro)bvwz8;}n0P>`<_@QY`n9x&T@JD7;A)=_FCq^y~
z0~aC3%dL*bSxG}Ya_3#ECbTbPIp|%iEy%*%9aOVq#8Kh0`vTA#Z_n_G99O#BG(&w+
zV#yM6n3NL%&KOp?cS~llCtgqpFMnd{FMr%J+hDbi%J0wa(8lH_ozDn8%KU!HK=?$g
z`!X-)z+;3f8+5iQF`x*T2aPkh-fP7Wk5V?d$cC%8tbJ3+JznF&1tRv_4S1rE<4FOc
zhhaPN+*Vgh#xGzTHt_Sm*<B!B@2IsU3dUF6p(7g*DlJJfD#d{7zG*u8K_aYOyQl5{
zLE^3tntYcO&b|T0K8OlO;#YOwbzV{6(-f^Xw@3~MG<+LFvc>|u-UiCdUr1}IRUfLL
zczbdxmtdu4#c4y0vu;R2Mx@SMRe)*H)hp&LLVPvK9hKRKO)Qa1(^Yg{O|J5D(u)wG
zJEQV;tZNOb(GHN(0GqIzYx6Otr$~6v@B+X-5j9DrSl%%uV&q@bnl}T`P~`2a(FaZB
zN&K&`7M*QGE}CIJmu(o)>Ie*1;PZHsyR^(l*W6>bq*G|jdR2nDghH5R*X_iZ@KGnv
zJ(x@|uY$5X=}P%QxY4kc<DUEOYV{p{L`k!$tE`OnFZY5yMaNYaS)fLSJHdw0tzXMi
z%Fv!|kEOK4q>ZQ6&0^DfkuWanqC|?-i{m<U#f7(XnJ!z=;2FA}mt;LoeAnD6CWFXF
zFfZv<*snUO?Yu$OBtFbF_f<?hK*^&6DVJLu7t9tl5RS88kq*L(U)ayFBQ(?>ui8y1
zJ}1A@{eeA`c-Y8K{SGUi2WO;Vm&(McC{^-YYo;sFD|!o^H?t8i{9sb)#)HPZ3(gZp
zu&(1)2=3Zx&M{XHWh`$#MQ@c96J;Ee*<#_)VRE*5E}l@?vb}l83daAiA1=ri%<QCe
zXwKIoHw0xQO|Z>2KSB)E)LrAR$9&xBju}|y71ufH&Ls(ZAdnd9rHN*qW^AR?>sIX9
zqlgk+)pIS_<${3TO$A!C7+f()cK=euqQyaE!;i!1NdPkv{fTq&=vb4sZFz5WF=d~@
z73m^&Wo#e=J2kM1BkR~k#ey-VnCYbty0<i*NzhA5Cb$<O&L{oiZHW0N?kRI5GD^n^
zgwn*&kf%$pwsa(OWGA(P5e?V&nj{h|P%;OUF2MeYoAmQlpRRxypUBqr?c7#qGJa7J
zHEi6U2C2I{|55iNWL|TF#p1{9alHY5{WIn14+YrG>Q5J@+s8L!QRV2ofzVq~_JQ}5
z4jy}Bc7HyFn6>t@Lvhkbd7;pA>d{WOuLpfUI*S*VI&)-(bS-ba8fj|HAdWakb5zZp
zg;x!=cD19e@A9r0U@b8+)h!O(V|0abg1ye>ggxY%n&L2LAJkL_0q3tr>UWawYy&?E
zUt(oOqU$6oPBV^~mleWQW%<q1Z^#4nW!DP0$>E5u737U+=IJh_<7sDK#R7<lbaKds
z_KDn^0kiG%5mSRs?Y2|(D$;m2ulrhQ>U9V!FxoqGD*DG`-?w=AQ^}lyy9VEboh)*t
zT?=GPm4l5ypcN2Y<r2PcuH(>kZad`-_vk5I7JJ1)-#T_e-5G9wTI)waI257dZjSZa
zg@%`|$MGC766QYz13Xa(`!)q{rD@o}O*|F^SZ=ai)nViRG~k5NmyWo0VPfFfM`5vu
zWhLW06?KjZ8~ubCCZ<Ti4(n;j!=6c5`dVln&c@7j1jmgY?3wJd>UUVo#o?j?H>2WJ
zDO)A%F_`39Is29`y5$EyRpLZ?Li&yAQ0q}UpNFBb)U;*4Qi%4%1}Qj9{7OCow5BL|
z)sg-)=tl#pK4EL}^(QzBiq{8tR7;S4Xhdc12^z@ch9y5npXdAwPYTi_e<YO;t#akX
zA@6YadL%Z_4*3q^cUI1d2&(6ANBB6xR!N2kM(Kr3645J&AG#M%KThU^6G3|=*|;UH
zB=oT}`m{s{aF7OE`&w*k-ETZFgdEdis0a>%H<&=8h{xejTI142UEXus7k&Mve|BUV
zEA7FFz<eZpG`cHk=99cDcv6)&SwvO4dm_dN`^ce$`Z;S0ZA$(whB8@rjMYNtx@&8#
z`n~uqO6Rbo3U&B$$a}^epVT!Moty!{i|di}s;r#h))BJYG+lh5^qazQiN*V%TQV1>
z&T;H*gQ_a6GXM|cq)-*V*~NU<EOF5FB|IXZn)0_|1YP+h{HnIV!@6@X@9}0zAx-eg
z`?m@i5w==|{u|=HL-kCH%b!6ZZBg!#W+QuabOD2Hp#dcr1{&iKlJ@?B>H`zZXaaG%
z>%9IStUljKnPlEMB78uwgyv6M2k?b^Ehg1gIG*RZ&iAi&3t#;j6d|s_h?QepJZ(hR
zHgfK-5Eus@_P>#u?;rHuD1_NG;_MM`hnr0K>I^B6?`o(-nZ283>pVwKSn`40apB3m
zRS7W(iQF_jriAkkY1*cZJzefXe~;DpVVo=+N>aE)n!B2$lnL4RRVzey8~3RF{5%<S
zMOG8&OwN^APK@6h7dw;8!$pdwD3_`xIqs%rNPDrpTao+hYcjusCp+lvLs9=^Fi~*4
z&CDg(;)N1mQdb-DD!O4(I%6g8dP{41ji=c&!{#+SF-fq~>E*@Qx6t7JJEWJJZca9}
zZLa!CokypDAUgsf2E02CN#`gV&hBS)>@RzK)g%;>2JE_<e(A&$xwvbWFZkc$4FHI5
zG}_&5zg-t6F_FZ4G*OeS`PAs>+MrM)hAC|4wLx*6y6eS7Cv%QTNI5k1X9`m&_QQS`
zhi1)^kf;?2Qdw%H5_DvU2x5BC8uO_~+Jw=aV$KP2Su|7!iq3sCg~azd>Z_vgpC7H@
z9f^kw-)R{As>g|!(+nBNeZHK}BCL&n&)w`&LLY1aA&5tN0#5;F&s>AR>(lxQqOdI#
zSRgTnTcs%?&d!M<DaFw$${sK1N&FseAsmptWPxs%LDAESJpG7q3#T3Al;&j-e)3Fx
z1scU+1l4}l`NzzJ^)3yy<Kf~Z2OrYm5cgG=+5`r%22-E=T5yI^Q`8xHCZTf%C5Ba7
zffxkv;8`r)|6KP#1<e*_S#MT%7?^Sw*`Pdr@S#5U%n{S@0C)L0_XNbUau}k}9Lq@%
z{+51+_%Z1;8>U~E@#NdNa1m_uI7RnMYH%zKyU%5VAV^-Io2~Ok%yhv}w^|MAB!pce
zmG%&TF#;JYg7GNV;zkFDUk3sV2)%Zh7@uz>L*v)H2VY~rbUd3_x7uD9*X2aceqPRD
z*OMA(ufPT+mep$@i-9Q!>J-&(BsV$Hl}?<n=;XJqae0lo8dhsrgw6A24PYVj+r9Q6
zsP`W|3$(F@`Q0qxOkCIK665Jy3P%5SbjAImYA}E`wqv?7^o0JNm|8I#r<XwGXEKe;
z&+5;PF5T$MZSf56xO&~fM)SojbUN{#&NTipR)3%4F8qlEiIHv4z%P69+;nj=hTm$N
z<O=ns#toK=lK7yUS9<jB+5`JoR#;0GqZ;&lk78NZY&gr^z^){{C++*BT#!YpEkFLN
zwYH2)L?lI%Zbh#XQ1*jGW`e#^umLrIsS(-GHPkrRkz86n6$5@)7|C0aX>68+hUGyA
zYyk-@Nd~;OK8^^+%@WUhZBY71XyuW3cKVZIPySg6xqF}>?)Q6)R)*9S&QvQ9MEhk9
zNkh?kd*H@u?uEYyk(Tt6`N=g^z!?gY<c<<2j$)Mg2QZ)zU}~ZWPJlAL91Hu33HIIv
z|DDnkz1<dtRyyg43P$)C>L+1iKfWGrbY9&@>#NT1Ct4p$@OD*LrkNO}M!>9B;G1~Z
zNr~(9Jz@hA;!nuOnTqrZ<(bMAqYkhebB8~e2NOJI(?sp_D`d-UqxCaSa1vhQwt5py
z$`RQHlgi3RS7#itn%vP|w)gVW0#j|66b-HQD=~zHW|T;1?ZI_V@LgfA)UNuHl^=Gf
z7>R?BF@pz$Q_9B*65DfmIcVC0>Q$evgpF1A!n$#^?RbI2?syTXw9f|>&E%(>riZXe
zQo47mX7$YvtCcirQZfClcIoltIQO%e=EoL>-6LE>{8`MVXn9;wQ{FFBj<0Bg&U=z_
z_0dr@`gNuCKl8{z*<fBplsJuqeqF$6s+krhHWN;Grn`2g%5pZN*bj!@-stVE5?dL~
z3_P-JC@%;HoasE3lPbaF?PY@h67B3(ew^qB3t_tL3ms5a7->N&EsgyGg&JojSH0P>
zz)(07X;g@YjHAE<#CgmtXpP#YVz=$;gi?OwgWK!0qo-JEc9YG03oR+%STx}urCfTa
zO=WhWML+jHF=746Bu`z|G9RbE&61zF9X+~L5j(t8Z;68|*S+jEVz&F%`TD1qM;54g
zSCZrf^r_G4v_kt<73CX;oim?NR3}0T;R9spe0;2J45<nrSnAFkb&&|iDFc^GdeXqu
z)T=gj;P@#6`Yq!yk#X3fZ^W1?ZM>2v%M>ZEn^bu;lx1NMePXa=;hlRRmi?r@Ycd6L
z?&8Bjyw)Y0Yo1g#zrMR_<c9O7#~`F1u4h~doq|x~N-r78U4M|Yu6K$uhH3q2p|~(K
z9|2ZM_TQ&rJYSt7&(QT9YEile{(M_8?Z))z<V&5}56Cu2G91`U!91nLa&LabJfzTe
zUk#2p+jXrE-w?|<Ma46xqX&b-tgr^Z6%^pcA%|uRJ7+fUQ1Dq5*gma!&;B1=gqdEC
zHK`*FICqlPSjJC@n#+lB_mM&SlJ=rsBH)Xfup+U%$Ou4Ah)SN(&sQmqS6L|dT}QY}
zLP)Bz6h*>W*+?cviM)neyH}06t)HFQ;*axj`nvTu4X<-E0rQjTMX|Y@er16JmcY{2
zJ~|Ikdtg6w3&0d`+_k_b<6>xEt7Nj4+1(bx*^tAlIj&1trtg0QH*62jU!L-MrRG~y
zX34$Vq|LNc{CYSl572J+9#V<B!C#~huA<pK4lVY7xjLMCkbie{^yR@AJ>Ik>6Wu2&
zGGnHxq3(7o^|rQ!v9z;gz94GW%Z1kTre1!j=F(!<`CxT++(c_c3A|RR^E3hF4?oze
zXUyp9a1PU%$k?v29%S3Hv{uF#QK|20r3+*c^h};nIxVo(pKFQV(>Ep=e@4c8DzF^Y
zNR#wcG9`&1oDBm1EP#Z>8hkLbnd(fCm9cLL3H(VzzjK%0d*1?Qv%L-~A!Z3Z6OAtZ
zG5SGS%up!a^^>LZ306ep%KpyOLfbbVW>uK9yeyt($I}hQdWcPjYzND~MW65#0^O-9
z6HH$h-H_?3{5#xm;$HbSfl4EdmE@(hfhA|z!<(nPcyDm@(W2h5ogV=PBkJ5$8OA84
z#`1585zV)Yz)2Q=aitGluMZ&LeWMIxEey*;I^o+o-ylXMk*=(-WB#!*{eF;`0l`VY
zl_+W92){Wfw8RVAv^jQC{hM4Tz<u5GXLa@R@^GWSoYd$X$-zeBa}G0?_K3!Mjg6Kw
z9&2(2_q(sju|0;DwmbYnx7?^6<nfOOmUfhMXRHw)%gY?t0COya2lY8c^WyiuGZlwg
zPTc1oXK9N&)|Ir;F3^;|i=FMl{l65^O68gc89hvbp+ikh`@9J|H1eNKO;Ubk#YM-O
zpbLsuH{>sXQ+K%=$ie?y(cjqD3psft4Kr$r+bCQPOk>~xQ*mCTQt2_qe%pSbTpOjA
za0_t7NohS#(V>^X`n3M_UUQc`QiheDgAuFc@M0~8z)COv5tY}Dbtx+lT0g@)G?cIQ
zyhUS3xG<*C8WX#|v%N4MSfpLL3ev<^aGKoHCAGE(ac<=7cT}^c?*%9+O>QtNbPv|x
zYd2d$+?~-K*2F=fnvGmPPAPj6;Qp2dl66`%5~Ca-Rj)4_21}EM8fOsk`S(4216mS4
z%du?l@?nLucRV&i5mA^W0}jFwLtDdPuY7$Aw%Up{>`K&!k%;KzsY?WsQmFPR+&_3;
zG3opVOsn{cR%LvGHDpP*(hWEG@vagA8=s=>(wn7;d_#q009w5v05!>_DZ)XEP<UvA
z<Pnl`pX1w(={GpS_Z$f+)nn<OTqZ&5mrP@5-72&}DPb-|^tbMZuDy*=u4_ou(}W&5
z3k~v)tD5WS`r2B|LSXN8-wBD|>cs{vE`(E6H>IBw5AgM76weC~xRmVK)EqtXS<clt
zn|z`3-(2qs6#{+J=Em<d`PB;aZ|iym_rk3{hd{K2KdB6C69W<LXAVXl0~f$;v$xgm
z;}%2A1F+-}6UK5gfyJpJ+5}a)lAdx(1O^!fqU%zae7p9h0_NvrBx}@lQ-q4YJF`BM
z={Xb!%kZGr1|!=yP~!9->4gHsn63RWiu6nvrS<44Klde8srTiXJZk?`f}9_f#;PEQ
zzXi(de#@lY=$JT#q?Ny3ke0ns<8w=|&~X)hK%AesnW1joj}nYSSF$psDryL2G8BTe
z_XNEqwXiNO+{r~qD0ozjJ#ZPpVTf@6B16wKfok=ek}Z+NFc{axl~hdKTJbWag9@``
z0yt(&S<m|<z7U=JN~`_)_(aFoJ?<G+yOYB3bXhcGf`68%IvB$E|N13A<v<qY(r@(5
z6g38VbPY3L!a172A*MO~`Nk5wB;kQ1OGM!#J!IIEw1iG0+w(W9XvlUMv??t$T0Ly<
z3Ztz>$-Q7zsZ#mb+h8^}K#=(!VZ@R0>!2HBbBSdu>R1tscnXS{Dr9sgM4ow9{Zvbe
zK8><NP~JmP%77u(5<liwmZ_(kEF-BUK%z&9^C~_pT`h8y#V0?YL+GattDvDx@O)kn
zvDe$wAInX-l!#;1<BwoahD2<_?Gug5%N8$!hUJwpr(lH)zg3J;3+k+6(^h}e;b5$~
z@1^z{jhe#Z3D{g7z9@xe*|D<1o$Y{34%q<eo_1_K^jzPbGI-5j@QN+B88(F9_9X5}
zon-Gnr?`E4D-t4jVfo0ufBb?@r9GJ;4mR2J-aQ-$EyGAeww9}#9@0YHm58U}w{EQ*
z3~G}a&pB!2x2V&_sCn3l(FepntK3EX$-drLA5Ab-&OL2{pqVZfkKM8wF9P(_W+EOw
ze%MLCZmD;Go);sH;e4c+$WaP;G~3fE@XdERSdD0#tau@}(L&v7CSlV>wIwQ7-&BFX
zZH-o~R~-FJljkc_Cecv6dCcD6vo{axS#tKbK-*I}*{Dpn_rf|$%bu&3FJRx-wBJKZ
zk)kOiS06JyeXeM-_gG>erjUyWKOX`cUc|vG>dQS!47=7FdsIh)%_JB}9`8*uArgkG
z6j_Lzrs~>ZaGWfQHh*XU@O}c9d`I<`4#?-e4Iv+_CAgf1KJtjWflQk;+9x+v-?@Ov
zxx=JuGc(>MReRgW0LI&1Hm$K{YH-0Z0fAZ&nMuoZzF`N|9#NwAJOtmkLwHn+Qb;jS
z@E&f?iZ@O<r$P&VDBr_qb$^f0YS@bf*aFjNpk;nJi(JVt^rsDqp)gT_C|U>*VZM2c
z;uSW*x337i7A?YBwv9yj<MI<j=y?84ucKo`Yfktz*j`BgW$G}n?cCxK^j$f*lpiDl
z>wD~jLpis>a!3#*k3r<AUhA4!y(yfbbPmMpovyd9pm<<SxcCc*YhdJBlb6~Z#{&<6
ztiZ*)n1W*JP58PtFG7nYC%cEts<XYb?nW>9j&ONAh-{Q5AseZ!v_2_~6m~$-Y#r+;
zBx%w_c@~|Q>tRM_SHb2sQmkFSXok?X=tjGqcKV$~k0bLzh)u(AkY3#M4G3%>sou!%
z7lv!-bg?it2hxVm42XrMx1~rFf<FzHi$8yG(os2~{(Ry^8`OpmpZ3q`SF(;C6|}h!
ziK3<coU90G1M!Q?hUkSHi`zYxJ6;)M4kUX`vfrwDSy^IDW(Gr8x&DSuHr2-In=e~k
zb!*qu>#}F4uX0ICpTrdjciQPS5_nMA$t?-JVHJTqWHBv88Th$C1?p@Z;|@-o+1@4h
zal6Mb$*uZ{$zu>I7~wAMyJw}evi^!UB^SHz@#RHT(D|`Tv|3_NxR8RR*%((Tp|>%~
zs0u&}LQ3g;)-I@|rg<r4;v%=fwkvokTY>oe;LEH8#eLSj{`eL&m>vf{VVKyOZRN3L
zyMs<ebDKprR?2pEe2~3k#4gUfkaTS$m4u}V0t!Qs^06HO*ruDu{q!6=G%D%f>n4iv
z$pp>CVuVuTHk62bDuSjlM>}tihUCp*r;8#(Mar?FsSl_3782y#n?O$CI0q(dBd|Ac
z1zVt2+RU^yn;=d-1@_^k%JJn_@TPNw(7nR6Vc8!5!zQ86Lae&**M`obLy?Av8g$mX
zl9Ac%I3k6?7vgr^z*~nV(<lQ53J_Dtkfq@7brG5LHIJQKp5U-^!LzP&77uWr7>``*
z&CV$$1xb2$MQVXT6%)W1jY~oH)po51)KgLQ5coD~2XSM90?=>uiQ2g3`exD{o<8j-
zi_mOR%OUh7UP$``zFG)BeT0=5cSqCCwGsWZ!nmTSFvW8<WK*fmhWwYn=Aa(FFMoLn
zw;e=+24)zM8a*&gNo!C5DoQ$}V>&>V*Dj>k9;DTBcxdq2I(BGUv28I$tk}YQM}{{q
zg`<+}NN7@R@Fhk$Wa~amZXqGcUc2!IkhcRmfrl}_gal|~-N7<Y`)E(B!5cui*e}ko
z+P>P~FCa6FPKIGTl)CykrCBRy*>;=lLy0sNHpov6#KO}8!h7v15BLcr6`Y|3DG0mf
z>*AZouDIzakrASZh(hm2M35s}mTfW>dns_d=RVI%Rjl++`JE|-&Pi9^t?gY#x;Wa$
z(Y<^lv@52Y5`O|jI=j9kve_@<1()*o#nT31K*rg;a^dkczLV&q@B}e({9H<%!WEL(
z9<nMpwJ*e18>CNh&Qh(QOfpSXj^hKfMx_rk7N%SzkDl1jF;E8&)%06RL3q4n6)HnV
zY`(%F)HAl5W_=GlN^)8V6)|}UcCP?UrpB#y&{%j#Jj})>Knr2C+xz%dNw-s`*HbeP
z(+LvHS2h;BzSnKp-c6DZcrqaMdWN9vd(L#TIB;j0Wx;hYe^+aAiPSsz`#_vt1*#5F
z(YrjSlC^6>(sP9?*AA@s@>H-<FZ{+lFu7s<&P73x&rQK_T3f5073cDPs6!?Fy}wyx
z3bFp1VP>fgTxJJXEx|L2S}Yh!(T9&r{m*7f@pFB4QL@B`HhdWg!!|Rp`%;m#6}|>W
zUJu7mZXer@$9FiyDvYZ@Jf}6WU{OwHW}Zc#;vs9%h_lm7eYOtj@12<h?;R;^)WLII
zwWzXIdI1z(;WO-Ca2p3`*7Ng7og&A|S7=4__EkI)A;c;M52B|-VR!sFA#%0DkOec!
zu_?*$C@CZHx}=@eF3IN`%4$%TEDm!<A-;s}mVb!wW6}n;=`A^FUz7=-_k_xQFdkVa
zqGc~ZjHaIOy?c1mWv)k$zNif&SacAEoS~f{*|TyB*k*axE-ToZVPu@dyI|$cWfc^E
z5XIN3@UlN>$0vXt?EF%BZ_quE>DMGLvum#8uMk`83HmA~F-DE(3#e4A2Z63Rp^m+b
zR}0me;R+(pCcxd&H!nBtSry?RHo#cTaS?e)bVL(b9BoCuL1ye+fVdA3wQXeLFu3=k
zDV@7|&1%6OPn~U8e{hmV#O2ZVhC)WvX+4ZIx8O=ju~{v$E794j#@5tyvuMEEEF_XJ
z<b9~?ER#N`_Jo&e4=LvxN&57Qp4}(xgP1i84p&UXIQW6#&v>;m&4sI(rye0>2^zZr
z9UP!U3!@*Aa4-Eius_%2GF7;F4mHd&PDD399c9mZ#;Uxk<$+}T8WX+<Y1|!dr<~*;
zfqCVwt{w#60<B{5RK+jF5rl5MoL(&P>iuPl!-bGdeYCZlwi8_K9pS(JtJD>ALufVk
zgnNYyIy2g}Iq+9AzFC)uqFEs_|8^kec>=*8*f=7o{3KT6Jf-~?Z)I1Q0hUd94*^*M
zI6kzHdM*7l8-rjBx@w3}_7m-%PkpWgt7Qyw+Aisou>jL>%76Stb66`Aj+z0#?^;@^
zU@Rbrch}AM5(2*T;THolK1dkLMX|(4CdiaxwsND@RA5R;!HL_IxtPhT%J3^p(r(x~
z7^%Gsr8P}_k$JE3X}P$Sdx5cAxmfzu{2fbE>Ypq63a_aQSIva8L-cQT^iylpvK$Rv
zzaMGLqLea`@kKw0a-yc<^|N?Qlv;L6iB@;^I2CO}IWKKMgj?tpUxs|Jb*-`IbIM7s
zN`Lq~<~pDR$YLJUJSwRBz+n@9pBzFM_}<ZnfD{bWd`Z<yFz;?oDej}X`YHzOBEON)
zy6Bgy_|5pGhJ-OsRJkb@^kqRj>3Sb5Y|oY65?MDgKaFj<g&?F|Ev!f*Xa$`Q8P__6
zkJD`68Kn$*UPS(+j4&%Zj*Vly7e;q=T9!hdM5a4t0sS~__{f<|sE*}<*tD&WO27=F
zBF>=tbM0-`#F8Lrg&POMpIMsah5%}5E4cN=Ql^|YlN0d0j_bH4V*$RbV9-ornW#G?
zle<z^?m_TBelZU8rTgy!3m0H&R`#A3@@pZ2&pC%BQE@F{4aBvaS!8SVHHS%?pciR<
znFsYFAavJ8K5j{l3T>Dle<qncvl>?4)7jn29Dl!8gOSl-?RpFaZgw3Ci!fBPfdl;Z
z&DIHJd^jvao*2(hP(4A_r~nxqo=K5>CFmQY^=wGAZ4i9+WCZ9UYv3o(VPDm9F)1if
zGjiXd;8BpGU`*lK<7#1i2>nZ^dTITB{0`_4o^w()nBjVO;O9Rqit1TjR%q6*5BuD!
zH~)EJUU$Ve5i13+f=;AJry?GGH|pCKKQjg{$Z-{NILoPM_fS7mEpjXQ21QhLCG*m}
zZc+XVXCvKcn5t{sN;RIU^BYU^$<0z(eb$Y}3-@8N#IG(|`?_`BuvLKK6oWY@1umj=
zeX9kkc}9b{Vi)6u%<j0`3Oew<x)<Y6{_wXY%w9HZxpWv+ZEF}lltyKo1<?_-0IpW(
zeM1Vj*P4=I4|=Lk^}LW?06{!Y0RL_*f{Z_91=R7eW<g3?v_i%@RgXfB5Kko6f<R(S
zIFinj%Hi9uBCAR=Ra&MQILZ8vw~rJooCNfa&vU1qVO5X2c(W=9q7b!Pe2|;6+Lu=Y
z674@CCCG>z9dXxjOva@eC}XIYH$`n3I$eSy<B`}_YgN!Q5w<S8=@_Mlc1J1RPT$UJ
zQhghJ>mHzUmW?(+u)D+P?b^@pFpb<&2<VM_6%L1HVb%&oyPK<qtFvERM-bN~aK7YM
zG#Y)s&@H6%2o9*2bMTW(290b>p++L^ux9W0&{af1YqBP3?n}?T4r@MplOMKfT@C-R
z_rBMheq1<p_vtl~%0?lz(+xlolrtGI43KS3y=Dt?L{rT{>YTe#-dmR(f|E{sm?mSa
z|K@nZ#ob9D>M19PFR`NUk#E><GyzDtz`l*X$5cOgitY106w<s;N0JKhodguT+7GXA
z8B1jHo91jtTP7-c|2bEbYMJpZR}GK{T<%B}P!`%7{`Lv8xc#87>9eYE?qd||d64;R
z{5}4y<PX@Q8O!HRIOmW03Gjmm!P-zFySq9<2B`r`$O?zT)H~lte#U~=hs^ZKD<-0s
z00n$TvzS}(;@`6FkMdxihY-bbD;}(`L56&^l?h6v6tO(~wFQ6fyTJmBdpF258>Hg1
z=V64XYXX?zTc#{3c{NXUQBA6^E96>_v4Ao^uyxA^F?yTm$^^JrhchKS=<9##6272v
z6P<;aRqf8_^N|oaT&flmz_Tj{_WWeiW|#{ixu*WTce$D_wn+<PBDxb=X*t$!3>r23
z-6~j#^7w;`2D-8#!#B-z|Lu>FR{SdOJh>ZEuO`n|o3_<SZmBF{GrA8N)vB%;%<q>5
z<#W8Wr<__jL{W!SkMA_BCFNB+b^;&#0(Qf!e$LF9mHd_ty$tMT#CznVDB`12!(zcJ
zITMYLX1=;NEzPcGmx1G`fDHSHaud_Hw9Ij^T2gOpgx*-Zp%RLlLamJ2_Z^~D#z7v-
z32(Vjw3?Og5{Q68#y?mY6&JpfY0^R+ud$x<<gAG;5j>ny(;ZE6k*o$^{4_tEx?|vl
zy;c1}FX2tm8@2D0+~Zd#qugY*gz}(OSv*@KyIU%zZ?k>?PTRvT)?3|r*S+<eJ&exL
z@GlkNp*v$9W_3eUISe&CH3IlTTL~t~_FXJw4kKp(mJLOz?qe)Ox58H7d`tzQ>Xs%K
zd-)PJy&L^bc(}D(f^gb!k6)3Ma54~CWaG&|aV~es`8;D#O3Jcdx-N4MlM{(VxG^k`
z3Yutxr{ABIgMlu4rXH7k$^^?GJbSKZ__rFmdjDSla2k*0J>9+FFp3?#v@3yl$VNlt
z)`TfnAP4C>P%bMq7b+Me`ugq`fNPWm3<E0qM08kCz&o8gcin5fBBdTQibmhoW=l3*
z+F{k}wlW=Til`wBL94trLNSHzXs(JP$kRpz%Hy0JDW%xfc6*5PUv@C{*H8|MNKfXq
zkinTXrsUJMrsJ29iQlT^n|@9Vza#}hY^yu~FJ&LKLviFQ_r_`Kl5@@`K46z7(g`ib
zEDzAALQRo8A4l{K--GSKyflyJ=jzc3;A_6}Xqh)@sehv8ub8kL-DZ^3_F;Y!-I5jX
zf1(lc?}u0mJ`e!oxn({Hj&aGlWlp5huXS=gsHh+0Svcbedw{$<*67Z6tZz;dUc=%F
z&P-bNvP;aSLKvhg2JOUeT+$qc;LdUM!>VikfPg%?JmmTXmPHObmt7q25|;tqZ-2|V
zdnkKFCd@mjeDBEqAc6OqR{?JH6b>guHb|Zl-Pnx_OOrluKBqywb-!wOAZ@egou&Yx
z9eya;=8&h_o`!7r8f53FbE4vU(pCwEqifJZZkKp+VI^|Cj&?+^Ts%xtPt!8T0jc_k
z`fW%X-lgCqc)<joTO3DRTv3)gi%!EA4#_RDa&Z4-Le1JKEEQim`z-%3Vo0OQO<_6<
z+Od`%zX`9}0I{VBQErJEy4J&=YzFJw%V-xvW{_S>&5gl7{jy(2Yj8k^<hWBmZ=4k~
zlhBd`O`mSoDY*co%pVgEJ86RXXsU|J>P!QB&yH|cyFZkbuQI{8K;W2{I(3ot-toO`
zlxi|tU<KPD{ygV{(Qu4=pB4Dy_GLpODZ+Nk35YBn%IoHHmi7oaSweC@xKqhp{!W9U
zV6Gbqw1pFqsYr?tYz`gD+eQgrW0?vmhvwp~oo-F-<}hSPjy3P2{Mv=7ZY1Du!>F(-
zI8ogr1DIn4qyEVK+lmW#Bu)unokHVPtA&{iE{nQnc^ojyJ^{B|D^+*mbXZbB=ou-}
zKIGJHSYLKXWEDV9%iDX#$;T8Iz6P%ESild(yl}$qMJTa}jdn`vxOJ16yE*Q^p0N@W
z)h83TUE#JpL4qKfzzj#<$M|8ta3AbC0O7hZ`%yt2)P+(9ooz)q2v^juoPn7pV|?YH
zvM;h(;=Cw2$#T=}amggCxk~wTJvaaB$oQJu_~I+O>2Qws)*Y#0$uyT4*&}N)>hMfL
zc@k?lhPo;*KB6TS$EVHC&kvkx5nmA;o~IiK9KZxGsu43D(El)Cy=MsIaNlSgeu-DB
zN@RBX1mA*Y$;IhDEp)$3{sxjE6s~Zt6^NjO4W!L98<uXHdsRzdiayDk#Xw`nqu`FL
z-WS{93pkF<1}##7z`&PhU!>afy5aW}DP4lgM|L?ct*ghUx5RlaSvnJ-{-`ICYIN|b
zH_Cl-W7WKZAmopuRN-7ve0mp94N<O1P+XOeyQ4MWp{y?2L=D1tqIi{TcUjRkk!I~o
z!ibCCwXsgaa#td-t<jmTDuWI(GAYK1l*XBcB4O$ok_R5!5ZMpmB(8H@U_M7Sx6*S5
z>4cK7Xq#Dw2tTiQkjP$;+XIkdGBW*is27RS4RcwxeZke(V<JQq<Yh)GqgEtjtPCyp
zPuT56$aR*{6d2@HHQ9^dxL)n{NaYaO5|x8Go<H<alZ79yFk$%}lxwiK?dJHJBtY2_
zY)Ml@SDhXJ^}&o9^LW!h%DO_Em)s>sd5g%h_kk{LRzBO-O7jN*ioA2?hVC4a(bnjO
z@b7}DEk5IAEKjgfXszRa(_Ng`8FUaIY&h%Klt_#422-JKjTWT(!iEb|0|H6QBI1;&
zfr^n2m|JEPDWNuw%Y*d!T!dcZ`#=;h%22V{PCVs`ZSCD9&p)DNp`iQf`gN*;E!1k&
zh^>py0!eb3i`mJJXg&y8?J9g8KW=(6P0*vOuT&@2TT0sdB^>gTOw5h*t0M+rwJ~ca
z4#G4O$kHj;S-3wSIN2XZjH^S9=zKXRU>IB&BIVtZtG&=rk`8iPZQir&MFRIW_iF1S
zP{=Li^t8}6xbdh2S&A|12uArAbwd3B6=y+GPdjLADW9|Kjr|RG#9#3pUD=?YH>zAh
zSwOjC`#K>N4$A^c#(dvh;ZU1Iwn>G@tjStpD&G2dHY{vbO3zC^3g$4@<^j$<_x4n?
zjDLqi|1^Sieoqx^x&61uG5&oGCnHDIPQ4$nzd!6O*WJ`5+PBO7cbBoS0&=cx1&LK5
zBZXKs!R}cE9>!_HFO5$NEmcc$`b^ivW~R$}{8)mqnHRBJj82*?lN+||b=|)5y4?g-
zh8$3xJ*N{$dK7_upkxhw76wWIlu4+_81hr<^nT;6jU%_nso0-AN~^OSUQsihaHv}#
z*ad@Yzo2EYDU-wEG<n^Jd2pxr-MU5e0alO&DqbL3d<&Kbm47$?OHUKg6t#ivQ*j-d
z)O_C0`E_WmB*$rzbP%bQPo=pnJ&L>#|4k1ZJ{`;Ei5U>&vt9bG2@ANc2<Mo&Xq_}9
zoU7|+K9BD1AS8??CQwn(@HmLvn#Ky$7hvnK&Bajaifjh`b|pefs(;$bD-;rE%Swqj
zjtjuehl}i>kByEW>?>g<*iW2N0#jitnCJ0@W1pO;KGLc{qI^Voku$41knV8Ah-(k_
z2FaHRLMfMMUGGb8r_{vM!MS_?k6hVC*HyngbrTHCVpmgd9(p_wF(|gYPdnuzV0WN9
zX$X5M2F?7f!cggxXj!Q;*MEy?IkD940Km(^Pi;mk`&3GkVZq>FOwoZt5wi-b((mHF
zxT%zaB-;12gX20xL1N<%58G%g!U#nOr6k1hra^N)y3$&y3Mj$f1o9&fO_G!XV8-3O
zpBpSdRv=16E!y!RQZ@X)GnnLGnStZ$R&+hA$6JD$G=V>ynOj{{2rO3YF^$M}5-L{(
z)$v`>k?dCzDEC3?E@lVUSFW;K7xE2U1WkwBgqoGuPTiL`Wnk@%pn(=i7Wq%VaYfr%
zcPkcSa+AE4ia&VBVC3`V6v(&S&fxr%T)^z}pcr3-ryT-ylmtmgSIU}BA5H0=Xu%4V
zY^cL11hf4i2!}j-fUG=(f~`dn9_bsdE>kQ<2D|H=fRul7nJJsJr)BIJD_fMFbc&C7
zwzjbtmkmwrbvp13Xn(uny17lYAZg*iT#MEb3C$zOn%ey&U+oZSRIm5`B-O-udq{(9
zxt8$AZ^@=L-I?mX9L(EdM3KPy+|?N!mfdN!kRK%h$X1AjCXMzZH9%T20f!k;Zfm<k
z3uwsmhFyJHMkF#nAUuPlW2FYA<PMS}?x|d*#Oznty05xMlt6)fK+hVsz#7#4+~J+w
z{SK!U$2_8g(;vKzm~>#O=Xx9Js<G6y@hg)Y>$*)X&~QVmptV>(a4N=&`(g`yJaXtr
zds4`U#=}#^uV1I(w3hI=%3TT-!!Gc}LqtqerC&4Ia!2u}#fWEZZ4iJ!zEdDfmr0=G
z;@?l|#mjPAbw3L=NLHqaD%hq`7_~Q`r*I%iI)!0}MRJq`gQ%aB#LZion_(FwO3!d|
zv@^4*ksn1%T+!Y>lU^elw$Fq}WuP?_)6;fp_y5%F>1s!ja=J$z<edPoLPZ0B4PG{D
z<->U$EGMD3z{fi~O#rkvLe4ACLMrs9X!jys(`p9SMoxFatf>3jXSJL9l<;;RejnI3
zF^AW27vYF$jzKil(y@5iYQfqJxfq5EZ$LlDEjGnB8?-YNBhtuN^6OBjO!ggikYW&F
zvP-~iy;u>VFM^W&FYDcBsIihX((;<swYpH@ZOoQC`OaGzwv73uCvHMDs^x515#rNe
zin=#W$ayZnS$=}XcQVSjo@pra401W)SSVoo?pGS~&+3T2p4-ydesURIDn#f6yL(F-
zoZ&&TtUHZ(I#R!<k>VvZ%Bk~*bS>0tapFslYBTwtApwr3PTswqCglPs_x8rs{wAk5
zBpQP?4TSmm*9naKW?fqIN1D00wUl~SaFc+W`f5@GGLQlImp({Kxlq#|h8qA~*q{Ld
zQW^-^>Atqnp)B(W=9NS14Eu59T7r@{`MN5BE|oSrlnKbEk;ueB)x7vXG-ZMdqBZU?
zmeoexYyu8I%rmcu5ZV)9L0IWw>=U7irIPyyE1BGufA*wj<A|MosD<5@6FOW=^(`SY
z7%i9)od^bQowt!-GtAaIXM>YoYH7uWiY6O~oPa~#IAv|L@~sh@9`hgG0rb0fEIIU2
zkApHeHE8mAh#}-qk#sq&j0I^&54T_^y6Q-Co(UfvjBNrU+DHH#WgB{mMFkLoIvexO
z+G3}WlRsE!9K~Bz^pUPb5v2EXz#}=nxf*=-=ws2<(M0~1G4X46gw%G$Hjdv|R|RbA
zgNC|#)^lJbm@b!9SZ8+m4(kTPLwJmunMHhthW6v{UhtQ?Ocwf%P>RPfp)}c$6S5SQ
zEoVx!%C(9HUs_^(<7cQgMFJ#Zv4MxW5%Pl}5CZV!S82yDQ)y@909-{q*d0B&v)O6_
zdADd2P~HvQ5OH=V#Y*_>RK7KMUtSu~vdn94dab;Rf5{c~bCSUns$WQ*XSHtDMY6pi
zt)3h*3i){O8LXq)13I(?EcbTlT5w_3OdX!sd|UpKRIU2!P}hzx)>|0$?t*^vz$HyL
zh){@I<sE&u3<K^%!RhO%F$n}n<8IL1n~CGHehngrKtKM*I0yUG3W%V|F|tF1khK5H
zYZ{)m-5OEEdsOpAZ8i|Ld6x1$q6%X?I7}Hm_dWt|yMOdDC|t@vR~3wUCze%1gUY^3
zypZf*vwwv+Dn3M7J~J|?1$v>xxg>O?zv3oZ%g^P#iw)B>#e(I?0ps_&Lh3YNjqKd>
zMAwD*(ZTbJ<NcVRT1B1n#)DfN-s7ve;Kb@u(2Fq=4$kj_3|7`U4kVNtadoEkT>QV=
zat;}An#KfsHcgY#ofGzf39Wg+FAUSqUz_JX@G=_~O<Me>F~+Ej&0eNG@4z;*OCCsW
zN-@pczt#q?qQV%f;KQSRFH_@a?n=U{;5&VxT6PC1d-AY~fQQ^%=cB-Hd;7N;S^=Mp
zhsF_SV7!+Sh0z@Paus)QK2?Y&BnD-)!OwisW)0#m0CYnw%+ZIACzF*ju86)ake_(7
z#ZR0g_Ln;`SEqh!3ZbT-QYO8fI&9GH?#$-)zSF~lcv!qu;FCfnE_pm&CqN7wJWHVL
z9L$r6c>!7(rI!xTE=I#3)V6{ab}4s2x*s12=Phgf6Kc8h>Dl4ES1GQ1824`nS)kU$
zol$)H18YfajmI^vg`g;#_l%_yl!qu6ay#dzb-Lf0Dx^zI8hBRjX_!%K`P)(5MBP}P
z7@momBxno#1Pez2oq~ij!%{iq0Qr}cLjHQ^hX%PYphhUWM*j*Ryp2_Ui%C@{Wdf5A
z&yJ~&Yg;tdrnmRMKF1?oMI|NE>&x%A=)1W(xJK^Um1S$U7RBAhe{3K(<ouDO<XG<q
zY@H4><bs46M@c$Jy-)0ObGF}ZSKeWLwA7n?iI{xV%*c-Dxkmm_C$O&;K~hyNb5y&J
zeIgvi$EnE^HTc0mCAot;eIfdw3(MK<(Qz?Cg<?|stiSHoFq3i2cXl?UXl|YT{qUwy
zsO~f2`76EF!c>#gAL32b#cu#${IKfGQ<O!;aA57yO_Ai?q{bR2pG-5Xw3jiUeei^s
z8hYXJl-*JxK9I@IdKEPBeb)16J)G(4eZojzgMW%$%2R8bEP|5iM?aDHfQhu?&P)}F
zzQ+~VMK8eAT<Wj0TrS=QYP3bQ-Io;Vla!;ST^o$9NP|PkBX_^FIoV`%RCv3oW8XX_
z%d`adp~%^;{67Ahn7?J>2&4pP37^`2!ZC1SoB2X-p12R6`)W2<2r0n!(`Goqn7x;q
z=kV0p2)G{{h`HEgrMvZ~xZ8md&P00SGKT*p2}iWXn+41LLBq@ajZ;qZH+CL2EA&sU
z>k;XfS2Bgsr5Ck33+jpkgX#m<VG8E=;_q(?BV&%c!gDaUA5&2{M+m@uU#+L&8oTkM
zS?;mH)cMQUP`WC>FX0Ygu&{CJa86Z{6ma6SJq^ZP-0>4c5Gl*xLcnq;38MgAh+gcU
zM^EWegDnO*x9wGOo%VBWk1YG+CO@B<;W5>&EO|WJu%SKBg81@!PoZ0ShVvH^P~~vO
zSmPeA(-Iwpcavd9#84WE)z9*hTOV=txTV9*QA!%{#P~@ucF@nfl5)RgEx;u>nC@NU
z53o^Il{Md;FP(!X47Z^JqKM9J?Hf!SRAkmu`=$AUZpXP2?Nv3Bsr6#BprLfiOkSgv
z>VH$==U<>ILfEQIUQdyO9nY;R5F@=~rpZ|9Ev~l7({kGL$EuARMtMwiSz}-8Zo-Ym
zPyf0&NDr#%COj3)av+TM_`Y5#7JscyK0y)*n<>z?{N)6w2Un3M$~UrtynyT`0VEe%
z$Tr&}PA#Iqg$O!jfW)fNn3@_ZKP;HCzbf++?1YoDQj53j9MKZK_^p+b>Hc7FM`}X%
zV)`^b{5=_azEvB*9?o4-biNa9mybg2rT~EsZ2^%+@3-=WwMBt7)xHDrXQs^4(|-|u
zY)lDi{XLsxe#Zoc2VZW6$Yp>EPBv7vDh(j0oJ31#UL`WcTw>NIQvh|3^dW4HJNi||
ztDYE@;HCTYwJ#{oLXHM)O1hX-VRt3s&`iRP0T$|OVWa$UO-$5JUwy#<1C{-vqjq;7
zU~>q;@;-0X&ROh&WLcoP5Ss$WLA5^Z<9@JKTYNdQz{4txYPkP6ZBpftcjcCfkV_$>
zm#Xe(;i=m81LQhOw)!ysHNt$Tq1E(|s_bs?fJB_?jdh+Lqe7X0;6ZdD3Vza?#Jn6*
zX7)I+j<vmNrvfcD!dx!U6*hO>D6%4lA5^fn)Xr(W*3XY@$_s`6B5!agE1nHOy#T=0
ztTqaOuVJKg-_$8hg1aofF3ML}79-MuKN5X3m=Bz{dE#@=?D?O1?KGDk&lQyEJB?ro
znp|z(J`CnZl<Vl!zC#HH3!T)sUNuobc&fnr?IDKdbU`p4o2WH-#|cdYwqoqsiCRZd
zg}U6$O~t(xb_)Il8a5GBg<;Np*Z=Qkb?FfW+T?}0%!dq-03^|*s!S6?>Ne&ZWc9?J
zNXzh;3w&xMZ{7yU?ltS%XY0sg6pgh8aQ9`hEM1W6zcVwDLQrSe{DIdM6;3EI0GqP-
zY%laTQa|%ix1nChV-|mE=<~7fa1vCcu9cC`n}hmth3qG_8hx`u&ODeU4u&~?Xj(KN
zbr0E4^)HQD_CX%bM3Q!_J6!Bd)HqikB%H+tEgwSI;t>UO%RD>tKG=fG>?jkwja}sn
zf|Kc-5Rie<@uFSy-rJ9le$C9+;eLf_BV-zDy_i=uZL@5#6nAf7D%(*LkFdZ->ITG8
zm)sd+=A;-x+PslUo1Oi)n~^b-Y3m_gCrE3c8`JJ*JxC{|sdzSCohgc~O|JVg(&2jy
zzxJC&>e$p<H@S2NBOcm6o=hb6S?-i)8e&LfqgREAsX3sza_ve1(vK74yTQgnt?@J6
z=NIwD2e<Ui>h$GZVsZwv;8#K%``(b5X-O1&xpR*rXLL(7Ym_T1I@GTzrM5p_En9dm
zD|cx?It2wl&1^V36<!t$OUMT3sF+K9%<d;>N~FHp1mj~<)f2`hdGe`@q%PE5#hv%>
z&(CW;&0to}w1sf~iG@@D&wLGdVp!wgbqh61Ris5vJq#I~Z;zY<XbV%KPu|tvi+x<h
zi&e@^y8j%{NW&jn(BtH}Rb|vwjqIKRp|C^eq}f~wdpAMIHzo5?3HPdk!H4=p9AFHL
zUST-?e@cxbBh27!OSokQN}A1Kqk660WZcOyfKwn~9@equQDlqVU=M#UbnUHy(Rmkz
zH)i4c{#tC|f5wI<cya0EU9|ZDG`IZeh{h!UMA9l4K5jhbktlN#j)HbQg5XqbGPk~f
z<zOZkXhbHs_T>-e+!Wtw0)<j%Z`SE}@d~O2I6(bq^ape$1u6Vm_fL9;v{Zhbv+&?_
z(Ru!Cc<Mv=%P5B_wK2(1CThn!xz77V?>N=D6dDSYOEHg`04w^s`SKuVOmpN^MKb_F
z{oehufxwi)wASxg5QJ$)$TV7eSHk#DqsS#Vbnhfg?nLa8-e!^{H>uSl;HtfNGTNZd
zw%63BLGAfb9;bCsFa9PFbB{WM&Y<0JwB?yHbki}0zRVM~t|y?B-7}aVz?KEzHl}Uc
zwr$(CZQGu<ZQHhO+qS!R-bNd|WW$OpsH(5-`A-r7H|rptQ-ZY>+^%_R){fI5Y5OPl
z;oeSxQk|>o{gLx2jB?c*d90NyD57Y&RF?xP3cW>j#gAmfBd{c*197AU5?9{#Xt$Q+
zZng#q%F>lyzlyaVg1nam>B2DK!ZFjyqv$ie6XpFRzW=$ScYVXye#vBK{b*@NB;^gc
zktW1<aSTl*x=R27_YAsG(R1DUT9&5HLGBg$5fQ&~>PiL56>+GX(@W917nyLg@+NOz
zM!-rDYK@6WC+`j?V!x)K;~ZjOu?`cLev(VH7)^h9V(BjYW83v#Yu9eDKNpL`g_&NW
z+u9C*a58x69OswW>$at-L8u@rZ+|mpnm;I>5rmaWt{e<s8ccVr@(5mfK-*(tEsaqu
zZdo>g<V1joYMS7`0^V|4A_Q_(nN>G{B#Uu0*%`|kv`EQ~8R^TqM8%1ivV`ZJP*Sii
z*`738paY(Quc_0t3m-1in@ln7DmNz@UH)+O&+8$Xap6{R-XIyhOU7O?5eZS$hC71J
z;jQhC1iP*idJN~oYL3acIva8SPB~PLU{TM(45uCI%cg>b#&SKHg3?X>a1sgrF6YFd
zx)?LdGA7d1(wePDtMct_x|AQT4&(E^U~toOU>LG@?MilRD2e-l@GzVp6vgvX4fZLE
zsXWYmY#wxdljR$n@UuV6=^`k0dI*B(YTpz(Z|+lgIyNA(>CI~tvK^2;k>qc*^WK6%
z4N0p;L)+by+(U_;gU(G^oAlgAVu_eITEkKM6`=L`>4$@D`pV`1SM{6uzpCGi|4031
z;9zI^@9H-_EBpUL{dO+WX{E)+!l_JD{txzDYRHANDLUG+jPq|nhQsy#;jwFX>ofAd
z>bKZ(+iKTI7imrNN*U%MDUqUbJ`!Uy8(8G#X8Rn&Jd-2vQW6VtjJ0*NG|jcid69Os
zHGto=wY{-1p)5ZUnv@rp0udTM8$Yo5A9&CVfV6Z#X>Dz7dt(3uFK{_BDJLj2fMY#a
zN=(d+--fS4Ao`atxFfrw#^QDy|F7F1thte$epK_*htz{#PI*v>fB&xlfTk&|{QlvA
z1u02gApVMix-j^FO~9PrHiGlZYa<IF_*ND`O-^8nfKyl50Ia{d0Mb|2(loyo)9F8S
zsek{mwtH6q%=KSbU{CaboL|86K-m7WEh#fIez|}tGXPFxL|m!8O!oEQ>Ko~6Z5cmG
zeR`eg-@Shv?HhhC4L;F6rAo|iuMVsajL-Lg-ZE7bl)rXeI@HnIzY;gsALyX#*gh{U
zEbR_H<fMKF{^{K!I?>eES%9?saH)Qd7{GufDKoIFqB(v}za&of4PR;)H8Ux$C%-Cy
zG<}ffwnxU6z|O8eQje*hx#wSMx&FV!UPR#FShN4o@8AER-*wIoU>j-kgAD&czYiYR
zTHA|pum}0HL?&jYaQlYd>^jRU2mSPI`Ahv^!XI*Zgq~X=h?Kmzu;lzH=AdH*c6N^K
zBms_pvnQT@od|#7#Q%xkcfa9Xzjg;dZZSWpLSMh0Ilo-ZZIv-GUGX$`dIJ8xZAJhM
z{(iI=Qi;FQ+hZ#WBagqxce^|+{=YSf80I&YMn_M(`+kd1>*>EL!Ld;}*So4xW)4zt
zdF2JgQ4yehQ)4H$RX-+bjZC1N>*^bTzm^}h4nSvV=or6PC6)#jhBmLWck4eefY;K$
zUhjS;A8Pl|8G?+Gk(kWCpWPpHe8asTon@Y!0D!-Kh%g*pi@(}mgM`%9w}5vAMhE&J
zbM^EM-#o5<aJXnxiL<{1hrQuxKUOqmNB{N+aL%B}%-9&{>-#5e*01{M2P-BmJEHrX
zk4toTcx>kBmh@9@kgqPHGI3GQ=GW!sm-QR(($B^Xe-79jsI|k`ZUD(TIVJr@F)W#j
znV@2Ebr6IX`9OZ7OjCJJ&uI5L6}4wj#o|pDOK_KY_<Q}L-Tmaw0bj<4eIdv~Xg>Q-
z*TnCQkl?bGe}lx%jDQpRSe{;Fnr{ulwM#dCCrtkIOqp*RSAJ2I-UDn1-}1}>h|{<5
zvD|^`UYzX5ng5a1Ln8rf;h`21gr&~@I)}%?_#)A8-`U-q{ws{}Eg-IpEH*Vp972lS
z1cpLGRB8UAPvJP(Q6*N`A@Y|Mnrw#1o0oF(RQe3uRUcNeO%EzV#;7l3G2a16d3HR_
zX$0nl<1Jb>sb<@f6{9$Gipu-vVG8#-EnnqOC^ILynK;|+ZNLx0OT${hxrXvv*d50n
zmmsJF0WD{SIyg2$FP|86;8JXv2;kj&23)qhq(u1M*{i1PtL?y-&iodigm$)roz&m&
z{en!YL<%7iNW7DC*D9!9?nrlSgg)0$n^$3aye*a`*f4vHl{gLlDmzNqwCirg?u~)p
zE&Y#63LH`_U#MlYEYqh&UYq0J5oZl`@Z@Y8h)Y}Mrf@bH5!jN~cHX;|<BG$29jP3A
zdE&T(OflL#T1;TiyGgzJbi|_iY~1NY*vT`=;$bi$o;S)E98Pa%C4X}4N>&%!ZFDwH
zc?U!dLRX~Hndel?v;17|XGd8O%s?WS<0sC$?!u4wR85}>@(J3_hq*c=69%3w-LsT>
zNtpT0Pb}i(f}K3T`g7CVE`n;0Ts>d+MX0QdJpSh_z24pPOIVA`M5NQoTI%Law&y3Y
z3UZc6^}k?VY_MoY0Ot}qX*oKzXw+}gig>|JC-2F?K<PU0X<`#84SPp=B3N@LUl^m!
zoY1@T19CK8p_S>$8ElFAWa0_}boIeuWJ|tuP(?Xbl5A3YN*_L>CVzr31qP!sUvM+K
zS1j4V?<|rPkfkB<oB4vvL#}<UmJPFREN4vsTe)JlZ|qz{Gqk1z9`!yYVfSnSV}%*F
zpbmPIds@HJ2~SHY6?8=`tPr$GJ#BqCnR&%FhIf`wc`kHT@tHKoOcDYkMO1DNiuuK~
z3K4vCiw>#vdz?_J%P6&|8uwirhFI(_37@#}f6}p;0pr0nO(`V({}l?VSiGKz?GE1M
z{QH+dOjUcw(CxW68dBuk<ek7V`qy%NEu%i_8ThA78`pjZIrpf`oWSLX3k+jLpA7Y#
z6thIs72&3-^a>*4k!TiBA}&DzW7X;wlTW7)#e+@y5IfHYp(%<3z~6d$MY(~j;dHqz
z$KWb1#|8)A<?gP=nn)F$n|JT7WwO}2EimttqKGN@+|^Zs7u9WA(^bY6dByCAW6>KD
z*2UT>A@F+I@7w1h+?=5^7dwMeg{m@L7Fu(7F)GR8IfqywwXXL(!Z@nMDBcFMWG{26
z3fOgJGtx5roZ+lM{pKR%&NUG&l##@y)wo&k%Lp3#rJ{Pgt|H?>-oV<^04kCWtVbbt
z#6h--@6qBX>~B!Xv`o0fodB%{n9RKnHBwMKdL~l;rHgkzP2*3lE1l36*k#F(D(w=U
z8kP5E)XwOW@$T)_`)LSU&x9dcHFLxo)i&Is-vpj6V}IT7#0iUagb(om?2m6=*inAC
z-+G}pb0C<StAyPmuvGgGubt=v;a<m66~~wVs%YMsI5^SRA7pI+tCflKe};?wbfZ{?
z&o{xel0<5{Flf2vscS(#&kq@QO#v$RHtX_*xEr1sB#uo0sLG<1?Mlv;K8D(Lw|UX_
zjBKhtj=uDl65u?j)kD&ujz?-ij*#P%%oLu?=os=tj$osct#KeHCw?!nz&o)SQdk>W
zESp~tm~XobqLbnBATPt?vjQ2KYF`x?vkeZ>uXgHOm@|DX%FNZQTHM(cWYUB5Uz|Xi
zeSO>s<r{X}rwe;HzoV|3#t%`hTvIV^HsQ1^;tD(_kJB!Dw32m%Q{mJ1FH!XbFW%an
zwkAC}?KXN0&W?EM5nJh*c-%QOGrejhjZGe!(OD*FQJpaQsRNnrdspU%4j*CxSlxuK
z*McS2qDGMzQn**i3<aLRe;4k9_HNRg!;bzhr@r^>h-&8uV5zk_eJ!M_AOFhdW#UgY
z2n4{;G9m*#OKuLKj`6)w=q7w7SqW{@ZNqfMqLyWN==xFgW$qZ}LP)|w&Ug%<^T&BK
z9-<~a=UlhP10Hf~v{pxRu;O*kOf##nwAFFM;An&k@{m9ya;Hss@bOkny+*!v_-i7U
zVEpwu8z?RlA}}hQ4n<#Wa>3R*q-c^mlBP95n0VZNQO%K}^Hx@R{j_pD6d!Ax{HuML
zFQO1Tx`tJ<C~Nx8R7B8|b>U2@NSH-UsLobpQ(7*N$$iehsyRmR^jf#)GaC)tGH576
z3tVeCO}T{=FVpv;l6J~lYwO-K_*B7}9M_BtTk0gM5$dIx)vi^hLljvxKtpqC^m`A)
zDd&TYYjPigr?vXkrWUzuBzz%cu@GP%0f}=i7Wqo?*9RzwTXrWLLPHI*&$5}EEj5b)
zX58R|GCC%_ckF~h9n%|!3T)~rR@+h>SV=mX5N{wKXNv%ut8rW8pD^JIm;Pde9f(vl
zO&ywwA+fq%3u^D6L!Xy&GR}Q?{u=u^-!y^$*pabkunZ4DKK($cH0Sf>b+(s57RlSw
zi`sEz1RvW^Ca}*{xA1()h;i}4gV4FKZp}eb3@0|nJNcUhhc=h6{JzC=k!cjYp%Y9<
z?xuFm1iVcOvi%FiXM-Zya#<nXHWN&(O~a{@v(5JDJ2@3BU_a+WwH2s1%o-nuvO-1~
z3EukXqG0Jxm1a#k4E$n=RD(KHLZx)u+d7!erx?r6ZeZ>?#PjUO=A-y^&L+Ebd#&ni
z$WEl{MIdIN@<h7cbPUWFmS?A9bD0#-jGS$x>9#iG#oR7KbkbfARjyaWb}szGz`T+)
z+R3TF;u+P@hZ21(A)bhP!*L1;Gl?qmxofA>WL#t($<1>|hlGF~$G4aA?bgyZFk`zQ
z$glbWtP1Hr_}Qf}3@>&Xb_pJT>tSWrG(%cUZSKG~?_AEE&hL)l3a?DCEqZ)enc3S9
z*zE4~jC>_?-oTVbC^@JqZU3g+X`Li}(O>2_Ou(Da{>l0a7~{cwk@ST@H?7lmg?3id
z^W08t63a08h5L~>$CCkrMgrb@3e`D+>-pB5%NB?rt^fC<&75NmX!O$Pvm6W|1X5d7
z61A)k*@RQLkoZd$pZCO3PaVKu_zjq22G0iGFPo@<hi`3HKz(d1+P`E9x`7|jFS2bV
zMJKcgX<p<>Cn4@7WXq?rAuV+f<SagEI5_>hN)=rTS<3}FgW_Z<X?6_X{R%A;xgAT$
zoNnq~f$@Y`N%i@U5DKz``pa3E5{)`AVBkufeb(lXO{n-!2!k*bq+B`A4ok*@w57Vs
zCoh!;Z>yoScC^d#`u=kM-VhS7i(8q4<o#H*OG-22Mw@_V_Dw^5FQ4)-jTqIa1L*On
zjF_Ojl=??Ir|=g|Rue&ZA#WgwubJ!ymm+nzf8kv^T^6d`2fKk34Oq(cDBn~1SMZuT
z)K!(_;HiRq(?7*5*~@A_0#ddwYZkByX@C%edn-ZZL#Q7E@%b^tii7N&G}`SH0_92p
z>K4{2T6CcKj)7K8BdS7>T-^xSBZIc>YW9_moY#}UudEXtat+!;Su=XLN$43Zi~XP5
zIsZz-RH?>k65k%Lk%Ft3e-*44R7$E`Z1#!}enXE<<;W;VADR{vY<dnYpNCl)eJy@^
z(u4CQ!1KWXKl^u`J4|{b-Ff0<$+Q2(i&qA&k)0#i;WAx|_j^RKEr{AU4@)QEPqF16
za+beWy@vUM%msNG;SB9@+Dkt5Wqlc`yON8TquM%yqhaQ0IiJy7PJYh5K?Z#v^&7Og
zWhoyNttrM#!IGh!c%4sw7ahcP&1{fAL1qtAA;Iv&F|sAxzv45}-?<k1s&7Ft)lPPb
zx1uGq%~;3HQoXdYqVR^|;~i7j9X(CV=aNaZR3Lp<=Rm#>Qn&Oq9wZG3Znty>O)tE-
z1aw$lMuq0h$AJ{Yf`ZWL7?@#{M>XTN%c?jKUfoAq;Dw{>f1Y8RVg!MWrdJ2BGm0g1
z7Ed#4{{w7lGKBV9b+qeSE!Z2|ic2W(8Hm{dwPvsDUJ$PXpQC_>HGMHR6(t!Xq9$~3
z*;AKdq*kT$u(+{)OaX(}wZ+x4*D}g2Vb`hj{h1c|=rDQaROOIEp>ywIj`%wYrMSf`
zs?{A#c&~I~2_FH&cAz_AI&O`RvT5OzEUjvFok1i4z^mipkIR+Yt=@;GiVYr*qKVO6
zF~`D?tOBQC?oX}p@qrD}r_@SzA4m(}nBYNWMQ#7pZs+l9Rh;6Ox@<53BvhBhK%7|a
zAtgd1<<DouL-jABhPDDGV6kyFC0c_T%%ecHpx7&Yz!i{^3R0Hez*L8+t65{3-u~-#
zHINhS`xHWF*I`1VOfj{ntl|myg`Vj&O)fSNKxaC=xk$&TNuMe}Pv?e{VOTg|c$1jb
z86FwQY?0ZLRaC^Xu>p)@@B_-bGh^jw@4<}v<UrzB$O2O>e^!P}b9)yu9&c5|{A~`U
z+_(;TG;pjkl9`HhBPn{rKv1AWu{W&@b%nJ)>~(!_I98}-JpZz^@_BfRC(jD7s)=f)
z#Tbv(IoN_7=i`-%Av)oH3df>o?1e~`<AAn)i7W8#9CzUEg^Kl_8ioCu5KxVW@mfAs
zVMgtcG(pw&?o-C~=yDTagd|XK#0|RpsMq5I8-f43BG7ZWsq@q}4OF7UTfKZV646$B
zx%FxyF8JB0*GT{Q>ft4gMz*d=X0^yx+uu~J8oM?299n5ioHg<n?7@w+A51HfP2J8A
ziDlm@b^l}|_4x3s{@_a4h=tK9^hAHSAEoYSFOY>a`VP|YSy9keB7XN`R9OqQ<%)=Z
zIK^zTZ^Q=Xk9Fq6Wd3g0^>W<A*B?(Y@;l#Y!1jMVK0KHsv^NIUQy(tz`q!cAuZ5cZ
zgUQ49q?s_3eB^6YWI{L-x#AH(N8pef{tBpQ=U3Q#d)c(W=wu8;=a|u`!<Y<ORifWa
z05~?W>NH9ECmxQn@#uhJeoqm&8drC-ZiltLj%o%CCGC;z+S!GdO(p~J8NiKN7L+LQ
z83=}D-RB@I@uPxdT~c&ewX$@&$Tj^C(6DP|tl0pl$t(mO$x_1n=l9VAjlVy}bJzj#
zqUL3=Y7+aJEJR_{f{ckY=6ZW(lXSinnXkjgqk0`oFchxLwD&E!+yKo6yJPdnwybfw
z)?&^-*$~fyMp=eRDEW&|u|x(sRPIiVxr7;IS72L0q5i^i<7DFKUZ~>KEUr4b){^U`
z@eIqJpYtV^o#RZfUO)=-2v2<g8shWB@<r%pO>1XQb?MdG^0EqeDr*J&XJY5+(v?*t
z4^Ow!wwU&z|9HLw!>NPq$1EqqRr4==_n^t|+?%W9q*t8B^H*MhXd(AP@;`BQj)9Zf
zTy(FB*X}UB9<h`<>gLUU0E#O|9lR6Qs$MO@t@i^m>!;qAEBDTBJU7nS*l^gstN@E`
z%CA7+*suogZEEubHSSP)BuU~GhPqyyrDW7_M6i;@RK10d@9#}R$til6e{lKKe!tFu
ze?D%aygQKzd@l0(0mZzYkX(+(i#ba-jn{i>C(YV=&7S~I-w4KPS9&0M@;U&e*^U68
ze683VZ)R`Q;5{w|(F>v(=^}~%F~#)YaR<?ezcDE##iZY?m3CYQtaTj)xNRO$QY(;r
z^4&pk`x$8QEV@sRsDNY+a)j-OgV7STB1jmxxEc@ApA!`4ZMISdyZmxK(%^-uXpB6$
zAI&+Jwz)f5nZ+f}XJ?sA$|_s)d_Ao}{;XuMqL`Iy(n_kPz+m*Ezw$HT=6=8!Emu5u
zWM$$J6A8wd)!<a@ZI@URPv5_T-+gB*-hb**!M^jp%709XQ%RN=nSs1*&|Yu5J}L8Y
zNx#%verPEkH6JC7HxjafD3^)xIsNOBbO>cnv@RA^`prr>6`IRarBwJWxRlXX1L}Q7
z8O%X$O84(d_E7k1JrO={mi#+iihA6dMaoaSaO<)m>D&CLcGB|!<n#usKku+u{kF_{
zThP6u*;~`0AgHUx8Rw;Nf-n905%((bT!rG(iy?ipeu?gkJd0^u5A@PSR{HNq7#Tct
z{AwKEDa=^!PO<G(V9h3xVv@-b%7sA4c{e%jtl7eMniyZtmip-0`e+q_2MIQ%QsDSa
zRNq;bAU?i#(RNSBKj%`Ex-r)hbY4|gX(iagnsza!ikM_GN?JT(aF(3;hZSO6Gejz|
zUGCtgjp#<_E<4hbiCP8YO++{6np@2yGJ4Ay-V|ceo~7ch%Lq2N0~H{cSOI-$gZK%O
zp!mex;@bu$kQi0AS3paj{K2}dja<VosehJNRkD1?>j2Tkiu>92CN19_U=q(43R<Zi
zyHRnnOLQz1Lih_BCD%RBxa8Hg$#%x_=Sb#aRP>b^htidfZoq)g6Dj-3xg|h>sNyn2
zY@ZV;MXwTr;Rg~3Zdw8up<O>rlxE~2`I$1pTd#`;c5h+v5>IO1Vu1dda|n+Ra-}mH
zuBaJ}ih{`pT?c$N_|PlZE9LZCh>Qzs4Lzv}ZI<#A%51x|&Svce?MQP7X2NBK7D>gX
z(3f0+p+G9fb+NJ$3iyW`%ECVpC4zUjZ%A&fI87snc1fDoIuc6Iif;CYvK9vK;SV6l
zo3-pDmSzs2!P*Z%TkFEM`88=mBVioTCt+y)ySqHSqa%?1^mFhZQCjx_O4;M4ajDA#
zlILWu2{yrpIxk|76t*MA0i<3=ZQz#(^DSzb5w_0fY4ka&;w+8mjA<Py<>G9$zcVPF
z99)_aQ}YHyF}KM3T{$}Vhhm?Hu7$T9UCC~=s2$dwfgL??0@9FpXwufZ^}o@MhcX}N
z<Izw&>0Lo)sL6XM`0d!t1xUr#Bj@soKlw`E^EjlQ;NR*DttiI@aQgPG4;hZ&WTJxJ
zsktlhFI9~sBEV_oCneE*fMmTJ1i#^hb}5A1!O(&u&F^8}$$tS3qkHpRwlNR&;IiM5
zGD^n@Nu=_iJ74luDARp-`%1{Z8PPr`;P*s^=a3`hOAURJcEpiJy>^_PSoJo)afrx5
zNx0T8&Z;w@M2Oi*ySKEAq$x79;~XLV&?B~c-0diDu)jbE-w=W#>I-VIbUD9NJ+Sdc
zsxsNsAuUwKIXfNJc?BzUUm#eA$@Q3*4sz{Ft3#!a=6YC=WmGJb=A29n=Bf~ZQwzoh
z<qspSD84ozY=B=3^Au2H1Ef!i;zLoXz;L^)2+v`}SS|6HV`oi1GKN+Y@oDrpl`gp0
z{k?n&<ZUC>TMML6rUf(;?1j(<i<LufGQw))YTeKWKx~r5hM1LLkHjUtCz1*0_~>1Q
z-W9{o<wdBM{8F^kYWcX40~)aWtvbMzAn|!a{-vZK`?<nzN4Y8sg)}2DI*60@T;^=$
z&AQ*u`v`Hl&jE<4TGTWD0$*J4FNHHbj(gV)QrCl(<i+CGIpH2HeydC(;r3~czgN~f
zH5m;cYbdafdTQ)%s|%G!{R{-o?mFLI$`o8*hs27Lq%f*$4c92#ISF+zy&|Mgc|ldX
zIn;xsvY&1C8$NXsbEDoz1x%;6)uGjBNtvBO2EQLUw1(>S7&oYD!jmanGy&K9iK9gk
zD^Ce6eit~a-jfBUF~n(QZH&Zw_M`XVlmahJm(Y~Ed;{>AA=!1S33@Oq{S3yP+_NIw
zJSIb}%8}qTxLaVatbY?+9EB!e0Y5ze_J<PL2wKAKtT42c&B_pFw>}26zzqo8$9Yz=
z`_h=3R1OZ;e!X`!i(CR^N2B@IeT;Yvd`ykwx=752l&oNhMB7fKgW1JnyB*mlKL!}+
z!xSeTHmvG2+Fmbk;>beg+a2B^Vv7O!k<31gV+Jtt#<u>k))N%H;wTRBN&C`WN<}Z}
z*+^8hXF^*E=<EKd1uRK(iF#;Iv{S`Zf}w_nc^0W5<UW1-aD_zK>aX#xp6wi7wNiT@
zS4ddnymd6JTMEr4aCuK{=zgz=Gk@#}i19znlt<P+jG4#bI~W562j>>f+xOpsKM^aF
zzG{6pgw`X7rog4jAvGc3`>KmAHH^3iLZjJ0&KZ8VpI@5CWWx~?vl}$%I*+*kWJhPg
zm5g?bhCyvMr<gQ$<Bj?jBEaL`myrY+NNtdR7Tk>wt5}LH7pdqE4U05})PEFgX%%rm
z7hGo%_$kJKE9{S=OYryw6P*p{`fUvhNp?O+RK0%M4=CkY_wQ(lM7vQzfW|@jvlphV
z{?@7-5+(%5D2by2Mmn;q(!X5kZGtF@@h8tU1WyZbbHFRrA&4&%JCb~G;a>%ugpL6n
zwZ=;qPW>yh)Bbg%5s;-;o0DJ(2rN?uU3XJ?@D>D9Ex2`I4fO4R9%FsQjDzO@acTIm
zkL85U<KUV0#c3ps_Pxd6dLQTRNVLxf5n{3F6<QZ9F!x*Qb`8R(r?hlz4o%e7begB;
zF}+jAPuBW@d`Jj1DW*R=W{x@UI=M6lc|D&Zi`T<QwG4^ps7pVg@3J=r801`K&fPWE
z7wHx)EoSJK@{hb)s)Qwuh*z}YEXE!188#XGZLfQ~C)GSyfz1%mtIZQ1BiXjQ{Zb@2
zp<T5oJ`S0nLtx>iDfED*+QM1sQL2M$bf~%<Jd|aL2EdzrRKUy0F$2L6rbn234x)Vi
zPBrnKkPez!As6*rKTIthxYIcW>RIJ2&(T6yfW1&_wfc|8Zi5cHjl#L2(+F!#7}lR6
z8sEGnDB{_K0+xNU)EsbTj*m{b1Dxg@Xcr2QKWI99)N`svMZriWHr;{TJTDE7<s#{E
zdq}wUWhE!fGG3O~v{GP`3BCz&NocmG=2tW#;9-;f^-{x#OqiKov)3s+_vJgh+Fg4j
zLwwMDoOcM^$`ssX)pc5WwZtIf1C*O0_$`OkElb{;T7suar$Zd0+r$+hxop+Xa>ww=
zz@|57?rEAx6<X$Wi97nC=K>LpiJ2o)J6}UmGj0y&A3x(64v>~V9H8K~nWIaekppXj
z6+@V6qdVmcT9l~0N;Aq;^$S2$9X=BDu@&R;tu3KS+~lGs{aTbAC$6ZM(RK>zfrdg{
zt)h>E2TJmor22EMa*m7}BAZ%ueAZ%~uq!XvR}HB`T`<w@f+}C?W7(B)5RbwdLae?<
z-~F7&MrC6~#Pc?s2s@X7IXsA}wQjxPXS4W;Ih}zE!!FWrFTQWv<VhPcmu%hO?}dpb
z1RRohyd+&azIxvCAV*0*M|Kz`X5-nK734AR+>Hg<wHuTw$AUw!Xd+OJ5iy;eq)ATf
zOBP58<uQcbwcH0Zk3C(_QLi}rMmdQ?lF;g?ZHfuxF5o8Sk2cNff<xfTa57vZCg<Id
z=?d5niZ!L)FE{WE&x}Jjv}hnR(zr&!DqO^!$63W1fW47Kgg5|gMwy2a&8Jv!_<1A7
zSqR7nY2y>j0Du&h-5j@&?^I8n*%J--+oauw0ZeccQ|_b7Xh2)DN)`(eOTL>*GZg}M
z!S+<{3#7nE!@0qwUG(5n4pn!+mHUEqwEMgDh-iUKpHWqAAWeL2C_Yc{z1Y;MgXAm6
z1s1jog{QkzdYHUWK=JMM8~qD26};lk3Zp@yDYQw8_B-a|>6v@*cp59x*lIeeTvxQJ
z|H7fX#@j({&_Jt9)_*#Ox$`@A1asYOD5P_lbBw=SJE+h!u-UPrHNp+yxwLeh8v1*I
z1lE>(Ki;+MZ3zmP4Qc+~rM#5f5Dy?pAbR7Ll^_sSU4i!CU0lDBWrP^rhr%0g$qgKx
zj6s_*Lg(N(NyQoIfQ>2W*tm#Ual(!&Yg^*hLX>nR5FI~CTvwbE<oNwcOndMnlC~~p
zQbXHxUXdR?AvdZFx@lMR=E|X9JX7&GM9j-+=A?i*`-;MP^(;L6wdz2COBy%YpDYx~
zD3Pr55RvQRkYUfjFhGf&rpu?6?jatOB>}t!#%XAgUrQv`oXteYMuql*qT^Lq=!z+h
z&(Qqg&#SvOX(Ok}byECPq_^BC3uG&^`>sCEnEJqw_i=$<-y9#`wh2V-(m+N>-f0x4
zC$-*O3g}S5aOm?B{K%J@l|R_1LjIQku^@rcc!I9|+u;l~N&M2s7v3_C{LwF#H$*E*
zRe->+$I7Mzz*=OVQ`demN)JNW#1_M;txf`GlC!<q;P%9sj&Q?*;nG4w_G>^$@h*LO
zz4H;q55<b4B?_-q$H^V#tnA5PdF=HN%ltNfXRa(KB$IO3sfrBj?!-{#m>p|X#zt4l
zxdL}ZC?uR}>Z6#el@^W`!)k9DZcEQ&7nYuLUT;P%#`TZ>6G5<7LMV1PCAyd;)ygL0
zoH4}of!mPIC1nBP^k|yH9Yw!dKCHuigs-sVPf6&XG%=Mq2NkmVP=$OIzNPZey{^o~
zp(ue;{<Qa<sz7A4It-15?&(J~Wd1{sLK#HM*~wRuLLfN7rbdQxjjxq|DzINH_P|yl
zTvOYM3f~G!Ryyme0flXmtr-z%c=wr{$3&|DbraVELZB<+T|EGm6pfo1-}xgio0i`$
zgGKiEQS^7~MTL^rC}GS{hKdizbm>mO&Q|~UDJO>5EWn+=N2N9M5ijpc>?kp|3UB7!
zByb$|95L5X+D*7pIWXS9c99_9;kKN6*Vjqfo45wne!G3a00AAr*u8#k24c-9VysRI
zcer}eri|w3^NEG~WtZb))%{9pF%F?Q<|4}eXyqbEeUKiai)%ypXx5XbWshn6vkCgS
z7+VGK_Go;<Q4V$az3THguG8L0HC<<-b5_Mks2*v!Xokjv^cp2wK)c>LWTSr9)rVR|
zYIjtctB(F_+?U?wBP^3F+?V~KQs83^^ANvzCTyHAl+@w&a@lY$hMl`YJ#XLrYbM3}
zlG<|QIiVcval3BurXE{J%&t%OFI>4#ov5sdbA3%6rD~?`>}7w`sKeR2Q{b+P-|@mZ
zf|-0~7~F9P*_Uq8JaFjlMX^eHsqX>nW_O315F4@!^S@jd;{2HOk}}YKPG^^z-JFUX
z!dt?IQ<#yRjuzqt)kWIYc(*zFF>>XM5)@?IgaKe~%O4_Z(hFW@l%y4n;CZRy-)ycA
z7E^6^?8$qofZCW<E7VcwQC3xV@#k*Wk$8hqo)k-jonpL48AuCb4A#(Tqm%Q#%Ya|=
z$nWuXy=JbBWDNga$@=qsw`BsG26X0QVTbz|oyn>cRLIAuVJIZ*etE3=HB6H@>o9j<
zhY9O$XO=O40hSu@q^u1DskYqyEJ#OQ9E=~bXt`n?b4D1UNQa>c#E1pX<IPA@7$YVn
z;ptB<b0pw3ZC3Ojjq37NmJYQ2tgV@8_NxI#ifF-K)Rx4j)Mpn-*X;Bd`rhbyb$muS
z@xQ7>Id4go+ltHk6ZlXlI-0-Q_OjBJ|7IM?Jk^<5Yaq1!!Ms}Zw$wDlO0#sWRrv-d
zFbh!R9>%`6metXg{_3%k^0RLG9**K!_>q{gAL&sB7`6uG0gB1@8r+drgM0G6;%zrw
zI9MB?@Zulg25oAfmFjS-vQ#Y>cl9`H^0Bbm6|zSL7YAPT+Z)3O1ar)3<E<qq3%}C&
z`C<WE+8wW->dtDt<gOwU8c{M?no_V~wr<;nE+b_p`=$8U;2H$YZj8+e=ErHm`V=XL
z&Bs$)Y1id$t?Mt>*+>?N{Go(Gql%<3-1K@tO!0PWgJcZPw66yHqHG%KW>{rDLgq8s
zUx^qzdqLswBidhH?jBl9R51zm`lE^__Zz_y<TAZTCkBQb7gEV4>fSeSUSTu@Pc?*V
zG$PJxez)8w{OP>(J;TUuZrV~OPYPeQ6FHIE8cN6*`sy(J)o<mq{Z%-L?K$XI;rogj
zg5v(2>1NF^tdY~x8zRxNNdl5R_ahb8xh}u;94>lK+J;q&Ey@QEml%3p;aI@wJ!Z@&
z&?1#B0%nI|_OL=5B~?Nd(Zf{w*B&t6Aoq9k@qKj}?pZu?E9mFwLp-YqAv;1lzqqmM
zIb!=d+A)3L5J!sF;Q`uIdZv>(2yGQS@Qb=MkElq$!CETKkX2-~N*sd-xwDjxWxu>_
zYkB#K?xJGT<3_JC7fz9Ix4VKwGbt$<b;v#p6cj6dveCG(`-H0yz_>JHtNW{yqF}c_
z;+AyRe^9?xD_BdnGNM;m^cZKHQy6+3$?x%2Z^uBx)BC$@ZOUQH6*+SQh;!V#s#CK(
zvp-Vf)t&qN`CojQ5-sm)BnAF<;^(Xg6J`1}8STr|wz~ZN2MPMsVyV_((4CKZeBCS)
ztFX!YG@>-fIX+R?<u0r{4Lz#BQmIY=&=})c0K^0((y4rmJlex~q%DTH^ZXXfB+P&%
zsgq!&n^H}b+XW;C=kM65uOxNsRMnAa`X5g;4&3j%i}lIo?3SmE;nvK-7_9UTJWB>A
z^5N@Lb99$^=8cfm`D`B3@<$qgM7KU=Ax%y?GhX*ZqdjX~n2H^uz|Bz3tUG^;kW0pn
zOK)~!gV7Rem==5_QTbXybG89|>Ljq@L)H^+UF+1&{E-8+tu1myox<3xh7ic0jTA#P
zch&cn>SS<f`!Yb%-gWQ5mhTpkmHyaXq3yQ@MF>S)%}XFSBUBR=Psq5gMb+U^aSpru
z4*^|s{rl=OI&}>$`CfeX@-l(&n`qfMem5+*@EqmtcykA|;6;_O1E$oL?&uJUBD?W=
zy3@yI^)Y|Z97RLtiQ?n|Q^e$@rosHnCHxJD==7X#M2}BdRZqFk&A)u~WPqOCfO0r7
z5Xd)SVBdyg@H0^sUHP#%;=phV>3m;F;1pK}NaL-F#>@Z4xQ65Sp+!1aqZeFf%mfx}
z-UKdewANKl(BtR^Ttv95B>Do8Z0?~}Ujmg%Of8+z@T0Oyn_25XFR4BGhv$V)We)pV
z>t2b)u&jj!eHn7oTG5^;{Jen=a4zffwqPa_S^eOM3EO=Kg6~T9-MX^4@&f<K>)(Hb
z&*Te^@u|>n*g!rLE+grVB^-sX)4)u!7-A@#rEUAb;z@|?M5$dDR_)opvvqmoh;VoQ
zsi)WVRp0m`SVZm>epcPxZv4y)faiETEYB$lV6NbTG9+f`>fE(n9*jHys`4SXD3-JW
zoH`Cv(CG@%y<f%jO{5@@hc!vzN@@vz&(0n3@G}UvapL<*8ar)Jj9r$RY%f9{CIu8O
zAFz%rdB=k3RpxzLeN=f~EOi@eB;N<BVZ3qK`ky6C4z<Msx)wDkA4HP~0on1Oxz<-9
zeBcKxJ^+XAIznGhJ<<N1{CMd`)>E`Q#ymVoCw^Tr25WVU#@!F9E-ODrJH!z(4_F)n
zx01Rb7ZD=`2EsKG=BFfDFcDWU)CdDrqOf}Qr(7I~`Z&lsaltM-4H4n+l)bhJQ?2jt
zGPCkM*I=b%hM67F7^&h-cem!^I>QFX0H@H&y0MxXvwyj?JFS-PMU4@{<($$$Rp5VD
z%=WBV(F&JT2u#9nNLL7UUhdJXTb&Wwk;7t<d&I@VT|Q$aQLjQDaCw3nB!siI5Y!)L
z-EmaXa`*I~N;6NimhkzYiJhoT=it}v`vsc$`PwX0(w~{nNnjq<<f#t6(}v-SGYuC8
z4ddk`;WsFNm?co-F`=o~Hh;YzEs$o%ep=cpQp=KwsVbE8UZvizUe$<V@u*P<3z?FI
z4np2W|8U-d2EO^MqA{&7KLwOOnWSg)x7ufCCq`KuKlT;NHr0A0aNU{>Fe|hgy{n<b
zlEBDEdJ$ab9C{)0eH|1NIJ<d9eE>{pM{m1ztA~OpH<<7VI2xLr4?7&g8ZMaIn=|W&
zj^|&{GMF)fLj?6fR}7^m`*#Y<pY$!9S%{f;=TR&a()k&o)j&Z4u@3MvJGMN><|kI9
ztKqc{maxn;re%#UI7;aY=|V&tX@&!0pG*{}i+ng+%RP0s>UV8eEsW%4fZDYO4Y|tl
z)SEScwUGs!ti^dKDfkeAltDajhYMOe;`b6Z9NKA5C~Jxg9LM#DK06)agfOwQhy>i-
z*vLD-EQ_R-X=PLb2U7C33`C->s5fCbeRVtqutH-M<Cky&O>5e^fZX%Zd3co83X6Mv
ze9&$u#7aAQ;-{m{Yf26@RYmAP?wjIi=eF`|f1Wn=JT*OFuOlsRmqwCRN|u@25EWr0
z<R#rlml${5PPocamuegL1lzjMP41ls7KI2W+|0b3K1PB&@&$z$Uk8&@GAT3lBzEQ_
z`Z_{S;`YV&+QCCvuL7HdzYsYSqHNpHp->Uk%1gh#q2QX(Ri*`883os*x(|{A>621O
z`t4+B7P3izr9{=Sgh!Fy+tqSK2T`QK?%!;*(vX5^91}P0AezddA=*Rhkd8d1of++e
z@>F~!@_%*nndFRQv<_(y)1|gBY;0_aKfEWZdeFb?<G08s@otuiU|lw@_m=H#8o)0`
z3$D0659fDu8-+VCmi=lQQ>JqXiBQHiZrpB>kBW_t<HZdIWVEFse(qkh7pS|X#MXt&
ztjSxiI3I2mZk1NP2Lww6-F1z_6w*zD#NR!Z*z3SY>J}-mUkrQ0Qs?H!t-%VbJrSx6
zcaowe1W((}dpw7G{`HYi0nkWK(0$`KF!$ll*3b?~z>R|%pgzd?-iX{z_x00zG)P_K
z0T_%@a2O=s#;vYcM}_xO?68XUcSu_Q*q|wYDOG6e4V}TvEOan)2*>5Can>)Mo4nV4
zg?Ks=EgHwzT1^mM@M|>b$UIr45#F;VYx0)npLU9-S(P(r=UG1Pcpp3c)<OJgAbU3R
zN(cw(o_qX)iovLL?)C2^_Uis!#ZXK{3X*v-DuA8*L5J-2^;^D7zQJU$Mkl4^nZD;i
z5SrBW$aI1yQJ`=#a0_i}jWF<4$njl$msdSLqBKi!m_TNYXR9nx>VmgJZwnC$y|^pX
zDhrkIcF1piDX<&ia6w-woM7vzU**?8c&h~xYTTFj(?EY}5>eg0%P>?UrXwkRf$1+c
z5LUr&XNiuYM%rT`&YU;g$w5R>BQGHHog$zGvVEE4rp2h9PpsZF2v|AfvP-a4fQgO;
zsu(c}NAa=Bp%QYWbgHy{+-C$KC`5Qk8r<?X{*~P4wy{6SGr{mTL4(_x1x`Ek>oXBj
zimJJ1%^md%vA@W7M<;M3@2Rgi95l#@gxWVeQoiXHp>CJY6w^M|iB+-r?0EC7+v(D_
z?-Q-X!FtF-2=~#Hj`vkWlzrNn|B`--!K}mxF1K0Wb1*2)8B70y;+L6CnSjh(Oc{06
zx8Pbx>DUaGK#N`ISct%;`#iEK+XZxWGVW5g5MLcN4;}x#)~<Rk+u6G^0@G{cb|>o=
zYvAXwk(yKUoi{N-PZQSvm%+8%Ms+Vts){<nxy^=0{ySQ9P-aA><!uhXJxHRc3o7=(
z7z1y64hurYSH~y1P@@z!RNLU!wX3~u$LVFNL*8!p=W7aZw4F-3o}JjA6MNJ6uJM68
z0Etlu)ny*w&Er}iPGTN`n|DgUIoU%6ln$g4N8+T+rL1YJ5#;)g)aW{YKT80K7>N&z
z)A+FY<Ztyf+O#2}S{o_$mg*^zNsbaC&S&O)8JrOMdUMy@=k4!@V#3vnhl<*1>H8+c
z+}&aq=~~HPREaXuBrB%vdD?h<h%%Qr^iBd?8)PYH%X-f%e^#x^l_BC6Nz6<flf16a
zNW-TrO+Cj;&RgSvwAL_RSVybq!i}e{GrN!TS6XJ#1O7MshwUf6TLUUGpUB7(UtQ`Q
z4B+8)8%dw}TKZ;ck@T&FK+d;(A+EyUj}g3dMzmF}fPTFa2~i7hCpLuaGwNk6_!UeV
zXfKG)vEmOMdiaTm0kLI7W);Pr%)e(Do2{RC<&$NAYdI6dLszt~YuZxyJWV%T8bOf9
zR|YGbvO_0G!pwtlFk7;z1gU+N1!>-={slNWp2PQ9NvSm=mi9F;2KjYAITCmrr`8XS
zFFc|p*M8sv(Ur}*G7OF-&jPmjJB-3z9tQfg#Mp51o{f>ft>j<v>9_4a-BnM3nbVNg
zkvN}44tBJZydj<K9C6$tUq8;RR+#I-#Gj;RQd27kutORUoY(>^d`*h5rR=z_rOr1A
z*z!`crRnjA(3A3@us`c^S&Q$_qv7S&WifN6RCsTZTFvs>${GUBofMC9J4qn$1>Ot9
zOt&7GW3O&wO6&ox3ua*FCstwj2eN;t0O{_m4(Fi(e-Z*zfUgy|j9^raE{{Chr<5v{
zEZVVPI%F@==d738ksyW4DEN+%Fa?AMk-pFuLp!`NDlNGneS3clQyM=|k{>O+0fRka
zjZ397Smm|!isv@!U=MZ}$Z{&X!z}s?KN&iAN}=o$ci$1nm=QB1r#3?DJhRZ8LF~Lt
zk`hze4>7bfgkhx}W>I>8L?d^m+H6bYz-41-L6qA;$EnYa{#N_==9Pu&g}(Om3pPwz
zUWH*HX331k(>F%vM&W1S8&(MpXfFqKeAs!0LnKMVTF-&ZHizTrmF3uES`ZI6>$|h%
zh71IHea&xgRzqIpG%GzpqS$|OvTcD7?~}_v$Fy5>UaaFC)(vu*X4k&Ye=t@s{dDe$
z2xXo}MOjwgFCWV(nf59>RhjWEgpE#e4FYKOrfzen3vVeV94;`TF4*Za;qkrQ`Uc5n
z%A}2f^niGFIobTpyKm+Z45qhT>*5@K;EUyP{`$JN9)<r~`2aY*_|eQ#%R*Oeu#d!V
z8ql$wd*CfrS6s=mAg0gbd*;_H0L7HB=@IKGtNSID-t4}Ip=y25q>Zt-80o30O$0AN
zs7hD-_But-Pw=cB(T^Yhh|!v}uJ6{te?dAm@JSo!5#RDnSj(xv@NT@|BMYTF2_nnD
zi|}1tA?o{O>r|<c&l-cN5<plbn4+bOxqgG?`icNyqpMxRhEQT}2ziYuTE5Tga8C+t
z)iRiGQc=f--k-=nez|oMfeknW%P}4SQEmJz!IEff(jKNOboj!SPl>hFV@Dw)EQ~Kn
zcqC{b?^glO(w`5@!DMp^$)X;y+5>ZTv{y8BK(hChZ$&fx1)464x|e{)8A#kinQdvb
z*&my>t&|AGb1fH{8@t+WJp%N8$^JL|F%;zrT({_<$kHA*-*1898l*(7@H)pVD<%FT
z4$lTk_+Ra+`=<ss#1`G|9RMuK)<D`G8jSTU&w(prS7Uh+6t4W`S$BhgfIzEax{H>K
z6JL6~tSpKgvd2{3x;pzYW3OK}2PqLNQ=`|U;NOb4faOmUk*9##cc0Tg#%X-j%%dxl
zHp@Q94p^DuB>#ARA;RlC`9tby2JYLT6o7J&G=1E|c9$h{N4_mPv0|b4-s3cnayZ))
z5bX5SU2XPFbjb1NXE*k5VhRYvEXC<m{4n3gx0X>)V8PGy%aw)J-^}eWvgv%1u%hVU
zqPGaRps9mZ5+<mLqJzQMr0c9{GA9$v0@D%$!0sz9tW|z2PdwK{I%D;C;?_mC1v7Q7
zf9-hEh9=-k<oJaTQmpRM7*c2gif%oL%ZMV~WEK#`7Gh>q_q0H3DPWKhc*2krbyIYT
zFha?ORRzOBRNT2Nv5}spsk0t)L_xvzNNMOEw4N#fYdM0s8jx0Q-3DifkJ@YGMQeE}
zS%*vZ9yM%?V0K~0*>v03N+z5G+R6;zQ%M!|_&?!2;PoRJ8xZhmMOI$s(rvsqfh+Se
zHy<lDtGVK0!VrAekK-D^)R4O@mQn_%Fj`ehHjq73(U0{^#uIKIJCqYFO0rZ;pg!b<
zpKYRU8`ws|RB%vxw-`@tswWA&+)4w;Mi9%?IM%JZlU5{WgD!a00dNvvkq@4#F<EJF
zc>#J|GA>Jkuc$1Cocbgw(DM#`Z{~Jl8Svs)mFZOQoJzCRE7KcFkXV8b+H^7?O5i$M
zj|Up<VABMDBKXV=>4Qrq8aMJ8fbsStmt)KnwlOrz(Q;Nw1y(BiiIgHXDy_B-GgA5T
zI^9H%BH>LOf}G(FvKwMp;TqFZ&myi!<dPU(E5t2tCWCfpR7$~!{^8u*>inPxYe|kF
z9^kUc6!c3yF`}tu`;Z|i`(%@-^_CbCeCsX=6QGi5*NMchp*G@X4%R~6*=-&MytVEa
zNk@6oTuP~HnW)_5X3O_Je;AAIA9LFRwS$@uf{84JVv1~;b|UYU$pf$lZHOCDiVWpb
z4VOSikvugoy9Kmy^sV|^GQ}nW%=^ZSA5_|1(XL=7YQD@zEEv;h46?gu$}qVcR;!oa
z7eDcXWF_~#U!lzgpU4WngpT-j)|z#pA}UUs1gJyoHH#7^JlPKcRDS!QvJP)9HkHUf
zhmP%>6C?`L@L_c?k0uORM&lt({Su#ka8N-mm#WK=<?hXTWZk%Y>E~>rmZ64)U{)Sw
zYcw}J<+L5<9q%)BR`cUj-gZFB>rfI@iRZfk%=c%jXKw};1Ta0Y$71y^NltY@UDk_`
zbi7d7!;FOo#6y!#<<zjZi+3|sxAlU(y&a)v<j4!*@ii~W$aXk(thG|3Tow}sB1)O$
zfljaRBlyCr{SU`}yz~(SNIq#GX73gNKXXL*@DgMu@6i}5&GgnR0Pi;0oh~)Z5a)UY
z-@tS{B7A@Y5fy@0ndS3%Dw}<nM8_$ui9ZdLF!R#bpq$|zab-U7q1Kz*P4~>)u=mrg
z3h->=a$b`J>`{OzbExza{}kI#+Yg_qgE<(urW(RAX&2}B*vz{{=J+Zn4sJ|$5pz8S
z+}DR8xOq8`5xHfRNo-m0jP7Go7w3OSl5er1mf9#6!fg$gQ?oN9q$1B+#1`5;OxHNb
z);sjIEf3y@hPFFJI5<Gw{D>EA4__}11zwYSRWrg-!XDH98lMK{Y0!%V@CVaab|wR>
z^hSS;!@l;HYjCS!0FVT*t3Xj(0_2{1erQWW^!5)t(qy@Y>?22f>V{TrDMA8)O&5qu
z4}LoAlo2`s<>-pfjmdy*@GuX3!F-P4%SSaz9<=_z35bLCnqcTGQjqjir}ey&d6(Z%
zq$aLrb%JUzHsHb7MocpiG+>uxWy31Yb$&|5>Wv&75%UVV0~(EQsP64Ao<Q5OXa-*6
zzVEnYb%7MD(%{uW$#YK%=WtK{n?33nOzf@AVOU3<6%$@y0~;={`1Bo;AOq=M^NE2w
zv!pExsdX~z#_*40EIr$HKJZ3k*jW6uuxQ+tcPJft8v@@61a1{o;s*ZmZYa`6`DcEm
z)Z3MFeivb;%FLFNDJZ*FnuZC8xv~$%n-SAQxeqgaJXmF8b2v;ohL^9I;zO`S1<|tt
zym~;US`Pc&=mWh-Wkl=pGTqccJoZs{tcm&ezmLIkW~M>~&)mgSs<;;=zy+f4@emC<
zLad+C>TDZKj`9-77w&m4*2DrwfOOv>7N?3Ta)BA>G7g#=@pNg!24AS36t*{-#)d+1
zCS)WxJEvexUd=6bt-(`!#k7EW_(xfuU&(J5O09gYQ6{&*i%CsT`w?ip>>n&!3^4x&
z;rQiuRH#GGVVOIR1s;D407Y&Si;+Ch{!#bx^7U#SncOhEJJP!DN5DlBzD$7D@(L$x
z_b;_J$Xr?K)h)Oatdd3kz*K{7VvX~};RDMd4;pj?B%EL(=(f`3M24Z2j{57)qB#V0
z-DxfMNqOtx?HT{L*}0^~ILsXCVBTCyp^WQzL_k^2;ZyT)f{y_#-BTX#if)>xhQP3{
z4~4J!gus($YgBL(;Kg5pcOktWd%h-6JR}p?E3(>~z?A_86+TKWUW!U1z`m2NT=UlS
z3s&sQv$>=B$P+ftfzmfhCy`+qa8pV}Xm{0t!T}qtF$TdqVB1y5nEkHs$0dx!zw71}
z3qyzKcH{_L5H^nUwDtU(HOxBRU7sU%qs8~A_oP_}k?ALD=jh8p13I6oJZS9WF@R~7
zjp%&EDWOUIbI&2J)}4j6duLF*<rZ>J2uNDSixABi&Ivqb+g7`?S#p%>OM8-`lrIKp
z0~!ya#iE5{JMar*c?f{yLBPmpJU<us7L40M-~av*E`ZlM%=F-YsC&zxINt5w7bmy{
zw-6kHyAuem!QI{686<cH2`+;LcX#*T?k>UIojcjT|GDSf+NnBwzr0h`RXxvi)x7BS
z>F=^$4?9l~fsERu<{}1J89M!AIhOf}o<=HPk$SWqRV<66CI{6V;X<SJ;^t2jF)@IK
z+IGQb5GPK}=AhT9|Mm%kznzCbOr9K6+j&;?id<NFSkpMk^!8@Go!E$c)5oMzn$ohQ
zsi?(m5kj}Cuo5R-^P_#0$dZp<FF*?;7JEh!cFu$5DTuRT{d=p2N^8^Mb8MOPs!T8S
z`2fkR!jF0-?Id+vccBo57pOIAtcbG$3iZ4Lk@Y6K1|MUX^Q|e5=$;&;U#twPSf;Cd
zkX)CWWua6_OQgaWbGI6upw}aeF1DIl6**DLux@H$9Vk1*bDFJ-E%>I+#_6;!M-pJY
zk_bKz#W(Avp6UG}T9Y{{IZQoaO9o^g=P9`R%yFS`*v4NdE_RAN?8&^kbDf1jK$>qQ
ziupn_#N&NRp^arItPg)C)%x}Dt1~P>(*;m#vl#k|11qCS7!fh_>x$D@r%dHIsLeDv
zjPp{Ct1m>->|-~NXqx5r44OsQ^)EzkoYK1U$l7@Rs9BRjOxb>G-gzsToF;Y=*TOI6
zcUc9W56HSRXTJ*)3xwq7d9igZ5!c=8Fno8k;_;yvR2JFNaRSen;a^Ye$)s8uW*xN2
zvkiF?QxxzH?Ae5#oJys}GSci-sJ}9Y7zET$^_tqH78;&MM%lkhseE6`gMhqI&4Bv_
z==5B9{rdwJ9(bQxnCOt6POHI4Pc;`s!1@V25iQ}`9I9H)9&v@tB~l~PZ;8=jAtWW)
zVwbiW+>=gaBNz@y=nFQP&G#)CQ1hwQ1WSCYOP@XraH2l#*mh=ne=Xb>4EE_m)K@8E
z*`>%Qj$L?3PD!ARN;M08XZ7OolNN!OJZdNAL|gV3s%74j4a^!@tnmt_M@+XFsv#XJ
z2{4xd-#7gWEm~oZy71XKnxhdLFqT^W5*=y8mgmCgB`0D$f5`%gpE!(S%eo+2g%wL_
ztq5mCH2$N!4b$5u=%V4Xlygvo`mDc3(l@A&D{oBQXaeVVXPp2ZMVCI_*EPBtIk$zt
zN-0A98{$%dQK>lQ7ZUapkuOwOaHJNWKX6<UkJJ#{_kXj9_9GL}hDq&3d#HFyvzf$z
zt-UZAfks!{MWVfwYjKonHT%&r#M4Um#9;l^@4X!N6W{V(PXbW+GKlK;&X&Zuv1<vk
z>A@uZQ~@J;39Gh@6;{D-!If#LLj=Kf?%P`T>*?jkF?SlwA0m4`baOU{5*^XzNTX*}
zLH9+}0f%L8jc9~8Or=4Fy6f8Tv>=)@7C#TX3)qECEe9m;VsrcPthG<_e%y92xp)qJ
zdG+veBo_N#4P#EDbWm-0IA{Q8s4?OUsLnE!zHCef!T4L%6jKKO_pW7y<I5*ATf+dm
zC_baKP_j&kVx^fqu}7MF=$hDXi5VyRQIsbXKGwI%^aA*1??s?)H~S5w7gL)ccdVM}
zbKk8AgVNh?s+QYCXiGJ5ll@noKJ`*`l}$S{OL`qE+{QKP&d)WWzjf&E;+9f8NL#!|
zle4lZn7&uOUNbc%HbNJqQ(gDQqTMh)WjB`$d^CtJcnp&e)i<(_mMC@KpINqATTS@s
zJ&MYP(e?bD@fR;59#JaVf*H}Y5;3=EP%^DD{J@ym>hNiL;?doBq<4v46oauPJemi-
zV6Hi*ByGw?(;V9{!g36j=Jh7yu_0sC6aJM@MH<rWChoI(rGO?>%nZU|Epa^u4N^R5
zBm)vcVMDnz-pr<E?Cm?-ALuvO?h5P9Q5d3bpktpe%55?v$S!zY&;k=5;!4Rgzmdg`
z&8mIR*3(3q92hy5<<pCn)?Rs!``z$^m62saZ0jz1O-sXa@p;6QTGGdJh=%zrzheYI
z5|Dt^9cP7FIsU-bF*}4!1uib?7I9u5r7~>Ud`30QZC|UUM8Ha{u@~20xRBxLN&!t|
z9U|t4pnBpWMA{i&O>7pv89%T<&1>FIdLEa_>2i9+<a7OuZ`v^5hElx&^Os*k1DGch
z5<PyOy+T^IUc)PA-fN!W4p8PQnKn{1ockZAk>>qG_kCPQ8yLXH%6!6FIMKA)hgoZ9
z@J2fpcWu>y5=;V-p`%D2bZr^Y`HC0qSY6l=kfY|asShMFYLP#RX2HZxb(HvH)to7<
zoh>ak9(vF820n1QXj9r6Z<dcGsPQAyU5+Zh)m5w?U#8&u%KYR;x1W#Ln@>c@#ZHY(
zW#);pzU3JEh{=`dTSp#5gLqp0p<tlf9lbik`<1|*PN<cETR?uFG!<|#R~TO^&rj}|
z%Pq&zZ9>F%aAGx*%Xjuc-23R}sO)zkgQXfG$yj9inbL!_{cbLEV^1Od_=4r$<4yeJ
zb#hFFcpnYRgJTBj(%^K5#}y=sm(RX*@&GF*K{N6G^dR{I1p_{JKr)VmKGhlNZ-<YE
z-9MqfRZ%TU2WV80i?xjOU{k0PczNkO8jJI+(|nQF?>l^3!45UK$zPnloUdBF@SYS<
zNiJi~$H;6PKSMN<lqDB84=!11y7g$=w*q<cSum{0Z0B_cIepzUMPzKS_qc4S%acPb
z-EwVWc5$aPotT1NlXc&ah@Rv2UOga~gy86?)OEhie1t%7rwyF@G%1ZFwYFYV)wh6H
zxe{6(u1b_8&TE8K#x&k_6+O5z#$?@wWP5V^#j9|oxS${jnTTNCbqT>sEoj|W`Ai8Z
zPgjRnZvjsMgD#s-6@|XWyW0=iz~py+*zMCd;v1S1%$3NVSOp<Muaci}MPd3r6;BCX
z2$HnI#Hz79E>Kx%#CP<?=C_{E)oq@0Ivf>dP1ujzPoCJn5I8xxMnQeXIR3NGQ+|r=
zINGg*V`o8HKkPStJu1xO0@rT7Orx&sRz=vWZhg!X{$hcse;u8V;u_nbgY<=w%{c#d
z=@upbL}pHd5ut30ih0cU+$g+YIZ=jXf;;19wJ?mVCQB>C__n>T+x99~#89-3l0#hm
z8&4MGazprL(l<D?*vlBTto<Ygy|-XgGV*9uZHK5^X>&8;!t1Sb6P_+VCb%+;nNq|S
zmhe1FtcB|B(R-BOgD*#)vOEYFAoZ5XQ&{9v@$C+)xl28yFVTFBvN}%S&<)JXA2ab|
zuV3`B8;BI3E)65)bfOk6QVTQ)Vcm{9AL%*!k6T<FLu2v9hBvfKX}{HEXiCI9W+KEc
zm_9D1GhpXZ3AaXJqJL$a!g3)xEvtCr`|z#%n(ONdaPKn)?FRB)OXp$!w31Wvd^O<H
z3wepqjc#$KWoL^4=Zvo6=^$4W8jr&UMwSFS(8-c8dxZ|hu^&7EE459<(W0>;zE32k
zrl#OKO`!bKqwDp>tm0e4eqq8rvYE+KjcJdxvKhDU79R3K0%rC$lC1oCy;<Gw(&nIV
zDFSS9qKfA+Zrd{l6X^X6AX+Q)a~+}iaT6MW6ea5uWY60w$Y3w1jG#HI;b}1)=j#Z?
zu_`fD6*@`mQ6(!x7D*#aIMMZEE<^}(+l5`EeRr>tZ;^dZLnxxrQHGNhqVo$*g^8op
zFaA>E<5THg)tNfy$2_4k8qPL0kEOO~>#R28U#fFOrW;SAZ%Y?VlZgGT`&uX<%T6I6
zU}9s=L%c^u5!K)89%cq+vntcm;R<xe$j2j$+XW=MdVM3VD`l<QF7IcwYvUNW6eb+B
z7cT8fA~S__2Y$zR64GqEEy}+s4H$SP|D1eiXU9bKt3Z43GggLebnsUGfWp|>qYzu3
zYXR$ffMg;iwNOt5`(*EJ;ERIV9=%o6O7Dvh=dLE^WxywFYQX^=*gzDYC3Jj%4{2sD
z_6;CX_An+w#PB1XjWIPyBaq$VF?oG{EBLxH^3$b?1HiRK_;9kYt6$eok~nqTDwKU@
zOnaYi*vly;?1N@S3kWaofw$LtHTCdZT@CFXi5T9K-j<rif_BBs9zR0^yG(;?lRy+-
zyj5$9Wkw8HPm`dl{_5&p7a6(p1zU&gmMwxgvrZu|I7Xtf4-$KM76;*D@ujXkCR*KP
zZnt<vcbn>gRVra3JX*XtzHDiy;wHx=rSYbcjInrz;eAV8bk4_SHm{%NXSh41bh_`b
zVVWfIGzhxP{68f)Z+~8p+c51isxw2U)seuYWW+}hi=trXf2QZCOlUI+Q^;LEjDbVD
zs#pzIiE|=L2q7cd)0660g+mmUu6#J|j?K0?O^&kfKPGGl(uPjkL9#`(oYa=neLvAS
zyZp#Rgq9YvFS@oC#lscYmDhdB>MuSx1t_(!z&ZNprhwEEC47-o0WjB(<K|vXjfm8X
zXJaNcH6t;uwupl1Cw#)rS@9|G=wJBsaqC7ox>O&4pYDf%$EKkVQELHNcho!&mya=+
zRW!6)4bCl0zVN645Sxb=a4bj~I<j%Cyw!J?$b)*%PN()O;ixvyeox-A@O?QLIk)yO
zE32Z@a0jRv?K*H%7octD#FStne-8Xy!ZQEZ0U3~0izaN1G|(i?W{Q+A%n6AQiZ&uY
zgUiqm)q3Tm=>y@-b!)A0Cp3_fBqf(>@i_Cb&c5a0HF)JP(Co8bym82nH+9j7^{6-W
zu*EFFmtn>StnBif(ir2H7OyhLWt@ih<vG9bu^lYTONM1ddTU}(l!c78^m%EJ-IKH-
z14ZS<Jn^t1zHuIo2l8>Kvt!_wI=bj@Z@v-k!XR?4Bp+VSqF1G;LM3X3XJj$=m_6Y7
zenp@S9;5Rx>g?{PN_zb4sw=u7>x3RAaElO=>-&?{_4>RmvkDhOe3C5X9apcVb&8O1
z17xm``_%6O#8e4;$W(;FB*?nP*r`knN0JBAKEOxp`{`ZB7_wFG%S3z%xvgcYd3B^T
zQVW9EEWSvlVjMh<DKGc?fam^PwYkPV?>ztNZ}HhJ1Qz8S$C{A*mKS+Tyn^~pObJmw
z=|Rz7qCPG58oj>jh0Pl8J)0h5i-Wqv*@DP?Rll`Q3H96vhNq{o$o44;AkT(|VMFL6
z7xAvt#;x~54Ecc)B#tPkp3vra!Hd|p%<TGNtry1s`=fGX>%jw^hsQ?qKyeVg?tIn}
zEBx8DC3YBO)go<<D`7Yz`U957OA@MY48<c6zFt=?-+HMVlZi#9$YFC`vLw}1=0VXj
z`*^utKAT=4N_HPVOYiSq3zvkYwOuo1sP!;!LZUSNA~c|4Rc^9DI?89KIVSlNd80*K
zCGlVThWKUKR~j3xwJvcZCbz1cy#z!YX(Lz4w;@edp$o?@4ZF<n=b+XL^+caK!;h*E
zgcsEag8VUZFmK9t-y!!$qu9-I6@R*vot70+eeq@2Wz!~*B;8&vMFtFVZv}=#8xkH*
zSeMS?r7qLSGpPQ!{k-7wwCF;ulj0z=+9>uewn;<pt5X8i##DSKy3JerBC7DA?!EoS
zkIr7ApAp9VoK;rm1sWYdw4B%1$FG6?Tw*p|#t^Mxja-xcp$<ahbyKJGXt)W~Y<CVs
zgT#qnhzB6{=y=_6-5smban=F=Y~C3|iu*7hZ|dso!TsI6*RXq&KB4-ho!+af`HUW;
zD88ji2_-6K8>L=onK`tV6Y_e2L&`tL@mJyVnHhN~0t6UR0+)h^F#;<;+V>k?O#&(U
zGpPq1iG0YZPB&PWbru}fIFC}i9*$X$bk{>dkdZhR?Xa<JuCj*r*dD@&P|Hl(XG~as
z>X=5y!#~8|$d;wX8m#JuYnf*F0V;6ji$97GPPA6bo*l{F<e^YEL;mpOj5ZOhxCwq2
zPt<QPV!0EE%_L%LN^~BCr|L<ffIo1n-pgjH390hv+2o@p&V~Ln!XiX%;~PQyMzc8@
zQKoRk`udhyHYl2yK;HbiL*(0}aVhjW=%Dx7i2WJPzd!OtNVuCc<!YJg_@ohL#88KR
zo(<bJFF7mqKhD7vnTcHFY+cX^?y;SL`1Tud@q6=O7&Fy#9xFM0Atil~<nv3|*jLuj
zbowz4g%5~1>gWMvhc?vuW=KC42u}*5OB5+Wt)lyoVMGM94@m2DKGF`^7T7@4`FJJs
z?N;LF6OH(L6XnPo;&ss%weYiE0K}+g{JAaFy_>v#@a=^wo)U-F)6gpfZw?qIqbjCL
z9hVU1G9&3aiq%k&z=V!Oj3Aa1qrz6gPv;iiCoBblJbYUz7M3xiF|IXjsaYQYX~cTg
znb#t36`WLdt5vWqRJzjXRN3206{H~nB_#qXX71~33FS@2R+~*y_1To|@F*^a<elub
zOQ9Td@%Fg2eIkd*?%w8pyrFh{RO!pMgyYWSqqLhf&l3ndtEg)qTp>85aO~B*_pHbQ
zqK!$GVJ~KKsfAPV7B62|JDsvIO+Bulj}LYlu&y*j>^zBmzG<WUd{>9*C+zKE;d;Uk
zTpv)~H=~%nCX>6hBLIB)=6?o5P9Z=#B3ZwfvmclA*9nT8|29GK12-$zhkrhZOyZU{
zE~ZW&n8a-iT};228rz$gA_@v3I=eWT8rmXyEG4LR*)4OUw;j<phC&>C=CBmQ#Kwb@
z9E5UOs4x5`W*<yGSp=!gYVr`%!L&ZYjr8d@hQx+HdTpfjjzZbp&v!Y$oK#;V)pzdD
zyAe2Fgk<@0cM4ir<uG;^2sV7h6d`?Fo*mA4<cwV9964+USY?FtJk!QzhManw^hDmg
zQIP7P^>n;H{(3jQDbbuKEpSU#p9kQ-J@8p6sF9jZTBXcw$s9Q1AGI&)Pu}bC0iO6R
z6n)h7jW4&D6f$fk>fijDGOl%`v4HF=kbyo#2_C#r*QmKl8Uny{KZMo8s!@5#PWz4P
z>t2w|nA%{U!~@C*c}Xr+v$_z_YaN*;e0Yri)GqIvjM`Ju{8qTXP}`O<J^&FQO?^~%
z*durv!Q;1>9;o$fXmqwZZoik6c)eVi#CWSe*1{;;vdKX~Up8+wTYp4>{Nh)RZ|S+X
z-kA1l*7jNJuvCkh{+n4%ZeY>ZUteh&WBR(MAab%?d#en{$Fvz2G-x_$EUOti3p)$x
zz(dQ#BN<S}!?-m$4A@tu{Hw^YMi%@h11DL@jgniMD*HxN@2t6`ARIZCYRQgC{S9T9
zj-a^7S2}+JK1B8Cr8~DlGUD-V#CppZ47+ZruPj6Hm6J=HcrrO!m3wc5#P!O?IWirF
zaCmfsR6_QV$44sNHJzAtaZ_1#lVSzfwJ+^{&W52Fdo;-OpWCs44^ljP|G^eA+j;}w
zL%oSC(eP4MSA_hM4DTp3yY@Jq!uK+K(GFX@X`kjessVv&2IKuw4Zo;1?4$9n3Jsyv
z>4=-X3FWHSKrc-{$TRFCrttAcq^+7U%vAb0&5u_Ru4XgCyljW2=^-S>V~bti<3O8}
zQe!-oJF-NB_${Cv;9jrP*uoKo`OX1_`Ih9^`6-3@zI!a23-2KQTKW!f1+*i3MfuN?
z*C4)k+YWMj5~%AgmhIYOFvW|~+yi`iL216F{aS#T=qK3fF-WS{x)Z%OA@j;!vk7hn
zdD|8k{qfvo>7gD)q}5f}Fk-32L$?2=^K)3v>&x{?`Sg1<GcVGKqKK<&=Ab+Eeoy|K
z?}FFmO>J}+N7iZ2aLXvCi+_cY>u)gf{5LR0ElHxcJ<)rik;|c?u6o!skO$4!8i)5#
zeZ>>=)@E^I`wqj&y36d>@lqTS{{D9ik#O{7N%38V+|y=7Ma#_1wv1#=w3aW@`~{~g
zS=rl~er#x{*|FUo^GK%XzNlY^*Gq1@ym13kjaO;Is2lef=|@jKmjZ3l(^CC79}fXk
z@F0yGlFS~VgDAS2h0uk>$BP;JJr_aZsm0ub;a7sGmyUwvf6<t=x9t-F{o1JB2Y2&?
z5|QYOHhsg5ni>C4SQiyMIx)YEKdBQJyrBHOWwXa|pZcOuZ-bjfW|p)?6ENxS9)+Ld
zdFPhi<H<+VZ&Ii9-Tq?BHDm49a;V9L>0yrxk>2d~*>IF?%kpn0#F#>tYP}LcqOJPm
zyS*2l!amZy+qJi6Hv}uG4wR>Zt)L^*^<Yb())14t8?CY>xAdUww}aFliAg=_E#W$=
z{S|p*#je$?7MwD4$ppbf!GTDcpFW{5BWg(yA#rZC{tk=`h62V?9YFyRfyE+S@dh&s
zQ|G|2I8|uz9tIRVLe?%I3LXO<1KY=trWPda6<yc26KaGH<>G0aK`&U$HV3A<5avOI
z8Z!#LjteEu<y@Vx#_&2$_6Rmj^3M_8+6DuzJB0#gl=7plXe$n}$(F3AVyl&K3dwOp
z=;@m&__vHQr+v>NZ2iwl)uCoGYjU*&RffoCZi9Ghw*7c(Cs^Ak%${7iqc}j8^+aH}
zu_htA#-x<DRDoxp*%AJ>Wfh0<eglees)GB}PNPJ<DRMfP#Ht3P7oL0UqXKWOs#hW}
zx;cGCrX!|}Hlr8H3r)YzK69EP&V9TxCVEw4Krc;m@wU?G3F8qSJ`cxOHh$hg9{5C1
zI604TQXJ?4#0`cF%pHIFzX$~*i2Ne&6e3T})6)_M`jHj~vTXmKB_<x@ihoeN*K-OX
zVd7D)*g-zibE0bBk@XpVQ|Z1O55;AA==q-Xd_m+jm}$ldVjJD@Yc1f372-xY8~Pzi
zk9U*EBS%jSlm3qTo#RSJX2+^+>ei&}84RzgV+Kj+`Rh_65Z7fgbdBpSi3;Fr8TSUc
zQGFTlR~)(jh9ld5gQH!VFnZe+_OfX1j&Q3r)TabfOGxsE35w#J;)bCBSF1rdq;1jH
z_9wIcB9)s75*q<Dd$Tp~jo-V|S5$7R=ZD)ndL<RF{5t(}1G7mdw9QRLp|?)gFW39N
zdo4$^SLQeTs3-jjhxJ<XM*yK!*vn>nw4^b8O+V^{SKIZse1qs_-;4Ea6xHJFt84~6
zg$tL3zPRx1`IE(gWzx`LcKNfa{+Of}mv^IYvjtF^FNB(tvGxG|v(VQX);p~uMU5L>
z4kC;Jp44zkj+WGwlPA0P`iIzz<ZSAz?ddzXz6}F0^r>8DdxNC&FLq%#NE}ZDj2Q(N
z>HMo=H3d~y+wFeJ4R2IDwKj|e(_Nw&Q<F|byd6ksBJI)3IqEfULx(wczh8D{K>)(r
z$&k}4%jHMM<&Fo7>>DM9R5SdC_NQ$yhx4lmEr>d+K^0B%5vH}w7M#v>$%w)4gMmm=
zG$VgFgeOAe+^F3QL<vR)f~by=fl$Do!d*KCDho<iK>r*u6!B384m^4=huC;1cqnX8
zAx$MnX}Y>~LJc`E=xRA~gx$ZP!5n%fKvD-;Ys{ei;ZV7Jz7yXV9^`C~`iH|nZ^D0Z
zD6=L5ia4Tl7?m>0mukdy?IDl`%VHo{7G*2lGn|=r&B3y`_m3<F&ien6MQ5-iV*N`N
zEB+;moW{ThPV@h-EH1~gWZ-DCgaEYJ@*iBN<s!UI2>T~AD_`GxBh6(7>Q@4q3t*2F
zvI`7e@NcSLk=MYhSeZBGz2@mW?6{?D61%WRvV}>mRA6aD5$Ffie8UCfl@kY2HW*Cd
zj%9P>8N?qCj{{ZBOO5@SmKy8*0oq)Z8k;x+OQTCHTPxoneiP4N%C5{mOP?ssj~OV<
z&t(SjpOu_KGL)S1^}ymOGnjG@mPUp*WrSfNQax#~E`pD7L)wSwkyPjWl;>zB%cvbR
z`Oy}oor!niZHDrzzYsoCL!pWgU6GpGd*cr{e41|YmsbT9Jgxb)GRpTL=w4^vjoiS7
z>nq4zBjMW|!TuFQo_`0C^S^OYRpjh*nBTWv(YJ<59|ZKf(FD*%1U<xneN?OB*)=s7
z&?hCRcfYcqj^?lQrld;*YD1?sS=ch}Z1#4SujUjE?aW{L(%&<=9gm5czFd1RdUJ7U
zs;4W|r<0VG+4{cl3Wn`vm(;sq*0{vkE}1pE?#O8qkJq(TzbdfaPSldVFC{mtxGm_&
zlU|yg8+;z!Q5a|wiX6Ig|84PdIe$8iciYCadATGjQSh3o$X$O=p*_JWbj-}D`}AQZ
zJ$?%NvR=Lgi;C6jN=DrXaN@jPd*fg!cRW#8P+OR0)!aH+!`n0|(&|e{?Y&g{G^SnF
z``qVCRInM@+<ddEAZi<$FA%m!&Bix+yK*P--ABahsIKs2jp4W1vqnZ~vY)Na+17l+
z8lv1;1YZ%z=XiJs)zrIkO?s$K1G{gj#`a~rfs)?s35f%dh@;sZjiFA*yW|TW=+mmG
zbf_+wIN7GiF(TZl_}4CBqYo@dbe}%3oU<KoR?_Fv?C40+&U1XuDi1GS|GiO7XeK64
zMHa`))`@Sbg22oNi?k7VKGjqnPf^<SHR54|W|SK~%3l?@L1zXW;We;@7@1rZ3RUhu
zpKmP>iDvx82I2R4u{Lo0C6?dy9jK-QiV=@oLGp(A%UIaGQvm<h-Y1C6u({th7<U<+
z)I<2j)JCWDpF_9}GLe&S+0>H_iCE#;fJ}=U_=MevJlnvdeR`_5C!(7mZ-oi(?l$U;
z<m5X7PvlJGZ+gH@9`xOZ7*3?{IKvFsZ6i+}*xDoRg2vD@_Z(idZ!MfIng<ixja+5E
z9pk?9qGEp6q7E+@J<H0QFR;BPPHlfPVlL)?H{$;`()vHtr0svH$?u8W|51|&lW)Nn
zWPa<uY}+cuUV~k8#?NNEZRQ$wg1*cB<%Ie^AfLIz7NO{*q8rB>SN8}zyThtBq8DZ%
zul0x1$D@bu;4tfCZv3#`s}LLjwIl6>W((->`kB6DsyxmE?q#}2T%V5p`n8KyjQ2#f
zpI1BdP+Mp-(4IUv9iOZDo;>c398wA%80M0%SH?q`H@8RWTnf=#XteHn6!ri_?KD@K
z*2IWu*(LqtQ#=w#j)get<4=Hns;VMA1722{*_)G3>es6zq9Yb-?dj%zK6tmrwOU1D
z6VsV{2azkQV&4mju9SsZi<i;%QGJ((4_wYC3SDaTO2*E5nqwZePxoiHwkO-rJOF&2
zb-C6(QA^BQK1+3gC6o$5Ez35wl6*}G!2avQsO0eQ>(DApxn1|X5_KQ17Cw6#CAMs^
zNQ6j#Fva^sI6T-gXjE82@OIf<AtG?%@^P+k;*jEBvkaV=>1>KT#XBd9;c|8p34kam
z2K{hAIA91qWY5ZM@MwAY(t3*&)Vq?&e3Z>g61@8B>LGuuZjMif-do(qv$XTp8^9{)
zu_i9~dL3#&R~t&@Hm}nRCD^m;yTNSSrC^z+Soz*;h(c*7Pxsj+#GGoOWuyuBift6{
z%6X6^mBEv%%c?VkrFEGStSSiL;||Exxw9I>na0xZlg36BWokQNAVz&cnZ8SsYCDMM
z{=E!Ow4)4v+UF6hC~v<wf!Q?%FSNHfr9!YB1~1o{;FI8tz#H?K;YT(PCJjG%pSeto
zOxN8Boyswy6tJvh>&>Nr{XTXVLgz75J<twvw7ipi05~*~99#HJVZK5w@6@kH%|ot0
z&4VT36yi?JBia!M$_5L}pP*}Jm%==i>7S*Vcu-YCJgAH5A64;^{wLA`f8CN9>uq?>
z#Z09Be--8Ke=ACF61&7De8H|Rt&;^z&Kh%p-A+&z79>0w>hm3RW1FlpciKUjZ~@)e
zG~oP<DQen71lM;z(%IK}>#!*N6{11<#_q2$viv)YoNO%rv1^z9e=5q!|3y(^rAyjs
zLt{2g&G^q2$3`2HA~R=J=N_-TY4uJP(nE$G5ALe(?Ch$`d5LZL1T?jmA1{yYY4k$M
zmlmW}@{ep*Ij7UfE0zx(icf-X{9>z~@?WwHh#jBI{2mF5b9HUgNBruZrn8Bj4=S$;
zzCOIx6P4+|F1_OSwBJ#duB9Gqd}`0ak^d-2j#0aT#s`#}7esVr%5w>BEt@_KgN$4|
zUdc_?U0UO?6#(&JYhRzxG92coj@e(AM%hiF+pi4_lFXl!UxDk4&H<^81ZzbdeD=5e
zGj0}3ZH+Io4%924$5#UkB}w}_%(~U4o0XN!p&aZQCfT#Y^V>D;>;qj#g@h}-r|N3a
z#nZBecqht4v8qcO?YFDtK<wq!6av^N0)fWRM6L2x)1SohIHzQJJ>}>*F%D#Iu+Z)~
zg$5s3Vv>2VcVqDG@oya#u$A6BG3Le}#ePss`Z9aw<gkcR^N~6(Mk*bD1IStolbjyN
z<M7#IQzuCpb8dsC&&MI&3<NFFNhj`r%_eR}T*(QsyMH>o8PD>}QBks)gLH}9nW@ZK
zwsBSYL4g?DMOxvH)c*@hNdWl=+sAegxx)Jk!A|O*t~}bkZPe%CPW=nv$xqTTWT*?&
z(D^ST*k*X4BVEYb&YT|oA(bLy+}pg55rc1J5(1lRSl1a2X9(O>yKUkmJ=8r}xW*o+
zL7sxSqas@-$>v>+8XyJ86W%deyPcarl>$Qt_t0_%VlSe*{JQzHRd)fE>+Fu8;SR#J
z45Txr9u;4zzZx>@zZ)_;5Bq;?$hL8RTqE?5vuDIJG9|C9<Y*t&cj3LChYXjC!`0fT
zZYjhN;qYTP-v9z6lC!F)pQFv!@H?R61}pB`GD{v|0M8?f@U^b?6!&TC1N~{TUl^Vc
zUG5OGgd-PkgWJ-yPP2`tN9B)Q76mjf<cF;*3{(Vg$~Kh_?Ge^`^5O5VO=0~+Pct7c
zmlBrtXawU|XCp{<Q&6fiKkr7o`dbcbbyS|GINL#MmvOoV8`tVFYZ!J}oa$Ja!U-J*
zowhO?700hSI4Ni@-UakMX)k}So8DWJQO5CEFj-DXySm?jwHFVdy$P1{)2b=*U1wJp
zCIotG_P+<&O7OiMRYBqKgD%Q>RtOA!@P+Qk%porA(sU8sz_!L;0rSLucxKJ|HDzzP
zkmMQ}<{4R%@{gI98;$h2mp$_mFh%Ol|8`5)L?y;`-U-{%-bvn~-w8@49}m+#0bRph
z25#|O#&Unr!?2HW-oo>gJ3d`Pm-rQ^@xeYEYbWqjBy(_slw)wiFP#5=pe45Jt@{$u
zI)~<qJ<psI$>yhS$5s)ck1hmJFJfODk7<pJK<X5|`<f&BaxH@3L7V1c$5xJOG{D`S
zc?Wya4vN!NUnvuN;3{bY$d18*Od93x#r6=cWM!P;^<JT8T6^;a8b+2V^sn~A_V4z?
z$;0`d+Yk6bxW$a#wx{lamaqifM@kWZNVbUZgWE<s!KzMOQkDx-O7UIj>(dP(R;ii*
zx?INkNb3cX^WBjv-=u0o1J=#2#WVr;r1sjl<Ga)MtE<0tJuPkc__ehct@s6ms^;72
z8{7=;7OB8a@A<~Bw0eF$I=*TE75>xLhJZl9<JZ;K^msXCB*8`f!)?U|EYwSH0V3jq
z0q0%eTWpi~>w(+7X6wvkj~x)7!K#|(yaCXey-nJ1`s+%zUCC>4=t(vsW~{0!kIwqO
zCg-VOOzR6jTK!^eD5DnR;`JzTM(ArCyWp4Ur1omxRY2ua#nj5=trIUXg9$T9nObQz
zy!zzmTdVb_v{&x!$g5Y+YT2ol^X8%Q>b>dm+L!W|%E{O(Y;BxAx@y~-6IrLsY9Zjt
zjN=0lTHba7H1ZY)q&T;5R%)0)41{pRIV}}@B}5j&FK&n#^d|WC{@r~_9v@u&zclNh
zTd_h&CWSy0XHTdFbTvUSi+aiG-XnJ#RT{hekb)vcz$-GRFtBi${LT#O*AWle1VS;R
z_2B0cVud&#62*jWB3qg8BPU5g9CCtMZZ5pcuP&O0{aRiRk=2oM9e>5d=FlgRWRl+|
z;J{ss3|RNBL7|{9h8q<-gC`gs-~{KU!fwLu;`Hu?7@dJ<-#)en$*gk?)Jd*$Ne~}A
z;RHRk+Zo{&2vVV14K2jf?YM#hDq8U|9}_NER@74N61G;hVYliA!WCR8W-}fA)b|OM
znJ8Jjb?)&zM2N`z3ua_!EF_DPR3;L4GrqM*7om;^j90ip&95AF30Px1PqE(Tsh-P>
z2kP^p0uCtd^QJeF=DL{D;X8oHfE2NFdMvJv{o$WX2yZsW*~?QjOulEg2n_UESQ_ZA
z*7i@wGwCs6LRvxFyM4<t7%%XSC*OMiN&@@8lfcFHpC?~!RsWFizV&~R;3f&B8f3C3
zx>aebT0E;q#<D6IlZB4@a(8q=?y-0aK_C>JX@1<b^5UVZpt5Q^O5w9}JPK-d)=!bT
zb#_KRJ-saFC6(plCsa=J0tpC!Nhp~!1Cv0fUpIPgFj!K*lxfXY-A-+JHCdYx=)3UJ
z@*q%z+O`pQ(zFO3@U^+o&egN#)ShNm-*0C%!th}Fs61V#Aj|JGW}9tsRLa-cLz}GU
zQh88lGhgfKhtX3XgY$0iFrR>t{7#qn*wEnO@rpgUr+C%%b87WLS~@?<OM&t|x~%@=
z(zV=E*QtH{V#>{f#|mMg0lx;Lrmb6j_RVW64<6^s*ha$Tt0%i`bJYh`RrR!KN97d2
zG$8rIZ=b=bY8|-zvbQuDi#|6lB`lfW&OX6(T5v_k42dj=D@NL#CWS@J4Z)SQ+&=^R
zJ`*5)$NG(k18*5PNFhqjI3b!Hs-@#z)hrG>=%b?YvUrVq!mne1FJVmUiTtSu6_%f(
zG{8IvX(P;Sqm#vOey_Sl8!25$az0MBmqifI9MIayeYcIgy+!)^i`L$H17YuT5H}Je
z+ARf^ft1cZ))@3Wo59{^v#4$UUqo84y5G9Vz6Tv5ddOQ74|r}jZBG!tKmQ2g!szPw
zpr5xLS1{s;xFu<Dxp|uqXh-;%muo!|fYp6YtfT-dFDW4EAi_(IA}~n_#T^&Mjluo(
zGa#FcZV=b2A9Ozvv!enpL<WcwL^yD9FXBGv^`zT$&dD8*MD(?r8QBvT6Xc@1ZTcEM
z9;qbA6;b^&OG3Am4{&;w8?nq^bjhiZy1py*y~1<q(mKOs|0>s9P=WW-bE;}QPTnWu
zdSy?F$tR;0ys)$<_2;KM&#B<?I7Z(L*3+wagv<ijM}E_qqx2^I_NTmU+ipHD-<x9R
z^=0@;f%?<z9<2G=-A5AA7IL+wXzyPG9VkG(p$(>st3t)ULc{Uz(C{$-v#r{P5C8Y_
z;U8$0dC^;s)GNW|!!8aX3MEEl=v)(Mo@;11@K!WYy*mhDrlR#|4|mZu)@imsM1ivo
zO?-SrA?}#~!7-l>zbLjVu?YQ(Mj-`V5XX9b=+<fZs{5WUk;1z1v9X`7003L4!1oSM
zVXNKKg0z40nuk&dxVqdFJzp9fC*nig4+<3AdZm5kJ8~Ovr!*hm+dN?hvW4#yY+fhE
z3hLk@y!Q(X<b1&@*sdh)K)ex?PVv84;CdqN=f=50!e;4+^~Uc9G@oOP8>rJTJZAIh
zw(mDAQwmf*T)cQhw1i%m>VHZydZlL|QJCg)eAXH9ll29KT&S+RNii@8WT1sMydx07
zDM{GhmMKo&eQRG8V|jDv*B+aGB14p0kMIO<ra*$W%qYKn!GR+~t8Ub^SQ^27>sMwu
zpq-&#-sC!zAon=w^|dS_Z;Dne#D58j22urP0n;CL1Hz*Nasv`(?I&bbhLz`c_)54+
zvQEf}8Hj{BEj);sPSy<wMS+j`S|)fi%sM^aB#M-DvK!)Q856Nt>Bc4s4atI3TVZ=Y
zh1AR}OP*I|8PLO7K}_^8LNszfz{`(S3GZ)J?oX)6Fko#!dSp^wC$F%frBY^nuGF0N
z_sSkig`0UNXicC<URsk6c|(xGVau$ZW1pJe)1F~fR$j33(vZ#=GNF|nr9z<r!Bj+R
zN4ra|{yr(lo@O`BzEeqS4GEt@=BERDY?M|#gVB~w8B6|3rSuNMuHr|Diys;@UrPfm
z(V4j6o2>`+{JXUrfxKGIz_wJLc$2KuGLZ+QDB5*r*$0L*Sgy8RkAUQH>43EC_9=r3
z!;HZ;TC==CijcZkNS2gFM3cz~<6==Qe9;H~92l-1Ht8YrE3Xl|`&UW4UzI1=n<JV5
zeuJraJ1U?G!{CUsivCa7`Yi+?hRh^rg(Ge_u3=x|!(|!bnc9^kS(_$%nTym*`XTc&
z>5(nQb|`oWEl=pH=I?6f(Z*%$+>NnkFs(Za>Mjquqr8d$gXm2l(M46FlU9@?qunTr
z?f}ZbhVt7kK2s6ux?;l1nd@x_*Wz3I$m^ZzZAZi6+nVXy6YrDIuj99@>bGxWuV_?H
zLh*}lWF~K+R8Ji&o{9Obqmp13kmI#!o3^=)aMd$)Do*4KyY0OPc?io|Ms3flzwWh!
z{pQU@-+Z*(g;2T1A<18<{##Q3`+r{e_(L@~sA&6FzV(DG5@oX_;v@b$CDdg#+qxxd
zM0qYHC=7}}PGK7U(rPh$?A}FM3Ktr)x)4dGdHsq0>Zj(}!@<ZGQ;U_-=OzJIb#>Ks
zov3oYb-wv#w<Y4Une7R#@zc`sVY#}4+dV>sHJ8!Nspjw-fg%U{r_;DV!NbSCmy9Me
zH}0sM$yYQal-uD~pYzf)gdvI9q{5%I`gVAC;FA7G+1wKRoUZn*ge)tZ^Vgi+xc1d{
zYxp0zdSuMZ*;aR>nrhd}L~WFk008OuY2$&LZ>3K0L)<#icn@DuE5a_C=uwF8U4D6o
z#g+GX()7Vbmgs>&ol@EIL+skOr<2K|hdX0HjNY1pd~hviY3YGBS4Mp#Vf|_vQleJH
zOm(?>z|j(ux44p?JC^%NQ}eA*cGRV6CA%ai8LVu$Fi^NV;1|3IG|+#J<VkqfQOEhi
zRR9byp6&<+NCf1U=*%^U-&b-(iT-8^E8a(ojb}{|h;Pko#cXBf0o=@|z3gLGy)x$u
zgKX!RswLC;<j<0bIt;`8QOEF0+lMGIePZWUD0w{Xjd*9}4d8a~I8OsyJyB?M^G16w
zEsq|A`Z0x7YM3(yS84VK5T-=j%dr*=A<BGoR+qyIMOU!vLsR~fhHrV%*dys{Z_!(H
z0B(1S1IARk88O_-7q))>q7`e8_#gBA+DG7GWJOI3ORXt}@ys|5aK3Y=j*O`~h~?0n
z)_BW=Ux?jg3-*SRX#}d>1lHJMOuGM^rUXyfH+8+&gQ_}}#bvitNPHc1GqAB&nwgOe
zKADKCc@1*X^b1^cp3B6{yrBkHL1@|cO4<4Z#x%{pG?lXb)S0u~uQSYKFss^PGppTl
zM*bv-CE+8^L@&7ARJ+CYsC9u4(7H(B2LIZy#dfWIfzMlZ4_qv}M_w<zm%ZsT-v=+d
z;DE0LANRrc$N(}yQ3WzV<#O;<NEa#hTWl9>g81)-rM_t*UXa$j-xA8`5qU01eJ<B1
z9z2u!?Aq(+KzEAW6B@DY7=$DdcZ{*yR-A|xAx}Znd+*>wbUA8@_xzxskIOsjm~qB;
zcLB+JLv1oCreA(w9X=~c{0!?>x1j!4IRDlpz|Q)gM{MAZoqxfJy(XG>8`B>Gk^e6^
z(f&7_(0|~3c^s8pIw^*qxVj%5NrO-RPKu_$HjlytNYt^u*)P~+uLr2TtPuVD{@u$$
z!F!$bto6*teJ$N?riZ&$;kA&lWcjgvowx$(q27L7K3%A$9bA)io%d6kkkA!;T4!4y
zTf3b$=^@tm_T0+X4m;KGL<ciFpY=0W9{_R_^~)Klo8yI<^HoRj2C|4b|M<CU?zpVG
z9{E$EO<32&X!c|&r*wbq^c$%pz$-Hf;J0iaYSE-}HBiYzJm5a%fjxRQFc#hNs;Ub}
zz^TuW5A8_{Mb68am39RX(uF*nae7zmu8-Y!c>;tyu&0+Q%iZ>$T^4EqE7#y;u~nPW
zaK5Hk``PO<TkK`Efb)DnNt(lVSEbUQd7rDpX$|OLlwep<Dag<X4Pa_fxnXQNEqfpd
zq44o|ROj&VVDZEP{0F-VC-Qa)ASH$gfT_w>%yaB8Y;y+e2J9p6@Osh<XmIw5mW0y2
zBEKv90ZtY>K?18>bWA%DHVMcNVI7Gc597Bv9Qqd^_dsdT+EG{?8!K=Njzofou}_!<
zTOJ5HHF1TOm}6oql_9YX@<ejCHP#3=>^edLa6LCQ$hdQ}d#M@5FHQNpM}GI_@>d*i
zddO#6VzLUUu6PcA(#boQrJFw}E95FU4SMkOmXld(7r{I(!)|jhd)od&N@EfC{gY4r
z{gB-ifjg7b%8YbxB-1}(L*IY?YYtc>IBKudidAojoDQx@PBwsZ%J;BG1@495lk4}T
zLOD8fnTEH4H+#M2FZdUl=BT5vM<W+NBycX7@ko{?$(7*IX2CpE6Lb&6PrnC2eIM_|
z$ebJwANOENzX!I<`-Ei5`%p!JNlU*6w+i@xTLlD--KYh7xTpn{eJ}7I!OQ7+;4AaM
zT+#}D(~lURP>vXnrx!H7Rq~-~zX$gUP(vH?STf7dz8WVgJr8(mM7*iQbzn<yJn+#L
zMcso-lx<4)iq8CU4x1LS0^P0ZTuvgB;X&W<IwGUl+65wwxnB>~MGAPwu5{0Y079Xg
z{-2unJ6vDC8o922@q_f?pZ{ye0M}nS2H08uix^T#&c1{hz4b}GF;wy(z(`hV36YGF
z;)m0IXlmgh_AzB03>-f9&icG{DE|#Dyn&m$S&FW=+nwV(wQ2^kFjBmipmyR&{||l)
zysT-qS}dGdp<ECMPQ|xO*FCjc5YXv+PosJzn`^OmS~EF8;jU@5A#qc0H20*uLf!6)
zuCB^(zY7mPANZ5q6^1N1*(DT>P73iHQs^6a%^u<s2r&2C7?qf=adL|)bbm)9kRPMV
z%WF!`rqm+GC{POD#B6G?IA&G!;l6`r_tveWkh8qE{QhDvHnu5%upSAMa~(eR#$A)p
zv*5~+g)}6Y_!S|+*23J6o*-+p46-7Ay9s(om&jB7XTpSgFog0xR!n>qFuz4tTzL4+
zYJ`4hC^V6|%YrANSHVzns0zXOR&;7W;yaHyQZLfT0DWi@s?b1~bM!n8^@y`#+84pM
zAnsew9qRA_1f$~~@+g5g>hblDfLo%O@=9v+5+cW~=P$oB_0Rjad^UZJ#u+3SqLWfG
zr{sc>x_Vm5wSfH#$_~`W_20c!3CfeQtoc-$239i;%8jnu$vf8Jt_lD8X#Pu`nvLVX
zcr?dW*?j{iW8H=NT*Km#VMLcGMI<e!B$4HQqO0n52`b4LX!j#~J}oikS|vbkx4iK^
zuWv3+7XdkSX7S1r-;XVnbTI?#&l0F=7Uiq5MJ>_7h#qopt-100q-&Yiaau$WTO1t7
zH2EArg##1xdcn_P&pgZ>8|E|n^(9AzBwwt9frX(=-Swzmh2Lbl*C~R2qOrFrI-v~4
zDjj0l$J(pzE$>V5YXx|)&c7QR-G{e2i=v=)=sLy=PH?p&X>V-30sjYQgb9AHIrZ(_
zO5gx&cyve(#wFra#~ernRjXm&7J||fm5A_8Z2m3#5Pd6DU@c`>jy#u49^k{$S2d3<
zsys*^3!EaVei5Xmc@VNKElk7kl_z?vc0|gvJENy!NLRd7#gENn7d{oMR>iz6Df)o^
z$}E)XO=`p3Ux&Ag3-TTp+4Akxved`UIJeIq-pJX`@PQsMpOyOC9Ru$G&iujN$O@64
zACXDc)Xv<+;seKjbH|kbx0vQ$3mbpf0E=mWR4!3{S*k_dJPeg)u0c$I=qo^gZ9-L@
zCx}0Kz15Sg-dso!6`;jwex74aE0}E!r|n<im!W0PIh<xx*mZ3WK^Z@7cSk2^qfC^A
z>)bYemw%W^<Y^zVtk~V_B|EgCmEs3(DcF^`Kha;(_AP(DQe1r;y2l(GDDL2y>hawS
z-11gboMNifolRE9;FqD*HL#XmommjyV-Tk|Vo7OH$*;{Jbrg5kLyxGYdvp|A|LJzs
zxzp2jl)aWFvt2fCbaBAhvbHw39oOK|K)Eoa)A!uM(cp@=bPH!8DrsMXL8$}!dGvOo
z@=b&P_~N^BYr_W-E%@US3+eqOWyo8T@{O{*m9QJThKtVlw-YgtV>@t$!eNHOg%sbk
zs~?sX*#b3}G3#Tfsw^b&kKK6UBB`=_=jt`y-e#p{p-}R~FbpPyGey5}Ze-Iy?raZf
z4GcZ8Fy(HZi^U;2XYRxii5KpTdnBXs1S7i<dF4{Vns!u-8~|RB20dVNg<%W(>MOM&
zaT$8eo=tc%hkId4rI~!PfVl>j;V95alTDj!SC4&8&M6gblwy&UdH};&pWsPpMLh)Y
zKJ$A&Phl0~Ax(M7SN%xWD%?I-j^9qJ$PcuKp{%p0<_P1Y=GBxh%u3ruQ?y&@U4*Z$
z#NWHL+?5aTCt!O^)ba1VXl<vBa9;Mq<?IR6&-l)mN-UUj{%U!DsX(&-H!+sdupL+`
zR-fos#ByCev*hQZkx__X(ta15`J(PFb4)gmiDzQ)qoad_P^oSRikr>ZjQdJs#g`;d
zr=I%V7QA>zmc0b|2b#5^s82X)YCx3zO=96dP|gee3ws%Nr*Kjqpl05xv0<kT=Z416
zY?Z2Pv-Fw^_DwWN;+{}NcQtu~z1T~U*Wi~xadtgO+tAAQ8XP>t3;mG_^*>TkYFuJE
z$+-ZQ3MqILbJK*WDOsa^guEYN%0<1*nx_l|e8{qbI**X4ZRJ|A`RvctT>xg>g7b_Z
zQt`mAkFrVN(Ds+%aWjD;tSQBCVbBmxZCghY%EY<aH}#>5DahYrQLS{7()%q+{<Jsv
zl0zEe)2oc+S-&gruJ8+n(h`|P0Vv^qd1SN$AAyT#Qlz8|yPz`kiV4y^R3V;cOmaqk
zJ#7&4c-IQW<*Rvb)g8Hz6u&&WF)@i_X$rshD|l0xJp;Qa4xVA8-tcKGGIN2=$iBXz
zqux})GjliOgGBI^LS$}jJ5^T-O<#AZH+cuLHv=n<&~sSalur>?;2dtHdu|k)>w7-k
zH6o>*;g>}dKcHm~JT?5)82(bl`~TTv$d>+%U!|d)wf)Ow)uDaVwS#_R);-Dl_-iB^
zb&g$Cm~ka(+=WynE{^(1&+n5F8nU~L3A88s5}?U++@PPO5dfXZhQd{T<_Cu}ufn_C
zDUYK%4WH)G@&@<I`yGZgTZDR6b(S}~QCs4W`ZmZV^r>Z=F9^5=`;UHnvmbEKyq~v-
z&KSaC<{+QwQZbuvRs01~=sm1yxs2MDZTXph`*~xSgP6!U5QsPsRwcwWD;)!np;MuY
zkP3epVk-+v<nowGSSiuJf=U}utIFd~$2Y?$je2Fe`DXM#qKc+`kf+JOR0~qa*ot?M
z_o=TQB8sBC`YMFV{Ti5TotI2lGyVRkzqCTMw~!Q$Y%W5d+Z{Y?#%D%KLxR5dm`D4g
z*(k9URcQ*A**N~3oCX`%Lmp8JYr>6mjRWn9vPspC3_A9%B=xtyze~s_?c4k5h-y{(
zoj2mhYDV2y_gKtPSFTBiOB4|Cm@Qk^quG5mkYXbx1f20Gc8TyXKM(QK9Wbh#TUGIb
zbAh}#RE2BTlU}Y(78CL^nojGVeHwd0YE);w`m5#rrT)jl%=$k*%YTr-RRbq*|GfX9
zZh;0l_02kVD^%sr`yX)N-a3oGK0V@aDU{x7vgOd{r_0fWU9%tGA`2#cTJfVYT<+$_
z2U19H#~sytv42;1tYj7ljxBX?{mI}i-~E%pCE9Xm^L|!)1>L?F^C#S$h~6>VDd@*u
z^(=IxhkD#RuF&?*pCs!l7PR_Byp6dHjLJtWUU@#bEk&_G?iHLIF!1<#b;$N#=2Yco
z`E9wZoDSIMJ<|nAhRkx;J`ped$>AnA0VwaPi9-*%3w+>OM2ejjZVQI7U+!Ai(O`ty
zNfjLLnA?xP_x%LDNpnRP9HestXm^G3$-o)ey4E*IhE*?qQdD~kcImI5l76G>EA`L3
z-HkFPUMYCDKJd4Z8Wrtpu~q<tdbbUy2Q1mz`;EwYt944}hd)O3yq(f^k&QE{uQNVG
zs6HwWaSa=h1^gy}!~eoc1FQ3+3m%WaK!GelJ_PZGPHL{d(;xg=txAJ5915BoNwIt{
z4%`CG?KyMjvc=HXoQTiZJ*rUnT6oDtP%Pgayo*Fa2;htgw1PB(mW8FX(QDo{%z&zC
zj)X@rs-)|gKZIH@Kpdh%JL%MamK2tZ@iby^D6zwmQSOLRk63PfT2DK0VyBhQ=kDSm
zP{H6wjOK)?Gg**eqiTVzFFFY#Hbz1o8Y-jT2w{1cI0x_U{>1=T6qb0jMFcb6J<sjg
zWe(JeZIs%gYN|t(VDLoa-APJyRw^|FLP`ziWN8ezT~;hcHCG$*am(x;Ah0PYyfQpx
zNi4F!!T0gEpeB2Ib*DawbVS_eeumKQ|5ZU`kLOgJnuPR>J*x`Grp`|)>d6}=#uSm%
z8IY#sOtR!wezc+QsLPuaap(namtag<t*ra?>nEm0$cDpi$Orn}|A)P|jB#vPwg!ut
znVGq&i<zN|8M~O7nVFfHnVFfni@A%LnQQ8t`{sUEqoenIPntg?jb&N6x6{sT=W^t7
z#)^pEy2(DyZPX)>obZ|HzUq@mDT2wj`F!-L4Ip!~Lo^1VsULZ(=7+|8I?I_Mwgai2
z+((1f$I%ztm8y|_>NS`F4Ldw+p)94;f#Bc+9lw7APj&Rg@Lkmk%M8F*3d96^B#xR<
zniOptyu>M&TC|NspKlOflyu#P;?`~$L}VtLRE$kGEZ!GVA4;SemzuR~lJXrMviFco
zzEImbq`U{3sYEl6-=slh2JN-`IflHMGvvJTy`6*F<*Ly)sJJFI^{c8=cTQ{UNH0NM
z)vG*nS_IkDtGMPg^}DE3zn>PPk{A_g)qheLdS{7{l+qC8jT0d`qQ&Xfhk|5?khFfE
z_)Z*=PD9jZf(S_%A#U0b8b}u*u4aS?c}7E2HA*z6;&xosjVo%6aV67vVpDVQ$t`^n
zC7lC&Pc%&J$~C->KA`8rkMeP>xc6!mP=SB83%?1+2J?^h<R84KtgQcot>gcPp1@hK
zmO2mxDfWNq2^1vq|MH@8*__W>Hz5hz4F9;>isZBYR;`n%FXC;Gbf@eNWvxr<v`c(C
z{1IVe6VcR`sm8?1s?cJo&c+{f_>;`V7P{`KX_?{PTCi=U*sVjwCi=l=da!kR_ydG{
z_i^zf+i~-)?;iP8Vz6szfUC+!2gxv^f|mV0<!r`jog3m?JVi3nUUvHkvUy~R)jH((
zog6fy93E&8(4MGBL^(yM_E|IA6hM5-D-+yhjdg@ogJ$qP2yI5_t8e@=|3p!zt=oEQ
z@Io2Y%>}(42L%F?HYML)j7qeN_fl3ys)x<#m-okU=J#$(O|d1l%jILy+0#@;1eyZf
zkbZ-@Xmwe4RavY(dvqP;&^QM#mAs^Vr61dQD*C=$aBcnlZ%*maJb7)0Y!WFr;Yrpl
zz0o+Teg`@dMHH4$R^8vcL&Jeq{cuSGeh_EGN>b4<EcKZdhNW>;UaIG64h{zBVYnSs
zU*{gDfNcusXdkBZ#6^7THzlz8Cn6-rQq!JzYstfVsI?F5z}jw+yhhWC@>mA(=QBsW
zlQsIIDw8z^k3KsxdgMCixw2EA4*~^0KkgYhSe`Qfl0F)KCO!QqSr+LKZ{_b|Kia4z
zjU4b6NP%+z_o3nu^%>rS!HV2X$GT%!8_~rJ_E12>l3I~7{qdt`{doRa{V?){WDEmK
zYPhX(q_Ptq3&y4+-A?+r1kvGg?P;_Vdl=GjtvARO`!NbW>7nf$UMn>QR;$zi7=1dE
z_I;z*D9s|-L8=vQdmM_dg(#Ds*RIvn6{dQ*rSqO=)6$#qd1<$MDXN2)>eo79#Me7k
z)D>=JiS#}CGLjzuKLY(9JfEzr{{!KbQlItT!kND`=@)+bT!bsAf>4m+IL`mKaONB6
zAaEaVS!~WfT{@72XXyU|`o)H)$C`9~IZvyEC;d<Z_S)Y-_xd-`IlgSG?H8Y&T@hrx
zG%q*QHxAeJ*B6espte{K){E>77QO;hE}N{Y31<;o=r5NG&$i&2PRQ>QAUw-vY*4Nb
zvIqpVAp83Ge(^0>-4QHwCB}z3DIQ!vE`^w4?GHIJJN|`qyg|U|zma}ayKCZ>k4fP6
z#GAQkwvMxIT8A)i;?9xz=JM!_E{*>-{@`>+WnI%8y7kNSL|0(cv<)SROM&I|Y2<T%
z@Lo~}|GSb17ddlH(vxCy5_Y@S$*b4hX?{?mWZjBoa-2oE&({BfI^H+biI#e?$qqpg
z#}QPL@ia;x5wQ+fjwJep3G|z90ttzf1CO-Ciz&<zta9_c!lNM910j<F_yZphsmO?g
zqG*n=FwG4u{-7dTSX-YH)fuQBTzSdz5U75qu))8uD!e~(^86*F=8g=>neZbgq`BzE
z71WYv+E*2?KsN18DRDTJ<n9G0^+6eCJ^aCG@}2ww<!pbR^I5T>hr?B=EQgn@<WSoI
zHj^>AvgelD!d<5^A`p5Z!eJU#v!DGO-SdRO<+x<d;NFCLdb^LJx{MLNzzgP3ua(?r
zr!NpTi0#K+M-*>Vj?r|gY<%QSpJC>N_vHB9wXW;>Vx&bsUsS5Ya<|yhCWFlOJQ(fZ
zdaaFTY2}O&uc;m{t7%>|tOmP7^R+`jvQd@#G~*5@KQ;CHMz}4|%ivD_j>EN`&xGWk
z`NcyUus<3b%k6gfIZ4$Qu|qzecv1K~m3VBw@e3CXr2i4=|KJV$f61hyHm*@iOR@d8
zCcV4;FQi9swWM@ACcGczMOa&hH+N=geTSK;wwY`3@P!_A)4EtAw7fL0us+y|^{p3s
z_9$6LzWU6LcP&ntK)4QGZ2e`@J#1bfKS*{TAFgH2{-l7c?wUixeH^?xY_h@&Zc0*)
z)mfK!@(4U}q=Q*A;x6Kx6`uwG+(2=}&E=O*5URb?$}$6xT=7l^a9L#-VAG=MzY0bi
z<$dWIx%|E_tI^V7v)O$t5AW=NR)>QMj7*l4ZzD!6(8h-+Cole!(N1^8$4Sn-K3+k&
zKAzPfbn+^6claMTx3Qoq?8F;cNz^~U!A~kB<@m<A^CYxAg^-&1t7#V5{wzfehb$5q
zIFT{-6~n$b>JA5bJY_hRFm|nJj=n)Zt6s?X&cA>z7L27n&BC-Wy3|eWT-nxE7cm(9
zE6E3r=fPL23_8NQF(qy;=i+%DtnU7sC6Jl$!dpr1-9W9lX9Cu6i{>yLl~F`7N<1Dp
z=pL#v8d4uFGrslMmeeNDJI|IFf4%0)7kR&?W~O^ge>*7Dn7+o}(&o?df01bAYG*oF
zuOJQU_2f#1v<2`b;T7=h*@3}|+DS*bVO$t|3(|Nfp<&4^DH_w#{#f0cz0)`hd&V2W
zz>@80Eg7h6!NUf({F-VdyPpr`aJld>+=AT$;k4Qo?1X(A4jF&ddIYDN908+SsPlt1
znMU)r-mafyncyJP0=G2^%F#rWA;@9ZV&(>0KGW2G&9-Lk!SJ-O(>WXdh5s+0dq)jF
z!7a^`y+ZTz>&5;f(Eq`+_CGc0|LIv{4O~%13F9X^sI^*Ds^}`-c8tT~kHvZY<}Q}$
z&K(N=+4z5%bgOD_?#plP(*14HsfQA<*Bt(fyQr#)XlBP$b8K>b1KpErBW&aw=$9nb
zcUE;)r}MC&HW{}UT5XJ{Ujo!m8f+;D$DwB?PabnlHeObLAsr~^Y#tlb(}O$|0bS7c
zntmDHX?+9IiMohHe^bfLBk+MR9n6gp7Z&&Q8|mOTKw^I*T~Y0(fqM=nw$lS&=Caia
z&Z<!@^aQyJN9Kdeqa&;s-s|AC-3@gaT~qjKy3w(gz_3v(d>pqF>%pUkkCX9ZX?f)B
z=xG{q%HrT7(dr=BTBqInPiMO+;ei4TYlhK&`e}~KH`H;4|AspJH`Ea=wqcR){2`1W
zs3qa6<^v(3Zqx5b^oS4`*IoM&{)4>8{TF!&92X$ye}zCrj>{8Gc8L0BFiltCVJvKH
z&T%Vs6}L0pC3vyaPI|v%9$6J$?>M_F32C~*K(fZtW`x$~Tswi9@l1NDV}6%Rx>5Z!
zkcxltgqwV&3bq=0<uv|Ga1M96HN$-`TUCX_Q$8(;87pgB)&e$(Bt5I)o=M+AzaZKH
zdf3f$7)WM-djr+soWy-CyT|yzjQQcp5N=^UIbez(;D&K2q3+6{C$uv)`EFexyJwR2
z<1m55z?A{>s2=~jGy^xrmNUy<`n4?a;YQO{oD*9l@^hmA^bPAdHa^vvLmIrsH+Rt}
zH3&fao4e#1ssD?+%>D;=dB*YF_;b7Q<k<9*(bt0Z)MULeVp3$mWq6a*Beo|7UlH<6
zUf}Kwvi<?-46OgT`KaV(YeYaVt8cF8Xaz+tL%_(u@Ynx-dk4qwCewHO-}~b1|3z>9
z_uB#&S|$$bqAj1*I({)@4{8w?5ouD!G;Q7Smg+;!nXZ@oBs4d{v@~=t4_6!}@@N2n
zarFw;WW;)H0L_`V=akT~RIGL=VpOb5@ko5ig9bsU$aZFR`o!6@RU}CL{6oWW;>;?-
zBp;})HMlC<_We(}`b3hz8tCdmR!A5eXmSoX=1AbSKxB+c0d7YALhQJ9LL}^|s|c{)
zE>o(`<UzRxFk|uDnt~3<dnYJz;BsM0_9d95<LaRFA}oen+Q@zbJV@|F0a&$AgLW{a
zUTdhA6k7v$bIy2;ITi|#STGih@oh_P=eX#gAq^xm)V*JLVrxI6S?~aF*qPAd7)Lpd
zn3$&Z;Hw3ARr+yoNDe&Vu{i)?1gAt~^5i&V^GT9jrLQR@aOP-8MrGOCA^O8vnEk25
zBMR1Xq4E!dUH$y>5RW`I{EPDtM{$MCYRhXgKz71}1x%nZwiN~!2>PhR>aZ(;BIFja
zDu7Pn;aJGauwlTc;#c@}S?YR|{lk(-ss+@vPB0`?8T^TKZ8H@}v4Vmd%=+Z%u-eUN
z%b+$J6)7;KQ)}5%Mdcubp_XT-)SzwmaviJ<<^b#gA^MtqX)Zygh7pN}g#w6kA;Wwv
zO3fAc7*g4z)dKt6;sFw2k0l2~jSw9IlV^=*d>ch7`2j#!XrxH&qyYv=&9TfEurUyD
zAX$3@x0!|!cj&|g1f=<=0FYoaA(7gUSLhI(W2n!G5W>OG!VrMclyeOT<p4$XU<-lw
zbcGZu3fJM#sS&6RKxu$Z6d<-?50j`wR{@k2xVoPP;1RiERvK1^X!K(Wu98=ES<<DS
zjOuNAV!Yq&t~zuIS!`EycG-Cc^@E2{A_&olX;h4=+B5VShxK2){%}|G%qY?<9f${1
zQ@(=BtO557?H0>82M^KLg_{Y}9Fo2;J1k`z*(OY@uIlRyU2xFVswE0ZP3UovvQSwp
z6Vl>O_?4cJbh`Ksk)qAivR|PEhCjJwjNZN0vS2Eif2g6UOGv7yN~+YWI@YvOR&OLG
zcwO!hcBU)|sA%!hi=;Lizi3ShI)1KEBi*1!7%Gn!vCb40>ZxsEKk*xCeR~_)tmkf-
zJB;#t8TS7uVobExS0pu5=D?jU9jd6)vCaIFRsi0~7g1$dY=yHlXE{f0dQ}wpXro(6
zpG9}V_D~gaT$-r9WQegcda61X@H&;qx*f0dzF=PXYM5bj4QPlDw*q<l*qR}W=99zx
z0EXY5REI2DLg;A+F<)Bqg4iqzR7<`R%GICjo|?FRY)Jv?CxtECSauClcpHzQ80;?H
z9H{8asy)Y7H$Xu>3QDlo(Z3F|haCSURyDTZ8%Hrt+qxegI?4ytjTB@+$~e7QRFPjr
z?M~FlLppas&T+t5r<QK0gHAjIw+4+Rq@GSFZ>vfT=YnhcyQOw6SNL76yPA2r;-T9w
zl(l#+dC%P|sQ!8m1y0O2?G|f}#(oe1mO1`#MTjz=q*SOk$|X*<ohlShV2>JWMyF0<
zvBCU8QX*3##6BL(EdQ)kiTzZ98r;EEJ<y2~C7O_9Jf#|%Ru2u(4QbzE6gx|3E8IR0
zP0vE-Cvoi8%I-OTz{9ego1ncOfqY#oB3Y`@S|FSWLRl(-pm!$!+8F^~uP6L>t=19%
z1L!efA`}W3WRa*aa9zN7T=I>O{slLsFf8q64#AJcxRe62h5J8fE`w1uFP{s$^rCo$
zj)qpp?KO&!Sc+Iiu_@hE^b-0P3iBA<Ff~Vyfv!v1U)}fad2^D>DH^lCbifRPi%3_L
z!3~=!Uf&1EMl0wCw6zm1=HfipNuS2Y_;5~`=mCjcE%~)891HWT#yh=uojJs|;X8w-
za82o)ku1H#EWUobpecp~#vX)R7dR(v1TdrsdbMZA7z)wfx^Q%Bs+y}oF(M@9X;EKc
zqP^7ah^@0w`AXU50y+Z7Smm>SH7p&lkOOAJU11nZAhoR6+JHqFytBOy80oyXX_^6>
zZE)r*H$QbdblzphQrYN5(H8VlF%uJ$8ZiHAjK%HkwRd!aTZj@}&a(q68I}hJQH~~v
zgTz9v8a*IwZ5;C@V;vSjDxhxMKIcD!YE0SYxmG~YmtayhEX|^jj%3^1u0=?1(?dUI
zGSdq-0%f&<Ya(EyMaqv5MPck;C3sAf91RoViny1AcTywl#$~@9+?n)cvrS|VCRW8K
zo`LeGw;fJ*Wrc%nW!tP3cift@kD?p0?*z)s#{^FytPL8(<zXmqhZz4%dl~);@I2O`
z+z3IjqQ<CPkH1jl)aW(3be46~H^O4op*^%<uQG0!-LbI@f2p!rMKq2jzXnuAFmIk)
zV6eMCG4_9)F%?>Zya)#E$hSYT;Lsp|6(F13%VJL&jsZ@F6vU`?*-jUU6y1y?Y=jh<
zh8$8hWXj4<T@`C$070N&M)W)UVdu>hz5+9}llZZLWC1FuXgIlUMEEtEx%>m^PtQ1)
zCrIFQn0aam<YY}R<qk5wr#K{?Qb!<Orm5$)K1mk!=%lZZBYm49>vN#ofs!tDUMV<l
zSYb(=Ee4Eu;%fkq|N2*=5F75e7<lQxE$=P+@9}L<^^3^;mz0pQI1n29aV$jruDt+f
z5wZYUdje=&(m*Dx5uJkWmt)?MYrj1Fo=HMsjxM@fImVw&e53Km-2U#EIUg1eUqrHd
zDIC1y%@A(muz{q{^N5OOyF~Q8A`z;>X>5&O;oX97?ycOJk`zY}N3o_V`ijjXvT>;w
zDVdMp(egNrnS}Y^#|dMCYj)a9I_b$ES?J8$UY*&was4Uux6Suyf2gPH=$i)WqKXP)
z6>d&c<JkhoXR&eHBxH`=jp(#iW%loOwSoYLx>rJ?wfNe6s7HL88?}Egve!tUzmac1
z#*qA}l*k7wy0EMv1C5kG5;arTHd7eVG^>vt{>4%@YNfQ+z(^#noX#9?uS(g4R(YNN
z)AmV5ffscrg7MyL_Rkv|3Y7FAU~rdmk}yeO`hE|4iPwMv>2ot61U9mOhM=6wF(U4z
z4G|zeP2q1O9RMJo(5tUhj7uS?y;^v-Aalm|2?0&t-v<CrgW08ih&6Aql>F}G(Ck6?
z6|Ma0ZGugd4{8g)`n2tGBY&S5@TDX!=iw5?D(Wh%AyuY~9t-65EC=v=3J{CshKo!k
zw{D=J?#0?)t|l)an089rarkgNIij33@|)ib7FcQ>(=&^^o`VSZYs2;R=cD*iRH<vb
zi1a4^DLD)J#KJ+r$YpFx;lEKo$wymDu`5nA5|-VYm-xX+ylbFXeEgMO;p5+CK9zG7
z3d&m@g&O?`j*$?*HGLKkeihY)<f#%c$WKmg*<ELXoxswO5Sc{Vrx~lEr8vZ5saX#k
zVoORFezxx#u<42l0zDOA>5`c^QLW`&_;ku|iQ;a9Cyu`Qt*0q`RZF5;aa79!$A3Yt
z-tB!r#w)!PyqF3wEbTfC*c;e_=R2>9sJWxjmo=GR+_U6Ri#)7Y)sLChDMoH}@Vq%Z
z{{04i)bOAmi#r{>dZPb$t;>ohSYYSg6`5$`Wy|n}^zn|cd?DmHFq0=A3cCXFXv#5K
zOzo-LG6FaMa%bu21zeJT#h8hF#j_u_Wk9cT;>zsG+=XW%-#PNV;O|QhqB?q{xG04A
zF<Q?JxiyD(8=%t|sh*q1!PVP6+6{qOjxr^GG|P;f6L%Q1Sq^^2M6Cu&uSQO-dUk5p
z|5@>;T@9~Vb?2N!o9MHQtDE`b!I~YdYdr*-e`|R)g5MJVdnaO<UiPa&K$Ko3qbN}P
zUZI~&@=U>ALi*?Uqr?f}(fTN*(FqhtfPO0QQ?EzBuR7=sSy^o!K0b@u^`UYAxx>nG
zunN0TQ>{8om)m2<-uwt5H2gYX=TW={@^ea$;p_(VbH4LX+y>;w_%8jDa~3XxBY;#%
zlwo@YtSZqxJ+MXPW(-;%5kSEBIT2s7a#~xa^3U|D@d>#nHZV&D?3rRY88S29?(+%V
zOc{Vce6bv+M?_&z+fHpmN-6N#&KJ&}vEsmK9l)GD%Qlh=DY5zQVB!So-h*Q-0j>aO
zfDOkuX*1{Ym$oDR4NEoO*Omv%Bw8^FCZK1gj7V;)iF4%1CLAxl3A^@Az=s-ZH1%PX
z9Fqr;>o6~4ofXwbt~VadNh}2U4*vb)+mnP58!QBliR$G&$2v^%=VYE85FO%fWSLzL
zkumJY<Sqj~LJcy4V<S#n_YD})h(34w8H{{JzUT&eSaw(wgIJ9GaE;Hh6+z4vhGg0s
zST^|PoyBK(v8+{4C;IH}Zg$VF?d$z9E<Bf1VpN_$VW6B{qNU1G-QsrJ$JxPOazbZX
zg3`AL=7ldAi>ADpUfC*Y6xB+pxsqpYU1RQe%)Of61!rp#w$7-`8Ii$@W9E3<H5ddT
zx_kHKP8xhQeQfwG0JaP$SvpNK*yyLw0KsqU*;8l-gN`arxYDYt4@9`|ViPay9~yvY
z<Z!b74HfbqB;1<HPHMJBUM%ov{2Y#UXxZ9PSSC&Lwz#N!=$>?QJEpPGcK78POEu*a
z3_rCi+`Y!p*tbj0C#W*uKJqPS&|4<#-8|hM-)}*ZC&nVG4}dVIg?4n*@~RFQ(iNN7
zhsHEA^V4~?P$h@Mr%4r87?{*^>H%jpk7SeeFgv(i-{<#;iFjAT*EK?>nh7aWhg8c2
zjdw^4lVv20gQ7dqlph!@gO;lM=4I-ZUn=e*8kT#9)KkU@1g6_U6Kv$w&Gw|`EcRI2
znXgJ&J<x1PVIfBqDrVr|X11~jJ&dXC0x!63<Fv3abUXuODz*iKflNJU$|Q~9@Vi_e
zZYhyka0Le*GgJ*A<;*UF(m1(i2(HHntglz&;o*`q)HcQ_b9lKrJ%eOk-b-XkU_9ew
zi)(;2HpiA9Rlu^`z)fQ)mpo?XUOXA!cH`3N&>Cj&azvQfBvjdWQzg+cPAnr&w);T_
z3g>)4yDJz>yo28j>+v>f?b<b>z*fy!$rX>Vb)zn2L=_>L$<7!90rzSw)3u`%S*NNM
zBRf554honOfneZ&uj(_8LD2&eE*l04?72hj5Q0?i1wUFs&Z2|Y<b2$RDmCB=THv-T
zm)#V(iNr7fD{E*5l)@rw9?w;cuZ34MZ0XXjf~KM$$!2%?ylyjpcsT>LU#>1}n+iwv
zU6t%c0J_XbweF(QScHvLKvJ|$N3EH`?+%^qvBgYFTEh+NM%5StY~nRDUr=n<JG721
z_N7P+USyK$O$Xr>=wCn@s?+zicIsVW!(8sa(8*@&Xm?grOeEB&aRtP6qkd-DqOSau
z-5TcCQQPv>Q%L{0p4aBc-6OH8{GABLrX=v?<E)`&#OCLvyH-cS{o}9v$xS;9^NQig
z29!G)AY!gA%(}bZESWkR04@l@J~+~1)S*4-4Lk0&{WC9?ekzbv(^&5MU4%hdo<ny>
zvN2n{)s7n_IuZS~_rKuZ>%DSy$lD(*+V3syt6q(OV@DaVg=jPA#bWpEgHviGOk?A9
zU=?fEV+|TGB@ptn7a{JFWAImTUZ;0_IP@bjFfq(f&Z&M0sYDu7e)yhGloIx25xoxV
zv0cf;;Nj$CmdumJyt!zh42v=i+o<e4?YF6H<`HFEBT}u|LMtr#L-$<ychC)GPf%Vo
zb{5fU_1K-YSYp4S4{copORvke+Q_A(Y>k=X;p^b>@J=5bSR<nIzvK-{OGRZgKM~TI
za1AB-Gl-HtMH~iKZH+nddfe|X-rNKv+=-Fn5@J}d&BGIjyKhknEtywHoY787%3iai
zN*}sSCW;9+h~U*Svu})f^144=-futm9jva;J8}o}n>ueC%?Nn<zQ6yrCK~YlQO#Fa
zOT>&n?8xCv2%Jp&mCQ5p)WkDVA+0&d`<MyGh&E`|>q%~_WeH}jCEWPcG8Js^kQ0+G
zKgD+U=IPbp;ic*BP9Agpczs@iPHr`nkH(p3F;*#$MUOp>>U5w_fM6(*)$M1%u!gva
z;|IrRrqTGrqbuQpm($1lbGLH(WYsd>*%oq#VhD(PIOjX1Pk58ovO{q?xbRWyDu&Zt
z>svL49y@NODkiWDfr3tS@{Z`MK6BUQs^A~9w|8?MJd?(VY&zRC%Z?P@yn}?Ad`}Zx
zg>GG!uC_|frxmW|<t_l@A^_Wysau#!yyne31zKtL8=-reRn3+|3G^4RE^_Z*NIeJ9
zU8`e&JsvLh?)S~GuOUZ!j_t2H*5~*zU*>UtxIR5-T-XM?H|_tJ*IwNd44<Rec6vD8
zzc%JAY0|QKOZe`wwYr(siSX3Mv$$+zRR#G`9030@pM1r=1q*TxuGg$!l+%{9>S|v@
z;N(ez>xGl95%7$S>W$Bat2erDAP65pWZQBVRY03eY$mHUYlEb%R3c6E-HM7#dd#9p
zN7tgu%rvnHi^W7jus?Yz3K9?`4G>{?oCr-8(fe(`j)pmHj%}TE1F~$7z6tIPo*$Ne
z=Eu#c#kaU-QOrE0RqPI0G-q&fe@#9iIHKAn@&3E6w`V6N?7_`{gAomH`i+Y()|ax=
zXZX(~lozx6p?E2V7oq3)MufnkQZ~%A*5%msMfrw0zp@hC-hc`~qQo=A%Rq82tf4Ji
zuVDBsI`AUje!EzyDCE_E`NcsHRHM9p4{X)q8=Q2bb%|7w!;2vMrFt+9j9YXiO=`5(
zqzp0nDo5euxs}(d)joh$SOt48K=&?S&t;I7rC3f0RXy04+_`_Tkn9kXC)_~wjm5fC
z+{;OBqvc251yi3`RgH>cIC@_QZs7#X)qRhZ>Rt0EIpXihhLVCI`=HM#=kb{N_dtcY
zmV%vA;}f5PI4pFn*PO2cN%V#ZN?m*L>H=bpvx2s-ix#xSdWj}^!o>|0r9B%>YmuhS
ztkcVs6%CX#vY~o96Xfw#E3vw_1WFm8@S8%UI#c9HHiXP`nQK~lzoyb=>2i^h)V7VE
zldPZ6rI(2;pQER&_066x<ry`+O`}rjhQlm7r?FC9eq?oA(o;{8>C9uvfIsY>`s=Yn
z<2#TSUH$S!63j6dV~a@|%d{*BBraqW?jYSF$d)U(HZsdi!`WsFuXGo<a!8oV=B?NY
zEj0G?&@{oaMlvY61F%mwCTtKlZu!G5Zuy@z$21W=!QmHf-rFRq&BvC5_S@5YHX7}(
zw#}_QQaAGJE*6A{QwQ^e*{agHTvN7=`e=ciYNl^8Ad?WJGx%M=#pb*V*YnE{xGY=i
zgGxDKqzyOw6didDtmISfYmhfo%Fd_3_zo|(f27BhYecl*uC-R@qWipeN-yAgiq6az
zp?Q4>M8c^8*$jn03vF73Uu`Q`yIIcVw-=dfHtcYN5Vfqwd4bo=-@z10)bDutqqT!g
z8*+Vjb8>g))I>S^d~O!4+lt<uyLd<6nc-9Vhb?4A_J0U<`2VJ}#Nz*Vmi+(Dl5b~8
za*{#EKOvptA7TVJ{wGS0za^3=-!bmC|B^`7lawRC$m#IIt)ofTu2c{+ssq;<jy@88
zd2?LxkXywDWD>Gm*%?!WwY{7jeT>K48K@QMG`*tA+!<l^m%n(pNLgF2xwYc-JAUHz
zV+@~d&a*u@q=mfMs<ce~@?g7|(II-V#?q!UXszcZU)*uy>r$+ZZuf%ts!kp_J+S4(
z#QoCE?ekvA(IGl>@DFb|9jMH&_vyD^o0(@b>Nc<16K0E=^7KtmtVd%p_gZ{v%o!FT
zDHDT`id5e7W}gW44iSn38G5QIabVw&mUU22HvgdB;?7D@Rd^v)|56*SJufkFdK1v>
zbn)T9(*bg%$g*rPhmox_n%sW7L|Axr$gn94SvHVM9u7`|y~B_h*+fX*yvK?Ec~({a
z>YWj2_HH{<sCE17cz_mHf$Oc1W0f#ojJS@2%(=_%mti)AOea(5HHaxIoT=0n)_Tp1
zXs3YMs?Ka$2<x7BET62K3dDmYZplOy#xBDzHZdWZsYJVCC5DVht{f5{Ey<baNO96u
zC?pXRpDjrg{81rW@UZxN{l_{k(osV&E7;q@3C_|DkH~4f6>N*)()UL+G?C+U>T6<T
z8JeA%K^eJ8(orje_11`O@%ocpfZY@J*Z9d-Ch$xwcmk0<FudJBY(U*)5U&uPV;3M9
zeDW^KeqZyfIlE9y7c=OLhwezzAf+35=z!*CUj(5IS%A+#xBG5GNiaLE0}aDlu2a=c
z2-IuUK}iuP9masmI1djy<&#vO1V;guJCoa7ti*68OlGA)&ur|&IW&*Ml&2|$MmBhl
zu6G14=viTxV`f9Jx5;;2L=(({9!boCB`b_Qb-Tm{#?kqM$TE|@Zx_xJ;u=m5$^LKo
zX=C84XooEVlVyfs-u7h1hm_2^t7;RNEYh=?TI;dbAJ>vgvNICjjblo6ADN~~zL2N*
z2lVf#K929GKENr`Gn(t^#L6E<8cS&kv}erM%0H*Ccfa#z-^c&pmH5uu8@(Ud)VUvN
z(knTwwk-Lbc=S8f+h2)C|H?A@oq+qhTc^i&;!(G;wAcSSRY_`6sVO<Vy&V7JxoIDU
zr)giR>ATnW;a>^4@7`9BlVi$KW^lX~cikRF#Hy2{{Khm`!Qf44v4tXTpcRQX(sigJ
z2+UV8Y_VSWd!;lPa%v0E!&sY)ypwnPm(wgpP5<0MFMrUXg&@Ptw*=U<(nxY6p9~fy
zz?p1#L)nDf+VSRx({WDKd<p2@7s!5sqoAUA|C3c<|Ib*j|Hg&N!N&G)4=NM$w=Mbm
z_P_U_7N!5P(P}{qzIvc`#m68D!LbK}u$OP?ssqMwR=}nR*T;sA22h0}dCC9krhl5O
za6MSKRNp$D%yWLR<z?TG)w~twSfwaSEmBap;8E^u-cp!JZ?rg_QV{cKNf*U8OU$m+
z#4>%lG=q^Qjq%jO{6O(sH<u9!CQ-cC5R+!@uqhx#mk_yu9chuo=Xt>>i+mwJ3buo3
zG`UgQFz)#kZ<rDhzZs7*!uNBLgvCW1qqtFl&`mm@sjZ3R#Q#&6XRP#C2_~l@ct@Fn
zQq0+kW3f_Ki9%fxyr8|1YPPJANudfu+YHy(Y_^Lz#BQNau8*xfAaAyI4X4##9(OZE
z+upoQ>cd<XcFz|?GyzIf47^H(0Su#xg+bJ!tju5UTKo(brXqf=(S`B&x24pc@9&M?
zYa)re#`5L3Ms^H13)3lGcF}TLp%0o^6ugAW*!EtjRRY*&@$kgLU^oyeQYRQ8hQJf|
zR0)P;xGn-cnqRs8;x%Q<gy=-I$~M4}P(1DVie+-ePbcA+!khNlmjVzR?eSpXky^zQ
z2-pVwplgMM`txoORh+*nFraCGhM6B{59jkB8sShGphQOXFy{0^^?)M*BVsVcZCl7D
z<~N@mnN}c?+!@n$4EQyKv?d~@=7EBbFsDIb#7b9u`v&za8#E9JF#FC%<OH<btRW&`
zo?PWyxe82DFbRGjb?K`tn4#N*UbF60J8AI?YSyr>QxV~P2=ud6;_`)@_`=JS&6xSZ
z5l&<JnWpX9{p9a*j{MTH0z;0}v={P0n8>7T)AIcK7&?&sdbyzb_4)94xU+jZRpP^g
zqhn&w*7`ob^J)G`{;S*LX|Dp^rswVDVO=#O1|R;a#zu$N!|Uy8pn(k^o(=wMaG{g^
z@#d)FP3N-O$D_3i7*}QHgyqxD%jN6(eR<$le8fabNNmLR#l^#0h|blG`ryV*gHPMb
z!Q;onRLJT=b8(KfPj{Q>8R{BBf5SxxzH{Wq%S2C?$HQ(%3ST$J#zxBf`NLv}EIz!8
z4})&tO#<N4=S%3s^S8}!_F~#ibLhrk^!ywO)xqV}+SJkF?fD42@w+~?bpbuW02+bQ
z*6Qhz_2Em0uNnc*Y~dNH8{jaBmYv*wc^`2L>~@-CfIn2cTWTw}k*&FDyd(6&CY&1t
zJIknvBg+4^>xDg-7TcUwVi9f9+e|Px-+ZLz7db_<$(((2nE@uQdUJjxBpN0&&bmvO
z0rmsbMmDPC%6}A+WRj&{OFZLdgBgowtAmz*_;9;T3qJ=fT|IRI_|ginK8S42HR=eg
zfuSJD_0Oa$-XIBQ8Q`jA%p&2I=DW`Xfr+4_L#0!gi76`(k}<QUpYU;dh@0oxf<gQ%
zTD#dL#v}jf5kkzYli5w=6GM*J<*f=Dfz|vVaUjp+5ziKJpqB|mj$r>I-j6@DTeQsK
zASW4^GXc7=CeiR`29z@>SfJ1XhM=S*f%F9I=otA}no6p2<JS5~ea$*T4CN+<YrOPL
ziiu)uL12-vpde!?&zR~0NmRe3VM7wdfe-|D@-2VY`JW=96(O^f4(LYMu`j90$xsg|
zfHf#X?eI_v>WNSzI&0yBB4I&B(Na-U6frgKSN;@)&>7pN80M|IM=>J7NMkN@Ww1D}
zCTIjzn%a3A-;k4}P{iCTr4}r80u`B2ayMSG%*a&|9aGX2a6(F@J=LAig^tpaR+SUo
zW3XCS5l%##APAgouu6NDQYwhR`n7PJFN)4W7Qgf;qRS^j8G@&|HJ56QeX(?2H}rjZ
z{vmMVr^1+-<Cb|v^tnna(w2R(nqlT?K$^@oxfrA0rzg?=e&AH`EOHEG(kyKVS`{=~
z&prbRS4Kxe&g(^^L`m%LuEi*!AiJLqz0|_uLUuPve9V&ELq=&3F=;*y7-u+DuS&9Z
z`i@a!Ptzg>M}8gssW;>su&U3XS^FG-P=L`ScT7zolP%BzyJm%v_DQh<6rO|`_^ss8
z#0VdjV_r>Jy+9etG?zlETnJ~BP>a}Z%<_mJr?{kzT<uT&e(+Dytjk*6g$)1u&;xo8
z%2=z<)1;YQgyYvfjL4GxP+C|PY2b3Fe||(JoxL%%NwY02$}3bjoxS@oxRWmFt`?Tp
z(cEzf#WZHjJuOE+em?{uQ=PE13*OKA+C6qm|6CM^n9`KelW7r3`E@*aK@-Uml{j&N
zd(_0=?4pScr21N}XjDasph{uq#5Ay^KZPLb5+N(N!>F#blnmH@Dk+U9n2rfbCfAWO
zjx|GkL}Q%?Cr8fNZiN^@;RL+_g|iR^%85d0nrenk!U?F#2?`d%Lg*=|MD>@;KPpG*
z<n97UsL_POAQ4%GV(=MQff_A@!yyj4W}exorexGsAi0I>*HO@*1r6obstF5BC%4nn
zj8jtGvV%|+D2bNAL}_{ag%_~2fN=~>)BWT-pz<L+S&hojP_8@#4Hn~F71c(45sIof
zOlU%o=2_pWGpO=;cDqIpKik8Y;rHL2mAU!#h+9koq0vpXMYST2FQ@yl*nh0g$rcpJ
z9HWR~f48ay<WiT40{!T)k6M^U+gd%TF1mq6`N);`B~V0#%0sQ-hG_OZW!9}BQDQ2v
zN6>DQu#u!V1k+xHjNnS^AzlT+s*P#(*Ee-J44#%2b!lb*jQNFqq6UR89{`8xII3$J
z!yXqrOR)k~9^X5Q2XoQ)&WKTD_uGXoJR5VPqSg~7IR+uLo)j1ha^n+Lw2RRPL$TgS
zhaVHU{WsxoDH~V^OMg>kZfbW_TVHDn!`f7<cG}eh8hOZ{-Hl8)oH0lXPCaSz$`gV}
zRFtWq2p&7pTnFgw%)@SnJ|BPmPL7tgX`}HpysULC7an|Nhd}j%Dy@ItrxaA>`mOcT
z{Cjg;0!ayNQPTgm9CWHGwk+_?Fy6zDdEdY8^I@}U@#>ZF@5-Y6(+{U8LtL#I`eUV}
z#`36~QBTs!-Cp-<pKL#e*&L*KFi%e0EM0%}w~B#9kTTopVg2D{{(>?TwwWA5TL+1_
zC>w$s>i$-Oqe3Nsm#~}c-=w73DSx;E(@?V-tl;=eV6x?RLOhH}3TyYC#QO~5(J#8L
z4i?Of19y)nY(HLP%+}Ol%@kqDCMJ>JYzX7|P+*Nby5XOl?06o9am<Xkg|fRsapBE~
z->K<gUX4ca0Wx@S%j{DU)u4UFJZxLg998@FSRxp+kHs9}aR*RPL09g@{MzdbAcx)X
zfRwchTu;(av83oUMJ*2BlQk%yGK7ADyJ!G@{Q^>L^oailuE@mjpL-?$9am&${Wn)+
z<X~m_Kjw<5zspJ5V1uvTC{IEA)#8mNYH0%bt&!pp?9uJ(Vudi&SoMrqNo?@T#xIk#
zs;hgqa2C8aatv2{K2(jQT8n56@?qoEV%Y-q0c`?Tfj$cHnM<(;KBd>8;=y8@)6cPe
ztCEZN3atdMXLBv*%Dsd%sSO3@P}IdrG}r(PBW}UO{GqUv7fMILlweg{12V0&`d~4h
z1n+xDm8W`N{}Ac_Iq`|PZ?huEq@luQ28u8p01=&*C@}BIZ6_qd7UR^R5nbvNU;Qb%
z%1~emM~lRPi78)T-o(FN8+y|yOTqdDk}gz!(@46qa%<i}5f?R22%^g)o*OqErGac!
zlNchpW;#b{TIj<RC%PtDuWg~kTu&&CEm7nvVpd>|pWz%&f?czw`x9+7wwTXiZ48DW
zU5G19z%taC$$$~>tdU}!ed)xc63r~G@{ON33q})jN<(4ATQ5SD)@XuLU`z$jEIL@-
zA@<O8ifUG(fv#3`g}Hoqe#pRZ?_^qUoE2_<xZQ;rXj3Ee9G%nb(962Lr1jjPAxf3n
z6z-z%bDnBe#vYh7!Z}{k0_{e{GQ%uT6EGtq^kPweuQ4IWv5h6a%1}2VUUZ52w;8JR
z?JR{&QSFBympT=uI-rOY;<tio>X2-0EZsR<rb4j3b8QR?2<^8D<NB>BIdxsse2ns9
z0Y_s(lGTq;=lXgKSuXVmDiPu>^QlMZ3q#b#@BVA?%H>&=(vg)}iH$STT$H+9NEe0P
zIe34(o}TYhzCO<k-tJ5AKW}axZcf;~-d{>=+5_-jhf?5ST?9%lW@2P}-rk>T)^+0S
zzAE1`>%iMHJ>-}_5RXOKfwGFgd_2B+<yT5Ykk(pFKta`s0A`;!6$H=h;xlMlr4?7n
zi8-zS571BdOI$z!_@C_)ppIz~3iR2z_b4mZaHZhY#CTblgD@ARQouNXQ3Nr*VshSk
z^_=`n{sspPxG6loh#S*V&z$IwTURbJ@Xv6iC1JpcHyxb*%>eH=M>c%kpJNz3AMXce
zY?|?&O1E$L_j}%E7x*zK2QOzcf4tvkeOk1iYOG3eZY&gLd!BD^uJ-I}V~_>PiET+%
zrj8&5B?X;9ItlOHzpig`yqt|71JP@npL#WjZPoQq!D-lfas1364Z#bcb;<9cOn^8A
z(;;RBCEC8eR)}hlM$jm|Bi_T0l#03(k~oX7&a+S1k7WXPAzp%g^&Ul%`R#k}@kWvS
zCZJZQ;%p8_ar|0A+9~g*5;<hjQWTBZjDnwfH;8O`jL1K}NBZtZY2t!p*bS%vaCq=L
zezPz328bPEB7vYI!FSCgNmnZ<ub(ho-VR;*Pl^z;)wI;*Xzt4>Mw$EJ-YGBQ2Bv_c
zP^cIfZ48XP7+BF0e`LP8YN!menvJhKbJsTJv2ef(&<bR$5+r95ffmj43e$WY`_FUg
zjSN>en*4eA&Mr<D6I{9=R-@rqH_NW<w8P2mK(L0{IY%>kdS=^kpiV;Jygmif?ZD(U
z58ZIp%l7{L`-E{=2JN=mPWyxe#r*H@sIbU8eqJ{G)vfBfZoBF&^a!fN^`L*9Iz_{S
zN^-BX61}6uJYwF<N8B6n7PgPt!pBnwa13yVM@|Eznp%o}+mL7G`+H~w>dq`3_5?f!
zG5c$JJbU^RBqev+U6niOpBqY2pg-#?e_AVE^yyvFi=s94aF-7s1jicum_r(c6G!ir
zX{CU~-4SJI8;+-F{-L&HGK5Y}(zJW8X#5eW#N(iTIsM(fg(W;Kp?Ofu^oHsLw_lhM
z36kOao`cey`UQBd4~9Pg1(~jt<j(f`oq)v<G?KZB64vOw0=cM!TuJ-4-jYZoJiL^i
zN*gWvh7f0z4GD_b2f_&$y6PPXiZWM_BzcdSs{Hm;dG3HXfkqPhN`fz5T!eU-25q8T
zVnbb718+K%nVFfU$%>9TuqoeVki6xrA>{cQ7vMQR2)>8*UPAqzkX=}YQ}FK^r6u^L
zSAGZwfs!143CM}qx}A=;ZYREqkh7q*$hv>E#`cAeXekqJFg3j`B|=lCUA@~CT<xdC
z?^fp+n2BjY(>Bh@W8BFGjSG!_#M_9vLnoreL>dL{MwU$u--ASwi~Ez-vg>vO;DM-P
zDcTJy85?lYyM{Rgx)$<LUqWYZ@dBP+FGCVF1b7vnrm5OcJ*=NXcnU#?1=HYGvV&8b
zfO7#Veovn#E~dh|hpubND~&4<XAV`<27m)dF=;GrthKKIbA4mgu4P50RfTZB0_n8(
zxFz3Shwm~TzK3n1>$yPK)s<{WG<y!!UB%)Ft?<tCC=jhN9a6Tuq9-(A$)t=+Lmn#J
zj7|N0$SzLhB;2Dg1Y7ODMsEm6GF7MlX`ifYJn?$2LvO~*!_Ivhg;awR$^L`7AuZ|T
zLtfij$;9};(AY#U5sAKNpSye#?*SxIp}llg!mPRS2ZM%+x`v9~U+>@!B$`M2_opHU
z=K|hy5}r64#Z*j-I)NEP^XmlROq%5VekGOqtCAhBvT@FL_un*L<=@xcAj<=|SJZG{
zw(D8m8^xS2n^Yu=dZY+yorvUB-sySz1>lLtKJZWUfbl=GKm51!fcbyX19p!8s#fs7
z-%1KfTQ4USfeF4k`lbhlr?G|Q{9wY@KZOU;QzN_>383~Uc)Q0=tG6ULH@ym)9#20G
z<X-0aJk}^;?ad1kg$KW8h?5u(e9MT&^NN?AD|k)(=MMNGMH7T12WM^w$xnme3reCN
zfbj{U$Y#C^p^cb!#W0s}Z!jip1g%(5fu{a3D^`%e+oSnup)Ny<N+DP@SwcBO=o7)t
z4shGTyw9<3<B@Qb>u^gU52%o4G+tJi-oU*KmJ`gq+}bSIZ^5z%b&Ks(yjE@~N2xqI
z%4ohfA1h^XXdR0(gS#ypl1Z@mNfnh)SuobJ(aqU#UFhR5+A`NGrNv<ft_c{A{8fTx
zjbaC{X+glZM<T#PdZ`aQmI&e-Pua!$vv`5D)$k{7OT;cRAJaZqDA#H+e;C)8kyLbF
zRv?hzp^RYFEzTo|v;r{*@9$Ve2>_rZAi$D$7zV(xDFhu9$2LrJMX=n7Faz5zoB(=k
z?FS@2f6cBY@{Q92$GEd3MTyo$P0{|UEq_mU)X|u26`O3x7WX#(eRmG9k_aXX*tP`r
zppXrT%sHfuV!QFMksD=td-5{lcw`cD<?KCvEkEXDpD5pb4!U&PYcE?8H2;n7=Ou>y
zWfNP!ZgE`VhiUA0F%;LLHzc|*$jHUJ#H~iN_S542=%UD`agSu}WA3ktTpEJ6QS8e&
z3%f;gt(J)wZ;a6mjF)7-j`r`#JTl73;24A}wdI8M>}&F0l0d65ZOcTjT@O6{@p*oJ
zOR@3k3~P)-{ag=;;TvtUI@00u@p`*kFMeBw_jY@`dZ_E+^?o&kR7R{P9i6r0dwzYq
zYQ|wkc=F}*dU<`K2C&TW`Fwv3<+>Q@`P5r+DSDaZ1`|8pusSAQLeoe@Ajy$H-1d``
zzO6ijSV5`X05USc(#_Q4_4SH|-X>b`oi$Mf>Fmk1w$K8u)tQaE@fP?lhXvY4$y7SJ
zc+f3aWxUVubSr~#m(9>V{cDR4&=v-?5S&0wH+6>})DoV*pd_glTm^LUsh5`qfk@X!
zRpROcFceKqkbsgv3&2GXpD>S7unk{=!~@z;%-*OoYQOz~CH@0NeJrdqbM`cWVmvD`
z69OM%YG39D2vw4t7l1Kz<PW9%z6vaKLM6y@tQ$R}Txqk-ulAgz8*e@jXgXD}a^vkX
z%se3ZB<MTdRejPZ>(nJ^Gg-6?22^~^1t!2ADV0(lroc6jCZ1d#g!4yVTMyqgyx1$h
zDeD;af#PPdGkE?E9)b-#X{#8p@1~SSt5muMS=9#cx6~8H&s=Qb2B6X2d<VOgibL5G
zU@b|0AW1&yr@0q^hS!p<ky23})0N*)w~0loKzO`pTl_FlV7<z8qld|<q4W|sF16X1
zyHV$ayXrLISweME(FM8WnUt0(aEE>ICf7W2UwLt0xx8qTM}a$##-f&CsnT9lIdD8n
zIS+IBFHGfI(M26OQzZ#ZWp{@1eaqqW(vRUWti{E7r#;tj7^?btWc3K)JXjJVr7>(}
z*;>&g!GfRTrSWX#<bTac9#i>#)G4H?HVX=tvVv$*r~CuIg%Txmd0`}#+xLL^pveg0
zP!Q<7SqOhWj~7Xm*8zG)JBsz!aR2HQxJDep8k;Cj+7IsUcJ{ApxfOi?nMA`dDHRU5
z!wXem&a)v^$`@z{pc5=bQljdsXqm>sv~Y?+dBS#9d1`c&I{~lq&PDs7>6Cst(mqXG
z|27bISG5`fR>XY$6S#sow3){*9<3Kt?+Q6^4v3W2c8>FO)1lP5D_{|6rAKH%k)o*R
zuu176v&mK=V@_cJ+<#F4x@jRXS>t;SXnz@&HYy$Q)N0T+$w?}kmCb?LA_~$f7Zt5t
z+OT)T+0B)!Xj}+rC`(klswy_${pvrV`AJyBhj89e8hn+X52CmYXqsUcmp3O%`P9oz
z%aG_OPc;TlsyWk+<jjAdPN|dBiELkrWTZg(=L*8Hdb}pmQdP7mI#wp`Pln&t6kgur
z=no_kDz=ENscZ;iqwP#W1e97{fCkWE%<dYP`|QOgWTAxc@xpwzvQO_!&`FgYvElWP
zN;IvHEQq8Gkt^dEISZzrsUFIKo&qZs@E=rEJ>gq5ucuW`?Cl-pI>8<_-k7fu=vY(+
z5nC(yO+_mfsUC`f_Ob*GJhcW8O8z(A4<IDaFh7(YQM<f&0IR$O-@?bc^IK}ja~EYz
zb>vr7?;jU|tGtG0yYuU7jxWw2A<$BVauvc{8oWvBLO_a;OcZ2)Y;82baqTS?Rx@h+
z;Po^-kpJzmK)is*abH@>P30crPL#Tg7@UlnWF%sv&Qq)Y`x~jkdr@UJ<bhQ3!zf*h
z4^N8c3PNu&9%939AH5%ZmwEQJ$GGQ+<}AdSGDhmZHC1G5lTn2x!pH0LTPpv$w9J3?
zfHXZ&65^mV2;WM5eJk6M6vI@cd7{TntrmSl4<p|D=-2FESuDk&aR>JZ)F^O<Ir}=B
z{UQF3i~4^i*7<K2b@qQR>P-JGYWIJzsB8SHAZvl`=IiNCaf3cti$8#e@S9c8pvY_N
zqHbdkRBVr(O5SWK;_Hbw;TEgs`N4)=qdij~7DKB31G^Alnu|o|7v$}qeL*1g+<?&f
z?HoeRNS2Oc3AEosIEIKg%3H{ic46f-qX)-2u7K4HXv12+f`}sa%~3NW>7Hr$S33~I
zor}(abd!}G0lR#Qw*^5~#e!6oxskqpCawbB)KEsujP@K-ygkRr@m~|Q%5`n)9VjCL
zrQ*0EfsNJ$(Y(!a0c=Gg)&t+y#{?XVl0raqme<pxfG2{|#|8-3$0P+r_WKS60BNh7
z7r5(Dy)Kf-hWd72^BRK2olJzA0TiQa-x%Ld0!&<Hs!AKtQ*%K5#v)x1u9popf%8*+
z7pwN4{HYJRN=|1nR=;a-TUHCN!c^~TX|NkNkciqADv+%^8;XSe=E%s}nW|MtBE<pz
zEEQkLjn6g8%Uh}@OwD>*uJq`4UH}EVKTjs>KN%th81%!J7^0s;Nn7F>Bc3>soBz`<
ze;*L+DwG617CuG+0l$)@h>BNDV!R(pO@a!kn_vFm$L>Hlo!j&;?Knw<hRS$J#%7%c
zNy=s&4t3~ogh^5lNU-N?CSA`R<!u;aCpS<!@}+Rq<N8ecn*0eCXuJ^>5>{<eRo|F6
z)m2mpF>O@Uc$-lzwf?p-ZK_cCrSeyz876>?eRXH`xTYgz$D~pofofyZw^rBjM{3pZ
zCyw!W<UhboB^(o7`wZj==t;7?-JVaI6Cpi5o}ahFl!H(nPs=Ire50I}XFZ+Y>-j_5
z;+^sH?c<*+)W)9A=f|EB%aX_CtDY9$Pu$oaJ6H_r{Az&pYFECFz@cEa<lI1_pb%`n
zTR_F6y6gZ&;?)2}s#yG50K*z6x}1RL5Z+<@P=K^Rj&NK>K#wTSN8gO%;7pP6^X^R5
z=@j4l>4gvG(JGre;PxE+a#Q(9OJzgSON`<L%#Ycn4NApV0&p0)7k4KRi7L>R)+9Vx
zq0!!XI3qL5E(Mu<=IQ|DPEBU)P(|CwWKx}Q7S_KEzb#+YS01n*Hn+vsBDmaJp$klE
z6^QPPyzEut1f-Uj=I<8UgAiIRxf0*C;c1D*8hj?%Ou6@*8lT}%u2Fq_IsQ*_2(Hye
z8VrTs^V0XJG~2P4Ttj$X^(Sx%Ca647zg<*2j#?vm%a0AZ1!^|~{x8zr0xGVjNgTxn
zcXx>3?hNiExVyW%>)-@;cY?dS2M7drhXi*E7VHiAef#a6|Ji;2ednFS6g73LyQ{0J
z`&RedTb*JgHqQ@UAU2Ng1m{~(vgkk?HGvvoRQREOAQ(t$tk?1F(!jI}IMm{N8VGP7
zdJIr*m2+1ahZ#155-n$t#!<Zkc?H%c4w`mA%rlbfO4?IJiLQU}vM<<t&MxLH6#@Su
zimI-AXX+M6i+`~Q|7^C|CaMDot19?z6Jq6$CL40Op9r}18?2O@LWu?L=t}5363uNV
zlV~fG=-KZ)P%Gfv@|B9wm5zj(VaQq&rHbRFj#-LfN#6uSek*_Hq~P7=r?^uV%-)=1
zj@)MUZavDm4&*IA#2G$x6{%gEc1jcH82iONhB}?A2o<tbvrBB2=cm(=1-Gp`@e=63
z+LKRc=16RY23AV_sGQ0yUCiupumO0<IzV4;h&ycP1~WDfRZY;nh!(?k1OwJ)xt~IG
zC3j5az{8f4gg8f`QFj?xeZfue5nPV!vy1E`BfH7~eayhgPpiAEl#%jcb>fmmgm+pj
z;O>QL5{E4#Or^{sh^oqMNAW-@)qyIc;pV?bB-ulZK!sA|(Go^bfq46!A|#YlMHXv(
zEP61B30By`$LRD0zvc*AUR93L#O%)Lp5|;Hg&-D|=YHaXZ_fHvpH)Cx0(N7yD6G(O
zsaic~qh?SA0jeSnN->d;HoK!HduashCF~R~3S%U!>U3TYEQ~gLD?OotIP}B6s4CM!
zU@LII{2+u<6-BRx4C=Bp@SFwFMms1I_HD4&c3K9OCpjq7^r=Z{s7Wa#W7^SDEJ-2!
zkU}U|dXg>j(55leq|sB2=s{RUd&fj4K3;w>DYAj_RL?#)hOj+`a2t6z-RbRoihZ|7
zc)Fj=?53_TOM^U1)6<)Fh<)OcVUdR1>A5W3#p1T;>LfeLk<+^26Oe7}@AC}4**<S)
zZu-AGtoHEh_wbQb4<2?uYzqp4pZERq@Nm0&;b2*TW_;5*KWVn1+mgU;X5gGXe(%s{
zZ+l7=bGKa*I~fqO{yh6h=a~>-%`1M5&A}^o`|*}fGA`41Pb;HK=63Y9c3px8_Q#lj
z0|qQo^2`pgk8=W*H~OOtZr(B#ix3Q4Mj$MPz5GvC^b7o3{Xz1CbLeJIsgG1yh3F&B
zY0D8i6ccXooij;g#5yaZPak2umQL|l@~HL90W9o~Npjzz0yed(>I#<}H^>GBa;X!R
zuEnwY!L8v)V79W70P1(ppOUyWzQs7>tDaoPJ7IfvWJD#?K;1`XpzGZRM3Nh!b7y2n
z?mdPZofH5kJaNUZpSR2ElpEw9u*Ex{*`9fj4b8|}5k=YMo$;ayJOh@yEcDE@#yBUx
zh*kpi6K6*h(Rp=0CSBB{>A(n+J2<a?^s#xJ?K3b(aLy<X=TCW%Gb9<!@~RYB-e6z+
zJ<?n(?*2S`eX86s<RJKGoE)dypXjW3N~TnHN=7lWHvLnk5t}T+CO=t+T<hn7sA+}I
z<0hp$W~?e-MTa&fxs+`w$K}qjE3NH*b8}X;HC33u-sN;RjfiPxX2^WKv2~X{(e+~E
zNKYtUJ&SlIo27L7^|<6FQgfJuoLa2Rs*@qNoF-P1#lhbuSU|bR7|`h(nlVx|{MgLV
z=BBxW$^MZ&agR<NCX>C*jrT+!uj5O`XDTu4v&XQ3FQlZFVTVzAc9X{{me2mL?JKe~
zp||IY9c9s4rCd!hJ^s#Tl!q&iTo>uOex3H(4c6<vujre1Mr@Xv%{iov^xC{=?Q+d>
zXXA_r0?K+bSewJmSYAc^_9kWHU&Z52V5g)$=Z4x?32CuoW2LZjaQ@8D^?a)*(QRzf
z8ikrOQNR|Qi1xtb>zoV^J!VmaWx!lD{Z|V9OJ5t_{Kaa{MBy@;qOy;hoCM0s<p<6(
zvSkI!MYujdLv^zA)rAu!<#K#CSsLZIG8)wzODe@94?3&jwf-5W6f?~h<I6Sos<zLI
zP;2~8Pc@7Ff}7n>64hybOr5J9bT{OvbQSaFVe1Q|)j^*LH$QJ;J<oN2*}K%Xn0<tx
zKdt%UfnNl<z>2>LiVCVrzHL=THS!KyFyTP!y34A~A}xZ{Vdalzx60h9tm|q+DdtQ>
zar$xk#o5#v*Xx`{x6P@p$BImXV8Gz_ccrt}wY;d0L&8#Rv$e?6qrUC0$Ab~I7gOGo
z2zY%2mMWih>xXVN4e}`6FIf(WB}Q}!-KAnmJVsmi#x|6@otS1lyqwsS@l{eg();7k
zoOrR2wMM%nna@llT;w!74!ZV(i+){`E86Iz!e*u}Fxw<m&>|}}7xQIl99!HZRobC>
zvp$ib=f24OcR?%o7=iyxhy4E%w6ZX>1OFYha<DP~uk%7g>)1LI4afh@3sKUZP=773
zEmcGt?>n=cB2V=vp<Jg(N7~`@83qh~r&zxKv#TJI!2`WprB9uDLv;F?xmSDc8RS6~
zSESj`DtDyJEP;FRa6PjA_i&vNNX9-hC5_A*CcdLciG+N~AvEHG<oye{BuHk4qE0nC
zBaAl9;=^nG=|cyJ`rMlbKJVyt=+$z6^>eAIBVF~kmCCcKBByCmQtJs4Ws28+79<*f
zADm_yZGra=NB2X!xJ{N7Ij$HH3oh=o=p#NJ3t=`>vA7I^{q=j?nu;$DtP&ib#@{c0
zbsC{hxRY4;<Q8vK=me!CB_t)L)MxcMk5zW6r1d*bfI3Bkd%N9eBr#EwQpy;D)~;E6
z<g8}BW>WD-%@3>Mbs&Kbag2(a5hf`J5g@&kl7mf5q!Rs(iC_k$vB+{9EVI@_i)E*3
ztJH_scNH;%-;EL5X;^bD5vLgCZktDV(ogbP+Uu_HU52A+6<n@GLn|xIF!PdhIg9b<
zr9oWHip;tnDl}cmJ(K3YHNo&ur4Ze~k|fW=G)s_~awsw#6jtsL3Tu#>a+tf`nF^8m
zLThS{hm^)lf45=;CMS2XG9GvHeyDh$^wN}dVcp;<EUeFP7pF?p<q{B^7p}xKZ`K6Y
zlN)(tVA3=NBSTpMuzp8Z0Z><g6aEp5?xH8W5Nz`8CJWr4UUD*8+7it7QLAZMA34rx
zu~b!kx{5En6*y{Yk)L&^zr3e3>!lhNn@Hj*bgtK`KZ=%<VHu%lk{#ouIRHBR;-M^p
zt*_S_?OhzPB7hT;j~l_gkHE?}!K6%_cXHr1s@IAE$#Z_&L3S~5QYOUGe}1{E@w-!g
zTsp;leYF4acx$<7ySca6j)uu7z%*ejKD0xQZ2a8q-EnHS`E=>GZhJ?DR^is=?eQ|i
zIn<dlYtOW!w)p<;th?t$!YDjQChhxEnXg^T>WlpPZT0G?iY#;;AI|aZsaG4&=FWcO
zZgsre_r!JGaH62!-}|P6>~Yvnu<k%V?U#Mc%LN?l&QrW<htK^?m0CfMe0$2tnpmCc
z;_K4JMnryu7~7|1g0kGR?oAu}l_n`+`{kW4b7l>zuXp)f%gv7i&B~|D*<ERNt2eQ%
ziFt%9i9OgFJw9ZN=;i6$f|4W!?ss7~>+X+&)f1~2V`Ay<6}O%}eKNRE%FY|MXB{_T
ze!kaBcJ6oPRY4tbt9+g3YDN$H1FmS~aKirYT79t_Oo|WQp_f2)oaOWK+I?5`DBc^F
zyCgaYZ@J-34%7}~o!(x(mG=K&Wd?zj*|w;Tw%(4Fu0qZxey82!>U(XuT9ZnfQm2$<
zbhm!$*6`BSv$1}+>b4$9oF_fLxPyo;Q;a$-P2}%gz@xBPb*{Ef5YB!R_j=Hg{7Z#Z
zPd?gvQ_ntEz5^z0hJ6CruaJ>#DY5(<PUqK&cctUi<eCbLr=^^43=c8WNBLpeUCn5L
z-xc$~v#(|TUHb0N-Gbg1<jJz)N6y}Kiq6Q!p6w{-E!ls59j@&h3iuo>tonJ{6CuS#
zRzbNKZasZ;4VIue5pPr2C6@b^0TFMRbYFWovGm=q*Iud>y}b)ZR`QiO$Wr<B-l4VQ
zc(Eh4p5(i65K+ouu`km|^=huXGmUES_udDDqXVBF&8i%Zr?OqBUa?5B_M7$&Pk}O;
z*G$CUvFdm0v%Vc&z@l^HzB0)thow_F_*JuZ1i|LmQ3&w14d*Pwecj{xI_qoGyQx=)
zCA;q5Kf-PwqrEiPsMX|km}H;us4BN_3SlP4rIeFqFD%DHD4zXdu}}z&$3xzQ-7IGs
zmZ$T~47+)^ET4LD$CId~^Gpv%e|-tt=5y1qJQ5b^m>c+}91vFpXU1ve7s~Td(Df8J
zN26=TFkxSFxuUK|)rsP4A4~szg$6!%<#d2XPFtICgVgQOZ;K*reUKJ{2^_m9w7e=)
z)j2GF(II7am`%DY!y{}vTkt<SyF0}-aYE0sTP_)oH=wZT$jxb%uYdPE;#(Oo_B+ql
z>*fB8*q0IiVc)Ir38iG}k(L}Q(@a2rc-y(dJ858}>h=A*YI`#G5}mT_;4;p~-SVC;
zoFEJipGrhz!$JA{XZm<~qe;64zs(v_w|(dR>yRzO9t?@sI^PZCu)Rz)^>aaCw<8{C
z%cL~qU{QYHyi9iX5f?fyLSG`2#pV70{yt;crs`@<&SK`PwY*+Ho|EuPb;@0)wnL%u
z#Dd+3*w>CBLo%g408{5t^Wy2B_u5v41&EVK9o?*oBw2LnH!nwJ=V$Hnz8#o7PY-_A
zOGXv>y#ThZKQmSKF3Kx6>s*pAR<qdgxu(_YG?)WzjNf-Ap9@J>{ZqXz4p?7rPTlPV
zbf5QezLmf~PfA^1#KW&f7T)c<`EHKNWMchjfo8}EIipikH?P{}MiKDz!jFZb`n2`Y
z)9Rc4N{8L^lE5by?lzMcJl54FtH+Ne2{Kbvq27Ckhffv3T{m04LA}2}z1(;{97EUN
z%<o+)sx7uT@JaL6)Q#w8SH|diyC0~Z-j1i*wUB9}%|6?^|9G@}{GmiIcw#uF5Tf(3
z+8Xjx;6mo`t`gKAGtbw(IxKa5c|H`U#Z|kO6`$!&h`=L8N<JI1h2BL}s9A!P)7vub
zNG1D;?n<X_y~&BRr`6f{Y3=5YNB{2vjf>DMf$@04?rjexgxuIK%Fp&Zm&Jx%T%YC<
zT!+Pxj~i2n_2m~^?H^BTS(sn@JIGe`{ZCh)4X)yIifF~y+~!Nl2R^WvZC2H^gzXN<
z36Os=+06Fo%P2!Q_M_>~d{Mk>*FSH`%p7pqv|k_>R$j|lDY*?_ra9Ah8Ov#`n?1dw
zx~s3&`9%CXt@ItVW17U;sl;*z@twX~=DcvzbYc`U)=|^BJ^ss0*|Eo2oW9@T%d&t#
zRXdwv#H@#V67NZ95F7j;QvUw&_>lw`Mp3yQfy_tJowTe^w;{vNg}VA=sH=9q6Rdbm
zS2cGDB^zt}*=06mqMX*+w&h6S?LA~#;ulrc8=);%pO>?Z(YMv*^wXEsNo!)@_M3aK
zRyP<|so*Wt8iqKpnOyn1%+q}>T~|CD7Z}-d9=1Z#l`%A&M;B&OYxk4vikSviL!?3;
zQZbmepMHF0k{@>b!@9jIev&=yy;rEurY>Nb!P--gZ@sg7!do}L6j5bVj$r6F5j^zD
z*KSp6Iirr8U3YzKhMWCoCrrLyYL2(RjE)UCoOehW?SXp+B0$DiVi-AAlceF6E+Z0i
z8;+4(pxZHxcINj|_$*GJBfUWs{fw4Klu27K3#W>`NzC)_)T0=}s6Y-5q~biw=wIp<
zF!!=Upv^Kp?ik1(u?QlFcq8!<j2~^8DceL_DV|zeDF|H9Ypk|qkai(!eGo1EPKe)R
z1TE(buq#VxhK4_VvLcdv5pA_R;7rlgsNiyz`!mckc)CUejG;kaP)%yA(f>a0^3Bt|
zGojzj1}SS8b4WwZupwX<LYtgg{4y;gBIf4}wc7MFB*!g1+?<@|JvZ!}9L|O#aw&z#
zIAoJpGxwH*=9lQdB0~~+XF}VY*rL|rsOcIyql|clM1S>xE*qVmzLD&UNYHbf?YK&v
z)rwNIxQfVZ2rjzAHwF&MG8O5^h)1b*DskWgtgR?zgtd1ZH^=M_12U<`#0%3t=R>S<
z17IL9Xf=MBUDInQvO=_c(tzE<{X7D(_MQL>q-_M;8uBB65F2M2@c1TLUXHNF>Zt@_
z7ordmLG)RCOU+`S@Yfy@t5&2d!8H!W^)X%`4n^Src4NF39^IM9YOyX2fpIc6*JL*#
z7P$8T7u5-*+i9f69Q+c4<Ia>R$?rdu*=olZM9cI>2bn_IWxdA)eJU&1Nh4L}(4*W+
zp#g)T)lAT;nCsTi)91{3r%;V*gGwi#nKur8v%zRiotlE(PK@>yAC7?8!i|Jce1A1Z
zq?BF0q0!ESLhU^(uQ5bzhaCw5_N7Gay<nCUXd7ZzRs*DuRH9{N#20+U!?ZDgQb#X(
zq3Bs-NUJPTB!%tY8X&^3d?;co!_uh3Iu$V8223}gtK<P^MksZ!!i?4{uil9$h;yo#
z+ldI>U$d|-n=YUZm(2sB3?W9-#KiN>1g=XCI7JvY#tcCX?a6bozv-VgCZ4rgoCTev
zqMBP`5*<>KmN3bWQdA1et4}KxqmV(HLJj*~oHx=4goF@jL@tp6Z5O`_HUbBv7=@BZ
zkYa5yrNf7M14ZO-WB8CQGPGYfMljnA?&{)L4DN=Rxe9mE%}Nc#KVw6GY0VdBiOUc(
zj?iW`K|b&;aIq9VPc#K4oxA1K3%^xlHiR98(T#dwv6}Tu;vjI7D%El<FIPw+XtQL8
zRw`#3ym^@<28;_=)lP!`!bQ9*M9Wv)Y8nIFx=S~Pz4zt3W{3XwSQ7uS1CN#jogRdG
z@4J@=?U&0yfZ~^{D?7qAt^oE;YB<(iiAX1M?<<BR=;a1t2PAN&9pAk)NWWas-FuJM
zKOqL<<9V&|r(i}388D+HxZg?44EMgp!IpjJ&gI{JIK!&9WyF9O8YDo`S4H9+ab;PJ
zd3Wib+5ufmC*`cVSIN()g$eT-UySl<D+=+ciJcUXu~C5rcOiuoOTO>=&HBzFt8Cnf
zXN@b1C9$DH&=4a6`g0<{+$DkQcyh6y4GZN1*Hg8;qKv&{=6{D9<-x4jgZX%<a0i79
z<pCW`C&>nP6qn|ZV$Ii9#Vq{AM=9WDRM<7Jp^GlAdZP=A!YO`rt1Wlw)y|~#X;7Sh
zW#I?=-kg!pd7h=#=?$06bR+&0lhU+CFFsv0%5x~mr^Bu&)lv!4S>x(r$tu`j%=#OE
z@s6$*3JfWvU4Vclh^WcZ781R#j2c)xtRx!?Xcf=FZDk>$>y19tfV4Xa0<<bBFW^Eb
z(psgk@TXXL(f3A&r`+Nb5l%Ug6B71yVT+5Gz~~RXkJi+oRq>Dj?n{7gqr6pcCJ`dD
zUOJJNN{AmfA-E})7SPk(r7jpYAzEP<Y5iVKTI*KN%h&xDxBdjr)MD0Pe;)ZMz;uD{
zVC>`QZT{Vx*ls<B>w--xMni422i4s<RLc}q-6_;cB`lM>8`7Z_v`qvN=FJ60VTF<&
z+(L;uo~yM&>435>vp*Y2Yg3?d8d_hz(jO)1A-$Rd!&p+T@hc2GQAMg(PTi#>Dk<C!
zlJGU8Po^vv8W!p#2#j|CgjJQMEh>JwKH|e{lMeHzZbH#3ofD88R;56=IjA3yO`o)n
zL4+*4`yf%-AO8ry4?z^dyAPqgw+pFFxkPMGNL;Q(d25bFZY2P8d(>@OJ#QqejrowH
z2EE`|IcL?&bJWd4vACdkH&4^m;-kEkw4rUjpI-DPJ6J3&6pd~ThkwIi#?s$8;SI`O
z_9UiSco1R?YBB46VBL0rCsQ@ptswz?5YZ(!9*F4C6vB?!5*k#xmfZG_Two#&H6sL<
zm2>Ii$CyZrRz;GN_YTJLs)0=A8)`{>*U10asIufXWY$YaO0{487?R+FPqM)e{EF&U
z&tEoj2;EVtF_TND?(=p%vtGSUE&8z@zt-|^KsPy2ie3zZ%n#ss6nv8rt2FpmS1}#7
z?^8fT^YDz2En+VzAZ;_~0HTs<81gG0-Mt~Zo(i`|4rEGNC6XgZ`EmU)uU_Tr#egli
zBYz<JQ!gYP+UDnS17N!;SZNBV7hQp(Ky><=9Oj_~YMX#<oO@pbN-EJ|(wV3jW3S3z
z;a22R<E5e;=zQ<KSdrJN2X4TXe<VKO-~7Zq^{7zxaqS<$!3^B(`GoLa^2EScbl}Fd
zY`{i{@|Thk7Ikpk;`UfkHeHAh@AA3>p6iCD(+2TtPA{^V<c&!;_pB_C^@$s<-+hl=
z_tQVU^~Q#tu9Lb~OE;dsG-W+X^{RdRv|CiXJvIbdAD&W4BF?XLd*qLRJbU9@JH4ZZ
zZCyR@J2U=h9Lls7UwHr1+by{U<pFG;igiChAEMZTY_jfP$<LF8_~9Wt-!f!cK9ZgY
zD9sK8t(S~m-Ws=Ts&_n-4hmU$Xoa!#V-5-#k->DyWB2fZXW9E*;dAA2reV9D8gQHC
z0Xz7Gfh<ZRG<ILP?|q&4VArTYo)?ch%Hku=e0N~`Ndn`jH$0P|*9<SI;V>V;vtmRg
zrGs=OCD>&|{v}eZ6FCOI@>O;j)2<@w1xE;6kGu=hZr?c;ha}&<Z~1@aQ^vgoPodM?
z;37nv5{^HJiTcjjb`{Y4a@T#f2meu5g+%&U9^YudSUkqAP;m_kSRR?A@t{rf0vpmN
zf#K1gKy!yTt*St$6B3fn3z?A#al|3TLUPJ!=iRNPFY10zwl@Harq3|K7ZD($KX?IF
zXv4W8vV-|%(7N**z}q0GeN<XK`$wg*MtKg*!Ym5hX?gh60rN06Sxq~IKHj^1XVSJE
z|I!qAGB$)B6v_|W1W&5zI8b(DA-!ivIPXYLx|5{v2|J#<S5EKl>tqI~h<k=@2=(;*
ze@!VE3!ym|dt=24(oNSkY=tEa<32?tfPE0G`Iil7_meJM=<1(3!J7)4F}qMh@Xk-P
zwT9*+5dG=}yGy5U+Mn!YV3z2OCWI!uU~OvR^TVBYV5IGhcE<1{-&E1k@H@GEv})Ph
zUX)*}yStw@5@n!|-Bjm13A+E{EDFdTKCfu&ISb3}Rt;|_`$xh7GOfI*0{zR68HK_*
z#_rrN;nn#pPXvfgb5qFvQbqNv#w~&>9nXgM*L>Ke0~bW3JQRlaMU~sTVY!kxghYNW
zpRZ{g(XNMiqq<(K9;F(zDQjldQpyag(xYpdIh0Khl`=~13$z+kCUC|l%!^iPXM$78
z5b_S`Q&S_4xRhzKE$pc0`sdbEsO2opXw6=HqGZdJV~<`Tf927T|2JEuZ>Ku`=T@nL
zlf8+nv8fX|wS$S7im3(z3nM!-BP%-%B9pMIi-o-twGJB>C-}pq!_3UA{03a!fZ-dk
zdjmSm?BEYGv)CIjeFJX)#%$g&`#0e525kR*;cv;m75ztXm49FRTeU3>lZvH_jp_fo
zdDXY_&i}5&=?$a$H|;-Kk$j`Lz5&ZOVDbh`X_%B<jjT+KU8rf8M4e0xUH*?=%e-;@
zmFho*)Oe%)N3DN5{I|ltT>nA-+vopO)ZfPbD#MtDN!rxY-QLN>nHsF<zinFD+ld*v
zn37YA@vt(pax!zU09n|Xf$VHrU=CS(lmGdmij$#(gQ*EQlZ2s-vne8zf{M5XqlBxC
zjgg_9-J4}qEG(VL!GHf)n4VnC)X5p#961{!3p)!3kd>2zje(Vs^FK@vE?}xbE@SCt
zN)DEUl}XIh$kNb`hTPi0(8a>p)R5fmU*$O&*%-l&A|UYQWp76jva^AYf%{M2`aief
z-`)}VCp$AY*MB}b!p6$Q`M-T#BwJfL_9OVwk;$4;V@i;_)cdlqIt}t#G=yK0=v1lr
z0#S>nvI7mgk)D;?Pk-X(YD+$*tOV)cuGMJ%l9UoxQbK_mGSx*#PZc7EVyI1kw-ly*
zze53dE~5vkVjL+IK8YP;E?|lwN{7dZM@WMbm^7eB*?uH`D>RpSKaELy&|vum^7nG5
zSU9zk=6e815hDwnVafu<f~v&lfHGAAia~%;nazj3ewYyZet=TKC1!`hry=Re&Vs1a
z``qu{_1}g(ed3o=zxtVQsub{L4v1LCnLF!HhkoZ=0*FL^9ttTvs%cbUq%8`u!X?s#
zq7W5kl*Y88pQniT(p#V~O}cRGRZ?Z7U<5+cMMFu$?)C?nA<|AKT$js#rNC>_zEWDK
z)X<cYiI^`kz$J3kmC_rGCDfRqveJ<%!>mA*)>W_jAvjlyI4i}MeEOwPMPbrfkM@Xo
zzFuYY51qVcIVO)^WeUd^>=docZ_eG-bi9Fl%7!SI3^9C>bzk^S>DXE7!ez(kF)Qjc
zG-PBe>CJG0s%t2(ptM~{3=FJoVoG2=b?d+HXqZ?F>-P07Kvnj0);5v9)_-4s5+{kO
z3V~4+YsO{$4j*3?5?Ih{{VtfcuK)XjhsGsCJ%=8qNOl+k3unB^p(IH_mAIdvn2B}#
zy*_SaxCWD6Pj~B|y+6-C{T^P>$Z9r8^e9@BF0P60?{*P{g=cWzN#xCb4xGXr37lh4
zVl>4BJpBB1RI&ZIEoy%A^=4o&lZ*FTIbRMJbUnaRowIk87Q5vW`d4pi)&w4sOYS0H
zy_+jd>s<Aeiyd3$`d}*ipm++trx;sw1o=akk9<u;l6=*T(lV7fjkjMeJrJ6!@et(K
z<v#zsxTp_dz{Xh#Z2}1y5;X)Op2Ah>2w2OOP8W5DtzsmKw1(7+kR-oXXql42fp6(E
z{rEGONH-_xwryT@c(wYj+^}FGIJ8V#Q*|JThXF--CA72r7<gWfwMx~5z#px=EGIF~
zog7m)RHiMfx~<ZpHT=VrXzjzR|CL|FeBz8iT_NAZgdN<|w@({N{4e_nk;}M-$xB6F
z<oS>*m)=1&QS+Et-D=wBu9g^I+w$u6?|9aqk?LK3G+urZM2u~O^zg$th7}?(5;Vw-
zN-YPf-G#NF4X>j4``~yE1+aM#-5~hk1S40WnwXZOny&A-ru$tD1n!K*I}R5w=CGvd
zD`G}92whzPlNRL*FiazMs98m&e@^C!5mGkhmb$E%Td)Z_J*ANXy)=v6VIBtSTV4ou
z)FVeBs#(MGDV}{7T!5vMnl+de3Iz@CfF)6h=oywywWtZil|d$O(pR*mqx!&`KhAJ-
zs|w|8?++`hQB#uChMO?vYY<U!duVwyo~~{h_UDc#Nx}(ZGO{o(D`NVBh8lU&(uZ7K
znc<0vqo9V%phwFVb9q4`pS3nO;Att4QE{d4N$8W%iDjS)gA#kM2^8eIUVw1*Q7KvR
z?=QHazn&Sn*0vriSu5Fs6Kc+evZ3z$5cWS4nn`QlNn|f(eBP0u;nkqQfjeHB?_+hV
zK<z-KF3|%sELgg!QXI3Fuy}}*I5g#zLRaHsyOLADRX}?l37zTxa#{$OA;u3<Pqx$}
zsbuKwRY_DB{Xo6j9F?cC%2TArb7Vdn+kN;!c$^;?MJ_Kvy7%6SC9sKGgTE3L_E4(0
zqIi<GT)@0Lf;D){CeBR&y0S?6_UQw?7&K7h3ng5|$YFtHme+F|bUq8i#3#&Fo-{V7
zZ;s_I^FmQzWxeOtc(TgO*M<8tgELt}lFb;t6BVxSh)l{}JUB2@ZuYzNz}%_Ti&OX6
zQA5mu(N&@<4x)nXeU7k0tq1IMAFDA0Y6=1Lxduj0@qCNp(aj`K!L)I@8CU4LF=sxk
zI-Io3k0S#FIRw;{tt4mExi^i0)II?bPNKg`d?Bt(N(f3;a;qIvr$J45puMI@YQGf*
z*FiqZ*PNmFqcZDLOzl0#8%iHlWj4o(<QgSnX}r&dYO@MApUbQ88|7P_Dwa-jwQ&ul
z2R1QZHLs>iYr76{CF7)d{^b!F_dXyDaTu4|is9S*V;syaqsj7wce(i1(H8AXwK-q!
zUvW5*xSR78GXcb_3ZWbe(6i0__{Y`7W9SPQ$+j|SXCH|2<$R^|u1)bHtHF5^Ldk<}
zk-6=rguP*jwN0mTAV<*N&&tk9!Rj8U`cCM9sl2A4rtwkQu5S>9J$O$I>^2-Zg&;Tl
zwBl*|zouA8+ghRtQgbGrR1*4IB0L)G;H=}QDGiCP?IHizYlUqpVj`S4uQI0$ylDBL
zY*C^;*ylvLqlc*b%5E<~^yXY3TpcHSh~@@rhuRehu(F5kuwC$EjJXiok%~RJ5T6<G
zC(Z(e-Lu%;3iC*yqiv}yhK05--z)H%5*5a+kRKS31W$0+G8j9!;USkT%T0uTV>kP!
znntVHCA2^_Tf#@-sd8E7)8M>KK*h^iz?$5LJ_3y5Gw^c-pf%~r!IXaiofXCF-}KdB
z7uZKl8glWOZj6~~&4ppGVLzi5Mk0zvVHoIlS)*X4r@Zz-xbZf8Gv7`0=m<f)y0J|$
z!2;XrF9To`iMNd}-oU+;amARWls<v$k@D@*(L*n0)5^@!w?Z{-LNq_c3WjL7p3?c0
zAxe@a3B_~?vX0%FBjrDHQ0&37*E0CCvBd=LZTXK0%#>s^lqPEki8*MKT>W!Jay+TP
z&yz{Ry0yif%*EMrYXm7#l-bObB;!ECF<^#MKi{rGD(fUAv2V*jnmzk^u;Zc@;UIf~
zl%s>vCisxXwHsG7&-S@T8Rf|;7g>yVybZr5?s=>N;Y(UBFNaTm>U{9>GPzfL$Z5_P
zh#k$7>ngL|?Fj3X&z9;4r!K*I{3+RGrpf<-M%!ynqyH>+$`9te)TF{K*qr2c;3k$U
ztfWxdZPR{|@8>DoMBgpZim;)1QC0bYckQVt(faI`KmW@&EBss@YTWO^CL)~qX<I%I
zSGT9;oBTSP{I$t~uRh+77e8->$Gk-43Bp4RMA#u^m|43EmysDspLc&=yr_O9VH*-4
ze6%=O{ep*Npweqgy3#sAuW*UmYY{XJTZn=y%8VQ5&Jag@-uZlYGaNRsUXkV=g>T6`
zWASzRSC|=@E~Wqq1DQ`Jzth|qs-XYv;PcPafqh35-eZVp9zxefhyq99Z4tXR{#B1o
zZ|;rAcvtH`$HU-XH`e#_m2R3x_}(Kiq+6ThQ*29GT0LeC+@NGmF};`D>hbe_+|;-8
z?|R+6Uq5}k3*-ISd<Fv=(`9uZS(C;6V<Pq}{yegA(=og~ezR`gLwZK`Eh+ft{&vT3
z$3R63AI+&(U#9$?BWrA;XO@jbg5vMfJ%gXW$8qrvNVNAfIEU+jDYmfnmm{r@IDpjS
zM3fcH-sL58<J@T9W(f0b#RXp?mp^(mC-j*{UnAogCE7L6Bd4h!$w)%>&%L-=PM4Tx
z20!6_&}HzL_n2$74#HX=fdyv@=2)2503T)gms)QhBsbfa<!-}Q7h1B-j~j?E4a02c
zKWYq=g*zDw(;d$D%@}yIngL(`OVR+1ijW`ANj2iBTLj>PZOB-40XpSoFC_meb*u5f
zC++2i%#V#7A8?~2*K(aaUeZuRuQ)4z1*r0MK#O$T81{L5d!`wVljn3W=c!y<UPoTB
zMP7MET8TyySg-RSnzT#GKCJbEQ#*%<YQ$l#g@SDr+}4nrtxJ^bEhouylt4;Prmsx^
zw7#BhI1?@-B&u@N9>JPi*0;l`rGgAz9JkIS<;(_;@HI%FS!dKddJ)kSe1E2XUU!U1
zueSoIRgn|;(y<zd!(w(_JO;hIzc%@iRXYPSl2!X2TFrz!EOmoBg<o4O<D?w*bO`G-
zLRqlJAM*;S2K$fbNFu{^zf_X$X}<2>JB|4OC?-C^Z(WD=N~T<{7Vgd#Eshp<_8xmf
zRV3Iawjah6p8fCO4FD5ozozUqr_NTV?3R8rK~J1u=*^)As`7SRiPZivT~z*a2O}@y
zqbSN<b?FvX<R5tPF%DrQ*hb``i?Ap4V)}>_cH6!Pxhg3?nHAZC@@KB^!*m4M3L|P&
zR(>*Dqf_YRoq#|_4MHPgcypv^g{1`t$}^)~2eupRYl8RQ<83O&Q(pe-{pq%0+n-;5
ze*W3JT_2p^<i@?LCs4h`Q{NO2SR)o>r)Dob&AOSS{r!6NeX&K5r256vpV<}A)BbYl
z;QadMW}Ykd{JckB^ZxAUY&>)7E7NA@LjX|6_wn-Lr#xXA1%BNtB<zct^MAK6XZx?|
z+WuKQX(*YRAu@foGcolb*JdUMAG6J(i^!yG>1FyCM`Thb*JdGSBY&eR+1tB-sjP3W
z>?^$$GP8gC`nKAC`zl7R&BrRj!70JX!O6|S%*n;g!NS4L31sDF2a1S`aBy?8iwcnc
z-&w#d{p(mA7G`elf6mAJzd!U(k>%~9?e6({^C`t>7;hwnf%nVlQXQ7Glh644XP+-5
z`bNW!zqs+J9IKIeM$x{#kF8FFA}ZBEvM{y%3CDQMXMj8hWs1*P-Uq?*<!4GXPH4bl
zXnuHIogF!%O!fxZe(c>baI)~nUDp0Hf5gPYDC+P+Z=_t<z#;SS0YB<{wlETv9T+=j
z?#>mLfcGT>9ezI!FBrJTG3-<gk*HxQ4Ctr<rg_9F5KI>VQVM{lJi<8;L@B`4fM^NT
z4dgWiLD(lZ1>+WkyTvk`k3^`-wlwHq56YAabDn^14g{9~rx8DZTHj*kMlJ58*MJD&
zcQ5wT;(@pg_^`#^3eN~6saAxehSi&b&?1;3oq`P{hw0BFIRTL5VYvj@O+kD03DlAi
zL%{XDbHM)%E<?=?`=<{AA4YVG8z0tq3&w(Dc$$~T<->3FjyFEZUI=_Z;uc>k1Y_Va
zpfUwim4{vXSiviVWC17HM~)9=o*h$Q2Q9Y+CCm({#*G8_P-rhI2*CkUCXmJfQ6^~C
z05%KKypNt6Zm-wV0a6DhwU@6Ixf)bq0I37L*oWePwFcFK^xa{ORb0r~Y>rrzDv52b
zW4zb>O&2^uO!!dyy%Y)_T<(cTV*%G&P}H4m-h(Kk!fZ=`TM*|I>TRH1E!p3W;6qXM
zMggmR*M+~mn633fr%ezPJBJ=-8S<E?jS!eQg{Eg2Qof;o;{W;oUr7HWs+kE+MNLDn
zCo^OklV0M=o&GCbX~+kFy|PWiu>}xNKqBlTMTgBcKt~S1GJwj0ao*xfgG&|uxyvt&
z+WUhECkS&3ZdI5K9!k#zA(K_+1R0BW;M)Xq9ozqZ^aCh8fdJ4k&eHNV==<6JdLBY8
z(gXSdWY-qHLCe)P$T(RLFae3G(arp>*`ayKX;_Gb8d5+3#SOHc2kjDQ$c#$aOXGmb
z{LDIbi0sTWkTt<v!uCI*QSH5duX`!(xk$vY{kM$Rl$Bj!li%*}9iRoCS&jG7{KF9C
zC~xWaKjRhlAR;o=nhqfF^`dd36EWoBGzN9$5xx1LxB|X3%l`VWkoQ}#MR}0yf!PLR
z_^@nS$lOR1K}dTVQirP50msOT(9Iwt1*qmgB?Z_~VZa3xSt*QaFIu@U3Oq<l0j@dV
z)Bvs&4kZuc1V%QG-~`HBh>n_d_}ouyxX~dkViEsdiwPfavb6$ahC|GzNoRo47Dms4
zDCzs_lj_3xvlPV93D}dLR6#toDC{7!DO9gM7;Xf!UYyW{|8Iln!xl^$gmE5ohd!_{
zNQ2u-n~Ee4J}Hp4mgEGYqgRg^CI*`S^kW*Vc^`#B8oG9(-4-PHYy{M~AlND7KoFj+
z8wh%f93N&|Nb>4q8u(>5>@?VtK4fm_i9k&U)W2f@Jj??;9S9C~YXSidD4$-4ws7&e
zhR=QV!wxB-+Yu2JExo~}ZX$;_6Z%;smAVSiwu=rQ#5je58c=8ep9RsdMNA!4KBkWh
z*4~sqkfaxc)hdKB#olEbvq(HIx@J~Y7$GHG2R!pn=9Cr!P_qqdPf)8*5PYur1-Bvc
zbC&j=TVEGqu~)qJRde4;h@_Nr7+P>P2wG(leLj%HAQMRHB|Lv>Kvl{$97d{K4apv)
zn+G_7_3qW1GKq<;I_$G0hZi9q>&T|E0`TVz5a)Udp`Cqg@3m`NUPJR5zvQx1y;IDF
zPF>=Izkj`7gLL+JaOB@XpLZ7X%_Ae`7+%ybJ7Qsm5A6d1b9}=PMYmw^0mcSUQEA)y
zRKMdK9T2nFhbN_yGI!RinRR>@(;#Q}?wcf4NgShhWqw*O=u^AnTHG-~bc>GZ!aAE#
zVDO7Qa-vlS(3z<^;#3C`ZE?2RHS-riZMw~TPnHO69H=x-6CqrjVRXiA0^;xBa)tz&
z!J!?!L*|RtIb>C^;k*w=yieU*RVF;1(F2(sIyHO=yFu?(EAoYKo|!jm&{u*SpGixs
zX477(TH{netEK50I5D+XfXtkUj;cw6FPf1x>$yuRw@8twc8Of(7acW}BdnP*tQ|f<
zVNwK_wMaRGI5o59uYi&(eyC49WB0zp8odNCM;jiJHb`0}9h>6w5Qg<aHb}mz0(7Lw
z0Ab?-S9-h5^wEZvNm68N-bUB8Xb{NKXz9LtjJx-K=@Qky!<Usa2rYx93pKsQPJSGZ
zMzcYYxOGPr%V(H+sieW+rwEwdx->Y=g3c=r0}VIZ@`yu=_8~Hg;^}-RPW0`|BU}$X
z8wYFI^_^z{1NCHTI0G~IDcxqbn}VumJkSZqE%z%_MRpm_Ue2v8Bk91N-}--G4`Bg6
zrE(X{Ft&Jql@4dU_$$~UQA2BwFFV9K`FIlF1UNZ2ck8twZowt*A#>5F4{H2gBSeQ-
z6>b80kKV+A2j*O$MD0M9?_`D6Boe+N+mhu3mN6}Rz`8*yP~JOvoA<;5{e)$|D9Ond
zblz0>&Us_S$1V+zf$9ii-Jk}jzfog_cL3A1g?_+d{#9|_cZ+<>`|7~ZxqA<V_BUiV
zXDlh(zVD_|H5oD7fB`Zx_((#v8yI{@tSxMOu6^iDg`1CQMqhbq*$h^){8_&qSu!I{
zq+P$--(JzTBD=(IX8~V-chQaoR2Xbzk&&pJA+|yyhTSD@s-6+Xk#mhdw2{%2M1F3U
z*@8iJF7_8ToFBUh2*dt4W!0Bo(o-W#tR<eIO4y@?ewbOK6oAB7@u@*8&dfK{>1365
zrV{t-!{uUf1=M|g1W^S*)sOL54a0<^7xbI<20Y^YuL)Hztz4-rm{T90M-69b_bLlU
ztrFWt&y)=DECBcQ?ZY|&lNBJQ3UA@>I<mQHjL1_)O#?coM5C^(QBv>eHBUJgyFrNj
zAZyl&$1zvx3gSgeHMF3hJ)Sdl)Ec$e00HM{+inbXN3~1SrcE95j^yov3EJ{X)Xd`<
z<<UdLI8-dDs+b@eW~=JVOr}_+3hu_yraV<2O@7mAb^3B1lqn;eFg5eBrZk10$Xor>
zM||Ao)sI7}TRbzgi}7ml);O7JOqs_o<B*24<&4Y>X~(*gwC>Tu<xO0WRE4V5O)lQy
zc4^05?z~Ib5unwG4bE!WYHHWiK!!c7wy(r^Y3(h6hze%av9Aq?yLp%%<2O6NU7tKc
zmGt9lG}?VGb+gm~9YGhXL?r%oj=evlyiFrGd3M=$zZ?G*ec4spcgvw|pb(DVbQ#$;
zgVbuV*aOUJf9Oo29KU;~Fm>`!W8Hvq0x)!bHdAPGn*#7uepfBjq1qBrac{v%mZx2W
zMg{q*sXQv~-m+Z~7JLAQXcpqUuZr`cTh#b4zG~R*Mg1&OXt2G^`1Sp(<zB>5ff#f=
z9;r@Ly)gI<1us;|L~td&pXms#73>lq56N?2*+4%m)@Rk&XG_DfAlo~+kr+Why7tcU
zBTr^DMy+XSPTp~f)-!;^a>Usm{oQI1(N+eIxPHZ&EBT_G)-(~PGf({V6bv34md>yY
zcO|u7uvNe1*~2*vI0J8QlUO`X*LfVGHpCHyRMmlKje0fmmVQ54r`qrM+%k~=7usHy
zPb%8eC8SCLq!bFY0ZIsCpXaou%>}+5xlO^N(*%4%5R*DF`t7wsI%Oq3Vt+)CmL^m!
zmWflepR|2t1C@z+kO;3mJzJvy`4e;9K=7cx;M_+MkR){iU7EmcATFpBiwOrQMS-e4
z?f)d!CNNYmZ7lbz6zBO@65>*lP{ub9&hwt`=sO)KOU!}2+c_F8s5nX%E4%YO=i%Ou
zY0U0Ih#%zwEtyTvB(edRB_7Tsl)pb`ak?J|BjTr|L7R`0SSg{_V0t9CB2Q#??33a!
zLhVTDK&y}Ox3=}P=lEjjaNW&OC3W=O=!&<D`R+N2w-5}WUK9JpTQ|yz_m25q+pf4k
zp;1HHId9%r;zab}D!?{pJkDd_NIj5|F2-ec=B=&y*NTQ9Dgb=;y3d7oX@yZPG9O=P
zcgqPB_!UXu(;jv8aIE@JYsc5ndc@mK^DyQzAE#+{2_aIeGTkyBKXi?91gO>iSc~oX
z+=knGb5gv%=U1=Q#^2edM~)o0C-)MD_^y}YyqIyy?yTlaH6I+c=}p<5nd?=Hpm9Uo
zrayMbQx7L!6%TCXwfWzF-bID1PUpHwf25kEa}PGOXo2lzaennt4DoXyM$ULlCEabE
z+xS*=TO63iw4rDR6BVR*-P&EShMxw5wMX+D>O|GYp2jL~8|j4UOf^m5u)@dNu9^nf
zu=SqXI_D2rg+ok^K1G6I>R7`g8|_%o576;}tXH@lpU!rD`%b%MPPMPs1%sYATDPof
z3-TD>q(CwJqMFVgfs-|sygIrwAWSV{oXtxdz^A*6=<-a;tFB-VAW`=@A!4mGn3YQQ
zl+&M|1+Rzlwxu9!pz)gnUFo>nM3mK-&5I`Z@XlwPfogPKir~{21J$l|D%4%MdJ~f4
zJkJnm!*uB2>t`%;c$uif)amZ)XRLD`N!9Jh2Ifid={13{s2TgQ3FYr+5Cba>zDf0X
zZgf}nlj8#h_B<8V_4~q;5tz+#(D;go11n_$689c18b;rJN0g|XI>f=&0Gj6XSg75w
zL(i;}pAlAJW8cqtt8&A&`Zop=(WQbid^Y&>)22k@3`5Rp3`Ry79M<{D3d=Wk+~6fD
z*ElC<bARQ0DjOK}o8M!elz)}YT$83|puPw?w`zcQX!ts3oRbTav<ekLU=o@9@+ZN`
z)Sx06WzM<*TP4tmj_5&umsW_Qw7z~8E@>N8#*(D8u0C>8i4npLWUF?YkV-Qwn=zA}
zw=t@B$+Am`FHAShIAvlQd9EIi2`r**48W!9xC=+@5axx~S|%H(9v)AaVXR#o-S{KD
z%L8#IC)H7P(kHdsnycgmanewa##f&8LR5sukBq9doco-L7}kr?JDxftK=*hMj40XX
zIAw1t|C@#jLZYuW*>@p-4L%LVSS`gyJ*I=JXx3nKRLTEqmfVMSE;rb$=6e6us($1&
z0A8;q-Q(V*>Q=FQ*;$irWBpAmojZaKG^R=4XwE7hH&?IgSvc^Eo#yg&=47iaj(qiT
z3Sl>WTjj}Mi&aB7q~lU*GgNGH$EzgZt0pa6k(%BM(QcgC7b0J295uZb&&Zfee`$?+
z70cc6ddo6I2PTKIq0u0D>M)8os`+kLW=%^p*P<@p83E(G_Cvf84k1Je{b}RXPuo$)
z9-A+F3{4sj-;hS%jbjGV=MT=1(g{9u(OSf|UbkX5U;RKDJq4832NOx_s=dtc)4H%B
zclG*BS?|{Pv>-l0*wK^JUJa`5qC%-?^4RR&l}Sp!1*lC|T)wJp#%eEpi?kxSMUZs<
z!veNWbS_`ym{7DIL{&%M4a{hU_A$&BVaWaQa&FY%=_}s2W5+keVo8m&wv3YX(fdXD
z6LWz<R%6in{<YdYTeHq(hL-dxfHNIkI#KP?A%wGuOj+&n#nskI{nGw0OO-y=%3{r4
zM+*Y4%4oN1^yen#opoxfe#h6xCh83Jn+(;bT?TPh464)%mqXlE+l@u1wwXl9=}x#b
zRiAuOAl?>nisbMUZpD51gqNos)IP9g7s#+HUG?iwPz93ey8S8!&U3=hvqsfN+nFP_
z((`0YU32ix)F#JSgOvB0L+x_A)@dny|7&thLqx89_Z27T%L9qpWu(7SHK{}QD3_M_
zS(+2Cr}NHep|M}8>XX&AxHd#0e@*@=BlboSl-u5+3#s_E_qa7U-sQb!7uYo2sNz|A
zuoJqwejL|+CfKzwCOy6i8(a0!X&XQmWdc|KVGm`!2C94~o>|afV?$6C72&!akKbm)
zIF@D@NIxn2@KoygXzJ*PoTuqG(^;6SYKLfG^U~;9;@bRu@?vj=;B0q)GUG0x40Pb>
zGosIYK5n<m2=#SjGV`v;UI{h+>OP^zYNG><Cd}3I?06%dmTSy=e}Hxkwq*>Dr@(v5
zZj1GtOro(&SqpDkZHvYJBmq%d2uRIw&MQ!)Lr@BXv-iC1opc>Ys9?Hrmr6bSyu<5C
zZm==Dl3W1*M{J(2zdnrG?vJ;1`Iv*FwsS215|Ix}fQ+$6pmH_A2_(bb^XjWw4DY<V
zNZxp-zkZDly2IU{7(JlPhD#0N+E{m??mC>;*yGqU?LogM8isq@xkKvX1UUI9w%DIk
z2Cp^4DYzJrA2WPR`kmyn*LMhfNYTCL$xY=mZhQb$AHwy^RRN;00U@@}!)ef{YyjI<
z<AYA7q-ghVOkD`)t;z=}N|kKD^;YKt?QzmJ>2o$X%<yrg0Y@3f;PsJN_ue>*_4G3G
zG=tq~gOtI5VfR_8(mL}s*@L7wex0I8w?Gf)dQ`-Rwmzc>l;6v}snJH)B>vf?N{Df0
zK}EPkID5Pg>a^tJuroF)8Rx&%lPaJz>gz~g#@yePLA|%NfkrV1d$2pyur-*Kz_MW9
zq%#=_onU(WqA^i7ZPTspuMMcG;gt?1C-NS1KYvuQfTn+3f=@q>CIbhy%tnQ?KK92-
zsLq^?Js!2w*4^%Y!hkNKZkxP8SS_31>F0*A<=l;qQHqubgXs9oo5qT&)8^3noH3c_
zm!!KI9`!6%-97Ukmkn-4h7hiOtJN|;HL%zvv(+q8{!rqS%V0X3+WVMRKRS{MtUNk1
z+^w}VUZE{{)y*vFi&9<x6U<p>xm~H+<&yh`5D)NMyv&S~bizzzZt@n$eq8G<j`qB^
zPYfnXh3vPrupCjnZdp?Z7~~@2(pc~Nl^LzQckZJSvZ|*0TeRx=0;1$0EO)LL2|I|j
zcJPU9p4J9HfqaS{kC&9`|6-|1Cq!`po%>*)TBZPi%3FDIe~oeCFd!=Bc&$LAQHOH_
znbod8_4o^gcum1*^)X&=a#e{S83SeJVBIobWd(3<LCo+Oucv6#cu*vH>rdw|X-?k4
zTnFs0k4H(zP;h;IQ^qfn`YHj*lt{i?pa;QQj$Ie>-<rk!EDqnEa-Yh|!MlJ5T(9Pv
z<!gUkOnxYd7j>u?;o9F+0hrQAp9QnMCvQ*91t}Ns`C5>zChKgJ_+18IcJFn86h|&7
zax3eh=dNe8j@j%dYpco2{>$e!PY0au33Pk@ehv9ZTe#AeSz=t{udC`64w+e+TRD?F
zm0rWns1=U<g-zU*B3jOxu9~r&H`7Bb8g!7P71aVo{kRG?^F90u9%oeDrNq@$woMsS
z*()XtvxK2l$=c(}D5Hi;gq&3mm3!9v$)c?sHd@!(<Fk*+6|hYZ0}D`U*C)!fb1dXu
zjiE}`t5th&_R5gQ3mQyK3kDp5U3tXTr3-EWlobz594;s<i{PEwiii2(<bjbU@+D))
zS*vmm7aeQ3cSs)HMOk*gZ?IegyKqb4%R!0OpISp5taWiq;|HzzK>!_Vi)Oc8D3lCE
z>!mjEg_ZOHT#L3sDALNqaN~8Y9zg$HImp$T9$K5IUTCo;nRf!DBGlD{fkfP1(2oG}
zvfjqU52p+BvMz1mv<$d}OU{I<q}I-C?k$_73v;!uWYFfd4!AB!79DGn-U<X%^OvlN
zZ|DSdLhUhJ@OIG}>=PbUt;1A0Yj;=$bi4}BTUX91xeP!}>(has#bxlHsIStbq6PqX
zH>b+7kXi%aY#;~oV3MG5y~h6tr321-<mIme-}c-Zd%fFLN}>D|P^^R?=Acp{$B!a^
zp<`GNnYjmgA@L~{<5|FT2qgkYZPj8-O&dCdBLOc@n!8%U@YtPRPtT)G0bS%a0h9cN
z#1+$K9YSG%?)yNe_IDmxrEs-swzuR5&Ap3GUf=)jTKjIr0bbg~<_EU$xs3wOZAnK8
z4ImDZk({3IGZ2G>9(fT%87><1%bB4xK5+4x1*|FVHi{XhSs-RL_9AU67-H5&+1(}q
z{IFE*2*(21aPLpMMzaB_y=@|AkE7!OKZF9uw5#@^X>}mE`dlk+{50B#qCojYJJQ0r
z?VoWpSkhRnALmk=;L$CW7%NbynnslRv7L(e)K-bqP&1X5lKP9WXTFYBx*4>Te^2gD
zO`5cl?-py50<qKd%ujqOOJYoIU}p(&C<FaJ-CcP&RPFnp#YDz7ma$~Z62l;6&oX2i
zBD}~_$dc?#mWIhr_OVRzO3^e5S&EQ-Eg5AgORqhekc1%_S-z*<-+#Y9zyE#zxzBa(
z>vMn3eVylXo$EaJeV+4-4F<MLo0V=z*Z24;d6S+3yZlc2)GM*ePB0D2jJem>T(;-9
zWrL2qu=ad%!e7zPu$Xs`VqhONr$!5PA_t*_&B7ekoL!z6$aLn~HjcF~<xzh)tTDT;
zKU)hrBCB!69oYX0Zc#&^!iIq#>6BU!up$j>LG6z$*vv(q?|tN~Pc<#~us<>cRufho
zmOI{76;9>b=tnUhxlm#+p%TqkSlFEoy!*D<gw0mIY!#>q@78sH&^>J7@bjPZpB2EA
zkCv&1Kiw?<U26mXHU9DENj?WeVPV7T6?MNYVGX~`zV(qWm8y4t%62!87O7?t{qj^P
zsPXK_4(AKq*?oLxof<vu+Wf2HI&dFF`|JWFuKIR4$*Zc`+$?P7ph%{R%4EKC9^bKk
zH#`~u*A2ZeaMtlJ&6O7oLqwksu(Y_o$@cAVT0`s^EB64`^}hn;H{x-AQ9BhlzD`cU
z_)dZrO-`_j$r@zqJaywhZWLr6Y*`zBqjr4qa7&r5A?|)j*JQM6)n&=+U<>4ncX4&o
zH*lQS>)k?Jnr6hACeh1tguraXbu3EWK&A&-8Do_MO!p)8pTL~OY#yH22IUhizt_3_
zDy{%JBoLEMefamOY6CG1w!JQExagF&R39+Ke>RUXXwpBXf>k+YjMmtZ=d%Ryy;ivS
zB_F31`lO2lTZ$o1qfW!Gp>IyrObZ5!7S+u^{cab?@iqaDY<*&_c}vX_@%(Ejy%sj-
zK<VZS7C}2~9Nwp1fi`xr!*z!r71>v6#r4FCvE`T-CK*RRWHHZGgzN4wPkOEeao@_7
zy@Li9<n<PLHo^3EVnr`?m3^#|@=MIqg-2hzl4KSuhwXqB@2S%&!1+>8<bIlQVI|BV
zcRH{uNZ2mR=rtB8C==ndQCu2Pd)y6M`zn6!s1;Y$syUFESCBXD(e*8bHK2rtmx4b;
z(6UdZp6FMEAMW^3KbhmiIx{_MV6LQnECE~lQWO8tK)TqT!!CC@Pn(rXDpjQ)d>WGF
zT=p2tDSGXCs9t~fveYf}2hXf}$Q8?ZhS^e)GRiV|j-pzt;2Kf#%zOA-5E@%srj@GO
z4|f|h_@%l9FnQL8ar@+k>fuKWOpe*Lu_0r}-CY}pHx}il-QsEyGm1Ame=NKYulic_
z+-grbpc#V*PN?eid06gb*9hX2h0;nOyO{E%YwIJ6ttG6s2|`w!atfhd=Q)pq^va-9
zo#D@*D)HBb-5%b3yDE52pXWNbDwa03)`H%q^KWO*xN*9{!lTYDgRa8c_O9x9D*cK|
zPO1O6b|2ZSUItmMn~Lhv2YONt%j{8hGt3+bP-JsV3$Ai8&|4%TsyJzu5kVCJAGmP|
z*A0t3u@wOIxV<qT5-w^U%>6=-0B>iKNT<W|c8>hB(cBr1rbi;^fMU>}Ho)ZHSop04
zO`XpwZqoQ)j_`xAFUyWsW}`(fgZR6oftpZ)JSP4!6gJWRj1hS$ZoG94(&+do&=6MB
z{<bagy_cqn->12lbhQ!hCh>QAf2&#h3i{+bs&ss77>=kekF^*mE$KCwJIcnfG-lWg
zcp)dRU*dR|uk0^an%{=pTQTC311BqZe5jo>IBFvsutVSbMt*jB{gmbeZ?meAt5#5t
z6+u#0I~6^a&fpZ(oj$Hz;`o#w643Ndw=_|<5nl2S-DTMK6P>dxHf(^UaT=<H7GAd3
zKY|xa8kEyq;Oc?xyk-iIM(JR9WBVn|QD@y_duZXqYMg9+hVDS0en(PNM+GEVff#oc
z{<F=d2@z{LP*$5EH?W8lZFjcwcN*7{^>i@4L;DN)fUPXeV&LP8%C8woPhuWJ(iHvp
zv#&AYlavT1FheKfY#JW!>TJbvFCXE9@9I`M@fSyFeocb&ei8Uqevs4doh7?A3rh|@
z7sk3{z?Cl%VD(_9-7p?1UmW-R^No8afH}JyM)0HWrwGdmE7sN$ZYE4q($VLcioG~a
z+c=<5{S9=bWV~B!KS^{N6dr$Sm@PaGIm{ZKpi{!m)D_MJx%#_v$uWHV>z{Tk(_+a~
zx7bXnx*l63Mrj#p(g_ac6^0(TfrKFmZY;m6(M(04M{XIC(K*X!3Z88|A+9Q1jK`ja
zS;qMU3%!j!JC81aEp-ZOa4Le@=ztWe<6&0W577;GmeZ1owJ43P;6TXnc*JF1`E<d1
z6RAZY1^1g=J(&f}C72Acx4mZQ{i%!x$!y9mQbuYW%9Dy2iLBe&$p~d3v7Q8hj=I#%
zbN2-DdwumaG3f#vuYx6w&wH?NKMDfys*vfFtA=cn?k|+>HM}a1T4uS`-CrVlG6e#A
zhb5Ac0OX+&Th6~O?$^!W8RE!+Kt->WwDC_50GA>l+PiH3EjuaZJ^80mx`5N~D1>sx
zzoY`c0cNj;CDe{(pH>5Bp9FH`O?il6gSqcpqU<&B`$IaB>Ygbl&$9`-zu`KcEO4fG
zL;C;XznwYsshdv@vR+RXSI%Ei^kwqZcFjbp$}y!r_r+ODzRIoanM`+hLUZ=itKH2F
ztL<<1+LRvPdt<btp047^c~3Y2{l(MKv-V9$KhC#5&@0d54vj5W!6R?+?yD5^(ZQ9^
z1=mI-2IMG4SvQcSwfhZjIrg5Las%pnMjZagrdl+QE5+H7Z2%{T%i*||BR^pA(&%v@
zvZogP>xv--UFgLK7?98zT5xJewa|WpJ43DIy@61#-Mvfe=uh9xHEKhfe?o@HrpGw8
zdnqi~A&=J)wp@Gb(WkC4)<+}`y_A&)ZcmUUbQ0~D)87=1t%mdkT3`Hdug=yBQC=ss
zqObNtbW)+3s3Z|%-xqr)BwZ+*L!j{*@^2|ekr}X>YGl28ip3Wb`&)~8Q|Xg-stl%k
z7dp^Ro|-iZUti8!g~wYALL6-e=_`$832#FCYhQZhV2g~_(F0^;SZ|1SRI<gXC}dmR
zJ|V(-t!wj}@Vz45+LQ-vosT0^m;W*M#~tf;k3vrIkKH_FX-~b;+)pOhQwO+7LuPnO
zbI0EYl$}-kekP1&-hrk){iq^NlTW>jq5evl#!#bD8$c*meOm~HplOZDrh6Guv3S01
zh)uM7&a4G>)ovd{<>wg#p=uf{VK%epU$C3vGlQ73=|+a^7^<IlI0UsW<$KJ=C2AHz
zX-&AUj$xE#*`2^JmKDfcfVCip$I?em7>a;sk0w%^K~$3(#19aOsK~Vjl@)6<3k8)B
zwpm9`0-yoITL5Uw&;(31Q9}HHl87!+LQDV%6DkaVv`1I~KH(8rz-p31fB;8w9uVN2
z`WRpx;<5k&2Dy&Qis@KTyh{kS#18|H2;00lujA>~ZDq%@=p$H(HEF7eDnx{vM6~2u
z*?mZ`?HTJbW-JTjXeH3C*PI__@`b@@2m3K^_;tD-gGt0Jt|ox{cIq^6oj%}Bk%%0K
zJHIXW9SC5#?f}=^;lTq56A^ZRM^u^)ctlTDfMvzjBr9M*G8bO31IcrZZ~zG>7_tLI
z%qigj^t+7jF<G%0lTnD3?xgvP6Q&GNvYI%pCDi~!jX=~xQFGW)Fr`&O$B?=%{eBgT
z*Rcpvr1cmIm6Il-t~XK$*`|HK0ARG-Y@#`HvZ_fF%VVslIWtN-7Qa?eh^9p#G9jp%
z#WNff0?!CmoEFCQB$gB;RS1lg3E`M2Ba&Q)p+41FiY3iihKicvf6yhe>0yRjfOGRw
zEQym#0R*gSsfMClEkb!r@u(zvYCL^v{SMvEpG+{IN}kk1mYbU5#Sl~(iuP-IPj*@&
zsYcv@kw`LGX99c#!5o~83#<*`H={)4(Ifd@-_54KGW>SIfV!&i&sHMoCs#-Tw~fo<
zV!<g20fAr#*w)x)%uMkzh^n!m3*^Zg+JK80@qVL)$JMZpwP!aoRSB2QBElmp_NkNJ
zlE)f-eIb2X+#obMc_Do`iHW)e=XL!ogrIIkgcA@>ua&Iic74*6aCcZ}3|N|yK@7vk
z%m_l0yqjwEp!o5w#G7mkK00n8{j!7sgNbT#`gZX{+l@v7AqQ?KTa00_6mBMvSfeQm
z>2DDawe?Ilje>JsD{Q5#bkCS4()Vp%@$1hsQCBaaa?bvXZ;)Xk5sOZ}KWn8+?pvW*
z>*CT(J@gqVF1An;#v7@e9d4SWP#ni`s>!L8(0u^(>8sl#f@g;5eE_smHEg`@pwGA#
zoid;Pj1Zd9k))DLM<DNJ{nlXeZw*TRt3kFu8f4?%^vG8x5!wEzaQlx6v;L@1_>T&O
z|EO@~j|w&as4!7a@^k~Vgz#D18j4}O0h8|ol)q970r~FA!#os#vo^z)IMVt_U#Ly{
zJy|9Sf#3REHho_+@i~V$jc?_oO|(*u%SGy{Pvi~MkeHLe2h-Bl+4C*Zv;;#fKv&TB
z;5>R%N@!?lt8tCJDZ^gkhXvK|(wHY|NaDG2J{u8>7r7fEPP><42}m<R14bc8j;LhQ
zwHD4|s1rQ*VKy#npG=fIY@*MN;KgZU=7$@v3P?4Q3lZxcHZGYR5Q=sT#KRwKig(0`
zm@<6H_b^m*6v;KhVd%u1H1KE}Ek4Sno4(AVuiCi8WI<^YrP{BnAMH2)5oQ>QIJ>%c
zkWW&Qw6UZ<E#c3mJEv|3+PKto?~ebtixSw?GKsja<1`_*=QcE<DYPoQ6vpvk57oKp
zyi`zqqi8~NXLZ}-zG&5l#<0kyYNOa~f(Yk$h=+*l_{vu;i7Lqnt?sJTnp-cNqk=6~
z+kQGl?WLz_`LB2S@r6y!j$YymbL@k5JT05jdhpeL)I;QmUe=@KvYW5=)7d-Vc8^xn
zx{_OTBktBX<4bw>mq$S#z9_Hm58Rdi+Bn#8kVXctUjB8$C}Etjcq1*j(r@4wuSjF{
z)||bH|COLkd7<akdUMswiDBnlZB=|{s)({eb>DX_j26ca#Q6Cv9ERp4VUor>yr*|O
z!Ff>Fg%(zgt?!V*cl^N*ZeNb9jFlfk1|RUirWG^VmrC^%=c~o|Ht4tEyD@gKR$p&$
z%@@IRJ=FXIZ`PY@uKdAsW?;_e4E>Gvj&ayWd<p#QUY=AzTS-SLx!7`b9oTl{hegR;
z4ZW@>g-K03;tFQ6xoSxgZ&$hol#FyPs_@Rz>xNd(eZ!Un(a`SBqx^KAlJy02^Oezw
z;Ar1xWkK_m>6b_Eu5V%HTAG}j%W9^-b(7lz>fq^n<adhbujkZSo|V;1{B}Cl!8f~O
z+j}8k*P?@zI<b%)$-(GeN3Rp!{0rX<im?3N?4Csoq!R2xy+-=p;@jGn{HMDXrX7f9
z8yBKGWw1pjctdmz7wTh&h#K(Z1wv?V1V~qPVgFEeklDD1uRcUqe&Jm#F;zoeY>{Bl
z8v)f7UI>UCO1jCik;4_lOc3fl2NGl2RP6M|v@!?p5x3;)HM|dZ!JSyWXqMo1;+Eu;
z#*@8W>rO}ASx$|vzNSl}OVgrD)%HanB8$QVLk5>Otvma*2dTvxOPd;*#1Fgwof+a2
zI#NOh&}r8RS+AG=r_7-9P+d{YgO;hym>(C)tN;H8w>Lts;zDoYfDdeeC5Wfh)y}d?
KOB-D@X8jK(isR@2

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/algebraic-equations.html b/docs/2_39/stan-users-guide/algebraic-equations.html
new file mode 100644
index 000000000..171d090df
--- /dev/null
+++ b/docs/2_39/stan-users-guide/algebraic-equations.html
@@ -0,0 +1,1342 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Solving Algebraic Equations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/odes.html" rel="next">
+<link href="../stan-users-guide/hyperspherical-models.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Solving Algebraic Equations">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/algebraic-equations.html">Solving Algebraic Equations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#algebra-solver.chapter" id="toc-algebra-solver.chapter" class="nav-link active" data-scroll-target="#algebra-solver.chapter">Solving Algebraic Equations</a>
+  <ul class="collapse">
+  <li><a href="#example-system-of-nonlinear-algebraic-equations" id="toc-example-system-of-nonlinear-algebraic-equations" class="nav-link" data-scroll-target="#example-system-of-nonlinear-algebraic-equations">Example: system of nonlinear algebraic equations</a></li>
+  <li><a href="#coding-an-algebraic-system" id="toc-coding-an-algebraic-system" class="nav-link" data-scroll-target="#coding-an-algebraic-system">Coding an algebraic system</a></li>
+  <li><a href="#calling-the-algebraic-solver" id="toc-calling-the-algebraic-solver" class="nav-link" data-scroll-target="#calling-the-algebraic-solver">Calling the algebraic solver</a>
+  <ul class="collapse">
+  <li><a href="#data-versus-parameters" id="toc-data-versus-parameters" class="nav-link" data-scroll-target="#data-versus-parameters">Data versus parameters</a></li>
+  <li><a href="#length-of-the-algebraic-function-and-of-the-vector-of-unknowns" id="toc-length-of-the-algebraic-function-and-of-the-vector-of-unknowns" class="nav-link" data-scroll-target="#length-of-the-algebraic-function-and-of-the-vector-of-unknowns">Length of the algebraic function and of the vector of unknowns</a></li>
+  <li><a href="#pathological-solutions" id="toc-pathological-solutions" class="nav-link" data-scroll-target="#pathological-solutions">Pathological solutions</a></li>
+  </ul></li>
+  <li><a href="#algebra-control.section" id="toc-algebra-control.section" class="nav-link" data-scroll-target="#algebra-control.section">Control parameters for the algebraic solver</a>
+  <ul class="collapse">
+  <li><a href="#tolerance" id="toc-tolerance" class="nav-link" data-scroll-target="#tolerance">Tolerance</a></li>
+  <li><a href="#maximum-number-of-steps" id="toc-maximum-number-of-steps" class="nav-link" data-scroll-target="#maximum-number-of-steps">Maximum number of steps</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/algebraic-equations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/algebraic-equations.html">Solving Algebraic Equations</a></li></ol></nav></header>
+
+
+
+
+<section id="algebra-solver.chapter" class="level1">
+<h1>Solving Algebraic Equations</h1>
+<p>Stan provides a built-in mechanism for specifying systems of algebraic equations. These systems can be solved either with the Newton method, as implemented in the Kinsol package <span class="citation" data-cites="Hindmarsh:2005">(<a href="#ref-Hindmarsh:2005" role="doc-biblioref">Hindmarsh et al. 2005</a>)</span>, or with the Powell hybrid method <span class="citation" data-cites="Powell:1970">(<a href="#ref-Powell:1970" role="doc-biblioref">Powell 1970</a>)</span>. The function signatures for Stan’s algebraic solvers are fully described in the algebraic solver section of the reference manual.</p>
+<p>Solving any system of algebraic equations can be translated into a root-finding problem, that is, given a function <span class="math inline">\(f\)</span>, we wish to find <span class="math inline">\(y\)</span> such that <span class="math inline">\(f(y) = 0\)</span>.</p>
+<section id="example-system-of-nonlinear-algebraic-equations" class="level2">
+<h2 class="anchored" data-anchor-id="example-system-of-nonlinear-algebraic-equations">Example: system of nonlinear algebraic equations</h2>
+<p>For systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.</p>
+<p>As an illustrative example, we consider the following nonlinear system of two equations with two unknowns: <span class="math display">\[\begin{align*}
+z_1 &amp;= y_1 - \theta_1 \\
+z_2 &amp;= y_1 y_2 + \theta_2
+\end{align*}\]</span></p>
+<p>Our goal is to simultaneously solve all equations for <span class="math inline">\(y_1\)</span> and <span class="math inline">\(y_2\)</span>, such that the vector <span class="math inline">\(z\)</span> goes to 0.</p>
+</section>
+<section id="coding-an-algebraic-system" class="level2">
+<h2 class="anchored" data-anchor-id="coding-an-algebraic-system">Coding an algebraic system</h2>
+<p>A system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the <a href="../stan-users-guide/user-functions.html">user-defined functions section</a> for more information on coding user-defined functions).</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> system(<span class="dt">vector</span> y,              <span class="co">// unknowns</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>              <span class="dt">vector</span> theta,          <span class="co">// parameters</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>              <span class="kw">data</span> <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="co">// data (real)</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>              <span class="dt">array</span>[] <span class="dt">int</span> x_i) {     <span class="co">// data (integer)</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] z;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  z[<span class="dv">1</span>] = y[<span class="dv">1</span>] - theta[<span class="dv">1</span>];</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  z[<span class="dv">2</span>] = y[<span class="dv">1</span>] * y[<span class="dv">2</span>] - theta[<span class="dv">2</span>];</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> z;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function takes the unknowns we wish to solve for in <code>y</code> (a vector), the system parameters in <code>theta</code> (a vector), the real data in <code>x_r</code> (a real array) and the integer data in <code>x_i</code> (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.</p>
+<p>The body of the system function here could also be coded using a row vector constructor and transposition,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="cf">return</span> [ y[<span class="dv">1</span>] - theta[<span class="dv">1</span>],</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>         y[<span class="dv">1</span>] * y[<span class="dv">2</span>] - theta[<span class="dv">2</span>] ]';</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As systems get more complicated, naming the intermediate expressions goes a long way toward readability.</p>
+<section id="strict-signature" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="strict-signature">Strict signature</h4>
+<p>The function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.</p>
+</section>
+</section>
+<section id="calling-the-algebraic-solver" class="level2">
+<h2 class="anchored" data-anchor-id="calling-the-algebraic-solver">Calling the algebraic solver</h2>
+<p>Let’s suppose <span class="math inline">\(\theta = (3, 6)\)</span>. To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.</p>
+<p>The solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about <a href="../stan-users-guide/algebraic-equations.html#algebra-control.section">algebraic solvers with control parameters</a>.</p>
+<p>The following code returns the solution to our nonlinear algebraic system:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] y_guess = [<span class="dv">1</span>, <span class="dv">1</span>]';</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">real</span> x_r;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">int</span> x_i;</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] theta = [<span class="dv">3</span>, <span class="dv">6</span>]';</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] y;</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>  y = solve_newton(system, y_guess, theta, x_r, x_i);</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>which returns <span class="math inline">\(y = (3, -2)\)</span>.</p>
+<section id="data-versus-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-versus-parameters">Data versus parameters</h3>
+<p>The arguments for the real data <code>x_r</code> and the integer data <code>x_i</code> must be expressions that only involve data or transformed data variables. <code>theta</code>, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, <code>y_guess</code>, which may be a data or a parameter vector.</p>
+</section>
+<section id="length-of-the-algebraic-function-and-of-the-vector-of-unknowns" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="length-of-the-algebraic-function-and-of-the-vector-of-unknowns">Length of the algebraic function and of the vector of unknowns</h3>
+<p>The Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function <span class="math inline">\(f\)</span> with respect to the unknowns <span class="math inline">\(x\)</span> must be invertible. This requires the Jacobian to be square, meaning <span class="math inline">\(f(y)\)</span> and <span class="math inline">\(y\)</span> have the same length or, in other words <em>the number of equations in the system is the same as the number of unknowns.</em></p>
+</section>
+<section id="pathological-solutions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="pathological-solutions">Pathological solutions</h3>
+<p>Certain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.</p>
+<p>On the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.</p>
+</section>
+</section>
+<section id="algebra-control.section" class="level2">
+<h2 class="anchored" data-anchor-id="algebra-control.section">Control parameters for the algebraic solver</h2>
+<p>The call to the algebraic solver shown previously uses the default control settings. The <code>_tol</code> variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>y = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>                     theta, x_r, x_i);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.</p>
+<p>The default values for the control arguments are respectively <code>scaling_step = 1e-3</code> (<span class="math inline">\(10^{-3}\)</span>), <code>rel_tol = 1e-10</code> (<span class="math inline">\(10^{-10}\)</span>), <code>f_tol = 1e-6</code> (<span class="math inline">\(10^{-6}\)</span>), and <code>max_steps = 200</code> (<span class="math inline">\(200\)</span>).</p>
+<section id="tolerance" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="tolerance">Tolerance</h3>
+<p>The relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.</p>
+<p>Smaller relative tolerances produce more accurate solutions but require more computational time.</p>
+<section id="sensitivity-analysis" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="sensitivity-analysis">Sensitivity analysis</h4>
+<p>The tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by <span class="citation" data-cites="Timonen+etal:2023:ODE-PSIS">Timonen et al. (<a href="#ref-Timonen+etal:2023:ODE-PSIS" role="doc-biblioref">2023</a>)</span>.</p>
+</section>
+</section>
+<section id="maximum-number-of-steps" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="maximum-number-of-steps">Maximum number of steps</h3>
+<p>The maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Hindmarsh:2005" class="csl-entry" role="listitem">
+Hindmarsh, Alan C, Peter N Brown, Keith E Grant, Steven L Lee, Radu Serban, Dan E Shumaker, and Carol S Woodward. 2005. <span>“<span>SUNDIALS</span>: Suite of Nonlinear and Differential/Algebraic Equation Solvers.”</span> <em>ACM Transactions on Mathematical Software (TOMS)</em> 31 (3): 363–96.
+</div>
+<div id="ref-Powell:1970" class="csl-entry" role="listitem">
+Powell, Michael J. D. 1970. <span>“A Hybrid Method for Nonlinear Equations.”</span> In <em>Numerical Methods for Nonlinear Algebraic Equations</em>, edited by P. Rabinowitz. Gordon; Breach.
+</div>
+<div id="ref-Timonen+etal:2023:ODE-PSIS" class="csl-entry" role="listitem">
+Timonen, Juho, Nikolas Siccha, Ben Bales, Harri Lähdesmäki, and Aki Vehtari. 2023. <span>“An Importance Sampling Approach for Reliable and Efficient Inference in <span>Bayesian</span> Ordinary Differential Equation Models.”</span> <em>Stat</em> 12 (1): e614.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/hyperspherical-models.html" class="pagination-link" aria-label="Directions, Rotations, and Hyperspheres">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Directions, Rotations, and Hyperspheres</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/odes.html" class="pagination-link" aria-label="Ordinary Differential Equations">
+        <span class="nav-page-text">Ordinary Differential Equations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/algebraic-equations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/bootstrap.html b/docs/2_39/stan-users-guide/bootstrap.html
new file mode 100644
index 000000000..041801f23
--- /dev/null
+++ b/docs/2_39/stan-users-guide/bootstrap.html
@@ -0,0 +1,1363 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>The Bootstrap and Bagging</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/multiple-imputation.html" rel="next">
+<link href="../stan-users-guide/decision-analysis.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="The Bootstrap and Bagging">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/bootstrap.html">The Bootstrap and Bagging</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#the-bootstrap-and-bagging" id="toc-the-bootstrap-and-bagging" class="nav-link active" data-scroll-target="#the-bootstrap-and-bagging">The Bootstrap and Bagging</a>
+  <ul class="collapse">
+  <li><a href="#the-bootstrap" id="toc-the-bootstrap" class="nav-link" data-scroll-target="#the-bootstrap">The bootstrap</a>
+  <ul class="collapse">
+  <li><a href="#estimators" id="toc-estimators" class="nav-link" data-scroll-target="#estimators">Estimators</a></li>
+  <li><a href="#the-bootstrap-in-pseudocode" id="toc-the-bootstrap-in-pseudocode" class="nav-link" data-scroll-target="#the-bootstrap-in-pseudocode">The bootstrap in pseudocode</a></li>
+  </ul></li>
+  <li><a href="#coding-the-bootstrap-in-stan" id="toc-coding-the-bootstrap-in-stan" class="nav-link" data-scroll-target="#coding-the-bootstrap-in-stan">Coding the bootstrap in Stan</a></li>
+  <li><a href="#error-statistics-from-the-bootstrap" id="toc-error-statistics-from-the-bootstrap" class="nav-link" data-scroll-target="#error-statistics-from-the-bootstrap">Error statistics from the bootstrap</a>
+  <ul class="collapse">
+  <li><a href="#standard-errors" id="toc-standard-errors" class="nav-link" data-scroll-target="#standard-errors">Standard errors</a></li>
+  <li><a href="#confidence-intervals" id="toc-confidence-intervals" class="nav-link" data-scroll-target="#confidence-intervals">Confidence intervals</a></li>
+  </ul></li>
+  <li><a href="#bagging" id="toc-bagging" class="nav-link" data-scroll-target="#bagging">Bagging</a></li>
+  <li><a href="#bayesian-bootstrap-and-bagging" id="toc-bayesian-bootstrap-and-bagging" class="nav-link" data-scroll-target="#bayesian-bootstrap-and-bagging">Bayesian bootstrap and bagging</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/bootstrap.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/bootstrap.html">The Bootstrap and Bagging</a></li></ol></nav></header>
+
+
+
+
+<section id="the-bootstrap-and-bagging" class="level1">
+<h1>The Bootstrap and Bagging</h1>
+<p>The bootstrap is a technique for approximately sampling from the error distribution for an estimator. Thus it can be used as a Monte Carlo method to estimate standard errors and confidence intervals for point estimates <span class="citation" data-cites="EfronTibshirani1986 EfronTibshirani1994">(<a href="#ref-EfronTibshirani1986" role="doc-biblioref">Efron and Tibshirani 1986</a>; <a href="#ref-EfronTibshirani1994" role="doc-biblioref">1994</a>)</span>. It works by subsampling the original data and computing sample estimates from the subsample. Like other Monte Carlo methods, the bootstrap is plug-and-play, allowing great flexibility in both model choice and estimator.</p>
+<p>Bagging is a technique for combining bootstrapped estimators for model criticism and more robust inference <span class="citation" data-cites="Breiman:1996 HugginsMiller:2019">(<a href="#ref-Breiman:1996" role="doc-biblioref">Breiman 1996</a>; <a href="#ref-HugginsMiller:2019" role="doc-biblioref">Huggins and Miller 2019</a>)</span>.</p>
+<section id="the-bootstrap" class="level2">
+<h2 class="anchored" data-anchor-id="the-bootstrap">The bootstrap</h2>
+<section id="estimators" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimators">Estimators</h3>
+<p>An estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set <span class="math inline">\(y_{1,\ldots, N}\)</span> to a number by <span class="math display">\[
+\textrm{mean}(y) = \frac{1}{N} \sum_{n=1}^N y_n,
+\]</span> and hence meets the definition of an estimator. Given the likelihood function <span class="math display">\[
+p(y \mid \mu) = \prod_{n=1}^N \textrm{normal}(y_n \mid \mu, 1),
+\]</span> the mean is the maximum likelihood estimator,</p>
+<p><span class="math display">\[
+\textrm{mean}(y) = \textrm{arg max}_{\mu} \ p(y \mid \mu, 1)
+\]</span> A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.</p>
+<p>In analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.</p>
+</section>
+<section id="the-bootstrap-in-pseudocode" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="the-bootstrap-in-pseudocode">The bootstrap in pseudocode</h3>
+<p>The bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.</p>
+<p>The following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate <span class="math inline">\(\hat{\theta}\)</span> that is a function of data <span class="math inline">\(y\)</span>.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  y_rep[m] &lt;- sample_uniform(y)</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  theta_hat[m] &lt;- estimate_theta(y_rep[m])</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>std_error = sd(theta_hat)</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>conf_95pct = [ quantile(theta_hat, <span class="fl">0.025</span>),</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>               quantile(theta_hat, <span class="fl">0.975</span>) ]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>sample_uniform</code> function works by independently assigning each element of <code>y_rep</code> an element of <code>y</code> drawn uniformly at random. This produces a sample <em>with replacement.</em> That is, some elements of <code>y</code> may show up more than once in <code>y_rep</code> and some may not appear at all.</p>
+</section>
+</section>
+<section id="coding-the-bootstrap-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="coding-the-bootstrap-in-stan">Coding the bootstrap in Stan</h2>
+<p>The bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector <code>x</code> of predictors in addition to outcomes <code>y</code>. To allow a single program to fit both the original data and random subsamples, the variable <code>resample</code> is set to 1 to resample and 0 to use the original data.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; resample;</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[N] uniform = rep_vector(<span class="fl">1.0</span> / N, N);</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; boot_idxs;</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>    boot_idxs[n] = resample ? categorical_rng(uniform) : n;</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The model accepts data in the usual form for a linear regression as a number of observations <span class="math inline">\(N\)</span> with a size <span class="math inline">\(N\)</span> vector <span class="math inline">\(x\)</span> of predictors and a size <span class="math inline">\(N\)</span> vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of <code>boot_idxs</code> from <code>1:N</code>, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array <code>boot_idxs</code> is defined to be the sequence <code>1:N</code>, because <code>x == x[1:N]</code> and <code>y = y[1:N]</code>.</p>
+<p>For example, when <code>resample == 1</code>, if <span class="math inline">\(N = 4,\)</span> the value of <code>boot_idxs</code> might be <code>{2, 1, 1, 3}</code>, resulting in a bootstrap sample <code>{y[2], y[1], y[1], y[3]}</code> with the first element repeated twice and the fourth element not sampled at all.</p>
+<p>The parameters are the usual regression coefficients for the intercept <code>alpha</code>, slope <code>beta</code>, and error scale <code>sigma</code>. The model uses the bootstrap index variable <code>boot_idx</code> to index the predictors as <code>x[boot_idx]</code> and outcomes as <code>y[boot_idx]</code>. This generates a new size-<span class="math inline">\(N\)</span> vector whose entries are defined by <code>x[boot_idx][n] = x[boot_idx[n]]</code> and similarly for <code>y</code>. For example, if <span class="math inline">\(N = 4\)</span> and <code>boot_idxs = {2, 1, 1, 3}</code>, then <code>x[boot_idxs] = [x[2], x[1], x[1], x[3]]'</code> and <code>y[boot_idxs] = [y[2], y[1], y[1], y[3]]'</code>. The predictor and outcome vectors remain aligned, with both elements of the pair <code>x[1]</code> and <code>y[1]</code> repeated twice.</p>
+<p>With the model defined this way, if <code>resample</code> is 1, the model is fit to a bootstrap subsample of the data. If <code>resample</code> is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.</p>
+</section>
+<section id="error-statistics-from-the-bootstrap" class="level2">
+<h2 class="anchored" data-anchor-id="error-statistics-from-the-bootstrap">Error statistics from the bootstrap</h2>
+<p>Running the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.</p>
+<p>To estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data <span class="math inline">\(y\)</span>, it is usually not worth calculating bootstrap estimates to high precision.</p>
+<section id="standard-errors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="standard-errors">Standard errors</h3>
+<p>Here’s the result of calculating standard errors for the linear regression model above with <span class="math inline">\(N = 50\)</span> data points, <span class="math inline">\(\alpha = 1.2, \beta
+= -0.5,\)</span> and <span class="math inline">\(\sigma = 1.5.\)</span> With a total of <span class="math inline">\(M = 100\)</span> bootstrap samples, there are 100 estimates of <span class="math inline">\(\alpha\)</span>, 100 of <span class="math inline">\(\beta\)</span>, and 100 of <span class="math inline">\(\sigma\)</span>. These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for <span class="math inline">\(\alpha\)</span> provide the bootstrap estimate of the standard error in the estimate for <span class="math inline">\(\alpha\)</span>. Here’s what it looks like for the above model with <span class="math inline">\(M =
+100\)</span></p>
+<pre><code> parameter   estimate    std err
+ ---------   --------    -------
+     alpha      1.359      0.218
+      beta     -0.610      0.204
+     sigma      1.537      0.142</code></pre>
+<p>With the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:</p>
+<pre><code> parameter   estimate    std err
+ ---------   --------    -------
+     alpha      1.359      0.206
+     alpha      1.359      0.240
+     alpha      1.359      0.234
+     alpha      1.359      0.249
+     alpha      1.359      0.227</code></pre>
+<p>Increasing <span class="math inline">\(M\)</span> will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample <span class="math inline">\(y\)</span>.</p>
+</section>
+<section id="confidence-intervals" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="confidence-intervals">Confidence intervals</h3>
+<p>As usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are <span class="math inline">\(M = 1000\)</span> estimates of <span class="math inline">\(\hat{\alpha}\)</span> in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set <span class="math inline">\(y\)</span>. To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).</p>
+</section>
+</section>
+<section id="bagging" class="level2">
+<h2 class="anchored" data-anchor-id="bagging">Bagging</h2>
+<p>When bootstrapping is carried through inference it is known as bootstrap aggregation, or <em>bagging</em>, in the machine-learning literature <span class="citation" data-cites="Breiman:1996">(<a href="#ref-Breiman:1996" role="doc-biblioref">Breiman 1996</a>)</span>. In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate <span class="math inline">\(\hat{\sigma}\)</span> from the original data set, bootstrapped data sets <span class="math inline">\(y^{\textrm{boot}(1)}, \ldots,
+y^{\textrm{boot}(N)}\)</span> are generated. Each is used to generate an estimate <span class="math inline">\(\hat{\sigma}^{\textrm{boot}(n)}.\)</span> The final estimate is <span class="math display">\[
+\hat{\sigma} = \frac{1}{N} \sum_{n = 1}^N \hat{\sigma}^{\textrm{boot}(n)}.
+\]</span> The same would be done to estimate a predictive quantity <span class="math inline">\(\tilde{y}\)</span> for as yet unseen data. <span class="math display">\[
+\hat{\tilde{y}} = \frac{1}{N} \sum_{n = 1}^N
+\hat{\tilde{y}}^{\textrm{boot}(n)}.
+\]</span> For discrete parameters, voting is used to select the outcome.</p>
+<p>One way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.</p>
+</section>
+<section id="bayesian-bootstrap-and-bagging" class="level2">
+<h2 class="anchored" data-anchor-id="bayesian-bootstrap-and-bagging">Bayesian bootstrap and bagging</h2>
+<p>A Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.</p>
+<p><span class="citation" data-cites="HugginsMiller:2019">(<a href="#ref-HugginsMiller:2019" role="doc-biblioref">Huggins and Miller 2019</a>)</span> use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Breiman:1996" class="csl-entry" role="listitem">
+Breiman, Leo. 1996. <span>“Bagging Predictors.”</span> <em>Machine Learning</em> 24 (2): 123–40.
+</div>
+<div id="ref-EfronTibshirani1986" class="csl-entry" role="listitem">
+Efron, Bradley, and Robert Tibshirani. 1986. <span>“Bootstrap Methods for Standard Errors, Confidence Intervals, and Other Measures of Statistical Accuracy.”</span> <em>Statistical Science</em> 1 (1): 54–75.
+</div>
+<div id="ref-EfronTibshirani1994" class="csl-entry" role="listitem">
+Efron, Bradley, and Robert J Tibshirani. 1994. <em>An Introduction to the Bootstrap</em>. Chapman &amp; Hall/CRC.
+</div>
+<div id="ref-HugginsMiller:2019" class="csl-entry" role="listitem">
+Huggins, Jonathan H, and Jeffrey W Miller. 2019. <span>“Using Bagged Posteriors for Robust Inference and Model Criticism.”</span> <em>arXiv</em>, no. 1912.07104.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/decision-analysis.html" class="pagination-link" aria-label="Decision Analysis">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Decision Analysis</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/multiple-imputation.html" class="pagination-link" aria-label="Multiple imputation">
+        <span class="nav-page-text">Multiple imputation</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/bootstrap.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/clustering.html b/docs/2_39/stan-users-guide/clustering.html
new file mode 100644
index 000000000..c69c45ec7
--- /dev/null
+++ b/docs/2_39/stan-users-guide/clustering.html
@@ -0,0 +1,1655 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Clustering Models</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/gaussian-processes.html" rel="next">
+<link href="../stan-users-guide/sparse-ragged.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Clustering Models">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/clustering.html">Clustering Models</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#clustering.chapter" id="toc-clustering.chapter" class="nav-link active" data-scroll-target="#clustering.chapter">Clustering Models</a>
+  <ul class="collapse">
+  <li><a href="#relation-to-finite-mixture-models" id="toc-relation-to-finite-mixture-models" class="nav-link" data-scroll-target="#relation-to-finite-mixture-models">Relation to finite mixture models</a></li>
+  <li><a href="#soft-k-means" id="toc-soft-k-means" class="nav-link" data-scroll-target="#soft-k-means">Soft <em>K</em>-means</a>
+  <ul class="collapse">
+  <li><a href="#geometric-hard-k-means-clustering" id="toc-geometric-hard-k-means-clustering" class="nav-link" data-scroll-target="#geometric-hard-k-means-clustering">Geometric hard <em>K</em>-means clustering</a></li>
+  <li><a href="#soft-k-means-clustering" id="toc-soft-k-means-clustering" class="nav-link" data-scroll-target="#soft-k-means-clustering">Soft <em>K</em>-means clustering</a></li>
+  <li><a href="#stan-implementation-of-soft-k-means" id="toc-stan-implementation-of-soft-k-means" class="nav-link" data-scroll-target="#stan-implementation-of-soft-k-means">Stan implementation of soft <em>K</em>-means</a></li>
+  <li><a href="#generalizing-soft-k-means" id="toc-generalizing-soft-k-means" class="nav-link" data-scroll-target="#generalizing-soft-k-means">Generalizing soft <em>K</em>-means</a></li>
+  </ul></li>
+  <li><a href="#the-difficulty-of-bayesian-inference-for-clustering" id="toc-the-difficulty-of-bayesian-inference-for-clustering" class="nav-link" data-scroll-target="#the-difficulty-of-bayesian-inference-for-clustering">The difficulty of Bayesian inference for clustering</a>
+  <ul class="collapse">
+  <li><a href="#non-identifiability" id="toc-non-identifiability" class="nav-link" data-scroll-target="#non-identifiability">Non-identifiability</a></li>
+  <li><a href="#multimodality" id="toc-multimodality" class="nav-link" data-scroll-target="#multimodality">Multimodality</a></li>
+  </ul></li>
+  <li><a href="#naive-bayes-classification-and-clustering" id="toc-naive-bayes-classification-and-clustering" class="nav-link" data-scroll-target="#naive-bayes-classification-and-clustering">Naive Bayes classification and clustering</a>
+  <ul class="collapse">
+  <li><a href="#coding-ragged-arrays" id="toc-coding-ragged-arrays" class="nav-link" data-scroll-target="#coding-ragged-arrays">Coding ragged arrays</a></li>
+  <li><a href="#estimation-with-category-labeled-training-data" id="toc-estimation-with-category-labeled-training-data" class="nav-link" data-scroll-target="#estimation-with-category-labeled-training-data">Estimation with category-labeled training data</a></li>
+  <li><a href="#estimation-without-category-labeled-training-data" id="toc-estimation-without-category-labeled-training-data" class="nav-link" data-scroll-target="#estimation-without-category-labeled-training-data">Estimation without category-labeled training data</a></li>
+  <li><a href="#full-bayesian-inference-for-naive-bayes" id="toc-full-bayesian-inference-for-naive-bayes" class="nav-link" data-scroll-target="#full-bayesian-inference-for-naive-bayes">Full Bayesian inference for naive Bayes</a></li>
+  <li><a href="#prediction-without-model-updates" id="toc-prediction-without-model-updates" class="nav-link" data-scroll-target="#prediction-without-model-updates">Prediction without model updates</a></li>
+  </ul></li>
+  <li><a href="#latent-dirichlet-allocation" id="toc-latent-dirichlet-allocation" class="nav-link" data-scroll-target="#latent-dirichlet-allocation">Latent Dirichlet allocation</a>
+  <ul class="collapse">
+  <li><a href="#the-lda-model" id="toc-the-lda-model" class="nav-link" data-scroll-target="#the-lda-model">The LDA Model</a></li>
+  <li><a href="#summing-out-the-discrete-parameters" id="toc-summing-out-the-discrete-parameters" class="nav-link" data-scroll-target="#summing-out-the-discrete-parameters">Summing out the discrete parameters</a></li>
+  <li><a href="#implementation-of-lda" id="toc-implementation-of-lda" class="nav-link" data-scroll-target="#implementation-of-lda">Implementation of LDA</a></li>
+  <li><a href="#correlated-topic-model" id="toc-correlated-topic-model" class="nav-link" data-scroll-target="#correlated-topic-model">Correlated topic model</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/clustering.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/clustering.html">Clustering Models</a></li></ol></nav></header>
+
+
+
+
+<section id="clustering.chapter" class="level1">
+<h1>Clustering Models</h1>
+<p>Unsupervised methods for organizing data into groups are collectively referred to as clustering. This chapter describes the implementation in Stan of two widely used statistical clustering models, soft <span class="math inline">\(K\)</span>-means and latent Dirichlet allocation (LDA). In addition, this chapter includes naive Bayesian classification, which can be viewed as a form of clustering which may be supervised. These models are typically expressed using discrete parameters for cluster assignments. Nevertheless, they can be implemented in Stan like any other mixture model by marginalizing out the discrete parameters (see the <a href="../stan-users-guide/finite-mixtures.html">mixture modeling chapter</a>).</p>
+<section id="relation-to-finite-mixture-models" class="level2">
+<h2 class="anchored" data-anchor-id="relation-to-finite-mixture-models">Relation to finite mixture models</h2>
+<p>As mentioned in the <a href="../stan-users-guide/finite-mixtures.html#clustering-mixture.section">clustering section</a>, clustering models and finite mixture models are really just two sides of the same coin. The “soft” <span class="math inline">\(K\)</span>-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of <span class="math inline">\(K\)</span>-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.</p>
+</section>
+<section id="soft-k-means" class="level2">
+<h2 class="anchored" data-anchor-id="soft-k-means">Soft <em>K</em>-means</h2>
+<p><span class="math inline">\(K\)</span>-means clustering is a method of clustering data represented as <span class="math inline">\(D\)</span>-dimensional vectors. Specifically, there will be <span class="math inline">\(N\)</span> items to be clustered, each represented as a vector <span class="math inline">\(y_n \in \mathbb{R}^D\)</span>. In the “soft” version of <span class="math inline">\(K\)</span>-means, the assignments to clusters will be probabilistic.</p>
+<section id="geometric-hard-k-means-clustering" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="geometric-hard-k-means-clustering">Geometric hard <em>K</em>-means clustering</h3>
+<p><span class="math inline">\(K\)</span>-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters <span class="math inline">\(K\)</span> and data vectors <span class="math inline">\(y\)</span> as input.</p>
+<ol type="1">
+<li>For each <span class="math inline">\(n\)</span> in <span class="math inline">\(\{1,\dotsc,N\}\)</span>, randomly assign vector <span class="math inline">\(y_n\)</span> to a cluster in <span class="math inline">\(\{1,\dotsc,K\}\)</span>;</li>
+<li>Repeat
+<ol type="1">
+<li>For each cluster <span class="math inline">\(k\)</span> in <span class="math inline">\(\{1,\dotsc,K\}\)</span>, compute the cluster centroid <span class="math inline">\(\mu_k\)</span> by averaging the vectors assigned to that cluster;</li>
+<li>For each <span class="math inline">\(n\)</span> in <span class="math inline">\(\{1,\dotsc,N\}\)</span>, reassign <span class="math inline">\(y_n\)</span> to the cluster <span class="math inline">\(k\)</span> for which the (Euclidean) distance from <span class="math inline">\(y_n\)</span> to <span class="math inline">\(\mu_k\)</span> is smallest;</li>
+<li>If no vectors changed cluster, return the cluster assignments.</li>
+</ol></li>
+</ol>
+<p>This algorithm is guaranteed to terminate.</p>
+</section>
+<section id="soft-k-means-clustering" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="soft-k-means-clustering">Soft <em>K</em>-means clustering</h3>
+<p>Soft <span class="math inline">\(K\)</span>-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft <span class="math inline">\(K\)</span>-means can be expressed (and coded in Stan) as a multivariate normal mixture model.</p>
+<p>In the full generative model, each data point <span class="math inline">\(n\)</span> in <span class="math inline">\(\{1,\dotsc,N\}\)</span> is assigned a cluster <span class="math inline">\(z_n \in \{1,\dotsc,K\}\)</span> with symmetric uniform probability, <span class="math display">\[
+z_n \sim \textsf{categorical}(1/K),
+\]</span> where <span class="math inline">\(1\)</span> is the unit vector of <span class="math inline">\(K\)</span> dimensions, so that <span class="math inline">\(1/K\)</span> is the symmetric <span class="math inline">\(K\)</span>-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.</p>
+<p>The data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment <span class="math inline">\(z_n\)</span>, <span class="math display">\[
+y_n \sim  \textsf{normal}(\mu_{z[n]},\Sigma_{z[n]})
+\]</span></p>
+<p>The sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters <span class="math inline">\(k\)</span>, <span class="math display">\[
+\Sigma_k = \mathrm{diag\_matrix}({\bf 1}),
+\]</span> so that the log multivariate normal can be implemented directly up to a proportion by <span class="math display">\[
+\mathrm{normal}\left( y_n \mid \mu_k, \mathrm{diag\_matrix}({\bf 1}) \right)
+\propto \exp \left (- \frac{1}{2} \sum_{d=1}^D \left( \mu_{k,d} - y_{n,d}
+  \right)^2 \right).
+\]</span> The spatial perspective on <span class="math inline">\(K\)</span>-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean <span class="math inline">\(\mu_k\)</span> to the data point <span class="math inline">\(y_n\)</span>.</p>
+</section>
+<section id="stan-implementation-of-soft-k-means" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-implementation-of-soft-k-means">Stan implementation of soft <em>K</em>-means</h3>
+<p>Consider the following Stan program for implementing <span class="math inline">\(K\)</span>-means clustering.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;        <span class="co">// number of data points</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;        <span class="co">// number of dimensions</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;        <span class="co">// number of clusters</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] y;  <span class="co">// observations</span></span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt; neg_log_K;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  neg_log_K = -log(K);</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">vector</span>[D] mu; <span class="co">// cluster means</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N, K] <span class="dt">real</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt; soft_z; <span class="co">// log unnormalized clusters</span></span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>      soft_z[n, k] = neg_log_K</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>                     - <span class="fl">0.5</span> * dot_self(mu[k] - y[n]);</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-23"><a href="#cb1-23" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-24"><a href="#cb1-24" aria-hidden="true" tabindex="-1"></a>  <span class="co">// prior</span></span>
+<span id="cb1-25"><a href="#cb1-25" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb1-26"><a href="#cb1-26" aria-hidden="true" tabindex="-1"></a>    mu[k] ~ std_normal();</span>
+<span id="cb1-27"><a href="#cb1-27" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-28"><a href="#cb1-28" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-29"><a href="#cb1-29" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood</span></span>
+<span id="cb1-30"><a href="#cb1-30" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-31"><a href="#cb1-31" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_sum_exp(soft_z[n]);</span>
+<span id="cb1-32"><a href="#cb1-32" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-33"><a href="#cb1-33" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.</p>
+<p>The only parameter is <code>mu</code>, where <code>mu[k]</code> is the centroid for cluster <span class="math inline">\(k\)</span>. The transformed parameters <code>soft_z[n]</code> contain the log of the unnormalized cluster assignment probabilities. The vector <code>soft_z[n]</code> can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.</p>
+</section>
+<section id="generalizing-soft-k-means" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="generalizing-soft-k-means">Generalizing soft <em>K</em>-means</h3>
+<p>The multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., <span class="math inline">\(L_2\)</span> distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on <span class="math inline">\(L_1\)</span> distance (i.e., Manhattan or taxicab distance).</p>
+<p>Within the multivariate normal version of <span class="math inline">\(K\)</span>-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.</p>
+<p>Although there is no global spatial analog, it is common to see soft <span class="math inline">\(K\)</span>-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.</p>
+</section>
+</section>
+<section id="the-difficulty-of-bayesian-inference-for-clustering" class="level2">
+<h2 class="anchored" data-anchor-id="the-difficulty-of-bayesian-inference-for-clustering">The difficulty of Bayesian inference for clustering</h2>
+<p>Two problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the <a href="../stan-users-guide/problematic-posteriors.html#label-switching-problematic.section">label switching section</a>.</p>
+<section id="non-identifiability" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="non-identifiability">Non-identifiability</h3>
+<p>Cluster assignments are not identified—permuting the cluster mean vectors <code>mu</code> leads to a model with identical likelihoods. For instance, permuting the first two indexes in <code>mu</code> and the first two indexes in each <code>soft_z[n]</code> leads to an identical likelihood (and prior).</p>
+<p>The lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft <span class="math inline">\(K\)</span>-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.</p>
+</section>
+<section id="multimodality" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multimodality">Multimodality</h3>
+<p>The other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.</p>
+<p>Bayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.</p>
+<p>In light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.</p>
+</section>
+</section>
+<section id="naive-bayes-classification-and-clustering" class="level2">
+<h2 class="anchored" data-anchor-id="naive-bayes-classification-and-clustering">Naive Bayes classification and clustering</h2>
+<p>Naive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>Multinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.</p>
+<p>Naive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.</p>
+<p>The observed data consists of a sequence of <span class="math inline">\(M\)</span> documents made up of bags of words drawn from a vocabulary of <span class="math inline">\(V\)</span> distinct words. A document <span class="math inline">\(m\)</span> has <span class="math inline">\(N_m\)</span> words, which are indexed as <span class="math inline">\(w_{m,1}, \dotsc,
+w_{m,N[m]} \in \{1,\dotsc,V\}\)</span>. Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) <span class="math inline">\(K\)</span> is fixed.</p>
+<p>The multinomial mixture model generates a single category <span class="math inline">\(z_m \in
+\{1,\dotsc,K\}\)</span> for each document <span class="math inline">\(m \in \{1,\dotsc,M\}\)</span> according to a categorical distribution, <span class="math display">\[
+z_m \sim \textsf{categorical}(\theta).
+\]</span> The <span class="math inline">\(K\)</span>-simplex parameter <span class="math inline">\(\theta\)</span> represents the prevalence of each category in the data.</p>
+<p>Next, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word <span class="math inline">\(n\)</span> of document <span class="math inline">\(m\)</span> being generated as <span class="math display">\[
+w_{m,n} \sim \textsf{categorical}(\phi_{z[m]}).
+\]</span> The parameter <span class="math inline">\(\phi_{z[m]}\)</span> is a <span class="math inline">\(V\)</span>-simplex representing the probability of each word in the vocabulary in documents of category <span class="math inline">\(z_m\)</span>.</p>
+<p>The parameters <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span> are typically given symmetric Dirichlet priors. The prevalence <span class="math inline">\(\theta\)</span> is sometimes fixed to produce equal probabilities for each category <span class="math inline">\(k \in \{1,\dotsc,K\}\)</span>.</p>
+<section id="coding-ragged-arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="coding-ragged-arrays">Coding ragged arrays</h3>
+<p>The specification for naive Bayes in the previous sections have used a ragged array notation for the words <span class="math inline">\(w\)</span>. Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.</p>
+<p><span class="math display">\[
+\begin{array}{lll}
+\hline
+\mathrm{n} \qquad\qquad\qquad\qquad &amp; \mathrm{w[n]} \qquad &amp; \mathrm{doc[n]} \\
+\hline
+1         &amp; w_{1,1}    &amp; 1 \\
+2         &amp; w_{1,2}    &amp; 1 \\
+\vdots &amp; \vdots    &amp; \vdots \\
+N_1     &amp; w_{1,N[1]} &amp; 1 \\
+N_1 + 1 &amp; w_{2,1}    &amp; 2 \\
+N_1 + 2 &amp; w_{2,2}    &amp; 2 \\
+\vdots &amp; \vdots    &amp; \vdots \\
+N_1 + N_2     &amp; w_{2,N[2]} &amp; 2 \\
+N_1 + N_2 + 1 &amp; w_{3,1}    &amp; 3 \\
+\vdots       &amp; \vdots    &amp; \vdots \\
+N = \sum_{m=1}^M N_m &amp; w_{M,N[M]} &amp; M \\
+\hline
+\end{array}
+\]</span></p>
+<p>The relevant variables for the program are <code>N</code>, the total number of words in all the documents, the word array <code>w</code>, and the document identity array <code>doc</code>.</p>
+</section>
+<section id="estimation-with-category-labeled-training-data" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimation-with-category-labeled-training-data">Estimation with category-labeled training data</h3>
+<p>A naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// training data</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;               <span class="co">// num topics</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; V;               <span class="co">// num words</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; M;               <span class="co">// num docs</span></span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;               <span class="co">// total word instances</span></span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[M] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; z;    <span class="co">// topic for doc m</span></span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=V&gt; w;    <span class="co">// word n</span></span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=M&gt; doc;  <span class="co">// doc ID for word n</span></span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// hyperparameters</span></span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] alpha;     <span class="co">// topic prior</span></span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[V] beta;      <span class="co">// word prior</span></span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] theta;             <span class="co">// topic prevalence</span></span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">simplex</span>[V] phi;      <span class="co">// word dist for topic k</span></span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>  theta ~ dirichlet(alpha);</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>    phi[k] ~ dirichlet(beta);</span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb2-24"><a href="#cb2-24" aria-hidden="true" tabindex="-1"></a>    z[m] ~ categorical(theta);</span>
+<span id="cb2-25"><a href="#cb2-25" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-26"><a href="#cb2-26" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-27"><a href="#cb2-27" aria-hidden="true" tabindex="-1"></a>    w[n] ~ categorical(phi[z[doc[n]]]);</span>
+<span id="cb2-28"><a href="#cb2-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-29"><a href="#cb2-29" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The topic identifiers <span class="math inline">\(z_m\)</span> are declared as data and the latent category assignments are included as part of the likelihood function.</p>
+</section>
+<section id="estimation-without-category-labeled-training-data" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimation-without-category-labeled-training-data">Estimation without category-labeled training data</h3>
+<p>Naive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number <span class="math inline">\(K\)</span> of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels <code>z</code>. Because <code>z</code> is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability</p>
+<p><span class="math display">\[\begin{align*}
+\log\, &amp;p(w_{m,1},\dotsc,w_{m,N_m} \mid \theta,\phi) \\
+&amp;= \log \sum_{k=1}^K
+    \left( \textsf{categorical}(k \mid \theta)
+           \times \prod_{n=1}^{N_m} \textsf{categorical}(w_{m,n} \mid \phi_k)
+    \right) \\
+&amp;= \log \sum_{k=1}^K \exp \left(
+    \log \textsf{categorical}(k \mid \theta)
+     + \sum_{n=1}^{N_m} \log \textsf{categorical}(w_{m,n} \mid \phi_k)
+    \right).
+\end{align*}\]</span></p>
+<p>The last step shows how the <code>log_sum_exp</code> function can be used to stabilize the numerical calculation and return a result on the log scale.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[M, K] <span class="dt">real</span> gamma;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  theta ~ dirichlet(alpha);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>    phi[k] ~ dirichlet(beta);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>      gamma[m, k] = categorical_lpmf(k | theta);</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>      gamma[doc[n], k] = gamma[doc[n], k]</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>                         + categorical_lpmf(w[n] | phi[k]);</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-17"><a href="#cb3-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-18"><a href="#cb3-18" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb3-19"><a href="#cb3-19" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_sum_exp(gamma[m]);</span>
+<span id="cb3-20"><a href="#cb3-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-21"><a href="#cb3-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The local variable <code>gamma[m, k]</code> represents the value <span class="math display">\[
+\gamma_{m,k} = \log \textsf{categorical}(k \mid \theta)
++ \sum_{n=1}^{N_m} \log \textsf{categorical}(w_{m,n} \mid \phi_k).
+\]</span></p>
+<p>Given <span class="math inline">\(\gamma\)</span>, the posterior probability that document <span class="math inline">\(m\)</span> is assigned category <span class="math inline">\(k\)</span> is <span class="math display">\[
+\Pr[z_m = k \mid w,\alpha,\beta]
+=
+\exp \left(
+\gamma_{m,k}
+- \log \sum_{k=1}^K \exp \left( \gamma_{m,k} \right)
+\right).
+\]</span></p>
+<p>If the variable <code>gamma</code> were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.</p>
+</section>
+<section id="full-bayesian-inference-for-naive-bayes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="full-bayesian-inference-for-naive-bayes">Full Bayesian inference for naive Bayes</h3>
+<p>Full Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> and <span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007</a>)</span> for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.</p>
+<p>To specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables <code>M2</code> <code>N2</code>, <code>w2</code>, and <code>doc2</code>. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.</p>
+</section>
+<section id="prediction-without-model-updates" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="prediction-without-model-updates">Prediction without model updates</h3>
+<p>An alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of <code>gamma</code> for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.</p>
+</section>
+</section>
+<section id="latent-dirichlet-allocation" class="level2">
+<h2 class="anchored" data-anchor-id="latent-dirichlet-allocation">Latent Dirichlet allocation</h2>
+<p>Latent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model <span class="citation" data-cites="BleiNgJordan:2003">(<a href="#ref-BleiNgJordan:2003" role="doc-biblioref">Blei, Ng, and Jordan 2003</a>)</span> that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.</p>
+<section id="the-lda-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="the-lda-model">The LDA Model</h3>
+<p>The basic model assumes each document is generated independently based on fixed hyperparameters. For document <span class="math inline">\(m\)</span>, the first step is to draw a topic distribution simplex <span class="math inline">\(\theta_m\)</span> over the <span class="math inline">\(K\)</span> topics, <span class="math display">\[
+\theta_m \sim \textsf{Dirichlet}(\alpha).
+\]</span></p>
+<p>The prior hyperparameter <span class="math inline">\(\alpha\)</span> is fixed to a <span class="math inline">\(K\)</span>-vector of positive values. Each word in the document is generated independently conditional on the distribution <span class="math inline">\(\theta_m\)</span>. First, a topic <span class="math inline">\(z_{m,n} \in \{1,\dotsc,K\}\)</span> is drawn for the word based on the document-specific topic-distribution, <span class="math display">\[
+z_{m,n} \sim \textsf{categorical}(\theta_m).
+\]</span></p>
+<p>Finally, the word <span class="math inline">\(w_{m,n}\)</span> is drawn according to the word distribution for topic <span class="math inline">\(z_{m,n}\)</span>, <span class="math display">\[
+w_{m,n} \sim \textsf{categorical}(\phi_{z[m,n]}).
+\]</span> The distributions <span class="math inline">\(\phi_k\)</span> over words for topic <span class="math inline">\(k\)</span> are also given a Dirichlet prior, <span class="math display">\[
+\phi_k \sim \textsf{Dirichlet}(\beta)
+\]</span></p>
+<p>where <span class="math inline">\(\beta\)</span> is a fixed <span class="math inline">\(V\)</span>-vector of positive values.</p>
+</section>
+<section id="summing-out-the-discrete-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="summing-out-the-discrete-parameters">Summing out the discrete parameters</h3>
+<p>Although Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is <span class="math display">\[\begin{align*}
+p(\theta,\phi \mid w,\alpha,\beta)
+&amp;\propto p(\theta \mid \alpha) \, p(\phi \mid \beta) \, p(w \mid \theta,\phi) \\
+&amp;= \prod_{m=1}^M p(\theta_m \mid \alpha)
+    \times \prod_{k=1}^K p(\phi_k \mid \beta)
+    \times \prod_{m=1}^M \prod_{n=1}^{M[n]} p(w_{m,n} \mid \theta_m,\phi).
+\end{align*}\]</span></p>
+<p>The inner word-probability term is defined by summing out the topic assignments, <span class="math display">\[\begin{align*}
+p(w_{m,n} \mid \theta_m,\phi)
+&amp;= \sum_{z=1}^K p(z,w_{m,n} \mid \theta_m,\phi) \\
+&amp;= \sum_{z=1}^K p(z \mid \theta_m) \, p(w_{m,n} \mid \phi_z).
+\end{align*}\]</span></p>
+<p>Plugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, <span class="math display">\[\begin{align*}
+\log\, &amp;p(\theta,\phi \mid w,\alpha,\beta) \\
+&amp;= \sum_{m=1}^M \log \textsf{Dirichlet}(\theta_m \mid \alpha)
+    + \sum_{k=1}^K \log \textsf{Dirichlet}(\phi_k \mid \beta) \\
+&amp;\qquad + \sum_{m=1}^M \sum_{n=1}^{N[m]} \log \left(
+    \sum_{z=1}^K \textsf{categorical}(z \mid \theta_m)
+    \times \textsf{categorical}(w_{m,n} \mid \phi_z)
+  \right)
+\end{align*}\]</span></p>
+</section>
+<section id="implementation-of-lda" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="implementation-of-lda">Implementation of LDA</h3>
+<p>Applying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;               <span class="co">// num topics</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; V;               <span class="co">// num words</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; M;               <span class="co">// num docs</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;               <span class="co">// total word instances</span></span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=V&gt; w;    <span class="co">// word n</span></span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=M&gt; doc;  <span class="co">// doc ID for word n</span></span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] alpha;     <span class="co">// topic prior</span></span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[V] beta;      <span class="co">// word prior</span></span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[M] <span class="dt">simplex</span>[K] theta;    <span class="co">// topic dist for doc m</span></span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">simplex</span>[V] phi;      <span class="co">// word dist for topic k</span></span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>    theta[m] ~ dirichlet(alpha);  <span class="co">// prior</span></span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>    phi[k] ~ dirichlet(beta);     <span class="co">// prior</span></span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb4-23"><a href="#cb4-23" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[K] <span class="dt">real</span> gamma;</span>
+<span id="cb4-24"><a href="#cb4-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb4-25"><a href="#cb4-25" aria-hidden="true" tabindex="-1"></a>      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);</span>
+<span id="cb4-26"><a href="#cb4-26" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb4-27"><a href="#cb4-27" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_sum_exp(gamma);  <span class="co">// likelihood;</span></span>
+<span id="cb4-28"><a href="#cb4-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-29"><a href="#cb4-29" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.</p>
+</section>
+<section id="correlated-topic-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="correlated-topic-model">Correlated topic model</h3>
+<p>To account for correlations in the distribution of topics for documents, <span class="citation" data-cites="BleiLafferty:2007">Blei and Lafferty (<a href="#ref-BleiLafferty:2007" role="doc-biblioref">2007</a>)</span> introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.</p>
+<p>The authors treat the prior as a fixed hyperparameter. They use an <span class="math inline">\(L_1\)</span>-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.</p>
+<section id="fixed-hyperparameter-correlated-topic-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="fixed-hyperparameter-correlated-topic-model">Fixed hyperparameter correlated topic model</h4>
+<p>The Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior <code>alpha</code> in the data declaration with the mean and covariance of the multivariate logistic normal prior.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... data as before without alpha ...</span></span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;          <span class="co">// topic mean</span></span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;   <span class="co">// topic covariance</span></span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Rather than drawing the simplex parameter <code>theta</code> from a Dirichlet, a parameter <code>eta</code> is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">simplex</span>[V] phi;     <span class="co">// word dist for topic k</span></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[M] <span class="dt">vector</span>[K] eta;      <span class="co">// topic dist for doc m</span></span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[M] <span class="dt">simplex</span>[K] theta;</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>    theta[m] = softmax(eta[m]);</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>    eta[m] ~ multi_normal(mu, Sigma);</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... model as before w/o prior for theta ...</span></span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="full-bayes-correlated-topic-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="full-bayes-correlated-topic-model">Full Bayes correlated topic model</h4>
+<p>By adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean <code>mu</code> and covariance <code>Sigma</code> from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix <code>Sigma</code> may be encoded as follows.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ... data block as before, but without alpha ...</span></span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;              <span class="co">// topic mean</span></span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[K] Omega;      <span class="co">// correlation matrix</span></span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] sigma;  <span class="co">// scales</span></span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[M] <span class="dt">vector</span>[K] eta;    <span class="co">// logit topic dist for doc m</span></span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">simplex</span>[V] phi;   <span class="co">// word dist for topic k</span></span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... eta as above ...</span></span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;       <span class="co">// covariance matrix</span></span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a>    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];</span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:(K<span class="dv">-1</span>)) {</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> (m+<span class="dv">1</span>):K) {</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];</span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a>      Sigma[n, m] = Sigma[m, n];</span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);      <span class="co">// vectorized, diffuse</span></span>
+<span id="cb7-24"><a href="#cb7-24" aria-hidden="true" tabindex="-1"></a>  Omega ~ lkj_corr(<span class="fl">2.0</span>);  <span class="co">// regularize to unit correlation</span></span>
+<span id="cb7-25"><a href="#cb7-25" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);   <span class="co">// half-Cauchy due to constraint</span></span>
+<span id="cb7-26"><a href="#cb7-26" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... words sampled as above ...</span></span>
+<span id="cb7-27"><a href="#cb7-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <span class="math inline">\(\textsf{LKJCorr}\)</span> distribution with shape <span class="math inline">\(\alpha &gt; 0\)</span> has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by <span class="math display">\[
+\mathsf{LkjCorr}(\Omega\mid\alpha) \propto \mathrm{det}(\Omega)^{\alpha - 1}
+\]</span> With a scale of <span class="math inline">\(\alpha = 2\)</span>, the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix <span class="math inline">\(\Sigma\)</span> for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on <code>sigma</code>.</p>
+
+
+
+</section>
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-BleiLafferty:2007" class="csl-entry" role="listitem">
+Blei, David M., and John D. Lafferty. 2007. <span>“A Correlated Topic Model of <em><span>S</span>cience</em>.”</span> <em>The Annals of Applied Statistics</em> 1 (1): 17–37.
+</div>
+<div id="ref-BleiNgJordan:2003" class="csl-entry" role="listitem">
+Blei, David M., Andrew Y. Ng, and Michael I. Jordan. 2003. <span>“Latent <span>D</span>irichlet Allocation.”</span> <em>Journal of Machine Learning Research</em> 3: 993–1022.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-GelmanHill:2007" class="csl-entry" role="listitem">
+Gelman, Andrew, and Jennifer Hill. 2007. <em>Data Analysis Using Regression and Multilevel-Hierarchical Models</em>. Cambridge, United Kingdom: Cambridge University Press.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>For clustering, the non-identifiability problems for all mixture models present a problem, whereas there is no such problem for classification. Despite the difficulties with full Bayesian inference for clustering, researchers continue to use it, often in an exploratory data analysis setting rather than for predictive modeling.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/sparse-ragged.html" class="pagination-link" aria-label="Sparse and Ragged Data Structures">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Sparse and Ragged Data Structures</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/gaussian-processes.html" class="pagination-link" aria-label="Gaussian Processes">
+        <span class="nav-page-text">Gaussian Processes</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/clustering.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/complex-numbers.html b/docs/2_39/stan-users-guide/complex-numbers.html
new file mode 100644
index 000000000..09e110b52
--- /dev/null
+++ b/docs/2_39/stan-users-guide/complex-numbers.html
@@ -0,0 +1,1350 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Complex Numbers</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/dae.html" rel="next">
+<link href="../stan-users-guide/one-dimensional-integrals.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Complex Numbers">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/complex-numbers.html">Complex Numbers</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#complex-numbers" id="toc-complex-numbers" class="nav-link active" data-scroll-target="#complex-numbers">Complex Numbers</a>
+  <ul class="collapse">
+  <li><a href="#working-with-complex-numbers" id="toc-working-with-complex-numbers" class="nav-link" data-scroll-target="#working-with-complex-numbers">Working with complex numbers</a>
+  <ul class="collapse">
+  <li><a href="#constructing-and-accessing-complex-numbers" id="toc-constructing-and-accessing-complex-numbers" class="nav-link" data-scroll-target="#constructing-and-accessing-complex-numbers">Constructing and accessing complex numbers</a></li>
+  <li><a href="#complex-assignment-and-promotion" id="toc-complex-assignment-and-promotion" class="nav-link" data-scroll-target="#complex-assignment-and-promotion">Complex assignment and promotion</a></li>
+  <li><a href="#complex-arrays" id="toc-complex-arrays" class="nav-link" data-scroll-target="#complex-arrays">Complex arrays</a></li>
+  <li><a href="#complex-functions" id="toc-complex-functions" class="nav-link" data-scroll-target="#complex-functions">Complex functions</a></li>
+  </ul></li>
+  <li><a href="#complex-random-variables" id="toc-complex-random-variables" class="nav-link" data-scroll-target="#complex-random-variables">Complex random variables</a></li>
+  <li><a href="#complex-matrices-and-vectors" id="toc-complex-matrices-and-vectors" class="nav-link" data-scroll-target="#complex-matrices-and-vectors">Complex matrices and vectors</a></li>
+  <li><a href="#complex-linear-regression" id="toc-complex-linear-regression" class="nav-link" data-scroll-target="#complex-linear-regression">Complex linear regression</a>
+  <ul class="collapse">
+  <li><a href="#independent-real-and-imaginary-error" id="toc-independent-real-and-imaginary-error" class="nav-link" data-scroll-target="#independent-real-and-imaginary-error">Independent real and imaginary error</a></li>
+  <li><a href="#dependent-complex-error" id="toc-dependent-complex-error" class="nav-link" data-scroll-target="#dependent-complex-error">Dependent complex error</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/complex-numbers.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/complex-numbers.html">Complex Numbers</a></li></ol></nav></header>
+
+
+
+
+<section id="complex-numbers" class="level1">
+<h1>Complex Numbers</h1>
+<p>Stan supports complex scalars, matrices, and vectors as well as real-based ones.</p>
+<section id="working-with-complex-numbers" class="level2">
+<h2 class="anchored" data-anchor-id="working-with-complex-numbers">Working with complex numbers</h2>
+<p>This section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.</p>
+<section id="constructing-and-accessing-complex-numbers" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constructing-and-accessing-complex-numbers">Constructing and accessing complex numbers</h3>
+<p>Complex numbers can be constructed using imaginary literals. For example,</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = -<span class="fl">1.1</span> + <span class="fl">2.3i</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>produces the complex number <span class="math inline">\(-1.1 + 2.3i\)</span>. This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the <code>to_complex()</code> function may be used. For example, the following code will work if <code>x</code> and <code>y</code> are parameters, transformed data, or local variables in a function or model block.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="co">// ...</span></span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> y = <span class="co">// ...</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = to_complex(x, y);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The real and imaginary parts of the complex number can be accessed with getters as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = get_real(z);  <span class="co">// x = -1.1</span></span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> y = get_imag(z);  <span class="co">// y = 2.3</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> a = <span class="fl">3.2</span> + <span class="fl">2i</span>;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> b = to_complex(<span class="fl">3.2</span>, <span class="dv">2</span>);</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (a == b) <span class="kw">print</span>(<span class="st">"hello"</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="complex-assignment-and-promotion" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-assignment-and-promotion">Complex assignment and promotion</h3>
+<p>Integer- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z1 = <span class="dv">3</span>;  <span class="co">// int promoted to 3 + 0i</span></span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z2 = <span class="fl">3.2</span>;  <span class="co">// real promoted to 3.2 + 0.0i</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="complex-arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-arrays">Complex arrays</h3>
+<p>Arrays of complex numbers work as usual and allow the usual curly bracket constructors.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z1;  <span class="dt">complex</span> z2;  <span class="dt">complex</span> z3;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">complex</span> zs = { z1, z2, z3 };</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (z <span class="cf">in</span> zs) {</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">print</span>(z);</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.</p>
+</section>
+<section id="complex-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complex-functions">Complex functions</h3>
+<p>All of the standard complex functions are available, including natural logarithm <code>log(z)</code>, natural exponentiation <code>exp(z)</code>, and powers <code>pow(z1, z2)</code>, as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as <code>sin(z)</code>, <code>acos(z)</code>, <code>tanh(z)</code> and <code>asinh(z)</code>.</p>
+<p>Promotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> times_i(<span class="dt">complex</span> z) {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex</span> i = to_complex(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> i * z;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For instance, <code>times_i(1)</code> evaluates to the imaginary base <span class="math inline">\(i\)</span>, as does <code>times_i(1.0)</code>.</p>
+</section>
+</section>
+<section id="complex-random-variables" class="level2">
+<h2 class="anchored" data-anchor-id="complex-random-variables">Complex random variables</h2>
+<p>The simplest way to model a distribution over a complex random number <span class="math inline">\(z = x + yi\)</span> is to consider its real part <span class="math inline">\(x\)</span> and imaginary part <span class="math inline">\(y\)</span> to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z;</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">2</span>] mu;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a><span class="dt">cholesky_factor_cov</span>[<span class="dv">2</span>] L_Sigma;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For example, if <code>z</code> is data, this can be used to estimate <code>mu</code> and the covariance Cholesky factor <code>L_Sigma</code>. Alternatively, if <code>z</code> is a parameter, <code>mu</code> and <code>L_Sigma</code> may constants defining a prior or further parameters defining a hierarchical model.</p>
+</section>
+<section id="complex-matrices-and-vectors" class="level2">
+<h2 class="anchored" data-anchor-id="complex-matrices-and-vectors">Complex matrices and vectors</h2>
+<p>Stan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_vector</span>[<span class="dv">3</span>] v;</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_row_vector</span>[<span class="dv">2</span>] rv;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex_matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] m;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>We can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of <code>rv</code> above, we could assign it to a constructed row vector.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>rv =  [<span class="dv">2</span> + <span class="fl">3i</span>, <span class="fl">1.9</span> - <span class="fl">2.3i</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Complex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.</p>
+<p>Furthermore, it is possible to convert back and forth between arrays and matrices using the <code>to_array</code> functions.</p>
+</section>
+<section id="complex-linear-regression" class="level2">
+<h2 class="anchored" data-anchor-id="complex-linear-regression">Complex linear regression</h2>
+<p>Complex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take <code>x</code> to be predictors, <code>y</code> to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex_vector</span>[N] x;</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex_vector</span>[N] y;</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex</span> alpha;</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex</span> beta;</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">complex_vector</span>[N] eps = y - (alpha + beta * x);</span>
+<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a>  eps ~  <span class="co">// ...error distribution...</span></span>
+<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.</p>
+<section id="independent-real-and-imaginary-error" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="independent-real-and-imaginary-error">Independent real and imaginary error</h3>
+<p>The simplest approach to error in complex regression is to give the real and imaginary parts of <code>eps_n</code> independent independent normal distributions, as follows.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] sigma;</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>  get_real(eps) ~ normal(<span class="dv">0</span>, sigma[<span class="dv">1</span>]);</span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>  get_imag(eps) ~ normal(<span class="dv">0</span>, sigma[<span class="dv">2</span>]);</span>
+<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a>  sigma ~ <span class="co">//...hyperprior...</span></span>
+<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A new error scale vector <code>sigma</code> is introduced, and it should itself get a prior based on the expected scale of error for the problem.</p>
+</section>
+<section id="dependent-complex-error" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="dependent-complex-error">Dependent complex error</h3>
+<p>The next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[<span class="dv">2</span>] L_Omega;  <span class="co">// correlation matrix</span></span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] sigma;                  <span class="co">// real, imag error scales</span></span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[<span class="dv">2</span>] eps_arr;</span>
+<span id="cb13-9"><a href="#cb13-9" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb13-10"><a href="#cb13-10" aria-hidden="true" tabindex="-1"></a>    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };</span>
+<span id="cb13-11"><a href="#cb13-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb13-12"><a href="#cb13-12" aria-hidden="true" tabindex="-1"></a>  eps_arr ~ multi_normal_cholesky([<span class="dv">0</span>, <span class="dv">0</span>]',</span>
+<span id="cb13-13"><a href="#cb13-13" aria-hidden="true" tabindex="-1"></a>                                  diag_pre_multiply(sigma, L_Omega));</span>
+<span id="cb13-14"><a href="#cb13-14" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_cholesky(<span class="dv">4</span>);  <span class="co">// shrink toward diagonal correlation</span></span>
+<span id="cb13-15"><a href="#cb13-15" aria-hidden="true" tabindex="-1"></a>  sigma ~ <span class="co">// ... hyperprior ...</span></span>
+<span id="cb13-16"><a href="#cb13-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector <code>sigma</code> and a Cholesky factor or a correlation matrix, <code>L_Omega</code>. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to <code>multi_normal_cholesky</code>, the vector of complex numbers needs to be converted to an array of size 2 vectors.</p>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/one-dimensional-integrals.html" class="pagination-link" aria-label="Computing One Dimensional Integrals">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Computing One Dimensional Integrals</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/dae.html" class="pagination-link" aria-label="Differential-Algebraic Equations">
+        <span class="nav-page-text">Differential-Algebraic Equations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/complex-numbers.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/copulas.html b/docs/2_39/stan-users-guide/copulas.html
new file mode 100644
index 000000000..b90c1957d
--- /dev/null
+++ b/docs/2_39/stan-users-guide/copulas.html
@@ -0,0 +1,1441 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Copulas</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/hyperspherical-models.html" rel="next">
+<link href="../stan-users-guide/gaussian-processes.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Copulas">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/copulas.html">Copulas</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#copulas.chapter" id="toc-copulas.chapter" class="nav-link active" data-scroll-target="#copulas.chapter">Copulas</a>
+  <ul class="collapse">
+  <li><a href="#what-are-copulas" id="toc-what-are-copulas" class="nav-link" data-scroll-target="#what-are-copulas">What Are Copulas?</a></li>
+  <li><a href="#general-structure-of-copula-models-in-stan" id="toc-general-structure-of-copula-models-in-stan" class="nav-link" data-scroll-target="#general-structure-of-copula-models-in-stan">General Structure of Copula Models in Stan</a></li>
+  <li><a href="#gaussian-copula-example" id="toc-gaussian-copula-example" class="nav-link" data-scroll-target="#gaussian-copula-example">Gaussian Copula Example</a></li>
+  <li><a href="#advantages-of-copulas" id="toc-advantages-of-copulas" class="nav-link" data-scroll-target="#advantages-of-copulas">Advantages of Copulas</a></li>
+  <li><a href="#common-pitfalls-and-considerations" id="toc-common-pitfalls-and-considerations" class="nav-link" data-scroll-target="#common-pitfalls-and-considerations">Common Pitfalls and Considerations</a></li>
+  <li><a href="#common-copula-families" id="toc-common-copula-families" class="nav-link" data-scroll-target="#common-copula-families">Common Copula Families</a></li>
+  <li><a href="#further-readingviewing" id="toc-further-readingviewing" class="nav-link" data-scroll-target="#further-readingviewing">Further reading/viewing</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/copulas.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/copulas.html">Copulas</a></li></ol></nav></header>
+
+
+
+
+<section id="copulas.chapter" class="level1">
+<h1>Copulas</h1>
+<p>Copulas provide a flexible way to model multivariate distributions by separating the marginal cumulative distribution functions from the dependence structure. This chapter introduces copulas in Stan, focusing on implementation techniques and practical examples. This chapter was derived from Brynjólfur Gauti Guðrúnar Jónsson’s <a href="https://bggj.is/posts/gaussian-copula/">A gentle introduction: the Gaussian copula</a>.</p>
+<section id="what-are-copulas" class="level2">
+<h2 class="anchored" data-anchor-id="what-are-copulas">What Are Copulas?</h2>
+<p>According to Sklar’s theorem <span class="citation" data-cites="sklar1959">(<a href="#ref-sklar1959" role="doc-biblioref">Sklar 1959</a>)</span>, any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.</p>
+<p>For a multivariate random variable <span class="math inline">\(\mathbf{X} = [X_1 \cdots X_D]^\top\)</span> with marginal cumulative distribution functions <span class="math inline">\(F_i\)</span>, the joint cumulative distribution function can be written as:</p>
+<p><span class="math display">\[
+F_{\mathbf{X}}(\mathbf{x}) = C(F_1(x_1), \ldots, F_D(x_D)) = \Pr[X_1 \leq x_1, \ldots, X_D \leq x_D]
+\]</span></p>
+<p>where <span class="math inline">\(C\)</span> is the copula function, <span class="math inline">\(F_{\mathbf{X}}\)</span> is the joint cumulative distribution function, and <span class="math inline">\(F_i\)</span> are the marginal cumulative distribution functions. The copula function <span class="math inline">\(C\)</span> must be a joint cumulative distribution function over the unit hypercube <span class="math inline">\([0, 1]^D\)</span>.</p>
+</section>
+<section id="general-structure-of-copula-models-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="general-structure-of-copula-models-in-stan">General Structure of Copula Models in Stan</h2>
+<p>This section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.</p>
+<p>The log density of a multivariate distribution using a copula can be written as:</p>
+<p><span class="math display">\[
+\log h(\mathbf{x}) = \log c\left(u_1, \dots, u_D \vert \boldsymbol{\alpha}\right) + \sum_{i=1}^D \log f_i(x_i \vert \boldsymbol{\beta}_i)
+\]</span></p>
+<p>where:</p>
+<ul>
+<li><span class="math inline">\(u_i = F_i(x_i \vert \boldsymbol{\beta}_i)\)</span> are the probability integral transforms of the data</li>
+<li><span class="math inline">\(\log c\left(u_1, \dots, u_D \vert \boldsymbol{\alpha}\right)\)</span> is the log density of the copula</li>
+<li><span class="math inline">\(\sum_{i=1}^D \log f_i(x_i \vert \boldsymbol{\beta}_i)\)</span> is the sum of the log densities of the marginals</li>
+<li><span class="math inline">\(\boldsymbol{\alpha}\)</span> represents the parameters describing the parametric form of the copula</li>
+<li><span class="math inline">\(\boldsymbol{\beta}_i\)</span> represents the parameters describing the parametric form of the <span class="math inline">\(i\)</span>-th marginal distribution</li>
+</ul>
+<p>The implementation of copulas in Stan has two key requirements:</p>
+<ol type="1">
+<li>Both the probability density functions and cumulative distribution functions of the marginal distributions must be available</li>
+<li>A function that computes the log density of the copula for the transformed data must be implemented</li>
+</ol>
+<p>Most copula implementations in Stan follow a three-step process:</p>
+<ol type="1">
+<li><strong>Accumulate marginal log likelihoods</strong>: Calculate and add the log density of each marginal distribution to the target log density</li>
+<li><strong>Transform to uniform variables</strong>: Apply the marginal CDFs to transform the data to uniform variables on the unit interval</li>
+<li><strong>Calculate copula density</strong>: Compute the log density of the copula based on these uniform variables and add it to the target log density</li>
+</ol>
+<p>This process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.</p>
+<p>In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where <span class="math inline">\(\log c(\mathbf{u}) = 0\)</span>, resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.</p>
+<p>Most parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).</p>
+</section>
+<section id="gaussian-copula-example" class="level2">
+<h2 class="anchored" data-anchor-id="gaussian-copula-example">Gaussian Copula Example</h2>
+<p>The Gaussian copula is constructed using the multivariate normal distribution. For a <span class="math inline">\(D\)</span>-dimensional random vector <span class="math inline">\(\mathbf{X}\)</span> with marginals <span class="math inline">\(F_i\)</span>, the log Gaussian copula density is given by:</p>
+<p><span class="math display">\[
+\begin{aligned}
+\log c(\mathbf{u}) &amp;=
+-\frac{1}{2} \log |\boldsymbol{\Omega}| -\frac{1}{2} \mathbf{z}^\top (\boldsymbol{\Omega}^{-1} - \mathbf{I}) \mathbf{z} \\
+&amp; =
+-\frac{1}{2} \log |\boldsymbol{\Omega}| -\frac{1}{2} \mathbf{z}^\top \boldsymbol{\Omega}^{-1} \mathbf{z} + \frac{1}{2} \mathbf{z}^\top \mathbf{z} \\
+&amp;= \log \mathcal{N}(\mathbf{z} \mid \mathbf{0}, \boldsymbol{\Omega}) - \log \mathcal{N}(\mathbf{z} \mid \mathbf{0}, \mathbf{I})
+\end{aligned}
+\]</span></p>
+<p>where <span class="math inline">\(\mathbf{z} = [\Phi^{-1}(u_1), \ldots, \Phi^{-1}(u_D)]^\top\)</span> are the inverse normal CDF transforms of the uniform marginals, <span class="math inline">\(\boldsymbol{\Omega}\)</span> is the correlation matrix, and <span class="math inline">\(\mathbf{I}\)</span> is the identity matrix. The joint log density is then:</p>
+<p><span class="math display">\[
+\log h(\mathbf{x}) = \log c(F_1(x_1), \ldots, F_D(x_D)) + \sum_{i=1}^D \log f_i(x_i)
+\]</span></p>
+<p>Following the three-step process for implementing copulas in Stan:</p>
+<ol type="1">
+<li><strong>Accumulate marginal log likelihoods</strong>: The exponential log densities are added to the target in the line <code>target += exponential_lpdf(y[n] | lambda)</code></li>
+<li><strong>Transform to uniform variables</strong>: The exponential CDF transforms the data to uniform variables: <code>exponential_cdf(y[n, d] | lambda[d])</code></li>
+<li><strong>Calculate copula density</strong>: The transformed variables are converted to normal scale using <code>inv_Phi</code> and the multivariate normal log density is computed: <code>z ~ multi_normal_cholesky(zeros, L_Omega)</code></li>
+</ol>
+<p>The following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;  <span class="co">// number of observations</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; D;  <span class="co">// number of dimensions</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[D] y[N];  <span class="co">// data</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[D] zeros = rep_vector(<span class="dv">0</span>, D);</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Parameters for exponential marginal distributions</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[D] lambda;  <span class="co">// rate parameters</span></span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  </span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Correlation matrix for Gaussian copula</span></span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Priors</span></span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>  lambda ~ gamma(<span class="dv">2</span>, <span class="dv">1</span>);  <span class="co">// prior for rate parameters</span></span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
+<span id="cb1-23"><a href="#cb1-23" aria-hidden="true" tabindex="-1"></a>  </span>
+<span id="cb1-24"><a href="#cb1-24" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Likelihood using Gaussian copula with exponential marginals</span></span>
+<span id="cb1-25"><a href="#cb1-25" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-26"><a href="#cb1-26" aria-hidden="true" tabindex="-1"></a>    <span class="co">// Add exponential log density to target</span></span>
+<span id="cb1-27"><a href="#cb1-27" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> exponential_lpdf(y[n] | lambda);</span>
+<span id="cb1-28"><a href="#cb1-28" aria-hidden="true" tabindex="-1"></a>    </span>
+<span id="cb1-29"><a href="#cb1-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[D] z;</span>
+<span id="cb1-30"><a href="#cb1-30" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
+<span id="cb1-31"><a href="#cb1-31" aria-hidden="true" tabindex="-1"></a>      <span class="co">// Transform to uniform using exponential CDF</span></span>
+<span id="cb1-32"><a href="#cb1-32" aria-hidden="true" tabindex="-1"></a>      <span class="dt">real</span> u_d = exponential_cdf(y[n, d] | lambda[d]);</span>
+<span id="cb1-33"><a href="#cb1-33" aria-hidden="true" tabindex="-1"></a>      </span>
+<span id="cb1-34"><a href="#cb1-34" aria-hidden="true" tabindex="-1"></a>      <span class="co">// Transform to standard normal</span></span>
+<span id="cb1-35"><a href="#cb1-35" aria-hidden="true" tabindex="-1"></a>      z[d] = inv_Phi(u_d);</span>
+<span id="cb1-36"><a href="#cb1-36" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-37"><a href="#cb1-37" aria-hidden="true" tabindex="-1"></a>    <span class="co">// Multivariate normal log density with correlation matrix</span></span>
+<span id="cb1-38"><a href="#cb1-38" aria-hidden="true" tabindex="-1"></a>    z ~ multi_normal_cholesky(zeros, L_Omega);</span>
+<span id="cb1-39"><a href="#cb1-39" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-40"><a href="#cb1-40" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-41"><a href="#cb1-41" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-42"><a href="#cb1-42" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-43"><a href="#cb1-43" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Optional: Recover correlation matrix from Cholesky factor</span></span>
+<span id="cb1-44"><a href="#cb1-44" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);</span>
+<span id="cb1-45"><a href="#cb1-45" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="advantages-of-copulas" class="level2">
+<h2 class="anchored" data-anchor-id="advantages-of-copulas">Advantages of Copulas</h2>
+<p>Copulas offer several advantages in statistical modeling:</p>
+<ol type="1">
+<li><p><strong>Flexibility</strong>: They allow combining any marginal distributions with various dependence structures. For example:</p>
+<ul>
+<li>Modeling financial returns with heavy-tailed marginals and complex dependence structures</li>
+<li>Combining different types of distributions (e.g., normal and gamma) in a single model</li>
+<li>Capturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic</li>
+<li>Modeling different types of tail dependence in different parts of the distribution</li>
+</ul></li>
+<li><p><strong>Factorability</strong>: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.</p>
+<p>For example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.</p></li>
+<li><p><strong>Tail dependence</strong>: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.</p></li>
+<li><p><strong>Universal Framework</strong>: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.</p></li>
+</ol>
+</section>
+<section id="common-pitfalls-and-considerations" class="level2">
+<h2 class="anchored" data-anchor-id="common-pitfalls-and-considerations">Common Pitfalls and Considerations</h2>
+<p>When implementing copulas in Stan, several considerations should be kept in mind:</p>
+<ol type="1">
+<li><p><strong>Computational efficiency</strong>: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.</p></li>
+<li><p><strong>Parameter identifiability</strong>: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.</p></li>
+<li><p><strong>Model selection</strong>: The choice of copula family should be guided by the specific dependence structure of the data. For example:</p>
+<ul>
+<li>The Gaussian copula may underestimate the probability of joint extreme events in financial data</li>
+<li>The Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications</li>
+<li>Archimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions</li>
+</ul></li>
+<li><p><strong>Numerical stability</strong>: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.</p></li>
+<li><p><strong>Symmetry considerations</strong>: Many copula families exhibit strong symmetries that may not match the data:</p>
+<ul>
+<li><strong>Radial symmetry</strong>: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns</li>
+<li><strong>Exchangeability</strong>: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.</li>
+</ul></li>
+<li><p><strong>Tail dependence</strong>: Understanding and choosing appropriate tail dependence is crucial:</p>
+<ul>
+<li>The upper (lower) tail dependence coefficient <span class="math inline">\(\lambda_U (\lambda_L)\)</span> is the probability that one variable is extremely large (small) given that another is extremely large (small).</li>
+<li>Different copula families exhibit different tail dependence properties:
+<ul>
+<li>Some copulas (like Gaussian) have zero tail dependence</li>
+<li>Others can model symmetric tail dependence (<span class="math inline">\(\lambda_U = \lambda_L\)</span>)</li>
+<li>Some can capture asymmetric tail dependence (<span class="math inline">\(\lambda_U \neq \lambda_L\)</span>)</li>
+<li>Certain copulas allow for tail dependence even with zero correlation</li>
+</ul></li>
+<li>The choice of copula should be guided by the expected tail behavior in the application:
+<ul>
+<li>Financial data often requires modeling joint lower extreme events</li>
+<li>Risk management applications may need asymmetric tail dependence</li>
+<li>Some applications may require different tail behavior in different parts of the distribution</li>
+</ul></li>
+</ul></li>
+<li><p><strong>High-dimensional modeling</strong>: As dimensionality increases:</p>
+<ul>
+<li>The number of dependence parameters grows</li>
+<li>Some copula families become less flexible</li>
+<li>Vine copulas or factor copulas may be more appropriate</li>
+</ul></li>
+</ol>
+</section>
+<section id="common-copula-families" class="level2">
+<h2 class="anchored" data-anchor-id="common-copula-families">Common Copula Families</h2>
+<p>Several copula families are available for modeling different dependence structures in the correlation component:</p>
+<ul>
+<li><strong>Gaussian copula</strong>: Based on the multivariate normal distribution, offering symmetric dependence</li>
+<li><strong>Student-t copula</strong>: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula</li>
+<li><strong>Archimedean copulas</strong>: A class of copulas defined through generator functions, including:
+<ul>
+<li>Clayton copula: Stronger lower tail dependence</li>
+<li>Gumbel copula: Stronger upper tail dependence</li>
+<li>Frank copula: Symmetric dependence</li>
+</ul></li>
+<li><strong>Vine copulas</strong>: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas</li>
+</ul>
+</section>
+<section id="further-readingviewing" class="level2">
+<h2 class="anchored" data-anchor-id="further-readingviewing">Further reading/viewing</h2>
+<ul>
+<li>Jónsson’s three part blog series, <em>Copulas in Stan</em>:
+<ul>
+<li>Part I: <a href="https://bggj.is/posts/stan-copulas-1/">If it bleeds, we can kill it</a></li>
+<li>Part II: <a href="https://bggj.is/posts/gaussian-copula/">A gentle introduction: the Gaussian copula</a></li>
+<li>Part III: <a href="https://bggj.is/posts/t-copula/">It was the best of tails, it was the worst of tails: The T-Copula</a></li>
+</ul></li>
+<li>Brynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, <a href="https://www.youtube.com/watch?v=aHZ-rnepUNk">Copulas in Stan: Modeling Spatial Dependence</a></li>
+<li>Sean Pinkney’s <em>Helpful Stan functions</em>: <a href="https://spinkney.github.io/helpful_stan_functions/group__copula.html">copula functions</a></li>
+</ul>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-sklar1959" class="csl-entry" role="listitem">
+Sklar, Abe. 1959. <span>“Abe Sklar’s "FONCTIONS DE REPARTITION a n DIMENSIONS ET LEURS MARGES": The Original Document and an English Translation.”</span> Translated by Ben Van Vliet. <em>The Original Document and an English Translation (March 3, 2023)</em>. <a href="https://dx.doi.org/10.2139/ssrn.4198458">https://dx.doi.org/10.2139/ssrn.4198458</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/gaussian-processes.html" class="pagination-link" aria-label="Gaussian Processes">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Gaussian Processes</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/hyperspherical-models.html" class="pagination-link" aria-label="Directions, Rotations, and Hyperspheres">
+        <span class="nav-page-text">Directions, Rotations, and Hyperspheres</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/copulas.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/cross-validation.html b/docs/2_39/stan-users-guide/cross-validation.html
new file mode 100644
index 000000000..c2c63070e
--- /dev/null
+++ b/docs/2_39/stan-users-guide/cross-validation.html
@@ -0,0 +1,1520 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Held-Out Evaluation and Cross-Validation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/poststratification.html" rel="next">
+<link href="../stan-users-guide/posterior-predictive-checks.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Held-Out Evaluation and Cross-Validation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/cross-validation.html">Held-Out Evaluation and Cross-Validation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#held-out-evaluation-and-cross-validation" id="toc-held-out-evaluation-and-cross-validation" class="nav-link active" data-scroll-target="#held-out-evaluation-and-cross-validation">Held-Out Evaluation and Cross-Validation</a>
+  <ul class="collapse">
+  <li><a href="#evaluating-posterior-predictive.section" id="toc-evaluating-posterior-predictive.section" class="nav-link" data-scroll-target="#evaluating-posterior-predictive.section">Evaluating posterior predictive densities</a>
+  <ul class="collapse">
+  <li><a href="#stan-program" id="toc-stan-program" class="nav-link" data-scroll-target="#stan-program">Stan program</a></li>
+  </ul></li>
+  <li><a href="#estimation-error" id="toc-estimation-error" class="nav-link" data-scroll-target="#estimation-error">Estimation error</a>
+  <ul class="collapse">
+  <li><a href="#parameter-estimates" id="toc-parameter-estimates" class="nav-link" data-scroll-target="#parameter-estimates">Parameter estimates</a></li>
+  <li><a href="#predictive-estimates" id="toc-predictive-estimates" class="nav-link" data-scroll-target="#predictive-estimates">Predictive estimates</a></li>
+  <li><a href="#predictive-estimates-in-stan" id="toc-predictive-estimates-in-stan" class="nav-link" data-scroll-target="#predictive-estimates-in-stan">Predictive estimates in Stan</a></li>
+  </ul></li>
+  <li><a href="#cross-validation" id="toc-cross-validation" class="nav-link" data-scroll-target="#cross-validation">Cross-validation</a>
+  <ul class="collapse">
+  <li><a href="#stan-implementation-with-random-folds" id="toc-stan-implementation-with-random-folds" class="nav-link" data-scroll-target="#stan-implementation-with-random-folds">Stan implementation with random folds</a></li>
+  <li><a href="#user-defined-permutations" id="toc-user-defined-permutations" class="nav-link" data-scroll-target="#user-defined-permutations">User-defined permutations</a></li>
+  <li><a href="#cross-validation-with-structured-data" id="toc-cross-validation-with-structured-data" class="nav-link" data-scroll-target="#cross-validation-with-structured-data">Cross-validation with structured data</a></li>
+  <li><a href="#cross-validation-with-spatio-temporal-data" id="toc-cross-validation-with-spatio-temporal-data" class="nav-link" data-scroll-target="#cross-validation-with-spatio-temporal-data">Cross-validation with spatio-temporal data</a></li>
+  <li><a href="#approximate-cross-validation" id="toc-approximate-cross-validation" class="nav-link" data-scroll-target="#approximate-cross-validation">Approximate cross-validation</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/cross-validation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/cross-validation.html">Held-Out Evaluation and Cross-Validation</a></li></ol></nav></header>
+
+
+
+
+<section id="held-out-evaluation-and-cross-validation" class="level1">
+<h1>Held-Out Evaluation and Cross-Validation</h1>
+<p>Held-out evaluation involves splitting a data set into two parts, a training data set and a test data set. The training data is used to estimate the model and the test data is used for evaluation. Held-out evaluation is commonly used to declare winners in predictive modeling competitions such as those run by <a href="https://kaggle.com">Kaggle</a>.</p>
+<p>Cross-validation involves repeated held-out evaluations performed by partitioning a single data set in different ways. The training/test split can be done either by randomly selecting the test set, or by partitioning the data set into several equally-sized subsets and then using each subset in turn as the test data with the other folds as training data.</p>
+<p>Held-out evaluation and cross-validation may involve any kind of predictive statistics, with common choices being the predictive log density on test data, squared error of parameter estimates, or accuracy in a classification task.</p>
+<section id="evaluating-posterior-predictive.section" class="level2">
+<h2 class="anchored" data-anchor-id="evaluating-posterior-predictive.section">Evaluating posterior predictive densities</h2>
+<p>Given training data <span class="math inline">\((x, y)\)</span> consisting of parallel sequences of predictors and observations and test data <span class="math inline">\((\tilde{x}, \tilde{y})\)</span> of the same structure, the posterior predictive density is <span class="math display">\[
+p(\tilde{y} \mid \tilde{x}, x, y)
+=
+\int
+  p(\tilde{y} \mid \tilde{x}, \theta)
+  \cdot p(\theta \mid x, y)
+\, \textrm{d}\theta,
+\]</span></p>
+<p>where <span class="math inline">\(\theta\)</span> is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors <span class="math inline">\(\tilde{x}\)</span> and the training data <span class="math inline">\((x, y).\)</span></p>
+<p>This integral may be calculated with Monte Carlo methods as usual, <span class="math display">\[
+p(\tilde{y} \mid \tilde{x}, x, y)
+\approx
+\frac{1}{M} \sum_{m = 1}^M p(\tilde{y} \mid \tilde{x}, \theta^{(m)}),
+\]</span> where the <span class="math inline">\(\theta^{(m)} \sim p(\theta \mid x, y)\)</span> are draws from the posterior given only the training data <span class="math inline">\((x, y).\)</span></p>
+<p>To avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, <span class="math display">\[\begin{eqnarray*}
+\log p(\tilde{y} \mid \tilde{x}, x, y)
+&amp; \approx &amp;
+\log \frac{1}{M} \sum_{m = 1}^M p(\tilde{y} \mid \tilde{x}, \theta^{(m)}),
+\\[4pt]
+&amp; = &amp; -\log M + \log \sum_{m = 1}^M p(\tilde{y} \mid \tilde{x}, \theta^{(m)}),
+\\[4pt]
+&amp; = &amp; -\log M + \log \sum_{m = 1}^M \exp(\log p(\tilde{y} \mid \tilde{x}, \theta^{(m)}))
+\\[4pt]
+&amp; = &amp; -\log M + \textrm{log-sum-exp}_{m = 1}^M \log p(\tilde{y} \mid \tilde{x}, \theta^{(m)})
+\end{eqnarray*}\]</span> where the log sum of exponentials function is defined so as to make the above equation hold, <span class="math display">\[
+\textrm{log-sum-exp}_{m = 1}^M \, \mu_m
+= \log \sum_{m=1}^M \exp(\mu_m).
+\]</span> The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as <span class="math display">\[
+\textrm{log-sum-exp}_{m = 1}^M \mu_m
+= \textrm{max}(\mu)
++ \log \sum_{m = 1}^M \exp(\mu_m - \textrm{max}(\mu)).
+\]</span> Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms <span class="math inline">\(\mu_m - \textrm{max}(\mu) \leq 0\)</span>, and thus will not overflow.</p>
+<section id="stan-program" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-program">Stan program</h3>
+<p>To evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of <span class="math inline">\(\log p(\tilde{y} \mid \tilde{x},
+\theta^{(m)}).\)</span></p>
+<p>Here is the code for evaluating the log posterior predictive density in a simple linear regression of the test data <span class="math inline">\(\tilde{y}\)</span> given predictors <span class="math inline">\(\tilde{x}\)</span> and training data <span class="math inline">\((x, y).\)</span></p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] y_tilde;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Only the training data <code>x</code> and <code>y</code> are used in the model block. The test data <code>y_tilde</code> and test predictors <code>x_tilde</code> appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use <code>x_tilde</code> in the model block—only the test observations <code>y_tilde</code> are unknown before they are predicted.</p>
+<p>Given <span class="math inline">\(M\)</span> posterior draws from Stan, the sequence <code>log_p[1:M]</code> will be available, so that the log posterior predictive density of the test data given training data and predictors is just <code>log_sum_exp(log_p) - log(M)</code>.</p>
+</section>
+</section>
+<section id="estimation-error" class="level2">
+<h2 class="anchored" data-anchor-id="estimation-error">Estimation error</h2>
+<section id="parameter-estimates" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="parameter-estimates">Parameter estimates</h3>
+<p>Estimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If <span class="math inline">\(\theta\)</span> is the true value and <span class="math inline">\(\hat{\theta}\)</span> the estimate, then error is just the difference between the prediction and the true value, <span class="math display">\[
+\textrm{err} = \hat{\theta} - \theta.
+\]</span></p>
+<p>If the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. <em>Absolute error</em> is defined as <span class="math display">\[
+\textrm{abs-err} = \left| \hat{\theta} - \theta \right|
+\]</span> and <em>squared error</em> as <span class="math display">\[
+\textrm{sq-err} = \left( \hat{\theta} - \theta \right)^2.
+\]</span> <span class="citation" data-cites="GneitingRaftery:2007">Gneiting and Raftery (<a href="#ref-GneitingRaftery:2007" role="doc-biblioref">2007</a>)</span> provide a thorough overview of such scoring rules and their properties.</p>
+<p>Bayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.</p>
+</section>
+<section id="predictive-estimates" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="predictive-estimates">Predictive estimates</h3>
+<p>In addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, <span class="math inline">\(z_1, \ldots, z_N,\)</span> then it is common to report <em>mean square error</em> (MSE), <span class="math display">\[
+\textrm{mse}
+= \frac{1}{N} \sum_{n = 1}^N \left( \hat{z}_n - z_n\right)^2.
+\]</span> To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as <em>root mean square error</em> (RMSE), <span class="math display">\[
+\textrm{rmse} = \sqrt{\textrm{mean-sq-err}}.
+\]</span></p>
+</section>
+<section id="predictive-estimates-in-stan" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="predictive-estimates-in-stan">Predictive estimates in Stan</h3>
+<p>Consider a simple linear regression model, parameters for the intercept <span class="math inline">\(\alpha\)</span> and slope <span class="math inline">\(\beta\)</span>, along with predictors <span class="math inline">\(\tilde{x}_n\)</span>. The standard Bayesian estimate is the expected value of <span class="math inline">\(\tilde{y}\)</span> given the predictors and training data, <span class="math display">\[\begin{eqnarray*}
+\hat{\tilde{y}}_n
+&amp; = &amp; \mathbb{E}[\tilde{y}_n \mid \tilde{x}_n, x, y]
+\\[4pt]
+&amp; \approx &amp; \frac{1}{M} \sum_{m = 1}^M \tilde{y}_n^{(m)}
+\end{eqnarray*}\]</span> where <span class="math inline">\(\tilde{y}_n^{(m)}\)</span> is drawn from the data model <span class="math display">\[
+\tilde{y}_n^{(m)}
+\sim p(\tilde{y}_n \mid \tilde{x}_n, \alpha^{(m)}, \beta^{(m)}),
+\]</span> for parameters <span class="math inline">\(\alpha^{(m)}\)</span> and <span class="math inline">\(\beta^{(m)}\)</span> drawn from the posterior, <span class="math display">\[
+(\alpha^{(m)}, \beta^{(m)}) \sim p(\alpha, \beta \mid x, y).
+\]</span></p>
+<p>In the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws <span class="math inline">\(\tilde{y}_n^{(m)}\)</span>, the same result is obtained by averaging linear predictions, <span class="math display">\[\begin{eqnarray*}
+\hat{\tilde{y}}_n
+&amp; = &amp; \mathbb{E}\left[
+          \alpha + \beta \cdot \tilde{x}_n
+          \mid \tilde{x}_n, x, y
+       \right]
+\\[4pt]
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m = 1}^M
+  \alpha^{(m)} + \beta^{(m)} \cdot \tilde{x}_n.
+\end{eqnarray*}\]</span> This is possible because <span class="math display">\[
+\tilde{y}_n^{(m)} \sim \textrm{normal}(\tilde{y}_n \mid \alpha^{(m)} +
+\beta^{(m)} \cdot \tilde{x}_n, \sigma^{(m)}),
+\]</span> and the normal distribution has symmetric error so that the expectation of <span class="math inline">\(\tilde{y}_n^{(m)}\)</span> is the same as <span class="math inline">\(\alpha^{(m)} + \beta^{(m)} \cdot
+\tilde{x}_n\)</span>. Replacing the sampled quantity <span class="math inline">\(\tilde{y}_n^{(m)}\)</span> with its expectation is a general variance reduction technique for Monte Carlo estimates known as <em>Rao-Blackwellization</em> <span class="citation" data-cites="Rao:1945 Blackwell:1947">(<a href="#ref-Rao:1945" role="doc-biblioref">Rao 1945</a>; <a href="#ref-Blackwell:1947" role="doc-biblioref">Blackwell 1947</a>)</span>.</p>
+<p>In the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, <span class="math display">\[\begin{eqnarray*}
+\tilde{y}_n^{(m)}
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m = 1}^M
+  \left( \alpha^{(m)} + \beta^{(m)} \cdot \tilde{x}_n \right)
+\\[4pt]
+&amp; = &amp; \frac{1}{M} \sum_{m = 1}^M \alpha^{(m)}
+      + \frac{1}{M} \sum_{m = 1}^M (\beta^{(m)} \cdot \tilde{x}_n)
+\\[4pt]
+&amp; = &amp; \frac{1}{M} \sum_{m = 1}^M \alpha^{(m)}
+      + \left( \frac{1}{M} \sum_{m = 1}^M \beta^{(m)}\right) \cdot \tilde{x}_n
+\\[4pt]
+&amp; = &amp; \hat{\alpha} + \hat{\beta} \cdot \tilde{x}_n.
+\end{eqnarray*}\]</span></p>
+<p>In Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] tilde_y = alpha + beta * x_tilde;</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The posterior mean of <code>tilde_y</code> calculated by Stan is the Bayesian estimate <span class="math inline">\(\hat{\tilde{y}}.\)</span> The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.</p>
+</section>
+</section>
+<section id="cross-validation" class="level2">
+<h2 class="anchored" data-anchor-id="cross-validation">Cross-validation</h2>
+<p>Cross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation <span class="citation" data-cites="VehtariEtAl:2017">(<a href="#ref-VehtariEtAl:2017" role="doc-biblioref">Vehtari, Gelman, and Gabry 2017</a>)</span>.</p>
+<p>Partitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.</p>
+<section id="stan-implementation-with-random-folds" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-implementation-with-random-folds">Stan implementation with random folds</h3>
+<p>For the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[] <span class="dt">int</span> permutation_rng(<span class="dt">int</span> N) {</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[N] <span class="dt">int</span> y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span> : N) {</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>      y[n] = n;</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[N] theta = rep_vector(<span class="fl">1.0</span> / N, N);</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span> : size(y)) {</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> i = categorical_rng(theta);</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> temp = y[n];</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>      y[n] = y[i];</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>      y[i] = temp;</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> y;</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The name of the function must end in <code>_rng</code> because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers <code>1:N</code>.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>The transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=N&gt; N_test;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N_train = N - N_test;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> permutation = permutation_rng(N);</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_train] x_train = x[permutation[<span class="dv">1</span> : N_train]];</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_train] y_train = y[permutation[<span class="dv">1</span> : N_train]];</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_test] x_test = x[permutation[N_train + <span class="dv">1</span> : N]];</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_test] y_test = y[permutation[N_train + <span class="dv">1</span> : N]];</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Recall that in Stan, <code>permutation[1:N_train]</code> is an array of integers, so that <code>x[permutation[1 : N_train]]</code> is a vector defined for <code>i in 1:N_train</code> by</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a>x[permutation[<span class="dv">1</span> : N_train]][i] = x[permutation[<span class="dv">1</span>:N_train][i]]</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>                               = x[permutation[i]]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Given the test/train split, the rest of the model is straightforward.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  y_train ~ normal(alpha + beta * x_train, sigma);</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  { alpha, beta, sigma } ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] err = y_test_sim - y_hat;</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The prediction <code>y_test_hat</code> is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, <span class="math display">\[\begin{eqnarray*}
+\hat{y}^{\textrm{test}}
+&amp; = &amp; \mathbb{E}\left[ y^{\textrm{test}} \mid x^{\textrm{test}}, x^{\textrm{train}} y^{\textrm{train}} \right]
+\\[4pt]
+&amp; \approx &amp; \frac{1}{M} \sum_{m = 1}^M \hat{y}^{\textrm{test}(m)}.
+\end{eqnarray*}\]</span></p>
+<p>Because the test set is constant and the expectation operator is linear, the posterior mean of <code>err</code> as defined in the Stan program will be the error of the posterior mean estimate, <span class="math display">\[\begin{eqnarray*}
+  \hat{y}^{\textrm{test}} - y^{\textrm{test}}
+&amp; = &amp;
+\mathbb{E}\left[
+  \hat{y}^{\textrm{test}}
+  \mid x^{\textrm{test}}, x^{\textrm{train}}, y^{\textrm{train}}
+\right]
+  - y^{\textrm{test}}
+\\[4pt]
+&amp; = &amp;
+\mathbb{E}\left[
+  \hat{y}^{\textrm{test}} - y^{\textrm{test}}
+  \mid x^{\textrm{test}}, x^{\textrm{train}}, y^{\textrm{train}}
+\right]
+\\[4pt]
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m = 1}^M \hat{y}^{\textrm{test}(m)} - y^{\textrm{test}},
+\end{eqnarray*}\]</span> where <span class="math display">\[
+\hat{y}^{\textrm{test}(m)}
+\sim p(y \mid x^{\textrm{test}}, x^{\textrm{train}},
+y^{\textrm{train}}).
+\]</span> This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should <em>not</em> be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.</p>
+<p>Because the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.</p>
+</section>
+<section id="user-defined-permutations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="user-defined-permutations">User-defined permutations</h3>
+<p>It is straightforward to declare the variable <code>permutation</code> in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.</p>
+</section>
+<section id="cross-validation-with-structured-data" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cross-validation-with-structured-data">Cross-validation with structured data</h3>
+<p>Cross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to <a href="../stan-users-guide/posterior-predictive-checks.html#mixed-replication">mixed replication in posterior predictive checking</a>, where there is a choice to simulate new elements of existing groups or generate entirely new groups.</p>
+<p>Education testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.</p>
+</section>
+<section id="cross-validation-with-spatio-temporal-data" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cross-validation-with-spatio-temporal-data">Cross-validation with spatio-temporal data</h3>
+<p>Often data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.</p>
+</section>
+<section id="approximate-cross-validation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="approximate-cross-validation">Approximate cross-validation</h3>
+<p><span class="citation" data-cites="VehtariEtAl:2017">Vehtari, Gelman, and Gabry (<a href="#ref-VehtariEtAl:2017" role="doc-biblioref">2017</a>)</span> introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo <span class="citation" data-cites="GabryEtAl:2019">(<a href="#ref-GabryEtAl:2019" role="doc-biblioref">Gabry et al. 2019</a>)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Blackwell:1947" class="csl-entry" role="listitem">
+Blackwell, David. 1947. <span>“Conditional Expectation and Unbiased Sequential Estimation.”</span> <em>The Annals of Mathematical Statistics</em> 18 (1): 105–10. <a href="https://doi.org/10.1214/aoms/1177730497">https://doi.org/10.1214/aoms/1177730497</a>.
+</div>
+<div id="ref-GabryEtAl:2019" class="csl-entry" role="listitem">
+Gabry, Jonah, Daniel Simpson, Aki Vehtari, Michael Betancourt, and Andrew Gelman. 2019. <span>“Visualization in <span>B</span>ayesian Workflow.”</span> <em>Journal of the Royal Statistical Society: Series A (Statistics in Society)</em> 182 (2): 389–402.
+</div>
+<div id="ref-GneitingRaftery:2007" class="csl-entry" role="listitem">
+Gneiting, Tilmann, and Adrian E Raftery. 2007. <span>“Strictly Proper Scoring Rules, Prediction, and Estimation.”</span> <em>Journal of the American Statistical Association</em> 102 (477): 359–78.
+</div>
+<div id="ref-Rao:1945" class="csl-entry" role="listitem">
+Rao, C. Radhakrishna. 1945. <span>“Information and Accuracy Attainable in the Estimation of Statistical Parameters.”</span> <em>Bulletin of the Calcutta Math Society</em> 37 (3): 81–91.
+</div>
+<div id="ref-VehtariEtAl:2017" class="csl-entry" role="listitem">
+Vehtari, Aki, Andrew Gelman, and Jonah Gabry. 2017. <span>“Practical <span>B</span>ayesian Model Evaluation Using Leave-One-Out Cross-Validation and <span>WAIC</span>.”</span> <em>Statistics and Computing</em> 27 (5): 1413–32.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The traditional approach is to walk through a vector and replace each item with a random element from the remaining elements, which is guaranteed to only move each item once. This was not done here as it’d require new categorical <code>theta</code> because Stan does not have a uniform discrete RNG built in.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/posterior-predictive-checks.html" class="pagination-link" aria-label="Posterior and Prior Predictive Checks">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Posterior and Prior Predictive Checks</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/poststratification.html" class="pagination-link" aria-label="Poststratification">
+        <span class="nav-page-text">Poststratification</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/cross-validation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/custom-probability.html b/docs/2_39/stan-users-guide/custom-probability.html
new file mode 100644
index 000000000..dcd0d5da8
--- /dev/null
+++ b/docs/2_39/stan-users-guide/custom-probability.html
@@ -0,0 +1,1326 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Custom Probability Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/proportionality-constants.html" rel="next">
+<link href="../stan-users-guide/user-functions.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Custom Probability Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/custom-probability.html">Custom Probability Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#custom-probability-functions.chapter" id="toc-custom-probability-functions.chapter" class="nav-link active" data-scroll-target="#custom-probability-functions.chapter">Custom Probability Functions</a>
+  <ul class="collapse">
+  <li><a href="#examples" id="toc-examples" class="nav-link" data-scroll-target="#examples">Examples</a>
+  <ul class="collapse">
+  <li><a href="#triangle-distribution" id="toc-triangle-distribution" class="nav-link" data-scroll-target="#triangle-distribution">Triangle distribution</a></li>
+  <li><a href="#exponential-distribution" id="toc-exponential-distribution" class="nav-link" data-scroll-target="#exponential-distribution">Exponential distribution</a></li>
+  <li><a href="#bivariate-normal-cumulative-distribution-function" id="toc-bivariate-normal-cumulative-distribution-function" class="nav-link" data-scroll-target="#bivariate-normal-cumulative-distribution-function">Bivariate normal cumulative distribution function</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/custom-probability.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/custom-probability.html">Custom Probability Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="custom-probability-functions.chapter" class="level1">
+<h1>Custom Probability Functions</h1>
+<p>Custom distributions may also be implemented directly within Stan’s programming language. The only thing that is needed is to increment the total log probability. The rest of the chapter provides examples.</p>
+<section id="examples" class="level2">
+<h2 class="anchored" data-anchor-id="examples">Examples</h2>
+<section id="triangle-distribution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="triangle-distribution">Triangle distribution</h3>
+<p>A simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If <span class="math inline">\(\alpha \in \mathbb{R}\)</span> and <span class="math inline">\(\beta \in \mathbb{R}\)</span> are the bounds, with <span class="math inline">\(\alpha &lt; \beta\)</span>, then <span class="math inline">\(y \in (\alpha,\beta)\)</span> has a density defined as <span class="math display">\[
+\textsf{triangle}(y \mid \alpha,\beta)
+=
+\frac{1}{(\beta - \alpha)^2}
+\cdot
+\textrm{min}(y - \alpha, \beta - y).
+\]</span></p>
+<p>The general form of triangle can be coded in Stan as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=alpha&gt; beta;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=alpha, <span class="kw">upper</span>=beta&gt; y;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -<span class="dv">2</span> * log(beta - alpha)</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>    + log(fmin(y - alpha, beta - y));</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because the bounds are specified as data here, the term <code>-2 * log(beta - alpha)</code> could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.</p>
+<p>If <span class="math inline">\(\alpha = -1\)</span>, <span class="math inline">\(\beta = 1\)</span>, and <span class="math inline">\(y \in (-1,1)\)</span>, then <code>fmin(y - alpha, beta - y)</code> is <code>fmin(y + 1, -1 - y)</code>, which is <code>fmin(y + 1, -(y + 1))</code>, which reduces to <code>1 - abs(y)</code>. Therfore, the density, dropping constants, reduces to <span class="math display">\[
+\textsf{triangle}(y \mid -1, 1) \propto 1 - |y|.
+\]</span> Consider the following Stan implementation of <span class="math inline">\(\textsf{triangle}(-1,1)\)</span> for sampling.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log1m(abs(y));</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The single scalar parameter <code>y</code> is declared as lying in the interval <code>(-1, 1)</code>. The total log probability is incremented with the joint log probability of all parameters, i.e., <span class="math inline">\(\log \mathsf{Triangle}(y \mid -1, 1)\)</span>. This value is coded in Stan as <code>log1m(abs(y))</code>. The function <code>log1m</code> is defined so that <code>log1m(x)</code> has the same value as <span class="math inline">\(\log(1 - x)\)</span>, but the computation is faster, more accurate, and more stable.</p>
+<p>The constrained type <code>real&lt;lower=-1, upper=1&gt;</code> declared for <code>y</code> is critical for correct sampling behavior. If the constraint on <code>y</code> is removed from the program, say by declaring <code>y</code> as having the unconstrained scalar type <code>real</code>, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of <code>y</code> outside of <span class="math inline">\((-1,1)\)</span>.</p>
+<p>Now suppose the log probability function were extended to all of <span class="math inline">\(\mathbb{R}\)</span> as follows by defining the probability to be <code>log(0.0)</code>, i.e., <span class="math inline">\(-\infty\)</span>, for values outside of <span class="math inline">\((-1, 1)\)</span>.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log(fmax(<span class="fl">0.0</span>,<span class="dv">1</span> - abs(y)));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With the constraint on <code>y</code> in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on <code>y</code> is removed, the model will compile and run without arithmetic errors, but will not sample properly.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+</section>
+<section id="exponential-distribution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="exponential-distribution">Exponential distribution</h3>
+<p>If Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where <code>lambda</code> is the inverse scale and <code>y</code> the sampled variate.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log(lambda) - y * lambda;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This encoding will work for any <code>lambda</code> and <code>y</code>; they can be parameters, data, or one of each, or even local variables.</p>
+<p>The assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a>y ~ exponential(lambda);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There are two notable differences. First, the distribution statement will check the inputs to make sure both <code>lambda</code> is positive and <code>y</code> is non-negative (which includes checking that neither is the special not-a-number value).</p>
+<p>The second difference is that if <code>lambda</code> is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the <code>log(lambda)</code> term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If <code>lambda</code> and <code>y</code> are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).</p>
+</section>
+<section id="bivariate-normal-cumulative-distribution-function" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="bivariate-normal-cumulative-distribution-function">Bivariate normal cumulative distribution function</h3>
+<p>For another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation <code>rho</code>. That is, it computes <span class="math display">\[
+\texttt{binormal}\mathtt{\_}\texttt{cdf}(z_1, z_2, \rho) = \Pr[Z_1 \leq z_1 \text{ and } Z_2 \leq z_2]
+\]</span> where the random 2-vector <span class="math inline">\(Z\)</span> has the distribution <span class="math display">\[
+Z \sim \textsf{multivariate normal}\left(
+\begin{bmatrix}
+0 \\
+0
+\end{bmatrix}, \
+\begin{bmatrix}
+1 &amp; \rho
+\\
+\rho &amp; 1
+\end{bmatrix}
+\right).
+\]</span></p>
+<p>The following Stan program implements this function,</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> binormal_cdf(<span class="dt">tuple</span>(<span class="dt">real</span>, <span class="dt">real</span>) z, <span class="dt">real</span> rho) {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> z1 = z<span class="fl">.1</span>;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> z2 = z<span class="fl">.2</span>;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (z1 == <span class="dv">0</span> &amp;&amp; z2 == <span class="dv">0</span>) {</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> <span class="fl">0.25</span> + asin(rho) / (<span class="dv">2</span> * pi());</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> denom = sqrt((<span class="dv">1</span> + rho) * (<span class="dv">1</span> - rho));</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> term1 = z1 == <span class="dv">0</span></span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>    ? (z2 &gt; <span class="dv">0</span> ? <span class="fl">0.25</span> : -<span class="fl">0.25</span>)</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>    :  owens_t(z1, (z2 / z1 - rho) / denom);</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> term2 = z2 == <span class="dv">0</span></span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>    ? (z1 &gt; <span class="dv">0</span> ? <span class="fl">0.25</span> : -<span class="fl">0.25</span>)</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>    : owens_t(z2, (z1 / z2 - rho) / denom);</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> z1z2 = z1 * z2;</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = z1z2 &lt; <span class="dv">0</span> || (z1z2 == <span class="dv">0</span> &amp;&amp; (z1 + z2) &lt; <span class="dv">0</span>);</span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> <span class="fl">0.5</span> * (Phi(z1) + Phi(z2) - delta) - term1 - term2;</span>
+<span id="cb6-17"><a href="#cb6-17" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is written using a tuple argument so that it may be called as <code>binormal_cdf((z1, z2) | rho)</code>. The two ternary operators defining <code>term1</code> and <code>term2</code> are derived by taking the limit of the <code>owens_t</code> function when the second argument goes to infinity.</p>
+<p>One way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (zzr <span class="cf">in</span> {[<span class="dv">0</span>, <span class="dv">0</span>, <span class="fl">0.5</span>],</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>               [<span class="dv">0</span>, <span class="dv">1</span>, <span class="fl">0.5</span>],</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>               [<span class="dv">1</span>, <span class="dv">0</span>, -<span class="fl">0.2</span>],</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>               [<span class="dv">1</span>, <span class="dv">3</span>, <span class="fl">0.9</span>]}) {</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> z1  = zzr[<span class="dv">1</span>];</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> z2 = zzr[<span class="dv">2</span>];</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> rho = zzr[<span class="dv">3</span>];</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>    <span class="kw">print</span>(<span class="st">"binomial_cdf(("</span>,  z1, <span class="st">", "</span>, z2, <span class="st">") | "</span>, rho, <span class="st">")"</span>,</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>          <span class="st">"="</span>, binormal_cdf((z1, z2) | rho));</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case, we verified that the results match those of the <code>pbivnorm</code> package in R.</p>
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The problem is the (extremely!) light tails of the triangle distribution. The standard HMC and NUTS samplers can’t get into the corners of the triangle properly. Because the Stan code declares <code>y</code> to be of type <code>real&lt;lower=-1, upper=1&gt;</code>, the inverse logit transform is applied to the unconstrained variable and its log absolute derivative added to the log probability. The resulting distribution on the logit-transformed <code>y</code> is well behaved.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/user-functions.html" class="pagination-link" aria-label="User-Defined Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">User-Defined Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/proportionality-constants.html" class="pagination-link" aria-label="Proportionality Constants">
+        <span class="nav-page-text">Proportionality Constants</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/custom-probability.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/dae.html b/docs/2_39/stan-users-guide/dae.html
new file mode 100644
index 000000000..6c2b59173
--- /dev/null
+++ b/docs/2_39/stan-users-guide/dae.html
@@ -0,0 +1,1358 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Differential-Algebraic Equations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/survival.html" rel="next">
+<link href="../stan-users-guide/complex-numbers.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Differential-Algebraic Equations">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/dae.html">Differential-Algebraic Equations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#dae-solver.chapter" id="toc-dae-solver.chapter" class="nav-link active" data-scroll-target="#dae-solver.chapter">Differential-Algebraic Equations</a>
+  <ul class="collapse">
+  <li><a href="#notation" id="toc-notation" class="nav-link" data-scroll-target="#notation">Notation</a></li>
+  <li><a href="#example-chemical-kinetics" id="toc-example-chemical-kinetics" class="nav-link" data-scroll-target="#example-chemical-kinetics">Example: chemical kinetics</a></li>
+  <li><a href="#index-of-daes" id="toc-index-of-daes" class="nav-link" data-scroll-target="#index-of-daes">Index of DAEs</a></li>
+  <li><a href="#coding-the-dae-system-function" id="toc-coding-the-dae-system-function" class="nav-link" data-scroll-target="#coding-the-dae-system-function">Coding the DAE system function</a>
+  <ul class="collapse">
+  <li><a href="#strict-signature" id="toc-strict-signature" class="nav-link" data-scroll-target="#strict-signature">Strict signature</a></li>
+  </ul></li>
+  <li><a href="#solving-daes" id="toc-solving-daes" class="nav-link" data-scroll-target="#solving-daes">Solving DAEs</a></li>
+  <li><a href="#control-dae.section" id="toc-control-dae.section" class="nav-link" data-scroll-target="#control-dae.section">Control parameters for DAE solving</a>
+  <ul class="collapse">
+  <li><a href="#maximum-number-of-steps" id="toc-maximum-number-of-steps" class="nav-link" data-scroll-target="#maximum-number-of-steps">Maximum number of steps</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/dae.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/dae.html">Differential-Algebraic Equations</a></li></ol></nav></header>
+
+
+
+
+<section id="dae-solver.chapter" class="level1">
+<h1>Differential-Algebraic Equations</h1>
+<p>Stan support solving systems of differential-algebraic equations (DAEs) of index 1 <span class="citation" data-cites="serban_user:2021">(<a href="#ref-serban_user:2021" role="doc-biblioref">Serban et al. 2021</a>)</span>. The solver adaptively refines the solutions in order to satisfy given tolerances.</p>
+<p>One can think a differential-algebraic system of equations as ODEs with additional algebraic constraints applied to some of the variables. In such a system, the variable derivatives may not be expressed explicitly with a right-hand-side as in ODEs, but implicitly constrained.</p>
+<p>Similar to ODE solvers, the DAE solvers must not only provide the solution to the DAE itself, but also the gradient of the DAE solution with respect to parameters (the sensitivities). Stan’s DAE solver uses the <em>forward sensitivity</em> calculation to expand the base DAE system with additional DAE equations for the gradients of the solution. For each parameter, an additional full set of <span class="math inline">\(N\)</span> sensitivity states are added meaning that the full DAE solved has <span class="math inline">\(N \, + N \cdot M\)</span> states.</p>
+<p>Two interfaces are provided for the forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. The same principles in the <a href="../stan-users-guide/odes.html#control-ode.section">control parameters section</a> apply here.</p>
+<p>Internally Stan’s DAE solver uses a variable-step, variable-order, backward-differentiation formula implementation <span class="citation" data-cites="CohenHindmarsh:1996 SerbanHindmarsh:2005">(<a href="#ref-CohenHindmarsh:1996" role="doc-biblioref">Cohen and Hindmarsh 1996</a>; <a href="#ref-SerbanHindmarsh:2005" role="doc-biblioref">Serban and Hindmarsh 2005</a>)</span>.</p>
+<section id="notation" class="level2">
+<h2 class="anchored" data-anchor-id="notation">Notation</h2>
+<p>A DAE is defined by a set of expressions for the <em>residuals</em> of differential equations and algebraic equations <span class="math inline">\(r(y', y, t, \theta)\)</span>, and <em>consistent</em> initial conditions <span class="math inline">\(y(t_0, \theta) = y_0, y'(t_0, \theta)=y'_0\)</span>. The DAE is define by residual function as <span class="math inline">\(r(y', y, t, \theta)=0\)</span>. The <span class="math inline">\(\theta\)</span> dependence is included in the notation to highlight that the solution <span class="math inline">\(y(t)\)</span> is a function of any parameters used in the computation.</p>
+</section>
+<section id="example-chemical-kinetics" class="level2">
+<h2 class="anchored" data-anchor-id="example-chemical-kinetics">Example: chemical kinetics</h2>
+<p>As an example of a system of DAEs, consider following chemical kinetics problem<span class="citation" data-cites="robertson_solution:1966">(<a href="#ref-robertson_solution:1966" role="doc-biblioref">Robertson 1966</a>)</span>. The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants <span class="math inline">\(y_1\)</span> and <span class="math inline">\(y_2\)</span> to the product <span class="math inline">\(y_3\)</span>, and the algebraic equation describes the mass conservation. <span class="citation" data-cites="serban_example:2021">(<a href="#ref-serban_example:2021" role="doc-biblioref">Serban and Hindmarsh 2021</a>)</span>.</p>
+<p><span class="math display">\[\begin{align*}
+\frac{dy_1}{dt} + \alpha y_1 - \beta y_2 y_3 = 0 \\
+\frac{dy_2}{dt} - \alpha y_1 + \beta y_2 y_3 + \gamma y_2^2 = 0 \\
+y_1 + y_2 + y_3 - 1.0 = 0
+\end{align*}\]</span></p>
+<p>The state equations implicitly defines the state <span class="math inline">\((y_1(t), y_2(t), y_3(t))\)</span> at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients <span class="math inline">\((\alpha, \beta, \gamma)\)</span>.</p>
+<p>Unlike solving ODEs, solving DAEs requires a <em>consistent</em> initial condition. That is, one must specify both <span class="math inline">\(y(t_0)\)</span> and <span class="math inline">\(y'(t_0)\)</span> so that residual function becomes zero at initial time <span class="math inline">\(t_0\)</span> <span class="math display">\[\begin{equation*}
+r(y'(t_0), y(t_0), t_0) = 0
+\end{equation*}\]</span></p>
+</section>
+<section id="index-of-daes" class="level2">
+<h2 class="anchored" data-anchor-id="index-of-daes">Index of DAEs</h2>
+<p>The index along a DAE solution <span class="math inline">\(y(t)\)</span> is the minimum number of differentiations of some of the components of the system required to solve for <span class="math inline">\(y'\)</span> uniquely in terms of <span class="math inline">\(y\)</span> and <span class="math inline">\(t\)</span>, so that the DAE is converted into an ODE for <span class="math inline">\(y\)</span>. Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for <span class="math inline">\(y_3\)</span>.</p>
+<p>Most DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically <span class="citation" data-cites="ascher_computer:1998">(<a href="#ref-ascher_computer:1998" role="doc-biblioref">Ascher and Petzold 1998</a>)</span>.</p>
+</section>
+<section id="coding-the-dae-system-function" class="level2">
+<h2 class="anchored" data-anchor-id="coding-the-dae-system-function">Coding the DAE system function</h2>
+<p>The first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.</p>
+<p>The first argument to the residual function is time, passed as a <code>real</code>; the second argument to the residual function is the system state <span class="math inline">\(y\)</span>, passed as a <code>vector</code>, the third argument to the residual function is the state derivative <span class="math inline">\(y'\)</span>, also passed as a <code>vector</code>. The residual function’s return value is a <code>vector</code> of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(\beta\)</span>, and <span class="math inline">\(\gamma\)</span>), data, or any quantities that are needed to define the DAE.</p>
+<p>The above reaction be coded using the following function in Stan (see the <a href="../stan-users-guide/user-functions.html">user-defined functions chapter</a> for more information on coding user-defined functions).</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> chem(<span class="dt">real</span> t, <span class="dt">vector</span> yy, <span class="dt">vector</span> yp,</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>            <span class="dt">real</span> alpha, <span class="dt">real</span> beta, <span class="dt">real</span> gamma) {</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] res;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  res[<span class="dv">1</span>] = yp[<span class="dv">1</span>] + alpha * yy[<span class="dv">1</span>] - beta * yy[<span class="dv">2</span>] * yy[<span class="dv">3</span>];</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  res[<span class="dv">2</span>] = yp[<span class="dv">2</span>] - alpha * yy[<span class="dv">1</span>] + beta * yy[<span class="dv">2</span>] * yy[<span class="dv">3</span>] + gamma * yy[<span class="dv">2</span>] * yy[<span class="dv">2</span>];</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  res[<span class="dv">3</span>] = yy[<span class="dv">1</span>] + yy[<span class="dv">2</span>] + yy[<span class="dv">3</span>] - <span class="fl">1.0</span>;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> res;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function takes in a time <code>t</code> (a <code>real</code>), the system state <code>yy</code> (a <code>vector</code>), state derivative <code>yp</code> (a <code>vector</code>), as well as parameter <code>alpha</code> (a <code>real</code>), <code>beta</code> (a <code>real</code>), and <code>gamma</code> (a <code>real</code>). The function returns a <code>vector</code> of the residuals at time <code>t</code>. The DAE coded here does not explicitly depend on <code>t</code>, however one still needs to specify <code>t</code> as an argument.</p>
+<section id="strict-signature" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="strict-signature">Strict signature</h3>
+<p>The types in the DAE residual function are strict. The first argument is the time passed as a <code>real</code>, the second argument is the state passed as a <code>vector</code>, the third argument is the state derivative passed as a <code>vector</code>, and the return type is a <code>vector</code>. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.</p>
+<p>All of these are possible DAE signatures:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> my_dae1(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">vector</span> yp, <span class="dt">real</span> a0);</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> my_dae2(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">vector</span> yp, <span class="dt">array</span>[] <span class="dt">int</span> a0, <span class="dt">vector</span> a1);</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> my_dae3(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">vector</span> yp, <span class="dt">matrix</span> a0, <span class="dt">array</span>[] <span class="dt">real</span> a1, <span class="dt">row_vector</span> a2);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>but these are not allowed:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> my_dae1(<span class="dt">real</span> t, <span class="dt">array</span>[] <span class="dt">real</span> y, <span class="dt">vector</span> yp);</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="co">// Second argument is not a vector</span></span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[] <span class="dt">real</span> my_dae2(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">vector</span> yp);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a><span class="co">// Return type is not a vector</span></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> my_dae3(<span class="dt">real</span> t, <span class="dt">vector</span> y);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a><span class="co">// First argument is not a real and missing the third argument</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="solving-daes" class="level2">
+<h2 class="anchored" data-anchor-id="solving-daes">Solving DAEs</h2>
+<p>Stan provides a <code>dae</code> function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] yy0;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] yp0;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> t0;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> ts;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[<span class="dv">3</span>] y;</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> gamma;</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);</span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Since <code>gamma</code> is a parameter, the DAE solver is called in the transformed parameters block.</p>
+</section>
+<section id="control-dae.section" class="level2">
+<h2 class="anchored" data-anchor-id="control-dae.section">Control parameters for DAE solving</h2>
+<p>Using <code>dae_tol</code> one can specify the <code>relative_tolerance</code>, <code>absolute_tolerance</code>, and <code>max_num_steps</code> parameters in order to control the DAE solution.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">3</span>] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>                             relative_tolerance,</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>                             absolute_tolerance,</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>                             max_num_steps,</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>                             alpha, beta, gamma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><code>relative_tolerance</code> and <code>absolute_tolerance</code> control accuracy the solver tries to achieve, and <code>max_num_steps</code> specifies the maximum number of steps the solver will take between output time points before throwing an error.</p>
+<p>The control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the <code>data</code> qualifier.</p>
+<p>The default value of relative and absolute tolerances are <span class="math inline">\(10^{-10}\)</span> and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.</p>
+<section id="maximum-number-of-steps" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="maximum-number-of-steps">Maximum number of steps</h3>
+<p>The maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-ascher_computer:1998" class="csl-entry" role="listitem">
+Ascher, Uri M., and Linda R. Petzold. 1998. <em>Computer <span>Methods</span> for <span>Ordinary</span> <span>Differential</span> <span>Equations</span> and <span>Differential</span>-<span>Algebraic</span> <span>Equations</span></em>. Philadelphia: SIAM: Society for Industrial; Applied Mathematics.
+</div>
+<div id="ref-CohenHindmarsh:1996" class="csl-entry" role="listitem">
+Cohen, Scott D, and Alan C Hindmarsh. 1996. <span>“<span>CVODE</span>, a Stiff/Nonstiff <span>ODE</span> Solver in <span>C</span>.”</span> <em>Computers in Physics</em> 10 (2): 138–43.
+</div>
+<div id="ref-robertson_solution:1966" class="csl-entry" role="listitem">
+Robertson, H. H. 1966. <span>“The Solution of a Set of Reaction Rate Equations.”</span> In <em>Numerical Analysis, an Introduction</em>, 178–82. Lodon; New York: Academic Press.
+</div>
+<div id="ref-SerbanHindmarsh:2005" class="csl-entry" role="listitem">
+Serban, Radu, and Alan C Hindmarsh. 2005. <span>“<span>CVODES</span>: The Sensitivity-Enabled <span>ODE</span> Solver in <span>SUNDIALS</span>.”</span> In <em>ASME 2005 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference</em>, 257–69. American Society of Mechanical Engineers.
+</div>
+<div id="ref-serban_example:2021" class="csl-entry" role="listitem">
+Serban, Radu, and Alan C. Hindmarsh. 2021. <span>“Example <span>Programs</span> for <span>IDAS</span>.”</span> LLNL-TR-437091. Lawrence Livermore National Laboratory.
+</div>
+<div id="ref-serban_user:2021" class="csl-entry" role="listitem">
+Serban, Radu, Cosmin Petra, Alan C. Hindmarsh, Cody J. Balos, David J. Gardner, Daniel R. Reynolds, and Carol S. Woodward. 2021. <span>“User <span>Documentation</span> for <span>IDAS</span> V5.0.0.”</span> Lawrence Livermore National Laboratory.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/complex-numbers.html" class="pagination-link" aria-label="Complex Numbers">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Complex Numbers</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/survival.html" class="pagination-link" aria-label="Survival Models">
+        <span class="nav-page-text">Survival Models</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/dae.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/decision-analysis.html b/docs/2_39/stan-users-guide/decision-analysis.html
new file mode 100644
index 000000000..dd3672d7c
--- /dev/null
+++ b/docs/2_39/stan-users-guide/decision-analysis.html
@@ -0,0 +1,1372 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Decision Analysis</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/bootstrap.html" rel="next">
+<link href="../stan-users-guide/poststratification.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Decision Analysis">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/decision-analysis.html">Decision Analysis</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#decision-analysis" id="toc-decision-analysis" class="nav-link active" data-scroll-target="#decision-analysis">Decision Analysis</a>
+  <ul class="collapse">
+  <li><a href="#outline-of-decision-analysis" id="toc-outline-of-decision-analysis" class="nav-link" data-scroll-target="#outline-of-decision-analysis">Outline of decision analysis</a></li>
+  <li><a href="#example-decision-analysis" id="toc-example-decision-analysis" class="nav-link" data-scroll-target="#example-decision-analysis">Example decision analysis</a>
+  <ul class="collapse">
+  <li><a href="#step-1.-define-decisions-and-outcomes" id="toc-step-1.-define-decisions-and-outcomes" class="nav-link" data-scroll-target="#step-1.-define-decisions-and-outcomes">Step 1. Define decisions and outcomes</a></li>
+  <li><a href="#step-2.-define-density-of-outcome-conditioned-on-decision" id="toc-step-2.-define-density-of-outcome-conditioned-on-decision" class="nav-link" data-scroll-target="#step-2.-define-density-of-outcome-conditioned-on-decision">Step 2. Define density of outcome conditioned on decision</a></li>
+  <li><a href="#step-3.-define-the-utility-function" id="toc-step-3.-define-the-utility-function" class="nav-link" data-scroll-target="#step-3.-define-the-utility-function">Step 3. Define the utility function</a></li>
+  <li><a href="#step-4.-maximize-expected-utility" id="toc-step-4.-maximize-expected-utility" class="nav-link" data-scroll-target="#step-4.-maximize-expected-utility">Step 4. Maximize expected utility</a></li>
+  </ul></li>
+  <li><a href="#continuous-choices" id="toc-continuous-choices" class="nav-link" data-scroll-target="#continuous-choices">Continuous choices</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/decision-analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/decision-analysis.html">Decision Analysis</a></li></ol></nav></header>
+
+
+
+
+<section id="decision-analysis" class="level1">
+<h1>Decision Analysis</h1>
+<p>Statistical decision analysis is about making decisions under uncertainty. In order to make decisions, outcomes must have some notion of “utility” associated with them. The so-called “Bayes optimal” decision is the one that maximizes expected utility (or equivalently, minimizes expected loss). This chapter shows how Stan can be used to simultaneously estimate the distribution of outcomes based on decisions and compute the required expected utilities.</p>
+<section id="outline-of-decision-analysis" class="level2">
+<h2 class="anchored" data-anchor-id="outline-of-decision-analysis">Outline of decision analysis</h2>
+<p>Following <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span>, Bayesian decision analysis can be factored into the following four steps.</p>
+<ol type="1">
+<li><p>Define a set <span class="math inline">\(X\)</span> of possible outcomes and a set <span class="math inline">\(D\)</span> of possible decisions.</p></li>
+<li><p>Define a probability distribution of outcomes conditional on decisions through a conditional density function <span class="math inline">\(p(x \mid d)\)</span> for <span class="math inline">\(x \in X\)</span> and <span class="math inline">\(d \in D.\)</span></p></li>
+<li><p>Define a utility function <span class="math inline">\(U : X \rightarrow \mathbb{R}\)</span> mapping outcomes to their utility.</p></li>
+<li><p>Choose action <span class="math inline">\(d^* \in D\)</span> with highest expected utility, <span class="math display">\[
+d^* = \textrm{arg max}_d \ \mathbb{E}[U(x) \mid d].
+\]</span></p></li>
+</ol>
+<p>The outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).</p>
+</section>
+<section id="example-decision-analysis" class="level2">
+<h2 class="anchored" data-anchor-id="example-decision-analysis">Example decision analysis</h2>
+<p>This section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.</p>
+<section id="step-1.-define-decisions-and-outcomes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="step-1.-define-decisions-and-outcomes">Step 1. Define decisions and outcomes</h3>
+<p>A decision consists of the choice of commute mode and the outcome is a time and cost. More formally,</p>
+<ul>
+<li><p>the set of decisions is <span class="math inline">\(D = 1:4\)</span>, corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and</p></li>
+<li><p>the set of outcomes <span class="math inline">\(X = \mathbb{R} \times \mathbb{R}_+\)</span> contains pairs of numbers <span class="math inline">\(x = (c, t)\)</span> consisting of a cost <span class="math inline">\(c\)</span> and time <span class="math inline">\(t \geq 0\)</span>.</p></li>
+</ul>
+</section>
+<section id="step-2.-define-density-of-outcome-conditioned-on-decision" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="step-2.-define-density-of-outcome-conditioned-on-decision">Step 2. Define density of outcome conditioned on decision</h3>
+<p>The density required is <span class="math inline">\(p(x \mid d),\)</span> where <span class="math inline">\(d \in D\)</span> is a decision and <span class="math inline">\(x = (c, t) \in X\)</span> is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters <span class="math inline">\(\theta,\)</span> <span class="math display">\[
+p(x \mid d, x^{\textrm{obs}}, d^{\textrm{obs}})
+=
+\int
+  p(x \mid d, \theta)
+  \cdot p(\theta \mid x^{\textrm{obs}}, d^{\textrm{obs}})
+  \, \textrm{d}\theta.
+\]</span> The observed data for a year of commutes consists of choice of the chosen commute mode <span class="math inline">\(d^{\textrm{obs}}_n\)</span> and observed costs and times <span class="math inline">\(x^{\textrm{obs}}_n = (c^{\textrm{obs}}_n, t^{\textrm{obs}}_n)\)</span> for <span class="math inline">\(n
+\in 1:200.\)</span></p>
+<p>For simplicity, commute time <span class="math inline">\(t_n\)</span> for trip <span class="math inline">\(n\)</span> will be modeled as lognormal for a given choice of transportation <span class="math inline">\(d_n \in 1:4,\)</span> <span class="math display">\[
+t_n \sim \textrm{lognormal}(\mu_{d[n]}, \sigma_{d[n]}).
+\]</span> To understand the notation, <span class="math inline">\(d_n\)</span>, also written <span class="math inline">\(d[n]\)</span>, is the mode of transportation used for trip <span class="math inline">\(n\)</span>. For example if trip <span class="math inline">\(n\)</span> was by bicycle, then <span class="math inline">\(t_n \sim \textrm{lognormal}(\mu_2, \sigma_2),\)</span> where <span class="math inline">\(\mu_2\)</span> and <span class="math inline">\(\sigma_2\)</span> are the lognormal parameters for bicycling.</p>
+<p>Simple fixed priors are used for each mode of transportation <span class="math inline">\(k \in 1:4,\)</span> <span class="math display">\[\begin{eqnarray*}
+\mu_k &amp; \sim &amp; \textrm{normal}(0, 5)
+\\[2pt]
+\sigma_k &amp; \sim &amp; \textrm{lognormal}(0, 1).
+\end{eqnarray*}\]</span> These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.</p>
+<p>Cost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, <span class="math display">\[
+c_n \sim \textrm{lognormal}(\nu_{d[n]}, \tau_{d[n]}).
+\]</span> Again, the priors are fixed for the modes of transportation, <span class="math display">\[\begin{eqnarray*}
+\nu_k &amp; \sim &amp; \textrm{normal}(0, 5)
+\\[2pt]
+\tau_k &amp; \sim &amp; \textrm{lognormal}(0, 1).
+\end{eqnarray*}\]</span> A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.</p>
+<p>The full set of parameters that are marginalized in the posterior predictive distribution is <span class="math display">\[
+\theta = (\mu_{1:4}, \sigma_{1:4}, \nu_{1:4}, \tau_{1:4}).
+\]</span></p>
+</section>
+<section id="step-3.-define-the-utility-function" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="step-3.-define-the-utility-function">Step 3. Define the utility function</h3>
+<p>For the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as</p>
+<p><span class="math display">\[
+U(c, t) = -(c + 25 \cdot t)
+\]</span></p>
+<p>The sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.</p>
+</section>
+<section id="step-4.-maximize-expected-utility" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="step-4.-maximize-expected-utility">Step 4. Maximize expected utility</h3>
+<p>At this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> U(<span class="dt">real</span> c, <span class="dt">real</span> t) {</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> -(c + <span class="dv">25</span> * t);</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">4</span>&gt; d;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> c;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; t;</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">4</span>] mu;</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[<span class="dv">4</span>] sigma;</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">real</span> nu;</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; tau;</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="fl">0.25</span>);</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>  nu ~ normal(<span class="dv">0</span>, <span class="dv">20</span>);</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>  tau ~ lognormal(<span class="dv">0</span>, <span class="fl">0.25</span>);</span>
+<span id="cb1-23"><a href="#cb1-23" aria-hidden="true" tabindex="-1"></a>  t ~ lognormal(mu[d], sigma[d]);</span>
+<span id="cb1-24"><a href="#cb1-24" aria-hidden="true" tabindex="-1"></a>  c ~ lognormal(nu[d], tau[d]);</span>
+<span id="cb1-25"><a href="#cb1-25" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-26"><a href="#cb1-26" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-27"><a href="#cb1-27" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">real</span> util;</span>
+<span id="cb1-28"><a href="#cb1-28" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">4</span>) {</span>
+<span id="cb1-29"><a href="#cb1-29" aria-hidden="true" tabindex="-1"></a>    util[k] = U(lognormal_rng(nu[k], tau[k]),</span>
+<span id="cb1-30"><a href="#cb1-30" aria-hidden="true" tabindex="-1"></a>                lognormal_rng(mu[k], sigma[k]));</span>
+<span id="cb1-31"><a href="#cb1-31" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-32"><a href="#cb1-32" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The generated quantities block defines an array variable <code>util</code> where <code>util[k]</code>, which will hold the utility derived from a random commute for choice <code>k</code> generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.</p>
+<p>For simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, <code>nu</code> and <code>tau</code> would be declared as univariate, and the RNG for cost would only be employed when <code>k == 4</code>.</p>
+<p>Defining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.</p>
+<p>All that is left is to run Stan. The posterior mean for <code>util[k]</code> is the expected utility, which written out with full conditioning, is <span class="math display">\[\begin{eqnarray*}
+\mathbb{E}\!\left[U(x) \mid d = k, d^{\textrm{obs}}, x^{\textrm{obs}}\right]
+&amp; = &amp;
+\int
+  U(x)
+  \cdot p(x \mid d = k, \theta)
+  \cdot p(\theta \mid d^{\textrm{obs}}, x^{\textrm{obs}})
+  \, \textrm{d}\theta
+\\[4pt]
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m = 1}^M  U(x^{(m)} ),
+\end{eqnarray*}\]</span> where <span class="math display">\[
+x^{(m)} \sim p(x \mid d = k, \theta^{(m)} )
+\]</span> and <span class="math display">\[
+\theta^{(m)}
+\sim p(\theta \mid d^{\textrm{obs}}, x^{\textrm{obs}}).
+\]</span></p>
+<p>In terms of Stan’s execution, the random generation of <span class="math inline">\(x^{(m)}\)</span> is carried out with the <code>lognormal_rng</code> operations after <span class="math inline">\(\theta^{(m)}\)</span> is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.</p>
+<p>It only remains to make the decision <code>k</code> with highest expected utility, which will correspond to the choice with the highest posterior mean for <code>util[k]</code>. This can be read off of the <code>mean</code> column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.</p>
+</section>
+</section>
+<section id="continuous-choices" class="level2">
+<h2 class="anchored" data-anchor-id="continuous-choices">Continuous choices</h2>
+<p>Many choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/poststratification.html" class="pagination-link" aria-label="Poststratification">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Poststratification</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/bootstrap.html" class="pagination-link" aria-label="The Bootstrap and Bagging">
+        <span class="nav-page-text">The Bootstrap and Bagging</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/decision-analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/efficiency-tuning.html b/docs/2_39/stan-users-guide/efficiency-tuning.html
new file mode 100644
index 000000000..f89182f93
--- /dev/null
+++ b/docs/2_39/stan-users-guide/efficiency-tuning.html
@@ -0,0 +1,2128 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Efficiency Tuning</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/parallelization.html" rel="next">
+<link href="../stan-users-guide/reparameterization.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Efficiency Tuning">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/funnel.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/efficiency-tuning.html">Efficiency Tuning</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#efficiency-tuning.chapter" id="toc-efficiency-tuning.chapter" class="nav-link active" data-scroll-target="#efficiency-tuning.chapter">Efficiency Tuning</a>
+  <ul class="collapse">
+  <li><a href="#what-is-efficiency" id="toc-what-is-efficiency" class="nav-link" data-scroll-target="#what-is-efficiency">What is efficiency?</a></li>
+  <li><a href="#efficiency-for-probabilistic-models-and-algorithms" id="toc-efficiency-for-probabilistic-models-and-algorithms" class="nav-link" data-scroll-target="#efficiency-for-probabilistic-models-and-algorithms">Efficiency for probabilistic models and algorithms</a></li>
+  <li><a href="#statistical-vs.-computational-efficiency" id="toc-statistical-vs.-computational-efficiency" class="nav-link" data-scroll-target="#statistical-vs.-computational-efficiency">Statistical vs.&nbsp; computational efficiency</a></li>
+  <li><a href="#model-conditioning-and-curvature" id="toc-model-conditioning-and-curvature" class="nav-link" data-scroll-target="#model-conditioning-and-curvature">Model conditioning and curvature</a>
+  <ul class="collapse">
+  <li><a href="#condition-number-and-adaptation" id="toc-condition-number-and-adaptation" class="nav-link" data-scroll-target="#condition-number-and-adaptation">Condition number and adaptation</a></li>
+  <li><a href="#unit-scales-without-correlation" id="toc-unit-scales-without-correlation" class="nav-link" data-scroll-target="#unit-scales-without-correlation">Unit scales without correlation</a></li>
+  <li><a href="#varying-curvature" id="toc-varying-curvature" class="nav-link" data-scroll-target="#varying-curvature">Varying curvature</a></li>
+  <li><a href="#reparameterizing-with-a-change-of-variables" id="toc-reparameterizing-with-a-change-of-variables" class="nav-link" data-scroll-target="#reparameterizing-with-a-change-of-variables">Reparameterizing with a change of variables</a></li>
+  </ul></li>
+  <li><a href="#well-specified-models" id="toc-well-specified-models" class="nav-link" data-scroll-target="#well-specified-models">Well-specified models</a></li>
+  <li><a href="#avoiding-validation" id="toc-avoiding-validation" class="nav-link" data-scroll-target="#avoiding-validation">Avoiding validation</a></li>
+  <li><a href="#reparameterization.section" id="toc-reparameterization.section" class="nav-link" data-scroll-target="#reparameterization.section">Reparameterization</a>
+  <ul class="collapse">
+  <li><a href="#funnel.figure" id="toc-funnel.figure" class="nav-link" data-scroll-target="#funnel.figure">Example: Neal’s funnel</a></li>
+  <li><a href="#reparameterizing-the-cauchy" id="toc-reparameterizing-the-cauchy" class="nav-link" data-scroll-target="#reparameterizing-the-cauchy">Reparameterizing the Cauchy</a></li>
+  <li><a href="#reparameterizing-a-student-t-distribution" id="toc-reparameterizing-a-student-t-distribution" class="nav-link" data-scroll-target="#reparameterizing-a-student-t-distribution">Reparameterizing a Student-t distribution</a></li>
+  <li><a href="#hierarchical-models-and-the-non-centered-parameterization" id="toc-hierarchical-models-and-the-non-centered-parameterization" class="nav-link" data-scroll-target="#hierarchical-models-and-the-non-centered-parameterization">Hierarchical models and the non-centered parameterization</a></li>
+  <li><a href="#non-centered-parameterization" id="toc-non-centered-parameterization" class="nav-link" data-scroll-target="#non-centered-parameterization">Non-centered parameterization</a></li>
+  <li><a href="#multivariate-reparameterizations" id="toc-multivariate-reparameterizations" class="nav-link" data-scroll-target="#multivariate-reparameterizations">Multivariate reparameterizations</a></li>
+  </ul></li>
+  <li><a href="#vectorization" id="toc-vectorization" class="nav-link" data-scroll-target="#vectorization">Vectorization</a>
+  <ul class="collapse">
+  <li><a href="#gradient-bottleneck" id="toc-gradient-bottleneck" class="nav-link" data-scroll-target="#gradient-bottleneck">Gradient bottleneck</a></li>
+  <li><a href="#vectorizing-summations" id="toc-vectorizing-summations" class="nav-link" data-scroll-target="#vectorizing-summations">Vectorizing summations</a></li>
+  <li><a href="#vectorization-through-matrix-operations" id="toc-vectorization-through-matrix-operations" class="nav-link" data-scroll-target="#vectorization-through-matrix-operations">Vectorization through matrix operations</a></li>
+  <li><a href="#vectorized-probability-functions" id="toc-vectorized-probability-functions" class="nav-link" data-scroll-target="#vectorized-probability-functions">Vectorized probability functions</a></li>
+  <li><a href="#reshaping-data-for-vectorization" id="toc-reshaping-data-for-vectorization" class="nav-link" data-scroll-target="#reshaping-data-for-vectorization">Reshaping data for vectorization</a></li>
+  </ul></li>
+  <li><a href="#exploiting-sufficient-statistics" id="toc-exploiting-sufficient-statistics" class="nav-link" data-scroll-target="#exploiting-sufficient-statistics">Exploiting sufficient statistics</a>
+  <ul class="collapse">
+  <li><a href="#bernoulli-sufficient-statistics" id="toc-bernoulli-sufficient-statistics" class="nav-link" data-scroll-target="#bernoulli-sufficient-statistics">Bernoulli sufficient statistics</a></li>
+  <li><a href="#normal-sufficient-statistics" id="toc-normal-sufficient-statistics" class="nav-link" data-scroll-target="#normal-sufficient-statistics">Normal sufficient statistics</a></li>
+  <li><a href="#poisson-sufficient-statistics" id="toc-poisson-sufficient-statistics" class="nav-link" data-scroll-target="#poisson-sufficient-statistics">Poisson sufficient statistics</a></li>
+  </ul></li>
+  <li><a href="#aggregating-common-subexpressions" id="toc-aggregating-common-subexpressions" class="nav-link" data-scroll-target="#aggregating-common-subexpressions">Aggregating common subexpressions</a></li>
+  <li><a href="#exploiting-conjugacy" id="toc-exploiting-conjugacy" class="nav-link" data-scroll-target="#exploiting-conjugacy">Exploiting conjugacy</a></li>
+  <li><a href="#standardizing-predictors" id="toc-standardizing-predictors" class="nav-link" data-scroll-target="#standardizing-predictors">Standardizing predictors</a>
+  <ul class="collapse">
+  <li><a href="#standard-normal-distribution" id="toc-standard-normal-distribution" class="nav-link" data-scroll-target="#standard-normal-distribution">Standard normal distribution</a></li>
+  </ul></li>
+  <li><a href="#using-map-reduce" id="toc-using-map-reduce" class="nav-link" data-scroll-target="#using-map-reduce">Using map-reduce</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/efficiency-tuning.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/efficiency-tuning.html">Efficiency Tuning</a></li></ol></nav></header>
+
+
+
+
+<section id="efficiency-tuning.chapter" class="level1">
+<h1>Efficiency Tuning</h1>
+<p>This chapter provides a grab bag of techniques for optimizing Stan code, including vectorization, sufficient statistics, and conjugacy. At a coarse level, efficiency involves both the amount of time required for a computation and the amount of memory required. For practical applied statistical modeling, we are mainly concerned with reducing wall time (how long a program takes as measured by a clock on the wall) and keeping memory requirements within available bounds.</p>
+<section id="what-is-efficiency" class="level2">
+<h2 class="anchored" data-anchor-id="what-is-efficiency">What is efficiency?</h2>
+<p>The standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).</p>
+</section>
+<section id="efficiency-for-probabilistic-models-and-algorithms" class="level2">
+<h2 class="anchored" data-anchor-id="efficiency-for-probabilistic-models-and-algorithms">Efficiency for probabilistic models and algorithms</h2>
+<p>Stan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.</p>
+<p>With probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.</p>
+<p>A well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.</p>
+</section>
+<section id="statistical-vs.-computational-efficiency" class="level2">
+<h2 class="anchored" data-anchor-id="statistical-vs.-computational-efficiency">Statistical vs.&nbsp; computational efficiency</h2>
+<p>There is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.</p>
+<p>Statistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.</p>
+</section>
+<section id="model-conditioning-and-curvature" class="level2">
+<h2 class="anchored" data-anchor-id="model-conditioning-and-curvature">Model conditioning and curvature</h2>
+<p>Because Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.</p>
+<p>A second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density <span class="math inline">\(\log
+p(\theta)\)</span> with respect to the parameter vector <span class="math inline">\(\theta\)</span>, defined as <span class="math display">\[
+H(\theta) = \nabla \, \nabla \, \log p(\theta \mid y),
+\]</span> so that <span class="math display">\[
+H_{i, j}(\theta) = \frac{\partial^2 \log p(\theta \mid y)}
+                {\partial \theta_i \ \partial \theta_j}.
+\]</span> For pure penalized maximum likelihood problems, the posterior log density <span class="math inline">\(\log p(\theta \mid y)\)</span> is replaced by the penalized likelihood function <span class="math inline">\(\mathcal{L}(\theta) = \log p(y \mid \theta) - \lambda(\theta)\)</span>.</p>
+<section id="condition-number-and-adaptation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="condition-number-and-adaptation">Condition number and adaptation</h3>
+<p>A good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix <span class="math inline">\(H\)</span>, which is the ratio of the largest to the smallest eigenvalue of <span class="math inline">\(H\)</span> (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.</p>
+</section>
+<section id="unit-scales-without-correlation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unit-scales-without-correlation">Unit scales without correlation</h3>
+<p>Ideally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.</p>
+</section>
+<section id="varying-curvature" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="varying-curvature">Varying curvature</h3>
+<p>In all but very simple models (such as multivariate normals), the Hessian will vary as <span class="math inline">\(\theta\)</span> varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.</p>
+<p>For all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.</p>
+<p>With sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.</p>
+</section>
+<section id="reparameterizing-with-a-change-of-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reparameterizing-with-a-change-of-variables">Reparameterizing with a change of variables</h3>
+<p>Improving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the <a href="../stan-users-guide/reparameterization.html">change of variables chapter</a>.</p>
+<p>One has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.</p>
+</section>
+</section>
+<section id="well-specified-models" class="level2">
+<h2 class="anchored" data-anchor-id="well-specified-models">Well-specified models</h2>
+<p>Model misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.</p>
+<p>Counterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as <code>normal(0, 1000)</code>). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as <code>normal(mu,   sigma)</code>, where <code>mu</code> and <code>sigma</code> are new parameters with their own hyperpriors, can be beneficial.</p>
+</section>
+<section id="avoiding-validation" class="level2">
+<h2 class="anchored" data-anchor-id="avoiding-validation">Avoiding validation</h2>
+<p>Stan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}                               <span class="co">// first validation</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, Sigma);  <span class="co">// second validation</span></span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because <code>Sigma</code> is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that <code>Sigma</code> is positive definite. This test is expensive, having cubic run time (i.e., <span class="math inline">\(\mathcal{O}(N^3)\)</span> for <span class="math inline">\(N \times N\)</span> matrices), so it should not be done twice.</p>
+<p>The test may be avoided by simply declaring <code>Sigma</code> to be a simple unconstrained matrix.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] Sigma;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, Sigma);  <span class="co">// only validation</span></span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now the only validation is carried out by the multivariate normal.</p>
+</section>
+<section id="reparameterization.section" class="level2">
+<h2 class="anchored" data-anchor-id="reparameterization.section">Reparameterization</h2>
+<p>Stan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.</p>
+<section id="funnel.figure" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="funnel.figure">Example: Neal’s funnel</h3>
+<p>In this section, we discuss a general transform from a centered to a non-centered parameterization <span class="citation" data-cites="papa-et-al:2007">(<a href="#ref-papa-et-al:2007" role="doc-biblioref">Papaspiliopoulos, Roberts, and Sköld 2007</a>)</span>.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>This reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.</p>
+<p><span class="citation" data-cites="Neal:2003">Neal (<a href="#ref-Neal:2003" role="doc-biblioref">2003</a>)</span> defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for <span class="math inline">\(y \in
+\mathbb{R}\)</span> and <span class="math inline">\(x \in \mathbb{R}^9\)</span> with density</p>
+<p><span class="math display">\[
+p(y,x) = \textsf{normal}(y \mid 0,3) \times \prod_{n=1}^9
+\textsf{normal}(x_n \mid 0,\exp(y/2)).
+\]</span></p>
+<p>The probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to <span class="math inline">\(y\)</span>. A plot of the log marginal density of <span class="math inline">\(y\)</span> and the first dimension <span class="math inline">\(x_1\)</span> is shown in the following plot.</p>
+<p>The funnel can be implemented directly in Stan as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">9</span>] x;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(<span class="dv">0</span>, <span class="dv">3</span>);</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  x ~ normal(<span class="dv">0</span>, exp(y/<span class="dv">2</span>));</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where <span class="math inline">\(y\)</span> is small and thus <span class="math inline">\(x\)</span> is constrained to be near 0. This is due to the fact that the density’s scale changes with <span class="math inline">\(y\)</span>, so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.</p>
+<div id="fig-funnel" class="quarto-layout-panel">
+<figure class="quarto-float quarto-float-fig figure">
+<div aria-describedby="fig-funnel-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<div class="quarto-layout-row">
+<div class="quarto-layout-cell" style="flex-basis: 55.0%;justify-content: center;">
+<p><img src="img/funnel.png" class="img-fluid figure-img"></p>
+</div>
+<div class="quarto-layout-cell" style="flex-basis: 45.0%;justify-content: center;">
+<p><img src="img/funnel-fit.png" class="img-fluid figure-img"></p>
+</div>
+</div>
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-funnel-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+Figure&nbsp;1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable <span class="math inline">\(y\)</span> and one lower-level variable <span class="math inline">\(x_1\)</span> (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of <span class="math inline">\(x_1\)</span> and <span class="math inline">\(y\)</span> values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density <span class="math inline">\(p(y) = \textsf{normal}(y \mid 0,3)\)</span>, which has mean 0 and standard deviation 3.
+</figcaption>
+</figure>
+</div>
+<p>In this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y_raw;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">9</span>] x_raw;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">9</span>] x;</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  y = <span class="fl">3.0</span> * y_raw;</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>  x = exp(y/<span class="dv">2</span>) * x_raw;</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  y_raw ~ std_normal(); <span class="co">// implies y ~ normal(0, 3)</span></span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>  x_raw ~ std_normal(); <span class="co">// implies x ~ normal(0, exp(y/2))</span></span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this second model, the parameters <code>x_raw</code> and <code>y_raw</code> are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.</p>
+</section>
+<section id="reparameterizing-the-cauchy" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reparameterizing-the-cauchy">Reparameterizing the Cauchy</h3>
+<p>Sampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.</p>
+<p>The practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.</p>
+<p>Suppose a random variable of interest <span class="math inline">\(X\)</span> has a Cauchy distribution with location <span class="math inline">\(\mu\)</span> and scale <span class="math inline">\(\tau\)</span>, so that <span class="math inline">\(X \sim
+\textsf{Cauchy}(\mu,\tau)\)</span>. The variable <span class="math inline">\(X\)</span> has a cumulative distribution function <span class="math inline">\(F_X:\mathbb{R} \rightarrow (0,1)\)</span> defined by <span class="math display">\[
+F_X(x) = \frac{1}{\pi} \arctan \left( \frac{x - \mu}{\tau} \right) +
+\frac{1}{2}.
+\]</span> The inverse of the cumulative distribution function, <span class="math inline">\(F_X^{-1}:(0,1) \rightarrow \mathbb{R}\)</span>, is thus</p>
+<p><span class="math display">\[
+F^{-1}_X(y) = \mu + \tau \tan \left( \pi \left( y - \frac{1}{2} \right) \right).
+\]</span> Thus if the random variable <span class="math inline">\(Y\)</span> has a unit uniform distribution, <span class="math inline">\(Y
+\sim \textsf{uniform}(0,1)\)</span>, then <span class="math inline">\(F^{-1}_X(Y)\)</span> has a Cauchy distribution with location <span class="math inline">\(\mu\)</span> and scale <span class="math inline">\(\tau\)</span>, i.e., <span class="math inline">\(F^{-1}_X(Y) \sim
+\textsf{Cauchy}(\mu,\tau)\)</span>.</p>
+<p>Consider a Stan program involving a Cauchy-distributed parameter <code>beta</code>.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  beta ~ cauchy(mu, tau);</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This declaration of <code>beta</code> as a parameter may be replaced with a transformed parameter <code>beta</code> defined in terms of a uniform-distributed parameter <code>beta_unif</code>.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=-pi() / <span class="dv">2</span>, <span class="kw">upper</span>=pi() / <span class="dv">2</span>&gt; beta_unif;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  beta = mu + tau * tan(beta_unif);  <span class="co">// beta ~ cauchy(mu, tau)</span></span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>  beta_unif ~ uniform(-pi() / <span class="dv">2</span>, pi() / <span class="dv">2</span>);  <span class="co">// not necessary</span></span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is more convenient in Stan to transform a uniform variable on <span class="math inline">\((-\pi/2, \pi/2)\)</span> than one on <span class="math inline">\((0,1)\)</span>. The Cauchy location and scale parameters, <code>mu</code> and <code>tau</code>, may be defined as data or may themselves be parameters. The variable <code>beta</code> could also be defined as a local variable if it does not need to be included in the sampler’s output.</p>
+<p>The uniform distribution on <code>beta_unif</code> is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because <span class="math inline">\(\log
+\textsf{uniform}(\beta_{\textsf{unif}} \mid -\pi/2,\pi/2) =
+-\log \pi\)</span> is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that <code>beta_unif</code> is between <code>-pi() / 2</code> and <code>pi() / 2</code>, but this condition is guaranteed by the constraints in the declaration of <code>beta_unif</code>.</p>
+</section>
+<section id="reparameterizing-a-student-t-distribution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reparameterizing-a-student-t-distribution">Reparameterizing a Student-t distribution</h3>
+<p>One thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable <span class="math inline">\(\beta\)</span>, <span class="math display">\[
+\beta \sim \textsf{Student-t}(\nu, 0, 1),
+\]</span> by first generating a gamma-distributed precision (inverse variance) <span class="math inline">\(\tau\)</span> according to <span class="math display">\[
+\tau \sim \textsf{Gamma}(\nu/2, \nu/2),
+\]</span> and then generating <span class="math inline">\(\beta\)</span> from the normal distribution, <span class="math display">\[
+\beta \sim \textsf{normal}\left(0,\tau^{-\frac{1}{2}}\right).
+\]</span></p>
+<p>Because <span class="math inline">\(\tau\)</span> is precision, <span class="math inline">\(\tau^{-\frac{1}{2}}\)</span> is the scale (standard deviation), which is the parameterization used by Stan.</p>
+<p>The marginal distribution of <span class="math inline">\(\beta\)</span> when you integrate out <span class="math inline">\(\tau\)</span> is <span class="math inline">\(\textsf{Student-t}(\nu, 0, 1)\)</span>, i.e., <span class="math display">\[
+\textsf{Student-t}(\beta \mid \nu, 0, 1)
+=
+\int_0^{\infty}
+\,
+\textsf{normal}\left(\beta \middle| 0, \tau^{-0.5}\right)
+\times
+\textsf{Gamma}\left(\tau \middle| \nu/2, \nu/2\right)
+\
+\text{d} \tau.
+\]</span></p>
+<p>To go one step further, instead of defining a <span class="math inline">\(\beta\)</span> drawn from a normal with precision <span class="math inline">\(\tau\)</span>, define <span class="math inline">\(\alpha\)</span> to be drawn from a unit normal, <span class="math display">\[
+\alpha \sim \textsf{normal}(0,1)
+\]</span> and rescale by defining <span class="math display">\[
+\beta = \alpha \, \tau^{-\frac{1}{2}}.
+\]</span></p>
+<p>Now suppose <span class="math inline">\(\mu = \beta x\)</span> is the product of <span class="math inline">\(\beta\)</span> with a regression predictor <span class="math inline">\(x\)</span>. Then the reparameterization <span class="math inline">\(\mu = \alpha
+\tau^{-\frac{1}{2}} x\)</span> has the same distribution, but in the original, direct parameterization, <span class="math inline">\(\beta\)</span> has (potentially) heavy tails, whereas in the second, neither <span class="math inline">\(\tau\)</span> nor <span class="math inline">\(\alpha\)</span> have heavy tails.</p>
+<p>To translate into Stan notation, this reparameterization replaces</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; nu;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  beta ~ student_t(nu, <span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; nu;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; tau;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>  beta = alpha / sqrt(tau);</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> half_nu;</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  half_nu = <span class="fl">0.5</span> * nu;</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>  tau ~ gamma(half_nu, half_nu);</span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although set to <code>0</code> here, in most cases, the lower bound for the degrees of freedom parameter <code>nu</code> can be set to <code>1</code> or higher; when <code>nu</code> is 1, the result is a Cauchy distribution with fat tails and as <code>nu</code> approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter <code>nu</code> characterizes the heaviness of the tails of the model.</p>
+</section>
+<section id="hierarchical-models-and-the-non-centered-parameterization" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="hierarchical-models-and-the-non-centered-parameterization">Hierarchical models and the non-centered parameterization</h3>
+<p>Unfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.</p>
+<section id="centered-parameterization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="centered-parameterization">Centered parameterization</h4>
+<p>For example, a vectorized hierarchical model might draw a vector of coefficients <span class="math inline">\(\beta\)</span> with definitions as follows. The so-called centered parameterization is as follows.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(mu_beta, sigma_beta);</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although not shown, a full model will have priors on both <code>mu_beta</code> and <code>sigma_beta</code> along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix <code>x</code> and binary outcome vector <code>y</code> would include a likelihood statement such as form <code>y ~ bernoulli_logit(x * beta)</code>, leading to an analytically intractable posterior.</p>
+<p>A hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of <code>beta</code>, <code>mu_beta</code> and <code>sigma_beta</code> are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See <span class="citation" data-cites="Betancourt-Girolami:2013">Betancourt and Girolami (<a href="#ref-Betancourt-Girolami:2013" role="doc-biblioref">2013</a>)</span> for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.</p>
+</section>
+</section>
+<section id="non-centered-parameterization" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="non-centered-parameterization">Non-centered parameterization</h3>
+<p>Sometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see <span class="citation" data-cites="Betancourt-Girolami:2013">Betancourt and Girolami (<a href="#ref-Betancourt-Girolami:2013" role="doc-biblioref">2013</a>)</span>), and in more extreme cases will be necessary to achieve convergence.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta_raw;</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// implies: beta ~ normal(mu_beta, sigma_beta)</span></span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a>  beta = mu_beta + sigma_beta * beta_raw;</span>
+<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-12"><a href="#cb10-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-13"><a href="#cb10-13" aria-hidden="true" tabindex="-1"></a>  beta_raw ~ std_normal();</span>
+<span id="cb10-14"><a href="#cb10-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb10-15"><a href="#cb10-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Any priors defined for <code>mu_beta</code> and <code>sigma_beta</code> remain as defined in the original model.</p>
+<p>Alternatively, Stan’s <a href="https://mc-stan.org/docs/reference-manual/types.html#affinely-transformed-real">affine transform</a> can be used to decouple <code>sigma</code> and <code>beta</code>:</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">offset</span>=mu_beta, <span class="dt">multiplier</span>=sigma_beta&gt;[K] beta;</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(mu_beta, sigma_beta);</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Reparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let <span class="math inline">\(\beta = l + s\alpha\)</span> where <span class="math inline">\(l\)</span> is a location parameter and <span class="math inline">\(s\)</span> is a scale parameter. The parameter <span class="math inline">\(l\)</span> need not be the mean, <span class="math inline">\(s\)</span> need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> are from the same distributional family but <span class="math inline">\(\alpha\)</span> has location zero and unit scale, while <span class="math inline">\(\beta\)</span> has location <span class="math inline">\(l\)</span> and scale <span class="math inline">\(s\)</span>, then that distribution is a location-scale distribution. Thus, if <span class="math inline">\(\alpha\)</span> were a parameter and <span class="math inline">\(\beta\)</span> were a transformed parameter, then a prior distribution from the location-scale family on <span class="math inline">\(\alpha\)</span> with location zero and unit scale implies a prior distribution on <span class="math inline">\(\beta\)</span> with location <span class="math inline">\(l\)</span> and scale <span class="math inline">\(s\)</span>. Doing so would reduce the dependence between <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(l\)</span>, and <span class="math inline">\(s\)</span>.</p>
+<p>There are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if <span class="math inline">\(\alpha\)</span> is distributed standard normal, then <span class="math inline">\(\beta\)</span> is distributed normal with mean <span class="math inline">\(\mu = l\)</span> and standard deviation <span class="math inline">\(\sigma = s\)</span>. The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.</p>
+<p>Also, if <span class="math inline">\(z\)</span> is distributed standard normal, then <span class="math inline">\(z^2\)</span> is distributed chi-squared with one degree of freedom. By summing the squares of <span class="math inline">\(K\)</span> independent standard normal variates, one can obtain a single variate that is distributed chi-squared with <span class="math inline">\(K\)</span> degrees of freedom. However, for large <span class="math inline">\(K\)</span>, the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying <span class="math inline">\(K\)</span> primitive parameters just to obtain one transformed parameter to use in a model.</p>
+</section>
+<section id="multivariate-reparameterizations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multivariate-reparameterizations">Multivariate reparameterizations</h3>
+<p>The benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for <span class="math inline">\(\beta\)</span> to be multivariate normal with mean vector <span class="math inline">\(\mu\)</span> and covariance matrix <span class="math inline">\(\Sigma\)</span>. Such a belief is reflected by the following code.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb12-12"><a href="#cb12-12" aria-hidden="true" tabindex="-1"></a>  beta ~ multi_normal(mu, Sigma);</span>
+<span id="cb12-13"><a href="#cb12-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-14"><a href="#cb12-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case <code>mu</code> and <code>Sigma</code> are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of <code>beta</code>.</p>
+<p>If <span class="math inline">\(\alpha\)</span> has the same dimensions as <span class="math inline">\(\beta\)</span> but the elements of <span class="math inline">\(\alpha\)</span> are independently and identically distributed standard normal such that <span class="math inline">\(\beta = \mu + L\alpha\)</span>, where <span class="math inline">\(LL^\top = \Sigma\)</span>, then <span class="math inline">\(\beta\)</span> is distributed multivariate normal with mean vector <span class="math inline">\(\mu\)</span> and covariance matrix <span class="math inline">\(\Sigma\)</span>. One choice for <span class="math inline">\(L\)</span> is the Cholesky factor of <span class="math inline">\(\Sigma\)</span>. Thus, the model above could be reparameterized as follows.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L;</span>
+<span id="cb13-9"><a href="#cb13-9" aria-hidden="true" tabindex="-1"></a>  L = cholesky_decompose(Sigma);</span>
+<span id="cb13-10"><a href="#cb13-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-11"><a href="#cb13-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb13-12"><a href="#cb13-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] alpha;</span>
+<span id="cb13-13"><a href="#cb13-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-14"><a href="#cb13-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-15"><a href="#cb13-15" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb13-16"><a href="#cb13-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb13-17"><a href="#cb13-17" aria-hidden="true" tabindex="-1"></a>  beta = mu + L * alpha;</span>
+<span id="cb13-18"><a href="#cb13-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-19"><a href="#cb13-19" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb13-20"><a href="#cb13-20" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb13-21"><a href="#cb13-21" aria-hidden="true" tabindex="-1"></a>  <span class="co">// implies: beta ~ multi_normal(mu, Sigma)</span></span>
+<span id="cb13-22"><a href="#cb13-22" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-23"><a href="#cb13-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of <code>alpha</code> and second, it avoids the need to invert <code>Sigma</code> every time <code>multi_normal</code> is evaluated.</p>
+<p>The Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix <code>Tau</code> were fixed data but the vector of standard deviations <code>sigma</code> were unknown parameters, then a reparameterization of <code>beta</code> in terms of <code>alpha</code> could be implemented as follows.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[K] Tau;</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L;</span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a>  L = cholesky_decompose(Tau);</span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb14-12"><a href="#cb14-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] alpha;</span>
+<span id="cb14-13"><a href="#cb14-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] sigma;</span>
+<span id="cb14-14"><a href="#cb14-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb14-15"><a href="#cb14-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-16"><a href="#cb14-16" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb14-17"><a href="#cb14-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb14-18"><a href="#cb14-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// This equals mu + diag_matrix(sigma) * L * alpha;</span></span>
+<span id="cb14-19"><a href="#cb14-19" aria-hidden="true" tabindex="-1"></a>  beta = mu + sigma .* (L * alpha);</span>
+<span id="cb14-20"><a href="#cb14-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-21"><a href="#cb14-21" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb14-22"><a href="#cb14-22" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb14-23"><a href="#cb14-23" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb14-24"><a href="#cb14-24" aria-hidden="true" tabindex="-1"></a>  <span class="co">// implies: beta ~ multi_normal(mu,</span></span>
+<span id="cb14-25"><a href="#cb14-25" aria-hidden="true" tabindex="-1"></a>  <span class="co">//  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))</span></span>
+<span id="cb14-26"><a href="#cb14-26" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb14-27"><a href="#cb14-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.</p>
+<p>A Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let <span class="math inline">\(L\)</span> be the Cholesky factor of a <span class="math inline">\(K \times K\)</span> positive definite scale matrix <span class="math inline">\(S\)</span> and let <span class="math inline">\(\nu\)</span> be the degrees of freedom. If <span class="math display">\[
+A = \begin{pmatrix}
+\sqrt{c_{1}} &amp; 0            &amp; \cdots                &amp; 0 \\
+z_{21}       &amp; \sqrt{c_{2}} &amp; \ddots                &amp; \vdots \\
+\vdots       &amp; \ddots       &amp; \ddots                &amp; 0 \\
+z_{K1}       &amp; \cdots       &amp; z_{K\left(K-1\right)} &amp; \sqrt{c_{K}}
+\end{pmatrix},
+\]</span> where each <span class="math inline">\(c_i\)</span> is distributed chi-squared with <span class="math inline">\(\nu - i + 1\)</span> degrees of freedom and each <span class="math inline">\(z_{ij}\)</span> is distributed standard normal, then <span class="math inline">\(W = LAA^{\top}L^{\top}\)</span> is distributed Wishart with scale matrix <span class="math inline">\(S = LL^{\top}\)</span> and degrees of freedom <span class="math inline">\(\nu\)</span>. Such a reparameterization can be implemented by the following Stan code:</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=K + <span class="dv">2</span>&gt; nu</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L; <span class="co">// Cholesky factor of scale matrix</span></span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] y;</span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb15-11"><a href="#cb15-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] c;</span>
+<span id="cb15-12"><a href="#cb15-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="fl">0.5</span> * K * (K - <span class="dv">1</span>)] z;</span>
+<span id="cb15-13"><a href="#cb15-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-14"><a href="#cb15-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-15"><a href="#cb15-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-16"><a href="#cb15-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] A;</span>
+<span id="cb15-17"><a href="#cb15-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> count = <span class="dv">1</span>;</span>
+<span id="cb15-18"><a href="#cb15-18" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:(K - <span class="dv">1</span>)) {</span>
+<span id="cb15-19"><a href="#cb15-19" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> (j + <span class="dv">1</span>):K) {</span>
+<span id="cb15-20"><a href="#cb15-20" aria-hidden="true" tabindex="-1"></a>      A[i, j] = z[count];</span>
+<span id="cb15-21"><a href="#cb15-21" aria-hidden="true" tabindex="-1"></a>      count += <span class="dv">1</span>;</span>
+<span id="cb15-22"><a href="#cb15-22" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-23"><a href="#cb15-23" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:(j - <span class="dv">1</span>)) {</span>
+<span id="cb15-24"><a href="#cb15-24" aria-hidden="true" tabindex="-1"></a>      A[i, j] = <span class="fl">0.0</span>;</span>
+<span id="cb15-25"><a href="#cb15-25" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-26"><a href="#cb15-26" aria-hidden="true" tabindex="-1"></a>    A[j, j] = sqrt(c[j]);</span>
+<span id="cb15-27"><a href="#cb15-27" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-28"><a href="#cb15-28" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:(K - <span class="dv">1</span>)) {</span>
+<span id="cb15-29"><a href="#cb15-29" aria-hidden="true" tabindex="-1"></a>    A[i, K] = <span class="dv">0</span>;</span>
+<span id="cb15-30"><a href="#cb15-30" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-31"><a href="#cb15-31" aria-hidden="true" tabindex="-1"></a>  A[K, K] = sqrt(c[K]);</span>
+<span id="cb15-32"><a href="#cb15-32" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-33"><a href="#cb15-33" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb15-34"><a href="#cb15-34" aria-hidden="true" tabindex="-1"></a>    c[i] ~ chi_square(nu - i + <span class="dv">1</span>);</span>
+<span id="cb15-35"><a href="#cb15-35" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-36"><a href="#cb15-36" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-37"><a href="#cb15-37" aria-hidden="true" tabindex="-1"></a>  z ~ std_normal();</span>
+<span id="cb15-38"><a href="#cb15-38" aria-hidden="true" tabindex="-1"></a>  <span class="co">// implies: L * A * A' * L' ~ wishart(nu, L * L')</span></span>
+<span id="cb15-39"><a href="#cb15-39" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal_cholesky(mu, L * A);</span>
+<span id="cb15-40"><a href="#cb15-40" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-41"><a href="#cb15-41" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of <code>z</code> and second, it avoids the need to invert the covariance matrix, <span class="math inline">\(W\)</span> every time <code>wishart</code> is evaluated. Third, if <span class="math inline">\(W\)</span> is to be used with a multivariate normal distribution, you can pass <span class="math inline">\(L A\)</span> to the more efficient <code>multi_normal_cholesky</code> function, rather than passing <span class="math inline">\(W\)</span> to <code>multi_normal</code>.</p>
+<p>If <span class="math inline">\(W\)</span> is distributed Wishart with scale matrix <span class="math inline">\(S\)</span> and degrees of freedom <span class="math inline">\(\nu\)</span>, then <span class="math inline">\(W^{-1}\)</span> is distributed inverse Wishart with inverse scale matrix <span class="math inline">\(S^{-1}\)</span> and degrees of freedom <span class="math inline">\(\nu\)</span>. Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since <span class="math inline">\(W = L A A^{\top} L^{\top}\)</span>, <span class="math inline">\(W^{-1} = L^{{\top}^{-1}} A^{{\top}^{-1}} A^{-1} L^{-1}\)</span>, where all four inverses exist, but <span class="math inline">\(L^{{-1}^{\top}} = L^{{\top}^{-1}}\)</span> and <span class="math inline">\(A^{{-1}^{\top}} = A^{{\top}^{-1}}\)</span>. We can slightly modify the above Stan code for this case:</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=K + <span class="dv">2</span>&gt; nu</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L; <span class="co">// Cholesky factor of scale matrix</span></span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] eye;</span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L_inv;</span>
+<span id="cb16-10"><a href="#cb16-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb16-11"><a href="#cb16-11" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb16-12"><a href="#cb16-12" aria-hidden="true" tabindex="-1"></a>      eye[i, j] = <span class="fl">0.0</span>;</span>
+<span id="cb16-13"><a href="#cb16-13" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb16-14"><a href="#cb16-14" aria-hidden="true" tabindex="-1"></a>    eye[j, j] = <span class="fl">1.0</span>;</span>
+<span id="cb16-15"><a href="#cb16-15" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-16"><a href="#cb16-16" aria-hidden="true" tabindex="-1"></a>  L_inv = mdivide_left_tri_low(L, eye);</span>
+<span id="cb16-17"><a href="#cb16-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-18"><a href="#cb16-18" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb16-19"><a href="#cb16-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] c;</span>
+<span id="cb16-20"><a href="#cb16-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="fl">0.5</span> * K * (K - <span class="dv">1</span>)] z;</span>
+<span id="cb16-21"><a href="#cb16-21" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-22"><a href="#cb16-22" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-23"><a href="#cb16-23" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-24"><a href="#cb16-24" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] A;</span>
+<span id="cb16-25"><a href="#cb16-25" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] A_inv_L_inv;</span>
+<span id="cb16-26"><a href="#cb16-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> count;</span>
+<span id="cb16-27"><a href="#cb16-27" aria-hidden="true" tabindex="-1"></a>  count = <span class="dv">1</span>;</span>
+<span id="cb16-28"><a href="#cb16-28" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:(K - <span class="dv">1</span>)) {</span>
+<span id="cb16-29"><a href="#cb16-29" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> (j + <span class="dv">1</span>):K) {</span>
+<span id="cb16-30"><a href="#cb16-30" aria-hidden="true" tabindex="-1"></a>      A[i, j] = z[count];</span>
+<span id="cb16-31"><a href="#cb16-31" aria-hidden="true" tabindex="-1"></a>      count += <span class="dv">1</span>;</span>
+<span id="cb16-32"><a href="#cb16-32" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb16-33"><a href="#cb16-33" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:(j - <span class="dv">1</span>)) {</span>
+<span id="cb16-34"><a href="#cb16-34" aria-hidden="true" tabindex="-1"></a>      A[i, j] = <span class="fl">0.0</span>;</span>
+<span id="cb16-35"><a href="#cb16-35" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb16-36"><a href="#cb16-36" aria-hidden="true" tabindex="-1"></a>    A[j, j] = sqrt(c[j]);</span>
+<span id="cb16-37"><a href="#cb16-37" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-38"><a href="#cb16-38" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:(K - <span class="dv">1</span>)) {</span>
+<span id="cb16-39"><a href="#cb16-39" aria-hidden="true" tabindex="-1"></a>    A[i, K] = <span class="dv">0</span>;</span>
+<span id="cb16-40"><a href="#cb16-40" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-41"><a href="#cb16-41" aria-hidden="true" tabindex="-1"></a>  A[K, K] = sqrt(c[K]);</span>
+<span id="cb16-42"><a href="#cb16-42" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb16-43"><a href="#cb16-43" aria-hidden="true" tabindex="-1"></a>  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);</span>
+<span id="cb16-44"><a href="#cb16-44" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb16-45"><a href="#cb16-45" aria-hidden="true" tabindex="-1"></a>    c[i] ~ chi_square(nu - i + <span class="dv">1</span>);</span>
+<span id="cb16-46"><a href="#cb16-46" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-47"><a href="#cb16-47" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb16-48"><a href="#cb16-48" aria-hidden="true" tabindex="-1"></a>  z ~ std_normal(); <span class="co">// implies: crossprod(A_inv_L_inv) ~</span></span>
+<span id="cb16-49"><a href="#cb16-49" aria-hidden="true" tabindex="-1"></a>  <span class="co">// inv_wishart(nu, L_inv' * L_inv)</span></span>
+<span id="cb16-50"><a href="#cb16-50" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-51"><a href="#cb16-51" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Another candidate for reparameterization is the Dirichlet distribution with all <span class="math inline">\(K\)</span> shape parameters equal. <span class="citation" data-cites="ZyczkowskiSommers:2001">Zyczkowski and Sommers (<a href="#ref-ZyczkowskiSommers:2001" role="doc-biblioref">2001</a>)</span> shows that if <span class="math inline">\(\theta_i\)</span> is equal to the sum of <span class="math inline">\(\beta\)</span> independent squared standard normal variates and <span class="math inline">\(\rho_i = \frac{\theta_i}{\sum \theta_i}\)</span>, then the <span class="math inline">\(K\)</span>-vector <span class="math inline">\(\rho\)</span> is distributed Dirichlet with all shape parameters equal to <span class="math inline">\(\frac{\beta}{2}\)</span>. In particular, if <span class="math inline">\(\beta = 2\)</span>, then <span class="math inline">\(\rho\)</span> is uniformly distributed on the unit simplex. Thus, we can make <span class="math inline">\(\rho\)</span> be a transformed parameter to reduce dependence, as in:</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; beta;</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">vector</span>[beta] z;</span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb17-10"><a href="#cb17-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] rho;</span>
+<span id="cb17-11"><a href="#cb17-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb17-12"><a href="#cb17-12" aria-hidden="true" tabindex="-1"></a>    rho[k] = dot_self(z[k]); <span class="co">// sum-of-squares</span></span>
+<span id="cb17-13"><a href="#cb17-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-14"><a href="#cb17-14" aria-hidden="true" tabindex="-1"></a>  rho = rho / sum(rho);</span>
+<span id="cb17-15"><a href="#cb17-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-16"><a href="#cb17-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb17-17"><a href="#cb17-17" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb17-18"><a href="#cb17-18" aria-hidden="true" tabindex="-1"></a>    z[k] ~ std_normal();</span>
+<span id="cb17-19"><a href="#cb17-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-20"><a href="#cb17-20" aria-hidden="true" tabindex="-1"></a>  <span class="co">// implies: rho ~ dirichlet(0.5 * beta * ones)</span></span>
+<span id="cb17-21"><a href="#cb17-21" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb17-22"><a href="#cb17-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="vectorization" class="level2">
+<h2 class="anchored" data-anchor-id="vectorization">Vectorization</h2>
+<section id="gradient-bottleneck" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="gradient-bottleneck">Gradient bottleneck</h3>
+<p>Stan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.</p>
+</section>
+<section id="vectorizing-summations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectorizing-summations">Vectorizing summations</h3>
+<p>Because of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the <code>sum()</code> operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where <code>foo()</code> is some operation that depends on <code>n</code>.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  total += foo(n,...);</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This code has to create intermediate representations for each of the <code>N</code> summands.</p>
+<p>A faster alternative is to copy the values into a vector, then apply the <code>sum()</code> operator, as in the following refactoring.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] summands;</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>    summands[n] = foo(n,...);</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>  total = sum(summands);</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Syntactically, the replacement is a statement block delineated by curly brackets (<code>{</code>, <code>}</code>), starting with the definition of the local variable <code>summands</code>.</p>
+<p>Even though it involves extra work to allocate the <code>summands</code> vector and copy <code>N</code> values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing <code>total</code> within the loop.</p>
+</section>
+<section id="vectorization-through-matrix-operations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectorization-through-matrix-operations">Vectorization through matrix operations</h3>
+<p>The following program directly encodes a linear regression with fixed unit noise using a two-dimensional array <code>x</code> of predictors, an array <code>y</code> of outcomes, and an array <code>beta</code> of regression coefficients.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K, N] <span class="dt">real</span> x;</span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">real</span> beta;</span>
+<span id="cb20-9"><a href="#cb20-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-10"><a href="#cb20-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb20-11"><a href="#cb20-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb20-12"><a href="#cb20-12" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> gamma = <span class="dv">0</span>;</span>
+<span id="cb20-13"><a href="#cb20-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb20-14"><a href="#cb20-14" aria-hidden="true" tabindex="-1"></a>      gamma += x[n, k] * beta[k];</span>
+<span id="cb20-15"><a href="#cb20-15" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb20-16"><a href="#cb20-16" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(gamma, <span class="dv">1</span>);</span>
+<span id="cb20-17"><a href="#cb20-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-18"><a href="#cb20-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] x;</span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(dot_product(x[n], beta), <span class="dv">1</span>);</span>
+<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-14"><a href="#cb21-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ normal(x[n]' * beta, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>The inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">row_vector</span>[K] beta_t;</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>  beta_t = beta';</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(beta_t * x[n], <span class="dv">1</span>);</span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The problem with transposition could be completely solved by directly encoding the <code>x</code> as a row vector, as in the following example.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">row_vector</span>[K] x;</span>
+<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb24-8"><a href="#cb24-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb24-9"><a href="#cb24-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb24-10"><a href="#cb24-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb24-11"><a href="#cb24-11" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(x[n] * beta, <span class="dv">1</span>);</span>
+<span id="cb24-12"><a href="#cb24-12" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb24-13"><a href="#cb24-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Declaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.</p>
+<p>Having said all this, the most efficient way to code this model is with direct matrix multiplication, as in</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb25-9"><a href="#cb25-9" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, <span class="dv">1</span>);</span>
+<span id="cb25-10"><a href="#cb25-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.</p>
+</section>
+<section id="vectorized-probability-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectorized-probability-functions">Vectorized probability functions</h3>
+<p>The final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.</p>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb26-2"><a href="#cb26-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb26-3"><a href="#cb26-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb26-4"><a href="#cb26-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
+<span id="cb26-5"><a href="#cb26-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb26-6"><a href="#cb26-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb26-7"><a href="#cb26-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb26-8"><a href="#cb26-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb26-9"><a href="#cb26-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb26-10"><a href="#cb26-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb26-11"><a href="#cb26-11" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, <span class="dv">1</span>);</span>
+<span id="cb26-12"><a href="#cb26-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication <code>x * beta</code> in the model block is a vector of the same length as <code>y</code>.</p>
+<p>The probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, <code>y</code> is a vector of size <code>N</code>, <code>x * beta</code> is a vector of size <code>N</code>, and <code>1</code> is a scalar.</p>
+</section>
+<section id="reshaping-data-for-vectorization" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reshaping-data-for-vectorization">Reshaping data for vectorization</h3>
+<p>Sometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).</p>
+<p>John Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu[ii[n]], sigma);</span>
+<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The brute force vectorization would build up a mean vector and then vectorize all at once.</p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu_ii;</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb28-4"><a href="#cb28-4" aria-hidden="true" tabindex="-1"></a>    mu_ii[n] = mu[ii[n]];</span>
+<span id="cb28-5"><a href="#cb28-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb28-6"><a href="#cb28-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu_ii, sigma);</span>
+<span id="cb28-7"><a href="#cb28-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If there aren’t many levels (values <code>ii[n]</code> can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector <code>y</code>, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>y_1 ~ normal(mu[<span class="dv">1</span>], sigma);</span>
+<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>y_5 ~ normal(mu[<span class="dv">5</span>], sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This way, both the <code>mu</code> and <code>sigma</code> parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.</p>
+</section>
+</section>
+<section id="exploiting-sufficient-statistics" class="level2">
+<h2 class="anchored" data-anchor-id="exploiting-sufficient-statistics">Exploiting sufficient statistics</h2>
+<p>In some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.</p>
+<section id="bernoulli-sufficient-statistics" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="bernoulli-sufficient-statistics">Bernoulli sufficient statistics</h3>
+<p>Consider the following Bernoulli sampling model.</p>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb30-4"><a href="#cb30-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb30-5"><a href="#cb30-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; beta;</span>
+<span id="cb30-6"><a href="#cb30-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb30-7"><a href="#cb30-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb30-8"><a href="#cb30-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb30-9"><a href="#cb30-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb30-10"><a href="#cb30-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb30-11"><a href="#cb30-11" aria-hidden="true" tabindex="-1"></a>  theta ~ beta(alpha, beta);</span>
+<span id="cb30-12"><a href="#cb30-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb30-13"><a href="#cb30-13" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli(theta);</span>
+<span id="cb30-14"><a href="#cb30-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb30-15"><a href="#cb30-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this model, the sum of positive outcomes in <code>y</code> is a sufficient statistic for the chance of success <code>theta</code>. The model may be recoded using the binomial distribution as follows.</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a>theta ~ beta(alpha, beta);</span>
+<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a>sum(y) ~ binomial(N, theta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because truth is represented as one and falsehood as zero, the sum <code>sum(y)</code> of a binary vector <code>y</code> is equal to the number of positive outcomes out of a total of <code>N</code> trials.</p>
+<p>This can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only <span class="math inline">\(K\)</span> possible discrete outcomes, <span class="math inline">\(z_1, \dotsc, z_K\)</span>, but there are <span class="math inline">\(N\)</span> observations, where <span class="math inline">\(N\)</span> is much larger than <span class="math inline">\(K\)</span>. If <span class="math inline">\(f_k\)</span> is the frequency of outcome <span class="math inline">\(z_k\)</span>, then the entire likelihood with distribution <code>foo</code> can be coded as follows.</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> f[k] * foo_lpmf(z[k] | ...);</span>
+<span id="cb32-3"><a href="#cb32-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the ellipses are the parameters of the log probability mass function for distribution <code>foo</code> (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).</p>
+<p>The resulting program looks like a “weighted” regression, but here the weights <code>f[k]</code> are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies <code>f[k]</code> should be counted once in the transformed data block and stored.</p>
+<p>The same trick works for combining multiple binomial observations.</p>
+</section>
+<section id="normal-sufficient-statistics" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="normal-sufficient-statistics">Normal sufficient statistics</h3>
+<p>Consider the following Stan model for fitting a normal distribution to data.</p>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb33-2"><a href="#cb33-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb33-3"><a href="#cb33-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb33-4"><a href="#cb33-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb33-5"><a href="#cb33-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb33-6"><a href="#cb33-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb33-7"><a href="#cb33-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb33-8"><a href="#cb33-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb33-9"><a href="#cb33-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb33-10"><a href="#cb33-10" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
+<span id="cb33-11"><a href="#cb33-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With the vectorized form used for <code>y</code>, Stan is clever enough to only evaluate <code>log(sigma)</code> once, but it still has to evaluate the normal for all of <code>y[1]</code> to <code>y[N]</code>, which involves adding up all the squared differences from the mean and then dividing by <code>sigma</code> squared.</p>
+<p>An equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.</p>
+<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb34-2"><a href="#cb34-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb34-3"><a href="#cb34-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb34-4"><a href="#cb34-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb34-5"><a href="#cb34-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb34-6"><a href="#cb34-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mean_y = mean(y);</span>
+<span id="cb34-7"><a href="#cb34-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; var_y = variance(y);</span>
+<span id="cb34-8"><a href="#cb34-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> nm1_over2 = <span class="fl">0.5</span> * (N - <span class="dv">1</span>);</span>
+<span id="cb34-9"><a href="#cb34-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sqrt_N = sqrt(N);</span>
+<span id="cb34-10"><a href="#cb34-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb34-11"><a href="#cb34-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb34-12"><a href="#cb34-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb34-13"><a href="#cb34-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb34-14"><a href="#cb34-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb34-15"><a href="#cb34-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb34-16"><a href="#cb34-16" aria-hidden="true" tabindex="-1"></a>  mean_y ~ normal(mu, sigma / sqrt_N);</span>
+<span id="cb34-17"><a href="#cb34-17" aria-hidden="true" tabindex="-1"></a>  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^<span class="dv">2</span>);</span>
+<span id="cb34-18"><a href="#cb34-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.</p>
+<p>The original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on <code>mu</code> and <code>sigma</code> are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.</p>
+<p>This transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.</p>
+</section>
+<section id="poisson-sufficient-statistics" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="poisson-sufficient-statistics">Poisson sufficient statistics</h3>
+<p>The Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace</p>
+<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a>y ~ poisson(lambda);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with</p>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a>sum(y) ~ poisson(size(y) * lambda);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This will work even if <code>y</code> is a parameter vector because no Jacobian adjustment is required for summation.</p>
+</section>
+</section>
+<section id="aggregating-common-subexpressions" class="level2">
+<h2 class="anchored" data-anchor-id="aggregating-common-subexpressions">Aggregating common subexpressions</h2>
+<p>If an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using <code>exp(theta)</code> in multiple places, declare a local variable to store its value and reuse the local variable.</p>
+<p>Another case that may not be so obvious is with two multilevel parameters, say <code>a[ii[n]] + b[jj[n]]</code>. If <code>a</code> and <code>b</code> are small (i.e., do not have many levels), then a table <code>a_b</code> of their sums can be created, with</p>
+<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[size(a), size(b)] a_b;</span>
+<span id="cb37-2"><a href="#cb37-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:size(a)) {</span>
+<span id="cb37-3"><a href="#cb37-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:size(b)) {</span>
+<span id="cb37-4"><a href="#cb37-4" aria-hidden="true" tabindex="-1"></a>    a_b[i, j] = a[i] + b[j];</span>
+<span id="cb37-5"><a href="#cb37-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb37-6"><a href="#cb37-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then the sum can be replaced with <code>a_b[ii[n], jj[n]]</code>.</p>
+</section>
+<section id="exploiting-conjugacy" class="level2">
+<h2 class="anchored" data-anchor-id="exploiting-conjugacy">Exploiting conjugacy</h2>
+<p>Continuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.</p>
+<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a>theta ~ beta(alpha + sum(y), beta + N - sum(y));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To make the model even more efficient, a transformed data variable defined to be <code>sum(y)</code> could be used in the place of <code>sum(y)</code>.</p>
+</section>
+<section id="standardizing-predictors" class="level2">
+<h2 class="anchored" data-anchor-id="standardizing-predictors">Standardizing predictors</h2>
+<p>Standardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:</p>
+<ul>
+<li>It helps in faster convergence of MCMC chains.</li>
+<li>It makes the model less sensitive to the specifics of the parameterization.</li>
+<li>It aids in the interpretation and comparison of the importance of coefficients across different predictors.</li>
+</ul>
+<p>When there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.</p>
+<p>Suppose that <span class="math inline">\(y = (y_1,\dotsc,y_N)\)</span> is a vector of <span class="math inline">\(N\)</span> outcomes and <span class="math inline">\(x = (x_1,\dotsc,x_N)\)</span> the corresponding vector of <span class="math inline">\(N\)</span> predictors. A simple linear regression involving an intercept coefficient <span class="math inline">\(\alpha\)</span> and slope coefficient <span class="math inline">\(\beta\)</span> can be expressed as <span class="math display">\[
+y_n = \alpha + \beta x_n + \epsilon_n,
+\]</span> where <span class="math display">\[
+\epsilon_n \sim \textsf{normal}(0,\sigma).
+\]</span></p>
+<p>If <span class="math inline">\(x\)</span> has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values <span class="math inline">\(x_n\)</span>. First the elements of <span class="math inline">\(x\)</span> are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.</p>
+<p>The mean of <span class="math inline">\(x\)</span> is given by:</p>
+<p><span class="math display">\[
+mean_x = \frac{1}{N} \sum_{n=1}^{N} x_n
+\]</span></p>
+<p>The standard deviation of <span class="math inline">\(x\)</span> is calculated as: <span class="math display">\[
+sd_x = {\left({\frac{1}{N} \sum_{n=1}^{N} (x_n - mean_x)^2}\right)}^{1/2}
+\]</span></p>
+<p>With these, we compute the <span class="math inline">\(z\)</span>, the standardized predictors</p>
+<p><span class="math display">\[
+z_n = \frac{x_n - mean_x}{sd_x}
+\]</span></p>
+<p>where <span class="math inline">\(z_n\)</span> is the standardized value corresponding to <span class="math inline">\(x_n\)</span>.</p>
+<p>The inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.</p>
+<p><span class="math display">\[
+x_n = z_n sd_x + mean_x
+\]</span></p>
+<p>Standardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.</p>
+<p>Consider the following initial model.</p>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb39-2"><a href="#cb39-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb39-3"><a href="#cb39-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb39-4"><a href="#cb39-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb39-5"><a href="#cb39-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb39-6"><a href="#cb39-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb39-7"><a href="#cb39-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb39-8"><a href="#cb39-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb39-9"><a href="#cb39-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb39-10"><a href="#cb39-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb39-11"><a href="#cb39-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb39-12"><a href="#cb39-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// priors</span></span>
+<span id="cb39-13"><a href="#cb39-13" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb39-14"><a href="#cb39-14" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb39-15"><a href="#cb39-15" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb39-16"><a href="#cb39-16" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood</span></span>
+<span id="cb39-17"><a href="#cb39-17" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta + alpha, sigma);</span>
+<span id="cb39-18"><a href="#cb39-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.</p>
+<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb40-2"><a href="#cb40-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb40-3"><a href="#cb40-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb40-4"><a href="#cb40-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb40-5"><a href="#cb40-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb40-6"><a href="#cb40-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb40-7"><a href="#cb40-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mean_x = mean(x);</span>
+<span id="cb40-8"><a href="#cb40-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sd_x = sd(x);</span>
+<span id="cb40-9"><a href="#cb40-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_std = (x - mean_x) / sd_x;</span>
+<span id="cb40-10"><a href="#cb40-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb40-11"><a href="#cb40-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb40-12"><a href="#cb40-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha_std;</span>
+<span id="cb40-13"><a href="#cb40-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta_std;</span>
+<span id="cb40-14"><a href="#cb40-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_std;</span>
+<span id="cb40-15"><a href="#cb40-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb40-16"><a href="#cb40-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb40-17"><a href="#cb40-17" aria-hidden="true" tabindex="-1"></a>  alpha_std ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb40-18"><a href="#cb40-18" aria-hidden="true" tabindex="-1"></a>  beta_std ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb40-19"><a href="#cb40-19" aria-hidden="true" tabindex="-1"></a>  sigma_std ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb40-20"><a href="#cb40-20" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x_std * beta_std + alpha_std, sigma_std);</span>
+<span id="cb40-21"><a href="#cb40-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data <code>x_std</code> is defined in terms of variables <code>mean_x</code> and <code>sd_x</code>; by declaring these variables in the <code>transformed data</code> block, they will be available in all following blocks, and therefore can be used in the <code>generated quantities</code> block to record the “natural” parameters <code>alpha</code> and <code>beta</code>.</p>
+<p>The fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.</p>
+<p>The original regression <span class="math display">\[
+y_n = \alpha + \beta x_n + \epsilon_n
+\]</span> has been transformed to a regression on the standardized data variable <span class="math inline">\(z\)</span>,</p>
+<p><span class="math display">\[
+y_n = \alpha' + \beta' z_n + \epsilon_n.
+\]</span></p>
+<p>The likelihood is specified in terms of the standardized parameters. The original slope <span class="math inline">\(\beta\)</span> is the standardized slope <span class="math inline">\(\beta'\)</span> scaled by the inverse of the standard deviation of <span class="math inline">\(x\)</span>. The original intercept <span class="math inline">\(\alpha\)</span> is the intercept from the standardized model <span class="math inline">\(\alpha'\)</span>, corrected for the effect of scaling and centering <span class="math inline">\(x\)</span>. Thus, the formulas to retrieve <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> from <span class="math inline">\(\alpha'\)</span> and <span class="math inline">\(\beta'\)</span> are:</p>
+<p><span class="math display">\[\begin{align*}
+\beta = \frac{\beta'}{\sigma_x} \\
+\alpha = \alpha' - \beta' \frac{\mu_x}{\sigma_x}
+\end{align*}\]</span></p>
+<p>These recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,</p>
+<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb41-2"><a href="#cb41-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta = beta_std / sd_x;</span>
+<span id="cb41-3"><a href="#cb41-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha = alpha_std - beta_std * mean_x / sd_x;</span>
+<span id="cb41-4"><a href="#cb41-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb41-5"><a href="#cb41-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When there are multiple real-valued predictors, i.e., when <code>K</code> is the number of predictors, <code>x</code> is an <span class="math inline">\(N \times K\)</span> matrix, and <code>beta</code> ia <span class="math inline">\(K\)</span>-vector of coefficients, then <code>x * beta</code> is an <span class="math inline">\(N\)</span>-vector of predictions, one for each of the <span class="math inline">\(N\)</span> data items. When <span class="math inline">\(K \ll N\)</span> the <a href="../stan-users-guide/regression.html#QR-reparameterization.section">QR reparameterization</a> is recommended for linear and generalized linear models unless there is an informative prior on the location of <span class="math inline">\(\beta\)</span>.</p>
+<section id="standard-normal-distribution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="standard-normal-distribution">Standard normal distribution</h3>
+<p>For many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use</p>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a>y ~ std_normal();</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>than to use</p>
+<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>because the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.</p>
+</section>
+</section>
+<section id="using-map-reduce" class="level2">
+<h2 class="anchored" data-anchor-id="using-map-reduce">Using map-reduce</h2>
+<p>The map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the <a href="../stan-users-guide/parallelization.html">map-reduce chapter</a> for more information on implementing map-reduce operations.</p>
+<p>Map-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Betancourt-Girolami:2013" class="csl-entry" role="listitem">
+Betancourt, Michael, and Mark Girolami. 2013. <span>“<span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo for Hierarchical Models.”</span> <em>arXiv</em> 1312.0906. <a href="http://arxiv.org/abs/1312.0906">http://arxiv.org/abs/1312.0906</a>.
+</div>
+<div id="ref-Neal:2003" class="csl-entry" role="listitem">
+Neal, Radford M. 2003. <span>“Slice Sampling.”</span> <em>Annals of Statistics</em> 31 (3): 705–67.
+</div>
+<div id="ref-papa-et-al:2007" class="csl-entry" role="listitem">
+Papaspiliopoulos, Omiros, Gareth O. Roberts, and Martin Sköld. 2007. <span>“A General Framework for the Parametrization of Hierarchical Models.”</span> <em>Statistical Science</em> 22 (1): 59–73.
+</div>
+<div id="ref-ZyczkowskiSommers:2001" class="csl-entry" role="listitem">
+Zyczkowski, K., and H. J. Sommers. 2001. <span>“Induced Measures in the Space of Mixed Quantum States.”</span> <em>Journal of Physics A: Mathematical and General</em> 34 (35): 7111.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>This parameterization came to be known on our mailing lists as the “Matt trick” after Matt Hoffman, who independently came up with it while fitting hierarchical models in Stan.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Stan uses its own arena-based allocation, so allocation and deallocation are faster than with a raw call to <code>new</code>.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>Future versions of Stan may remove this inefficiency by more fully exploiting expression templates inside the Eigen C++ matrix library. This will require enhancing Eigen to deal with mixed-type arguments, such as the type <code>double</code> used for constants and the algorithmic differentiation type <code>stan::math::var</code> used for variables.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/reparameterization.html" class="pagination-link" aria-label="Reparameterization and Change of Variables">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Reparameterization and Change of Variables</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/parallelization.html" class="pagination-link" aria-label="Parallelization">
+        <span class="nav-page-text">Parallelization</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/efficiency-tuning.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/finite-mixtures.html b/docs/2_39/stan-users-guide/finite-mixtures.html
new file mode 100644
index 000000000..b3a240090
--- /dev/null
+++ b/docs/2_39/stan-users-guide/finite-mixtures.html
@@ -0,0 +1,1659 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Finite Mixtures</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/measurement-error.html" rel="next">
+<link href="../stan-users-guide/truncation-censoring.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Finite Mixtures">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/finite-mixtures.html">Finite Mixtures</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#mixture-modeling.chapter" id="toc-mixture-modeling.chapter" class="nav-link active" data-scroll-target="#mixture-modeling.chapter">Finite Mixtures</a>
+  <ul class="collapse">
+  <li><a href="#clustering-mixture.section" id="toc-clustering-mixture.section" class="nav-link" data-scroll-target="#clustering-mixture.section">Relation to clustering</a></li>
+  <li><a href="#latent-discrete-parameterization" id="toc-latent-discrete-parameterization" class="nav-link" data-scroll-target="#latent-discrete-parameterization">Latent discrete parameterization</a></li>
+  <li><a href="#summing-out-the-responsibility-parameter" id="toc-summing-out-the-responsibility-parameter" class="nav-link" data-scroll-target="#summing-out-the-responsibility-parameter">Summing out the responsibility parameter</a>
+  <ul class="collapse">
+  <li><a href="#log-sum-of-exponentials-linear-sums-on-the-log-scale" id="toc-log-sum-of-exponentials-linear-sums-on-the-log-scale" class="nav-link" data-scroll-target="#log-sum-of-exponentials-linear-sums-on-the-log-scale">Log sum of exponentials: linear Sums on the log scale</a></li>
+  <li><a href="#dropping-uniform-mixture-ratios" id="toc-dropping-uniform-mixture-ratios" class="nav-link" data-scroll-target="#dropping-uniform-mixture-ratios">Dropping uniform mixture ratios</a></li>
+  <li><a href="#recovering-posterior-mixture-proportions" id="toc-recovering-posterior-mixture-proportions" class="nav-link" data-scroll-target="#recovering-posterior-mixture-proportions">Recovering posterior mixture proportions</a></li>
+  <li><a href="#estimating-parameters-of-a-mixture" id="toc-estimating-parameters-of-a-mixture" class="nav-link" data-scroll-target="#estimating-parameters-of-a-mixture">Estimating parameters of a mixture</a></li>
+  </ul></li>
+  <li><a href="#vectorizing-mixtures" id="toc-vectorizing-mixtures" class="nav-link" data-scroll-target="#vectorizing-mixtures">Vectorizing mixtures</a></li>
+  <li><a href="#mixture-inference.section" id="toc-mixture-inference.section" class="nav-link" data-scroll-target="#mixture-inference.section">Inferences supported by mixtures</a>
+  <ul class="collapse">
+  <li><a href="#mixtures-with-unidentifiable-components" id="toc-mixtures-with-unidentifiable-components" class="nav-link" data-scroll-target="#mixtures-with-unidentifiable-components">Mixtures with unidentifiable components</a></li>
+  <li><a href="#inference-under-label-switching" id="toc-inference-under-label-switching" class="nav-link" data-scroll-target="#inference-under-label-switching">Inference under label switching</a></li>
+  </ul></li>
+  <li><a href="#zero-inflated.section" id="toc-zero-inflated.section" class="nav-link" data-scroll-target="#zero-inflated.section">Zero-inflated and hurdle models</a>
+  <ul class="collapse">
+  <li><a href="#zero-inflation" id="toc-zero-inflation" class="nav-link" data-scroll-target="#zero-inflation">Zero inflation</a></li>
+  <li><a href="#hurdle-models" id="toc-hurdle-models" class="nav-link" data-scroll-target="#hurdle-models">Hurdle models</a></li>
+  </ul></li>
+  <li><a href="#priors-and-effective-data-size-in-mixture-models" id="toc-priors-and-effective-data-size-in-mixture-models" class="nav-link" data-scroll-target="#priors-and-effective-data-size-in-mixture-models">Priors and effective data size in mixture models</a>
+  <ul class="collapse">
+  <li><a href="#comparison-to-model-averaging" id="toc-comparison-to-model-averaging" class="nav-link" data-scroll-target="#comparison-to-model-averaging">Comparison to model averaging</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/finite-mixtures.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/finite-mixtures.html">Finite Mixtures</a></li></ol></nav></header>
+
+
+
+
+<section id="mixture-modeling.chapter" class="level1">
+<h1>Finite Mixtures</h1>
+<p>Finite mixture models of an outcome assume that the outcome is drawn from one of several distributions, the identity of which is controlled by a categorical mixing distribution. Mixture models typically have multimodal densities with modes near the modes of the mixture components. Mixture models may be parameterized in several ways, as described in the following sections. Mixture models may be used directly for modeling data with multimodal distributions, or they may be used as priors for other parameters.</p>
+<section id="clustering-mixture.section" class="level2">
+<h2 class="anchored" data-anchor-id="clustering-mixture.section">Relation to clustering</h2>
+<p>Clustering models, as discussed in the <a href="../stan-users-guide/clustering.html">clustering chapter</a>, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the <span class="math inline">\(K\)</span>-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.</p>
+</section>
+<section id="latent-discrete-parameterization" class="level2">
+<h2 class="anchored" data-anchor-id="latent-discrete-parameterization">Latent discrete parameterization</h2>
+<p>One way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider <span class="math inline">\(K\)</span> normal distributions with locations <span class="math inline">\(\mu_k \in \mathbb{R}\)</span> and scales <span class="math inline">\(\sigma_k \in (0,\infty)\)</span>. Now consider mixing them in proportion <span class="math inline">\(\lambda\)</span>, where <span class="math inline">\(\lambda_k \geq 0\)</span> and <span class="math inline">\(\sum_{k=1}^K \lambda_k = 1\)</span> (i.e., <span class="math inline">\(\lambda\)</span> lies in the unit <span class="math inline">\(K\)</span>-simplex). For each outcome <span class="math inline">\(y_n\)</span> there is a latent variable <span class="math inline">\(z_n\)</span> in <span class="math inline">\(\{ 1,\dotsc,K \}\)</span> with a categorical distribution parameterized by <span class="math inline">\(\lambda\)</span>, <span class="math display">\[
+z_n \sim \textsf{categorical}(\lambda).
+\]</span></p>
+<p>The variable <span class="math inline">\(y_n\)</span> is distributed according to the parameters of the mixture component <span class="math inline">\(z_n\)</span>, <span class="math display">\[
+y_n \sim \textsf{normal}(\mu_{z[n]},\sigma_{z[n]}).
+\]</span></p>
+<p>This model is not directly supported by Stan because it involves discrete parameters <span class="math inline">\(z_n\)</span>, but Stan can sample <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\sigma\)</span> by summing out the <span class="math inline">\(z\)</span> parameter as described in the next section.</p>
+</section>
+<section id="summing-out-the-responsibility-parameter" class="level2">
+<h2 class="anchored" data-anchor-id="summing-out-the-responsibility-parameter">Summing out the responsibility parameter</h2>
+<p>To implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If <span class="math inline">\(Y\)</span> is a mixture of <span class="math inline">\(K\)</span> normal distributions with locations <span class="math inline">\(\mu_k\)</span> and scales <span class="math inline">\(\sigma_k\)</span> with mixing proportions <span class="math inline">\(\lambda\)</span> in the unit <span class="math inline">\(K\)</span>-simplex, then <span class="math display">\[
+p_Y\left(y \mid \lambda, \mu, \sigma \right)
+=
+\sum_{k=1}^K \lambda_k \, \textsf{normal}\left(y \mid \mu_k, \sigma_k\right).
+\]</span></p>
+<section id="log-sum-of-exponentials-linear-sums-on-the-log-scale" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="log-sum-of-exponentials-linear-sums-on-the-log-scale">Log sum of exponentials: linear Sums on the log scale</h3>
+<p>The log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by <span class="math display">\[
+\texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}(a, b) = \log \left(\exp(a) + \exp(b)\right).
+\]</span></p>
+<p>If <span class="math inline">\(a\)</span> and <span class="math inline">\(b\)</span> are probabilities on the log scale, then <span class="math inline">\(\exp(a) +
+\exp(b)\)</span> is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in <code>log_sum_exp</code> function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as <span class="math display">\[
+\log \left( \exp(a) + \exp(b)\right)
+= c + \log \left( \exp(a - c) + \exp(b - c) \right),
+\]</span> where <span class="math inline">\(c = \max(a, b)\)</span>. In this evaluation, one of the terms, <span class="math inline">\(a - c\)</span> or <span class="math inline">\(b - c\)</span>, is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the <span class="math inline">\(\max(a, b)\)</span> out of the log-exp round trip.</p>
+<p>For example, the mixture of <span class="math inline">\(\textsf{normal}(-1, 2)\)</span> with <span class="math inline">\(\textsf{normal}(3, 1)\)</span>, with mixing proportion <span class="math inline">\(\lambda =
+[0.3,0.7]^{\top}\)</span>, can be implemented in Stan as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log_sum_exp(log(<span class="fl">0.3</span>) + normal_lpdf(y | -<span class="dv">1</span>, <span class="dv">2</span>),</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>                        log(<span class="fl">0.7</span>) + normal_lpdf(y | <span class="dv">3</span>, <span class="dv">1</span>));</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The log probability term is derived by taking <span class="math display">\[\begin{align*}
+\log\, &amp;p\left(y \mid \lambda,\mu,\sigma \right) \\
+&amp;= \log\big( 0.3 \times \textsf{normal}\left(y \mid -1,2 \right)
+            + 0.7 \times \textsf{normal}\left(y \mid 3,1 \right) \big) \\
+&amp;= \log\bigg( \exp\Big(\log\big(0.3 \times \textsf{normal}\left(y \mid -1,2 \right)\big)\Big)
+             + \exp\Big(\log\big(0.7 \times \textsf{normal}\left(y \mid 3,1 \right)\big)\Big) \bigg) \\
+&amp;= \texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}\big(
+                         \log(0.3) + \log \textsf{normal}\left(y \mid -1,2 \right),
+                         \log(0.7) + \log \textsf{normal}\left(y \mid 3,1 \right) \big).
+\end{align*}\]</span></p>
+</section>
+<section id="dropping-uniform-mixture-ratios" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="dropping-uniform-mixture-ratios">Dropping uniform mixture ratios</h3>
+<p>If a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>log_half = log(<span class="fl">0.5</span>);</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>        log_sum_exp(log_half + normal_lpdf(y[n] | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>]),</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>                    log_half + normal_lpdf(y[n] | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]));</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>then the <span class="math inline">\(\log 0.5\)</span> term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log_sum_exp(normal_lpdf(y[n] | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>]),</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>                        normal_lpdf(y[n] | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]));</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The same result holds if there are <span class="math inline">\(K\)</span> components and the mixing simplex <span class="math inline">\(\lambda\)</span> is symmetric, i.e., <span class="math display">\[
+\lambda = \left( \frac{1}{K}, \dotsc, \frac{1}{K} \right).
+\]</span></p>
+<p>The result follows from the identity <span class="math display">\[
+\texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}(c + a, c + b)
+=
+c + \texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}(a, b)
+\]</span> and the fact that adding a constant <span class="math inline">\(c\)</span> to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.</p>
+</section>
+<section id="recovering-posterior-mixture-proportions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="recovering-posterior-mixture-proportions">Recovering posterior mixture proportions</h3>
+<p>The posterior <span class="math inline">\(p(z_n \mid y_n, \mu, \sigma)\)</span> over the mixture indicator <span class="math inline">\(z_n
+\in 1:K\)</span> is often of interest as <span class="math inline">\(p(z_n = k \mid y, \mu, \sigma)\)</span> is the posterior probability that that observation <span class="math inline">\(y_n\)</span> was generated by mixture component <span class="math inline">\(k\)</span>. The posterior can be computed via Bayes’s rule, <span class="math display">\[\begin{align*}
+\Pr\!\left[z_n = k \mid y_n, \mu, \sigma, \lambda \right]
+  &amp;\propto p\left(y_n \mid z_n = k, \mu, \sigma\right)\, p\left(z_n = k \mid \lambda\right) \\
+  &amp;= \textsf{normal}\left(y_n \mid \mu_k, \sigma_k\right) \cdot \lambda_k.
+\end{align*}\]</span></p>
+<p>The normalization can be done via summation, because <span class="math inline">\(z_n \in 1{:}K\)</span> only takes on finitely many values. In detail, <span class="math display">\[
+p\left(z_n = k \mid y_n, \mu, \sigma, \lambda \right) =
+\frac{p\left(y_n \mid z_n = k, \mu, \sigma \right) \cdot p\left(z_n = k \mid \lambda \right)}
+     {\sum_{k' = 1}^K p\left(y_n \mid z_n = k', \mu, \sigma \right)
+                    \cdot p\left(z_n = k' \mid \lambda \right)}.
+\]</span></p>
+<p>On the log scale, the normalized probability is computed as <span class="math display">\[\begin{align*}
+\log\,&amp;\Pr\!\left[z_n = k \mid y_n, \mu, \sigma, \lambda\right] \\
+&amp;= \log p\left(y_n \mid z_n = k, \mu, \sigma\right) + \log \Pr\!\left[z_n = k \mid \lambda\right] \\
+&amp;\quad  - \texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}_{k' = 1}^K \big(\log p\left(y_n \mid z_n = k', \mu, \sigma\right)
+                                                 + \log p\left(z_n = k' \mid \lambda\right)\big).
+\end{align*}\]</span> This can be coded up directly in Stan; the change-point model in the <a href="../stan-users-guide/latent-discrete.html#change-point.section">change point section</a> provides an example.</p>
+</section>
+<section id="estimating-parameters-of-a-mixture" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimating-parameters-of-a-mixture">Estimating parameters of a mixture</h3>
+<p>Given the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;          <span class="co">// number of mixture components</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;          <span class="co">// number of data points</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;         <span class="co">// observations</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] theta;          <span class="co">// mixing proportions</span></span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">ordered</span>[K] mu;             <span class="co">// locations of mixture components</span></span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] sigma;  <span class="co">// scales of mixture components</span></span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] log_theta = log(theta);  <span class="co">// cache log calculation</span></span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[K] lps = log_theta;</span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a>      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_sum_exp(lps);</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The model involves <code>K</code> mixture components and <code>N</code> data points. The mixing proportion parameter <code>theta</code> is declared to be a unit <span class="math inline">\(K\)</span>-simplex, whereas the component location parameter <code>mu</code> and scale parameter <code>sigma</code> are both defined to be <code>K</code>-vectors.</p>
+<p>The location parameter <code>mu</code> is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components <code>mu[k]</code> is symmetric, as it is here (each component has an independent <span class="math inline">\(\textsf{normal}(0, 10)\)</span> prior). It would even be possible to include a hierarchical prior for the components.</p>
+<p>The values in the scale array <code>sigma</code> are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.</p>
+<p>The model declares a local array variable <code>lps</code> to be size <code>K</code> and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points <code>n</code>. For each such point, the log of <span class="math inline">\(\theta_k \times
+\textsf{normal}\left(y_n \mid \mu_k,\sigma_k\right)\)</span> is calculated and added to the array <code>lps</code>. Then the log probability is incremented with the log sum of exponentials of those values.</p>
+</section>
+</section>
+<section id="vectorizing-mixtures" class="level2">
+<h2 class="anchored" data-anchor-id="vectorizing-mixtures">Vectorizing mixtures</h2>
+<p>There is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that <code>lambda</code>, <code>y[n]</code>, <code>mu[1], mu[2]</code>, and <code>sigma[1], sigma[2]</code> are all scalars and <code>lambda</code> is between 0 and 1.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log_sum_exp(log(lambda)</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>                          + normal_lpdf(y[n] | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>]),</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>                        log1m(lambda)</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>                          + normal_lpdf(y[n] | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>or equivalently</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log_mix(lambda,</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>                    normal_lpdf(y[n] | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>]),</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>                    normal_lpdf(y[n] | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]))</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  };</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This definition assumes that each observation <span class="math inline">\(y_n\)</span> may have arisen from either of the mixture components. The density is <span class="math display">\[
+p\left(y \mid \lambda, \mu, \sigma\right)
+= \prod_{n=1}^N \big(\lambda \times \textsf{normal}\left(y_n \mid \mu_1, \sigma_1 \right)
+                 + (1 - \lambda) \times \textsf{normal}\left(y_n \mid \mu_2, \sigma_2 \right)\big).
+\]</span></p>
+<p>Contrast the previous model with the following (erroneous) attempt to vectorize the model.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log_sum_exp(log(lambda)</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>                        + normal_lpdf(y | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>]),</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>                      log1m(lambda)</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>                        + normal_lpdf(y | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>or equivalently,</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log_mix(lambda,</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>                  normal_lpdf(y | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>]),</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>                  normal_lpdf(y | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This second definition implies that the entire sequence <span class="math inline">\(y_1, \dotsc, y_n\)</span> of observations comes form one component or the other, defining a different density, <span class="math display">\[
+p\left(y \mid \lambda, \mu, \sigma \right)
+= \lambda \times \prod_{n=1}^N \textsf{normal}\left(y_n \mid \mu_1, \sigma_1\right)
++ (1 - \lambda) \times \prod_{n=1}^N \textsf{normal}\left(y_n \mid \mu_2, \sigma_2\right).
+\]</span></p>
+</section>
+<section id="mixture-inference.section" class="level2">
+<h2 class="anchored" data-anchor-id="mixture-inference.section">Inferences supported by mixtures</h2>
+<p>In many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.</p>
+<p>We have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs.&nbsp;female, or Democrat vs.&nbsp;Republican).</p>
+<p>In other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.</p>
+<section id="mixtures-with-unidentifiable-components" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mixtures-with-unidentifiable-components">Mixtures with unidentifiable components</h3>
+<p>As an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters <span class="math inline">\((\mu_1, \sigma_1)\)</span> and <span class="math inline">\((\mu_2, \sigma_2)\)</span>, <span class="math display">\[\begin{align*}
+\mu_1, \mu_2       &amp;\sim \textsf{normal}(0, 10) \\
+\sigma_1, \sigma_2 &amp;\sim \textsf{halfnormal}(0, 10) \\
+\end{align*}\]</span></p>
+<p>The prior on the mixture ratio is uniform, <span class="math display">\[
+\lambda \sim \textsf{uniform}(0, 1),
+\]</span> so that with the likelihood <span class="math display">\[
+p\left(y_n \mid \mu, \sigma\right)
+= \lambda \times \textsf{normal}\left(y_n \mid \mu_1, \sigma_1\right)
++ (1 - \lambda) \times \textsf{normal}\left(y_n \mid \mu_2, \sigma_2\right),
+\]</span> the joint distribution <span class="math inline">\(p(y, \mu, \sigma, \lambda)\)</span> is exchangeable in the parameters <span class="math inline">\((\mu_1, \sigma_1)\)</span> and <span class="math inline">\((\mu_2, \sigma_2)\)</span> with <span class="math inline">\(\lambda\)</span> flipping to <span class="math inline">\(1 - \lambda\)</span>.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+</section>
+<section id="inference-under-label-switching" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="inference-under-label-switching">Inference under label switching</h3>
+<p>In cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.</p>
+<section id="posterior-predictive-distribution" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="posterior-predictive-distribution">Posterior predictive distribution</h4>
+<p>Posterior predictive distribution for a new observation <span class="math inline">\(\tilde{y}\)</span> given the complete parameter vector <span class="math inline">\(\theta\)</span> will be <span class="math display">\[
+p(\tilde{y} \mid y)
+=
+\int_{\theta}
+p(\tilde{y} \mid \theta)
+\, p(\theta \mid y)
+\, \textsf{d}\theta.
+\]</span></p>
+<p>The normal mixture example from the previous section, with <span class="math inline">\(\theta =
+(\mu, \sigma, \lambda)\)</span>, shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing <span class="math inline">\(p(\tilde{y} \mid y)\)</span>, which is more efficient statistically in terms of effective sample size, or simulating draws of <span class="math inline">\(\tilde{y}\)</span>, which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] y_tilde;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] log_p_y_tilde;</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_tilde) {</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>    log_p_y_tilde[n]</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>      = log_mix(lambda,</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>                normal_lpdf(y_tilde[n] | mu[<span class="dv">1</span>], sigma[<span class="dv">1</span>])</span>
+<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>                normal_lpdf(y_tilde[n] | mu[<span class="dv">2</span>], sigma[<span class="dv">2</span>]));</span>
+<span id="cb9-13"><a href="#cb9-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-14"><a href="#cb9-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply <code>log_sum_exp</code> of the posterior draws of <code>log_p_y_tilde</code>. The average log predictive density is then given by subtracting <code>log(N_new)</code>.</p>
+</section>
+<section id="clustering-and-similarity" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="clustering-and-similarity">Clustering and similarity</h4>
+<p>Often a mixture model will be applied to a clustering problem and there might be two data items <span class="math inline">\(y_i\)</span> and <span class="math inline">\(y_j\)</span> for which there is a question of whether they arose from the same mixture component. If we take <span class="math inline">\(z_i\)</span> and <span class="math inline">\(z_j\)</span> to be the component responsibility discrete variables, then the quantity of interest is <span class="math inline">\(z_i = z_j\)</span>, which can be summarized as an event probability <span class="math display">\[
+\Pr[z_i = z_j \mid y]
+=
+\int_{\theta}
+\frac{\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \mid \theta)}
+     {\sum_{k=0}^1 \sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \mid \theta)}
+\,
+p(\theta \mid y)
+\,
+\textsf{d}\theta.
+\]</span></p>
+<p>As with other event probabilities, this can be calculated in the generated quantities block either by sampling <span class="math inline">\(z_i\)</span> and <span class="math inline">\(z_j\)</span> and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.</p>
+</section>
+</section>
+</section>
+<section id="zero-inflated.section" class="level2">
+<h2 class="anchored" data-anchor-id="zero-inflated.section">Zero-inflated and hurdle models</h2>
+<p>Zero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by <span class="citation" data-cites="Lambert:1992">Lambert (<a href="#ref-Lambert:1992" role="doc-biblioref">1992</a>)</span>, add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.</p>
+<p>Zero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.</p>
+<section id="zero-inflation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="zero-inflation">Zero inflation</h3>
+<p>Consider the following example for zero-inflated Poisson distributions. There is a probability <span class="math inline">\(\theta\)</span> of observing a zero, and a probability <span class="math inline">\(1 - \theta\)</span> of observing a count with a <span class="math inline">\(\textsf{Poisson}(\lambda)\)</span> distribution (now <span class="math inline">\(\theta\)</span> is being used for mixing proportions because <span class="math inline">\(\lambda\)</span> is the traditional notation for a Poisson mean parameter). Given the probability <span class="math inline">\(\theta\)</span> and the intensity <span class="math inline">\(\lambda\)</span>, the distribution for <span class="math inline">\(y_n\)</span> can be written as <span class="math display">\[\begin{align*}
+y_n &amp; = 0 &amp; \quad\text{with probability } \theta, \text{ and}\\
+y_n &amp; \sim \textsf{Poisson}(y_n \mid \lambda) &amp; \quad\text{with probability } 1-\theta.
+\end{align*}\]</span></p>
+<p>Stan does not support conditional distribution statements (with <code>~</code>) conditional on some parameter, and we need to consider the corresponding likelihood <span class="math display">\[
+p(y_n \mid \theta,\lambda)
+=
+\begin{cases}
+\theta + (1 - \theta) \times \textsf{Poisson}(0 \mid \lambda) &amp; \quad\text{if } y_n = 0, \text{ and}\\
+(1-\theta) \times \textsf{Poisson}(y_n \mid \lambda) &amp;\quad\text{if } y_n &gt; 0.
+\end{cases}
+\]</span> The log likelihood can be coded directly in Stan (with <code>target +=</code>) as follows.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (y[n] == <span class="dv">0</span>) {</span>
+<span id="cb10-12"><a href="#cb10-12" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> log_sum_exp(log(theta),</span>
+<span id="cb10-13"><a href="#cb10-13" aria-hidden="true" tabindex="-1"></a>                            log1m(theta)</span>
+<span id="cb10-14"><a href="#cb10-14" aria-hidden="true" tabindex="-1"></a>                              + poisson_lpmf(y[n] | lambda));</span>
+<span id="cb10-15"><a href="#cb10-15" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> {</span>
+<span id="cb10-16"><a href="#cb10-16" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> log1m(theta)</span>
+<span id="cb10-17"><a href="#cb10-17" aria-hidden="true" tabindex="-1"></a>                  + poisson_lpmf(y[n] | lambda);</span>
+<span id="cb10-18"><a href="#cb10-18" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb10-19"><a href="#cb10-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb10-20"><a href="#cb10-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>log1m(theta)</code> computes <code>log(1-theta)</code>, but is more computationally stable. The <code>log_sum_exp(lp1,lp2)</code> function adds the log probabilities on the linear scale; it is defined to be equal to <code>log(exp(lp1) + exp(lp2))</code>, but is more computationally stable and faster.</p>
+<section id="optimizing-the-zero-inflated-poisson-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="optimizing-the-zero-inflated-poisson-model">Optimizing the zero-inflated Poisson model</h4>
+<p>The code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also <code>poisson_lpmf(0 | lambda)</code> every time the first condition body executes. The use of the redundant terms is conditioned on <code>y</code>, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.</p>
+<p>The number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare <code>y_nonzero</code>, so it must be computed in a function.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> num_zeros(<span class="dt">array</span>[] <span class="dt">int</span> y) {</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> sum = <span class="dv">0</span>;</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:size(y)) {</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>      sum += (y[n] == <span class="dv">0</span>);</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> sum;</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_zero = num_zeros(y);</span>
+<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N - N_zero] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; y_nonzero;</span>
+<span id="cb11-14"><a href="#cb11-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N_nonzero = <span class="dv">0</span>;</span>
+<span id="cb11-15"><a href="#cb11-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb11-16"><a href="#cb11-16" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (y[n] == <span class="dv">0</span>) <span class="cf">continue</span>;</span>
+<span id="cb11-17"><a href="#cb11-17" aria-hidden="true" tabindex="-1"></a>    N_nonzero += <span class="dv">1</span>;</span>
+<span id="cb11-18"><a href="#cb11-18" aria-hidden="true" tabindex="-1"></a>    y_nonzero[N_nonzero] = y[n];</span>
+<span id="cb11-19"><a href="#cb11-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-20"><a href="#cb11-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb11-21"><a href="#cb11-21" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb11-22"><a href="#cb11-22" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb11-23"><a href="#cb11-23" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-24"><a href="#cb11-24" aria-hidden="true" tabindex="-1"></a>   target</span>
+<span id="cb11-25"><a href="#cb11-25" aria-hidden="true" tabindex="-1"></a>     += N_zero</span>
+<span id="cb11-26"><a href="#cb11-26" aria-hidden="true" tabindex="-1"></a>          * log_sum_exp(log(theta),</span>
+<span id="cb11-27"><a href="#cb11-27" aria-hidden="true" tabindex="-1"></a>                        log1m(theta)</span>
+<span id="cb11-28"><a href="#cb11-28" aria-hidden="true" tabindex="-1"></a>                          + poisson_lpmf(<span class="dv">0</span> | lambda));</span>
+<span id="cb11-29"><a href="#cb11-29" aria-hidden="true" tabindex="-1"></a>   <span class="kw">target +=</span> N_nonzero * log1m(theta);</span>
+<span id="cb11-30"><a href="#cb11-30" aria-hidden="true" tabindex="-1"></a>   <span class="kw">target +=</span> poisson_lpmf(y_nonzero | lambda);</span>
+<span id="cb11-31"><a href="#cb11-31" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-32"><a href="#cb11-32" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if <code>y_nonzero</code> is empty, <code>N_nonzero</code> will be zero, and the last two target increment terms will add zeros.</p>
+</section>
+</section>
+<section id="hurdle-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="hurdle-models">Hurdle models</h3>
+<p>The hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability <span class="math inline">\(\theta\)</span> and the intensity <span class="math inline">\(\lambda\)</span>, the distribution for <span class="math inline">\(y_n\)</span> can be written as [ <span class="math display">\[\begin{align*}
+y_n &amp; = 0 \quad\text{with probability } \theta, \text{ and}\\
+y_n &amp; \sim \textsf{Poisson}_{x\neq 0}(y_n \mid \lambda) \quad\text{with probability } 1-\theta,
+\end{align*}\]</span> ] Where <span class="math inline">\(\textsf{Poisson}_{x\neq 0}\)</span> is a truncated Poisson distribution, truncated at <span class="math inline">\(0\)</span>.</p>
+<p>The corresponding likelihood function for the hurdle model is defined by <span class="math display">\[
+p(y\mid\theta,\lambda)
+=
+\begin{cases}
+\theta &amp;\quad\text{if } y = 0, \text{ and}\\
+(1 - \theta)
+   \frac{\displaystyle \textsf{Poisson}(y \mid \lambda)}
+        {\displaystyle  1 - \textsf{PoissonCDF}(0 \mid \lambda)}
+&amp;\quad\text{if } y &gt; 0,
+\end{cases}
+\]</span> where <span class="math inline">\(\textsf{PoissonCDF}\)</span> is the cumulative distribution function for the Poisson distribution and and <span class="math inline">\(1 - \textsf{PoissonCDF}(0 \mid \lambda)\)</span> is the relative normalization term for the truncated Poisson (truncated at <span class="math inline">\(0\)</span>).</p>
+<p>The hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (y[n] == <span class="dv">0</span>) {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log(theta);</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log1m(theta) + poisson_lpmf(y[n] | lambda)</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>            - poisson_lccdf(<span class="dv">0</span> | lambda));</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Julian King pointed out that because <span class="math display">\[\begin{align*}
+\log \left( 1 - \textsf{PoissonCDF}(0 \mid \lambda) \right)
+  &amp;= \log \left( 1 - \textsf{Poisson}(0 \mid \lambda) \right) \\
+  &amp;= \log(1 - \exp(-\lambda))
+\end{align*}\]</span> the CCDF in the else clause can be replaced with a simpler expression.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> log1m(theta) + poisson_lpmf(y[n] | lambda)</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>          - log1m_exp(-lambda));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The resulting code is about 15% faster than the code with the CCDF.</p>
+<p>This is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size <span class="math inline">\(N=200\)</span> and parameters <span class="math inline">\(\theta=0.3\)</span> and <span class="math inline">\(\lambda = 8\)</span>, the speedup is a factor of 10; it will be lower for smaller <span class="math inline">\(N\)</span> and greater for larger <span class="math inline">\(N\)</span>; it will also be greater for larger <span class="math inline">\(\theta\)</span>.</p>
+<p>To achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> num_zero(<span class="dt">array</span>[] <span class="dt">int</span> y) {</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> nz = <span class="dv">0</span>;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:size(y)) {</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (y[n] == <span class="dv">0</span>) {</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>        nz += <span class="dv">1</span>;</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> nz;</span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then a transformed data block can be used to store the sufficient statistics,</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=N&gt; N0 = num_zero(y);</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=N&gt; Ngt0 = N - N0;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N - num_zero(y)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; y_nz;</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> pos = <span class="dv">1</span>;</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (y[n] != <span class="dv">0</span>) {</span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>        y_nz[pos] = y[n];</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a>        pos += <span class="dv">1</span>;</span>
+<span id="cb15-11"><a href="#cb15-11" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb15-12"><a href="#cb15-12" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-13"><a href="#cb15-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-14"><a href="#cb15-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The model block is then reduced to three statements.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  N0 ~ binomial(N, theta);</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>  y_nz ~ poisson(lambda);</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -Ngt0 * log1m_exp(-lambda);</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant <code>Ngt0</code>; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.</p>
+</section>
+</section>
+<section id="priors-and-effective-data-size-in-mixture-models" class="level2">
+<h2 class="anchored" data-anchor-id="priors-and-effective-data-size-in-mixture-models">Priors and effective data size in mixture models</h2>
+<p>Suppose we have a two-component mixture model with mixing rate <span class="math inline">\(\lambda \in (0, 1)\)</span>. Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are <span class="math inline">\(N\)</span> observations, the mixture components will be estimated with effective data sizes of <span class="math inline">\(\theta \, N\)</span> and <span class="math inline">\((1 - \theta)
+\, N\)</span> for the two components for some <span class="math inline">\(\theta \in (0, 1)\)</span>. The effective weighting size is determined by posterior responsibility, not simply by the mixing rate <span class="math inline">\(\lambda\)</span>.</p>
+<section id="comparison-to-model-averaging" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="comparison-to-model-averaging">Comparison to model averaging</h3>
+<p>In contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data <span class="math inline">\(y\)</span>.</p>
+<p>If different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see <span class="citation" data-cites="HoetingEtAl:1999">Hoeting et al. (<a href="#ref-HoetingEtAl:1999" role="doc-biblioref">1999</a>)</span> and <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> for theoretical and computational details).</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-HoetingEtAl:1999" class="csl-entry" role="listitem">
+Hoeting, Jennifer A., David Madigan, Adrian E Raftery, and Chris T. Volinsky. 1999. <span>“Bayesian Model Averaging: A Tutorial.”</span> <em>Statistical Science</em> 14 (4): 382–417.
+</div>
+<div id="ref-Lambert:1992" class="csl-entry" role="listitem">
+Lambert, Diane. 1992. <span>“Zero-Inflated <span>P</span>oisson Regression, with an Application to Defects in Manufacturing.”</span> <em>Technometrics</em> 34 (1).
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Imposing a constraint such as <span class="math inline">\(\theta &lt; 0.5\)</span> will resolve the symmetry, but fundamentally changes the model and its posterior inferences.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/truncation-censoring.html" class="pagination-link" aria-label="Truncated or Censored Data">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Truncated or Censored Data</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/measurement-error.html" class="pagination-link" aria-label="Measurement Error and Meta-Analysis">
+        <span class="nav-page-text">Measurement Error and Meta-Analysis</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/finite-mixtures.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/floating-point.html b/docs/2_39/stan-users-guide/floating-point.html
new file mode 100644
index 000000000..ceb78bfdf
--- /dev/null
+++ b/docs/2_39/stan-users-guide/floating-point.html
@@ -0,0 +1,1387 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Floating Point Arithmetic</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/matrices-arrays.html" rel="next">
+<link href="../stan-users-guide/wiener_diffusion_model.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Floating Point Arithmetic">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Floating Point Arithmetic</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#floating-point-arithmetic" id="toc-floating-point-arithmetic" class="nav-link active" data-scroll-target="#floating-point-arithmetic">Floating Point Arithmetic</a>
+  <ul class="collapse">
+  <li><a href="#floating-point-representations" id="toc-floating-point-representations" class="nav-link" data-scroll-target="#floating-point-representations">Floating-point representations</a>
+  <ul class="collapse">
+  <li><a href="#finite-values" id="toc-finite-values" class="nav-link" data-scroll-target="#finite-values">Finite values</a></li>
+  <li><a href="#normality" id="toc-normality" class="nav-link" data-scroll-target="#normality">Normality</a></li>
+  <li><a href="#ranges-and-extreme-values" id="toc-ranges-and-extreme-values" class="nav-link" data-scroll-target="#ranges-and-extreme-values">Ranges and extreme values</a></li>
+  <li><a href="#signed-zero" id="toc-signed-zero" class="nav-link" data-scroll-target="#signed-zero">Signed zero</a></li>
+  <li><a href="#not-a-number-values" id="toc-not-a-number-values" class="nav-link" data-scroll-target="#not-a-number-values">Not-a-number values</a></li>
+  <li><a href="#positive-and-negative-infinity" id="toc-positive-and-negative-infinity" class="nav-link" data-scroll-target="#positive-and-negative-infinity">Positive and negative infinity</a></li>
+  </ul></li>
+  <li><a href="#literals-decimal-and-scientific-notation" id="toc-literals-decimal-and-scientific-notation" class="nav-link" data-scroll-target="#literals-decimal-and-scientific-notation">Literals: decimal and scientific notation</a></li>
+  <li><a href="#arithmetic-precision" id="toc-arithmetic-precision" class="nav-link" data-scroll-target="#arithmetic-precision">Arithmetic precision</a>
+  <ul class="collapse">
+  <li><a href="#rounding-and-probabilities" id="toc-rounding-and-probabilities" class="nav-link" data-scroll-target="#rounding-and-probabilities">Rounding and probabilities</a></li>
+  <li><a href="#machine-precision-and-the-asymmetry-of-0-and-1" id="toc-machine-precision-and-the-asymmetry-of-0-and-1" class="nav-link" data-scroll-target="#machine-precision-and-the-asymmetry-of-0-and-1">Machine precision and the asymmetry of 0 and 1</a></li>
+  <li><a href="#complementary-and-epsilon-functions" id="toc-complementary-and-epsilon-functions" class="nav-link" data-scroll-target="#complementary-and-epsilon-functions">Complementary and epsilon functions</a></li>
+  <li><a href="#catastrophic-cancellation" id="toc-catastrophic-cancellation" class="nav-link" data-scroll-target="#catastrophic-cancellation">Catastrophic cancellation</a></li>
+  <li><a href="#overflow" id="toc-overflow" class="nav-link" data-scroll-target="#overflow">Overflow</a></li>
+  <li><a href="#underflow-and-the-log-scale" id="toc-underflow-and-the-log-scale" class="nav-link" data-scroll-target="#underflow-and-the-log-scale">Underflow and the log scale</a></li>
+  </ul></li>
+  <li><a href="#log-sum-of-exponentials" id="toc-log-sum-of-exponentials" class="nav-link" data-scroll-target="#log-sum-of-exponentials">Log sum of exponentials</a>
+  <ul class="collapse">
+  <li><a href="#log-sum-exp-function" id="toc-log-sum-exp-function" class="nav-link" data-scroll-target="#log-sum-exp-function">Log-sum-exp function</a></li>
+  <li><a href="#applying-log-sum-exp-to-a-sequence" id="toc-applying-log-sum-exp-to-a-sequence" class="nav-link" data-scroll-target="#applying-log-sum-exp-to-a-sequence">Applying log-sum-exp to a sequence</a></li>
+  <li><a href="#calculating-means-with-log-sum-exp" id="toc-calculating-means-with-log-sum-exp" class="nav-link" data-scroll-target="#calculating-means-with-log-sum-exp">Calculating means with log-sum-exp</a></li>
+  </ul></li>
+  <li><a href="#comparing-floating-point-numbers" id="toc-comparing-floating-point-numbers" class="nav-link" data-scroll-target="#comparing-floating-point-numbers">Comparing floating-point numbers</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/floating-point.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Floating Point Arithmetic</a></li></ol></nav></header>
+
+
+
+
+<section id="floating-point-arithmetic" class="level1">
+<h1>Floating Point Arithmetic</h1>
+<p>Computers approximate real values in <span class="math inline">\(\mathbb{R}\)</span> using a fixed number of bits. This chapter explains how this is done and why it is important for writing robust Stan (and other numerical) programs. The subfield of computer science devoted to studying how real arithmetic works on computers is called <em>numerical analysis</em>.</p>
+<section id="floating-point-representations" class="level2">
+<h2 class="anchored" data-anchor-id="floating-point-representations">Floating-point representations</h2>
+<p>Stan’s arithmetic is implemented using double-precision arithmetic. The behavior of most<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> modern computers follows the floating-point arithmetic, <em>IEEE Standard for Floating-Point Arithmetic</em> (IEEE 754).</p>
+<section id="finite-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="finite-values">Finite values</h3>
+<p>The double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:</p>
+<ul>
+<li><p><em>significand</em> (53 bits): base value representing significant digits</p></li>
+<li><p><em>exponent</em> (11 bits): power of two multiplied by the base</p></li>
+</ul>
+<p>The <em>value</em> of a finite floating point number is</p>
+<p><span class="math display">\[
+v = (-1)^s \times c \, 2^q
+\]</span></p>
+</section>
+<section id="normality" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="normality">Normality</h3>
+<p>A <em>normal</em> floating-point value does not use any leading zeros in its significand; <em>subnormal</em> numbers may use leading zeros. Not all I/O systems support subnormal numbers.</p>
+</section>
+<section id="ranges-and-extreme-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ranges-and-extreme-values">Ranges and extreme values</h3>
+<p>There are some reserved exponent values so that legal exponent values range between<span class="math inline">\(-(2^{10}) + 2 = -1022\)</span> and <span class="math inline">\(2^{10} - 1 = 1023\)</span>. Legal significand values are between <span class="math inline">\(-2^{52}\)</span> and <span class="math inline">\(2^{52} - 1\)</span>. Floating point allows the representation of both really big and really small values. Some extreme values are</p>
+<ul>
+<li><p><em>largest normal finite number</em>: <span class="math inline">\(\approx 1.8 \times 10^{308}\)</span></p></li>
+<li><p><em>largest subnormal finite number</em>: <span class="math inline">\(\approx 2.2 \times 10^{308}\)</span></p></li>
+<li><p><em>smallest positive normal number</em>: <span class="math inline">\(\approx 2.2 \times 10^{-308}\)</span></p></li>
+<li><p><em>smallest positive subnormal number</em>: <span class="math inline">\(\approx 4.9 \times 10^{-324}\)</span></p></li>
+</ul>
+</section>
+<section id="signed-zero" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="signed-zero">Signed zero</h3>
+<p>Because of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., <code>0 == -0</code> evaluates to true).</p>
+</section>
+<section id="not-a-number-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="not-a-number-values">Not-a-number values</h3>
+<p>A specially chosen bit pattern is used for the <em>not-a-number</em> value (often written as <code>NaN</code> in programming language output, including Stan’s).</p>
+<p>Stan provides a value function <code>not_a_number()</code> that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.</p>
+<p>Stan also provides a test function <code>is_nan(x)</code> that returns 1 if <code>x</code> is not-a-number and 0 otherwise.</p>
+<p>Not-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as <code>log</code> and <code>exp</code> have the same behavior, propagating not-a-number values.</p>
+<p>Most of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.</p>
+<p>Comparisons with not-a-number always return false, up to and including comparison with itself. That is, <code>not_a_number() == not_a_number()</code> somewhat confusingly returns false. That is why there is a built-in <code>is_nan()</code> function in Stan (and in C++). The only exception is negation, which remains coherent. This means <code>not_a_number() != not_a_number()</code> returns true.</p>
+<p>Undefined operations often return not-a-number values. For example, <code>sqrt(-1)</code> will evaluate to not-a-number.</p>
+</section>
+<section id="positive-and-negative-infinity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="positive-and-negative-infinity">Positive and negative infinity</h3>
+<p>There are also two special values representing positive infinity (<span class="math inline">\(\infty)\)</span> and negative infinity (<span class="math inline">\(-\infty\)</span>). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, <code>log(0)</code> evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that <code>0 == exp(log(0))</code>. Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.</p>
+<p>There are value functions <code>positive_infinity()</code> and <code>negative_infinity()</code> as well as a test function <code>is_inf()</code>.</p>
+<p>Positive and negative infinity have the expected comparison behavior, so that <code>negative_infinity() &lt; 0</code> evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.</p>
+<p>Positive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.</p>
+</section>
+</section>
+<section id="literals-decimal-and-scientific-notation" class="level2">
+<h2 class="anchored" data-anchor-id="literals-decimal-and-scientific-notation">Literals: decimal and scientific notation</h2>
+<p>In programming languages such as Stan, numbers may be represented in standard <em>decimal</em> (base 10) notation. For example, <code>2.39</code> or <code>-1567846.276452</code>. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using <em>scientific notation</em>, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, <code>36.29e-3</code> represents the number <span class="math inline">\(36.29 \times
+10^{-3}\)</span>, which is the same number as is represented by <code>0.03629</code>.</p>
+</section>
+<section id="arithmetic-precision" class="level2">
+<h2 class="anchored" data-anchor-id="arithmetic-precision">Arithmetic precision</h2>
+<p>The choice of significand provides <span class="math inline">\(\log_{10} 2^{53} \approx 15.95\)</span> decimal (base 10) digits of <em>arithmetic precision</em>. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.</p>
+<section id="rounding-and-probabilities" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="rounding-and-probabilities">Rounding and probabilities</h3>
+<p>In practice, the finite amount of arithmetic precision leads to <em>rounding</em>, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,</p>
+<pre><code>1 + 1e-20 == 1</code></pre>
+<p>The closest floating point number to <span class="math inline">\(1 + 10^{-20}\)</span> turns out to be <span class="math inline">\(1\)</span> itself. By contrast,</p>
+<pre><code>0 + 1e-20 == 1e-20</code></pre>
+<p>This highlights the fact that precision depends on scale. Even though <code>1 + 1e-20 == 1</code>, we have <code>1e-20 + 1e-20 == 2e-20</code>, as expected.</p>
+<p>Rounding also manifests itself in a lack of <em>transitivity</em>. In particular, it does <em>not</em> usually hold for three floating point numbers <span class="math inline">\(a, b, c\)</span> that <span class="math inline">\((a + b) + c = a + (b + c)\)</span>.</p>
+<p>In statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix <span class="math inline">\(L\)</span> with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix <span class="math inline">\(L \, L^{\top}\)</span>. In practice, rounding and loss of precision may render the result <span class="math inline">\(L \, L^{\top}\)</span> neither symmetric nor positive definite.</p>
+<p>In practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding <span class="math inline">\(L \, L^{\top}\)</span> with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.</p>
+</section>
+<section id="machine-precision-and-the-asymmetry-of-0-and-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="machine-precision-and-the-asymmetry-of-0-and-1">Machine precision and the asymmetry of 0 and 1</h3>
+<p>The smallest number greater than zero is roughly <span class="math inline">\(0 + 10^{-323}\)</span>. The largest number less than one is roughly <span class="math inline">\(1 - 10^{-15.95}\)</span>. The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of <span class="math inline">\(0.9999999999999999\)</span>.</p>
+<p>For this reason, the <em>machine precision</em> is said to be roughly <span class="math inline">\(10^{-15.95}\)</span>. This constant is available as <code>machine_precision()</code> in Stan.</p>
+</section>
+<section id="complementary-and-epsilon-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="complementary-and-epsilon-functions">Complementary and epsilon functions</h3>
+<p>Special operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation <code>log(1 + x)</code> for positive <code>x</code>. When <code>x</code> is small (less than <span class="math inline">\(10^{-16}\)</span> for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing <span class="math inline">\(f(x) = \log (1 + x)\)</span>. In Stan (as in C++), this operation is written as <code>log1p(x)</code>. Because <code>x</code> itself may be close to zero, the function <code>log1p(x)</code> can take the logarithm of values very close to one, the results of which are close to zero.</p>
+<p>Similarly, the complementary cumulative distribution functions (CCDF), defined by <span class="math inline">\(F^{\complement}_Y(y) = 1 - F_Y(y)\)</span>, where <span class="math inline">\(F_Y\)</span> is the cumulative distribution function (CDF) for the random variable <span class="math inline">\(Y\)</span>. This allows values very close to one to be represented in complementary form.</p>
+</section>
+<section id="catastrophic-cancellation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="catastrophic-cancellation">Catastrophic cancellation</h3>
+<p>Another downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider <span class="math display">\[\begin{align*}
+  1&amp;.23456789012345 \\
+- 1&amp;.23456789012344 \\
+= 0&amp;.00000000000001
+\end{align*}\]</span> We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.</p>
+<p>Catastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses <em>Welford’s algorithm</em> for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.</p>
+</section>
+<section id="overflow" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="overflow">Overflow</h3>
+<p>Even though <code>1e200</code> may be represented as a double precision floating point value, there is no finite value large enough to represent <code>1e200 * 1e200</code>. The result of <code>1e200 * 1e200</code> is said to <em>overflow</em>. The IEEE 754 standard requires the result to be positive infinity.</p>
+<p>Overflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.</p>
+</section>
+<section id="underflow-and-the-log-scale" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="underflow-and-the-log-scale">Underflow and the log scale</h3>
+<p>When there is no number small enough to represent a result, it is said to <em>underflow</em>. For instance, <code>1e-200</code> may be represented, but <code>1e-200 * 1e-200</code> underflows so that the result is zero.</p>
+<p>Underflow is a ubiquitous problem in likelihood calculations, For example, if <span class="math inline">\(p(y_n \mid \theta) &lt; 0.1\)</span>, then <span class="math display">\[
+p(y \mid \theta) = \prod_{n=1}^N p(y_n \mid \theta)
+\]</span> will underflow as soon as <span class="math inline">\(N &gt; 350\)</span> or so.</p>
+<p>To deal with underflow, work on the log scale. Even though <span class="math inline">\(p(y \mid
+\theta)\)</span> can’t be represented, there is no problem representing <span class="math display">\[
+\begin{array}{rcl}
+\log p(y \mid \theta)
+&amp; = &amp; \log \prod_{n=1}^N p(y_n \mid \theta)
+\\[4pt]
+&amp; = &amp; \sum_{n = 1}^N \log p(y_n \mid \theta)
+\end{array}
+\]</span></p>
+<p>This is why all of Stan’s probability functions operate on the log scale.</p>
+</section>
+</section>
+<section id="log-sum-of-exponentials" class="level2">
+<h2 class="anchored" data-anchor-id="log-sum-of-exponentials">Log sum of exponentials</h2>
+<p>Working on the log scale, multiplication is converted to addition, <span class="math display">\[
+\log (a \cdot b) = \log a + \log b.
+\]</span> Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given <span class="math inline">\(\log a\)</span> and <span class="math inline">\(\log b\)</span>, how do we get <span class="math inline">\(\log (a + b)\)</span>? Working out the algebra, <span class="math display">\[
+\log (a + b)
+=
+\log (\exp(\log a) + \exp(\log b)).
+\]</span></p>
+<section id="log-sum-exp-function" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="log-sum-exp-function">Log-sum-exp function</h3>
+<p>The nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, <span class="math display">\[
+\textrm{log-sum-exp}(u, v)
+=
+\log (\exp(u) + \exp(v)).
+\]</span> so that <span class="math display">\[
+\log (a + b)
+=
+\textrm{log-sum-exp}(\log a, \log b).
+\]</span></p>
+<p>Although it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, <span class="math display">\[
+\textrm{log-sum-exp}(u, v)
+=
+\max(u, v) + \log\big( \exp(u - \max(u, v)) + \exp(v - \max(u, v)) \big).
+\]</span></p>
+<p>Because the terms inside the exponentiations are <span class="math inline">\(u - \max(u, v)\)</span> and <span class="math inline">\(v - \max(u, v)\)</span>, one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that <span class="math inline">\(u \geq v\)</span>, so that <span class="math display">\[
+\textrm{log-sum-exp}(u, v) = u + \log\big(1 + \exp(v - u)\big).
+\]</span></p>
+<p>Although the inner term may itself be evaluated using the built-in function <code>log1p</code>, there is only limited gain because <span class="math inline">\(\exp(v - u)\)</span> is only near zero when <span class="math inline">\(u\)</span> is much larger than <span class="math inline">\(v\)</span>, meaning the final result is likely to round to <span class="math inline">\(u\)</span> anyway.</p>
+<p>To conclude, when evaluating <span class="math inline">\(\log (a + b)\)</span> given <span class="math inline">\(\log a\)</span> and <span class="math inline">\(\log
+b\)</span>, and assuming <span class="math inline">\(\log a &gt; \log b\)</span>, return</p>
+<p><span class="math display">\[
+\log (a + b) =
+\log a + \textrm{log1p}\big(\exp(\log b - \log a)\big).
+\]</span></p>
+</section>
+<section id="applying-log-sum-exp-to-a-sequence" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="applying-log-sum-exp-to-a-sequence">Applying log-sum-exp to a sequence</h3>
+<p>The log sum of exponentials function may be generalized to sequences in the obvious way, so that if <span class="math inline">\(v = v_1, \ldots, v_N\)</span>, then <span class="math display">\[\begin{eqnarray*}
+\textrm{log-sum-exp}(v)
+&amp; = &amp; \log \sum_{n = 1}^N \exp(v_n)
+\\[4pt]
+&amp; = &amp; \max(v) + \log \sum_{n = 1}^N \exp(v_n - \max(v)).
+\end{eqnarray*}\]</span> The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate <span class="math inline">\(\log (u_1 + \cdots + u_N)\)</span> given only <span class="math inline">\(\log u_n\)</span>.</p>
+</section>
+<section id="calculating-means-with-log-sum-exp" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="calculating-means-with-log-sum-exp">Calculating means with log-sum-exp</h3>
+<p>An immediate application is to computing the mean of a vector <span class="math inline">\(u\)</span> entirely on the log scale. That is, given <span class="math inline">\(\log u\)</span> and returning <span class="math inline">\(\log \textrm{mean}(u)\)</span>. <span class="math display">\[\begin{eqnarray*}
+\log \left( \frac{1}{N} \sum_{n = 1}^N u_n \right)
+&amp; = &amp; \log \frac{1}{N} + \log \sum_{n = 1}^N \exp(\log u_n)
+\\[4pt]
+&amp; = &amp; -\log N + \textrm{log-sum-exp}(\log u).
+\end{eqnarray*}\]</span> where <span class="math inline">\(\log u = (\log u_1, \ldots, \log u_N)\)</span> is understood elementwise.</p>
+</section>
+</section>
+<section id="comparing-floating-point-numbers" class="level2">
+<h2 class="anchored" data-anchor-id="comparing-floating-point-numbers">Comparing floating-point numbers</h2>
+<p>Because floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing <code>x == y</code>, an approximate test may be used given an absolute or relative tolerance.</p>
+<p>Given a positive <em>absolute tolerance</em> of <code>epsilon</code>, <code>x</code> can be compared to <code>y</code> using the conditional</p>
+<pre><code>abs(x - y) &lt;= epsilon.</code></pre>
+<p>Absolute tolerances work when the scale of <code>x</code> and <code>y</code> and the relevant comparison is known.</p>
+<p>Given a positive <em>relative tolerance</em> of <code>epsilon</code>, a typical comparison is</p>
+<pre><code>2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.</code></pre>
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The notable exception is Intel’s optimizing compilers under certain optimization settings.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/wiener_diffusion_model.html" class="pagination-link" aria-label="Wiener diffusion model">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Wiener diffusion model</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/matrices-arrays.html" class="pagination-link" aria-label="Matrices, Vectors, Arrays, and Tuples">
+        <span class="nav-page-text">Matrices, Vectors, Arrays, and Tuples</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/floating-point.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/for-bugs-users.html b/docs/2_39/stan-users-guide/for-bugs-users.html
new file mode 100644
index 000000000..52a424609
--- /dev/null
+++ b/docs/2_39/stan-users-guide/for-bugs-users.html
@@ -0,0 +1,1450 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Transitioning from BUGS</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/style-guide.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Transitioning from BUGS">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/for-bugs-users.html">Transitioning from BUGS</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#stan-for-bugs.appendix" id="toc-stan-for-bugs.appendix" class="nav-link active" data-scroll-target="#stan-for-bugs.appendix">Transitioning from BUGS</a>
+  <ul class="collapse">
+  <li><a href="#some-differences-in-how-bugs-and-stan-work" id="toc-some-differences-in-how-bugs-and-stan-work" class="nav-link" data-scroll-target="#some-differences-in-how-bugs-and-stan-work">Some differences in how BUGS and Stan work</a>
+  <ul class="collapse">
+  <li><a href="#bugs-is-interpreted-stan-is-compiled" id="toc-bugs-is-interpreted-stan-is-compiled" class="nav-link" data-scroll-target="#bugs-is-interpreted-stan-is-compiled">BUGS is interpreted, Stan is compiled</a></li>
+  <li><a href="#bugs-performs-mcmc-updating-one-scalar-parameter-at-a-time-stan-uses-hmc-which-moves-in-the-entire-space-of-all-the-parameters-at-each-step" id="toc-bugs-performs-mcmc-updating-one-scalar-parameter-at-a-time-stan-uses-hmc-which-moves-in-the-entire-space-of-all-the-parameters-at-each-step" class="nav-link" data-scroll-target="#bugs-performs-mcmc-updating-one-scalar-parameter-at-a-time-stan-uses-hmc-which-moves-in-the-entire-space-of-all-the-parameters-at-each-step">BUGS performs MCMC updating one scalar parameter at a time, Stan uses HMC which moves in the entire space of all the parameters at each step</a></li>
+  <li><a href="#differences-in-tuning-during-warmup" id="toc-differences-in-tuning-during-warmup" class="nav-link" data-scroll-target="#differences-in-tuning-during-warmup">Differences in tuning during warmup</a></li>
+  <li><a href="#the-stan-language-is-directly-executable-the-bugs-modeling-language-is-not" id="toc-the-stan-language-is-directly-executable-the-bugs-modeling-language-is-not" class="nav-link" data-scroll-target="#the-stan-language-is-directly-executable-the-bugs-modeling-language-is-not">The Stan language is directly executable, the BUGS modeling language is not</a></li>
+  <li><a href="#differences-in-statement-order" id="toc-differences-in-statement-order" class="nav-link" data-scroll-target="#differences-in-statement-order">Differences in statement order</a></li>
+  <li><a href="#stan-computes-the-gradient-of-the-log-density-bugs-computes-the-log-density-but-not-its-gradient" id="toc-stan-computes-the-gradient-of-the-log-density-bugs-computes-the-log-density-but-not-its-gradient" class="nav-link" data-scroll-target="#stan-computes-the-gradient-of-the-log-density-bugs-computes-the-log-density-but-not-its-gradient">Stan computes the gradient of the log density, BUGS computes the log density but not its gradient</a></li>
+  <li><a href="#both-bugs-and-stan-are-semi-automatic" id="toc-both-bugs-and-stan-are-semi-automatic" class="nav-link" data-scroll-target="#both-bugs-and-stan-are-semi-automatic">Both BUGS and Stan are semi-automatic</a></li>
+  <li><a href="#licensing" id="toc-licensing" class="nav-link" data-scroll-target="#licensing">Licensing</a></li>
+  <li><a href="#interfaces" id="toc-interfaces" class="nav-link" data-scroll-target="#interfaces">Interfaces</a></li>
+  <li><a href="#platforms" id="toc-platforms" class="nav-link" data-scroll-target="#platforms">Platforms</a></li>
+  </ul></li>
+  <li><a href="#some-differences-in-the-modeling-languages" id="toc-some-differences-in-the-modeling-languages" class="nav-link" data-scroll-target="#some-differences-in-the-modeling-languages">Some differences in the modeling languages</a></li>
+  <li><a href="#some-differences-in-the-statistical-models-that-are-allowed" id="toc-some-differences-in-the-statistical-models-that-are-allowed" class="nav-link" data-scroll-target="#some-differences-in-the-statistical-models-that-are-allowed">Some differences in the statistical models that are allowed</a></li>
+  <li><a href="#some-differences-when-running-from-r" id="toc-some-differences-when-running-from-r" class="nav-link" data-scroll-target="#some-differences-when-running-from-r">Some differences when running from R</a></li>
+  <li><a href="#the-stan-community" id="toc-the-stan-community" class="nav-link" data-scroll-target="#the-stan-community">The Stan community</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/for-bugs-users.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/for-bugs-users.html">Transitioning from BUGS</a></li></ol></nav></header>
+
+
+
+
+<section id="stan-for-bugs.appendix" class="level1">
+<h1>Transitioning from BUGS</h1>
+<p>From the outside, Stan and BUGS<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a> are similar—they use statistically-themed modeling languages (which are similar but with some differences; see below), they can be called from R, running some specified number of chains to some specified length, producing posterior simulations that can be assessed using standard convergence diagnostics. This is not a coincidence: in designing Stan: we wanted to keep many of the useful features of Bugs.</p>
+<section id="some-differences-in-how-bugs-and-stan-work" class="level2">
+<h2 class="anchored" data-anchor-id="some-differences-in-how-bugs-and-stan-work">Some differences in how BUGS and Stan work</h2>
+<section id="bugs-is-interpreted-stan-is-compiled" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="bugs-is-interpreted-stan-is-compiled">BUGS is interpreted, Stan is compiled</h3>
+<p>Stan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).</p>
+</section>
+<section id="bugs-performs-mcmc-updating-one-scalar-parameter-at-a-time-stan-uses-hmc-which-moves-in-the-entire-space-of-all-the-parameters-at-each-step" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="bugs-performs-mcmc-updating-one-scalar-parameter-at-a-time-stan-uses-hmc-which-moves-in-the-entire-space-of-all-the-parameters-at-each-step">BUGS performs MCMC updating one scalar parameter at a time, Stan uses HMC which moves in the entire space of all the parameters at each step</h3>
+<p>BUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.</p>
+</section>
+<section id="differences-in-tuning-during-warmup" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="differences-in-tuning-during-warmup">Differences in tuning during warmup</h3>
+<p>BUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).</p>
+</section>
+<section id="the-stan-language-is-directly-executable-the-bugs-modeling-language-is-not" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="the-stan-language-is-directly-executable-the-bugs-modeling-language-is-not">The Stan language is directly executable, the BUGS modeling language is not</h3>
+<p>The BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).</p>
+</section>
+<section id="differences-in-statement-order" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="differences-in-statement-order">Differences in statement order</h3>
+<p>In BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets <code>mu</code> before using it to sample <code>y</code>:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>mu = a + b * x;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>translates to the following C++ code:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>mu = a + b * x;</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(y | mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Contrast this with the following Stan program:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma);</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>mu = a + b * x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This program is well formed, but is almost certainly a coding error, because it attempts to use <code>mu</code> before it is set. The direct translation to C++ code highlights the potential error of using <code>mu</code> in the first statement:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(y | mu, sigma);</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>mu = a + b * x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To trap these kinds of errors, variables are initialized to the special not-a-number (<code>NaN</code>) value. If <code>NaN</code> is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.</p>
+</section>
+<section id="stan-computes-the-gradient-of-the-log-density-bugs-computes-the-log-density-but-not-its-gradient" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-computes-the-gradient-of-the-log-density-bugs-computes-the-log-density-but-not-its-gradient">Stan computes the gradient of the log density, BUGS computes the log density but not its gradient</h3>
+<p>Stan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.</p>
+</section>
+<section id="both-bugs-and-stan-are-semi-automatic" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="both-bugs-and-stan-are-semi-automatic">Both BUGS and Stan are semi-automatic</h3>
+<p>Both BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (<span class="math inline">\(\hat{R}\)</span>) and the effective sample size to judge when to stop (stopping when <span class="math inline">\(\hat{R}\)</span> values do not counter-indicate convergence and when high enough effective sample size have been obtained).</p>
+</section>
+<section id="licensing" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h3>
+<p>WinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.</p>
+</section>
+<section id="interfaces" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="interfaces">Interfaces</h3>
+<p>Like WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.</p>
+</section>
+<section id="platforms" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="platforms">Platforms</h3>
+<p>Like OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.</p>
+</section>
+</section>
+<section id="some-differences-in-the-modeling-languages" class="level2">
+<h2 class="anchored" data-anchor-id="some-differences-in-the-modeling-languages">Some differences in the modeling languages</h2>
+<p>The BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n) y[i] ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Or, equivalently (recall that a line break is just another form of whitespace),</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n)</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  y[i] ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and also equivalently,</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n) {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  y[i] ~ normal(mu, sigma);</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.</p>
+<p>In Stan, variables can have names constructed using letters, numbers, and the underscore (<code>_</code>) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (<code>.</code>) symbol.</p>
+<p>In Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in <em>Bayesian Data Analysis</em>) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is <code>normal(1,2)</code>, not <code>normal(1,4)</code> or <code>normal(1,0.25)</code>.</p>
+<p>Similarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.</p>
+<p>The distributions have slightly different names:</p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: center;">BUGS</th>
+<th style="text-align: center;">Stan</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: center;"><code>dnorm</code></td>
+<td style="text-align: center;"><code>normal</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;"><code>dbinom</code></td>
+<td style="text-align: center;"><code>binomial</code></td>
+</tr>
+<tr class="odd">
+<td style="text-align: center;"><code>dpois</code></td>
+<td style="text-align: center;"><code>poisson</code></td>
+</tr>
+<tr class="even">
+<td style="text-align: center;">…</td>
+<td style="text-align: center;">…</td>
+</tr>
+</tbody>
+</table>
+<p>Stan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  total = <span class="dv">0</span>;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n) {</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>    theta[i] ~ normal(total, sigma);</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>    total = total + theta[i];</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In BUGS, the above model would not be legal because the variable <code>total</code> is defined more than once. But in Stan, the loop is executed in order, so <code>total</code> is overwritten in each step.</p>
+<p>Stan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.</p>
+<p>In Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.</p>
+<p>Stan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> allowing</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n) {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  y[i] ~ normal(mu[i], sigma[i]);</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>to be expressed more compactly as</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.</p>
+<p>Stan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of <code>K</code> parameters for each of <code>J</code> groups; this can be declared using</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[J] <span class="dt">vector</span>[K] theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then <code>theta[j]</code> is an expression denoting a <code>K</code>-vector and may be used in the code just like any other vector variable.</p>
+<p>An alternative encoding would be with a two-dimensional array, as in</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[J, K] <span class="dt">real</span> theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The vector version can have some advantages, both in convenience and in computational speed for some operations.</p>
+<p>A third encoding would use a matrix:</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[J, K] theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>but in this case, <code>theta[j]</code> is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in <code>theta[j]'</code>, may be used to convert the row vector <code>theta[j]</code> to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.</p>
+<p>Stan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the <a href="../stan-users-guide/finite-mixtures.html#zero-inflated.section">zero inflation section</a>.</p>
+<p>Stan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ...</span></span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>        <span class="co">// ... do something with n ....</span></span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>may be recoded using the following while loop.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> n;</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ...</span></span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>    n = <span class="dv">1</span>;</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">while</span> (n &lt;= N) {</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>        <span class="co">// ... do something with n ...</span></span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a>        n = n + <span class="dv">1</span>;</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="some-differences-in-the-statistical-models-that-are-allowed" class="level2">
+<h2 class="anchored" data-anchor-id="some-differences-in-the-statistical-models-that-are-allowed">Some differences in the statistical models that are allowed</h2>
+<p>Stan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.</p>
+<p>Stan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in <span class="citation" data-cites="LewandowskiKurowickaJoe:2009">Lewandowski, Kurowicka, and Joe (<a href="#ref-LewandowskiKurowickaJoe:2009" role="doc-biblioref">2009</a>)</span>. In particular, the Lewandowski et al.&nbsp;prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.</p>
+<p>In BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter <code>p</code> declared as</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; p;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and then have no distribution statement for <code>p</code> in the <code>model</code> block, then you are implicitly assigning a uniform <span class="math inline">\([0,1]\)</span> prior on <code>p</code>.</p>
+<p>On the other hand, if you have a parameter <code>theta</code> declared with</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and have no distribution statement for <code>theta</code> in the <code>model</code> block, then you are implicitly assigning an improper uniform prior on <span class="math inline">\((-\infty,\infty)\)</span> to <code>theta</code>.</p>
+<p>BUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> { }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although parameters in Stan models may have improper priors, we do not want improper <em>posterior</em> distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.</p>
+<p>Each statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>,<span class="dv">1</span>);</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">2</span>,<span class="dv">3</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The second line here does <em>not</em> simply overwrite the first; rather, <em>both</em> statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, <span class="math inline">\(\textsf{normal}(y \mid 0,1) * \textsf{normal}(y \mid 2,3)\)</span>, into the density function.</p>
+<p>For a perhaps more confusing example, consider the following two lines in a Stan model:</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a>x ~ normal(<span class="fl">0.8</span> * y, sigma);</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="fl">0.8</span> * x, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>At first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are <em>not</em> interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, <span class="math inline">\(\textsf{normal}(x \mid 0.8y,\sigma)
+\times \textsf{normal}(y \mid 0.8x,\sigma)\)</span>, which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.</p>
+<p>With censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> and <span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007</a>)</span> may be implemented in Stan directly as written.</p>
+<p>Stan, like BUGS, can benefit from human intervention in the form of reparameterization.</p>
+</section>
+<section id="some-differences-when-running-from-r" class="level2">
+<h2 class="anchored" data-anchor-id="some-differences-when-running-from-r">Some differences when running from R</h2>
+<p>Stan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the <a href="https://mc-stan.org">Stan web site</a>. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.</p>
+<p>In practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the <a href="https://mc-stan.org">Stan web site</a>.</p>
+<p>When you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)</p>
+<p>When running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable <code>x</code> from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.</p>
+<p>As in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using <code>plot()</code> and <code>print()</code>. You can access the simulations themselves using various extractor functions, as described in the RStan documentation.</p>
+<p>Various information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.</p>
+</section>
+<section id="the-stan-community" class="level2">
+<h2 class="anchored" data-anchor-id="the-stan-community">The Stan community</h2>
+<p>Stan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the <a href="https://mc-stan.org">Stan web site</a> for information on subscribing and posting and to look at archives.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-GelmanHill:2007" class="csl-entry" role="listitem">
+Gelman, Andrew, and Jennifer Hill. 2007. <em>Data Analysis Using Regression and Multilevel-Hierarchical Models</em>. Cambridge, United Kingdom: Cambridge University Press.
+</div>
+<div id="ref-LewandowskiKurowickaJoe:2009" class="csl-entry" role="listitem">
+Lewandowski, Daniel, Dorota Kurowicka, and Harry Joe. 2009. <span>“Generating Random Correlation Matrices Based on Vines and Extended Onion Method.”</span> <em>Journal of Multivariate Analysis</em> 100: 1989–2001.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Except where otherwise noted, we use “BUGS” to refer to WinBUGS, OpenBUGS, and JAGS, indiscriminately.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Most distributions have been vectorized, but currently the truncated versions may not exist.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/style-guide.html" class="pagination-link" aria-label="Stan Program Style Guide">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Stan Program Style Guide</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/for-bugs-users.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/gaussian-processes.html b/docs/2_39/stan-users-guide/gaussian-processes.html
new file mode 100644
index 000000000..4f53b356e
--- /dev/null
+++ b/docs/2_39/stan-users-guide/gaussian-processes.html
@@ -0,0 +1,2113 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Gaussian Processes</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/copulas.html" rel="next">
+<link href="../stan-users-guide/clustering.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Gaussian Processes">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/gaussian-processes.html">Gaussian Processes</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#gaussian-processes.chapter" id="toc-gaussian-processes.chapter" class="nav-link active" data-scroll-target="#gaussian-processes.chapter">Gaussian Processes</a>
+  <ul class="collapse">
+  <li><a href="#gaussian-process-regression" id="toc-gaussian-process-regression" class="nav-link" data-scroll-target="#gaussian-process-regression">Gaussian process regression</a></li>
+  <li><a href="#simulating-from-a-gaussian-process" id="toc-simulating-from-a-gaussian-process" class="nav-link" data-scroll-target="#simulating-from-a-gaussian-process">Simulating from a Gaussian process</a>
+  <ul class="collapse">
+  <li><a href="#multivariate-inputs" id="toc-multivariate-inputs" class="nav-link" data-scroll-target="#multivariate-inputs">Multivariate inputs</a></li>
+  <li><a href="#cholesky-factored-and-transformed-implementation" id="toc-cholesky-factored-and-transformed-implementation" class="nav-link" data-scroll-target="#cholesky-factored-and-transformed-implementation">Cholesky factored and transformed implementation</a></li>
+  </ul></li>
+  <li><a href="#fit-gp.section" id="toc-fit-gp.section" class="nav-link" data-scroll-target="#fit-gp.section">Fitting a Gaussian process</a>
+  <ul class="collapse">
+  <li><a href="#gp-with-a-normal-outcome" id="toc-gp-with-a-normal-outcome" class="nav-link" data-scroll-target="#gp-with-a-normal-outcome">GP with a normal outcome</a></li>
+  <li><a href="#discrete-outcomes-with-gaussian-processes" id="toc-discrete-outcomes-with-gaussian-processes" class="nav-link" data-scroll-target="#discrete-outcomes-with-gaussian-processes">Discrete outcomes with Gaussian processes</a></li>
+  <li><a href="#automatic-relevance-determination" id="toc-automatic-relevance-determination" class="nav-link" data-scroll-target="#automatic-relevance-determination">Automatic relevance determination</a></li>
+  <li><a href="#priors-gp.section" id="toc-priors-gp.section" class="nav-link" data-scroll-target="#priors-gp.section">Priors for Gaussian process parameters</a></li>
+  <li><a href="#predictive-inference-with-a-gaussian-process" id="toc-predictive-inference-with-a-gaussian-process" class="nav-link" data-scroll-target="#predictive-inference-with-a-gaussian-process">Predictive inference with a Gaussian process</a></li>
+  <li><a href="#multiple-output-gaussian-processes" id="toc-multiple-output-gaussian-processes" class="nav-link" data-scroll-target="#multiple-output-gaussian-processes">Multiple-output Gaussian processes</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/gaussian-processes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/gaussian-processes.html">Gaussian Processes</a></li></ol></nav></header>
+
+
+
+
+<section id="gaussian-processes.chapter" class="level1">
+<h1>Gaussian Processes</h1>
+<p>Gaussian processes are continuous stochastic processes and thus may be interpreted as providing a probability distribution over functions. A probability distribution over continuous functions may be viewed, roughly, as an uncountably infinite collection of random variables, one for each valid input. The generality of the supported functions makes Gaussian priors popular choices for priors in general multivariate (non-linear) regression problems.</p>
+<p>The defining feature of a Gaussian process is that the joint distribution of the function’s value at a finite number of input points is a multivariate normal distribution. This makes it tractable to both fit models from finite amounts of observed data and make predictions for finitely many new data points.</p>
+<p>Unlike a simple multivariate normal distribution, which is parameterized by a mean vector and covariance matrix, a Gaussian process is parameterized by a mean function and covariance function. The mean and covariance functions apply to vectors of inputs and return a mean vector and covariance matrix which provide the mean and covariance of the outputs corresponding to those input points in the functions drawn from the process.</p>
+<p>Gaussian processes can be encoded in Stan by implementing their mean and covariance functions or by using the specialized covariance functions outlined below, and plugging the result into the Gaussian model. This form of model is straightforward and may be used for simulation, model fitting, or posterior predictive inference. A more efficient Stan implementation for the GP with a normally distributed outcome marginalizes over the latent Gaussian process, and applies a Cholesky-factor reparameterization of the Gaussian to compute the likelihood and the posterior predictive distribution analytically.</p>
+<p>After defining Gaussian processes, this chapter covers the basic implementations for simulation, hyperparameter estimation, and posterior predictive inference for univariate regressions, multivariate regressions, and multivariate logistic regressions. Gaussian processes are general, and by necessity this chapter only touches on some basic models. For more information, see <span class="citation" data-cites="RasmussenWilliams:2006">Rasmussen and Williams (<a href="#ref-RasmussenWilliams:2006" role="doc-biblioref">2006</a>)</span>.</p>
+<p>Note that fitting Gaussian processes as described below using exact inference by computing Cholesky of the covariance matrix scales cubicly with the size of data. Due to how Stan autodiff is implemented, Stan is also slower than Gaussian process specialized software. It is likely that Gaussian processes using exact inference by computing Cholesky of the covariance matrix with <span class="math inline">\(N&gt;1000\)</span> are too slow for practical purposes in Stan. There are many approximations to speed-up Gaussian process computation, from which the basis function approaches for 1-3 dimensional <span class="math inline">\(x\)</span> are easiest to implement in Stan (see, e.g., <span class="citation" data-cites="Riutort-Mayol:2023:HSGP">Riutort-Mayol et al. (<a href="#ref-Riutort-Mayol:2023:HSGP" role="doc-biblioref">2023</a>)</span>).</p>
+<section id="gaussian-process-regression" class="level2">
+<h2 class="anchored" data-anchor-id="gaussian-process-regression">Gaussian process regression</h2>
+<p>The data for a multivariate Gaussian process regression consists of a series of <span class="math inline">\(N\)</span> inputs <span class="math inline">\(x_1,\dotsc,x_N \in \mathbb{R}^D\)</span> paired with outputs <span class="math inline">\(y_1,\dotsc,y_N \in \mathbb{R}\)</span>. The defining feature of Gaussian processes is that the probability of a finite number of outputs <span class="math inline">\(y\)</span> conditioned on their inputs <span class="math inline">\(x\)</span> is Gaussian: <span class="math display">\[
+y \sim \textsf{multivariate normal}(m(x), K(x \mid \theta)),
+\]</span> where <span class="math inline">\(m(x)\)</span> is an <span class="math inline">\(N\)</span>-vector and <span class="math inline">\(K(x \mid \theta)\)</span> is an <span class="math inline">\(N \times N\)</span> covariance matrix. The mean function <span class="math inline">\(m : \mathbb{R}^{N \times D}
+\rightarrow \mathbb{R}^{N}\)</span> can be anything, but the covariance function <span class="math inline">\(K : \mathbb{R}^{N \times D} \rightarrow \mathbb{R}^{N \times N}\)</span> must produce a positive-definite matrix for any input <span class="math inline">\(x\)</span>.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>A popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, <span class="math display">\[
+  K(x \mid \alpha, \rho, \sigma)_{i, j}
+= \alpha^2
+\exp \left(
+- \dfrac{1}{2 \rho^2} \sum_{d=1}^D (x_{i,d} - x_{j,d})^2
+\right)
++ \delta_{i, j} \sigma^2,
+\]</span> where <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(\rho\)</span>, and <span class="math inline">\(\sigma\)</span> are hyperparameters defining the covariance function and where <span class="math inline">\(\delta_{i, j}\)</span> is the Kronecker delta function with value 1 if <span class="math inline">\(i = j\)</span> and value 0 otherwise; this test is between the indexes <span class="math inline">\(i\)</span> and <span class="math inline">\(j\)</span>, not between values <span class="math inline">\(x_i\)</span> and <span class="math inline">\(x_j\)</span>. This kernel is obtained through a convolution of two independent Gaussian processes, <span class="math inline">\(f_1\)</span> and <span class="math inline">\(f_2\)</span>, with kernels <span class="math display">\[
+  K_1(x \mid \alpha, \rho)_{i, j}
+= \alpha^2
+\exp \left(
+- \dfrac{1}{2 \rho^2} \sum_{d=1}^D (x_{i,d} - x_{j,d})^2
+\right)
+\]</span> and <span class="math display">\[
+  K_2(x \mid \sigma)_{i, j}
+=
+\delta_{i, j} \sigma^2.
+\]</span></p>
+<p>The addition of <span class="math inline">\(\sigma^2\)</span> on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs <span class="math inline">\(x_i = x_j\)</span>. In statistical terms, <span class="math inline">\(\sigma\)</span> is the scale of the noise term in the regression.</p>
+<p>The hyperparameter <span class="math inline">\(\rho\)</span> is the <em>length-scale</em>, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of <span class="math inline">\(\rho\)</span> closer to zero lead the GP to represent high-frequency functions, whereas larger values of <span class="math inline">\(\rho\)</span> lead to low-frequency functions. The hyperparameter <span class="math inline">\(\alpha\)</span> is the <em>marginal standard deviation</em>. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP <span class="math inline">\(f_1\)</span> prior at a single input <span class="math inline">\(x\)</span> conditional on one value of <span class="math inline">\(\alpha\)</span> one would recover <span class="math inline">\(\alpha\)</span>.</p>
+<p>The only term in the squared exponential covariance function involving the inputs <span class="math inline">\(x_i\)</span> and <span class="math inline">\(x_j\)</span> is their vector difference, <span class="math inline">\(x_i - x_j\)</span>. This produces a process with stationary covariance in the sense that if an input vector <span class="math inline">\(x\)</span> is translated by a vector <span class="math inline">\(\epsilon\)</span> to <span class="math inline">\(x +
+\epsilon\)</span>, the covariance at any pair of outputs is unchanged, because <span class="math inline">\(K(x \mid \theta) = K(x + \epsilon \mid \theta)\)</span>.</p>
+<p>The summation involved is just the squared Euclidean distance between <span class="math inline">\(x_i\)</span> and <span class="math inline">\(x_j\)</span> (i.e., the squared <span class="math inline">\(L_2\)</span> norm of their difference, <span class="math inline">\(x_i -
+x_j\)</span>). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(\rho\)</span>, and <span class="math inline">\(\sigma\)</span>.</p>
+<p>Changing the notion of distance from Euclidean to taxicab distance (i.e., an <span class="math inline">\(L_1\)</span> norm) changes the support to functions which are continuous but not smooth.</p>
+</section>
+<section id="simulating-from-a-gaussian-process" class="level2">
+<h2 class="anchored" data-anchor-id="simulating-from-a-gaussian-process">Simulating from a Gaussian process</h2>
+<p>It is simplest to start with a Stan model that does nothing more than simulate draws of functions <span class="math inline">\(f\)</span> from a Gaussian process. In practical terms, the model will draw values <span class="math inline">\(y_n = f(x_n)\)</span> for finitely many input points <span class="math inline">\(x_n\)</span>.</p>
+<p>The Stan model defines the mean and covariance functions in a transformed data block and then samples outputs <span class="math inline">\(y\)</span> in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to <span class="math inline">\(\alpha^2 = 1\)</span>, <span class="math inline">\(\rho^2 = 1\)</span>, and <span class="math inline">\(\sigma^2 = 0.1\)</span>, and the mean function <span class="math inline">\(m\)</span> is defined to always return the zero vector, <span class="math inline">\(m(x) = \textbf{0}\)</span>. Consider the following implementation of a Gaussian process simulator.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, N] K;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu = rep_vector(<span class="dv">0</span>, N);</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:(N - <span class="dv">1</span>)) {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>    K[i, i] = <span class="dv">1</span> + <span class="fl">0.1</span>;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (j <span class="cf">in</span> (i + <span class="dv">1</span>):N) {</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>      K[i, j] = exp(-<span class="fl">0.5</span> * square(x[i] - x[j]));</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>      K[j, i] = K[i, j];</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  K[N, N] = <span class="dv">1</span> + <span class="fl">0.1</span>;</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, K);</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, <span class="fl">1.0</span>, <span class="fl">1.0</span>);</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu = rep_vector(<span class="dv">0</span>, N);</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>    K[n, n] = K[n, n] + <span class="fl">0.1</span>;</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, K);</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The input data are just the vector of inputs <code>x</code> and its size <code>N</code>. Such a model can be used with values of <code>x</code> evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.</p>
+<section id="multivariate-inputs" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multivariate-inputs">Multivariate inputs</h3>
+<p>Only the input data needs to change in moving from a univariate model to a multivariate model.</p>
+<p>The only lines that change from the univariate model above are as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] x;</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data are now declared as an array of vectors instead of an array of scalars; the dimensionality <code>D</code> is also declared.</p>
+<p>In the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.</p>
+</section>
+<section id="cholesky-factored-and-transformed-implementation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cholesky-factored-and-transformed-implementation">Cholesky factored and transformed implementation</h3>
+<p>A more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose <span class="math inline">\(\eta\)</span> is an an isotropic standard normal variate <span class="math display">\[
+\eta \sim \textsf{normal}(\textbf{0}, \textrm{I}),
+\]</span> where <span class="math inline">\(\textbf{0}\)</span> is an <span class="math inline">\(N\)</span>-vector of 0 values and <span class="math inline">\(\textrm{I}\)</span> is the <span class="math inline">\(N
+\times N\)</span> identity matrix. Let <span class="math inline">\(L\)</span> be the Cholesky decomposition of <span class="math inline">\(K(x \mid \theta)\)</span>, i.e., the lower-triangular matrix <span class="math inline">\(L\)</span> such that <span class="math inline">\(LL^{\top} =
+K(x \mid \theta)\)</span>. Then the transformed variable <span class="math inline">\(\mu + L\eta\)</span> has the intended target distribution, <span class="math display">\[
+  \mu + L\eta \sim \textsf{multivariate normal}(\mu(x), K(x \mid \theta)).
+\]</span></p>
+<p>This transform can be applied directly to Gaussian process simulation.</p>
+<p>This model has the same data declarations for <code>N</code> and <code>x</code>, and the same transformed data definitions of <code>mu</code> and <code>K</code> as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, N] L;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  L = cholesky_decompose(K);</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>  y = mu + L * eta;</span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix <code>K</code> computed. The isotropic normal distribution for <code>eta</code> is specified as a vectorized univariate distribution for efficiency; this specifies that each <code>eta[n]</code> has an independent standard normal distribution. The sampled vector <code>y</code> is then defined as a generated quantity using a direct encoding of the transform described above.</p>
+</section>
+</section>
+<section id="fit-gp.section" class="level2">
+<h2 class="anchored" data-anchor-id="fit-gp.section">Fitting a Gaussian process</h2>
+<section id="gp-with-a-normal-outcome" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="gp-with-a-normal-outcome">GP with a normal outcome</h3>
+<p>The full generative model for a GP with a normal outcome, <span class="math inline">\(y \in \mathbb{R}^N\)</span>, with inputs <span class="math inline">\(x \in \mathbb{R}^N\)</span>, for a finite <span class="math inline">\(N\)</span>: <span class="math display">\[\begin{align*}
+\rho   &amp;\sim \textsf{InvGamma}(5, 5) \\
+\alpha &amp;\sim \textsf{normal}(0, 1) \\
+\sigma &amp;\sim \textsf{normal}(0, 1) \\
+f      &amp;\sim \textsf{multivariate normal}\left(0, K(x \mid \alpha, \rho)\right) \\
+y_i    &amp;\sim \textsf{normal}(f_i, \sigma) \, \forall i \in \{1, \dots, N\}
+\end{align*}\]</span> With a normal outcome, it is possible to integrate out the Gaussian process <span class="math inline">\(f\)</span>, yielding the more parsimonious model: <span class="math display">\[\begin{align*}
+\rho   &amp;\sim \textsf{InvGamma}(5, 5) \\
+\alpha &amp;\sim \textsf{normal}(0, 1) \\
+\sigma &amp;\sim \textsf{normal}(0, 1) \\
+y      &amp;\sim \textsf{multivariate normal}
+         \left(0, K(x \mid \alpha, \rho) + \textbf{I}_N \sigma^2\right) \\
+\end{align*}\]</span></p>
+<p>It can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.</p>
+<p>The hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (<span class="math inline">\(\alpha\)</span>), the scale of the output noise (<span class="math inline">\(\sigma\)</span>), and the scale at which distances are measured among inputs (<span class="math inline">\(\rho\)</span>). See the <a href="#priors-gp.section">Gaussian process priors section</a> for more information about how to specify appropriate priors for the hyperparameters.</p>
+<p>The Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix <code>K</code> in the model block, rather than the transformed data block.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu = rep_vector(<span class="dv">0</span>, N);</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sq_sigma = square(sigma);</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>  <span class="co">// diagonal elements</span></span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>    K[n, n] = K[n, n] + sq_sigma;</span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>  L_K = cholesky_decompose(K);</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb5-27"><a href="#cb5-27" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb5-28"><a href="#cb5-28" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb5-29"><a href="#cb5-29" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-30"><a href="#cb5-30" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal_cholesky(mu, L_K);</span>
+<span id="cb5-31"><a href="#cb5-31" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data block declares a vector <code>y</code> of observed values <code>y[n]</code> for inputs <code>x[n]</code>. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix <code>K</code> is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value <code>y</code> is known and the covariance matrix <code>K</code> is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.</p>
+<p>We have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the <code>cholesky_decompose</code> function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (<span class="math inline">\(N \gtrsim 100\)</span>) the Cholesky decomposition version will be faster.</p>
+<p>Hamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model <span class="citation" data-cites="Neal:1997">(<a href="#ref-Neal:1997" role="doc-biblioref">Neal 1997</a>)</span>. If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.</p>
+<section id="latent-variable-gp" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="latent-variable-gp">Latent variable GP</h4>
+<p>We can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, <span class="math inline">\(\delta\)</span> to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb6-17"><a href="#cb6-17" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb6-18"><a href="#cb6-18" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb6-19"><a href="#cb6-19" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb6-20"><a href="#cb6-20" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-21"><a href="#cb6-21" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb6-22"><a href="#cb6-22" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb6-23"><a href="#cb6-23" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb6-24"><a href="#cb6-24" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb6-25"><a href="#cb6-25" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-26"><a href="#cb6-26" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb6-27"><a href="#cb6-27" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb6-28"><a href="#cb6-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-29"><a href="#cb6-29" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-30"><a href="#cb6-30" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb6-31"><a href="#cb6-31" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb6-32"><a href="#cb6-32" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb6-33"><a href="#cb6-33" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb6-34"><a href="#cb6-34" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-35"><a href="#cb6-35" aria-hidden="true" tabindex="-1"></a>  y ~ normal(f, sigma);</span>
+<span id="cb6-36"><a href="#cb6-36" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Two differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length <span class="math inline">\(N\)</span> called <code>eta</code>. This is used in the model block to generate a multivariate normal vector called <span class="math inline">\(f\)</span>, corresponding to the latent GP. We put a <span class="math inline">\(\textsf{normal}(0,1)\)</span> prior on <code>eta</code> like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for <span class="math inline">\(y\)</span> with one <span class="math inline">\(N\)</span>-dimensional multivariate normal with an identity covariance matrix multiplied by <span class="math inline">\(\sigma^2\)</span>, we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.</p>
+</section>
+</section>
+<section id="discrete-outcomes-with-gaussian-processes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="discrete-outcomes-with-gaussian-processes">Discrete outcomes with Gaussian processes</h3>
+<p>Gaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.</p>
+<section id="poisson-gp" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="poisson-gp">Poisson GP</h4>
+<p>If we want to model count data, we can remove the <span class="math inline">\(\sigma\)</span> parameter, and use <code>poisson_log</code>, which implements a Poisson distribution with log link function, rather than <code>normal</code>. We can also add an overall mean parameter, <span class="math inline">\(a\)</span>, which will account for the marginal expected value for <span class="math inline">\(y\)</span>. We do this because we cannot center count data like we would for normally distributed data.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-24"><a href="#cb7-24" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb7-25"><a href="#cb7-25" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb7-26"><a href="#cb7-26" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-27"><a href="#cb7-27" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb7-28"><a href="#cb7-28" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb7-29"><a href="#cb7-29" aria-hidden="true" tabindex="-1"></a>  a ~ std_normal();</span>
+<span id="cb7-30"><a href="#cb7-30" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb7-31"><a href="#cb7-31" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-32"><a href="#cb7-32" aria-hidden="true" tabindex="-1"></a>  y ~ poisson_log(a + f);</span>
+<span id="cb7-33"><a href="#cb7-33" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="poisson-gp-using-an-embedded-laplace-approximation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="poisson-gp-using-an-embedded-laplace-approximation">Poisson GP using an embedded Laplace approximation</h4>
+<p>For computational reasons, we may want to integrate out the latent variable <span class="math inline">\(f\)</span>, as was done in the normal output model. Unfortunately, exact marginalization over <span class="math inline">\(f\)</span> is not possible when the outcome model is not normal. Instead, we may perform <em>approximate</em> marginalization with an <em>embedded Laplace approximation</em> <span class="citation" data-cites="RasmussenWilliams:2006 Rue:2009 Margossian:2020">(<a href="#ref-RasmussenWilliams:2006" role="doc-biblioref">Rasmussen and Williams 2006</a>; <a href="#ref-Rue:2009" role="doc-biblioref">Rue, Martino, and Chopin 2009</a>; <a href="#ref-Margossian:2020" role="doc-biblioref">Margossian et al. 2020</a>)</span>. To do so, we first use the function <code>laplace_marginal</code> to approximate the marginal likelihood <span class="math inline">\(p(y \mid \rho, \alpha, a)\)</span> and sample the hyperparameters with Hamiltonian Monte Carlo sampling. Then, we recover the integrated out <span class="math inline">\(f\)</span> in the <code>generated quantities</code> block using <code>laplace_latent_rng</code>.</p>
+<p>The embedded Laplace approximation computes a Gaussian approximation of the conditional posterior, <span class="math display">\[
+  \hat p_\mathcal{L}(f \mid \rho, \alpha, a, y) \approx p(f \mid \rho, \alpha, a, y),
+\]</span> where <span class="math inline">\(\hat p_\mathcal{L}\)</span> is a Gaussian that matches the mode and curvature of <span class="math inline">\(p(f \mid \rho, \alpha, a, y)\)</span>. We then obtain an approximation of the marginal likelihood as follows: <span class="math display">\[
+  \hat p_\mathcal{L}(y \mid \rho, \alpha, a)
+    = \frac{p(f^* \mid \alpha, \rho) p(y \mid f^*, a)}{
+    \hat p_\mathcal{L}(f^* \mid \rho, \alpha, a, y)},
+\]</span> where <span class="math inline">\(f^*\)</span> is the mode of <span class="math inline">\(p(f \mid \rho, \alpha, a, y)\)</span>, obtained via numerical optimization.</p>
+<p>To use Stan’s embedded Laplace approximation, we must define the prior covariance function and the log likelihood function in the functions block.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// log likelihood function</span></span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ll_function(<span class="dt">vector</span> f, <span class="dt">real</span> a, <span class="dt">array</span>[] <span class="dt">int</span> y) {</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">return</span> poisson_log_lpmf(y | a + f);</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// covariance function</span></span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span> cov_function(<span class="dt">real</span> rho, <span class="dt">real</span> alpha, <span class="dt">array</span>[] <span class="dt">real</span> x, <span class="dt">int</span> N, <span class="dt">real</span> delta) {</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> add_diag(K, delta)</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The embedded Laplace relies on calculations of the log likelihood’s Hessian, <span class="math inline">\(\partial^2 \log p(y \mid f, a, \rho, \alpha) / \partial f^2\)</span>, and these calculations can be much faster when the Hessian is sparse. In particular, it is expected that the Hessian is block diagonal. In the <code>transformed data</code> block we can specify the block size of the Hessian.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> hessian_block_size = <span class="dv">1</span>;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For example, if <span class="math inline">\(y_i\)</span> depends only on <span class="math inline">\(f_i\)</span>, then the Hessian of the log likelihood is diagonal and the block size is 1. On the other hand, if the Hessian is not sparse, then we set the hessian block size to <span class="math inline">\(N\)</span>, where <span class="math inline">\(N\)</span> is the dimension of <span class="math inline">\(f\)</span>. Currently, Stan does not check the block size of the Hessian and so the user is responsible for correctly specifying the block size.</p>
+<p>Finally, we increment <code>target</code> in the model block with the approximation to <span class="math inline">\(\log p(y \mid \rho, \alpha, a)\)</span>.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>                             cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Notice that we do not need to construct <span class="math inline">\(f\)</span> explicitly, since it is marginalized out. Instead, we can recover the latent variables in <code>generated quantities</code>:</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Users can set the control parameters of the embedded Laplace approximation, via <code>laplace_marginal_tol</code> and <code>laplace_latent_tol_rng</code>. When using these functions, the user must set <em>all</em> the control options and store them in a tuple. These control parameters mostly concern the numerical optimizer used to find the mode <span class="math inline">\(f^*\)</span> of <span class="math inline">\(p(f \mid \rho, \alpha, a)\)</span>.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">tuple</span>(<span class="dt">vector</span>[N], <span class="dt">real</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>) laplace_ops;</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.1</span> = rep_vector(<span class="dv">0</span>, N);  <span class="co">// starting point for Laplace optimizer</span></span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.2</span> = <span class="fl">1.49e-8</span>;           <span class="co">// tolerance for optimizer</span></span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.3</span> = <span class="dv">500</span>;               <span class="co">// maximum number of steps for optimizer.</span></span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.4</span> = <span class="dv">1</span>;              <span class="co">// solver type being used.</span></span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.5</span> = <span class="dv">1000</span>;           <span class="co">// max number of steps for linesearch.</span></span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.6</span> = <span class="dv">1</span>;              <span class="co">// allow_fallback (1: TRUE, 0: FALSE)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If users want to depart from the defaults for only some of the control parameters, a tuple with the default values (as above) can be created with the helper callable <code>generate_laplace_options()</code>, and the specific control parameter can then be modified,</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">tuple</span>(<span class="dt">vector</span>[N], <span class="dt">real</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>) laplace_ops =</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>    generate_laplace_options(N);</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.2</span> = <span class="fl">1e-6</span>; <span class="co">// make tolerance of the optimizer less strict.</span></span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The tuple <code>laplace_ops</code> is then passed to <code>laplace_marginal_tol</code> and <code>laplace_rng_tol</code>.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal_tol(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>                                 cov_function, (rho, alpha, x, N, delta),</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>                                 laplace_ops);</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta),</span>
+<span id="cb14-12"><a href="#cb14-12" aria-hidden="true" tabindex="-1"></a>                                   laplace_ops);</span>
+<span id="cb14-13"><a href="#cb14-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Stan also provides support for a limited menu of built-in functions, including the Poisson distribution with a log link and and prior mean <span class="math inline">\(m\)</span>. When using such a built-in function, the user does not need to specify a likelihood in the <code>functions</code> block. However, the user must strictly follow the signature of the likelihood: in this case, <span class="math inline">\(m\)</span> must be a vector of length <span class="math inline">\(N\)</span> (to allow for different offsets for each observation <span class="math inline">\(y_i\)</span>) and we must indicate which element of <span class="math inline">\(f\)</span> each component of <span class="math inline">\(y\)</span> matches using the variable <span class="math inline">\(y_\text{index}\)</span>. In our example, there is a simple pairing <span class="math inline">\((y_i, f_i)\)</span>, however we could imagine a scenario where multiple observations <span class="math inline">\((y_{j1}, y_{j2}, ...)\)</span> are observed for a single <span class="math inline">\(f_j\)</span>.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[n_obs] <span class="dt">int</span> y_index;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n_obs) y_index[i] = i - <span class="dv">1</span>;</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>transformed parameter {</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] m = rep_vector(a, N);</span>
+<span id="cb15-11"><a href="#cb15-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-12"><a href="#cb15-12" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-13"><a href="#cb15-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-14"><a href="#cb15-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-15"><a href="#cb15-15" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal_poisson_log_lpmf(y | y_index, m,</span>
+<span id="cb15-16"><a href="#cb15-16" aria-hidden="true" tabindex="-1"></a>                                       cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb15-17"><a href="#cb15-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-18"><a href="#cb15-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-19"><a href="#cb15-19" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb15-20"><a href="#cb15-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_poisson_log_rng(y, y_index, m,</span>
+<span id="cb15-21"><a href="#cb15-21" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb15-22"><a href="#cb15-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As before, we could specify the control parameters for the embedded Laplace approximation using <code>laplace_marginal_tol_poisson_log_lpmf</code> and <code>laplace_latent_tol_poisson_log_nrg</code>.</p>
+<p>Marginalization with a Laplace approximation can lead to faster inference, however it also introduces an approximation error. In practice, this error is negligible when using a Poisson likelihood and the approximation works well for log concave likelihoods <span class="citation" data-cites="Kuss:2005 Vanhatalo:2010 Cseke:2011 Vehtari:2016">(<a href="#ref-Kuss:2005" role="doc-biblioref">Kuss and Rasmussen 2005</a>; <a href="#ref-Vanhatalo:2010" role="doc-biblioref">Vanhatalo, Pietiläinen, and Vehtari 2010</a>; <a href="#ref-Cseke:2011" role="doc-biblioref">Cseke and Heskes 2011</a>; <a href="#ref-Vehtari:2016" role="doc-biblioref">Vehtari et al. 2016</a>)</span>. Still, users should exercise caution, especially when trying unconventional likelihoods.</p>
+</section>
+<section id="logistic-gp-regression" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="logistic-gp-regression">Logistic GP regression</h4>
+<p>For binary classification problems, the observed outputs <span class="math inline">\(z_n \in
+\{ 0,1 \}\)</span> are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs <span class="math inline">\(y_n\)</span> through the logistic link, <span class="math display">\[
+z_n \sim \textsf{Bernoulli}(\operatorname{logit}^{-1}(y_n)),
+\]</span> or in other words, <span class="math display">\[
+\Pr[z_n = 1] = \operatorname{logit}^{-1}(y_n).
+\]</span></p>
+<p>We can extend our latent variable GP Stan program to deal with classification problems. Below <code>a</code> is the bias term, which can help account for imbalanced classes in the training data:</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>  z ~ bernoulli_logit(a + f);</span>
+<span id="cb16-10"><a href="#cb16-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="logistic-gp-regression-with-an-embedded-laplace-approximation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="logistic-gp-regression-with-an-embedded-laplace-approximation">Logistic GP regression with an embedded Laplace approximation</h4>
+<p>As with the Poisson GP, we cannot marginalize the latent variables exactly, however we can resort to an embedded Laplace approximation.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// log likelihood function</span></span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ll_function(<span class="dt">vector</span> f, <span class="dt">real</span> a, <span class="dt">array</span>[] <span class="dt">int</span> z) {</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">return</span> bernoulli_logit_lpmf(z | a + f);</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// covariance function</span></span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span> cov_function(<span class="dt">real</span> rho, <span class="dt">real</span> alpha, <span class="dt">array</span>[] <span class="dt">real</span> x, <span class="dt">int</span> N, <span class="dt">real</span> delta) {</span>
+<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb17-10"><a href="#cb17-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> add_diag(K, delta)</span>
+<span id="cb17-11"><a href="#cb17-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-12"><a href="#cb17-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-13"><a href="#cb17-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-14"><a href="#cb17-14" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb17-15"><a href="#cb17-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-16"><a href="#cb17-16" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb17-17"><a href="#cb17-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> hessian_block_size = <span class="dv">1</span>;</span>
+<span id="cb17-18"><a href="#cb17-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-19"><a href="#cb17-19" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-20"><a href="#cb17-20" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb17-21"><a href="#cb17-21" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-22"><a href="#cb17-22" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb17-23"><a href="#cb17-23" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal(ll_function, (a, z), hessian_block_size,</span>
+<span id="cb17-24"><a href="#cb17-24" aria-hidden="true" tabindex="-1"></a>                             cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb17-25"><a href="#cb17-25" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-26"><a href="#cb17-26" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-27"><a href="#cb17-27" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb17-28"><a href="#cb17-28" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_rng(ll_function, (a, z), hessian_block_size,</span>
+<span id="cb17-29"><a href="#cb17-29" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb17-30"><a href="#cb17-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>While marginalization with a Laplace approximation can lead to faster inference, it also introduces an approximation error. In practice, this error may not be negligible with a Bernoulli likelihood; for more discussion see, e.g. <span class="citation" data-cites="Vehtari:2016 Margossian:2020">(<a href="#ref-Vehtari:2016" role="doc-biblioref">Vehtari et al. 2016</a>; <a href="#ref-Margossian:2020" role="doc-biblioref">Margossian et al. 2020</a>)</span>.</p>
+</section>
+</section>
+<section id="automatic-relevance-determination" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="automatic-relevance-determination">Automatic relevance determination</h3>
+<p>If we have multivariate inputs <span class="math inline">\(x \in \mathbb{R}^D\)</span>, the squared exponential covariance function can be further generalized by fitting a scale parameter <span class="math inline">\(\rho_d\)</span> for each dimension <span class="math inline">\(d\)</span>, <span class="math display">\[
+  k(x \mid \alpha, \vec{\rho}, \sigma)_{i, j} = \alpha^2 \exp
+\left(-\dfrac{1}{2}
+\sum_{d=1}^D \dfrac{1}{\rho_d^2} (x_{i,d} - x_{j,d})^2
+\right)
++ \delta_{i, j}\sigma^2.
+\]</span> The estimation of <span class="math inline">\(\rho\)</span> was termed “automatic relevance determination” by <span class="citation" data-cites="Neal:1996">Neal (<a href="#ref-Neal:1996" role="doc-biblioref">1996</a>)</span>, but this is misleading, because the magnitude of the scale of the posterior for each <span class="math inline">\(\rho_d\)</span> is dependent on the scaling of the input data along dimension <span class="math inline">\(d\)</span>. Moreover, the scale of the parameters <span class="math inline">\(\rho_d\)</span> measures non-linearity along the <span class="math inline">\(d\)</span>-th dimension, rather than “relevance” <span class="citation" data-cites="PiironenVehtari:2016">(<a href="#ref-PiironenVehtari:2016" role="doc-biblioref">Piironen and Vehtari 2016</a>)</span>.</p>
+<p>A priori, the closer <span class="math inline">\(\rho_d\)</span> is to zero, the more nonlinear the conditional mean in dimension <span class="math inline">\(d\)</span> is. A posteriori, the actual dependencies between <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span> play a role. With one covariate <span class="math inline">\(x_1\)</span> having a linear effect and another covariate <span class="math inline">\(x_2\)</span> having a nonlinear effect, it is possible that <span class="math inline">\(\rho_1 &gt; \rho_2\)</span> even if the predictive relevance of <span class="math inline">\(x_1\)</span> is higher <span class="citation" data-cites="RasmussenWilliams:2006">(<a href="#ref-RasmussenWilliams:2006" role="doc-biblioref">Rasmussen and Williams 2006, 80</a>)</span>. The collection of <span class="math inline">\(\rho_d\)</span> (or <span class="math inline">\(1/\rho_d\)</span>) parameters can also be modeled hierarchically.</p>
+<p>The implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying <code>rho</code> to be an array.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] x;</span>
+<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb18-12"><a href="#cb18-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb18-13"><a href="#cb18-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb18-14"><a href="#cb18-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb18-15"><a href="#cb18-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-16"><a href="#cb18-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb18-17"><a href="#cb18-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb18-18"><a href="#cb18-18" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb18-19"><a href="#cb18-19" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb18-20"><a href="#cb18-20" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb18-21"><a href="#cb18-21" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb18-22"><a href="#cb18-22" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb18-23"><a href="#cb18-23" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb18-24"><a href="#cb18-24" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb18-25"><a href="#cb18-25" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb18-26"><a href="#cb18-26" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb18-27"><a href="#cb18-27" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb18-28"><a href="#cb18-28" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb18-29"><a href="#cb18-29" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb18-30"><a href="#cb18-30" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb18-31"><a href="#cb18-31" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb18-32"><a href="#cb18-32" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb18-33"><a href="#cb18-33" aria-hidden="true" tabindex="-1"></a>  y ~ normal(f, sigma);</span>
+<span id="cb18-34"><a href="#cb18-34" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="priors-gp.section" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="priors-gp.section">Priors for Gaussian process parameters</h3>
+<p>Formulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.</p>
+<p>Perhaps most importantly, the parameters <span class="math inline">\(\rho\)</span> and <span class="math inline">\(\alpha\)</span> are weakly identified <span class="citation" data-cites="zhang-gp:2004">(<a href="#ref-zhang-gp:2004" role="doc-biblioref">Zhang 2004</a>)</span>. The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.</p>
+<section id="priors-for-length-scale" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="priors-for-length-scale">Priors for length-scale</h4>
+<p>GPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.</p>
+<p>We may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.</p>
+<p>If our model consists of only the GP, i.e.: <span class="math display">\[\begin{align*}
+f   &amp;\sim \textsf{multivariate normal}\left(0, K(x \mid \alpha, \rho)\right) \\
+y_i &amp;\sim \textsf{normal}(f_i, \sigma) \, \forall i \in \{1, \dots, N\} \\
+    &amp; x \in \mathbb{R}^{N \times D}, \quad
+      f \in \mathbb{R}^N
+\end{align*}\]</span></p>
+<p>we likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, <code>inv_gamma_lpdf</code> in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a>.</p>
+<p>If we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: <span class="math display">\[\begin{align*}
+f   &amp;\sim \textsf{multivariate normal}\big(0, K(x \mid \alpha, \rho)\big) \\
+y_i &amp;\sim \textsf{normal}\left(\beta_0 + x_i \beta_{[1:D]} + f_i, \sigma\right) \, \forall i
+  \in \{1, \dots, N\} \\
+    &amp; x_i^T, \beta_{[1:D]} \in \mathbb{R}^D,\quad
+      x \in \mathbb{R}^{N \times D},\quad
+      f \in \mathbb{R}^N
+\end{align*}\]</span></p>
+<p>we’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: <span class="math display">\[\begin{align*}
+f(x \mid a, b, p) &amp;= \dfrac{\left(a/b\right)^{p/2}}{2K_p\left(\sqrt{ab}\right)} x^{p - 1}\exp\big(-(ax + b
+  / x)/2\big) \\
+  &amp; x, a, b \in \mathbb{R}^{+},\quad
+    p \in \mathbb{Z}
+\end{align*}\]</span></p>
+<p>which has an inverse gamma left tail if <span class="math inline">\(p \leq 0\)</span> and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> generalized_inverse_gaussian_lpdf(<span class="dt">real</span> x, <span class="dt">int</span> p,</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>                                        <span class="dt">real</span> a, <span class="dt">real</span> b) {</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> p * <span class="fl">0.5</span> * log(a / b)</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>      - log(<span class="dv">2</span> * modified_bessel_second_kind(p, sqrt(a * b)))</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>      + (p - <span class="dv">1</span>) * log(x)</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>      - (a * x + b / x) * <span class="fl">0.5</span>;</span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a> }</span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of <span class="math inline">\(\rho\)</span> in a fixed-domain <span class="math inline">\([0,T]\)</span>, we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is <span class="math inline">\(T / \pi
+\rho\)</span>. You can see that as <span class="math inline">\(\rho\)</span> decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.</p>
+</section>
+<section id="priors-for-marginal-standard-deviation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="priors-for-marginal-standard-deviation">Priors for marginal standard deviation</h4>
+<p>The parameter <span class="math inline">\(\alpha\)</span> corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-<span class="math inline">\(t\)</span> prior on <span class="math inline">\(\alpha\)</span>.</p>
+<p>A half-<span class="math inline">\(t\)</span> or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.</p>
+</section>
+</section>
+<section id="predictive-inference-with-a-gaussian-process" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="predictive-inference-with-a-gaussian-process">Predictive inference with a Gaussian process</h3>
+<p>Suppose for a given sequence of inputs <span class="math inline">\(x\)</span> that the corresponding outputs <span class="math inline">\(y\)</span> are observed. Given a new sequence of inputs <span class="math inline">\(\tilde{x}\)</span>, the posterior predictive distribution of their labels is computed by sampling outputs <span class="math inline">\(\tilde{y}\)</span> according to <span class="math display">\[
+p\left(\tilde{y} \mid \tilde{x},x,y\right)
+\ = \
+\frac{p\left(\tilde{y}, y \mid \tilde{x},x\right)}
+     {p(y \mid x)}
+\ \propto \
+p\left(\tilde{y}, y \mid \tilde{x},x\right).
+\]</span></p>
+<p>A direct implementation in Stan defines a model in terms of the joint distribution of the observed <span class="math inline">\(y\)</span> and unobserved <span class="math inline">\(\tilde{y}\)</span>.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] y1;</span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
+<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb20-9"><a href="#cb20-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb20-10"><a href="#cb20-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N = N1 + N2;</span>
+<span id="cb20-11"><a href="#cb20-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb20-12"><a href="#cb20-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb20-13"><a href="#cb20-13" aria-hidden="true" tabindex="-1"></a>    x[n1] = x1[n1];</span>
+<span id="cb20-14"><a href="#cb20-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-15"><a href="#cb20-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb20-16"><a href="#cb20-16" aria-hidden="true" tabindex="-1"></a>    x[N1 + n2] = x2[n2];</span>
+<span id="cb20-17"><a href="#cb20-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-18"><a href="#cb20-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-19"><a href="#cb20-19" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-20"><a href="#cb20-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb20-21"><a href="#cb20-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb20-22"><a href="#cb20-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb20-23"><a href="#cb20-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb20-24"><a href="#cb20-24" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-25"><a href="#cb20-25" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb20-26"><a href="#cb20-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb20-27"><a href="#cb20-27" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb20-28"><a href="#cb20-28" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb20-29"><a href="#cb20-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb20-30"><a href="#cb20-30" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-31"><a href="#cb20-31" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb20-32"><a href="#cb20-32" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb20-33"><a href="#cb20-33" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb20-34"><a href="#cb20-34" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb20-35"><a href="#cb20-35" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-36"><a href="#cb20-36" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb20-37"><a href="#cb20-37" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb20-38"><a href="#cb20-38" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-39"><a href="#cb20-39" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-40"><a href="#cb20-40" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb20-41"><a href="#cb20-41" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb20-42"><a href="#cb20-42" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb20-43"><a href="#cb20-43" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb20-44"><a href="#cb20-44" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb20-45"><a href="#cb20-45" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-46"><a href="#cb20-46" aria-hidden="true" tabindex="-1"></a>  y1 ~ normal(f[<span class="dv">1</span>:N1], sigma);</span>
+<span id="cb20-47"><a href="#cb20-47" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-48"><a href="#cb20-48" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb20-49"><a href="#cb20-49" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] y2;</span>
+<span id="cb20-50"><a href="#cb20-50" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb20-51"><a href="#cb20-51" aria-hidden="true" tabindex="-1"></a>    y2[n2] = normal_rng(f[N1 + n2], sigma);</span>
+<span id="cb20-52"><a href="#cb20-52" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-53"><a href="#cb20-53" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The input vectors <code>x1</code> and <code>x2</code> are declared as data, as is the observed output vector <code>y1</code>. The unknown output vector <code>y2</code>, which corresponds to input vector <code>x2</code>, is declared in the generated quantities block and will be sampled when the model is executed.</p>
+<p>A transformed data block is used to combine the input vectors <code>x1</code> and <code>x2</code> into a single vector <code>x</code>.</p>
+<p>The model block declares and defines a local variable for the combined output vector <code>f</code>, which consists of the concatenation of the conditional mean for known outputs <code>y1</code> and unknown outputs <code>y2</code>. Thus the combined output vector <code>f</code> is aligned with the combined input vector <code>x</code>. All that is left is to define the univariate normal distribution statement for <code>y</code>.</p>
+<p>The generated quantities block defines the quantity <code>y2</code>. We generate <code>y2</code> by randomly generating <code>N2</code> values from univariate normals with each mean corresponding to the appropriate element in <code>f</code>.</p>
+<section id="predictive-inference-in-non-gaussian-gps" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="predictive-inference-in-non-gaussian-gps">Predictive inference in non-Gaussian GPs</h4>
+<p>We can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.</p>
+<p>Consider the following full model for prediction using logistic Gaussian process regression.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z1;</span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N = N1 + N2;</span>
+<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a>    x[n1] = x1[n1];</span>
+<span id="cb21-14"><a href="#cb21-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-15"><a href="#cb21-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb21-16"><a href="#cb21-16" aria-hidden="true" tabindex="-1"></a>    x[N1 + n2] = x2[n2];</span>
+<span id="cb21-17"><a href="#cb21-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-18"><a href="#cb21-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-19"><a href="#cb21-19" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-20"><a href="#cb21-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb21-21"><a href="#cb21-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb21-22"><a href="#cb21-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb21-23"><a href="#cb21-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb21-24"><a href="#cb21-24" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-25"><a href="#cb21-25" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb21-26"><a href="#cb21-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb21-27"><a href="#cb21-27" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb21-28"><a href="#cb21-28" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb21-29"><a href="#cb21-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb21-30"><a href="#cb21-30" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-31"><a href="#cb21-31" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb21-32"><a href="#cb21-32" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb21-33"><a href="#cb21-33" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb21-34"><a href="#cb21-34" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb21-35"><a href="#cb21-35" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-36"><a href="#cb21-36" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb21-37"><a href="#cb21-37" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb21-38"><a href="#cb21-38" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-39"><a href="#cb21-39" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-40"><a href="#cb21-40" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb21-41"><a href="#cb21-41" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb21-42"><a href="#cb21-42" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb21-43"><a href="#cb21-43" aria-hidden="true" tabindex="-1"></a>  a ~ std_normal();</span>
+<span id="cb21-44"><a href="#cb21-44" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb21-45"><a href="#cb21-45" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-46"><a href="#cb21-46" aria-hidden="true" tabindex="-1"></a>  z1 ~ bernoulli_logit(a + f[<span class="dv">1</span>:N1]);</span>
+<span id="cb21-47"><a href="#cb21-47" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-48"><a href="#cb21-48" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb21-49"><a href="#cb21-49" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">int</span> z2;</span>
+<span id="cb21-50"><a href="#cb21-50" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb21-51"><a href="#cb21-51" aria-hidden="true" tabindex="-1"></a>    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);</span>
+<span id="cb21-52"><a href="#cb21-52" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-53"><a href="#cb21-53" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="analytical-form-of-joint-predictive-inference" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="analytical-form-of-joint-predictive-inference">Analytical form of joint predictive inference</h4>
+<p>Bayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.</p>
+<p>Jumping straight to the result, <span class="math display">\[
+p\left(\tilde{y} \mid \tilde{x},y,x\right)
+=
+\textsf{normal}\left(K^{\top}\Sigma^{-1}y,\
+                \Omega - K^{\top}\Sigma^{-1}K\right),
+\]</span> where <span class="math inline">\(\Sigma = K(x \mid \alpha, \rho, \sigma)\)</span> is the result of applying the covariance function to the inputs <span class="math inline">\(x\)</span> with observed outputs <span class="math inline">\(y\)</span>, <span class="math inline">\(\Omega =
+K(\tilde{x} \mid \alpha, \rho)\)</span> is the result of applying the covariance function to the inputs <span class="math inline">\(\tilde{x}\)</span> for which predictions are to be inferred, and <span class="math inline">\(K\)</span> is the matrix of covariances between inputs <span class="math inline">\(x\)</span> and <span class="math inline">\(\tilde{x}\)</span>, which in the case of the exponentiated quadratic covariance function would be <span class="math display">\[
+K(x \mid \alpha, \rho)_{i, j} = \alpha^2 \exp\left(-\dfrac{1}{2 \rho^2}
+\sum_{d=1}^D \left(x_{i,d} - \tilde{x}_{j,d}\right)^2\right).
+\]</span></p>
+<p>There is no noise term including <span class="math inline">\(\sigma^2\)</span> because the indexes of elements in <span class="math inline">\(x\)</span> and <span class="math inline">\(\tilde{x}\)</span> are never the same.</p>
+<p>This Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called <code>gp_pred_rng</code> which will generate a draw from the posterior predictive mean conditioned on observed data <code>y1</code>. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of <span class="math inline">\(p(\tilde{y})\)</span>.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> gp_pred_rng(<span class="dt">array</span>[] <span class="dt">real</span> x2,</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">vector</span> y1,</span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">array</span>[] <span class="dt">real</span> x1,</span>
+<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> alpha,</span>
+<span id="cb22-6"><a href="#cb22-6" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> rho,</span>
+<span id="cb22-7"><a href="#cb22-7" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> sigma,</span>
+<span id="cb22-8"><a href="#cb22-8" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> delta) {</span>
+<span id="cb22-9"><a href="#cb22-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N1 = rows(y1);</span>
+<span id="cb22-10"><a href="#cb22-10" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N2 = size(x2);</span>
+<span id="cb22-11"><a href="#cb22-11" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[N2] f2;</span>
+<span id="cb22-12"><a href="#cb22-12" aria-hidden="true" tabindex="-1"></a>    {</span>
+<span id="cb22-13"><a href="#cb22-13" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N1] L_K;</span>
+<span id="cb22-14"><a href="#cb22-14" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[N1] K_div_y1;</span>
+<span id="cb22-15"><a href="#cb22-15" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N2] k_x1_x2;</span>
+<span id="cb22-16"><a href="#cb22-16" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N2] v_pred;</span>
+<span id="cb22-17"><a href="#cb22-17" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[N2] f2_mu;</span>
+<span id="cb22-18"><a href="#cb22-18" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N2, N2] cov_f2;</span>
+<span id="cb22-19"><a href="#cb22-19" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N2, N2] diag_delta;</span>
+<span id="cb22-20"><a href="#cb22-20" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N1] K;</span>
+<span id="cb22-21"><a href="#cb22-21" aria-hidden="true" tabindex="-1"></a>      K = gp_exp_quad_cov(x1, alpha, rho);</span>
+<span id="cb22-22"><a href="#cb22-22" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb22-23"><a href="#cb22-23" aria-hidden="true" tabindex="-1"></a>        K[n, n] = K[n, n] + square(sigma);</span>
+<span id="cb22-24"><a href="#cb22-24" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb22-25"><a href="#cb22-25" aria-hidden="true" tabindex="-1"></a>      L_K = cholesky_decompose(K);</span>
+<span id="cb22-26"><a href="#cb22-26" aria-hidden="true" tabindex="-1"></a>      K_div_y1 = mdivide_left_tri_low(L_K, y1);</span>
+<span id="cb22-27"><a href="#cb22-27" aria-hidden="true" tabindex="-1"></a>      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';</span>
+<span id="cb22-28"><a href="#cb22-28" aria-hidden="true" tabindex="-1"></a>      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);</span>
+<span id="cb22-29"><a href="#cb22-29" aria-hidden="true" tabindex="-1"></a>      f2_mu = (k_x1_x2' * K_div_y1);</span>
+<span id="cb22-30"><a href="#cb22-30" aria-hidden="true" tabindex="-1"></a>      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);</span>
+<span id="cb22-31"><a href="#cb22-31" aria-hidden="true" tabindex="-1"></a>      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;</span>
+<span id="cb22-32"><a href="#cb22-32" aria-hidden="true" tabindex="-1"></a>      diag_delta = diag_matrix(rep_vector(delta, N2));</span>
+<span id="cb22-33"><a href="#cb22-33" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-34"><a href="#cb22-34" aria-hidden="true" tabindex="-1"></a>      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);</span>
+<span id="cb22-35"><a href="#cb22-35" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb22-36"><a href="#cb22-36" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> f2;</span>
+<span id="cb22-37"><a href="#cb22-37" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb22-38"><a href="#cb22-38" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-39"><a href="#cb22-39" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb22-40"><a href="#cb22-40" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
+<span id="cb22-41"><a href="#cb22-41" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
+<span id="cb22-42"><a href="#cb22-42" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] y1;</span>
+<span id="cb22-43"><a href="#cb22-43" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
+<span id="cb22-44"><a href="#cb22-44" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
+<span id="cb22-45"><a href="#cb22-45" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-46"><a href="#cb22-46" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb22-47"><a href="#cb22-47" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] mu = rep_vector(<span class="dv">0</span>, N1);</span>
+<span id="cb22-48"><a href="#cb22-48" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb22-49"><a href="#cb22-49" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-50"><a href="#cb22-50" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb22-51"><a href="#cb22-51" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb22-52"><a href="#cb22-52" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb22-53"><a href="#cb22-53" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb22-54"><a href="#cb22-54" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-55"><a href="#cb22-55" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb22-56"><a href="#cb22-56" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N1, N1] L_K;</span>
+<span id="cb22-57"><a href="#cb22-57" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb22-58"><a href="#cb22-58" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);</span>
+<span id="cb22-59"><a href="#cb22-59" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> sq_sigma = square(sigma);</span>
+<span id="cb22-60"><a href="#cb22-60" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-61"><a href="#cb22-61" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb22-62"><a href="#cb22-62" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb22-63"><a href="#cb22-63" aria-hidden="true" tabindex="-1"></a>      K[n1, n1] = K[n1, n1] + sq_sigma;</span>
+<span id="cb22-64"><a href="#cb22-64" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb22-65"><a href="#cb22-65" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-66"><a href="#cb22-66" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb22-67"><a href="#cb22-67" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb22-68"><a href="#cb22-68" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-69"><a href="#cb22-69" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb22-70"><a href="#cb22-70" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb22-71"><a href="#cb22-71" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb22-72"><a href="#cb22-72" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-73"><a href="#cb22-73" aria-hidden="true" tabindex="-1"></a>  y1 ~ multi_normal_cholesky(mu, L_K);</span>
+<span id="cb22-74"><a href="#cb22-74" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-75"><a href="#cb22-75" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb22-76"><a href="#cb22-76" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] f2;</span>
+<span id="cb22-77"><a href="#cb22-77" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] y2;</span>
+<span id="cb22-78"><a href="#cb22-78" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-79"><a href="#cb22-79" aria-hidden="true" tabindex="-1"></a>  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);</span>
+<span id="cb22-80"><a href="#cb22-80" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb22-81"><a href="#cb22-81" aria-hidden="true" tabindex="-1"></a>    y2[n2] = normal_rng(f2[n2], sigma);</span>
+<span id="cb22-82"><a href="#cb22-82" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb22-83"><a href="#cb22-83" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="multiple-output-gaussian-processes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multiple-output-gaussian-processes">Multiple-output Gaussian processes</h3>
+<p>Suppose we have observations <span class="math inline">\(y_i \in \mathbb{R}^M\)</span> observed at <span class="math inline">\(x_i \in \mathbb{R}^K\)</span>. One can model the data like so: <span class="math display">\[\begin{align*}
+y_i  &amp;\sim \textsf{multivariate normal}\left(f(x_i), \textbf{I}_M \sigma^2\right) \\
+f(x) &amp;\sim \textsf{GP}\big(m(x), K(x \mid \theta, \phi)\big) \\
+     &amp; K(x \mid \theta) \in \mathbb{R}^{M \times M}, \quad
+       f(x), m(x) \in \mathbb{R}^M
+\end{align*}\]</span> where the <span class="math inline">\(K(x, x^\prime \mid \theta, \phi)_{[m, m^\prime]}\)</span> entry defines the covariance between <span class="math inline">\(f_m(x)\)</span> and <span class="math inline">\(f_{m^\prime}(x^\prime)(x)\)</span>. This construction of Gaussian processes allows us to learn the covariance between the output dimensions of <span class="math inline">\(f(x)\)</span>. If we parameterize our kernel <span class="math inline">\(K\)</span>: <span class="math display">\[
+K(x, x^\prime \mid \theta, \phi)_{[m, m^\prime]} = k\left(x, x^\prime \mid
+\theta\right) k\left(m, m^\prime \mid \phi\right)
+\]</span> then our finite dimensional generative model for the above is: <span class="math display">\[\begin{align*}
+f        &amp;\sim \textsf{matrixnormal}\big(m(x), K(x \mid \alpha, \rho), C(\phi)\big) \\
+y_{i, m} &amp;\sim \textsf{normal}(f_{i,m}, \sigma) \\
+f        &amp;\in  \mathbb{R}^{N \times M}
+\end{align*}\]</span> where <span class="math inline">\(K(x \mid \alpha, \rho)\)</span> is the exponentiated quadratic kernel we’ve used throughout this chapter, and <span class="math inline">\(C(\phi)\)</span> is a positive-definite matrix, parameterized by some vector <span class="math inline">\(\phi\)</span>.</p>
+<p>The matrix normal distribution has two covariance matrices: <span class="math inline">\(K(x \mid
+\alpha, \rho)\)</span> to encode column covariance, and <span class="math inline">\(C(\phi)\)</span> to define row covariance. The salient features of the matrix normal are that the rows of the matrix <span class="math inline">\(f\)</span> are distributed: <span class="math display">\[
+f_{[n,]} \sim \textsf{multivariate normal}\big(m(x)_{[n,]}, K(x \mid \alpha,
+\rho)_{[n,n]} C(\phi)\big)
+\]</span> and that the columns of the matrix <span class="math inline">\(f\)</span> are distributed: <span class="math display">\[
+f_{[,m]} \sim \textsf{multivariate normal}\big(m(x)_{[,m]}, K(x
+  \mid \alpha, \rho) C(\phi)_{[m,m]}\big)
+\]</span> This also means means that <span class="math inline">\(\mathbb{E}\left[f^T f\right]\)</span> is equal to <span class="math inline">\(\operatorname{trace}\!\big(K(x \mid \alpha, \rho)\big) \times C\)</span>, whereas <span class="math inline">\(\mathbb{E}\left[ff^T\right]\)</span> is <span class="math inline">\(\operatorname{trace}(C) \times K(x \mid \alpha, \rho)\)</span>. We can derive this using properties of expectation and the matrix normal density.</p>
+<p>We should set <span class="math inline">\(\alpha\)</span> to <span class="math inline">\(1.0\)</span> because the parameter is not identified unless we constrain <span class="math inline">\(\operatorname{trace}(C) = 1\)</span>. Otherwise, we can multiply <span class="math inline">\(\alpha\)</span> by a scalar <span class="math inline">\(d\)</span> and <span class="math inline">\(C\)</span> by <span class="math inline">\(1/d\)</span> and our likelihood will not change.</p>
+<p>We can generate a random variable <span class="math inline">\(f\)</span> from a matrix normal density in <span class="math inline">\(\mathbb{R}^{N \times M}\)</span> using the following algorithm: <span class="math display">\[\begin{align*}
+\eta_{i,j} &amp;\sim \textsf{normal}(0, 1) \, \forall i,j \\
+f          &amp;= L_{K(x \mid 1.0, \rho)} \, \eta \, L_C(\phi)^T \\
+f          &amp;\sim \textsf{matrixnormal}\big(0, K(x \mid 1.0, \rho), C(\phi)\big) \\
+\eta       &amp;\in \mathbb{R}^{N \times M} \\
+L_C(\phi)  &amp;= \texttt{cholesky}\mathtt{\_}\texttt{decompose}\big(C(\phi)\big) \\
+L_{K(x \mid 1.0, \rho)} &amp;= \texttt{cholesky}\mathtt{\_}\texttt{decompose}\big(K(x \mid 1.0, \rho)\big)
+\end{align*}\]</span></p>
+<p>This can be implemented in Stan using a latent-variable GP formulation. We’ve used <span class="math inline">\(\textsf{LKJCorr}\)</span> for <span class="math inline">\(C(\phi)\)</span>, but any positive-definite matrix will do.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] y;</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[D] alpha;</span>
+<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
+<span id="cb23-15"><a href="#cb23-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] eta;</span>
+<span id="cb23-16"><a href="#cb23-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-17"><a href="#cb23-17" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb23-18"><a href="#cb23-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] f;</span>
+<span id="cb23-19"><a href="#cb23-19" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb23-20"><a href="#cb23-20" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, <span class="fl">1.0</span>, rho);</span>
+<span id="cb23-21"><a href="#cb23-21" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb23-22"><a href="#cb23-22" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-23"><a href="#cb23-23" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb23-24"><a href="#cb23-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb23-25"><a href="#cb23-25" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb23-26"><a href="#cb23-26" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb23-27"><a href="#cb23-27" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-28"><a href="#cb23-28" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb23-29"><a href="#cb23-29" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta</span>
+<span id="cb23-30"><a href="#cb23-30" aria-hidden="true" tabindex="-1"></a>        * diag_pre_multiply(alpha, L_Omega)';</span>
+<span id="cb23-31"><a href="#cb23-31" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb23-32"><a href="#cb23-32" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-33"><a href="#cb23-33" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb23-34"><a href="#cb23-34" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb23-35"><a href="#cb23-35" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb23-36"><a href="#cb23-36" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">3</span>);</span>
+<span id="cb23-37"><a href="#cb23-37" aria-hidden="true" tabindex="-1"></a>  to_vector(eta) ~ std_normal();</span>
+<span id="cb23-38"><a href="#cb23-38" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-39"><a href="#cb23-39" aria-hidden="true" tabindex="-1"></a>  to_vector(y) ~ normal(to_vector(f), sigma);</span>
+<span id="cb23-40"><a href="#cb23-40" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-41"><a href="#cb23-41" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb23-42"><a href="#cb23-42" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, D] Omega;</span>
+<span id="cb23-43"><a href="#cb23-43" aria-hidden="true" tabindex="-1"></a>  Omega = L_Omega * L_Omega';</span>
+<span id="cb23-44"><a href="#cb23-44" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Cseke:2011" class="csl-entry" role="listitem">
+Cseke, Botond, and Tom Heskes. 2011. <span>“Approximate Marginals in Latent <span>Gaussian</span> Models.”</span> <em>Journal of Machine Learning Research</em> 12.
+</div>
+<div id="ref-Kuss:2005" class="csl-entry" role="listitem">
+Kuss, Malte, and Carl E Rasmussen. 2005. <span>“Assessing Approximate Inference for Binary <span>Gaussian</span> Process Classification.”</span> <em>Journal of Machine Learning Research</em> 6: 1679–1704.
+</div>
+<div id="ref-Margossian:2020" class="csl-entry" role="listitem">
+Margossian, Charles C, Aki Vehtari, Daniel Simpson, and Raj Agrawal. 2020. <span>“Hamiltonian Monte Carlo Using an Adjoint-Differentiated Laplace Approximation: Bayesian Inference for Latent Gaussian Models and Beyond.”</span> <em>Advances in Neural Information Processing Systems</em> 34.
+</div>
+<div id="ref-Neal:1996" class="csl-entry" role="listitem">
+Neal, Radford M. 1996. <em>Bayesian Learning for Neural Networks</em>. Lecture Notes in Statistics 118. New York: Springer.
+</div>
+<div id="ref-Neal:1997" class="csl-entry" role="listitem">
+———. 1997. <span>“Monte <span>C</span>arlo Implementation of <span>G</span>aussian Process Models for <span>B</span>ayesian Regression and Classification.”</span> 9702. University of Toronto, Department of Statistics.
+</div>
+<div id="ref-PiironenVehtari:2016" class="csl-entry" role="listitem">
+Piironen, Juho, and Aki Vehtari. 2016. <span>“Projection Predictive Model Selection for <span>Gaussian</span> Processes.”</span> In <em>Machine Learning for Signal Processing (MLSP), 2016 IEEE 26th International Workshop on</em>. IEEE.
+</div>
+<div id="ref-RasmussenWilliams:2006" class="csl-entry" role="listitem">
+Rasmussen, Carl Edward, and Christopher K. I. Williams. 2006. <em>Gaussian Processes for Machine Learning</em>. MIT Press.
+</div>
+<div id="ref-Riutort-Mayol:2023:HSGP" class="csl-entry" role="listitem">
+Riutort-Mayol, Gabriel, Paul-Christian Bürkner, Michael R Andersen, Arno Solin, and Aki Vehtari. 2023. <span>“Practical <span>Hilbert</span> Space Approximate <span>Bayesian</span> <span>Gaussian</span> Processes for Probabilistic Programming.”</span> <em>Statistics and Computing</em> 33 (1): 17.
+</div>
+<div id="ref-Rue:2009" class="csl-entry" role="listitem">
+Rue, Håvard, Sara Martino, and Nicolas Chopin. 2009. <span>“Approximate Bayesian Inference for Latent Gaussian Models by Using Integrated Nested Laplace Approximations.”</span> <em>Journal of the Royal Statistical Society: Series B (Statistical Methodology)</em> 71 (2): 319–92. <a href="https://doi.org/10.1111/j.1467-9868.2008.00700.x">https://doi.org/10.1111/j.1467-9868.2008.00700.x</a>.
+</div>
+<div id="ref-Vanhatalo:2010" class="csl-entry" role="listitem">
+Vanhatalo, Jarno, Ville Pietiläinen, and Aki Vehtari. 2010. <span>“Approximate Inference for Disease Mapping with Sparse <span>Gaussian</span> Processes.”</span> <em>Statistics in Medicine</em> 29 (15): 1580–1607.
+</div>
+<div id="ref-Vehtari:2016" class="csl-entry" role="listitem">
+Vehtari, Aki, Tommi Mononen, Ville Tolvanen, Tuomas Sivula, and Ole Winther. 2016. <span>“Bayesian Leave-One-Out Cross-Validation Approximations for <span>Gaussian</span> Latent Variable Models.”</span> <em>Journal of Machine Learning Research</em> 17 (103): 1–38. <a href="http://jmlr.org/papers/v17/14-540.html">http://jmlr.org/papers/v17/14-540.html</a>.
+</div>
+<div id="ref-zhang-gp:2004" class="csl-entry" role="listitem">
+Zhang, Hao. 2004. <span>“Inconsistent Estimation and Asymptotically Equal Interpolations in Model-Based Geostatistics.”</span> <em>Journal of the American Statistical Association</em> 99 (465): 250–61.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Gaussian processes can be extended to covariance functions producing positive semi-definite matrices, but Stan does not support inference in the resulting models because the resulting distribution does not have unconstrained support.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>A boundary-avoiding prior is just one where the limit of the density is zero at the boundary, the result of which is estimates that are pushed away from the boundary.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/clustering.html" class="pagination-link" aria-label="Clustering Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Clustering Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/copulas.html" class="pagination-link" aria-label="Copulas">
+        <span class="nav-page-text">Copulas</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/gaussian-processes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/hyperspherical-models.html b/docs/2_39/stan-users-guide/hyperspherical-models.html
new file mode 100644
index 000000000..08c37a052
--- /dev/null
+++ b/docs/2_39/stan-users-guide/hyperspherical-models.html
@@ -0,0 +1,1309 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Directions, Rotations, and Hyperspheres</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/algebraic-equations.html" rel="next">
+<link href="../stan-users-guide/copulas.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Directions, Rotations, and Hyperspheres">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/hyperspherical-models.html">Directions, Rotations, and Hyperspheres</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#directions-rotations-and-hyperspheres" id="toc-directions-rotations-and-hyperspheres" class="nav-link active" data-scroll-target="#directions-rotations-and-hyperspheres">Directions, Rotations, and Hyperspheres</a>
+  <ul class="collapse">
+  <li><a href="#unit-vectors" id="toc-unit-vectors" class="nav-link" data-scroll-target="#unit-vectors">Unit vectors</a></li>
+  <li><a href="#circles-spheres-and-hyperspheres" id="toc-circles-spheres-and-hyperspheres" class="nav-link" data-scroll-target="#circles-spheres-and-hyperspheres">Circles, spheres, and hyperspheres</a></li>
+  <li><a href="#transforming-to-unconstrained-parameters" id="toc-transforming-to-unconstrained-parameters" class="nav-link" data-scroll-target="#transforming-to-unconstrained-parameters">Transforming to unconstrained parameters</a></li>
+  <li><a href="#unit-vectors-and-rotations" id="toc-unit-vectors-and-rotations" class="nav-link" data-scroll-target="#unit-vectors-and-rotations">Unit vectors and rotations</a>
+  <ul class="collapse">
+  <li><a href="#angles-from-unit-vectors" id="toc-angles-from-unit-vectors" class="nav-link" data-scroll-target="#angles-from-unit-vectors">Angles from unit vectors</a></li>
+  </ul></li>
+  <li><a href="#circular-representations-of-days-and-years" id="toc-circular-representations-of-days-and-years" class="nav-link" data-scroll-target="#circular-representations-of-days-and-years">Circular representations of days and years</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/hyperspherical-models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/hyperspherical-models.html">Directions, Rotations, and Hyperspheres</a></li></ol></nav></header>
+
+
+
+
+<section id="directions-rotations-and-hyperspheres" class="level1">
+<h1>Directions, Rotations, and Hyperspheres</h1>
+<p>Directional statistics involve data and/or parameters that are constrained to be directions. The set of directions forms a sphere, the geometry of which is not smoothly mappable to that of a Euclidean space because you can move around a sphere and come back to where you started. This is why it is impossible to make a map of the globe on a flat piece of paper where all points that are close to each other on the globe are close to each other on the flat map. The fundamental problem is easy to visualize in two dimensions, because as you move around a circle, you wind up back where you started. In other words, 0 degrees and 360 degrees (equivalently, 0 and <span class="math inline">\(2 \pi\)</span> radians) pick out the same point, and the distance between 359 degrees and 2 degrees is the same as the distance between 137 and 140 degrees.</p>
+<p>Stan supports directional statistics by providing a unit-vector data type, the values of which determine points on a hypersphere (circle in two dimensions, sphere in three dimensions).</p>
+<section id="unit-vectors" class="level2">
+<h2 class="anchored" data-anchor-id="unit-vectors">Unit vectors</h2>
+<p>The length of a vector <span class="math inline">\(x \in \mathbb{R}^K\)</span> is given by <span class="math display">\[
+\Vert x \Vert
+= \sqrt{x^{\top}\,x}
+= \sqrt{x_1^2 + x_2^2 + \cdots + x_K^2}.
+\]</span> Unit vectors are defined to be vectors of unit length (i.e., length one).</p>
+<p>With a variable declaration such as</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">unit_vector</span>[K] x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>the value of <code>x</code> will be constrained to be a vector of size <code>K</code> with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.</p>
+<p><em>Warning:</em> An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (<code>lp__</code> in some interfaces). The underlying source of the problem is that a unit vector of size <span class="math inline">\(K\)</span> has only <span class="math inline">\(K - 1\)</span> degrees of freedom. But there is no way to map those <span class="math inline">\(K - 1\)</span> degrees of freedom continuously to <span class="math inline">\(\mathbb{R}^N\)</span>—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses <span class="math inline">\(K\)</span> unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.</p>
+</section>
+<section id="circles-spheres-and-hyperspheres" class="level2">
+<h2 class="anchored" data-anchor-id="circles-spheres-and-hyperspheres">Circles, spheres, and hyperspheres</h2>
+<p>An <span class="math inline">\(n\)</span>-sphere, written <span class="math inline">\(S^{n}\)</span>, is defined as the set of <span class="math inline">\((n +
+1)\)</span>-dimensional unit vectors, <span class="math display">\[
+S^{n} = \left\{ x \in \mathbb{R}^{n+1} \: : \: \Vert x \Vert = 1 \right\}.
+\]</span></p>
+<p>Even though <span class="math inline">\(S^n\)</span> is made up of points in <span class="math inline">\((n+1)\)</span> dimensions, it is only an <span class="math inline">\(n\)</span>-dimensional manifold. For example, <span class="math inline">\(S^2\)</span> is defined as a set of points in <span class="math inline">\(\mathbb{R}^3\)</span>, but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by <span class="math inline">\(S^2\)</span> in <span class="math inline">\(\mathbb{R}^3\)</span> behaves locally like a plane, i.e., <span class="math inline">\(\mathbb{R}^2\)</span>. However, the overall shape of <span class="math inline">\(S^2\)</span> is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (<span class="math inline">\(S^2\)</span>) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.</p>
+<p>Even though <span class="math inline">\(S^{n-1}\)</span> behaves locally like <span class="math inline">\(\mathbb{R}^{n-1}\)</span>, there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at <span class="math inline">\(2\pi\)</span> radians in natural units), they do not produce a smooth map.</p>
+<p>Like a bounded interval <span class="math inline">\((a, b)\)</span>, in geometric terms, a sphere is compact in that the distance between any two points is bounded.</p>
+</section>
+<section id="transforming-to-unconstrained-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="transforming-to-unconstrained-parameters">Transforming to unconstrained parameters</h2>
+<p>Stan (inverse) transforms arbitrary points in <span class="math inline">\(\mathbb{R}^{K+1}\)</span> to points in <span class="math inline">\(S^K\)</span> using the auxiliary variable approach of <span class="citation" data-cites="Muller:1959">Muller (<a href="#ref-Muller:1959" role="doc-biblioref">1959</a>)</span>. A point <span class="math inline">\(y \in \mathbb{R}^K\)</span> is transformed to a point <span class="math inline">\(x \in S^{K-1}\)</span> by <span class="math display">\[
+x = \frac{y}{\sqrt{y^{\top} y}}.
+\]</span></p>
+<p>The problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. <span class="citation" data-cites="Muller:1959">Muller (<a href="#ref-Muller:1959" role="doc-biblioref">1959</a>)</span> introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.</p>
+<section id="warning-undefined-at-zero" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="warning-undefined-at-zero">Warning: undefined at zero!</h4>
+<p>The above mapping from <span class="math inline">\(\mathbb{R}^n\)</span> to <span class="math inline">\(S^n\)</span> is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.</p>
+</section>
+</section>
+<section id="unit-vectors-and-rotations" class="level2">
+<h2 class="anchored" data-anchor-id="unit-vectors-and-rotations">Unit vectors and rotations</h2>
+<p>Unit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle <span class="math inline">\(\theta\)</span>. Given an angle <span class="math inline">\(\theta\)</span>, a matrix can be defined, the pre-multiplication by which rotates a point by an angle of <span class="math inline">\(\theta\)</span>. For angle <span class="math inline">\(\theta\)</span> (in two dimensions), the <span class="math inline">\(2 \times 2\)</span> rotation matrix is defined by <span class="math display">\[
+R_{\theta}
+=
+\begin{bmatrix}
+\cos \theta &amp; -\sin \theta \\
+\sin \theta &amp;  \cos \theta
+\end{bmatrix}.
+\]</span> Given a two-dimensional vector <span class="math inline">\(x\)</span>, <span class="math inline">\(R_{\theta} \, x\)</span> is the rotation of <span class="math inline">\(x\)</span> (around the origin) by <span class="math inline">\(\theta\)</span> degrees.</p>
+<section id="angles-from-unit-vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="angles-from-unit-vectors">Angles from unit vectors</h3>
+<p>Angles can be calculated from unit vectors. For example, a random variable <code>theta</code> representing an angle in <span class="math inline">\((-\pi, \pi)\)</span> radians can be declared as a two-dimensional unit vector then transformed to an angle.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">unit_vector</span>[<span class="dv">2</span>] xy;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=-pi(), <span class="kw">upper</span>=pi()&gt; theta = atan2(xy[<span class="dv">2</span>], xy[<span class="dv">1</span>]);</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the distribution of <span class="math inline">\((x, y)\)</span> is uniform over a circle, then the distribution of <span class="math inline">\(\arctan \frac{y}{x}\)</span> is uniform over <span class="math inline">\((-\pi, \pi)\)</span>.</p>
+<p>It might be tempting to try to just declare <code>theta</code> directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values <span class="math inline">\(-\pi\)</span> and <span class="math inline">\(\pi\)</span> are at <span class="math inline">\(-\infty\)</span> and <span class="math inline">\(\infty\)</span> on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.</p>
+<p>With a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.</p>
+</section>
+</section>
+<section id="circular-representations-of-days-and-years" class="level2">
+<h2 class="anchored" data-anchor-id="circular-representations-of-days-and-years">Circular representations of days and years</h2>
+<p>A 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.</p>
+<p>In human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Muller:1959" class="csl-entry" role="listitem">
+Muller, Mervin E. 1959. <span>“A Note on a Method for Generating Points Uniformly on n-Dimensional Spheres.”</span> <em>Commun. ACM</em> 2 (4): 19–20. <a href="https://doi.org/10.1145/377939.377946">https://doi.org/10.1145/377939.377946</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/copulas.html" class="pagination-link" aria-label="Copulas">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Copulas</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/algebraic-equations.html" class="pagination-link" aria-label="Solving Algebraic Equations">
+        <span class="nav-page-text">Solving Algebraic Equations</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/hyperspherical-models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/img/Figure_DDM.pdf b/docs/2_39/stan-users-guide/img/Figure_DDM.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..381d4b538fde6f9c7ce7248a0d68a842f69082ce
GIT binary patch
literal 100560
zcmV)9K*hf$P((&8F)lR<CD7Ex*a|ivFd%PYY6?6&3NK7$ZfA68AT}{LIW!<3ARr(L
zFGgu>bY*fNFGg%(bY(<kV{c?-3O+sxb98cLVQmU{+N`}z(q%c0U^lN-_y)?2{~%V7
z<CX?mKo3fTE}4}k+P-i=fbWY(@8}+cS*Az$2|kjSmlp&`5GcR?<5&6huYW85i+@)8
zv;3;Rdig!;`qkU-wfe9B{-6K$U%#x(zkYeae=EPc?)mHg`vtcD{kI=~*56h`{Qu{_
z|M&mQLjUdg&0l^WJ^1(adyilLUiEirN)P@Q73c5#Xa2|E_PT;A_`mXh<#$zWUcjE7
z0kJ>-^`m5MzsuaZyxIT!@%Qn&&1zNuk6-`R*Y7Gzvq=8+M~PK_Hy!rxQRM#pN0~i-
zkJ@Bsnfv!2Wop^KXD`xjcP{u@idQ)Hy3KvS&r);!mbE@lmVf_QW|!Z4k2!9k3x1Sp
zW&SRu4arg;@Uv9Y-?f!n)&2X=GPD0~efLkH3x1Xo`Q4|@(JYjI|Jn0=Pr80<DP#QQ
z<*ARyl%H?!XTi&Ul)LZjXTcBrQS!dIpCv!=N6F9Y`%5=J^k>=o68}>6Lw}adclnp1
z7yc}HU+7<oe&~;~pV#`c><{|0Y`)#U6ut0g$@_}`QuITAmd!W)m!cQ`DEWW>y7KLc
zpZ}fjJtd6Q%WP`nd>{Yg{Fi3Trr6K_nkY(){d>&P{!-+!C?nQSnZ5lib6J%6`G8sG
zGDbqTy`ehqH*VAL_i9=s`}-;Jqt~BOKl=SC`Lp+*qCflpDf{y&K4pJC$EWOHUe_)A
z^Ld{4_2WhSWjy{|?=vF*<5PqE>;LTy{MTO=4HpLb>)*SrdjETi&0Q1hkNRT){ck^3
zu&DJ``g8K#g%;@||8&by>)*EQ;%ffWEk~_?+wu{A|8&by>)*C~r0GBYo=>fR-?Gc*
zfBZ#fwSV8X3;TckRcEz--?pzF|NM|gwSU{Ttw#TJyHxr2ZL+M@{g1aB)&70kzE1x0
z&m7hMecQh9{^RdCtNq)yqs{V9wfdvFziI9)n*VH@|Ng1;Z(8;>_#bZg*7-N>p7s16
zZunE_-?Yry{tvgCRX!TVqdWXr#rRf4Os)98e*N43b2LkD=>PqXXz~B}*T*(6ouJlc
zbsdAD&sn>xAET5Srp2gpv|qLU)+Lhm#*ir$xOMI<Ir}OHwqfRaKwaIT%qALeO;h%b
ztzEN5sb<xHYROX8oCz6YbCjNNcY8m^hLoBX>{bykXx(nXtY4<c8SA}-nPl&%+RuS)
zA-@93KE0{hO2XQ)#(=H23tdAuSf&j?z&6zeuv&{|>(qd~V%!7G>gx%2sROuY+4=J^
z%IH8blOo|PMGmZtUO*d~23$kWTQ<0wExN$D+t!4;uIjL^nNWsJG-IYP&lPN!=0EOx
z>}+PQ<~_GZ=-G-k8v|o)j(XYd#bMogH1Ng5>HWYqV$^MlmA)L76|cwG>S}uSZ`D;+
zu+~O%*6rVtnXPA2-FmfG*4Kxbm9SznF4!ouQcgze`P;GPfGx4f5Nx(716HM6wpTG5
z?f0D2d)OMQ!-h5uB!%54L%rVTHLF$482;UtN>;a<*nc-xY$%g$nJ_a(C4Z{}o7;v@
z%6YPG*6OBIZR?V?ZR)H2Gv8#l|Gb4_r7RiA_SXGtQ)x}stn)Bye79h|sJ$0sR>w@X
zZ0Z`5rJBDT-M$rAwJ#x=78}oPUvDqjmpO~T>?0^ypM6_qNuz~I%ik`koi>K(ZC83-
zlpFTCvp!LqcE-BByZr5*&8F#QKqa3;O>FU7vp+Ag)+kohI^%5oF1^<Z>-u5L>;1dz
z5wCIlt~QYrW}_0Q*)WRL*~dy^b4y`$U)K50VzaYH&kUJl6<E=h%;vC-rkD&Inm@Jq
zIw&Wle3*$<>N6$!{&sb~5U~;N*|%jfb%I8Sdt+iE6=sc(d@K_K%2N#MWj_5DX|TRq
z&2m;VX7lfEi%T}EHle#Q(|JN{PqwJcrV*3^-N8(b4p@~D?HUs%s31ENwXOwJVw{Yq
zQomX6NxHo^{bj3|l$q6=(Av&gX+4>VEjCr9WbHN(6Vgh{%9e>xF(#7Y-<U~IXZGkT
znEiW)*#y1MBsOph)_P3p`FmSWU?x67z)V|i5mXDIvh%mPn3&5Zx~$XTWzX1r8`$R0
z8p(FA^ttaEbIz<}tGdJNA1zobOfCDTJ?>>|_z2jbYPQjviC$y&`P4)<PuYU51EcJ)
zF$Tj~O9}C3y)To>OqgmmFzxH3n2n)kF?+84%b%}ZWP;OKU5UinGGgz4=96xdT>@+C
z!%S|%Y@{U{Rg;?w)>DKT+ssURa~h4>gfX$nPuW|PU}p1Hnq4K=Wn*g(7|Uk<*1B%%
zs<A4oCfizQV%Uq#`ufmLITQ8qvh5?X&$Joqd(UiRn8d8GG2<H+_D37P?WJF&4Dm$k
z7|GW3p=&0sdLWavCRh1<WM}Vex#*NL>3Mr%Glu~vSL7N}X7BWusKL}kBh{7KV?0g3
znb>3>-DKrVWMY%8d|P#EplVW+QdVosYosdbiLv=%W{ZildVgw7`Np|PnVBtg|746=
zwfR#-65HY?ph}iAwML!InbplX8$Vs9FXT^`DZiX5OdF>BsVVI?WX8J2*KSK!w}l91
zT3416K0uSlCO;GHxi}3NHofzgHlLMK5_{D-aHe54Mybyl^2*j<dwOT~%qy9RO+R39
z-Sx0)eRDRL28T6YM2wlLEy-+R!|KZQ+LVay`<(VlY|19HubyEq8Y+S5BsTx}DH>K5
zGsD7%v$U~$vd!eQwbJzVEN8;RZCW*LDa+ZSWRK1=tJypIVthl(rhgg1Yz-OMFoted
z%s3Hel|||>9VVHL*>fFrN(wvkv?^ndp2%bjJCyeQA8krdBpb+7+AL){yzf1j05}~n
zX2NectFgcuvk&m>I_^;GVfL`T2Rbx=I_$U0?Xb18Z`P*;8?n+GnkG2ebr|z}IfhnY
zavEx}Ucqeg@=KdKJImRp;G08z$`R|Zy{a>NVhLuA4MqvuEFV@~$1`U0KYMJ3SmC#=
z_joX~NSbL=Fy3}Drjx0L(vG@j$Ocwv!d}}^b&c8RII@+zCt90lEoT}UFdIl1Uwh&s
zhG3hi*?o1(x@?e4T{Yd-W@vU}n&1@J8n<iHKClhzbLMZC3C;;a=S`-p-ljj>%*o%1
zTTjM@=3{N^`~bCMPl9X`fkaX54zs1v#6MbleCsgNES#A_be^{PQ#M#FXE_nKlNaTz
zZ!sqtBWvQnVb1ns_EH+ZUG^4xGQL+s_HAadK0D7pV;wRw_>>wm^DL$y@ke0Hvm4V|
z=Z8a&>JKup$rzdDKeI)fh0WQRJD<w_^Yowcz<~tCrJ=KGR<})T(pyX#NoMcP*%*>f
zU;@Pcy<sOFy+Ld$m{moOo2@-7nTS7qlM?mWT<fwJq-HaIBvYm)R^U<e4B2<%V`*z!
z&nCJX)QtT&s{~9;=)i1d>|mq&tEu)~T!eXpIXX(BX<r}3ifQXJnD@-4on~*B2~9?+
znD9>~1BussLCq)FCN$+2Q>v6Kw|%4GY@XzZn7M3I$k><{ZB9rqwENK-#pE_}1!k65
z4OUHP^3%P`(O_ZA^g-C%E13<V$^Oc2Nqn!yY!IoB&4C@v6ir)_?3>bjG+POaeF`uM
zEfltft)+Z>pZ=NfW9Pz`_3Yo=X1Ty@+-ESR4qtK(lqmxCfhjDs5RiRCG_Z2wm{!4J
zfLI8l478VzrpeCig}QQEF^@ugPdAJiCDYZ91rA#n{1uHNzLM#a!+&em=1OKl(-~S=
zFtgf~5w;0b^S35?8*SJkbaZ;yLNzqDS{56Pg8>BNsM(cadX-n#R>>~2ATe+@HCvXm
z|H>|qeVs=|0tp*)hFtO-CbrP=Qxm0GOiYGC>_uirY>8$xdsuyf)?WQ*<<m3!80bG*
z@+oB|fXPh8%|73F^TxD!W%fqn0nDN>DaE+ipGTuNW~LoGYc?BeNPc71b7v){1LjU|
z{IqTjIcLm7rZed)e`@N<X51=TOt4vpk7YIe?J^UaJchmAn)R9XEMBfw&32cf$=Et+
z5;xm0bA#<l69`rFx4WLXX;bLdXLgA)dNz}9ySCCaXKeH((?wVKY|l^4fSK%u*(A*c
z!fX{(CR^AwHnc$Y`GIylX0;(3TKDBKSrSvum?Edx?(CXB?d%1@a(lp-t@jg$0p&d=
zGqLHvH4`IyV%1dknh{~y?lif{Z>_XseQT=sZ_CguXL?5Ax7&u{vRyjvFt%8)Y<HWw
zjGvGRr<m(Pd+PPIT#1Y=?23YUake0Y9hy3wm0V5O$*DZ9Jx#pE&2*?V!4`54hHAu5
zg~c)1YK(xe!yL$#Z}sZz^UUgOe0T`^Dq@0~v0{>wmthaxlTCZ`{?iiAU}&)wZhE2a
z5HV(~H0GzV9R>`{1SdVQWuoP?+6q_vw`=XJ@9cT{4*I-o9hW2aEVDjkTN-9_GV4=&
zF^sXT7ctGuUKu8FVReP{Fnh%nF!bBhh!vZpGI&#SK{0!U9WmIJ&;S(C!DCGH+dydM
z1&wdy4^7B*M<v!~Lu>O`X8Dcm#%$Y1bKVz${h}O(p#}wEGu%s=FiIv^VXzuoL@+7t
zfig?1CT6L0UL7dxGMGtDIx@7~jwl-JFSiPgKb%=AsDmI`r*AOVzY33{$;@oq!&z*j
z!t~6|ksB_%xEqcC)?P=J>nu+&lZ7i;kk+y%qx;lE{AHn`K1J-4zim#wYiIR{j?F9g
zw`W4)cGqkTAIz@%>=<n(y{&%ETFDlKA%X2FS<aTe?!Pq~y9P5^;-hKZp4qI*bcxlm
z*_|OVIdp)9mAhRkyTs~DlVDbzyDIisMlw_RU1FJTdcHcCpg9+;=QFD}i{X)|Y_26z
zlTM%(>$)*hn0s{=TH;0**#3c;s#~+ViJcD`J<;NQVt4A%IzZ}QVK4Jc#ElANGTm9s
zy3K)E0olrU`(YYhS7+&kiNa`05Nyk_X|^e#COEgV9<3TP!RbuGme?r5b^325ECSfB
zKAe7FaZnp<d>1UhnCzzEm*Z?^YNqB-tugVRVsQr>zFo>>-Zh&wOSU&><LJ{H6UWSO
zXR(#Ev5jn&!^YlOZkajkza8`bR_rf__1lJQpS1SEMk4FHi>rod5KN1Ei#D5z`Ls60
zd`!%cJ`aX9Ys;N8j^KiAglJ*=@CfmZ1<jWVi_O)I@xAt8BL)S-U8=WX1#2)<cgA65
zXB&NEAQUzPV~&}@;yWBLx$T<#a-r4giuq}59hX|{ifnhat%n^^3EFzbv$mFxYs<F_
zEmI3?GP8wzhPGO2cA3p~|E=kM*=04=@i2}6f|>mEp(>#UZZl+j#crWc7_hdy%@Su?
zAn|3=(6MU_HG9LSsu`nX%w*?2EQ^@|GXjnE<*p3NMu+Li@>On{@LYeb&g_V_zFJ*u
zD@O0qBU;cn;~~<w)&$I%VkHh2jcr^0f~_oPZzcv{_iWb$)i_MgZtZ60cFK8S31W(r
z_#V!_aJ<u)!p8JGfh|5gP@Cti)Y#)TWP1G>N`z`MIUlvHn;OV|Am^v)Lu}0C=9%Yl
z25)T6eTK5ttj%n-Bd%J%KV2p^51T!k5R;qdr(t%$RNdzRJ2u%i$>~GXLnk(--Ua`J
zXSQp-8}X%|jj6rxXpqC$i#^Py_Yu3Uj|Owo^2nDS%pPfl9Fohn>mfGvqAO;yjOOa9
ziM5)|j&deRtbPmA&_@n+D3BRWzL+9-W-=3-&Q>qk<QkYluyr3nOl%$&rW0(9b|if}
z7VPF^Y=ra=+aI=v>s+5zVzDK0zQf()GdFvsOjlRi4l@b1rfVWiLnq*{ZyjM3dLC%6
ze$dX!Y$b8oZvr<|7_(Q(iiHzxJlGPKRlSz1ZJZSc9b6M_IcupcB-xKSc0CLD*}X7i
zlJck~P1a>5GkpQ?aR9k7V74B%RyJYt)n$JlDM&G3)MbAktK>td2~9bw$-c~l=3z1a
z%~mm)d9<I{Hh}5*uDiR%9A;LIuccjzBr_T9bF)hffFY@8ICqWdYQ{>hdRT0(!IWJ!
zl(^XB-P?Dv50jQn%*q)$N@18>F#9SGOKD~1O7|CQws%M!L!3{8;(*@}W>r1P=}cz!
ztdHkFb4n&-9wxU1HMzN6Bkny1&%O`u$Oy}MoBe{*hAYz{!SPI-ZiAT49nYZLnQ&9e
zbn}qRW~%gc^kf;gI@7?8`oHrnDhjAef17MEqs&~X;--|T+B@qr1J1i==ETT$r|BQV
znbhnb`k$LWHT8e9V&-<~KU*A-8<Xl2YPNjc7%^JfF@I)uZ+CTt)mFNhk73i;t*P*5
z+1M~Xv+X?`6JE_;X=ip$wyKkD`i}i}-m4*d#mATuIUBCZB2|3>uwHAHGxf$Nz|?j$
zV&#LY>o7f9o%&Su&aEONn;j-Pow0)(Y?kILz_Mi_e`{Uc$Epq8S^4kXHP+SlZI?~@
zu!vih<9OHE4Cgdu7h$*ouvYu9HT)}E5*G`UT@zP;ZSgw$j-Wl<qH@&>WERU&O-&S^
zWx|ks#M3tfLX3ls-gq5rl>M>gs)?`%o@_6dSva+<U?y*T1+cH=mb0}hwkzhXZ<z4R
zZH*d?X|#Ozqse6^KAmyk7HpaLbhgWF*R&a50ZbMiSqWR)7gTW@>acCgfp7MdoIcnw
zdNzP+IC?2|jJiFwZK*uvuwzuurqGCXjY;uEV2?O+F;@FgC-*#S1xSmp0^x3Dq#f;l
zON}YPF-7$i)$A&{?5O5jOa5@+xaUaaGWULCODufBj7|K=eCuo{!shZ~X(Yg;dkY#H
zY|1%P_qn8bdZ~U!<x87R+4)f0eOi38Px80w#wI4TtZYKv<AV7Zn}TZg3jPbepC~ao
zJBP*1H!)K&H&%Tw9Jff>`CC&5efPh#*kpy7)({(;Em`k==upPa7P@uSTL@<u<*-zI
z-&AXo$(b2U;-DTT;ye8X_tjvlGd1~uVxJhM+jsxXd?;C;aN^*5z4a{F%E|Ng!<3pT
z04%n|Z3ORQsc)>XtI*FH3yTd~TUMi8>=zU>_M+!mij1Ld`+l;!O9tEY6*tik_UdQ;
zzW;Bk{#l@n^>TLJG4(F<-59pdON<pR3Dpl%8s<#0&FNkB4Wk-tn&AzKT|I6w6+fFW
zGjnI5g+b_h`zw){<aDHR>k@=$byvH5tc?laM+c?mkf^&Yu&Gkd;sjg6PXJ8x#_O5*
z9{6cW51R|bRxoIVF=ed!sXE6RaY|;)3@|@T={r^@{AcF&^OaI>3!NPZSZx;}!u@<D
zG(}?rnFzB*KiPCK#o0_8lfahvVTx&X$M!Y0nEx<|lT(;YC}%r6XJhR@^p*A?S0Xq7
z&rf~g$104uE%dj36D#Lvs<+z@-G1UUM;L6138Uv^YQ|}mV#3%r4hz*BB<JPNU5|3A
zIErE5`XSk^jz(cC{NoJa+~Q|-b(dtLa*pn&^CO!WCO4gpcKpwbS?B#QWz8_TVWtZ3
zBG!rxVq+ZeF;1q4-TI`S<3R1VSrdAGmLi|t+1fk(46^1zZ_FgCv;KKl^;4<p3;r1=
zvscbT*b*3e1D=R%mave;Gnl>7*$S<=sl-edyh+<WC!|48X8jzsNC<y3SL*bm-`&Dz
z0w(Nba0dJ72RmQR#`-Yj$^qxMao}#u)HFZ)-M0K^?@XQTe$HzJYci2@_N%#u3I20_
z_G@c(&H8MWb{V0j^5=7G`<NJF?P1xg`?mUcavZ#q*-Uq~W~|_tF+M%<5xvNlOisU}
zRg4%DryP=(@KiI_8aL=UIFAiR*e>%Nu!mQQu|A5sCX{eWGGw29b%frM%w6hG6~Zmi
zm@RbfnP6XEyo~KmK+ZJWXpPzQ?J$$Zn$7a@O!pgzy(yS0$PG<6G(!BRdvi9!0Pt%W
z>~iDVF88-)scd&w*frK7{I)r9MQ@og-|_EuR}Y&$v7X`uKHN8fox_oAvyqIK$%?NB
zpK^DIXtUfF{Bv(?*tUyIcB;E!?3!$}<AtHEWTR#=4EG30UwvaHyG6p&LxY*ZTBxq~
z?3#V#KAL^(`%6<pm$MZvSg_5f)co_vr{5+!ooOnaX0a3w)A8&d*#zgbjMB(U6{Z~!
zkfHMY!>s#u;l@P7|K(@}usYPee{9N}uN&AZS4ME2iM8FK>FbR0*x|1W?A<4jjk4Zm
ztiSGMGHa|Lz%l*pnm{~a8s>tPo$vMe@f5zXhJ)&vy1<1OCVoM&Dd)_@s2AjrG@;p=
z?p3u)C@MgEweAXJ0$guN%vN_lvUhbK(G~)qB5R1=UMivhxWb977QgG_eYZJKdO{Wj
z-h^p(87~0+h%<J-C{b5JGX{@;RSUh_q0WAsuEJ;m;tJ+oS2J$$0IhBJ3NkUbZk@xv
zNJh!b;XV@L#Nqy4<MZpO?kr{Nu-vY%d1kJ?v2nZZ3mU#p`~DVH#D6@S-ahUw?zgs=
ze4b{TaCIlO9X_u>o%?vMspQ0Rq31#k(X1DW`T%n@KNFg<;vQEz>Jj!1mA#eYdUbzp
zwxdr>m3rH;wYc*<x^vJPw~rQHgQe8`c=$ewI;<<-I8(g3Ut`JRZil9=;U-u{eiHb&
z;hDrGl{(smws^wWhD;{oRH(7%12ZXs_h2nQTIQSu_e>6}?vGuGz-KIuMxAZ8d$C~Z
zil?S)$YO&l7FS?3Ve9RrpS?qqw(c6KE5DP@sg3(MHfz0Kh-)jb#U>fBkLfp1n>|)b
zXy)2-PtEyheApu5EVp8TC_w(}E&73bxf5;=Ar7-C@1B{x?oTsP#NBW?xtJF&Q+9x3
z@B2I(A7tPf^*;Ib5wcl4?jxEd>MpjOyF&*KZX&q@><8crfUoQ9mzlJoD>jK%zsw}B
zxSwp?!U49S?uu&-q^QGfy14(SEjgc`cPYmkw^Y3)&7fGvMOXT*b7=9gH90Hy=hI}b
z-whtSUTE@uddbyuRt@fZxQ8?6xVNN9O83gOx7I&Un-4DTSSumh4?O#7pDJw~axdpK
zN`51F+IM%%HJ!c015K_aEiZ{r=7qS<a!PmQtQkg=TYJ3qcFQ%sN}Qy_L(!R>xqEgO
z0=mBJ9@hF{$~wo48RK?`%~OY>df}|xY2ZfI%txH~tG4CN-LWqZG`n;1b1rz?9xYuy
zzI(*}I*?nV4yUYJa&vVED81bWHw#{-dz){$U9#oL9iPqm?6-Aw=TLW5&$X4yESS07
z{fm>C@~Fqpz1GV_U0Sj2s|j0oPj8dzO)gWm=DJ^t+6nt&K_1+&828uXL=GxVZ!7o+
za#6rE8M$4=O3~x8pL5Nuqk^e#P-ecf#kD2BXJF!U-nONoE8`I=ZCq%wmzx1KU9y4p
z2FAHUu<uxyo+b3M=4PlvY&^ezpt%fkKMMyFzU!GYOSrsrf*O^YpP8_=q8<&2`wIze
zFq(&e>Mhq2K`))vFGTEo6<j_Weubm4^>g7ync_Nyette8%9Fjwk&H{Ner2PUPv6bj
zQ@@~r=R!ifpJ1ca`#fyop4%O=Z!>};4=|lqfZI3nxf-JF15Nw_aizJ9+j8!=IR>Z3
z7j83kfIUUtW(%!*!TiiN_nEwtSc&_>#fkoKmx>Ep<2)W|sO1B#AR}bsBR858wyj^A
z@e`<k!&DW1ZtpfngFBkGW&&5;>#xC}{MOC<K=pFk+eK@7g(mFuGO0dhrd6&RER6dx
z47mx*exZHI)!|&_1NF?`<w{yW_1okFX*mKa-(jI4)b|gR?!!xaJz@3L6M;YG2Ws~L
zmJBHbY`CBOatG!7@YT;Cc}laTQsODsoVG)icH!t}Ja*cc4i#TDJe>Es{p4uIOODZR
z>-JU(L%i>Am-x)sGS_Xy46oJBeXV{AnqJQUX}QaZB6lgLt*!ekX0DHO|7+)qP<EI-
zzMJL=Lgo6liC?dBX|xSguGAac;gbh6Iko&c8oT-l?bYh?`)Ik|S3i7iewnSu%^=`$
z+nQYU)7yGCLC<}@l6SX&Y|}ZExPuOy4eu}()VN3Ea2LP&$HKJ|nr2kEn8(@mPACOa
zolVJlpOUQ{|8;l(z&880eqAqi!5p$fju-i!cQh1F5jHeO2umUeaHYVn|H+4O*PpAh
z^P<|iU;k?}EI*iu3-Q~8sW}tSU~uuz;0`Z6ApSZ2YmIYnhbDZ1z4m(Rv?XHGM||za
zg_}F|+-@AVLvyeU_+8zQl=W;O@)o+JIR{+J?^PZfIiM-gKDKB`_vgf4;OC>oZ8P9@
zQ`!AbK2K?r+3!}GsMOncw$B(h6W1Z-iD1#VLx-<@50wVU=GOFkAeYtQv&uwesH@d?
zVHjW}<+#kfI&svWw~^k|$JYd9e0RXj$%d;+yCxh%Pn8;O55QTu#DSlGN_n8LFFO5L
z^ZwixX7OvVeb4)Iv%QP|JYBzmHWb}&!Y=)E$yV9rKi_dZ)}i3xKDp{wFC_E=!-ihk
zFVyDOSEb)K={LbPuYxNUD`9n)qv@Z^Ey=c}|6EqS3ctJN_kT66h&$6WhwV#E0l}rY
z;a>$e2WP@jeG+1>%VmPY0A51j2?U&eWyiPO_x-gQCd1*5XcoX_;@5w!X&O~@?BzqO
z=~(B`ddJVTsos+IP5KbyKyzE)NJ_smUBZB?sGlfo+_jjo8dfw&`*eW5DW6cpuZ^31
zUoJE)uKIOtE>~Qr+b@or?UXZ?(0ts&M;ETQpndXwbDS&LdZE~B2ksi`OGQKR>+WXr
zeA*uD|Di~?TWl_`4_D>OR`c8Vm7A59*-Q1`PMNo$jB_8~B<Q+u7e1!t7jgE5i!0{b
zhHr8oP+eYd%|vd&qqL*Lw|wEOBj-*&2L4KDlHW(Stz38}QAzw#eiON@wR0(-Z+%~A
ziyl9=@A3mpP8J5dw;yd2-sRlsms@bGb1}cqCb!@uTnF&#e8BB@^ZRbM--}=pF#Y`2
z114}*6FMnqc@v{W3=Wvp(s>DEv|QTw3b>xn?V32W2kPKk^^=}_6psebcr$tN-kAws
zer@*ptO@bHWEgGYg=quiVldF;+AWFyoJUX?sa@3-hmS-dDC;)f7rq5t?uqjyAm>m1
zb1n5YbarjArN+MfoZQ_4R-LyY0_r&IqkW*9cnIs7%?5?*Ucl8uV*_$7!VmW6*BhIJ
zt3_8QSX{L7$ty+X?T~s@2Wj_BNa#T8<KE#=o4;EG`$||R_MHd3%Vxy=Jw3X`F5~{Z
z;l2;lnTnjScO?9rOOLnCj(52S(TE%AKwIs*N7m^3Q*z?_3}k`4P;kE|vcvtr&93T%
zXTjC5>izjr)|U1AZmw4Xqi}zpp(^`65Em7j$Nre@1I1siEfwoFqupn)&Rd`2dIwsm
z)Yi?NPTJMGXlyQC`RG!9TqqpuacEfkg~xFed-#g&jbq~=u1o2;_h8VL9}nVE=D2KB
z?(pF^xySvrX}2E6g`0$$kj318tv?uE3B*&W_tP;$YjJ@dzsuQfN%3oS;@ijfA#Flz
zhqJHyb6fkW|2!|I0_`&z4tLGZadQoeI5mXTi7JVLjk~e<@yQ2n?@()ZSydK+MckL`
zcMJ3*K=If;{W26owgy%zNQw*R@PfOvTVrPr9_VUyz%jV3A$}{fCghUH@~}2(>H`-Z
z348<V7Bs_DJ^szDZ*8vA;D(x~J(bu}(+9UpigJ%ZwY{L$b=xqB>SH^~)uPO#YOKxN
z#LGhq;KnV1bKxFxpcTxX;y)j$Fd3}au|WLhE?hp-O!b+GTJzBv97bGbe<55@N&ULv
zyK15eSZ7SM<}RL$qD`ofH82yr>cNI07!HsUjN(iy6o$G3libj19;-~f%H8$;S}J@0
zTr8CNfl@pO(W<$p)bIu?KD?`q+s;Ow!faK%BcXnAjC+Z3gQzzAn(Xriu5cD}C~VMi
zbiO|&aTRi?IVGFNgInDrG~K6k#~WqbOnZ)f%|^I4ycIZlZ}a|~eIwY2<9e8fM?>6&
zSGzMEUZF&E0N9*9S4>?gx7?~HUQ}wGARz_<lCs^OTC0sV{Q7h<<?26?W3=8UF;l|d
z_MSP6_zawn{ah(b$l)7MatEbxR!?9GBF}_Vj6*^<5hAOk7TlWK%=5&}>bE3E^bUK3
z#BnJ2i5MJlGah&@F%!;4)(5iP;PJV_UlxcC;XmJ9ZprEIi4|ep9!+7ZedBW&CR)SB
zA{4;(a2CmklvaA&adW_3G~cqj$PDi3m2rjgLKC?TWy}lhJ$p>GvhVlm$8iH_f;30R
z57dOVv8Pr-tD+D>t+wzma=4qvRjc8K0BZvu2KOPqKkfzdiA}_CC~~-PV(<ko)|JKy
z5Bn>dgPeUkWM=f{!CT2A&c-1N8yfhf=la|@B{A(1*up!;Apy;RBfR;5(q?fuuk0g2
z%=9YPl>0oJ*?_P6{&Zd4%a9eggn(|oi+|$8^|0?7`|#}j5}0URwArBMrf){%9|T76
z!R;e=tO$Nu2p++T-ovZ{F3v7?I8+4eT%rKV&JWwe2l?=N2%lhMY#-fGdbu5%?)9-F
z)MmoPMIfT%<xz9<@Z`hcD4(A%&X|9>Lw~=JQ*?*i3Qo8y$BSqlYu^#t=gEmmkVxIq
z=R$M+JdW}qaYZ1O$ws)T>cRio6_9MO;dbxQ|MKIS5rbyC8a@)5jpm=>T&;YdvNy@C
zgWN4^>^9-ofEUk7xk6|-!Umung;%OW=^oz@)&Bl`nhfC)S8Jb=Q-^mR<v-cN-wrmy
zHxY<)^zg8Uad*I>OpkH6Rvwu(_9Lc;7F_qN8K^`LIQJuNk4zpbgUOl~B^Mi`9_LW1
zbiGjgTPEJaWo8L%s+3Xm{@T{|;t?BX`oS3VXqVeh`)0E42_)EozYn?4Y1H8kS76}i
z+XF~)8LM8X`$UlBMJHq_IbkL9pu4%PRUMiE{eeO-0rby2;c8es>QN-E=1)Vr5Ke&W
z?OJX>3Dr3~re*mI95<64<`rd!yLco?GaJJoX#|g}{z+=ai2~Mww*hzNSB^V+arJ1D
zns$3bvu)L5Y~EEA-4sV$wB(XYKk?(5uO+sk?Ay#J?S$yu{quACI2+?`pT)Q(r(c-E
zQEbZX`iwxJb?N=Jz1B($(b9Iky%dfafjvJpVFCytQ*p$*zcvk_c)?Xa&6p(WUQ*hp
z>qM#{lByYJ**L2w&JF5u=+=e!4*l{N0;wKtccyb#(?<oki86qjTUrK<1%h}DoX==J
z5c1qZw$@qdZKo3`0l0GILQ)cULtWVP3-@VdH@*FtCw^>Aw|v+txK3c3jCEVKue&<D
z+o8=r#lfJrmb)9ktsc#_b@@E^o<v#nsIPi^X(n!G)Z5&tFECN?aKH+>nl~?yxC6vg
zC@FXtZFD_uDqTB(q49vwc$Vi3rqg&ha9DN4tNK-M#>4d0Y@Hx;?9{0TV!0*F@DYdi
z<0Z^o_ei;H%HD#rdZb)D5^s%@;Cn!s{X!cqb(nsc4vBk1Yb#ZB<=zTU!)^DPZ2MWJ
z-@N)T;aCJeMj7|9)qTo1dApK~FvIxMVJGH?(hGOTJO;7pjpM>fIHg}6kRsGRAmNFC
zqA{~NS4o#oBU?DzE!jMT4o7@p?sTsIaYm^T2rc07052Y;Z+{@+Qh>LS>7hsP!OUHH
zAdx&gU~#&IbnZ26{~QUI&=Ly4CV&zVv>g)d>6ZrsjIFeerKJqe)ZD7>9t4JRE3RcY
z;*uxlk`cPY^aC!E`QX|zo-J38sO7NdLfpB1Q9Z)*8gZ2X$PHdEIV0$}!^IK`%yXt*
z!;r_-G?jTEcUw=C=a>W?m;UDRZrx-0jvqc2EK}~5n|_gzh#`6I)rqOyJ<u~ALLbOI
zJ77!CqK=C>S1&k#O<kws7tTpO+XeFZIc~KMXYQwW_r_*?NqzKToBd6$`mNZ>=<Wo~
zU`U4BfK{%g05xX|p2uuvvn!Dk8u#Jl!^_3X`{Q$$mjI5blyp3I5zpKx?DeMW;@-5w
zk?yQO4+YP8azs-Cn%P-AEcdFn%85@jOwuLllNE|PIeM$KH=qu;w%fQVD!!VSDKhV1
zY`jMa6JB!X3WyUXu;!jg$4`==1-|tXD@VuGvAD!RPRtA4j#RfilcD-@aWY*sK<dI1
zSv8Wi>2s2$z9sAU8Uc3%q;trPW|+oe(1}v#>636`(%Z^Vlqk;DOY8%8BGqv@0$A6l
zPHw9o+UBQuJh7Kkk9rkUI97K$auXZs+i1|)?xWirJTdLdJ=yGdSaDVgl~QjPR6J<_
zbm!0$czd3(_GL@XS&vvM9&n!EE<I`WLnPb=N(8ZiHV=Idt+{ms6`Th=z<e%XoiN<X
zW7~85*e=x5){XrQ9PCp&{bVm25$w$!7g9rk@=*iV%uyVE&D2M{dN}Z;Y~-C__-)@F
zh^KUeYT8j9BaHRJX#|aTD(uBq4+_z4YuY1^fQ?4V2zRSEi3D1U`?5|Ko(<JwA&gtr
z7JoTq$DN?~C)CnUME{ZxOq{z&CeUUIT>{+^{h)f>0lAIY@~($U-0gr4*O05n56B1R
z3?4aGJ(kV9T`*Hu^w9Ky=$wV0m)`Hx=Pa9G2%eH44?;VEyCbWBeM0q?6<^2Phos#~
zH_L&;kt-R&yZv^e?}U!~1d4ESI@4#YbJ)s9UOj$nJ^|;Y^9*>G{B$_=oO1Ffe1tIP
zmaLmjBiyaiMSKMJZCU-+&|6ILQO2iUvtaem+H&pWBu`EThtclh$zj~-f<5)VMa@J%
z(|Pnf!RSQw#s#nXJ^OLXa?j#a8y9qeGWLoL5?gOkHo6Y|tYVmH(ODUv2h+yc8+50F
z4mGI}G><$whbQHXIF~kRQk-}Eotg2S@{&+_#7OXT7j53Mt>YeX+<{{M1*&XcIL*a_
zeSe}T0DJm|b~fr+-0BHH)-DxdzC-gst&89F=m$z{a6uJW09|Qqc6O)<rvR1QuIX$i
z_i<8oh{~-&h=qQ#d&s9_`%IeoS*+NG9p}?srG6ev-CM<}=FBGu_z{8b)09luv8{fc
zKIJ02it~&&4k-Illj5g>o-W9pZM=2HdyU+8?7r>R51r}h0TsL@oTguACnh+iZS{b6
z^VTptRykDC<Gt?~Xy`6-pTZ2;-4DIG0l~p{JvHd~JcaZfiC0SfL3N*1L(2ZtusL8p
z`B<jhr%;cz56v4|PCfpa(+ECh3y_}>I_&BA)G?CJx+m#v4&&I`;YTv7M~XV|DzDgu
zHRyGj?dpL_d=ziy1E?w7E9m{6FmmcypiEBW{2=fi(HNjw<km3#c8gQ_Ji;37EKt}I
zw_Y4Dy$0*yr)Sp9aq{0(>6acA)cxsh4@~@=<!SdH&tUWY%yC<RL&Y2@N#m?f_BzA7
zf-b@{oLoDxI*j6Hr3a?FH*lZl+AH^nhsXNa2kK?LoEK{oBL)k&*2#FoX7BBkNsJYA
z$Epk6>nu(LfV%D5Cw5jRn`86VyWCm8X$o=ZlFTcVhpfY6M4aK)Rz7_Y@}wTRJ_BQ>
z?pYkPx7!7E@!&8|wV;d84er30mp%s8o}T52Be#7iep81=90F*akzmMv5?|Bc-g`Ov
zfO_JYq*BiMM1a6sgU1&3<VvmFE_6NW&+(=Xh}7{<%sNtTzidf3IjKr!ym0EbMV+Qb
zF+9@nJT!x8IaT5hJC#qrWcm#E$gZDfFw=0H8d`eiKqYckm7b?+gKM`0cz}+GzTs5a
zS#cU-ZE)droD5avz6Wlt;&9<G&LM!pQH=YN<~G*gwkTIPppw@8@F3m$R3Izo5`oD*
z)`LgD4JB}J<8qBNj5`-NsE0}>a6`Yf&K@Yw3Sl0{m|zaM$vFs7E+@fO&7+Lb{8Gz-
zsiOeN5^{2J)!WYQczZgGEJ#k(LqGU|Zo9SW^xzNBG)4x7pof3(m=K{WLd63>K=&<r
zObFo2Q;9*iKaT+cO-T=39tFZfNa&@#c`N`fN#-UA)WbG8%@dTteX4pguA!yhvMdKd
zbKeoVVt$*6v2NMbS`V7WBGe;Rv*FU}?U-3HTMtd4JO}~kp&|HI^B5V5mg+G5;|9^p
zyH9ySj;{;%IN8_iJIi<G_g^CYTJNFR4sNh8Rh`Cda;L5{9na=*%bLN{P7Na+tf70J
zA?J3xB<&;ScbofGJBo*kIMFitrOShqd*Xu+<%X@lohCg8fe_~&H3FK>xPxm99yfyc
z8hQ9QxQKu!tJ8gZYafmJ%W2wgTW&Twi<Ec{_n9-j^l}H0^iq0gdNU48!c(W|migUd
z5$bt2z2*k#z#`O>7_a61Cg9ejl7iOV&RZnWORaL?IquOUs>gu2)pm&)-cg*Bj$Y5V
z%iFlE+1fn(E%ccO!h;Zk9#+C{0@vG`iKEgFRk`&pX=g58kJo{NvPZ%P=Kbg4KE~=}
z+vUy<cN1=B#3U~v4IF)OzOTNkm=gh4-G1Iq-;!Sr&bCJ+;`;M}Ccn?yi72>0lNu}Y
z6NX|ta9cfz3^V?&Gnp+)>bSq}HXvT937^y^Z~$dqJN>{h*U6b^rue+Q!oE30)6)zn
zD^e+DLOZA})Fo)%`-6JeXQ!XeiZAuGhWAxao4?u*-ThWT0*^yy`wRj1%})(;rC`Px
z&tv9$5SFkn1{R@*p@7bm!B~WPWaWr39uDQqW&0+_g-28qq-`eZ<;<+H(|HTH%FzPY
zAiBREL}oZN2LLCldZgsgyjwSLl&T|WItLF4yNnwYS0BYOsg;0t5hn~`J7RK{JR5*=
zz0&Dt5)g{89Wluk)_n#o{fzc(b=@wo$q;m=Jln3o1kGy^!?WWZIe9{)2Iba&J2XxV
z2R<R=<N~z^53_!_2S*7y_N@Rr`BX6S^b5L=Jel*V(A!g&=%MSjHHp;eEd2<8ei5en
zZho$!9QLmxC+F1<wqLG2Lo1PI6e>f`M>{1i%#H2W5Kgyb=eNjdbWE;!Z`urwj`zt7
zI&)VMxMOzD+}s@mT;J!botpYLhM*ojU}~C&(8W~*zh9FlFauLij~=i}UUV3K99YJo
zQf|G(Ug(ikbtN3t?HG3``xG7g%;Eh&IETX3zj~U{T!A<vI+BYx_L6YTBX?H2lb#b4
zw_8!I>*r}jVt$)FxGpte5gB0n)#};#kJV~jd&l?WkCa@QkH!oQZvV~Gud(B~peDOd
z1v3KfLD_9N8E9wf$xt0<Su)*JZ@tZvz50!&y!bHLi=0Pztd&ab^w69~POg+^^K7`C
zQIt==h!b^wFXKa*>*w7_`t8pXoc~m?^F|U>&Z4q*&VyNgpA946y6yXplC$*anwY>x
zQT}+UPpnEp5;#rAL#XT)Wz&;sdN8=@lsJj~Qr$;R+A30SSyPZ}&^&*g9{=oUT~g~0
z>KQ_v=DzMkeJDp#O3fvB`g(eXTQZ!MQ@=OhG+kDwN8CDUfMoVcr#oi2w8$O-T^>ub
zdR!^Flv9tOchV3)svUy~EANiOZfCubOkM0S#*xH2x3@~F)kE{r>m+2-`Ea?^G%j5I
zEKM?V>uwUNs2@QoGcW2v_c)4|+#`)ic=s`)hV)pU1p66hZQ^$S?LuGk$Vyom8)r5T
z0C5I)?ed{JGCGT7n3vi7uG|>~NhhmG#F}*$CtI*WZGe(F?a&O?GKm_qj^<@bS}a4*
zEivTRt9TOgU5-K~J0pe6=BI?GwH<-Aos!eZsiyDCB=&o}U(Nd<z>wEONv;X2d&@c1
z*x)8Feax_3(sTwcJ?emcyPu-H>G}Q$4KOa%nC18`-NpuaMV!`}QiWg`<yhYw?B22j
zw+Xt|trzh-4ozPwDCrqazV141SvG>_q5add?LkqTu674(OFIJ7%$IpbOg$14w5M|>
zRQnN_`x<|f*oysFsyl-uL(M8)QA4gK&{(DSNJpSaV1c8N5anQ87=mSw3~4gC+<&G%
zH%o8(!z0c*&8`j)26bAR5osK1^Zp22Jfv<}a@L+foHM1XUtm9<`W+T7psyn^aq9k1
z0;wJvsr8`SbD%nnR02+Y=hf+2w_nPY{6o`;VRFlQABDg)p6nO*q<0S*L4TpPFdmHn
zhnjY&#vI>r*DXv!Jpz*eh_{ALm^`QZigHZCvzqLE<sFe2BT4-AJD<~ICLbt2;5zbW
zZ*8|%rZMYzH@Q;@Bt_7H>039snp%&zJn{B}@^PLQcCR)<SF2G&nvRnzp6v{{mc053
z-R+1=EqT~G3D4L|l)aOV0FAu|jev4a{Ti9moyr4o=KGN!T62#G%FI4Audz&m%@aby
z=}1QYi}?Ng^b2$8z`RvEJ!A2RqB?F-l-E->H`*rm;Lfw6cP6ueI|>+~8E*Udu^gC-
z2)D`BS)9<2Z=Lp<Yy3W4KqkTGCEDs4)oF?ctUV*xGw#Loi&STFnE1`N$X%%2qlT5s
z5Vx6jbgX9}rHMVbj{>G6cZMe>p&muv%<o)=unF~O^27zYtqo6??R4avA)I5WQSbdo
zxhPX&z{kE_ARTo;y**g*3a+2^<X%bbLGNeFO^~C*a>{ob!X@XxJfxTs@y49Wye*j>
zF@oN2iTe)jeZsx^vOZLM33|0XZW#2V$(w1J&&qv(e)KP0^8=+HIjQTe&cv!8Eog+1
zcIufCou)}Tac>`IdA9FXZ`oPrk(BCIMN-N6O*G7;iwvxX&iiDjx&dPoVB$o*f@O@Z
zxqV4JP(LksDjwiHhn(&-eAIZ3EpGD~imhqkxm4kiD-so??K$z3-6d!`xI)o)4wcmO
za5>#)T~egyrMizIpcz#e+NZXl=_xa@2sOrAW7B-Puj3Bb_wCs6Lv(s&>hTZ^y7C*v
zO|KloLvQ?vi343NK_h8{(~TEtZ9mRwCQfI>5pbG;!zPYjRikXn#nGLz+JW(M7G6<&
z%&^?d2jbN3Fu#z6JCGkVHNWKCl6n|DhmlIl;c{;y4_Q7Fa+N_G*1WU~-6Pt{RdL~I
z;b)7UJ5&=+@1eunr@7qg1eR_XivxP@hB;jyxcxK*D4$A?Md5UR28B3)K_gX%)5YZ(
zkDO6%{}tu+zziC=Jk)qQ<KJ3*%F7R|W>k>s(c;L5I2=wha3B;_+F>5^EI%q3LDQ3D
z`Of(nmnI*`wNPjK!-j`vcdD__E#DTLhxQD^w0x^;pI6W`fDe>RtB2+_zQtaE`<y`2
zt!h<I4IFfgTd#=woOr6(exG)G7I~-X<h)#eS?4vFuBS;k7!I177|Ul$OyvGy;yx$s
zxE2Jok9r?C^5`C_?s5XH`4R{i>#B3}@CY@j^;LrAatoF70AsZK`@!a_!};my;C9D>
zdh47s-q*;jxJ8UoU#Xf;oZFBs+vB_gBDW-E<34C;=|IJNJn%CDW{CFlVrIG?ECO!2
zZqoe5(j6BQOBytCR+wVO`-s7wkpsx7-&H+S6IQkc54^LR0Ft`~{ZSTQtTSFSk6RaE
z9I<u}4NC{=;hdeu{?@__45>GBUwvT(YJ9dcEjuWec@C5d*N8e-MnGBw7A)~@R%9)5
znj^a+T?}xOck0j#dI;TYt2xbhLW`J^?r!2Vg#03?WDOy?GVB(V3?GN4SH$uVc5?*{
zyBoUQ-2@tG*Ptg^lhbsQTEnjaM%hTA;B@6&UU}NM2-1fC%b|MfB{lt#Na+Q#2$jZ#
zy4rqt!7k#RF*6U9Tftf(5~cPFZFbEX^i!?uSQI3ZMFIDS<<C>_1T;&cYtVH&a3UBw
z9dVkyh`;Y1Cr<NgAOyd}vLv(LvURw}iPL?{wjv>bWsO_bwE9?@B(Ebaj~*3YzhuxR
z=ys12r}>>?25O9WhTL12ff|zXUbKKNB1IfJFC|p$p}Ty_E-j%3?2?m|xRWJb#ICWg
zxd=QNvwYXv1N`i2K)qBw<CxR-83fazR#cE{u1DL6Z*QlmuujOFq@n*Wk9!k~(@x*~
zbwbxfwHYoF>Xxn3JRDuiZ>A=lw-0!L1<h?-sEk;gGBFf%k^(&FzI~Xa)jl-$p%=0H
zM-F);poB`rkU!zI%7@B~<tZaCEy9S*op9f3x2{;8wtgBjm2@P{Ri2m}ZXY#I7V@F-
zxSAJQ79lXCXUQ-DpeixsdgUFFD7dVAXwDjf`tQ&OH1BglB_t4=0|`gpu|VlniNu34
zM#plKePIf^-$`?~nOk;w*1kj6eTd~(Bo6)5Af8svX?h4RLQ}fqiPL$zYcn7eu)H3#
zh)r2zh4QR0Nq3DE>V|w>+;y*#G#3M|t4F1I*#ox7w{mDsupNmn50zWC?f4!vqYf=D
zU+3>X1xLvlp3!%Kc%BB#H^VjL5g#;9a_JsD>4luxM6EkzD7SHkCe>S3JSV+Vy<No_
zGU2%W+I!~YP`PE*ZxbJy0-wvB>*(<{!<_@yI)h!^q6%6GNFP)#+|m>3t)LBALGya<
z(&J5kVCr;S*dtcTp*iEYOLv~2lAz(jeI!!-7zdbw8b420tA!<~J&TSH^$S~vuG;~#
zR5a<neI><N(4OLin?~g<PHc@`O1PjEble)$<Bc8~`$$qent9Uwp;{%Lcf_L!*Fai{
zcaw{$c#h2+@e%t-;%OFh#N&a*Zy#tvxs7%0XK^yBw=sf=2Te0iJUiTu;mzPBP#HC8
zd}4whcFv@Ed$4q@c;~4@3jV|Ku&JF3-ncUdJe?t-gk%cZZm+b@9j{^a+bf&ZHQEYu
zPH;<HBa4P^EI9TpiMNoqNuVMnI+h6%1b;B;L35{8JX_I9dW)KUpfLmz8jT#sn0O26
zvJ91-<DK!i1doQjv+M_Y%!32>Z49UNq<WuBLbtUk`RS}|u4NorZ?(s}*<s9O@h#>V
zgSc;{m^caJ0o)kR`5noeobEc}h2=JoG!J5$-P*Tov+p>m-m;C0kB3G!87PjFN$ZDl
z3oqz}_-7~V2`5*<66&R^@1&tB3JC>9jgirOHDnYRIo`VND2NBsK2&aNruCJi`*w`r
z`dRPof`)-{awsL-@e)WpcA)1xs7Z4OPUeE7AGku$)yQN5H2Wro8##3;e!)R6?8m3J
z#Z{M}X3T5SdD|Mtq4Ut4bhaHIdCfbQ(Bn(!^j3?H{Drr-QZ&TLjin7;8Y>fXCh6`a
zhWqo%iF;Z4rMF{xoOhb%Z;Or*`Le_^l3CrHN^MC)ITvDIW_Q!W5U4dqJfj~9u`gTE
zq_@^yThFp7B?y1nB%vf6z2%5!ZVM<!n@MlCCMSbQZ)1y7wV?Ahwq(>xZ)?JFPSABB
z=OGDiW1EwNq_;79k3DG|K0(hm_MtI0gj$}O?;%AU!IJOG*~#h*`xV%ew%omgu1h_J
z(4m^Jcj@^WQcMsLSaJ|^6(po@z8SrJBCSr)`{S90a;<qIKs+#RPVdxEpq`5==<O-j
z3s%zGRWtV_>8*C7QzyJ#P!CxiG#Vp1W?3VtT)K*0x|32{yqH{D36NRmg*EB9wi&c1
z?W_Sw*QLnMq_+#B<(&NqcQ3bEO*)zOL2q05nO%5$%e9w$4=K}#XF2L0=pN(JJ7C+L
zw}Re|X_@5@bj^Ng4C9H#n2$Ct+*9W_G}W)=YX}oj#B;Y_+AlQ`J|Aiye$BTtZoQ^S
zj`mP{FsTnE4dA8g(zW|bS4-KEf$xBO>jPzIeJPPW6Y7OaWj|DVDe_V+`Q?U%DFo0g
zYkfmx{0pVF`3+mI^MzXPXy?*K#I07xLx;bRhf0%|+Jax+8c2ZH9=bd*X3(Bg>SqK#
zP)2JH-F?g#>Yel-XqI*PrT~XPc)tYox%2P`^c-N{wBiQ;QV*Z<O*vhe5AAi`$@PPZ
z)LV8YCHndwsEUk-&biA=mq-13sO<GpA*;qqRohGX%Iib5>VDzpy9>Apg79O4Uy6bK
zmaVINSG&1<U)vw3r-b;XeboI@@(I2*Agbm=XKV5Yx_f!)axaB%N@k_;($%#;bhpv|
zK=+t$4LZjDdOw?+b+gu&uBQTSa6KdDx11hk0<!5t`wyY?fotvkP{MM(6^t$8?Z6x&
zZ@p%ZmABsIVTQjc5w!k5_ttj|a6fygvJ>~xrThEk_yK2aFZFmn-?YvmZw(Fe=B1x?
z?*mWX--l{_;h631!2Gd1e4sRuhpyhsAE;-t{iZVeGZ#pI(De`0lhS=t?PI=WNf_~6
zmJ|r{TV!r>Ak5KUDpTGX%-$~TrMmiCgV`9XKGduUeQOvZFPvS5zVxFLeBcRn_)R@y
z+5=@TdFbw2|3KZp<eTc$XUMS&^^ur_?ksgYw0e%vZ%WvQKhRYliAns<`V2V%JnL!?
z9iEc$n{w0ntzr66={H^NSvZK|*^53@Mt`4Sm;MfoEvWj|YEq@Wwdx*^$R$m}cx2}E
zUCH&*HJ?R_TnSU&dQA*|D+Zfv39G+!l=ZC`K-bcj66o@+VQzVboU945>qEJY@CUl~
zcx%{gzmK-}R(+@reO@pI;v<Ey4|VH9YcKT=RLgj4Ac*5jyYKTnRO+)%5$3b5w+8Hr
z{?IvFduv#=&A055_6Q7OefDRaB0oVNZw-6)M;#+cz$kAGDo+Ij=6^;^&PmueG}GkK
z#fWb|#~-M?)7&G<m$zO*`@CJSMP|J<5W!%+l*r%@U0voMsLFmvLFTS6-5%cVp|a(r
z+O@njsA~UyJy?m)JjY7ZXPqJjXnA29kANlyXnEEtqJY-&($VUp+Hb$)DGw#T=3B$^
zBdv$7S5YM#Xxndpi32?zy2m>oJwolHjS*0GJW`WLucNn@?(s-XaxV;hX)CWnM&Qv$
z(IiOd+%I&G)NeY*dOOQ@!CL{bSLC5)zdj2U@f+n2UBH*+rQ7{D9_kr4z9}(1)=Rg3
z54mC~>bKC;;%5=<rAzg#q4m<<vQ5jk&{X1%mG6a03Hnv*Tf>n0Hbyk{w}Pqf&)0nP
z5GBM>`&R9t@+^N^JW}La!`5e(D}iD9_ZO@LRe8R4sElu+sYLI1CK*(RzBSmtwwJO2
z)m|!dztnp0Ei|FJ#!E?q@z$`ayiA^Hu6}j!TepX*yhZoi^&LhvS^hSLbHsO9_ZKaH
zpsL?WrD7=RdTFcrE#jfOw3ia)W<HcF7yYHHJ`<X_qV-!HgYGJC4QsS-6IO{W+WK1q
zmmOca+w*>!UtoHmjz=tz&Y|xYuJNrIDES=5GvXw47~jg6lG0)5TW@#QfQORG?xCa2
zZ)Hq9_jY|B0%z8juCnB<!StB&QnqC4OSgP4Q%ZUVd4v-Q9(MZ!UEgc1lHy@MA`??l
zUs%T@o-q`4=}VV>FH=gS(elVlLXDPhWlV`RTJ~GR^b^g8l4anbt5<pZYZiZbYasGd
ze<+z@-WtyBzi-Ok_&a1=aQ%J}WPa%TrIi5BFO@OsLtB3ynCuVlJSY9bvrv-$VLs~=
z`5)$fD`0<ry;Rp9y0APH$v2fM<*k8WdM_pY%zUXy^Ox$}-!`n031UBEPAZ7~2+if8
z^6N`g$6G_2&y*(K=!<sgZ=GX2LqM$2@@>FESFJByt&Ts?U7uBoXrryXeYGX#5xPVh
zZR;(o9+2vx#F_Y_<b~+Z6chIp8!z=x6yLP9G2R-?0G5~bN4k2hu;)dP9%8<oHO=M`
zZRCeo@9)=GZ&NmtAH{+c5mH{dn>-5|Ng`yu{n8=N^de2feg>R85qrM0w*AO=&qP$;
zvTf+M`Ks87Du19-zRg!9-e~E6p!nB6bk1jR2sipjJED!&_YnkN;+JZBp?&%NZHlZ9
z@7O9XlwXusqy1T<NF320wSdeK{TXpmM~r90$sIAqOM4&k*3fm9myS8V_xw~n1>GO0
z%x~*eNhh!#wSb8JYrgd7`RRowY#wn*D0+E&pk3Yp)35cZ6+)-)FQpM$>z;2}lXd&0
zt!w=QRsC9^pd0|NhY|pKymXf@c?q4Pte37*^bgd7-h5MO^-Eql-F3fok5=Crw6yUD
zy4w04NsqDlrPVwAK3E3UZ^IGTO5bm-rWn<i_BO_T=^paZF=mm6+Qk1uALg3trK`<n
z%sqHdd#Q}?!&4~nL*BA`eV^JvwSDiAI^Ex=MXmV#yYf=fv;Kjuy&l-wGv?Tc?>xsw
zd<Fp%@!R}$Dy6&?ke24{0Z~8SvTGKVhni4mZw>fFygl2!?mtj%^`$!7_<m`G9&Pdm
zD!ufF*53OcsP@$!+Gf`Wc8?G9&~L{M;ONiXCmqH2+6SHUPi-yFCi+mB`qnFbHF>G7
z`BGCB_Dj`d`Fg3mk48cYk^WNMkJKa~#qKX{b+12A`Sw|X_O9&@RCT^|Ox^XN(<53v
zbbs$JK-YN05V1##>PuH`FWs%qx7zJlfWN7F^6<aX=~wapKv#Q~DguyxJFPg~y}z`L
z{tP-fDZ2iFu5YIm=xAfU^^X1Rv;v);@8Fvf(6_#HO?j!bRsKNtdPbeJ6l=Z|p|xMS
z9vMwq3VFWQ6&Y8)X&t+)hqki&Tb3xeZ&_0+A2E%o`0cm?9Yg0!r(eN(sFY`|A`WR8
zf1q04?#is?9UMX{*O$H$gD|9zB0v<<_E4gbw(-(h+oK*3hP0QLHV<R-P|sohP5b=z
zT!9iE`VUmTJy)QUJkKxHS;s@yoX?Ppr#$!r^-Q?mbobT&Ky|$xQ}UaB(e2?Y{y;_M
z18aYVoYWNinINR5*w0!;YKpzy8oFmO`lh36d+R01+j{7nx?f71?my67#`|b8%A+0-
zkF=IQ&{dxyCmv~Cf1tZQYZVbm+w%nzlC;hGQu4(-^h6u~hVL5c$pqzrcw+9CuF+m<
zpWl3IAc5aomP90PSt)a@m#+EDF>+I^^475T>Tg-XJ$*f}r||mrTyeUjz0`!r`~%(c
zy;6BLr(Iv#%YM7iO?9dd{dNm#Cj-wP=o~s<ic9Q%sg(Ls>E&BFxY<R%>7L(?E6vkg
z)NcjfGey*IJJz!-RQ&_hZ^xD9k^9^CI(35Hf7`Jq3j8zV#6a)k4^+Fnbj(rSdS@Mf
zfbu;_PrO`xq0ahJne%;GT$AOc8SUdQRQu|0S=Dd9m3Cs6d@ob}7kNgU3>7k8+VWQY
zU#afzWvY^)Vm|{$hKenJprYTZLB(9u{s*eM>O)OM)3;t}f2dY=d7ptq2>MWclzZTM
zo|c%T^%?Q*(J5YPdcn71P`l@B-!J`f#pFYK>)%E!bocT0jARaP4YPc^tvJOc{h_Ok
zZzC4EhrU!EKVs-E^Lv$oddQe>s_ol9%jsj(RT5RKZ!@+Iznr{Zx}?1nZ+&^`mi^K`
z_P1(K$yBk&AE>9L{iYLNl9z7n@3WCt<E2vGX-;s`(q77)gZ5G_&!7{Wv_3PP;G}i@
zfl7UZCc#PTw;c=B)?V7?GyjQA+Mbn)*rdI_bo4&n8pg8|44w1)JPf+t9Z0B*XQd)v
z#rJWvQ=RRld%hBVdZf?(Qd!SRMZStP{y=p-+5s6Wq`b7%N5qn{Le^Vt@AAF<2i5%+
zJ?WRf>8$TW$5zzxR?8KZ^-_K8j=DVX^?c}VrM<0@Ug|ANda3XCLzvR~ZN)-Y{nqa*
zVM^bDBuweMs1T;~Te<!}fBjqe^&h{=uYdj9fBO~x`+xk;YOwsOzy6)emGS3)|IeTQ
z#ZkqLP~*S+*RTKSpQbeZD)IlH|Nh_qGYkDMe*fq0{4c8gOIPzZ|HU)^>pb|s5lF+d
zh5s*#|BFBV!~b&r=b2TT7clLWp|U^!^`qnoqGl=!&Hm?)zjJZCo?t<w@+8Pn6v@B-
zC_yMl#k!6n_wPT-a9zOn2BOUU`;Rgeo3RtjBD-_J&r&yWa+bQ_XQ>Rsn}zc4Kg-;k
zTv_OXAEiE<dQs{FewHHnJh4);Q2zaA8G@jrLfk?Z{4B+7TCS^Sq5S*Lp67ew%-_RD
z|9pAscZJSn{d`?N3tsl4+<j+13x42_lK0L1Ect;yN`7A7U%L6BKg)j7VgD@qp+C#!
zyZlSh3xAfpFZ3@(KlDf0&ujfz_6Pl0Hs9`FieC7$<bB0|Df*#5%jTQ@OVJB|lzb%L
z|K|7Zk|qOUSmhocBG>r{l0lj22E+VWBvF*W7jO_a@JEr$coq)ucl{ZFgNMOstKF^3
zXz^3_N4uY*KU)5j{Mr7e=+7QJW0d0cKjR73C-VA_djCi%|35x8*uVbY-oStTMNpqO
zlfV9*&?Em_^8Jco`@{bBv%h!Ah)T20pRYU1`1dUTaJy0I-?e+h!#~_^RQh-A9%1tj
zw;PrIUAsqe{nPLFRQk8=x)l7UUvSp?w=KI+o&R*pQS0Bf>@xbFe#u$u-?fZO=|9^l
zHU4dj&EuT^)2&9Wf7`OJL;v{mMy-F_vM+1@^v9I7{$0zbMPGmZsdl}_f4^1!&s#l9
z*+1Ltr_TR*yS^^UKilr7&S!aj7S2DfuivY%t*QT)TX1gV(9-|=AJOFh@vo0EO4Q1R
zul><b$&uNCJCo)STE3F@fH;z}PB0lD8K8ho^@J-TwtVNDnkbGDFA1irJCjjPKyV_<
z<F-Ul#i^+=3yGGAk|^thDqfye+&(&waoDoqP1<AeRMy$Sts-8`Y+cFvbaH4%$iN=?
zR0t>UA?wBSzpe{~gK1*;WoT}K$#=AW@%(K4Le~hzlj1~&xD-S<1@XwqC;lO^<SWJ`
z;qN@nSwO#MPDr~QLBWV6KRA;1fOI{3Gz$+WenmvGow+!WCUD&DC4Wuc42R8={s!EH
z-W0#Fe?kcoMIOI2!m~-A*~`rhGD52@+4*Dl5+hjJir8z*)3oegCjeSlsh)45I2<F*
zo(%6!M6%va6t8sxW;!I+_l1kp280O9$c6-uZ$vC6Mws^d^R{udAi|f&IzO;`{zoEr
zt%xW}n0Vr_17$?mO`zW?C%{&lClnm9;~Rk!2#=Mq-+`OcXQE^6Na;%G_`%^bv8`6b
zmm^5L-zOK(YudA<8BA6cA*eC}4uRtvN6*BXB3fX=E%pisAmu@O4^(^s5xFG7b;iA)
z?d4{qrUETaoy9Y{oB)ADjBgyq5sWF~`NB3w&cs(LYh=OZcgQ`IQpAuVF1&|DBtBA)
zFqc5jqE2w5$bm<!_|7>Tp^Y-}I#4-m77t!~Vtx@Sev+__$V8d$lu+>=p^vyiBi{tm
zkux|#1(n{jA?aikASTc%BTI^39rxfrIh43?HKwq-Q(gXCE9Y;-=<$R^0R=HYj2=%K
z3&ekm1RVs}xrE(qV(Z`n8LUona}XG(M+zmv!&fi2Bw|j3g!4}ibaRX`BEz%wOFSEi
zXtXf0E&n1MO$5LrD!ktmCmhXcConZp;Z5>-!BJMph9?<ve|apKeS5PyQC7F7V_X-K
zn-#b{GsS^oqSv9Wh`dQic(Q{2+9%TKOejP`*q+~SX5Y1Q(neU89U)DDenH$LutXL_
z;=z*z5lGI~g!aV4p{EQBM&>;jnd4f(*437h?O`%Q6mXHpW`M^%iHQ6Jf+trRu$RaZ
zLmYV0i~&a`RRrRDx-+}^VG}_?Y?2+hOo;YQawy<-Z;*hbRRWr1YS~Hqw(VC$5Lq>U
zEuPSa_#r!@{1O0u5R{+bA0@(F6Y0I+B7v>cBJBW>Of0}TvQAvGh<QLbkIeoK+?=Wr
z&10uj#vzM{2L$fOSkXjzC;7xLVsS)tNSG&XAm&Doj1~lbZycx*T*EJU2lRXa#MYSm
z_Gpkmju;vep+gDp&cAxBj2f#BS9azpXtHr+Cl^PD#JMP~X2C^BaUju`vlZcL?)IBY
z_hB~CtbrcjDVSgWCTvBo`7BBZWtd5|l0Vk%vzPuy=11bZ56;gBLqT%sfP~idw%3R#
zzd(Y<0*Q~e12#gg{_?mI5l9=jIXfeOgh3712f5ja6*07B$>CSNi4_r9XJN#cgtg3q
zLWK9uu^GW1Mr4-+st-D`;v>T@@!b)x#CXVzGembMq8<?Ol3lmYsC!ANk?e%n?xY0p
zNDgge!R;Z?2+ly@_JBG$E%S&C^W;h(uzMu{E&&^M3VB?uI4mPHgG9DU0=t{+^^zv>
zV*cW4#Wf~^FnDUS1MMlq;qa)c2X^A~0J))<CA&wq@mLFyBayi71ch;^sW{<f#SAB=
zf<#g`qPmkS8n|a<2S7je>)pgo7&WV01i$#@krR5&g3^g$K!}9WPheCa=c2IPXBON9
znDp=nkq(}4?p*Q*c9OIw^nASpLNH-+hKP$eL@xkpF1YF;>;31ZmB(ms4adGj;2IGu
z8-1O-z(5mTHGkf?MC(vz$VTqqunl;k9h5BDJvAv|4Mt=QCX%}k7f}r|P%1V+MOv$;
z#ySD&f!wtN+L~JP=ZkA<z!qUgu@(yFVII(+_ADrVH_GD~=!N7Nt-x_-Hr`akD*$T0
zgQSDAFya<WJ&#VT_UO<Gg$Tq23THb4S!0(l*@;;&+tFxXIPC+?96Lgb(CzN;4h*Fl
z`vpgFn2ty=0rU&p1Sa6*I^e)Eak%Qd(64V3mY_x8VWPI<4(o82o((NrNrA2J6S5b$
zNGkx;$hJbzcKjECeV$JehTk7-L_}+BDqQBAt+}#8>bJJl^B>EFi-ZEeNQ8&+<yW=|
zQP3jAJ&?%;+ado6*6#PU^`t>w0UM#B_2k~of{mLg>PcKBE9|&HIE=i;u<e`Jo}ckV
z2*<)3L<`q*J!J-*oL2wx7zS%(m2+_?OFdsoc4u`mWT+>^s>zfT0x-!&WR`Qx(^>@C
z7tgX={e+@PeBVGJMK19SYNS*V&r>OWLSZW1$e)|xSsY?5o}ZrtQ@$TT%-a@_D<`L9
zB)fr$eMK-^5sr?8W!lLb0F=Hj<m={iBw|hzlzsMd6g6kfmWO|G1k4*JY&T5Kvq+EB
z;{;@P^$RF%U8p3)V9TS14KH4M%QJ_36z<9RulJ4|20*`yO{9cY<IQk+hKK{e8L3`@
zp7tK7?LLi8QUwq`p~;clSk#p*5Ra*FjA)pk3Bz>kENJSU4YY?@h}X@f8K1dCX^%u|
znBx;I5vUQIm@w>hBj*U+-jD7K6ygj3Gus!2J39%moxdG~fKk796XS!GrQ*fN7MJk!
zultG{WD{M0xC~d$bl;6phnb?BeSm`Q1)L`ZfbR{k@##?=KF@r{o}dopqRkNpC{=DY
z5eDmldRKSCN-jT#dz>{9xUiW#AVg&9X1}44@B-M|(FTB<d;q|iy~NvzcmhCTBn2FW
za02T2$Y%cB2D$p6NsG|e1Y7re*1!=dKY*So3AiFb9UB|N+m?NlLL#acmSI9pl#%tx
zqHC~BTj<*#XT(Xpn{8|xULRf5-y93lB>7O{*mX*>PZAS%YoyKqn#lE^lXx?s&CTvV
z*A|J|fSVKmz!4S*;njznS%s6}2-y8<HBi>dI{PC2KTu;kjOgl(&;dlnh!i0(Pl&D_
zYH{SR$?()1Ms6a)sC$A1*oagjKq0dQaK;83yE$%k10@n20Bs3!wj;DVfz_Ff>Nb2$
zrmitC>AL1`mCMwy8ZQhtCc*{4+K7R~i>aLo?U7@Z!-<Iq;vV&*?S!^0FYivBXEPKg
zs6Ea!-Xg{bvHsN^{8qNp{Sq~hlwtwLBus#McDj;3Cv$kfeVGvEAEsfn`NkAs{x#!O
z22`T_J6s|Aa9x_+hrC5ReBZ(H=?k0p8O?N3AIjDunJU{_eskJk*ZjGXP{WyoKorkJ
z=&a|W#swESt@M0fTyKef_j$|NfG<5TGSmW<xBw0#xt^a_5ZfPUvM!-1&vOL~9fXT5
zjG)7u`WQ@lW`@BK7?@Z@_CK&?>uYNCbCj74Nh7eocS+-e5d3$0kAB`R?LpQGhwDBj
z>A>UJN6&te#ViIX@*1$E*Vydx5!QH-7+m!{dP`kmGe>K@o-hRoj$VjX?{7EEYB@&O
zVIx$nO_CNpRo2M%3*7NIDN;8(z*A$g-KNbX+&iY_v0Z?R5CCWXuB;Kwd&W#3`83ND
z7oTrJ6PhJ{zIrp3YBsd}X0*e4OPa#C`uX@qto__y;a;?Vx?w5!4kbbS{UhJ;Cb_-B
z)qjU2Gy=@Jb;Mn5v3(Ep<d7~*S{7_<1RkLiiM4_IJP-NVRxmddkYT-LeWQd(mB@_l
zyP*5Xx|P}+CoKC6Z<(p8VO?R@xJ)fA;Tjsvq)LEwrAOT5SHcD=yB99ZQ;aB1xjn>w
z_z41)*x0^6XlQ?&<PeC|my&gU&Q`16e+K#vYCx$+3)1{X35MR;#`iTVcJY9oG2U0V
z))F>9neuf|j>KWNmNn1e4s#C}C{uD01)W>Zuo)Xg+bzViu68++l8xRO3fLwpS-?fa
zFkk1hIx~XCKI%YZ3nKP;=LiufVI~0mx-)RbZamw>(^YHsyY)Uao1iQ7liLva1?6Zm
zNADo&d0(LivHg`$iS6PMv?7ZW+sxB#CN!pUCK~(9)mOpp;h>U<>+o#V$6O)kxsMC#
zT6U6~C?r;g$?Xq%MLoJ+Od$5CD<Sz^D~!zEK=vwNHPlD;t@}};nXK9Gg;Ooe_tg&*
zzt`+4YCnR?_qWSoy?xk58wb|d5n<2U<gSmStr_uvo*Z2LHJsda0U3g$Tc14|Hpd(L
zr?57T*ahx-2A9K;7SNQ4>SJaq^HCkyG56$kuIWc6L*OC;Kd|?a4Gq7`46NmBQ~)>o
zTpr9VcC6TN_nXXad_-P5Hrz`;YIF4q;{n$Qi=^h8;Ks=~YID(YJ1_f8Y7V(_I85C=
zS~JO>^VdphNSLS347QFrtg$e$gKuKj9VAUO#AnPQCYP5z;9ZHQ;i}0x2A_QdXTc$G
zxnEs&D^R23Tr`n-%*B<+Wa#lMw9ox4Yrsvs{bOs!yXB6_KAweUE6gluM*A^+IUvcl
z?ZY|#k|vF3c76Gk<%j^7t*@=enqG2NJEE8mGRt}ZLAejUi%(!o?8s3D6vFIdi)+0<
z*UlEGdRnud-Bc64dQADD=?%=mI;ey)!RZ6X!R!J^Y?HvSczwHIpM&b{YQ^^YI5NqC
zO-Uia7KDrX18u0rfh#tlembgl-?a!R{rC^sj4$cuIy>njd;=tQeSq4Jmnk1iuw}CM
zY@fvbBu5-|#LmkO#SAFnb{Z#toO*OZ>VsOz-q_1JLs51PI5$YRm8|aN8)J5kM|{A~
zu2sOZS7kbFJPJo9tNXqb8w@j{o-I1|8<RESR=^^`8;2+rmj<?~a+%~{TXIXQB^w6)
zx8$B136%)?Y>4G{3db#TpB+OIxMnj8A>lA`u)$cReBVuC4}q}c?pgsHkxUDyxHPCI
zrr6oep1N~;PjYn3LaJVJhKUT@><q(5vg>ep>Uf7k=24zF!YSm20D5NcfbMYNn~-vh
z4-R%3LGwMVYUJ4jvb_!KVPJ9EGe~m<EOTb#p4pcDPl@-yaemuppER<^bN<u%*^sqa
z8<J*`4CQPYS`SCUHE?P0=&>yu5x;QE&aD~FStVNsa<+5Z?aub=#(H)~HQ{EPi7jFc
zdymtjL704xbGu@IxD<j%1F(h92XJxF&aU&0DanlxLgd3nswZ3sAws_IS?5S{&V<bK
zh`bV_K{!xV-!7MtE1N-+tp}Fa4(9O5*DyAk!=C}RiH#4uAB1pZT^&x}n>%okzB;z$
zcm5px#rNB1p0IFC)d-BwuKr%PA@Tdgz!6DvIlLOkYmaf<F4RKy2aXT!95g6sp}@$Y
z&DoV3sj$p~3se^crh%>=L%b0^WuBvp6>n|-|E#@BvL(5WW!cYB_ypn!|DO>=yQCU5
zfNW$nGD;*`IsM|=0C=t)-tf$Zi@fJ<ww#BmN~Mw@32ZQMRXgOs9V_6>*2HIO^QXk)
zf}>Jk_Tv!XE_NhABK7$BnWycHgtMjWuxI%+pdaR&EkH#S1H*HPqpw<Z47E#GdpZuY
z<DC`u+F4=smUM=-d>5X>uq?L*X<^4j)){-n$lw^avnv7fz?$#VNS%f=sO`D2I5q1C
z-~+pxl(<(CMhvqZ6ws7Y1BY6^FUz4><<$f0jQR?B>g`C(*Vr7Sb#(8SPN&>7;vh{v
z4g6u-(u_aPY4?avgEMrco1Mpbsz%^O@*Wr|i}eh+CH*|kqXo#!v`oYxE;AK2>}P@T
z*m&&Np<$~#s$3``>mad}D7d?w51u*RR`T?PLf<t3cWVqCw<V6;S=lz*KBj*$nA#zx
zzij$!Lg-#v@$Ocj-YuU7me6L2@EIsv80H)4uqBrUMCsdzA2-Dc16ykW&!VI2_;&h{
zE^M&1E%|#te9#dbjRo{Qa~;9Sp<yX0RTMk}@if@Vh$<J!w*-@CU<4{enD1xg+iaU5
zL6ZLIw#8{Ij@;Sb@Eog*&&^JL&kSH+JR^Rxhry*L2g7V#k_1V?Y(UUxFm0^~T&!+&
z5N=(CiAtLhu<Bof=02i=HR^a)Xq~`u{H<gN^}=4JAGk+qL*N;FhmPiFj_V8s1S{=A
zx;Hcv8`%t)%d_(W2iY1lip~|!y$a_^KsmoTT<*~@N52=}Z4ZM=oepG?MXCk{_El`c
z1x_p$V9PcU>TnHn78pP=z|0=q5A`t)S&t~Y#kBE4r)bo|+#``jhtu2{TJ<#>wr+KI
z28^G$nI<nsVE1gcnACU*3Z@;GvBfd7=9qEQW4hNwVH9vU&G~CeemmHfF9Yr##b#O-
zC+3EP*NNu|)fxz238r=>rE_R5F_VoOGwfJ@RrxZIt>mIJ>Sm$p``N7gkXMtZ%69Kw
zEiuulH}{nhC5vKxfoa+Dfdy_gZya-Ubf!RGAQ@vEV%=aQxqF&c`Uup^aLRGvGu(GE
zv=HJNqpIVQ5@xmwH(Q%Pp1|Rh;f^IYYdrN*c&{g)X#sUqup}xKceoYXD|2zqxXG=3
zYhZGxMQyLK34yVRucyPz7V2MSmU@%3d&zPb19tkci){}tHF=g&oWaVznEaad=R#H6
zOF*TkF^DRf!R!VyaIv_-&>s83j<cBz@)%hf2}g$ftQMly&RD$)Hn(C>`DSz3JhEbx
zY+4+{$ZQ+GUMx!lvv2&M<&AAGAFtUza3IP9tafIDy!RP*#m*|%M~lw+`qo``WyCO;
zF?R&h^iI^-aqfc6{Xb!6<Td6?afH1Kk{7J2R<?a3ATxnr-8QG9>&=KNe>XI59d<?~
z!G15(2e$#s_1+IMbaIS4L(qN3Gjwi)C3-df$jkiimM<Yt?oM{&XO3yn7K>jCCg0pb
z%)5|Tfi>&MwxtIo-5)Hw`Gt~qqdP1)Zu-D9JP)~+#{IIXj-h1GKIr2*$06R5<hxG~
zgKQ$pV0Ir9bi8{u$HOt)D+w9z%rT16kPWjOfpil&Gp=T94OF}<ah>vT^erpk=KU$-
zS+4OwqcgiNi5d2&bMj8L^Wg3^u*vBsdlYyGIYjwS@okmk@qp*02$mhw5I%Z%EG)$Y
z`yj>uYtK`ZxZK$peN`}r!#v8}6?J)F86`^QGyV#uFr?A#mf8Za`7=(Ehb(g;P1#@2
z)ME@en+`u}L$$jSeF5|Y*|5!4pf-?Zj%%Nx)0|pyV~^{MJXrzYM&07h$7G=j>oFHi
zvWA$G?L2Qet}{12(`<9ibf57|#$67z3@HK6%`N%ucBy$goE+CI%K>2>jz!ki#e3?u
z&jXX!DqD0ue6zyBH4#+&ycL=5g4Tfc_dGK>aFMVc*lgn*jqWr4ij^HE+>#%J!Se+1
z>ti5V^PVC1!mjaM9>?y11YB&)18a<{iyR?U@N^fqJurC7(3?nM*)0-;*mIw$-sdCi
z>T0gmdwTM31^F-tD1BtUKR<8JT&+2Uu{+uv*_f;KLOd=Yo6N8j10EsO&0MXIiRaS<
z%MpD3v>Uuqz``nnZDXY|HD)7(Vj^r`2X+SFEtuSS3Po;xw`^ScrWtJU-<M)9M!?e_
zm`~W;_P8Z<w-cK2z~;djz`-=X<N$T={FuW-u(BZ&a1r1Pj9q%9xX%bI7=37y3&cFx
zk1e#~5%<7@44B;aJ0r!76QbPX!w3*vd`dbO@TDj;Yp;b9goxdNa)PQRIiJ4Wf#@V7
zf{q}nC#Xk?yP7NiR!V??XAPT0i@VT5r7Wt$j&?6)qch20J@Gbd;@+^dn^<n0gYzlQ
zRCHJxaC6A0M6<r50X-wcm=<5)9|Cl#3q#vX_XgI)K}TtA=eW0e77E?M7RD+cx>1UU
zo-xGl;jFuQvu9{)M-4<vG<q~AP07=(LnodN(3|!i8SaWpGAC8j7lWd^SEIG;74&??
zL#=pb?0v6x?qD0md*#ye`KrxI9>IFGfC{1#F55$mcs|^63y!Xi%?{yim)ac34q^KI
z{><Q3K`Oq54m~}JsfN{Rt@r-dCdABgCeKTE=Qw-GS2NnX;7Pw!!y4`YDLV0cP~gh0
zCFog9HBsbo+NB%oc3GFmK~G;vigTIS{S%>Qrmr3)GxDvM(3&b*=cnG*Sd4S_y7#*D
zj7&u}^pWShbZ`ow{MGjxE5yPr1@ko0SzXGR9<YTlO8J+P^ulsV_vb5FE7s|U&4Pnm
z2+-=jYEUy)@J$};7Twy_(FygAcV6|q8!dST%<`h=gqy&SOGmR<p)(Q9TR|zjRAmwO
z%~jEbr;tw5Z~tK8!<QE-_aCQCJGGqA@9C-8q2t)~q9c9k#p3D~I#oXp(G|QlB~n@L
z)y_1q6nllV+O2mcHrb)M)LV9H^IUEQzD9taeoNe%a<q>RJq;!wg-_G$MlJ-L8AkUy
zv6z-3$Jk2R7mlTZeS44(@C;MBrS8i|u|*n$DL?h-1gzIMO2Fr0l`(@>GPsMjQEv^E
z+~=a%tu$y;=5U-lNIZY3J|Sqm-`z2Gm9bjEZ~1ezL9%CP36E=0B}$fe$sV^}TnyB%
zdDnXjQ_Fi`pZiw3$ZRWGO1VvQCblG+?i)CebgF`}SDle=srWKNvj-lC;!%ws&D`;U
zcD!z~(bl39bomwurWN0PCunXesBq1CFD26*R#&ulC;LZLoFaLIZqY`XJm|#FH7J2-
z3ST@8v~Sr?b~Xht?zwNhqYjII9HT~`Az*3H-fTv$&(%2k679Rz`wWX@RHA*2TLa55
zJukUeYS2v@B<PvjK25X9AS6nLc%y6uP~Fr{WwiJphHk5dGj+!Bqg{uhvZv|dD<ASi
z6U;L{XI6D2LBQ#2)zpGkQgw*7weH0li#1?wF{Tb8>4t@ot~77auBZWdosD&?9i$}?
z&3Dcngwu74PMsb;4xOmsi?-hG)h0snqB+m|WadSShumB-w+v$H2<#<oQ0w<<8{Kcw
zoW0sf=nj#+B&VJ6K~1c-VYFR4Yie^|6|5f>6f};3vD_9!A6T%pd4FF~))me9bFC`3
zcE$3pv4S>EwSldFD(I-8iso{E-v`l_xG1jlxh<VS0Nbp$Qp`3hnp@ba+@nTI4(m$3
z*E;X?ba<#Xa)Y8aD0)_0_ZBuneF|G%w|^Xp#+_&{J%@Z%l9~{-xEm7m{NL`ktaXd#
zfr|yOk96r7`HB@S*IcF(Z)>o;?^MXb4^=?ua1TD{8L6{sK#TX*dyLnnx0B?(@`u_M
zbhP-)`0lM_t>v=Ot<tlEk6Cx=O)Kzw-cOfcGm;<3Z`K`nq36@jM$-%<V`-CiDxQ3f
z&%JJi;6q!FZqJC!nn9ksE!qpc23#(tzMvQcH_pswq?hOK{e70@JM{(LpA$T-pO2l&
z{dvu-45590DyWmtN$BS_o{3RTU-&5Rb&De6uh@GcF>bYIW}a5`uB8;|ofJLa(}+)O
zV@z5^{&aU2;F&yzqVHQ9_xF>T5|&50aN*>E*Y)cA-s#S4H$ypOxKz$F&=MmpI@^_S
zI6S;{CKq%|T7r$@;mbC3wmV-o^b9Mv?0MZ=EnAX%PKi#ULg)A_B+%K`(G#$FfTyTz
z<f*y97ilPM59vz~8XV&ZO4~&O6U-*N-m<7;7>zj(q3tG#i6NJ`8k9^5<uCJ__tyv=
zID$c2nfGG&OJ-5uvJfN|2LF8DE`)6ydgQb36DDffIIY5p>ktUi`Q`9`pNTj)=}dI@
zS`CWlm+3#m6qqIT!L-Dlb#+4zEo~jG5^}KGr`|f4HKy`4)?4<>G)p&&DRA%hFjJdg
zgQ^ET)@?Ikika7?BS~+9P3{^^h_uK^RzrjjLeE5P&E3|F?xhg8p&EF}_s-6}Z!25<
zL<f3?k)l`AQWz7^A=L@F8)biDxjHS?P^I7-l#du#+}uZx2iK~pY?`YEECdVzM=ULX
zL(&t(m$(};^cZ5+;>Tjmx9qWzyGeemk$O-P(;Dp_#Rlz}M~Av&t>8o=0+{-To(X@d
zY4IWY(2W2IwC8be>@-W&lcnOWR?zdNhlEi0)IN0Lg%6#1-`m}4Cq?e`ey_IW@i4VX
zA0q-=fylrU`m;-PkShNVqhpL)mOWe5u$t>&Ku_PFg1Oz2Y+Y>N^aw~`?d&rv9LXb&
zppAV_VD<1G!O0vqwb8#W#c5ZU8Y><PH((AdfPH^zB|We!zr;Al7BUVTK)Lt?ox>iW
z-)RYX?K-TzyN3e_a>ONAV=;-kwu1biX~y>`SN5(ZY%52#_|fCX0qt&hWx4uiwd-l0
z7}p6@z>!07W5dl0)U~^`RnBs6mCwa$=C~Z%?^wl(5R)ES-4z7dsy)N$lKWn)ZF^rN
zv?u}luu4mJ2TkC>)nqCvq+cexdnz&`x>jS@OKMnh1m3d8HmtgS9(mldXJKKYJh{$W
zl;kv4LH==9XYMYdw&5wtxs^m#SaHX^#5Y*1wQ&<}mkyLFre0Bua~&g&=H}cr_Bo-8
zP%G$6F<US=Vb(y~*!LEQg|!+={oGg^&D4+XugR`sI!AMiTTZ7JQ?UT`!l4fsTHR+8
zIyqQSeB_9p)j0-N$+=>kkDzv8num_`feYt7V2x!bdfE*a4`}W&hcWZAP$Oz{$2Jsn
zAp8y;ei8=7_qL0+=-UIIvus80j4)WzHMQ!QJmxK0&)}`!_qppWTL(j~!b|VdvJNET
zep+~EzmynT(d^sG$Z0p!D-^Su$}KwNfDIih-?BscA<z~(gve<VgD6TM)}<UG0aN)@
zy9i1qjCQ4V;VC%Ow6(8K%W$+b<;D=xDt0_!{ff=)F+#BpDE7iR!)-xxN7D*A-9HJ+
zY^L@;lgIv0qCiB;4g&{mC$=l0$>ncEiAsRxHDh9&Yy?!joa{z&*5Yj#3okCw$p$sF
z)XyD^+D+Ume=NL72w<-!A{HyYLyu*-)U?Du->ab_W`biajeE5<?Tsfkf&(hg6w%dt
z*W0!m*GkbDi=B+`l1ZEGzG8@BZvagTXk9Vi<0M6(2=rqSFU8S`E}UU5#oJAEx<&}J
zatbNhXTD`yMw~@q<9GgG6^Q=<G|%XutOD`tChmN@bG9hHBBC>Ptup&ZXh}f(^v#Du
zXTdK7TIZ)3=0e{>LGJ5Q?;==5VsJaQ<1+hY*(KuU{@mF{A4rk&J8YTl#|6+9?a0=i
z-PV~&qUYmp$o4%n#jeVs+Giq0e?@j$pbS#QA+%faOvufOwpMSgl_LsKWsP*&8jfJ|
zM}z|O7+ZCi9Nvs?WD_E2eW{_z6!Z*cTQ8^ebc-IVq*w*@nnLIdc5CmJU4stF*|5r<
zuPrPIlud@UH<~K<d2tF;4K>1>L>JB?(CTq0xg^AclGRX>*HDsVw2UTqFcTft8H)#x
z!%DICI;+NZD=p}%0$Uib2a<D5FN_e{`+J>l19S0BaNxk>72${NqC(JKSpP{5VqUhi
zL@tt|D?vv?`~up-PA(knux$e^E#KQZD=?EkJf`)y6lYC6>#-T~z1rH^6yd-r-)hMk
zXZndfPAp5-UZ2qjiV&UJj%3-Ae6>*IgkOy&_hS&<+6Y7IqkBInoaRvkdb-!8RpIl{
z?okk;C_qS&Bi026n!;hm(cyDnX8^WJQ~o?c*FJ|1XIHS1+jIfRe6LQ(&=&0<c&=S^
zq2+pwf*VB)$%(}HQ1}_U6Vvn@@1=Iu0Kw9+?^||D!xI`dZjtcG)U8g}f)U)fkpM&F
z5K7)Yqaj5D#ZAH}%bQCl5t>t?bkVtr{Ot8!#*E<vy3xj%IwTzd&tiO81<jT6pyw53
zqj3NUoH~b~g<Y2jipZp1Ob-zcs9NP?^FXBh$k}xL6k7#}j47eL&yB7#a2u`ekPJQV
zKkF5o1yiIIK);dLTuIEta9T9UPTj&8_o88$yzwf5&f>Nd6afj-CiQKxf{J6Y)^4rD
zcSQUlaTd@r%e@$(N_w>|+=HR5;QWV5&Yt~Vta02T%E4UFj#irJh`pwQA&Fn|lzdQ(
zQJAOfE<@{D_kQsWHCkG^_dETCR}adENf2b(LH&hivbOGBmP=lWYQGOXpx8nyggE20
zKC_GH2Swjhw7M88G|OiJT4McCaQB+_edDpwFCD07AXCI?Y0!GcZq;xl8br^NtHsaI
znJ;Z^S&Qf1ZUy}e-Ui#}TBNYF-<L7(@9`Qkns2vb9f-5ZG>?M&46`Te&gmWy6}AhI
z(U=-U>usCmjX)dUqPr#=BsquKm-=USs)_c}PKurY$ATI#XI2@|)ZGw*$`xZO!pTj)
zg3x2BM&~ZKY&nCsBz@0o_S1%f0wMk;&>km%1+hoM=cmNAK4sfPR0CS(r{43?GA|Kk
z-J;vFgU2#Mx>k8yI<RacE4)@r?d&3Qv<{u)(+qPmML^?#_`t=tAxI?LrJF5J=sny#
z^bG8bmF%F#USPrPm-F_iV6Esj<6m;C0j3=D5)Q^=salNfTlB2wliCqizgyPSZ<C_a
zO2Dc-CR2_b=~h4(qS3lzZpFVEoeimod5oSYYOkX5DR;``WHjBumE|CI0peua)gnbF
zjxo>`N9u*;;C9XDalc<&)A>NB#H9PZ8XCTOw$a+HU>AJlfVKLn2fEnghL#+IAli&9
zWhJHNnuyTMnK!iWJDYR1#D+L@%@~<Doho_}XAv+ya3Q(@9W6#@q#VmM+RMK5EO574
zx3@Sggc3+6T4QJA&|bWopINcqt0^@tFs*XoX%SP?ZaWr3ht4yQtAe)14>~X$r*yva
zz1WhIKlB=_M2?`p0qpfd_Zi9cc3Qi$G+M-RI!W4i>z9sUO@5}c_@oxYUpIPs`M4B0
z2a^O3H=)9jH7jT6+R;N68*ZwAb=SSvF?FWiSq%+u=$V(2Uc_n57{@rG_iHtGf&}(v
zBgk3H=UdVOWuaw!ZfnrFMPm_PSPv40EK$JGZp#pxD_ZvW+!m6eMn5kHt;kgM8cMi3
zi%vP7BppB*L@W|n4qzYA^HGaLmqS{GBaGNB?IOLhqAZ<GOP1e>A{_fIOAxo3hRCp6
zw(1lDid5638*vWg68H37k3v*lO-5HN?hDR1)!;qU2coPgI+pK#9m+<t(Z18Yg+U<X
zlpE$PDVHO$6y_4{Nt}&9Yg+gEr=!>$b!_GM$Tkars@HH`41v4b6otx*UB4zm8*Pd@
zV~Zb)Rem?D!Epz#m*RLSI=hcxRjqFw?7U82=RqY}1-)z1s@~f4a#9Nq(8>oht>dS<
zePDymWeuLHdk#ap^n9S@<^<CyH`I(cDPM}KuhpA8+8a8NX_1)!j>8Y=Bxq@9*}U!(
zFG8e&bKU3N$#Mql*lmOnxGd(1@wws(iPwp;281;YMNmPsd9)ccBTOJ#`?zHZR=2X(
zCV&otHKBR!TecE%ZuB;nEaTJq&{TYXqsk7SDtp?;O}>iJkBW!HNaLKiYKc+xY^U3=
zd$qwgCptWMNK|_6qMvHd8!rZX9rq!YVilpT)h##(JOQ3*{L)?Q&7cZ?8~04z&1ra1
z)FN#*!TiZAmI;XxOfd^U2YMc8wC>n)twm$TG8$V?(1qJ1bcGXDMbe*0y&)IDdFw?s
z!01@{LkG_zn)<ExEIpH6!&$~gS70j(GfN35Vm`i^(1u(O^R3|+OWWzEf{Avu;5~w=
ztPT<Pfab6Rt29ci#b=7`h00zhhY6xNK7V#lljYSoW0u=y(23g#bj3zyCDb4rOFoLj
z*oFd5mo)~KZoSrna6lD@F|qC%tfIv63x=a)m4HsOpNB-h=;nkU8WLvEj)RQRe4mMs
zbAu)9ud19Amha<q<8kT4x521K*^A0E0Gir`r*n3}e!mXlU^tN$jIB55?A9zj^w*E+
z{4e6EjLz^$hjNl=)Wz?hS;ufgTanSZIBD}dCcUpu*`gCA(by@gO-BrZt;Z`w_1~{m
z<8ZHXDv4Z-sOv*O*+f{DM`e^-Ye!JQg_ZAZ4R&0OhAT9b%}UV}UL%!tA#Nw4(PLQ_
zZW}3)lfFfXp-|Cw_wOp}Lb95Ox<^taR*l+*!sw$PqD+~rC!MeQL$l<9p`+glGAq-+
zOh!82qJ?Fh(Qvh@U5>Iw!@(Ii*t8SO(TGF@7Mqt^jV&eU^3gAbE_^cBg9Wj%5_BYc
z-KYol3w63J>ggPytf%$5T8h0&XiRqPE-EWwm(7~HYL0WdW1y`&Y<nn!l;~>l;gSr%
zKk<8*2wc56Y*8&fL%0w^M+`>tIkL&swuXX^Vbc-@@GV=5cW)f|%B9k-fwuXnp*0^-
zYK!ytI@AFLkKJ7_9Nv{#j^&`{T>~wn-f9WRv1+kIgU+_>6GJ(924!o`Y1$tecbodY
z&#)!vBpPdTsAR#E)hKpODgoHZTD&L-$0^xN^v-ZUzh$!ocSLW$N5L}N%!I}Mp=SW^
zMqQ!|T02MQ&@tMrw`G>eqOm4~YOfI5<6OLk?!Rd44?=gbsj1x6Ohw4DW~Z^|S8-FV
zW({!IcAt38>N%fk{nW6m-;~f17Skb%XuG#{Y|*Z_gO`DK5S)6O!-z}*PO_nhrg>?W
zCMYzG7*|Od(CzarDK?4glH;ga{9G&M5!|40>k<vmXXqSe3!qAQ>ur@Ca?uEBf$mwn
zY)ZG@-q}Wl&OUDqLm{f7x0{nYU}dXeTq6`e^<@(jLT<N)V@%nkk$ZGfB_}Y8nd~q<
zDA{qGrg3j<PmZ6V5j34Et3vxsj#U?{2E;~e*2{6^4O9LdL6;<j9t+W73LDt^yU{rD
zsu~JxcFH9@kg5%cMn7XKQ$n`cDd+a7p(`hVXqvaI9ZwEva^5=_tt^MP2{E~<eY^-z
zeCQcrn$g(dtk}MU_lf4u#m)d^^V2cu!c^KG*&ym&ylG)H9LAA3bf7I_3DkU8YOFB3
zXH^UT6X3Gv!UO5T(rl>WfUN{Pi#V$@grx+LQxWYh=K;0E2%_HHP1`-&;VpRjlTA+N
zt^2~UDbrea>Uxc$V_Iu}Tzwk*fy-N)mC~hT2sY}HG*Du~o#qSA2y3%bD(eRw9M*`2
z^I>B_xQEm5_k^x(BPDu_J?Qx=Y<3C*x@n8wYyl1%*QGPr<a8#6Xju+4vO^a-5!`0^
zBo|(Tq1d%ap?8e5bT8F2R<>yO?M|pd%|wguK{TW=@jijB+X)cE#NrTj@mlEAM!0Al
zw_bLOy@qt3fWsyw{M)V^0^F682f^^0hHgeY(XeWp&S!^RYq)MO^n9(`%8q`Ep0B4=
zgYNv^IJah(L+~QJK!o~?_rkF|{aiH8>Kn)TY?q0~)(n(`8DLwVJLMIL^;@pflw$RB
zA)Pe}rh0!o*#`mE>5~C)vhopa`_hGBX0((KCAoxX`8?@)?a8iHMsxr<(Nw4RNm&9C
z^(yMdLX(!~HkjQ8Y4{Nlj&JE>;tN0=6OR+oh?8j59TY@kHQy9KZy6tVm_ZF^EJ_HD
zd#}d`nhss+TXaOnPT_;m@I5P4^t8pk)oR04aKwj&NG5_LqT%b+szSV+Moa~i!xi9)
zP{=0dzh!W^83tTvl8DA+((p=b)oJksOhcs^4{+#=g%xzLgDC1VdTXB1LofoApsQPZ
zpHD0vfs_Q^x7SR!)|S{T0FE#<pxWsZ^$eR(Lar?t=ZJL)R7VY>#h2)fv;+v0D7R1?
z)3he>M^EQ@zj?fYkth+34UtB6g5?#|`v^Ms63d^bC3&5_n(#quECRjCyCMLN?XRYF
zw?mtRQLG7@OQQ*rugqvfSTz(_k~eqVA4-Bmdo$O43p%}f<1caRJ=e%yL!1~wIsG#l
z8@df4O}ZMi#7SgBNt4qs(Qx&G$^}p~;#nJFn%F*~wci@%Cj3-1O!r3M15WJF90wUq
zVH4f1mcv;XI;+2&9J~09$c^R|v@jr$6v7)zG*)blqy^E%BjpMd4?x%r$ArypQxZ{N
zJ7|^Cg=lZLY~484h`MAEl-wa+L*BOr7ByZrf{CC?zNIu-D;i8$HIvh^ZVf}H2co^)
zXCN7()7o#nv|cru_bt1W`B5}ohya0AaD!%`o1-7m8DE){xN<hy+of5VE=2FZIHdw*
zFq}mjI-7<fYO*0LX59;4nOm-oWG29zGu{_6n>c%kj(Z&xHMc>xup6;|xjBqbi8kfw
zH;m>_tz(qr^u8Ii+lmd9b3{NFbQPPK&U>r30nP&5hRQj@)2N}gS-rIt5*c)O2Pde_
zAA*Di)th_d`Jt=XV2OK94T=hcHFsYzzmn7dWfRf)fux=^IAT~LtwFC|L5s0c!L;@a
zy@u65W`k~%)A=gOtyT*x!SuO-uGKoUdG0vqkVc$nL_k2-wAtyrvQHzNzf><9&WKIl
zp;=wCwOj3(JG^(DPyZHWfp2f8xqhwaiVogl9+H7?pYFNjxHPA2#=xv`1FQk@`WK8&
zqabuM1~{7g&`q8Q(Y9{|NO5=tHM)|5=7+tTTE;Egm@5Tyj9%=GGD?g3q8+$x^qQZ+
zn$2t*dghJUZfwwNy$X#DO6ZIZUSj*EaZpmsdY`4;W<g-qi`I1?c57c!w0M{69GKx*
zB^uutRI0_I$*29!$zwC){fxBU1W=s@h(<7Z=lr($NSWi~@0=@lA3frLxtqGJSI{qP
zwR@+{?NjS{OFmBK`ndzl0(&*N_>kC>((ZZ<@r`yVdsrjf<$1?f#J|N9ai73#v+E2)
zvnSal#^%1ax;6C~O}EaiMB1V;2X=gsa5)i;-D~JxK2W9z;_fT~xA%@z{Zr~J0=L;-
zXY}(a2r3Of(fhDSbtWGwlP4*K9_Z~I=}$yQvY~Q>6h5?<WVXQAu8t_Wyn}Q2cWW4m
zViffR574PyF45ScgpQKEh6pC<Xm;9!!=d=Sh2`xy_NQq!shth_I;KmM$S)LgC~#Dp
znTC9Moo15o81*j{xSH8s<Er>{B@?!Ba)io2$4$D~%rsV<FhNI8Ucqd)+E%)mUe|)|
z(vKQmL7cUAT+_R)O~>l@2?l9Y495BpN|HpQ^}<<(*<ItlERqIHZ|YS(wI0LQtg*4e
zv!#<EVNSnRY+gg9R3Vz82GS(>du7w_LLa7w)EyjV>Gz9z${OhHS#$1NDT~eR^uk)U
zeRc$3uS>Ug?nBe9wUNiiXpfJEaGx~1hS<=7l2*oPxn-H>?Iq0mx!NGliPLl`+rIV|
z!iBY?>RGb4OgLvlmBHL-y%#$R@#(D)HWqZ(<TWHW9XryV)fTB^L<B`w=$j3cV??k8
z)Tic~Rs~pU^O7sw-xnWF#t6^aPL!bc`4N6%9R<UloW|$TB#-Y`x*eJD$3DDhzeP!W
zJsbu*3v2(nM>x}Wf;vYJdKoIe#=^n8H!Z1}({#%cY2y!7KG)wUxH|O;$f2;3SCC_#
zQWS0t8UV#WI=|lmVZ3Yj3rO>pC3Sbu=tW(kO4is{OQ5q9+&yT$^;VxhIPdNO6+vjQ
z-@EN@uN;`}&-?IEr^g0R?;bm2^mA9|v-ajziXEqfpZg7j2Tzx_s(;HpPX3^y_#oLs
z?5CsoS-in}oQgnK@lmqhuMXMIn!IdSh#lR}T7$-;kz1h-e_-Fm?88&H8T5k=C;}@+
zL${@)my0cLG+r<H6`GHeJwo%KBZ~S$vXf$=`!^$Jv!H}$IlYzodU0F{qE3#5QM_eS
zyhRBTN6~!mbP<Iv+V-b{wWiqsBvQYZ+HLy9S`09#c|`YaF;Srfvb=M&JM@~Gi2pve
zoT0>h8jX|M&Ju8$$%Thy3t9;|bXq@jY%|@_*Sj^G)t0yR)d_>{<fCM}oenL~)S_&(
z4I(nhN6FSd&o=xWMDkDAvBPW7dYfkM><@O04dg(xw|Bc2+Gpqv7&9G;J*5Q=a|1n4
zd6|fY8HFNf^+B>lpmIkPv{ru&^}-?>)?oQiRya<Boo-&7l~1L}-35)lAwedems|PV
z?CHHePPTBChR*2SEgGBD3*Bvimiw(>D3HKtyl|~k1Lj-n%F&JD^cgsZH(|PU?mCOV
ztnea$?&!v$y3;xInKORxM+9Bxakl&#a`XtwzJ}9r>9P2HN|H=qP|hrzmP^lyx&%#^
zw!FQ)>-t=+;Xvke-1{XhMR(dR9VxY-e$D_L$yOR8mYZGK&V8t_Grn;g(6ZaOk^BvM
z>{g$b>g@Moo6UNs^<GTLSPlA4;=q^EKjs>r?NAzYZ!b6I2Og<4`<22yV&FaNlVyz~
zm7%O7{>E##4G-LnYqWQ`4MSJE*SbZU6tM*32x{OV-rNkRect=(qx%<LV^$jY$#*xo
zTsmBE`60@PMB>mhp8BLzLzQC?u;3f`=U?7!p<}qkRb%)a$<IlQ&f5x{hx<!#UFGwt
z2*T`9gMYb$Obv4db>m=IRF1=PD?nN}XucJc6;0CuRUf#jXRUyp&$0xax5{?L?x5rT
zp1IUtLRekUopR7y^vq_>FYh@Gp{%om=6efsX1CpIXxt1U2Xl0!h%SI~R^)WuvIIT`
z^&xQ}7kYbgoa#eo@r7yR+k@sB3z%BktpTqDe+i}DvZaUVTs;Xh^xW<Yns3<}A9G2!
z)>fNeDqM|#+b8jweE~I!k`X$b7mgfrRgLpa=nN~px)e8*FrrH@JfnBO_!tQ?A3iHp
z&%6UY%{}72%6+Om#}4Z2=us8y>g}C<vktmvzcdFulgaXy33=9mdiBad*N4s(4VL0#
zLJ48bOE28p7pFB`7zWG082eIu`m6$Xe7GS`GbARl1D1eHJ`U@>t4XR+r{(^B-lmOH
z^`ShRF*AklB9t6fPSdSmE18&6uR;3iK2nw)eMTu&%cTM7^uuAap|aEZ+O+IwLj*&s
zn_VRi>jzRVwMAK<#GoVk@)d;`<alT)Xumf`$Y~B55CcCh)^5?WnsEEQ?0dJvH|q|o
z_xGoB^{DjT>s&r9lwI!6R~u3Aw%|^~?SK-RjnFa1r3O(_8apbD@kr?p%B8m~!e;&o
zJV6Jvq`59Fmnv-9s2i>Wm8#V)?U%0QhEIvHnpNPA#$NMYtfyTsZI_NcE)7UO2{`VP
zp^#UC)}Unb4%+YCnxB-xTfteDxFJwhzmK~$x)^Qu8Q4#Z8e*VD-7WifS3+203<azN
zb)yB!o_Ej@wA8Yq-Wa|RxvM!|#DxSl26UrxAhTEJrMK)^TcvnA_r3@fi<gsc1;;MA
zEgoMvLWAu#xu5qA@<8-*>4ju7cbIOq2s2-Lt6kf@OrZD1w(|3ehWgr=p5fH#xHO(N
zWUL3-AZEQK^Q9#44H~d053a%JU?hZCZVQx%4}s1Ydm`R)kP%|dTlB)cQ{2|)TD`lY
z3G|j7N?PgkUTe&~FRekRk|;VYL058DlM&vdz>!w8;kdW372N0amff6Q2lXdIDm2ea
zOVFHdt@*N1>ZJkk8H|?@7BoVk@hODL=0GR({yZi&5=7kY9kO;%mJUJhO?4%?a(dxv
z^?}HHN3Fg*vm=Uq>AaNvm7w=#@p`=UHrA5b*Eii{JPYd2fT!AM@e;ytkz?U)?iMhU
zo#26M)_`GC%cFQ)I)kEj_Co2gJMBT+omxXpO+f}R9Q}YgvUOMjj<j!c-;%bg`vqs&
zF)__Bsui77UW&weyqR(zU?W)jQpvqKwfCh+yWD=EaxQh+f@X9&A6ihld?^}{L2FQ!
z3BOQ$(}L!p2*tm&BaHXb+47|U&42cbA!+8&*Cq-l?++u3?|LaZR6%FZjfm4r_sIK;
zqGB-Gd}x;Nj?;eWbT5-XXp!j4L-!n)wxFz-+e2G+FOxs%T(`kLwAZz6*>b6r7QQKR
zi!Ytms{IL9O8KGY#%wUyCL;TRBdsrp;?Adsf8fe(eZx%rsY74HAW)P_`h^ns_|Vg1
z$1jwSyN5Py_I=Z?1X>Se_bBKJItVp->1wyRKNMB(my)CT7s`J6L-$=<{DkYig@DMe
zeY*IE5{>!`9e1pAdt2Ib^V8%XD%F{9TEmg{58Be)ztC{x{Dbzg?w9_9h7aK%6o>ei
zPNmv<=-791`p~7d-9vFYdgu;^ogWl2%5<j^w+f*%mzU=Hx!n)k*HHVw!U_LFYcBOI
zTW6UMY_;rPC<mJl?VJlfba1Zr(%JG$CkHw&<v`@2i^GJ6o)=kvp?4?yL&+e~cX{a^
zFU<WeV1LkZcklk7bLN-U8fUhD&^_k^TVD_Bd6i%2XSMu;+6MMR_s;LNNT)a+x)c)i
zP|WLx;;sHoCD}dNOGnx-&6pK0UEy!?lNKfa{-%}b;H3<u4{hb?x+sU%^j3?%|9<G8
z-@H_rgTLwIeqXv;d+C8sqa(uK(=T-N`c26p(Z_tL1lPXlF$GI|Xi200LdRHd4V5^-
zm#*2$L(ikl{Lr3O`h|{FU&>a-3)k8YJgb@f(v^N`1mq3YLw8N-ZN^QY!52k8G>!4l
zoYVe=&RqIKOU?Zky7PJ|rx)#^NCefF&N5$$Ox#-o>0`=6>0|k!Gyl>IbGGtUuxfjs
zVeO~y6GYR~H&t*{d*STkEt+e8<|*PTM|tV_N}d*1=<d_Qw$$n0vaQbc3*B?R_j?u*
z>7^sDT3<TXcqy5i%1ieyFI~#c{7rY>Z<Fok<0Dm9dwRK&eWm%9%_DtRYnjE$FLdQT
zA6i>pFKwyhw+6<m{?J~^{DrQXUP|)y@leucwU>^%>qA#*<rlhIe``3#<M&LpBtiE*
zQz_ZrQvrf|Pm7`LHy(y;(tc=t1}5Mb-?P+`L}Pv{Xv($uO-Ts#)_ay{>8<x%?AtU@
zCdV~?p=<584anv8(w%mHYuGrke$k?YBITt!wU?rN{?>3zi|0cr|9U4sa(wEj?xD|o
zMB>n1-W%x81Oc4o+ejqAX|3rOx@&#u?Dgpj2us)f3!Uw0afGm2f6MOO)^}M_O7!m`
zlkBuontbT&^R3}5!oRIC<;)%Rt>G*=z9%WN^X9SN8kAlA8}>Tee(7TvkXg9Dy&S>(
zUm9V~^=*nN1@hXz(7nre=<fSza+Znv88ML*P2Y9_8K7HgFD)s}U+Bosh>4_VewrLf
z(fqwcA_I2I-wRkJ&@Sbzp^wtv8b*004m3sU_X?H-;H|#TK-||0N8N9&NPfPxZmCuM
zNl7iyo-tGTUA`Bv(9xesirf-?zV)7A@Y^sTA$%M4r7ewb!$6tm$9`$o-t0rye43mD
z6Z3loyGSsx@-K9(_SUe<dLM1Awf94J+s_MTrTC0OEEM;*Wxo&=P5BqPQ+sPb8}g+U
z-G-OW{LE9Ristdw&~|>>997X$-Wo<O{Vj{mMtdkZF3U@IZEp>$*T)<q%|uIY4SPNn
zkZ0m)G15#t;~-fk`eTccWumuV=zP05Stdq$>pfFXeQ)7RGUKfQ34{L9Im=rEK61a%
zJ=fa`v#<QpmDe+Bl4oMgm+rdLTLWn)zF!Y3#iyUMP~6Wvr974|EbSSf(G*S3JcX!e
zPA?@2Z@+Zqr^!(jeO3$$S<zD7XTS^ptpP2G_kPc7cqS^vUbna2)3lE@hFU{=MokHE
zd@o?vc6rNhYS~J@X-Tg^rkuC^fgy<TCtTwhGnJwDy;-&l-U?3Fe{6&8M}QAq^Osv7
zb?r-!$&%k^SYx%9BI+?-y7qUIZ&r%?d(hlU3;k^yK*!#14W*{?mMuGd51Jck(eyo0
zZKOqWersqmzZdH@&$ohS19&Gz4XlUGT)s_v=zONXofO^w*0A=|%eS<{zrWy(e);pY
zLudOQG&lJso=yhcZND{?-pfnbfhsSZeZ90P!uFwiw3m_#`K@7mZ$rc8)6KV1J$&2t
z&^_Ox<XC^3ap%?Ei=D~fX>t~dJN-iU{x&MN(vysrmb|}*c<4&yr3l^hhqjcy=PBsO
zPlrZcG=H08(3R&~17QLAp+uj(HIS70rK>#e2YJz7e!tr@ERcHQ`@jiE2*E?y%$zS>
z?b{gJq@HN|t+%FL-Wn=70Uz2*|2D>;YvlJKdK>MfBaQjiKxo}d*^<pKUGsaI+9aQt
z&)`J%iB*1~&;I--`^0*NOl8e`VQ$ax%tCRG{nD|&r>Tv^XnMv>G)B|6F}4vIP3x_p
zpP~GH2J)M~W$~VV@As_u(^~`5PxYZIm+{uH5JP_`d*g4DopIqAGs!<u*Gn_{&o7<r
z`*>=TfZ};zB{_ZiIr%4^iIVga{h6moKhf7)0sH&orF;Bx3!{(u(wS#^Yp^tA-*naa
zz3&IzbH8-=^}UAO<eykio0ETHJ%c7|#r<uMLHBNN4dppI*ock3Xr=nr+1k?t2#u!i
z1uS&r{L)c!`-QIj%u*<gmh|51nPQ$n7p2iM-m<%t{oXI}|KF7C6ZPq0WS^+dm`V1D
zmS0**Yi|u_0X$#&oEB}8PxR+SkbI)QH+w9bXJ{k$#CU%{+ITND(|8OEQcukE(j|Yx
zhpsu^``T?j-HY54>uGRuPptmZT%H4c=-T-$TiX6!zHZiud;UUa`d+?nltxqih2n4j
z(B7ZML1^?D?GzXPKElMk_@%qN@O=68y%d=v-nQNNRen))9qThkp{-dTvjEl7`m{Lm
zuaDCTR8lzK6iwQ<hPr!tY3q;E%0@!Fk6-BQ->cWX6>nEwx*xBVjU4@Wn}IBo`K8Z+
z<|c>1d>d?k%}*tf8)AJKjZjop-?C?{TQ5DYJpV%X{+ghm8~~4pp5gM#4ijD9J}c1P
z(s=1e-<|`aBY!nGbe3<Q715Q~OIIuTtzoCqexaj`@0LVxw0>FjqTdJ0!2G>%1eR3S
zTkA25@=I&^c3BZ!ZN9X%p5{YO$N%LX`sm}Oqx7fE$v-j5OK1B&JcX|GZ4!%)?^8SI
zF5i2kqU-y#s3cNQeDBRcbNXHhiGDj2084q=oHgRx&sifrjldG|d-*Gh1kgjtY4hFz
z(w}eH(bJv}B@w|}13nS&ovoVJFLamu(%nn@erf2BmiY^vHPwgaTI(-#*HIo?dfgAK
zXZf68`t8^OZ1w5;q@nnp`=Gu5cCE9T9KX=n_gn8Voy&ab9{r`qSXeLJdFJub`92z%
zDCVp$-Rl`O$waa0OG{qk7dpRvR-m=+@(bO&zqFmk$qywx$S-t#?=L_{dxjxINYmag
zJ;V7+NlN`zdsfx!rDt7{zpsW6X=%UEQJ$#^CDL!F7134eOG~Ryqmz!J?qBHmc3Oe9
z()wF(Ti;GA(B67}DYCx#rK8W6&N9+3bd9Ig$w@K#OBuA*OUE-tlagXS-zzyO=9~}B
zZKd&05~9Cl%UI(rdyL9wm>$af)*jm0?k_##$ohrO^vqR5Va&hKo!;A>J*T&E5LnJH
zeMbzmNFPH0Dbn&#q)1D9>6ttp^8hK*nqFFJt-nyEa{s2ae|xS#QHlPA&Tr2ZXqWQ)
zLwC>Zp`-Vw$;nC4=Pz`Q^wL#F{e|xF-q<GJ>K7#|#Vo(jIr{^1eVUx46zl0AB&Ar-
zTt!lfHQpLdt^1{|?(){#_I{e2rQ%*MMX&o8x>9=|ZMO872g}orK6K=#$)#XEexWNr
za}}DT<@tiqBrQF^R8ha*G*rj`gzp?mVv1+DAumOJnw-28t-SPne*LYXueRQ@=-<9&
zXX<Udbo8f>k(gqnw}usZ=zq|Sn)shIlbPb%b47H_^3trR*N3k8JyT_JQ>^?_X-_=x
zm`?ej-)=!-YyU!f+x?}u#IBdllwXQ|?0o4?^QEhQJFY-C@_;X$`FkCk(I(CN7rMV4
zSD+bTw1?u@TVFcs)8r(msO=ZJ>wIbJExq;j+<t-cyOZWDp`I7+o?kk9f1eiD<oVL8
z{PC0SI_g_??{B}A(^iCkeos@G3>EWfaZ*&w{?hUz(EbPA>wB8Yq^Vd>gOR3U&0px;
z-=;xkxwz{ubnlTLiX7xy?<~JmJ6C$2frJSAq5CoJf#Z2v<Vo|>;^>p+{nBF>d>aNC
zfzod`4$&vh1X{~?#GUA>?Y%RSI=nUX^zF7Hic9)KM`_=SSm<i|rStJ4hOX4VXDR6D
z>!rJV`)7$hi@Hq0it)XS%}q&<)=S4MFU4DbzI4s?(%ROyX^=@-vDz<mtohc^@g;fb
z+Vy=l@@u?wrnj3nNz=cSn1k}tot{QVtTaD89ktTjexWlzgC=UF`FkA;-KD&=^r!zL
zR$88!3bE3fUwV3JduwRVPB66h@AEL|cy}P7vpq9amt1>$>F(vFtG^>Uc`N!eQ<1k~
zJTnz}D@OZ;?(tX$B(9k0r6oT@ESW3jc&n{-esBLl_j-$xuVp>7=eMJ?R@~{WmJpWl
z(*3MEGK<CiSO=(<rt)5Gt<>MLyS-KxvZeWZ6$?Gxp+7YL(vawuzLN^NrQgQ&|NYm0
zOMm^3ztUg-{@?zuzx;pypZ=K-On>FS{*4$*|L1@IkDvcV=m-DLzy8aA9+T;><p2Nq
z=l}j6QRsj3``>Toe^c$RH~ODE@xR++|2L#Hj;-*&{`wRDXZ#N(e>5=v7k}(D)cNPX
zpNjvT+W!Yh#5u4z1`C({j~{=xD00R2lf)AZ9WN4p{ZRsG$j#c?i`>8eC_}J-b_Tr6
z{ritHnH6J27dCF^f}f>s>f|VO!Ov15${U5^??21jIIbvk!H-g(Ej=&w0Y6KTy&n0L
zC=`GHSq3$942WCkf}f=bs3mkg3dP@l-g&$ygnl~N!hXIyG~u@Cn8n-rS?;n=vHQk;
zmixd@!TZ{N7W}|Z!OyGv;}$>kN6}9%?4LzH^hePjQp=x`7yc;tGeqxG@<TsG|Jbs<
zMSsvAMdPLZQS!nc1@DXfqvVJFC>pQ#kCGRD3O-}q|KT?tqspmPJ*e3?@jvaqWa)*S
zsrSc!zRb%g$#ma8iv;o#B%8>bp8qIvnV0$bd{O2y8iE7=#UWF4eEho#yPST%pAtXs
z`cv%ZeSgaSxcg7hANT($`SU40MSniWr|chZ>z4ibJkR_3@gn|ckKgNkTI7FzYB+!V
ze|3TX{9~ff0I>i1H-u3B59XI2^N8Q)kNdq-mRA~e{_J{DMgz+Hhp+3E{@rzLfcl58
z>y`f9b#1)+hp+3E{@ryodX|6u1$(W3du0vA|MV-4TL1RS8r=IoePyrpZ?CM$;-7xa
zQS0Ab8JE_7_NvtQw^uAuUjL`B>b3sul{F{+<In50{_T}D$JT%P%3kZ=UD>Xf{;6wz
zYWuG*rm1oLqgVX5Po4kub#0#gXRrII^S{2XP1^tLbw73f*Vnb_`=7n;r_RTYcr1k9
z4T*1a;}{SBj~2@R`MLbH_J7z*==h%>o!;DaTor4DmPs;E38^;WTiC=0pc44(Suhdt
zuebnId;z*CcmnsRG8zcks1D^4R=ZLKMwOFXrPaKQd$5UGBQUM|<3(N$<+U#>3a*XH
zY*iwQ9hcsnGf@(?ch881bbsdHpS;<j-3*+Z`;*Und17yb`>WQ*%8?#O#y1((_uc{V
zWD^|8Qy`(+pINJu_B>V7zXd9YzmogW6Az%<-#G8DNEF|j><O8Qic*qiO9u{4`zv15
zu8ogqe>wB2nsT>W647eCh~<*PDE8y?SK081yII8M)g`UT%kaOesN~)!*~H3K<dxcr
z``e2H{)*PJCu~Ngzj40LCC(FNhdGght0L9qsVRViBob92(IsFh<YuXO1ej(XY~jLE
zQR2-euZ%u_<3{RKN*`w}@#jhmZXXvaRG-WAVM@^G^Q&*l-i-c8OcjUvm3WC6ml=&C
z{fbJUrxQhoUs3wNwrf22S&p8!S*wz2$E`0ZnpWb3Je>q^k#3J~F67@)=I6gtWr+?z
zG}ZakFqrlSDN$TSrZJfcH=6u~BhE^Kg6>a4Bo!z56<0OS42mMZLh~zs7$(t)_I}!Z
z>e`<812#CtuT9zR<2t14)2~|_DXxRGxkiwoRLBN<Cof~gttt5+fOi}x_zV|g+&i*7
zYL#7PPoat!|7iog9@0t0ZCX-Y-dWxku3m~-eYJA$b7zh<x%ugW-oS|vI3=?uXhDd-
zk_n~izxua`S6tPvPiy+?!6%M9DD(|df;mK_ixF{owi3|fWSKipIWD42^{OPtrqT{0
zpEcb#ksHckQxHC187e9cPnX7CUdoxi0xI7uLxVB~8ysikHQ0uKq(n<`T3;n@z2a%k
zjSPL^)y2`#-Y4TKw@KdUD2Q)tQ^81xj`n`#43qI;yG%Avq4!n61}5T*^gbyjDS5hv
z_g12$(3fC@SjBX@fm2;r;q%2=sp$1HHqGYUjS#7Amx(NEXzr7fP6bzOQIN=XO(oT(
z?P1Ht&T}o3kw$Ny2;8dF!*Sb#BwW_I+w^>aq_5JIYaRj5*IflSK5hY(I*IuqG@y~%
zZ{!TA<b`eb3DNIwl*A=4uD(fl!`xJ}=Y2(IpIkATN#-6GE{-cJ`C~o7A1eFGQo=1Q
z&(npRJ}Ex*<g5FI@N;X{9GhfH4r}44vPv*uMx9lO8LE;e)%ba1p3}X%m@-oOOjMLx
zy~SO2l=LaMfBblGD9ht0Gn+FM+<{#w4h>F%p1F#YevB=Y!!;+L2iO4>SI_L7;{DK<
z5s%<H*%NpGQGB*p*@gL0)o+|la&d_j8q6pS>q5Pno-YuMUcJc`lREKFm?o8F^@lBv
zCFz!)iVj8nGh<Zif}X(`@q7i-*DcY5_;o|bQ}405iZ_D#l@rNICavA0&c4;^Q;vM`
za}wCuOkZ+5&P1<JpLlI6E<W`wFs!b1OD?i6RP~z`*IrVW!=BhKFOQIZ=OB{dyLuWC
zbo4W7!;Gq)5DYdshOCGVd+umt^!b)3cy1v@$&3h<Ca&$N!%)%hviDFI-agpk!p32{
z&1UODX$DpDsd%DQMD&r{(bUs=xo{&8!9ZU;E1(Q!jvuRzrDt4yC~&??%jW~p718+L
zYE(&HYGWn&$f|tus_~3E{(1;=8nz>u){JJ`$R}uqTK%@hMtVSS#zahr6S@#pS#K;=
z7;tmuSQU8dwi5wPD+{K*d`cm{iL-Sk{N-7}fP^RO!6P3Qy7^T*r^G}T|DXwOLJ^`$
zd)bFkVw<osYSA8c+$YIFWGEGK&IVlS!#Mo+7H|*%$j@hODejYChtqO0QghtWRM;6X
z*BWpUvIQ*Jg|~qOJN;FW^XvDPcH@|lpL^OGr15n_sG64`M5gdlIcXVFPLoR7uWe~;
zU<5-@ZZ1zThfcoY2)LQ<*-sF}#{*yOLq76=ij1#Wsd)l6*^d=(Bk!v~#l=^$V9gdK
zJATG$w3SobI%|a{tFaX820-$B>P0NK%8|(D{&XQGynuLAs^t9qeBq6#54y)?w!-1*
zm%aA5p*=|6zzvOLRypeU3WisWZW963*^pZER2fL)D-vGDvgZLr13$UdmONV&jF_KF
z5QnEiMin3D68(6Vdg1c!yYnrGDvZ7|>A4x}xB=+CbG~eBv!9P$IpVg-4%KZrFyju#
zPG&xz>3GDm%#JZkH=CZY2Pyna^}NJh%oYP4gTOm^?{RUm->4x|9&zJm)tRbwK%d~I
zYGUQ5lzbNw%%hTL2HzJ>G9}e4x83dVG*v0q1%@SW0Q%>BMpCnBP{HrjEdT8}x6r`%
zZ*EWjWMSk2spQ*^>-IDB>ZfNly5gsOlYXRPiF!L0C7rcpNcOC{DBow2GgPvid74?E
zqTKnGB3T%^_Z2ONN_K<2<F?P*y$L3k$PUj$MdwSJS;YuR!o8r9JR@Er0~A>q_&jql
zn*h!GiqWglBKy4)kEE`fl>oUpJf9bhTjT~27b18{Z<b`JQS#Ukw$e+*)4id0U$;IR
ze8Oum^mXWJ#Zzfbt?z90m-_JW0^c&V&$xeI2sd!q>?t47xNkb<L4c+1xa?Ard9{uY
z`9LL5W=QkwKIq#Q(u~l{Mjr9&y>x;j4^AJF@F&w|G<#|B9pX+#4hF8HWReaCn;Lu!
zd)N4R(Zh0^X_=)7Tx=S!=Iz-`6r}AN%SEo<JP{Kj_6e;9?mc-zTp+nl)iH8>pZJjx
z#Lj50zT|L`dx5L3v7@B-D!%Qw%uL<phrQE|yPO(3xc2JzE|<?O(uHuehP7`m<31v3
zsWd8ecBTY5pdL1Iwln8l$l6<tK?^&~N`PZacz=<s$EO({+(dJDKGtm(N-0rgRor@2
zT5Iq7gcvP5T_sUx{F*H^%^0h!4_u@%mn~C$4rHl?DuZl5PoRiieIn5U{k*jqw|Sl?
z$&Mm6dO0LMP;<dq37052*(YLrsHH|hnNKiw2b^tL4TV{}#ZB&MBgR>q+KvFy^%Xe=
z8ofrp!b^<w3d{w&M8SMUUJX|x`Dz8(`dn<9KI27;UnA`$cs`91C2OrV+7Z^!0xC#d
z(_%6M9nJcV41>(AJOlQO3PZJopYika%{3naKet38)8_p$etmwEY$9sq{+c9a8b`9^
zFAT1%4$O|(x{rR=PrKMCpYL_HLen*07PCo#*sRqOG1_&AAsT96Qcw+NS(X|GX?#1x
z4^4H-GyK{1APV($+BS@(c7p6_Gs$_sD=u*?5;K~|t|NZiGMWpN+@9hLILWq(A>J-K
z6G&Ds?@X!z_qJUemJ-F`=*TUwBr3k|iY|R2eL~f^<t;vWuxe|Kwszd+YGeq(Jdv4g
zaHnJ+W7+EiNA`WhUEE$;_(pea@HXE(fH6YSK5Rre%Zfl}Q8Io19O3yy(gePFEoU7P
zlFBAm8zZiA3<O*Dk;Nu1Q<p!DR<6bx`(?JAswH4gq)Vt>LF{pv^G#=sEommVnsM+`
z=j}?eJ_DY?*Y2qI7H~FFC1@|s6aFDLudwqfpU_+98tk}LI{fFBwt~(%py;>R_99CH
ztPQ_7kwL>&eP>DGlM#VIV`(MHaJ2)DY0JJETeZl~>lJ7f?5QAuEGn(wjO%E%KvK}@
z-#jrvHl(&?`^zBq^ymR3@N;$=zziWi=%}u<P}g|jn_LLaH`MqUWP6zKEU&hi%(dik
z;Y?;SpXG{w5$rD|Hwkx(DvxPB^f}GF;?jgl+b6a>c`#D*iY?dM_NM_(ECHG^?-3<)
z#U!T(dHPb=_ReeSXa?~l(yvL&kR4VgJl#<9#{n8TFDM0S)%q@i;Zb4R)o*d6xH1vm
z`n(cv)uAjo-f2wou-U9}&)hts#<Px{+xF!B#00o~W^yAM$AB)&S}3l3y0Fb6XL&!P
zLUK9E8aTGSbpQnAtBz_U?QR_$xRK4N(ogsMA=F2UrXqK#?C?w|x!Y+%@#mI_XA2Ti
zAUUf_RL)e{J=rwy3{Wy<aeWVXrczt&HH~-%G1017@ow4!wOxUo>;79P$yb|Lj=i(u
zamL<m=ndO0Kt*0b<1z*6j2|zIbQ#Qko(NmXR(q!lk9eBh_;a-g46#BYn{6gk^|Q<+
z!@Vq9oB8%3>hXmX2N<sQxq=p_gEf7U$)X(JbUwgw;<D`rG~$z7=}45|#wbwv8mlx@
zM!=>paVB%=M>|~AmV3P=e~X3mHA5FFEd@J9$(AX5zLXVgG+XQ)o5{ZhLV22*%xlJ#
zCl2YntmYOFa`CO3y?h$J#}jU{a8}h(qA8?bKrhW^E+7<NI7iW9*0VPP6%)?tVKwi$
zn1Gx>)vO5j;jkzooI!4Fo2h_<Onc1Hwg5ySUYRU0B6?d<V_nmBwQ{?^JzuMzL&XU-
zLhy-~#Mi!kX$2g-4{h;z(sI<{6$mcFWW=w}trgoKOulQrbUN5I`;=wpNNB((bzHK;
zHk+n^Gubf-Uv8A#N@Q)vB=y2G=;}~|Otr(kbD8}qH$Ar|lJI!rGw(4i1|#;HLyhWo
z1#<8;+abx?b%)b7My`(0)9x#NUDc7;WX?1F=@+a6chBz%oYfE9`S=_ji?r(a11*{h
zH~9*Hd&R?y&Qi2Wn#pI2#+gacPWrsV9ML&=^>Rnj6f$_4br~cdcfc)&hQ*8+x4sqa
z&0R*aR3zZfmu;uvIYx5L2|8zXI~(=)I;!5+c~rk~LUM+k^7)Pp%J1Eb)zQ5_5{)y>
zKw~W$^;H7A>C4S0O}5kL^fipH!hxH-1Ps?CfOptC$&{Ha$RyzKEFx`$JFlh19Ul9_
zOy$N-fx}rzO29}O&K1CoGXrCD%H}D6*n4NdW4Ir+M#-}<vKa)Sq#1Crq_LeG+l@u)
zT~Xb-kRJn~>7H`(^YP+?`(Qu8&)oOu0gG%DN-j^13)JSe0-wD54g4aC8@A03?@)8s
zww+`}S`<8HyjefplL({lUaBqh>KU|KKrLah(5mluqwe0SEo0h<>1&W<#wL+7gtU0Z
zIporCrba&vLD^k#+*(%o;Tdo>Gx7Rz6Bb?f*6oOdbh~z&MB8*bbKkyY4IABTomjM>
z>RyJ8bcA8|fNR(R$?$V<B;6IgO@+x9TWGf@c{DVo--^rB!51hSt=#)*#X9hgyxg!g
zZHnIEYJiw~TXx=l%*-9v8EM((;;~!3TTgPzjBRtYXHP`QT_dTyBz*4#sIYDAO?vKB
zbllI_yKIB8ji?iwO{<Ce<3jrkzB=qBXPw>WBAZOJ<M|lQ84Y73=<!XRwN)pdMpeB3
zh>|6%18?>a{Nv>iTRsg-`_9W@tvD*_2)5PuIpR4wBi#K1&n(^TSa&vmZH#B~X+Ttb
zUZyy0e%*QvI2cPAu#`E@8H&d+Wx6$XS}SQXIYy9wk7pH?-KT*Lx&uy5*Q&<M9*@Jh
z;<Ck2>G5-^{_s8aFqk;L?6&CXDJCfVM!PSLQu0w}6Sk5xfuF1RLSd6QrVcw&+dWE_
zJ$ympJ3*+mcTx)YCcd*XD%b%@p6(a}oYTyp<Y8tgwbyCKPp5HP6vXh<W@JykosCYr
zVzRNUO7|DRLBH*zvKJYTI2DkMrH7A(!&Mm&u^y8j3jooWu1OZ(bhc$jahUD<bjY(8
zbzfFG%|;U7VyLNdK&t_BBZmOTVWZnOjtYAsg@6X9WTppj$g80!>uZ#(ip16tTd^kE
z@|4?JYScJpMweZ2*g)l`GT0M6LhBD-VFO38Mak;ksnKOG)lNy)^kXTYJUf_tw!aHh
z*;}<!67oq@hl|u^+WangU#I|V-WvDeaJwnK2nG+f$n;&d4*skl2QJ-=?%_*sbk~WE
zq}{6LPlvhn8V*JXu}eZj_Wm}D7I;Qry@}ZQMulBlE;>!=Iey&~axEXOaj$$PoMI++
z@Ao=g+2ys%XR{HEvnMTx+=~SPb_H8&k>nczJG*y|c5c$Wp*fcAzvbmq^VXqh(j;@*
zyIi=ot*F^<jc42qKNa~YHZRE{i=!DV%i$hMz$q6*a*%ABJy2e+Zq<^qWd&4>wMOOE
zqGZv+q)RfQI;%O`j9t=-A6G3H@NDzFeH7Q_FhOe^d#?mEUq>XgYl&Z@r>~!{ol4yK
zP-wLbgM6<uX0;22b0}Kt<^o1$D%-Ar?dn!vo%b@!ypgb+{P+Z=_1yfty^j>v-g1nQ
zIW@e_O1>96Z=dc&2s`iDToTS&U^&d_*J&n<lTlRGGj=LloMU+tacDbyZ8FoXvsQD@
z8);pnKx!>HmYL}2t9Q*XX9)KZgH5SvlSx-}vBQb?v7AQZJycS>ELqf38;cJ2G8Z{Q
z)dNn>Xw<Drza^I<sO3FYTU=<_aRHef+l&is#ZSw9tc+-(ZClq%j%`+wm6rA*<OXZ@
z6r$>ErP_tu>lnDSHpWkFnste$x^3$*(+%%L+J?%S!_R1C8;U^NXOevIh&oZ8v-h)n
zLkhTbDok3SHR3{@^|+RV?vz8c8vT|ZclWG?v6j`MrTY49Yx*>+8n3$=&{xN2-d%B%
zOYY%X9j6inH+H^n>+}mpv$K-2+&Kgd^)vk1&V7n6x0A|c+urW*ojnxPF@EeD-?saN
zR(e7aK;F7NU3f-dIs2CI9YR07pe%w>)7%RY$dq6%rsbAwjIA;u+-n}0bjBz1jvevi
zGaA=oUNt?%tt0lAZHlV*GG`FAQ{0t(J&JBR`xES&>|KMcWHaFz7|k?=@oU@ps4edC
z7`w^$e3QR3R$SYrOeIx-vEbUa2COMw4+}0rOXWx9N9VwqYz=sfUR!}8mR^&&jG!!2
z0&c!|&CMl6$=b7>m!vWMlO7!PnkNA1h+4pzZ4Fq8_e&h!i#c)O$`EW9TLEGjWrxRV
z8O1kqOWe>z&rv6rZN`JoS#62JPF`FFuQQt`&|40;tGRI1;_lf<q<1;xaMbF44clfM
z&3na9YG7vw-Zq!|aS8J16$yXsp4Q9G5Nvb%@Kq(+=u))Mkxt4{WzE>)S_+L@lI$pE
z?I!oLM;iSM!R9wT{I~>o-ng#|`;M#kj?cXUGFpj}BhP9c);X-CNp=<+*Uu}r<rTu{
zXH>EQX66+o=U&A_6Y{-JR;a#><(rA9qUU9Ec8e*JB^R3ga>*~=*;#VgO6acTVW+be
zb68wHV6x=e+Z=mSjJJ#KT(+&^@~yIH#?!U4f2NC!EK6`MKYv8^JZ0PRik(c@kbeVb
zYGnlE40hPG;JygJg{u?}zsvWM5JGRocDzn!w(V#n&l6ti?07|ZF>omQbW5u<>OAdh
zv&wlBGp^1?vS7=;{zV17Rhw1LFzW5G>AP7I2>U7{;AqW)YpTy5gt51oZGS5an@y~X
zEd$&1^DOkwo0$wv*)fgXF=7;RafO{R*tXo=M*!t>|8zf>ICGe-0FMdmqvht@(T@Al
zaOX4B&C~Snu}y!A?AlFR0iIRKif1`?v8@2lLNG<0Qyc9}T`M}<SzVnG*tgvFz4{s3
z%)22cKRE;Hla=A;N4y!y^RlxoM@s#`g`b72ykbKhG&lEeNY&54p(V2gPmAPaD@M`D
z$-73<WtJ6`Wt5((xf{ITPIfxA`m}ioDgoE3Tp9I38=i57PV@f$tW)Q*k9F4>bdUmt
zL)32bpV($St>B_k?!%r>cgRH;_oMs!m5asn?Cy(1(J5aY2I0PLy;}z@I<fUrF40|W
z25*RQ89ghoPB@rs65kDiW^L+gU~KVZIK4GHlrM9xTej40E0^={e9M;IZRyf`>oaKW
z-MVX^b(NqkcXLnd;Z-MkzTRy6VXfS&^=jt^XJ}5hXvfF1Dy7U0c2<;W+sk$DBB)K1
zco5Y3fpgmLPnYjK=sLr*ZJzc0x&GSAed*Gt5w~^_ZJO_1ltURmLp}?wv@f0Vfshwz
zSwYXLs&(^I%ROIuM(OScV&m3u7QRW#>ydN6^i1~E{OodDx8AxiMhez)ZUIkk--5^d
z>z8(&+Lk3#q+EK|bOqYoxfXa{%o!RA&elFrh~mdWt*NB@dnH3IX9^p;2U}Q>1>cnV
zxzJL`R?+_Nr%stGh0Szd`3ut9|0+6uC|)aVk3&RZ<<;{DdQ9MB0~D-Uk6;lU_&gR&
z<&tO6B3q>pOxN~>+A=Ouo{INX$+U8Yfzow;m4mYV-za-buW`$w#w?n{a)DxVby|T9
z>wCL>jWZV%Y=R%Rd%Eyxs3P&I82{ucd&Z<f4%(D&y>p<&&z+@R0-ioQ3DV|RzkcY_
z)6U)cecathIYG%>;%jbNEa_fXivdML=~ma;ImRB=Y+K!BXLoWCT;!XeYh-sfhnB|p
zR!lYc(wf<u;s%g!y_6*L>N8`cds|K0V9=G~l`~%8=TkUD+VRF1huD8F(qA4q=ukWu
zjZkv^P<b#HZPbrj?<ASCy|LMDD@>viRl}&Cr&#0=*0Y7>0X<V&GWt<z!lbO{(nWzb
z(YjskvTDimg_RFwgyY6?JVR&4^Oy*{@v&kW>4eaAzLBCOe88z5t|Pv9u>LO@@TEzP
z@iNjX0?)gP2@eZ}UGBn)`A4mtY<|#V|D_Kl58huzE1G$)HMB`t@Nww3+A~1Y-uP1F
zexHCCXi;3lZ?%i8Uw>utxYT{CMVh=rSFZEcyU928SJ9?$zhyV6nsl|%VxS^^dYvZK
z)#4=(5XqV9fo$&8OO;=zu<9t$?O4{G5l*nhyBX3O+M>n1B%qa)C^StfAJ~}JOZiv0
zG>egi188Y#owAIwN@IZz9VCB%k_+SB0DE$aCOC6z-a#ih2cXLjrv*{#*IoH(DR4C$
z&S{oB+r3xz0PKbB^r>~Dtxl~ihmkXJPpfLe^L?d!DT>|{jX)+~*2rJ7LKSP!f{wKA
zpw)aUD2OoYx%#!8h`MCAyGU_3T1Ih-43yY=Ip3=_4!Mgq?PF@1gxQ^mHY-O@%%ZHt
zSoCGi*n7QY2aA_X>YyDhGYPfpDq4~s{y@*yc818j6or4E?0hVz`(&0RY21q~G##N`
z%EC@Ja){h62ch@LHcEof*q%dwCN<8jdE7nL)ek%WANC%~jQ+U@xwBp?)n$jJ(ER!w
zYrSx7YcEtqTTs^Ig_|8|ekvd)up*N|s>2VRM95^@<NMMzL(;J9_(pklmFy8H0UP)l
zWULks)B(<HO-S<ZTM}7=rK0;<7z;<{GfiemAVkA1V(!k{ll)@4mp^vJ>A`3{=KC>g
zX_-vWi!`H`PE1KIHtx5u3BWssDsOS*7*(|sxh1$YbDg(dA~lD)7YB7gYi3!<kd|ud
zmreo<>7e@#m2A3t+nleo->Z=;q3JcGnl9~~<7^st%ox*7C+6<$DOOm!hf*TTDdzLQ
zXMK2HN+u81t{;lGAul)YeBtW$Zpi>}n}&dAy@#{3_%y<?1x{DtJuIK@&@)%`qSQ6r
z4O4WZF_Ad|iSHRb^N#F(wtey*^t|#jOCao;JdR9sav&nP*X82`g65i&kbBL(VJx~y
zE&=t8Iba*Ots$z~wvV7FnisStvZgM#QlN{~Au6TAeb}I*5m|=5ZR<V@p<SNPn?OEz
z$8ZZB%s$fj4SO3!^E04lXe!G)X_g+6s-0u6TXfUmDVaz7KL1N+!IJZJ^9_cYNVZ7*
z955Gbf3%&?_jrpQQ}j$;Xh!G!so)qxXKsQH_rHMRakY?CJ43vG`p|)GWTNEz(k6#-
zS_r9xOBV_h{GtB1voozTR&xFo>l>8I6W(Xs3?4C;YU_eOOy3WM&Vt8Lwgk-gYJ{IO
z-NWz$bLeWgZK&nar-q%-+}B(3;5yNP6b1Ea0v2Vy=g5OcFY)O>*~mkd+>f}k8fxhg
z*qFMIm{vXQaJLb`fgD9vOp*`x{H1JbaZHU{wl1PtkrbaLdI#&lZS~HTqT8O8U&f}E
zCi(o_!s*yeePQktPg{uIo9bYwbl<XLXU0nw>-F<wr?RR~CzE_U=*omT@s*8T?wt|X
zt9rwy3y8<Lx;Xm+`_f&yeNWY;Lvq&E_tKZMYh9O;!NtC()b@u?VmlBO-_<ug=<{?~
z<gI)6`2o1q%Lx#0*A{5)&Ay(A3s2C-W+#m>N54gx!j>wC16SxVF0$sLwSCuIba$Jf
zj?rIkS&}iUhFbf**<(-G40WuleDC&H4mMtP_vr+luXp%V<Zc6^o6QoaasfL%o#wm~
z3>DmrjWU>U4uw(E-496=$2d(}ECGXZaBp))7@WY3a|h;mi%(kXmD5-9mdAe!S~Kn)
zSpA<7j8!4w1DJa*iU4dZ3uLE^m>gIWn#D|r$qd*Bt1%8!>$uk+JU8o}9kw}eA-2y7
zC-2YiXG6GxvN^d~NdHUdt`xqIcCUX1=%LMtV?4R}MyD*O?H4*(F1K0E1Xx6m!5p2d
zdq5F)n{y17%xZkcg-%jFa8xtoTkdX@Y5HPEq2SKA=B>8K&Aaef?)aF{r{Xi_8Cq_a
zqNG&yy#0=J9Ft%ov6oWwzV(vMsAy4{d%iX39D_yaXJ}{pqTBlOTr(zPLPKniXDB8)
zASd!_@Dc+sAwr{S{d@kMZqcO(LN<wVtn5)RZ(LaIbxj8(We|cWb~uzT9Z2nSU#-U5
z8JN#YJ3o2(A(kj1@hUreG{PYC=@?)%Lh|DaDE^>b?>A*%7|S@o0fc)PquzQO{*G#~
z=1vcyMBhU9h@A*D%p1BE^s$~k^`dHJuP3<=i6}vpdWt*eAQ#2PA8$STNuDUHNws;q
z54M<XY1>(2Jm8#MeVlF7SfW7L2T>A?=!#P(!}=U;<_W3@tirx#{ggdcE=yIMZc6S!
z*{#vC6q|coN^FQ~m?=h-NkHIK4W}ctTecFbz=+cNthZhU673_D6`P2Ro4C*0KEgE}
zv^d!s>2^-Hf>DW#WMNxYz7H`rULi%UQ}_2%UPfw#YZPHW4_mBu<p2-L`Jo2YI&K9@
z?ZrqZ+dUv|m%0HZA`nIVXKF>abEKTb)Xzu+6lT=exra`|N9YC9m5HDQhcsSpPww+1
zy79Oy3R{iIO$tKOKBqPJy@i%Bz>A^;qIa<<WP3?hTVY<aD8zw~PP<}<S_@)4y~@z~
z@|ML!)2Ot2PWR4c!A*ouuiTR73JjLY#xO5k4VQkNL816gcf!F`!>#+6i;k?3d3`2_
z%^q^nw^XX@!HIH?%eG*hF{Es~H%GQ@%vvwi32Wg%xP<w&|0TPW?Z3I=6W@3sH|@FR
zmGAY5)Y16IV)Yhn4aJ*++kE@nz_HJ&8um<_zM_K;97ePYTFb3>6httJvb)FkOzZ)<
zFtl~<;U&<ilLYGBecpTElvpq9GY4#TIkQe5R+}-TZIY-kRB|t4-D*)<G7*Lq2P<rW
z=(v(&ksbY(Z<B?%xrKU^il{3Rl<B^rz{2Vtcfu5lEa0+W_Fbb5Y_p@w^`vfhjey!g
zcUJhp2aataXxrUcfW62USZR;ahhk=#I*CV>Lq~I6j8<HH2|9DK1E*uxx;+G)$>YUS
z`rIwDfy125V-A5bFq?Zv?2<D)^xS%gLO3O8KbsoRx&|b4%9DgBGLKuy+R)=L7SG;P
zLrFi&ZmL!`&bQc;?4glyx4N{a=pq^C3-V45b!`Y7o=QLyKg2mWPKSWInLYz2zyUj(
z9#UCIA9y|uIg6$=J`X!sywekr*|<f|)z5v?KT9e1Qm3nS(eNSyp23MD6m8LrTfs3R
z7$rC$U-OpTbGTD%tMRObFvHnQ+{e9dMfbD_wS7;)56zmWC`&?{ptwz#l*V*bGSN9b
zkUYXtp{ET^kEb<rQxM$J&0x~}QS_|Lm%h9++WQ#JES|SX1f_c}Tog=Mx4p*OX;>3V
zJtHe*<CeAfzSM6EDBq)eg~LSR$e>6+gQamE_ny`nzn%7UD>!|?6dumR<YyX%h|%2S
z35>JJzH}g>QP{I>(S)cx;uBFcyvM7-I}10HBGIjMX^|RY!M*2x=2&vYhRPmM&Z48*
z;b)2yXy%hryNaGQ-nd2E!qtRR#vaG;7^w}7T#i4xM@vA_LDgyC8NJ#qiw>sfxxaF6
zcNR96I8{yEmx<72hm7oC-&<H5k?2%*?)&@mrs&~_ruF-@NZP=5P66eBVpG%UoiY4!
zB7}Q&k#~Q*96|=CH8?xqJ19;RDogb1EnDc<`kGE#+^d}-8dHE!>lNb(2UzSK7IRIx
zHDGUa=w~ZNb`1KoI-!dXniU~qEJu9XV<NJaAcRM9>elAxf!WMcD;{fh-5O?R3#G^-
zFXk53ARN1@Z2C0LQY&EM-pP%P3fxb)n7g+M^4aOAHYPD5wDbR8+qvv&k|f7@-%oLt
zGmArI+&#7gNCMoA)H{j`fgB0~G?WG;sMl9~=HZ^Fs;dd82>~<RT~_gDM&2U)Vs7sF
z%(z#Se3N+55%UW}0g}L_a`3m9rN~&Ro=RLZykF2w;+nT894r#*16oE^)8u#rL@wAT
zxe?E0WeQ!*ou-`&#1MBt3BnZ>OF}o)ld7`Fo9V@f8Ql%9V<ucbmk)rTzdO}4hpxpk
zP$htLV|NF&W5#vQJBF<cx{=#p6<yDJ4QEl`o(iljRtbIDo^c{`SrT3D1DHkdAH$|i
zPdM$-N*>@${PColiY70rnpoLNPpovrbk<EcMWuz%g(a!HPH=#Dr5fTJ?4}vo09D;E
z8(*m-QFbPwy~g{2FSy-)?U}~{T?lZnl8T>s(c{@I1rd%h@{T>%{wQB`jP>`p$ondB
zw}OiZO3G6YqLwS}BGg1af%redA|lx@@(DEpB}_V0B1Jx7qm7KBZ0BO~DUZDzD-}AC
zwRkWmr%T=fEsp9;{{UO(scPsAP_e!})dK@ouE>?piThB!qAyX-pm+Q{x_1JSq^qHk
zd%<JW7CA`gtzL;9AFS&DXRU|@m=Ce9>B;rr4DiubG9OhW)4YVUbhYVzR~~eCq?xP8
z&DYUq^Lj4%B$J6%MI!a`RC;W9Mopu0A>fi(qJ}I-mG)21!@_NB?n*s9kE_(VKDH8O
zqEXkl+&wD$x<?1?6){kjv0dJYbp(&J-d&dk;!*dtWB17LVTC;;E%fu%^Hg&p&>^ml
z)kGIk>{@)E{mzw=U606>@Uul|dfaP*Hk4z}M%R|#nWyKS*-PvbHhl{#jTA9T9U{<J
zf6-msYdY)1jZDzbVHCFo&W0Sz4<eG6eZnf<5?z;la(sta5|EJ;`-H&7$dl1i+IOb;
z%tZTWj~f%O4NN@M30A}rUl5rt&8tT;Pc{?e)~Y7#J?AyI8zb#0HK!&0Nv<HgTx{gh
zo$eDVMgkYqI9{`n8!pNB;bmN9$LmSLQ}RZSDv15jV9*bVcNE1<EKn~45uGk}QP3s%
z4q`J`U96~7jEbqM+Q)ZnRmrO^rMzcr2tk3K5BDxpsv1PY19ZeU3M!A2|7ceP6?C}3
z&!ZhJ=ZkgZg>acm!yq3NALzJk=ggK?Q$7~;@(uJD??RtYr~(zegKr4Mga9`RV6t{E
z`&uCN%z!(-ax7{&YrLrastF2}loVt8<r0IIF8T?IKe6LT8BX*QVjSz*1E7-%PC5Jg
zE~~fF(I3x^34*CaPGy}DH<$YYqe!mBZh6%4TwM=q;Z=irg)XVQYh=jp#x#(35U(Am
zU1^W=bnj<Qxj~1yUPZ|sc`{S!qBxY1(Ey!<9$3rIyz|0DQ+FFW&xL5LK<Kk0I=4M=
zWif)zV!s0{6i4ZEuGzDXsyb&&maw34_NdbFgN7>yYSV3aBjFxsqh!S08_S_x0TCYf
zeAbIjJ|bSDC(Jk3EpzjASSxu2Ly|0<EaKqVdX8<QM9fGsW#XPl?5pIJE#?1wlWd6^
zd}8?=^J*Ljg^jD@u1j?X-7s1G1)uQ{Tn(<u_1jk`L+E;(r*|Gle55E*s#3M=8C?=p
z>e>924V#c^8-m38KJMB_WsN>z?S|u;^{%L@vm-m2Lwlo8maoGi0j_HM>gB|tyH-@8
z*u32J+3)O;bI{}3b>~Fk@6W54X*tw|D0R??)Jg|J@*mGNEidZ#f<<LKEoZQml2+8L
zy?213+IEE@V>O{{b9mWy8<8OJny%$Ri5Y2E7b5h^ZPGs@P{clphgP0gm1zQik<6-G
zZ%2iky=7QjO|veHyK4xN!QI{6ZEy?jZiBl!gy8P(?he5nf;$A40Kw%W&-1=}pX)pS
z&iv@<uI{?4R@Kxj>+Z_8D!f5v1+9K$B_K%dx=z)-g|0>fuNQ&^xA&{ph*hUj(h=&o
zz8P{MTty~@Muz}pEZGZoC1&<2-;WfVU9&5xya8zhUb{4FC=z28$ynfb%f6K53u#p4
zmb#w&Cb?kU|ExfB>i(4_5uXi`{!iO%k@^i7wE2SUvIbpatE&}gx9EWPBM&)0!ouoE
z0^y#)v)!W1c%3dzD?*FC{5N0)kr2Pc?+NbcYB;(}PtDqecs>S)KMY%JuoeugAsQ(S
zx__*KGoy|@s*BDQ;s>P2lz+^2zyeRlps`$U7vWb}MUHbDys$Sh>(pfgZj+qreJd_P
zR<Fj4=c2fOa>`UaDz1MW+$<SZR08@bkv|m@cv%Ms>2%Qpe^J$$&r=$BzW(;eu_oGa
zM{gmrUzy@|4?}xcakRF;c3>Ah>AqIVZIEFLcCuajc#7%t!rBDqSf;L7PaF#)C1e*S
zVy7^E7zx?E<>in}rJ)`+^_bsW<Ng}iBtz!Xvr`ZPt7xkh?*s?F`SonJ-kYyC4(Sc0
zv3Zvq!)0(uNG=TQyG=)Ukjg1VpPGL8b?N?cN0bvZwphN{;vf2{C?Z<qa>YfLu;YSE
z<Kb#om#7@gtCi!`B+WP*J51b}fn{@I&+A&*eJ(tv)2u*`K|+tP?9^mfKg=zB=S&aO
zyzC?v0T_<wi!I#<F!b%un?9i(ixB*r75YjAUo{L>QQn<2eu9;u1~rPMM;D3d#HWVK
zbniJ?k}~PGgIW%G{w!?dtsQ-SW><9O5ND*Yt|^ix1?P%A!DH%|%1h{VdX%znB06GW
zwEx~Z0<F<N>-5Q&Ay*T9taKq+G!A=@zlBGuNT_QI#(F#&f#7)9#vU<**@+R?&%tYw
zlV2R-_A`5CF%_S6eOvo|)RBfRTBEu>r&58xa_&PajbzNiSJ4aQ6Oz`l8-5{;C6w+a
zyy&W8%_G;zto?AHl2*l1eTv#m?Eo1Q#IiaHe{mABePGjdDVDg0aMj(G5G7t5MZL;6
zBmt*k*-KMMXFBk7lO_^NN^js|W6eS9C<HfB<V$lRg}{Zb#R7iu)>9Xu7mkmFL0DP8
zU~1*2$AtCju%{%=L%#CHGkxjJs82nK1vFR5{r$Gibl%}>jw`gHb8%&sz-`v+WBbW@
z$PEgg*?nu=)z2Qptnte~?)hi=^QOID&yw!2c$HW+y!fX!cr{LcW%NHeEx(=(s|la-
zLL%8{+<g`9zZ9P={|bS%lOtvumTz|!P5AQxvtusF54`H&_vTkGu|Xmbe>Wq2gymN@
zPf2lHBtwlm&$z5Z$-eQTO1gYpUB{V{R(jQ%QSkQilt@v+F;?Pcr>L<4E0kM4ATAF6
zo;0W|DAQzRt1R1eNJqIBN5+A@1;WhChnIT$;>}X`61lhelrzN}GC1q}mZOL(d1j-W
z$M5LbPK`Var%vlYAmudEWfc`k<L64AEcxxa7vc7Ligl~YA8cwn+9i(b)+~tt-*pEF
zp_CS9|4a=Cg6v(16oEq?)r?c`ZN7JcZjHEGIZqdHxFK2fanbC_p`jF+aQC3kF|Un5
zDLUr#pp1?Z+Lbp#Mu)9Jy$+Eq^WOnIOC*bKSsp+a|GvWs&mXb4ts8>!KDaY7q|T^~
zHshoV)a~%`E3`pS2*dqZhKLnavL+kwXr__`RRfU$`bD2C_Pu#`)5zY+EgpY|<YkZI
z$8vHn@w%Q2skd6(5mXq|GPw8PcJlnd@Li}&aB442crFMJ6&FFrDJx&9_5f33*lYcH
zKV>7DTUd@GxGAQM1w8BHr@G6hN=PjkCCPx`>A>{H(S{CCe+|%-^g;!SBb9gG!Mac6
zglA5rS(L~I?=lj113**w1%(xIxeIz;PYj6!A*<KzT{ll(mkL!hCnhO<m0WGGLA1GM
zBH*)`eUBD=PBQ%VQ@yolNxvhv8FRTi9Gd@R=OnvEy0$3o^=-rRY}8aUE(dEUAe2ko
zw(Y@szEx{Fgn%RWVL-2G#FFWDcD}Low-KrdF2E&1!bM`0u40sLH|urF48PFmImn_^
zP_~a&Cfp+K*aAIEUCoIl4vfE?!0&4L<G}}2e@oR^&%QqztviI5BQ(4yU4Yss7a1D&
z62?#5*6=W1`h1=^!Z{r>uglKpUcsk3x|mx5`rfjUlYg$Ma%>=!K(m>;qPL)Sl-Q6_
zr!G1)hh6Ek_!dk*EB`~^WUfP&%;6{?4&9!PA8D{yxuJ_m14x*DpnSEvqD(n~09<^i
zJuI9$IUSr|QJ*?~X_>#)M&0|8_+)&b<4m0g2aA)B3C7S=JU;@4mr!25a<8|7@h%^T
znQZ)V_Jb7=g#}lw!cmhW{8K>42pm))Jo&uxQF(;X>=CwyDpCVSt@~vqlvZ5=2;5sN
zJM`fTt2bMxYrT$>l#jF5#W+{yn*W^PmO&1+%0wFw(kd$HjoBiz9#M~xm4A;BL$5>z
zMgxvg#T!p{dJ=zBvHy5xG@yaCa$ltS1%m%evw12Fp*F;ZVujxERuIzSqQAp~-u!iz
z=%lO-CYxM#>rQFi%ws=h0KHNopmD5gC?SgiHaz3Xf$rWdp@RB$l+LbXy3eA;lfQI5
z_m`}U8Gtti1z1-~GakFBsAL}hV};;jH#N>btFcDh0mfJK*gx$M!_XIIBGHzpG_<Ak
z$1$7nL{3NwCysNZ1`mY}+xdJ#k{Khz1T%TQ%whsr506tYEX!j9k?KjmVTk?cu<*cA
zRsYGmDf!ZSF6Prdo+vbp(IGY!>Tefe2eqBsef#wt?NCg=Qs(`M(98R+tn=PCU*a=X
zT6Fn^-tX@+dQCk1onqODpywxcI`=!m5CwhTT0&=tLwetdgq`QCUTx3OBGh(_CJAiR
zBkOVz(9djgwTNXVfOPL*7_kOPH6O4Ot?1(^-!gr!IkLW<2_bl*FeTJNzv?f-iQOHK
z9~Z+hw$5mLCkg>n&#rJ#w$QjTaV*|SGwj}dnbuvy=e7?Wyv0qKqFbu7;keGP1;T_k
zpzLmq)31pz!0v>_+DG9O+=N`0gxp<mqj@oImTlM7iIrAN)<J`&#;{@Lz+bT97tN>G
ztrN6LBhilogpCqtJ5b<*fN)esz#XcS#I^4>qMaXqtcSP8FQrBR3gyKh0!(|ey&`Wa
zQ&1q`ph7P7RQMCJWjW7qSgQ(Au41%Sb>*bnS!tX`{`Sm;n7lpLXOdHX{I1!ils|-N
zHvy66j@l5zlXWBWt*>{3!@M)UE1!sV$`427TF@;!b0LvId~194i0mvAqN?gO5b!&O
z0V;u%q7<(A!@8{Iyk`~gbPHteESFgw7U#h=VrB5n1O8GpKj5F8G&6FYTOo$Ed@f2j
zff@MO%d|LN=%vV%CoOPV$g33L&Ln|9NAeudGg+?rE{o(pQ;jX_v>;;~m1FRsqZj@v
znjKx%0=<r7eIs-1VE8QPZ+DQ<)`$*uoM+I3e|_uLY3y+zuikf*hXGy+y|&OwCTY})
z?FL`a?y`+xcaQSqUDvijW^ryA(M&Un*~MB%kvCN0mPu1L$n)wGEH38;`2FZEkGYt^
zue3VJWnW4jUbCP4bLu&ggeZi+)A`v-Hi%~^!1f6Ki(@T)q(lEXygeaSfJ2^jx7*dj
zayNH;|9G7jr#^Mu4Zdt(q;N(`*E-*RZnKmTC7n_H2-qSFaucAwAjqef*O*bBOQ?!+
z0c_$sDZb3I(OQ@i$H!fRSV?3Wle2<(R!<4d<B$9=h<jpZy=x$qqvX|y)FtkKvKL7;
zy8Y2SJ#A`F`~3tVE%%UWM#<Xd&SEPVTVD%DCv4pGkJTQlrR^sLUUnHttk%udl8OZB
zB8sY20jff-_<#J9&9oYDr`2mBJIZi&76K%tAY*ARiPasBJ5eclzGgU-=47?C%2W9w
z2McD|DBrA+MLv$OmAq;bIm+pSfZlRf%Oz{br<GbzO}h$K5fojy6;W1!mqxTTquIN2
z;?J+U5o@H+gG<~&^jyov#2IM5?){!n>?0ILO$S5uP==6oVbk~lcozQsaY?Vw(YRTY
zYI#$L&-ugKJrxq8DN*caC#HF6JrMjH^{i4Kr-0noHj{*g`k`un{1ef`Sc1h~#?N+x
z_ZwCy;!P8ldKFE4XCTu`Y^}t;SHQQ5PYPJ(K~u7RQ9rW%;a=_hbn~NORL6_d=W}gP
zus8us=uDA!tEMx7!uBm>;=9yBdBf~jJ=GUtvh4OQD^Wx8CSfG%KMOn2wzxLGb9a=O
z%$aSY+g;wM-h25MCzE|s@*LgzQZqheI*mhk%Eo!?*wyb$Yd&QiJ*-lCH1SLGY2Ayo
zhX@{W#^o8*n1;#4S*XhfVianj)OeV?aV736-rvG3%7ru&FcRE+F5hpmtBXv<e#Iap
zkJVbT#s4HJ{-sfue%(R_GLk4-q^s&O4YpJ~-MhTC%p;h_<72f76s9*|Uc?s|BjhYD
zCS1wO9A~HK(X3E8qPDd~9nI3ozc+8~H#_u9->nH158H+J16jnpjo>I-iC^7G&zkGY
zgbds{@+#{DZQmDJo(;TCr_<$9?pV8)mU{_5$#z$L71`fjMZsqF_i0K_Po?S^oRXsp
zqyP|%Vs)?@gybw=hpJ@E$1cxTcR4HHSvxFL<XbUX^E|k(AX=hCt1ZH5R6WFsjwNjt
zOGC?HCppzg>+^uV($han>%p>i(^FfaJl36!iKq*-V<23BD)^|>fCKQ@4W6QuC)4^;
zS!&7F`cf#>SEkX9Ax|E36c_Rdt2{3ja(rusEMJYsh-D$?nDuu3nIX)@pJl&@X%WbL
zALGN*PhfqLxMWok1GMi|@}!p`&UHt{)IVNEo&hC!irDQ7`0^TkUx9CQxmsnWD%X|k
zOUPJ;%3_XT;#voVrjV{Q(rvyvf`KIORE_({P+6*1AT8(nHibH)=~cP!l7VCaEHI6s
zhOAK_1{Z&#wj{g~!0Y1bYosvsH$X`;xm<z>GmqZ)H-v0b05;<ON?;LJ)67x22<+#|
zc`3Og%-PiAb8XHWK3?b1tvHz%VfcwON|F+LB%h%D1o(bN;vB4qwFo_`GggYvUvL0g
z38)9!drDCtPd#GgDt0y&A<lXVUEBIC-X`($ol>(UwyZ%S_Yh%?L1N*>$SgKp_!3u<
zHzQc3$g+KK1M*B*fJaD&y?nTa=7bD=@XP+bMdJ|#GLuL#<gItV?${D<uEOu^{ORW`
zs}iGNw9`)HdGMqkM<iS8Sq}AnSFsqRRUQo%2mZDSCRLJ?V};p4fk^#)IEB8zI;IGh
znA%CoSAy{GXzq1S*)a&c)VJEb22m%xyH@;~VJa2ZZ7adPB=b<MJK$)KmOJnY+B~@y
zB3xpYJ5lu|1c_eE#-skmFilW77eCHRc=YHEQ3;PRH>lG61%$^_TYU~U`t)<q6Ej<V
z&YBSv_fG@Y+~kY{Aq@-UE@58ssDis#`JX+Pq*8xytN!d%5H!nvYmGb{MU_@*#AR2_
zz|%QVNaS!h9Voz0QDG{x$g_&Ng6X=)$a3fo|ILT)oNf_9qOM3)w3(eFf^0MjM18Uj
zR9MG_h=;;GBV1MW_<<$kKqs8ExozSz&DX0{M#L|^cG=mEZb)d<F6iOkaHNoR`f|>G
zLD>nwk`J#H!xhULJcQdy0?wwl9#W;F9iN_et(%Ho*5z4!$;H9%KVubSWZb%oqhO)&
z$4n6Qh=Kb#k>@h`An=TcC(^X9W;0kEw2Co-fq|yH(PrVZ?f-o_mGxt7FDa93`xDOs
zpPfnsTwB)XHIC9gI^qM<-^4qA`sj|YW7QQ5b7^3BRPt~^7(6E@roZz`!4Y+hJLkn^
zbgfq&AI!3Ckvppxl5(<3!v)6lCBa#@MVuppeV(0gr`ktdReO$31M^`>8(ZAZXW{7&
zJ5KX0_1gzBF$!ldob@&}JBIQ<e~xN&LT%)b<BY{8!BF`j^?TzI-r0Ui*3<L|$@DDZ
z1^QgLePMwP)hm<Qm0UGkXOcP-(f~}=#l(F!Mp^9RS$-<;p=9IoQLCJ>%V-WR(9`I9
zK0lk3u%#8ss>zP-LLur5Hp*8Ij)Ku<y?C^%%)tY|VM1kAd`v3WK&xn>xXy@>zf+rN
zVG!YR8|>jc`9W8Is4~p7y*r}sUEw8oAVn>a&5zT7K!zu}Z`7XQ8~!4VMm1m2Mbid_
zV|?JXpQF9P4<!c>KH(TbD-sAkkn7e<ULqB1YhJH5Z_D~K?8}Mqz;S<X8_dUh0{ztH
zhAv>vyKJrcwPK?zFNe?|CSpp0EmHtP?)dE0T*Zd#;eeQa!j{wqf8Gb3gK?E?3b38&
zrKrT(j6o+l7o*1d^-15dL$@NBgm^N|Q@ct=_N4Ui#^YM<Q0?IL*S+DNAcYH9R2%V4
z)78myCu`d-$y~uq;MWr(s_V@sktqzWYq(S5*kx>jW`ZcXH7xCUC&`z9UF1W#NxqEP
zh@hIrDgd3F%9kJ3o?#kX)aatKP^?`N#qf!52qzXwq8es%v?7rdG$!!I1{H$QGZ>R*
z!dh=#`i3AgaO@QI$}xuraPm<UrE2kP(>a#@8pGIiqNC4!>10c}hnKK&GV5`8@y8L*
zutS4Yhe<s-7F?&!V}&z=ucC8eIF?t@LdR>>YWd>G>teKtuC!+;cbkDRHdT7xKA(8x
zD*K#Uk-1i#glJM>eC{;j7YWbpSbsgXymv~e6xhn0TKwv4AWS>(e({~K+#@(rqEN4s
z5Y@-faBfml-+ty%2%U0a*^xdhgh$ue{&?B`p1WF7e*e#TG~v1jFeK9eckz#c2GpuR
zeMcNq<3yvq%;-V>r&`^1hs#1FWP1AC+3ylmPb&2U^}_ump{(lwfBX#NX%V%T1O-7W
z?B5gKh6;v#;&6n77(afZHN%S6NlbT2g4~H`A`0P`zM%`v$d`3-1W=`Dek0n!bh)%o
zHs)m^Y)9~-bloh^b3s=ZM+MC=te>Z>!AX`jV%H`=o##SFn)nSdjBl9kAlYqvj+&wR
zggmivx^gKMpUSP$LH`@9<+HE|-br0>pn{lut0w#x1T}oPmR<|vH4JtT3KyRIA|(An
z4vPkl#huWp=$;40t?(Xn;Z=Ozut`1Niic^<(k<WR=x25}7Nde^ZYnXG`wUNg?hxU=
zrLehjeuhs4&^J)}xuH@Q`5sXD6on(|Vtm5X^XNSN_d0|6=<LLQG@iV$4<9=+R7hXJ
zxzH<!Xi_FJYHNaKRZ<Kr6~FPSp4cY2Hwz_SkTjZ6CM_-$t5Dh5NGBRVtu!X8>9NHb
zXzW6GlC2BnK{)$oMJI<i!^OBtA+p<Xx#H7dVtC%2WoqIEB<xA(UVbPr@h9&tGFjyl
z2`Y-3VPhr)=pm?{Oly*s*X);;vJ}ONcS-HGkIy%blbu9{#1bkE!0$QIuMG<yiD&tL
z#2>6QFY}Q3j-88oH$yrRm{Ss?1IH#s0z_QrmMvrvwM!|;2UvoD5O6WQti*LB29W?8
zi1z+sO-(%qm82>7JZB(W>lj9in+4^UXW<{Yhw?{b@o?hCfLURVE4}^$g&Kx3kvwt@
z*>DFg7E!*zG>s0;#|tbOy=4;kerHYe5iL-h(6WiQ8b}Aqtxh@&;bAF?2E@Ox^E-V#
zs*~uT#RS)i3ne#_G&g+4I1laK$--jnwaPI=5P5WZh@`q(@Q@bh+h+uChD)g^OGDD*
zMBrcKB{gZ9{lz(vyJ^FDyLW5pIN=<X#yss=Ta?LMIC2q@L4M3V$q_+`oj6DJ&<h&z
zi=-ql_aJ|mOWTvpHp#7Cs{IJF4o&&1lUsT>#n|lykc|wD9nXoP+PaghvG&#aa~loI
z{SA#13zm*XK8$W~v^lEjuMkx#(`C^vDjokA+g{uurY0dRVSX&5WLTgYA#H!Scnm)}
zO|Frul3vTV3ZaEyc4`gy+r0Fe$ipl`4`|g^i?V`WoT%fc#*s4%Z<}}^1bL<M%22(|
zB%s_fo`l63<C7#Xfr4b()`AkGTm|c59sIQ~pAih^0XXmrUl@kHodcA%LOPs14Xux@
z;`cSot4>GR1R>o8HW!nFq;klHtMLfX=X-jmINCs%QZO#XRc*6QC`C4&rIvnA;pzIb
zq_xF5Lf+ZDDfrXM?mTF|*~|cUW?h3EE;&IYbngejwe=w7dVSVZK?8wjWiC0dcXK>x
ze1Aek`)t1qXx>25%o6GfM<&mq*s)t1j-0+wWaT@b9mX1I6yz%X7%7yB1u=HyCbTdj
zVk85V^^dR<{`>{+G`RCI?#iq9555tLZCTBcH-83I`9$Ybc$C0k-=3!l>rWP3V_O%*
zmpw;c0?Of+cncJl^8*IE;Mtw*{SmVu%)WtQO%;av7U|d}Kg;5w<`v3IKZoyB(`cvN
zAL<NkIE|YpGJ!L2QOsnK{%*7&VqTvEVSg7opbV$WYzZml`oJyio788E>uv6P(jm@+
zXy?jwPc%N%G)fd1l|5Lom|74nlUKZLMT(oz6>T=3<%4iF)?kghpZnEsjn&7~PFzVJ
z8YfV6SVz!)<X=)#%~yxKi=J<}D|f<F)=X98#3<wG@jd#+-Y$g~+{+G@nzQN>3MqDG
zPoF+-I_b#!rM4jS)Eq7ZL*oJ?)JvH5qLfJXQo7N_eT3!|u3~0}E%$H)*?~;zRElt$
z;}bN&oZuj-ZM-QM_M*DoyboI_(BcL#PejgWJ=&$VY}SQaYv#<DRBJLdrYbvT+H^-0
z&@UYFK4qkaxuL9b(C*sBjZAyevTb(~dw{EYO4k-MLXEdyme*TcW$^>dLMC@5cDYh|
zapK0)v?ok8j8TMki3A2iG`x2#{FiX(6y8~8F;5X6GPYn;o_S8&WfHNPFe4#*s4oVe
z-KS_+@{*4vA2iN#WfnD;<XQM<lM=jth$GTB><Zw1L0b{_d?Ri2Xz?1r-T&OCjwnwv
zBQ33!!X=xD90Q!oOq@1MqiD}dtsN7{+qd4QIARxR<%A!GQHl8}s|!lBRt&T%Nno>u
zKkG!z7SAHiMsRn_nHz{^h7z$S>lQZ_!mJRsr{$!b7MSYq+8?hkg5trijIAc>J`m?b
zO1Pk#7h4jVgJMfGi;hJ{liN30O2;*@;wrTlz_LLKu)g-trJti?G_*fz)f06-|5RWK
z)gPJIAr~bXRKRZcz=jAbjGOKoJvsBOPEwgjG#8MBE_{D0j)9$sM^&^c1IBMlA+@dw
z%QdxklmE?CYVr0PEQc*H+l8K-Coy0Q4gkSd{l^|o<|@?Ly7J(1@mQfqGmqMOZZ}(d
z2jS@^fau2CI%I8nL>lk>mJjJmIbku0ZqJr6)a1|tU*9cBV<UwyvM?qF_=&mu?A#Yr
zk2=9$=u{d?pS(?~;Xot=1rCa}esYZ`#phw)v*JskOq-fGWE<VIkH3W$5YPfoQhjtX
z{$yP~2EnFGwSc;)Ix6L@<kj8ri$iXgLv+{7DxGyn05U8UlH1V8NpMMPgJGmRS3iX-
zqJyKd4+>9meJVSd*TTz{o@NU|m#%iHttci1%sQP0VZ_;p;jq}Diz6!q_)u^@PZ#?a
z_;oOS)?b)mRr<)>#cCY&kv7v_m!bLLuJ%z49mj<>ciEsGnGVW1rpxVd*sHSa;~m#a
zXb$RiSM$mamh*7IdM{ahZ?o7D-vW*hB^#=D$O0U5e?l>Dvjkcrc#N_q_RnCw{2hdg
zXHd|ixiV(vy{~RJ0^me>>03k+j^g(3D{8eB?q(?~d5a^O)Eb?6!7#K{P7{&YW(`vV
zC@*1LEl+)1jqIu@zkn6h95*_K;Oqgd&zlXi{ECvPdA|(XpG8rJi>4ve#sTugpgQEV
znO{5eJV9aLG^6Ipkr+1e1=djHG`bP*X(`Q!h0tjSJB|SiZr}@*I-hJw+r7VWk<gR2
zyB6yPt=r5PhMC`4AtSRCz$qqTawK3YCdZ-=sFcyK9ev95fn(C;OSy+L@34%gO6Lw~
zaC0v|UO!JaMyL4J5;STXQGHNW+%&+=)=xaHB2OI)LlxI0OU&2DerC6_QoE|Ec}`c?
zoDT*LLTBmyQIBTJ$|sB!tgF0B*um*Zp}a71j;`o1F0)F?2~Hrque4aq4}o0_rDjU?
zCWBlnbK(EWkeD-JM=Dzu8N5t)L}%xHz8{f6jYX1h(OJZ~4pqJ&=tQZgO;j4Zi!<V_
zL&|7aei+fnQyqmw1wO{$!8%S2K;4n<tlv7$4g;FZPhvwcu}Zfp3a%uGmX)jM*(F&Z
zBlF~C`NK3U9(^L_#!QgE%Kyn<=eGHH>X_D$JeX8?;DCJVj-ob4+0-RUTioioEP0Av
zU^+PCQwR4Y2=9eSEb#k^0Vq0kDquZt*X|)$W6I>$o_ZY2VTLknrsy<<2M0xX6731B
zPz(iWDNhkgT3*0}8CZmqB;Cx6C+lHGHJ%boLZwbh6MsHTzq`_e^fyJGw_J)%c*i2%
zni%}njnFUrjd&b(fVedC<v>4F8u2X#j$&wcf_80w)X#x|IyA}yf<MqF%{UP9$kZEi
zgIT6Svtv2J_62^#um@D*KZi3K$||2%#}9v-EU1rd<8~E%j^VLz*=pod+HX^W#Xcyk
zm{ontcC}~OK>G%N#RYefVb-G7TP2R;6HMi_rQGr{*+6TldbaLS`JrFDvXvKM`{W|?
z+UH*AxU*n#Z#}G4i_3HHk*Sxu6}9oY%}bm4N!xzm1jrU2MD)2cH;S$u@e7p1sLu^w
zt*)e#m!Dusv8EA-wYpp7M_6ZsJ`ZWRI5XDr?#7R+ulm2BDm1_&^++)gWybY6R%bUO
zR?#+0d{!tpDwS`&6Zvf}p}fOj2#Eu_SK#jU;`7lDT5>aZKzIkj2OF&cYse3JyB@Nk
zUit;zcevz5r#dd)5%{pTzs0VnTKhsjy>NBG7A^8=<?Lnv$v^LkbjYaduRZ;aH`9Yc
z=Qqb(D<Zile=u(AjLsIFj@{hfzIg5YSo|CoN73*d|7TJ|jk@Nv!7o8U{G;a^pB3O?
z&t-OVo)AwS_-shp_L&`I<4bIMqe%bP);vhyDN&&iAJ4FQy3Vn(4wjHjc#@{S?qumM
zwK6MW(;~K=RqXlgMYq2DXoL&q8r04hlPCurMc~K=Hny{wN*YOj!7*##9ML#rBYPOb
zqr%tKSd6SHMlTNYT5;12yOZ@jiZ$&zIc0S_A~s#>TjX*oFBP?A!t&T6^~@d2C(qo(
z9G5?-(M6b&Gt$TG*0r3z09&w!*MvAEK?mNRLvL!vRn??^lZHYZRRub%?|-r@P&uj@
z&>!%%R@cq>bW&(X^F+@H!F0%OI3e<eUgr<fXi=FKWzicCt3sAFHw4ddkT*Yi;L@2~
zL)Rfn$BJvaf?#_V=jw6dSR_Tkf!N~VRTymt^6~~ZR75-NcoxEqD?wa_t^o|I%rXm9
z7?@s1^goxD3lLBoP6@+S`K<G8xr_;io?d(y2g*DS52CF7T|=L~rdYkn*V<nk)7c%H
z$5R~PYy{8Ee_SP$oS^!Mds>jziEN9}iX`s)mlvfQndw(gq#sepU#JqL%oH+W+umgk
z5P?3$^QnJJ;6x@J<P2zLHbPeOUg3TsuY&p7cMi1>FLzW$Py~AuOO58H;nR|d!4pW)
zGz;qjYoHZdFy7^^P3zQ!uRdMCPMraMR`y34UHxO5fucOlD%smx{-e74&b!^p1@>W*
z<dbP@iFAhoWSVODEz=n?4z<Ds^%H9v@<US#&;t#+&wX<{CKqmfCt5!m0&^VI(XzM5
zgC$=j22km(LROdkS0TlgI913u_L;pl*SOkW>f$Nk7a4O}Rz}R<x~k#~Bvo4EQ{ZeI
zOrru1>x)j$?XAl{*`5#{87h49lQ2NRVmH*%cqQa{?$m0O(!MAn>p^NW_G{u>OdoNP
zTA%?LaSW-AyVWj=!g<PqCRjblBPBR--mlp)`k(pnbV=N03y00?wB{-L*O{}(_XjzN
zSvqJvJ2fBt5-I9=!iAh>S4a7OYfp^8+4J@K+!ta`*v*8n#eGNZ!(UnW&aW_{L;6sZ
zd$q@ZlAa7+pF4d=HS0^sVd{p-7;VHUWZKt>>K*%a^$<G%Q5~I1``KU79WTQ|yX*F)
zQf4ze@MMUB=(RI!0|>yTKjYsdjnvG5O$3}Qh`HSWi1t|#8<FVLjkM|AT+_wSz><E-
z#&AYZpo4;yN$k3YL&Qt4cIx;CBno8*uXHVSKna^r*u-hM(>G9(CiZEcMNVT;U~}30
zHOQ2+{24tvJGJ@$v|f`)VZ!q*Vd!7HAx`_8h-Yx$Rx%a2U-4c$=Z<s!yqg$wb>7sG
zH>1oAlbe1!+r0Lep1z=zsJF~H4Zw-lBm*|`chZ8=HYA;0W`(998AUYMz;MbvA*Hdj
z#!0_*E*9iXnjqfN#q0UaS!!pO{&u_>Sz?b9iYwp|ueWZgfCa>KhEbZh#q!lOZJpWP
zh-+XELG@f-n*@JudJC_g&<TU`jE*B=A!IospOJTyR5}ZMq6K~T)F7Toy5JAx6|We^
zkgUnmW2gdNEBY6t;289Ais9w|pwA_PQx{AmVBno?@J19D%7r-z#-sw?ton$=yrOG=
z+vq0B)s<shHk>`jOiZT9hvq*h{u!bmXNik`#Le2w#oRnm9T`8<5FAB!8%<jV-Okgd
zh&%tuqw5C-&4q58MBI$@FJ$V7$H2p;6Nnw8&eHhXfMB*=XceWFl-St;FIbuQ@)>7)
zqh_`CcnEtoh)N9T!&~^&BNr$Sb%UzsrksveOg@I$ITa6v7{y$KVcq`xL*`=+)>rV&
zAJrn9_rfm~742|5E+}Pc>-C&74M+>eCSumy&{Rmf{z+8M;31Z~^F>qCAL&i96Y1^7
zA!Zs<xc{J=?(@PX9SIF)AgCpZCDrc_)vgslBxaUyxeez=CP7-diZqGJgWbb5=F_n(
zjB;mV|3K^`v=lu{3tK!7`qZ3Fuo6L>I|jU7+1{&L>eE3w_n2W0y}5eW$KapZ1${TE
zxHLQr1cQ)vnrz*hlP`ZVQSa)i3K=UPsAkR7upS#nHbuEXxO*(k1hv>%)&pG*w~3Zx
zpq9dnK4W>36*&fBiEu(;vVCG$B&<cU;^h*Gf`(a5l#d}e$1#G}wJ{@MXy&sJ%~MwV
zVZ=Qs*k6k7cpXU>=qj~9!ceUP2OilQdZS^hosFm-K@@v#=n))&K(jU1n8wPLGxC#D
za(tEVES+we1n#G1G~gS+ghi~9&Len_DPyL6&zBSl!Fpk(z^(AJ&%Mjm6E~(Bib_AJ
zy0cB081b#|y(U{B=BoEjR4VJy%A@%`x#1vh5DfAeH|0xg($SfMPokQu@KgGuq*dF_
zH7p#(S517y>@%D1;@(ngDaufhW7W;+J_Rcflq?kM!Ceyh6$)EPt9tZ@P5UJ*cWp`F
zxy4g_qK;Lawe%}qb_JegFfxNclM@WZXcY7cYv>80N|BgmZi1%GR@PFLxz^gJ!Oj^n
z(+K=)oL%G8Rqxj{m#ByxaOD*CuLrEcN8H_HXIYe@zsgUoOBMNG)#=GFfIjXvzd#W-
z&8SPI79Q&>r8q&Xe6mCa{Tafp1yfl)g8F<N7!;eU+*r*Z@Fuq}#JJwc@P$a(<DM{u
zFi<o3r_sH-g!~H19gIiH1y!%aA(5KrZBA4?QgKL#3E-B%M7dO<wBLi+_O`Ljnn~jD
zT)zge2F9<yHrCFXRD}?W2El!WO}akRo=yj!41Rj3lO!?-?<gvNWHjDs{(~AhCAxzM
zI+@Ss^Zu0q#fndiX|)n`N~&7rTw((`PRonC5M145RHr)MN<C9O#}9C3Oc&u}{4ADh
zD&le9-EJ-tzrS->G|WqDy!vw6Mv!Y6qYyWPA^SahWF@`$<yZWn)h*5o(Bt7_vz>);
zv7<<Q|1aT-Rh|b#Uj%#sbeO(9H*%s~V<nzQF`3w&D<U6@IkWxd2BX3(?h^R~&$)Sz
zikJfMHVWKf(b&i0RTbE}7+q5MPKsU7Yj&9TcMXXVWv;Et@mn@%mrp|SooA;xdPN5C
z0#vUlWEzrDE%OIE+7=xi2Tav>3G=zd=D+H6U*?A#O%^|!Wy5(EK1P_0=D6|j9=XI!
zndKe^@mBE8eL0or4zXHFwMqxxz#UHEzQ@`HkN~1>d?bcH)iClLrwpj2HDy+*4fc{r
z&q)ydEEJo9Xpk5swF&NQ+n7CU!)bcl)E;*i<5wy(m$oXRfZLN4jT7{_<e+*QLh|`l
z)eFWy?oj4y%~%B2$lsk#)N?tSXRAEPUKrQzL^VO^E3oQS9=lK`A<&{hOoikgLyo8n
z5mH!gdb+|%M54+~tv+DE0Mib1TNtygy9*+TE^T^l(-H*Y&&a9<*KHz_`tvpA>LnKo
ziJdnVEk{G<!jB>E)Vfx<21kmq4m#;Y;4F*;uP%zigjk-gDQ4V=<NijsyXYhr$(83v
z2`krKipn3@xAvSmO?8bUdSsT{+nFKGM+7;xf}xY%)Ko)p%;pH=nuhj3hI>c#WN}F5
zMv3f|3E+||LaDev3SZX4QXM1XB&^-xITUE;+5bwGhFhQKEL^$%nEqBos6V%XfjCK=
z+a%}0=jCLRr|^E!xD-xpQdr88Efe{Jm2AMsO>&hsn>axEhA^Vo(&OvMlsw{fD*3Q$
z{1%Y)2;H38J#a={iLjJYmNqe6c_t~@w-2+hD~o2_EwI8#wR!2je@d3WH&*8*I|2p@
zAH?4>N_b@En8T2lQBtzVyWJUsXC>@|Xf~?KZ(T`#Hce(2FD&fJ25jY7^@_#k!5EsE
z&u-n~jN(H`A_`AMin=5^AR4KZ^8flrJwT$UhKzc33KBq?4>|EMBO@plmNF*hx{W7N
z#4zmC<P}Czm~!7L&(*R{gqZS_<nG|rpFzp1q;nz3T+~^w0|P!6iZT|}h|P74CGY~l
z2SGeHmm7YFP^9hfCm@IZ6lcPDCKo&qM-%i#d9y9DOJvB}yro({7R*{tf7$k&0qG*)
zQZVjd1jbS7DBIDvFh(LRH86tTe5T~lt;6mP%AxlGt<^d0yxpm%`loEZ3VkNpZC%O1
zU~)RAQLR{vKagc3w<{*zJm{guoV7r|dPY9fLq^)@(2<i`rYj!@H3bbhpMvjt#VB4i
z_yEaJx&fyK<Kx;~UXEJeo=-$c9~LKW(TXS!<6aVE_pu+I0V(bFDJX6+1)z8nWBI27
zgm8k~4wOVxPX>xRYtiH^GcLYu&*pdj@O6SwF%*N>8zS8krHJYeIhRd_=Fgt^IX{Zx
zCISzBtY&3sip$>esov>0ETLfo>Whh2jrrjqM|FpvBF-r)GRerTbP^>I-h|fN)JO#%
zdv&y@Ql&GQ>^BQU)$jFG#g#fFWDa@vq&~@YP8I_SKN0Rk;8gRWFz3Sk8^xIOo{Kw$
z&?ogp?6bAE^P_FcMP{z^O95%Qhi#3S@9Wo8p`IBNLTPsfHVqDTQMZA$n~Nr0zZ~Ut
z>4dPE+m3z(L1nQSIwgX5+H9z^Z#C1BNQQk&a)j5Rn$$D}*o;{Gi|Gf+yf`3=5Q_ct
z70MxBQsQ#5<KEwXd%sL7`aL`xPhaN565jQ_zk__GVMgFYAYP_xz&y!j%d&f3-lp$i
zdpvRI6}Y1ARUZ#)V4oKC-#1{J&aH|Z-(T-}W(1!`7X|)k66?p|FEgT?zCWA4HV|Ar
zQhmt@9=i~H-%-?mmzcqQnf|i%ax;A?bsO`=^1i>3>*Ica{^kAD?|mTWqUkke`ZQ|@
zxN!1<55XJuFrMT0r0C~ERLJ@A`tX<|aDn#a_l=<M%f{u~pL@Ueo2_i6Y1plc$6wP_
z%?8g8iR~eAo}d2A>%V4FKK~Nh(z`kAN}Rsee!B`n{8dxo_BFHT+*6up!g=ix>^`n$
zDQq3R{rkIfO^Xj+k<pN|9P&fWVM4ocH$O)6G2zWtcMsuK=Hc2}-uR1m%?+xc_rpy8
zYM4yHy?qBhc2#Jzpul*+na$=m=iYl3BUPligO@|WxAFJ4SM#Q=*BZaK6v4N*o7TY|
z+AaN^d%>3>)_d9qZ^oUiju}Pooj+eXfo0{>e(yY&srnrghf54M{C9EHEasO>Bi@3G
z2EWc~=z`JgaJ6E@d<b_GyxqD4KYgTcPY~sgsd3NId(jY>F?AF?)8)Kj6x`~0Er}EK
ze%S_hKinyb!-`$A_k>XP@^W8Zh8Oe&WBw6#^!END_;x#dd3oP5(A}}Yyy8|Z=v^X5
zecbBzu9$K=InXuifx2g&AUIk)tBT^{J?e~Q!Ng{jA3VaaGx+6-bhfi0^K0AC^h4Z3
zl>W=@`{@k@^K#}`hwmOwh400%>QiBA%j=j7e$d4y{wu%dq4_u0nftissXsSaXQ-Sg
z{VeA6Z>ky3Tkg`q8(dIbZcjCxeFDEw9)EK@oTu8?94fGyKUyz|q|ts4KV*LUX-zzN
z(?-5`MXNYGyiP2*InYgb6J>3fo^#39!x}P2@RH>BuJ0^#!h&FGXW|TUGBvdQTeCN^
zLSP3l0semT@gXp(dN`N@7=ec7rU;Cp_I4msJCHMg<70p_0;962v%QOxv8gkF?O%et
zsfnebh`l>Nhv|dB%F4vR&JAGx_~K+^)kXL~ar$5fFf;!%T0r2x^8RxFM;_oqh>g9I
zii4rCDS-9wxWB?dAX6th0L#Bkad(iU3dj&-3it;HfWRp6p%;LU55OqF$_)6J3H6sr
zf|Uut_1`KhfcrmHmJfaY>ZT0%5YGs7vNu*S1p#zEgosH17*$Q(K>%HZ|JCfTgnu>r
zU)o<a{{OVUn*K`z02m$2%^aM}jQ~3TqdB9PsjH>2sj{Sq?q9k8F*Eajk#yLZSsAzi
z|8FG=7drzx;QuT+xw#qG0Bj$0*8i0MQaCw390A~D;rx)r$;R^0%gMzK;P_`W$44I*
zfQyOa1MA;$tengMb`~y%4~u?`U;%La!@<qQ@}K^HG5_WM4?g=}ZuSqp|DgXX@1K&H
z>7(<1SpOk$`~&{4;{P7{|6Tt-DSv_4K2W&+5y8gJ`GLpE%=TZief0i2`XAbdvTXlp
zbN!E6|1t321N*0+jfn-o#>CFRru#oG`L82&{=xcR`hS0-IvmW*49x#a|8J=1u(Pp!
z96TEv(_hnn)IZ$G%JP@O$@!r%8#mWKCDVsF*|}N%YR=C5;S4sGe<-Z%A9U6a<8uPI
zKlJ%<3JcQ*59eQVGP8eRv9bSk0UHPFUmniCSRd~B2aEZyynh_T$;!dN@xPk=jnV%E
z6{DP~ojJ&Y8NkfO%7pNr)`v$qSULaO`Z(_YRsozG%&dTaJ|CfFY4Z_#AK_&45pZIr
z#`Y$rf5YQH{}=A7F(7Ewg^iyASu#QX{-QFhASpLDX}AHVUNlxF)<pkC5<;002n?AN
zP*CA_nPEV#R2j&-p!>f2zU%#6d#lB)YAdtrzTtl5IrES^voe$dLMGM$Hk^kb6ax(h
zY79*E_O`JQTqw5`fJP<KKWd2}-X{9yl;DsqiY7%xKymX03>yk6xc@;_m<mY&6$R45
zs{=Av9K0WcrT;sN5J2c7nf4nlTI@IQ(m?OrSTKHJe+~i|;qMORz5K7}KyzFyvkX6a
z@O1+s|JuezviD1n|94m@aOfZdFn+W#=-q%7bR;L()ZRWI+}QgU@>Ww_AeAB>F|oeB
zzVJ6#Wwcm_<ij1<2N5L(Fk?83z^ec*2w!BR8CVaYH#rKaet0IA0Fw6y`E3O%eKZJ&
zUas{#1309J(W_8y5j3!gUtpN@y8Z!_P<ODE&#=9y52rR@{ek`O4EGv$H6p;}OB1wD
zNcUGzeJ^=T=Kh!qgb+-?C2X<Bya#YV>K-Ftgb;psKge4M==Kc~=BmRPO-OeX4y+&P
zVaZ*J3F0I|1c4;)qDKPjNe#w%%!)}8yEP0JDpchA%ZVa(q`yD&@CD=Tq1lFkgT7|3
zekdl^ucKRQI7xd&V%AR!<M<z2qTxdAkvy&)A_;>J1}J~eS2h9*#|7Jkbwz(t+g#Yf
zyym5P8IMT=Q1xP7z&rFKfx`##2s8PSflGZuh4L417Y&4XW4%(064U&&4F_EShJt2J
z6}^9X@`-csI4-8m8)^a70C~_s@e|m~pXd8A98pF|u8ECz`-jVCY!(&uuA1_Jd-Gkt
z;l!jJUw<%gbW4B#a4s|eQX&?tlo5{bJDo9H;M*e1)#X?ER7`?|-zB;e+^-d;IX$0x
zUsHJ8!QSbLQHN)6AhzE!4mgAy0iSNr-vpgs-NxSIc6-WS?CIVwLl?R>HvGcc{D$9i
zm_{H^F5Yp$2DI(5B)~l#qEQUqk(>Nql&XjZ(NE5QZz<7Z!4*J}r?Q4lLi8I23e^ev
zQYjOjz&f%8=l3%Baor5t-^p5E06z_)#Rk85er3`U+V&e_f`IHWBtjz#P4yE2kV)8<
zZY@2GhZ@}8P75p;7_{y&?nH#kf(gMvjpRT%|M?_KfCh+GDkUAf<ZJ>L31s3zNPy^_
zAb@fRB$e{jZD=GDg1a66vxb3_3<j6?HjoGb_xKV<jf{5vYWr3ITgwR%V0q2y!i4-b
z{`QI?1n@;;8ZE)&de95DmKQsfPMWjT4~Svbdg;2wltmAy>xUI96=;I*STsq<Jmbz{
z`J>p?e}ejz_cp0Z!70V82R7IGa^W$G5Pn^0%c*<s8<(?BFn<z;!^9wm=w`5*L45fH
z$qmA6nCS?C=&W?*_R+d|zci71OYxy`BqOeTM0=sDNrkVMxteq{17Skr<5W8HN!38^
zap(i;Q-_B&5&X_+9f#ulE$a%%WKB`}yALh3I!q`GE|phWhmlB=Pd~>UH&FrA;sr8G
zStPo8I%6lGW5ny=hCP9%S2#%fcZIS3Ba7mRbj({G;?q?HJ#CGKy&b6!<E%n_wZwJ7
zMBqc7$<S<J`K2@mYfVmXvc*gAFb~!`g2)REo`hV%p5hl=WO16P6S$WAfI~G9fxd>I
z+8N1EgWgV?vswhf)1N<KJr}j0WKAb&!6P@0x*etWPb2JC+X#mN5k#n{e!g<c!31@q
zj&`ki4@N_`bIlBn3^?E_a`}04xZe%JmK2~=idq(Gkx)=)Ejx>7ksAyTAB?<XM@sE4
zQ~w-o4~IW~TORP=nXXt93?LZ|@L^BMPf~+I{^l>_OlTT%@$3V9r3*25RdBk5J_F*^
zF)-oP|I82T^^Rr!4eo#^*(ZMiTFA6wS+;;o|A{t*(DNzayA89W%4aqnvZ1Hg9|mp|
zvqeOcmWo)Bc}3$~5MhV852iOPvp4B(3;HAQ1SFm}@M6)Q`8OV9?uu^~pRx;8RLQ?l
zGKg-XukO>pQcCEGQgD1TS#Mcj_o|Wd$RMI7HyzFX@i3I9Vnly>_$pffqH)yB&#9s6
z>h7v)U2Ks2^CRcw!ZPSh9$H+1zoLa3kGruYMZzXT+kQ$d*Z`Tx5!!mgNx25AUC=a@
zn|g9jp*E$dtfY1b4X#bl+fDjP={RNiE60_9dw2#-Fb0;2Gk-&Z{2t_(aqLx?0ztAZ
zCiI|{c3kUGn1chv5$k?2Jz^U-xp|`9#9hkGdnN+s`XV(gB<bE3s*F<sKJ-_UBmJn&
z8vZUq(~qf(oVNABPiTz?qLW;ps9!fOW?M{0KZ`EWXVRWm_VQ*Q&_#z?Q|`TOte2G*
zDxVZJ07leN^6bc4Sms9Z66L537-IALXT^`F!{tjp7md0q5Dqk=ARCKE)9X}3&ar2<
zUPsy-H$-{WkZEV_lS(mb!7Drm8iM8uZNX*j>rX0e;$<vkt#zy()dI2FcBsi$*n2*c
zJKxS#Sd4OA1l4|Uiwk_Ws|lMqQc*W<csE%hhpr^o7WF(jKAl`kJ_`}dUM8a|+{7z&
zPbiBST95}_jYKbbHjxW%?&eJ{Ta^jXYp-{#oljx`E^K=zX`8#E1FV}+iEdVfP0oK#
ztT0s(q+bMf^{E7Q51Awyd1PP|u{2MmHY2<_B#6NX`)%R!1^Ppyt2c8G<8LEj3!rnG
z-X>gAh&%siaU1qBI&<Urm~-lAm$pg9eBngv8`<G?(B``vU)tTeC^0n~f(xq^=}kS3
zonh0rDjUX^8d>@hy%v4WXr*0YIbtMzo|c5cFte2L0}@2(G~17hv7*S)P*Ik%>nZ1n
zS4gqa4QsqUsL-z`g=!wp9C%59RBNb2mQmu(e{*NICqtP4O+JgLP$HrKv<P)h#s*rU
zAG*vrOjUUIZaGur81=9DIZ^df!?$by_rzIcmOU*CZ(Tu@%DUj(Yh-v=_g1<~HD-~#
zDTXDN<?0i8uvd|pU`i8E%2Cw5by&SeUkGg{6wwrvNa@viB&&f)O<q#ZM3+a9&=NgD
zC_0&tpq}CjC81emT5BYmzBL!|#g7vJJ>$S4fs{fCmmeaz#Yi4Mv&Ik7Wgc}j#9#mE
zIwqQ(&L!({-%;Os!NW*b9@9!cdWhpr>q$FEX}C7n=?kYnjDECm+*2O+4xdqtMWb@L
z$RbV1+7-~ki~s(^iBA981sJ?xMn{(Tax&u;(x5@i+(j_KKL2D;8<%rxO2Ppe_B&7g
zez)CC(vrG<XuO7h{qPnjUH{o0yuFsbZuN9KIRLrbxXf9YyCcI~*jYG)M_Z%jn(s!_
zqn%9z9+t4GT{CfWavQhK2i60Y@wA+?Z4<Rgdo$D=pC7$O2f%LbV)*%vC5n_&j&Z=l
zD%<$^y8&aNQ-RsM3non`@Z8x-FsZ8PNu|N*Vf4^x<y>BN0A2r{Q@Ucsaa}yrMMZGu
zi8oKco0<S{W>-YxRddvRUDTvHdF0<YYug2>bPP-?r7^uF5Fd;-8gh5DRvs^s8SJXZ
z4-SD~0QaEGlZRAjZ5$vUG>;?52qAI)$?fG&tG_SW-=kfaY|wCVI(PP{waWc`=$nv?
zK(mVRHDkofn;_77kWiZ1-Agh(YQL8@u2m@@@&QxbeYx2grQ^@%#ZbA^Zp=f3aHn4^
zeAFK5ru_thk)1=5Us4`$AoU-3vQrGCp1!wQ!FwujD-S<*3E?R`f~-*5D~9_v<fU`B
zoB44^(oa802SfDbD;rZd4|hlTph4Z}%Nh0<P`o>>`6*3eh?#rWEwnoK_ljthFYn)U
z0rewS1m4e%PJT}mgrRJ8?cnvQ%<duJ^<_VJ6|ra!^#Y$51Ip1)oKYa|0)*{7F;gm&
z+R_C(_iSZsv1;ZG^kP)p%A9e(A=eRK*h55hH6mD|PDZ`7?sw{`$O>rP%~5?m&HEHw
zJJs&rV>vokGrQ7K5P3YCR3x?0IMR(K36oV1#$8tDAJ4k+`@%bJn!#}y^OEX~|3zuq
z_iO$hS#Oi?dgH^Y?a@37sXSz@_1sfj!dhE{ZIU3otQ1Kcc7}a~)bmbo7^x6s(=m&d
z8`ZJg02f^MNtT}%+C<U@j)i;TCX<s>Amw!MFq7kNCN?64v)FaFcO%{-kG(JwD<hRv
z&Cu!CnlNVOIG`CjSP`j)D|-`>iZI`Vsl5d+Wravd`>S{%Py6H)d`6Y8A9fVgM$M^&
z1o)RmF$;NOO>$o)OErs)sSYqx`&wt6F|fnP8imuCKWkwvti>K;;iAc~Mc#505T-gK
znkFef;N9f$_M-4s9ZG}<$Xxob1qJ9#ZaT3f1Xe^wJP7Os5ewQJvj$sMU7#(a?Nxs(
z+>)iVS^l}j$5otOQ*6ZWZrJMPd~aMDT_Z+x{66VMi=_`ULcYZ3Ny%wNPlX^y`wF-&
za%4Ux%<QgY*+9vET1~4YRmC=-*n(z27BI<aEWpm~5+hQ&#)d~nvcMh6(m<TEEr#00
zHp_t=y3$Jhax#KuH`kK1HR(3Q{KcOuk9$|A4+qj7Gn2m0Uf}plb3rj_ViVmyUWOyj
z3PtOETkw88DKD#Xi@OEbz!Z8rOo&nl4C{WoEfir^RIiUVT7iSKJ-|39OHo+ILryiE
zrjDw-U`jANfXeJ2dT;g(JjpQ1UpqY&V}9X(ZR(u#YIAiR1XiS!3^T+PV=b9Zb)xzX
zT1tAvkdMIa+*lUB6uuM-^y3{UFJRHut(RM}&qz*N%krI2`2ByJy;G1V0lQ^cb;`DF
z+qQYi)+yVzZQHhO+qP|U>hIgPrzc`2W?~-lAsLx38Iifx{&w8<^j@J|;Orj$q_x+A
z1d?ta?ldkc1=T~{skDo5+1vhdrz$dyVv?o#LEJEM`^0I5cgY?Fs<gS*+dl;<^+}Q1
zH04zz-A2XUv|UZCs<d0QCV!VW2v_X_5axm-HUGgQgv$ED4&gWmUh_t%8r%=P7OT~5
zTTS${Oo2cOjbX9fj`S$LxpZ>zDEnPrOUp@FW)4f53*`;wpqf0h27eID6qyD?td4s*
zsKvBvEJZDw?p<ms=)**v@pj7cYLe<<EH;{G2bqoP=o@1rNo_Cc<hMDkRodrNc67Wy
zFQ4(Jt(D1%O;_T?i<KJ3@aVD96Sd4erUW9YH&wV%iE5>YI7IU#Zez#KcRapQ+jccg
zMHG}9M$XNnHo>lB&(wxXDcp_g-T~C-oK{^B6!4Y1R0d$3Zhda=cQC|t-7*7bhnhjv
zrRQ(<fmvk~HCvD&VY7vu%o_+?LCco-?rCrn<i_oj2kh9z#y1xNoL|Zxvvd~pwfyyt
zOme{veNp;E@t2X#6usg_b1d8?ti<!CyQX-}r*jfUB1O}Jhi;6NL>a6}T`S1p39n6G
zl(*?R{T}nx#8l@^=5dxq`6k^*<yb5!c|C!Lx12sjEd^!CYp<4fLRVyIzy0ah`B_@C
zCwR81<z`BW!&!`>TX4)mPvQw{*V?Npot6d46dgpgzFa#SUh_<`l+deRq-e>II{V_E
zrA8y`#7Y$51I)3y`I#!{{>!eu6*n}h{}vtS8kPsA_6d~*>Vp(irwTAc_=z5X+~}iV
z!{>d(v5l%B#x;u}XdDmHy5&1eKe6u>o1WB#;!5L}B+TsE4t(FMw+H0S#Okj%=<sU>
z46m6DGadb!E|_koN~@x0(`IzoK@-upOy9y&3TLGokeuHF5pk$^%w=@D=(N-&pVDM^
z`&`)JDv5*=LFMf2K^xWCrnQJx<ko!T0+mH&_v7p6UI+tIuL2s9=T2%C@UbNTEihTa
zXJEpM;7qTAmXl^V*END-KvAf17n+%rDbEN^z2UP!xC5?9=tp?Kvf5_NmCyt*`aQ*>
zsZ_Z6nO~5&JtOx=^l`98a5>C_<SaM3sI^n5Z9i4TlJD}FOvtjZm1>9W9%*hSNGAoS
z+|=P@1Nb}Q=WoPbK7u8QhJfeQ$YM?@y7D7u-i!C-0CCxN655|6mZ1V2VFB-0rnY&g
zx)85&tng&icuV#|1Cn`!o_8}KA?P^>H>Pi)6}iwOnD)lBBcg)XKDn8mb+WulXsb5T
z;t<qll-*>!kxan*mY@9FdLIYWXUd%3_$_(KNrUyp4;UL`%UOy(?2QFE$*SsAwCqge
zSm(toXk?ND(P+PcZk*fPUu8xk0w5)mLEjzQO_>M8+7{svhcM%ipEa)I(W9E8^O@^2
z@@Pk5$A(z1NPZQQ3U*d1evt99@f}<$KpSd8NAIB5`GH0sJ0vJix8Kf78lAtcS0Y^x
zI$<I9IiV9Opa-H!lHojXZm%)cEizS&xpX>T5&90gI#l$nGN`Ux41kqq3YOkVXlJ1-
z{pX{lp>vf>FC)Zdw+`yXO#+IgqjM&Zd@{HKM!NZO?sYrO^%&+4BoP$Z?8NC=b-Rhc
zuw4*Xlaw%WN8ii~!0>i_wTm11jPT-s#_&1|2~2~nY1{R0`#AQVBBXeP3AozxG8F@5
z&XT~|i|byJ3^i`CPsa(*s}T?t|Ck=1>8~FpcyHhj1@tfBI;Wk3yy2WTU`fyNMiLXI
z<_H7+*-KU#lg@P99zodbI=*xGnc09WQ7P-$Ee-X?j-D-@oLY0p)D+RaxMi=Npk(M$
zfH-&3u~8GIA0(Ee)I1taIbsMY-bqmO$RR@3bxU?Q>Yx8BvscA(YxI0kBuCiMPt)bZ
z_%W+nmBzGNjpsT7Z^hTv_!Nf1NA!H%-}*Pg&%Fx6@bx_I(2h1^jPQvmz=)w7e9|UQ
znm)cMxcY@Jk!R>!L2=gNLb9u~!-Nn)kw;|(UAvyOHD#REep;gfF?3U@B2H#@$On2M
zEo%J0MlZH~D6;cd`Qjon7)A~XckC8xJ$RaGEw%?v8bM(CsN_bhN9OI?j`uNgcQUhs
z2aKN21ZcjPjI4+W+}wqs^4yU#yf0xkU33ep6>$re?fHs0$(KE<t#M~9W0hGUvTb%&
zeR?$8mD!k0sG{H$AX+8D_(sQOEpzJ5tCJC=&G_KD5t6Qc8L*pga6(+rT8Xwl-yx^f
zE@x9+g%!7JVmmyR6a6M*8X=r|zD$R!qS8l%X6#~=z28aW#X;;-+3&WUv|_V$G54*$
zAH^E*Ucz8L^|E|EuCq{RONAH_q4+o_?Ww%Nd|jPrE6*i09Tom$u@<Zlcfr=Gv89}D
zN~gr!ZFSjTz_PwC;x3!<NI#+#y%*hRDJw8df1`jz4kGlHwLxmWO2p@2OypayCzIJf
zNOcd>I&TEPNO*>>k60)j-{DjNeUzv*TGh>Cah1l6u`r%?{f(jJTip{@h%1*g-#6fu
zrb$bG5^GdpNYXUdS}<ko@Y6}<z*k+m&d~avMvot%P6iKm?`hD;<Eg)0U$_?auMkL%
zm56;=9Jvb6IOZuSi4tnz!?<^=zsMK;+-ZrxNj3WqzxB4DjGU<8j_$SLzRo05TrZH5
zgiCo2AB!5rS~=@?Y$X|@pq-2XWZv2+zy?o6kN0!agySKFHK`le*|yhW1w<WQRT`_J
z<Fky3oW2CrjXDtFbG9{YLD~LSqLm+I2N*-8O^|)aA5TnCToDB$rMkJK;&P6%B-{;F
zloe%jn{ltvJ3&OlaSy4p%Sf~s?Hks0JMTS?`6Na~tt{G;zgzjA!nM3?zh5pcj<GY!
z#d+)dx+AL0t+svH$36jLs)TJj<P6x2{Yg(xs2ra5Z+rVZ1qC9Xwj<Qn>0&_En9X7H
zUzU`#nh+4xVG2%gQJ!O;%}U9ta5TO;LaaPm>1*QV?itTl?tTJNLdur`WIH>wlTP+u
zN|nA&lXihV<V-nb6|IDO-3)hGCQaKnm(?FCGOL6&pB-n5m>^v>gLK(lR_f3RR-fco
z_n@wW#My*ixgQAH%98C0uS6}=Y9!&ID_Q)ob{M*WQ$w<2f7)}V#i5w7I%tC{NqC?Q
z99?3C+nVQ0wI>+Wrl@3{N0;_6b|u&Q74Z;UwRBOm-}GThnADvi&%kGtn<|LnZw$y{
z2G9*NZFi26>A4BCXSJi-t*_GLIu_Q~$Gly5KEV#kID}9mGW2Z1X8DWT^uS8v${B2*
zw+UCdGBeI_m3=}5>ZV5kmEpod{G|6!Za-ifDa-K|WtSjRj7DNybw|Xc(2u!<OUlom
ze_FY^ckMPfmF-6K67Du#y#0om+6~$Yu8y`XiV2t{?6tkfo-pzDlHl6G&WT7s@G->`
z=!zEz<_^5w$4na^%|q|MA_jraz7Vjm97gskA1dpq&fo780M$2f7=ESmpb;E6q)KkV
zZixJf^5vpEJZCN(I4&%9z8r*2Gol_*pAGXbitPj+#g~PO&>11r){XLRKqro^H=1^W
z1!kFUHLA6F{~U|Nsh#ylANjtpBSiJ6(LUqiN}v238;Sz0xekD;Q*N$=hd6s5?U%Zn
zXh08#K6wGzbd_CmwfcQzeOc0=GQuJ=At?`+P7V*Mrn|j3aIV8aKe~hdo<P!>q#FIt
zwMF+fG!_wnHrj~<w&8}4j3_Q)Y(RvKt(k{wNik(|z8`wkb43FGpSy|-CyaUS-KsuP
z*;di3f$IANwX2{*+$VT0NSR7XDeX*;z)9!6-Afj@y7HF}&mqyl-cz|X9bG{f!Y5K7
z)=lK?vXS&yQ)if~NmNHJcv*&4=a@F*HH*qO9C-4k?MtQ_df1XV628`3KywD%KXotp
zizkOj%iU)_28K|CJ6o@Y;&_Q<x6dY4H+nT`el`Yc|H30&J_6GIJn+!%z+GKC(V4^w
z2l$&8>Ne+}A|~umlmjMbT0x*BM*8eZ76y84i?s%iR;uH?ha3%=EyA}glPF9GWD*5J
z?7)wvdp*&)9iDPK2}Pl>ZhCRJm(O6JJg~#DZx0u(+I3b{OblJ_M0!mcRFAbx%~iNJ
z`IPQ`GS&v@oJRIn<{m^#m;{^{t?inY>T5leCrqK%KaD4+9-0k0gWrxvIi=^ez(Mt}
zX5=l&I@&Bn6>R3f%ESabXt$6i@;9(@THP`yOV^Uiu0*pvSo&@@`YCpmO~$TjPMo*?
zqp9+}J^QYhM>#G%ElkD5RNbhFelG<SWP<eqsFH(w^g0r35358y@6_1^!<Mnt{@yNf
zAoSvMqU9Wv2Gk{w>gI{P*}BNDQg+*D4k3<)u=DzixE`Le#d4~b8*UjJ(s=7wAT3fH
zR7flWSPblaLFUn$p`$^4(8g)Z+kyROU9r-p%uCp$#*9(lb<*tu+*B)%PFt48TV66*
zQ{!Ef3{<jm<0>rC<qWjrYV{F;GL{Gbv=u)*0{X31MVNS9n=`<{XO1q`TnC;FCD9<!
zsyN}5PKj2=m?9K9QJ>+MkpjjAfiS;&tJNk!=X^T9iLpaVa>vNt<d`yT3&s)xElu*|
z?D1D>lsl)JB4%SZMK;YObhExSGWp=|Oif*ngX7ftXCao0e4@yY8^~3+?t2=sM7>rg
zzd*2=a7d@`7YL5p+*kNqEO_@PEkWmnmY9*9iW{J$+ul&c)h8n-PmIf>Aytz=MH2`A
z!-L8tPuzo4-3{SFB`|BDqxA}1MK&1YZVAgXEB^Kl4c&zPMl|%JiF5IDx;Yz@7-8{Q
zh6Y8GL>uH|`7t>gNmtvG(!XwbJQkd87qVNnL7C6Zcy-ThGI#5KNvXI(*XdNeDF?&?
zy<u-YboDZ^Ki157m9Y~k$=-v55<Vkzg6BmK4iue3xlm~%?mO8%=*t9j<h?Hg@#6Rc
z1Z5(sg|`Y-+m|!swEL$-{RecC#o2Ohx`H55o!YQN&C}f^rE*iFjgj9<tDH-sa|*u!
zfxUU_&IA{rlW_XRm=2x118c3%-PC$oE}O@%Kd1-xdRRE{CtQgE0sv~0xAC0-LEz9e
z%idg?j(tx`*cVDsBs`zv?7Cvp^8M2~3<bBgENAwkeXshglB9vIkuUtJLmow}e)6a@
z&1Fni9)*)&?BiXl-yQp=BcL)W4bu$(0cYe-&=5M>(>Y5<E=h!dZskxUMN~5!^|D+<
z0qMDdmoh|a*zQqf?Qkr91bn)u>BDD8j&P{0cgcxeb><$-G)&4*`Hykj{FD`C8hJdq
zYWRY(Y&HNX+;5%Nna+Z?)#LUolegBCdjPM#?#f?U-*@y1CJ4|x)*Ur((enPyxuB-;
zyh5oFx$Z**B7Y|>h3gVt<9C+Sr)69Hm=wG$`@7+!=#P8=bn{CQ{=>Zguigj4f82@x
zX2$;|e*8zR5H@o7XK!ZfXk-6hs)ek+)vpqwEGVolEJY<GBO}4_yWLXX<W~dvuLA`C
zhXO+L>srt-ve5qu7i_;$0So=F0V3e=?>`0zE64AVLi)DiMrJ0azpLy_kiUw=f9<1T
z_|HxWM}14Pe*)GfmPYtA48I!`9gVD1e#HVvI<^0K7BdsW??p`Y?f+wNkkjeYInp`P
zxlsJqE&OupUq9r(JeB{=xcH9|^4|~mzo0l13;TbwZ^mCb&-{N2#aY?e8U8;*aa9nd
z1+ShG1bKJ}Korr)=sB4Ac^HL$bRg2&C<eWPBF*+fkT_Y2I0XqaKr`rASP+OXt|zyh
z6W*JylTDBBlgpnEFD~Ak9}b{;Omrx64*bi!b8vPIel>VvAo)3#cmQAie1H5{a1MCW
zKSCXXKjLG4#`1|^(10m#Hts0M$dSWiN`0o8X|#V}XYU<Bz$5+yk{AY(81du#f`b!(
z(Fp-dg62X#`*Q#|f&glOvo8i<m-_P?{G}%|VWyqvi3YF+hvQe%(1`zO;o={I3HKN1
zwe_V%zk+reFbee51{?+lhYlb9CIW8BWWtO}L_oZMentiEa04flmxbj11h7G0$It_Q
z^3Cteh4t0F`t#UpJ-ZVH5f21NI!zzLXYenOXdNrk|Bvk-Rdis;mtGFnew<u#09v+Q
zHAS%8n85xY^ovi1U9e9UOyAnv+n#~1qc19W=<f{-Xg~bTSUq-BP?taK{aSedUQRdy
z%%D?n00TJQ82;Kk@a#UdXz0g~%_A6}wAj${s*J$Cc3)phSzy2*Lis$vJbKkVoXGDe
zplpULtpSkzQ!tQX9<e{=awx(6g0_kykY73#P}qAgFJBJMfqgjGK25v3VaSX@1DiZ~
z>E%A^yTXS)6PdZDzU<(@@rm)E{9AzX(1Yrys<uefi0_&dpHh((J3DA${C}|RX!yVM
zqvbavHdlZyK|pf<X3P8R{8Yc5L}~p3ae#naz-M_f^vutjQ+9^MyBVAoH)Lp_+CO`~
zWY7R#?%!VZefkrCfG$JdR^J|Nf|%o!qAyXe-fW*XN%f3>fNoLX0Re%92L%B1_09hJ
z5?}BB%#efXt?DZKem^e-auEW|`-Bh4oausid^LGp_k{KVy_!(-0K?YF1H9}2w19>M
z;J4!K{jg2?bbQ*Jexn@y0KEDv7hIp6ziUo>NqzA(>anF#J=5{P)?}MvKRY3?gZBO`
zEMmWhRJY56rIu{?n6u33VS^>^%2xm25>5)?1p9HK_s+tf@X5asv3-fqtD<9@``-F~
zz2^b|2MXl=%<idy*!6iabKAOpR{8IXKg=YW+k*G+p>+rkk^tx-Lq?K}RH3^dfxSW=
zn4SU*eTvfai^E6rA7}xzK7Rs6gCoDyc!f&<272a_57JS*@evAf|0Tz+@wUNDdDeZ|
zAr2N~kOzO8ME*mF^<GQcc-1O;mfX@iUi%4bJH#Cz*6RZLH2#nn&zw}k$>j?1Bt+5F
z^YQ*^!g4*((LjCDok0Urh)^{aOy~2SwML2=Rn<xSLod0*q|jKV^z#h6W(AngQQ>`Z
z?QI;R1+IC^)#&?novn}v8pJ%ZZNp6eWHWKR9VAzUUacZ!s+gG+1vQv9tiea=21i{x
zlV+$#BmlM9*6;BTPUz|Tc4RQBGfoP%N|@LFr;2^fKzG6@?lnOoCkPS=A?CN{SyU>R
zD=>s@JId~zvZlyvEA!b)r!&w!g;(Oo{Iozby!HaKWhOj(j=}tE14wo^6q=1F0G^=A
zSo%`oqKmi%ahVvgF0;$6iX`%wZ4A`|&ijjawWzzo5a;jEAk(t4$NC1>BOGgr$_QpK
z2<sv;xro$kJ!WB@Jym`62V>1xbi`AXVhx%%^YN0ZC(TweUp@1K9miMhvFNrP^~9=`
zUj+DmP7(3QHb5m(0CIM82#9;RI!#djmktb~QXxIMz*>i3_wY@3Cj;76yx>u`Jd&)G
z%ATlInK=>1?0WtUVXl6M-q)CGY%yp>oL6arhi!H|N_OrDvD5jZq10gPpsw+k)7+DC
zk0Q#iI{Mie>>|y#^Fr3HgCpgQR8rr9Z?J-aTysFd*~s7Y8L#7%`U*+6R48dv#{N)I
zH;EGS@{g0CIFc6H`enSPykjPf9ahc066aDm`on0thBcydpmkd1Ow3jG-+DqO^K#(l
zt%PAS`_rzU2e!tSd2mI?YC}x~fkDd74pkyyotAE@2$LW!UOy`E$uw!Id(IXaW+H6p
zH#fH-T>h=rR9%WPUBW)Ad<*_<MzJyrD5?GtU4PbYWpE7j0--3Ay~X&W{tVH8Po)a4
zT<ZwlHwv|xODbxMxdlaw+(BZ4&6-^|2eV^A9yO3;ePRk4I(r)0Wi}=SGXH(-grtnC
z-=#`V3zAQcvyuEbAW6TKq=0I<Kr^T^WT;A;oRn)x90x&v<a$u_@>zhdyy;M1@Z6SN
zG6vs@mj`!}n4lFAZy(D#$Yisykuos7cQwTx!oZjG7}!{Tl{w7_l2M1IAA?IF2RIi^
z&zv-;o|V?z<X!2-+v2?&w7NPFWF2$kx<IeX!R|$WO{GigoKnsi$Wp2oZia(cJ=n#0
zmy~5<D$H({N<!9oE4K2PK6NE1A3Se=cnY<nEk}zU&X@&HY0V6zBoAg8Ld|(#Zz&7=
zAXpZE$2w^>THt8ISqBp&Yh+4vjxqS;NKf#z1San?{f4x9wJ(}B9QtHJ3xQ6HbZtZt
z?WP3mi8cv%Y@yn}iK6kIdskM}`uDxj57^Q+-(2GN&BYZ$Fj{uKh`CGp(HiPHS7Xxw
z!xzH3#8NTU7^g(ywC~!B1FsvWw5}z0qBUDbqh+w*OG93)lEHKs%!!xAFMYI?X$0JD
zqcMx%;uzQZj6#F*_ujdjL|k8wE@XNhI~&j&ke0Q6FylYBxJJXTOxN{V(_q8s1r4Cg
z_W2}K)zKC$4C12Uja<h{<%izBe$1zvfjlSyLe{e3B6h`Vmfl%YrOxQ>N^t37?N+m~
ztQp&-#rD?IIi(X928Xw_&DG2H_(}U0%+sm~bZ4D-P7pSj%u>LzlN8aaaz|#+YGFZx
z-6}YHX&bmc(($J~2!v=a{w7eNozn6k?FCxH@H>Jx4i53e%`dEiq4{Tshq#?KOK@?@
zE5y{4ri=<>pXFYZ{JXBNwO6)eUfU+Ak4B1ptK<*d_#O@ylWz`{U^|R9j0I)<&MXs%
zNlzIL*BjdR<tnh;E;Fcp!l$_d%(%}dOcH<Fna(8d$@bl`XD|X`7PKkLGAC4bHg7w$
zabCkm>f!3S<k+@QiEeUYISJ?9e0Q^45YQGMx`^(&&Mm}MB6kJIC^En<#uD!w4nsKU
zexr(BXKH1OzIeTro1wLT-8D9>T3*T(x~AevpVGhGPta`~<Tk2j5)ijk9-}1gqYm$k
zyI5Oj#xq2a{j3g?s%+R?^Me|r6g2*P<=%5S2n`wqg8Y4cZO%)H!ar!5^<%zUKBfP=
z2Zv+*m8u-nSdoAM^UhJS=*W^a`J!IM3pV39N*3P2b@oDN|9O1!Gk41DCml%>oK5T-
zl0KMVy>W3Kj~Qa-X58U43FWXhwW?KYXC;569F~HuI7Cvx0FeYr8*UR)OIMc?XP{rj
zY(L0{iuwYYuz94+%nsnewvc|qIqA5B$Ohthql(Cy2&sa2s~KL8*8EUb=@klQDlVSk
z;cRT?)g@nx)}-mq!nZ3`BPK?=0re4;!792sIVmlxltX#i-XNQNM?51M6ccz|{#oLf
z4#irH>h;ys7=gR_Ip!(ys=!e#hyA)TDnr*J{m!aq{Zep!_+?S=y~pWR-4$s0T?TS8
zzp^8!<&HaJS<YT=#D~bI+!BTx)mPk?^<~}T&}yHMXs8t0LFD_eZ&4y3Q&;~8l_VuC
zU&;J7why}Y4}V~U$oq_2!Sfv9i)vjHX@e6Q<3U)b#q8#8WS%l|Bm`%0`#8l^oIvkD
z@yhHw+I~5%fk1gbL2c=L*`46a$*7u0KZI8P@H7v*BDO;OZQ=2n4hv^rny0KOOE$&$
zRiev~P1SYG6s@6%*`Y}ce;I5|^PSp+gJcg0;`^J_3`@)riqEXj!MY;ap$1M6oV3sd
zijg~aAyS`XdqWgKAs(vIX@DhEIHnL4)WdOn3E)Vuix4RJpt{TBCF?y{d@bq1Z@_Uf
zOo41qq^wrXje>?k6U;<*$YF(9K>?9Oftaous$>GAPG-oCqn`>yxLzigTc)ex7sq;h
z;)admsNbFxvu2mB=Xj`!fJ6g_1^UrWIUkq!H6<&nBj3BEQqnSi^t(f8$tq7byA=J-
z3#CMVXW9rULsLVrn)Rf6dYcZKJYn_@8nIaK?bG@ph<c;Y@uXj!HkDTNW?a$Vpc+a(
z3SfE2bdx=cySKm4ucDcvx*Lvfv8|Z=BAK=+#yW|5(NFToL449t5#kQvlK0SIV_vYD
zWX_>e4${QBxvVuCR*kE4{|oy2k3NpLO%uOCLKNDPcIb;1;!FO$;qR-!EBt6XW9YR<
zW^>w(<F|3^>Kvgs$S3^U*;=t!Pol!i`b6eqO=3<6tBw|`8hj{(rHth{bERjO2A|TG
zaSVWHT}N$!waNw5Wgq;u0#S39M%t7GTY16Hy5j|xn@za0SkW;54-Q;UFC$x3$riue
zaL{>S+?ENfqz<!vBwR|xdcZ{}$Yhv(i&1jv>(N;v4J#$>CggZBQ**8*xgE0+iL$c3
z{Lk~*Mm%T!h9>0a#5NPUjAk!E<2F(GZ>M$#hIBdtW=(D0Dm><r=q0rW1bnT*__A$h
z0wZqtjFoDyM?KwRn@q#>iiaF@avq%Vhua&e^-4OaO&iE5mF=Xet+^^~)9qdLr3Ki8
zf3H_Dud%;oL<_y!&Qu$E@=IpoEm8^xco*lEm>)3n*zc&_2?_vEH4V+1__?R}XZ)Sh
za8$z-;ph3)Wso9RfjsN=Hcfxp4~{kjx7GHcgx=uh%X^0CxzDQc+^t)nd`%wj8Db5_
zFEtfLq<35k?&2kBn|hCf(|vEKOCPi?Y0I&+fuvQMpQCV7yHgJj!YRa(jpDX_Rl!_=
z-_GBYBFUQayPm_3on1bvx8a8*9nfLdBCz@sONxW_IvpNdPi0vise7I{_?*R;bJz?2
zDO#q!_sbD*i<A$MLix|;&lP21Pqyr5Xa)&aG%B`!rS{l&IdXFw`7$`iq6>W`tKj>?
zR2%1qICV9DUcF$Zn@1~n5jAyCX^@gcPY8<{VM<(|7Mu83S=U0(!tcx+0U{m5PKSnx
z$Uy^h-m??Z_va?-{!zSkKOJ}w1b2t$lYErinhbL7Xi56DBNiEVwi>d69oZ+g^|kDW
zgoVIS+Y=yVR!uY)DJOjy7nhJqIEPAe3E0sSxP$!JvV#f7dkeeb_77VsBh7Z+IA+a$
z_VEte8EBC#J8w>9hu=;Yw7MIHDc8vR?C-Km>!>nB7gBo3m1{j*cr1<YL!kM1p6$r}
zr!Edep0&U~=vzb$`%S&YE&eKcT}k-0Ja=z2Kx6aXmr5SU^AT%BgNDyrjbR9Qw^hX3
z(`zWn?QYAZ>*}B*2cS-rz6rzoj5^T*CXu1}46(yCbGT{`Z~NR`wo)-jTSNuDxy5{X
zc|}bXBA)7rh8Q_o&E936a=s{Z02{2+p_!g&JUV_q9DL07ZsqIxADB%sv!Zin@uqY4
zLhXZu+p^bO8n2*o+xE7~Xv&mc3>?;$?JR1+`WBuZdN~BB_{I)7_VrFIWnZVV!H!9d
zVVe~%Za{L4Y~VS;Pe&39nr83D=$Vv(^g*GDm~MjXZ7$=dKGMF&6P@;@7taq5hA*Bg
zA~Oua-*i@f{#U}!Zt@SGIOp|(2&D|hQYd_C;csSdG=nO~>tZ`?DOndX7Hi+C;+&m(
zpU0pcZ;i^la^Uj@4Mruek%V*l#zHSId=qeg)nYt&EtF+RygI(N5;cb^LS{AZwwc$2
z9*qF9&2<ofK)bv+5$4@7tBhj}mqPOTsdC`ME1+oL%^;7<Fb_1XO~(s=v*=i*S0jO4
zNK&Spltt39vz;eO0rDC9;pCqHH4VC5Ps8g?_gS*>Oo4h2lRBaG(i=NQ1JoA=f0HJe
zauoctY^<{cFa;#F>K1r_D@uC$t^!e9SfbZrOux@IwxuNX-^^2o=+8AQk;qSi1_&63
zHzoMV`nuWS+#VAUEO<_Y`mYR?2<?n8+Bdj1)$?#(_5}})ncU12<AUOsVhA3eXee1u
zW>R~|q_+!R7lZhR3tDB3!xp31KHjR+y3g@4PuQ(5ajc8<3r+2kvk0)32I;`@y_hpq
zkf{eE<$11|yp+dsRlM`JBSl92jp_WaFDYoJqh{JLA}Q(XVCr9rpEH-W3ZmE$Rt%1e
zIAOKg+zdP)ti!I$i93)_7L$7$;uAbCiC)jJ&d4Ib$i@!#Wmqutnn)uOZa##j{z$-C
z$KoZG*gT{iEMaOgoY}O7&68lZatK#FD5g9Vp@T&+I%cI@W>}&Gc3}U5tHBzxlb?n0
zQU@peAh(C;?&tU}NaMYk1A8+_dU@k>w%|R7(M>2$u{5Ns_&}zgEx>YcP35ss*eF}g
zy%c(}S~s;UCFz-sZJsXabX-3MjS`bi8#~_a3+&Sln3a~Sr-$*|7V(X1v=*BUmpdAN
z$wk5aTOL@6v(=1OIP^m4Ai>bmPB+ymo)8YwM#gO}Ym@@^m24^TD05HhAaP?8Ew#a5
z0Ux5nb3jj@L*cJ+8GJbtHO5sAxPx}a#4t>RqZEO*3G4YDDpj(*>S>t=RpY^+EPc6v
zk(bpk7C2YP<UsI3T9w4XK5IoX<${%gip5RUKyP6}$Nsjsp1G9oq0(ydaGBcDV*6h9
zXmT!2_y`fe&NXiJjFhs@oQT-Rkg2Q|-%dY$LIi>quNugQU12hcb5Kp3<;^rYC71Ig
z{Sz0;Hq@Re?8k{6PdC|~zqs!(VYzs^OtFPEHJO5icm67^#fiXHRKa*`SepDkRV7E*
zg$P`Oq!*Alc&4D@(Vo@4$FWY%o4DI{*~d*WFkN@fgGv`^x1SS>FcvlzzWF!Cq5#zU
zDO0IoqilURLtbY9*d0%MPp9-6Zf4&y*E9CXi=3S-l~wd+?H9~yTxnU)g(~FSv!L<H
z`d@c^X%YRy*u&i-*sI!WuCFRgLy>8myAIiGE;^awVBE0sXJX{$>U*42`DZ8{F8qvk
z+T!o-F(dxTxq~&?`t<7J0x9CILf0TgS(ldFb8H69QkUI4Xt~`HEci+!e0+Tp)rA2J
zmF0TX<6cO4ZjkP-x5U9aH8R+vuEX|}p62K06MGxZJJgJBOW_}odW^fWwxNN7Xr6}!
zgBHaQCan5z2C{?n>lLc&P}BKqWuQS_iEyI6q0Si*H4(IPssbzqBH?dDmO>%FhHfA_
zv50Gv@{X%+TilXB*kkyhaZwb<Y#CC$LHrMm2SZKWqTi{40GmQ##gX6kKrhZXEmhZ8
zN7-9vGM)FPdXkZgbM{K|=Qu5)W>x*H8v190%3*rTc+|+!U6>^jHIE96+LC=0ulF;-
zNAETH#H=`zjs!rkJB83kgr&9grzxcKEET!3hG2qNLR4e`%iUDr>_Y;Y!c|Bpv2}o8
z<(#N0kW*;cog$YcB}X!nhKla+DRz`ef%YtG_in7Rb<(c8Tv58U3*qqJ6&9q>5@pOG
zyYm5xw-Ix7;*SKkNWu|HgWkRy7(WyiM?z0n@`0>nyLUkig0<?w&%m`oB%h=mAHf^Y
zcWrw?!5!zzVThl_OQP1Qo1JjFZ@Z8O4f+af$raW%;~ixq%i}TCa{OX%gLX3>%4d@<
zh`T3)L<&fqPtW#<kA5s&uDDV1jhUr8&7@qORDszcdGm8?OIyt$Z(+d(MZ1n7+rxip
zxwXE^=q~XVmX9WGXfKquPD9yxCG<-vmv9kUjP=}#5v&H(9-kSAY&;CHHJnV6O}nva
zV3kE+oQbYZe<dv6vLD;1<#^*>Ou$CbVoD>!gEg9ffP;#r?@sW{voq6Aeds$yt<miz
zM&dL4vrcGL3J$J5%Vz)fH6hm_6XLo>t-WdJm8#3X`{Vqq=b=e19?Eu(1>VdbQT^ro
zdTx8z!rmBeb#0s4n}t}S;jNFC`V=u~7bz~~{A--<r-#`VLq*>g#dG5!&B%f@fNiBJ
ztkj0blEs%mSF=fxqqi)obl7Tp__#%Wwwn|+kA))%TANX!Xd6Gvn1ovw^HZr?V2qC(
zQSeCjD4<UHymM*tGYz}OH`!CicOO^;S1Smuqb?IJf(tU(YJ~Gj+`T$~T>Vcz0QdKA
zvPq<BCys3A=X}$;1FkkrkSOs&<{6Pvdw}!It3dE6FgmVS<&P45242Pj))!g2@}Xc8
zf_<8kIjB)5^@3Dp1@Cl)Tv>X((ncd8+9@!SH*_FBdYY04%?n>#NVAqke>I%VA}$R{
zkJI#fnhnW_<vdW;15n`faM~?&0}r9fy6e+kK`lqka6MjnzvV{y8tKjNlxc5|y8Q|4
zt&NG@(Tfn`$&qZTn@663^1=dilZT$lnw%t`4z&c#jai7@a1ZvEf7P}p`5~uaf(Y}k
z+_&u*9`-3Ek^NL+9SD-Fnn&2c+(Ap4H9T1cz*L2V-;TgWHNUZRwfmFXt&(t0edB$G
zgh`@|7G5P~)L5#KV2MW9&s9;i&TSVs&xc$E?jb}7Q^C6#DzYulTZv%#n<`WUD-$8n
zJCf6m!g=vW1qe#J`axY8>Yg33GJC7I`A>%PEwtRN#V(9OqYDIu?j1HbgjbE)g%>I=
z)iwRD651z?PKm4gmX$J_uuA`|Zj1MB=h?sWQs)rF4=rS!Fk6sl3aXvOi<bXx(wL5x
zcAiQ*N4*F*`XrUCg{4*33PWKbtTjZ>y6Ve;gJ`nOz~(~LXHXC-l|!hA+|zfwrWI;(
zT|YAc=uSMX<p|7TFKEhyP_P$@Q?h$uJF;&KDjHu4?do|8i)RD0o2`~RZbsw@dAr5$
z$XQ*IPm7DgYZXd9-h@j-gS|*|+~2S7k4KbTg_n>Hrj0u{M)9g(rbqrIlmsO@bU$Ws
z&jESSR>9$-V}9^BVu*mn)VDB2?NK<exs%{=y!vmpw$BvQSLg7ZE-K7y4S3f>95%+~
zhBv<(BYg+DJ&#|zJxLPoKS&jBm$1C1>u+C)r+r{%{`F<KXs2pKqYWF_87-7C{!~6X
z_mRGy+81Nr=7HSLOk?%H%}bZh%#<x%!E5kG{cWL=W9MeASP7yo`>ll4GKY2Yf|ybT
zKmS!0Uo;i0;A#zF;JUfs+1ABBXi2Zq4C}Z=$aoNZ5UFDcdR1yuEX7E?6S$IK)iY?^
z;A)ydF8+jY4H<o+jaBl`{ojDY|HP~Ro2&i5fkQDtAq6#sUvQ|v`v1TU>HjA;WZ?LX
zDF6TGhQEo{|DuNf!HfS7)X?3?-sb;;4F5-F{C`V;{|0gY&qw=T$dHBpe?dY<e0DY#
zmj4Eb|9g&>gN>Et|1&ad1W`)7LS?1i+9Cn^&Ct3?xj_Gq5G^zeVG;AZWI$kGqPUAV
zo4B}C%b)q%yc>^ckLkCcrf!=ljhRgkU5^`&ktQ3__7nHadS`j8SVZJMCx=J-2tcD2
z=iz@`U0>UN7kUhgr#@S>SzQfjQ6czAAn2gbFECm-Ai%vCP7wUhIcX$Fz*4t%0M|Fb
zPC<jtegkf90A5_&IUfu`eFFfHT(ni#+%XXO06~7-)W~~zHeK}9K7?=A{{TZ&O%;He
z(o*^nJQv?eDsa$d5J7-+{#K&Vr#n&2Tqu4xb12Y0uOBMlh(IFjYtrGVtE;Oi5XYxo
z|9<R3H4cDny;dTAv_n9DH@{1Oo(T|BKy3jZ$b)`E;8QMto^OQ<fvdf3esu8LQd__c
ze|+2+*lc(%c<5VKHhwu}pxi4^|4*aJyA(jM-J4|q*EZK5oNJq#9lzdKzG;3<&C&`i
z_<g8gc7V%1T72K;cu=%WuxUU5!AozfKib^6G2i@j_^TkM52m+dcKG}fDqsMfl{Yu%
z+?M)!Y^`7^V5>(iL7P1T+^p&d24rwfP9OrixT!r4iU0%r%`w_9N8d~;Cg6)$`#&pd
zxX^VCKid9TC0H$hz?at$3d*k=`+tMpVx|Fs!NBitZ^0w(0r{^1Vq>a^UT${ZyRCEi
z@NaCM{(bv>Tg$Wge*J^U<*(&Ay@OlW0XhZw)2pAI%m4EO{H@u4cm%TQXSD}_6$^y!
z|0ee+#xnX%_hxrbUkle8c@pQ^36Q<h<C8;;+hzn7)cX4U@7ttx^S-`(T|@HH2lAVZ
zm<;R*@YUfy;t%^9up2-p_opDh-iXkykC+=Q@J|-V3}1ISyaze|V~rtR$74-k*LODW
z#7;yX<m-)^#;n+|KLGU){AQ+D?8x)oz{ijD10U^=&*eAH(U0Km&y?8M!q5@i_&wJ5
zPcFY|ZN$yaE0HeGUd-l7Ag?vRp^r=yzfV;K=_u^w*?UhZCH8A8P(3_YiVqvYWghS|
z;2ICufECV9{<yAtzz)>$pJ3l?A)ekIJY)beI=Go1y%<fSi_ce6SMK{eIAC|J^q+18
z1Ouq1k1lObe?b6zXwb(YPwcw_V({A&FD)Ui*cm)fpw>2g`PTxd-MUVGUmoP(H;-&W
zPXOz!o$?-HK!BF-9+GXbX1joIA!K0NfFB`b5ZjJ#qWgP*R)QY_-gm(t8c6PS-fr(7
zK)&sZXT1G?rZGO1A^@A-_^Z0tKg!irWYEv8mp-Sjs}Vk9K02uIqhN*+96u=LRJg}V
zNXz!>BvMCHZ5)&wL#<|y$w>=7xXb5mQ4z`g8fh5ye?CYu7-xfeUbW+nw2b|pr+s?r
zP_LX6<7FRYdaIQEuLq5pjyKuQQtJG@*5taF(HAq{88rc2t^tGlrp2ScugoLcuCobS
zXCA_6yF3!Tc%PEJvAmz06iq4=#BnZ~9pjmn@rSuvUDFw6NG*ahaZ#ov0;WY>_XF6x
zzAeJ(O)mXZX#8-PZt;_QW_KKu=&sF&ZOXKAW(orKF@h*KVCFgYZh;eeHf75(*D8NP
z`iJNoc%u-58addp%_1^iwYWSnlJvGr#l2jbR&G79ZbYpi@gvjDq2@#toFsVaP$G3Q
z3hUP>)4_G7ZlLqm%o<WO&EJex&2AD!yYM;R&H(9&q*Iil5^&CV>J6!v{ibqK6myTg
zxD{+uZ&j4D7R|1Pv-?kzFJ+VNeJ9qWyiHd}%_V9%n^*TT;?CUC=jg6B&>~PB&mVce
zw!9qTkmzTAQ%e2lW<_xzd0$jb)Qm5qWD4RSe23IKa;pc7QnyVctHhSX3Kw3&PefCK
z$@|uhI%|KLtKk}Xe#CDc3#i!b;;_3hO+{OK`Qwd`o0DqIf=q&foL`A@#En;guWYkl
z(YE*xqDI*J$g55whSUwO!4YKOtGMwxeEL=K>^n^?o4^2r*dl5&)f>($kgMU|L2KWR
z$O_^MpcOR3O8=sChE&TsC$#Q3mPiYeB=?|+-b*hZc(6a1>tMq0_Ie}K9YOuD6k3b%
zwVKJ9q_d#XPBp2ntf;y&dsg4PbLd`A8%Z5eGr5S*^-@nNQ4|&rYMe}_EV;Y+Ti(t0
z!3yUUrmKK6?SAGz@B2}%pKR{Q?<Te1&532Y{!G8+-pip$oKdjrYxZ}{TwAdDbS@>&
zby9fIS!FW>jtdqEV25?;Af`lLkE8~03zbzDIcFGdDXP$8ye^5C67-OzXrb~vbD(s`
zE?Y~3dL`C6IZPPDpG2Y_dvp=~OUtp`EyIoWz+19Ovs~YNEW=TIO10IwGodB;8i0#r
z@A7T!tYNc;cC){H^J{>j%dnx9NbONtK|VWIKSqt0<-vRZgniXG^K5~PD!LL;fpNQ{
z(*oTbj*hRbCE35XxzaCku%|{UUz9dw)CT6z+!13n1tQ{D*$#iRc^~I>Ti&H~4#P$q
z<3?9wR{kAHZr1%WyFmQ3bzqH^xPCL+aU3EzQupc>#<liPn4OL!&^XORUk9IftG<M0
zG-*{neSjg8xy35;Q<)6=@|tMc2J8fTUHD{27M<j@%B6Y-0yx9QuW4_Hw<^~DOyUT`
zVfhxF%$EJwS_3h<gz~(feITSrreRt0Nu2qlz5cI=f^miV)H->4$~&RlM(f*q=Y`q1
zns$<a`t3`pHK7psw~kA`zP#d;CF?1QLNQwa7ceYaqIen}S-5shH-KTIJ7B6nix{M}
zVmcO;YxlkG3AeVw=!Im4Fsbls9{+D`eWlWUDZeG6TW~|5x!L*+MHRe`ASy@w03kub
z^b4h;a>ERwD8=%;(@gtA9$l!`8Ef_nVdOO)rTFjY<>CEew=KHoIUB3YniogfSXLD{
zA8=Isx%E3{ma2s5{)Jhm1&+_P`|Z1W!c_BF2~UTe=gB{V^{;NknFbAi4nPrhjKc`A
zS0ksZ>9xw;*lZ}K8v*Vd>9xLUAHNYL=7~b8>iI`INy&ZS2pSnPFrOB`jgPd=rL0RX
zdMVSY+gj4mhEuCIYRd64sOfZYfHH?_?+FcHZl5gyo|IWG3?_&y!raVQtt-MGE<*;n
zf$<Owt*#<(5bDPnnL{tD7H1YOU}KK2_ZM5?wG93N#k6~#p=nWZo>roxhI?6(()|wV
zn!H|f^x~P1&$)_Nv&Hj{2Tbiuz)|WECB1L?q3^lhud*FNz5*FWc~4g&wJ?F0!0xAK
z<*1XT6EzGef|n_%-&Dh?;BNCeMoSrm4FMgyd+CrfxaNHVPC{eVe{N%dr#KG28ChZ(
zDiOr!+x<vX@j@_vq@{xuCH2YUa94gm$yNZRT}=>)jqE+R+{%3FzA3tSnKkBlnRUsO
zzen|IxKpC05vpg}7VjyHL5Zun_6~waTWGHImOC|7)-+=7I_+keZyGgFst1!5GtEUF
z6}Ss+(XgJNETB8t#;V<W<*gIf2!Hn)q2uxJaj2N0FPzxxnmCaPw;@!CA|J=g%pCSS
zOnRK~*v6V`xpn_*!catl5kCez9!89Xf_B0N)9dh8_-YOZ+xjnEa!-NVqycr=LhJnC
zKals-td}{vv~e!#v12VSsuCPtG0p_|Crpp>UPk>Ba?as9X3TP40{U--_APWn32q!@
zx%Wn~yH+7e#x~d`lu9NMUgD&xyITULgR?6F<HO!FutR`k?LubWGG)r&H}|d{rSUlV
za8@aIm7XQzW=;QM=)#+|0x&yYl%DWar)d?CTy5hg%kj`^MPd65q}*CuyO@IF4Avm@
zez135-8I)Yc09R`Kiy;(Keglj(DGRI?jf9K%6+!;U%UjhnS=#@?6j-E+4Y<Q2lxp6
zE!PKU+AB$f5r0)H4;OSAoR3FU#fW^SWFhx|qZYPk|BOuS>m+u{k_-K9h<chp)5y$`
z`Hn0PX0HSv+!mdSt};{gMtynlUA>M&!-}9RQilQ~GA=h8x9IUX)3w!{35n%%k*lsr
zj;F@0iJr}qY@~?hS|q-yNSE7g{2d<GL9NdKfvR)v7tL`Oc3PGsl7EcJ2+TQx_3zRS
zt?yvg8Q1DZrFReKwt?)|L<A~41OiHAm8K0Rxg$3u{#}YmL6=xA-HVx})y_G-%srG#
z4J_*<Ywx}x1v4V-C<{sciiIjb>DEgZ!f~7@>v^x)2wNEs0Cx?4`IMZKBKL9ynJ619
z0dQXZ<*e1%@l$}t&bKtnGt+m4vnwEEX{XRfJ5~Kb!Q4^zRD-`4>-2f1__6jQFy;xX
zsHWy0%JlWzE~YBqpE+?2Zc*+f=gOidF%`A~9#QgE9B>_yc8^fO&muaBvL7<hSpH0E
zCP?9o!{9up%0Z5ST=(W19Suj%)-OAbGx(h-vUo2PcOE<pi|vTX0qk+pS%7v{ytg+Q
z6ivLg!G{!EaVqmZ*Av;BTB+|{hdhA7JBtK)EGmU~9_-EIsa<O#fhKU<@O)t4#hrJ;
zbok^z;I5CSN_LQCpomdsv?R9>V*|cyWO2Fd+S;3c46EInC>)vwsR^Q7aNSu;I^NQt
zc?QnB327A3y!IQ8Xfh7YT7p7tQVpIo+-T{1U0T1K7qCRS;xr-nsD9|fgb<r~9IDO6
zxtc?Ayre;V_kZVYo7dh`8<6)3-5i4y{rRg8;zAyv@nsNcTq8&7LAQYERF?HERow|Z
zQb}hw4dR`+Dz(iMb=j6$K#|&-Z@Qi6f3umsQuIf#@ewl(S_=1?7&B*_2-_CLYt$#b
zTXj1g7u!+Z1l^}WNont#N4*kzTxgy`xZEN>tgE7*_#7T1IGN6gWrhpdn^R9IZMll_
zi-e}~MbQTVZ>gd>5edm|rwsmg1m}{co#K-WQ@$cGgKI}${%yfrl&nZfjdBNkkUQ-S
z8Wjd2ijPIkuGOb&&H|-0RR|{7gU*t>>mh=bsFRjsV6Fw+KiJTBLf@0x0L(P5P+h#s
z6E9!vcDvAhT=rCrQm1YwWzU}XGhBiU4Q13qPd}nDY!m93P)0v4%6kH1Jj{OtJPue&
zm_X}~vK}M*<D^E*4rSEsikeP~W)nt4m7$Dy(de&rS44H&a;QW8X$Vbd2ap1-?W}ve
z=*z6Jxigkz2+9;W^yf(s&t~(v_LVFgg9xqt#ExO^c^(TQ1us;~@|I@Ia~0vE24X)}
zMQSF0IsQn)Y==jD#dOP~;@BSrnsDD3S5Uw5F<uhbn^vHN0e7l#<;QT~a+znIT`X7o
zan?REe_U2h)UacdvK2rA^ZG#c_y6;6Z;X8)hSi~W*<EA;O%vCS!DO`%fb;)2&L5c8
z>Zaby=Q5HC|G*yEeDUO-K)>7{*)SBzDpBMOT4A19t5BB@Y{>m>-FTy`_t)>NBit-=
zP9yF5BJHe)MFdrC7#oQ00DbW@+0ln_5>L)dpT7KCASGwn?PKq;cTdx&84K0YrGXq;
zvlEU>_mI?-TlqnuPpgg9orKdsJ}ezY&jCNeE&l2!pdig~uXf0?HF6GC#`_|gBrg{e
zzJ_n?eOKK*zy=W}wAQYvVT(G7MIp`ouAp@(AT4Rt8A56yxYlS&?iVxFTSqv(X6PQw
z8lkbNFrW_O2_VBkQaOWQYB2!8yCrs8y@Z5Q(I9RTXIyF{vSxx#!Pn+V$bEl>64xI|
z1O_HVN+M`Jpz+pw=~s8yl5@z&&@92wh_Y9pCh=$fS^;a$5iFx0WAeWJ-ztki&1QCb
zLuP*>OoehB7$4VtOdzW2{#~(J&~(Cywn2BcSq>!E#Y)&??Vb5>prVMhW6L9V!aYy?
z=7|>EYBZ&)X^FN{994%6rnd&);M7)AT_cA7KK<VN6QsIEO^EcFWCo;e9TFk;QirEr
zEh3sxY@-QZPa@w&erN>SwZqF?Q*B$v)lz@sQim$}5F|)@uh;q71PZvw3He6kAqnBW
zd(n#IIOLA@zQE4(x!WAIXN<8SR$wzq)ONiAHvG*QGBy}Iw%3HpO-|x7(6!=DjET^F
zlfyg+!|&8C3j*J2)e=*-;}G2`iZW!J3og8d$(IQ!f9V~rYqLoVLtDQgSesbu90bT{
zUI$?)9Hbw~i}(57l0cSS?VDojxC|c9jIeq6a7cu|Az-v5Cgh|vnxoE|HN;}~On}He
zUou)J^jIfI=5%M^EM6jf*(m?0z<Az`zKWcOn$HRmK8i|Onrvj|YfGQRV=6_M?iNsa
zQC#u!MGjhdwALjbVZBomR8qmSc^gVQ-;=t->egh_zwuW}sAV*W1a+nG9szkN%3aIE
zrq|`z*Qi?dh(JNAuYxvia#cmC=!?|<N^Calg6HrVniiSUvoXl3-)%Qu5LGFIhY`ND
zQ;4v9UdCw*@iFM?C%*`|=jAO;a;|w;wgw^gWO{y|y_T`gAb!ax>%6NUA_r0Tau>B6
z0@IcDjBtH<8?L-;%0*=JQ7{w4;F#K1O3KCPqpBOs9@FG*vV3qR897?fd@N{@ytM)3
zX9fPHVr*TWLOe>^r5T<30$mA-Y?#C9@NBh5aCNpP4T98ked1SYzx6{+y-T3^lcJWA
zx@BDwFu+C9VzX7d%F@F6EizGdIrznVv1jCa9&BT*FYm>f1dU<*odKpkyy}P>iUj`U
zlqHPa^Ct;Jhk{nQkvQso%(2W2zXIDeq3@+1RXxgV$8GB2B3o>>YakP09I+@Lqly+L
zjSVsHmDvJtt*JYHO!C&$lq`k}3^h#@INf{wUVc+x6GVnBD<T$a9)02nOPht%Z***b
zYHr0Fk{XQ=zu<wbgCZPe-B)RGKok`Vg=TJ`>yO|PS%FEN=QU9J-P$99d7(9$5>*bk
zX79kAHf95F{zoUZnymgeZ}N=M?2JiF30YJM_Ty?PoN}Zhon<?}^j{L-(V&v(VwH6~
zRw!sE^!k^@L630?Nig`=yRD$(#i#(MODWJ02I~;!J3etG-JY(kA&+f6w(Wz~pql6K
zb{1k>fbBsd#S7h72EX#ZUEjVSxvQCCQ)#s~I^ao9I1kLgOp#gF3K87k^vaZ>OB-fP
z1dikqr?(nyUC!5Ek0OHCPiV!;Z$ZE^7Wp5H)AHF<Q{%3XV0IT1YAZ8pxqEDYJ4J`9
z|3`K27+zVkZVe~hLC3aj+qT`YZ9D1MHafO#vxAOp+jidcv(K~lgT2o=-*>%#-dwqo
zHSbw-&Z@QMs99C_SYx<uG1(U9dL<7IWPMp8ZAzxXVA^0YW<M1#wsd>b37OY?sbbHV
zdu~y`-Xm4uoZ&4td$HNIAJV9erfLh?t<yg+9VAi2nqTErRuKs|%30FPDNwe-qOouU
z1B?w89Z4;Y^Tb3Hqg7d6yELt{Tu<Xz^saoY!`*bMPbK6}(DsnT9n(D>^~-hhFlUJ6
zEWOs;Zxs0WQsPND^8S}pKN||JJK1#;|Gtu(3mo}3dqt8u@PNLf7t!|g885-nt4(WY
z0#<m7gCYqVr4uPe0}`#3<{&F?qHVU90mDq}K~T2F06ReyR!T7C9uQP5rWf!9DEsmq
zI1OQprF8$2@6ESGeJ|0K<w0&w`dCx2$SXYiP=v3cfW*dAgSKa*g&elb!1a(Y{9SUK
zDl{@eJw6p@d`<S$hY=2wE*_=zB==x5?ft}P{)d8iKe0`GOqD{b^U%Q)hV|Vn%IMYb
zV|>f5v^#7KqTMsKy)q<GqAk87=gi;u4H~L@#3||5Plyf+1_1Y074yqe*zZs^Ivyyh
zgM|%=Izl2aluEA&Sq>!6t-vn{Us8>>(RFz%(GqZnQlrGpMV1;|3z!zr6Y=d(21BCh
zrC+m2B@n}@Du32Nv(M<7LyCn1*<F%i>YU?HdlqVDi^8<dIda{QusXSvOZ{?uNNde3
zCGvde()QJdUOYf^w}GQ4h@srr3_7259u_JJ%A3tOjt_z)G=BW<p4bke>b0q;x9U1`
zgZV03p&Om09d)Kg`ca0d`AjbvPc$co8&_QsVwKFB=g};0UTUgiNy<&TxnV)&7Ol*B
zcEL<wfm_j}SJ<e(%7x8bE#BwhMtp#o%EZIky*^=q2w>ebW3UzO{;)QcQx`B0mAgov
z+f)AQszoYm3=5_^0Rmk0OaWD*QY~O<L-;cMG$s5kX@3XIc#2IDn7)E&8?&o<01hM3
zrQocv_#5bt6VeJfqqYTKI{$*AcOoRe78^#(xtq0H>COrT{sM+fi@+IT1ji!7pHt8Q
z73=<aTzorL<j47baVPTdwd3X(&HGO93F{D4B-^Y72Nx~ZZAQ$R`&Bas@Fk0c<R#6w
zX0Xq#utZEjLZ2f{M>`I$Dm6#0s?7tIPNyr~x#Zw`WI1IRIre%+Fuv2zPJzp1<vgy{
zOC5_!!(VfMOS};DR2R3QPnC!k@;ps75D&#VRU&QWtvwGP1&uX~FKe=GP9b1GSPD$s
zu@ARHmp)eaO^*Mu<BvlC<`B@I4oI}$5|~Ph2ia;2BwdYSxQ2u~oL&%B9Yh(fF{jb#
zwbjKDOItqVeFG|FaQ;#f5p-e2{^<hO3t^qNMYWU$4OgWCqi<wMDfs@n@<)4XTRqoe
z-4nI#_soX{VZ(NJS`VI8GPR1SU26G$7Fk;1u6F|OoBi>{td#KD7ugD(b_c~fFM2fM
zwz(pp>kP24NJ@;{1kM(Z@?HK$S9qO!IGG$zk1&ZB_&ir|1vxzA);TCk5p3eS+@ttF
z9s4I<P#o2(*MT^HtgYKk%G_~9(T~KY41&fk)ZS75vE#<Q;(F~Ud5)xtI2c%ZN8UV0
zGVbA*i972diRfhCfxXBdSr}>a;53&ucGnZ|Y)1wyzK+ZX)Ku3r8~boF{n90&*JNF~
zYt2F7j>Qhvxl**~$X7pi4F{g6y}<RBCmi$MfGk3a$MI{mj3`2)7D+Il$d0|L9{o2K
zqJjxSmTRVE5lQv*DsTO~k~PQ#5K)P;Pm*$tsmerzv#2MIwZ+d&I8jM2i<wg#rMf-c
zWtt#m3~hbg(wt?I;gk_*)F;Dub@8(N^Dh^fWAFjDhpzH<qX;1~>B)jfNqRx5qXG$2
zZ~WXKII_PLlWwH+bnG|<(Xgwki3PcL3AqpCYdlxh%Cl3ggjPcZxSWG+&^+nX0p1aF
zS3x?asn@{n4?s9fPx;|2T;mVU;x%xK(e@TCRiLl&h9u4Qe?7~0%g7T#GOca{SyFqP
z&eWC+7Rq+y?U#hmzFJT++N;5E_eNIE*fLCA)9JW2Z(#}OJb#>1vhO|>w~Z)wtlYKD
z1m|X=x^9HHQ;vL_Yb!!17lE_?IoonKa0CWU60&@J^=WHJ$FK%@E-N#yWH6v-?T8H^
zG@XUQ=w<T@J%3-T$%x;VO7S5+iV<jJlc}1PGj=n5SJe|$4GsU~Z`R_D1Q?qx?JqB^
z3Qy`ftonR=o;4CyXf^x3K?8A*&i?J-<!wpLphk9+a?kC-+680`<#5nqrA~{do%}>d
zQ|Dt+!j_*x)MX&WtQB%BL4;^jMlVJ|@Gr*u)|Wclsk|>6#zD{@&DwFS)BL`2X^I34
z@M!^i%TI$W81Wr+w;_hsgy%&!9-O_0v3fUvs@POqz3uLVTH1Cc9L9aO95>=iE@H40
za0PyQlA{fh-FN$Nlya(2IqNN@-ETWey*rmySqCd9GAGHn^|FxYf|-t~9c1|CyUv>t
zGl#Kp5StiZNL(;;J>vPuy)ubckmMn4d7IA(PO`HgL)k|O8~rxQM6;f%XrqpZipgS>
zur>aZF2Q!Q%)^CM@9Cs<%8T!_(+lT$T1IyVg9@wi)_9oPg&r-*?b(nMEEC9pBNUD9
zgJeX6U&a$O!9;$qIouw_R(E(q-iJK0UVFxlSbDMDvCjD5qHmEb(nXEj7AC4hd?@`H
zrnBC<&HcAxS3Q1*V?JX3$r&S}V63P)%Uhz57GYvaEDZ0!{nV2x7Fl9mLR_Z@PoqBD
z^xRWDYa>s}*Kx_ztMg~nT_QC~p*IaN+dZRjk489>M15;=fLp)48wQLb8cUg1$z!Y*
zKyB3Bd92uNC#yvz;7}|Ygdx;2x0nTm^t*?lF46=k$==>1e0WYJ5OkLBi+1{@+6pxe
z!S_vkAccOK1YU`vB(<|n)VXRJ{(7%$#gzPAia%#Q^ykkzr(z(4B62ZQDJvvm#5FQS
z58x+hC}E4SZ{q!XFSh1#4V*K;bmLuI&+Etrq?_4@Es6^{;yJC`b}34l1TwwI=@b~I
zPmmNsHu1j)u7!#^7blT~jc;s5+`f=+vI7x#9d%&jI+=6$X<nJjugNp$m@qN_SYSd^
z1wE&a^(0Lm2m<CxIJ#*uWt~nA<FDpI24Ki&D`m$L8{Umv;Bl~jWqq*Sx`9WqbT)1+
zi3P~N1~KxoY3`m=0#Py(H5d(tvn84L?wZf>!N(#Z4LQgV6}E}4k-aSCCzLu6UuFJE
z<wLOYiR1_M8S5BpH!Ic;0K)5mDb)peRrZvE3sY-<siNX#V5nTaDi~7)StVYe!PogL
zzk!2ZX?|Qa=^ye|#GtcC+;saSBoeQQ<gj2L!hkbFnB>t#z)fd0kz6b(a9Kf`+=>`S
z{0Ieh_8QEsg&uRIc7p-k5EmnTl6x{b#FNKwbPKdu4lB~{{6a*-3jS%svO8X@T?}~=
zw)xes4K6Z5yQ|CkyFVm2BzV7+CU2-fEPAbiThHM<vrT)Qw6<qn*q$OZg8nRSV}>Je
z@xM6Hf9jGNEpF(2nUb|NP1xf_S^&7X?Pvy*8XG^D4Ot$M_FZjHZ-bB3n#_knQ5i9Y
z%jqGz>xFzJwByi>H6$ll=<EN4eeBnBnlL_lig7X4oa(k8?xvC%4=ZUymSY#52&Uz8
z>y4I%WlUPcp}bSKMwKyd+M!#$m?2Ll3~v$96EwcwWgLy~b%b03k%e8!EkZS>EgvTa
zFHnYNd|1vvKM=fvT(dFq9-s>;6mra`f-Pu))l4zC7-**$A>IVJ!8VSeO^1B$S4G9X
z-CyOjV<8tmbGO*#7;9Q-KuLDC7P9{Qg*v}wT?meTN?H#~F1AbKN8l%&0lVh>*7z}u
zh0GBMmD(G}ou&^b9|6%|$e%DQu?<_8kI3gF30Ehw%pI!=)f{fZ9oA@**!odD^F}q%
zAm5{Gb6q$v4lolXY%WF2-t+K@D>q^jLr=#YO+@T#vVKftBHVnZIt!@lmw6r+9?Bp)
z8tx=bu+m%~;OUf0$7c)~2f^DlC1fr00;TArr`oGCyd~IJC}VHG8h@|v3YfTUW|*rb
z3f!xzjj@Z97MVUn*)8lU7>=ev0rp*Dxau#YyqY&&wwdNTsF<fn((az8#gw>7`cy31
z&tHTC)7E<{|1crY8=vk_9BZ3ygZGg=tmiq4Hiao6cd@gO^<E+ClWy~1Uk5FGH(XlD
z(9P~lir{8?OdWbCcNdwe@Ssvt{RG4HN%&^^ZQ<<!V@5{CAFZ->J%aYIUD0oQjWXz>
zvAbG|cJVMkG(9hlnIN9HO?lH3XE2m%KgSGn`3iXe5bl<3+CmGw?H-nj2lMsR9-a>T
z<O0tK+R9$US{|a0hQo4(Pu2>2=BAX2nPPs`xN2NG0t~g>q?*Rq@R<_Rr%ol~dej7G
z6{lt~w1^t}#V!rB3XR-~KuzW;u^_^DO?=4qf{s6Qh;c#B2ul>K8@S@q5d`Ua42Y)V
z6SIQ|q*bvS!;S#W5nV6uh-;rdC75*v_{pHuD`py|pgi9o;@Unn+NCWE&Tb&$(u0Y4
ze@z%1<!rMj&2^GoL=bnmsK8wHx1fAz`lqB!DYwjGNLbNgJ2U9}eZA6Ry0jdf328@(
z`;>g@Ht`msFs}kJd8aoW`>f72EsUk{oyzEg)KeAX0u$FCrP1HnkS8x<gqDjS^7|Cs
z7S3FDVZjmrjcXKoP|b{X6w)p*Ar6V33nhccat263Sdv~2uaJqwItdjB6=w4s6JE56
zn#(vgx^RCOYsG^ix!R{cmoN?4%)E$Gciy*1I#ncfqeMy#op6Dp%P9kSDQul94_<WE
zwbcrvO?e6P6rlQgOC5$#zQ<Gu<(m_uo6wC4k}c-3p9%_*V>g{?v&6)+v<ah4s<~A#
zIZ302G_+z4G6v596<fkW4g@E*+erq-!Vk%m!R<+P#uaJWe|17Sq10Jz@0>an-t2V$
z*({0t)jZ!7Z3!{qtbND?eLJ#Yn=Tp1?v?cD1mhcl3RhewT1Q92<qlPOKAf8%^CD<$
z><0Q~z<@D!WxLW5GQ`g6152-T+bS!gz$IL23*jhk;pNHP_HOQ>Uqi)R93cP{GmjYJ
zcTX!Yb+<;xjTmd4Y<c)0akGj*cA`@#zE5-X2DVXIS4P)Z#dqe6X28^^z1$O8s_wCu
zpna7~5N-!eeNO#9qK_-%m-*t>DAc3{^7HG9Q3vc~L7TvOQC94I^Xjnk>KbhCi!>m`
zks8m<2jaVfYO|Q<SibPx)U^zD_)J@1TtR7F35kQ9(&Pkw&XZ@K!RI<|2OgjNq%p<n
zzSRj#99GqNbtE$oVO3`=FDUFP)Z96pVk)IXR)#_n^_GHv@Uh5DN=w#V%om~X6{0i+
zdx~RB!595yhm?7~ysEP6B62J~vBF})@MnDyBlwgPGr<|t%!LHxSk!S#tG|K63Jz~a
zm3NdH);rb0LgSQ$%hBp-c|t&K=JRP2G(y)UnM98JheLYk9(?^{KQlyOkK(0`IfO?s
zs)}>f0ets`0uY_69s^J0Iih(WQMh5zw>d(cHa<zBT0?GX3T{qZ-}A&ce7h=Wy;cIh
zbHOkseOVGw3O(Ef9<9fe;E*?XXL!#0?$0}!XaMSgVDB%K^~{<-U(l7IzXc*Cu$tZ_
zdvHw77ZlqO-lMbJ;4VmK)2@A~aY}iEP4X1MS<<{dI}xp>^@Z8)+LUVL9gJ|DYk`V=
z){uhdl$A|`xyjRVSB_HsaeSRbJg=<a9{&1u^SZCUEcE#DtHYl1tpo(OzoQnuWCMm+
zQ}o6rT)RJq>piPp%0m=F2v8)6K*H-N=Y3!&`zi6MM4XWcm!0Y&YZI7LvcR%Fvz26y
zEZPM@!AqdL;(5fvDpoR6-biFdi-qYs&c$Qj%-?Bu^#8Ia`Q51HKkIb}@QKMt3;&|s
z{Rg=YhJVU+FfjjLq22vAVjc1{iZq5arZko`Ry5W$)>eizHZ=A$e<kDn#=VpOBN^{^
z$CO{rH2);yF|hvAyW*FA2kl4u{;}_mWISe8CdU6Y8BY;dQBkwC01;JCyafbMh)3_h
z3h0n}NZ06_kFHTN8aohNUaMRS2ce)iKTw{MVtk@7Kl@GO<@@^UoBMo=^Y{8Owr9`G
z#yhW!pCg7klv$8MY+W;QqUqnNa796iO3W?Dpn!Y?f%$OZ0B9<~FoV9m(tURMse?+1
z84&+y4WI-G8Q!}*)5XdujS>K4=8F5RCx=%T3{6)I4fYG^r%z(<5Wzh^0A{-~Y)EBQ
zV2c9yd#V7mln|FdKReqEY>vDfqx=AE21ZOwOx$@#f>w0$)7^ljg)f2_VC!AEK*G}j
znm`5i>9yg07wsn55AV%3^!IOTYXj=G`pSnmAhwCjGXW844}#uZMH@~63I4)HD+_Z1
z^30Wj`whip3v%$S0V5aFR!avGGMEK-0~QrNEb4-nUl$d(gLUFVz@sQC5YOAs(d+Yr
zkM|u2DJAL4xr29(H?c2)H?SWbY&-^f_>B+2_ivFlK#)b16$HW81DBt>MGs_fQGJ9q
zx4ScRSU;V71ARHdgj7^v2J*f9{DuYt(cLx4T84`p`M{PlG0ym)+2QA3lC!d+BZ`T9
zQOm=`3Ift?-5ND<9_J{O(-!UJjT~u1ICHrVD+^N>O$L8<1~{$w{JA>~z<W7ehc$~;
zRE?h>6Ue{$v+f$KadzgRl*dGsw)AGE(F+iumkA>JK`_oIWB_OXO&$Oq$^`_FK3z^7
zK(F`J%W)*$uc%)F1Z;RgKU^ok9ymDGAnjk;2XMGS&Y<c(Z8noY0=;^CI9T}=q96tc
zKD<D^J9hA=k_VzE4tu|5JgjpgA;q&y$nXlHlj2fQ10y0M#{r0oi+}q3mNEbU`92MD
zaQ;;)j{uzXwN?Fv`L$GkoCj#*nIKZV%X@E1U{e+1^VO@Aoq06t7hi9%_s-%siXCtL
zhjGexl%e-pJVzLyt)#T`q<8P8ZYvxAaU6h*I&9W900%i2G~9OtGw5e3HBD5+v(39q
zEh$V?8X&_wgpAQ&eipdZ>rSl7SZ7dv=v}t3`|I(KmMpJiOhA}m47|FVJm<(Bh+l4S
zxZ65#D;}Odhwbmupg_`Ode61{`DbPEI1=Mi0|M#O>fiu;ks*Pzwav<;)ryu)zNaDn
z8~_wUipdQWoJj|85ZnXcg<VQ3`z3-091-zDjqTt8s=pQRofPo0OXp%tTe55Zef}w_
zLYLMW73g9U3mpG>BL7Wy*|wa%G^s0TbENhf3+^(}b91W+L!sGgj6E*}9pQ}pZi2iZ
zo6+yyP<Q+MMb+Ay3;VLc>yp)l_|^(KazlDOXE>5~t!t<eMVc0xG_>DTk)BE~#)zXk
z`up%&Xy1`h>|^V-?2y%D!qIL|81%_7*CA9EM%6=fDn$YkN^bgUH6NnkP(MQyMqOP&
zWPbnGeD`FN)JQE@p>F*s#ig7S84u)gJq8dSZ?D@2L+#W9tqv!d2tg%OtZOE1s*02B
zJ};vu|3S5%Uwtviw}MIx$u8$p!d7~$Mq!sF#?qzsIz2KbXpSZi@(R%<vvD3;hH?Sn
zQi$+SzsV09Zf#o-pJiAy6~|~~9HA|>imY3-z*SPqUhDHZT1~Kvg4j#f$D3q~jka?1
z#bw7K*oSrE3iBLzI%nzhb*B>y_34iB_m;svG!07>{5;foGsuvXu_)xO=4x!RJMyX=
z)x6si75lzr+!s=OBL9>6T%yC13*n&ojx1e_%Sphnn9ax9nd^K1yv!1JzD59m&1=mX
zk&l|Lcwmzp#`I!$%n)&=m^yk-YS9@t`qLuJPxf31@q%4}pxbe_TWt!3b5A4$M+uqB
z55u!-FHkZq-6RrjO!&9^#lSRyFXra=<V;2*CRSv63R`Hkgy8p`()V8~-M;~cpKGy^
zRi<MdtsKpFzSmW*+B&W4Z>tU;x4{}%I*<oIw{Xw6Lx0#>WY4HhZzVa5JNKt&Z3w4O
zc8$!Fs^63DQEjqRNoLX5E2>3tNgsW-bu3bl?0XzSfy(R{<6{04I2O5)zv5e<(4+;u
zG<JWm;Rfz3?MJ6F_w~uC*psX~BAgUZ2)UBat&s10^-<Nb!FuftyW6kTVv$S|L%+04
zP9eXzwsmEmK6tP`CFg;<E7Z>rMd`F|{IrtXuBiEnxnjL2#5v^6fy#-u+p)8^D^s`B
z$!sfT_IV(;yh09H4aR@Ps(cMlXr=y2Mrk%tmyp=e<X{>YQ0xFED6e5-G8q5W@DoL3
z>l|PMIs%3`pjR9Jok3j^MJ=oavxF+AuEr-f9Mo1@ZOO^E@W{6*$GMlUAAW^KKIFwy
zD4=FPVwRH-wGMl~u{8HtIr7s$1@&BMjno%auqRX9t#P2FDqDRQZOOefiJsbq_D-79
zF(wGc6XF8}QD|zoMntz`Iwby<x#T%0V7GK8*dqY>J)X5gT!&V=WnP~{-h#nf9~_h7
zfk=kIT!2UV7VY74T(YuUET1NKp{3z5bC12^mXuIj4zfV79oRlgeoc|lHiWe1W^)<6
zXx0^q=+sWNt~55{*O5}B_Bde)k_<&2vY|jQ%}wl~)fC%|c+*V_U2>#=@bSLc)C7(E
zx{`vFEmzZ8Th7}M&W*&P>fFyuBKLzqLf0yC`-npb9>D1MMiNxFjWGrs1-qk*)naM_
zT$7SmIEUmS)L-Z;yJuO4iRaGHv~D=7H1{)@8QG8&stu@%z#`rBmLO^BNlcL|_j8EM
zwq#DGZaoi~GQb~X4@I@mXRhD8+)FFGiKyw%UXuboNj+Liyu4v1rf>4~34<M~XG>De
zg;RRmkyEwVkNpIBVNYA>BSOSqQ9y8*)wtqbe3dc4^d9wlcs!iq$mzFKb0Yg>FMH$|
zyyZS~mhk#XRCxVD{t%t!NpeGpaO9qU$^8<D{FS@^YQ5ALn)<lEZk+AG$z0&Jsl)g%
zb(ov}(W2LhJ|k{xc+{G#XZZ&Zai&HfE>3-pOT&B+8_j-@_v1ysXEQYcOUVK5qTBuf
znu-KoG?D1qcvbWFY%wJ;InqT_Y$7Br17vRwHE)RHL@u62#nX{w`}KWNwk#80$vKhF
zb40r(*hL#5D`7+KyN{~)fTz^UF5@`QMM;B^T=mpKH9^}wL(*n`hs#wzExw_oGurP*
zOsXAAAdhr0K{|#L?|wSgl8Y|f9d)gGs2Fc@JX||Te>Y+wmzr71HxKC*eA(Rg<io4?
z<;$It8+|E|5qU06#>-h;JgyY+HI@d9C}7vGPo(suYB7V;a1~N^J@T{CQaODgG9nb<
zFF|6~KIfWHQ}smp+H9x`rzf7B)AM+tY8&Lk_E_x{r(EnRT=bxM<|$=b*z$I)YEnFP
zKiSct?R&r|B5~hY+{lV5+#Y88rc=TY&52-_;iXYP4Rx)p`(|?CuL&Izn6!d+An|z|
zLi>>~X)Xl4q^85~xhPfuSXRz$vb)~ndN^*5qb{>@W=8a^(Fn1v=4R5@z&XL6r}-v}
z*L*vXvxl|4XEX2XBCBmM_v-+*e;@DK)fV+i+3;nPeY};G)SC~4o61rS2b<xRQxtyk
z4>+f&xj+-~y7vg9@H%;d`fj@8Fv46rszPdY6@JSTEv}8HKr!5{UM-`&w_KS*X-XE1
zoi@K(6PFvXYXSDOdhQAXiB;dV?|LXm!#SiV^RZH=1)TyvBip6cOU#4nv(UW?+H#0=
zD8_JM0gB3KGoj$R7sJ~991E@8*|PRzlo*Q0Mf_(-f@tx8eQH53J<oe+z|OVhb~_pz
zxTvSBH<+x96{d%L5D`-Bm>MnWrcO`D$~sdfh+d33Ssa;#5MRM!_z#0`%gHlTja^?>
z{eieNzsmN1yt~x+3V6j_j%&bHrD!1nppa{+f*G#E7Ks>Avn4xBfc!SF4LUmFh22u&
zCF*hd#Tp<+v$Kia<<bOZZ9;rkj*4ZEGsp_$Xv@euW3=9@8(b42m(QXqr4GRGGoN9T
zO03gTmod0)C`7%Kj0KI&o)B4V;SU*cr_C&WIB|quRb-UL69QMFI_^=Rq!0niW`oU&
zEnTi*{}kmlGT&+BAKuQg!|9;K6pER3S+0uWWvi}LAxs%OAgUakUg@xY?6oBzngbVU
z14YW!bTDc*EwG=gIL<#(4R-$C-8${j??iQGw<??i*kKkSPVa<0R8IjG7IQ+viCz92
z=jhWqlkTJXiHFr{wzH)ZjhWu!n5uhcsS%<nNO+;BA{MwH&l@Z50{-}p&RB|5(*iQ*
z<GX-(na0M1-B2C#yUlo#c3oN3J^cL1)7vh}--%G|^J@3m=H_QuSo$v}Wph~;h?qs+
z7JyaE8XoE>SoDqLB*P%>L!6_FTDY*-;Twjtia*Jg7h6^kW(7D|rFa-}kz67_VG8Cy
z+Zyv)MiIbMkvgqBr;--l3?~N^FGG||4@%bdTVr+K`;=&^%#Hk@p=-z`g6>&QEwYLa
z(psE`##P{j99=f7Qfqz_jW^d(b<kGx%EM}C8a@=28XqQb-5NZ6$U}9ze+mx><NxNO
ze4=>Da+Ee49XNA39h`S43DnckTpj)FMGM|b<5op%cE#Hj-T!vjb6x(J9Garfuq}eX
zr<T{vHf$T;c0grCPTv&<L*j(Y26I*wZtG6RU>Q85x0oVAJR7{t*n<rzNclwAf$s6c
z#1Ec`qZ`i4OdLkDou;GfE>HT@&3omPKlGktZ)NZFN-R}?vT_Ll18vg2`Jh^Cf9u4b
zLtDZu@esg31(A%l3MF$jguk^Vx~$kUCjD+%#Xi)&4ZMb&6I7rvnNZcu2|79Dk;jv=
zc9@YImg`4y7%|wMZ;LYzGn@KXGddel=hrRQ;}>0SU_2{XQ-gARlcBkjD$faDDI2iS
zqU>rdvZOG)mBNd0m9sC^wOU6EKBkH?GEkNMQ5Es#imsYl8srXkhgp@vm6I#B=o>@R
zWjw-2){0M&dQNh(83Q--T)<IRr9Z>KuFvTB56>s<^V67)K(!h)%(HUS_9)retSbxZ
z%5%h%W-SfvBRjm#c?onDpNC!0DOyA$xD_Y_#N&ed%KQg!x1R{{fT6Mqk969SOvU%)
z3H8(19UqSpT+JyFyW_39UD+8bAG-6Q^&P*~efG9$4v{)*D{oqz+_J97H|Y40_>9+?
zLz<~ZpzIh_5}iR>JRct?<!8>@9mj7KF)+TlJAHijnVmuf>hVg(MWYBB)WN^)KHTg)
z2yy9{>th<bwTFDm$Z6-nYm?2R%#~pCywPUI=-5}7HhR<iDbi274v8su9hC<)A^-`=
znFq4vld4^jb~M<DBh47=?mVzqIfzIf-5hi@$SJwBS^Mas7k1AZmR+>dX<K1`f)R9G
z)U7gV<~2I3ZS-lCvE&oA((BlW-zd32R$tWC8ZmiuL8wFELJ&87WAbA-DB63ZiDCx9
z3dZCclkef<Y|prX@G`=HW4pBEwEN|PV7zI|3S&go0Ap)pdauS{IO3%d>|i4)hHnh0
z#;j}-W;(n9Q;n9*px&mQr}*Kldy=4%necr55c>!ypoZ%;lD@uv@N8gAZ2><U46$v9
z`jRN*s!rv}ZhRa|VQ;FnI`?Emb!)12t=#jQkI~6l(zIXGIlr1gQWD8DxGdi0SJjyW
zY#%LgwoV$RjBP;}=9)>9t+$@bAIvadLrcQOc)kAd-QOovg|@JX1U(%GNZ^H6r$T;8
z-@D9S;4hWRy6a1nq}V|?HbYoychkc<STNIoTAIeV^$m|ux!KT<C=0ywDmg{~==i|q
zLhi4|H6I)upge9dHU=xqPG$b6e!}B1gQ&&=hfd`oa`a}PDma%Pr)3Y?>Nt1*VpYRP
z;U*k*twz=17G9^{_96$0njo`M2|TtvDIr?rtW5OTuoY=#r--`6SAq&`MS&0Ne)9|)
zhAy#COkH1{kn#?2;icr{6Ms#mork%qSO}btIrKO=ep@ce7lpA`2vbvoCY0hY-1v+)
z%uz@T3^l2PdGjw>#$^hn_}7)-JlWS{T5WrAvK~*Kae_-Z#1`=(_V#DusLXcNvcn$2
z5>}Wh9#tPp7$k{?EHLO%Ba}&4%Xh)fpJe(7mv+HAoFlOC=xxa*mCvAv5%U$;Ga?to
zzXwWxUp8zWqlPYFlpheG1b{A}`A%BRh7g2-KJu2|b$LY0WO_TIJ7K5^6Y{L!WwAkK
z2j2jA)q@Yy5(CkLfdrb?;PqkUdbtP8CiA`3(11mmWIPFFozY;FJwAFW*W_4QPqich
z4`ZkaZs<Jcc{wy}47J2|>l_G{VC*|i(gEt+VdE=$B*A_Ps?~=pX<fmn|E%NjR~pae
zG&imLd2UMpF_Sd>m>q@wa1A!(n;v)JL<6z2Aegj6?~RG;AeK_}M8dt{f)=}=ghhaZ
z;)1PS;g3w1brjhXlinf8ZRK;w@&kEB%HANy!g~{L=g9rpneE9&y-n2Zx9f?rrt}^9
z*QH9iOGX+4M!}>v`qY7<<TThNm!^y;5!O(PX*$+RRD~y1Ht#&w=h4|li5s6ARY1-s
zeck@zo@~1*Vo`I@YcYI7oTL3)jpKmf<7EDTJDmzmsB5E=In|Ua!zg!imV2_Am0qnK
zy$UT(oLnZu%$>&*hQvZS(rm)=22}56xt{NC8$_o>=kwTFr)N=IcxhJW;=PaYow3pt
zSecwk6M0rG_mtHe4Lf2}0#!ar@P~)ac3<5^os&(&s~hL<AZh(FYzJ=2ubRFctHBiZ
zY;9`M<rif04jpZWAbfEWVG?sglf%V)ckp33OsC1G`dS&GJU(YpB6XWGRIxCk-u{E&
zVJy*}bycOt3pmdqI?AXg5c~}8J7bK0O7rkzL68uuMBHHHcoHlf9h@OUs5CGFS?b76
zES25T`BjY&w=t9moJpD5cu)$%WFK_#PGl#_WTNGYfD+tlBzoa!YfIiLKIof`nF2!<
z-I00Qs4N-$W6lA!cy}>qKSG)Oxue(yjcxvj^|tOn!qy!Tco5PdK;b<g?S3{R7enM?
z8vg2JEdSJz?h}N+OphyYhtwk`&)qgH!Mm(YBX~o4m;bTj6Qi3EY)5vg=3L@<leb?t
zA`9Y@2>T-C3UB(F@O(GEK<qXn9cvD$OZScA%7EcnVsPBnk{gAQyIBDbKRGO{^R}3M
zBklUbvP<kfV*Ap@PYH9xN-hMZ-SdYA!ZnWAq>hfpP1jS>iMh5leJZ`Yc{_$=JqGl#
zH(w^X;Ts>IRn+|jxA~C|K2+++U{Bl8ZKkL82?X&mL$O>2y|#M;!`qFt>-{L~ZemTe
z*iW;%q|bRiKfXgNghnTD?`z}H2Vf#G42bWBYMHD!w*`}V;~$4bmg?+D6>51{2}uvl
z0qwijUbM6TK2iDy)j#eks7xg+nOylj_h(N8kvRo<u$P5LA4;5iaZ=uEa9RGK7eV48
z&>^%X-VX*~ID}QtaxX7f&yYRCz@D8pgvB!0MSoP8*$0L8qS#{>E^<Iq{<IN(BVVhI
zW^C;^W26qRrrXoC-PxdI&pz-9byqae=7~#CP~%e&77Q!hK7^@jjJvaOOx0je-N0b9
zC+EBy&OU`9hbRN7yx_Qml;#w*SYO#|{Z>`uO@P^gmVu3?XP(eB{qm|$uiulNjxbf_
zG!A2-8<1VL(DGQ`V+?SdM4KAFC|4<M-R6rwNGtL*>fuYa3Z@e_f!;yt#)9?UHn4d`
zHtqPMZCq2%WKWD_Hb?EW6|uz8$&4YhPK(c%*GYxhbFcP>do!rGA=PI*h=z=HjWS%8
zN&aTQY;Y>u0d@`ApOcfqmWz=ODM?oECN_~ih{8i9S8Vv?Y|aN?)}XC-Gt>jur*VF&
zPf6^d4T(g^xkR~~?(Fj}<kOZKM@+#nD_n3_8pb$qF{UHUYqZRdYxeChvWas9L#5+Y
zcvZ3jO1UrDYc|07Gjk<VnC04R?nJXlFRuJod|M_Mb$h}g-|8*;HJ}`FO_uqYw`dGU
zwl0dH>1iZZE|~Rhu{A;`)J|%$WkcEo_hXL-F^Gb24wEqwyKGWy6?#EsI^-pdBpP;V
zg;n5w!uYY2eC1~mA(1ug8U^|6mfMg~1g4m0Y~YX(;^-`=A5`jkBGBXa7%Am25onqq
zcBZI<wtHbE`*mj^ky7fLlff`Lhe`h!utX54Lu_#Cw1kM_)#cYt20t=}+7TyMMTc`g
zE%!vD{cbGRJs<1Ks}Ny;t^laB<eOlK$=29FWvCZhxu0&Gp~gD6lQ?b$A1RsQV7!eJ
zVuWT2M{<4qcu&1f4^`Z8eRJdWN|`{&F`<Fl-A)c0X-`KkTYtnvSnXr4y4eBMzFyQ@
zI>Gr{KKpJF;pwyy*L>i{@#c^u-&)h_gae+<ot#cV9Abfz9HkX4FZ=RAdGD2WcCL%J
zMZcBZK;T8Mz*?XbxQdQJe~$`<z9aG(@%X!n!aeVw(qYXp!B|r6-P)1Cc#T*6eP!hH
zYaI7Xi#JnIL#NhpF7wtv<ZwS3332D0eW$#ez-pABha3{cr@)T2g=>_RmTJPokz-C0
zVjg$^y|_LhT5y?06`CMOCTS7-=9=uyjT<9qF;vOpS`jgAc%feO*H?Iatv5i<ASQ?f
zR2D|79hIx1LOS!vi2w{0v75%*)4GoFRL#;jEgAcwBWZq&e(oG1)f}I#BsQo0LZvBJ
z<HTMRK}5Jr1W~r#iJqKuX?6!pU(VWicMI_t_c81n2VF=_zU~##Z9|?aSXNwOe6124
z0i=Lp8+nq?JkN%ZRAKOm>SIVQ1Q(DF*&#K7`RU(tq!LTSe?r?x4%6fCWe`7%R+9xc
z>N8DVzNFz?ny8kR2H?RCG%A6CK+=jtBl_{teUacXEAeqG=jP6j!IM>a>SE6f)kB(~
zIjA|O%2Ne3K523EnZ6lE#z<pFqKvfyvD%+C_BF1AyyIMx2%)#~Wf4_W8zC=Nye?Eq
zu#$%&(&4S#w|8*H%>|Qogdu==?zWhAq&@sg2ST&VX2PmUIRlOpE!VFC)5?K*C1c8S
zSlbB~TE0J_k4{4)WLUN{zWLP=h(!Xj#R2Z#aT^yqT&-uK#GTe<UNoa<7|+;Dd0{EL
zq3WFk=#SiTf&&^G4m$zujzR{B*vM2_j4Hr?cO;XGGkZm$)skNNH`&79z3qP`&;BQo
z4l#Z?5mmmAY@vece@GYpQuk+LV#TGSWBf=LvatM1-Ty<f<Nu$!KjVMa>98@hGqpDO
zWgcK?XK3(8#SZ#kIl(_A2^s&U*zsXF@Lx(CxF3=u|AVEyrJ|Je7A<W1zKRP>!qVq{
z9Dm+`=o5cY%!YFD`Xvg&0*nw3Qh$N>*QW?D<x=cm;&E%EwTFP_=i8RFgBq@d5VxI^
zpBc*%+6!V&?(0I%&Z_DfX3LP{bBPU@GUGBQ_10nEu@*dAel|!kak5@DkVz*^MV4iX
z4R|R^Wo5y%dN+EL2x+Q}3~kDP4=hQsu~{$O++pRxVhF0HLCuZlY2Z<@yzc8G*k1um
zE0A19XGqUIR02(l9F9~{xKp7!0Wgh;E-MS&d=ehdaZ2*pS>1Q7x@}-<m^AAzus*xs
zyvfYFz)S78$v-XK^IP?n+mm)ms^Cy9(&1FAR5gvN$Vz*4)mCSIHSNOKeXXsPoNB!&
z32ms`x6)qlTK8Bm>3>Wuzw*FZEO>DuZkA!g6tUWB^6&|ox?BPTxeNp#z{Hi6>cdX{
z%o}<^M*O)7C<{!K4PXW~1r+hyO251duI)F$77chac>vL*K!B3$31y$|=1;T&Zjzc0
zkUe@;`VKMIY!0K{qMZ8Bfx;;5kiDj(hJvVLqyS;G>3l78x~zp`fU2ltS;E7;HpM_$
zR=v&YIF-%~sFke`jpvnPHuZw&L6Br@Ep~^Oz<S!Irm*84ZG0zG6B8S;qy_BrkEBUf
za-t<E2pRfTd>&S$cWXw$`R5W%H=Ya{VIO~HVmhp)o6e7koou)Cx3UVtp;RTb2FV_V
z)9|vj(QtY#6Wp}P#yB$w)?Csb6h-7;(e9lCI_Pdc$lZg@Q9k1#1V9)5ZeYMrV7Ia0
zOR=N&HY#8r22dCMv(g)>?p#gARKD73Af0vKUWcj|5`*0|TrggTgb(?S7ZS$ZGz5Oy
zMrbeM{Xc1*Y7QiNcrert8W=())>|F<OGD>?kOgNKmYI*2pjB^jnOLbc+uC;~y-$~Y
zU;YmFfA8h-A&KLt=iqAdVc_wr3;1x*;McV`{KXpho1@2{oIh}X^9GUsg&sS5hhMH6
zxNN^9aCHA`gO2W_n@}+|a4@mQrDOUHku-|dN>--&)&_>S^uHkaCyf3o^1EQrKa%yh
zEFT1f|80ieQt1~)5!-&psFUa?C4Yl`fvu|L62;PXe!Z3$Ts}O^r}O!35u3$30Cb+H
zOyiU8l~)&48O2r0pM)MeC$s4-_Sz|8clP#>XJ=Ox>^PDfFZ2b>x2*1NdNiIY<v+X5
zYLz<7(l)zeRV?dy=8X(IWv2GpO}T)dtDae|jpebPTxz&jp0iC!FlQQ#7SxAXYDW_n
z_TlXzywcug98M7sqSEcvNtf7<p=oWv3sB%J*oT*xELXMTtEe1+`D<E=14n*%s(`w~
ze{FkXh1#n!bJmP5uy`^fkr8;`yx{_x-tS=cobIT79`8Bnd-C8sgA%00rc5eh=2#wo
zdaGwcWquu4O}M<XWt42G#+6c1Nt?D+Nb#EPNW<Or7@Dfl04XScPm?g|cVdx8kl3~N
z2q0GjDTbs<qyt~kQ|~e;DyD4=sH$Vl@$KhBhX_O11xE%o^&KJ<AfTQQ$PQA|aIS6<
z1|ITIRD4~%K{@5r&_NT?Cv%0QEQSW=BrNkW_J`OAcG~Eo)1BY1sa1#gE-yMCC)r2G
zjcV-G*2Q|i4Y|FA;Jp*y$aLX<jmw`N@D0>G7OaYp`UY&@=NPT(8kb?v7TW;2+_!>n
zGg^q>5ty^EK61P5a>e!l8r%hLEG>9LJGye}a!6s171)-L_R;29oUak;Rep+TuM14~
z0lK*G6G>hM9xLAWn81t!c))Zq!SpZbMYx?4un9X5T>8=;CZcvFewna5BANaUEv<^_
z4qKcF7tL~0`@#X-EM|N5x%inFNY5)pefz>>nHWN<aud1KYgjz@uJZks8H%oHBtNaM
z^ZMsGO&^(L*l%2?7&1w)9yt#e_QgoJW!%oL^H8{D)E+raPp|Vnc71>5JnY#QOX8M6
zyLTP-A6Fpu%&1ws9b2JUf1SEy-b8BWcAITJKXpIIP`jVl3^7-;_l$^Ig)3bd<l5fm
z0TCWIynqP`EOh-lp8sfF^RK7(e~0Hk3vK-c&c779{sd>nUnV%e!}-_7?`Zx9cK?Xx
zf8hIHwmS5`e0KidM{~b{yVae*v@~mx_0Outcx$5+s_Jo~J>(A(M)Bbi2{AU?euHeC
z;~bXG2BIVl{(?h%z4=m0C#kOb0l2?`qXJY?%yR|w9buKrt1bGd?(A6$uGHeqtaL`;
zTkoO!mx6Im%clZQ<NM5j#gN-amlcd)9nP=Rs+LX-nYXXz_OupH5v>&4r`smU4*Gnl
zB^Att+XeKm1)eN?!;i7~hRu+Y>hCN`6QQS8dBlms`w!qsb&&FCYLxnjP5lkuhGZ3V
ztsrzwZ3ThD{1}muNrsUqp{Ic2!~%sivV!<w>YL8=93#MEAB&2fS`TTKT${S+BM0QK
zP`}D!fxC!){XG5!r3>z~%SXSbV%*TU5t%(dx`HGnSl@$o?8U{$_G}Pspr7RJ2bs0`
z2G~BIKPv<v%pD<uj+n*)!r13|2AzFAqsVQ}ADC*untnZaVg5(pF4E>$gRa}n+Y?xb
z7x?KcaNV7_+WFgYg`>9MTcSEwn|BGmRv1tDdFG=&2t5~A@-o2EqFe$t{Okl!T$4nf
zSmT0O;aKJPyt47ghfv%G(jO+GcN9T+@Vp|q!44gqi&+o*T*)`>bMwX{z&)*IM)!q<
zxLL^Wn&m>qBjmYRLThreL^RttygpzWy3AR8#VnJ$z9$w|={j{~p5btCooCD~!+zyD
zRXrXj?U8Y_vM)yCkx};%SlSo+wd>w>DquX0+B1Xx>^dGSvrzJh)3EmVd$V@ObN;qv
z4~3iOO^N;b^8Ex?{aJP|!hGG{6FN>SfpT+{`@o<#WO&BN2F_}GVesG4O!r4wx&IxU
z|4iWfGn{{80{&+>)6snh)cua<e~g#^i08lBWYPa|?C@VtXEs{e|4&cnWp>!MW0k5P
zfWvMk9zuC)$j^BOpV@9cOI=)4ME31~1se)9z&zeZ)taYS#t4MZ+B9=;U<Nv8dU20=
zbb3cJTnmM2UpDc`Xr?o*Hw0~+RjfMiYhuc*>!0X*YjSxZ@f3RAqsnY`xSHS$Y~HXD
z@r17~H%HBvMaA)ZU=5`Ea&Ntny>T2n4c_;kJaV>u540tYd1g#slOGcq!Pnh_k(@~M
z-g3aa+g8c$5G6yX!hSaD?1dcE^V-DQQ!%3HU64Vc-$I)K&lloxC>%j~b!sI9Vdm+;
zk+Hp}?KlbBl^^xa@833cBMHn5dyFcMnx^dtedbBdPZ+%3yg=^e|Djh8@97~1Jbkj>
z*h;#G1LY#&>c%6OeW*@XYPCMNt-IWB%F!{PN46_Y$&#i7H~3b`bm}s4J;Kx|-CQHa
zqefGrRISL~KMX6Y4nXuR6hQq`6#yo{IST;iw=Q)?fEfiU>q~$+0A?UofK>FX8fA#j
z-_XB;U;>ng^eKix`k@D!yIhbeg;Lxk()ah*@C)~&p)*nXSA{lyOYXNxbi)9{Mn<&g
z=Y;}rTLAvBgWBjCVnhC-iHVH0WL9kWi4)+w)*n00eUGmei17=vIl3&(2hHXidbhFE
zXET4tImRq4g4zP8J+m9BX@`Y*h4R_k_}2G^>3&!>ikX;DAczejrhck^kDbu1*bAtB
zk1MFJ;v?>Oeg<NQk-}4QBMm~5>7`5q@fgY>oZXn>JEYw3VeFwwv`|~X`@GInix9G3
zDI(=W%p<E+7=mIU6>F79y}Rjl-6P35ciT4#<}qntuT{xx7-m$^Pnbp-npyhG7*qPH
zl<9Ga5mbCA)odjlMJe=2GmYanj9i4p9xb$@7NOR0z&@ijwBx6<QC!fdBKW0;UNe=+
zUcSyI=Wm&-_|@)S+ijR)4Q?1UR|<H#7yG3qKF5Yo-CL4rDR>DzmMMj(uJo&HY=}}=
z>u5Pog1_08C4hjcTz&tRg!vpw-C3#%It+6{mmWjFJtr?cLO0f5&Q@^Tw2A6*w9dIr
zG`opWQuE$E_I6MFyq>i9-ZA`k&-}dJdg85Z{0^G(ilcbvJ+b)SZSWpM|IFim@;+hk
z&O$ss72k+<m^fWNz~js**l3)~A*0|XcTk^`^vSx9c3|1hX#uAy?CpT4t)unfshaE!
zG%?{&`R^UhzgOt|?}iV5SpxV=hx2PO;ZGgT$5i5<!-tR8zxOu(RqhXSfnS}=ALjye
ze`hhv>KYr`f2<)`JATYdK29cKYG9A6{@bua^N)1kzmJIobU%y(t&PEHemlj-;>2Gr
z_OWX5uc3;powb3ZzM&m1sg$XNqTyFu8v~<{e^hkTjI`ABj328E0(OSFAHIjUq#xnm
zH%+ar1a%z@aY+T)>1pYi=xAw~=@=N87#JvM=}Bm5Nj}O+SsVOMMHKCHZEOq;aA|~f
zE$j_H>Z>U9m0H-*!a`5i%Ia5@6irO+aX)_lQ8lW6R1KGrnvRZ|7MJw*Dw6%-?QpR(
zGy<o^WuOJ8{pSxZGZPa76Rr{NKVr=E^dB_n-}d5K{VT@uZ(?kJk9{=tugB7{v3`6A
zzkdFCzK_?xW`KW+u`$y#eB23tj<L|weta5#j<GTPB7Ofk#!AQfw_`tC8vkBD2F8zj
z>rclr(|;^R{XO<?j%8u{Yde_f*_b|_m47-P3oSG2U$4tT`$0MXYdMyG6Jw>L|LZka
zS!wD1dJR@qdZxd%mx=z@Ap1{sVPj(aSbqCUjG2M$ukB@HVPO7io!M9!82?s|jfvr}
zbz%Eh;rl~f9PD&Id^PRBIXQ7@6inR=f8A>|^48W4xW8KVt3MF4GP1_~4P?JI1aZ|l
z8R%FT7z7#kS()hg_yzg+1ZY`>1z8061R435nD~VRxN-mIBAU1#lMQ<ZT|0;0?r}CY
PR$2ydA|fFfVetP0cskg;

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/Figure_DDM.png b/docs/2_39/stan-users-guide/img/Figure_DDM.png
new file mode 100644
index 0000000000000000000000000000000000000000..859185e8f65bfff0cc7dd3e07cfcbee6bc1c8155
GIT binary patch
literal 74686
zcmaI8cRbba|37X=CCUonWXsGbBOD=ngvj0wnc16z%(6$e6q3F7EPInZ%if#p?{!qK
zSG|7k&+m`ibZ|Yd@fi2V{c*pZ0rIk9*jGueA|WARONfgoAR(daA|atbFfqU<bM0Bk
zNJz+ZCc?t<rov*vmS&dLidMS%hC&9GdWH&OkMHyFa3dknhw1C+D2OxBx9Z-~(P`~s
zWVmYWpy2QST0zI7slFZZrM@er0h0VyLu2MP?#yQ-G&T|f*Q(}v%v9D@ug=EEmTTn-
z3PBO@Jqk#-EDQZTT^*fHq{65UsjJX6a)>%L18Jugk{%H<8K!|Ha>2meH9Q?I+yk=w
zxKIcTa)KJ#6@zBP0blzr<(H2RHTzftfh@fZPmcVAv{U2zE-md9ZGbxjQ&NFZ5?XBI
zE@b%B$oWZt7ZuwF1%gMr;rq7uudl!EynP$`?Z(qH9^5-guk~QKdYGQL@l*}<Gj)-0
z_&m<z&JQtgUHA=D)C(n_w{dq-y*&j`KiBfKrHLBK>)(!jO?(@38(EZ6(v;FV1ndYB
zx8ZXMBN-W_d*FLaBxFAmBsB0H*i0l6@DFTo3d-+K&~;N#e}9hxLHzJEfh8OX35p~k
zBBbPqyf%5|Exzbj{kDTqbpG6RFGwsp0d*i6@=bKq+gR3mH$K81Z$CPdhH*D#-@Xt3
zcHKY4i~bI#B;idotXtHJ*BduF_l7Ok?fCdeN7#z^_;&LaiPI~qDl3nNy$d&os^<&V
zDDly7p-9O8;|u&FrsMm5#zaa_Bos`@|M+qn*(=`__y789`m4A;Zzx$bss6`bB1S{g
zJ-YtCMn!z~6d!^~DymWW<bO^K84a^>hxmU_8W}B>9*X-Qu-x{||1oP^pLve|V@jAJ
zPdy=K_%D}9{?~qh&mfo`u>WI9xc;d~LL>Cw`?>!2W<j91(LMjilpwgGU@M<~Ni+Q)
z$AFj$1PkToPrhr;)FVqNXXhKNtA%yNGmlm%&ebD_y%!T@lKLQm6;@~{dhgk-Ow|FZ
z77Et>6u8*RFp-4h&KEwML%+Wtj%$c`Bu2UmVqkb9(T_6+soE13dJdRcp=$+A5MkOC
z^<Cau*xl(L{1dDL#dvME#4}{IOa>*tEK;$>t4NLgeXs#cGX-nmtw<Ami1kBVF_)x=
z52tb3Vae$0^`^)CPU|1|7L!)P0trObinqi+`z{x0WT_mAzq(49CK=$k=J-M_uC>9N
z(&fP4Y}}~L7n{;_=))DO7^Z@?W9=tnDu)~7%ah~!hwD+5c37?FX~NilpAm!+tVcC7
z9|!Sq%5zLKl<3!FrfnUp89mN}Lc_Vt{dvj5XcAR+14w2K-XB~J)z~eQXarAn+KNr0
zI}=S~>vOd7%+*6QtOLX_u5ZQRV^oHC{k>(L(2wAV<*`>Ty`kxG{bt!*%^{Q1>9%KV
zT3iBCe_fqeZe=@TM)F1htc>u{Rvezs!rQYAJ}Fvh%=Q><wL%vM=T#QXN7hXQTEyJF
zRP*hk4%d0!gy{dry$Wc636~~ac`t-8Hg4Z~6>aqS+$xlV$Lf@Wp8_WA<5hfb>`M-Q
z62;q-mMSu6$CnZ~Wuu>MR<&LV9YKB`*1@vdF7wdj#%L<>@cr97um4_$ygzWhYd7Pi
z?jrnPPFv^9x#PwQjJef3W(I97gCACT6Yf<H(~T0y9~rl`P^j-TlL|%DLO1POaYHn0
z!&JHjr>V49kw;HGP`L5_ItBV#Fmljlk^UUQ1{yJq<!wgtJyjn`dB@4|&qu^@>GP#j
zSiPgMtRV<3n=YMey&0A+Bf3#L4)ZFck!n5|!CUZddh>U{c|{GyEe%bnxPEC@Q|Je4
z1#6G;kckDS!jJbwpS;<T=t-KYc2M$xA?&6#OlY@%qM(H$Ss>`Ja<y8alCxoi(v?%=
zZ<nU$1lIEEN!P2(=(R(+wc8VY#^qj@txjisG*8awzg#qci$9$*|4ul%`rKJQ$(@&8
zz106}F0=Ktn^whvS{@t61J>m0V}H-JNiM}vhVGtw?d#i!eSuJ*B)gMDb|;<{+GUMs
zFCMln^+`1GB4wc&b~w*<^SKoo$%I#Ld8sf-%`L=x_h-jyH^0V`WHa7O`A=m0k_IfV
z+%<awAMAp>D|(XiJqRiW+6PI3uviW^dtR>}$J$Zpkn(-y<ZP`(YN{uW^YzG@)r^oF
z^%KLLIh}1ctJin_Iw~kc9l8Iefz;>APvMP(y-j6FZV%{l)B;@3?83ihD!LBf%neyg
z3v_51Gtu`5Qbd0C6YUllr|Nk$8-XYH`og6r=jQrf(I5Vu1)Np6pipxNFsv_dPeYG~
z9!tF`ZLwoH@nW>nTN1CX1`zm66t{(+Y~qY$5;1L$l)2Bo(Q=x*na5AXU2y07%xov~
z-)oNr(cXws_sZq*>s~cRI%p>KMwlcB2UN4oM<R8F$sY~BJ(jYl<JZa1%a@k)Z6V)n
z-ye2qT9wg=5IJMAn*P(ja0OAeql{VIldd@+{G0l9eWJrTf#=!xpf5M-SA;DGnyEO9
zsM|cwXN!n-Z*d5hySdbLCQ8+g_#8<!{AHWa2`rG}s>Ir{5suU(HnZJ=N=x+OaATQ6
zM=DCMG-G!UM3Q#PI-0hcL+ANnZBbc1X2`<{+)#cOrVOmjKWBv)K*ZA)t!Yl^zBIyg
zrqG^*+-N6IL0gMWrX0wilQ`Nz(<yRkW>kJ|z1xKx%R3lO!u3U~x|7cP%C5`TwdxVX
zp8vabU*dqpJa?gdeFYq_I1-$g!>d0_1^Qx`FqVT(;>8d(Qha>N`8PkIirsEcvvJze
zWHBw5E;?*dy%<uyv46)wd?(5Pj+K7D;sJO-s3*$mkZLNAtfHc*QIBjmAu}fB!n}UI
zLmtg%zbgh`={vOQrRT!Ls;yhdw6>3|I)kXpZ~mF9;f)}<tG^x7xQ3XR6<qM5cx@ux
zUA0hTE-TNqKLbZQvu9IGZ=o|q!$A|OoQ-QTK15|%w;mDAjP+$<IAZs2lQX`$HDJkU
zw@iw#V~BTs?>pq9d7>~eiv3%jU66>wH@TWMrH0|ByUE8E5>@u@PBU;%{?Bh;Rqf=q
z5&yAYAYR#`zr`Ze)OHp_MBN|B^?aLX^D9}X{bhM#wLco^$Z?7uQ0;$vUffz}ezvpQ
zLrIrNz0=Jnz(?P)sY*33CHVJfLg<){gi4{imx~W($ak$U^e~)a9n1<Mqn8eEkHV@a
z$72q`K6RG@&P<OZcqbAHU8^q8a$35Z7s(W0N6ugp{MU(!XaI2XMPeEEa-~YpI`ZiP
zeP{9!Vh&e(+2=;azQ-b!wYHfwDw+hLN4v}%jP`rR$;B3LoIWHjUuMOBkV|V2l3>)l
zXi(|pjGYJ_w;EE8;!e*FU9jpGR9QFa5*$;-jCu~*W<y*I4=uu`YEXA(ypbiOs8r0R
zxOh+z6ZmH^|JO(yx>U2`te1nSVs^$8Mam|k*xy45>5QAlOAg6vhBe@qXiCLk7`BFb
zGHK8sm^}~cdmphjTJXsGBYfcgpJodQeFKIog6lrK^!Em|VDgf=c1G0xocNX-=Qll>
z-x{5F3v7$zbtc9e^!V(zhtue>4$YVCHe*oVHc&P@7vVwJ)xXngO$L5&TJe$)F<$+B
zwDbMEl+%NFcd{v}uDJUrJGV8S@+Cyh6%Z}$v?cIJ^nVO%<v&E5-y1+JbmoGOe0=wx
z`Qg$+dA-<hA6y>mRhF&Qyy#++*_dc%{sU=gqXt3^PImq~JbG<shgaq+7QMm~n0D5R
z1S-y*-6=?kHJvXr`aiQPB!XbZ*I<iX&Ufdoj%u{+I!QiEOO4&bJv=ceQZ@fb=k*E$
zhY{L*AR|hlMPe(@qT^gnnX5t;VGqNnCELF&1dF+TFUH6Hbm9?ME3OmVVY+0lg@XC$
zWV!}ThpLA(j9gA|k}-&guI-tMlfie94z27CR{a(Xs{V(Ge>*&;0ih19nGxQ><;Qc1
zI@1DDah%&y3gNAR1QH6hkBS&~B~*<w-cV3PuI65A4JM-$_H~>~i*Z$WhtiYNOO8bJ
z_#XoS@X`kt4EOkvk;N@=HuQJke&ko<bfy&Ljxg&9Z#cp805(%`KjV@>AdB~6>h|Q?
zb*>luMS9n+OI`WftNej=2(x{>OuGQNG+mXF6&Ph*A9Gr>MHAR|d@B>{(Ur<$^YvrO
z#7i;g4GKcy2c3)|0xMDBG)}>yyPW0s{<O9)5x_Vco#Q61fluXqP~1+A=d<#dT!dJX
zZI|kam;<+-;xB=)sSZE^axA|(tSPYi!abZ!gAP|6q9prgc>?z#n#q{+xxhq1fGT_q
zXkG33d`$H9J4<(w)UCQH*WW#*3h7K(<|xwfIFb1PlFFl9X(&_2z5bUkU;b&Z&=l@#
zY<Q*Ig8a>h)BWOKpP6wyAEV8rTKh<1Wf(SO@cL0S3eIdsGSkuc^yz^O!E)bcs?FNf
z2<LsrT6Etu)IUcP=SDVaq@AEz^aTbBZ@ubz?j*m^RWz3#Ukz+RN`0=T#C$dE)wPJZ
zX!UA#H|OJ2&CR-=R87%yo!fs!AACMA5Mh;AvnOJpk(-zp%sk7~b!HRv2oR%`)y&j3
z;=~(9)>bx+G5;wGrO>IUH7&+@u4LF%A?<1pUt#;6@1Fy1->a7%ada!3AVO?CCJKeC
zPQU_o&Fl?U<dMf)EbUSWcF`))OcLSeXZ$<M8PPS*W3*}IjFiou3gG4bRogX*dB)~9
zEvnvE01JntqoE*6DJYy4n#Qz_l}IQQP!>_@uON%QWTSA!Emp)wGuYI+Z$kg%^_FBT
z`=bl`@PWSyHhdezGIy`EZXuSLddu;`P0Ia36M!|~LqRKd@%!wq;2gAnZ<4TXCMI{9
zk4h=FSdXc&i6u}M3Gwdu6AnV}Bdq?uqxkik!0P?EkQ33zt1GnaH}F-<SS0Q#=@1%Z
z^S^Jr9b%}u^F2uDs725#Jet;evqsBl201U1zG3kl+i3;HpZ1DRftXz~tVtX(ySF!g
z%;8BX=iI7Te*kvvt<$|%g18|q`W_grg;u)Z{G>xvAp#_kigNoM{b|bIaC{m6oE{R&
zeI{@Ox81p~BT6ZMQRKw4kxx<6Nyv;lEff+873@Xi`rAAN2X9Z^JJ56~T$tzul3VM}
z=c9B#SAXB%Iig>KHSGTpX`zq<G>|}DdVFpo@X&qct^MI>CRIvy+mA6wX3~P0jbs`M
z;bP>@t#n1OlMlMWVc$M`TOThYtJcJ66udg2E+_eOtzS}stLtxPve6=r)fc_ee!t`N
zD7p<@J&UmDj{c9Q1f#vdcM9fePrTUe&E(Hd6{V9#;w^Rp(^>z6o&)H?Iz?PpCOE+O
zp>l8@8?T+uIMEvJ6=In63wZaDWf?WcE@3T>z?MXXHD%evtw@fZ>m$pItpxE%A<60@
ze+Dxq1R#YsvLM_Ho{bJl1LGbee7@m~#JvbFkYRWoF6|?=NGC(4Xl^dz41m=Sble!E
ze&Epl$xV<O=;C6yemGI;h)9`wH{U*oMkek@njA%l>ye=XCyP+>OC~nE7xK9kPhg(|
z_g3b^p1yZg$Rl6I`u-!##`Mpyh)Y>7*LU}}(&hMtM2_1{DQGB^4pVJ0^o1(l#qCk}
zU7Xj^Cm&fKP6|1l?WZKzhHC={02n!f@c|#l!9SI}Dwc-hw^!_j?UC6x3@dsTcO|?r
zus?_f5%yO*+3c>2m1KSW_>`*SU`+%vhLV|?xxg_|YObECT4YqrO2l!-f%m|O932&r
z<MdPzL`AURl;y>&j6q-OtpYyzC(2A{k6}#*OJM?xrMkjKO@u*yEs$TElF|x_niYnl
z<&0#5i}N#fhYgL<BAA@-{SuRb=5q{nylv+i>%FC?Jl1pm*_u_#(X84m(NRh7>y$DT
z*d4cYoR;4CeXy9UB<_Zx!i}$HtCdJ%+_*!3jff@k+h@Oy2sr`bu8t<9X*6B^m6vR{
z5jOPzUCW&Z;&I_3Kjqv(r=r~@Ju>1d*<`A81xL-ghhNBPe=vobF6hyD`SAH`r*#TK
z%x1zox;EqAGp0wkUmLk=xAHZtxR{gUwJ$bl`)1vRMc!=a(X>*P{d%}%-Pw>9=RV<R
zjZ2w{cBTRipUhYKiXY#Df_5I1+uylnJmayu&{e3@fO53-KE&Yk$Mymds&mJ&s_SxE
z%=aG~98)BM60kj;OGE>ahF)awSR2THQYNTaE9|^@d(E_)7Nz1Ay-K&fN3QFEm!?3b
zruy8$xOabr4K$irDmgJMOV-%uchsdWhwiTors*{W=6bo8@MR%oKlu3K!rQ$!MabyD
zOj0tC2evs?W2F4OGnQjg;3EW=0q%ZwJc`E>zHa%9oRvt!3jUgmPnnZID3#DFKj3(-
z!V<BIF(`@|Y6mJYBDpm~jA>GMK3z1VUDgb}Iq@oltg_)%8$0d5Z4pt@kK)G21vl00
zc&W#6m(&Bm++Np=$B;7&h&=6rS6ybC-^|a32--Uwow--+^-FQ2aaobd3qg`m;&%7L
zrC7rFHp+zsE>3OwMo0La_pDqxx)uk`?_41uTnO~`YrQNS`|uq>P(CAg8AfV(x=P{D
zS559lmjtygY|;AAL5GUrxTfP#-z`HJDJOu{9&dU4L6J_tA>c4zCivH}0?TGrFU5Cb
z)~X)ScrjjTzB7Gs9#f3g=PQ?gPT@EMzq2{g_B28NJvVQw+=G{UY!w8EZRkgdf1tVp
zGXCE3Y}oOBEIO%)_HhS&Z*JmnH$jxa&gWcJs{S-7s6{1%4l8d%_Q-1mvtOHlgn~|F
zoRYQS8%fMz<YV?!6bUYV4OEuJ^k<0QcuBoAq%KrYmq;z_w^J9y?VNqzjZs2^yy8$R
zS>kDLO;roiL{#|qm>z*^&3om<cYX&^`3V4?`Qvn3<DxyikWl*YW-?ZmM*9&#P(B;o
zOQ=JOfs1~BB_I|{J=@Z^frt-IEA!|;<$L&CqJyiw`}2a`Nc6c0o3la-vLd~p$<GG$
zzhVw<1PY@{ZYNvrP>womq`S*E{r>X;y;p1FWvmaoI7<3nHMpOwKSUV4i?nyc>QIhN
zS<t8O=xC!2nU>q8G`B3!-RujD6z((&<CbziqG|RA7K#tAe>nZTP}jSaYuIr;A^!s<
zy_Z-$jX9h#Z*T0{XPPzw6-NMBr2u!>fhU)5MMQA}5q}ZkGzEeZcVYKt!Q)uFZ*cVr
z4*8nd!VdxWQ%7QMtCYmt?AouNo&iLl@ZrVaC+me+So>;T9m_Wkv(J3-Y`R{yt3xs`
zhO!x=7*$Fy+=d$y^4$1$TB*`0WTLE=dhgn-z)`ONY-AB8<c-m$3QMG^9Es36iBy!m
z-!h!5X_`^wtie*f^*B?#EEjMlEllUO!<r6l6dtEH-H+~+4FY$xEE}d-q9(Yf#CDjj
z|FvM{yP+U3;nKxvkBhUNvbwXbr22lbu`)|mY|;mjc(U9L=7KN3i`bmbMaV@55->`>
zfAVIou!AKculBgzJVKEs&q8(oSuZT+S@K$kw`U?+3^-hHGEiFhHCE}O>M14D&%Qe8
z`fJb>dbD&ooabHf8DomNiaC*yN~SAcp--=SoBo3X?$gM>e}aAXDb#7S;j?~lSCZWe
zE{ll~{tS%@>!EDbqF|>aGW?R<stw-5>W#@N^<>wbfc<U23EFS<O?osF!=BsEa4LET
zon+agVq9liUxQ?ES<i{DworH!uQfGf;|TnyuiR;8oD|q|w@pwhH5(1rt+UCq4;>)k
zFnay0*461`kuWc9wof4Lq2)%YoTOx-MeUIhBNrFffMcUW_5P50w3_h+8J~UenrHrg
zS`E}!Yqol~yXX;l2#?Vf@0`zmIDW|%qj0su%I@3i(A@pSWDm6%K;nX*X6KMvls6Qj
zRt~*vAzLh^5O8%&KTjY#va{(PKJ*f(>aD%1wmDjq;WBM+Rg*GFkhPok{7rPj<>ufA
z)|Z+KWr+kRWyvKMKdvAr5`vV0lqa+rKzFJf%f8$&0!JOlenWaBDr9kH{F_|grVAa1
zWMkaWcKn9kkyNsR(XgiaaY#<^%vKs7`}-&d_v9|Q%UXf-;%gGFcK)d}sTk>&X~$V?
zn`)3lhZ?-EaF7j?<yG$lj{D*H3_|9D$@sG8J7Qp%2fT1u7v-E1=lzxIXD0HaNewFp
zgYU&~@ccV_=?(N-L+@IQXCyhd_nZ0>k=5b(4?@=6UG_(MQ)e|xUPSF5)?G}lOL^xL
zX}cCYD>RT;ZKF#rUQ@cZ!7w(2b^HQXwXEnK*q7tLFGYwid_p7Q^!-06w+YNgXT{U!
ze!`#jM`4|uujBnlU=IO86q16Ot}Yg9R4AQLl(@V23NLstwTznE@cl+6z#w}eW|<0U
zoo0%OFF&y+z8tNvfo0jRk69o+^^P$gD=y$HcbIDplRh&n5r)0s6_VpyheiuY^2@;*
zc-PnRreRAf#RIY&E$2r)9!4V+weR!?1QH--OWP&`=@n<D@l!P}>Tu|gU=fo>d2c`D
zBS#W&ZGmFOvfLu8#0SG#E;&_#S?xtV0;{Dp-vb#7U+9fiC{I3+t)qJK8jv^uW=xhk
zOgX`#&9ANy2EbzG>vt+R!o2h6d`d2!$X|vR_N&#qvhtsv;qhF!S~nO)>M|;$tqktA
zE+*5peUP|JxJfQv*8LLhE_}Bcg$?ffv-)?}sOji}7Ub=;4w%Fv-VY|BXHEOqx9DFn
zmI;2wXS}-6k*k++4VUl761n^N(PC55%@qqj<&By{Er;D$SklW?wLbLfV(DZ7&9(Z>
zOxS+Q(PDD)phZ!h?Q;Ld{p21x3iluHYTQMK-Z5($ucgOZSdWhj!D8?0`&Q);xcKHw
z*pHiU2)GbtsrP(hw_TEFP%q8q*_(3NY>Q-2Wak>tD10YsGctxe)T?Q6_8~&ndbJX&
zs+<O~2y?bqeG}y7MAUj55qId$eM$E~WYGwjFx&gn^ymT(CELPPFJ|w_T11Weqr#_Z
zYJf$BpPo!>jaIA9CD@=?PdT}(R_x?auz#O<P=WqCP(xHuDq9DKwLKEQae17(0&B=X
z0E(5tEZA|yVv@_3?!>sPS6OTL$a)@crV=uOBoZCJ2aE<csWM5CGCj|S2=gtD*?&Uf
zIaw<*>`vez4(21RG>+kaue_#_*PY8W&Gq9`kkPynXFUmtvHN<-FgvHmb`#N?HydAH
z6@%Cl`_1u(F9zK1C0j=I@k*8~*SG`rJs1)F5PD@oDttFvQfPv5f31*&=wt`**bg?1
z65t@RekrMU*-X2VXZy_gDoT?O)I-&(hhM<OUNpSsd$!n`Wo@i#!DnA|gJ}VyW<oz$
z>+Rn-q7sGIbT2iCCA`9BF|funvSOt94XF&YPmB9rOVGvXZqLScZ*2TVcc>)S$*OI)
zbg$dVK|rEHJ^xsS97UOc-HK8{osng8s!+QhxwE+Kbl-cirtaw%D+j4+7AL35Ae+$~
z=oxM-1Ju;Pc1;EKtL!oW=!#w2ligQV$;*f~!u5|zUJSOIQ)ppfcaJ$4f5wSl_bL}%
z88zf#vtLm4OwQkxZpK2Sj?=3jMj0Lm9@Ep)B{TI`auw;!9!j(<CJyGVhuzl?HyC0{
z1|(P5%UQ4R@LQo_RHz4Q*hbCGi;9c6zmhP76@?_yb+1osQc*cqgCW`dIJ1gOtFIVY
zDw^q~LkNZ7RA>6&XzXas8hX~m0Waz>1J0u@PoM6{!L#u)%NUT`B+1=>=oZ*?#f^Tv
z0REn!NVzjr{!+h;g6~SM*;sMgWa(~mr5&B)+D2>otuYp6W*GoKVq`k@Cf^PBWa=7M
z`uEx&*#*rNT_eNUKfQ*3I(+r0?g8vFntyrg*$tT}k=ixg;j6su!g-wz2ZfB!+3>(D
z#PvY6x4VuYQr!Zpy@U12d}fBLeUiSwxKqjphF?hP%RA`O@bdnFz{wg1F}s0yw8JOl
zj^5a$IN1J92WwBE`48EyL=qCuuX#JB_q-5MxSCVW+;P-wF@}WGI25b1_Tv2Ho8NAU
zTO$^o0k_<y1Zt~#nFSY*Zld?QBcnt&)f&^O$=*U(Pf~A}^nUMXZtRT(Ju&x5%A+;w
z`_gf9j{)(^5^9Xd^VEU$C!1H+T}3Iqui$-@jD(Tr9cBiJYB*>1bmOc>!IH4rBjhmB
zET=tfR|z9;QdRfY@+WUQx*c5dEC_09jwyu!^|2Ia&-R|F<*Ajt@YSqr{YvIA5vHrH
z^?=j!XvQ0-J*RBy1xg`FWa2~1Bx5xs8SR-!L@NJ4Q@#*qnxJ{)KAl!1{aDgmb8wJ_
zMlup}l#$r0i};3*`I6aKuR@_gTW{CLue-B^%P^uSQmK5|BP{9$a_+Jq@O%The9L6R
zi{{b7i-Oq#m3TS=A$!?1RxOmaHu8!K_vF)am+%QIZx#2`zHmi3>{d@?ch!<WeDtPO
z6Lz&<p~T+@>X<tyc>Y7U5jCjW&h~++aqk1pmaGBVvP}~C!I@8ln)0g?Ha$@-<Sx>+
zCmVbbeD;k#Sj2>9fnTMc%T&xnl5oB7=ipHGj+}3gvbiB)OoHwJBaG^`C-LMT($iA4
z4a?BHG^UOLPm_ckz!I97a3gk)I=X%_S^Yq;_APq+b2<K(8>L1vA=L+QJ>4eJ)Mv*|
z0JJfg^fj^y`(YUD8hI52>`KhY{)z}tRKIscP%ffn5b=s$^=ahDs^;s!fy451aSeXp
z!R)g6WP+mw3UH>LfxNX<6iu+e_napEUyA#P3*&~1A)-AOfC!vm;k8|w87cwQ1d#(@
z*<pH-Lt-lOh;fe(v%mU>q(qy#Y9eW1F%75Oh^B}?DI7=ex@4w;US3=~(@43~#|d#-
zcqG1NKBL=qu52PF8)>mk)REzjEzd-pSRFH3Hv3;>FjV6zD6VNfyXB~r-1}UL@pRgG
zm3g$vAy30)y=0?ce|0#14|yZl?4HB4TUjE>ih^pDea`q})I!0ty$G|3^1ewTt<2{I
zsqx7}n}>cB-T65!-VHjwZr?~j>eBUH)pXr%B98N!8ZR{LFj($SBM$bnTzajWSg{(p
zb)Ik*=E~}?zH~+~H!8a0eyf5N)SRb#6%N1zW;{^FOI6v;;|x7$TB!&aZ~4r7-@HJ{
zNL+~8tdk25z;g=^r?gKRN#Qh}iMc|KIU;C=`})m=9FhZ~)Ih{LOrP!t*g1=~e&2*f
zwr1XUUmZl6!Vx@$@Vguo+pHeiRaF6(wYl!jqFt+@<up%VGL#+5X*SZ!FULs2=?20a
z$J(`NZmSv3(I~wOs;<S|R}}VT`&uo*WUCf_D*gTabJ`cjF;(iD-`W}TY|Bcl`$UP_
z3}ZAGHeB`xRVJO4qd+w8cqi=F`R-tSyhGDrGAdK`2M6Z#4gR|s`tR0vRtei+M7&8}
zUxiT@D{3yhG+3j%X)sBp<h%<Irz1+ewVVM^w*V#Tm7Tg*l+I`xrT%Egd)s3i?BR0*
za&8J`xg!_yayZQbJL2_Pcf54WZt@edAp8&U=N0%PMX-?AoE6ULd|mH5rWl!}>t8b?
zYyzE<`5Y{K*8%eqzYZ$#sp{42k>mZ<_{wfHGBzF0C9~13)FVSsW1011R-YfQe)H=p
zJ=h0e(!gnLror1n=VwHeS(3&61d712cU97wsA*@%`$fY@;>q9Cvy_>Mxn3xx+gH9_
z<=d>Lnyhko*zzR!4g=VHEAXlC_@wQC%))bE4Wg)p-;V6|Qv|$wZ`CGF7Oan#`H{CP
z_2j8jCu^n)&A;^ProJKLc9Z<HN8h0-x1uGOcNgzTfs5I7MDN`{8#<ipu3Fq0zCIoy
zLd2d53H3*5NTH!{GYG)Yu{-mXj1sY^n`A7TTqs#Rtr7pygLe2+CIMx-S`D}d+MY}n
z+%%47Wr&jH?HdxVTZw6^pkCpbea*2JK#0n5XBZn>77##vO#W%(al^LX8ebXd0%poy
zB1<_(1oh3BENYZmsrk6kPHA_t;Ivtp$-tB2!%a)0_7`kB0H;Q0B)hBKpb%i{;Xh0u
z&F6fxE_hGrBcz*2Wqmi8ZzFWKn@^5YE(^mpLnfZ~aKa{}CS_&Q$jsUD8;9;<XDn4l
zqJ7*z8mKwN>^CMDKKo*GTyrx?aXj8zUO1KwR!PJ@ZYH%fh>Kda*HhtKe;h*2zm^<8
z7&{r4e2`$%eLu)zdH4rc@5x5hW}t7cTA778O5qE9HBNMIlCHP+g*k7p<7-}*Zb=F9
ztU2B3DAa4Z>IATqA+YeBu7}gDz<d+at=kzgMluUYr1G?C=8jkM9-dHrQ|^T=-e3zM
z&@?mu{v}`slw)zPdiIXYCo8#xS$Wmf1f^EqCs$+vos26jJ7-`FHg==JC5E`AqBGl^
zGQa&n%^M?)h6TkT80zmiO;HTm;Ll~^)5q#Q%9h+y5-&MFlB+Ax7i0TC@a#9ROfg2w
zE&qPC<A$?ZTU*;fo#MUc&$5)cHtq9!zOK(?8D1%3;`1~?l=Y)G`fsGGb;Um%4scp9
zywbpQ!gvx*Hd4Iysw%rJf)2(zd>&Y5c!i3YZ_v{v$);8-4;*OpLHx5!rOa@{p6rsf
z>ilRj*AfqmZG)Bf(y{DA`XT&Aqe7$EmHBJLl^Z;~Pxz3`_ctcTCl_Cfo>+_(XBzgP
zbQLCN+ppBroE?luli_!MO+GNa;`k<`wByqVveK^fp3@zNiI>d<pfXYhfGrz@!?=?Y
z*pk{owmw*s`f<WuY-U;Q+Gu4oNz}b(?S-gcH)tror6+>WSmE8K-*o;8UXk1$#Tf7M
z45vn=#v>zFc`!@m;bHDchIDLrOBR0eLMJF`XrF?OhKmq_4CBc^&T7o6JQs0|1wtl7
z=|Nb2_!<T7Nkidb)}|kV{zWT8s6Fy##RhL4|3IT@28)i7*%Ri{Usx}N`>OGZSC7Dn
zG_6dWfsyeToR?$37SM=*cU&~HY`ftSJDS9sr3edLS(8b0%PJx+Vgw+Sqe$GY^b=ts
z^CxYw>0*<Jui2^t-2=&bMp;Jt1RgeCXJSjz&-pM{(?xpo%2`zlzjU&cc;_wf@_KU$
zif+m%Vj5ehlj#FOx>LGq>rv&=GN0h0NF(!J<&l&9yG6yGIz*#){3R_2akxY%nw38c
zx4-Wxp#e(rd0>!VyNrku^x_?_MFLHTAbIg_i<;hr%d}Q~1C_({dQn|r%XF;2-T?70
z$Oj4ahT02_+>!l;i*lc;J~HF26aIBH5Lf73;Xrm5%hge6bKhz0ln*W9>IIfn(|1S;
zPX|k7E8{gn>d~Wm*Zz_$u_>3H@CdnsaSw-6`}0Dbm*v=(OxLgE3Su}ADwbkyS8#%7
z%6~;k_8>2g=m^rV#=796UjeO^AU|lHmR2k;m&8KPVH@bBtu^5@xikcMo)MIA5_8ZR
zd(g-#`XkO1!Git#0)LHi`Zcl7n;Copd56{<Tq2#+F;x?b9xsoR6wjOwt+LnR(>4pQ
z(X$E%GO}TJB4T!G$o}fde3%%@L3=rx{<YWA|Fm48Z|iH_PA2%{>JGNPecpILfTpME
znUG&_?sJ<0vCA~@L_X%sXC0QT&GQ8j+jOGK_iTfpdYEhDu(pp+g`AG&Q-GjKq40x$
zP136g$M<r)_HzzX0ox2SEWe8+5)`rEJ7GP77a9kf)2x;?`_YM%FW@e*QY;lB-8_qT
zGL>`U5<Je2Hr(%Ff{qZ7w7A}TjfFS}G<ez6)%KEMj5%){*8@snw1BJd1Sx7yC`6|X
zXzh0TrG$=l<)n;y2wm*7!rlqH*~<^<=iTni-bMRy+bZZ+osKI62jr4?R~*+m=ZjH~
zfsZd|n-3mFtmOe`@z(_+vHPc1w-J3QTp46d`K577=FtKwiF?Y_Z1^&!$FkD!Y4>X&
z#Xo0vw(U)-xT{cG-8$2tg>ali;pjS>fTPTr+}~>zM5#O73Hov7E+H#hTK7^(!S>r%
z9X5Kg5OOXCEZZtj{v9pH+2H=EO#_WqZpYrjy29BQi^)gqPVAYOdW&sr;g0y+wyohf
zg2i+D!`?dsa?2dtw)Sg)I_+ZG1JaV#=2@G?q~AsXb_<hW<|N}C$yYEHINHs(k46hU
zOqpmY%)BPp%L1TbSdq-D_lvZNn8d>otzgg`E;M@6%3OvvUp{-vuH&gqD|<aKo&B%A
z8r1bF8{PwfWJZTq09DM<vl;knPkBs-W0Qt|kR>zvHt*jggq8|PInoonm<V^6!fFRX
z%QHLP0-MBTJuNNJyRPmpW%giwUoRW;>o5=i8$T8tGUfZlMg&gl!(1|}BH=pi&wnSj
z&;u%kzDMY}f%V!btHbICtfF4*M$+=pta8&JoH#?3y$ieIZib#o1^wAaY*GYJc-z(D
zH9`Q`T2_BJeI?E1(DTubzsmXMh3nCVAKpzUWC!IMA=<n*4%5+(*%XlRr#!NhDbG$^
z=EJ3``->TnM_aQ7sDKq9W{CLG2ZS4>Zu+^1|9^?B*K|Oiiv=1+<EP=pFGQhq>Et)O
zg#bzgfMeRj5@^ZZn_TlMw1FMNBjgAxe(2?CIKER+E1lb9r9gC|emihr5Lo;7oe#I^
z5C(S_9vR8W<NDm-?D*Erl`sq7(Qzp5b%s05?gz0n+O9Vv*T)C)hVtKuPuFTK5BE@M
zRytBM1f-a8CZoFj)6xU61s9kxdf%)Dcv8w+tZX&IQEk{kNAEiXuvNH;U)0iKbu*u6
zt8mVbf{FNHclh?KccHyGuVk7eB`kS8%%*2wa&wA2Q!CMviGt{l>FP;<t!3iJZblTQ
zBC0b-i>N^KC3JLbnw(%0{IFf+JrEcBm)3{VL`%QLjm)bpG?;qSaic%baqWj`l|6aI
z3`*F5+@WMkaA)|N1Dh0&KgJFQ64-<1kMAI~Sl`7{R9Ndyf4rxuvZs--X6KmyzSL*U
zhzVpT(XP91Xd-CQ=YeCIOikz;bw#tX9_YD}FOMo}Yw>K0V1L~?Ob)r!j{Z8jR|qj1
zTS)~%5CkM2px38R@=4oye`lm<Aw1BE8le<TWlk}<=6<e<S{Utmm`GCuQwQw}_raCH
z0q?@eN~ccg(Fj>edeZepp^JZQ0`s@BoZ8=WA*6c7%k%Y}<*0ZIO5kRX>@Dz+sjQYl
zX@I^OG?da=chZo@pb)YQpx~5E>?b;0nHbeqAdY0|l2#Sq{bNuO1+dA~T+4_apuaAh
z)B>}W)Dq3GD~{Tv*Aq$Vy5K!uU+7o#gH1W@8Pn$1doycZ=+Ph2)3r;199&U$`>jS{
zM$$^y!8(8c2dpB)_viBEzcvCq1ge4PGp)9CbRil|9#{!hkK+=rC>tDE^EEz3v~eG!
zffl<6YC#-_nj%qWtbSXmxd+!rC5aC}DeY8ZGJDgF%R*Y`fc5T+Kc+m#1lTheMKFST
zd3iVAEY^^<c{XO;^03{eCoB-cLVoWd92$u}lv89dRlPTSdbG@AgmdS|YU+G@K(X0!
zm~zge0~cq&Nj?H<B98S7h15TkBhX4D#-c3=gwNc9&cs`KEwH#8(sG9*g~Y3iKqyi6
zS6xK3&rUKNGJ@Q4G#G#o!WnAcM=7<(Wf~pz&Dm}>@Jxy##MpgQt>jUEb%nTOV2jr6
z(2H{ePrZM2Z6I~N#fp~?iH`@pI1vKw6SF$wjTU~Fa36GsTl<RjZdABraAtKT)#{OO
z8mWG5shB{ZDKlIQCpbDksmoO@`6ND&PJTjTAFL_^n+bdJ=8L=22Wvx=Vh@R@!oR&8
z*Ix`c=2Q1i9WksifEAs*;occ)w>l(IWYqP+a=I2l+9Jy5QI&LQ;>@S-pG1m1J_G`W
zk0^1g1=fajwyO(ambKrm;RYP@%+%hbmqzX=oC^_TQZ1F<l~-RV(R($*VZuCG$+aua
zxoGB%aV(kare0SoUgJ#8^9?OzsZ-6U&**_Z&Oc!dx_t{owWh^=z?`&AI!rUeq4nqY
zv}~3KGD1!Pya&1fd_?)vxte@&nh{wP@hL~ECJt(bpNy_a721k2?M@n5)N`YiqsMW+
zo3HIkd}}iL1{wT0F1J;E__PPniQ(wQxr9fKDpes6JCHu`|8W*8mKk)F+@reokwQlB
z2x|yJU_J2?>>Ta7P!2$7@qxr8vHK(5|ClW9HJ~tmzl9GvoA(|9nyC74$|e0!_p@&s
zp4_9mG?E|k#05YDs$xD`on6D}{Pd_lSF`H+$q2$;DDPSgYm#_kkJX_`vOAGb_0hRq
z#&TRtG^Ww=r=3h=g6FuZbprfsRl43Aa)PBU4hJfwbsE%Mrh#Wc<{@XB_re4(@H4e0
zsRlo#^`y&q5M?Sf>DPTZkM<n75TLS(uc7}b<+}{&ln52SLioih-R<j=JxTnApfoe;
z0mTy&2xivlS++~PLY!NlL}85^r5>qPUd6E_MF<n5_k?ytn;`T{=`zc8nD?M634vh!
z_!XY%&jG={7DEG+=x360FZ1oPXvRweoXn>!Vl|2eGGH2<Oj!*v%<<%}2O}|~?g35Q
z)1A5)v3p9mhtF!O#oA+rJ<51kLo!#7j{A}Fhozto8TAkTX+mTAhu&dpO!j&>?K*&6
z|DZ%A$?<qOqo^8DI!-{$VpudW>y0wM$buPu^u(ZvX1Ynt?o$sHV&<U$dJkbJ5s}%5
z{MvBAMOAtx4p2FrM4ji>g^rbU6nY#nFO<MSSf0}_lsH&AzwmK85xGGz<ic-{^ibo2
zQs$mcHip4?@%CHd8bAWN2#MThmB~smx2*y{I-`+F-yAqoyI7qsxN!T(VK;EN>=J&)
z>Htd~x^Q-qo`k^?`rM8;c2y5LcIzsO<WQTIS<Pp^3QzMV=~wwqKOFxzGPMvwFkQ^v
zYm|-Dx>ZfMY}Qzu0BXJUT&(mgZ;h!P9Lx8O^{JXFb=K^5z)tAr8B~|&v$WWOyy=g&
zmsKsG-J3+b5XHAKb7hs%`ML%nShqaXJ8Cd1qpr~}!{nxdBFz9JGqanvn%eQpF72$7
zgqWh7&4aOQ+8<eVZlh9YZyipg6kjQD?t5XR?5cATu71%Mo)%smD=Q*X8+PN>|1s0>
zx<$C<mb2DU#XIu^XGe9IhwLUvw(?;rJ;_&=?8St^Gn0I6#h3Wc`iSRF4v7W_2L|4(
zO(h;`r;5DzGWf}<bNWntKW5mS6TKst?5OU7qkBv8?D#bPe%?kgBAFK6DhIufdB9WJ
z<Aep*r_npkBPgofdgTJVH|C{vj4F~DTs~2pC)z7f>(mB*H^|t56gY_)pycbO1e0As
zES2J*k3lpU)O&f;=O52RKp})XVbO1aPF2S%BCVw<9u8g2K(A*9<8>QnTc2?zWxM$g
zCzw$4$DlS1yvQRv_SOZd8lkIz39)XGMhBT*PN(zj{T<L7=~(J}zb;xK`?`*FF8DV>
zf};`kCNryKKfi0U*c}z4<;);=OEE*1Yc+OBg7@iCc2Re@fy=%9F9CSFUvb`C!fy7f
z#xMFq1di!vWnF?GX0+j_>u(=96&_K<*c|XICd?{v)X2p@=$afZF;(GnQ0e3MePBLT
zuKx}|nr*t7(?ePsnx<~K*6G6^E{zj6e}ziwLeKL;O^@ZNOvQ{x7wU3Eq2v$D-X>Cv
zG3vJjmkfxnJ#^c$Y!*2ec?mTBxLEL&xZQg}dN8I`k;gg+p0l!zp(hIuzYk9qFs0WQ
z3lMFlf1oA!4r+mRBO{O6coDhVg;FO`sLFa?(k50;;OI;B!>>^|441$B^d<<_ca^&?
zRw66yRwV&Po6vQA6BH9spp{&4^4L3{q_wN(;?ZRH$3MpgkD`9<8L=Si5j;_8XPUI(
zGLSBly1cvx8<mKljkUpQTnGqlMoc0J=-d_p7*CqdL+4f0LHo>HxNC40O|2u6VRpfL
zveZ0Y^*%Fao>^(4)juiiCzOSJLw^iw1RXk#Ve)9hwrFN`1Hg_I8uwB10Lt|G9d3C<
zUu0Fl_>ztdeF)N^({6j|J{DcV3^`!W;?!~Sq~EeNOb!<4i!nS+^%(_JW1i@R*a(oJ
z9daY|Hh(>_5YNghpY)*xzT151+C+t-f}$eOa)Ej(b^zo0ZOCM$U57bQ@nzy|eGl!+
z!HjBwH#_eVGOLvf+wQEC^PQ_4ZF*e9>o&D_C-POvG>1^IKKrB;c7u$!{N%s82MJB|
z?A95)g^4DjI6Yr#d=G{B8X*&|OhyJmTBVR7OQyJpgP0+_73g>$JWdDfXT)qUU-w?R
zlU+h2s_Y>UhZO8B544<?_Khyiy*WEOoCZli%Xv91JW@nsV*48z$O3<?5Ml}kymafQ
zXIT^8PFRh+NBUeDYbTp^mdCviCg3%oFmMK0k;*_F^kJR8y(0Rux*rE!eWs!blB`)M
zNZ$bL3mXE0Mxn;p?r5#3X9Ajqp~Z#d(nO&0)(udHLA}OH82@=_61r_om=2q<JI3Hi
zrxfmxgg)+yVukg5uguT;Z#O|OB^8oSZ2NT0bour3#3QvlE&k=c_kPve&7|xGt+)}E
zi{FD-qr<4gB(1Frr4zWH#&McZ+AMV5H@R9Y`?nAtaQe?tWN(a&m6*n-gEjhk*eLQ@
z)n+<nuUcQ;TY`9?xE}I*Sd1p6wRN5xtTO`P9EUb`-$hu7=`hRZi{9M2bLRo0gSFAr
zI`|skuYG4n?YsY860jLFc#X>;i601H@uC$^-jL8#4*8!aWKK7BX8QE_<N%T{mOgTT
z=gy_A(7y%cnAee^Q)##L;4>N<HB|-gA|fqE-RQ-lzwf<)!%jd0L_X|BvQ0M+E%wy;
z*@>ujrQjpXlMdE8abVID8wrrLl3^{=+5#tV?d|aPr4hrIFhheOyt#Xb1plX(0__AK
z9u3E0lF8l)jr8u~kUzRzX|l2j#0b7_vbYBe@s<x(W4ME)Va8uB6~a&vaNXodWJLF$
zhJ$dDR;`(;&03D-Rgwp0^gvl6d4L`Aa;_eZnl1{cv@f&U<vC>*lPg0dp#4aG1P+XS
z&BnShnnCf&Sg{F}tk&rDH<#cE*B^2D(cilNAByaEz#qLPo%Bc*I61B9aBdWr>5#$n
z)@?D{2U!@}Tx426|1mK{;kFm1J`4ze0qO1$C}d`1xIX(*+XTd6UEljRgzjCg9)vVH
zGo|Ot`2RUKcqdAhN?tTD#v8^rhEEBx8>3k?$I$e+FteOA-2{^F@Y>P~$0(*t-^8xB
zn5tG?8OkvYULPs=oS>4U#$_X$A)Cma@cHLH%^@!NI4L^+Dd|N#4iZtHDjFp1?1_ki
z`smn=FL;reyQQD!4Kr3^#za^^Ll7D4m=<^x+vc$!hAxo6nW5xYx&cMm3h7TT3`Ag4
zf2rZWjuM<KHwwb#u^M0CDg{Lh5Ar0%I3!|`lKbQ+qy?UypKJozpmtm?Q%5*ptz(bh
zJe*px&s)YU&QS!M?xK>@<$WaoaIpRK4gJi28wwIiuxC1&<8IeO8ryH#TVIi7$0w@6
z>_FSyL>A<lAgAdP)0k}aGFhO#((A`xlBL_qva7KeEi^2Vy!1hOY7l;3p>-Ml$8rE4
z5B@&2p|2OckTb;8Jt&>s$|#?A8Iz@EPhnsZ_?TgX7rK!UOU6YnK00U#+4k54?_OR=
z!PNvq_U)%3dUbyVY6Aesb035HrZZIy9jnn|x>gft?@W%RWn~Gwu8f0xRK#!8c^e@e
zM6~>C(rt1A@aW=1D-)1%$33qimJU@y7<yDp_b*oMpYd^HL9e=l&1weP2Q^z-Q##aP
zxd#FTo{X4$mA(0Z8m#dE#9Zw571>1WEC#G%r93T7L=_Q7H6#It@}dZTra%n;4&3Qu
z=6u2Ok8Jj*TY{>KQ8}Bf%>Cl*u*2<el8)D9Fg7tx1yA$2R8UcQG={F&>)T;>o9cnH
zF#>&Gk%>y3yW4Iq{5jp$YG)kRl^;K4G%E-bk?lnP&&J%+)Byp_H$^s)S0PmdD<Zfa
z%YR|;b9zS0?{@(-^W=ec!@^h4u$KT@-5;(3FLyq#Aj4?QKzftm!#?I+v|st!wl@1S
z_cQ`@pw8W`acP+U>=6_qgpQEpgMh^DpbdOuSF6%?`8tPg-SG-zERT&*t`8>xoRALF
zWy~Zapk?;{$>V|R+;D9=hc$!@65zUr>$bEkmxHEk0enE1#HI4*O&{P%;!rq<0p$)k
zN2(=CALSe7?Ccb7UBAr+-KagWDAQCXkYNPb=!#7exdTdOBA<hOyx$ew5}Bqzf-Z!v
zt8Hh#LuQg^m_O|Q_qf4^fxPOQ0(6SsJAKT@OLM@zk0THkJFB`a><GKsyG<9n6ASHD
zm1FrFa!nAwGx#aWMH1}tq-sH&GZZynR_B53OE#~37Ggv!PLW3}=PC!$e-e|p5*#!s
z<JAS{BNe~IV(}poTOH`(Y7@xdrObn#=Yr<NbI|wB)~XqsJUBU{6|d6$j1kKL3m)A>
zQ{W=XR-AI(mFe3<lkFoelf|{){@!GCIBr?T47Aubaa))E0<OP+3)#rAQh%lZ1W-$_
z2Q>-h$>HY1htt40GSx~xWv=R}fs_{w^s4Q^4q*qcx_uIP?LPT*YuFw4`XY>UJux=l
zXx76+>64NK29U#U@PeXU1UB>HXn$C{>$|4M$@uhy5{8BFuc%7^68SSN-peraufK|m
zz(K$@Di=5y(YL@Y1%ee?%oWEara?eav@zyY>)F4$qPy6WyvZ_^mWi2AXuGVSIOu8W
z$e>+YjX*b(WO2bdpzO6i-ENTzxJ5HDt8{7fSH%FK^^M!+KPV24cB>b<93ps5A3-Z~
z3p>U65jg-8Vi?G<1}0d>g4JZTQ;|s>fRKHV>{`>&@|W)iq?tO{_d$rPI!OrB$}sAR
z>rRgZc|k@jg)cFo*aG$E@?KGfD$%Qaqx^4BrVjOFNh&ewB8gbEYma20rGbLp8at-X
z6<w}!&<@qC0Npyn=~}l!{gxXKe~^&)*zvaA!^~>olo4tRr=dYeHpq;7Q{Xm%J|PYK
z(4QIL6T-dQr56A5umIYW1|{sqT<yB4iAt98sPORcfV;NMpKRm@ZoejdC|hbd%>u|4
zG`;;1kVg|}H(X?k)VQEk=Hm}I@JfMKZ19YtTTEj7+`&{za0T&We*bTM(l4ulrd~Df
zwFCAA&gr%>%g~krXv=PC8o8{GmCzz(k@Au%2c<J$GFN_7v9`8e%Drw!r=6Wn)f>~Z
z$kAeRb;4-L6B`k*;r@s>chJ)%{>uzpK7qQ@lO?~zY?S?>=O=O?{^<FFfRhf%7)rur
ziKrw#gQErAdKz)LP^qOg<qY$sstEgftgm1_Lw9lkdu7b=6F2^KpZ{Oj1KLKQ{pa?o
zTgx2WF-ebSRmn}L$^%zVtifG~%5mAIwrhMq5?8be6m{&IEms@QXDKc&v8~f$B5VxL
z$4=xjQ3gH%9V*e;?q~nW58subutwh)a(-u@&0BPApROSEbD-_h@%4#!eq0h%3aGi$
zg#B*xgJOzl+&5{W%yPO{=1^@k4cglK_SeQB*ivo=vVY{$KaZKZ5WuM`?33=wCP+Y@
z3T)l_dSk9eMgN`y$jkBeB)w$ZRyygXbH#{`@j`xgdD`WM;@5(d-uQ4a@K0;O%|HRJ
zMu1(<WztVAyBRv({syYdp_>RXUOJNH^#WqnaWoZ01%<~r@y9?NJ_nloocoG${DX6T
zgFP`Uol!BV$kV$$g4&1@D`vp>Qu2X&7bN7HULQ#RO32W<2f0EVKvr|6+UuJCjV23X
z1~x<#H6H-Oa8PXovQ*CRJIkaD@(2QJS7Lnj$$;>X*LlylRzSN_ei)&vwe9%%x{teH
zXP6DFSbpoEFITk9kpFV_$?|kvU2Gq?l%dkRC}XWsaSIc>5jb}NlT_i=#uYuyX+YsG
zD~=nF1N<D+vl#K}8c{o_875Y-DFi0IN=4l>+-a=5Vy#*Mn7%a7ovSdD;kYzY@oSD$
zlXow#Nc_8j0rdHx9_WK+N7QlPVp3D09-3YA!O*oUH3d#9qr*ja2fO|S@qIlXAi$Qa
zbl9{gO%w#Aghs|aAhUK@t;n(6nBXasF}Yj_E;`t@r_eO1KetW%5&Z1rFg||h5YP(Z
z3!SkMj+j_hE!yr!X{QL21S!qPr3|2Up!!x?=!%aqAKyX+E*OE(@+S+pb#^}I&qY>K
z(D0EeK-^=*X9W*wT6MbrTikXTg8iQ&(_v<Pu39VXpw4QF)w5l?&PWPc&ktd|J55B|
zt>8WrgDRN~319N7xjBQrrec5#=I`(>R3(8L&9E;vPbU5W*E*L<_D04}Thm6iZX}$B
zwY2`}7Z6{xg!=fR{BV|jQHCiLUhjRcYb>o+cy$-&Rio#?ia6GACP#qcYqWYPHRyv#
zz^!+F*3X7LvlfDCwi>UcfeK3D-Q!mpiDFcjM>0YHZal6DkUsf?um>+TfX<YTTX66|
z9X6-TpJeesOCACF>q4SnU|#*{(e7kJ9niWIT22eZ^1GCd9**dTRJ1fQj>&d#4Qt@J
z)60#+678ZVfQBD^gHDbp?ee}jR0*W(C$FM@YwIo{4&*6XEdVmhgIO_V)3uWm)&L@O
zodE0|MvwN=CHeSznfW*^f;HA~+aIJIwp3G#tI&4acvqn<e3KJ1t4Q@T(ePFCh<Imk
z>&Ir`^@B^ZYJ!4%_4^j*Kj~3Vt{$t|WJ5*M#}wR?FbeHPkvz7u+i7f`vgzpX55B`g
z2@rS~;D_P{+|Tw^4s8K956<jmnJ^<sD2U6IRr$(G1=U|AghPp#FW(|kz>Ik5OES*;
ze;WiK7T+c)2b)7fLn8{*A%;(u&~)W^=J6fA@Y%0ftt#m$&SYjJg(D<nA0#3=k+{lN
zz*{<SIuPVNsF)s9**?wf4WE>Lu33!G-M^E`yzC27mxI8Ac7bGd@~`87#)2RmMD-3}
zre8J><8N^4sC-gl=zEkT`%(d)d%U_OslM^mH3a6j2Ewi&44pclt8@tEjd$ux72P;U
zmx-UKRK9VCJ5*Qw;a&o`LM0A>DF>PU&szlFK~9mGfb}H)(U|mv0y^?){>pGmM>O;5
zLA&(x7)m2X1lrM8udq&w-98>O%?rTzT4K!WmC%k534Zav$!;;hT^5_p3SRKjpDsh{
zK=BJl{C{5wwL_mGR0)L03<Ijbw`0J~i8es_!h(4+ud;I)h;?GFO;!z)bsT~Yt<gid
zw`@Wn^3(1v_Be#_M1j==Mr7GFzb0mP{bA(<-e-~Uc?Y1jSZBbqO}PEc!~f%TP_6Il
z5i-K=MvZqFFRZ#(_X9CjigA6g2Lrl}0kt#<H^Eijt$3kh2TCiWCs{6A9#P=q4q{k>
zZYk%wXEc~-Rl~4~w%g!Z79b;>elkhOXiA<CrsTgac0(Z{D92oT*?|~K#XzOmR2@Rh
z@x-EXnYPUWnQp*eFi;w*C<$)xG}qLE@Mp9JZYQ#%^&;-2@N9p<lc~bMCJ_gA_`#qp
z+#*zYfDj?Fp!ERknT-q|z{HF{w^bcHm*PLqjX{w;mxp1@fop>EzN8kRApQ&pBRtIo
zsN_0NfR=-amxJB5G>xa8;{IJW&@{pmAH=_<W?N9QmC8VCQM0e?a9|2<HHmnygG#|R
z9*_;qa}jPkZFG}vXu6tLf8w@MB%$X57=Ie5CzJ&p#9RGZ=iCN9W<*G3%q@Qmv~t`<
znjq-y<|1=rPw_NiP4NP2_4O=5QUR)?BC=3CdBNom5z_IIk@!4Ea}gY9x)+-l=RJ2w
z`RudDL-0oI9ON%Szqm1o#8_uPDxCg|_pL$o2vj<dA@}e|u!nb*i9xfn*miB?DX5$%
zfqFOssCla<2f^L19176r%25m;1EE_T2)U%at#^)nujm4j!*u7ImqXW%&<p_Lt|Qqb
z9qdLv1_pM!&dXbDa3AnWl=#q)#P~nnC;%x%`+r>hcRber`#z2%vXV{6zR121%HCyX
zy9k-t6_ve5R>@x3A-l^Sk-aktAq_&x9$BG&=cQ+__vicj?{#~=o;*F)c^>C+9OrQd
zMxgv=eb4)xstZrfT3*sfWr2tE!d_4-x_3SfOt(fNqp&x2YoE_7e~?>(@;%{!_>m{l
z8+$2DAB>h`7C<jQs2(9KThTye;;qz7u=qsN@GI$Qf3L(an|}5KIHxlf7EUgXYAyrD
z9P)p}OHdx!C`-YxAV8NKjAcbFYKYkY+L_&L`I5MVV7V3g2omJn4}u48U&#3V3cX_e
zDgqZAWSC7HQ)yq$GMO1YEpcTqh#{zX4Mki;iu@-Gj2cHwXidkKUIQ#hpoBeGB+ZGh
zJu89Jh5x2wCD`Qux5xb7gbdz3_l?i|vd5btjRyx?^U=4-dK;EuG8l56#4x<c&}qxG
zz(u&sH@yJ?W~H}RdD$?Yo<2|&94pq}xIcX42J@Oaj1~FNfZSlR6@T@qyYazwn1Yq;
zI+rz2`~Z6{w_~K-0oc58*xNd-SwOAGIsf*arepm7B=y9S;n0U>ir+76oJ2w_4DRdr
zgoJM6?>sbQ8MCFAldbVAMX#6X{rn9j<eT$0(;1xH5k0$>M>HgD5?<zt`#`oyJkWE5
zK(cg7!{~3ZE>q~-L@phLN9X9`a~T=_zVH8>1pgs>@QMjwU=3z|vxOa9%*Jq<Y{2XO
zTnURdto@Zg%A$Sj*JD!g8Xh?3E~0=iOhZc$y92SG`$7+C6%>L`+Qe=@S&^C6UDD_e
zFxuvJ)h}O}{RTb4FCH+SMGJHkDC?2g-$9)$$^&7Sq38K`7;FQ2GLRJK|D~Kjh8(6M
zIUB->zXYR~<{izU+$LCDU}K(16|&a&JqhaJ<xZ3vW$4=E%WQEwn5^>li-==&Kp@;9
zb9#H+De-I~jpXQMz{y}fUEdIwC?hOZl4$5@D&G6-wUj1;*C?LyxA8_|^wY|$D<1z`
z%W!YxG>BkI;D-;gfv^Xj+z|OIkb>BlUk;~&no$ZQvrs1rkbRP+r0;WpMB6~?&wMH;
zmhV>3o@UezHp*ZrOAyNgz96sB^3}tYWT&)Ima?q;CC3WXf8W9!PCh#Ut>tN+(7*42
z$V|>I((!E~t_cm&oDZ|G0zgz1%BRv98)XCdm}~@E94B}@2+~XUP`AWemrF{)G$Ap#
z4Ys*&rX)N@P@vbs)!q_b9jW!%U%yTEi%wlrGifvNjQA5lDLdDR5`zNg=_>Ke4xv*V
zg^R&a!7`qA{-3`h2t@%Z$8ajfQvA9pV54P!VC;tM+Nwaj$0bxsa`HFxTlWK)O6{=>
z)qx+H8Bg0<S<lh6f!$bxL5>tttIsnv4tleAwIZ`xv7hd3|82_{hWlTb&A$(B6HmdY
zgtWD3ID&tGqA4&TMv>Os`)lMSH|czQX5Ud42I3#oAKMY0-Gm+QON~ZUYV6-CXi2Ws
z;2O>CotBV14wz5v``;n|-}h$0q%F-QeS#2xOP`}h?iw+Cv6DkU8}dD2xH7!Et9ZYc
zX18U*dz5f=voDVQn}BFhg);5;yuY6S89(;zj%(`4{{L@{VwuDd$Fo!n$!S~{3?|uM
z|Fp9A9(E!c_dYo7{7l+O7j<e{qTMYs_h8Wn=4J)glJzgQ12!s{(u>vqKe1r3yx5mm
zk`3lFVr(a5-W_2!nCsO^in4sS`!S51$CpYTZw#TTSq)2@Or_$z#HwwFrsSdgh<gGU
zd&2)erlPupeL}y1QTw4Q#)Gb*4(o;-uL67Es~L(~Zc^JNMmB3_+)1cpi;cwzCuJVq
zdwpLbc}<mE3Jt7PJ=BS&#@&yGB3M4aFslKRiCK+L&dZu+kXeU0AbdkDVhuBh93cpv
zk%3bMLuf~t0#=Qa#X0+rn{On?EGhh4oz^oCIC<y*G+82_v98R?Pd$uM13MZ*dbAjC
zIn1TXBA(B8$H&JtT&6&etYps#Flzbyllg`~!z@lCZ2jZ|*uNb({J1-hzPwP){&REV
zM`%I9_h#FCNuC4q|E|TyLseZp0d$MS(x>*BB0hVo_754WPu8cauVXC!kSY;JM9tH+
zdX?_NDU*8)j;SZvcK!c_50OpcKqa6!QQL$qRaVTVvDD8|#c%!(-B&{iDTP1z*vHO;
zDhxO<`G0?7tNnkA*VpefZ(W1cg2W8h9=h9s^bO~5ag>EoEp=ZuuCyL|VHVTmT+Q*v
ziuEqNO*=Gk8wgnWc?1KF<U~>H2qdB{Pe#-dm!T><&N)n>E!Xcy)(QzToqmf47*oux
zk%U+Oy*vVVoHw%gZUvkG4&46)tBy-)KF3g)Klzw}MtakO=g;r`*6yVp@n@V7U+)02
zfSi`mTk+q&xSn*MT6^H{=S+$?zu853<3X>nJ~v@TU@@iKL;S>EH3*H!2S)Q7-;0rq
zrUS@?P&zaARsiFW1tLWGNLXm{54_cHQh}dhyl-tSciJtsz|i(TA10JYQ^Lm=ggSi9
zeuA>#v9SU|p~ngdy~w>4EoHk*2YQ?r7`Rxx4a}nu0`0M34l?+CrJehRFU>#s+I~@0
z&yhzB{VB6u|JbN3ybS!~J5!_9I`h~(FUw51b(q(lPEWwr!O-JB_Ajh3MvAN0{nqD8
zQpH9s>b~th)SU#jTODHzU++=%($Tp4#PSsKQ?X;l-0zk``hUR{g^t|L9t^3PpwL>l
z?H336KY0MNg~0Vz|C5j2_R6jA$_fn2O>{Aeqar{PmTR~oax~}84%tHqzgz>0(U9Ko
z+Cl7LWlJ8{eDBjMDGzaq@aRG=Q@Vd{_7Zo0UhPi{wHjKqRfR6?!j$n;J0=I@=G~UX
zb&PghcB}Pu-#h7sGm%!xV^hZm^H0Cxw!#o^>jez|5H(-}`flET*+2mx3Y-iE|MQhD
zfZLmVccmlQc5#$vJNyB5g*viD9|LUbcL15Z+67Fn4%qY4+Gy0b<e#2WFsVEn{e-!S
zN4O$aza>@=OAIcUKdgHS*cINl-0AVO$I3bx<JfTo9m@ePf3~Xjs+H%o+PwSoZF#WK
zAF*zBrsCI7@Fq){xgE&0aSciuds}ni%mXlN%$~K}Uz=HqNk~Hr;GLrOATDBra$`W#
ze}rz(RdN+IwRp&uh!C)fPfzCo&@FOV1LHf1OZO`Ydiv*hNtTXk7vJ&4IXdC4;@jT0
zUi&r@(r5duH>Sl@f^K1@XaTz%>vF&eC9|efGu;ceX(G5~NjsFXY<h)49;KO?=Rgue
zjUvh9_9?6L5;AV_u61|~KxYSDdX5xI+{r{-3}LA5(VbT=@V(*QoWhQJms%1}<nMrX
zkj0vabtkBzd;2h1FhLD4Rx!{%jeGVNjpp3L#pEkgTMVBem9luvSt>jAP5&ppZ{`yT
z%MM&|0QPQ<=_}Us8KYRiG;6H^e|{HcO)!7)KPsR81QcG``D)0%vy#IYPB8ks)`Xu&
zaCUFH^l}ziJA}unF91vwmZlN8RCbwfZ3(si_4ta`NTdy!p<Re0h-A^t&`rKEoE;80
zjeCgsQP23G?v4~y=BLtV@%7IwGtC1jNMyQs2VsXftg~(1d-n0nX>k)Ghs6pZFYR-G
zmEs^2`Dn%?*qJChG_eCyjoZcK;nayhmwX(ww~$#`4>#nQkY!FMntAPi^3dc7hFUN4
zigWckRuCJ_Xa-sfsOOuQaXhRkuSG8D@>xI321S1yJw1zia69aa=}r2>1Na0h)ib-5
zADT#>>KuZoX!Ehb^-d}9RDI)_fiGXLyty?*%J2lN3wdu{V*E;;4{tuym9&p>9l|T!
zFcE0_o_BF%iREwCj>AMSt7`uGtM0G_;X=*fueKXW#$W(5LEhuNY|pQRjgalo=|OAn
zgW&)UL0flUnWN79R3MLkyZzB1`iYz~p?!|t&g<nbMrNtJrMfUPSr0P}EjJ&p&j5v=
zh7DbL_gCDxj^(_Y!{=todM-q=tO8+UdAnKS#0g96fxIF~({h>7rSB!x*0mKIZ@F^+
zti@dWW4ud8KkE^LG;7iqyZiaH#X_+VfAv=@m1Ft;%4`v`*qL#K)kQ!7aKIOX{;pBG
z={Dn1-!J=n>*?kvywB%P<X=ODo!)!FTXN(p1Mxb8RW6+{3~_m5lO+Bi+66j8<g$4+
zeVl*Tm%}Ds!(R558)rHIU{eB1*SZ<0zdj0|_yBzT$|3pF)OLl1Od}f_Q&0x`fEym-
z2!?<~`q#q*52FMg6n^0woqlQZ;H?13n`_?KB6eqYfP>gDl`z_nNIs@%p?x?AIzy11
zU#9nR@!?w5e}R{MDS*{{D_tr6KA%YLoI`R@DnQ$${Fnje)^qeCD#KDcZ_cs&3i)$?
zL^{RtR=(KJ(gH2a?xNn%=4$|`UclUFVkw*<htTf`Gt68NRVdzH3_%5h&{mb~gd~FA
zW_+OiIknJ_8IvE%cK~ZdU32rkKQLlLufFjPRfL`;ow)Q;`An&UgTwmd`a{W|v?#3r
zEg96O&tRd_4@SeBB`iPEeAcq{@Ee0&s+a$yTvgV_g8xuS2tXgZ+Gl=$XB;I>z$)YR
zKTeZ!v4Z~CZAj%5@Qkbn<W-g@rMC7&k-W;AQdZW%CGHL;QLgj|!ShkB+^-Pg5nB!8
zoYLl_qSju;s~n_#E=1p2hIqd*vhn0-p=D|KO=g<pWsmiiBjaeF4>Q9um+h_|t$N71
zoWW+Yx(nVE+^<*#A3(p$<g0!o3`vM))P?1rp_3C%zn})2jmDkUlf4oScVnwaj&TDm
z23%i!Qpr{HsH}g@d3FojWd>}jXw5jZS`=PxO0-7VPrB`@aD*X{*qZj~TZ_8KDPzce
z6$LS@#t(s>C%Z5IBq%msq%cHO^dM`U>RwKH2ijN`-RCV2Remes`nAhud!c`No%oJo
zXbQve9g>Sz1slQYv?Hd|cSGtf2eTl~wKjqa*zBX{a5Xx_&PW-F{`&{X<Z_GNZP@s@
z6DzslOzF^bdBqQALqcA#_&>VN_)uRZDy;r9!`)``926|oD11!;m89Nr)VZpk(_!X@
z)Rt1yFp-{y&azgjAS;5@`oqgkf(t>vlmi{MHH_{K+)XOm!|mTwNjF$KfED-72kv}?
z5rsWQ1D@|80!RDeqQmPzr3vB}Zi%gImJ{K8*i|Rp2)-_k0aV`-cyVo`lUttb;#_9G
zKrDvceb&Z@dvV^KHJdeShrAWxT}>f&?aj*H3)3dX{k`K|OC#M#3ODz0AURxd&N?}U
z8c_IVpZF3c`a2hp`AhQnj8uht>Q890BgWMb;uu1XB96l~O|2pu@LvDr5vc^Zc=!Ui
zV%dRpR7bPSPGJ(i`o6NqT+NIB)Wi?Jy(BPDKTDI_642$=sH8h?lD;EG%~!2Rx%TlE
z0fA~>Jc3X4HGu7`X1ztPv`~o)BTf{dk=i6@xVN81+#D}2@@9QC-)0hoFl$!Q%Rz-T
zFWfq=Qtur!cDwbB{wcjEkN%Q}@?@2^4~pAjQtrfRz;0&X16k|=37bj&Wh9Me$8DXR
zI?NTe5$H2u^IDcZLC;+I?}2JB5p3Rz$*Y0>)&EW%nK|pHqMHhNFqbfvA2wxY0i0yL
zkkCa|<3Y#lJD_Zbm=MQfRaw9is2Yv2O`IHkGw|`?sic^;#C<Gx!XDU>#g!ad;4DMo
z4wTtNt$EmjCyspaj#1pu>V`h!qqfit4wf^KYVqir`q^OOZ@YDi$1GM}mCCaPCM9@^
z8Wpc$2i9Sm|3l~45dc*i(YbC0)>6FnVx`9ZHigCSq}7P8p^nFeRFq^LsKM_$IvSY`
zkB3A2$UgIy*XUNkT4^{N0wuXpR;?i1&=52E<n+;=UU&j};@Z?21NPqsUZjV+`CAea
zJrV%Yan9U%<+44hNXl{ZPNS_GJRSVq3Fu>43+lcFdckdeu!0(bV;C9~V{o6)_?-B;
zw{qLCVbR3>c>Zp{(M|1k!|J>BE^q@D$r`{6plA*5E8q5O^FDFokEe+~;`;;Hf%bvO
z8CEYA^ZZz^!{DO`{pURF)Ma(XhUuXvpEmqcj0mHDC7?V$J?xD4;ak$tn4K8%^8+uG
zgKNa?X>vIgnD#{uy!GT>*F^Kzp-i%p)Z);w6%hR9|3S=ut~+#@AlhoV%hrm64~1-8
zi0=na0+aa#f!c9h!ll7F_{>f(T#lbyknpB@&8L*VFT{l)nrm1OCnLz&<AF{?XpBH5
zzf9S@rD74bO(42*=hYfbqGI<s>bocvs%DFuY@X;xay<gM>L~)sPhWBLEX%+RwttL;
z)FLSCrHumN<g#?e4MIuZgG~`tmM#utq;~9p35UyU3iWtlv+vh1D)|%0bGif|^{AjL
zfQFb4EtifcZ4bKb%m*;QtHs6cO)9wwCmmZpH!FpxGu6T6+iz0u-CAm<`uHL#RefK5
zI8yu7H20qgHp{f~9H)w{MU#PzR}^X}71Z%V)>@sefk$$KrYAnx|1Qsw#p1@4W6rYJ
zzc*o=4hJfeq1#MGs961HZ>R0a*59I@EPE@H$k-IAM8&RTN;39=Su-uaNg*f-e(YGR
z;ZH?Et#c1fHUhGlhLt<EgCh}IWT~j+2-DEOu!!zkq71Pw5Av$=&o8F_%5Gu4ZICzZ
zJ}~{wmo;ju{?Wk`MFbzRUVB!!mWhITC9tpuM~tk@v!9~%*ThQ-XZlcl9&I)!<QeXZ
z*Ev*C9Qe=uB3vsdld-~ZzNYrQ@D|Wdv94mDt;RcS0{?WlF>@?h;=MkD%$<K#>m5Pj
zewV=yod)Uut7KmZHpzy@Z^RpxSN72N0}B(snS*(U@$qQXiXpw|3fo0Bx(LE4-BCAV
z_^B!axem8m@s}L-8tR_0S$Z`pMYK`<P|+;`LBL>01QowPyltq4s>K#k=dv_1<oJ*i
zCw#>nCQ$92wfZl2#&PnwdH(#_|MMHX+tDs<Fy3ZIWH{^pNMVdkCd)kVri_hw<EHI~
zJhq5T#r<@zlZ1hzwS<cBnR?hUC`MsIRKFAzgm60QGGZCqAw4KXYP0POhUHSYTuD)v
z;cnLZ@8!KuCP+AQ@u%tJzkk5UCP$InNh&KP)wNOo#Xl%<$}Zb49x{Cd<jH*?^(0iu
z@Zr@5r2Wuj+#g{foQ1|Z(zIm2J`q~H7+Wxndt=*`nvj=*G+y+UE^qub@>q0i3fr*~
zD*pIs(W5V-ONv}VhEoTNq_|v5s(sOW!;%{sg<6^EQpcN}tPYb%Wd=U9QJ}9r<9VyG
z;2RZERt2GW_w}oAT}oTg*2CGL#ERUN<ry2NFsSGoZ78eIY)0PHLZB<0OzTn~*5b!$
zMyM>A(Cm2e7LumG%XZpxwRvzjE!CCZEdK9>WyLAVf+(~m8;BM<|2YbTM8zt=zn5Tl
z$B%_JKcc)d*GC0@xW*c{!Gl^c>g3y%R@DmNPC{wbW1k=vq7`iz5q!M`p#>Ux2tG9(
z0xffk?sh77tLh=KXb&SBvpSqc6;uxd9PWhsD;4E2VS*+?!rB3h-H(}L{&R3$yFJP?
ze^g?njB>#I!CPfZZefZ+Jo1;T3D^0;Afv8P&TY*6K3fV=^4>ZGDqS0jm*P85Udz@s
zZ>?FrlQs~KE(xI!GsJdkP5FC`7PyW<fvQs`Ijb^%doL7dZ0v}$$HXAEZ{&C@b3F3?
zK^kzNnCj1~%$MbfK|qh?T3FkpS1}f*sL+Y2w89&5=D)!jX~|fNu^dyscX)baq0yl$
zmQv5<j*F`{_2)4A4<XSEJP77vlyZk7&#M>%EABjz6j^R-TJ8a2Q9K)Ao2{1p&!uK0
z$!vTD?%>?CR%yFHOHn-d8i3iA*p}GJsrOt;AJv|eoD>LHBT-WJNjB#w+<$(5OOl?T
zYa}^wg~!~?j?u}HFqfEc(2LM+qLMKvcJtle*V6<CI|0of)N~~B-=hUM#X_ttn)`qw
z_cVIL8DJr6^_AX=mc0))<+bk}uaQ<bV!rDl85}a(gOuBmRSMG4q22Y=RY`%q7bV@D
zO-Kx!H9Fi_N*@*6I;@Ne)nrJm#v#XKR$hW=14nkY%K}F05}}B$e5z-zl8Cu@3YymU
za8HpJAJy~-vu5{}tDvSYZL$eyHH3xNQ#FWS{1In)ao#UdM=Q|e#oJd<h*gjtV&wlv
z?yPo|J&_?R{i3X!ElHxdGrHYXpY`SXb|fQc1<k)#`Dsh*Pj*~CYW{b5+Pr`k-xE3P
zTmP#~bOgsg$8PjDxX#CCx_{u7>ch~QbL~GN_KS%0yUi36_)~t(etukGI)S-m+LX9T
zHQlQ9>B}g@Dr{5eJ#<RG*KH@piREw~)NYdWNBJSE2<-v?gSl{$7)5%4sy;SWFUycK
z)F0)F(>N?E87)k8JKizovn~|j$bRJ3E406CmWRV>Xl$D#xR~Xz4_EL9Z0P5^5Gzq;
z?$aD)C9TwkDAvphq-e=Na?`kH(e<yQ1h!!m;SLNGhWS@_mHOP86J22^fYsDzyqc_n
z!5JOYq>|FOSAgGQ_3tX92#)tBo1-!t9R9BYOW}Ton8gLS`Wwl{x6?xTMi91^j;p{h
zhY+oH5F=|!UP#%S4MJU_xh5Ai42`q;dt5|lSP^22GLFMiCnmm`H)eGYs#!AAHOm2!
z00lK2DSOs%VDU4POMem}oZrV_G<c#Vb`Czjc)92*qG4MwFn!QSVasA=hFW-pW)7GH
z^nInwV|kpL|CyQmlQ(aoAX$|oLo#=-%J$iZN26IH4!SsW6&B}eUPg>C*z&z@u0oKB
zpC{C!((Q134E&tdI~cuZuax=`3p7|%eQ+!-9(+*jKE`PNziP#fdFBN69{+phC|pmh
z?wz}(<#j~KXO=QO%v0+I6gkK4rnwJ>Y{Q11nR9}G-RVm%iHivx5-fZo_Mzx|rF=(}
zJ{2KvyjB;-*e2_4>iWyRcyy&&f^hnFz?5Y;$1~?DRqe-lY@e*6`Nx&IQ^>Mo<Jd@k
z%m(2zsiYpV*p^5PnTS0GF*^KR!G|EFwJ)s!QzILpQs$~3Q;{meyYMJ3K|rG_+m|U_
zsv_4{Q;V}g43};w5=Kk6!K}d?)k8E3_Q^*!j@z*(;F(aMiz_hhNOMw;Jy7rBAF$9z
z1MTdg_FKWUm*}f@f55-ORl@@H`X-sL2jIU`u*q@_i@Di4Mx4`ghvb4F3+ov%(sw`<
zlHy`Vp^DqSdYhkO@0x44Xxui2TB@g7<Hh(F@5x;?<Cuwe(?!hlLkx7xa~uP~PnkX?
zZ0a3V*+(lN)RPzf+5jVednwH}+9b#-$>C6~CM<;;!PmYTezgk0hl?<;bL~Q`zQ1`l
zlj(!oAb$dEAaWv>c{nf7lJgoz7rzkJZN$|057cGYVl&RwvlB$(Oq9)4Q5MH2{(I#F
zQeKhd?AJmIe#_BNLN<))HsEhWY=h%dC3E^KYN+skh{{``4dE<pM)I`#Z<+hRh=b5=
z#K2xFega%!-<?fA*zX8kVAJgqW}JYr01%=&!7O1kPWAW^gDN2*x21C1L!f!Fr4F9)
zpcy>ppIoV#?u$-310H4{h3yg(ysYxaC*yiIG+9XLo)3qmk-2V4x&nmmeHEPI(#+0j
z$y!exiw09gz1TBTwICi_p5G+}p(>~e`6oWmx|}X^xriFNE96nE`oK>qS_#zIX6Lw~
zpOO&{QmAKvzY>YCI~sY5Ae$)*!3>~(k+C$Z_Nza(<737?eE+(X$p!4fxovHd{sF~(
z0QxfvIU(nX{-~v;&vd+)TR&~kw(Kkvb>n9foryU3_HXFth8~+vFh%BQPk#m}C+Gon
zd;gZH+7#r8-Y{^gqEbD8sp2f&Dnv3R@9`SEbQctJhBWBIMZvYKldVHEzm;BA#VWgY
zPLFEk@9HamZpESD13VuFAeelaRz0~F!Iv+#R0ZQo0i*D1&(%r_8r#z5;7TM*zO%s@
zL|#|#{<K{-$t!!_Jv_rd1qffg8FTv6%Y&xEiDTGdNNI){RQ|rG3W4}*;HVWY^4Yw2
zQOp??>_#k1BRmJ0j%pfh*8;tuC<lo>!CEj8S4vYsC8BF`^Gz5f9@q!cBABX)MM!Ra
zuB7^f8JUgOTccJJ1J+)?*Zo~o6Yx&;5{KtBkJ}TxfH4So(^1xaXCSG>x1)<ZFJ-QL
z&~vm?j;!pWm?PsEgBTkzua)xp_%eFYTV{KgI~9XCun4QIa8kA+1Pe3b^W_H=(aRT9
zYp>cJ`yOUwKrka=>vPZNtpI>5>j-wanSI_6g}ceztXcX;T4?h_Nirlhu6O$HIxWuQ
zV3Kb)AK0ez{&olQ={#wiXX~~_wB2VY9;uuPLiU&k9#r?J4f5yGb1D$7@xeaqHfhJ#
zYg<zb`V7s@bJKxjrBxJ-t9F+AT;8-?Vkx_Si_P$s!mPqMVa)*O-9)|j$+PF=l78F!
zijY<o1vb2~i^fVT!L?!IEe;pfCmQ{hh3^+T9wmn1g6@5?Imr>CG3l8PCRhXVF6686
z4OCnqXz3bOc^N|AO%&_yXKra-vaxKFu%;{?uKSgem*RJdf-p~RHxR&+cLNbq$lq_W
zMjY2OM+V9@g%n}^^zCO0?fA?ZV!#bq$hJ}4kh%phepk%nvkdDgjH1ETkK*E1l`04n
z3|-(27o-xx-*N}1!V7oY4>y=SV+SM+Bb%)#XdZI-jlec4t0iryz-1~&I^#j?T{D$h
zB8Qf4nwz0{Yz)(F%m!n>2HW-hoVaDp1R12gA1t?)&W9zB2jN~q>wJft#E`ECSkJvr
zKoRQvo$C37SODFILlR#n-}M^mMJ~aldR+ORx7V4&6b(CwSdo7bZ;m;j{F_Llj`_b&
zG|2ENR=>LT_RqDMD;fce5QM1Rk>&sE2P^we@MS1|rDgiXc>4VIf&(ceb0tn7be2);
z2%@ge54t6yx#U@=;NN~BIEH1wkL?<9g5|o|j{hP`r?8b$$QWOTYePR-qkZIs3`tVy
zCjR0b!}3TDP1bF);JO>xGL|Aj#enrm!}AR@LWb+b70n$AlHM(I8K~=YSY4O%>%pmf
zvMm||Ti*g?=r`wBt%St|@CkW^Q%^Abi5sw1VQfOV%<f77q3Wy`pMN#~N2F<A#`Mi|
zjvh9DUs4xp?D&q^rSMN}Fk@U)31W1gVx<GX7tDsQ%txB}7XbSx$szcb1laWTMh6XC
zzS7d;cd>_l1BQh`KXs6=TEiQKU?FUh0k097IZx|oESMi1Z_pW^v-Rfk$Qwxte|0fP
zIlE_ys&f6}x6d(a%KcS{)6#)?(NPFnJgOsUC8S(I;9;edT`%m}9$Q$GPB)OMOif+c
zhg4+^rO!)jmP;Ug)kvs)REYXolVkL+1CEmc&INpEM)$Xn&d3Lbe*w`e+#y%W8KX|L
z&uoHwnus(>x64uO_Kf)b>`2Uun<IiNvX)`eTm#ma%-|<jGY9_3l+Bw_iut^2<uLD%
zjOTm|=O&GO;jsF#lQ;LmlV0s%3I7@)^5`3@5O1Ab&&@}h=&263B!B0ttp$e_v>oSB
zjY_pCn2e-@q{2+?%76&c3UQx`{dEY+h87bLc!{Y`i81TI^8I`(VFwph*evJMILw+J
zs7A~G9tJ}1N8SHj*cqJGig0<)bN}lH#5ll5sPq;30kC<LFU(b(Y8CBGIAZi)a}-fx
z;j{$kwPpiQtRH|14N;PM$89@WK0wqCFF>a_uUh|(@xCF;>DrD~>2*S0<=dte?^nps
zp5Njv$^hP}Y9PKbL{l}VyUY6h`dMS2k-_)LxJS*9@EJ@3XW5SOwn9i{RNy$FfUx_^
zuyDBfamXD>k$i$7S2xcNCr?8^J(D)L!LwKpXdupgpVcQRk+|UUf{=M9+1CG+kVk<@
z=v{YZDW<=ewMY_Ih2{Rq?qqnlMEqm)aLDI}lV-HzS0NxQoN%h?l_OIurtk#IIbY8y
z+hZHHQm9DW%B#Zo_Dq!3Q0{$~MWwIW*#g&a?pbVPY)7p3ZceC~ZIS`SIJygF!e?E2
z%bmj)14Qm{TSA;;d5wNS#A0D08nqV7#mjI|N8wafc@2jyi<f2eke`Fe`+F`rkr|nd
zKfD26Oik@h3a%H`tQ>sF__#J=@A0+|F)Tv0LG!;>meUQi*38$Gwln|L3~%LG_^+z(
zK{wpo{2eZP%zE$&f?nqc9j8)m`HPExjbI<hQoH$bY1Q@gt>t+lR~Fi`RyRX0GV=4t
zE%?VbSo3yB*Wfkmi2umLu&ow$HABkFO@)lJL*Qmz^32q<9Ljy=J^!w^OGy9#nvU%}
z?Mk4JoobEMscOl(mRG<nt-+6F?+H#MaY{$aEArso6Pnr$eexz7fY~D_vW;~JOPlJt
z!DuB~Cp6_!|LJR<rP&f6MF7-d(xb&Q43l^DrzQV=WHn?Mm`UL2aoW}F*9c;hn)uK1
za37m_9dt#X|6)WXs&QY*D}?AhsfWwfz)=SLU=a<D5Mimq!cV;Ev&Keu<m%Ka=@QV>
z+}kzCs`Ge!db9I>me0d}3kvvKgHifb-t|Benxvvtg2aVYnCex}i!a3H&2Z~)8nym1
zz!)xU+5)lX_G1@}K~HoxS%kH$8%H-hhK<{ol2AYkguHN?>f5q|P-)W1TjB3b?+@V;
z<BV79{#w3(cqkKK#NkePShiwXF%<Ue=YJ#eV>wYW^d)Ce?T61!OSr_f!0@@NamtN~
zet-%)sMMIoO?>3IAdqyvePtJv=kxtHQ{Q<3vx5LV@E(Q+so&H4d2re}_#eUBNk)n@
zGWzZKkz5V<dju>cUc(@KwAIunu7bKPOs1>ZvuI+KxijThTCTR7E&((g7(P@{7}`R4
znl7|M`S{1=%Uc~A?6_QF6ac-noa8$c-rZIdql`ss;?Yn*;WQHl>-FrPfmI6qDr1Ud
zH_v9RXJX}X0enPOJ!}kT&6b=QG@o?${*S)F0m|JesOsqS%^b(Z0nsf$+g%NN{3Gj@
z;59d%!qxis2hd{@SQQ^34#sw&&4FgFQ8gqs6S}|;Mq&r;uJk~ca-+$4F({f$WNqy5
zDpgu?X|z053^i_kIsPya4Kl8Z6^C<$m_IC8J)(_`4MnuGEr*5L@@<FNwxr5o6L}G~
z4;{)S(#&7FsWHP7f|HcS3?UTuQ;@En9mXL>@ROTEK=RMl(*Q(l+jgXxO`NjAv_gRP
z8iw%6?ZYAFi~YIrMK<W(kgZM-k}sRtt2ZyB;QRt6^<a#S2bdC#@sJ6n-23}Ne=CZ8
zHUWxoM&{+Kkq;R5Edq<wiL6knD^Yk&h*@7lvkRGoe@jzVNl4MqN*b|PIun1S^XV~@
zv_z3PJdu7u^LVcezrE>xZ2XI{`aIPS)U&LmcS$}cah_fNaV$jx{|zGcwrBxkvQ_AR
z{jk|LMwG;2>2Pjho`|<Bw(i;8urIBcq=q6%ZoM1(N4f!5SS@vaeulC^h$LC`dbaU#
za2k0NpTJ7R+<yh^txGoV@>MN+R5UvB23mkXe0L{Gr-SLisC>H<qcXqEOMhOs;$WzT
zy>RsEXX49xXk|I<CuU@y+gveroUOwDDGlS)@S(0OyLk$pzR}C9o8UR*6^UUZ%&QUC
zlXFlRW6^2R$dnMpsJ#k>j=syDXr6D_W!`o(4YsS4n}e@lX+Iu<sd=PU<5oi|x7#L7
zrRev+OS1?<KgdfMx`4UCADD2O(d>}ylCTO(6YWNreeZKe+9rlP2Wt*Io4@=4<BVB3
z*T0XYT`tmj-oZWl@@9G9W7{4A)q2ldpBZHTgmx>>*N6JO-DImMl?Ol(#1=b-+X_%5
zcilYbRxng-+QB(UwT_7__4@R0Li)sRK~cAkjKjCbfR>_jIBb@x0%6{_p>5nX>{P?;
z2%#N=19t+yAA^M`SQzro)`uKhDU#@B;1<#9zeIwe!akn+F{no-+FOQ-;`))Uw2mPT
zIR}4iU+mJ9jK%lDpcBW}^7Gvs_G^ybx!3R!+0)@F4ZDs6_q_rhD`bsWHC6k{d*Re&
zrt1|B8(Ul6=Cp*8UfT<TZyYb>DY!;$Q&IC(t45#*8D7qF%w*6$d!yA)-QvAVq-HwJ
zHnlHGtM>VneSl{6I8fSh<#5^4`Sd{%v0-_=@B|sI#$-=d4MiX5$)OUXARAlBPlqw{
zs1(&M#H*hz;GJ)Gw1Ar45BM_c9FtT7kKPq0*Wp+QTieNYdTXHk%i>qjF$?dz90rVh
zV8F`4!S}fw_jmd6S)qdF_gQ6?cUJ)zB<nwGDEk&hK^VPAth09i2T+gS!;cz&%d*}z
z1A5AtgYT(=U7GAcbKZdqKa%xw=YHW|tbZY>OAQr3aM~8uaIv{9#XqIW@$dhf7u(E_
zo(JFG<10>8G2K{^9bn2YR66na@s1-F4TogdUsjda+*KZ$p$}dOT>b@^&Y7fh_X6DE
zA{-puj-IUYXD)Td)ZUIws^pQ0&O?T{k8dIPR+MU0r=XQd{bWBnBZir2(NATC2nUHr
z9ed6)IBO4m`}Ci*cp<2S6<!a=(I!hw0g&@!-2@Y?BqP}Xoz&r`%gM^EKj~Bn-4{eD
zW$sGZ3*l)mbJ81#*)TOv$-+aA?V4XMeW`U+)<g8UjuOr-c0U7$C8%H$%yaHZ|N8zs
zgrY9qKV;%UfVfK@Q|aEBJL2;L1~0&b+a7S?i!26ztx;m17{aYnXJuuJ`+c*T8hbwc
zff*c?rdY*O7k$VGXN$1za<{%Ba*liro!UaW?GFdlMBT4`iqYjs^}j_6-inyOHG<_6
zA}`;+Zv`t^0=jJQIwYz(DDMX5#g+BJqJb|L^2Sud_H#JqjLCT`vjv8k1S4}V4jI$R
zhsU#k;G1Jfp5Zbbyv9HTo7-8^KJIQnv~upMwP9{yKO8odB?bGf{V;EuP;kb+LGf>n
z-dwX^uRVhQ3kQ4fY9-<xMt_$f4-7T1dXaq1A_$L0C4BTl5*nj(hUAA&@#<f00+&RQ
z*X^#&nL(lE0Ya%s>zVLtNzMObG$gOJsZF{2Zuu?4wbUSSSv323-i+T<LB%`8N}Zud
ze%z_q4Sy?T;uTCVHdi`_r5D+JE=?_8(20E6V?xPiTj8!xp^L`80xH~BoxfJ!b5;_x
znFQZRsr0wD1|?Q0SlBV?yCC>L-ss)dJj&F2<!uk)P!Lw30ocXmmFR&rdQOPS(L}ql
zto+2d)!dJ^mO{es6D(ki0jqnaZ=pH7569YfST{ys#*0IfV_YkWI;O<NA?$oGco*-U
zU(<gUiYHl-D`o#9;piABL`MJ418MdcsqH>7>pzrQegEU)^f8kNHSU<^DW+Sd!=~bu
z%Fos9)|Go*+|6Uj`i|wUXsrn6^wkZ`E1yK>Sf6y!MLE;0A#Ee0+`xX>4<Ic0co7*u
zs3fc^dtTyujvUht)>OKJd6QZNp1jWL13!vv?!*hqGf5~Ja6J&xP-xuezp^6oxmiPb
z!b?~n;zHlidzq7ZiVLqu*hjO-MmLmO$g+e8Z?5E9B)wio@outFJiQlhVDK!|_IL_w
zQ?hK<|HkIC_PcaCs~g5#tFoAP4l_i3pu?$Sx3{DH_yXXb+>sIl&b=no!`lkv@Vv)X
zDzt~gr^78q4e=H_eYeZ;(JxznH_20b)ffbaE?UMQZo=j`>GB7NOyZKVF~8T2*AHg1
z^w<lyY<0EdXe^q91m_fm!XDfChaZZSA_OfTn;>gH0b3r*gOQU%N+2k~5xqu&`8#AD
zO0!4&%L2e@jCZSDFz!XHDiHIJ$14`Ae2GP0g7Ci7=_-KEhWSGZPriF(38Q{vX689&
zq`h)S&jnh?!QTI&*y}L%*7cUI-^W0iPponj0>5*QE!D0@)_7ihbirB{*GX>+&iT27
zAs|x~NF$?>4y#9n_$Jfjqvpj*Ofvak)m4KW`q))y!<LT6=00mlQYSyh`e7?50jw9|
z82?N<y`vfPNIcs0+K<As72&1>a3b@>AXB*SQ^m-MuPfb4>{@LUDae|b37loy3ZsjA
zVNc#}Hj&%U5ZKsh@zsBHd+6FBB#&BK(5by@sWL@vWGV!K$&x~}OY9Wx02OYCO<9Ee
zq&k%T%8a~LDed-`K*TUJjWC8hW1!tQ9rJ6rakY!dH0Zj<H1iplaxM;*&{SU{of~-I
z);D>b$qO-Ulb{)#M@fo0^aVvLUM#ATJutC(?h!*g*m`A+!3vPDQt?Uv(;M7|L?>}q
zaJj9jtsPxv$kJg7wGUNoL^$71L<6C8?aV>G`IDEl^Zd5Xj${-xBcI{a1i4ooRq?;u
zNa*(-!C&?;3=28TVs;B6^f!+>X05iXIq-?Sibmg%xzzV?Ao*A=ov7I?!36l_EK;d5
zW3Jn*#dOJJZqEPTqf3ke(n**6zSn7W=T(v`>F2MTVv9wyb*Vg1wjr!u<*umv;21<^
zE4z@b|HP@h)-XMSSowVAXYZ9TjsPoN<fp?yqR95WX9<Z>xCY_O{+=ws7>iS=jb3aL
zm(^}tO8E?Zy)0NVFvm_gi9FDpba}wi8<XleAv&Z})^Em@KJN*4Kjbd2f4ghfz?5e|
z?n8nTIQN1F4p+vfr{Y!mp7$o{`kfH9&w@CjN;%?Ris{c+ufK{nAo=28NuQ9-?}bw&
zQu(HRr-`5J(VKI{i>qY*3GJH=^D*iwJ#?3D&M)4%*;d0Lo4r2s-)ATTSZ?OG=pNlE
ziTcjDN27JY^1@S;6-a=V&{~YnQV2NP{FXcZkt=bZi&cn#P@ce;G8Ul}$FZe7j)_(x
z%oiZN2Xc&$c3b=dVw}yV0ms%;h9`A3zCa<sapbMC8M+xidDMMrB$iS`mA&*Ir`-L>
zw7^9;uJW21-;ffe*A>}%HLJznHPdwEair5Ra5;z?RW4O^slA5vXyoe)d;%=RjMs8F
z$UVV&d9&(BY^>g#&-HF+a$wOVw>wabRl%SVykdHPVo&G%U<hRXiyFh9o&XWcVrazd
zAL?C&x{xRTYZ}sQ+B3J-y_MRmg>f)#;|#9Nbu3mI*>i&oR5$|Jx|OjCarnk6Qy7NI
z7%uQkJXu+hF<~P5XZz2e7*Q>eHIg8I9X{f0ISs4Gw()Cm4|A;;jxk1#*tSr;6(~+s
z(ALANsf>-^aH|_Uayg?z6r}FuYf{P7x2GxznwNWn;Z2uXgA2@yJARy3lX+D>oF!-A
z&4DwMmb4#0)>2~$0``6Gvp6Ph^4qa)h~S9ISuBqilX$;<9CTK)<t+k`Y1+^|U?W1e
zca?@on53>^LLCy8=5U^nw?!4*YW4pSH02*bd(Y*>LrU89<-aeOixL`$EzXkTY4d*{
zie+;N)VTTkqapX3pFX7USR__Kt>SCf@Tj~AYULfHV!gHHu?8!8!j9ubm~Pb%+)QeY
z_rjbnY@Hd@inS^+mRu0Dkb`r4Q}C%N4Y9g$`^PZs#%%j57@|XYz*Ws#=xB0nk^*lH
zwu@5N!|h?e^;PE{#s|SQ=q|JCXd+Y{&k+`|TY3wxlu(8eb?durV2Qk&s}}0QCf0@^
z4u4RaQ13;`b-j--4>Y#pi@=-H4+QW>icW`JzwPZuDfgQL6Pm92A5Olz)c4>Q>@MaN
zgumrE3UB`AVM?BWZuWH|?jL?|=l_I)vpCe3EWO%}I+=fyX?zum-T*Ca8~jAy7$qZR
z??A?-&G*sVc00kFv40RkpK1YwSvFbic}hDg-q|!{d>8b`_rcl+(ME8FDZZoc=?$r7
z7Fd~6^E!7e=Y)-gnhoJBqpZt~SOwNEc^03TP#uTgt6pJ!4v#1jQ-8v6Ey!G9_1dt3
zR*hT|3`sSm(em3un!CMGSRdOS7iVS%{Pxm8-ShM8TfCwZiL$Rx)K(B%?&6tDmFm8w
z?O3R*f}%nj-?c0(ztb@HsQB)^_n$3bgEwgN>y<IJDcy?vK|=~Sq}uOFQKu@8M=zsC
zDD4ff)T&dTB(0W8*L1hweVMqreX`zV>)0s3kZH0vP4laxGxG~S9^us2h^O60E&t3n
zZ3M^oK~eE&Dbp?YpEm-2>9HXCQ-?YafIXqpztJeSG<VW=hV&ficjKoLeK#L}VHNCO
z@P%n<$uUoPU!Gz<AR?P>Oi5LEXXeXqv7nsUxgFSpS4oC-fXy~%wmM~GlH?VfHG1qg
zhasJg9yQ44fJW}kLdi<adZF=FLPm7T-6EMqpI_kI<FL!}x?S}~c(7*1De!~Ad(vYu
zu1Gp1WtnYBVw!w0>^)t!?y=w1i}zhhNn#>HZRzT(r>vq25UbEM(@N7ajywsEK$jM|
zdaXm0z+=2tY0q>qZt!k%06T-+(i3{tO-O6ZM=CVj$O7YIan-GXEqamy%?BY!E~LKy
z{;MydSF^p@Eg8-YGuX1<8`rsBfSk7PQk$yEeW3goTH9YD&`%7t^@hmgR>s1NZ7`Tc
z&dOGL&Mgw@3lbILcL!PF7nI=?D74^`R*_Y}2|CW3ts%aiz=kP)!yN1<g^-?`+XY#W
zMDbg@UO(aT;zcP8pZS3lhs*`B$z?z|D1A`+@->lM?%tBj39R$uQJ{@pnnzMdV{A~C
z-qN_AF}hn$GlrK`ke<NovYC*~&hCPNNZ+}c>}^z%p$WIKWf*nGZ7|9#6gi(!%E=S{
zCgH}9_9Sbs*dykjS28c5Eww0)erKo#z92i(D@M#*-qtrX`zFEKD^1mXDYcKs^6-;B
zijWu4T*iR|Ou`+tj(5l}#PQwazDr+esVraop;EGdQ-S8P3Vbq%dXZZ-BZKdsVpU6Q
zdKsPm*gro-EJILK9GdLuLCcmRUpMgpP1rF(?GVM?29&5`;p){X%qTHtrb}KRX<{Gd
zDksx-hWrSx%CC3KX}Xt61CJD#)0R51meXH1PKB+atnwCeo;Vm>YIKHQS0gUq@`aYD
z=$9ELL)Er{_`dTzsPT;Oi*72O*vj$Th1<d*K`DAt5vqg*3u54|9+<w{vWD?<bn1Ul
zB$ckQQhct7cfg%09>>A?@G(bMq4{hiZnsm4g#^DlE}LHGFz4?glBlgTma51}mUoqE
za*aL2N-;aSH5iiHG`s0y@0XO>osbf3Y=0C3S;i&uY~F07{(T~DOWVxc6Lb0?N7gnw
zvN*Z-6bnF6X}+qV7YoZPO2Qzg#4+`&V#Dd|11P8yQ6(D+<)qlYbJVYg6*~Sm$TgA|
zT_98sre-2Y%PNWZLs_aZEYTDHBTaWtPB*V=;H+_i*Pt5Lj(eR(dHr1&l?5k~{SR+&
z;yPEfJ(8~k6-|`Iq4vw^uan~^3+xLa=**@jy`x`AwBD*@NFHhS240&C&TqjHA@=N+
zLy(@9nscYi|7`sEyLVMSTwQ1kcS}Bi^`6ew1`YL>U!dN|?|K2O%~vF+Q*q03wuxL`
z-SVAb;}K^{K!ZJ)H4u`7sM9Al3RF<i2UcNv&RvKSkY6q(7v~?(@2k_u9~&|QA2oMH
z(!WtHdp2m~r3`~(f^}>ikspnffo0it!9|xX@l0Xcs1h&ZvpB|s;lk<mR}30j7j>J}
z{mzq>tG3@FrHz>;$NjUB@+*lp;8$jGxRCd`BxO0IsYB8d<3EKTn?05}=#Na;5^`*T
zoJk)T$R9RPm~Xrgv|_|?P&ES6BGo~5IaXtuS4i}DYs)#JGBEZY=i4vDiZz7P^;L@W
zSV=Kc@L4hZ?r|eJreREYgv+D)qeI_{uMWvuo%SZvyr(Uehn+<@s9-$eLGh1oWp^vP
zY4~u!19EPA1I!2KWHLb!J=g2SeN7Ucx{Y?L@a{zp=Z|-|4hEU|90@eLm+HR)RK7u-
z+Dhg07H^3-l6#i}Ns0lLf`zDpVtL{D=i`A3b01k*fhijc3=P{<V&n-Hoe|@)ibJz$
zZ{56<Z-3iRwK7+X3)!JRnc;`Ac(5!RPdcAJ_J`GPMfb23CPV+jkl@(jnZ$mIag;vS
z&se5Q6qMh1IyZRr(g042u5DMH!&TubaJqSIS4^iMHBHa=a|KAmoFhj2t(?S7a&mGd
zR<Cz-l?Rf$$Z9olf5tzl9^C&EROj_ghWBr66vQcD*CwY(#_#y#KEvOhVU}C`%lR&N
z=2gXfIFGCgoLS>_iBQt_bnC~j!sd~J6_QCydj+g=Ed^ra1Bz-YjA7{xfG3Bi&y#-I
z8owELQJmQ41%@!nQ6uQWH9Nl7<!79WxWQ$~s$m?Tan_2Amu}rlO8X0lJpIbFW>i%1
zfY*~tmdr3V*rHq(b3zr(_tp&Q%}GuV<al%w(M95_RK>IrMKa2Pj#nXVcEa(kx_$>)
zCW6!2<W5=W;{Frvx5m^}^vL|sqyfIyJtZl3L29THA`?<$t33!S_OEZ3)2pi0KO%b>
zyLHeW?c1(K@a5)jZnrq}F%u;*e~tELU)a6ha0Xn5oziYIA0zhb&gm6>Z{G4(kJiP*
zhH*G#^8@s-2?ssK--_TI-~R63wLx<EW5W$g>v{WE1VH;72iLD<RgS#*M^+vZ=!b?b
zwv*^4JRZXRW84C9CSy8l!1MoTGCX*iPQi?RNv7I=hQ)ErH2JLK%VG|-PglPd=WhL0
zSdmhCPN9pDo8Q)GuZ;;C=4~>+B@vsF3;%0%8$X9FO*Urf%WQQmF!<!-3r|-O&&}D*
zw#P$QL07ntA27;$s86F*t3mx}{Lx-XX$8H0;m4hV(6Eo=gJlZ4POch=Rp4tg<oJNQ
zJ#wgz4c;15P+0)3q_8*ItPze34YoZ)T4K=x^2Q5;tqG3p6HfWUem%x>geU9vG}jD+
z=!q5&I#<utwp$i+!0%0cN@=4I<96a%%rj9Ep(jeZ6-7itVK@)IDaH=39v4`WXx)dm
z#bX(txL>(SXgnH29mLm4Y2PxZ3CGH|JrtxhdcKzA4LS%>7S(>Y)rPz;{*Iq4vHQCo
z(qMH%*NY=ea)(cCiR#RzX0Lt_z-el=zMuc)MV(Kb%;kUY@Q`8-U^{n$BCEtu8~9CH
zaO_kzKSiQ88$OGxwHwRL3o|&YL|?m%riC?K?2_uFHk0TP%+W2KsfOuDt9P^p+Q)+F
zFJtykE&4o90@tf+h}%mnO2iG3h*7}I;W(`L*pGrA(`q!eZfQU9rVvg>xpq@eDCwxH
zG{fp6{>?LSXv8nJAR{5xq}xKa)wVGp9G%VjWGu>{R1@E)ZKTA$@~b6M)p&g+_NFW2
zKu=bQv)viUx;#fnOqYvqI1m~7<ftc04wqXAzY94jhg-trrrhC{)g5~WF%*^5vaA?6
z>Hg{A<m%|yk1mO}4WEl-FCj@;?+is9BzQeBPge`m^hz8h3sZX+CXC8P64;HS>2H2a
znT%EhA`amY723UoVpVReJB-QG?sZgdw|n=N4BMaA>Rr<FEitg>qP&c*l&zK5g<`cj
z_};DhQc(9GMuzs^{Y_HH;X$kdoBwW0sFwWJq;uCFdCs(GCeP|c?XK9O=Ty4}btHAF
zY@H$|&A+vMs4pjp&1fZy^ijNWJ%rIx{@bYnRf7w|(H7HU7;U?Z{t03rxVe)yj`nsJ
z<e8+#mhqH&K920vM)%1|rOs{jiv+B<9E8myP0Bt9ajHkH3R78^NE7K(aPzPnSS`NC
zqcx39DLV!Fq6GW|^8VHC)3zyS()eF6!Ru8Q`^h)4o8-!JsC}ikInxmz%ZR;zuqkiu
zWwk6OB`^K)Tg{N0f+lPQ(h90IuY4$EtY}o254L|a^$C#611_`M0j<^f4dXAt<z&VD
zXZNboML@=jLE4~QQ;-K6BKWc;``pHBW8RTt8BwTiH9CyEO|+8UV@zIq6i;n65}b54
ztEUOMeKM&YviI1@cfIkp$ahIiag`|y48jl)rb8<(WZ17IKlZvPkDf(V+yJ9EhJNFT
zkuy-4I*@JULfuzBwlIsxZ-x(5araw1`gjRTb`rC{CvWh~NkemOOi_4obMGdwPHu+s
z<y3YaF(agGMq{_`8}4ag9-JH=IT@k46;+=^^DEZhEFT+~;gl}P1GcYERpk0CsS`L{
zQ)SGxzUOKzFjo_zOSzLnpS0Af|6pdL!%19_BqGq_=1U8H`H=Bx%0Y%=8~F5P?V2m&
z#W;((*#jmTNZ$XrzANFrnks#f%42nWmAv*$F`+|QzMSb5;YLc`k*FyfIBBmFMfI33
zGJogA-Nb-c&gnji-kRK<+_DHu^W?mIPL{~(>YIrgd{n7B_*K{b7|rgBtMAdw|0u3&
zK1SjcWj4xmb3Olav(%;R$Ics%h?Fy7e7IJ(mLWY_maF=y^xjm?F~(LHa$P7JvuYq5
z*f|(-o!PI(1<N3}x(^sYHI(@;Az}@}-#p{x&x-YMJ>zb3IA@w)BnA5`=58+&1(n32
ziv+1Kc-4@B&pJt6CNG&g#`A4%Q6#0Wh<(u`h5=!%l(57avI}e`tSmnT2Nbc#+^la+
z)1Uu>ah;5whnq~Cd%bYksiKjySqZZ$(Mh!pc18qdp{K+`zF6YU$A)Vvi)>^J)B}}@
z&@NudRr(^FH8zDbnS`~L)eAdzEJn&66)GU8@&ngDk?FHJeswsUuG>^tRAv}kpzFk(
zRB8&G97$CF^`uP1cCg5h^2MH|G-5>+rap7>YiyRqz2f%G$F~)Z^Z1e}#QbN&v}v_S
z&eFT695z8Tb$H89?BW60h*48=%IY_$psDw%Tu2VSNlgTDK6*t*fb;S6Z9ZN0cl&_q
zQ+R5{vogIpWUh}9Y|>rw^dfJ)dZ+33ia`64F0yr3JVVI(+UJCK>ZERTg0@*1XE452
z1x6J`XQPAPgZ^PheZYq9)J-yxZZbYedT9X?#<Y)|Pgjw2<_=H=37aFQuFTVNKy__)
zGK@6ELmwAW%gR{Ye#Z6q9=d~w*Csl24J39!6nO^VKk=uN-Y*Kv-Xh2Nce){P<(;C`
z>VPA*RS3QTmBTKDE9YuqPbfWIfeLpe*ekDiTHGyJhhu9gnR=e_yqG^g>@82IKU3z1
zjo|Nsi%mrx_la`hyeQKO?GreQlG78LmVt~9m|dPaPcHnlo(KI_*07++I`3Dx$Q^E_
zGgc5?9IpG6d$s`gcY3;^nTE9ao6VO~fw+94&!v}GCM<<cay0mEOL%^KAIqwUqeHR5
zTQl53H4~^CJTi{i$19Q3n056WmB(My=lnsRvoLuMcNcH##gDHaj8=7(EHz+Lzk|X`
zOiQ=hUxu{#!Mq1{7`^#pjG`Cvl&6akQ9xiguta>A;g+$B(ZXU2*vN?Ee1E-pj-yQe
z*iIl+<D$?U;npr-R@`8bqi5kQAyz>6Ogc6xaX*?k<M;vg?s<207hfmrrZ09Qy3S69
zVBIX#l7(T`^4Wb}8wOUlDB5RS*p=rMgfyJ%rf*xOojhN`k5rrCie}AGMyn7qOb+fI
zr5beTW{uD{EMLeLGGkUzTHlbMso&T=Cg{7F>f{4)mEVw4Juw)wdh9Q<qF)(`McV<f
zR~tO;F(;7)cKJ+ih%FvgR57y##qV@qZulxTvm$|X-Ht;8P_J%%wYC~U?lW9pt+)T<
zpJSAZ5c6SbI4Mypbw!qm2#+Hh_g1c|Go`a0pU#Z4Z9!YPG&g%^)Z%WCw})UYq_P<v
zLbkEQ@e7%iq!(5Em2V0@4qQ2($X0NcRHD+VtoWC%fvAC2uTE>C+xhY6e!5ycdqQ$;
zM`j^;y&@OJ5EM@>X&klV^I{Hb_+P^<iC8nbN3-<F6tQvzKUu4a5@kG9_0^aQ%!zwC
zeje-oip_T7UJ9xWOiY3>ksB&qNiSckpq}sxgzu^~+Mr&$DhkGHM;%#R)Nry}M&j}*
zUFFdM{*K~-K=((zuN&eEmt7PF1of~VKJ$P6^);*6$kqABIp=Q6I-;n$no*%2-$h@o
z$b3B?J6MEbb??LRhorDx&B9P_<h;Z;G<%L9l4!Gkad8M^8ht-Gs3sT~$nkaLGSa*4
z3a91RfViQOR_<^1#P89&zG^7$v>*8K`UpNpqw&ybAF=spS@jytg`?w$&BAj;HgOR}
zS%=OR#B*Hdv1O3n^6g8)uV2S59`YqdIkmi2hpE72F6?|VySHVcI^m#tIP6(p$vB6+
zyxme0OOc#H`G%gwiZ}%Qobk`xs}aAe>K)6UeV2=0)LB39#b|&3=*_;T(AQ8A*QIA;
zm}O<8q?`zfEHGIRE{`xFFP~Y6`6WrO&IaZrgSg_e(qm1aOsRLB6<3>o%R(TD%=(+`
zK9{)lS2a>;4L@bVXd;qDy%weS`<3tJqb^4qHMxA4w`~2e)D|7+xF2MzTzkYa<5zoR
zp(gaFgbt$;b+kI39sB=L_uuhY|Iz<A9xfSWuOzyxY$B9>5!u<YGs_ktg^<hMdnBWf
zy^732wo=KaNJd6xlp^2rY`@;`_viNe`}bd6m+LtmkMlT>^El^z?sx5-3(q+fo*S>6
zSv$!VKyA$-&`GS|S?^`y2^rI<)2c=>t8ttrqqI92JnzrsG6_6D>kxJ+7V;^)3_CFU
z*#@A(ePWrZ$ILx(A5a!tLwee5Bx(+gVs?VH&EoFl;an$jWrYz5d)O_N<$Bitm4NkK
zxuQYf#lO#17HvnLkQ}Mu3SGRxgGlkhy!D(~t6av0_o0P0Ve`0wqAb&%=mYptxwgB;
zG>;gDslv`5XL*FKtvfMdKyIS*+&@lFZsOq5CjweSg@+{+ocFd3Ig|?QD7R${s5j}R
z#UFP9KwF>IP|sG-sU}&d;MH~6X4OFGZ{R$9n@)T*^y1OAw0V5RvZ;m_`-7GmerCs9
zC#P<9-QdS}g4)TnG)4cqZ}R4V)QQQxTt3Tp?msf{KPL~vMX1|Om0Z$yW9srAbb}F-
z*{Vs|EPesRB9#(OVL006xYtxbwkDhFGg67=*tM6Dk6wbts#(vDZZ$7iT)~|#+@7aA
zDtrS6GRQ_R5*$84-uSgbi~JSwIx!C5?!N{58sgR$;d#FQaV%${2<_I^GlbINQ+xN7
zr<MJO$Ai7Xg<l&cJqwFoE2nvP)cerJbOfDOQL)wT`42%^9VIr&v2FbRQ>W)DJDGY7
zT_7;f+^TpsMu1V0>pQ7H-%3NKLh)4B|70?^9*OdTx@DJ-{Iy75AL)%MDFEZ$>PZ%T
zFmv#U2xxfcb<ye%BvH~TI45IzD6<nea<K~Ryl(VDWZqQx;TA!Z^mY<vbK?76!OEU$
zbxL1rrOHpsA8@gx9t`Ity1lfSQ&sBQefjn(lR(#2ykgPRmx)h|<`=iWF{M1!4qQcv
zo>*7qyuQ*AwljU>qYdM%AmqT2YJxZwuXh%4(BAKE!=D)WLSmxcv)tVIPBo(^zBE{H
zGzT8j5$O(&?BuaO&V<#XUz<CYM83bk?s}Lhn9WSKMb9OGJEYA>()O;80(;m|!gk22
zF&#j{;@Gwz0*76abxC;<e|6cR{ff7JFW&!XF=aU**qP{uKXb&Im0R8HWD*0*O2G^9
zp25`A_6BWPTwcOq$BLM5rSWzPEm?)-6W$L3aHA))4WOhI%!z$3Z$Tc}X#Mp5Y0{Hn
zP}ENej1?m9m{?7kQk=EaPWFwh;KW!}Y1vd$zbgFbMdTXx<ZhWP|M-E#D*0ru;4Afz
zrGF|ou+14_7c(h9k*RVlO%GIx-M@#b<usNte95{#m&V<XdDqZ=;&$%02RxrhXm8;3
zuBYz8ImncXvR&3ki+}hs^LYNF{fGlucGvGz(eaSnO2P9vUL`=-+M>V9p|7`TIN#85
zO<Q^-wqjkC@VuVN$h~OsqF0}#j#EQ1NTXJf{L@ok#gVz<*F&E#ueN6h2)x8TT48Wv
zB*^;R$BV>?DhhD^;SDrbVG+ec&aawv-xoCeh{q)x?|}NnJDdoDo0Lpb<>8dcLXvFL
z<stTj?RHWJi%#Z`-t%u-$iIEdg_-OvC!UuV+ZtNF{V5)MchngKCpG+j!AtFj+0Q`w
z-8R-gIz`!<q*G2|##roRI~#*M)65t^R#QZ25-P|!$#X~jBO_?;su<wY-zvq!+8zV2
zVGVE1h~+{Fb|fT&t6e)yn=Fq@A1}QnyT;Oz5;xgfX3^jdRc3<W-IMxbuSoaS<oL&H
zzM4bF8uRs}mlVq_hi{X8?Ne}8sw^pNw*Lm2Ga62esrjVzAxG1;b5{~(ar7zba2pN~
za(-WE<We*auO2c|Eyky!oVGYb6vKE+x6<`?>~r${-Ti<U#Iyt~9{r?^@pL%473&@p
z^h9DYxEcEwtFKzi+<Q*1k-@1tDe>Lq1Osm0?Uch47YH9m)IZ5)8jdW@?`WQ<q`+xu
z3f>=|)gBccJZ~&Uy2ID*vAdVZPBfpkBt59#H#z?|bUYF;ht~9P*QOSzTx1_M*9xf-
z4dNslEQra)^mv$kX<2#r`g|Sf^&eB1T}JMox&(A&;wN;!b1upE=Mkuw#7zVH?&EH#
zuT$8cjYP`$7I)+g$rZg$$x(M$7uMgpIw)+u9fl2h`L3NwCooNcIjJDpJu5ggh%-vh
z-L;iz5hFamc)EralUE1$gJ9-`3?TeUD_RlOkCPpnn-J#z*vn6U`4}4o(8RrN6^k(c
zct$bF8>)!xpVCN@0uuB6zTyo7dPOF-j1!dyDtbM<)MK}(^Y(q%c&)iS#itV3BSfM{
zhO0Gk6)8&v6__+Gn{IwNMeyqRe6&M{YvW~QQ5vf1#n-(Zgk6ti!zR#NmFTIlO-s2p
z!}sac<AFh4ncgdgainQUZSx^763vR%q*Ahk?_}#97q->Jr3JLhC6&b`37uz{8{5li
z6VSAkL!xAWIyh0>EHZiRZmOrT1HIXmg_~h{hnE_K`+HBW$m}$F8ZU`?PyDm17f|DF
zafy@;jMgzCCr*C~&Z9WG@x(sT%*&P{1<^@W@)SYD)It~92-qA;c&$26vX+du7B&n9
zV$BI|k>k?hD&VXWuj4K#oxFX^kbwMJ8*uuzQr<!xLaDQpO3d$S$&d$=7nAz|e0}KM
z)i4E8&r^+TkrxX*KD=a3448@!Ddi5svKNcb9};{;9hS=O)~x8;#ywM7GsfMnxmT%6
z-$fBOuBJ+GT#2($P9gRxAkuUP56SD^CpX+tG9W(n0FSqsvPy!K_xQB0I@c81L#xkH
z5qr7*QcQumG*5wr!qu~_;_2rOD6r&;<h;Ne&ZOgFnR-z4G2gh<Ck!g+Pb{N^5>!r=
zZHR>XJt@Ka!o-<=L-4K}FEcg!i75$B$ho7~dBa|gS!G=9S6;9?b@m-#h)Bl2X3!L$
zJkPKmj^s&y<8COBg8!@b?ZO{1FflWF<KQ%vr8ZF(@gUdS5o?ZcReVu`${Un=^wvZu
zf`G+MKsu6!kzdb#5wBSAMsquCt<B*W3+*h&>oVrgp-4@rfIEp#YB}=zG4JKdZb-5m
z*=`GeR-n|Hc(de^(viZ@Zu^g~wXNi$dd011atee-*}be<@F)&*g;P@NaWYS*dr;(;
zbmj<UJR-crr*+8c^S#1WmMpZm7@3j=;i}vWm$f9D@N7A>OUxjpMuSral}vq|N9Ih@
ziIwK$T3jiHM;$H}n0ZUnj^7}Zc&uaK5_+un@yP-avq-%DMAsyS8)+C{Goh*Wn8gN7
zjOJO|yC*rkD6&^&b8f5)u7sd^G*$#hC?p}ielDI;yEjx7y-2g_ESJnu74)*2!s3ll
zU-$dp1uG=o!ADV0&nU~M(c$|8gB6L{Aq-Uuv<|#VPGN8}V!x1Rl-}FtUYXQ##Z$Se
zA_TGXrJfY(^XiTG%!KQT`#W8fy0$DIl+IwasoHr_MDj#P=N|zL=S)oK=zin}89%X$
z6t{LHjo(|_6g5Dq)V}(a_6&!;DgKaf5H55G5L(lvq^F=R#NKBy(DMW0AvlJC3EIZ>
zkdG4+idXYU)-eGGlPRY*r)uNa2WVh8atL(@D`Nc&C}xrlAD3=ynP)p6oL@i6{=Fdx
zH{R^+9sO2Xxy2`)wWP$pF{5*{SVs}B?f_i+4RT7s3)&LZT2Zq=6nKk;VIy(Xp%5nx
z;m5(|mQTpohWIvCbRQJponCE*rn;jgs+<I_5&cgRHLttjkjIX5y4KBSx4H;0%VllO
zGVMXVE@-B{(UFpfr?tMxe0uZWapeLuwK1^~|4;b;Q$ox^q`=a0wo)6#dj*`IGl}L?
zKU(&Z5>jAXrp*W0a=ExRQv*WGhgpVBgaP8K#NF_u%K#j=di_`v1%%lrIUPpBN-=6R
zZpW4awefiH`<O3ei?#*QxGBA&=E8Z`Gg5&f596*bCDgCRuS+C}bN&?ZbsK<P8VI+*
zLxU23YNB3QiuzIM{_wOUgLlCMY}VZG<@bo0j4l<4j765q-V1d1^cO%0P{c?ruVtMh
zTr&(L*Zr{BEQ;ENdhRh&*|vm|Ln5cJGFd4`!J!^quib9t8h<&0)XH+|<9*sEMXGzR
zh0ojy|B^>|vrFCgpq_@R@3w;QqI)5oEcftGia`arK{x^vClW}lB_)&kq*+w!ym)hX
zNN{84r}_N|;<Az6i9<Q>t$u9sKm_r6K<kX2kQ=*Qz{96ji5NXD8P%?*z>9hEqx1=q
zQ7#&r3=<wc$;<fV2Kd(0ZQ09&Ko%Igt(=xxqJ76Q`Jvd2Pah5{h|z7I?aNl8F+7`-
z|CPcJ+o~TfVWD$a_NW!A1%H87V<&ph!PStGRD?s5&dsSVnf!}3&FTB<i~X5h>n12;
z_i-18EoR)jt8)|6FBp`Ip!%1TDMW0bbSg1oHa^GBioL8WL1cUtLAaf0>K$gA$)aBI
zc{FWp;>vhOg4okcXh!?eWvSHnUw4K!DIz{bVF#N~G*SYQwHIZh@3mzL*{Jvn9M|8e
z*t+?1SNNIz<v;*<_lrNeJ8cn~01PPLLoM*|812=4K`m0@gjRz-(w!>h2k?fkvcK^X
z(R)AWz*+{XxQn7U<h!zm7SD8EPG-}L6*_jdJMaByeY<0VSA@UU8vuCYg<-4c;+G2U
zMXSVA#%nxhAb!5XDx(j?EapUU6+wN1)ES!udPQ9)5-Y0Y%^l@SyduYS?r0W7%5W0X
zWLSaZN%iP5yFC5;7BuiRjXG6t9jr<wF|58?g6j*(js6#BODmJYDW@qZ4O*!E)HjD8
zFH>qvBi+2q!PuxUo1~!KCEzweudN66rF4}6`;~y<KuR(+aX4fA9sG<<7mqGxb~GKb
z^xXZj{_bt(_1jpd;tP}WHow|LfrA1eY)Hbqayk-l6x|#s%9J80PsS+z@Ss$E#C9l#
zPuakSnOTvBylnK@SZpxM2`++nh0P73DLu*W$xGG+m&t4iEH@Asvu1#32sS7&;Bc!7
zPWu}K4Z@3$mRwh66Am)O+K(7ck<FKFhGYj%i9uXKJn&_eIc^Z?-%2N8f=rZHbSgIL
zbIOCaK*IgdC{Nko=v+osQTM~M<++4PQQh;5mZqY|xgn{io#4h>EHw1#=3wzBZPZ)V
z7)i>Towix4hZbEE$1870j6b66t;|||1NvuG3UZbb(x+gXa8T+A(K}bNIDhjLz~*Ds
z%Cu%#Jlj4q)9eFeOUK3jqeHK%>Djo7`bP_2uRM9>Th8}>%Kg-z9WoMAN0^G`qZ1oG
zg)4IIfSZEOc|;%O{o6OD-m=-rY`m%5cH~P1!&L!QBy#Z>3mP5+!x-h{T4&L9GNi&$
zLJAD<)V##<CLLQUh8z-SXbITjEh9&<rm+;{O+YLwP-Lhie7$Aba1>I5y6oe+c!2P;
zU)$V+ot83`va1&ASf^V)>*KsLRi>nsq?t&trpfVrZhFf!nbXnvwo~Qf#0pr-SLAR-
zLVZe@KT)z|O>ZGAO?YRR{C3w5jwih(4+s$`ZM;1R5M<r7h;WIEp)>E#EQ#O$A~6B#
z1=-^B3Q;TU#*UQ=V{A)oNqG0&rUBio@rBgpESs07pYg<yD#aVcDy&TH64Mavi^=(4
zJ)A$h%Nz@U;OhwMA5#=i;94i^w~-}}&F4`Rpl~;)OXQ8)A5tU~AiZw)xWDAt-w=`K
zW){%Nr|H%3fb$f|W9+ez)NaaK+O0yRX_z|kVUwgCBDOecHs0PdXp0!JxzvO5wmyyV
zOV*(&+)<tgS_&m;l{EwzsQSBXj*{Ok!{%zw>nu%u1^}3_Udfx|bVP)8O4<^S={OY>
z+7L-499D|UxN6;b4-vt$QhggP8LiL@fHJuq?!&eWt#S0B7n1HV#;;Q_IM?QQau{IE
zrCKA{;vTH4vYUMn=J@ia2>_~cRIZi+l+ihTPGo{0tav58H3fUi^@?Bnxwe0f2Y@4B
z2<VXvm~$TYM%oO}xIl<=b<D{9_=(rSjg~0ruICy$a*jXfw7m`&zsR0xF0XY$*ORlP
zVr)ng7Fs{8RfBfx3|d($ncGdBZ>Ul`CAJRAxGtIt9UjVVqEK}8x}Y~~6|Y{~K`cjD
zTteWIcThe)Q-uH}rL(X;>M^5;Q86gv6JVXej}%HArD8dK!N@KzTYCZ^hH_SdL=iNY
z>ZFSG%W+tX*kbKYA@rzM+hyW+JuVkbQf?sqf1DkM=vN5T0z<HUkFz8B6us8_Cq^I_
z_f!?_`@|MneEU>i;K}(Q(8G@L3F)Q6?{}`fr4LSEw+J~*Pr3HOgRK1i`d>N$2+eKn
z5dZs$=F0xLY(f-L!Pnh+`ZT|*fTgHDl%l`^1-Xu`)K#3!6jF$6{rc}~WL@Fc_K_3K
z*|r|rV)9<5$X|JVPOb67470mcy!JPqy0&DD)mg62RQIU7cwZ>>4v|{(aR@UFt-F>a
ztCSv{FSL8^^}d7c*!=Vt0bCXxg;wr&>AKV0-2<+nT|ne5wS4QDE@t36px(Qcu|;a`
z$s^WnMAJ&o6#N8<zJW6nU7QAzr}KS4*DEjYuHS@^uMh|rNc$Zd2mJEUhs>+mo*k1h
z2ZK~5n%(t3^A*XNFG-?9PLJ8%vj)@!d(D`5_)_Cl0R<v&IBOGReFe>;F|ryDlXEe@
zn*d>0WRYP(e_0tLq>jKiD)`S|tjqLl-=23nm#t5TxrB*z2C|}-7eW`FusyDacI$rZ
zS+#G9>m2yU-d1c#2OTIp$eik6GU`j75i5tXif0itE;AZ8?^ZqPBak!&Ey7<0Zb&|t
z(%?z~C5oUKNOn3qyF1l9ZZu`K*zqNo?YJT7#Pq^a$zX>aX}{z;q1_AOCZ`L}-6}(R
zi9f3H6qqQORjmT#XCUY54sdL+vrz#~bG7v`ziEALC)#wUZ1c|We1^C3v4UBm{B8#+
zX|q0Yr}e6Nz|E;OUG`vNsew}QYq#|Ch%6ru^TGuwd;vN1i%OBWH-`jN5WvVuwMMV?
zkDkAiwLkXj8O5pL9kQKJIHT|d90(KOf3mxAuty-?jmztb6hJ{9();X#K|Z=V&T}qX
zg!zC4DkmVxKlEvSM@(y?A8}$8F(JKr(%@qw{?8Cz&Va(PfCQemdq+KRx#VT0YPTXS
z(E(Gzse8~|nf}`6`BJ!nA9T$IPS?%h*|IZ+tIf`447@ukK8den)MX$|<#VB#23Kc`
zvly%DRVtTyi6+m!e8PtRg>I-S;l)C|JkO|ah$`f3bu%~20l6Nk*iVgDr}83|MjoyS
z@YsI~WsUA);`V0nrZzkLuI`X*q|pVo5<=IVPs!umVDQIvm!6Rqq`VZn^(ge_+HfzU
z^r_e%{FlE`=tlx}Vf%R=6BPr@IkE>XHo_(`_mLs!8PsJmm{tMgYe^{@^pvNR9B06Z
z)1o&j6`NOT&$$!M$bc34{-Dybe5C$4UkD{_HLL2*2RZ8Y?z;4GMZTqoi8KICt-9j-
zho}YVCQZ7OeA#{Oa+6BQ!0d{7N{-_;d5s{^p;x8NwAtEhyy2Q%7r6`v2~>HBQ)dIx
zlU#2?B6`8X>_`T17k2}hPr}O6kR+`8u$i~h8ga+_<f+X}7f(Wi?A@qU3cJAP2CuHE
zyscp2%)<^=P4G7dzwncauVATpFn^G__eU?&@AGIR<|ZtRxy}y7NK6D86MUKJgD_1T
zS9+{CitLP9@8DI3nmi@DxR!byR!tm~W_ic!d{rkm7=Ml8lj_@Wx>9~*Y>Dq8-&Z(<
zLyzAbd|5cx#DOd>Y+^GbgXi<0d`mb7*As@iP^CR-^RpFDrMw@7>3`o{bVTN6qdw2q
zJ=tnVe;#?@`jq4}%8j;MU%Gsq`a7`XoIv_BTAz{OPdb?iO&};D@DZ8EfH>?OIm%s{
zMk<SW886-4Piwy}ndR~I^-Tn!x-suWO7snmEhrHoHb)<q-a$XU&*~tZa5W_cTRLkn
z@tcznN%k4>K+JBul(bnpE{w}^{ETvy5>Ho6nOz66n?n1ua7x9%Lys$H<qLU~Yl#s{
z{y2E8x(jYgDaK?h!~#b7haLq%WyIZ%v~ZBNqU#!ID?X*@u!o|Bin%IzF*UQrMTmVL
zjv7d1_KKtV45oV!B)#r4m6BkagKVsV9*YANGwiJL5~74^mpl9M<V2TPi&J8d2LA3^
zxmhv4Gvz)#?!1R>%Z-@jHWGEtD=UrU(P3DuR2S&jE6mPdgjJuLsW1$3_KM)VpU?7>
z7A|)$J9p9f^`BDANn9t!fFXe-yQK?V^TR6?NavyUVU;xP?!ec8u1rMwTV#_IS?)pU
z;s?{)c|Gm1unUoxC<^~nJJlSlcC|D|XuMg|jlufyqAj|FiaOxBV2%-AEGXKVJbILt
zr~&Z3`1n^np9vB%9A_<j1CHzt0^hVa%~ql`1`M2D_3@unJyDJ5IPsL|>vp%aY2s7q
z(UI>VXQd1b!il>;c&HH)nTKKSjW?LRt^d^~FWp|d>0}TlYR^HT<5H)#t@oF;cL8cQ
zq0|%)$*g?)?hiA@#<skbZid!P0Mg{Iv(71!zVs)+2G^!}0Ag(X=5=OPO)|p7v#lE^
zBQEgv9AkP2b*Ne_N=j|VI?RPVoFRmwfB-etI<Rx2IZm(y@Vi?=iI-x4`$s{vzo1Jj
zKliLj(mip<?Wx6FoRwA7qP%X_C^-=WRaT;`R9}R+MYqej&S=K7TW-oD&<leeh}p+v
zCRnTFn!!0EVZ_f9$T+*<)Xc!^c|_weR)56Nyf9Wk{Mt&!s-_D6Jm1SmyjHrqhCOEK
ztlg;tPufKVCT63I7ba~(O>u9B_%QFn#Tmji=YhP0Wa(?B1<|{|NhBl#M+R>>`)<zY
zT=ogVDFPk^5_!GKa3)DEW*4on^^-xk+x~c@N-HqO*3ftMF6W3|+<>x(82z;$q+*;n
zIE9A`Stx1q*re|v{68obwrQXt*OV)Ey2pu~gdijB<eRKi=!bC6JVLBZsFs-T%KPx8
zoP|!?v0Ck9|HB}sx<l(pyRK*z7UA|rQ#;%o(Wb6=7OMz~38rUV<JwGyO<G*CmI>V$
zFL>7ht!b9@J+I+-HCRZdxV3j}gC_Vzv2jWUV-C(r?J5!+Sw3z;a9m)cGc3D#<W%SN
z7FkSBJYnK$T4IGtXukNL-hm&ZKpg|ELerIQ%B-61D6}?>m(*g5;WszFv<@-olQ=o+
z4($U`;mgn!L-e+aO}Y;J4g=%opByJr=;nu)^C4FAM+dIuBe0vNHwO|BqR!Nfo6zP{
zbK83i4l2h>F3DHTYPaB$KfyU0eArIDwMUV@DwZua*2#<bHL3MfB9~4)knpmwYv7IG
zf5(j3^STNujllkJ-<oal5-%Yf%IdHl>@}3iB`MLa3<$2?4KUCh4U3l;@r8y+iErLp
zciIPvgTU#};M<0BPNZn}g=3?rl0*!}Gy-5igbLimB0HTyH-MP?s@B@wtG)=?RD|Q^
z9mw=`wqA?fd@a)ZI`7(V#@U}kg8}j^oGU(Egb#&*oWa<Z`;ufILc66NsxM{R*=d;a
zRmtW-XVVX23kVk1I2H7|cn|D~G8Y1#;z2f+2K+I?ks3V-*<#t9$^Q3c3j5lppb5Sq
zr@=>MBrn;BdlXoe;!AF5<#N3KkmkA$K?}Cc4AR_w6%_&4vtic(rbF!^PVS*ntKf}V
z)cg)|2z;C=b(yNXmGWmgywJWh1s-8=L5&7T11GqVKiWDn)vHE@&z5JFyz!0kTU(LZ
z<5|%7YR!QSj*mC35VveLxJzf046|rS#$0v!LPKLS`v(S@>$>x(?;Ae)8|;`vLA(E{
zt7R+0z9vc(cj!n!@>w8FiBRupzj{`Vth(+TLhTgtM$K;a-T05gM1dm86Z-BkZmwB!
zrD59fcB?GMMdqN5g{7p2RqxyQH02-{u>$TBJ(euWWGmSFE>^)2r7HcPHBZ2zo)acg
zdmBi%&Z8fO30ievZFKz5yR|3CePSzt6-f`hTM*LT4B+5E%fxM>n}L>;p3rJ{_#2SS
zhgA$5r%nZqgaXkeXQDJd8&2hQT41aj5DiGdi<J3h_}qh%^YZM|B5S)uyiPYf1-@<E
z*a{!NPib)<Wy?I7-_nej{-|l_Bp+WF(GmMaZ}cnE9Ign4ux?P7OHsgu+y13RsOjX*
zY`Lb^z~eb~jdcf07ZM5*bL+eH`o8@LefYu-yuJ?}eD5Sm*N6OuTF#63Qtb}fgq<e#
z7xyWB&*ob^#e85ll4f^&{9Qiyu6BERu=!E(CnP{d<8+s6efJzS)j+_rAjjFUM!-gK
z5vlGWtdNFKmwcU%wn(cfN9P=~3McB4*;~-)hGN%!7LSIu7oJIa`pO7dPWCOmD*itF
z<J$;d*Zhc85>IOTd>OINTrbHWJH^rOkL{ntQi^`xN9`f}%_Q|H1kQnXpbPI$9P^_n
zkx-#<cJ|h}>>9#0?h%tA+}P4DZAc&OLi5_>>$BIHZFF~=GX*GA#tME?@bF;|=#1nE
z&wL*m*yBD?V`ST<w@RxT?AZ4C({iQBxOS@?rG1a2yzS6-z5yle#CozqH!(dBac*h1
z@^$!=m9RJ{Hh)p5(FsmF25k+Y;lo>${T@I}sG_;Ku|btJaz^O}EdlB)fMk<F4VV87
z-wUgcBXFo`wb#kImh_@JF`V|IQBy{qNsdE0yKPdf`p4T?8p?ltUA82cadOCQ2l{{_
zR`l`H;ypGBS9-YaW7VD^j2xYmG8meTml3CwRT&dRr)iKx9EOqEfAs{;PAGc-?k!P<
zJnMTehMX2Y@*1YNNNxm1#n9bYpiOyQpIAX&5hJ;wv_=+!Ssp6A9gekOZsq6a++b*D
zXXAZ61>bc|u@<?p8R=LjRV>qu_I(9O)@OJ1Q1b2|_7)g_4bb^c@~w}7dafG)mI<f&
z&afKb#*$Mdxny{$p3+8|cJRq&AaQp(qoIVQ2S*M_lrLU-&RJ!D;p63Q$8<Rh${KR<
ze|PTy;2<Cx7e2>tfigk`K)c$RkJsQ>raEj$FXm?5W#)tMvH(GxQcQcUJhL0kz8L-@
z`;V@IY%|FI@60}ZsrjjUIG*ttx&KJSYCvfk_RbksORW@4Sow78EVqRno|`eZo}~JF
z&LnKR2JQ3}$kesmLOb&~Hvs0GXn)!O`runnaK9OnZ;d~oLu_dl-V=?PKS+NAy0}mc
zxbZFsWcr<^T5I^#_K44L*ie%4LAc*Ubgt0*3Fw~uF4>-y_Qj^OrwW^q^jx7)Mt?gU
z8x^-+|4$7_1zDL3sdd2=1w^>6q7DOs#Q8KUpgPN3*$pjg`EM-fWq4^=F4!1i@@mfq
zb~OI<Qh2D`>_!mal8a|1u>+XO`SVg+16ZsQ@rn~$nSGQX$PgW5njCdE@A(S+_;IHU
zbqd)^5-S!F4I>MY?^6D1&lE=rnN&}9QYHJh1=>FIAhFEj=`mHWBcqw$rD%Bs*FFKg
zJ8_=Fdz8zxMz%R8&zl_~d<waEo>viCh1$>-w}Z0bT?ZQTAa?mjYts*ILZuTw#*j4r
z2;WV+qukctV+(*NC@9_`j#@3p*Kxifzp9R$zO24+8GVMXiTst|e_fQySr`aAueyg6
z1@yQO`kw!G6e!--<gi%t`u9Nl$CU4V>3DMLd2X6$rtsJ2?j{+$nF%Q(JW>v|GtFy+
zS2PlWR7%zqo32B~;55soQSnjQ+W`iZsZHbvUa`N<eg5G0Om3Yb#sEnRJwHfhkrjX4
zGe<l)S$9Ty`rW=-EA~qzZzDTtcr5H>Zq{|xC>b<b3>#i%^sEji$0=?heu^}qiUIia
zsq@YQHxlLGbN*Dc)UNSjH^=EruC%O4^og+KjP8G5-4EW_`EKr%EpMNg2Arpy#G^&l
zwrnNFz;e^SXTiQn;Af8ejVwNYL1GxRQ!9>Fuktr6Qx5$X3-X`EUh(?81UsO<N%nx}
zNaP21{xSDwg3Q;6ic-!zcz7_7tR&x+_D1S39s`*T_?+}2Z@V9?DXOR{dAh*1hy?FU
z>Lix5@Gbq)M^p->XCN|ad`elztCvSCH<a{MC~7hRc=$t3pk?x~HZqHrmkcY;cdd#z
z&VXiN@r>EJg&AcL$#$p8NNvWD6=jXCP5D?s2s`Avh=y?4Bv`Gb4$w33XlZqH0Q<N$
z%~we}AZy9Mzg{$Xb@6<rq?pO6v8Z)LvGZ+vA&gIUTVAA?A4Zsvs#fZkf}QT1-Oh74
z^WhI2!H+G7op}b-=6`bmK$erO)?|cmfHnZ(EmD)U8AI|!nNyEKUfiY}oH-h`Yvs=F
zqIEt0A!UKRh^cnxujUB`glw83hI31kTxBaI#=7W{drvgM@ikzWMc)_Tuu*6QN0CBO
zG#z$`*;IuWcp|%D`)102b+0VRlICvW!Wz<3R|UGj!6%r%65~J78kw=iQ^gt60zWub
zo~QH}s;;IDMzfUcjw$@6xHVNLx-vr(AMa+WvPTG%=srxOGYe8O`0mm?FZEq*Z6C#6
z5g|jz*~uIN{SLL!*u@Xu+CCrZC2LwH{GTxkFXHU+Z}ZRxj4}Ij+d#t&sdR;QKK0@q
z0x@0rFB|H2#H+Qmn1&Fdv5NiJv!r%cCw>xmZ3ScW6^QI=TMjZ{EjCydV2!F@CTR2I
z*}BTaCVdo${_qA1;yDCok1KFtjiuw&tdHsRGV?;)RN|EzY{A^8(N`1*FO6$9AdO<X
zGaN@B;k+uwUVQd4_K}Op3A2<RQA^~tNA(_|7n%+$gr&1vrp-NpfePNNhvFb5*KI2o
zE7HYN<x71?q$n^VKZU1?NH<u}^H@}heSj9`5dq8FB+JJ&t2;gh$vxTCVi;j2N1FHz
z;xQyijzfXc$37kZr;v0cAf6m(8YxZq9Z`$NahOt(1mVe-a7w9Yc`Eq&>St*4Tq6%A
z$YvT?+%J)ogP)F_!qvtn#tVI#YA)x&6_E^ieU<h_d!p`YDB^G+xsoGvxyw+6rQs#i
zjORYpL|s1G5>ayRSh(_zdpPg&WPwW!M<};}1ea$dQ8&evnQ$1<VBky~C8OuDu`%H{
z3&AS*^58rBl!%nNJAte&ZE#}`cd_u-I||n$$LQp!Z{r%$blt?v7kx}cb;AC_)Nnv9
zE6jkXw|+Sk{cx)a*Nvtf3S4~|X>HvSJG|i;dMx~L6_t4b9dmjA>=2Qd>qrfR?DVh@
zu18dJ7)R!j{Pc}0or$KpVt53kYWfBiLDsN!b!W>4#sd-J@Rpd@l^>xH!X|@6sP_tA
zHTqC?UwfrMS7>C)CvpmBkmobN@znvm+p;f_xp*f6>kBdzfy4LD6MC>)GqGhgN_&(f
zvRm6pvzK&f4m9jcLt5edxyU0Fr0-u8XPb5@4OBdvfg?ZPRnApPO8xKmt{<wth4)Op
zWl_-;hzf#p4O$6IuH&fO<(Y=OC8;O5ihte>7K^%VVv4t4BSW*3@^HRcv5T}7pHv`H
zoji8YV^Hs@BYiawOR7_-5iTXINEvkx^sfFVJ1Kl9;Nt1x8Xh>}Xg(s0y^24`qukh4
zu?xKc1T?fn2DwLy$6LJ3kyj5O#M!N6765{g?`zszC-}HJ4xL~Zkz42-xT%}DMeS8V
znL)rb3HbU8eeW!|p885e1HqECUQgBWTU1qKERX#5zq`JQAJP(&o0}tsN?q%bLAKW)
z9tgV4bjh<ZKjDcDfD?gb6|@aw??zHh_FLjsl~)^6e2|<i?h(y8TjHmAFpxrk8UzK)
z2AmJj2oniC$b!0uxexS61?4L|w&$YM?{2bG;HB}hs01w>g4JuG@6H6IhZ^_9O{nJ3
zF*P>R1Qc8S8~imB3X=tN?B8=X^Dcw)8U%J2q1!;PJ@_hAlKMxew?J6(uwNx0^adOV
zZ-vLED7(g9;nZIi43bv0Z=W@`yn|Inu`R>Ve3MjGI*WS^^6~6IKN2DOR9pHUP&BIt
z?&UdWYw8(~*0aF6Z+36m8tSE0u4N()ZB<S<5oAakSvO$IqcUOwE!wH09K6?MG1+fV
zplg}S9x(5&AJ7l=X*UXeDafvoP+r_~wsktC#Q$N)%aFz>WdlPB&GTjtz4*qfl)A3}
z*qVNZ>j6DzbBS9$Qy=vI=3Kd1DuCjM12QyLnua|$&?}@HO4w<bAg1H?>su~Q_C#aL
z52I!aD0ya&9(h13U*>vjQX|fh8*<RL(Z_kW4vnHhQMnUPrs}z?)%Lzkm0CMhFs#c#
zH>Wwe1{l}QwG+@Dzi9CqBrd5*3Ts~jLuPaQ5**wF{B~4@97ihTvmD|ozM8D1SzOd7
z>g^3XQz5T;OJi`h7U0~6rYn(Ny|NXL|G1}1NFgrmK>gp)PSY2KjwJ<rb0hIElhh5Y
z{DsIFNTn2h_gDEKOO#jup<{XO5bg%ibJ0Azx#gIpqKvLfPALuYTxkMlIvk1R7N0wg
z8mNb*yF<eLIW5<X2EI{>8Sw=;ADtQ@<+MxbGU(}OyT5q%dq)F}-a34o0DR;pVk?hj
zi8$!rI_|P7RtJYKs**#5lsOc2WdhifvO@f@7fCzBwujePlI3i!|1F6?!aoYqD9Mj;
zS0?{t4LOMe<a*SP#_p+<Y<NS&6k*?x9#loF@U?Zg23p<48usR8_S05kmV^{C(B0a6
za`q!YS0#D*+EIp^m0qFooFx<fZ8QtU&J0^7F(}72pxK~FUaym@nIn$-h%gA?2{DO?
zNLTEC_S;YC^4kUk-ywU$u@*nt{b7i;a<m@6=AAfJaA_Xoiq6e8{bmjM3A1nmuUL94
z+4)x&71&P!3oZ-urboby*o8IiZ7t}*aoq_YcJ$*H;dxv`o<h7#3y7%N3i`{nG52>9
zn!{rI1s!$QV-jQGHq_}VJ?&m@>?_<hwn!$CtKHbtF#jvn*TV@0>hg|5?+rLO#)*{r
zAz-t=);fQ-yH<1Lk`;0u5r_0`Gl!v#zS;HJity8dIB~t#<gJ}Y7XBO1=*c#R)fM3u
zGyEpyg{(6!PFHR!7G#_q-C)huv}yKBE2QR0%#~nh6Kun{-jm0&J=U(u2YNY;oz7?+
zOHvvqK1-m*{6YM&yYs@2^iA9sqAX2H(eC-=g{5kiARvdUUdHnqg6___b^(3Kgl1J7
zqT(0$x!hR(#}z1AI?%xJ4;M9BOf5NF8*RXligiLIUHiw)W=|u-dM4NT<sUP!<wBto
z5^D+L-JwmU8>Ho4-qk716OR-OFb_OAZhJ2(9x9xo$Fy-vAR`2r2o9v;JfWsH40*IA
zYuI$u(HAS3{3kM{eQzl7=M^p$oLP7GNQ`a<O1D@aAWOck*H)>WKBV>PuF32sGf$`j
zts$EU8O5x6T;WDo6QHi-;w6$^T#nNEY6$0~f6n<BX_TX4EE)akZ-FToQ$}1)g*+^K
z4mEKL+`m}i1uGp<1Z8q1B73nglB*q=RH6M!E^~I1RGpz<jrhI9Mi=YZVuWZnK@3Sv
zc}2UK0~_f1ihwKdh#>u*+Qo#v;FMV9iWxcE7O$%c2iFmh|H^Bxy%dDIdg0cK5onUs
zHV%YbSdpg4@qfSzV37x)uE2BP?(M%1y@-bsPoLw@j<yIxxM6vBPJx8%)h2x&7K?^!
ztC^n4lsvkj2$68TMtFZJRxe&qH{tn~sk#KW1^#UmZXWy0@c@mF&`)?|h&k@PJE6qh
zeNI$5i-q^64{^^#hW+-7`z{}BtfZ{n-RM1g98qtof+Wd(vj^HWa}54U2GLI-7ZWCZ
zv5^|#eEWgm{PA$QbO9h2Usk_Y`>$uyWyDc3h)ZR$&~l|LPytP`_M#Z(@c^X}i1YXh
zE~Z_IE)p!pCWTW-;P>IggW84kq30g~lRtXNp(=_ICxAVmk9Fk;Zmhd&V)OyC`ki>U
zS7Si@k%7Q6oxT9{##gg1o*JZP%np46%FIgcu=HfRyl2E&l^~dn&vTgiy-HaPP*+}3
zx$$o`<6L$>Vy41(onQPlQ$dtw!gQLLU4L%u)|~(#K4<7g*gu4x?E`&r>CT9}nUIGc
zl+HAeuG*6am1;)bd$i!w9sZ`AEPQkK3u=}%ZHc+h6SnB|nA@bxJA{%-c~}yOx!o4J
z<juGj1>4?+lDOoXL6+c<NkS@_r#4$v590UOu3eV7*&+{_*{f&fDcP5yAV;M;61#Wm
za^K-ua1MzN8n6BCar!yZqzLfmX+Pgz&=_Cj0XCZMh-j#%OkMB3f)I}90?|!3m4;;8
zLljZ5yanQu`YWwEfz+hPPg2T6r<?vBC7<BUc(3Y2?YPHbh$8vcm<_{pl3@Gm+0j)_
z5*$O#nPPmRLt>MV)Q|OFLhu7?0qhk8FXt~qLG5s^SC+KzhRC3~Oq+C;Y6(eb5Uo)`
zi>u3jt%51w;)Tb`kA84){xsr%WLZouZ7#ENou->mQYP2L={Hw(+z@2Dk2%7*`GKnS
zu#kbyvj@@{EEYg!IgOIhRH(_=9=0UH2_@c)Q)E)MtY*_0Q*YMaCtfog1yn);0$t@a
zleqeUJk%%jhLSJ7S*6dV@VydDE7Atl4auJ)>9NGdAO8&<APT&WD#4aY<~swe-!D_P
zWkv~`GT<hE0JdXsO39OOn%x83iu!rU^|nu5BKjXL14<yrZr~n*cX;b;$)aXfLG7%Z
z#3IjwWfTM)TE#I?iKzmgL;L9z3I5*wQNm}iwRUse&!Al9T*I=<KkVU_yL0%+NldXi
zKl84#_GzcFhgdqgPw)ROabzpvN8}IaE<XB8-ZMZRU{Hz(9}-lZTw$V2#+vT3N=WH6
zXk~T;+`(CVp@QnJp2;Ils~a+Mm~~2i|D^;G9O{O)$k=46h2)ijb>i+ibb=V2WqH#!
zaZ&G+y^k&;Ov>MuquY$JU0%LS2u0yR|7F-kNoo!`g}%&-<mDhv`puC0Yx3q;&Dk9{
z1+O0ZlPtyDLBE82%T+e7OP2<3A#_I@AV?A{uS2~Hy#awCjz)HU#L7+FG%7w=;+j;H
zKARznuzH3^oJ+|g`3z9+w0lHeg-ELj0Yv2llGU%D<QY`NgwKFCkVrdeAy>f}>K&Cd
zu>Rd-Y5ntE$UxPFd;QyL5s3+cRG+a0=1s;Q@BY$_5aGsS-jra)2r6xVIfLj{#4?)I
zpuE~|!lA}44r!R|28o;3ALmMtU8EbtB`DK5^_@5=MORLrcs9LPZIrSR{~b<d$&Ns|
zQoT5~H-kCgHv*C;Rfd_3>z}7#sCr*TT>HBYVNi6BBCGS<z{kIz@`d}yyUY4;CrG|7
z*K(l*!a`zC)X^CbI>;NGftsAUuhr{2Hf2*wH=uHj9TSW*ED($LYL~7_bh2ybTD(xm
zD;9Hcf6Q4wsLaCQOoxab!Z~ZS^>6fbE;}4?Xe_nPz+e2!mJ6hC)lxXn7I|M-#fBgi
zPP={2hf&(c%V7S+B*<@HMIgND2sL}&_;bS@0FAw_iO_46^9*0porJvTOZy03@CuvO
zQ6fWxzVq6PHC@>gb`Qv7>(l9I1)0Z7&Y6zFNojxb?Hzw*tviQ~`#Ib=vh7P@9mhC&
zK?{Y#Ng|iM=StU(Vz*ssvR4zfcUDVWqKi^ZthEXa?;v?pi*v3GUzc1~eiUUr@>wA{
zZJ}(n@qX#uhWnR4q%B;^_x`>mX>&*1!~UBd>+|vZZxQ`?MCZPpm>kQO91`_YAQc_!
zjB7@&uR~?)UugK?Qh&XN!#Cn<pbNkJ5#g@-?sQ%D0!c?-rj$gXtidW<*fw@>`|XzZ
zw;Gqpw>-(?{YzRCV7zV(2hV3n;_SH#TDFDe&dHpAJLt;xQ-YfugFb~%*m`NseYTAJ
zhuU(%L%8Du#TX_}&0@$VeW8UFyxwc}ZmK1)rq*#(s9~VxArMM1je>UXaIS7|7gNnq
z^7B_W`nkM5f{U7>>2)exICiz)tWs@3AX~};RO@+=gysJ2ccee>00Pg#6llWRC&<Ek
z<w3C?J`HF?2@@7eT{Nz7w7YiX)TPKLR(oH!_Nq>gH4amGwujSr-sX3yeRptuXpd3F
z;N@=BTj3j@Y|Fjc^62D;g1>(Cn-MwgI!`U~GP454w5p6t96v-V9QP;|bj03kNwVJm
zDPfG<3{BX3`R7e>P(mu0z*I;aZmrsZNWe`oq3new%T;`fY68cZnXTBU5+3F2n{&@=
z>JEB)2!E5m53-CS`O)u{6h7b=)7_J@K$=wLT>?C5Ofau0$g!nUWMRE$f4xy86}+Nr
zOd`Y2TW7l+vLzQ4`amznIRvSQHt#o+-tBJ>wcc1Dlx(UxXyIYbe}}*6lXjtLL(R1j
zR(^!m-8)JgeL<m%pqQ=%XDN<Vk!gL*E|v-R>t%%AZo#$Eg#D+@5Taw5aZwhhW@OzM
ze%TE<Z2^cO>uQWBrDTD&cYmHLo$>4z*0kC_sl=pCJN^yT(4kgD_;rIa+MvwvLB1_<
z5P-q-Q!^``-#(qvHBqnFNOV-!D<FRNq~<%9;6uMA3J2AXK}(R7#i{=JfRgsZSatD3
z`@y2F?Q{Xp@4Gv9%dM9_0XJe75P|5>Vn2fTNZhdYKrrAsMz%9epWO7-QjfN8+U6aa
z5~ce63Sj$$iRVENcwu7X&<~@)L2(ygtVu__QVOo?i)0rxNq886syIT{8=sOwHz-NJ
zK=`S8b$L17?@j-1ar%jtrBVhDi`fdf12xB9iv1ZJf6O_e#NCVd^d8?P4*$BHbvg#X
z2N6wqle=@LHvq5{I%l$#KXRaCD21J{!EgU2(ZN8i`L({JE+GyM#j9n&o6GXt?)M`7
zF~~F~@UZxP7!#&n*GyDH^Avwc{49<}G(2-bOuC)P3(UXWS-|-{%)rNNBOoTWaeW?L
z``k5ywI&DW-*?yb37Fnb(q1pm;`uYQD!eCszFaB1P@ee#Z4jpe5y|c#<mR3Wq&!>$
z^0m>H$B5c<{ECYA@70VLUl#`FRt5bR%{um9pK_XI!-4cJ-qpeKi!w#;fNz%#z`ZFL
zG&+J#AbH{SeMKjYt2Uq!+h5~kyZ}#nx8Wt}uPOMkWQ7diRS(LTIFtPPbo$T-hwtmR
zazBn$CjgLC7s&k%x|&{C0s77_1ZU(v0?TqP_$x)*_DlahSphGYDNk?0oHVGsBbUmK
zZHCSXR`F1MknH^^Dxgs^9OwZnE@)d!yB(0jXo+hpi6A-8RLC{{>nA^d7f%Swjg2gQ
z{I>&X#T{d<0da2-Nz-5ucABWIbP{$PIoU7)F7w@;Xs9yw3iN`7=NeVpE82Gd|9C7V
z9FkUh0Rxhr#oSpSC{ez7^D$CB7V_S9yq%fdg(vU$GP7Eo0vm<N3H{}`!6q>5bZA23
zZkYoU(!XXl^dr$YJU(YQ4&K3$tzP~HweM>oGF>pl_i7Z&crQMlxVUwzBA0bdnG<pT
zKkSMSt+@a$S^sOyfdjujczI8e%np?3spY}D?QQ`>v@9sgQ29yKoOF3B2?HHfI#eV6
zifTw`?fAdreG-Oy=g6-sWAf-g2I%%f=QrX#Ip?B|ey`tl=Hn;)1_+K`lG79B<FBcP
zF5mn66!0$@Vr$lb-;bs1iNl!T03ai+!$$M!M-5D?tivE(rpV&mwdh&u4Y(u67sFso
zy~%vW%g&#$Tb~SA{=J%%#m?bBFL}-!2n)1GD`iI}CQM{zG1j2*&Lruj2`t*|AV`}4
zPN)~$<P3CIhdZ95|JN9~D?tQZWp}O4<JYU6;Uy<Tk(Q^%=d*&EXuv1nx7c2J<KSHj
z+8&}sQDt6I7pqG?UhRKdrB|d|Y4iO*s~GHP_SB!pY@>LIUIwkrJUCu}kKxGCRGP-O
zfc@6Ub~RnZfwt%#?tdPo{D7DR&+3EuzTftIBtYxlxo(65R0MtDuIva{8;!fQDD$5?
zfR)%#Wpg;%cu!9V{F-n!wrm`OJHqS%r`vyQ2rLBCcR5NpIw=3Wm##mWHw*9g>}G>f
z_tw%k<I|2KQz>M}&Y1U_kt4knp1y!#as$}xw~zl?6x<#x;qUOb6AIf8bWPU#e)D_+
zKv&D|7cc0_S3nr3YFs#^q)+Xa-Ti!Oe-6xf1C3jM{3*F2T8Oa)q?)Ziue{NI<ZSwi
zse<Q1*~29`Rir>NuL~9~=74B4v)(_K0qeQZs=uvq;B(9`gK8(XC9i=}a!u4hv-{`o
z*!Lo99kPyNXh*Ped4l)i5HaT(KXPtoiuoN5{dgeF9TY5&$CvvUS${qth6thu35!<L
z!poe%H>!c<;tLx<K3TOu`0z6Ce>NDjW+E8RUF!nHpSv^;lmQP~5Re>YL$Yj_r5e%8
zU^Fv;1y2GVZ=;7C|NG}+77_6|OuuIaNa~k{Qvmt5V|NfFq1UEl$EiG5-!=_}W>)y<
z{*HH$zs^1bUb%gt;<pRsZo;6+$?v$kxgjno{`PGLf~k!gZT14Aa4c`?m%#b?ZodXg
z;Shv{W5ImCM--FjpMS!j!aRVagWu?3m-Eu(lP7TOLKiXe{Qi?4zr^H%g|dop{u(KW
zFA0TR-kQK8^1%r_k%U2XBpHbPcrTIt$3_p}vV6kM>il^GHqI-Mf*VWT+1c?dl(N!*
zaysgBKK*Zph5VhaJ~F{W94~x-&q^5nQBlVu)fzClY!bI8AKue6ko`UENX#b(HhCyc
z^N)jRqeIAm2ZF(iE0jLC_3O=lzD^rNQw$nG;=k=59gT4$7xk3t<JbLnZ5_asBTpv#
zv#-WxHxqGWpn)fG-T2z-+cyF_KewIVPZ3TD{yOpS*6$(1L^cEGKR*@6`A-grw7?TJ
zoZl1u<A<Coun%ep4g48RSJY7v`<Y=;yZ*NSeQonIlmA-UG$CwAn)X8_Tp)Xw40GOY
zCrk712%^dPFd8GaWJABrCcqN^A%`QxDK?;o(q3c5`k%j^A|l6bvHkV2M1)Au8__va
zAbFY8>@57B!B*kLkk?Sk_&78EdLtG7m)Wv-uW;&9f8XxM7Xk9H_nhEXkoY~)Q;Gqw
zBrN~T(hnET23Hfc;6D1t!CT;Tx@P}(a4Bga06s}W^6$R=d*G4jk4H?pj^6vvp747X
zeoZ_Pe6OAw9{fK({{3G}H#}(aM5NSz9{=xj1z?dEQByek`$hh_H}Y56GhlwI&K*p@
zFZdsO12^3Y-&5t^W&Y2g|5*`eO*dH2?5C~B{vX#zTfp~3IzTG-fBykz9eASn+m!_W
zkL$-)!1vff+kfJV|2$zL5`K0+AHn;7Twe)^*;;jl{{)l&W1Lj@!60TXE&q37|4zpL
zn&SU4PCsHDw^Q5yo%Fvq`TzgvA0>K}Yf*^rxje$mqhAsWA&Dv+O7ncbZTAD<Qu}xe
z<Ns~iXkXZ$^4ARc3@R|rPxd^oLqaAmJG)sX;r$sm0Hk#x0XR6_PfJ{Xhx^}uVdjWp
zqtOa4KnbW2w7xLcAp@KXMb{0;IQ1?L7SBRWHwECcy^#Gv-#Zr-k8t2k*E@%o-Fo_G
zMgC{HFxL;gq!Pu~FbC;oH<0EVVa<Y0U0{t7Q1SFkg&fe3(A5u`bpSY-0e#36nf;v?
z;Bu${&*Y$eh5gun0+GK{++*yYS)qIr_uJ>y6p*~`VXB5R=LQ^1c5p_zblyPf5`aaQ
zjq~h?;I>qCX+r<EsauDFrL)O&f$GJ`L%bU!kj$00A1-73d^nxX6RK36<3Nu;1I4M(
z@o1I*dod}wz+!_a9U_Kkzcwh{z7go;fr4L0`Auk7oq^oNcX9mx`=?!pk)QfRhmbPH
z11|fUqT{Q(K^LPh*p61^U4+ENN`dbl3H9HX{`X(#QpCrsuOn#VyJ8ki3reyh5GJZX
z{L(mAW!{4OKR1v+i$&-xtWL_T_t5gbchAPgMj1%Y_qOg5KTj9gJk4iV^}n{0-?t5T
z1{sXoZS%G6^XK<%#l*z!28^((SKAM6+=hPgr~h*o`sZh7+!9ZoTG=d^Xou6;?j{Ns
z=ZT34bJZKDNkusHi2mQBqjdJi=-9@|jw~-PW22AJW1}Wya_-n+|Kp7QSTBtkx>Cs^
z=t)$#Ye1arK-ZML>jk>Bs%DQj)9Lxv#&?I);uG98QRGT>{-5`gAc3!StW$Ys=ggli
z`?rHlGHm{K&8q}QcElj=ULzB6(C#TF@vB_|_v>H62`}vM(feP0jvdEA^(dmL-4ssU
zu?@FZmsLmC_+;W;RsdHY9qj)(=%U|nOvcuPLaCzjW)Hie{kT`c_a>@!k<(@gi9-vk
z(X%g~Fz0<v{Ls=nZuX!GKQPS9X=8lhP#8Xbw~X^$MYYeHOP|{_j@_))_;md>aw6tD
zpY8g-L+dtC>p9m%Cw3uxS=n2Dn*^+na|WP6*pW{2R@45MPMPFv5p@a-`VKySLkTAH
z=2+6?9>41Mjg+<xz5APcZXaQ7V#JAO<{R!>H7+(fq&QqM4eo8Y7U8v-DOI+^Y`S}W
z_3c`W)7QXpGMQtflAcQ^s;rf}*^6{33iZ`<Z+OHN={>bEEjJphTr#@<fIe`j&#X>`
zrnCID)j$dTV_H53&Be9;AKx8ZwoEaae|}Umk~d_S@!R{m$xViVWpOs0uFvbOZu%S^
zUaBfEOlky`!pyrh>-k+(Lki_f<4z$AiY}YcvGsXx8<v=){TBPE_q#G=n)bF{-1PL^
z@x3|iS5$zE@-H*Cf;_<9z1jKsZ?n_Hvou(~mGqI|1Gs7IYhT{u?5s3oPk3BV62E%I
zde>{ST>k`OY5|EvyK8|w#x=E7PPe;;N-a~yYwD_e8+Olx^h$WvR#}CN7~&O#oj-b@
zREI-yY52tn^<%H{`o<zk_4yRjMM(%`MUEIf$hQi-YgDH7^fn`d?Vtxqz1M?-{HLtA
zUwq?y;lA|x{qF2H`m*Kkv&An)!Bq!@BT-A3#(jWme18S33+U_DujecJ?<AW%d+0JL
z?YH!5nP@315f_)xy3v2z?8erhf5uhq0JV%GiOGBleP;qOjvX>y&ak7$b62=z@i3eY
zcO0WObAG#=UKPrJ+GQwjq<JZsk7dpw`_rnx#^~jB0a<U)p>l!#>;R?Bx#+8_OlIX5
zCo8wM7Q+UL=Ti++SUi@*FAdq8Su5JVc=>ORQc2*)sx#vgsiF{c5-~lVe%ggeOQ*J0
z;Ys2C84-t}XFOHNz8=s0vLKGtr74MzBg+Ime}^uSO!`XhHJ9%cS-RGwFavg6<cEU;
z3+w92X6C}vOh+qZt@myWKeOz4mtFjL<>EkY`O<NBwOYlZe4Lk`F5VFr+W_IT(YSL_
zGo7Px!6Z)u^xUq}<ta=3cHFUP5ELbRG5NN&MCTz)Qb9|H9ZbZR5Lq9Nc@4nkCKgqg
zTl4}hql7qYRyg#6ES&ew(R6WSS*zX_MOh}0KF6WJWXlmjkSpRKKUC)N5%5VYlYXn|
zhSqQDuQXQ|=zc$XQfk{kP{v&B?vtd`4zy;XPDgLPxtMa)r&K3oKSjSFq_6$2yU}bG
zyjNg9=atV!xN9DNeem>Z^7S2a$m^?9eK?{w6E}Q&g@(`QVZ}J!T+>6R6RrkQzVnep
zu#%N>nv2VW)mBTgA3E`Dyiyz$O))3d=*&UvB$4=ozrMq8c(?DzvFlrlR}2Zf)5T6@
zc#CY~Ud?7B$*gf)7@_8?A1JrxGs32%vtR<t8@G@1bc+8TCVz|_{s$ssqq;NS28Hi2
zW_S!4Za~-FXekuZIYkL29EN!JIrr43PvTZc6TrR%4>NlCa`#iR2on!I&2!Pbv7-+j
zEVL)dxsXO3z_p?4>UFmn<c_1^F)^&`7g4S3)hlk5G74<@;BMPD`05_x!qD@$q0;te
z(Z{Z3^jW5)%W`vsye)oZU-I{Cl&g*G)kD6Zv+%6=G3&CH2u%y@`?&=f7eN9r-78?E
zJAtDYX_uTSes=|NP_J;Ie>(BD3u*UMOLXHm=jO;rb<AB`z8bWD^DFX~k|mQjmwKvf
z#3OhwGVwcT3~O?JCAJ@?wVO1{HI?Y>a~c<~33<YATus^CH@9#1%JQFwNxYBj_)89+
z1RqJh0aSj3RR)r}h%6hecyHtNO>0OF%^jy^K6S&{FHPcFnYE3e(RF4{Zm_L9+$Y~m
zHCG->uwBle=%Z78QIY88eg#E-5;rF{pM>17exQ2u^knVE^V61r<&}XK2Md_{EARD1
zLuKOmQ6CSbyw>~3^#9z5(WOYFHs!itZVXXf(%=a$!y09!Z6tLKbjnj%3lK?E+Yv(q
zXmReD6E6Vf9)p_7;N*GaSa>EKmx|-yY@DK;(4|PL{z6o)3eAleMs>fwAk2wFAIuF@
zlaFOS`7VC!ij>#MuVv4B&-T|nVC_6!{y^#e`FB-AbX(`NK2aF_suv*7lx>L|d0uV4
zlC9~8V&D)zlX4z97b+g)b3kJTnWHJ>e~~us6Q|RZH9L?EI5EM0duqR3;Fgd_5zl<a
z`>xFw1YS{(GBWY+E1*sBw^ux7i_1+6A9f|RN3<t93|x8MY!I@lMDVm<7{|0h+;gSB
zJ&xbB?V#Dxz`M?@r`n0X!v*l{S`eUwXBFi$dO@}4nT!%`#PDWyDSf<&cLynhXyjW)
zuI>3Xs63?lKW6gTnU#I*E^{n&Z}`avircBjO34d+aKY>FntbR<xfpzzq|yEc&Zc*?
zOd6f<a<f6TgG7H*18FR9LM?`to%(B~Xia*EsPsGT_emjVY**B+mMdt_G2!P<A0exR
zIwB<*O{GVuON;Z7y}V@*|B%Vf^#9l1TSrB?eqqBj7=#KUD2;#uqNJ##AR<yy(j28z
zL2^hD3n}SFM4F)oi4jb4XdD_;Qc8(Y`rVIu#B<j9egA&zUF-D^*2oOc%-qi%dtdvy
zt{wa(|E1niNtjH0zKXbyY+FjGhf5f1sHnrGq8_F6zrzWb(Ed<!YOep01=caPvlb^L
zjHYhi8t0X{IW`_kdIfvtB|Y1H5)vkXBCm4(rAf9Fu`gA2_z#EC`dapB<?c>Ox8{5J
z2Va-l>TI#yy!v4O@5N3;@)bVxJNG9C;U+@)tbct@uXcCi^{^oG20Eo-53V}?H_?Qy
zx)7_}x)?#UH$mTHYr{qF*6@qXFGKIqf}L4m*D07?H!SKyIra6)jA;2af7Oh<$$SZ=
zsg00r?3earrwbQ%=#T4`HExlOOP)LD@x=_t9Llfi(>O>P>Don;|078;KXOJe^e+?6
z>pw3zTI3(9ur>9RI7deXK(6!7pL66q*XR629WQ?wFWP^UtW#-pR1Tj>|9x3Yuf}t(
z|25teTh*IVz?(FBvDP<q>Ix}(ZJXrB0gilE4-#l*dD7;!vpL9ICKd?l7L%>Xnv6Mt
z{_a!lFE>0Ajv`Stz#UDHwaj^l^?#7h0yw-0JdijmhTIa%pq^;!r%QJcQ+Jzu&lnVG
zD+2M1^z`5F;TICKlAIq7-s(~nb{@M}E3LI>`3|m|x7i1oEXs6HXh_@mK+ADB-1FOz
zz0FWU+U5OMCCNyRF<dQu7hUOM+|^tsCe>%$5m36g^g2s)`(Py6SaI!}``DWUrL$M?
zrKPK-$({FJz5JQNU5u!MBGoe}_Dw9TeK94Rq5Iwx&HrgDu4<@;AbkDZ-c=|hJ5bW#
zRdS&i18j){&af&nr*=!)TQJ6*an1GYTGS&@27t=DkHQF;N++N%aRegUGx6JWUH9CC
z_DyDf^bKR7L{I=(Po<cL(3S4=?YgI76yu-s&QG`=jQ~O>MrtCoXiwrk*-^e-W8*=n
zO5eo?{KGXiTHnWbQ}`yD3mTMpCAadWjO}_jN4=t0X|sM}JfqoEqK<>Tdy|-0CACnE
z2rJ}mie6yZQ>CyNV1&NMG>daNYyll{pQ>iv9a&*=QHOURpRuI%1bcyID9c&<ZiM+B
z+pd~HM*{U@1QUPGv*d@CqUO$SSAp9_44-e_IE5SvZ|R{4P`HuzO^wTEj(Kja`afc}
zsOmb!;a{i82<b<3m<&xDn-4v!LUWCUHSebMKxq~iZ|ZD=Kk=F1X>4n1&(p2&)4w1U
zCH+xoY8xSFi0Li0ReudMmhboViU5sX!am6wQ49MD9aMYICHo#FCu?Faf|}eklrL0+
zG@Hk&28#X)WU_m+_Z*#w&BI6Ni5`K?4V!9c)Z65J&w-`uxQOnmWeQVUL^ZO{Upan*
z(5JsRj3S2U*}Nq#?Q+VI%hRi;Pm@sHA%_@bJl;xP_%Hc(Z%JJ2g^=B|sV{3cj&iHS
zRtbAQ=ZfU905b$ifV%DDWn>IlJ=ob9jiVHaVrcjM+5qT}A?2BlzC?%(=BS(R$d-_0
zUXH&;S_Cc^GDn1n@&-gV>z?2IRK^Fjz(}B2zK#+3Dg))(uQ|@@$e8XXquAT<*mJ2w
z^|4>X_F7@Mrz$reIr)b-TgOhb+*3%qvfJ8BN6Q~{Iss+2(M}{e<ic+D8muS*?o5$@
z9CvWg1*6u-_I)p;2KG9J3rdU}34e0FG~)tw=U(k5T51UhA&|E@Mh<;b50p<PNxG!J
z*OyBZAVIR_TBOVAeC#KNcaNAp>D2-m&h4aREob2Fjt7T?r`|nh_z|p+?J2XHj3f3d
z83%Lu#rE?YOHKFrg=*+dVwb44xJ?teEURSs#vXbe&;Wfmx68zHx2;pjiPQVoCsqVc
zt@f8(_7$&h#U(CH3UtXF7-ai7a!OjG;jvv)^^q4s4g5q416?dq{CzQ&Z7FjjSA!1<
zZaCM-LoI3ZGx(AY_$Y-QnZ{dO3ShU-Jf}R5fL^CBk*s`{TqFz-ui~8K7b|LmGE8ep
z;-#ew*OobCZ$+^ZXDMqDQ&*iZIz&^|nljgPkFBc2PVc*=XlmXT6SG8XgqGS(^LF<7
zG4oGt+6mIOCl13^A^a<&yZXZ0v)dJlNM~)OA{bP!7d3l7vFR_%fh12N!Z^Ku>cwFR
zM|zy{03fK<s(~*nD=Q~L(%x%2gb^-5sn8qPz=VF-^4!bRO=njhf6#qzr8h`!YjO>@
zYerRR_bY%yv;f=T0&Zh*wE)0GRIj|e1b%Eq$IE5eN8|PhkcwMpR34|K5eTPb+uQK{
zt9fNkH``uEX<#g%O?{6|#BPMgzm_t_(XP8SHSUxAK&Ete@2<8pae%;(89yb`{{{X+
znq|GEM}5*OlA=-6$%?ZJeEku_h$LklarO6TV>|a<ar?Ysmp%>(niSlEJ|GF4v?~bE
z1jH^bkgOKSq{WH9mz@O?nuALh;?XKDlT9%RgTr6Dy}6)zJabO7l;T3Is3}Ib(8Qr|
z62#3Ct6wKo_~;))jp!1<*Q(6r+mE4>HoC#?!B83x?YS@h-mvjZ)Gq~X*?m<_5IrRm
zm3pmHc1S>xJkJTcqKi~CvF{uvR(P<(W+$f(1J&Me5$Kam>59Q)ut7;U9T0yE;6cY;
zfA8WXaK$Z-y<&~ca15d`JKnT5`;kS&WC%f4KIi8P9M1U0|Bz1<awghWC-h4)?(0q)
z&&6prT%3v>ZVS&t3PDXeW*_ABTBR!LyjS;C&I;FO9HY93OCXDET6uC~(Rp8cp|<5o
zy(QXv;Y@tZ(WI0LmaTjJaGhy(+gqWl^~xTsC+j%w*;Sr9s`l{bQk&Ri{lyvPF5#V;
z^s0G>$^EygP}~w7QCVw-W5q#)CBhe$o3L4(|3|k{sl&inGw=im$>A(@m2JCrEh!e*
znU*Iztn4?TVT65=de~t}n`&qwdocz0XC2;U5`#Aaq0HG#)4LqKx?jv?;&MEkLAA{=
z-Yto8E1)A*O((AE&2bpOEN01UHkyF!e8Vg-qaQ_ph*r#|>qRlH6gP7LnmdOhhg7b1
z&VJPW>=jVbU)h@U+p4GVQ$I7!%_L?UMg82iPzyT&RH?4vyx~BaGdj?~JA1QSqKI%J
zB&p0j>wO5zSzS^5G?bB@eDAPvdO3j~aKs#W{+u7ES{qFP_GmEPbha7VAdS0CxFpzI
z8hlQd)4~VCY>m%JU>aKW_dO;3svDRIP)0msL(>P2gWsr786PZ{PR9R?NBJV_T9Y~4
zTjIPAI5fG0$w;?tE;eh#ht{`JWdzrCKI3p-93uf&%%!2b)JwCeKy=c0b}0ol5$`!K
zSV8~1)RKH?bA44L;zV22%uhI+eTeL`km_a+zLkf{aBMtLHfg=0b`$4qWiwT=(D^Lb
zj>J@P{*%K=IkgHv0JnUh<=Ue7YBO(&$Ey)(>vs0qVSa>$;)a6S)*z$~o;{k)piXa)
zxw}id-yzc~uiAA+7r&h3Dg=}sziRh3p6qF~LHY5|r$N~Fz>eg(?|_!NR8J3z&$bHq
zyBin~+r;KgK1yC*W2rBmyBz*rtgux{tYgNw_*#&i*T`DEu-w%psjo&7yu9^da=Q2o
zgD>AYFgzL#0|~?FZl-?M6`qti4s$&II$Lx7XjKi!Cwxe|5+$Duwa}7oNl#D)+#x7e
zTws3t3uItFM)GY)&+LWSSfx(hM5RIKxACLLct5g3Mo|Y{9|w6;sH(ukXF=S->+}>e
z)mSBOO5DQmEv*2UEpkh=!LLjQo2MU33jR6*&7aGf)4k6iz9IB{l*CVbns^YH7FD4C
z$!C-89<_kH#fW;I<($V!4wf#hF52*rL#NUk7FIJE^G8~1eA!&>sq?2Rar~Q_E5-a?
z4*jirUR`kXGc}c1J2jfG5-HIwR=A-g{<Rqh)<Jo48ikuWddA~U!2*{P#c@s|Ku&H@
z4Qc}3cHsyCSRO)U#{8Gt^lC%gL0k3Wtm6P|NFz=tpv%=eXz5m%pR0b(sC)UDF9jAQ
zQF%Mq#?vFxJ+;adyf|o`8q)SRuD5hq{0NsXhSrtiDa_{w?IRUT$Es%Jo%?O%^efc1
zw=Cq@rO3q{Z&~!0g>Njzl24RZdfN<O>c*sL5^Aoa8y${1Z?5PsF8EQl@LTlp2pM1o
zJ+10$q@xutg)Vy!8j1ZJHyfd}!Uj7U9jEvBW!DTZAC{pvB^&D^a>9wsY0n1>Og_C0
zW(?jX?UiR~e}Db$GSv#6YfnOTPht(e1G{3;9Mks^_!kan?Luic#0InXxSz49!#5t>
z`8-Kou}SM1WtR%uDl#WwIV-RU3azk>&AeBjDmSWQOVjr;6EqMB3!Y*QkqA>%7{XO~
zoEW*>SVpCds=ruClaTw&dz=vJ82!dWwmlX%Ro<5dV>7u5P^9)YXMNNjT#jfvwOE?a
z@M5}?XNy<N7@hKSWo6$WIrD0r(bT;8%>yLlLK}HnXh3yBBdA*K1XX|s$yxU*n4BCU
z_s?ro-D-@z9wr#BB1GL3sb0mC8XYIJsLzwCCyQ~!i3f^lpQ=QYEOw<@#&;Nf+CI-c
zESdc~S-x`AtaDHA;Io9<?nKwR<1BoiL6{l?hhLx8h{vZ8r0(F?p8XMBpFu^p#Ef<5
zL>r$yg{2R+M@yPul_?t^Ce^qkY7;xZW3uWEz;afgEO7C;o)*Bqozr=1Y}>Bp^6sF7
zO(?!Wb_ua^>@lMJaOH5aXuW+Ht5iCs9&Jr`bt&4b3-;AnhF~LG4N2j;Lw)Vb`&abD
zZm>EX4>l}GF|L`p<d#g6J)yexT8>BUkAP2dPj~h{)-=0!j|4<M=kV8lS41$MlctL_
z+iND^mpQ$!h=fNhf$2Ucx5V(FC@?i?;VUP1Zk3daKPSnZ`RTsV%^)sy8J6&snNl1P
zt<KnjA8lN~R7u8P=WnCE@B9sJ_>h%Jd;}%vb;^RyfSOxPI9xgLTQ2H^vKt@dw=Zy^
z9M_`v1pyH#o-OOOlpNH@9VpWl#ka?JWxhVlEPaP^V?T*KNPT4(Jk2m_flfSrd4e-(
z1{TS;iUe5lTM|Lg6P<1HoH>n<3F2tf+iS6BtYamn{mqpYXZlLB@HOqi!lw^2-Fakx
z^<ag$RCU9Qa>kL>*&3p)2!e>s(G8x6mo~M+q`NJ)6|9IzDt=_@;miS`a%xCd)VCb#
zO0!PcLL2XCp~}>sh)xw~Y~{Hh&FmU#kR<zpNoKxF-54Z3Mgv-?;d6S6i?2|kjsniZ
zo}ts85l3I&JKKq7N$vP3&<7wW1iwTi)4XYNjY#laQ{eUt`+dAQ7Rh_~EX{JCOFRuI
zMpB0n0#Jrmu{N7caq?q%Q~k|lpt8WcMI`G@z{wDVhF8yO>w(>L**42>K%ZZA-e4)S
zLkh8~S#xMIupW7VjWa4CoaCWIuio^*`qA<@<@<REdCcn3Yu>kyO-&VoRaT6B7;JsQ
zYfB+B-Q=ZEfuV4!F(;m_kEevOl=-OezB{T455#|bU@CghaiY1xWKskfO^-YVQ)3?X
z!;iC4b~m1<B{v_HHEH0x8>*^Ican4<L1^vTZ8XQ>Tbp0b;|Vu<&fD>Zbg8@B*XfWJ
z4ppl}tl}QuICkVWb19pSB%-iuh8$?tDvA_;iJ39=b!jc-L+~K}WW9YbCxZB*PqeAl
zKEzAR?ftp(;x4c;wnrOKw@5Fwh!iKPl>34QjN;rR`CmZqH@|Y(VWmQGB5{TP6TRqy
z<5(GRry6?X#~X&xLE=0bng$i-($!=qL~<W63ACEx+tgF0SHdGBKjGi<&kj_y6=w40
ztIPf5SpX*>$;&oom0{=<DPz#s_GWGPUS|3zl@q%lj@TbQft$X0)mf9kcWE@ysVvK`
zA@h|7K_6d3!RVMHX{;`-kE*&$TDZ9;x!iDKcZ3%b%tJ)qaFtk<iB3|Ha4RGq?+i_z
zpXS~Q&q2Dm=c2j!nd>$_v2F8hZ;os`kd`n<Z9<r-&_!asM&gtp?vS9W@}=2yt26KT
ziZf(yn(MYshRH~@q{GH!AKFh>nvS*M%Hkqt<Qj;<7%^9v%TxnE29e*>1iPYYfD<%b
zi~0Nc`xBumIKd!dtIU}3$Sdgw3ss|-JXhEC@{25g?@d=s<K|eL=A<$7m#^z4S6xt8
ze5*T~rFXLsjxeKm;iKla*T^RrFy10*x&a$&c#b98WZxG@DpxI8xKFU3eLH17O+>^o
ztgIlX?QWT9ff;mGkyShMAnFjqw+GG^Kc^FW_@#gb?_{r(ay$Gbd$W$SwFJwMz5VlF
ze1{0kQD!;BI@IXT8KkzG%ULqfP$6lAzEuwMy@=(aZF`~l3IP3khBKr@u_?vO8TmSS
zF%`<3U3}l8Z-QD;ln9mavgYHFbOoJft|}9y{2k<p@1Ry5+<#QKj=B6pJ%g$~UEfm=
zAP;^`c;>?**IVXyXAoJL;l3mZc1tVx((z<>lFrPKxO2rEOSxA9lr`z-aea+<WN(tp
zyss301~qH?i!<+6dED&+9<i}SO{Tag*2h2JS`&!W6JY1pHE$lQ?3-xzshjXD&<$b5
zzpX6!O|-h_PeH?l3a~xVsgtM6;JL(g;n9~0T&m275Tjl@zD*1#XP%;EybXPy54ubE
zSeQ~ytwFEYXTS@`gb>0*S`IL~*K=uq-h!NT(Qsv9WJJjBFW}B>d8L2L#JXdZNsDsk
zcNrpk+}Q=+1ID&t(+zFC0N@4>i3wlK3vAd2g#O@P47Mnv(MKXBEwSnLp0maNaSO)I
zI5Cdt?i^V-Bg5(G=^hDIT3gOXB+euxnQ<Jj7CiZK&08c~KETM=QzKK&_1OjL%|Y{c
zp;h9@E`Gy<C-SdRxKPEVFz8>Y7MH`fY$)8&`*pW9bcvmc1Qt#MDEQ@}TaH{eRM#7T
z$kXCp^mPjZNcM0h?H5RlHV759Vx%c4yx_J=FtnC+dwH82$Y?Y8w+BTTvn4KGm6Ma3
zoyc|_f)Z@{=WBW?z#GU0Tdny@O;yeX2Ve_NgkTGwB1dA_Uvrf+s>St7pCc;;f(^$u
zqsRNAmwn3^=fOu{d_f#wwAv$uC!zRJYd|`_^0{XTFOma{r^{XmgL{y<aF-;&|M~?h
z+~a8eQdQqxfhFef1|i>lWiNXn9*yUoC%fMDr8s8#F?{&wtW@bEw;cLP>h==t6Wyk}
z@pj|n2|eRz3ok5%c~_>repT5+fO*1Q+Hk^>%~Vw}yHb*WLVdFAg5;d2#0xRO)$zUT
z_V!;iQ9-su3s1wOOR)Rv0R&;22TM^w@$ai(p^+dl$SOQCc~*G<<FL6|mL}H)M+jH$
z2{#Mi_+dW7G?5tDY>BPq<iX-WuUL7%;M#Ceb1a+|LnG*KCmjb*`x)<<OmYcjlKp<I
zb{>2^3C&-31AQ(^`6(fA=oXmt#L3-r;_DV5Fug%5&1(!|8r+__cCkKd^m2>9M_ydj
zt6hs85y0q`)qbd#H~;(q$PwIbF7NUW+`+{uHHQ2++5|PQRa{UOsoO9>7lv8IzuCBU
zyHsxJZGr688a{1T>;6Zo3%o|lJ%v~IwN87jj+669DZXgz(lPUrJwj7u;`wZXmZo8C
zrLV+ZpTTpv!{5l`uNY-1+4gd49VqtvOld8k;*-#C*pmo2snswa{|m|-jMVB_B-VI#
z5pv29gH))~7&`%X!q~R37hW__LLkF}o%!tjZy3&xG6|--fz}BWqaX+h>2-ddUGIf-
zxz<yN0&mwzN?pXlulT61W?=4DFMZoIZPYa{DepvahvSG9_mjZWq1T6Bt>4?ra;DpH
zeSKux@yfH!9F4OD?QS45Dj(9tJ^G$Y+}FENRi&9KH%4D=(v4}*TjUz>81Jk)y42^1
zf85{<#8~mScQA<jz{qb%9Ej5*+4Fv7eEoaeKE8Q?NqVq-^Ky5xUytR-$ng{e2NGp$
z52#}G5#|0id|24eCfKvw2-RTOr`LA&7W?-YbeP1UTdKQvpTZuOU;J<ntzVU?H1nZ)
zx>Ip1N~;8re*nA^zUI@;X#B40DzJKgkxQJ^ro7*ol3TngK2UDQ+pkiJStTh5Kzy>n
z`I&+jNxaXp$2fM^WFq+v%ON{Bco0^5hwM{Km+UjOL<uL1TshDa2hrcAyOdgir#PUc
z8yJC@K2D$es@GTRS33wvv~OLfxcD|GvO_(vMk{f6WA|hPDiQyZIZZq4x9x*i{oxKf
z^Usw}W8Z>s$bSfYJ8RBg0{XoLO!vjak&SmW%g+Kn0xM=<2_<P#E-?i<E8T28v>%;T
zyDT9mE9E#66BWGQC<7K!NucAj-3Id%kY93fHwUWSoeqcVJZ(_cI+@)8nB%OO)v!3r
z-phq@l*EYuFHdV1U1s$dC62sgm&e5QiAD0(R5)H6xStlCPUW%S=!2`-bd4@hziX(U
zn(P;FRA4HMEP+k8>&lds0;<NAXe3eedtvLDQA41M&sn8G<UTynVrWYj^Rv7c@bGwe
zETc0JnbHRGVkeF^xUlfHNo_8ByRR*$C8o0FX2;@OH~EUJ)Ekl|MW%+`3<FU8Hmi<v
z{U0W?+l1*FT-KKxU7!hOozLocTD8TVwn$o<pP?ib`wi{@=ovgHMl4tX9~ix>y%LDD
zWhw_|ODzs~yQ&iPc;k0JNAjxRb4j~0tCG*UzYq0v6PJ_6*Hq0X)MeUH=r{w)F(kj9
zptoUs;L$DBTw*^NCqF6Rzit{yIo33et=>{?Gr*4pxvGBY@h)`V%>pCLJP+Mqj1EMF
zxKrVm+gHAI*ohKbyMueCZ?e0f*DM5ZRhf*|gP-2yb)!&F12B%1f71SS`8iLWom9DT
zO(U7coJhxHN4~zuj0ezpt3sH?oK-`$QMH-(&~_wY=v`ik3vbJfeJnp0luj4S<ZAZh
z>KGqLBdAC0m5{ospOWl%>KR|0S?PLL>Q=XeSYUi57<1>Gg}GkaO13-%yo9KaQ`6;p
zr2fLCa{pv`EYp4e@;ZEV<urF;&Bufz>eoz^7$f;boA-7`9UZFbxL(3}?(KN9@zSLT
zaJcoD>kqG4M^9|+HjSrAcm4PvlqpF39POdelODQ?9@a*rzs6tNfLPTyZS@Rds5fxF
zi|n(#dKZF@QP3Q@ni0Iqr*Al=KTiww&~`y}&L<Bm6`TOM3PXoUM!%sBFZ^Yt^@`5Q
z^%h1HjyO-!{1f)0ll#f-2eZ5EZWsApx&uK%BkU>`sMv(jJgG1OyC5#(Tjnt`(TW0-
z>#!v$pf=x4bO_>#RZ7)`2RYje6Ra33{cM`oX2GmgQoqKdcw|W@wSw@dhjpO;?3pg!
z+ePtPcPeEd9AzgZ7k6^Q&B$_iE;fck-TZ}|&tY&+DrqtOs6P>GJh4q~vX3=yRnzdY
z5~YYdnty`Qr&`o$|7zFGdYc!g>M2YpX>PU5=!C9b!Igdu;QP&kpo+Sfmd$SefJXj0
zS(7qUZY5v382kXU$Ed(b>LPHE#2PoxPcZg;Ikir39kY^gxh(AK)xe{qZ@%t^AKT^f
zwnoa*iuy%DDDjpywknR5B(!nF)x)vlwtpV2!VU{Mtpx}m&6gX9n~?gj<VP3VpYO6O
zuKTjEOC~=!_ddp(x^MbJZ{*N2+wle_tJ6GvFSuN1_8$^ZGd5K>D80~PCy!=|V?QTk
zbcp2!HQAfJ{p_8Esn;p4v`eZ~xiflI9;9sEp`!%|BRpXd4oW^HAjOQ|Z_D{{j$h%@
z&=^G^{l#~O(JcuK!mCeSO?HevAv|UgeiQLBkYzpB`}N&u3BGEZNPej0@G!>mtHj)o
zcb$I7WQ|X~oGS<#`@v=%R8?+jCPTIfcCl*p^dB_8&|H5efy*tjMq83NS-oo%ON;np
z>>3eBb$Tr8b@`8yHMim@5}u#p;$FZ`HqG{-?FZbx7qOhVKZnb~o<V(c%dk>c0+=?5
z=qZ!GPRnWX5&-iiEt-u)rf$atP#E=9E)@mFFiRZldfF4gD)oxS_KB2<i?eV&*IKCl
zZ>WPVdy0gTmNg2bHHo1lDD$SMW5Rx0-rJ9n4W=2~6*53_A`EiK+7uEJ2IRIVdLK9X
zA$yQ*B3pJ{f2N`+y~f9#>b;*;;}YdJ;hwGT^q?1x3l}F_L*riR;H(Qnp}KRiIffJT
zG5P*up)6=7O<isYyck#4X91pMfBrnZU~6?f<9BMRi4U24#J&Cco6A)gm8x7_U_*N7
z>Qs!!{lR@1KhRs{nun0lC)T~*FcJ>kg)=^phLZo9t-fN~HK45*Tx|=?`%$6bqYN0T
zHtwj&IwNF2?<=;Q=w#GaE3<a3q(mV>>_7)`R)PehO}{p2><88iYUa~dYG1@9m2v`q
z;7VRhoE||jSI6a<txo-z3d1h4^uVM9aSj<%ZXxwI8+AgIH232qa?MAi^$7C}eON<t
z{E1Vm4zDF0D;!)bnhzG2Qe2gA*uB8$6U`Ya8A4$&P<GS)I5BpZhUM151KK=7j`@r;
zzk+Fqffx+UI?@%dXlR(dFx8WAvKfzCs1Cg&&w(6UtF13a(5W|G?P6Hxd0ur%2#D34
zeq=;Hvd{axX|-jAhP)--mHhmAcTrKcv|4jExmYzZBHI^~tT#K^3n<hehsb!laeqOK
z-`VDcOMhJ<g}+!k7!`f+FQPC22F+Qj_!D_L!8G3wddiCLEczdhG7uhk(*wyL3`*I|
z-Lrk4`iFDRg!c?%(Q|Keu~y|uc?5egnHH6kQ(MpJm+Uq_5akncd-N&ybeWVSxa*MS
z_v8jy%_-(x-wW))c^$)Q%*Et0{nas}wwb;u&xL-a(LZx(BUpp+>-Lo*fByEzQ$K!y
zqemt!%|9hKKLg00QDiQ{x49Ja>AzXZL?jnL<iz${<n)89vGapH2=*SW`^^YP>qn*T
z-#=~_c~krXzyAG#0D90>t8$#Ucj4d9{JIF$t3{^&aapn}e83wFdVYlY-*5P95dM@D
z|NH$fJ^`u|{YPi({~j<|MAh{_E-U;0zVN?C=Ksb!%zl{Uk9*&#ZwBy?Fi^^(Ap9fD
zjW76Zqn><X1E*FbNC6>UELbC71Uc-U@+ZF`JD8aaH;hE&{t#=(R_zAA8OF{ErxE#B
z(>z^4B+?%n8|#(Lp|BG(W0uq@G%;qDa1;pr)ByM_!puQb(|rn{JUR-|a!77hG!Vg?
zKE4fh@BviY{KKoD$G{V_3`)!1SqOA5jxKWS{P&_=0LUQL@ur0)QefWs*cxGvTL98i
z;<aMih{Huox=ehZD{D_z+=KW7P#1x;`~t9+xDkoKQI(hmP+HrQmiDU!Rup$_tS)U5
z<vVtn<);MhaPo+tNRYvGKTDqF?`Yd!wx7yFt`i`%CMZL4feldzY7R&NVYxCOINtpX
z)So1s<#u}UfY&`bB6$VuLWKv;y8$H$qgUYW4?T`N6sx;gD~eq;ssB)*Wp|S--zn%0
zOHKSO2JAGgo7sgzXoyVyJ=Kp?8R)86`JZ6_f^^6(59{LLP__jv6UuV-mTdqb?_s8l
z+3(yAg1Z!0=^n?sci3zhA^wc)$>6GcrBEOB)&yMq2yw%boa#5g4!BR7JWscvM9Pw@
zLi3M3iTw)f<{1<gzh_K9*@1SV9D@8@`;+sXg;@1eX+qpr5G=?rQSK;v0OAD2_quiE
zslWd&&6&zttryg^tE4tZm(T+hPENFw<H@QM#gYei#&sMhafY5LV(vfd*04vLXZTZG
zgbXv~G3qQ7IOlGHwlw_lr`Soka`%N2i)KXfIqu1n^ph$9%M%)2_C<d@QK|g*I}-&w
zV(<Vm0XHu%@6m61K$p`8iD<XKDYzuxZ%IC@2i_FWa@C=<8Idu+*jls8K~#UPgWz;s
z0*%+QF)z(D5N5)_bh!)=t=vZ7hAV@shyrQ^3=8BS4I2M8-}6ak<I8KPux%U02$`M$
zovuNJqadi-8Ksn<SexEz*xmFu)uAO(qOFz!=L7B(8j`?yF$Yaa`nQ%H1w*(IalM$9
zGi_tXAG<e@+xipB5uJ1pHS4*nppO4k$ZK#&(mA6Bil9LWc62MiUV4!01iX+p6Oje5
z`dHd$z`ZF7k_R1dHZl)PgW7R9J`$vzVP9uVw+)c=e}Iq?saEoG5Paq@0*0j6Cg#vD
z+TrgA3w|6N?qd;a=+A=@{K$fVFQ5y3vThm{$l*cHYOtJ2;#BkXE=c}GFunk1qz4$_
ziktuV9_L^hG?tEC{3C3+%C0YWl{#h~$XOJkys8OfBR4_h3dRf$LcNr%9(5Gj<;4_#
z-J@l5M3rwrK_v!yP05v%z81>1A;N(8*7qUN+^QF}ZfpijXS|U&H3~5WnM5sjD2*f7
zlrcV~sQ<czNthXRoh{urM0V@2l7Mtm-9Bv)2Yp-~vB?a-r)CztC=E9(doF(`e2G$p
zx!$+Y`+F|Qva6CkX1KHBsHa7yKsXe_cxR>GcDTRSqy)MvWFcHYlLLO900lR6=V)yL
zfSA|(Drk>Gn0{?1Jq^nl^Da85ht})1#hp5@23=+wcrf1JRn}zeK4j6{AE`Sa8LE|?
zOmTO6DRsMd4eZZUd>12A>21%G@2rSvvO{rsw;X4Ke#9%t4~<50&}=eE#4&{_qG~Uw
zftWO6WQ$M{q2mj#$*sQndw|Jk_yn9I?B?IYR%hLzmVgp>Gk`pB5EM|pD<IY~9TNjf
zn%ZvzL(nYVHy#8AzeYi}4-$%)p>VF9QDN8M41h-BTDN5LJ+uNTsMB-X&dlUzuQWtA
z1k{PD%Fix?VXd;9TtI;%9-RzyWHHQx+TlDsq1tnh75phUN}EM;HOGd1f6QvW2wC$w
z5x4sM{MV)(<rNOV{JEC+eI_DcP=jv_k0MvSKB}5vhk2d@Ifb}s0j3X&s=%QPjZm1K
zZH@Z`9V;>WS3=p&1_WrC@NRm>t0CuZ!O}h|%y+m_5HnllsI7DI&@$zxPHLdIB!ET8
z@m$r}hq0!*GZhX4JjGb&>jR~szehSe4mQv?N!B}K?=Ra)wx2qmr(aD-K%A=Dchjrm
zWkXF4lq%r>z3Non`;1WM00g6Ahu$5TWcYLBG6mKTeqR_lKoB#C&jxVd@&gcNto^4<
zT1jd2#;csCF4C?4&>Snt$)ZYaQ}mqri?83)GvB$~jVQQjtQI%a?(MuoUB~`Ir>@=c
z%UxaiA%X#A3Bveixa6cUC~d!!%ki1Ooc}6qkBDb}-4=ES@^@iu3#RMeM*M_8mKR@Z
zqWuBbC<RF{>)~AeoSvZZw`LxuqdctfqhwDdcjl1S?mO~NviF3sG#t0Rew+URD;LIi
z)@y8<vqCWjfk1mvi=VOMAkVO9i90o%i)^%42Bi^d*EyI=;MvFv+=YT#wjF_7KnWyi
zf{ARX-$G}qj0bf<zDh^m0!#?$6rEl5I6vR8;zaK11TOLEdoI0e1MkpEm>$-=;j){b
zE*#_I`=a|ahB*!3R;+0S<Xg2z^pWiV>?itlaY>R|Cx9x_5k*?u1&yf3AFG^*FgwW?
zCcS1oYx#aBzHb69oRE_v6TwRF@8N=hJDyaBC?B`i@U`|unxqAX#qGnP*2b{g8ulnX
z=nu*c`wblo2OU)dfcUIpX*#pGA?VQog$;g$orDndwkDvG3u??HPDGBq>i|!$mruEA
zU@7#_+ehu}r3I&m0xIWyu8s<idWya2<~Ho$$Kf1}g}b5z;-OKoXo!6VD&g=FzWGtX
z6FjyQ7Z$*W74vOOtlN%F+&&IVfd875VI`);9<fRpNQ3#kVNaGi8{(2aCuVMC)tu<`
z;+Pk775df;Xdo|wWmoVc5}Kd0BY+#+DO5=~@9hL&*;h%xA37GmVhu1U=YWKLAGN5Z
zg+-q}>0@TH9_4kesTmOIuZo+^oDR*uTKsB+bpgFBwh3qGYW@4sEY?q;g%rP^Q@C9V
zj?->>)M+OSF)bKeV`K3o-o;j3o4U-lT`TYn_&G$u=L*K;m|&~0mQwzNOHz>OI!FK+
z+Y{_194-X_xn_VO1cDe_()xbg1-U}GuskUfXttN$vy~xbo28n_gKXPa9>2h2Q&Lmq
z$o{YSa-cyxYUlp`$d#AfW_K1BtkqpWu0W03-ngp82C^&@wbLKS2KXiB9PjS=9Vm4w
zz%oiW$9Lku?`5>Jpx{xbmVZEH=Tha(p1*h6ou!IA3g>~Xv1?53X8iwnJ_ULH|F^%_
zwhRC-4HZX6M}z&R(ea&&O_l#6f6Nvof!L?@%#46dPj2)E95S&71r0KA*aLS|6P4Zu
z{b!=BQAv)m@8bj_-{ZM~istWMKXvgOQaN+GX|r;Ce&^==Jq=-qFCGqnE>sO*S;r62
z8LQV6hYcJ)2t%AfS`{>!WvlmZKKbvVJc$U4!4FeF!NkPGfb6=mQr#pXl-QZmzaMp8
zl3jWo3K6_zYZ8;|5QtkDgpsRf*K1emibi~eoq`{)>QeDvZ^XnA4g?EiCorydW2j0+
zWCq|^+`KzdM7+A4;fkgWSy?p!5+Q!vrt{PPx<Wixn&)4oB_FK}dw}VGgf7HS=gxza
zB8CsgZczW%>tPlZ`Oj22L$k0ZLnkalxw@Mh<`Iw~9f83LJE}uBX3^*guf6AI7svnn
zx^#TSb>P0p!r)(!{Zen?-PEob<)S~&)SXcZ*e&!0@#%pi0F>mi$GyJvyym9*|9H7z
z5m>uS_vFReLwG8m$-=ig?4QmR|M*7WFxY;sNM^mSEd#t;VWX_j&Q2;+lINsCadTjH
z4uL<S6o1*uJwJHP%*Q+Z4IiY#My37G?H^a&S#q++$Vx61(^mia_dCzRBW?s?k>g9)
z2{Qiq(q~Ab!YoGk)6!yRbco6J(}O6QogVjJvkHC}&<Pu<U<mX2Kd-v;X?G#$5Zyz|
zo!|W9k*p2iRxuYseE!$<$si)&p;G)GsF-Y;C~PMxPv1%Z<ElF&AbT60KX51gABHM_
fd{zzSfNffI3FCn{j<vl+@K0G$O#v-$cKiPT<0s}b

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/change-point-posterior.png b/docs/2_39/stan-users-guide/img/change-point-posterior.png
new file mode 100644
index 0000000000000000000000000000000000000000..41541256cb13f008de78223f3e9b98e1ad149ddc
GIT binary patch
literal 37215
zcmdSBg;!Kv`#wH14l#r@(jgsE(jeX44bmVYARrw>gVNogl!P#(gvbC22uLZZz@Q>Z
zC>9~%cOT#9{XFklzrWzO7P@@qIdjfFd*A!Muj{(6O^T6$CMgjE5ex<+)z(rofx!?~
z&>up4@X1s6!nZIOoactBs?kkVO;ulSUw>0SCl@zmS6^p06HOIS85wC9j5o{0(a}Up
zkax_9($R73r2rq1f3Qh%a;k}A^bmFk`y4xgeO8%WXlI9`BgYNHkRl8;5q%?A!Uw`T
zv6Dl&qr}}NCiik;ezLMn83^(6J2^T|!rJmCbcvL9S=ekj`4}f1V9uA|%!IDK@YYxC
zyHt)6<Oj@{;!HLVwv!8)FAI?FS7UbH=_X<Dl)NAIv$1_D%&y37fPL>zykAq3z?~As
zMyO*Vprh0g5T)#9>>j3(63Zb{Z9*geBl}k%bt*|J2_2o%#zo^Z8FHrbRA&!zXF@c2
z3CAE7H;@|>vq`2hSzQq06m#*yg+FHTg~jX_;?N2ghM&s3s8M$_a-l0qrKcnO2H&*R
zk;3;S-~%&<^3$>ofWa<NLVw^M6?0%>m0;Rx%4Q+(-7-|6+4Ig(GPV<bCru=EroOV0
zNOCff51a9<QmV$!m{`ZuL=W+YDH@`blt{QSIjZ|Ie@Qn5UJP$H8r_C8Hp(~dnsu5b
z7=C}fHt|b&-<?DIlk=Su_uf9ARhy{!$ge$h24$x2#%4M7IF%6p_tQ+l``JdnDH*AS
z{_l?xMJ4*VFanhc`{$=y4~@v5P_kzJ_ah_hyWs!(0vc(@tpuNNt87aD?>*2+UdL~8
z|MMIns!H^xn309%zmKNI95J)ODgXOZfHK1R)AKfR{QD7+)#=6m8$~910^GERBzHdR
zzXQvJ;R!|&{5x{28-f~Rhd;jk??-3^?ZAcqds|MN7@WvT=z`mwe?L-!-&_Cpk(7Q#
z;HCvJZL<G2FiJ2TU;lk1>i^Hvfs?*WND#C4=CQ7mlasXUumdNB^!d^I8kGbh6C)$G
z<KtuP(Ia+wWP@4#<van){0jsZ$G)ug4ZjjS;7n0^ewU`krqk>DFqz`mRGC~(`Zv`O
z!d&6&4h@zqLP-~yv!6V9vL7fISN^)?x?)^=n_>b((=L4*{=ZYHby*26fFe-*cU6@R
zz^b}UXG;9fN0eZZw55+?{*9(K9`L>J#HRoGD4Wm-t1~o6_U}q&UW51E@Eaqy9Z)>`
zX}a7O-154~+LS@k`(YP@oP62y|BDJ|H4xBoFV0VYOm8M!axjXrrMwsXc}_j-=clZ~
z=ZF8^k&PPT{S@gZc>FeC`o*8K^*=9;w<?5RmPlDnabFU$<JHX<q>y%rt9-8ZZ&(+@
zZehexIg2&A1^gS^r2$h}&Gvmdk8C=tZ@j9Vd+d9i&$g@S$;p?#?d}(!C%OL{ddckI
zr6qIDpI^HEkP<AAA%6;niKi?u+eY$j_idLs(Ww7!6v|{NxDbWS<ew}KO*6grinCo0
zlTNRBXD&neZ)(<kL66wd9q-Xzxy!srUn1oz<UUZ7@!9fOgfkIOOg2|^*hN0`hTJ)u
z#P)(+nvVpiA9#CWrqU>;;J1d4nWs#pw=@rRrcF~QxsJl;jOHS0&~9*&153id6j+-E
zN?AASVDF7bU^lPt`RQ?+%V_$8X${{my*omDmd&NJ>>m^2VY_?#qtreK@@8Jggz)s4
zzPTD*-82S?q4n9qLvk{q(z?l~NNS8=_ufjUF0({m{dR;LoOqMp?4jdqww9CAeI|v7
zr8W{Wd{vmCU&zV`lXBw=Y6C=Axrk<=K&>AoPCg;P!6ZCS_eXW-n|ha@zpvs65)aKd
zuovm4D>&5Z(ciOC`+fM;H<XE>b#kmE_DZc15mWT(Hyx$5f#`u1Wd<#trH5DGOX@0b
zBKC30ogKd62x6oPtoV=De8Y<j#dBV{;4dWAWZtjKVbm8y><8nb9URKfQ0zj+15sy6
zVVWGl#JgcAas!PtN}q16pcTHVfTI7)YMFf?nyn`vb@j&{231uT;HQz`(#Ef5{5O)>
zSdiUno}$y_r5JgnZPVAMCk7Z~R-vT#Ql-L>FfA48SsIp1@surKYbwvY(L%dMC$Ad(
zJFu7Wu`qm87s2VM145UtxZ+~RjaM4D&a;<N>$w6gPud>-%)es$VW(vvU-r(nUy?cn
z&C?WOczgzfv|r_?B$Cwd-KEyKMoVG(tL`Mh-=1&Y!`{w@p$b86bG!So?n0UDootDl
z3uD{!jq?}VObukz>1T$9<+@_8-<da;l-L(ReZYr5YPIR{ubC{7nx0XjKPF`m&)gVJ
z+8U{pA%ZseSf)}+m2Gj)*SGjA`2=-_6+-I~KlF{YcVj;<x7Uf^TxjAIE|GexUPyY)
zcY3~$keG<bx$~Y8v6&S@O>ioBEmZTytv%1-f!W?WI}wY{pY7w9a@)cWefoU9hbSR`
z3aLa4bl(%H)b^WcT2oJ6{r*<{6Rc2~!V5!ES?#_tbj@{){7%60Hb)RwJ%WWqDHQ>C
z(X*75uiz?lsz(nk9@)rTOW1@`0NWOnIb3v{eoisR-GI5wYEh!m;vxGu&Xa~fg4lhN
zzXg)q1Jsgt@+VB;-Y8C?KQ&)qb{(Es5eLg2^zDNq87L}lKM$u7Wawp7rjihOB|49|
zeM~e{i2<3K|A#Ud=FVYvpE0>l#%-K>zUo;RJ*!BXwzDxMTg(_Jgfz!|A7EP8a`b|g
z%rboGp)tn^k*N-w4}EMNqxv=6)DX4RsTP8hoEQwjS@b27YblEEM$=@7Ub(!rL-?MA
z%$M%>YNIRS0{0?vOJwgfur&<oORc)vQ2)BfC^>mpmQc(5?r&b%FNCRXhL9A_=O|?&
z=OnB&Ly^H}t6!r$miMx`bancI_Hq@Qd{%o=@zRae!L&Ac8=`-gk_++LadXbwUyw6X
zg^eoi&Mqt-oj~}rN+T`%k?mt5WJA}jH{;0^9A*mn7dCh_d3gHp%W`LbeanO5{DA=)
zVe#4{^Na^hjQGsZR0pQ6%e!cIxm|PFl8eZ!UjO`bOYyOtMZC)zNQt5jPW;kQAP<gN
zBkV3uF4u2jwUjvRA7!n<lUJk9e*el*j>FeZAfn0cFxRj%l>N*(;r6^g{IJ0)q1S+J
z^(ur0lpuJKH_<-P`W#HEpmhxPqSDnp>L8eN(Tw&`+q+M*Rr8b&%#Q=NI~tNk{|3P%
z*r*AiK}%AUW7+6Dca$8gBI@^zBI(-??uX;8iU+}D=cU<?Wb$c7(-`Xvr6MF6srDvI
zWlX>HA3D&>_*V;Bw>{gBa0i<xQ)vl67oroA)@aU5t{AhI*{pJ~$KK1%%Qvg+6P8j!
zzCFLxSfe4$uWwfy$d9v@9FK!442}cN&~rjqqMxN?gsik4s>mAbrINeC<3>69u{y$;
zULcq<(5zT1|Eo5bQWZ2q)5A&X6K#YLIaILkE!!?H-vB#S57Q@7=4x5DX8y1Mw)a-M
z@Csk;E1rBVX!#Yvu)gO=zPTJjPKjxM<?r>OTw1(Q)>!C{<wk>fBc-A6!J4Obs};LD
zRC|?@!L}fpb;Ss9l2gM4<gSExPjl8R+^`_XhbjL4#L+&iew?CvSG_1g6)JY@Tt)y)
zQqMho-eCLYv2Vf9(2!M{41e77Hiigg8ft2-sHiBj-m=?@X}cX_&AS6UQ^gXxi5Ho@
zY)FH|FBd&md?vQ}#&_UWn8HP~dJ}bi%jTvawJ&S^Z`uUO@HH!)pWlt;a`v`c>Iv}h
zV|JRzN8);G!hUUUOBEFqcvwsB5>Vt_PF5qeAb-8gjfyWsXO&rWt#-aKCmh_~ZfueC
z#viiep2$ti6U~8QG6LX^37#46MC3a??2GRXpXX{eKbH6k`I{AgGPwN8ZCv{5))JWL
zSS}?g(!uW@JDfNaTyS8rOwQ6#-l5*42I~Ubwoh=^MWgpW<QVeP_J{q-UT*i$oo_I|
zw-@pgNh|Po%E<_g;x^%C7TO49O2VpPgFgF&lqZfqFFH*KMi@F<Kji`I5`zI*mu8{J
zX+VqZ0wU<5TwI5ne{UoCaP4{{33o*uSd93)FuW!seF8aE#4-h8->?C5<9@!aPb(Ae
znS=Ga4P|f>Lh?#d3nX`nJtw>lsc^D;b>e}I-;Mc(JU5vZvU0y6RF}Tz@DXM?XE~@8
z4~lpkF8~;)44Fz7DP~ks<}*ukTV>bNT!J*d%F2`NzQDi8#LFn@&Fe+0T<Jup=srmY
zaEiCF!usFXC?xbt5z%WOe4440kL<uXTcpw2O{SmZ`MUjVK66M|E}s-O@J<4`wHItF
z_24f{0hGKN*c6<}ZZ%rWByG(UT(nv)$mZ>x4=%b4iXdH9inEk9mJd!SL`WLL9B2F3
zhlWt-)|;+83JwjeorOnb7VmXs==$GJ^U52V?Y@YBKi2@Saaj{2mk+h6%-k{Wkf|ht
z1y^j{dAq3<<&K!*x3u*Qyq}lmYCkHd0Bfm*B?I~Mv~p5X3#*Fg@gi*AdolTBAnAGe
z<Z!*)JcwZmKwAWcaz>j+bCe!*5v@&a97p{4`0}|U+W^QpVi8CC&Z7@9@9(`xEv@oW
zJ>pG`i{yk;Ci8%a{8k#AE~|p1#N*#>3g0RVhDvXcI3h%iRnptab(LQ2Mfv1xJcb<X
zD{ru`?!Gh&s5Itu>^Lo$k{)_zC0?dbm)|KCZ+K;58cGJ_v|y-1VpO{XsA_V|_X~L@
zufMxA)5S8<57}NUC4k<BO$qTcj)pBlQXgS~+&xvYE^Ew~CAZ7L2|%>$NEhIOfhWrh
zC~0oTErRjp%M7t^DBj(M()oTmtbnlJWujP{O#ty(XivE^RyR&Wj`k(dShS&m`0p08
z`s~0jZ>@U+KflwMZ;FF9j2#EqFcP_d7{28K_v3GsAhZ-7IBgU6d22PEC&1&ko_X)Z
z*r)3R(NWv=b=Cfx<2};e=_z2CvC!fUYP%G?S5`WSX=AwMb*exYyVc<5=Ju1U;yRi9
zKAMJ~<uA*-4kaiWDMpT%pM-^YYEfsu7Ttbr&$mULg%w@(F!<z5QY)q7QL~qN&ZAez
zW~JbI8QMM~<Ty$vI`^1JHc(%FUid}DBv%#f*}c$gFHl$Ne-}@m1-c9#4W|8j(p6jb
zm?qS-RkB>Fsy~3cxV<c*fg&2s2!{CyUwuItWMUckt)IRN<j9#(mT<iUN-=@qh`;l8
z4^Y7Yt#)Sc&?dqq`m?W~gwgjN?k=0+lQO7<H#K?wa0UO*8@$GNKHD_d-1})t4YLE*
z!%ODUUaxuEhbN&851)qKlSz~jtc70=EcF}&E%*D8wDzW;?;l>gcDAN`rwpzVhOWvM
z+5x3-KNSn$oZ!w#^H3V2lubUg<H4sOD+soXwK;(QFhof&n}4cKtt)fg1Oe2_1tPi2
zbUG$SsROmN$M3z+RP?&ulxHQ*6{;VXL3CKS)}=G^#Dq{`R?L%^QpNo<4iDq1cw#ES
zV~N7>mlQy#jv!5fY4Z8ab!3Bfd6(3e0CP#**-U{VH1i5P%zyenZ%LSozQ~2FI#}tv
z+4yvKtzT|Nq4jU1fxMRe6_zia&Ll&by@p{-LF(k6YR}$V1SEim<=qK61ps2j5M#W^
zB0m;NH2b7F2+u~NnvzS$uiRuK{?pYhzLOWAlwVV|^uiX<Ut+4=2WkY(QyJ0itf%Zo
z<dP4l%H^j_Lp}~Q@Pw5cC6eV^R8M9YW5qn)?_{w?9VrtL6Gz_4Nt59)>hiYcGp$W!
zk`2^4_`3demoxu+8LR`lIi8m@oJ1B-9#Qx8iL$(w#Y3Z*A0L>k*|_eLY7C9*ylx3k
zHf2zIfTls->C#5fBH6sBK@|Y_kt>hiJR8tn-2x$d!71E*H3Uv7;G^Dm-S{DhGCrl^
zy=^92vyt(2>7G<~<Vlbh=B^7${bG64{xkdkaO?7e&p)OW3t}!1j3<&ZX0LpnZ!Acq
zzp7EG?;^%+RHH+)+;ywOF6wx*Eog70TIEAQJjV<fjerI9(u6Yk8mNSokxuVFA}a?Q
z9Qu3b-fnGf7F|lkX8r!PLolc7g+UBXX=vl<{=w<~!j6HjV&24nP#xRf7|tjg3)V(r
z?qRMEL_LoepuRl7X?6<1kPC1xOaywj8s&v4H6p4_EnpD5nwB@K<>cAxvnuP=nu=q3
zY1<oE|8wun;)hR$iM<~@?v6gb@Y?Y>QMNJh;ovw?g=|r}M!UxtgBBYHVy2sfMZSB<
zztY-`#KBwRW8Eo0DOR*>in{Xh0Mq^}z~FMR57h5{XrERHu(8W?*>?`G16PVG9<6hH
z@VJ)@)hxTqN{oU9N^K03S_MkILo{)hY^g>JbOozaOx<$Hk3xsyQE!WGAd^DD^fyyg
z;!oVEZpuWKQfO(G4B?(1=#_}yxP16FAmhd$vs}<vE}y9@U_s+_Q8SgwUOXkAQS#vZ
zl@QhtddHJ5*Fh|kA01Ce3z3&LQ9jA$@Dv5lGnNe^K`c~j=cxHkB%12(iF~uFuUL!u
z{Lr|{r2V+)Z>nSmsgfqODIPh;<2R?x^WtmNq1-SPd&sr3pRfMOW$7J0t1B3$tM{mm
zezinjnaIDAGv8=A@hpXE6B~2kIq{j4$KO;?G`9f~MhsLk#uHhlxncxSN?3+#VS!mA
zT*fZ*k8Gt(_jL1Lj=T%(UPLpJG(Ia!um}uE1!<QNv9Lq6{CY9~Y<DJ#?(K#~WOVq!
zu5s(n$@~r`^AQzoWa(0m3}22J-(U3weF0#68T7NuoHzowryZ|dkM7FU6e+uHt$qnN
z&^2y9c3RK?g()23Ab)7sV3cF_2#X*ex?gMk;XREa!F3fJ95G5z@}O2^gi>FzQbSpZ
zb@*FzPmRQ5Ft0_>yrLw^0bC^KLtt(XPYvCEKVp4Q(noaV+ddmeHw*wQ)WGklgCRfA
zcLbA?e*xF5@}EG-#0ORO2KFy!@}%->bD7aD@-%gXve_vc0BKQ1?u?xK*Aa@R`*9FV
z0~_>tVF)}z;AiuYYSKmeuBniWM5{t=kh$RCPV`Fe`8YFCaOr*~z=*w<F!w(NK6}3N
z&juU^**14tYK?rRk0?#$gUDKy)TFfl>RBErobP3SFg<QTI=;CIQN#_MV07Q2I4at~
zG?<1vY+CcJCO4Fu{o<5TI5cw@X<E~F3BA{!+Ji*Y-Wcx@t^&KLBw+TC4G#}5yP~4P
zn30%PF!TKEq@)~dsvFjieV99~YU6jw{tJSg$siDX9$&@v>DoF4w<WQ8la(<JM>DYr
zbcw)DAB&GHaOrE!AOSeaToV?a?!SBCp*BJEVc!ie%qjQOfwMLaY^io8so)mtci(o|
z`nSRp^J#LFcEL*BJ+nl^PT6p-9>qCgQ9tu;L6s1Oay10i;ef>Y0&0nJrf+eQ$p5gI
zp8jJFL<+=FewUoYx6=C4>EK@+h!5juBQ?6WlKMGfcp((~+}fprLmSq&=;)2Q1cxx(
zuEWhYU5`@fL{)6SyS)(IRH^%{z;Vev%Y`lGpIq*g;h&7oaz~1O>khxdl;gdnQ1uVo
z@(axG2a}mh4OT3xSSUzR2l<d5^~d{)O@~%fP{`$@9~h_MPYN8TrY~5sl|Dg)Wi=P0
z+M|Rn&hngmkmr{`;dw|8v^JBd61tDP3_r#lvX$td%Z3FpWHOr2WqS2gi*&rIw_L?P
z=Ja#Mh)`6PB#OMmKgqs|EcMdxcgIpNCsQ5LlViXQ9#BpC<S0R6HXdPx{+1BDt~l#0
z7kOINjYE2k{`tX*GEa`yq8I>s&`3%!!8+B0hnzS{I8SjFkYT!CGc>+PMBDFscyYAf
zp9s)qdy4doqgn6S8oS0>DZ#NHnpfQ6&M8q*5y#;-Z*GY?PkrlQ#7O`43C4{wSq$Xm
z2lC;&;SW(uH#AgMj|+pQc#qo4B-;Rt5xcb_Zr$dptyh1032ZeDqz;ozaq2~p>FbF9
zkTI2LMz@>kgt%JU`zm;Ryu6vGrxBW4)wE)Bsnz)#Tf2n+#G4?&MW)QvzF-ea>+$6b
z`R~c!u@%|&yDlvb{jQURqUmSu^fC|jK>2PB3*f5)tPF9V+GQP-A|cW%bt5KEBSPyz
zC_+W{`}gmcuSu%S5MI8WB0{+BeKiQlRxPStpyXA^(VhiVBg4|L=(-((t9K`PV&1YT
zky>bd%wlND)QeY^dOPDtZ#E%Vfa<OvY?~8&`43fGIROuz0D2Hfr&aG7(4&dsMX!=x
ze5>-oE1AU$qc<RXUl!cw;FU7?3yNQe++bb_-db*Z?%%O;5}w1Y&pG=e9OxFV@1CW!
zISu1a{J5RT9_CHa3P@HtAXzhp6mn2mx&-H8PiNy@uK7Cf$3f`nk+bGduCcw1mTEE!
z*xxaplsKUw&nd=meQ+%t&SRhsrdimp(<#4L@)rN$TDb4XP+s$5N;si?ehXNT4FI17
zA_9h3fI5Lx!={+d0?a|2x|T+XY<U`aa@YQv#uoQSh`)bImdqkt&r$jbqV(=sY3nHX
zHVcK{0|^y+0ag#i^CK6F{wj`@ru}cOl!T=4fNQ~?y<_QrA*>>@!_@fbTvR8enCt4-
zuN4vBQ)cVCLb-JFbN|t2s_*P9bZuHg?BOl);ULlBDJB@)Ux9LlJw3Hp{FPu$SoESw
zu`#no=)=R$brt!)b`~c}WdeBZ=#`yWP->vT(nM&$(5DC9ni<}>IRJ3UvZOaxwX&{x
zySu(Q!CIBV(eU=dqY~^_aw;(WUm$G$Ws*x%1$=`N-oB|=H(wVPWoUoy@`<yK(>fjb
z0D_`{Tj|%Jl2NE6%bwm*aPU7;XPfu3m?&lKT>avE_O>vPS~I}Jd?yra1pGRg3@-Qe
ztJ_3=8W~A?E>i1%I#sU9*p$QIy4!eZ9gg<BfiE5e%L6frjTcm}*vr#F0ogF$Oc|L+
z`Ema5&LWpEPKkpRR06fMf?4>`+@p|}9U;wqE&S`X<=#7u5kFo^LH!WFHOoeeEA{q$
zU8T22=*JV;e}~A#N>kuS#SGdby_<p@-jqF~2{a`RYVwEx8ZbF0d>z?j-xu`4)bJYD
zkFDtf&;+=f_xQb-+xXbBN5z!_JO32B=(ArcaWiZKm%ZPh;A?DUM&mEV4EhIb)7GwB
zDKux42UQm*44=kp0^4$;;`}F1==T>q5KK_75&+V@S&M`HnR&CFHF;#z*R_7HQPTL9
z=(C;c+dxeIl4I-BqkeuE{l~Ha6vV)%jK1JQx#QnEEqUCJ)KVx-sx(NID;Mt42+%gN
zME?2t%Fvv3W3<+3IMD=TP3O%*3PZ#b<&l6jQLjC#*I-7k>iBpLPHwwH>_7x5PlKs!
z?R}pv=ejjjrjxn-sxBjNZ^isZoq^0Ib&OF?F#!IT0e3gmO$MWBe)4lp_d70*Z2ZHE
z9JkjGjs0IrELN9i&u)=2N{$f=KVhrCq)upQX?Ht8{Kj1MLkGdLO2OTmlvJsP;s5Ex
zh;m^WUG;d*KVdpujO{8+a*7TQU;7($I9_$1G+Jb2=&zf5??Fk(12>fKi9XqT>^~*5
zHAX7+mJJf8>rLzUugw1Z{K^f7jhTBA8lbvCl*|^`={c)XcJD`*@465bk3lZKdBa%e
zqC?|$0;#1W%BFFvBHF=xw%x<9hFx_v`A<UcZT2{OI+LiVr}6+w>U7KgIsM@_mdUO&
zCbtn!<@zF<Yuk|K!+;f)pMPAS3HJgyhy0Ou|G>>Fg*1pJ#Lt~YLjb_dcAciT<ss5b
zn26><ER3Wv=Ap&H54?|$-e+v~?lk8B0_`TQBE8%9xwa(NVh02Z&62<Jrx)Dp9~Xhj
zjyGgJn$A2w$0lyEzh?K?$NtW@-}}R4=2VzVagn4*4bOGQ<P5iGAhk5uE6=oTNSi)E
zEv*9S7CBmC_9dNVkxZJ_zCY~F%5A((+DR01{DSCQzKl$P8z@|-9Jmzp{L{$r<4;u?
z)xlra)H(PB1#{yF$QlM7Op33uy?)jb^-JK+w+7QX0|pOT;a;~1+Pst!2QRNzwQ1u+
zUo@(PpB}07JL{#O7$oedTfGTmvia7KO+J|h1u5_JMS?Xd7*3ZzkC}8jT5JuvUY>Uy
zY>>|YneoUFOT{R-i_p`*>M><z!^$-AaR|+&#_{IS>w_i*Row(d&ruU9Z)~RHy@W%N
zxzV}wteCSDo^sxsi_PW}g;opVx-+qZm23RVH<X&DHbs3=xkC0f66mkZvxiEL(_b>p
z_$cgSFMOZ3ypCcak`4UC_UC9+QH}UZui*7np2`-vLgB_x{{go{FD4O3H16wk^m+T~
z+1F^PM|3tnAj)u~uFTzryNTfER}>kGaakRv?N0mkLRL<@QE9`XtY2zav2CLA{Va?s
zd5}4|U#!k0`@k%xSD3!vhXa2erEh4Y)M(KSx)f9y&-Ib+c@L47kX%x24E)lw<^Ckq
z%Un34M-wLH7fB`6_-lmGm-^I+uMcX(f}+~~VDc~-=0l=EX2>G3rQNxNbS+?z@S^^F
z+nFcy_0|iYbpH)$q~MgP`_<1S=mefnx|HFwA;55+-T0I|nV5HOZS^sHGtV-2kFJB%
zO5xcTWzP1J>Z!+67`CRt&N%Z}T#B0+;MQoPgc8XPe1pBc{Z%x6;Ggre4MwB%1Hxj8
z(~bLM<I@h*iR(Z{_*JI6y4iH%(Bg2dhA*KW5rO{>euW=3z;#;dNKo?9OT%o0aXv}~
zra_{AYmI9DcIA4=66xh*Zbv6SzaI2zSoHaiig_#F{$C4CoGyCuc1qIPFmKKTm(M<f
zv3MZ6Rq&egrY`s0*{N-KeMu@}HZ@zpr*8h!T7Tc{XYJ6LZDXvC3S}`FrhZT#E*0l3
zWfkvkB(p?8$ViT1LjqNXXO1J@8sXhGnY#S=)*J4OYX^l4??IPZ5Rz#{T}GA?-=i*2
zd!FBZ`Rv!#?H|<m<qp5#WGOb-yT9c(pP=!SKhEh2n;ZTLX&V5I+0F9sFA%A2y!T+a
z!1YzFzI)PZQs2{~Z?&UTWROrM^kjeNnV?(ppOu>}^~7TNA|U>qJ9UjN&FP|3E}Hxj
zed4vJW|UK~bO{oC$F3mAF@BNpnF$zY%(dq}-c>skL9|0$21^$B{x>zG0lUpW`!Wd1
zKN56!B3&O$-hs>Ee}#O6K=SaQ4v122OqI&$U}Ga`x-1!&<vrgBgcM5Ygx$!*M{5KY
zf~o_Q8M~5w0#d}T>PFj$pp(b_oXR=(t*aG40kuawMOj&u4;!dRQRQDS=`=CePVyD%
z&O#dsNUP3wnbmKp_zit0C&|AbGfb)G94i&Vxnv~@wq_>U66wc0Cwbrfmg5cTu4+&n
z!~}*QS&S8R1lIK>^L_uJOp9i_`qz5{N6&@z5(HKa>7EGNbsMdLwqMWpae+Hs*aF%P
z-~AT}bcugkYB(-9PZq2+_gh;=Qwb2H>7Z)qqAwdxaD^&XK=q;MD-oGSz8>m)ey3?z
z44E9mOB0}y<!uG+ezb0PPxJS?%cU>KH6%qFVLJ#T-wyvvy!MOH2YGc;i`898xF2Ax
zOx>wcEemy$DzvfDi1HBsU>7>mj1l(Jm%dR|Dtq0=70nm=eEnjE(naWIO3rQ=ei3oh
ze8?)uqcSsOFa~uYZ=J;NYwqQ7PgsJN6-HU96D##$Xh1d)d`}Z5g90O;W5e~~-<+$~
z{yuqi`r%TnyAP_8Z|=*bN-qDagD1Zu4_6B;@-Y1>nI7J=qkkKcd{#B;B2q-Eyh+UL
zOrt}!<cxw3`Nq<E(#CmBf@zC?0ZROGOr!#&VxxBtMShri)3L6^>n7Y+O_KFO*=7Ub
z)xSv8C9&}MzQ38V8qqA7MF<ZO{s>`Yz_MFGJwe>Fa*Y>%jGPAozFlL9$NaeDLxM@I
z*txOLWSzfg>Wu(Q%>Gk^%itMl{?%{T2dvcrk<AaJ6$(E>SmM#qhs{m(+l1<oiG4Y2
z*}l5*`zt3is$}(|@9J5LVV((@%{?C(ViX&ADk4nkA7@F0>^pWcT_V^ZdhbRg4BwsF
zUTnE8x<Pa;u@*Z??;4`{tb7vzXYN?@77VkzzkD0wks-R+$6PAAZ{u#u>=Je;HUSA?
z0w`fE2=1LDv|q_go13mO&MtnT=iOa_CtTPSZEPae*0-PQ=)9QgH>()OaD*_>J4J!s
zSrF;E$_aY2Pj6s4RE>AzXHopS-1^15CA9LlJ|oG#0%Vc<#-~r;1N0&h9GI|QUK3sQ
ze3)m$aV!*o+`nwllsE6rasmo5sKmaA{DZKIhXg4omWA7KI_AL~Xuj_zKd7cR!zuUU
z(Ds!+KVym!ite$uI4L33l=qTduR%31h~N}K=n8sPXD+l{uMpR{;@AJ@{6za$wuAIk
zUo;E!T6=@`yy%6gY+^y5cx)_-!wujB<)FM<qp7eUnsk<HHqQm0%>b2!4|J<dyJAAz
zpaG?W0ny4oa4DFc5FQW&@KKmS6rsf7q^ai|`%Q$h{8^zEn-;=qWTdUrdZpDRNv}vO
zK_Ph(Cc18l@C$kZfb%8^JsBpseWmx#O~W^Th2aBa1QiA+<q0IO*wy%J;Rj}h)}L4+
zP|8O`%|nn72_e<3^6=&+t5%gBk54oA<IZ@kVv+LPF!;{5QC8tOYHV7vbv*ow;Li{-
ztA2KPH<jp#0Yf=f4{Q&^zqx>#Ln9S<Xevh%ud|nR_zLch3<;?`J6G*JJ9wim%-%q9
z&JhDc*+u@ES1*oDq2l3>)^%SU45p8YDg^KY{Jhh~TEdAH9<fCJG(P^)_w@rP*jFwq
zIAUQjNr!;S@RKdqm4`L(9|Fj8DQ|<3l8_kgNu==SG+^S3pT~VJ78cN!VT&o{Ay1B>
zl0OdGT@pnJ#S5N-rJyF1UY7V&tw%jFuAp%bCSC{J6mD8|p8Aa8C2>du1;#&<BZ>Md
zVugSLk;KULn@#M&kbDbDsh!An0mKruL;`Gv<g;0PoUayE3h`8kV%ym17}<wv=|8X;
zqz$2K^IhrC#xvD;qB2r9SAE1_f%vIRy-<|mJK$%pp4_Uoq4NqiLaZO#!ii&ruj2{Q
zVHO)L4S)aq%wec_)<IPAV^)*(#b6wP8z{Zwge-5~T@U-r^@kV3MLC9sMUuP0(&|Dq
z9A_V>e+Q7TXkj|-QmtnQ8Y$0?+o5DHZz8w#n=Ze`zkThkC?q;59N{iVNH4iV8235%
z6Oe@G(;u1HUNPjwm;i$YEP{N4Z2dF9nzu_L5}8764v<ap0<W8;96G$_`A<fy0E0$f
zfnS%f!jNLqKDA8<nNFbFJ*ExBf|58)o<;Wkn36Le{R-?pZ{QfPtJPs6L;KUfydtvc
zY1T(3>Xo68ocd5?+7MwNX)ew8-lJlf7WF|4zAkQMbsqk;|F9Z(4a8hVNM6lw_-*b(
z%z)AN3A^vWS=XHNo$c3r1zTtrVs<!(i&J3E$Pjy}V30Ni=1qw4BO8e#$y>$#U*6W}
zUNQkNdcP2=n9msE?=oD90V09zMdEa=)`;Yepgo(3HXt<nP*t{K-l$nuup*lz9V*X&
zVL+@jhm~qjQVTZ#?Fn)lBr3?lAp7Dm?0qmd<$ew<E#a*CW!GEI%-(N9O+H`v#X;Jk
z^aG_Cu>h!Xb69`KH)l<DwS3|0&Ep&f>{O^DZESrjrXteWmJ<g$BIoixZ>1+pyBlXH
z$;h&m7NDnK&NDU#LHaiF>Di?1wDVjz;VBPMSqs~_GAKihndB8+7eE-0#FXWJYlv<t
zK$em(wYlAM6YX!DX(J~C-U3%WtQ1G^Pd^YabIta0G{NMO){rFvbcGtGT&&$8`1J0c
z@HlXDn6V(XK{G?Se@l^a{~o%mRiwY;M1snjhY?}Gf+)>>QA;LY0JXm@;X9y(ZGoV6
zzBt5u;XJVS_+!a0TQ{U)#F5`;AZr-2tY@OHBRg96fyjMVBJ}O;(G=sT*4Rn57&K~M
znH`66{L#u|v^}kvQ$fw<3aFNrjGG{km?BRO_(}xdcKbDjAAT86yTFt-!WZ)#xYXW>
ze=y4U1&U=Z)Ql*WxcTAa>%gUMGMV?OS91^YCKfhId)wS5h|bAbzXP7Tv0I(dAL}bw
z9sTEI{khHaUk3mon)w9WW&SB$1(M$O@93j_Cz}4@_aChy6OouVtR*jj+&UjAJM5)$
zn<2s*KTC(Hzy8GEax?=ahe!*e_yW9UFzi&MDmnVpi;89Mo4Pv@D|{02r6JchPESww
z?_^#xLHd!_+x4_jHV1uO)93%g1#2hvw14cg%I_ERs_T|5y{L0Mknv^7HM$9~PifO5
z=z}FQ`TSn<#y0}nR3kzDlPB_;Tk)MnV1X`tE(V3PTU5e($w&xP@1BpvH$MQop`{Hx
zwKxbCGkJ~ijuvd)(9XNk&GmB3LkdUfx`!|6PD%uzsDY74@zV9aZ!sv79Uo&JvN$=|
z3rOG@rC>*d#FoniH#gaI>HyxVMz%;%EEgmZi3ocka&|QoA8>STWQ6VIc>8CdW09M6
z()aagW4>)*67T^wci327kL-l{w;rS0dAjlE2EnrT{rp^pd@SVI{8DDtLv`vTgeYoN
z1-jepE|a83%3cXb5n;9&af|#BhgUTYl2qeS?3=DzIiO40<XPmm+Iz?Ua5Y$f<1k=K
zMzdD0=mE*Sw7y9@%CU5q>W1rRh6kh|!%$09`}%Qi3wdChTR4Pn($p;5bo<>Gd#GwH
zzj^w-U?3Od0Aoy*z#^4|)%i=2(?`F)uCI3}dv~E{GMH~@y>={mKr|1;&{9Lq%!&7)
zX;D$frfU+HcP$d+6?gkb*6p>F4{$xqDkJYR6ooKHNhgV*TIhsECk2?O#Uy<M@@yLy
z)<}Ap^24z)H+u8iJ>jQT<d4Z4@qmBP`~c%uabfTtkZ`fgh!|okI{dCd4vmfL=ie{&
zfTBJh<a|xulggsWiha>OxD0DS%z`U|x-AAlk7W6B^jSlMP)KF;#Cv5UkCNM5<b`Ow
z`$W32paYRdWgx8#a6m(OHLej)N-88MDNde_=Fg9cKW6MflDAf>-^ydZaZsULb(F%p
zCFNKZ!YW1Q*-IwJMJtdGG+^v6n#8+FV9cHLZqdWgt|d-q_yd|pd@hGT5gqT@Di?S^
znqjFM2RfRjl9mA<CygtzG8owxVxoVPtdD))oXC?)f=DdDZIte{yS8JfPe#)2t8e>)
zg$wj2aUzOyQ1&NDg;#Q1>)(wSBRu*@umd8-v=2#j?Qp2k4}^a1y+nn`-@LwG+#|>v
zknB1Nzz8%bq{dA-mI&K5blj=CW>+#V;kzc^8F-cveLCDk_(Pjb3|T4|DaS`u1aGh6
zBxi%5($V1D&^%z%79S4znJ*oP1dST7%W{+MGg8ggrUmKWAX{vAe`%sS@2<)y<x?vE
zb4JxYsmKtK#a>x3di$EBPFUy@P`E*&zf^U366AkBP>rQXF(qNR7=m2KJHUb<yXz2f
zcUC_xIN;Wn8MU0HW;)Z2Ww>{T=Ug?4InBm|=zwiiT>AZ=F^&x0myhf}JM@RWC=_+M
z`!;VWFF35!hoLl^sEcV;+;w_gij+1Mqj0)EynI+F0d=W%gY+dlXOsW|ceQl^Ovrlz
zR6n&CCE1DhN0B6~3i3q5&aB#Ofyf96f3nq2%lAupaITBg`K1&$u!2x62Km7wsPLGR
zTn1sqavhlA9FO1d-?qF@!wfW`n?R#%I^Oqu1|Y6P!s?@t1}Po#C*%t$DM<gUzdnGZ
z=k^KtAt9MaLC|Ap16bA=++UM={BLWpW5qgI)Fnwk(%kXV2E9iq#T}AeMHC&S-|rI<
z`Jy5q_vRxf7u5v~-z!rl<HP&xO@X|g1KJZ02S3>-fEFReky~6A4;3iXF@aI6J(Dpm
z9B^UC$Ao9TK+k}%Fi>wjB~Y}gM-7(o+0~eAY}s*UzWB;<Nybl)6Nd*nH4a8@W%n;T
zcAh4=DREOKUxlA5K*q_w0%XLGxL;sXj)NxmYo8{79;Sozplbhp>Adm!nGWvxwMD8k
z$K$KRMS_2ntIYq?s;FcYjIQOYviL+Gz~QaU<M#bTs{uPm@O#i7#)nh~pPp!ggpyG5
z3dnXhcV>*qPzvovEC!d0w6IqYcIN^2ICoyCBTgefyle8;Q>NH|J$44hkPl?p`c`&Y
zc1neTUkkR|{mTQOh>n9o)N;|KL<ES}f0GsXTP+1$hgh%=Ugh$AdhAY_0sOc|LfyF7
zYWa4u{qrawSClYAdgy<zkIx3z)9zZZak4YZ_^T~ru;oO@7O1Vjodtd1s(RQpVVk%4
z*H;NzTi}O~5pIarWa$uE<N2g7_$#e+=ME4$o}B;QQ9u5Vw~mus=zsf~|7$!Ln%6!^
zM5-8IP2mE!cY(Z|4|G%`AZ=w9zWm3y&VCu4rfzi&h}^2iEhFj70{@60TFU@id_PVE
znXVLK8u@2)vvz|@=dTPE^MHjo+ko}uR)!{o$gvWLC9}Vd*bN=abd2#oR^J`LrvQ>z
zk?-qc+djo@+y?^dcvXf1|Gs#{g0%;I{)-wt1)9WQ$xc?ukf^$(!~H2L$YB0Fblkgq
zE%MO5l_*mo^t&s_T;XkV&Q1~+P~Q|}42xdouzKY2nQBOwK2SqVnI>W*FJS<s<D8cN
zk=kVY*Z`Dv(mv9-qcXYR2AA6u@)ubZyvJgT2C1-z6sUE}DbxKQP74>V8HQ}i7a6xr
zUGER;bc6rn2Li0YO{OspvWon0Rp(!FNjdZYqFKO3BU&QrG=yF@@L^TCir3TA?-QFq
z3=cL_*iN{}EbGeerxEFG{B_{z4335ZRXC&#gZE|u&2QQeIPJX+n?VcCV&KA~!FTe@
z4=#3~7v8F}eY_*wUcdh`M~)Mh#1f|Ui?YwyeX5jCHIc-)q6IMM(7n#Z4*<yIMxFhx
zmtTTI;&>}zPR{b{Z-P)*_!VG*{unT;`$$7V3N(hcU)Q??CPUT+K4ioz$47QOCS5+O
z4c-m3!VXQnvwgw7q36e2dE>_q?g^QMO1<@q>+oJSIsaLw`1g<kP}S9S3XZcNex-#a
zMqonK_qU9zG-w)GVhjEQ=irn^(48ZJ?W#dpp8Ye@{5EKq<+kKK%aj&*0QJb<2EDn}
z%X=O4+`(NU*kJIRKR?Zy3`9lFGo;SG>AcCaq@HXE$uJ(!ji3(7uZN(=uYK+2>xWr_
zwq1H%-pkq4v|GFF)A^rY)is_{8On=(W&%eU<Pd?mo|?&{+SpR8xureV(VKRUjJvgi
zt5F}DLgJB=gg|6c=1AxOlU4oHGlqDujOzVGm3<W(;6yRPn(!V?v%_U|0txRv*8*Ar
zAgaEmu}NT@;EP>qOSSrpz_WSx30PLK==0MD=XPPAYqrmttlRH?-^btf<P?b0HAnx^
z0*JP(N^K>R6Z_5<S_5F!f*#CXK3}OEm%E-*el48#(I&?TO}y|fzgoy&TrqX+(9XKm
zIblV4Y(}>HWu$?{6VKVpwwfvb_|u@f2&2F#l9iz@IPe|`Q*rBMwmkja;Ayv?hPOfF
z5c{p})!yTEp)i#37SJi02(3czkA%+n-Fn+?40+VdNg;KdlN|Vh5F}5z;7P~`8Vs(R
zWW)FT_u@c<bX;%71$+`C-;HEyDMdnaskIQ7tO`N<^cGG>O`M0~Ns&FpZIh8MN?Nv<
zLg*%X+{MFzW#|?}{+CGT3=hCPzpKI$GHdW2RbK!mpc~{f>69Dv>%6nZ4=89pW}!vl
zawQu0(UKoW>?mRIb#F{^32<FhGIKF>s%xfbm;u@GBhvGg=TxGm7B(HY(&SX+S!BSH
z()IAUB&q6P>-Al<`lfxdlq!)3>&qm8&%EVnk)L=)X<tn$hv{Kc`XA(ou>|J;Upg-+
zX+)bH7*ijL$||+TT5aXhlwinA9%U{;_kjIk_411_D1_NY31@0WN4ax{zvY1kUccS`
zH>D=i!dJvKZa!Ulvk$c%Lq5h{V~4Z=Bj_f)bqQreGN3}SSFi>BJ%cf%f|g8kDncD_
z5JBW0_r&MRTS%*H8T<IM#Qdlf9FO27djnB(V0W_gdi{_BFKU<LZ?4B0BS`9oTi_^8
zs(pJb1Ywm{Ay86X&LkgVgtAH2l?DU~>eba4Rw!t@{d0VYEdohrzyq~{@eBcoyE&Xl
z>i?X^atwqP<eKyc&46)mre-saOu>kgT%}I^9xZTx*+7GG0}1=N$~H11^%@}uT;W>M
zZHuKkJMPNp00GP9+|VB%X~I26ma?Y(WP47p;Fg*Qk9zO<fnJwRckgc7#OsXP2fNGK
z2s2<?a|NzSf8bJWP*}gSXm1g@B?`9w{?7>DOeYSK(=QTJhw8*XUi*4CVrL*n`4uR=
z^YsB)C~+vy=QeOBoicmTMWoR{qV3tq;RN9GJfOWyvt$dB_Xv3)*l-^P30QMMIk8`h
zI=NNzTug9Gk5XC(D@3;knB*7)vDFb412R?aMo~c6)kp$iB%_a&g72Iiu3Ji}hTsd?
z_jYi-24uEjQ<ZwiA8_nK3P%x}t75=M=}WkD_A%AQO`4g5e{$juBy4l(<Qf6Jdy_K{
zI0|XvzRCT6t~arL1Pq6IC6fC2g4Wq#$rajcb$>Oh7i2fSy?dU%@V)?|^N+z{n*Q&s
zC^`zi@2<Pmsm7}i$mBI(TU~zzw8|ToteH3hoRz>(i48@qSU_44-UNPZ<B0_Kh>DPR
zsL-89kC9tOSh1>S5#%$2o-aMYd6&N#9726ygHd8mURs+c*(U(048uR<!kKv$B6>>R
zfY`^D#e%e(4T7MNwOBIhS+7D+bXnxVvi~c6DLyRJqa>o?|3Kn>@mVB-L}N$6bU_Pp
z07nH3{5kvj$e;kwvB_EfL^m!Vdr+|98d;li3>B;$BEG59gRYkUD6D`)b`^BAe?Jn7
zO=>p|3G}!O(y!mN5y<0%>4VUL4oU5fTW>5-`{3VQa%H0+kIr#A{wf*%;@cPxjxany
zRwU_hW={NdnVhoVfNuEVYxJMK$)c;8xCn7Y#p~&EU!-rm?QZp-_FIoU)CMN%vo98v
zdxINZ;GAJ)<XFPU#<MQF9{&eY+TEhu&#uRDcUD#^MOv;t%(>HT2wS-=Ei&@U?eNR$
z7bz?QDXRMXV|nOD`r~2Nb4M!+k9W$1U3T>p*M8o9if;&8oB?op0%+r$!otEw7u{wr
z$fF6eBIX@G!=2xKd3xgWE9nUY_G(n+4oQDUJk3Va!||&mAT?({QEA)P;Em_Th<EP(
zz8z?12~jZ&Iyo3n?9by*q%xJe^4b;6I{gKAYDws_v;}}~7|I4Qb)1=akS@j=ecB2g
zP%{qRkRn!S_gZMGH+!*?kzbW#{huSWY$6B?NgHO_K*FZ$-i63hPFy4U4(Lkr%H;<V
zSoH<nqT&2~KZzw9oWH0Im!0?zLq5{OH8eI;p4O~*9=|1()I`Kk?B7Wtm+Li<-iWUr
z97?{}M@$hXBC4Dji%%>!<mj06sQ+5nPjcx;E?}dAx&?6kl4noPXHJGHcVTU%XCTjf
zoGdZ{7Ff5Li?^uDy&Z8EuUmnYDPwcEe-MMGs=bR;$@Gl3La#TGDI6&Y_5U=PV*DRS
zTa0eejeooz)#nZ5P29`FY+WI3MBJ@?a8Ar5`igB=E>P#qefst!=@p}%l-^me{yru5
zRSTE`@-TS!1y}t@wbQ-<G4a0;IRwx(Qrgrv6z3m1UKt2*+tsA|f*$(#f)_YOsGfvd
zul1NN-@Ynt{Xb`u$t|EQ4m`d0Z7B6R2gMK<!Ua7akv|rhb*j8b*YEtCsOlBe{sJCO
za+AMs8LRgJj3PJuOONKR{L`dBx?Nh}hR403WKUhr$``V4u#>1}=&U=SD}T9B3};!p
zGl&00Bns??|4d=bwBjE-2(Vuc<~-=OY;h0+halWaN(!eBIWvhNfFp5h#SGMh`$6!_
z?xp6%kP-t>S;Ca^5pkJA$Ohgo+B<)HXTV!j0dpn;9_Q{tzl8E(mc=ho<mVJ^Nt}*A
z&w~@c%m5vLUkLtp*H7PF`VPPmQU1l&IP%0D*2te(SY_r|c@DLA!0BcdicI*j)LJDL
zsHt}!9K&slI0tD7q7YIspiaE-sEBl5dGR8L=c;!)A2Um`Sd7b#uR`j(;+xIJiXBO;
zQC4_i8W@lE*8nWii+g6uk_-|j?zxTU8ut;=Ta2bM{D6=hn;oPB{G?_O>fQziNX)xs
zr`Dij9$-h^Zvpv=Jbp5!B+<;4f5{CpEZyjGpDusojq$r^Cq_aiqC_iX_fY)$*gug-
zs0il~x>K5q`lGe1N~y(-YsA-;S_OD4^w+%Q{3xVHfJ&Q3J`dQ*aOJe0kPbmrG0Syj
zm|V7F<71#g*Mq|d-?J6~(;otboRA(aoi^fL135YmNXX`aNGvKpq6ge@F-piLHOC{T
zZ=-oy@n81{>Grf3<bs;~G_#pySa{%nP^E!$+T&ddV};BB?>_z0?|P3DrvQB`HWslt
z%dF5mS?=}I=*CR5_C5_@_?!^KPb6bCumTYUDAY^crAnle7cjH%A9#ba%^!e~hU(cv
zx63hhoJQ|H`Sf!{k&>*v-KdEKB{DrY?P%%RfCydJg@Nk{2?>d#g`ug~SUh4NVV;kQ
znXEZocm_lpyI+lQj-7;?@dR-|bIheZ78OL|=I|(3$N3^q|NUJ67n!6pKU7K!p1CMS
zT}E%7f$!;n?vWeVT+b*?5QH<NfAWlCLC-FsqhP8RZ$lyZLnHnU<RS4W53ThqJ&JMi
zV534wb77ga4YXUCQ1Ll1pOx-V)fs0ncCa0URE~OM*9ABUZT$K%(SM?Tb)@hZv=7)N
zPPf(=Vqf@h@VL$S7J2-$dC>t-uZ_*fsiy^moL9<B=q2lHWI6>dz1l_a=sySk*`7ef
ztRs@bPFy5QsTSGI!Ds3(pMf0GM1QdulH%FT3WqM3l73I(_WJ18;`P48@C9mJL_Jva
ziA5l-Za&MG%XU7_!&P!<IYI~tSQogT^H%re5gKszdFw!%MlJ$ZN4Rkt=h^{sMQs_;
zJ1lfq7kCQt%^EHAr(}1UobAzROLY}d4YB5n;NUlon#Qf@36JB?-0;8VN`QtqAgxTJ
zeeE#*3V6M4W!M)Ba#>Op3ojLzXRobx&Cs+lNZ2L#biKog__hIHI~w1K6EuFH#@jt8
zAdN7T@_Mb+JaGQIT4Att6C&Q;i^TFtho5FS>b+}OKh+KvmS&k6=l*<E&$}v4H?6<q
z|GAdk`>OL>)_5)3HrEulz>C6fAJH8q_P8U#)v2jJB7lau9{s03%sHlOqr~@V7G&W$
ziUlmw<mt`H-X|c<wW~I7w%GSspe!|LhQ7`QOkb;GvMY~1KZZO_o5ZNw6t}am1)1zN
zsbg*xnspDBuey?Ql4~F!KcL%VI@k(mM`*-Dj6X_bhLVe$(Ef;6Oc|}>_!*q1djZ;O
zckj4?^Kii0v{f2V4XicW((yy!XvakN?e`y6gV*zb3{5{`Bxpf@b?!0-aGlog;Szwy
z(@T06HX5kE+GZmLWz1C#wrJF+7Az1HA0odCEq5jD2&t$ZN{;b*)UlgeM26!WSg?@R
zFOdLs^0$Cyf%YhdDTevhOcca>dMdQ|v1Ha%7|LcNnF{}QjY%RcHgW_InO$I?l*ZbP
z6(QV~cg5D?d@K5Vbh>r9D}n1N)LMxG<~L#__BC?AKDi*7ryeS;Ect+G8l$W*6|vF6
z+`%a6(AYaBBdZRrl90O|B@Pn#Ij{na1K@eRi60*=lo(G>!G$Y9=LuDSQCjV-3`;sJ
z$@G*x{k<&^3XpR(;7oH()>nt=0<09&n7E%wUDdJS1>KU0*^LKJfN5+^^5X(<o?+ca
zE;}ps$FR#Gq)PCRzRPVjr9mq<_&j?4v;xa*gGmJ0w@NHH(C)Vw6EEZ6hNPJ?{t(sj
zQ~vC;VYEqKE5&{&@WFd=!M{Y{8a99=F&#<H&U$|h4T?jwVUfOtC!QcOe#AgkaG1VF
zf6TjEvFr{w#vb~)rofDCf>@{<+tC1l9~XB-Ib`?2>EOUAONm<q1b(xFQ}neiNH(Qa
z0NuC0T75D6iuD@L<$EeNoS;E$TYt8(wNH-q)n0!#{7T?}GY|HWkSzabxRVw>|4Txm
zQ&5pW6BbMRRln%lp*^-({6;p>(G$=PnO$2o?MttbVRl>9aY6sM)AL%^MDhKoZM_Ls
z9;u?8AHAp_PH93oeWA2))$G^pKJ+}8&hoo`g)4Q>rpf1UH}CNkVQ22E)@Y<LjHmFH
zzJB+BB)ogzY|mHc2jRy24pN`%BV8=;_1#=4TFfZ98pJnTSGKBZ`0>&Hq&EszoW3L4
zToIN-6}5O2xsk!ggk&?@C=Ue<qB{d5nk`~Gcy@cpEWn>fGFXMXon*B$i$T!%>za!)
z89gxIsRs<se<R@%e@Q)LLIsIFH>+kr5FB8A=;bJ#CVz_iMOnG5Pczt%<=2YQfex6>
zJKaf*!lbi{`3wF&KlyqofWP6GRqj<ua5uXWc8GMaMY5k>+h>vJ4KtxtH@jc?klbQ-
z*ZblZp+sMq?t9pj5GH88jy9ALcH9^RIyYH8Fn(#UzeRahvat<T36s3s;D)cgc4`Ou
zo1^Up>XI3ufdM5`^5q+8mrC!x6PCEO+H2l(9lc5;Zl?wMZ2DJf6?W=eIWB|@Qx)FO
z#g4V|4tp#)yZproY%!S>Kr8L^-_(iON0c-?zjsN11hhgehN$hB?E+`j`N?*^Vftg>
zoJ`>kx1ZD0j#UoDTkw~~AKbA`*>bYp04jw=khMre(g-4!iu3vsfulA>Gd*R_#UE&9
zQN3j<^Z|Kn=f`3zj$seZIb1m9m_9;d<)*Cv)XSdq;fEjz1ZH}~kI$kICV=7~<?wF@
z%qIpyM{>j*pYC<N5yTU;H<WDwuH#FmHQb(6Y>vvwI=e0<lDAwpj9d1A{#r|<*DyWY
zd?~daYKCcOZT*l4Km-aF%gB{$BaC^J_U=F@%?-KsUp@W!+Kx<NXR`yCD|zPAYvc$m
z$*m|~fwuW0$kQ=#^o_+~#mN<jp<baN1K}Oml6Th{Ry23Tao|L9<dCJZw0G(x7JzXX
z9L5Y<*#TNvLQNw$kO?4U2^zvH^Yvh$Ja-I;?L?dn^j6%7?t8`RiHtk}zyBhaH4|VX
z-ouL+W>tlVK;<K6|0Fb3-F<`c?pKV^Jt2en%1uF5u`H!rNTN!vkrW(Y7k_zr{9_Y<
zYvWS#2Ooirw{a3!-|2gAhG2S>W7+#yQF8bj`%}F4*C82kN~={cSvsD&DOJm@AgRWB
z@^MgfmJK7*kh69^AlGb<38m*9APn^we`@xFIMjCEuM_p0Ve6E53~DCSK*po(gRw`v
z1n1cIR>DEo<OBG{CVAk58E;*>UmbZqvMTr#%V!sQm17BwUaiqU#a=2vrH+YYARicF
zQ`}dy>3KLSQNw5W2C1OG!bey%qK0rF=l!)+_2S;$wC(s|{bvVp;mD?*_1mM&4Rk)`
zV0W31d;u1Md?-_Ik}*o^MIBFEXE4tT2;CpXi{`j(2%#2nfW0tG#5}-xm58A?R5-4t
zd=`hh5G}HVjw!3Pt0i>CFYQD{yB)>hzn%;(YcIAM*F`MdFWYV*A;V#AN(3m_1B<{s
zmmLSi1<6-P)!(3s@_Q_XBtyLx$}UVmk7fqT(?^A_E76M6!xChR3X>^a_Yc}5|6g_A
z9ZvNh{>|yAV-;nTB0~0vjF6QiWN+a}lo28;>sXml*?UB`5XnAN6d73|WMt3COnUB*
z^!+{0|IhVY*VE<dx(?@j&S$*e_xrwI_v`g~i+ojFCm6LD)^Um!-3eI#T@?Ot2ZFBl
z8R_9l$g8>Pb)g?c)o(uRI`P;_U#0m<*QCRGbhvzI2X{@`glqx*2sPiB>d~7{_^G3^
z{NtJe0g(i}87n5hjymR`JjO!cNQO~%)%)y4{JHxYgWJ2OF?1Ico5M~;I^R^2=rLP=
zXLIQ|?=6a7QL}8QH%mG|$GTjM(6Q)3pH34<9i|GxO*Z2*`VJEnthAp*{=0FjgB{D-
zg;{nNBqY7uq;f9~g};-b-6D6=pcw1(6X~29+_j--BsWldkt$KCiEmj-9E~7yQ#?E7
zd^XQObmgMp-M;Jx%Oghhll#J9kW8GweLcNv+8wE_Csg2y_h+Fu{h;iabzk;v05ug|
zm-RW1wiv(;Y5E+-B&Y~elLr;h#lG=8nklOL#TOLD;(r3;F;`T=z9&b0yrm~0WP#w7
zSY0Bs$#-aGi+!3N6fK?TB#f9!%goM-)%4-Ryx8-g>E(aZyyX07=6YuHW<2y2s${t9
zLA048xwlbKA3fO`%k|t8?elu4^|qP>sLPRJr3zife_I%J@}p{^Fb@wcsg5nSvpt+%
zyyr4W!{o2^`RpN*`(KI&$oRhzBaO#57I=rVyC$YbK0s@XiQjo&LJYCdM-JuO*a_GU
z>j@c**M@uqgAr8(O?6KjEH48>p3AK5B;T@?5Nv9+!7tA)JK5Qw%Ja4xQhWiZN~u8o
z=+h%RHcXZn4PYiktKI=yrGV6wl^uC>;c~g2#rJr-Ye>@CkT|CHGyi2>>HLX^zqS4o
z+@Lz<P8uTr%``oaF|F7E$De|8a(Fpb^?9Oi&Z4%_Wfy9-PeEf)5LI`o@+~1F*%8ER
zC|{@WxK-(=7ybyK1V6!OfdOsI7-p|@H0!dE%zQu)CD0iZFKaCbhRvqGmo1|ls=kXh
zA)^#jMlkJkPk-4N5Nh02t#J)-6nQbkiYL`vYUjBokMg1fc^!ZGCy-+**S>!Kt@B~+
z#-Mc!;gqTJvBWWi6Iln%sf;n%6EBv{X1B&U^Kf>zWblH7`H;jtx}P<wN7d?&u=q))
z$Y?iKXslx>_(C`m&}m+0MeVU&Cs_EDKPM?|B%pm+(p;z$F1@e?buzw9_LjrA9+h~%
z>Abncr=w;M*$;<3>+`)(6HZA#<%7Ee(pzS6w)8EBE9=(Tdy`sMBXB3yCE!qrx*Kt5
zUq(0-9kFcjqEL%!@7+`Dr_C9$V=`x*`U7UJ7c!rSSj~0P^PD2VQ75mQT~o$4(sroe
zNu(Gy_%%@~J!`X%sy>wQL5ZTkxP#UpViL`@{U6nukqvJ4Zfz0_%X@OC8{8?41Gxxt
z%bsrz2olaeI&B-oK)%s0gd8R8*Cd^-jvT6lF^wo4xOEDZ4wK5BdW7=l4PGkiXj*)V
zV*;4w!{C}bxC_)(7NaWxw7s>~lJ)VHf^6bgJdS#h@|9T~g?(PiW|p9hKb%cA1~DrN
zZ;CAHYyIfNRc`9<{nkWSep!AnugCeglzZ*P_<ovicU15$grrQ44Bor<CyNM~91%ux
z#s~mB!==_=cq@^1mM<wUd&_j<45(o*pRN78FDx=;eUd}m#Y;V^)EeB^x>RK@G^H%0
zq_dG{vFKKL(X#LSS~uHh;&b%}_kfq^I~?FLbU*M@s+0*}ld(N!Hhuwjj4(zb!w*he
zo;m+HxbH||ipX4fyn3Fx{?tBaSQFKW*1tp}L>l^hzM{%@RzT-vC$|h<iEUsYKcP1|
z_;p(W@09x$1P2E~BC7hdYiGbAQBj5)%Kcs5&_m2EK<H8O;b-Dq<CdA_3Ffxny)ofe
zyvnU8F*_CwYa9-!90WcV0B-qGKLKU}OJ%@kr-=l!vTn7rsvbhswzV))9a?!RSZH`Q
zjICor=!-uTztR;F>;1qJpnhqJNGa{z=VY&IyE7<>(bs_bg3(z<%@|PFI?DzwIZox6
z%cA{8K!v@UWNjdN?I@7yhCA)_!pS<GK9npl3#dK9Kmz|*q@<(@hQmDT?eEFF3py}K
zGwWb3Ah0(IJfw<|2d(C(yu{gl-iF+DDBxM>o6~YzLy(T<?$|N!;)OK?ocY8S+e@r6
zrz~-nivX6#(&=Nlof{3;SUn;pCns5*nH-kK9x(@fPdrR&Vw0)FSE~guP1Njgbp_`p
z)GhhuSHEAlnn>VJCBE)acJ0Fm#Fwq>j7DFxl*^QIzwq39ZZ52X2?t7aBq+k-s5xTn
zblK3HEJXxVS7>+0AU3+byP9WEp2y67BjKX@B=kU)u1)@fF(uIgB}OfMwjbN&NjEFs
z;kr~7N($zB2J;ZvGH~OG^2OyP>&FG>Khx#fRbxsdf_`p%wG%&AX3<8c-F}^4<r;OV
z+Fjz&bMK_{(vumS!_czGxv^F(#Ap~v#Wh^NqibrF?5v{IHhA0X4Q9*24=H0By5E8{
zcGBkv!xO2!9}ybR5WYf(bk0f}8fCaoym)0{t7nQ9a2!6&b)N*hH*yv=59Maq2uqqD
z|LQPt`Cjo*+KeZCS;J~d$!VT9=6=!Pi_iMdm9>=?K!p}~Fr%i)ZQJh|ue0=na1d*L
ztRFzhnkAJuei?8Kd+F?61H^!qUyHffDV$qeAc47&DEtw`wE27X69~cAjOnJi)Z@>&
z4Y}%8VyfPsPhECbgv^eH@Lu(88UngS*FxgbwcVn{x-Q}@zawHVh8Q7nB)W<4hCFs=
zvsXbBVpQxEF7?X)38vq<HA)<9U*h8)4m49yI;vV~@y2(uEEGzpLFbPjexokxfZ7Hr
zz1^xRn}Tn|MhcW9;ZKrP^`u6emue1<yRp)<2Q=B-50h_Z=lof(pqG=NBm*+a&;w%i
z*p4jSHv}sm%^qd<7w;Y^-v+WZ-&v0pAG&Uk&*Tig(_ozHv-H)0#4JSv6mHqst{O-4
z9CCIb_B59zV;*)f=~yWcyg+aH$sa1BgZ_6K@a{OnM?s1d{ZT=<R$}dwU>1mWQHe%z
zk<w;PaC!NiLV)gR)qk~_s}QG683zY2+qqj{Iw^*FyV+x%Zd7a0Tv+tdV)}MMgf&Bv
z=bNM1VniKf?P=|P+tL&tMNV_SYG~NNDESfKlKt)vx1^E^>shpV52lbXJ?nB}NGwu4
z^=esPc)P5j<xz_Fbl2flslI`-?=j>$R%bOlb+Q;X=hm1bB*y!LbL(g<Zt2>%c(!E%
z)?C(5nlZ5u#|CuK6aMc)&ptEqHIu$8YW1aPamI2&>d$t>@kU0YX^oJo!=woxV68T|
zw5pU$%S}A)3bn#*U|8k`-jFi;((#1pigLs+CmbDkg53FXR7h{&$J)G&h>%{eXJ38y
zI|-?gY^%@3$NlO~4m_H*HvLj$)IFCFIiIhe9B);d)1YV_LPqL7euUH{11jB*&_8h>
z01~(9+gS_QVqbDyu<VO!5gK08A2CEZOIl)814N?+EmMr_)Y?D1Zd{T8P!qC`dSblA
zNy4nLmn1(P6vpJ#2I=0_VJkDXC^^OCdcjQze}Zrx5cU)AsLu#mgdt=<+c#r}^+8cI
z$n0~Y3jn5Eo7aqbwa{HeIoSZX(p?nF2JHNHHu4la){)SgPbYXJq;O{CByb7ii1A4T
zvX8YJ%PB6nF^w`l8n4|P$ya78sHqpPtjlPB=kNSUm^$l3;)N2GI@Jgq8^nQ&7?~MR
z8^*#n>8CuL<5)j(nhWhH|702z_IaY$FgyfTWKN>ad7Zqa=?SgURbYGP*&h;naguCe
zjA_&Vgo=WlKldUjA)C0lT2o6p!}!&P<Rd-ve=XUrqEw0k_&)?%aNNHpv}9)?Yyi=d
zXX3WDDUe+r1biQh{d9s*X_I#3D^oPkq7App44_f}Z0HZTI?SDW{gT$$H67Tj^bR9#
z(*rT)KBJia?ab>xGsv98_Iw2t$-|EuB0gkJDCaM-{-LJ<F*i-K7X0Jw>~!APM93uD
z;N=642eVb%b743(CI8?XZ%3DbBOQqVMo4?X!_aA_Q|4aw4eso3Z~N&*zWtbJ=9=9D
zyffQAYs`O3`&Z_Lxs3SB)IGC3I;SYj7~mdj5$hR|UU&iO>?>o65nn!}wB)J0nzzoA
zSNf8=HJ7P;%MI?VG}e}`_${iMyncJ?bos>~^Y=hSW1rai`RK{%EXRf$hb+b+iPUPE
z1p)s0P-v-%peh(7JZ&$cud@6Qo@#Q(eey-a%RTXXY(ETUex5N}dJI+v%xxOClAVp`
z#`Z8=Qu5WOJ|8VJ)+zD%v0?~1K_@RTJzO$iF){TzE=Qt_C%IK@YAcwBq9GG{j1rpe
zT0Kit!2ib~eEl)1?*fgf7cDKJQiqRGjjR#xkKXn_sonGW@e#atYmEW=D+y^+0J4(2
zfctVTxsHX=xV}0Q_WM?Qi<%(k#4RbL1p&k@=N}}RUoVtqEAsdSO0s>l(o_fNJz1+k
zK%W95hi_c{3?6RUu7`#)H-S(AfjJs6Dcr8~H9R8mSRf!4H%x=|e<}TjdB55S67+en
zko0LGw{)!e#?^t&)}@=@hzjcW)?WhObT><nEDwyPsENEEtq^@VI&>*+`1?aQ;R*ux
z>#<I-ihd^{tft4E%uc=>eVRMTs&?pZsMD?-!kdo2lk&8BsY<DwOm!JBzdC{Mey<P7
zKd_J5fTEF9D-RUAy0SDi6@mc~1Qi5<b}IeMP%f-cc8U``KBp;dCge1yH4DBx^5S`I
zHrTh<m{KmU-hkYc(AfMb0%PXeqwM(;5*R#IWG0iS>Pkod<N=qUh0@aMlGo7~+==pv
zNP&iM-$JC6z-^*yiUTsp3pUqc?vX;P3@H#?Um|o^<y1+d(X*}W+CT?Km{dsQmsaxK
zQt5nfApof1gI|NwE#An#8$mFBoC|uW3^|U=+INe#6mh!V<lo?+*7KF&^20b1-74-3
z*XL)tVokhqjko!3qRR%q=I9!qw4y72E3w}<qa*B>HJ`^01e0^?JPtd)n%pFPQ7qN1
zop|d-gV?WySJBu2JwKIaye_ath;y_H5K^qjF4E)q`&p3wpj?Rb3*q9MIzL?!Jb^BM
zu^)7(<(d?PNn=cX+`mB`y9%g?QnV#x=lU0>j!5G_<Q#8(qeB0UsMA~+%sr%0m#*T2
zJ*}-y@pA3_p<cNcOtDWP1u3LFu|X&;pYR!;At8i{x5&BMEVXA!_o{Q{w#7q50#&IX
zf@5vH@J6)CRito4#Wp(g8kSdNCj+AZN~|Z$=rZzx0eirc&mUw_RMa0W^h<J%*If?<
z{XqfqwQzP?vY?kg0R~ipc%Ow}R1NG&VT}Vb(4*GI-BB0>=0zkCDNRm&4<%sfo~YA|
z=l@_<W}Bt@Z1(fn0?gghdd#5DwF5Et9HHoiQu!CHAv(WDFIc1M-k+23*eb?v4LtT|
z&4nVjN?rTz9LAkn7E=ur43Wuz$HXv8B7r-f5EDE{*mN$e=@U%Hi&Gr7C79Zz=y#(X
zI*a(L$30Z{khGKYeDDQ|Q=y&5qo4aNDM%bm@Xa>h=Ba#|wtZ_g7W@x)C)@gvA{!r$
zbN-E`ReK1mk5$k(a<Dc-yPYgQ%bNZQx4!w#d2GWnwd^SW)kx{QhRo0GNl_QXA`qq;
z{KTIqCS@l-ajf8+%tzC*0);PP@vkbxx^<ad^21m3-Vpbn(q5fq8(M1~fr{ZfFCp8p
z3id;Uq#@yuRC1lv(}#6A88M5fo#IW$%1<Sd>+rt~RmtDp-hO>l&yS%EQFTI7lD&^y
z;&;uDrqf+;X1AIhm=3+h6kqTQB@vwVtmJRci6GqY-g?`;15||!$)u!3Dd%UgT)Ha7
zS^;f00Jp>+wd-e1<oN^C<(pTpUbi7$dS-u5@ayl#^J30{v@|q|GBPsbxli4X{ra>o
ziIXAYRlGO;2$vjVs)BI4k@g58>w>%A5^(rTgY1me=Zt`buN8cshCANaQ53BNhUylD
zUoaD85~Ea|pePnqWlEs(D;gC$LP#oNyO%;nVE%-t`a2?3L$D5NaYMd8S*1;O@2!sV
zUo*ffUart%X6gWkl~v-^D)%=tqpb`Z{-a;Fl$^Nf0_yIx=#bW_$3ItYKDEB~Q^NF2
z<*w(aXt(j@kHhgu5*zK#+^;u#1S>k%bjTjgv3u{v%4XE>T{U3Mzt!s__(F8hXdh#O
zGWvxiL11>^k2zwZ0@pT?Yv3}AjQ$C9mm7gFi8r)jW6!XC?^t2SZoiE$JZK9_=V0`R
zyuHSQ*<^dausc7prOObtUJyKJ`GXq+@9w~rW0#|@E=o`%on%i)EL#TRPyDo34tYWX
zSSi)kGXBm#Xo%5gfw-T@I|bY~2cz|%CR;ciYDr#4q6Zx*75Tl3qjaMoTtyTLJ%G_a
zp4Hm-JV8!!^e7?QSO+-2<vTN0b>G^x`T?{F!>g)5n_y{yN+lLy!f0gO#%Oe68!B$V
znkKQ)Z`Sqv+A(TYiConGSSl-o;!Z3tL)T3)Uxh{my4E-$u=bk7P^i?WJ#N(Lf|klh
zz?CO~9v`{+&D+GbYW95Z6@Lq#Y<<@qBk1gv*?luc)K`_j>WsXGSN%H~;D1ZXVp4DC
z7LIK8dIL-Cmd5L752JH6po}wzn8LaAnKkz2vmQ#<u*II-$Q^5F7vU}GB)Y@TN*O~-
zlY77Q1R$znzpvVq+>VMB?yL>5eRUDfIhuVWDv1$soYFt+%^>b<qY@_gK?b6(EX#7m
z$U`L+@x_bRsf#oS$xaE#4H&egO4ZM(K?5`klKy?J#nN%S^DrWY_&o~hjdNS~0p3R3
z3f^tGFJ)NVLoR`W1f|oE>L~QOAaxYtGmzr1l(WQH`I%qT>o=3JUZ0FL$upNVI(&cm
z2}mT;Uv15y2&)&XFtzx<j@VD!(WAdTfyY(~u)BV}S1nM60v|2%-gpJ=$iJ+NxewZJ
z0bhu=K8()mu|$F(PuREkA!clak;}d(R%KwK5P~j%-w*zm`}XSOl`k*ZT|S5q=7W1I
zVrWrL+Nr`U;Ied6AMA;docH*1Z2d|jX^7C=Er$qP<YHBS(82j2V<rq~g;ohZ{CT%G
zUwpYn)9Xj5$p<L&xb(!{9hE&<<o)yWwU)y<cHioJ6-J;LtvQ)q1-@ax!gy&?Uxz@a
zlWVEaFR(djuhEn|QcSZC3IYj7@-O{j66hC{ca<2twoJ?0L(7f#=@;hM#}gDa*nY&J
zx;(%5!EBdB_64a!sE5@{NNDXqATdY`npaK=A_)-=4gG_W8{zIkyI6z^)&+~u(5bWt
z<{#RA0rwvRuarXIh&Zw4y<F^2c;cmthEU1zl0+O@{D>nwbbG8ZKyhu7IaTjsR>eFW
zIgPNrvr?7(^o74R)Qkq`<1UfR{cDGgNiaf6%;XYzNAxjkIPZJ<k1IEIRlY)1Vlz^s
zzg=zx6ySH-;zZB)8XIMG<17q5!f46y&jrYn^VklgP}-;q5-ZJ@8n#Y&4IaQE?d-Cb
zo%^jDlUpJ5K65@BSk--D;S57TK6GV;+ZTo^1s|asFWp`^AIQZbm^cKgqE>LHdpqEk
zXxvc79z^v9NBu7CZNJ4yG<O>j!$xZ~S2I29A0k;oC2zsT--?7>Lu=37uMsLmV!q5D
z5G-CK`>vWnEyGC{YMK6o0i%dt?LDmYDp38GKk<o>U0Z1`9KD}jUSk(i<5(2Cx-9K<
zeg~RSjUO3Q1jO$Gaq&}%CvcBQt3O&sWCi!v@9<&~2cpqjPf$BuukE%aKybE(t{%{f
zApP<GxShyS=?O}p5OIzPuZ|4E=YWzsqd?;n-SU~ej0Ms5dINW!qW@OICUxWND7681
zRN!Y|DINq-NWB)EBhA<@>7;<1D{~wLCiL&C+rKNc)eT@9aHkVEw#upyEDtF2f;U?O
zqzbHG*6JV*&{-ij!G9+_?KOjZoK10oJ6-+rvWAeRr0&C9bZc^_p;6F{)WWV`T*1hx
zM^O`I;9K#uKO5ZB_lhYdFpbm;1;vKsyr;sBYdx|5jC<n;=Eu1q*AOe}=NX^Y;&2;*
zrO-%{EO-5s(2Hc7(w%6E-BXR2n`<+}-}_g}fY)-%2xMg`*maIPanexo4}N6dcClmU
zAET=|1)-EVR!t3LEQ$3W5Q0Asc&tB_&X9s!>&1uu60<DuAp99PK5?d)1DQK3i6EZb
zP5rZb4G>dP_M5Yk7gP$*o9@m5dh+d}LeRAtmIgoWE~p@rsP3GiCB{}@JW>#Px?Z8H
zRr<u|TY7Qe;)!H}X~1Y`;qmKD``Ip=FK}OSD)Fg!%<dET<W!JIO)WDl_e1J#G>WOX
zf#ahH+@0K(PAyT+sB)XCZ=^fBQ4q;~aA;|1F0q2Upe(*b4`f&Pv?9==%z-az#nuRh
z>o$Y=+iTOULzRxveB3;7@ibFDVHhkiLcP6iz{!26(f+gT^b0Smbg8JY&NY_(EE(;C
z*laguT~N7a(LD*2w1vF1ZD~5tAB1pVm>}P-(}=79w{s}fD-9A5W1SX6ZA;I`VIjQh
znqppo4jLVF4p!v>dPNd)ju|)7ApE_MUkKB*ma3?$wCpsh1RpAGk7{t5n&-~SRU{8~
zZ6KmW5RJUd=?vdXyG>9PIx|tOyp<_^XypcWIgYNv&jRH|$UjOn?gZP?HHry-{GE~b
z=r)7tZ!LffeCTPG4nd<rX;A-UxoEjer6u%|v+Ctw0j$SfGlfvCO<?rhdW>Z3zCqdD
zbZ1^9A1k3Ox_G+Ycv1G;=n*I2e4Ta<ROR_=#0O7ENqU*vKcDr@%>=<ah$o`ho#y#I
zyvM5s#RdL_lF@k*VkSpK8kdm|?3IsX{7IOW=bI`qn_j=Ze&aN`9lWQOvC%V5DXFuc
zq?EAWMaySzxHF=3Zqb65AKR9hVgzpBs5D!7e}EkBsd(eL(Oe?L29Z)U(XS<LKjB6m
zF+N3l6cyu1tA8#U*LzeN{qeP9DuLb3O275@j+VQ`PgHqw1@ZgFys%(kZ$PVqbcuR(
zZLQGI#h$7)P`Me?zew<at_si2f{G=GxIZQScMUK@4HTrE{1TT~nL?}wvqMKosWO=7
zjk@^+x@YDZVe`wIC^B2g0Rbcc$OTKk=3-Z;vwrQlu8dCr8+YlijgSyS>1m-Nz5U$Y
zg+XthlKJj2&X9}!{@7GJj!1+|c?rxA5Am0ic~dQQ9J1aQzQ3XTU_r!cTflGp?Kk#4
zWoe>WpEyC=T#q=xi3ss+b?o+&e^>qqBWfgbL0rmqo2vMn@WjI-Y3db^oe46|cvH!j
z9k3~%9QEnxrF`y~3|T!N?Hc$w+v}7ZvIpfy#_*q5ZhZ|%e%XNGtNAkbio<Gvi^lnK
zW5mB5W^&LmA7vf2@hjO|3YT9saE?$pf}A@1gxc@t#UVVH=8S*zPOQB2%QVsT^t(E$
zH&=5TLbjCO66YibI{kjmPnS1PnsLxifZ&`cT?N<v;wrh0vPEk6RWpIlXU-#c&sCW*
znVKr1`uN-&2R7j0!=XjEgr(Knf3uc5D6;1kmwzF;I=QKwr?DGAEI+xCDn)XI`xA1<
zke@H$E9O9MZcqJL$=!b8?=yh|pQKECT1^>^nH+Gl6d7EkOdl)BKS<^%u7hRf)!T!g
zVhEC5Ey3tHE{exc_#Csj$loP9Lk_>NN0oKK^|LM_igT(|$!C@JZz@r=Xh<z=AKTX}
zz*pPL;^+1*La5V|8gdNo{3ybX7jRa>A>7Y<%QIQSG?b)&*N4pO_@zB6mlM=rgz1&O
z%WGSKe)C{f!{?RSE3E@XJmNx+XfF4|mIC8Fl=Egb^sPx_j!h(hzrDxQB1(vPw+(>e
zdtxC#{Af9dz5-wfq31ntgXzcq8u&zd5WTRZIw;4w7~=RHjR+<b*jn8Y5>BMAR3bTX
zL)t`9sgT)UUt~tGYZk4P0xwbH{LdWp)`a=R#cI&I=0W%xHPAXN_1JdsX=-Mt7qyp0
zQbv7fO~Z6%dodZ--gaK~h?Q{uA^9at4;kMg;)Du{p?GTT?fbd|290Aff1ge>I%y>J
zjBcD6QQKn|2b;9qx7$!;N8jrb`Q{P%0E5y>mGt}`rK!Tv(##u-;ClV5bP`#>#Cx8T
zL~HjT&4RftQax=7sLFcISMqHPK~<w4zVD-rO#cw+s!g{1Du*9>+^{J^_?f4`q}5@(
zSq@3OZSA^F<$O_fCJ=y8gtwt@-~f#Yp>I%tt^!$tV{8s&+_6ko9(@q{a+8@Y$=+As
z%;sBHiR76{m7Wrnmha=S$p>6mP*8|%vbh)@^}biqgN39?1;}>iVfUC==sVN4bs-(;
zP6E)ADniI(fm5Yg=<@4C*y0KRCTI_U6zDqS5WNmVP<C(u_4*kifbE@le>&SIp}d8r
zX<7a{9|s*h0wnG~#!Myr>~D_YhtG{AW9Yd1)tHijnf-oh8Cro)z@OaMGRJ-eGB`E!
zv{^-E?K~~YoNFaSP7vXP10_5oq}!n3E^-g<)?<>BJREe<|NC*5_M&V()%ngUk<*lE
zK@ky&@awQMdq5C^TG<ELXopc0aiweHia8D*Vmx{p`EdRvl}+1_F-3zkOzB`x8W6K1
z2F&vjk6wa)<H<AD;ioR333ayYO!Emlb7^1XHiuZ}n!gL8bO4SD-RllzM^xNt7Ln->
zH9`y@qi`u~Ny~k99Nl#*5{|Gl%$B>r9jZsn4^uj6_fiBBpY{{${n}in#YliWMLQvz
z*2JUa!w3nc#J>-0-F*C1`b|u8Z;;;14dT)%$SGAz)%Jbf0j&!?YIR{O#m<nT{V=W`
z3O;1d3GQ-Z+w%DNb!^>w()5nra|y}2u7zba>I{aw>$Qj#fmo7~S<Cwp?Bt<Mr58Vm
z->KkztBEp^eO(X%1FM1)*hRTr2^6-1LPF7+C_9Nl<;f@BKCy2qc#UZE&xpf|gg$p(
z?U!1bDE?%`<;UTkzjiv3EeHp{=ecSC*yjkiPSdxYDiZ}M52r&pK|1E}7+8&|f8IvO
zGk8@hvQa6hJpmc^gAHRiIU~2^Pd}Tu{)zMVlt_Jv(K-J=Jlxj>0Fx|>r+D!r*$3MW
z1jNpZ6C&Y(?1J9TV+(FI4+E?(DD!$jm7WpU?@Afebhjh}mB=|Xjg_Cc*l^wrk0Vsn
zPw#L8z4dp*D@*+<c;@E6@^mLxJVa|rNUE{A^<#qm#_=*On2e@f$0J6lbGUGfz2HQ{
ziZ#7{Ojt+3U_PsLR+n4k+?cOvbP_k$*c7?XQw|Sw*=VS4yZYADwM~VN6OxW87o%5d
z=K#y#@NkUmXTGfZ!EwjEG2Fdi56j9f)CN|+_ZUt~dbrM|C;dJH-)#MsA}J{oS}M7>
zTw4Mf7}|x7hmC$CA(mer+-O;_)Z-(kJB80z-1@f%&n;Q}^_6LGJiJa}P{8<tu?t1x
zp(?`_t>SA><Z~C=238;0w5!Y9)-1G*fKZ`zDM!IHG$4zqfR*h$PnHN6uU<$*T!@i8
zg3%eBbl5vwiG|W-7bUk~jc!g;0L)Qk?ov<M4k?N)Q{moVnfl8h-MF~XxiOf6I`;{f
z!VWY7Po$GT@vSwPsa(V#R(6=H`#P@K87=PxJ}EsKjQ$)LU66TkuUZlXCM!@hK_GhW
zD{FRx?gGi!X;Oxf^eO)%hX3%BAaO$Z<Jb2NM~QM?eCIGvw~&n(EkM^}FH#OIPE3hE
zC2SyO>W0{ea5|^Azik)TYW4lbEpj|5T1<%L!Sl=Dz(EibZ2Iy@wZxXZ83`NVBFo7{
zjTMS_Fi`kg@#`lNqmq_T$EFn=s&)G>Ax?tjq5R-J`~!Q|?K|ZuAuFwt)i>GDLT&|?
zS@o^vn)V#1mA~9L*>~2EP^lYFexi~5b>_Lcw2gaN5o~oC+%ZJz?Th9zBznP_Z`V^$
zI&Y@l7Knf&$_Ib0)Pq;?Bzif=j@B^Uo>e%9)S6G_ij|~-sG@LA$~S<Cf2H$0*+}vd
zN;%sG^QejUZv=dTTSlMU-Ddk<LR;+dXR732H8~zcNO<?pL=r5C_8X>&1lcwKmTRQs
zj?L$Tn>*72NA-Ujcwv#f!6>5%))0H4Q@~$&#x_)9)~tA7h+A^MlV+5v*z)fxb^JpD
zIHVR(R_PKpl7a8J==C>U<sE=t(p|jEc%MWtt+^d>5rZDU-4j6*Eo4Lm6rOiZSun~)
zHWd9m;Wx63MO=a<Kff=vh&1^rmeIPs%1a-U&2;$h(E^|neTq3%U;Sikq@)JkLH1*|
zfQl>z>~%C<2FMN~zus~CQUSlow>Ookie+JsmrpqT*ioT6z@ztl6#gbK>E|e>qff!&
zzQcKLZ*1SCoPp_+G;h8J`H?on&I!&_0kX@vaqvrG0$Jujfed0kUUv7(9#9ChN)n4h
z`GZcXFui@d^!M!Lf~1plEnr(Fe-o`or!Zrg#Y%s_%JM`fm!Lcse`TG(6>^<H6WTv2
z(s6|OquqP!{ycx{?kCa~=)L+{_haF&^8SeK-LShpNX5Q>T-lIOflyqH7;_I3OX46n
z(rG0_NM9J0HxqQ!8^`>7Z%kAm*@BZy@4;cU^c;SwGT)m9Bo_?~=(3~-F<lz|8k#ni
z#GO+Wsc)BUHdYrV2NrCdH{DX>ol`Hnr?l3!OAaijIHs)0mEU<pQ_v{MR}ese!Z6|p
zP;Ml0C3hYXaCICz@}5GPfDla~jdew-R}o$`@_+lQpji0t+c1nz&P(H><z)~5d(AIF
zScM6kZXVn}42F?HgONnU`#hdxJDpwo2dvy*0iU$KapOkc$cQaHJ$=?7_1CL^kN*yx
z;jnfGRY3VGh;?}jRH$W;qE&q=D&n1ASSSOsuA!%=r-Mui6PzE0Q7sEqX}c!@Eu`w5
zx%~W}!>nh8&5EC0=A4SRusxO#S9t5_lS4B5M`p~Fj<VfT2K4x?diyvG@p6OOR@Y1P
z1GLClOPgclW~VyPdeJ}PNhL^xvTR3cF0BX&Q0`xLj<Ao1?Lv}jyp`)Wv>vswQu(S(
zzP-iKW5_K@BSX%?qf_R{08hoxt$f(`@9pO(XIivg);fJ(AJX%0YioNuc$ka%?>Tdz
z_NF@G!|#B4<ppUiNPq5t|9Q@wsQ_9p-pb8?&H6X*1oNlXrTy=gS72O+cOQK|$hUD0
z-=g)(yxq>y^1k~SA*-nJ^742|y@xCZ_re-gwe9HQQo*BN$}(@5KNVUVcZAEjH)i=g
zK*R&kDk(@$Z=gGUy0yY=!+-w~<1xfFTjQbnn+tp!2^Y-d`y8UT&JpSm^(sa!2^h;P
zoFPH(4ZQ2SBa^PAx%|w^tECv;#sZ0&po1Zj1+8IiQz4Uut2pgERq}(UNZ?a|8Yto{
z*a)q~6qo8;<~z54<v3{$B8``v1<|3kC;hZr;O+AJkA}b}300M+_GjnUFYBpWe5it?
zUKQldB7bBLl;$y7xJqk!N%F$Np@ZL^W0l5H=gv*F{X6a@^vpT#K!IO>2PuJ14bkDl
zhiw2C?d$F~`=4><_R@ryh`5+DoksNV!SU@;FrI|H_dMno7axbt9PUg`PJRupR{1Yq
zMojCjqKy2bZ;GEe{dZ(j9`7u?58g$qnWHWBzpbjb1qowbUSjw=NTjiHL}31C3GCQU
zu4KIrkWn1g`+@Xe4ydX4*fd0kzx}pbnbfQ`Q->)gi8j0Z?<(`8B}JUuvz674<>4Pw
zVyQ9BR{uVC)5o9+zDVc3)M*>?gBw6Xm6MeybplJh#e^05cZ@pe!lB)Gx8jgL1K7}m
zPxF)c7`%Z`_wh>-6omYHXeV-}5~aGY>Q!|R8UA|&K_nE|t7YSEU;llt7$K^l3qZ1i
z!w2OMm=&Kh_)&B3?=SMMW3gB~IDgq*zI?f_r|0(K;$q+D&*~+1J_Px<@*f?#^7q*i
z68~IZ9{^5wq%CT=x7YlC?(a)cc2ogL?lbl0Glhju@)X3ws;b0dv#-r@JkGu);Bk2t
zmy;cm8&h`Z$NIbND0xlwQ*|j1)R;XbXZQR_X?c}NiX{4gePInvgICr?l}&fFKbZ|<
zxwSuimfMncUK+9tK0AZa)EY9hv#}{d3aW%`*nY1qL|pRdjFHcyM~64LbiBi0>zYGK
z0VFdwyDqH*I@galFQs4mX!@95ybtsdc91q#K<4|H{R22e0|xANDnh0i_-lP(g?dLA
zT(M0~O4}$Ny{sPcl0_s`9diiMW;Zxbm|U}?(Mb#P-RjmdUFQ=z()FdiH`4fQ=Qs+C
zw_8->_#24yAn$*k($dWEP?f~j6DQO)-3<pWWvJg1mRj%86gF!L!#gljV?Xmz>(Sr-
zH`ZA(u;<;SX;bi$_yiSZ@~fMdNGY(<nu(r&;{Z;Y-3PYe`Wkm(Kzzs+>{`A&euHg5
zi6k-i997hcBgmI4wxKx~N00bA2M4XH2p>pZ_ZvYRX{R;IzMEr;6;(bR{TJ~?Q6$AQ
zJO;OUcz7;}yaJRy<nDWjMEcrP|7P1$nmEvXI=Q%ft*BcsEC|eySq7k|-QdL;zT?>F
zEAgJlVgkcTS{)5dm1mCP+hk!Kl}}A+81Z!H8T0>!MV$0D+JXEb)UZ`TA|kQNIDv&=
z0<J@31OLXO8*1Qi?Yz~M!<D+m$@k^{{hA1s>o+3w+1}7Fah6l|vW{$vWQiQ>8K&mH
z-w9uwiMiWZy|=!(nP?lGH89P^L5bDYxM$COa0CiPEclkDx5Vc2n-#aPws{>~IHLdx
zt@kw7YJ~mo<i|uQ;Ry0?b}bzF7loMdCksAk3E5HY|H5aG`sB%z#ZWLBs;Wi^2?^=D
zyH`6sco5cq`aGWQ!b9h<gX#C4VrG8+6~Lv2c6N3ijz&gCeFFotzD2GUr)VLJF$-7A
z9)!hll9>@dpc%Mn*iUYzxs}LsQeq!RN&Nb|RucG_l7iT_OBg)%78a@8525wm)0n&e
zyD`E(ieyDUr-+MNHYl;z{{7a*2e1%Y_sFXQ41_xBCkHSG7Z>J_husKoh`1G%$>Fpu
zg&KeSduSu5!otEAiBd<ho|4sIihZs8B@RMD4jmzp9+a@9F)=hQpI}FU($}wrPk+qE
zaJ{@yXq6Xv=-+uNQX&In*2LSpUg+XQJu|b3AO<?ZzM$1tj{kz!3q07fg;ZAhRzR$8
zcu??ffx)0*a1&KO2|c)Z7%FaQoHjo_*THlkKsCT)YT*6;-)kcH0JU;+$iWrkq>Bj%
zb&5$V|9g#mfT+9V+`*MIQos?Z7i4+=y;eg$@c;T0)I?4yHSj$(RPg}<d9<U;0++uu
ztmVkI4dD{23W=8-ErNtN3{-qlQW#*VDu=c~KX|vaR9bIP%VHq}zJ{F~5+&S9+u{Y;
z-X&1+WaIk^4evqIMi@+e#KEDxwQ@dV88NH5`prV}^qELirj{sUY^EeZmiL|B`amcr
z2B4k0h&yzYe=gTy-VuURQOexhTuydN;raRHWta7fkY~;1`p#QIR4Yn3DmKG7my3Ux
z5n3!Qs`D?}qXmd5u;$O0&K^t>3U)P^HUGEY!2kK~`BSTuTi!n4fHVliy4mwX-8c0z
zTd4ZWtz&BebMNcy#3ob+YQ>}Sn0tEx;v2sGJlZYlO!VH=tX=99SZ}(2BUK<ohC*7&
zewu>}uqQWgyp1+uHIM!qZHzG52<Y+;CIbO#01X+<><Zm~g9QhJh45~H@4rvnKEa4@
zdGXfzzu`uRG9%Fo)TV&hVsw0Dc=%rP8OJbshYw6X&A)+2qHki7`WtAl<xNfUy0%b_
z>gejy^3~tzU~mTw-gt)~&rBCW;uP3h1)S<GxMPJ&soqrJN!&8%iCN3d2GgwfU@3I#
z_wV0*U0t^V>7C;e#h4>|0Z&^5k51><gn5BApdDPC|4DAzdS}p%XL?#w_2D1WxjFEX
zOa<v&uQB+my0q6$UtYdH-3k4{F7`{QT?#0{5F^!qH;$6=lvpA!iH`-;KHj?%XJ3PG
z4mOd}z~a`y8-Eavd>fWX)rYM+sQQ_V(1P6DkCV_R9)|r`!!reBRL`p<-Tn4!WOJMk
z;reVqON(t!;{F)^(6KTox2l8npBlC6sQUT&6&*$D=I&Be5?lwUqBlFHx33X>%6sYK
zLzp_rATB1hV0QMGo<>TFKTqmwpP<{f^Vul<3-d=tjCEC)Yt~errTOL~Vo%AfhGXX^
z9VJ=s-}ca0ct|qezWvj-7Y5~S(5kEuDLZzs#_bhpia>1yQ04|q{_N;#s)02LxA<Jn
z^C7<3a_cWu%|K{tMRq)?sPawkRuSjvez5>}aXCPueXti)jFx=iEvrl7aEZ@W*E8$t
zBzgF(PHp#egK@AqSfdqNdy0XeZ_#tvJ#{uze?4<|YRW=<XXg3DgO~MtRlk>9>R*A8
z#MgF-HRa8ORjRZ?<J!S;4bIM04jZ?6<0yu;TsAlVT&{A1VXt3{MZI$qXaBq@fBm|u
zuL<Y~UxA;i3j0vT8-Y6U$JtARlGBk?0@)fpksCBk$!nH5UYO!X+<JM0gui|3VCw|s
zVJ}dG>-4!~@*QF&QbG35Hq_DWqP%v-6k@jnVM0P>@4cM(hIMoLsGnE8CLW7a@$GeZ
zObPxtH%HfVVMf>0-Q8UxHNeGCY_*WBoS)NkPBYcPq|pHEdR~KUuDhzLY5_Q_uk(wl
zI7hO6^_xXhj?Mx3FXG61l9rJ_dz)Xuy~YOGh0cI?SI56AFD#xcSJBdlTf2T9Mj<qE
zUqQ^Pe0a~=+A(vEp$FMS`>6Z&WULWIp^hSJJ6rR-u&F!&!Lg?6axz*JTP7Uy90lUj
z-D=bqK6rlc&7!01T(tRVtyf4lNB1e}r;-wflTD04S&xvYm6lg_Gqm(ay~Y3}6WPlu
zT%6NQeO7l-VHUOB6FH|~_5f16<))jgw(qAoFnv6iJsyIA*zh<pThi#Oq37RHtH^t+
zW;Od)y=rqL9#(j?XY)H{-M_3zw;Y#1CBc<z-p|&~#z(oUm$yI8E^0}YVafH2wGN@^
zVjW@nHhl|jI=Z`P=N$@jRt=ja(@lq#0>g$arMbCo+V^9DXO|3Kj`N|pXF?6;^8*wM
zS%R<EH%3!oXG3KVd4ni!qN=Qv$Se43PVc10{PoQOO$`yO8a}`8D(mB?k-1(U=SfY0
z5c0xyxF?f$2$0XF^6;6)z~V{G3f=jRwL-Q_{)Jr0-aFIHm(CmH8AlyIQUw!(gq1J0
zG^G)yHtQEsLY>kh9p_lmijF)!^ckn-bZur%%IIUJN#D-+hS7<5g&v2DC64mU9C1#e
zZ4)m<r%GCM+~f=2gU4)LRhgeqk`&sE1x0#BZt-zF)CHNN+B{5FKe}+1<+6ua{{$4~
z7eqD292-3EwoC%2;<@MdtSS?s4s2?SQ&%Rb${38THTG$qbZO82zVqSbF%~!a6ehJ3
zsGZWw^98SNwg%PZAZ9|eH)q?xlIi{9fsY&Vy#g;x7Qx%I<k6!?Wf^<NB9|0j{=SRf
zWt03tEWS1BcMovor8y3AMOJefHTvA>D-FG{utkzsksSWkmdR(%`p5cfg$-*Vh5PuL
zCyx5H$J~aiBn!psqDOsa3^YPlK(apC&h;B}eL?cO2b}|5yGH**l8N^1sGEXiSXTEH
zZdnxU++6|^(Koy#zb|~z_D|4|(dC~dxk7%^1U#_7Ly`B!Rg*0PlVLT`?%W!jZ|SU2
z+A({<)pt>{k<CawhmW!?Kg*NI$aJQ4^)4iTmi0TmFO_wjICHv)j^v&FJ*K;o)pzaN
z@fp6&v#`+}W3u<~mt5}@yKVgAXQyZ}nz}!tZ;{SD`$Ap2GnQTi_8Hl)8R21uklS5S
zz8r9W+C#C37FBTjSjts7X`X8dXO+w{>fjrij=Z%F3ol|v2BdXW%PctWJ<u)4Dg{Tu
z_KVv&b!gwF!?`u}0vDW&JS*opUqhhjuq8W5sFSC*>R5i>*m=ObUFpzWC-WBugdLK<
zY(%{EZ{f0VI@gCQWUf2C_<pFgc+&6nb#O|Ft6H_5J=0y8c9y3~bh2ixUFmd3PHB>=
zvCUS5=^RIx__U_OFvb4OJcruR=!-4%O5VOe9Bk*=6n2UiPm;i*)ez5nwGZC~HkFe1
z%WGnPpFQVv_L&3e-H-r2@9kDwwW_)xPtu-bmyubW4O)(^65U{3>hre4F629eE2gb%
zwxN^IV4GL8w)V(CargQaHEMG?Fa@u1)ELA)Y$;pUf)RI{qv_xj_BdHRDes&<GTFSq
zQhFC#bJ%f?^NN!$5v!4qEBIJd1@rqEf02t<OmGQ}(wAOi$(>i2n24h4*%?*z=>rZm
z-xu<HK2nITBymDCa>FXQAFF>gbepzj+eS=ZL7C1g7UD^Y9;`2l)ctr|&_F%i{bbEP
zwy{T34;lA^OnAkOQGGe|&ud0T?j%t^S(6hUn5REW=+cbjOBE#$DGx*OIo+AH%IV>I
z=wxEai6J5l7JQOz>o;vdcRDM+7WU`z`Z6AO7g5s;Pg;_J)$FPxyJhmbOu`=|;X;#F
z#)obrw<L0s$*0&UE7(ihE<sv&Ch${dOxy1#V;+iB93C)ms8?OWVEM3>H#x0zMO9xm
z?KS%1Ndrwj_9vflcT;~}laJk<PtmvPRpW}On!gydqw2jASuJ*Bq6E5gNtt|hg@3@k
zyb7<ZdwA#k{d%K6Ig9xLS4~I;Ul@EU^E6`3%BAno&bqI6c5fx{H|JBYPg<(7Kf+6U
z_-@~z${SUD^=O_{db6NgPb{@3A?t@@Rgko9;2drBR9zKtIE*Tlcq>^X4=c_4^DJ+k
zc;9TZ&JqES*Du{EW%i5~y`{Z-r_^szbrH#F%{QusC3CqHaBI`6=yviiMl~NtJ*5zx
zjMBWhl4+Q|W$3Xb;<ZrHT-ES8n1}h{4U;7+qR+Hf=H6v^!ATp=&m4;~C^>GZ8I?^q
zI3(GuWbQ2i)wq##UfSI2uePDII`~~^BS_v0AtT4u1GmwLwy>zHM~aomtZLfm)_nBp
z+n6t6naJp~0<x5CodA(R-Br=Mp<ZDT#uFYS$2qzAzCGl8hb+}9bGPaX0&P{LqB`Il
zDb+eN8y2b-OuRdVcscKmx~DT*Rd&h6S(SH@P3*+>k88|0KYyLl)M}|5RHbKrvd1&g
zDuWkG^lnExnNBY&rgsgQe=;eRhO`iPJ%&Y7SEE~uct2gz4s!45U_%uQrOGJ&m&vB1
zPkv_z32po@QMCoYXGlKkpsFWC%R(j}FrLBxFV%fwh$G;tpkzI$j|qI#AbU=K!~U<X
z#e9c?zU7`A@j*=|%@_r_F*`5GL4i+zswBb`JDgTW{i_Wy2#)uEp2E`AxUWIS0olEZ
zs&-y<5TWl|3_2=0io-L1xxDg#*_3ry+F&zKS-C3U1F)lqZ&clBONf5pIP=aSxD|3^
zXA$l_m<5Zrm6?XyGZrJcU;D%mKb%~T@e9y}=^Kt+c1-=1^qB6w-8VVI7|I-Z!uD~i
z0-^9zglGYh-S2Et_cu8u#)Fta<>=`%i3szn1faSS95!pqhD0wBa-M41$!k(O>Mnk!
z!z_q0z5-HE6Yg2{U&s!l1zB0|#UNJ=HQCMZ-04nz4Nc+CM=$?W`}LmB>#3HpDj;S{
z8{eSdU-EVZdFQ}2$yLf<jkI;6zAwwDKS&e!<Us+ESH(Gjqm*V$#K+eCWZhf^j++JS
zM5uuM4D8JANfY}9qN4Kas!K5Ot*xzJ0e86HG?3l`Y3t594YNFiWQ%u&WN^f#ZjJue
zK!?|q?bKx;gI0s(`pqZ30QvCM{(R;Mp4b<O9>`eQ+jlNbgHPcDz6e;la!$Iz?SG6a
zIgx~*#63}XkOzsJ{N*kIFp_WC<~o4#WKC6zL-P=aO@`)yN^~j-s5-5QQr>&m0&T83
z;{tIz{=RdQ#-T)peh-&43liaF^0YAk|Gzw(cQ5p1Od5+XmUWN28EnDUQp>AOXG&}v
z8}#zOjshFtS<4EV5S?d;qbaZo3%O*0rue`)YFL7J)4Xsxu~P%N^cSKWaxf!ydJVg7
zi35AtDApKIT#sL1NK+13`CFvsu%g!7Oi@K_E9pH}8l0hv`*nLdb@|mqgw?RdL$H_}
zcMB$+Ge)=>i@-H|edXvSmLIG!l+tM}FiP`kP?tQ=6biI&fHXTS{R4dBdtob%&zYz?
zcY6Q<e-O?fsSww52_miB>qah+j@%!BGt%9ij8qZ(2s^3S0>?abdOH9~^by|cCbE;1
zr1)$}QBivxkVx1jf52s0UBTsJte-Qu(L(Xbn8vk*7(%(vqz7{M(ND*$m-?o|7LQfW
z>Y|lYcL|&ymxnm@R({d^v!5Nu6dfm11fkU%>tQ5oT@0W#E(RvWS3vrGLz>nu^4I0&
za_l7$p(%3cVQLTNDvc1MFf15$oCYq7w*guzUox`#*qmDF;li97iP_$Ck@y;W|8}Z4
zb2`#|Czj`ZlQnB?4bW>4pQgl&hWa#};(k$n7=mjH5DC`?EF)wJ%`F4=M=ceM+7_#c
zG#It9craYmH-uCucXGC*nc)@GMZF%IScr{NEnOa338&{l59gvOaQmzCl`B_v&53kN
zWPU7FPQC2<NtPg@)GfvCim22`i+iXQWC?ukLLj=DU*}eVk&~D9>1LDKzK+yC^XDa|
z;>?EP*onlE&i{~%PbVbQw&0ZFa3acU=E~WS`j|Iq!PkW@%|Ii5V!$gr|E1@mwU6e7
z+jF*Ib~xn5=o22U{9@C)+C(9Zuh9ApiV%CFfH?*&r5IG+yUcuo=!?c&?3S0yl`TK@
z*m<obx>7!VrD0`f7ctGsD*^kcEm<zyT6<>$!-f$B(U3Cd_Q*MuI<+Vaj^VJ`gq1aj
z6i?0Gz$(!#`ZK|x^uhbF8zRF|KdXuedcLP4;^*qZa0|oDk4LO3h8#jWLST2xW~tZU
zPc(d)4IlEV7$<!j5QlS`y*29S47Rkik4NN4O2H-qn{e;45WQ26x-Bo~qzPnZ&Km=F
zs?+9Raz3c2*#_sNIh{(_#bWr=lGkr!Gelc;!5V9tt+TtMN1Q*h#=fB`MHXUPvbZW1
zA@4qF-J6u6$$sBxB4F;Dp?WW?a$bvo*?*fhh}_=g)r;o62gkf7xlBG9i1v+NJXyVo
zt*PnzT;K0We5AU%y1Vajxh=UDY)ui-LAKo5e7N#hQx&VxcK@RD_}9?#2bB_1k?sXV
z-1ic10}6Z((=&VT-5&cBSV(*lsQRm2Ud|78ze`n>_~CzADSGj-6&-8^F8s#3>Xw?{
z1UR33x5<y2sxzo|F04NO0|-v{bt!kKFM3&nIjb-~|ENJFG_`&!Nu0w;lLPkGYd^W#
z2ai*E;`zlT#X|#r7B`zQu;Q=>Km$WDy+s(N_&IdX!!qddxNz^);i&Ip&!8dD)e!LL
z37NdFkwI?sc5aHt>h<~A+4~Uye{O&w;o*@x{13%y%#fZG!HLH)`?$8WKG8*~Ubw&3
z6&(<mL|mea*!TN!lgHDD4#47;^POM5OdAwQuJ|152eSV?*mu?ZtB5#)^0G+d-cKJi
zO5nhy1b#!IWGz$wmRLd*x8$M!7U-uCm_q(P|EtfRLlNjK)k+J=>jd!c`c)OVB3Z2O
F{{ln~tqA}C

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/funnel-fit.png b/docs/2_39/stan-users-guide/img/funnel-fit.png
new file mode 100644
index 0000000000000000000000000000000000000000..d06ff5daf9c880a9816aa3ab72f457d08d3a1d06
GIT binary patch
literal 77471
zcmeFZXH=8v8a65jh!q*6SFun+5$PR90i{X_Jp@5|Gju}N8IjOLq&KA#dWS#=h#^W9
zLhpjqP=wIy_eN*--rwx==lnWrowa|s1d_aY>iyjJRi5xC+G<xXv0OTL?%dVKj~?os
zJ4aDY`bSL#{Dp9{su%d>w!?!5PaZ#b!1cu4&DO!$=G-|c+CRb5ZsP_w6vEx~au^i3
zjMc&r*fOxbSpzFeNjuf!xZE{$hTP>5{@dSWB*P0?FN`nOB6a*jC8>_dlt$0fJd(S4
zmCDM0bhoAYv3?jo^6?}8rB5_B@2*jTzcqSW@#udt-#P4*o|ayeU9=c~C=Ye0Nnr-R
zPADo(@4X!#m(TDnLkrPgHK1{RplDfJFt)S%H=$Pgy&l3hb_8|jTI~9%_U<hVJYu3R
zoLi^Gm7YD2dH;%T*ML*4#tqDcqkV@ustF?754;2gc@cSHVA^lKeK;O@?VE}hqrdVK
z=u4=ZsN|YPY=!HdLeDdOj{637Wkqq@##WWQed^x(O*jTtCvrw3l<Kv;3VCUKl$P6I
zT~8A}1==-e4p$tm*=GBpQ4TF@c>FY&azEy<KY-kqQ+K=YZJ_7Wq8b!ha$|~WiqgOD
zdH!XsBS$Y0e>SFyzMVUSq|djchP~{5tIA;v*)ZI%9<g4EE%>bOF!OY#RXq@y3{8O<
zF8ndsS!J5sMSd&h879B`d&q?l#p;r#0?O`w8A;JBhAj;vC)d(K<+0Kf6Wz-v_mgEU
zVm|Y^G{%2?$Ju!e<^6%VlR>?`Z~1O;i7niYc5H;QF@tgJP3C0(swWkb--u$9jPL>#
z+c;!E&#E@NMQ!Qf!&}-6f>wit>asacGVZ_ogjSW1yL12ZAAh!WJ{95C>Hp1n(oZ^O
z4);f7B_moeOE*^*SN0t9rNUTIf0XQ))2eFq9;+Q)PPybr>MFHHt4!-P%1iDpGW7iT
zK;9i<la7}GgN*i*tz)G?-q*T4qZ=%T-7%-HQpz<z+qZPSiinx5(Q#d%v-gthSIBpA
zQe3?z*Fn{|-?P^2<m7n_BSuX|xZotpUE6Cv?vIpJ2p+rWk;_`b5eXkUet3Vp=zEJ@
z-kc17=PJQ%#^mdm-dII;?#e;v*IRd66+bI<^*4O`iagCaEp#n_7BCJ>%R2tSESq;b
zkUtP~!W+mFD14%c@H9bjpWKGTH^&N5N;htxGw+P}mqDM>O(G!ngHD;^Jo?%l{@)qr
z-yJs{SIV)L-)Mkx#N~fS|5&}a+kNGP9nIu<smf8S#$Rxgf=P_#4plnUaJvEak0MV7
zlkv>BJUxle?IYW_^1aQ|<`WRUG8qV-N8uY0I-hW}qU3cMP5T!6cgyq6@0`Zb-II}-
zaS_nG^I3E@T?S^TvVF-j^=mgIh-&&;*5e}YA&CxpqtP;C8Q;Z#j{5tMLA5aYZZZ1a
zd0q*hdi2ML?Ay&Rxuchqr;eIcyeF@?teZUyJNqb0Ftj%-pbK2PzJp;Id-X#*U_mr1
z;=&FJJ-e9w=sQ)czD)V_5&N>n?4)O1pKr(>xR>XlET<&PS*aw8ER_3<q0fe>XmSqb
z>&yzGJ9kd$+~bGJ2EJsglNTpI;J|w`9-*Dl&qwPz?mCkFSsBhnPcSqpFf7R$LU5BG
zegG4OQ162tZQ_N7=&>aX1Be%<b*T66d0*)1K74rn8J#jitoZb?HP3?9UB9>{zqq(B
zqL^v4+x_JEw8gaKu49j$Wcny8^}FO0)YrL`Le7!>`-kL9#R(3m+Y$fyw_op4(pNgC
zq4JNn1226xRXPVE%*IOn`yqcnmQ29w|9be(zjHZ}krfw5sLlS@cT-T55Bx_{{(4Gi
zF(pNLr4UE>t^d2Fz{}TL^e+7OQz{v*w{)O<Z%6-EAAk<9|E~i||C@n-uloNjp`SbM
zf1CL4m_`1-DD*4N{{J|PT3)3fP4~S<@?i^q7n!+8(f)WNi{#Sg!xhhN@!=d+*ZP|!
zrZxO7%?Ar5!TxJCMq@A>yPagVp0=enrBK87x7&5WM|C9=irXJ(wA9$W3<zjO3RT+u
zvEk-~9{IfqPb|HW7)Ei(!hfxLt#&-mrFJy$V9#O?Wi;lun65Tj?L|7aEH9ELLcC36
z7Lu$@&7_vzzwEyZ-Tydr<QH6#rK?Gd<~qaE?(dWZr`L|VHioYU5K8K_87+LjJ=}A!
zs{&p3ve`iIt#`D<1Ru{Ws00mPpi7}WIvc?$Bq$J*dvi7d7G;;$8#g%Yd-u>Afp}{U
zV$wG7s)YQ0D~&eDu$akS{_-G{YZzq_e1P@X{}HAr7r3kebrh<1pYQNIJy^^?npjkb
zD6_vvPEXF0?x`npx<7}H3EIv^=$zdnEb9cr55I;fY=<#h&?dcUkzP!Tn|+1$Kya>d
ze|1bz3^ggG%eH{ByMMm9te<&kP*px`Y#p_|<`yiw*07jCG}@<^9s()}Z1ETBhpXEQ
zscsovJ*PgxaZ}R|MeXdY8DF<N0jDP$PMjHG4#Du9B53*)F=OuvTeYKFc#F}of2Y~h
z+uiBlvu)wCE#c-X<Dm=tUanq&(Ggb4L(_ZBdo?eNHkGTJSnm4Lmx#=zZN6Y^vMktP
zxcxEJl={=&kOuUyf8ikb%@tsRTwe;<GLu^~F6{)0?Xr)~bWwM?r!U3tV%P6zOMUxe
zyP-EwXfo(%ZMjEwLo<xs+nA_EptkKkdUmDQuZ(O<C!|G_I?Z=BynCH(13uiSddb4e
zcB@OYSMzAKN=Ih)^{xIW$*GZpiaiQyO%pB@Wt|WEgzsDwUh%k1@ya`7{k_wc3kqMa
zn4(YC+;n^<y@r(tCI!vpx-_%62xT6!V&z-`wUBe6o0lI-u4l(6plZGmdt`;t2i<P2
z=*&e%a$}<T;g&Y36%Rp?i*k$7-4!M6$us+4UUI7ooB_{KKp)#19LBkoEW^pwIQ+i9
zuvReD!;W}IXek1#p+EEjM^eU4v7I>XlbN$$<}j4YbQZl^Q^wApO>X=EW|9P?Qo!gO
zPbuz>Sf|SKwKr!okH43&7d5nes}a%?a53NzpQ#<j8RXlmSynsd*qvao*R&{~I_Hi(
z*JS2W14f-567ac+7GuYZFO}PcCDAVopeHLv_zz(mDcaKW?{18X(sJE{gfh*o59sT-
z+_3@%naO@;m)hVH6SJv#%~BIDg~HKjZyuqvv9!}_XV6HeH*noUKeDR!WOrP$vC!OW
z5Zry?#U~#`I4SMR<GHSAir$JZbM`)qHu&Tf;CUx9OtPD}gE}{~y^<rpcineIKQkp-
zcD<e>Q^>{Jyf~S<3!y~u!HR1Vnc7z~WD1f8Hj$>V-xrO3VE1gORNc@?LCo>D_aZMn
z3!J6Pv_Jj*>~I(p)aPOX%*;I9Yo!PNb)h5r)fpxp-QtXs4rw4{4`W2BeW>Z>Pcs;>
ze%f9(W$DkxZQXcH2(4@0DOpa$9NUP7icoyUoDwiSJNg$X%eRV)JkpUn!%)S8dAHJ{
zkgl^br2yqgPrMd{qfc1DR7=IFNIBnFg32Wr3$lpK*sadb_5B0BxLsR^7DO+p2(uKV
z>Mf$C=(z52gfi`|q&Ss2xCmr$u!K6(h3ZlUZAL&nOAdL0w|HI{LHKQfFvKMs>btd2
zWU+&!EGf?h4?&Ny&>Y3XY{d>x1)X|SGdJCfT2>&rzjEux`1Gm72~qLPW0=1$ZGRiD
z!^!VE*c@9@HzlL8m*Y}5SsUn0^rGnGwzmbAnhZH3Gyq155!IH9!odffi>RU&@+)qt
zzZD$bd=?|W%j95Y$PJt}p<gaNtZNmlo2ab8o4GdJIhj*DwI-Z2p|7{3)cuwSC8x0c
zeTH><U-fXXENC!($hUYMnEcGrgSM^_=c?`$<zl4JV_Gphr2~^z=Wk4zN}hDD!$9xw
zuUK7^vdqxoEqWBQw)q|`d6aHHM@A{6<vWGd$QwJwqtVOWOf<fT52_2W-&#^|dWm-A
z6qStiOIZ>8`2F%Q`C|9ea?!nUHw-Cs);iTtr>KCg(~*(K(Z8-Jn>st)6CQDh)G?s9
zw<v!jNx?vCPlHH^n406*<Aok-kcyw=%W}()bxhxU+<!XDzTt~|;YZw3J?CVhZmMwc
zdJ<n~=J9tP$Z{043n#5$F6W1IJp4uzY%D%5M^2HT%O&(MoACT-z^`@eG<o<}iqaB5
z{jhE$-FwVI+1A>59C<-h(7520+m%_1w%_A(zfrC?5qmu6<BhyanC2*>w!VJ^8jn6;
zTIspkUjG0H`%7t)@M(Lo{`o}c6r59d$iyzadO0suD2dTcsByK@R<hi_kWMG5@o=R`
z-%~PNp<i~R$)n0kOHMm*gfC1?AYtLr94XcVEA|w0vTCR3v~V)!RHg--P8MTU5rX}O
zW-jp@$1@l5?MrUZP$prQfzZCLzUvl5Otsz1g2~I&LC=Ax!q~hJm9Ld&=e`(i0{iIH
zi%7ffGXYcc2-%O47~^%R{BqZ!485x!e3&0RR$SJ$UWCjjZCponuX($IRyrK=(-VSD
ziJ6iysr!{Np~7?1Q~n$289)Bu$%zLUd6y3DD0F=P9;c(dnC@fYAq(pQ7N<8K)#xd(
zwB~LwGH$|?3i&zv3<8MVlF;(x>63MzWgzlsO&>2vrsDk5%07w++0IV8Hm)Ub04vy5
z+^}CP2G2|21Uj~yuXy56rSi0zKQeEZ)vXCe3}9IdKYvY%i{109vWZ+Z*UGcSeGiyI
z@TXfG74JC=(BZ`3(T{mXJ}-(&QT#6DtyGr{V-E4s!YG<V9T|ScXrg0%JpQWW#G9#u
z9v_LA4JoRuQx$aZb~b0amF178c@F8IyC>+e&S%mq4r%!^1m(f2+V89fXV&}fgI1>A
z>r+7H3XL}{W2Jhs=-r%UYiSoN_7$Ru0cn}4Ckh-s=wg|Tp4mN|m5o@p$8-O&W*{(%
zVI;GIK4A(t=V{zy0>0nUH;;%Upz;DSZOs{paFK9pZ;>Omm4`w&oR#5P=Gi>-)CNvW
zot{XcpI+lU_^gIPG1f8^5F)aEXj0^HKQaF<hL}SVradx`Cc^U<UtAViQdi3<sjp|{
z6ST_=^q${E2(LAdgM{Q7hd?MlLE}Z>{ME~@XL7y~T3%`+D<3f=6Xk<Bnv9yAb!(M4
zM1Q$?72R({@tg%4C&_~CUd-~NiQvBemG=c*6z{Zd9&9+`Y5$K`l1(4($0#25#k?Br
z%Ia)h_bPK@3qIbAqSKM9*euiuMCwMO2YFpB86jmJp;>$$bAoPu8rH)QpAt<UyGiF@
zC`BmC*|yHv0Iv{IMsl1O2$mOY9$^$X<2>8sjN*P&YY#`T7PnX6IL3L&oi7gwPv4Ln
zjaro7@Qa#VDB7r<O^PHM!HCACsHJIS0GsbjXvq$OwO3W-wNEy+#P+0*FN5+Qsv>uP
z&!&D~H^jM(l_DJG#NhA8s9ewA+&DYkh&tJu4%RKIJf)Y^>Yi6TS=LERP41W`ot_Yf
zpas!P=_jU$kO#vko72M+`StoaR4tUE4OQ#4^dYP^;i$__0l!p-;hq`V2R6}^Y4eK8
zwME3^;Idj-0U!C3;IoZ0ulT1J_}2rI0Nl0v5Z`Zqb`v3+e-wxA8hQTdBDW5UaSGV*
zpJshVyqv)NuZ2RMQ!9y=xlniK!-4#ED@{;2444_WzC+#P`f=X&Y3ENBwEAUEOJ;89
znm|VDV8|X^<coDJ^7-u8?JyaU<vHuLVI$I(YK+c@c#r0Ktr|28!z_^2dEXr*fcgs6
zS-8?!)$~=!)KV;3dLb#4+$fzfN2T*k=u4P%&Ihx+zAK;T1^ZCi6X|{>aA<Xwlib-s
zFDJj!Zw^e0iQ3?d!|n*yO=J@rkXxP@rgqrDzL^~NTO>~*<+f9dj=9I1p0nSk1MgHf
zk@02J5BdVhiGnA5)|hZ5D;?I(9QDYkOw6cT*i=~p!2o^vEL%4|=6(pcn#1u%32)e#
zPYw_H@M4Dly3&y_Ba0CvCgcE^vd?X#6z<BZQ9SNFk>f|m1+iK3;C}syYLy=qqUkS(
zJ_M2F_PEqyI3+@MfON6+AY1830%?Jnp%h+voYdK1)jwTy=kx8AW%mo^+^?T+_o+#8
z@0)>7Bf2A$&Y&lQQ1Mxo=U)q<=bYZbO}wVku^Tb`7>D$i<M=JbGooS?yZ6Y`rNhlP
zVRAVK=S`!@icL_Z_QDaq@=?~cqE{OR!X4N+=TOc3T1C+%`!I)$yJW?F{A*oX{2_x7
zE-)h)_JJK^dOVtWGWwPA{2JriJ=8Fo%F!h50n3;TVm&=7J6gu^v7azIBD9mWFU9t&
z!vs|3<+@}-<3frP3p!{tdg%9g;rKRZkBLsz&Sv{$gTv@Kx4x+C%@Vgjr@kD+L3}+~
zIP0gQ>9f;Y`OCU7a091l&Z7|=i&Jq$`=fEhkSPpbGhz@QLv2Y931{6b+Ahdb*-KaL
zi}I+LtW$tGLNXU%EP{(!^SuRzYXp@ao5#dnw+78wm4Urou}=hEjD4WwXR+{ehJE1j
z3dccYsi$|I>=aJ;Q@XzE&n5o~mjjG0RXL)9S<TVL46K7F&5u$xyRmo4GRjoAsHd@N
z|D-P{O^e7jkr8c=m0xPeLRvg@dP}Wt7${M~U|i)5>9gR#%h-<N-#kc9TG*@@AQX9b
z`kwSBHdhn&eYfH=CmM`PrY~C#qI`)xY3B@m4h$d$*gtolhOrNMRq|sdA}ao1A*KAc
zP)`NDtJp)|WAoVhpX7<S!Vwio@CErRv%eQIZ6KMyBQAe^&g+6aRpoN~$~9Dv@$3k=
zdMJs?2G8GIkY7@!aeP>p{}dH=4K=Nt2;x{C7Y*D5`HKh{WHasMY)0D2Wy8M5s!24)
zJoI^iMa|Y+h=u3oA~RGW;X?diTdc8S)TGZeS~rrW&b~9W-Zwh+hgDiyhl|RaThR?m
zaO0XX_bKSv4s=gF$SDY{kcEAgQ!}A`Ac^{<y7vupc7$>BK9D>9(Z&j^m@blq<zSe2
zN^8o3gcJ0J^8?dPyQhz}62T7MK-_5DTns*4jIxs_m_#KueZSL<Vj}w(Bx&6DLeR8H
z+Yvs;Iz1A9s@zDU%98ExP<=lt7;TFEbnKqI<t7rXYQ*4Ac_|C4=`gn8=6SO;#a-4&
z^Uz;h^DbCt`f#YuJKhVNjP=VdzPuI;-9Z|G>7QmZ`IJ=v8UJ3%h0XI1J-<b5l<xNl
zl2?F_gcemM*>h@^>IpW3em)Es#f}##yVCWVxJAXYgN*4^-38s=*+#*h3)p%i`Nmr~
zA7>M_u`}&kA>y4oRw+1VZ7-`a98RyaOdYq2!R}W=a5&pkmjo=$_}a30&oH@hXQ9)0
zAOJ<9{E$M(*%##}m*p&Q-5z3$f>ax3<*`e{><}@ymmeZ^fB#P5^t%Tv9os4z**#5(
z8`*_^o7GM3eU!m-Mx#vnc}xNI?k85|-$J#w=T7@g-h3vT`Xp?J-g+YQ5R|^uvxyZB
zg3U0no?nIcsUXPwK`yuKe}lZ{UPI^E%kTEZ1Xa^wS`rVIjvI|ZNk}-0b#b*D&y?EC
zENSn($ayH)Dis8W+{S77E{Ygjc{?R_vW;ietsBdKw5Xr4^^WpL5Px0rbsv4!G_V`p
zPj)dd;`k`KV#0L*Pu*233~i3)GV=^Hb~aW2xh1WG;{9CCj*@V4dD0nXWdULSktYm_
z=c>z$fH$_}<WX-W90{HU0VNTW!ydZX$Uhw<>!U}{%gf>YHNJW2zPa4(Y}k+13zFvt
zVb&}TH25AqoCh`sW0Vik$@X_Psq4UiZ3a!e=a1@Ld8w8*##9?hmMS4~VjpNr*v9gu
zhR^nKimpvSWwX+-PAb@z>O<#B6Typ<Cda3|CHa2Wgz(q%(x+2xPIx)U`(872rw?b+
z8o!2DjnKr;d*{)%qzuAE)~qE~RD}MX<Bqu^qTA-1ic$~9#TwT|885hS2spn#t)gig
z^;P1v|6a@Pe4>vz#g+TK{>Ll*PCr&=V~YZv)Cd2BG@R|rhKoHUx8|DJ$}0#w>x1X}
zP8=(d6;KOpqbg#^aEz%5a~gIPf!;wJCB^U>q`go0Fs&S-#o|iWX?$5gu3byE2WMMz
zOzK;tD?GMm_Oai2Hz8E}?#hdLV?r)=#i6=9e6#3KOT;*ELe5L3784`G^pZt*D4hTl
zKVI|Rv+X6*)y4|k)P~}Qx{DGKJn|A|jWzk`0QQ$H!V*!Yq%8`-KFyHzzTZS$BdL<D
zl$McPu<DN~pKb`Ek*G_II!5e2XDacx7Jh>6aTIX_b25(clIP{>XSEjdRyb$H&_f@$
zubt{G4{<yP>ZnC?yepfN<p~mCd4||{WkHkuTFD+$$R7gEd-@LEFyiN9*sS=`d2Df}
zaw2Qq!ao^!H!gGb%gcGI_~1FY&6MG$JUgdZcQZV-W&BYzrpiD#FR)8hdGQI4yOe?|
zx)~{FeClIc&Fxy2x(R=)*tDZDep`XACFPIGK!Z!bj6@5G^bNY0pIOPh#^%8?&EZ#-
zq&xJfJLwE2z6%VcWXos59_V^m-)KI`3u5$+Q-6@-MuNs<q{CF?_}r5Cx{5N?85p@^
zhlB9-1Qkr+&XEAQH@LN_(slcg6~m_RBb3Hj0lVG_J>42Xsib=inZ(-=^CZ!QPzip;
zi2DUAyUg==w-qCuThp6G>s<O1k$YPa2QXTx&o=~VkQpuezzS{NSUK6EJqkWMnx6JK
z*`=|LOb050_aMc-oqJS&Y)FzO#xHI6@Y=F-@|f`Srb*e5$cZVfP;IxeB9Nt>Zj^2z
zLQE&VM%~}J&#ZP1w4$4)_A1$3(MWSvw!|n|vob^IP6++nCK#geFk<~_@Ma9*t%daw
zR!SP37erR9yJCYOt_uVP|1m7q+iK$nasP5B-s|?tB$r@F=i0;0)u3q{{)%O{9{TgQ
zr0!hp@6Q7w;elWdMsV<4*q0vbJ1<O$_b*y@Ydp3~>a}3>O26=3Z#q$DdqwTudR(VF
zMS0f6)pEQ?1~ILa&fIsGbvk$Y&&izr>zE{i_@jGsAFIvMeWv{4F{^?@EG`o}S@K$K
z&C#J5MHvQ$^8-#6$EF<C57PmJEXuw@bj4r?%Pc8cH{ss-D7Bd7Jnz=mqE7SG9H=~`
zcu?P;`&Kuasx>-)84`T5vlg)zctUgbg9frA5fr#CxznjuoyF#T2%O0S@vq9wyPmy6
zDXh(Hr+M@&cQ}a^ae5<Uo($#+kehYNgKH*$sv=t(2ue8G7C3->$az-&8d1S8IO^Rw
zL%G(v71{ZkvcR>kJ2wqqzYiGP%^G(r`;~=_ZZtKHG*rq*;ZIMKfTN_^ceK-D2%HBp
zDrH*^W$HUbA%i|OIZ&0`7Fv9N-n9U%W@!c0hGzbl5{qEPt=Je(b{?4E_-It)@L`n#
zkk}1rJb#6sCj8zw%s5;?BwTMnKou{r`!=D~1Ae^ZT$O44(;Bldffic5oBWDL?%PGk
za&$$&tBZH5Wj_%f(8yOirHZt8U>u~IJrCF1Dtc#U1zmmpz1o--n8LRrVezx=mLtpI
zIssNjbw#<JC-Mj1;~G2eZv6&%>aT@faqzT`^ys{YFPL<Z-J_k9m+j;-#c!v==Wo?|
zj5*}*nAI%WflJD*ejEnjdA$P1aLdVd8Kd*;uCa+v^O<DtQHhh{@D!T?%A+{pX`STK
z*5=88iJeu*oNm6wbdknYOBJJ`eQ@vy-y#s*oAK@i`Po$Ev$h5NEu5_<x-qd!Q)Yvv
z;w#2w30Y`z<LseM3m}3Y{Tx|2&QCPr>5CQh3oJH>gm5&+Be9ZmLjr>l=Fy|SYEa%`
zqX>4{zQY<~QpDagP$V;$)i7iNnOGK+ohThzsJiId=iIKM2LP~ez;3hR^@1?GiW&a7
zLMPuN+jGKsN(F5Y$~{FZeZl|)mXQjH7EaHQWu1L#|CSSDh0_m!LzckvHI5e5X)@pb
zpi7DI<>bEDt#YenyW1_OiZ--A*Ru{GJb|o+z?XC7;|7W{+Ubs8rDvr=2T#MU{;|Cx
z<OMJX+nRc10iRy7sHQjn_*2+mVe#W-;njvOPYLjAIEMiY(mmq+<NBVD27`g#Z(KLf
zTSEkYuR^k7->fCQJhoo?xI^vE4;j%cEr<jR7&oG~Vn2}0bm_fgqe#2m-~cE@+!4JB
zlm@vQF>F1px`vzb5OBh#*tBjwax>716MyON?fpUfyL-SVk$bB<gIqYR$WBwQ6UCMD
zmQ<s5+MUIQY1cI=sGkWv+5X1C_os?RddCi;wF<PjmnvN-cB8e%K&*2^U0ozWf4J$X
z?J9uO$44-dC3PG~_X<$yX=Cv_db9<PuveY<P4DHlU6U19D{s5@fTxlARW2z3<O@KM
zC54${d=XIoc1Dd}-*!>GzK?bVUgL~HD7PAn3tC7>$+YZ3pl-DSi)Y`EyGmWv5MfJY
zvonaFgBjY)zU1~!W6;3d-k^J3cH{G{nAIac3{6CSTdtLPQmG0epVzya)l2o~aLyJQ
zK}>jsX7pffv4{)IpJF^AnTulC_9o-;X-GLQL9uVl8uDigzNJY6K4SvIc&o_T(iDcp
z_Sp-$B<PSgNbHeH(Ml<}89=(dx(Sljg*%lV<5k-yv2D^)$0WGf`+lBPQ-J}+>&68d
zne2U9tFJ0Nl@q|iWczzzx`V$;wbzbVMU-|MLG(QZ4PG)k4RFy^J`oMglGTy<{^mh@
zLeVVkjP>|XwS>WOhg5%Q#E^JAAqqen4OkGN<{*62z~#Z&$NGi#Yi|(HR}9XWrKwmT
zPs<lblGJdMn0Vf{{hdd#;Xt^b{Q$&amN~1+b;@ww;o7@&Aet81_r)Vb&iCI<b-w>q
zMgzi2ZcF=2YW|%*za-dGyfAIrZGm^Enu)d8DGr7=5yv>S)-t2XVJ;C@jw^lP3oV`A
zLQIlLuqhB~PGFx`XsKO%-aSqAeVK2+u2C)bM)0m(a00i~!IGMy%*jFKnIG{L(&!1s
z={#MMZscadEOhY(&eK)DkNY|DamTzlEp>K4S&yT9)CR7H&ou>M*q>e%7KFiT%jRZz
zjlI}G7#3fi65h3)*l|y2R&<xXz^LQC#-I0GpuJ@!MR6%o@gPo-*|RTPKvH9eJIq=l
zOeVH-vm%XNtdt%xRCw9@W=Ks)8!LGn-RlG{tdwHh__r&o4g;yB7Lqeqy3$uYJ30MN
zJXHwlW+JCb{2YVnv+V&B`h9;spGZwx+2wv@bmaxWKU_RG=nc+5grfEP(iG_@dG)L7
zSH$VN(^4f5X^&a#?k%7Qr5ad#ehOK97DyivjBROmkY2@k5f#;)YGFgYPtAh85|#l3
zsJHao(4Y}?vSH-&hKICvM608I-&z6a9>i9FoD-l7{8P^mJbrmuATa#<EzZ$nMoA3J
zHF;=&@Vke(mv7p^7=S(E8X7XjQ@%G?=}kMtj+&I@gBK%nGg3P~#&|d{YT1gM>K<$=
z1>+#XIImT$*`r&D9?$z!nv04NB0Z>utD@OgLZbRF`6r!RL9PDhs(kO|{B9nxd9l%u
z;05WcnThQEO77Ht`N`JegZC@9#h24`_^)E|o2W_#vbwsac|iyfCz9<LJU8MhH2yj@
zYS!UhcQ6*ugXd8^_+lWpQ^x!dsXj+S=eA3I&>(u9iNeGBjV7x%CpRGXtyvXMH)B%C
z-q~LS@WJD}4fx#gbMqZfbH?KA?}9-Mr7J^HUA5dHgVJEgh_{gF(h+jbmIb$r1Q+Rf
zMn#qupS-AwEwk@q#AxF%!%Yj~4prR}PPIM!RB#unY9tDBZAGNR9W5k=1Ba<S)-UtQ
zq9jP=%cO&-V9A4U7BXy)LmF9ekBeHwr~M3ms*#Q3f~}K!!hC%+`f)(;clT?hmkb4J
z=zTn65JX~m8;gPATam7`TzG7oNa}<1&-43C4|WI*;Ygxp83B>On%qZ|w}+pLeDPQ!
z)PY*_upP-C17wsFBbx3h+oyMv>v{sQTCD(~IdEf=GD7V~o`tmp{rT;@r>JCA0umtS
zWLz58YL{U#{^syWg2gafBFNN4>%K4jLCN9o8=jv}0;Pm%Y`y|0(A~g^?Q!U4b++g3
zsGa0X<F<D$=1YMy=Q$sdC_6*~n;IMJQwzj3Y*eSrrbuYZdX$auoWinQWZsh$(M*>m
zB6d9cCEK2YYkK?3i5vARHVu<(G3fg)>A^?obh3%;xFd_BqQj>-Mv~)C`$eK%u2f-y
zI(zic1u{ViqDyS9=|^23w6~FECmnu#nOgAR*dZjGFQn$3UbPUwp{d@e|MrSi)!d_c
zkin@&WXzCg0YahV*Bo@5qGQ2w*U+GfL}-xTAwY}_iDNuek7W*Xc;pMELyiENXt!fC
zKxWt>m~;8H5Tw_Cia2H`yu72=K55(^jjP|1qS`E7;Q-6EIye`0tx1s>w$vV}xkv*!
z+A5R$SKUh_u+7=8zEh}C%*-z8{^=O+QL(9A+N4iWtzt*htb<y1@sKGG&O-$RqYDsm
z+NoYSf|*b5B71aoZ?sWtB$^7#Mg;u|GDf-9Y^~qY$c=%z!(GbOqks<Vup~EyJ813*
zOdBs7GaRh>-Mq7_t-2cY*6;dTHS$~3k@wpNs2dCC`N4DE*s@{`<^dF^LC?!$9G)pV
zUiCPIo<@>sPsPBQsO1LNpQp>sVx0k)%&|BjQ$uXaNMDN6S3auKC_`$X4D%QL{DV|G
zAKZv1aR=Qqt?$PRkKDmp(_!5ngII$>I;U5Yt_=avIERV-b~=$cs3+tE$h0!FW!I_=
z<P7Dm!nLE<HOhid2&vYN=(CdzXaa0cy={qGUmW}AUcFg@!J3^OF%86^-m51cRjQvZ
z`C@Km2rN!zbzGpklil&V^2YC6K-_p|wCFP=weq=Ja=Q4v+I3PH@H0<{eDcTnGQGhS
zCySm;a^v$xbeDUwTi#2`r%Qk(V=6S)V>~yyaAH$NQ4TN?;NwRp6Sb`g&)nQjz#MmC
zoT*B=L%`abo_PQ&Qx+gwi|Gy8tx6TtmdT{@+Qc&MP2d3LWoXrVP}rAVYiREA?|Y5Y
zC;cZI!DqH#9dYbF#m8PJ0IR;*Q0901ex=>jI1CB>Hv0uJm*TFdeQV}qD-qLlAdHgy
zl#QKxN(EGCiTW56<nZC}N2W*n8>j0V9VXUije~|}U{@kZD8OQKeW@c#%P-GE<&69@
zIzyGKMl^(;hKd<CaPV+u#J0MFWsgRD?vYdU_~&^xwrNoMd0V6SwenYh6l1i!$hi66
zBO_p!ySj>EK3egDRHQ8@wbxRX=Y4R^6uot?OG;@id8KIvq{zCc-_hYRwB|OLs9pUv
zTu6TNMg>Y|bKc|2okUL?QPWrfpqA6T^?87~1ySJ$*P`7<@X#A-;n+Mht28V<<C#5?
zhxB|y>BAvz6fJXX{Y4SoT%~<it36m2|LhiE&lh?B+|w&6SOmMd^QUmY%|HOkc2!8O
z8jdUdDNG<~8j5wkI}68IkYTAU2W{L-nVN2U6z9W4Ck_I|O3xV%dc$kf+00=<30?DS
z=3X3Tjz9Q%LPg$V9DGCnmBswHBel;09=<f!)Wa%wCl1=rwH_x~hr2=vr&MicZ|S&4
z7`;=7)PP1#dVKqn(Kc~t{?U_w1B0twOgejJrG<`~amePKinFDZ9TGeR;_EY)UBK*=
zxM{UflvzDUpQP!C>UBB?+QJO5h5$2vbjO;op_l1AGWKNCDj^1MKY3%WG!P&=9H?A>
z0)5`}kr8cY$GvCHy}QOx9HsZ5u?Uzk*LrgE4~Yfy3b7@&6$h|J3#_{IVlr~gAdpc|
ztD@Oy5s3Aur=|SkPt~mo2{ct@D8bafgS>IiA=s<ri47*rA!&#-k$2oMCKo^5@-NC`
zqOp0moE0I2G<V5x`wUX%w_V4aozi>gBTBu9_f>{-dw%LySk#6qd)C^I>ibsHyr=vY
zdrZ3bI00&q-%OPqATF27PFKqKhfSPq!A>8H?FNLJBG%j(6rdfH0iK+ztNpZoL7yLR
z+^;QO^i5Ossd-A!tcLrwu};02{gMjCyI<%j3Mwo`@s_BYtpl4GsbQb6vP^b??Q49v
zZwI){Z~=08FNt=ty$H3XD(Qi@ByUK2Ij1a;u74%?`5@THtA}I%L@&67*;h^6{NBG(
zCv&hssTVTnn7+kS2sVD_%?G8yvYFVf9#`<oyC+3}0^@yzY3YNRP?`WsB)sw~i4Tvy
z6Xwr){vxid5r{wf@TCWwDOpY$SB55vKoif+I5+w)aQIpdQ)Q>(H*uWP;=zTYSu}y3
zz?fJ~R3MiDFgO@#n~#%Uhs6^wBr{JoAoCu}T`YCPNOwh~p6TW@H8qL8sRGm>`C;;_
zPqSSsozM=P`I~be!+A22EVkYz=HL`|E4stv9u~E_1<bKR9N3?!deTR>y2<7vUA5CY
z1?B?>v9=I$;~sjxZn%8l*OwdyCg|zi+B$F#r{}_Bp7eJ}R5T~R483%<T2D*`3U`4J
zsWU^!gR^nWiCM$vufcP&U1BLp*Kp3)ytotRtgH3R14^WVy)728rr_Zl5pFKK2|&8E
z+o6d_ROUL2G}219Rx}U<=vr5W9)S7o8J(ixy{>~SI-I|a$`3ptKsld-CL@08A`sAM
zYv+8XuhJR!c1f_;!5^6Q+<e)^v*XOOIHW0MGckR`Spn>lR_G5TOXku;pOi1K8#JUM
z4K4egp^>BTc`6y(cHu{8^5$&sZ)+o1&E}}f`xay&*!$bfd+0K!ST*?-U5B-r;q?^7
zl}KRdU~Ss~lOu$l7j$#*ri>h7gNZQeu(R19T)%^r$LE9I`43_j^h||5(XAu%94KC<
zgY<~)K?wv0x!2h%g@%1rumL5fYNZIy7@2(PN?e?a*2{~UM{MSK_j_Vvd_Udls4KzH
zelK>s*>CzD7r}Y5q_zw&=!)<F$6YIOX|^iqfpGKa7;J~rK(9Z9#{i-qiHb=ZE^6-o
zB8}}=wPdE4Z7+3P{WDHS!EXtT)i26}Ro(OaUDV<xxfB>5tHzNGzgllz6991G2H+Db
zw!rV10EhoB!iMOR>QehyC-@NG*iSbpWW}{YD%>HNQTK1P?-ja-FHF+}{dftIG#Q*<
z1~e|r4LxefyDLc9Ub5mPS97m8BNJX1(OX~5#4X0$2Z#<F`I}0SzNU7IHy|!*en^{v
z8f3^*@RK5sObtJ%xWz;r=435aa+Gk<70+~iw1Hb3XP7z<XDRtG$2KXAeR(%)e4iH&
zoLuW2>p%J&Z?)@*T@|`VBGdUZ_doH}VqBV|Wwv6@_G2JK#?(zf2UuHa^TmpAaJ)y}
z@b@6i7o>uRFZLni);SlR0`*^*UIH$Z0_i%EEvr7*ZCQ3T1HMw&ogj}^kPN;Tx}<tN
zhH)b;Xupj!&pJlF*28;x4*<}WGm%j8xsOfAMyu(o+DmPT=vA9Sq`1m4oV~;$T59%n
zY%p02>v8bOaIlt+Uxz;b`4=SVjsJpsul(NXp+Gy}1Mz9biw9RQ_vsZt-}tgfPDbSV
zz0Fhu6$=CUzw;aSsa%7<U51X-1RQglDTJ00+U|5E?ZOU8RVZjj<kXT)woMxVEYB>M
zA58+>{;SD>C!IJg|3%MMN=5_kvK8~vl|YJf$P90@K^COr39QGXbRWG?Jd6i8Mr7sA
z0x3P*2B2`80)D#10*mQyfY@u~$#~^+b-cobM4Rh#(5Xd*J-+GM)6-7Ov4#x5sF`gr
z`qS<#TLX-uRm4F6?Z1yMZmf7e4%~J}4glP`b~9{qG<VGk_v42jTJn>t5Q(a~l8vYG
zdkwwm>TrZ`PighTLxge=D^{2oh&S+rI5aJ5xjEh-3+r%scOY;mus7KQ<VWTWL9`DE
zjTU79@URDV-G~(JS*lUL6(;s@R^i}_j$xj>0FA=_jB4LXz8yyg4Z>0<yB=0Hp)F#e
zoy~MsXkNYL2D@d_;0hBpZaA*DW0>Rs?Lmj~0bNUh?nJR<j~uZJ_^eysBu-=xh`sN1
zuY8^IXv8W_M-}v<^RoWbm`7n-h}|f>6MFYR1%zrRX%#!x-uXpfbT7cQ=8E%$;e8RP
zem#nd{%fek<1By2VQ+5ElNF7!fOpqaLTWa308xYdo=MOX3UXH6{LY;Y3{0Efm8$~_
zKV`2e)gekn!QntwUIprn-Sa^cAo9eqgI1CxO||b*os+DJ#-|<?+^ItJc(=rOvD0*Z
zX3*j1BmNAlu^Kcf&(o#&gK=_!-pIVK#?@D7m)*GLn@k3XJF_jQY~hJcwou~Y9{|CZ
zX9cbR_kd92J|&#ve5ff2M2NxkMx^^oQgIB5sT$GVwAO1do}Y#db&e>P5!ZQ#P4UO4
zubVW9Y-~ru`Brj$s<d(&NFEiR4ONi^lLq5n3^v9wDvQ!#_|PUW3IkgeTr|l4N$zy;
z0boJfrpU?;7=yB?zP2G_r~oJ%cg=DQGqLy_l1Al?**Va4?@^n?Qo7#*9NMG4ML>^d
zRYiK#K~JtT=E8L8m0!&!=2H@|DwDf_X3EmSS?wHwhO9G4xw`Uq4v#Jo$g6p*X5V}j
zEe~PV!DW871fx<^q}r$Tp<>c|99+sdukV@ZBsfjLx(6(8j~SI6(7xv<PHb#2m^t_H
zqL5;U)&f@2nZ^oW2mUPBvdpOZ*M5M8X?nze1I>7~d}<?T&*_Fv;!6Cxg<~n#TLotE
z34wc)b-I!ILmnqtxrW5wYweSD8-f-kjwo-gxPKT0ai%knAT|~iP@O8pSx2G$<6)`2
z2@h7V<dgYBHO@fK2H$CnMj-(&r>z6XEIU&3J_k=!GfukrPnoN=RM_wI5%UU}bc)JG
z>lKaxK^LKP?t?w0^wxsUbl_g;+Mj9^M;V6$`4WiG4c1=3<CNOE+LInB`s4{vdHG&t
zZ9#bDL39!Uz*5O+^G93<oZhQwe;(sk>}4_c=|HfVZuS_Q5hruD4oGYxsP49#2xrT|
z(|znjLg){Y7o??jq1wHEJs$7`4+_4B&8MuJXrPn`WFQ_6#_S$)hwOwoEd?n?{<%>d
zGg=u56mED0g#`CNuuSb48jz<oMpxCBg>OvwB~z3y<mQNW*#WYMeaeE9&dL?9<CGh%
zg;_<ay*9TBz+#zehXEBcEmFeBtj}r<Uwdv|wegBl8ffD@{UL<waUTW5xT%c8*tj1o
zeZTQ`6nG@>?x_sdxHATxMIUM8Z}h+em_qxNh1bW3QmX`#&e2Wd!|r!_h|STVDi_H0
z%`IPs*6N?KAT>aZ6opK<8QWtuGo@PXyQ1mQccnS7?y^NpRfZvDNZurQkr8_|u_yC%
zIa7k*FR}znG--UwCX1E2eR%?oHUU`IU6^TPa{us6QzLq7EpdXe2!pX`5-Zsc<Pdo-
zR3zc4FtVao7=`f8#rVD*_<Y2dQ4Z|O`_Q%EY%j+}v2`Wo$sYP|;vIXA#gg4{)WA7I
zfWa55*WD=!_D@Gx^PBE7Kzvv>Up|^Js0p4rfCI`s*BcMce_A-w28dO!gHztVyg-0e
z)s<Uo(3MkZa7)JR8B#o$^DjmYJ|BoB(OQ*W06;%OR?Hjcz)>YV81BHysR1w_?Bi$c
zG(XZ8`=!<w9DoxpzrRPIOd6m8ZZl5eWpu8zHN_;Did?ijY5o`muQ)rZVeMw%0c6Ab
zZ##0($E1qJ9UpPZzL(gOCA(UoL{Wjrzia3%?$|qCP&TK!OOngRppAN?q55i*&;|?7
zm<U&aSj>{ulOEwS)i)o|SY3bh75DS~AWp0lR}?ix0z=zr6ja`Sv_>|eEYWn>8j6?E
z&OW^{cO}Ys(1Dfg{bOy~X95{AD6SJ)&j_Kf*_zWwQ+q5hXRn1K;t3!(9IH8iUASa*
z!Ar*IVUX&HcB-vBJ5bPVSw!9$eeJrC?&a`qkJd6Kf#RhmFmz}@V=_h#x&m<0uEK4z
z9|XFGX0ivujq`o$wMBpEMZO#Z_fnL<x@@V^ph-yTF45K`?i&LUtI?+E6i80TORcSw
zgubPZr(*Z0V8T5Kcgb^=<XZWt-Uy(~!PnkHgp771Gfsa10@3;d<2__=H>(yGm5DIw
zZ7$Ofe?FaIUjRHx`l-vL=Q-{B?LM}~>d690o3RY=5Nw66yXx@1xA#be)?<~u1hLg@
zpE;40vVeII=)&Gw4QBU%#($(uPh|HAfBQ$rYci0DuIk{40CpWoHrwmO(yLuRBC~9M
z=SNNDWZfBm((qcX4;Pn-5Xzxr@A62&0s15Ci0ks%>GWADVl(H}mLpUuI+P?WT>eC-
z<D7|JEz9mgyl;7N@+nSwZl4LDV~!$}PO_$-ip-qw>NvAWX!X!z7*luPEDgprj^M-N
zMa4|SPz~S88n3?G&Vg5%KZJNxzp~t$?TnWx^lR?@W?E=3KM^3MAs5?7-9-Zy+x!}7
zB{ybpCH&q1oMBz?&ClX5X+d0lqn~*#t}JM$2m(?7z{-sjBLZ_3lULjVocj9+c@bn6
z(9Am3wyzAv=Szu*-J0nW<VQUsQK)N0PN9IqGyBvOeXh#I4BP$WR=aO_=_MQH?W4xC
z(<4d5<#lAYpp4rDCPe_%M#lKMg2pPcY$=bnd-G@>L#$eIHKAx@5oy*r)<X^OwR8E8
z;WH$j_&xLqoz<xAb2_eKx|}PD4&#8^+x$k@x_roW@LK3gAGQjgcXE<o<~kbi=Hc;}
zOovzxiUW6iI|jdd+%Dx*v9Xjgu#VsGiu|%ml0d@dq9kkG9J9(Icu-!slI@G89oqm}
z@L^Odf4JfP!WodcnUqJs>O38$ec5~HV>4j^`C6!q=psr(f4}yZ23dEMs*{#|0^&2V
z2l5<`L^`A$0f_z$(beDL2%TZexNWhVvLIbAuchlMvM^;e%m-^0dXW;fM5(K7YI$N|
z^)9UK-K+Fq9=|T{^~QiqKs>d*DbhnIT5^YE=yzg7t`VEGK-_JVA>rpjImlJ^qJJ7H
z=y{?HJgXcPrsRUe1zdUcMM~BA+jp58=+$E&De@soeLL?*Wfn<VTh#8DQ|j65a5JgU
zZpGwQkFjtu>qB$uXEd<-)cP=w5iwr!w!&j0xlkec(5MBBJw(s^dR|uMab<N^!&V1K
zA10cHxWF!UM$ZT-WIDTzt@!N>!CGJt=_Il(T&aJ-Vu9j@WM#Y{Gon<Wkh-~c*TPd#
zorE=|mbfF%+48sJHLwpZ>4{8f5}^XD821GUErOIF>T*uOOOk_!2xa5kMEU*(tx{-*
z(JsESzi2!WqLW@y|G_oE4(2hfu2KCwe9mzoL1_LJg7wux1+ySawvADM5B+(RB$2)3
z>0NngqhsZOp_OG$z2+Zab550_kWOI$_>wocvM?prGs{MQHfzdfN|ka~_+AN6b!g8W
ztpI7XH^MM;Q@H0b403&O#@-7Y89vx#VGqR-=IN7mMN&1=KY!b2%FD@Sd|d75nmM%@
z?UQpWS}H5ha19D@r_HZz-jK`^?(|GzA@NR!xlRG(KXFD#)mYvB)LyeBy<o+douAe(
z6K_|t(y=AG^7MWzD$S}se7`~n$9b<)Ob7?HwiFCQy8QN+wf5C@U!Z(QvJ$}VI`q%M
zYBt85ep!RLXhKQ#+~pTWu}CI0?<wGaH?ra<_h|q6U><LIpDJrENr|X@sYixTY!I1Q
zVVNQMx9<Jo!-&;gKGZh`onY5Yamg~u@`bztn|Gux05)Wk8Io>_pToPHE8O&{;LML#
zKVRsISBm_0Kzd0zqo|aTLEy=pB)PG;kh#fUeqviSOT3ok8cXEPOU`3U{)`kc+2l8R
zzgleROx}0M&#|Ddjitkqs#TNb^hON1v4_$3`<lN-<vafqId28P0Rb=hjtwbs^B*5K
z-9Z1<qlE`t8P!vMSFe|pNsL&&JUCbIIrp3)CBo!am(BqUgLBQG;d1?4v{^Yp5_8ZE
zd08i9+(NQj0M3l}-sCtAe`gdR&o@071JaI^mvwcvUEeN~6-x_vn%w<m)Z3ze-8#X!
z^(L(Pw|#(sxr_dG)y*Eg_u|*M6x6j~8KG$;VR`NShISUG9GZZc^9}%6?cAM;BUy)x
zSGNA@%QH~96&67P8r+l_Lw%a&*p~rF$fVXG;oe{40W1*UxWEJZuy1SzWU+~;286P;
zQ@sSBNq&NU2Y)rGl70kmOW`l_q<vMNUaOnV0L&#uHB<h|zs77pe`d9+S35t=T^-~>
z!pIMg0M#L2d<(i6BHo)Tyg};yJ#eVKL)DkNsh%ehwzGH1N$w&zAgJ$qAq6NK@FX^B
z$X}Q=On`yp@M6=djj)rA<9tW_a!#DK`sVPvx1eV=zrIiNu_fTC5|`S0w`ylo$CKe}
zMY5Aks;>hsF?99blz-j7o2E9?uOY0}cS-gX>p2s8Ay0mHB>P}2n6JeC1|9XUQTX-H
zpI=Ny-f(H@_avt;j~IWh!oTMIXRH3-8|Lp%S-1av7XaD+%=15I_J8jGe=WEFU0?7s
zD*o^Kg1<r!FdF|~U0)C?Ua6h_>8jv!z=@d;WGZqH7?%F0ozc_{z0;qU>Y7qJ9dzUg
zlv9!=#uZjeE~cTs&ZB=v$a`AAMIa6<F|cVs38*0hh*tVZ@i@b5(!Em<Pb$r;H=2)j
z&D$^hoc4ch10dW_P#B6K`05@$?tM;9a{u&)DIS0G1hU|3)w2L;@+Jc?)#VQw;$M1l
z-MjpEdhzS1AS2hO2W*A>XZ=1wfJ-iKE?QP-5<po>qDYQ!`57`Adv8{Az`}{QNrWVj
z@OrUE`h?4CfCXq7P&$F8f7$x}vq_%~fg4c9vrOck_W7>6xcQqMj`ju(nfd~bv!4Q}
zen3$&p28>$xObwrzvu+rCz()64l2^t{}8<P*9rEo(<4-qa{FdywI$VKk~6P{%eRIk
zpJ}5}Yy$bxkDE5Q+f<FukSf!$?ZW(w6j$`l(7%qse?5eZT>28V<-9Tf`MhPohU_fJ
ztiu@4AD2<Rq)$b{*Q?u`Nl8UNz-HjbUNQb>XMP4<HbCd@E)Y+K_XOs*E+Mk?<+yW|
z79p^VG1aZvk7R~7<NPVhM(_TWX8e0v-(-iJGmOr?uax2n=m)j|%)GSb`;eL2Bpf{w
zL2VU}S7C%yTb|o@Z^r!B_nHIdAa}m-n|K(&X$cxzM+nv$|7f00)smd7-q~~uJ{ljG
zdRw#;`rj?q1U`A6@AcET3LwQ1Fuv2?!s5BYRxjWF@%}&OAINu&nJB-A$M+B3`8qng
z(g7-4HO(82bSiDIZB^^L<n)d&<ApEa3yi6BbV{!owvc5f`MUNj{=Mz~>a7egOfo9A
z-%qGmo!<;Wd?pR3GD)UHns-}(QxSAB%ZM>9ceaZtrsUl9BTk>4Ocz#xtTVA7#WDNr
zD2j4^o37ixa=5?OCWSf~gjyq<`FQ+vpp|3|T?Sk*P%*~o{dUQIQf^M!8LwPUDT=;%
zPc>9EJGLX(dZJhhuuk6z-SUmb05|&Bnzd}ojnDJ!P{>fS;nTUNh94`Pb9$vg1S~lN
zF9tH*JNN?49zuc+=9jB7PgY-e&2|G`kxB91E97%GUW<%%1kX*Bx&TjP2cF1bdyHMq
z8L+r`-B`0_vnlCK@5Qc!5NmjChlXPq-Pb>#OO+KT09<G~zWHc9wMXG-)fPpXb<g8X
z+)onF0U#YTf$J2WdYqE{pWBlgL6Ye~hs!^W^hJgx>ci0BK#(w@xJsL_b5Gc?)I521
zS}$*Rhq6C`J9!VaCfQ)>{2gMLombjq%s(s~d;*$2G7}EnO7fN)E;1kYm`idUF4!Xg
zoi~@sBbW`P<Ejbk@I<d<S)3=IE(TLnB9ZA5Yj`8x^yrMaw0D!D_2-5M@IFYxYgna~
zV{nJ%mD8a8Z(;c)lBMRj3h=>K1J^>-Oz99*BD?9WJ(E{>u6x0Njo=nN>V33UH!S?p
zec(!DqB3#jwWVM6q1d24SG#Z|U$g%N#b8w=N!EOW#*{Cg<Ya~bmq*N-1v}lou2X9P
z$eB}Hi}w5Q20EdDA-{B#rMU~pOCfbL93Rp)HdZee1HN-U7TyV9fARH9<{e%vZdgch
z0N}Y&Xgk&CzJOP2R3XXnD%|&eAdH&0fw9~psJl~?j=eMjE{Jd;-6Mf|@i4o#Wo7IG
z^MCCvo9n=QF+br!f7u@tFa@XkEhH}&SM{hHnZr~>XD#w>1@QrwDV*Q|LUJIo9SX=?
z=heD_TIaG<)6IV@())D44NJvJ32)P=^{|+w9@5}QrbfCpZMO4~8?OaE8^>V~CoVvU
z(HJf9y7<q1)5*^jQd8W7AN8du0xk<E%8<nXCNtAa(sdh!Y@i}>{N;Zp+DfYA?JXSz
z_wL5OyDU`yzNQH9KNs$d{rCH#$xZ)?d&JDM0Mw6<ms8KzxB-eXzKpx|BRA%(RrWr*
z1HOkl$9MkGDR<yl=%{%{^~fmF%d1a?Q9LnH?b{Sk#_(|zj=CY<8%e7Sx00%X0D|y;
z1lphD%%uTr+bV&C*57IX#Zomx6R^nohXGyb&=Hc6wn+wn<jpwcXK7;|IhE8Hk}eDR
z*M|S~L+HOFE<nq8v|6qKlnY#Dqlt(bY6k8+=m!X~h8vlQctGg+_L|g8$X3aJM^9%;
z&zoJf_6#~B2zpRIU`k7fRy^Ie-D_Ra1C;kXc}&A3!HuR<8QL4QKO2E372^ww^9+(e
zH1ZtKRm%mO(+O4oH89lAKwKg5E`3)L$^~mjxaH@VdmoX652ngUYEvu*=>VL&+9a-F
z4DjM36(nN*dC<Q}2TlZUmHPoNthjaos_&<YMf;~Km?~=!9O+jBqzVpxV7_N}|1q#{
zihxg+-VuF<BI#glHHvu@Hp7=!ippG18+V4x@Ian1=IlZ*SZW9j+__oj2>a){Y1sxs
zbRLw-TC0_G2s=m73swT^=l^>}hRRU;UV6%|uFLU{7s$xPfxE<zux}kK|7e)f`)lOn
z#$LQCe*ZrB{#s81#!Bak+W!PO{XcvSpwRhnWB$J~p(oZrCd9&6TKr!})e09^2sljd
z-GzU4OeqTJn4@`prRsm=ML!!^@Ddoa|8?g7u=mw(QFU$GFoGf=f>KJ2ASEr*;RpgM
z5(b@;5<_>73M#20og&iRIiM2KAl*ne!qDZn_JH^Ey!ZPbe8>BXhYEY|wXSvLdBys_
zlleb3`<Dd%|0XkiX|`iVK_}XFo^${xe9BfZn@XOtYUc#|gw!13)FRq{Oo8Gw5R4`5
zt;K`+0<<&Qs{b<0s5?NZoDD0W{kOLBseOv)a4nXwGVU+P3pC}sLWF6=(!ETpevydC
zA~LwLGtgoc5qXDLErt@r71Bde>;CgVYoHHKUn{wa#N`T5RFEwDxvO;qO366vj5_!8
zf3}a$>{UJRw4M>@*XGa8_v==q%lzIP#zAGo(NLlVW+G9w#!SZm1C(UptfuGb*e}i>
z&XNZb13btxeeJ4=;?@!8wLQ0bQ{5g$Ec|vpgn=AIs#$z#%dP3-Qf^sW1y#_VqHD?$
zhu%1>j-6X+U*han59OM1MJL#?iVRkmSbO44o-HFb+0tm0Gb9BWsNNUZ+Q4N49Si+2
zu9Ab?5oX-OTs+~;n{^i+Y17W#nhB4OwUkvL>V7n{p_hw^4BlxL`(vl|fY4stnuqyz
zOR<$s@Jj-#9%h~vmA*ps=>7Q8-Y-pk;lf^XjvzPM|Dl6-9E6B5l<H4zL_pILS}Nvt
z9|!PsfgV>T$S-WE;GszfMwk+15y<j5S6>xOE&pOVusK)cF}QJ~eb~$8N4il%M0k7n
z728!O;-C(pRIV_ga$C+uKIHz(S}@UU@3qjM19R1omteGr%^h7Z8Ah>j6ZU?;>g|NA
zc0|3aSWY!7U)!QB-}#uC0l7KH9lK?wek7s(jOI=I4>{W=OSWCt%lP76=&ka^ZCt0(
zODct8`kQ9CEjRK2KF>LT#_59v4^^98)L+=wshmWbq!nsU9oc!S^*p)xM_XoTzF_Os
z&GwqAhOlpCEKkO!ezsbOK^_plk3bWqq~g|~Xy+cC#9+0Ovo%b;d#ltLGIQ!@anFSD
zoXJr5%Q}R({Tdlxi$H`2Rc+#>uuaID`tLfOrX@~oxirz?ahJYyt_l9XO(hVJP%;qm
z2NqWUPrDCsc_16yM}RT5_kUS_JQgPKlPjj~|8{-}NU0yy!dIETX`G7Qi3fmz19U))
z^GC`#7&iBSI)9!(Ux8rA>VLdB5e<WRZgFXev`7QYAjIw+Ay`9eu@4ctoZBmk^&gis
zFU^N9kCM6I->wvxTyzV+s5UwU?RgpcF6t`b^G9l6@DDQ#Z&tH5Qq5lH@TL-+21uf%
zzm(L;w^53?U-J~n*8WO6h8^rIVTQXO5c|i2TAoS`ir00gq+uc`i@$=~b_36kl=o~O
zsXFDq-DCpN$SNO7bx<BGhb6SK1#S(Zzrw{hM;am2Px5Y(E7a<;JppN2sSjq%s96oi
z;%57{)QOW42!~S(!lG6Fl0<;K(>Kr0_lEqx6g3V?6cDdB{nm`{zX_l$4LofN^|XS!
zfBhcbE4ZWDtW%ae&~7wpoxd7AkUROPJ{#^!9FltDXD1v1XTlY-yF%fY47Qpm3#%aQ
zOXD^E-ye{}<;q)B&)E3q1BgUv1koA^sWn7D{S?54CK+Yd!=aV2a4DD4petrfe{Ov}
z{<|x@6o_-^oZXo_)oKUGJY7BR{aq_KzBe$h)~jUg=1-^qeR~2Buzw!==P&!`pQCz}
z8cqL2m0D$3Ty~X#arHw##v&VE;N|^IBqfrk4cDuVZafJO1yKPf_r$K=$d60V<L`2P
zxL@ss2Yviso+SY}h?LULo_wRWfN1_f-g2GO2?L(J)Opn&&O&6&mWZ$F=gWBiUn>Dy
z_a^OxEnTG4{r7w2$vc}Z-ukHKk^{%th%XePk}L&40PA0<=+tul&B|f94MhfMDT^p1
zgai>Jd}4hNvo^Znr#U2`2-x(AvwZ2t{{0Qmy~qL<W`fTsf?XI~4b{$iUKd2H<t!={
z;7!V6sCIiBBju&OA&Np`GI*yPaLF$Qhd>n8cY$yd?2xRznTU%u1p?6j{yiY&Yrx+|
z8Bm_A=<k0=zCxHTiE=Hme6nIFgcha2DPZ=`GXFl^`X0!I{nLSh2iUiTGb$f5R)K`3
zfRH_ikhqQQ`+zI~O82B2%e#{3I^7b3D_>JiHB4u8H)(qVjAJ1AN261nw?|O+t|OH*
z`PB*kM_vR1y~k!+{y*R%6fu5LPD`Hhb#{61W6l7XBtSC~TrQSy>0ZIV!VXa=?+^ql
zE!LJ7|LYtd-a)j4vnLK4bK%p<7jyPSwR{BTnIa1Ba^wY6d!HfzW{xbcYu^Av%H#nx
z`wXZS13bsx4wN75dJG^Ptn5eD#JUHb-)g8P8vZ+(Igm4N=!b>o`;$fgr3*yfT0A_U
zwQve_{)^TKIG(7M*lgd`C+y&a;c8$H+VZmC{HQB^i0bk~hv5Hx3M`C3sK`vjuJMY&
zruJ3xB-D$(o~3Ce@%eZO&MO3)5*@z)>+@2iHQF=d?2c~$!{qF)-p^Ay5q%$o5_o(3
zEy<HLO;JP|)f!c?mkONz7%aFOLWOf`s<CSIlp*C{E7^2&Y{SrrqL_(X2!DMzkNnFj
zf0+ZYln7ACnu<EZEW9saUKFukr9zDW*I%+h)4XkaAV4+0yhuCfnAKC&X$-~2{LoIA
z@R_5qOn(5gf|=G3&;0XRFgCA(RIzasfNuXyL5QYj$w6H@B27bc*`Fil-duvW7gQMm
zve%;@`@nG(?ajudx?JA|=x{VZ(nRD3qoRtO1k3w>8kqPiO8wXSB5`qdwE30qwM5><
z%`&cw+f3mBqu(4hnpjhJhb@MTLKS{qdrkWR40{aG4Tw}aU0Yh5PfYw*|DKCBFGV!`
zp#`z<Boh6aZi_GuF^zXMff}dx&*mC5g85oHqU&{+yP;Zs<NL3J{#x>XM6H1<5x=iY
zCN;0RutQSJ6AU^-uDI6wQ<#CWLu5&TCaC`@<N&QEx{+uw<T3>~i`uJ8dVk#f^%z0A
zcCF9te}l1hK;ty$KlQuU4SL5Sp)#M+K6MDk2SCuFL$3R1rhqjlDK|S?2xykOx$0aa
zHVR7L+Bz8Lw;H_<8_fWcZo#!wKM#u@39&L&0Ib0&gXquuGOrPVgSvq`7j}aJL-0_q
zR-%8iML(JyuuIA@^|t3?dTI&mao5J;8>V1&(kk<D8N6bH!_0!>J}B|cT{iKes6+Cg
zh-X&dl$}9V-^$HgEjgR`ykId~6*y!>Hhv&V;CsD$3s-5feY;3%*h^4l%d-NilCpPH
zc%W%h$pDA0-0HgMz5Bh-;@>#~2=0H53NuV6j!d!dkPw3M8Es$Gz~IeI(0XRlB>)`e
ztxeFxTOktZ?m#yJa2iUT#=oL*gjb}Qt{2rzh6^j?nHUj2FlIZ~!U`ZdkCGPj4AicZ
z?P*zav~BkNf{<nP+Ro&)i>~)-LOij^gRc+KkHPe_6@G1s8u!^z8f4U4Rg;y5st?pf
zf##3_#@aN$=UAu>>J;SIg7Q0Bxqqu8{^16Kp8E)*tSek?F;*4qqo6G|yJrpLO0xUc
zCo?j09$=?Pf;qUOw3x+^<N1)shnTvr_z%ZOBh!5Jj9HtJSWC&v3j16V9(|d95GY9m
z5;D_8Y@qU8kNe)<{!;F^2ZP=ORIC3Jj3KR<z-+XEX@?(3BsM!8Lfl6_#|mtWgR=C}
zY`!wt(R+cT#j6!df@&eDYz^*Dj_{H!TQ*c0yxGR@t6Kt>-Fx8KdPaZwv#(N2Rc|*|
zVp-nTw0m;T9RNpwUDeo?U&DWLvwy`|lC7moBCNSm=d)z)XYG~^)rKjaer*LNo$?gq
z;TK*ltz~B``s=K=;t_v7uEPxqOIyVwmi%S;+H~LFw-xvN4eA)o?^J=AcJfek(7lkF
z5KqbRlfVgT!IX#E=ror?HYS6tJq!!K^s_C6FIg6K$u0ny<pN2n#6H-}1*xc}`rux>
z@-K&R4U!QbbeCqC6OX4==pReNSy2`74B^qaUvd(jT#+XNylHp2bA-9{(iL6_u*KmQ
z%-&!qMKVY}Wm8lQLBndnP?>lLdMxtft33dbnO2Kj|97Yi18SZ!Z^J;DAlt6d;}C)q
zM(2%1918ayTbC7cVO{kj5JE1ehOV1}N7g8lS3?sHjW9=`%qJA;c2BMbV)=EB&!i8V
z%LA<hMY9PW8%+x{Jl?LFh8KDheJAlz#^jNA12T;tq80v2Pb2irfiI&F<Ygv?2EZ$&
z<pfx&6-nQ*v-gg(&eu$Jo0k^^?{#<|W4b|xqFZwf0Emj9_M1uzD%45zg00lRh!T52
zWy>2#sbSf&wu<1n8~<l!onG_B^$>3lNGI;k&hs2k^8DPJOGs#@)HaQCRCVk5OyBDq
zr^`SO8jiA8HZ@K1_n<tI#d`qSS~b1=Rw_6vqY8r{SNp`y`F%H_8jZsQ-!5?3fF7Rh
z<GiU*mv3f!2)|vnW|8<8=FvzL^!S~#&7zOOw?I|V&|(Ez|BkhZF1#VllEDj}Lz5&>
zn(hVTqqgx-Y7C*ZeRxrHOY}Bzg;!2XuYev9ATx`+<Ghe-GJfo9+AZ=G75I~qe5?7B
znpyS(QLA6tqGV8PQ!1INZ3aIhVoh+fm@g}up0WMZC0Sp)e9^7*<&dcgIyhV%<T>>A
z7u9Np4iACwSjP|aX<B4A&m6RkfWY*?Q+(5O{i*cVpihG)2EJv&x*+h5!R-K5^PVg}
z#Y5gMdl2sd3w)DT+MQQa)t~0mvpxnFIm%Y4L}JUpDYGICqkD`Mc$03QIOleG(Ft_j
zIuDB@$%LfV=_Hz6-9bBux_;-LM8Sl#bqY(ThT@|>vH_ciO4;|q9UE=+s0^cJL+FgF
zvU;8jyJ$NTfs-E(mLr=d7DkrOgcEtGER?pXmv~kzW@P?!8b<a@wqWbeBF~`OW3CD%
z*elP=P>Quk*kv!zeO{jVfIafo0uXPR6m2G%c^^P#w6tdBpu&ETnyBSJVZf;s{Go!N
z`lc;!^4<XnECx|HwMkY~yOmp<jyI*{04Ut50A3AGt7eO@hLX1$D)O}C39{vg6#%c8
z6umLT8GY{S2(;-fdz*zX5rU1}Z;smo6lUO6Gj^CE?~>M<|6u#yiCTv)3V)x=qqn##
zQg3tZ<p^sJQ={D-AA}C{vN9vklY5Y2?YX4YMqBsfZW4$7+^(w1U?nJe)G9RWY;WiG
z&AtdJ+}fK&&o7j;)&^LPwaqai1di0p`-}9AGo4^)CJVLXn|FO`F99o5e7Dvt2MzM>
z+yh5Y0t0_7a|{|)y#Q4kE`4_SFaZR%+iea1ub}*u7`PCKo9sB=U;Akf+1kO$8}W?P
z-9qT~pxJ|SV92yY=?0j`c>|iFo!!9zELxGS`yKsoe3m_qnXFRX$gRx|JBcu-|2SlL
zvLMTJwT&yFIEmq)8_YT0>?=hViyW(5U%Z^Kdsh=cGu>dmdNDi*s4J~~VBMmj?!3N3
z(8tR>3MTVfpVq*ou<|y+mpARocjHAkt&Gk?pNRBCwXaqFc+YUfv_^YI-xbg@yx)1S
zl%5s+@}i`09Mp*ADR4E&;pctUFHl<)d&3WZ{XQ@=H{J`(0Zg8kR9A!s-nxM)(wgY4
z2tb7=WHz4!DV)Uq6EO4I1Gk%B&1m$z7t}@xNs!4>CK#t%SdPHUp496lKT~}h^nzGO
z*5@tSWBy|fNh5Q77+h19tU)2%G_b^BdQpKNlg@#`fD0fEq$Ijh)P16a5UK|aV1%c8
zvifS`OOl&5?qY%c#iTax)dCDYV)yyigM13g0aOZFXIlaE#wh_Nn<(;)^+Nd{2f&8O
z&>YXfpXN{=XBb(kqY0$hl;8d_(AubIB7$$nXdjWDs-zZrT$ns1uhY*D=L-jX)+#yu
z?=YG?1j*$FLd$p4vd-T?0^ISP443pjuIyil(FIt?0pkb6e-5`7gtE4?=_-xlbJZ1s
zkzi6X%8e*Xb#1wc?%#~%Q=3Kf93`;C%yxuT&52TomRE%2g$3lwYK5GXZ=nIwxUfM0
z7#OQvzw}??T5Ct-a8}9CyRZ}Ap7Lkeh_UGbz*<s4pyd%>sr?`O5ov~tijCq%8!LF2
z+Sq+?G>$&dReegs`VIv`?6bb`^CuPoVsu%u$myUxwuM3x>?)}vz{|?6BVS{M!07r!
z%JxOyU;V#tt55Gf&}qwmnv9+I{r3#-0S^mrR#Bf8*U#3{aCO2M8mmqR!#HqLh_A4D
zK={JDa`4Sr{Ohj&`Q}-Qd16Gp8%*HkqDBrktUC+SJ>sm>tRzXWd5N@kk=ao4gks8C
zxI%lyT$)!-qcL-kWmV~Z5yl0>+aRB>GTY|RiJ!TyvUSNOw$01cDPD|8u24HCZZB%4
zs?}>HLyn{;-R<LNZBlrr9*CA}LX@C*K_3R2Ds!c26l8g)QVFTBK3Mw~4XThj5V7g!
z0I8X+*foyFZXYbZ6k*!(q3Vg)QO|jVLJe^;_WZ*^@NMC$jSM>>5|c6oj!kHc!umbn
zCJ0=f7oQUyZ@0qbXc^?Hyq{847+;5`+i7_M1*}u%nScyw<iCwon+#Y`Z;6B{SrsKq
z==we_FTQ)7Pmy*ZNCpvimFRQOzX<oP)JJc0gPS2elX5<*)XFO}uJuGJLrLckb%Q<k
ztyaUSY8M6hP7V(26Mh+&>h~1;oAwx{98j`Mfz=VQ;B=?6R!+_kQs4lAfhkGb8t3Ot
z=u_77hTmmt$>;6!Vp?pP_E*QTu<P4CVZRi{iErjNmYp8xH^Jh*W-?pnvCyR1*~Vwy
z|JCjyD7uv5CNhP8WAA_PY~l5ZNr&(U@|yCvTO}WVLmCBaK9T~&ts3m<(J7dX79J%i
z&97k3s`Vf4Aed%J{PNU_nNh+^{c08of&PVABX+Eu@jAn}f6wLonB)FlsM;7|_p%#p
zHC-vO1_kybVb#Q!;hkKaUb=6ThC66R<zuAvHvrB5f%ULL6+l?#n#D5tc*S6K6Thcz
zAp%9-qorT$W~#LMU&`uK$C&7Va|R>M8oI3yGq3O<n*M`kVqbO{sy~4Ja5^!FCXWzm
zZy*?jURGas$;BPrul{`np<yc8_~-53it3EraXZ?_E}L7LyO(=*mT7{9Bcb5kC1_*R
zg$O!pSy9Zifql_~@3yyuEdJs{Da;_S=5jswFAbbn%X_E)SvZ}|4&{khJfQ+{n;Y4^
zQ}TM^3bhvM`PqT9^&k=s>2Tg_kw9sHvbt#B#~);|bUcak!&MLw%f>UsP==XP1MIZv
zl_2&f<%B)!Cdt9B(XvOsgfBQA{6%O2xxr#8;sk&c@fP@)>@#(1K)8rKJA7U=qMo}u
zfS`40tEXz#{r%#^lnC_2mr~&}CnraOYA2eN=cz6#))4CF@^7Y!bIK<zd~k!m5uRuj
zgxkdeQ6DY_SYz4$?h6+7&LzL3{=@qzZD>HA1jHQQ%B+<~NeM9NYxuFICYda3xvLwR
zi_T2lR!;ylOPvcC+eHEygoA3u7v;%eESIm=ZxL_dkE?3mc&y5=6MIQ1Q56<9m#S%O
zMI5!TX?si$;CRx=JGgT%u5cCwhI0CZ+5Wl*lHywwHrMpV3&Qxp!UJzi$XY)YS!&Cy
zeVmhOvge}p06kZn^TSyJ#A1t&f`QWl2w-?Aa)xM#k8E?~q3c&HJQh*D-azBdz4X!-
z_GR>2`52kHcCCn+Et5QHT8z&W+KTzc?J1z?u0|xUv?lw|_(yRHzGdkor=Fc`&(Hb!
z@Uebmg-DabVE+c8a0h4ZT3KmMRRfp#J-_vUAx2zS9N=nD1$CJXPrIk5^eWr%;0>)A
z0j(M4V!YY=wR`E_W|)PwsV6WD<AgRZ53@v+2cf<~!=Nr8F)s$D8OEtc#{GK5&eDDl
z&oT-JMfzi1wxj&+XjUd7xKT?ddnTZ^3~aA~*-Q))N@XHks-j{x*(kU&4eng>5c!ut
zQfl>D+uOWh_;Zl0Y$w!qI!~HL`J2eTIG`L{s=zbRVCYh)q4C|hKI>!8=FjxthaF?_
z;8$vZB{*tPCjktkERPaBcYgA0TKgW2sjow!u%g27Rn^2S??RiZa5FB9Fp(K|w<wQy
zSXb4@A`gIvWq<=tv+DyaUmJVAb*ZiWx)s~E68qco*}51_U3tgC;|Z9GqiQwHh_y)g
zGH0mm$423jd|KCH2=$oe&ZC{19>G+NEL^VDw0bstD#$4;6O*>)o6>9Q8W$8!(=5Dd
zfV}o9BRA!J5IndwygMHjZr5P^3cH?f>U~WuK1ri_2zyc5xcuW5<Pk&4-N!r>tHt!y
zQFmS<;_iIE(@!HPQCM(THRl~W)1y&pF#O`@6nXq=2`xwHx}HDgabMiS)n#8qHJ)hA
z4(-dmh+G`iW8>?-r@9cejAybg<vk+AQGR7QJbw8`HXdZ7UclO$rWz@0uhyh<K`0a=
z2v6Ii)#=%dnnCuOXU*m#sI<XH&9}cav;4XFK?+uQ{r8{2$^-q%jE&z@ODw1nnh)<t
zAHXi<@1J9PlLWh|NO3fnf$Vy`u**HKuJxxgf5oA>$9u?3?1^UMao3_l5xU6#@t5A<
z?hIt{3l+Jz3~(!-tq&f-{d?D^XJ?O&U4nDIiz;om47ul4?+jUZcRUm}U8~lOpDWz1
z-dTR_*)ikZ{7yCVNS$53*t(UN0PceWBX~;LX+2=%lSJe(icUk$=s$h!d9bapy*yqs
z9xH4sMMu|QqLQ1j^7<2RJlocOZ(CJcZe|5v*F_XJ-ey0k9+6HO|2Y`Jb6i4o%~wb~
zQUaF5@#PUv>h-n3B+&|Q^Xj5dmXnrb&jGm4V~r(^tTk!$iK51wc}Mz3`(Oc~ey^5r
zE)y>=$v)-JZvyOv{EguRc%+EL5}yP6+-<nS@>3s}YU;>CKZF~s3Z-PN(8OkiVzDmS
zK<i?*8@cW{NK?)q#9C}P-ZFAFki)?R&+7kxk>~kUxxKAN+s!=S7ZQq4R|PY3dovEL
zIlv7SyfTe3=E%_sN0T=z!=)XTg<aZk>mwuN7E^AvRbxpFg5pU+i7Rqrd+FZHfm7em
zl9Nh_=UmRs6s6$|j4E>Nvh+jbW{2UR6~Adh{??95e@m)&=(g2(Rlkbi51Q0=FXWS%
zxdacoVYxT78Lh~<nR0hrjUASy(PZnL!C(v71N-uowPBs~4T%{evz>$OCb{mhQ135M
zPGd$?Z+#A3tqxfBin<l2IvnK+xD~#sS2`iBc5=MwzOaRlz(thV-kG{Jv!}Lhh6*XU
zB2BLr{9@L>ZJ&{ruS7F?+d62o-ezoI!&7XNi6^BKGd=TBNU3+*b&A-y#><|QDib{?
z6Gt19<F4_CyDyj1Pn6s7V4yhTLunO50Q>8T^6n0$32;wEcF~whJ97=e^WG94kMUWH
z3gnDeL;vm12XB)SG}!gGrrNWvz&7sGKJJ|4vqU9bbl?{4lkL;#=$9q&4B7kr{bI+f
zccvLNPR3%nE3H13F6WDk7sPU`O#iIqjID(Qnoqk2r(0Q;aK76T$WA5BKXz(OG3;Hb
zG3#cb`8KVQ5W;`1*Uw5z=&Zo<{TKy<&7k|}6yvrIFu*Hn&oP+)+_{3TI%_z4A7N+$
z82Wy;yU9yoX>I1^eIEo7{_+L5Z_p))d$=53zH|>XSE<HPXU?^*Pc^>2Bx-xCjqnbB
zbEtCjAg{46Tg!C%4>_&4ODC;}T^+sE(O}S*ttMGXRMy(JZ{MEY8RPZcUOV8oPib=e
z{mpA-AYDphHYjt)!b0MY-vt_$Pfa_>aU!D_qK@yR>V+mjwrg49PRj%LQA?P~<g@r<
zn<8qaoX&xwM}^;r1Qh!QK5eVK?Mg*<>0FgOYT5WXt~BDvu{7LZE5u3{)7tOsft=<v
zY-Z51Wq3_OcQZtCKdmyL|1Ck5ZnfJ+{6n-g4>vc<2MV(?nBBP<^*=6`@HuxLP6)u>
z63z=fPe+Roh4rgfU6L5RPlStiiGpBDLQ;m_;<LBcmq-qgYmk%&s`<eBT>9f%$q<l`
z{2<Sx`W|&|P(%~AJ)ki{m&#vi@D`oWQ`!(y(N#5te!~$rqY&`eE2=~vim{Jm*rzU0
zOc=0q`DRB^&Pgkjy{WkHX`<UWP^X=u%M8Ir04GJ^+<wBO{tEXM0@ng3;O-lOSCsD<
zNUQE&$ZmE)64|tuaJB9d6!FyJKR9T}Z|mIISxwpt2(F*KCyv!b4a&8XoY>VN;CVPy
z3#dC{^ub$8)E@fX3m8NFUsQfEVKg%FQa4E$p&dD}A#gLLEmNctd2n4ZQClfRxh|1Z
zxSo=BDDpLtubOdHf<x2Ol!fjN2R$7-H})!!I#kXz&np^D91q3`uW0?q-;ZDqfp0s~
z54(NLrRE|<;Gn$l+4Q%TI+n5PiHCXW&2LK7uwHJvLGViR3AY_bVmCRe5vV<SXuGd{
zj)#^v=CmU0>pv-wJL0kF_dfc35q5@r;`KKis-Sm(W+((qf&0Pi?5s)B#9R|`<8#62
zSn1Dq7#{&Og@aNBUn_2InIM4k-o@o^_bW5Et#1_}?{|9|<hPZ1V|HBE((zhdjWaFR
z6Up0;+qr@ZU$cSdaPguElxxkC<i;Cd&>;I(&`Mwu9uYwkQjY^(MT?5$`LJ2<*qTpb
zTHozov?%hZ_^7jZ#A9Zu=_uMR<bs3}O-)R~@{(BNz+}FnOowL-MSTMwc112=6ZM6Z
zcV7Vd!j<8*_=>oS0k3VW$s#_n!nSDx%70(7w{%ouiVF=i>`z*aL?2Z|Hww`YUGu5!
zz}p>~e)vWQsOx2pyB=5yJ|g&?!VNU~n`oFqyI#4xo2_lv@>q4{>~i5~blpv{OIngJ
z6+#s&t!hQLVN<IixbpRhc<Z0R_F4A=3<=<HpvQ$TVt0J<2lYD~s^c@7S#?q~BNong
z){*@YU9Winx9wU8&l7W3Z$6g%W8TSe?>9j*YwaIgMaWuYA;O5VsL}Bmby{*a{Aicz
z6d0hb;^O(SfCry*$W~#ON&tVyu3W6#rSj`eBlA~h0cVz{$zqw2Mo~jW8G(ieA0<lp
zOa@~eLn`z4g+VJvv*l<-@v1}iR(F1-Yq22v8CITJ*RaD&q56|x(U<h}e@f$^9PnK(
zhp!Gf#c8O`Jyyd!T@T<(xv}0|QsI;38d(1>%bw3dFD%d#Gg4|jJ{`OZZca;X{#_)K
z_}*u2Z}9nta515yt4I-XKRLJvu&zy9?B}ofAl~Cx-|}}5_Ct8U_AOKyEip-rak*`-
zlc6k0*UsBmk(d>~DVE3GaW88XE6NS~R6jg3bdw6kh`<fNV67wkiT=Dbj6c-eHg)?$
zx`7K*#+3iH6v_zHCo)8$AuhVnfp`q#a7eu1iW^CfdQYTWdv)(TV*i9l$&92+zA9O^
zy*fM87<T{r!3&J?qazFbafq&HCn|h?tg5NIj=<$e>4j6C(AkZ2eo-nv(=~&1D3Z<Q
z>7%8ZY#pKTdl@A=bV5dzS=5EwzOnpU`J4R$@#(oPu3n!yP@Pj@rU)Xu1}<<yBTGu#
z1c8e1FG5$@XD&^>{o$)boB6ZXb<81tpgm$>^6iu*Ar(IQ7w47psCO`hnuyndzWJmG
zcohzQ*6)<SH-M9`U#pF&HE*HLRLine=~gJ+q9(j1DyPwQEj~iSHQC){-!!~5Ttn9t
zd1U&T8CD3}7oVby*?LBV>H*^Lr6g(kB_s}{Uwphmx$bPpiZy>>Oe4V?$vuZA5yc6y
zA>5(;K^THzl=t-nRH`ilQXS-e)I<-|Y#LW0k0cOHix3$(yaK#Ukp_HZ_@x7|JSF`f
z<NUkF5v3W+ayTeHh|HM9P%-cwNVi>Q=ba|n`Na8yI?$0!rZ46Gm{S#tA@R&1ToEg$
zzX6kdi@C`NNvTFHI=z(6)!v$U-`(WGej&qTt4=JB^pWHzqyxRQhIP#ApmE8iBSD7B
zaVtz9v=i`!Hn1}R%x9sKf9b*ut0If3b}7?{UP{gu?e)5@eoI!=MsDYe3>O!4IC+Pb
zuc$i(k{S->XGm7&Z(fn>Et4)T6{#dy9oR0b4CyC$N--gxaYNP}&ks7YfhD<3@EL?P
zY&E7Kh(>6e^4#TAj){1~1(W+G5RIB(9)H9JUrE%%H0ll7KO67|lPCfE^t&(c4Wu<6
z5do`KB|@{cfY>0wBfGb7;NLFev*?wOHQb@|k<_qkYW*tK)2jk>XsFya3Wm1D3&-_*
zB)S>!ZD8XqSA5nJ`+1A_u*(MxOv+mpS>^}p_dc*gzoiRgX-dcaCLsG8yfR(^lRzpC
z1$(QZniU=t%berrNObF9$+@WxLSo14jmhw%{i?V&7h0_$c!$NCtL$@vbz5)~JP-&2
z{#wr#xxxgU7hG8jUWXq`|JjL`X)SU_JykuiF`ZvF?8|d|Nt(*I?~=nuDwD)rCAqHB
zb?ww(5uCUOM<!c~y(Z6|eWjIf>*3_)CgC`rB$Id-=+E<{ud(*_8___(1F-_nZsSr{
z5+)yeYB|}raz6GI9$dR#npSs-u&nTm5U|c{1(^V}H%(N&ZtRy>=jQUDZKC35lNHL$
z--})2i$t8H{kxn@6=W+nuS-hd9!VtR{vnIUi)VjQyYVXN3V0Ok;uC%}`i{lWZ6nD~
zV;+mPQVr_2R3C=R*xC`J2JjyrYd#Pu9q8bglm==9tdepL7kUgq4olaeHli?i{Ns_`
zK;E<O)3Jj!-Zdh&lXqt@ss7SoC1Gb4KE0IWiJtCH&W&+PvmN0J&~=v3d&Ie|{=+ff
za3c`02=f?CMt_@r4RjiL-lW!LMkyTBdxF5dAiY4UJob1eLuYynD5u0|R1W4Bw6l&1
zW(KV!Y-tzQtnH##3JeI#J7%fm#jn{La^e|~=jqv7S$Z6mU#*hW9{}(3tM<AJxiWAc
zpC-5Os|CMjM|-IoEUHtaHKb`uX>v!^WX4wXzUV#DqLZVsAK=^4maz?*i7dV%kDh)D
zJx8hw&cx-3>|IEI-g}793ruml)_1)_ndMW%{bh2?SjEudZ2s~%9^b8^a`o30BEDA}
z8y?Mx_)&XI(Lc^Yi*%9T;uYX8dg-;%2UW8JQGej-*mMD6U>wvB{067@Lya*G$mRG;
z`|1qq<?EYUZdtB7Gi_C71iu|a8$>_g!=x`I`UoWVGOB_FJn#|ny3aLtzbOyR`o5^*
zrPM=VT@mL}%Rz(Lxw+Jr+}zyJqm|A|r7J}4@fP0hRm+3hJQj_%C(4)<0i7I{8aS{s
zmw&TU27cn5nBvX*66YsYv{krnlKobKrOBAC|6pYM$vYgLp{qh1Mci}E(FvXefhs<^
z)%iK%t|2l&Sk+ad#j$K^PE>d7r#{&&<0IPC5N{e6J2d|}d#S?rYHASeQm}@}Bd4n7
zDxE#!dTrCyEZdA}t)4!ictJxp|J?TcYxSC}H6{DI@kAOJp@kR$dY~I|P@Uk|qylcz
zpYWvw0r8fBb(-w~r!6PdOPL0lMi0Bd^q{4<$GUw3ZA#l;B0sl<nV#&g0O&R2^t_9>
zcxSn4V-D}|Fnnd|ya%5Ra676_f0V?ukkvM%JFblJy~;XkOfMbGcW{{oe1%sQC<Q}p
zM>|;@)YtPipLoldEd(aA0$e<Ur}5wQ3gse~4_fTl#d6x3hZ<LZ(TuBOwzsZmidDx;
z2PC*)sX+RhBGG4tLq-fN1WtzQna>{{uzfc?<Q608gaykoW6zeu!NQuwGbVA@Th6PY
zA={O#vjh8KR~nZU`fBcv+s=Prh7oK6?U}hoq3?BiTGU@~r~<&f5UJDxXIseadMsk3
z6bDWD+_7IAB2Dz^^}Aun=!3|Gu~%Q~v)*?1HMYZ(WlXGDUG!2)AGW^Udgmu`v8o_-
zf<U+Z1HFaRor6H*Ipjh4o#xEUzq0^>1{}Ox?dOk>bIrS0MX&%=$^&O)6ETP-$UCWs
z&!^mbAb>9{l<~b7nZh$N9G_c(D5FQt4!zcj*?aZm&)_}J-B<0XbyDh#+IMI*;@-$I
zG8<q8ArS;JBMVpFGI%Xy$|TILshGcFpSH5DHQPjo&aTSvy7=Ql-h@vby?J~IE#bF|
zPWKwQ&%CxfPkH?e)|G(sl7fRdqrvtKD|{!xf&M&<{I?(aMI9v7o2-Chx7|&k4?Vzl
z`K<S#N8?O_7`w-i1EtNxTL-Iy?G++%>1`XgF>W+VMI@8haM_MiG3Ah6$I^|3lGXei
zy+(5R{_++N*S)!wHKv9tku4FySCJUIXSV@kyej}|{-avN$cY2`Mp>VeDj;1OucNIk
zEMDeArf&aFMKYmIi&XpWMkYn$OPab>(j6R5&Xe+b@HGpLqt+%;lo#O1t*No0EKrJ`
z0B$^RZw1$F7;P^P8UlfB1NN~Er2KF1m4e|IeX}O|K*XQ;2$`DYWVU4l5UG2TEQlaE
zrQQ2n=fz)#D|Ndx#(Re|uxQ*^h7>VCO5(u^-K@H{hN|65_rQ&PhE|}OrgZ?lTh9}{
zwbUO2E`@9s2O|pGEWe*P54Z3h^JLXU>t)#*$Ms6)?~k~K2{_GrO3J_!Jfs?D#B^Kp
z_ojI_cRlTrFF+Dl%L_Je!7z^SGt_w2dQS2;uKnTe`ZuS8oi!n$F_+Kf;JV;8khnK3
zqse?1@Dh3?(5FMDNGT@j4%5iHfZoWBH|*gA5t*g$4D=!*1GxDb`-ac=V(Rh}IwUZ>
z65u1GL0}ip!;^I3l};-+p&KAg{MA5ZPCnFe&^p@|(>6JoR1S)>f<{oqMMbxwrp3TL
z9n)4>pq?RREO`?cj7R5f@&z(x+7<e$YlrimNn44L6!8GlUtoSvMLc2;&s7*Z&Y~{Z
zp4loraAX)9zs@aYHE2&||FQm^qqh6gpeIAQz^~8JsqEwA;fVpx_;rh`+b3K|>I+uK
z7XC?>bVF8pwMx`szS9Axyl@dTVsATW>1x}2#h(Q=bG*>x!L~CO#d$@XK&@3I=!h5r
zae`wC7Y$afrY^k$=SenFg6IucE}2RfL&wO`_^s|b8WIn7wDFY;NX-uJsOJ>cy>J}o
z_;!eWCLe@){>IGab@|9)!9x4FcC!wS+^UV_$uJ4F1aC)B<`E4VV7~&S&Y<7`TrpGt
zP5sr6`Hgk--n~IAQdJr6Ex-_c1>r&OdsCk$1kUkWcTrypUZ9SiX8-fKRnlNjoRcQR
ztIyT^+-N+~8yC4WUZ^QwZKBh*MDrzBX}_NGvo&`S#B6>yfNz(^v7<h+k?FI2&6-nh
zyC7n)t#u9*l{JDJWt4b_ORW=V|9~d+4}-yR*!2@6b5H07HrS9YHW;o92WBn|2#%#p
z4QNZ>UMjP!a-#PfS>5r>xvAS-?qQfj2eBItEFJQOe-knY+thM$az2zCUc9g5_Ccf2
zM8Ql@tHd&jGH-izRBW&6<-ng@bF;^{u%UF>u6dY{|5uGEw(zfFZ&H?w19%h~-}}S_
zQW$ty7e!w{dX>0a@-<S1I8I(UF};F-t&2Ymbwygcb;q#qLtr4}idpnQ=5rXj0@rV{
z_v1?{?3&&a)al-O^QuXVSZlg7XY7o4pYm~PRMqP=F~6Ws&9NFt>7Y;nC(eV0cZQsO
zbtuTf0w<49{kvvv>vJKik;rnrHF1yKyup0KMu*Zv(Q6k^RQzMAPJCRQSGJ{&TKTpi
zEl#R2Nk?{BP2ZipqL_F}STNmG0B03;u2eQ50T29jXvVr2ZhMV*fV}_B#$dXb?e`Zw
z=dj=OK|Fv-*3-Lx${O~6S=muy(b5<vKl|JrP3k&8?YMlc!OEVP#yL$yDun6=_Z7KZ
zh1Pt*Hm$iEQ@*WgAq;lQDdpV^fI8Hl0mGAJ?Sq`Vu_U-0GX@%?6TJdz+swJEet%mK
z<T=rYzbQJMkDY8ErSe>7?&F@_5?H8=+iohzKd{s~u-sXqEs!*+Eh+A(P@)+YR1|ZH
zuy+=TXbYKh!rvE97-|db0(K0wi9bt!aJebuPtf!xhEB%ME}|FoVV=i;+gT!YDxI$E
z1#<3y@egW|YSoJ>CSU^3>S}ZD?X7gBD|_%=mXKnL{^<VdRmae4B=q6~Mr`}CoO$5>
z#-_RKrZ^u~#n?B(kEJp>Yyf0Ry=Tf$&<^0Wf*5gUI6T2(UXqfEYP2mecQ~HI!lPR0
zQgid8(iU?=J_oJJamitHMck<MaIfb!ols`nSSoN>9__T5SGsMPGmTenF=b|3g%+>y
zo2uR(>`%5Hi)3HRU1auspkYyd-Da(a;l=xQ6$5TDImd*gZIz*hRc7zqCawnzB6dbL
zOLUdCoa@lw3cEPgwfgC5p^!~HqoBFDxo-JZrq7|{Q7#Ti_eWrl^Y&+jYND=8L(Rta
zP!}|}Zqo!bk)pLdPK<wUeYE`9cR#Zy2{lQ=$ku>6gk|`_cUTaeW3Y{v0)Dw>J8czs
z42ypTiv)6n^5njAA9>DgIAU~I6I=s^=3N~;N^-5{zUpPZjZ<>Y;rAX>ce5O{%0%{Q
zJL!dqAQ=uul5^>#<G0-UZJinewu)W$W<lLbYD2W2v$je{$c>dub+>aPPoy=L$onc;
zB~xT@Q1OtZ!yBJ1x_JB0X7x9iT9vuCU0Q?G1s7^XQ&mmim#+u|`37bc?XxC>WVj?M
zI8Gf+el2=)E6AE)WpSv#jmrWp*~iF1(v%*ezNksUHx_A6H#8)apUeYE6dsT$Y4tDu
zeDPq$#VoiRG;^@Z6*H}v3C4!kLXGXS7J3Im(+ZJ??90$BiH#+1J0VJ}sG&eE0)Io9
zC)nVPmO*EHfxk@t!nO=@Sdz(sQ(-RigM3(taf-s!_5zsKoWN(suZ!EBx}GBu#pF}_
z8~8F8ye}v<Pk@g0n;RXHg2*AG_~Xq~n!Sl_XioHOmmEVJEp%&`5i}bl##DUjN{QJQ
z(*^hag;cf0&osGOrB)RdW@cvl(Nt#Ve*oUBk3OWAw5ILW)QSo6@bW+4NZKeTdVP_h
z<n?D)tEGd7qqa-i7~E|B5^IKo!Fvj~c^aM3Sss}E*_GUzvZ}+P2b;MpO{Nlyo=(~#
zs?1JD<?mJYheR^PTQnIY+=?DVM3tyaA;l7+CLRbH_<y<nbnUy9;s$2Wi`%?A*K7^v
zlKH%5Lu|N*Zm^bcL`Ur;=DmS#{(gQ_qwtU0t{rVx`<(;JQ$4ssw)pdQNp~VUS}6^{
z=t{IHbSJrzoa9`reJgZ{Jmw72EpLhnsQtMZ%f?w91y<h&t{~6AyW#g%&JDJqwlMOd
zL>q5qAfKuFE{zHKuwG3DO~}#Vx1)m{q2uqwu6*-H(YnW?+Cc5qTumLOD(=Pf4z@=t
ztrbp{kOjoIIFpPjale>;nd7`RhWvhEKWi@ZD|27g(P|xfugr6#-*{@c#27Pg+dMRE
zYv>yOqUi7uordY&8HD&xa$(TlZZfvkA{RXHCq2%jVXE!HF~Mt(=QW2)uaj;+=GFc5
z%al7okF|f@(8K0j-w#!t)w+s_7euX48T9?xjV0d=zn*GC3#04F0U%f+pFen!zqTjM
zHTaIZ=ry@T+SY;vZT?t+vApRY=hCJj!97ifIpj^nFvmsn_#1p(f>l1crcqxd#=mv^
zcmr%mtuT;|Aj3G*f{zRtB|gmO*!0_e=RsP#h=p_*KM?)Hp9@{!ORZVux|jSxPS*<n
zLs&zFizfw~0K!kgd3>9rsw3QzdHC10A!C3CQwb22tS`$jEcr?+O^lot*yMYi$1ORx
z+u4uWJ&8)#K;)>Be0iO(++q=$$T7XjzDKtWyru?iFsxXYYz>@UN*wp|Hd>=ICuWfn
z-%nXMEygRRe8Ir4E^V%R*by^dkkD>HAVPetfxDk~yWGzuxZdgkgz+NOK&V#nfkZsS
zgKZm(kP|%mt?jy^Ra*h3SDGe?fXM`=C4-RLZa{THK0`f*$xN0+o;1fb$#$VV9Q_^J
ziCSFe(0&2@AzgV-?qUf&1;6RaU-1A-g#-5<V2PQmyD8nx32h&R*oRAyF~ra$z1kl!
z@_QUNWkMv}x2gB!Uu|D=)(hHFve7&f580G_!XnrA0@Fkb1p_U?RB=iQ`;<W(jU)NP
z+QS9H72;#E@j0K{z67~04~U=OH$FZ9&*8kC%BH2`&?VCi3qOqN0e7!=9k#C>Dq3X$
zI}%wWVmJMxs4YNuGikv>7YozDd7t7E5`R}0(V@MeB~<FEF`R=p5^2<)GYMEAhc@cm
z2}LsijWYaYdXB*?m|M{Ga;>0(Ug+&2xKR~d4hH44f>C#bpRJF^i!H##t8%x^S3#6u
zb{2AfUB@u-LZh?zCytQ@Z2yIhcesHn!*(%8V!#!*prJU4>aKNpYX!{BpL$}hD+5@h
zgA@X%cKhnYl0YeL6g2p6){n-DS6*H|)C*k8pw<R@2dqHNV`>zaR!L-iy-cBDV{lWP
zrd>q1XO>>g7`P@WH$UctG(F@$Dt-qo!0tEpn_4;=rVr=Ne%Kz%?hq(wYB?a>VLCf~
zgcrgKGZahsp2kObvQCi6xw@;72l9EClG5rnkHaYjAu;FGSR>5*=a6V<NOgG>m>b)1
z6B_p?f18KH8HWT*(7WusMMP8TQ^-P}j{;78=nU4#4r))=CD#BudraET+~=(Yh8mbM
z+k1bq1dXTm=}QeO?T0*_h<74KBes@@xifhn`~Xn>G!HRNmQdAH4(@|!1NFiNpt`Xm
zO8Pa4@%B)aYhgK<@!SUWfVjq<x~i`(Q4je#guVFMX@}8x5?51Q=x$}Tn!Kuj*#Q~b
z)npu|(N96c1W!7;QlCA{FM8G%iFFM}fxEgmRuy(dEl1mI`pn>reLm!$meVYOTYGL=
z^kwmZVHDfb<BPqST2V1~uyXm6avsAdj2!$uxR8Wu?d;oMC?ASut$MuvCVl!*tg(mO
z6Nv->=`%uNUIu|!l}=IbbJg>|#e%^qq23#6gsxbE6>)S<8#r51S>}YbS3eIM@se5b
zb9pF6EYZ)6S+L8wF{cV3@1{=yB}3@}B2icg|La4Zf?~B#;hDSdi>-$~HB#j_P<*()
z1gq|-+8WVV_ierXCBG=HO@VkxbT)|HkczI*{uDs29e{Khvc8Gxgu`iZJ+Dco1kB$k
zH)s8o_4D@Ynw?P%5fDcyO@646Ds$gi)#Mwp&5rR5|GJ#7dFI^Wb-psbrX|5ibHmn~
zFrnO>xivS*{%>ts+dEeRXsB|rG>SeCvRt!r)+dfKDI4j8Q~^$^ue6SCnzM)+(bM}R
z3psC(hlt+i(Rau3)a`z1EabaCf($~=O<+CzL{H!?tkCeEIbq5u7*bn#f#_jWS&*yI
z2nbn~;!{FOW}_4nbseHIqeTZy`c2R%9!}2ITM{-CwXIHLvJra@B5Jy5jRV6V4vs#_
zonJkvlJ=0o_)ZS+xK>9d^OLomWls!HWY??PpY4lyYwe+v@O<cd+-;uOEXk5=Nt2z*
zsSPY#4H$D0%BTLhN_OpJO}?;wSaqK5*RPV?%Z|%Bb)63F+y<QLDXJu9p{guzT{<f)
zG)wiU$OQ!}OST7B*BJfKv+rn;XwQ1kK>#tpFW{Q^3K`!$rO|G=4_ckX@zE7?+Vf9B
zz8G2=@@>T;Te}(}MOSaAstlEoxxe9ynPxEXYWPic@0SOrBRu@SOjFY(fYxx=%BX3w
z`sgaf>8Mdzfsi{FfVr>Y(Ge>-cB{<gFYX7l`?9bAT>`2$yhIBRDZY`0WVyuDX|S@q
zg(4xo{;1LbNzuL=uKRUVZsG|dX4v$Jj}O4djG0{^pa#PD`tUdL^UxDHG>AT7x10(?
z=PsTS^{v-s0e95+hFmwJhmKVGrS})|GlK89<7tDmZZGIO;&Tmx+(%_@U=>E*P-00y
zaF?rdyFL`H4lJ1-Z%Bz@2jSC>bWF?QcXX(C@6C^ImvZ|XoP#rtmSa^8_8O*!+Z&z@
z@6d$=F)g{?%Qva-ki@R%7Zh|JU;ThZwZTVjk<Nd9-dP`8F#*5pifKO1+P<J};aJQ!
zU9Vx$NhvnyV(kB^ZiMli^oarK%m6A^)%6N3q<o}(h%$xyJ}Hu~;umb1a(Snu(G@+R
zuozVf(J{`-8Lr4*6dww%j~jO89!<@RG#&%J0Fg`EJH_X_+8eGKa5kLaTyR0azxmk&
z>sU#LD>K&0EH)t`W3@-JG;M4zK!|UF<hdoIz3uI?^&%npeWodDuOPk)*3FQ+;S&N3
z8Uypt$-_PfUR)jZ`!fc-y;U2OW3$7J`-A03OG&=nqWmdiiOQp`LyZ;B?Q%oP;*Myg
z{lz2n6h&K@b5>ivBmE`m7|hJf4=g5$g0frn)slB#L+9Xwn1k&b+^q9WrY|&m>{dm}
zCm57||Bv7X*<}f<FzSei2-dc_+1U(-n37(5)9gpx)GJ0e!9A|qF7u}2<wdBvU&BHU
zt$`=+s_lf_Y)|~RPymX6o1u~dM)`DUO=n^!&ri&1IFLMTtIoE~Y<Se8yWu$Xv7`^N
z#&eZ*Et39wa9n@<(FrertUNde;Xc$!q8z|utf)`qybix)%A)IX8ZB<B4$G$+wnZzI
zz7<Kx+8sRkG!CpD=vaSGVSntda9~^n$?Dqo_)-hL<VF{A-wX>HpDq(q>lKb^9hM!h
z+Qn>_l1axHrhJf#kqyLhq&_h|XFKdv3M0wh1>E16)YBEJuFd*69;XH4Sk!W`WJ#wK
z%&5~4td4n2Uz(Lm+wVED?y*l*{J!Jc&xDsV{AeYbsQyx&M}ar}1li2j=(~@k>I@rq
zu4K0DA!#)_+{UWicT5Bf8BSD37X%EMuoK{&h?6-?lB%&+Hn3)k+J4nd#o^iKro(HE
z1-3s#lpVE)(F=QqjkbpQxsAVRMgMe81wFg)gA6=KT?nYfmn&I*%EW~@RC&ilokhEh
z{KgdCQh`ee^-WwWA`_S89?|SX(qX0t712|fX-HM?YNZ|LsZQkJUvb_C1Y8{Mx{O^7
zE}l8ze86$^cjA(m_L}UZ*MUpzGw2+DDn?o7z$A6c$|OOVb)gl#TTaA{1=^PFe06fC
z#b`y>fMj1#X*hv<^(SDSI>ar$u6p)rIt4kC4gnO`X0rg|^4aRoO+7L6Ft;^#wcbu|
z<0#+O(ha2JIO(uw2Lt1|QvpgO0s^GN29>9sNWldA1m!J<XDrqPr6I#CnKjET7CfFa
z>Z-TJa@@0?)0W23^#_B*jfHr1ZhM)^KKYE6aV+)FkhW^Jrg2}6j>Yu%{TBKXB@MCh
zsZWiVxCkfgd;et&NZ&j}P(3LVzU_)C>T2Tpw)$ScX6|FZ#?oX-KABN*VdIGDxRX`F
zfUfmOnJr5)xn1=AHy5xf=sGZ}%=v-Wu}BW-OCL1BEF-VCN}a6xOzG;MH^IH8$bkL%
zV;SVZ4=*#brq+%QrIh%X7z?gfpW@DC-+4-_o^YF@vKvxCK*8RD5`)tPWVcRse{}v;
zh1Sb}-x~s1=E+3QX81przSHB<($>CJ>AbehE$Zk0J@`reouMJGnZe?5sm905R6-TX
zQ)PyywhMI)nB%KI&B-`Uso6?kq|>wH`SVlObLG^Ihw%4SuefJFics<yU7;%OZ!DN?
zA4)ygD%Sa!D@kcj@43eZMex96Y5KszDWB1tG74!LJTZWbqT(zvno2IPo=k{k$5EFA
zt?YU<DpJt!VanCBaHz3BsZl%JRCfem`7s`E4wyp3KlHykmH17_7VI#?aZZHqCl!I?
zr_nDTgUj1T9Ek`WXEVCx=BVAP9PO4rhd8a=I4j1pG!Q0?!7_vcJ9t(eIV%O(q<fkG
z+YhoU^%LJ-8@E=B6@I@N-FKfm9FrIzZ27g$kx;_Adwp@cw7hp^Eqb6I2$$ik#PIIQ
zI`>?O{l-aR9dQqE+%<tKpb87?hl*1Cs9@_DQltDhET^{{v4xx}qj^cAMPu*CXYz=|
zDn}!KeoDr0W{YR(^W2ZCtkdWMmjbY2Rtwrw<QWPu)TMPqCQsx{zZO=Z-`Q91N`)z+
zaF1!p)RF4wt*cwn5>Om2HLvO1TUxi&cJuB=idk9i3AP0spD?}=IJ)M?&rUT5$fy&z
zKFe+_pjuNn!$qUUU@fW_cWlh}MuzPo&G^Vjtfz=;&P{Z1>4Wj8`{xQCv5t#+ZU{f#
zX=`!6cOad!^KOF$<RKC8WS~3VMF;wx>dY@NyZmgP*#}~d1GNn%0kG}z@1!@Bj18T{
z{DQ?SNSx6*x|O!C+W9zM=ghVEe`-f_TQ1hgoEGCO0%pzNkh^_CHgFcTT-x`NJ`uv)
zMi}dq?J3*3rq$o<N(FwhPZLc+>KP+9V_k>c9*U{yx^JIGaEKDH9!siiY;hPcO<)C#
zW4a^5rj<3TJx;Wx-rano`4wJVd#?h$2_$9sOz4)}6r@`FP3hfM0ru<uCHcTtoTnO=
zNCxp~?L4qXLC54cr|!x&o}bNo3&S3&`IYor3>pR`6uE(h7e;=7;*OwWTFxDpw0^WJ
zknV$m$%_;T9m1}d;oatjA4__DAyePU^n+d(s@?_GWgYC(-x>J^q}H$uLW?Tl!P(kH
z*PmmpQ`B2@kyb?K<tdA5hwarDJUIB-q;jTO)p4%iNGd4xo6%q8dKD`F@{v<JB^GrA
z;>|Z%-S^2{_v!4lFm<<brY++<!Q{~&LMd4;9tZo=@-!#E>IX{ZEZ$_www|!Xbw<+t
zB=JobK3-D_{{3JMnK^ES>;xSMx-GpTKPs}gZNX2~faN^u2tZuhY@xxw)NO&4P(Mp;
zs`%ZHwA30mE%ol(MNi!mE3FZ;3Iq7*mBJ=<tA}ou;Idk?n-O$8J%z#bLHs?mfhWFT
zEj2`4+ukOuqpp7oW>_9+#nh2Qbn#|F`}g=X!SJl9f)4u9m0_<wTcfswVjd1|v(&~=
zC$|K;j@437k<vGyvs`HFz(-|&qS|>)fS?%adfH(@6Y63g^zYS5Qc^oGax|Okb4kMm
zGfhF&d-f8lgChz@4u3K*5%p?YWx=WD-a{w+^DYVOkX$V+7JDwF&o%U?8iM%wiH}~0
zW$JesD{5dma0=gZObnG<%&tbXujY?@_3}!5FkFzoX-gMDYi%znaeMfLuEZ)JJ+7=4
zTy_G%Am9&h6AWG<LseRBLJL-wB`bCMa+k~K8iOqoR*nmfOFYv|qlD^WuQfc64i&Hz
zB{@2MPG1$gC+nQn>1i?7);k=uqzAAJRV|kVEAR9-VoJ8ettylF9BYzFoE|0!mLNmt
zvRp?zMYmb(dsmhko;U}V9GIn6MS2bw)A3B5Zj>36XPwk40m%(Sz}FZ{R(giZ?pb{Z
z2u@u&T-kM;?e#QhuQbm>m(SZ~n-1Em-X0feX_ys0)%-=E<Sx_G(B#xj6+wZ5@hE`k
z#aw1pyc-q6ty<<w==qR!qNKNR*=dlvM5lKyt9&>*)7-McQ&rpQaK8+&MfX(SV98Xu
zeb1HDv=I8?c8%f?rJ5u@5kIUtsa1y<#BrwS#7=;s%SG*vXZnME$OWZx<U*yIe*XNR
zp{_2>-o;`Ve@neiY%%BMYQLku+i`xDT7P>h*01XXE^wCDNEF4Xuw5dbuh35<qExuo
zV!$IdX0)1xcC-tnUfOJlu31E*XSd#VJFj>Q@3r1Pl@}=x6A*GWyPO;VT$zoxBw)JA
z1G7yv*(;tFz~ML>&X=(nTH-bq{cFw%6VzC$vwHW+xsj0}>f447z=h(_UU<y`XMWc8
z`fvj0!4*(RFQmHovVT#sPe-m;By8D&x^1nU!l{MriZ}`G6-@A;piGCPC-9!zMn<%|
zc*8;l%UWhTFlpk~=uVgu0lxy?*~B@+d)aoJHW#EUe+YsU!ZLp6;bGN#Z}Fd(g|~X1
zn11*mpjYE{)G8_HcTL%NboyX>fPQH1*J}n6DgDOcDdbR5y3I%d?43i2AtlK)t;2Ax
zcKJ_{ja~WDwkdZ8Z{v`;dh3weV@wAsp-u*Ab31e$vWER%ty&64Qyaq#0rZOgS<)h>
zp)DGovEt}m@|p?0*0_97`0K7hvI-vXnN;?x)0<t2g5=<9rVip{B~<6}v$id-lqc1K
z38HNJDz|7f?KKSP%|H$F*Mc^;i-y@Vx##;lpXyA3nz8nWy+*75kGQvvin41Tg=GW<
zR76QdLQzpr5Ri}@L6k;8x}}8yk&*@#=@yVi1to^=Mx|R3kXE{tZaDj9<{o|Z_kHU<
zXPvdq^T)HsVeWnJeeJ6=(w&%Fq-`oXH?I4|s|YKld)2EAj$L|FH=OnINBUMB&Nyns
zU6AF^bhZtoS_!kYB`K8ant3;8X2u<NPP>nrg=EOt)=4ltPMdy>cBnA6ZJqlf|GPQH
zp~985$9*p*mhsnb#12MH`>jWHotrY-+VzuW8mZ%a?Y2@xkm*4BXy_8Hz>{me!gYfO
zW#knawR1u&@_>(?8^qkCn0Z1rBHgy*8P?`Y0Nz7hv?nJictP>(#mI)Wxr=tK+=6#n
z1=spd{h;NyZ8kGPU619w<w|R(6eF}{eTC1}xy(#@XtiRlf5XhI!Z|5+PGNM6KaR%N
zMbqXeO@!ZDJr<fYK<DYcbHHUKQ{mdwo#Q04B(Crs6jH=X%+?1pZpXx(kT4o~sUXY#
z*r=>*B7AKSTzwFiR-w73gSWl@Exe|K-WLZ5>v?)c>H_UJPvu%O>l~rUskjqJH)m~r
zappFXSZVWvnrZNI!G5X3mf`8mGr<o;dW@J9V-*6G^D7ICdxvN<8&&uM(b_kcY1;}l
zwa|)smVGltHh2xFnO5!P60@U%MxW0Za2wS54x7!q+7b>2BZ|_?>V#N-Hhd90I6H7d
z{^htSUyF^c1ww-J(>D5&&RE|gjM)MRlaM*V_pHG)sq9;8mzfhr{omKsn<^W)CbQ3s
zDv%o`7wHEC=~2`)@8>6jpI-tF6h&<A8}u{-Y02FmWb*V@k{YT@o8+R2@qMydM*d@=
zrE)a<1<fO^GaWo>{sBV;4X68BQGG`jim!O)j5v>;(UW%O1x-BwX;=LMdiODl7B6Zn
zpP(ZPI<kQNSg-QPdfr@DguaP$0X4Cj*-<xVp*>^SzmB5WnJxIdW20VBO4V*UoT0q-
z`5Gg%67nIig5=fjE#BPM8gy<UfCE570Q_n``;|mL@4)&@?U@IDNl8gl^oEm*x!gVv
z>|&>TU#^7S*bI1m<#qsuLtj&ZWSOS@B39|sN0eOqEOs-|>l>w%wT77Us7mLm!H5T+
z`BDnMR%cmihIYsL^tc$TJP|Si2JgrC79$wlki!Tu0?POYK5{fN)FqQ|^e#0@JXk2+
zk_=3-s_!L<jo|n6e}9E%O0TbEXGu%hB%0^Y7FGZyQ~(CLX}gR~zH!Gp^8Cu9dKG61
z-p#1OhA--l-UAip9C32mc5!E;Hb%eM(P(UG;f+I#?h9ptTiEJm2sWhu_Jr}waL-jI
zYj0I1dfGHPQTpzUqB-aJiZ44W1+<xFS_5qZ1$sqW%TH8M{mv!j{NZI?bxE0nMq86R
z^8y7jW^H_i-WG(Gb#Sr34p5|~yW@M{oigG%e~7cWH+1h(hlvmIg^t_P9<8FZ8F{B3
zpgHQ^jEL1)pA$o^WRu!YZQyLy|H$?SUZryhzGPpVqzQbut!z*RbatissvO4)PnnKC
zKW59_q9~VMM-$@F>H&(JAj$Ut_y8$JQc1~RsmMy2Q$_k1o$5XFQ9{dg$ku*GbdD;G
zIa|E1y+U#Pco@$Jd?TFTep|LD_dN$dMtC2Y9G1AZAWGli?3_Hke3?eMPO-~{uuk)R
z&UPY7l_oZqzryi~MP=W`f^Jd28{DG9`}j4f4e}EAOp*9l@kFYEq+$`#s%lbk*uMB<
zt5(TKcB*51T}ZDKbv`afAkkTg#_r6VrnPk8vGQ07)5#muAWd%nD1NWqYE<`y75Lx1
zBhQ&-M=4G()^^Vg*v`)sELfjGwJ{Z=dU9$Ylk<J!rtfpj)6xft)<60SCuZ}TS3NpK
zyuXtBXUK`RbNK+)Bp@_)ze{jTQ1+hXTS;|r%YE4^ysD@<p{cihO{+kHqe8aOTRj_A
z{7=phvfm!OjtYq7VS<s5AqF7dqj10H*u@yB+Qm+4j;4I<NX+C9q`mVXH*;V_HIs^x
zGRM~OXUCE;m0bc_Pj#zbD5bnhD`e=~QF(O*3?&i*g@vW;G9#?Kwjo4I*V~wN@@{^s
z7F+9NZz?c<Yjue~qt7mX;C#kK=hDVNdyJ$NzX@PWoF-0!h4EeH_Qdw~L<<w03&&>r
zOLP)T!)Uwb>MGL9(M#Vn>-Sw(2izg$rKSY<#jsn?$G6rMIh3LnliVxB)_!;7rGI>w
zl~(-StqX-h<Hsn3VZ^A)f;f?Bxfdn&Vh~3r&-uxHmRJi8G81-gp3fvJ=<BYqt+Ugf
zSwK0K51CV~<fHnH`YMby`fhJh<XhFzE}3>8$z_ix8r{!{{tkFS(;aVa?DpM|#}OzB
z-sK`21uNhC4$|eQKBWO|9_~$ys^W%J0}Skonp_Tsfr0c?{6%7JXOEPT1|L1F%Ymc?
z{C7Qj2X*t|xK6{1IrT5-<wsq-iX$mswAz%DEKE2+LJnsCYyi)-q_6Lr*T5t3ny{#b
zK>bNp8{meE!9`|%uSZfj=RvX6dB3r-vGyO<LUFCaGiQF8C{PyBiS`<`C~}k3!8)`$
zxQK!yn9N5I6giKu@i{8yyqt~xcw1AAKRiyfHn(ot@9N0dO2oM?kSwMO63Lp|OPoJb
zbJBLcpu0nBZjc*zBll#3S{>RJn>%*^Pw$eBB_)<D4<Jqul2dS6e-JD1?UmvaLYehC
zIF$;IT6NyA-d5b=@h_j*h^m!x#n6mA5JX{{#G$ZJZOz<%=Le%jYuSzsvj3^-V9}S0
zKVLOZalW8Y|EyxpZ>M4BTHmeNe1b!(76T6kE&!dB^h+Dq*H_(7n0xb#HH*B(C~!)M
zTg9>8d`oD;f1=MddCsZasZZ52buM}}cgG%fxjg{S*9e^v!vY)s4f*7oM@f4KT{BXH
zLpHwmvd1I@u=ieNq+^MRumN!=j&GgK*fo4xDMmGW+Lrx_UdO9`0pHrkeVq;Jn^gYV
zM$(%wqbf!8yLV==y~Uc?RrUC0mL3R+T#v#44xfUrefFFZ{B|*Aws5H0eI3I5%G<Y~
zvI}UDtHhHJ#KWH6xz|^w=TI373K8xP?384h(`*$)OATd{4CGh{@5>9|sWk=X(e&kn
z0ebW{l1jtLht6j6D_*e5vW89V>*H8f3nzkHcmPYLR)S4tl43Cq+Gm6FXT~PTO4zx*
zKLqG4f#QWq+P;AT)NFQ+k;zYx32hPBJ*BsW_1F9pKmiG!#suD}9#I018~DGf*jv@R
z2zst|)#YxI`&$I|SmxH|qHrOMWk9V-h41uX4n^2Zc(i=7!zNp@&1lE)?ta`Gun~7z
z02G_X7JckxkcPUk-Y2gaSFiqsGMY&39bC1FA7J-+9AYa59GRZ0z1=#-_9w93GQw?X
zkg;uqfKMKD_B-V*4w8uZFCOOvwR#r55IsKq&3<-$wFi~aS5zlb9&SG;AXD4yQ)yOK
zX1}6hKi6100UK`Ckm2TNWe7t}oFIqzw5r(ga=9-J(`g;e)9?Gf_GIv91Qbv68qCdE
z3d+TvnZj7?2M{AbqI1o!FgarIpL9~ZiqgvIeTNwRzz|*cmUCU(L!-L>90T9T`1tti
zk<rmj57q4BY1QqHo~ooPowwCO+tf8pzh9bv(2C~$hV}X1g&3Yek{9gSczWdMYvY6?
zw3KHpDh214G_Kb!E-va5+INYe-?#`Zm7w&Gi3Cgpw3uQyA$|w21wK=Wix}9RzYwJO
zm0ILOO5?fFvCyzm+V0iM5B2&C1^aeY>wsQiJVxg`mT$qzJ(xO&5{6O*AV!to>Q1ra
ziIbylfDKPmida{ch$WY(Iqa5<JfBtWPB3BR$lPMnm&G}}eh?HL{6nf7KxolV4{|9C
zJh<1vJ6BdA#?)5<^qN+??HtQeUZF&l(qB4I4j*`BBRF*!ktN1a3wGcIK#A%{cnx$b
zzT&X!w&HG(shjTQIu%iYMVa<kAPeWwkwjRzzznhx^}Pgz=pRNuWBo<GlJ9IyQ(D=m
z;^6lN#b1#Zd?Tv4BhliX-x*Af@B#N#@Wy<jrh4ppL2hp9ybV%!sBSrJj`}MaRmRHI
z-Qy0=D|Bw^EwIrT&pd4xrs7mP-<KdMl!sw3I%psoYD*SJVXY;Wc%59{3|W#8(#ZoH
z8QLp7t=;;~ITf}6so6b1Kkc)Hb1RThkW|r7&VdC15@GKV`i>xy`A4zUSdD^#Kp`r2
zp=6rTQgYsA+{)^T<JFPUT7$V8`yvc*5ln<V<`nk64!BDqP*jgn@ySH56nwkSL>XV<
zpwai<hYSS^3p4;%R-7NkmjGDNQ4ypMN6B4IuPT;9s!8kD4+Z)-UB}wvrvcmfe1*Ur
z=6P2TC~}Tea`^|AUS|W|<WDw!t!?jgVlN`Y#*wpuGt~zr=7mY_z{7t-2H38y?|JSG
z*6}<mo1H7gQ%-5s-&fOT?iXr@@tT0#p?#?T09W9VtA8HGT$QS;QQ}dW<jCcivrkIB
z`j}*S0^04z{U=(d$QV%=^#jR32X?oBCJ2Uw1d<??{I<1alB9Nv-qtjGlICaj8*!a?
zR_!*t&c0w>GGN6npA3<Dh2~4ai16@ckeFj=JJ-v#VYj_*nXCKhP_E64l2yZnh*usS
z9+yTgUcx-|;AB=!{Hq6#^-NC2llw<4+l?e|mNb?Xpd!Rr`Hecuj4@?22M}{W&3iah
z6nZLvmfsM}h*DYt^I$)KQtE3~tsw2LXM2%tFUP(?O6!3VS|zyW1vbB;nmOUfsn@>8
z6zej3E1L7?occFxX6Lhow3XR|3PL=irNVZd#~b>wIC@u;p$E|gx@{V}>~y;h#DKXz
zor;ILMykEZ#=-cmtYXl_@(81_!*cRQ;NJT2T#z5VR9d-JYS@{oA5q!v4&|W((0@Jg
z$d}RlbF~rr&FbbW{y^vSzV+r)deMCLUN@C0(azixXrZnz1wSxJ!63=`LXa832<^`&
zp7u@ZgAXz)HXmAnENj_$0m+pN#^T(@t6I=-n{hD5kxLGA>+<i9TzN~sQ@;%cFvWw?
zr@hhHGSHo`nMrs<VyPD>Ybr)8pD~JOke8df^kwGiZ`K2=w((o&z%9S2RW^t6vNAJ}
zh}tqgP-+zA5hYq@&P2rJ()GmNs<!|~LQaSV%B<&`?XN}0L4UI9(IK+9U-(O4Jg=^N
zGnBM8tnM9kVrRWhm|J_JQTbgt)Qd8zuzU_>IM<us;Y_7z9Sn`CZ5i5~8;Sx+Q_`zo
z+><sB+mjDR{EB{I6*l)rpd6RY9+kVgrfO|Dr2IVe0gSH5Ds6@-M0nGqy`8n_%I&7g
znC;T-#isqVe=U$wBq#K}AH|c`y|)p+cjK5n$a){2vv4hiWWjtcTPyF#(Iv;qHAg59
z>D8e31rIj%@0f(;1Bk%AP9b_<Azz7#1p|r?PMg!*R#sLH%I$M^EvFOuxgMI52r^+k
zs|@XN$ej4ma>JG=jeIDND75q?G#5$&Z^9~mD<-(Iizg2_XZZJ1_)NaBxh(i{`YIQE
z*j$-u$=q(t{3f#ml17{AwXcq1-d;tDpUCiImqw)7+)pUGRE(-Hvn-&TP=UO(MRTH>
z8<x7?S%S>)q=Y7>^ys&~CF+E(^(Zg0yx7}r+cR@nQFiSFS^k~dpXqmkp;8FvIR}gP
zdgY*SK~m45)hdn?Wi?uVg0V(`4vX3C8c1SkA+d`fUnyNsiRW~mu<2}x-it{pb%C>w
z9~{L}@T3yJ|L}hIm6GoARwpp?7Rs*YaRsk47|l&sq&qIO{yh;&U@``+Mmd&1@6)!~
zzCyN4d-AQH<!gf`XQFv5xM_IIW6USYA-7aBXkW)E=mc7@$!ISju!7_x@i+cJpLyBF
zsL!iSeBp&`w}O*f%j_4f4_N4Bw~38}<!9H;S$$T)2@+m{+Y6^u<g+$D32%RDs02fz
zO2?sD^UZa_yOvAsqQ?y~e6}CPZg1@j?Kv+kMpQa2+N-hwUeFABdUE!y<tq$6@3|9p
zTrZ;FxWKQf#2JvuJV%N{^YaM%noFhHg9pELi=8d~N#WGHmA4SdW*UQbo>wtP&%4BT
z+s<CpTW=P{x@bu!<rsSSvlhkULG4Qv7(5|(TS#~s8iv-A2FAknxB2sdtosd+xMScl
z*X`00<2YX&iueXb%m(vVzu%&FUcSe;H>=n2PG;3$D%bIS7ibsZw><*m<`j{*Zrv`8
znk~2amf1+eISLQvv=5vZv=})-T|=v)+Efgsn;U62OlS)()-#5RgaM8K2F_L*+Ji(+
z*2nL^fL^Ne4Z<5|Qb425uCmqqOs>W5oy{kc$F(z0da752d;^{G;-mOn>C}acq=oLF
z8%9764lX(8X$2-xPyq1?xEHUV3aAJsjsfG<Liwr3*z_B_>GpHR2RyesbKVHQ`%y5q
z@(X{~p=oUn9Uu9r>5_EyYODE1zROn7u&=-@`;NooyE}DrnItkidfS^J8mPEMjhA=2
zEQeyyQ{<oQfAGKPFj!_A%Y86rs#X!cK+`tG@biP}&+FalgQiX0d1X{1M%E>!F@}a?
zsx5-H^6?t!mw!f>9oZ0eSo|=*w`#Fx2z64Xq`W^$v!|%6X)w4LUP$hu;zXkY6+Lat
zRPN49wL=8k0%{2O23Q@=R?=Or;oEvL!9QdIHE8`N^K+td-M8|Zj4OfVrx(Y@+jeK$
z$|e%AiBj1|>cyV9+_x_c>(ma6$VW{i1EOgz**YKohPN?*CsKKHx;R~Cz;X&s=-4^p
zH?K_CR6n*itqFQLj)V5jo1<25c7M?C>Mq_;S#*tD5#99}f;ic$i1Jy3(7vU;QMq0H
z%^vv@)FjIJaQIH;522;Wf{?@5N)1vkg{s2!H@@7KhD%TGheYhGtq(e6HrAesk`W|u
zz!=a$Pz>_vN|^eP_EJ>0v{1FC(BZVnvv#+3rT~#723@ezNA(q?C)<zcZ7Q!>-%B+{
z?67En)RR13Y}})5elqx+*G_j*!(Cstg5*&lb)FgH_&FC?S3Te^05pPA9FD8A>|lna
zXA6`&YBe5d6g9NnGIX71$jRz<Q>dK;`y|McO|S=uCA5*XULe#VQ5hM=<5_WhfV-eF
zV>0^i97e;S_CZE?oG#Q!bRD-FGg!)R(`;#it;V2?1M%;+m-53ZtJkLFTzj2n_2@aF
z&bBR}pUM$sO=5M>i*u&g<b5pn6r-I^@$n6cO5ycq!q))4+qDSqjtIvss~$D)<I7ae
z)Dr9fS>Sq{ePT2v=}uX;v2~#3=c@es9Luq1qH`m=TR+*ho7wvUCVmjr(-Q^dhiCe}
zy%R{nhIfvYf_G=RqXn?=_j99L%}@i)fpIZ4gl+nmk&*h|+l#i@AbdOVVzPqi$uL<w
zZ%0OQJj!@$regKU=iIh+u^s0C2`5Fpj=hl<qc~=i^`I*IG`C`{Tyc7TBpdzs)f(9~
zHjVc&OVn1OmC5riwDWz1ft9!7CcPQrJFaxwEo#kv|M1M)@Wy<ezQ6D8UCBnix2!r1
zR4cAPE0H0-3-KMv%Br!$+fv$_@=Nb{%1su3%&mg%yFfbhJ55PxbGYuUxmxV5Xz$sK
zw`|`L-n?hEyD<@MS3aY8vyD-+Dpj7m9hECI)WoFXh&R&F_1<3tKa7IJx%Rb;6C$Tz
z@trqUfCWlK?4?b_VePip)+01^8HMNtl_M%PZ9hwP8e0(?jMo?2N2qpcDstGD0%mS3
z_U0FZwmlQSyc^#8(^^QkEiS26yc?sxJ)&wq7@Abt1y)^YKK;J5R&Q5Jun}ZCQwSAz
zOoMr69KdVBESMV)a)V`jl%SE2E#X(yWgoH$ukQ`*ZMlZJ?kx%L>5VSs^dDDs1zn<1
z(Af7^HZxtZUkm3+oV)LbO_!Lb)uqyq?O+g095l3*78y_RmF9>TWLhd_pB%}Vql;KA
z@t-}ikY(jTJ>-9*`bkJUU0p2WtzAEu@$L}%y>nXuO`NNt{;;7RI|E&o-^!ylY^XO5
z*cJCp=QOdhpe$(0etEFPI34N&Wglv#P20=wg}DIh-$MblZF;KW9UG|BxL*9WW>@=B
z@v(1fulpJ^?G|8V=NE8=$?0ooCR-NvMe)468@g_$&7<wzG{5GZEukI1fX!Qkd;}q^
zI)dB3K8=I@j3yed<Yzjk-<K15duKi|<V_fP^z>k)H5e%>BlNRwC*OzJB31IBGunN7
z^NkZpH_+;sTGr+VW;*5)3l>XrH@i=iAAW;nYlPP4bcXIsew1PgosgF2y4WXVR3w4N
z{BC=3SYH^cd@M5p-D9Q`3^N*CN?x82G+-AVyDY~JT0O_wkAO<#4N-~Q+ED(exXrKc
zqrh+<J%?{QS&LJxFW4Z>$OLZ&bhKZgn~QUtk1P-bB@yp3?O`Ajt`mL}2_;V&jUQN@
z>dTXfu;TI$!{&@Ikjg^(hZC>ky#>i#hwm0UT^2I)1)E9Qn@J6sBOZEOp_QB81J*#r
zX~)3PMKvFNXnc?UmEwb@ZE=F3r8z!_BKQkNl~2z-<8fDdimlJyJW*f5dD&E3%i5kG
zv24eFB(Gd}M$ofGfxfBlC)wM%nT~MSR|y)~=pRwBmtv@4-k`X7Su1~~M^2s>Q~>l{
z$G;q&f3M1+mNSlhgF<5Kh6<$fN(vjzfI+aawHI6!7lwV=;>>Ycjt_zLYcaJ##cQ9f
zC|dLadxTbtd85bg^u^VUUkEmA8KcV#kr{AN98b84h52&|l@qo_ym$IaY|`wzO9u^H
zKngp}<?rHnY3ajZgyMKJ<Ixt}P)BzrJ26^Nr)c&q>nv};w&EvW))HgB4Dzi0k3Gv(
zZQA`H(|<e-<lI}Yf$qNMU76a#xNLsHqjC&sS)6583>m{?<eN>or3$cCUxE)Z$4;l=
za8}cVjzso!4bWppMLW}~uA7t^wwzu9<2m0nkPbS#j_|mA>Id^t$JZA|LO=F^3LT-P
zzRI1x@ux>vU!58Ojgj3qstyZX+sOyYm$Ix-mdm>nm{L>_H%mv}yZL@nC6DL?HLT{|
zy$P9Ipy;o2T2qO_PaWj|rUXYuRBn%rcf{{5>&G$P3823GAQtFQ<6v@Z{4AL-h;34D
zrM<{m0v+_rH$1f0JsK=`VV4o1_0{jYr{kS9?|bE<N&-SRu;*hSfvy&{>!Wi<YV8R*
zo5N?{=2J#7x?r9|VB{yy<8s7`kgmJPHgnbvB{*@Tt%0w4KT8+#mE1y7nfdteo7aD7
zd!KH>Z#P?^JyVu<Z)kxXrkFe?Krv~8IlOnmk9!XRkJNsV1<;8dzi8ilnLO4_Jf=)D
zHh%@y$AP8!?y0xG7?sVk?Y(dt9~GO;NouoM)vbZ`F&i!Qfe38Ghg2MLLW|6>5TNbA
z=;qM2PY%C69BzS%Iqa`iG8fq|8RlA!)IQI-K!MTWYZ<>&+HAJ+y;IpNz}56m%7R7g
zx?)?-)3>wQsan2QG#Zu0xDcxn@HZ>c8Yn+F4QHX8-EAD>Z#dPx!-mlQqttyLkPnYc
zipYeyM=hsw1m=Yy<m)#si7RU2!3ixdfG%>qEBFc)=pdQ0q&`&g_3a780%<;rfg$r5
zTZ56uXYkG^<YdCV05JRhvxy7jy$DKnBY(Jg=}Ub~f%sSM^w_$Vj`^k7(ua;imL{;m
zWeK@OOa;9TVJ!>L`v&BZj#7p^Jp9?Jj{NatWY84GJ>w;>Blhy%333?g+>i!%^S4u_
zZy&~5KGa@d=k8a?b>VnTFGAIhMB--SRY{3xw5@0`(W%NAj+fisu-9<<$S;=XD|3_k
z?K&W%AD<!!(bj&1v!E(G%oCKscDT*E6qcJC6#~PXz71=J1wobb%Uv)1V641{7ync?
z^sB)(DZ$xzUYlsrCGLh;*ab&g2!37E?Kz44^#S)*f^gtl0pSZ^=bsB@6)HaF8&Evw
zqvc8c*4QZk8&UZuxP1HkcZKW34Z+F+A-szL9x?HtSrNvA?+(=q<LdgscYLJn1ZOXw
z?0KqqxPwYI8@LHTC$V=ZYjnp!IyP%rncBry0a(W;Lr?<P9zIprOZYH?c!h+c3I}rL
zoO=yTifhu)BZ{kGH9C6f92a3XnT`hJ(#s#zhh0>RC)&jrFQ1%p&X(7&kql<(>O6<_
zCzYH~`P_27dA~0h1YoQ?YX1{b5O2OJb>cP~_WF<s*~qX`pBInU?km1zKBQRp4yy2n
zo=Suw)3N#I$*{i^M>2GZD0U`gdt9B-1-VeEBcGwJNf12#M!{?Hc!fcFzWie3LMWfj
zLYpfrtUQGU;|5S690We<@B{JgT<Dc^YPel+;axssqbe}{IHRkb8jkJro%l|ImVFh8
z%RfLO@kPkfmo?ST*6Q29z%vIoVVQaSGkl34?!kUBQ(SMY&!|s}vuCFuR%o!Z!2j*M
zlgw6kY=jFI9;8r~Tv69EW9&Ky+#3<*!F81AS*x%Q)5{CA+JPv&g0_m0cF+-K|2MTD
z1JyKUMz#xHA<~WY*DXq|xku;OeD3FOOu~T_VGwSrNx^~sD3T!@5kdZT(@U&P;-1+7
zrmJT$#6o%(a2KTX3#_XZ)D9AVJLk_jSbt_!rN>F;h-_0+lOeVZn^~$_osD^geP@)P
zEcX4^RKUrfY4Ra3&QSo_UaD8o(b2LZmL`%P_B#%NolmNOm(+cihH0NrpXmNok{<qu
z)YYb_KD+qXjzZ^9Tg}$a^Rk;T4f7yaN+MsV!jJ(AQ`R$Y>d%?-2cQ>{+Q%vT&S#*%
z<*XX`*TJkFBpur6_?wnDz~Ru7WLX6Cg}g7r-kd2!<pvZMvsHuc2KlFi^{@dd<Q~B2
zP3?o{uq1_tFMG0GEdCwACBZ=nGCqEYV7Iro^#iv~smO*lN2rRuV_xdLIkTMYj0>w>
z2hp0O;tYK_c*{sesJaYt&63d*l$j#6!vX3yF7aXUAISmU7osY85{{RCCFp-3TYv5z
zVd1x7qF^#T_;&&bCoe|<)0^k9aK>FE^n+6%$`<DS5q8}o6D;u7%+pG;_5InO^C#sD
zXP6@4+zxS30Qu81;*{_@Rkx5n!Uf_6@y^C(pOYo&<1KM&E8%m{AcjIDm>q_>djNq5
zRZK6lMh`ESochv~{Wmz2cpdnrld=Q!Va1T~#3h#Pm$ZSgmovXpIQCb=7d;2EE*^rW
zdpJk@y(2{gT$g2w``*4dYmcS8<nzGwe3Hr^{t+JjGzb-B_H+09D=J4}X(bW#ZDA?r
zy1)pNtIjn*D!{pV3`}XphBVz}@s+GOj)Y=?2=Vn3kB$i4Jfu&Gd+Gpy9haNfIk*Cf
zI^L~<CdK@RBq$xa)e#V3b0yKXYuk1{Xz3ay%!NRc9hSM33;ftcyflzhK_XOk%j=p}
z_alX^=QSr9ov`q2hhzlES%ba9q{d=ZSC{W!ChgHN+v9Vdqn3%AE8`9z)hpI}PTfrl
zI-y}U+g=ISQRm)iQn;Lm;V!d|uTau4Ml<Bi%OAzgV*RRzE&}@3IdZ!Q``k4WyrDO8
zN~LILrS!>Hs83i|6HwQp4^)a!_xX^*F;G9r1plY<(sri#!9&Tiq#-cOyg}KIEItlC
zw}6pTn0Qz5x~;kgx}cW!^JOqsdw-8dML>2bL%25s_MUY7K<9|aH*fBB8ybkAGe&e_
zGW;vX#&AssPt$AhvR2K@jDVrq@p*bRS?<HIe<HyZFp!;0yc4)fYalEVV=jCWCjIn0
zEI!H!e<>3>>Ke(z+hkzHyDU&#^S<Z}<7w=7)TI#d?;K^o#@k7$^N>oPz4_x0l_%oE
zfrwLp{5bA@>x`{hT+9I>$7tyy(4N6+xY~REqS3on{Mpm-z-4AwLu|f1d$i({w^Dl4
zj^#O1Az|SmU3}?IPte^q{O5?kG*OK6Y7ZwZz0<D`k#iY5v)$<G*LqDpHMMDdy~B}!
z+2m~IvXS7eP2>n39`^FP)Jq<;rjNFZ{<z~`{&=jhVRSLS<@8!1399vqcBP(kvzDcw
zqI=m}P_cN)$H172aKn&)+@dIbcvYatIamJ^R0#LQCH!Zfz(=BG=htUFl0gQoJyYTl
z<r2N2jman9SSi16*=EZ<yffx1u)r17o7y_g__JksE4PQz?3kiX^Nb{D#A1YcYDv#;
z>&KsUQbSonV(@$ejh?FBe3!UYu$<t^*T`!9*g!f#zT7~6Ykae_QCuXY@6HO^&Gj09
z;i~9Hycd4DP=@o8l(fJwSx|i2qLo^OMN^vz-3V#WkOSeyiV#V@uKr~^EeUtgu61Bx
zd?Jjx#6sk^;oyT%N>cEseN8Yb6Y312K0y{al#JOEzzHDZ5<~}@>DCL=O%dr+WPzwg
zhtx~IiwGKSvsVIaSFbJz(;?!>*qvIwUUt|xSsfH_f!2G-RFejpc*x2d`hU^f;A|7H
zx>yWs0Gckx7w03774tjj>gqn$tU1k;i{|vM-!k73_BK3yzU(F*%a581TrdqQGt7MI
z7QJOE{EWF;MX@%>X)~%+G6oyFJmCIkm%M%6OB#j@8xAw?Q(|XhW_Sv-@7n;MPA`1M
zQM=b2B-M)t-p!QB5B%9EnTBCa%?WQe&bi{)9hkFXrTP9p&p}-G!#Pr1$o>yrkAxEH
zoM-}Y!~uF;U)t=vxi#Z~_f8jTd=2R`;wo-!RgzCEB``hWKr$WGC1#SOc)?RxVOphv
zUv$v)3X4n}mTmm;iYg=sOPeMfx$_PFn_HmU4n=vY-NUP(9`{9jbo5f>yF7pXO8=BR
zvSaWG+~Yw^Z}>h=Zl0ukbmWeCl>Dtm{)=-W6dC<DFr}B!o4$fdPDSMaSnF{^;v?n<
zq{Rp>(>@F<D9Iempm|2;3ENDd)g=&FaYx~_E^xpyd=bh|Wfv3o3oXEUezI=IE$IhB
zLa+U&VhR2PHPM-`I^E@P0yfPVg3xXip-nrT)%u|q;jGYKpv%En%`KcTVQ|6I{PmgO
zb^EDr8&2cdDABqqIEfo@SkV7b#V7|(f;d66PX6+CHOv+ng4_V>4y!sBg1f->_Xq5h
zS5n%rkx4&MH~y^J>{fl)++aL0xj3KprZ_QqjIJ=#*GE&V?skq-BK+qYka$1%$sA^O
z&=tdo#0xHxCBTM(82*ZZ$UUbt*=fF767lpr_Gb*q0BaIrj^e`Ee(F_}$+hXoRqPd$
zs<~;OJ^DLQ-3=KC`b59n>9(OZu+z4oJ-!6zckuxP|Dr0zfkk*cd<Tl}^J!Wek=UZK
zM1pmSkBJcxqu5vLm?6l6YYT=)F(v}s!vZQ4#{6PZ!%h40Y)lLhMv90lft^U(nn=I}
zF7QbqwEkOnLEcC7J5zK%COwL|_h(e!l?e}5K0l8`X93pllmY7k=sL|OusZgM5-9Mx
zne?BrDj$R-@b4P{4okXtOOV35!=W=Nbpeo{yt6Be4me0FBVA&%GPN16M0M|Q)%AbB
z0WU+y3vO&o=NImfzn4KW2m$n2NGCdhh3KEJ2hVL2zkc>X3VsFXR3}zh!c@GnVd^1T
zc)<mRnH)f%&0HhUMijGh0+l<r@H5HUf;|U`012!-W{?1wFZozi0+U+_j^rhm(DG5$
zsplk2IM1OJRQb@~7Gym^qf5;-{dID*$U_dSzB57zpjubHs}BoJZ=$=`CCzS1zzGZ&
z!D(x}afkQ2izL=W_4_B)B_?ox4E249*^lu}uxT}<fjGSBKhgLRSc0N8BpAMJ_e=Qw
zrEx`!%sv;+!d%VxQWdSrF}x`_se|yS1d=N61v!5<%+;O%&D-B@t^t=9^BeBe&!Ow{
zGB}5Oy^acEK9ik<<?$O>^Pz_FacULMUW9kZ5OlyJ5gBw8c1M*HKRT?Gxh;*`K^|MX
z!HKk{;@$|oL)eGB#K18hs1m`f{y#9Oo*dwwG^H(e!v_#k#J?g4J2cffFnKl%iqMtW
z7&O53e=s~C?7s!r;JG6br(twO_8|=7Efj6b=GLh(d82^QQ-Hhry&eGZJH%NR)HREv
z#L5a_N+0HukWv8LohPq+!tEoQP@f|15?@UYUp1`43qd&i_7)r_MKavPQ?G7heiW8u
zx)q*30;hyAkODPOiBPN-2kS!@@4h!yeEEhh5TP7QSqGyiRNtQZAh@s+OmAxjWes|C
z4wGkdjoRoqg@rA+&HIg8h4;1_Dk)gj(?8co1pUH1Hz|Ykku4z&qwu>byOGy2jVZQ=
zfSQ0^uwdvv`W6ya{(to?gkY#pMH$mSo8uz~rTH%MnO+%q@mtR1fgQHL5%jaGqqTc=
z-0Ay4$SvxT`x6YZfMN*at^6RK3B2Y%%+VsOm>%*t&H*k3c^>e_bE;k?y0g9cO-`eX
z^pTgfG@TY|#zUl7*`Z*%udnYoaHexTKHT-?%NPBqpH36}p{3aw#p6*W9be#N>O)t)
zTfJA*axic%7G;qOfD6;xq5CH*kK1AeQW(=i@cPOB{T_lqA`0PM<_n{ESNmS17Uut<
z2jVAbJ09NmGXVZ^!->||4lQLZ`!R_c7RS3nd(&VdZR|$=2R1OTboCm}{Z-^fo;)iY
z%Ur{9^&_0=P<;^)X?D7}$CFWIyg5B{^VYIi1Ge1g$TraODZu_CS)5LR#gNdAT@VR5
zpG(UR+Ya9l&J>C3D<5>@d9Zy&kj)TJSekxWZTr0nFE=aen#oEclVAf{HTPYjCp*-~
zub`*YE;CqLbQH$f5pqJ!M~$VQOk+y);WV?e5LG7{*6Ec`$yh`+7lF51El=Ld>;WC2
z1W|UqCht!bTlwNjym4%dffTB3f2CZ<Spceu9P;7jdwJ=<o9~fQdm<-G(`#Sf4u>J>
z3z=C>l6hz9K#a4+Q)v0D)$F(IDe1-6KL!4vW-PUU#>ilIs}#hfL;uk#kFWQiR(VJU
zTK6}l>6@EwYu45cu={C~^p)eRXxRq&PoYuu7=`(q&1xngT{`>G6S=0{tc>x&V?r54
z!^BIe%H@q{P@OfPxv}CnxI_naf^6LBCjpA}H85S+RiNjBo%aHuj~z5u#aLo#EsK(X
zi$W%P$(xMni*z{JP)Im9bLvMdX_`|0+8)~gf4L_UXRFu7DZv1ni1^R*YE5m%6|V1#
zSLgY1YfUn;^*fi<T#`BF%AcJ|6-QV~eVC3O+BP)22#q;y=~ULH9eu8?rpB?D*`Noy
zSa;0p?VO82{6e@t#~*DW)9xU;dMV|oQdN0SgOmvCBd6Z66}z>*VQMOw4C2faUUF4C
z@v;kIcWyycw_mjzar3A1`VUX5k;;`W^_IfQ9n=zNruG$djAN@u)%^z##z3Bd5{Th!
z^sOr1{nR2byRp-3u+0AP;cC!+Q)j+tgk%W>=}?c;enI0uT6l)cFfDmlzJEOmYZBNi
zhcE`8>N{uF(@-UvM|-<F<Dj%kD>m_3=cw%vh}K8qI=r>6g%_69fZlFxugsP07s4`~
z3=GlC&Zs^Q4}ztSR27eMHb9u<<xoxGHE+SDOrOTg_fSL0U#dplSLfS0f-}c9Ha1Ki
z=d>>F_}_^F0qo=SLd|P)^${wA{4WJZU5z*LSZL}5S@jmFU$~VO3V<3urVn&{*2dH~
z$qWqeHu_c9ls%5q2^fz$@`CJ-QRU*Xc&FCLpEiaL#3n}P={vGrb~f`un;QZUN+i<C
zrF-)~cU$YViI(mAAALFoa;CV1WUTs}z)Vk0Xzud+y0#k=XRLNsy4$vq(R8!xHLG)d
zm(8us20N{1y3(Wjk6p3|U%%NK=1{m~)?f4)%E0T87<(3DH~Wo!KLr2y?D^xD<rj5#
zKB@JgE0L4CH6PI<HXPP<XpnX=KveHB-<xmBRAc;iKenPh1^yOvU1ZpHpxy?(@1k%3
z6%|#TchHlrMPb@eJ{P~^E*_8c(?Ewk;eq|;s?gBh^Hv)xA3<x*=-gI5Q~i@rPd0Fa
zRqOb@%vbWc=GT?H{9w&V?)yTeeO&j`od%-Gy;5cZT%^YhNsH&BZ1Z&zRtXdV{ida=
zd|w6ZGRc0NGk%I|`ttjj;m!<na7CX_UrkdrFHT*+NegGF_()+dyO<5sQ>mS(*Hi$V
z)r^=bcbO_bxwLR~ziMJFZ8k*%qrN4tgCITfnHem(a}b*uv#H*X*Ax}mv2hwOYZcGM
zBP&CrZD&+8+X9rvHh7)~wz7=Y)(dNpZP#`kNL#p#+CB>c>BCxoqz{WgYK?dcNnMDr
zF6&;mH4p)zs?d1lR|gN1NbmYc-=l`IMHV9n=S1f#a$HFU=0|WA?j^GN62~UW{AKpi
zN;7F}0U96)d%(TJm^j^b<}>LV<0Z<$5Va$3!-Z{FQ2}ReaG=)SP?Z~lW`=*z5PU>G
zAVwc(P_(@oJ-9y+TUH(5+7J)VfOTlF_T{gEpr>bK<wkTRyB(OuzT93zrEL}m^0s6O
zabfAHBlQFKR<$>KLf|lY;<>6!Qj#YZltCko!R>LLXwdUkS1k`3A^Z@uJGNcQdoS2x
z1`3owAAMONCMlQByYpXi0-^TD-JqG7Z74SU4AhQnx82SUwFfzHiu*4sz)B7ZQ$S4}
z2UiOUBLiKxM<5v*YDNdL!<N7VY(}13BhasG9HRKC34<|`DwrjS-$n|>r}}>+i+2ob
zzASPM1CLvviwO&G{7^nGn=1RK`xprnD1_R1Mjv4-ySBs&3(-FnYHFYg*Ay{xJy0ut
zoD_UqKccOs7xd`>4UawP9O{7#t%W>Rp{{AlnHlUHnpnxiVB=me|50+le;fNn0zWw0
zfT|rw2dKDhjx>t9u`#h5_EGkt(MhbnOPz=u&^|lbjeoefA0H!gs~Et`k3H@ogqv28
z5;EQ`d~zYzXeB9`!cdc%0DF2v0;Oj2od1EH0=5o?mWI`izrh)rY7{~IiA7SWiVqH9
zV+-z-P*S>EV-So)U=RTL2YGGXMXWZ>iYIiCC1ZsFb_pVNND-?`nV0mBJt!u`7-#Mu
zj)2yU-y?!xo&QfD=0U>w4@WPOL7}SLs&P;ztS1PYk_>i=59ulYwZ8_(0|;+1fJ=TV
z#(?v+{`qPrXzN2S7;#7Yf{+o~YTP-Ro)|0+2Y=p>0(0V0xH|*(wt$<8==x=sV=M~4
zZ5jZgnk?e#-)hb8PeCkT0}(F{!tcOgH24WS)E!8d!V3>Z)XAF(Qxhr+rq2bhoL0ct
z4IrdOtiq`I<O?<m0RI${urJO5zw{**v6z1suM&sMH0vJ{xE3y0We7Y^?NgQd8ej5{
zr*!d`jg%XVbf=t!IJtI?UX8zD4D((>FW}*|r27rg-=A(lF8b-Qv%K)NCOEtE*=gs~
zQ*h#b6bOuchyJI^|LzbJazSl29^E5^>m^8y6_7%usZ$fmTysN)E|{K808CRLXe|k*
z1QGsucFb*vOS)Bxy!!e$1nZn2WN7zy+bUM=_wV$-k&D>{XgUF$Q-92D!EFwx=?5UT
zqH<mY>u37&4YgMQbeq$_oYOxZz+7zA0dV8Tbb<+?_DuVLH5N^nGE$b<d#9HdOr?i4
z`M@+h;K)%|;K0TCU>w534>|cglk{SK%y}H3!OSg9<=>s~2%CY#_}y6IKlcy}#qrH<
z62jgA(n-nhRIUqDrZS3!+UX6GxMC-Pi(sZu#~(-!Zy3n{bv(GocKid*M`B4_uc^}S
zO)=3+N`cs`Ea!L6$weQp+ml{32noUuLAqVj)%NhgtC2jAlpyx{YTltN!E-OqWaZ`4
zz0S+;6**2mEAsl4WYQ}i24VsdABvkw?m6#f+s~4QlM?Fd<+Ke=?C!4dt+m=3jh9;u
zjV;AlwK+#jI1f&2IQ``0T;u;K;s$1e+LH%WHR|18@y1MNWk5n-Lh~Ji81abEX9yIO
zp~0unmj{84N_J+Nj3Cy7i`3Ep8ddBD215}0ot_O?#STCwB(0A_qCT1=6QO5`jM=d(
ziQyUP7_dL=m4?@bUveqP4PVLqX*x7>cz?Ax{1}hUP#+u2zmZY+8X~XW85<wLEQCyu
zK(Ko&I@;@M%LFe^OVx4j?-aB}16xfL>%4<qUp(8bTEXYpTNCUunF!XivthM|Fu!&~
zX5*hP^r5gQ#|*7iJqaKnpO=RjmJDVwBi-FKM7~uSCP!gG(Jq5Tq|ta&V0Ve#4D>P+
z@T{c5^*j&kFX4kRAQPh9qKoVpv!7!Ba`y~am3xv-wPr(O^78%?;?dyrz31Kt!l!>p
zY%OCeVVHDXJW6c;D36)HpZpkeNU0FM1KbYmPk>EJonLwO_ltuR04)A|WQDMZJd5sv
z0T#lBN5plw_z?kef8PZ|`+KnCwFbSBN5pw~dCyNUj^6j|a4HJCQ2XhskW-y+c6zMe
zgj(?SRbR{92Yf+v-h0TrFX{y|Z5He@7BL~_s($%Tn9`#HHXHSNhX$s+?l7v|8_i~M
z@AQ&;$u~neIXioFj6Nsyq71QR&r5B0X!_tf$C#W>^DB16Sf5H2Cb>9`{8o`&d^L6T
zrS8jusp?hKnbFK-1S3Ow52C#9UieZE+QcjE9+FEG2}_jK`I>yM)_VF)i1f4D0cRqm
z#ZIBc5G>DbcId;`9D)E|8S|u)6ovOn<1JV$#{Y7T!96TooIH<qEI0GPS~psitiY^C
z{;s!^9e-TXi9Lm1Ann#uCYpP!G?!pbQ()vF=tveeXfaU2g;vW?%`v;EjZk<eec9&H
zZht^F_n6doki$k&I-4U^8Ou?-b)oT=`;+imP>&;J^dVELoR19ERquY}rK@?wnntCI
z^OVKU#hkTm!s^_a;UWD6i~04!!pp9Rg+|XOEAl6ph|uS53nn=SL>w=9el6*PM3(A<
z7zRV1)alC^g6UHCLb)#Bxl`*|${ygjpqQ7Re=PaJ17U`Yuaxtr>oS&7P+2c7oFmn{
zI_Q1(T}DWT<s4sp$%}{W2?Rc8l<k`7mg{N*U|e$|JbK+eEKqBP086A7AR@i2)+>oc
z4C6OQM@L7p!1;E}BgkNa7<a|d$r*ku{_7k-w0=E(Ap8yX?{5$%RlFak`e89L$in?q
zbx3`qo!?5p{#S|jkO&t4d!H|2@8>t>y88e^@c4N255mH(d79vS))#6D)Vbe;h235<
zp}3CXrDph6eJ+#6vrZB{tMMz>+8;CTjs`z4xFFvvfqg*S+f83t67l-f{?Z9Zuv-G+
zf(&Rt?eCp{9pqdB_)(K=!uj`Z{vSOOLTYSkD)r{H4SdvT%Al&$Xvv!}n!6C3Pg<l=
z7sRk&ds4s{M9Ej&|Km`A_=EHyetuZF`(Ix2*CW^b5OZ^LpPpRzgu(ut!F^2OFE~L<
zJ_JkN_q;O#FFAGc(W+v}@M|s%j4uHgi+Sg{`~zbS=4;Gv=eo>bOmNpFe>C2C?EybI
z2I&7V+<XKIg$h#)QKcZ}#GppiHKL;Dk6tCvzWDo~NS#OEz0a(2ss6q}Ftmk8T;~{B
z-X|EA@W#PVXjvq}^pW{0eB@EdD(Cda6tb~r_s?BS2EeK*ZsUc0+cjQ13G|2J9$(DA
zjmrQAkNDI619hYhA=r#)RW8dvn?a5M<UVaNONcqppT`gdH*y*gr-_lVAoYXnQ9SyM
zn1<{A7f3#QKoutrr^BdwT{9s)J$(w*egMV^Iq*`!6ctz?!aoF8b{bsS*XJUlzfTUs
z!U4##X^9#>_)>5icK-@UDd{5l>3xL(?AAB~sm)^?*tgj`*dNZ%hxuWU9ay>Sj>_IY
zE1#kQuzPZVEB?<TLWlnE9|?MiR99D*S4{E5t^txS@s-}hFyr(N+!4jq0Glznw{ime
z&Erd?ZLN=AD3KgYliG(g10w*=DWaE#|DdH0GlRS0*ZLVA?9mf$CzO{9R!6|PQ4lvc
zVxHJ7$A6B9p9-!dmG=q^OTy)g-)GT-`e^JN@6%Tq+&hhRSN2&GSp6{rgMHXDhH1q9
zo9eT`VO~6YKlL{zxSs=%r#RcVDgK{43aD9sJ>s57(A(SVfab{fiBb>bSHW?;yE^xu
z+N3aFKxCG$s;T7YOMn3zVqi3{j4=vzjMK20FxRk!cx%4Yf?CkQT2d_)!yf;5o%=38
z(@ma6X>Z^Sx}ae<U~{%&jaw#ONbrp7_QH><gU4sny;lC&`F%j1*A#zHJL%0=?(IGT
zWwpzZWj^_ppcL>&YFY;I5SW>8*!7y@KbP?0Uezg;PhJDbqJqoqqCbF_KK=gCmoF#5
zxFco0riasCQ<9{J;z6?YBj73VXl-qkF9vg1O0DUAK@m(*PW-9A@fV2^B4K!WCcrEE
z1Jqh3&$1gfzsh>_<6~Bzd8uMoOPpZo%jV7RhlQyrud0#r@$pIaJ@{e@3I<6)l^Oaa
zmna4v<jPety}mS#f|y%1p?XB5+spV>Zp*x6>YbzrbDX^ms5w%)kXbTM7pk_ft2|}`
zl1Y9{y^m*QbHHN4Z`4VY2Z$P`o054C_!W-?3M`5z{Q|wwPuqfXH9nT(^gQ!eoSMg6
z!=OV!PEawD2fGQ<WD2iBdT0Q)v4sU&`NKl>i&)Nl)YDCh&{MSdX5Ew7?7EL&ma|lT
zAT|3rr)h=w={0eXrhXJuYA{LhSOrFqm=_M%?##c%lG9xc_a`sxVjduxZeg$9I2_<)
z?kE}WJei!py%aD_*OmyVJt<e#xU8yA+*!`=!7AUsQ3aU_sq3kaj-uJY`0}*btU5Z!
z`@a>hLyX_Aq<u*4!ICgF37(P8G0(>b1G0e=_RG$tM#Z*R{z3tD2iK(*o=;M2<>X2k
z8mc7|owEFM<iky5htpoc^kqZ*GXCzR21TY<4={*_#3&ym+6I}GyAGWnfC?$qv=G=x
zzy1BN?$nng#lEH^TeG=xyc?sTpGF%(o6^c)Qj|oDou?py8D}rVS}BA1`pDuBeYNp1
zQra115+(vOTTH#=j!dH*!03`sLpuXQ9t1}ypYN;<RTe(5Nv{MGz&?UT_A|_V>aQWa
zRxjYjySy8NpVBf5)|5O~rl_bn4W+$!@aVW6-v0(>V{Fo(Jq}V%Zd!K~Svh4@e@#(1
ztPx0?bM_jTl`Su-7qipfdb>HqEGauS(Qs+TXXMH9<maRp)%FVP_!p&L-Q5HOV#}{8
zv5kn94iEJ&{QL>&(G0ak$)Wg4w;3zJ_?y&lVQ|tP;p*lo=xA<~+EmDgCx%_1(o)F*
zYH*h)c(D?~^xet8-Al6-Tk~PghxnF!wYSezOHp`opmJNd!|=m%KfX9;qK%k3>81U2
z+Bxf1iCPBN!4t_c8YKGK2DvOcW=1@=JNf;u=H)I4_l*b#k3h)deg%kFphs%<mH@vy
ztJGO2%9u2^?mKDcx@PwYR9}2Zom+@g^`&AXSvZPbv>veO_ONOb>~_^|WEjL6l|lqD
zmfcK&{@BYsND#x+kL<s#q$w1g>g1QMUKnoPwF~@?N$QKt;gQ!(<W-6u>8`hX3f&*P
za7DX9V0+DV&(5F+t5DuGbAOc(lE}Pjg#3vSplU#s?HsMIb`(eT7iDKCRW~7;)-HN=
zb9KvtijMc6mZv&Qm6fl?Kl(_d+YoZ$wbP+)5J4dg5M2G1%CW*)Ek*c{%>ZpT^td=9
z;w%WU6qofnw&D-(WDc`jRFl~t_Y#h76pnvze1uA8jlanVt(V!D&HZla)X3saki&HB
zP^VyuzH2Od><;y6_p>K$G3$jY_CA$wFQhe{-f&>s8<?X%x#44qvT$fT{58B`<nh`S
zaXQDyVIfVc7Oo#)3`Sa-2dG`Y=f$Jpr!Xf-JY1e31m+*Ds>=zG&46m-V}MC3r_2Jj
zs4giIW>Mc~L7+(txkfd^UkGj0(pap2<hQQ9-$g2Hw{ZOZ+K_AI{BsmD|C@W|#MRM!
zb~561!hR7~)w@jXX>zWT{?w*^zb;2Dfi|mqoRH3wahN%LcX2`L?lwP%OOIy$Lq!Oi
zpPhWgX(Zd}0(hpDS0Ob!z(<!%-@-6&<yo|PA*(Cx*SB9<HtfFXdWuK_!SGf$Q-&9W
zUecy3nVoG|!sqyot=yi~oZ-NlrK1Xv<c+88V8IY30~wInk%S6sx5M|gGzU5KYW)_?
zDu3o1jFe|d|08?yh>M<QH3UZrySniHsyaCHP^{Zg{%agZ5y(m`nfCH{_)`i<zE>q{
zia{9ucj+CZUZjk)Rx=xJuVBpn>*~PhZ#ePIqH^a@=hG)_=qHp)WiEoUQW30I2TBuW
zT248pJ*n}bP>dcfH~pXYa4kJJ<*laUT)uMB{X2K=B(KR=J;_~OB)33oI!fr;N89}?
zsv!15>T$C`7sTxp9e1}UgfrFVP0k9jPk%T}k2dQC;Q=YRlw|ZZSR^SWI7RB{;#tRF
zP9mNT*!Dxw4%bh%pr*Zj#rr8J@UfU=^9F^ME80+QV_4(y8>2yuWw4w?Y9r=dJ0k*6
zSYl&@w9=Hm?<FHU8I)I4NJ|FSrUBL7d@>q*r-arXR3ZPEPRk@YF1UQxpbym2k&*Cz
zcWcanm%|Daq#F>s{ndj0_f8?Bhf{W;g3ymLJOlT8({gijUnXyY(%I9)f;oUuf66gv
z`nc34Og}vKY|I2?sr~>XdCcrT!*(W4z%b+kBuF_XvvZ7pI{EEKP&g|pu=Q|hyd{ZN
z?+bH<L)5T=r|Y_>tBQk<u3TR_RyFjwe7E9OdwG`6Dc5q0q%;8rl%$pQq4ol(d{Vbf
zaP4Se|6?ggIl2(sneWpO%H*kHzPr6)-aklDD}Yf3+cS|jiaRZC62Jod7|_ul9(b6W
z+HYPu_i;jaZ-exw-3lnNdjTNMW^HZ#vNbOXuwIBRsKJ1<+^C77e=zkXWRLh<9pJBN
zA_6txL)JX<Gnf~+e<doizGidwGXn-mzVA`}A{U`2_#0&oxZejJog_Yi)3CBH?&+4B
z``bJaqX?>EemJ@kjk_7_hoBQcoAtw&<b<sZ-0uE<<St|T;2_;FPB#FKtM{buuAb|K
zRTASIB*1S-SxKURlo=K!bn=yGjO7GOt3o%<tQYG23oz(#FTy``p23}qfz<=v!Uh$!
zu#1ziiAhS?;*g%;d5mWBg%z0OtAr->SiKyF5eeS`MiA^Bm=yZ^FAQ_wFD8LOtFRdm
zb)_9B$fK0J*!#R|4Nm{Y6|oFkC3ig`9^l1BhFvP*6j-6<uATJ15HA5FGW|{-%RF$F
z2XOlRpFQ(7%JOa<_N+j)4~HBt1*g&ZSc=TRk8ux6do=xze}usmY;nUkg$_O>4#1Bu
z0EqqZNCG=#K>Metq+?<44%v{?Ncrb~V+At79y)if9nOt5jUQtYRuW_kvi6~ee@WpH
zU7<htFI4`<!{2ZL-UH76!h7(E1)$mB2%e+*ke)6Jj1oiuy)PV1{U6LyeQ*sSEB9%z
z7sg<RFzwE0(Tumo8tL_r3R^IogoWeple&KdWJkd|$_G27C)E?)zGbwwwN=t`jwrx6
zf}a2ssion^7-_$cgLDH8QqE_+a5GZ=7YenIe|7NN3EWlSKD=N6xQ;JS3R^sjm#(>N
zueX2sB5yyI9k_{Ay!EAkRiM7`MeXmUkw3sYiK!mb;?#ZVg#3Ii$O&l1l!Sv8F@Tm=
z3iIDjA8;Rqpv9xe_&24vA)kT^Cwh?T^BX`gMD5-0H7KXth8J@eAOe$MxzQ@w=j1fM
zP+`7i&_lFVHL6rykNErfyVoHSgS{4i{`C~(P5G}LIY<udRNC)!KCrqT0@#^kmjYH=
z{kiCWjVL(w;fDMO9sOA<#zBcQ-oXb3Rj_w`!Oq*ie~f)s5M+6OOi2A7JP%6X>Qv?O
z|Di_0^H%^lUQ3?F;R5c%|KD-^B#VSPPf!OTjf{){OT0YCsr&IcX@txM88!Rx-dH2(
z=|fZNcPg}!WcnUty@!dM=>~ov#hELkatyou+!UfZJDV%IK+UR|H2dJ7y32pXwy$k&
z7+-i-X8h;8Q*EihyE_4AOr$3`t98F755KeY>063%d6nCr&P%#;uH`ZQmWAWC*u%#6
z$Qu%>^P1ZuDfN{aO(gcNR3l|~)|GcS-?<tDci&)dHPv^VSKcn4GkVg-ZLdnBr?IuB
zs<o-K>_Wc5ABL~=S$OC2wCncId0>(%d_5;4A<3{C!M2MSD6-Up%pv*CR3<S*zAJuR
zcK_$k!EY!ZN`@dnb7nS0Dp&KZI?<4vsdSdzGoWKhcSj%cDoq}8dq+cvhAF^AL5Sd;
zbfkP|Ht^SP2x?vkiT7CNQ5E16+HIlTTjhqVBD*h%5*srrw1Od3sRt#3Yl&XUS^D1z
z>Nk$D$~9%x=7#bPD#fE|{SMJqzL#p?Hv86|@HVBO2Qh}Pv|T($f4n82_80BYD>|W>
z9|H2*IW%M{mM2GkytJ6l1`hn;fqv@ii<9ghPPhvDS~7-3E87L{*-u`aZ`ta<!2#1z
z7@{>n+Cb$xhQ!E$hbpc8N4GspS1?H;Hvx@)p=Z2`lLau-`R^Y&nugrCz4OW8?qL`~
zm(K#<)8POaoYXxFh?w#q_N56afCOa5>>mBSnFH>-fZ=SC<;4C;Wc!Q!zj_9_i4S};
zB7hvm!YU{qp1>U8{|1}E!REi`OwNRW^?E?oKB%8;ObZ<5hUxx=7;q)4yrTVAw(4X7
zU_T4Xg%z0u1_U5jf~gd9tWYq4xSOed9@ec4Uq4_-2|!aCdxn#ucTWKC&v9B>+AssG
zL;QgIb;wP7mbe%9Bix7fG{CbmuXDp3&WEMFr{`2-Ca5r?q!MYv2oLxB(DUbxp1%1H
z&ufPwl@=UbFcZeHFTz9-!1E|{+X??gpFw~=4x!m`|DX>81%N^d;j`QDc789nd=cPk
zwBJAOL_9<aP`4$;sISBERNUE+2|R#R|8}+hUW`=S18kf5W5NmC_2NF10}Wv1qMjg3
z1yzWvSFgU_euzR{&3c}JJ*NruPEqo_p84}~2M#bFlDKvA{=Xjn1ZD7kEfcdpE(;4l
z@n_)ur{5$T!@U;wp&ZER6-hmq3V%Z!c^ZK23n>8#UM^8OAvc%vr8dxQWK?IaW4&jW
zqQKIh&t+8NR>XdA_X8(0xS|s!^?Oas4gCEGPyyA{)I>HqdwbRD=zQSIIL!pEfLERC
z-@85!6qv9BZ<zmpFm{vw{WJDt`*GlIkOH~XCAfn%5jeoLhh@8&u)>@W#+d|TfGT3X
zMZ2)lw0QzD5XOj}FGxIx1Bu97z&4|3@#_9@v64<I09GqUM+2)zzo8WcdE{padH=B-
z+>r6$oz9EjV4wS?ILNs9KYnEm*)Q^s!!;U#`+Gq465LKIckku7=04UG;c3tm?$NE~
zcooWr(*>rlE->1UL9yg-z*i?3SWTzmFZs_!M6rOaYwT4ARG1HxJOV<|^A5dpUN@gn
zOB99Bg16jr`=#*}HC(y_3`;j8*135zZvrH}5kVd}kSRBB^%L~R;sjayLqwn@(;k&9
z-&EqY2hUl=xA1d=aSvlYU@7X_`t351`}hNlZaoTyIH%v|u$lh)sQ#?yOIDgQz}5L=
zEKK?1r{lsJCFpCEB_IkfBA_>u*CCrWs|Sbc+M|)Crlx?2{Y2%$<nI0gQl4XqEDw7H
z=I?^Z%F1D7y}<wSG39;SX~^+;1B|cvNe4AV@c?sZ+53UtFli-Fc=ySZ$=9A5vMTpD
zV2kvBwfE)GQ2*cmH4##lBvMgI(h#!CGOe;?DPuQ7lE{`VS(|L7M9R{;Ebj`Lv1Tj#
z5|Sc2*+N6Mm?=xv-+jsG_p9^yo$tTjb3UEZIn!z8wLG8qd6&n1+<Vtf54Ke5qIHPI
z?s1RL1O~OJ4Uu=g<&UQI-Ib_&p@xgd3*lAFcj@>H5U4SZ6M^wRHZ<rMO8<EowlmL=
zqx|fg)ng=`p;iMr-?9I;;Pm{%vfjDWUSik04ft7o_f>e4o6TwOxTd4RPJ@b&Je^P{
zErKsDJa+^qFAL<;cTNrxNCpO2>#4Rr(;Zgm@0b2MTh_qOQM1q;x!mMAQN#tUa_~a0
zD_D-|4$eQmt(rppQO+!?+kJvm*z-4BA_8)3ef!P2P73tp(Hj?c9(+$E7jMt)cI$jM
z2&kX3i{b;)%d)@2N&)zx5YrpXV{seZsSW!F6f;I*Y#h3HCju;K=~0@tV0?~^muC)d
z)w9u;A$*cZ>jAXa-qORE0!@docEMKWTP}IbkLOh+o_)MA>xyOE{e3EdRDlq2A(;#Q
zPi*J~RV2jg#B;@~Vq#+CMjoX-r1#({pbRG^dTn|*U<ZzAanaPf=x=8abdJ`RJR6v>
z60dU;W@wz_zDD7_W#INMvgG{jt8N#hKSow@bGaU}ub1GZs6IFlz{GLD--$X5%#1XW
z3RG1vr9;P3Y{w^wmqh95%*hW9J`&BickL<^UKs@J?sHfgbWrQeg<eg(wFZy)2mk>M
zB}sTRCSevTjq3db=2G5NH8XQJC9^if@BA9K#V@h;$2~}g)VRz`75floW!$|U!PbQd
ztWPSqC}bpMH$F$6{5cfIsvaqvEY#plfeM)|X#KR5PuiL{v8mZ>l0o|b>;#F7b6anf
zNe;AIgl?e--6Rjsji$4LJ}`}zx}tX-+cJ;fWYDx$FX!-KXswykncru`9+x<qAnKj$
z@(Ma(k;;6(2dVd5W!#T;tQE=gFZh;EIyanc-Bb=wf=7fJlS{Wm#xDtE=jG}1c*B7R
zgw#^bOWVM81Y4$sj<`j`7Q0>dl$T~Fk=}$IBmS*h_h@fjInBz8P0Cp7C(d<p=0h>p
z(u|_2>47gIO9YVFGg`(0m%t#t!-Z1G6?Al>P-2BBqg2-mv*-JuC9H_4lPoBRce%0>
z5b+YpeYke;{BMayp*QCh4zQomG($kVt0uhlP`Pm?%}I)VkJuQBoN3|{YSDG_2F8uP
ze}YVPzc)e2<p-tzQix3pGb|rF;3c7@wnOt2q>+-`SgVbSJY0t>+A0$5N&P#vkarG6
z%2X}KDnQz2Z!1TV`}rxPGhl*#p_bm0d?ib<o`pq333OdMWY1@d8vXJ>i^Xp-d=3Hd
z_?Bm@DU&qwvZex84TGf5fK;0a2eb}*ozDVuvAH4uC(LowI2UuR=XxKswY1OkG;=*`
zQt9p1b+y_Nnnhs+vis-C4rIV=@u*kx@dswQdA2={`G1c(W-k~qC${O<*49QtESyeX
zci5kDBxNJdLF&qx^}inq<yYnaOD0@wX-rVBeXKU&ch(Gxz3`Vy@^KHSn~i1C&pt&z
z4^DSq9Ho+LEAm}Oj~9-K9mvSXdCY<u)su1n;Yu~irHLR-WwM=l0J`qMGxqx6V6Ew7
zD;@|f6)V+hh%g_!xb39_bT+o^yt_l<_)nZ+M8w}3O9*%bHcn#Goem}D4rNVR%BFII
zX$9YmerGU$-)T*&0$+W)am(j{51*{+V|OooW?hQ!RUBB%>?a41RZ%42zCWXoGl)xT
zwC}u10c1aL?tKC0YGC;Vp(H~bo-MIn(V~&0P^r!^TI7{k82T5|Uh2qvm;2iWd7Kjp
z$qOdDga9%iZh2uSf*gD40(B&_nwMlNKlk-n1;EVc+P~kDrk8D-pn`5!eox9GXuK_m
zEy^o`A+%coS>_9PRiv<zD0gc&C!d=5?R@;&dO|bgDa3OA9WF5k=d%y&BQs1lzdB4U
zb64$yexaQG-l>CiF*~b5c5>l7h7%@1HY0PRu5?{`Y*%h*6~j5=O*Fr*{P%S)fg>+<
zLb1%E?unel6OWPPzN5WAH3d{TALvn9nj|3u=m^#(Xgs$wE$KrmEo@nn$8+ET0^ql~
z@EGOv7{G>oU_Msx{!q?0S!w->=`4W#Tj8b-UKIGlUklR%!OK2<>Bjg!M=aHbHBykq
z$}F`}qM}6*xRvL2+8fJv9*Vsx1w-T4w-ROl_&Fy2Cs>{jp~s$zM2ndpAGS?#KOIRU
zU#y8~7z5ly;iR~qH$X!o-It56G^vD8ua1qF$6_Ms`X8JfsMz<e2}$y1K|b+qeCb$6
z)(I~IjqAZPU6Sm3!b}OXVM)LnP-EzC+4)PI@U^y1O?+#=^D2$;e66T&*y4zwA-hLP
z=$|{b9`$G|_x<En8{t2n{KjAK+?s{_pCfwzyMJ<;Gp$-)AqC}7NIC5eg&Y-7U7?n~
zWv8_N4v9<tk0h)jaUe|QZ*CEWbdG{F0Iz_($73|L@d~g7%pcq3G94YyJ}>(<-me@=
zRD*KjSWd^kMJX>1YgK@i;yj@Fgu56TTW$_6kLE?x=l;WH!Ur^o$5C6j&h*bX+CHcB
zPHM}oBiN%t=8AW*KCBk+Zv0uB{P-y+UEsk<IHms3g<Vbo&a`){q*dJ$$3|7b0MOa5
z>z&9UP|3;yxwUJrpnrPB{6q;JXF8?i@#E{{Z7nbUg(`=Is`zjVY=j{M;#!0iB8*dw
zz}}@pNb>pBWI6M^FutH6I*l_^BVG<wb-}r7cSNZ+ec+s@EN|~DpUt*0Ewj%n&t)7w
zBrZabnymjvagKdO#ZR^2@jSAN+eGZ_t*x0<c3nLw+vm?QCtM7zLmZ)jW`JOv*GuFX
z!QfcQet$1f@j^#aQuwjEVA%6gB9>H+^pp*RK9dEy;}{NuDwpsEZ7wJQ;lAUSmnBHJ
zv}yActWE)?_dTP$n+`qCl((G6ptnZo!YE$z)Z*u_*QhAi0OsDg);wakK4Q;hp<}ap
zI(V7FZ)!IJLbxMR6pyn;vJ%ls-X85>)zPCiR>jMU6G1_IKCfPSAW;i^av#<p;pcPk
z%nt;$SisxaOgU01TbVKq)QAco)#5v<!Pl%|ghHJI3(zptdARYjQ;0?Ny)7!cU3IT#
z_Fl8d$Me%OZoYFPQ&#vr%ke!rg#A9O)f;y#-wr%wNjJxN`O6I1eX=mr+RAKdp8b40
zju0&i#b$R7;V%yqrTR?F_=tMVR`7kx42Gn-gYSH2g#Y;kDy-X3V01Jw1~C(jQDo|z
zdre<?0P_h5K(1ql6%S-Uuke@ecI>@snc_S>)`2Ah&qU0fU=`;rZ35Zb1pljTMjM?r
zMv8Z1>l6rsOx$-?@*{qnh}<dUQk2^n+e1q~)A9WDmdky?XyNhsT0O5x${2PXp_<RP
zFuX9hJ7nhH5s~Yhc@wa{eb;=3uJzkm2XB}IsT3B~D_RDUNwRRYXbsXwxIiHN0>Qc|
zF+lm}17)F@5NuB$_!>g>z#sxLA*I+EX(U_$u5FLVb*YawA;W}VAuUMoR9&zzX;Fhg
zt*T4B_r@ibcCrh&ZvAQ+f4?_sK<AEFqCkenvRB6u3?Z5#qD6WC!kuGkDe(lWV)hTo
za;MksYL?^M+A5bNBa)C(FLNM0lPm2fDCf$bg(~I0RKmG&$CaA#W6UN|;L|d3nK&N1
z)ixHD<i4CQ+p|%AN7Dw*^RXBUy1SKCuZR(teRSaqin}_k7G(I<M)rTq!#pb%t$w!M
zQbXNrYH>%F>pGNZ@$7{Jm%=El7mu0Z`8GkeDgu{-LYpf>u_-tl=*J@e!sc~SzzV6$
z1EEnN32-E-Xl8~2?vk-V_GO*nSSPY#5&*#R97keR4P!4HC&ku8r>Us!K!tFLZQqHI
zE5q5&%z^wrpv0ZL*nA9rjA#dAorlON%&33TSj1@iDO+ePrwwNgJ_s|SgQAaKvlkU8
zNaux6*3vRXP-!}AG}~pk@m>G>M-mml)<M@6eb9I!D?XM{!4mO7OM!CIvIj(a0|!E`
zYhk(iBo2I$y-M-sVh_X>xuxdW&frQ7G<tMH=y>*q{g~)$KRr;we*c^xp@$xk8Qz?O
z79P42Z*tjQ;P=GZaXPdsMGr$gaq5`#mDZGxQvY(wyb;_MnMCNis~G7G0l#`d@z24>
z<U>GW5<5eXa~|}!2vy8=S<XMEr}ODUG;eUMH`9JWP>)v02T&Dz!<ylidIP?){Da(d
zJ^~-#PSjEkp4!p0)u3cf<I$CyrQoWW=Hv3bzNSTSRJl^v%P82`M6Ebn?D$O4=p!S5
zp0-f1^UdWxXdPhquq~U#g2&$u0clW*xd!VaZF;(`>|)zF%ypRv=Im#vu-q9GWnpt?
zHq>p#A2aZR&(ByGq0sk3Bh)sc9h=U|V7tHDz$TK2#Oc9Nijd!wrYXScagaIJLXmI^
zWJVYEXkc$dCH(t=r|HD^EE&RckgVtXTTaANR^&TXYH25*gvuTuhWdhhUk#Q_?B0M*
z){6AlD9kpY-ZIsrOBo}k$>2jd)9@w5#2%z{Kx8II+t>{l4Bni1Ay8MwCYBhqzGE+A
z^IkLIR>E^e!jnzhFsC8Ko~#)kJ5ih2bczSo==FTG)w2^6rp@|zZ&$1Z?DiWmh>KMT
zIUQd-;%vFmX(Bu0PG9E7#J@&Wxi~NGj=>m~8VIYq3u-H@A1HAbZyOk6YM%F&n@(9-
z1<yG0`#jkm1Mh~Xr?6T(6$Ya%JPO5K?4(~y3$<{awmYJMB8iH0GHwpNA+D%Uckg(-
z@_LI*<-MIuU-V07>m6*Tg!xb#eip-+&*$Q%-udcE{i$3++iz*9%nr(w7)A)(lL@+~
zB|U0w18H=l?v4(_oUXRxb{!CUPh-1)wHJ7-G@5tg#h^XfK{qlD7{dAKoDP{S3`u6>
zM=%ozKKOevha00KOW##iE4<cBff9yh&%|?u|IXcFpNf0KmEN}qlSM)^KtrkdSx`qH
zaQc=L`)S_54tP*69egEnEUQCyw)Gy)<#-!|*;oC;EaUDlLi@u|K}As$ZT5$8t{+fF
z-*2pY%CvcK-bwNha<@aDUVKd|X)|#?>kjIAPH1Xb8V|u0WW1PMBZ^>lNI9$TY{kc^
zloyCydg9WkT67@&LS>F$<`pWbWJ6ozZomb73)(dy4IIML<QH%xkpXR`jH=nheW%8Q
z7y_>0-DTlZ>c)5Y#o8xo8<LQ{ARXNIRCTc{-}TC^(4|74rjm>tA!Z9eRGy8g7ITi(
z*`np8c}33AtEP4xpYlDrYSd74JcqQ8vRtXQ#xtag7#r*3=Rcjkfyu#)&nEB2w<M6F
zY-(Fuv8JE4LU|C!i``i{7WwJZt&ypzB&faLhlwvJ)h-dO=DAXBJ}x`%(~M6|hO4xt
zt0h`!qR_jqzO_X6@6e_IVeimIa9RH~$}l^2D~hF60wRebm$ME(-=~73k{8>pW@%HK
zo15EOlzx+rQSx+YwEbO@5z}VfQiEMDqN#M$eXys3ksjf8g#E>$Dg^!zYo-)i-$4$C
z#f6{WD+mm{2Kq(bitLUsm1G6T6D?VAN=S2O>g57AYlXYnIwivqnh9Y-ASYVyVQ`ws
z>z5^_hdK))W8|DK<()^P=|8Z&&BUCukQd^Izp7Jg-2Z^-mQ?oSc>D4OC64^n+7m|?
z4*FIR&Zj^I^$2R~NVU)WxXhP-kaz%W7%$FyszuFI9PI_n*+9drOW4+yqgmjD>2Eav
zrd8^-JmQy*CvwgsJzB?5*SDUzpdmPJt|2HwR6_Z2h;3&Wx+IFa&L+0>o$ZwOTOy{h
zH|p65C<|H)4TGs#CkleFdz8v1gVAXK_QA9d%|2*H-xwdRpGg7bbuEFFl0rlrr<Yxw
z8h}tjGh~vj{go@E_h4rJ^J#Wrl<c2(8Z*?X%BItkSE_z(UAaZ3OWSp@sq3|K1sC*c
z5wmW5er$%afo~QwtYkpm1QH!fwN0NuJWR{Z7dp?%*E8(WUO{A>>F*8?&K;E}yo9qD
zIjI@%j;ivdTu7@PX1n}mhTNX@^e*h1i`PBphG1|e6kt5ahKOQY0Id={ah0C<EtbJG
z>1k2ND(PB*GTVG50<o*F0CGLliSWF{u2a9PkhA_ln`y|j^)@g)q`zDF(*GK&l0$6i
z4wti{eesHIj^(t}C}7b0u3Y{4@-eT4A?OZa(h?Qz0?}g<v9%?O&4C|3KU+5ts;r~{
zlLySfW!$Tjx)+z|4)!{zX!;Y0_|7W!`hf*#wQZcN(Zv?$;ohvHf{n-oD%f}AC95&a
zj3?JyGz{IqT&2Ie++T&O55mVc^hQ~P1|!}{;BwaE+I@!&Up{wwHs!>CjB8=TZ-{XN
zf~Y-H3KBzT!AOukpf3tz^^4uyp`p!Y#3i=w1{jzN8uD+KKYA1$q(FXnz8uJJ`YX#T
zPj`3N<;^iPBljJp4Rpb1vjr}hX%D3ZRLBx2P+fn-3s{KjHsPg!FxtD9+*6>w6@vgd
z!=2Kf0Yut!PY)>#XN)$n6iD-5ibtl)q><ip><QyvZy|rG?Bm1Z;ARJ~l+<yVl+KFI
zGGJzc&YH=Bt5f&I8(f_&ohs733?2g^GJy$x6lyuhcff@AgeO8tHV4eB``%7b^a3=C
z=jiV+giy3m#7raUsFIQXF3|Y0nXTthIUo(Gw=7XL%$wUl8mJ1P!;yHL{S;|5H9YhI
z50j+Eo*q{8(jEk(Q=6#@j@pF8{2DRxIp)>6P=6W1A(pt)igXF8H{rBAO%Nt{F>x>d
z3{L7{d)o<Wm->iCCs_+V+0NvY-qOv}T&wE&9R+U^g#wA$&~>yrli$!7LWyeJ%||2M
ztzij_+i;M9ndDo)?)Ozz(GM4dw+1gu<E^csLI;sznm%RJy5Cn>eYjzSxp&Pg`!BwR
zHbT`Y_oZ9+`z=;4*9du@h;Or(cmEl&$4>&*5U|X+ZtcFb56^-5iSDK}{6#s!^ub$h
zT4rCjhDFdmoXIPF;Uc58%H4xog>OZo!U|wKHaob2it&t=ceN=Yd!QZ5=(+q$U=WU|
z9S$`6tNviL2inM}x1oWZMR{RTsrF4s^)6?P*}1lo9tXKqwjp5wl6qv(L=PG}|Emtd
zk5z!_3ySum`A#58rd$k?%gA*T_BR1wq&|-ruElR9kYURIlmkrsAI4FuIa-VoJ@EAF
zUyX3Cu8CPYeJhkfYC?ZZZB5)3&keg)scmml0hM;zX|#KdOCTa<-}OoiqG$U6sJiU1
zGxbIHKhleXDGTMG8hY>z_MIwsx!I>7HDf=DzfO)|6BPx}<<~mx&+ytv+f#Z@pp9-Q
zM+TRC>8ak|1VJSiq9%{ENP*3t9LuVA>zui@?;az<WzL(@Tiw`74yd_6)Z?l+5#G=t
zQJo{?e*%$qBez?U`Dw3-P+^g?L;Gqzo&Br6Ku73KAB2R|Y1B=$2ipi4H7NtRdD_1>
z9m)WAKf|o-CX(a{Y$54$=^o2lO_TRQMutbzu9}t#xoBY4Ld5o=@jumu14kj7XO|a?
zbcI1EyxpzFv}`&z@djP1HvKrQ@b=)X(zcSNDMth|2arT}O18&;5iAwso$Vr<V9HC{
zr%N<oH_RX2WNYvUO+YuF`>eJ6Lkpp)lW>&8uQ1OlrJxErz>sW~4SM;ODST~3zD4v$
zj~*@HM5ZrQgbS#bL%>^LmqkmThdqQ#o(msY<7giLj7$rO2H4KvwUZ@z>a!nfAavr$
zNc&_DslLVzEz_A$rW6>XP?CCOxomr|j6(^8AQdn&iXAY@qwO|hv~&Cd7;L1&2Lfn2
z9l{=sF!T%EuOc+qmh~4<=-7_twjYq*2y*XDPHT!5w<vdQGvxv<a6y`;?Gntfb9@C}
z2a-YDbq|Jjycxl+ay>BdqHxfQU2JIrx!JE&MTvUfvFG{_b~m`BXZya#EqfuPqLw-8
z@xmt)FagcP9Woa0NNN)lUM$0I7wwk(=i8AP^vKk<APAhKlMl7f5`KR@Jm)O5Q{hqi
z@h>ZO_ZQ!#A>VR=D3fzJX@!o%2~m-Ud(ooRGw!FGVT@RlEB5Jk$mgBOR*Ikr(g^T5
zi27=BG?Cmue>#K*oBz-ea=loUl?hOja8$6Vo=C+GlBUVAN9-ChM1#}{Xw$DG{uf}B
zphfRDv;Mlyp6@95zRGNsOtDT2($6L8n={r2Wd><#1JEfg=Qj;3W-S52lJ#SMIWb0r
zrWBo!(&q=5k&VC3eHbf2qE7Z%=ZA&~IC$?muoNZTdZl?=tkp7%Dm;P=HX<2WtqGwZ
zLFf9&-D29ssC!nxDX4)_hLVfI#u%iLn)$*kr4OdikW%hQAb=3euA03=blpyqh3g!S
z8})kliORpuBxaBvFyVs9iRGmU>>?s~!;NKBz*`1DwAG_ns?-^2v?YQ9m<zI8{@f}f
z@1ovk=(ctdDg1~VFPR%5RRGL8FKj{MH|}8t^?v=<Q>E4CeZB3~k4E(hoacCpMQ<H!
z*W_d9x1(sE6~1c3!Vnrs;VE6;U(ZVDVtnI?^B%u~AI(%wdg7!OI+{wOWqm;^nL~A2
z1Xs6%3u)!ch(IEt-g4k(9`CZJDTi&x;4m=iw5qIr^#N)q&ezACS|jLb>uIt)1QPZC
zc?J{4g)u8-=3dxo;@lC?yhJiWaA(YXLPN#>D4`_0XDWnZzCOs~&zu0BRKj>uk`^n{
zJ+K^sEk^sN4Gu@j(4>=!&=QBfj;u0h!&C_Y%v^zBF|av{u!+okv!x8j6-J4p+iEMR
zv~(CI7K|Wum#O#a5j-M@_)3o8rD1dC{W&xl5jCtn+P^~|jE-%~w1Lt$E81#&!NS6&
z-)|9JT@7ax`1o6_VqEpG1p^6w%X8_=MmhEn7n@VR)L=)EQ9W!kzaJN=%&vYhvdYj7
zlOsSS5R$=^f*>^?s)`!R0A<!o?F3s=v2sf75ygyXfLl}1zeNzZr^Y9zR*5ukpO-?j
zrCo^WA?ETyMPr0I14(+psqaUV7AuknB<VptUiaQsi?x@4ZV+e(_7=Ximk3pwyF5!O
znIjS1v>r=~n1I>Cj@ob)Po+8pKt7Myk{N3dx3*(~Nc!T`dHkxJy=Q`20sr%wasq%u
zB@@RLP2~-+pQD%|TR-epg-sD`<r;;A=r2t`MNV+knGuH76Rb)=PF)|+Ig)KW3tb$`
zZqj2%J&adSVu<oc($#Ypiq>`x2h<|!R@KGTiyaZ1gc+R&O?-ZmvE%n3#8h(4AYwio
zS65emH9{mo<jq0Med7UWAfJ9}7R}XpODuhnB5hJYqtqC3$xgRhfUpgP6^LK^)|COI
zl)Y^aPSPWVCtF)y&^lCT3w++3a)^cbPd^ZelEJGpKABCc?WNrbuy!>#J;4=!6R=mQ
zEbQYn`$f}AjBGC;{YdQ3_<a!?4+^Sxwf1WKU-uEyK-DWRd&xp$y!_pEl>SQdcMpZS
zTV-r2VXRukZNb>~)Hwn5NEc(P)gB4j8nAU2s!Y>q@7@P73Nun1D03UyDY_ZAz2$^N
z!ne%bAqt84tL}m=QV9DW2;I<7;Z@kE*ssmh)+Hdc1-smi#NKat+1Wpg57eM2?o-+5
z+}>5cZ`=9!gzr)3#S7Bf{$55yHgJu$KYbqZ=YJ7UpgK8aXY(~_OF>(lw>QBH`0r2t
pyJEkO5273XD=`1@yp%38obI|cbCdu6&UNrd^_1qxwBu$s{|^?nMAiTR

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/funnel.png b/docs/2_39/stan-users-guide/img/funnel.png
new file mode 100644
index 0000000000000000000000000000000000000000..f3f1863e3e30d5e1742e552ebe3f93b4bc438e8f
GIT binary patch
literal 99805
zcmeFZcT|&G_bw_(Q!F4=ks_!_M=4Q|rUD8A(xn9z=@0~|301bDAWi9AA@n9Kbg)ni
zO?oc^0YeoCgc{BY%KqH@r+1t?#y#VXd&l>e!pmFMT(ds&nd`+JRmG!+PaQsR;K0#a
zH*aVhI6yvh-~d@Z6$SXsSI6f|;D@u8*RS8Xb^SWW9S1uzOB>Sz2QJe-)gHAQ)S{Pu
zVW*M8D8q4A(a*mwPg~QVjOA2LJ;klp$ay73B({(D?4$(bMFtDyAhsk^4dDZ!*d~$f
zKY0A+)iXya9w7SHYKv}Z`tfGox{3Jq>G+xR^W@rJE1Vy2Yko3Z-E6utdSzO28ajAG
z%Gk0v=A`!Xh^*ZBwzJ`{(-^}NRQx*%yOa-hWnonXLYrDn2)<{)x6XcL^`~l@4_zRr
zu5s1C1BTjPaH`e97+5_{t{+iv?y@dXrmvygTDL?|46U60-YHH-nnfDv5qHA1y@6Xw
z^^DA+pf9|$3^`Pl6jv+z%!E+)jSn(6zW%BWV=3n}3oXofb>E@wt5C2#>f-Kk!B{8U
zoRRO{tvF<vN$K$yBYbOG)h}dvi*cvFE9V>MmEF4UMZO-q+2Ky=%C3&fc;(?ZGOcKA
zoI^iCF+z@Lf0Rb8vSsCT5pkONUHj^}*{DXY%e_wKCkj)Tycc!Wi~3Byg{C)ZT8`Zx
zd#~h?8EqV6uQT=6a8sdvbTcVe@WYz4mPv0)Z<(T;%5?IU4he|JTgGK&U29lwhTK4I
zjGj96&ZTHcXmBHsZAJKpF!rWn<<4(THZdyIw`0$H<(R?EX$ShqD-xInLK25N<{T-Q
z-TGuIC4{CZP7fM&X*?)Nu2uYY;|7;1qriji3?<2wI|-M<K2<4*Up;rJ@vo;Jo9<ua
zRO>ilGwgOHc%tF2mjz5!0&mrkk_~x}YCgZaE2G&@vd#XWa88)ToGv9F(ib~NrTkvv
z{W0>x4z>~uyqz8wR#z(3oU|AvRPQWr%X(aRuHM?ecxtmHnD8tnUzrcbrS|2bsKGoP
z2PK__6Qn~r&DvUK?$}i{Ma6pSe6_W;<F@@u;BbI#gLpoyzU0GtU*0=`ZCede$$N1B
zh&Sl(&L0lBa<NJol60OsI%_wk_hmq1;9W9uX2a(T*ZKD{jnd5>WnaH!65bLrVClx`
zOkJarR)3w8Og-zt>%q5k!GqgFXy>-Sqh2}Z&RL`I>QF)QD;0}XiRbzddB*qYhW(8!
zx~&t%xHVPLh)E_)*mmW1!Bv)g`Z8m-*J+bg-{%gkwH(=DtzvdOTxg|Ij1X8NV-`Jr
zjv}6-w_dC6uPjGKy}`uSsT$&q^?f+5G-ty&OoYFyM1nu}%@>OSYM&a$MMzNzmFv*;
z^ZBU<R|&7_ge4@#Z251WYZMuem$287#~pTz{nBFONillt?X4{59&t2-Zhu~89?v0n
zbm=9dZbd(a7Ey*a%ms0m(y9*u$!Dv5IfK5*jcirRI}abRT`;)exBEeIR>#61oz7#P
zX0_(jz_ahF?o%Rf11ML^tH!63Z%$H#YD(mfZn0vO$A=wJ?XKQ<?KbWklI)OEySb24
zlE$37jP0f?GI|H>U2RzS=nfo^J#g!WoR%xe+%RS7S?y@yF%e;r6q0A5SCtfAQ9U_w
z^Mu?*uF?P>v83D!&l~M-2v3wZz8%k;^*_v;d${q$4ZVwRNuO|Bc*4Q)<Y5oVv(sC7
zc`g;G>rRzUl0%-eZVR18*sd#f+W0DbmgZ=*Hahhor>yq@5>hfM8jgScC-rQ8SJ+yK
z$Zs!#Kfl9B^T_4Q|G4ePgM3KI$cCt8waET(H@Nlae{VhTUorhh3I3}Rd$RbiO8i$P
z{^gASs}lcJiT}T=L{#7}X^2l)ZVh1*m+MlMrSz_W4UvMXnz!wCv!kM6R`X%8tGvhx
z|MsREy;w2`!-iZxmeGp1{`2k&8uxT~V*UB==z)yJ;#|J#(s<EA`D}^4lt5E26fy6(
z+O4am+<!RUjnqEp>PmE*ja8qvu2DFnpkZH8Wh43MW)G^vU}QO>+TYMB1KCl~Bp4fE
z04tkO>vEY0gs8IE7O4-oO~-e3CK7Pf=#tKFhprOVCoBou8!8N21bj`R^RQEIDu&Tf
z=zHa~l+ez4gN)#!uz72o;|3-&O+~yvCByr^r>>}WJt^tI6U_Ko&&4Xds(GBff}Zu-
zTtD^$wb1QBo5B`v-05dO^0KJ2BvM1%SQ~3TpEJ~-cB-}(D(sSqnQ==L1D7fJ498w@
zCp1<mP}<vV0#P+dIims%3m)DYPK|`ZEr++KV(k*3J2Sdzy_Hj9mQ|}gSqA22+hoht
zgPb&@vyS*-m$c*VV+XPGc3ueBy|Uirdbh2KV43PuHo?v}+SV%XA+lMfT;xxW-nhq1
zI^j9Y{<!%fcCN5ZHTdfCP5;p{J9uYohn?q2Ed6^Ix}V8j(oK>pLjhe{jC(&btNZZf
zRbKM=?#q5V7FKS#`H5Rv{h0Vm3vF%d1EltJ*jeLU_?4vXr$X*$R_U_YWyL6+pBva!
z6$^$pKOv{?@RP>9@nIeHp`wZM^l=@Z9rHb&%k~XyjB?`kBzM-z$i8#%KSA?INJ-<$
zUM_p@2pOB_`d7zKw2CS60gOn#+|ma}tg`*(skgx{sa4DsC=Yn;?sT%c45>RVndptk
zQL=`;MR}R}LwB>X3zSTI_((`aZX_)gdQ65Yrru;R-%ph3c{s+PDX9f}7Fiobbpste
z>@q54gqSZj(J<-ezsB(d9a^|vxzft_w)CWWJx^NPD^-~toRvY3v6sg|q>fnS8R0^*
zafN}ee(X?Nt)ahyx-NwZHr&`(VI|oQ<!{f1U-y<BB@NZxOivD)LF_t93RdAd=5-7d
z?@f|dZ)1egr0_$9wo+WIHz#q`_z^@?9)9Stnm@F;hOnB~DbHjSc50<U_i(^Si~8Ee
zbYfzU;o}zcgFKPD9J%4Q)Yvr<@ZB|3O*b;?xfA<FqhiNU*<)+SO2Pr3lV@oxO{IZp
zqyW7~4rQ*hu5Yw?NwYdMo)Jk8kGb~aGmmJ>)_O#32B~<<>fIZubhC_i>UksQEgMv8
zm6g-eJ8d%RfftBmtip4eBZj*Cea26v&F0leSB;Si-JDK9!1UcP;Vinb>~cP(p{~HP
z#Jg>PY1-+V$sv4Ps%DLxES(;_^ku%=cf-53nc=Q;ziCrO_k(+rX!@!Ve7|`jE-dXT
zvY-)c(%yC_*^%&=Zl)G;{ZKg%AJYiE^B%Zlb^*NWD6rSZeA^YHJA$mvefP^|`A@6(
zP#qQ{d@2)n5_$ye6-lRd^&QM_{WO1Zm`w2Ooo1RSAJR~T)V0sKTZ=U^J54p)O-F1p
z&eQ&=kv`IVXRj0PVtu`5phxcBd#xgVYdY?+>J=}SkvBD)-Fc}pyBlq8S^QkWT;!a{
z8@3};t{=EG;-Y~LZ;;yiE(hAVn;2aWmLaS=_8Pf#3yAFduWyR7wN>n5TR&34i}ZJ4
zWf-*Y^(iEq8?IfkN$P7uzFy|-By*p}Vb2eTxVz@@1(8*^7uF<bMaaxbWY4YYBg5O9
zb^?sIAM9!HzHmtJ)!4b4ds|np#?cWH2cOBXT$`J^Z&<z2VwWmISPGu#y&!h}OR_9k
zH=FC2PkGfTUppth78(xW39^zA_-xVFXDmI%vyT(E4E^7!-QIh!_W_>_D$!k(RCvdn
zGWg-*3D>!I4IMeCkpZoV;$Em^UD97wsuwvh^Qh60qQTd|U`4HavQshdlC*e{z!_w~
zjBwYWPTktG5p}nZymCeJq@`HrmYpAVUunUZ$RGZzQLcQco+T}^a`K!&=TlLJp8R`S
zGwpYB%KOtN&P5u&BFxqhvf?(+W-kJJEBbJlFGZ~9E>})*k3P1)&C?ma@HjE{0S8pm
z+NTji7-y=v9B*HL(uqYL9yk#sDJY*Am~q?pvTm|_3jrhK#WUSpvo>I@Ie92J@up_}
zz4uR+%_Vbt<DCb1=0T{j4A<7d25jbe?O1nuE=^v8vf<aO(^FI}!v(*B%gRI-V-&qI
z-6kZ|B-yjvB$SwdOX2Re9c{bq9>hYql5-lBxLn@P{lqeHcYAtPJaDmtXv47nB=d2K
zTB4155>^r;^P+bNi@T~WqaH49!V{NKORT8b+=^7GwLTO5nu!MoFwDJYG;Y@=q3vxy
zEinliMwN_uZsC?AYbq_hG3S&cQV<JeX+iwC!KJDcQb)&EdUI8l^tOOBbi^yc9%^j=
zb${FTUAqEH_`%Aha*Ou2scnW84w5+mqki@af|cJtEHBpA1u&(#zT!%N`Y%SW8`fRv
zcOJBD>Q&z!7Uu2b@H8*7njPKU8BJ63Tz=i%ZJ$e+kRc@V4BvJ-2txD^8yTj0=rtAv
zeiEkf;di}*u8xfH9NC_N9G_edq%2!T^Tp1bSZ#Wu92n~?=G89Ggo%hGzpcOE1>A<%
z{A=>o*^=F*U@t@T6MiJW#b}Xj^0dTkwkDc171_Y<Rhj!zRU}y>xDswA7uP{_qv$7L
ztb^=qdbRJu$p2EZ_e8||SD%MpA@l-oIJ6e;xrwc!?_4cWhfS=~t*9XeI?Dx<?LW#T
zWT{t}6>=D04MsP0EE`x|nw?q-TPddb?YyR<&BZlJ;@e=1Q*0AyGCP(Fo_sk)8VhS~
zPMJo|UFwSAO--bZDqD8>?<tTLD_W`RFhk&_V$rQno@w%wOjH{amLp?>Hp+Xeawj~C
zS?D*6?16`i9LL{Fke}a3uS}G5fnkwC9+p^C&5o^Ycq&es8XZx<ZK$N?wb>bK!`qNk
zS?^-wY6?-#Y}o$dC*4(K%vhA2o45e+-QF&?Qg$}Uas;s{q6Y-a^8x-w^VW>pf~+36
zcI2Fbau%ngV4Fr#nDlCfT6js{gMi#1J;b(CuMr9dRG}nog2`=Iae`t)GMQnjj>&lF
zA!=hPR&8S|r+bN?`v($q{yGO-H)d}ONl5T!Q_gZ9{n+5?j^Ez+IWt(~?Qlzsm<Qj&
zJ5CTy@&Y~Nc)<ukW%mfA1Y#$RP<rDy=0kZ*Eb2>Ng7b>+u~PFYf90&468&7;wMoT_
z%&A-!T-hn4wfO9?K{Zc=3#)Sl|Lhe*|5|HEXVSq1J-hEivy-|to7#|W+cFjCSjBT$
z&!&TBe4N4+*wk3@;%UU_7PMkGriX#n)&2W>N@YrR_eVUiN=HO|7mNvV-*`1sH+<Y>
zRx7nhsHU=rusMr2^rK?-nwA{pVTVVN;N`ps%h|gaV*)Y^iJj>eZZ%(<yoM#J^@_i3
z#hmf(s<CqaYf2B+!ZitWg5d>uKOa~sXYLl-o?=Hb*<q-%w{muOYjoFhBsJn^Tp%ky
z5%N)LE1(mEavycN=Zdw)xVKD5lVGWmIOo*fO(UQm*eh*rsxWjM6n#oZs(@{ihx!Kc
z;IYdb^7>CbQ!j8Wi!N4gEqFw>x7jzCCSG|wl|szh%DP+NO4EXE2A8d-t-kcyq@?<X
zk=3l%2`#&pLm4eOIL~#k6TbW9;!k!x%kx~#daT)fqjfjq_UL*&B*ximXV!T161i??
zy$63IX#h4eD>rlSF*Eidr&ymz;zqNBRqWh(StF0~_|Jq6HND!3xGCg2mEj0l9ve;-
zm6RZ?lfUCRd)w1q+e=jo`h9C33EOoXvHdE+xCDG9x4hA2G-vTJFrXgC=8FxE#Rx;8
zcROeNPQO|xoe0WZvqY(<#QEcNmRLofTXuPx--w-%th7hNo<IaIsr^iOzUj*z!{=t$
zfB@z5{?J3ILd~os^T~THaG!|X_8_k?gPqsx8_B!aL_!9SuDwtWvw{?76pb%ag7bi-
zK-Q@Zi^48S9%V0$R@2?s(e2bYZit<q@$P0XR<8}NQ$F^D+TroU^t=d96X&8Aw0e!0
zH(J*0tj#tG`|rtfh+YDOZ-<<O&91ZvnXHSg*IqW!Y@oSn9Lc&@95NIa#u{^Wfe+w;
z)W#!11R2ew(z7NsZiI*n)v^qV28pRh2M?_XA^Mv-K657;x)d>Maw^{nEuiOiYO08S
zalo$UdNr)`GBhGH1*X?7Vs8skZ7=a0_;ObjwPK&ZW$bQ9os-(=!FEu>EaMt_W5{IN
zzy35f+{5Ih_j*u3<=1y=D9wo6Y<DONWpK_M`?N`6i#Xr(so?D~+7R1Hl2g5!5y#TD
zJq-KpLh?RtZ_DRPDWB11%X_08YjQR8GWQZ);zqWgYx%f8UrfEg^PN$AH9mZ$VzIDI
zI>(<+==h=z<jQ!QsoJBkl{&@Nb>_%3_4l4z3rIiie=S#L+93A5J=jxpXN*#&(a5;#
zgZg$nt!;oeVlC{Nk;udfeXk1!kq<+j{0Sk;a{J63KSZnqC^$Kgw?!U9^<VOT>A*c-
z2O&h)zah3zR>kN{=&mRmU$Ekb!o{uB#JR+IjjOOx%TwX-8&X`9FsQ);y&mhDFmgVJ
zSh|hBxa8%<pIa>u`sog`J}zL~D0PRO8{|Ljs1%)!X5U?JFwC;K(=9HxQ=|Rys&4Y5
z7Bjg#!y0}Is8uBF+zeY;EY~^3k>Z45vJIKZIG28vlA#@a%%ePkx83+hS#8>;qd*Lj
z_^6$riy`Fr7<3c7U*{OaHm~11S*1a0wM!Er4XqhvSn}9@X?`_1<5>7C2Vy1#2MoWO
ze!A2dO&X5OW<p?hcWQQf9Y$|9ye4<DcCp?dj35>TOiB8O3JTbEk&QJ85W>wKks8M@
za+H__UCZkjF^3Oo)1ujJ#z>XW<(!V5`rfLwfvl@RH+y0$3fxx3D&#cKPiD%t>KAwC
zrj(op>mmK#<z-u@zB!R2gf&!TL9A0?h1T@MsZEf@8x1`Ac=U18bMYJSH_n<hy&lNs
z`-Il?t~cG*BhzPCu9kLBMbQ1k_XiF)ejSNuP$DJsr?+7EcG1G>aZ=@S8=|MMC`;}R
z*R=CU?KAb(X=!{}P})TySG|numB%vK^z?%?CZTy!&XsNOLlFfyVb+6F?O&*vZHn3z
zR7`JNxmapgVVU{<FK1I1eaK7it?{~m$MKvEPHC2X(^mAk{h$OB+c1?nSHV_vnWCz>
zLXO^-9ToeG_EE6O)No(G+L{Q)L!oA^IB!)`t9R>DDYM(b*91^@ER?@QKil!tEjrwx
zP%8>!b{45q3EEhhh8Q81SR+?^7E_;hfJiVhw1XOQXwo3-mQ>+RaDBkDIQ3}W1@A~X
z<hQy(9oy)Z5Q&gc?$+MKrO6Dx$g7lwm3PCnM^F-ew8CvHUE(wCO@VpLH4_4MZpsVq
z3d2-wXGa#D*q<KTa#0w3X3A$$yNSM%Ug=!E(Vua9_Rx>){4)d1qr{=dc^EPZ!7sdH
z)gXXIj;xiS7}h%HWhyi`8ZI6eB9P>5zt-7`=~d1z@Ty3<Pxh>wwUHkkkfaV3=5d!)
z0Z9yhntkrqljd>l8Hb!^KVC(2)nsJzGpO(k3|R4Q##lJTRboC;l`2X_=$S>IPh(Cm
z=@gxNEH-Cd(v8=f(7X_T+F-TAwpY`MnFMdqaNm!$$1&%Vi)zJUY2})^+s9b+i&(39
z=~y=muAD#NK>7|W2yg$9i!{1Qb;n(I$gxwUY_xsptVW)c<-``@vzn)aCgQEid_wMX
zX9cwlsv{8nPqkH5alSU$Awz3g==fIeEnyii`zo8_ZXFfR`={hE&{Sqz$+)&#O9xAM
z!S=dJ-9yXvn36yb>7^HHqPByUa&gndv?HT(owXk}GhLn+2ALZ@84fRkw`_R^a8$nd
z*?#&G?Cp(^v06JLn<V6;v&*4N_{=}!6lvs)&GaM+bMb&Y<<La?EHuq)yTQw5lTyMk
z+1$J$!|1!ioz>^E@*`p)7)O}yYQ!uja~K9{%Ufs^y!$rqMMQX$-o<Qh1TGck7|izC
zErW))o9!u4g5n+0qUG{RPoQc1#~op6U#UK!VxBQP3YN7si8Uc)8ShwSWljmUQMlXi
zn(yx5+lEUTpR%vlQpN<aRHbZ1Z*Jr^)-zd1c&>X)a<OC64kPJQFP}kG-n8t3p;v<8
zaUc<&cWyGmFSyi5qK>JLI8qGxZEQ5vlYh2dA)*?F$<EXhJjs=@3C5cpkulCq)Km>E
zG3&>$=IDCY&$fyYZB?fBJBE1mz6;x}7X9hFMzoe;<K;O!dX~uWm<~xN-Uj%-+oj?U
z-%-C$#3+1d=^EjIsr(l@9_J9js3DIBkGefm<$5OiOoDAQ$MvmqwAsSmS;{j}7qC4q
zJLqLbf)`v%h*NnW+g8sr`#xGS-Q2W7U0T<-+>9MSD2EkSR{;ceOAop|;|{TV@TfUC
zNP^d6`E}y$(VYdCPGlrBrr_!*_Nblfbx6f=$5<lBb@k@M3Aw^rtBJP}D_@;EZgGp{
zv}x#x3h=?@81KM%FHu0*K7|d}D(T7QVeedwT96#wXidco^FJpsGr7}U&qGoD(no70
zb<S?rk(h|*_{$j@><}ylO3maL$yD|36qtR!H*#!cB~r@1yQ@){Dgc_sd|;cHO?SfF
zdX*w(xJ}g_cG<?RhNvyEcC>{dozTNM-D8zyEG}h_!o1A2VnXmnG^2G(@izOo4vts2
zqL<ZXFDK~~%I{Q_L{fdiNyvv>u|vZTP75rQ+T2-fgwLfCQ!SO%7qT}ZvPv!8P}Bv`
zLKV4Oer-jxhGYsq>vdFJ^SAj^2QP=$n^m3H6bI5Q*_4=>wz|_hw>g_<ESBN1^8UD|
zu&ElKD0#gV%F;dYsB>Sb#lz_d1=)xbskxrCjSFicPCa_&V#Cdcn#^Uk`khQw8=3<l
zBYj#n(2yL<WKu>$_oX@v3s0atE7&5QS>#Y@Y1YQvt*|g8-7US=#qxQs^aNzvt^Tnn
zrzVs0GqjLrr_zGGwkHAu%432TpJzuR&~O)*Vh?qeqd<x-A2~@!_1Fm2F)p<vlqu`g
zfh_v{>nFQ=>pzS?GrX-gkQs)g=N;AR2ma+0WEL*X@(|)O$lriiSoUNJ+FX38RPn(R
zz?VET29kAn1(1$7hvN4kf*?|Osms{{5@Jpu^kq@TCFe=!4&k6wI{e{AKog0d0@hq=
zRxdCB9y4kdYwLiL0kynCq}uogovM?%P&bTWkh%;}87Dq_+t9S1Pu5hE60-$JCKgCW
zFJ<0ZUuJ<Cy5z4bx|$xy%~`#gW9QUeWamD?qB1D2HqXDK*}43dX$XKoD)LkDA5#Ml
z-=!h>Ue7ad`h8EvZGhV8`g8l7N~;{%ey0M>zq~dItC&9(@C{Dr&hvEY9b_#2nlJ4|
z`>mRoKE;G$Hevcb^G~Lw7TtIflo~0%-R{4aYKa<lR(4Zk(?$n4^)sKt7Uy}`4gge(
zF96|b+bu$9S7e_r%XXH0NtD!Z;n0R<pjc0OvySmfH6)X->dep1dlAi-k5%L^>&U2%
zsNNg2MX4Ybt6c6=x}!A{H16ts<h1~K-|Xy{9JtaN%OiFzsRB8I!=Plh_ZEOjHN=Gt
z%RzZRgAQgZ&vVD{%|@uq+!h}*)$B~3w#l-#>}u;y%I%kx{Fda~L92MWebQyVxHpxE
zT$)}P?hJt9-r}AxRdUP&5LpwhpN!?Wu1}L8LReTN2*$gPzi<+JUUX^EV-YbMA=dhE
z!Zih?O%*VEvUsjB>0!-?NQpYNiLRM?jR|MXcsqAq4s^g6G>iK-{0bEJux7Q#VAlCF
zMPGJoUh~7UHa@$oyxANSbk#$~uE3r{!`=K-V5FSt$MtQ~E6tB%nIZl4rvBWg$DfUX
zbUMxebJI&}Qgo(9#A@)=C}kkeI@yr*)5@PkYB>nPi#lj$GOF9Lvbbxl6os)R%MI_w
zbfmf^i;>-2^E(vRD71rbu5?_MKQP?{it-|yJ1B^D2V6^E#J5qGD&|XVucZFau#e|i
z$V31lY_9@w<b~1pZK8FnH3)ieEV6TL&@M3nz&4RyZx!wNCeVcB%~O`9rw%Se8hg5J
z?ksvGCJ=CGsBDmp?aX)eA;eiGC5n}7+Y6GUb&3wh9OWJJ+?q#0y5X5V3raqrP~^q*
zla`9hsg)Sb97WubAn0<S54v)Ksmz8su|{&CG&$pt{n5bjKmvH29i~gf{-7`%)1Wr=
z=p=5jMtwJ49dT`?p37o&SiK_gZ%b1Eb-DuW3NSoo5}!)1&1dR4a@{6N<8>-60kLF>
zpt^~PV0y=~=qev32{Zl69;y`WEH`_#LS16dz6=|@{X$<~?T{P-if60lg*z1pZCzUV
zdfe-BPRW23W@J&wYon=&WF3I3sq=*7A_o*dMxM-a7l2P)+I;Xbb^ty&HCdT2cYOYI
znc6o>pS*>_sp`h}Li-r2qzVlvRC?uG%kG!!$C1o8N}XeZyb)pSPRx$A9$?LtEZlA;
zMqM##_^pbCaz6Q(qQ*9_Zf#v8Repooh|88^@438f;=&E4D_I`pR7+an0b0EW_i}u>
z5HjkKW6wuiIfw8aAN51kM>E-Ig_$QszfBx1u&p{hd)#aNI$D{X#ur~4Lt)!HyF}sC
zqT~MHOG|)YdHQL`JpS1!u_u==mRcQgE!&*QN<#rZf(7n+(X9849i8R9Wl~JLZ&i1v
zDDa`|O(gFJ(25e_9{6FDb9+%$+idh{$|mZ36~%KDtdD}lMcG7JRT8eJ49Co9b+!0H
zXD@zhGfpth!WmSRAbaPz61U$|vKz7;xJtpi`%O_MMpXiv;vlr9eyBt-7QS<AvHOkv
zR>o<e#Vr5^tM^^~xyV3}A3>B#h~C1-lA%%A`cgdQTo;R*M6i^S=`<!hXj4GHDBV;n
zKFcgd1u`tU1GeRMyRT$NJx)SEui&f)R@qo7)~?#E438L534u1nEzA?EWC#*b4ltFD
zHZQ`;&~->^(u+V&rtM0%6$g|M{66!vN)AO(^Pn?t({lCWaRp^s%`9$>Q<{uY>tnuU
zkiMzrXc3GRy}Ck8IP>Im-qLx6F1I_qcj~w`wCe&hnT1!Wjn}U&^pS<zu0SWA>A9=b
z9mw&Gs>ix0jFj1|4qv2_Z1`E3)a%MBMK4LYSgCr~K0j16(|_eb_vz0wlbD2HM-tF@
zYcJ|@p<kI-RlY7sKVWsQ&Y8qznwMSdNo=0SymgaIXzN!1@xwMp@YRu?z`{{;tHXCS
zQVwkNHAw5U9`XWkxhQ!N3yOWp9cSU$`Z(dW=VoVOk2Y5a`jhv&Voi5Q{c=6elQrKM
z&ZlZ4^zUoebE6k{G<roT*tj0)e+r9yE%^jMC*@W~OG`K8$m&^EzA@3>sRZ!!Y>?Qu
zXgLqUGHC&R)N8jkhDEB^gl39a<jo*YE8^;*{B9kC@PqS0o=X8u2iNMOsO^*O%G2yO
z<3oETtCt>&4nxsDy_wlFZ||+<%fb4flj1|VL(#ed7gi`%LTdOF$41Kw+|5tQxUc3h
zquKNxoyx#S&q>pZDVt7+HPET>u6P`-tb-DU(Gz<dn8a_iEOJ<4d5q<j*|=2>dZ05U
zEwL}OL}G1A^x&XkibTO5M%0aE1d`y10d;4vX2-y2;rqAM5oY=z(et$V>|OM}@F^7l
zi0vA+$|Rt28%=4u?2eQ|rfL`Qu8^Te*E@6DXQlFb-P}D{RzjFUJ<sMHQwcxy<q39s
z8#5edxmr>4ZAyCk`$r7D*8G!sd#<^YZLSIf4ow102bmQ^q$tk8+y~Cg09-qm!KiL%
zc~2+kjsVkkV6i^)L+NG5yvS3+?0G=Nl_WP7s>o(}w`17GL>O71ixU+sW@_B>y9yr-
zylxe^CcrXWJi1H3H%Vo%SeFhv4~8RT#0B%3C3P8IEI>_U>u&DSNZm@FMobzJ@Ls$4
zW$z3(l6IwxhSQI4w`FkhZXKHuqj0x9?I;RY-u><;lbv;KG8`5M#h3RUEu!3wvCi$k
z`*hizY>4tAhTJOPaOre-XQrYT^!s-rUK>U=K4mm<G@OUw?pUilz!qC4l5*cbs+R%<
z>|U{fb~R7^t>^X}gcX_H!nigm4e`1bKhRvInO>a=!pOUvU``u7={9|7QG`zQ{C-X^
z3Ea7}L=P>8a58xW6pkHQV_<#MCiVQO6R8^ylymfZx2~}fDn5FgjM%Q5WMr(07*uK1
zmW*$m1+B&%hlR<~ul69&rg8;c=7fAd?UP6q7r0E16@zY|b;Y%paGh#-f&pVjDLs~P
z6w#{ZgIMKiDoLDHNNob01RhAYA||x7^1fF(=wtjwiHToUs37Q@rkYmqKlizjj9<Nc
zTXSa{<FPA$9AEkPOMv1~3TS<KhI6*uf!0dGdg_us1!|^PzrZYcu$9drq8WvtaTRVO
zOOGLQ4)+uK(+Qq!ThbtUFW(q^`NNT?)6qQ2n-IS4U3>OWbG#%3RTipPmWTzS|2&-Y
z%?tYY_@yao5S|-`7`H*7Z83s<?II(#EK;>N)E|(YC4#=i0z+l!=@;H6dD!%~`1uCV
z+QMPgyd+Rx>`T9fx%i$ZoeNLX6s%q!Uk-daK(PS&MwTCsLb8oSL-1}@Ltycp>d~O(
zlabLUxma1q({j60ftB%wen_RLe5k@%JM`9ey%WH)P#D{h?gA?M<A_sT(~Cr)5vJ*d
zuNcJ$E}Gr%l1gCgH;pvv`0$ikldPS~6)Jha{)j{z9`#9Gk+pkCvO=5nLC8d9M(^Ug
z%1c&}7^BDUJi9z=6f0I>*QiQ+6I2#|f($TnvZ0Fb0#?wBR^`Y;nTTG3$FGTW6CMw>
z5nflSkzF)vgR$vMQp&8q54+0=vB%+Fj*YM(=0e9qx0%BVv=^3nSZ!`QAA7yCHYkKD
zA)>>Q$2hQtNO}mSn%Jpyso5IFo)dDrb;b5NZfybbeH}=)ibqU2E~!%MOc^)qIrRj*
z@8N5e87msM5D@>3BO>G7rlOr@ADDrn*iK*kIk}rpp&a?p#QEzZj-dZHC9nYR@2y@d
z9_r#CX0Wi6OHzI7j=DpqKLe-*<^Aj_Em_cF7fAR3RvCn=oV~Ef69iLf1>!EpsndD~
z$6g>>RF?cU$UM@k^E_5MBCCs>OVaSf9_zwdc&up|-`3sBf;A$Xb*dvR&nj1jDSeGV
z=qE<TLqLZT^XYQS$f&jxWubyScUi97LBw-fJpOzx5?3BOzpH!iDTB^Ux2mym^$p|g
zaV8JR9neLz_J^i~LL@0%u73TCuES4e`)jJ)!7b_9sCnl1@4kheW((Ae^ZjN!Z0V3o
zn7-v5?1@XpY9?-K`F9D^o(_;zsgXlbdDnBi5A5{k%LU~edL?K1irTrx2qxLQl<+GP
zFGXf%PD9QU3zaSbmaf~-F&kc~3Pgs>k%wh7nOWlVC)6M~7F=+x4KyDC8tPbbAy>K1
zcMskVSm(trKPfU^|G?C-gC9+-P^);#o#^i1za&m^rLJZ9z9+#Mbj?zy5**Mx7#4dK
zzVf@@li7FDEwhYVhKu7UG730_6g^WKJ}YZ$^`{M6<0fI<w~|JS;5k7MAVYl*rb!RG
zBC9OSe!=Ds5)JteptVlNoj+I56oHU|h+nnyuBo>@HBAJMwKXq@=8BBQ*Gi5o1q~fN
zF7Yi1_boYiD%Tn%OkvkND|Kq);l$F;tys?50zkS!f>K%lg<bU8-D<)fanKd#kHbA?
zICzE_Sv-7|lxPiu=-I1OX~)pB?_-$$S}dNcpEo$#wOM=cxF~F3)%}R~K%cb)|M(OJ
z6KM?9RdruWolv+oET`W*i75@jDfvc?A4*skw2tH4bx9$|noEGjg99aBV45JOz^yEv
zJhmm#IwD4>Q!yzRyod>QpX4f=O?~_|U&hN0F*q0#!=!d>2l(a3oy)4!oVz<q4eX;u
z{nA@=nkarDR*kvA#lV763Vmc7pTrp3(UX3_QIC>eyx^F!adE%#11DA?+RbZZRLvKB
zhy}aK{EH%*rROdvP2A+mgwE8ALpKip6(~NCo)H4t#GI^dlc6e2g96XJ--C8fY;3R?
z=}|YV^*}hJb(WKcUg{l})O$Vi`0&9n0Gx(*CQs%z>wPms4nYSl23^mSYZ}|>K~;yd
zY!851)d81Qa)I<6ZN5$1GW?4;tkX&wuCfTy0;eo4@7Ufg_!nPfj}9+sdthg{E;Ks!
z71F+fbBgN@3k1%R_RWY;n7ET|Z$G%vcyeS0v7}-L*Btan#8*xui5RfT;5IQaxH7DE
z#G>EQfA$>wzA<Rm$BV~Dth%Y)Ix@4u#G^4TS!nO?jmX2I#|`DjDxLk?%uSS%Y)kr6
zonJ3+xpIDm)NH8i3Ac7M*W|qxth!iY?ecbOqw=A(?P(dWw*;t#6JfvYxJlE<CIkVJ
zy~jb?Z?b}%v(Cg$w*ZQfY*3<02!CU{R${t|c+`EKS^F52vXfrKwYRtzYVL2tkQR*s
z$sIlXLKaSm^$@^pGG3_GtF5u`%$+Kfy#;gAlg5;61uIDvjP#bR9&d`QnlnZ63%8fB
zzPqGccnAGtz6|pa%de#8iH$IQ_#QR1Q!u8{<l(Qsw3_E7h41UX=Z4I7fe$9G+nAEH
zYV^dGIubUfBljXYk!BQDz+u6Ih?7ALp;x0<*M&?c7;!@{?<Z#6H!fV-FgA~Oa<Hwz
zA%&dgH$eGW9!7pY7TphuLW79V)wgy7(yitZ;u#p}Vm=*!d8F>R^xgN<V6?~@&Q8PY
zS%s!vDy+RfT<{$YpIp;|g|alf(>1UyE|+*};LMdd{Smii6pAdelJM+0gSeIyM*c<F
z-pDDYdO19A5-Osu>U{lO=K^LG7^5vu#l|ax$hj#)9_kbnKeuwS7J>C-gEjs&=Bb@t
zy_W8gl@%tZz23W3W%WnSif98g?2)q#)LFB1n>BtqayZIzKnC&PY%iSHg{pJ857At^
zaSaRb5H;A9;??JiYdXXZ#2(0-GoYoHAf?{*a0L)vk_!g@X6tbKFfWY|yQDr;FCWxO
zog!lAke$~IyY32)*5Vi4&3T&I43D0uli7fbx<BS=ItoQo9L^pTB0b;lZROPT_B088
zmm2MOc)MZaGVaUu*du)Jy=@M3X>1>K2eJH9s*@&}Z~;9v5&+K-m)*-Uw9jD>m2**F
za))3&#fM9E2c%i^^x7shBD~6CXD~34W&j6!BfM{fToL-lIL%8sr3ageAnD={_T-rz
z)<<hpg5HOo^t$27-OB)O(J`<MM>JDCH$ONXQSknrvR+J-hT-dBRXw<W2Y_N6nsN*)
z?CC7}G<!Z)X9W4NI^S*Up21geJ|XniB2}EynN6jvntD|x76VI13YYkmi1mxX!d6k5
zR1Q7Fs3N`^baqSl`DO~=Q?P_ViH+=oje4!P4#4Ue*KnuQof*Kc$^gzqd5)6t;L+I?
z-3%K;NY0Uhu3N}b`-IhNc{Whbz?EG<w@E3gV>Fmth+Rz(s#`3m=OK9d2OV&0Xo3$$
z<^|S!@UE!)**qH7=uwt-*{#1cI4nGeFVE{Tjj9phb!tRZft(@MX58-CBceI(^=db1
zfumDwktU<UofFO^!4qpnm0b%l!oirM9XiKSQ8Tp6gNiiLPv`=-MnHKiP^ECjf;pT8
zwDycR`$CXcEgWnd1IM6bxdkK#l-rwxTGc1Si&y$haO&=Z#0FByQ<u9}zIh`ea#P$_
z(ytc}<g!hboeH-WZ@NFcRvs`itGDpI1C*2>G|0XtY&4rk_(tTAuG+VlnOrM!O0=ro
zAyTUbC3K}BODQ*nIzgR1R5B#9gH@}dhX@RMf&|Sra{{CqLo4r6myJS?w%$e0su<V@
z_m)bFkGM`~MEEBcVImA75VgIzcJAzuwgQ}i6pBmPw_~#Xu5-mkx_Q><8(Nyg1xtWg
zD8;^QaE*vb%P}Uio!G83_X5c_Y~$fI=Dc_Hr)+Cp7vh|3gBg7ArRt1o>f4<%JDONx
zhY2e$1Xt%-d=YUO3TCf15E)C`Q>*!TMQ*&nZZ0;1=7{)YRej64f*)WEoU_wx`jOg@
zkQyDLDs`!Q$OU?LG|b)Ji0y;SSfx|D{WGT0@DB}V#|SdIN;Mg_1oK4=9@3MPrytdZ
zwceWFkerG#;gv^u#9Il!vHi-ba||@A(7ah;(i@-E*1G6QjJo}tauv>vv8h!(+(@CW
z(BWbpv5Wz-Bob4utjf=upI!98eV8{bpM0gPFEI!D`c?E|wu5HAVG&cU@dmnWr93Ax
z(&5pmo1jDSHen@hx~rRO@w8(hKkSae@nO$2n@89Ptmy!&49uZEuvw?qBt<TIPl5yD
zbxxfN8NSY*#j{gwiK8Wva7<L_rF7>S&k@Ce*yLd8@@<?7&fSimE8)TO#nnp7Odz8x
z*&qhbXYeC>D@M8}X%zmhm}*EVR9XmVQBX4Vlj<q?IT8tF2ajJBHr?n$e7FnRCK?fG
zC08$)s!0x%=ExqgijnF`c<>1Il$xnzxMUd3JhV^)C}kRJZH^T!X4~c2Ojd3oHT>Qo
z1zP7ZK2j?bc5{>uI@+}R%pl4dN1~!Il#71(m2IG<5?(-m8cS>$jhZ@EOo${by$l=v
zYi8Yp$@h-g`z~03%Ef$>BdxKJ`I=pVTL5LX4M_2Ls2zJ*A<BNQ3k)}%no<$xE0)4N
z*FUnSW<wnOC00f1%r7giY7l~&o}}#1tB*{xF1jkB>-lf#*?k)VW!hn)foVnr@m763
zYIjpFCah9LgE6Gd2s-)Uz89e&7Q&<aI6@jQzf$KivJzq$y&gY=L~6SC6tH!(;&(wO
zH2#>(_V=OpE|Lpzwluz&`r-tPgHYluvn<LU^jiu~uLcHC$FkRro+PpvV+T+me257Y
z{^V?Md?}W#359O~&yb|+qdxx?IvCU>Sdl$6SfvSn<$6Pst+yJmZmeB;Kkj&mr4D?o
zViZH8pBzCpO0r_93ydGJdcB3%GB#%lq-NpeaU-)cvR7DCiv18c{g)%Np?;F^^NN}A
z_?q1vBy*{-(6omtLu#c($h4IfytMbNeC}ka4p)^Tq}S4h&MlnCUyO|4*V8c28pP+o
z>0Kl(tdQRyw_Zsr0^GThZNs|hoL*63Kz1>Y8DPsOT>$A3`o14w6OfMIWCVq()kJ1A
zF8VVck%0HrAE+NTgoI{EQGbu1)L1~a4%@ICyrI}1>VK9(El|#E;)Qp*!I5m1Q63`0
z+DJWATp({`=NQ3A*Y&uF@8))KrpDy(@kE`8=c1?N(XU(wQWANpJMn{dH4%x!115E@
zqI%y;yR43cmnK-Dg7LfGkMD*U<Z$L~U9RQuGM^N`V8EOab!w2^1V^OL?YUh|+9(D|
zbFYVczerNxP?ij71^H>%4p#)=hem*Z<KI$#CDJRyFeZX{GX6}zM=kfO<~c%;Di`pT
z6@b*(8SgTpM|i>ZtbfeYNk$XPF#6`Ojth_Q=V@N$=sTXT#_Cz@ZZEwiZ$RfAzxrKv
z1M#p$te0P=jJ8UpP}VH+Z6``koW>DNI?DquQLtNFPiTS-h0XQ=Ru%5cZUgAZmZQyl
z-a9Prls~;5_Zilc!ha#Aph&P=|Ax>Ck<7g~GS&Sgb^W~O^bNZ~BJ*>5M`vT0ZZ$A*
zeTPo%!pjbNr<RJK3Coz(USqE@ZPur4#*9&_{YUec-Cnk7Z>!KU_PcWNu%uVHvlX)7
zHA`Mti%VDZk1b#lh1-_vBWujZ?ILWxPZbf{cI#4CW;*jd%|FpSTAGxd4?B7xZUpf9
zdjyyN28>^W5=YHY{QQrowl^v;2&+IoZj(a6oXFT%I!n83h5w2U_LwulI0_c~5kM#4
zRYAf+aX{tbHIAm?#De$Uu_euLA$<I1N`^2s)oou{@n>oqQ|4=AX2rO<L&)W$opLER
zGTFLVe!w~R>JQ-}vR&)NjD7dF6?pOIBgI4>*Ni`r!J$d^Rnb;rGUU3(%lGP%M>)Mb
z=a-(X5<qWM6|E5+hA?7M_xsi|d|5si4^0fR;Qx5l2g`VQrCCJNHeV`;X=9#zFwFqE
zu~)|3o1&&FZQFjF{dB*ovt)Yy-h@SjK_PgFxzC|16|gAMh;tQA-9(0z2K0L=Ecv7`
zov$OPDM~Ob$J%2du;({oay`?C`H4!!)T*o9KqD#Xb;jpu`c`Q-Khh;0-)H>a%OoXT
z&}1l*`H9-y`#{c#--@ipAtj#LaEuo161I8;Sdh^!)hqElEO!cWm0o{o@_7G*yjxuH
z*!P<xjsuoy+L5cvQ-C!qG|g)_#LYreWFP8A%5{!r!g?EE<e6ETBQDp5Da`hTXLK0E
z!b7GaKf$sCKJGprk4f^^CHWP1@_{P5s7K$rTkuo{o(S1?);=e4;WQ@M5<ho}588Bg
zRT3%RSMR|({l%$)<9Ia!E{ogJ`Z`z7Xy;k!!$^G8Pdp!)SG&4&qgctFYbMF}SmuAm
zOrk)EyrqRBlAiZg9Cgn}k=;X-6%ePfpq{dM{ef6;rkW;~j14b(y6E43%8=RZ@7@Y#
zS!|p<&sV#BR)%(TWq->cJcCr8spEJxe4EJk^>-lRQmHKyaApjPfC@v&y>T)kK$Y6&
zhJzW~*D-LnUi1A43}I=b!Z8E<Dr)UY=k~QYiieb#kTq2$LHZxQUQT*KYf}XZcn5@n
zqVO8=i|&f<tR(Kb!?y-Up*gc?zC6QLOE#%7^KnN2<)pQU5AyGTV#n5MWDmDuGX~fv
zvVKj`ZVY%)L64;X<J;|$>wwy1J0OFu0?9xJ0CRhbT1iQhIca>)djYm}md>=L8bW`g
zJRZ<nouVQ8+Lu8^RjMZ6R=xl8AzIh|CTK+K)s^O;oIvQ?b`v<@D$JM8U$CL^P$Oc(
zIAUR{AMa8*!Kxy4FN@E-HddtdGS@t4JgWE(^@!Vo2#R@J+M&dU1cM6ah?H?GXkKOw
zr!07o^|t{u2}h)C!OBjOCZDB&1di9=xf)NW`BKdO>r<xGN-a?N*g5@dz1AxeTMrV(
zR{I;u>@SNV!4aW$6FQsYje3P<Na9qCVEhf-I3Q{7UZtOwJ1q)Jh<o(m5YKWBDrTW#
zvD&eYK{th3u#{sM7_+VKJaBWCS1fq?MwB|2)oCvSBs}Z=T6!muH2Qu`{mPXYxjpqU
z3jxY>L?U{Bw1Or|&<Aw?(`?&mOQkozC3gZ&<5bCD;5ZueQH5B+08V$Nyx<A|b>B+J
z(2Z7tc@H7VPOf#f>D)w~wE&z*DAb}!05z9JwMD}mRZAQNQI+19oLK@ryeu4`9^^|b
z9{p*kNf!Xdyf<v}Q<2}3&cjO;k56l=UO86<y4)uXe2Dbac@QwWHxnH@Zk72y(Sbyf
z*a60Sh2thREmHG9)s$EsPg?J{R6XK&)F!dt;(2(CIMy><TR1#t$3ps%CY`m>tu^in
z*4lVW-(Pp%ZzNd|CxsZb+4lX$v%`j7JF6HU4Iwb?LY$32f_P9fRJc*r<^db6N>O|<
zlNt+CBC>EYwFm4QST)JUxlIAQ?uW*j0K#=m%K8hfO94zms@B=>6!bxNze#3E)O%#^
zwS#<foO69K5mPK3NFpZIRah0fUu^n<2I#ikR59Hzb@m<24~?z;boeHZ$h$P?#~=3h
z^Mm)3Az3i<M1toO3zz@<H~;t#9UsS-^#A_d2X@Xu_Rj+NS$XY$&i~CP{rB!Y=lEZ#
z{Y)i)`rm!?|9@0#|Ep{N)wTb9ANOBv`Kc|!hzADJ*I9`Cpfr#&@a$zWM8+jwcK$Ve
zfVOwSXLBOOhaEeK!*eGG99s4%i~q3c+DD*O=r?QJ*S-ia2FK;k+-akLDK5|i20z6*
zSHP@B5g7K07`*!<>)4Mce>;SoL(ROVnqCzH)>>o)Ft(gZoa(jw{OX*Mwq;tpvr?XG
z9bbor^5+X(3{ipq<8I<(zB~n071Pb+%Fkc|FeXT1CLQ;QPE9MT41i53U<9x_PQh$p
zs0>RiHEY1wA==W6mt~L1N9;QMXI}9ikrNQR)WN8!zxT%>DK^#1AT#VFj?Y<%C9D;t
zA&G59OE6gt`q+?Q=}ilX`J&DaB1>szxla`y&Fi<Hfc`UHN4@fn<J25yS!iW4d?aU-
z?*UNb+W;?<)y$W&x&o%P9Zp3Hl=QvT$9Ag8ND-MXd(-FtQrSJxC3%6sPndsJJq;%K
zjB`F_>lJnodphIA!+S${F%>%?VOEJWbhuT~lEby1s{i9rK7zz9?_a0tT3}!-#j2Py
zycs0mSt~gs+0~mfmcB>D({rmqlQ47fZ?51!w*G(*0X$#}#WIsXEF*~NP^Qyeko;sV
zN>;r{LYS6M#aPc!80PTfmHs8%pU<fR$Rd79?2^GKXl*+=>6rwI%}fopWS7rmc6Sp2
zfy{XxW23M8Y2&{hf;-DGHbU3?&E_g%ZLS>ch#rPk*iXo2gf<QDPc{Ezy(FZXbm_qZ
zsyY0K<9CwYDQ*!bMpxSx0kBwZn+&M>o#af0A;gs0(>Z6M5R@>fAYnuPCvE-b`|GuV
zg4}K!0*%+G2xHF;twRyJV4M@1c@9k2x^99*R3Bw)5{xeow>2Yjep2hZ#BIgfjKsVM
z?j(4_lhnUU>6js4qPTwkWp|g$*AGv-%ZL~<7L=!Xttj{k0Qz_Zpp~UCr*js@s4qu+
zq{#qDCvzM3t^)L-qs)BJ(;`q6GOCxU-fdWm1g2unJG9?V=G=7sS*qy+aid<H^u-9d
zf2E$aas$SSnz#4tRDmGbx5p@*wmvn3`n0{MnUr*g>eF4)@I=ow)NvBhPQE)E?rWD8
z8gM-~v1C-Dhu7o8{iC>r^+~0ZlbxiyejFL_@~wB+rDRu~^6XAvYH^GL`eB0Q-`*~>
z5tT!NSWzeg!g&!W_fkMdu&i=B0kLq4$=KZ<j93<SfN`}59X-=<V)m^aTmZ%@v`6vF
zW*~Xz@l3tcy&&{mhT(I2A4>dr8s)r2-0O_4x68*(kw+5S{Xe<Z=t!qYCr@~=C9ef0
zwkzc|E!p)yL?tbsJW8I}xm3d#p`Tgp^?DRJ;ic-K(6D>DrzB~~#d2Qa2i-xohP1Fi
zW>QLbByWUj&IO!|kV@p51M*r*hIr_-RBB)9rMi1rL_?KWZsg(%=pbw@CYAuM&&+oI
z$W`1U57@^2?vM8W*iV3jG)GrC>jM_2x##ub5MrT2!Fy*E^B~F3IiHh^Go1iV1dy=>
zK$=ROd{UKoA7PSP<_htM2_6d7ocK)uZ5lk==UD#Bk2nG)F{@<|wi~H*$Xi`$laVoc
z^qw3GPF+wXx}#!KFb=E+=$s}4wy0qZvR=E=ug_-6UAe}f4flVZ@9;0n@E}@Zop_9-
z{2eJzPRJzin+XfRp%;O1WO@?(z;<Zy#$fU}fNIQGh~Lclts@pz0C0?nV?lY26<KdD
zK&K;OM?tOB?Ua>Q4bPl}=Lw!3VEK6HNZ^RiKScG@tfC6MPcLYmKV3tbTe*z3LISSW
z#~H-mqrHDejD7znMRYxz34k(jypu>tVq<m!CSwU8g~hvRjf-qpGv3!f83q5mckg|v
z=Rs*CR`X_owQL}-Y%0cq6-2EXq#~Q!*_L=olx;lcP2*_b5y6<NQS|B`{Ff$8h=$o=
z*N-DkD=m;bov(m#twKCzz)5;hg=~;mOwMG2VQUeA!sbi=8?Kx<YYe$OLu|qCQJ&yZ
z;E<2KV=vxSUy{P!Qd5$gFZ#cVSj1CEQtv(+_Pao{(FZi)n}bBOk{GAZ76pq<NCa?Z
zj>KaJtn;c$U_09zE1kr7Nzhg-MhkK+ZGQO(3A!Pr_2frM{!>59s%zu=sXIHAg``$k
zGWSqD;X5Go_&BJwy??8_1&#r^%LEXO6mn|D6yM|0u9HvZ)s0E;I_517`xx1t@?Ovf
zb*_avhFIsy0#v3gWOj_`0L&BI%z&2&a4?{+8sIYDIwm<60?3`OD_ZiI4woL^I&xY1
zSLHtsPAc$lXf2L70Xp>|K;L}3*lOPqj}D->G#j4=jknuG<gE;XNC`f<s)zx+xko;i
z|9Ku<7>HU%R79N+W(7~^e3RK+4bF|*dQi9z5d6?c+!+vGC}Z|0K1s>78my>%;xxf;
zb^i;<7OJ>HkN)-|Sg|>fS*t?HzpU=>XR(4$l04D>&j9nEXE}mH?-p$yQvZMQ<v#uK
zkay%h@2U8UwJe1x9q50%irc>4Yzv*<nKyrXl_u&c5a4IYY5m*3mr*ANFz2J`{%3!m
z>3k;Avnn6wf+2^0rZ7MDQ}`%R5B~e|{~^5p<6<vNdvjYJ*z(@49;Uy*a5v)ZdChx0
z55*0#ehCuHr9e9la=Xa>K|5Z6{S#Trw)&?5|Gm{m41w1en>_a8D3br70wTmz|9pbB
z?Qbg_Ko_BA7Igx*_vzOrx#rkr|AcQhnZ835cAeLL(YYihaB32wy+Gg(e!_z2Qb_9`
z{C*i6UjjfWBcxyQH?;mQe<1+eltJy?KiB{noJf<9Zz}n{lFKF$3ngmifM4YL>+@xW
zfDo%J`Xl}k0RAJaBATe-2W7y5UY6Bn>zRxk=>AC`C;my2YLZrFtk};3WnLzM?c#FP
z(AuA}{Fh?Ys{w&;w??1Z9}oV|TZnsBO9i?t!DF}7&;4Fb%rXGKZCH$J={wbQMcPCy
z=h-jl=^bzH-RQ&nXMjeTfelGJd3WjeG7&xr*8JK&q4?GB7oDmHJ~YOMVgJa0U*!ev
zcn!8w`sB18-|t0wF9x{v|I;sbduX7?8?I0S<5l>FYZTGF*ORS>so$t^oJuee03~3p
zFqzr$iKL)gNq0%)wjM}@|9XRu{fYI&TDKZ8FdlPmeZ=nM+K-}<0mzT<1oU4E^m#!?
zW0P2_nfo5dHA)BgW=o8=D>sQ(pW}&w{dVV%O;5d^^h&9I7uoqH>o(ZV+qt0aFqwc5
z_(ges`EDOl^2tX+G@9y~3?Ts?nNQY!{j=K1L+i#4%dKUMKyU6jlH(q+4?j*e`@`1$
z%tp1sf4{W~x?pP^+X7>L(Vbt^w3ZoU#KG;KHMIYrgGOM{?NSh{UxoOK)ti9gR9b;s
zS^Hm-`}rr}>&b=`@k8k+f4S$Eg}e*}Vq%Wx*Z#dL@M#Cs&;K_U@6(@bASOms5g&W{
zC^75P);X<vpPz*A(a6}cwY{~!sls|-5L2yZqeuVuf{Z#()Sf5W{J$5(?Fb-<hw*kf
ze-MNvIQ+yT<{pD&0;z*q@U$CCjlRr9Pi-BOA8$+I6<*mj>k`U>7gR-ylbD>j&64wk
zxFz=*NeQn`EgyR1`iIXVdIph+t5)Q{|E?oM&(O$S`}Gg|Yev=hRa&HV;-!WByE0)+
zV*)MHQ-%KoB)Qtrj_b0CrPtSXcN0B*>0I7f2!j{Z->$XGyeN4;bH5|_FNc#1PJxq-
znU?>30VaE|83c%ixxX?5fB(?{7&yI<>;^qwq~jp3`t3!p&994h<Qdzo7kp#0r_t9>
zeEH~c^ZT(se29oLSiV`jTF(9y>EBD+`;X5VU_HkVjokl}bin1d?3E0$UsV2=CkUSf
zI&tQ`<R5gRnIBw^c9JgseJxA&jA(mjUH$%`GSTP2<-*pqo?qVhtL29PA^MRm=KtZ_
z3c+Rbo{g?kzsl}Ek79lW6gFgx^$%gxjHu7e5{dfzXRrNhhp&U<1ldnie%#NA{uLRU
z!R17~X-(^Y{(j%Z-Mb02_WBc-yTAXoC_SQ5&$?Uv0Z}`0>)u*O&@6G~fLqW(hBE5>
zv4KAAOi^N++FI9tKDYMy;cs81-*PJy|KU3-iHbPup7Do7=w&*P!t(^??muK)OGGK;
ztswu9$_T#&j62qLx@3K_RK?nR@tqSASCE1j@%hO(8v)|3`|wa-UHTLjz^p++Fgtz1
z{p0<k(7^)_e(cH1r`}&a-;FyT{!8Mv@2!aUsxyI?uZpC2{y_m4fe=mXW3>K|Gbk8=
zV};aZ{ugWS8P(L*eGe<5hz%7HkY+&;5CLg|K(K()rArGPq$9m0h$1KoC|z2RPUtOk
zP!NzVCG;X<D4|J7C?W9QNAGigPx1PGdB5Co2P2-c&)$2jJ=dIb-N9=oJ<*RiXmVC7
zBL3@hRqhwRJ`#h#e&3nbgVa{)3Ek~w%X`s#V&+amWfd;D&{+7LIs?`nrAiR@KZK=C
zYB2D~nH<w#2S8u`Y3~U*2ZCcj@?r%UB0^51z{}2J*yI&&PrR73c$D#h-%o~FC2(XF
z&3`+-4I$&42`*|u16sY|z&)-67&~SFaI_N3@wb7?e|~3aHi}~Tp}uCM8>4_9gg+sG
z!<F1vVkgjk2rmKz^%6FynYGR>6)*Ur_4B;gVe@~<Eurfqg@1T7UAU|<pOuS>>{n3$
zKuO<`LeZal)+0!bzTb@5k`JH)>%bqhN|ylfARtA8cVT`}D>tt1y0#(0RUGK`(ffZ&
zX{^YE0hKObfuBwC{D);lJ~}%&UN?4sp~9>sY#n&KNkH&AZX42jLHrr3lwc8S^aOAZ
zj0*w4S*0C}SQwf{r}gl?)rj#`@JSf(A%BN7=sQKMJG#^TT=jk?lu_L23~9#{zCzNr
zLtj!XNsuqqsF8Oc-_sv&>uRn`0@!d4@LqD-z5ZcO;(~}O`yB?W#(nq5b72bCvb0PA
zh5J6RoqY_@qz*_34golLG!SwmZzDNsUXIRA{SXXqFMa;6#Sflpg^nR`5B1tlxON~q
z7D|%BD8D<hO%)^N(xNRw$&#w}AJv8oSs_j6O$^5&f6wk$@CXR_id1jO0YnX!jlzF`
zIDSD8LPdHzz*i)@AF7`y0E1O0*V?hT8HpJ&$RqUM7SX@OAH?DNT)kAonF~$;Xx7YL
z?aPoV-5&popZ4GXgrORo9G;2teSZPwphG3FB}1Fr0k1pk#tI;#*ed{Z*g&`1<FS|}
zNj-5j;tUL`R*kv+&CmMJC7Z+m0)+`P6}*+VY*fv?-E(a^awZ>`{JP+?GSZps%ir50
z%q;+Rr}C2zsikX`)|y;v$Oa~a^@%Oy(U+rZz{kzU)BV8L_3TG5u#R58lD5+4MkiA4
zY8#!Uh87lsfZxSEhpXhSH;*p$-^TvbNk4{mkY2%uBUO^ICnX2E#Af1MfU!oJT9&Ky
z6qpB!dpliw@!*@*L9B1(ECNIkZIztMQ2PwQazJn@7J2b+n%7?u!t3lq0495nH=p?V
zl1bqpmM8ejx{w=aVEWp{kO*}Su&v<2gS7y(EdlD^G{{4;@d?^_x;@LD?c+9NK4}2T
za#-`f-E04IS%2uGTz5ppPJAIEK`{p(f&c`$$eCzBbXo^Y#Oro|qUSn)X@&%pH63Tb
zxu92$Ga&*Xb_GJ-n6@O7Z@*8q$sksY=a0VG@CIh0^VUB=buNPH9s&@<2F}=q9@o}?
z+z7e$r31Iiyvhj(c?K#sHq9BR>Le{d@wUH}=2^8(QVfqG7v&+&H|{KY(<I)!{r`8?
z_w(6r5VvL^i?o90uLDCX9tbP7*4sc3hd2$8Tez&AJG)PhUb_#KJw}##fu(@=vH@=f
z7RZUZ0M^^f(f^<4P`~<v$y38>G9mT49>?i^f(BDpK+X_i$(y|fEP(a2*T76l1KidM
zd!2E`br(pd`cufhivQ^%AO(39a5$`C+8_@^%zW;DMVb`u2i%?}<{DkWUSOLyiU(q2
z#@^xqhVzC{(G=A3Dy@}l^=^U+gI#Iy_<ug!zg8C5g1yQ&{6vmawWDBldFLRIeaxro
zU6b9VXj0{T-iST@{Na<I=|J0*=*HVWLjeEO!JXr8e1ody_eIp@r~S27{N7zi3M9oO
zHiONtDsLLFnDXKrMAtT_LI-lA?$v}e9Q)TT|8JuKf7u2hdij;z=1tzZoz*N%H0ikL
z=$pz8nEOnEd%v&|a4pS>6a9zXxCXF9H|#~Od)opF0BofM*owK-8e9a&K*<@~0f~<L
zVocI2d;VdQ0LFbEYHZNmBaYq4GOm&0z=C*wdd&hNR6?>`JQ@^HFM<{C1Yz#S;OCu^
zLgu@^mNLKn>jn5LDW<kX6XJJLyp^(k9l!vjuzzFarX}{A1Mn8J0xaa5&ON+#oqWR6
zCCDb*3s_#FNw%O)irx5!#Us53@&fY6aYPpsuOQ)Tqm8*Z@N(+~ma%H@a%*>eiJg@4
zMaiubs+u*0!sie3771nEGgH5%qj82_|M9K62R|LuKS{5D3I`MSaydDmN+r8~-ad0G
z==zHgQh>;%o5zA)1ig?h7W$Gi7D?!eld-TI-88Ss)aK(UTTW2dWL&vuKT49R-YB|w
z$@g)W<c2>)%bXl_?y|#y0=2R+@QblY^W4Axu=GBPlTY?hp5OQ1|BIu3jTIgKPIAU=
z5o9_b$HK|if}JmXhQNn;W>pQ<ktTsIni{NG8Gf;M(}h4Q<PU8t!ycucdD^}<>e@uK
z)$k(Me*0UJz?<dY?&3-@e^$WH)a9@u7DI!{XQcXd`y_b^#d2-A;y{d|a+6kBzCjh{
z%hB-no*g!DT9FB#wfdqpohQ^ATd89zb^?>*Dm8w5e_zx{h?;*=Du~?aPr5=B7q|o7
zYF27L$7ofT$GGmp3MX@jrQ85U|03sj^fiNXtQ+Oc&%uYu=M$ATuSbjr$lto7_=JAn
zHz7*yOJC9tYrRuCb_-l2=IH)fk~qCr-}4W)hxtFkm~_Xu+x)%DKAX^kar^Ly)~FtF
zm^86IoDT|B1=d)!BM3{vW~LxAlAc4mSbp8LW`FOxW3n8hq%_VSrkl`n?jRKXPcjIB
z*QL=7om~EU+Ni#UZ99Z|137!p2l^WPK9n(zY_(D>gD+wnm!tDc>uU_XC}07oeY8sr
z6U}!I``x<q!Jm{wVAwrz4ZJ4l{WADnAp8Rr^g$c$8ToA9zIgGUKQ4Im9omQddrCbQ
z!Oh$$a&Dt;-nr`XYF{>{O6F`8G^~{>PrN-?O>3#&G{006JT*H_^3>|;x|`_yYs~%Z
zX^ozHmr<Lt{lol$ekS`AJnyCM4od3(o?U-`xP26393Q}d{(oNR+|LUQu~AlL^$xO~
z`X=K_`Q=M)$mOW>UwkPK;XI!la@-4Mk-*baJllzv^{q9ORTU*?u8K7jC&fuVxBEF3
z49XNaj%=OC%iMn$nixy)UAhDvW1szQS^V|(lk8yl9f>@u&?Nco`T;{MK`Y&WNc8I+
zEjXR2#`*Xm<8dwhXE`0>4^soNdP~#|!^J6d^iv)s+ukDG$#W7+x@0X3+xS${r3r@=
z6AoXe#Dk->78gH@{d#su=-F$eI8FZW?6VZmoZm+m{!h;Y%`GozZvFprp+O<n_3ii<
zmP+R>Q<8L;(r(iqUT@$qGB|AGujrH;{J@fy8{F^<WxyP=!y>eXAxN{v(sT_AUr)3a
z+(`UDNS}R6^>b=8+<QW|)G&~7m+}uYMe;nDDKtsF?7!Kr{`JDg*uV$`=BvCs{Ey#d
zAElS{zT+FL(itsCjyKt`!7hPgQ4X3`9w)T=$X|I)YuLQ@v;Eu<lLjca37#+dT{}N#
zt|HjIeu#aV(QPtxc)eC}I!)^5Q>=6D*DBwztous$r-xzy4@H&J;0SFR|1_(%!@*)9
z5Ak^%{KF6&kOfb^f1ZCdH|)v4$($Y$%xme$%Gi{95i4R|2<1<|F7L<TXJ1PrEm|d4
zL3ZdQX}OG#ywqIL4R&s_q%#9)Jv4)AE`K<(F}|AN`iJSna2nd8<nEn5^oN&Qj{?tK
zgHxIR=T|$=&rbQpku5BQ{q=Rq2&tr=Q&&yBH|M|4>Ysoeq`P2xU-bgbUnAp(yzl_$
zO-Ddh8h*<e4;|g8dsH)1#Fo1fuJGLH=N<SrDIZQsZ6fYd{^5pSsK6v&ey054;GbU4
z2YNju+v)v(dcFPog+@t9w2u2J$jaxn^NS{?ViJ5c@o}Ab=JHUS`T0TpKAwL-B=<Pl
zxRW+IA63#cJ!(?;b!TZ!@VIa~xrKj>v|y*W55<wxD%95t{}YG*e#s<m@U+@8r*HoK
zmw(;-|K<yS545)DmN-5B@Y<O{`{L5Iz=iJImbUw<;U<{*@$NTK&vWfpELOcrww*an
z$3J^gIn^N=Ied7DT|2u&6z8-U96DI^dPT}(->N^is8*Kn3rZ(hZ<s_H+Bn;u@6zUG
zb^~*wRIY`Y(ZQc15)k>}1(`=UXY9WIA0GFM4S1!EN~%YH3L*he5V_MGO!=p5rH124
zfOo$46C%}<Cj;l&yW^Rk6v#HoY}=SsR)$g#+=I~u+439u1}tm2d{V-~=buh9Cfu;~
zsunZ7HFw%F3Ey_Hv~bHHu*WWV$KJXp+xwoO8@KZD8~cTOD#`vVj%Em)6auR~<|p{`
zZTR!SFr&0bg%17S(Es~~^NaWZR0C{FfQHxfvT(aK0B~<ac<(A3o-PM&O%1@{t$WoD
z_#jbFjlqq(t%K>wvLLZR&Y#-x``zwfz-|=&%)zrwFYGMDY5Wn>EWy#jnL?rj6dZKH
zTIbC&j!~#3g=44Kj7#05l+`3i`j0|VpZB7meS|M0{5aaC<JZ#C@*|nP=E|Ul11kV#
zyf=YRqF+(gmk*@vKt{rO*OS{=1VzZPl*xelh5Sett;vdX2i}>_AsKAsbmOB_V49X}
z$36PZG-Y^9k*%G70x}XkeFQSNk8xDVY*%Sw0=lPuv#Qb?*5RdP6UZv&@Wfq+6ec+d
z+3QOA%<D}ADYlp3M>)t?*Hes{oD0uR!6tV>SBb0D^9IXD+{ZmFQ#OGm1{?gxQ+j`p
z&(~{nN>V`c+)tU~6Po&p<n1&89oMQs9u8^P8NuYc;ZG{V;$|$i<5EzGH%~hY6<MyP
zdFIKptZ(&SKrItv+tHhP**KyLUGDLa;FXmkZEXirW~R6aG9{XsWP01T_H1i#7rJ&!
zU+Ye7{6G!M%@QdT<<4Qoc2bAW={OW-HgIcFk*7iNE69fr16Ktto=+X36B5k81wL1Z
z&qC!ZAPOB{21;*B!$PxQU`EQ$3|f~=_NlAbJ$zVdt!dg=tF-O>n$>S>jKz9=^R&P}
z>^|&%&<T<m$ZfhSAOoeV@ah|_m)@#;1?b07qFpc`?}*f?9lUHv8Xl0<f;9U9keiw+
z>&x!*U~li9wM_Q^-E=xRD@XzOw7TWMW=;?O?t6<3d@WYXtIsgTr7c<Jl`aEhF4S*B
zsMm^uIEOSmS~(9gEZawb)T#lTZqG(;0nm)uR&XR9F!hLsfyu1p4RG#aYRteJTKNEF
z7b_C{XIdkp{UkJBUWkW3_}}9Nw2lrvlg`Z(fUYLtM3htUQUi_?EXmcX%09s4=LE+<
zm@_22I|7JUW}85>l&CJjRSe{n2W|4n1-qXU9Lu~QpGiI1rKnvXavQima#x~x_9VEa
z%XaMjDtYO-vtGe1dEyV=Jx=4gnO>irBXjK5d6e|hgR9xaa4q#3?&GemWL)n{1uZL^
z+S1nSb;)!!JL?BKIU;xy8%&tx$&f81#?y{e6caIFzI>~-|GQbm_?q}?#ko~;;VEaK
zZ&ar2QBz8}A1^*2dM!^}ah`ZfaG>-!1Y(%c@FH*yHUHk}N5#sJCU=H*6Ocg9?JiF2
z^7;^qQM}hiai_aHJDux+8x!3^L&Wq7a6Z2N?z2PU)UxAgkT%{RMP~kP)T9TfbHKsq
z!b>MWIt~=J>DL$^2m8q>s5wu86Rx$@uB$OgDH9IRgGpewYARGA&J+jr*9rS3V%!EM
z50KX(JU}a?QTp)3>C%>-1#s3^-3N&iElTC3(26GKsSD~6=c|rq{AQ6qk#;x?FdDi^
zs0+*H_CVcuj4i5qnMW<Xlw1%fd36D7Jd*;Uo{ApWY&hN7t1;xhUdVRmN0thLJ02`$
zAb#4s^JsVi93mUq;KZUIn6jGO3xZA`57PTXj)$lG&D)}$L!VW3-1c;O9Rhf5$!JFL
z8IHdgxDMgKKMF$;R|nbAb{^KvvP^boSP7Z;po1JAv-d)wF(8Wv@4Q2j0r^(0q(<q$
zsEIAi*0m(pf1{Zb*g+q%L`@`OkVV-3UH;%r`kGl8DetC|+)rg*_>tF`mU#vE>S>)l
zNFlb_ZYdQWKLXsTnX+pKfPVP<sO@zm_;RI9_q-*U0!8d!-y#X5*M1v-Cp{hLLM%C_
z)Ybr;vH>((Iz0|tiM5SbXrD<N5#4KUy*R@5yP@1i$$i-ovgG72fI9ZbSBVrpqU|A?
z>JD<OtUz1i89SFDA&c;ebSrz>aqH6R+$(z@sH9E;O)KBQxX-234snilFk2^|Yt3N$
zi64XvW|=s;fGpb)n7o&A496`p(62QhYk_}fAVj4P)*R?q!!Oeye>`I!;jktfFNBCq
zGoV<0?)8u%aEn9NX8zuGKKt354&imhlD2ePfP8p*;lMuMc;DTvk0qUDK)^Q}VU;z)
zP6i?EEKaX}+aWJ}@ntvZ`of)e;VM#M!17^#i@$e<xaPDTY@5cemNd3}eG-28yHwfp
zR7I5g$)(?&nEc?h{5te{eq&Ze23wXqU)adA_W~j5{qrIoj&2j$NQ{Yr5~ggV4s2Ns
zQ0Frv;>>&jc4ykqS<^eiZL#6dlo^vpr^^no%~Mof2Ek2koHp@?e*u^ooLiXs+S!7a
zzP_MqnAGG28qXJ9o$@GPDeI(32BnR2A%GqEIRN2d!ZU$2t5SN=<@Ow5R(+>lt7C*p
zf>;bZ1l+G+$qQhW3D{zWEL88%!##`(Xq<1zKHJALdM^b<b+HvgX4l<~8oQruo|M7o
zBtm=c^<s6OtjgLC-kd}tYp{E?bSKuBb+bdoIiJB5wsI<R%<O>y%{WFrI#(MPkY<kf
zv7%w#JhB*(h{h3pIFF9CEUTAQ(k@)h37LU^EiMonNH(p{vJWvOypE5*DjE`*dtG8D
zJ=fw=;4OR7ZepxnmqQO;)Nqku-a{<W7Bl?z3MV)F8)rMsLH+0H+dmG$ja~3$<F!O1
z+GlGQ1XYc1tl776XxNqhfD{@!Wp>%)P$!bhw!lpicsY37Y2Fr*^8&`#ZW7BXgLjvJ
z&(;alSmZkBR7ZD0?E9N~n&Yv-aSqVVd}qQS_Zx=h48&^}c~!O>dnOcSYYR09qgnRs
z@=jKt4wqYwi3k=~l*fI0#iy(c`9;gY?E~a+G5Vp17f!$@=^obhU>V}DBWR-{x(dad
zK*%^!l~G|(ers<tbnpJ^OjschSNn?s$z(9~?s1VpqigBh**1tRAcd_q?~;13-N=VL
zHF&1%xml(m*uOHX%J!bA4;zw~g;{U^SDyT#;N;i6o^OdlIwysP?885be6+rP3F^Cd
zj&@#Yi_U>yu<WG`?brBXCaVz!0@PZ8v+LcpN~O^}-BRuyur}=<WkEH!;36fv4PwyP
ztco7io;KZW0YGcV<<9m-i|R1wy@d$~bT8eWDs9P%kq8k7xmmo?b)m)P1FS!um}$DK
zRp%boW;-&3CMgH%j@5!bfnxbrMRgeU=JHOEq(Lpf-fF#f1bm*<7js*2XF3Cz>|^42
z;*4oYz?u$qj%|9q&8(_1SxQ2cttcWGxV9>R6moZot@^jHh2$lp@P_n*OElrjo4}uR
zVDJW(Ll=^<1a8zUBZ7mK?VgUaLHQfx&xdfkM1C9U!@dIsSFsba^E_@91zbE^g6UQ<
zVOw<h9x^G%FBgX~WZsy9y2%tMY4V0ZM$Qt8Rw_rQtE{TBBNx&OkwDYJ4v#tRn@K%v
z@eTsR-oG7J6y57&=Skrhh4RejU~@or;M4YWePL6KUH9?8G?Fk0thSs}LKp6w@vR0A
zcO2>_)s?6|Z06BKdcgWoD&|Ksct{?UjMse#OG{<BdBr$O<q=h*`_6x-JANN8f6=EN
z3k`QA-v#KaO61ZY+AaLuN?>j>MzcB8dvt0itLb}17~@u_du@0`DYb1}J4Um6cMxPi
z&y>bHXN<EmV8GduvNXsGkjZK8vAAk{3&}OnzAh#rut*cG`);m)=G>0)1jEBZljmLb
zIsDq&L_T118(w67-S*lHm<~tvYzt-7$dCu^1KPH9nPt(_O58N60eF>>^=MO}!>R7M
z-7n&<Q29UU8CuKRn523)-pJS^&siAHF?TH3@d(A5ymYYj@{<9JjhVQ)c<(0z$L;&=
z_VlzAN^NWO#|TBl5!2!U^F@uU;Bh6%wW1YI<uPT^J*#E6n`ARXgd<!HU7c+l?E3g5
zwddvoq8wbyM(Q4)r1l-D=BS=$8!VY`W(Ku3Tr~6JhL@F;EJ0hoN?W$Fk*~s(UMlBj
z4D^<3+E>ocNScPQw5C!jX6WVG`+mB_ddiltrPvI46dEzPY&WDn1O>Q^yF=;smO|-)
zorLjFk-1b=q!e(*YJiOOe}GIRC32vHKHm$QT@R-AP$6~rgDp`0qwCKzV2Xs=aix@8
z?nArxmZRnh98O*}s1zFV?=v$8#v%Ec=oZpugP8~9wzR#_V`aJmwAjkX*>AzaWf!1m
zNWjUvEZb;WVNn<>bTAF!f|}-{Ne1k2L46Hudq3ZA(<sAJ{2eZrkDKs}V5ycDJAu+;
z{C!sGc%m;bX)NJvXpQ8dm2~DQ0D!hE&Zpj`myK8uxV4a%rgYmPy+LTO3^b%meIc+O
zs5_a*@n~dzD9D#|vYw5DXNn$qC5th;%5n0xAwt<IJkxV*9<KdpDnebE8no3^GOjS@
zHew6eY}kfWXzN3%fp4JW?O>MwZ;`KQ`J`Ts3>sjeSEQKjgseAEx*zvAJeW(|srAgz
zXURIBt=0PER@5lnlJ(Pm(M&UQTjzeLDMD^?za<Et%X)Luuh8GkVu`W7i*d&=Z*4QC
zUR(-!WScWGw^K0fj#nW)dHHXc^l!i^qkc;J$~g`lPWRFx?hU91QPGgxX<4hZdvx>E
z;oR@?O+oWPE?arI&MRZ~jjo`0&Rya#aqa=n-iWzxMEq;4C#c{VJgQUNujlFe`9h#e
z4qD28_MyiMWf#0kFJXoON%5$h#+Qxefj>@&(z!HD^2II|*){pa8}}fs1JB6Qp=p>t
zTw*fdt%YqG)UydH*tl{JC8=$xZnd>dbZRnDW-QhtBEw8!8LvY-ddyw>+pN*i1ods;
za6a%sqr-gNs>l|$ZhAvEyLxY_(9}dsF7i&dFQPWbqQ*zof}xlEK*A(Ksv3>c;eti7
zR&!@8kgAy_Jq+(<tl1wajb%7L{?%ipytCU=w74<KXDZc_<IS3~nZ;3OvHASaib1^n
z*tk=yf^Eg}sa}yoqSHcAyRE9UpQnxUt_JjQLtg&L=Wmw2=Rl_f5k3KN%3^V46o%#y
z)EI+tF;2D&25Rj*C^``!abf3kzu2!;Q8&f1`~vU07B_JW3WH47VZ>p!KsID~7y2q0
z*vSe&o)0XGHQ@8oba^xh)&udxP+>cNVmT0;m_1zEy71QJNxSETH!C5|@SUVnzlDPQ
zq%J0Lyy~b?zRo}HTFQt<mB4F#D)Rg9&nT9=hgH8lG!tXh!z&Z0o4XeiSl9M1F96Vl
z=|&a?c)GEB$`43#X>=8|IQQl|z;ecbMTF=_AN@iT8B$gg`wmE@!>;6`Kg9~$JRPb9
z7{$%l5~bU@uZNE&nWkO1QD}|jW_&(8AbpnIISJavzl;U0{SGNgx=E>cy2-byP1kzh
z*m7S{u9P<I7i9ey2Z0`_%3`*_fg<L6sU=7^e5vUTTvWZav<-O-)`^GgWQt=Ww8e=~
z7CDkF0ylIYbCpYME$Gkz>0<_GMlEoy2z2lzjkFv>gDl)IiWO<Xe`yLd7(n&C+R#*z
zodt2IxQi8E66;uIGY1*BJ!n>P>;=GsCHwEOJa$UOvK?v_kK18T^S%HmwUR@6Hvv=d
zC9~CmPN-Y(Vq?9B!7>3vSF;Gqmta`<;z4Oz4k_5MFadZw9cBZ2fRJH$3`AzwO^oRb
zmfK!<QYxBmY5K{q1UA)dxO6lw#LcF-4_HgxlW&Q+`vo*Qit8xbh0Z3&K^=WVch1-s
z@?7S)U=NbjmNMF~+h)fO0t_=SndzS73;mkfhcbfNYz45=ac7f|sl~=Cs^Zb+!f@|N
znB7_0&)qhnJiA_P-4<nD*2i>^q_^rE+HsQ2T3YYI>c=XuPCkz;`+Oudi%c}V<|<Z%
z$q9=SqmPC4sx+hdhK_4isXpqO$QS(Hc+}1<<wFd;9V#b}E;ky!w1ppQXQ4uB8r#D*
zcaJ6~2X*zTcXLa6u&MOaiuxFe&+O_abXE!L<`D8+#>u2xgq=35>++MQ3sUS~n_Yd&
zW7d47eLe6sKbEP;Tb*+sgDadBpSd=1*n+Yqic^`*wNG_2r?~S7IH|chb#fpfpUcZz
zB160294FY~+Z#8|E;=p&xHr%2C@R4jsRSQ$w*@HD=%GZlY;}675#%o{pzU&*&^O(J
z^NGrWi$oxR6nNaAwFIe|<#r+3&3Bx9>%&IKUiD+D+;HjoV#B^YzYoBqpX~&&piRvs
z9wVhcG4TXmoc>f&Kl~PD^zyay;bU(w@8T6FJP^}@C~DA7NXG+mjpLs1S%H`o`c5sE
z--p}#7LW4p0K$SzIDa3Ng4~7c0Z*lU0vUqd9H#hf%^>Lr1)s;qqxGOMY#sY5KO~;?
z_!*eh+o7fa0%G34WF`PQNPpB4BsI7+tAJi&1t2Q<azjAXB;;}VBX&W=uiDVyG!7ik
ztv&iuSg}7QLHw2}2dzd<Gp)T{0ewVxW7gZCUO`Q0pI&~^#dXjU5&MxNkNg-gIH-gx
z?5lSK;;7x@KEt434_UoCaL7Zf8DkI$f>uH3Brc9n>=w`*5mzdQdN<#iTVn$m<s_6i
zPTir?dkjvNGxsdC^Yj^QFHT5hG7=l|XnY}5Lh9A}8&DZV+`k!B@VglPL7#fiKUYZz
z2Kqc5!&Fw2%Rv)?dT@FlCgz0X+qsr2#b!5fOo2X%d&1S=Gpi`a*|r1Ubzf2BPD`3X
zbO11r>9BMcw#z01_^PV_gmht#!PF{o#gJyH%m)BJT{r~}8qe-OZCYRA)k?F>w~fhB
z?DqE-{cZ+qq?Y}mWB*G9jCIXiDB{3ZknGv4m(8q!RvfUM8Q)z?V!7P8b8n87)`0+j
zpvRN0{AGam!IJ{Ng|Pu%J_TyhKOi63lHGLxuZ;cxdAhr-1k}~fqO>qw>Bb1&+t))(
z#JASg3&3|l=y6~!`cZ{L^k@tXhd{5@GsU<!$OTijk*njIHgS2=wz4BRW)s(9<dQ#$
zq(I%0Mz6wK4E4qwvbV6t;j@b?u|97QHn&(H3*veZXzsaMKU(cO;!a8m0C_3clFyM9
zBYwLj3+)fTw*@v_#y*augxAB|%MhCeVh23RIlQcq_ihjB*9T0F+1@>lEVC=J-u{|S
zm&Y-H4M<QTId9oG-Qbu~bAk9CA4MV;g#^ZOK$L^kZWR$WpoTj0uAN6qJ*|rqx2z|a
zuXr<9`o=ITR@wWk?LhXUMytZRH-Hwh0U#6KiW>m&hEWlj-b-~>?)m;U{yFCSNqe@C
z%nm22t3sn4>~G+b@94M=`F*@(6t~#E+bpudwLPbDwxX~ze*?3nK~mY!+hDGo_*PKz
zwMA!G<fYA?Y9V^QQKZ;*Pn9s-EVEY1H_+~|XT!8aM=ps*t<8IRs~?0BdR>(pVc`Ul
z0x5J=snH@1b(Y+RXJX%G8`&u#HxqBPZneBh(+V}8MiB+R_!wjlWZ|3FvawV1%_yQ&
z!9X~n7EcJ>Xz4+;bV}aI>B?q;tsq*P9(BiH*2D$^BU>u^U0|OEoVrUVB$k(1cDD;$
zNaTzyan8hKZ=<Q|SsvUS<-|~#v`Ch{V}&Jr8EvViA7LXJHN+XywNl?GwAXjni(lt-
zD!l{Qn4%K*?ldl}O;2PepCuEzxeTU#3+KdTSRIj4)DqJIq0}6aE08<A&ik>YX2_Uq
zkdgo-p2v&d3KvqZLy+7g(45nV=-7lufQp4qK?jU&`%B=VxOGv3A5i1=k6JHBoV(Z{
z23;4L-EMd-1BLXOI^Yi(!oTbQ?SO?u?hEfFg45pR8AL#A(k{3g{#8c|MvkC-TBvw3
z@0>tO9C7M<xXKAj6vEA0YpPV~qAy+@CBMqs;3H_oD7@l3?m6?^H@EHZ@u~MxH&mVE
z8W?U0D1qX70-*8yCcBY+@nIF$hBI?LK@s-?0q7&Wn*w8xM?S3Mv-3|thg?9Db>`=K
z>t^+qIj*)Hu-3i|zQ&4}CCrtRma4*g$6_5FM9gT1h^E;wN)$~ZZo*1#uk`E=kL1kk
z#oi5;BkTa=(>_5L<U0IW=XwLX#1z#Hl`J6nF%5w1Z9-lKEGVu$J35loFOpVj3Yv?D
z;70;kh=L;xE~Fabo~>O3FKmQ{zZEkzzeU4gs}I!&Y8mD(044<Kr2Dn^zlF%`PkDSj
zz{ap(nHOkR?+Iv0{?L({Pz*e|=#`D2le1herV=R?0-*BUNzh}L7#~|ib8VP&RH>O9
z@7&qa4;S5HiA&mO2K|bcijVYSwsK1C0+le@-TBXhPZ;kf{rCnx+O0D~P)}B)spo*8
zbDm339HEr@+g)J!Fw|~%67pT<9}0%x)sI@`ad+Dw+%2RwcA*_guD)02im*jAh#&|P
zhc*Gk9fqsTEhr0?CmZvOJC{r&YM>wSDiNNY5WkO~B8jJ}Q$U6o1s&!iG~dmJtzI|T
zXqAV`!f5VC;MY)<FKoMLdX?h0_!H{a@0Wh}VJxJ2$#<P*?%O6lD8pdw&i$np3|<XM
z$I|*><6HLUxmw|rW@KR%dO|a?%&f9TEU*mMn_wHW<k<Z_==-1pQpuTvls=(9{Xhy7
z2;>Z=KfhEN;vLdAMeQ@s52`LI)aZ7Djl?Y6;a{ya9Khw{8QhPbF(mi$hE|mER18Gg
zC<mxUuc?FxOe7ic=ds##sv%~mqSdu!-3lM>hn?xOa2f6z-0@fn=4##=bh(7{1V-5&
zG8mb)U^2o7>06z@20tz@ZhEro!WmfJj1k%lyVRuJla;EwORD_%f|Hv6MI2oBMLg@N
zK~U=ni?TzV9Z4lXCrRiWeVfFQ2$`7F1CHp>pigm4a^#RN$aRaiw@VS3M&X5Qq`_ug
zBofDCX!Uunq}dygTSZL_Ft$2VkU2XO)K}y=)aXewf9o4S>fjz94zPYvJi9B*B=XpS
zse8B(cwz_AAZG78CroEQK^#aj;P|XH4*PzAW=?)BTPybngoB!qz;XU%kIYNZm3&_H
zEv_xv(rWu&<o#n35aprvDD{7VZ@&ss$FBK7S;~98yzi<|NwdkUK&~BX`V@?95D`G1
z2!ABNvUN*51oRyL0cNtuDDb8^UbY9=$p`gt<GUaI=FpHOtb@<25!>B%mudiUqvl+k
zmY%Wg7@0nuP#)ySz=)ykG)DLWqZd7)_QQ|E`zy9>vwKmXBYLDru|-F06G!3VA!VQq
zxNVh6T&-q3LE-c1CNuf|ncvNJqw~~*K?|TyO=-$9rk(_V%oD!ADED?oi(4G~wPCYT
z(*u$cg#fh+04{w}s~;2^g<R)eaf}#(ZVQwIJHOB63WJoG+!E1*6~J*9f>N+N$`*YJ
zjy7k2l9dEOy>KUi7pyT)$E)rOU>lfS7k>NSz$G+xezok5;1ul-Iz7HG>_K2o#M+bI
zSth=lec#ouoZngpEaj|c`$b3g00WH)st;RS7vdb%WBk^*pNjd*n~75y7VSmUdk|4S
z9(ep2gj>Js_PZdy1%>RObOY`y;<rp@slmEmH;}lo-`cM6TbF-zz>Lc$LElDeWab$g
zChi*5Wj)_nzq8FfN(t_!-<kuL+*SBQbH6E++B@e4dsXO5PBANs4ZL7uve3qD5hLK<
zA>Rv%3?ikxl}WuTbQ4n%D6Rvj@aQ}o1AV%@7Ht$Bq5oRCU;UjJIRpPzn|FUc9h`VA
z$E77QL#&dps&h%GuLhA%<1pj=O|_Qfjb4?rLDSa-p6jXzX>yh*J3M7+6=Ic*D;;n4
zU-Ig#x${`H>rGMRk`?UCTIuEu@l!9F6*AP!SDl*_7^@QIUHGCp?K*cA{U{m{ho~+^
zZ6{pX{{S<At+^zi+*Ez~2I$>T&ttgN>In6WOjR1zKFV{?4OF3;@IL?ool~SFWmj6(
zGO%e1944f)i`K#UtKlaQ78f~e%rgW~Qk4KG$#h691B72AbI9pewvrq@_YTT_kK3Cg
zd<Gj>`|ZO9E+5_jw9+Gcq`JKk^wu}}x*H1sg#smH*;gMPl2PrW?%gRe0gq-9KE#&#
za$(bv;SnSUcoqjwj)u<`0k8^{KQ|VBf7rsCLXs?VQ?A$w=4?*cx|XtwTcyp#qeNj`
z?{~CRSR*qy6sjdPRdAOHpB6BmSFzvCkew+)mQQbFAe=L=nzAW|sMLm}k#9UO*MKY6
zWqNVv=PdVVMTTS@`rLN$E9Yo2_siSXJ-TPwTDeI%s<caD$EP<F=hcgb(2mmWcARNW
zK1K4idrstasZ8pPPg%0wkL}&O^KOAVHs<P7&-tUgx~7{cmaoOKW2;_IS5%M__Vm0q
zOx`iQEXR0tosBN&Gy0^4^CxW%VvJ2pF3Q|eo;b<B6$72@CIS3sw-XaU>Pb^z;u(S(
zUy6!sE-+d0IDUt~RV5bhkt^V+hp=ixTU7x-#FqxjCC{1LVoBh%&BVH%L?y%2is3#s
zc$D;J68wp0eOSulp+gPfNLwUVf!pG6ffoT^kM6JC>P8OA)<O^h&phPySa2LJCJRxr
zQihfhjKZ`_m3n<Jhf1@2=a~(|&+)ziGqPsLsdsVtV_CU>jd)p6_(wdO8ONyWM3nhL
zZe&6Uaq0u8>1ZKFvust7Iuns&@idAWlB?*;UpZ~tbm`ZU%$2a5nr)hd@SF-QC!a2Y
zoYG-$*yw06Z0!EbrrANn9&URMVN`G6)^O4ys=EC8eT}W0mPdK;RPVy?r6_Ir8cPM6
z6S7wo%QGbJo%qQkSf}@Mv6A)O3Kad_KK8HO*w63$lK_&m3$1eJUFm!iuHCD0`PH!w
zT(RFY*}a72SMc*&hKz>HZ7#60Bq?=Y(XKh9L!!Zk_QqP}T7n_&le*H?1O!8(U<Fz8
zOkq*VYITBs4e_d8@`Fz6dFG0k_{jOQagHAY<I^4|R;ksXiKjW43N_qWaQs?Qo7M3i
z2$r7WOXrt6E`KD^^{6wRn{U;tyFJqS{k3u_h5lSaTTH^GsnXY$Dl2;Nq@1&L8CldU
zW;1FAreiZM_c?W!84LTAJlEj8W4h|u<IVvJC!FbmiwydsieRIjo@98GA3m-Kvx;{W
zO_X5!rnc?-th_xbr!wCut&&sv+^^&G7X^wThavAdcDuhhT)#^tw-dk&AdlbK{6$mz
z{g58G1xhb|vmEjNU4QtuEBYZh0eP`*<K*t|lDoft_(A%lQE#8$Y`W^>qJ{LkVtTP@
z#|Y^Pji_65O2s=-hF0F#=`+P>^Ci|nQio<6!SI^OL|5gWQ|^`%nyr#4bi?RDZ>=FZ
zq`@I9)>Dk#UXl7<oBf;E;tS|Ud3>TUP5sjz7cZ5uouM+$<8KRnR)3`U>VrsBu`Oo#
z)y$}6+c+0?`d#?t;LFIM69lFr?oI3P{`lh$6&x_!BF|Y=p2~^(XfCGdMWvrY2W!$r
ze+R|?WZj7-)y+9Qjg6P1!5THW@g=jwN|cx|m!MLTcY(%)OEg&NEd*~z(Ys1}f#(``
z+ZHWbe~!S1AGG&TuN&9|PyH^;{XKHr(6Hg|3mpFYFaHY#q}&JNue#uB67XBP%P)dn
zD|ON&Tukj))a3_SGko`GG)NlC>zX8;ouSD71B>ZI@5B@4Xu5aJY2u3;5-@dP-LY($
zipsqn-O9QiC##T9&QMG1j195YwY!`TAFcOW<Grr1&Q~DR=@_wz0~{(sFK-Gh2?s{P
ztE{`?)EBYWxh>9E3E>t=9yPFzlF<BjhU=ou`9<RMVwgx@b5m7}5B!>np~IWrRmL29
z*2_;L#3va+1DFkmWs5v1pD2WYG(w_cc=^_t8cUb%kM42`!gr4j80xAZZ}fEP#xk>k
zyKkM<AYIq%$|CPRv$GSKO{1sLk_q$;YRk{UEX~GBza_ql+gsDSxs+W|!h7v)MNC_!
zUcxeC@bh(YDO$Krii@VF+9Q5s>(*R$50Ypqn#ywL*9LNo!EgPgxY7Qv*ZzT*^z#EG
zu>pu+oTlap&F|O5&%g5jfM{Bq*Vy;{=9KoY?>I{Z3IH2-_`U+MV=MJ+B(^=WvN(;D
z{P4Z{{E@T1(`hbu@7yP7C{!hCw9g0X2AnMF5mkz9>-@ZuUC!ndw?<AQU1-(~nXt$Z
zpKz{DR`jH>p7ri<gCWIBr-efzLH5Dx&^+8$A&Pz&k?$#WCUlbdW<N5|aC{&>2En2*
zvz>;IYEdU^HYu0QzovsF7d7eRp?0O>E)0zGCrEU?3N%T9qq7pV)}?f04CSrrM9(?a
zIVFxxy(D;fhZnwQhu=)-)|Q_!n5!fq2L|q)3m$OY@R_=DpTOofy1`Stp&$BwXrx=X
za;D6hPbF^5x-;J}D}PSO_`&x3YE+FyB9TJu*I-pr_HeGu7M*wc!(cs;wgQyNXxV!*
zfBMk}aqy#j(~mrV`VqL&Bj`$vXzD&|dG{`!w0?8>@X(hUH73g?Qbxn@!n8{Hl+&Ia
zSyxJQ5yFd>6d$#-NfBmPLA4G|O>RCFk&~KO@o0gm64)i$V#PD02=wx_@O#pQ-0X@B
zTo3cIcRLsFL>%)nw)}|4Xf!&sT35oNX4=ys`Q3z8q<z_dZ5NB>cvR1|uI^jfi%RAi
z8iX?)$JBMseXUsZWRu|ZMT*QJ8H&XuG!@NDOKRN8ixR>`r)56g)R13hwz_35AM9(W
zK8CKAt@epJnjz7}<&ZgE4i=x<@UobSRoWWsC+{m=@&@<BDwoDXEN{NH61E_d=m<M&
zh|xOR9q*9UMD%0L@z{Kw1k`wu{QO4l+B^{I?(DX7Zd_Vhm2!=C7|yWrrseup!u0ts
zT4sCLzR+kLij*6FS{c%S^k(pXgy!GwTl@;(19hEdX#71f=&z*_1Y%Fn5zjv?uzh|y
zr$NL3kF@DSdtVeICtBCCW~FcYJ*HoAM4>HJOpFPWRyt^ljhWrOI_}o)08ih<SC2|v
zBi<V78WNJXOOGB_Lq@G}l^Pg?tlqa{pAnsIGc}51Z)WCboc71c^rE8rnozrm;Z^U_
z*-+NcG=Qw4#et9Yf*2-OlCF5QVFqVw8#`ZBmfKUYl3BU-PIMw#$q5m?@W6db^i5W<
zKK4*_qJzZgjIQktX|k(%TG&IpXX(j_==9vix?-)8%#T-|5~V&{ThqFgy+wzL5<PCH
z?=qUGdL>GYz4Z6RWC>ui>>bu#HD|BplQznnkX(j{M5M!<f@?lvqQ^zNhQ}rPam8jU
zNh43PgS#WQf~2GSl=;`Rm-an%AGJfqv5o(}T=Umo4?+Cm1KmnD{v=W%D?<d`<qJE%
zWmo+ALjTICdLP{o6V_BFsn|HAx5jE(?OE||I@Q`FeCl(DQ8c-Hsg1c$KhiHn=PM#A
zqD8wH*}*-XMG))Vtjs1PVxnmaP2vaA8ujt`0kLZ?rLR|wKHgm*4a*6q_XvE~!9B)b
zJD3Ji5{)pduwmx-Q7HwidS#D%!;m|=%x<Q}qKZ4*^5&wsInN5$qKRSZTdmiKj}kCM
zZgt5_gtkI->twPG(a=6}U?S!V_r@unMbEisRgTvcRYWz_H!jRPzw#|#4ySIF;4WNX
zAYD|m9(HVM*E31zjf@YnD|n^dE8U)r#o%o42D6X&Y8vlkTbAU$9l<vA(NA_--h}H&
zOlRXO{E{YiEA^Y)#GY%JB<@*PT<mH3Fg^`&gKM=ryH_n?obl-0jUKx+<V<r8Z`jda
z>mxHvT6BU%V~!K^udwi&QuICw1VnPFU1E*@(`UGK;Yn(H(z-@E3KEC^BwpqLp|2yr
z;Q|02=?}FgmRNsJW4H4<dSKx5`V*6XUSYD~A08S67Qrr4-z$d6->U%(OUEWgOaE7V
zwdE<nP!R#D1{nkTz|zC?XwIeTc{aH<xTPT17lDSk7eP*2nCyzXCB!Pn92c!Z)RurZ
zXOJD~TtZJ4O(O!^qE)dY6t#pr5sA_6Tt2DhRuP3>yHM7c?1b+Q#~1a}`5B~pnJ683
zZ)vD(<)jkyeunJS^Ch?-;*-<2jDj(x<L+$vVrUa0PUC8#A6~)1va+V+yN^v_c1(?S
ztGl6WT3f$5rtjecoL7P<d^er80^YA^U&Ac9$*iZ$EgAG)a#FZ1Qo~c(+ZuKj_pB`K
z8^OHY&_SUzdaCLi6*|$dZ*oKwr6IXI;J>+Rr`%3!pZ`YIJFDIK{*J~XSCU0RM&MzN
z1{C+Q0fDvsQK9~+jq<v#`y0c(@bqWz#gRCZ7KgfcM8}2BT&-6%gI(1zy|qC#{aohW
zE6L{Ipm$sQ{9kgI2VsCmr|@(>!y2lNZ7x{Cm^e5&EnVgzqB8_1Z`OfB@(*CHFR>M9
z+SK~TJev`!CSOY2g?R#C<)`ei+l&ON@Ml>82{r&oL9c_uJj*1&9rYkU@-%2noF^;&
z6LbXDqM)R<4U?Am=R}_WJe1PB2~o2j9@w=rCxNozn7AF!zy@d-p*0MLT2mRL5JYOo
zM4=p%^O$C!dagetMYaX~>Qr@3|MO%!{~!cxMYwW;OfH0OK`efot)dAe>2mmDM>r;#
zY|iJcb7SQbtzPVB=m!k|$Q{4hL#k8HN=Nk!DKN7r5VC4&t<^FT-`gFX#2u@V=Or+b
zG~s*v7}LT>9V-#2P`&2(;SEA{EgCi<mN2Z0_}pux$e3M=UudZm<l@X_I!DTTmtQ=*
zr8%5<jj1f7HFcp!jf;~feCp0)l9dh4xi6-N#QbTh!6q$^O1^AhzA}5(>7mj=O~K%i
zk=E?U;g*gS(_2>2r0W(VHkg^&Qawor1*;LB8n=nWq18aJMQKWi5n3Y(#m$AWbX3a?
z=QB73o9jXay5~Bp&sS@epJUq9EyiSwN}VWr?Ov2fto&M`61nnKzW*CZBZccNj;uCd
zs5F{StB$V^dpdV<HmJzwYJ~;12G$(KJ+(|mzaFp=g^}0vw<4HSTyai`y;e5(b8-r}
z>4>!p&6P+LwuXqWi7(e=0c@I?GQ-YVnsQRRDQs<OEioN|Q^+i_RoUUo!iW8J-ix83
zz$H4aO)}b@In8esGi|(gRRS>89&z9%Ku3CRi64G6wh2|h8Cf^~k7`<HNUk}$Nhl%W
zvDpoP{XFo)cxx4B4D2fS)-3^DD)&0nXb^!lnJy~U;)VYqVm-O|B=vRj_4A&PANIM#
zaX`yA?uZrEf#>U26lvN6uja$#72)A1h3emKp+Bb|e*rX$2lMo1HbFOw%4-4**??xp
z2^n?go?cwNo~!t$Cy?fXa?pAak4Ugg8=x+o5RekJhH0^ev8Z1btz@KSdL0?nD(B{v
z?SMYJOwIn*%AB<<D>gIB&ifszZ-33Huty)v=ajhANqM?O#^2UGmoxIs7lP~t%BmE2
zg5o0{qhF-O@hnkw>r3(w4FpvX8*znJBiy3)Dv_U!Bb{2A8l=YOh`pO7hS=}um9UVZ
zk_oIV8hQ1#%Hdl36q#4`mM`^g*F1P!Z9dP{X6<`dGY>5x9I!K;RKb+Wtv9Yxddpf%
zQb5941V$+I@D)#U_EB?HfQ3dBvd(1LXjBN{@=p0mqQ#?TM6DtR-p;z$4Ow%N<ZA26
z`<hbZ(zZFOuJH}s#dccSrpC543P&fbPiBo6g$NAMFw!~CdoJLSxevFh<BzL#ERH^o
zKV{N#@tbFaan^$5eb|KHx%R7JwJ>gaLknku$qMPos=mgn#@d-OZCeG;uWQz&)(ak0
z^F4*+oy~8^hn5c0+AqmrdwW@r69yB9=sN#8ELX~dK$!P!k!H$wXNs)>;&eOWBDFEh
zUVv@I;sy$4TCwzh*lGAhp#(J8(6Aq)xnM}fYd7QqCbT^ebsL>qUTg*=a$5L$GJUCK
zq!-lC@X}yYwzC+-QE~(tjgCf2*Z)mii#d{NSp;n^Y5X<qZh$-7Jm2u9gqjGtT4@KY
z0$@&x8Da*2y+0A9G)<?ie<}~u%Z4=MTJ4KkIOJFv|L2K;%y_n@U8C&Mr@!QuP_kB&
zG>#K-0QKV#-W?DWj?JKaksg2)%L*z_oGw=fUeT6v5iUH@mRFMG$v7V~Tp0hb(Xmp|
zN>Z8=jxUw%sKgjBua#J?EcYR~UG%i5u4B6~7`og+P8QS#Pic75>5&9w&zlJzXG7l<
z7CN=lJ=16tBB;xRXOI(%<n7WH=a4UKtfz6quXww7aeLHirmBsilb?b(Yi@Xt<qGgI
z+j_n%jQF_uu(dquQEL>7y7QE##>o6j&dL7cSAw1Mhn7m;=Nj1YnBx!;1pQKGJ7NIM
ziK?Q-5r@7u-dXPGoavlCTy4UHoS*0C&lOZ#E4?kezOWYD_k8#KgAR+HBT?E+?gyGK
z=PE{ZR-GHv8e&@-e_zQ53{j*K{nxl3E%!=fG>>{S+l(8()t%!$<IAUIz9Ts{yoT#z
zAey$~T{rHd7{wNG@Jm^_PK#-7PRn<&r89__*f4s3zNmNCh=M53!d&Kz)>^6oj5=y{
zYUS*!3Wbrh(t#;r+tQ?F-GNU5_^fI@kH^(7jj#WeuU!Uk0GvX%PP+oIo#^3UMg@Xc
zTT~cE1`zIXD6ZcC&9m{e7uAgkc<dzTC}eTxM3YvP^B;D)52H{lcMm{V8o%Pj947Wk
zfC6=K0cZtvXygjS=RFPtW!_VLP_ym=h#n5%@Eys;+cvJEpp~W$cICO~@S{s8nZGc<
ze=8P$aizeVx7)2d-b*NW^a~XAQcT)i^xV2!$XX?{Hg?ew5JCMRCn4T7V2i8?)djY{
zDK#x-J3QJdU3$KqOsp&>w)#Hz+|JpQchgr&Z}QjhzZw%hXS<az<u0PiT#Swz=4+N7
z(z?!y)`!Cj4Xv)WIu){LD&QicBUJl+;E5=F_K$HjxOl##15ap0zWJQfBA0@gC;qLl
z*gLr3Zc945dBp2LzL~|CP~i(IQ*}-Hv`sa3#kfTtv1mC5{eA_~#}N#%npabH3xtGj
z1)4@aUd{MR0Sb@Pf)q5|Y2?$SoD4eN{dlm!{oX!ID2v^^nir+~dNlg&CBNkHI1k3N
zxUyB_dPa5gYL#g;xQ|flVb|xWiQ%(FUe@8aW-J)!3Dw|!IFr7I>*hI$Wx@h)4}}Y+
zw`;lH<92>)TD4P?s_)&JQB8W_3Xbe=B3ABN$Y+orAdjUD1Uu*qvtE>YM;I8%2_tt%
zi1z#nJj0N9*M#iAGv*HRFd*01d4H;+r6Whb;=1ms2_AO!()A8g@<hR>1c3}J_jGcY
zcrW8hYW!aFm)1utA6+hH-#MU)36-tY3AP@McHHwL`CieaYg1TkUU5T5535Y^X5Y->
z#7n;G6BN$EE8zBM6Pl*%a694APS<(FUcZ{Md)UhG+4^*SbFvU`@AW8JXSwFkyhx`9
zD~*`aG^Yym1eX03%AZY}zS@+3QI$C~mfbyi)}^%NMq)&G|C_}2gKTkK^5(~~uHHny
zaq{L)xuv6Wj6Q!T4@UVQvC&g(<8=1>ogiM^9McSD#x7ckrw%*VRtTGGa#UvFNy(El
z6kQ%SFa2cnqsiYsf51!o*y)6g%CX1BZK4^rmrV`E48Q5U6up^UQdjaxf|#>FM@M8+
zzH+EFpiUI+W&Le5l*YBLAM@>HeSBf&&NWo@Knq^H!$z<R`8K!urAx7zlDC`rR*J*{
zhnAj3<aDXF?Y&YsQL?Zo&7@^b(mZFexuVCd<)t$jraqdeOy|6@QEheevT4+wp>@~i
z=fA*H4Y{Pz@N9XS0^wy!49Gn*mEs&{{yCG_=T~vc=s8Gjj+-sIA>5Bb;zyRxHyr<W
z$?fy=q1YI|$++u;?bxD|&G{a7`Dndcn<z_RCN($+WX6J>U6VF}(;LH0ZWrvb{rqth
z$gd`$F7$&SfE4T2f%0my6OuwoT<$lVIKTJje?-G3m``{u(Fb0PvZpMZQr7c#NPO%e
zs%T0}agyNzh(|P!EFn{u$>dkpj&uZXEx0cElc3g}SMVVGiOYj^;##6EIgr)zki_Sk
z()*AGK36f_n22}d+3MFA;SZK66`d~OQm~=$7QlL$${!HMIx9|vBb&w^@P#!+XE?v>
zSKrPZ5A1W`c{{wi9TVhNCYcaoH%C;{%<Dt4sw`)$W^vr*ts&;3GuC{lnpW`QS^ci(
z^0Dif$kJ+}h7u~n#hd=xi~J2?n@DL_d`8{)_fp=ctnP8_1O5SvgC62%^@YWrM$(u}
zXoaS^3_HBB4*2w?t!O!)36+RR2ssEQB7Up(x*j^-UFcrd!y%%Z(@1tvVi9i5iD6Sb
zp~obXQ-+SmP-U(md%mAZd!sC#_+9c!&CXL_wl^tr_EOuM)e})#1MTG_8>Wh*oDsQ2
zp7`-eyR3&hBUE)m*+UO72o(PM__Io0?FWRax{SEU_0G4$*HGb7Y)-b3b-S&l89wsM
zkFi5}TBw@Ae2K(Hx78auSxy*@8i|SCjkn|4RvD>FUY@X56`9_N=lG<gBKt{RSz(Xi
z+jZJLNO4W(S7;g5U+9_<hB64XT3mX6%kx43uxcwn#GNJKq77NK^5F)9#9+~JhHDky
z!?eWo#_TaS{s0>o*@I*k#Z;HrKW-jvpu3o_Q_K;M+@8xV@iNYy%xwon*Aa1YDyL)?
z;g31fgGA_bHbkhN25bb-uGeiOiVf<(x;Z>hdlouIYQfhUf%VZzDKMh~fg@VcJ32^8
z`vb_VF_F^;{{R|Ba02<dr<-Jc1@P-~g!rRScSzxuT@WWZ^nF$ujRE8DNLv)|QLt;?
zD+kR~>LKxtQTErkW>6Y!0Pg)gimQKemzaiXBw`c`t^}ybVl}Ze|2QaeYu5t~;eg`|
zAh$v?-`Uw=(DQ^9BC{4u1~9o@?R`b-mBLh77(BtD5UG0V6JcjMTkDd;fqoA`+3Xon
zcyaam$(xw8=iDQ=l~ms{@p-cC*jo|Stkp~VwBSCiQSKMUezZjwDD{8ajlcFjIb=Q^
zIkIJ<H@sW|w}vghx@&4Gw$rnzrj?~)eV{TXC{1m|im~nGV#W$ymkWl4+v@4Hwh^NG
z0=z4jveJ3I=C$e@FS!Y8ddCOf<X1P*S9wW>8wd~ZWHl!^?M$a8v~KNi72b?JcT_5W
zS`Z%)NmCm$e5E&9fUS4_qc_|^MdOT)Tx0chv4O`Xx)$F*UF5Man;xTmtyw5g`aV5>
zra4<jll-*}XIE|4bznKeY1AM^tI;*uE6UO?l{+u6kh(k3Sv{CsxXv4pofe*$z<rr@
znb4u44nstZ^p+TV>3<TP+`bicir}-CMJmOs(wSCNNB9*@HJ5UytgSB0SI#>TB;TJK
z>2K|enE&9a)U~RKS7)s%8NbS8PuErBUcy}hPnG(<1#ed#R&_Y0TkC>+zh{ZpM~=M{
zl6p+i)vhLRU#iu$uF^yV&~Fu<8Hs*PCKWUKOrH5Qwf<$GbPt2ccS!UC@%1@iOwvyR
z-Qx++d8KUK40L5YM#bu3Gzhe!IM3_<o5cT}fG0uxownln>{wb_+K(dAV-9uXdZ05M
zeYl7Oht@#gZvd0{x}WCJ4v?jn=le7<{&TIN;Vzha#`q4et5bw!oINC*LjeOr^Uukb
z{(@@%qi!cpo5{2O+LK(+(-e^ZjX~{h!$8(uN<UVh?5?5uVx>k~PA1;0|DJl1r@zs%
zeatL8?OXN0u6j^X?Q8x^ePL_sUnBW#@ON?_auhB_&?e^2cm!+azKp@lT+*N{;YiNr
z`eckHmG^0#dsRGo-%C%oO#}OSt<|QE$Bm13*>coNFmkxHTRB&GW^CsLBGXqZ+A~!k
zgU)osZG59@I@LA<(?bk$j-X4UjmKB=pn8baA{FJ#^)X+XS(rkb_hKA1cJF-mOTTt`
zgOcZBDo&<Dh$p3@L`|~USv3~hiW%6@GVhZ-pONU)H-b_N2?}bZGHf3q&UnAQM#{tv
z*hvKNviibQk3|rxEGJ35ojh#S@L~tw;k@$l9ax{$6{pZBYMj3_ym2AUSyP;YyjEj+
z%e{W$X2i%U9CzKIs5__#Cp^M=eSNqirj9;4(dm7MfR2x0G&#CN*OoJ@=fa%{1#}^5
zZTEIn<N#7sXUj&9l;xy8v9@y+DagL)qWMZPSVkt^x{$uv#I2{<0PelLcWhbb|M7Lz
zaZzqtTTw(sMI{9!43H9#ZY+?N?hq8DyBQFXlu+r`0|V0C%z&hVFv<W!Bh4^$!w}zo
zImhpu<MrPE_%Y1O%-VbPTF+C&@r7r?w92B#8d=nZ4dWlEbkb-2D;X8O^x5;WYqr9l
zDcXMx=+6THBQu8Iz3h95@jVj3d+h^dz?sB*56D2caR+Q~iJiFo@PF9#V+0Q&Xq5*+
zYbRvh%j(yu0dIYYwOQHlcBcGQ*FP9@=HMR7W%lZwu7<`#l&2?D1d95~BYMwS>PunL
zweGzyeWI${T%;v864NW=EFPTRds2}&!#iWXHs8jg*`dorNI2JBc*#^^k-OvO24lp1
zt3jPL5Q=8V?TkLtb{AnymWs7FlbNt+$`Z!OnNs@D&~}KV&%;u6d|*>s_1>gA>-|D~
zJ?|&klN!|G6wP{5EXg*ua$%?>kCE`Q)HGE7pz27gds}AZD@7Iw=P!#>As=FSGxpRl
zA2C6i<*ma9guQVUscCEC3s1eQS%%GEOCh){t;?sol+?2YmGpg@%jR4e>=CbXJ*xT=
zs_Q-?<Lzv@Tm0?esKqrh#{!HET?^tOUbIHd!&YNsUr_a#lR*m1Pymq{RWh-lBRm81
z&CX0QK_*SBb-Fq$>~1;<TPxwGENeb3DPhoEouZaSnbo+4@4F&vNhaAbphE_Kb|Wy|
zSP@YwZ=8{$JazcIP2Ao6WNrmh6GE(Db2(N8g&2!SIL&y>H~Y+_*ryyDa?k%aCqL&-
z{?QHdWxNhhDj0E(8-5FV@=kIer$9_uys1*0Wp-Qd%@$Zv-S`p+EQJwp=VrA3<sLmw
zPQol7_v={RV7YYnJtg?lB|_%`+FJzb=I1kXmZ|+P$9fZrv4bzNzHCQs7ZsJt>*p!)
zQpe^z<8%q)>n761>zns^sQ2sbf4Aq%PtM!Ay5-lU*|EZ-iQ-ejiP5R@yQy&KEV-VR
zt<Y?aywK7egQ4g-ry>D6gj-{2mwdZuG^!VgW%FOfKQd~@M7wA48tlXF>Mw;v7tAaP
zB*;2fAtf9+*d(GkQZv^F(A-0J+c841IKLdG$Pp#&=P5kVN*3qbE68F?4Lrg;l3h!1
zIF{tC-K#A%3|$AZHd>TZ<LqA;Zg*`&)k)>uMKx!F>F-cjmHi3@1A?euOtgHetUpi8
z8MRiy<9w(lX0q9KwJ5bMU=CfXSCm{D?m?#mqOqv`p&54_flbDUyY8coqx8{hCEFgc
zD;xe4o&ihP>viEQi*-?nVkIH%w?y8lOL?mAbEX?J76_c2-vLL~IZw%c3+XL_nLY>n
zj;}evkPOwC|9R49p;b(Rpt#?r5&Meu8kI_wwmcg(s5R_p<zYyl7r&w|Kt=j?5o8wM
zoNoI5?yUd1Ns^oZt^yIeZtj_WS}n`x<Hirfo$M}GkggKz3EiaP@MNX%%ey%+RjeY$
zLy<dAwBWj6KA*{&Mj>$vHd|^lKw`inkY|#RY%VtH)5Z)pOG!Jg3pSA+lHwQ3yfY`9
zO^V)L!9<?UV3V#>y*k;asOA>%DjV(;j8WWu*F4dhpfDjLL6n;CxO{w5MZ|f~c&e+C
z?c-CU))p1|J*^<!rE_+&Rf{1avD9Xc`PbD^mgQ_B)dsY0Sto{UZaB6nFqvF0i!x!(
zQ@0LBx;9zg;#dOaAj>n>h7(Ej@u2;?vu1f&UZ0`5Qux^R3MEYyC){esQZD*ZtTmNm
z{O$hd%5x=3O`>pv8a9&_F9|lpW^nMjcu{?SNI^EfKI(w9)!4MnP#}wADTbQ!z#WI1
z%y3Mp8_<|@rSmIIDNb!k(nL9aS^FXy)cPhjlTM}0Jyug=La&Z6eBVid@11Iyo`(HG
zvA3+o;=Wh(p=*;tBsB%W_`*oxEB+ZlBObF?RNzSGPKJzv38>$$DgD%Y-+RFLb3S<R
z5B>l9M=sivUABG~r1jJv98YAuWqg7F9K49h&-ecQhyN(x0Ol#CA+{GmXa0NG0j}h`
zJE6x<pJVVXreVZ(XqX(^NgS+N_%<1!oG+i2<yKLJ9JcCNh)LtGKqqQ6*F4Df;9kM1
zRcmYO$l3y%pt?UJz*|C{WoT|VWi>x7fDT2ND>c)E^fYxo6-HJX)HFL6L@(hw2Xgib
z0y4^t<-GA2Y>9W|-Li+WdHh%f!n!MC1L+TjyVQ33x1(sdR9Bw$6snM(;9YS2Zbbid
z$DaM(TLWQs2i1qRgXi4w;v6(NX1nK=9dRu5!e!hbJa_dO&ecyeptq2RWA59Kjd*+P
z<Hh;gNp1FWwAP~&2T&WOx>ICIsRE1q%zTC|FvfOA0d;lMs2_Ft5Qnt&o9x}bt(5*Y
zkr4ddt*i<It5!^5DYfccRo2{YZPH9e0Z+0btblbo!OJHD5#;Cyb2!1o-}UH=)*b!_
z95q=V-CiqiU2jCOYgFltJG}E9oK<^I{A_fMO?UF{dzpHv%z+m*?%v&pA<}anuQPg|
zmVAHGciffkW~j=a7-0X!b#qxjB_Ini#oFK4j8G=u-ODzI9Od6nen<<iOZ}t$E8iA`
zrvxoiHiC-r_iX^Fs?Z1F$3Xj87sR)J6gpwMs`&51zz1lB?7cnFH__kbTJ?6Ml@LY?
zfGQ>f>v<%vK<h*16h1(pmXQ|ZRbv3>wxE4lXjoL+PaZss9#kJkA`mZ?F3w)|Ef#*G
zgBLrPH#EPDizZc^QlEQPedj#gN@7^PWu2$>ckZdk*;|?Ww`IFajFVl}rOb&elKU&m
zBj>}E%HgFk&vkuw<)-{FcRPD>6v{>>R!lf>+s2E+$Zm}}GulMBlh=sGkhjIdnOm@l
z>Mkc0w~hoY4D8-%1Pjb%)JL|!<zA5Et_SHgQD=^b8VA*e%{T#@P<>HV`^qyB{ZHi;
z$PJ}gDaK6y7+Wms@8DEN;S&c9R0;&i4B@M#OKg_)t&y%BuYQz`sEfhEwhM>qb5b=>
z<PsDuYFr~uraJMfE1|)-yH9Uo{Hmp*ISe~1Y(Io%R>VFSG#EDQ76W8@eOd;F&j>oc
z<#tx}X6UYM;kt?T^;q&_b&7GHhQFyX##2^gPP}e^#B+^DDOId254L<heWlIgVE>(f
zs#01r-8W;(iI=$}mmlLj)_c<q%n`a)78`+GuR)Um>ZgCDng?+4<RgG2_7&8PINIfW
zO~H{+B+3lpiGv$&kdbpr5-Vr_i;+1YU9XW}p;Rt0dzZvQKKBrUje_Xp+`tkwE#2}9
zqSEnQ2($`<dO2#0)EhxnE29m{k>5=p`l_6qPgEhMEZQt4ttBMRF<S#j7@|pDt6%b%
zj-Y8EOrs}YpTc(PD!{uLi!%v7qvYPE+5p%r`#(#7(sUyPdsYY*49nV9K(UJI96`^h
z%&F@(wg08FEU9@0+Un-O1Ay&kM}299asee90MmP(B_4Ddv*XDXx-0HLJ;rr02yqfd
zImQ4%G-%>s0QBgSw-rwu*lJQ0UIY$Ms!&~p8sSX%t<L~Hu$1+p7&97O4KI=oi_(<s
z*)WFb<p@qcD0W74D-}%TY*?sDx9WR3lf1h(5Lg*!>$97S%kjv-AbU2twrdSpk-QXZ
z*TT#ODx5(nQ`gmHSG^b59^v%p3l1syd3UIM3>Nuv^8!L3*LKCJM<9!-e2#6(jL%)j
z#9nxMd^3sz6s6j3rDP{<Npj~;sjUZ8CsXY1PIz-QHRg}K9Usrh!hPovFgc)>S?H_0
zed<HoE#2mrw%A7n@xp|zx~t}v3fNrFtL~^Ta~xW4TsmI{SAGg6wD+hsNXVSm?eC}0
zvIhTjxYR(0spAWpM+w(Mz8v|r*n7@nIFz|8dG}3A>`6UT+`FY5wJ(C4OF}Wb@rz>b
zcH?JMEkvFfiMLKUv}ti{&;k6F<|hEjo)|D3p8>`okbL4rz2v3B84|W35m)UVwjoY{
zC%r6jxF(Tz3c#NA6@HhZDeF3DE`HWdotQDu<{|>E5(h#{!w`xN@|HO;ArWWNm#lXC
zwDtbWyh=U8#chDRbg^ouffC6jo740mkZwTdl_~u^Y>e3Oar0;JzTG%bc7hEFRsI%i
zkoEdw!*OwgLkcp9t5j;Da^=R&AsUzkXbb@?f!n}E5Pe|opaJyR288#)(r@NbjjP`1
z;>T@`J5+ZkZvNI%0{CB<9{|EwzkTmI-AYg|?hi@;+|{W2u<6quem}7QAY#g1OQ=!o
zcNd_Newt;jeFb8+$2xIBpPV)EV!a!@K1|7`wU}?on?gNBbgtANoYlLNmhU}MfNzwN
z=IdDi3RyM|C$n3V&(qlUO$#C|To5GH2D?`Z3o<3mhezV@ZWaTztBe@5y%>B&2cD0A
zNoasFJZ(<QiP&5qxVguEs)t%YwENAf@KSH>QUSMFYg;<(S=F!zXmojJAE?spIE;k%
ztURN355J$!UsF7_-^MZl#X$J%>HNUmFm9pC$*yc~^=wz1`K%x4`xLo%FqBO#9Va~~
z4(aF=n%dbZ<zpz|s1W)bCnqp>#)LR&v$}eu)gN5x#RV--tk`FVb~^jixdYYm36Jbo
z)gv2XLm0(VX|*vhj%OmADM?x(?eciiSoDHflK}akj7f4QHlL!#q6a^P-bLc@QjRj#
z(XhiTQDzbXfK{4*ao^kJ+x48!J}d0EE#0TuqK`&E)D4kd?l|Y~uL0y%xq8$5klRF<
zLt2jpBP4p*Lk3uEujb+y6pQ~ZQabgzK3hHKT3`h*Mz_RNjlmLUup$I5ubgo5B*V8+
z$YU-Yd=mL>9OZlbx*fPiQgl$&T-hU|37`;BC6{-o3W31?6EtV_W1T$Zw<fgfBn3Ma
z$>HXw6S6&rs7P<}1A7J0V7<0)qG}$Uj;?1s7ZF~)9C^t7MQsdL4O-23tq6+n9?*Yz
zcT?^Uw0xzs;3LWROOO}d(tqXmn`lAuJv9L_WelkZRir0_`h;qaMhni9Haghx-eQ<q
zr$}7`>^A8l1RyEI{1Kq)uD=4BB=$)$4dfP9XQGDZv#UEa&|$mx9=T5EgylGh>r|kP
z_nkUTbFdUf#`n3)R8wt{Z^L1aJx^d2=8R1|c)BV>;av*1(9fet%n|FUe%xh#nRqkF
z=65A5R@A{EH(e*P#*01(aLK!7cPBPAkMw4!Xb@90?We5fwNg@=Q*FSmZ<?J`z2-R7
zFUvm`s-G_k7gSKp*#9B)(Ie#b=neH=xkS0~B$bv(m7Zs}BwCvgjj?MJ^`a^;VFO2u
zt1uakPn9du=A)9lpvZ<!8_`;tZr}h}udNGf0CF9T&p$}jq)7~~p~)LP*JG@?;gy1{
zn|vdOzuU7q(QIpzHTO}p?o2*w+G<9Vm{RnbH{2VigG^zKp7Cr`#a}cOES7Hx_~^mY
z-Az<?rJ`?N%U{pOG`-yG6D|EyuSIPemv$umy7=3G`dg%0BHF~QS3OP5PaX(q4Lpuz
z^WF-Td%6F0v(Fpf(Xi;*w2Lh%`_8_UI>xu5BvVp|$F<+qtNa$t0ER5rd3?8%^&~t)
zY=3v8AnK<&Mqm)QzgV+2W#QEiX6!CVHdTW8e_klllX+S9yLTCVKJ(Gm?7aKu{;?g~
z^k<r6ggHreqd-v}5<dm&2K%xAUV49kXZ!cq%3N<w?q#n7*Lzdsf?@Aq8~=m5z<4u*
zICX?{p7-qW(_bllY4cD3%JV@b>1!p=FtZ#k{uqmkN(wDH8QJnFo1R0+ve>y=U7LBo
zfzOY2=w2C=Xupo=4&3(AudAwjlok@nuxUzdJ~g9N+;k%x!^%25lCGm~K0vQyDNtUp
zAC7kezSm#?87+(3Iam|d(ZSsoDY8^n#*XdQp80T?IJey^ZAEJkcX=Gm_u>Ac&~~X^
z4Ng17*W&T3%e4kIJ!|D&9@X-~n^CZ-K$Pkx>`HX!nE1{7x9j|GD-;`X)yxTt&djY{
z99)-ESYvFaSrmob%UKd7n3}pi*=NcWl=ig}#;Uy*L&UP{#Z6;91Gm)Rs=CN1y4TGc
z>pW?dDP5!(aW^H^L32cDXoa!){?{Nmi@eoJc^CFC`$8K$Va)>}?OC4H>mN!fJZV!+
zFXDOiuOn?V@^r{)t^tudg}u;!Ympk2&nIXU(8E(}bs7+b+)v(k5LauPmH_A~ga_3;
zwehLXK}dzlmxF}Ab#mNQ7_mh}@_VZWfr?Xb8qivGA>8)>HIOBeeQ(fU^8*R&`SL2S
z|EXKR7OqEuu|m&~)Ry0sx$E~XQswv%TaI1pHE{U@A?anAtJ<Hm2g=en_rQj_1sNb~
zc67|<ZOhdv`a|8N0I_L-fLb_QR3|IjFGxv@{nl~D(fC%Y{m{}c4@tdnE{PX2WE8ku
znkkm|Y#gk|T5h)+9(2jNdf0<Wod5`(DGv=+UR`l~0HO*2NQ2VKkq-sV_tE=Yn^~qr
z_KxMYU_$|mpxOu1LgcG?VXr+p{p>$9&WFe|BlvT_rD^J{nREuSFiPd|MwHpu&c}Dj
zVRlRH#d-{~iBi3hqWm$E1vKH#rqQDl;h#zc>=C8=eKV%4(rHNT^8EsgX-a}D-!Qod
z7%bT>h)Kw{EU&0A$h`5u*Z-;W^1J#hrpo>hYt+ptlawTOq-n00A*XbKevhhX^UZgq
znZ_F8A04#aFEmzGKaXM@6_^ljWQAj51g7|g7i#w-mq?ak=IYiHRy0=8NHHOCRRKSG
z+(@|lkBtEilw||!{tzLtN{SHm;z5eh!qips7Y82ST)I4vdg0Sm1t#V`sVdJh){N;(
zNA0gXV;e>F#n^Bz)Uo(nCsTYr>}bekGuqjk(Y5K7m$aFuICp=*K&PdG;$G*7kU56~
zkmk+}Bb10nr7Kf@N^`OtWvV+^CZvZpp39tul*D%?mjP~&TT2KurHQNk(j^L@3jgH8
zICl!<u;zZ3hzx$`)qsP2Z+m(O@avF39PsV*iFStoCI~Hyr>q0fAB6td)#km^Zb%oy
zq1!9F_Gjo1aQj2O8~(xr|5vMnfs2`7bDw;JBS+}U)j!(ibS)guqclpv0BQdiV5T__
z`L6q>(-6=wlwVv!Ll^^%5T@+W*y;g*>urEA0zaMubr2&i2(^Q1|5fboEaWkQ6H-3_
z_~j$Pu~(!G5Gz%{4Ae|C=>nVBt|Us{tC2MBP%pfZkW%lW#g@5FLg<KjW-<k|LwR1S
zv=U5eLooJNh@hv-UcQy1nq4}InBI!}rk0u@zo7=JlKGIC>lBP@D^Tw4-!--`t3J=X
zXsrQnao1Ber(;8`f0p6lpzRT^wpZv$={lp>-t3*I4&O?gtImX}Ykm$(&d#h|(n%H0
zn>o*+cWak1-FV)h-8DOgAX(>4%Y3U@$7?NJcG9doZcLUef!-GyFlq9&o(CJXS)?Td
zFImLlkrv&K<od-=?)9wg&S!_(8p72PQTjTLDqyvwWjN*3VmY_9mUDIu3CrNS=$a)g
zjxmgxS-OAKU3So0K6JxgpclbkvojIGt0!OD_wWNDDHZ{?d<B&Zm$b>=-7~&*<JsN_
z#UGN_<%Vg{#aP(24aRVcRr)h_0S>dO1?d!}tgFKY^EGv;0I2jpSbD3LX_I#ufBmDS
z6!&LyrPyGhC|CrVPNd+-92s5pt*D*ifQCRn-7vNRY%(oPyHHwbByj)`&LE7929ONQ
zDZK?tZKqyI?seKrCbcO1K1%B+Eks4`V80q|K3KEpJCzMuxu*e=g{@Zc)8O270JFyj
z;kS-i+j=I7Dlr`yF;If}A2&t*#}g7yt->&641ni@poQ(gL?K`4u!74pWMcbF6Rl4P
z3hRMtd;*9cZQuRi{9W`(WlG>&b@^sI|5rv7Q@L=8yATDt)<uv{Nb`D6!|zgqy8m=E
zsL`#3B)0q)GY^HdMp_Uev6g2{B~WbS%{L{N;s><Ui6#Xi*RdXX37;JklY8&*3zuz{
zh{F}OhH^1Zi5{lqGn#RgVuBs;-M1~EJSMD0^KAD#DNW3~<khaA)|Reg=&+p-m__BH
z_tb8du}en7wW7Bcms9!(OZcN(<^^`t?#oW?h%v75Kgx^XnR&VBN^kFsjf%7$h}*0|
zsceXftaZh%VHMwgTp4R;)wRgQd$8&|Vl;|_v(;sbx<$A;)o94f=3TS&+S=B(&gV?F
zMosNFh+99~qbe8m@rl=ZbK4*7zEZ!L_#t7z|MmEtn3;m9^+jWN_a!{MC1Ah_t=ok$
zv>L<|mCn1We>)$bC!tywrDP7;=Fd{VKIe!V?|H7_i)C$_YF_3kTdXs%po05g=7#Dv
z!Cbt6-ldp`Q~Xo?%3$tAOPT@JTCqr5@INFBVy@3OfcueuOoqE~TsIG+=UxH&|Gk>`
zrwRt=!lkv6iG58CK>)`EfEk@`XMFr?YXhq1e<LKQmCiYpjpb}Crj9|BcI_R)H4X>9
zR_+R`aVFzb1~<j1c!$hh;E*gL_hl5J-OpKQ79sq(Aou9W`STY@7uOUJA`AUAnCV2!
zVky^I4HKCF<)ApKpKg0r+*dhXZJu9dWXUCwYxA#P`^kZad|Igauv`mM+3XZqI#*u&
z!o=<jyH4q?zPZgZOH2TSOm27LiV;en&q!<xu{#~q9vPd0)-TXiEE_?GXzM&P=hL;g
z?zN?D{t#X=YkSk;lS(J0MU*vME-6M1?6Ye>od}MBWg~LLgRJUY6-gcTwNdhp_q1wn
zEus@H8W`SU^~f6^m~_fns6!Vi({m~w^VE~1A{BRw-;1*KY{wbQ#OP|8>_%g>QCDa*
z(6MVSxuK`~i_`9J6k!4*Sij!ko|5uIc6awVp1(O!CLWM{P~Ni~Bg)K{rp2PGQNI5f
zYoabSU8UQ({c5Y0E|sQn`ZX%yBHI|nX?<5vRRKt8d=?AAa$Uv4Z+BB1l-dQahdr%k
z_Gg{?(erw;G&Y8Nu-SWRdvZw?ZD6)76kT3mt)Odlpjg(w5c*un+Tc5{!>A!JS+T5+
ztyf)z<ZjIPy+B~fwrXoflSu}KZPRu*{MQlcTZ4<*#2bU{uiU0LzNU&>YKdX+Ump$u
z?LGAZBvdE9$)t_{fiO=_A}bM0BQQEWo)j1NT)uLUh#AyUuDlF7#eEWB*ag}D;Q47J
zcrN`w^vXL`yXNnlKaMYq5}&1;*(n~{`{tR>G&DnedjV?rP(1+yTM>3Mq{2Vn_7As~
zcE29EQTAMUH!SHpBbQ@3F=d0P>AfT{!{%_>j`_=n2mfOn;o=L)<Di6gMD_$M%H3;#
zx5-+25sV;rebigSl`h1anA{T*{QgH6pJ^VlYc+qSJduL6Vgr3oTELCIbld5vr|_kU
z*t9h3joRLt>31#QvPhul)*2=^HfwL&XJc&)$SX4IP9<-qqO@xTdz~tJGcr`n3v@Wb
zGx@KqCFjTPVqd#2SmtPw?qa@hk+Sw83#iLgD%eGYO9PzL9xz2^RAU%5%fXd08+s@K
zUlUg?U2-1D3#l853UE?b?})?I7dgebM^KwLD^WPbTd-GNbGU9>@IAX_rjjbvAayu1
zZ-J0_+|pE-6Dc*hF>^H*+ZRs$eZI?KB&m)FunLx=wbX=s3feYUP(<W%^RZS+qMF@;
zn=6yEB}!HH;?~zU(0KTu5^hQyg;+|56prjt;Z|6#-IDy5rK+x~8)lF(abmhyr@EjQ
z{wF5}WU0ZaK_=nfKYo=2ihAV-fU^C~i$CA;M(&tQyq-$bZ|h_y<m>qrlH(r1b59AT
z%j#=7I7IWm;0BK21{CmI9+rOtJ=vH*XrRzwW|-Yg%)+xVh4xByxe#dI3+zzGe49{k
z8o_O}Bz;_qy4azC^vrWL1550f3G1iThO&p=7OcS;`D|BTo3n9DfX(Vr&S}N){2JDh
z=gO>=3nmnTei$ABdWxp89zLAvsR_F39VEX-qUIqPt8R)JtSNz}M7-N=<=oegpnoZK
zrBxEszE%>W`86Tv0_XkwJvEb0)(D0Q?I7xE&anlyl<_%@jCZ0_GjXJeniCK8mnLkA
zpXdcuvcWck30J~w#~PgUazuT=%-y>eG*|3BQ<j{PD?^LZ7CEz%sx@~Us)Q$zDVvO&
z`mM2dHmT<Zn3wLtMuR4WRIEG)WQ$Tk9yv1C;~D_$Uq<-b6Mr8H7ccPC^a=SlF8=PE
zxOjZI56$<!HcQX3TS;ZtoS`7lzy9Vnc2-A~>VIt2gXS71P$d%E(aTqUJ$LA6gHGSd
zD*IS}n%Rh_Az5km;>R~2C1ZOP%<{X>@Yl7W(uatyhLOxid(b?W``jiJ(!vBFylp!$
zO(*T_z^_qu_KWP>P%ze$GOFi1WPQit+2kqDFH+H1srOIx79_7|lHuKTBEO6YXsV0f
zJU6GYH~k8mk?MTvrF!*Si&~acp}2~-7HyK&f>qCztwwD%o-Zhzm9UI-F&$QROjH}V
zV%V};OTt&A`*Gadd62}mwUb46!FhUBhg;L+Q!MEi)?UA>N(R#rGF&-i)Voo0n#pGZ
zWjup0%sz}FH$-LTg%?k4J<?}r$^0tEuMj4r?Rk)Zo%EY(m9}7OPKgP&me1^lDmz>6
zv1>*9r7$+~*rKG`%`Ok`6Z+0R-1vh>0=E>bF-@jDS^K3aAwrpAb&NZpDAoXq&DSv2
zXtfKF{170=`C`vIB;s;6ddXUtzX8etW}#SVwkGb_Pw)Ak3MpwM4*Kyh@muKM(mKta
zAUGI|hy73)tvr^9#Ga+~t(OE`Z+J;p`p*A>l_6ONnzo*G>nFXTgQt=tFeoTjJ#By+
zw$#!UgV)d+%WQC6QQ_F+{?IJG$i@n@Fsm8E#~soNbENOB>nvG33k?_Qy`bUIRxP#^
zONz}WN-GsYkm#h{DJwI&`xR0C@h;mgA67p*HBX`C+M?_X%&Ut>V06NRHKwSjTdA4V
z!Qn!1!bN8=dh^oP%;vURhl_1%oBYnb80{WLCtZKsCtP*r0LEEjf^lg1TM0F#$LO>B
z%CPyhZZ~DjWF@J6s@=f+^aR1%3cDv>V(fZR;ZJ-zX5CcXEZE%;PCj+<<;erGrQgwV
z83ufLvAb_#zI!a#S9L2^+dgb`6$nzvb7CVOET=6%<+o%;uLV{tq0v+5O@~xhq}b8q
z`)=3^F@c^Dru}KNf<5J>>mF0Y3n0VvlVjK?d17bs$?<<^#0yCjG(cfCYfkk$8~T<0
zI6h!v{NYMP*>u6ifvx%Xd%f^YAI>qxle?+MT;ClmUH=~ojmLbCLzRc)vp4jh-W1pD
zNrzHfL}yL&{6;~vjq5dxL4l^+$wXCui=E&S^yQd_)q6b15##QMIwnG_U60%Z5ENzP
ztLNBl+3#B-nRx3FOz`q#3rY;4&{Q{2ha5hTBLv%&8ADGFlp+$)9U7w1n;#=XV}~(k
z{~+d!GA*-ciB<JjK+1=M`S_N{Dw^x}FIF!a@Qbu&rsoMg{iwCaXEMF7wQLRm!uu%Y
zn@p6VwQp%uap;g?R5ACxn&XIEnl^Y*m7zVdtv)&}xH4Ttx3`WGHg={obIzR}xYB&H
zk$e%01Mwlj=#e4>>>2*B#Fg{$JB<zBdRujgnGVs(u$R=APANzJFbVrqXz_>CIb}-5
zHWLhAC0$xdF3oV##CHXU%liENqzX=y^uaIhrUkP}U#9;aq<n$<((y}=l^H&MHld2U
zaw-m+y`XDCxj*b#Q&XK;T|-y5lw^toL_j+=eDXaDx`zc(F`f6B55>j0M9d(%;B&sF
zW7>j#+{J$Mgo%CvYA22ppi?Cn{w9$BNeF)qDY@tGf{ktVlUVJP5=;@LMU<#G0KVAw
ztZ!^7vqoGWzW5g*>)%wqP7FqYo<xs||Kpjv`V8W{Wn|uYqpdx82eAzp=Ny}jiws<%
z1QbJoX8B=J;DKTs-(+ZqRg$)Nk#4k49kZ$L3V9bkijX2I>U(W?A}7F7;;}UmcMq}t
zXJa1MgHv_FNM09*d9}P4SKhFaloqFmDdAw;h`e!is{<+N^_7S=?y|Z8RF;%D37(jH
zNSa!GXZIjW7P0b2_m&5NR?gIvY=M85<t@v?H;;>DOQIX#ncRbGT7&j<T?uWDk<+Zv
zY!N~%+OyBEyMJe5MCp=8onLiV)L}Eo*q2YRjh2}aWXdS;pM3U>Rp5>eOrg2y?WFc`
z6BGA3IdiNff=eA;Hsd_y7&-H0BC|C`V~sCyN6R>W{qsbuCN6k~#=kC-7(;6K+F$%e
z5oT>Ob5L%T_IZwuCUv6NG)qr(%_L|XjYd1w=AZe0S<>ddAl&jJS4;U-_}`cb;D(A@
zGMHFIqE$??)`^n=_uf~-pWX_vdh-|6?yp3;@9_jKm9C>|VZXlan?m-ev1sEdvu2j6
zmBtx^l9Q+8>o05_Pkh`65lsL3oEOJ@doC3=!kmZ7WT98(F3zqWciU-HCywA_7rk)@
zcG?j8DSP9ML@2Lk!?P9h2zUA}qQn@wIIWQD@#0=Z8tu5-V%=P}oSd2iJ(1#6CA_uk
zzOnWmt<2eVm{itp3rcjk)a`s3*(4H4vUE0A*5;MyLTS)l-q9^8#o5e+Ytnfw1C-(_
znOfJ9b5wMMl*9$#Uc_73J;v|y<`Rdj3^Nv@TVfBA%_$9j+~AAok2L>G7ogKdik|zj
z<{?d75c|D>)DTVdM|jvxas^B>GJs_L$&<S7+c9!^vOc3|(IPeMl7oCI%xU%U_tg_B
zYene`92<-oVGVxja?Ws*+hOpiUThsvNpcdzm0a)ZV=K4B*-v}%(P)GnkXOCR$?Viu
zu_}IdewFi2K}r@j4!`~nxpLhgfZH-%`0|aCmMM`i_$g4m6URk{L?qw+gLD1oBaRM*
z{f1E>{@%Ss=wnEZo^iT7xb@~f|I>*!V}1(baDkA!oH<us*LCEZ3Vn0Y&`il9n-SvB
z2%$_4dsf}d%kM~y-j~c=?He->ojR8v{U%n<#8`=ZJb!64wY4KfcD#Biwe6ww=F~#i
z4g2cMv!D8aWN>ir=vFNS>uI!l2N$e|U6hR>xj@BgomVNh^09ia0ShCheL>bSX#-&p
zw55ONb0#lNgJVhU(F_wTxmC7HG_N$>@Oqm26ufc~i*xi6r8i7qrg_B1$f>#dwS9W3
zt1gYGO>HgOfCj$V6}JEvoJC(+F0p~z_sUICcUYueP*cfc%VUO=C1YWziVYWe5~7-(
zO_AJ_aKnzG7t6kl^7TY+Med7HBbX(P%1&l4g=*$d4&40hH9k%NyKVFdfnt8V^|!|T
zzv_V>-V2<36@?9R|3-YoX#J0e)EA%a&bVI`rqt&tM7i3h#bPy}R|N+L@tz$#asJv7
zH~A^O_hii5;7FL#I!*Kii0waH%G9q*iE(O@_`eu8-@~aG;og`ggij^kh&I-b9e<g*
z&))pnCOul|b$8@Sl<}iXXZjbp!t(NypAcxbg`%PEgSo)^eHr<!IJ2j`hg`hHbqN}r
z_1zZ!?m^_8!WPZ$T9rd8Cb_Yqa?X4rXQ(ACX?ZfkZYa8{Z%O7)-V{|WT{X*6f>*Qp
zXK=QqnCHh8BxoemseJ8UFXPe)LJ0H6Y&l&E?l<s&X?XHFsxLe@KQMLZJm87QO4-UH
zdo5N_H{fk`Jx?S9Ui4kgXY@WJg{Myb_pgT|E)=O^OJ+K*^JdJbp`aaF=;D79^VJzU
zJRHtJ)L9y%N7pj9PI^47f9YB9^_BNOWqhOuG0F@FF9$DdRr$8I$kkm;6{!;e&$xAF
zTS7+WA@#p5{on7^N`G;6@%72b<X<h+w+I66ADS!=bS1){3GM1}zRJ)Icj=aSQEx=7
zWk)~1!zT<Kr`9V;=Au_?f=fldH$Hke^tUOaCKxbi2lv8h(L(23_Ta3M=j^nGwd(9F
znjtj?A!|u3&YZ8LQo||r!!UG%ayTUur&dvT0q>I1_++ec<q}(y^QeT%!hP882P5ov
z8J#?hJmx$TZVX|}*K9Hx1X%$rm~kl6kh6#KDK%OwJX2%#Im^hVaf^6l1@BVw*ZV9=
z=DtQwrMvzeDj-nDu`9?K77*>G)&)3Ziz*3ac$B3kH(PQxo1q3RT`rm-L{r`M!%UBU
z8Noe~GbOxpi2Z8WlnM#|&zpKmrv4?Kp(CT{=P&;MmFdfI2l|#fv0oj@sq)u#KKon|
z(<6|Qdm$-sGFw-<S~*Sl|Nr#gKh$45+czfuD$?fj-~Z<5FY4{iMm3zckoiE|g1;)5
zi~zLGfN>h{i24~|{Q4Q<!QovAKGI(Xt_n%)-4{E4YQ!YN?4nnAVBO{%?v>BxAFc>Y
z^q3hbnzhk?*2#g#k2g~0>}NBRD2FXX)7iy~pW}uFS>&Yphi8zeaP)|Q`wR1`9YDeG
zt3{y;?Gu)ZG<g|elj?ygI$4JH@+A=-+^S`%C2J_x6mh2#G*NoS)U;)q1LqW5ZD*HI
zew5+l!~Sd?nP8se#C73bu6JMU<!R))T3T`n{Nl{KM16c?3_Q7)=F_YU^6E;6(pz+L
zresODIeMaF+grQ7_$knISm2AKJEROWcYTsoB;?HDhXO79e`<pz0b03D?(pK4j(}#C
z#Ph$r?ECfGp7jCG1%DCd&kJp|LpX0|=xxIBtME>0=rb(-$j7YNisgB<dZeFlIo$gE
zVD^-+IAvku|2Amzc1u4E+N_DMqA}%+{Rr_&7tM}Mnv_`BdWw`*N`Rb+lXSbsfyf4L
zzGvFw{F2H=_u=FwXU^uv{GL_8*~rMAA2|Hxl>4SwQfAmfK}NK39cF6^Mt>zMhK6N+
zcfimM*4vf6nspY=&HQ0KbK?0-bZi@(dqvK@&0abZqy7QWAje5|Z_n&%$GKu!v^|D6
z<@$A*)Qt9?NQ@P?0bVOa{A#0^3=Mq&(%U$;3#0ZS5$zbg#S)fE*FI(6yCxouE6AAM
zXu^ZYVtY5a$ufcu&OMkIHU#ILE=4FF98wHA{zGJTlJzfZGzoq+BSxoBju*H{M-qjT
zd@2;7GsaRLT&0Wwc;5sP<g-?C7jFM*^#A`@41iyrnX-@k`HP>voQsNpn|7$_6(ZKL
zW7_R<woSfQ7w{}xsUV2kaT>aQq?Y=ZZ#CcToT%0`39d-E-8`uH(h_r@(ar6H9yLZR
z3`;j?$-7EowBeznJ%Uo1U*?taLR@UGX-ez`E8AP<qe#(S=LCM8P|}oGxqE?uDtfD>
zN^oyZzC2){fnCBWuE8;m6{($?He9a#&7(4TvME{aRySd=sA=2WW%BKn;5j$hf>@n_
ztlq-xDcKb?8e=Du#!CY4o81}BE8Q1t-LVu4>o>@)(GEhToSf$D)eq7&-1Sj6AL@cM
z{k;!i0wc`!Tb&8k1=a7Qs|Gt6pbG5Fiw^t<!n2VNqer?}yVlu@>h4w(oP2Y?fxDHK
zG-bCtGXOA2p6d%bANY&ZW*;ydTpj?nAKucsW*CG@9=r+!A4SMs8LPj0+xhTde=LM8
z|8(m9yjkN364A%=hY%3>a_x*5=iVyk;p+PBVlN`~qBjqIe_XSVjEpFUIKdj{a!%cD
z&J%Qo!qvO$ju|cMTgegtqRAiNUb~DunIEj+)brk%8k<hEy#SzkKCMT7A#&NBakl&A
zl+UF}h8FYdoc_)qN&)ANAiH-tW>p59Y`qu3iJI|U-MS6kv7Q2NXTgb3o(2HSJ2=oU
z_OYgVqI0AF^}=AeHF0K=gRC1MQJ@*`+*<OQ5g9WC^bw8wTce{M97pTO6ZD5}g82E8
z8Bgm<Y3ORTvwxGxo$N>ym_$4x08?V_d($3|Ci0auBctR@s#^ov?1ys<^EiBDSP!Sw
z3^eIDOUVv}&sT_vr$pF-7;CiAq-VFaWz``qtw+@UtNHwL>3)oycx_y5dRqTie2Rve
zgA7{MenigxGrC%Cf^CasOAg1X4>BXU1^9GP%6g-Z`9aHBk~?yS^jPPt2yTuI_L&^y
zCP`Edk}8C|CvqE0G?(A29-3uvD|Xt_>EaBc8|Aq)tWNib+CR>unw6`Ncu^VqiHQE3
z&xWQ?&#uti$FznVgN<x}yhED0-IRYV-}D(oE!C)rT#a_2q=P6LG04rbC;Ph_3#E~Y
z!YraQLZ~wk@ogi7iX>YHSiKsvGE=Dkc<uMgwW7M<+r^0tG?^rhc-PS&<#3Uquu#(Z
zb!Wp4p1&2EG+dVUS(gTlYQ@t>3`c<123hDh(}}!_l5dsRszP-n1)Qhhe}$<#*obc4
zj8D`pw>o3%flh2_38OK4$4b1h@PPUEcDwL~udlwHVU5c|5`(6`v$`ea_9bcBO9lX_
z`&AvjrjA%uRn>8NsfPx_mVoG*2Zucu9aAAhszD4^M(>DmW>oP2Pyo<vE)EYOUGDv~
zFn@)h=gK9mg+~%otInVD+zbmp<{N#Hs*iAsDsvCOhTCv(p~!2*0+M#XRxpLZb^~RA
zNw^TxUgy%D;x@oHM_gXiH^07sQ%&9NCdBRUq#o|LqyX|K!6rj-!GF8^pECXNgYXo8
zfalQI0ZrWsO<pQAp*;3Z?)Di7F1Zn+aQI^Y5U@=w=er_uHc06OJs>`Xl1^bZtC89&
zuq1dHz#;d#2ml_0!tmW{o8fA^$+yx0&o4(<=2vBqk0SKnl0sKymc_Tr?*fTX`YuC{
z4i+2XKcjD2_e_cY4{B)xw`&#$-`e*+q{45@Yibl(kqxl%#)^F{)@Y}95*E_V6X9Q9
zZtMwE$-r<c?HYBSFz>0J+wk%#TbFr{CS3EL+Gm&O<1OF)tlXx$D!)daiqJvf--tir
z%bSr@%68%AK*i1k*GzO>FJo=KclBysLaT6B)fJj{99`;XG^iQ?D@&A^aXah+ofb8&
z1J{O+I?{mfs{SXQ*!wCpsy!ZUu(Rk7q+&Pt29Ct|p7r7(_V)guUI$;TCVY#(G3UwT
z$<yrW3_h*M?gWoYJv}2pnL0`f0o2%S^k`NX$F~yBRfSVoOgR^L>IX5J9W?rElR)Gc
zOJCmb0I~~`f%{rt8<osv5P&ppz*+FWWan_=bImXDRw<dHQ!HZ<fZg(bsLw$>%8Mz<
zeLCK;Ma*aRLtq1Bbpj!N+OgTR2pf(8h~1$iN4IQEzsALXo5#^=xY~Kz>63)n(!eOf
z(=|xq!asERAGbtM(AW1oOUcFd+1cl8E}o&I-lMRsGKyZ4&+$IHnLeyu{=%;+N%tCv
zIj#1a1ZM$%m<?py8o~4mg@uLoVkU`*zPyAfXJu<ANa=WT4ePwW?}ys2n$Yg~3AMDg
zHlATl`j%}B;HN&2+59k?({y=8*wWH+Z_oZ*h9m%&rF$e6m&VTuKHPfdi$lYLgWBe;
zD(t1vB}D}V_?LYABL?sA(4}DGzF^UH4f^vBG<69Ysm0UDx=?cCFOJa^fC#jjk*kz=
zv1EncdVmaqm8tMHj*Ho5=sDavJTTk_JUcVv1j`T$(yQ&Y!MgpyF+c=rpECpo_V=)Y
z&)K&8eVsmd8XFsLzL^$NT5Cz(lM@0SN~zTQQc+Yf&ohY4bQ$0>d*-!U50-b_z(*eQ
zr6cPrd~!cXB7B6jugX!nH%sb?SZ4gI1Ui-`_tb+x&K$Nd+{p7u`v#F6=EDB3hso^w
zf0RX5x3sKwMnq_W4IK?c_z4=0CPF2Xz)Vb$UY3&o@)S4jr8KoCA`+yEU8Q<YsP8fE
z?K#cltVOpnpl~+luQ7MFXxw=O0)pn!r#2znL#=CyQ=ppVKQiZ>64PSEi4aE%E)gBe
zc7|_&g|BKT0~b|o-k$HN6M#De(iRKTNM3N;UNSyOv6k1}+|9=MGoq4QI;}SPX-lyV
zWIUkFtLN6PJ90U{cD1e7?;&6c*$PA}Nr*KK-@>jeH<AQGb3jL!8$p0)VY8xwUMgXF
zVbi&~5yo8;g08iv{q?396%vl;AH2OJ;dmvThUgT&$H)ojQ7j0m+@DVPsiTB(xzO-P
z*?1)?l5Ys$A&gyJa+9_k6I+a_0|XN@?cU$Ex~?u_ukCuBZz{OkDb8Mr>a|n+#ikww
zjK1(QYMKVI8m;xw)nL4(5uiCxh;IPtq5xJr<FJDG=!=Q*4TyPb8KN`TJ6JmOcAKCX
ztz`dT+NI_AI+#9g`YKBcS)6UuA}F5z>tF0MU2Oam7BT24gsBY+D~fNI5<X6l3{uj=
zte(Z;n#NJU-5mf>9|iXsCH5L$lJJDAwD?R2cwf)et;knrTC_`rH9g#=n%IYsjkept
zh3viWY&t5@x5j$`f3S}i$-mPfA|%vr-`{@)Vrl1`<rVS4dbteYAuHV+w*v<o$Epm-
z+Y=}x3{SD~I(vjR%-||){X5vZYbIh}8Ym-{Pd#nbS7`GoB_?H1bMLAZH3pH=`rU@?
zWlN^v7mjB9ZRb0d`O8(AYM9BVd=vLc)C){>P5evWzO~5R+9sb^|42zKA35h<RGJW8
z0F^9^s7u852|d`9K8w6<c1CFn&uN-_W!=raRmSQb)73>%IYH_!jTXA?TuVkO0aJ)>
ziykkF9Ulg<M`Fj`(tYXzvGUj!gQXh=o>@cr`kuv$v0cUTmo@YlkgeW;OibN;3$PX1
zKN@|ld8cyL-FKU7NU*}54k8CjSt_{D$qBg`#X!w3hZ5pGSqF8AHa~EOmz-@+T2-56
zF%dMfb9ECht|+JseWIi6pRn2VAK_|#%wD(3R(H8m;&89GQN^sGvMG5U!Z>JzB7|iD
zpEZH#hZ7NidkR|SB7yw*U=K#S`Uf4QpomykEqr@Dxx37MqD8UTIf+?Z&^Eht7j1w3
zPJX#$Qe)+@r?)jdnPM_wYt^Fp7N9IuZ+x;=Qzx`{ZknS*U7ihIV53xqq&mLjN||=k
zmh)_rVa(u|_}15Snq$+Q*b6}6rYWe}2)7_Al)Q(V5YY!j%I%2TTki~JPU+Km>I2AF
zu3V7VTW0s!J3m_oY#4H54bT^4kuSkqs1k@dYq|D%ZTB@G`=jBKbq1Ti?(eP5Rg=tX
ztAm-Z?ffs5Fdp9z+IB#d>*ygwl-_%ISL9kUp))X<-5y33LmdZT2G#o09?unTY^-+{
zdE-7ngh-J&%YCkq61nUTEnt<RFKj-=_2%V0CgCw;t^-?g(1oq>-ro&xZ+rAFO@VB=
z<PewE*@fQJG^}?4`c2TV?Ojl}KId02$yqrHuzJRzDB1GuKE%cLa_!1QaBK3M{adly
z;hfGhTM&y^ub;{U=Db-r{oW5Eo=p86o`8~I`^hB;Lj@+{GCx~6((g_J!V!(H3*p4=
zNB<wj2m-);Z3z;gr(=(eR8`&2Ewh(mbA3h!!kjK*5PQd&BGCFdVyjDLH3t_ITUxTY
zwV0V!JwV!0T2wR_CoQ!q-Xon+OqX8NR&ow~Cb{W2M+>d(bXmvVJJ_^h){s;Ba{S<e
z1dWW|mPZOa*>zKf8OC`fozwrFnFZaTLfwcsw{kKnjyZWEE~eF>pwmo9iB9#%oaghJ
zdNQvJm{etDf$VJoZa-pD)ipX5xA+ee^Im}Pl*^}mh&q`q#e=$_miIhFKimMW2W2N<
zhF%66Yt>QDl+Ep1w}DT+2m#i*RP~b(Su@j3egOgMcLTdR!1N@NzW^~q&%nN#(4oa$
ze-nDTlQtPeRH45YVbx^C!Hpz^V$W@^?j4FjOr9>K!NF>5g-Zzltwk+9CmQg2gGFph
z_V%}`0tVmySbd?0*Si=4PwSlgN=q<-zhtxEky@}?d0jl9GF~Vgum#1d`dt3U_xk&0
z)ZaaQZvG`+kICL%EFd)a^+G#&jp*(5QLnX*h1*Td`%_wWJ`f4r*?68c@->K<khmqd
z4iF6v>{DO8dY(De%K<-d=BLy8E4Jn8BiQg@*&F!f8-);VfuK2q{5DfSm^;8flt`7<
zTx=%XQt}!n&0FT>exTr0k9DA>;kP<{gu@DkWp#R53AZX=(A}Sm3SR;wpJO?O`xf|K
zh)cfEexhN;YmhMMA61EeIEVTqkURbWW{Y~FvUPYKBK^}P;Z9cl&@KRx7XzNtSf{F9
z2j{v?%Q^}qU?NhI2(rv3zc+(A+(R|`9H0(`L_xrjO9sFdX&N3<>UUo#;kk3C%R|!Y
zX5>-rn(Dc)e$D2q?e+n;re)))z>m`;<jD!y;t!m5Q{U{}5`R=y?U@}1uWZa77?}@z
zgW}cW6x}J&u(sJ9o9Xc|?kC!!=z15J7alfs3bebb0^bp3#n9gw7$cQ6JH?Mm1pzP`
zR~3?R+i^<w7J?VKemtuvygPgSMR!-D5_3yV3w(!dgWLHz?>ou{rw?6?kI;cJG414Y
zvU8k0$p^{7`-+2X(XUA!h51NN;7i3kH%mt=qFXq^e&!#b<upPD-7>IFYu<maWxY#{
z&SUn&*g9H?#0OvDGdS1VNFqTy{s;cMH2u|ZiGytool=0|GMFhbE4KRKFuK6&*%A4q
z-2LyAzr?XdR|v;B$7?p)x1Y^tWCQc}4FjD9Qg5Z-)4p6uWTCY1S$mA#N;73(Af3m@
zvcdz~!%bVpgQ)OIwgV_Ql&eVp)z$x*7uH{fG!hm}2h(L2kJ<pS4+#Y3R0))<5%}wp
zq9!gL?rw#l{eP<Q(RB|{dQZk=vrX1^-Bi`(t?Y~X@ssS8oSiwDj^K(3AXmcj+!l`p
zslmQhuW-Z%6DYANbYGiJS*+`XK1_Yv*(SH|Smkh>K}V0<jeFX7DKWCkG%|%-yJ{kb
z*O}cz=cN}rT+8jMRFalgf@?|4Ls#CutzGost98pU*|-6kbS8BuftT@~CHIjXsr-&x
z^{H?NwBh9f(*&gKT1N<%0Gq9WO>al9ITkmQb@yvrp?vFg6i%_jq-rIY)xrEgc7iRl
znc9pVl;|AHLpQXl<B<U-Iwd-qY1#{}IeJ9q@H(HO+<#>4z6TVAKSoB^H0NPn-!z7r
zZJ0gZN?-f8wo>VUq`onb;`$ZA#FI@oLGSVM=ZZ=8ENcIMWF-Hz9rbp^pk6(+X1o4N
zs$T!*bQ9Io#KfjSC*QD3CF<t&<YT_gr?u=1Lt|=>{?@3TP}Uxfu3QWgJ?C*+rgHPM
z8i`EHJw2Qh<Gec}g=w1pERFw5P)*U{h+wNTRT%kBw5f*I4&f9R)(IQD+A5$@kkLb2
z)Fx+-GU@jZS4rC7rSQxOS4rWWcFWRZ;aKoIiiDV<NJxA(cA9BH)0{m_%9$>~AERBR
zx8dAlUuC+%*BzJLON!Ua&RKV5yP9fwba@Le6YPkjM?Wu<7%)Q9ucgPH^w5V|E2uXI
z|8Wzm86GDdh}mN)NA*8?Fl2v}W*`DGy`WNmD0AWRWBXr};is{^2nx?z!vB=ykt&|^
zl|0wPq_{Igdevk|1jyX35H#~iG#<MjJ%{TDf%!a^H^Y<Go7sN2<l8)YD#0Jc8z#l)
zVJa3J?fpXJ@SVYR>;r7$<hC2%flIEXad84al*_nFNo514rXwc91`}t4Hy4*gm{J#e
zZ+2+UfL>vVjjPvyxu^Lk8VAN~GZV_QI~Bh0@I^qq_Bz3{AMhn^*eSOKcYX7u4b!xN
zrM^t5%LRz&m?kSFT(Emr9S?WM-JjnK=!P9QX8Le<e%W8Yyky~P&qd1PiB;A^jFRsi
zeZAO}W}Z#m_=lzb+aLiUdb1jR8kY0@*YEzmhfa4TAI0Bh$bR{!<RT?m#GP64J*nAh
zl~0cwSWXs|-M?IdQ2xxL5_FaBURl1%*EO9yYI^a`!O5r2ho83H%2MLYta1}Fqxj#g
zi%e&ccNy>Alg?T)k5@?nmq`U#n&0I-u+_CKY|rAJF%938E9yfFi9gO4HiTp3vhqz=
zt1ZarZaogcrGVUSA9gED`p-_<EV*wN-roQ_Xt9;Hn{Qm^nFb>`xvS1_xZYQA{k!e2
zLiy|K6^4M5*1Tkh{zZ}gdYMZQyi8<vJM7i(#i5}rd2UO+3B&tug@$gJum4G#s;QV|
zhzS_L|DGQMto2sUB%mD<5Ykem?0hu~Xaz)~#Oh<>#|hYO7=42IzvsJubfBlS->?7p
zynYuNH^7!IC$(LqR?pHgVCO|M7ng^q)m37i%d4{`AgQl>q4zH{RVecfo@C4}dRJ6b
ztEZ!h&{4=)=(_*KqQ|Cu)p<2_eJry>>fl)4x^GExH_ckqZqz1lX59?E(K^whs%u)e
zVGI`6-LE#$Vo%7Cq{AH}PYBTLOr|?Wukup6>%%b~^MOANS|Kd~iRiNA_WtPM{&L*`
z=I|_u#GYMI@XE@{^60^!4>6QDKo!5&?l&$o6gm*>WClfB>dO62x@0HE^8v+i1wb-w
zlt^+c5~Fxcykf5`KD`g(al?-?mw{ARfK+almHlQ)$LP3kR#w(>wkCqL>ETxV1lZdq
z2D?a}4HElkiL!?F+AYM=d5JY12_DCqvc3a|FuJK2q~ID#psGLDZyEpVffl|bDEoHK
zdzxSCQ<m2IERazYmKBOeE<h_pVCScz0SC6AmUGR-a>y(8UlQLwjnQ*d5M2ApV^-`B
zAPev8eWKsHdlZPVaWVwp+Y&9Io(l->{Uxb!yP#d}#JdO*N~=}gUYFc-<sie9R#GJs
z&sM4*?vxfXg{xn_h>mP+yQT08>kJF270lqN=E^KmYBwBYOTK*vfihz|ue^A@X>Z;T
zy!*AsyF6ax_wTu@Z;p<z&zpyMHr45U&ll0W?;kUw4LIG0pLon@0=8XCPQ=+GM_$ix
zwtKaL*iZ25%`g^s-O6;~YV}z_SZl563<#-@=qjHq43sUD#z%sd#;e!2t8eK!_oy!n
zFGIOT<KW0<c|7~U4M`(^!c`?1wxQ!4|6!C*=Z4k~)zmC_Merl+n&Rd$hCZ+wwz>mc
z9UYWWAr+PQt@<u@g2IN&DukL1G5l?offA{P1~`;956gc<Mn4q_1d#&6PO5zswubnF
zl~(#u@l@PodZc3ZP{;>U(1gql2L+QJdkpK(H~f9BzVb!@_%XV)pF}ZlD9#y>7r1_{
zKFhf1;R>qe!UV>Z1>rRsM~Pwmw=?r#@PcM-x&v2IDpf+QTcZ&qN7r?10$ny~U%dt+
z43nZ$6${%(0$tnh`?RvMhif)5=_&pGea@L(lXJa9Lk`~)q)jObs!I=J2rXI6KI5ad
z{?xJ*a76dVa06klKA^Scv9$t`9iwgqKd}Jf7K@aymaG|YFvEq4L;W=9TIRmjBSi0j
zI<pYO?wOZSx9ra;FVXmsdw=8(jDF`43V_F>$~ZyIumR$Y{Y#8Si*;FSgCu$Xc{CTP
zJbbrb>_ByY>p*C6RL^b9e*w2X*+_q2w+)yh1{K8t&h&KH&PeC%t7E>;E*1Vjvgfk?
zX2ztD*B2`4T!30WHOot&0{^P=%>e|FP&Ozw5=0^mYor;u2vr(fkf4sSG6wixwgxbY
zU^v13IPoHzZT`dGSYyYtRzsECO^@$z`s}pdA6lV{fT|IU;*!pwO~N=LvDs(Rn0kOs
zEvNheMX$?H1$ZpzdU#xkfdF{2_i^r_*ikLwKP^o?EkMkz5vhr6bw+IwO)B?hPB>u_
zqO03h&})LYjvIoz+inK<rl%u;-){tX0@zCzCSr_uvc_qg0oJwlQFYX?#2e-_9(1I{
z=Ur4Zvy{x$;VbY{MZ{U^UrL5D=Z4a`drgMJ?s;vRrU@IWhI!>vtGer8%P0r!{;aos
zqWGvFAv0AZM{UfrXIDLRUyik=%irS3pi?AG?6-O3%_wc;HGAcK6MDRs+l<KQl@VZJ
z%4r5_Ty61Q)02>S@!2%<;kQlNim!!~&K)Qnefwa_z<(ApG#)b8JPsl$Hf&QmSl&^;
za*rsCztK;Q)&%)x_^xAn%oIq~oGhjB^Ua%8)SjdG5z2DgQ9ALpc+}VEhoDvmTu&hW
z>|M)%zcuGiT`C+P=&-|BxIQ5q7`>Xnc*rz$?>y%J$a~Xxth(rLyhNFbQjv%fO2&{7
zu5w3am3c0iMdr+1sZ^#?$gE@@GtU$yndi)zr_7nB|2mf;O26mD^Wy(}p67mb?p)`b
zefC~^?X|w^yVm+tWbQ3^^jq4gSplRFP{&ev2jR_|`y8sTCp0P(-LQwA9Tbv~Cpu{S
z--SZxMc<;N;W(S!(e}*>=_2Ek(?i?IveLzZ=@fcoFhU5I_$$Oj`{{1{HZeB6Nk+vV
z;5p-lRk4KOQ%$`ombvoL4p2$l1c&LE7DY|*84TtMApmV&rl3o(AMP#(Kmur-!q~*g
zXcP1b^rqkMVgFOJFn`f%HQuuW2R@btP86+x)Ia@y2xp(h0{NGmTk03+sMfXm`TgJp
z<M!5yMI^Lje?p2mYv7&X2YbsWHzm1gk;1Z_P)7YOA1bw4_mM&|(#oN!37B(<R}gyw
zSBZU14pBN-<=WE+*#`yLG?~hKc1OjfXis<g)@CN{t<563a}A=}C&eYhEu2~T)kgXi
zCW-46tmZ$)DW>y?>h`pK`{p&-!&mwIvC_14SRA2Vhgi4R{@WitW#4{gJR9tvbF<vg
zy?c*#%>45NpB_#xPUa1xt%?TAWMVm~l-Xdm4{;`0Y^q(F8_fcjKCvuanUh)_6i7YV
zuNjo_eYI`OforL(a;Z$^>U$WyI99#X^U}Jf`a6i3ddM;T+Ll90Wm<{N?B<YgE}unA
zTEoj1G!@V<lhh6-j+X%`C}dZlz?|!jJ@VHcN1r@+AJGhI3pGBL9L<nukv7!^O*lrT
z?kCwV&U~A;Bb~Svo$P@mb0)I{{xzAu<r6JSIzTPn7Cmdi$t&$9m);X(erRvF%se!7
zRuox*x-in2L1VXPN87wwJZ$Ph85)^ELmWeN&*r#71kr)KV+lR2Io5)`JZc>HA94&t
zc1;PLWVI&u$&r;0U5JW)tuxYm=@6(2&b#I*N>Z`5=BK4%XqTfc$bKX=85S|L2q(|o
z%Jl0;_2~;hP}%-@Z%v-h7a#H>(<oVTO};+8(|v4A<?>}W&-mmWom3b<`92})d(fIK
zMVNq^kSH2Al@ZEYxPZuUL~sH-<z#<^@eHOMPGjz<g_l}|g~>>2qnr^Rfc0@I+0-Zv
zEo`2pr|IVdtMUIj@Cwf37jbF{m>p-Nv8R_!P|xST1X;<(I&K7`wlhEV)GI9xkq#h~
zfteAq@IHWfjn#QiqBMHJ2~P+}_1<9vmYur?7+cC%2cT^jRKj+4^lpTgB&OWdx%iBC
z_QMD;M|LdAq4tA)Er3v^;5^tXTFdSbh5*k4N!;z!ld@IHE+EaH`;6eHk0^O9Q);y(
zPe5bm@rLv~&KzjO>?eX>e8Sdn?v=p&tM!u&>};>;hBZ^4I`JABBJo~c-;=cLy&J?I
zz5*6vJCU7Lh7yOzd|>e?#qm<BeG-x~vrMnMd>e(Mt#6l#;LhsnIM~R#SE`jOeY+W*
z{W?MEW~S8op2~#<1G;iaxu!m6yFIx1y?$=dCWF(548T^vI7<7l{YvsPRkR6YIy}nq
zl0ExFLqj`h;2?g=_8kG{<!TWshoBv<zU#z_SpF}uTfmAys?ZQ%nuH*T#;97Z8WAS6
z46*hWJ>)!8L}YR}j`Od*vKwm(|57KO6fP2Jzh7>kMuG&M<l2Qo560&$3^hnajU!$J
zU}W<Goum$(q@w9z2NCU4&vf}*U8MF7i%I-O`3D$I`Xd^S$RU7q59elwm^CNHa-Hkr
z%Gdi~FG<UV8ykS^V4XPH9Y#e-P#3PbccssGMSHj(Nwi1G%P9*i&WsS42-_}k|Edz?
zjh7O_!uD1NlWy|3!MaKsNJ7y(YoUxukSLA`u&)cxH}yZ^cICws5@M;i@`2UU`5uu$
zgA9pqarAUin((VE#A$Y+ddI`f^XzlGNYD*iz2Mu9YuoGfm+O<_-KoW<ZbJ2MeO1yY
zIbzK0LR5l|)>5wf8R*L--*c?Z?l&X{)h`@N`EZN~efiKhp4JQPT7{iv2KH4wvJ1#_
zpZu-(kQ+2!z@Ld~49N-p+I1MM6zd08sBCB>>&|~B^nBnPeqi%<vO|dOsT9hxo>~x9
z%Jm*Qr+XoWh{p}bujG}G7}1ft@PEHtd+8cwhN-9$)PRn;&&~aG6OWy?er&wy`(<}7
zG5dtW`n@7IuUuO^Z1&!s#f^pYX|HKaD|01Dsce6^q77<D<{?*`+F&a2V&_)8qd*@4
zgZ5k8{R@kh6AZ~Dchg6BL4`Cgj7i9YJ{&*0^XCu`_<@9|hRRgm!0u}7)X7`nXJb1^
z?)K$I^!0FIY$0KyB^IqacTS5-O3qwMFb`G6GI1z3Nav7`2=ZO+_|j0`J5+|2%VuHK
zP3ZN_kI>ZAJh@^!G@y<KH!^l=VtKS|G3;mfqdej_^R|i!jmyf-r0f?Q>2+3uc^AzF
zmuUC)8l4!uD=lO>-mPIa{zRV|VZt>+u1^sWb=dtBzS_%_g9_9R#J=;aH6|yI*{V@9
zuMjqWlB*15(cHJsgxSA*h~;iX_~hqrN^ySlhqyechf=TnoE^mNZZZ^?l<N?Zkn9<8
zbL*vQ9-7uys<CUR5WjLr-(jJn%$L_=Nm10cXB3@`k;saaF^plTLLQt;=8}7zQ`bwZ
zMs!L(+JIQSv^q3J;dm|OGL5v7Li)$|j(x^&gKYA%Ur0N)QRU&xuJ}>qf_06`$CD8_
zJEJ@{Z54t&GuC13Y#Usu+d4=hU{_?!ATA*>J+EsnaOA03mT^ZON0UV`UXgl0ktT7!
zsUf5K_-OKAbsH@%VY2}@bIlh@p9%YyZn$FiFaNs#@?t_e&Gg)9rYb%NyX_jPZDHqa
zTTEGW8MRFQ)iQgv@q7r*cfx@p@}p?4BVg+%wJV8XYv+Hk{(;_Sw@busVdVVic~^)d
z7TyVv5_TjoY)Okd{{SJ>{A@~&@a%h+kKNT#>5DUaIiOVVK+|`1l6UE<2gy<zCxKw3
zwCc?ds$kE&%ZGSzn_!t63o}>k91nNUb1%qKJBT4|x8f?9oMu*oR|37lX6?XFi~E!;
z;KQ7o=D^_VSQ{H`lb>U+m7J5OPtQt$b;6H_V#@IKPACGna%GSJ!or?7w6=B8UI)Kt
zPOXRv`g0jasua82ARO0DCVx)nhC@}#={eWMc6WhbkC*1{^Bu*k{6Tm8n6FbsjRq(U
z#b<qyIZ>2**d@27WY7h(y1tmzwJ090V*m!@)cd}DHPH@XnWFs{y*G%ME{DH%TwzSY
zYC>MF!H*uI_n$Y-!tmR68f_T@Xh@QS!?JZ#ZW=bIey|>S(>?zn=WYJD>?|M)0USo_
z|FFz>Z!9-EY1FtWiCHYBp4=-%h^fOvI-&ogWUa!OK|{|(bMcpG2(Sr;DiY3$PQ|$>
zsTuN&TDLo=+&L?yByU4G4Em!M(W>&#EBPtI_>US>xd~14wCf(d-6bU<yDU1@7+<$7
zwqzqiy^38IN?Fge!|8y0a#blT$L!?Kf#ZKFj}QZ_2?O)T2{jgR4o46V6_kkDo}H!%
zzAuCTrXu@6B>q2ML;c>haedxcIl#!yYowMq$A_Ep+S`HBcBFYoP=ID=?tXjwS6$Ay
zan}a-zECHf$fLt&Q+;Nl>HuE6;iSNvDw;sl6GdL;_50+&jCp__y;LhHFs-J(^J(w^
z<pElB?B%nn{X36tz!=W;Wj;q6so(hQ<|4xd-zn-RwjSb-HWsp!1+$PsX@-i*A~E$-
z%edv*p`P?=9OS8@&EFMH6x!cDFLnHQ4DgUs<fAf5Ng<Jh|1(`JEF2f7ETtHu&lf`}
z&7R^*H*aQ|=o#HsL%?ukE%?xV=CWe;2gGM>s<SzJ#|&-K6KsSYpvLfF{B4=l?0=3Q
z)rhVC_E?t(#z^%>u(HXf=)bVCDt0Urt8B_qGb!QnqqCVN2(GM&URundB=uTDy*5Sm
z_r)6|yGrQOI+MRZ8vi}p7k@W9fh$0VxKf||eb_o+mh+XhYIk>iiVa_fsBE*U*Qk(K
zf@ZL&o}Mb^i?b&4`j-Rbo+_Rb)|Z_WFF$uAW9vRbQ4#b6doJ8(&Dz2I#H@sz7<`d_
ziGk17)%Y!edqv^TD2a%Vzem0G`33rTPdoFJCN~mqY_5bf#pY4!b4)4xobFqRO(@sq
z47Wb(u+ZbD-}b??x>`Sz@$iy)kP`WQK7-t;M=9|bwKB49bY^nIRZ~1G@uJ~OXf<C4
zei%bhc2qa)S(04cqWoZxf@i@wmzK=FjATRoJ!n)qIL}rHTX{!~!ulFEk+W&@e&kU*
z{3MI%#lE8>mIOm<9PX{Z7{Ei<m?_%4)8wndyTHeamxa>Uh1fs^r^(kT=ZWTE#aPMV
zgZ9ZYrDXAQt8o&jb&lyHqE=3(TRq<iP=9>Q9<k&DhY$IEmU;vcvv-(9Ja64Gl~oIg
z2H+(nzdZiF??b=_yo~uoC+ZiDSmckcAA_vUts~l;0EJtK)l3-Az4M>R#Np3Q+zCQE
zpvK4)o?`?hitl9LzBVTVchN#8`|h8WvktxPrWd4Ed_7+ARen8raHMxkV+@ecn53SU
zi1E~90YSn3`Tj8bV83s?vgBqUfGRbp!>87kCF}HyRsZ&|=lep)FYIA`_5=({Ky1)4
z?X;Ae<KDbbIDqammEobbNILf783!R&3}jc?d(oq&{`_3D7YW2}j0`TAwPXlBKt%v%
zX~jTG^^bXZBMj+{*7sW1hx9~BqrYM?`Lc0a+6A!~u37mSzw#RXCsCq7_u06wmvixW
zkG&V|j@R)Lr{~~ZsxVWMkIGGv>Ya&RcDJs%Kf;Cxr(=qTr_N$HhQ+!ue8M-Vl@h&z
zz5mYdpv8v#4m_>9f9)b0R7kCH$uvq{K8L_iE<nWzd+!YxGde)dX?@Uth%wshC@77v
z&#CJt>aReQCPPPPRjj<G7ni?v&dd$hsb|mhGpnmFJz2cQC#B4wwDuVv;!?&2EzU`W
z@_4_IOL{)IjeJMwgjyLiArrdm`sakN&umC3a`+HNV=}5XzWevrkV%4s6>^UWN$jvC
zL8?@dkgt|>{SGKBJPdM*?&}WStH{2-zNcM?buUcu;m-^HCq~3mUfU2G#jk5_m|{bj
zEnDheYSUt<E%JF?cN~y;J;Rq<K?RwlUa`+9Uu`8Ap?4o1_|e_E%f0!<AG5e2FDM^~
ze~z7x9Qy0;x#wOu8pK&GX4UX%Cf9T|)382;Q=F(PmYihc=It6JHUseyCF3uz?ZJ=^
zY|GIcf#J!9y9Vd<+=TKx-d2CL5{Xk)c5C8?ZlF<(Qz@dFE*aN`8Q=guD;GSIdofSj
zg7Je&2}3v%afMxkZikFpZr0rD7OxQ;f5h2GUk1aJKa$-*<<E6O{}*yJYu07?{ZL;1
zo!Z%(y1TG#u0X6<O592E_d2bw9Jr#)qnU0^Wp}%QY8FV)(_ufF6C3d7GdDi8KnjTx
z-`H4o#>*jWvTf;A6me?B4F2Vp?k+EU5H%nnu@2%GwFxVbVj-tA`Y5{gT^BwWJEp3K
zeR#h&%t3Trr`m{6KY4ZVM!1jmf9jCBehEy(sE@~k5AG}+yU<6FWAf@7`S_Pmcb2IY
z@Uc#Mfq9V(gtyt#a|iKjFRsjWD+^3#b$p0zHPn5!eP!0>^41}kIqGElMpdiFQuaeP
zJE55*lluVGN+%V<^e;rXWaF+nmqOqCNk-WGZiu}v){NMM_dv%YDg^#y<>cfv4j#>4
zUBR??ghtRUosbhlV3}Q6+DnI-zBlsJZhC62I{bdc*Y47#x_fw>K6|om>OG@W1m4;i
zP8pevM0amENiLq=rpWfn!ukbQ{tUEgKL!Bt%DbfR4wEJH&JjRV?Sv+k1b|`TL77{F
zr@XCz?6r`O)XS^4HIJ+$`Q*k{3MVY?xSj))uvK%}NnJ)olA8=i3?0Wm?rYJ4w3Xn+
z%wPlc;#Wx4O(y~&&jGb7c|4SAInrVb%1xSL`l7mX_q`ES161=bp@M%&<k&4$rQ)Y7
z^*RgqxP_Eft<|HYW&no*PTZ#rlyQh}$$k2|)Ge^mU3T}M>@P6I1g;g-GoBDEETc$O
zV#@h}!}2DRam5<RB?xuriMAhG>{xYJY~5yu{~i%Vz=Id(_w}bohgpraQ#QmZ);=$H
z`1zq2sbs_aJs227mzIFvt$eP-os#zop%@A2@H$Qc{2SP{`H-)hR`Sgv2ego*W^GDI
ziQQ@;=u)wB!7KDRp)_$Zb!yEK3MPIHXIyDH2O<T}ZxvmQ2VjBBPV@vpoxHgA4v1CV
z>d3RFbn5~Bzka`Rx>4ZgnY1U{4v@{cgporTfT`hUHQ=j_Mp6;+{gOuLO)b_s&OUnd
zJpV`ppv7ZcvhSspjgd~k1ONCj(!=)_Bvg&We|C=jM~Y_%FH$9-_(Oz2=3u!0CkS76
zI3sMf5r-htZ)a}-el=pSp7&Dl5s%|IjUp2AE985m-(2*cn$ey2e)Mv!B{cG)*6W4<
z7z2R1^lkWo%XBaDD+kgKGO9Z^F$Tx^;SqYkx9tuX&Tv^ySQ$$A1hpbbkqk0o&74Yp
zOz@w*@{*Zljb{bKu~si4t<}XCcyA885Ilglx(o!@hF(CkxnKRP<UjIec~|=!O08G}
zuP8sk|H57J*vtEybhTG0+Y)aTzHoDxch1(Sbni8OU#xVo?~SHStnhM2AOI`Jxr_|?
zTJu(g?_ek6xL@pg+T>HQD6P>qru%aN5)~HiVP{=Z4r%`=_)Vvu>e193Te(i^ymn3h
zcT84a2s><@K<;VT_>2O;8J8+Q^RL3byp2f#`+6<UKVCs2s3_NV_Q{+XYjIK0f7$oB
z7V6^5KulFh$-WSsRuz^xiQY{zZan)yBC?Y)`$0Z?sbnnZ37$iURYSLF)qv0IH56(G
zbz>_7a&v*%_z59{jh_R?G7@D8WDh?6j~#dshUhXi<kC73<Cb7uZt&@rGi5__$qH=a
zGb)h!(FsIarQsOOY^$*YkcvhB3ATVzcH81@X&w>G*=?f|QvEgfYKLK8#~CrrP^tKm
zQa4viV=LJ;GvXx=!e`>+r`_h#qmPgRVA9o7zjq9j;W#kh<VpH7o)ZiPJ~F=uu!rE1
zFL#<z8+p*L_w<kFLh6X@e(bpg1tH^<F^p>n`h6eop%<{G8-BOq+Nq~Kq$KlE(YFzH
zZeDYyS%iY=UAL7>$AH^8p`)+Q06BaK0{h-J-bIro+gk?IKpQ&hOYoMFMacyz>XqZ?
z`T5(2=A4N5zdP<3m^k5cLZP`^m49F~tnwN60KP2o?Tp;H)%LkD-KE*O>U^vTzUvD)
zu5+$oLZQ|9PBl{vwrvQxoOhYRInD~FYw8)#ylY!F4Y+tP=L1(^f@3b7!=Tm@Ty1sN
zS=#3-Fh#N~U#GtwH1V)ixi}n~{0R}+bjVOHWg$Wcsan_L?9I@xbbu{O(4){gl5u~`
zskf`lhG`a#u%`kOg~BQU7QfO_DhT@@r+H57its9a(f5z+E&?Xi#nq~4!2`$g_Ep<Z
zTFi1DHgEy{WJf*lC+B)CX07E@zRU&b_N`{J?=POsUYs2Z&itiA1)TEdpCtq*d~Q#S
zMTO?$bnS^=({dp#So2nIvfCam*oab(nmcxp1b-jBNyudeM`LP2;HEO#c2Q7+n6Q4z
z>OMTszW5Wg%=~^r5C%*6Fx<XyT@P1%D5wbBcO|Wy`QE_WeOKnnBn>XaEK)Yamv^|k
zx%aTO>Nwx`Q|<UUbt7oq@e3>$N+k-LC^u^1-Y_;WXZNEQj-Nu4-eFZ*SD1Z5_{8qb
z6vvMbpXs>bTl3jMeV_WZGT<`(*7%!4Wkei@TcKLzCN(1TOi?Qt=P*IP$4)@3u>Dl3
zHj{)Yq8yr4vo9cwV4PC+)IZEQ8SLl|z4%ts>QbzU-Qozd1<}KWkov0JA2>4$6rEeN
z!$3k+&9vNLC~VjX>w|HD6m9znZNBTbO-;`iDg5N>urrGh;A*`m_#N`B6!IOGbC$Ca
z8wTl{SwUizUYrH&4diCjEKWH2eEX9R(r#{fmUn}6%wzrNXJWlibNX95)?Q*0Mz-Q$
zbMiM)`^xho+g^))j#|M>Fm1WYLQr-<I2iy^#4;&sX#?5*HU;}KeJaqhS#QMb&y>c3
z5P~B0lttebjjH~RIS9lUft;5z<+x54?~CVOsP-^QACEJ(Kc`|d7#2s&K2j7Mgd`+%
zO0D!reVWaig+{cmr1-fn4J%gaBSEyC`~Y+q0q)sk^K4lbZXO=;73eqb(Fr689ov`7
z2!%RhrrDqniQynw4T(CHL2?1JWHxl84}-Zc7`2jB087SK>@TB1kXjmFj0?MPRQIVN
z&u!0VyU;k`m9y-s8%|FLjia{M-1x-9;a6xU1kQCWtU&2>M#seaB@csfBXAsvyfTxG
z*;nAVZkxwVX+(>I5hhgQh*vBjaA}&!*jj83`ovSLjE=1^Uz+>j^!YM{<@Z~i2q&r~
z-I%4_XWaoYmk((oD`7B%z{Q!sfHK$sZoR!p&w0cVgCD3nsZt+?ZglV(-ng(;jtS#n
z`HXeXF4+E{p#iGLF3v9&9UnA>4bm5ph-jkaNHfyXHN}CBj&8yB#b0O|MF5Iv_3GkB
z*^8*KCc>&a<oM`%641;xGo>}A5Nm@7oWXK{d5uVE%@fg`p*(8R$bONXxCa=qMt_F7
z5P1r0^BXr);&ULPDnzqPZe>FW&3%FzO1J{VtS7T3v!?hTOApyP7kaO=m<Dd4wUeEB
z)4!=|CVwLdvBuhS>V+dd#S8SYy_@sB{X#(mr0`r++&`dIGKJkIwf0)-?|1#v;(vkP
z)7e`?QD|GPeZRl`E1o$3ceb;P_5ZEQWh5@pn_>W@Ip9wb^WN+BV@dHI^((RU(R$@q
zK5|GIo{5W*X5{dv@#6pNa-$<z{LnUC6LLy=xkoOuca8G7seFE449LLiB5q7>wF|fG
z2(&ebXr)v4(d}QmDS-nr5Zsjv>)8OI%rA#DU8DNqr3S2spkf)SK@I=^&Pu7tx2v^B
zFcZog+0=xk4A54lf0Zy8xRjq7;*<-)Q7WNLw}6oO6Qs3$C$L@AFg=0N?zS97`18;y
zi2f0h9(v2v)U?Fn+%*j{FkgwfV~?}{H8@lU{?c&%ioyM1<qq=j@bFKa&Qy8tO(0=%
zNo<Z~wKx9}lCEUtEdVFP6_xLfgJSADIwt_ykHj}ml#f;_IP%vcVclYA#v#Y&B9gNj
z_jx94=Ew4wqa)&L@Q>ZLja}R)Fn9TWQ>offH6+z7^=vOLVIUZiZ~dXe*k4*^Ko||J
zi`WA|#UTm)YNoR+d2?OkevLVF9OQ0tl(4BV4w{dur6SV$SgBJFqabXSUuqT3LaBjH
zu#*NnTV!4ou;N{s<kEQ~bT@W$c7I<iT>C+wK6k}HegO``j2^XOvJsPX8}O&<KS2C=
z9P&ol>;?p>@SzYvJ~=}E8Hafr?!=^V_~iWxLn!=~$|f&Yf^0|i5Ouk=oDhgwA|uvK
zH=8%hwQn4r>MyXfT_+lpic=#<b`%Ztw|DHj6i0veoLc2*Wtnx9gWmd~kq?1rQl;gx
zyH);E*^uBqJ|v&dgH)p@NC%(!vc)7+mp6Ml{h4M|!-Gh5`Lp`c=i7m7%c^5G@{8?3
ze6|=DARRgXr2(x`L(&oc*i~e#Xze#@Mzib@m&YV#BLtQQ6M>}yq$j(Ow#*D^2ShU^
z@bWc*!pitnBi7PX35>UULs1cF31aj}Lh$z#y_<^-(0IEY$u~kNYNP}(Vk#(Y+g%!o
z*av<ZQN^-Z7;^}?CDr8O6D5UP<#bbaLU#t+FmLbx8H&h$PUvVK%#(|eoh-AMXmng|
zM7*rH76_eo!nv}_Cf{IbQ%1T-Ocj&`*|_5w#eDJhm0+l7fkx3HjrzToo9~vB_>tYQ
zY!o(vWU<<fJ5dpWvv!v$L*B^p-UN~lIHDqD+g$qzypX26ps^k|W4ms<^fnq4H82B%
z9%O9r#~C9~$!s^G@|wzMmL%d!YI&M6WWK#>kkw}yxf$jJ#GCbfE+_i~+C$jfZ)s`A
zWbdDl5kQy>mh-ieym9uUKsg*mf)eo&&6~IUzMx1cfDQqNFbQ8k&{_={&YG&=Erix*
z_3{0Rl{RvVmx7S)KA4&s2uScf!$fl~6P)~U#I3)D@IdVl5|9Fb8kn*Pe{Cj(>u9c=
z&i$$eV^HE>E-YS<h%d#$9%(qil{+YboTAaK!I@99hcwCj3lH!b(v?6YgAuT5b!#-1
z(N6Gul>k;U*g0y&fx}Rf!SQeq5JW07zn_>=MJjTLNZH)T_?cAN73Q=HeE`hOv0WlZ
z*H1@OIySl6kb|>!vbSumb`p6%pLS&9l%3wk4O6nlMs@v-PBakj8)N4=rW(ptFOnAY
z^`^f=qC}jWePz8!vG<MeS2ngME8|tO8QSKcoc$3D&#=}V+HlF?`esZm`x(UREngZh
z*b}#C@?nLR#ad*p#wYKvB;RzM<uS0D)(V!q|BEfKS5jknA?$Ncd+uU;mF~|m0*`0U
zEN<D^mA_l$;jE|XR{m8GPVk;!)EH_3DAyx2RtzB<X)PDlOg6Q=G*xLla?gXyuUtFh
z$3p@of{E_#Zm%3zqDYW}1e=V%@L<B&g7)0{z>2^FDNR3vDMuHfJi2$%{S#JSJX~YP
zUcXP6EdMDNJopQefIo#x_+wm1VypBB;nqQ<N<jE@RRcmOjcb|s7|ljlj_{c|vLW$A
z8&Bb0@Z(OTY~!|F!ul1<gw3YPBjbDCYn<7aFGrbPwSkj<Ypgw&^147Dn}_cKs*E|}
zi<Gh}wc@W{y;7V(c<hYKX_XbgtR-DLfbtec4QPw^MJnA3X6gw*F<0fTzUr|M+cX%j
z0RR&_%HXf^$}vMJo7p3-xkM)sv&cC!DN-Uns+$wfc;iQaSp{A&G_U7nNYyNMCl<_R
z?Ei49@QLU&!f{hNd7I(cW0#Ctz9-)E9jl9u>7(tCGky!8GN?21afuFvMjb+5g#g+3
zQPs!WKXrXC{E9|p{ebD48QfRez%w+8WZqVgYG!r-wGV@i^Z+VuMsjxBG_BpcSLjD%
zsC%?)FP_x>E4?;(E*!?Ly^#N<VFV^iUz*9E)p4j<HLwCl^iL#Swz?QF+64Q~7?;I<
zP>fgK3OMMLy}WP4rEF?mF?>b}Y+#6iyJM0)EQTA(^UNh5piXIg<lAks=E46~5};{*
zXkbHQL{dueiFhA3wk{9oT6k}9Bt3m7ZzwwE3L#Ogh=MOtXT@uU!mNx9!&j8SdGtE!
zx4gwhqLLN;R}yb7E2bN9yceLER(i91bExO%B746cpPuqW^QHsDI){bi&$l?WW}4{?
zaB!3XP};%F#m|4D23!Vz;CJbQY|UQTG<1GnKuS@9O??all+e%aP|ajkMev_aDEq_G
zl8z~DBH@1l<D9;LMEXjJ_Zv!d^JDkk(BOIg^;h~Z&@owGyjrQf9B4BXg~72-f_pNy
zpMWVVzn94v50DQdAMUZ7spftd&>46fOZK%LNZqWgSWu!0fW{T+1c4-qxp1!BvvyF=
zr}Rq(_$K#ensnqX)AIqE#AUHA5lhMe6EPCv82a^v8)lBiRUALJm$A8=22<&})4=PR
z)}}dkwB{09u^(rB)(roPbka{UYQ9iJZS5=2Xp;M1*n<vY00pl6`SQ3A5`gB(=X4$E
zM2_kEz%nF747a@ba^ok6baV;}EL^u;Vq1Ra!i;gdG>iaEPt2UkTuYnToE^BRn#HPD
z@0wk{m&Nm-GjTp)bvpma^l+myf-H4He-M2<2ZTyI5n&#1zw22iaAgUEDJ^6M(>SoO
z521zg1U*~t?I2(e3Vz^086}*~h2aDugsF~T@+J?2DNO#mU(DNn5Mk+!LxGw!46dFD
zK@v>axUD0@>=?dt#F`(9gPpif%w`lnMvBbDIlps&BzHWddJYuN&@rH2{7`)2xFmP)
zYt{=FELkPah-sWbjMd(h<Ll{0Rp)^CprfvOLgaRa&q9Yt5agTIvuq|&XpCduzVADS
zk^weA+d<+R;k@qpo0{jhEW&uuCjrpn?_t?hNWZQNwhJ736c_hwEv0bk_J0PAX`>X<
zM0j!qQyE5M9&tqx@N@lV1o#|&G>S-C*URe1`EL}~%!1(uYvu7a-lX7J@F}>}TPE(x
z4vGbtIE@T=E>WZ1ai%|1^!gJRUx9_J^-%U7e<nmqYIOP;hwdQt$hT`98n3F`a(f5~
z`iqR-svy%j9v;t#FyC+1(Y@bB!Z<Ux8Q|zJgh=<?@2_&G=)A*`E9diAoD&TLwUZ4~
z80c94L!$7yN&Eja{>aqO&A)6+nO2I~oOk(iYn%qkGcFixE(M-{{uKHRVjcDB4-(i0
zntB>H;dB`{4Nu>oXwfoO+M{~tP1D(1%-Pt@n)PoQPx!x}HP->OmvLw{U2K<1_^H{p
zS(MyV*7Uq{?#t>+OiT*jYIS8wy_%<NuF=Y`@>D#tR$R8e_~yjxzRH{vuDZsr1LxE0
z7T>kM@i3Zx*JRi{-NqyiwT=O^N&GQ-H8C11x#K$&);L*YAiPT8>gM)%uA7VzT~Z=K
z_1|7yQ7)s_i>T0JMQ>Y^4V})<mI}4hKL?PLq{}1J5l2XP%0p?mF@eZUDAaLZSpQx0
z#rze==DvZJSH9{YWh*Mp=}y&{JkEh*yIY>wYBQMsBL{Cmk-EdOx-ztSHjm^@({*Cr
zJcGt~h*i~FX;yU#V|N`Gc=qgBB49&c<yb#`By7qvyblffRW>UV0wxU|wW9A_@(w+H
zOW2kEtjh$-{7lPsIL9Rq6N%R*5w^6l5EwKvxbcPv3kfN$x_@z$qFg>l+(G(zKES?c
z4So#l$*IKdLSWSPv6-f#^6EC<*{eWit9U7U9wV^>V0GFS%baIUM=g;Dzc^nSmy<JH
zSr(kBqSjv@Yp?AtpG?#|>i!_}r+c?)eT`wC`W1>{>L<3q(elYuk5?Qv%3Wa^iSO61
zlUvBvR+HDH%WCQSX&q?a-X>R<Rs+%{ucTrNyNApWq{oRbU+6Canl4^Zak#a)A1Okr
z@Rq0U;7`pNiTbO?==zQ#d=FdZe~ii#-_ZUYl_yp}0TD)vTc~(4h2!Q~si^prfVGs~
zjXgf#4w|sdMCKp#S;9<s-$}>aWRMBDlM*l&h6%KysQ)`_1M>nwsCR4sRejf!f6%Sk
z4#Ex}y%@bwakz>TEkx5n!p6^^q_G@!2?~zl48PpSomd;PHyu1u&q(wyNMk`XL@ws>
z%a@NGpLNJP4c*omp0=wMB#n`WV^9$668md(&8!;=aO0%wJo=R28x&O_Gr|r94O?)Y
zV6MLH{GOP<>R+$D{D~ltB0Asbgs%Ho3)hh8&84JLzaI=pNK|~RDc-hesyc#xERbmt
zcFgN91-*9ipg)3>%m01K07?w)^pii*R{pYuzXS7xXl99=a&3L8oUzKkTx%G9_H2m%
znbLHUmq0lQ$y8Q&QWovDr=zT}@u}*wZhlbO@vSfutY9Hgg*RQ0+i~5A6)u24mkPz)
z-ucgDbddGste57tt}X|l+fuUkwAt-b9>YP}482fM+1qS8Nuv$I>$(3s2O;>V?xFwq
z9t|XyiD;gnVEhjWSQl%ELZ0wz-g1s+(N*d85nWSF*6y@F6b99)vKp>qsF9m1fV!&D
zDP`pK6Fj4BNA~7ML3_g9@q7+u%Z|TDyoMYoJCqY%-<q$ni|(;RUE9FkV&5SI8xpom
zI+#Jr53eI+<drHoH(!Qam<v;{ax@0o{A)L2KS5Y@(h}((H=UPlLoYCeHCJa=oc_n3
zrGaF}FH?IraqWXp0=SR?IU4H^(QJysTx3LYncFRhf#;xIRBy@rBS21=4KsAh5r<4*
z+jndZ${Z=+EKer;5A9OOO#NgrF(1n{cx_QY$Sj7BI<&}ro2qiMXS)=*frKCrEI{Ex
zWd`C}UE+T%M>C?WZTF`9LsjiL2`Yp-yqbE)#e~}5AYchIgu3fv`BNnJUeV}`Fy3Fp
zx_<+#1-oImk!2|@?zK)a-!h*O6cGOnbfE<BzIV1uC5I*KJld*c7_F_PzWHB>^V&le
zpcJ#MK)H9`&P$1^#Al=+Owq}X*Ia)-y7ORUy}sAS@Wu)4vvD`RLPp+}H#5e-I+lKm
zG2Qx)o6AH@b`(Up@O<r;o$PX0oBtvA2>IkfAl)t|mZg}i57wTk%1}--!)I?Tq?d57
z<BIG1u8I<B;bwP7ZU8dt|GY7EWTr3@!4)i4yPZc0$K8wbZ6-gXlpN*?sh$kSXEV1#
z+zW{TBSk+us|1Pm1C###_dI)BGKlm=tqc)vn~9B&-g_Cg5_MUZR58-Q=P{DZcyfii
z@dJ3m;|5SSr}Hdyw^UJt)IH45_|7KYBb+7XDXD$sRC{*o*a#oNi1x6#owOS<P$wW&
zyWg4PGGY=ryR<qpiX#~x%}+7c^wX%lXZa!Vp3ddN40{g?974U+y{t<Rx7$$Q)a#{>
z`oqQesnd!B`|#6(CwH@XiLxEJ{QB^Gh{D0sm#}+$CviQlP+$5)^Qb7Xgsy4XOsGGJ
zoj048mp7T+Y;4wnzpxdW%Jm0L@!5X&Rj&wITo^_wKtlC^PNJJNrsaPMGDiGSnz(N|
z+sT$$-zxz%fDggoM8E0k>ETWFS5}ta4!&e<YNY4ubx!^X)K{W0x3cP=Y(p@Q{y?h(
z`zXz&O#_jdyBl-4<swoARgi0%{UpN_Sq-2@NerDg?fIGrz3dZVvFP-Ybt)@xIy$C;
zW*^atc|US@cW=!kd*1;?bw6d8bbIDlZh6?2p@*E`E%r^OuYpybNjZ%XFQnLaL?^&R
z=yGDN&Q(of{`iid4jx|K!cY%>KY?HC9A%C~phJoc%-bwBI-GHsz7*VH{DUrPJ<)AV
z;BL@U|7b5#&4Ti4_rN}-Wx`Y-Hp(9HrtoRzhOk`WKy~P=n;w@YuO%7(1+}a_gMl>h
zy?C#sMgAl-LA`Gqt}Gvaf=0*{7q|$;dt+}J7#Pe!B`rVO`}!)P+6%u1_o&hQgMUlN
zNd1F9@p@t3eJHr;nKhD9D|xG{=vj%0BQGM;juC+yhspMtN8@8Np1rwSDO4rNW_sv7
zL-tsn?LUY6M7@A;wv#L28zmQ%3~*R%>7dhsQo$@u9=t61OF+&Vj3|T*QZs|JN?&&R
zB6$^|FG6d2jUyOkHCr%s`ddk@9h1D~RsrDfc<{h>>X+<g33W)|eb_zGUEJ>;43QXw
z_H`y!DK+|s-HOIo?(6}i_|+TMTUYo?cz4{FxH6YPg7BYCpLy5+bLRf)MYCl<rLB=d
zRadO6tb%22pr%40Qi|g7RhOfWA(oy{Y1bQ&f$+o@riV*QnstKMG`~Cw^$-kRR@|rs
zz|UjFAAcD-3obGiJ$rIc*T8^a;$setusdO>kemJ+o2Y9R*j@MCmJJNl(B9tOt#htS
zZ(t8hT{xZ0sjhszblT#op+)wYl>=^?oGYtoF)`4Y=3e((f}+Z+pOu$#V$@w^{e~~P
zmw!(;iT&aemx<H0YJ3;7K|-3bas_bY77u)jc~;O@?oaT|ey-EG5Q)N`zWeR<*-Y9y
zU1hVNHcrkqCuxa|+7ULeC?=g=s#jl@{SLW}o(6lfr-s#6dD6|xH4=t>9c$i;h!nKN
zE(V>S!Q%)*5oQ$oFi~QK3VTQj3!blg&ZK;fAj4--dYbV1g{wE+s>27uFUy-|kE9*{
z#mT(6p>&yd@1E@&N_d|Jq$kmt?h4SMKD_Q96OoDCk5v1>uPEm1@L^UZ?;@OCvV>~0
z&dHVhyj+4b!F}t#jjlL)bQG=A>GdjtO*uktt2(&mTu>m@Au7y!v+?<o;j~?0S8(_j
zUulG+V~f+G8FbNKd!p-eXYQE2@rq}ncDWx!8>a+HKKxUufXIFui8j?z*^`@`oE)3f
zb)JnzXlb<LnrIxJHNwb?x*0K}+yGT9sN`Z54o<RT3il=6>-^-9FgSc?iDbGf?C!$1
z7I~XW0ke)Q3j_70_gv3VDtp-Sx|Eh0x{W<0ha>z<+-ib`m5f@_B=Tpv>lq^^^Mzwi
zwleZv_+_)O%wt~I$#`-#yN--fa=N+o+>F*lvXpCavjUM$)I8hJ)abqF#!<D;!Bz%V
z^t8ySf%3-2ZXIrIkyo?UQphybQcS(Lx-?bz{ez~NwM*l)t3j&ck3Cdmi2)aDml19n
z5?o;(J8qowlE6=X@<Az7LpCT_+)q~sMH>&_MpZWCU`_O&&+8I61<0*(c?&UTC0<z|
zZ#t^0oU4Ok(#ewm-<<U))`Xar!}1R}#`Z5Rx82jl;w_8+xlkV?FK_uaX6O`gu`Wdn
zu-6DbJ60?B;AiT6#+qQSj#W&?k9VB!^>GTZ-7Qi}Su`eLa;)+qfA_^y&h&x+ZZn(Z
zLItz*Ya@v99dZz~O)YM&^kvHDdwM1c?J9jEmyak8IXODcZzw48%hd6m?NhSRl6;l@
zvtqrk)>X*5-by7VzL)TLi1jtMUd{W5*=29oLVslBJ9XulFj@VT+_a{A2qg+U&kIf+
zpR6Z1%7}|}x>oE>Nl&Mr2iAo19f8qU)$BCPGqjY2+$oyWE&Km*kPw38pMRn6s|ET#
zE^;oz+cuy#NZrY*K3X=@`IvWdY#G$>P`X!H@`|36f_HPrY`Z{zBv81aJ+8gq>@x&j
zee1$PLup=TbBNnr)N^nl>Sa*7P3Ks#uHULIVyUXqT&a{VZ4pZ$&9@1v9OIHCsM`j)
z43ZMxGw$LK5^!|5JmNSVqke?KhX<HF_6Rrm)lDdWEJenyEpBS~uOo;A=CdsBWPCI1
z<3d<X&#YF5#gSVfA&ZWyvyRRc6%{>xz+=1El>RrKi#f%-2yh`jDL;Fz+h)Y&D3i@%
zWl6`1|73@;08n-iilvW$csrMIJ4Jj(*YkUoe6y$ZkUsI>9aooMX(GK>w$0C)W|z){
z+*~=MMq+t{o{1^IsQ`Ic@%(LVZJ7lNEDjSc#Eti<1RU%Xa&0UCE2h7qB1xC;cASeD
ziY5ksrLx&-QU5j+(I@e7Gl&i*9%L4n{7O?8c|G(>KwBP2yo%TD7kqi5cTJA@L9yF>
z!dG+>3>Hf32Y%H9tcjD6d+vr!86F=v-@na`3=ze`8kKz*YsE%2SFrEX3EQ6sK|cpS
zI&x0#G<r5@+7Aw_r}NM$`p4xczFR+c3Q3|u@R4Vr3}2rLmkK_CsR5j!ws=#D?+EZG
zNO^d8Oxc?v#C;F-RRkVaQdT}5mNm2)?^rV=)+P}xgNK_D5zXt^_$DK#jDPXxt54o~
zP_b!;IVP8fqtawisC;AkpP{U($wz6FlswJm1G_HGn~!a~3p%Oo5wJJ7Kh(**?caaD
zJ|qeUyVXBy@$dd&Yri4C3U9n_bA@l?joY6Cx5sw2*R@4VuKhbo4m&GLd7q&^xAk7e
z|MqVChDZfa>dyyNwkBF^{mS-(#P=MFKb_e6Z(H6PsQ@Fn%|p7gf3yp2g&o_u;{KR#
z{hg>neDk~K0_dx2YO<q0Tif)%*zS|29zL_wCs)dQ*8jJhZF}>3cpu;Az?kjc%;wC2
zzk)Sds52tDeL^-I!WXwe>m0U<5C04G!h8~-8j!d$u9&?+A`H2WoUQ#LuU|!U;b*+q
zHm1FY$>#i99$fvp`SL$6a5EN<?XVv)Ka)jNFV?$LTc&dBPakl$E)#BSW<QihTh=sA
zs-fQ5i)ECD`v(IBx7Ws~y8&ycG{o?aNjpzN{UgXXzCn81!f(DVG7OAHCJV3aFw*7(
z`hoydipv|E*|rWFW5cbx0RrHi^=(`1VDqQwGF;pYo$wgWzoO4;lcj%_`iGjBs@9HM
z`2k8ByZL<Wjr|*OjrZ3=!op~;##L{I*pL^!7lf}j{}n2?Ii_tFs6g0D!>zlG{sBk~
z!b(U93+Qk6)HfedrwHFyLHT25b3(RVyf^?7_5a_?(P`!!g6B4bsOgZdUWuG-s`pkw
zEvMFaV$1%^lV|5GiYkO7hs%yMih6U=wq0;Ifpq*Ie0vI~-G154of+bT5)(@}y&XK3
z*4fSef2{=kuf(r;$j&i+^8k-TY5Zp(h9jT3&F;b0KMOMQKdAuLLOyoYwm17nMn?}Y
zG3EQe<+A+wo!H}U`8n3xMPn=b6XN2ST&V=|tdGCzqs0+&BM(y-`^KU3j;%Rh`GR!Z
z^f2m*n<M{fw)wAt>XR#ujbKENjEv+mPzj7#T3Aer`cz%Qnjd41i;LqY<&;XFR;;yi
zoLN$9*3qWN2sg?X^t#p+5N<Tt`y=-fa;NUKp4>|?x!b8TBiNSE1NF1^qU7B{;wqoP
zf#NQ?fyUb}`}o{HUQ#wAcs*2jQ3$8iKU;k&b=I1T<)ypT=8g@yj{Tn`c^G{ijbU?2
z3F!;ERqm(ir@9T}fnxR;tTj4gw;`#3iyw9&{HnK55Qn<_vKd~HqKZj_Ve2m~vuV|+
z=9LwqB7DigMlAR6OHY%VHK|!vPPpYuAyWg0iyC`Kcfaa#bB;Z^!RVBc*{r(B2)+25
z7h>Nk%~|-z-e7jB%qh@<&-@Ic>8hrj+*rJ2HRF^oe-N13Jqw6HW84WNmE-6*<@CMF
zO2=C|88|D_%J#KR_uh*l8ngd3JZ{t2WyE&vc>0`2z=Re!&OyOuQ*lp&t-8l<!si!N
zW>2sjSrSOkTx4tDUnOk_`vUX@NdjbAL(~ARS7NcA-B7TbS`TxvVFGdfRXA-;c@NQQ
zi1^x+WN-A7(7?r7S+d8y&%E_<C^zh~dkF_)Z*Q~oHV$MIwK6Ph(~xo8)_Z=dFElw=
zVU4(QAGRso&4cy@6Ntazbb7--bekKveg@bv3OgTsU%f%Kr7b8M>^!s(As8CxLs`ac
z`fziK?udaUqP!fJy7SPaY0zopUcX<BCdW@nsY|?fb=y&gx{9TpAz##YYUdGBOP>MB
z80dWZdq%bmM&>m9`Jlhc#sTohP*88djv(Np*syl|t`v}uN>ocQY{8dNUtoPI6`T#2
zw>^3DGwX!l{JRh@`j7Q-h4s;iE#&-1y#akNJ=BiNhy7!HzQFphH+1@mZOX&8i&QW@
z65L{L?7TkIny@~{hVPaCV`$A_dNh8KjrwC!)~E6X8#>K3Ro-ml*-^y0dThq0j!;G&
z+;YQv(E_ei8H<Q7oAa^l0+owk3$$Z-TW;C%gBJ`Sp8w<Bv7J!KbaVR=Ue@@pl$hD_
z?6#$-`w#3JzeI+Am{3V{VBMvjXjI!-gd(4TkV_~W`HTI#X8$RYBtEfiuu-vHn+LQE
zZi+%7-;w3WZL=VTSXk2$4O_l#a}NT-5{S$K_E_3KW&r`Y{>Qsv7U=J%Z!9C>)ex;m
zS6iLC`jxhRQr#8A$c<$9N9maJ03iouGB;{$z32CMBOfD1u8h|f$=|xCe35DB%TjS#
zp9T(?hEII^+amXG`KI^cFoM?>k$t}(yJ<wj1<D?ohVIe$Ew^m><%{z$4gWiL6G=!Y
zI_o37L}0tW*uR@dqFhM8fOd<ll+(g7ZZ_@H{EzMC1P(iwZgJ#4@)S6HV2=I|zCDFg
zH&@1GW7$He5Qx~Jf@gabjA_KNa<<t&mI-)fBH-}j(8B5wsk$ATD*zhe{uh{jB6vPP
zCxm$cjl}BJJFV6oCD84(ZTUz%qMWnw(c)Nse&tc9>2%Hx!d(X-Iy1)ZL9)}Gz_#~=
zS=2JxZ&9!5!FSXfpxcH7^$z1)HQ+>#1Jug01i88D7bl8H3ZEYI9v>qu6WLK@39q0E
z?1>fJ-b{!2sNo|S4g*6&qS0K|lRft!$GhMA<uPv4JNF^x^eK0?-CtRoPCE<Qi(Lsg
z=m->jy2kJ8OgKNWf<vR__3HXtlq~83h^l`^aCUZfA(CADP-78@ylYfEFznuI0Z^?x
z_hX*`8Pen|(QbZHWdU$2GI-gE($dqSJ4c6DId&U7CoE5hm+t`lE3^03q}O+7#Pn$}
zB3O;surk=d*tn932Jq{}o>!lev`ViC3k&}nO9Lz}p*4O~9y&w?icl)GuVhb>NM1Zy
zG`o@}un;OB+F;ZHt#C5N@)j3LM=7@Kf-TGP8gVb_WN5UNleHyo#AUoLgVJ8Fy7Mfx
z%ibJ>D1c9Z!_q|YEBk*VXAf*4q?Bt4P1TSNlp>TEnA0&Nkj%W&Sc-rC&PVd`O?U^T
zNl+x9f6jr^(ye~I_R5-mz~G8C{2%~61p54DePs{3R}KJ$+dIHvI4=E4rSD`3-am%G
zdIq${WfM-xzO=Nohu;WOENv^xJ`;($8DzHXeCliHVm-dz^46<&=+@ygLDefz0o%)d
z7SQi7YU6-jwGzn6OLz>?I~h|wQwfNQePuBrSv`2G2zPbw>mMAQl0PfIJ5S`oe|X9v
z-<I(P8duEyx>YEG=T|GqL(8jmoZ`Y9C}S16dn;F&nSxZR?{mtAqOnD%Z&8Fg`JQ2~
z;tQa1KrdMx4R(3bk$2RJ<AlN6XmLcs0b8vJMvtVmFuPLnUg&Y^1wk-WWNJ#gt9*ZD
zzMo+7XytvDi%pdu)AN@{q;pav1e(Nx{5bS(%zfd?d&!#)F(#ifnwPuKUx?paG`anh
zMVSYi4oqqHCwDTAxU6;*5Q$Z?O4V|;k=<QjW^8)UCBjb2F0#p8(AD2v`^-OMQM=Ni
z-(`kG>e4|7QLt$k809a|Ac=Sn?dC)5FTdNf<7Ol#d;Q?ps8g4IY#9M5Ga<6_H=>{i
zBAs|JKVKQ>1(D&*WKcHupa$@a__q-&#Hl$R{|L}fObfEi7JG#Hs%^I4CufnEOyLL8
zi$riK=FO5N5;lQ!3r5EQ=w!h|fRrTf1hTNi4Hs^+zDGz;8B9QC=v<;{@3+J|C7usO
zks836mBTpC4gg2lzrZJdsTVrceCb)Cn*S>N>AEnb7e&k3?)QTDvf48xCzWKVwefg<
zs-Hgf*~ei5<_u~~-3^FXh}U$lAb|^{{zdB{j#u<jk`<o=A#`_`f@t=#j*iaLgwfH_
z&b-vcgVa?XNa)N=lU+avaxt5wj3R)MXN0~;H`UF)%HL{nPXHPc9DVvqX;PDY6}+7e
zwF=@$=LeP9)6!N_Ft6{Cm?zKd1Oy1psESIXzPvtr)6==FcZ$lY)ua=mMnoS1GCn)M
zS(z&qsL1W&88c~tEk4S(LLYzJ8Umv9!{Nm#qI*V>;N?>)X@;gCpoEVn2QSBZimC31
zsH2}81~p=qu43Kmb#f(BkGo~QKRDP{TE8-fEz$w9Z5|=Dcyk->jUChVX#5E^_B+ei
zD;aSMihsd-pbGcRoSb*%SrbP-WbR~xqxT*Kss2%MHeJ8F*rT#sDStg_=Lql*9=D3m
zE=>dYjgP#e@F5O?>Bni=7`a}l-u83*Ge{42Z8LT^)z&f4d{Wg!D8lGsc@-8@ICV0e
z^NX}}Zx`4XJ+<;4b_5y2SK_<dobpg}-{xK0E?Dn_*7)czJN^9P@X~`l2hPNl&CJXs
zO@98dFw<h(q0<iPfR>v}^NZ^X^<h4dlpW`8H<lB89TkV71<QA<4*M8ojbmYa3j?7d
z@v4?hwEyc6hH=(lc<Ubd?!xQ9uct&Rp4=|r`8?g$ZOT3`QNYL2xidLa5P~RFe(I_T
z?gps>^F6OJd2&`3hTkS9v%6AoCw1_ArnYx*5i5C}9U!qnz8zTKv`M9vegx9%dGoWR
zqH$)Ri0gYxqzi2RLVbc~_YGdQ?<)m+h)zc!1!7tC>mqx;uNzgH_pe{bFvIbYsoIuc
z*O=WdJASY3xKYMQBuhAVC>2RUgSz7!MN{*4hw7pb#!LDC5GjE;)zb&lIXQq|27u6Z
zLIL!zO!Oo^Z!(uAO9fSY?Avp!Zvi;c35VqYDZXi$QM<f&M0SvbB3(E*>8&?GoUKMD
zFx%InFzr5-0o?2XQ0oJe6naR;<MPZLL#l-DdD;Mr2E|*zcnFge`xkKmAkuS>4BnS^
zIZ0W;fqQRxy3rZvED?9=yJBZDv=>i0Q{2|J6*>8Ao0x6P7n$FK2a8op<b6ZWNa*{t
zvsNdlR!tP($Q`Y(u0DtybpGWhbW+mO(k_VG%Wj9BHV!d8SavBz4=7)8fvbhD;(q#u
z)4K74wKoGq4?(BkRNiS~GKg4>evIeyyrS$+RUC9;2Cq(;n%|C_-<3Q6b6AuiHZF&I
zf$K4Ld<UN|{aN23S%m_P=rz}Bg=E+bU!p?%h^dJowKO78uQvR@Q~Xn9YvIu+3{^GD
zGZ*TvsQd#m#!g~pJvF?m`CosIJ_&NO|Eks&>wjaa5I><0W}AJLX4lT!AN&8^yQ}2f
zuDre+!h(i&TW2ppl)e)xAeHArnR*JuVAHM$@U<O*80M{_ryp0drNDed7&@PzIf<t&
z>_;YXKaR6yz?Pwaa@DsWnr$yD0_Fer-~!(v9~U7Hk#QNR%>9ZG;6shpM7UMgFyFwF
z5iFRCe)8@-E}?3{Sn1p23}_T5j@@i*cz6;bc^YT1nA(hu1<w|wVvFJ-nh28^Guz?l
z7*#9M42e;g5(n+m@m92Y%OlWzcde6YxW5rTq;|=b!~)LQiD^OcRha+Ab=LelZx7+!
z`R@l%#SoU-t*80qX>c|pjZpVT_kI}&$+$xYMqjsqw90vs$fyS$KAgR>7txpT=0>^C
zF#96@sRkKX@6Nlq4bn4!%k>Fq-F{mmBp?oS3x&g6C(_??jLTMDUOoeKOCgj;YBkPV
zDuIkhngEk8$KJ-3Q<?Ti<d6(V{eXj?aCISJm0ZnsBvmEt<K9-c>J{@q8nEQM4Rjsi
z>7*rg=d;8_MKNLtl8m>RA^!xDjBdtFahM*7yKwa!rwdk$e8PbasF{NQa0GE1pb8I-
zX#ymdaujLZfe34%AE&{gcc8}T!~=Oakt0?_X`%w@<P9Z3XX#47tR`@#v{W#nTcm<Q
z-H-4{=pI2~rzaVP60Ke0&PYlVSbi*-UqDHR;UN80pf$$|Q~Ix){KXJ+M$&hE1t`Yb
z$S7Z2+lzICfK2^JZRQJ<EY4D1EJMU6TtblotICrTnHZ>5{|tD@`%RMN%n8kdp`epU
z0dsu$iVn1*=!9BH>LSlGgnn7&FSeaVop7?%3AVR<ivHC#QbHeGhO=jC&;-x#vkR#+
z!m4^@I>z?H0h-Ivvzu$)W;C1m*@QDsK}<dFUeN6!a3VfyD7sZZA@>#e-qiQ5Y|S?-
zpmHugnR@-S3^`g98JbLm1$H#ck8GW32`^0SwVs4*sg@ze%9u7x{0B5!7k@bNLC=J8
zI`84g#1h1xNuV;v%IVR*w7Iu~(CD#*>Szw6N(xB1cb_Lk>I2vSqvy%lo8ga;b~F{`
zmX+#=-?kKQ1vyN(Mg7!vtBV~(J}Ran2^zK@hFT-!p;~Mu|Am~|-5qo40_nj>KM*z9
zP#I!$G>Doarzj4hfKmHF)C?Z@)t~An%vn0B6IAT%eISL>^y|r6ePzBPX>&U0&&uL1
z?9V4#ocVvfWwrS=uyiY7HY24Bu|nS$Yw1G7-4kIZJW9O%`T0F?AfIH6YB{)6-hA4v
zljzUJgVdgQ&7xuFWlT$tR>GaLS0s?tIz*QBU@S<GO9oSkvrZBqNXs<ebMw$LUbLB_
z@wTzCv6TB_%QHw$OD9u!O7vtn6R}MAhzf!Gr*KbY=tqdt0Z~sNXCnRP1Hl@8VSxrp
z4;v$Jx@KLSay}P5BQbO2KD0V{I{))7J8pkAC3nwD*bF4Bc?@P|N6`AnKoAWc=>`Ky
zHIc=1-;oi9f$+HdvOr@&CJclCyb$9z>^hQfxQx}B=zz#3KKaEIDkJ$(6bFK-M_X)i
zVX3F&72<M2Q94q@d+hlId*k&_%#h^O?Nc96ySJECaMKs<qmWNJ7U~t5`2=N+^+n!V
z()1vpRvkX6(z0~uw*;{N%>^;p2dKL^&HBnYHOZ%lb&`wsy_ZYk16^67m8@Nk&$nlM
zq6i9ay$Ya$j^mo`2=HMwQp%Da@cNeXE_4S%RpMe)$zYwm^!V)?vQWWlGbZ#3j%&h-
zeE=;KqIOU};GASjd`2BM&175;;ys}k5FoX<P1TWQ9Pu49T*V!RGjqhnF~+%<fPlqN
z-O}iTF+QD7tI$hi|HP3-@AQLTvrvQ-Wx)ahKcPp8U^uIgvv}k#EPL1k_B9tTYbanZ
z<xw2f<~qr0h?Qr1;CSkB5I5>6Ky-aq4@R;S06c)-V2I!d1Pn4uiM1fuhXcmOLUaoF
z`jDDy4{RMZDzt<&K0d^eHHcwTLgaOTR>N-k)Jv3x0&Fk}IU_C_X|vrOd<K#~_aer_
zxb1F(jK%Ab{wNjs-2UbvlNp;z=fE$|!H|v$aR34$!^P3Dkc(sB;w)kig(J9Rp+`b8
zA3^Zxi@dLN(m`FU-QAgZJR(PKGQq};&X(_n$5!2gLGN#*ZHkn(i(Wb+uX4TWI|5Xs
zq#jwRzB=(UFGEc|Q`w{)!bzqqULL#9=W$%C&%eDxBaMDAA-mg843{JF!9;5aWvG%b
z@pv0MNn7ko7_Khj_&*tPH<Y2{|7u?1NqY-=@QOGg3BR61MvOuGWhnlW!awujCPfbE
zJr?xoX9dLY+>=;Y0Vn|&uE0gfgSzKoxFtN|7B^{NxcXT)1d-v&U_AyuIC_Yo#TBda
z(v<idtQn<E#pE+0W0j11w@Ug_Qc@1T2aYEx#jKTdKzeFwYW`1XP6@+hQ$mLO`Yh~L
zJlf-EI%3u!|BzQQY!<MD;qhXeh>Iivh<`hWD(9smsz97NV?}r0?!n;GgT1RS|5M=_
zL->VA3wh5kMH-7D^nrCc+@FUNX9JD|AFe5uG-<C66`p;0^E|rmP-O2#%GgJL=m#NE
z)}Z6L&UGwzkvGN(ry@kGa$0z!%ChWKvl@7Lw|c*MX|?r3`aKckkTabdL;7$zKqpw|
zpyUduzgKk+2u?G{3_Wm29k~aO#ZnD?`D|SbW?~g>1&J;`TFI?el-$@=?p+@_Rxciz
zdM|g$)sSP@uKn5rv=-tnpbhi}MMe<SIEA38#vPnj%d@kx5IpM>Q+n$S=?>^WOGjZ*
zcO6GB?7(xKFr+tcSdy3cxe8?!xe^!X8pnDD-DM(sX>{RS>_uY}#bwoD3N$0`VUK5B
zY_-9cP1?JX7P1?l_HQ(c^rgKn#k!7GjL5x?jXc8CV({VGQR%5jK9Cy|eE9d}+PYL1
z_Ok9n=O-%XCzVx%?W^BiJ;_Xav*H45Oyb6zvPxIPW&D=O=o4{qY;K~vG*1rtS9~L3
z6)H9JbV1RAxX|jypR9U=^~Fqc#ziSi7&iZ}Ut!*0Yf$e8>DC+g%#d~qwpu5~mn;J7
znSbgoG_z8%Y|`p%(g~KBETUEAUmlcl26f9Pa$J|bU9&kw4`aS*sb*V_pnOjE+x7)=
zn`tNX>Up|a;&S++QorbN`rUPv6TE&5TNIA%FxB^mbcH}<kJ(BD`d@`HXb+y91`UL=
zghME~woxn99LR!9=hPj2;Z*rV>!WfADav?SIiTJQX|$r4tWB%eJ8B4kj!1z$5D9zL
z<cNO;^I91-HI`fpsm6(BlZ<H+(oFI5`*uqDjE3D2n~SHa%HXTnXM?i1qG4^_n%Rc{
zPevOOxPG;qO3iD=3JeVF1PF}@7sJE0TbT+@IP?9)a5}U?rmc5PZcqoysV5^L@FI-^
zfP^K3P~DRo@r_XyupNhN7~D`&aKc7BvBxTB(zwKBwHjsL4F_z~efxJp7_G#+d$*MU
zp6n5%To!@0cO`JZ?fTe*@gN;S_4BMXq|XN`7xSqXO!k&eW!vX<0RO)v6Tra4)s;p1
zVnAXe<6&c}N`)O9fQ>m_mnapIiR0SeH!6^7d<oRw@dT+l40hoF5t32*xB6H_{Bs(J
z<>)y{C*k*6IhilMjsG>%-eFtjw+Hu(N(n5R2Vdo%uDgkp*t%9h_5ZbZ{r^<oUtC6?
zB#|;!O}DMK^1ZC3Rl2rnaZU2=av54{zEx7TLg+@?R3Ep9Y2tpi!pF^&y7T=qh0xTJ
zd?;UPG$M5K?Yi0hJYV;V?lt~{^^-@}eS5#oIj{3Nuje`Ec`&zTXL+=m7=_9*^giX%
z=;)^s3P-+iP49f)eNb3|aQpK`G$^d{nI@M2Pt1?uYIQ>`vh3L$jv?VGRHau@1Q__{
zz;daxI^x@ddVcj>eFFP{bWH<1R_ZU*27{W;gDb_j7qPLb0M&n22l_>S^%rb-u^Bj}
z1&R#S`4-4pk11FBlBkJ-u%UB#@qFJ1yn{e!ul|h0Q{zIa28rC+k#L|m)r=jzD`Iu3
zqdt3?M2az&@@5u2M>Bkha-aEEw`Wyh3Qs<UKp&{+Z0vamj)EKbyJu+bC=@2ukX?O!
zcKbpIOKW$1hD6$@A#IY;{8qU#DeVpA?s;aZ_4P>%hfR=mk57LcQL23^asW8MH7Em^
zr#_dpt1BZ8ak1I%nuIIC-S-Gxkcam;peWPB)POrhH{aS!x;gGF6ROO*Z(pW|3eq3V
zvAErTeX2x_h3{GLb$a@JvLgzw6p53A(&Rwyu<-zlb0|DF*5}Cqa}oe-8vRwXJj<Cw
z=%T{T=_e%%0H*0b=_LidH=w<_IeA69X=S4|E|CFy3t=e1BRLNmT=4AXF#|Co6Jv6h
zKg?g$f4OwkAU}cX{Xdl_p%{|!01YZh?hiMbyloBy+YD)lot!9~+IRDgu4>LZZLShy
z;Eht8yF)4qKP@}FFNJ46UpZV&FC)Bq8$m9Z4saxwoTeGMti-gMeM;}k2=SDOsqV#0
zZG3&X{j#Z10VU?I1O;)ijhGs}rPw2xQlb*OY?_bFTnSzlTRq){-Ubm+67`IXZA<}x
zLFQqmAsR;HNYsc4V9Bza1XT@B*Isc}@qZ^iF7pN+@GrynrfJ^OT0zhR<t<a!o&tcq
z4DiLfqGM`!Xom7~I8PdT+u!JBq^}sI<GZL=l|NCHaNn*SH*9qTxQhm6lEIW{z~CPh
zjMa!^cZ|)Bth?ZO-FF9Jgc8=Ar}cMr%_7&g(vT<BX#4r-$as&}60@{9@-o==6Sl*`
zFy|j65@4q59DbVYL(*33y~lwhf8?FYTm1@qJ<w!DYg1ElE&qXg;Ly1>)@{fQxD7R<
z?m0f7S6-%+0OiDlYDQduwDyGM$JAK#VKf<C*AewRx$^E|c}+XH0e3ctGLR)@>_1~7
z&^S?W@Xuj`T*No@W~y4yd%`Nq;j07fQJQlTxQ&8&?{1C8mH8+780=mbo9jdQCu+^I
zDr?z3L#kmmP{y`M(f*N^Lz)FKt5!tZUR4#{S(`+6dVn^A8@Ld*_B0ck;zyx9*(LWB
zT8Jqp$x6P+36S!$u~ss{FS>d$u2_!&?C(x*&b}%yOTB$CrpIp{0@N;+W|`S+4;(>{
zap57OY6r|(8?xmMjrAFRf6s7{i&VoabzQms&aTj)1)0<fDBiJ^@tZs@h^90IHd{Ak
z$|@q+2Figv!B?X|QN9l=>vpDp-oUv{95urGQhK3ct1518zydYS$?V@(k4aeM2;eyj
zB*-S2jz_w>k<nb3Xym5d{P3ZA-8%WbqQrcLzfP~(u~M=;YOE@m#YM0KZoAHskDHsC
z+`i!;ZoPf1XIDs?@X?<&Jl>;IC}l6n)epKkZzp{HPBgWY)J-DNd<NW*e#gOD`dAxP
z-Y_@kIPglsBK=+tdbQT40Q%g`9mPnq+TTq2L&E3ei@<EX%dL@qex;?+5Ftb%q@U!j
zcc~&@W=xWW)*#PdI^RZWB$Og$a(ufM2K)K$-D^z!cE*Us!Q#^A6}nz?f?5VsAKp6i
z?hRmDZ&$`GX)`bI7xsh5kC3qH^<h)&hxs^Vdx$UK6JCNUf&&~ETFA1vx73P7<Ncas
zq9d~s5p_AC5x)Wy?AYKdE05OB__SAF9tMp2(z;+IzjBP+i&-w&QKWl_5Pj}^Tz<U4
zz+s)dX;JLcu}{fGN2(DX|6EQqAp=#_0((th`EgxOS61St{H`BQ9;hwoi2EZfCFrlF
zI}yM5I5>~^-2Q$?1Al>Hhe;=jOOtT3muwVP5EFsRqkF6Ws61`R8U>-<LYJRC+*rea
z<D&`xjjA8|9CwZO9V&HaJ$>y!i4g9QBIZ=yeQk-|0|Wa8!QcQ$@UFGnp?j((yIwzu
z^_@FFL1Nrpy}h>lilH_Ui24j~p17pQxkkAFVRrY@v!!0qf5m>JmqtSMx)Ei?H}wE^
zsag_Q%$_M4&=hG|;z)h-;v_Q-R=a+QvyLIloapH@amn7$3vwGW`>9%kjI)g<9dIX2
zc^1|u*ApK(Dl8QVkA0w+v|(AII<>wrgCRX#=0ueZcGusCd{q^>;E^B5q6#j3$*0xc
zj1N*v&QGxQaHv!iaW|4Q+YgESrOXBPpD+nMD%k_$tzxz~w0icfjf+>%^yISTueFI%
z%h5&S#_uC@{(x9MUXr-seZ^rn_ehZq&$2;Ut_E(&C+DAgfs)mx^+iYgc9Ue2NxzT$
zdum-ogMn0O->VQTw{GLeqqB$BBf|V-L!i9pLEMea@AFd{_pHS3GtZO}dPPD&on*8z
zUDVPx{(1F5ZCTBhcuACSqRSpfW2r7<tlNp-B;)?v_UFOR>n;`%M$2yn$Oy=tA*={E
z#=()RZ%T`0)@XrVZ*6&+NRV~{8PAJ@=w*xX0!iggen+)6W5be{6|WS}fNgo$E`>sR
zkMvkusAl9@Y;oDMq>w#Gv|cQ1>7N_q5_ZS<=}Cd44N+#(EaJsU<R#r|YqW^3$RB&g
zQMa+g@PiwKwcmr0@;eW`bH`q;iY%_1q;(x=J+RIHEy>QZe}f6Q={SI&YwQk`Ippz%
zvX5OSoxOll7hJr64A4@~NwN+`$dqx-iyJGIH4|;Rbc@aeM{+2r+r-hTZ>OxgM%jz0
z(sKw>ZPolkAnz?pLG|_+*MCoNFo{<*$u3Yisq3-d%9VE{#Ht%#_sC7K{q?&%O*=By
z_&=KAnG^vqZ}qE-6`Ra1Zb#KkczD(piJpKxl2Hr;ou-im><)r#)|wPureu${uWu9B
zk?fDv0i*X5eH1My7=Av3vxzIMS<Pfp;Sns5M<lSrY#UQ<f~{3do+jFV9tLa;!}oP-
zxiWtz)+03(0TYSnrJX7FCETEgk*A+cgkr?Fo=M5hi88OVc-P4F3cEen$TTrP@g)Xf
z#u*u=TxqH+lUYXCS_yUYf`r9MX9XAo>dbo3W<{gX#27uI1ohmC1Chhm(sU*}nx{Ix
z0j*ga8KE|NSeaBH;K2KY_j_^3OET=@gooXl*@VxUv*Xlic^jv)w;}R*N9%-w<;#v{
z+>x|lhBo&-WL`|+B!*?hLKG*di>EFpQ)1fSW5XQ>H6%9p`ms_gU!jdNQ8{S3&LV?(
zzMPJC8Yi~ooVYSll|d#_c&>HFb|6(AIfIPlsLfuP0%F^eSg)&JY{~gRO@1}EvaktV
z!aR$lm3!%0u6g@DCw<DH4O}#-(iz6{0mw3P^|{3vx$>Ul?jA&5e!u8DdMQM96Eg7K
z2FJW8*WF3FL2hN0k!bFsb^Odd5j0xu(U7d<rL8$Bp%r5ar8q2SW@2An-sKNv@ANnE
zP`RX0+}OHvV(kh)P}50*uw{JVG}n;^$TcVN;sFIs%RCE}QV)Na+!t|r>I(tWp9Dg*
zNhkk=?pS`X2-`!Yl@k{N#aH7(<(%dJT|BkOZe52m(u=(QkeBnsv#D$bUajG*(=_RJ
zan)UYOU?^<{Z6QZVo!>_{jCkAX}(+yG%U5duTdfLfBo{Aq}ZFg?A$b_o{UrUPVh*;
zhQBFG>>Zs@>|Me*TO+Nhn<ROrEy5h1LgAWe=#H|0P<sBO?xRXdv#-6Qq%_0Gdb5?2
pf}sCVz2p#nZSCS+)1W;nnz43|vDdB(u}b)7y~S?x&mVgq`#%yaC!PQR

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/logo_tm.png b/docs/2_39/stan-users-guide/img/logo_tm.png
new file mode 100644
index 0000000000000000000000000000000000000000..48c9769c7de2c7c8585b52a5fdee686aaa47c8b5
GIT binary patch
literal 21065
zcmZtt1yr0r@F)zki%W59ad$6Xq`2$iwz#{yJH@TI7I$|RDJ|~qQrz7?e*b&ld%yRb
zclYdhCdo)9n<tsfOd^#OBtId2L4tyU`XntSrUC^8(ERU*0QXUYYqKl-@c=lhNQyvJ
zj}sn#6qHOgrOo8!q3Azg1SlW?6AJo2l#ee|xFr<K|G-dC)E`eMDCj)E|830!{y(h%
z%{=J;3;%~Wr3x+c!9>dvr0Jq5FUMzWZ_8|GVsB*1>|yKh9|F`j555o3*3`w2)Wg=s
z&Y90cfc$?j_&(tOs#(ZM{|CjzT7X<rUWruH-pQ1di<y;~m0S>sl$7+FlZhFhikQUz
zCjTf2kXyL8IPkHsxVyVEyK^wxJDIbv@$&Mru(GqTvon2QFgbhLxfpsd**R1EFDCyF
zA2CyBV<$@o7fX9P(*O7x8ri$L2#}NiC+Pot{#QR;EY1F}Bs=H-t=5NvEdRBzuraf;
z{2$&Qq~HFl<x_OBH2skLAAdo%Z~ue&|El}na=x+rC;b1fGyhA||Ec{@RS@YL%l{cR
zL8Q<6Tkud&!cfv;-$5RL(`<MTLh<^C3iBkVRK~x&`-D<(jIzH0ay&dqGyJU7;XB0c
z=l+m$k8iE;lK-IkG0p;ZP=4bF7n&?B?t%%-Wm8SH5>uQPSKp5~flJFNT%#xBC+hwi
zejWTXGkUMbOUoH&M^;cWVgU%~0CHiN2q+9v;QvP<4;%0c1OYK)T!i&n6SRDCks9A3
zLfu!YAY$*`l5297u1ONMcf2q`MRD1T;o>h!KilKYUbSSem2e#W-KMr5GOmf`!kP^#
z*{PJMX-Rc;JBaIQ>v((U-jl6dt0|rkBh`eFg1|+)gv?z_e1}}%M!WnP{FlCJ7tVHy
zioa?Y+NeM<;I>GLwDHN7E8a$750ZY<n^c(2sW36-NIh>geF>K|Fnr6$V>I2MmneEa
zDB8pG9&nYY^eI6i0NbYMwl(HdPfs)5r4x(Urh(TfV~n+U4(T<MF@6>dhefX{zot`a
z;i>&qGf{35+kP9T4(<+GdmvQ8Y|&w&;M&S`Ktfy`9U%XTKwfG9T9{3{x}LCVs1+Z}
zei@}kCed4fu$M1Gb`gSLzsTTk!zbcGH_vL8%73jdFv0RkAU@W0KJN512vcD$*CF`O
z)2c8xh<@)lG*nrQX?*aM%g?j$XSVYCHetkhdmd8$)o@}VWJ(CtO4SIg5raA}sdZ{E
zP;2r%czBTdGDz-2pZLIvy7n|=<QeUitdTFHK7R82nM(7cNu#KoGUfU|Zysd7BqNM1
zT~6oB;_ve%A*!#Yq-oUXM@rX&(Me$mP=@I%cP?EC_PN{-Qq`O4+KR`<T`Gd@V0k<K
zk|KtV=&&0Xvv3$MW!^InKjN|e-m45ARuKX`sg?D_nIv31zPx~(X<o|LQDkK@!7Abm
z_$GAeNS9H<L5PH+n4Z$_Jcm1)g?oR!ahd4vk3@!xEP+K}&NbgIWR_$rJ%$%^H3nU^
z)j#kFDa-sd;Ox0VSAO~wHMEEwYZU6Wel(_GGE)M^hws(l|0&9#x3)rJuPAs{E2y;e
zyM}ibN_1)?`AO6A0FcaLc#a|y5O%_2*`+Y?zCn?;PmG+$(+)C)IT$LbN4=EHauq+8
z^G3_?Ggt>MkXUU7OO-u|vfV5ojJU}m5t1YVYtTL~U;IhDYy>jhbYZ5$6oH|4@TM!?
z?=DjhX+~eRe4Bjq^=J-6=~2{}834K$n}i9&Zq)dvBWdqOZ$tIC3Zbw+py(F!@>5F+
z{B2h8bxfyNhS=r^P`QP}lKvI%0V=jt91!w7UB2k>%GvuIEJu&K4jL%cwNh$xe+)_o
zMCp164~Z*BZ#~c7$)L*FEa~S~KV`8L1-d=V{(#>}reEF3!aYmIvsO`)LG&pVY{iwn
z_f-<pLh@I>cXNLxl%}Mn@~d^ESPvs38@Vr_`?#IBm#uow3D^?#@<Eah5wH2bE;tz;
zW%TC4W;cjRckT;}<SDw!Ssgy%Z3q!BOVacc(6P-M6xvsb_sb*7PDw+`OLfFu%Db4x
z_wVK|s8XuHpacubsA0q+W8sX*&(p&wg<EI*7Fp_?UE)aUuzHt$!zqqU?Xzj_a~BK*
zd~DrcsmM2J0>y3w({m7yv5Zc+ztsi=O*}*@H;+g`hQp@sgXXJ5-_>+x4SK4#Wv!Qa
zne=Xn$n>@Xg@FnY^26ZVB7!1<vmVBhzoJ}xm8<}jS|n&UjhB&)Zktj*!R;DvtB3jw
zw<yHTSPY`1aA7BG9%}RYl*Ig<)V*jt_dlxiDcejRMC*x(WF-4puX}|-G;ZqsdNj)x
zF=}u{ppZ0}0sPZhMvt?5ynR6_<*v>UYEdqgT6W=^UlApObgf>Boc?vKh?re}6+I&Y
zFku{~M4lh~+Ts0Mc=oBNhD+ELsf3a4K*aU_-bXvO#<H2V=cm<ymyJg<Zk?b!5}1KF
zKOa8+R3Zw!+vgEyi(>*Id@pZAI!$}&$Mo~Prj6eq&CK<uBbbiHnc>KWxEI=vE0_Uh
z6fBM<4d|VsXky(g)e@f{2^!F^7dQvkL-W|*Ui5JMOEaizvGZibhvi8H!j{a;8YKO_
zFIR>B>bVuANg}EL=HDFrK|n)TI#acz_Nm{5F*S-D_??WEIa>|whok<}<X)q3wQjF8
zWLWHK3Jp1DRT*+_-J}#?{WD&Ljf@mL3!R%oR~4&ZuE9L<w=P^bq_#W}+a_S`@uXdT
z|GX+iov1Q=7Am@6Br?DX0UQ(DeEIs|+SB8){$&oK`1^M$$kUP~!+q7)R6=P_^AY9N
ztmf7`EONbHRzM?68x?9B!}hS~9PRod)w#ZM5RXc&pUWtgHTT52>X0DsG6JHXu0q<!
z`ijnK2tsk$4+^E&-2D90L3uiFl_CROS6n19=(dhyK=Zg{@1@&~;YZU`L~OlvfUp^1
zZ?YeX0TljMxB~y!_^CepXL+ByRh%ytVOC<4u*aqv-#YBp0=6Y~N)E_aXGdgd2{4od
z0EWe}^895WbJHEv7T+oklZ%`gvrZ;UR$ILt`TY-3LDL2cIIB<MpF5RizGKvM_KItP
z6!J!7aFEnA%OyylO$w%I`}bRpcN(09)I3#1nThEZ_1lHzOnh5diZ3dVcrsJl?toDW
zWf{WQXWZYkzFnEKr3c2ney*#JINpqTw--_G_vdIHAd`(W)I^}uM9Xd`_kS*}VPMQ9
zp}uE83puFGKy+zQDJf9lq-~O?HrDxR(!nn1?Dr$XGeqWNuG;;H{sHHn%)3!}4D)hc
zA_g<7*c;-0q<J%tlQOnh34p@SVYP^a=0{cV6W{J#tqE2ylb|Pu_12broj7_)Ye~H<
zR)^2${q)IK?eoWI7@+`opodh3N*FAPy%y~1!Y->rPF6L#Mu<PRNkzXbPS{uJq+N(s
z95%2pJcRr?@UkkNM?he3bqcFhQH(8tn~}NAzzO&NeSNU9*xO)f2o#}!jYiMWYw<j4
zp;^3}Hj6iZ*f{P;V8;YZe7qjcn)OP*0+LiwBzDjP6w2yA6Gjh?Vg7WVUD5_kcR`Pi
zj+%Oymq>S!w>o}xg|lsmZqA=7p0yi(NV8_UWM>4*xPGCZE216aDi+zE7I%bpa<goP
zWk(>M`XFg~d~bWi|28jOQE$?kQi|Y@AoYNZtzLd1nH_2eGM+p%AVf}@9j0n!XJc^p
zO>Z39Hw#hJcv`#(7g~MM-cojJ$bR-9Go|@3nGc-uF=f<Rl?LBB2|0Tc)jZp>9sM+3
z0y@*w=(77n#Iu6*#@kFhS^P)L_fC95msBprdd2geoj~~@$s#dge7KjT)|1s|)|~HR
zKfhuWgWtQ&<X@`i7X?}YTbqA62FnH^<de6HLZ7N;%1V`njKt_Mo>;b%E?7%tUT`wm
z`f=$P$$zcn(&}cql`=l{=6m%jGs}nx>jNXH;b!Vvt?7=iL6zJ5RNOH5mC^2E1{id9
zx)_dloBCRxH*zM47_FeP6!OGifHae)#bJlLpv?d;hQNNV=)a9T0={n6_Dg+Rx{C~p
z`$Ba1FRq*?bhg`<Ta$*fVUrjU+&L&2+bZR+0F{EixBWsw8`KR4JDqqQhws5oDIwjv
zwcVmQdTByuBS;U;oj3)eaT%18Ujon6%MnBbFwzRG`lk#CiX5A*ODI488#2$<sQ47=
zo5pT~bFwkNuehu~-d%37x9mE1*Cliq1@+%dALM%nJz1m1TjZvZ%<_}5n1SuYb9p(_
z$ry1?I7sY&c6Vj@e+SfadPu@)nOs7n$G&lhSAukRgpl+~`~FM`{rUP@T`gtKkwwD4
z+ll3umN$*uuK*EjsdE^;Gmh0$vkizwrNao!fe8T;1s^||qoMd&^yExWCXgT!Cf>Zf
zwEB6@ZHZA<VxMqWbE7@|${?b(4kVT6M=4n$^I0L!dKCw~FF^yw4<`D2Z<E#eomyt>
zMiEP1yZa;{%<o04*HS6h7j<S=xRXaUvUY9adDwN>x4MlWoF7LlEt?p6#P8EKlVc-)
z{#YWzqZsCc4c0C^hs|mD;s`BwrMD*OX5rouBlMKTJ3$ycri|lH;e(&fgix#W+o;y*
z_5OtYOK(~ZU~c2|fA6YwGF?UcmrMbF1SN{08ub;Zw^L?Rk<8&(kqp>9CB$ukqW>lD
zNR%4A^hX?*iCbz0=`M{y<A}YIj>_8#9G}V`XmbA}VA0HrPw**MU~_l0$DyhKI}Aqb
zB<RiAe%2`fU*7(RJGtcKAu-t+|E=DOOZp^={<rcV<=jsmX&kI|vNtba!~g|KwSC#D
zN|3xk_1r(}`JmTxw5ttT+E9Wfq@3Ne%m9y<+z@CX=kR!y(@gcnf1ACJV~EgK>$674
zoMm8=$J)*O7nQg4{xb1hN@shR0Toc){Cnqjo?lVNul?WfUF~TPbnyUiT#?t}j<45O
z9jKOjdd^69WlaPuYYuYv0Nd+=6rE9i-ocaiRS1exI27f~73e%qT)i6i?_m}Zx5GLy
z6Q_<U(p{(`Qi%<-vyUtRw$&l-HDkr1+ml>YX$CiTT18@Twz-nauV~>illbxJX}hKj
z<yyrmX1S;4b!EpabxYSZn+zYkxR?Fy9e>m_I=0vs5a&`j8SYc`!>=b^n-@&$h`oz)
zuQ3k{!c!L?@!`1&K{ZqAI<&bU!+@xiKxq#5Y<gvhb|v#B$`ZjJ=x38a1bNL`|ChYY
z?5g4yD<&wa&;Hb;zf}g)k6sK>>bLy_6>=klH(?fQFrOxTnJ%1?nZDm5-PNISZ`770
zcUjWJ0Zh+fe#)FXRcW)n><-f}pD0q-f<_NvMir>DGn2EgPY-EkZ}jH=TL(jp1qJLj
zR!q!V-NF+y5D~0h$xs1|_}_KPO51Y#q`5%)rdiT6SoO|Fv&+nL^o!;TRUEzswtULd
ze=G|{n$`tqH5L|+R>@EXFx6$mNYh~)R(=MBoIE|oFge0B=-VHwoVxU1;_j%cRIj@k
zJ6uPaR+t+j^hQepVt)T+_@Ds6Op1)!mdU+B&-Z)-vEQiI&@(Mc3a&c6|H9l)dN>!8
z7Kg!fh>JjWL;#b0f-Wu9kiw=6=&!S}^O;|@wJOZ#gN=Tx7HwAJjsC*dW>sa~<XSL&
z7128v9R{t-ml0G+JC-qg9|r~RmbLOJX|9<d=qOp!vBQ|4dR8S{_szAVxn?**yqGTv
z0t8im?Jhs<MsYv<d=V&T?u)zP*_E+ePLF)2SaD^*@(qr=3{i=4m>quys4ED!>7(mN
z6;z`{4%8;={oKLD=6%L*m7u|ibT{>JWU`~1-`6?O<zz1_P70zL3V=cB<qhoD!Y2`a
zXiy;sW>ECzEkTOt{%Y|STgB_)2o)|MBNrmY@mR^~1k68&Y*wQuX=wZ~u0~|6*A$%u
zfQm<97Goby+Ig)U)>H;H_qVi%B#U%SyN5by6$W55zhQ#^YG!m^0r5<yPym7*#pI{N
z*!N{h(S%OOcL`0IPA$n&k6dc)h~e$t$z=wbW!GWst#IMcDg`2b$h0$3^qd8Vn|f<o
zlh={buU7-99AvU&P2j;_q2boW_tI@}HSa^kfK}e$!1O-_@>Rh@*;Yu*fSdfdgp(7T
zaZi0(Zh4flAR?yXx1AYkh5XcDhx15zU`=N4COIKy9%3aaR0;j8FxYM+l4&8I^c^LI
zmCRev=O{^Dm5VT?yQIn>`TR+1393cWoYW2dV>2R;5f+$P8X7bouR#>=5H*7kTf_^J
z?Umc4MPJSp&nVcrR@umd_%M{10znS=Ibw$)R0KEJnA}%FzDgGl5N+X(s6-VQGc*_v
zM+`cR;s=$sWwL;0KZ6yY0<a-1CF&D8<Z7-W$t^pg#+G%)Nt23iwFtV^ar&>6hD@@e
zDJDQ36YN0Qy~dvNL)QfH*UR@7@QN*Bg;X2;%;2lT&)-<rAE0QO%-mN1YQIk54|mv(
zPv4usB>$N*spN<J(ZQ+x!)sGJj^Xng*<-?D-34MW5dk{Ds@Bu{Qn1}K5eT`tQsF5e
zxnd<2?Et*46!4xcI?PEL;o?Y%9^tM{^iDJy1}6^@tZ(1^_3G*)#Y%vO=*sllA?SX*
zpj0Daa$#2DGXkrdtm<CsA$#(!8U{qUb|g*q`RKv9Q8ZA)ciZ*XF*R_ND`no;^Ha=+
zL8WU>;rND9eStK5VWEV5auvBRV6ZNeP`}&Rkp{kDzyL&Jez<f5v3N>O#kNcZ4fcj$
zquxc*YV-&iI==lX>rUD+l9zEp*NostiSPg>e>i|HGJCfkPM#cP!oPx6gPPI{HR7Fa
zpQ#TVw|H)FHTixdTcfeM1Ry*}fI%aX$rU~w1NW;@Zz^d$x2ApfW452<afR>QWdbp!
zqTxK;1felVGym%W0E;$uBK0?WsxewCB^oU4m*`$An`eGH4YXa~Z;Wq<=Rc)~=Ok^C
zvk|gHdSf|GahRY?nFxn779)mh9yZwef0URcUPc??o?N96sB~A&Ei}g8(#yK*v7QtK
zr4+N$haiOT0%+)`kK6*L%|O2%_~v7fa5fmU0XE~g1aYz>j}O@1?<F+`4%_>MSH3Ev
zL4&+i+}ey;VQ?k~3>ZCXHpe272<RUc!Vy0p7P$gMtGd7+P9YD4%R>CZgbGg7?*Yo{
ztv2lT#?f9?_?pQ3^-RQynr7l)=ZYZ{20qE6C)D{xkAOdn3@PeK6A4)m?Id;V`KW0M
zv1swKA##jN79guZ#q~ASAW0QZQ(B-#eIUq|ts!l>GAr3yaovxo&8E>HZAx5BI0+`C
zuW~@rbAjhDF4fWzIv9>a@cuoy-T&<{&wymvhSc?lh-G`Q=cXb+pE+@j@o8_^S3=g-
z)H^AZuAtM~QdTqPOSHFzKn{a9cPJHuRdO@J<_<66P~FtDSy<dX%|_hLuBxr%#Y%#O
z(H*0&<aX2w?ud5BiHrkEg@;e3cw*P;CWVIN50kF)LK3N{UW5q8X^I~-L(*~im^is3
zQ+zl{%20aq77N~B*GFg9hR|cl;=wl(eFkeO(Ixr93WLA){Xg3^{l7>rWXT8*xci!7
z6v`th4vK7m<npjc&IjeXJUUsdOqh|IU}&-?Go^0nc)yH}&9y&NjqF`Vr_4>i^gm~w
zFRM!ot5x$Ksw@N4nUg`WKp3wU8S&8k2B}v|J7XF#d7{0Cnkp#&yL~VBg+JwBhLxiC
z63)`Iywa`oMKgZ)(i1d%D~m6G^Von!(`DFIN`@&>OV1Ra);qpiR+c#wq^yOrkNg6g
zKEn!iwr>W#6#5i4zctLRyr#tgT0QG+)(Y3d5>3j<HihAR*dUZX($UJ1?LIP>*N!fs
zQAHcKyrNt6k}M<7O0lsCV?kAiAwv{pSi1q|WheqibI}rsPPYT$vG5AQ56Mw#3X)Uo
zOyWJ$g%4V^(`OY~z#8#Ps>a4PMolg3O2!_eI>x(eE8pxTxccOhF}noCBi>*D@wX(U
zVacyfvPat9{!YbM;YMspaq5<8I=k$L6?+o8g8Y5>*^kqw`?EJGEj8Emnzg8tLNdcm
zEVJvmJub=jo^&)A2jryG*#nwowKAdN^nx0jtz}B}$(=Od!M%n{y6m`ZoSv&U`g>ke
z<yuK?;mg*jvq7|;LM#sa7%9XaH|6ME3Q`|QYu0$_h5uMYUxYSnCYz8uG$t?s#_a*W
zfluB@nvl24!k|!{XwM+I)2IidKVCl`bT+Y}N}f~N%`TLGMyV+!CsQcRKz7bv?^79a
z03Iyy+CcSM`Kk*CQs88h5Z0gHp|Bhc{%;Do<k7<SrAp#IN^0CN8G-D-$dP}cRN1!l
zlGLSt_&kL<=-g{`(v(~V79;$euouO{P`ik@vAP|^g!RaSu|XlzFd_UP-q;_hU)cy*
zaLzZqEDf8Irfp_H-TFfwRzbYaU70MER~pp;jEf<tJN-qvuj2kku9p%4on%G2ip2%5
z_Ie6WB9Ta@^xMf%t;IYEL6M9qocFbB5H0^COWxtuQ3GCRmD<*S>C`PRRQ}7Z(GpPU
zz@r~CNJTZ@X(sO}{3gW<kf14*79jqC;ty5eSFWFDdTnl#XgE_tgf}#VU_4+%C~OV0
z87UWMQIQ|8TG9u_VtTFR#sNCOVZ8)fA0{M-Fpc(y=cR@-+T#p+kL@q~h&Kv44gdK2
z&VQYdB3*A4<sppA2$mNa*#WB(k|CM6E&eJG70Gv6rqEgv<AUs-{+7&_vJrb&ou%K*
zzt|}MhjMzVfZ)|H{XmP9=rUS(!Q)W6N5#G2!4F3k&oWEcl>gqlhq=dZu18OcXL|%m
za_TR*?Mn%s(PhaE^g}MMpQ~sD)q>mYsl`tu*~Sj`AB}m`%V&^H86jqQ?#p5Vr@JTZ
z3Ikk5^E)9PUOHF<!;TUELyPT$*yyaGRe8b4pFT2Ty*j`tz6DlD)B3;j;r+Cr!SqNx
zHyq*fH!S@4Ua(i{`@&3$0OIC6iBbr`E>ZP9DRoYW4mK0n>EFoM2Z_9TXxvA)!N045
znTlwgnL<a;SPl_k@GlVGBTjb(^IC_2vY~^_dzb;IDT-1tw4L&j0<Qqkho1#vo)d9S
zPZD6U_OE+Q$NOh@d-S<Xnn|c<WP3-*M)#hBeL-!aYq^Dm0*Ddb3NYNfF;O7RUwtA1
zVH=?%*i_``D8RbJYr)EWGlQcmEeMxOedw;h=6{rU+liVl4F$V_0qIkCQ7-<cfAN1k
zA44ry60^@3lB2@_<;_y}tvLlhVT@|&m&VC7>G0Xa7@<ALwCng0O?%&pz#X!!p-MbR
zURw0@ob4(U!C!R2P)JNAsf27`(_42($ZyWdBI=xvUh?U-DdL_iC7mps&{FMvvFOR=
zfFjp!i$eDazfd-yoqgR-ATj7O68LFh3C2V8(!c{?eE8%ub~37<7&=o%X<)7EY6-fS
z(yu(J9i&6r%sA68*+|Uq^pKixgGF}=g{nfXeEcUr-(ordJK%~{B_OFSbw9tpebL;H
zlQLxG8+B(q@LJt4a91hIx#6W6`C7{#Uabxtn4vD%kH_0d9NhK1=GyOH`W1I(VgRpG
z;QDP}3uT;O_PSL)F2Yfz?#ofQRX`UHnJW}c;Q&3|$GJckF-F=gM$hs3C3wAY;+QOm
zhwca!br);r6G+ihg-v^Y30fd~?zdyUl58gLOTe>j4yb%>NU>=OE-?`SMub#Qo2CyB
zDXmH{X#uHuZrJ&m@dx{j%w|CXjVRV#hIA;+AjNyNRV+)u+JhC9I~~p)W+XNv@P}Dz
z?@D{8dhOgP;T$xTKO#cz0&IU5al&OwMnH}8wx|pQO>MGVHgKCOctoX`rvzYJ4D5>#
z_C$^e*&CtqxqXy6PYLSUx$PVE{egP5jIv4nlOIO7a$Uu^yz)HY-YpsOt0u0mT3`@t
zXARv&hg45J-n+m!`B{y=^^0MmoS|Z`4g70R%n}yYCaspik6RfC^&oHc9#6veze2!1
zyDp0&uSyMg37L(6iDwc&b);N>cl`1Z1_$H^)fKar^sR>${yeMm`*==ITod_pK;nAE
z%HC|y0RaYrD6k)^XrxZ-igppT!#}~KFTTP+J|hv4m1n9(*0iIvZVu%mKnb*NCJBOu
zj|&>xUK1`KTLqXJ*oKyqbcddlnTzx|stQjh6r@xcmy*h3Iyi)<Cq1be^m*UPLTLW{
z^~Dj}7r<&Y0a;Wo47a0tG(e>sH3vKFIw0~4LRIAcU0#AK&fOOs9G@Vw#?)@)@ez)i
zS4?;L`491{tISTPrC4sy%jJm1mxe^u(bh4CT+gmD*6AygAwXwiL?AY$7kn&dUvw^E
zND#~AR8K+uP+^bmi@akbD|mS+UeF83ifDd8ixG%G9&jn<A?sqk>fXY1c2KqCt6hty
zAZ2i4`jbNHD4$!o$@9!e7q_C(smg}6|JoX@cezhM$+Hs^ZzY>6uN;a`R)>S|(M2HH
z|DE?H*A3Ss<|j-{ABa{sBML+TKuP*LE+3<PLDUQ1^>?8p2sYeT(Du&)fyl(Z(I+TR
z(QmnEyK>IqIB$2ChMG;LLdm#9;@L;<JN~3j2;w?dhj68<ube1ZYy{I`J4sltMDRh$
z)pH1dGjO!hY%jJ=iC%`pCH%{=<JSF`xqE|j#n76^dP058E-bgxZz1o#M_Nf%c6I>;
zL2-FNrseM(6T!P=2V7%PGf%T^$y%x~%QgXZT3(92Z!ejnKAPv^lUnXD1GT;46sy^n
zubOHNIK&1`X!ABt#9yqfd{{^$G+e&ULm=`b!g4ZvDCvHvMW_T=1IuOf<mlgku0)lJ
zYl|I`5MG@N-R`b_g<$J$w-sLK&o5^-1~q(~@9p*sTYpnF`q4U2^O~%LZi^(k2hH!o
z134zb2Ts0!Nk|`{1g>K_Xv@*@`?*&SsR6sRYg>BMd8|n2G|nyEN9eXuACgX56E-%G
zWef+)?Hz2I7IVo#uJYDuuqOu3s0iZx7W0Hsf2KVe3Dl;j3`OKgsUYmIrnEPdk~K(N
zY*Tdz8Q$DZ0yu<!hv2sgf)Ee8F)U6S`>*^@1Mh_%;g-L!4fv;1<y+kQ-f&T(eF?-p
zwTY1LpTkRz2>6X?GT6|z2=k6zxkjBWi>2GW0f=@G6}5j)U3M_MK7X&nQun)o)#O|D
zt&ua<_?n6%x^qbp|IGpaCf5sBySddLm5Mx14Y<l%!9C^O1a?dvt(ZgVQbV`gFPP%A
zKHKcpSu=C&&D~f-JFi?v&1M?&H9%M3om6b8JBc`^3PXKT<V8+|ANfjwuaH&BR8BK#
zXI|)yRj&cNV&k~&oYm5-X+I}6-j)SHF;e>LC1ZrxTX;?s($m!d%a=(ZSJxg$9)^l)
z=RvZ~Uavfr6By-&U5!%cJ^qQFuSf(qW_IApTD!pSN#=#Vc}LQSIvpPC>jwxX<AFP7
zwq4ReyBF&EJS-dA&M!X{A3!IF0X-}i>StWiDngBkO`c~4Ohm61$Ln(Sj9tpkB4%3T
z`0(Ni56=;?PYewVr1I}+FWCg@<Se2Z_`m7>QP=fM)z&j-tQxC6&`as)cA;Auy`1UA
zPs?K_lm0?}K>=8J`YZOV^y1Y^gdK-;H|qC`j-O!wnu*FDQTO0C=kd8tEF?GQmGbJh
zR#)ghvfKzsoj{r9t8jaVx;~48@F=c8>Va>Q1^xMy@t={g>ijmaZLp171(-fUW&F&a
zxxapEc%PtPuF6K$y!wOt@Pv#4j}fxTus&SY<g<Dh(7ydzou)}djcxn%$1#s1y>T8p
zky{RN;ESHDnMS*HjwJH@;jRdS>Q!QIP(BJuHQ;l=xvch9xlm=;6BdrXzWh=r+2oWG
zvSf>WNIgr7$Z|v-X&9Q33P)#K@)bH@5aB@h>Ogu^8KRx#(`tknss+I(!2B1-vsCIX
z$P{cKbO+EJ%Ql>sx&1n1*dTlxs8^ka&-JGlkqzTIpt(_5kwiR^(rsLTS@EV6%wHu;
zys5uVWp2VD)lU}G<L7}-{JbK$2~tjGV9Cr|vSCu&Bh{A_T7jVxmH58=2QhxupH=*7
z+9u9Q(IJ}X$un`eFvzx{&g4t*ZOG~DPPBxB$9BKy#o{#y(0#h~w#+5ydd#p%<aDd6
zTEj&J>8z0z(EtmAPeecnC{2x^nzS<}Ueo<Xuh{@yK%Sb8_K76Ece8^;MybtDY<Ldd
zPR8ui#>%11i|E~qpYqcexL1dk(4z1CmP4>q`H~Jxw(7(`u$Ec{b8c&^ztJzcRm@0L
zv_5rPhY=8Ly$jUAr7(m+3X14~1&Uu}qjgz~U9zm-Yf&1l3KLb-sAv!iZGmnyQVk?4
z360Qsl8YiONN!sV$8xQ|?}m6CZMpTO59OuKtK7MFa2M_H9BH*UQ%)%Wbqv)P6*w#y
ze`Rfia=1iVz4M8_M<T#T3ES64`=1gc%~xSIEy`O0=J7Dn7DyNZuP7wR3#zte|J=3H
zF9;fKblrYah_!`tR(%K!M=5!m7SVAT-QuDS)IoN0;Uj6TQAZv_s)}hUUPg-`(hkb6
z(vyo(JAFkhzw<?tUJ@mB?)yquLG&GBCbGhmQ7J$-yjOlZc-YsWR?FtKj}F!H`}A-@
z^2S|<=CM6&@KXT8X$F04yFU4%+`o`Re0W3lDqQ5~@lOgA+j5C30tu`1nnW{(|A?`S
zAPFGRCj0`Uuar!YN1p%~Hvbm!bn>=4CUlj%Lh3nx|MqsK+lvGPC4M)oa&A=pJOV4k
z#WtKFMa~){E)nB7h|wq3TDGiNIm$=)zK=&M={~SUn6`f`d^EsB{Yrv^pR3xnt7W3+
zHX5YWIU17*xd@bn_QekG7#y?7eMwO6e2UCq+U5OaS3^=eafRo|P;&$LY5H-`td8R~
z?OG^QZIjK|dCMWnE&$vp&wgX!4JRXFgst6V$z8abeJ|wp{JXWhsU}y_mpu|dcyNav
z`@wq9UW^gM^RJfJx;?D`iu_QVmnW?03}0aN>p<cYS}1^HBJjZvyIJ(u_e<%^$J$0?
zvZ_-{C<XD7&0@p<>5V8xlXP6$@~4j~_u|{)^OlaqR5u#YE$7xWqg6NwOqRmGyRwa{
zAXliuJtjfn`OE_F>UM>t%{aW-R<fFnc(?c5v4h8TIN0v2E}bSTL&o+}EFf@Y3aSNZ
z{%@O(?MaypA8K`}{0>X^mqfUTQ%_oPLI7c56$>vyhkAEZ#Q@j#fVuYoS;Pfg$C~VN
zRnT7Vs#&Z<fG?MUjWN5~7Eblp)6=Nj=f1=M1XfjFFwM!+0#fPXBNY$PU9Alb=Me~)
z#dgl5>B5^w7Nq*<`T9ee58rJah~Hp=h;O{A!04riYdXses5=%}TF^1v?_lXz?X)o9
zI=w5_NOaB(LvZ*PrXmcJ2#*>pt?`1C7~Z;M#g&(MXvBn3+uYQSOMmiuz=ZzYj5d@8
z@Kt8*NUlW`1`}Gx!wA8y0%eMg)$DDLmHu3Xb&v|K#i?9g4Mw|y9z1cxtvA&n7XMjh
zOFV$>U$g33ECG0C%?y==5@IcXbio$S?vl)4l@w9WLQFy@P7J8aj*nQMbJ#^k;d4eU
zipxOvMDJ@H#N4s_q0)j|8JiH4@+=5Wgwet>?@eZ~hmg#RAw|+~gFmIvq^d_hz`+YX
za;NM3sgf1d+gAn=f?TaR+tg}X(*jpUiDgT$^s&$reoE}>BB<7A|GVM>H#&waob)jn
zt3E0(zOGwW=dojVDy(!dc;b^gRDd0OQ^OFj&+IYWK^_EhT~AypjYtjZ&X-2f_w_Z+
zee0a|Q{=Fm?jVwSw(+$u#dbK13Qkkhs=iPu$1Z6F*-mp+)e%p8D5RovR|t*jX~v{K
z7xJy%6;KA{>e8Fs;Pb{^tdrZ-Y9AJl)CsJsrnD2e>B&Ba<v0W6I^GoUzgnZhdd{ot
z|D~+xHw!IX;qD7`Ww4}5my-+DjL|8a4Gwah=IbH7le;g!`AvHubSszcHpY(6>&`Y%
zZ0;r`pGy%CfS<W`{boI>>&XbzCLosIO?)o#Yq8Ugkv+hW2egPT-5QB&F8{f#PIFMI
zS?@tUin=e1PFtZ#P>PaqWf1ONo<e;V^22*l6oiHw|CZaSfDL9ePe7${CQulw3L2c$
z8QQDN8b*o<h{$819)QdK3GEtw-!|VqgG(LAHa9TGgR4__ENf9A0HAPUaf8L44q3&Z
zrR0N4?yzoNAq&E72+omTokF@~>zdllK*)R<F@S`~)&!%}U6&>FNrbfOVApiFZR4N3
zZjv;zq;D6!*5u7mx6S?Ja~G%HUn}jeia2w0kBu!04xnQOsKCXAR7!J@adARtx$$di
zPg!PD=gm!hXP4dTmB&AJFsMyW{u-emlhWow&#VN*<@c_4>CeO|N!xLY5I%C451g_+
z37V-XIVYngJ-mtkI3^#PT3Ux?h<ZEX8?scPBRTfaDtF=g3pE}><NKIZTdU<^x+?vA
z!vL%6@1M!sYhzRNk9z*<4OFjuzCM#FFfJ#bRCfj*#h!#eQNTl%*URabq|$p4P2{r>
zMaMIcg#*ul;dG*i5)6vN@28#rW}>f3h4BP6z_u-pS6)Kd>GRz?@!yg8>bUa?aF7<}
zm-3r%<2#8FB0h14XLmrQREDFt@X*4rx*@Z4vS@_+y9vM@tfFEdlo`dWuG)S0hrS<@
z!!kff3`G7xEm&V@wJ~oDw-<C(XFJf@<OHn-E4h4_3Ou{xh7WDX(tqIsTT18)Q1Ckj
ze>mYh3eYd|K-*Ws;+~f&@l^`|Q5LSLKH+bBMRH#J7>X5U(ry|vm2|Mi3cvdzhABE0
zxK>@<)IXskCc`CmFWN}F{7vmt8S)!nlS=>VpP5h>-4z(h7*Epe*{HNNBR8$a+d=rr
zJ|^h88f>al^itjj=#PEKaPsrMwd4V7A~udqQ^`LMP^|<Jf0X|ZbFcRJ*l;vJ-Kod#
z1L^9tHLI}S(?*SUm!T`l$oDKVM>I-x$H=Tyl!hkFe~<9EMk&7@0$huqca<UAy^ntP
zX_He9)TBma5mL|FRLR!P;$_zSM`4BGmOrb9`QHv=<<V2pVIr4;)tD_kB8Re1@=9<$
zLAAEH)Z4#2+hb!Tp#ah~;kytDU6%beHw%&IW>g$otk|{bX6~a7z`&%!)IUzv!>i2R
z2xMhi*l^*7vB>sm->XtE8Ylmq)b9IMjy4Q2&M+4&-q(n<bad<A5sF6m`S~}jZGL43
zeziaO^E{4eSkL+@ET<WyUR759-g%V|7Y2<DTUC};$!#?*_F}dF5nw)a5z=ZFHJ|7=
z3}Lt6@j_3r#h}x{S;ZxH`+f__<{{=+gPjiPL7?Uk`d60H)lO@0rQ)Y!3_sx2o*s)8
zcm^|2T}ICCPeH%@R2a_6B@!3PwAL_V4vUCiP+)kKB0JSVstwUDJ-`>x<X%lOzY^;!
zgk@lsiVUI$)%FYM$qhvcy94_sxX7&Yjydtu9Nh|NT%ltZxQH1<2BY=;8*yaZb))>6
zfAVh)oYhRaTE`aLig|50O3Y6Z1ncA2WDyUX;L|OxBTVzg!4xFT<o9^U^UmP-NO+J(
zu&vBnq40@amk_u8buF;kQ1~Np*Y_zdksZcJzO%F?(Oa!HZ^vL;>M^tnM|eLXfE#8Y
z$9M@p+n{qmOiooRk^x=bc~Dtf{NE>Gp(_N%xb|(|cwWDD5{((Eb#$=8!><;fF8ak~
z+%>aRzJFWmNO{g;y&Ax<()3&(+f)4gM#QwE`9fEQ)Du1jL>8(?ga{(mkL19i>zl+Q
ze$BVTp?3dsHHdmPxZ`vC%+!poyUC4~0?;TetPx{Jm!$+6k1IJX@i{FfVC6_5-~kIO
zi|(&tUO%MgylPgz3t=EMtqU}%vH_txkhRA?%sPPynR$`EuMuUgfw88sbva?#b&Rh0
zHK+I1cf7}ULt2{$|Gfr!_2U>@Z~{ZYQga?q*EK6z5)p9n@ej63Lx;EP2e2uM%dJrB
zE|RYHjE=Q(qEgi$YOZK^j>~@TqKU^|yZ=|~`|<V)44Hysuq*FZ=}J~%+6NnP>?HeM
zE4p{l;n0O5a@Wpg$y#7Cm?sPi9c`)jSY;*zsyC0yYZPHeC||vxhB)=`CiyWCI00L4
zoZxnkF}Kr3Z38EC0Zde3bJZH=$b4LWliTPN_vBp~a9w|Fx6G+3#BXNuQTZzeka;>^
z3<e}gN=-%Nt}Xv@SRX$%TPmoxt<=b_%*UjqXQKkCg}ZjqWDw=v%=HS=w21*UI+Hha
zX8;F^!jT9T!NQWhKSm~+QLA*Eop-gH$!ni(j?c+ifIsb5EQng{u(5QBo1+fPs}+&e
zP2^(D;H~%`KI`F4@nb0Q2d2-Wpw|R8j6W~cGp&<2_9MJX{Giw??;Xp)L=JyesV>m2
z!LHQbUj&jS1gYf}mP+j*x{x&3)>H(w7rgR(cOyM%;V_pAs^gdlP}swe)cqt8nXhod
z?^z5U$#gm2aSGnJ)g#;^1s+kOXQm=<b3Tl=tu=FOgd5`aD}IvC;DzNYw0KdA<gfoV
z_87>$=T!_YWIuAxtVkrn<u5#xvM*I3zQsED_e4eRAGz1w#1ldq6cJ72-jpLZ#pVDB
zayASfI(=wd@0<Kvlh(Vm1b!GMK<v!e#e`kExDP6rHWTd26rn@In4||_kQQ%QBrr%t
zEyl?Ww9XlMJp_(`J8Zu;YB$k#wVRH|><wwGb4^qbwRF{k#hsH1J-=IW#0ROlZzQ<@
zFeg9D3BeZBxi+@pa?BDw0YWr_)TepP<wNSQe)qQ5G(PU<xs9{Ig(<^z$tJolL6!5n
z)_7&6U10<|kb)y3gDWFBz*ji$04(ON^(NTpW#bAnv*P}hcUM3q%SRM_R$y`Q@7tm6
zl0F&+_^X%O)CLlt4x_N4XJ@hOV_$4Q!~+#De1Ui!TE6wNFhgW)aB!$E7Ka-kM$>zG
zEDfP|y#7oTQ@T_Xz0&?|=tTZeU#GTjpMe7R&O?=Uj|{jaBRqYmmya{V+>U9c>z-yU
zi)*ZV5IlDH2~P8|N29mb53x3|Qe)7r=SI_QQfH4o<iZPOK1C~Wy&r0o13M`Pi5mEY
zoHPu5z>Q}q$*iq%`RJ8D|55C-b`zwD_=f~V=j)$6hu;r2wLDPGd4}@RpMjX>6+U!J
zG_c;SQLn*=TP%<+Umr`<Jt`oCrk4^?B|OftORV>EzQrJV(AxH#s);aAQc@BE|L#sT
zDzZmE-E1tJ0}a49ecg`YIv+9+mtK#9ylI}}u;9(kH(occ59T$9k=7&f*r>bQE~BM%
z^r$o9il~SX`rPcMu=;8KR;)c=8psdn+j|=j9Y>R<p1+yOZu{GveE5og5?B0dsE^u!
zKCe@Z6ch0PMdF4p9jIE}2`jOu2Xs9AS#N|d$ODaJ(7A=Z*WQCG)ix~iINOIQU;8>s
z;p=~G86k|lVAm`VEZTxi(rn+H9E`BJT8l4;f9%zi#e@3cZwZmm<;0b0|LRWu?R^s%
z!Ik}Vt*3)6Rpw%A0FzPq($syG<+r|122}1fWx;KncfV?#K`uQ1xV&*ApV#yYlfL_z
z5XaVW<7dZSL@XDif5d6?-=Jy-k-%deu_)a~`Vl<8eEbcVVhHfU(r^g*(({%hj%zT?
zf6FpC!tgxE+cN83AKf*ou=J4<a3Db{NI~$Mu@jiVFp~F@shD3}-npW6Pco4sh^hYV
zD@pzIYDY*F1NGifPS!WcDRI}Km7Xj4wy2C(mD;)N7wIe#3Rnik4P`nE8B#WHbV_}v
zY=&OB!+lt=-61c3N>&ES#A+HmUJgee1yv>%*f?@%#zh1Uc#7*WpKJ8;h1{X)$g3`|
zY7m9(4J5OG4j^}+2h@qkc@Tn5UI+_4m&o|2aZ{Oj35eUh=+jZU;OiJ@VLhU@QL~Uh
zzwM{I?3nB=Vm4ZQBIg#2UhBbn747;Ai}ObRkbSG-udo2*lSu)ZWLzKqW>)Zt2fm#E
z@ZEWhG9r%cX7u16g1lW@Ao)&y41TlijzBkq23e3l5-y94<Avnq5A4c3ssHgx^qK!{
z6NReIOoy?Yn(1ObB|SUUTovZf(TBdKFwR`+*#ao1O!(F6U3K|euixF^uTS4*uSb3)
zTg3HN+63$~L$%uDitZihd^YKN`aQC+ik~sLs#&tKdF-n9_}UB>2WUCkxZ1+6E~?>o
zeROzJAaJTRb7V{vUry@&E#r(g)Oz0B)TGPJocw-&7GMA3ZjIN?f8>)CM7`x&NT%0r
zKw9p<OFkabc|u;=dO`^hUg%fNiuJ14I7dUkW0xCrLGMro>e`iX)a^ucp3aYjt_=U8
zO4u-0ele1dZaRpP68YNs@y0@4@}c;8;P(}984JwTZZJxEneBK+TxmJs6e{&o!=ZWJ
zN&*-)m8<mkzd-pd`(qr)kfX1|use?J3{re%Bm{sFl<M3P=y^T@)zP6U=gKi%=oU)=
zhDFgIYz*5sM2eaxZ{CzJ%Lwf`2&D<VQiR5O?5U8?uZ~>2o92VLI<yjcLiCuh(+|fJ
zpxswWH|A1X7eNw>GQfMT-(~MoHZ&I4IWK7P_9M)M5YrFN@&A?vQN6~`;uO~3qsRC4
z4ifc5rqiH0PaE}UW_oYA&Va6>03)j-Os<-}e}aA$Il2<=2(DhB2N>q(EagI2hJXko
zYJVI5)g~6JE(HnucI-?z#mApQ_MUq7(+I7@@;|))rs#kw^oYShb?0#QRu@X)9@2p{
z{`rBe>K|Wu@^)Vnum_KX+I{WXHS4}u2z-R;MG%q_gkltE1czi%8UZc>7znU|7<4Gi
zebN=zZ~%O7G#t_j59>)zhE0R6O3Y9LLuCmMZPY#bKYz?1QBbV^nKm)w6K6YRp$0-q
z(d01!7f|R}>d0Xn>C-F^KkS<wT0NNpaP%5!=|oAj;nnS2i)^08Ufg)C;>Ut;z6k9#
z1!=IM-@iH`NPP6X3BA`g5j^D(fE)&(%-C&^_Oab)Hwqh0Ingj^drTCf3VYo#xF0Um
z74N#D(_}nRYP#Gi#UBxPKqD9kr#84y)+gU20$eWo<7!uU@^tfku2SP&tL#kLoYf`&
zcic|#Uu2#9oI7|RoAdO-n^!jF_59)A0#^<1v>2py?}U<`!2<@4YP)C-{I6aeOAb=e
zQj>8N2mJQRNw?U|1MB1`46O9uPwf;BOm=)~s{gRiF3P!Hv%UV-EuJMi`rNye9r9zu
zCj@w$<Z^^`RL<p>g6C2*MxoA$NcYN3dJtqAex<x?pYfTrfI|aXI<;y~9R!3F?F`Gl
zk&}LeN|C{QXGPiE*kCiRq>nr>-7D7=3S4Z)4Q7<IYKog`n7~n<ZtJ~iPXYHNlv?&B
zCQY)q!6RL2w&o3K|8{^InjSL^z)(J>e=?x2$%;#&9%(Nmh5PiW5oKq<4|&q^nwonL
zF+{jTG;z%f=Ze+{5*1|cCGcL6D3MMjs7emZr(|XPyy*if-~VJpO0Hcd5J?u;bN)|B
zv)oO(U$on=uJ*la<?1)X)E99^O)?8(f*#naMUp>JY6UB$qumGy3-tM*>o6!3DW%ca
z16h>Z@#u}=d1FHTB}DA7NgAg>(`N0H5g-HPdM;|+bD=x&eeF^`{a>3GpD4^nZb$Nw
z4N5gRc7>*8aWQBW>Y~q`FzO;(<pROvcl{cJ{YZ8J=tq=Rt;i;I$BqPCyTDiM3edfB
zn^x<a!3hc}cEx;h`GEji<ZaC>O7qS<%S|KR>a*4hEVmB0EU}6HjXTb-PG^ec&WDQN
zyQw5D0HarX^d&c1MomuJk`1vY;2sTV^SgHmwFFlu&-dr$#ABvE%b2Q5j+8R8y36;>
zPj;*ad9(kN>_p_B_2iijqI+v8;mpx?9mb*H@clI#q%|wg&*0ICfhKif)dW+rXc?n-
zIw`5bj>zg?!G}Uu8YGp$vMF{bDI|h01JkpY_^nSUF?C+l%vw0=73xX&<SSZ(TNE)n
zPrf*G>YO*mZiF`e?w0GGG9v}8unb=uTJ_zXV<4V};XjOTKGx1&0RUF@JKph|`4*Qm
zxreJt^0SXNFMWajDbKhe4k|3T(nN-yjMNAf)O2i*5XSSp4lFKe=VLpPB$=6oZOMS&
zRi4kMQaB~^s+_MNwnxA4D<EG?DRQUcIGWdn9$---4$MD6YH>9ml4V%$ay_%PQylhT
zrNym`f7JZ^TodBbT0AInX=HjPDvv|+HWj+_UddSZnKXRsKmn)G-xF5)=V_m|0Iy|J
zPil}6Lj+%5fg^m>T^RKWGyrG2fAD)(ON}4X`|l`fkAC<^T5EjtgSP`AX!G}(Nvlt3
zlrI|GdByFY)+#NeE1tYL1ya}Qj+$5Tf~hQ(a4IckIcu;9Y3=ungXV<GX}-e1ZFb{w
zTuk{(;3}vWS8%6C3AvPiWbqPs0kD&uN-b_nVA`p%VJ_lc7vV>o!iSrPVz_M)meIeb
zv7%O7<9S7%7CMCKiS_O1X_oD9ObX%P9Lf2-EL>IgbPz+T;=%CrbBa?c0%8?at7Iy-
zY=+Wq%f;5gWxBBOc{;TP(>m*;{`U?7x^#l7)6@XR9Tdo7nJNe1mnnN%zdtMYJ2G68
z`tgK}s{v7)25fMfit8qc-`jAp=XEmuhHBeK#%Cxg|5)~Wf@xGY0!^pGEe}-{I=Y>z
zQA3YqDX3(Vfr_5;eIjZbGoiA(eDgvJwioT7!K%b{J?aWoV(Y_};5BU4!-{EFsG%_F
ziLX$%X)mjn_j+{+Jm~O3Ku9ZdJU9`;VFDSJMJ>I@udyN-VT`Ev2!}qw(yQyDi!xMV
zn-E16G;9~~zjwKuZHOqlOZl@bbY{HEdm@h_i1(K%d?d0g9wz(Iv=rl}TE#$o1*=GK
zV&6AFKp(xoTTt|B?4YLg!m~Za=s;hR)l`1AMdKR1=@65Myo#$~xV7yA3v(@+s*58L
z%$s|Uoj4?%lrp7KZ3L-PG|IYiEBi6&(5G9(Re$MJz4O<EkZMF+Ko*w6SC=DZq!cHd
zQ<O4T2AI~j+o0!I&a09EbD3c*^tjhTRmIr+sJ~?x4~ry6Ol2unu0-Q6{?WKc3l?Q(
z)xWD^l>Vwj5dyU?UlRN`_|EV1Z5;o6%$t{{nt$i~>%bimL~`9@++k-a@_x>@)!PFh
zK>zbkk*C!HLQ9OhbJljHkDYmEM<_Uv+Ur6iwz*waAu^0z)T+PsK9v#iaHE{w&&n2e
zD99qX8b+h9RD{@dgU?rXfispq`H1pHUjA#OHR#e{x#dFeANtn2n;M7DR;B-^@Yvk>
zw<M=q4AVuE1=-D$#wC_F)`4L)D7CSgvm=~GG*BbNp#N7|@FIF0|Ixt5wT$f;AqAz|
z{>X9%`7vBnZz~1**>hGHFw>9XFl&wHwQ4`d>D&z+zI#PKJkp)ea^>0rd5tcW3tLl$
zTxR3vzWP$y#_Pet?hEnk7$h~Gmx^)<(;_<vDZMwF$FRbU^qSh3p5gl+t`rQ1bet68
zPePU2Vcs@uzoJ&e)Z%uVa0P|d|BPr?$w6l_z@LxLZVk-hXQg8fyC4Z(#eF1C`leff
zF!p*%1N#+xb~k>+y}=*zEIQbvU}79Z3&#Zhv>t7SwNkDY(Vu&jtMN3QDTq=(#8P{C
zy_qXS%E5!EgoM%)KMwL(xuic8B8BYXvx>*$pZ%76TRV`sz2+$o3Mf0Mr3k1Ct<m@O
zUhlB7N<YG`3~=7DhOHs8-R7=9OrEYs@CRXxz3-}k;6596Mrc-MVa7ch`Sx7MSDh+l
zFsG25!8{E%psjas-a9olsj7*>=sEChAK8z&6z!@t_4QU8Z{0D!C=VL<Xom4ICGD_-
za8ut(jhK2Oezv0xDc{c(s0_MCU6I(iJ6o&-Cv`SajmI8~GK4-at64&jl<|5xJ>%#4
zqWDkIxzL96&Q$dPs(p^1%N_m+1tGZ#X_q`ZzJTPErYL%zNl1u3Wd$e?5s`i_`M2Tw
zmeVVn)B`g|uQ@eap}+1l7U{OXqD>TsbEevdr&gJc2~in%Q{75kW-ZE9`lW|kUaL+A
zRBOGh_gYE9*z3G~<!AQz|J8Ax(Qr6X8&`uBMDHac#Ol$@#_B@!)uToiB^HZBThVI>
z5+$tOgJ311too1MiC&_Y)oXUWoA>kkex7^g%$#%YoSA3t^E0ldsjxr+Tuj@u?hMt)
zi5w&9!Sc_$RoFBApX$fym#L(6dNEfpiO{1A-uvOUqnF&Me^w4gv;J68B=~&1U!9KJ
ztnhKkuT-lrt?iI)*`0!a1zOrUWsbT9p|P`y?CEYYLl-d6A)|D`9%TL3u@T_JP0<Sv
z?}5}m7C>r2NsEY&<MYH#HAlmfE3{|!@x&ZRa$LV1uI3h)?4g^WZCgy?1kM(1AcGbA
zf==}MstrmUk81&PLKm5xxx6d+dp!s`m7Cae4DP!yi?o{sG2|9aO?<h5fad*3<E5(t
z2W-T%@=473wG2vikQ}d+C^Z|pF?g=<traEp(L*{3!F$hNi(A1n%4>Cd*4xZNfEfs!
z<-v_N|5eAU@Da=vHqdfP&^hxp&PSuX!fHL#UROoG^q?~M;|qKLP?nn_U5#R(pmDWV
zB6k40SFa0(;y&fndP%Sg!v5@O+``vCI;xHW%3Pof#oFDbKs|oLW`amJEmYiv4W@sy
z6ln1~*ykHlR0P$;8)?XD8j!c5`vfAyi+F)6xlcok>;D?};39P;c{J3cPjMHwi~j!N
z@7j%%_lEe*@EQ49ry)&8@r3w7_5}a|KXs`HW|v~Ou}G~a6Iw6n{fK4`84-RK34n7=
zT2H;IUAsGQP-0CYgn(JBb^|l-b2>@q+>za;zoAvrMg-%TyyKqud_ph$J!i>u>2Pb}
z&GnA&*>jNW7eIcaN6DVNo<(6#nSP`raG|(+dYmt<bDh^}>YujvH2kBBf7zm&d6DD>
zP}FrCP7famM2HHjIV--LL5*U4^He25>)8BwlZuB`wT<7OPbkw=oDf-*tcl&l1Pu1U
z5c;mM<qSc|4j<FN_H>nd-X~(ju#MHh@QYL@`YLcRDxOumK*@Armh9MSyQlEZA>rBM
zs!4Oql%mo`R#+z)pV$s0iq-%mG*Od*6YpXK379Tt?Kg*G^OoK!eK4`CaU~YWx49VX
zB^#@|IS{W8NUb%DWKFktRUFq`;<#-&Cwa`no2T_=`N>>@UZo;uswF=GFZ=9ys)RFU
z5vpiZk!1U^GLTq+>o`~-vHH2I`_Tm47JOPlS;xVddG&+iIG*LRJGa&ole%D>Q5;O9
zKr5=bz+4)k=6?7EIR+K>$RQ&q7SJ~}sY#4o7h^qI|Bg}@aXLs|+i%s+@08S1IOOa?
z)5R*zf|Sl(bHmIuQ<w%quUE$dD;t&=9)`|7F&q07^(dS6+(3{sTT<nnDK}ZD;jznd
z@r#}<A4r+KAnv^6BEZtxHYb}{(6u;D9YAeP+v)tQm`wmC`5xthF9;78;yC-_(cwI>
z<ZCsVLcI)EF366>KaoIR^%rk5x=a4_%TAz4i$<p|Dz3&w1S2tESL@N0{HXPSg?dtT
zpW^XCfy>~erQ643QX*aO6GVvEK2jia1N-=Be9~Ci3}_bKTrJn0H-VBOjE*gRgtpKH
z`ip=BO~9)yOe)2{|4uv~gK_5z;;&XFwPoBj`j{^j_D|Dgjdz)#0698+O?=w-_M&bm
zKW(r``?U}NC05E!xF75gy?j+8_P5O#xdCFW5$v=;#p$-3qz`3IBD+XT_$FMgGQ3;Q
z%d9@Uz>DBN?q>2|kJg6XF%PqKTBG8ITVD5+oXQW5Z+g#x#d5r~+$;lVN3~ZHlHu;t
zf&;g(0Hnm=LPOAA)qu@{!qmP)9@2uV0{PB23C9iDAZ7AhiJ6cUv4+kN?|rOP13lJw
zPdp!aTfcRxA?thOVT7C|w%Xdx#l<?3#mo}OZ4O$6UJ_?nV|U4mV+?Bv1`8Mg;yS5N
z=U~1*Apk!${=e4SWA*7-&!)WAbJR6Sgh8IgZcKA*o*!%zBAHcO=Q`#M7Rm+tJP7yZ
zO@YgIG-VeXd?<OTmZzQrr%zl83e`1IY4jWq3+TWT`)6v=j{{rWe`m4CEFZWRA<M+^
z?8LB}f;wlZ=kC}EmFEqcQk2bASToNVj%uR~2a^`{c34(c`Y(_UWNPehqKhC$ls}FW
zJKgmfW|t-A%}0?$RDbT3xW|UEkZEqO$0=E2Pce?)^~wy|Z2MIK7aDVnpccEQGdoBG
ztEs)HeAq5G!Kx~k1(nnNIKQ1J(U=XOMtchUK>c{XPul4$e!`<?h`7&R{>4A7<p_?a
z>qI}(babltb{noGQm>(!5tH-hWb#|yol{8SvqrnM4w(RRA_UG)6%z&{dTKwrTd>=E
zjC*QtsJZ=ta3}`1`1JHvW(nSe`pz}4mPVX~<D4x<q}y-aW)h|OqN9;p&qDjjZq~TL
zfTm2=)syxOs;&oOdDFglhkDxoTrh`3YYC$(c(*?jVM14Jw{LF#isQ3zUwu|f)~h`H
zPoTVVoPs%}qf2g0FF9%gC!dkv`=WsRwL4L1J_qb{=C!S-*E{=;BAYhzb`hpx+>FHW
z*Rg=$7zqMi(Civd+v{_y(@v*l!N<2LbX8Qok(sD4RZO3;$GZ}$OaojhGZ3W~<d>lc
z$o&`cOlcS>0LqVWfC2pL8&m0Jn<@NcOBXvh=h-$*HT45>5<{Kgt)j^ik|oRfZiM8u
z3cm=ync|Mk*?7qZUS5wXk7z^)gHk3n{uHgI1d5Ct_OfZG=Gvy*+Ke8cnf5ZozrSxl
z_(^IitNKN}b8;+KX9$A|F4~x8RNQ+_bvo^FBZZ3Q_nT!k2h`2TVX1(#V!C3LjYL(&
zK^VUxdl8i?xRA)Oa8=Vv{^d^$>L-T91Mb$%ODyopR4}l7WmhWuG+l}{KE2Ki7)MGJ
z8TP#IwV3)9T5@WhGw~0pnNKG!lut*{eb^}SIU!B1SElr-nyxT(6~acXAA?KT+p2QH
znuQavXQZ;RszY6oE&c9=njT|Lsq|5(q&b%*0$66o{XjHa^$~>7S^8_MJS?~)(f}Vs
zI?7{8#dE@D2D3RV<(xWX`ouAWj*$enC=?s=tAaTQm`p}fS%yNEElOFz{l*+6hguKG
zr`hR0CtDp|_0BdpEar_B?{UNvo;_vM>!HT~iffpjrlYkqcK^0ehxww`Xl<j(jKL+)
zw8(3$mBiNXPJGNOw!12qa9Qe>=%~=`Z6GsgKCI*N+$m(S^m3$-LeX($M~Aw|wiUqx
z{Sr9hkwsmaY2DSaMC&M=759h$MC#DpflwD%u<+Jfy5Qj32`j*na)5Ej%g)m;zZK+V
zErNdhNYq1v_;Mc;w9$g&8;d4}(c6x<3{EWaUPQ1?I=%=cww&k$|Gt{xwd#$w8Fka<
zN4WK|r*m<(9R2<aT`ySqP3zI1|M0)<uKFK5w2--jztVe$xtBPC!EWE5A!jyg)G46>
z83dt$@E&1F`pRQloAxFo?OY#4^`)kCeTkZjc#pj_|BuCAq|lbePL2o!*>#l<(M*3&
zkKbSxFIT;fpt#(uIV`pt6qQ30E41IY5t-*%%v<f~rM0@&DTJy+M?r}I=}air@^%%@
z*D^fTg2rv~X86UifX|mDA9c5c)4Cs&GRr%vZ{al5M_}**8*|<MX^{9|r-9b*d2j!_
zn6T?A3xpA@hi=>#{ve~#(;pR=M#*RUTE@9>(Q}TxXV3C}Ary^*>OqG$e^Yr0UzOwC
ztkB}$9N)YbX#PAB?4~lEZ-&X+9rUKXcOylN^smqvB2pF2=zcOgJ9`*lyBBdv74%3T
z37O<8cP-qYn>kOlq3Fq!ek&*HPirl(*5b4xKxspfBg;5X<$28}w1_;q4U}Tl?alZj
zz3!~_j3z(w-ZKJnr;CG25t-wQ@t=+ixxzy79N14T_KB#fyj?@ia3(@`?byinr-Tkw
zZ6L1?$bRF7?$wo<70X!6=zhDu*XIL$Ea|()`Nr81eMU@~hxA(>>Jy7=En^+R50?3_
zb;IbY(1Y-ho@sgL+)}F~dbP1}e!@5PG_}FyiVaC$s!5`Rjva;k$Fz=Fn}w;j`hNdz
zt%W^m`bk25*k5^R&T?<!C`)fF%x^1d=)k{=F<>bnxsj}<{_-6+b`-Xn0&h-A<s@=3
z{=hk^?EmXzV^}#XxfYwe`PatIshy|Ku#!DPGV7s$!&8HvIO-V5dXvJXYkz3zp$J-N
zh@YZ?8<MjP8w%=X6HOYvCErRR9rS;O^^k`4Vn?fMKuirTFOlf8F%le`CpfIQq{{M6
zycDo;{ViYvWF*sKQ*}x;#jZd?U!$bslPcIoJ*vI%fu`%3*Z%uED;s2R+D|(|b0MIi
z*mRIqrv^fa>savq%~QNS2iP8a6L3CRRyE|FgoGbJl(Q`bevY4E@ojTC`x~KSwun7P
zcQ2E0@|0|*<fm)Fmb9Hop!lzIaeOhmbdEk4GHdnwX43ZMa=VxR-iWg!rh%5nIXVAG
z1{)+f-5HugB)>hn6G=XURvSt{LD;whwBD8ck;h?n+Qf|x=(9t_ZB!3#q62cz|7?x9
zj%}*Fk=x2`%oUyA<r*;2Vdv{LHhM9<MoIE^%6_c%0&NN1rOQn2a&}3qdC*pqp%NLb
z`*p;}7D^pN`LSS1?nO=)B0_ug2RmT!`u9%YP793aKb--2-K=3r%toV_c}DX<K>-_5
zB75~kJbQt}B3B}SGx8yUk?4w6W#rvA)b7<KBbA^Fsi#qfcGQ@B6T-O><u7{8hS`p?
zDJ1yhB@aoNe;jQGSsHL@Y%=?f5xF;t?pb$#9ewJxC#Q#Q`&a0!B12o)jsEl>w2rPu
z^|U$KqjKgiXNEQ8Wzr;;Mbo9Y_~T~s-KTCUqyj1>d0ppW$f0ePrTeknCDhPpdB4=7
z`&ZUm4$|3FTjZRxMyVl#3*W|!_b3#e{64$juk~`D=V#uIpTSjsP$E!4d^p7St&eh!
zhn-===?`A<oGGWC#_f-Pc^^2xFEJy$t7%NbkW23L;T~(ow^pB&8RH%j@XfS_XH*>l
z1IVNC{Gakc3NJezu=#=rWo+QaU+dr8vU>+qo~SSa8jUD=QsKrbB4wKQR<Xs>zehS;
zE*bN`IZ#GH*c@SaIvL-09*HGox%f=0?-Yw<&*i#T<uBKtV>=eh7a+wQ2%Fveg{+zO
zDCm(Sy-kGompv|TS4Lo@<aAfI>AIK}J`Ij2#?|57{%-iB^NB`*N054sC!#wSIw-|7
z<>u(R&B|<WNtvX567Z-pUP^{){@_ThRn0<#x%g`jcZzpANm($KHl~a-_U_Ym{~-~>
zvzFmWPw@nC=AvKaWA7;M0Y1;!MLevY<}}6g8>vE#j0G)GJAX{t4MV#%)q5P6T(q_&
z9{l^Tu0n05{P&Ot43K(%)K%?<2(V=D#ctn93;ycnRb{?*NDz>l{AJDnTp45_*VB(!
ztyFQwJ^{JMjoaOa^u~Kgxl&<e_q%?z(PWRkJevZ=TXa@?m4OfsaT7``JgPTY&2}HR
zG(3>{0(`DO-z+XPMugD`Vk&zjNWPw2A{N_&C!A6%_t50)x*R0vUILz@vNXxBH3#0G
zQJ;MW6V|(!-g8z3v6~%GZi?=DZ&~-U>+<b&*YC0E`WH<68nBczi1l$6A_>l2>ev<X
zH^MMd<t#FPU<W7i{$D)4|DV+Nzaym7;dfNZPC%~4;k{8Dml5x|nyzY%vMv0709UFk
Au>b%7

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/non-identified-plus-prior.png b/docs/2_39/stan-users-guide/img/non-identified-plus-prior.png
new file mode 100644
index 0000000000000000000000000000000000000000..e9d242084406bc5024c8eb73fbaa8e93ad5acde7
GIT binary patch
literal 96043
zcmeFZWmr{P8#Rnbx3mHhN=TzfcXxM(NO#8uX#oKh0qGLy?(Xhxkl56w1U9|lTb$=S
z&%yKKyRP@w`{(`Pb!~U7wdOtVIp%$jF(zS33X+&;&(PrD;4r17#8lwm9tQ#cC{U1q
zcf30aJK*3DURsKZDoKlqQYbmuzqhn8gM;IG60E*rzo3RM5N5CPgHVt{OENUPt3+K@
zw~pjlQ8$uwO!mGEVfM}}{mTs=&agZZ<b|E;d_`}6PNbVhLUWI?rT8c@kxabj&RQ#_
zRYU3XrKP;LJFzKg_7T-r8{AB2RXg>dR|DKD+*`a`1`A^RhL%-{Pt`xh6%?n8yo`;>
zA^esqA3k0&Df@V`U`K)J%fJvWa~r|=Fl3c19A#kt%fYR}*^3shh{drmD#cbi0y5Ck
z3k>DKN$YA^{1)Wv3(Fd$#S_w_2@VAK0{D1PGOov14=uj}r66k5C_RLr2&EB;uW|N0
zOO3JNW8&VJRdqX(daCzdDvG`uJB_TeL_5~7-C;8)xt{Ln2Wnj>XV;n5Vuvj=oT)X3
z3C>iVk{-#H8kW>a8~Y(%L|={jz<ZD@pXPl9xh!u<8X6YiFC#4@dXG8epvzxdxw3ha
z5|@oZsUhEcUvSO1n&VddAo_5qaZx#Iy8R`$SJm>D@vk-+P=118qNB!U(9%GKPQoDk
zi|BVPIYS#CkUt1k7B%J~4vq70zR4s!mesVjE6x+1FHY1}Mu)vl;5CTueP!Da+y0Gw
z0IS|D<LLmQO!wFhjbG7wFY_n!vxp6;MDw51md5v7kcd5J1si!-Hjzjd^e0tJs<T=p
zx5dO>C=fE4Oy$Y&{!mJN{jIY}f`gCxb#G8`$AB>#mEt(A&5|c~^m<Rwr*fhurcC8*
z-kuVNmcBAALDjiOH{>Q2d#ohpct1)xXOs3&WZQV!un^IlY<URiCqQ)2lSV~XH9{T*
zrQ;hR5Zy=R;km<SS3}XaA&I5345u#?N7>kQ_wgu@@hn_9#|3h%tp)e6`1+9=E{6A;
zt*u>d98V&bB5Zp&O6|I<+b?EI%9w6!Rp5Dzyu#x$`j6b&Q9WLe@#{UBpvHvQ|I!|v
zSD7!%%HF;7AALd7CfF-5I9|6pntz*ln`f76m`gOd!fO@ul=tUL5Iu+iMhBt=vB2cQ
zU9{_|U@!G!o4+t4ayJ|{rBTm%mlzu3Ersh_Oj)PBrBzkv_ue1^f4gbCDd!_8#ji7b
z9+R`tbhL+hHiQ8qYa(_*tFV%<@@D#tK+KL!jg*2k)2-GORNz9Wy^t33Q-!0qd-n82
zj+<UGI4<0SCpDZ_D(o;qv9o9G4SY>%<AuQmP3h0a(AyY1Rvw;Twv2Mry>Hf1cpNpv
zPtjbGMu+r$kyfxWr3>7qIr<4S=SuQRUZMK**S^-Dk_;snVkZ~@({Z@hHnm4&y=)Go
zirN-lzHSzGTf(qC&=m{4Yv+Y%Sm@^Bf%b8rEzjmdjud=0-(*H0L+hK?wz8x)kiMw$
zl&)Np?Z~bzxzvn#d^lGhp%vpL=X`co%=wJhkm`<b%<M!^8_v<ghJ*nR4o(P8T1;5Y
z<I!FUYVtdE*E@Un5rK&gj5ojFQLH|#h0xJw`YGX#)r#u{cGlw5L_a(4>||nQV`C=A
zEN(+wR0`0=CwPk#|1!K`jE<guErjvShsPKbOHjcU?`4p#rlzK=ySir6l|gRt`9Z^B
z^L6u~x;$)|`#d=;E>_4N4uJyh5sDBTJkE#vm;PlhSh?NI2T?`;`_A7V3cdV5u_5u-
zcmDZ0>rVunAXBeZZrgvZ@Ykm(A_xC`>p#~wl`$Inp30fCL<=u?(T3el)N;|4>Xnw9
zi~*S`i7tViEwGUK9JQ@(a<+KR);gLh!bUy#P#mHodt43rwVSv24^e0IoaLJ#qU2Yj
zrOsQ!KA=US<=eBG&>R71-jfd1)Z=eQS2HcXbYD)dc~Tv_pRU$JiU&qutJr=XvwnB5
z6X=v4@zi0BuVIC-K@6m%MPTv3tp;82?sBAKb112!rWsn~u$>W#<%B|t4j)LHPM3v`
zQi$(g)3k3JuG6p=4??YZem&b$J<NO9IK8;ID1#a8`%LJAw~ZOg+TNZGWbn<1-kgGF
zUJc38wnXe<YKb2VQo#Oom3Obi7g`X<?-1d0dfTu9v%5e_h_uL?DO>RN?1>_az1{%c
zJQyv#f4FReii}5wBp;jcChQZg7q<keiV8*2JrNw9k~Jc2ebmR9tuR;Qf}Kuu@E>}P
z93lH1A+J_WC_qeD5{1&e;TKIQefF&l(cPOM`W<i5E(Z_V#++WF_ELF(V>P?WT7Dt2
zpdgbvv`VG+hy-pL6n|a4F6z&ksS>=d5JZ<G2e~ZnFFKE4meZ?JjJ|_MX{Rw<3Q1m2
z^t+lk>ZBE5Ydi{OvFUy4_^d<1Xz_rpPF>)#Pccf+V=$b<CO+_YA*(;O&#%X;D|fIJ
z(-O|69r+GJ*py+>(G!gl-UD#gT=c^E!kSpU_3!$bCT9DKLUVowJ*PhEOCEF_i)W8E
zaX9wFu~skn&g(`n6KQLe*gCvFkK8Udd3S*04WIj7bdZMtSH4S`xgoz@C2vnTaLI94
z3!@!gkm&Nj`}G{Me|5KKE{MLFIK7}y`CJ-4Uf)ey9Ip8W%o2T4C3*aA9>a_Dz^r*?
zUhm?O+S0zAdqfMT!10&pApHinuFD$pG|(Ab)F34Vs|PUjwm0MrFPe{IOE9cwiuZRK
zM+EO~jy{?5z3>4u80PJrSLX0DB|EgEI%c|?J!UYK2lAkh#n1Qn&}*iFzD;DqPT)X1
zE>f1IiQS|Y%w+#mfoJDOhGe^Ip+Tndg|GGQTkqcyhNJ>lW6?<bwYQ_80{AbnJn8hC
z1huV^9w8XNDK=dnUSoEzlf3}OL~;N+^wtzxy-%qlAfdVVfZy3IayyFzQn&JRXe|eF
zB_jF}cN$A<5)<h;2z#ttFULopi?etsOWH>^uF|tOkj48wLW04jC#~H=c^4^fcG7Mw
z@q_E>aX+H8gICf{S$z-)&(=wr%U)-)y(l#sh&!6DX?9)K);%rpbX5~FdWBHi6DJZ0
zk_c|KHoTs<m>cz|v0Hlf{H!!IvVZjH78=g4Yg<GL9U{NWc)tYq)22fZi_g_W+VG6_
z2O(7y3X~)#Y7(#KlBAAeTnDx5^ycT0_C8tV!)OR`qCo7pPiNh=P_=Zz(ia`___l?y
zttJ=STnrlBX<E;9%>|ZhVspMaz05gIoN1(CI>z$v#%VRHWQ#Aaqjg$XSmy9qMK|nK
zR5~%n4|=~;@9L?qqXe%#pi8U?MNHkq*<Wtr)gd8B^chNgyL!}tjeXv72kS^V^x3Mk
zUT{aOzihd?Zn;otH)?wS{(ZYY@^fN4-;*Tdy1w4*cP1v8`?ve_Ehj05odT7%6F=o0
z8S|FxfzK(>bE96IYw=IbINCWz=!$U?6tU!3JP$&sN_C*$u?tU(8{Sg9-B4`*BKdr7
zf>I;VQxVxPqq}L*bRTOj`uc}2=}?Y(X*N1bBkRoutsv@*fwyg`kJWDjHIGy8C5VAz
zQ?sFtruWhS{ce_?i{cD<<>K!scfHf*=hfHJKBr5c!v(gkKDp{%Ig)8YPJVZD0&|z9
z(l<jyU(e#=_apP=BCq>_TmgEsAqYCPDk*rS+^oM6SZ`ndD%>Yeh%9frfd{s2u*Y+@
zZlJI}YTkCd5;6j8WvgouEK#P7ZJ2gASUlV5-;s8r?=GSx+xIzAqL&+BzPiIr^|Nmo
zqH~sxS^dsVbw#n&#vjK4U%5kB0pNJO1-3sN%!Bgj<)drl6nKAv7o*I1#V{dVjXM_4
zKG|c0vNOtvfiuw`wrK$Hk1Ql=McD3<=f{B;`ys02UV^t5>$>RB?9NJLavz5USU+fS
z?>(pWoFG0EwfMo$Z!uI06AQHw;7?ZCR*}<_lam2Gs;_riU<)m_2^0sLtg#%=lEAJu
zOB*zt+NF0c3Vs)%ezWCLR<SVNqQaLa3O&*b8rRQd*QE|&^@@`%Fs?}H^?k{+%$Xso
z%W4}tS=n>%hRltHooDVRzDuq%RS=qKObDzpd1Dq<9~F4(vkkv<{o{t;6#J`AO7T|K
zb7*dVld<J1!p1W&=G`&Noywfzd2HB#Qw(CfJe<Z^({^UWadm`#4|sXNjN_TJjGX-m
zlP+iT??{IAd_Rp!7}=qRq`-8nf!5Gr4S2C2Lcc;GSz@)~yet}2=wz7ebKIAr5N3-*
z%nhGtDCFN7{$#sH@O>Ob^|825FJ&kr*YcI11rTw2-S@EkJ;Ph%TVT~f8=LoDSwik$
zEq67)1i-L?fu85yHku|yH>P3%qwVcAd#H9ZJuz@^(Z?DcS=1;Wd7d|{G(m1BxaY*K
zE;^htd{tpGtcoDlljiwzt!Jh!hj?ULzoY$pMrP#OkM`~Tyq~?lZC)fF%gZuuPK!Iv
zs$7|I98JU4&zVr*Epl49xlFtBtoN~9VNx{Ch85Yl7F6%=#@=rY4A(mi9gH#1_b1a#
zRZ3=p{6}>0UB&B8hu~xP(r%mS-_tLi|Nhwjef<2F$YHeSpsCJgq{{QNaK@6&^984a
zsjgyeUaqXuxrxWIJ3Bi!&)F2S;(QI2C`R$&2PrG<z55;TJ~PPF^}@0jG7$z7Wh<1*
z{3aQanoqM?y522$mBjBa)F9WPzv3M0jwFF3b=(&r4l@nQZ)XK@PpX;JC^?^svrdRX
z)iyYwACUPd_<-U7+i6wKPR+7Gujs%r+B@jb@9sL!s?EJF#j5cRYIygyByusGom0fF
z<odAXu5tfNjJ=+6UNGpS+BJVW)-$IWsj6tk+Xm;nPuB2i{Ak#{xgtR5JVoz#<KU)O
zQs6bI^B~*Fq8Ru!EQpqm(q|1XEqAYg0Q`8WASJ?iK5lty--bZGAq~lq3_^<h`|@Hk
zM?jt1hSpvAZQ~*6P*)oBhh8Z#3Kn|Mq~_WF0+Bug1e$htvjtH<nTDKi29n?W^0Xx^
zkUHN{IRrc1MT7QCJG9MH)Qi!snp=QPkf_>rK_gF*!93Yf;<N4n-8X}lm2qz9i2+&F
zNNO0^2wMm@k!ye5Iqh1nc^Do+B2`HB8`BmWDT2n;&L;O(a*W`Y3T$0H-{q6knqPf!
zX%6#(+p(a#?daRwY|ws>Q)TZQcKGyt1=PrrR5O!*m}R=#2PD(YdSeRVUXRkirc2ek
zRoQi!!?>-2@#<z3Kc1Ac3UZDj&Bir?6Fo(6>iLWxZ001&c~V)D=_;t+PJgOQ4q=#=
zJW_dU_8_c<y1Y0|#0<ne$Ux>=S3Z9}OK}i$56`iXBJcIXyW2zEcthVyKAAJ63@A`s
z)*YYf%xF&^1}}c!e<YUe=gq}hvf@V2;VTF=$tLb}#qSXI6L1;Yf28OOvb|N6bRTm`
zfi`_^jySz7xw|T{R9$fQhSRW2JDVyasy|ebJa63MXSk+`AF6DMhtz)a_RKgms@oF;
zzfw%tI<s;{)%E*L<fr_!V%Ys;bEwF=xQ+(XTYOpb<l2_|+=^w!MorM$Rkogwiw?&E
z?U9}YuuFz+@M#>Azo|3b4z;?<$Z&_T-rf%baisbNTW4PJt~bs5pzh@COM^_E#jx?b
z<lz-1^e1Oi-r~FEazsuGZK>UU-}#wT(qZc)`gs`fT&c0>K&ny>jqM6hX1dqBG#N16
zV~@{X!sZyd+)>+rhnB}T6*`fk_m7QZX9o8>eP@|8&$*gv?m==)YrL8&Db_O|AfG{%
zUi=wK3%y?VyWR4eKdAXeK9=Pm2tCen!I@D$LWW)D3d*rEXfAlC8+h*K<!Bsm%PK3|
zt7!WHJF}Lh;jO_qF(EzSuL5fKhDihFBk1^mt6qQ8;sNh0qdyb+OI30iv!%kb%i4*H
zp=Y|56^pS`H}_l3{J>k1sS$&vsmsz1EZqa}dp{Hm%tr{Tqk7@*G2T(%vNf!QGw8(K
zRAu|6PYBlU=&u#1TApJk!j~~>+*f`hym0|@F1n2Ere=4#%g1p{+II7?Zh{i0>-kA4
z*th9$$or|?1;d%a8r_ovFE3~XFZ*Iau8YC_DhVu;D<;qa&uGkD0k-&6G(Tr-0}R<k
zW(JR|!fk!REw}6$AOuc=uGd*EN;ArP{|SNCpef}NIiYK<(_yc!h;Sq!{pRzunwC{e
z`|m7}=8DoA^!@34UF}rZp81IN!oKTdRMYane9n@4DSRN4#yiYH<Y$nLvT`lI{p!Cl
z&#q*S4%lZsM>`gd-3#ScE!>~x1-AA+q1X=smA`^f8%;+VY1AWPXL2hhZk-Pp-o9@3
zJ*!Al7fH_{V^Xp3m?|)c)oAe2yl`=@(yc72lT@Fd#SSqNRHJuI(YOW5z*ocMa<<I1
zzXFbm-J2NrcBB<$)X|$6V#R8DUau1GA*-kAh&tB~1NAT0VVMtH1r1WghfQ@E24lhx
zKA<v0ybfW|EOz?rDeTbCkF{nhJy|0n)-Hi%T9IV|+4VkpV|-&yZ1MF*%<16#>VbNR
z#I1Za>R0;G3nBa(9Q(e{CS7ep>sHdRhs>Dc?Y8u+^{#vvn4-c}MCjzBRT3v7szES_
z;YdN+?@#6kAf(uGVeu1`w3FnmjfDYp1~$La#T6Mw>*938W$j9aq17qA@|s34lc#Pr
z+OGT4$#Y#8-<4xY)Uml@i|M!ipEvdeDke*(KRJuCOv&J1NWuwoImO$YA}tntv~~|U
zQ)AZJ=&Wv2oUc-Jb+4aDncrlPTbO7#MvIMruRr~Ia(G{o58_>XiS~JU`12-O!eH+G
zYQl)mdkmw|_Y;luFw-9t?BMlnWDS_4F1LLa5BF*iUcErb`~u#c(HD)tx_PU&gfO^0
z2^_ZpK4=5(VPJg^l55&_s8;_|I)N+UT)l553whf-kgD-)%kzo!V$%l5=;27H(^jl3
zCpNB%H>l>Na4rw%7P7*WAOGu3K^dW`a#S@oXj`;S>Kk1wm)h`*(^yA~Q!5GohN{tz
zDQ2|%30pSKHkk=0=C1w_0=@f?fO=uInv$>z_oRizpHwK2v>G1KX-ZalgpaRz=ryfA
z^;mlT4xswFBPm%zP5tw*G7V7`l~C{JxR^$qfh>qzd>8Wb^w?AkHYG0Hkn78n<D~p!
z%s`gE&82B1_zDtoDRNz`0pbLp_z(}q%0fh)IrG+Mas4^bU+oSx0{Oc<!)xW{G$c>5
z1^p~vKLJK2DRAl1%kPT3lP#vP$xZsNW|9hwDh=X;m+tpl3;n`FYya8+M2Y?aGh4_8
z|5whvllJ@iy*N5O^ey-`im;ZzvPxXTrLXnOq0CBw{Kei5r9|DNpgq`@QPk)giW>P!
z<aqB>>nzl5$q-+cRHyz|Y|sG2813=P{`_*Svbr}XiIZvYBi5roGY>eZYR2qjN_xsS
zj$`e>R=O-u@*|zYw_;vRD_&_()1Oeg?>W^4m&*`-j^68yv1nhH&m>Nl-_cI|u6A;I
zX6CD`v&_UTrBCaDx@<-B8`N5`|6Db}GGH2Fu5`@=Jy|{+k%S>z@!QJtY`GnJp}33+
zBZIMfn*ppvLWh6D0&O=Jyq@zpUyl|PVAd_3FcK~}_R<d6#l{L5(^q?J7w<+Z_+>s|
zsG?x<CFh(B7^m>`3N~%m%K$3jSgp`JC(ZX;o_G9#1)n7ZWH*|oWoOSt2xnu)x~frF
z6O(Rrr;qnldDRETwt(!yUTyc2po*JL9powO++Zg~*Jo-l<jmtI2A<&9lVg;)>(OZR
zI`cD?-GI_mbLF1GnWkH4$sriWc41mvctJvA4)Y}VOy3GdaZo0_OiMWTcJm3}*{o?G
zqIP1a_?nVkvR7Y|w%}T9(kT2#g#9d1TL$$+H$lu}p6Ui9*p1J&j=moZ8rw1L*Q&uW
zt#PW)$Y7lKg0@BpwINjCG{aL?yWs5zZ|ufXWB)1>U)hKsIGlaX!p^A!*^DSmW}?Z5
zI?K&E1O41wgJw7@_WQ$r*@sg!K4-hwTDy1C)$3AZ2onBq0S@@Z;~%5Pj(OC-KF2<G
z!Q$TdERL;tCu#L+8lLROfy@scm_vOB*+F_&wfH(@>MD=?4xG7hrm63$F0iF6ID4hc
z+<4SpT)!XUsYwPMwZJ)_O+UmNQXk_J8DfLg_+8iZCl~QNtK(zP3N%<N2|C*e^=h&$
zf!x<^ATmhf_2p&lUirXoDC9!O%D}XM4qe1H-rGVOC$1=P-tLPnlyE~Phr9c)_NhLb
z4|PW0U4Qh(o`AZZg;uI&^0Bb6PDn=Yn!vvX8Bn@pUQ8ytOToGa@NgZz!Ne=&C@!Y2
z@&tuT1dR{f;1W4iS;~Ff=w%ZGUoZ&ONh$Sbehc>vni@Dgb!oDahIn1IeE+Z?a`_dj
zxu3VF8)V|i)sbMH`qbdOekS<5`MjykNj=2gXPrcbsFDH`k$bdp((k6n4}%cbugz`A
z&SD-``D~_aus$sfdmu$eKuzxedMy_x0J6_MtT^@&cGWV7mo?UA-=-A94XU?iG`Epr
z&NhRjUwIw{k{5FJ<KSuScvqpOHC=7xEt%VWWO*eT*ct!~1wGTaqZTK3B*ubsB!h&I
zNZ)0w@jVoE8;T?wrlheVDg2^;>;_*bUGL>DcJk8OXvaFtfe1XUp=+@d|Ca}+<9L=Z
zI!y$rrh&J01{=FM`1tX9PST~x4$vX~BF=qZogv69-UT)L^kW3j@&Go*yIAG<1iOgE
zS*XzJUd2dM!}j&jeY-pgO26I}ygLy*Dc;qvD2hMitc$G13%Y9l8u7-g5rrLCYc)>0
z4GHl6G&ovyVKnmTZQP%z{jUU7fL4CLk=tYGLms*aKX+k6{N%NAYRi=%E0hE3T4xX|
z3&P42v+%?MIDnryXJDAtL1e4dGA;Pa?xWWGp3<TFsW}2T;|8c#pE5pzHB!*dOZo@J
zCD9fEdqSyR&+%LRJTuVN<l`TVUD-v?ZP~Naly?)#Lq(j3<kvOx6U)0^W)qK|>uF0_
zrCNiM^RSnbWYhl&G^+6I;<fAIForOfoo4b$ilmVAHbfXP&V;(~P?gB)1S?s65^-5%
zUj@|kSLasw%CAJ%_2o0VffuFo=h6*h1J7JGzDR};HU2^rc7~+_9HmC0rLSc}<;z$J
z{X+H~-)G&B*)HnsdT9XSV8}L{WJ`y>=LHDxvH_)xyBi90=i^RaAYR<CAcP;AoPGz+
zX%Q!=s=ABfcwxW2?Ay&tZao3`>V||nRm)R-ps*RA+R%%WOTXPX23i@T5s+oEX5Hjc
zUjIV>0GOK_eC0NP=6DNGVzuRYOVI^a*cT_myd~v|oD67`egHYxolX<3l27isZU9uX
zQnO6}BuB3uJi_)6l9O4%bcYG~K?@A^6DG1@yV&&gI|a)>fe2F+o2IO$PeTNiAc95>
zPeXlcSJt1O_a^OLi#q4@ue|WW85{1`yj`9+kNrA1vxPl<VzY^ClJOfs0#>1U$2TQt
zAKCID*~k=?0)@oelDo_RWirjB<#NbA>g1q)UzAF|O`soajP0%E>DmUKX+Enm9!cd#
zw_~yQ81Y_KMZGNE_SoNC>!mKx6F4i2R@bZUUzlft<lV{|9Dn_epieoH1}!x2;<nzO
zk5XL%+8rRk4Xdb%9wMV4$6@ZR&W{;jZsUmG&A=TEELE_$zHxk8x>C!jGft72ILHy!
z>rvplt}D8Itp3CzUZvkI>&o{l@y&63OR~GVJCQs|e}Iy!*XJMCTm)dA!kl60g}CEo
zNK@lC)6$XtC&bVZ1K5e$U1JgHE3bgo52&-)m}m$ZpIR4|&p3hw3mf<Pj>DlXT9Sk~
z8BTj?x2HJ~`?xg=jo;%C-{tOk_rJ<<sVV^nS!;)X=x^Q(nAh2E+~4ueF7jyFRWH#w
z)!H5L-P1e+0c7v8o1pjWmGXDs*y>H4`PW44)%%dx5zk>~`fDBb$++y%T+RrHtl4sP
z1v9<o$#arUkjY7r4k5U#{SMFwl!)HxFh%UmmVYdd1p!sP;_$*u0EYU!meq%Q^1jh6
z&qVEj_S#W{o5fjME=S3eywg<<ne=pPeNt7ih?NT{+s2i=ylNV=Vtw~Agv^!}r54U|
zyt_6PwrmFmu8O_q1`MX%TAa*kW|Xf_y;!W%uG~xcgOpu0!7s=!rF`vp(i#@Q%=-H?
zwpo9-GW%~yz$2;McB#adP|kQ*9BQW>3)4OY>m))c&O9d%B7AtOWo9A#+C3-a0T2ON
z3D+u?Mhsp$_i6FQ=k9u?#O7)X=)Y2(Sem!RS<ucz)5j$}#ry1`V$I3uN?|DjPhkD~
zvBM&d#QVXaf;YyMG}CDgaSImO#hK*_u@Qxcp=54abyke=$c|}roJo4(17LvD&;#@s
zU!`KJhC&}RO+%1tv;2t`L~)}^`COrnPFx5LPNi6J;6EG$UW0Ip;E_LeA?Y!L;iU{~
znPDE}y|{FrJqV(^B@ezz2uBSKm$l5FX*h^I<~JA297xjSTM{cdcQRuw#7J%hI+@+H
zY~H0Eg;jvB{N4K0NeJaMxK(LD9f{KkLE~iSrxDW@Y7XPedokknbVR_xaSnH&W{|ql
zZ0GqJREhW6ahYeoazcmD-_hVYxX^9KWI9bDXMG!X&rc2D@@&QN@x!X?!k7l2Zx6X>
z=2ha+o6`Um?c^*5I<9A;8+Y26opVWF?zCG=Beq#3xSbxT!+~wvX37|P8Ia;O3xFIz
zAmBxyEtHej$68Y|q%zi=DW#h+r*cz9gfG0VX`FWKA=9)hx$ZkA+T7nougC5`Pi2gV
z_UL?}6E~j;o$V2*&dyndzTP`mt<gYBnV~4oih8p>5*np_j&%XBbJ@G811A7O)6f6)
zfXEwVx{Sq}c*?CMX=(30*|{u<cU{zc&Wv~6E!n5w(LR<fgMDyf<gWS1^F@sVNIbw~
z?ZknBzC}^!GMo-aiw(X2W(wlIOGga|GT{9k>x<G?G@0t4ajPt~aO!*Lz0y$SeSQ$d
z2N9Im=+{0F9pJ(CKISSBsPgW9ldAj#5uI%Gc(jV~q9Xt{ds99wZ|^k`2S&7BkT87<
z*%DdmxdOjFob^nKEC5AYN(I&`p5llwOYbDcuh<?~$O;xSwdgJETX~O5)4IAmwxzut
z=ox(+;Pv$EIxn6)JMce*X(QvV4Y>I;$&N}mbzI45%t3T~%MF;U;f;Mz^)>(X+{i(#
z;ruMbbq{J#?1%gUCFKSn`V30uD4Ip%odtFb7+aFS4NitEP+{3R#W+e<MnDX!@>`|D
zkwW;^zIt;qg594pRZKTl17AEDLB%&e(SEd}=;t-I7zhQe+})ms<_N-eVy&{#m7?mr
zzVOEB@en{&d-MwFD53Y^JOW6#a$8j}az&kg8)NWE^$MU0Mr^6NktEGn$T0xB23>nW
zan%rQvg90NecNVF^-z4?^;(`H$ZAkEw-O(n0B@RX>BCR}pZ(0wCB?wns1pjTbfAgQ
z!9toCSiP5nkf<9<;)CJ)f4JQ@rX6j_<jEQ*=Wb8FoJ0<2Y4tZXe(ETLpy3b$gj=(7
z3(y5@n%iMV1$ttGA&f#slc<ch5l)jMpMQ&{OTs6*XJ5W1hiyxCECJ6M)jp8Q-h+}1
z$UW!n5r#KwQMN?SLbSKmu>FR#ZO+$auPqJcV$yClxS|W0pA+qJPFY6{AqyP&8xf{)
zg^s2s{*<>}8CqgY2A$2V@~&JeaP8)1%~xl<4ia9ZRrH>p2&jW}PvR|9+d@9-xBt$-
z23cIp&omQmMbpFE=495fgv0&C9oxPc;<nYiFYMQNs@G>nc+g#N4@X=+DS~!Yg|Yp-
zn7nG0(nj*Rjoh10Qi%5RpW99@LKwKwi1yP%kz||Vy}1zNuZ4`h33x+Cl(^CC$wx=2
z5s(y&M1J=BlgB<5F$zM_^SuNERMLeA21V8{X0&xbK&q91gww>b(3I}x)B}QWx`3*3
z;v?;KwRd@{RE{v>meX<Bz@ZiN{)I}V;gU&7mP3v$MjcGEr;}`_8xxBXfprw!@lkVq
z<&y(t-Ff>mE3~d|Xg5@j7eFAX^jyZo*_gsb4j_`T2a?)V-c3XkZ0{m9RvO&)!?CHq
z5eiMW+THJ_zJ@(^-;2eP?&-={(zyHi4j97J4=WC>j|MLeh=NQ^S!<1N_oU%`Z`R49
zHHEQdPH`>(5sEypp#4=8I>2Ss>RApuTcEh$28q6rLV|yU@`?SC`1xdD06thXY!fV?
z0I2j{FOHH{ZfJVN*}>LTt&XdO6~ntlcsim+y{Nxl^K1}(TowM0V(&$5ba_Yb9P>8%
zC;**#yNYC9)MyVi+4pLvju;;UeV+YH4sMhW5I8sWRfkdjBm3i|UJjaTI4M}AzD9({
z*-riNUcCv>y9V8rT7SNX;-WG%e*`a22^3@fu^W7U`vs`s(Y&DFa`>oAzGSZhdK&1S
z9i|vfXsz2yCGkCuN!y4txM(NZ9AFemmj-Hj`TE3U^#7O*(#3w+-)`CmQ2FH3eVnv5
zI7uAWAf$xz4@Mar8VcS56#2_dNxrow!cTl&;aE@sYA9X_{T9N190jcKXeLkqM&FwM
z1~QR@E{*>Fk>!2?0;!=~%MXq4`kQ+)^aRT$xD^FY2q)^O%XIs1b3+#j-v%W%Ds763
ziVaQt*!}X4*l%}3`SjMm$@@H{+<W_;BD`=gE6i{>j-&wwd<q6{{5@dG#@~_lgE3QS
zZu6G+YT6Smz%GG^lI*pc$F^0#0Wl7Lo)0Jy@8CBmZH}kFiy>cIJ-1Sbid$~4d~AA1
zo#kfB&1C-y5}dy;&IO>SKe;_Pyk&MP5?hclPz2^UAK@KY;|S;guef$CoH$!EjResm
zQz*UfK6=gd$t3@>YG46CxDOX9|DVDA&u@_^2PZtcQ7};VcS8Kv-?0sl0ke;=lK%Vu
zzr7sn32*_WeVXykfqz;-MuEP$^f4*Nzpi5O8kmll66P(W|K0Gf3#LB?-bd!L3HaAl
z{^KIhAIALKVDk5`gph#@#B!MZds6@FEBXH!D^3vEuTs>1T_pynu#<dTxjFykKOqD3
zX5Lm~ynjWE2FM4F%UvhCe}#+(M+nF5LkG>jt|EV*0ryk+uaLP=e4yC-e}Vshf&agk
z@zz_@R#jGS3po-#6m!lr_Gx<D);Tt9y|MJdoN3(Ato8EsDbe)&)bT9sw_|O8fnSuh
zh9Al&BOHYH=H?x-ng&l#1r*GjoL=wEmc7t(ol-YGTy9!-*)=vXp-Mfhg_)KccZ9I=
z^OG?*?3I_{b*ieWnw~$Mk#BZ|Os9`r9V|8U?7k_EA)B#o1~y)Yp|>-gGhF&Epc2Si
zPEJmR9J1+ReG))eBy`)KCjdMZ-Mfp>rJsjfZ0SPjh5($wsCeJOcG1NRGp~?clEZx&
zrEz314(2M%lNI<2b!(W>GC|j8kyD<Kp05ZQ2T;fBu}aH{pAP#~Gf^ox)1D>T74P>1
z)~}50mm579M|1;}R&3YLnu`4<TuP!B9xL!~tB##7KxYie<9oo-!}>FT55X4~FpQ+R
z?2!BkA+0z7CSd*UGPd{^CR|AaaH&{-iuotb8GQsGC-h>CmHl5}0Rs$jd$jsjA%6lO
zYbP*VQ6ER*XaB+?UtA%avlwR3zl?UoCIB!}HgjS88!!cd`(8`OH&Z^e>peojU;_#W
z97EEOf8%l5eVBj9uKa6QA>RQy^dk$F=)d7!5xB4Kc((n&h9wH@CS09Fp&9?$2exSe
zycgUV=ls{OC;`kQU5n@#{TF7Q#sJ&N|9SXL$IpQ@>fVs!mVfwQ>=CVr?^CeM-$=nT
zBW&$`b~aw9E~*k$3|VS$pV`$4{$mb8!hj)vFYfM*6gP3mP$39wa5t<{65ahHlK$#c
z9|pGR{kKa06)=q?;MY|pt=~xvnX^xpWO_C>{(KgUSAc9AdfNW!pi~DP14K1VtNamt
z2mX8R!%S#}Lp7o#rP%vPmQTO?5&@l!%lN9~=GeB&jWI;NfrN|vpV9d%zx)vFfL}8I
zIa*t!)F7OPD7ff?L)q~~5!#>^cHy7-!XOWazAA&4E%?vx|H{`FBEYm!5m{Y?zN#=H
znJ^-)9rWJ;o39=Gn5t)Q8Pz=^`ePYva{{y3g38%=H*A-x-nK=jlF~wJ%*D0}8Dz4#
z!0gz0F#8w;fKv}^3QXM7CWYda@DUGFSNH@yrmD2!mz?)8)*s7-F&K`pJzL2%4b4B_
zLbunp@|o)-Q=oySi-s0vOvEd~jsPs#hw;|31hO!eM#k^qn?P|O5d=(v81p0@=7g#Y
zBfCirJPmsK@>{=AL(8<vgF~wrJt<Z6<v%ABjAw+kdi-=e@QIXNR!Xogn_($GJF1Xv
z!p9SorpVn&1A^ckkNxWm!apMs0z9!D-PliSc;?Bk>5y;XnX2(-=o#~|GfVBW+8%DZ
zNw$AHoDkKAVAe0cK*6X|7|idV$b}ZxFd+YYrHk35KV`B#!tQpsoWiqC_UBE$0ykk6
z9>e(K-48+&_cl2<@z7sB<2M3`@xS5T^sU7&4+OHKr!NIh#qZvzYdtl4b*ecR%pRl-
zv58B6WA=mbVM3(K0`UVbZ1W1?B^>&bS>~(UAfuc$5#{2{U|q{6U)12)z^@CDQg$Y3
z7yewJw5_Bf2SLp^pJQ1OeZSyLHVP8CN>dQRmn1!^bvOm-I5&sZ%Tc{?v!B}eZuNyR
zb~&--&oHF4Qk^AuIZsoL@;^d-O=KpGQ2#-eN`ar+QFC8Cyku!Wmq0@lb?CDA2Qyz8
zu|`7qL%{OC1kmavLFw@y*&}4chypY>gz=tG#qwQ_{NLa7r3MbN5(F(4DC2A7yIbGn
zX<OS->7c{73<|(IJ41P>hkrg)BoOCkI2xL%LQW(_wc&AZjgmwa5)JlF>Gfnl=Hk?`
z1r}&r>uv9SK<5Vx?SOD+>$_UbA(@HxaAMYFfwj~;<C|`tDwd&(G?6UgH^bL%!n`U{
z^8;NVja=(vmk9B&Ssp{U8*Zrd%de>>>G4T{4_3@46#rmfP;~Fn)6Lu`pJPXvg7G|q
zM)Rd?+$C_`z$zb`%L&-RI-IaNoJJE~;tYI|8fYrVcn}GH#}@#o)-ViTiu~!y?gjFt
zjsQs_eUC$DkXG&>%dBS1AcTEIM@tKB=u9g?_xmRC*RPpG?UfG@(SrF=<|mHxc`R^3
z9UUWXZ;jm3pFfVpH^%x{<K^kU$y%Iu>C4idU~LtWhqd?lL}hYFf7%5tRPPi*P`}Q)
zV<*Hth%qKW(d(^7)VUunJM3h~hYYZ^@Y&3jQ!oBP;{15nMEc&&j#<yKi=ZqmFE8)j
zT^$02%$$#RNmaTUYYb9Pk_zq!QKuRf)^0jK`aWN+Bh60;#SO)c#|<{H(dG6IR<|Ne
z)coFQwHxjJ5aHNU06YjTK4S6x9r2eRR?lNK_{3|m8RYFT8u%PPif1?1b77ZY<#O1)
zy+`o&w?iLw{6-Ru{2(bHwys{RM<3=#`}9y6e7^XKl4MUgnNGc}*D5_&*Hwb!>8k^E
z)rRG}GUMp7fy7n(%_1@)WDe#1o&+PJ$30jz-GRVEAQqq8^Ggl@?XKqaRa|jeBo=TJ
z2frESdXemRd$}sD=%w;`sze`Xum$kCjOJ-5bEzU$yNC=)um7@g16EiSyg!la5eg&F
zlR5Nl`}C0C9);Es>($ulS0X(3-1uS)nJ!YsvAVw3M(NP}02!^bx}0!Ywj3khp!g?k
znvYX$7n9wSB-lldAeMQm=lj~-{DSs5^E9%ULUO=p24$A8<JU%=Xa(T<E$0`+Kxy5$
zvsA{AK9yX9&4$l9*~b)#j$>N{c9F;(Aa;{q*#GSALLI0-!qhPZhi+4ruK`|d_A)3h
zEQf#Ja?FdCEtiU6^@$|ulT^SB1wKI$1RUJK_hutU@2eM-FFBQCF-bUr?nUj@>Y66o
zMMIFqrS^!I#;kn{%!EW6*X*s0$d{w9M$Th#%LC>=;oj&do;)5{sMn2S{Wj}Nq+IuT
zm)iF>+!NiDq^jaF<V>aY=`n#@PrSy1Ow0mSY3(O#T4P#wC4wRB#vOV4`RW?LB?x%5
zX*+Fk{eW3gWY`Bp%|oy=QZ)bBtV8P~WVLE*v#sHjRX{*7EYxg$rcIKh@h<zsq<yq7
z$P22*9$I?a&#SbR>egQ1W0xVxraqcr(ts{yA3-<xsQf`CHYx-FubG}&ozu_q&%|;I
z*nq@bN$h($V7NJ;Ll@D&ySR!KYYQf@jiFWn10EeG&j~4(0hw*)L>^9U)QAb?NoB<E
zuL^^tU;aGLEQ<R|&FU9F$`H{xOM@y(<lkpP%oIE=pp6mp;TxS8u?Ba=1MnQh0KAWS
z-L8%6WQWA6B6@&u&xU4^Q|t@q1zSm+_RGmE)~f)i;;9$!y$x`Wr11-|xk^jbTELS;
z4EW?$DkI;|+mY`+$n5)<jAB4#A@Z(1cLg|%f$vaLH)J~)P+7G8+>9ap80}|f-G_Uo
zaxH&RBnFYywxs0{=p;1sz4xnxMJlDKhiF{@AzyKJ8CR>Cin*JjhEPXy?5!H1C_~@d
z$@4?@64tutpJHdt$bJtuZPs028+cqBaBpbs&;-o)9xo9ix%Wytj<XEhms~wReD4y=
zcB8MEYw~tqR#YX_!;c%G8KMSG*|fV>F8-_m)8VWq3$K*mguMFo3BlCUAOks3nKxkQ
zv^#P7L8^4f!^MvY7?NurD+t0)^-lTkNg*egoOl2$LER^2jc@Z*vG?agm^k*D_@b4j
zxs!&>ol$Aa8$UfKQ_G`b<QIRxuS+bH9Hh0dHzLgU`O{U|khbCTeYKredicZOo=T4F
z;_+mWv}-L$s`^9N`h)_MS>!B}&7kXc%bb+BhM?uE=^8BehJ_`1Nk;VcD(e}I>MGq=
zHI}p}egK6Qbk7%`xJ~)=#3=As8o)|5^jy`AVc(2m=-8(SLm3z{j!}O?wdT$grQ6i2
zH6gS#aukVhpL$+ZP1KL8)~dh8`%!Jp_FUVhMrXKh_@F(TS(}u?U*W_NnF|B!EgXE`
z)GZBXT_0Q&hhFGtT}~cC$J_GTXo<Y<Dv3F_<dH8h9zW_*P6rMLF(?9y>sZb-8rSR4
zqlZ@{T+AzwWRZd<sf4^ZS&<c%n^&`Arq7(+L*~BNsDE!8r?Aczqc<}{y>BiV7FN-q
z9B<zecqgW$=x02a%p%bqGYf9H+AWZ8ycP`_BfdPA%tMxV^X84I`*Qtq%=;l0<U<{S
zp|NdYz>Xann8kLL9j#*G05qu?H%hKMO_9T?^&`;nEuZB6#JAQa0$cpvXF-nFobp=+
zXN)8{kl{M4D4^fKE+p6!*Qm#5lrZ^A6Ih^4FykDv>(r@>449vcfS<TOXppbjL;f+|
zz{|3BGtqj}mShu6``03mMO@urDecS{x2`f)+s4{ceaYBg0-MkDE<PPM)AO|PD#VHG
z7m<G4ohS-gPK$cFq*3AMQB?C(Vn!LJ^o@dk<&D7R`@1E8@w@3!P3t&*9REJ8DlNsA
zeY<Uh3(Xj2VYSw{+jXETUNjxO;mCj+gd4c-HUv#qJg@IO;v(oW9CdGaej2&(7$=1{
zz#;kIY>HbGVHM5!4qD<z)@5`hxeM4K(G|VMW7deY8$@FfjX4dSz~tvwUoE`8jH*x&
z#p(@87bei1N`s{5s+=k1ZNovPl64l70^0I>5;7&<4T<$dRRet@>ORkx=_Zip>AG%}
zvw8V-&|%5uv{C&7`)~+O0J4#G9TWZK5In1C;deK#1J)<`^ONsAK~5Ckuar+aWfiYC
zZoi1VYm)Z_%@pfjTJfhDH)lAagDKwNx--K^)CHB}dmTMR3=Iq0CVKZ7;m0%l0cH-@
z<o26*#*lKte0)ODuy!poKbwqMYK{K<FgiIOkI;XbIO6y4mm~AW&o}5k?L!RjygUHZ
zN{$d7TIU6SQ{@R4PKn=y@6A>N9}loZAEw;?;Fy@05N3X}uB4|Y{q0iLE|Yn@KaURW
zTv_|<bhpW)Sr}E2=%b2PsxK9o#f6PWkY7AMzIW0oTx9;5c=qwb+Ak2Oy!0MTuK-Sr
zj{8$uT!zxjLD+s59UUFO$rrk_`kmX)qF7-=>@A01Wtp`dh=wp#6hvd4oURewjUGP?
zBS02o0q92d%cVlv8EJ%T{nK_io%Ox1pw?lptjDdwdU^BXi>TiC7HE+@cmNt~?{~fM
zn=4=N=BNXbMAF2}U|fmOPtC54mc@``XWX<{^hQJ}X;#5nwy7(oN(uBXP6VP6CsIRk
z)o@He|5m)|IYUOR@&iy0NoiH_W5T}n#C+$ISaSwQzO2#jE`Wna?4&?t!B2jzQB$w}
z-2W8NCVf8l7RECa*HZZaAoOV}oWu2WOmZfP-B&-e6eq;DKJhpwzQ<#Xv{i)oXPuwb
zTndRLui%6T+-Cczm~nCrvA%4TB5rU64F>eJKiwA#7Q{|zVMm|Nn+H|sht?w;irm51
z#a)+MY2?1ZL80#F_*|Qr5zJ`7!8UKgedU8xGltB7{$!7}u(5Hvno-WiRZKSXjG)ZI
zC?W$7V-G;wOhQhn15EG@+l!sz$ZTtvUyo$JNSdMA#D|NBup))_y=akQoeoYzt-l$Q
zDzjAxY5NKu+d}pF_4H6RsR(4FIZL&Wo#UX)C0zQHGW&mLAOTY4-eLTjxeMU$;}46z
zJ;D9Bgr6M+xVHB1P3VZ)j|6{*DR%1l=cKslC^b`}wmapAnMQHbEJ6&<cQ;uhZC@}L
zqs@Osj%HKl5F`ku{M@CK$;7~nGm+RrWKyaRDm&D3QI73WBJ7bvoM+}K)`;P9wcXiy
z^;JubAJ3ot%dZFRq7gAr<i~pud^@UFh5MfE;tiBD2#DXgA%zv~X7T!j0HL=o3$k6R
zf4c~Xo#e8bW7(ze9?%IuA7SJOEa9%FRkoA^z_Ft9`s#%^)jT7%`{Ow6sr=sU_togx
z!4e4;-%;zu*RIcN2w!7awZUp6Te$PC%AezF?YjBvJPGZx@~5iZ?6T&k>*TBs(h0QD
zmOyIO3Fvu+K?!VZXLbY6Lch-Ts1V~jLw*kX12WeCBI-W{kYjrCS7cE|KTZ-wrL?~b
z;RY{LIfL0(W50GQO3WzW*;{>Xh?J>)^-FfFWD|{MNn?_^@ul(h33`*f6Xno#GV2oy
zkPL-_!eF<A^D{X)lM;79u7j-M=!acsfg}ZXcA1yxuU;H2o=0IKP5G~0x#%+V%4J;P
zJ64jqvWLAEbQ&ur4jTlaH|Yt<h|v^eFoxg;F{5qeWGH!|<#k$lp+$@6kj7B|^Xx!2
z!~tDo+wrCa=mo{hP~piS{~N>v$9UzH?SYe`pe9TPGAVEQjZ9W;Oy>yWp~%U!=0W!t
z@=v6uNbG)OK0IMdjQ@x+D3EH%@xiLRe~IxFNVm`WI3X46@OIiUkQ~O~xq(HoEuH$V
zi2Cm#RDwakxrv-_y__59eVCQEshqt_GE1P!U6I|s9n-H~N$%VJ7JOk%@WX0Lk|1xH
z9Ocsy+=tL}1(=*?1*=c_jNmPqGZ4~S%6qfSuoV~9oKgAdEe_uE*ca}auD&*;AVmH`
z=VH7~H4^;G=TYv5H+V5;U%5Ux<2(sQ$20$1yu0r}g#>||ycUiX(2N8ovQK8aef04l
zUGJw|bv=~bvv}6i`ryJK$IzSTAVZj$WEu0<+>?Fr)~EYAgu%j=Y_Ff>8UdHajhJI7
z=u!Fp7!Vi#SDd9sBcRYvNY=VE@gS%e5n@l9n0~(PKam*}XCtOI`gD`fotxO+A8*Bv
zBHtWGXmvORqmd)4em!BzQwTpS#opca&O3>WZT>YZ?~`PE7|<}dB%&S@Ow(1M{gaNR
z$AI6am`~L!=${?tUVvi71@QTWV*>HUIEtLO`{=enBABueA)(J1@5)DnOw=wb7-GS`
z!?Ze@i2j{1;J`p<gOPEQU^;N9S410=BdacWgI=#4(TM1u<nd<)Slb2!@rtn*rQYXu
zB$Z<Nzg2`1>heR)U-imS!l^2>?P?P(0BC=~FjJ4iSWRuLMcT+$!a5{4>8+0cQ|Qe^
z9lGTu-5(^65jVnj(OOckJ>FA<;$|<!@2Shl(G%j|0|p1@Fa%ig88*@4#M3%;*~mw?
z;5PNU5KB%j{igQ&cDLb2Ydj4t4KjEfRxPj1V-vl(jPK@_jqP5O3?vV|O+LC+DNSHk
zEa!PP{V+pEN&HWfYHOM-F#VrUmD(=c219bGA`@ek#>8X5tP)Qnzewo2nEf`ADbAAR
z)^<PthQk<?+3=N6Sw{7x%2du-ILEGrusB7L808SvXPH(S<Jxau${uu<a0s|SSxnW&
z`!HvL?UL=D^{7A>D9&lHU$MyrHh*VSFwbsWJHAIZQ)oGUoW5igqr2i5!;HJA>cNk`
ztDG}CSWN85lI9Sa1!3E}x2N^(GP$lb_&@6>;{%3l)I;(AnVJ7J*|U!LsC&yG*Hr<l
z%+bI>eDcR(0z|&d_z@qocor^ASnJ~5dfL16#xydOteMfMF`)S$+kN!gS@D9kQzR+G
z@d^2~A84N2v2em>UeprTNI1D@4ROI_>TpEyAAG$C-_tNll~OCymC-!P?L4s?)5Qor
zUo8qZmFe~TLj5{Sqy(|EKKZ|Y3qbiUoFxW1Nx6!?vaK+#TGED#lbgCDE0&5WxRV)>
zyZ<Yf!ofG;jM^KaA#{feb-A+FhsYH6fz<3WZogir=+C)MitEe)px5X|{({JE-061Q
z%ZqvuU^DC)bTL$~!@{)jGJf8^=2UfmQB^NJ)gZ2XxKRD-ZCCXq_a8Kx);FOB!}5p&
zvBwX?cMtBwAYfw|YY9ni!$nam3VM%f6BAA~1=i^B!7?t#xbg=N%@y~))?zlZHhv-{
z!JI3ePOOax!dkuY3@%IA#D4FL!cX00^F6sdIozx3N;_caTSmnL;%oE^oHGjVnz3mN
z>W__s=S34@sh(!TX7mb=^wfN!3{~8m9qkMgWpnqtd_<~#gVa{!e~ZrDbkRwGu)~_;
z#06ZHw_U-W!9{W8rXsuxnbCs~M!%-1x&rHPp?dR?swIqIjt8MMm1*@;jN0jQp2xxQ
z53ef4_hi?(T}u{074^$&Lgzv)VNWZa{kd<DuhFTtaL`mrSf9tHjZvg&of<HV|3nJ3
zY((fdxrz1D%@cq8YnS86X5>DWG}*(P2Am9(obU*rgs91a#nMqb@ZZrOeYeDW*WFgf
zdTItr4v+qZobj8fYel4$nli`M%w45~MNe8E|H>CPT3V>v=nWd<n^%t8b~(Xxa&7N>
zHML639|+><3h5V6ot1W;hqt_6$LxtfD>N1#9|MULDnH9Ux1tZne<x9#=|m#$e8-pj
zsUiBx$zGC)F}NI_#yW$=dP!djnp)^9-kPuoBvU3^?kB`}<wP{59@l-<4SF@dKUUF8
zAOpk#&dX?(5@f&EMb@L|qIVM8sySmYZAXY1<A!*#pbM`sCtFQeOEYKVh6H=(QuRvB
zQvco-HqSpnGfk31kioBKWppC&N^RdUKo;}<C&&De@F@B%0$G@Irg{ClUPM>ZuPKIY
zyw!AYDrdZXx%=`NKYgCW7etmR69Ya}P;(+nZ7m~X&~lCB3s+@}a#jhdDetkG5$#Wb
zV;QDlS)+WSB#I0nl$gfXXO;{0{hRU!!_j?!W#zvtPOArQs|Z)>OiG18&<9;~c`>s%
z5$xCN5Vl@9e4jb5_o&~voY^$1iR-ByzjUZ`$kJ(aWkz)1%SmXo2QNyU*iy$UgBebZ
z-iQNz&JXFlSM<o{50<J?A+Tlk>^eD@KS;UzmuNr$0}y3#a&as;b6B%8KX<VrCVU|%
z1KG8&t;fQTR;Fpvq5rzkEXS{}8m-ZsP#sLu+&;llFR2SeW6p}7UXs!Jk$uFYjar7U
z!OY9>zpDc`#Bb5+*{$T(Msble^7U=Eql&I}$1B>Rb~ztcSf>*+!3_;|m|PSFjv`o6
zVhpssujG#wNd&)ATi`>ptWMj-2~H_~{Eti^<P7X(N#3^H6$CF>&G>hyGjgIo88BZR
zbB7cU6)p!3<{H@I-<bJMj-3Ut9W?$TCkk~n=JGShh-iID8bhb{Ey0OCTm+vm^NDgT
zg%8I+#bGy~gmRpXJDD(u3ypC>!`E2QU%xRIQDh+sp=rrHih*P7G(cmG3CieJt--w_
zu2QU_MioWn-p`uEdu}<|m7$Bi^m^NbC9M6OwA>}?NhQ7$LnJp=pXThodip=>4>(<k
zV60mFa_iAQmB9Vf1D-mR_UHa%K0c`U!tVz$+0p5N>b(+8Vjx{Wp*b~R5I>KuQ+{D!
zy4A@G&9)@&owbcNtyk>%X<Tq@=^j<SZQrw8FYUXpP}ce=V2NsjFgQ#H#$>+356LYg
zi{Q>`mLOUGZ&K0|fD&V<u{c>h4Fx}QbV-z7a8(UISo$T+YP=W_D3P1I=8rt|?h|K$
zDG9cSP4CUO+6M>qYJFD<o33Ty$0E>q-7nSTasB$L)Sm95E7KWoGyl*=7~iJ1ay=7x
zYxc<ew11ugErs<TT1_Yhwb1go&(RKN8~IUKI~806Dl<%w`^moFZ;x>!jxt4!dUkWA
zlN<IeMQ2R)l1$r}>NEshV{SPSMW)8wrN!jJpgbBe<J$K_Vp<K|ofZa6tF)`H#^U$F
zm=-v(axKhwE%*%hA<#r@OO~XkmRoD;9OI>+>q;o6u`^vs|M?%)9ryk4>*A98;80)s
zf72b46CcvHLXOSNd5yX0Z|Gx$>mrSx-yNK>ksQHQA`7o<E&j#BI)Se8?j*B5At8an
zIJdmJGo#ZC_MQ251M6AR9J!`X9o?s?>4fSe=<A6<H-YMJVSO(@OnBy`{y`wv!td0q
zSfn9lJzy37+C}|SfvonH=*GgKL;2AOwR4^5`Ty8@>$oVlaBWx-lu}_B8l^j=TVm)T
zhZ!1_?oLHf5C-Y)o}s%_MMOek=#rA|5D5hYzlVLcy3cvP|F-dSFV=IfJFn}yRVp1{
zn31_E@+8|yp$hZaP96+0bJ_dZVaDW#I7tm|t%{dS?x=C@%-9^_SGpkIO=3DHy(TeJ
z&p%p1#-7!X@kmwj=4XCQ0B5zJKk00IupVPEz#mQdKj&$Rh&nm8IFgg4BcQO@OX6^^
z&lr=GRk)w|)rO}$X^=I6+?oUU|G(uQ&3I3tE_miWo$;XzVEY2zP@SuGn%yX?CyFos
znji~!vLYSO&;bH}+>iq5M7>+9e2w(%z@qlKhMA{@`mM31_Oi_AV^R~Pac_n&h_O;8
z^<V2~E4qTnosnApEY*-%`owLMIxTPjRzNYCuuQX{Kb3^F15qzUwFrK28r1pxz3H&D
zx@lJ((S%6>i1{nNL(Z+>)XY>|fr2@6Pi>N5j_rxm#Njov-j?LBa5+24+dS-J;1I=$
z{xT=<D;2PHQm+`EI&;+*DDVyv{tk3_+piMu=LZyLUy0b{QC*fxJ#|T34mDMspnAP9
z$*otaR4|w<bF*>A=eYdmU(vpgb<^2B?j;^Sz`0wm2Iku;;tFQR=`1S?V8LH^Tz2%H
z`fLy`-_k%T?5uBL6n-fQvI8NhUPqPCgi!@Xn>r7O4_{ZuX89~?(~j6a>6S;Wb0o}g
z{nuFw;^lpLnk(k3_<ztJ*RWJsfZ=@4#uPD@4wl`c%s0`!^0vQhSl<0u>71~7QQG3d
zQkwZapH84=h|~j5=L>)1po%J=8Ce^bty+e#?FA<7Sz(!PwT|a;-P53nZp0cthselh
zN&`9@^m);AGeNNt1%X2jP(1lpKe!kc{>~v(n)3}@qM?Fc==ca#K8W$n<U;D>G2XxY
z>px?!^5kaKF8|k=GK*3i3T)q@cpnXX-bAEvp4li?uP$kFU-iLMg4nOAx#J{C6wWG6
zRwPOk?K^DVIz;~r;@DHPM-QK5W7Y@<^q=ANP|k?v;sxspI}XjZZ@q1YQ216+){nY*
zz}UP>#kbLkK&<v@kfNMZB1RGuzx*3Nm5BjU8e285%7wHkpa!eMEuxZyLb;9l%A8n8
zJYXpUoC^2r6^nc4nX#;xGw=l*-oA)5SQpfD?b`AnKylknXHR#D?AnoaQj#RAP}r9S
zEe78fu5`_nN7*x78vD0g^E^mK$B?15xN{Bb8=FVoLV7-j0<Ok$sy_TQ2rUmFDE~c7
z?iKohLoduNisrgOzxF~R-r+$4{R2%URBd&OuQ)!QARZr1s*3KLcE2%gXqi`S23G&4
z1raNA7oOUiP>Qb0rDwr?2`C))`b#D{lM76$@>mPkYif}eEA5N3>N$?0ckmqQo_7f~
z5T0_Fd=dhksX&-xAt4Vd@=J8#YaX)C{7?6THv{|U<brEdB<w#U;eWoK3@rc(*<qYD
z+Y&cOEhwl7ef&|dNh*ptb3f0ZZp5ATu1Mu}X=N+R1kl~-QFFG&)Vi%`n)Rb~2Q-U<
zp%TR0t`i&v8<r7=+l3iH&a3cmPX|Rr9h^~*l3$H~!qRVw?fPga9%fL>=KF@yLnQJL
z9|I8;+5Z0VYy;BL6}8J#EJbF3x2nNGu1#CtWnQoQz{AG-K~K2nBdli_&RHs;wv6_s
zupO?~9X>v(yMN>1&yxFU>*)h8b#C!51Qu?;1;Az3v$x)U+??$`c#C6BGiWmDTYvd5
z-RJOj6@y>p=Fjfk7n6Qj=J<p}Vy~$LfH=$7vTz0g=A*k5nbp|3^mlX*XICMV3W@Ao
zp%Bt$!!CX91}VPS!Lz;&V^A<pZ(Lu2%EzXCUjqDP^uSr4cV=}6x|m5;pQ*BuR?{G8
zWW%{d5%{Tg$K+9A^5~$I;7ErZTBCi1c0p=;9*yOqe4Jgdc@{)ShiN>ebO;@7Z=gUm
zvPP<*>2e34SiaM}sXG#l6msDDZjQxQm0P(N&~+fdvn0On_R!}--x#ysF#dM=*nqat
zWAgZZ_cN0(>m18J-c_7GpS|QZ!t5uyJrsWBs#fv#k~p>AaTWL%=Tb7Nv#mk5&QErY
z#y!Y=G?j=G5j4hGO;S_~#z+-Q702c%x5W)I@y4@F@scS+HAX%MvesrUf(xozu)Xs_
zkbEH|sw7j;h72jFkOq5BonIQ_fhWLMyvoMA>o>|m8$QU%^$JrY-Dr5$gkNn-T@!TN
z+BuyFnsS+{(s*;nQN!$Pu?m>QJH9>mYH^?>4fdK}K3SIi;&kR1G4nm$+1{@^a{1}@
ztPR)fx2DWD6^*USM`E<y=)=~_vky|1#!ds&T!KxlGweiv@m)~tFi$X&RGqB;htU2R
zx8cl6fx+e8gIlR8(nt9Oq+XsEz9c?9y*B${Y$Z<dsjhWR1uS7qgajhm%}5P4>BWS4
zk?TdEIb|n<cHYrsAg;>VtSF{omqe?gc@^l!Td7K;h!xD(ADbC~OER(f0<7g<{e3;b
zudhYZ1g{(Ng8tS8|Kx8sm;pbW(<C<4M8Z}U#9(GEl>RK4wOmi9u8Q3@+-M%LO4j1b
zfj{2oG`0>ZF;SCpan7w3AX{S2jxXZw@j_U?MfXp?{ZKq>g>WvYA?m&D@Yqq%;VCzw
z(XQ4AZ#2u9nCtuMVIw+mi1lt4_6f;a`gDz3zm=}sak(H3Q<Wyu)4!QS>D6@XY|@`N
zf-+!5c+&%)86Tr1oruR3TDXDUL~GeHwP_z|u3lI3a{FKmnXLRS_A>285gmW8h-iBK
zsPPAtZA0fAo+)FRmhH{5@n<hVukoSFhq!ZYGNVNG?`NSRE+-bB)P-8|4CBYL2i?X*
z{qE#O_j&7z2;bSqB$bCJWpbj2@DBb$x`4SI_9Cw_P8V+y_N<|()4KD7bapS0%AmEy
z@dIF56w&-8ws{Q?Czw%1F^AP0j#D|xD7c#jY7M)@_Qhh^8ez*`xYw5E2WzJLqpkS8
zvDhZ~5R&ggoG1cJQz=eHwQ?Tyths*Hdde$|8q}|g!OfQ76#JtZ?Q~Vb6FIaNj)Pf8
z+ut5^l|2yOOM|m+$hG0bcEoboaQKF+YQs~Ppc5ZXfrkAqXGhko67x6hEZJibi%n)0
z%(vt-txRYgQyQ<@lBO3AuG23rKZP36f$5wT@scZBvP~Bzb<U$o-fk_rW+71-p9rTG
z=(JzYLr}kns?ch5$-eW(+vMK&#s9RK+7|GBR1SThO$M-W|81cV`L}V9v_YDB$^Add
z$vDT?By^g4>?JfwX&P2SYpe2q2F)uc$<|KyELlf;&bn_L5{ek4g^1qeRaHPe_2<)%
zI8)dC#1;SU#G$9i#%e$}0sbgS&PVfj)nHqu`Oe`OS;NBt&brIO!i#Oj{CvPzw}!KY
zo|=#~D<(lCvXF19J-Et(#lP!1yh;GKEB*~)H?EMBysmd{1$bjKg}hCP;YVE=xYbd|
zk=Rp^KF2OYbjc&UZ4qa|9xIPu2&S9$Wo^9ismezyCwo{N^b|8HqSLL=N77=cTVTsB
ztTFw}AaodOogR}mPUS%h#;D*`+*E1aLB1GrT-?if-7Z##8D9s@oA9h`{7Fpa2j;5i
z8~?$7P<ULq@Aub5FBa*vA|=y+YNey-y1G33!Lu<9VN1M9Fk3;<Vct;~cFTvj-csAk
z$8{8jF{%B$6z4;;Qt#rts;vqj1_bq4IBby+*O5C3Yt&trmVAPIq87RznQ#;L%=?lU
zrXr^8p4Zs7{NU<9J4llRiqqDIPe4ND%TG_^+8%MS{i&l&c>#+GT<nL@4Uq;gXrhcp
z>fl+0ynQf=Z8e!@p$eL(h`lTxFe#p3efS5nGC!V${w+XMAhh1SiZWEz;M$>ii&HN;
z7}~*1usB^03Nd_e%*vVfhy^U-VnFSq_rB*C(LNOZjnulLl*QSi+9M4yL{)CXFieK7
z=M7zPw3+{t^<2X~lcA#mBq<-5{6A(D1DKWQ6*VW4HlQ6!alw|mszz2<%w_kE#H}yW
zXz^#WMlXy_^>NKlr-maUsV^=p*?JU{R*B)xxqAi0{sJE5pEvloRJI>3^k)G4v#OQj
z_*4L4)V|_eVk*Ef|9(6@R#Vmy0d_5*aa8mI`nv9<1=mEVY6~<qrVeaspCX;vJD_aV
zfBQ8<F2F~8K~nG%mn%Sq9>4xD15;7kl$*Aa%<?1ErHG&{^!fc9>y>fF=$jsMvvoaN
z>TlVrBEr6bp=0c(YVO++JXr+`V11ToNsjNsY+p<EvTGZ%<riB0=D#>jd?_pL^dxL9
zbWY8}Pa8~GV?TQf@nV}CRuh{}J$(!D(8R0}LFq~~NwF~Ugg^h<xgU_e?ob<w2e*C0
z%@gy48cVYV<TddZs}{4t^+Dbs!{UTdSY8XzvVv#<iW0BX(L9*j8R2&9mc@r{{MDJ8
z4?~-i?pZKMbOr5-h*-rcl@D;gn8?OljEXMJLbgYvQEdVq0C0+EqTE9jz14^26^`(#
zp_qP=j{w6P!b?Vu<oL9F@8JJV9sz&Q3v?tZxHNqO+rmg<h~nt%I~O?T{r)kqBEwMb
zXi1<6?VkCnUtTps2a3x9we>viC-gGZl^+hpQ@A;v^@!=_@(XPngV>Y|K@n{Od>!+b
zNz)LrkDm>GK|`DI%QLapkLqrveYV^y-FicW1Z9X@m-Jy_I-VnwEMS>7hT7{<Y+$0l
zC(<{-`t0#5aW~<=cK447n*O-L6<<1Z;L;P#F#~>2m=@aDyBfM}wFs~A;Y=mBGuF^;
zMHG4&Y#`89LtH)r0(<UTUhH2?g$oBcSgr2W`2dOV;k9vHKbA?@<L`be5GoVta_#Fj
zD|=!#&bd9i0tN(g3@Ob?iha51*R8~kE*-mPp(*vXPS&a&(2bBE1b^Box)A_sty61+
z82e1sbh4sU4vZ9!=tvG_>0IkS{BFQB#B_~08!@eTf6j3)7e<t}Q0qhLL+eF82!@Tt
z_u9Pb|Ds0W3#6Xxh>>TBDS<=sOBv0vEC+=?<AeQWJ&e&D0H$YoqTIpBn8zF#f0<xC
z3xznqRZJ@nGy5n*ZvD4@dkVORzUA(DA_}&EB#3L5xj9!Dq>~l{O=_A~FdMG8{}gfE
zNZ187Hr#AnW=H!@@UGTA2b4`Z9VeYn0yUYc0hpILwGm8$#2G3=oEyy&z|NO42v<9A
zxu)|qwcH3O0sKWs`%HIWFlDQEb%|Va$RJclrMxRuP6r>F7i;-%+|dUJQ;TuK)r{CT
zHFH8F?H45y3uK9<1cn9K!%Yyc-RC@IH_ZD2`<WK=VU;Elq{1%V;W-ne><HcV1PH_n
z^rRN_EU5);Y`y|3g`VOw&$6hkvY*-ZK49z1eUIOWz#Dx+SPL3RTh+39?{puD@GCgT
z-^hoC#-4?@4`El<=qUSlnfHN^pq|ihmxTv^tz4!N+kF|*C^2;Be@DY>*lDuY2mk?6
zMjJkB?68hw{?z94d@dv$=oinUV?!~g;WwokK_$>6QT4@{rePqxa@H+oGml9v7X7a2
z)hzU&C9)-vS3pv6JI7f1ojw)*_Miu`@0%5Ss%`THiW%qKQR7a*Nv%!xlNlE$8r3cJ
zWe-lX*tU6Yq<2LFw;`n5tuDeakwe8ad*k3QoZR=yO+=EH)AI^qbD*_y`ICo}`1@4r
ztuB*6mOZ3!v0vr_9&;HPb(V({j(uc<#6|uO(ry9W7h}_wC$&$5*MPQ~7Aa2L@GK;8
zg`eXCrB|v>p@jl{%Blq7J%@*ffZHRKv08k}m}J+J{&Iy@NHnD6WO}$;lKn9Z-cr(A
z2pgEzU*RcLPsLB-%}5O;Ey?>+f&;@OpyF)8r4yDjO~C^idrIsU#tB+5HG@nXvsnLl
ztv;-0B_mt>s1(fI1l|PY@Fgw+LdH_#EGfu6qflZ63#HZy@oe};G5c6$Z<zv(&R42?
zM#Ze=naq{R%-Y+abPoGUixQP-pP3ii=+@Dc(dX?oyj+=vDZB@q0Yz@UsRRxaO(Fdp
zPyR#d*!WHN<Hg6;mPvG)DNr{~rQml;P-Q4CjAK(<j~|h0k|^FaKUdRKtxDBbeKmw_
z?nzyLj9^_%>M)|`H6c_~HEXCG(Uz}f^?{ST=4f^9<|QuDD~VUo+y(`6fC}6hGx<yN
zu=<Lc^gV~yC)&$+!o^DtO4-<=diUn^I(+{2D6Zprvc&vy-kFUk{}aRiO~Qr*vGlJ%
z?_$~2%)pFvoXog$gx5Cf@kvp)HE9|<CZ+>?C#q}xv<Da};JwqBc$kl#;y0<8Pi1#?
z9>5xu_t9`6pOdH+L-uqEG`lvSR|jX1me&PKzc@<$0_L9s(|0^l!wXLmsqO3Tr&(!r
zl1vRMaH6{MmiBT9n5!CvKhvSpOEf;8O}BWd&WadlE#v=fX<WnR02~P^O__{o9Uc=*
zG!@4H#^A&386B%!`y4r<u@?qp_cj?pW@6uLGc+l3qkW?w_34qZ%FGJ@$-+dEwdQGv
zRI=9A=sM}v!~2}@LYNx#Cp(22%Vpi%HnUXy>k}BY2V;GM`#l{%rB7X|3q01P=;S4S
zYg?@%mt#U@tZ5|VGauOYCT!!$e;n-<O45s=OwP0ePZ`W%_Kb@q#vPWTKq*vg*}|Uh
z=}KZ&LuUVIVNH7l7A?q;xoYpE8(^kZ)!B?RxDzIZAxWs>WUlgY16tHg{OR=rg*#1K
zoZ0cVXqT=s<S<7HUX!2;-+)g4Rzlm0QI=mV&?TWrc5<ZnV#+Uw_Hc)N0kM&=uz^M=
z@^6_&r^W0~SN6^HHdZ8EmBZw?8KA*ycDQz|?ZJZ2Eu(5(9FOM&*<|kzV_RJ7M5cA<
z*8Fgq+lp0U?vE4Fy8hVw8im2Z7?M=s&T$BugzR_}r$D~&Ea~{cJ-G|T?Z<?I@j*pp
z0lq!<8-XepgS}QlrSYa?6a|HoYs1<?6A*C0f`Be19A=pWqs|Vt{6-F=s3iOQ6=f3r
ze{qyfRR0Gv`#&=u3mE|7N?~lN;r1M_&Ya9Ab;Oe0<CiKBuek{xLl@~FP~&|ug1Kpy
z@5ALriQQT}1<vnA-*9=A#9L`uIS+W;>zWNr`Z~Y8lx~Dq>j(GiS7;J8xB<_hY`k9}
zyeY~a;0IMjon0gp4uy5;>Em5|P8v|!tLlLkd8f&3>b8g4G9&Ko<^C5ULJ#2i3FwAQ
zv7Vh6J2>Z3?-J%;^V7}@LV;RT)tOQ26=<wQEOpfqquMTpXy*g5z!La?4CWx*gP``*
zZ#W-;9{zUQUJBJK-+);Mb(Qpc3F0Zf)N`fA8{ObfYt9+KVDvLX)wbnk5P}-zsuV1-
zO@41?l@d#b=1kt@4@UMYpg7bw(li8NY$DqGS{_`hx~hK@wPg*UtG4HJZT-s=tgxzL
z<#d+NpT_Er3LjBpx@4K7+=0`QRt%hQj|LXmLfmYl`ZuK6`>YOG;ez8&Whj{IMXHnV
zAmmRxCXxx>H&uEit&u?GOVoOv*g;*KGf$48<bka`*Nf0axv(JXcHRktb3kG&%{K9g
z7yXt=p1+tvrmr_J&xHiAfffE_Mt_bZO@@sRD;gDY$SSjT<AYnX5E8&B8mD|?3wafO
zB)lo=$OuY))mqe4uf+iK>)`lO;q8F0r{&}pZl=PN*5<*9!fWMvj5>??>M1f@CT$em
zE?|i3_YzMz(&qfci%h1wJbRCSnxEbtEh3@j9v;as^=h(6{=0$19rV!Yx8H?#zyb5Z
z`>3w(KgB%$Pq)HC6|l6aN4^@*XL0jZcHC(rRc0_nNz%k|20ZD5g<?;OnK)5cO+{?G
zfrIM;1X%{_2<;fk@j**nISeZzh7!Y`fteM{?JHQlg?Q9v`Oe{U?`#leCp}JqxA{qn
z4xa&oSHu%FgFBo&xy~wM&5pIFK^)W~1=uHR!D+{=B2$)PYNy+Y_bnzaRV#9*^{zhw
zF_y#s)G%$L0RK7aqs|%)y%xj5UhXJMzgmdZ^aB=my&*~h(fC;@zeicO995`>8KWzN
zM1~FBpQcN1$Ugol<54S++E<A8z=K*RHYI%Xu9E?fqqh&`rudF&jbqBI{Z^<-q-w9X
zNR&)WHl@^i-#6`=C@&S9bUkm-oqB8Kv?CH}P*KH$?^WjbA}_NMla#<z2fcv7z$4`$
zkk=m`|GvhqEB==nk6<BQ0RQ!Tlc@%$X|A9O-269|0^xu!J_3U*q<#$m(~s=BI#Ae7
z^OqOa?EQ6#SglQoL+cmCbF*cDwY8~~H7})QrhcX4hpNYh@{$MfP{fn+<I9H&FGH${
z;<w<$iielxzn}szeNMkU)QDA6=TNuf>+rzwQ_J;IF9Gstwufy|LK*d!LcH>*qCenD
z3UZ*Va%&#RV8yn_j$=a{=Z&bR$b_MXum+yQPH7w7azVWx@dkxLm8!o8yO45w=~g;5
zD@QP36i@`^j&Fowax!>DU=Jz?nYXlKc@z9ct4Yfd3fHY9!-^ZM7+`1PTVDNY9$iku
z<6~Qm233@#$Aum-ZEBrxohCLl3C+~|8JMVbm_cIY*OEGdztD}GSnQF?8pSLcR-8XM
z`oG_t;Hu%lOWh!obAWBhjDVft81$kOc_~MB++yFj72ww=a29GS1WAqu0s{_h)=702
zP5>(BlSKtNx0;ab6QMAXEq7}E<RyrcR-OOE_#SKH&0a%WlV>BoFQXD09Nr7Q)?(*f
z(FKO1^`5^jCY8ch+I8Ex%%hGUaH4jpx+@MAr3_o4e^5<B^1ubMJklA!!+Hpss^8v<
zWAh9UwDZ7Ce4`2BTKxuZbU#I?U}?lYv}u6)leYJFreO`T;r6RJX3J0SP!X)fDO~f4
zB^NPz$peDF-xJ@Od7IfB+pB;G!-UZ?0=m3!BiDqxyjh68f2p1adQFexQ;wXnZVYdO
zwUj)lm(vPJ3}cHFz~Tg#ziap*HQ+?nkMsucu^y5DGpYTOm0jo{9Wp^aWt$x+KO=S9
zJ!2B0AY|f+f|Y~u6K`7FuNV4-UxJ$oXe1TSeGT#VSx2)DegHkS&<{)$%d?#nbDrvT
z`1X&prET|FXu`dD5I#Hl<P+A_*2cK@(O^*}i;-p)*hY6_kw6H<4E;N|9MT64()Gwt
zgAnT-i58Dk{^O9&5iK^o>=6MrhC(Pkc3-U0T)pS}i;2~CpjT$@`b7$-AoPtjT4Ul(
z4V~wQ#4^A6k<JW)7BvccT7z4dmXwa-aFyeHF!_s^sp#VNjheK3dd_9jdJvEE%L-f(
zN3GU&!3JBB{)u?Dqr*s9Z%5cU2S+K?z4o~JkD~(=N`$ZfViufUZ`=4Uf8#2wn&JWI
zI-cuPZ)9?cu_f&;CU%bcxvX{DZ3We=rE0+~iB(UMn36S{<9gqDoO?I6JzbzMSm-^)
z2U2w5LRaf9ZdWKXp8#j+<DPGY6}v-e@RV68p&Y@&p<uA8yO6rrU7ggj=o)Z=pHQ;a
z_HgIeyw1rN=^*uV0J7?FSIDTy=(ZJcBxdOpAlGEoK4Mp;*KLou>+(D6jF1GXu}_a7
z8UUw3m2P$Z(D*r2T$(c1r7Kt2%?pg&b8c~(nHG(HFtFLzk@v<eZKm*8&+EKwGWAZn
zYiyFd2|yP_4>dN2s0}MQ8S8$tQ10{9@g!I0$xE9lVjv&cJ#T!qCS~nrU8dW77b82N
zfnuf|*0*@c^&t0YvLdWVeAQ7N9ESb7Rt1Cu0T%cH$CcXVA|Tg-d+YuaQ&E8M(jpbd
zMXP7*rqBe4*zCUkfe<|!lKXgey!~cAJD5lx4-0@)S?hazGyo%cIy`A`Y-82jRZ#@(
zDNQe=8}C4$7e5il#q7uHtD*<uCred#E0_9;e?cf(M2`}{^BRZ5<(a^gHYj=Dp*!q@
zb&W;Gt}~yStP2zGsHos~n8Jmop%&$_9Uhzc+*rb(z}P1ypoL+7N5G5$Kl3kRhr$Uk
zO{qukecqqX{j*-6>3nW~zMk$Upm9zjb$Jl|T9v1ni8d9oO8YPA?yk~3FT|rkEbKFN
zIsmR=5U){V2`&h$TUZkg<<nH9hix#9v|u&vG-ArVOEBd=_$#TQ^5ic;uIG+#!g-tX
z$RL%Xat6C`om%zIJ%<)7;aNe}&oM1Sh&L$&!&~v!5Si0?&ylHoypA(_?OJ4Vksrd}
z#M+Cs4A{#qix6|+Q5}j*zhS7;{u}w&u2}e+B`RYogeTfkhpB_|d?_dF>uPU)033Eo
zn+)QUozEmu5_?T!L4W7f|C2oelVCby{+1Jo-rd~{j{!W8&$|Rm$}uUU21mR<OuRQg
z09`Vg0m;LD*Kr*^WbL5TjC*nU?JGH$2Xmujz&JSuXb8(p2;Qw?5J)vTvSd$C7+has
zp2X-H>n)J2Vk1H;G8@uog{`=OrqxDIJ{)g{+fV*c;aQ3~gRVmOd`Hk{7N=`Hb7M2U
zGU@GB2%bD|uH9_F`IQvu)hYI>*(x|KHFqpO%j61`@NwiF&83vS5Okz<NV%1Mz}HPk
zlXfe0l?(B*25im@#`}|3<thT@>?^$;k@u(SSGz>Wd0IBmS-#`XSq#8B!j4l!SNhej
zZ`={3+h4w7^gWrV%_4pD>umBCMm1qd@@yk_SIOa|RLBJYSe?n`;0g*jRm6eh-0vmo
z*O<%dy|uhVyD6Bv3FGtOIKx7n(?AKZGSXLCRO(o6!X&P4Z?MzljX|#YY<blJ=gSP5
z?V4Bv(f1S{ODd)X7J2u=fdpDfU#fy&^!c8-WR?QmT=o3g-kBKbUaouiaSMgjxQ0u6
zSE*!V^%Ng5a<ljA?=C_c8!)1-Xd6$f?@<QXt!F)}Fp&FsYYC9e$Cy`@md?NbOAZX9
zP`kG?+bl2(Xe)*%wO{dCe=gjT#=YXfjabjq^Ace}8*@xK!s?0_I5=S8sk$5l0^{Vm
za^=Vv(4>-CiqUQ6Gvgn2FG=6M@sBS$q!V(}FavweE56L%W%?S&i^#;9tuw6Aw|7Q`
zOV@`oz%SCgv=zET&U?gD@9DeL=x$9(_yDkDobFg6HK)4D%9)(q3~rX<5@WLI%aHbv
zK<Pg&jR<gQ_{L>dtdvDSy2^Zh^lCTI2B2)L*UDeHT}S7Z+<R}&^>v--zR3)r{dKp4
z_QT`LZ^^_#hqy34Tr3chF9izH#;-ERQj^uI;UG~jh*INGa7e>rCB+H&T>YzstpdOA
zam}3;Tew923s;A%q<DW1H#U?AAF5Tym=HRZztsK!Plnuq=jYgD1fCRmwaU-@W!qov
z<72)9cd9=IEsBo@6NE$d2Nh`s9}fmLxO@S<9_8<|t0^lu_)m)`IPxDt?I(a9Xa3-d
z?(s2{HH*R-kQVGtAfEHS0GKI~!aDY$fp?{ze3%n<ng#>?=7L86^_QT*;s)SUWDZx+
zY-0gxlAWiX!j(J%REqTL_mhxb+!g7HGp&Ap05GQ|e45r^fdiO26~nheB){2$92N40
znWW+qc{4r0{#0<5EhcCmgzt_)w?H)|V@z7ep}0iWZrR0A1%tz4)V=NYu1Ki)K9<07
z>2V#6=BzH4YD8vwxcqi!S+Nl@GM4F2-ASPfxJS|^evn;ZhC5VrDTJ}#cTt+^&91I=
z(3R+D=bAf#iQ}((65jH%PXJ|;uG66N-Nwg_uhD-&YySgLx<anubdTsp0n1VHcWs{m
zgJv&lD_3||gHuwSQ>?agEr2agPe#%iV^gM|%hM^0dwz%2zBU-cggOpnb6M~4q@9LO
zQ+Q_gK}Cgxh)+Z&VWip8Da}vTlElaRd<*^flVLM!x>5$KZ#A3q3wBdwAGE+r9}mLx
zo6^rD+tJeXdR3G!C-1jnqBmo;EHa}UEIwD~0-ef(1<@DAPU@%R^_9B2W-;&jphf%+
zWN^~5AmST;??D^52c1@4P*CwFE6?g04U)7P_lx7n!p@&2NgplU`Z=9si@C;LigdH>
zi){6=^XF-MMa(pBXD|b#xECgUz{&(E$7##&`JpJaQM;2Iyga1*VqU(+!LZ|(mrkC2
zM|Av;*8A?y<}dN5;e@ecVW#Na8`a5QBHKJBHILbdx6HNooj>SFR30uDWlmVUGuBtf
zF3Iy3te!7@g)Sg79~;lEoC$yyvoR6o+c5hAL$Hc>-Y(DgCuvPQ@LVFFW%MR|Nb>1y
z^6T=7>^^wIJvY;`cg5B7t2n&hh(0$~cppG^m}gIw?zyKf@nFCPXanNvo%$vKlswuq
z_jK=9T6jB}50A&5+ISU8%&|)XimsV@(mbR!Wl80w%kJ}m;)8Q<Dy!srAZyYHuxVF<
z3J<=}o$SDM5B$;RZ*Nmiq||Gnd$5fQSP~1hmedZ<(W~6v`CfQx<W5_fGm>aq`b$J*
z9M>zu2`TZcb>1>O$i2xkmEGs_=pauRd3nl&B1EVj-hm)bLt<Biv`-BU80>?B4QeB)
zCujU@S=BZP0<IR8`EB}4N!**<Z`j<ldnP0Ba3}foK{?%Z^>WJYILV%_gITfl4?wbV
zC;0G{68p<fdP~KV^si|^SITh%5?Ne9&_Ebj&LNdh6+{#6;u=V;ZyG&ko!Q(FVAJ%K
zeRDA&Xt+Ou_4(tb)n3QgDZWh-3AkzE4<aLb)ry2^b26z=mYzJO_PUCd<C*Wt3*i{F
ze|$)M2y|G=>psY?TB<(WT*FUCA#*yvVLo?OWV|4w3$oq%m`YVL`1s!U`V)@oNN$!=
ztFpeXdZ#zxzz^w|>x@YVIX;ZEA3Cf3Q3?YRM&_(@D|W>FQ9$agYkO~*D*8S2Fn6#;
z`&)$3un=wIzL$vwX6_fT#enwlvL_}U$%|7<o$QfprNsxnl0o8)TCdm_#mD|v7|7(v
z#M@_cO|@dydbum|Fe;}7(C1q_+&Z@SKBD>{Qkrbt$<bB8y_{I&LhF9v5<`hl)k{ee
zq9~HZ{A`E!dkS;=%WSxbzjW@7oppL`>b3j=f>EtCAZBFgHP^kHwk;qPH7WU!H`xXu
zv0rrgazu9@cXL+H1)Jt6%|8alk^B$m(@M68^&=RUkl_02tg6SVB|Ackv=L;@Sa<4W
z#fh@F3W(p%Ut6^^P=M%?J$zSJ7FYJveN#>MBZ8dUH&wHs%)>6*QO_vaWaYx%2}4yo
zW1uvz<hfJl)ttAb|ESV?TI2+PhfsZ~@LL^eu~0go%ozJA>ye(B@@dw3oMq3nXLod`
zx8E;8MgU9hcc<td#+u$K$EQR%KntB<ZS3j{o0Wope*4tJaOEBrfOup!*KF6KdeEX~
z@0s>kpBNc-ljA^;S|NFDwV18>Sz2TAtQh=gP;`=)Bnp(wv(a0T0}y1LQt|69XSJ!`
zG%TM!OViwN&MVe9KW*J?@{RVO?0cmpYh(T{8vcy-Ne-Ue_8X~6SNjqoq&+#(41kpK
z@Qq$68~)=?0lpOgh<6?8aOhU`m<3bTrJp`1j$|2$cSqlhA4k#008LO=R*VgE7)lMS
zcx>s#GtYvK35(o`*4F+6T~}FFe`*klH6WjgjoQ*VD$p~y)*=J=AQBQJ8{(Nu@(Fji
ziqrUsV5KWZBc!WE|AE=Y#U{mON;$%YRKmxZH@s*~#rGxj4XP&)gtRj>tBbM6?P4+z
zZ*)o*hCQj@@A%X7pyM?KJm>WrE<U0-uVU%m0`%l|=44|~U{0MwVGi3BNq)IN?(V}n
z6B0lI{@<wxae};=XFeh^l-cQP2GX9X)QnfJ|9IF+F#lx;L^DZ_#1zTv5xX{HT-K}=
z-xN<ehza1KVqQv!`VbwPG}pVqP;G^RnOL?g!26~eoizN4H2gl97rW(MF1-+Ly>Z8w
zocq#VPoDr|C3tfbR(@>tpxk;B9NClCJk5a5Nb{%T&+CfyxppUjHu**pE{$6U@#*x{
zdd26>tyk-#W?uN8J(&J@SNd&hs@Lxp&%ZhOL*h2b6iLOUVC#6fGid{`VXmTOtK{qH
z;7h65%ag{t|E$isb+z=R{c2Z>e&dxc7{E@|YpBHvFk@QKg$$BRjATeZ!zL{c;(kfS
z+?QrFI14TEh_P}qHy<er+U6h=ZS3!XY}llgJ$3Fpyv6)zed2>9{n3J8t7gqc{w+@6
zOO9(zKr}T`{sLJgxhaJkdnH?)9;7Xuux?%0UI!T4X8n3~FNm#=pKTKHS?=O_yo6o<
zXiBO|W)HS5q`T|S7rc2To@KhQd#5mmpEc`=IY9aqFt2!fmD((>#B1IGg`LShT||-x
zUuo?>9ZCgMTw6~pMsK**ZAXja5&~3EX$|Mtc(SKf&LctEAg|9d@E4GB`^%i5(KI2r
zo0M7E=&bCs{Ji5wtbFSbZcKN43|vYE?<3RYa{~q#noFW;TeMWbEV;`(&^}db{2<(K
zjvKLE%)`C?(Ej1Jj$cP>84upXjXV0}!jAg!q3t_#6T0gK4Z$D6jm<hcx>yuBLQ_Tl
zGc(+v1u~IeQzv7GFr|B>mN-AJL@d`qS%4y@@cUkF!c_#xRb|sO`EG9<{zWjjjh86q
zlN-!-i3=_V3a&{~2uSF+5!NS1#-A*FRBD~W!YJD+5GNU|GIc#S*?pIJqS~&u&yf`d
zIlMtC*fq160&)W-Y{@C`*&0q1H)@{u8g49ep24MGWC(Wi;lsD3>vdk@i!wSa7N~=e
zozE4WIJsUoaG6mkpl~LK6*>FaJ@5SC9OwK3rY09Qv&Y05fVOweH36GUFWsxH4E+rJ
z)Q`7hc=E570=kjzrm@_99V=r1wtwycf+kgJp+A{$ns9YGv9!d+OgFiyPrZC&bi>1p
z3<uSbI(aT-*x1Tc1s)!Qg6!aM5R#|aTq{*e|ChO(k)8~7{iJu|lpZ+l1<iQpE==vr
zy(HYp>9GO7P8|>NvxV#BRZsh-^;uPmwsfv4Y?*u1TS3OUz-Vc@=6_(V8@a@wKZ(=^
zHE{dcL-zsQl0`tHe7D&LCmf*gJ_23Us64GdmH_16qdqrp)8{?r%D<iQzf-m<ZQ!Tp
zRB%V0!$zKB=<{T9?108!!Z9aVZ4$<y#@C=b?tNZij`D<n3rk;GnWJ@?Kxw+sf@SDv
zO_^(teGv`Pd6ARLcW*w`Sb9z}84wC@008HH>elR>D<`QVFk4#E?821hh=S@4H?RlQ
zyUb|Si@pA`S4NE$LVc!Y%!oI$eq$O>p`UllQvyc4yP#}@e_$zRfO5;!<JKdfN~@!f
zF{aPu6<lQ0>tPCBk={aUEfo<3gS1CSS?~pjSprzT_x;Ffwo6O)GEUY~t4N*RW6<#8
zY100hg6B}U%Q4Tx{mg1dVq3LcgSn;1`h5KRcuQ9fdiAI`a8SSV#d)~{BmTo%=3#;s
zzzq58uiEeTT;p_f&Kj4GR%m>$$+i<Qakg`~SrQYO=`2C@KX7qblmD!2;=#vdzzqv4
zqEq?GN*7Y*ABX6W88D~`qP!Wba_<t@xSjRgSx0ZZV$;{;%Xr22`R4WD@9$70t!D!d
z$uGpLMQ!M@<o$7pt~1DBhqNiv%Rof6m1}5$U)_wnrmUfXUoQ43CKgg;<6eRDSz4i)
z<OMDSiM2G&{Ecqnd@%%zR`csp$1pC9vbDx#4v+am6uZCT(a^jOBqKNH_5|Ow;)w<>
zU&FDEcyMu?^I_}rsExZ9rt<<!Dp6n3PZ<LcXc~NW4Hf!ncXfI#)4xYiq`D4jk5x>#
zAimDP%Z9D39TME4M7#2k-aHGt>l2BxNnsr1M*xDJ25dZ}BgPh}wK_y|j)T@0hQ;tr
z`Ms3=5#S%NQ=Fc`SWH;qu@oZyejp_4D3Vuu`#}`yh49TTxoTh5Vrw|4wAG|$&^(>W
zi5mY?m?FLz!VA7fTN+GyR}9UqLO*2opA&(uhj&x;-2Y5YO!xQ3uKmw1z*yHgf{;(t
zGMI=dW^iTP;v>nASR^EvjX@(63BoF(fmr!O9A<JiIQnm%?-)20TnlS?jYWF5mnN>y
z)9bVD-GZ@2D*DW%ViBE-959+LmT<>#iTIX2kzt9E<$@$fxYhyxxK>8lO9E;I+LA|M
z&`_!6BSE~VN8K^-ADFmjZ5Q4?7U`7f*~sYLeXpipMWXGmIp~Qj{@3Y&`;?FOySh&%
zY_bQB7}BdXL!GJzD)?BDRrq{Z?2fioyqZGvt+Jvk5vj`wwjXZBP0%x`^%$%~(s?OP
ztWd-c<~(5aj6Lr>4q77sZ78UqD5%i6-)+1Px#|s6zp5QpGnO@E5m~GVmnmfk6sqst
z{=S1v(LwZ`N(CjHlrs_m)_A3G=0eyOh5Xmvu5M3G!9U2g=u`DQ;J+Ee{7I-;gSEz)
zcEcmDTWKnz*hA@5>2w;hs)~zY2w_#h8|}%=&wFrDL>(HItT+L^cr^UcOH`K=*_8x!
zjE-~Kx3Z_yE(8Val5AP5xz#^qmj-L$zH_o3y>z@U#dFs6W#_XNK6luASmN+a`cN$#
z%$hwaw%s<K^7?~r4c1TOSC8P|PfCadSfH%Gf*$aDCDHz+$QGSIH);A{+)T$2)8PUZ
zv&V+tm;^AGHBpBX(c6RAAKXDVwF~U&9E)PmlT;9UuPPM2S-KsXGAPV9R6RF4R=(1w
z#;!LP(|Oj{-P_FzNpVrQKEis@<Alkw=xAteHi%#MKYDcy?dgp#fqfVJezrh|33>Cz
z<IsN(L=-?z7xSxMGN{WD^5>L9|Ge9)ztO4Sk8kFdea}V%yNe5pnS*{$8^N!Kw><wX
z&GIHGnYjIXr~0rqVUGewCHgR<0-`W<a5=jgdtIZX9<F+4Un5$E`e!p0ZhaL@9-{Nr
z{71<|=xuvO&~OgjFk!oaIy)jh%jogOV1gpe(4BV1=P<`d!YmCKnN$gpprj;Y=Ad#F
zOWBCr*s5wLIDAJn;=cVGN6NPbzi$&bCoeCyGVZNll6S!G>0@w3**mK4OW|u5h3YOD
zJHCW4h42NBv8A6y(te?_#@rm{XN?h?%hvjow6B*!B19zM;mJ_6%j5~}{HRpzJd`ZZ
zeuwtLp{n7<?)J@brD;!}z>d|G!O)_6P6JTf;cVlA4!4t?y`hPl+P0SD9pu;)^tHkg
z%?Z4Q%)OA5huB_QV?D<3kXvnkuaXcK_~6$+$g=_;T=B+r%_OX2f*_T6XR7FLa}=KC
znS-~3%3z>YwUi;Q&q|SHtLfv64uwy}Pj7^}Lu~UZ<SH{wg|~IVbPi9^sRN%bT|*?S
zu(|dFzfEFy^6oK3S1&9pMG&^zy!3V!7VCZ|J3gj?Qf{B%cEAq=8UF~ND99rcW)8+<
z54!!mLta4U)!#Qlfs4fybVD{&3h>hr99R`(_8N&z-wYa<5qmh0bq*>PJgl3muHP(Y
zUK1TF{$whHZ}qbV+ipn(SR=GqBFAYB)bstbPg?_HOM$(sif5X9rd*tJX~bhca}~t<
z@p8nxy@a}@hG%g(yE>am^y(U3nAU!UP06^!wQ;G(Ag)?P1#NPw2m%#+#`}_g?ahyY
z=)K|b=11s%-pubmr3l2;qNfSGo7gCi-Nk@a0MbYj@?Z#6*|Z0<tF<$_RcS#9O2@=u
zeU5%+xbVV;5`J4~wy=(nMv$D&ja;old)hHs_<H}{^{S^EY-P$!qudU+Q2Wijw<Ml>
zshI6lsX~<AWpaj2oezPPi^5qpKU383XpKXUEd4(Fc07-0_UJ-}#v6hXiL%gyUD?OR
zbclRj7~bM4s_}n?JDIR+cxGWY2Z0w>WC!@f7wQw;>JwtUVdC^i$Q`R??-)pe;Brws
zFVRQ4Q6nTa{$^NNPw3-h3Wk?(v^9N}B38LHk;ta;r~A0fVU?`d?2a-UMd1CrQKwC|
z7Q4^iLyq=arFf$BrEOcMO(e<F2~*E7N!#pj8GJwELm%xTkJw}YRo(R!mh&GoL_qw5
zsz4ldfTPx`#q|fH{VAr?+6RXG9R=9jd9`udErp4Ai3(B$X58IL6J=O!Y`5|VgYI~P
zS?!?HXpBHK{=Ed>L%7ZHAfC04PwsfQc8CW!2I>S#gw!TR#m_k2M%^98`&NQoZKR|5
zBVeZh8!Wu3PO>M;jtH+~WJkFE`1>RhfOJ6rChh~^g4k|i(FKndh8wa~CIs>Q)&j6B
z6maes3UbwsC6JAiv>f}+k>K>R9f)wg9#dghQulC26BzN8=|r=ceR}Xv(=c)sR@^jV
ziO)VmPaZYsp}JxXeYO}se&YK=cEc>sHX8)nzKf444zhJ#7go|^i47Qg9!b%Dx7P-W
zZ&r&fk(soE7>((D!mp{8m`8+L{oCB=u-LH<w$<#H#E9wspU^yZ0}Fw>y~uueKeUZP
z!E_jH%-q2ZAtLePh2xnC<P%YMsFriTBk)+$|4~%x{MBQ8Aig;TGcHbJ_M!S?g1F9Y
z%*`8MB}dtfu-w=2-l|B>P%_0ZDZp`Fs-4O``<XaXbNqVaF{u)qfAN-rM^{8I2P$}=
zt3L4>iyofS?Sj|yTmH}g#^Eh8z{#}u9ldp9;1wqNq)hLuqzu=<kG$Bt@7FR>q4J5|
z9!V}N2z0;<z2@jn*yImlhUXz7u;`J`p)m*o`eRp(GS-f?hs@7gUTK;LU=@B=GV7fn
zf;QPHUh)bnvrLdrQ;1T@VxM}X6GK=lHU=ghdT*X7-;F(farlrRk(QHJ5&t%M*#{<e
z>@cq1dEXW_aHz8vmv5gj@UjqHW0yY=LTYXCKVo4Tn)nqEKuHOSOI=%A<9!y@MEOOV
z|EAYnrc!}8d1h<(J0mHv(g%7`WQmZ7a0fV9yqCXiAe23J6zPm%!Cy+Q<V+yj(t^sD
zt2f{F&@qCr42xV^cOJ(!6lf7@J3gdeo3KmOcYDAmS6t4{#E!`MV|zJK#B}s!%dZEG
zSg|Z_U)Lj$-H4G_B=`hpLb@g6GAg%+7#gM_*n_f3WK}tGxf7p%zVvVQy@fNNtPi>D
z@ZZS|p~U$tCtGBDV`q!zJTSZ=Iq$*O*V#WlcAg_KVC8LEFShB(jKGu@d|YbKimeG#
zXe~1hSr9qX5J5w+4E0==T*Ic!CP>XyF<)zOf#6@e3TRPiYo#39-97C36P5o2;N$RX
zr0r*ZCvQD~TO`)6FEL7Zf74|9H8moiXe*V?TzSnLXZxB0JqdexIcEMQv6MZBammls
zUXzkuJGNp+{e5Bh(c5<_UIjmy-E#S>N&-O)VRkF15-r+y<vjsScUtu5cbYK}<<Zl{
z*$TgMdho!>(`C6pw!F;y=N!e}&!6Gz;N?HtnG4w!H)h7>F^1!U^?!%oF>)Y;93Q`Z
z+9;pHLsV6z&S;$pV%W?Z3S>)9(@Ktq`18Mb-de!x&Wp#W@Wfi9Wx>XBNIB))076j?
zZdpg=EODBlzPi<{?V26R3$XX}eN=X$`GK0BUX+YedLy(;%``emR+4S}QMmG*`rgbd
znHV$D9@h-8{cCNcA9X<J>oDHuh}gxy_ju*xD(;K8B;k(T!Wu2W5Y=xd?pVvFSmtN(
zvKr~*f)e5&q7d#3UX;NE*!*EmEbo_B?YPNaHrT`~YAxX}=H@W&k!mPK@S>f=MU|A6
za)wZ(iq%egXckD$?V6nj&S(hT+3PvO^CyO^-r&Z>{=4YBlgWE)VcH>B=2K${@sKIR
z9JpD_wmJ?nB|{jS(j9-)*7eu(MK}S;O83d(>Krlf(M@~eCXq*;DkBN(Y0QX}Xrinl
zVuo3GkmO9sha<ip45z&-J>1ZuYZ)ajkk*{90aK_c-Me%M_S4!Go?`XEbnc#ry?~HQ
zOEsnj82`<Q0^uL>oiJ_+@)a$|9AM9qrW>Vtb!JgYnm)@3>tW65lI^{LbP<i(l2>4F
z1#Pa{s;BbWxS9$|&hdG?Qn>g3eAMe$_t_D*8heeH{vWSL6aWS)5_T#{RxEkbe)_5>
z2XR0+6v53OI$;TdrhMT~kSGd<7t|HO8qq|IV^TWD9bt1cIYX)VHclp#h*b@U63m@?
z;(~b_Ki?^^Bh;FMHbL>A+l%Zz2h^&73e8{oI=Y-O2(ylCwz{yBo~M1g^L$>?v38j2
zJGqXwazQH8_AroH2X8oOF;@oIo$^1fx^3zz+24^enx_O_6mb9u$S~TKkn%mhYL_~O
zROmSh)Lq7_R2d!C?By}-e3?wX@8U(DtTkkm^)Nt~<Mc%+o@RODo4*cdSuQw;yC&G6
zv{4kj-4LV(&5sPs6`QU<H#B077?B2H>KdkJT8ZTtUui_eh6*twZ82YAA7d(id%P|V
z;5}Fz_k3#wE|X2RO@%(9#pc;v?J4f9)JoR60xbfYep}KS<`Np^bTG3qEcRoITrDMi
z;A0KY9E%gfh1?08`Q?btic6Yp*Lf$)A=_;VRfGatxOIT^+7Oe{QK_@<*y&N#>C~ES
z3A#R=a~ls<H5WYRQdr>-beqy36RZe>J2D~Lg8$r%{?*NtNg2#vm6s2%oG`nJAc+-(
zi0(`J5cR9s0PsM@XKN;H4w*KYSuW#IyyU6Q$&V_-_dC=R#e!ncvD3Uac~ldn6Y0lg
z=f>X2O$(|B#E6x*V;_reYpk0tBN#%PX@iwSL=_)V6G6htg=O&FzOYSYeN;j96<Pyv
z-~W&AH6>(-Da)b^^yxNY`2R7iN5nahDNeyMAk><cF)S?RmTa94nlr@p+#xJU0`j#z
zS|}ZZip`)MTr@_l0Pa{q6t)y>zD4b*{KH&vtdyTMQB(=$rL}9vR`jE1TDB<a$79`<
zI(UbIOSmhX(9$n0J5!aQIpU-B@hU1m1+o^+;?C#tD1GnU#&?4JA+Rir@g6yh44z1I
zXo05uH*=z(#{CKG)xs~#^0oIC$OiU4wtLSwC7T>t?l1x!jgxG?{TN)L0DPe8=znro
z1Ed|ovMP_)@YJK5x;l(Lr7FPon8{e|`nJl%(5H4muS~jWhg*C-s*cxQO_>tsqMni7
z>}Y`;aXxS}&X2W+!*yL=SdT^&LFFxMrY8p02Pnx|G-}F9?n_NG5e~`ecqJ`u<q<vP
zAC6e;y(%9=+Vo`>fQ1itqovMly*Zn#iyr{qWqbSSp)(Lhl!<t?MGd$kQS_d&3=w%m
zibinUxCtiwEMifj+|RW$H9}9n!jW2d$wja4;Q`VC<w0#$!gbiq9|;oPcMSn0Bn~4)
zT|RVHv$14DAL;Oe%FT1c#_vZHgt&9TYyS0PiwZWn(1t9&Ah8KSwXE5=GWG=O#F4SF
zL??xSXG&&h%8vnm-v!pyyI5rZcI9L!__4yt-xoL)s;6_o;z0AE<+L$Q@GxWygdIDa
zUpOd;S6w|41eq>D<$k?!{m>5&&QN%w3|=fWLoICwVP?*}e3FSD28C@sLQa!C(&ptp
z-6m`QOnQ;9qLLUZ|4c&9h;no6Wa}1#tse6Wywi^R5f$Jx8#qt`Va6dG$Y7#Br7xKB
zdUO)Uw}-BKzaQ{#9|-srUTo|lqC3|Gg@G)~bcg<0{tyBJ8@3GePPobD#y|o&7Dy!G
zpas_s49(8ctPKd!<TTJnfL=TWr5NMMKMX5%XI3#?Y2TR&YX?J)0_p?jZpUDk^PL6T
z<02pI561Twc~ac^aQ5Dk>mIp9n}~+Ilebf=wNG|2+(?I6{&fSh0vc2@tkdpnpec$+
zrXu)Jh!r3E@$d7234zq=)k9v5WWfBk=wsp_5m<(7p1h36ssIY_o|^}e5Go_X9d2+3
z5%r|eM&S=t6_wP(DlS394?PVUlE+SVlcl=e?AOO|kk!)6VGt0t`D@=cbX${>nRrF!
zI<UJ?JR+GyfjMJjwQjTAmf<_Q1Ltv41h~Dx)4=h*I6id<{e<k<qv)#tB+K>DK-se^
z%{XxNB*sJh;~|8UB+=GD{mH0gs(c%gWQvEXuUDA>NA)SwIWE{$tciWWLVau|mSc5j
zaC`ST@u|wgnowvQB|TDKBn0h%pC1bJP~^x*w2tIy>Q*k+<p&+N($0IddYSW0Cde9A
zsP4U@pu+1)oT~!K8i&1v$KR&$*bC_uN9GYtRs9ngwg?0ARg;55Nm!$PnCO*$L21<G
z<}z{_HK>8A4YXTOCveqWWn70%J69gS0>-ehvTSdvz-4^-^NDKVooeoJ_^?`6XpwEC
z+0P%=^hk3~dTTm;P#R@q){~ogeIZsFZAUE<1qbr9u+!~FR`<~1Hz^T4dHs0#MiAx9
zHZvM9V<w&RJ?-^SS4WyLn-xtB?M{#vo59P^8FDFnjLgWjTiPKZ(>!=61z8I2v`@Oh
zKtYmg_&dR}APJy$Zla&{993UD=%$@~&#apC>HJq+mFb69RUd9r2XQw{G3>N#R|K}6
zc7~I`2KFniFLqJ85Nb+ac(w^sZ1@~vL?++Q`FwfUpDq3P{Nd$~<yDVCsb44a&-SZU
zlZ@{8zODN}qA5u{AhG|p{|J~jJv!OWB+7caMwvP4HE9A!jxOR{{LB!v)7Ez=uWtt?
z!k{~%3;qxic~JXK>|xM<`fXP)D4><-2_Wm*-ou(rbNO>!`n&6m)Hk~gGe`SZFuxh+
zH%_8}^n3f;r-!~jboqZC6-wWh{<(Pf<G0m5)so={pI9$trmpt41~&q8zQvk89=Q8W
z8=|g7P&*)1%zm{D@AK3Zny%?|E&b!K%l%)$BJ)oeYnQR*k1x;nC#Cz(H*%xzTmec#
zHDNr;xZ&UOL^T0NsVjjMPWG-Ky-JQb;`EU`R=nM(vkw}d)37)w!*=zi*xoY;zZR>_
zMYFn7A|z(O^FMQ17hkKv*LKJ;#`yxBcd}BUte2un2F!kDyQXnAT?tBjo?&wrt_U4|
zI|pIfNucv4^im&VeZxGQ@WF=5TEFn^dq~W<`rTBWbT(w<kT!~Di$M;KUknltlLTF5
z_F>U&3gjkWjfcmGQ^E+BqaJRdp4aBh_ww9+U3|X_>@+0*2y9bMe3G2CbhvKx<xIEf
zYde7|=Spv`#ak@sN}<$o@wMZM=~V!RXz~MCDvHXEmYUf4T<fI)*nsq0p5;Lhi3V=v
z+i%bxz;4c@&*AP6h<*ZV=l@66TZcv2c3tC&#Lz>Cl)%s_CEY`JOUKYDAcAzxP>LWW
z-O`=XQX&chDyf1<N(j;+@H+>e_x+yl``v$d9Jhzc%ysQ^pKGtZ*53Z$r}J@w>k4cc
zWeKO^Xu=9Pfj-s~_0oI#ku0kVKOqityoAUms!ol~wr<sKR3IvCPWHf!U`q;O6-zp)
zqr1(2gsOVAK0Q|vDQdFdl6>bbo5Ovdo**~lLgG2aPo_3x=T#=|8!u})6kb<l9$+mN
zY2hJucXgBcXqj>W_W-_n+#8{3^P>0G;#cv)bL_MM>jj8pP2q{%VQ|ZGetWVIT=7$C
z_sy*hJeF<rx>aK^UcKKT%RNna>?1E|_|VxwVflf-+Z<=Wxkz;aq1?4tQCWirmbf-s
z<~{zTYI5$9gC3x}cgjH@rgsFw-*dMYCt9TImuA2PSapp(at;<~PP=xZwQb{PU?@IQ
z<LyNxvJRph>kS~UwE>BzUW{rO<!bSG2^JRNr}SjFF`M>7eo|H+<LPJ{%#jp%cZ{qg
z-dg?Eb(}1B*o#)7wUb$7o@R2lI|0kHM-`kfV}zR<z0>GiE7gi&h5IPt;OT9PLL%|r
z16$#elQ1+XH=0ce2r_Kd0T-4El*WCsKR?3Jr%6W#6WS1a7~m}Dx4b5OET&rJU{bD&
z(}3(j;ophU*Aim^vC;D`Q`Y&Ge_ukGb9jaCl!>E#Nm^POdZKa^*lEeuxM&AX-RGX_
zmN_+fC8KInLp`f_eD>UTWV<NdAh~JW5?kxLOJFmBiG|>4b8RpW!z{u3U!{3;zVrT$
z!#sS_KDEERN&7NGB(rHr7%-Y6Zm7WUNJmo=;q@@jl5!3<)NVh{yA%!vlvo=^ImYZK
zTV_pBkwK_0lS2lN;X$!>^@C8?XF9rsFdqzx;3oRReho%3h3W4q_idAM;}O9wJsYvE
zSyr7RH3;PsE<Ba3EwSLlz8T|K1RM`hVoKS&`0r6LV1AT3{+1;4SYo?LUvxFyHlXWP
zqC#T(9jhSS<o-8*z+8U<^pbN?u+LU)4!!<i>d^0@Bv29uB+AwH|G7nf5_a^oIjXX#
z6DVAE<gh<`k>CY_^spJAG((~m0Csm&Jjy(q+!Mn^@W|wlWYctyEZ)gz5yn5I?yYfc
z=L+Ib?yc24^xg5Bzx?KU28-Wa-mTJo8{aPsRkBv{gZPRNU{?HDfL&HE^&Gk!NIf&N
z;L+SX?7sQX+uQrQly}vuKEa{-+&C^BKda-h!+^HXzY@w6qZ6GTL1~s?K?v3I$#Jw$
z&@B#?_Yn$wm;TQZ-#>!;0hySi1%fMDo=A+j<M%eEG{Z}sdzB_(SBuf~^E=Sp)5+S9
zR|Qp`mq4|n4WHq4c}#g5b2q9%eECqft<4hquntq@5`5^()lC>dm+bx3k;}oG$J^y)
z%6qJTcAVa);I^s7d{pFDH-rlfV~D$LNc*@{a=O%o=bc&6DN)O-*joH{RLk);b;V&b
zQKx)T8BGYwX}>@Y`MTgZ3^`kJKb|~~G+j^r6Uw#Q?B$vAe5sl4)m2qBWj@8n90|+c
z6U)PKu?_>gpi0v=GjjzpzgMDn>m$~l@v<KtEr{M2c$pf|>G3U4s=M~w(X>m-n@!%&
zs<t(7DDbP()s5YEObIDaUSuSChvdmv-9p3f_RWncGO%5~`iv1qK;9(gZI{Q&VeA;=
zPTR?#@tEQK9ML0>tjYzUKi|ZuTo!t3II;aLwJQ?L<=2g6V0{FJ)NnY=+$(~h@DL=q
zpCBV79;_!3M!6TxnC<rz?ws@)dmR79Lv(#_O_u%EnwM}Oqi3css(PuJfl>VdrltZ@
za!dI`B0l#jtE0bgS<^DHSCGm4=dHq((B1wv(zi}evZ4ahW}ZEBw%S_|Ra3h^$-rB7
zJo-Dny?s=YFZWJf^w;;bi=~@%OPy<@jY6cqGW*y^`adT%*kpPQLr6z8awWYK<dyG!
z2sm%88H3a5NX<WyT9A+)+uY7L4;1^|G5`DSjz^IuMDF3IhQ%nFl9v~`On;t{!9{Ir
zW!R7)VLnkk_n9&&U2d<%PaJi584qs7H}VY5&+kZzxT*f(z1rh_Sn-Pq7ojObr}Z3M
z)252~9W)+!En(lL$nh#Sl2X2U^{J7rd{c#u%>3vIQimieTs(EfUv$K*F@TX5eR-GO
zP6V}7y0pfg?GJ#H@4_49<oL~s&fXAFI1cc)KjksH?953>xb*xz)ES@om{tP6;O>CV
zJKo>ZPcJHX2h0g{RSaz2Z52I-CS1dWIprIwR-b-m^_8}CfBbL(R#mHhmxsF7z2)Os
z>2|g_6*FjadD%NsYOWe|y7ID8s*7YZ59v9OZ>Nk#X=E72tzlakma-d<KVZR7><#M3
zvd9OU%($G?NBiDCQKoEmA>7Rc^Jnii)CCeEMyQ6pWfE(Dt6+BVw^Y5Mi+aa7_u??x
zcfN<jY)9UrC@=nZNQnJ(t{wIR4O>eI(osrXpE1CnmkcDvI?&S5T{Ch*39_f+UA>9S
z-^|6lBH`<=vhO#ywgwvZ4E262S|H*QC2){lLDprf7_E0ZX#Ck!8y$%z`#<#TKiHnH
z1l?63$@^&Cs=sUJF+VRK=C!>k-}P`X?smxBj+YOTSUV%>YWM2`^1XcE2d~ZA3#Tc4
z7_H`km@r7gD5W>8I$?S-d@5DR`l&iv_4M>{l}OR!UEQrE97Roc#-VaXHd5O*_)AWo
z8t8+Dj8AtYkPU31bob);+hJC5Hn@x9GWl+rF!Rq|^*mTL+r;j=K_i{+a|!3hSx2W!
zEJgL-J3@^+)XIWkHbOS9(uY}&Iqnmjs-c^LWr~<nrmodV&LAOsBltEUi+O2Of(~{s
zyN5eyaziiuzfX_Vo{RUj&SBwtI*eut<s;s&33{)N+-$DIVk7M{^~1})b^Rr74Dhj_
zUQ;2Jl;R$|Ah)5HAxB!cd?^);^edped8=0~_N8IkT-}7EO7=-O7OOkx$>F~Ix(Iog
zZt-<Pntv!YL65$&S$qHBbY|k~oaEkV+ufly{t1E@1P2EuN7`0pOhtS#^fH}I6_)s~
z@Bf7kW2JZ-y$<LL^9Ni+rhjAS&Lh3aY1x&FqZNrS;v<d#eNgG_@3j%9s1TLyb4t8`
zKA^Jp)y9mG4Q9}R<9U71!57qXy8l7m)|><RRdTkycR00nxWD<Op5ez~li%a<LMSll
zjN6&80q-gM`olhq9NKo(X>ymwdw-f01b?MV<Irx8{i+U3rTb3&fGuZxyM#iVL*XtT
zI}F9g>WXLKgUa|1U!z1H{^--uh~OB3iyj)p$6f(U*SNXYO&_>bUscnSJVYPbDf1FB
zU20}?Fl{2HYkAj1=RE7P?4~LNCyxB$I?gT_eiTC=s=Wb;t7;}|ypJO|5E9FudhwSV
zd7mY8w`E*2+r;Y=ib*YC+r1jQHj#E4ub_~Sk;?bUYB5n^zMD@~@aw!6s+-e-WhJ(#
zIz5Rd|8SLvzq1U#OPSXC-(6}hJ+AGvRE1G1G>@mk(^Uw<F|G+%>jM(mJ&G;=jdSX&
zg{uoI(Y<#}{jyV}-GAQFETT;nzGO38ep&7<!!;z&<9OJinX!9Zv)?Xk>CcVRrGo=j
z^?%vTHrTiG2fu26(S+7}FXl%QZM3R0&$s<5$r6w$i|cMya;RD)J9p6QCtLpVKgBSB
zO|s}9rM!z-SNrE-$95Xw^Q{c@z69Uz)58E~8#I(l9j)}2>{-g*m)U!faAI%V9po>%
z)4AWUIj?oCB66B2oUc$G2?6LJ8Wbwk3zs2Bjd@s!2E6!p?j3xF)0RUie)4-IfdoX3
zb<?QjlMt$XGUD3}MnNHSf#Lg+u2s!hw^I3Zzj|hJB7F&IQpAi_G+ynC_p{e<d^u_}
zhz`c<pjj4qlRm1{0UfNP8TM$nw2N3btr5if97XCaS?}Ubnl)Qj$8^8h&HRMnsi^dU
zZk;8`xiI+e9i<DWiT-*)wYk32%x>y6OvH!K1|Lj-+u$KY=dTVLC)IuT{fa{G&gs(H
z0if8h=gZ>4<@UlSb*Tl%yx=cm^6I>=e_+fOIy;D-{D=%Z>DvV>nFP=nxXjpya0$Jn
z+!H`a)NnGJ(bi@6`EPu%lX|SNy<EQ(D0_B-XMIStKuNWmBB~4?H(qpDbV9&-+5C+d
z=+uYU33+9NvgoT9eO?>bdRPt1ux!2vu-A}UEFgSTHG{0>_d}>;lk%%-!h!k77A+k}
zmb+M)g_9X9##3>kg@5AN0m%Ml7Q$)SCvD=+U-5*ya5e$&yYJ@goEC`1;YLc}Mqr`O
z@!D^lT{4_TwE*$<AOGdV6y_6py%-*B7An#A2hgb>y(Vn8Z7ld6Ezpcfl{^6`?;1B(
znKpA)eg)M+zm~*yip4o@2;a4oulw%`ZzXO)Ihx};`MPjf@}60R%|H#Rf%N?8;X?tH
zPW+7CN>UJH@uN2MV$<{QpP}f-9R!JMmvBp?wZAw56u67BI)%g_yD3;hAOIydB@_>-
z!ZdW=>)c!!F>Gq{&Os`cu+qKXSsr|sF`JWe*tEh4d&_4-MK(IX1B)n&)FKKdS3Qm}
ze4f$MJm|`uH1S1UT-Z%G`ma+9(m-h{c68Is%v%p41MLUQc9As}e{wO@`X9rgM1QmD
zRiv69SVFXJC;LrH{8#ZVcv-w)7~b2(QI8PsHzoTFI<;L#ro+hklHcl9XL_+bT{E@`
zQhh*+N{ik>;;>5N(aHa#Rxjheb5fEzRDgHo{BC=&sB)FRdD|m<zm4+JK5#DXOqqFf
z)y+C&p#Tkm1cZTLC$Q<Ud|#5z_oJvzgi>tBdUd0Z2i{FKRQi3$^M~4$sqmQ;2rewb
z27847i!(E+Pvr&zXaouL1=7gsvzcAG9x9WR=hBXh!8NJ@={U|jEl~VhpZ#skbL?Ra
z7ew(XWqte+$$5WJ=ceCJMAdpn=j+<9KtGUGiK5~#C_=CV0KHbaB<-EG|1=%z0!VHw
zil1*a`ou4wrHvXFw@&-TJLae6wmWvZ7lE0k^DpJh+#2#~U*<|*`g}DCy|fNrv--;o
z&b#CT;Lk`DBkNidKdJ>2e2H2X(_wv~f|KW+Ah{QFY3LF<K%$J)27BHMf+Ek<x|<!C
zU{%d*eW(AEfItMGSrBT48wFXA+usq<ckeSF<OcpOSDA<Y*{gZd#T7;X6u};McXJs2
zl5ZzwkZ&+d-DZqrYbiw+vxYfegO9%p^>dJR7h_c{s=V}CdW#~{5{0=nO(?Z%)dkhc
zJ&Wd>=_iulc57QnW0LyovYCMNY?d1hM_<uKnC7$a6DD$ABi!(4RD$UYbBX{tI|@>X
zOiHV86*Wrb*Im;PQ($*~+A{dGl|v>o66N)?Q$5XhIWsT2HYudI{if@A1KtRSN2ITK
zf*VoYvLe}{!jo*XGFMZN?X~y5<nHdmYmYTy25|Bp5kjapc<BXvL0>$7a9xK!Gs($x
z=*4tcPj3p~q-A7T2F@jq7UW;E^>Nnw)>sI{7SIhFM(NcRoC6C*VQ6skkoemsgGgw=
z6+_U;EttYPZh`EG;-At5E)-oLnl5l3U8182`zA5{s0z%1u3uT|F?mVNJ!W%$J;={u
zVVE_8jk66E=uxoas^e0BOl_v>Rp>gy{qRA!z=%F4HY(0VN|;*#Qcl|u!BWg0d7^4K
zmvFX8IrW5AHhOa3(lTfPZZH-f>mW;@uNTwK9dsYG?D))*;0(Rj{kk$jOekF&s3X=F
zbR03^u~O$jcQ0hm{eQQushNK!+)?LL+O^5GQ;CF&P2u4#8V443qUDV))Zr=QHib15
z?RrkhI1_?;GgA&8pWl6Q9^f^>Sy$k#aG9d8d!ICR(%g2K=A<-5DqgQ3!+>;RcY)y@
zH5@lTZF}J8y&Ez~!fO+wuSfRL=VN`sdZCYBtiQu!GdwmFL;r&S)=aPF#9}`902)#$
zSQ9XW9int7<+$(MGpDRetLyp1(nDQ>>zKIp$|*zYyklt6L>vBEd}7rY9iBJk-l!C8
zJ2F4kP$^*c_G)1|q!7mS1$w5$czRw~@&4RF-$rWauVxb7eJ&{xy|tEycDRG66u(5l
zXSR`#ZLQ#CNS9_?`O*>MmX?|yVoC)9tJ{+nHSwNvsa+0%R3&fy9(wo9JdpJpO7PIN
z|H4gCk;#GLQ@X|q|8-JzBr&1Br*Cc%BSZf#(ch%`2%LQML}aWQ>-$RA>UV&G?4)OQ
zbM4n51GUuhJ7GNk|K7N~9H2t6Od3}{yxa^ys=vH9>bAm$t`vV?tob4{fxvv{P@OJ1
z40O<8VQ*QlPv;_IonQU-#C`WUyU8`XO5gSHscPCsspC_72&QvfGCPSlu_XbYkA~v<
zU#rTapP2eNOMe&<5gEkR!PhL`eAugneDr)-V1n2&pxgYfX<=Sj#+(vor<Dvx&eKCZ
zmqy`kx{1<A*I{`E&Vpd>41vr_<nAWV#IxgGCiKe+gAmY$M-q24e|z+S?3M?g%Gmyx
z41-(G#oR^9*&7^bXbs@(2x(A)a5;dq$HL;b>RVw`*xg`lT;@1d)+ZTYiDn57+0f6!
zqj9T~i;ce%DGxe)o_-TqXk5To2)SmLaztPSPwVK337c=)fAB-7lCd=-#ojjloJShB
zM$7E)dCT;fs=X_;Z7TXiHtIyL7w;1ZxjotPf?AEw*(tHA*kv<mMl<6TDyxDca5!*G
zWA@u4^)bO7BjTr6r<$rTQ@h0?`&fqu)u#OGc9>F*vey7GAJD&xsOi3B?Ryn-n!HwY
zGWXA@0CWDQz4iy-YN`f^f6si~1yt~QBMa$V&@vM}ZZuthQ(j2b7CmTZT=Ey<pdV3w
z1*EtbdW-5i>4UUH_~|z?xLvX5-g26obVj@5&z!_swUm)>@sy@rc`s}B`^Dv;QWIQ8
zN-`R(X=S(7eB{^7VYi1(;dhP1HKVmvou`x!yjaM5v!43<o%_x{DoP<Q>~2*M{EnDy
znD2jD<=B1A$HvH3Gln4rjAc$5^Demm?!D|GfZW}#bn8A{ZO8o!P{C36H)9c5w9*hl
z{{0$wPAY)*?pxU(PFi>%F1FlKs;M5K?`U}_`}a-WJy|w`73L2$MafM*J?u9MJ5)xH
zCcZ<7I+XT|IykyXQ%ONe1RtyiRbvzc8uXg;dx{k$H8mzG4-%(d0ts>kX9hSlh>)Zd
zoOIYZJ0966JHbD>&q3@BX^Sng675!)5mAfB)>`zlB4t`H1vDov<V?_#b^_L3Qq>JH
zz?JKV76Gwe6?lWl-zNCJ!uqca$J`A6s*d07v!}+=Fc|Fz?U%<&|F`ColOc?I`lQm5
z%5-5y8e<bjxkw{NT<ChuW_*FNj|OR<E-EjQ42`%SVS+@&uc;wrS9E5;f+TdsE7b^B
zGnR`btJi9M3Y8)Le!w&rkX2khW39|0*%>1pgA|CJs5BHaNQG`IROs82tW60MDj)vo
zG&=s{DwB3TnkAt%tywrfMLu_jz{)#DL2#zwma2a~X`!`diLEom|GX+OjA~M!35p73
ztO|AE4?9uQ{c`yMS9no^TrDi<6ZVB411LZvY;R_~pkytE?w(}!fv%cRFM*p@lWbMc
z(vumk+b8WZ8;9R!E#2~^lPsFly>A)6icVylc{e6UO=50ZZ$QoFHDK<O?Wa+3(481e
zj)m(Z8{usb7H%GgOJypDr~5-ruS{q6x@;Tg!eDUA9H=r%STrQiq^F%vcaypgA|*PC
z#_XE$u86y#vUu>T!9RG}IA7dY?oJB%+>l!`Hzy$8-2+^O^By81avCy%)FAPm;RjUZ
z`>yvN9oCxNmaaW6a)T$hvoaUN*U*q(VV}~^&0M(|zjD?gt#9iRQXG1*o{`WK&^0!=
zo%+0-I0g|~3=Mv5s-kC%t2_5U)lmu~5ZlG!)L#F;P~@9JMNoDt!3n(eob;Orio`gO
zw+F^!^T~K;yl)d3iRr302AiU~IcFOMJ0nu@?kYOyn&Zx5@;qiN7jDDQ+)k4sIOmeC
zP1)*QJ!%U6#>g)?n&UY9Sc3y8_K|}VRrlI-T@B_&0K>w&hGgNi;n>W}Nj)yN?S>pU
zut6Jm&`&Fy<zFfU2}}06aO20!%%p*PbA^akYTaD4`@jC2PuvXvIEd*}tFP4hRkAd?
z<k#8pLbh}G<)_Hwn$kEQc5!&lu~#)kb_USM6!y)%7Zra^$^3#~iYU!B+~I{3o)JRp
z=RH09w!Ua3=Y<X9#lg2ok5l&5tw8<~S3IeSXVn}j{xOxIlZX>1tVCxZO~|hyadLS`
zFv@dzPi#*IC{Fz~DklMb)|_bHbA5%7D7@)cmO!mo%fNa1I%ESf899)e$Lp?Jc!G{C
z<@lD*%)_%+mfyow21_tUw^r8X<+Qa}i{5S(<{BssO<*hGACf0OMU-f5{VhBEGkAV%
zYijiC8{EDc8_kW>nHFZS;@;hkyZK7vzCNxp`bCy`U?vt-)|ba|(orE0m*Fp<6sx*k
zh3Ht9zyQPtg&KjnN`Yx$BAm@M1bOG<XV;m|lxh2r7uu#>byDA-_nC?jKD*~)&;Zdp
z!L}r&aMf%uV)L67)w4wkd2@)VUg%8Fb0CQw!WMeV$uS5i(61$|3bXcz8XjRz31S(4
zzf4pY{kyUdExd5W5(=ej5MD#xC+wP5j<Uc_R?`x+Sx1yO-<|8IFnWkH+OW+k88_)v
zklYVo3lnNxBWXfGU;_xOv?%yZ$9^s`We~!3KCFeiZVefWO*}i?b}2}PKkHKvl&UWQ
z$140x?gmv6O-1?7`23UCHiL1yz-h|*4c+W%Tg%1w==`JP^zzGR*Q$D@8Q0pbzO4C!
z5^4Hx)vfc?HC;z@yP(gr;W!7%V@lyUr=mK~#Xi1=-(E|9y=U6QZtgp4zZi}~2LbIQ
zn{=7b?iNrwyZ1+6;h(vgrKpki6-sh^Oh#X);0ympCI1YHp;=8>U~hgn1r>e{o$b83
zJnsiZCxs>GN+wBIHQ1^8fi6=&p~yEVy1j~SIlcU{RHz>!CUv@;+;0WuL?pGN+nAu?
z89LsB-khkBEm~hZ&kF!$E_z=xCRC4?k}TB7!pBf_qpYY8Op4qwL`fw<DUK$*exlzs
z7lw0Th5-vE9d&lOD;|wIPtFXrJ;39A_j{k_9pBE3y)=mb{PH=evzDR6Z0mTl95Nx!
z&nU^|ExGL1PHB9W$>!*E`<0&+FqaRhZ{G7f;xd4zJk~_yI<Z>muUmyuN@v;*34$-Z
zL41=rrJNP!wkz@#A$z`2SsUd61`?1?)Qf7Q^TE)CHKK7;SRdmU82X~n15YFY*Ru#V
zkrFpLjuC<B5gqDE5^LGs0s;^gFHo8qDDLirUIo7X5rJ=Tc)nrOQ&C*o)|Nsc7YlZb
z_LSRs{{GgsI4)02xLMQj(C_S5sgGn?YH40bdqvew(3*t$Tb?<^UpaT&kFt5M5yEql
z-Q9_2<QPxEM`p<lS9N1RHLf)c)1j0`AHj|B7%DD^h3IOW*J3@W%svzTvL}W=g*F%^
zN}T6CqtA-K^F3C?6I@MCFcdbhe38v^`Y364BHK@3LIZ<B`x<0M+?X$~q!AGRn?8Zh
z+o?0m@NNz?;H^lh#Q~*RH%MWITzW2{75g2$xk{Iu%3uhwVmkYKP$ZYy>G3XO6_kLG
zY&kjwE7zmZBuYW{7YopBqRtct{(xPu&DLTnU;FbaLvGf?c4yIF#mRl3RZ&pf)LEUN
zp8^t%ZZ7$nDmwJa!erI(bS*NSSUr9egNO7dgGAv2F5|P6=?tcpSBdXa(ofsy1+8Wu
zF^)-QdRXoVUj>ZAit9SQLb8s-i}C_PrE0nO-iki)h_vl48GkFgOd8Wz{N?jj(e8x0
zm>vfvH9ALsp}|m8;q+e1d9q<U3$}l%zsr}YWIY|108Q9i2HmM#M%alGi9#vp5Z=r9
zbXS%YAX9CRq`M*uLEbEhBHthDqqhjFL10sa^N&98Y~UMC4CjkHbauuCzyulfOnQGr
z@Mgg)>6&x$Un+3p;?lvkYtY4sCMpdxOwlxfInQGt+KWM0*?mZrJ2CcSH(eQYHkpe)
zgI5R=l?;qubh425b2mNqub_!jZfg{&@YEQp)~PPZCWo_Z8_;Hx_l3tg1js-4MThUr
zn_FgP-`&}wq8X<k5hKklEm4P4X0_8F6PE|^4dmDy7@XP7+?!MFigLH^u?)Gl%(8|-
zZE_gl3NOmAR{-^uJVqE~imLmpCGo5NFIG3uK0LUlK|o=f7jO342<b}N$0*72@x}j>
zBp5I>8*mWbjYICwtcU6=7gtj<1nkvB!S`<#VRKA%z$IND#Qpy1mHMp|i4;xY*@4PH
z+IrUU4}gwIBk7B#4>2BH8cHArYfhhc9EcW5L11$fZJ-!F{pEEnQ37?o0i63ay+}>D
z0*ivr%b<DNTQ*Xy^pZZBi=F-y(wP<EyZnYa-vfio%f=N^wYDZH5VBW%+YnCJgWAUx
z57<j9$d%T<@aLA=p5~&;X*OUQXTiljAgcfsrxt$@Ia9}j;oS*xsa$~Y_|iz69zG;L
ztln`Cjq<+IVNkAE{)%_F^#`aks;TSgKH<)T!;vU<R6^8ua}zp2lt&LtQIc1G_XFmP
z%tS{rh14&85PzDRmYl3tvH~)jgI-Tl5O2`%0tO%)-E5dbBqjF!mNdBkh=&ceKHabU
z6-6EpK}FBYVqLdLO`rC#)%Cc=_70^9w@E0M-4D^0S{U-FKQ&Jit6qn>bYhXm7OTX`
z>x`SO)Huf3%vw2f74%=U@Fu=pGX=s>`TvyiZ;X|#H=0`NR<ladzO6QnaSzXLs&k&T
zr3=`br*nrom%H?3ijHr$F3??q7V^d<UEtOWclz*Pye?Vit!f75=IUx5AQ|G8H9Yva
zea?hTX4U+}Ix~-9rZ20@Z3BaG|I}Z~zD^s#4A!oAa(JV-t`K|NoU})Jl+5Sre@{<w
zjKyX*DGH~P6Ax*>#TpSWh}i81cpWdq=3tk!JmXe_XI3%aWhpiBi7BpoWbrdC8|VWZ
zek1`t;s)7&`WV(#Om%NF_4`h6Um5(!<L#sq7f(9=c{749G@b(yiB*#3pX?G<WJBO4
z63)Y+Q^zxaBXc4rHG~hzy~E}nViBScL3vUdNEmQ_367~q01Wn+xguzwzqe2OFLjvT
zB2Vi#0c(EtBPJ>p|38Cw`d=v1mDi}q06y%TsF*7F`W8gV1DC^e|G}@z4=nEzn7x0E
z$VX;VZe;p{9^C$%eQlpWR4ODiBPqT8#NS`yz)_e=U!~v<&9D=qh64|A_)LI<<5}<t
z<vh;w)VJ!Jff7&jVgiitbIALESq|u?flr1I=NfsuNr(R^h_=KF?b<G<J4Wh!>P=Io
z_ZR!Y%d;~dTy*eAVN4FY{`LS+4H+(;0})a6IsO^LFGo!BH{Eg3{w15_7eU`%>rFOx
z(tF?siaWGfRa?Eh<ksnL9K=qtz9g31l}#}D4NmzY00~53(uzYu`<ELWlr+YL*5Mq$
zxkH3@?l54WLlq4HHn75{(y(E=k!+#228TlZxR0aXc!P^RC9BaDA{885dIi>VQ=_Fk
z-yJaQcCIdZuI>oWSVvY#X_aW_<yzQE#C5-~Fsy8%-rVgP0lF{irNCP6z*GM1j<UtV
z*QIbMyet9~Wv;cvSb@<DgInkkTts5aiF=NNuHfg#$XYPoNQGjl!3^q3SXGxkWvQ`;
zoa{c-?o*zaN`7si_p9z<-6!6{SKR`ht(ebNCnWe&pW!5Iuon^wFLS~)?A2fzOopI}
z`tkBRX{t_qi#F`O=2hl^5b6e6CmdcM74}=6JwvI|Fu+Xy<ZQDx+WVte=`;dN8qhvc
zgZ#<48j`uJXZ7rz4;!KjbW2lV-G(ARs#D`_v>dD-auj&OfBBS$1`wf0vGUk1I9@n+
zm`c?8!)GQYNOp<)sl?$@#6UB~=lGG9@U9d{Z3aR?hm%JA1<`cWmf%+0%@_+Rneto>
zx(>3UDA^|>^BWTP$sBD*_C8J8KW4kYng&f-jsqYl;Ygj*BJd(hRbkjNQMHltw$kBt
zrim>8Tdqc3q7?Adhf95#_Zw|PYw4tcQuY%3T2NGXljDCA0X$!~!Bsfknsxk<Tk$yw
z*9rIO>{tjz2$)RdoTmfr0j+@O*+xer+FN!LjnUnIA33wT&fWNX)BG08_n>)Pdn@;e
zx0|I}JenUP+Zj_c*^quG$Mb=;HUU3l*`j!Zf2mV-OKjCeCMmvq3uXd5yENSe|GcoZ
zS6S`tk!=5ew4y%RCCg3!Sf9II!(1uGK2quZ?gs}V0)w8C5~bOv9NPCI=eWJgvL#Fg
zaGH{NX}5V>SGr0Za|+~DU<bX~t&3*2YOEVQpuaNV%q`^)Q~?QLU6|nl8w~+bTPB3L
zJD=BmRa<i!K41aZm@cAeq{z==Ybp98CWQe8Jh&PfkT_B)l1*=s2QzfAL#*1h5vx$`
zpjx<jP3wGsNi!D`nhB-IWK9^(ULH8t`dV_RC<fl=2}t{z_1|17tNZSB@1p2lGg@rj
z={ZqMeM&5SVm%6d3<grDPj#Ly0$EBZaVYim>pYBk^K(!g@FF*s_k~$de5ZGA6^~(a
zD0tbVa|OH5eN*pSdnOxa`kLT^h6U?XkkP3anzJ;5Tj^m@&#9_)i=KDI+dFCFu<P#k
zF;&=K01{q7g^lp}ZP~##(ECMv=mI<(gXspAJM%FfcXS~ys!k-ra2}^7^q`2|-p&&Z
zJ^3LESLa-2ZuO{^j`QCg{Q&qIPh0?2URNS%s0!2y8Dpzi9yNGOU*wm0dwu-nM1O(#
zBE_o!$=7wZ>r5Yk4ORL_%mO<B(;dt7?mM3qBPid!)4}<pjzKzp(#<Rs6qrr@U+DsT
zZ{ux9Q?n2Xm&~l<XqqzAnD%4*cL0NXMRIYdB+<}b_8r4pbS>kvjzb?*XQQXKCA1#A
z)AU(uPu{6UqouVqy0vTDeeN=jtETL=0mdd(0bSwULW-y6XQk6cK~H8lF|deBdKy5Z
z@>eY+`mS$!Nx`5D67*+;s(@<dLdaeU{ei4Un0|1=9vU38Y$>xK=~Mdr_utwV6x0>?
zLCWbzO;0CNUpKYTbd2*J)1S8AFVLtV&18ElUxv(h!>*Z6`d~xb<<oDRpx-^#f)gWQ
zup%^nX^R=Iq{{x;V6<4Ee*pJXKA)&iwm0yD**xjxZnVt7ESEud^OUK2FlZ#+CFDiH
zW9%JOWE-t}F{;fP>|!69g|pmdNC)j1?Sv4-Ip{x<{_foWxchx4L?j5Xz+>;$0m7pH
zD03C$-njo!A{q<_I&`vPmLDbS0}bTS#hQ-ac*HV)cf&jtZK#T4FTk+3LVT%({oG@u
z%*8M5-aFAmWy!j}5q+ocxoR;J5q~$rLcd*XM+&^xU)=@yoE-Be@LqzF=UWtCl#}mE
z1zUz7a{9EDo?$T#E=^&k<xwtD$@iM4&3uv^-#_g8M0Q6(?~lh=E8*}wGRS<)-RL_<
zF}YqRab<t!NR5TQRj+zhGUyBFeID-W65mA;8l=TKgkhq|P(`ITq2!hCrDn{|6Fc5!
z_O9iHT7G4cB#UJ}=fb)BSVsx6i2q`pX5l0%q!t^dAdm?(${Qs|CqeRZ^CWI=R+iUg
zE11?71(u<iwhnu~z}>|DBXt+c6-VZvfH+lz=bO(c8GOMAllYdi3!7#`eA+qutmo>y
zCzYJ>6^~8e@tkX|Cz!ByG+FGhvCB1(YGg@;o*L(-1xdBAjTh?cL*WaaidQb<UWW<<
za!uT{zndoaJyOTRf!CyzhF7j+*Gy;la*s`u^$kzLT{EBL(=XT;*G+3h8^1_`KMrRv
z1VWi8|3q>OT=dNy7k6Wd7|v4lfa>8`0+`!YSBjmw3rfVP;``9_+>o<hRoaR25LFPT
zJ`P+mVt;r0Zw+vig{$Z+xe~!<7zEJbSKb3(vXCAjZAkBkw%AI_ZbIl^`8zMr(M1d0
zUAVeD?$_6rdR5=61Cc8k*XFr1z#mBv@JA?Fhro*$Jg$ev?T?*hvjzVt5?-U*v5y~+
z!d-f@&(0IMP@Y*L+7L67Rc*H55TupW;GxWk9}O5(;Gd?e)0<5+myrxVdF2AD!O_Fs
zd-^&Vez10pP5xCUWTQ8*uD@o4th%}_<@8+W<3XJr$Je;7(x9lv`AgVptqyV14TpXK
zb8p~StlWzI6uy^K)6<k-l9u`#`j&KA@}JAY$wK}tDR1lbTymj5Z^|xqY*bV}z9LDK
zaFb3z<uW|&x{VaQ02<G6!fg*A;|%Ko0@WgsW8KycftXlXEJZz9BW03%ffI^%um&WS
zauY?@?(X;vm5pVPfg}GP$O3W@J1d4}A(5~MvC+`x#XQd5lG#qRayRxu-bKfuPC%ZR
zx5JwSV{Y)AZ-R0=uf`Sff@K|`K3tG8&1Znz_KX?ngo7RW-DeuBim4Nj%Z3(_cVB%H
zzf);)QJemhE7INF{bnFX361;1_FpB7f_bMJqT5ZM4Qis2UIRLXaNQ_=GE{^XR7l~B
zFp!2;w-$@gu5163vf5TB=NCe#{jWM1srO!<3aHooWhvQeYCZm;mkRyemgaP39a7z#
zh@UKks*5*3Ux*6Rq;-f_^`PiMzHr535=Q&Qbs(5fA}eiV9hBHUh1CQCe+`6!*jvgm
z28}aibb%4pn%`ydRA2%(gU`P%5LuniO{=La%Ffji(GM`4XUY|S(HzcCNZ3-;Q79mX
zH*^C&N9b$P^uEe9A<2nx!In7;uGSm;+-xWgdX2F~s}Mr7(81V(KfI+8`O$JMHLY7k
zCiI4ksF@;s`}ZknC%3Xu6F@b_2Ls`B!Y;2Ij*Sk&=I649$FQp0Gvmz}3il8c1M<dg
zuXY(QDHc6ywSdF%;^jv84Wd}etZAnGasDA%vWI}%`tJ6K!QCjD_+8FR0L?#X2h!?3
z0l2)f-TwnmYa;YWHU=A<x2Q-kKca@bu-2T1n*}u^j8W`q*};Lj#S*N8y-lnzvC<;y
zx)dA`Zo$#VsePYpv&eM47&=uI&G`aO-zEc^6&m5i<0MPycVw!3134agQgPV!C}dJG
zpAR{XxWTXRAg_RYOC2nRWYATwCg=84?42Tg=l8`dgAWnG?45oG+@fu}Ky!bzsw8op
zM&6+h9FPSRQZXAlyKE3tS0C}~(ZS<N=qTnQPYNF8vn0V^f9+^NJrjSzt&89LPN#w#
zo(Y^HHj!<55cOo}looR4y;z^JZW#sVb;lzgo5ac49kMjH``j=`;9If4WXA2lYNYHU
z1Q9@X((nh&o%{8Y$PSNMgG%K=LGOWxffe+#W3Ina787?<kCjVZ`fa2qT2_<`%}D-V
z<0^<B0Ic6wL!Dj_6fMnFWY|CEYb@AtV$^YhLsCZ?vE-!WS2c@MzXLu*ne4M%d{@xv
zBa^2#J%QR;!qp2c<zqhwpWebn^dwFIt#Y^=BZ%-!@YdnV9ZQd(ZvA2OlVElQxN;w*
zdgcQQ`hw#xMHU4bVtSs~(=a~~+0^mIAE)(W{wgeh+dkLj*TtQLr0RIa0=WLUM3W@d
z6Ibisj0*tL=Rcj3F-8e3hIiK!xk82Ma53>TA=4m<dX5TBBSAJaJf=}l(*29*k#@Up
zg2aHYW5J3pQm3Y-M(kJKr_Q$J$?{(;(O<2<oF!591NREk>8G^TG0vgg*q(_+oLf~a
z)uj=d`nuSS^S<0amag9Zv5k%9evUe)ekb)~L3CbOX|G#uBbyv=Y=W`P;IW4^8;URB
z#7T^!iQ-kQe6H1SCLEu7aMM#2@oOu<^7LHAG&wp!NvHTzpE3}wgyPfZkd9eX%|?Xo
z93IWDLKwN8oH3Pu9oi$g8eQ}N-9g(^;lOV+Eu`{tg>M4RHTDB{ZbHtuy8mYuz<UT~
z9fwp)ZrOu!%dyE!uZtryCwL6;FZ2Xv1dwY@H~ZA~T((OOgPUS$G3*x6IImLkrIS8f
zG?Do5ODQP~>EicUQ{?dVHqgU!{-~XoR-lD^d(t)EVXNbz9OPw9MTMEhZIYb~f6PJ4
z&Vq6w%}VZT+us`E*|M<e4)NrxKqxK%XWxr`A(XLpOj0@Ohf8Wo-<Hv);l{#(y6xWw
zkq#lxg)d&fPhOis8bs~Yu5=lk1ZOa83=*(pLQaTf5B;tV{l;vanRZS6LNxcH9%#bZ
zISNehB4i723v9Py9YRh&SXJO|Zq}vtqrDza%4dtBpV~yEhM?%S2Bxh-FFfCKlpZ)m
zSuI-y{TbW^7N3JMnh)4zb}FpMYVovt7UF3h%|l&_Pf2snD-3U)fBIS!89~jenMqzp
zcMP)o$G&>}0ItYf;S|r#D?*d-O<M(|bcG&-K4uKK;&6v=iA&gfNUPreJmugL6e!_x
z#(O)~%4#$K7bYZb+`@yU0#j2Xp!pVa{GF(BXGv%&!CtyfJ9o=WFaOh-&nsx!+uv`;
zCuexn5JorZm8gcHdO??yUju3Q%z<^o=dM%;CJ3;x`|L_w530&azos3h2wyB-ov!5(
zr-l&Ff3W^BkCVk_o9}o3srbyCp43#@Wti}|*o4PpC?y#ZIxYq9A`9dvK7vGznpHfJ
z_jft*?OH}j#Q(*<HAl;;yW09Bzc=8gZ9W?3K6>^R6E^UG6jU^>N6bevw!{~Qe1czg
zRn;IkdzD6NkWC?tvM74CQYG|Qn^e-#*J?3YQJhok8IIuK5~?9sQa#7x7{s}FyULKB
zl}#nkLThCMTrQ4~m6pBDy~WS2Wk`1L*BF{8M16*B;8ko2|7&@Cb`QfZw8G(?(_z!l
z#GpI(b4#ax)l)xHy@>=Koa1sy=8`EJ43I<lKY)11pC~IUM<YoN>CYN~3~_6oQaw+u
zmkcDR2e&ZjHqucEN)3?(?wSbTq}kt{P%DNWGU=&Jk(-XQTXMaCLj~DT!SU{k=dV;d
zu;I}CU0wXUgOqA1WS~RCmgyyl_&2Vp>LI4=4EEl<@Us645+bVh$?tWZNNC^@nwVY$
zARNDSA*ACc%EudxPEXz`OBZeqJ`joNry}lQAIp(2I%XUYZ@r8=XFy=LX0H?TOaX>a
zhY-`8XQUy)2D&KA!C|1^P*NwOXIw*_W_9dO&r%G{l_DDrW8l<OS3z7^A=Ok7r?lhq
zm*(f{)8r3WQS1F^JLXEvUM9$(ld(6x^scUM)VI8N7hf;_r);erPYz-Q0P8Lo@67lY
zxSEJtzdd@&s_`abk2h!$KU1>_GHiQxrW{qw#?Y>E5)p_cS7ngMje%dG>v{FB_c#I1
zN0Qr+>SBRBq)!bu<(L@BSD_c*syrZt<1so7egw2}%SL>5lVWKcj<%%!<w^-`E!;$c
zSvOsjL-T&5SI47Q&|CpG;L0ofk1NkO2lC9-73fq9Jk;;m@!Nw|rI0$L010ztSNQ=-
z#{Bb-O&_#J&@MGmz@PVn#*zr5iJ0pmzkB^4#VoZ-SuP;e0tcPZ1)<5&8=%#HT}T}N
zU%(BZxO^E*W))5N`T0JzlGN!BzBguFnyztBv<L*<WHBN|k51lbZwFL`D?9kPL=s!j
zJKMFEi4{`R!q4&q?Bj}=tw=!^W&t%SvLFe}e?4|HMwb38=D~-PF7ht^P@8{5n-lup
zT1`YH#GPTK>bOS&hYnBF>D*Jb7)woh4Ffo(ea$UH+(k>vx^nMV5xd49=1=<U_sYwW
zF6vX{?eE_0>N4~LTI0=zLcXree7(I+v)UEnbdoybf`}qqmkl|iaS19h15!V5RoMk@
zT3h@;lUu&ueT+2Q1!ictl4O$JckTr5;G^e~|AzSjEE-K5vKx6*fy9-79w=PHG%9i*
zW8*Q3Eb)9ITQ!dbR{uBhcb_IXy)B33#yqs<Ev~a|%e5!wu%dA)IZ)U9<-cQ995h;t
zN+_hSlckA{bs#W8>jwDRA?fg$^Ya-`M<Y51nQxaf-2sqic(4Ie(K3s7L}!4)GgZ|&
z<&j1=vH+r)lCdK~aSEq>_&R^oa^v<caF~5+Mgk#ii-RT>Ortf^Z%AUE1T+#~))Qdu
z#DfUNVEo~NC?WqzS7^F!rdI}R_JSUkAw|1X6Q7g|sn}4niLnhaVh2YotSphKt8u=@
zM(fHaO&@A^$XM^QsllqiHQKAf^NGj>mJ08WL|dl~=x>%qP8vWeLvd&k#ZcpF<St+;
zh-A(t5+1uVr1XL*peRD&8HR7bs(E{74G?O0d1qfr`$DgdL!%Qy4vN}C>LgPdZ2BXp
zpMWqNO$r!hd5Hu)NY!agU*EY_EM!Xen$!-W%t<bfr&JU4O~B-Y7@Fq=FS=MVsbuE+
zIOPaBqB9UBj3(+a!2V&g%U%*~UYsHIN6{j5aA=?D@8|ud;J`pLIG(O8`n&frdmO6V
z|IDCU&46OB>E6Kl=T~NW_U3iVTRsHid_skOMcWX<$HBa(>p>3GcbIJ+OmLU~P6N87
z^ahizdBI1x(X@|L-eB*|q+fsw-S*_DwdXGIoeHqXtt8xfKZ_oYj+Rh5KbRN`S`|cW
zR=3`JLPIvKLgD?*YY2sQ`vM`an!MB6NCw$%bTGw#j|xLWK2MUe%NDZ!9N5>|dzck!
zkSnmOV14d2(2DK^$e?Y7C%>Qc&;b4-C+S?ClKC<hm{Tl7Ah$_y9p2-^+3M4XLeX}#
z8~i~#feV$JTknIPhQw>G)kh^>i$QdlF~K09?8e;*O7RNe7t5cv*{vh>)?I&lez*Fi
z8O!A{&CpZuYV%fy0ogzRmHNF~q1nB;w~g>{L{+FQ=ZwgVaQU^YVjURhI3NBT+MU=c
z-vuVya88E&#qHoOFGGZeuSpgJ2x#bhXAr%^3~>2+>3bKu+|ihUn>ml(b1YU`UAUYK
zYln~mIrk1`6-$)?C@0>A$uQ@g0;c?K-A5>M-1P?q!^rO3B=(`zD9q?h>g$~^AnVU2
zFT{b$VLlf&U`5$St{<1Ahsr&`Pn%4$JbS`@-e<p93|{N&>9=pSaFgBj9ub&PNXHps
z@Ka(ABo;&0(RyTXAyniUlu$Q1<W0)vyC6L$3KiQ!^R>@_6;C$LZv4U%c!`s0?hKsC
zCB6gQHJ-nh$35B>4c^eVObL<#W{xrYz#A%9xpqxTfU#smU@Vz~8mxq563!wU-x5(P
z$eKyzrDyzhR!=XcJPZOFB`q2}<gcK3whsm-RV^Ko$!7FA@{SLe*uLm|bJf=`<xZ}X
z%g+QtG6F_0_l{%kdU6RfiIclKf&0>nj}_vtTC4pl1gQh>5@8rJA!m%Gd1iuw>#(SV
z0ki9I(E(^EpPWUsS@%JW`D_svbUay0Ty>i2chA)&dV9(GqWrZAGYnE?i3uzhQP@|i
zDliRa2uBPlo(h>tD#{3G=)69qQek`jEQV@6Y93F4&*(MRdACI;!CPpI|FoS>m@Urj
zU>&2*U0nqMDnrY>%ar`l|4dEy&+O~h#YcNBkx^H_jf*&?S@dN6&|;{V?{bBQR%P=A
zpB=dEz2g(l#(<xG&bz#CZSm8nd+)YpgI;;h^XJb!!MNQA6@?&6U6(!o+uV@L2jdTq
z&T8iAPp!Q!GYmCh1HAcQ^TlzW9M0WhCv#v?yza^ao^iWwJOfN}KFPZ}%|jbaEAV1G
zKcR0j<KPB<bm<4u74c0Fy>JR?YD0MQVUiQ;_tLuBS;7PL=<}8Lc}V9tR(8|k?h@N{
zsfTLyv^_b#X~?L)J9*xOgi_m<(M5OYzt(y^PKB-k)}9mQHdAxUrJ@p9=k=cfp4QCJ
zH*`qD@6;H&r-s~Uv1lKt%}Uir>&WD+R=`sr1iDXmzoS`4pR6Kq5$hz2+%tcz^F(sJ
zlBW<_Pu6#fSQ2Up;By$yZsG|^wgV831`F|MSWJQ~ve#GjmvVg{8Wg$*&>lTM-E--~
z;@z(Z`D#cTa<X{#1Xv3;#pT@|$$aaoijv@M0RwQ*t;}4AU)ho;*{A{I11V)e^;^EN
zwZ=A(GEylkA*aOxI?1>le*ZURQ29Str|RmF4b5MEGS{!Q=I{C2KHRh59!Zs=)cwaG
za+&Bu9zn9u6?xsYqPk!HhR2JpZcU<ESNHG0n}I1FZ(-f@n;tke76ED~dz~_(+!r#B
z-nw!EZm$hiDbO!wxqL3Rnr=y^Kkq#zwFo@m<L*B?Z@Cpk%PfnD{g<fZK`IpE_LfkL
z%vR;`0GT$XTb=Z$6^oq!qc`-z4ayFB$h3xku_C(-yS~<J;^2>h_5;}MT!6PzHYLB9
z@KrIh#$=}eMk%I;Nl^t-7e57&=<{VbUmyMfK37lB_)ZluC*#6-`&f0joa5=M`R#og
zxT~vo33*{leZ83q!V!pv(yKTgClUUKtgC}OOD-13asM-2(ko0CCPV(tg?!5^<SmsW
zJJnY06=+JaF-3xk!e|CSCqt5fdysb`7?Z>uG+_rqA}>D`0tj(0Xg_PgKp+O9s2~ow
z;`T*Q{1Cmg3h{pp4s$=6=bB(d3rfzj)HJ7G&imgkjpKpM@1?aGcp>u}wt+s93cYWl
z;(7nae=R`k^=k%^Or8n!DN|T8Z*nM<D-h&(7SJhh@vKZ+8f{EbJzF=a1~V)o9BwH-
z?wctuh92b>6t0V0G)^65KF`SQAa>igLo2uIJ`jJ6h=e|1L%F29M3VOnT=#=0@i3}+
zNbfC7nK2LQ1aQ0SS9^Zps_oPpHvZ)#RSTty<I_#KIQq(q7Tsigrgp6&sr!Ccw0|i<
zh7LVqA^MRX+Tp_aOQDI0krs?pDcN=$%e^ZML>Zb6{(#Iw0(nlc@>jR~h-`x$t6p}D
zAH;6trEd!UuNG&F>?q?8Ska{TgQd45?lE_9@>kuv4yS^N7S)G;bbtqYwEI11f3(qa
z475v2is?47J=#_%n&k|+*C=>+cot)ro{yga(uI|wsp*}4GY3);JJ$O@C*+0V8$<X(
zCMx0WxNdVS^UDfk-QjI%QxhL9LxKrq|9amBdh3$E!$AYX7|Pxr=JFyV7)px^H=b@8
ztw*v2FY{Okjja9rEh_vLzqNVHBl-UB(ZrT3g$0X3-pS7r-DX`DWV|chKMPu<;*A}d
zNxaQ3ZOa~kkvXP{p>7h<P|YoNv>K~2AbNakJ}$WlO#=YWZCl}2;jeCFX!g<*@Xk2M
zn3f@(6wo2kE)?Vzwkyr?t(c+sU+D+NUWTj?w+ra@ZzL~uUOzl6<u}uB2w5jItcpGT
zJQXVLsOIQOb*^Ila+Iu6>S65+7uJMBw$<f-J2OM4Q>XC7Pt~Vy2Ykgg@m?yZKQW}i
zCDD<^tA5Oh@Bzvfh>$7*8I1#qKI~^V;NoU>Ut?WTUs%HZE>G9EPUrlV;(ZeKp7M6T
zo-Sv4g8NTBQfwqsg?q3E%slqUU0@D>uz&Jrsb74vYU!TTb0=bWOpD?@W~5?BK=wa!
zj)37GLd``k14(QFbEa3r%JK(9U-Q=k<4b&D0~uHV$KbwX@p?apje{8iIeI=+Z`1Qo
zJ7M^kk8oCPH7W#cm4IZC^n>XA@1vW~i)m`QjKlMNp@g@#V0sa!<^`v5<rC25RWDY_
zayS|ZWv-O$LyJYg?f{}b#Og@KS^r&MjxhN83ZMa|KX=ddA*QNKHgH1O<1R+kb6j^(
z$U1C#yi$d06_g12Vtr_-q6G>uHF=c}WTT=t(d_-Ob(=QDQMp`!fM|?m-=QofkJ}fH
z0(&~cJQX~-4IR#S-PyEI6SW?ZQ}gkXt}b1X1m9QCi_&79+FnaC@X%xb&4xyQ{2`&J
zw1=Iu4@&R9@^I^?{MkHhiLXK(`{-mbI$F$o_24T_OVRG9X(F7H2|{LXG@r;6(fa4;
zf}~G1nbn!BQ@Pf-p7H<HgP?DSe?%i6)X6QKy~jG0EZwMB`D)RXFGm}?w%5H!--Z2I
zzikJ=O%|C)JVeGjtvbR<^eQ)d_kR|T)_zD(f!+#*iu2BQ%fM&m<~B4{AcD+(*NFq%
zxz;fdKwGSuoH!BO1(QY)<!X?uS`LlSn5c$TPMRPc^Md^%6RsWyR}3jygH>x-VN#(~
zh;D;zA*q^C4wp^7^WI$uE=?jWYYy?p(cFZcLmhDo=YzSUqdyg-Kj%{6`nc|rooAWw
zK}~S#p-$*oKjR`8IumI}+X6pK4^k^1O;u$x5WQ|3AsIFW7@;MsnjXCa&GG0`3yQIU
zg=cLw*e@v+OAv%sWOxM&#goviCOw~ZETXO|h8sr`qx7#QfdOG7-G=D8f?z{^;41+>
z40ayd{Vxl#CPM=F^gepV($7P;*UjxQtBVuxKC5kZ@zQfu-x5=}N0s>f==5k1&&vGo
zQ?)clC2`OTKeH>OQcgoV6kP~!mU)PL^OUdsJm=Op+L!5CLsfTg2P{yCJq!LxmZzgP
z<66`Z_v>@1BZVE?b|vr!s#UUSguQpRfBTGT5zMe5uv|OZvz>dQ+pp-Q&5T^pPMiMs
zM)@{Cw=A6lm9A@m{{pvDa)Lnk@lYCJ@NG7;nzwBpr$Sr5H>E2saTmS!95->L>fTou
zQr2t`<!CAqO!ykY@8{42EQqeIJ=mOf%BskHt#p$gljPa#_~hz9U8Pxyven7;;i~@l
zPw~$Td3;`rh;=X4Vv#PiF#=>~fGE&ba!&Q4fxcd2u+7t0V1S7Rj#;73ZFi5&$sRo{
zNY2aq<X_x>O>FAJ*N~q!^{x(R_vL#NeEuWQVEb?9S{Ng+IV0qLM}l*m5|tQ1UPzEa
zS<%>8%sk@GGK72`zFR=K;?3wZuSF_xcrM3`nrq8`Nm%&OkK??%xWs9nAecqCEuFYo
z$3}rGsZUECdeese?|<qc70k=iu~6)Vvqdnd2>)cWn&H<7(B4!|G9*BF29@?s>WLBA
zygh#(d1&^7R%gXIbsy~iXe~%SnJ$CLHOr+#ZP%WIYf-^9@WO!c1s-jok?oQAQb|e4
z4Y+8^&&`u@oQ9_nyJ*U=%XcO4b{_&<V;}yS!-BviiO~oDyv(1Gob9*Z{I#i@eGMc>
zf4yurG<?k`BJoUxw&nt|4G<xW-9o}@=_NCnnahfC(FrB*O^eKbwCi!SE*aVX6%sZL
zb<X0ZH87BM3d^;78w@pw`_9(R9j9{ki{LJ!yF;wYkFtv=e#u)-ONHWj>{`j{)%^ke
z{`*jI7tUfX6rFe+QHKHtCfytRv&ZUByPYOBXN@JVhXp)i#poj8&LbmluMOUI4Keux
zncOYOdWdIyer$DhXtlAVKy(MKWSCk2M!`j3o~oJ|0@mLmU*-f8fQ$t50NrE0;2|L{
zehqg$Ei9p0XFyk5?ho`z+_m?<2;m|M=<cwoU#C?W%pgMZomdBP6s>Ba-LeM*Tnmq@
zU?q^^f!4BKhG;~}X@KlzBiDTez4Th{CTh$i2jS76d&PWfez?74dX%%dBt3>98JaW-
z6oaZZodz^uN=yu5r?`%*-~Q}i7y-m$O@#j#faYg6Aakr)ZlBi3>rJHazPwTE2I_}a
z(mo)6&{6%aMB{GoiT;TCFW8e(*^=Wsr+V%ceRSk&(R%W-=PwJo4!Ifo@Ne);a827P
zwIMP8C1AXMtitJ3A^4{=qR!y7$H-p%1F9Gpu<&@5d`2bhyT4ujKepZiDC+l%8dqF&
zDOq4?5tfo}5Rhi+kdm$iDG6yMrIzjv>6Y#ikVfh5RyssLK*IO4>i3=B|DFE~qs%yi
z%M<sx_uO;NMJOoz4ly5<`UQ(Rpt`RSpYnNC6$4HV5!aIm=H8PDf#EOmk|@P<SE&7%
z868z;)SzsXBUZk16D+9pY>Dqpc`;HMb4o@7KGe)co#UxBHa<5;^$I4Jo!x=bA4jVc
zw1kF5+z7v_jWz@(YsB}9!-S5z3A(EK9+MNXXlG660JQLnUo-9HSL7y6fQRc`VIL3T
z9t?n6pLL<En+(#xMo^m|2jG$4y9Bg4fJ1p|{+5*2|E3u(!D1MU-8cUGw0RJEEiAlp
z47FfCIJfM$A3MV>T@RszOVRh(%j}5BJr^E#Fheic;kT(e`bjHh>wulZ&ZIC-L`pVm
z^YQK4){OU{ZRfs8D+OR<x`pe<@1d+0q6RjCh?9~z<`cCN-IL2+`}2$Sil70c5^a>R
z6do|8laMZ8cGzUSJ7f)L5XWp|cm-k0Ja&7T4Ag_T99aw~{`wlg5udi@vIA~g<?XM%
zy=L*#Aip|)%3q^F?zOC`z5$h%&%;mtsB`|lh<|F(EF%yKXtDqC`AHD>h!8gl$U>#;
z!}C6m9;1Ar0CZ;<Q1k)B*tc(i%)ay$75?N!s$@XOVf}&~;{UP*Xagcaf1E}BbcF%@
z<t5-Wewj_2o%@OzgR;)Zhu5=qX?MT~k1jPZLLy)CIM?8XcU_$?pLF!-Xt^>7Ax>9o
zBIA}|KU5C9QiY{&Y=U8V-=*aGel&eb%qaBMwOG%tOFGf1QKSp#=#t?ENOY>a@jDW9
z@U>J4q%S)Vkf0D>6zqn3u~%WN5X}DWa-jM1HOgA85*Oeg52tWH(R($jMfo2Q;FI}T
z+>_^t2k$mZ#%Hx!sHXr$$Bval`QP^?3w)Yj*(<>abet@GMyX0~qL1#}5aISH%8`+g
ziiZt-lh~ti(_og(2neKxPijJ?jap-;kR|)g3Vh`a3H}kWMfAN+@(Hn`I_lbjY(!L>
z0afp*+|!?A0_DqRf*~H+?V^b6iBLfOau>HthLEid;@b?rv-hB+DUjoK`33iEeQj+3
zfGisKpQQXF3IO(v$HhM~{={DwKcMZLqQ-C@u|)D7%`J!YlPX{}y^2vq;KDi}lnN1g
zVuVSzPp4RB!9}8{HJc}ItVR27ryi=asY;t;6}O)`U-;9G-C^IxE2s2Mt-vPjSb2od
zFJ~LuJUoqtq4z1}j66F~loW;p3Lr}yZuf=g(9Qzxa6PyBS?u0ywQNlXC{mcSCCE{M
z8zm9;O-yva$E@w_l;N~q6J=x!_#A>_l+G{2B`c7`u6GY8Nk3F7Q%~cklC<2lSWpA+
zNXS}he5<5HNP|o*z%&QCblLeT57IkAk}`jBQa*d5u*a0%f&+Tf8$JzZguk9m;{b{_
z@paNziJ(*&Y84AG)kon=LYsi2$44$xdtL!g^tP7HZmqPpK7IfB|1GRg=iHP6@XR&6
z!8p17yL>*ME{+Mhs?8M6l!ZNcv@j6EGf?*r#{m6L6~oKv%{`1Ssq`rw3TlquhD^k7
z1RXvr$8bn@iBC~8W$|=0OW;!WRNa*`=h@Zp=kKYJBa}RC={)C)chc7)!wSt`7%J@8
zOjjgt;II|ckhoTkf^(BF1MOvoM1VXI0??EGnssD(LjWV}OpBpse=cRX6^MxrRo(8`
zD#2ve$gKN&4>4k@i7rbB*tFjVxRp)KC1fc0Xrwaq&dVlm%G3BZ-WOI6rAOrAGhoI!
z+t7~JM0HLU#WCS1xKMA)*pfX#e}iR<{<&*jc8@%7^q2LZY<7h@b})g&*H=QFQGW}z
zq6eU_!a1Sj6MN(ao?#^qc)~oR{-<8?0iTK)SB98ZwpDtyF(Q|B1jZB&cHF+2*zLM?
zC&65H!$Y(5vG7jEm&~@7s-MHPEOHsbvdSHJWfnhqOZk@sOJCU7eZ$PIRm0yFUZ~t9
zZ%Gp*1m<!`>}Mm#85dp#^eVHo)CO>Ke1XaW(<ZS{BUX%Wj2qflG`?wmzF&f7O(;kQ
zD@7~XkNum*OB18;@nZ1CXGvKlcBoxmzCwR;rawu+ZkpRHbG@sI9dU&B;fMEz1Lu^j
zkt_bj{$w-07~9MKt9r5K_u&vBcE?t`6orNhIpY!%TW<(%4QDlHnd#n4xNF7&;><%V
zy(F_zDP=qsXeH}37KM|_!E6NSQaWp`;=0y*DmwlG2EchD0o1ab*Y<AKf5U|s>NQ8@
zFkEOCu+QF#AM~GFw^9MBKlbOT=YMr1+$dDqR!f<Orc=8BpNzjT!-&(fR)ZqoO@1rZ
zP{wOml|-Ahc)tJcR){&-%joilsu*UaLN7I@30!4ONVbETvv3cZC4&N^ca68<Hu=_=
zQ^_A^wZGS}yHzoteey(P2+aZAq8Lm17Z^Z~QdR6;0t6|a-T8?a^)!D<h6561MG?(q
z=-(TA$M=?C=qo_b#TRA@WTQhW7rfWcUd9T3A#jO_$e$+@dN`TBF^1nOD_aozQ5pvi
z9YYc8XRziEOd1JAm-lvZ8m3kGeCCXdgI$tN<GWu+HmyvHGAo4wqLtrI=5zrC3uLXF
zO+}0`HlY>+L{#|7_`)*4_W?CWJ!G}_)O647;f5U`&%FCM?Vt9S6Jn2&rw<AmsWWU1
zdv+h@*)M+_2dsO@I^$^tIpLWq-}j3i&_qal;xV->_?bPNAg0G&v@)>nE)j!M=g<J*
z8JxkO4X@Tg#thSrVPhxX>;2xentW@u(9Orr#)5JipI_VbvTuJkmq8sXQHy*q^M^6V
z;R>OmeDrL@vGI+B1iwCLF5~Y(!}LWx`QZxqL$j>>oWT5ta1Tu%62T^>cGViJiO`-K
zdqo9*NMi;cgU{yd6zuXgfFsdUd=LJM)Z&9Q{*)K;?ETr3C+?3iC|@}ogl3I&CBWNd
zW)~I~IxmiuG0jADU+QDefYmJPm0{tuNW907fh`gTpp&1EdCcuXo^sKGuH(#QBj$qt
zU%(jH%}|sb{kq7r;rD<4@qUO;Nr2_%S}vJjs?G2|X6lER>ksnn5B0Gqho%X?j1(7i
zrr>W)*z0-Q?xr)nwTq`I=S3*~*btI{`xp6HiJw|Fl+vBLAk!Tm(m5p54s>b)_RF<S
zX~n)J&Rv&_=5l5%r&Hdz(gq$LU|U;Ro()8<Eb|6a_E3Q;*Q&FeurA;pgXq+vR&fSp
z{r1oo8#_XtP(B89rd?f8Wq)IFmN8fZa9Erfb#d{-_62P|{kE1%g``sVBGh<x*FUp`
zT)d|K7Zc&;o>5<b4+iBPQSZ)Ic0C0KHf3e<mo#oelKuhQ4#U$?6|A*$DMP(|O~fn|
z&T3fXyT;a;D-d111LnKIh&ONI@DY?+IWlJa(GoVMwoVy*w(^O!eOFs3dKDVX=XZP@
zWfvxold?+yFoWO`LTnB-nh5oN-2>(jL<G;hPfHa<MxQBTqCBo;qM4p^O+W~0ff@|Y
zo;o>}Ytn2%CK!JU{j+H}Hrd)P(F?ZCq!@Nvb9Fx=f}x9g;u1}H%GALE9nE{BS6!)6
z(9F2>XBfyO!p_A)ueK#U1%(<LDOkyU^an;V#d+Ft+0hfSi4IhZpz6Q9audMNT6=%1
zZ|-lM^@#&zAFbqGxu2EZnVXvWK3S))oQR1LK!44LxPE|ewLlS`q&wA#0&`rVHJh|U
zy^3C}w3Gctg+;!zA985N!Zi>_5)hdST?ZGA=lUC_tK0bjAZh%)&}{q}OG&VmlM^7-
zdHhH4>Pr&)=Q^n9gRTwo$lmz0#WiHT!tM}}L5<GALyKL4{6SIv>s$Crz{I4ri~3%a
zeU61_V(I++Q_cVmX=295>pgQqh0il-vtv~-r0&)fDZJRWkH~?gIR=>APZ&+$FDw0S
z0?XA@{V?6+Zs~8CZu7)9fOE(2OfZANPY185vSdJ`pG_6gS{mz`9R8c^=%hNg=Pe&k
zsoI-aPu7=*GYsc1TRWzgml;T?FqA;zd|8(9XZ{9gq2*5kN;Jgj0#rgqRiQ?i#l)!E
zJ*yY*)3d;KYYo#6Q~NDsCuJ)la^Su5xs!kBau&9)RG1+yBje<&HQg+!zx<9A?uk*=
zZ!c5x=%kG>-M&c5ir1=y2hb6R*wXzZ3D3@IhPA>IsJHwHRBL5+<mGbo--JfJ){Szf
z>;e}f3sQ&nRueLl2($*<y>8#qA;0f|6cYWE_Qf&JgvuGay(5(~nhI31fsqLeAT<1V
zwj=iA)*>?Qa+8^<WIN$=ISK8N=b7iqe_s^gKNjM$(Ozk))U*Cj%o}|2c!?*-Ht7mE
zxvV;=L!#ep=;O#(N(aQ(FHC}y4R~@UdB)Zl*}C%eAEV;Iu1R#MzRe_>IT2A-Hz_?f
z+kL+WN5Iwsu-d&q4kMsioa_n*OEcolr0PGIa6x}z7nG+0+W<b<9kd^TDnl%}V3W#*
zRy1SP6?Ixz>Cx6m%RfjN6U?G}n#zETtLv4Nje$(=kmHZ|paV+wjd0Sk4-eg^Rc+sY
z<S%xTm@XnU80@f3U8MK+b$mw~^|@2>_(o~6qaVGj%5%q0N$O(}8}&(1Y%?K1<?#W{
zQz?UPf*-0TSg6#&HizO^pT<Mi+sedPLDZGq`p@LCvx#D%($R1KUfVms;Y1V4i%B*~
z^0XG8ZbZ3E!`~oxy4orOP-5fGLAN~mU84(KJw2HZ0}`hjBkr967(em_j7(nyf5vr+
zRiZXqYuhVZZN>RrHnVW0S0=jkN;C>^a&7k5lwKr1J^47D`CE1vcZAFgpviTN$W;H<
zJ;XY)1<(CF;;s|Y8ZK)D1LDB*Ih<FUZ)tY6x50ocx;JZ{aEa6?+G6-Usb;A*9>5OM
zJzXFLCfI_+gEh;l;rRZq^o=b%5vf9M)wg<mxUe$~vHMA^dYW(GB<DD4{NukPuJrE5
zcWTgew=*&z0ZfrRwp=mIPV`e_b5QP0{)b1gby>)%xs<E5?ca8`2bB1|-P<oHNa`SH
z6O=;V7asg%ai+t&r6^f<g=!}SQ~Pf^eO4SZ-;jBVfo?ncmD8c66YzPO^&e}<tekGN
zC<?%zCd2sv?d3(XvJ%PYM!>mi%!kI$>tnn(y^ZSJyDi87dsXhT40*k@vZ<b{g)a%5
zY|YAdKDBN%Y2|KTNqgPc>Ee*|!83spWu5Z}{l8pD7yS0ZQW}#pnj<S7`a#ZwPCQx+
zGhMc$oL2Fzx;)y1)5%>OJ-T;};l9Q553RAM%;T|M;=Ozxl38&RVzey=xp@lV*@GBS
zJc%8tM6<^HW@^MVb?6l5PCoGHW-%%`Gcfv#FjWncEG#rC|Gf$;W*C4Rse!q>WN7$>
z06MMYog<1{5w(P-%e3b&1T6|_g709gf&D(0|NVHzHq?M;`(pO3v5^Gk)Q1o)G0Z|`
zmwc7@`iu1buEq#U3=H+yZCdP?qroXO+cV!kH4z2VP~)0(xoh1M2OTT4C%%RMAf-c!
zV>=kY=6OoLEA8jDpN7?iKMn;mgCAGtnk-)~+CEGUdNL+A)rqghV8lV!uQWrDBU;V~
z`sd6r4~&MWbj`%E_-6sa8vt0gD>ABM(J%(E;=+bzt`t+o@k1VW!qcZZT53rIc+y(d
zEZ<StTu$T3lEfjS<p~&ERskCm8XBlt0p?>SV2Ya5)t5-Vq>t}Pf++hbxC)q5<DH;`
zOVx$C6gH`FnMnyPOu+w(uF#<RH9|wsz$Z%$9Dw5Fk}#o-2%RW40)BR#xltly8E@Ay
z3?5els^Y-&k6Xehf9Rc>1(d!8MJZeZ%p9lAzW_~M4wN$%VyzQ-vzek@YpIGFZTerM
zSL$<{kP)hLU5`MudL#dTmHodnTb*!$5etXEcRX6QZa|y9S;p}U!Bm0p?rSmGJa9p+
ztYQi#y?hE2bBN}p>}{}ILa}@frzU<x0@sUI@}7NEZd62SZ6?{OUy5iUVsI4=vB%tl
z9D}DDuD_=(%4!Y;&Rx%)&lN=vH(J(~orXoO_FG4`6ju(946G~Y(5N{AKyAOKDu@6s
z$yF$gOl0lR?I+)R=+wu6GDO+|Fh3=3ziMe;5J`Fl`sy~*vit(o;2dOHLD7`Zy?*=x
zh=dz%_cT!c1wGG}k0T>pS^(cft&TkuAOAS;>BsDg9;54?$hNCvF~3U6%+yU5G-@g|
z>PmW0CBE7w7Ec$MD4N)$(@)zaGCX`I1BCXs&fUZ&tS+;-=C_>pxA-G(E?k7y_kK%%
z@!>>VTBjV3o)DOXGUEi;ESXp;6%2^Xi<=Afe)>RcjXB|`(-b|u<(I$}>F^|fRr!ni
z=jOelI(r4iJ!Y&Le6`B=CaN#+?&59pqFuA}jW|^dm5OAdL~}i&*IlA-2{l`dZoe6^
zWdKt)h)`;tFqG3;51`96Tm;%*J%HJQ=Etve?8Qd9p5Of`ZNfT?Zl;V_*_yYh8^GRs
zdMivKz?Au>siTqG{Sp0Byvkz3$8T4C*DHSisxZJloU19kNME^3U$^kyv<O9c2KKby
zEVN%X9=VNp?AJKty&VfP)M&3g`UzOu9Tikyg!zDjfv-=y1hVI2^?MBJVwoueFH3k$
zsTno{I9@8jaw^*%voCTncl0zt^rt8-5@g}w$Q*gqL_KLz&a;_U$US=M=+1(h>azf7
z0K%GSn$nzh;#K%7wz-Cus(J%yyvXT5Edi;_8j2djSa6BReXRSeqPbyW6;)3+(4rDA
zo6PO{pwn}3Jv@LO`6N0MuSE36x=eTHIm*!2`)+yayI+OxbYD*D<n*@#us`!MfY&uM
zGaT&tLC#ec;f(Fa<n5{-e<W0#dMzspe1j54zdI&=r;5^@*_rJVxtz*4-ySS95Wl-j
zdS`XF+V$eB#r1#*bwt<Gk321ofj<f_CJRNT&iyK{i(a3MMgCb3)S6I(sa~0z_v1~6
z&p3CfiSm>UuLsl@HEfWUBI)N!mo`}wpN<%MoZEeeKJ@r$Q(Kh2cT;!LU9}FXskdDa
zF=wye3uT>l9Y`EAnb;gN0pR=g>5A7xdwk(Urh8;PmIqUh>UiIQ{5TU5*#q?MzhA=@
zo&E&_>#0v91{QfL)w6{EcINcXa~HmRIiU7dkuWGDnI(gtq@s^9lj-MC!%W-f^)Z<z
z*Jt0m2zt6i(-|H<qM~}(@v6XpB#@9mHN0+ecM0vYNMgTbQl~v$@)R5dbkZ<953G%x
z6sRtrtN?s~+8$%-HLRyB@8SW;Yw1M?u6J}{C@_qnL-bzNwE=)fAN*Pr0GcrKhZ))=
zdq@|5zzd;93-LmcGubwI#M<aeSG3!faAw?_ts3(4WMX1bN(=-S15L+^>0)}}gV`1w
zHnKvfFmoV#+Pav<3$<3Y{Fh&G=p#FE;+iYJq=Yf}mGg8+8JNYC%vU|#U1!>rPkN5^
zIIcW}+nf-dJk5HC#?_~4RBc?h5HZHAn6<{)+xV^Wn|>u&jd+DzMn2r0;J{8M&iwL>
z%P&cEw9{-_u3JdJNEb7O`veaCq6Wqx1P;|D_`2`ibx`K7e4}4<4ckLghp`H5j2S@0
zb&F1kAb(n)mM05bhLdakE4b~}*1OxSunNy_%snVy?Nw@b0JXiE0|aNkYeqP)MynkG
zeacn+6(BG6+Eov`I%;pf0ED8enq7x{H;($=0p|nGywC9<cmL<h_ZwBT5n0*f&$pYw
zEevCDM(vYLX6-%Kqh*81p4Dv0AfZ+4nMbT>jMiv1p1-eFMwZqOTaJEu{Svx4ome<q
zdgkVJGY|ZDdloE&s1V(dDr$}4Sr!<12?I-12Uy?g6QW7e@*&Kyy`y#P#D>6|HNON7
z7Q6S^##Y?9`a$`_d#%IEmZqBmhBdJRR{|}|`Vx$d0YAR+X^P*D=3Y$ybfD?6jmP)3
z*484q#K?{OwafcDeI$vB>&5r9R@ZylV$8sge&A4C=UbK9TWQCekN~M>qOME?eW5%}
zNR{xkp>R5CwB2mBMC&!KqVWPvKVxxE<Foxi#jcH;`LnjrCHw^F&eTOU*e%U3A1xy&
zs56Vh{gE64{<tFBEix^M_n)LY$2Q~&pKR1rP=<jJq9VLQI)8u5EpY9UjM8a9mt}KQ
zYKb{xfXO&F8|0}C3A3`d{B=(}TKLKS7T>M{o_5ASea(V{ZIS4(l2TxpC7~P7Fbvz`
z%VqRp8<K?OYS<wKF3l4m7-?!}n>0P1&$CyvwA_lsTE6ZApO1l={QX!y&t4oxyjswI
zo(xO{&;S-$GsDM=@TGr%^_xgO?zl`totjzJOgsY<1`$W1g^p6!Qj9?fsuZBbv4n1o
zxrdHGDKH38@ph?7Iog;*2M7a+Ej9I}?0H9BLAVR^JCX+{ZZ{Hs_5+Em-Z;~P$H&JH
znax^iL}p{C_WgtM7(@R=<I@Ls;X{!WdKi4WNYOHu(s00AbRa&#j=rY*fIf1%EX=6O
z18~YkEBTfC!4{`>0AdgiHS*T4s1e=I@w(<D-m(x9mh$e(u*Tf!+YhwyWTmv#qo<W&
z`GnomS9yDpvv5__yl4?FZ92Z7+81`_fBA;$Kw!DGuu`;|dg;Ec2@aj!v#i<H6N{oK
zZO5@xi6^;F@w9o-aIW;LGnXR)1oz>aN9;d++Q=d<QjGR_vFb{&I#HTTc2r|z9dI-}
z4bFJ--rny)G|YYw=nkk}Pl(>cJ@9S<rhByNfeaI&wr-29J>TMgf}2_XtjAKbi(3c?
zkrv>JI@-XOaX-fX3S|xiyFas;A&)Hp#>4=Jxqu~EtienmC|!^CejlanWDj#r$YxXJ
z=8Kt&y2JbP_NubaO8vO$uNPh7F++DG-fY1kiSLLbwe3?ONvBB(>U<TI_|sd&2NL1q
z2DzMgGr9|{9vkyjocgi@9rBr~#DR$)(5fDBO%5+yLngkf1FyKY8`{K#`lg{_NY(~v
zOxxS3f#xcvPYSKxRacGS=Hl3G`)2G;3dY`SH=XR)622`A#SHy&!3?yG*yVYbQY14L
zsiFHN9(r!v@hi$Y{@4sP-Z#GvH!1quSQE}TpOKw)__73eeUIjm`k#1l7moylroDo4
z>>cr7rcD<;n^b?c$R4ukHuXa5g^^l<mMP;$fFv*cWcBvuN|>EWvornKoBFHF<YddE
zkveVxQbSzuXBI2pK_{J+*0OU0+*cDz<ug%qV2>LND{oX{M1|%hLsh1~BSNOx=_nqd
zGcYbMn<0TI@T)QpLC3R^np6oYkg7CHI?O(6o;&ysGF3v5T?#$ORth+rFpGa|qJ*d$
zt}<(4lrH$9K~&eoL#Dp0<^7CuTHF!RJA0aNRZ2SO(DY<_XO%kbe8CbvO0I9Uls{dL
z#0!M$VBq+Lf=khSarPRBd&Y2#`*_22tT6|c?T=9|&HS%dP)F&=FWQG)2PKilRY~YO
zI{++gJ+C}rOGCAVKC6%X*c6rsdG>Y}gp7!t85vIWTfMw~_s7t9_g`S?Q18v0&9{LN
z6~cH!GJg+~F*xsvL85Ox;+I!SZV@vj9M%~u;phSUWi?``OH(9qf!7W67#Dsq&zmjC
z2Q*op>`@yl)^cZC*rh&3OgciR2Y9IM1$U*vJf=?!6m>nm^uL9}BC2vJ?6RCj&{g%C
zV)&HS4wofGqzyZKuFAqBdM#5Gk$7*S81nKef8J{i@d2|i`(A<;svkN&N}oc9S<!7W
z!@20^pV;M+8pOkYcs^Y3y#xsP!*8sP0Y9a*1)Ij^>9~=?ETQ>2=hUr+*H0am`lS3S
zqbM^9*#6|UqgSBaoanjc?UX7FsJRrD0s<@v$G5(VryLLFQ4cLIdi5Z~il;@&<k1%8
zTN;|u%MuXz4IU3Y=)G5*G8s@lb+FLZ6lP&r%`lY(Ogh*M+_N3vhwbM4ew3t1zbnm(
zgTxTaG#XC1wRt!480)>dbY_UitDNYtGNrCY#WZl;n^!McoiW`Bd^@Nst!K*D0qNbv
z$EV0oxp@qxzhy*!;ib7@Y0AjokIu%;38TIMlHANek0q3P?PrsP&<5$jO)cb4es7n3
zw#s=RT_5K)Q&E<Al;nlc>FuVLTKFdo6I;QTRqe;gih~FzuR)OUUvy=M+o1Xp>t9>R
z4Z^2#-!}6zOOek|S4$Kj?ZcWKSI{5vIw*c3;YN4QG_=@dM`k^(eyT5Pz`!N_%m^_>
zS)S`%HUa~|(4z6}@->4R!<b5Fv=FrmHBbB$spvz5$b4PoVXyjzyFL^uaNyLCeL+r+
z?J9xphko)zEAftNG!{8Y8DT*=t`e@?9KP8r6#R*2n?mnb7pQ}BCtpX6@;OEw;GWmF
zo3G7X`hEf}+!Is3zUV!U0U(vy$Sl#u1(J<nzl}q`XI8^|IwS6y=R0kJo9D5i|0SM(
zQ*xj&1ohSWxK1dN2{4dbE&x9<!L3EW_D#2<kLhg(I&tj+Bq1I(_M@rDI+NiIB<3qe
zo<2bR{n_*45+|Jkm^bkmeUFF{H9_jIywF?4qXn?2Uvlqv`HgaKnNPW0u)R%5xJ|1!
z_hbQG1huQw=9%s^Bvl|{*u9s!-X2|cBxc8XTneymEnZB=$RZj;PMyHprU(hPe0?kj
zmYMTC-n6H$`It4joK(+#2tN|wQvI&mzO4U*W*uwHV1I>ix`v~$C()~|qAsq!#Vic{
zC1b<;io%H5#<5$UDknn=y{i@dw8lhX%;^R^y)^A<1ZUoSeTH=Bx!cKy`ft(c$)=|i
zt!9~}ZsEqf;d@T?>7PR?g}6-O@mf~wZ}QSBm}{+sF28D2xt&<@Ch|tEK0STjUVr9&
zGqiQLhDPSY_lwfAwS`OUZIy@o-^p*&Pvn_@d|I)0^DiyyA^g7D;V<cS<k{O}(N!sQ
zyeoodjuKw|Rt6U5_GtZd;`!6_6X9E@Pe&JkTU&V}(C)wSoB6oO+XIy4OKr&`82B?a
z_0b~301=UThVC)HebxnFwtcb=7;k*9TX^lcLB4QUFn4zkzx|?z{a150(A{tBU*&!O
z8<j3vvCXyv`e6f=>Xiy9w%nUvUwK~tc>#ReXd#EW3-U8#KZpjE@Yz4hu13BM>m(Kg
z8O_ys_-n0(h8TttqkSni9eRBBO<>oa=1kCfc4wV*hR6)xoqYPrWQi&#Ld}+y$ZA`_
zNGllpl1&{iyOo4}Yps*3z+`K}E5TtyWY+%m4&N7&!Qb5kPv5?4ii~wHcep>j9Htbl
z|C5MiN?)eIZR+;s!I!vspf<!7yc>YBCEQ#GnA1rWG}k5VznF>bp1=DwA)LcdOwWL#
z0fZ&Bo#xlre*~(0w&mNO4aBrf?|@=;cT53qwlRB?m>Fd%Z}8I9=z8Bmw54%9lsR++
z7(I|dzX6cCn_DWY?V@iPN-0p!S_}I@i>VIy>M(}lDj;Y~zL%1?8hFOXD6T})+nUvg
zI)}SId7XWabf~^Dft9X2o-VJ1>ayaAs><3yYRV2K%L4WLtUTTfkgbPG-l}wt2U>0j
zQA0QzyuCg!vP8kYOZTa~7z=7&2mu>P(%m}|J=jRKRIe+fX<sCdOtuXB`A8Zw46jBx
zXezp-Iy_2`D(#2-Zt}<)Fjv9m0j$Z2z}K0-Rv?~^6{fSbG2VYU_}20qhn(+=MDnLV
zV`i+i@4Dd2EhYP8z%-_Z$mMOj-rhqB!LOR4LZFLl8SLBX*Efk?kr#zhea}IfondUv
z{XI0AA|Hc-u|r*A^n-#sh(E5PQ_M`_qK4J^QSp}&4my0!4JMxcmSkjjy?k_&F_P`A
zBzia?F&p(<$%*Jsp|S0OrrkWgYvwq_ib(f0XXy|1DwdLCCL<DLv3@LjUsR3x8V9N>
zqx3T29j=28xNO$^S&oOamm4dQsP#hSDyQ}x#_$A9iHgdnw{IQsW~*Jlk;K&xarSe{
z%2~tSC08A})Mp;ds$G9PsCj8|_0z?0Mp+Ue^Uc;n*vM**Cq$H>k8g4#>L{0w=@fXJ
zIKG3x9FpD=pjjZl*M?9)t7IUgXq}JBj7cuyL@x=9_*?*D*X0j1eN{QLn9M4fmRnRs
z_7+C_4nRDhdJT5089(9+u>&B3SBX@L`160WViZjAH$7&#+omWnc&z&hk$wXR?W!gR
z<_~n<C0j!l67E&gi1Sn7sx`p=K9cW1j>3-;CC*d3F^PKlj0pP0Z^&&*owdi6`4I*S
zWZDgFFNpZ9Egm4MuxcJ{IZ_zahb{Ucs+hlA``O6K&BjFfnYJuXs3s>wcT$?G^X=e0
zH;plk{tXE;{wf~sijeCLonvrfi7kQb85ay@E!EqadcU^9$!B9lPVhA%)v~{^B43tn
z*PHapwf;p~SQ30dYic3It^a-*7H(f>N>=p~x1&b6z*Fi<MPB33PKsw;Q?bTeZ!HR1
z=vjq7g=5t~Obsn;Ar8l(w~J+Qg5-=+IblKsUg5$^`OW+`4^la&fK6L#BjYy92vTU6
ze9yhXsqE!Fm~`Bj$QuK#UEgF4mRMBJ6}3k6O~7soa<j1{7mzQS7YiyQGf-QG!pfj%
z_yG#p`SpMEL+VNikRTjpn!H^MM_{K?GfTnhN;|M(h-0LRPoTK;^%Bmi1ZXT$I-%{_
z>iknv!_$P0tR@wcW^ABba--+{Y&m69T!cXh(E;b0CLc8&0$ve#ss<GeIGqWWyjFIu
zT`-k*n4njvQC+sAuXxR8TiQ5z-!8Rov={h~rofURouIrj)Dp2o0P8jWwz%%!dX<v+
zLXp_f)1A2?67vYvuBGvwWDIusMiXl6XUMt!{+G4ZVZ7yl%r7m(`e62#jZ%b$uesu*
z;V*ku(-HwY!RWbF1%_lqR?eLUvOC;N82khhoPGnNl>&l#wA7ka_=n5t+O%UACm@72
zF;ai8bF}xHri1AawQ4eLEg^8(eDm=b-np<GP}lvvdjDJLZ@)tWaFE~JlN5gsDmKK|
z*{gf~MFYNL7)YzbE6#_-+<#+;)lz%B-mdQKja6|?dM!y}pkTu+Qbp}I-hW;I`49Kf
zjb2T6=&Nh%92HqBDiX)#=SL;7I>CCRcxG^Wr$DY%b%BWjkD2M|9P#Gsv?&&f8BByW
zEtlR}@r*0gktGQ<)s6SA)I6bYv@_7gpOEkG0aZDiWZC-p4~;k>;^Ts0|LwXgu<K!<
zeDnWyJry+W#UlkX^$;V}lMGs*-P_IKS;&dbVu4Pt&;5pBDiCf;jDScFQyIbBz`?9i
zqa`AO@gO&7=T^(8I<?vPZEYdGhU@WI8u~J=l^OP5ety}`okWgG&PugUHbLr|DwR}a
zplUf;?gosH94P4+1x~P3`*{~Da3WV(`tqYROLX=Zo9bp8j;JJy{w^a+EN}#<#fww^
zI|6*Z4^{hDL8;;o6LUsO^nun}v>;jyRe{7Ta$I37i8|Ufm?y_`dO|e7$8?D@lna`~
zDqQPD;8AY~38Tt??87rdoI}Vt&8w|t`T3AB{FxnncXF7PjzRGFqT0aQgy0C*$%Ewy
zj%;o8!K#>pm)i6O(DBJYOjYehhCKOB2qx2?`@(S-qZ3J5jQsCYkpc%(@s2gjS7T9-
z#J$9v6_^M1Y?1q^-6PhY88)3*31;XWU&mw)8cQAV0`FF_;p5rGR8-oH)mT4K`kGSd
z6qbJ$^kS6NW!dAS8;{zAH7h~Dsbq??H3F0<eh0w7w21?h(zw@Snr*g-B-759g~@GW
zy^GVWQ%LG$a|iRGg-G-cgoGV7Yg4~2$;ZF@i-i|laY^(l|K#8H{we+bet%C!jHuJs
zBd;0uH}MdsQNiWuYONboOSV3;PpXxy=~vQAPnjsKTz<$56W959c*fK)+eA0Y>St!q
zWYhpDm}1-ACM}^L5~=x=vbU|#RL(Jt5x8gB9PVz)O;Ft(C7{-57wWLPky?j~WfCdX
zQlN9cctAkyscu2xr|Kr_^ZOq@M$>nd+cU!m+XBZtjfSLmSw6-%#3YQ$Z7ZP)w@7(0
zzwbqFoz?zSabh6%AhE`%!%7ULlvPYh`5B)^xzyFp7jY5>!JRoI<Y7i<;4XP*H|{qS
zIK)G*A`FAMx^=$+mQ$M>`WG+pCwj&SQn7Zzo9uh_x2I-^);o0AxZc;x^5A(vpN{{S
zA7`x<kq(vg8>lRgiFIjgHD3yOVpsXAQ524-?edDpf*?DxzFD)%>9$YBPBV{&`;Vug
zc79M?%+LNI!w3974>54`=_=p-_Rb`mW|v>k-S%S`bcBPXc<_AC_Z}lH(?OOF3h9|x
z9SJ!DJW0f*4iVB8Xjhj0m&1Oe@i<SO44GJlY}-m1>0Kj9u*EfZ#R*=4ayc==keq#$
z$Leo2U7XNUt9-cf-_lwMBkKa6xt5<?(GW^Q50iYJ@~JE@giA@)d2B&cWn(y;vE2z!
z;p!o9zbW{g|J(Q*aUX_MRMy_0687wsV3953%U|>X)T^Q29@gGri8rq?tB`j1u^eHG
z2%P<Khl_1gqHGGO8`>VD%M==-W<1X3L|EX#pxMz}7wv!-$c8zjmP|Crh^4}9(qRsc
zj5BT2`!LjXLRgQ?9U9q^7H);nN>19XL^5EDCpSl1FYGWj+Nzl)tJJ<}*PNE<%X*4B
zd)rhh-XC2jcm5(j|NAE}XsBO;AB3yZ!0_jd;%Ed_Lf1v+>+u~+4n;z-=o0HHBV!uV
zjajtu&!c-gLJPAwto60gyIA^kg^l_wrn~T-i;-+qjgCDGqoykhR5$nSa*6MBmnijZ
z@*ZVAz+qkspDLsqyEm!*j(}hg0MH^ygT^1>%)Hno%OA1K<y_mtC|+oG8C;OU#9XPn
zU2m~7lHAdhNYiS%Q-wGztH!_o<j=b<h9CVuiHcYzeo>f&ov(`@d=W12;xHcERS?2O
z9o-v&-$YObA5DZJVAq4J6L7#j&Q`nR={I7hhp_47U~ry>vO`2u1^zSo8Az{e0>js2
zF>Y3Zm^DYTUG`t5c6Ck37qq*U_q*2SEs0&GCkHlJO<gHWc)HmW)<npfFcg$*I!AX8
z+9{sZ&$)NrbEsNI;qd^JCJT9ZDSBtA?N6uz?0{rdx8ImH_dDz8LDjT$u=X(a!~BZU
z+t@0gT6PiKQ^;9~aMjsHl5ThH(9clw<vg!4x{4>G*fBDng%SA~A4R07U9_>d>Cg|x
zbx9c6iV!Sli`~iEtgXZKc$V~=2`=PMBkQHIp4KY<w4zW-3Fd%9=YL2MBWvS0uyx&u
z0es64IF(&j`z~AmVv+P6RoP>3o{OrinH6hyY$AFXS<X%X0gIRd<8JUesb(kSrFU@%
z!{XDW)LEhsaJj6KuYIm!{Io=>#$mRT9@bJ_Z7tOCAg`Ebv$O6S#8n?tqQr8VHEuev
z>1$(f&r%Bjem-=`)~0Dx{K}*VlkX$)dysE)SQbv!sglE?y=?$26F)|G;imm^%c1VS
zaPtp}1PrKBeOLk{ffnJvFzPRRM*-HMelgv^6F`W=9fx>vV$7UR7~%0;OU?00wX(kI
zKAwVJu9d1$X}}cQ^hrLsl`ft@Qev}ZpF?A$BkN(V_Pq&Yydb7#i!;Tyx80@wadBf>
z&F<CtuhC||#yfp1YUn7lH)@PIoNpce!J6peifctj+lt<!qnI{TS&Oe0l&@lq3CpGi
za_v$OEjZ}x^IqHE*313WJ94w?P{&+at0(M6p%S}H=kh-#d2AQ>Hqkgh#ARxb)NQUc
zBvr|-{SNlbSn0-Flg?k%&EbUe8zQfkvz_jzibK_cRmeW?;rSSvE!fN9_>#GZJkFV#
z|Jp18&yrE(#cQ-C=1Oh~+_MILx^~Sr)0`Ay@pW^QJXAtf3Rp%G)M?vLpAr6>?_#3^
z&*CBrR|xF^Ngn{<G}m$5bEd|Y+Y)ofm~b`Z8ZXqtRj3P0-hX(wPCbNfLbqE``6I#C
zT<sb`@!pDkKa7(<D9~o^(UT{VrO4Uv{@i_au5CqdSMb}=GY+G{km?2MB3*sbiV+8^
zjjP}dHQY|Kf{)MPFez(H3jbwb3j+gy>>*mna;f0&>9@xhgQ|7v^4sQ&PG;@E<aMzo
z)MO1Bqlcksd=W$~=PCr6A`K6zKgu6}T_5>5r)ig`jU^|EmndAGo=j5Z!#Rn}j~0Z_
zDSqH!Dh-g$!|^qXw$i?|CzYCfQOw1^>AcBs>RMZMcidf;0+>4TVnybTZc0iB757BO
zyPN>y&S($oy2dJ%kZK}9HWVy~jdAx1jOrKvZz2hNPa+NElK;Do6hNI|3nltW*Zz=?
zMu%jh6F*6XnnEg<s(}(+0nr?$S)P_4RsJFDcF|p1_ZxrTaeR|sq#_bM^2`rGH;88%
zM!ply`x$f+7pHGMGdC~#E6X_I5ocJg-n|98JOSd;kg-VhyiqD`W%okI^eJ8Pns|E9
z=3G5~Vq0x5YGZ+wL;@>`DYAo_&e#ZmA~qCVhbW-L=v4Na+7nRAhXNSK?4ow+pC)#x
zQM>opOiA{?^k`6u3ckh%6b5eR@8LdEWI<VEoE-t%m@ajQF3T#5s1P;+ou*reZ(#Ci
z*@3!ArZF3AG>45k0N*kYC^sJaKXjTID;jvytygYAE3p$FB3q!U+*3d0P)?k;g8eqm
z$y^~`W3$oAgH(L#r=$&6U*n!VWrVooqXor%XwxI^3e#aFpu-HTD;x+t%oiG{iR{EE
z2rrDdbz7<jkYltJ#r)}C0M@|lC|QQ92asI>W)nU6Nrt-b-`!pwxF7sYNP|DAVL&+O
z{0ozhar||B`F5emN4!9eDB2#>+skt4Va0eVsE&BNL~v|@C!Cv37ybH4<vA0vZ&i6E
zen=_v`yXQ569>-F1i}hTes0sv&O|BceAZ$XlW@RztEXC3K{JYX6j?N-PQ9uWE+>pM
z7C=yZW!MU@8=ov@G-xqZj^AOV#iR&Co$mV?s67!jfm4{+Gkya^RF0g<GhZ9td#1So
zx)I_TeJ3S{?a}P@^)U8f`u_^)Pq7#;p-KTTT?Q!_g_2$+aktLbLoDJil|)4JcdDR)
z(Va{dWTjma!G}2_hOo|Y-Pv>n$Ws;X5j=QrUtNUEW1OWe7b{k51;qQ{Jwc@5sjZ8d
zASRnM#N@ckQG55r0_Cx>Kmx-l!t*gB=8dissVxl4R#_X95jgbs{~UTMAa^$NE(&e{
zn$6i_i>&#1X3^I5HgjtFE>kahfSP{IU=h%M^*j+d4p|4NB0Ydm)p}(-k`Neoxu({v
z^=B*p_r@ek0cjXftZ4<ng?JDe<+X3xgEH8mbLo7VqYY&w=d)p028J-`nJgI`i2P~(
zE4_ACKL$)tT8s^-O-=xdi|?;_79whC{v;YyXbQB^hq+dCUma`aQp`$s?BWn~d3=j9
zUDPB*ovj{x%{{iDq_J_jnh7PRSgZqmGr2(*4^n8Lr#~n%+Zg?;!jv<`nY>4rZM=(j
zDP5lNi?7YKa5duTFXIV31r`VN$7p9wbrU9yX^idGGG!txkyQm8zg(EJaZUQTar8_{
zH6=dCF2pbj@oS<&gReM@O1u&KdG*w*CRu<uB4W`#`n|QI=Vf72(!QR?QNxm}p2yHC
zVBawbyI<oESa%eoAht!d3!tSy^k&gP=J@@V(39uZ%>RD7K6sWSbo?Im2YSq6>hUL?
z!(UTuH_ji)tvnaB@(-(W{_NbLjVY$j%LwLo1W|t>8*1cRlm&otOm6t&J#-8PfAFA$
zvIN`&Q!0RF#<`ri`(_f9jnTGgNm~+itgXerV@=`3b|@mK_P*A-`=hN1Xj6+AIoTY>
za$+IU<T`dOLx<^~g33<rhXLo8OK1dzv0eZ^x8sL9J?!39EzWH>2d`hxP}n|jNcsj)
zR}2q_=N%8@GOzP))ZR1NwOyKHbm|0X{6%@}SwLz~M_XMiCCp1Q!K0CG%{;45e#G+Z
zrJga@={{`G?li1V_s4|0emvzeNfcbjmzx``2g4)v5*tOjDf^l<In*Wcz<EcD&&5fs
zF$ChgOpC*V3sn0YN-6V>iZ=MCi4Nxp<@VBsAT2P=T-c1mAXzsTNil|`f)>J08^|Aa
zQTc;CSQf(8RrCHmc0+Qj$2k$;L}lgx2*2FE1oyal0f?8Ix|+g!_ub+X!+OBe$dA)r
zsTJ@X*t9Jq_*b1(YM?Cv=v{sIgO|tNqc-^riVe>dLr84`_1{5>RSJmljKS=f(~`vg
zhvg}LB`+`~0ub7#kdQS@vSRv1zEWfz1Gk1Q%%JDg8P4>rk-`$Nu)!+8YIl5`pimN`
zPTt2F^6grAzHYf(rhojk)lIF(z?Gt2zKqE?!~OQ~@zlX&|358=Q)bldeCrroJJa^F
zt_RS)S+{GfCoP$@d?W7%x;%Z$j~Ygi<)VNF&;asS#B+9rz;30@t4gvs{a>*E7bbnc
z&X8!T{Z9@y4*)x+ohk!MEeZk0?zWilBg!}XvsK2TXM9)ZM-4Aq@B^?U)bIT*M5PFd
zTB7Q8^Am8{$5IobhdccO6Qa*Njy=Lb-?djOhhx@#{idSXpE*zr9VyN=^Ny89>c~*;
z*N8x@DbUYV5;qH5EGK&%v@vo6T~T?JcU9=Om|G!HgaGk~!R-6$Cf-HI&r_fbf#Ys%
ztdxzwe2&t9gQ_y_X-!?vw9b>K0G@(rMi*|8^#-_jw{*b8>m7M0o{XL}hWyg(vX&^>
z5)u-kGZbPk9~ye-N4{5a{iR}-y#gs^abD+duNa+)AQ-uniP<_M3kYiL_b|S}M<2MU
zB+R(vxDAgZj{=ZQA+y=XO&Eo}p<eaC)Et6d?f}CQGr5I@9FAP40)}ugj!!a^G1nBF
z!6adV(?8Y<NY|P7gnWh1@-I&KT!lDQi9_(j?jO$(nNQ#Ka?<zC-0h6<78+G-qK`4%
z+}OiZhv-twIF<{G5(g1THMjO&Z$Im$&59%@sDK-t`8=zv^b2Clv7ablRjtB2x_*zl
zu;qR$LW6cyxqQWRJuwcER0Lva1f{nDB-Y42TkSP>5*3^SV3t_`MtLymEOY3j4$l2r
zKO<_6wYI_nlk_h>%e$ou>$p<h4i6XDpLgG_&|i3lHF!sh=Yahu;)-<3;m7OrOc&vk
zZ#R9ye41Wz0=zEWx3($4Tre}f4w8Q8V?UPM05#ISiHvKnnL`_*b;#Yo?OdLLRdy?-
zyy3Wkw9!!4N$5{Si~ZzU0vfi_q7tmGYtP)sOV}OY{|77(O{%$nSO#}1W0_;BmIf`<
z_Z+Q}+(YQ<DFj<pvSKRS=QD!C)|-aI+{)o=D*_d#EV?0_?p4pddoQnoRyKfEmuG!R
z@F#hoh|`yA4%=?I?dT1<K2D5#F4EKpbR;ZXy4Z(MxsA7#u)y9TmDYEdJV(a82M?h^
z4=KpL>ZQB91|u@ADsN&wnhhkPSwGv+*%dXaxr1Yj{(D--=>0od+*Q_LHwA5+3w)oE
zMg%iZfE(70)TcjdNd3jR_CmrHv7Hlr{f<8IODr`o4fTS+^xUonZNB1ZX<%;du}7rQ
zo8u2GA&&};tsjNBBG#}5DE*>h6I=L0-+&%%tP9GYs)wGNajxh?6DLZ#Gm(Lezk6L(
zUtNLP*QDk?nhjP&jYJo{WP5xwDTS!r!0vd%5Er-6W*|qb=3RLF+RIa@_s89!)rdz!
z<WKERoeiD5NJaX=jps`(1FyC2UrM{Ywg!NhO03_lfW6FQ_mAuBOQ45Ux%a_Tw6rCJ
zZ4>w0om}M6i~n0L{rhHB8>j?aV_E=P7cBh25iRY^8Z|$A&J}l{{k=j~{9h{d3Ibq#
z#xa@_y<HDyYCn3`7Y(ymSS4UtlO@*Gl2xC2ooAe7?80zD>=L+Z5V~92BmV7gDaun*
zLa7a89nd4{Jhks`RpXC8IBQH&<8tuQB`v0zRnmn`kR^xgEx9LK<Jg8Fxl8@@<zT#0
zx)`Qxe90AM;j7Qqq(6vpiz^%fvG+z2b>K4z4{aOYtr9<<97TR<^SYW>vL{sgmzeub
zQI8TKv9}L%H`xT$n~bClJLRF{*ofeg?Q#S7PwXn`arZ<iUd4neSN3_GNU;T($|_CH
zT@#$uZvHb^XF-8BE+t!zQ<Dr0CS$PLG=X_!=QeKk7qZU8C`y+{ahanaQ-P%L*k(Hl
zYjDY}1e97(^3Fnl7xf~R@0*<HkA}d^0TNU%?({AYn#+Tn0VOptih$Pgs-(+3e&oZ?
z+4Bo2IS7d(UiGF`cC2l^-BPnj%r~M`|KWdS82eo~jIInhQCe(`^uzKB*v^pxw?aaK
z%L2=gm?K2E%BHrm;LL%os@bN5NLv642hr>@h?{)W+9Ogh(8g^2z&xrrSycdW2!2bu
zCX94LZomH!fw#k${A@$(Kvt$|7?=z=Cy-NYE;=1UNE`^fU3lc2Z`M8ai-9l-U_7=!
z+HEirq;gh5QU3UDC>8DM05E@m$0A;(K&~<v5Xg^Oc+OiDx(w+D{r4KWgCGFJWz%Ci
z>3D+F?uz0;;^nYnj>%#OnaX50gvKe)ucDQVy$^POhJ3-6V@VBFPnLqK5IdT^+kdV+
zKJAE>c$n=%JPu9P#>KN|B`Ap)UveIIdep0*D#O$;=}b%~g^qqaZ>ike^Szruaxuo+
zuSy*{&IZp}stxJ99|N=`#On2AI11nXB>0P^J*I<XItmLmAyTG+M{%KT3N<nN;N}pZ
zkZwIY*!Bgify~dXDE?7u|7%wTe~JZAtVUXW=7|enPDcGVk@?MHyf&aKOaxFXhf6^N
zQNlm9kAZ)hej;XhiN)4vY(}4J__v;i8^?ft9zAD9Xx}TWil4KazrRDzg5+gJ%$;LA
zDdom`H2tZ__$H&pcD*^|cTqwm_9mZu#G>D3q9C5Hsb$dA!`H)yyhxIRw`+#-9cuhm
z1l`q7m+SdWIK!))B53q_BgLL_5^oOZKPw9ui#$$gA+C{~2IwH4)krK+2=Zte055o~
zHbFEt0fS2xlrO;>EUyudWpz}djxG^nq>Gx8bvzd3T}$D-3AlEizuXh%a8VNfgP{P>
zQyv3IvMVRj{nizZNXfz#pSGdigdkyq!K$H^^KD+~wvu>(U`gZSYLK7)S`K!Ds?!?F
zl7NHUw|R)g7-fek#EYAIg6+=D`VUqqSWqqE^TyYgr3!g$y^^%jm82C;s$V|ZvgMCT
ze&kw#B5T7|xne#(7MDN1Bn7?6Wo33KL&p8hX{dpV^Vp3ERMUZZ0gPhloxyJdybv3V
z;McKMv+U&;J^dQ&AHR351D(j77#(Y}e}_<|w&&p(uTlw;VG#OrLTM$8_ic8nv@}!L
z+H{#|B*Z}W6ER4OzVdY#oY*|yzBV}jd9$SrA%QLPO1Km>;>?e6a2x7NvH*#+W34sw
zgFnm{HuFTgD@eFr1APa>XZ;=?r5>qwxaE0c;<3@j&}1IV>9d+RuBMbKbb6MEkBpRm
z4SWz`X(0RtJ?=M4GT6jZ7CkF|PKX-S?G-0yX?w*qfytJS;+KD8nwK0=K#|r^_l*{S
zq3=F#S|wxbA#@yt!wQ<L27@%2o~p4p!3OzWlRAlSYl;*2KpT1;82vcd&REZ2L>u!M
zd*lbaymA_25KBDEf{*<kSm&z5NNvmkF@AM()<+HIL#n@m9q@v~pT6p2RNl!v(Z+dd
z8!DP<83*vlg4L+zKib4bs6ktp%XFTbF)GTVD$Y}UVCf0n$xyWGWx#cht?7+l56Z!X
z8EA}aEWZ1F({pjrh}>B|U}p*Z(^~lttVZoYr9N;1lfu4oolB0pO+!BXdp5*G@no41
zn|2NP{s>0qvVcXb#Li0W*ah~($c+ecnXpZ^>4W_Iht||~d28cx1N;x4j?_<$_9l~T
zy>IdZvP!@F%8bZP=7XUb=zOx3Tpn(D3Q3<XsiD9c6B$t#(<i08PoG{TZ`jRp>|Vu$
zb_!N^dNtC;w>4Vq4ypw^IKM1dBhuJi%>xiPDzsKHV0(x&{5Jpl-1p#AQm}@voug9O
z#g%H;`B#>q60a?`AkKT!go%Zt%cTJW40fqt*@|jGg5dfQ&b@ucwRHiLoJeC{7}!fG
zyNnoaKO89{e<yW2DuHy+w9}Ei*K>XLs=;j|mCO|KvUeB;mfUK(+MXrZyP?@aoRXL{
zuwoYTAo>L*x*aghruF2Z_c>9(bSW#a_I2MX#{P>YR2X@);ulnYV8MCAOCJtrJmdva
zLdpxbU-#8bn(SVm9|ne%2Xm1nbNzVYGNefRx;dKHhmGn<j_685+maedV0%3{uMiq=
zz>7?brS_j3)goNuVUq@Nxz_0DFbga<#R&i};V&g?L@@kSseM#|VoZhE_ZJn341fC(
z+NQuWV10du=i|Rh0h`OWxvp{w(>e`HQt?eeH8lfPCr7Cl4AZ|qI0X$=ji#q<SAIb%
zY~@WQO+RaxN34-EM&>OPVl;lnSQJx&Cgw~S$l~SJt+VA7%;>UOTMK{~*F-|;qO};o
zWk<DeY7iq1LD1k8d_Ywjo<LE^P`VO#=9^RNp{)9dC0K+>A-ii5XyklK#S4Dk#!GH^
z@BfY7z#<af)dA0he(Xi{e^J#q45p%k(IqtVir0%GYwkgpqxqvsC=fQYI_3lSwlK_n
z6$Z$x!BwDj&3P7r*$Fno_gF;{U1N5Re%`qN94DivKFj1ppse#7z!X--$+QcA$-ju^
z+FwG0y8Vl?y3aARrfTsw>m1%$(Ve;ue)b>kmj~mbsO)t#KsDmJ&s+W9>f!jxLmdZ7
zi@BdFEm&7S=k(PG>4c~`JStSThjnVN1U)dv6kh3Tt%2I=kJXBCMs$^)b30g+=R*B@
z;iqG-aI!Ni8vqEJWJYaXX|7pfQJE3O&;6rFON!#;dzI;MN@K>)orePk^-(FMgog{&
zIC?aZpH&-&_M8s#DgXB9G<|^ro=I#P@-JXgp`{lSxgIXQG6Aawv+$^zF_N03Q0(1G
z1i;eNpym(~kQUI-Bq=~=>!#`hh;u&N;5r%^>d!)^Z)sHt+)9bzr!Gm4`~l+2vvA(v
zZzM#>Nhx!ud4g#t+lgu7Krx&cd5UStaf_EmlM&(F+Az!k56Q`S;pZOin7TU1Ke4_P
zl_c@{l*9fz^s^ZI3vWiLodX#HkXSTWyAtvtEHG?5B>00G|NNy=8uf1@JxKEcy{Zge
zfP;%RUY9}b7j}LFMIHF4oPtcuVRmD*dB;*57dJr@sIX9`nI(XJfu*hok334^6KVI(
zHiHS@p5mkV+h!L8VIGhy4Jl2h=FOVvKhH#9U=|KA<kccEY*2i1xN#kLQp8ouS~lW8
zoaz5A8};b_q3bK4s@l461wl#aMj8Z30a1|dlu{bxfOHE;cZY=1-JQ}P-Q6K6-5}Ea
z*6|wm-T%EY9C!x%u=m<)tvT0>Z+;UQ?k)L_Pj(9DZ_^kZ(@-6;#;z;|Lf5L!{^5PK
zCP%->r`;~`m>UYlcAHWVIkg%oD(_K!aTz!>c^-)?sKL1`;_`KXD>;EZf6+TLgLYeO
zO;GYy1->k7&vS&e_pM{`MO;gF*K7ZLT0m`VrOU7-YDNK9_tgP*Fa1fZ4Nl?{yy3jQ
zDW0z}aNSR~*KOiB`faLpJC&r=nCT(yl1^Z_^b){Yrryl^2e1Z}hjdKwQ*q%iPT5)(
zNeo{VLC4a2%$kfczOl$OI;L45Gh5XC8h;x}0yBAuc>-Uw3GWc2!jmN-hW`sWlF*%(
zs(gyL{P#+gM!kni>$9ama93b$C?>BaJZ?VSZGv8ZMV=xgZ8vUBRC2^oQdl&xv;m)O
zYr|xER)q096;)?1GgFG8Wk{bweq-f`2Amj#;KWe&`rPp+KKT4G41*5HE$QM(2rse^
zuPg;{DF^urdJcM58&B6ZX~s4+n4tzAp)(Dn#Ueh%Y!wTO{V|eege%xaz71cBps)o4
z`I1)9+m()$se}5&&}BO2HLTMbz7D&h9y<Bc2{Z~8A^&%Fb7^6vEovJEiT7<k;|dBu
zb0L`^Bvgd0di*==Mo~fq8%c`-S-Z#1V30|h!R8Io#8y~e8bI+ncJ_H~jKtt+L{5bY
zze~wQJ1N@GTyhb~HP}&NRr_i&_@UU5yp-k>93<;^nDUZtG<U>WKiG^h00$BB1R#J)
z4~4~K<h!at`fP;2x3${fO0QMHII4Z5Jt6{IQVixPRDVQ*Oshd9re^+sDpa70s6|7D
z3}NPb*#9&24mPWAFuxjNjagOL;*&Tf!N*dk@(d=bEiMrijw151+N4a0`t@bJw2RXL
z?JN@*3hS2*MJ3;bkH#<z@8xSohYy98Vyh7fF;f*;OmowizLe?Z6K^IAh=06SU_{<#
z+%+vaqW8A>Grp4GT8pa+{<RL#Y-1_&dOux(_kA>yp5BU|U>iM3ITuz&_2@0}zd5;(
zME;^!A-I~C@8nhTSUC3d17-|XHzLwl#QNb*=wNe;+v0ZP`473&75%k^OTzt=SoKU$
zRheTj#B+sAe+W|fhLshr%21KXw0bX|*=qzz8JF2eh?pvjxI{Lwtvt#YlBI^Ldxx7w
z3!H%{kN#+IYWKG<wk~TF{XFe^fiY(JRW6V6=UD)qxmBvB+8S&hw$#FwX}|LUVMVoZ
zLHv+2D`R1(X(;waUg!FozsMuADXvC|;#NnyDR^So(fq)U6@PVP$_**KfRzH1GQ+i~
z7GFd`1|MMDN~^eOh}|KYz*Io@3~zHC+3U<NDEeB-k16t0%3pYIkVf%d4MD2KgnS$=
z;2W%OEYD&IFamqb@(bAcq0uQtAjSI6=2v+FONV6bu9R;B1I%ZDk3hHfZ^KrW7i(yF
zObypkfO7V63GIoYTJNY*L|T!Nr}+p)^vhWNgvZxH7%^XJk!{Uo7<hOeWf$=ETKnG-
zHdWyMocm!B#y~DpsdC@2y1Fe8AGh<J;Cm)-+85>JAK;^7&4XHEykYy{h@j>NF`viQ
z==#~E0@`0S_uEKv5eE1ZUPqO<yx)fzrN>K9XLK3Y?(dRmcGA?exR;*cP8(W}YXTWt
zOp@R{5LQt7-lDMax?z7Cg$G|+<tQX9-xBtjT-kssg<}u92~y!@Y|+j!n|?Ue3%DpL
zB*Yz_q2NL>;-~cF)C^&$`H#Lym+RJ<5dpANf=0BzDhorn3hc6L^`vj+3}9NalPIo9
z{|P!H#seqFa!Yp5uM-4R0(pT=*ZLaJ02JoZ*R%I$jtEC0a-UM{qWe&lJllLFN+sf~
zSJYBTlpD~RRby3aXi^u9+g@D=BaBY5NBgM&i~T^1EEePh_}Sj`bZ+ML*efb(ywKT3
z6dD?S)>)m&VXaAiIdQwjc)0o~p^{heNR^KRUu<;Mxw;fYiEuAZ9v+=<{sW4E0tib#
zj@tg2()^bxpK;S#bnL4VV&z&tDg{)a1Aybmg>REkTX-C-hv)%68_mEhp~^(2yS2PB
zO+u%R`1hD7vFE7a_~~}{;+oyN;snz|DJbY5W(k$r92mn9wOObe6l1UhB?C{854)cF
z(7nZo&6pu83C{jj(rUz1i3R-fii~4SpKe)b-DW0M4zEX)YhlqJm5c82`ca9?<-ihX
zC|%s~wAq`st?=`nM8G)Hta#zPWObOYJ0&ALHHEux#RFumx{#Z`)v_l^EmpOlh9!2Y
z6?UUv!#V5D2STD&b~4c~Mu*JQw5agIoRDHlG4!NNFw3%d-YQ7DPdRXT3yzcnTB=E&
zO@cxpT!=p{yHjS6;1(51Y6_575K=!ttQpH6)B3AGe#NI!_7S@Tr%prVp8ejEtrBFD
zgMj+}RDNQzA{u>($av!U#A4<)b;O8dOdE$e@rMc*^HSu|<l<wt{D%{*C@9u#4T7CM
zW22H0+h-;GI!_WdS+U*)ixl^%$gj~QSRw1qydFQq7Bvo(XCwdU>)BFXRu_)pEJC+9
z0t#;|g;rhr%?(dY4FZgHMfTD|1fb5I6cw=Xgtd+TL1luW4XJ}T2ssjl_@d*c$*%a?
zUK-095#r7Z5%MLIxmx|wLpT${Xvc;obRUfHi2S*F>#+%Hl3dt+;qNi%ZnhMZIb8Gv
z;pKYUM8LBUA?p{<(bX+n`VN^DwUW^zniO7<CmLi=LT=jV_R$&h>CKhxBCN({SeI!c
zT76`r4Gny#zyfKYZc<SEqw(gc+-chcJL0$`86L48tfZ9`9KxN5(|&A@Zs`d@SIfwH
zz6cksT**g3(f@!`?4ZK8fWCY(BQqjJS~|FT$VBzka5-j{ZRlZaD(u~)?7|W<HpNUt
z>)8pK5#`Y~V%(PDU|wUTZkc$|8Q3t_661jgrKW_?`$>pC_-SDW8vdAu`tgO2xSKNn
z0pdZEiSQV?9>0?h{CwzSIY?<?I+gVb*m3q!it(}M=$_<xEd`>F11%{NIWTaQYJ&yi
zFvK^}fEO%eP4w^?xmfeG7}YQxmh&cwBugx|k1$Vv3w8-6$BsxWL!l2sOV)AXM0Zz!
ziv7y)J1*Wgc<x<u)w7P;4V9(wv2>Tj5G@4l7kHbT2<(&)Lw=~O7b2K)B5xa^#0Bc#
ze{Zl~A6P7brAJ&XmPhm(HMU3{mWx*s7JD$*XQ6`J>ls_rDjD;LqEP5wa`O*SHDDLi
zR9e`uPhi!G?01(?ai}UjH#bk_xWkD=$5=L%S;68o$T-2HAx8J<6C^hwfjwwB+-RG!
zAS9hbf3$<J;>xeOR3C9)PUH2m`BF?npMuIfnb9-uN9pQ_y2Lf1Gy-No(X9dJ;yEe7
z_#X%8h|xn@dr+Y{cF3YZYj93X6Pu#wdv!3QApI<BZjfPe!K!XDrU^o;8M`$i#q@JN
zBMS304do5<RGJ4EZU7DF2)r5`cYR04Hj0Rsc%@M+7nz!$crkpk;H%7nnlW;Y@F+^+
zm|;1-Kzg}Mx^RTm8D*!!Vf&k2Lro`F;3yuJ4HNv?bG$TRX)G40<AhcwYM+|eGKZuD
za9WrjgUU3G3W(5EF1BaM;zZNSa6`oPxsUj}szlHAJPUlzkkj8hu+WX}88ha7>eGL}
zmqG^<Mm)eaa1=sb?n)P56@2d4T-y{Jfen|HYtpD96eF>Ht<%{2L%&`SDvcCFz}hYO
z<QMWwX@HESKYunYyHs3_V&zVwC_x35ofL!R3ieJVE=47)Kk{Du3%fWn_C??48R!HN
zUp!4E$OTeEv+7H*D<z*P;JIZ9Z2KT&_;>gsa(F)${5Vp6<t-y6B9IbmL)oq)-XQEb
zpMj4`{$kmBKgW8T%<<V+PWEC+M~}QpcCic4v|KPQdatrm!N=c6{68P?DWGtX;ns=5
zUtswr7R;_O<nTc90&f!Mc5QTj>RB<yT%OA@h+f%86Qr-5NPg7e;H_N`xa@HB`bpPy
zAChh(sp9#!TYN^rZgmv!?AA8RK$E=ZX(g4Kn6Hjd${!NlY-|`Wpx@CEx4RvSc1Xzw
zIU%ssCwMFP<YC>@I_4Jwt{u|d;U0Dy>Ob)!R0ziwf6#o|v<R>E(zE&`n}zS-PQJ^@
zt8^FoM*hI8e^GqX(P-BSjoO0T<bsyu3O-sM%QwBBu2`sV@rOejg`XpylGOJYx)%>U
z^uMTHe&Cwrvr^40m$dS4qJr?Ob}iD`y`K9_t)WyuGB0>XNn$&9u=eEJ1ZRDUt(fF9
z)tXR2NiZ5Rkc*%}gl_YH7V=&azZ$aLUNGEA&Xb9+=W#D&C$$lz@hL_Gha!iy`dXrz
z^+zZ=YkBe9GcpX-QFS%RO<*3;2uh1m4OjEt?U#<nG*uD%iWen_B%qJ(<v;jko-vGu
z6(fmG6eG*OyYkLaKARPf33Bg)dXHlxr03~{)31}Nc~#Fcp$ZjW4Qy_q%MIJ*-wNRi
z+{x~Xjc5BUB}_=UBKUEumeXVE{w<crZ9HrukdLh&tS?V`-ucN}jQzmUmqeQOsSf(~
zB(|8Pz$epCSN!pm8qP#ZIkv(ii&X$Vw?<Y*3X^A_aJfPV9XV`wbk}V9fMe^B5o&)$
zxs3^9pHHOYDy_8WbYPPu7Q!3vMg5ezKL+EkZRL?lMTWd6$NPS+{I^!U1E80mvqDnn
z^`r^xifyps&y97cGcq!VP70X%EXE*HlsN4VkU4~!KTa}?%C9`Fjtq%Yr(;Ta-9%X6
zzPPm6i#KXFc9AluVuz1mNcd^Z)i|rO1iNEOg^aRG+Td~E#=R=V<d;TtZIbw44UgTq
z_DG171h-ark>GJK@Lz%%13YQAlYew6HB6ehFD(krYj~2U<n6G$^ox%r@8E^>yr)4?
z#8zv=eR)R6hsL2Rpx+b2G665hGX1$S>?2d^3T-fNeXUD~tC5hADt}yex4*t)n>U%y
z6ZF`p4>zq5OV!3lL${S-*CI@)VtJZvC`w)ilb_0W<~_r}mxM0c`IPPGpF!9H^E&s?
zFoA(`g$!>w_H*DI!bJ<6+E}5_7}*f(l(Z-!DspP!UG5Q95W&9|XY&w$Td!S2zv3u4
z+D4+670f<r{FG<7ULA=s6CNIUaj})MJxhkF08QSn1O-M#ghY=wk~a7~f`I47+Y!h8
zA>;QuQ;?{ZYpWb+ZaNi!LY}jSsy%!B`-HJ%hmB1}Ga*PyW6R27eWTqvhI<KTzS*aY
zpG*4~BJMrdGRIkEZY{An@x@ma#p+SJns~Q(nhMO<P6bfuvxtxw;bT2oh*5^E)*$F4
zPt#HQT=aXK?RHCGH+8uTK(oUx%B3-NR7zCag0!kpSWC1E)T>OTKu^XLvhA%^{jXL+
zhXI+79Jl*_=8Hm(QjWUh*+S2|xrZJqgyTzs79ZqMd(|VR0xEpvg-Qr|Kn2a(5~l*U
zL`m$(L!0tGMf{@lNRJS0);ak6)a<kpMMXs`*LSQ1AH+gl3FwK6#p@>$(-j6oB>G-q
z8#!cH6kC5HFG?R?C?_2}Q|l>@ct0rA2_EkS4O*}y=kH1WwgbOCN3ii1ym$hc!u8o4
zP?ge7etoRrL)W6R+^NRLjwcw#YFv~qgB9@M00GzQ1+AX(V~WssW+$=gQP?k|J@m6I
zpH|rV3o`-7v16y03t4{oa)md8OzrRjyVzgqg)UPG2l)aT#D8AznhNGUEv7Hjx&xI_
zR0#KPFt+dgQA8m={M1M_)56%EeQDO8PX>@DT8rRLtSCJDGww)PRY6*F?t7)8MhQ;f
zU<O1)U<zX#JMJ@6gbq{`FpI2}t72wni3WBfV)<ltnX+dgb@03EznFBMAKvLoCn=|i
zete-OgcXoWV#XTCy%j3_cLbfNYdXgl_pe!I9p{;6W=VZ72b_pIE(g}RcTSUT@A*R1
z-PT$y8;`*+-PY$$-|%S#Tz8W4ZOj2M_c%&W9DF35qI@+=+qcQcVwN@)9FK?x==z3?
zvs`pDRZr(WKQlJZNEgcrI$*t^6)-5lt?YZJoW*~1avmj!HaKz*L$V*_q8yOuREPt3
z35(LLM0mQS@5Mk!$gz#ECRsV-)6@k<WumP8Ti&Ac6x}#o8*ra0yWyI!QJluRwHEfH
zuy3&dWC^ivw&`h?pTsCGn<;RF#rK|@m&v>G)S1t5T|U&h86m=ib=Xf&Z#~xK<=|M8
zwQu!B6*`r&!FTI$n3{cdJMXu9_`!*E;&N^NT$QxLV4%MEy{gH;c};`swd1YBo88Md
zy$5H`%lmkhfbBv7&Uc%+Ju+?T{`C(rq5Sd3pk-D!Z_MrwnJ0Ywot>S2d-E90&KNA5
zoHfq;mnqrUEq-XEB(=;i9A?vR9l9T&C8E!NaVS$27WLaRNFWv4bD83uJk<3Ed5JHl
zSI0Hh*{Vpr%=o3)j&ismQuz?)b9{-7!jWvE+{Z1CqYod2yA;uzek^a07V|)+%cVvj
z`LvBxiu7_uy|q9@<@EDPTuYy>xO<$wb(St8L?x!$(+2;b-B1mQo<22^32WRhl4XM2
z_rS7NrOJG40*j;zqU_R(=Ntjpn5N>o174H5G2UO9KeULgNbu_*F$J2=>)P4Y?vLd5
zifb=4c<rXp<KDDuFi3rk)}qZcK=M#0^k-mxUDWcU+{MZZR96(AxAtS%$s%2@;8O)|
zymfHty-13EGTF>}2W3|QU%|(TE>n-<<Nlpa#v$o7nxl%)_aS6#ob8S)!wI3ts>=Se
zvIPdgZO2E_N5f`21-UFoh|n;z_E&IxGrdr-{ih57m}M+oaI6<J!_NbHRMFLGEby(?
zrbFOP<YUu<lxmzkYSDd4i!5g*ED@bQ+PdRX-IJeWF(LK&_5?v1w%Z1eTi{(GIS99C
zzA)Oxh>syGWM{NTT?&8C8M?XIGtxzz7J=KQfUV$n%r+v=Vt!Pm+xY;?=W-bn8qz?4
z1Kx&*eQ+(}Z)N8(PU8*`?k}tQlvHZpg?zi=)LH62Rz!I2P0mH{t%{Ie%W~wH7NhpH
z+G<RgSO;AmFN%vI0F(P3#J?oHWU<yUEE3(1QZFU8y7KFaEKi%3ZJ`OTa()?QD5iMF
zs62~*1*9@IJuELacH<v?+f0U@Ea_#%T#NjB2!Cur4{G!<A4|lNu~)y<96#I^T#F?B
z_V$<g*(ae<O%1iU+r;LX)$gL#L|$~)3QpWE{)k0f?`9J>4xj@CW(wjE9>|QON?p=7
zHe8{ihhh50k4NORsmtfDr|psPhEZ6&DxNU-slZ50zG6t6c;n>|MdK~4k!)U8@t^|4
zpoX%cQS8zC*+zf;h<h*;mJ|g`BH!_(aQM+<oWCCQp={noe|r#UeMQi5(KO(=rE&?1
z!raWvafvuMf^4_Nq!~I)ObZr03@sT^C*|t`AJQXmMG^hb(FAM^yjuQ3ZKUw>guwI)
z9;3uQY_G}bj2xSj=(Q$z>&pvuzAoYcV*mR3irwk=6nT)4g&7eD(#F~gVn@jfzQCHI
zETAQCkp6h?>LLS-JE2*hr|3cCun)pxBsL_Zp3D&%iQykJnOM|JpqFxfg-{E_bvU+^
zAoyyaFcK(rVh5G>4XtS_zC?zd7!=namDna=-LL2P@*XJm3x9Vw2q_PP%<JEakWz(6
z76MQm8G=_L4T9bg{q?WZ7nvN7wAmc8RW`vPmh>iUBORL}Ks1FwK|xrGo9Ez1M#iK;
zuh%dcjpEx2RVQTfnA|6VVWMHNJE2kLJT|Lc@(4nBwXP;sZnMS0knh8F2uGA2S1sA3
z>qVS@X8u<q0R-aiLjzjw*;0Sb9pRMtI}@KZC^xo|5ZQH25UkxgLagh_nG(xUgEYs=
z_$F%wQ9P<ce@ts{(gD^gY@&(y)u9QjwS36Lm!IBDq0f6rKD{5L)?dlG)6>75c@iy4
ziaz~gV9QqTvCoHaB`3Z~B4Zhdh=LQvkOkZd>$HWe^x>}*8c0w(z}n*VKoL+`pW%Mc
zj7%@I%~;u+VH%b1N~p~%i;~Gyz5HaP30x9c@eqMDe-0W9GY;<>zY#{JWb(6ug$;_W
zp80g!>aaC6A&Nv+G+}a`r$J6ZYA!rKXf-Q8O^mYO64mCLUN@AXL9Kqrc(5iGe5Cy^
z|E$R%$O_>h?nr!<WABYG6>&{!B3ae<Q0rOsZD?be`2IRA_}B%ZY@ovB2+IUnl~kO3
z2*u=H8FO)Y6}DeSM#@`u1-_j6(j@fWx0Ts$OcL_=GBPX_CcAK(w6w9+4ZMwp`A{Rx
zm5fYa3QbnR8vD19ZkWJ^7hC459tdub_i(=<x6{Lr2x6t!JhFlN(5!<;M>?dXA8H-x
zM`RFlwT0V?ZLg~<gUl;L(o(`T;qjWHSPPiN#w`ls3;rGGj#Dr-A=dtalIo^vDe;nb
zL@zN8<>yLU3sRLLvWK1k-Fe~3f4)42XFX2!cj<pSK~dpQ@NLU0<v&`Uf`t^OKp#We
zm+AzUKP#yz=^HDF_Pt~U22TuAn!5vSwG9XypI{71hE^S{;f^7txF6GW>fw_X$4%A3
z2%I4GCR3bRhtWNl;_;8Wq}OrA4BQC{Vo*PhKcfh!_w2LzJQ9v@bQf$zS9{60^xvLE
z6F)lHhNLh#piQ9&i-H~U(M}LL8~1_u3*OiuH{(-;ZKL~s%WWg^`$BPs0%KU9scI0f
z7BF4=@+~7|?UA%Re!U*gq>UI!VOp3kZ2{)w<h{`%1b==s0O#9j_DiD!RTlpU0tyF6
zm&Tn5ovEq?IvOJhzX@#Xp->5ER8jWV_Qn!q8En5P9TnQq%RL$W;m0D(@Q&C7mEMG0
zK!0nK&M@5uyIVmWV?z@hjrEtN>}-ED3k(f<8ag>%8jCl%VA%4z-Ud~)7a|ZO{+r++
zYX@E*NMPF((*vD3$u1lMahd|}VX<;hQmu}^#9_<~{htC|N}d){U2Tw}L{WkZ48%##
z+KxZF<+npd0^LV<(F7F#IR3yVdZ^Sg!uniO4Sr72KXb@)u<x6~mI;z7AHGs+$vq{v
z9-m0`kWF4LC$q&R+T{mvdugS|{H*#1=?ya<yR6h6EZ2gx_1(dYOiQcf{78@lH^!7B
zX&KKbEd3Mh5KalYlg5JVa3!v7p_N8b^T{#V*|1b})LVJwN_SVg)+UNkBf@r_e&wR?
z$rRm+!uMgFn~~TE-kJ6aPG0K+2Zv=tECF+!BF@GyYS2jWJcXP~;h6g2i0R))fbMg`
z#(^5jhEdIXW7y_A2~L_2xYR}2x{F#t^v`SP5PW97@3ksHG3)6KH%TqUn4xsy+qljC
z;*#+LDXiAn>HclU$MWUOAlh3AoxZ3>(DU=p>VSTiK;@>^18`(6owyW)xxoBbo6Wyg
zng^iQ(t0h~^cpeU=upyK%GJ*vHsiIhw3E%0bG;WMdCHcZ*lIggDNM~s`Kia&1~Oty
z4x29n{D_JG6Yk&l|NWp146rYFevq-F{=Fg&`_KVtPSVgk1y`j*QNjK@0fT2GU}p0e
z^Pw)88<^I9SO7b~BB+K;|7@h3aP4ePJ?j2iTyqzncu+Z^?`jw!gZqIFg#7Zorr2Yo
zEefN01$lmH5x5Cx&Brv~-TvQl45m^agCF}nf{?7Xg}kWsLkJ*~lNYz56gg{GYr*#O
zZ#c7N7SW|DGpNS)HN{gOF=M~mK?YmH@eWI$Ir4=OHDUghHyvSQHLn9mBbh*5CB{t?
z)4wmG;?D2iGD4wJL@ta?v7z%J)l3sXR8!@y(9SrcFG7dC-camCldYA?LUGmRY;jOS
zEo}nM+br-7BBdzYqm@*W*zXm_!x0L2f#{DL6k+^TMcNWjmH}MBvImuFD#%juDz%aC
z>%y}PgBd4UdUV7!)ll(X#P*m)(3fClWtO}n`q0pV^jvvmxH{`Iv9DZkJ`|)@asgId
zIk`ske}pCpDl}N30-yc}4M<jFJ-7cZcd+satG&VHes6`R3R1i|wW&H%zb$svkcL<c
zZWH&O4S8Wl(EIU&;Zv?ck3}Zq6H#BLyZsr|;dj`~q;xiExa=Mytv=%BvQEVDh=hVN
z-kFY<{nlpyF@oUn*n?;LKAVRNqcY@P5?y-8`y%?clGSk(#LGCJr12C!S*{_f!6)aM
zLyBtAoy!Aq(&$it-Hsr#1*nOq?Mgo1xx?x!AuRCf-TRec6{(!A^dWf+k8vm(z4pt~
zxWA8LK=$~GxFa*9XN(lePl(&*&l;!04BL^B$?^!(?8jh>p_id_FC?4aMp0X9Yo>N8
zKRE}s&pVR^YpjYfAj*?sn^n(87?PNBBBH&DxFUPg6PkhXS2X$Iq9Xal)B;&cq(#Y4
z>9uh@*yltQS5~b8TL;|Pj*XOH!+F=4j=!-&=yo39e%ed7g~2zYLR#0)Q^Qbs2qYrE
zRW<#O74qu?{~!j?%r#<Zh<`M*f7248HuVl5a?vIz@IH7vsu0nv&-zWe*2}j&lp`2#
zqS07$@!%>pMoEyURGzvJZBn6(2GcYP_btr?IctDDL?&W;W9b$(Cli4f&13mof-lL+
z_Gc>Z*E5?~F*mwiPiYCM^u8AV;iyF-PBY8j>cW09>Ql~y`C<~6#O1SfF{MrBCZ5tZ
zq8S_!>r9!MrD@BPISEC45*Oh_V7RC}fKBfm4!HQcnZWOyVEqLdOdtnBHec;-RZvPt
za2S)xepqyP4q_`sV+c|@(nfK^>H1e=D4VkBy@3cJU_@`j<2AgQHgMyJ$6qLLV`Cfc
z^k2=UwF8C=YF4+Z`~!0O#g^c4LAa1;Am;~|W5<7w=OH0DbRX6CJG6(Fu}yL2F$m>4
z970adh7E||2;e<@?S=65A2dt8at!+XFuwNr(bja$)>P*PlAu>4dj4C4-gmPKch>Bs
z`_|jprSz)ZZgr1CDxhjJ7;fp2Bf*P*8v?&m7T2J@y@W66lRn+=AQMPY(F}#>muLvZ
z7|}WRYpeQ=Gmpx?+D0NMBC)&|J!pBkf`VIr5mcKVZ7G7<OMdBKTBqku;MvW-gA>*K
zc$nuHa6>+KJmrFBtfhhIh9br@{Yr!YZR&ADHiY33;m%q>cSM`vU;o301=z_~M)*MS
z`&PU!T6`>DjSpMxmlZ_2J4zjS{ocH#h>5Zbg;C)-#RD7%YzpA3&w{%~$LCytSnHqj
zlTuEEjM3)}fsf5J9658<gQ$RAMSS1(Fk2tn3Q=|QkXmQPRsHKFPDObeF#@yT2*m9G
z-Yun8Y(FUiS*QhxOOcYoii|Ph@cWh^j(Z1|z%(mx*66*5!j1^`$m1!Ls%;E@lV*Kb
z0a1!o+BQvL$`56#cw)BXvFq#8BeI2s0em5X_Kt63)xu(1(sD#oh-V>C;Cx6j@K-tq
z<#_%_R^C9ZnBLATiN8jDQz-^utjwwln3zO=4;LQ`wy}nIe|!0^AR$vafeMX2mhD5_
z#&l`6kK^hlc8KkS_?yht-dwPKZ^L4x)sjsx>7~}??+wv`xrT@O)&AnBe<fbK1Yu{P
zu@TgwSGYf+IRY`v&RW4!lvUY|U|}vNO>$TkABcIhB!f><kf8cA6=t~DZcn+sXkm!L
zuo7HIA~ZK$HX8Y_9Xknj>|8D##le@%VvGCn0bLowj|zn$DoFX_8QL4=53*SLcCS(t
zu1fn|YV*BGj|tw=Obe@frwT7#y|opE!-Zn=pG1s_{<hme1}27I%s7x0cC*iBRCv*J
za>NmP#F1)b)BjdMVGG+&Oh{Q)NT5z!UcM$oR5MUSgoKAsDiO;=7lcdg8OCfr{8gU+
z%r?p$n9M|#HNWS)yjt8>-&@-OH@BMU*Uy<#Wd@-Z(X4&tF6}3v{ip+9zkb>C-SW}E
z<Lyc2Oyr*sm49}a2Q@~M<n!L^kpkCI{gzhEPo9^LDjNwdE#6Y?Ah7}6Jdhp9;?*vP
zr`$HbG+%m;HLyO~1hG|foLH$}#Ado9$eP2|D6-GNz!=p6N{Zki0#u|1c9sv(^>;r$
z256Jtr6i}l*m3*yfyDM+t&`77SBHbiwncSIpeHnAQRB_UQmP+`akBB${R6t}1O*Mp
z&CH;!beGiMgqpxs&_;v{p+QQIH$rAq3>35>x;(K1SyHu$BY9R#qimKV#ltVJDZ=hm
ze-KU1WFy)}EF{4jM`2-0T&{Z`d>6VP-2Sf_VGlvzE#ZEjLLQWbf614~e4`pJh-5+#
zS~3v`?VDS@8l@gMS-)osuhdx6BlB!-MTZ4Y?$N2B>^I)J?k^9e3bnW6y-ASy$*1#c
z{`9Ej@Lt8j@iJQZ^D)?DN(WhKB6Ap1Fu^^1--hkW&RzQ)B}m(rk`7(UBt?~Sj?1Ve
zj1fMcdZKqkd&3q%B0x|q6>yZY1N%ZGJ+5vs(Qf`LuGjWNp|#XE&_7$*bf&T&o_rn3
zw_Qj3WjW-<7hqrPeT)*=vU*y-m6spLHt@VPe(t{|r|Q8UK0r%OA$E9Nf9ZZa^r-pa
zMl3rSUdLQ|$I-HjJn#K57Dt9pH-{^<_1MY~L&zkMbxztykr_f=Fs|*wjtqGngC(F}
zP&Q)A-kv555s`4BC_(p_4qX<Mv2|sFh1S&7JGSvwXx&~+)*U~1BHk=%5BkmFx*dtR
zjcJLUf!2}@XPdHaTkD5r9lZ6HSed3%C8^L}aT*Dr-JUCtOUZ54v-;er2A8_3=8yj4
zaz7Jrf2A5|7(XDIR)B_Q@qr}7<r`?9e#Gens4YyazUoDvbG9v<Hr`%`y#HHC`f6vo
zB7d3tKA`4R08G%z_B=2qB^<x;*TAJ}lEiBAiKuWoSd#7RlERnR;S90v^ZLddTRb#V
zBZ8K%HTQWkv9-iCL~HHwYr;TjN!ffA*(p%AGw~z}x4))V0aOd}L*o(;h^1aMMY8oe
zgM9)8xA)I20scsNr{u)iYNz?bq)DqH*CtMe@Q8Seh8kDN%fHH{L<cCB#1<BB(j{hZ
z2vWzpAQl@LT$9_W&;*|#3^xa&G7(MDhnXTsjVP?2ln=5onP){?=bjE4f}@8852lbj
zA*hF9JL&dn-AU3Pl(o;MyOZpGew1XN=Hhs`*1zlW9uP6}@^$bWjVa0=U)RioSnjKx
zxw<N7mGN4zKG)>aynmfBs}%0E3DloWaim^MN#~xTyYF07R%(~1smHLzF<$&Ynfc-L
zT|pztphrAw+)?ntKa;(W(yMQ!LQPiX-KZ$s8ac?hacko$)>|!$;#m$l3|Q~9T6M}Z
z=c>bW)5rK!xbuOov)AZVL*;kY&i8f$9wlwO20>;f^t6A$=1bz>TGZ@^7+uYj73^@#
zq9$5u=EO&J@QX6YaYN0YGK}k8u1O9XjMJU4_QNZ+0I&<*2ZVO}e$3Kc-U=uxRiGN_
z_8y7z_vR%iHD9RkP|T<-Up>!OIHlM$j1Delq{b_YGPm2C_k7d8s-ym%vYD7+y&DPv
zjHNtLU(j4Ogv^FdO&U*3;ER$~(Y;kI2{d!UrYU;c+JyHit3W0Pz6PnTcVi>R$KJdz
z{%&dAbqcY;Kqusj-Ciu2Xt%ap(x%{P`1y;PeIJHB<l}LR^}_|uL6;aw-gU@j)d1Uo
z0k8Dni~Y!TUem#Z^*O1*t8dMZuB}YxdKB=bh-E{XSe6`I`DdSRN8R6XSrqIsc$E~(
zG$$Zoygz6A<>B#ifwWMw?Rm{)iI=#_pmSc)h9C1d8VR?`>yu`ZL6E~%iUTr(gqf6(
zz%YM%b6TY-I0)DtmwcF=A)WrV0^+XCK9cqQ8lGVYS^F;gP6oDz)_kYd?Iv1>ajM6y
z3Lg1pwKETTdC)Fhk^qlDYl!y7UL(mDlNfVX6}}nq_K3|G-%W?pt@^!%Tb1>(HFf2U
zwBd-jI-8}7SkFg35QAqyF9fx-E_C_cVP!p+fplH8yIv*8<(0Ny)63=SJ_5(t&Q6FU
zwcx%N>6LbPxcKSL*X8cDI_A9Fu=T3hIkTvD(nGPz^;@sYm=@9QVa|kxh7NK)>%ws^
z<0v={c^%_9ZXbhW{x!qVVI^E_dI2A;_j*~c2~Xr)=DcL~S<M@e6%l>rm~>v^QgpZ<
z_w-bj_dKJ~cXfgDG<HIu!E&#im1ANjAqt=v19%>ZzfHz%taP?!^n1$eL(VA0Mp-9D
z2O$W)R}&TXG@yVwf#YBuaeLQB6F<u`F~Rg<=K4IR4>{SP`sqLc`zHb5L`@wF?}HM<
zrTctm0?KB#UCayS^lxlAP7qFh0K(N`GXi0hhDB71n8W+Edo;B%XlIT@p3^pXyzE-v
zJ@IUmFG>B;_^!SWUH@oHL=fmAmA0-K8GwDtk5i3DiL!O8e{bVttsbXAMF9Orr_MEU
zv;#AK3KqA;?3kYn`jYP;9$D)NThgtK-$4V0w{s=xO#R`D4{qD_SJeC+TrshT&8umJ
zGas<&Q`aB02^t0tSrbpY-nA8vl_j3oPR~>)>}}{Foau|9m7bJ01{PU6?Nfhc&*dY>
zMX8sFEn04p*L-}SbkD&4a<>XE8PxFC8s`h~ZiLVvxB+I|YWzNnJXyZ$qb1v7C(Tk&
zMVj|%T1yjsU=>!)WUKC&Z|^)z?ti3B7r2A=1uJXSd%!?=%=_H~4ZDWvJE$q6)4JYL
z#yjR`nTsaJKCeWpYQPSkGu!ICJlQoNeZS6TnmqKU%K`9KVLUDtSi`qq^4}Pp#~Wyp
zj|RK5XHw>>qLwm<#<o6=$zGds<TDF`#L%!kgB%f9H>>4bd0XFkH7*A=P91mFlihCQ
z@rVZ>?KPbDTi}wOcZp3i8RNZiwh|B?_DK`6B8gUbhrxTXk*--5>k?yT88E7I@zwd&
z9}N5u5gcoPOW+H;uQnehYKqXC{gI7|?n>j0pm4X&y(JB5EVCrabSZiiT;ZL{zFE8{
zSbOF=t^jv06a;lA*{Erei!&oET(;HQ{rIjRZ>m}^GNHV=?SK!Ve1pl~-iPdM|0t+7
zT!FPJ3Q{E!ekml$Q**6h_WIFZpT<OnOa$UW!qm}Xubi2^foWJ%$x_VL9aGm?TF4OU
zFR%u@GOKT6nRkV>jozPm)a<U-VN9TCS@%C>x>tIl>2f?Sg$Z=|to5Lo?8eN(&3mUt
zr}Y;L=OLj&*RcjkUtQFoY3N`7hOEK4h0&zfb`ug1-Tfp!dsnvXX;A8Rea`YdiVc)b
zy=GLK2IWe+Dl@6{AaMqt1hOr|ah7S8jo_dbON+t*6InRl0RCMu-57HxhrLK~*W0VX
z+u8S6*T<lL@)BSdR)4n+rB|K7$oN@j=-;>Z0~oM4$=5pu^>O}=cp8=hc3A(<LPb3r
ziE1vQg?GJ-@B!cG3K*$0;kOa+9#=a}Xi6eJwqA`?uFb@#8iMw#Nh(86^&dX~b}1e2
z#c*h)*MNrW`KlP|tFoapF5p)Fe%*W-vI6hLtN^_t&DseZo&MH5JUW{eNfp)WY2}Bb
zFS9BUdf!%m(9a{G=$`aiA|Aa{D-L%N1~pf>%c0z|yZI6X9{d{8<hH_^=<iUQO-o>1
z*VvK|D!ljIZZDC4ppqkO=h86!s@_xWV=?}-u|VydSV}_?i&x#)Bn~D{1V76Pru5=e
zc$H<hU4#Rod{U$}F$LQMrtd~VH_D0B-WxoBemATzGVYFbu@)29CI*8V&})agWiy-i
zICly31*TIgjB3R9%E<Vq;;uHKX*~M3KhYG2zEO_bfvs`-m8q6NnEcqBmthh{32~Pb
zYu6-7^l`m+Q*U}rr3z9hW-WT557Fze*c!uJ^C94akt}0%ygRLOyR7Q8Kfa2_U|2J=
za8m!t+;Vfd&)m`388laCcZnR7`m@62m*0ZqQI6{iUqZ7dc)ylqMT<U&Tx^)hX0~N1
zl%BRabwn1M!G}}pD11pStwcDZe+bIOFf1nX?=rgIk#S(mVdN0!DjUQ7tyTEBe23^w
zl3n>pA1NkLmL^m)vo>DJmfCM?>!B9Vgd5SSNW~b@j+j*#XYyd+J3|p@H*6CR|I)Sc
z4`QhpYLcfh-5prz7yd>NGIGr}$WY#H3!+FbG(~Gz4>8vgaJOXEMN*QXVBA4jS?hm=
zZeb|VDGDTY79gAqSYQ2Okb@uw3YX^vzSfh#n?m=^DexQ@9=U*N&7Z;WPz7nn1Rr6X
zKf4ZmFp(f*v^L5Oa%a|?2V9kkYACS=I1Ikuo>&Jpp6@>@40=5_9R4{}?as(K^CSVr
zhBcJF^Zf4E(ck#wKOxqirR*g1oSOeq7!)OG01-vP$vXVtxmUY+B{H4jH`!_XML|OI
zjG%Zz^YqAVLp9{D9rT3}LEAEg!YQ|)802JuZ8Z4r-L-QNcBZU6=PU~7->NSRif3Sa
zz-9pUEc}v=h|b(H`$<I<dfsC%Dc}NICKv8W|H}vVZTcijQhJ1a{{^GBep@d#&a9z-
z>z7q@Upn+Mj!NrmM2)Hk`X&krD-D$tEHQA{uk`;f^b7E!KQ=kT7*v17wqPT%r-9R2
zeZUoskU%6~mptab#v50Z&UrJdoTmhL5!$rm#(J?o&#wD7^6@*Wk`DG1HxhB;-|!8{
z?8(4Fa<{dVRqV(aCkfT|T1xPlR(V-avm8oB&Da@zq&8*DL{g#=yY$MVlS9@(#m;X<
zLBYc8VY5&W#6HUtAR1?$Oukbqk{a-3gh3&o{c8|>S#b<MQrp+vwZxQ{{`{mM(_Ml_
z@-gcVCIH-qg%8FXqW2`@VN@%+FNRH)dK)?VvdD~m{{mkcrd13@854i#`^bR+01*bl
z;Fq1j`FFDdu37OEOaS$@f4#)w^QP*#M#)UkJtMlEyopqVBRb%y@VZ#WRsNRt-=agz
z46ZCw=WPF7R*-J=<)3q5FZ9~PbhLRk(mv-oD~i`CQF!6!?%8uRST-X4?=Ry@kj;uA
z50d($XJo|a%4m;)x*idF3)S7q(~;NM(TXQx5%DGcqA#uiFGopoa?;u#XgGLd%U7^q
z!wAobs}p3gWCSg1s@snlZDb%EF~HOWpX{U$xBc4-j0)nvmmqK^(nR3MTglIs`Up4%
zN?ogQ*%u(VOan~+>9(gy|9!(F1n8~0yoT8SeJiCX{@PGhq#8Kghl{4qMrsVmqep6L
zaw->~g1j#)clcM3L70@)Lz38cW@8q}h8^8pscz@T^%P{`6g+?rn?jZqu>QYds<xm=
zs?XNFfVp7eUZw7!O@F~e#(VwR%3@874N?$PzCJ8q{{uz`&W8sMMc5L$@(6=%^hm*|
z65PEH?i!-^cLTOu|2<&nXHZQ~g@A`B+5hhg_mV-oqWl+y#gqm2qjL9^3o9PMUPDW+
z7m{MufT#Fpdjg3^_c@Q_kpsKX09!?$18e*Un$nne6OH@La|Tt-O@tURZ}10_hCq7`
z1LZtXIL8F&x}V)0Ir{f=CzAN9=#6;Jr2AgJkE(E!jB74MA8MS{P0cd-_t$=o`wp0>
z=XcQLtNsiId?=}ft7Y`fe^D!n<*`Z$oW~!fuVnuY_n)s0DT7zO+_EzI{NF1j&d@Mv
zXr|-8r5Cyo9u2??oKg;z`t&DV|6Aa}V+5+58OA^T&nCeE6Vc$P5m)fvD-5LI%BpT|
z(tlqnr5LIa$CpRm{#zP;zV11=!bB={O!cqX|M@yDfaj~Em?WC!-|O=8O0O8ulIJRU
zkN&HrHt4~YaXGGL{~vgw3NN_Q+o{0wHxKcTRP(=qW+?JIGrRw-9)2$E4`^VHH)>1b
zuP*-Qy)FWQG5UswNk-|vSE7KK`u|%R(M#jXxSc0hF|o09gdapp0dI%yUOFlB3dk9y
zrwiHb_v?@DyS#ndR@M5Wqb%#Yg^Q!3ql)>={wKG6U0q!v+re3e)BeUo$fa|y6KMw&
z$`(vi>9oGURSc=yztSM>7<>jgAJhe}gX%SU5$!hbmOe?_B3CEBJESyF-idu_!$^Hr
zPEOTX5GQNT%}!DS^vl+fqT?(k67Fv6(1=42+aUL7BZ5kAO~J{Oe;VXU{B{w|4Yb9+
z2{z!?tq3nZw{N}ZvuM0N(U~mLY~P!!o4g`0gr%4L1|k%+d-bPUw*(92Mx#AwQ@WH1
zmMw@NQF<bM&x{M$3)5JM>#k?|n;D;8#;xysQ`!fy=jy%M?Les83MGPjQvEFNg$Wjv
zZ}2L%arxmsEue`TDViMkhIuyWc0K8td<<pG_j3AFTlN5!`>>a(uHgB0K^fPUEjac_
zT#n>#HQajD=UF4%dQsLC$2E^JVZJ6iZ5kOe{f|5Z0(nrzsPz3GdC-E&gOse0_5ZkF
zuc6gCfhA8P{=4gsFg@0iRW~ycjhHA(65#x+#j$x|drSdO7H6H(8y;zQZPr1@i85{m
z7c^N#MRgijVfpzwI}0nofx@U5x|KDFSguyM1ehwaV1th?ywp0*3*R(0oPABc{;IRw
z@u7DrzZMMB-F@Wc#=@V<m)~a+3LVhc4>r9f&HiQ$eqDo0y+QrQeEj>fDGXCi=Yrhj
zyO;e;0Uj0SF|@aXj*d>+RE0@&V_g4QnniSl_)nIS@2zN;7i<$$rRD;er-<A)XUp&s
zn|7d+S%T}uW>8W|(s`Qvu|_NCc6B+p1&s)2bFb3f9L`&R8uJbbaCc1#MO0tVWas7G
z<7=f#GT1#$;&F1Wiy#_aZ&EPKEkx*p=GV=c@^&d<{TqLFO<d=@JzoO7eIBw|uoJdJ
zgSi?GtC7g}B5RH()PuH*8yoC5QvIe~EU$mZz1z>%lNX96U+!a^f#A7@^J(1zqt;Ox
z9-|uiR_y7#wMJj?1LEg$Kie%TH(rdx3MBjR(IRpk0RFkpQ{_gAN+vUYa7Tx;erRTh
ziZ{@`t^wLjwuAmIY}<LD)e<hV)7M_ytHk_457iC8LDwnGFE1KOkjS(H59=IsHDK3J
zc9wj5o~&bR+>`C?PxSENF&}m*A@BF5yX}Bd?P(R6Hcwpstj(xQ`y;;d?j;Zr-{tjx
z*+UwCH6rlb6u?A>1$B#_ysCN!hyDZFl?3g3c{Wes^UIwwJYwlVX8~cJukOj`jpy-=
zRwgWb3lXKIPi|FLd&ynrp;EX6oX|+y#z)EDT0V^o9<~Hu)WE-4bKVdleO$6*(2V8R
zRf^p(wQ9Eq!rjW;x95Y*XNmKgN1Q`4!krYB&cKl7=5?)gL`5W<*lK)Lg7O74qOx1`
zi>GCxy&Ap+9_E`H<TXmtbMb?Tss`JrOqN37StMNF{G!^P(N~tvr*n3A$!BekG_bWm
z7t<TRbss#fF1kn(vZh84Xm7#fLxrq#g9dE1`q8NM>(JY?&|N;NH=v6iF0(_o)a+~Q
z-ka#gtJS5oc#HS)XLfF<b~lG1Dz7%tU@si@Lf37$%YcLX;+id<ccM^zJ&ON}J!rG0
z^o_UY9J+09*jzgw+S#iY_#CGVdT^oc1~vVFVj>-$=S4py`~JbgIsI(#_DYuYyr690
z@y$21o2GV>1sa{XR=u5?L6@TB+xgQ5W$BkMaX}ffO|QU9Jml8<yx(D}P|QKd{j3!<
z@m$)Jy}jL?Gcm;;C+a-7FNcT#?~ykP5SLgYFgO&rf%4c$aVqi%M&3DOqEu~m4F{12
z9|F|w5;oDQ;EYkooBG(3A7g$IKqZv#&4hi7aPO06nx~tauKrEgOyaIq)^6Y4Db>(i
zO#;uM!~6HAeVo<<EZIKtg~7)1ucU-@b<+>^GAH~bKGATV*ExKzvWCMB!l7510+WD=
zG^B(~{zUd$;5b%$ZE9*WowcfJY=08Zj8V-btY)8lZ;2z)&V2n$2r1v*D{Ma$0?d3}
zhKZJUym(?O`T6;Y$mJIyTA$wrKwCd*Ni{TBKTXat9~?TZQ@aAOMO~nG4hO914mjnA
z_i;i#>%yy5wc71$9~vX9Fr?52n8!UeiMlKqUm&vhGF+D|jw9p-$kU3yINf_aJj&1B
zZPerr!{KsaV^FI0(02Uem;>yV(yKRb_HA&?&CTB<r8&K1m1*Nv%-t}{zVwb9`^=)-
z5gY+ukdspanx+BXwI1CbR8@RLn1OY1?^GeARG5WX^R&F9%_Wn(F`zo*xoq>wB=Nr6
zcyuP85q~%Nj_+X!mxJ!REY6fjma>;1X-puO$w-@fnKL7q!c18~T;2VqW;C)VCBwEh
zEjNS7C}RL?h#$_@)^`4yW_5M7(r`E}7+JWFEh})$mywaN67afbOK*{oup0&s*i~%N
z8+{9Oxjf+vJ!g)CO&A;wmU4>dxgJoO7l;%%wFS&@Tf9dDPK@8In{*dpU{jGS(I7bq
zZzsZGrJ|Xh&Z+D@p!u{{G5-3Mgr>sPsQ5;fEj6Px{f7b1jN7ASMNQSQLb~xBLy7pq
z@`vza$I221LmQ=AOa(@Q|4fph*?NaeCgBqu;prAl`$eCDuGVSs_59mTz3GRlUl+lE
z9=@D6V$OllVUPUOQ1#T&t~MhuIi5WwKK_i@lKL!&zZ&)xLhCVA)dOs~eC^sOfht;K
zoGYL@6SFTc->p-Reb~JJfY>sMp0o;nXpZ^Z;`Zh`5h-<zdjwA{nb+mxvqgvd2ssq?
zp#KzSA-`EWL(TIxZdEGY!D~hrv&2&_<++TY=NTJFmb_xmI3UplVxc4hK9e?hZ{{YC
zPFLipM`WBIRAXL=kZHg?_Fi=Ss164|Z@-Lc?U&uS@=@vwrm>;PYxIbVlih7bG{<iX
zYa7GquMoUarX6b5y!)PF9BQm`x$)Ua;zwo_I*EQ793Nojp+>;`=KSCz-TZ2;d!6mJ
z(tUofdQHn_c#?a`cS+#U&=Fp{J0S?IQkCT(8fJVR<hz}vi@Fy)jDBqXLsn;hUv#TT
z_9owhVK!rUQ2YRkZ0#T@R(;7-r?X}&59egTbRO&DHy4cSQ!mGFhFo-~I=h8c^Kq7q
zmK;wW%@2n@qYe(agVVCn=K0Z^A*DO_jpHWuBV}90g1kJ*Gl6fnI^A!JSZ_IAAqYKC
zy00hJ_X45zp%zX2lvPos&>e~xJ&t`uoa9pxIIj)*k9&_aYu;tr(fjS8O=_gyM9B07
zRJ{r5M99EP={0mDo#=SQj}=<1RnN8fSocSc{vM9_L!HXgwDyS1jUbu6<x5=4I+tyP
z&zx_{zH&5hwX41xSZ@iad+a#Kds$FJ8J=W4z{z2^H>-5NYn`7S6PFA<5R6<7zncmL
z#$~7utN;G$GvhH@qUV;{L&gR*KXM8SkN1ei#3SP3$bixbsHL2vwpe!CTM&KL`RNDV
zP)6&XhG*_O#o&*{=-Jauqk8{HakR|N)q38Z00mAKKX^ODvjse7U$J$bRDtOv*9)U!
zJ-Nu{_8>fU2?>dOL47Q1w=d_!h9cdtaF{@BX)M2Iu))6~e<i}C8OLVuQrk!F(qgKl
zdERNC=TT6RW?fumG7ChohvE7SL#z$VYA-dlt`_omq1|YfjE94vW!Om`I_4q&iFi@h
znk$)`={Y%mx23NFEpMw3WM@p;_3A(@n>EK{I)dH3uf+#B@!UrdCwS1~)nf)BGB>az
z&xt+?!NI*!RWH;3@RWmtD{~kPUPk5{>1+0yn7ARWLrh~5q0?Sk2clbRnKz3xw_n8f
zSZ_q%xW|a`i3(63g)-zFKAjsnzB!Dwek#s>)61E>{IC_A&J#+7YI=Jv9_eDCm7v2D
z+oIO(3gwD5nX30Xf;T0<4-2u_J46Bg(+ep7v5~dp)8=wfYDGINTfSEc`5%i|vM!r*
zcE>z(is<5kz5}yJErO$F=4fv&p6=0zq+Y`hK+`7u!l?Sz*mAAro`po8mswgv$HMDb
z9z3s}oj1u_Y?Ow4QJqrmhn!jmKcbb=-5Nnt&Uu%cPmg3dKG5dHWJx(c$QXq@gsnlA
zBk!9F;Ss`>^Z5>l-3A{|Vm*2}^<%ZyY=6Ffz%>sJEJZzm^3;*8JWk7UXXs3`*Zo8!
zojll_k<FLp`^i+##w!_mHTXX%Y`+ns(U}jQUaXJ0g&hR`feas+*GDHS@ommGUJ*#T
zoa@$COVxm*K3go*{BY>>#;j^i{}YW&&j$Y)ahHF<2lOsAvg^#4f->Fva9m3KqmtE;
z&5NymGk4Lr!>sh=sE?KTt1K2sRP1d<#-pPZYRz*z2udheVcWP_Q@irWt}l)c)jl<z
zuMri;tG=UoB%8$kfwdrG;!42l9<@cwAuLcEA<8VagDIkSCQozK?ye6eCxiQn^{n%M
zS&}hYhOP^5X<CK^><ha+c1@#Kge=<bqcZF>;mb@7XF3*CY^b8?%=s%vz#LW6)J!a^
zHW*Cu@EW7q?>sa_LO(R@p#G##b5}UG{pNV`%_-B!*{KW21JyHqQ_PZv`h4fght*AR
z&y$z2<X*#00<*>ztFWM;{E2SITbDb|58ldjJU}R?<BuNZ2$&)Zge62+i&WM+2+WFd
zpVV^dcS_o8?tiwhQ$b-YM<64WN*{F*e!lWamdl!SQ_QzWYj3)u*Vx$ji-3W7?PgZ_
zvoG0BniC>IV6Rd>sEN>7Y701NzEApmu8QyGOxJ*)HW9QN%FIH3re0}Udb;4Y(5>br
z3++f>^9ALyO&n|;-sTIy`z|wAA8F%Ookcw2;%!E43*aPltJW!uPbMhC>VZDl>7Sy6
zh23I3>1A%Tryk}Jy1DfIf<s3#YuSpb<s+B>v?Rb)+I+TpxF*_iwI}jrzUO1_n_L<w
z$Dm@+UAh~vdPz@SnJjq1BTKJufaKWQ3y)m4ErJweF7uH?ksw^o-Z{N1Mh##O*taR+
zna*b9LTvW*KOXVy6r=y(3=1HfAqz`{4&=)e7X|Fa6MX$RU?Nz%eN=mhM#^hiizaa!
z!Om$eA_A`|(h3*Kuhqs4E6|mSV2lcq-}e(qU~8+C`53Mv1#AW4IjyrsjbnsPb}$ve
z^9=9rjpZs1G?_snKMs17P<7w;xP_~|n#sOw3};0bxC0~hZK_na(qe&k6d}tX*`@L@
z^fC<TvR|NYnc&T-SeED=r79~K%JjyJCERb$OhsphoN^imYlbso?nsc!>W}oW2xRVe
ze*ZvsfDtA5;OtLJ_WL(KuMP2n1AL}4QUink4h$UnuOA*=KyHJ7eOYb_wl=apH@BN$
zx2hyGB3<a`7s(vKAEx7BNqU1vV1H<ut}rPC`FxIdE8^=G4h}Wfz{wM_z9Odm=W&UQ
zO)0F24`H-O!Bc1zJT5LA<9S^i^V3ORye|Og6N9W8eu-b>9cqAeo~bNmuAzhu5e@+g
zr=pvLqHmN6RKnjqME*5aaOkT+jgM#0aeJV^u`6qU<IEH^WqjYuSfv82R+sbDXeD)Z
zlLfci`VWLpdjO?atiVfC&tJ20jsUciSg6a-`@O?F1El@CBT~#2h3Zw)X9vsm-0V$#
z!31mtMMVRIgoFu!X;mNZ{~9BD?Hx@~p-0pJN(r7_2`4EjxdSpg5;-Ww$xNEHQ=j#E
z3qTLLO5^do%Fhs<Uvs`f$Bzt2a^DGo4gkG@Ur#U7Z2rqvKw<cn$MFc^7QPGP*CG&u
zS975JxqiYgQ}Xh9)YQ}xiqgaBiQxr*&VPzKINxyPSU9o42vX$WvSi}*fg$<y|5SDD
zK}lv&Tn$RC>BT0NO1foMi@HXrtfr)`_R?rYXUWnYV?Nfjuvi~E7UrW>Gbf6z)WpqE
z!|)Z!nyXHa^|7I1wFYRCo2#Z^nu5(@wddo5cH|G{gSp?i=ezfJe&?La<=oDoooE9i
zwgv}|UMssyLVBkXOb~BTR<@zMuflJ=y&Vb->U$I6^33?%dAjFHGXH6&$G;x_s;2!g
zt55+Qh)HfNe-s&PTkvSxw-9w7N4K(Jhs(nh)tT|$q)SN#oDG^wBDNmf$JAk@J&4!P
zAj@2~HV(-0G;?gAsnNzWQS$H18%|^wItFlG2Pm*1XVMyZa9$AhX7bpk>Ma#%h3=;w
zy2eT+w<G@k1D)@Ubm%{04I^(cG`&!AuI$E6vfZ|qe#b2LX6Nb#bTKrjHN|D@0EM%C
zUqN}*yNXn;<KSWcT+1VHUTPi`4qtzSvQH0lgNt6E8y{1X^1!0Hv9YoALZ|zqUk2fy
z!hX0Xz188xHBJf#wOSn#6m;w?c>HEYtv5QKwX1kOmVJ`_ClS?aFs^a^qhgOyZ@&E=
zt`sOb9thClxTq&Hb73XqAACMP#PRaS7b7m{-5E~Qj2g*0@f0!@E-Mn}rrL13<<;Td
zP}uoq5lbE&!eB%Ld!F*M3XpEha!AoAYa0?q<ldj=IFAODD2S;?%xHGRJBiG#$Q%=~
zYpkq_%!^wl7vr6TuqyIR&hhS>s$9<lp`qi7_2X^ed4#YC)}?u?C{{Ns^%$@-2N{nm
zn0pZ!V$2+_r*gEsIQ9NZD62WJ|H{>?GB`cqgbIqj*ToX%02vxM5kP*;<yb5@ahdYZ
zdL4L7zYqdXrY&_|lm)Y~Bb>_VYMFcY#0=X6TjgRUyzK*Z5OK-xFzzZJE%O#a<?F;J
zq$FN%TV-{#9L~zHi!@q&X1jW%Xcr;&yv{<YC5U3r+nEOu#r9&j&f#%zghr%K^e+(4
zO&;9$`~-dWLlfkOf?lj6En=EqSyiX%6L}OYv2qpPrUzc%VjqR)>S)B4MhC*m+1zhU
zgyCQA$VTE2Sd-+?Dmpk0eZaIh?wgKuGn^6@2JVu7=oKEo3#re7D;lF_Ctj+WwG*U!
z7Lf9KMI{&no<;;Z4OxGU2()3tR)rq0<SR5DU2tB`g9NRtsx(HWg;O(DYI<<@2(DzZ
zeJ+7NQIdmqu0Pbsk|88fT3Y&7DiP&f-Zd=PbixdUg#e0!y}XFC+U00^XCZ}DL+3Bw
zybU&x3hHITXwF4z%G0H5y_5nEsGM8WF#*dAO2^m6D6mZD!q1rxesWAisM%J+HhJ5u
ztx`#A*A1>BuV`ABZ?+q9a8mFZvOFpv;G#$dKco<o2)^<5%M6A=vwa`xb`l?g`j7k;
zDs%i5BeE!BdZas^^qP_|?A}KY4yMemig}uR4-*Fzc3o5>6bd^%r6^YzV?N?|KJ6aH
zI;(z!h2ir^85f9FhW+>;4a*w79=U<IM?D0;`Xhl0rP2#3*oU^HHo*x@?%fW7h1X*6
zTmwU$15_DVTIgm`SWJb~y7ECdV6;5tpAz3J`R_^7-^Kmfu+ZQ=@*AMNSRh@AbZv`4
zVr2Prw|8;=GnGgY=bO1?*Z16spYIn07Pxlyl|n*UpRA7g%Gr6AJ7v&aZs9Ia1xQU0
z2}bNm43a;dnG%y<gf%5IATlGnxp~~m4r9q8Sgorghi=$tQC(?51QD(?*6s)*&a$Cf
z#mpt$EP$mom;pDsxoPzDtP^gGIqD2yp~h~=NI&*^uI&eQA0BQELoG}1$EixdN2^!5
zwaz*|zvUOe6ZiK^5nj|gaJagFgiq3}{00Aig<LUlqwl&~Hi-_&<~OI>b^#H$Qj)LO
z%oEe@3(*82QJMX@JA^bIGvM38TmG{7VS^{4HM^~9zSf1<kBiJ3M{ei*cIufCymoyX
KxTDJFaN2)0$)#xk

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/non-identified.png b/docs/2_39/stan-users-guide/img/non-identified.png
new file mode 100644
index 0000000000000000000000000000000000000000..cb6f2bd7b39aec8c22d89106771571d7b03ea213
GIT binary patch
literal 64857
zcmeFZbzfE8_CBmgqk_^6A|Tx$jihvUNjFG0h_XdVX#wf(X44=@w{(kiD-D}?=Jr14
z+}!7UpLg*416l6R+H1`@@*39|laERYQV-CH(C^&2^FaEgxXPV7_X5EW1u8Q5#Jjzy
z{mvc4r&eNOO44Ft6iQAG=2o_5ckVoY1W{jfm{r3O`skpNg(pm*B^4ChQL3)0TSrV(
z+=(n5oxLlAm%TYf|8#|q=VLxG%Is!MfueT+5Ax+*k?DJwF9j$cARBv6pR`s<s|L{*
zNWb*n_=ZVIvx}s@)Zk`JtNKmv_@a+@k$0Vc-C$N+(9o(nkx>0ZTwzJ_(9_uHT)dbR
z`QY!BW3u<g3O5y)BKrm&Gk?WB9o$|b4My$TjoiCdIH79s3Yi=E_(ZYQ9+%XY@a(>F
z|CmjUEKUmw?92*^Ja<U)Ym^%ip%5Y7_se6CkuF+61xjJG@L~FG++x&5WP!#ha~7zP
z;XQ)x=p}V~;`%4%k(I?!MovRZED?@Swi`_5B-f)I{dZchFWGe_wU{3l7*5ohKMGG)
zACdf$Ei){wlQ!~4I*YjY?t|b#u6&dq<?FJrE@fy~jI)5efaE>$Iu}D8X6?%6O+rvU
za!kGbxtr?wq^rf_$}Idh`x<9eQzjdcdEKg3Ka75SmGLb|FidpRSPPu*tJF#8N1%!@
zZOI*2d4uvsxT?4@4{6{#9}h<c-hr&9jeSY}i<y!{ZDour_5^-|h;BN&hS;_k^1g@l
zZs~-5cru+Mn>7B#=3W+$W~PuDQt)R!rOtofaX}{VoDy#2V_8EcnbjXtF|NsMmD&&&
zr&7RUG9J&D;m=Y^VUPLNB*`s6&E6ddY40;)d!qRLvF*GkZ^Uv}U}yz?6H|tAHh)*?
z>z1B!En(H^yO-p~l{>7&7T8&3JX1+KsIp)AzCJ`kce3Ndr62Wuc6`{V=&FXtr=WCj
zDdPL=gYw|?KGDTM#P$2cGFgTrD#c+oPTgH>3KVQhSDx=exi&V!I}ZhVksHnicbjc&
zTrM3C!{$Tmy12{iJ8Rm`rb^41F6~qh_+NPi$EElFa%)5LpduC2yE{t#VB6t`_VA3#
zOnGMZ)_K4%70p-SZlV6~bxXqq*BRIO_IZYR_+yLw)`5ilIZu7*eHpHv`O^BbT*(Ey
zXxBfvda57W9LbEt+pyo1N<HOWYG{N#AFOXVZj;JItE$lJy@J0QbJ=)VAwXP)Q)fsP
zox9TXYX|LQ;Qke96M+kQrL}yuH`8xK0!~b7<YeT@PPLA}LKi&k+0^J974GiNsUxaf
zH@z>balsyZDZ#WaKkkPpe(U<lfdEZuJTo|>Da*Ncd>xI=%E$M^j!}-fo8xCPpQDDv
z5xPs#@PNJ_^5Vk`=|Z;&?p|EY>C%EyIy9f&T6X<$sUX||PTZl@XWZ_!O>H5WPn+L8
z3Ez0J0Be?To4;?jrz;+G)5gE8VX2#k?YoP0+(I<-{+EK!8b?M5%5i<u&-KigE69<m
zd}WI;(oNZ)^DfX4k2k04L$u=j<UB+-B|Jp@hEHzrM$8U{weL83*b*~f-?<}lM_T-a
zn#bLpWR%9I64gOAEJ@h-aaAi6QPC94g^x8P#Kk{7#l*sS^eDzsG?f4g`<IAXtAtq(
z1p+!M>XS`D+<R#hjy+aJ?PkvP{H51iP_F{-gSnoo@oTr7hWh&YEq*Rv*b%R*^~+8g
zvA1^+Del}w6}f|e^#*<!F!pFeT0dfw^`9T#MSYI&Ebu<!-#7mG4juIy(a$PhNA3RW
z^Jjsfx1Ra0TUj}?uk1acH)mbKi54lIG}c{-HYHXkvkonmsSi-IB3qXugzfhl_L~dp
z7e+@-Btr?7u1gyadd=ruC%KYrD?So(S*xiZE7dlk>G_?nnlH6Ml&LVE|F}~*V_hf#
zvs$cOiIVOA-9^67O5#4}xYl?$Xe0YHEN;2VYPA1nM@VKfxy$ei^Ko^}u#;&^(g20#
z79OuTZdH2OcxJ9y+sZJ{>+|)2&uk|BVwi%a+*7aKnR5r&R8!rRAf)C>ZS}vJ^rx%0
zIr2YP=^1c9)C?3!dQ^M%GbD3S{0x@oHhUe}N8&oBoZ!7zM_Xy!5zM`9c9=N2xR_dZ
zu~nGl{uSB4>gTtQ<>kLmTW*>HG`EWi3_Vw4LwyN<o^8Owcor;YACA#?$#fkAqG1~^
zuNdB3XwFtxcD+~ct&6L1Ke%d|aMZ=fHUE;?f^ogSg(7_R)m$&iri%Bjp5!jeTg1Dl
zF8BdR7<iv$S(?Unnqhlub`IW234E^B?yhBT7cO@l{5kf$CL8*tX_M7nSOvSV-_=H3
z^x9^cgp~U3+L#}YYwoY?tyCwMe+kN>Bus8gfWdxr-29I4pOZ`E>``~=B6TOwo{-+8
z0V=mYAN59c5e%?M^_>9Lp2`MRJK4}&x|KFZFASR=suw<)2#yOD%B_o@)N>9ypWpXe
z$Fi*&kzB(tJQJP#@rsd0Dkiy~*Y5PxEBu4=kRVUN-b%D0H-~xuhm+}-<agJwaunB&
z2oOjmepHPra3s=dF>15i&*Xh18^_32X+FTNDDXS*97r-*^Ri*_s$|-+i*(=`wx8-=
zl*FFmy~k^!Us<0odz6hDkicQliPO68v(9hNY!~^HG=4zKKD=9}CjsW|Y^t|G->_2=
z0`vA*4&zDWbzZCG?VdFB_f5}O$$oPO-3UEU4xe(k2XT;Zn~}qIs;r>?*N=CYiP(w)
zBbTcz{vVUw+t6q~3!cu6%4bE8zRKN2yWG-j2}Z(*WOzLjy|C4CzLxZ`-8{+8{Gs5{
zaE;Vkl34?BdAnD+bBhhGbAHRji>y5JF3}?)<BOMTJd(Xe&=yX+tvmZ^A?a9dpUN!h
z5h;`i1Crerrsq7@lh^K%pNYuP`2bxFJm8so75@C^hhF*&wOOpP0L+;jrrqIBw5dMd
z^?+R|t6sxeZW^+v1?i`LK3JJDzU%UJRoBsff5Nfh>SQ)4!tW?kfo^R0)7uZCQ%J&5
zMD<E@n1%0qpCf@lAku(E!0%R6w~-NYkZ6)wxVjp7H$Z{v^&szhE%d_r^z8TYKmZEa
zl$Cy36INcrozv6Ak=5k0O+|mcUhkt3sexVSBH3<bBpHYqn~_|C5RolZTj;!-Irs@V
zhLPW#N!g~7qPh^>Z72#Hskao`1zxYu<vx5G;(!|Sr1h!^C%HGGCtA|x?84<_+5#ru
ztSWvde&pe!6RHRS<2V+>J)J`6*ChLTt1%~}(NDx**5sRjE1FMNn3*>nPjEQ+!)C+u
z_sdLCodikx{Pi$m22%uQnKqh$IvRcuqiY*7Lw1P-y{e{-LzpsD4ZL@`sBA*<GGhx<
zW{Q#8$KC3lTDE-8f~f9#IhEAf>-k+P85*WmBGBob45y3D9}8bDVJ5jZA10d>;z4gN
z*M;|5X7X-~8#{??YTbGz#P@K--ruLwjb|BVB)j%ng09_pnc~$z>v{3Ps9Bi&VmPS(
zA^#3mii%sqqq#nVszIch1uq-t{iajNYU{z+x5M|zg#>yTYDPY=MK#8)@zSFX9&L^e
za105U;4$f#?&w(Clh8Ask$LYdokXugV0=<0+S12?(k6WQh%L*c`_K4GX6*z%My{+y
z?{U?yt0fXq_Gw%R`5i>3ep^$u{(WpLy)3%iPRk_H1VZQee8%G8+5m^5;IW#E-sy|5
zB$=ou!t^g8iuHytltCva+izMYyTcA&OXRJue+3DDjyLe?sdgOTn&jLrstni7Si9Qy
zze$_3CuZx6YSE{AaiOaz{eAJ()$Xx$Uu03s^@Tk&^2^n6;&1(dD__=@NvGu+HA7-H
zX5`U$Q{7{z5=-o*Y>x2t?})XfAab`M;hu^ececAQsnjt9gmGp$a^@Nva;!|e+2e8c
zB{qJw%nr$FLE~>9LLZbwmiwb-I#;E4col!gVMr1<1<pCxczv;5ei+5SRJ}`1?><%X
zD5l)#nuDRZN@e7t+T%d8<zhV4_qn2m_G`ndU%QpTi3}zV!-Oyt=HAn;V?ndtm{8v|
zI<0v2>Sx>a+a<cc=E_n)cz$l;5Jku|p(|~T7g5G!ouu!~VsbMazZ>Bz{i$5}Tid9t
z{bZ7NPssE;2p)5dj7Z(644Lq9D91r<3=wg2g=E;H@zTZ2Ehy%cryWLDYK>hWSVch8
zhzCpDk+(vMrnUd?pB)_qP9{yHU8Aq?-rL78CSAck4ZHY1*ylLu(8-Ari1xSHslbsi
zBPTlpVfhPBPLUHK9$md1ZtGMXOXDw5wVE#^DEx6^+!qhdAe*~Bef={wrKC2R2Rs?e
zEoU7p88sS}z@&0u;uT>h5{4ZE7i&B%g#0(3O~OA8k}|I6Wzci$`1+Ig?^n;&GQCfK
zv@AMFBxi_TeGF~6va>BBaho!N1V>JGAabGv0NwNme~+eBWS$QG74bzA)&H!Wywk6>
zN0?-(*%yXQ2hr4XlrSI5Q^rkh*eWPAFB&hmoNaTHrmQj=<~Rw{mvAlgRjDmf&$7I_
z7CBg>l^EA)4mPmpV}c&#b~f+oR`+`^!WJ^tA^mN;`tI}9Pgbk71$=U|lv-f(`z7_l
zZjbA{%Z`q5U?upc#>cdW#J^i#;r|HxT|RNRVrBvhBcrOH$`v}!*Gg(D*Ra(MUZRfI
zdu&s<m_OOB&mWPiL)J~i)MtJp^wgmV@qSk~FulE?7hfG36pV@*kMGcFF(;fyM)*8+
zODVBpitaJ?gt<+}PkDN6KFV=&BH7V;xc}9<vbxOd$ce*L!yeX6716I15mPoT0VU?`
zR6-3gj(Di%{4ZIC5pkA{bUUWY-p^s_sj)U>iR}y{#=lZ1!mTfQk)|ZL;>rv%-b~B=
z<j+;zC-Kck*;=E!5Be;Peu^@O>(qM8R9P_}WL%lxJ5yHh;F;i7zFqGLJ7O?Am)YU#
za9hDkw^UDz*@>PD1hJA_Ge{4wY4(#a^JK<p<GSElc^BECHm<MDZ?l0+y120})B2lh
z*u3%lg+h+HhIU7{{>$wHEi+!vsX6yWUrFXFjYk9g8jP7E0xRfQC)fKo*A52QE6CZN
zcb&YeWqcPLVcf%YPiBM>JiBDf#@d?Hd<UM>Xr{9EbmVDScM_aWK%33{&if2I6Vupe
z-399S$(_0>JDVwbov~y4s(jlW?CVzJrt18pY`C`k(-&KW*IKT4m)xOYx#r#ME4jPo
zmqPXpyERj0vUMQ^KR{&0)vb}(cCog8k)LIS;#4e`19Am_LM$|uBo(#`H8e<YBJ}$*
z!57zl5y7|h%o%`1h_>V@%&um-^XTNH^WZA{TBDZ<s%PCnz%ZF(9X=WD%r#kTF>SrK
z-?PoVbvhI5$Troy!8_cH7>qR?|7fkGVJ{efkh=kc<7WH~tMAIOX1Jy^pwGQBo!1g9
zI*T+9*N29_0GLPEv{Wwk{DQ@u0EWUk1I2ti_+a4rw>#@<@DpgFXdJZV`nVJ+fOfI-
zVtZP6r9@v^Q+8KZg4_BXOUtMNf2<D`ar1*p|9+G8mVMv-KA+zqhW*{7k`vM}Ddxjo
zhSF~{HtQ(nQR|yCUe1?Cql#U<!9mHBv5_64&>{a>*Jfk0X#Rnp>Dx4mp#@<YX#Lv}
zy#*1Fg|k^mk!!6b5#wnFE<Uf41IAJ@bRm&eEG$+upU`9__2$x|Nbn-LyVz!;NZV7s
zf(3^i4%v(y?5It&TcTSEf0@EWY1|&boG1kJTpH%PRJj~Bk|<>ZY)|&ZoZ=_nw~$ZQ
z2~*EYG?lp;{#xIFtqb4O39J2@ZN3u{d%C_gliis@KkIusZ?f(cFYljjy`t5p#q5ST
zf%ASjoS!+`#k<-WjU)B!8?CXOhJRNmVYGJb>vRXV1$T~4Eh?r!qaf=gk`%*DH-`Ov
z>*ZH|rLp0&=Aj{9PqIHZMlB#7m{Mj3-C@pTiaLIUBYM;2SzJBQ+s%}&73Cell62X;
zQ_&Zk99NE)8%oC6=G+Wo1s<c}M#O%1Bo&A{wqFl~R43zq;u&O)&4vAv+#4DpKjYu4
zZPz69#W&9I47wKZE%X<@+EmPRMz@=iJ|!d$=XFaNtacbc$94A{_^G@RAn)uaLtQ80
z3!7aGcZ9y+W%r_>k3;Tj(rW5B*gkKifEv?@rO#!*xmvurGW^=dV-k9~AtfAK-Bm)W
zS;l0P6Em~ousY;_?X2~*M!&CJpEfjChP)dwldxjUHSr4evRSE2gPAMSw*JRk4CCcN
zW@Y6AlI?)U4CL4E$KD(FAfE>}eHe>*V_F~?ULCqBditsfwK~8wO8(=nZlA9~YB3kL
z2F;vRCxIB{{f}MH!lst(ov)V$45ZKj0{nNovR_Q?SvL2YmfjPuJ**f%MpQ!8efrbu
zS>S36bJP6&MH-m!77EF0k!b@4tT2|+{mclV`X9WI>$5GRjHnn>3Jmjd=n%wy1zko}
z+UrLEim|V<KaOSA_blEd%8&u>Sw^aD`Sh{-WX7hMkj@#k0mNTJekYaX&eTE1R;+U&
z<J$BsWa#Ox&Vy4Ot|k+&S%m4%AS#ueA6!$z%b_LTA&JrPv<$WT@0wR}l^E4wSo3tt
z`QK40%zIQWK|Tb+dA45uqw7N$9|^JWLENVbWk&7Ov%E$Ff%hgu@3UCkbZD%Mp0CEJ
z9`tI^OC@XVYEAG$_P(Ou*I?e<H)9{xVsm(F9>#{#=953}<bPUyf9+JKr`B$wR>>cE
z6n}{7y!dwx{>9S%yu!9Lo=J-=S{hc(%Q9)Ryu>v2i*h`S#HgL{WPE&>V_$(r52jv8
z=92sidl87rK}6}3h0XW+(ox$A*>Mpo14Hi>SPEjOOOsttRFXiQRcu<G0dtCfplvzJ
z>|s#m!{PC#QmB?<ZjNQw%(Q!U5X$(@DAo4E=G!a6`=@Vl<9Zmgu;7t&@QdHYro_!G
zZwd3uK2Y0i`r&sV(9k`xqnwoE%GWj#Z<+nn@MKs$$`tKp;k0?3aUVU3&3$h!Blln8
zi$Y<4VHL;dS`4yXl>K6d0MA&BsZ4z<ekJ?1!DGn}6SWw*<U38L3q1vyzDTubT~NlM
z9<_WDSA%L3?e(YoUt@KYT}$+2+E4H?rbdg{;*cR_gs1L38dd<z$#mLPc6Pji{d|RU
z{<7J2Vi84!4^^$^b6AsD)J({m37W#ER=(M+R=%Q0DRl8w2FjmzzBImuZk&KRbcJRI
zp9za(n9mqp)EN|Dl6B}W)2b(Xvq|0O$}?P7)FLLv(4yb2s8u5W9OU2MnUKFfYc;RZ
zP9I{=^H!=hCxzJ-H$6C^u+ewf!8S|XB=-Jp{Shb%=ciC4TXc^ai1s8H@()@u{eHcp
zAL?eYU-Tj(POp6RhBVfGf?$N`wGRFu?Kb<?#466DiZuJgjbxTRvEo3K8KpMo#c6u=
zVVkKTYhVL4)U|Y+FJEZ$)NsR~u$67XdMBjrxX14>X^~)_DyArEeZoe*1d^L3Wz<T1
z^+5S`mIf`-cs{2#ex26HYo?B0A`D$xQ{nbg_MKNv1i4-}2D-e=TjTM?H8Z_|Lp(j+
z3T$yJ?m~AnLf!K2XJ@HJMPXoRX524WC^zZqEWjkkGDdmKKYsapJiiBJNhV`zQ=o;l
zsYaHXSgzw~thNMpF*o&)a#pj93YABmqLE70h_$Lyz@}+($5vsCcln`iq{E~EXYfT1
z#2q8sJp7ShOpr9jh^$}g<qGY9JkPvo6@f>ks>(^SKuRKApQw}d(SAEK#^VOtgKIMW
z&r{Oj{!D`g&miX$^u}2(?}L*c=$K!hKwDOmt?xM!+5|)#;Fb^T5M^Fz*i9X~F*H{h
zOa<nD5}n#&!>fs~GwGdQlw~Ve->)8D&M@$mCb}1O`b9;6cVZf`49hL=GD>*Hc>d-4
z-HC(8u~|AdpJ*4$;u_-g<pecWR^tuBckySZT}I_YNP|~r>p04NchzPklcUwLQhg4j
z&En<94GJHV^`Lm{YCe7<XlongTrpT?TdPlZ`s<5^?e+_e;IL$vb=92ofjd@BeuBx$
z2dZ$^#hdF>c@KHW85HAd!wvV_GVe2uq+|bCJLrR?nNPb}JN%vk5Iw;nYyWqAr4x(Z
znf~NGL6s})Ld`>v=HtgN$*8W85?moX+^BL3-!a@5JS|)<V(3)oYV^0qrqa}vn;9v4
z@KKJn?4Q(&kn*52HK-POZL1LQo9c6$z-FwD3Hof5&^Zd~TH@POhZb!LyCj+}qK;Bz
z7Mx{%uV`6di%{+;om2Z)#kyrq9Mb0-`pd0`Yst&di)2sW-QO-gveP>2)LGxYTQD`F
zYIN1JyK*mMf~x8e>u^H-{*VnKrani{24KE0Sae6-ROp^*NN3N5Bpb>CYAe<7`*#X6
zMLfG6jpbS7D>iNs<&JpkHG*<yo_W5v4SV&4<<8XyGtX9)n7>@GOqvU}+SrK_3w+xk
z;a}(8|MJz;W!A$whY`ckVP#*eLZ{xvYNuj+Zs?dlGWJ&+28)Yp_qG@?4|cq9x{zbQ
z>EQ2r524B{51+fygz?iTJ}ib1=O|HD$#U1a>bEZ!&rBq~JC-XZGdbRo<sTd0fO+Q&
zO&p%iY&j~IHa|6Fr%vU$C6Vc4l-ZKi#8j=`$EtZ>dw9cj`sb1RpYeAoP8K|+cKBYc
zh2fH%#SnRL^rH&!8#a}%tPwE<%ebD%rwTP?STn2|D4XHpRRoY3W7ueV=yY@>3QBTH
zi5;SHwH!^hT)WyfYgCL>tW9w(&-s^6zCF376x6PXyMo7qII}2EDt~egi$OVqk`;{n
z#x`y$HB?u<z4UZZ0v#0BIZ5bve56$H8o}%jRr^KcEVdAQ7VgBP!Ys&MJ5rn7hOfyx
z{yLN5>rdlNDf#0jMJc{kLG?tw{w(8NUe^Y?g79FFC91X2tQt($#TO4~r=Z>vosL4n
z`J#>iHHW75EWe&$6=rp7I_c&Yip@qW?xg;DhvrdqyhH4sjgJJ5u;S#7y?E20H>;iM
z*>ll$w(D0Kp^qAbGRhxG4%Dru*p$@%L^p~5#K!GqX&ONHO*R_FMLRC(%gIgRm;OY3
z-4!Sn!YH;8cIi7+t=cYIw`;v`QE{}{yt;3(<nuzEV6P8}$)|fP1ho{)dT#1iAk3Za
zLGil+i<YaCQBBPpjfnD+z?lq%@&BlApANFe&-55S-j4MyP8;$)1=Ztf@l2VNObWD~
z4cB)@pN|SzR@M!rui$m7LAlf6^Jw3`!ZZ=dgD+1p<<l^uah?yL$X&Cc_m$}<g82Di
zt?gyqsDIPP@q2X{!dtrbJv~o(T3CEHWz9Q72|G(r;sx}M{Xh|+(OY4Crk;`(k9yJM
z4V6!3E)3Ln%lVnO(peIZ4w$_reyP*;+!U`~q_yFxi_z3;oblKyaT{S0XD4E&EUc|-
zCZ2VBd0>_4mwjvoWv^b;Js~N@4$z1Q;d7<JBam_3a=@V@Mcm@2q$cg(FnfgHX0+It
zBpsQdT%#A(rnXD+?bn>>Y){frhD6B0+7DblmV|gU)Z+=pO`S-A+(!agf$yCuj$i*+
zoS`ip@-su!;gxK|lR1ZOM<-g1<|Jqu`i@^7rrvP6-MwhML(3=iB5=nmYH5?p&K>9o
zUwh(dC_C+DO7#6tDk1!u>@#_Wev*AMfn>xB`eQt@vg3gh2;|0wr*%q^V|tUO{Jx~~
zzvj2t;;Mec7#F0y$Xdg)Jg&{tWQjh-!D;h~w7c<ALY*~~Bpq+ddaGITyXPRuq1m%e
zwSdnapKee$vlNfP=-`VC?xg8WFPD&ZGgd^Fx>WGJd^dz*GAf7;v+O5*iYcF68$*u7
zb@v&o+>T(Y61!~Ks&5L6Wux-T`J`srcRUBQ;qkLwml}5#vTE`ou~C$O-qlH1w#1b|
z^ZNBU8r`SFecwra_qy}I>zVl-<^myZ=TNJe7vIBLh#;TdQ242FUpFKyXGAthFQ|g-
zA!(3QuPb{8QdWK0u(!2D662Jt`f{JxzLrv>=n*D%50Cq#fw$dGV(gA(v*zchJ+-Cs
z*A0&ZqA-7*&N(Z3^aR)7iu&DLow^6e<7BI0x>_j=5KyV?FxRa-&Yjz>91@ZY`K}pt
z^!hrJ{OTEZs_Z(op^BP{W;**kcS<5<pD0OEWW3LFS)oSt@5q9+&Fg~uAA`59DXFO%
zBR0&$DH#ks7DumiL&j*=F=-}eLYc=ThzCeada_yRW9u5FIESLg1P}&D@WZy$oqVYG
zdBf_Nk84tZ&u^Ozt1~@c|Kco{eTc{0a(STj!CB&+&9}Mem88AYs~rpb1^w(F*;VvE
zq&`0s_|*z9lpB>tH#LJ?%hj%f3A64}x#+A{p?@c#MJ;3)_W%Y6b+^L_^jLXW1CGva
zP>ixT_p@~ogniU?Xd0KeUu-TkOp|2$7+b3RhitlTs>1MGnM(d_rz(f3cCZCi1U32K
zW#Zd$0}dDB@hhF)ANMc?u+*bn@Q+bMAxfTJky+(C(Gz@f%+xCp;|Kb?S|zJQ)n7*Y
z+2ZuLGrv>6CG+|9LUohKX}7lyZ@TK>i+Om6<S~b*qV$3@k*ofeogjz{6MbogI%&cp
z@OrE7M0&f}_q1U#EH>J;Y_73!3JrG;j(jEW68fkFHaY;j<p*7XF1+;ztoeu75%P_d
zcgjT~WuvHMsTX6p2Y8G(bL&F!P)Y6K(5W>zV+EzC9*WWZ4(|xLtDet?vSmQl+PkX~
zY2jqQ#y4H&jGp+%khzcQ+A7Lw;yo(C7JpcXiYMt@X~X=B=Yv|sI@jrOIP`&hhNnM&
zNf@v}Ugsve1<z-)G##n%ERI9d7s$xQ!_5}MCH;AvL#Zd@M@g2QtU7{m|MWTPj7NUE
zt&(t+iE$_LG7|-evxJa)@090R*D-AKD9WJSg{6sD-@QvLAj)tld9yng8dY3oFQlxU
zY*GAxx)1iwZBKbez~;e0kc{8WajAb;GW2&RF+O9eZTyV+BDXnn{YJ*y)mSU4r{=bA
z-fCUBQ7(r*m{U0-+2+P(xY?;3!ei3ic;ox}V{)mRj*WaLasr*&-0Jj3a)LXc&^*`W
zlliJQcn5wO>4U6q7CRO?qerhD@@m0z!nJCH?%k?E)z2p=%*|&%8&9;|Yo26pCjk*2
z0OnC(LYSD8+hp=7<7}iNAeGAxcSt5OOxy>*@RnB)ZM*r_MNMu5{x)4y>$)&oFT>ph
zCY2w;13NXE4>Y;Y=~skQ`z%NWrDo2lPfMSO_M%isd$nvMUt^QOBW197WnYEMsmP3x
z7StV8J?<0DX&;a6)|1^&f)1LKxH{42vF()d8&A&e6B9f;gfny#m@K+i0}L_NOL}=J
zPM(m9??teN8Sc-EHrdNH+RMhQzH~IS{Hf2m?Tgs8ik_Y8zZ7u@mu8|E9zajOSvFz9
zB!nUJn-mj=m&M@n<FlAd)&$SHs4diw#KyF}a_~C|voimlCG9y`jnQXOBx~8peEOrL
z)G0#v+Cx$%HLUfs=5^Y;hxF5fEA1aF()Ya%l&Ao?Uj~G>ZPr=;jgPurheuzxm}{cv
z>!5BQLknmnnI3)8W_%FmUG5dK3Oxu&Cp7sc891=9D)pRs=jTF#s-hqlhwsJFRC|+w
zhE{xIKOn^L7)qUYFCZOQ^*&+mei6A9_x689A^%d|r)wI{x0M%4MiiY7sG4j{(=AOP
z?QKrjX}S%ZNMoB8IvN)1E#dQBWib3WzTDLZTC4^Y%yd-EwrBb~*+vx=)+>3=g`|Bw
z-mf!p+CXt^voxeNltAEYQ`bY{wUJK9FU3k_reHRvxP29=?Y&!_SwU=9rCi0NjjM`k
zv-Bm#h8kAf*iz{G?z~BZ&pwF@PV5_3R2<HjEU)S=UTBnOthiGGcgI7k*2NXw?!;?I
z--!5RueU^=|BF&|mNv!1c;`j!r<YrfhtV0DFA0x#%k8rXn;cJJJ^PqGKkg9{cQz;;
zW;Cv}zs&U5cw135^8-_2)@J?<U|!0$-{LZ6*65sm%G-r-Z+&f20xXQqRqHVDc6=<a
zG}&;L-C55nLP4$B>Q!_yO+rD`t@KPa`mD>K$<dHRQ|r^;YMb{G_JPo*cNT*#%&CSt
zJ3Eo^^?o$yjYDHTzLq=Nh4&7rw>^J;q(x70@1@I)5f~yb(g+_9UtJNbRYp(07!761
z+4cdYI&=42U(qo1I%N|{&7&wrREx=?x^QM}3kCx`)_CY29_uLk^V|+e_moHev6*W5
zMUR66LO*4dyR6C;s-Ek(`WndoK1HGf{@;D_O~21^Ni?X%$V#Z#x_tk_AYV1<r2~n=
zgT-O`vgg~K$S2eRtZQ;;5w(iG%M%rZm(>io-2uuV5|xeblH2b<U(#9fkRsYD$*#un
zs@&eBM=XEAp_4$_MKcWcS|~lJq^^2`wEl-q0#Q)nc0R9okJ8b#&y5{<WB+?%`wjQQ
z=Heu41B1V~SxjSW_9FgQJM>1ZpQi%Y`QgoEW9LHO-^Pb1SHA)C(_aK?2tS^SlGDo<
zhK@`o)2Xu?`@ozAehd8kKt;b7K9l`MnAo=LMaOy-$9gdT2n`ajm91(0tKD){NCy3G
zpR<WzUWPAseX7WEYFk4prQm&3Et?qcsjChaKjmLWViHneL}o&OtR#ZrF)4c9FR<>;
znZQ)bI|7}+9zgstQ{yVba;7DI_<$y){;LF?m<h?lvb~@+FlfMr&YX@~d{S&AZ9g&J
zujhYtycQ`%W>n$vNCo77+Apu)mY3AqZ+=`@Sa`zJV=n6`@Z-(B{>AGP9w<dq#_u@U
zi6{GHsJkRnZzYOjb6>%+rg?V-MK`kfjHjLC)xl=Xti3af|D|8{VtuN1p;A%p@h8@v
zR<3e{d;C3VIulJoHIhc@&ej8ku8fmjP1Zv;odlzOqkYT`{Sq!rxGQ2#WfS7~Z>+bA
zp(O<jyMuKFgH<n<0P*SN#^FgWjhs@jUY%Dch;~2nZlz$7WFcQbt94<qx%ogMF4uVd
zI)Qd-jA_F6pVMAvYE|<5h69$#DaJN>60w|)Vs<U33*KR(Og4`UC3Hu1)wMO<<{Y~!
z&^0RZi`R_SlX(l%8Q&R~K>a4{#<Y0EPYv0hS?J50(CxyG$*81Lg1s14e|6v+a&nhN
z={}Y)`f)6hUj8h^jaNcg+oG0byKxKZ95yFQ#Dqa%^NJddATrIg(MG*ucn|wIOt!VI
zK}Ve3&E8V5z<!$#nIkS~yxcsLZJzeXT~+{xRO8f@NwyNU*N`M*KoDy(g7&5FIDI~x
z$3*Xb&XCF-hvsv0O^5;Bro<VZt}WYVdR}NdY8sZ?=hm6xx@?>@Hyay{nR^lDN^3+|
zFL~|~<8(JM4nw2x660ad(crD@YQ#U?U3}#h^Wq077J%HZw=PWg3BqnQt7=Djc}NML
zexcAutTM29uBm&73aEu1mltVJ@Kt@(sciNqW=gm-t?I7%@iKfN0H#^=giMJ3H7x@l
zr7-1OAEh0brFj89%6tFDJnZTqz9m#adAMM=W|n!zf=Un&tfN#mUg&?OVDv>rUX+Fd
z!a0w}tIuIFGvs6%K6axN{zgA^K>)~0xxCc>YI}s6c=HLEvtafd)%`PCv+)=q_*s5;
zlnnY4SJ32fmmb<cMx+?FL~s+-J&Y~&p{|Fg16!Pq&oK({rh5#3)B8W;JP&EpC^Yc(
zU6jW{oCkTkECtmc!>4?l5K9^lWT1$#y>tghivAa)PPwleLjc+7R{naE=+Eqr3r|3V
z?~cjZuo$Lma!lU_wE4JrZ;!R<-5o?^`Byzr)HE)bO`z}(GhmJU4)8c@Q$_CjpCO`Q
zBt&?tSv+K1F_~{%;CM_`f0&^j$eA;^6&_h9n$@qo(zEbovs|!Wk+!+?`42a>tfzSQ
zf6tc9^rBHrvF!{eCEm@oyGLdf{rn0d;*C|i8mHU8nVY&tlqr!@?IHL_wNY=1-jQGa
zKpEJXuGG&mp6AEO3XZXdZ$D1BDd-=AO)7F=d;kX{wM6XSgHC_c@%J#xe+Ho7*3bcT
zrpoA3|M1UEe+@?6Ma@bA>aSl_-2Y=T|9(yY4xnu0%71P0{bOS4)Bo=}Agmp2x!eDK
z=%4qB)C0-q=w2U@{c8&AulK%W14;A$E&BgS`u~~7|2ksuN@|a(%9w6Fy(iV7OW3`Y
zWM<DoXt#hq`+9Q<v*#hPVrrxA_;Px2J?o?$bHMerABW!t;Yo&A=mNt1@|zg1ol}_;
z=Z!o{^NkO&dJDfI84`@zz5BaI#0K(|rzzQc?IXx4qX>olY6*FmlX(1AmvW%)wc4(2
zs2oOZuif{yY(iOWin+cQ)x{l1);G$Tf6y+^=n|?eGQEIRP7avvT%$LX`oM~_*!$v&
zU(X(2Y;10p?|y!@HkfP^$EdARYR#XKLMbM&?b3uw!0VJnnz`KHpTMDdTGseo<|6Fa
z{o7fQ1yxrBxx<m}{`PjEMK!6irjA)N7}!aGVBpu^kW%LHajBfTUKy5WH2X7bk2Stm
z2P0~%u*MOawf+PXFeSB6wq0e?Q4ifDqeXrSqA!SFncX$oP4(%fs+~ozWsEeWyRW6o
zQ$t(?&MCAvavG|(Ql!~-Hnojz-VkO6;b&j_HJGsE3;L^9znWqd92-!LtbeMnh@Kwv
z6-*kAWe+Oko1XnzQlQkSvDMDytYh9CYmdSa6=clWO}9GQ7%?BokjPgJ9N$kD3(DE!
zDIxC%)0HaI4s?O_(Cppjk`^nXZ7c7u`S`vv^({>LI#<pg`0K=k$J}pr#?tlErb$Hr
z&2ukMt%au7_^AYIAx?pTO==~2MwBLc^{zUn-J~I%T5)|HL?->fMV$`P)U>=+fKy-(
z&v1Vm%qK;Wp~a{FHK2<1lJYKNAl>8oD*w6=FNXB`y?H4rhIDq~6`?JCsd?93Br%5c
zgGN{$^f7&D%3fwc1Q@Wql_ORF23gZ`xr#&RG1V7uQpsapU&Aqv6<zf_9T|32JAda+
zejH%l#tRvARU6lpXH`Rwg~0NjeY(A{VFHk<!j@*@s_7FgThvdSWYQB8l2r42vb%Mr
zC}h)1mWm;<h>0)gaM=E_qb)@sm20`#n$FlS4VDUYYGPve-*>nJ%x5&-aNfV2L0oA>
zloI&vW%J!$I@m?B=p78B+?NFpcS-mr28WYU_2*ucWXkZKjudU@9iVbjq5nf(F2W!W
z?ZrEewWqfizwjuUCsjgsr4Y(x<q!HK!tscm`yaa+2g$QRr}6jT19G+>lq7TzzLX@@
zjdiTYmNbK5g11-V!T_FYA8Q-n7e+wsh{;GwnXN?knsPvb@VHOKG(JnIyvQcM<}wOr
z=3#s`0r*|O7AjP5^}1M;Njg2o)yqsenDK$o?QE>XfA?+g*kCxv?GJ%1xNrh(lQK1L
z5eYkGLCcXQ8oPZtxA_W)&^T2|HDzvtf;3e)3MyE!wmj_~Nw39VfE2`uaB#J*;{0Rj
z6ah+N8?9@jE7qu?;dFu5{jU_#lN9rO5C%N>eE#vgB>3|xU9f$L1v?u;X&z!((x3(U
z-hNot+dD7JR4C2)(Eed#A{422@Y;N6^piOJE?>_=vn4JxJrvT{u<0LJ=Ad#{+}`8x
zS0HtMEst?-!9uL9qOBOw<dOUU^R)rpytU}t>bVnu0Pe&P*_<f)%@FjHIR~fUbROH%
ze6Dt6o?hVs>DCGYUSOkg1-{UfMB0^4gK9Eo_Nd9R-3L$p`Q+1DMg+YLgwR-+W6$4h
zwjs_uB2mIG*t&%sS>;FKR%NgbV<v(pF*Z#or=TE1e{hjhp|(|atek(#e0ls)y0eRV
zby}%|#=^PjEuR9H)9Yr>|M>}k;pk|uDCHuDHFbT(BNyVzA8!?vpZLF`Rs&tuL_+SO
zuA`Ii%852lYd$zA{2TR^!9g77eHyuEIauhTt}^L^8)G6I&md7`|G4j2U_Q}3n<8x)
zN}OP0ne=cZXY}N3CC@ruh~Gn_=ebIF;@LZ#!m!OXocD)lu2t90=v=8vbYE0AN(*=)
zTIi%?Lbt{H2Er<D@@jxfN_}eHIM^r+ry!>l!kK&NDw#$Y(}o^|&~fYT`ouSw_ol~_
zvz%|Q8@`24if?!W&*kd5Eq;D;V$L+0?9)P1uWG+H&@Geq?Tcto0OsHJ6M*v=0MN+c
zu252Q8ik^q_HjB$L4pX!_^)in+ksH$FU~y4>sT(QYfO<&ZS);41ToR7fU%C?_I+C~
zKzJd;;m-=kWqAKNB}FGj)U_<MT4^s+1~IY6=?C_0DU*u4Nn<Cac$$cM%W%MD+9S~4
z<$i0OG*M>^nONVBC<?STr&`Cc7D|tb%jzys&2vS&nhBiH**s8xqFbW8jid0H_4|qE
zD=u+Zwk!G-5v1NLE_fzEKU2vQG4T=7tpF$jh4)C^&`_>)%iD+pD`nCjjyotAm9_d{
z`KnD)42S+#l=TC4Fsn%Oz($#-GD@56RcLf#j@s;(HseDyb*cb!@qpWrJwSotjp!7b
zN`We|=F#bAvTgDNRdjX7K}Z$fsdBRJAN$9p`k!~WAvR(f&2ohz!^^;1Huy%w1Q@R^
z<6=#LFT|BFHrJAlA)QDsIlC-0RtXo`86Az2isB!!wh9Hr#FI~AsydP7FU<7x67FUz
z;SbbBfan=4M9(_mMkYzFuHmr%JR<@f@s=T{*}$vfuxEYl;>wb%R6DVv$&a9Tk1Zu-
z(xVLH+87|c*!(e~krcNU=z;+czmcX$S6cNhb%a7bAeW5sn(4hPbS77afZwY^1>$K5
zj41yIvBRPWh_P0bKFkmPOXIOX<26x!7$x(QJd^c;;qDru`U?#ao%&L)PoX4`lY&Yd
z;JgK*2m>DwZY}%{0uwkyGNp1^Tvw}!?t1)MAw5Qv8eu?-S&)GQ1!2In3o-SUNav-2
zb@H$3)M%=p$2L|E>!Qcv!}s}Kgo6n8mLvo0pMs}`sT5}@F6a<wqq{N(n3H@(!<Lkc
z6Ezc`5XG=*{Y<YWPVQQDc1vFy;$U-|;`t-yfBV;;ZMsstfs{SO;%^EbmdR`OI320o
z&I<>Tyy|iA2g6(?2aPJpbQQxzfztt$+rBiD3?Ns7Y@YP)p-eh6eqF@F+Hm>^5e~$}
zk>~$V1wK#(U!OCms@Dd17CGT3989{xxok@6a$qHE><mRGT;3<Pnh}gDfItf%WHT1$
zQXf+Ti7C4r0=FX7L|my``GD^ocyIE`(a#W;M9>-gDwirGSlS66_uK)~ljfj_(aRs#
z@39go)el~0Yu^9)5$V?ECL4U7JAvIyM#5E`cA^&GkH~DLV=T+gk^8ZtW`WFEh>4CL
z{vj_40AXCQCcwPEy*(Jb*OIT;=+VjJv=}7eu$$pLgrAi#i4dHidsN<bj~_lE{DplZ
zLi&+fRL8IEcmpR_My;~e9l)G5qb}Z3GW^mSoHNnlsNeYrPI`!E-G4|fG^dT=K4pXg
zlA%W~*47zl2odMv2wk5omlPF2dRCa;^~M)&8U>M>bI-e|YVH8GfE%<^s}uO$OX|Q#
z_&}Lad;dyLbhP@It8mrh0%L@Mm0tk*Emu6Tr&c2iFn21Nx^)`L(gAGkUCfZWrv9om
z0lVo3-lah7eo10GobhK2OK!7v7TgOSCe9Pu776Ev>q&rC>Vuz?DJt(IQm&Ju!PZ8W
z3Ma8r;9bS31m}OGx}fk0|AQXdq{Bh}p~mA0UCB7{@g=4Y%3t{$JhGL>CHFFcbYrq{
z_<-T-wI+AFj+u1=sna3QjYvtr&-eU0F%Bz$h=xK5X3{;{wXfA6Ybft~Uuo^uIxhFa
zr^D6=WX`<-M`VJ&1ZMY$1_<X|C-qHw(@HI>$2B;-_nV-VIP|Z=io$g5s$+%c${Y{|
zVkEbVA;E1jX$jwpp8rKu{n5rSJa(c`V%IrcsL2r?bxrCz2)0-mfWudoy2L7bJ~Mu>
z-7E72Xt=)f0GFi9DrgZ7<3NUa+H^6fWYYoC>0CksTY$S;hzQ5SN94D3Y$tOE@*3-{
zXBYp%Qe5CC;@pfkaWYQ$$5x<5@Ei3NzPUWy(`C-tE@`me(W>R}I<l;s(6LDXz1`wh
zoTFuov$viFzwCV^LdhtV_wBcGrX(*?NzCJQMa<$iI-0|@5?H>gv6^!wbD~`M<!Xtn
zMsHP7@tJg1*02Ew)|(wXd8}ngt=0jO+tRUyXct6;gO}CzR-CNj0?tYEVf}d0kjEd8
z{ByB5oaXHhctT6+J$@TQivrobyuC9ojfssCnwy12<seFS2Quk3RHw=a1E@ii!=df1
zw}OHQ3RtLZS(#~1!_jwHs_cCea@gm{?Oa6_RNtsR%et&=CLb_clfP)Y#S>sIvwSiE
zB4*6mH|786|4zUJ$bhzK_v4s!&2toziRuF3Gu7tt`feZf+0;<0n_t!8#L;D82?0KF
z4NT_r7fK$IVuJHWjhA1|`;%wmD{%_g5eA-TrK{wHIt7?_Mi_P^cS&8Ls{QQMZz)t(
zWkB=Lgn{k0#12^YRFAvMQ@=Ec3JRULq|Od{^Anz511`W4iiA})GjW;>FMr~9N<D$w
zz`XHcPoLGpbj9oR&i%;)rugskZnI4LYDi6y)haUd`B$j@$^6j0uwJHjM%70+)Y>Cv
zkW?Orl#uSEj1uA|R-I-a&#en4PaFLSRu4^`m6G2E8>_FNr-r>hqNru%&Qz*rVMW(g
zU$$HP)~#g3f*x|gqa+3%BkR@G=0#FYb5M$zRTRIk9T`v=a+$`Mvi7D=83ElpF!j8S
z(cE)?UYtRBqKzHhI$920u5!N_MpHR|?<bggu)<^3`|NI8MhJ??<XCvnqmdfbMI()8
z4=}&<4)9_&C`FNAQhpqsptoJ`E{pl#VmW`&=&UW<+xyFQ!bLH@aFlmq6poP~r<H;0
zOjL%(HYKl}G`v!->d|Al;GD?5pG!w~{dRBHYb3q~pJIz2-6&Dg+vXxzRLf$owE3by
z75R#s4(Su`$&|@_SkZ8{D5NYZN$>1+X_7eqQ-lGWc!U7+;L0V`zw?X8HYhD7l#vyM
z&a>*k*~cgNady!82@=nhNSgbI(T!ggJT~o8=Y$&{Cop2~4JB~U&d8To5Rh7aMFy57
zcKUx<)*8T)McDf-H&^>>2YoCNdSjdsijO2#w+-nxYxQJS4fw#+xQfI)n<1ze9OVt2
zGHjP^?>aDr#$^plgZQ-}nyo~*iZC!X>O02u<`zo5i~1BbW<ujJWSU9t2hc`t9q8A5
zVb5!2MIm+h#QUrDULXCt<{A(ehwQBV7DEX+eB2hGYsgmeV%e`9VU2J9t5U$qdjpCi
zbc$H|4BmI6Ir7dgNKAM6xAHRKC##gUVj4}rM4x`u>nr$K&J%DBrjfnZ<7_D?^VK6=
z+j5Do?7N=`b<oDiKoZFMSsWNH3lK9<{oF?_bmz<4c_5Dkn9_@1YEN5iN}ITyumwR|
zs6YeR^&BM5ISGpK@Ats53FydNd-bFN|3t-fr9~2a#E@BWB!@;m5r==lp5K5@t>MY(
z$lXch)$_c>1KMNzYLm`~9N4s5mf2QTk#Ao--z%@__JQOL61Bqy&>LdQQ^jmkBEEf|
zA6kRa7uVFhT;G^uMRJvz?wEOm`@s&_=KFV~x8eaF&>9KiB?`IEumEhQvIH3^+;a(n
zKPS-Uq~1I442EmMg6Z=5!THF2#xN%gK@eN#6Co-%*^FR1ZeGu^16O&NP#a4Yso*2M
z{cpH&V6^IDz)=PEZB#)d1rOP_`(UDM!<LDV5~(~YE1L>TgRJj>Lz<Ert-AJg9mTlM
zUANTmUBuCiYRA5Td^-S+sVVoE&E$ka@c{U$wr$n<T!~}OF}}FP1mS~zj<s*ipp{Mk
z{^k#ui><`->Pz4GI5S8#po3mbs)K>zK&KW4B9-rd0mn3LkeqCu79*9)BE8Aao`E02
zeAe0zlAVS5;1{b47;N5uoC|=3!zE1r+FgAKn@yhef9KG@_@QD-fHA1twJQd}mx+d$
z{Sx1Hd~xg~H2?Y*X${~LcKB&io1*gfFqUoQ+#CS?gn;a6r5&hU^oq3Es~*!Us+?9*
z01*G=HuUrY5xxk&hi@`dVx23fED6w6<GJ0wF0HO=p1km^ATG0Xr$~uTEe=O4BO>@V
zjDN)n&$wa{&V@80ARBi#Rs)75o7WNN@3JCI2;8#W9;s9C^v557wvKJtsSF2b@I2ec
z{~(At3)H3Dx9d`CinMjdTQ$^QRdUuR;6ga)P5SjDz!hkKD?D8Wu7Clc(5rvA0Gy|(
z|Cv=`jRug)y2Kx(Lhw(dk{Lk{QhbKUEd-VK9)M7;s%NuU2#ASvVdMXYMHSEj1qgBs
zT4KRt2qUEaAwcQ7Kr)i1OizbHm)mbAPCG<YdR1bc!CS9@L-*m|@KNGM)P}-)>9I<J
zRKs!#%?mOcwQ2;GK#v`D=(o4>1}qTW2m!UBnD$4kX!TPka_ZqwY`JQX1Vag}E-vD#
z4f3jP)o^J{AjH;WYgys|r-%THd^#*=r9cAtn`U?>4+;`a^6;9^Z~n6HAQ27)dSI{j
zL55}Yxn*!_FF>uV-kS03?+6ZmOo06pP+KHrO=tU}6$frT0n&eaq9Ix+FdFmkzXi~h
z-6q1}>wD;gTXu_}`D6_cYCfL-y2}Cs#M*I1NG6Stw$=LH0GtpBFjqq7w>TPA`A?V_
zhM(XMGm#wttROHgk*5DOEehb^tX*%5A5#8hS-UjDq0t6;IP~uTdc(Ml-Xu`5)N)V>
z{=q@Pc{YB)v&}OvG4C1zNQIBa8I4^9eay!UuvLH&#@{wf5dpM7_L!Xy@&@?ELbq=W
z5SkBS;%1Dp9RGjn7!ez|@Z*Fv$)DZ@7zlZz81NNz10OU01N|t9fiH~GpEGYm{|NsW
z)K&_;|CjK!Kw)P@a7!ET%a?R;5*XKq$BiEp1pZ)^($gU4;{#7<y(MWC1t6Bydug-v
z`>~?O41d`SM}T>~UDYW0f8@*R0WTOVUOupsd;@$I>F58h_5pmg6ss{OBYyz$a{!RT
z-UDj~5tjt-Z+``+rLoj#T%Y8&|7W9rnPdg9boR&fiLH&E82z=5CK<2m!J0qL>Y~Li
z&r88zcHa^LCdC=1eQcCm8$dQw>pwaka9+G^gf4J?Veb=F(=*WHc2_U#?subp0*^`0
z$$)IxUO)f4&`<&FJ1dfCPDm%C?t}v#qt+`g@2)Nou;e4SAU&dS@)aMa8&k+Q%y(Eh
z9F4i56+_w7!AwJB#5OH!7NwN4x-Hs9!W<{g-K}+mu;h8oW|bPN6X(b=ZT}K}Q(@$-
z<A%KkHGX5HOi~A?qc~C0UlXDr)TiiB*igoaj{IF72MmHJF?G(Kg`$o{F)Zh(1@x-|
z7CRVB^&k19>7w56$J>hA8~z98fGasK21Gd-Y2^|d)IbeU*I)qeN8n|jag*A=IDU6h
zb+Wz8+p#Md&N6VUN*1C>w*mrKvR58B0m~QFyz@Yr<0Msrepm>g`%53Cs(wxuBN(I$
z&Ec~)H+r{*^VqWUW76MC_=FC3Q7n#RMsX3D^v40_=p$^Mf8yY0%(}l9PYma7-r;Pv
zr=fb-7@5+k+1=|?Sz&}s_s?WRY|R-*3cuUPq+k1sc4Z(A;KcV;^PKL^n$FagO|NN3
z%?iL(3DTpVK=v27;a5X<<#eHph2=$+(7TC~<39L1?r$(6XbV>lT5}ZN3SfWej{rWb
zR#7dK4{9Ulh@uSXUf6?ljONs8br(gd1X_o0c_E34c`>@VtTxnYDMjzu;GL4^xn;Su
zvv~VLTa(Ym%s>vj)ifAR0G7_LYF|}<7qL+#uYl?=^s|c7?gxF{1Ud{Rlv+TI&zNf?
zf^U&QRq#f`N&F)w+b3{`e@{9{H|8Ts4a@^{cs{E>yZwf1@CGotTvdjI*jUY4fm84R
zz|SaDkOqNQ87<xWPq1-$dRK5i>K5(#=aK>fd?`n1!=;WL0i+EJkk`8g;J#1n07xDB
zp@s-A0N0L>2S)L4kyUA7KvOl5td=scN55p!b@1TCTql~03@|6EPn{VN_5HOfd5r%=
zDphcD=OhZH*h&nZ2$^vH>_-OMGKZ5R{9mIaAPu+yFfzp)l}2i(2*4MppLOIwJcq)Q
z;f82pn8cRSBj2%+pMP(x3KKLtsv;OI(f%u;<UDQ_ol=`CZqTBMW&?kSz%edqqB`ST
z^TcNL_~>XXd;f5aiFLU%%rTDn81QDi32=8%F-lADJj7e0&l`<1Eq*#})1OP1-_m?z
zwGF-4_TRVK_9lWc5#}B{=rmoYxPIUNwm8W-G0MHbHF$hq>a7SB2-VzC_DSQ##9M%9
znJH2avr{;e$};dZZ1l({<aMgG@%WRWx-w@(IKa!f!6AvlZHix*7@&-kpC=#Tyzb=k
z7~8Kp(Z>uqu7{2v2TcX`9_cFTE?&u|Rvqh}I%jh(OmYn>qb9u;lRl>g>^}RQBdM&G
zmPH{S76s{u%K4+-uN7u}TQlxM<xqDY#*4g?857z8x>gx(ZTS%z!J=?N{(iYHp9hL}
zoUJ|@_f>Xv&Xk@{+dO+{{a2_O)$KgeU;}O2Y}?VD^(Xn2*Lv>Ve)q?=-RIr&-M*nq
z@-OIUNv9=5QU#5>N~7QW09E{BhM<>N%Cq&JV#{(@3bXC+_U!7r&KJ#i`kqho%rAeR
za&Xl8Q`ItiZj<FMdJ;unG`UzE`$zL7jDs_&s+;E{J?z`xYBI6hGR7>OXa9sa<v^Kq
znavk8pxq^qpQ$3N9VN>7^6z+Yj#v?Gm*}rw@?Wew;8JM-mgqqEageM?0a6+dNDn<@
zaI@ZS0_9F@E;SwS?t_bqwat*f13kqsA_!CRZ~Ce0Q{nko6=C4ps=i$s;hfa+<rJ9r
z`IqyA!70_3^<&BO2!MP=<y6($`6Ca|*=W1p^LI!&hEsPA8DsHY2!FU&iEkGYotY8l
z95x|B1<oN7+87xT1)oDi62X6qY!Scf9VxGxQAL|am}f><^M{J)+-zjli*AV0pDJe$
zLvT(<7_tARphYs_&_mfYAOAld{wJ%8Bw+)qUqi_7OqPak9R<Gik%sX90UuTHzw3gL
z=G^0g1zgyR_%BkH1#0f)U$tR-g;$EHCy?2(UHS0tZ}4x(cu>gXdYltf_km7h3Lu&9
z{*Cm9y@1$?<L@$U82@o>z5j)WVPJ}D>RH7YM!=iD;2|k!9#G-XLwZ4@qP^{Jcx#Cc
zoS74e13%zD;{Ky=K*0f8Jyp(9IsP%~EqM6fG*tjHoD*VwRS+cd5~Qr&kU1$d&gK7)
zu<wrNdVj+%AtTu&Wt6=_q{z<Rdy}$fnNgG-iITmtH_0lyw8)4uvX$(S%pyGZ`>Q&g
z-|zQ4&p+pNPRBXE-_Pg$zQ=W4_jQwy+OQmBclg6c`c0C+PaU={!ScJk&(G=v1&$D=
z!@&7=?;q}Q7Ld8plC$}TuwuxUdYhUx^R(c9wJ%sPogYtNymV>*A1HEG6)IYHImMtq
zM{}=E0r;7!spB4b@w_Gqt|m}4t9GKA9HQ*OVD3L9le!!{K=(ie9g;j$2)a9AB1A|T
z;M%Qm)^?-^rv4Oce%grT%KkXxR!a&+fNdxwN@hQ+;KQ%T8w6DT3Zus$*d8OSV(4_)
zJA6GSpth-o`^2w^lQ=V%G|}-_ZjeY5*l(I0H~Hc(bPe<EOq75j?+c5KaOd#*SKId%
zI}tP?;UbS8Ih&SXRsEG#E<r-NeD<RssMSzvAm2+3x$s@sf)@BnY;k^HqXnq4OSdhg
z?l6quk=W#{O29)_#6nd&{A9M5%kN&#&kcR$+@TrwgQR=!p$a-Z1*9d=;OIQ0m@fVL
zpYwB@lp#;~&_R*^coXlX;d^*nsHlm=OF%^~ga;LQ4ywp4u@ILSxRv)eILQkZUs!l<
zc>5nw6ja~eV)L^~Ud0pmz!(s3^*c{>dZ4_k$|Qh?^$H(8{aM=M-~J0n8iQouO?E>%
zE#Ox@GYAKW>BxVddNz4DKsb;8I((Q1nzjY3*}N~!hZj7|B+#15KtYbZ@IP8pd^kV&
z4D(WH5AdKfb&S9q`U~=D393h1Dky&i2tQqfxKetR5c&yY&d$9NL->*R$$y}331Rra
zIa-UWHH7$k1sFa0z{mgl179(<x`-*5{@ek`+!T;+<E7F+Oe|tPtrC1%i~s(#3`iKE
z#qBFRdKcA14SQ{L21}>1P{g&(-_iKlMW`?i2h&YSrK#Y{vmX<?@svi?+d`RL+x5C`
zS*)}@V=GG``#?7sq`k%vzHO7b>(vz(9#8aNzP!-)b6zkjbv=o4{q$w=kVDByk4O7K
z_iveGB~Go7vNlyCy?!~(v~jLMsq&;Q?*Qh_U6B9y`x`N2qmRgOttRC#7J#b!@GIW3
z2hFbEs^rZD#Kv9=t|(s+mpXw?8+-7%Y;^;JSMV=5P8IS(?2eIg3<)#ej*&9)`x^*g
zaKv6ENd5iw{scF6CZNiLq;gwlPK-@aY(5c3;d-<-g5ynX<obq@XU~59-rC-Nmde3F
z5!F}9|3R%y<HCy{W{}z3=f{s8<|R!h03HEc>GBnNsP2VI^rzQkT?xvM>fKN^_drBG
zZ{loTLJ(1+mV6qGLelTgBVmZ%yw38fh&y%`1;^TYL?nCATZqwoBZkx5L7$<C8_}HB
z?<Y1Od`uCB`@0nMQ$;Pw3|ZDTNnX@mU{6s?qmko-yD6DQ6B6{N1*nOa=yfi<BO>4T
zSE$%K<h;;)236l9^ebUOhrH{Gs=1R`dV~drCD-)nW#>AdtG<FC<1$BK6_=@=8k>8Y
zTQ0o~w}X9trlsl?rb0F8t&N;@JPI^+;GeI=iuiGhnJy2!L;es|b!5JX36BXE)Foen
zQ{U|FwoezDnLFo#{%Ls9$WisCZs%OT7V8BZRr^u<BJ(ptS6+lYs%R1meFxrIb>N)k
zboc`<t;0ny=F|B0gCbBO8yjkXL~RgB0LlD>${XGQx>+>;^lpF&tK-qK6(f~9(g9C`
z5uE2Mf=|fB%L$a-r{RR<S&E8^TBW%6@@eHL^NB1}Xhpxn#h#VfD}p%yf_Om-7x8vc
z5S6a3j9Z>M!Ex?Cz4EjG90a1M_-JAFfSxR5%cV-QIwk*`VB`m6_%xyiU=SwhRFlx}
z5WE*9{Sx78_lB1&g=&rfd#QVJq9hS$L}BaskB=uyUmQLSV+HwzpZ{+}Vqt!283~&b
zDmHilv}vx3>Ad>*n-wasKsffFO=a3J`ehnYkLuSGtC!z$wiBMJQ`ad@P-ZDR6mjg_
z`(&55?8t;Ka`q#bMYOlDK6C)fgY%hQp_*q{RT3@gaQHZyaj|hR4Jiu-gUIB_hTljk
zPkDt^lfcBvm|dIdZ#{!VE#hbb0s#q4iShvZzK^%a*Jt`Mu0NnO%*_~}_gU?-=@<h|
z{824jADqJCs>FUE2vf4JfHlk{%*_d5IKnm|>{$$^q2X<x4hJd`;EjYokfh%_@b~%$
z`{B3{O@i;>9lHTAe@RiAGO$jeoQN5?=Q8yHOdh`qtu(mLqd|-^HZg$v*#f-+oAX%r
zaKD0N0GO`Au{>EuWvCC0+&?UHnluE&UP<-VSZ>$KqRL)FE12CB0;D<Hh(F_f!NZ$7
zUV7R}N@(^(MvWFM-MR$7BnrdyRK~>FeSY4g&=K~u)HNM5cm&L9I$^$mO;WsG{KPZX
z-)CU~0eFX(??-vS&`1)f_B#&I@Mjeqr3{>m(KEd|)y!Qn^n;v5`3k`&nCTt5(FA{5
zT`;F{<lpt@Er<xy3Wq)$%^m1Xq&j6f9~;y8qrSf0vJk&@b#0A?qN%KzjaMFPL^029
z4{#}fQsT?lXSd(wkRiZKrLRuhlbU{DEZds>=|zZ~Oe~MdQO7MJgTwW&x~67~lnnBA
z6iY}``(_5k$O>a;<&8#rtlNdw4&k0`mAMo9^aBjYkZKXc^k4QlNqLbusZ27)t)1_3
zNJ1#Nk1zA+kMVP!z032H`dfnH*NO6Tx2=M<+dir}C%pTJ_5aKhi<vRFi3SyUoAwu|
zIUh47w9ZMmsA;`Frs}1awIIYpllyIC@=ev3)H4ZlBL%*>+_uS_Gc<Y0tU6B~fAjfT
z<=gde9TmctyaEZ#nZ-9tl^KK72|t=$d5+^%KY*+oHRO)R#665XWS}cOXJ@!MeA{<@
zt-XUisOH(fa*rkZ|1bBXYo{e>WCjm@J1hB0-kpi}ag7j(d!37Yq2gS<BT!bdy9zUm
zY59V-%HCc7w5>SqA>zy4b!&zY0nt=Pw5O#3UgaU#yjzo$vrUKCnUN5~gsJXbvv*=n
zR@}id3T`t~`A;;*@By*nhSc>K0IWB!cD4_L1T@IcYID=7ntaZdyIVL1C<KL+AobRt
z(@#p6z23(%tyB#6F<)*UekgvkVe27F!ga!QVyn2=hF1Q(BLrYFAYeORyLgoePaq2V
zl#?^M>Yme!nk_0kT&L~2^XfR+oFAy1X=#nSv^1IITodxDF2E%DlZpLH72>zHic!@&
zA(wMZx8tPb3yA%^Bb4rL2yg@%?&_TCo^2@;d+&II8+F^3jWW6}YDsNtv}&1c+k~!O
zvXsjac0W%QyX!6aIGYeYbvcS~4*j?3Qu^?lZ#?ZwGAF|}lmuM!RX*SrXK5pT7v>m=
zfDso+1ZIJ1-2Bq3&(pH-A(!v+>+dQJjGb<%Qv}!$)Uk(^+aa$$UQ{_*Ogm0MedUc&
z!PZtWeg0j?aTB>$YUxh!WBsGM=NJCLY9x>r^9-fu&a0otgk*4wEq3ml$_!oQI(Prw
zgXBrnoY*K32tv7`N`f$7_F(wy`I%0(pc9h6NrE&=$UsJtqE}UL{O?OWz4A#4a?T-Y
zY-Wv@3mj~3=N>gKRtbUCtCUQ0fP@kr^a)1+9Zx<|r#byJ8bB?jbh6&R^>&F85~p$$
zxLHbp*p}hGHk}@lvGC>f3u5tOGQ>k*CK(md|L@5Y26alv^E#jGls_THFq~d|1UT@%
zA&8MyKv4<;tLQO*soeVA8AQ68j7)cBwr43kxb&vqPofb28fW?d>F*9TT(i;WgoM|c
zQJt8+;Yq_fv8@KqhHLd^rJ_%+`8c_=*7l^mnuE`5`$zF<gBmPH21=Zm{+qko?O=pE
zWau^Z@Mgog`O+8R6f*66$znb=)t|#SGty4OWM^Dg;&~>#3R;A|#oqR$k8x4{mnKBQ
z70$((5ZC^~zZk!E=44^_g6kKfrH=3j@_8*=l)s_79$g%XMPkFe7vAj-K_?-Ps9^~0
z`hNsOX)@^DU^jGHqRITseW!t_Oi4KBnq%aqNBEVaC42P+{SI$QM$ohSeX-c^pA(@*
zQrVQ<%^3Qa123q4hj0~CjWK*FeAG}=4W?NIB__`=b(|ePRDSU}*2;UOQwhvUvGIbv
z`Y*OOc&v2v-s1m9mqY+vG1tEA<7E}ob(17f3}E<L7o+ZTd#{}US&S-Of5q{=c;ATz
zx@zTOSFy1i6%f$>>hmzP+J94}Hiw`GeSIZc7e<LB_sW<E#L*G_CphWHg4%H`3m#~x
zE%4#&xxn|LrTOnKNW^^4q`EpiF3sh4viMz1!W-7d8`R*az{fICerAN<^J<Sw`3z*r
zXFs*yRi`8B=~Zx~%VFvnRc(efBZzbPi&^x04%Nc3^U`+E4A}i*7XL`U{spP}38T8=
zw0Z#`Lj@;EID&*90Ep;|J~hW7fpQy-#T)<h#S-U$cwv{jJ^mk$yau82Jd|9+{t~RD
zOS)2(=k{n7bD-jMO}&bsQ}0vwoLl_+KdZQhP@7M#w(v~l;IW^h3aY9@9V$T=Bk}J>
z4;AyhiV-W&lLeFPwakU^(H}!p{x>S@CkYS!81{Kp0IiNnGXP|c15Ml6>rhYS^L8HM
z$vWrzM*&2_3L^Tm#OuGG_^)OuD;T~&_wB?8skCxDxTc23_Zt4k0B-{I0&TOwi-*4_
z1rufA@!RHAlfbASH@aLblmB^W*iiQ03#^V(dxvH~1_0oc|Kf-8(4z=SW5eB#R}m|b
zM^V6xdZ!nhK@tCEZqv8`Qi_(j@K>vf`CE`s5}JFyjWLx>1ECLRx-jSx0mW{Cvd-k(
zml=w^Tb1?+mF^Mogb?Z1m=I$qP>-IsqU+~B1f~H*mvC%Q`=0#OHlG53$b#m-+GZSx
z8dSHW;J$X!pnJ>>O|Poce;;!`I4hx_E*|`MWhLQ_2LkZOq8WDV8fdo23S9S~i6VS~
z4>8iIs=uUK|G-uxHaO@R=fF(Izb>XN;Fs~T`5&zD+}QU5(=00lrV8o*2uxYXj$<}t
z-#&&;6KoixF%LexaCI?rmIkW)<u$<irP30P9r=C6+0l^2|D4hb;vHVj-5{A(#Hz0^
zOl)ZAFLB0v@?*qU?QRqIa<%@|hD4Av0tvvv>$!NNyh&DLX5lzJ;G<_YiY!bNid!uZ
zgq@T6f2G^f&*tq_bMF`rOj*r)dVo1-e@_298BOc$>cxB8NAmi<)_m1mxJ?=Efw?y)
zm>hoPa7`e)cw;W*ejt%2hUZ2R=MHBI`0lXs&%eY*X-06dJPIvlsIeJHP(^JCa%8E?
zC`bN{=t(Fd-jdlrC+YXC{q-B70K;SnEhPqm7%~F0_7t7~IW?c3)#vjVbom|&*3F%j
zhsddj*8NAfmYU!L78S7Ioc2Xs{;A8}0COKAxu=;2YUKS4p{KqA)182Q!~+q<ZhU@7
zemV0YS5CBBjNJqd?FVLOZ^Hd}`oF;ESGX(Ta`f$1y;MHhQ&e|;(p;d|;iJZIXa@aw
zS=G~-Z(C(q+UeS5>3ckm_*Lw5$)RZQN2J_Yj{?44)hdkEut`F!&<Tg|qYtpazrQ>+
z86a@f1Tj2+Ne9vQ&kIESBbx>;7CH$~{&xWKbD$*iEg|)zxVb#de-P)s+owz)niPo{
z*NN3$sW<%a6qC&xegV!B8>p`bP)r&w95y=Ow=NTVA?o%Yb)ZBpK!)?$EnQCVGbO)>
zF%JMhU;mpEmyoy!5*LvD@=2=toT1fmE-p4|xKV9TVRN5G$jP`^r()O(3I|rQqkt+@
zgrOh<^jeDo=*a&HGWRsGAaO@lgN&{X{Jy4k-a%o*GuLfD0tRtOU@g~;Ouvyk^vxc(
z&CSg=kVx4t)^C`ufV7O3J0RT~{<YD@O=yYvp$w`j+-N)fuHu7889Dp<u%}r$cpK>X
z)Gq}>!7^g((Qm2IYnz*5B^ycSXzDdV*_VSfx9}BSSn$)F_L$c%AMu_`5MBTH;utv2
z9xrCdAdx*`{SRG+-UXY?Siki{LGjJ5$TnnIUw27!5*gKO?1JB_0LC7l#yZ&?gr>!H
zWkLKrVDqv9ehfogrgdjM^w$)Bbl+rmKqx~y^yUsub>C!jlk-a{UO(puM3p;Oxof#C
zfVJn!BxtRKz(K~dttp5w^M<3mhR@G$Z-#GHZoMQ|69CDZdXaMOJY1@<EF0)VJ1Nk?
zxU&Zd&O=;_W^9&038#YF52~60<TjhrC~kaEXmDuW&Hq|M^y>g(v5z-ql+5@+@TJz}
z+w$21+}Yy`OKLbCwcTTsK9Q|<&@%KqsH5)|dEQk;8nttv?!MG#=CjswcVP83t@DMr
zxU%o{Uwz)S0lsq#jA=A{a~=NqNCFtu&Oj~Xh8zJ_48lImg7sw7x3>IHu+PAamG*X4
zv2Xb0Fqtj9(Skg|!3TZ!s%{<lc{EYRFfY)Keh==y7(mo^X3&8=mCn%&1fpJV(~Kp+
zjlj7alzt*Ot6)xi`pJ;XogZ*WzOLU~3ZTC*C1f|gHM;AYb%^fTnBU~m3t}Ot(i>`r
z;6d1RFLIVVJyG)$><H!~Zca<nhb;@69|$TX&$>Vh;WPGE*4Krs-*Vz#V4R791W!`p
zBWjjvxm54jd@Gn+)ToMP-hRk$8FmhyVLRwu>5o6iR(bni$23quw-WFf-nU23IK-dZ
zUj5K62WN&&jTQov%Pvx(Px03{^p}17COh9uR$!Ui0QTB7UjCm11tPyiYpGs!WkIG)
zv^Yr`!XUP52g?;F*}F$6shsB=DVnONi0?w}WT*ncq=yO$52nvCf9sB?`LO`Ey3?Kx
z{lRb>@+;4%kB(x?UW5k1?YX)35`aSjs8&7)&MudykmG>c%C{TH&mtlB(B|b$G0-R)
z*C*P43}1a5>BFT)bLbJL-p3F+@8))E?FM57a2PNXX|>faNqM4yH)tU2?vQk&d`;Hs
zOY?pAqRQ>U<vXEsj#>koJYLtUpEPF8_DQuE+P=I^dOt~S+b%cfIhoDNuj-dg8rNjI
zML+JG+5B0tZe$#C_;{jlg8fL8MafqIW@o?mKEt~vYD&9;9h5TVG26Z6FOe%hUiIsT
zlS?{L6wxw%7qvs8j5!_<e1e4lZ}K>}C$79CmAkO@>WsGR+ib45yUL$;^VNI`=9#r8
z!Li6h;K?#L6_68{8`x3U;O`$+Pc%1mR+tJi?yhBS{_?P@VJqp8r#YK{-dx#sK)~jL
zv<7k2gQFtjuPys@!-nJz8q4WyNW8SKjdADJ7>R%UqU~kqK!E@^w<hnL|L4OIl@C`I
zf=r&$nzV~C#fNWqzc62qEN-ZL*zRfFPi)^{lodPow+K_4)&`!Jxz$MBwX7=rAi;V0
z4sQvCL1*K<)zYZfzJ*_H>mF@2+S)#-)(dts0~Xue*?x`i7Fv9gbFJn0D%#HB8*KRj
zF5-M_>Qk5sFfqUhypQ8}SFrcWR-$#D<*^%4NjA)LTekJm)kUh{r0{Uyp0mL&N9es8
z@DPdzPxKxDIyb?+a3bEeI$Tv=A5TP5(cLD~iA9ryKqja_dd<-+i0;UTUQcI6WuwM=
zvOLi1dmmCqwbAsjT=uc8kX|(5yQbbw^{pVU`~B5%*+KE>6-7P$R~oy^Cmk=F+<t$<
zOwE3DZR5P-<>ArRp05>aCgXKIpleRz1&3{`Pfq=gIH}(V(h2N~#UGLEz{QP7X#{WK
z*Xv3P^8F5O1JDz5Gz9ujerKw-RR}Zj=znEhxh6zlZgeN)HDt;KSq7M;n<OvrJJ8#d
zoXwk*|74)Y2=NS@!?$GAVx_Y8+owOF;NY|djga(=#V=hbn*Jpo(abZ1YH~~S5L9qp
zH3iX4BO_PcbqI8X_Hg=HH3><xrZ3b%#4C!9tgVe;R^db>nUZ-f*jzD|D3bY!WsJRB
zL5`?`(&r+B@}t9bH9e71`a;p#ko2;GP||yck{+n*B3eP63asi%uhb(77zpl_>XE}B
z41SxW@*N)vWfe3zqr#H70>zkPazb_5D)Xw2?<Drz(EE^@B^9roRaos*-a6~d|C>Mf
zRgz24?XMOO&L5bdxKS!(7UH?Tx8WiEY$Ufq*-K~eo($evf=wH(RWGRJYt~t+f4D4s
zzZ+C~JX<OW8k{QDvEo3xypf_HU8#u))TxqJM-YSuQ3B2XXVkcY)eqwAsoqi;n`qZQ
z9jQL?q5{15Wov!!dp`WOIrpg6efy_xzH^MDo94dc9$lm2j8>*W7;p~^>#z@zVhdbg
zz!^YQ7Dj+amE=iZ%mmD%xos>)0@=xdrI5YQqi0rJz^)L)8(x=CB$BTkQ=?usqG@R_
znzQZza|x6wsSek(!z^TCgr1TRm<R0F5`~1f{i`@>1w_D~7%IvOy-oX-(*GGgUQl4j
zuc!y0Tt{#fS;5Aj#Fe>iG5k(!%fYQ^=rvRvdL_+iks(T#q6@k5*@R9K*j1m=@jez*
zs;L=`R150<=pE{Vdd8K$=YA1kJ`Anu@ZB7V<stXcro^?U<k-y9yiwdVZ+!2&VNk!z
zUE@NKG+ie`G=4Za5k~Jgkb2AiiA}p!W63u#B{o+*`w3gC#Tb%0b>g5?E%$6@FM;or
z$SBQ1>v0_=^=3h>8BdR^8))XKsO&9=MuHUW9V%`Y!VlEKy6@^x!GavMEDN>AQo!=&
z6bEZzhMD2Xw+`b(^_>_Gncq|#6f4aq8Qw*JF!}Ubz1239*auEkUIm7r7H7bRrnog%
zX2K^b!wW3^zMGW8Ac||hWt<EI&ywoNqfm?_UkaIv-39ZO1dxT7I{3ty`BvYLn*9J?
zwLh)oZd_)Z$K&gghcb~Ib*}>jPaJ`%_d2J<2(o}a4Zp{wFa}P7DLoey#!%6MO1=B9
z)NElH0rEq$0=9r-U<{z1wirUfd8i9Ml??8H(fRCKufF%3kLuo+*o!<r=KKAVmu1XV
z7?+8^bE@I#*kf6X9asQ$hT8O6Ix01v&+bBj6_2#_!d39LleRZT#Rt9y5>Yg%-Xn7i
z+=ne3d@EWidt~GrA6#-#-8q;jiqre>dF>U%^UI~z>W!X`n7k4$C@EcN;nsDYuiClZ
zrxA1X%B;X$m*YKxFoKT;1-J6_2x3@+pvB#T20^-!xUrG&?#An6u<$M6151A)j3IJ<
zZmQkkb6@V&;qE|WJp&7lstmx~z|CRQXG0KJB%_M2LKe*D7oU!|^);*u#xeJG!$Iu(
zIZnTOC02nwW%D{Cju!amj<zucslA=q{rPS;{!866WVx8^HJQL-%p?8Q^;}$^RaIk@
zeL9J*R46}cmi~v9`#-LOw1n%71@jFUVKB+edb&M!P@EUp_P2p6lKGhdo0Fh5&z5LU
z5<1m;r9FQON`hmpgq~un>BI@S4}AxWy=SsD*1_W2ezj5kDD3vp;>o`EBii_D=M?dz
zkt<yskBObB_YAXZr{uu|l{5V>9m?C!SS=1!qoRTjj#IeKBj4uF{kHf)r5~7NuS2aB
z!3fZgog3V)a*A0zKP!mD7xo4+`}(MFK2!fSW}PT4LD1KouQhzE@eWT7TiK4nj}PZ-
zFS#y&hoyuon1<4@>t(;EZdk%82b-=$LY}7#E-KvGrn<rgBmjusOj2j$ntX4N7ooU9
zab7M~0q1>})XKZ_E8$ya(?4OrZQ#bD>%-G6IJ1N?j!({0G<iJ0TliZ;BSA~n*(y>U
z*053k)tx6jQf$*kaze!^?JH~-36Fk`oW?(=Hg9WBvGeB_1tjX_;iVd;UIG^`kKXF~
z^?_EIvpUb>cw6lBzY1J>iMN94Wrv0zlTN7@U5X`5)SOD?=)AJ}A+gT+{K)AEk@_gm
zZHx!f;*+S+9sMnQ<O1b#Xi?ySQeX$m{wNvOn3w#BZ#<j6DL8}5f%`G)x~Gd9Sv}RN
zb?g?0F4U$*MQ}b<<W?Z8p}+c)bn#~XieMz4@?CX2pKZ<Wt7Zi0##WS{v=s|B*$b_5
zS>eXSF)RT3Jdg_79>qIs$y#jL>Y;zgItdYyd)L1R&lQdD^bOFlM;n%_>Atdi4_A+h
zlaA*ay@>jo6yVodo2B|4^M#j(d#`UEfJBuRx)-Wv2Ow0Fq2?dv44|bH!^n!Kb+Ho7
z*a$NzTG_R7j<hp##p&vAB=U}h``5b}TD)`dahSXcrh@M$4Plah7Gazh0WuI`LfFAa
zL0&jCUO>tX$$^|~oEIZ4O1q5Tys#X~n#y~s;d@4x|H{Li)!cS~#g*2lVvU^mU?rTQ
zwk+Yu13T6<>YjhpSdgZm#ZzHj0&{ZFnn_srHtZb;-0$t>C|bDVxkg*>rdoSePtyOm
zJvvc~3$UsEt=H`6^5IGYaH=D>CqB}K5x{KgAi}5Q2KXjcLyCD^&g|W1_7XP->y_`H
z`VtP(U29k-FW~ZeMHv#mQD+@cW_|q1v;C7m$apwqa&YN&!bKajVZ2`Km{;wGNOi*L
z44IjS(01sjyrI8Yhdj+Md(CSjA&{_XG;-=vr^d8fUA2)M1oqXSe~RU<dX6_s?^NeT
zcn*zZrj_GEulufUsVO<w{nEG3mXtlBcHL}GO2*LU8Vd20tV$xZEiY<{Gp(KP&9Td3
z-$K=*x!ycq2Kb`&bOtuSdVaybghZUAHm|=SR1~r<jFnMwhlS94QeqzeKCI)C68kL$
zZZ1KSE+jN`CE)&?O@0ONqBSyO@EoT!U!B!OZjJ<!8i0u00zDqQ)DlU6#B3483Ki3y
z)c+X+{-MWHb2l?JE8Xoy#bu6(Zcb;mLwNt{|Juej6iu(%+?A}3u)F(W*mHEac>l-S
z9<6Jutl~cuwAY|2KIZ7Tf=gg7q!iciH=>fZ&yTL~@a)ZTO-Fb8A9(Hb4l;(Y@2njr
zjf}3&#F~-^m1cjhK8wCB7hJ(^Q_O2Uh;&$7a`%tx-9SnvHrb0$P1>xpY?C=U^)1Ww
zA2mM;k{Fkj<Szbn44opt^Z@W|+5p}YJ5gg*Bf9DRdGr>tl5SB{#T>O^ENT~vcS@)z
z*Z=|GiFfV@iK6&5G%*1u*#<0Wklsv^-L3gI-4PAwqdpn&LG-f~!KgC9LAdSne?cnh
zIDU+x9*5JnNAoQ7f*9QK;6WcIrD%sd_y>QMsK+Z3aJx+-Y}~+hS3&cvLJD$hIQ``U
zCq!imG62)vMKNGHEk1Ul!Y5*X9?<`}Gnc;aq&w}Ve}GbI^?A+P!uv7lAlT(`N7}$6
zzve6c#!H#-SM^Owt+R3M9KVON+<&qlMqoY|795P;Mf0y?Bn=Va7F^+fXdU$^i5F8O
zv30@+2SWiw6;TMN!r(6d;Aa0qef>^g2Y+N{4HkGGrYAQ(4l|x@(DGWr#2DWS%<Lhf
zqr^~h)<pMSgEwGkD`vI5)_3ZCtHZk4&N8(tn8GF^!)(BR%KNR75)i+poWTDo{0Lvh
zmulC^Ha?*p@#|Aq%z7GAyKk9JCs?qw1lHjNs$3IS!3)OmV!Fj8b>&kU8ynt<#W0>@
zIDNz7d3WaShkQPO9cw~L`GvXk{FkdIZePNt+V3S#(nR|}LW)i~^}!qE$(xPtwTs1z
z35yA-Ga?_Je9i5cS-d`{+f%%$t)_N!YopL?*gekc$3$HF@<eMp8Wx|Ug9w54-*FlU
zaZ{59+9t4X{v)}B!%9GUWPtZ&H?f)%k-16G%Sl+{XaO$JF6Ay0Z|m+qGAMt|W&E`B
z{BrZj$*1EHr@_M9Zp1ys&CM+yBd!6A!mCfWNZ%YiyREYTI;d!AX^pshmH`Lm+QJ-n
z1Ya}GO-&S#2r)1MoQjt|eD0-*;sU2XZ>5&p8#p;!WLI>h?(Ur&A5I9;cW#n~(*S76
zxgRH5L6hWg{;gi)dhf~~&xw668X6h3gL=6gIUtHb64T$m0UD1aD2dzaVA^sLe$5%s
zOz_hGXqH>PV(dBWe*Z``^VRkF{@xSc0p*KFj-7_vU@7Qq&?V%c-@p~cybKvRl`;Sz
z7Je`&NbCo&53TK7*tRIH1BS9FW({q3t92ewU6xC&8u?&`ZYm9ins~5?8X(}{^x%s)
zkl&NDF5}7E-k?GI{)&%8>u+^<HawU*D(p0L7SX@1FdS)U?g@v<$Udcfy+u^pDMJx$
zBr1xpV153;SS43A#cmJ~E}?cvKCD`7vrdiM9&?GpR&7UIyJV;nVq>X{OqyID`mFPf
zZZCSaLkVatV{2~W`LLqy1l|XJtyScbBC&tJHD;<!HH}88g&L#zebq#cLep)DqyBHt
z!&|Cq3_0$hpL)bCpwoH_iGD052J5`a?>~K@uZ0$MEqF_Ywp6=L)o^sp!4AOUTgf0B
zNDy*1O=ak=c`irq(;XY5(-kVISKZJsf-1~jOE++lkS)nxY$#DqG=-Cv$i0b4Ah&Lv
zB5;RtfRBNdfE#9<US2$ldRP*YJi;>(Y9523JLbJdB}1!&gK^6Nj-Yl_mNYev(&@xM
z9V(CwTQn14vcCF+*Q2ekjg9vix`kz`Z*!>MeUBEEmg_jNNG6`Pj_~2Wq1k`j<KuHK
z!;0qfKbKx~`4EDX<(OoAut7#7O18hlSRR94`g6L#7r|?z#pU{Oo3vV1;21VpP*_sd
zE^@VZxSk2i#Kqrfdvi@TRwNr$VkV2r1%iG&^<aX1p8@T_bX>npda?~jZR7hqj|xsn
z=GEA$;x!4pqw_%&JP8S$=jRv>_wpz$cH-kB*U;DP40={-?(VndXcPBS(XtuKAYWTA
zch_k+7fsKac6!URF2P(q6(x_=aqa!Z(%PbqGK_@jO0XlgG9-Jp;2Ct}j^|?HY`A;W
zj7k+Bk_74q^SPnXHi>iCOazpE#p&kyschP`?$39(lBbleP_fBO;OrVFCC&wyNiW!*
zdUz&|t&h&s{;+9p(8j&<3%t+8<v-AzuRVpEBI&ne{6jHaOLb`|=jQj4l6s2s69ewK
z>y_H#%LKBl9f+|dB|8kqC!GAqhxax)KZoPb6MW?|AiteiS7e$JPA57PPC)0YC~#3_
zQ5O2-A};`)7y6+ZAa|rs_aQKe%6_aLrDS*ca;7c#rj~Kke3{)~!b?h8Q#&;@B4wo2
zoDu&zkU@F8sLL8%E*ZQc)B0YR#pa+2;)pA!dY!@BHlM_%J$%W(jXa3MD#)CRAcm6Q
zg+6FNDZ)SXs8h^-qz;au4yrluONeF>^K4y9Hi`7)O7^qst-T5B>r2)(E>9hvY}=I2
zIi**&^(ehRuCvkUK|T;Ar7$RkRB-G1YGl`V&UH<85Smdk$WCn6<C<)zVyE{C)dVuQ
zF_HQ9e6jhI>q~U6aS7yR_7z98x9+d9aGNy@CHt)RSLHRmOTG3vv_4;!27QP86Jy5m
zDFH;lHqE%kN==-#H<y!JnqWPs)S(rXA?#?2A3#!3RYKH(nXj<B^NjOR5wDtfbzW=Q
z7ezxS@9oMMSz>XYFVmtwKZMaK9kS!ReV10qkZ@G5VVSbFF|9V#Si&R<r&Qa;DRoC!
z#`FA*YYjc+rN{O3e(Z=6w(4wo?YzU%88UKYV4v89lFD;tqDzk6v7~<8@HFh2<mvl)
zD8awdFFT4V{sf+n@UcHyeUJnMNl0(sleiFbYR9`Z_hXqj%{4lh?ayd$?$nwU+5Vs=
z?&u4aDg%xzOzREb?m8n#Q1DSbtPPT#8i>fy9w@WCQ(t6B&Ep8ooGk5VtvyfUQjfDp
z@nXhNEe`DTB7f*zEV>lBMi>eBu7gKL;yYO88|3=x912lwq;P+TjCU7(K+t$;qbtIO
z^u|Olg*Q)XS7OtWlH5+B)CQqbT@J5@wxW{?zy_;_(afF!>%|SmLnrRC#Do<y8b5JL
zsa~v^3maKiGTYSDkj(q~6LqZ)`=}R)u06-0zTju^6PJ0~c-chKBz1STWXrz;U(vJu
z)K~geseC3SbG2CcM$^jUdi4wr&dx&Km(K?A=)HTOD%_%+e7375!W1ZlUc0B>Yj&uU
zOrYe{jfYVDWZ>-HT;>5)i467pY)dz|kYMmaVN$~PowY&G<v1nvl-R7X9CzWE!eZvr
z2#l~dj}psW(ycw6G925mnc1e{lEPwqBP{uC2SZ-_!5psJh%jo0<;*qpAuluT(ABtD
zHgFc2Np4CSO1F&O7{Pz5>8|s}g#I*(_}R#kY<8V`hVCzou$4A;21d#4c3>Tx=;!Mo
z7&3g!R{rhB)I;Z++-JYmt%OsB)KvOwn8Uq>R~~(<98Jkz#iM#Vai@tFloiq|7t9aq
zROFCou%RjQG%CF2?DBXc>!|Suw`(E4-5HVE-}p0Y?Z!Z)I$eOG3p?zo-tJ?|$TRCb
zUPvTE9W%DF!S2>KTsT;$J|*(CuDtb3hZLLigCVKyF+Pp-v1gSFeQS?$C!WsYd;((z
z>sA+VkDRnO+cKmsA)wxJ-d%U@HNG*MpF4mj31wFtrkoL5e;nz~t_B{&sBl=40lSad
zd#@KWYshNuz4g(~=O5H``-&Ohxn>l`yP3C4aaZ8fW*$HJKx~r%!Gcc!&yOy-n5m*5
zvTm^VB83q}bwX?l$-*zDenHBL26%Vyi<-{3?ZAE(0kQxPM^(LM@Y(&!4%@k3^j?ST
z-3F<0$nhJvP=36D6PW;8uC=W;S7&;g*yhqd$4%Gqwxq0rPV;d;*d1N}G@+d`w`-)s
zN_}Ax6h)<L@3?ynaBknHG3b@tj!N4n;A9qc$-Vh+-IfFuzQCEof+v$^$**@4Ei|@W
zTdbZ6AG~IIX(`#YgRh=6An_e+Gwe{E11PYTNHHH`)F4!xkHv9oHlBYO>+3Wqx9a}@
zv|a8XUWVa2_*lLdqWdl-YQ0<bz9*Mkd?H$aPc%C-Lip`Y*p<zh+(F(7?~~`hy(C9A
zE;IQYeW*?7KEL0Zxm#k*c>ruEPgTKmDN7j{f%3jq_{TefdjjtF%4s}!Y^McU()hOn
zGru#I9p-59e*zR+5<qiB7KLTY3m|2#LTR$-dOla6p-0X>0VId6ud-@CTI(odZ(m3m
z4_)bvIFFbOAm0-(bSr?ZlSSnnVK1ja+lFB}ji!(!di>4Q*XJLzA9K9R^Ol4{R3s7I
zkx9US7bq2RVm0DF?tYKlEhQ9<6kXR}xw*S>^ELP31)Y-t#`07hI)%D4q2@m9T}idM
zKI*vDtwUgVERVlG6d8WGiJN!$Xn)p!*a$vzGpUp?A1kqmAMG+XzcIZ@qb+NXi$JPc
z#tmhhsCjmuS*?d**y(TI%Da7gMLqbnhb)jl_Ct5Kg;3?!X*{YA2NkZ$RXp}F9{?z}
zZK&F{9lVJyK54lk%Ti3+z2U}OeEd8t=Orl2S*)AINgvd^@F1PHvb{ab+*H^H7Q0ns
zs53fF3~qHr`gWfOmMxo8r$uL*;Of`<jYP}VSQ$~xdm&(YHF6AanfpJ4Ynu0?n;=GV
z1v4*o1_oReHZ#;|?pN|!!$@<3?Cs;fMw)Gz($)gX$XH@F2hp;f{;wr1iq`<S?N1Zn
z=j8^P*-q=TdT<s{->2Skp8pOTO{^`F1-l5-e|rH$399e68%mVkHoIs28Jgu5ZD*lR
zuZ`CH)!+UE_NC#+;OLGO_5GbMtl~42-*;hw>lr|x%Nm}SktizXr`V}3UyeQErXZ!R
zKqY;50w@5XxxPn2%dQyodHU3TGl~N4(I5d2`W{NG_U{?$7j95%f*Jx35D<A<zs*hI
z({+rif$|w&w~6#ue!RO0g2mA+ZpI3kAY*m3PTv0FdrbjH^Qi{=xSBTrO*pM~Qjsr9
z9TT*RCNgDLx?G%9KpmL4&)_@iL~3_$zf3vvuun=Y^wlK=Le;NN#}C)0Z$D5Ivt2zq
z8Gkv9l797uHI25M1&)sTG3(<owe#FhhAw^iTExeGr9<-Iae--{oCANAl`~Ad0;Pu!
z)ApK)n0|ZAIvZj5vr6XC^j7(nlSzeoC?O1<60Scwu9tDa`xFFdiUFXLn@Sk9qtg&W
zVtnmW5R#XlV#m`_BiMl*GR0lOO=vT9Z%xh(X2&ZA+zLKdW1qi%Et^?Sv3>W<2JG8)
zg~*WDda7|KvCvWh_VOIKzE&t(8GhdvijN+JX9oB^wO}?n1p^o#HPHZuVtl{^4DA2n
zX#6TjI{9a~@S7^yS}(4iru0V}X80kLr(0EO)URWACDx2Eg-Xd_fn>qgq5(+dsnDj8
zZu;B&%W@v}xRjcG)B%e-VWv!X^0<Qd4I05y>dZh?Z3ibhj}2sKRL%u5HRgszTfNqS
z1hR?G)JSaZ()(sMH>rd!M~Z|X|EJr{_40=a2(=GRcE9iE9bm+A$<g#tvosQm{(zu)
zQ7CC%W!R7H<khM7P<726uC1FLfW@+Q#Yq6EviX-1GV;T3-1&6_Uc(Kz|LX?an$R5s
zK0f0!IzssS@ufuNN#yS`xM?w)=jDm{C=A5CT%gI2;O>d4c&4LZq{Q}F<FSCt7x{9)
zoDxB}(1#X!zY4n#@*#<|ci1;H@?=3jl<Ahy>C)n;^oa{-ITLxHzg))wcg<79S|gHf
zTYAlV%bZMf-dR0?Y4bwTFu8mR!)i!+r6;FuQ7P4F83GR0HI@SPa|>&gMS(MbJ2<#k
z2d%Fv-K8p5B&n*b>|YBEOTO@1L=2DzD(`0l@G@V}rbJ4g6TGf>hMKF<Gl*R<4_lKP
z&yv}-%ils&>x)WIj3`KJaVhQ235`CznmG($jG>{SchPK|t)rtLUK8DY4TL{ExK&P8
zV2FXHlr(guTC<%BiJk(1H_&zi0ODT`ylF@An++;hr7tvY-W9QHTq3fV&?an3@f_)I
zOA>nb6IQWTY!Pu}<k1a0QlnMLVN@WgVT$WY)s&W&*0z9+?p=>X?!8)tF%Hk%u0sE%
zc@SH?enf9l^y$EcWG?-oidX?9_QSsEF)IBEi=rj;N(JnROA>9jM#eXf<$rlPVe<8k
z)ZI*;?~>=%Yluh9mTPS6N`#*Gs#<21JT$*LvA$G#?wY~W$P4%9QLl?7Zwd!kKUtI^
z&3xfH!5Kt|<v(?1ZK$U*&Og_<cCou5c2>^x`Si=`$S~=jPpFz)N-{Z$=mHlrTYeHe
z+PK~;HaNt@#_?qLWPa<T3BB!<89rF>Zb)Yp_t=(`j~R!6i#7N{T(D;Rr`M?$`YPG$
zt5W4TnEPBbMLV2o^Qi(i%XVQC&`Tuz@o*fjai5!4P6f<pGV6hAnOS&&kDZU?$E2IH
zz6py^^#%jk^Ob)dI@+V6W$S7${W_nP3wyj_uz$4qS)$;OpP;?rWsN+R19xLi5xi(?
z2$kyLW@$dJk#|Y=$m<MsgLK|-ne=6;=SNJ5WYQyi%=NuA@<aefmB11Hkz1=sp?Efe
zPE_|5M=DczS^VmF$ku+gJLB&@$J^VUT(XOAOm}Y~Xqxt{wL8M)!=}K{<KKC=4JWZ0
z8g9qf-acE-5VWndM82nC^SNWPV4--*uw|S{8!R>5=22<BRC?owOS$=rV4gm^bJAv=
z6w2~O?nP#e4Y45umJXX4Sze=MmP&E<mC0R0Lk<;mp0HE?zK92RWFLJJ*Qd>&u#&K|
za`Zj_00C_ZQY7;|XL)tk?-h&s_0Lh|R8)k>cyg=C*nZA%%P^F6d$)DKI_W9wiRvlI
z?2Ak@P~3dTa;GatC8<onA(LK|Lfd)feRfVQ{R?}V;Mb;3$>Af*N2v4<(#}38AII^2
z5KV?9$cXo96HaW4bjhheTlt108|Q|nTb8BQvdx2MPC|^rfc7MLVsLBSTX16(I*(Uw
z+T_&A!GBm81GM9L()O1K&Aurb&R_8T8FgU&kDgZQKm&L3g3mKq_9S@*2san2M1CN_
zoN!Io$nMRF+v62_`r?*Z!_h#>!s6>$XYBQ&IA2+dZyC4(EJw`r!gp9{(a~ks2HJ}H
zt3F=4+np2gLa*W)%%<}X?Z=B&fX<<%I#L)+ASNEMAL^T1b5t>ZycSq<=SfZ}E>^qt
zdv|R&Ow0I|ks-3<!U4a}{DvKFhwS>h@ue2}Jz+I@5^~-|eq2<kz3r%8;on_V+qLMs
zL&VX=?_6?98IZ2B>!fxmMrtiF8qW?@B~OP^LEY!agm>0HohUYg6{;frLryB`cbW`g
z;MYh~-{L%USOFI#dY_xWh|+S5xqX*U-iF_EP#doas*(h>$Qo@G=7*4+M--oiGhwzh
z(B~k~C4e2tzSg1HNse!VbE8Y2nWR>UfGQoQNAajQVm6*#B!v=wbF3)nN34{|OsV;D
zC3BFbWmWc4MxgAVW*{gcBmcXO50$z;3jS$<Mw;=@q_e<B(9H~4jrgVF9dbTqB^#ma
zq{Jh`$DAD?(L*3*&_p4qePRjN#K&75%7+^ipo~5@2`lQ(0Go2LEXnrp*iWeLPZ}Oz
zFPwq~d3a&?Aym(7-ecft+3!fhkLqE_sF+;eN$&Sdfm`3HM=RggSdMu8z{oS*Vh%}6
zAGk5>X+-)Rc<+99cR(OZOVY^vK>3uno$Ymzl=-HR;}LP>B?-0fO6}-&2RNe+BbB0;
zkZGk|j2A4lc1V?KRU*NRP#j```dgbGyxWq&7d?3}s~p_mU6T4dfmaY+7J5W1RJeiu
z8mith#wN*`zL>_%zFd`An(#?1$SG3F@{<?`vMjmCfZvTnh&M-8@tL_s>?(t#<(O#z
z4_;-$PejPdjMaIqWYoJ?@GKz_BBB$CYBXD2)6@9TRw86wx<GeH&-iN4Iaq5Q9+=U4
zqyhd+_P(|WGHvS*e-b-N8cu{Y!-zNwSCE3FuxZ;qQA_uYCw}oY(Eq50ej+o%Mc<O)
zaR;^}eAAcK@Bg7#Rw+puXh_IvAD0m@0>dG?*p!EtL7;gW^>_=?_oXc<3nv8Zd$~?>
zj=`*obsrh*(WyLRSAd<~*z*!fFN?$CKbH(;pMQi`PC>R;`0_!X@5nlZ8G^kN#0p>i
z9)7cMd{)qvdm(QYS{ZoL{Ffqc{>=O_qjuA-5BDwq$M^3qc?`Sjt;99{%pT-v1O>pP
z+0KMnn^5Bqm6dlupL=XyOy{(GLaCML)vY1d&z27_E6L#Tr63cb{w~rTU#__KP~ubR
zbwzh@4+l$7;X%zh&fKYdeomFb5ZI{D=T|_^bLgnwr@QV}7FsTM@E#0}U+GNinZ8R-
z9cXkwJOT0k9?s8zYZx$-MQhzwkS@UHfR|$%$R{I{=(nfkCSgXh>_pzZN-k?qI_6_A
z@XRncn$=dY1C^6YmwbwV3c<>WYjb@`GIyv9s0V<!I4ZhY_;1nAHuJroRpi?XjB5wO
zBr^P$Kogf_LW_j2AHSiwAY*8}tUzMnHU5A$i_W=myUIj}jWYwQje@0G)s8!b4$&H9
z!Klfy>FcMARoSXOlhSuo3Ul)W5M;9klN9aMzio!MZlP~4w!iq><&@E}chZ8Z3>uQx
zpy8xt#ho}>uy!?j<l2DIS4`*7`=r6r7s6!{(&Ne6<JT%7$KG%{0x`5==#@=M`UQ$T
zA`~|cG0ZET$O#S%xe6ifmSIN`b3Aa^wXJu-CL<373(1CI@lN+O5Jp^6;G2FsQ(0Zz
zUqaozDH+;~s^y<M8;u)xF1#TN2aSbCMThzu82=3|aoxM-0n*P5kOxi;OjC={G_WCy
zEH0{4IBNTuIy=>^kIGGx3nyp`2~^|dw5*=3EoNcCt9)Gdu;PbzwF2pxNRh>}Ti0Oc
z@q+LRWJ>5TQdg~{MV5xZdiW!>v~V&f$!S(I_z)(j9MQ*WVCznSB67tO%1Xvg<wL<I
z;q*%29XfvIWp^Jd8(#nID-x5B$%GW<O-dnICn1#0M?0`<hY9~JazK@xBikJp-+6&+
zFay}$x`!*U2cDjx=l=vP^u-t$c-?GFO8Y!rh*vLqk64Aml%eiY<IY@TTdvp|yV5wE
zC_Y4foC5{tri?Y{Y$nBbSH#<4ZPLZ!TQ00VaKtuySYj~Ikl`Fwha%J-=q)^cX31P(
zJX<wzOXU7sUi+bjQk+KvH=4J-TPN4Bq@j@IkOX52mKwneO>~*io5xp!NN-;A^uh&o
z4pD#|CeuUc_yLWIn3<m!0MMkCXyh8GF+UzBR_uwP&hU`B_KVHBTlGw9OwKkD=HH67
zd%BI4lu5##A>slI^e#WqxCTJcJe@aeoVrixy3L%mz#+C#{jUEFLJ|6C>>gZK`E?%(
z?w}*|F2ofa_`p)n7$$Kz<SxoDDLtT5$YxqgTc6u`xch@16+2FGQPq65;$MFgT|BJ~
z1r6BQbkLjlc0G$m^WAaAE}_G?9fE0&tE=kC574SBR17X^+L@7424-Z`<F0i!cq1yl
z3wJI?m40U_<SyMnv*~?>2}4CpneR7y67nDTU4vUmcnX^tW)>N@bcm;-!|--?=ZSGR
z`O^peK1r%4M96N;ypIeH(~iO656u$oK#BxT-0OC`HFG6*UXFoOSmQLfb=#A6^WcRE
zohW?hA#`t=_6(O9RXvElCiiHrW%cOXaaEB83z!vlSZvsGT0v&rs2#;^KmUC!;QFkh
zj@dhUFMpGsSP~e@_VoAU?RB``qobFA_7R4TguKQag0!!wBhrPAs2p`fGnQ!yPtVAq
zH0iCDmhO9d#zDE`LVUEyEXp;n^wmzrT+bW25QEeK@c^W<7nA|kK<ntq+KvWF6VWQx
z9QE#D5VYvJy&cq`i{R$wuG#_EvrTMk{&IkU80f$z!LfTzQ>aEm@o?02>O&{?37H3c
z#BuuD5#FzWdCXY$fw#(~vE?(}5eLd90WR@ncK#F_l|K)7Ol@?En2-0D#dFRhb#-%w
zLf@w|BL~hKP!d+k`1sVL&2bBBFtvB2IyUDVQG5_@`@LxqOu48Xi(021%#<4ynDcUM
znPv(?V|X3V+kMShL&6KkIp_Hb9H+iy<R8aT6<yP|uYWLS{rJm;0T_CR-9DK$+*LrV
zC84nqq*`y5aa;b<QZ<C&U4~A*u^iH@bFPyg`<hr|hQUp|4TiqMKPNQX#rFa;GS}AS
zjBYeLD@wNH(|^&94W##!tbIxsE@<YzSHg3K*#ORlb(m|cmHPpGvSGzsn8}IOl=qND
zbt;m3v1>3AeE%gt&MNy)%7*iT<l9xIrVdVa`fScQM!XN_udkw+LkXf^JvASQHad#_
zO0H>G*t2sXxEit#2sut?`G=|7uYqKK|A&tsZBAbA?TMi(0~N^~Jj(LJ1<$}8+(G@w
zf+oZN;Id>M=lb?H>3k1lC~8tjS3+l}O!gUd&`}W}O+0Ur2gNF#I?xUz$z9qpXt8fe
z2w@)$?n~J0_=5eZlK{V|(!TE~&Ry;Iv99E-s`FL?KMV#UnzA)Tj@kI6U$8PVN*X48
z-g9;BI$2<QdwW$_)sv4Ewh!*h@ZH6ijDKyA`R*_L23@flMwREXfk9YSI-Qry)>r9h
zx;Y&R`C<1s2pc~P4A@nSsL{4;uqcdGf4Pt(a_WZ8ts^mU;OEryRe|J;fWT7@yr$dO
z2lxJ`eqxB#)`>*#>9Q?MGDJFa6hIBljMp4i)$nr7g(tuGBL{Ua7sc&~3hG#$(pc7q
zH{RRKg_L&)7xob~<8(6{>{DW%nw;@(&MOk36T5LOzrc$pV@Fg;x1i~cK3?G!&1p>T
zEkan!Ov2wO5n7S81*mx02cDxfrdDsG*=6UJQHC9KK`En{jmt29tm)(9%bCYySh-`W
z9QrSviaf^7X&n#JYT6ZG)M-82C#V}X9dmn3v@c7MP<3$at=Qw8#!kk#-pN7ehsc4M
zZ@lyg#i#>*s#-KpIu@Nov*ZWodW+Fit%;|O!$@4iDl_eZmXWdpEPFvF(N^H{Ak*sU
zDe2$AJWAO;@#E2A&8g|+G`GH=%R7%ZbD20V|KeXx;uSi|P7`A;AWD@e9e)-IhLZs!
zhnIMd!4nmf9D@jR+G#EXc4j`2jl85FX{`#@{nrgXYm{UU$otBdzRMUot5;%dP*h+5
zI)dRob)28=Qz>a&I(qIy?$s54ANoH$G5ZN#^CoPROsN+h*U>TIZAA?Ui&dsiXOP{1
zbfwyrGh`aXfikk(L~_s-SiLOkl<vC}pQnChY+!wEv+sjTID>Mla+eyqwlqv*MS(}X
zF}TZ?D!ee<U4|q4*mi*ue0~$btc1g8ZM5gWLLmku5Hj4_*xec3Ew@a*_n`%7(D^lx
ztv*kN4UxlDk3m<F^k|GvD_8aIThv7u9vK`QY=_du!Ob-Oi|Qs6^B?Cy*;K#eQqVyf
zn4ica3{z<m4A-$?{?@xA<<~GJ2W)$x!y>bR;##leuzk5Vv%!H{&g-zC(?Al9t8>nR
zA$VlndPN_oAH-}{&}y&2t2k&&Vjs5{@>D@J1A*H~YHe|lf4(y$cHafh>+Z(rn#<Vc
zvpM015u?})oeYv9G)bh<Y*|6G$G?h`oq1B<_{cMtLDhqKMKP7xmfEGc)U<s@SI$c^
zyuhYV&nZ7`_C=da&d;~N;zPZiJyfN-%`H;bfX+fPP0Zegs>MV>`!*SB6N?R@?#WdQ
zjc_K0Rr34Km3K*Ka;Zc`+kn|I<SK039vFXdSfmXNz|-Zf!PW0g$Q(ttlwTgj8ZmUD
z{Qf>s3CxHXNdnpM$0EbZw**p-)?EPbqc&tr?)0{@RU6E5IfBV{chG&VFxROLLE`gE
zw}x59k#?|lzU+>PP~LiUg~w9(>5gelAx9d#ri%6>+yD5~p<nLJ#K8Fp5EiSFnpqCi
zg}g$uUIizwUPFZ9-!MGvOqH`I(!O={$)vjJ5xL{;(LVrVPnca`8M;Oq7=ir=6LG-t
z6Kb*!*yeB!paWZOlX&6z5p`iFyAtB{%D0vqv)LbIUoTu;m>j#3;`%arW41R<Jmq=0
zajCHDMAHiIs1A|v$DGuVM%7y5%XL+mZI`xh6MQ)iY8>K&tSA?de_zcNcU+xjw^h20
zHiddT_yL<GT5aer6q6>Oy?SX?JVTo{gx>dQm$Z?xP5CBhHWHAhiQ!SD9;FY0*yf2=
zcN#EeUbmQ03UkJuLerG*lb;g!agld!K4<D9q3VzG?Y!65*Ovr@a^<@XdLgIeu1qR&
z|57UJ7(Te`B+Ynx$LXkNhv5+;K*?%qYQ*#jbimD<$GYnQtnsquc7sJLuX?4d&qeEK
zz4_GBshM|q)bQg?@U>i(N$>dO7s_yg#9@Grt!X+sV4h5Z#L9D#217n4GN_x0pTB}L
zU@<Hw$N*ZE53?82Ct&XO$RVh0BqSJMpw#B|gYM`9qWPQn({h-trvCs0P-|NtbL&H@
z!qOOZr9HN6xDCIf<J>~&UN_Hw8Dqp<l8HtB?7di!i@tj!Nx543Gt7oIi;#W(ydtjc
z@)#0M0L@yan*Q^g=1K6536l@SkKh`;wicjtX2mw=$O%=&JXHy2j4K*D&?wp#+-(K~
zh~(5KkO{!{8(;?NC*{7g1ljlnA{I%(&@-fg3hLP^m=58HHVUh9P{yf+{Cx>Xn%VYa
zo6EAMn*S{fNVvsyP5)cj&%;+P%aI6^)$RRi6%1n!2Dd-7I62cNG@ta~4uu-aIkNa!
zE!$p&adC;XDtrb!`|G|m#}QDYOpa34%s(kBPXWt8+3dlGfjCJ<ye+A_!h1)W0}I}=
z0YT5!P!+|uSPFA16O$BMz=9vJ#)-Uam*J=g`O#k8I@eUneca0aVs&?K`}3o6b5+^n
z9Yc?c5`_;xN<8y~J#e-r+P*Jkpzo4TGiS6Xq%0G#bOHtB_)C0jySw|mOM9&<zK%<h
ztfw3o+-q-SP-W{xZo`s#{vXAXgbMjZB6AfrXPN+4@pG{nK^d42xz{#g{uY?zQj*>#
zE`D6hlbxnSAkL@X9E?xJi)MH2{2Loxmz<JP@8k~2xt3(sKpS$`L@tK5Ch|3!%x&}R
z_2~u&r-r^NRY#6qVo%ZC7<jDb&jv^oljEgy+xtX^KN7+w39L`QfKi!{<k27;QZxo5
z(g9lqO00a=i<FFms#(5toV@$6&2bpsBMB~ONF(<sej-MQ$lO8Z@OFX|BO;y&(>yzx
zm}J+__NpY?GTPSBB&l@*2db@o*zaZcNyb;vWH#=7Jw9|&pydng5?1<iC-NAv%#+70
zaI>l&i?h6_dS843k5eW3U#123z%;E%piT3$&jk{eh<I8InkFGgjxYm0j9@f{AArpj
z5fI#_HDX7NcwlW^ylHKdHhU89V3@!LMpX}WJqjCI<9{k&nBGqDleV6^hb?2|uaX<;
zdUP|4LHPw&m)zdLCfPeZ?s@=buEQsP-I)$G%f)N{5_b65Va2B<dtRSt>aIIY5}2RB
zA-uPJ=71!mgc^FNEoT2lanNR7l0f#<I+vOT!KD=%7|yh{v*XA$oNcYSn;k+-pSt-t
zTI3evIiY*O!ND$tIW=`jbN5fmC3bW$>9y+7vluuCEEX~QRP{SD@~{7hjrCHU-BtDR
z@tNGoopCiGuAeH87U)w%)%fCB&hb8LBRRQ{?eNyO@ibG<$eVjv-#%L@P869vkX>It
z?C`NP-ELEW<aihLm#gEi6I7DUBmnqB{F1wHl8(%u?1E91Lw=$d(FM6xOWW=|n7$~X
zmnw^9!W|qK;AagKbn=lbSslHm)M~S}9*U_?y`pJtavB;+K5tP(K$i5a((&#n++LB|
z{jh4<w_CxL4ZNd>u}_2JNZmOlX`!RFlcJ$nqsph(Yi-Gn*k`C<o2w|F41fazcUuwN
zZQG20j-!r}gsOl-38MvdQ$fmRQv@R(Vd#}#3yzoYzS|QlcuSdpK@am6C?=buy1Gd9
zCl8btl-pwG&3n*sb7yimu~(#w?cqP0GW#Syp1PsEw_}=%1PMTt(h{!qrm$6jHw$^d
zCq%`u`61sG05o4B25%C&pt4r!9OZ;d5ugny2f{U`t`87?b2{&7MkwIEezc@S$U{G(
z>`CR0nH$ghT_UQ3UJapd{HGd!mKng1dcRZcg2V8E!)GxC$g6^E4tmM>f|-o+zvDnN
zl^>c+G6sE+6>z>?&-#iCE;EtsLh?+&42?h$9$8X-0Ot$ZL(&FPs)e%oxiCu3@PY1e
zs9jDSO|wHwAfl9bT8y65LLHX*2BDy!oIU2+%oy^eGUYWnw)s)(yJDEQ{OlIGM#J$C
zB)+#22?T7}h>3SF@Ser4IB4#_MWSn?j@*6Pq1kgZdZ6>Q`DCO3{r6`F+n}+xM%Htf
z=HF{&G9h0pH}U58keki5bFj&RI`ny@?`o#i|JKdw9jOPjdsJ|AYh<J7NV0%!)VYO<
z4{d*g3fD&wq9D5?3W&nWVp0`+<^ikl9ZbG>mWFY|T>a;a^hhueX~<fBs)f3To2S~&
zcbs;#^uAt9I@>au*ZT2Rui=u3q)rEQpbK$KQr>$s|Lyu-fwlj#s3X9jx)Nv<jt^G0
z=T!1jg$Y;YPX7dhKRyHo+1lDZUSm~`zX+SD=vTl7u<!gYAmMb7dnRr67YK1tp&5J5
z5$dA9x*L0l6y)eY0<QeKyMY>W$PlY?_X`?~mOoUJQ&F7AbrXaTYHbBNg>8e*jgCR7
zC8Gdod(@|K-GsW(T_>Xt3?hs?-$BoBM3{MMw!&!%IEG3~OC7gMojSTVJI=2t&+&6S
z(W#DAWXb{V>4UKADlI1X;>Lh%wucmzAfNs&K@6~cVTyWi89_-W)Ac?+Ls|!9TR>}J
z1*6hkmV2Vv0sAhTuNRSJ9a-S1|I^r2hE=t7Yd|`bE=52^I;0d3+#n#`(jiJCAq@(W
z1|=dPDXDaaq=-dGN{NI)NsEMn$Q^6%jmLBD{c+Fp9KTNvd+oVqj5*>R@7uH<dp}|h
zIu(`%pZ(q_eZ6FT7L?LT6>sEKNG+CyAoL5q(BP%ZPrUKvB<7^;SIF8Jac9Mo2OMv@
zc9`k)$;VrD@d8bE%9(3f-(G#(-dJ#IVH)J73(|AMGxGyqdo0rz0J2Nvqp9Lf#&H%t
zALn;G$vSgB9U-%1wz(^V@fR*4v9YHBCS9!Q__cMwa%wbZV8E317b(lk=CZH*{mV<&
z0^D7nb}#+lTp7d^c?lwWm*1&;@_ooUgD)-mSbbN=?3}L?Xwz&{(X~l|)j4`lgH0<8
zKe*>MI|*U@SKD`P9~R?uC}Qzmw_Sd+blj@xfr1Nknv1tZ6y3hL<$W{pjraQs>%O0m
zn#uaodZ76H5o)1ce=Bwmd#EmMmg$+%F`f%(<bSd@|7a(*Hi_F$j`Fcx#T*1qaGUC}
z0(vMnA)(JVOUu{u`SM~jSCwy&P*EzEUWCLqdZ<w);%Vod^)C-s$aIryN>1e(kWSsn
zJ7XuE8T-t)=|NU(t~_hCk56z*Amv=uImWP?50-DGr7a8gPp@x!DnUkswJD2vxi*h0
zyYWCViAR2Ud7o*(&uK+R(%lo5aH<ZWL6g1<_;lp2bNjL{G`tWfHMYpB0V^L_6fRJ=
zEltuoW8=R%FH-zM@%Tn@TuHGgf)N)|U8h2Cu!3~(2j+S2n`!19C7LW=Q?PuPZ*jkq
zA*{gUt;+0OIl4|GGEgKnDzEM}I%qvOaLq@o3UIc2a{uwA$pwNW{m;-{mFk<A7<!_S
zUtC|mEi=fk-_8(<AXvJ-_ndP<@==HT+*6!t%5kk|&)7hzd<JTwNff`_pz-hh%!yS~
zx2&!-hUSiGz3J0VqqJsjNZN*if<l{R@B*dcnKS6y1-$DAa4Hmlhw0l+%U!un4z#;J
z{k1SZClA!+N~=AK&7nAo!o#m|b~?SZR8BnK!#|L4G{7P6-hsv+KbxczI`dSiLx7Yr
zwHm+UmG0LSKq^IzDOA&iZJvnZ(sO01i<LCcjemdv)(YH-?5O_&jEIQ^$+O4iCo8RT
zc^dD8m639W{fllTgPx>qdlRfHElaX^8c;i^q*tr=oz(;tzJ`gOQZCg`2))M^{#7TZ
z>K(iel`~$8BY7h;U9ZmawjP7~x}1nQRzQL*r(tG|%wo6Zode6jL(0%FkxnEiLHuV{
z{M2|Yb;!6Q(cMo+(s99e(+&6NcY?p{%CQzhop|HSR^N!HUen!S+FK>rHMarXNJx$E
zkc#@Udg^tYc;H3^Dp4$~ZfNSW@3`kMozo8BVq@t-Nq>CqhaU7Su{MF8re(;K4(V_f
zrq={&#X@4so50iecVWXanItFIxY)DCn~l(@f4!h*EuHUvK$C-%BBlO^$X<*RiIOxr
zT9NxFS~)>NOLJ9uqB6C{UpCK)Bb=jHS-(oNE~<VmYgB1M;gDIKn{|1C&Bpv-(1)u(
z&L=<n*+>%yNOx^-|F;iOrcchW6YHM!^}x`d6{KLCSI^V`IIlAbtX-MmL$np?TJnJG
zITu7ibMhfz+v?^}rlOZQJM>%ug`m*65L^_kA`loks^y_g(hQwsZ_F2#?V(pedlT;d
zBP>#6pOE|azpzwB*@p>9S&z_{dyEQFrolmir0mK+)x98QPA#=dKVKw}%?LCsUWWXD
z*mp?2{777j<`^9lr!sb&>xbmc+h0GwBc!XBG-Q5SC8llFLjvb|$;VTp3%$9>?Ld}Y
zbQ|$8`g|>8rScbs0iR~>BXM!|Q){rEXHqxXWd2D@RcjzU(N*V>HiRzlSB92c5$sxu
zU?EhHk21z<>*?e|LD5*7!K{r%N#5bm`CLdUPZevYp7pKht5?W{$|0wqLw*-|`b<}f
zZSn2;6)3>WSLpr>2QH>}icc^RkUFy6{G-t^N=L-_1*t5d8T@1t%by>#@5GCjazZTr
zo^2Hxa(a!ICjxey+@U<cWVD9pU%g7H@kLlAx{%6?WvbVx{p-oou8_k~q*AqJZv>ne
zx~K%bNvBP`x{}3avLXaXYijJpMpKch=L(*lucq`y-1y$=d||1$ZJK8Se_?R_ET~S}
zEgbP0vGk(2l1Qi%7sRI4XVnL=GRO)j38MF&)U3>C3QDBVpp=yPf!+xiW3?J8$PyO$
zEYhR?-ZPZi%a{|vj*i~RqV`asb^%xIDg!O_Dc9ElUOGy=E|G1(i(Xw_eTbqu`}mKG
z0!m09c9N*QPGUXq)Z9iSJ&{uVqBs+%gNAxq4%W+7Kgp>M8Qna#WItCM9JeXkyEj1<
zW8e5>SI=))$4`Y@DdgBW>uZb7+BF|c#iT2To~1jO1#eRK%dLsJP=8<`-PqXBuBvJt
z&~#+kbswIrC*#pt`ZDtLn(ivEcXqi6FOh|pq_xPR=)`YhS@S20chba5Eq%<i0+d!e
zx60ToswLHJ%O*6gi`=`)Y*QQ|s}bX3Lf&yu%p}C3+q1-0n&L2h)e}<L=B~q%F9*c%
zeN(=;6%=1^H5jZ=eq%|JNYhwz=#<yqSN6IF2XYfB^&NW8j(D;T{H@Wd{NqOqmYe7G
zO;Vae?IUZXp7&8!2AEx9qF5hqyZ_?q>A{Yej9np>41VX%!Mf`+^DbkCS9d6?wXQ3#
zxJ!>QydOPmxph32;KH#>OQY3EM;c_ljM*%X(-bV365K{;AdAE_wdIFPEnJR=J9HA2
z-)H;<PVzYaafO3v(${O~?sX;#)=pB;GU-hoI&qML@X(o$fmt}#QF@d|is?MI24%vx
zqdgJkT)}{E5jBSt4lA`2fT&UuS3Rb^t@?aRU5v^mHz*NIW!Tzs50qEZdwJkr79r~N
zX7^mqmS>~m2LqWHi1*g+rW{FGu4*E)0#W>kE+tdl{R9<Z@*Ci0)hgacvRuD?^2CBi
zJLJxlt;YfSz%~1JXpX7DbV5B8BL_n%J^o6D{5h!*_ax}tm5}aGU(0$AE_G)yef)6q
zo!=98`ofscpeTo0P6*>;wl6D^DCpTZTn(Mu&@O-{;A!%`I1~~@YsH}*vALAInM&yN
zP@E=?tf7pE13G|)vZWkC19=Jrq5<UD-|ateS8iHa>o_S!v^C=f9&zhhk~WrzN=qA?
zO!0Fz2_Mm6v~lQC{b{0;=Uk4%w^VPZ=KkYyFpglXA!Jfki(-U>s)dO^U(SAqKl`a%
zW$sL)!{~r}u8@O;PpaeTv+dFh3+K%<8)+<P=YxYK0Qlf*d5<dWGMfBUO91jlbuYOP
z)Fg5?k?ya-#uYhZHQEAGUsd~X5@F3oWk$a&tu5kL7SPATY59<oG~PZe6Dmw!tBZA0
z!Z%)+_1yH#(yHq=o|Dg9O9cmnH2yYOK1N`BU9Dv|XV_g?1v?G<o@yXUoeH#sbl<gK
zLjQ=Z3o&9>pb_gpNXq7%Za9C~^Jf1#*RROMS=sHIaJEM!Z$}b+ZPsP(*j?q&Gyf=`
zoDr|({YT7OFB&Jf%#F_;>l637pqr(8(fF)*ujTA+P}b}ArybVrL(4`1m!hm;Dun@d
zNFj<CFo7Rn++HW3n<{{Mx#~2}X^3BSU@;$Vb6eqHk}HKs&gm2Y?bGdTxvNa^G%ucm
z?gi|tS8LItAzy1GbMCQ#tN8buGa;1rArHD^Ft!_F7?7g_+YOZ`PAxRX)A6?#t+>b}
z^puO2GN3ihgOJ^E<7g_9R0Y?nAQ~x(Yu^2<D9+Z)0t@M_*M`+$(PPJ?Ugtr(d{ONu
zfXsg<w~m#t$SJIAt409i1_$pHeBf;`;1`gGq6TRoI+sME^N!TNqVqr$3Bt^YdmCNv
z`?GjJ{+R@cQ{!~A*GR$;>N*I7S?npg-s)8XO_mQ_d<j8??t~J|8=Gk|It2iKRX}Bx
z<>$8#3?J&E39Gj-E3*@rXf)@Y4Q{NLd^cd~a#FMANz;u{(=5~LKE(hWWtcd>R`ud4
zMY>eTEPGvj+H*&rk{h`#DIzzVAEbJEI@zpZ(uF#^ZM=+WXVBSNCA9n5%Mb5YYuJYE
ziOvbZx+GHIl7RF4{<L;p{VV(=PEaX8x2EyZ-YVzPGv{J_QTRGsj%BB{Y%S}ry){LH
zqMZ5u5$R<IqY)#CWY=Z^38I<#kBb2_0j#4L!i{u)O%(Eftq>a1NR6{EFqx3iV@Q4J
z;*?CXJIh7$MVV2{VH!Y;Ig&~<^w_Ce{z!3$#d4*{eD+w#RMYOdo!pl8q^I6b^AD2M
z`47no8`Ae67mwWztB|VoXWs7z(^30ZD^n!!QG$rL6j~RW?L;8kR2t%+!h+|W!WbAa
z8N99$qfMyv02I4?6?UuDL&$2sc+k^}-k%v>Bw#SIfl>GD@ZUK$t5_UaYhj5ca%Rkg
zU>_S8c*>C2mjS({V2?nX619U5n$2?YDlkGTPOg(f6C_s(oYSjTO{Hbfl2wV2EnoXA
z4@%%w<_2TTMP3#Fu|rZgW*>w=BUI$Zk5%*GCu|tkx&@LEk;nUXL3|znAzL@vpiblP
z2uA!|q8XqnZvmTr`Jl{?kLF~pNQ254WKL`#Y}D?VpPzpVl~R}JGoOh?K%JKB(E;qv
zcOijhKTuOsfXw3!SXIJ(3{fN{it?mrAGIfS$aD_U_Zcp0Z~^Ivcmc?*Bw(AH08H%!
zzy;pWeZs1U^g1H@0sxsW=l(uLdA@LdNj1w~egI+vmlMk!%H6_wl2GQJ0T#GZ!1q(=
z^cj*qVhG0_bOc4!f!Cvk037BrLOSe0KL8s~Vdu3T2IeK(zZHqV-p7C?0zpj(6yERd
z?rswvfx158L?QG{c~Q{n_y(Z`L}YqT8z*v=BMuR>t!yNz8+OU&f)J#HAqHd~g%Xi&
z^nD7@)`1gnd<Y&_XJ@|%ubTWdb6W9@3ZkM6h^+vKT+1%WR@t~h!pZk?ub2?Xr;2zl
zn>uov*16^SLgnx6#PyXWf$x|!E#;gT;_)e>>8G&W*Q0EKY(*_+fhZ!)=J&6k+N#js
z{Lu4jBRF>1jXFfs-U3UI<XbN5EG!uZp#w$wSJ1>vD?1WK$%LK^t%Df*RsiZ}5xsvk
zO>2cf?w9W20KZAih<*(aN4mF?a6UvDgPocBK0HQxPfYYoRDl2>*RC3pzRct2OY%#f
z{r2l~GsVvGdIDU67eS!&l*B&;1=wccuxs(&ao_ivNwQ6560v#6e*RY6U{-K1A24wZ
z-8Zq*+;|gp*p|Q&uncCJXMi`eV)=q4W|#eOG~kNry+8M+aFb}!aTEjn-!K8_|JB!m
z2x@^cRQgyp=9kw|U{Wvgp;PX;iWx~qC(fE;Z=y48x?Hkzv8IaGRgj?jK2&rsyjez3
zX;dpHSjHY+04Vf3-aaG4TKaCfDNe`F?~PKbXGB%o6(-}SWgL~=m-p=A;=dg_O#hn~
zdhwSkgs1FM{hk?YS3148<q0X0(}jl%q_2%Tz4&Nf@2%_Jlyvs}^wv!E!d~%H{p&s>
zM;eRBR@i+<Eo%sSpep&jrO(zFG0SC+&)rj|iV7VC2{h}x#`FF82k$io_rX22<b=qf
zT~P4k_04$py~w|$Vb3<aC4I2mU`<-ygpvzP6W<Bpmf2>uf#1Q>9M9CegSCRM0j|v~
zxsQQR0p+Q!)j*rqN4`r#MHAJ~)YKqIASStb$#A>k@y%4DomuW{8MjS7m^6rtd0JWa
zMzjjPFzXyNj-FW9HnHvI6{+0VcyS;h<+663O-Sv<W-fA(os&ozlI^>EX1I|hBcu+)
zT7~GtI>-pF_=wSlNp9}Z*YEri93mQAl7Y*U14ugFmsvWMxwu~0qS!cP06ufo*a=nT
z6%qfvzP)dJ(Pk?m)cCRR8fvl5f}@+~M%LLUCq*TG<ZmKn<^D+d=Lp5aEsDeSriIhw
z{&#p7dytYiq~siZ^UPkKWH0jV8BLs;nXx(U*id8pZLhf=r{Fnx({Oe38OL*YZ4Mc%
zxa&-b)dK?qaU-@SP!8ZFeFcr)-=sQ`#7eI<n4R?Ide!g_{W@#aIf8-Fdhs(0H~X8`
z%_pnub@vXdYLks-`$EdmOR97^fLyart7Ey~wam58FDn+baq*_Xq|o>J0f0i)nXgnZ
znRUDOHmSqL?tnMVQWt~wvq+gE%Kx!K+bZBbp_YMTQ*`uFvtj>sY7V}oOTfk&L1uv1
zF|4Z+%La9z&wAhxpdGECbt)9)!L`)(CiP|3wFrZO-t>MT3SgzXUr34d+G1#tMn81p
zZG0kf1C|cgiiPD4t@%2VIzrws*Yv`xM~GLIp#vwK;Dl{gvNZ&nw}Z!i4BT5|n%&%0
z+B-{!uLzL&E?t1kGl@Cqy*l2x;o(w(&cUO^5qn?75hjGsEM!4@E6JY*a%gIZi~CCF
zh}~lVIIMngu^yr!B|t`m;vi-tCgbAK+eO_jOvH(2(`!it5WVMN4!v=`vO5rmj2_iY
zldR5|O~uCkvBy6~$c?`;B2&F97hm|chSof;*ACs@x=iJwN~OqpF+J-(WJ506$bHa>
zqp?%W(ju|6JB8f6dbqf|BNv#h@I-9h-tk)nE)&H!VMIOBTc)(quJAv6jc>6jCX6Ep
z6`TrSz2+k`5DwcKxpGJZFzXFstt$arQX>A_9=cDTuOZ29q^VIVqjwF#&<m$!V#4r1
z6cvI&=)AvUhxo<cHhoyU1ofsbI)C3@E)K1~@Ox(Clg$SF`K`pi-N_tui@FdAYGACw
zx1@WU9NdxW+DbA5oq@H^nSaC1FQcnCz|0dF+<gR;W4tOic;+O{(6%)w5~(G8C4pKp
z^&98&-=d>Gp3v$A416`4Y*%t?KVKc(806%*_@7!_m%j;++WZSuzR7y`U#Rk<AF!4Z
z#iop{9*V#nJ_SeN79?%uGj#$<we6IVi6s|@XI|<F^xi{yX%){j0odU!<dM{%a_66S
z$5aLBLp4>9Ha~)1#Fmdpv$7um34CObd)y5G3eccgNq?C;GU*!fyn`?ca$`uPq6FBK
zDohN@9zxQn7jXy>PK&FfRp{PxnbZa4_fa~y|8|l98Alm9j>^Bq=OkGw7K$aO`ge`B
z!VRyhj7350S?AZo_zsA}R5P2{hdou)Xk6i)v>m&AlnC`Z=!7MaScOZQ{0K62<ue+X
zqZ*Wa5(8Gv2KnkVcL1wq3-v|@zvU`_EHu?3g3l)83IjRtCRqymT!(bH`NRytebHTp
zDuxcF-L1LoQ=HMT&w@BUo~0$+EE%Kt_qzZuRvI|BvXAg2_gzbVWM65cS&GXan2PqQ
zEpu+SE^+DQL%z4T6HRP#0uI%K3>7Q`I1p7J+v0SfJF6#zD@Vf$sEN<RX!oVUV~>Gd
z4d+|hPp(lK)(+tZ+jv~%rlkS*)mt%>tIAF?_*Gas7{SQf)M1Xp&b{2|)2jgPF>m;T
z*ZU4R*9kbBP{34EYXhf&A5qR!V2kI1zSA6g#}7q(XpmkwRqrer)*S6Swg><T2=NIy
z)7Y^P3&Lb@oaURi#gMjZOkHGtzICz?I6Q!6O5oY)(pN>cOV_3m3{PcU`*w}koTlyS
z*%MEDlXw}@Z~rWD=zT7}q3XFiRn20}uy8?-Z3Qm;_@vs$LR*zh&mOzJyhE5x1fUE`
zH#}V8^$^$stgCrgmnsmof<}<=LRq7}zW8TWuMLp%w<6F!rBHB(sftAZ{3h1bzX%%f
z^gz!6Z~z6}W4+`FauHDw2tu-&GITfa0@f<v;BO<KPKW2->`~(vJUP%B!8)-5Ic}^M
zJG|K`#+9aZm>XhdewomCv3nkYi_ub*RLUCZIND{4`L3~YflTk;*eTdNdWiI0hkf=t
z?7mz_%AipotB!Oa^cb~42<!_f@UUNEVHMOopJW`+)-X;PE|`--;roylEf9%-{;gP$
zmwE3Bn1?*r?UvJq{m7c3qZ)s}?IGCgC8z`TL&(B9siz0=upd%}=2fHJ@6%+nNAkl>
zkRPVmf^rW8ag|6RdX5!@5Z3RpR-*%-eFF<fj0D)<0~#zW*v|ifqe#1b8(N)d<gS;#
zT-sBAQYp9Vv^*|$tHfGH&RD8~-oxPZPiSS4?+@<)qK@@S8?B5}0<OIO@j$l4!>1jv
zhmS2r2;<Wthz34Sbgd-fB&K9?mnCI50x`c_0P--5B>%a87n4gOt=J;fKr`(AzQ|uY
zuLGv&FgqPgdLTq-J{b7MAMX>v3Gsz8rTN^#qOg_7kZjh?<4E8<^3@$)1|35iiXscz
z1pdq-2Uqz10EK+;0hE|3jgbC-A_bBh5m<npNUnXUPgFb>gB*ZhkfSkM?R*AP=PhKP
zoq=DzCT_aVx|l&q0_palBJ<q<TH;J`ysj$m^FX{gj$(%hF;58ALXVNFYyYdU#}J0g
z_b*bczzZ4OgxcDQ=<ucUO<nXnG>^hb$U=i$Z^4;4%w@Iz5UpKg{5PVJmq5}!|4FN`
z#R%h7LjqFBWhS?ylzZUyhDx__o`|=Bk&%h9V$#r1HIE$ZAz^t)X=JwB<Y70nF$b8G
zs8e9MMokTgIQvAyJ0j$&c`WB5<fgxt+>yJ_CJSiJu_v0Firshqv$w2Lk;K3=Tt*qp
zCdtqzKEs>&I@_WVp;Mu~#G%IWxJ2NWdSiawgj)LI+V$_oS9ythl828}Rd&B0v-1DI
zy#7SQ=P0%5+}rUJXAgakk6%cwrrp-BX-`wf!8!9*SzcC8&g}c>P`>)B?N24M&md>@
zh{zWF8Z{hQ#!56$8`&K}aWqPf8A(coDvT$&NX42wA*l&a`?oFeqCQd9e2%1HVfaCO
z;3V2uCNN<;GTE+R>hBJ}f=2Vn^fa$>lZ#mM^G)hNVx`DKA|jnHOhv*G@*_OFHRz1m
zxvkU~F1-cF6V~Ktr9XSz`=0YvG^jwH4@NmbdSz(ym#`!@D7Qh<4wN?-`Y&f`O~--}
z;?e`_<GGw8V8M}^tMx`B>+93K<!pyn3}KR-y^)Y>j8VtSJm4?>raSMybDRuwtfqM`
zWdHn%Bj=Y3I5&}Yz00bgB$7-L($y9Z?sPH}7!x5a<W^Kj-ob|3Y(iZ4{Zw|34cE2V
z-IueQ;Zr*+GCafD=*CNc!9IZBPfGP7#Ga`(Fc1ieRxSpUNOaFQ&=o3c<KaSb9w-u!
zQ_RQ+rXlFlTeJOZDGv|H8+aGMJv}DxbTg}o6w-k~i05)b0}@JikIxY9pLpz`3Ld~)
z&8IT>L6QyqBa|GHN$k3?oE-vQ?1&dFlKPVmLLLG3?;)+x=)PIba3ImM$Vlq7?{qR}
zbukqT{RY`4hV#=@c8W;i614|+=1xxx{r=#TxH0d;{N@sVveE3ZJ3|Dulk-ldr3d#v
zAq*4H?pj>0FD3Qj)jVj4=hruLyk|dc-c<kP#EOBFlkK%J=k!XMO^54(C*QN)?0(Rw
zDOQvAYOC(tA?M}Y#^xiX7zU}1`g;YxGw_q`yEn#=tla4D9IoEC^Mea+2lB@vQfB#@
zC&g^GSW4XFA<6p_zV+u>hfOuslOR_n26(Kz_ouv03x!b_IM~?{i?c?vIMwr-IQlv-
zzPM$*(9;OqXByU5u3VY<eb?rji|K>nYYV&S)jg-9qzk_cjcnAWzmgapDU{$i#kuV6
z56u6YD^1C7EWS2>{jNS~v91z~(!sCTP(Q^T&TVI4!7M2$`4$2?o!$=}o_+p-&ZAY_
z?pG3thYsN)kgg-Wv{h4~t|lfXGXeOF5^gztPb)Va-E?&y^c`X~+PU%WQTFTfSJxaD
zx33mXjclxx=UP`KgNkJ?<vIKBelWQ?llmLapssa>_@=v9ERUPdp#AytT&Bh*!-Tf+
z%G*Jj_JCNjf&Tq?dr^*JyT#=OHI;g`)4E8)wN*L0bAP|+rMG+}Ot!yAbOOEMg}+0|
zSrrAB-<QvrOAp_2?}s)9#R7JK`8X1=(aaZxb?fBC!chg{ga_4SfHlm|ftwXn*Md6W
ztm>z8^Y6Q3F>c6xs)f{B^PnoQ&=6A}@7x_mtIddTEexu#&F-2Nk7rx#k7)l`L8QkQ
zs(e4!Y(GV!qmCqziFbpSf%$@gsC2oE$|8OkofRI^cF`?o;~odI><1ptn9T-64b%1_
zJ3{BqzN@s??HNCMTb$EpZPd%?vEjKaJYOR4LDeu(<Sg{M!kKktrGsR8!9BPWwr4_*
z{<#!5(3iro6QuF9L86+pz^$O2$2WFCAqXFuy&y}~ERhLdV$2n)$Amcw^A?wU3dz(g
z%X5X^4s4!|ucR&LN>vaHmnmrU`R->#_#%fmCk;Fi-@2}2)IfZE^zYipAE@s<J3KzU
zE`#&W<4+;Cc}g>x{VDA{H#$$W*#q(lFqisV<Ulvjc$PtDL8icTGOP-*etrK~zk_(&
z<+tFl?@un~!u)r!0eqUd!^nNVv=4Iysu8A16)tfoM}a4HXiHn0P4<#2oPVOkbKd(*
zM{L&U8EVB|n_9(i?gk}A))6{8^xaW_4KtdMFhk^~dL~~Oz4`^p&nE+?VfVNX$-&>L
zG$|Qr!RpCA?^fS8(0K&N#_@rT(^G)iIJc-zk-LO&AwgEqbGD3UvVVNL{VDnvU>kYK
z3^&B?@y{oELB-@O$Ygk(CpQOrj1h2T;8l+<XM=)mOAC^BIO4dxKSQV}q!H)aT}UD$
zJ_$+;t~UJ_e7!{b6X+0(sTV{6P}<9RR}3)7`FCrjLh`8KC76?xZvAn3JP}-@-xQYE
zYWd;_X2beQ*t_3vU9t>tUtX(6MG;09A;r2edy(2?_UDDsQX!@+g`v^$K!Cw;S<cg}
zqk2JO$+28!impTIn?k0*4}ly&J)eI?f=?J3xD&xzcGwMDs=@3;WS#R40zGlzm7{Zi
zM8n18krR~jXL%XnAX%j6mOTvXuAFJtnKL5syw}T&{FGDGGCf`^X-H=d{_=MrPhW02
zJOoOc=H5Xzahl-o#KJCByTx1l=eS?Pn}*3KBulI5@Mh&d`TFdqnY(i8kmcLQMni?M
z+h3Fk2rmW(ow^vp#6=*OE3Ed+#iD&ELrGA%%KI}7`sU~a{6HAg&p^Y$q_xZ~ON)Od
z)ba%Bk{KzK6hDiFQHQAjj)oVgDGy;hVowiz?}L{lKnvid^Hkd>Foc)!40Dm|%K~D{
zKd5x{4k1S_S22icSK(;G*@25532I|#gL==g=MUL4{5Pa6KwSQUG-gRmaSI;GkAUW*
zxZPNnK1&u9;wOJA4n6jH?82*JFpL$!o?(tlyqd_>iG25)d!zfi6_qiJA5M^*-}zH<
z+Y#<`5m8f3uKFNq=xwMtK~9AzQ2rmExeD2gITO;sJE2fzb%GjmIAO9ZZNlzQHXc4E
z_b-P~s$mYhQaXx3JI`eEqnz=COqfHNp`Hd%sD=Apr+clf_@#;T1wW5f-n&g&6=zja
zmK`dkrX>N7_Q#j~U56W*q0^0*#tkL{cUMSZKRnuxe1gc4sO`D3ZQK6a=n>(<89TaR
zBSjj|+oNOnR&LhldzL@x0)*`K7wjk^p#c6MR|pEVZ^hp;z!?;7s_}HtEf09Q@Us-`
zu_@3Ee8|@F1*RbkSJZzaAe^9ZO4lr{C|GKU%A8GI>IRVgBVh>ZNAF0AnB+B>;m$7#
zAn=DBr`9z**c^a7hFmV_LeP-Dkkq6NkAUDq+s=qU6YYOzY=OyYvQ_wJ&&6<IW39sw
zYq`(L&1aQuU#C-{<!Qfmagj1C3)W?Yc+TbjVNvtU@tWhcnve>aLAQ!s9ln&s4>&^-
z80!OJ8y*SXE3l{$=#UA!XkZF8Xy<3Jxwmd79gSGe=MZnGcWvASfiY>x;4|3>^<7~T
z$u205x$n@P1ThNl)nVGvjZpM)B17Bi{{~e+`+)?RxqVWOnIB^1auK#$CIj~W6o+Bz
zF3Mqw-P<4@C?y5cduv5b1xhJ*IEmMl*w#<Yl`|6T?=bA$hsqd&Tw_r+ll9sC4KZ*8
z*$~emW=g+o3ytM4{UH!a|2J;L33`5_%xlR{#6j=Dk{l*fJD-Jc&eI>zq~wZ#8=23P
z7F!~Q!G&DND0X`A9`e?d67C-6>)IQ60*0cq0LAr{#qaNc)M><$pYUS|pCWfE`32If
zdj+T`{!d|^NFu#IVJZ+JAs9j-=WeNmuDg}=u+vXhaUBSX!3mD>#{1{+FyT&y4S>_R
z2N2>)h}iZa#0n&odXKr71@?CtcHQ!v5JGrtXAJuMiK(>IQajv%J`2uKBa(f>Tx^C9
z31zXfmlFgXvh%!<&ankyICPK5DY!{4Tm3~S$j!#4od@c5%bWn_1bHuMRyVM{Q_|Qk
zhbK7%Vs9mwJ1-ju_sJpQ$#+(^1ytq8T`&NZ{pEr_MvlO+vR0bXrXY|fbk;91wNFk-
zL1beW9VKetGyf1p8s=)UEy^APk@2lVtmYatar;36TL?ZJ;9Hs*Nk7z~T8-`Cv`&Y$
zl5+_{WUFSN$|#Cd&JyY0M6)<X1DOI+SQQv16-fuVZTEc=yu|rnb!)&XUUvtkfc1i?
zykkWBlVTMOlfvwSDTJp(&bIjq7bKwlTta@@ecxU^%JO_@-ShdajnN{Rv8U$V2YX)J
zNaH1N;Lq#=0b3gC874d#T>MPU?0av(4;I;)ym`~y$!Oh*U(XMq)QD7<OZts9<?msF
zH=b`7&(d|LAuaRz<n@lr=Z(=--1p)`0BM3|$uHrn`}r~)u(9A9Fr4sZ>IR+qq_cdf
zP`?i@dqjf}vtGDdmYMn77!RaV>=BJQg3OXVK&6(_g(0|YeqKjQOSc5sX^;@uFYWSg
z0yHie;g<mD=4<HKWP$FDi6b7pvSI4`trf6lYuH=xBCMz)97s$VT3Cs#>XWVO$9U#7
z;4QzJ!Ypic2hkxb-q5(LrL5nIlp#y~mmGnB4r!>Jf#%xE;c;DCTU#rpt5>gz0{MA5
z^gsGv^{>wSf=G}+&gHfKeGYB?^Z#+JfRwF8Zsd?3wU-;w#s2LUMDpUp5Xg~PM@!Cg
zHN4jPyaK2`A?V$X><}<4V>1>x1MCGzm3b~%3V~iDaJy<+o=Z_hQBZZSZFq~2%VF-Q
zxr9LQD4+=Z@?mN`mN+9@4P3I#EA`VIUv*BMnk!iW-pJ)%q^(kNDClLHDsUi_N-eGx
zI0B@r@P6J60FP0Hni1ds<9B)E14_FZ)8S7SWK)0rO-v#O9I1e}n?OroTN`Q!?ZZ)c
z#s&GF3quNtDapxMS6CWB=xEb)ITKh!%HNDjM;(#qzX~*$GteNhB%W;V?YR%2frEbx
zX<=G9U<*1w2=zp<H2UFRflaz7keB1L>Gx(jtsT<(oa#VO?=FAhMmA{tQpX<D!}Sqv
zI3^P9+LAIomXiHI0|)m4#^wq@dh#>)co@iq-WPmg9uXXxMvt#L-vKQ;BE@Zo&~Roo
zIFCI&MN0Xuzn^-VL29+Ls-}jXR}q^i`g2JiCAq}DNTt0;+7`PNVe#u2wLk05$eT#h
zb7rw@9?AmQ(i~1spZWnkI&Qf>Ix*o~WngKE$kp{V-`dXQRc((t9dTv&VfDnoskV;B
zhFvMK<#fK~>enxudc;<YR3K>IG#@_FAkou*qoLekqEdV|c}1n`=V<i;a3rXENoPKU
zWLr~Z<TV7jlo(_)R5DI^TI>1oE)dsJhB)!0k`J&8%m7P&l&Rd-sH5Qb%tQWME75-!
zda2Y9MGAqg8=&ha`e=92wqeSa<9tbtUrfdMOTduT(|w64pW`FZN(XgpQ9zA-JAUWM
z?L7?_Co}8A|FlHe4CLK^oXe54!gN(^Xm3k`Kkg(HmAjtw7;vmZW02&RWA`<ZMcbXl
zEk-i-E-VL33-IZ7*9CW{g5HK%!5mqLcnwl1MOe`?Eru#2y^O-*t;D)@Vohpl{17Ig
z>KGgQ#WmsHLkdQflnyHfyT!JBQReQDK=9mEckJ4QT)^US)tg_sCG%KQdDDP%0V(qB
zcn0ZeH!fYvpp9qvRV*epoC19v!0%n}C;E!ceJ?vXB-QhiG%Zmz2IOSRU?Q7-=u_QV
zd@D;*$gFs~k-@MzX?p$6q(K2|iTUrsO8QNZ#dTazskBxdDg33s9dPn|;D^cjm0~R#
zyj<`z$bK{e-7gvu(iWw;(No*o!C*mgk9Ki7*|Fp<UkG)`Y^u9P?af|<WliSet}v;C
z?!)!WYc04$v^F0E?*NdqXY8D29gQ;OJejU{NtGFDQ0vxM1PA%AEj0>?(mAd57eJ}n
z%(s2+DQ&UVc?Qu*H{G6Owz$L10XXfWlTWWN*jyW<4&A=mIXbd&PJLw~Cjj5ho+QAO
zWWsRBU(clsC<*ApENv#`Gw11PtkSvkPVpkKZIKp2oadW)iM;eyme!kQp`u2v0T$6i
zEG$8*WS8o~&eRu<3=S{6$*<<ozn{Y@K^00Tt9%??p8bg6MrIH)C|Tzl1sMXl_@98n
z_{yt{om^vFtLD`(!G|+8yX%8uKZDDp7<yZoe<xn|$DlO-y5QI2!OE9T)#uj&B<VYl
z3J*N>40`uO^E3iE{&wIwt^Nb8fl`0B1@~hWWsx_3gLvnKCt*8GiVSGL8OGLGQ?J%$
zYk?7ei33T8X=+9yB$I2!?V;bAkf7aBkeAyl-bRB@4J9o;94R2q(&Z77J#}qs{hekC
z!l`vXVl-94J@Qz1M{yRg$JNC-2)4%a8YJ*Il-3Em-Mw<k|K3xseTM+W1U67Fem~sv
zuR6XRty?D+mL6eS^;qcF(Rj$QD(FNH)4t~23eFB4qD?q#r`TE+yPW-5GZ;k;PBDQV
zU7W%_w|<eHvsav-ej;pezOayp5HbKY&7%2fc3d9WJTXl;!j*XC4IRdZbEGNae|0T2
z@TwmD$GVdt%{7hCuaFB>oV&HTbLtsP@|G48K?6*7h|Q!PPJWKSrNU*6c4Q7;T@nIz
zstV(@Q&$wLuI++ehhwzN_MG#=IcW9LF1rI-W|bd=t@guBXi~&h6%^fvS`jQP_;NnT
z7jR5ORoy=k)c-qDd7_SRRR6>?i`{vH`zdp8pp&KIRfZ5kWvPvykI*?O9;*~825$#p
zUKKZ?+5lOpmeymb`=ur9gn}zd-&)~1%@DBG&KV^hY|uX{=VXxS87j?DcYRwRmy`!V
z6B(*zkI9uj{Y=_mn;DuFaP2xMM0CO3$ViZOpP+oi_;vbnm*=1Oq}K$A4?Mv&egI|1
zx53#SU=}y8{6;w1^J;1$V^E9g{X<mEY<esNmG|iovieF#5)ivJKKxuIxOdXwb4JKP
zVd&NOadAG6=^{ovJOH<9&=zE(ufGIvE4Ozyvx0a4Rkir#gxw2Ag1~$$$@#qOtD)}@
zwT&Xf@?#_GRWD>b4mqj#!MuFtF;@Oq^(S{+N=ymztx#;jUUeF<Y%<*ueBOwP?T%rC
zj~GEj9S9?K0br#IsBYc@yCVo=(kz|7K7j^1qmK~~1Y!u@qTz<qxD6?tPb(ManxAtU
z7LGuwI8<p+4ysrDBHF!&{w{p~NoXj8M{rUh0j=1Kejtj1dbK!|86x1%%1~(Qw04B4
z^sRvUhTL?yZxdGx;RBptSD~-&Xto`+R~aLR%*z4>bEpQ@jxd+GNN*<Dhu==)yx7lZ
zLG2AqgEx>`4MkDt!}dgAltecV&LITN*tnV>oAoAvlJpIPq!vO1L!Sa+yzLJh|J>=A
zbw&}wHT?Etj)>Ktb=Oix)_n;^@Oa^m|9p`GieQ6k1T6biFdV$E93SN~6)DkGD-BRi
z@84<emg}JkHF{|M8WgRu_K7eMf&8BL8fu*&On(3Ln(9}=I?^CgBx{9pO#Lq-_7YQ*
znWu)hmJO-+>t!Upe(h&+o0Is2GL#aO`*WT4ABug|i)v8VrqeuVhQ$uU)9g7unvjSA
zf#pk)uP}JkL$CtIM`YQUU|(20?dSP3OKoRyIY|kQp+Rc?7VF}@YBQ7aI7+Edg1+{0
zY~M=(R4JMYuBWLIp%2xq7l)YiCOqr6NDje2qX{~S@qzVx!3TCgf>Kty*Ywa#OoE7J
z2fNz;2W$$634oAyRi2;RUo@b;ph9*GJ&d>A6L!q2L!coJA?aP1sguF+iu<WYI70Qu
zVd6_LFonE&GSPP|=hxU^)rebF`g8dI9u-n)7d>Xnua^Pkp;&M$)RKjk{xTgNsu;&k
z1<IZkZdtt^UJ=o1#CnlXfaSRrd+X2OFdIDwF)|)q3{Cr?7NDmUkRTGOD|eq5=(lR{
zg`T2}rbg0k_`Fen>>ywO5+GdSVMTA1&mU~Ax^U|p{9jdM*uHu(uvzMF-(_1&C##D`
za@a>yjsHsT%yDA2wu9h{Ae+sx%15aDgE03H@e-jBh~eJ@)+$7366W6u3;dZ85cGzf
zeI8;Y70y`aC&94RbnoOj|4c2ahWLzjo;FmXAcHxS64-Z;7+@)Zmu61x$9=HFrm17N
zAV48MN&$Gor2z^GzS|~d5kWe7=KHJ>4ZkKDodvY>q#K9+4LJNBV`{cUL6y#H!M$j_
zI(AZGVSZHgxo(M>w}XZ1#_>`%PG~BJbEa^upYfk}LFJNERb9PeY+~>V0FG@&WLLBq
zZ8p_u3!G2MLjIDj5)A+(MT?a%kiK4Z;3D;0Bs(vkc_)VO5v(NPzpMi~$PSWa4-9kF
zDs$KUGdDOvZp6^?-xqo0y<Kr=!G(+@w>fS4|D6B|MA0EoC^~gQvwY-GB`2v60?Gq>
zkR2Wc0(wccbIwQI(%Bx&pTo`HOqSDNqj-Db=>JSK4n8%KdGAZ0t+!AoOjc$*(Pf*C
z;IjAVasIuEF&80kQ7QR7a=*?4u!dBK$p*9>q<YmlI3tz~5V-uH1_jGMMu2keF0$Vk
z>h8^$UWjK{(H*foU1DH9End2{NgXL(dWdE9FLu+CT4}y3-~E8qC!+!!i}-0}$G?{Q
zfEt@gkPRWDKy4v{g-(8SneZRuRw0WyKey7hQ6ldjN-0VJ(11t8XR%J?pY=mtC`cAE
z8m?O3+}-zQ3Vx=7+z}47UO-9_OYJ1f2In)<$aO!_4<FMyfJBN~!03_;C5!_PUQj{<
zMR3{t<IsbD+41#6)DkEP2$Ia7&k%*Vu$Mu|2KSGG(kbk1pqh-=mGe;AOp^!@3OQHO
z=Q#fS4rWh;BR1(;j_5zH$EKx@<WjcdOHd+%^0CbW$Kdh&-8@(RI+AUOF%pKFG%cK&
z!}y%gi37(9H&660^)D@d{C-Be$J<=}*~ZABu4dvFPVes4&^0iyWoBj$kIU=28f9cZ
zQf8TSc&9h>{L;v6a;D1`m1{Y5`u*4af#l1GsfOjr#0oUAB{5eL7G7KMN6Me4H~FGX
zt_<6n4!thfsy}LUxOT_1bZYhk1T>)D6;+<A<x=24AxiU0@Ll5rBw37asbf`GHD0t!
z1JHW3D7y>7*YG-;#U!^f_G)daklLoe;Ja-|1K6hfZyHt3reC*NR@K#4=tK0%)@z%g
z<Itez@HL$4u@F6FrXgZgLYBcvXRdw5i7iPtv3BYnE;g|>_xVJyb6Uo2sx4go<xp|j
z$sbK)<n6m&PsM&%+AEbk)jF4bu$`og{X8C(vq()cs0Q60x^>GEfY-O!ZbsNUJJ;?q
z)v7nV?;Ua;9j!@@dS0kwvRd!!Zo4`sWFj7R`hEY=Dw@>VTJPlI=jZc}Kc!!P8QLSx
zdq$V^5mg8~P5J|vWwmIhi`7Mk9(?=q#^B;3!Yq;E;$n2>3#8QwkWMk8&k18!Vn`6#
zweFuBiKf)QvYn4kO}Op^vAF`L8q;nuD`>hMxF)r%O@I$cOkjk14`P}H8K*ti?eTaG
zN~#_$Abu;+M#I}XuU&Ka`IYZI++-Ij!!J8`h|gttzXXBZiy01(057!RnIrzsz6;_A
zWIJLw@n3u%Y7p@ey7YRf{y8#eJ99ww60|^ymeo!<Q!<{x8+AT^pMg#b=I+!1?<>oq
zdHA$yHozNEYx$L^`uNxl6tUg)`=Givq?RV>iCTW6PWT7~;7RDzG$RyiY5S<#rhf=h
zh|Wl-s0gsKU!_Rxle>gNOit1)bo-PO(MOvxyg$iMteGpgjU$jfD8jIg|L_0O5(p$u
z2f3V`|NHKK#AT=aH{KFRf91sWJ?H2?4+17Vz}os=?V-?fnOg^0Mi&yQ>Foy#F53KD
zoA%mG8aiP=j`wx1*89mR9a<-xH>}Qe35Ta>@#QIQA0Z8Hk;9*kK8DABn4RZNc<3Bw
z^N)5$8j5gSl1o!MD5Dd2om8~)DAY712`}I2J8WEB!yTU%bRX|5C8ecTde>f`u^?Gg
zV-*{GmYkmSq_DnY4-ceAsMF$_19rFjfvx!|nOfQ#_2~wRQnLn|?!j}vf1GC)bFk|6
zo}`P#<KSrdnM!N&(Tj(VPjt3S<80|xz-hbQmS5^qs2@j^<RhG@?(m^djPJv8ajAPh
zx<IkN!tw=SUBIrNQai+Byev{)ah&Ns$7V8wuWv2KLyXw*zS4H-tjFH#sKct0&HeyQ
z7T9z<`a|;)P?Wf>ew6(RyaQvn0HS&37x_f~=I@_Rip(1HYn;b<n0rl>Z)(kF&>X)I
z9mjLAz%b(HVH9rnERMc0o&sRRL-{P4O%Jllm|B$@S39(WZMhOKyC$I(8cjfag1?0;
zTlrmiHY@%*&l)HgbHIyck2U(fqg4)h3aIxK*6>jWpP9Ez?`xb=>}7k0vX$&XudTwW
zox09#Iy~)mv_`=>0770ka2vqNuUYKm{U{RX9t9GvNqH}e3icKmiP(RraeB*;-F=1n
zr`rMP*-777mUqloqweZw$`QDi%)ernp4?iQ6l9U~$ncy@6k}}AO6@6+3T_NDe+wdb
z>2{sX@l~AB(MJlE1sCVNmw}Ww2MGDHCcl2V+Y{tUm3}-X8HYbK|7W>3H>q9O%rBk3
zgdkv)O={6>xa+_`yW6)14K~Sw5207P14N1=s%VXi<^8xuTZKUCPeeUT=A6e|45Thc
z@WWtfXm&rnZ@TVnrmz+i*i58fYVI^IasRW{>|MK`wkE|Zh~&??6RZJ0EWZwyKKja2
zz`v6s?i33RD-Sb}&g+>Q!|n;R<S_l026TCH-}Zp!^!wxsz*(rSnz8x0y`zJANv2@I
z-QB%QBH&{SZe?)|y(G<}0~!=ZkAIAU_n0^ntIAEhI8uHd7{DFBcR*hJ8Pdspa^zj%
z-O>6dX0J!99c_mgXldI(qihON_`>`|`q!$%EJ4ZnvGpos@?(H@N8Px}&eE8%I$qm^
zwmR3cCtZ4`Q^kbUZ)4tit#<9B%U5DMv$yHeCSFUUbUJ~C{GkKB#0DAZc4olU`aZ7C
zwkKT}lxN?$m6EMrj;<kh)C_{%)izjYpq?=TH3$^TmB;i6-z7vG0zi^!2Cq2af@b<z
zBZQi?fy3gNyQhqHll1v#lhv>GK-IUu&fRwBtO17%k?@zMR_%l$RG-(6Sbu+Wil7{V
zb(O4m#Yjd=X}+~B2>B~lIQv_AlLEHCa_*$dQmlKhC9mq$I-4<_am_F)DvtIWjMekQ
z8!fhje)h+Ot6<sHW!KXm_|SMBEsfi&dAAcPX5ZQR@p;H`dSC#T)oJi@s_Z5`dfd59
z&VtxgW+#hTC5KY-W6GeqUWiQ74`%L*rRHYaTPrU5UkT1OwR};#aqx_oPh{X}Jbdh0
z*m-NHj|ykVrkM0EXY2C@1u)rn2F0g7PczT=P_5S}wR7kfu+QnI=?x-YTU%S)QLm`1
zwEq0^?lp!xlJ`<N3=W$&-!RfhleF&08GkWO7}fXP%w$9CYYQExX1eITpFNN8QhR<E
zC4T0L-$*O=Z6F94S`c>{9q_Yzv{fJ2gPRgYMC`-UA|0CU7HgzT3cXJX>#yA$g_7(_
zi_H|o>=S8cMXCMX)t@x8A&9b(9l5#vMYo~CuJ*A~c2<~E_q)rqy(Q0*t7JUZ4VMIp
z-i5wQ$@}h;`>Q(bqKk2XezCCv{}|-YofMA_MXlfP{QWet{y49yY2f#|m+h)c1fGG6
zX&uVXYCYGN?UWCk<NX@wtpC})CHA8r-UKNnrQd?GGY3~jad4k8VR4pS;YZky4(aoW
zF=B$56Axi;*jJi<FE_?1-fi6*%GZ5^A1XO}KOl23Py6@{n)JlLK*C@|!_b}}{G#or
zSDD=}6oS8xz3*pOXdHGQRPm#_J!OI0ZM!!VHOx_HT31oG%DnWsa3i@p{Rp$X@!cL=
z)!=6!ZI43D6?4}8)cOcKl9Y}<+faSz5y(8~RE!(^7#ej(JI9Znsjc(*b__T3u-+-b
zVF~sJ>I1$NZn@7MiE%_GoEVbM{xs|GH8Gv~DT%90L%Y?oTT@_z|BlyzjF2WVmx;00
zAKhCXf5gR0Cv`oP`6Sh9<}@x#{v!0goPO1hd&T39AFl*wxwqHx>Y>|vgF$>nIoXfy
z2Xms*h)TPQs5y1KQwtM}{r1$Sq%)}-1%K@Y^d3DUnHJ|+z6qq-<aHKj?n^#t-nE#F
z>pcg%?sq<!%IESx4UJv+uMbVD9W6C9G};c~%`&K-b&r>*WoKJ|TA{)vHQyMT5%I~I
zI$x@L@Nymwt4&hdb-9S1oz~m&b6l|#nv%dfw4;Bp>73~-v*%;^`P6N%j&F=Xb9f$1
zQOhqz14-vzegT=3Lv<aDG6hMV1UiRK^<D47>xljK>T!3-4t-g@{{qzsx8{=-POhH9
zl~bubUv*cQ=lp@a=yH_&Xol<os9&kwKQQH<K}<qMxhfwR*h`}*{7nvZZm#(5%=nt)
zyT?(YS4VoCs*Or~p*w1Hi(+0WPEt(P$IXkTTCtvN^7YTLx~PdGnkr@}PL`S(>d3g1
zA?+V@pOrqmYKWA)XV7u;1BZ@xxBJBv*<-m+P{&qC=v<oxe~@qWT$dxV-vMTx#RORj
zLQeJM6Glcx3ZzL&X;KR*JweCf_)u&x9-+_>&Rp&c+)P5ZzJ4%7W(VZPZ-OmegMQ5<
zz7DvE>-}q;$B)^3{dM{76EZFf0V*b4<GgV{aA0SXZy!9UV~Sk49m}0`cgGG*6`2GO
zZ&yf4bJIThlrEIuaPrx&wFnYI&kS*O$-uV=fA+7|`=qaDQ77mU*kkH(a4qL=O?^Co
zOEE;xXUGOQ*;BcFtfZ1+LGF5etC#0PHzfr#yRX>!GljW)xl!`Yz_49HFt$4K*>8r7
z?p=!u&Kw@md<C<PRhbKEvm&zvDR!%g?lMFT2jTX&MGAI?TLd*Olji9l2^({L_r<*!
z%zX&=2yW)-=&8>^ffEbg$J~1qy14uC-%0W$d7bWCzMWikQMpt|ap!s72Ya$CFZI4U
zVnNfV84+109aH893q_{o_40ohczx~Xn9uptUHJUl?0hqa=d$ycEwS#B&qH-{`ZlxM
zN{^VFcgoKig6(=<AZTmvxxRvkD{TGTS{^s?kk>ALX9u^lon+B@HeX!4)~ThDcUN#J
zh^64F)hY=K3lnyqQIqgmeE427qy>1#*p~y<Z}K^yd^YAUN?hM8BB4A^hYHbfbh|!#
zqV9f;wb3s;N$%?3yzTO7i8(dirPGZq?o(}4eXKFg?e%AJ79=dj<NB{i%(}16UmGql
zdwk*E#7ySo4#IA-K0nA?a^RCYo4`qNIX26nxx~xV-O((RTt>KQ`mLmdfcNA^*xkpT
zXCw?T1ujoCeW+d?rMvj7QB+~!)_gK9?~UWAw4KL1)I3VSuCX*Ub%5yL$oSh<kvh+X
zggLGk$yH*{fP(c?%IeuD{4<Y2E++Z)WJKeXOg<T9J_BXIry;V^Ix<AT7boMy&%pV9
zp3keod@|<IBSF7hZC#*eQa{}%e<8X7+~KABZ#{&VcPmIWpY8o_n3|qWdUdft?;1<a
z$s6a=cXR_Kw``kU>hd@SaygZQuyD^(glM-vysx7-?#jC-_u_aBy6ckP*q?0qc_fu|
zkj)jywP4ACQ%Mzy{4_E+GDOrTuMpv1ChfHrJxO)=hc=2R_$L%Ema^<sTPmzO7ias9
zlCww-I4cauT6-9iHeJp?Cp>w5MOtL*{_nQl!bIR-Oo0CB_aWV9fk8&w#pm8;ebM*Z
zIG)S*{FENkO8c$H05#lekI4ye7nvwkdfLI`xNH6t;2;kWN@}a(5KTniy1kKn2VW`h
zyTqCJYWG=<zU7JJ_1~^(Zb~c1Bjc++)K;o;KRQtU((oSYGtAiCl=9n$@T)pUKr6#0
zQGSKVZQ`_5>7CUKi)R5|sJueYOEkNu>Kq1n+yy^D{Y9-Xm@u?E|75-Qioob+6lpV#
z=h3&414GT=EJUPdL-C(#Hz<Fv$~Mrvw>(zn5fH2yIrD#U*(P<BA<A~%2#$*PaA1`7
z{-V4sy+}^{^T?eGQBIUfYG6-J3D#`h4gAfxF1)caIXU(yr#O1H=Mh8GA>}>${`PA`
z6wLAl+JevhV;DS_LsXQqh)0=!AOM1+cz!OWzso!bLq-Xl+G5xr4+`B~e9w`AT>T|a
zKh(OU1bt}a;dpewTsig--X2NZ&sxo`o5!u$f^qN^KX%0Px~)wq_fe^Vj-Si#Ypgnc
z&m{7`Um7aiIIepEXUO@BWK`sD4(l46S$CVkbB_)rU7S4P^o^L3lKviV-LXI%JYu%}
zKbjdX#c59(P!Y)E;NnxF_WuaK5ExkIW7<yl_YWyoo4}H+6yCu@p7E7GqqYM4&WC*g
z?{<_VBr%`<#R<3XGs#SO_zKi(V(6xppC8CSUTjh;ezUT&avJIkOzyvqaAAg!%ioeA
zoNIQVQ3n1VlZgkAllwsji0CD@!9i+=j(AQ<1}+lp7{Uz#T|Ru?;P6I&6GPCj#ZBqY
z$z}~+?Tha#kC$9w$G#Rg0A!i9=~K6m_oTcX*iC0pY%B(~{BrWc;%a`Ii`+<zH8y4{
z0oDhK^NMTVzkg}FtB!qpHaaj+dg~($$k(!Q2kq``sY1wNy|cO0U+rihb%^DX&FaVJ
zv5ASCDW_e;gR+`piLmqeu@1*fQJRzEJTk%{Lp->I)}VPZ6hEY1GfCLm3iL8`N=zld
zRy!A6HN1yi2X4|{xrfKt8jd4t;v*<@<A<x!vw)^lDQ^uPgObHvvY6x8H)F&#CiXp!
zUb`O}Y@D1I1T31b)<<ubxY9ya7Wqz5YBn~t?2aid<S!mnj6#<J2a1ezJ<{c<vCp7=
z1Y?ZoNl2wa78|+(C!3pSVHKUTC696MPjw?hMp*B4T{rZmP}05x>81*oNtrsYrQ{;h
zI^kDiBfwCCp06={V&ZWt!)WAPv>fnOns-WeM-k<0*6;g!sgW+DGjD%AVzOG`*yTdk
zt=qy7{{!7zmG~)GKAmjEC|Y+L%+lgh62tcknTF1y=QWpS6{@ozo@dbKl$B|L;JtQ%
z_2*aU;ZrIZz+!n68%2}Bc!)CsHN>dw<^1ZS2+M@Ig^@Q#vB3ZjXsgpw!!NXv2Z2nM
zl7Lw~ty>vX5MwD0i?_i&C719WANxK!3^0M?F-OoZcb4+0|4t1sy-S~X0=x?ss>r@V
zkYxqw2=)aX{|XS4Liig^p9M${7LBl?pAggnZZKiZJl7fc_84=#Lj7Xl?*}FCIjl{!
zuYur&N|WEl%TDKP*g6TaA%x1{%iX6}o?ZCPi21k&^8H)PGe3|83?x=d5$Dgz0kwq$
zh;vooi>;xRp~C*%Ns&>zHbv#L5`OD>fC5qk-J3@zHf<O%i_)(OwyA0ULAETc0Umdm
zZO>7fRSB|lK&zjxlN)X%3bCBXnw_(AiTi_4*4f!vB*icd!7#~6=RwR~&dCOMMRJJe
zO(?7^zFLa}RTF)aTo~yZz@|Q2=&Q8Vgxq8%LY82j&=vx3(*g8v+(;e+h$T2Gct2pw
zOw172YkRY%BS&C5KHkJl^9cC%`4u|>3p7!{DB8^RYpxnGow^))^?ikc#v>Ug!F#ho
zH8L`wzbcIIg#<==$$HMPXdb;9f<7&%dYInFj$Ll1$l9Xtf%c3WXg|G`?4PJ5QLmXk
zva8i@T3LZQG;c+>QxIXwA(<Z;i*Scu?axvmqtV;^`tFSrLjwA!vBhY^;FvCsAK^m=
zCm7V)DwibYsIQdix3$a<nK4qM$C@B4coIU(2Ov94YcyMfdQ#ZCoE6V=F_yvNhIUnT
znd5MYb4n#NFCyy&t*>lvpYqHGBAauM_y$O^yFnNpgQozjf)t<yaeOph&txY6T;>@6
z^}97k>1k<0Baar0pGRC)<T3>pT1j7D|NgEa#z3P^l2KCHxGlLaBJayh7JBI9yMl|n
zOn@A7%ok-FSO*#uhW7}mV}HG;qr}bMjZMo4l<&d#Y3H#K^VbJ38-nu3GaTGcd;C7g
zYO-ks<5>j!%w7CYD>y~xdCZRm^_t&vdCW-EZlF(*f#2B5#cR5Qj^xpsUi0?-2e{I?
z?hSd4htXlThXCb-q+y{xKhtSv)3=cq06bUI-rjy{i*7z=W0d+SWVS?HEMc!-o_S7)
z`3T$UBOF}z6@2teQ8S$JS<L{cdC4iyH}_ZtK_-~Oi2iFkFjfjqcf^Yn0!t%1pjQ&%
z-;`PKBkoj@b)wHCC8|XMA`60Buh5Tu#VO>!y?!j@*%{8Pdz&4=$Q%zRMR81imiYub
zb`F2RGJi8E2Rs>DU?345gP?gF@a~39t%Iwh3W|Jk2@2P<#?^_)885l$05xI~q_C}Q
zZ9jDSU><;jS5DCJW+IdyGk{f%44bZGFNPi~_hS9xWN^#c!K8KNWq~{R0Q%5Rrm*Z_
zzB;BH*<dCcM-XE`c_RqOVN};y_l;~UO-84koAY8AdS~zdfo~3HclBZ$?w>0f{hsY4
zl$0lDUz=m+68mN!3Bb5ngdN%cFy=Ecsc?4x{|}D)IXf9QR`Q>5uu)pUZ4+X8J#OFB
zW8Pj#7RE<#)=OdEyrX{{jHEZGMzfJ^`p;lcr11XSAIo`>ng8=+N~@qfT#^v?mED(1
RCOGhqvVw+uiQJ{&{{cj#UCRIf

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/one-param-identified.png b/docs/2_39/stan-users-guide/img/one-param-identified.png
new file mode 100644
index 0000000000000000000000000000000000000000..e90fedbe34e475d922a0532856557e6e5872541d
GIT binary patch
literal 61295
zcmeFZWmr~Q*9J-n52%22N{6I?pma!gcS$#bbO{n7(%s$NU82&R(nvSb4QD>M-}n3W
zxvul`{QG|FOL#WxS#ym!<~_!^$34~;d0DZiC<G`lFfdOg#Dx`MVBr10f20sZ@QM5P
z;_omp2$be;-^xq8eM>5DZ);+1WefxJ=9$0BqV23Q7T*_J#cUjYQgyM<K|jk>l(ZTM
z2}(K<C8BcnByn=Ke$i2`a=rdiK!`NERa+?M?(-V)23BAi9!;Ew>?xvw`}A3RwS>}V
zx<UzY_st(@WYl|)RDL%(8PF*G&_2ED<6Pw2;NH-k73S45uSvvL35+c&O&Ow$iOR!?
zPL&B7ts0ktA1~UHWsK+>z+n1@eLlGJn>Yy4w->R0CwoTG=Js`N<O{i6yA3w6C;r6~
zh5m8NS}CkHr0WawI>fnSqCaCC2yjJkah}N-E+btuys~8c$f3h@JJ=<VW<;LmUnb0T
z@ATmDx}$!p*bp|7n?zKVM82~h`pq0>SI2UX#+2lEva92(ex01tU|5g#Wr6-ox%CVG
zRLu#|AE|P^vIdEFA0J(WU5$FcxsWKF6hwMDENqDB>6KtDATB&|A9<hmSmxTok;R<|
zuVUnsYA3#%;?0z!DMnQ`uJ?iJMfESk&4~PNCG(|sOW!0t3lsDb?NrzO=KHEN6Z+vO
z!i?JT23Eb1y!oq3n)4qGjB>qZ%fva7QnR!vEfAb3P1I0$e9M}^tsB<;!n!G@BbuZS
zwb3a9zYj;Ub7YJ9V~L5I>9d(%kD5|(XTsCwM|T|%@mzoLH*+zsBNEN(j4K+{X0?lL
z3JX)n;xHOa6i9Ms%crtN|7a2A;Gtse_VfSV_l|{JZWP06-jy?KrOPk461RmhQz3`D
ztL%MSPlY<a(lqQ1i9yxwOF~o3?DE&Yl6E0d-?+Y^K0>j#=E9~M^Q1jJZkBUY#^I8c
zKe`d{qzzOUoIW798VI`!NGzA4KcSEtW_hKxhe?WrY3BHPlrPWHl7APKrx&s5VsNk3
z($e9^?l@%rt92JgxlLzn$HlL*3dS32ML2GKx1iXJ-ak$q$SxGbyxOp1R8Mzomo$cF
z6lW^3a<(shhAF7O@ptp}k2d@sF1*XUE3nDe%f}sG<hJm`=gy_{r1PY|rS+uoWWJRS
za?ofbzopcPX^mid#MyMvl1BB*y-e>N=6sNj*@R^pJB^ZTulp+QTJ%lxO(hRuIaY(-
z^QgSlmOs15X9G`eiCgd-P^v6sYTOwQ5%6B2Q6Z)vPIW5(^eb|}(U?t(%2nj(?)-H^
zk>{kHycQee!j&3CBmU*!tK5&SWj46F)aDD_3+nP*_|v;6%$HnTOV$k1RNZXLDO`4{
zA}1&gNy7s=UWkjRnG!`#lN`O+YSU$fWiOCDdh1zrCd58t54^%2TBGG~u5ao1nnl^_
zOCGu@xNzMn;xzxndS6TU^L+>Rj;fhfKBng$`e_@%Ou!#mk9D@puSlnjEz28O;;V=e
zN?hfO*Th>=%kvI(BQD<ODnm5F+$66F?n_@2aO;uZ<BS*|^J~D^xmXd>W5U1)z(@!S
zD!ah$rXV$A{+w(_8!hr~A^r5Uf>cfz0qX}19bC{;^v4`Q*v0wk=q?=cSn?4}v>ZV?
z6^54{nVEs<v>XB=BH7MIdkN<pwp9}QGW!EIl}-1TxBCOJ(+QmS30CXUoO#myC_>&a
z2&6Eu5CIrC^oJjPx}W`Gvd5r!`R|WmA$V}Key<Sz=MNvQc}C?egf|i-X7&I2oYt@U
ze|_gae=5(EX-~m{AStc??b9Px%gN&6)J2=7y=;S@LF=~Hdv$TV=W|MD)21W^+70%q
zTj-nw2xf*O`5TeqoaUpH8g({0HzQ+m{5no^c3l^)<7l11_!MzEP6^+Jo1AU&IjqzL
z$qy@^)5s>%5^`7t_Ia)+m`d)=)U0d}t;Pw2FCOUDYPDG!kLAkPcaweY4&%S0RVh-9
z;ry^JKHung%Ov`HHBVI(8Qoi89YHjNASc#U>ejTy+hD%QIjrS)kh{Wqw$@k|#vmXS
z#+so{rP*<3@>lc=CHo|^RK;2Ax_OKR-ZYORiBw5NmuV)Q##bg!>+$)nF0T*TaCOP<
zwK5|o4XLeNKNCBNn^An&7-UpO_17$yeiK){E9li0S3Msa9Wynwb?|9Ejr+GBX7gZa
z`j}~t+3(HuXi5Ww$x<D9Ols-y$r7!;*6V%xEU~aqJlG$ni7?jEMrXUz_Nx(Ml4*RN
z&7Cq=uGHylpXk*}U**(at|uAir8xC|_+p>LYt325GnRvcHQ2m3Hle>h)#R{6K9I=H
zhSq#GW0_gCqwtj0Poczhm5CN^@#B5V$>zx0tMdar`L)cKzsb;RICA^w+)uZSy>?49
zs{bfcMrKp~S&fqGJs#q>-Jd8@lPhgL>Wx19<PuT4D_8O;;ludCL4(z^^Lb}Io`zP>
z>)uM^zU2)CE`79j(H4XnycF;EYNmqrF{N>f9pN%}CSKdj(e}kL$Cb6+X|~=R^<{7R
zBB8~8N8{J#a@>w~2zxw2EkM{;diU|-H&R8ciSapl1-!)45dXcjJPZ;lk=dCgl43lT
ze=cL5_wu(K5Bv4oeL;Vd_*DflV&g0Au{^mrY<ksr-tHm3YpVdc&ol%m?_9PqB#y+F
z^zN_XqM7yhiZDEG4!i6pRMdLJNIawe%&o=gF@arYPfEtLd5>WZS6v~l_EG!daHWd|
z@5#v8x&j(k;e8j$N4zIUMDM*FfU_+{$cH`B6KTbHNqcZ!UOTI0+oEF9A5X^R!tc1P
ztg!ad0{)oRuhna<=Ly_B#c;K`lI!j1)Y`>Ln4CPy$ICU`1P<%OWcU4s`>*}a5F`b>
zIaJ}y9^C~RU!=E7ZQ35TUak|mf<>+*EZFxkij+<UiuE7Ep2lyWKm^Y&UD`g^%s+Sf
z_0GQcMIK&`d{O{ipn?))tjzB5x41{oNaZ&X3RQ}z!+4Iw6<HYj)Fh_rVSFOd(?eqX
z2!;^++lMpW4&=AS)&cjo@5DDn$k~j>gv+6Bzit_atRQQ99R569aXnt|DJp9v;1BD7
zMR|v1fsg+<Q6Nh+BpVziWaFEYF*(dZP9Fht$Z)6R+c49|G2UycLgUPd#9U6hc!e##
zwb~v+>9laoi(fONE_roIa=bG4UK(|yU%$jiW55p)W~VU9z<hJD<hwTSycmM3bpAd6
zSt1HH-H);@A1>J%51d5HtL4gbwA<VBMUz~G0pmD&HCa@iKLVVWt5Iu#6rtiSd=p;?
z8WZ?Gwq7cIZn&y5?!%dO-AFM`^E`W|L!R09qBr_Q)b?TYRO`j^<EVgzV#S=G7h%MQ
zQ+gk<Q(z}N99)Y-w=MYqBBg0$W^CjjHzEmOTM*AHeXgqg`BE!2N|sB4;I&P%3Xbmk
zp+XqWz+`hmshAjC>mMm$`#CCarM}`YMjo_Y%TAW*F~{<=kD*R5TvT*Dw;Y@w;v71~
zx9%f&MQl3EvG?5UvWx#t!K#U=`Yb#6WL_WXwsgmtA!<p4cm24D#o$Mv-);1^H?PVt
zM~3xx)NK#9E@J0uxiPO&-4;z8CT+a7vI7BYwgiNlt9(Sd-Sh(sQ`4$?x06gpON|qE
z{ma&v1UA#&<)2?=@SUfPai<N#c!M8jf4n<R%h9yo5h48W(N+a^bg@DD)7`@6K``++
zc8{+6o^e{DQ^bz5D32Wp8mnhp+}X9Jb`ztd5%Y|w`I=p<+O-x}z7ZO03&576%Z;?8
zyD&Cgv#=5pZa4(e@JBw%vc@H-nm8{g#ai}iXO{;RO5#E|Yq=;Sp3$oo<As^Nuvl+B
z{S>OfZ-2ORc6>FRE0d&N=&<WI@nR>{eLvwy!<6fEXDUpCxhI}$XMpkzF@+ZY!uqp5
zQ^kJ!l4tZ<wN^7X;|V7ruNUyDu$HAOB?7aIt8TbDi%G7lw`N)Q+V1aI`hwIujzv;d
z6WtYDw^}@ES)8YvEhrMhCOFwqS4w1^DjZET4ns8DxeR6&#vVDgTTj_hr-7g$9v5xC
z<NqsFavN`^$nE61>7819t6cH{z9s34*it@#9J_u_?;5hpqh1^xe?uaXc9ETp&f&qj
z4E7vTguzU`-O5Yb)mjI8z#yaLWIw_>8@g4}5?cX0nKwgw|KQ643AdA#kOmv?VP?n{
zcZZIsWjON~`A?g++cUon=7!ZsKU1mioMagcvPYL*%zNiG^-f8szPQ$JkLgNy_K*0^
z=RP3s<!IH}?9*oOPSu=)cue{jPPe(AaU&m@IRgM?Hbcbq+WD+)+t1>JGC$K4{?w<0
z9MTD)?1eboSDRU5d>hi=T|N3)x{R`mIk_nd(KSte8%{z@Ixh?0ECw)g7~BiL77Emw
zjgYULSSJ+Atg`H><KWu=Y^nFyy}q<Ebi!}8p-3F^$JLE@OervUW73b;6px(Nusw%u
z@7`PccHNQp>UO@I-UEaWWb2LO4}%r9RGylAD45F}X2YZ}WD6e?b2*5+$Sx9278@1j
zBTq5Z2_X+{d4J}{QVYv5?;AWXi=9q<_cg_FQUlqza?xjt4`BRFd=0@*o!#gS-C7o`
zj_B?vrmKn1w59WnkJpW7hnwwh??wc3wkL|Z+UO^x9i*={6d=1~)e8X#kkuq_rtj|&
zb?zc0<Tv^n!>D!2y~d+#TFw{TJ2mRO8sC;^)(FQvDbP#pjiGO{?JL%(TJ5F?!`gY2
z;J`fpNQXLevdJN;me^rS`TA@xN3YE*Q|e8MRKcrIraFu1${df=3AHl{bACt8@Yr)-
zG(Pd5EU63!J#QxW5}hWeQBTWP!we$TUI>x<*p3;Lb6+Kg!rZ5;S@>tVy<y>v=IZT`
zGcA+Gs9YQ88bDSs+U$CqZJgO?G()f!hpXqo@>}Ct_||7zuaIi+jQ4uXtMd{+_iR;+
zmuq^fLT@odP1lu*Y{#Kbgnjf2O=|oSj`(f+-Oce3@=%i;hwXw@tZHLf+v%j%zLdnq
zey!_vK_2zbp4CtDvQjg#dpwQK2TXGOx7D5&e5X1eK~^LV@&U}!SsF|lSuCAME$wbG
zLi$qa>4hQe4Xfh;;~T4mgF%m$o7VL#>uA;IQs1yFNohmH<J8*@I|&bu<w=AEcSJ-X
ztC0ewrnb6nTR9hI(WjLxT~j*FI=$rFV&98NPW(1Px%cbi`SPR;95@EABEB+$T#1a>
za#6ScHhH}ut83Bg+FD4v=q*;&nd47~fFY*1Tl33H$-o^xyD<6LAr0cBU+IFrk+G@U
z_Cq^mZG11%sAUqdxFz?V<4-g*P`#9R&Y0C3Z!x}TZ?uO)_Pe=y4ilqqi89T)KaAsO
z+BA)7yr%ZC&P<KR<q;*<q_@y+#^>jfcr_DTlhX&8u`jjY(UD!%ec>g{^BUzh_g;)W
zbK~DuxV*fbALgl9D1l9Nxz%+4ILd@OaY1mBqTlxU<ak7k<aQ|J)qTmgGQW$nszh6b
zJ=rKrkW_Y5okhx+swPZk@kLo{$)4X$tX;LZpUY~VS?o~-r#LUTA{(o^7O9pHfFqq!
z^$WSntD2?E6p{jG+&6t`#nE-?0ma`$^QO}*Yue+{-l)QbIeHWaDk&^DDbXzXzV3J>
zxCW1e9Yut_{F#)ex8s9{pGmypEBgoI(r$zAy4yfzCvoUl5Wv0J`u3Xhc06>vJAz_<
z1Y?kK1(BtX=tIYct73MGSo9&+$ilaZ1&TTG?2>!Iu?{!C48!n6CN(-U&#JPwtG8Oy
zGCwi-Hm-znV;ofk+yAImie`$Ab#HVS(-|@)JkAQ4b3bT7-s?Fcjzbl>X+%xlB>h}v
zeY1Fft9R3Solcq3L|UCE&28j0p2d28-IU?lh=f&~z|^8`gvaOMA~ahur-X&y_VNA>
z`7mWq_E-*=0bY+q$Hb&-YVDvvDEBmf$YeHFspVryLVLWiIJJtx@>H?LwM)a^krrFU
z;q^W|{|p{VMv~id!=h)P^lS-<^;-#-!#g|X%LTA7%#6<`PyAnIV}Dg&c-AdqdJA&Q
zAdOHt!z&8w`#&>jg&sMN3v|6K{HUV-++82gtmSlCu|AEB>GIn^d$Z&nKe&*MQouzy
z^R1V%{;5EIgs(Eik=sBePNICS6^Hlv_5}TJje*xXv`^0+%yArf^{o5LI1CtZW%X?t
zpE99{-J}kFnq9=a9iLuzAe)tSke^YaCT4BhR+zqqdeuBU%Fhz@VN-^$-;u4w)FksO
zBbd~sR#pz3=N+dmDr&j=h&&s+4ECq}LEgn>JSV+9;`MU2HhKHDNc3CKV=lr?4%S$k
znjk%E@QjWnYguR+R6W`AR}&o7Ji5&vs|U7VYLDwWXF4;d{ke=x>ZUTGRo^IPxyhR!
zi3L)Gg6F(jx@}ZavGHvrUKeEFo;2R3C;k9cM$_Jy0sn+YW6gN!FBgJun&dI)L))(c
zCp0ALRmXdDoac?t&}QtQDJed`SM$m4^jK^3nUao#QDR)qo1lq;pr)x|ULBH;Hx<+8
zoBi#HGOujj>{@QCb|gM3m-~8K9sO?PRk3XNm+K#xa)C8gD&yI%c4UP6KXYQXJo#th
z^{j=@1G#)_vSo{@(g&qpgw=RCgcUWusL+YY4p#s4^__*B*MuQmzn8~r2j(g^qsk4<
zzMMsC$1hoeIzstL&mGN|QQZ!r<oLemwQd&{3AcSz6&ukW^Yok2{vdPHdQ7Tzjg(>@
zd!~MlqABVeBqMtMXZx-IoS!$()#f>}Q9!gS=*2v-^lDyIVmVsAP8_O@c<fruoVV{$
ztfFdXw2f1&+v0Y3{PKA^dO(G0M2|DvG_BnhifM@W7h_(QoWm%GAtJAf-x94TV?R~W
zxCYY^bt(@pr$ExbLi7IWWb4o<nYsVsCl-?D)yXq<gK&vq+zNsWyY;LT2IsDUtvZD<
zP<&W-x|(@dj_WZ<6&WzrO(<bjW5$0Giht_Ny*KK0HPU%wL2;F3eSN3MG=s0mAKhD%
zpYB-vQP=Nf4NcoNiwf_>;k^94>tJ|~{-@5V^Pm#+?FO&&`2*~hQj|1y!{X1>`L>k>
zN_m|z!Ybk_oDSC=V?O+O6@<h~O5epVQ7x2d#`GD(qyo23ayFYIhB($4#MDD?UE1;t
z)_$I|r&&{|9T7PU3nHUrT&F|@84s7eGk&$nE`Iq|u>I&OjYWhM_w#9lSr$*bA@xq>
zAFMB<&}B?DE1&M0j%(^slO>^k$=y(WqNs7El#ndQ7-B*=C!0HOo-4(yw7-=))OET%
z6Q5=+%)LIC9=cg`$5g)rm%w5ekpK35n{U&;?@a1VahI~=YM1kO`^S;tj?}X2r?BVO
zDk)WYD1$qb7@TbGd&;|$#Ttetxbs7rqV+#mgR$C*tE@#zh`xxOTr7cdq5BOIaj{*1
zGRnKw+4w3CqWjMRb_eGU$@aIN1J1MEF+5!HD^K>FLyvXULlelJNXd|kEkVr0-g2kl
z>NrH$wdmtV^=KldsPs^(Uz<i|Hi&(SBG0a0(q+x)6J<l5UN`29lM`2sn`!iIf-l=h
zh?j8AU6Gwi$z>}&extNMXnJIqEJJu)6#^~#GNr;(TniIE_C0I9q}OY0PS8+@qr7t7
z%=qjUI;3+hXe}q3%TaCB_VG$vqHPEG{agxL-7JBY&3PUw%pNE)4!>1=yuV&F!1s}`
zZ~1ezQ`SXdJ(dgd@0;Wx<8nGL9aF(MUqt_UZRN+wy=$`C=u<v5H1xlNTBIjkSr)DP
zHDYN=Opm?fx?W5zPI=UYX!oq>@~(MpWUt__?$G)Pmh$%MzC?KLvgAW9_DN+LM6Sfc
zx;$orKGLu1or;n=chX|tld~oh3!KtNFVpA><@aSXuP6YWV;o5$Xi6~iZFntV8lg^T
z^Q<P{wXqDOrVL38ik*)1k6ut{;Igfy?q5b~pErhFPKz=Wm3D?Lv2+np#M=2$5)~3{
zW60i)<qXVxQCswjJwzWd9k*3ydK*6zm3A6mh}wK-W2R$0Q~iR>(s_gNckYVt;&(17
zvxA5+<BY;zDPC7whw5BfCGO38`ei}o!4~+7v9T)1AEeY6H@!?v-_!ZU_90d+PPp>x
zbtN!O4OM$w+88FpvWXuyd$r{mM(i9|w$6%{uG*ooHe3^KxgcJ0p}Y&X46DR_0^3)o
z^l8Pg4MjCiHYK>yzAamd)i52uAr5nkeUFD2RO4>(tY-LCDYwdx+VRI5Zb1pY?EbuF
zY+XdX+Dt*T$E24x3pnPgBjIsV$ma8>^J4keQfaR134Xj<Hk6HX8c(q|(<;?IpE-%m
z6W390Kyhy>E2(a3F$5&WV=2ZF@QVB0vi|Z1TSxMFLYv9N<JYmqTDRyrLb=yp!JmBi
zVAH@80gN5Ee><5Tx}P%mjj`%FQ&ba$Y%|!M>22$*I+O8>=xXSrMzQ^O=T#v}EzYAX
zPq!VhI?^YJ)&jgmNXHT|&LNHNe!NQYjn19K4(nC<@{zo+xE?P(DU{zXjwG(8SVsk9
zkh31$NW&luR$jYFZ;SAw?y<L*W)!h#vI(rlL37TFo~*7%Min>8y?JR~h5O;7io*l9
zjWnl&+y_{FlwLC~({m6d_g;|YYPf@9!XuBXimhw9dB`DGZc*c`slP!_9`i%$&Jmj_
zk`fY`mB0LK@EWt~E>+Ipw4lt!xRTsNn0$aoU4^{RK$LvLUF&+2^{V9ab8e~$cN#gB
zb&XnDztjw>cMvR8*z?9K<((V07keNNj>Z~jxtt-+pqy-&8_FMn<caXl3miNtY~S({
zI1h_pL+uj5>^3DFX?a;Vx0M$9!y-?cQD8fdNy-UH2mC`D%cw;iEBEGnuXyPDc<PZx
zn1(O&H+ZhV^hM{5be}=V{MiKQPfYKoVEs+gJST;xoPT%tKNgwwt}Uut^x{qpcE`~i
zO4*Z%3&~Z4oa*FAC4VP4n(s_@)FWD{v7Exe^}pt%3_qxipjVh<OSvM%J`0}>!F!kI
z?|}}Xg*}&YyS+M<aR0;lYyg0KeuBHXKR5B|@rGAJ<4XOI<7Et6{W)nN#`YY10WpI6
z>cS;aR|52L>|zyT?yH9Qw5S1dy6t-ee#Gp9^^A<b^b`$&<Qzw>OIOveysXI+D(G5{
zk2`5?{V*r0SgA9(R?A`0l&Vn2HaLNNX0Omxyq(+0sidDYHRYVF6fy2uy3aae?)RGw
z;0mw15|$D>f$*O&0q8NJbZk?ShkOu;bs&ZDX~vMJIkEW;N0mV3z`vq8K8n1h`e*cK
zRaw)@YqR0@R_uF+q3!%-0a{KUkGcqDs;PwZ&Je=Bf!B)+jY9`BmSr-5x$aU*4U`gr
z(Z40sXOze$4Fgod?Or)=^v}D}Gi5mK%~0XNR$&X}DdtHHIk*<EZB{zwT_dZII@%r+
z+BUyC!<@1E-mI-`d{Py%6$^ikac70qL5l}pF{?x7aW&1+96fa_YuEOnxu09lob7Eq
z^U>;K*fivNvHeuH?Sf*d^y`-7Z!Ouqu0yW7b8#!psoUxU$#O{?jK5V2QDd9n?}1-a
z@<jD~NsKrcFNk8X%2buCF~nV45+hXjY11FiBHCcN*xF(znTGsfFMSc3NxL_maGsNj
zP@CrjVLsG*xdxk4?0=uAlxW7>db-k>C$W`Vx#xZBTL)6|35K-el)M8vN5z8${>`&0
z{2^LlB(GYQI{IMg9WAxcO+L9#Z3s73gt`*zHijUasa>wK>0otujMCUbF2vh;XKshz
zq9u4bC*TOiykfuj{lqlcF5vjp-XY)wsm5A}qNrsgUtX8u94YQ~jTmbT{+atQM8z;`
zzaZsUOmST|@04WcS4!OmN@UBklAIvI7)!6}2vE`XQpIVL&rpM6kjjwdRT8h@^=_4%
zkmH+>r<}HL;?A}l=SftR!C^njdj8d5>y+1dL%L+xgfO{J^mIPEcL}iPByycC+T*Ri
zl_95zsO09{wZ2ze4ow3U2QepXv$fiz9+j?xATQCKk1TLc2dut&<spe$U~*MKl+c$Q
z)7d|gCg<EgZwC+Q7MpV|Z_zP<%P~NL3D_lNNU3i$vidZ#<&W|w#m$9U0qgDA`9;Is
z&>&VY0Pr!5o0-9>bvGS98D7I84{7p`Ryf#eQT#!+i<tt9j*gXXV(P?iIhV73wjt@u
zWOgo+Dxz4LA2qSDoyEp}FIJkEbw3QSgA1ji3}PQYb$<0hmtg+|Cd0Zx>r6a}^^X}_
z?5biO>hd%$M++9=yAQLKyLeT_J=8qcM}5qB9+RxB(da|=Xy_mZ)atEwW1-#a*^Z~|
zHfGe4j_e6@t7ba|h14Y~2_KJ-Ao!cTWuqJO7ud+=>hN&L8<VV~(dOpd6=yPCD_mi~
zu7Ens_89bJiE^t>>#1dIT5Ovn4Ng(6NCV7xSYPhZE9;q8Yiz(}EDubC5*TuLj1!V^
zOJGTroX%})*zwkaW8Wt;Zuot!debv?t@BlV=D?DAZ*r(sLDN-HB`J1MGXlf*!(`-I
zd=6=D5J91(MZs6lS27YL?Pfkk#-#qCXMcv4b0?rFYBGv3OwxnHs7cBD+U;bsdZ?b8
zGBJ+*iohN8Fh=<dHG+d1K3*_x46^iv%gmS-uW3rGjShZ_ZOGRjfKc<AG7`&N2jcQ`
znX^^dtX73wj)*ooY(=&&R-zJCd0^o9|Cno9`t%qhwd$-q*SZOij^cqu*270EYDUim
z&Eygsx|PL2L4SM*twJ^X{4lpwXT#`VrUL?=5>jUywqdGJTTZ^Twx7f}iyqoKPN35R
z<vR}fZ0)Ysy@@9E0KJmG{L0ltr(z2lqpzZ-U$h|i36#;&5EFUfH*GJQ;EIT=)Dva%
zp51q~GDs1aDR*zO?7>Xm#=enW>!hd2B)*csW^|a6hrCtGfvmFokhb+Z24rhgnIJQB
zd;Zo4T{E1F9FAYx|F$&d@)(C9uwc|wLti>5&1Q1Y<X}Jijb16PuFJcA`)#7M9;LN`
znIycb8&9e=Lq_`QDCz6z60N%I4TrN)O$^KEC&VLD=~c;zWi_^pFYcbZ@4n+o*kV0g
z7NgP8b>n>?MGQ3jLJBGmtZ=z}^QvkAc`dm)x1;Z0!@xWVQWd|8tKzk`a@t@PYotLQ
z2dBg0s=*+Qea45rn0eO-F-i-s@eJ8<*7KN@#l@#ht#d{2>-`Cai@g_BfGk7a`))%I
z7R1E*R7G^ResJjObO-k``Pd}))Qmq0u%vAtTNEL;7KX7y-F_Mx#-&vpfHni%EJ8>}
zv7NGPO2)kB`TWYMQkibcMFpQ)v`~I=)o-gQncB4fAxDY;8<nc;=-kIxM_HW!3y+u`
zgwtqA>z7fah;WoB5U#i0D~jEs+#5GvF=Cv<wQQ5t#1D|-BHp;SNxXj_^$XLQLtYlM
z{RUdy^__~EPE0ypcAvZ{4kJ9e#MbPVhI_xk%!_gl-<;qq<J;|`vQaAAQuo6wCFKB6
zBWg17&jg|hnB@(Y(ro0_*uKh5vFxkU?!6{Z<My4|qNU4Lyi2*<gz@3Va&zykew%B5
zsSxJ%@%8RiCBqPP^-~?uL~D8Dj_K#SD?ac@I0@sR50P@z0BZfl?v3QvRRv<6Oi57h
zbIU=uXt*&mOdNXpdM-S&LC3U!@3c-DDJ}UtNC{H)yLgRGqI=`IrFwD7Y6dVMS=C3`
zh5VLiMnXw1di0eF9Uk9;o>#^c`WpQkI6<$3*_Vuq))IYeDg22Ap^h>wq0z>bulnZ6
zDzt@ul-4f?`Ndfk@|+~@uT>R)oDmt=R1*8l?Yy6?a5d<G#<?2)Yj?Wp>%jM~Bq^D!
z1Ez#keQS<{BOcY52McnW#c9f2GpNi>yOq_~YI{dJdHdJDKv)~A{aRtz#g)|+_PBx9
zc#KTRr@2(Dt&Px$%JXDI4EeNh@!ra}wQugaux6kjQ(gPlhaq0fqp1yt9=lw!S-1XN
zC~WD{`O%e4SxX-e$2{Lib+^GdRz85V<obr|d0iTh+RrXyby&TzY>>qGOQEyRI_)^l
z(gOPYYAmOTE6}Qro|q?0Qh+R2Qij5G5Q1=NydUt=urv4<XcWej`^<Pfn@O8#C}((Y
zQr*=j%dhiUn0u*!OFvO<tvAN7#qBhRQM-OvmU>4n<*<SwlA6)5%D95xv0)mTLhU|@
z((%^5-~6?ed);qJ_#v*Rh{IsRFyTCO3nL|w-Lmte-=3SjSK+eD)mB~?gH+vF7QUyZ
z4U^)&&qTH3mYG5Z`m9ndF{>6|+4ji6K>%aeAnML@MNk$L24tSyP7gKkPn5@|1bA6+
zH8wZbr%Q#V3~i0Nxx2dA4Q<~>jvMA$bIxDCj<Pqd`aGPHs%G7v5N13wvAbno>HOT1
zchf3!9%t?v?yC}ny-b>uyDw(Ij&JcG=e9LAd!CN4?4iRmm1JDP^V_||H%tN9SXq9j
zPZ+R8<kI?=M{7s>O$R~jTy)#rYBHBoC0c65)7YFhf4={8+cFU`1rY{IW(;)_QlnmF
zw?4l83g2{jvXb5qV{WF6X1k6yWs#XX#EP%P)AR_Ca`@-i7EW{9OJ8cNl#7i6*X52K
zxVZ<ZZ)rOR90~2%su=<a(msLWrAj@(FIWnT9p2XQN=B>t399itYKkl94>MDH7WvXn
zru*4F)n}A&6qRr%0-2H{Z?6C5w-7;JM5>DlB0x*22PGnLVyi}i)`FM#vKZl*v)`S`
zi^q<HnFkJz5{7&CSdJ>uU*$KGKagcDnqQ1?NK!;KX1V_M$Cbv|zIIw@cC}yfL5ijB
zY3KPWgco0!nT`%Y^~uEQSP~bMAcOG{qSnWM#OnL#sz5MhU`GHQ76>T>Y7vUpN(>mE
zg<Kcc1aI^J)k2Cjt4~On;Z?gT@uSawH5bBh-fQsk#VtK{Fx%>PNw%$D(mp_?7^F&Y
zu|>i4tocLg0Y$8yudoh5pil^b!2Xp~1U~oS^SlaN>!4(+uaV>Q>W2kV4IgR9=6-|S
zO6SikO)rKVg)Gs-V_KqgbkU61h-ZJL9q2-Ev9Q=%K+d4eO1xHA{G)S-6Ds+TCxY$6
zZ&B~`TOI-JeiHuBUeMQ7wjkgn`>P8fqtw@k!Rl+wj*t!!e~Qma>8D7Jfk3+S2)56l
z;$&xvIrT)tG=)kgi6bOUYz@Z$B^`T6@;_n{bPRb4RcO0eyE5bL=f4U7BB<6xTx{vE
zizJS@^`c`=wQr0<EL0koeSAs*@>Zt>g7o)6A^bjLvfMI_PJ0#z6I5C7h)*DG!{QKF
zNOVv;r~rO8@SObqOoBtKA7lVdvq1?3+h5HJsS$!`Fy6Zx)k1Y3lZf?yj2j>=^uq@R
z!CpfD;?M8QR$KN>??_Yz5%#6eFALej=FtCyw{ssfQzUU%2W<>DyM8E8Ez!J5hKGt;
zbkSW7f8ws?CfTq`6h)ak>dHZFwnUv?<x7+N{?pCrlwPM|MJxBXxQa47u$$%chJWn#
zA3Y7GevjE|5)910Wj6U3V9gc-6tDlKvq5(SDw_yfI2h<n{+b{9HJOsYtPd;w-=7Pt
zB7+TXMTYs0cIO}epo7Kg$e-k~fBUb`BWQqB=aL!*=3i1B(n17zNS5Yt_tU=@|MUIA
zsW2$-{%_I$*@6GF^#8fX1Bm|5L4@G@s54MsiFpf1K!&{}g+fe|5nS%yn1dn-c1EuD
zM_`&1+?ss?p1B~YB!bt?q1fUNMy1@~aS$}`u6CF;YAj-GFA-m4NygCitPdm^?@X2m
z($(4;gO-m@NJq2PH0$Eg6Gc(uzStQ1wcZ|pcfQJ2i@lAvXR}hEWBp_3Jk_>ijW!Ep
z=)E&FmLH-u>mv)~Gd}HtzD?#q8jw(YmZ~R`P|A_ShAQVm6(8%dI&P0guv<=c2Vv8D
zV9M?oZKOD#F}8V_{CM(?Z!7BGNCe}<0}!g8!lN?aUr<CEE(aj$GFxupFTVX3Hjcyr
zMuq+9A^1l~_|N|k-~%AF9>wSSZ!|*hrT{3#(?MhTFF1mMyQK6MikG`u{kOvjkcI)A
zI)<bb{|j7UA*)dUsQ&+4d~~1bI#>)p#A>!y2D)>XD`EWER@0S=z^Vo66`3R&QAqzA
zt5FiciUifr{OK8eKzI|<b@}~Bqt;4;rvS)dWY2yXqL~9m{uNL(dZj!>nE%*OI0Rfc
zZowmTG?>geM(bPKT?nxl#{d5FD{kM9FPO$aT{c>%LeGto_Ma=Dfiw^c*il@?*}<>I
z6j5c_Vxmx@&Su^S2uj#sT5r#^poD9&W{qU3=b6R}&{BA1yTHTsJ{;?R@6tV$w*r2m
z&?e-ELB&PCevV|!DkySdftcafF5!5KuiIRmt=_cP^<Ky>S5$r#5LH0${GU+q&$8i;
z29sVhx0?QmHw=wPahiD{u}$nervQA@w3Jz|RU3yvJsj)v<NsU+{RRj(SbO=@dxW(Q
zbk0B+sjdm<RwkXmx(<4<ZD}0OU;nQVNQ$ls{O(=JG^Fno>;S=*qw6MAFBB^rL#JW{
z1P1KqxQv?7{~s3x_U5?5#xdzwWKS7qoITO!BI4dH_id>0x^>e4&51@c?SFEzhX)g&
z1wI>4shLGT4oHDzL<0-ZLSlfJCIf9z)2|C?KK$R63B*H|oXDkXzA$8>WDq;_nAv3r
z`J)`tLURd{StN@J<&eSpw}7q7(dN8w<J*>EA-QZ^WB&1>1T%$<$wsvvy>DME!HTIm
zZM^!lDW>eeaT^1BQe`%bevihp<&T%_kZ7A0fqr!AGpo_6ii3-R?z{QwW3=PEd)}Fs
zW5Tt~n4ovDtyf;Z6dWZ<HhzNm1TQ=~Sd*y*h(4l9-8}_OU!_zf>TZZ#{icu~Z=Lxg
z@m08iTHv~eeZ(v$u6NtH8bWGq4pA=C)j1g|D>oS+XshgX=cDy=ErUY|mx7c1_^APJ
zdIbN+{(KV+w!$sp9mv}xrdpcQ`LnO=M{h&@jNl6AZ08%<TaDguidL8X@_Ic@pVUaa
z*<E!9SV}5Kzd*eb?S@L_BGrNObKUUrox<o4!q<oe-K{Yyrm>b2g;b5~hR5pzt!sqY
z!PRXM7W!P`tS%a=HJ9nWCzeyCr>seS)u*oo7u^*gK%knx2I<+<rcUe$3uOy+a+wp1
zYk)Rc_xsl8YE2T4OhTDwdQ0xqFD_a8EAUdO^CTN5^XUA+D!B>^rP9s$d#p}^TC}54
zf#&5*QDH_7{G)R5ixPuhWm>^szD@6m1$AY`(8+DH%{<q2v>dJv@Zh9H7ydRwix56K
z*eq6sf~~;rEoBHBzWE0_ze0KR2VWpf@DA-SQ?KZ7P)`t%S^tg=JtKnD?N1)ML^RKa
z3p|N?CQ+rRRo~Y58h!?_J@eT^2zm&3VRa1J0N!5$F`e<@^3QdpunET@pk*4saMZ~y
z{kWgttW%TTSM7){Ynlet@vTg$cs9*+0f>(S8py>T$h^`Be!wNs^}z{&E#(PHl@^$C
zw=@;zepOKH+n83qV)nUn!Ei0~J16j4pO@(Bl}2BHmJ|1|jk)cH_1$GE(~eRW%T-=b
zN-M5oxC~O?cv;c@?@N%{4)D?F=wuYVJh7c9R9SCi%DUgo3hS579FU4zE;ImbKqSfU
z0Z#`O`R@+VqH>8p%RL}NH#0cbH%BziXpr#&r7l{(MjSn{HWfi>YzLC>wr#g2HWMXU
zuT`*01)<N+0L$)idNymD)HYpZ+J)5Ppx00KnTQ+&r_|l1gVxF^UH31ouT_w%&*_`|
zCt=y<;bVIi%;#xx4;H|8sRCRfJ>D3KgL2JKDRSh53Q-6Fdn9PaNCGZvwfs9OTUE$D
z59i+#{t=Rne)T;;+k-Nl_$hJD<V({*BF@7fm`dw+!2X03gl<Y=Wk3WDgYZdOAVhJ~
zT_&czZoy5e^=g}*B9zVpn)EAloZQv}nG*ijLfxLiK5=%rTOes#S)R^kfvBMJ-~4%d
zx#WvxwAvkU)DDjpNS6;(RZ*Y|ApuxgPj-3|W+Ei$4alWI#O#P6Q5_+CWk$+wG5)eI
zzFNrvbbEV2ot;X$ulMnm3hYp2gNd^!2Kjq95K>7=5nAu>+@UIA=`=naFQ6D^n|1o}
zOicX#?sg4`cG#sj9t&dWxD~FAwiT5&_TuGVp10jwN554uC~G;>95F5eWLzvNF&CaO
zYs~3}P`E$q0~BL|EqggZU5N+RNCe~nvaEK?!d3jDV(CqeJJf}=+t0s?e8uTm?TK0u
zHFZE#;3j(r0)k}e*JscOm%!y%sDYeuz^ihnjLfqhZwOZ*CWo|A-S5=kcqVT=+6I)U
zL3F-XUQh{R4yaQF3kZ6Vp%3&P1a^2eW2`rraDW$mmhuSdnFF*yKA07f#-ajADWy()
zZEM3tk&rDx1a+@V>h$X)sP>=Dr2i$Njsx@3Ufm)SF}Kqz_nRY~*3ng4!juLOsY4=*
zf--sw*wW}Wz|J!7&DM2Un46OxGe8dk7DBJOJF3Bp(htaR8KmimPgA|F%rAleV6Ki(
z<<O*gR8H4rN*YKAQFmfXM%ombjt~11v?@@knIb}Z5Z+6t_5W}cc^a^}qZ{Ih9*K<A
zXy`WoArD>)$bL%`Zg=JysvNdPf9@hl$5NlGRkF}U_03J(XAoqKV3Tn9<_{LA*supi
z4eF~i9S>16>R{~TjqO_X#var+qKMVmY+fUEofKHUHSM>PPD>2YRr#*A4cE2B#NfF*
zH#fos6*GT`QuoRGB&5ECEqrMIR^Vgy#89ke+Zc1*ojFqZD_H;XS10!O3@GnLq!l1v
z*OdZMGOm!7jW>Z(jeU~KdDi;%)<R2@LZE|u-%vQS|L}_y@W=*O-_`UyCm8MzW@?H~
z0N(3ky~PqTwWfNWg-(4W?ie5xAJltant^^7=8r4JfPeal&K!pb_BLaZ*YB@4kfytD
z?zf!lUIPCO2BtV@TV#kEZ~3I~v=Lk2eq3jmB8-PzDOZ~1BWQruDm=+zjU<`460?B_
zv~$BzY0+=!xQgTd%`DJ0;O@^q-kSy+obFC5DPIzDI~6ydj7r%931i}M_a{3@Y1MqA
ztzz8J_UJTCs*6x4-<4Sd(5Ef8;t-E;Lh%sB#{e959QeT5f0B+e<o=}BL%UzcrkUpH
zd8~fDTp`E<w1uuI@vjpV7G<hGK#>>0HpDZg9;W&=XQeYFiczZ;L6r#w6*y|UEkIW<
zOd9^b$xr)7<ot+yEt`g)>c}b->hMUY(XSU=-pX>WVJR3Y`Yj`GF7yHt3g2dXNf*Dc
zDF~4J*!M449%6-nJj5K~@0xsg(B(~!xZp3qBt;O@vk9tjAejxdV@UhFjKJ5|ayHY4
zx0LY8JW?TdLPjL)IqxcHn4>gXL7<gsyA)WF?kJV@-xYa(Z3pe-frUED(gx8pFm>U&
z7NfQrKxl(r!zxb8>jMChAeZa-3itpNh}MAYF%pUee0GY^ga*(<w0>m8%EF%>OxHnC
z?s{y3&uSES=yL_aOdnw3O-+zgUQn*adK0WR{ONc+a3^%iS%jm-?YK?3=y5D#!$!^y
zT?!c~U>ZJ23ja+LX|3Q212<dSBPhgrp*)do>t(KjJWA8}H|7+sO0}O_B{hXgCjgMT
z|NQ>W!i9NLC;|sy;WL1R50I$re-bTOxNU7+;y&lyUnA;hp1T!YMt_#RE!P>knkHnr
z6uYwnsC3MVF8Y+;dwFwtoD~R{04)?Kp9zpj3w1pDYaoFwWCq=4b<iy$2g=L53b%s6
z^$ES_Wc&!q$Wu94(D52DA%eFUy3!RkB=@cx8+f7sEcgnB^xkyZhy6lFg#)M)4xIpi
zCfE2Z27Ww$+Tr_AaKtBSEhZ#&Jx^5(twe9}3aLh8=+$t<w`dW8Bf{Q;gz{=AndqT<
zA<ahMu%4m3y*`fvV=<Dz&(><DblF6Wgy?Mae~wNIsf<s2%xJten<`E506eiz%;!(g
z>WTI#!06H86^;LTAcIDw@Zg)3C$7t1XjXga%c8$v()5C6^OH0<>R}LKUH2O{vK4bA
zJF&aL91TN-da<{Z^a#|Xt5AYxKOvGFx;lt(a63?pGXjXjoqwHwJ?aqfzG?5`%TpmX
zv*C1~X0X>La=*cJmf@TVs3qtiB6-13bJz=c>lG#IQ$BlGP#`&M4S8Yfqlpb2z>=@#
z#OY$pXNf{^eCSjP<K2|A*nqe&!w&fftOP*{BKa72aJnDZWhqE#*TXK8&LDvCCK_-t
zYwM#S?gJjET-gZB4~20=drOr`c|pgkG+Nwh>||KL^q9Al@FCQmSaLA_<u0N>piT_k
zbMy-k4^}}>7E6cD4HO1FKxdr|a&lz5H^|hm?1(2o9^D5BXhvM^_ddadubJXM$y-35
z03TZLqGLVS2mC6j54c~GGx3n_{H)8t6mEEGBLt^;M>QbKG3{?12ECeIP^;5e2*!fD
z3XleaNBOQ^)Qtdsfq_d!mw%1H#ikC5iv&K;Mk2>?h1AuznU^&36$T<J{6|2V6$ip>
z5Yyw%U($Cw{utnz2Efq8yXR&Dh7<foM)}fWoVLoe9B{drp&L6>Wh6o`p7qG2J0C1$
z?$bhBY$m=R?{A%*u?C@LK!Dbr=|y234~4e(6v8<3{L(ky0o`waeV%8)EYYaSotlpH
zf_?S|jF%aWW=m8$Y$<Z?RgW2sWQsge<_d!57wHbr9q9_o`|GJc5Nd7a=^BmR1ChdL
zm8slHBFqX4S?M2gVau87f@!sSFk7QwnayP={s19}07Bm7ix@p@g-<f(uX?*IZ6E_b
z+N~Pu1x+hbpit>^eNKtT00$kIl1}1~0rDs$PM`@0iJ8`y5HN?{mJ$=3T9b~43>rP~
zjStYi5X8X)$!_Buh<qjo`adE_?*ROkNf1+-{VLDUw(p_p6nXR>y88$TzzR@Z)`JxU
z29b+FO5T6xanwV@YVZTHHJT$ecue|DA9I*D8*~8ctIbA)fD+TNI2=)nToMf&1?U$<
zTarxwHGoh1(OO>!9Mq+He^nUffZinXIj4`eOgcDLM<5xN1jJ29S_g<g9j!lUfCu^r
z0K|Br?5Lpra4P-;dDHm<DL7j!<wWKxz}`I-)oelxxb31p5n|6Hpu;#Ox$>q@|E8Rx
zAR{zm7j^x6^9=+j;lR05TxA660@imO2+(HBhg2lwfGBUx6k<;MY=_NZLBQHG1Ht~2
zymawLun@mkfreqp0R4aB%M-|Es$dHK4VX*`X8n~d5%naM90lP{EFH+-S%HHoq{RU7
zQB!_X<YVUMKOdjHfx^FhJMe<X_4c#`4@b&HYU!Zuo}+gOSj;Tg@0HEmuh}OOywwO`
zVuA+b|CqN#Z{DYG_W$PGNXFE3ef9)o^=JUA{)|&a0#61gii4n9s*SIP=$N2aWilY_
ziH1QcuLwl$u`$rL0N}sYVzq1{Kv5(KG<1@QL_^TZ;Ez8HN(OU$55hx6fbyli(Qzl+
z{e1rj8jipaUI?qY_*xdjH>&b^r`aCxA@g?MQR7Z2o8&tX-+bxOsX)pvBnOklno(x(
zP$9v<()fUYk1uu%>P<Eraz~SJ7$0qnVP1bU{+semuW|a)dLgJ_T$@@>^8w+Y#CzTw
z3GhU~r%@8xYprKj6D><4&KE!Gfh32|9MnZ*xN((_;=Qk45V4sAzcL#Z0ICKuC{}sr
zWNd<mkB71e4NehHAHD<wGUv8aa<u%FxfcLaaFVxQQnCPP7j!VsWKMn;M1s!kO^}Dh
ztPqK#LT$GNE{{zmd-2zjFp$mQbS~6`vAi&-VicB^MdS&*4`?m+g4{d@6dCrQj_Csv
zGen3L;h+}<c?LEDU4=UH@8uS77H>>I6D0So7Z`KQ3H_l0aS_%BBchQKQPdKEEDC0!
zIQ5i(9$<M|s1xK+)BdHmT!aMT0vGJMh({|7#To#Y#na;}OeRm{1BQ;`dHn(5lmtD9
z^>1FF;++HkU4#}|0g1s6&<>I(FnyQ6LOzx$@)(*_BxD>mJy}J9O@fXN_Di28bAHIS
zU2M%14TV50eIx_yzQ`_c_u$1#$guN8IczKz!z&<Z!Ek&e;4RRKOyoRgXVQ5Uqoy12
z%@=8->CGxQz5)X8rR6C8;?M9ib-=P}zyJ2gL{$D9Z8Beh!hw}s{ynn8&ZJB%gNDJ4
z{yxymcEXvz&&gN?mL3L{Mr=WR!T(Pn`PJ-N1yt=yu}nI2AZv{P5s>XS9}PM!0vzbk
zw?KQ4tYD@|A?J@DrYx<WfdV3k1D_DoSqZ*LiWxzV|2yjUD<Q9Y!89o)@R&Ama^Cc3
zl52S*=l>i5Zl;#VVQq1SIhe-J-##ku0Iu!^Huu$bk0$=dSAy|agmPt5=w8pg_f2*f
zl|Vj!<0FvqEFUnv(~FJYQMtuIQfLHPv;L~Wz3>3Oya9Y6(BVIWB8ae0dw&9(@#b&_
zfIZwTAj`xO7VShwm;ZbU`rXe?WqyId=@F`-heXW_M8A_L4p->O^Y*4e06a2x!_{^X
zbfoen4iRsUpD@|dGaq==H|-2|sl4vBP&a<QCR7RCB1I@<5BL{7Y!^B@oP~n&ycFP+
zh|e)#3)q4Zq|<?t-9&z+hb9f0qfm=2p+IdM2GKVjrzod{<^nKq3z+g<3u-!!F%^>+
zAbfX&IiXF@+*RnGY1{q8J_3rr?sgw|<3-Q&7$6^ViN@@K5-1lWuqKfEu1^#XcniUU
zooBlMIS?Mw;^*gXX6=%4c4Xnw0*ql`{7itUY4?F=iK2M_2zs_c!H%!v(f0g(Xnq7;
z*EQ@a;c5Uf(D~Pks22?a?S^n5lZypgs0`rFG?>DAu6|^NNa8u*Ub(>8%CCc2g;ERX
z0&tDR`%r8EMUJ8kn2Lar6o?=TPi+S{!Gy+lk*>k#1_UbEVco>gm2!eCIUYWc@^4r7
z?wLABY@p5t7+fmQ4t>S;;_ie0sP=!}1+XS7i~i@c0Klz*^w0Yto**^4>50|`@kSil
zNi?dVcm)kUOBkC|Woazr$A7+~edRi#79cGKiNHI+Mh4RdT^?_IR*_x>2!8~n9)_?5
z&K@=_oD-J4>mGP>l5r+pL{Bmyi?2jF;9>27rF(!Wt<EqDzhZ;&P5_o*qx;zvMtvWW
zEjdVk3ip&cCht~u)GA1;&MDC_{DFJ^0$j0(>GwEBEtA;rPJ;>|dh#{Ue}$6;cU{63
ze@NSeeY^*Q^YOjdYwbXaCnjiQ<R_5a4u&cjK%+MtT1!25U5&VKM1=wfSaays2f>I}
zf6+t<4v<16H-IK)1*)|I-Im5Y*N8U=C=mjH;{}xzUj*6P0C6B$Tm{Zc`@=wy8XPTe
zgm|(i`Z)>@2UCIkrPHoDboM&rU5GC#7p-I_1w8@=<O#tm@$IQHW~dt$Yd5g|eog~5
zk))(Ien({FN<%m1@W9<A<g)iQNu^0rUZs!R$oQwu4WRvqYT9p<0yECEpn#Jd>4;e_
zop<`wnTJIlx2bTE-DOgjal#_`Q7mOL3>@i)GSRY0P@7MZP;wsi#*Dw<1L~j(o~xP~
zt=~cg)fzhHu*~h%_g!aNy;cw1qf@QH$W=AaMHu0K!jzp0<d(De9G7Nw1yag4DyOf1
z<;jsuTb8vL{VIQ#TI<=Am5~fsij(;)GVwhc4HVllv)MymIcT8)c(ynoc~t;($Cdz|
zI-S}+|NH>x2eF#P<>GH2#F6aL$eca;sIfRQ6MAEm?J7zITFg1Q5o~z~CsY+S^Nq>5
z8p;M|Qsz*?c3n$30cB4Y^b3S9t>znRhJ#em*9sJ}0u&&(R&_bp+UHO2pcv_S<uLZ;
zq;hu#e{8T5bh4zhcIEkO>HTyk1&6<>lxR-WomcJCcpMMBen<lVs!4-&wtebU%Mf~`
z#CPjeqSw}%-j`U`p;Dmuv+gX48DA=@4E5}49%+>8i=;m7tN%z2;Dna|*As`&W`f6d
zZSgJwnoVNV1C>y=$v~q0Pi(c9;SqrWY``0!wX+5s1=j=AyGZHxL{g2mef(%XS)2f<
zr|+}<;>2*URPsftj8)NYK&PGAY4Q-6Btb?{p1tY<4b7kek`;)%0Y%?L=5+-$9XL)n
zmKH~cBO^=qrH<2eE09j*(J&cI#>Xf?hX)Bw5=dwkfApX|Tsw;40;s-+mp1^evlqMx
zBle+t;d75J|6R($Y#o&D;XtGY&j7tsB{hHxcGl<MM{ow1ApqFLz#IlF=W##VT6UD|
z5&7d$KY>k1J_}_{N0a#&brx2DtN;Qy>pY;W*yJ=|eYnzR8B(m%G+7^?uap-$XOl2a
z1E44YQ+|q}@!$+3xv_u;Vm(Is0;T5K$-RZd+5TSEjETqNQqG{$INt9d+%xXH@J-2k
z7=05EMb$R*nuw@Gv|n+VQA4yLponk*1=+BBCjf3J&WWJAfb(q(X725$jpHOhxDN{K
zVFZ;1EgYBwsLLM`KmFYRqC;k;J=87&$sm9cJ$fKWNta&xfVT=rw0YgG0wOiKkGXB`
zU>BG#gS;HPkb@8x3p!gCOw7%pTuJr>s4|FQv4jqOMY29@D%RU25JAG%`r`Va?=|Qb
zWf^GzT}7<7>7LhTqZx0Jpm|7sn9u{REkl_ivB2vGJX}KddA;bN1dY33zBVU;?1T~%
z;2Hw*CSM20v?Lc>J&l3?Kcn>S1`Fl@A-gI=MEfE9bR+wKCXtp+^Pwn^$i2;0E7Rqa
zh!O_Kd5;c7#Wm2giU2s=2gr_Z4+WPva72;i-_#Flb`e~vowbPq0Vgphb6YF~3VpEx
zV9Z9CF^mu7{-P}X*4x>LAj|&(%oJ~dsR0E<1n+im_^CnC*Q_yE3yA4Q=|8ZZ!F{7l
zulfUs=`8J!x2KihbtU>Jf(0Ni7y?1hEj0egUo53AH~5Lt0EYRYtn6r(7}_-}XfS#s
zArz{WF>S9L107-{2!7+fMB*4AorDJ=@8vi-E7)0d2n3ehslx5l!VuOD@CC>y;x!8W
zFqmebznKi)K*0(awo8D~@C?WzAVM~ix0h9@8A;$ics~ZDKapK3S1!$_I0(Eaf&T$$
zy`2#phHPpqCZM!cxH`kSj%fB1@O+WrVM&WZgrG$oAQV97BPl2IDZwlK0<-eM(9r;I
z0nid(g<_i=-=#2g92o8kREX701y-v)N(tNzTq!OgKs>~()2O`PxWhy=oJaXX1mYqk
zkTL<h=+GWGY0^PvT|)$b1awNUDT0NJy(qwiLEu;)OrZy)P#oYyrGNv&(ZY%Z2)u!8
zK3@RuH0hOjJ&%Rz6AlONhYarbUqVD&M87YVaTGk<G<ZGCDs)l-+ust{Rt~`lv~j8&
zw9b!ZQ@AGbw0jT%yW;@v>96}X`r#CS46)$)(PJFEv2&F|W#n>7EMmCdq{9s#OrZ#M
zv!iz(3H*$kNQmqa$kge;r8ym4-ygg;JhUBDGo)xdf4q~ycmn7d7ded70}As(T7NiQ
zu=BPL(41Qz291Es3X2>pSg07g7s^_YA|OEhwm`F{a1Oj&ge)BK87ca6I8dC20XA3*
z%8(nnzT$}_Lk73R0Jo&d=wy4ir97n%XlnPg-QU`7Ew;6dm+CZi*UteC_J-#DR!*EG
zsDB4iI|h<oU+clxgO5=Hz=nQPe)I?=1*Cve0}hU%6XXQ+-@WE}DLAMZz?ftdP^v_s
zDC;pUQ#6ZVr=K(zRKrzC25|5x<1#e!^A@N;2zY{T5Ule5Q1#vMRJZ^C&gsaZb*zlS
zu}4uE*_^EGtTM97mXWL^9Hq<<A}c3bgEEqtv}nlAPDMsmLLvNK*Xi^5J-(lR?mzDP
zcHZl{UgP<EzMhw>OET;O7K!k@1Qn}p)h|F@i|C~u6dk%^A_;-a6mIs;S3w`N3_AE0
zz}5wIOXIn*5=o$^IAqn#FOI`%lYak6J!EI=1=J}`>K#5<Y&EDx+YyuhuSP5P96Va#
zavmV1TL;SOnfT!PH29gyF2f5@M!Z7+T3O?oJB_YB6yN_-C=raYsC@v2DkFXS<=tA9
zjKlyUeOAaye<$=LHE;_~zz~I}grfy62;th|LVyyAz%<w`!AF-@FVsK=i+lE*`GgXq
z1YDJj_karEwebclT;3Ovd-fQ$ii&~x#twMQ8j3b2`X+paq=zwHeUFX-F%=(`OuCH6
zC6l06*K}ccb7P+{Fk=jbgcJ~Kh@b|M>C87?>K=JS|Nf5hX%zXUKY0LQR1V^O*nn|O
zGM}vZq1Sb=7q(y4^5HeILQ77t<rno@`u2c=IDe!lJe`9kBSMF$fV*;nf>JJ6{JaH6
z1xJ9Zp0nZUAr8htq;-yn*8aCvW5}FIs{n@GFe#2NMczw~=)p$k<ZgvY;<w6Bs92uw
zz1|6Lg#$p9_ju-;|6-PbJ!o+y{U(j%foaNzj(%=13;rjJ0A;%~5a9OvdlEQh#K=rW
zQ0oy-!=3gs@vBe+EIStUVMmM;)X#C^aLt;dP$#FsCTvyTqb#78Cb28+yf(B^{{tM0
z#3NwcX>qKT+FOGOh^e*T9J(Gq>h5%FE{r2I8goGY)iqe4V2O2G{K!}M@37S$s}*k-
z0b?XxSNNE`0*^q@m_Gj-mJc7Gp=a=d8UnC9A+@_Ek1|u^q9h-smrnK1=r)HNkHCh9
zLs=dE58_1}8htH+GH57Z&=Z1A>L@)a8kX7ryiE$OdwLYS5#m^zE7}xxh`{TNq6cH6
z5evd5R3{qZ3~2b}JoF1YT^jgU(F-|&hxB=B?;%?c=bhc0ya+Im4F~mO^kxZT3FO{u
zr!DfYk~HWS&T;?O=BuB{;)zFk+<tiz1;0t+@%=s;PZF4U;15N@LBnd<{o#ESBR^IQ
z^9qDRF;GHSuyD;$`zt8#!FA&RU^C2BDhM#lRksJ&fZ5+d6`(Cdf9LVbC8-hlAffV`
zbVQ*YK#|6M|Cn<KOuLe~l`pkGk#s5}|KeV_8ZjJ!=SZtkS~tgFKP3uT-w26eawpIK
zE3kHAmNlP1G0A?7AONpo2|xLJtNfh=v%mV0;FXz44CyRB;x+`Lr~_;rXkCq6K`6NA
z-$WZI=u99BSqtdEULw)Z`^}HotPJ^{aB!2`8%t}1je(?;507>`hT+q{6-nr;^}mnc
z$=u+m^HRt1$fQbovj-N&Fzp}=wjQ^jBGuTpy;Gg20S~8cmL7$2Kk*X|2FOHLA#ovr
zX%c+b?+QVvx58?TLK{Ohzv^y5A{6YYOEJ(T|IbB9U<!8bazYhAm{!$ZyQnDqbrRx%
z0iC$jk#}4jkTXyX1O8*i$A>cD!j-0C<2(cxI0v-HF6qFM!wZF@dK)z^Yw>@80KF=T
zD}XQe74Gw!h#a;0!(=ug)?RDtgyO8U8z8PB=~SZsdJ_V`p|?!>@*<LYF*&#x*_`+a
zHN>N|;Q;PsI3n8lP?E3#Jnty3dteekLjDs<KXGl-+icTVcwl$_$16^o_{jb_7BR`l
zX{!83X`Rr}Hw6BX4Mw4jkZqgpY1)Zas03VPw3iFPd&E(C%<K>ZCcR55GgCTi0+6vI
zu9842Rd76kn~Ji5qe?R`iwQzzU|m^EMZw1-5xK%aZ4j%F`)sECUA|S7`&6zE!d%F_
zlmql&Fn7ELE>DDygiC=HLOKwC7VT$ve=ztZ9z63nvVY&0=2|{1(TG?Q1fzlzIVy|L
zytV+U!zS@1aXXd~5*7Q+7%hDiS6?w+gh(+4Nc=6dYOgV{#>BRTp#O56G|`q_06qfF
zsfYK*9s^VY5Da*UC)rUjK`#+%<JLr8btLIZhDa53lhzHAAFrTMskOYZEDO>;z%&qV
zz6sx6T(cf@*!BLIfvr$edFva80hhiA24yM5N;TT*inY^n=HkK<)=!O&dJS3RmM;)+
z&HDDhr6X+Rwjp$tKf4IuYqgHzp%8_pg2~^aY@6P6P9iEYBS|%*EVq&_8H@sGmJGIe
zjwuGPmo@D+OPZm^80#^&lUb)X<EXgGS=<N2gl1Z*8MJGSS&LF^Zm!Mu;-+24kp<kU
ztBd(>?JBn}tqf^5re|#!kT;VQy`lcf0es9N9$KE_vn1ezDJeLwr(d>DL6h;*pbvag
zObTCrgHL1!-*JPFi+nkBbJYcuReD;2{$iKf=m26Coo>JMDu(svJE;28px$@~<@y+E
z&Ve61*&S#M!jP{%S#<C&x-#W`otryZgL~EAHD3IAHeJpq7doR;uoTAM-*Lce#FG_L
zZEO8XR_C+~FXCWFT+pHbiThpFA9Nkg3+B7pK=>2hjI#9~|1mv!|LdcWT4aNOP%%WB
z71s25h`D$C>{uTj=MEAm-N_9ckn@vcU0octKtCQIfnmcu0{nC1LYIMTv{-CgVyv{Z
za<H>eg1>V)_`x`Pr`!nx6EaDsw=5l&Eofvo<u16dU5cdvlg9RAHyQc0ZW|5(i@+hj
z%E%7Yr+7G`e$iG&T;@na@WfFc6Nn~CuoNfhdq)NDRcw3W^a!xC7Qk`SstC11e$e-R
zuR&E)o5GI=7dc^KNAGgw0!V*kgV+3?v=3uGXmW~0Hk32~-*Ew~j$*|M0+2h~4c1p@
zUR;lE)0)=LjdvfAiV9Ft;U&;w9p1#>LM<2k0uQYZnmt&!-GuZ!NH_yimv5hzA-v{|
ze3BBxYb&h9ZV~b4iArT4o{P!ks{B@?+T+-;UB(=Gtl;0CzM(+e=)rFO@Dr+tq_1Oq
zH{cDt`n{9e_x%z2x0LL=3LTz6lgjtBJ>tybB@x|mo15$29Lkp}5i`Ph)Pr5-=S}E1
z7NW?tr5z{b6h;9d#WCc-rvCEj<{`mKmlf#>gzNrL@(-hAU+0Y1agi<rc_*sOk{jnN
zyBjVCv}kf>q?*>4#=1;h894wd-kOJcmh^-uP7K#`J237^Ka!9!(*oi7;_*@Pw%v!f
z8QddGH!(w~0`eP*dK=ZoTWzPl`?)#k91Y(BufLz=urwNfelw#(R%dp$U<2@14mQcz
z00m&651Q1REP%)wD6^{gj`0wIj(|Uc$>uPBp$j;DhK~r>Sz|`v5?Z!V=Rj)XJHv&c
zhc)WM^{HF*Eq~LS>Q01C`S01q&hcD#Mlj@4l~KX<gi$1Zn#t~><GyJBc06s3;#+tk
z&0#Mq@NbcSJgel~BdzoAmiN80tJ=S^lad{|Zu&0QshvlB_{3AFPwVo|?+!O&GTLUe
zX7g<E?E!UB9yQ46tAa*lw#xDwOI<g7(jkWFh&lMJiu2Y$tUL;>c&0{#YENx(z)qRT
zPGtJTKhBf@V&>>MjM-L**S3T1CGEQ*&6e*m4F!RvROoO`iU6CN2NbPlqTQqKh40+`
z(Zq55-4(S->`ONh`bpA9;I9y~n*y9ego#!JLNZT~=(3=VL<lQ!o&J{{R|*Z*4LaOH
zP3G)<FqVF+&sL%2?Pp+hJP#&J&)jXJlq57mJ#7f0K{{|;(_^A3;d7?5xo2r+lqVHn
z6v!5^ptuX%d@N@q0qQkI4MxsuRsY^4XcuB~*8M4s`+9C{=S?T?HYvVlIcVi|A<osV
zW4>}bSHY*F54Q<}6vVN6^Vx@HTLaSi>LJ3XNM(G>!Gs-g-^@6U6i?Ee<#G?M`SUgQ
zhU7)q-I<=8m)|gP^t%j%W?k)G@mJYWC#||}6$uPx=)P_5KP2wPc`Ru6wZkGZ<7eb&
z1yeJwt<jq0gdk{p=r|={KR~oL-`*Z`VS+Sr0Kp+>j@MRfBJK6;>BHPp&*<9Lr$EHB
z|M#y;bM<ze(lNSk+DsLrOX6mV&whXZDSm#uNURK^&5e)WFQuCkx=vly95^-QkzcAj
zp6B70@X%`SEGUJF71^H+`b>9@Z`}+1RQ2NS=Ob#8n2?ES3U*d7{YHr9>SU8B?@e-T
zC?`x@agZ;8Fz`>j;MeG!j?k|nCft*X+Mhp$HeQ`!brIdcQXUIJll$#SQF-gzPS&i0
zhgV!(6f{G^yd1LvT##II8zZ{$*RBIU&Ux<ciU*~F*(;$^SRm>9OAc8!H@Ro^Z9rPv
z%{ZNGZi4}_l%JIjv*t#YevbNSJ@-Vt;m(%jOSbLxn@sMXWR;=he?DTvRC>(i>DJ$0
zCw$+lGM>C$rCxX+3ka#-cfHTQE7Kj1uH1qQmJwJP(l!wWWuh(}1(Kq}h@t{PQ(K5~
zeXe_2idP2qV0Vu@8Yt#^e)ay;!SBI|Cr6Ssc;5%Jjy2K@Om%TI<X*{$7=dPAKLtRD
zs7+Qj>G-wX-GX^*ohA+kgHONHmFbJQ%RRWxxGZ<_{KzN6dal2Q>c9L`Uww4KKY4#y
zt?2}IdU_nV+M{i`Qz@%Dcj}1*Pb@OIMMxr=P=?guQhA1`jAKZI7=}5e>lT~OaJOxg
zl@;csIL+~~F%ykN6_NJ{?(GfNo~T(2*($jxv`vK;sojp1&eAQA;&It*a{!l+gR$Ue
zLBQ__2i*GI`-tVaVa0=dyukGlXt1^Owzt%VpFNo8sd6<$@iLMo6u+ZL0&C2)Z0bs9
zbJm@%$jP{9+-SY=S67f-T3;V8o7-f$uFaj?=jjO$Ji2O@nk@GQDz)(v{ICj%d8wM9
z=cuoobGRbGKMrLZLs}CMo##$5-(WwIdwz#bD-n9+_k`<@zWP5)w^|kx%a@cXt8g-O
z5HGm{jIqJr^7udf&uv5RO-wJ;7LV4?A3QG6*}f59%VM}|T<({SIk#f`T*6BQYXx2F
zJ=?;=Hh$RUR*WhKuk5@k*Mf?dIP%)|Mi>j%2%*>lm-Mur-eSXe=sd6)JI`#~Q|p7W
zI8NX+P%uNFbIOgm)3!Mbc(kTUk;;zQ(V?qjl`mI;VAcbmV#ZY`XC=wB!5C!re>Na!
zMwcX&Fx;K<wT7<9nD8xKZnvNBBC3~5<=%aXYUahM9V7kT4bCT;Yb&z{+QPGp@R0;)
zE9~);7dilQYXwI(&9q0THESj|C3Q3xN7mm`b9wC%xc%OKR;Ocz4^M|e8CUqgJ*c&$
z$xXQH&g|(&_Th5FMsn9ocO(XIW>j|-orKp>pY?VDc+{+Z;r(Gi0{gJhqF~>wY)B*s
zVFy+_H>#!uGT>#9h;}%?)CI6+JQA=wOC|BwN$y3cH8<61t|g`t>3A-m&uTV3q0DNY
z^p)-5HvfqQP;b0uZ2<J)$n)~z*PZ}C1!=g9K+wG-nGH;}Ok;|vjb(93pbI2QK;;O_
z%*9_GR*y)&_Rb2Ld^{I+d38MPgVRZ`*~`N}<syM*bu9w<h#LYMJKESGO_E_sWU@~C
z_nPWbyZY+E?PyqJH)PJ;?8eGUz%6~{dWBqpp&w?o(=Cy@R6C@f-8xX=IsI`4Bp8Ox
zJ-{k|`=RfB`OlZg*DseK44(N~f-TH8If<7k#-xMtEyH2r8GI{M>1+jm$<})p?m6Ws
z2A29b-!r`W4E)P#wl>#6*lJeeZzboS8tH(?WjBFN$z!`Mh(nnDj75MFCSb8E2fu(k
z(BVm71W7=HOd6$6d2G${5_^aio036p<C`p(Eb3sFA>C!T+a>ej7#)}6{;Hy1{Y4C;
zNX(j`R%i4?%6Ot?&glZ@&ByaoF7VC#fE^HpF_E*E3r$@3dnV&d?|>R2wsurRXJPP2
z!<?_#D^8>Z)f>FuD;__!_2zh5_>a&LAY#%xE`|*MXtaWGx4%nR%E#6?@3cbV#6<AN
z^e{p^2uB-J{n}4fM@_CDr*mHpw53Z0hZ{c*rHkj+==l9WV5({U#7nPJ<zAP<<&G06
zo29*o>a$*pz$U8o4?ZN2FPLZ~&`&byWoLJBRKgI-K0)0C|B~YETpeI-Smif2RMXNg
zy>ZL~gh1&bU2Kg3Z8BVnNE>3o;meu-c)mXemDwga(bK}lH~NahIc9ebE*96L8fJ2a
z!uds1r{(2CO`T4I<+#+xfE9g@^aH%7Bv)fIqiyH84awKb&k89j&){{qY!CKxwtE3W
z{4KWdJUI;4!A{lfEMK*mcKbDH(*6E>d0vo(NF#i2Nsey+HY3N+F~!>?++|6hRR@II
zKL^eB-G6-GB*U_8_f@N)qK$lSS_P4U&4RA}V(l%*)_K=#*V8xD?5`hYsLCpc8{jo@
zG8B)Q-jk(}*khdJaW$aLU)sn;bjhhq*Z#>NwBN^jd3aCyolNiS;NI|^w|jp|MfLdG
zle~`CA5C(ey?*#+LTIGF<qDEILM0&mt|pyV-*HUI;d;9c|LSdyW=Em90|P#eO|J~b
zRVU^+)EeABGTrfc%>haRO7|YENinGiP4lDhk6)s*HbaIfsrLAsPw_X~g!-fy1X+!b
z5BgiW))w4N&(DN*foJ3HDF-PWUgH{hj`kF-(T92nBLIYQV*LI6!AgB!@U}F&(4dm_
z{nvze#=}X8czZJ;?J7-6BoA;TQD?fL9?dMUF#IIJp`^=#TGB?w$^MF>{wF_SZ>^;Y
zb3mcj3N2T1`ajN`LLbK=^?F9L2GqFQ#Oba<IBzx}RVPg*84v3fYjpulMhXRx?|Y<T
z4MJ%bBFLN@7eTT)2F6UR!@kE5reM?pA;{>)HPGLcCdTUY6AA9H-|v>%MnJ!kt3^^g
z>3+@N_ZW4yf<*x%WBmyT8>z&qOUu)0ha|?Y{&z%&kpXCUKFo55H0agcz(i-pv&Nbi
zVI9Ck*^HFF9fgwdijnnOe0V$3W!c=~BfwTZi3J2A0?kFi53-)bq48H-ViS-RhoD=o
zw>&yslv6{Ko7)0+51gH_g$^2+B0s}*zbWKQmWL^>aUi6YWPLBd?GlZ2sVTbn+87d!
zr-=TP129$`x_FOhMgYb@M9Op<Pz|4@IHwNmgZ(i8wzKF_K-bp^&8smROQx*1#8DMS
zNB~0!=H${dfx?!U$cBZwOeVXSKVD7^TIHlP$wX1W?83o6#j+VSv=D)92l6k15~1Y^
z6a#x7MIpX|LZGh)Y04th{W|j_4LcrROrb|Np@oBHSO`XMQlTH!03lOqhm;0fNEhiR
zmv-t<<FGBg63|S}pDA-Z0@SxCiSbYW1)(r%w;$Xf=D*_;Xq8=58%2U?U?~nOPSV>j
zK@RyV)Lv+?PpWAd%!F36Wmb<7%Ms{udKnM=Lqs?^Afx-SZ(YD_GjX>h`uW49zJ3(Y
za$2^-qPPrb`{Xvqa<o1G_keI^=KS%Fxbmf?M8FL${lh&73)xMeA~}dK7}QSwu5Tq^
zjbUhz{gu^mXUJ_VIKP1NNYEtJCdbdR2&4wws~}~YHMr?MMhQX^Cg3P=ac)TdOYDbR
z{{!p*W+v|4u$*e`*P&?w%)_B_B0yF&*dTtyx|CQ}2vtM`-N#7AGr@-y={Z^sYx^$m
z;Y&-dsGkK%k+`%ZFCJYi69Gcw^}pJrdbQX9S+)BPINJgUbMDd^=>UgDCa2+m3X~0k
zBn=w2hgLF&D)>DAU5mnS2GBW8x+d5~z=j#&8yxiBBB)%6_i*Uni=xnc0vI`*yhL4)
zBwhlvZ(P)t$mLFu7RUw5)I^w^$p3eH;17hOp-(ufARn4@06^;^{)s?{_V{9-bU@Er
z!5Dlyz=uuzum`!r1*rmKTY{0(`Q_;z&~&9;ys3mRCxVCw7HGs<PziL1M1UgoP;t32
zJ3az*dS@)$ndD(@wlnvYJH<j_;}9>ewHv%s+uy~K(fy~8KHgDw0&K>4N<;&#bR;BE
zHZ=h@2r#^G_&CO)8_VWh@{0o7&`Kg2k#W0~my8hIbwr{9E{i?Lt9b?Z0nljGZHg$v
zF6PB1sH_!Bd-NH>ga|ip2_sT$leNJq#t~qR*Slzy(1nnySNsIRLC}Q&qp?FHL$InA
z1Va8@sRak9k5lT~Z&z{?&qbHh-h$^bIv0NdR>n;}81#o;y)FwRq&JiHok4-K57+&5
z65^^T>bF9o0ShJ4!Kd=2xhX*17dj-D?cn02j70JUV18)ptT}+*0YfUecCa6y7pIxN
z$F!%l7`BHK5g32=p)p#yCOE%nGkD_#02u%GQgZZW#YV><F)LBe37E(U3mEwdSdR!T
ztU{BfZ^mfZ*fLWk&)~_O2!(+TwAA%&@k{7VS)?%`Z1>K}c?~_iA|ZV3DkK1SeI-vs
z4#4Y^$0jJR+<tlXh&*&g{KAS@v=h!V7u(c5$qE>7doBD0nxdg<G&rS8f*65ozeq)6
zd2mNP`iO|p0=Z!;pzqViQ^ycRMj54_44sBgmHyk|Ia?}LjYtEpWt))>kkN!FAu2}*
z9$VsIoM|kLA<DpJfk7NJssK19C|FZq+#)~fxgflYeXtVy+xaBWe#*g5N*WLI|4In^
ztIjFoK4x0#2ow#8$d2o#MRlQA8SpEI8T3VY|F3S9V1oszX%b8=ahNlA6p%sShjePL
zVA<_srV3{=bTdAMaOsGYP!47q)OE;%$>uUUz64_4+_3Z{W^{ivjP2t~2*C%qn69{+
zg%x}4IJ}F6=?Essi;Thx00>975$-v4K#bDL+Wr;2L~R3_<I5#A=UK7+X!WJCQx#7a
z_3i)u`3X20E69*z-^kwt1Pm#*ieZWvX*nf*zw7Vsp{wQ;4Z;v$>VT|Efha(O^T0yS
znF!tRTMm~jUYHYmL?z|m!;9jfh?h5Myd0E%_>5Xioi&7WNkov+3x0+FOjw8}uW^CJ
zcpi!wMf(xWOeUSDFV0&tHI&yxBf*w+0iNe+LH+nY^Vo)1NJ*)0xP$cpwvj6_FT3+=
z?;CV0#Ki05-IRy64FeBKwVAKIgP&uBPaI2rL8aJ3nJz-9?*y3a(~JdX1*9bm7jA!a
zMqU1E6n|XarR6WW@FV5?aEDxn&Qb$!9Jw|F4H>{(@=%46HY{y2iJgA(;1}4ahvs3H
zi2aqi5dI>lZrV-j1<)W4Q3~t({r!>6%09*09EZ-rH3;XKTnC>#oB`<sdzz2h7xw>=
z@?X?}_6s@ezm~g_wwibpz4XIOBIc!8KF59n_3{o>&For?H|eo=Ap0)u6sJcw))Lp0
z6SgUOs9`AyN?l8sVMys$e~I!kO*jraT~`GY1+V4In|B$)zu<$Qv^P336TPB(;RUcW
zZh@ZT{p6G1H8;1Rzc|lSeDZ#(rPEIUuST!%Wu1&O<3<<5AEZhcTO!C}SMCu&3%-?t
zz{up|QD}t&kcFRAnStmHf3u56LY>G8yWY&~DF{GFO~P29oA3F^_>2TitfuhyqYsWh
zgz-Pt0S3d?MiJF9H{WkV0i7hMgMKTW$;d$xk@=M{Y}$1^>YvMsCgG8d5lAuaM_CNP
z(;>X_D7^}b9UvFj{JSa{yy9^tpxtEw$ps=ZXe30`z@cRVSS7VRlq3Q4aQb-tEIDI{
z6+$qabV;Zq`ro~C5JXNtd2pl?xq8C=Z?n}1r(xjrgu5h@xbc6Xon8bFK@O)!pm+=t
z!C0aM1NT2j`wF=E7Y78AjuDTesESK+L{ZgIeYB+LP7>5F<<`>sTwkIbk;sc47m#8o
zqTu+3mIcw=W5f*jqdd!bZ7Au+ZX&_9#_@*<^55BkJq_j7Q|m<^sabCncFs*8I9oyX
zPO#;@x$}PnnGk=_oD$5}MD#!RpfDoM3Mmx_r+yry<MK<?&yW^<Cmg*y+!a1z+^G<d
z4-k!LaFxC$MTT6cI^02#W}((o7wL;xCjg8!QGpn{z!2?75J8$06YLmOjL=W`1LM@=
zK#L|r<ZcE2g`w>_mk+f%*P}qXHp$c4RiUps=n}mryqF-u#(L}~Gqy&65c^7iCzbbr
zS#e_8W#87e`;}%jVz8oJhgmKdZKJbFfJD()=I?eO2SkF`ta$?ZL4U=+R^%C#sE?HY
z-}z>R5pHM0)%2$D<{!7PGO<$+Cf7-~dv!jQ+Bh{w&`VxiurGmWhqGT8#^P2?{WK@n
zEkCP=hQ5*hSd0r@%{A$2NjJr=C0RfKG3+`hc4WPv5>#h)zr^q9UkfOMC|3^iWBs-M
z=kBtK+j3Wlt--F)hSE>t-4B<2_n>4lPThDfC2^5vccoNe_Nj;b(8$)QpIFy%cChP!
zL0b>b=o|%n;VISSbr{oMdzq<{<qsPXFaW{n^(r9uRK{+H>;-YmO-MLSTlN}vj5KY-
zzHR)Ccgxlgd}J&7PZLNQp+{qbxtnx7i(KfFK#|IeP-{0x1eQmwChu$i<yDg)Xd~Z=
zma8BQgy9`VBXec|G7_Qg{cljB!6nh7AYBezQ4nxd$=(-eA^bz#`Ext@50o9MRH7Op
z6YLhAp)o4I8{P!TUbt>UHTXA1;-vpv*xv#+H*6<PGbZX<F3U|ZJfDR_?4{&P4ToIW
zR%V&akh25K%=10_n7gGz^!4M69bV}g=Lg;t(lknp_j7B$ySXttq&(*f{t!|VGs+e3
z0+sh<r7H)QHT1fCNSB%)xnJqg%Ea=@?BbQCXS8^HkT8)91uWqrOe|$RUof}!(d2G>
zElMEKU~u(VPT|YINe)1Nl;v;VQ1+8OJbwp)Ahf|n+F*pS+;C9I^@I4!9ZBFo)p2Oy
zmk$;Pe;eM3&9K&D7MQ|v(PMCpEJWhp&MU|9L^6qp1;3>L=L84IbkfCE77UKiqKjz6
z^4LR-eP~L!>%p)9j_{|N2E#j#gWn?BV+@4yq8+90?u4HrkSOP|wTG*HCk*$ghF<={
z@I0LnpH720)500?6@I=R2Oz%k>6#0I$*nK%TbAs?;x)|BI27>UinTYiDmO#|4r6f<
z&z~k8z~YIc;D|-p>(~eoKj9cWF##V=7D|IT<ao$m3b=y1csz?Onfz$MlSB`<QixRy
zn0W;DbOS)WxIhgR<0f9R7k-Bz9uBWz06ZYv(9uzq42Euucx^KIFwvr16Mk3P=Q+^d
z*(Gfq=)Dr~k!YmB<7E`%3Xs;Pjg5`v7&Ju*A{2NK%ujh4EdHhWC`_-cfL{7EkoB`E
zriRhpnk5|QQpYg<`!)b>^MItycu-kWtH}<7RSQ#Y-37mV9tS9i6F4GN?eltyVWfVG
zM0#Yq4?+x5bw^^=j#r{_aG5hh^6&y)T91Nmr~-iKK20_omR}hr%IKA6U~BDV1Lv$y
zSGk{*K0qpAT+ktt+l95u(L1un1m3t0TEbrU?&2?A--Y3xKtqC)u89-+dkI#NnweP4
z71AcGtzX`4pvfVA$7U3hl3S6&@JN|xk2?HD?g5nX_rY*d129GU63oNS`C}@00%47;
zy%&CuY)(CJYk8{c&A=1SAH(k*{_`ZR7t+C;Z1DCG2eng@F;_{OIJgWfY`AF*w!-CO
zz+OdX$#k6b(UTg$fbPfwC_DhhBc*Ee$4)0Pm}$Ycl)SV*mIjMYA+!%8wN!@$9Pz|d
zwt$u*Oz2^2$+FP=tpl#nY@zfvf;0-NC7jt8mbG>axwn!NW{lkc!iP4wV4#~B{wS)l
z$-Hrq)#XAu(62{3mnjQ)Gc&kO?uGr4|E}{A7rC3OyexG2zM$<FQR@4$pyiZywMX*b
z8{EnntxnAY+9?N&FxX~NQ0smSE=hn=MF$fGo-l5C!lbGMtyMx0<^#jSR;SkqJYWxE
zguz{4B@*ACjDwv>6E2PA3|oXW;Tw4P#;#1X10U@`UmaPiaMBs75#8W1bk6xq!=s0z
zMc-Rl3yU7xnxhYYj=^1FZMSI#>bf7qC|8vkYxoJ0a5ITA_*chh!YAp$#_`R_r61zH
z!a`BDyF`wlXCSM*xNlzO1&UH7cD#TtEOz4$22Sw3@!@DhJUxnBMR&Db@C16Ts#s_t
zt3Ytyr+28q5d!mvk2>-@u)G8!apy69BiQFrPnsah+Jg$PBA!c8tBx;)8{Vz<Q*!MM
zum18^h5G49=jrd|2*6zlR2}t)P4$@;>L+K=W66X>6L}$s;WbX%k2Zz(0?;dz79AjE
zvbG-<p)bxM5N@IA0Hy9KObG_`K3epepgsU;33u<*Lw7VLV2~x(0ad`>1_q<EI58rL
zWPy?1jTXCrk(rqzr>fkAF>QEQZV`BH8tf6mQ;>&d9f^RXePO4&Ix9iMj&f<W?$L3D
zaNxBEkzX{d#V3GHbWeFVdVf{{_#(ZlhXPqBphJ^QK5&;bt1?8hufhvZTq^G1`0;KV
zHMkMEv1_|Ofs>Bup2Ydk-*rEWS*l3HtFT|UXC)otQVkut6ZD;9ef&_^)IpNXqQCNI
zkQ9kqmcI;|nh)Em>nS#VaeX(JJDMcP2DWT!_nCbt0FvzX!ZtbwQ?oIyyF`vsSD*ss
z#*VG$dqVA@aT%9V;UWy46(pr~9L)5E2E45&w=87#oYVHJQ{@Vm|2(<@XkMH0Z|n7+
z31J>|d${$h_YJ~lJ&WBZ>s5)A^nrBzz2{lj;FGne*~5fEA-C-Vo!hi~>og#I9*z~V
zt_yfrA@KnS`tm--4GD&wJfKTopEiHy{wAJF?-}!HST~Yb0J(FKe7_W&G+I*9ukfL-
z>L`p9_FZbrwZ+d|KHizxefAKJPYgbh;fvYfx=&M!C3TpQ{8xJOH51&yAO-M*%H+v~
zab1`FUtr8TTSq7qPCLk)6FOC9TW>xGfy)_!LXqqNZhkW9C9XTfdI#GcPU#j7+5L~d
znp#ui%t2iDHVx%EX%`%QxZDnwgi~LwyBK<BKZb2qAQRG(yi94gvs?398>)3ce0Q~p
z73(;`IZ1a0?7r=SDag>Si_MVb6b5;M#u#069d=m=@q)Gt*L~8AEjkn`_~T^@DeTmE
zb08cbs24D9es-D>X*`_#cPjD{G%<4oXIB{m(Jhc37Ofy1{tX`05qf+@_SsF)j#NDR
zJAP;dLR8^m$5o(ti~D@<KoIuBNf&CcMQ-22y+2#c)P(vdS+tcxUvf2LV}|#)$TROI
z5yv19aBynt>3WRR%U8YmytyI^wA(L}kG??^#hB`$3wqOufO8E)|DJfMZ5!;&>rJ3(
zHh5>?>Y>vr@ui~}Q=kDmgV1tn6o8@%h=dA(^cBk2HwaO{9G$to7iEp<lMi+BM57KQ
z*AKswe*7K?Bct>gxxE-11CerVau1in%b8r-M-wS1cQn#UcNIJD=Tg4p=RSF&@m<p@
z0dEDX=IR0%+?bZFfPLih)7(A4201Vs93JD!ob>tzStsH&>D(Uotn(0Cn{AAAQLH77
z#v&1_!(g>{b!)QldJ<1Mk#H13#!m5E&GF%UOU(^!6j7Zj!nQWM9k}7*f2dJp?ajc|
zFWv`qqu@vW2)4tJ$GMUE8mYQUyP%{!*fUHJXxlkQ>ViOH`0kNQz#+GLmv-p~B+G^c
z?o)-8TaUGJS375VW52v){k42YYLIATFoUTT4q2qC900NN<qw(f33#-Xub+AfpilO-
zqN;ar?%phnBDFpWTs<Vp{ui=(7FQKz87A1?sYWTo4r9R{kKcd9V7ouCd#7PCTHRQP
zAsjtvA`g@S%CI#ndi_G%S}X}IIVei}0Jx<elweuPeu48)TT}zT=_r>wlUDhtBrvrD
z<UoC*MOh{*nEERKO)}jE$auEz@r$Jbas>5P4`M=umB4;w>S)FT=M(y#^!Rsf(*So5
zLAn~9sd8*(ayhn&93gzoex?oEFF=PDXxly}37UGxpV=3$!;kJY3hV&+kOio7)NKrb
z7KXbe#F2Xi{N5zd`EI|UNkIkJ_|qRT?Vxc@VqmD+yuCN`85|`9DmG7!gj2*n>z}tq
zq;EnWtn_UL@NJu5kO)KQaOOxD*M#aRf&4mda`zW7VRquwQ&U5WQTq=0bpo^{nROAZ
z=hpf@J-sw@&LW!OUfG{*dz41UC_d^aw){fk#cJ3vE!-hX#^(SJ_%#oNLlu02{wb3u
z;PeGo%XZ^Z7e6~|0D{4aOt0G3x{~*AGJ0(tQnm3!BMXep%Z>@mu;z}z-C104VUc{e
z@E#tOp_)fq-*~qA>MJ_f%B{2Zg}-GW5S$6!U%N(Ljp@zZ`($2fJ@xm%s+Hx(-*dy~
z;1JV8KgQ1v%KSuCEyE$oT&`8Hk@9x92E@EpR-@_;ltkgUg-LzMRBL*7Zeh%t%BvEY
zk{z%@CD1&^X%L<v6>8&z!FB3ibudkK6(w}>>$5TF;9qLO4w4YpL5qlh+2|;P(S92k
zDF`1sQTXrwW;B`?w;^^vU0nR06lM5O#I2KdQZnNU*ZF38;Xd#(keXqf(c|>iRs+tN
z-gWkv|1OY9S-77qX})z`3dUz)2Oc9~?tZ`7uCxb6VSL$RG&l&wDX<FVF9SMuRvwO-
zfC}@Z{d9N9laTcQq2%au+e{BZRH9(9R>x~0mZR#InQ*EpLA_ZRa?e5)y&zmI5no(a
zoEkOtsmxAWSK|n^6=^0?oT8+4M&aZo_Um7bw&TPEAp?l2?AD?Qf68zfXsi{q&rl(j
zw;o`9I{gx=2w0ihGa@+cN~`Ya;5BUEg-nZUufl#Mgv0FMi^|HK6|vi!_0anu<x>kX
z^~l}+y#h)w_Lu)d?F%E3IHrsqG@lXHH0)t3h&7${Z-uClHGfAjT`QPTcmn;aQ$*Qh
zWFr#eE^afqj3X@A*#k!V1lqu&y5g;xWPWPK-H$`Mb1*QvJLtNfQ|gO9k*NP?A9)(`
zkXtxNaZkhF%y#M>65rrEO?_y>A2m{#umltt(CgICetyQojN>~%BC+=u=8eLpifDpU
zW(_c3K+gRD*(GINQKA-eXZa{zd61<mp;LLCuVuy`4IA!=*D|V{;RHcX7$lJIF}a#o
zwUSY2YFl%mfR&YX+^@dw7x-1cjGjL_*%7v4W}Q5MHDh1QLoSHPDIPyH3J^pEU@IAg
zEW*@!i<qc03L)zW6l@CTn4AX^J`wO0G&{N?Yy8D5pM?GW#$!lA#SCmT&r>y`7rSO+
z;Al>rSvyD&if|IECn(~3AyIa(lyjpB6UP2rkYUsOd-N(QkS9gyv&~EQVN+U0Dh_V6
zsB^3u=fu&Dzv`o(oU?~;=-HCU`}qDoR%#4bup<T1^ITpyHdyJF49(fwE#^diEM~{~
z$iw!dYEESo!@%DcbU=0SWCIE)9onl5xaAYahN0q}1=Oe<coWC>CV_%HmnL~ZuWTj8
z{s6`lecyu^*8@PV6-dRI*v9fCo*)r#)oIgH9R1uUFvGjTemtz!|Dgxb;=r9-`KPFR
ztC7JPjCD={iW>}a$dC_d?+GnwG=cbEg4spCL@su*S<$N32>dq8i0{a*M=@p!dv}5-
zWC=}=`x+98O8&5p+Q3Zur4ozI8lK>pBZF1v(kdE2KitNtES&9^#O!EzN2xzO1B8wr
zj)Fr&9h_8|U0<L6=Vs*MuRN8J3qR#JuvSr_!0q|58EE;M>cNt2Vhe}3eEX6Eno17<
zUD<k)a5{K(Lb37PRB_Z-w0&)OJ~YkuaIucG-*Mv-NF@O<b(pdC^2hGz&8eSY)p!OD
z2=ZGoepJ%*(9sG6CZ2tdp7#$yx~zl49SxpSOyGX9OLMJWp44$*@vna5xX8cf*|cf7
zhf=~ba&7FrU_W4ZzS&v2qeM#_%_P6hKG(`;lPJVXkbyK^TxF552pazG^kLnNPY}+3
z%=mz9O<}*(Mj@O%mtPz0ZonlQ^=obz;LCkrtW)6z<zrif)X1mhi%|Bp6+$(VDsS_w
z_G^ulEC`{i9QQV&1+2>&zw7bd_H57LR{?dKFKf%qcBAsE$vhm|a>ZZ#a4PkI#_xL0
z_%U^e5<G7)IqEzrF#mpp_PXxX?Yqhcz3Rk&RJb2CmW2}jLx{z6k*z6DP{CAJ5nHg$
zw{639S2N@aALuv$!08Ntn16iF&V0_yNg>rmBiqpKH{#5y$@~l7Grb4E;b>AKBTJ|V
z1s$?Cc5Nh&Vx(LO2XxGwZiMM(71Yn(-M=6|w1<27j5+ZltRKbL0_R|>KuZSyJ)W>T
z(<7%|7)}@g9d+{Y#*k9wV+dNWY!9sF!-Y!>MlSz_ZBn;tR;W^fT_hgC);>am{cv_n
z$k}OqO=^7j%WPC_ynIyW)3&$IDiWEon}F%8d1w6I&zg@EO>F|=KomL~PJFT;J*+9r
z_qsdB)l@F&)|nsuE2)>R+k@h@I5K6blAe?qxzh3Uxa{i9NV{d9cP~FW;wSl<VGgE7
z@pjzz!c-JjiP((uCti5n_LL^rO71cuTWU#-aw}atEaN5d!b`rv48YKH-NiShj>QVO
zbvPbCatQCh_gRvE&$>L^W6_bXP|N(X*m_)WUIok{EZuX{!4j!xx48G;v#m-)+csKW
zl&w8=`syNylTMGTm9m2j*IInIidvv9*pT%6r{-Hg+kI}$&==C$pU~Zs8bmxQ2=8$w
z%fb`w4F2;wsZ_-6Oc&^&n&{a#-pTtKGEj!{B~r%)4cNaG!pp7MBc67-IVqCC?v&+F
zOo?LYYnN1gs|*NS+=8-no7QYBEKd`4ceH4-$x6TTD4haXb9JmlZSl}tTg~@<g&WW~
z&h@1PJ{Qt=YICvdR#bzdCOEgUK9)!Pw4VD+H=^oP9iQsBxM?67{Bf)y_p`X*#%#C8
zcOOGIbNtgR5ZQjc4vk-~?<+68r;<1Px#Gbd-S+skxZYsloSb8GyOISCGZU`o->UXB
zwD*dkp&`-eXdE*ebVk-SQm2bc54tI5UKa2sPDl`Qrs_vR-ZJ1D*7M)JTffBn%c%dY
zSnO#p8x}Y?Dc;tUCJZd?YwN~mi#9D5=w8^(L2GxrHg5Cd*58DlGP}>8->au&4st2A
z`R`;5^oVU<7xwLXt=!jYy}oGIqxs>fuy|chiK%ec(<M05TxPE4!|9DBS$#I^`p<(`
zRBEaxtlB5U{5GNgup)V@1YTkJKwbqMV`@$H2xW~j2;M>I>yE#!IFBGS*5Jp_Dj25A
zR2QpC_&puDe(f<I$EEqVV@U~tcB+dD_f<;6H=k~Omwo1wvv<9O$Y-ZxSaj#oy^>5X
zr_T<nV!n+Z^PY5a*#tfr_V~C3rib$859<Qwz1B8j&vjZjY{Vnq(bAI-%o_Y6EzKum
zV`9W^ov_~1OQ5q!Pxv*j?KEgOye)Z&R+8;d$W3szyfyW;`a3ZyxRdEL@_}gpF_>H0
zqON(^vf&><>AD(TJ~@O#Ey|iisZQ|Xt!^Qm8oU<i_7|H1*=CrH9lIRfDW~jBDo&T2
zhtX0+Z~(|^GKjUjuNTqy;K!cA-fi98FQ$)<wqfs?Zp?3OmUZ<Q4Gx9C9G(kwXNLC#
z);hz0W$)wJT!^e3-sV4E0s~$p8wY+h1y{G8ZYy70n)}w!nWm>QqEv5g`YStfdQv>L
zTE(d+XZ9zP(fWP8g;#8L@3TK$lo78&zy^cRc^Gx?1CT*+B6zgAarxnuTUVL+6Ca}b
z&1AY$4kqUyS*KSNl_qYkeR}r9J8XT>&HFSuZmSI-&WW5-!Y~}Y(KeF<tMpmcH`MD+
z#M~50_US^iqwmib?;-r2{s9G>D`1`FFx2*8SbW(ur1L5s-!Lp|DtzE)(>_wrRr&d+
z(Tj^e8|MPKM<|`acbIy#mBsI?lyF@noI`A|e4x#8-EnGyCpGN}2U7A?_m44ThK5yB
z)oA5+4G&9L+_iR)u?;@Sd^JEHO(o&AEDfNZMmer%*Y>WHrkL#4xOny>%|VuWO4e{>
zY8$q1WJBg%abwmcx_4qCI~Cl>)wMdJQ*Dkxt-<$46oXT%hsMXhg$&Ot{k?Z0Xm>sL
z9?Zt$UP;jmu`Ft%1QRA$1bU0x*7pIhFl(MVc9mq0T5DICRE@x;AwLjZqeg6~$a1Uu
zPZ<<a15>JHS1aw^YGvI0G<BXAX8KhdRh%4#DcV<?*&yzkleo26;#E2~{1$vFQ!;bv
z64uPl=|^YQ2Ob;_8TzEVxzt#y^2LjqQNN`|=lo=Cx7g+1)64SZ{SfEpfI?>!Zh3Li
zJQ15uCY$-`0Re;##FaP1(vQu~+-PD~ujKyJl>#WT*iL`*0Di#j_xfwu#+w7Bb8)r7
z9e|CU-AYNzY2Ed9iB`TVf|(OP7oXRS6CUXXPo?gz#nS5`fUcB;G^|3EA-R%A%`Z=|
zYa6}6Yf#50cyg7nQ|sI_xFs9UCNc&WaqY{*19{X3(VPag%=Afvv9C5p%FAlI_rN4p
z`T$JW^_RREp<_s_m-6-+sBB%oKulmgU#nA*nR*M(uD@;d?IEC3uL8R0a`r`ugEMts
zZNvzd%(gKAG_nk8-xXWF85%pz{q|h~H3?j%_3JtH%zL`j;6Ka78+zm(fvf3_r&mJ^
z7u6|z=mT|wU0KGZo?QVZ`@|NSW;06ES*;87tQytM4b_NtaVZAoyxlZoGc&lJhxB54
zY!RShl&T`tIol}%Kc|=7J6u>N0m0vpHUaJ>@7|!3*^fWBX6{WK=ML?8bJ~mhjUH*<
z#eW!Z=!;bcsGF@>fTBacoE`0zcgCIdZIU6>7};;PJb!IJ50hCRsepnqr|hk&m*Kti
zzI7$;p8=468i`vmO@3?b-*eJ%li|Cxm&vxfGI|qGK<EeR%6uzP<aiJM0NJmO>Fq5B
zSERzKwiM8bx^4ZPP|cQ!T(Y)ns?xrXRTDH&>w@X>iO0WHsZkS^hPDW<2)sWf!C4ki
za1<?@+wjS}0k&I#8YMMD|FHge{z3es8I9||$-SI@$#d>-0CzV`{~Qc7ZBuo5fBUAw
zYsGL~_it=_q_sZQ9WD=qgQ``Z@%#o?4Elkky9JLUkao`Zw_$SS#Wd%3kGR^3G>Iv%
z&!G67-d_i#UnVGPf1TioVvrurpFk(W73_$C`hMbPo{Cq{Du*h@=3&p`VZdq<?8b{3
zsmsb0*X?P>OmCQ@GQ52kCric!V*)T#L%d7wDHn$d^wQ3LTdhouleIYTIQVsu)uraT
zKYsy7>6$AmP4zN4KbGzKYM8tGgv~1xN@27|iXYvpUE6yK%zp6L0x!XFqHL}d;zAnq
z9S8i`I`-{!x48q79>s48F`;93)z+ZxxHbMtn*jz-o`c%-cwmS26YgtT5J+<cDI-SH
zv$wngR$Y4{I$v8--U5@7@TTU<t%sR~`JZg0WJ7*sey#+i#6i}Xr-M=(-DxLmSO=hJ
zcCUt>L$2aP!O_b<?V&Jkix2^z*yrfE3n}HVIWHyb`1<&5Er1fjG+uxoPuaEp3?0w@
z4Hk244+?63l+ZhN)0(UL>c9>Dg1#teJYcY7Y=wB|TrTgh_^D5)8%}5!(7JyrowaUi
zzZR!Ih2YB#NSKcSNhwIRZmkTnwlxj(3K^Pku)e_ZVQIB+?mKJ4B-<;62@_{=sr=H-
z)xSgU3ND}NGygGtX{#;ruKpQOb!Z;5b;g!fzcpy@PhF(7iMQKaOj~K*ap?B--BaEt
zYx5sppT4W9dx%n_v@$#BGP3-rYkunUsjBG4Eg@c?$VXYH8ey^~X7*?GvyBhZ-zGnA
zZ3a&4wtCE_r_|w*S$n@1vh>+70qN~0Od0ER!bpDO+QUa0!aPtG3{0!v>gG9W(3VRv
z8$Rv3Yx_xif7zbLwFmlOiZtu$v1f}$DxuxSR}*s)ppsGFqNpXeR+YOQmN@uqv9@3I
zrwxX&H~cw9QS)Y6c6E!keWdEOu$79Oe#Fjn-(eW=D>8b17MhBO07kW`ejwbdb6+dc
z90&Bqt@g#Z!SwDDMV1jiIL1is+g$yipX}u8ecnr6Ux83CGZ_Ug*iT3YO}2=oGe=S}
z5@KRDSt1Uc-db1S|Gd5TfiN!>kvwQM2JTrEu)@<A#wrh9+YAzpZB5DU9SR2GID53C
zE(d$uyHk8xb@OL};cvnD00dW$<NN}RY|<oPB<Zcw%MR*e;m3#+UTwhN@}*{8*i0XY
ze9_MuPC`(Q;C#;Jl5o*hpk^=s1JLJej0vmyIE$W?eV^32k1*|(m+HYTf5EYJ_+mT_
z_7<W2kGxHGPlVy$Bc(WjNGkL`b`XFBtF|9t`qH*)0pPJKl+ODj(6p{;n$*MAq9f$-
z8~3|#Vw`kO!Qx|mxh%{R#wW%6kA}#(DkawzxHmoSRqTGM%mkp-0ZMB&jFh@&VE>as
zNh-7ofpaql{quJu3*efC00G*6T4Pu~?9U4xg{{h?UjP#@I5+1eWMHt{cfhTvA8N(9
zIB_#J;}5&jMN8H*_IdygL3)H;<A2b44Q8;IW};gVCP(GNYW4mAK%UX4>MqK_ggpaT
z&I1j-$+Yy+>{~QrMz}=*N`c7hc&>9$;8&r6&Z!4k+yDLTAzJBpqlWiry2oyP;8Ww}
z@Z)$E3$jEWtrv}90uA>0j!DBtq3fMI4)9oLWnCi}9V&+|V-STBwX8e7d`5*(Mx{&F
zG*IXdBt7kO_tKrYK{|u$wlQOFU7G*=>@e-<gMVM|GB3Ph&aOt#MJN&QVCHg$HsdP<
z_BPz%XMbKDnz?VcaS!#J(#gfI*`|w1Y21#iAxB|g0lf!upcc8t*gpUHsn0hL<k(lG
zrt|M-S#+ChA{Ws4r)G?Dhp+`WSnM$Zzpg{T>4UH71*Ri<b$LRm9jfBCs=)NFgIkC@
zU^`EEHYSg^sjZy6&vdounghTm^x(kxAz*&QAh3gcxfKw602G6fQLq(6tZc75Fg`!G
zLRT(~#V`=E9i#8lPCnaQE*k_)>g~?$@7}t@%vk_vv=l1#xeo-x3mZE@8rGD-nB}O1
zEC!mi{i~)(9N6ijz=D{#Se8o$oRPu?A#U@^X`8n0;TibsZBGM(hC75#+?g4BVeuj^
zu09HSXH8Ab70?a!ftA2p8`<va<RogM03!$->=332ZzcexNM8%;Q3dk+q2@T-K?@^~
zk1!!Ma3XBYbRP_WqGQ^ok_t7yS9|n-^<xbIlzkS&iWgq9UfN>~4)`0N!Zy+ZnG*mr
zB!+|gp}fu0bQ@`+J(biav{;fFPQk#=JC8j1cRnN|1Z%f|QQzmy^=Dg(RcU*Z%qi9G
zJ!0!UWuZ=}TJ)bC=t+NWqzG!Q_fz{qKgfLQ15b%DKe|6a#2I-=#y;Da^*#zmu$bLZ
zf<<hP^INV0aU#OV><vk!|30l5iNq!bo<Z5a4(Bsu#i$cR%z<s*&qwPERDsUyGOT)q
z@H?e*j>*Uq9S+ut+E}T6CjMj{j>NE?kv{>jhaJXXZVBO~Fjn*sPM6tqHuU<>kN;!u
zGI#pzdx+zUToWRZZikzMQ+i{3x^BphxPD%33mhtv-}t-v`<XW|r=HWdLFC#GP|+<a
zS+rno4>(NWkqAGw<xjVUs4%A3Axf1G-R}La2UU%O^ez<F!y@1Q0{n{<>_mob=72Bz
zedb!I`@!7IB~5?unVmBY<s*|nVr!=dKqy|MSN+j4nd{$9yR{bpT~Kk2uG<#~wo=Mb
zJ%?++4C)AHJ3_$zfOr&t@PC%(xRD8pj~^;yKotEEdF1ku+a*gPkS;_+@BoSr7&}@t
zvk4&KSBbl2_Z&|#B22)2<kbXN*{7cQIAhx>C*-YCE2aUJ<Q6U3Y*rH2$?ajMFZ_cR
zj*o(!<0D|K+aZShIB|XK3W*lTW~rti;rM<y9qhVtMHn<M1|`|RhTXdK9(c3`wD5tM
z@<ylTBI)SAM^f40`W2z7dowGVCrlH$u;XaC5~l&Eo$diPS@B1D!BMl(=fJ_u!WT5;
z3Tl&|$mjkArN>{!z8c@654Vq8t-HE;J_(osDCNkoldM2+)hn>G{q}=*07}2N_XAl*
zp)JUni-E`$TCV5Y<M-(+=u6t}g1W5=d_eB=^`GAfR4aj9)+-CzkN46RD7K*AyG;xD
zy#oUS3==GO2fjR<mRA1C6be}NI;f)uK<$<_@U8CiMk$a&63a?ybXF~!!JpS{rxU@h
z`?d-W2#wLQCV&H+V5;guZQNR>RLwu|coY!98}6>1-~c}#_I)ghv=@ENL}~i>usgfD
zxu;_at8l;p<P4;KxC%R_4>5bwlGBCWAwN#fsiQS=f~v;>)vaP|$9jrjKMKZxix-oH
zQcF%(mrh*~q7PRCRY_`IY2BrfXi@NJI%lQw6@o@3tQIGwHGr7?=NP8FUIA!jy1p_u
zJQ%u^$5U|U-`j21_nekc088L7NdY$MvcX~ZwzD_GP`kW(@{(*^1aO!4Ce|a2740DH
z&A>@>PXJFWiYe6`KYoML2_>=_knZ^O3U_l}1FL~Ruhk#%lwNkT-<@FJm>=J@ZKG3l
zx!V-XRI1Jn1_32^i><xqq)k)D&O8u7Az?-?AGJ?MA}ffZccxxr=)*6g6>=8^t3jTv
zKgZW$FU}8zoZr)-;KI4DIILgb2FKs7oXxQ~@3+ojD<I&Qp?%{0*$br3<p@6L=XrVO
z-)@f}*#iV$=uhu|CYgK~w=3xpU%Qj^yX_!^9rOf?yWT>B%!3Icjf3FZ;AvI+b>HVd
zT0hwk%h!l^-D_VPyQZciME$9ty+(V=0rYCI%)jH67xa3<53r9HnaGo)8U*+)h=V^3
zqWibAEB&DS`7u~s)&TZwD?oSj1NfftqQ8$76fKPHDZ_qjA2j1as6Zw3P5oLugPOli
zCK6hTXl%56yGubMD`gxfl-cp%?sRTptAljim(7j!>?7AQ{Is?Z{Od3`1N96mq{W<2
zHkp2As<474T!djK`{5?=ZS=OPUIWA?;Vi3&uLkVd10CmUw@qey6f#-*h(H;qMt(JU
zApFf;>T}&Cv>QZ1aXA3HWflsy1J%M-A`Aw+XL+Gov5@A!MI!cNIU~YggTbboP1<#A
zmU`!W;*i$oGHP#@;;$tjPAoqowwVtO0>j*>z76s+M8yf(eU<+g18y=+f+bcLv^NDe
z5mSIU)0i*62aCt?QA%YpZr&gt2JjT=$OtWY<lbiibvGmjUAez7hu+gj;H?Kk#By8p
z!%@Fm(Hv7tsuH7iN)y=ZxMeE_2JFrIG(h?9S~_?hK}f*5etm!R?U<yoSL<YoatG!M
z*q?Yn;O%e|*Np)q65P;KO}b4b`wqwza8W*L6_{^3tVNIKA4+5tQ?SwE@3vrp4;HiI
zX+<mmE4hDmC?Q=J)CRfO3{O9xWeT79yh-3ma1|E`NC8kguiGjG{Q<pefU!TDGB{ww
zcuvs&dwk=)?bq61L#AJ<xCzuhvBO5o4ovbDL0;PadJHNq4Bpjh4y>^TfP!Eo%{fN@
z3r$~eU@GVe>GClUU8o${VNPhv!iQ_nMZm8aXz-ir;J%7bB{mhY?h8n90}m7~M-S#>
zgP@X_!9}R9_kVl<8PMk*w?BP4tT6RnA`2Y`SI*qt7lU}Midg4BT=2h$Mc~5NEc^wt
z1Wzz~>=2a}ypa!|N+9!w8h@fkIAYhpaq&jo)z(lRqhRzd+OXida@Xw&$KH|z-vg0(
zUbu)au5`LU;LGD-_D3;btc7u(%H@l@k3U8VFNj_pjKvVRKv&u_b(AyP#I#cc_PW;N
z9uX-lJ)04jlVE!Pe`xqB2D?KRDr&G*j6cU(VvdLRtlo}i!NC2q^;LW4fanTtr$71e
zKarjWt@LAk=-+De*gIBYbsy8?^EypZK*P2_91M2{V41V4wyeWn!>zAi5d{awBp;4H
zydrMX$G}9`+nZAXO-MY}DtQW=hn`5-)Q(*gsQcIE5R@?n6vzk)02*2@F`S6z=LMoN
z+Br8752X{p`hB3-o|1Sw`0wYBU}3=HCg_!})<s0S{rqd5LO#6pJ&O^86BD7Jpmf!b
z?IDFHlKSE`PCzxey%(6o*Nj?aQ+%dpvCS>gthBHr{Mpc`aW;r<?f(V;MD}oiNCrST
z;t4vY=n^T`4@elX`9L1>t(h97xVT07Zd@<-15h6eUifN+Rs^+)_sRnbPfEN_L#@dh
zgH?N`*}4aV!!ov;Ft7ifFRF7T);hx8fb8NN4*FtCVAecgQsaz-kW*|iTtH)A{8|1)
zX^agk7(zUT^GKxJP8K!(3$t>{j$j%$_g;>H4jT&JbFD<j7JewI36jMufTfR-_H{~<
zxkaaiszRg*C+yg{@929`u3;_pA#Fx+&j0?tKseIqfwQuO7oW}zLmO(WiJ?b>5uiV6
zlWRUJrt}jitoA*ci?Gs3w9<9Og~Mp59jm-2{XbvI8DsF^3AiW?g7Eq_%vJJ1k;ndc
zd)`yLfWauB3rE3et83>u(DR<KS}r;CZ()jJc4Zq~0zlPt@FJPyMEfV;q~)>r6ZwdO
zehB4XP+1)|S1n6+o^?6vJb;Uc@du-rcaps}N3G)3Jk`a-jexOwBjh%p)5$bdJ=ws|
zpDL|0|IIe;%SQ4Dkl&lng&PvqH^Ccrdrrqxbk@(G{q<pB;^BzG7`7Rjo`Sfcuho`0
zA67Sar{q7Q!$tgD&%Ub_#MG`KylMz+Och-J^YYT2d4QQNG&(NN4$3+VFFy}~UDm~=
zY2t?YB>11JyM`PRyxh4ySwWYxbad`ih~@akH~0Ll^q0@hObP0G_qZ56ndkA}hs0hG
zNjwY3MUoQz9-D`~DVJG%p}`iEl!V78cXoDnnPX^q=GX-6q$K#Iev92wA^Zv&k&H?H
z^Z_)3=SN9hx5oAT1Gg+32ZYOL_k3wh`m@<uz_GXyoUPZ-&`>s_sQWkc)-V(N!}FW%
z`3|;ABkEJx7lO82pLD?Jr~El4Ti*{dQ&MqXPK~HOmvO(K_IY$?D+jDgs=pdyUoE}(
zD5i<x|31=9Cb@5MQZpcFak{ytU1jxjhaQQ2Zso2YUcd^?6}8<C2MwmcK=|8#OJNl0
zOkike07e1kkjPd{08|ycM|Xn?<1$`^I0Jf;iZKn9wm0(BZ9<5;LU!J#>^GSRDh5%8
zR|RprP+#s0+GebTOA>n&Flz-kFYT~q^4fOF2zBi{F~HR0hf1RciL4|-@dMZ>PPFo%
z?8Gvj-)h;#r68U7mns5>#ahhZ%5=~`B2O`*)>K=KA_>*tt&VRNp!d<Y2dQ&!LvSFP
z0=1tYQpLcqX~if$5$h|n705WPiXlZ&0RDv64IcyX)sDf{vjC<}+ijNgn7IFiZzFvq
z1aDJ_s3q(E&+luX)1d+udniHwY3T7Od)wFIQG_zMefYU$Wy2D;$1$QJ^awLqeLFC?
zMG?x+t5DCDgJ$&m&j(oy{}PKUz3cujpq%l9LqR&k<O+7(g6A#HxY|r4H^H9ufV=l&
z8-GYGQpqHFEKr%%=;XM;io&enAdc`#xPLdxKH$c$G2I5QDLA|#g8ay6S5wfxnu4EV
zf1Zk~-*kI?elJAKtXZ`Zg7yRQ_BUWOG0>*;>-|ekBcFdEtsH$@unrmo2IlrG?@T1=
z2;BeK{Yp>|sxj<Wy4Vg5j?)hwnzdXw@AaQ3$amPWh0ukQ^4oVqH}+at7No0l%(@Yz
zv$QNbx#i9`RX$<*mqRhQw`{IJ+d`a6CO^6?wju&S=h05~qmUmqY5n{wK#mdsGe_fF
zV<@JhtQ*BpJU9tdv?p{$j1MudBv9_;U6z*xc?}LSt#gKvmW|!LQ{#D}v@f7tLSgwW
zBfElMaRuG5*1Pbp<CWOUKXl(UYe5xMP3|><H<wlu*1Z@K+D6rwkx6E>(j2D&Z$L3M
z+MQJ-5Ci4ij;N0%jHFRL%m!oEQoogH^uN9aj63P!hQJHM1N2xmTrDHfu$1-xo~D7x
zA6^Q%-Ns&azhExQzX+~zoP_xT5}oWXY#pL9g1eB6kKHO<l9+ft92(gHxEI)nU*@;n
zgK7n2Vm_FMpOF?-YS8r+J#(-C=FHj5J2CtOZp<)Ow}#|Zy8qYMSB6!&b<xTebc2P`
zAky7PNC=ydZfQk9Qc_7tH)(*<D7|S=5Gj!k6$K<kP#UGB1QZGJ&X@3<^F8-I&;5RV
zj|VpIUhi6S%{k^6W1cv!nZ~}C{+uKM?lk~2hM)s@t)g-B2nQTOCExZ32T%{OC|Mri
zG#0j9UKo6`03Jb*afb(Rj>uZE8K>Y!AHTE8=|%@BBL12PbEh&iBu9DovoZ1nrI;G>
zC)Vrt?uwZD-&e;N(%;em3iu!L&LJ6=xWr%OhyUz`0xfNS&{&CmTZbgKDjUW-pe>l-
zvmcz=gc{HkReAq2jC7EITIHa`VXEC<i6Djc%21K{oi><Fz@e9&5(_$yd}ew8J~3|y
z5bzgB(31<MZwSRs?e2kVDC0(CZni58Pj+Lp8f^D?*IOW4K#_H-*#NhSk^ZP|{QI5F
z9>uH$)BqZOQ*4EtV3e-MtTD2E?^oO;g@T?QC#?^8ObJl3HX>+@c5i+x7N56rn(<PJ
z)dMJ}h=}o907K$_be9&g8MHstTc2P;YO=I3$bksIkdE)=kc<Sp)l0G;M)Vg>u@<kp
zJQKn*YqvNJN7N6o&wdLfVXHHVpJ~+L-3WWeKUzb)85&|%fFm#2PcWcnV2^#MdJefQ
z)P{PkBpT-$5I>Nm_;(0_9X9?rp^X?@n@}Zmvc~mNsHY&rJ|>}%U_x4mm9}DVOFRW`
z-c=(Yn(l74tyFB&$uq$9rb@vhLd8Q;cXnop8mR`JgDP*}<xM7_t)=hK)?l+eE@^TC
zy3({rOJdq(SR&_ppCHBBf*0HXFEsv8P4sT)1}!Y!5K9tx3;^4Uzy;;}AQ(0=cS58J
zqFw=;?LHVug7AC7^cy(#?vs<JI?pvCxf_r#8Fg*%CqQP@=o9K9g@20ox!X7WOVziE
zN(SW{IdY|TpWUAb5gh<mw0@bypHorV1=ToR&L9lBWmKg^u8_sM2&LFWsDpMADPe!e
z(h#6f?Y6gY!%;$iSJ)6Ux}TuD^e3DS*MWKpan3z5ciJKh`33kb*th2zK>_?Jiba)P
zh8}5F8<UGbT}6yS_-??*0RaU1^70NygQ0(bjl0#IEK8oJ8Tg7-5sQrwMF9Y*j%mPI
ziGTR0vLEg7`EUY3;2mgn+H9#O^87(OGmFHB0Zfq6x>0<5pAH<TD@aDWvoDG7=3;qY
z^IokTG&*=LT@UvOU{B&hcU61Q^h;UTLnU1FJ^g{kff4&rX8q(1@5$*!enS5!_;xH8
z5UkxF?X%Qy814>{%74Ac7@kf@$Rywbr&Kf)(N4e0K(FSya?ulinGM%4N8KZ3T94gI
zjybz~nPIU^xT+vTWdIs1hJb$scN{(`v8|Rju;T*73D$Sl8Vdj`5gKT!_8`gTFh3!}
zrUwvC9*<_t$#)s`P~|5z%q>ucNTwuQ-23=?3Y1BuYZ0>BsX9)eQCA=*E<A()p~QF|
z#1eRB^wqohFcwpRf+JM~^Hl~Q#RoRA`NTrU@n!(r)if?$ya-GK=*2+)yYjl0#QyC;
z#Dv4%0Dc#wp#hDg*pd((c$}Cv*)-z2;q^#4FEckt8>X44L$Y&(Y=~CB^>u){O1Zqt
zM&zn#7AyVHu(+_37z_$-lM;aNiw5zX4)0zXMBG|H8J2)vt;3hs_Xv0ib52CVK6z{$
zSWHhWhg1weWCA2sIHW8?WiPy|nnK0;0!X!Ry37&v)<F`;Yj*!cU@x0`f(25@=@qoR
zJ+_G6`!hU{eW(nwaJ#AoO>Oze9nrj`bqfC)+jE3G39#IZE%V=R%ZDQz7V1j$T8bbB
zn@om9xZcoe7y~+BKABO!*4XV<g8~>m&5JZW=Amyn#!j5YjuaL$>tbT4@8nvTPoS>@
zD$$i{E|vM+BYigl>$PChD5}1tAPbv&4GsxD;}XE-{h%{Xx^Uy57+(DT`wzy40>Keb
zs%xjsf>*zI5aXF`^~X>%5ECKty7A*zPFzTXnKirX0W)zv+Jg^x`)ent;T4(3VCxdz
zKB(o3A8-nou#oN>L00_l?L|P);{cgn!3%IzAeyW<y9uv9OjR{I?=9e#4$wh=0lGAy
z!b5e+2V?sRE?FmBcZN?O6HlY_w3!KnF4zFf&~gAz41HZ)-L&ExT0c>ODFWsp5M6#M
z?v}6c?Z#E%4B)jVH$u1)y_z#F$v~ESrrTMYn+_k^WVg@ZX#vfxq*68GH$eRnALH*F
zqp^B)*;%>YcWv&oIXDNopil(nj@_>r6DEIw4@Nlz^2)>y+pA0`Eb~oQ0pNa)*bB60
z{8-WeLVFIEIZV#c8SorU#Y(tPwW|NJR?*m9p<f)ZJO{sl2hQ1tOLwHV(o&704PHa2
zX6iNlOl}|(u@Ev5<tn-e0J{<Gmt(8dho_&71BK=iFgXouOX6tH0d*g~b_X~YJAsh?
z#iH+`clL>N*XD4^m_@J{p2rVd5E~5c())?zaA37%31w(G6w;MdOCQ*fz(jBW<(P_T
z-fLPNz;_+{2?Q7qFXv>K0fd+2&z%7}Cdy#_mb5S;cSO9!M%gmPZ#`@FgUO)#a3jb7
zqP{P|WgP2eGUW%>J30-f(37!;NkAHYm2#q<%NB_+m5zk?AvmZY13RrBh&w4juX$@w
zBG4<pejMD0iy?2D%|ogra6aycUy!W@;+6&#;zK<pyuG4`8Pe<_yFhVpKYJn8z*FDK
z#%fVGe+&$bQG?5bvaD*M`aI01cmbO#$*%0?R|OaAcpZul_g#Nn^!NbR(YdJwz{X6V
zIiKf@)Vb?8C|DjM;9>h?sUe;@fL(0q&Mz^Y_$cS06btQc2SnaO5+hDi`)d(lCxfZv
zt21Dv{b5LHNLdoXLR(GX^upVY`-JSHy_@$I*vMACvhSYhWtLHm?d&5J_(AJBCq_IU
z>$3VdM7i91c;>t<=&Z{?#llZ^H09QpdtwtAd`pg;iwn4_tIes93fvI+m~Zrfw`R3F
zicg{g8Zp-a(sw_j2^7>GDOj}6#e)KYm8!ANj>tpAascIL`#KD&98)`Y@d>|uZUlBa
z7mRR&2`xD@-^%Q6v2g>SGs8M5uPuK&lU;I>I3!P)(=ZPDNN<qVB+=2>(r53%M=<W9
z<3jRzT7;#O^zdUlHQn@N+CB4@1R;fQ!{C$tvH7Ywxo}*EIy^<SP-1Q({M@7z_m7K(
z95=MP4>L5=;T1^dia0`0l&0G4As}&t6?erf;iCyjV`LQMn3RF;eL*}bJjSIrPz{LE
z^@%}8f#`FZ_cM+Om2Q&>)vA}kOO@g4*lfyZHuMxKPK5G=Yj;V;#9xX-d{d4(l9jfo
zHD5YzeejQ-62|1lR9oWPYpZ14U+MQ&o(~tVpm}Z#-#T(*D>eCShzxysF2hll?4w!u
zq&<TOE7gQ*?V&+lvY>sZJTD|X&OZSTBX|sQtJ;zfT<-~83Qe95V8b(2I>hIOThmCb
zG`CV2mQbW#+o9o_3@#LMDteyb3~j-2nG;!omSLLwPAVD9U^FfBuS|{D&yu!K2?w5m
z4M6+&?J+(cmZK0^`OzOGr;I-HjU<@#)U^KkY@PZUnRL2gBslXdqv^n`4-1=V*YYt(
zPGF@1*KN=wH2H0~<nc3-)A-Ed{%ewXByqP!hZNmd>ACS@zFvGl*Ml!>o$RLnP_bBI
zFXED_U`FPVHWNg65hu-OvaxqoCZVs#UpJg4nK)pg8|+f0!TAf3tejoRT{ts)JfWA}
z2V^mGpj|0KxOs18(=qSWXP?A#T<@k5r4@I5=WDaxF2~0El-n`QwOp0%Axrc_uEG?S
z@b6zM`m>0q+Op@UUp<isO32S4w(KF(7&@t+7zs31g+D}HB3Y^=wLF#YvIN>QS5eQ7
zEvBwY^*`yL2h8N<D?^Ld)~;R|vURtim1+fy^>Dx$+h$J2)y4!5B`hvtSe*PwU;J!H
zNyhRUjwp>UxEJY~{qWBe-h>Wsl(C!!-7sV!H{+Z;lV5{p93n|5fO{=(1IU0AksO{^
z`MBs;7ovIj7Qd0T$*sr5hI?zYz5MZxDg348MQP#Q<tgbup0l2{XJC$A{VHGBbvO>~
z)G@}6nLvp&<7Twnkj1_jbI_u~8<vj$QC*`UxETN5V92N`+Gz7epK;v)gbWYX&zo=f
z{7$ws+9qXePjFo?ewE3LYRoKk4e_$|!39(S=p^2Hp%j=^pN1=Coq-o?3>g(76mY)c
z4F)F30-*~ZwLFf=)ff)HJnu9)><e6E?}@I&^0lN2N&Vm51vxpGZi#JyN0MBXg;M#*
z1vLDujI538?pNrK-xl+@{Lw^cMC3{MLjNr9i?#E_^5-)WayBoh)lFq~qZVjo%Qt;y
z&PUZ;xZ3C{@PxQ-Ks&OpvV1r+Lz}}xswGcCqB7!12VI<q65pASweYyv4e0n-fZ5Q^
zD+dc#8Q+wPK#zI-clR33hsC!K$s1}l3_4eciOFN?R-4I4XB;nQ!{BwctA)!5YiqTz
zI@7m@Max71+)`bpLj5aaIV=3r&OBNRZrQAqIxrNoD5(-BJT&(N>>LC!?ycY2t!AZ8
zEa%V+a*h&yqq|tZfLK*Y3@s38D^ShUOz_s7vhbD{aPoaIeU+h~mFUXssXV@dUztuk
zYbDgkZfe=gv8P!JD3r?ZJdx$mZLO8&KuY|?pa+UHbrYpmam&AH-_3K<iQJ)Yk;701
z*|IG6+{s?3wNydI+!f-PWxK@ss5e9c!|ef&>%qq5rZ1i=-`&5v)>%xu0n_bJ?C{F1
zjz!<g`Zs2W?!28if=tQ%2_3qApl4s*Zl42?XeZ95k#ugnpZ%HtBj^<+amaCQ1%hj|
zjg-PxQcu!(=^U5c&#$y|Cpd=atWo}6tzDQuvY>dTvaIlFDNDy!d!>~TC_!~D)mSP!
zjLYbwHom68o}`~*`?HrtV@mORxNJd`%q@Mgz-lF{6zFQfU3F?b^YbvwviUJGbWO1K
zN*9G&`Fn|Gl}exLgZ4uL%KaS@UG;FccGD*GPx=mBqL5zhyb~XHzEfUocqLQg$`}8t
z@FvCLae2a|(<z7&q~g-tut58mmitbV2#U@Kv`ss@jlB+hJUZJ&s82mHBn!myTeRh2
zpMyDoC)Y=C{+<UfdLaU@%FVUt)Gg+(6=hawiKMqG@_=r#AZHza$i$43tHMO%#|F}%
z>u-a^BtAfmAc%Sjndr)fQA>Wkf>FJm`!pk}{B(-Rt;fMwUS^vR?KWPO{e*U!$mVof
zZRnFJO+aNbE}c|KiAI`uU+#~zSDwu-@=mxb*U$w(L2G`dmAi4NP2`Ff2prmI>Sr9&
zyvsEWVp|fft54lr>8wDy=;Em$y*cZIsez(FcbkQd8^{~+7IFjYom_d*zgIv4(-FTO
zvylEdaX(=+Bkf$G<&hpcPC#f*#S#4n(%9`p1>U#7)sfJf1kh88j5$lh?>pw=dL3jg
zc^D@k(Y&!D|5%$~rlagJtPp-CStJ#(tgK4sCzoWKsgkzdL4DS)RPZq;zF25oy*cxH
zILJh9ig3Mj>$v1Bo$`neCrzDjIv0S(iN*J4c5iYpF2wVdPL#e~3qHvbJrkKXwh5sN
zU`_Qo2_dQ=V{$M0wUMra%VJKY&g{cs<SM^@7ZPW<aGygt^ymr1Az$@Ej<zJR58LY<
zw?WN4kTgY9kIW*xh{?_5(t_aRa>o153eyR7P8S&4DfXH3+(*(DXbp+9i9g{8HOMzK
zgS$yZ-7#adSK?J(xuhiTRm|7Jd@n}>bSTg*(ovuk-0k4mL#gP-gi(DE8O2af*=SMf
z(qwG=y9PRnpSOR#l@>QAgb8}UimQ?WC4uT$o38JFwmSZ(wT;iT8G3BOXdHcS5bY!h
z6TwTj+sN4>dt9O1;Z!v*VqI0$_In2Sx~g`poM>N|<(6C&#2LNZn|XG4gFT|_FkK!f
z%0=vNkU#(E9k5?PcTt>!{4m**rRU%@2#!Pjf^%gmfcMG$h@DEpMsgHrjV^5GoU3ly
zp2pEg&K&;&e4r#O-w<lySrltFGNS>=Z&&iF9EVM{es{yf`GspGQI&5irGanidHonI
zLCV6Arqc}oA`X7=dYGR$^~7K<XW=-ffM2KCsPt}n!pMhHwsbA>0f!u@KHeGJB?7mG
ziO}HA@JCe?63i+X|L96DTLo>mEXj$Oi)vrzoYiL?VVI+^(F4|quKdi6cvawSCp&?c
z2ck0N;^ZubdPwZgV7=(0<y&hefLX0w$A$kG%a7M`>$@1e8N@Qvz6_`$B4#Jxi(0Vx
zh6)+3gNtFdy&&BCw?>r8k?9r8P}>hG8BQLYrDxAa>C{y3EY*_>JjSj1x9x!WxoRRC
z{cdLEt#{<h)s*I;^hKit+VkAXaz3uwCa-*51?oDLNhoX;d=@vlWF9Fyq%VdkE4fms
z%W_Q_D*6g^KT`CNG<|p{$K7B}t$=6R<SPG<YjGz)T$IyIXR9tE2d3&*lZ5qGrMJII
zZ?(oG?d=}iC@Nt}hN8KLg#9R<S%b{AD+emWnyUTt&Y)xn5kxKCEl%LE3}N(MUn%%x
zE>IH@9C?C`=vC@k$tmK=<RlVC;uCl#!ad0b>Z5Y|(c~84big9y;1LM|Qdi0{ME-#!
z?C{#8vW<+5ika*L-S@B`LDWLfO$fX3tWR&q5Td!%-uk1N4KO6lHqL~Vp{#4?F>O15
z&KI6eKsK!f>3^UZ-&}%>Yj|FDo)|5U@i-hi`1Fo{QqZnL2FO-%s0|5e7*rUS%5KAj
zf7=<=IQl^BAgYK&3zKVfiR)~j{hU@V@$R)EKtc)gj=_V$8Zg=P#~;IHZvjwPLoEq{
zRc8Xo4Q+P8(F&cp9W)Gv!4L=;V5_#|OYLN5&O&^!>K+pYeK89-fq-6K9%HBya%?xA
zyjI1uY9iJqA}})VmR?vtDWDbHO$`8+uxnxhTfud4am%0G>FO25ofCb)-{uH(CO~ZN
z07tUx89Iv(7(CC_sIS0H;fktMnRaJTUK0mLhur`m1py$OPTCPxFI9xchy>Lb#VDzG
zz+aG{wC6R%rfFW|Y+}A$)RmYIPo1$37*CZCnw}bkehk4rr#uY20s3gHL<mx?$Gd-p
zud(yVcO#*D^pRF}h=ci!CakodgUSO02*FFM(sD3mW7zBK5Tj!eX3){Jrc+U0*r52R
z6UkBlWCO9{01k_w2@CW7_3Dfd^gBCf+-3~es3j(0`LmCymJu&qG6lpTTNq7<{vF4%
zcZ3^xkcwRb(pB}d52FA5xd2ar_~8X~Xbwe5{{>8LA&3T$B3^(E%My81lpfgmJeRdU
z9fVm1fbAzZ(*V30TitST@2OAm;?BqdaoBwQ9%bzbG}`=At}Y>79f62LKa!RS3|;sg
zvi~fMj5iR%2vmYpij<K*Te2SG2UXY<@u3LGZahlV@;JGVOfqZYfSk}CZ05^mx_<9g
znKYhRb$@g}^p3GoTbk#8=uTawXQRKkx4!5a#4Tl{b_m#+<Qv60^>eW0O3JTk;-BKU
z0ZmXCXTmykf#masF7+-n%}0PJlA&I+e9^_HeC_*PEI)ah5(_a}1F>~`;jwrHw)kD7
z;Rr9#UkMF1nw?-F-*T=|sZ{@GFf;x8;rr)LDrJJS6M9=8;?0C25A1z59=;9`(Ppo3
zYd~g{3D3ba<U@MKf0O{g$@vCo>*C)QY~S7aJDvgM&|L}d2b@_JJOdup0Pyuwf9@XU
z3%<||AFx6M73<~Ji)OoDwEIYIYi{Mk*#H5KGjOGUw^6k9$IzoL=4iMr>#Y&*FCIO_
zY!LUpMg0));vyz)$`!~XLirt;p1lg}*Lk3EtV!4{@3L0*+EY-^AobKeFiCj<#aHLk
z26(wjgr|WJ2N<b>M^_B`#NHvnz!`}hOp$LV*$ze8n!AW`em!!?0NBJ2Y7mUxEAApm
zAOZvfEiv+DP^4YODgi6K+G3+V{wY4(Q$N_dB}aH)(tjsEu-C@G+(dAC7@iC@TrTix
zLX+U*#}mR(Oo-<L#e8o;qA(<x(rn}0`{@N0G=+d*`5ffH7qZ`z3L(qqMe%|$!p0w9
z?fw9QBRb~zryj^m>_XTuH3tLyg#rM7!kD;ONNSGZ2X)z<RSCn4qKxhJL)(8Yjz}-L
z>{@xjyT6-)4Ui!SR^ukAdBJ4_<cpnffM*@?vhu$BOgF)?z5R!wFTnP4T&eXzW_9w+
zf3pt@fd>)lzo)|(VSoo^Z+c0BZpi+<j?N^3J?+ZbHK2f@tP#n0jmGDEZq-J8sm|~G
z*TV<(SY38MHk}Vfa~`$>x!#Zh({K!a&`BrE{c!9E(ouN*OQy~eA!;D|^A{n&r>ip`
z=7FpA{pn>{TY1dmh>+|Av;`NgVGO{wa`?iQ-ssK!=phT65dy^rnDZT_<v(qZh~?jK
zP&!7uq&^A^>})E9ApyKvOErc4Wmd&&6GKQ0;3M|Z4dFuFBBg~BjzD+-&au(2(!_1;
zMY=q!70Y6O85f>e(7+=sf8ztiu9HibAm4%(_Dnw4xW_OIgx<X=zLRQoFq-HN1HDC(
zi*m*IB=PQbS%J3MG6noGA9TzWr<?Y;B=ixccm;~Lc2Bca09pDaH|Vjvpk+}6F*7KD
z#SS*coAemy3+~aeej+fwfE#VHpry*(H|n>*<7q0=%OgRN5#PuOvyzP`8oi0$eGo$O
zT>XLsU`_NH`7t&ejVut~+t#CnaKIss^&oV8r>1de(CjQp5rx)?45$_m(<t++eJ}MQ
zu>6iZD!#<II4o;S*B#GTPCd@Y2X{$4&_>iD7-V3ptXI1q2iu~wKeU?>I503U+X2p_
z28Qw-;U4`35C7UuF9bn5W>9B2jNy=X+48A`!XaPyD9OXCx|b3FTZj?2crxKSa1}Gi
z#FEU35dH$v;siyLtFWf|VNLOwOn$P^;{CG~1R^3ZI;{b3hA-&Yg!EIy%V#k3V+faA
z#Jc_T2Wr_pt{hzV=_ucgRe`p1D&<}GiO46<p&A=sy6EG}GA)6}Tzr5mG+MKj+4mMN
zk~9}4@4;dKS?^kb;vS*q7Z2{oVYr_x5i8+;MVlbPY19gEP8}evCw%JZ`V@E?LoYJM
z(Z9gDffjt4oC9R93Nr6>KVWk!<-<jKa1Ov^sOs}yQ_B~8!4J~56KDy=hNC1r7D@s5
z=rZY1ya<NV$XTFoM!{APcB9oh!K@<$sB}h`M$9RH3_e)ej*YkeYS^>hgG2M0&_;(!
zo%hvi!LI9Gu#-y3ST59i;Q1PYI{LMdbK~0kX5yXWg$I`iZh}o}eVhbok}!UwD^npM
zRw#JG&WDmB%9T)ruYz#Nr6yH&O%TAN#scCJv`E65s&HOH_mw$Lq8}lAp=3ArWfM*U
z$JzU1ihvHr0$<{@(650)$=~b-7gi#`dFArgmhq{nO|}o7>Aw>E4zF(Cd$lMF0}3xN
z`Ly&-tUU$fpd@~zsQAH`pWD}Awi8SWYZ4Md2IO&8eG#h$Ve;Yp(;h7J?@xveoxDMS
zIMqqLAJFurTKQIe*1Fxb<vef;N0#5y$S06gT-7DZkA{5uM4tVY*Pj?Exzh6K*2iH>
z$|DaK7Ad4=d`tBf836gxE!<u#O!V76rg}Sj+VR2HsSfB}&Vk0lmA5p{FIGADS6HuD
z^gqlXUnPWP(I<J<a?+3eI<<9OyTtWwW>DhSBm;z8I}x9ET&7MRsM-?H3jp^cq3Mek
z1pffgEQXhMbum?m(Sp6SHg{Il4tN0}HBp;Uy_|Yqeqa3QjoEpe9=LHmWm3B+vqW|3
zNkwVYxmVy$8vrh6KA57-aHUM>n{=+ONN9=V05{uFr0UJJ)Ry3HMfA8+Xl6|e-4jSa
zMasi(FXQKedTHl?;B(-_h)AaeAMz_%-d<*<Uws>SnfPuDGTIW^srk~BdHNa8cYju1
z5h3NZC(EE3m&&Q|%zh2rJ;gX$g|(v4TYCvMJ0pne`rghVhDH@5umncH;nW2q%MCf*
zm!>ouG(HRNsT-hi<<K=}$h{b`R}-L_FXIpd_SZH{kYm2gIQWr`{;0>MmN>o-#iuHd
zLa2)GE~MACMltL?_bmd~??%xjxz?vRsv%a4InbztK&*gd(b?*Yh3vl5B$9+~YhqN3
zU{vjZQ8;nT67+6E${pCL5KlZ#6R4rN&J(!O2Q%$wRCkd~^qFB%BVrEs)ay;2Ab!wI
zTlLuhAgdFh%S0loI$g={Jl|p7BSBG%0gp#IO$Z!5iD)}y8bE-1$mu;5=z8iwc8fSY
znf-3)EJYI_3T!y0!zhGA1gQnA`0F;cy;b?eGkSbhf9=@cJSO4<k`V})D@bX)5&P>-
zNB;yK0r6cz>{WE1Vk-77cW`B^B(7;KU%?$X!1)F70sy@wjID8dt&$S<<rFb%Msgkt
z%6;qC&740{C85h!yIDUlzKC9aMGvK@tV@bcabAUN*N5pRhMpV+UPa{qhbC;Moeove
zCfvQ1iP1F}2-^h^XTq3{URQ<oM5!4&IpkL_eb)d~K-HD7?q%NaMKfR-+f&XD7#$J&
z`mp-32Xua_qg`b2p%3gx4WIo2Q6*;|NCFnc&INjk3wK5Z67Mp=huM1R^T}WX*tKf$
z;`@+-Fg*f{XuO-|(xuX)$o6)C1<wtDA1^^R74INqSF|;+gn2A&*B*pvzy~LHQ%H_X
zw@e$1u_Sxc6?=cskQxGxL8(aDSdNEhhNcN-WEd^SHl64N6Fp>5q(|@HD;jOVf=f{W
zt*!^Jj<wqXi3!A9iSK<9zJDVTu;gI))kipD&k4rn_ELoiL%at}l5nisk|jio4w>bj
z1HRCuU_pk_jC#aR2I>kB$aKZz?`4zp9{!-E?}P4D+quTU8PxdzEWg$jK}FmW;$wX?
z`<B)3hQ@LNmc0V483lTXQW&Nt@Q~la3AJa`2)tEbR*wad<O9U#XRYqH_b^zD28vqo
zCCmsOfJ#^6ZIR0jI3x|9qt2qzh@r1s<>B-cK8w9Yj$!|CDo~E$Szv?8O2J?u3~O%`
zsC62Di}Vwi0u@}Yi25GrmmWp2xAeM(IH0!lW-603ozyq%{BsbhE>RXmxF^AJQH5KH
zE`c?Te)mzBjgfMuH%F7M>14>X350?4CLB8&Zb|HOP|F}rotrQbVlO`f68{b_9tgEv
zAm+K0_EqfNzSXI}7Gdt~+5rP;U4jh$zl-h<nRI<#)j%{y5ivs^tho|(z%}~O@_LxL
zaSfnW#($+~7@Q6YUfK)DYbZ%tb?<kT2NI%vGnkq18(BuLxP3it_Od^?doORmic#cS
zgRv@f!|!C&!|Lrw0YULagoeXshwzPNNymX$wfiW#N|+%9kA=bO$N=Bqhg0-_;12#>
z!psJXa%1k3zz03ih2#H99?-Y&!2<8pkivO(0xzY#3?>2vVZ02YBlN%{-i-qrZ0Ezt
z$UC{GA#5<x=Re2YK70yrL}d+hkb)_&jwYY2kR-^0E_&}h*2D42fDKdx?5A5Vmib5E
z$=AuJe9+@EphfBcJ+~c*zW@6D!f*KxB02F$-3W~PP1<ZYJ(hR?FU=_ioN3PCgP_L(
z(V$e=Z3SiM#-R3tJ3DVDvW1MxIENHMOU8f0WiZ?<#(-)d;x>F<4^W2^hLZbi#qP<&
zVps?e5eAZ#dgSG+SiK!1Luh04r|&Qj>%#^H<K0SxKAnFd^YY${MYt|!)ZzG*r>8T@
z|C8fsp%Q8<;A+zc_9IS1W<Nv5V)@{`2oG8U4YPJ!k+lNC@T6nYNpz4tAn^xZ+!#Yr
zw->?eJ9Y)IthWdd>JuEP%z<1AmX2^j^vcop8Dg{!0wcKpS~7E?@IJPD`R^-0>+0j6
zlt2iM@gda_L_m}?r1WUN4_D3x|Hs4Mz*3h+X?u?z1D_B38)URz*C6jij-?SOlC1n+
z^P=T1KtUi3zBO0&yN@7KL9aprj_@JjdVVN-)7(G-)%Z!6gG}Yg4u=$mR)o>1&C+1K
zcGr_w+aK)x-vX0LvAhdaMu%tSKPtw9l(|p(A=RRio)+caKTu%2X3UGUf}oPib^aVr
zLLBGtiVH2T0*_Hv|MXdL<ler4$3V!DNNNuJT5V2u?iLjcWvEr=DfD9(J#~XV@HslP
z$A|+WI|rUeh^rM%+ihr4PmUy=g){*1RG)*9q1U!>i^Uc@Zck9`uM;EuWW=k?m8R{m
z`#l$sW*q=WTy4@D))U>OZGUrbdfOdN(bH#@EpYN_%F-2UtMP*(on$$&*!Z}_PQ5ZH
ziW>a)&zro?C=kV5;!-|AX3E@3_fZxy)3OP6QPdg8Kxe%RpW6NN(f3hM^r}5QNlf09
zT*5e-SPxY{;(Ep{I+n;zwO<=D-~;+<4Vo?M^No^UNr)$@l0G2!g9Hu5*TesyN(dEo
z;Z)^_!?N!}W=i-jr^=iOASED@2>qyH`y0!&fx`0=ZerF)&E`x1a-K}Q$Uq!~MaeE4
zKg6J#vLC~o@YW<A)Q&_fBkW5$>-GTHR4`Gl7($hxA3w3lt_%ncs?U*9;QS&qmC6T`
zCsVNe;h%tqTt_hY)Oz8gP~8gKKez7)eB<|p5Dxun#Ar~pL^hpn5Vev8f@-bQtoF|s
z#vk7`%^P0#Pp}cD$i;6K=fN{8?oaG*QbMe#k?N>2v?*5|FqGY2F`TVBR`OtSYDnw6
zbj}#}`YComwW09;44XF}?vr+7UVyz}((L1TsIi-KAk%lB&C4-{VCmf4_<0SEqQ)8E
z_!YxZ`V0b2-3Y{Mr9Jq>FD#-Bop*%`{jd*f@g9Z=tZTMTu=dS49D`~^%xTvpZpwDZ
zV<@9T`WnY!yvP0PQISzI;2>WLlT8QVL~Ac+>2y%=7+W}1e-A<IetsB3evsTy1N7T=
z&Tw$XgzfUx9=gZbrtj5>)WELSwF<`L6oAeC!elAZAE4UQPAw%k+MuxZ1fm+6I*@Vy
zLMQ}u3%;A-#c+0==g4)sG0EaVa22Kl=@+@oJWN?2=bSk{gz$FYax7c)nUct^@|)@2
zWA@k);m4Bp6kM$vZZ?OWfC!L4Js#;uL)XXwkpL&VRPrHd2-t$zBLw&Ll_?3kR-Z-h
zkrGH-URIHAB3+cu@C{Nf!*?m{^!88!LHOdfV97Z5_kDs{K*J_*70SS`s8d61BBsT%
zsf!AbxU<Vae=huvdf8x#csWQQ!X=LY9)cjR-xH6nx}!=F(dp)_qgnI#u<{4@q2{WY
z?8JO1OWGaxE$XtX+>xOd9QCTAPqR>OcdRXr+J@wUf=Jxtg4^4h%j3-=0F<-QpBA=_
z8G?D&8drvk|C-YU&lf$5<EW<t^MOzI6+Hh29hD<5&*NP2eBe%sZ4W=%LT;)PU+lp6
z{r<BgcElFgwNKq+aZtjkQuPzCsH%XdoJ_B|xkJ$}*9afV9F-lC|H}YE{EypT%1ora
z^%%T)M;MUH&x#(y)h05@53X*NBl4H$W`QfEU<B(97Wdgr$3%c!G{gDifVT2k#Tn$=
zBYrM97M~keFE{)KYwtK}SK91iZSwug71YNG-s!+S%vUrtddK~cjs87~BfrW<#@X$#
zUiYtha6IC1&;~3%wX!(f^g76?-u%2&Y@uxXeiVv%9hoGL9LW6pk#3cG!gB>Y<>pM&
zi^?9Hu(CTOA~(gS5GdQf0^KRDANVA2IPw~UZC(LEFu5Bl7zz2TboTwXjlGSUjm)O{
z_+jYo(jxab9E-6vY^8H$UcPCkZw>G>7GCw3NO}Z{z66q9VpT<tW=*z4q{{5Zj5-j8
zhG~0V3s|aCXM7K>w9Cu9cSA^|Ylknil^cF3`}wx+8ll^a9&Vqg2*%ayxG_b@&)vMz
z-W2(n<&of-85*afCe8u8KA+;fP=uIQyMs{K;=N2mT%raLd)CrcTwuSoYwo((74=nX
zl8#@eHqDAEiLdmW79JYzh!Hw@Kd~CI7g}~($y4(kTn_%8^DL@p+m!NB2+XY63HR7k
z)WSL0HaQt0#75`bK)uT|;DeFWC%g3;!I{Yjn_3wb{!keJo-akfCY#Dz5UKDuFx02I
zZ4VTu5@oxiwjx@ho{Ouo9(^rsa&H|0d}vGmG}L&15{v?Mf{Dfg!AwTnpw;zb*k7pf
zwH{`~34j*+Xjb+6c4Ttee5gh(A-U(lzo6u9qjkwvHiwpMqWBoFU@!hIZau>37U4ms
zZu^CmM{$g`!22-=K|FZmHS2E6)=fT>e)p?Ewj6(IzdM5?fK3nHVIT*L#A(KU+9E`o
z53g26udUVnIZw9gJHd68L!?#{FD9|)O>uQ5q4^*}wpvS@h|)A%)*5T&RDE)d!3Q^b
z#h|t2NVS}y>;1nEKF*B8YKfi^B00}NiPy_;v`Lt>!dBoH__H#8WLX7=UG6e9f6<5!
z8v%&Lo+Uxhuy6Ooly6OhXRf_v>>HkPA3tzMsX%x)WXs_oj_z6UIo}yp1U`=?PJSzi
z+}#O$tB;f3&L408QY&!16QviprfGF+G2-rfZ`KvP-}B525l`z8`LHy29N0t0L*&-d
z=;afIOWVW4PGs%OjF7=9WTrXuWTWB{<A}+2`6q4%%0Brtk0&?v{cbdLPPI?56FVY0
z5w&mp=4c|sWn1U>doU+26;72))AuZq@iG5ie_aplwij6o=^M%$Iv9tnpKG<@wLY<m
z32G`LFsDb1@;d&ikuKsr3Io{^3_fR`VWxJjpVIieNMK^x6Kn_qvEPm}0oh(WUGs&d
z;V*Oxuh1s03bk98RJ}Krk`AmM;b>(ywif}ny?l-z!4A+2rcy=x?f{#%jNh~=ekA>8
zy(3Q|kmyj|pWll)<oRo7rsr7`+UNuqr4o9Y)GNlnw#i-P9|SIq>lG)XYVb_VltxBw
z1aFpp()1AdaE<1mS32KQA~Z6w+4>8eI`g>SxkIik?<}9LMfFGXt<EHB)Lq^18S@3j
z90loWaDHws`|sm~Bftdu{qXxX$L0fED6ev5hSm%D83Q3nIlMG0Ndqk<Xc5N`R+t*}
zzZpuO_^`U&JY}tAq?%h{dH;hZ6pv-uwa*bxP%2foSI0zSsEB9XD+6Ra8Y$H2SCSWf
z-nuGsnJgbN8(v79`T0@A>A=LXNoT6isq^6@p1<@`i)XJ+&r|CbJgeJqR`?6d0)ASI
zQsca6)IdyI2gGU`VOw2J$~hQn<zoFj-&E^DpIs&Av++Zb?a_kUD_7#bKeL`r9?h)w
zsV>}1hNHrfu|Y7D?<fMdCy(&)Qe5zt@%>l_6W7Q`kDI+q&@?&dR51}3meuXB5~WP?
zKoW+G=Y5t(tR@n)b$|an%Tn`<H_JHm2DIhk-QBLZtZDb1STx~uqR{t9Ef!p<T_^~p
z+xZGDSKiGdrOgN0W|Pda2+Og-fBx8aWJrOL1SkC*WP}=72|(lc4NlKDnhwwWvU^t1
z1TC}(w=+a7z+PJ1fX-Xe0fC{0*|kF9W3knt1K&I$;Qn}}jPUvhGA*+JuwniYB}xa*
zpiO=~Y<l&?=R4E$iEeqrjLCEzZPNYQ1yb-M7PY>l6A3JtbJS;lfu>Ka0T)2PqZpye
z&(7*}AmRFhN5MxZ#HYFhj1PAtf)T?U{<Z!WK=`@#?UdKrhakF8avo=q8D4WpA9eUP
z!LtpVVe@2zr~Of~AJrZICd2?mZNXPul`Fz{+CcLjSTP^fb-$GYOn!c<I|-cW@~D*9
zz=%43MMp3iz?y!TzLC_H;KpH6e!(Mgff?wr;$)IuEAR-HpV-a9h&h{t&mD<dh&Si0
zR<e+{)O2DHu=q)X1^2Qi2;{Hj!BFR<qmnv*`a--)!i39we;Mdkm!D)hCJa~IP0*cu
z^7NVWf)M9p@9{{*B;g}uhBME;Lfrv=gIm_N-SLQV>lBw~3cx7|Rt>u4-2hjJf19zn
z$|L_2>kJ7<cGYE-gdo<T4Zj@^W0l^*&JHq*)|;Rc59TG>557>u-n|Fyh#(prytu^D
zm_XEP`>42wONWBQvq4g-@%%ejq5<V4_0|`{sfA>e?4k|{#pN6RGwoL@Gpl^T{pRV+
zCnysnCbZ*AYg0Xj7|z$tzZgiGXIc5qbYfyg(N`tz!PGyy3H@3fuB~()FMt88QIyi5
z()OSDtxu04!!N)b#Dn~oJH@&q3_C;2<Q)vOov2(zm%x%na3W3M4v}t61Fo|43YZ79
zo5(HR9Da-NvEh;~zYM&B{dycj?}*Z?rM%-V=@^?X6-)7{?sgrLYTmzpA2CsHuZ#Nn
zyG0_Dru4~?JXYDRW%3G*yI&N~aOTkc;4m72JThjG^pi{5nWxWcU8#5yUjj3`eaL=|
zQQ<R$^jNSEKKIp2IGR7N6e~5-GjuuC6KE;MR!SznSJo>GDoMwN)7{*Dl!xdVV_q2x
z-Aj_XT}^RMbuoUBCm~g)q$axih|QsO5Us>}oS$3u-YC}t>Pfw_(jbGHESCZ9;l?X)
zkBL>Db$n7rQ1k7`WCF0hrjy{P^c9y#@0A|NHZr3yjq7?|PCT*K!QK1zs8euS(Q)>B
zVCe3m>{ua&)C9BB{p$mua0M}j8RoggU)*MAdj=s!;d4Cl>-%}>&L{h00m3;Lz;fyD
z0Y(0qehyN9Gbe_Y{bt_oi^hjG6eThD=RNVoUYTl!Xl1>RXiNR~BVce%Bi6K;-2czW
za7~kfb{XF3+bXRWa<OF;GI-jhru-A#C(u{4gAAX@$j0FA-Me*Nh*d#4-sXyM7v+-U
z>mzif?9}x%qFt0SLDYB0$d8d_X+5DYrKh+=G_ky#Bzkty_4Cj|aq`gFl0VU=9@bVJ
zN$bg9=On$Y+ILaty?rwPm4QRy|9+ES@J*&k{`Z@_07Rw#id@WY<8^=E?G6dr9eDK5
z+S$XmF0BdqpQRrKNlu#P|1SL}UJS;@``G`iP(2$SN;VZ%X#WrGprFs;HFf@^PWSoO
z*#b*tO?V+v4N-dnw!QymS_OY?Uig1rDPAhPQs4i5rS<Oc*9PqFF8{6EcRoala6CyW
z+NUMa{v^5#V{UHV|4{RfXhA_iY*f@K4t92>0l&b~jHIN~W3J>?O|##=K@!O;2n3i<
zG)0H}DN*X`>giL};2iVjQ^1>~z-#6Hy}kONsk|fh2RdDEhxtSMI{m&FYy)nR2|nMG
z;_TaT$>j+7q#F$J+FLunBrS9E(p-?MxAjj?PcMxvEGSS1k4$!C#%6bScj^2i+F(V5
zwh{H{5gRc5h&p6$GG8Zw5;p@z)X5J6g}!t?8h!?sme$~NM^8_$8C(`#xtN|-?#D-X
zgdBW)j}Z47ht$udOpa7-+5MA-|2ai?;joH_10Vfo!7(BTRq3LGOa6cU84GH=&cFu5
zcKly)DG{Gimo?Eg(fn9G_Th0)AON-k3UJW-D5+!#G8arFBh2ar>D>8T6fzk}>@Y#u
zR*XamN?a$y?+zI`xz!8$x%1CaFrB34GK`${=c$_~u<j@PzrCy*t-1gIyDNoZl@KDv
zIRpO>Ihw7I!dpVK%NZIZ9Gskjf1F)hNZOZyEA2d5k9XbPp7i%1biOj-%@E#BIrfTa
z>04o)ii=B0*l95xAjc#u`9m+SrxY^Ib;XUfMW=q#Mmnoko#-39gz9=5e)OPojT-UW
zRKFh@dLdOzU8t++R4mbm`{l(?7jy7~e)bTsG&|!R+tz<}nFG?m<kA-}UVP}3v0M%C
zgM_m?M>Ey^8{-HJZ!rSe0wYp`8O%%})8VOLMnKG?M^&w!CmsEzw|~h2JW1;k0R^f$
zJ3Z2i$TuvK1vHJzD{V4!Ys;qsJkKZKW{CpglX&Pl%{9Ds7KUe+>5dmvRvN=LWCgvH
z>N)z%2kTlV@!my!quu1>Sd+Gn@_myokG6P(-7ek$_nv0Rc*cRVJvl0Tx3#tPEXBkI
zS3i7WW_EVdYVS1P^RDjl5kIa}zm2?^lifcO6g4y+Hp}s?{%9=+Ad>L5k>bGN)&Pdi
zo*rhPW<J~e8t`zbPn)mk#Vf>DY^$ZvK_mJnr>$x3)zSV#Wp9h}^YdR-2c*e!m=oUM
zy~pxu*#_+Y#({v)#z}k38|b@TB6j6rBE1ip?1pdDm|N?8kh8I_(R#vtsxx#R%z-hh
zy^`wvM_j<c@o~MT(;Ar9*G*GQ0rq3Ty8x;bHeglt7@ZGp=PO9K^K<h3V*$BmFTW7^
za@GCWt^yZFwsaX}oX-0T+aPpo6W=D0@`wL&Y4Ux%x#M|_)#*}NTc`97#|tYasbSt~
zK!u*JccMXoF$?tV1!Tq$8>fC#!Ezj{@$na})d!}c-uKC9so<s_KnFj5G1sC+jYd_0
zOLD-u#@xxITh7L5NUNCpq#;ZtfsxxMCT$8UXE{9~AyRujwYo4f?+HUf%8Uxj54<QV
zYeuYL`5wq+)`&sUEPOr?I0q5X_|O5T8+$LxvPat9q-6%yP=Yx_?%|?c=(1t6$iN|q
zV$)0!DnI@0#L<mGoTyum+)J}phm2~x%GTe(lsh(%+Lh0!_Y!Oz4BY-wSknhX<00H!
zLtU5*S|+;lRd_1uMvuApL)k^jTaZD^chC;^XrEwkc>grkq_ASRsJxcSvS+w5&BBOL
z>(ZrfXJ%jm<X~SUA2QiQsw&GrP4qW-L@k^Rq@JF^2t_Kh;nza?^#0ns+aRc6$62dS
zJl`3ziZrV`!fi*hrP-$~I`6Uiy6as8*xv#YJ#A@yZEXeu0s^aK5=n)jh+u!R(aWbk
zJyE^X@Z(Mn4FBzyoYcB{)#;~q0|5nd_3eJcM%}c5h;OvsPh+s1yU&Izcd_lhLV2wJ
z8cU6D&rD?3z((n#CgtH#M?|XirDS_nJjX!G=?_z|9L9+t1#G5lv~p*_^)v<<jx*?&
z^zI~x@9Gw;f5c4~1~%Ey^rTp+1GpI#5)vX`Y@K)jb}ONlrDbj|#Dxm(tF@Pf_<LLd
zcv68t+?bX2fJKe0@))vSDxM+x&qT4X5m9#bv0C-3-b)vRXB=?Dc(U%$Rf_v2$YvvS
zgmF8AVgoU$sIYVel<=gTCyg#%4gUx0v!y1}CAVW^W1S%cRFpG%xL(%S?3za$GCfvc
zw)gw6kf$`)X{h;4F@y6)RNQzuJn(e-HCZL48=nGLfAe-l7XHvaocx8oOP)qmm(yq9
zCTz##Q^ft3L}R#B@k<T(uXNyE;c5cJ4ezoh%gTD&30<dOg18*tmk|EppbfeFdic*X
zGbME59Z84MY1Pe7vK%2dV#PVgIFa?rm;z@r(RHT!yf1Cp1qW%3Zg+X2a5xG}Jz1%W
z>HecIQ^;w7e0qA5wfJ;tky7NGSI3WsH9yob$Y~J7?<eHGQf-)&cK~qv$!8zLj{H#7
zzsT5t^+*I<_Y`quRD*9?2T_yvc4-FF%5ltxt`bb-d`?ES5#4Esa1D7+)#7xgO3wGm
z(?}Rp%x#?Rlb9)VN?}l|6Z4j*zm7*6{|!nYk<<@(=A~@!l!uk)YRV(UzcKxp#W84f
zS4=3OnLmWsNRK)2HJ2XL!|?=_^ZatNC>5x)coogpClQ~gifVlSvcv%Y6nW>p`YGUs
zz33Zo0RiQ(JH@qOzEAR#!7Gk(n8!DSgS=i@^b^msRjV_^yW@yTgO^?q)IAInnS5k_
zZh69s&E>~Co{V_d_-I#lT=beB%%xx8Nki0)@mD)HWv1M<P2%QOP7nLsO0Ge*7rf()
zP@_??8kqZdtUc0&mr%ompe(br2-lnRXJXRp?d`9(yg`$j^GY1QT?WJVEcxD~xo>r8
zsV>A$3cE6+XP(FuoduUb@r1<Vi?s38<@dxh>Ml5b^<5lNP)uk3%=X;;?eUwFYxniV
z3tVW`OG|VF9o`N20npxMbb+4k2!7P{!^W4LE{$HkCPQ!thb}ETpsRD}{<|L<hdP<u
zl`)ifjhFCs$U3oF2P<w}hi;|7LD}D1$cS{9_g&L0%SPEZ|2B+@L|Ao^!k)1l>&5rw
zB`#bg;d`*xz<U@lL{f5E{JE<2p`d2TBa;s1rKykgNZ*kqsiKS={5gdjxV1h`c$-xW
ze@{L)Mhi1a)!u}lfAQfOjNRPI?7>@nPOVPT?UvJGZhB-^C*i!?wpj98EyZK=>@%xm
z8qhv_v=UWk#63!Dq&uKhUBO-B5cZSgnZm0I=F2~>f9>jQvd@owA9L?p^81%>`+?TS
zO%=y<;~{F}@<$-GfAtgZw-^@Z8%+{(k$0l#$*jSgfVuR8p)Nxj-o&aZD(^1`0Yk9w
z!p-QCw)Lshju?`gS%T%o)zy*<L5ss@YTDVe)_z*%_`YuGtCBL7c=$ssZh)`8;`O#T
zkr&~0bD6@1b`I%2iR7=J;wy-W-LqLe8Ez<G4mK7S7iV%qzl9Z<YBL9I79S078xXsW
z1sxQr;qPmnxnyi?0c}TCOql+G^Mn%712F22y`Po41cPDo>%BdrGJK|?vQoSP#!FmV
z=wEM;6~!HUV?i`W)Xw5xP+3o!Btri9$#NF%^5U<?1m|ouiAX^PD*A{@tOnN@ejvO&
zaC!b@PIx%5(!=RMCE(SM3A!(7|48vw|L!RFARdoPmP1QRYxJiQ^i=e!<z9K8_yr~b
z=}ZmRyU$F5vc=nAMI0ZMq$;a*FOCl7@P3(Sx#AZ5v7<f@Ka&urFrB88!<FFM+SysY
z3OUAF)z~QOl4w)R^b@RubiH6c=5l9)4VcsV>VH3e`W!tlRYY6CRi4TvBB4a;T|hqZ
zjYB-$SH>YdYWzJ`@4W;=(ut?yx?tpvGNg4iPd#Z-NY;}p-+IDd#LAOk-B9wGhtF$u
z0Xs)3^^+0dOcD37$MH*`TOfL5-pvRUN}E!{q0qz`n!nH^W9e#4lw6|4Hx1sHZt<X2
zpQb33Dk%czt93A89h9*K8}}YZ=gnK2W+v|FFd9(}zpd|zE~+=x)gMr0Dh()WYe!Cp
z69PUH1syNyqqISj{dR?I#~*V`sI;>f>5oyJyWDShsFUXtOP>U}u51z`u$UXn$dK;t
zyyt@l4;CC(*SQzZ8ZX>u=VMiSJbEao()ciA*O4_SJ`$`6;avHC2+Y7s^LTO(9P;k}
zT!<iEBC>HB2|OXi-tP$R`%KWD5dBkHZhYI~MpSL#lP6IY-bKkBD}8>jEI<3SJ5`lv
zj(8noz3TVOBT7_tiVeI@^4e;N?-F+&S!`EB&<{^?HE-I)garB2_3qmVt8v};?-Mza
z&8sa%Ok~ygBR23eWxZ4CZtXHo{Ky#1y!?Ejnh*vF*W+xmSQlk{w3CDhR-|_(Q{VdC
zkB1*WGKpCqdi-kyP2&D2x*P1Bqfl*aZQY%n^y29TH*XfLzdd_ZnbGR|WR6C%AXD#b
zZmAbniH1Y^ZRO8w^%cyUvywxKkVBQ<dihgvQ@KWp5PeaUdCwVQuNI+?K!piXzq8Hg
z{7=b(nT66Zk0$Eie=!0jD9<taG&^4J{}eMih*{83#Od9CLFd;)eb6SvC?0!v8`S()
z?z~NRNfICUHUflDw_j8uyA|t2>~*%ed<vD|9Y6qMjPL3d300uS)plLsg#dOwSMNn@
zS6$;B-WP_SV4C5mRgJdoe@iG!c$^r*=6{MR<o}_w0m$J0_CRNkW3dM~a6p-!t)_J7
zzh9Fd3jIJMr<gqrKp5d^DW7xZn^tdkek>X0i+!p{+2i<PatSn&&M|<u%F!PJ&NtvE
z{<q%CWbgd+i5;pE4FL)%@HRoyn}$M}$RjMXXBmdZVh;uFfQ|l3Kzs<v;T1cz`;~yq
z2Ofoso2G|%{_+e7ENQB+_b!B7FYcd|l$6<`d+}m)&rkau0~k{n3Y|9eLvChw)tB_>
zMMXvPw(2%!xs-pNRi4K_*r`{$VL0exeMk3Qnljb|<>~1;Pt?=f+Y7g_x}01fbo=#?
z{L|#U9r7gp2N0g#%Vl=)-8Y{{1hV-u6LKijY}bZ2t-gPGBM5E?RDu?FJ#JQMD}I}u
z9rr(c#-=<)bmzf;0*esIjE?T&6C(sd^s7>ely(mrzP=4_g@L22P*y&Qii&cd5O*HG
zyNOKY%BJWtsXwx_4o~5I6g)e#J17y0twT{0KZIeQLx{`JVBatWGPuyw{ds!2Ecgyi
zn6LmG;64J}a#b+DoUWZBKpq-91bf%L??l!v7*=n9cMJ^&WcKH&XIE;Bp|d%+aFHzy
z8C5n8<dkSo$*Vx}JPsPv$S+^6c_c!-s4RqBg^x)g?8eB-4z;J=(Ld=d+r|fXibyzz
zJ6j4Ii@?k^H;n46!VCM)jEESgYG$aa9A@1s1G;Dd%T^3V=AIyK8HdiZTj+N%T1~WU
z4*+tPGL*;wKPn*52&fs`Ik$JvF+=W7KO2G2BPv~chYPntu(~c&)3yRQGVweQB7}eq
zYKXA-+_O85%$f>9Vmt4?4UIQd*c;U_c+6@Y2&!4npKC!p4rp-Ktww;Yd~)2K&e_>C
zWYfaK|Kifp)3Z#$>%nMTG&j$vu+%;Lh=YskEYp3j!^!u!9~=tq56zVs1s*wzw6wG{
z$?(omG6brs&CLldq1_$3MgCaKCEI7~`&(9G=Ep0S61R{Iw+X05Q>}QK{O(`MRLA5*
z+c-GHLP?k9kUED(wnhCJK$W~_1EO{whCad_UXZX@$I<MX19LkpS^5lkgH#jNdr+~3
zKJqBp2UO#@1<D^-y%3t)F}9~i-t!E(_3hn1Wyp*x54DBcF?x2^HdIAHduI*1!FluX
zam%0nyYDOu6aZU)bg%;oJO#}6k=XN#ed&)A9-W6Qm=^84SkoK~(oYNc*xOrK=(+QI
zU_^}{^jW*(G|TRx&?Z=@T9iTnk#7sIg>na;?>=@l>b)H0aOK3i-Pfn2gv-jhKut}Z
zSz0QTHvf6&%hOB3XN~2lKE(eIf~TZ{y`&KR<=VdNYZoc{fB!I+pMzKAUobbk<VL2Y
atM*)r7wp&W)gFNVDPPc#e<o`h@P7aaG)*M{

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/ppc-nb-pois.jpg b/docs/2_39/stan-users-guide/img/ppc-nb-pois.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..8a4f54567ead004aa821aa5f78edee974cfdb40f
GIT binary patch
literal 195082
zcmeFZcT`l*vM{>IQIZ6SgG802NDx7WAW1|JBnu;w8FEHo20=iwpn!q`f{I8Oa+V>Z
zpdvX*7!VNwnL!x`Ch#`?!a3)@bMN=vx7PdPu5~x=P0?Mux~santGc?j2h#^j0LLYL
zBYl913IGhjAK(B3Xz7J}x&eTR32+Jk07if+(G#ErArJ)+0B-=GE~5Ucgn_Y$<}aAG
z`mpj~nc=IEt}e{X+*IG_vfgjf4AV~j{@&E00N~>r9B6K!BWh!7C(1AbPy_q`3rLm4
z$vMbh+sy3pA%#QS!{7gU;U|Ch02q_|_x1mZ{U82fad8cD23cSP%l%w}oP9tz902GZ
zJNpL)0{{azSbi=n*#8hd>;rco*g+7!aR|Hr0fTh__su`xyT5U+nd^c$k3pEt*~7^N
zgeyQ;_U1p8yZ;mXM;|~8kdA??uWO*Ev*@8L%mV$~JiT3iSN!q#KUDt{=6?nHc!z*I
z|GpsLlQjr#VGRC00rGyw#p&W9EDypJ&Q5w3AS?mGv!R|=SN?=4{=qutAS?wAxL63>
zQX7Opji5Gk3)Hs+VUP}Wpht+|UvS<nk89RoIWJh=?R)cz>7VaTI|u24vINAya*Btm
z(V=`r0Dwl+GuY@F2!nKJGz0z25Anf1(73s{Uep6&SrAT!`xzhh0i;RO7!+!8Sbo@_
z>02JUhy6S(f9K_N$qa-+-K1G|^)@)f2iu|{`3IZ**#}w?U+*i2?SgI5UUUum%`ZrY
zRwUTtnju&YwoB_B9C+=|cWLjrdFmVeiC^mxXn0tD`0lK~_ir5q`$k6{7-D|NJJ=ST
zyi=f_0SJRM>C9YxEf3{+2!{ezR8D{^;0N3UpU!|UKn6qsbHD)50e|}gfnbdr;0btx
zC9YttD_G(R{`D7*-0$~-0619g355Qx3j&{JzcK%K^DkUcz#T001-|~qbp~`ro&zBu
zB@bW|EcN)CrYN8b{uKh2x&J40{^-m4pLG0a`Djnmo&|4p+RL;TXwT8A0iv`}S}j^l
z+KXVR8m;=@w9UYG{^IDcKkHyCf6xg4TXg-6X9e=!1@agSIDt<Lh~WnW{Y~Ltq;gLn
z<itN}Vgq5G&chklf0C6a_$K^s47q=isOkSn=5(LwzS2#A&l$Q;f71}9o1^;#{>*@~
z`3En4|ET{{%7-K6367}Wzu^7HvG@y18|2jc_nSVT6g)wmUH^$^xqqfH5Y>L{AFWN$
zHeVlO9f)%Cy(jRG(nBe@CSJi`0b*tUQF?mn^w#N#(@#%7JH0`3pQeH4HBAdmE6qDV
zl%|uWiv~^8PlKXq`y2ni(xJb{9ysjzp<W&K`%t_7;_#nk4Qec#4x1pGINL=w2%99^
z<-f5++4$Jbuo<#RgEfMGkJ~@=TlAkf2jyZ5w)U5N{(W1AbH(#-O5hk!9RW4K^B*n#
ziw;<VJ@y2`K$?M|ybtxw7l;6H|9R&9$q}sx?HO95|DfIH4t4%7+Vpz#7wNSDQTkK#
zYV>F5FCD^v&1HINuuctp>iu)yK>hd|<6k&+4R#Gbv?qYBpMON4r@Ke6sN(6<s-oJU
z>2nn|@^zM%6?O9V7X4inBpT=%<Qf?2>LL#u+RelF0N{uDZ+l3^JN%EbU=ski@Eu(H
zKK`T3E(8E7f&hR|=^te$z_pU65CEE0oI?Ub|DXpN>h}N;<aSo^_uo{kI^g=XaB%Pg
zT*nx)0Dzo%a6l|RI3SmR^gjVWr}sf8z{yBk&A>=QB?eG)Qqgcy9rOaAsiC5y`hERL
z7@(r2p{1i|U}R!u0TCKF0BR~48fsb^Iyzcf5IY*I2WUCzxQ?IErayYsi9sxYTk&@0
zQ$}%}nl2vm5v;`7n}M-R%)ETZ_yr`Tq)(hYrKGH)dQR=U?nOO)14E-r7S}ATtU>mi
zU0mJVJv_aFf<r>X!Xs|Q-HA^~yqlDqb^k$jPVU1;dBr8qp1&x4S@x>7uD+qMskx=K
zyQjCWf8fpA!O^ktiOH$ynOV&8%IDQDYhTyD;dXxf+{OPQ?Cl@&MFr6O!4~-ZgR}p@
z7bnOUH7zX-EyE#SRMcUIjC0b`9X~_QrG1sbDd4D>;%!E5oy?~-T}<L<&9OW;14o#7
zC6q9dxI@-{bM}A7SnU5N&i-cXAAHS$#(?_wMMF(ZLq|hHLq|^sUi6IghZiF=<L`_4
z&x`f<#ddhH|9u^RM5sU-w6wGg;6Dx)CKitW_H{51x{(0~vj8g%6{t)!oB$LclZum1
z0RMackCp+=!N-3)jx>eUIKTK1`<l}ATko@1M?Q!pEUQWE-6Fl}#VxclQ+V2?age@3
zMudVL=@M2MQ!}LjEqgst|2SLHAx`{u+Q^${B`AG!?G4yW>E=9cyBk?w+I@t58GYxW
zfk+7?H^R$CDaU#QkZwW-J7-HAd1Kkx!wmeg>u$xFCU`ecJ#+NJ`@_yy^6p&MugxON
zs_@R|ycrO}SI$yypncf)jM)+o0H<vR<X4qoDzfH?9yf6H-`%i>i`Z+Xvwq8koAcP*
zo>7u?TjJ&Mc59?^YSS=BPGopVVuT82CwVjc)8I5D>_`H{95HZyvS!(I#vOw|FAJ{k
zUNANL=)Km`+LHMq&(p=^!;azSy0b^KjNTTpM|9wz3H3u0V<B>OqoZ5+tpv}Nblql*
ze?{lxg2zGj*#r#HiP8lX)YPC%B7_O^OXcqcAnfZ`J(*59I+*ahRZ%@6+jdnnM9pG{
zv?5(4gV&5<K2iLycTX+YNQ2CmF{LfL(Qi0a--MGZ5&8>)Cn^jF5a%ifDXs^=(i7-T
zRt7)v^)ixts)&6ZoRvk)zrsQrc}E>#mvivgGWRTf#^jPRw{t*bUg`6VJgFAJp<*xI
z{n(2-ZT5p%@zRgE?kr7ZAw~<?8n#oAR2mW3;qGnPTTMy9`6)71Eurb0vgPsegH<~F
z_e$4c(#S5=1Hff5X|dKXR>KLim^3Be(>MA0WmkxA@XY5Yv;jS<Preuyi>?&`xP=&~
z;fdm8;NRV+(o4$LTlLCeUQ+?K_GQ43Mv4G+I8eu-wq;$uvN%`11WiMFvd_#}6D+Y5
z_&MeLjW+%j_@3i275gi-nhsq03^W<Jl|Zz6saKv`jkiK@O0!?(+fx#;dfAHS(*_OS
zl<(p)Hu<qM1CTn?%{_N)+cNj0#Rl(w{MMNA9h8!YLr=oxvjhAe7@%^|PqgT`uWHdb
zO%n=x?4E3Rw)$3jhK1%g>9XS9iq87aSl-lUcl=A{dB~ki82<q<#m;#EtlOC}`~=m|
zj0(JI`BbeYt#%w8FWu8-&W(_`@iJLW@*x)ZZgO?BVT;|tNMb};Fj`woRITZeB(tg3
zrK73Dso}!LFj5|@?*Jh3H29IFZjlBLfLO=@@D@sV4l28e`T=m!SPRlpe^n3QKD@0l
z8C!EiI}#kbwiUmGoSEvj8^*KWDL4Rnek?@aJyx}im<rBc&Wp<7yx}vQuS>1_otBFD
zCy$4G9^RI(J8H+|Tsp>m#Ty#$Qtu36IJM95yNR-ef2p+YIRLQc3VVvC$*|ou&jaB3
z<WH*I43aE_3i)y91#-y+N^($(LM|Ejlr3LSPTEvBE`0##pgt4vXTZKA;jl*O_$b}w
zCkp+clw+ps3j=N=CN979EhG?bOI9|x^oL!jIzb9!*fL2N&FjtdYt$7&4A6pnoT!|y
z{3#RUs(4-J00>#R@eOe&z)EJ%m@r4=r%W6G)pXAw$J$v|qziSM3Qnwz|2$dR80jh8
z*smZr$9yyMu9i_I({Y<4gMz=rDNK7r+fRr|2;H1L?vx%wYLxQ1>PaphCZTuMQL2V9
zPs$b7WIDNI;;F!fL;qYvs1l+~F;B}5fSXV3K6<G)M&}P_eE6gvu=;)eaZsw&D&iar
z0Zwaq=+e^^qJ=?339?&Axw!#yV%`M5Eh(47rGXN6+v+KOlp2Pb-v76X|L-QlKbM{4
zkBxoj&md7S-jK1Zu*bIx-QUeW0JM~V_X@oWLQ|6AN)A?IW?5%iau&i@T^%1CXA*U5
z+}NuF1uqUNjT-@6A#X6Ee^$!uUoN5s7X<3+Z4FaQ<@vqf*Sc2q4eilwH*&&9Ezt4Z
zXX(|p29#T1@6aS}af-S7()Ru~#*ci$4xr@ybD#dF!UrS&-pK78C&y5j!*%iTggx9*
zj6s_7`}8`@lj{cOUn<UCf4)_&8^DW9m+t!Mlns^Ph!(_OAdIX)3DT&ot5s2Kgb)40
z-TTj%J@-47qf#P}t7F#^&ymN(sRvTK_e5(?2qLkD8G9ELjv}Yg<;ZoPN<`8D&~**%
zyhv~;Jc7mq6SxN6;hJu5%~y#h1xY=9_QKH+tL7>9*3((|(j#u_cSQiAZa^ZizqYq)
zq|?<^jaEdPHiRDi5H_bvG4FYL08B@AE$$zA5)IA>+I(^~8skn7r0}l{OwExjmWckd
zmq(k1BcBb#bAH~<nu;H=;54V2{?friu7|G5o`;YV+l3E+qM45jKZ&rr=mtL$k96O8
zzut%rTFRSNA%rxJRaL6r;c8Ljh}?2kC0ik!pS-Sf+SO;!P~Ge$8`1#k!=5Bi*ljxF
zylkDJWSci<UNh<vJLz&R+*iTDM=aa)mB!P=W>hDdAUWJ$IAej2nuv139bIX<yE4*l
z5m;^GDOjO^5|Gj1|NcZ^a>2zyCIX(N>%<6=5sHa?znzL)YAiT9mXW>-7tG(%l=OUF
zXD>&;N~#yE_VKaa(nx<pgWE_ZLh)Rn%4e)Y>%YH1@RE$jim=1IhGG%NZCF??m>lQN
zeQV1U{zZ#LNiY8ti_8s06u6l7jG&#M`zH(c^e3Tv%xDHEd@=0+U=StO)fdhx;O!<8
zO1vPZlfDar_W2yj@Q8<33LM7GvZT_LJQ>ZaT%tbb6tQm@*i8H!W?xrr#W#fr=RA4Q
zCHek+Zw2o;r+DMH;x$XBUP=^nTK^#81S!mLq6y+7<meZIGm#_Qv!o&I@!WjAMBBJ;
zaYjOcUh}+<50iAp86$p&4@}dVA3%2Mqpq}LGd4q%a}v*FP>#Mf?=f<>XrK(POKv_K
zaJLT-Wc&76q`{uT#DmS(!J2NCyW+T5ZYZN53l8%RS33Q6WUD0Wo9;rsE9C4@ZHS9J
z>}2SlDr}G@@hIR;v*!8>>kH!tk~NNv_BS~OpAUUAY4OovU38Dlbnu0%>j%Ihv(SW`
zhg>RwaicvH<{~c>T%UciFYS>F+2xn>S*<nPE=f3|34b_hqzic>sdX$?yQ@p|3)*{^
z%wT}KlK3(l-m8+ixPGg$y|Kw^CvD_C>)_RqME^lUWl|fZSA;4N10(Us&tWRd+jAC<
zO|)B-YVAb-ik%(ApL<f1ZVNc?7O{T~!{?Dp+S!TnQKvcxPgneZ$p}_9q@~L8!Z+Sh
zo#~8kcI0_9z2+sO!bD}dd;lb(;eLcurdw`P2yyb`_A?BLq_f=bxs2gHH;?aypYT(T
zHTUF$LLYF)k;I=PS`ZV^nzbIk*=?hb^%-~ft7=~KU}of{Kp8hDaJV!xR*Y7Be21!+
z;=3JVKUwz^nYvi3x>>g5aCB*RxHp=`pa~oOIFx$S_^V_GJ!fLXQk)`_OH>;vPTr^w
zK&TPs@SvK-q9}t|vxLNzwA{IR=?4a_2Y~5Vl{bG9Mr8XEqq94OW3)_T$sdkK;dRI_
zenMu^OY^zC^QYc<TH{P4(@UMcM4(+?-;q%)U*a{;x8=CUHre%AfJ7nJf-YYPtman+
zg(iFjW00VnkLL4o6d!YMXduczQMB(1OPky=l3eao88}t$tM%FW6jNUS8ml|=*^icF
z)hMwS@a??)?P<OQf76wEGln-Xti>=FW$IQVS|t}Za53cyA<TK%!7@rFFmtZ_x_x<T
zr)_cOGqon>#q(vmYso}G3ezueAv{4lN;-l4j>bF`BM8V|^hAi&aS75j+?&(HoynII
z)&6yMSe(Bc^danr_atFQkekXO2f(w5{gXs?&_X>Wfp)4U5cDzhk^zvZdZ@2L28<R8
z-#$VTzzVP4(G(b89$9cHcrW!Je=Z^Yy5SZ5YXR#pJ#ZB@La33D<i`kpLgaErmwlUI
zh4{EYk4^fw5#MTgi;()F8bKZhe!=lp>s1tqyNP0catXbU!3>jSZ&6~<9)5T7kOO|P
zj*0}6mp(MTc2U)x+k<_fh3yRwKz~kc0)j1K6UVC&z8?Vf$jvQ`LbB?62)^&r5%{ZT
zx?V4ydi0;X>lm8EoMag(QyERO?8oUT!j4@iUtZ+6A{YFccQkUY%+zqHG4-uDq_|9(
z05VGJN+^37?%WUMH~`!-xWCu8xSfCfyd8Zqd1oZPl4L{KgPm9k=B(L|jynM8d<o#1
z&OD62k=wZ+OtC`Y1h#^Enf;ghSxv)N#yPln@i{RXLlmZUaDOq1Zb0>C@IsSUYZ?eh
z%#f$N0Hf@Pt8LY)jH|EPhbx>eH|I@rhwX?A_>bg<_9H|HL348ZY9nPH=i4acX{Tc-
zE`OA2?FqL3oY`Y{Pj|m;1J<`+&{$(=EEM|^>1x7fH||5Kgb(ZphkmZ^VTpJkndM%!
zc-%@$onHXWgZ4t-bz~xxfV~zQUn4A*2|^Po0#jQ*BS+sV4e>sDR&9URQf^|kceJAP
zv$^)Y0(Br-4da&q!CD^xvsZd3ybRqYr#%(~T0Y^s7ltm?HDp?6e%7pP^?dM?RrtZ<
zA6-@VG20~W0sJF!0Yzdi>I}Ih>ckdYvqx^n_W6sawB|xA@%B;n4m#|#7hi>YpFKT|
z)bfEYl@?Z;#8;@OKsiSjX6Jd&6=z(W7~$k_vb<&s#>fCR02O;UiwYaUHm!-qh~c_*
zO?q4wT;%*m1v!+G%XzC(KXJdzo2_?EE>r}(s%QOR`Iz1Xw#=dhY2TM>dKR9mUxQ=2
z9l`tbg_plU5Il&5A=$&t4EHAW+^IorhS3I@odF6H9qy<Fu4yw*1N!RS7&K0~{JlK&
zn)GH>rbGYPDdw@(m|JvUxDz%04UpNRC8!eh){(VSThTc2<qsv#-&VZ$QN3H~y}9Ca
zEbsgB*So!0LYt{Ka&9nncGBiiIyUyq$f*>Luh3)QN^RGdn6%aRB^OiaY(M3CJH{n@
zvuavPNQ9Z`bJ|0ZV@FEkv>?U<IC~0<D$djx!X3p^{PrmUv2OKca*#z%O!#B-;KQZy
zum!{C{@OQZXnA%~UgQ)|K&1#d+*I!nY=($`GEzJeWf2kRExUW{DO}NSe2X`?5k;x8
zzF}#rl19)y_Td}33bm#{VbR7enqpyn3W-R;c6mY5NezP|?_YoZD4e);-dGDi_FZVN
zO=_;~f*9-)K<*%ywj-An;*oW085Hqwqm>L>?Dx!|2#?XCtTRf+*|LcVA`?4&q@$|%
z5<(5}D&c7VtplJhCEqyz^-=TK$>~{px!Sv*6C-_@uUv_ovd5JSgu%F>OL^yi_1f<o
zSxeHO5hnKoG+Re!#RmbQ8*0AUJEkN~c8Za8FN%1?4&1vb1HOo<1E8s?Phn;S{~p`a
zb8J^e29n68WM6sXt&4QCv3g3Zf~}D2&T)(S6A~hir_#Tn8(}1F7`|eByTNaUrt?R5
zSeMEV4V9^c+4rlD<+XqK?kfDeNgmXJhTVJq5R>r%hP8apNirb@tvd>)ks2q!RpO({
z+4^Cn%(x7n39>4|w0oBQHtEdsaP{s<&(d`Pxq7Rd99#_OPng{;Vt0D`Jwl0!wC(}J
zUfrIH#5)l(eU`Ro$8>zW-VZAGy<Sawowhxb;&ayax}B`QNvaT+uuW{+gB@rIgq4tC
zfJ;pHiQ*-MV*HdUy(U@ajB>xfrR!!A51{N+D-(WzbGIJ*<xvn8(%B)OMRz{EaQUh;
zceCK6;E;Nfi`Fen-mIcQHL3=^^myurBc*qd#kcCStaEisN{~nr-Dlj(eL6FS>LPY&
za`WMw>NrZy_Jd*^2)ZU2eNc`UF_-Z>hE~HSM!*^68Qs@Ux;|BxnOm@ah^g9+4@M^~
zV(p;G=VdiDvFW+;*(L7<K4~*ge{D>#S<H5cGo#0J38ltm#-8nq0=LEtUscczO(br1
zOF}k|US<4d-TkXLxVGwtG>I_I0Z&19!tscbQ?JhmRH=Nn-o2m&iG$Q_lXxta4en@4
zvd2Lit77uxMz6~~YptCV<ll7jrc<39_<Tm_^ALwwdBW_sKxp`K&dFt?r2vUQU~n+7
zqV%4&5QiQD+}Cz_BG?E>3xWm4FP*os`vOvTd}Y*EQi~zqDMG0B#Jye3Ya>;|rf0}{
z(cr|Pogk#ViZUXnk(ge>gOvLyX~O*(5~+c9>P&ej)HZslmDN6H<^iNB>@bNZe;ImP
zQ-)+Qrm66}v8k94p{(veV^Y^(?^zXcEQE2X$;cP*d08D%v8jy-r)w-jRg&C6_hKk~
z?J_cnatzxyl|a&X<LgviH~MpaS(Zy>c99A7dGIH#%6CSg<h^pdIw7o>WB|VM@cJ|%
zd0s(Q$p_ghP^T;Q-8yH`XyH>PrT2^KT3nOx#(sSfdspF128DSais8d)<`T_k2Si#h
z8`7=SZPhV7Qf0E<HRQA<Vp8C{8MZd@S;4>;@0yOkqKrbYRvGDdQ=ID3K%6_aG){^>
z@`RkMw?u@`uc%Z(bE|^SnIyB~lHF_69TvFUZd2B9G=&)zjqS$|ss+ng^lvE7SE&rh
za-D8(PQ6W%S1qHdSw@%tyqLy&Zd^<h08RtK05=uEhj;|pWvnSpSX`ln)J<)vy3^j8
ze6BGWdh1>juPUe48JFnv@AH;se6B9$ui{<06&p8Ei|D4|{%v8(Fs#nE4yTT3|EOJM
z>u94Zc)eRXRkC+Oa8)oiwBUja4vGKR5Z7IZwL~$)1tj1;0YMdnOM8Q?^o8%5bb8sI
z#~xW|Un@h6LkTjbq(`tM=%(az<hr>T!%JirRP%iJ0f0{?KauT_jJlM!5bYbSzdw^V
z%Sns4&c>su_vQn;R5Nm}2~WcQ+KhKDCAjO1iQ}X*LYhtbUO%qoZme&7s~r$%v3C5)
zrDyP1qD_t&eFTgkVL*CR^$3grJ+U|d^yO|T^!rV-EGhJvrd1<Gk(?oEmde8NaKV->
z*Vaad<4?gQA=zd%rnb9V<86|b+Ra-Z8P=q7aV!)TV_e3il>P;_F}B+!u%d!lo`w-6
z?3<pz2N@Pp8*rLzS$_kys#8N0W@TasWoVH@PD6%hP4I`S)N+hzN4V0(kLT=BJ*#c4
zz1EoPcX=z*XwQftlPI%1JXeQ@Vy|X~pi=~d(s*^{;=9Hla*q4d+NMXB<IYl;BZ+R5
z4~z9vaU?km?9S-Iw^?*zgPQ!666MUsMxw;WT4MqJ<l&ajoGPr^hS%@kJ3+Iri$x~?
zR1hXu7~oI|@_Jt~QMO&to~OTJuZw1>;3aKbLN%;^fZinS*a7e$nJ`Rg4$n9ME{&pH
zzy;{Z*|z~io+z0tbK8U`ia}p<gHAwVsL_I;QTVqpJBB30M>Ir(g{VxDAE%s((#odv
z+_X9&hv_QV`F4HJ3LC2>J1vpr|B63@8IB#+`R#wdKsS78c)3V<FTi?>_qOYayD2jE
zys=`48(E89MI}Q`bax`p4ZKl2>r7z@z_DCyB_v_DF&6`QYR2kY8cMaUNaW$e8g+&~
z70;csn2WqM^}Vzc7Jm!U1&!<ABET>NTHaZ<LRN*&Ba&SXKL!aDHA&g6@6jh{u*hY|
zRvpoTq)vdZ8~`(m_$BOvk6xOJ*e}oFCEsxN+AR5xoWjQqeZUy@-5O>`AtiUZIQl_i
zA_;UoDMp~}R3I{YF7Auj5$h<-a@hMiggT-`I8ND+#*tv&@5f<tnVT;mT1Dj%{{F1t
z#muHU_m@1_!vELX!2Sk)i;UK|Ks;`<PT~Jp|Au6=GVgmdz|UPaUanHE*6jNA2YYYa
z&C3LqIwmC_P!$f76_)u(ytykP*?ej3!Y_nr^wHWE>Sf@?o|w#p)srGGOeQB8PT4G_
z2f<2^{OB7BF=!r5l`f(YNpDj5+IVnFh3{HJ&bOK1ghUgj#*@xD)I%(cKcpmq)uau_
zl>br@U((=S{dZM+c-c=7?kG!ODUZ@62v}H8D*Lfr_i_|`qtnD^VW-rnRelex6i>yf
zwQ2YTU5Fx_+}_G3sw2i6*|AOWh@-}l<z5t#@y{*iQz!Lj=Qi)_yp{EElhItecX9d*
z-SNOHTBsQaLHhE}bwnSG`SXJur_#%h_a#3KGAsc_>|c8DcX8}nS6@-aXGI!t<@zb6
z1e8##QEmOjrfcp-+O>2yGmG~Mfz^&?)Yo$SNu2vu%d-78FKR0SLkL!n4Bt#oznMM>
zJv03tAA=Z{ebmWG-V9lJjkpAwGA-n)v?I#_fGT7L{hv!yKl}ByffX=N&xeGY5>8H~
zkj~)jkTV;2OYA4aNt|oLFh08PXQom35$w50xVnv@qK1o4Sm3qOPSv-Y2@R{iNPN7;
zrUWC}z#d+f6pv!;+f1|6l{d$qY33U$Y{4Q)3gqMiz;%0iaar>zQD=eB`fc>fH<lE?
zdP2$7I(r+*HvgR=p-OJrOW!J5+?^X9hE;9%6=JUzPB(RRO_>o2og&5DnO^h+z4X<*
zU3oJT%GDVOu50K`Fl<AoPIw)SD={G#9spji1q6-B=D6VA@grqpem=@fnl{Dljpr^K
z60|JEPuXxW|0FM?K1<`@5WsEe5dIFi>Zf1wBBRf?Tm|FkjRN~OTno0_a_-_EEDeI1
z26rCt^+Yvy0@k}*e?-fppoiO>1K_^IVNBjJv~b_S59a@G<K#FfiSsDMNV*qEWV1)E
zpGM6kQ!b!gU}?}sboT+kx-lEJ*@USyy*{eoC!T8-DNJqR#I@=?`)X9eoK{k8F0-Iv
zw~jIi#aeG<pVnJ~a%ydN1}r_&hDE(DgS|6j*hFrTm!YiTa4;%aqMGcI<b)oTs%bPj
zMkBd3^?fUwiug|Ac&8A?=qnTC5e%!FCTMPmP$0xVti1ESc=qK0yOPe<^x6c^6lEWk
zr))Y?LSgnHdTz|#BDtrS*N-mw52@Z2kUQ1=kvt~0j(v6JzNdJgeU5nSH0rB5-ichG
zAx%^uK}Hcg;g*!i$+jwm?MJkMFqm<MN|@P|voSAQ)UR3w24qpKARBYHhL<4I2Y_yE
z<Lf#<&hR1ov}a>K?>^_eDJh8<36eTCz_w)R|NO~q=F=5Jdy$yJ3^dkyaaPq(M1x>d
zBA<a7D*j=w+>n#&Nn2C4q_bb+De2gieb+lgV=94wx|D(k11qsr3*plJZpM|o(4!8X
ze)NH*rr6Jc+wsWO`z+RS={skH9#8Gc&~n&y30GFYSx>*bc#+D|QlI*z5WoO11C%ID
z5pZ6ZLOly-4uEA1DC+^>gJRbBcoZ@#pYvX};e`18=iL(OK2<mIY1Ze?1?rCxcQI(r
zua2j&lqbt5R+wJdy1?u9Fsod%#;;#uU0KshKgQ*eJ|9MbNW$0ypx1Q9zz+AiuloG2
z7|FP+HBYn_#Bw~AT9YzhawXB9k++mClqZ1Ls!@6q8n#5mZv{st+&s}3*2%TTdfA}b
z58stqt_c&epF?$kbuA&u17NEU_id5Fd>&n=5takN-VqKJh7PKD;xmqKS($5WuBm6(
zk)_C`NDnkUWeCj_0!el}W7V?r$j4i@MBSyMBipw4wmt2M<F-%ks$aEPy2nj9;)ULM
zG!8~(UM(H~hURD;<Pr+aNOHxA;{(Su43>ztArOL2!_}9tIi2T=o94O_=2y5~nzH(g
zZ}`VQ#J-5pynsb*9S7%#JNcEj8xGYu&vk^Mdu(uswsEpG<aNuno6ih1D~2WJ3{P$G
zi~EbSe6(mGzXnZq5xN$ds`OqQ--xZa^(Ffym%*&umfW*EiT(qi)^Q#Ewc;zIY(<s6
z){PeWMx--4;nZ3WECIqCrHl<)fhM9V5)&#X`{yi2SB&qvG?_L}aqy;J3=uk$@oqtX
zliCDvZ=>T3Iq|R|$D=ib=!oUl&^{HMqaKp}aZ{jY>%9VrU^=J!Z(I6j`0c%x)=-x~
z>xMz~!dRkg3CJSSWvtKJ<`E4&cj48#YG#ZC|4C8Wtzhr+sw<PUp{PEE6d2YVGApxU
z+ONJC2dmCVbe{0p<RAAcbhR%}dCJKo*}eDK-hOslRA!KEJ3B>wE1Sr-v>m%ZPjJU(
zcCeDp)<-k+e4Jgdx9eLQ4!oh7V32tymt7+9SJv63Mk<XP)<39-GL(;+vY>+alWyUV
z@syL{n)c(ZmSZ9IE@Cpx9mB`NtK=N|XE@`y=q+8WtXNN<pxM~wCV{Jc>~Pf4k?)#f
z;orx<<*^l121Gz+l+?d2r)kTN4wSOjOL!FJxE#M0Xk+~qwaG^e@q{(9Cx$qjwI37v
zs1bmvPd45#6l!Sb<;p4#@@dayKIRjAvyTb(sEO5%00YPR3^^SkQxR&=FN1;4-H!8I
zK4x;4s=4Qi@hs8#14@rv2z3Vil>1A*EzsWKSUc$bmvF0E`zJaV^q0T$`$%U>daRf#
z7tM4ii@tL4Id;K=W}I??6h=T2&ykeIJ5G=cT0NBJgbcXP08@NniZrWpop)kJHQCJp
zvv<9xvY_mhNI!PM3HL4cWd|$v8P%@!;;*5XYre-WOs*Eq|3ZvJzr5}9<@B1@tEFHD
zx^x(ccWPOrED{IB901)O&jU(5^K>^91PXJEQ#O+1#%t3nlDtFLih|N#phKYU=sO}1
z0y`Gojb_nM^F1D?SDPm74g8pIUJN+9QvUX&o93g=B`PgQeCzNQXcFj7lQTn}Xl~&1
zu;1XbM?No@5ko2;2N>78JX75_<E(fVs4;wcMcN5i!4vWR<l+ty%Ge_7CPk8T;n5w%
zGqF1#yjk)E>w)&8bP_Jq{5C?0_j;-r;5a`FGQQ(1NfpNq+7A9s>Cq@%Psxfzp#5^s
zuONehtc!%;3y<`W8jDy1DE~$WGx>RxGZwM0&ucfC>p!Mj_yD$U@zS8{XWAEw^cMbg
zEePy?BOd=35szBv(&OAFfyB+L*xJZk`3<Ka4vEj32|M8*qp4rg%*JTWf~giXdtd)|
z6aWAJCVp!HRqmNbpB-7X5q5QYeKuI}`lnBqyTlTXU2D%$7vZtgmM8#tNFOP)yw%|r
z6y^quiSxxYpDOu~LME7+cFfpb&R3h;<JH=2Q;&C*_g%rs8bC-YBOJx+w2#)8<uPfv
zmd|+9whC9`Lv4Lm`GvMZ8mD_&YZ$=D>G3FV?;ZPs<V)zo2_twAy_y{GenrDfm?<1T
zG&hkErxdz%w{i&*89bMM+J0yio{F>X_an$AcD`4HB|$iU{u<Y~;Dz>RO*cLwva@i4
z_$c=>^|Hj(U+Jqrps5AS1=$7ZVE-z@Mt+pLToH<C7=|p&YfQ1y@79m4WlbKtw#z7@
zCbXZk7IS95_iL88fmnhSEr-4~!%HQqy<H+3V$qQubY+K7?BA@6Fer#8M^e?#5h;T2
zjW-qn9`Ao}vuXwV<tg^$FoZ-G;-r@q?sRa4v)dCgSIS39zdAx1;rV4X&0CJt#k=mI
z$2kpTd#T={zSuAGruYyK{UY;v%o4|>0E8qV;QG%01&_7q6z`b`-;&atVozu7=~KV>
zQEi}cZA5Z39O1(##Api4ytx_IcfN9B>%-*0Bav(_y0pX)g<w?1wphV9)lyg?)(ke+
zhOfablwgLZe}RkIbgM~%;f{}wudfxKPtEJqdCG>(Bizob_yN(gTk{U-kBD%@CrIsN
z13PHOeG8dQ%EC>1S(KmMs!+khmAD>sdb}8W)9~=bYO%MWwXERkDWoBC=?R+oCo~>e
ztI$ITE)Ni1Oxip?wf$U_A#cS(?~ySl`Aup7ig-%>^VDnbe3mI#AstEIZi{W|J<U`!
zB`cXO)hTymG!6el3Yo5^Q@wu!YXTs^4O17G@bkGYN*7E-p@r>RWfLQaR}o{2>}XeL
z!u=?1+}p^qFF}4U)~iqR@%_>>X8c&d!_E5kIIS37*3urwPGL-EBn;zNA^U(Hm{a3<
zosY!RwoDT~hewZT3U1s=u?a_ZO{(x#_1TWj?+Q0rs~+b#dvsMpE1ov<7;to)W->~h
zXoKvsfWqPVXb#G-0JDb(f4W<ez=zQ;xddp;!ne8h8E*E%L2(9xwW#Hs0^tM%%oLIi
zAG`b!>7Bv&1g<}BBxNL5>EoNHd_%ZWzT%9hj6U0?(-NPl-!>~p6PM706DZPAG-mrw
zeFjZ9bYT6(*F53h0qL);AEuX+LSGI~*BX6$7je@_BFiM)bSsSrr*tjWOK+}&-k2E}
zOc3_6KPg}D;sdct%e~fUAzAfS<h_;Rca6rIoj#$gq4%{Q6);wc1OdAI>$Z|#k_Z*r
z!B}Ui$Uen0H!iM69A%k#j5|c5oSIz>{RBcdVL*B|+_kv7*uA)K*6?$Y#0goJDt#5s
zw`|b8F_UL(LX96Vud`BPA{S>3D58qyebdq(-Mn=lrikp)fv!p+*baay3hx0h8vkc{
z-0zH-!>pM9p*Q=%=zo&b5(}<W7m;t*!}BS9Wcvfa*ayBJz0rkQGo~>4VGULcdbVd*
zdL1Qj+c8nvm`br|c8dh{_okQH>KY>?+8SN-ay&9w6w5?uvZ%zLBlYU3NUtdiP@vtG
z$c`92582|!UKCd7$?Grj`}rja8C3dBS0Ui$x6>L)59-AH1*F0|ySbvAmm&RIEcg_1
zM+YPdh9${)vbB|mYrOZ}d;fY|YE<TKZ9ucc%PxC}BBuImGNnXQBZ?QBxW#}icO+b0
z@o2Jvn`IkiYv)X^YI^vJ4I3nNI!&jR4Q8?L3}AoY&yk-~j#0)Pxkf`W5=BJYtFdb-
z!Cfkq?#~@fU%d76ixd=m&}x{9Fw#Z`IMqjQa%cgR6y(y=0v<Fjchd=Hp+DxV_?|6a
zm@<+T>e+O9;qBPSac5W>c;*S2nZnzWpJ}MB_|U9K#}`kvTNllD1^Et`b=2Qm|HZl4
zJ>V0-RXcD>_<j977vM1u4_Ve>)^2gu#3`Ejy8ZPM1KGe6F2Ml}*|(Qs*%&Xt_g)hQ
zh#ZJ<SWQ@e1VWWEP`$9xA<Sin)pLT(Dfoy!YPyg%@#xd9)BZq+^&6^u=a^oaU3H3Z
z1QA?lMie20bKptvo=V3JLM|3eW}^bbe*#gb2x%p{&%z4~Z#=Y%ZtJzc_!@D_7}4(N
zS53dszcGC(AeISxxR-oJnG@*Gl}`8xyPd3IHfZr7@$!Te<3P-i`;d!B{xk6l+F!vK
zs<Y|z(G|a5)3h4O1gw$zy1m0dga``mbZgKLg&K*|m3*fQbx)u!Z{*_Xurx^#Sc-|3
z0gWeu5E)c~cykN!{HcA@Ogm4N=P&b@LlL2VtdqJDpK%f+E{};*eqf095h4Crcn@Cv
zqsCcyu;h3<e+060;Ua>B2zi(5p9O8b`ADbWPMPqvsjsG~BnW2st|r$6iG6uPX<W%6
zbkbhE8S?DjyXQ9UFVCs5sxi<8%Iu9#KwXe?<@h6nk}{C89TAp7l8d@jR-5KK8yQt7
zFRXh7Q(&_bAYrKWxm;>FhGyTmrKPE^Dd+ol%LGQA0wbaLi+cQSH^TuoqYE;7k>m`G
z5GR}}g&Ev!awsKT-Va4r5?+1(pu;E!m&{&|xfcOxo$HHCC`Tc?tkB*Y3IxkrYw`A&
z+yrI0>TkJ$o<F_@yly!bAmgC^=}o%kPk&aZz916{%-6eNN(E{x7+Uec?hkG?2qGE2
zIi4KT8!<`n*sS!EvobHb*M?<N40!AoqY3W0W4jO&n(Bl|@@?V;0zJinCrYmrjB0)R
zmGgb%labe$#_2f&v3pmAG53bjZRoCV$qRr%Itt?&wirg^Q6Np&s`Sb$y{jw}UR64_
zDP?S|?x&`^C)=-*XQ1W(PMAI@zMtvm0YErquz7U_%yhc^QS)S!L?dM!XBxkhHMnN7
zkokH(?o9S&*;VZvX}O)FKN8guH+VX(5*)w*KTfXh;0QMclYwyJOX*LSO{@)X%WhTq
z%K7~yz7>zxkuKzq)s~4lZL7n(RG2TTrP)b-@q6pXg7+go=#VC~b$3vWRWzKaH%_s)
zCw|GHd{J-fl*4$qv_)JF<d-;hW=p)0E?kwwv3?s{*!TRqLVes`Ul4`oLwF%ZdTST#
zf)E%*I;QlM!Wngbv^Uqv&gG*KZIh~3>Tx-rv(d!{>alxu5dp<H9v>gajUl^4qWE!<
z7}X?%1mWzmU(YsYxZ|8-Lv=?(`-D<?n~a5I+z|J@B)259H_I5?YeM%@^#eNTw@JKM
z2*DD1#ME!A97UEwjhIe{ui<qF1{wQ&vdEt{C@{fx>E!`XSD1nHFpaqtjAX44w3;GG
zzmhMOHM=eMzFb$rLRNH1DcM#?KCsJ1-sUBgkpgDG8*Ftgq4`J}<R?+~*dOKxz#X>k
ztzcq)KAL^eAk5kGxPOOhKd(}{v{O#zIqleK{v!6nG}@d4U{Ftk?*O>2p$57O8L%;M
zp9lI+8L@@@8lJ>(f<eCs8&(93@=*EsmD)k4$GEk%KD8%!OVo2Yt;8WYaK@KG3jMSd
zpui6RxZHWa=^hHRB_<<%-%b#Agz#!d*efuOWp(~2yh88$CjaVsSGMP3W991NRxT)U
zZfxNu4ns^R)WL4%%wEi3@wGJ9oO$)>B&0W!hJW+2_y(*P45eB^XQu23>%<Fl8wjRR
zd1dghS!_>!bIThOnbMYGUtH~=%N>y52K}}w%9_c?vx6dW8RH!YX&9(5VQ!_cKjVm3
zD}B|ysz)zgG<jaAd^a4t5VHElKHFO3l<0j1_0Rlol&FZ6rPH>aR=l&0`OI^6w!Qay
zxL73u`k!9CcKDSP>hm^uimE{cCZcqifx5y#3LP8W=oa{u>MPNa9Py<gb#CL%u%EaI
z%YfJ^<z8Q#a}WinydKhacMo*r=)hFSSfm%Eeq5!HN0a-Q6Cti-j>Wc8DuCOr@)g%e
z>BC+zQ+_wb_wlakwlS8XG}CCzRADuWd=!j!#ifB^ko)i<iUOq<hP^x-v%s9c%-82f
z_ad>VOG#x*ZG89b^pLubo?%gIgwU?UsVr_@4Qhif3P5f@0Q}H1JqAq#I4%fdxt%O{
z2`09%ajp7w>~5z1^!WTm0hv43Z)i@seczgFQ4k2X$5Re<f-q!QjYHn~f>n)GDvK*>
zZ794lCU>z==F`p3dsFxCyb)A&mY@=-<Dw$E5(U3PYYTF*i)Q0!mvfm>;$>cJQ;HIK
zcaneA^a(VP?$``6n#9C+Q`AtE8*eq~q6EMp2Zyw@=Kx^R;1#ZaPtO%=8}Q|mi8#Mh
z-eo}5Nc^mj|JmD}JR8A)^$%%qh2O>@o?wuKlM1Qz?{%D5!k>_gUuYN5yp!*nzK$Q9
zQ@L=3!2&ksij+nOl9SP1NC<e|#tB*;#&EyuSht#@`ER!<w_h$!Uth;JHHGd!a65M6
z^&>9HEW^*9Vm`fRuJ}KzjX~OElkmtcKT_E9c{r4%L(uk9MP5Q7+w)hBI&605zF$5)
zcjo7(oB8SO?xW9|4`oDI5TPGqxUJ!eH8fsPzj2)C+W+XK!J1-fL>iovdGnJ{@3+KP
zFU{$!$l2iOYDPw3y><UjFkb3CEng|S(vrRxOg7RepV#*Y_<V7)Jvwhxto(ti`5^Tv
zpPQXD+o&W>jt(q4iEGge>s!VTnp;Cc7%krkedg<Oc@eL)*AgYH-$Ah#p<wD87kX0}
zoP1#Fn+egy<NR7O)U_~KL<BqeB&gi0H6cFiNW=4I3Kx2JYNN`8t|o*zPoI#O>V{wq
z&<QXuP5vk)LYxy$y!%#Dyu4pXr00zktLo;OESt$iPZ#?;I)m)AG5Q`fm|?<6Rnjwf
z7jk#F2f1%GK~}Q<@fK8VD#A(J{eEN<`cAv^(g8rVvH$(H4<xa-H7y}GB*dLRVLi`e
ztWQv1h0R`WC*1_88%#wYg6~y;O6`a(C8w((_kG7@){z9M%`=}UJuueFRemGy?TN%X
zFFw(!RAiS!he$XWHMibe#XMoh3@3y@dAzJBGW9K4bC+rh_JePGE8H6#=}hmUIbU}3
zGz-<^OxQ(mnrf01mi_1v186plvn#@4+d7l7J}BPhLCS)${X=8e_47$$*_PwkH+n=}
zIBp35qzmAvT7Fx{2H_W{GzAD9A+;Uj!DQ>6Hi<o5FW%fUCUpg=di60h<B!=4dB}QG
zm_-sd4VWZ{<LiZSv2Q_%^4s8|Tk;|K*VQqf=e)m=K8-z+yD3>E@6@lgN3dS^GkMns
ze^U=7oGjg%Sc*QW!Lza$XHp)XH^EEW(mL882qwc^{k7JVN*8*;^nti=XA%2L^io~H
zh0%J2nKm7)pu?$e9Ve!4WT{7ywmqG>vrMF*tyxN^qCZ7WH6y!Z+fPPg{HE)=GG2E2
ze3^VPuFO$6-lsgCax_{*$x5WxdUfp%(xq^YkHSoYyEV3q-ZD5c71u7w3CsC*)Lm>Q
zaUOp`rSEOd=}vGn^T7GZ>T{?11HeE;2>U$+17Y_p;Kt4ud+uLJ^EvY5NxL)Ck=Lz^
zD``P9mRj6YB+zhnX&_f67Zkyl3r>JLq5FRrlIp@P)6@d+XgKx`p{~C`7PKVAgj-)s
zIUhF2%*%h99n`+i<<U4S3+MmAa57;p@$%G9RV~Q7cI3_@@a*0BDo9Cyk|(qG*%Y)#
z3?!a%3a>Uu1DxEsB#9Tut&Gh!)aJJIW@OZGLj>Z=`%k{}skq7C_Q(zGXS#PuAr87h
z>_+Zuo+qCNdjYcaHk)vt(rsr=$(D?gB*LS32yd5BgcECb+5<31P<;my<xm=@N56&M
za+v5#h5Ko^nNM-{kK8^++X5<ZGipSEaB?k{<h%lgodsdosoP4RzsmXT2W&k<xF8=c
zw>nf2cIv|r_o?qX=|(iS--5}W6ee&_F9m;roOZtk`rRquhhvQjeUKDrBWkM{>Z-uv
zRceU+aKCb>Xm<8{fbZB$4PS1PVdAffgp<rPzpiST@juDmO$0YsOBn)Dn&kWS+ic;#
zw8k}*%cUw`zL~uBq@pp5N<_;t)#B7SD`B?Nk=AR8TWv%L<>MkF)H4YQN6w&)+oraB
z8yI(U>=_XgVNNga6L%x8(S3=v$dP^3$>16V@W*HlU>Qi9b9i%X-|M#Ueg}BJwO7dp
zRn<c0rqJzLb-m<^7GGU&Lfhdh+u#XFm+8zl9!WU6q`*Lu#wI+%R9ah4K5y|9&4}Kd
zY|QzPF46L!{rl~;jD*0mag%Kh;uPjb#OrXYGEABbKBR!ymw(aqX`H!7`(}vB&7-3H
zYo7VHNe&xpVBV%6UYXns<~h?ikY6IyDyArBnpFHf3pmH*^1gppXL4g6;^yr)J$dto
z>K^=yq5iK22@Il;nnLUq`Xj`nQL^HcDBr}@NPijJ7=;P+a8L5!3d#OB2qpVDq7l^8
zv&*;_8AJ(!Uza8n+gOgd0Pk+H#w*%?!&^)b$p&7(@X10ymoar@^Rd(t<EPZKnYZ^q
zjicQISGwFTWK(WRl+y|X)D+&hAhr&%afP34<@Rc<Q<p~K3v4FOwg)!*8i>U}pSFS@
z+d=a{hgdH0PQwM&m5dn8<Fc-@yAHrQZt%Qd+WGIckE3<w4Jdj@UE~px&$B2aFvE5j
z@>P{7#A#WO1^(3~b@kHlv8YInD~uYkj2g*U<eVVh7Q4OWj*VTzP)g2wwoW{b{b4X}
zt4hw<qB?ft^kwt1lcK_RZ?K%20)rRZsnDhafONE9CL<oqX$lKi&S03kwC#1@{*%K@
zxA2g<&El-<RY~X9CH`|pA9~%qV-Oh1?84bG4JM*=l+~27@tr2y_6sj<bX+u)KRjiz
z(~1d`lXwM?2=5V|pdi`7NL3*T^fjwrJUIZ0B~gVGv5dK4Fz=(piw_}-%^tW-5h7ga
z|7DrbFDQ5yUGL}Pvl<!xH2tRg>t@g_2DdTfiWsr9V<?Abtm-5Ma2q-Ru0LFIni2$K
zYwY7+vuSa21LZc?Z8YB8d&6G>+M`Y`D)Dng?1@lsSTd9s=>wa)h!}L_ow$;Tc_M9}
ztKHD>X?ec1?xuo*ZzlV;Evq><!}sEtqO<C%Ux#C$OX-WXxe0PxbNz}Nj*4xTA^|P!
zMB%ok=*(GJbFo5Q8BX#FOb9WDt^rR#k6`2Q4|DNN<ous)uMv|QW*N4w)rniyp#n)z
zv~`1)qlcv0&A|#F)Vh-@zikw{R94u~-{4hnY!vNoyWMVEuWxruSt{kjfQ8#}MXG|(
zN6nDFg_v`w|MuiTn<1ATvkKy+2sQ>J!aVyRe#v^JaO((Bfbt0{M53yqj6;PcK_B4Q
zT>d$wVaD~G_hDheck1X|lUbalAJjzTlNx?Vh(`|U(iBc^6QrUXl}5l1EDMhiqz5Rx
z6H(HY;a1-${l?LbgOU8M>Bfm|$sVH^&b$lv?|h;X%uEP84L~K)OGVd_U00%bO=-47
zc)*1@8})T~Svu95B20qfRC|x;tK^OK2)KSPQ#{$D5o)HfSyXW6jwtvYA&CbDrj)K$
zM0sI<ZSi4|+8sE$%GdGF_#iX)8YbqmckhfEare_&1qsSX9h0GR@gdL!{hZn~#Vvv#
zX}r8Y04Dc)L1;~j35X3{c(`A#e(RX|(g(YRFDg0FBTA`R)I3`Klj;GF0fbEy9l;3)
zr7-6bbqE=~i@YRT&B79o5xIFx6W!z_|EK4#v+rE7J9nEA;Wo@{+teaK=?|BoFsTyx
z!;OccjeBMjlV4=eYL=IoycHg_6Z9DVSSB*Drw$(GB@{LyNk`RJ+{%|BG~qfc{RO<^
zyvA-VP4#&@Jy)xZRQ1QA^A+I!uk}ehKAjPFuWXl#gG!x<n1>3<Au&h6V?Hhih;=))
zc-b(|`*D@NevK~4-cbAqM});$Mb<z8Fsu#Yr-UOh8L?3B_k!R&VE*X<?b(MVYzD*O
zDJJJ>V<lT699L8B&;%!_Im^D!(Qv`RCWVnwi}j_Xqd)Ma1b8F{DY7v04b81#zrxTq
z9#T5uE#CmGUX-{?bjy(l4HCVtK3JS%ZTZGwVE+XcfmTRvmn<d}teCcpN={T%eqR{e
zk&gW`Q<`?YM=6jlI7>0Kwwjhho11E61U8%)t9cGFean=@?QTNw#2vNAiR3S(1@+~<
zyA_gN;wRbg{7WutlZa81qJD?>aUGvv31^|HpAap&|0J<x0NL~_LBn`OVTNYeAR+4b
zv-yp@zTM}q=4Ncb&z;`AYr1!avyZE;hM5WcZ%m|%irB&6<I>&29V1OpveSWc2Y^vL
z;*<5tFB0buOkr<E0*r-(EF~FYUsPg@%irHzuo0W6a6fcqKb;Kz%aw)W>3@F$4Fz|;
zR9>f}=s=;jFTP)Xn|r0M8JAweAAz{WK}b-eys;^WA-@uD#-6{QI<<QomIY=ZZWE*@
zD~SfiKDflC?QBTK&!Lc}*4pez?g_bNwG@{M9m!tX<Hq-@n$3r5f^fr!{ViH-TI`G#
zfh59sJGdPPpgp)^D#kKzVcHS9f89U9wuD|i<~)Ptr<JVDE8hUJGT5Dh@L!j(&`sBs
zhYG$qEQ?$$4pM`P2+!5#X8laQPxjBMq906sfME5I%rH+_-9U<lWf)cRg~(p{#!cjE
zg;fsjxYMSDJM2{+p~RA4M2tjsT_Pc$d!c!PBckY*($wB0RaLsyjBS>9Gl|TOh7$G0
zKhUdY=_AZ>r8N7(1YK--H<T73=mBBTVDa^-hq2q=J<b}qb4GGCN9W8V$;><hQL2`7
zK*)^Y2l^ElKXA98^p_*a$1jn$!Ee${NbKatKJ{~`)CDORyL`12Q6L|lpW95-96l{J
z*LQp`s`o>{uXu#~Hzb{=BA78q=VhA2v~(f5T|&c{_5hfpF>XB{mL)rJxzKBe^yGH4
z9QaLkCE_V%9!%EohBctNN-)v#tFju_MVCX4@Gd)wdE&VvBhr~TFPG9Kp6U|3_uP=-
z!U}BK7_UHnp(F#jYZk7<0wEarIE~9SI2_40$@1|1c<KJJl3v*l6~`WlxeYr8!LneD
zrX;QwsC!1@`KEq!yrW|CM7<xolKH&Nw=>0#ecV|UH*Kn%X0sH}mdUL4YrUl{%&Dgw
zMT|nRW{~MHodY0VdnF^@x<d;4<?f|f!8YSZSqox)jrLwD`Is!B`}M-a^VCH?FGyB|
z86pDNr5zj&hP$jW!C-=Q_Ro*6Hw=6N{QOjNmM;<=$P3e?`Wr&}F=rLy5=5B+GCimi
z?v$~-wG438-U{jk59xO!<QaPG%k%A@*sVp?YUca-2oDaa1GWniZdSfFugbh@B!~%*
z$|=2IEXV$WTleIU2_dzBXXi$9bKc;W(J$i;ejdH5ToQcD>CT9oKF`t{#wc%*wS|dj
zx5b$qNF3jeUgdSKMUUpG7t9w`!>nuWzOZCq7_`a0d3`C{p_jVXdU>*msEU|DVy~dm
z1PZ|fq#>g=Txl%oVRsr`o0m+0%R6D8T|xDz%P7vJ%eVC-tR3Dxpu+R@!3bh&UF@8C
z0TI~zJ^@tRw#`u?zQ_<6j>r>7RzDQ|lo%vz#C;@zib1=D`_)%C(}W+-#oOZOi>K`5
zOOP+u1+r5E>l>3I-MLZ1^VQ<3vLk>E-MnDb%Bi;KIx|pIPbsq<C+qM3KkU7EJd}I?
zKYk&ULS@NPrbQ`|HYsFO2bF5HplowWC4>r*hD#(xsDvnuk|;5eCF>+>tKF2Pv1J=F
zV;QqvzVGQAOv~>+_kHgB{(ion$FD!mG1tu1HFI6>_xtsHy`Hb<>-BzFD9Vs-@L786
z+|~oNqiW@$hOKKYovc>(&wpBJVkBA*7H2H^X=16%49>m+ZZg{b6xB0}+JZT3dy2BO
zyD?MzWf#5wk!ongmt-Yl@&1F0OqL6M#xm3~mohgmp<kqGg?8*;@bFPvU1rcO_ng{6
zl^eU1?x>eonU9!KDP@gXC3peP^W})rbK?}9k`1zBRXgNgN8DY_7!}@8BzkR&=rzY~
ze3K5{1jx4rPE&7SrrB7fMiF{%+<vFH_t?s;u`OqBUX$3hmhigujGj4k6GE2bRx^^p
zdI8+M-?`$`)N&g@lDEU6Thuf&TpOEhK5Q!p&O1o9SdgTwoM*@^Qf4}orVDVFF``Pm
zH2gU_k3FB2MA_Ru=xKOnW8J*{)b%aSmv+o&tNJ-I0psXr*xQH?L!X3Ya7K}WPW>Yx
zJBQ9;e#4vOgzTMx42xU~JEoaQLwCR&>2-8=EeVLN^%Pq?&0m_of7ownX?3Wn+32~<
zyqJKt18Y*X&Yd10w_qX34;mjl067olE$Y(#9*nyZp?z@njs<0zNhJxk*E^;eOj}gw
zvTXLdjHklU(G}K32ZU(Zh}dr6v@1}tTvsX?iO7U^%W#I^B7Gk*+))~SV6aU?j)8vR
zh2+Iq8nw|@#UD;+u|8CGIQMwC8_lpf#6R*4E<PMv)dKnG0VmZF-AT${0fNa8JaRiz
zjOB{VNCDB<HfkPH;6fZ}bF~2^lki!_IS|lY!9I(+<c1E^ldt2LKC)z_)^r&=slGnn
zTO<6maGRQupw6^eeF>LetQ0w8;4KAaD4R;Wk}uCz;%uZ=UUdsDn0YWQE8j9$jeSnt
z`LMe8sXpZ(Q}cbM{<AJz+tPDFs2^0k`=X;^hDsRcL0%blw6+w*+ULkRLFf*rOYxW#
zz{{B6Wj$+4?+TGdu##4<+Skb@oKI-6JhWZqxw`Df&mp#Yzkgm$UfmmmT<Q^GMuj2!
zwc&mTG;7<}i2Dk4f5$>Fn3?d)3OylEchoi)uo5>bxoO-ecW!JaH_RoPPSAMuF)ySk
z?XLD8;WlZY7d&tSsvQ^h#DJi`92<wF9?}rQ0+%w~Jv8T-yc?BtWu?Zd?b8PQPB^Qt
zg$JI{f|zRb28J3fycARoRnGA2c!>y+M#Cg5Z9ESr)b)d)C|IuHljqxn)FmPDf~#ko
zvHZN?G&UbU#|Op0wi7P6R|e{>&txCJ(yq8;<(GI-N4ZY9<`B-R#93E2nL(j91*lq_
zxpw6-=D}!NSkm)sL+>&oHQ-BpVoQl+rF>+c@$+oSZS#StXuaC0z*Q%79pKF5<`7$N
zqoqdCGB~MPXWi4Sc|+6EEgE-t+nsk(+I0kZFVwwdrDdq{zPPztx0^;|?s<kIVrpz0
z?s)YBs^m-f;{G0m!u$B+z&F6u5>lP-9cRCrxl=dU2OXr}Byw=|!qpHB-x#*fzatxz
z{p_N42h4Yfr28zaT7RT^D864r${T;;s1z{Qr${82Y9%S!H?MF8<50(kMe24BuzeP8
zLy~RbCPDH?MElf_5`c;GbLTN9x<=G(B^tEhj#?wFuvJTL+Os^bS#7xEylv%hrFP_@
zRY%QM3`HI=JNaxK4*!6dAzCCk!>1ZOBXjWLr*-=xz20sNYCEt#D{60Sm6g#QT=jEN
zdBsZysAayomGpVEhA5gS-H37MiG9?HGbi2DHm~Zi|GdhrqVz#hagtMY_uV+ZjSETD
zHXCTiueYK_9q(=+bsJl2ZApJ#_&jo3z`>M`wQE#*&@YL%@xsMyZ@6fkF6N1+hXM7@
z-St$Zdpd8}dG>=Dw=c=<bysXct*tq0AogZi^Mzrx07Iqg1l)FRKC$(|S__VHI@K0%
zjZPlAPEWAjG3|0n(~h0o?dqD*EdAoBACz}n5!sugzlqFP>^5Wtv@l_=8XBtt6thG`
zECyz@=Ldr{SKB)cP(EXVW^Ctp*wQP-Juj>m-`O>{X5sUvejEG0Na2oIIm_0pSY8xp
zvZbq#XVZQI^iB>ny8~4+v+}adbP{gO;nVTTvSSLV(eAd=U$VEKjjvQt+faQ+aL=Y*
ztd9?wm}U?);NGgZf`fLIi0&ZHWD*jZ-(43kIpcdY<4)>!Bh@Ei{%L``)*Tfr!~+BH
zmV05;ai){GF~eP^CANRZdMSexhsybB+wWbk4)`LL4hp>HSdjs4x8l*%J@^1*W`o{B
z2Ia-xUM~rMXOel8Mlxm9l90_Z8uHOs7MUK9J-!gyAxK=WCyUgATg7~Uw}-W9lt^AO
zzpG6Bat)WfDCto1&Mazqd4zpZqjS*qz~6E2`~`LrWJF6bN<2M6>-|fWZCTOOzV4Sc
zv(`QsDyN?h7S}V|7IEVXeNgbQJTr26xrv<N+c;mz2I7aXt=vmTFj9<Q`r0dNR2z$u
zE7Bd_8nMbR$h&XQ!I04lMcFOMLpq2UV6axKqk3>XsM*L(b)OEO3i#Q)K=d(5|17L{
z#(DgJy5Ttswkc69d%(QradJArG(<*b$^MGH;lpz=omvBxz%Xe96I<^UO4A4dx}pS%
zjZ*gQ;^Wpk-x%lSq+i{l9c^r7q-dy{_U3cpUXru`R~&S_`FZ1||2^*ujMx5DNb!AZ
zGodbQ(_B(}KlxeM)zrdi<Q98GtM>OGR#E>8`%c+#<!2>9TRzq3r>mw)wCh9R*l6_A
zZjKDP*uzJ@fQlLLXn&-xYwY4SX!oS@xa_C%$4yJ-25<BYJ@`TEeT9&qfKxHPMT{}l
zA=)p+UV%H+aEzuAsyo9urjC6zyV32L{qW9rug)1fyT7xWJ*1{p0G*R8gL!$}WNOn(
zt;R5a-0pdyID^kxA2J5CH|^Z9K@X$dcFk!1?;WR`E#(`ZcoO`Gt=C~;jz<SJxNs#m
zRo7K0A>_49)5qqW0pj$8q+QCncNc})hIqu!DYD<-w{tb>o*vM^`&s&wDT&v$0g2l&
zl3lh_MQ*}q>06M8bb#JSh1r=Y-7iFH*T2&;s7=Z6R;W>&Idge~occh=!Fxf<atD6X
zmoe3U0a_I9gbM!bjl}DZ^ndF=(Jq%uh&b*fP+)lOU~k18s*-x?GmBv=+;F`Ol%A-U
z#5}k$!&Zy9YP#W}m|f|q(?g&m>6$*#6AhMr{r#iS)K8uRJYM%tt(xyw(XZqGx0)sj
zjo1G;j;W&U>YrOeuT0C@|H`?m?Lfv!lj~{k*G}&U8ccrAD}}BQBqN1D5v%V74WSs@
z0VBeiVKoksL1sV!-L??;v^`5*#~~Qcaa0O|K<eqDKgf{BMO1PTm==%B6^9W&bMzQ<
z94brei4za|b%|fk;@7<RwMPCqdm-t-uusC0$DtG4+%C(IyL-rMKGI^DWjo4F+a%Zk
zZWK9!rx%Yyb_AO^!F-qY4UZmu=%dpPS-*1*ftK1{zb2KF^I}O1eJ7P1gc?!=`9{L&
zap+zq1X?6r1M&38+?;WUSc$BMUv43cLsdJwf#e3wXxJ<*0CG+<L_nr-iyEHI8HYNI
zfFOPx;!aD22ZzU@5Zh{fMbxir{CXb0=Etw~^7rfqYAa4JFR!4=!tvGJbGqG%4~~>W
z5^G(pOV5F}g;?f1OC*mtWHBNJsTJuuK5?KIF3b08V0j+FYD~Y8p7hpX544eDGlt#d
zIgA+Q4YEO%)BQ-_wck2rC%$#ao_H-6vi-&>``#fdzwf)@*Cqb&Z+|_DDPRBBv-tHa
zctPAx?1f%p)q@*OQkY1yRf8^4>(;zdj#Iwj6>gDvy={;Q*%L<xBoSXVn?Dl$;|)Is
zp@dBgz}_yL3A30Ot^lR_hscx4%hdePF8yD5Z^ey`Aqs}m0;b&&EDzJ)r1HT$nzNem
zj@s0r9zai|k`F3mC>(wHse3iXnYP{hlrUg{sYctbo-#Zm4taUY6yb~*_*U#iA4R=Y
zjKIgMt|!d6A-S=1WyzamXCFUM{j#G-R|A^sDMPZih`pPc@R6eJx~`F2yr5)dPjy<T
zsl<c$0QZecRA0=g@;xEQooLAYH~%UX@cX@lMGVt%Xbpya7}P0{2qpqHinMLWwzK5X
zFX)jCgfj$j#=UW9QJxEH48>A35smp<M6Qjc9WAjf0ra+@5dF`x$~k7BD0JvUK|zV#
zn-A&7j(L_~hBls53%|;m@3k!P(ZQZZr%!2*cV#u0t$w9Dl-&}#{N}SXy4Z~M0$xF_
zK8iRD)u^;m98~37Q`sg?m>ugjKaWl6&K6hpY?g7JyJ6u`@fWt6<SRA~CIK3WMwshU
z1&ijR+Cwah3RP(Hz0RiG15|^(jls%w7lY~sikbJ<DbBUtdOz)~GW39^uk^;i6Mm`D
zsL`fXt`=fg*;baklv{RF=F5^BjVHv)Iv;ObyVy_8Kh(8P`1IXP+-I;IIxx5MlOjpi
z>{;K2^mDG|$6PdPN?M9{%w2!Z=IXs*p(0SBk<>%j%s5T6Z0jz4CzHR0O0n{{NDs-{
zZrgn0x0cB54`wPK9Tm8F^vqh-#2}wFoDEdW)%9TljAV=5I+mm(H_mTA8e|<Ny(Ktr
zSJIKI%gvPY2!g(wHv}=oiO$52zK-!*l{8+dKO0zF)7^jX+0}D}HfxZE4Pz}qta41<
zN_R&^Z;bKt&@b=5XhzE4|DE<q@oMstRa^E+Y?-eBWgE!Q5(gI0I@+TGaOQ`pj;(eV
zyGXcWPd*x*r<3&QksF4Ml~wyUS~uOd&E1zqkSE`W6YA5^Q1mp~O9S~`zm=Nj9YbK_
zD=N9^JA^RP&itON9eQB)o2#rai$1j>G+nL=XWoWy^k|O^?1<HOC|sD4rUtt$crdqF
zv*cv|f*0zq`t(_1+*5z6$Wju=0I}F&3QY1qjBPI1z*G`q&;Z2CgPh0$L0>WEn{xb+
zCK+zh;7VtXIN?5bR5BF?i33qckOkEFw_24G)T!7q{3ZNsoGF0;<o4x_qm{F^S37JV
zxSV#TuHAmbWwbi()Pn6xXG+{@y*gUB2DDY=2(-m+E7Q{`u+}lE=@@plJLWMxM@;X)
zthYNBLUA|tKyeb0L0!7?9ViKoKD7+rp)sR5v+MffMUjoigAUH(b?IMq%m})6IznD<
ziJ!RJ2*{`4UOo2tfPa=1-u8wdZRPnyGa;cr=Cna(j#NSO&IKKz%dS}aT~fRw?!Qp>
zm9)S}i_8m-E+e*;5acsEu-rzhPVeLWw@3Htlu?~dgeuQUe|$Y-bn|9PQ6{+wPd`FM
z(i@jn2A96QzA5yOlhZ2ag<JEby2|Cvm1t4Y0;2^ouNmvgmF6<8w6S4nhKi~G!J5><
zPU?oX>-C@;#p*9lge2PYYxjW?OwkjhyUS}wVd>IiXn*S^^|B|NrC!}IRK0w${7UrY
zjR&*u&L~?h)N+DwOIL_(i&UcL;tb{m;5N1U#zeefo~5-EX(l~Ca%!53s%oU|*nw;1
zaq@)39w-%EN}ioB+B(xSHUD(GI$A#n$24U1sllrWCVqy(mq)BfRs(7BPc5N45@0Ke
zX!11laVOh$Qm8JL7D!K}a>f)g-wlR8?2RHIVdBrNi~PzyTT8DFnWtmepuBFruOtzN
zqN%kFmdJe$fAMK)Z6-h&k%*J2_hUQ6YrW!SchQ@L&(<?hY+KxNYE=I_m5X}19LcMm
z`z^BT_tjWho^^TFtY)CxboJ=9d2^(%8_Gm5<vWpQ2d!>5DV{G2xwo$DhBTai+gAtk
z{Jl720UAX7tE9J3A662s5CTh(Z~`6?uo(R*W|IC#Fq2cvIE~`0>mK;njtv|@!rCf>
z%({>g@49))`F)~?v$mX`8%B@n#U@68z63C^BpFnEfR7I2*qRjF#`=XhD>tH#M;>5*
z=nROqWr^5Q^yiZXjPaqUjO=3c$FdL1#yz)yz|5;L)ckkhEcCtE@v3;1v)af5C{FdN
zzzad5Hu{L8V%U|tI*e%cUI(hK3ut3A!d+=R){%Y@Y~^t6)P1?3jc=W>@&9&WQh-7h
zBX$8<%1JccYqW0&6@VR<9*43}?9E=IkW3p<8r`Yo1iF9$RmwJpG1c$cq;n~M1Ij_G
zuB_OQ@|PM{v9n-@1xg#{*pJyibog}ea~t!Pgm&~o`&yN1126nQ6Ky@?e5(d$$tQ)V
z0d0KAihH#yOH^EUxNJ~5zRUZwTM1}oX`Y8vqi5}J2LcAXw5N=%<T($AnvZswpL2$P
z@6PHrc>wg?F`VbzXiarj`VC$deg_RSj$PMU7F2m^Y3JrTshe3#HY*A?Exd|f+9Na?
z+)5};#$F_g<*%c%KCBv0dN=p1P1u=nDF6BC{+eg2S)?S8Dw1K&q#3m}gwZgFn1)>&
zYVKBNH~W=NsV;|SnH~>qc>6Z3P&<BSlAMP;f%KSDg&GRq2I^!rxX<DwcXN{l@>ST{
zNM+#~l8@AhNIjW)-L2~D8=^4T#z$Tnt%)o-Y&aT_&{%w}+{1#2r=MbsvgL6ohHb)$
z+*`x>M(!tXIE=0BJ%dq>T0862yjvz+8(%Gt!#$Vpyn$&!XC}TJFmE4w0ncF=>u*_=
z8Z0w#=jhQ_n#osmfe~={ZB}s8O3-GujlIFA7A?4xVfTQst^H=Ck}S<GU?6-?LE!%E
zgS#G9<1(-5Je--H^eIsSwM4FHAxamNmy%yvYtU<{g+%t|w+%k?X^$VqJ~`waI}pd-
zU7YSef0y)fjn5k!SIcP~UtFL@CsEg<f$GNreNk^(-u3p(hTn%)Dd?d<+1-ibx9$6P
z9ro3v=m~J|butRrzm>NV{e0wW2J?mDe^UmfgioxO?wEH@+_^1YY(dt8!`IRl-NE_g
zd(b;*5#7b)M!PnRkbWm6WVbANK;32b($7~`9ef=e+O_$GQVT{}fHjYR13A!k$-p7b
z+8k=jP3@Wbzk?<S2CH$;4=p|kx~2+9-swCV{tkBmh$p&f8=iqu9*+<`8H!Y91$D!d
zBiYB1>8ZOe#P9w1+ta)YeO(1upmxmG=pT*u6myZYS6UiuT5w`!ll}GN;8!YYL#4_`
zkL&G`zbz(m3M1b^{2g`4W+~O9*xIh#W^MNQ&6ed+WS7|$2@9M|P%EM>BUV)|@r8XK
z4eF!L)i{}yzEHdj-DNC5$Er&UFz>Oo;8!#pAZt@yKqc%P^ImVC6F%?ALpbbwm;$P;
zdw(-!VgKmV_b;wJCFVEo1t`mGBXELK#vyAmQynzvmKlx;Lw7nBg7uGOt#IsV<aa+Z
z1S<aAIObmwLkKVfX@+Qq9WZydUt*>1163DwtmFgy%v}VfI`h&;q3{n7AhDyb^m(B6
zTbmCIr1sgUVFZx(dUD|k_%QeFHXSZH>WfZ2wt`Ppd^!&IULow^<`l|lmB7B*1)nb-
z%XFFXv_X9*mz&kf?3><w%lN5KvMT8voW3V98K`12WaHUeJYz&=dv5#fX-$gIkT55U
zt#Wkpn!Ew=(72sBUA|}xXH#77T^TA8-4>#tv8;GU)}@*jbe{C|IiN*lN(|_p0h+2F
zcJWu5&IsusUeQ}YqhA-v6S;y@?@uAjvCBH-d^F<WwU@2Q8xM9s*F06OwuXJw<^ZXo
zpQ|@H7)4)z;U??LqFua?ziA!pA5OrnW4Jp>TzJR`;-=Vqs@Ki1T&>jZrcZR!eUg_}
zZ>+KHAW;Ga?9~TMTls~4j$}f~D)o*Fwz)UF7ks`kv`c(w&pY{5R>S@;dLT$_l-xvV
zLth5HcoqFksUdkSeV^tQSpT+V?z|XF?~wWTHRnfKJ-PK|_J!~ZM#bpHTKW-Mpr4aw
zrPQZhJ;jE_0X}3~55aO}*CeIigFCr>Qm>_7ElCnx2;oXW<&7bASrpsaGwdZw#Q7*n
zGT+W&OqPD5jCR?1pj~Qq@G`N0#m8)#rq3Uc%cdr_p)MgRJ}t1cJ9?hQviOG#n=h6b
znGqhv>UE)yW?aY2O453|Z}oQhj)mNLY#oj($!8<Sil$!fk6V#=kh8X`K0dQP=3QIi
z-43yPGGiB+Lp3r(Lsv5N<O6VzS}Q{uHHtKvJla7;K22F~enaZ#XL)(s8x1nT-<tUy
z67)MeR~{qJOxK>Hk2>VjOcd2cwZj*@NWm#1AJjDmXOu-ndYm#$d!1I+r+hixcc<K|
zg>JJ17N`O3{&DR5K2U!u<f&1n{q~ag)}a!k(`sjG-Z>074{7CP91F`KTn8oPE{z^9
z^foe_w2H|7F@B6UCCGZulO4f(%x)RXaX<S;!0(I5i+JUukuT<Rlc{QL{X{XKtKQ$n
ztyOEWmt7tuEMY#dR%#_1<37CLsrMRvr8KLsP@G2_`qC%40$NyzyWB{DYLUl|m4T1Y
z?sJS*^?RC{?AR>qW)Z->8*$k6vBTQS>MPsNJ-lAnXQU8$T<?XTBb8{%piqhQFh;#=
zJEn17$#4%|g0p!MQeM5-uW#Ggyvt@co8m976zM&9t@+|s>7&s3!9X{xJm_;G>o$0~
zaQV|H0|vTH{zGrWvsrrg7C&7nu~sg}Wa;la?}w?+aC(OYn~EKeW}gB3<gAKydSI`#
zOxA3}=_e*d5)t0>PnY?NB_%sUyvF|N>)90ReJ`8#r@?DKeNm2CR-|~tU{<2&w4+Lw
zeAWL(&@C{k5lHM5cnG$s$Uw886t4>E7{_-#nMDr->3A8!SL28J*{**77x)xTIrk9&
zwG^SU7?sQL@xe<9$B;)I-ba`tD{LK4rIp#<KH9XyHvjoP=?lIo`PZ{WN7~Z{@;ApF
z>f28-^K!~i2q3YVZnYW2V8%M3bw0~{H{$)#5-X9pxK&hg@td_N+^`eZ-%?)O_&lsU
zBxx5B9hgSg^Z&82+`ke4jt=s>cIPBt5|}}7lO*?UftRn?cW&MJ2v3(%Nv3!cOnMG-
zgN?qr4S%D9kYuDD8fhR)5bc_3LFLliCYu&Yl&m1@{q)5mEBy1X-_J<!#oS$&=cm8W
z*UMazEyh{dkgrJX*w66qP~7j-j=gN2@~C-<N@VXzV>#(W>X%hR0v|R`6IAOzKPY)Z
zAz`2};ZbiE$>S{Cw=72b#&%-!*=dAUgOetrS8)5e_Yp~)Ox?*0{1ppY&8kef7|+=j
zR?RLUmiC+cRzTfp3rUf1v6s(l{D^ioGE)<U7!jU;jRQ1Hj-6+3bUq{^5UF`aG_F}3
zC_pYc7UM_6K+P_34ydnyaKMie?`5mPRefZ}$~u^-R2W#j9vyghalunYN_-Dps?Q?f
z!M>(Gvh8mL=MTjTi{7bNCT0?L^ZJFN1<B`;IRpu~O0^c@7$Gug^X=!Z+`IfYeZeF-
zNWxd3(HCWNxSi~WpN>0+bH~atw06`F-gc(k4&E4ZwsEzParSvgWc41k+Mc*slH#8G
zi<M^8)c!^d4Y$AFOS|#L#ZJ(&L~c~a7CJh%dAE;;Pc44P3oi(N0?N;w7Q|hu53*#n
z=FMrTbeSJsJf?h9xRzUEXw;^!WGnf?1vC>_sI<cWMea(k$Jw`5$-abM?&|gwIEB*v
zQa0!{%0ogh3nc7zQ_cO|oX)UISbF}4ZJ+ervHn!#C~d9<U4zfBP4egz8kVX^jrrFj
zLS6(ZMCO&BRpgRKL7B}BQk*%Sqd5+NuIrS%acC()KTfdl46U0J;z6dr0ekAGdn<mI
z1={p?$WpyI+(h(Av<PQoU@3ZrTd)pga6!S-XITMG#ap__FH)p$ztVYmWlN)U<l$_{
z)h30#goDDjW*3RiE4Fx`WS9--?X8Q-ZEo%ua~hOBykv7asACxsVw~3xH?#KHz2Rf#
zdTYO)mzPcz`t5Qq&vHECt9OLvF>sx|i7~GvH|(PB{`P0blI{21T6LH$JaSvvCLVWc
zEh<UbeA5pT?eT5>s?=llU)EFXkGV$L1;}3+RlXV%%s;231%`>2z_w){v&-}r*Wy;R
zk!5ce>^FZN+R%XNXZLVkCH$V0cURc-)#ih~Ix%Bp2~V}6LJ2mOvshdFQXrkF+k}7G
zSnAZhqq0)vW^zG}ZWcz<M?<n_!_Wh&eO7hFMVVs(LW0_}^(!=iB|axhq>8qN1-TnG
zaJIDWmSrJEYj@~Iu4dlIbGMgUz6!%WM6kZ^yUr~{aXV-tED{#?q3GcUi_{9If%t=m
zL%d~X3-9e0_BnwQtR4xQ?emeSl;Q1FCcg7!-zHhd^S2^T9K~jooa$*!o<(0wG;w%f
zCcD~dhlg<f=}Il~brVsSv0Jn7Wxg9h@wXCh^GudCrbusY?Rw)^=?dp954J}+x|_5*
z&qOK~o;QC0;x@B_^k;QXx@$A#mgHvShTM;lKcIWYu)}dl^3k<Vm1i4HZzV;*#d;i2
z0CwN84bPD}1k)am_AM8xfZOmsf^fAawGx@}gvhA(dRLGyd(FtvQK5L2>YO={sp($d
z$D$Aj-xK~GW3QL+Ey^8IQW}S@i2#q^1~`*Ff|)&E7s&We_W1+sY{abE9uws`8CO>w
zXxpNqh^i?N%OS|X6?$k^A9_SKPBIq#EK`z6k3twO)Ofaz$vA|!hX+AN53!Ng%Z1(o
z3Vrh9#AQ*BDvFk5MBf?_+@QiVPcW+lz}u+-;t@dr2fHCpO0U<Q-RZ`gt{2QYDLM2I
zwsSxY4`^_y<Te{pJBBHy#ZKlV3j*Q)H($i}3kbXJ|JA1-?7gD5EzG!mq7|>6hEH9A
z|JWGFVH=M_I4y>G#ui@M1kdWr4AfV$<Hjbzl^MjbVmJWBf7ZjjfAOJ1?(qlxGU`i@
zZ-zpyAP3UzEgtO7vfOlSv?$~PVoHh8;smG@Kc%1mB~oF4rhux+b^1!Ezx{RmKXjwL
z*KIZJ#$Bdfc=f02D5YMeP9Ncld3#x<K&E3NdX?{)8qv>!M221s5Fd&oGQBrVM595>
zH2KK?;^UpU&i6}H8(;5eg+eNOZKBNG$js@Dz_$WDAgThOa|dGfB+n59H;o@OEM^^m
z`5$1OCjrp|-|fmGsr|*H<iEEc2H|9`&BLN-r)7)0t}C7%ZI9-Ag6H+X%_u2^ab2<9
z2X6{=@qhnbzVp0}N&m%%{(I_XyvV#6j(};Pw|j|cppw(g-n+Ns4|7Vrya^m!DZ(0u
zwEJ+DV*@^6+~q<q_~-mzEQ!DPI59QqQQ=vOT}_62me^kC^Cav@d8f#=2CAv^n;}mk
z|M*Gs4XjPH0dx-kcR$^O&DYBi=&G?-$k@2{<DTf=+zqI=2eAC(S7ozu95Qu6hT%n4
zyZ?=^0zWg|-s{fA%LB`fh<JvCp(56Ov$E+SqKF;{@>3<bY7&37u>YY^ScTKbex8lg
z4!55eJGZg<ft@1Pu|X|_8pSC990Hmflcx%)mFg|EXu&*bT0cAA*YZNdK0}Y}g!9&?
zY*fdgupS}>kgHW8kf+^0|6pa5z4j!pSgf|`PH3L3Emy5ZErp6_vBSoUhT|kfwVm%j
zNVa%Az11%Sk;d~s-((=rD<3WAy1`uz3IE>nO{*EzG(DSfOU-TBhNALB>B|0rZHPC>
zI~FQ()EYG<nbUglJK&K;2i^W+8T`WqRNZUwAn~M`>!$gwhUKh%8=@aAkvYHPKxkYU
zSPiOUC{WW`&LNIg0KnOr?(V|mDV_^Y2vrcL(BUtZ!e4xxxIkn7bma}<!p?zjPNK?X
zdQG!|h<c$8n$^1$@+|-7hwLxb(MV0^4olvRj?76zzb;soivoon6}#Z)bpq{S4;}WV
z8D&HtbQjqm%G?i6-5bG{1b3>R9!0UHKM8_{(R^dD9C=@d?ry}g+lVg<*8bIpF8u`<
zyJ-53EIaw3Ci#)~-=P<p6I=}}a1jqp4_Ez5=KM!%AM<bmuIapcjTkA|v*F{?2d|6M
zcp%n)m*+-gpbdH7rBebA$%OY1U;FA>ezSo8HE8~J1;cckhx^~khMyGW-7BlTU|IHJ
z;IsR<u)d%p{*%`4bFHSz?^o>Ct?8L|5;H{DnPM%4Yz4~JK~yK0qDN~ox8FXQ<+^BO
z%~SO2Zy?6w3UFVP|G;mnfq!qEKCKyD`*4f>8q+>p){Dx*?*`Tat^kxbw`UxR{kO+;
zFSSqpcPd-CqB{F-_pIrg=QZTL?#5|=)lv&6TScZ=PZC79%Q<a#P9w<WXwZxWm?*!;
z+tGP8t>#q`U3V9PVJ|4>hyo$6v_J<xEc`=QJ@6>LuV&Hpm5Ij|Y!A7N3RSD7WCQ=R
zQ`mhp{OhHLyZ>X0^m`~w25zNOmhq08EV&P;&%60o9KHlKG#mXzW$4Mye`j4@9`Uv?
zu~1vlEb#!?zwf}(M-JeJB7TfWw`l*)CH@Pm_7@)~>P0;&_oaugGBmV3e4SELpqe#W
zIhsWr9YldmfaUn^GmOvTsMRiylU(=DewYxpz5Zck#4(+Y)DItk?8PR`I`LxRXHq>n
z1;T+T$#apP{OBSlzCzLpA4dN>&YeS}%|qXL8PVHjT`enKQg`~Baf>$}hxR4{TPOTP
z``;Nj&yiTOlb4@7#JM*qqB>Uu?yfs<&FR>*vmZdnG1`V6%|;s_nQuSA<(=cuXf-hw
zI%obLo2GOgG#}5nCDOCY$$x}r-!wItBE7^f=wWFAka*<t2>#*H`iCDLxZbP%2Q)8;
z+<>krSe%O&19w9ntz-p3)T~-}=_4tQ%a{1ga>9S!!S5G;0(9geMLesA{66Vliwu4O
zC9xX&#pXBelF3D>KRw-Bs`U*>(9aW@f(PLOjp2W55snN9A6l=zX}-*{tc$Zf7haS}
zIWRKF6y=2^$^fJctiJM0KkfaU>R5^$k_<y~!??o93H(P%LcTl^LJD3Iz`P9^5eB7L
zF=gK#LbprsL((o|=1enT7tnogYY5T@yq*`6(jD;%bf*AtN36}BQqD*+A>jOwOu@XJ
z;8|;{iJdE)kkJ#2L}V*Snp=^WG4Fr!9@N$ICu6JT{07m_4QtTgily<r_FXe^N!2eP
zz+a?@W%c1%vmcv!06oOD5CYd7!<}v-G8hEPFZxhp=#(>=_57ij@SmAKMK3#!rvrNQ
z9UOba>=S<=HIbwf*!)A}qZSDW%&=QTGPc_q6YRq!09PadIz94d1*A+)Je&;5KL~bD
zw41Am^}xj?J*yH~Fp+HB<U?iE+P@hKYyj#HS@<yk;B&-lsmt<Q_zosDi(2bZ3??3V
zjvfJ5tAabaU{)(=+V%7+nemm-Nay$MJ8(<KAq|*b2@*4@uYJW@^G(MgR5eUm1EOMu
zK}SEDIR`cXPre+vctYqFZ^%=#U&53x=oZxb990TV1u5*{R&#oUV!<thQNmaNir?1n
z%!a{bNT^Y8tI>~dPClHRB)CFl7YmRI_r!`*-fcs+Xxr6bxzo+SE}VZp4mU&wT!;x%
zu7kN5aI}v+#H*aaHDm|C9SD}y30(3++MvN~DokE45Ia%$Yw3X=5_$##q8nyQotnXv
zl&C~<cj(LAoOsAp3_E12Ew?Pyi_bGP2NC17%<?+8XK}h3mhWIvT_bcFK(S32QYeY|
zIZ9-s5Cd?36!1iQmhnAOmtyjeY+w(WH6zh84B49QJ&(B`w;9Z{6Fwolb>VRtTt!W#
z8jgStPb78CI<Z3<@4(zI9|9v4j2JM9l;_9+fxwAdvdC581UQXDpU@QzrVc&$AsNur
z&vP7FqCXFZ0oPvi$9m{?ybpGTyBG-K`7X!UsipM>!xY4hLj|Z|aCyL`UjqZp2E(CU
zJb8}fRw|L>kKwwxbgJD?^>{fB4T+<HF~61?<SRCenGILibRa_D27*R%zd3};zD@)(
z&{$lPAvIvQ)6C~s7Xr>|p&H={o;5vAVliL-LYxS$q(2Tl$G=WfbK&zVCf4KtXAPhv
zw9tcaf97lZ*97=vs?dt-0;1Op9h0f$98WNwAQ(+2se=qvg&_&aO$0c=^rvCzY(`2e
z_R6P~RI~85N{QXa?m0w7uGXbztE(BgtTRk_-TXLmp=*H9W++t%QQA$rjRe3mK~ILI
zV2v#$y+LJQIZ@=xWcunpn6-bY=R3$}7<rA#KnJ5wfF{;u>{6ClRHAX*n<Lj2Xv-gI
zUNtKqo0<OF^TQfH?K@qC5rp?-#xf!YIJ;74O2<yn!E%(fRA1PMf~Ppaluu~N!HDcs
zI0dYJ0vNXx=*Yy_o7=Z(<hYMwC|1^I@6KgMVGke~0_8itdIp$quC4d(wNndP6CPE7
zxu-0^dckrH@nF<t`KBHL#LZ3k`{R&BmkYXM2<E|3uFDMYbvZ1J(bVI^2wDX!ix3VV
zG$$;IvjcWWycTYTyVNB3p>=ZVSz`jgwOm-;T!S<9b&UN8dKtI|TxTM(iVyrux%-i=
zTD%DpYce^f#Pg;ws|n-~t}P8@10qekWbzEjgSEs?lq0wgo_{NI=20(sND@4w)E&H;
zmh)$_@hANg_~GXl812>n1xCNX=oc6T{TIRLjXiN_4$m|_;*`4UO+~ApGbSi~<cKdi
zme{C)7!Q`Xu;tv*+!Z(Sj~Cpg2ZMAfo2H1AEdW@(^C$tR`Wpy8q6IdVNZk!HrI2MX
zsVrg~%A}0!^c8C&p&NMac|P0#5EFRCmcU%v(14MLN_fco6@6=8V$V?e@D0y%2XvQ?
zttx!PJs}kO2YBFBSY@o3W0O9Wv%@&lw;nW9Xw3|?j3Kf@(8zQ3!&MV4Q$(>p3FzED
zNrCTm9l=tp-gR?)z+C_eq)OOfq86sPj@YRP0Qoua_skYdLCSI=i3#lIi>N*)KcXgb
ze1NwN?j|X2mh}{LB(oAZ7oA~BJgj|hQf8;<0S`@kIu3b(Bt=BF2k<P6CC3sF_bZcU
zD5WKWY#Ry5Q9=AvKkxq*Q~s;~`bj&HzXB0KldN7RcklD=;t?q~cdxH)-Wa;PV@sr+
z>&vA1H?~7&vQe~kZR#N-=KFy1zU@HEwcM1fbJIM+KjjdvqKCHWErUN~1B~p|Aq`*9
zN7d>Pz?P6pxgA#9XhOx09|&BkcY0P{v-86?J6BuB$M#VTtdNMls7vP$O{?A>*P%*@
zShBfDlk>uSWD{Ep^eoTjfd4;rg?k#}B!B;N<c0VmrDlIbe;#_Y-ce%GUsyll!W6TF
z3oP_j0Q3!lNt(#<XBj`f=evp5sQPR6z9}7?xhj4X%-O9^cJd#^&DuF}Qr@KVg<MpP
zu&#wWmw-uWy&wD=X@R_UB7+skt)y6`xZ9`!|2+wAMlxSd&M*=FSScd=AzXeb{;T`i
zuyk^gV!-84Yv4lS`&_&%myHD{?EIVpA{dP^5|gyQolJBbfx#xV-ji<333A^cFn}K^
z64~orM!_T*?H`%)pTDMS8qaXsM&l%EOYwav#81G_y^`8A^?K4L5r4;*Ps!#Bp(aDo
z{(SyG={b1@{D#a00msMW^f<xi8ZvPC2zK->++&#6_-zaYb@ANXsr*Lj&Rp^g;1Pm#
z;|6O(-7X;HWOX39=wZ!$DZnqCN^eX=M9vPVp&7M$H+VUMr<6xgd-4lwfJ2{NPW(Mw
z_rlfdp%20DXhYSHL*0Rh9<f7TuFkBIvxG(DP{HL$9LHtbKdn$$AtFCx+WF87=rtHo
zZCLDQi3dl^mqcJ?G`MHbtl3XuEs?F}=)oB1{hu=rO3A}B1wI19v_oxf))%-_)DFx5
zWxh$$gIfn8lQl?(`DSa$OK|5hbh87LoC@4_iw=+$pL0ltz;`gIrWXJYV@Shj9KwnM
zmryg%G7J>yjKHH*ctq-;+hp{Z4XXTRZX0poPOHraCoejr7tDE7_*1%lb<zYOa7KeU
zuM3#)4koZtAd+3oA&)l0s!;LdO--@dj;fh%Ae-I<#OLt!xS542U|n1T|JIT7&NZ{z
z6OU}&hU5^2HQ!H7?9Yju3mPCu3u87caQ=pkp0uv_Jh8}@L?l;hNT%9crwwL`@W`VI
za|58}NW6^h)Hn@V67z@{FY9lQ!A1z6<(at)NC;PmQ7FLBsa;Ox3%X21=S&2P;I9Z7
zTx%iRjQI>KX1gSFI%-UcX*RLGEY7m@;@!n6r|HC>Z2hvgsv{>(BD@$y>u^QJkT+IY
zldQ;FbNhP-TMk)ZWL}@N+a$EGJIwZ=thUl4@&385B<3vGp%djRxfyhzEaQTpqqC4>
z2zGW31XiFXTCL3SBk=^0sX?4W>_1!wdR{!-NsEEF0DSzw+OELU)**%K30wm9z9|zt
z<7p(2wLrJqadGxowhi&OBq*i8mFq;{o-gM*&gEllr;N}9(m4jFPQah<#$I%TR$c9i
zA^3^Qqlc4yz(l$TNIV7%-pqsUImrNYYJsWnYnUP$+!3iC0gkl$Cr}8PP0p2f#>kdj
z7j_t@^l(F00)V9!Ep9n*0#^4-%r}0VOSRyGK&Z4H-MPf|D~{HS5g)@9hTULN3_qS)
z2K+#0$%lYb1PVpLgX2E{Q*HptBGq)W4zh(#9DM~=mibq&;OCu}{({64@?#W_L(1#G
zNCbP~k?;a|qYM|tISf1)c`#(91>h;}_(Z$|+5@oWbHPOEY(f-)UV0O8#L;}X99z}h
z5A4=5GHBXb1PG0I3*XItf4Ts_2tL3+{r`#JV^)Hux@viOnKgwn-jN=GXDEk0l)C!c
zU2P?Owxbt<b%=fjWHn`=rx({4*abkLR;aaM*tw+Pbi!+Gh^0oiuG6%{H=^1`mUW%m
zQa=#9ivIh74SNi|ptat~tmy?$iRYUx8QaV63f(gOlp!V)MuVMLId|~B=>Em%5la$=
z=1;%G=zV~*q-BLCFvcu0($h(r%NJhK`DnJWJ+#Z<H&b!n$8nJ87JXFiS8)OB%~y@n
zAK8LGQZ~FjuK{b3f#OQ<27ZMQ0Fd_Ne#vRL6O7FBY<Z3$Eu=embVp{3j@47Y#?dbi
zv}TtsvAp!#wb<7w&pU;ht3)ZJODU`33a)ntbnKUxKeo_rNy9Tu9mQVFnSr6749(7L
znZ6yJ0PCQB_?jfJ=Rg!U4uJiAM<N3S{686eRMo|&Z-h=pD{P=#nucnKe+ZPdHYY4(
zTIxbp4)<n-ElGQ`E<^tY<_u;T{tpLviUS2;0Bp{m3I8buYFitgIU{ULA*XK=vrUAp
zJ&>&=)LtGBQ<37$cQEBBl>?jYg=R6(tQmk;gqibzn?~^Gk(^V!qBs#$8d{*@;htr&
z?P#Xhs}bzy-9T&?aF3;h_}PD20$}wC-T;uD0TcLTc^$4GYH>YKuR$hY60uFDSa5y<
z^pI)teYnDm%<bTWK-PThXKFC08XB650{nb*>e)$`SW6w<IRi#U;P>F-wTONcNMRds
zC{sT3=9&E5XAXHLn3W)0A{ioEbAkNjo$eOkZJZBd01{ZOhQfciTu6au>hioL5q)#)
zAz+BD49>Dp9N*N!bYP-=HwI1k&*1L{4(~cbNIR@rlEO!d{t?g-DqR>Wgj?zf{)LK)
zhwd&QDLg#8#;W5LC5StN@P3f7!iaN)EDnf9Q$RF=9h!}~gvaAhB`EJ2qY=vr2z2MV
z(G_}Vj&7)NPOPQ%(`oW@ezV>b$rFSK&fFMcXKUdC_``W%%d08+GN2~xd!VMfd>_X8
zqAiH6G2n@+aCU)F4p9Sj8Kj;FpR=I$PZPiZh5?o@R>2%05vsvs`X*2<nZoiDLUSh~
z!aU3j?>B_Uuths1gBsPG%Sr^!xi)4rWOV&&9G;@sXFmmX{GiPrz)nVh3*d|6P(w3!
z9*eZ@9EcslzmAxDttW=-R9OwDxN6~POf+D(yY*!z-9guVVDdTNMmj(z1K=1uf-AH@
zPzpS-TXD0rC&sD|2Vy5wI7~wKsYRwHz696N;<_UDeb=~Sn9~6r4Q7V>Rze?sXig6B
zFhV$e0h3{jQwx|pbBnpXycAIjnR8(!`co8>FV68Qiv{bE4N@6^DNl~T>;XT>o4`E-
zSaj3ZU*nCRb*}iUxb+u^t4&UfU&QFYM~w2ey%Jsrrlt8y{FpnT`xy47I}1$o@7{A0
z81d*29*0y^S|!=j!??1eMVvFdDgv4ipuNm_;d|;Z9fVxr;Z#xXMI$8Gs}<P<zj9<;
z0B!jtl0g>qOqFOPsTr0eb`}?|fIk#J1h99up8DCloXSRa>9j|2uWGv>flm4;2@nbi
za5weBL+YL$NYPnPFzN8l3}AGp1$B_?*H0YTS0D3r^{6>_Bl>6I(VCqlQ}OEZDH}AI
z|EsSwpFlxhRSI!P_BktzJI&0%k?-~7vUdJPLe>DxM@0dOi$igz@r4vljPuCWDlTx2
zx2zwW15Po+fFMcL1yon5OI8y{r1sEIoPeV&jiI<%suRX$GWTa1IE8|I0fqT*U?ftA
z>~f~UYkmcm;NTT-_me6sHyi$FHDfYp&avl_X8<JP5%)mLNc_+=FuGpoZoas>$7UY*
z3UD5<WC3ob?-^6^DroXz4`Rug&F~9&WR6V~{}JFBc;0PO4?HB8%X-E08rs#q)h?Ax
zWW0Ez#}|}3G5nFOqTCY1^iM+1#{=wEgYMkJ0KWpOKtQSq)e4+1z(KqTSHx0lkwQ%@
zH(nogBVvI3%$##)UNM0!2~K5gb}VshA|59&H);f8=}NUA==7=*hQCL$Uxch64}$rk
z(f=7j_9yKn{_DUt4)v3z1L-b|Q-V^UDz0I(<T*uY0ZPV!!d>KMmx}7ncib}IM;aJR
zv-LNCoPmq>XQ6}THMl|$gy<<#Ou5M25a7WXyNo3pm7<5{U|3zmw+&~1e6LDWYtpkl
ziyacK^GSnyOs`l@zCIvuW#WWy>jU;B=elq%g7ooYxM!=c77mzEl*u@h!*z^EO)iCL
zqmpcVLypBHFWFR!?kENxJK(4&bF)S&8PqzMS2w|zaGf#mJ&9_S`GZJhxlq^?$uS|N
z2OLKT0th>#;O>+`TEUZ|-2Ouh-{|>_Ynsh|#|b#C*);Wf{gm|qflT>Zz|0l_YY#35
zA_Wvc7Lq6u5jct03!*L(KMx`w!ARAAG*@iXV4eph@i1HS>;cGJ80^xC1S2_N-07tA
zR`Y_09V`u27kXrg{`^uN6YO3HMiqM+=|S9InOg$(W(@TutXqHn0N4Q$wt`7U$nnRf
zd0;z61dxoWVVrjdCW9Htci`<6c#dD8^f;840_&)9;tBPY-@g?n8BaWlWo|-4@Q2^v
zp6pb~jwntf@iGeN28b+kbr1YE0jMbL4PuNfkp$TwELSbkS2wG$gW(Mbk_9wlfQwq7
zu}2mmQ0)skkUVslEDfl#T6jdq0b-m1OH{JA64>hNj>Sa!%yi(Y8=E=4^jU%coD)XG
z92AT$<ey|E(l08Q-U)JYsyVP`t<$cwW6Z++chxG<ZTZCGVS^MRLliX*&5x5zJwcFW
z7_*IWvW)unM5>D)))c9+w09=fq|_{@Zwn!>KHl9m|42b9NH_s>zqSC+b-MvnQxovX
zmum_vMh{7$zG%AEPv~gk#Bv8Y0jYZNyd3^@3l`tO<hUw^@zJ>M-*LQCXZie$!M*(x
zt~W_YY$A~0Q#>DTm{QX4`SA7C=uR~kgoYTVC-Xrll-CY3XGroEE=k9F-W7fbodSD)
zBDaB`;+;IjYhD2OrwY8Z)go&d*Ft^-1+zBB69C*8iVFM#^8^y#k%@>VZ`Sg~h|hxt
zM$k+FYwx2x4j4K;LHqOm^5LYV&q6Rpz;il{jyPg6t;({}4YAIZmz&k<Wm?%Ef<e6T
zL#cKp*;=wbqBQT82ZU+V6EXhfi`SSN9Xyz<CIL2{GTMoF65HHxM}Yil&;|r?38b$R
zrB{-@WIw1+v*RQW`y*nOv$Bx63`gnLdmB0qRBS#sTY=LA>Ts)R!AM{MSeYVGSk9Tk
zn~dHRN2NvFa6Ms11ulmb%PlRx%y|?1$i^Ojtk<;pi}C#B&`~P1V7XAc7=7?#GO}l8
z8;KhTG8bH>%y!N1f&7FZaR5BTrDN?{!0E#yH(%MTvw6n`?&B+O2C%TU5{bZstTn$1
zxD|7b6czz)(G~nCEdd_z;?Dgg#rf|^ajJgo3H2WY?UCVjaydGPELtdnn~}|MA?O7w
zJ&Mz@x_D^@uI%#G*?4WOGA*TBly^RdL8OqW#ksEWu^TDv9f#;(Q&ks+6R{D@%SS}2
z#G7&Gs(BUyjCy9^fM`cpsGcMhAB4lIBjJxeT~1xNAuCCF_tn!mi_GR<s0!lr5P<DE
zikY(Oagt<9kjA{tr6cY(PpkoO)sY9YhaZm!<cM;7R{=-{DV{0!VA6moc>Sbo@lfMW
z-qfGpGLgLGD?S8@MZQWzk);!-e3N69*_qlqAx7Y-C!rP^mgBt*=^jlr%OwClr@RjR
z1-AHn;;T3SK)H|$;B3Ogjd%$wkksXO6DEl<l}vETcf!;&4Mf`TJT`?xc|>y|XR^>E
zaT0oyW_KdbItp|IS}XvoEV|cJDlJjt6UxGgCs?f#wQsK+H6ss#q*n>aY3lV<=Ik3^
zz%oXXg9F^L6^g~D1wCsscN~f$vh6_Sp!qQ?7d|(^M1bEHupaoX7|oSuJ!Kjg7_N6Y
zJDy2eYV88ZA~1n+(Xo(g_Bcf20?x-8#f*7mHprKOoqRNCEua&82oH$?_1l+BPJi-H
zDA64K-8;T|q1@|mXD3Hs96Ai*z5Ab@qeHPn7RW*n%K~z)*DCXqPX%%4j1L6T_&@|&
z?P7&0*I=r{v(f#B>*0}93uvg39o9C+x#dx+!IpC)b0r!J=M$bIDEL(%ZXYkVwj030
zK$1`!FK?>{;!eK{@GF;ExZkE0rYGPwVZnVA=@UD%dx>-ndm?*ZA%OloVrwjZ=+r*=
zgDjg_j!mm)fPXj*P=(=s;mC_@@_TJ=Xq@0w3KvX}Ne^I3E3I~qTFJRs&JM8}b<>jv
z=yMC8Pm*9N81_YuVSuQvGB|2h$bU4IeFNV&6(D~bJWovl@{fV>!=H!H(922hw1U`)
zy^Z9BeFpU1JWlX1Ogha;ni8GxWpJN7o^k?wS23p2w<H5TCtz|NEAj=$?%#yXlcy*J
z@TG692Kik8NT!Rw1W2aW{hMQHfe@a<&##hAUCc)t>(4isSoh$sB0t!hO(k}33Iw+g
zGF1>jPCu1R025Au7~P26$_K_u8MMsB(*^5@Biqb}b@YY3qiL?luEACnF3vM*95QY!
zTt|4Ow+en0%PY1ggn^aSd)`E}%>8>H{uNPAGBf`RBu+o+PVgTD=P&5S+sO8){^4=x
z$OzDnyI>{Ee*j^<`Gf?tZO_u@+=P6g`D!@<>`kct2W!`%K+~kITvila71*6PDwv4Y
zaIbQm0K3=qzC+k)wI|b(*UdhoEj5}1Kh1Puj-f%=8Uk;^OXp+TKlm(T<jT3{b%to3
zA2qd%)`zS%LQu1Sl?G?o-8Fz}8URtLFbI`OO+-f+4d!NEGB2&lpGHfMiiXA}M-zap
zZM_?lm7EOPa$e0%=_IAbl<Z7dAF@d3)zwy66~V&h$tX~Kvm&xJl00Y@76f-L#`AJV
zPX>N?D^9|B0tB=AiJdc(xWJ39N}kJikTk_Hev8!!--h^*Gh?dC-`YHAvgTxpwEIRO
z<<yI(wE=4jlw5rJH>pNmWQ`pp45ucR5`9QLK!TCuh-EN?xub}C9V2_Nk&9xPh@vKV
zhAF@XO7e*aA39w9=dODDqXs2YHrye4dY=%=pxLpKUB+%R#o8USre06^7Oubz3X2AU
z3V*=zg3Ix#rzR0pM}tozm8$g;J2VE`UIMqxp+5<zGD@*D?mVt5iZv&d&aa>H-SjnO
zHd$FhxiM96;=8GdKafxM;K%VvBaON}4pH}`nZmp*j)w(u)|6?4O~!x!mOjepFR*zK
z{g1+a=$zB{H1*fj@hzEMCqVrS3_9vZl;$bGTH~-omLp{LybLn;5mv%Pw1rIFQ_0-8
z0^L8;o6|w~(8mG)(C;k#s%Qu3XK6Xuy}+m3MBrDo%m0#Umm;5yJ|z4MJqb@1EW{vg
zZy;#OE%ZBnqfF=Q#V5%*`fD+YM<s~r=Oj70Xgc7C?wv0ok97e*_(qZtWpEz+3QbGo
z`etfyXV-JylfIBI5*_dX#QAO-N(@YqwoRI?w|$$p8Ix=h+Hz!#VelNNbbUyk+hZfS
z7rG;ampdV~{>otdkR2=l(#9o3mQ6h^_6NVfS8v@931XWiHZ5#TFzayIz|)}PfvTs$
zR}HC#hp;|NfN%Q<I1jrbIyC#qKbFvB!apX5Ilx?huK_y%nSIJZvPeil2a<Wk(yN=*
zgk_C|v8tvbpo!v`%u>=s#f<)Zw?Fd4%&)4iKT_wsaL3hrs_?bBejKW^=dq4_!dP%8
zGkz?Kp*;La0gbQPMBc)TPvrHK!k;|WmvaLl#@8@|uR_LLov(HVI>MhJ#`4R*rubXn
zK2O%jr0^BbOr9|s!l&|5y&KDyJpP(27Z|VkuLkJ<tPk)AK0;t>OA5lIQWA1wH&)%e
zj+kc7i50zornFOlglrioy=J5UipQu7O_#4jkKChNNM*NYb0Sc3Ku%@{XzrT8R1shF
zQMmOEe#npmM)2@0;>d6AW7a%VKCKh13#|?edsZp%e3uY%dnRLD3{W9nOUP!|$G78|
z(-Xg12b~|^J0Wy?GK|R6U<&r34-!X!7Gvu56z9p$!#^@@(L%OvAdc3fM;7TT7OG7u
zoJpor%I8UL@EgFk<SQAS43DPKzYnrgEl7wSa4EduSa&^jHl1jM+!QFE0B5sF9v)0W
za9`#ys0XN=nu5}w*~OIIU&C)`K=%;xxQW|^Y<(UxkU#Z$azgQ!g;Ksn@=2#i;tfAu
zCx6!uhQBJ*v49`T<0Ovo+$4U9b7%5YZx;z59YN5kXi7Zj>}0Ci!pG51J{!?4@UC<v
zvIQ#!w=@uABYgIf`ntb+r%JS%{Et}oyvmRBkUf|eDEb@}OP%7+_{?i(*DeaEMFJA>
zMqkMidzzyM{89-92%~HzfXi^N$-FpX3{^bHTs;_srLXQKvKC}QxJ|%+5^aX5CLnBQ
zxp0FUU68u3^gUzwJWeE63n?AY0{tn?i*E{yd2pnjA$!CyY(X--kwVud&WFE^!|Nck
zmP(E<2*fE5BCbrqxXF~wn~JoP@1{=h?vN>Is!s~oV|=P-%JM&Tr{7AkQofOCO^$ri
z&qP?{AH`bwat9`C+hmBHEWyXQPq~d|saF6SU<r?f4QWR3m8@0?lR;jlV*A(ngPg{m
zd<-~k(O1oXO!t%{$uGyso|0)zt?a6N4PfwUzp7@%dXi)LDptWhe6^sObRO`de1UFY
zalxu@kA2thR6pNjq-9F@)zr&{E>D;P9)bXVTMYbW6_rQu2fbp@?a;hLhR&OdDqto7
zD(0zoAQD8lyj*Ai`ymW5gs+h4iyX16wT*?UgiQDdnkkQ{z#jrq;GQ=QjK{uW>~-zk
zoM`e342wb2K%z2ncG%-MQ2r9G!vH{pRay${9X(F+D&_kGHt8}c+vgp2=oD6h*uNRj
z-aVsnVnB`fUA5|Ytse+;eN+nehk7F7Ccp-nvo-`!2iuS5v~S*5YgD_BV!UG&;kc#e
zVsyR`E_7>Zm%Hh!&TBfPW;`_$0Y9&LH<gG;sG}L5p|hF%RjlV=jymiIkL>4lM)42E
zAe(HWlQB+iD^s5&8s9#ba>VoV6_9q=urvK>Yckj}WRAW(4<x?=WuvF@7i;hREYiHG
zUKyurzUsF31LJ;>ynolE1rndCcBSxR`{ZP%a8!`2?jB-8&h1JD2hWpIr``so>CQgE
zT*v5qV4`1vWd^E>pp>Z)D2n6nMjk^kgFw(SDdnD;&F1^tFbvLZ1dw5Stuo#9KwdEH
z6_J7;l@<t|5^T%B2LRlJRPv-{2c|p_=Rrh25;bVdOT+@{Lg$Rb$O!R$+SCaR>%VEc
z@Yst_%l*~-iT-Tc@c)5u{=W+4J`O$=cxg8cKVmxZs@@{U+>@+?Tt=jh-6hfK1BIW+
zhlQWFrY+0UV0aF`B(mmYVmUK{IVXe$V;FDfH@WHVVHaSjZmLK`+XD1~>`OI`tr*=K
zOTx9kESXs@14g{K+y}7XRUv?WZQX_Kz^Vg!bnAAPM!nISzc1MEB8)wzuLPEqq#Dhf
zJ6!?8MQ(G{ddMFG>ACfVXa?JaYL?PwEWTLLTD6^fJo|>x8QPAy!o{ckyQJ_A9DPES
z24h*ji%R+K6N@9!)KHs|vFBN+w^gG3&LKZ21L`cy@ZPfV({5PiKF(Ht1ep%b`DopY
zO)i>QaewvN$J^iUhRz%?zsA8a2HRXdHW^Uq_u4~52F=24Zggo{o|9c7pWS$-h`a6T
zk;0?nP+%wr_#O8YO~;`Zw}@rFdd&NUYp5y%pQ*LwJ{ziZrR-W;b~bqXAdN3K9esZF
zfs^avGI`%kU^|ym?=<hZTh8gOxvHwOoF$|$<M^vf#{ZjLGR*5W9VI0^U9dE>M4=YT
z9&M+FIGSOTbwk6tMW!lOuRWMW)0bUNxIk>{AOb|_F|3EF%QV0)$4WX>0&>v_#6W_;
ziBy2PzE4#UcM@GR=J%54BJ*+5w1#l|r2!AqM6%bP+Ew)6s^~MTHS?7Wd>sClxK#^)
zTeZemV?cOK$joWg8o|HYK^farXbXKb+3UDbk3ery{Mf5FF_|gl`+nq29kII+!|o(q
z2kVX2Hu!PyrG-dBdx`|bOd;iM)q}G`D|ag{Y>6a)2!ci$)mmM?u|JOYsqnxk^-Y<Q
zdbRFSr`C6SJ~=tF8cUVW-wKMidYvZnyTo?k^wat%$7*xU>9*MgG1D*2k-ISR*bSN#
z8UB*m17iKZF11fy{jtUBk)vmf{>G(S)(cn!5hrIvJX-^}ad><5V7NZ&|FHMoaZPU9
z)^I3-(xL*=i3);>fTAeUf`uj`qGFU56%hdemEIBr1wo2{fPxSd5tS09_s|iM-jo)4
zF9{_CQv4SC?1Jb$=j?O$ec$_gzx@ZrNS-`ptu@CSbIh><0Hnt(A63zzmy%#BdyBVY
zb3nv+ft<VR7De-!f-e|&fDREWI&kniNCg`05A22j6-BlY4Lp-6TL83eAS%%-q^-X}
zG>j0HK=Sbw!v`O#r-S^r+8@18lsu@o9soi-EPkX_3N?xfNLH(rIkF?jcc<ZC%Z^kC
z;@|DAquBKyp+3{AhOf{X;S1S&$zjD^WL=;(I!|n%;~bzg>40K!u%*Up*^NL+p+Z2=
zyg-q0a~7bJ7ApH;%aX-l9eKZ+|GeDhb$AspuVxfdzp3-nWPX?ILk3^{1ET$AVYcDd
zf|bzk9p&wR^xQf8E&=NRMc8$Ee3K*hf1=!X01xXQp685zxRUUKLi6vS?T<?H8c}|p
z)f&0}ZYI7p*laKTEeOiLd3F8Pkjb(}mA_W4|3?OzV)>mV$<OVw{)6W7KOxfpKhC<S
z_7gDFr+h)UsoTI|s_AeWVbQa^AHKX_RtWv2-v*2io8<lph}1Nmyc_genUB(Z4go0-
zA94$#3r9GMTvn)Ups-j1MSzT#>P~D9?y6BiGE|VNUi{(N^o$LaGew){!v{@GZJU7f
z_od8R9&2vzsE{Y+Oo38@7_tIK*}RCQ3D~KLf)tE%6LFFr0DlO0d>dy_=GYB-qVAIy
zK}rUzK$8yx%4{%VZ~>&{c+tWWSaLGy3|R>4f;pGhwy2bh*}R;Cuveb1NhiFV2|Ipi
z+JH^Y1;XWBEP%}@257M6xd#g<fMT|y$L$eShZv^N88~WGgCC4^z!JG)Ac8)*f&qpE
zxK)lqk?sVyKrhQ;xZqRQnsE#76zTqp@a2s?tGYr528}h<pTA2%|C>rdchOtnL!ln(
zVBNL=pr{#<mYxUqhs|jf?*vq1Ml#^9r5ai1k*nlR7beq!8lyKg3OF^zQF?F|_Tl7o
zL+3%dZW75Jj1&>@P1CTW;{SZV!xD_j!%o*hfz|mM34F_v>i<Lup|foOWEKuA@*mv6
zx6TN{FvPQg!q6a(t(4_6?$T<~<-6~wsd?NWB2wq>MzP1oh6I-f>4MyY9W)*^2wi=h
zv)>@Wh(_dl1oa6{0R066lBKcRi1JK^DHx#<BZ?bKAcp{<Uk4fp{UXWz1JdY9^oN|%
zI`>M!h<fZ>;|aHh7)RHG4azYx66!tT_dj&T@2=^CC?ZPbNwbx(WjUxfp2X7*)cEh*
zAmo4|cwY!L5)DK!UrH0v<Hn97yyJk+MGs|=S?N`rAgwty7!MtzfG71hwt+H|k*xXq
zub@`=8w>Wg`t&y@<$qRv$~&JljUK>oyTC~fZSvSnRvm6K@zD`s(sfT}?e-&X$@v8)
zUs)0uDSm{F_y)0+T|O*gL@$ekt%yKSV8)dn_izK@B4hwg6#~K(;wO0j3sn_#lO_>H
z=LBkU@~{hyhk*nrU7m1?<^=QHnb=j1>m50`xC@ktywNq)cra3q=z%a~K->lS8QwA=
zVW%`f;DA@mQZ{FSz+r>rJ<ISDehnKqxvmD*%0P^Dcb;kkOy35=B)z;7O%E|73a*!^
zW-=guH1WaS_z5n%$uy0-kDoJ#3j@AsNe!$vUy=t-{Rqsw*yvt`x|9=CezNQ<GQ6}{
zwN=(;5lYs~93R|lnEC0-ZnJZjU1|<wtE1srORUp_V3zZh!blS1@D;fMFI5o#eEVrn
z(POgj(0xoDFFa&vuTZQWNaA%<Ho}=i^x?OM&V5dnPdFKK)ue7-R1q=XgtSGMsse#p
zCZ57N0Z^_N{N~^Aj`iPycWwV~71jP?c$bB{23;1U0}A_#-zwqlMt)t~P6Gxc*5oJ<
ztVFXw#aMcbX<&6bb=7)u4oFDu5QON(9@;>}U5DiF_=?{KDAIaWCUi%GKYSrCc~B5S
zf`7%+bw<lWDEygN+TLK?OHVx@YeufRh{l3m`G11bAPL5>Dm#BcT$!x5>OQi6e5Vl-
zIWAFK>r1@cXQw)`TgWXzOnBe@!2LteMCXV;Sk@>tItnyt_@c+ldc9VOUcYcE67Xy1
z%i&~BM*tHKq04LNfa7j_3-+a9$p>c}O`U$}1y9mFcrm;}RFE+HFv(ZSb<swqe(u5p
zr>b;4{!njk?Gi~Q@#oITygH?Z7ZW7Q8(8+LRaCwh|Jb8oe($c-3p#KC8STNHg4C`p
zFeVAgygQ>>k53}EC38xi3n2AsZ#TaXs;zvRA)Y-B#K&I^l{%~kl-O$v9>#TwZNE?|
z)*)YOe96>sxBJsTt1}F$nG9zF_F6k4;tpMX{Fz~eTWxa_lF!4GmhYiNNEkR%M3~X@
zYj~UOr*OYJ=ychxPg0{6k0~h9f!vPg=<-ji8$X>a)}soem*gR+XKO`Arks?+^p+LG
zG>5W*SrS>lP{G0I=B^vo_$=3Adj%%Z3+_%+OZC>hzt?pb-Pa3|5UZK;dV43x;Mi{o
z=O0x~AnW|i5O+;G$EuIh>y}lG62vKP-yfT7IQ6di$L|$b+naFWJe+nLJr0f*zbcga
ztp%y^w=9s=d`+K`K^0kt4mu*HG4wi<2ncVfOEMk)67&n10edU?-i|d*NR6xqR?7)%
z^u%`!e%Bc{vMPbf?C-|65w>f}ok*C6D=kx5g`hNGn@H$IG8CtGIylXeB3c6(sN{>r
zt~Ue@1bv}YtK4Lov>Kv|qic4OAAz}4kNpODg+*=xf%Us*<R*I`Zml9l@*>DqB00wC
zP&8Lf5*dytJSe}^nR9Pb$PN5nWl!p;<m&@~?Pvux7|%S42D&DbUXNZJOhb=N8iI)1
zO-q0eb)qfcE5r`?f@7+O=Ko4vRQC^9HEW4)iWfvk*(xMS67*pS^Ae^Zyg#5%azJqp
z8Dw%2#|QwUMC0iZdTOvOh$Je&!C?)3FbY3#B$5g)EL)FJ9T;dZK+DW=gIbVByW`D?
zKYoYs{?-MR^2k~$xek9LuqH#G`qn~PxLOM~)sEWr@RUTe91zlp0sQm3!SL(0g`(qw
z`q4M9+|UqXHjCSErQ5LEP$77j<Ccb+>JtXjU)c}N>c2fU{+!3gUagm5u3*<s`y2?@
z;tLum)io5Lo-jlqrRd$rxm$`kv<Lg(3yLfAlvQ>J<3y7h!galC9%vz%wVfL|D0KMk
z6*bkEF(yJVcmz;DfAqy;=I$&3ts-X&$YBgTw2^=SDLn=s8k_}9ibDT@t6^TMZ8X49
z3|WwYDCgR?Ut(VJz&i?pBnF-!X26;D5U4r=v#|kS<A@)$*N7@dl0{|1@<E&zXk!{c
z(qusoVEs?XK$oB&;3@?Blg6ly4Agpo)*%KGzRYe}4ik)Cq{O4gZvi~*nL~)7pH3RY
z&`@Kr6(e2`Z6FperMQ5Qr%uqM`RD7=19P!{lQYn+YWgWzZ}^mMGkU?i65;{>rPn}t
zdMJhzoEf3TZuc7bap?KgQ_S>R|42+XQNV<(20ES(Xh5`fIxtYANp(=VkDD`nLnZx?
zXT}Mj-0cN7>)su75PE;>S=PmkS2QwpCaoXZruGE*d8>!sRSZm1cl^~9Ig1FJItu1)
zIba^YT>J*|$K%+Xzs)XxZg$!JH(>oIeB}QOU`2neVWOhQccC;6!K>7?vmKMw!8jY7
z7mAbkI<H^lZVS>=Q?n?pxv$j5wn^aX-QcsCeGwn~9&;%nh7eQzV736~#s-?-0-m<}
zlwPM2UFREQ)kjudmbTjsLOTsIK5$Kd`5CRPz>wL&B2>fsj2J$MEyoAL)8Umuvwnn$
z{gsK@uQU}}cLpDpmZolSa<yINKmU9H$7el_xsRN?0TaPZ&i;~?0zhsB06F{^+{R9T
zTwdNQPmm7rTkN9AeWeP$`gkoXKaxkr^!BB9`vtZ$#1*oUyXf0ODjHUVvd~m0kl74O
zPUv+N&_UwO048L_D**fh$lhM?<DhT>6L#q^A%Ar=PbK@DZI!O_f1?mQFmw5=;{9?P
zCo!=PSb$~vZ>|~%y(j?WP6^<#WQxH;>WxCzI2EkcAQGry#Uk-sv+F;TltkL!?il1w
zvTbqXyu0G5F*XtP&O8rn0{$-aUS5rYNFKH{pO#9CD0C$(^*c|+_NQ78$P`D5M8188
zPmxpJ`ibMErqF?{d6W;4H4vwGXvRb(cC~kI1rWRhS+W)Gt2;elAB@c83VLXg5yxnW
zU^fJsopDR14XnDc2Z`@4x3k-vUT|K#IrpTR;eM8p?bn~XHL?!krhNd5AU!y{LPyfL
zMLar?r%+$ub9Z}zvJ5*6=!*j#bq)+GL4fP32SC8D0=q)|05XaZY>O2YT(h^MwIcLE
zChn1_G*O-0l9?)3t4ldK6|_A_WJa!H`@TCl@vNzy7(j?4D6;4#b?|8rJ2<%w;-1Re
zk0i9A)0b#Gm*51JdC)n5$h+J-wo(=H+QqMYDdB93yt_kw_H!`}abNvBeyOXz#~QZ7
ztRSoIssC>4fW+h4?VoRcR#B6{kgP+Nhk%Jl4K^ALfc(SW!(=33Cv*#$-5^}E^;L(>
zbmvX)gZy!uE=O^t>_6qxl7%t=fYprN0$)(=MdNc<&l<Jp89hjMh12G&BrcxFDm|wp
zDn$sZCRNZ96N_?!^+Zj1T{mwTF8wHOLBEWc#XLk0$-=3upu-%<Z~mWaUqJ1k<I!^t
zq7q=Ncq##X>`rV~?V9$6AOAld8YBg@XM=QQ`XTfepA_^iY(Vn&FT_IwR%>_^dr}}M
zDO%E=14^nIXwI72vxp);!T<kjI~a3uSFp=_P;vlEz`;mr?8Ew5&&Ym}qMt$r;->UK
z@1%lMf9GmBDgZur9ku~CwI>yj`WF_7yq*O_P5Nn{K7jh<akRa0tL3PR<Skq4h~@o5
zva#tdkF?JR=N<`Vn~Pq=dLidbP=L7mQj-AQe*FRiMTE#rY9&Makiym|D3SAZ;Ywua
zUD6UuZ*R{FxgDin49&&euCooo`*C2E_y$>?BTCX4yU=s$sOzX!kR;qeZpKZ?f||h#
zgjXZ$bWSkb62?R|p~cXptl(@gM2ien$tFqQM*fBDP~Eor%@vIwM_<_rg|<4@U_a`S
z4dLAIG21p!`dV%Fe1uI@WkA<_pf|x6C;Ml?)fDgl$r8CDKz4PscPGU34-XGvRVwI0
z&9B0=`3nk)(zo3(s}X-5<|=`U&Zm1jgaqPPQ3BKhWQ&&`Cy5u_M5uJ$W{CjF?6_+L
zTI^ZXYZ^@ZQud#5iHF+1ER>=PVo4w^e)))%4(SCu^G<p|iRFEW3e46nmlLI`@{caG
z%Q6S?=64!bT{`Kjx%^69|G_!eLk#1qMD_PwpY7mk{^$t(JyZYNGU+>kDi1^bP23z0
z;LF@s5c8$Pg9FG4YrBVZsqAdx#bh3RtB#Kkqgvj*OPPT{R3S`WQY)|~^w1tV`qBb+
z1UrDCKAGhMgW2pCEqy#Qux&^H7OuY2fZ1Z2fs^xB=FfC2&Tc`Ks_ri&vLxESZGb(R
zPPi=d+BNp<G27$5ceMn>2U|Nzd6!rp;D$<+Er2*W0!M$C9S`+e<!UH)bT+_5NWuWk
zais_Ti%2$tgs1Xw)6amQM=#As1C6Zz|89+}4Hc}p;d{uVK!uynN{_5WZbfvVH{pgp
zp+L6LF>?=alphckPaueHSWXD_36dH8B?hD*t7r#R+(#=!54|Ld(JOi+f%wDtENeON
zoS||r(RG4ffIN^6&_m|rFNL}v@}Yn9g5_i0k5Xf}!U1YB0`QKlxcgTxL~H~8t64P$
zs>+9|>E&-n?VKEc@pU5lb^QB}1fBhI{>HZ|PL0mIxb>LTYDq<sR8EV^`~Yg>_vxEF
zKBH@62L=VcL0(I?d#Skn!%~qL1SiGNq8C%obe(EJZ9gI3L#^RB@JQ;tsrIG4GN;uz
z!dS)GmV~<OKJ;kY6rdyTJU4?sfEfHLivIs|OU++pRHdrR67$9~+V@fF<<~EgCL(98
z?=at=3frU6ak2W^CYHkTGr*eDLTK{iv|f4a0jAOCN5?kNm$D#eMS~^Sv@U4t)*%5@
z`oRrN%fYtx=#eaF&mAZkw{Tnz)expIVMmZlgZ^;UZB$X>nTVHTaoY(7L5m@NFAlGG
z$k|HSaoc>7*jS>ZX}k~*9ceC3qy^xH#+1PfM)Dz7wB=Dn0Hi6>>yZn(3%D8pKHiNf
zOB3AtqP(*v^>|#2O0orQldoKk&wb?LQoHB)G4Wu7hvU6oa11^GKX<{y42<B|6QFu4
znkz_uBoFn%WRQ~4_V29V&_iEmu)ADehvv+<ieLL&9rr3s_dLjE9BI#SV=?1sF!qTT
z7?$t-ATXgv-lA{z5Xm8<(S4I>mP^(Bl$eVA=n~tx>$wbSkLP1cc8j<pIKM$SCMieJ
zO-Gu+VgwSCH%!ZTY!13F9yfO<3Pa*(0kn+?3QWiEuWq8yoRHBsTOndos5uDVaA~45
zyed>+wA=WV8CxFnZPrr|x{dX;-4`Fa2F@Q%+A#^_{k^zzxc*qQh3tyN3KC#ossYf2
zlMM*=B};rAC?0}t@Uz27meLPuPvi>N+}L~FKV9!rr3Rp%s9pNVCWEM{E{Z}=PvtF?
zuw{U_LOm>m>XtHr&ybxOeBJ%Xk?i<di_N!&OEC;yG_3{;NB&S2p@%NQ9>7?#&>Z8U
zoUMgh%<jA#lUBOanLR7h-Hm&;=h*Y0sk^Sqt2?%>zb8aV(rrL<0xU5~FJG~vKiMTY
zE%Hm!xoWZ_7)5`gf*KyfFwvRFc9+tLEPj2QiVbI5Uh6z^P3RRpYIf!0Cv9bKx!vt_
zRw^q|^n?2Vr+tZ_<H(CQpO^T!wF?WwK2^qcG2hj$iHDBL6n`bF1G45Sv5Q{30EB13
zIDL@veRt=aR^X+eYarU$JL?;SeY}|CLw1w=jcn^K`JQ3p%rk;=)X||U8!I7bc^<^V
zu^%qz#{uL?aYlaxd>x#!H8X@J0I(NgA!=au0fx2E+128WOF`*H#l*@7DNbjVyZnQD
zw^oYriw&Qvh8YgSzdthw>Iz`^;P9aW6A(XH_37G3u8xJMKKuh5YcXu1E{%Ky97>l`
zMi7Ya7Os>yA}da~ebBfEJB6nlH8!qfAm4c{`*b3&oERQ5dzN+n1M;p5u&|Wv51|FR
zh0Nz++oX}CYqaNwYQlDbXRcjjA13S!j2N@*R_88`@W9UttITUA>$gzTJJMN~<EpPK
zF<qZ<w5I#(Xb}W`IAR<RqWMsU3p;p{NP0vyG9O8($7Mv(xIJa((Dc37aU`QEWGp&1
zE+f%HrNst&OCoq+(|MPhpdBND%PkqA4nfKNkexcxL|vl#I!eAl_^or|b`fR07HB8V
z<Ir;M{?1XM?^!N+D=~DwU;h{(C?u*5K4%BZM9{n)a5S!w;_cXRvvSx*Rjx5pQKG2-
z3m3Vn@-yxI0V>CPRk~!t<hpq0-1e9Sx^+N*c&e*YcMRKaThOlurE-z4+6*AKTDt9P
zCfn(TZe5vEvT(S$c}9N+^S$R=mD=>k7Rax)bpPV0Wq>|-O%VR7yi{SLV_K-ifFARg
ziOpk&{mrhW&+P2&+;QK3zXlIOwIKutT0imS&=r^m$g_NO&~%y1%#bZEVikgch8p(<
zXx!&;xjn;OT~?dx4o;n`%i0{Na8|pdkI}kr-6(NDZ6#|Gj7o~PXA%+j3puYQPP8jg
zW1c#1;;z#OB4~>}xxpIj>vaTCjpj&AQ;VraiE8-M%@q@KuUwehDBYg5*bo3?lRtdF
zbPxx{MinS!>0Ip)4fd4!Y(#RU#m-_ma$sDURbd0C3pOwB8x3D7(!Dte%V3@APkic4
zRcaBRKiFn$2D&A!LCx|utsu*W{SxijbXuWG@>lD_wF5dA4qmE{+9LJH(*O9F76$`!
zaIRQ8j#p|$6A6}(sM6`a4FvUv1WGUZrjTsRrz1wpDBFXvT`Y<VX=ZeR4$@tK_3aYr
zdcXK8%;R5av{+@IO=EB!)Q@|%Zv$F4ZXeO&o<k98q}GsV8<g0q!R=YpL*0M+nVQCs
z6Z7@W8gAa4uFCKX{M4+SoE{W}H1QxGDabMh*D(sPjw|SwEk%pEc#l@9$`Vw3-@fZ*
z<~G*N{YW?5*do4J!2oin@hnRcRT9ksAM+&4(mAC;NXj3lDh-4%TWwp(%tr()cpg_W
z(iYEkw;nS$a`CDOJD1|Ly?UPuFGw%Gn-2vctRCCRO8o^YsFrdR$A^|HZBx6linVgC
zjeM;Vm8^kFG1><}kmNITesm+HA5RqoSd8d^E^{W7)BW&VU{Pttg9I#1Xx)tgi`x2i
zwq=pixisG(HaOv(ONK^9W^2={%xLSegy4QYft$}Ba<OQf3J(kb@%8>}zaI@+4A$V%
zL@+4%E;!DD^`%fRVJx4es4rQl2VVX`SrJ6;JvPEI&U+ym44rxPZiqcUeq?K*(m(^m
zD!MJvH)3@F=SL7CNuBo(>h6Gq@$&kc-HY3RLU^dris?tm7S+Y#gpnA#m$dLxt+UcQ
z@Yg@njk%RAbC(k&+2rmZ6N>rCM+usx(Tf;=>VZC8_?fnS{Dpj5?4Uxtd$q(v8$+8j
z7G=Hg5c$|slNVd*qT@MyQ7y{zr@CsD(?3|4HJsCy@PZ$}(AnfE$F)P`M7DgGybi#%
zHKG@vK3a|6Oua@n8juyF8WcAz+{~QT-S-mnX!iwItsy0?A@fIGh4U@E6Nc4O1r!BK
zOK#cnIyt@lp`UWA;C%*e9sPEtld`P6WR{N7iwDT)v$;mek9HJ%3UgrN{bF%c2!XR&
z-5he=^bT3~SRKpJ-vfbl^BUsOLb;6UK=P`xc^HCDeK`^7h1U03xhx~CwL4so^~`VL
zF$-3}J_8Hk03HVD*e-y2N<gMHn6;8ID7zXXwok$dk6ju-kjPFL7FYG4JX*dJxR;>G
zlyw*L-N5uxb)!8-Tk8eDgj9D;p!kU&n^QAmL<C1@L*Ct8=}!}V_v>T^GTz%EsWAAQ
zh(!J{B3}_;t3fnwE3Q?ha93Zg$WHezfm9h$k*|@R#}fx`h0lG^$0UNl|NX}*Vb<T2
zW}})U{CM_CeaM^6iEKXg`_9|!_Yn`A$a+5+p|IV@r<55Hy6(O$atdh@@DB=AEc*2}
zj_)7q)b-J7>?(~7FuykIW#8s68MnZ&>SOsA_79Iv&0NXeXOUwR^rdPr5P|Hc6u3XT
zvDq*+&p=?WcEcftY9uJ|FILkW&?DAFD2W1cyLjRen)^6@MVl8jVPbW7kE5Kt_ff<Z
zZL=Cy?5R04zlS7QesJ0i&EryCx=k@H`R<nMxiNQ~w!AsX?9Ivs8CQeVRRY<WFuEF3
z6++qU7&4T#619F3f-TJ&%CJ5tE)#v_1AQ?fM;()o-jZcD^21mk!TZnYefl!u1&^_k
zTG_rHx2(IV5Ng2>Rd{4`^+P(8D$=1W+g}7VBA5rwJbRf~W=U$c{c4q-{*3X;=;+>8
zqs_VK?*pP5EC~0{8l!P;9!Cgfea`nCYvEhTb|r2o%-iwVkz>k6<yGXW3t_IM3-6+H
zN-$oABX3<(C-BhB$>izdmNJqNOOauhFBKHnw>=HJ$+X<?qeoft!qU2>5`nNF<n)m+
z{URqyoJ`q1_^Gu4P%=@c<9e=k%emn4qaQmDG*|)5y&F5$p+$QJw-7ngA8+Spc-{R%
zwwir6bk@C1so`q@=`N^DbbBbY@M~1s`;Ec~dm+?3K%cL3LTg3_q5zm8-+Nb`O0ESn
zF>B#$Vt$?KK-up34+>)U!UKMS`sk&|xpP!=khr6g0cThG3+(HMt}e?PJni2Co!&)7
zgSO$!NPBIrz6QHJ8wQU`4qr=vpZeigXr;v;<OejcPJEa27<5}<07{w=Woh!*^9G=#
z!4u7|W2dHKoFi=KK3EDIt-t&<yul4Zx14M*2F#x$*&a|Z5+Q(+2t~{sp&rCe4FjDH
zUWbbL1_?+Nf%PFvbeS%ucQ~N^Ivo?~G0KR;^AZGN9XqS0aooF-Eov}-OctE_98n=l
zVebYchC!b^P1wb!FR28_Gl?kPNo4hMpiJwW4Sk&(r#n0pnxC;*Qvy^e*5hM@2P2qS
zX0(^+$&PaTih+v9Rb;b92qB@b{a*Z}DSs1ni<xC2+5f7^@tz~hc}7msn{R7NnjH{h
zc1#J9T}@mag|Hbp<*5f)n(s2jVZcMA+b2)Uc1g}qg+PkS1d@l&h?T52s5*2;P{dGV
zLVx|!0lA~)fkPIQw!8uK12_xSoCen5O>pm9j<6|fg-i<k$elTxs|`_G`$8h_g7&Mr
zBF|6?6DX540u%Gx3HXYpJSt_~bQ>?+LBRAicsWwQ{D+rr6TFygC2}p-@3zC%J(`d(
z&AnP6#Z|Iz56gjP?Sqerz-SBTvXE&d=nAVQFlc}g;u!!E|9w;Fd{S|yprRh>(C!;j
z8Il5B?HTaQso&lPY?D(P*1YwVV5}FvqJssU8C?d9_P+l=#aKTWQ6?mW6hc1B@Du;r
zWF5`oOw8F}g_N*uO7Ba;G+Cl97iW6Y1|#bvuJDbBd%LV!HFgQc`oY0rFyJufzCkR%
zpc@t8lrcaBf<9*>i}UZd-%HYf_i0l1px2(~nr&C~X-QYVLDaxoW)tYVucO-iw8F05
zxW^pNQ&XxisV4k@H*f7<@SxXyhS_<LF~mEdmzuso*pO?k9_YFLaO8VUZ&~wajm@)D
z!#H1ssdEx$<t*aCKo4V70oe54HghS^y-RBEUC@+&S>MIbt<8$#4f^gJfS<PQ+Vj}a
zI<>Ky^y@hQSC5=E59PaO$+%`in0$)+q(<}1u#xx$jnl8s2uTLgqz+Qfl+*8lE5U10
zSNFE;{_eDE;!Z~12vevy)x+b!!@voOYyVw~Kvs~>W<#gNi0pk^2qxHjS6h1Ilil-V
zIEV9|&fFVh&-63)oR7;QdIJopfxD@GE}82g4{4iT2u%c;f|4Wfl_MWEjSp7Sw`F!<
zbToCGbGuJl@7PdMDsl>3&DL^~3GKP;zFkFtqNzMEv+R^t--H1PYXv3)J)KH?G!RwQ
zz~RDP(P%04+N!xeL&#kt_^en?n0{7ZP{=b@+lb{1q8|N}^CO^xIq(KOexhSxQ@@=L
zo|(#x?Q}^fE8g(F#nEVM-G&OK@+y;42bjZ-UKC4m`_!t+h-&Ceq}3<A@lbBD$S;f-
ziPXU74TU~B2#a|u4jS_66Fp1&?UwgL1waBmYzHzC>EuPGA0mn!h$xF#pR_^-yS`+8
zs*1ZqPWWZU-t*nN4YqE1KEt3k*$JZVsedQx0!`-+YIC#>Y;!RbWRw_xm@zTRXP;y2
zDXo_pxfrLphq(rQ_q-H%%h$6?5NhVnbU^el4=F-&AKqI=cBXI-00?p>XX=gEBmHj6
zb=F^~;l7TE2mp^Ha$SVDz(AetSl>*zcz*qj(0l8#n=GG9w$LItCH6fM3KBYI*;&<=
z<LWdrOszS^F=S_}Cm9W)+e-alFEN9Ppvchc1%nDRj+ECOP|gV4U>|poHi3OX-Avxn
zpL(OVLXgkv(4@5HnP-wYaVq_Tkq7fGYfC=X02~7U4<aLZnxHkBshG(n<Wf<`BF=AR
zl_09yRawz&Co2x}1E*5qFm3rC;L{tXxykjti?~;*O8v;v9?7!K9Pi9o<}Y1iZj)t8
z*#SnMVpG2;n^$QoU1B`S`{{=#KzIH@uY`(Ux9-n)xD)*PW*;qbf=WeYFUN<r8Jm==
z*}mF2H<Vmzc^+N~iO>-SqP^wUzgV+jaC63vPzv9oEG-F{g!>Lm=i^vCq{%9TZ;*?g
z_NB#aADZok9)J=!JIev>pz#`IVGht-WT}om3H$DCA?vZ)Q{-m&P&tYZ{iX6PdJG{?
z7)EP5W7ttpE72|Q2_mE3Ddvrb^5O=5A%t`!24+?XA^Qfm7YX!Qe1`9vUDAZv(;>Ch
z>h{k^X;>==bpu)R))I!7%4fNSVYGc=tTpbL_0y7uEVPEI7!mDth-N#}U_cD^PwJ^4
zhq1>V=YJj8cYEwJ8^+Ak{j~=xdBY2<_UNi`0_9A0QOjtsi-c}}LgY4%;G+{M47_#~
z7#0+(3(lmFDC_URnlvF8OIZ%@)+sCu@{+6AQV*F{6T%T<srvzrS}Mnp59O>%MhweM
z2)1v{Jl~GT?Q2S#TBt-6L<^B$1@s_yoKAanpVP;?!U7QC>&Lgj3+g_QzNhjT5@sHN
z)L8mRg?{z67tH3~#;?=^q`c1@kGypyIFkWQ=f?TM7F5a3pkWEjC8`PRhf@F|n(si^
zg>@qXviJ?MnVeLz;+Y*T@phm{v{WYboRL#rvG1-piDC1H8JuJ8`};Z${*Y(J3_^YI
ztjgLJ@kZqE(n!T8it6_-M(&ZBGBVQGA1VRu(QlEjSpQ)QsbUqtqn#yzD>R;rJK2hz
zIpXTHn!BH-T*$CFxht4w>um){bj#Ju2$kG4?lSboVm7i+#kRP;gs6Kdhl&iV(c$)7
zL~+!$Lw4TU!U<biUSI*sdGDyAg9j)*0bsEnbbJ@GAgw=fJS)(H_84^+ma&`Xbn+>k
zv2Xt%<0!?G7h$BQvEvJO?tFvk6NdUGBXp+|Y=J9;z5?&p`9A#msoQ*qHDpP#el(_p
zZ;)aDW?0tl5B=TAv=kL*27mgbnW1@^H{h)#yHBUrQ8p?2H8FP&SJIVZOlJboC94P8
z^-rNk#~S;V3jp05#7zAPVv6ETjf{;yzBf!^me4+T##Q(%tJE4pmlmYeVg26vkSoET
zrr5U~l~i#vIQI4nSN?wq)b4=|!yL?hhT;;|ocnGD8Gsaj^?{MX<`!p~$TeR~*U*%`
zUsKjMACh`wEo51RXP7^3@IN-_VfdFje*VOsJtfFmFGfF0Yx~}mG@+4?{E*<1mG&YY
z20-(cNxwliXyFcnnRyfNQATIKLE=JD;MrAlc&bBg!wHY9VDrtgbvSY~E^&-6647LG
ziugqLE87{@b4vFnO+0xRK3T2s_RuDckB{+qsa~_MZWILS`hR*Pc>lIm_~<KS44Np9
z*H;sUQg?RxQ5PP~+tDRL$m&_p<#E0dJXO~jC_;CM`nW`2&sHDAizPFK2SUp-aDU?6
z0Qzste(?>WvU7FM0{;a17Ml}vH}z$(5^<XV7oRFix=2x0VZeZ9V2Xt#^XRfck;Tb^
zR@-(L5#Hk)mi<|`W7IsY-@NovfUP+%*qXaC`zD@?b?sp_*IZ1&#BL@qG}hjX_6Nm2
zuG=uqHyZ?8NjBS`b0$eDBTUxa>OXUX)?H%qwm6|;?+JmPK`EhRVeaQln3)b6b%?!N
zZ;0btJ>kc4yVXm(wPvR{uYLFMt&9IV*BF}7XeT|L*6$pcIl+7T3<H4UARS$M=hIdK
zAbsdRp4=hynvwS}KYSaE16eJn)vW&Dik|crusMs7;Yhgs+e`@-^|>XBJ7Q@n3=3)A
z|D8|x?hzUu7UUd$XR<FuVE*OJ4(3JGLN2M)|IYncHwGG&rQa|@pZzw*S+?c9Hn!5<
zTJq_&oOZX?<N{0H?3&TGIt$i4<R8ZIznsZ;4R83kTI>ruMSM(5=iqLoGybLX=-!KQ
zgrEgxG(G28j*zy>!3ow>yC&ec|9R3-zCH17xZ1~eZ*xY(g@@uwRyuV5#Awf6(A_zz
z8-|Dt_3>h9*3)~#+tj2ctkU3%9m$I{58E&D)mOWIQ^S(g+WGm{+%be9=Md?z(?_Cj
z4m3CHw)18Ug_OVix!5bpJaSAjE^JAcOXpqImORRuD;l9eDUa$n5^k(Pd=Tm{AgK(v
zW$VnpaQwC3@8lecwkdk*vQ+U|yQaEsywdTn7-Zjti`J$Er6Se&X>LWiX|Z8RWtv*r
zy8y`GqAawx7pY^<UB&ycM10hX^~JljLFUgcryvnuzswB994k~j_hmNZ^JV8zN&YGo
z#dha^z9#?VApdl&Iw~S{tnfo|@CYLCoy89AhV8+9=O#KihZiTi%ZLxIu<$=89QYzQ
zdGpUjhJSGu@-L$~2gdJS<_mGx5*B`bg~JVD^d~0L`nbAoMM=(g_QTr9Udsy|-0LUK
zXx}m>_2H%0Q5!Q54P`1TWC9A*mi-%GYSvO~W8G>9{g+G6ccWr`WP2jWIb`Tqm!RxE
zvilhG;PBC-1Apnl6rOOYE~%4WV(;fmIkjQ??@Vm422w=ck5<tehNl;D^p&r|Y?-*@
zc(_Vd5Xe7qJM*QFoQ`YIjjq3VZqBxpd2qzojjj7HoT5J&3`35(EOVsE2LD0!EX~hn
zTtrLS(SPl@-oM}}=c(Fx>4y1@$3r%fA_uzJM%Tpbe~H!Bjj4YTrL0e!YqXOU-R3{Y
zs1U)WJ%PKImj9Q9a1$24#BR$+`(q;t>vB7_$1!JBtlhxTF=Z=02d;`6hxY(PtM`{j
zS;PwGaI4c>p(>6o%(uc@-R@k*|2MGWPX@tgri#R<uB)Mw#yo_{Qu4K@T8euJ2fkK0
zEG6{nm(VbG%#wB9UpNligL-{jRT2~Pm9NHM<P5ug*S?y!9Yo-N2HF4QFt9k;l$P1$
zzYBzTnWoGz`{zrH`G&ih0tyMki<R9i302!P_sN~Ptw+f+`fJyoz!lD+e8;Sph3@?_
z+Tw#FNBIXbK%L{?&rSZ-g_<@?c(?PzHYBK@4}QkP3#^b=9jckW{bZAlmjR=NI@lz?
zeP+dB+zPJD?k8*}Hr3zu+|BgjWnlt+TP%*ogs0P?s|k{!6Hv23(G)VA91U4?lLCes
zmLR)nQ~?Bi-#-$$2T6c1Ra^Q7!TEr+CmxFU`yi0N@5$fe<nOujS1kp3aCY)LIJaj~
zZO(qrm>YMyGH}lC$FuWSx2(^_ZY9^ta+zvc>`Fd!Jy82`|D7-mMc66G^iAw;P<Z@q
zar$m?^HMPlnEQT(TR&)-+Q$Aq$hrmp?|VXy{(GGKJ$F{Y&VR6;q`<f)-wuc<<6lW^
z7Jl9A>;M^65X~wAJ4hkoCgW87PPR$5h}I?UpI-L47yC)*H87oV<K-QO5i+NNfJPMb
z!4l9!z-YDrS~=bwNCQd%f8NNN++?Z`WbJ40^n0`4AZS*=Icrj-mC!4!I2sj3)rGhf
z{e6(X@5$fe<nOujmn{Y7%EpkFx{w>b$cSkl9#!aNs6q?o{IV4u>}UHzVULnQp=LWf
zdSzLU#xU8GT#o|nQy&@e^y2d%fpmiqRvp{e{ug|yxc4KdmGEgldV^_^=?OwI{ThZ;
z{taS*H3b{L92xMV`cC5}yU}Adro*N}#1r%rPWZ(c_`>0MsQj-DjK6IU^?&Op6JAvi
z+}wQpVh`xT>WD{RRz2229dVNdpa~1B7SC9GmpDoda>kQD<G;GJrx5a}%RF79V3q>(
zs4pWyDg1xIgIGNW)kLs!kk=vn3ds+Q5?<_OQ4j<lv%~<$+(r-)8MDUzL3FhI=t1*8
zG~%QN?7ENx@XA@AUOqxc1;)4YMNadSJg?8CjKhS}Ih*Tz12bxc3ay{qtkF16Deb;d
zT-@lP{)!x49*quI$SWVuk4-^5{-gydm)peaKc;#<TusOU0mfH4x)SG)6fQI;@>MIo
zeupeMjF`Uq;xfC-=G$GLeGLLneg5Ej-S_Un9fS8ma##5svJR=2JU)oKcd?hTy-)~c
zV}o_v!;-*O$?QAPxkc!ykFL;EH#=vo_rZ2A)#;bTOypxPQC*!4Aq%n(;ajPiSChym
z4DfecUsi=KXJ=0fDer%>pD{J}T_pFxd$ARv0mEYCO3E=*t17&s5Kd8=AjNSJj`-fN
zwp{%B!swaZ^`npq5q^&^-S1+wjtk)TC9A<+Eg(kHl8HK`VA>N8gBR#BQipV*J7L~5
zS4Zq!wQcs&;jM?y7;6M*v$03mP$IN0GGAGOXOC8gh<I#mE@T~mF<2dEw>hDY&aMhA
z-Cd9sq)&Z6S1nOj@>pQMYaPUMP#xu}*@x>&s_ul4{b3bSu#HwnZrY}JC`9TXI$Jg9
zaJxY%L4;hQ)P9<~MmS&*{`CmSuU%FCX8x!z)%995GFVyY4oUY6^M<XbBh{bmsBiBL
z4%;ox0tsY+Ks+6hGN@*JW+0q#D$|o&t=Xs|?jZ4Gu}0MEw^#UwB%et%YMV^Gd$Zs8
zIU_{eQ`6KjjVeKQ9k3GydWNm?J%UQOtvA`ztB7%Gjc<C@*iyE<H+k5d#60?jhk*!t
zNIwWH!PpL+Ai@Upyd608$}du`<1$-bIp$|Re7vb!B1*{p{nmS}eT<qwE6qXQL3SR<
zx#tk_P?R?mquah?ezQV()u&V1LiZjE9=Uk2R7}iUgVRHba*qBL&vk`>e?;Hpa!jJF
z@cf%cMU3`)rL!VZ!e+8#UqxSg=Vbm~tt&M0%2DiVv^mW82yT<=1{b?B#l%)bRHb-f
zY~N#@jqW*eQ$ei}<_1@e?Ah^9{5J6H<6Ctb{F7g-SdYCxppry;{|)kWvLpE;iWmL)
zJpBuLYH_j^g7BpR)yLzW9GZIwjWfABxg50{{n56IUQ|u@ZDknA1O=1djE4X1pNP`p
zL*!cG!G82~XnzAE8FPuES&kF#N^#n(Qk4J5ItRIX_cOE0muN2#zIPcvt8DO;B90E6
zMs6(LOirjCj<7vct4Y~?YjcPF7T(<Li=x&to*8PQ1?nF2lzkI;Fii-0YD8pZkbahl
zv+2cRj>Z#q!=0beO^y4uRK4Oa2^Y7J(YU?AlXDfr1A%-sv{PshsJ_6?Y$7N6I?9%t
zsgr_^S;Z0N;@d`)th?@Ox+guZ9+9i$4}D*M*3o?L#_g~-Jcx2U)KNAB$Dk}W7y?7J
z7RzK3S%MM`%kNx|jA1`CbLr4k>O=m6M}rTYV;jyb5nhhEmq^RRS#k=|x7~;tRD%1W
zgj+AwBw7*;cU7dHy#K)2LU))yq_EC+`2N*XLre?#0pt^XIn{Ony~5;j*{Obtrf1|}
z#leyeLye4E$vlSl&GzxU+$(X}fb{t7C!W1V%=56U>S<o0QD1eH9pg<Dlif7Ch-vc)
ziSXg$s4hqGAPzS3o1F&Bp9q0@RdNTTA+&8Taf*e<`eeIWv<O3(Ez<N|mywf^AB3KH
zC~@B3+5IwTn)&XnieQ!tw>OFTD+;b0BwZ!XTT+gD9Pp`}wvANGdR}Mo)@a8rvAjT*
z9?=;mt(26!dO3vnTXCkb=;LYv$_zS4D5?#|U8tQh;f9>5Bm^6iYpcp_7R$JU4(i-m
zPS+}*iBjwBXDI3ZsKqMKKihRepL~yEWQk<;V2cbV@9CXrx31D~ow<>n9epv3kBj$O
z?>)q@(vxlV{ALU1aALQL1zqg=q0{{Y_YzUwN9K{w`a90=fM_e%vAnt-eO$3HD9oM9
z2f~0%yYSl{6MyC>!z_jqJ!E3HG6T8<tQ(~o=-<U?U?Tlp3@#4BJ}-0yPSYk!v~=L3
zqXXpczwUeS+x{2-)=yr7U3o(wj{A^o7gHsgfi7tc>on9zWa@QsC@L!4#VzWlA;>>y
zQ6X{i+Hmk#?u4!zdiowQ<{Vj!U<#5;Q{lXAy*pbie9KSY^_TcMtQ1R9Ft-U63ylzS
zVwhBe={K383GbkSMpOmD1XZKhMzdnpoF_W4PWUz^^VE_2TMzUhf!U=G1lYrC<i~ix
z4MQnVkCJ;RW*+W1BCK4`&uzG<XeQ!~v94pPNx2=P;n(Zxc2=*YoXQXG3o0<vx>F8G
zeGL6FAvKt)H;@zztZ#5%GhJ|lBQK!-%H|yDMwP<5cIrGE3Vl#`toac0#;DIuU$~fV
zYYEJSaJZZx&5$wiUDV@6<PfvIXO2947ZnvRX^iun3;LK58T92gYVz47^fsaG5LB=}
zk{|RIXR-!)7!BYygMQ$z=3KEv_A)DxAs=>L8crbMg2xLl2fHYSi1NK@8(~Vus76>O
z-XDE)7Ur>$e8=7Hj+AMpp=VuBAtR??eNTq+bhdPJn1F#t8TI)F@jwVFMIZSTvsE77
zIZr}(w+vYLE1TI7kCg3n8Emt;o*CTKH~)0^ZhdX_eNM@~o&0HIALUo%`(=47MGq5-
z!v{GT&mWFWH7s|wt+*O!zumE0X+JBo(@`FV#*R^RrCuj`a1mEM?c9eg!ECgQ+a0Br
z;XQ)R^k<G&zM~w7oZerZb^2rOWcV`bAz}bo1`QGHeFSFT6=H~YA?F*a8j<gzU9Zr_
zed!;LT@|VDDx7<#b@u92_PJ?=qvnI#VQ0R0fZ^cnQ;W9Y1$r(OK`Dz+<}`7EOSPO8
z=!%j#)(_YwH!>8|=(DLr@?33yPBn(n)Z=KI{Na<Cxv`x(mZBP+1v%9MC#C2785|pW
z_sVJFf7wR%&&;9KzhzjI^}`J@yat9TneY`zs@uo!A>yw&$WL}j`JX|A%7X(-<&&}j
z7>+`1!V2Z4iB-Lp*^MDhh2_3*I>X*%6z{T8Kdcjao)9u6J%HO_b*7yzMOa*j;|(~h
z5V4h9z_G+Be7v7egyT5whBdJTNtR6@SJT-99Q%W|BH<@|t&!Z;PPLf5N)N@h+Bl9?
zX<s*9+4Ke2z>Q(1>IfwsJIK5G(JZ=r5pQ<A&%+I3vQ1E=vz|`d=1LdzO^INab;O7!
zl_#KGs0;nT2_aQ16l8U&$J3xfE#zazc#($>1wn-nqpHvM+4l?8vg?sG93=|#C=OxO
z#RrFpun3+!yC<~i1km^Woalj=chNEOJEc$*RyBMDzR+4x3MWn>2H{gigpNVyAiCg0
z4u(y2udL6l=|%0WBPEx`r*7=8I9cgmFLFB}`z)nnx((${_9I;)=Mh>4&ZH8LkPt+b
z$l$mgPEq<zj_P(F3c`$*+{NzX*dI_k&Odj&xFr+fT3rSEiW!tB=)aVS55i?~J&$IN
z(tiph$J1A{?)Pl4+qW-B9Ie{(1u_&!)TeimvZGnOQOuVnotq{O=R49eosU@F;~|!F
zFlr;h7;`uHN&2*Q*xaG)r+Sj-2kKy#X>UDrlAxSpVVhnyE?l8V*e~4=G;o^f+twpy
z^C|7~OqxBN&ERzx{wA6u)1LB1M&9+xGDJwfS$<jbn3v0d7hJE+2lDAIbbIjPau7vs
z%Ue_zURd4}Hro7ByD#GL%!#|U2_l*T(!#k^>$i3fkU16)GVc&NO5>CoXA70K3R&wK
zPK(GZx$Dop%bCH98uV=`4CGK4C=OV4V6;eOC%F+ER^$`AtUBAVyYt%Q1^BFuPcyK+
zlhX>4bMn@%J;)td0$C9)OhS-52!qGTm{(L`F!6jU-C9*WQ$KiL2@(Cu4^y&;Xn4}c
z`SoL!oVr_<M*>BM{&}2OKoP1e=51^6Ep+U6-fEfukmJGwy93vv>^?Kr%u5A6*;Rb8
zhKFHF6;3`u=opao&4O{c<Q&h4ZeHFcD;qAT%M*6}SwfV6Hp6m`&-29wd7>VbTc^T}
zo=7OK=oQB0x%WsNYh!lvNYQ2stIpSIQGUS8%Jjj0@}gsdFEXdP!gH&hAALX3+3%3a
zw($)`i!|>VQgVzF|6|Ap*<O(gt~>ItGg?o<yg}oG6yadWVH6`F-olr~GEL?>M90xy
z;z+H8LQTn8_yaZ5R))z#{{e(_TPnjY2cWy4w74R-LccZHzZ$^FmC$9*E_4Slp=|6d
z`t)7?uO@2!C-Lzro@OmWf<1lxoKc_^7#fnfy(X2+I4!^1JcM?w>%#?%<#gu#Cm5~C
zvSL+_li{TJY%(|DQTg00QI$FQRFiAU<vcrHKH+fO@+4nDWe{1uc*LaHq0aqe>md~@
z`53)co8g>14N5A<n{#TO0bx>e?(|KygzP=aW&^YSL3cYN!}~{~jyKuDluS&|A8OWs
z_AIjrLzgFd=D`lG*%zJi9iIz&2I_3cO0nkAe^d5yd*?gWcqYb_Rf*+tsB{V+jBx&;
zd4#ikcEI((l>_hhU?{^6u`O=P%pPD3p&n+S#iPDn1yZ^37F6|+YZxu(GX)+I8;6L>
zai$%rS8|HyLW8A^mx{@dNN=~;qUo8GRb6Q626U^TY+y%9ak3OeZxY8|xGTj)X)yZB
z%_ujMn6I6$V};mVPU~a~kB0LvLp^+=Y5YXL`wOxnWTOiNXaCg7_NWx2qO1ywx7sXn
zNf~>F_0Fn^g(4wjvp&gbPXe11hTUT`5fy|bD}QRRuuhdqAr)WU_l50&rljPpm`lUi
zj9T#=G3c-ODiQ|f!8w2nzzZ!DSyXh2H|km@b(iyIE!M?H<Y;nSGq}O!{APy{tLi{1
zz2)F|J3bp*^49VAHfY4LYmj-v{zw@;+94^^D1sizgcuh^JZ=|)hEX-40m`=e37G+H
zS{dmV&x@Xo2oV?KHQ4wr@5b`e85kVZt;&Ub2+%~C-Ugv@hw>C7TCD90fiM<p->#j3
zfe{KSnyrrU&~`qDAVF>Zy>G<BZbL-0<(DIX!RO$4`y-V_RwbY>wa049P0c-X&wAdw
z!9oXVQtqbY^AuMy_9WWZup>BgV<!7kdV_s(NtiU1RAFO&W#{>#s9Bq?=Z~)+l_)MH
z?;)VeVN9y9OJt_9u0-GN%pAV0kX`wC;iA{E!E(2Gp#5i)!vh~fPjuMMrX12ePSMGs
zm<(qXrXS<lG;e-#7dI2;nU^qP5L=4jMYX`#J#+^!ut_wtydAF#t}^49OxDqg#`I&Q
zn@@W4eenwR8f1IqwHYx5_GBR&4B&xpB>y6`bEyor#S!|T!^w1COJ=)wNBQSDfwY$g
zOx{LZs^Eu++n2#|T^DGvWa|O=7MFsOPR)nZ%P(sBi%eH`E(tokipsuSVY3T(twXmv
zV&goF2uxLmOViH%(2#cL$W8szOuE)b`@*!<2eq0^UYi};5}0OTn$G0qGLFH%`A_J&
z|2tG0{0PoYmwk21Si%x1iqNW#r3$<$w{gw!*_DiY#&7@i2y5WVa$7i2>1+n=6)?mA
zyNp3IqbL>+l!dPHJ)C_wEZn`%`GB7YTYilBXl4+ZmO-SH(?y7yrFud)d!-_5GV=~4
zzuooB_oM`~+^gIfGY14y9xpM3%8l#?lZ~)+iHuEU30taYPK=28sI&ili{Fz1?V&x7
zd8$vJ%mooq=a=nX=EdlnYOEQJJ&cPCDw-$HR@F^>c=jozB#%0|&|7?r7EXt_)b?o+
zbBT__5Aam+{IFglPhaipU-|JBakuoZ7GGU*h5~LFN!~+6gb=z)!)CuGrKQ&%sG8Dh
zpB;YgT;{JZYf_X{V%Iq_(JmAt+qvO-P6Y4Z^%k+cAItY%@Cn?~T%1iVA|#i?1eKZI
zdZ-ewz0bh0)f{q}w;2hX3BFXr%yNF&`vO7N!=xYg0G4A>W+Mn^aAB|P73rX=+~nBC
zCm5zBV{^rHGw*@t8(@ykr+7e*6oi$!lMKa@&y!;GM96~!^r(RbKBG?U4W#I|!is^9
zYwWUjHtmb?(*KZtK7uK3F^VGJkR&;sL#*#Sw_%|mdb3uIzlGirrnf_{t-H`S09gVE
zgMJNMs>fM!cpAXh21NFXdyx~wW1CBb(le^J^|9NU8)vsM7#QfQ0biBlWV^uzQPfxD
zwk-Gq`Yt4gOL93v&h)bGUc2lVkGDLJyCM8jj<%?a<4byH8CG~H=BAeI$PC#K<azTg
zHP=J8Ut_rucC&r^!PeRB12Gf<L*!S~XHy2hbIhmO?|v-sg+@qMVEOuo(+4|?J=y^V
zfrIpdo|I&X%ybS|D1w)EUY|S^DY*IeQt)%zU7>0p>jj==5A`!(n^a^^w=9?(ZqaDI
zqGKeq(ad|J8Nb$m|M8u>r+C#Ez4kR6a%$ARBja>3u?zB{Eb#dT=I|*N>q=Eta!a2@
zNM;cdNfr@#n8|mXeSF^(r;oN`{O#+A07-NN`P#+_hQ5wm#Hi^DinPZ~@=+rDCF1B*
z@_@8z?P(&Dfb1b@17joKnkye75FGD5R~`*|EK=~4TMc;5ZP@?5@djBS(Mr)<k`$Yr
z!Fc&7(`VY$x#f>glf(<#MC6j=yo>Q<ONqC#<sodfX%-aqg_v!5_4cDrH`hF$U&=sk
z8%L&(RzV+`sx~XTQZBlLRbP1D$gg|#h2@S1@!_|xN%THE9G2iB^tFm7SKnu4%P&dD
zzm=b_$?0L|$nJDCK|7lJ2~gf;>qrTKqCYa}0-N5O=miGDg<rCtZ$H@|=zl1&F&e~q
ztL}1HDt$Z-XY;t^A(D2?C7hOG+a<p9=$3w)_~%<EzLu8i8|zK=P>viU$l#6Iv+8V-
zV6m^5-{dHA&V27@o-i?!sp+84m!D=E4{nD*R!{*D;%uNO4>@lj6@JF2Up6R#pxe5&
zz(i-){Uc>ApS<2|yK=m5KD51%kIu6+9Sh16GhLwlaZ;APv$zWs1jK=?M}=a6ZgS5C
z^TZ9*FCJpw6~o3=WgDrmBui7}4!(-$#r)j-)~(E0m%kLqh4Nh4JAFL4WK>|*Bxsd@
z00~hoCbN2K-jlcj-<TMgB^m?(OMSeZ(dYYhqf9x&{N|k~)54Bchd<4KuIEet&-0-k
z4=CzXc5)rj-8kn_32v)6A;i}<MLja{`b&-ZGt2B+F9ht1gvIulGw-gEpMh^w-R9!l
zS6x0W)j#WdL?tcS&!e98*&Ss+hxB5l4DRZy4}=*^pQ%G!_$mR^@Ngnte0$ZN5H)bf
zlr6)1o6b}5mhHs{PZ{kfKm96=FOW;>Bf4J?^%cKhI*lX50orOnYC3;6(|{iNJU-i*
z##dlRl}y2r11NlIS#7r*q!nvT3I&4`?QA%l5yzO1yEZ|>&OR!ey<&8;&w?@QO+4f3
z&LqXg<RpqR-b0-{+y{NsXs!RHv@bnzT5#*^o$T|HpZdK_bB6?1g8#9B>}uHgPrsHL
zd~tI!IW%YVP}UglC~!Yt!`JPGtmEU58XfhfpkpR9jY9v59!repI=*_gfAS0Ji&8{=
zM)J`Qk@06AQuSgq$*%8cv9FuOG@XzV3T;h@y9rY1e9Qg}!A{W>?lM*0CS}fkPuWts
zkk3@-jSxGLtU}kCEmgT2LY!1CEsBS^b*QUDJf02e`NFc!U<Xq5&J&!gbUfKl<~}V-
zd~`E1iltYqi6iX+M)cK49(w(Au?>JM^Od35;ntk86)UC*1K6N#z^2K@o+m0&=S8QQ
z?Ujog2Da{G(JdFdJ>-Z@d-H3}cG<q)4@@WmWe^vFI?#}+>CPHh&YW@Sl(NvP%c_>_
zoAJ)V3`qT^|7cxYyN>>QYvj+iq<~ko-^N?&tE)QsN8~DuT0<vI9@v|La>Zj0^kOS|
zOj=D;8>|DQ*fIQ48;lTQ*$b8h&x^x6450M*>&rJSQ9R99yO&u}pV=ts26_G3Gb-<f
zZ;{dT<0N2vB}-!l8!;u5HMQRtS-W_Y#WT$xQKmss8o*Y2vl^f!Ik|u+NqrzhYoi<7
zQ{zYf_8~yL^XmuL>i<_S^NJDT!d9;5Bz?vUy}MTs2rC2aE21q$LiCli0q8Qs81P{R
zDT)wKNxbiVX2Fg!5%T+|-VOM#UT$ciN+<T1s@q8r(uIYOz(1u<szRF&O=fvR4?Kyt
zdzOAzdx5Dr`;gr-n-RLmP!$-$Jq|#3$kT6lfuT>|GOg5vn|%lW^+ZxX?Ph}h(-)g^
z2yx*u7-y&6D=vp^7;=?@xm=PRb3oO3yNTu`AN_%vuqIWk(>Vc+zkn`NNpd?vTQqud
zmIe0DUNpGYQm%M?>1;`V*~f__&iYN&iB<+ZN9e~7;}+ZC3p><!<Z&!xE4PT%452=0
zXGU0Y?|FfZ$gj8D?MMKyQ=5P-!Bg*Mu6fq~#Sr-J<#P>XZn~9O-3=ow_fE(Nd27#B
zTKsm}CdI%pNK`O{W{(`b^yE(-I0Wju@Kr+Vi>iPZ{K(tXRQpoH!%rI@a-@#f){n5h
zR)H+OwLuS#$S!p(cbp%<<E$6Z<7%Q7zhBgMXZgqY{V!g=`Nc&~J74E@_-E16y4(Lh
z_P#t2%D(IWl1dR;EJaL(6qPn<W0YHxy0us;Y1(ZuDO<v&RVit)bZZ%<qQq24mQ1om
zi`|Gawz7<EEMv@a`JGW1@;vu*zt8==@9+0}`$LQ_*Ua_(p6^*c=X1_Ub5%@siWC@z
zCV~5>dphFoZ>efteq|w?B>V5P(+s2Cl=KU?&i2m9T)XD;>rS@NuGSxWYTCGa-r!s{
z1ki&Y8{aPprV0B0;I}4)t5xq;yi-2f<E?zSM*Op<{^0Uwp6*3(&^duBf(82gqf_+{
z&Y*(oNhgk-WW49CNtB$Azq>p1B66LgH6ALxhz82cg(O6Dj<;wBPTs~PzARwzM~Saz
zgic@9(sXw|!q+JN^%Oe5lX-r~@w$I1xC!^%mcSH?K-I+&y5q`TJF`#gJEu&6CdrOi
zJ?11-7q25%I5QzgT4L?0Iq!~nJ2q?{kS?gf;G(<u<Dd^c{*&_|c#_5#k;gYI-2Ulv
z_>G;FI+YeDW4#v%$ZY0aIElBHG-zOF)Fr$ZLQwKu#XyDC_oEJ_XCwbsDc?ikzOA{&
z1Z>~qyTrh!$F8h8O*bJXYdk35y@{yU{_L6X{9N2wR~<+n`1hdntFS(24>Z!+Cnvz1
zvpf-4+JCNl|B>u|sM4OTy{w|`_|30}wrV+Mg-biUXhCa5;4oM!he{~*fASH<bi`j4
z<#UzpH_DQ%ZoW4yGtT+U=o<UPT(64MGv2>j&U=Np(Hm&F+4&;>_}R^uCVy@hYEtv>
z7v)cMO1sQUwm)yQ**}umk>%TjIKDw?gGFn_alZxc9crfLhVT41pvjZq1YdQ>vsrld
z_`j^;A3bn0-jI@Vxs0$-y-vmAj*xc7f~czbv)^<z-|Y~M5SjP6eMzag(Fkq7B7`9-
z{R_rq(DcM^d*g0ob96UpiTnDA;-B|iVw)y@zKH4fQ0Ol)6Ct?pUq0f4%VqA=D%xP~
z`pm*GS^d^z`8%Wyl%0uSPVIY<3jEu=v5L>C#FdPbWZHTzCFyy5$i?fbLk;P7v29H@
z?nP$K>_v7_dqV20|HMiONX4l8ic<;>7DhXizo(7YMI3bYpT48Ms{ghv^YOLgQLGdF
z__Y3Czo+oaHv9X_cIxSkb?JF>d`0a^%Y(;KWWlo9z~Dze?pY@j#N3}1zV3M<H9C5>
z_^TIIA_|lAPmDPp4s%X59Q%eK<>@1igg@~g%=I6#aQexGhmr$w6tzo*&gq!uI&P*C
z;f%i(rPEchQ*(sh^*`&MpYV~-y_)TvDFuh4lg}r=NiyMn$W<yebL~G3<*d`#S`l74
zW_sZ-D&Q&Bm`J{}n9ZRZIZ@MIm!&Bz`78$&qTiJfd0Sf+*9~YguYfg0o%9!PJ*e`{
zE<XU>@u6ya@*Cv~yS^sH=b0v!Q}HEbU^J4^67S^$9#q7@If$PVIsxDBSp9=vN8Zij
zPP*4<x9C<*%I?qhdsfyJ;Wqdt8ISjTM}Ca%L^@>KT0)+fh*VzfenjctLo+UkF5hVL
zR~-dtv<EK86Bjb>#zuz7;rL?>g4&#ng@N2hi*0xMYyJn4h&$D3zv=6a_mN+{*OdrI
zd#-tM(O=x=s<HA0+_;Q&7J1dLAWWb;cwOY#{gDhBJsalzgrO^uZ<K%8VhqYtlssj8
z!!l-2@c3^sFU{r$RqkO{N0oeg0Ji9h7B36i=e->LlQsWGTVeHEn}h$OU4~NFzO$BT
zcF3_?u474-#%J+0?L>AR4+^ne?@TS6-}3FGaG+n#rTFC{4W|8%5KV@mjaU3~c}~PS
zlRxup`_B>hz@zSCQwsJKMnzNK(-s|B<Dj$c<mEw~(cCiFn&&UvRJeHQjS>a59A(a<
z3AQ>Xb)!VX!_l`oR4f6`vD()@;uPBZUl<>OM3E#xA?&7MV*qsu$sGC;xtUICmm+qX
z_n3_#vaVxm{jkPfMD6uJpez0sUn_JBY^~W-`cZ*6q(2BR%%{}Owa2=wS>f1P2p{+g
z_~Qt_L%!Aj2;J2k-eJ0l{v{eb(PsibL)Xe57pS#BT6Cs7f!#=APke0Z3JBR1KQhnR
zm^W_o*w#UBKCak}jOl>3JI1D}4@92w?p9nkTYT3C^!7fp3PB&epNSx(P3C%J^i64N
z^swk|_VR~jd2t}*Hz+hhou%<bFCIA``l2F^oj=lz(gTx&pRD|xPbsE>28qcVtI!Yb
zt?+7N7uvk*_9^jRWh{d}2I%TI<x2GIB%;8J0rs=jQXH#?guTRf!rk(~%}3c#ZCmiP
zkL%jUcYWR#%mw9Fj^~9WsSS5aE0ayZ^UX*3gBkA1wc4Fxm~&BmSotjMjJ5C$K1Vq^
z9u98{{tND7f*RdJ6-%k%<#X$WZ&TnQ>Hh5cUgUl8hIkZiL&fojVE^%WfT|&s5P`l`
z-N-hAn~SDJebB?!th6d1@Wz=b43PJnuo0cSlK8WLA`{rPXy?owE91FmqEXsXq1*jA
zR0~&pyKoGst7~Q|_MqM0O5|dRhPqP&`WLx>_-(X%B5nfw|3>s>k&wA1?0#Gp1n{$-
zs}lrX@F}u=U&3H92fL=m<1l@AK02&u{cM3HC8p8ZWqj&s?D64l{@B{*p0YjLJhVa1
zti1pS2n}&P4|h>l%|`=~y?hLj|7BK7XxI1vvY*usJ2mLv`1C)LfPYeAYFkfgmxjU}
zHz0NC%6bSoBzE^0c<vb=2o=BC-5xwz?~Hp!5x~shI^g!P@SNZzb6`7GrVU&pxwehM
zJ#7pss#g78wCfw_Tq=(5$_L@{g#g8PBKmfB4JHf^;M<uv<Ry~XCkQk>xUDp`#F#hQ
ze3E4fIGQum$xle^aiJ2k1fVQ&3fxJz7kNp_d8STG$F)zyx2U#1*)V{$HFH7Q6t(cJ
zXj5FeIkC#0#Q?wlL@#nMde(qoiwam*xXrF!#KI1k+>BmiCv-~vzFSCi;TytHdLyai
zWn%NMk*gz*TpgkHTC`mTp<U7+OZ4ZBiF_dVDni(szqUj5`Y$yv^-InFQuDvm{4X_+
zyf+hU6AG=if245|Nup?0Ev+k;?fVsGE`)Qvz|#by6%jhYavQc*i*ILdZ;oV#bI&-S
z{OITIpp)r}wvT=aW0nB>4UBgMt7;_K{Qz^mFmQ$9GeP%#WHmoR`xZgP{!7ho(f#Gk
z|MKR4dGo%1t2ZCJr@&eMMv&Xg1%9`tpUkUC2$i^*hGWzm!@99|yY$^9P9Jo(v03f9
zsLAT^$t%aQ;*L^EA)MooI1Ol~_u@I>+431m`ZLPv*DK}u9_)(>rQ@T~C<vj+5d1_z
z*qwmApkl9Tuvi$M@T1JkGW<-+S86dn%F(&Z&8WrFa~nHPL_w9E$qqTF^L?!7K(z6>
zxECn`w6rVLP9k0t?UF-zM|KVdMG};eu&IV*RqRDwJ*6b!W-g0M!V<MQLPd-EgUf_R
z=vyt&?+`5BfM~0TQY-%u;vI02po?^W91jB+`B)c)PU@0=XD-n6S$AsUpcOFhGVU1%
z0W5VL=Dc<w_*o`^`=~p#V!A=NRwFKb^1ub!`+tX6%#+MJ3c;9Z&Pec1!Ovx`|7nB3
z*?vEG?aS>~5<1Mi7rUF8mlDys3S$n0&u5`TZiic8hL;fUc9-!aw<RVdgQA+E!!;Hc
zT|Uh)*fUkQ-e4!!8;ao~VaiZ_Xd|t=7lG~9#pP)T++~gO(cj8dw))ktlcW>8@omLu
ztQAj6!2;u7+!e_dNeVKoiMew~LpdmK$)ZL>UBggOC)hQcdRn^_04eUV9@`*HhmDE0
z<9@KR9_hEpG5%jj!I!i^t{E)zL9om}63>G;gCL+}s+I&|@OB*)OAN=(%n^3YKp9<<
z98yWb&c<YO07y3XA_+XZz=42V{M8a{wI+`Jjq(vf+>IbE;s#g(xkDgUzC*=ypeeE^
zdfQoNK8Hy$zyjaxgwnzHgjRA#ba)2EiwX!uf%kHDe--usrc|QfU`=am1&#Nqe$Y|5
zMRD`AiT%n+qw6Y!U&I$%pvkni=2I-=2Y(*$JLE$oxgNRT$KcrD_E_T?9eCamwBdWx
z?>>gvCpRMuYJHkX*ayo%$CSyvNHB#1*|Zq*gSume<~LpAQr#jk{5@Fv{k_QAr9$uf
z=v5X~c$iGw;$OEbP%xeO5n!<&^RjFoSPa3M22XYozIMt(PXIdd&h9Pw2H@RbI<l`Q
z{j1I_?wU5r8=_BQcsfyL=a-7@HrFgN^LLD1GR1LL=F7>8HO8D=u{Xonxr5M5g0Y}6
z)igG{5bca!iB!!)E5mw`QBFNO?NA9f|6>aZZ$wE>%jI%8CXjVa)lV7q`F%%rRqmvg
zE49_?8xH`)_Q7jjzAmoFLU%!PvqmS!rbO(WTe#RPF8$kr=BP{F>y9hN$qGob*AA~T
z-1L+MTx}LP+zn0DQK#g`*4%CAI4)DrWW1>R++p`yqCGhNeDh0CYJWNsg+;+&?yl~+
zoVE!!lo9kf`X<u+8@_23<6U)kx`y9YH#D<i9;L*owdrPd^2I+~!*85j_CkBzCa+tY
zg??dX{|{njs>y|&Q5vRH&EKvWS<cQmku5Xt_@$={m}0jkT#;C;o9oW<#}sF_iD74R
zC!@3^W6Y5JG#QXhpgv}jn4^j)oqFaSt8{U(xHg8QH=%QZ+jWdrhPH1yj%$9vu%%YZ
zrskf3!4e)KD%jPBbS=5J>%TgJeynf5KmXe}%iz{9=^p>lPAjB)TnUFfAfUOviZvcc
z2a0g&3YejifoPQm!-_YymC`nwfT_LmVV+<ZUx@Zv#a$i$vba^n*Am@mZt(JIV8G|l
z8K+8Tu+4J=n`klN++dmyp1H6AWp7hJ(BDFWAugiOBYTl0CbrB6e8`^6hUCHfV!6KK
zJ`?~<k9u{CXHVvx$l_TGPwo%F29jpg^w=M>%0neqe)k4Ez9-BEQ2PXUr-(vOji*3M
z+V&!iOM4M^Wt`7|>Qg{rRZ2edWZBdu$3b;p|8AlS-Hdtmx%p_+F)b3K2hY#;=t@WQ
zCnJtQhV6>OO{}IDS+w0*&`W@!QvLe5(E~CYxa%kt->4oG9sUOHqs2Rh-Sb}LEZ_=|
z8%gbx02PHCZ{yhku-yI&*q6y^yas+g;tP#?avr`SxTi#A-~#u|+n@zGfSUs>8yRRk
z#7pQ~$=(s#U;Kyt0?A;!1FlE&3juo*glA7Bzp>?g;=M#`sm>@zbUVn}KFatAgxn}Y
zWHBKQCL5tV8TB4_av7z}nE8&HGM2j$^<f*I>1^IwR+dw6Fr=#1oY!dO03^~=-a7da
zn%Y}i*S6)J?|L`t9{u+M@`Andgy&qFdhyYWu4%%)T=2_$0^QQEGxKp6wh7a0qKHuJ
zUY)3EQpC5pJgT}WE~kzk4=2IS{OawHMbj<Dc`f|yK2JAP@*0#v;+WS&l(rO}>_wh>
znBZ7*44Rd*=0CEj!#>X<pO1hto80rbxVlV#Nw1G2(G;bI_#nF|w`lu=8n>gAtUn!&
z4w>^}X4R3fTh3(2LseDMU;y-Jr&g|PhKbjFsO`SxF%}+}G#PC;W}dlJh^^Olnpw@_
z)Zlz8B-(=FJ@?MO9c3r8%vCLDk<wHL)%BmpPCp%cOR<<;_{dvR{tCVhOw>xu1#gA+
zXpkiEP5TNcosU;y7rO90<G%0-KTbmHUNli^#Ugm%MdU?J7I+accyyv0DD9KL<&J*D
z&(u6HccS%-<7c)__@)(cU=60mEX@aVX0ZduDgGgvbq+oXB<xC5>>EgostEXNO+gG}
z3fynF*OFjIVD((&bz#m$toYV49*(_P1lRB5TZCw_li*Y`-63-%&jX#tJE0BZM}Fi}
zwgtCJMlIueGuFbf3@dgdvHNC^Xtmvo!Jn-MvO34sfNurqJNjvu0}r<o!F0e$iI(I6
zajCz3%_@!Y=sOxO1_~@9pGRxGA+&=Fdbk&1+BQW%T0Dcm;K}li;n<U+nJ+b@(X38F
z*L$>no%ae~{>dK5jM1+RCd|B7l*4Rb3<NORpZmci60jWx4}u0uTu;cO0aYU(;=?`V
zlzX4Tp7>;C<H5ofqEi(9Ez&gEYM#i}MJ4coo;qu^Xz-Y^LKYj4Mzhx)9lqT2EE31b
zD^F2|6<kf5U?a;MzyH2%H`ArDWYLmw_AQD_$M`Fs-EzHxvVUo_9423{a@9m_xvE6C
zFLqdG1)6Qg0*T%)0b20E(L9lSLR%ZL%m6#J(iIb{ChwzqEn{`m_%s<aWGxw;ZN{$Q
z^p*Cp>&D3-N*(sJ(Hh%=R#U%;duFx4@bFdD!MsZ=iI|VQE=~eGI-zbHS4s>s@&b9^
z$Ao|4)T?qh{+Q?(XeUI0py}BO_@^we?JMF5+H7<an3Pl3vDfjmk1KhX$yqp-sulN^
z6p1+|1#_e}IfEB|3{=`GJi9>!bY3pk8~%cRKR_TDlst#l&c=VXt&6<dmdibp#qraX
zcBb`+?ec_2QlAyqHTd(uZfgxm?a^UV2Gf267cJvzkCfEY`Z@i#d8~AQ+~=U*i_GHE
zx;~?7X;E(}wS=2klK)rTsRL0(<URDQ8I4!N^*d-5eF5IU0`2y1Un=C9E-*?kQDX|I
z`t^PE*@Ut|mp4{0vlk_p^c%RQQ$q~?6bu@~lxq{!>O3@H{R%<q<a<QxLC-E_h;Yxg
z;4AU$abzRGSD_PZ`J&w1iSH4z^gIA(U*A!>ydG^ImB`NKrv;Jf{<J#(dpX?aj3nm*
z7qki&4%kd#tQu8I51l}eMYHjYaGrOZ25(|9_an74qq?<66@!oTR0eGOJvquBWr%uH
z?nP3&6)J?M*Q8U}yNemOJ^x@F!#1{7s`5PCf_o98GH;zF<U_pM_}Uci3XmKhBXBxy
z&@=5mQLK!6G$;~i)7}DTJN<Aaf?d$m=u>&-*rABxW+Q-H30)BR|M7U~_w?%kj1YIh
z!o7}NPGEQYv-8HP^lzKXoynYFf(aIvLA9yUNO0~lh9Qzp-`$2De2G?3;b4;Z7zA@$
z!eN2OK=dyFFANCIN~l7C=rw_@(G`8D1H-HK8FKo=5r$fMtMkJk{+F2miRCF>(3cPp
zo+8#Z3_lxghZdR77v!ew6Lc5|#rRq!I5x25#i9d(Q4Z?#ii!^hmTYBuR1&3U)OPfh
zhG4YW<+lo3!=!X&<2xsrtKAXwB?9=LQK)x>Rtf)icn|StWhkZHF_^@h0J7d`>7h?>
zfG}+hT3e%mZ<+?XRiB8kd{atzK#a3vfq3mx1dA6opI#lpv0%;s)+D)_z&r0w#V*B%
zO8Q!ZgmYF^)3|?Ry(Y4I7^+yXrpJDQm4^i=tmQ0m@PYaDMFro;m#l?Lq3rD+Szu>w
zlh;Q$*#w>GFhTDD+F;_$o6?B3PfF}F3-zP?>_4GzWw)?1!1e;0Vekb^66=>v%R;nP
z$qqJj8-&i%8;06z<O_qq_0+@ax}wbz$<knFi|{Sz&aZ>^ZwR(fUWl(r+X8QHCx|Gi
z!eR_3tsE_(Fgj#8bT0i7gGh^G<2(7YTYqFc;+GP;z@T+Cb@>dMY9y^cH6N}~f!k@%
zC$r><*i`3snbBIc+NIN}_Jcp?^_M`6{RVb<fPZZvh&l{-wl2E`!G&GKGnB8b*e4k9
zS^R0U&@NeDN)20s{SDu#WZ66vcR-HyVL2r%W1eRgC#;96Vo5v$XZJIjh`)^-ix;%X
zW`LH|uxqeHe23=NfzyoF>P0FaqV1w^_FCe&O_A-FDdzZLzwy?&%)5%#)^de<k-d{Z
z$yL0>UnaEg#AcwSX^(r6hFx(LUT-PwN4)^5dI|)QeBv%#mPXHfbRxcKZ!usFJCXJ>
zj(=4*_l9eR24|`hjVDo-GX3Wf))B&0IiRO?YM{r;u;J2R#!_`DwebxUriKlLvo&RU
zsALkQ@;0gc$X2vehSQTpcvj4UKOAOuLq`qLpr33ghq&X40*U8zZIDN$-fvoWbvfxw
zga5w)@MpxRXizCL?fk*p%hNw#mLm9SpdfB6yx-ebIGzWNHyWZXJojQgEPpEe$)>%7
zB0z&iE%PPwDV<Y(XW+PJ53n`bA3}-a+B}|oFS2H7D=g_-Gh;xH)Q~s*GB#*lA=;AV
zUhF`rj@*brtSipRxJR_S7fD91K6yMi)Yp3fi-fJ!<0EaFAF2A)XwP^_z~J6fR}K1P
zL8SK4fI7M3>)_TlU_in2bcbQigAwAtJ407UkU<kYE6Gz5(={jT0L`mAUBCa_!*_Wv
z&`npsU30_-?lb$A^2B+9xA9xx*0|uiHz9jjNtj_T;)ib<1H(CC0F?{nAzn68(x+nS
z+br(-9~KxIM*SSTlP}SSncyNJ#_bbYO%33wbywwrTSveGvUV@o{eiO*&GzgY@+|?1
z^TDD#kxoOvz}L~SL`sL|I<hU{yE-FtKWzDqZocTs?Rr7!YQ=SdJPGh*2Z(v<{V{s<
zf`Pytskz^aY^QD87@mziOL_Rm;C6&i*nW8I<$Qqo26?Ij;RhAY_OPhjQVY)bPw~5x
z(9Qw%1^&X=21*OQLmKo~F!!_*tsB<^#ZUdhB@mh?*a`&gbA)9fEb6Cvs$q}d1jv)v
z^|X%>gCZ{wn+wM50~d%qv7xm~g367C%?tHSFzIaV5rGJa3MM5_e^=icN{&>%`^(t=
zGPeD3@t~~!e_?E&v`(bQu*_3;&tq8Ln)un$+FiS=bJ52K_iXI%+Du$Ndh-2fiE+#8
z^%3iQ!Zku$JMj{!Q=PRRN*oqYdKUPzj%#Hav!(gmLhoP#H&|EkCRFn;fum~vhScp$
zt~*EwaAvD;Y*?)(teb78E6x33X88%D=0oKcA={%&uZ`ZTa<5AIQ|SbWCcBOozwG%h
zdoIUk>dds1Mx?*p;Tz#sSzHltMx+~@hrE^&^5cV6|Drv|-9lezH5)OgRmPcgng~lQ
zJ#1sa%$V4~nH=d`tmFGX*Lmb6wV!avdf*WQW##1@ur_&(_%@4hv|ZDQ(zcn@vqIYl
zOJ(BNYAG{9C5uR((wKp)V#ejVYWz7JB{v_A#ikc+yPqAOPB;)7U%qncezPy57d?CP
zHf?mxoj-(eLB&jRlK~Km@#(FciE(7!G<$;dzv&_5bC!o0G63PsA#bS`Uy6wxIRid|
zzJQPL#5TB0$~~~aII1D<?emuht~F{puUIkR!Sit~YCcSVAG99-5Uo5w;F==F`kO}W
z+-<SRHJngJGa$yUxBs@orDgfFvfr1qZq5@L4Br83q?73OD7^wt7vO&PQv#^613Ur5
zu>gHSeDrWz?RN)bFa`U8z*`5Yzg;pAa+?o^^#egOo%<6LAo}J=lmR4a5V;(1Cje_S
z!qy}bIsV*?LF`hH4X`UtfG8j(bWvd7Gb-R`41x~=bbuEv0_7nCU}wqFA-=?8cUSJ|
zW|V#nSntt&4v0waNHy>OSx)f3m;?A9jpEDjR?^~pA%45}DgZcRZ=mV{vnWQG3Yw}L
ziyy^Z%TPV_jdir_Oo(Eb^8qTZ?4ic;RHd1oBrE%*XSWX26j8qgH&uc247bST`E=5F
zb2b>4t->I5NGEcpYHIN2xFZ}Irf2~du~njak<-^9s)_6JX{zI_-|m%Hp};u!!T+4(
zxpU{3>t9!Rl0S0R9F_2hU^ahhy~WHxwY%8)5WAHCR$)|4<hd50%vXq)Z$E$}SjTSi
zSg`OVRGc;l3)np#gbM~@+-04)thOSQ9@2*m3GfmI1aH)d;b6V_d_W{{i_jFx$3&70
zk3$FpbRDUwA4rF->qX2_AnoTu3>Ah(LOeyb63B!d<yCb({9f|ya%bLPS+`4tz>5I@
zK!AVXoDj-SCh$2u5RD85ks=4HG<r@zTTzsw>0+A665u8v(nHt>nibSB)npcS5lisr
zjmr~R)Th7)ae<?z19$;+X@F-KC~gS^R0+dJ<LU6QOH6@KXIyQ-y{2)V!kh>BO7@1X
z5_46C5@slyPT<HmqdXON1heBW3FjC;i+|mQcMI_AEXbthm&7p?JQFzU(Ck?3BidCH
zjn>n5vM&GfJbrl|f3N4U>&4ULq^->SX`^mU*0#Ovk<YQRlY8OAS!Ns0p5MgwENFC5
zQ;ENN@5)$*b+t1pyF(=buzCjR=x$<xe-rSRJ<B`E-)>?uR_Xq1xh^7C0xaf5m}xZR
zHT)I6@-CtMsCzK)bQ8YQ)ME|7o)5tu=Y_RqLzwFVe|0aidIQ2SXDWdXBUQWd&hG@g
z^#$2CvYAlokF7&5(3msr2<*jWDQeKo=RFIcsM)S<Q?QiPphX`v6Q)r=k!10;AF)v=
zHR}S}`6@!VW+k)rQzZ7gSQVA$6ANiV#kgwCW?GCn_m*rLnImZf$5PvR6s1{~k84zG
z0j@#90<!yX4SmSLdb0fZxiCP8A1pS%lqW(D=1y5xP>qiBSU>SDadF*7)2yTS_MtQE
zI!|!;p+Cumr_LZ$Mu$JLhE<Bs<Qc6`3d(Ge{~~fAZ+-syB=W3naRH|mj~5ZSo8bWv
zy4aFx1D?2hJSJk?i)4UX6vG26pPS?rXhm3ex%$iWy+NZ#pSVQrST+1wYUkdn3*Qpj
zR8L1c9I!bUtXFvZOZK5d)+W~sYp)%j?XAfNspr7gFEHiL!Ib~Ta@VLG7M{_KtL)2H
zr(LR9{x~i?41K$`oY#o7&ELkn!wXqz`FpED5f_~<8#KjmbpB1TwCCGzghiW8mKdir
z<EjZ-^#p&A!k*)ZR;G0KBA2LDxf#SBHJwEO9DlqEVk)^y9PujlDyB{H!f`k7u4{g=
ztNIw5*g_)0JHK0;Hc#l#nCTBc1y1EY37@e)glkHE>7fb-X-o}~Gkrd#b5op$w(c0u
zB#4j4fJwTP3^zW+Kb-VxHI6GLD%OjrPCMR<?D>F)o@#j_>88y1rZo^T5Yj1}iCqve
z5D8M$GTJi{t{)1pe3M=YrSsq`#A6$nQIStH8E{KK>IkFM+za^5gvc7XpA#v%ip8Ae
zXg10R75DPAz;T#-op~81&;T4Sk7bI(Y2Tn2%skTfo3=VEjZMS?R8*T4`rIr2A(jiq
z0<;mgIImbVRg+S?l?CCxWhaS8CS3dFCiT%Yf4NEjmn2*ut2Dadj8@jW@;{W<-;2#7
zKDc_TdGpPxt%miBl2o;02u}!YPrblH&({S}7q_Rd5|M^ThdNBx1??)>(9z1;ju(Xl
zW~dvSma#8rTuT>5<H~hWRX6ABtu2ajw$r+%8!vNgUfZ+Lr=;@&m(Ms|Z!NsCAL$+_
zfG*A*>`bFZA8bBjC-C#2yFDaNQ&J<C!x^~9e&h@d@#kDl=mfQX0^C6Hv#F28RrD=I
zi|2~V@m@F(HkkhgP)ihrlhq@a6J<FCTMNYPR$Q(1sMZUhSCsMm7-h+>W~bet(ym<T
z%VT~xHpoR9lC7~2$m2K-<O}L|gDWu5LB;eVi-4^O2abQcbpB?c-8Rr7P3$xJ8tt64
zJ6Q0Q?$jbaD_oEfQLP@I2OnyTa~DK%YovtI`LURJPA?KK%e^T95#C-SNuESX1Uax5
zanR^l22}MoK{&Po;w7CA>2xNID>ViCT?ayn)I=2XC0|f)rm*Iya&LKpA9LCX{*fk3
z-Yil^d4}Uc^e~0ew3e@G<2+0WwpcpazJ&Wb_#k3CL3DzO8K#Id(<&pkX9>ld)Wv{o
zwuUY1<Ihf%E^Zn%-s=OQ{RlyV(xeTJPkk4864`+lLaD7xQ9w&ifTNO6*9C#JrU1=i
z5~XQx5uawUff2$amT6!{t^9b_8PA)0E%}hQPC0867?*8~T#Vu*$(!$+qD+2)O+m8*
zxFD|NFT%j&*hjuF2O&3He1mT{_JI5DQ2>?y+kA9M9B+}E0cO~Giqa$u<oWuuq2iDG
z1*N~F+wS5(owxV$hCVLF-1a^=XCR+4b}%jEEXeH~kW^=E3RI9E>m<;3eW>Gl!2<#}
zg1QOF=ka)bEbe^;RPRAN2}OE{S$_sd0K0*_(*sBX{^&3=D64{%e$hJxiU5%eK?(+C
z5Y_q4^#Cyu5-jzz5U2s{V_NjFE$DIr(2c3K05_vADi`kkZ>;T(?fl-?Z%5#OH-p)w
zsH53i82-F-&XRy_RNAGmK;%^nGV#o5MflFU^pl~IKYJd8BP?TVqgW=5x!eM+ziG>I
z_ZI!x>!5<u(aM#Q#eQGhfu)9^*u_xE^O4QO($m;F$_11;lMYcrMao=q9A%$mj~oV|
zOQ}y3TK~=+VLb%ZCmEA6e>lNXlj)j4=~?3MP3!8YHs&p`Wo3G6NB*gk0W*;P1k_>J
z@*oyCWrx0fz|g{y8t|Om6cu_`7&ac(V^(lFjoo@5swCVIo>ZT3MMaamUN0`+u%>pR
z52@fb?{);5`-IY+hMq;XMkW%uX8_ZDjMlHc&RMm2`pq3^>$zS;)KZ{Jp?r!u&^02^
z&sbaz)wG(zV*BuN>ARu^<b4PU(?Xz)$-t2r%^Xhdh=y2fSg}cfP!{N^KE-MF%@2zP
zMfK_qAf*Gy&LA~OieFQ*sV@_SG_T%)Juc^c!`8kGX-*#ec>q-vY*j)Z%FY~sW+mPU
zaxf$P1pAmE>Vln0>|P1-Geafz@l9VMZ9Hfn;9feRD{CX8fO^I8?>6WNm2eRRaMZpC
z_Qn(1MN0SYQ6@}6n|2g~#tWoB8IJFC3UR4@s=KZkXb`gtIV&J*dl|yiuqeCgl_P^Z
zh^GQ)76@ZDOBj0tk+1%%Bj|KWZ5&sZPiiv-7Eyv-@}ICrVuilOb>q@ndlt6(5`Z{Q
zVJpOWtROgY!FXJkV1ahpF99C#-SW*MX#rrSJhQ;N9|i6+4b*DsEBL2*1M6rC*zsYc
z5nY_4S1O@>EB7E!qP!yE7O4aY>Sw}Jb8Y}-8d^#METH~~d0H%D-fGaa1D-jm3T5+(
z(9VZDxw7O6|I!$2lf@fQo*bZoG;E9&>U_{oUl^o{MBa(^NyE2K!L?{2rc|_czGD}T
zJ?@DwRp94A=|FpA!(j7H185ME76nft(9w<`@AALOVU*9=yKF7&luZS!>6IrWeawUg
zU9flJ!qEEJx2y3zqS>7UU|yQ^*K-YUL?R{PVLz^V!GJA1^k^9n@j*JKW}rYnmb@`E
zupR?!%=)1X;s${CqlK7mF!(A4T)ITOK)1Jn3-XdR1gMyiw*}g<z_`#tXq6%h^6Cxi
zJ?qtip|8CcMeyT%L2pAEdZ$1%IMAN$JYTg8cQib=5Y%CqVEf^f@w~J1@SQ6Wt~y*N
zJc65HUXHeHHtt!@fyOao8vry>OHmG$j6?U~d!}KkXz5wLcC}?aPATS{2cO4^1=z+u
znK(X&F0%UZzs%<x?q=*(+Wm#2iDsvItb{O>31MhhiMKkIkG2!Mlz=<+h%Z!uHJVMd
zSjg(q&eNC+8=-1M#SMl$#<$(XHip-bcwY55wuQ8ZA;F$I8_vUARvqy35=F3}3~-->
z2!*m|(AqmJb?h}TmCx_QtnAo_+dTa`Jn(J%qLR-%^a)o-e9lc=u%_me_8j50zDWWa
zd1hgpP<C7}#v-<TFD8`!m0uF%s*IFh`fqGT@XIgxC&{M2=9dU};F+DQREF_uysXnE
zS*S-Wid{9{L)&Ims?{;&MX^ziM<3)o5_S{~CF!a|pLQGnGD*yLHo!<HJ5YvcXE;~t
zB4%{%B&dZC0M2|duPa8p0=&{uxs(NN<u%_(p>C8+xrfglzB}mNOmucuezrvUw?}H}
zmrlPWmnaN+Hw0f?n=u8l+!|Q2rX+8&9g|--@T+axU^n6aMBBD#t#qgZ#H3gnJ8q^`
z%x7d?y>IJMC1+$88Q^OoaW^1v-?*GB262c}S2lM&_!asLSuEo=p%*dS{S2*g;f@CU
zFgO&gJdAQptyq!Ol4(R)t~%@b!Agou!<`w4K5y8`*Ypk^xxUhXX(eAecloX>OB(eN
zbk$P6xAJouEafv)g4dRzR(Ljpitlue@-WcF8T<*N(~3Vp=vjeB(9++~`uJ$cnqY1a
z<vb2d28C@tA31J6yl90WVg4DFq|~OR%mrHmUQ)eFkx<)LaA0tbKapwq-ehGcq1~dB
z#*vMs@g5N*HjDi1DaCcmHbIQ5qLE2ps=^yjmu_oDq&fK<88kc6jet>{alE-flri+q
z>hc<&biygzGoLs1##O~!7g@R6Sj>cG<w+l>O>S{P#!GwGy#$KK26cuJCg05ji+lH6
zIXHGD6s|*(|2bT|Z1c2g;)UH)+A5xrE#B>Y+{ly(9yf1(K<tvw$_~Qw_FRfoG??X$
zx3rD|=wYoZzBS|{xW@f0@qQsf--8>WUx?8E3?fwEUoKfAnPH5u-le*0oboV#WbHE9
z&B*9>XiUPjsbyw10STrv9A=tKsUPEINBxY8_vZ!>Mx&)SA)Ky%2lHVhhg+d8YLgx#
zvbxR>(=N4E)L@)@^rYeW_KGBnFP4V$^smk9vo{44P4yOkmTGK)g@<eM%kx#$dCnr+
zsDbToLnS|rc-Z_|L|U0Ec21til?Akok%;i322iP#eY_AIRkyV2*4M9AWqPehH2Pp>
zxwb{->a6u^)?b#gzci-ugn41g_X49>p)A8bcC+zy-hR(biw7teS_92Vq6SGx^#j2J
zd4%GjL?Xnv!^@0d(tUsY<p9KN@j~=7`b`~rpY#)+rGy5WP6|{I9W559E!sSG0r=hh
zyk=A0U<<2(@LXJ=fM`f%(ZCgw+Y2eRI6EL$@xf8)qmLSfaGv|v-MHTW#y%y#e8qq6
zD=PA?VoC(=Ox|ssg<OMzdG=jmU({OHeG<>xKjp&1_Vu3kwQ`h|kCjlC+clHU5ZVvn
z6e#po05J+m>xF@tRP_}<iBi8I58ryxA6sR>T^$aIy74<jcnq+mS#}*T(1`=nY9${T
zKH-|RU4AytZVJuKVAJiZI*orEO4ui>ea?-1I<gr$X(V9=eXYw&1Mfr7>C1vT?=nCx
z02R;9Za}-LBGCHlggEvfQbv)&SCUZR9uRnLwm7yjwV7Xl+TZy17ys$+;y;no@#U@9
z@2AsgU9I70gR&{-I6BCV8vef~LW~EVk(}vh7Ae!8Gtt?<HjZIB^cS&03{$?h<=~Z1
zYPZM7ka%u%5_@GyiV`^<Z=Kt-jPxlNOLorXDcBJvL8&tBxd!(Z$s6A^3*Tu$6=A%?
zj0owHkc}Wf<K?8>PXHJjzs4~q;n?%(r&(XY4I@{J0Of-WjdOGrEk*#$%+msg18a!K
z3VtU0GXNq`p~WWhE|w9oBC?Nq3!XJcmK(&=!ZMm6gJEl)79cc&x96F`8G~jUF@kv)
z*ksISj0aQ}c_uLou#KlH0f*6~@YL?4%p%8o%>jYF;sN>crw`*HvA>O<YX|6MG+G<O
z5`i>0pu8Ju$P3BFEP%LFNU!nCg*F7XIke0$D5G%Cw4gN2+dEhG|D-$S7vRPpuP|UU
ze*tb#K>a@fxE=ATz_)Mlm;_LO0~r6X1?p97&S@qC;>L@W{+uaJB;GVfU7RFTVYV65
zdKRqo;ay}0V};+3m})CGSV0fV;2HHIHhJIiY_CSIG!np=#=NscK<3pr$9}YaPEq3d
z;A=~rNE}fbIEALnx*xsaesR8saK8V$`o46a^4#B8``bQhYo^R(-J`AX+@XBJIb_Au
zx3gyGhzb`Tp+o1fO#>a-lI1Zu0@D`21ONTKh?f<A4q#-PR+jNeuLX?Vq$--ITScKs
zF<Ie^On!!1^4W<=$PV2qlRL(?2!`tX%0Kpz${DsCt14OUA3j+`HYX3?`C=t2B;^xJ
zACfhD8Rb|?C#2<pS3J89`5(!>N}=(RD79!Cj(1_F0e0yn`FvlBKs8U5dA)Glv`kXO
zlxEI#!i$|wi~ZJ?2J9LAsFdx?-3~6<iFe94#pv1Ro83CK*95R9eZ9I%mAkC28Qxp(
zJ%uF#S*5p~LL6T8jV~>C)$Vg|pF2TS%TGL-V^Fk@A__iZJl<A=tq7OBUZsH<kc)Fa
zSygLX!sc*C7jk#hd@zi&G~2YHW%li`%Zs-zIV-=<WVcR$P)qEOoX-FHJg3N$qswz1
zDU}{d%oWXSSj`Ez6w$b-BE^{Z#7o7ER%QPB9K+aKb6J72-V+Opo_EualGN6Je%>5t
zd0cHprd>M)lRw712e8sN0$2Dsq;()iU72!j+leD<7mYP&^4#Qw!)R!`lOm*5$eybJ
z?VA<t;Z;tGlQH0$F9tYF0Ww}n)HNKwYo6zJ6MNO!%2W3B)>%r9L-+f$I#>`F?2~xL
z4N6fP{u^%{wLXE&En9+Z3TGaJcTj+T40ck}otY!1R}uCqCJ5H`>un;uGX~g2_eyM7
zGaKshKg+aQ3%|2#!TZqK_tK<**aF@DR@w3sqbzuW*$RNOW%9WxdM`pHefVj{5nI=c
zoLu$Rspg%c)IMOn!_s-?aF0_d3gnlRJ>;IbUa(Zp@Q~@vt0enaBO`czeSC)uT+d?3
zQ#7Vj37ELYz+~`!g|GDI2B8S#a8sdo#4hJ=(m(l!p_26+c)KgNb2;B)A`J$9j1bBJ
zMx~?C{=7qp8sb^AY;~VsS*EpRw-O^aQ&Fi+@-;f$jq*k-+_<>2bpD+R=iiQAqK-S$
ze09g~rsr$rxNgvm^D{4kQhz+S7eSgRJ&N9V_pZ11?aeA5Padxzgw;QBS*LeyG|thN
z3k4g)uBF%~D@56PDY1x^#EJB6H*Xu5<Si|a{A0c8;!UeX+jXSI$lti`jdTJJhF^OD
zU%&bUj96DV`qbR*2WHOT$Td6t=A74J8gOdktx4Eh@<(#{#Ts%3@ie*8h<J|8sejfO
zL?s`UnX1wur+R)SuakIbtoEXk6wtB5w#FOpcKC3}x0X37W+ul@={EFGv)yUUahw$_
zbSEW`SX+V>Km2TDzu?{aoSrKQ>ry+$u3M8z4xk()l||Ncr{ny+CbyxNR_~Y{7S_1<
z^Zbk?i=CO%RbroOe3-75HX*dlU;|69n#2ugl?AEYzZu{GpHRtCK+tfX!Z3F`FbcR{
zWO}qJR4pb{8egJ3pZ%(u;M43<vMi)I@ILAA`QNakO;;-C9*(B&K1S7K)ew%jSzsH5
zGg0OW3vlr++9E73KE(%!ft8`8cE@h$(j{I9N5+(&0p^z+;A?AIftM^F4wpY>UGvr%
zF=stO9YO>`j?r0wX5Z%LGsymvA^Cj5JkxnNtqx&*e3iuaTCSgBIRneD84r3Z%FXYD
zYf~|hr^mF>v9vN2^WTY(IiRqH)fiA?yoRT@mjV~?4Y6l2=`z~I2}K(Yn=({>PH#gL
znaY84nXTVuz`KxW%(v7&^cKnTENp|uDSXqW0<`mB2I7Gy0N|2}M{8d;;+PtZ1kUgL
zMk26UBue`(XGxB1mj*BVN53vk6h}Nu$?N^f-=%If2G=^4d#N|)iYndv+`2VkZqfb{
zib8YBe6AjA4QC>Ed5PC_x_H>`-SiN@<bruyHtg0|T*a(Da71U1gzg77kr51kjPjCc
z@{{7Xl@?BzB79n2@-?+a>kYYl3iB#MGDA|NUM}m>*Mz4<U4A+x)k-#p?yW!c@;B$4
z*W+@oEgq+HE=i-zi1~~cnk>%OS(Dqy7w<nUaC(<2ojj@}Ki%f*-ORePH`<N2S_^lP
zvTt(ydgSz^9xvYPt?lO}jixzx6%jgDzXM8z3xVrTaL3Xp3SR0S`%r~)kM&gA&WlDb
zo}bv0mns`nkbF0{W&g(7IrKXLLQH=8NB)0j%ZdL&f2UD9@Nf{HX<n~7Y3-c*>>X1w
zzOIR4Th1YC5kJJZXxZ0I%kTj1D)t52(>{ykN_jzxx9Zevb&dgAU5^-~vtC2j7hT-4
zjX;|{2dDUaiXwz~cFjG!O~>P#4LO?JeY{X7)Ti6fHMESrQzbCR)_a$c__LR{(v3f5
z1uIGA;9i|BNU0W>myGj=5z8(3%d}4EwpL68(#LPRn$ei>CPRi@m3gNB1Qyg}hFrd|
z;O2d!CGTHQk7h5Lyhd5iHg=BW`x@v7?BxrMd;_rMe9V!1m?7)uv8Gx~e4DO?z47Lx
z#PfzH79P32;p1KJ_!-j`*BNVqh^4*6R-t+3X%rUUeNdt{jgb;+yK|PK;IXw3H0NXw
zg*&~-`jr)t)#IF(-?w-q9&u9X?0TEkw^KV{XraOHcHby0R~-27qAb^5gwKVob?4nW
zjp@qbyNc~3yz!tjt3IO&lzR0l8kbM?x=L@5EQ@5WWQ8%psy}1P+wR;AOACK@(9S`o
zrE&dNny5W7aIfQ~&TlDg_eT8lddjCH=5p49HqtJ}<*;gdJ%gb6C(`6D`#a2cxPSF*
zM&WOpUs`V)H){ITYs)LC6|tK|YQdPKF85l!YRsl>Y4447H4dqQN7z8X##0!LN-s7U
zr`3CkH`Vka_H?G+I&euI#B++ns!2WL!uh>-t%W;$ve0MY8TiRwxs(YyQ@7j7EpDuk
zylH-oQJZ*m&tA`?hc4zgj-P)Pzs&TA`=ONM(7Mx+>jE?|vJct@+#LgQspuTxjww~L
zP16|HU(=+xDmC->Z_B%rrB^(!RdqX2#b)D?4biDmuR~vmAXjtUjW1H3T4fqf_t?fD
zUW)ZJUTdUq%ryDc$wSG32PfsKuk_g>jFhhSHrA$hw)?`A+@Ietj5~%S&)wEWYiebz
zs`BJvm)Y?btqShPUR(7@?#r0lZ;Vt!zD+a<+!il-)Kk4xuKYPX_kysli_obnR5H`}
zr(MMl-$v#rUxcN{yve(a&G8^GjL&J3#BAkOFSN!Ko)$cmHQ%}}^MqC4t%&Wrq|z=;
zUbC7(&_YKPx8No+N6}+5YY}4^D$9!5od50l^^Hj}Vm8jo*N$I5TjmlwXIb0PV=VIz
zz)VeJRGl|1UcK6|Qs_jV#L{^~x{OH)u^QnJZEE#*oTyU7)j7ki_=sm#q_9|yZD}=i
zspL-}^eLm-Cfu2tz|efa`y3w?k-a+TP=(diqRjlqE3OBMC{2CaNsi>213(O3Kws-~
zNrc7$T6?aB51~gJ)4Y41(j3K$JjjceO>1pxU5StF!P$3s+)jJ{>1B<}=jPmL^xZUl
z#HpLT%J}%ar<+Et%bGzH#+_3mF{OHu4u%Mi3)Qq{DpEq!(uS>RHpi!h%w8Od_`IE}
z%~RLlh@w3(Run_KB8K!lTLyS5cm|~N_|`g@Czf1Zeab7(tlRwnVGQ++|A7~!$typR
zK6G?De(hN<?@~iv5uNCZw26ayGUm4P?frFHjU_3qgcm8FDd)zzjXPy<&hUCO=@V+p
zIkfskJD0q)sTbK1?}GX)uaj;F*es&2cyaOPCwDxiN=OKIP#Y}n%zS#SD|D{1W0QDy
zxx=x`f$5M`9Nr$)doe}uWBNDE?M3RUB{@CN+33lJIl@GN=fmERFn*&b3W;Z%e$RW@
z6UHt5qPBf<)6#scKg4E~^`vU&#vL7;Ll`<*@UQWL@A+&_s5rT4#YePYMh%aG#c2#*
z_E@JCuT`5TI$|_f$XkiiD^D|jtm@N-z8Qwv!s#-`C#KI-UeCP#{ZT<#?#;dBc~6o9
zE?Sg@VC6Rg&f8w&UIaf8rR8FAxYBJPD(SfH4!CeTjCRk<tHZa$rgy^%JP`cQ@W?B9
zsvCmJY<4`n(ES^|J;hj??ev5qxMU9m$0G_ZUZhUTho;Z#S}t$5n&T?^=zH`3f3x45
z8>LyL%|D$YpCYbxlB;seiWuRyRV=w8C*R;}`X%{5gB9;4II^g9yHNfKJ9=$aQL+w;
zei3pV<y87t!A|%rkW`6kR!h{mVvp71R{O{I8M97YvjJxs*MybD2-^;@>RD%TIfZ4I
z`fq*?E%z_)*>iH%;m2<(J#rj7oXB}61iR3NN@l^!;iiVI&h+B>VRJne#<MQSo{tk(
z+*VBzYn0Or|I~8dGhS<r%QUUENt^dyTWoNb{0bT$k93oNpxLVHnj9I%nKTidNi5m*
zl>C~a)l=*r>A81UuD_CGnJuPz?TySl<WkGqU5+Iw4nJ$)PeTSrdf9Ww^3Guz9-6Gw
zDx!kh>8<(7g|T+eR=*><s^^&7X}paZb?WGetGlw&$)%Dt1{b}Q+#)k}9tsbAa3|(5
zZTD-H<iG=yc71H0ab(7fWhK*)J13}(mZ0r=A;^}fVKqapnjitbvJ(w_#tm-fOl{WP
zd)WKk%UF2V2<zr3)@jq2SMg4XSIXyYUaE3B;mktfms?19yk_@7DWRUxB>skOz5UGA
z3wLyt7=J?xE&8;rp&4k~agE|l{QJgMNCSa?gxTh=32_EZCm4D)TIm<@GH%A3^Q)ep
z-VpQrU{d0>6{}0NT4u~!u#B4~f<Nm)9pUpng|yWhLX1e$afV)H_a%=tWVcAL122k~
ziZw5uY3HnbS@os0uZ*}TLYkSKt-YLlbHq&>$q?T6u;Wbk*vb%TDsQ7#Z3@ILCuW@`
zJI`&of30ziL$>SEYcu@PRc?6K98p{7Y1hIQXd{T9_S3s0vIv|}+(T_>I%T|Dym?XV
zw0TaOcE4d)?Mn{4Z*3nv*?inXL`-^{_xjP@(riC{<haJA`@E-;zHxe$0d4BmIwyGN
zX*_cCAzW3*3xy<}W5zLv4IM~5U)$&qfl)GFVW6{#lUP%jeyp%ItE>tk)-_D(MN!=O
z!(jzC4Nsh}%GvbIm>CqXLRj?XveO@%nB?48@b`#Z_p2|m9<h%5hrYT+;;E9UW%V9Q
zAeX!dtJy44BaEgl*IBJ3?YS*2CoMuNMz|Nb-^K2(Y!zu##IevS$q9sA<?^p2*FAlD
z>;4~lnz8Eg0m`GykKURrlx15c$aXImRA4`8-65(gY*5ojPWDCU<r@O(8|>=zUv9Jj
z=O|d-@e;d)AgG$RQigO{91D=rpt;;CE^%}T^Zbfl1l;VX?XjDO{kMXIqkpHse~Va#
zHEb<wob<l%w<^~o<m+#h9?mWM7&yO}^FHP`@3U1jk&_KOb}NG06N?9@FK^Q-s7#Z-
zUQfyKp1Sm3wH?7jA92Xy6m3%5a^KLtPCwy+coO=q@S&&OmBHSRy5o6g@mzifVieCV
zLIv!{J_zIV^ltSc@4$xe4b(q-6Uu{zDHgP}C7Z;?XiK-dyG)HX6X+iqD$aA=7}V0a
z;ad9XJyTs$Tg#peG0x!(fuXG8LpCq8{@6)U?DvmXd_n(DY@7&z_+Ueb<(GtzSqBcx
z@ab76dCd2nprlv`40=U|EK|;%Da0jf-%lkwuMnxXmHS>J_HU;^dzLVhc_G~699`Vl
zFK)BCH^y5|u4q}2(UtZlBit>gdHnQ0I6f<~Y+0Ys%9|vdg3-eVRm?3#7eBoP;K!Zk
z5iKw_cC*M=GW5Ec%-zY_#t2KpPh4#jQUAOl@we;A@9bUYU7w_qd{z8$UMPOarLR8H
zbvoe2obvQI%#gg))$EmJk|$HXm6Dh+Ed1K!^#p>o5dyx>pbH!Y?q&dn4w5Hw<w%?J
zKPEkPJ2ijnT$x6_!!y?3T|14-55H>mDy$$U{Uw`*XGZd^CHElfm2(X$N~>?%zRAjF
zRMgj^t44`ZW$BOK)J$vm-~lrClSPT)yJ`LXJU5G^;$zSzBEzU^U0ltoUSx8k-0Ej^
zOt#4BgrmK)@|j&<ma2%IJG;<(X7)6t%i<oO+kUzuU*|*2#a-oJ6VC&#0r0U4(ROW5
zv6|e{-0JWPcq{M^dXWiEJiT?67O4evN%wZk^-mX45|`+IU4AC$4e6%ZOYJ2DPgq8N
zRqW_5tezS36=55NAISnCBcA*lOPeF&vA<0<!+5+~?vA2P)u;KIt1Plh<g<5go4#0N
zkMEw%7Q)`^*;G1u4qua5$7|&Uy@Yrwl@-q!!#&8LWN65`y)!84RDMR)v?wo?vCB4F
zBc1+X)?oaVi;fr<X}zXA%gw+^czq;I#-s_2a;xkV7bFyXjd5$aZ|7{W>ipO1AC8SX
zsAAIm<`!u~PZp+-1sN|u-T_u<0L+@Q2Y|wnoE5e)v|_mIHP>S|gLtlsGJ()A(V}6u
z;?qJES=X)WzMXIhNZ9-|WwX!J6{9;cJj(~Wj^&BesQ$FpsvLDJizKBpfoqt^AYRlI
zSG1cO8=0hHVf0A(`UldX0=>1Ti8=h(>kGuuBkUG#I9$OI(nma0tI4I^)#qY3i`>*J
zc@OV~`4Q9DabKI$UKO1x%~)|G!eC<1d`{{W6W4DkM1O325~F=r&(2^Unu6mA(C_88
zaQ+H^B+3B;z@~e>CdhfHGN!!2<`#8nk9pE*Z<f>U?Aw0mN)~bD`l>bFqe{W#*1d8@
zarnKt7b%mmg{-hFFI-XV`p)J<^p9s97JM~a+hp2Pyl0Pu!9KgrYd^-VjpQ+aIo+T8
zDL9KJRL3}TXK6~oH@W6l_rYTAOE`WcZ|SA-?kvTz)Avs!M_Pvt)tSQ$UYCN}j_MfG
z5S8|OqOLkiF4vSnlr0#j1VfP7-;y0^bow%*ydq08J#V$$Ex$^p_IYw}G>KWo*+|)c
zv8wLstK?SzQGAKPFKC)>rC?V95;8O_ZNW=MdCje0t+U276voc?*uCg{{lhi(Oyl`3
z4oeOuf8cbKPi+l(rH{C#;GSN~{-&XPlql9M@3QDl+=XNJa#h|xA73eS`?6R^Hrte6
zhWHajQ`l(G+(ig#_aCua-S_AeBCelepdqH*>_JQ5yccF$#@`#Ovd7Rv_<DIsg6NQe
zbfW|f;>^4{1WZvD2j$sj{1her6kytMBs^-GAr^znX?aT<=CvHS`t`DI_9Z0;ZT%xd
z_@crZu{IB1-Py8QO`)_hWn6Y^#x}X>b=YpubtzCvR=mvwl%8KMe_)QukjV9sgb$!>
z!4NdU!|u!1l}81GJ_A9&r7z6?Y_l~b*v4V@`Kj-&8F9l^pJhPmSit7X4h~Zd5CX2%
zfM)=vx37>@=ufMvi+`S+7xCcSH0lAy6JLoVuP2c*AKdeXSlkM9Wd~iB*G8!_VDg(A
z={!SeVwm(%UnHD&I~+pSCxmV&j6AEeN`3h*L#at&m(QAv!cSEDASJZZVwEAuBTw0{
zF=sr_hV-cn<?<WAduXssbLNa93EWN!6?h->IiaGcz{OjJyw9$C-s)oNi>94lGj-hx
z_1Hb6UEA<6a0q+4E@CV~TV%lIG5DrtJm$F%c14yu2m1)Nv93geQ-!{X-l7XGqQKQ*
zoK+BsleSY@!FGry=p&{^@^ea8yD%ap&deC?70_TmU)8?j&9!BVf;Rt-mwTp<Y!ken
zzrY)*eU%b29&)>X_Oa&D4<3rhe=_Nehr{MD5rxHz%&h<9q5sNWsFiLH6EU(b4yc{z
zJ<*`btUh#{N?hEnb9bb!ToyVg(o;kbSl)kgFXjh^sLxs>7uoiD4&v`M_wU=~xqdpS
z>kqy-KPj(?(Dvxzix)4mNRtiYAAS|V-QM)~ZA8OCtEqdZPl{%odvo&237l=r-*)Ta
zqBrp!k$42TqjE*!ruTAgBksQ}Ji~7+9rirte30+BysM|y!ZZAx=Rr?A?p2D%`5^5t
zE)mnR3LXMkKrQ)2P|VZgRfbb191NJ^pZVlZ57m$}^T4;H?DO@}hSRj+j@-xYb4G=2
z-@Z!5&N^FFIbfP<`+dO!_U8w7$i0W3#~n|ee38*QwanyWp2@@|RN?7==hB$_7L<(D
z7(G{h(xu1Vj;w9FGc(s|s3af$kgX!5+8Jr`Z|R{Sk9Ftpld?sVPA$Ip)-p_3=d|>p
z;kp<1Up>wyS9g@@mRFrra`?j(!2<XBB9SVA0nrD7Dzx?&1&$_VNzwj3DNV^Zw7=cs
zJoo9bc<W40NBYLUai@3oTOWzDCc-_w$}us2aI-`GO1lq^#tuQ~^e1Y6$Z`*tqM=&d
z?Rm_k8^29zQdM!eWJUcO_hQ?%%pBFp21L1q^=lD>8s9j5#5|~=z=gJSypq$qIjUU|
zLrmAdY3R&k%xwK8r%pA<G?|l{weiy5c`^*77QRlJsqg1rr7Mg-YX~lhsm;76-;OCi
zUDx6swAqlmK62?_IUSF};!n+azhzy7U$HJSonun;SN`}X6aP>5dZ&^43JZ^?>zWJa
zg_27C818xe6(c<SP)?1iJh>rFE;FiV4uZ?M`!@>zPyzJ_GCO+Z;3+M&g^NQgC5rz>
zWsz52US9Y1!u~H*+moc6hu|OHUS+)c<`3`bpSe53E!LkLN3#>^D@I+I;dg4n0+Z>I
zYciG$@vFd1`=@(2bUB87tkIJ!EG(|=xuEp@A;Ro+QZF2OasGO?iU|IJ#b0xIhwtzZ
zkCyp7#@WP8>7a3<=&7-u%d7uZB^0D?v3+Xn#QO)@HbjWv&Km0@CU#=7`&x<eOQv7j
zJlXiG;om8_&z{E|*d8)2XZDc?>&CiQ{++5q({#()8KRFDB9+M8qAfqF<woe-eb2Xp
zbO8oTA@RUi3IP07(OX;2w@1PkVYmNp`PPm*OXz@;^8&}wcJG2NNW0}3dYGOIt*Zs!
zxhz!j=)Dv@7@3s(zvWvpkVVMci&4<#jMg*XeGIG!VTD@<wl1la1M%{`Aa5PN|1E3r
zW9Of*)`sRg;#%0NyKqLv*yZs#ry{*0w1p+FU_mfHLkwm281sba`H>&JRDnP=iiV|Z
z>qT_kk+zWj7IlBi_lHUMtqaztP7+cELTsa9sK}PxMzUGtQ%mf`s#qSLcMT&}joxm)
zC4Sz=-$zZ*5u1`(ky1=I>F;LvSH26;X`+;GV5tw|neJVU!%mHK6;x5yYS}}_b^s{#
zY`fLkXui01`4^y9fKPZg4mlpxUs?ROd^f-Gg?pH_WUsTPU$?)LnB7e)>#x3>wiYXD
z>76s%He=?=sWaNe5qD?lulTlCm<Ti_)B`q`_!G!qH@naFkFv&Q7Var=606tL)~T2A
z{Z?WcG7V9ld~K%M>xl*;QHTw++50QY!@pLjl_ZYVT0&<ipjq1Gspn#Hu=k0yt}jr#
zL=A;zLPO3`iXZt_@^K)1S$6XPlFT>KoqqOjS)1<)e~gs5QgthiCI4FHD$T7XT}E#7
zl*vkV^c&6GH_2D->>rixzF-sLTx70LC3$|gFzclK{JN9LdV9$kV@E&C{I<$}Uc7GA
z)Bx4#V^gCu2P(})|AN@+Jb*r1-b-n0rVq%zG<k<2KhZb#CI|_1zJyxyzG}M>TM_<e
z&e7`lYS~QEIg)6cwxl5gd&ax&SZ%mbMn0-V`SK3wtLqo*%ir7cHeo@|!ee=pwKt?_
z@e(mv(ghr%dE(ui<PY24T+%$TPvpaMME^})^n`iMmzyFjP;qE=xe3387hs>VoWA%%
zc!ARe@y6??Rd_cRKF8@TeZ0)yKuV}dM~;=&R+p)`oa!~*&CZPDdVcj%N#-Rn;_~(5
z_MF}LVd~kNj+w<r-Bx>qEB-+k%ctOS#n1=bc}3V<mj7Zq+9c=k&eI+^`dCl;vVT6-
z={N6m(Q<M!aE@0p>N5NhkHv|3Pm+m`(L*1qjqV-{CtqD!;hd^A>X6itTTj2Xt)Q2@
zlA$cXSBA?_%|m3<P8dx#`{udz>#P}-KL3ZkH;;#UZ~w<fk`R-S5YwU*l}b{`sFPGG
zT2NW09XXOD((sm&B26nN9gR{^Vp=R^nMwANCCij;Y-P`k7-N=qziT=Nqx#<G-1q0c
z@9+2Xd-R9QdwP4XuWNf=&+EGOonJ7%9XevU0jIK>-k=piou1Fk&aT(w9ZOx3;I4el
z-3U5EI~~z%T)K43m2{)=ZFy;yA+#hy7A>$h*NKJA@z_qV*SfBL$7NRTYuy8j01&i1
z;1~V*RMta=EPnz=hq-rE5=%Llue00TuTf|2osH%NmMJL(D*KOX<*}U{-WR?od@+Wm
zgB&1Sz+;Y*CqfID)^6;giVW|>4y@)l3+42X`Z2dcbCy1d(Oa~#ZnNS7vE>-S4vsZ%
z15t@n;={EKAeyV!$EqF=b6Rr5&fZOZ{he3TUAP>pN9rtx2oiN3AJ5rsa?Y(eYAbK4
zrM!cTj#$e^gQ&yn1Fi>6)t$GTi1p<M3htxJ(PHWXcP!DIqh!=hp2)|?WgC_#oevLm
zR`$d#yteI_x%A3I>%+HR^nk&3esvZswZ@*T6R*e2h{#p;D3rIqm~dk3ERSNJ1xBTk
zl{aSzQfUXs(ws8{*@GJM6l!Y<$}DFkB;@AE8~#3V|AoZ~r|M*+xm&XxwU`6vlZ-4U
zacr%Et>5-U5Dg}O(lAiT%2Ll7JMs9~iECT~n5i#lLH>#y1e<Mvm)l`^`{5Xsx6bm_
zE#2Ac^yI^jX<p80QOLt6v6Q=Z4&W!CEBN+;n{G~*-DE7T_sfjE*phQ!JC_r2Vf!hK
z4`Th%*z!nzyWkPuyoqv_te7Fmo5R6xiC7<+8&L16_j0Z09+|M~>f1E+jHWl9HT5Y*
z%;H(`>~m?HP~Gki`(ya}X&do5hAB5(@4c^IYv6a;D9(PP;f1_qMW?-H#;N3D{K!qZ
z@Ak5NEBIpgQ%Q(yZY*Jb`mB_Di62?OKzUL)9glzWdajzsMNG3gqbs?gB!q!AQP{!J
zY5Gv492tH^FV{&^K?Q8>vHR!L-*5Z0YKLj-ctxj2gQs$6OiSm4qpl<c*jekWU6bab
z>oFAz5^Y~Ud%p777;78n>_;#5>M0d(>e`b@X3ym`H(|pF(yi_)sn?q<WgSw;#}`T8
zYmmTBefR)lGws^1ayu-%UtoB0LoJX0iQia9j`bJJB1_(GfDq^u+lzedFPPtA1&?43
zcqbi1Q<ek%!t``^?f{OfX8YXs3?I0R7wb|5CrAR>IPo{mfcpsxh5nhJ-Wq%uURD>?
za^S0_R^gi-Jq|4i&+Yy+)5J0NijGd+4CGSwZu{SzN}J2{SFF!pQ+}vFY|H+XCy&pp
zYrRJ8@>qR0iuA~89pw$FW>a2sqjH|^8GN!FNo{-*FR5JLecll!zerMj!^I`-SK4l8
zTC%%2k;D~DWmLXL)`A^n*`!$79==E6`JL1oSiRXMy=i;!Jm)#$9hPSp3VeB%XEwr&
zAevODZGF3;<dBx@aaf)j1`>9s!_KORAD&XY=7^k8M=zx2X>tkv3Pf#YNp|P1^rRC@
z*1h*#@fz)I%zwJRD7(|~mZ^jf{T$vIf2tnKT9eSgP~coxVPsR<g(OjCgsZ$XgZ@(d
zGL^7%3vO-Ro<0e)#fP$=Qa}j7%OxQp2y^oZ{CLh~1<#mud)+)wjNcQ#<XFmq{b8(1
zRcq7iXKyh6LBtt?+o%HZ$g_%H(WwWY5NcgkIyNnze=p$Sq4hDRX3Uv>c3G=Qw5>`r
zxyFF)$vKg@u`$y#*VQD(RV%{yn9XO0c^hk&T$*Wqt?kv6N5=Az4fumb?owzm?P<?>
zFUINWS)?E@%ZZ^I_3xH<zkT=UQqlg$Q#NPrb~*N_Wl>!f)k*NX$qN5Gf+RYnedmIE
z#y#?NcAAlyMQQdBG8dg%Z1z!c`Gv*n=inBMJFa<IjO)Q{Zc5EA360e8In`M8d|dK=
zwO=phKi52$KJU`n$!iNc#OD;W?6bC1<nQIUHIS#cXo$0Hk|K|#Cgy6gFBU4MYs@*N
zeC)ba?i8zs>%1K${I~k(H8EZ~=(5Y1R2pwZ5maZe9^8w1xWg~1JDRsK`^l-P8x-g0
zeppt$aEif!8JRk4I&%qr67(&=pEenreyJ(BYFF<9RTCVrRh<vqaOv8xJ79@ezqo%L
z@)V|rpEf33wsWdByIb=97pEzoi5YtZ58ZmwmHID`<F<EHE#=;f@KCRds&$>Gw>qo&
zvVn@t8vDRIQv;Xje^ecKQ4-`hjf3cB>yTyN{}P=vZv5?#v0Eiy2VI_>XJa;P;$ZQb
zb;k4p9CH(*;EHDk`;<32J~px^i66|<Ip*FF(o&$RuYT1+=9xyKvv}VD>gp=mZqobs
z9Mv}{?E&Y|ad$7BbBCNf;vC{!4jlP(>|ykT$HrnO2@Z4_mQnsrBwII_I*o<Kr8ks5
z{P6t!@5e91^PXjO?K|>d1*{xA-&b%Gm8&Q2VfJ)=U`m?0A_{eDuXJ$FM|c}7dS|Cp
zzb&TV^(9;nZsxLg54@M>cWGrKV$Nh4zRYcoc=nBA@@>1bGhOVCUO2L&juiiVo=1Sx
z3(PLt90Jp_2h*YKsLsN6%QjCoaGc3GdE7{|*kph51)I;(trutgZlJ#O_8!{m;U)z`
z%?kd`Z*hjnC~wr3%-4rP054i+y0FoODU3`x!fNG*c{A9+vNfCUr#9MK4OsrqxPC+X
z8mDF<s{X3Pl`I=t14*hA=SnwTG}A1uIx{mj{YByQb2R#zlP|ocU7cs>8K_sq$<NbJ
zV@g6#=qVg(|Iu!o1?kQ6sB5(|TILK|=`GR>7!$Z|*%+Fk$((#3r9iV$Palh>ql7k8
zgL4DRJ`|n(VddePX-7_Aa9y7lt=QDu;rc+3nOtX)>(YRKpPxNx?^f2Q`!cm|t^4mR
zP_iI>Qc)`^5iD9FHb6LKytk2YR%1#ze}AJ<c*a}()r)I3lv#*f5?8AGP!!iNNp5k>
z%&~`8Cf<CUu{^r{rV<mq-WaB;J3D(_%9*!y_;khb@{$dhgh-6e$BkRrIfAGpvYdm@
z?dIIw3G;RLm50nvD2V#hWYsg(W6An1#f7yEzp7wT)>7_uvYxUha~3s_&ej`ot~8Nx
zE)!iJIY+##e^t$zd}4q2$8|ehCM=MC8gyZylw_b!@pz2Dnsc0GltobFPsd)!P>S<=
zYW7>fZSSy#>awY8RrXuYbm{OtK|ke6OT~BPHVXD9S&)@Oug~zD`lPzrzScCzN9)an
z#fdAL7wBXIjnD7E>xQ1L&v%J>94|MyllM3wf&|C<O5-HEg&IvqO=(AsAErO9w9x%!
zJ53I8hN?Wv#{83AeC*wv1@oIv?oGSwr~ct`?*`f4O>a~d*nd9tq78$Q#t4*|P8W=4
zu@th|=v?Ne_;HW#I`|bfxMk0L@Mzh--F4f?splTQcb8|umn)+fZ;tIi=Z3sO#-*7*
zm{(e>bw%^MwY(HbL2u&W$L_`o=WScbT~REx7m<W+uO|F>qRR7&L=#q2e63xzg<9V9
z^whH(R?YiOTk+hr6TeRLOxnYENj~kb#);y2@i$lS9h@UC+-N-U{LtGfYsQ4sErxey
zn;9>TuFqPVHO=XKpctDLfc%QQZ*8(Xdz(oKHnvjO@*G)EZ}e<-vhsvYn6M9KTZ;TU
zE>Gzb*Y6V$4R}ZRQ<=y)RHD;yL2CJ~BzlIHs^gBjEY*DNe6_gCGhT_!*{gZ7bm7>e
z6nDXOqUU|$T8?Z3ZA#~2m42Dm^6EJWuPyZBlhX_8Kb3`>d<<AoMbo58`rxwhN%U@A
zmTL|}ZKh!ns+hskh`LjoKK^!qrLoR$73a^|%-Uko?mh3t<8ic|cf*(MJ$`T3r92ru
zg#b*;Wj_i3W$lhQmUs7lR*56$KF>Ia<8;quQu5uBvwOq6o>jE$TE1}2%f&^}dJ809
zZ7V3sk;b$k&OKv%2<+XQW@akL)7V?+ODdfbSv6pha$u4D%O|VOPoA-9vf0HL-MYHu
ztK-JfJA?f7IWaH&4FusPOMFf=<GLr9=Q>R4*q5AI<+!ehYu_1eW9Pkh%6^C9bDyzS
zukQDV6oe(GrBaj6u#3$iWlryy>9yR}T*XH6*|Zj`v3-9Q;w%0Wzx^y4dj>HswEk`^
z8ldC&e?Vq_&_wn>6H!iomKyk}-p(SfYWo2j^EE0qHdCgZs*um3ydX_<aa_#2@U)5i
zZm;={)fzKcey6Lp{Qe|VYTF~-?j*bQYa5Qc-*jb1G%M%1k0VZCX3|3W#}3(UdY*2f
z@9LcKaZ=&b%L}&22dv)bMAt}WnRhis;F6rW=QdeZhj&_p=Vu%0KYQCJcKGDJ%NuXn
z6<)fy{??UaQkYkY3Xe)9+fsv<$m+eM;>wK|#hA6cb6HwiZ5KPR=j51ae2mglM{CyB
zW~Xd?VE9tih@4FO_Q~AeNlNK5$3GdQ$rddddqh6K_jhTQCCA<J{Z=&C2Ex$;F<-ke
zuqFJRkt{QG;3a`4M?=dfzym(hC%A@h^K$1;K8WX&92s}<FA<W4Qkk)v#*N)gnW}4!
z_qF#tTzx}*YF>$|!y1jHwZ9dY`lvrP)d+gvQpKfSKYpj~eAbBx7uT-XVy4ZwMJdF;
z56yR!7yQPDV&M0m*{4%Ztq)gU*|Vf>%cmNz%$!5<D%sa&?e$R2a2q({ust4w?X2@a
zbvb?wmbe;or)Cm%yDKw0{k3vcS?%%tX;&A|owI4V&M8xz6mgf}0e@QBKrmGUEvjA+
z*36jPsk~bgw@+_JzMP!-b(PQso2ES4b2cQGXR8IxGI7r;mk?x#^U_n+7T?`{KmGQL
z^E35wr|uhHJ}dAOZcedBu78k^Zli5BA=t##M<HAFSdeDN<aP7s$Aqj?JR)8Y<B8p~
zk%)sMsR7B18INXD__qO5E~6d7zW+4vqP?kZZ_c{A3^TbyN3U;28A>lnH<dyU*?$^4
z4yU_nZFe&Q!NeK9c{6S*-%Jo|Y_^NLpB8q=asGn2ViT&~=tg$wRgQaiY5U<^!dogh
zaT5F~tjWMolUkGB;HS#7cKWRG(RGUBI^6QQ={lcNPNZz~J-^v(?7FezALpwESzC>H
za54Zh6(hOQbdv2yr|wXI0Oa|)oSFvR>QvM6*_BH&F|+PHR)01ab513Xz07F!r46i4
z_&M6q&4)|cqs)vaI;@<PP~^HqZ?3ty*tKhCq#qeiG#{&q8N}u=Qnk81$8hAD^HlNY
zZdE4o!(H8$>U*V!`_bmPtIb(-o>~0PYVpbyScvn+@;NR|mgD$3mBxFUI=z!Bi0=OG
zRVs=xq5JJucg0@Zci<RQSV(QHr(Q_{?@QvHt3{KVF2?!rLHS2VFU_0gkZ8C#&fylV
zTHK_+Yso*0+kM6E{%^pW8gqH8`HL(=VKNVKbA$Ey6(8#2E%x7}hqt~tb+aaF)fK^8
z*WPCKzKZ0MYKx}Umvve>hu0o0Ii;$z;zh*80jo`0-ktCJRp!^g#l#3oDGhkc$*I|t
z_t1$M$w{H7`hLg_pgs*8e==lOkEV0N++7N-=RY-E96wFtS~mKUCZkUo$6r=~PUGZN
z)c8a+DRmz79$)w&&2Cbj#9Y2nXhy&>IWY{hvG5OJ>4qSH_w)p&g@s>X={OpiRmCV+
zj*+pr5ypyl(^_aYZ7N$+wYTKQ`T{$+4&vZP%IJGK<e9R4{a=y$3{z1VeH;{J-s;<k
zZEQVD`PD?nnY-dNH~hYfVhT6)U74<9m2=k0*2PkRS7(0E+;!OeAf^|oE^j;*&pOe7
zKg*vLJuSl4EQDRWc=5i-%AA`}1(!TyoLP6cN)z?4;0E4>d~y<B?=5jUYrO?l#^>ov
zc44ez_WhcNv$r}^cNVCZOT6FazBL$o>HIG8Tj|f($$Ul5FU_x;A8_)s(l$|uHQx5M
zJ95fX$^uj!(%)eg1wzxQSsY1r@Rg>4ZUtN*D#dZKpfcscCSKV}pZ)mgHBE;W{q!^E
z=NzlPbY@k~8IN9R7NdrgXG_K+F67fxgU3lH2A9|&Ny!<e5#gH8_i8UN`fUJn!pqpU
z!{AKD+(yzk7nj>U=IfHmj&ch3r#?M7ziEPR)Z>7~CvM)O8PQxy23J^~#P^s(v$y%C
z8g)%bO$6JwEJMQJ{1Vcn+z0hZ65UA!mANfft_|+FyjTyXHQ`N=_%f~YEXE^jw(ca3
zQDfrMW{Qek=FNoW*FC9c_iee55{JE7l<+FT>Cy}Qi){mxd$%~U%q@%@@!W+z%4Q^M
zo6iz2-<eKX_j?a!fp}}wySg67r=}AA09t3QC)zFzWzNXY4lnw}C)6hOj1{&%uukI$
zmeMoHP%t5JZ%+Mqrc>D7l;FIb<H(d*H?B{!<84noy~$@@ptW1Vr{&QTO3r9ZCS~Dd
z`MQncW8{&&?cw1MqxZIk%(B(0eWb~~;!ycVcMoB<<)8VHq!LGkHWWO^6=tb)B6M%m
z!ttc31o1<2$)IWvtvdn@*~wixI`F|9`Dgr*p?!(oJWt|6_p{LM#OuWC@vjfoE9`!_
z^BRxmmi4(pzGPv<E=e0}O5_*Jz9>r@!5zeTzBxOPwPeCH=9zQ5@9varkr{MjZz}NJ
z>s_s;ZhFo0%?aD?<-FY-gT@?_xg3i-I~W9??(U$dB%MR5*rlk1^PT#2^@{7|O7A?C
zeI_2+@BSXwzy~u!9q&w4t(?}YpDw7ra3{0sV=FG*-6uv$rTdzuS2XtC9LjGQ4!nK-
z#r_hk1u>0SMb5+b2dlT;lljd*%Wrm}avWpogy0S-rD9~XoJM0G>3seLm3lMaKM#55
zg=Z)DsWg&meqp<@NN1Cv)0TJcy`F8ApJpZ}2lGt0_{A>UwHLIWv_4{KGfnPe=0NsM
zX{P%rD*X#){Do7yZwaWDOIy2^KDIml*7I}W2ko9imh-Q?!4OaJ{D5MvoI^Qv_#jn2
z^+cxKOWXLkvoltGws3zOb9GJ3<=5A)sK1=J`0yN#6K|(~k>f(<#(U43PSj~-`{r0t
zYPt?JY;U>XC0-;?BxkNVzUsQB7p4XKT&ExzS~=jy61P5Bn%aaN<D5$GT@Ir#9|sp*
ztBGHmwekp34X1DB8Sy>o{>%9Gtoiw;h;w7N^B0#^#w1rpzG<Z1Y?i*IG;n&b_pMTI
z@3|xsl_28%hLW&q-7MW2@1_g4npm19Q>zj`JWWquTWywfrN}05$C$ullT<8K2A>$p
z6COpb3A^KiY&vCL`s=RpU+3HITgonbxG;|za&}A{dkk%fiI_9j{4D-g25F~B=H794
zcY93@)z2)lSh85Rde5?>hpo??kT+E&C#QBBHQ=Tam)>aT$$GPrxW5TeHD2nG8>_Qn
z#@V#={Q+6t;=49qUTm>T&cBkQ#AFkK2;+_CGST*HpOQiqEdqme*w>}P{bJJwFT6AP
zEdeXZ*Wg$)QLck!mdT{@B^HHGR8QD#Z;EjBvT8UagBDCXVSC3)Le|zY)aGr^lKbx-
zdOs?ZQECXfwEpr6rfY97dkyDlH2ONVUPo|~YFG%b_)VYc4aUy;qA>Q3oO4sPCP_4I
zUrkYaRK0TAwQ1tq8hhLGR5gI)gU8yShRly-${iJXzm)3b)${9!DTjCHF<hV;GL^R*
zYWz~WrDEG~Pg`HKN90z2q}<Q;Si+9ZOio*^vSxqH1<LcfBXJ?#8m@gkFVkjlT(ecp
zgXN3j@9wCnmNX8ZxaM&y*&#W&q2vr+j;d}x&wnhlDk^%gVP1lh+f?0Yj;5C?#u1Oz
zZuI`wGNS#uFsx1t+AYwN@LWuSZ^7tv*Z;Xm@prMvDNgWyPrut)vJcL99DcZRSH`a%
zm1ou4#9O&Ko~ZmA0_sB%2*A9D#Eja4f!N>te{61}-w`MIUk<64m};KC0(GKR<0O<y
zEgId9N%nqB8K3=h-4r}m32eytkCZPMV_Ou3$S>DwoVwEUZ@%07&Gz5jhia^wIHl-i
z?Uag!)ax2z+)wwAm&@aU(XIp~=8vKO^)~1sedycozTN(t`?#AoMjf)1&&M1%6tGSO
zJa&!PH;oRybt#s=&ivI$R&!S~%8i!u4XOQtIqeB`9yfKM)Nd~|3Ix{QH(!X${i}Q3
z?`-t*Xg+RJtGMCNjI~U?a=bof*Je91^8$4F`J9XNbirH$+^8+g5yM2%9oH1M*v?}M
z4^Do#a*_S&CwXhz)<2us7az`MvoFp{erPda5hm{JEo;*@%v3Q4W&X@a8lOer@gST~
zG9R^NIyLiSRBVUzeZ)=a)vZ_OHrtQ!GZ5c~Z&YQr3dSVhdvcIQ2imAVS$|Wt>dvln
zW1cR0F=lP=Iw<_IlElI*rw8NPlw6>l;!L3KSEi<>CIz{5@7cb`GVScGZB92aO=;&c
z=2B`F?&u~yIWBf*C8R#K;YZ)w?y`S;@4G84;$_sj$!@6^nqMy`EE4ZFiJ8_wxBO<o
z{b&JFZz|I@ZH}f+h+CYqP6IPAcz^U$hJj3faqhqV|Cnsw*AwHVFZ+nU5sKXX15EPS
z`HvRNCuY_etUa}in~d>HvS}UlFu!>lQ;z<V|G&i?#MUa$=y1)rk2#lq7pvu0h?V^H
zm5#pmZ&uiko`frIXgR`za+q2_i6#4JTvPdL*>cb55_E;tzbJ)&N1EO9;THGU=p1et
zvv$%NA9b`}?=N)OU+KDU{vW$)bT%%6zqFjP1ZJ1vlA|s43iDL#?(qe0{ak@3`dwn~
z0)O>e3Z?(VaW|{Q4>42IC`LDUIaZRF|Jhft5gVk>yKI(8`OP-*{lsQaM=&!kq|RN|
zbxy8^#Dv(fDl;2M93`z$iwEZ5{gcb(`{l;-A=s~8r%|iDF6vI9hMZyZ$N!|(J-U!%
z>-cQXMLVe}6B_re0)mg<R$R7W#i6~IF%L5)?mI94)$qvv+ic3WwrOigP=9XN0oTLw
z=fpA!m}`aF3+&vtyM3oYB$fRdyT!L<rdA}cSX%w*Q;A@8`Krt7lzA6kT;q>V_*(|x
zYR227%$3)Ax*DJFn=)>DWr4=2!}^w}JVg#E)h<W*=BQH57tCmrXRYZtnSKcOI`G}v
zy2DS_SfW9uQ(+M}PKEYF*0BN(iLbL~*;~!+x2?_wNLVezsAI&Fk8bhVYZp{>q~`Sg
zjtQ3n@vIJ~e{$J)ZLyP43qAX(HTA-}M&9McW7^U7nKPM*<WYk6jTnJT5e>GHU@tHA
z=2>rHO?nYYrLVYyH0p1@cRap7%3nDy{<<=qSDsx^JN(I4OmiRbfwn#25bRGDdD@-c
z{jN~xrC>b!{>OsRI_jBOYADkutl53Y_|ajzKkf_j;&v*&acU(LvxR)YFx~+_?7_F&
znCyRTAEN)1st+DhZ5rRF!5xzMz5xs8BnG<6xTf@cC>iljz!mxC0{+z#{^-K#FS*TY
z-X=Vo^IL8GJ|C6#D{0riv9o{IvHNT3dNexjPMrB-viD>|g<p=`A{2WDWIdbay~XTU
zAI7%W(HK&Xi9GZ@(sq{;<C?{Hy4_*mKP%%|puRKXb=HDYcMcj(9v5Kz>2FEo4;;4#
zKKxltDlvLyE}6>9ceR=3Q#&xNa}r88=K9EB6K#NvN3}_!8!$xrfAqhDMBT(yHM3I3
z*B#z>UIS^^HtJOW!=oo&z)AU>X|%|4`KaBm-DV2?o*rQTomW5VV#!)T{RL+y)tsIu
zKA~_*!3wnhFTGh0_r~rq8+UjOE~z<CQFcrwMq1@1S87qZ#{0jgQ^RhUmL~3Z`Lv|d
zWxr>*Q;^EJe$Dfdp?{%NjZG%8B#Qk@Tt2#WCYLtj*F?)r*2T6h?Y+ZvNhvEht+YQ#
ze2n2l(*k{pU2t~fi)9P{W)$_EGP|!7bY6LqLzeqg&o=`t+r(f#t+*EJFBPdDUpH~J
zV&NghWv@Q2%`=bt4|uB%iY5y5PwOVLsR!!Q_x(cR``&h!?bOcQDKU2H#C;@+pdW^q
z{=2m(BGY;4_F8tM_28C$BuNeOUxzzCaz;ghqwn2z3-F{m&*(^`{$^SzGeZO4^8nKt
zm(a?rN>)?!QQEO-{9^NqNY4ZJQR@BQ<w<R?vPl)g8G(tARQLG2^+#&;+wZ>J{#z=X
ztCh{@#grqR)i?phCUbwAM~2}7@`O`z;jFnTCDSp+7j(9KW!~by{I^>0zZf3DBwK)Q
z5%Yv90GmghN3eq!d%!uY9~usm)ym6gEpy$_erHZBx(ayV_VmHX;vam6`PlX+#ZBkX
zYt%H2RHZABek}4iV9^l&P5TpYgNFw1Jh1lOC@lP^2)rx><E_T^fPHT6zzyINjwDz5
zP*4-dDR^?}I$vQH2_R0ZiHxeYsRt{kp-O_*k*(XW4kd)dR#5oFT$K9-W7t8o8|d~Y
z!QdAF21&os7=%^_%TRv-r3~P8jK@%FdZg3cL&MrKS|tM>!qd`m>O(y3<^D`=xy#pX
zbw-wxkq5wcKTf-tI6K{4WMh~-ZLkE#86>06@O^MJ(Ux*U1Rn7xqLZqdiL<_7bpNxX
z8y}23oab>!fz}4D1<>`SvlB)&P8&kGkD>tLeBXMkDm4gcMhV;&q;L8T!&Lt1M1-+K
zENTe0xAVsLia-5=NxW<`%MFm20=W`2ZJp_qtAFPGl}b5I=&HSfyhPvO@q!g%whr+2
zlk+Q(wi!=!2qGI_n-dA2m^cI;Esi9VibVrm<x+8|SsQe8Yyeksw#XnCcaX8k!EFq#
zRICpPUG>CId2lOn1*|h?|1faBBfZbUQFOG*5SZpXMEbXBA@u6LfwiCC3vVhEH5;UL
zK0}4T)t{LNkqOUL)(6|jy-JYT@&PWtsWK68jOo+=kcU>++oH{##{5Yxm3=MJh^aOS
z?h{@#4_2OG_OSS2SBZ>xPV|Oe5WTYa{xVoxApz2#Ehi!cj?Jhz+(1ZQ+Ne!T9y}V1
zXzL_fc?W7RRHf)p9+J7^qvd9Za9m|+a3pXjo{?WY)^Ue1(i8EYw0a+|9~Sd&{vwbh
z1AGtCl?+p(8%Z1%ID#jjF;e1;;e{<-L&sNDgKT6VRd5(k6?Twuab!W?EKH8@a=u>D
zP&b)pY2C)N1{P573!KF-{4vRnti{daM(!V;L4OF0Lk9%!+=k>}yWw`qU+ZqZ*4_{r
z@%^cIGc>uV9?5Tk0pew}sH61?a9+<oLZ|(kR=z*H1_!mlQi+2@6$_%hO>ERh=G)By
z<PV0v3xCqc(r`FNe+f6_L;QVaEZ-kg5Zp~%^aO{IId<{fSpMllNM{UU2t3t4Re_rK
zNcWIysowCoEwBT3QLl5uduTk4MG8#l=yNSJQ2WEy(F5-ORP+Br4`oaO84EJy$Ur58
z%Q<iKCgw&CxmVGBz;@6^!ku#kT@{9Qe8Fh{9W_r0)7?CCMztlcrBp&%=NF#N7U=2W
zg73Mq(M`|FY+h9{Mn%h|-E>kLqgN61o^KLyE3AA)kaJi58z<)<eT(rtIwbR{`{5Be
zn0G3pLEEp1Y>21@QT=bNoen^I!5TW67K@&qBkrAxu==q)a@F&l9T@=Z4tI(D`X1q3
z%4X`J#`-w!M_Mtg$PJ*jWX;j%l+Pkd`z9SVfp#W-NR*e!d>)%-4s*UBi~L<L)n(k_
zB{XCLs?-lTGdg-e-MnubSw6qheEeLJKJc<(OigSkzS0iKU=-LYIHZnz9$pP<kyt*l
z1XdT03yaYi&-#oP97fu=A00MU?|m##N5HSh=o^P?9#r7oM$mXpfVbN2H<dW-s13K&
zM(g|ySn^@s$yq+}DE>A)djPjrEfFgcYBF0pkdgu1CsM2ZHPN=9#bCwq`||}yk&g%O
z+R@-Rog^-(&Z5c@v9GPbw<z561HDulw7cpf*xG4PdhNRCX&t#?sW(KV;Xr&^e*vi6
zRZqO!2c<w(sIGP|X(0K!f+KkD#D`HZJJuFv$6|_wIzaf7kWL;Hdk@NDozA3`Pe_X-
z+>?W5^n~?8PkFd!qU{O?>=~qGHoo>CCO#2f|IX$Pe4qRdC5p&~;_w4|20BvL9HzHR
zLLW~3U^@~defyApCeo)M@>o-WKqm7`pSxgec?Rq}eTPXX|2|b%ne<%?!i8}|p34#4
zLFJF|bejj^WtqWo`HcQrL>s)kp__70Kv#hj^x#@twMCo%$xyZ6&=>!Z3eV&+A2Pb?
z7vVzGFCJEB<37alPwIZb)RP8bVeLc;;9U#gg+%Qh-bL~jO#fXOi>4X$DwWw9H6nxK
z$lNwqU1p{d$79dpTXa3XPCYC&^K}@#joJHkEx5xgERgQ(1m&Z1R@6FJNy-mdHTXS_
z0mqt#)LMQ3scgayj>ios4BE`Xm?004-9CsT3zKRAB~CTzITtLSm5km0Vg=LVR0toe
zd+;?MI(mAeGRnKbckN^3GI}lfkVn`x9~_F+`iNeV7#HLetIeVa#z*Q31`Ig#w%Dwm
zv#B^`e_0Np$r_Cs7vY_t##k)#>u>Zf`WCxHnIlJLj%|Ie&u_5pDd{>1LssrGLF695
zWQDw@0FyLkKHmrY3E(?Dq02kaMBB#=xs;=gQBFdpr@)OOIFS#gXVQ??Pq&ovdrf(O
z>}={xVn)=aabx?V<qZF5PDUhms&%-CI7#tf4a(sn0^v;^95GVEa;=V>h7QBE0%a~;
z?>OkCE3nYlpkC!ZmIAkuG9<y%fXo`i3!7P)Dh)3$Pf)EF58XfkF<!LY>-l}8zyJ&n
zywTqPX&VxY4kCEjxGM-9-=Pzp82b{AV<7NEZwD;$vK$-_&p~*eRg?~LuRK`j#4vtG
z(6BvqtAn>cDO=EvMD4MuWAIPT!`DX)R?QmT!8b1oM4Sgc9JmPa*DN=|1^H0~M=S+n
ztzUh02EP2Dw^gERE}<7oLeoeBiTJysTXEt6UuFG2Vw-e&5q3K#o`ru&z_~9$q&eB!
z%N{CNKK5|mA7^j(QYGf~#7k+3mYx?}`vYlHpdaroJWTpvuw{WvS9VnFGmWoHxq$R-
zCW8E*I#Gfe5f1ymAS+O3`MhqPO}Xns%Z;Qy-**$2acvis_U!rDwNLE=8fP7|da(%8
zyiBW#wy3jD9F@uf-h9ap7+-K?048|USV?vX9m?qz?gR1Wx%ggrORzmfRmZ;nu2j-s
zNaE~{5qc#m{2wE2*5^h(Zxe%tV}DRyY>zSq#|u5HZ9+}FB|dFrZfSr4QN-{i#*bLU
zVdt$zw0C(*UkQg<fpkWKpP4i4XG(P=%^!6K=!5jFO$?fA9}Egg{w_+IBTgDtbZ^uI
z4xpY+tN^nRhJqqSGj9UE^{1wsD~yG1Y$9ZsB#Bk~JQY}{^4{=+JoIZuJ}((o*v|oJ
z&MyM*4IbrPINT{%-+FMblIXSI5rK+0P5aP;g2Cvjd~d7Ps5=PGX9~j8Kei2HE73&Z
zprPn<l3+ajguQYIt(mLK{fO_IX*xAm=$MU|J+HzMb7(rX8+u+M_6Xe^J@CLf9Y9|c
zj7K7g|7P3pgOL@DfDRqR^-7hSToxKSk=P=ZN#P@}!HAQ<FlyO=%v^&(O34{Ia3j5N
zBQu4uRAD8pWtJy87(Jn5NR&tV8$%%6ikcon`rob>@sxwJ@x2PTPx_vfL!nW$bXpHT
zDA6QFI1c&JPOiwt@RL##HF#d%Si_OM{C{NYZv4P1#)(T0smRnB0NDCNWi$TT?y8aC
zNJKkTmf>4;+));~=1IKmYIqU}G+zFoDki-N8I-~bMS*VQpnMKuEls>|3F_H*4=bNh
zCa+4=7!L2^s~iM>sw#B4r#mQlQkCHhU|m4V1s#QT5b1LYT{VD%A#`JH@6tuKtj6gJ
z*TFF^jvtUneFDMcvykpS{GgbFALJZL%`hHAv90j88?^jiB`L^3K^jN(S{<hwN^2o=
z%F!1)Bha%4Q72dK9+o#Bl)EbOP6o7hnJl5G8>7Vg%$tT3-Y4R@9W<`TGs=sDf<eJW
zd`krtCx9`t%WAD8>5ib0NNVH(8hdAxPU<KdHvEz#+&amfZb)Z6z7y<lB9Z_o1;q!t
zL1Oj{#i+I}jk?dh`Yk){9gf}nIUk(Si-l*DxK0=mKmYn1abJGW@kG={{WmQZi-bxm
zKyz`NAGky{Sb;L(=`~J15cZ^CBm{{H^jNoGG!zEM{m8V~whH_Mpm{jH<#{Kby#RuO
zHr_#ufb;{j?I!~F6M_4Q!2SOMfy@2gz#tnO7eFr&q}h@uc96iG9zPfrJ%;9o-)E7q
zw1&NfL-bN#RrO*~<*^&rFHd||aY!%KLnWVR+u-1Lo#lyc=xNXr5Pc%PU@WQw@JIRx
zN*u$NRnAVV$=9D3uhE^p;<2FPMQUd0q1Fpk<&NbCA`gtebhl+gQPJTs0qZnx2H1Ya
zLfZpF1V++%LVIR8IMCfirILLHOE!)zkrA}LyPqG|bUVH}0?;v9Cj?4*+noQ1_=lq|
z*GVb;CH*P1{vKSvNW?#c2fBuiw47enj;}JPD4*MlJaoT*?+<TS7@&TnpfQ%9>17~L
z5(&nJJq0fm^z64VAbsNx07Hz;h*N8Uu1s_1k@&gbhTerpsLnu;kM0@qYjs4?z{H&*
zKp`f#65lct9tfa~mD02aNWq#`2-8157y%|k;XpePoNy$p{L99|Kj<~|)0F*8CjD<r
zCTYhnl8({r8xzZ$Pv^4`UQ1<^7P`uY(G$WqgP#A}(9$&JTz6GxU-Ph@gkO(#yKYWR
zJ-6C!?-{l0@)P&#1zBODu{F3Sbw~*W0o09)x?97%jem{yo_3NDh-1>fV3H^TEd07k
z;G6k}_!p9K`Wr19QqQB)iN37UZ`-R6+;=;eYCAEBYI<}7=F#-MnYGyTe8;O|13llT
zvGg0#2;HH)dF@w{n)BOdf2bjK@mRIE!3IC=1V;C9-aH5ULcnQcB$SKxyo2@+%Sr@~
zqOqR0W3RhyI5x0UPjI`sL-FSFrPpdN%f32%Tty>U*ig+3qm&97W1113+dc~bnG${g
z^ik-%b9E%}A4Wb~<8Lub_^0!EVWvu6j6oTDKXZWaW+eBId>%$*4Myd_I^n`b@?*n0
zE#<EO`}ke<%i#MwnbkO@p+u($=$S1tKV`}=o*#c3DX?J(n)zY-Z4x2On{I~;kb0Mh
zdFKEsI|<kS2DzSiEKL`5bHQ4Ku8$sgUOfy`j81_ih-Sxz_vzD>fj@<m5TIPE(zIKN
zfR|1y|EhJR+o0FV5|$)C=o7vH&mB*(5Z#KN@HGP@8uWQ6)wChKW74>V!e{~0Gz1n3
zf7ifqPGfoJ`0<lcgo9zGi=$15c24vJf-+KwXT=Hvl63_WEBVFrzTBpsK`j(6>Ct2`
z8Mjqw*Q+r)T_vI9UMEBh-`Si)<2qKdqQ1AiCVG?uf+&0omA@R|qhJe|cJf<ip6h#>
zljWc;l-YOfehafK?-^pybNaf2<LeN*EqWlXvfBb(WkBe!{DQgu;13YRrv5paLPs;3
z(d2U?QzpjE8hUL*kzsN!eJETsoqF^SOmpz7eI3cB3i<Iva}a9bBo!_~WCrF(O8*I4
z7(xl=_X7z!?=i{T8ZzV_c7AlnH}@g{OBV;|427vSBIfi_rzp?~Y>lC0Pz7fzik}Te
z9jUOL?_e1OhjHh&3d}HjMA1xwsG%KR7&QRjT$~lmgGXkVlaNju9J3wC)D_M1^oJRt
zE7`RELdYH1YTpzEI=@<Au`i|Hx9$H=TUVK=t*gs2bUl$Kb(``vPZVz72z~#yeq#O<
ziQ1sTF1<I0Y0SsMoKM+mCLNVcpmEZe?lr;aMsIjakM5GtGdj%30Q_BD&US_r(stX&
z6^cl!7<>o+HBg+l>Y3n<?itW9lWOG1{l!!qB^7CiruBM6kl7PJX!;y5L^DD(bWp(#
zF7Pc8ghtS~Ce<{2XP*Kpl{(=H(W@$a(PiW1$GNrDC%KnQagTis_(Ah&^$9r0?-iK+
zf#7U$iChrtj!Y>3C0YK-rVsdn$~Pz0(=s!6B;2k)i77U1)v73!R5o)=FQ?2~bt8@~
zS-e~_wzN|9>#GwXfczwb{J%#A!JjiWYt+TI?=<JnGMQ0+6q_WKLGD(2mO!k#>wVmR
z?}F~<YLCljt}ydpe7jCLTBx)WC}&C#{!=HR_!VtVn>hDZW=|NZ*5O+hb>>pDpXBC@
zNYfdqDHYL?E^3MLN`-qbys-WJn)MpBAmm_qQvfQ{2~RJ1uZ(W8e1T)j<GAxXPjNm7
z8`U&P3eFo+AX96qVMp-bn*e+w#{u=bS2)-9R*h};wE<jq&)~wI5CVH)JB_QJj3F+E
zxIwb774|AfZe;KzgtXImeHNu#oVVM7FV6ChXpq55R=ZA|P1=9>sl{Q|!bv)FOdp(I
zcx&0m3>{R%?jcfU2x7xrN<a-vaj|{21-Pe#?qF|LPm>M@gfy!v(s<9yrk;|J#t*+{
zCrMNn#6;D35(uSYKe;c$>AkKHrv3>f|8L;F{N2c#TEu}&+iY*90!lRoD}J||phAC7
zR>rp!qhk<y{yC&CGfuqOknR4y9JT1lCiA8or3vO<$T*OBgMAbNc^<PB%~J<YX8qo1
zY}=JBV9CPh0x!tRhdEZzT4s8Ie?u6IjQIl>{)q|yY1v1a{Na^_-9DO9bHDvZnD7r;
z!~BO$JvIV=2sfx-M?hbO<GE8PFAoad3!Wh@bT5P_y&bUJJv1qBV86A{V4x3yW?6}}
zH0$EKRAOPrGz=$%N*4rQkH{rYy1r>Vb9T~7^_8FJ$2W!W?AX&FMMG&K2+sP_(etU%
zSV%l3_3IAuZlI_Er%VugA1SG246tZ<aWJfnMtGw5<PA^9M|$1w1s{?(Je_(4=B&$H
z_(%<;+@$U88PvyPAhRw^wZCkl2L17nQ@Us^GI0Xn%#?z6JhBltl1oo)r}fGUf%yBv
z)Ysk7=}XKZ&IoVQ!8T-=(3v!x5=^~^<ez~g7qCgIX1TSKS&ebv^6QK8s1zU_`YlK2
zxhsJ)7%J#Ty(&1VJ=Fpn*GdvMltZ<PG~j}h$YuKit#K%O36$eMJN#~q=)oo|6>@i+
z#gkZ0+<5na8-vnKt|=p*|A4UoEfV*Hkbra@Vl~KBu7|@(q9$PL60rgiUW}2h6!MxO
zcnWPG<DVoNyd?AvK-=Um81})gIKZuP2`quKz!S%nPh>ySRYLOn2>nG!+nRvoAwoyH
zlc<GKn+0Y3&dZ6ZFOkkq_&Qv1!15Z=BxSre)X@qcymP<}M&2LJ8sWeMGPmLfC4@`?
z@15u|>p^ouJo($NOmXJ^1+?BH{_ut$odp-u-5!^(J2)Sii0|56iS#Az?7{CN3@*oG
zkaw$*w$wyya|Az}b{5B82w5%L`IzHwe)&uuy1}R^0*tXc8tg`pv#mw2w3ck@T;yFg
zFtyr_RfsV^Kp299tIRT`wWPMw*t#w>-WKXq3Zy0*q6xireGHy*5<_sGAZI06$6#(M
z8C*6O-zDS6Z>GHK;KLsV9zzPV)<MR^uBTPqKi}%7ISu}Q*PO~-v45Ho$s<~=?R<RP
zTJtDd>5?ES4k>v^8^}h2{H^?J2`be2K03|a=&45Rl!IwYqtwH$BxK%gd6c8l{CeV9
z)4~<5ExH^Re6=JRvLF+<87jhZEQ~<Fin?2i!x|s>{E|gsPrkGGm1)Vv`egnqdRu<i
zEtY2!iR48%w-X<hVSB$bWg!Q*BWhY^TDVV@wX@<fTg`FOMPhwG+gRR)p*%HNfD}gY
z$3i-l1ddz#3EMm`wo`7&pdKxN+|L5_se2H^DTdauo)T0%ak=zjq;YyV_QR2D%}U*j
zr?&jxm9r^4tP7}<70%+^*%kA2Z=#KFtJm0EIy+rztW&!+{%JCHus;IjxHBo)lncNF
zKn>auRsQi9XVkdr;lp`eZ`O>Tc80s}!kn0AUaQSy10G;7a)B8!sH!FVB;W|46r>>!
zLA3+$x6aL@O#emm(b}_y;af2LwGS#;*e)zfEuS_qLyI+zpzd>qM>wI9R3?`B?0C^C
zOQXXZTbhrR{qFUGk{X8@;X3_9-Ty_@{q27p7x*_j3snN3^SN>&U~Z-cSml`colTKH
za1}zF0r?Wl$E#Ck@v(x7CKDWp%PL~Z?^L$Z-?n6CZ0*_Z-ZGb@Z#QM0uNOrfI+F1O
zLVsj`D(H4E_@o-kw}U`z8Ale_mnNEkXMLochu*7%23j|Tx0;jR)C%2HD4am4zPi1(
zPm-RW-(G#o=fL&NFZQ2bSfF)rr6SQu0N3YCmW0+T<e(+ki7Ut5i0K$e2`{^$W;=I+
z-r^bI&abCpmTO)i%=ItE^%Bsj(2Ek;2&X!IM-W46q45Fuczg<6=(oYqSa4CkyG9C)
zwiidjDVQwt*_*$WZ(b2oK`@WlWVa`rwer!X{kr72w#(mTdWP_qQmROt*{S>xf&$V>
z1-~e{O}uZ*5EbU9hjTmNr-$=z^KeeotLrsOp3~oVZdQaPGv%(w6pu@3%iRhFT5sj=
zaisvS_9_uV%UgQcjDe13a6YXhI|L(DG=z&ZW-kc>&DlODP=P(d>JX@Ma4bt}&Ku^|
z)tr+F+wELZy;BV~CoMf0t~Bu#D^UEY#;TUjRRJp|53T6oqMOPR+g}>VgHogc)d1~I
zTJPj%T#{g{jhU;+=ixl-A-w3E_$s)xf=rx%2n?($_2He|I496mr2;iztu5)1hh1ze
z0VvjXhsHVcv1PPAg-t9RKWHCUw>Nr%_7FR6xZuV**zOWiKr;|{<6eoRVnyg>uM$Rh
z*Un>l&WX~?jyh^VJUzs`9tkm?gYljQk8~gg^$NBNC-;s83cN4ARfHqn=BO;fFS9P@
zjBu~5#2j6K7cEnZyg>To9V0}yOs7-`Q#+!`1z>oU!bM4V^@AQ+sVJ5+gp0#(F;4EE
z1fVMwv@YCku>J8((mdo3rjXWs<ef=2v|_f+Q50iSy3C=v2%po#o0v=ysJIc7Y^!OH
zas2ZTwb8m(RnhwPCljqrCB92GQ1!m|d8{P!yuNj2TytD&Xa^{R%XEVYY>jfjh<!^X
zkZN7PEpprek;4n-UjB+(gbHyBmKxv|kcD5rkD?Ndt)g&d)A;c46(H-e9;Qh2b7v8v
zXv~espaMD@q32x#9opHqKHz7akg1wT{{LUE6Z$WfgHl{B=6fI$^kfUV0}T=Fn4N#7
z0-~`cG({w9RfcjSS>Yh6cBQfVO}2|c#N;z8?@P$-x)d`{@~TJF^E_89s<E+w#GO4r
z;OOD_2Wfg#9i(tGizI+H--0>54bI~I+097)3Z3TGF1IF2B^TAHkrQl8ragV)ypP;1
z@qYCY_qyc6i>6J<KgcIho)JN+XG=7@^QO+H^=(Whu7E|K4ss*`4#@coh-eRxL6AC|
zw0YjBv<y)0RE|Rc`34V?R4SQc%}(ut5Pl{hhsK-O50d0gP^4Eg9!v7W5Qqo5ei2Gi
zCWg3G2pPX;Oa|F4HW5eYCFk&c&*Pg_c}7hP{`LG9?%Y9^78;>X+-s^)$y<gLB0^5g
zZh-;F#H1^1aQiPqhz?RB3z}xVMg;rt*KYvt%jUYHS7&K;gBvvq8c`Vt&Z|bCS65(s
zlKqpAEPYzb77m2*4t-<+g>bo_TL#@xs{q-tXmS^LzeU)Y82^wV9b#Q)fUa6f>#xN3
zP2cuFw0JgJmZ1bFNd^8Bvh#!M0~Uer_Y2zEpJ+3j%v&nT22T(vmKsh7jydryreZta
z6~dWC2$fcxMxHF-5n|xpFwH~UZyir?!E6yI{3`)&P7D$cdX?fN!Vx|qXzagNE`(b!
z$8OI1g0Xf(KO@uaHy*1O*%-DRB1KVNqnY9(RWbWT=-}3Q9Fc-4jMAu*DE)f62v@vR
za*+@ma&SKl;8HD;hL1X8UB$5!_~r9r21_N4;2v;vTt7Vf>kqC03Wwqo@ttt9de*~F
zNBlGBX(zBGBWaG#e$Q{eY7=kk`G;8yht@7os9IVLSl)P&U~+#J{|GD;I!he%t1Ud;
z57F59lxHXoDGVlpZ~HhHJl@jo_{LW}QsV&sH}70sz?2yRx#|FusX~0}6P`0inIA4N
zKy$lcEgt-u58*4ypcekHs`jwqpJbiJR~TL>TN+Lj(X7i;=qCu|q&df^O@csL%U#ZE
zG)DlZ{w0vA4&*Xl{GO6QeRmKf`X_h-cNfzeGX4AfbIH92>y_dC)`4AGzdss#cPNRj
zGelRg?>}k+0sO_^ppqqSOd`nWbXO9fd8KCiY^b7JBiycLQ72-wgE9Qh$1%Ql9`60v
z&A_~bNAd>)t$@}-xPhjVKETRU=7)Yp=pOvIk;*-himitaMyTuo3pgX?T68lJD*G_z
zrY7P*c#U!`hIi2j5-HcR7tTZ<CbDCcKHdb}_F`r}UsA~IZ}%Ftttb}ub)AZo?<USj
z_rs#5wuBv!%{HBCHk9cXLT%>>Oe^x>@)Z_n78}RarcR@z{!#pL#5Y24stey|O<ZY;
zJBa2Wy?Xpju$p9c0{RzT$+Gr~=Lx;D5|$B~BNXyAcDA`dtN$jK5#Mac(`#zwUuV3q
z<Z@WLXapqJoE+<T`T)}rfgbRPO@<(|Dq~G%I{P5#<3NrVD8+6()W;-XYz5%@0p~#&
zrv-0?IPJQQkn`Y71I~jyt!sCMkn^yf6^RkRp9aE36p&eZD0K-16ya^_!5UD6bWnu&
zCMN;Wg}`&9t^SE|{2Q2s|HYEyq(1jSX*(oy$F<lPq+m;ppaX-te&Y?eQaK4=8O`(X
zY?TUp->uG*(b(^8KH)IO65Sw^$6znBL)vZ}<oE_ie{b*UVD5YPijYNK{29%PM1%zU
zv+>>T?@pm>Xy*|2^iBeIaXob|C7HGdJE)3+J$@IQfV92##c=LHbcB6Moe!h@ekYCt
zMMs{5w0qj->9f6aX@h#C07n1kCQHGBgPgp874d&*-gUbgzuNo85<zeuLojE(WwjA%
z@dYFFt@U-m4Z^Sv3Y>-O9x1rXGC<SY@vy>Yb@s=gQt&)xMC35IJ$IlQ*cT&E9MWo{
z)nNBUodHudst$yTaY!wi{EBYfMB{LS1tp#K3B$&9RBFXeq;tlwG9JsOu|6yFI{BUV
zS?-KvE*|m%-xW+>70WOCGmT<1Tnz>^pJ7K}gyf?rtc%_lt6PuTiRm`pfj)pHI_Ror
zk=+?1pNC03qJ)Uf(K(mle4pHF(acJGm}u(^j7}iSmXDZ0A}Jp|?V2f?b%~bSBaD30
z7*g<7Ar)~HF>HxX01^|AG+_Q28up3Wd+j=rO0nT*Bl!`&RV+bQ5~Pv3Dq7Sv|IPG{
z)^ENkl`LaC(dGY64#0QK!1vkFCI6%RhFXMa1xXyPyCAgM9j%}Q>wNDE0LM|e3mn5)
zx_Jb)79?|~DlxnLa(NO+p#>{KaE?nsgC&SY(Bk_$3s1@N6>KQ=H}NTHz;?MX<87GM
zcY}(mGue+PyyPlhKbG%pH0G8jNit=fv6W<dkkX(S6#J@kx?%DoSq`kw1P1?-138;i
zGtg`vYBHTec$s!HB6|MKrw?}Q*s$m6^?ma8n=U=y;xujPl!lAPyG?QTuiBh>@E|hm
zWgJFwnfP6&Qn8<u$V}X4kq$inX-cGyUd~NTCMl3;U=T`le`tQY-D>^Cq~4gc;47a<
z7ZnQG35J6Nb`#k7-IcVixqx5XDwXv6f{|U$(*wU^&C@TK7bi&Q3=+qs<UJV;(I=zg
zwFpaz&xQoFJA{sF@kuvtn@?;1>a0wsqzKc3WT@1IRH<w^cmvkNb^mMKmE}Bb{u;W8
zJX%DX;y=UZfSODH3hjY4^7CsCthFnS+U3^q@O0z8Jtn(Po$B6>Fj<TXj6NwCQHyy4
z7DrKx3pEVBt=t*{cYwsDH|z3^y_-`7V~9r3##|5SHS^DG=jPJSX6QH4CK9w~702G)
zQMdkaV4McbKeSpmZ69;7PqfsqOFmTAp15GR)*aIf)?_<l>xS4u^fC3`^<i&SZMfY^
z)kvwu(cuI=9m1q<3fUbDAdLF1<o&4F&LMy?6<Dm}+G(HgpOv7SO9cPaQN{pnFl3~A
zf=D&{^MfLF?sLSIw}E#YRwAsI-=>Y;hZsUl*!Q3~ZdezO6DemGEps2<LGq!?C?d$V
zzMetJYKUhvg|UeJS9^l}Z&tVebb~PetQ*wV!`5K+1p6z(yw|5NAAmhz;Ct+bSIk4!
z&arw6FRwYbR7yoT_^LwqR3`vPqRR-Akn9{2;5U>$Z?fd|L!4f?<oj>L*ma8G18M#V
zciIivS1+5p#=p0#&wJ8bwC&l(4mFpV)8v1jq*wd4RBSYj<!2TCf3FH33+j+Fm&^~O
z0`WLK0Ws3%$51LuzWpj18$<g?lofCCKk~!^j8mx19I#xo)mat*ychubk@D4n3TE`~
zW+*~%T8{J?0Wsi+Fe><-z5?WzZ=5ecVszi>pz8_eX<e4UsadAeSp5UcMHsa-k@(jo
zv)p;ot_at9g^>2e-wZZ&8fQMl&KunUUWpZkE#>HVSjA3EEC7bK=(T*u-Hv3YIck9q
zK)QClrS&C3xd15{itV8j2-Vdv$!qvyfb`FfI)Lr|w>#<(bkJHt@2`fbf}nZ$K11$p
zs;%G|t>v8;yz1>p+fgx&tq^3cfu&<PxQquK;KsEBtwfuu`M=c@;<+~(@UL6u<GF%T
zd|!$-=Q6d1vk}B<E2{*h9mNXfhfp;EQ@p1uT*@n$wt@h_E+-}ni)ktwWh_!)g4>Hn
z%je#5dG&5<*2D!Xn-efDn{BAsC%#}zdJ%>se-2NnhtDhgf+3Qb(b7r6Y+JRD#U}^9
ztz5dYL+okl2YMww0ps!ZYtiV|mOosa;gC`G4MDI5AS(OG5Tn|Tl)DVaot=CwG^MqT
zR`vWw35QP>)^R4@021CI|4@PJb;vG|;M~H-Z<qujFem&GacF=KI9x|M!Y{Pl+Gxq)
z-$0+k`RtRuOhA#53Td6>enqj+k!n#TM8wk>L75Nrpr+xt9t(aa^vVMGOlmjU2OZ5u
zcBoo1%A0xM>&-vajdWgyP8W$^FHrKo*#_aKYx56X8=h06GXHvTzB`vE6c9sBB=tQ9
zK#JMBSv4rN4+P{wfXQqjj?X19-T7B2V*!y8-vC&tS3>p6@4mtOY?%6gvtjBW4bqMf
z>0Y#bNgnW`wC*Qd_2Pn*fAif@u|Ds=Nb)3pX2UADa)l`=IBkxG`Zl1;kwRC_Nnu$_
zJS1I?luCv}o@z9~wvzebTN<t@PEN=x7{Gm>@!>%O5(z+lXrPZDpMZa=uQ5MhzQ=wD
z9{yQ`-Yxw7kS+3ubKlA$$d!uyz$c*qlFA6d_wK_<K=7XcT*_^IVk{+JBnvk<4+Zq+
z@hPAJynWVLr1P!oVu~wc1G8ni5qaQ^lg)+UkC2fY>i74v>h-Hl`j10AACQ_*bfa*>
z>!Z8DXfR=$?W-@CRAv4RIe4SHEwBQ)0KBv?!OL?;rKN-oY!Jw%*Fi=Hcq!M|48EcQ
zx(SM~(5y5B4WXQyBi>DTM?%*i=NRnit^luBQ?)6{$R7Nl3aW*?3lgI12RRSG@BJo`
zM06q?Alicv7;)e<1dUm+cG%I_^PS(aQs`&tss3*Wy@v^Mw5}DgK*Z{gmWEOX0zZOr
zmH`xXgu8W?&zlnl-O8}s2ZRZBLo#}`&{5A6=f<B6XPC=rg)^i{qT#uONwzhJ6s_fz
z|8il169n$=W80jo7~6Jo_i<KgQ%N>inWMLgX%Iq5+Gs+MYWJ3vNV+lcRrlhjmpWF(
z@pV2grZxyNke1t=nP^U(7G%!stt5eSlx^c*M((o$U+)V<p;eRw{2qjao08$jYRb>P
zsmF3dAK$asdt#<ttwq}6LXB;gI<V{cg_Kl$0~z0YWT(5}R6karWQt2t{ORL}{|z7K
zzttaOkk&lVt=0gg9UqJhl+PODy&o1fjZf6HNxi;Tihay<r7YbjZ&ACcM)MF)hEjs#
z_zb)R&t6_1&zA|wG-#lAr#E0j$O=pF7#9<HT6tpsan0?YW<Lvit2fnh?ksby>5Dj}
zlFdl2`=GYTA`s2uTd**(_Ojj#C0a`yCm+o=3<l1~VVGQx1?;WmMTQ%gil9{V9E7d$
zpKu^Sg(rP#(0mzpfr1+WYXu7Y+gc(hC4$^WU?A+8=ZL)112z}*Mr0}ZCGemnG##?L
zIV)U+nP7UZ{&%DqQ=B5?eZ;=cg;9$+so>9eFx5~haUERs8^V-Q$>hJJ?A{gpjt34e
zNZ}*&2K<@SI-a?ZR1*gGHSVtP?oBo36Mp5-0Qp@g%qkFkL4NZI3br-)@wApA6C;FP
z|5bc}<RfI^Ar`1C5FET3$UKY-nY$@DeMh}aIfc>K8dy2M_P4(I`fAcXyT`QPJRBt1
zHgE=$w+Lf)z#P#u+9pHC+sl^<zrS%0($PPg5C5B+4?_}dCO5is3IG-n{DVOwnKs(Q
zFBo^)pdnTPp;M#2VD4P9lKez~on!f*3A}D3?=gR~$MFWT1c#ibup{><SL50Vzo}am
z&gr#V{q{jhy{XaRwUm$JsFi{X_~w#kbHQ1|{NA0IehTNfZ78j6)mwZ|R|S&&D#HP~
zB~QKtGXW1!-bl|8?{i|;us)RJ^qgjh6Hgp>noM5$)IFi7ig|hO`QRk`xraP@TO;`&
zK}|cC3@#<VFF`ZBN(Abz1o{8X$v(eymN~lNR09AN`ZR&|P3jCvs{b6&3m1u#K_2gc
zK;PXe<j!>v#P_=R>+|vF#_iz$a#yV1z{572o<JM}g4QW4W!;E_-cmE7vF%!$%e<s|
zVk0#r)X(7-lRCH28gZ5Q)<_t!<cjL^Kcf}U6~u!uk=CW_2|lK6K&fQ^Op~xUyH(|{
z0%mwkKZ&kM^!Z~b0OfTFXSV&8jeA*}mR(I(Dm#oR<k7EjZ#44MUi<x0bW_>vUG}{3
zieeFoAfiJJqRU(j{rak}_L_;OEt)cYo#aWvd?F%<%TVSSbFQpoCej-JKla`{9Lo0n
zA0LsO3E9^WvR1aNLyIM$%^oV1BujQ;#*(FMArxhlELoE^+t}C2zGoQ}VwkZ`X6bj;
z_x(JNp6d9%KgaR?@ALaZ)S-Ls`@XL0JYVPQe4Xc+I$^nVu79>}UFeyJ2~?!@V)(pD
z6tXo?Gvq}bh`PdKzc(0p&UOLhuqyKg$#n1$kk2G5`Rn$>`4-`9sD4>b(Pu-xg)F`P
z5h)iQ)~s>v3DkV~J#@PR4D$a!mEeCol|YGB^zD<&_pX1OW=;fg2krM{kNPl~V2jJ~
zum!RaP>o8G`kIFai0`0U3rG=ieV2!T)&iDbT-kT>Bn^10fFk^GhhT+MVoU6DN3O2h
zbAwx6RF^|qff9}=R8*-h1AG(3ER4Wc0Md)6fO-k`zmtOd1+<zqZEn+Gfd1|vXM^CD
z*dlD_DCn>a`iA@rNNrh$KTctd&ru;t8ElO*$t@YT*d%E90v{lbpMt=<&e`rkG_`$i
z&f0SyEZ_8{i~ss7#rDMCEoo+8bY$@b7&n=B%>QJMLPzOiXXotlHT#Ep(~Yi1W-*TA
zHUIyVh@VT^f2en1|1a9rq&TTI84f5uyRoJLGwv6<n|l8fA+Ck%<8s^Uvvi7|tGqFS
z1l5trHjgQmjbyqsswmZ94UGlLvG17ffX+9FCn6DUifAxE4HE$aQe<uLLokXNggMBK
z4xg=7mcsF#*3#lS%GQr93PMb0q+}%&TKP*2De$&$GO4b$olRnhc&l&c9hfIq42C`5
zw>?h=t;9V9faxuYEFy`f)L9n2LR>8Z7yxuqXc;EQ8)^w0!C8?{$3`KR<#P~=Cx$7l
zw|_y%kbovM?CW7UUc?C63FyF`^3Y!h;f;v3%Qi}Yl4u+U&FDdskO7F<I0}2#_Uo~(
zZUVIf0Ozvbz20w{<T%2?P!vTXc)nsNur$3%n>fQ?AaV~)ivqx$0nczQ<TJC=$|Qe$
z0`P<>bw`SRBOz0uz}S|A6O)0Lfo%zb%+m8iu$!x2fxiJV$jbw$jpvI9d+7JSgH}*5
zfIJtEpNIEOftoiF#7M&{@aR1>k-y;H;pV}u?nb$<^3a&6&Fyr=;%$KDzc)V?%&;xj
zZ4tbmMvp@Tl_II})n+}nuVLYd^4eDrcO2PHcdI5JV#`fBpe4>-Px-p&Aqt(YsgCH?
z10evj91M>M*h}z6w1eXeBe0gnlDL82K1;JOwXBS@pDBt&-DFZbg|m$+D{wTr%RBGr
zIHB|TGiwKn<QX@$LqZ$JW>A+D23;@=&d?p;^m{_#GKdC^%y?+5H(Rq;jh{ezr{8p}
zbWB+%qRAURD7c{-OXOIbB1UTo)}~2PO`HfqChQDLnuMQw@V#$Y*a|5Zw%{p<m+41+
z<v|&esfk|Jwp-pkn!=#FTWCSSWL}PqHvh!CbK`kmWfzJ=gIJF4VS839LAj7C)d40g
zK8M*Ob`Gf;leVVNrulF+BE`U})5bV7BIVMyvYY5Rox<@neOi-qZRd^lLi{}0Z>)`s
zUFx`Pbth_8L*68Z@~2TEm1A|$Xn+eGRi`3~GWtEnkD1I@KAJg}0Q0TQ15W3sPty`8
zKIID|Du^$y0`^249HhE&qSTNO(r`fGJ<O`t#cn`>WJLA0B2DhC=y}GxLo2j%KN$S~
zO-j6J{}VG%I=r%$3mbmes-kD3S9SdBu;{&YzUyWpZy{up_M3p~X<J^*(icHBTT2>F
zGmDC3qqPK*0WbBJX1}NzHoi(W3Td9&T+K!-ikj@8?hsCrKlFdpKa7ui8G9ts?yk4W
zMo}m9r(e5Bfy)F!Xg)+6vNe{l&og$e?1wUX*FTg};7q<BDQCd$tU19-kR+ca<c{i4
zoyh;zm|)2kf|{%P(a0`A7jSm%c^!^JbrgsW&mT=<#bQD50HQa20zlFoE&j(rKvCOG
z4z;1QA^q}Cpjia@J<sPdG-`BwD|m>gmUGF$FS-B~KE|7BVM8}7_8G0Ozc1{(wn#|N
zOr0;V27!1SI7=eUDEu*yoSQ9M`F=M5sozBIY+|>`iN>~EXK6rS(4^(>1H3`pj4hGR
z27w#f{stMh+1b8Ojn{p3%Z$}}Vgk>mm*SA=E*jr)Ao1D6_`xUlMdRI!#lXUrqHZIr
zu`c9hCV*>houz3h+X`Q%q)pg*hFwnxVJ}Z_nj&$4QB=5qfIOg4Cjo{cr3YjQyU&8x
z;hhderEblV9uG#SX5~FZ%r~G~6F{-l)>Y7YFbn*CAQM>;Zzs1iyps|G`nj!!gDN_I
zcAPzW4WCzkq5Dpj%=&mzWke(S_Du$+>~b!5tHCwFWj26g`=+)8tAR&4Vm1ug*cqLg
z{PIkBA50akZj5uEMg?z<qZobi+AO`E<c=h=jh|`O=k{byj}pkOCo52V?6ollLOnWy
z8n}RdA7T|~@939_g0rh($bFMLPzs;IE*qp#grT}ubBQoppwXQMAnQ!``QKN~spcpU
zIPCO1SV8Tft}iILuuVgj54WvG!uDF#m!p^z8(yBu9k;1%S1KPCuse>UQb{{RTW{CT
z-K~g?B`SMk&H-SQ1GqP5Kx4EIP!I|-q?pc636WM8ZQq-<z`h=u1b+T?>kop@77p*4
zDr>{+#qrLzg{pJT3hs|2@m@$mI|~t*9$7V783+?`S}}(^p^U12X7z~uAkxhCL6Mty
zzO#W4SI=#lP}E$#x;-Gx5>ey`Kn5<YKp7Anfj-^jHYo&&#YsyL`h&Rpg}9f(pzspn
z#HRs3C=ucT{$RIdJxlYO8dKZf)sH~f|J_6b|KZo}SXzSIzi{_OEJe<w6x7^w9Y12^
zF4&a4OIkpo{7Y5Hmdn3uQ3Ex96&Dnhli3g*kBII7Mov%y2zC{uB!J2r<|6o2qK+1T
zIA!2>dz0C*q<{tjFbN%}u)3uCRmS9~%J0qhfn-{Im6*(bg4{v^kk4WGdpkKI+cw!C
z2CUFkGT;eI7bwx4pi~)l;~T^tu((i726#J}ZI~=HVQn>bYp3uz-;GH&dEXZn$Lbf9
z+MezVXDuQthL?M@=%_4XvdA~&t6Hk<dkDkx!_CmIho?fPwvK>a-t6!FRkQ?a#$XsE
zK^Y%nS=_5YGQQ|odwNF7=Y;udYW=6e!S3}Pe8U6YL#XT3Nt5btp4W?^p1K6X)-2V*
zquNm?0#kcucwDg58-pO6zyks~vPKXMV`tzOF{oy-I<duY$V)t_AbL&fgRgiV4}@e2
zEailP)0F`pqWZs~RUL+O9}e@k#1D{F%J9jBOg1u-V;z{lvhs?xF|$W1?^tw<j>o)h
zJ(dnF`>r$H-7&O65wXdGAh}xpGtQ71^*a^CV$Cmdi*)+$?Cp=N$L={uU<@N@h{g-R
za3vU221Zu@Rrj<@Pr=@`TWX=jRrSx9KC{hTBa;59fn51VQ6aTUY4P_jXxJHn^zPqt
zkg|6V@cVI(h;RSbeq#C0?2VJpmzfJ+&b*{#b*!GsYo~|iaj&&5QMdxNJ_H*kN&o`G
zbertgw(|`V3ULoAm>m1kx;4Z*ZZ(p^onEOedR}Liu`oRB{4t?m+I$%~G_uiZX`$dc
z9)Yeesy^tEcN<H$Dp=-6^Y;~{-q*?bY#Wi%$O5;fIEltCvKd)71$0yQruD0la(ysn
z_K(l`IB2U}sn@`#Go+xJp!bqEqg|j)Amn%x$ZU1Gyc?$X#oC~0l6FXX*@I-;$Ro2^
zn9Tj4@6hsrBsHH8N8_Xp(PmSILR_WN@KU%7x}?k+I+_+3-I#Ia;`W92w_n>}*r+R<
z@{J!^yBZrg4k6I&wSb~2CZhmWvjxs(8*NNtTLl#TbS{X=A_`<g8tWiEid>!n3=Mu9
zsfgS94`#-Vg1Iw9*$-OpKin?LClFS>ia+Ion9oNP;elDDEm$RP#%_XZLDn6Sl>SF*
z-zKfC*TV(Ud?#MU>nl`yd3&kcxN#_6e=eazk%lNW6$;xhu2BX_k31K$aP0{)i`CTE
zCBY@~eng|x-ER;LI})NXP$klZ&W9C7JF$TC(bb7Hh(}+K@3?({<BdTeRqQ_N(0T~9
zh&ufz6xxP(2tbMeG>XV>1yt=zxCLd0Y=i5y%6jOMgcuk-Fc`aTaSiAg%~14+1*DG)
zkO-Jg5l`KAOykGG?iYQ?n=z)nnNqeS`8;i<k^7Tmjl`62$z?C@nd;a-Jn(O>=D0xS
z!zzboR;FrU;rFxjy&1;O43+O|>}wEXX#T|REpqv~F%6|~HS!OaT(5?Pq38-%g)0)!
z+u?iKUG@43N&rYPDX>R-H)^P2E{B6qC*0yR_1Dt8OLGZ}f@NwmX7`=o40uipd}h(f
z?J-2Skn)8SPO&Nj`zj3yX6oC*mED2gAiN6DvFwIn7)uTR75A4Tqc$B`3d#hkEOWb<
zt4*Eu)-TUM=2d879#M=;__)+pRGB1&`c9KrXAS$$gbfWDYH|isN;yAl;XePW_N`O;
zwKYd^GW%*Haz}iU7dTdv3J^Ej{PeP6yEmKXmFcN)9+M58wB%{0Js;Kvhs<w;8wRV`
z#W9=24N{oWJk)M~0j`5$+O%5S)VNSTw)fMg{?ZadrBtf;4w0uNYHwKI8H*+)*hKDL
z(4RizHygSz!w{Psh#|`7EqbM~^cHL^Wd(k;JavWM>La`V^Q6PomtJvS3JGa11jEMv
zvklqdOnX*%dGn4hoFN=`b5(i@n%F%=G}ht)bUsOFLf|C<N}B=JBJ?gcn|yye40L-l
z0VGGWS<bEv{Lhv-O9Rcv-)#^ON#jNC!=E@cC{`ZoG;=qP*8V2?Ok%V_9{Z_|#u#d&
zy>{Oq|Na!Ifn5){%T!T@_YgIvV;Lu(={CXl^?Kx$>qNbGxWdS~A5l}CVWMWF!rhQh
zFb)@_TakSpL>BaMm5#~1lWAp7HA_fb6vRH5{08YZ2kAR(p<QdmeQbZj*nG3Kh2daw
z=(4k-%1j99cAQ0PziJjHuQF8VT2NY6poAlWp3LG<qRS@;C8{<Ii+d$HE@AkU5KtfC
z`BN3xiMD$R79OekJm$A28#&4!=zN&hd$4S(q9U}%&G7>nLhec9CgqpP7`NCk^t&iT
z2C#m*ooRTYi+QD8l(FuD-jR?PJ(VK=pNz0C;@{>Xs{$+6#A^^yh*95*9<dKioQ+ec
z_Vn;nF*T)os6EHhS1a}RLjGh=BsFVAqihTeyzB-Zb-)Ug3Ae!;NUR_cK!`9n^@qcF
zKdtFf@1_)!*_o26lK8_1E0WlKuAhf5M~pIRL=-9fW(yRJ-BkA4+CH0n-2QC(94=%d
zclXOqhXR!O%eL`#$Qd88uv#2ab$V8o`O<O!;Wk>mnP?q$7Bh;A%n%5LC34q-{j*W`
zQyA=iiXP7wI+|@@Bf<fXw(0%**C5r9pPZ&h;D6mk5hthtFRPv5oxRaRL3P1}iCpY$
z$F+DVt=3*V)I&qJh4ZC>OUAgNudwd+s7z0LgL_asC#5Lk@I81pvUK$x%JH#cR}s~s
z*_bNcx}fh)4sFoGNjm!5Yc@t{=RVvSxEiP#NFIZ&fBMfIZ6KVGU)R9VK_N(f)qgs}
z`%&fH#III}DHOR@aNM5kUw~Sj0<8={@c55&*mbt{%3*+Y)HI#iUIkdwGQ~N|-k@ab
zCR<l%>ZddI{ITi}LjHac?C>`2>Qbvjd!6-}o2_)yX(S0s=<YYzODadqcaD=m?6~=5
z^=~)+GcY*E)64x`m_AQ_0)A}~@z4EQ@7}vVEjzuqr6|jm+7;vuhzW+XWlP;)J%2se
zE}QWpwaMG7r^9vafcc}hlq!u;J7=kZQ<h_T>Ke!B?krBay3%^2?=1q8@st$ve{OT>
zGtfW~w#u-!7$-#jq9%RsCX;|M^C`x4rF3EXCD(VkY@NQ$+9#gQ)TQmdJ;-_T`2Toq
zbV{{=XjTo_DR6jpHc<b?`cUl{WjTu_!rxQfz3SB|M%QN7eY$J+M{mB(1}Elxhlr{Y
ze)O0)9`!GWow;jGmNZvZF>Z9~IWzUGBG&0ce_A594YQws$u1kJ#Zhx`*EILn9)^-m
z@*mcpS-l&NeoN-tiS?UI-?O_NX!Y3VvS%E#=ilnT@I3pf;&b;a4FI?H|M>&>GB>lL
z^BtUC;KTLa4CPyDD<=f(xujdGLwVcPOvvL{x1!Zaya_h)eEdb^cTav%h{~z2;4aCo
zu>YT(y1t5`An5v7ZJqAhi1u!+y8gKs8X*=x#@<hS)83QgX8Yc!l~&yjFbE%sD)@6u
z?_Os&tukL#zD~7hIywz(Ji-b)736P_(~*adZ5R{FLr9$6=JfNIf7hThxOdLfE)c@4
zPCtfC0LKb}w&rc~yGhW<j#w_(l(?Eox~l8T^X>t`O&K!H$12}5tiAiRLVgE&7H#;j
zO7~=kPXf%7&43peo3k1@0ewbET6j5<;iv3*JLIF8`|hjkUXtGd(Tx8Aw&1HPLCaB0
z-L#ZR%0Tn;tbORXVP)mJd?23p`VH&G{nv#LkL(Q&dI~-L2~ffRhKt*(Odq$r${Tuk
zUB9ZRuc-XnU~Bdf4`R1Z4hRjF+EOhjaSpw&I)vF+5KdBBeB^PqGoT^IMV$@Xf3603
zc0ipDO-Rjq!eeng`AmhcXxxX--Zze0JdW2F7ah*{g8`#S{*Bka3I$Ge_|#kMr;2+D
z=B24^7WO3uw5DSgY`oSp48@NxUw)`pUpV{se)-AM9Sh(@yEJN?{)GOqzf4RR{Hsg~
z-5;v8tzLAg<MkISuTGhSg-`NH@$QockRTSn*9|E^R`c=RI)}}SjrF(Bp18H;_w38!
zeGcU7_BZ-`3Dho%8N7E``1s~xq<S<3mR$WWVRa&L$|s}qm(>lN_%{xJwEGZ!kei;M
zEUW9!DzS&j1;i{k#hEVxBG|CV+zC;~eMtaQ&#`1rX!S|)XQ(qyo5KF>+y4{_fCmCs
z!%j3;!1=+|GP~6csp2h1PEghuNB5TK<epZdzKFfkSN!Yhe%Wn>cRNn76R<^aK-~0d
zVyDRT1t|Q3FLE^pr`2K^98ye%?|C?Ej9VO3+Hx~LKGY-6YVjhXa6npLT8mpe<;$+E
zYxb!}cCLTAggLPJUGv=Z3Q>Zx9ro`ncB`wwmdRthD?rx>D;JJ&#$2<wk|J|-LNw;&
zfyCZK3e)oxRD2=rMLN6f^X?l@hpG?M2x6DaQOJvfM3!`0n~5X7_DI8Ei)-$K&qfW7
zXucej`*&317ybwT<yvmTi<M2ywq4}$4%slU;y<e+i<#+1>p4@Me8l=9$BbTIs*;cO
zj$xlBCsA}jPP1G2dpw$p5&zUz;NCW*y1MHH*uZd2Gh6owhk@>r!#gs+P|=FQojp0h
z?)CdR*A}bo{K%lv`KLY9TbbB?w9K>R@?L)!7JX6VyP2A!By|?&;@#`VR!3nK>VhwB
z!7c~M6p;I~o0Jo%5N;CNuSO!%_Frah%p)GcWyZE4!*_IBikP*<kXiUp(ReOdthhE~
z5yjM}WE;z)I$YROew_b2GiBGW+o9+=__Rsx0X8%2RLF**<j&#F3XY_G4U%D8mjx6}
zAu&+(;Xhlc%Wy$|NpW%Xb{I%tz5WsJ021rp5p)rtoByE^ep*IGl0f%y>~L7pR8y50
zGIOk~Nw8#WTy&p{RM&Ep#79>O04R5(8LQ~@iDyS{duM%0e)qxXNR(aXPuGTA;2Eck
zv%d}Kv@8IrW<3!MK9tQoQDd#O)9)G1qbh~E?zm@bEA+8b=*)XcyFw_%(_6f!{s1L^
zb0QYhVVOS0jc&%up*3<ZZ}wfv(YNgElyVoQd{}75P-$OW*z|h~B-Xom!aYY9I+`Bq
zwfcmO@i5)==V$p}Zj6!81O?X1a_<(C=KO{q`Cr9HX2Cz8pqtQ0xI}sC`F@YEAkpA%
zhiB0Pd3!MH^m2^TPhZcxz{}(0Pm^A%Q`+Nf8_P=;Q@k=ndtP!@pX_vcdE|(ug=d3Q
zTu5J_2AAVWR(P9JH%T5~`o3?F{|av!mFkHsdo{r}x&eMP*abHJ2mpUVPMbL?PgGAP
z!)^pRJ(8V39-4i{y>xaLq$|n_@v3qhSvBky@}HXxiaiZ!+KqnxOHiy+s!M!3{sR=t
zZH<0|Af;Rg+W<-Cq5iO%285ft!W)vDiTqR(bo8=e+p6%DaQ+yF-T$wUW+puHHkrnU
z!5m*S!*vbSU!%tJUV)=hUbQ-3OKdIgX{CBx3&hW~`W3HLFXt#ad5-3&*z8oT-EKs`
zh6qA(E69z-Hn{np#s(yQNVH@BSmhfAWa`Q9g3fip7oT~>3AL$A<7~jm+{MW326zjq
zy&{5k|F2^Ns;!+T0kSKI5-s&|?Naehx4SW$$-7p1$&H_SMT#Nv=)##6J(Xj;Mgfw$
zjOU)@-ETbq)#vrk4&{>0WOlA|8|=NimU>rILL+y|bNA8T?Rv^~b)mAb)VP=8`Me@A
z<jB=go2RowWjFsdLaWf=0+uwmxsiOX{XtO$WL_Pjp^-xw@uKRVPx}e#LS_}82{uk1
zj2VLM$GaE|GsVAY9NxO~vGhuW^!|2UXIJ?s`#qWca5jITn*JL#3H1U3P=RHNOv~9%
zK=t3Px`g_Nw6T@-zep-1vSGz%((Yr$n=O@|Bl7T1WJH5M_Z)Z1;JJsnl}hYjCz~F(
zqW4)$&=Tl+>0?zdAUJSkNI`J#f~6CwSWGI?NtO0S)N9GpRT5Vo>K)TymVDM7RL3=W
z$-vut8U{61U+ET&%^WYOF7A>)a$H|L-q>=Z=nvueZ;<D0UGyr|i5CXw9RC9^vf|WV
z_(~@=vfN&WK;m97ysm^WHQNWh_x{sb{Kn}0;=ccgA-eFWMjOf@&!oCxqDaQAZ6_Ye
zJ9n=zHMeL@b9#zXzV2oR8~qnTXLwVFdSz$6+w2tY#mfmJKW#P2uwJ6dv<tN3j#9`p
ziOn#6g{J&TLH>bzT~v?d*dCpG%S$QgbBvok?tkQVe#beOmJNpy8SZ9U+%&6W?v+)T
zP-u5@y4H{Ok$G28bz=YZ(^*@e%7YR*3YUIw;Z^tr(aMcX$Hy$Pur%S1T3wYEO)D4E
za6?AO?=+v`51KFF2hFGVe`r3y(CA*=Yc8NoBU}?Qi^O<OBS*bkEMb8}+l*VP1@jTp
zoQJ!v9ys!%G>i|HD$F-$ALvt*1AO={!rBut-5s`*xxAB<4c^kEjkTJALcc@On!ad^
z+lQ5QgCpwkM*6wPe<Qn1=pXhIy1-v;`mT)dx>gA`>^^mJ&LhWEz2yT|zGFwZ8H+>d
zKk4oDrVMG%kNvN~)vttA6_OoqT%Xc1@vOBP&ChF+qGp~EW6(uw?=y2LF`D7k0ZwuH
zW88}UyTgN`FFZ^Wv(23gb(e01i*DKgp5j+nS^UBWkc!+jPA8huzB7PXIuSn@K(49P
z12+ea`eSrl9YK=j4_M@Pn*tb;{{fm|7WhBRUpaQ7`v>!<!ecJ|gZb0l#r)lB4aXwU
z%vmMeL<;n$wXIp_M8_%3VuIO~-A>?Z>Q*o*uZEW(GI2A>MyXG3GaRRCEJ|a{!opQ5
z!Hal`+g#T<k`}H&@x^PB25+A=*%FYAXKUITR(c~HXCUyzIbLVMtuQ}_{>?*Lp*fz^
zi`e9*3HP%|)1eyn^e>#lGF28U3JV95bXt#o_+k;GmSA0*6zcxUq9Lbw;gu&(=kt<>
ze9n+*&(Hb0*C9|H^-CCP%?TQZG5DPMC`gaD*YAxUF?l>{(dP0wwBsz<+`9hStV*jR
zAgliH+Uad~aVD|5!JZl7pPjyo=-m^bd*N`)jj0<!#P*OY2}1U-=YuQ|yDm~8=fPW@
zgNd`|Y>zt7p+1W`{SyZddAXYv{4e3gPaK@?|CNKIE(%|bo_(ht@qCPvWQqCUK2$90
z8Jv7HdjvD?__`xjbb9-9T>vQ^<={dO@8GbM&78HC-8PSHjyGf!F<$G48mGD;H?Qt5
zc-W46)n)>nliHfb-&bRIB|^q?Tu*K$#&+{Q|G-rf)A6`UYRGh4+6IO9ZhZ6K(RSo9
zb;R@WJKrFnj`zRguKvX>CRaUq(AjBQqBVJ|xJm3dC_bX?W~cc(;*cZ_g?Qg_badvY
zs9>b)m>1W4L=nCVWh^r9Mw({3{Os=K{t0Yk>?SAwNK$8zO5LC%7;x6VhXGSPij>~&
z!U1U#<yr1L81XoZsa+Vch!#)4fgHvgM7Y8#w@<fG^&rMG67^bF0Qu<&hp|lN$4N&W
zB{c1HLfQbuO2ydQ2U|XM9nJ*)FpHx`(DUvWTYsVeH$$ckjpQD{8GAKF7Z#kP{f+IJ
zGpGRkE{IWk#41_$zN?ZlBv`lB6S?ai(tkPx50_4UgLFOn4-vom2TT1ENZbve|HN~P
zR}OSKcqNchC_i)Zp7zh6d5uZ?7wB`MUa?}hdw7BIV28>vwV0ng@xMUd>D@v9uH~Xk
z{xc9;{<$MSj?C4YGFO6P&fLt^TYPxrLeKZE1vKA&Guq;>wt_|XQPVG`)^2=*>_e=u
zeS>6MohICPo;Jyllfv+L=w>Rn8O5XNxoZNdJv8(4L^dedP}cwirfGm_%5dBeORp#^
zFO53kz<Ppf{+7hKgk?#;zxylv)sgD^_phMX256<u<AHABGJxK__iaHO+8_pU^5dQ;
z0*@SGBjy9jNGyyuVvM`@T68P7`5xMB>WuQz(l=2p_Vil!=5OEHZ$G>zP#P3H|J9)C
z@9$<|@cyYarzv*Wvgr_F<ALS2>b1Bw1*iSF{CxX5n}k9J_W3=45EDQ;_#~o30@qCn
zn*oBYYehpl1&-J4uVzZ8$4r>dMGOdu56tBMcxylUJ$&!^@SiUlB+&wtzL<f6f<0eg
zJBA10KrYZAfSTjPT_9KAoZ4ZUfcG5Dd!<W}7)Gle%w$^jy11B8x?r|qDtgkA>Vv91
zWeTKX*9HH&FvZ0(*jMTa5NI8OZLpc$6jA4a5GBE#Y9ZPua8l$Nk5v+{BK*ssnzfp7
zvGkB>uT?UkRpH_X4OhuefA@I&E~5C|7d!qd$UuXHR~!R+iga%kv2ogLnct+U_)VRO
z)ws{fo;9inK1xgqWUXj)m8{3vWK7`ja3KjlsD-CQ#0j_Y+&c-8poCELP#FE)S!2&%
zy}rIT$N?q9cqwn<Lm)jG>82saSq~3S6|WLgR+|I~6k~`5#SO#3XPIAUi-h!hjgGfJ
zuh*<p1Zzm;$Ve*^P0fDsGQ|SqpYCP5_l(qfIGG-#Kg22#%PPQ=r+3+A4DNkjP%sBK
z{CG&+_rvhnCXqTOh<^wHg6K#ms)E@1?i5M3vZ(l$gi!F<!Qe5)2MP;0dvbOtS`7Z~
zdA9d2HOY5vCnzRv(1*?SsDpHa^#+P07Yxu~g92=>PaS9nY^#vuM15P|OIG6(Cofms
zzY$%W=*gJP7X#gIHEuO_QmE|yfhc?W2kbfl;rl7R>#uiT{KwD!aRhImJbFn4ddVUQ
z#6~X9&g6>}TIZ`bWkoFP;PkH<v%f%rD82-!3BgAv*ny-~aTv%Ko*W-uv^Fa*Eh~xb
zC<zscRT<|A<<`JTw)}RPcP~9VUX1`2yn{dldB7qB6tURnN1DlXcNhzJ*X4f4aes-J
zQuT`xM!qOk`G$0cM}OEkaArh<2?F%+1ux?DCB%;9CZX-(iU&p$H)POdp6xNs&0-$3
ztd|G$h`&3L=r$q;lzalofLa<i1zK>7<F>%735PAt7DG&_*<SgIcc=XA3y(_d2@gZf
zUFt(E^+8D-AoUM{KrYi;{?>>P9l@!n2hzG?Y!?}&2qSUsg=hG)f>S9!Ee9$;ZFltf
z{SFR>Mq|(2b>Xi!t`gP4cW2#d6xNC4yTyJH0^x#y0ml%}->nne7|{S?)450=WpXP8
zs6tE_UScw>dQ(!X&A-g@QSxb!a5!XVx`!r&+$p$rdx`<Ie5oI{@l0xg=h``4M#udU
zEc+n284!vQ3gh1%l)nxQ^xJRx`TuQ=-$x^6VQrvcgZL~>L>(9Wy<cW_sQ;1a;Wb4K
zKcDmH&q((J=1ut_4%e@6ouaxYdkqp=)U(R+1^UwG?#00j^m2jy(>rN)$D3jsQR;~4
z<SklFI-dla<oTKb{}adJPjnq^adI{jN4;A?@IahHa3NbEq^FwF_`B$Cj7YAP52~?F
zKGDfz!jHC(y1Y{QRk?M}(>+Ty$3&?a3y;4UUNnTegY#kJvEU4cI9R3~xs$!LRm2-x
zL)qXOMX{km^JQ-UdOF~IfJ=iPb(x#DAtHP1lFFvTRWtE*1&V>+Y!r&G2%O+d8|EC2
za_a@oz9DvWso6`_FgA4ga++=V_;{&R`SI(drSW%!nvTZM#Wm#VseU3SFlo{T1p5L1
zyXfIEi?i?YnCi}437dw^*r<qTwD{Lzm0PDuIiBN+W})}Vv_4`?E@n8foOe56xvjj0
znmw1!bl#J1shzM;3j9hTUmSL4BOn}e2RA-l5wj6X3m>XuebcWH;Mr_CHuTu|QMfDP
zn&46Mi3_f>8Fnv-<5q!Murmu;8836*=rr`!3xzz?UeM0|+fn>w*J<X5G4_Ppcr&zV
z^ea3sx?s&wKs`}_bqJrtFBi;XVfE-}wwm2M&6C5_ybH#owk>JGxb9`rbT^03o3EQP
zR_Ds%R%Au#d34zH(jTTX2Qu&V@5MQ8-M|}QK@l^k=%(&x_~OlDZV+vRxi?$_RixXC
z-E&C8RMF%ZRj`iZ1nPS)_o;A@KK>+5eq^A<g)9rluoVz_`b(iSM793eA$7#8Rh&bn
zRlBmN@AVT>g=V59xvg~c=JEQp+E~H;3!qt*V5k6h!c5t4tj8)Bza*^eiyzO}$sDIy
z-c0k(CT^z(bS}~vm-(w>C?aPLfOsT<On)7_eY#HbdU<4qLPU7FM@4O=>CUR8`?7}A
z#>xz>b9(&HC#twJnf&J0j=pZ7@?+5wu)v;}8H&U;q0^#qA}1oo+Fps!6+8(kt8G4W
z-u!WtJ)`D@ptpLr8E0y*qK?3(A6Hx$!;8$&gc>AO^p*AWhKzLWi$=jj%@!wqGL7tO
zau_LsjSuCtSh8ovn6W^`tgl^5RwCV`U0hrH_9?_*db^tb<m^fmYtvHhSMjr?Cv~hu
zSP4E32Q%tt+I~=mSIqk=x5{NG%H;E=bJ4kyjy3v|HbYGjWAj;Rv*klL@zyDJIIaE+
zO7Kf;-k0b+n-lYTIrnrTxi3<%Yf;{5+uuDNJ8K(NM@N+C_p}X94(@7JZ7taQ2s+oM
z!t!`7{*10<)zf8@MP%xO8N_so?V1xP_BWf_p)#49!k#Tam?zA%0jbi$4u4yOl~J|R
znGr2sl9Eo^!3U5AtG)MR-!{}=y`#nNhnFJLyLQsyw!`<w^)xxfeRbH}GOjOBo;6CG
z4spcIo0JTx8*aTaH@=!v9Pl`FZJ2Jqh^t!38eFXID9*T54lkh`q|g-`p(%p*O@EIj
zoRYUTw|txOd2)i;BJz~mvYu^CC;!Fk4+5DrJMj#-rpBpKnx=7E0iP?DBPFGrUT%jS
zoMi19ZjQu&S$C%OA7AoOK7?ohoA!?(e6v|=Z;V}u%<%~^OKvajm-$px`7mnv!G2i`
zUP7!)O4*`!1GR5aIfc36R&o{mIm7rzZHWoKV!cdjE^kw+U=0b%b-4xsO;Qoe(bv<M
zizts{LN>`k#y@sAUb7jLm0hVRdDdK|BD#n2%^M+@1mg*o9v`z=Brv5xFdm`?)^RJb
zrWB_XH+zVKU);Jh^zs$?*Y{bH6r%N5kLCv@xUE8KlZ&S6Sg_vr$O1lof{b`I!w{Wb
z3}Y;RjmYhr*X%^%kfx$3#2S3~jNUg0@I+&)^<*}@7t#eA+c=Ia|Du|L$nj#+6>Z>i
zHF-z3|75`D$yo_y(DMyd#@V)HJUWYhnkE%9>dEAC_OXS7SB%l9v$cV<>&qiCihEPJ
zMA2>xdo+Oux(6m|%fP~Po6pT+G&OFvZ9PX9y|D_V)mbfBJ-obrWNo^_M*ZXZMTAom
zsg|G%@3vvAR$D$e*k@q}^K2e2&l%nPdNycNvur0wOYj&Ru!DjHKw@ni4#UQOv|y65
z?pxt$yI4XsBlUPdX^v@~%GK@pTN=0c)A8`wqOR2uTudmzg4|=nR`ON)1}bSm=a|VM
zU)dfS?Vuxqr%BrXEa~p${m;+Lj;(N+UVj?2hHYz_aB@e#hn0o%3YXis@cCwmS?Q6+
zn7JU^)Jf`Lgp=)9f?*Z@+4K~LIZ*+d*xZ*Dcf%>(SC)fQs&?Xqw}^=25g`tm+m}7J
zR`C)>1$VOGU2xH`iBA6GGmWDWU(SAfc@M63IH0n~mGrbUFMxL%+Vo6w^tdD*VHg)U
zd(^-=UJt=}6*xI8Y&3gzwK?EI+s^j#DvmtH1)Ro6^gHF(d-XhdZ6B@B2bp;do1G7&
zKJNCQeUuQ7lB?4w#)4Mza?_L5R|u~>rD!oX+IX8fyeY(o4~fAjY;zQTfjk#E%??{w
zzW4Q*Id)kmZe+vcT)c4)eOis$Rx*yMJCqFZA(*`}yBqh?S{s_ga5!mE{hkmr-1MPf
zV<1dTb01NgghJeOf?mK)HN|o{sCtESWN5rRXKOjuI{o71<4YEu7ADtBK3a)}EZUm+
zb6~qMwOKyo5392-kseM~jgI>p@7|R_TcjMncT(Hnm=WbP^68x=TLE&{&O9dC4tqkU
zr;f&l&7iDTChd#;z~+6egtKY-Pws~)CCTLEMXZ?wN$R<}(F`I=Rxyq<2u^>U8az90
zw&@NY2KL_=d&HY@>t*XtEY@=e-uZe+uEa}-veXL+diDMb_+79_++Y1LH$yZHvgg^Q
z?r3!xyV+P8$2$6fPw@<}edg8<>-uziF2;NjyFbc6O%ZriN1OW#69j&XJdCP{Nk(Io
z=Fy-A-}<7>QbT?1kW6rg%e9E6Lktf)L}TFRn<Ua7(z&mZWBio|Ts&jL(>Mn=RSa7`
zZ6(^N_bDA+GuR_v+)Ax4I{X=AI1{iiG_rnm^yt!Yyk|yMTFI?3-^$9X*Q?%Mp9)L5
z(WL~0Pm*((b15V?JP2q-^BRy_d;3uvDXI|s(yUyYOBDPVehhJ=do;_+TJ2*00ZyLD
zw{DR~EeiKNI^`LsdG<@bmptd?blIU>(>1y{!L4jkT-k(RGmNK<fQSI8>`lke(yF(?
z6GCr`lZ;L<6=&E(5~<!CK93S%2a)f{VL$<D<FdDcaL*RkW#@<NqAzXI$-JL^6RW+J
zEznpOL8{v53b?RkhHErRFDv!(W=Z>kvpGI0crI>3EKsw!tg526JZ$U3sT78L79t)h
zX1d&_RGuKke_{B;H^>{)0`~&=dnEI!^GHqP%LZy=>+;c11a5$R<L59@@yR)ZxAeJ3
z^`{|HYi(P#q#6$t)S&_fCGYw2V%vs!?@IA)4uiX*I^24f!t_Pm*OItTQ3V<xlQ;5O
z(>RK8Z6Re-dvCdQ>_tXM%a(q9c`ry(*OTq!a%c#o{)mWB)Fw4b0dX4{^bNvDrh#`P
zTXHrP@GESln)bhMX=0NZ?2;ualqgvgf7pM;O#D_<56vP?Q-IF4Io=l?V0+=wM;luQ
zQ~CEIc5<SxMb!8%UyC>N4<OtiUdNYB+p_s9;O}5BT!=KpaYh$q2Bn!_d7`u$##QmP
zx1HUfQ~&Z?3c~;gY%*EjECfs%2E~C2o?k-=hWMg3D8bW59naR-hcCH+Eqr|c`jv$J
zN3Gu_bRP888+*l>%KP@94%^=39LS~|{y1VfTW|x0cOer2g|Svww~V)Nz|l9=^C)Jw
zIF1MCw!}3zl4k}an?zGbUTMAJYJHz^k8ikMKnjqN&T_$C#rUO})j$tx-9{-<gcFZl
zMe5DM<FkI<vLB3;s{k$F+FsOS8-`?2SH#)YkLfl;X+PZPSByBB!@Easg#XPcA*w)J
zrDQ$p(PN^QduTj&QwWzhTc8n^a*eU!JM+!P`@-%rF;8TBoN<YJ!*Og_-P!y4Sz~VX
z;k%dSBmEDcWBC`9H92uHsmw($CvUG86mt5ze0uHK{ta?HP^2gDCgpkpnmAP<jyaH?
z-Aqz$9rJENKyRIT*)J{pMt@@FxQTUCTumxh41>zVcJN*ijqdu@G0ItJI9$1dAP?_5
z<mfxY@K|8}^pNx{OPk+=Xt{SeIjI&bTdz+Zc$jg>E`{)*Z@3A_dZpN(y$(6mqrh<X
z0rx#uD=VvPFT22;jCkWA)Zsd=Tgiwr7fh+965gxtBv!mB?}D9^x9}svg?o=rQu}`F
zGg_O?ZO!G|Q@;=DPs&2wghlvq#`>!fG>DYt!Lw~-F=JJOyucz3<n0stK3PejZYkWW
zyqtr&6IrN^Xx%Bo6cdh&XyY!lOv=2$KS3K<*pWwGTlrunx#ljX_9pGWoc@sVP`xNe
zRN)=*M#^sxIyvJ0$eFYPfB&K0(y#|pu(fyT#TB#tpX9zGEeYAPbx&OKiL5wsvfJ~F
z${tU}`I3N?<1g=q1Z%WFJjt{;c=RlwaBog3VK>4I)h?AiSlM%@QZ$#>k>csmc}M1E
z7VA^~QXux9A1w5g8x7YltGwEk^@Ve*cm55x|1#W-hXDGg{WUD{H{Uf~NjY^eOL(^+
z;~egiS67as-@V2zgHkuIF1IgM?x6>w?&L_1G>9`iG-p|xCmw)d&p8l}Y6`~+I^dRc
zU}aFIfcTFR$CU$+zTx&yVw3k);Xco&WDuMMF{@*0TYlK<qf=L83*)=E358;{WjftO
zo%f)xHj4)6Azz!eZ1E%)!f}5U09%q~+8r!cl3j;XGarkyTxxw@A>{H{BqWikUF}i)
z#e|IjxGlUb%0q6fdn*|G4B26)NUS)1r16zOhgYB~BVELf8Sn(VX^ihbv(KYAYHTJU
zc<Y%d-jfYtJsWvlmG>bb3TI1n4*RIMQbP;;*!`_QI0zcLmD-MmH=8EuttypiJd2RX
ze2{nMO}mjA*UV*OFyyTTn`Z*(gJ^$SEJreqvq|$<uhT~DrPG(Kx~$Bz#`hP|N#|Xi
z-qWu4^(~ji($aUqs}`wXBbMdX1$fWo!M?IuevH?c?iD`0MfcB#KDNkQR`YV^H`b^g
z?u60T7!y1#{SRUbiW-Z;{WO!&?*-44$Bb#5KkMSjbbtMwNp|?v&QZ_vIa*i3t^~uA
zNy(tPv;<d9Fvmh;JZ$l&;6gqwFsIjQi3cwR#H$;Ke+W~q_qcj`<ve%LF$h86-&pH4
zQGdn->H|Afg%f<>$)!*r+u{A*-Ceki73$?Yhow31!#DKFm!=gFtVxH*r7yxcaoEaa
zgMLM>O%bM=I^#6lDe2=Xd`u0rj}l(zU1L1_IS{6yC^wEU$7d6a@RSWOx|R9YSd~HH
zvbRrO(C)m7d!!<{^j7#>BQw`rQmdBjP%iN$9+dSk;YGHpNondw&@xI);-`y`h$?L@
zuqp`NQaQ%2ub!kLx~8u_{Dm`OlcrByBllJ>cFVNn!?Ero&#$zhwFJ`uW!H*`1$B1D
ztxy*K!=#i=17mu;T|?2_Ex!$HX48Ak&&P|yEvWlC8xKFAqrKEl!C=?}KMe0dl!Uh+
zq5dkQm!1YLoOrWVw}QgdB5fY6q2axWN?m1af!5yPRDlhN?WUIAL{@SGO@kb(V;^xl
zB{ssIRnPZ^c|k$v-1gUJ-Lqz(EGb*nLm389C@|)#?B67qMYl~-OjqhacKAXz_XNUp
zBnoZocV6yH1#d9jnp;Wv1~D$JbpO)VTxk5+G3O^lOrN~HrE2{M<c{y~XJ+Oq#J`u1
zn2p|l{mEzqJ!CH4RcZ~k6-TPg8mzH+;ls8CADD}2Wjelh!1M7euQOc$%0A@X1wvis
z%X(?8(Y0W!jd9RH5SUCmEORDyzU6#3M`CVS@ze2R`lrV)>ALQhmx6G$vR8hfYh`AF
z+mk<{nA91t84bYV5XG@z5LdWv|AFC&2Pf$q`0{;yuU)m#)fnwKR{sG4hE6Qmf=$Rm
zmWB@s?t|Tu<M8JqrPb)#y&S>Yw9<VdIMr64s=g*>J6T)PUH$H<gbni01~~OZ4Lt9x
z%NWC1{4g%O85k~gQ2LdJPq=s^;Dm`CQx@{3tl0JA+%uCa`Kem|flo%7jzM<9jj^`G
z{l#(^<T=pT-)8XoiKF>SUyp?Gf`Xy@K66pq>^moHpCk#EwY9kfPpb8)A0SxbPj{W^
z-5h4Ia<sgnd7i&szTLmW&eO2IDmo@|-N~7PbVOJG+_mVCj6iJ~p^%;sE#|*SwC~@G
z5>bwVV`8&gaI6hdF5F{tkZ3Vug)DYZiplRS4RQX;e>UuL^`pJ#8RR7v6KpRIGtG2|
z!6|S<*%?ckay<oQR;$o=vI<|b4ZEErt_Ad_O&vX?5!0Ij!8>VkBN|Q=dR&=#t0}qY
zH8!jJIhuFbv~_Ry`?)9Z4;K*Fc`AFdrQPQuJ5oIA=7ia7c}U^Hq+-9BF0LrV*|guJ
zBKmY|OJr_)ZgZkqhNnlzn3+lwM5AJ6wFO1vzaOiy9Yb{VI8a(<U3~nd@9>3(UMJ|0
zHOsm^j3xWh<FLqO4K#{#%*Syi86!6{Zv5KXXUAca_gI9;_PNq+*3&?4+NB(SNzC`a
zQB6~-KnKTf5N<Mz+zaDUKZR}DavYRVmqN!<qvBpI^*1I%6$nr0Tluf?8c5MLN{AkC
z7ow6;*2Xsww8<X?Is7%S^ILI)GBvhnTkg*CD(6}%2TH3;92duB2aG5@C;3<@sm%^p
zI80#;2vmYpHTp#c>yx)M4?NXB@af6b@(QVhsn6c9XizZ~{B*JBBN5V9rhawSco<`&
zEa*cMjOL<TaDxI*LgISy7D4$XVF}LAqnMXa<w(-(LwDyHu)c*oDSYuq{|Ec_n|Wsc
z*#&#T^0M3V9K2I5wcYl9*ulUuqL*pBUi2RH7xQ}IfuO{%qiu3es?LYiXiF#WV_Q#8
z5)H%+?=kSE=^>qhKQRMan*(O|87JOskMF>`v?|*&``dZs%ZXJ)c30GhQY6RvImrZ}
zR;BiFWi1!~kgFexDU>C1;ZLCJo)Abb&0`y4W23(DC&Lbj9J8tEQaBAl@&jnhK@Wb)
zGd;Z(h0h$BUOPY8mwtwPtSGvUo!sV~BG>#_^QgZQF0;GYmSY4WeWuq}sXP?EB*_u*
z?q2-kkz9SJ%-2mTJioBxl7~JWZV>I_J#BZsfsHqn^4O(^NBtP%HecOLhf?780+B!4
z)`LFJ0J8?K?If8O)(no3PTImT(=J*aV(l*<QQ5uKu~IResk{7rrEc-?_%rk=IKF>#
zYk2G<)0KUnFMK)$B+06Y^k^fss+|W*>YfCS$rc<lVWy}a%0J<+G^}`j^{&f);k#1d
z?r$TsI!fiQKQY&<ersY`F)cXKd6O(;;eQdoK6@8+KTW7#NioWs#h1e?Uc}kde7g=)
z&>6@g-*Mhmi)KiW$%Ax3AMcK~mb=uwm6(=j$MgXmwmuugaVDnCRdM_`xeEVo!Zk!k
z@sDm3-sHj4c?5Ql{3s;CK=R`dV&jG7rb`r(8HIKt9>zhNVlmL=BFqw^gp}90PAqz2
z<(*uoF>xfiY3UpoHv$d|Jp%0=PUZoW31E2;SHM8)ybb~^wMr*J)8?xZY1!`1qLv^H
z(d7tH4ZN&Se+e8pr;$m#nS;i79Z8|tvvlXC8Ep9pS`-Z5w0${w5J&w44TIi-OAKZ1
zGj~wxxBjrK#jsz(I>z{QkraLt3vGaviig0^Y`EYW)^U}T)-6UljQDtk-#n-R_E10E
zrbFK}mj@!o<XPn?X_*=*PJA{kqW~{Fm@#o9sm{6?*A|?+vR2vokp78!5ABDKJ$n!h
zM@#vSL+`=GaB0uEW67+B&SsG!6tfl47pu}{-`U&drjQO6+Beiev8#dVGFd+Ijb4sd
zUOMc{=fD;k4ch37uDxqcx}jdeYAjZD>!-~o6@dpCInk*W#1>s0NaI4SYp?pG>Hz!~
z+vHafUb71FShp;V9Y2GS)7?>7n77=m?om;o2)c0!Ybf_s354da2e-*V>;aTBt78WB
zleG9@^i)uqP$??J45zLyQ-Ibx6~1~VlIIEI3qy>k&S5HqPth@zOi~J*3bgi6BPzq7
zhzb~5=>76G(t)5n3LB>4RYxh=r24JQYe-gLpaex9sjY}!X-wJZ=RU>S11W}T7t1KS
zD{eGM*3bH13p_o?7ijWm?jbZ_P|gD*Z&*Fx>1$U}To$J9Ecz&j=i~#6(T-_}v-;h)
zh#)iywzIf|WG3nj-=6XDU?5!;eQkPvKUdr+_AXj9Lhy{41%kjs^uT33t%Kr`xcPH~
zM6pl~1M}O&w}mgxGml<#Ff!)9wyxtVH2e|t4g&HdSSXpLp9q~!W$!PIR-|v)cksTn
zfqb=+f!dLSkDtchq<pj6k7ZQ@Nwi*riMD7$AmodFw%IeH1=c7VCxUK^P^Wq6Xoy~V
z<~3LV{}{(T#x*CTAJH6V@8)b=cjPJIEX}-qELp<ekW}XHkVE9a)47g`H$78qww#a9
zyRwzjA@stIsuiy$e4gj*(NA{mg6V?PWJVv(Y5S1NSo<JpVgx7j&K)tHnHB%bUGme;
zb|Z|EG5aZ6{8`W_ytr*Ml6s^r84ZmcD-X994KodH?VK4Cdl<;@DDs?%v=-#4J;co6
zIT_edo{eWN#)GopL{06R(zd^|zt5`1`>vSYM|qvmfyY+@8`?Gecp&pImz?f<ni51t
ze3Kh(-v+;vT}^_ZB<*~~TkUU<J$gVzS?9Ph;Vd)x(ONXO=jbZy9{T+vY9;+#^NRJv
zi+C#_m5EBDtHW}}$QXV|4JdmP{l@(IiAPD|^N!CKT(!9w>EoeI;#;El3C{4gCN-Cs
z?b-TM>IpY)cR!VhrT(al+*Tr1eq<ADh>@U_+u{<$rVw567+i4Ie4nC*MZt0w<)=i^
z$`ohlnf$B5h$;4+02>EvELu5)Oz*?fRxdyo^x}0%S55rB{!3@@QwsSIMoiFCP*<r%
z4-rWIH8H#^2`MN<o|xpU#u+cy&eG=Z>(G2$J{Aaca_Sg+-Nz&j84TXcn{Wu9+teyp
zlHG(cYBYIdwXjuwbb2m-=t!gk#fDmj_~QG!gP)!iK6lUvB_6ut<QBcZxSdiYBRW4H
zO0~~bd&xGffWV3;pveWf`UF*f^E#|%XuiQ`gG>X8su)%s2HgV>uLU*62h~HkndP<k
zlAE)PBT!gy6OaW<HD4TF{50DaM%3Z3lU9*|y!`T?>uWZor8f7}T_rt3+<-FqQ$$`I
z;=NFbgGJV8jhD$>UA}7JHr^?m<Q8H3`F6FuvUD3gO@HYxys6m};$?jH3x5tg;+4NH
z?qK34@{>%ilf5INZbnpn(x)4)#Vgj^bAPSm8_qISj_`-Qz?U3HVTU6MK-%}r0Q&CT
z$D-W&!Vj-JBJ;Nit*N9l3WkwkJ@7+5oTu<1=p?(WrKF`htQ{5aqS6FZ_9UIfy_TqF
zrKIdA`T!eaTP7a>Cp_3+sINxiWXgv`_WYEDlgbLNx(XG0McYoyIXfdl2e#QK98kr^
zEiPdtlMYA<vcfF^F6?_^)`{$oW04uut2OO}L4&)wI?WVGqBm;$r)idp&@MqcVVN@o
zjOly}S!u{S^1U+{>v==>^1Ha__9I+@=1Di~28#nBnwh)L`hR|=oVLXuk|U-Xwvj^f
z5vjw&FgnCsqK`bezRH`NpnFLWa}cXW=HW!Y{61?DXlY(6Pc(RL6_o_!3FxdQcdZLA
zrBzpMC@$5p_Mp(fqDP~VFpTXehv^X9c_Phzbk=+u)8#Z#ajc{5RXX(j>PS!9G;|Cx
zJKR`%?w)dT#q8Nq-FgR^nKW-x%DRQZ7q|V~ZAL+eO^GuJ4~N}=vYLPFeC1hWU~hN(
z-r9;Yt3AZK`J76@YOL)WuQ;4{i>ga=H!VJFslHC5FF4<_Eai!Md(XZ+=_|9Mfv3Y6
znfpvaWFhOe%;XmML7aGuM-d;n7epe?aN%crLq;81p~AhF_G!m?@9We%-<eY#!8@8P
z6c`{wIOb!s-hia2*^90{#c1Z0qVYUUiPuu)y0+AFf!uf6J7#6QKa-p6SU}z+lMS<G
z$&Ev>_e-m}Sh+ESX>X!}T$8>gKd9XRVkDR9;3bUh4T7n9^psrk=TpNP-gg#mQr7&7
z6<(i^--^d7W#X33k)F+!qKa{1NQ$S1ZVxjv+g4^DOEbPHDd9pQBx<Dm%!=^_$AL$*
zYOGQY^9L{ham~WY;%l{+FVa5RZmL42!t3L#xIk8&3n~2@<QizrM06c2o#I_o9UKnv
zcgfgmWq7su`c{Uswj_0Q>J@Q2sXSS~@3j%gwh3dcK22Ku3ZaEDGD|knKmI-~#k<#~
zH$z~}D-9RDuO4~t{~p9-ll_VTuq}sj<ONf6mrbjsNUi$V4Me;;pkq%y{9G)HxH@$o
zyz#=@POzKMC~vY17IUvle9OA5Vw82(e&VQ}XuZb$k#`hom)KGWtJ*!K2*#V`n*3PO
zOD`VhRC=p%?R|}>_FtCF5vUIcxJmi<?7#0c#hcCdF8bJ^=ztl1zeVZN2_-sNXZwd}
z;Rgq$mjeQ+hBAF@G3;3H3;0!R=MuUOIirKE@gCmS;qF@Ay7WGb)wV<8V0f5ozR@bW
zmDvY^7ybhJAS4s_i?_%-q$nf{K=B7MtZ%2?cB&khiwdTOGLmR*?7|PE-VLSMSgN<7
z0GeDu`VqWb1CNK_jfCvNkr^X>@g#4%u4(~LBx!6(#PotTr?%`w`K%K!eB8Ifhc*gS
z$GsDs930)Lj;tJv8JQhO)G?2DRS`|u0F81GZ^T=s_Tbk~;k>GGOP6qpVac#mrBrwR
zp0Gu)xZ<v<&dPc93=hx+{`z#kU{fsqkbPvn;*EVLW@|Y;Lo^|YlWM*(uxYP;iA|BA
zx(}o*e-qH80-UF|oyq5tjj-Y(nHSP`s%BP1s3bMqMRE$=w<1VS)Wr!}ql*<4oa=aS
zpx9jcJtMWB6nvN_Sey{sWK@=WmtWBNx{Gps8C5LajEJzzN`-<>QhtP1I8Wa})|RXm
zn$dfoj!~Pf^HN3S@#o`3rCzC|L(h-`iE_T^*wK(CCV1GPGGEzErl`Xw-QP8etOzKY
z(e#D}JTZ}$mM+a>aFR0wWpdK9dRHDO(#N&&gq}Qmj6w^-+*t5ib==>ghtHgtO>S&>
zi(n9v%UPSpCZr(WJKpY-Hs98^1Z-NL54HUso?6gSJs$Zo(URJFMoatnJ$o-dJHF4A
zf)9`k{(`uK2Ia{;xRzIM`(4V*&y^c6=D5W-6EPkd%(Nm0g3fE>uaTRPyf3hLyuA}{
zlITO59e?}%ll9Qz>G802^5>~!-e5cYkrv2xauX$)%{E4mzwWSw|Bg;}o?`8zdjB^s
zli3U#789dR8WUzv#(D2GS%_j-M1wJ<IpDnxE0}WlddDB<alDV@{hSy`@$9vHAOs>o
zp+(uZ@MF6CZi`*S1ca{JjZ-D4dl>cWvrPYAdsiL~W#6_<wrp7=iRlSRwnxd%i0m;*
z_L#DS#+W=5GK{QQGM11fgJg+JmO-|$6G`^17!=uOD27>{?|z=+c)xe+kN3~#J-+X+
znK_Q%J;#0gj{ClT*L9xfb)9NfV?IdViFfFlrFbJ#X9>gk<pDlUeYMkhLERzg_vQt3
zPf?(Yy^)ucK$MBm&1~D^oSV#bCZ8>@o6J71Oaxi4kMSq#X+Dv(9sRL+cHriSr&Er&
z{)w&A>BnqMUS9jwcF1=bH!#l=itU1stXq&)q+kI@?~`toucah^6TSZKw1=xYtB2Qc
z4b9BhaZ3d|HM&HIloxA9^Xlr{)r3Y)H6(rBpbaE=J)`#(2`XytxG8ggi9Y@vkeyJ~
zDd4?_9~FJHHAmJQdrECB%x^R<FK!u1B_<|6ewj2bx}G6FtpfUf{2vmRzp358&#-tU
zTHa6_<Zu<tNIyxJ9}l4XKq&5-Kif>&5&gwgi$kc4aeo)ntu)=q5Qg;(G0|Ls2pLP&
zV(>k%rSeW|Z{rb}K<IK_k;z=$7n^8(p3nP!{y(YJ9f1@4+1ldv0wzjVY!yqGpjqJb
z5wA0|q)c2>FGjDwuetJK-KBu+wx^2}KpCMEv1~NA`=pf~HM5Q#cu+Gnq(e_MXuo)5
zN{q2n1&a3_I5`if0^OPhMHa7uMFAdSWMxq)!G}rstXFf6lRn{oIv0RSnqKv>k*XV+
zx&_*U#@42P4B)wFU^Lfwze2UW(Ov)1Map8`zoPXtAQz5!itf$qz*2eDc9IB|>&r?r
z)_>TE$HYW-Bm<0~0sRqx-E|+d4o&M$u6#K+&1h7>2I2<4=kpfJJndvjlSjxtF=x5z
z>%+b9o8G%`7mO#G#fT!C;za0-oay)v80+kueO+9DLZt$pC0^@M3$s+tnbd3iNnb^M
zzB<M4uTivmVO59?7~gj5c^VV>^JFK&-$rr2j;t?J-4LktEQ&kFK-D2VDxoReCjS_e
z!Rc4(<xNsD8OEu;D~G}FU5YZuf~E`Q!AsX}$VIDft*^;#p<^1u%8F<jq{_@nPpa9F
zSyjx+z0P2ATf$jmw}z7kkTyuw(_xxkHKmNa(zA~i)V^Ho13$5KW=9?<@IF;kZHLY}
zr8AxJOf4c}-j@E;TZR*-tdMaXFpxGVaA`<bINd5rHKQt7x~cJ=*_AA!`w<5WXMUbv
z<c19ywrNF`TqD$ez!$)<;*gjPm2*-io2-wTZ1e0s?fO!$s!QuDs^%irKb&Lk2opxJ
zQDUh!!01E3<KqgqF8HQ-L6Ezz$rWw`$^O{%xCSY3o-bF*2^-`noPKs<->~6h3XAaQ
zY|EP)*N*ci7YhL+Q%G{V$zdJ8Go1=&3L9ZaGNDK1h8PZ$9|7lt`3*XmnK*$fU(>xn
zDSL&|kG&qDUKm05w5fA<?H(O$Y7o8M-UfTkgaVibmJef=7$RB%B{yyodlN`*rBfG=
zgart6B}-@$O?`=n;L>s!$y)rTGfkNk+ZlRs|8PDI|IGqly(<;Reck9@OK^tQ(@^-B
zu6$VU;3D+#5IK!FpXl3qiqf-c?_z-z!6Ji9n3W{H&p0uo=WI_aTTHt-Ex<Mcg9~Y<
zkxPJ%>EunCAsG>Vq-#(l4en*PRbAlusn1mQNqkYpj1xDL?mN<6YU8m-#a>>sla~0H
zDYdJQuboAovhDu6K>BY-z~4UEe`d4$kJ0o0zn}Z3XZ<g_AIv}V6#R?c>wmbPG$tSt
zCwDI*r!)83DIQc?imMZ5<B6lw144v`*g)#aRXOo4p9?}<32HZ(-`YX+c;^r0v`ofp
zH$zOTCj){f);g^sv~q&e(o;pBYHCt1BpZhD&96PgZUW((^(MyPO;a|Mg?73(FQu<T
zgwYc^g1q*iKC84(Qscw7%!U*BymN==aPpK{GEAQ$M#|}i2%}7joxt4FrU9<j_>#=1
znB(f|-X0!_f^7<EK<1y&*z@vJw2i0~iAj9Wu7*%GcAIL1^%H;Z&_l`9Tv;oMXZW`s
zr5&UEVL-)dJa!+Q2$P@%2Ir%6Ynn1fI-ylJjYhaH8hSH_txCo@*p&E5Hu04BaoZKp
z^U*{snxEi`<#7%hhE#iYIlC*xFufhCSn+TwxLcuS{*3>ZhPVrk_AehRWpMdN0Ct5Z
z3tKJd@xa7PIAo@TsGCL>T{1^J7@xJAc)i5xmS9zH<IR0*W&bxoA$#FL?2PGo_`yL9
zc4-or)NH<l-Z&s39nvhCC{Grj2U@nY1Q%I(zMNUJ9=dm<tY7v9sBUL~TD1>&J>cqE
zYFg|<G?t!Q#VbFFWa=zc|8Vqs`K1F3Dge*6U&4c@wU(GsX4pkZO7;hf^KffVo{H>;
z^h#U3lZJxt(~RPl5j+tmVtS0=KkkHc%8_;%T)af65XD1RB5c%!wYj0Hg+!n64#Y9P
z;WWr8=P-+46EI9S)iKZij+=x<0K<7z2#ExaW@$3ic&2Xd>?mUUf$UJa{A2B@xF7Kr
zeW7C?*-NZo5rle6{Ohn$Y&}qVOv`jk?sgVYLJLE0&6*9T!|_f{BWC<*#}iF!WMc2Q
zaIEa7&K*VEN}8Bvxz<s@|3?+o#`F-RC-{dr+6JC5t7R0YSdt{1Rc9m~k%=@Mvfhx}
zF=uc=(CCt|_qxJZ=-bo@BXV?hMqIzLb;%mL==Eb4SW(WT(LF)RIyH<U0|k;&>qUVe
zm;S|JAC$(eJxrKzN4H_KClbpSE&jsQ<kN@gWXv;n=0oFGBH5K$CCz^Ga|S*mxu-J>
z^?Rq3MQL(mz;c5?`LRftLhgA{{HPi<)rmXfVKQ)4etX5j*)7lb6XE6w&!U==4m^>H
z(@t6<>dTjKx+Gp2|DuvO1)=RmHKQ3%Dxrcax)c>{N9>D*NBWq9{h<MAEu952A+ncq
zqosJ>gQx|QDTZhr7h$q-4}>B;xhcJXi-z!{oFMV796=vb5SMy(GEzkNMcs@QzL?+L
zc%o@!W-7?mvr_;ufe=gV>}+9gR=S#UH1ar-y{Im#=Q}l9o~cW+-1E}|2Mws$l50mB
zdrxxLC^Lc9OjoUDKF`eS)vH=dcPZ_=s2(f%W^@_1Jg}z&D1Q~g=3qSE3&;sQGa0d@
z*tSQtuP=18*A1cra4UmK+Xg*mS@-VTPhQ#gVzTWurW!H&(SiVRoC7Th8%9^pCtWhf
za9<AQX)V7U)?v}=#t)vEI36#S;C)O|E5%v-Y1sgq9`CpQc=C2Q_=E}?10<oc<Pd{|
zjXj1`jYFP5V6e#19=XT<^8;Dp7w1L0L;hV0@vqv5zxr?V??3Tr`)@usa4~?l1(*fB
z5FFO;B0o}rpwZ!-=PqrkfHHO~j%VMn8mrA%wjVoI(b2kGT>ZbDV7zX6;3wE>_RUy!
zK#Wh!Ik}Iy6cU|zlVafQ_u@>S|FU_svlJgX{9xDsKwEv~j2TC_Ne|aM32Z1s@@SXY
z$U21o2m8*Q%;l5{ALr39s7A)uVg)*}G<cD4rd0=}K*5j__9DnD8!)1@@;5ZUwrP>Y
zZDeGfm-v^^OQ6*{2p*BVv~4(xK-)ujv{?&jf3$lLVi@AijmVf<DngezQP?S{1L=bn
zOnHm>%{2emB0(KOtl+zQozc~*6msg|*SS@t`<R2Jy>Cy^U-Z)tAui5j+=6i|LjxsC
zhIAl>ge|nX>X1GoGV?K4V`DtTxeU*jhX+aD3(Dus1Qh%0#kdVE(h&9HQ{}ZgP?#3g
za{c`fjZ1y6I}}Q;@Gc?FW7q)V==5&>T`WMmRYE+{-U9%MKkN~@W^(f)*^R-GUJl?u
zz?w~2;%k6zY9Dbizh68z72fSkk}#LaN)_Gw%A^5RKB=bA#O+{=q?!P03{l2ZXNv9i
zo<BWBo1=2hnSwVgE(rGHROqXH09r~?mrj>VvP#eJF0!#7IKmcqj=d1w&gVJiw1DfP
zdP5Ic9^w<abk2qZ0$B4>xp;8>hFFQ?9Pr0CS|K+-DXyv<pl)j$Q!8gS_|gEEYLiYf
zZVWwTx966&XG7sAq43&eXki5I`nQ%<n2IPTyc-sLXT*?-QMtnYg6}=k=XhEVJ)H1W
z7FG)z?FQO~Qyf3Ce`f~deeAJeBUttgh7O98T+$Kp5~k)d`o-KB-6XzYpD|*vkyfa@
z$71)jgeDNU_XtlKY6HhXco=8Q_KX*}GaZ|Xnzc235~S*tqjeVYWF1<*9w8v*4qSkg
z=W&o-_k4}o=4CP$gHLB78{DgKsyX74l%~t4ZfTlak{W*St*df}dD1~XFR<6drxN@V
zVG01bQct}TG$TGa^Y!~-XdE)@y3<?M&%~?W`G1L!lBAd`iOT^!9|(GiE>|xV!uXKK
zL2;APS1{`6_8hAkTN7~~@=u%hNwR|O+Zf+HH2ag!hkyx0Jaz}32j~^5`UiezvY3eu
zU7<>>PBg+cUXUKn3_dVo2n@?e(wh?0i~T80X)rc)I~?+5QX81W>z1C$$iZ{cLG(1j
zmkC`SfF8|S*ruyp)&s|4NjInvF*u|?v*WW!XJE4T!}JfCg@!MjZ|)*^J~h7V@0Fi6
zWgE|7-APn^lr9;67|0L;Y)A|k34lg^hxd1;^YBpuu@qb_lMtpAWUf!alj3LKIB*bO
z-7NgUtj_00NR`JLEJG}j@1NLlx94)#wY$iYUNE@83t-+LS1=c2#D#0J78r%=)u*uS
zKKgpWy&>Z?DEOC#?9asXp*ak9{>};JTBcyn9_$n|gVU61&{@<@&Rpb;rkP|5UY&4p
z9|T2xiV`5su2!Wr4P1>(X~PkK(33IKf~5%(dM5Ht2c+GI80cJmKu+zO<a%WvXO;>&
z@1!opoHTv_@MIh%o1#D(s7^0LX;h%JYbRR8QD;as&*rbiEW&;CoE=xQ7X`|szqH#3
zoEtbA&XUXKe`o<*pc{9Z9;GX7Ao0K``$A%>B^%D2T_){^3do<qfsuF){_`DH`Fc{p
zqDY6fRQ5SI>rhp1?+9QU^+LuH3{WJ{O~?<(t$IKv>)blc4!jd7s*_7Y8(LIv%Fv!V
z#kj};<w<%_R*sS(C+-I}vb`>O{541>=|O#CJ(pyW2Wl$oy~praD)B<!(RG5?{7m_x
z9Ze`ZRBi7bJ)`+}*(~J_yutTAoPOC>XU^&@1h1kXuevv!Yr3nYom6_<fT<F+PoN2$
zgJGtIOw_V=l5np=v5dFDFCYhkp<%g`DJWKoK_QKmqO$;ta%71Kf$~gOu(=jo-E9(;
zc~-i*{C<nFBH9C=^!g9$h`((8#{4tesQ>D-7kHSw2(u6(W&>Zgnr+;jlS73+zcb-f
zuIznh@{C7xhbG(gBMnlai)_4(haJs<i}f3H5^M>xho&juj@k!3uYC>ZoI@Jk1zm|Y
zkGg-p-Tu{<XnR{<K5s8>>ebLqdMz5UEy9>_ksMhDe8!`Vr&ZK_e{J{WJ~KD>%ec8A
zDEQ@Ua&=$PeKtbydM+t5k|s-3;&q@v+cgb~qhHv|nulneyM9!X<;&(3CihfR9aE{?
zRQQWO#etlf8JWd!Va)CF8G}&xSv5m#xI7{$%a#S)U+%>Ke?;l{BD5jJv);+41jQ?4
zIZvUh!#OAau6xpbNH0KpZ%0FF*b<6>{-eRc&(CA<e$ua-GbV=oso$1%Ro_?|ePw4q
zv6Kt+JVgTfQJPQ*gA<ew)PrUtTlv5w(7Y4ALe8ZUx9d}iV+XG}uw>e4`Ae7r*nfWt
z3>fMlq|1x?l`@{+y?~8x6{{huW+8uVh!?B<B>7C%;#*sc?M#=4iTYU_O^X5}Z7;~g
z0w5+gP0MZJwUkEk>{Q2_^1iksTQ@vDg@f)$5ka{iJwO&@7z_jz33_}j_1%V&u8d=^
zPVEz+cpp47{^RED6hzWdTs?dRSmTbIc3lfrtPJC$SUMEDZfzD$0|aTRsj_=&@Y!(f
zScZf(l@Ab`SmA-@5OE+&N+|_Knm%Q8LB8KWph@i4fWF<kKd=-xK(uSjc)TBlCYgEw
z%2Hre421p^{S6aj44`ZoFxuU4Ep#LR*y#$RP^9spL|uOPFeSmbrYVHOS2Es9|5NX&
z-)ciVO{S7l3;Is1RQh9VE9Kg-JPa^OapderU9ljzC8;jUPs`a9UhLTI&Gliebfo&q
ztC#(SWv&;ancq}SgmbRokw+=ku(wIO&ci-<vB!SZ<~d%)0lgL#YbL$;0F`Sj_d0o+
z%~>1et@1lwg$wcOu^(w=2ae}L^t*VP#UYRZ9R^k72y(e(B=DFm&GUYyKzj~#iV1}<
zAF5&ccMgAq+SBzj$auZ}S!bIVV;1JIYwR6h(p3vJ%W;6xB*bQimBWv3Rt#3e#s@e^
z-ntZ0)Bol);|xJBpapF5aZ|YSN>wBj?{dYKEN_J=%s*12ElJ$wpfGptwF;7SpLj0N
zg$K|n9=DG;IT<cDnNZIpp08ZF<M!yCRGvk^&l|pf(%buQZ2w<q4d`cP4*$&&^8e_)
IGygvQKikB`7ytkO

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/ppc-pois-pois.jpg b/docs/2_39/stan-users-guide/img/ppc-pois-pois.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..03234fe14e2a6b2ec7b3a4b7e0fa3928eb0b522f
GIT binary patch
literal 179965
zcmeFZcU)7=wlKUQAianb0VPNgkS1LO34#cS2%_{ZHBv$sG!R6jNmoHoDFT8hL^{$!
zM?|HAfDnq6nowdO?cI3Jd7g9bz3+S9dw=iu$M?I>hP|^gvu4e#HEXX~vt}l=PqcY}
z`@(tS^8g(k02qOP0Br%#H3)}$0)VM0Z~_1TW`Hgp4lsZahyn<KHvrHV(ElTaiMass
z518TgekN^^>4&ktzO%W7*?HrO2ER!&eR2&5@TC_806+hbAPYl1Q5#!3QKl(?9uNfB
zK&ou6ZovU(&CM_FQ`pDd|NPgBF!8$uz^MFxod2)b|K=N;yGO7a$O1E%j&Kil^8?{<
z0APIL77!Ey08G4K`l%Zs0sHWN8+e1j27<85KJ4{37|a8Bul)^9_>FVbLLbC=0>Yeb
z-mdN-{0fBSuKiQG*FV93YXgV@(lPY#_XvW!iSEn7JP6?l_x1Rl@wdl+lKoGZ|0mGT
zHx%Ug_XP!??7==)Ou$cRkoQ=3*K_-@0tmlyb2YdE!V(}n9R|0$^mmvV5Ta)R!pFf5
z7Yp^VJPX30M$j912A#JAVUP}ekawujKj7S(-dC-`bbc_s)BoBfv%i=9<QA+C$`TL*
z)2ZGb#{2RW0RV_7JjD1a2!nJWIzb4FeSEMD5KniHa|R$R2f|4{2$TIbfHWcX!C_bS
z)A!r+>87{-emnQm-(PpVU=G5dZbBA4d=2;U!MY&ifDrS)w}C;#-}lmfy<lAo=RAUc
z^9#~p5DD?VY6PZ(^)mQ|1YP}mS%!2^_<7^M<5znJ8SSU<mz@sq{jH;5+ZgGCLM`@r
z2kT-~a1Amr1YwXSqq&E_<-R=k;V{68&K2+g5WqF?=?3@%6hIWP01N>=@G}4i0&_e8
zIN%GWc!0ScV2TI$?H@Suzsm&!K43Z=2>YEE3_i_&WB#q=KX63>FEG^~`0*Rp4bT@U
z0YX7a-oPrD>iq{zQ9vL377C_${cq^}tt~(QP6xptz;Kd51-!KwE;5{9IK`j|h%&$!
zbQyFQ&Vi|#3|fECHV4c6gQNZS`~++H8=XL~MvvckRv_;kAdew{EBIUhF%UrT9~AyY
zD(OO@(tqZ}1UbXq2Gel=Br7;r(&rBh`G1k9*?%T;#xION7{|fq6yxVVXoxb-Fn$LA
zra;;J$qVAo{J%?izo+0}k0SmB@89;tKd{b%ocjJQ=?6*y4)W~rPdv;2GmSoE%i%w3
z8)s;A7-jE6ditje{h7Ki1&{bkgiAn-+@Gl@Cr_@O96wojviRf*<UXVh(h7M4X@a~5
zL?P{v4hRO)3qeDg|G@u0>Chj25A3&mU$6Gty{}#W;P9Vi4Qec>9;Yy;IOjP|D5oUn
z#Xqn_IR!YCIE^@^z#QQ}`t48s7X4?=LAlt1wf#dr|8ZUW?+W}6N?;$*aex{C|Fgz_
z(E&@a#c<#TNHYkO_rAXQ0}&wZKfif@=ZHasL5acm-)Q%#eVzXY?E?k}&K)=lh#oj`
zK=XjofeZWaKi=g7Qed7Y_%!(Ewt@Qb2gbi}>Jj1*zCWG-`iOvtAh?%zh^X?(lNzFD
z!J*GX)Y#umK~B`w*H`p+R<LN0N3cgwn1{Opus?3@mji$;i{IlR9sl5;X(6TnaAqB}
zeM5hy*@XhYt6%^SQ2jGa8nl&s`2f(U?iLyp_BVRqK>Yy#f<0A~e}AWA*8}a>9F4XG
z+A*d~0H9>hXryN}8l?!N{}}+<eQE8$L1u>6Ow15EF@XLc9poS#ts4M`8ahV0-`C#>
z19bEd2F3$S%q*;IAVM8CKu-sO&@(_785tNr>_{*lU^vLgb4=;%fkT&FnZyElm2YPh
zGK=d~b?{jX<0MqB1;wzi@(Uam6p}nHB`tG8RZU&vl;&yua|Y)Pjf^i`xoT-;4YKFv
z?&0a>4Zj{75*l_RJmTh^*t>D@35iLW_a8jW%6^oS`>d$A<aue?i}LE4+PeCN#y3r!
zUEMvseQ*2Ujf{?sPfUKAnqF93`ttSL_a8r(@ta%QJA_}v-MxLj=m5yy*aDw_<Luw?
zbr9r>o`C_vz_ia79sP}c#t$+u9#cBNbM`WmYv3U<<=f1>dKraP9W3H17C63ZLBp*4
z5~>T5_<h!XbM}A7Sj>Nmvp*R7ldl<YFrfc^LFnlrj1UNf@c<)u9bi7Ne=)N%|GrrN
zezE_)IQK8EKQ0<bgbt*^z`(!+{&TajuyOyFFWM~VMh4QR0d@!-s7#Q901TjzpCw5H
zf8GDTmH~t|^dI&kBm)5@9z0w`yl-lCC5GdRXbf!rYCi5Yl92}JE9B6C+vzj_A4(31
z8idwF(tx!qB)#h>mhH7L80QD+F@(;z97jNbecOdqYyFSwZx%7esD-{L1sd?I&W#4d
z<e+xNqG1ITE!z>bP_Ge0L96Am%R_g_N^R~4`rDO74Kj}uA(^OO=s?ybURjbw9J+p5
zS*>qs8y{G@Y{1Qf_3Ut22i}?~*)}QI0~Bi-z)($TA_wB#?3*b0I;z(Z6+S2JTIcSS
zI~<vHFl5>z*~e?YTP}TXC|YG1uo_THN4jmmYfcB=2af(qB&$7ls(k$7`b1VXqQO_i
z>`2hjKDN7_3}3dFazCT{xMsKrEW|1NcR~$rqj!VR$$N;1x^U0P*kW_pyVAI-5$-Xa
z2zl`?Kl{7jnLy{NZQ!y2?|*jF<&^u_>BO@-tXSjVcr9P+#)o1SpFYwB(VY_Pv8!pV
z8S`^%9sQoVJa_p))^X0{)9%V&pq-c_sQG&5O@s~$4XAn!qa15OSEJ_hoHsqNte91P
z8nAa+9*BJUci!j>q7GBvqUzASm}xF5E8QX&2l2b@W{X;ZjZc8h4}wdV2Xw0PTB2k~
zzw$sWG*7SFJ$Aa(FY5bm$?Qw#_iekES5e|!zqn=78rMnRf$aZcb3gEhh9*SnsH`4F
z8a7|;yb0%bCaC<x?uji&!!E)WT!?JcI7by4069hj7JuT2TkWIjZ(9sXwL}JPu)vC}
zH}NDHREHAsEXjf_T1#vw4d0vW8k%+dJblmDS7hPAjLyi+ggaE@`6+bK7f@qF!r#g$
zA#GlJu5vz<hUQOmZ5D9-h>^LLr<N<gBcA+H=B@bHK@00m1m0}EZ|x4K5byB{X-iN}
zV%4PXn7?+TVr_qk-|0mU;SY6d#RbQ|SxU&hJODiKeKA83qV_x2Dm*5ck=cfHWT)**
z9_|d?a=b9iB8x!zhrPd-d8v2NXw&%9N?x!o)OZ-|+bfPEs&EOH@`6m`b|6$@^d(o1
zkY8`rLWR9i?Fp&2+C1igBMQ529u66K$`OmvM@xv=T}@xo_!G;guXAy;uOPI89x_E6
z=>^DM$B4t`A7fun_Bu~#^yVqynYve6#TI9rM%UK-dLIYNtxF_XthE@k>#~P*sO^2%
zW#qm5U%xHJ)a12{>wMO%mStt#+u@#}DF-ecuuXtS-M04GDx$L1(SYOnONhO_5!m(u
zBM=n}Z>r3GH4|B5)DpLcl|;Uku)PwntRW*VhMxSIO+L6my|@P>aV<N)J3%ompih!6
zpntuJG+U^+v{O5)VO87QY&ugI7yc#P@04$Ol{gdnh}Bn=G7Z3)!gh_&$uwXkK7j^o
zS4czly#I#}ArRW|e{6zxy+FOULiPW0LVbeS(}1A~%!?%Ii8jXYjitUWblT1QM3~pn
z6B{=8q^Eik4K<^WuRfU3&TmE3JTo#nEa=x_{X>zg)<@=Bp(fIR1QaeIp9W-<LG}z=
zT53nvq%fEq`3k>`k}TW9VunH-@89=<v*=7Sz<RTk#3qkUp#k&H>!T!bM%+IzGdhxE
z3(`7QUuUsZ9fqVHz5r6Buk7l{0=v{2I(^>^1qnwR;U(Md-3&AHQ%k|b%Zq3%WZjUt
z?vZxnL0BNQ{}g&ckto%-dJMXHzu<b56ux4-z5);F4lLH-nV}!DnU<L)*tiG0`udsa
zA)!ptQn%CJg$vzwhJ1sFBAp_uj<rdX4V%1GXO0;1Dgl!MH<Te?XWC<<M|8L>0Q2|V
zCo*AN6(|H3QQB+GYthk)Yf7Wz=d&MS7pvA|<|e)t%>IfRjx4+F_wD5O>*ezyOpK|{
zWd6xTkr%h{umu{>>0J_73eVMFQ54G0GP$>sC_h%6`YO>k?0Z3Q>T^se%nK7M0wr?c
zd^#~~+M52y?if_3NcjR=vyIq5m8F;cGM+j)?elcH(7R28Yb7**@gyZ9^r_AYAs4so
zGtKd3&YTol@g&fs*1cF`&-~!4;vnt8lS@*rz!HH(2%tP`6QPb`*;lEO<TE+3%1SYt
zAAQ*#3)ccIhZrT?=>=_$D5rN-GWp;U3#hwoDr5~jS0@b+Y?m5A>cb^p#RDxDyM6^5
z=4GBEhMdVUKxt!fhA_dEHdaar(hV22cb?yFB0FGIKmURA&nsny9os42uB5&Z{HY7I
zqfSE`1p3f1(7Io7RE}`Y%}yxC^+4|pWi!wF7fjc>3>74##7~^sMo)1aB*q)yvg3DP
zeBr5Ua;i^-rhNvQ6IH5(Z-poui@Qm&mK*n=)S);F?5<I1%>G**58JL<`87$%CElta
zTZ3csF{CDLhH<=(!oN8^#LtQj7`}hRw=cS~T}G-vvw*AD?+TTbk>pPuhaFVQXQOt{
z9kM^ebm5x#8y|akCGxq1q5NzQDKZtJFk+!iuPL0*Fu|1E2BQHTju;Qh8RRYeeX}%M
zTsLRTjW@L;8}j0F?Odl#C5ne)x~08B9n$F}l6E%;Je0;sf+z;>49YNRj<dCKWQPqe
z|Ea<9-j&xaw~;vq3!n5#I7si>4iz#auZv;)wUtPL)E-!E!`&Sy;21vr@uMP-ooVUp
z&!$g)%GOCg*%5~`BVX4jS3w3cQf~(99{siIL$V8ZBeW15sH!8YsVNmdRqma!H&u?v
z_j*zm8jb~OmC@4Oko0#!SMP7;ug!oO!#x?rNSqj$mufc^QRG1ywQgE{`M{pd;`z(f
zwB_JgL95(vmZ2_hrVA?dvpOjiq%%<iP<o8#%QL2JtUpHJMcpR;h~u|b9%Jg{+qT%Q
z!JU*EJ|KB5KduM&L@`jO`|$z<W6-3;rA^D!2A0E{?rsZx%$972c{ZQ*f9)_M2pp40
z-ncxT&Wm(O0`?A=+Edx?gu-b6i*MKm#<mXWY8#}w#3sHO>J{KQI|HYDcTsWbBk`j;
zE^5Q^0*kfIHJCNUg04bh{WrL8h2pcjtNm^9nzb$U>G=v)CoFBlVu*UtPbMqKC%{&`
zp}rH*v5`kQhL#c8%+rB^7c;68&4ccZbzA4LgkmE~I~qPp13WRGuuIq;8`O12I$yHm
zqS3l<<>FXvtCLkj>Uz6qLwZT^yS!@=KI!w{5${hAF-<kKRXq0=vRrSjm^E}Og%9-y
z2M4M!Bpr6O*fd%Vx3px)3<s>|8+u^G6Nui8WaopE3)ZXp3kO=<Bog_y2^injRvV&<
zq5r3(Il5mBrj1LCEp1k9tze5A$>+w%ETm&jaJ#D?81j_IS8Ht@IVVV3at%SrsbNhs
zzOJ)2cRDJasmv4{N*8xy^*x5qt%0L5oJ#olt<goJ;(5My*6YsFGF|-J?_&+xbkSTX
zWJ~;GD%&xf%EjzO!|up4J6v3L=xK$b%r3*0h9sSD9-}~GB5&l<8{H%K$0mO&u4<AZ
z!(oKxQH2)R3tOqUmdIkg<ISVz9xuGPbIi1-NaMY^65Vi{_W-BO(eXXjADQBYVsTas
z+~?0Sm8p?mBlU*;Fm71l39BdC^g><wG$8iQ#W$(Eo#j9G7JC)t$3iy8Oy#__7@d|>
z>Gtry%~)yM3F-ibgDk#?xGTV4KRMm|ig4q=`mF!e?$_GUjMq-yp!<OIdWq^VLS2WZ
z3ZdfKgreR$9rU}=6Y%3^2vuxi;_bz%Ws8~flO_xni*w9HPnr<<LCx%9^9N9dpj<zr
z`UQx$Xn-37X3tnM5*+CmdWfNgL^gsZ&UF2;Zw8iFA~lZw#tIpf+dAEAv!iv)W(d`s
z_bnPGL-rxrpgOEj9?dh*L&(rNpA`Oe^Fp<wTBeW9G6r=nzZn^4_isNUm0KF4_e6-&
zO}USPD{@7RJ0A$aCgn;Njw;yM%IygAEM9>}9$?Ri2mGz9Y`N1}&O1a0ok&}&Tlk>T
z1FMGhe9*GG?G=aZvU|A61l-N-JkWHv_tvWn8KyDoZ)Rlfb;6<<6WMk&r$2^`a@^7Z
zCt_2{mmpGn{>>gMK$!wP9LT)cOg>=+*SCsVm^pg7w@3TzQfriYw%O>@qz+!z8PgA=
zitoi|9~wBQCNneY)o{JXvXVVO>AWR-4KERkaD+}1XU<ufH+4x`96}R<g>HQrA_W}p
z%M54zK4iSTW+e4i+tC$4vH~rkoxk?cB}SiC{+aF?c$)8}FX~RharzRc&hT8h{<WP-
zI@tdKb=fFUNG`C<5{%8e0H*<I$gUBk8skd?gvsGVsS@%b;?e>fsf0J|a1uA(c<Aaa
zn#WkzcrP%q-@AlE%p<~(;WRhrFliF(tW6qlUx8>z16&$tz(}IcRs)sI3wK{nhaacd
zCu_uTy5&Y+*Dp|<*bY_Y*|RNs-NP3Iras<erSO3KMWFh#k)T0%4fgxbbG39^H`azh
zwUeX)aiMrS<Y6N0S=S;efipumyAdl{ap!Zi>%IPjM5)G5mkx-u`9Nm(L+A+(1eGmv
zaYbn{<@85Ku@M)^q|vvywc?~wBJv)3`*zy;z}78K%vlALtJ&0OJF1qQ2GnyUwHTJ^
zm(N@mINjPlq39=Iob+v8x#;UcN)dPV0a1_6hh`Tx$l%nHk%5w;E!=V)%UAn2GiT|N
z@^XYu?>o)g<R6gD_yri5mj)1Wh-!Gx(FMfX!$P^Nlt5mA$wN~Lk6+jnSJXP(y=%cz
z2&`+l36FU|t0=ydDx@*#lD9M4c8UqXc%*O=c}1s!{&VjN;sl3U#dojth|Tl)E9XSH
zjXUPuj$`u8h%)FkW1Jb`8GbNU4IO)M?t@hP>iX&iqH@ik`t$isiL4cVn|GF6xtDev
z{O6B6L78Idr7~CMQC$N$iSXmN^~_Z>3%ieErU2?Vtom01>UxSWA!9NW5q(<KF1x$`
z+SuKlSBd^V61HA6AM?x*-*s<4jCgK<x(cm1P>1ibAxafp$JRcKy=j;_wV{Khm%IGz
zd^Qg==dl~fp4_nuZ7q}YG=O{vMzxS!LhTXoP-<46^N7GbvYOWtjMc}9drT^CV$z|y
zu`1Wz>)_y#8%nhohc(&x+OJu@4hRU;g~nqAVe@4$LF#d`Wg}6E#2tQF?S)&Gk@&dC
z=ef^L5f0h}+!uY0ro40a{TTzgj|Vz<zIlb}2+&q?lS&F@c=INPSJ{}cq9Ja>eY$0h
zf8y<Cm3ZxSdpvpwR50fnv(*=%n*M6SGu^ZAy;rP}R1`-SyqI$Som8r9(8r+}ST-Vk
z^__!j;)%S6Rf{x$D1%zHp5IP4!}A}tT09xZnF9|EcSs$O2t16^1b!88Wx-sWi88Qz
zQKyjxP<kCLcvurE$tyLu6gwks&z4wmHo~t`{&V_$25XnAWm>Ecm>8%7)|+ngm?RN4
zA{x}Fb2w}dG%l>+bkyxdv%@wHXOE7-*-6h2mUNvy|FP(O_IpH&F~|LWG0dxFK>Aq{
zAiH^o$_Bm?$1^~a;HL8wjH4*L+==@4bMMnAqJ%R!Coh8rxJF)~0fyE;dr*?78t57w
zSnUB~1n&7{mvap|wl31GBV*RVn6D>4&SU(T%G;?&Y_YdHjx|-T%VNJ#$}#h=?UFqw
zCOdS*YciUQ+E_ir@wvnIw|MFH#6e2ocxaqz{wEq>KD3h~Mje|}b384ttnT_vu;7uY
z?6aL?uGUuKw;Smr06Yw5Fe_{U4&@#j+{-t74eDPmK7Qjjl&anbhaWnnvtELlSKD2C
z9@Pa-CCn9P#QUtABq+}m&>@azjJV2tHqSqw77%?~pX)2Zd+k6N^MYRBn|M14v1J&k
zLXxHq<o0Ao`!NVPccIv^vvx&pA=S3_#sMQn0T5%==;kLu@6B!kL~#OGf6OyGq#)k!
zSBG=--Z>{3e9x<xiFJbC$ds+4(zM@FoU+K`f*E^3dC|fp0{fkB{n|btqjh7oJNOUv
z*}-tv=*1r25Xo#;;VOc$Wo7H@^Z23Q+$}6oJiC*kwP@W36Gp=E(SBDfaG@=VeiAeQ
zM!u*P|2TAjzh9XlNtf$o)-oOXr&BbHALE81o^UF}GxcR*8DQ|IVo-Ks`SQ}LRqlh!
zt*?}w4oJ)UuYHy1P|<kzigV;U{M>y0ClnuOsLbx}q*%Sb9q8c|sOp}ag71C*=4h`0
z>I?>FhlsV4iqj6Bho#Jk#7-?c9XEN*VZ68t87xi6oAymbj$|qyqI?=1!MI0plXWq~
z7bMZ}8;ksv-y97ca7lw<-=SwFr&vFJM%~oy!xwPHg0sJz{QB%Z=yY12i0V+3oa3j@
z)802jTR*(6t+n^$F;+VN`RbG8Ff@YFMgtJ`&?)f?qb;%vY|Ludgqvh%@4C7-bAA=_
z8p>=roqS$%%ND1`A5>XMMw<(7p{UGvX@I2Rq8H`%D=-XDFoptWk@VrDg^e|-dF-^l
zX|+b~A$#~w#;#tEhC7!>r6LSkx(aLse9C+gyX~as(OUPQhrp4}k9@1NYl(xir`AVi
z^kdxk8X4yK)xN3$Y?0BCuxuuLWf}nq_WV=ibsVBs%PbmOA0O{F>c6f~SQTlqigjtO
z&Wn5|)f5YroDX`3eA@|>hR@ZlqnKgyRj8Wy<O;bp9cfUQknx<Wmo^~IKNs~hH@`N_
zs(+>dH|cJF-t@teO>i&*l+vVa=!(CzxnxGXeXDq?ByUZ_DoZR~E=5efJ1n#BM0HKQ
z#(eS;FCAerfvo?GXwtQp4~{=!Swfdm{M5eAcIQlSv^EJ!WK`Y1YbmjafvA}gr3KcS
zNmt3|CXkAxL*b^r+ec**QomOCJ%6+LDC0=O4S_YnDfVV#BOLP=f8-(OYHV#o*Hf>>
z{8c<2vVe|*3RS*&<~VH|k3>A_y=e5YFM5@}=Fm{4&u%Xz7UQo*Nl+wP<1mX2chLT-
zt$JjWF!}LoW2=gy&d1MrdnqLfu<BTaNS_tgN+{qe)b;><K`QGSN%7~(5;i#(d2=vM
zG~`h0hv(T5a1jC!F#?x79?O4BU#t8}G>e4wO7?$c{?kr$BP>ypP;dm>vJTLMuY3B$
zNuu_#S@US<>uGm1-^k%PB?*$Nr^hv>Z1E%8-&%_ZHz`><j8Pw~Zazlx4Hu6ZCS-k{
zyZCET19ja#@*98mNQ)aHmni-6>F4iBz-HQv3E?}AYZX;EW$!fAkkH~$(K`a)J7<wl
z#gmEgesiDQEk58xqs7PeyN9hj$gQ}+J}gmoV@<f2O9Uw~oNb8vIvi1vJ3U}h8)Gwu
zI~HkEn~8RL*R6g!&!-D%U1C6TM0Hprv=1$*NKV{?CRWxq+|j&JN(|Dx(TSNgzWXw(
zFXC|OC7->AG0gHVsS^`wafh#pi2>#1MRZp=0Z23>fYO0-q2^2TYX?`sz`6)D4pl#S
zSHznoZB4+$Gt2J5!Tym2o}=}XO1I9xd~!rn98TH8VH36$4~L^J;jQBopN(y&IL&Fj
zG%2$W2|lYFXL#ku*$+`cV6&*4MAV|xYEeb_+~TLBx7v9+%07K8KpZqt4RAcyht3*q
z+k~|WiHvulI)vNAmh<E&IiY3;mTOe8xHzBS89RQ>DX$ws4&|KW3nTf3Ku7_XATbTJ
zjB><yoEu>lZ{3xapb_=~*U+Ur_i|=tN~Z0>^GBn0o%g-8ikp<4U*d&qalcV(HzApT
zjlsZeuzfH(ZDFZc^3edt)hsr9y6?JN?>fI*?UtUJZ#j|Bck1#;n-{Tb{;1Sq8ksNi
zSGyfqZ^>l5#ksWPi=1-xjD&f=s@K%;h5QQC)uAqqTBVtp-YNs0*mJskmRDT>CV+0p
zoN22IXGmp5kqnVi#8w~NeZsqu*!mGi1D{)=f$q8fC8ZJF3{NxpQdwN+?x>41l)g{B
zsj+>I*@an?sTHLKjzkVHPIy+HJkfOVt&8QDnwGrNN4TzVo~(>~aiZ}$@8|xs7wFH3
zXpIJ;{Pz5Kp_I`asV3Z9np|?9<kQu~YsNbgLrfKuXBKDAe>cQEuX{Pgtsv)H>6QHY
zEf=rvK*ncxYki0KZpB@0qzULa4-8Uxw&}=w<fEXmYC}54cSI?eLE%|*O8D8e-p5`G
zj_M!Z&-9L5Qef=9zpEL+XQRt6(bcRzF3Rt5zZZ-Hf!4a33LRRHl3O|LIG4CE6F$9c
z@2y%PW?3JNeD7Z8p1X6INzQ%u0lo3z<1JlO=26^+0p58{yn;Aj+hCH7i%OIroSoFM
zoc;U)cyR00c}wVqR~dSl-sV}JR7?nuq3DTsT_@g!i{9yAFlSn5=u{-i`QO!Or`)n7
zHa|0z>Q3QUi0SOQgyj_)Zv4z+F#9bs|J@9UoOR?yGEb-4X4*k=0La=~V*VvW(oN{y
zb+XpN#O1)Fce64|Yvc@WaBch4<x+a*-Z0@#p&x+~LB!4p;h|(+Wkt^G>VdOx9nLYe
z$3&OFO~h#KxjJ9esOCj~rt;Krii<hZ7Xk92msD2SMd-9)w}vy5j<UVo46$==e0^1M
z&AQ+8{3)}IXW>!Hrb18WpYx?_yeo-9jG+@y+(>cU%Z}51-7t~sEG0iq^Kt}+21*^f
z%1vK(q^6m>tVr_FT_96-x0~qf!vroV%vWctpJw;*@u$j6^kv<-&e&4@1`b?q7feq-
z%{ji4zaLLLUiyrz1iD%8-gzd?^a!PV-5I))rli`qc-8ReSnCi{L__q9ZULA0Bx=5R
ze-g5$wzQJZ6xBmz_F)q~r*L$wu&r&gzdiSx@!*vF*tVr6@C6a2bbApq?M7w&K$0tV
zQXzgVu0$`CWT`!xvxPVCUw4OvuB)b`97Hsgd}f~q!kKmqsVu`_1cGQtIv#%bXV`)o
zmw-*jBFFh4AQN79c*3R&YP29Dlx8q!;j_jOtUsb{`V1R>qnkei`!n|mkG;dp`{trt
z!K+0@kTK;<4ggt2f<}FcqNAX-Su_%~wOEG9`o+9y)+3VX#*(ix=`Sw@Fy$~j(t=)p
z62lsE;`>@882XYToG1FWDd8nJ%6(23wbrMV^md4>FIWfjyFE?sWmeYVR4$-@UqI!i
z_I7R&K({vLi4$3DcwO?e#9n>X!V8A@Z+U|}#-4+p>WPIxQH?A1&Nz!mJsPl-M&_03
za)wk&Wy*5sm_oXS<-D|@HsPgNmRHh`+$&-D#r)={?$M1{6x@s`YrUq=PCY|B_FN%*
zAz$50!p;_+MR4!wFnT`T(67l9H@&Li?&1ddh=D_AoR<&@D6b7+>HxH^H1^xD&as-G
z+3Er>eCt<&c{KvwEryTnMo1_3tHPRn`KF6KFwdbJ#2W^<`1smE)^Pt}In@f0X;~AK
z@LnEiQQeFJ0xN(X$S2APbegjD!fM!%W?4YfWHh3}BQErE<Mi?&L%6COs?o^^`>9|@
z6Bq+!aE1o-=Yv5JSBlB5ffJtO9W_07fRaXr`rIg+LHQFK>^W5{j!7KMW{KYVNOU_Q
zeMzOZrr}*toI4Ih6vvWtwowG=80LZHE(^($tcNq}ff8l-OJThfqbV{BH-SME&1D%s
zx2gRg8S`21g+E&1_(KnvSSEWc{uAWqAR96qYy^fn5(dS+*5QHQIgcCv_`RN=U}yRC
zRZ>gZu^qqBi=_oFtHs~*I70Hl5$ib&9I<}kkIyWmtqm*~`q&Z#5t$$4)tJT7=N(SW
zEKV<Thl@#G>2l49>)L~W!9+%FN{p|(YgH97HH$FuqpFAZ=JCpeggdO|&)x#SVIY7J
zDF9*}p}fRg!%W)~BJ+)L8p-!<@DsNq{342ujnrvH2qW+Djp@PH+)^wHJcC}%tBBD6
zZaaYTeORqkyt_#)Eg^*MZ5@Zol^CtZ_K$3|ujnhsr=D}|u(Dv*c}N}y9Y<9f@Kvl$
zg$7iLfYX_gTJ$iQC?U{oYSxRb_Qp|Ps$8V9PoPyB*`DNAHyPYGJm?^MX_@W^FI_u5
zMiVuki`vX^2953{*Fi99VQ0>?4Xz6Q>!RuZu<uN?k^jmnmOo}b7hD|HQ4c>v7;46w
zrBI>a@m<rhHY?1Ebw`h{kBrs6Y6fz>A1Y%zRR*t9GraeAnm~4+f$BL%^tmzu^EQKo
zud69dU(Y*gVpH5_|5Rr)LZahm=VyZ+lz}s6i(e7a7$*?lf?8OKZ5nOScSS@pWITCc
zyTp}de)S1=1Os=(fwl-r4aS3ui#iEqk-!;zefRSwHAkjpxQ^dq01+YpHwM0Rogh~-
ziGK}X(dAl<@8U|r;LOoPsV5px_Rx*^1|2C}2v3c@;hBn_wOFlB&2=d|h0`&5Cbfts
zJZ4$1>_wp?h(x5}Ga{_3sq~(<aah*rODR74{(d|QFaHnsjKS)%<CxquqJ;m`7sM?>
zAKtn<!sYnfpwSpw>cO`04%?$~?);6DaAv|4d*?$`4N!IUVTjf<QG7)Gh1SWtNJ-?f
ze@|Uq&(qfKTemMqlsWO~3oYHq7MZ1t;{ZwwB_8G0%0)O!^w5d@b$OYpfM>|~>|>XF
z!Aeu)<T-Oa;cuI+>@y~SZO)UspZRoQ!l?NY%w~oe<GyZlqJFwfZoA^4i)Jx82Z<Wy
z#5;I_tEK{7)6QA`{$GqGj9RAnCEj&cVpXck+O;#B@8sjm(_-=kwc(_2qG7KHCr$)|
z_EsPILGP&7Wzy7Co7@$$CJJ9nDRN8-n(}8ljtQflAZvh|D!AfB#EGtpJ#NLstam;#
z@56b{OFGD9WHNIJyAGx6x+sIAP<0+7B@s1W(8k|{r?lO5U9z|Ld7(Nvo;kN<im@#?
z=NIQ||M{V125bDg`E9Ybqr^Ti>T{KdDzkCA?hHYCL>4VyNXC8gp;K6s2+*<V#=o-Y
zE+xIM&iLW;6~jzaBoIO6f`@O}-Dz**&F|A;9sefzp{{s3?rC_2yP=lk3l-bJ*YoSL
zdGYTpuQ(MhB8bx1wMqOv!#6l+r`a9r)12tCBKy|0rzw=LW5Tzs!|y%oV`Imy3wt?n
zG6NN0V?v2bc%Ht+>;#PG2G88!t%V=kE00Mg5l`N!lA1M(Qm@87Sxq%a&oU3&FxO>#
zls{o=O}I-*k2*!xZ=_^}u9;lIqY}R6t<OdB{4&k+wpuc?hL5%2m>H@Vg4Ej<?j-G^
zs4PV!7GgHRyLOx^L$*8Ew}j@gmwys=`E3o#%O=~_-B{AX@_Ckl9?95}8MtO!wFPe8
zSs<~K{fFB$71t0X8H2!xmo4~ZjR~uZF@y=~tYy@&-@(4T`?qt%=pI!e9uB`X=PHhp
z1~u^*%ma2Lr;>amw={ageL#^AIowz`cdJ(5npGc3MvQI-l0<eRR^YalpyWdkLgvtV
zi_Bu-i->fCvKfb&_=bCztCVYlrswRl0_dbWvV(6w40UPuBPDfFSs#+T!f%<y>*C|v
z?*}e2>gM@;x^Crp-ib}E%1ZN0XLa&65a@vD%Ae9FyrQ&Fb!?D<lvk>WVEjQ?1-Ru?
zCer`4y1`A%#@6=j#eUx8K4f)Rn9Ik{=1fyyq+yM7r8mE7E@hFT%)c4pcgu9F^sQ#<
z^zz&-*Tyi)*KW7mZ8M_8V45nLKdP51f?KN?F-0w~%^I3DCqVO5<j1lD{YBz5Cz%R6
z1mCLt0{zGMHa=p>{29bTL$V*^+}%YsW4X#}Oz_Pwhu=XLLQc86{<5JzZN2p%ijn#m
zi!;qn1ZV$kEI%7WPT?GP>22lxB-&~~7wNbX%8PERAWOcFN5-7xn8b<q#gM)0=&yZ*
zqOk(v8aOOo9o5!cFcd0j%}-F}I9!V?BhkfTy`}{jIxwHrR_DN=Q%n*S3l*$|YeR7P
z3%q*=&}F4lC799em6gY3wryK_BWWkFX^ZLV9a-$m>^ga1fl#n3wY3d#`rV57t))>z
zjWh-E2k^b{bO-e({`zuTgQv4>72K_4HnXl;`PQf)qmN+8{7Y0LOc$1PSP~pPjb%5?
zK1oq8eqYd_M1yPTL%D5dLAvc^HDU^0WkKq82sVz&H;%Nv0k1Sx9r~%Rf;Hy#oEp<A
zX8vj|B`7Yq*fvf{(uSeKuMjMVZVOmmc*AL(3fH18z7J2oC?WzGZa<chuwV>v{n0K2
z{DAp5@2Vk^Q3R=Q&?z_Fvi`C`=3b&2Y}}&(J3HVA@k1bV?<>jdr%)w^q(-(Lsuh;m
zkvF-Fd;Z4H$H!0DRC{Z{P1%t5a)I3Xs$O}SL?$4Jevc-Kr*&cyX#l&bVp0J40xqg*
zK}L=3>9LAz>(O;N_C<*%PB#JO$DQkrGr`HpJG^uxaF>@k)8@{?tw&TkREG;%0ck7t
zvpzddacl+|e=q@Q%-%bd_=A^HsdB%4J%9scUbm|W_Te$gZ8hiV3&OB_&dgTQRM8*c
zCJL|I;VD-~t0~SOV;6cXr4q6}ebn{PvW*+>>($`^WqFs1b!OZ)Y7-=1B$|dU`XMe|
z@k88t0iVP>y}NQ(Mb%mIBNN?_OkkWsca|vB2<%Z`ZLjAxNEV!RbV?_OnGK#9V-uXM
zKA~RICw}`AQ)99hR-}%QO+GY0z1X~9O^LzMfD4l7PQ__qLL{XL<DJhsNHRmZ;(9AW
zU(Gq&!*k8zd{t)>{nxgC8Gns1c%WzOSFspsdBEH~qxkdp^=!&7%3W=ZMMdVQekcbB
zd{9_17++QWZgNlSlgRWp0hM=Q{k(;amzYn926#hotlw>jC7>`aQqVIUH=;x%4|S@4
z2t88_3CY?E%kZ8#cMou>b{Kf0uXQjR=`L2DHp#_+>JWpuVOd|9qJQpT?tRr|J8E({
zJmc}CUqb!TnM9pk^aoT@evRUq8HqV+Q~-7a6f1_yfmqtbGwWdDAz$$lTTs_BVB&%D
zLy*R<ejFBMExq1Ujd_PTHw6mp63iEoV2<ik!*vnDp0*fDZ;}ad)OF_kQ*y)1h0Fru
z-=?{)q)J1+f(rz)s4h$l4Pe%`#f>gtI$<2}HX$EV<vjvH_OPX7hQDX(DUroX?&;J;
z-RI{e*!L#44M;k#QPCKVM<F^w!)a7w1Lu!SYnr43QJ*HMhw3SXQBv~1(saa#HCMTp
zPphZFkCz0y`nz~0ex8;sFxVuwW=K6e{0Wlz2|ARvdarC{FMf3ZH?D12h$RNwj)rKR
zQQz#lQlIs4S#$eGPf<;ma#2}&&$r`-C)$(m<N{<VP=&k$L2)0#H)=y`_Tx7uZs<9)
z>HH;CY|Zyo(vtH%LRNo7!~#i8y~-qCzOvH8#0Pbfp5?4uxeYywL-g5im^^eMmIdas
z@u`N;jMwPdh;8G+jISbgj$ZhF2~$7%b!=5ZPEAnAOl)3{#n8ON)~3kcbl_nJtP09n
zM9d-igOgkdT>ctg;oyA!LE?kCpX(K+X4k|wXOGvsNuB$V?QZocxS9QOw98E(N;|3#
zTy_QD#`!|3EFRIDmZqaNb}8E}Tz>ud-G>{M>t96zllUJxNgqoT40xstcw99<V7RIX
z?o!P{RiR?0WxnG==EUE`#z>B%*QFTJHuQxJAYMbg22pzCqmw8<un&)C-$R~VXhz31
z=t$st%YVwhlWh5QI%JJo_UMO-wd(5bui+knPaC-oyA(%W4kdY_I`nA3b$clstljKR
zO*rcrqr+a43jC;Tch@b|8x6(`2TuxWi}$Cx(Y^MDENRu@OiRgpr58rqL~Lw-MnOll
z{Yrik2=&~>{8eg?;-)QAPqiC#m+iKH`SJ_BT#tF^=S7esl1LI!V^9vH;d7#wBk`$g
zk)!vV=$P73Hlvo7hAr{t*&Dth66qV41v`)`va%0_tDHwP)|eueWMR=5ZZ!3X{7m?B
zDA%$DT(9?d7H+Ph<`IKZd5vnD=WID|W`L5eNT0#)p<9ls5;X$lj*>!N!Xeg}Dt!bR
z+JvjC#JX%!3i3WictmzQ3o$(J0f{>*OK6H_B^Q7S)Q{>fwncp!EJgjeS59$8zsC?|
znW9^Q@xlb>S{x1N!g4M<NsSA1TOE3Asxaa+bDKv$KuG@x>sl>8P=As|q?Qg!=24_x
z90cvt>$Vp(py=iHB1Q`{pY6ng&j;6krc_LD-_t@k^t*1>-h5snZo3>-J@vU<`;C>-
z#j^R6UyisC(8EH2EEw_fMzcDHugo)b=O2Iv<{;`Y9Lt2Tx0>?WgiHy22&GzfO1gmS
zyBU=^aj8}DA|BI?jTdN;;UhGAlwRl0Q0hpBZ)gXv?T9)Y6R)Hbbvc4Ect|EtCt$Ck
zeW)0eP%-5>xX>o?rqQtZ{$+I`f6l_^4j=M0U2j(__?5_Yy|cW=m+Yy0M~xSXyP7t0
zQ(lXDFp<jCqnMQLyLBnMT=9JoT;5YSYIQSERB*`;0N2$!5G}V#jC2komFk*o)ef6d
zD{cTV)yt<+*}M>N){ji2_-TweXt4%CzQhS)xEpbyr#R}?(09b)i~d5Ip0&zN4NW!a
zkt>_lgb2!Qj5lIBcu6^G2wPw6U<>!twzh!pC9p)kVBSrdS?y1&wyOl(T`C&BUKPHF
z-q=Wyf8L05YWDIvWFprTITqL_l;!D}7*k&M1>6n;hD#?w?-W$;%p$xBwS5A$XYNCQ
zElRDOCCQz|^T9k(3Dav&e<jwfJa$yumTlF!RpyaJ6hC{Ur_cv!bd-6HsL8hrniqzf
zL{0pSh3sxrf;E;0$yG_bH2urCE|n$JPkFiHc<7VNs=97(!RRly?<WypL>aSH;U#Q}
zi1-NDpIp(`J@L=8+0C78BnJGzRg>Zt0jrTJT?H{xi3$C?L$2(N0bHK@V+KP3B|}H7
z<yWpk%B}fZyVY}^_d+a=QZU1n48!0?f*PUH`qPt(`D^wh>HXkpOc?0bcj~~1sHGhL
zwH-|_KNX95eXj__nabX9uVj^?m$IorR%aVe({=2Ty7F-rgEQF+c_~bHqfU_d8z(|<
zS#I(ZHcT<2>&HXqRm-X#&9+{=9=+3dbL)fIg1{OEw5rO){6cWCPyl&G5!7OO@2{CM
z3i6WGb+z?558h9_cDS$hp}P=J{KaC)n4t?g4r%lG(<+_!NQ%TOgcKtB7BpV9)x=;F
zu_O3CsKwkgysD7S%{sC|W$mKOvh2Y9c1)EbnOC55(h1fu^hz7_VM34R?p51=hBvIJ
zGN-)$aPL-TdG#^#T?dg^>!G~kM8SE*LrC@}N?vKb4~pf(v&wkhz>IHa{dmQsZ3S<O
zxZ7JkNRRShxZ^yCjztQPkf;vTmWYb8^UkbhOZkk~w|KrykND|dYB8!7f0{X#^XmJj
zfE#>+`rl5xDByabqXa55pAK}HUm!QT9V_VUy_Z>}tuEAV(rkQ$D@#e+hqp?Gx+Pb9
zQ)p9>D30!fCS<{QTMS4V;aP-RST4g-2kf-{`)`*^XQw_K2`ZNO?j9r;TYabDW(*+L
zJ%!gG^CAc<#GazhM6Ly?j+alFtj8`_%n?5JQa6`5+F!O947=Tp3%|ipgJxogyaDRN
zFeR33LQ=K6P6K$U^2|&4$wc|Ak6O}SY%w*jOkPz>WWG2tY9dw`SzI0qD9L>S$U`{(
zBBpj{qU;CqoLo-GidNd;m4n&YU(a$SxaF_jP5i+%{Y)zj^5WD7lm*|eIM~G^pdv+r
zyC6d9v4^8LcQ1oFb+N(z)=L>jS5u2Q@kjL+y2Ts}GI`&yuH2Pt2AvP0B9)bc6iH?p
zULc0zeHIL`iRkSwo1(4bs-z!^C4I~uxhLHkAqAn1A2k)a*9*8t!7T-#@k2wXy-u1j
zgr6?0c+t8M-y*4RdG_~W;Rhponl5~wGDU(8HiI>;gz%809bVI89dNj4vPEyjzmICy
z14r=ii*%k3-FtyfQge>YZ)hO%qWku&UgE6{v)i4RN1r6*E>GAW)p@7hp0QRTg{XDB
zIPVrA;q_HfC;(Q7jfJtch~o?w>Tly}+Uvi~7~pmqW4QYi#d(};-NgknZ3T;i0O3dT
zbUgq{>lbD-2MncEl`B$?*->+4cfT|fW$+o8`3e*KF(eZ@p9R%v0{UPx5R)eA$4!Gr
zSEQy}FjAiD8vecEE6zi+01Ae|Ibe4%jHp>WbZhh_A(#Lfk^2!uXBu#=9vu9OPfmV?
zlKIURG<wXEc*tTf4sGFz@LG706Umxz{%J6ZfYMqOaehn)$9hsy8p2d+mkQAnY0)?R
zi;9q}#NguVgz8V&Peg=K%;H>#_wzXKSBuuE=UfCGZMfg*iYAMALndCP6YtUhF|I_^
zs$c}TufnNm?<+DC5qM)hpL^Qw+^CuQ)3A)*B-`0(e%A7$OyhaFiD+0plqkkUelVDb
zAuv~iGoU#14{*&EGoJ=yic%me6XWqbzZT7I?+mic%>4`toauiF&$ePwd9`Zm=94g_
z1j#&L#7A1vg}%p-58b0$7%ag`6EWLm*4x9*gHo#;xGPIicQ6PnKXIUk3Ypx7aeYqX
z-nC01WQY2BN#zF#$=3`B_1;Le=GfviK<S6jkw1hxQ?I}@C@H7osp%%p;Pk5_fGiUk
zynu<No{CfU-6pkKUwVoQkZ4Jr^FDtwNd7DNWg_l-hjVf|s>ALpF#{I8LnKQejU!J@
z2%9ts-N+<ZJnJVELGQn=VigW!Ikn<k2dlMiN3leWI5Y2HV&|0PIqJ)9udNr+mDDsR
ztj8?SiAxHXfAP0?YsBsiiKfJBN0cDXQ8KCgD|U_}j>7V+0~g;+N0?NoOV;H$gmN?~
zM|Lm-oJGGBe8}6JUS1@T2hd}3xDLSgw}F^}-dSR3<e@GqZ+5TFfwr6ySB_VEsc8Xy
z%ni~LJs!0=5B1VX8ZQ3Utg&G*Zj|pJQCe!vr_c=da>1-~V36H*#mJ@k&1i)}?8YIE
z>hr&TX^A!@ZpFdGQJiERW~LnKU;`cr+K!l-4@(G+lVc&Si>6*zQ@`pgysl32Xn0JS
zoZ)s3<GV%J+=^mHb(n>iB{dDHPdoM&E}*zy74tlw?t9yi^O{bI1Abih`xY9E-4mtK
zR)sLtQFl>GFcGr@h4@3=0dj{P2R9u(Jt%vuF6E<kua_m<fup#*9M_!x60AF#+M|xC
z>+2$E<#U*FK3&4H?N-CH53w5r^WIUmeZTb3+C>n`<i`2Pvw#cZ=U;*Ld)8X>(5s=*
zAXD?;9r2>%m$ID9qF)Y9?1!XZwyBJ2E0P4lUmCVF8lkl>-+2CWBBAeFra^PkvTjkE
z*zNu^ZH?#;igB>IjkR6UQEzjyATa}1ztLSUBzDj5%nqB$^Q^-U!+O{?CYl%W7<#!m
zzHmJ_WnLJ;m4e{}w|FSd-0GO#=rc=l;=>Hrnlan;B4G>7XHGA8wd5^_yWU@|mx>&q
zt|eE0HWEIjUOMnJ8u}cHGsNtw{Gb7=uoo!lm-t_cu-n>nMWA|`@6Z4ya@gp|N~h4T
zWPgdaq=;{I$uldlg9vd`wmz{FYTf=er=W^31%s&li2VsN+Y%x^zq&g7Rv%Sl45?C{
zf-uis&3#%tVJ%{x06Ng8i`IOP9C2AnT0qSgVQ^+^`h?uL(Z$lvJM1v-kXx=kYQ-PL
z?pfC9*{N<Fjx}DA>kxeXs?_HsYcX(3fP8w3zz4qm?l4>+*f*Pg5vU#-@aZ0vB?l+m
zaRsSIteCGI+CF?dZnpGv-Wx4JFTb)EyaHz8dBt}v^kVv=jfk*zXhZ(mFVZP*0uA7Z
zQb3+)#p}aOzSz%R&k}g;R4qLt^tkMfLBPFars?;yvrUtdOJ;0QSS+|(UJ%YNqKJDk
zcK~<(;mHsItD2wF9%c8x-b!}JKfM`{1MysSj~2JOrVI1MxJ5D%u8dHHmxZ!d<9QTg
zL*>s-e>->IJj01_stDZunSM|3EX!-|M&~{?BDhbCVUz|uNFokW8^gg#--QuO0rs}@
zffgwej4T8C?44$e@qQ|2V+*nzA{q`E^%_6F4{(eaz3)51J|Dms1qB7K6UB=3z&Wpd
zUBn&^?`ywFy<70P3!PBMMp0Ov_^7WoQtRrKrIx&Lq*6HJEqEx)fB!Jh|KOX8Hu@i~
zg#14}Dq9)cD^%G~7;a`e(V|v-6lL|gChwN1rU$p$7yL2z=nEY&f`=K|&jNWqMbDyP
z+Ax>FXbdls1=fv7gC4CY8U@cu+`8hZw`7`lC*|~r`oo3Thn{qakI2oB%1<27`bD<~
z{m)M@{%0o~|H}X0%aD#X_W!+-fGcMIPnGU3djC$I5Zd^^@5f4}MftC@sfi5rSNi?G
zEnKbjHP5K}sLZ^SwDIy|6c~$pOOK_tU7-QGH-Swk(MSzEw#(u`1F+v|fDmj!6r7Cz
zkPM~)sC=@IND}aA|Bwdg5*Iuiq)<`#A}S9DiW*}L`?X60daHgEaHRc3;xAeJr5As_
zBmZ~yg6ldFUgnGiV{VkDZ}us9m#|7x_@}$Q#=U_~B~JRdm(*+d1Tc=}g?1{SZ$T`o
zzj*O!nSi%-{NSGgNV;XDAA0kqT|Ad?0f$Y3ZVK1a05mvlFRTLsF}QRv47Sblh6Z2;
zsj8^XvuJQN@WOX0xEG!{afX+!3Or1B07C=L%I1^E;8rd}vV;%~pfFUTHnwO$ircWR
zF!V1Pf63!7{rKyB`R~~es2)<Mxw&=7&Zj>wLi@XL@wGmHQ#shN2aN4?IqwSCQ=2iH
zcH69gY=^ebtLuce<1H~0d*Rod6`3=u%idqV1e{uN`sJ({zC}IXyg~8+LjqaR%Kz$=
zdH<_J_UfiI#O+_4vVZH4^*xJL|BJ-m{M)}|@u#o<mn{C0#r`DjKd=`%r9;J;euvDm
ztQ0m7hg6jMB?=@mZ{D}7%UIah1>l&SH7;tjtWz|rG#KBZP5k>wC|Z-4PXk2RP<y*(
z6uOn+e>d~Q<$^``i~qN8naoouLgyLoGCZO0OVibq`edC$K1uw5n^}^-ODM%*ZG>ut
z+&kyKpEUEw8(9Z1Du}1x4E2)(&IfS-HzWBvkcPzkdFR+DVI2ixe7QnK2}>r|sUErR
zf#YH2#V6K{JG2!6o^ZZNxZTsdx{|4t)4?j!*si19azBQvQY|@)i|&S=5<k&?a46cF
zp7P)MS)_j|BhvcMelPmi{&{vKGm<N~o;-;*r*jg<>r?f^usiFpZFO`InuAzK1CBK#
z_I^S4gw2_9_kR=$+z$3%yPfR4@TCjjigx!3=1$mNAK8juDPr?eYiqaHz^7^tkMM4@
z%~QUz_mlUI+&o_T!sf@c-&|Gs#rg@Wll|J84YR4z4{E9itSqW@H<RY0gpp=A!=6EK
zw^*kmj%4m<W~_3i@oB~P1`dhv&wTzo>Z0x(?QWWaZ%=KMfngy$nkQ-q#nb{_OtJ51
zlfoal8CFpUh8o~b268H~Nn>kWyU$dFc^vhgRfS0a#duwj0fTT9xIavBL8DJL<^146
z?@LKa?*qPd@yuy&*4{7Wr<BEF4&O-)US<r;)1-jgO=0m(U*?1-wXJHF)vF_d`#cb8
zBi*xI#yqMKPRWJ$ARU4QTxCDdXNZA#`-Sg4ANX3%;#M5)*j1<08@YYXJU5$VRLmyf
zzC-uGJxKXPX;PFDSsiDVteQqgd}(K@Wj}c>^Oli&l4Bb0*?Y|w%dRC$TS+vd>7&o6
zC+%`z{4rC}K1F&Wiv996YhokcR|{*BBifvls1xeHW|Q{%%*NzGeS{;;Od3<yKd!4~
z2|jbd_X{V#6jwNF!Fl$G5YS-2hc8__ieFmHzKgVW#re#8#eSPadc2x4yhWJQC1j~@
znMj;iJ>@v_%&p0y3N47uEPyO)DGG-hUdEe(>pOC4w|r7iCW%r_)y5Drh9H-hr&F(T
z4<zrU*)7X%!U%_lkk$+N(>E5g;+LN4UT+hvt&&BB94+Sgtk&(jdbC}>e_3~rl@jpZ
z3a4`+iQrsJYy~xWgKFY*+X;o6q#8_s2Oc)aS*`TZth)n!`I~B3)Qlp9w{F`H`E_Y<
zS7-yXmQ78v1HSAZI{z=<XnCA-Z+=`pH_{T^P1$*~Z+d5t-Dvo_IvNq^k5e|fhS(V{
z2smma$;SC)K6wW`926f%rdxQb->ak8>ZoNnvb5#z<#B`g4Xe&o_V>o3Ktbjupnx+g
z>6^A7_)BhNU0iSek|N9Jx^Eff$Fin9lWf=%4p(j{co+idfe!@_i^p(;?Em&2vcEj)
zQ+`eT{e=NEuS57NwUUz6+(3i6H-}n3UpTsydOXoFCQkSX#~o4reqOrmS-y6%Hu32^
zIw^`fzR!tuOlRuZd-uy)y*NMbR0;O#@{HP@GiO#h>agH1cwEC#tEVLg@AbUT&`h=U
z^HcB_)oVHQtxwQe1fR`Ix6{h^j;PWn!b414Afb4PQkHjYMk?FB;M5l~#=+lkI=TJ|
z!nxQoYC$QXGI>wtNssQJczZlxcO1pX`Mv!Q^=F<xkr3OLn3HhIrXi1|R~j<wjeevJ
zA-PcpVLV6!fxAe}#pv9J1G{1P*$%a`#%lo#h!ZEW+<sk6?<){Qm;L}MVLe#x7N&W&
z@XD6JMR}NR5_0$a-m)y}B-%XYJY&MP!=%GnmEdc8;0b5xPv;!)1iGi6+u~%ZwlhAS
zP>Cb|5~};~zu0^4fF|~BZFmqBMI{yx8^n%+hz$!Zs2dRxumK7}R8$1SsE8nK5DN+-
zHbk)m3!)-K1w?5P>ADr9ML`Ko>5v2nX)}B)SP12wd(XM|oO8eX?mxCWnGBQ6%=)eM
ztY<xIP3z?c^`SI8c+^XqG{3BGHu4LC6t&ls%v?Ohmq7O-vehbEGbVhy{JMNrVLU{c
zp&|QY?%t{%Rp@bV{Ce(}AxDe(M4_G699C%4dzF)VIwHb3_sv!b#hVJOt&dV-zo;3^
zEIYURg5ntYi)$1@_!DdJ*sa@dluh=`3c9m6>ykX0e8X4U?B$11h}AzpI!iZ@322Tq
zEJ4~nv|Y-oW5Afn__xGNQ9ADsQM9<W`EwQ7zZnm%%nPvij%RomDb7r8klFcS<)N{`
z+|YV*WC*`Qh=M1HV^L4}X!G_gjP_byJoVMQ<)pn=w+KJgoVso&7;4w3KZVh3Og&G8
z-fz<9tWW$>(`z8P@F+cRxnsx+(w))wRH=dkYVG%taMg3-<|QrCE~m-CXK!nRB6Fi7
zONUYww6YzWXWUu+{AARk+6xEoPT(y6)0uql?~G;rW1mzbjHLWNw+q=%;(By6aEPbK
zZ7?e31yQK$(Sayb(&b4t%Rq=}TMZ*c9XnO~vPs9HCBpN6b?PDUI{C$|pzJpq=g1>b
zM<wN@V;nji9ype7{Z`4pCZ~0Mszag5M*|O-x5Zw<^Q_Pi%_yG~+N?#*n0u#iUWST`
z5oM|J-u2$!j%1+Re9I*KBWdL3Dky?d<=qwR2*=zU3qIS&e|y!oqxMDZ258_s5+-_y
zUDs51;a=jsVkH={r;JrvT#*%!cW_pXZqc|aFJ|aY9#l5<9CcQmbVpzXEj@~Sk|~?K
zm?ii$zghYH*rT?=hr5vEmxr1PViyP)4>uYh3j7gl!^)Ciwizy~VP8ocyXwFZhk4F1
zl$|5Z_nt5L^Ur6gTKCpHoZvQ*#&|5sCvq;YhJINMEH<2Af<0`8(ykVsm%5PQGj88i
zJ%wVORq9JhLd_D4cfB!Ocr|0^nvPQurRf*4-K_Xj?m=FMaH1EHXBV+A@mgE5q3ixj
zj-89@51Y*lT{JRR@!E#(%X8<1d%c`kbII&0DL(Q|vt?E1Yjh0HSbydGm_UW*o4a@C
zYDS&a0Vm-2$H>64d5bf>DuqjZ3Q2=z@$B#OR##oQsyv=;f2#SiX^Q{mgstlzeDsRH
zp#5OP^M~b;a>N-E(xwu1c#Er;Z)`NUg{)M%Q1{OgpDFCe4{kl#=6$O<TxgR1{KS;?
z@^du4EiYX#VcFj4DQa8>YY7R42RV{X>J3i1SQTHggEN1U9uZpS_wBvm(6oN1uV$m3
z6n3YEmm>TlyMpfLGqIpGS^C}|GmAQMpV-VYQam>5K;=DIr5E>iUU)X`rk7u`8@HMr
zQk%{!wXf6&YI0V_O~x~u)hVNAeLFjU>)XKd-*sOre>H>gCyF$$kv#J%f<H&J=oRbh
z6|E8aJD}Wcu)yo(w(0x#?A|0rxLLFN@_Vo2P<^juFO3DAxLc5(LgTpj9M;k&yAt-|
z!(!GSzqj%8A5W8xH+)YaXLKP68ciMd++`^f9F%7Tny2g=QRZ+lDlk_?jgvWZ_g+2I
zi8o}29yFU+P2WL0Z99wQmTqI8ZM!JJQ`b5>lt~$#6QSh1fjIZNb;$g@8NR6RqgH+5
zu>$7}nXf0GL2mPuNVnAGrTBM*%g|i57m&3vlrAJ<jAgyI&pw~`OpX_M3fg!>Jb%et
zK>quui^4ip=J$dZ$z4b#O%xd2h1f9p>K;Q7g|^UOQjHVL<#<6B%yp_Om4qL0QaOJ&
zb~b?X@n^3g@$A9f-FQPDcNd1;usSn7p|P$*iK-(`f0n-d2>}xO`q>6I;$|Dm=H{3N
zTQxH&{Da<v%af?Vu4`?pDcb7O4nE$F9?U7<i`Ji|nPRU}CoIcAeYH})o!k>o8UD0H
zeI16qsNgpYtG#CYR60tP@g99{8X4tNK#CiGPq@-OZ0KnB)qgxKh?eFI5xo$q?ABeF
z)I92Z__|jxp3F?NBE0^#0t>%h88k^_cDm7v(*<8iN%F(S$W$QFVW5+sPbTf8oKPOd
z3#z7{)tk%aUL2Y<^sJY9Q#5Ui{flh}c85H;@TMYa`PORWg1gGOir~*$qI`gsF5b*Q
zB3H=_i_#xYqImCpSJB$k7U8v+=jtqX`~fcji?%H<)`_!Tpj>rCpMFH=Nz$`oV~y2Y
z>2fHrSD(x(vyV542oV|c)vJ%&C0_DY`gV!4eiX;_{lxj(+fKZ$LlD^xW?5q;=}hV@
zugNFOSwTr(8_LI~*!;0_tYVn8chHo(np3WBe{!v5^zqBb4bw@bh1^|ie?Mo<ym96A
zdXr10pYmbaxedym>GDu{M_>)sFz&5<?u>^rQxUIBp^DcU*6dKBjeGDL;?M)TiG*Z(
z$IkKGOIhqQ2b!zKjSieGdusX~+p=L(nkOW%A}fifaTTAhsJtsl(Q5X+2Rz#r>$vz3
zw{-QQ^N)sKG#mMF+0#u6)=sRRiYW@UMJ^0znA^@)&pzQbH*%|JQGW5g_~Nklm8rL@
zW$!3-p5${16gZr-ae5O^c|EGg3o6w}(<pPRI)A5%JyUN=NzAv{q@=Z_2631FwD8+D
z$nWUbiDnb|&$Y(r6Sw(%q08tHtI*>fjKJvjPwJYj!?QxKx*c5e?CrCxhM8xc`>vai
zJN3v&DJ8Wsm|aMo(m)=oC+)70rS<1&?^T=(!$WF^3*I|6AK)xYirW+XigvLJaiWyE
zz1CaKb6%FlJQ3!{dzXPPao=AZXlilIV2tb0cT#>WLto#Ux%=vC=!s^jRW{LOQ%AdR
zuEYw}zB-Jb6FNlBvU$<Cc|tSSHl?TDAn|{;J@~wrTP5jq`Ggd9aFFYS_7v4LH=}v}
zk4Ua#4CliT#0^HeLylOT!fuD`bb0K!=#2Wjs$&l>rZyOYDqQdNASV{xoY%;rbAx%s
zE>&iwiWzNnR5_9EG(7v`bianxN0ZK2Tq(PEa^BGTtrxzWTqVC7@oe=!LY|11lg1xu
zJ(D`;X{Z5@R5|feeM#&{y*tyN&XZd-;njv&JJ#I|Rv+&Co*dP|-g{km7|%(Zf6Mv4
zgYxY07E2F3*^nj|;yvY1)(P2%Q3nus3<T}@5@DmwrZ;7qpP`G&TV{sMPMdtmU}U7s
z;N8lnebxVgq+6;(!=GLw1>$~297%vgd*rY>MZQaz0<J%l;}vM#&W||IQ1pKRpCZb}
zpee#JS~@FC<qW#F_4K4Y_@ip?UHtaBc1{PMW!c@>U1nsL{Bo20ao_0ViwQFARnMA}
zb;GwcY;LshaE_aFiXkYwR%sAs)>(rr_L=RwoO*&JHxD1{HJ`;yf43+a3*L9}&&Jo6
zzO~Kd5bQ&)`#+<Z{v(NApTFs7m4tT32ybZo0JMyN-A?iFmF)xTJVV?mnFRi*GL-QW
zYJ;V|Iwk*rR*Y~}kA#;n<>>Zhg_T{%*fQI%jiKWQ2|f&)K6LJh<coLXB7DtmFHZ8)
zpX%#jNf644=9MH*W>s(Iov5C?*|~~*#xnZRml-No>-QT^kdI`w%;!jbT0VG?T9aoh
zVc(>P=G2Hs^)DE1N707aVe*&O(!U%XOsg>1zd`1#*JkW4PViDF+8;+fYsD^@A3q_?
zeYDl~FO(o_hZTNvk=kcd8CTJC51+Txb}bl%s7b^PX-_CiqBS<`@f0pUux+T+Rn5aP
z;a^690_3DqI<;Lk4(g?2x{y<d7vQK-^+HuNzk$h{SA_DFQ~f_KA^G2)p7N9zeXou?
zuE8qe{-&}9rrjSYp4;vXk-3>OTXsY6m5ayIl%hQGF*G?eU$qby8R81hrZ|k9w{gxN
z`hy-$K;(R-N_~m8JFA#})L~x7yj;l>c*~56TW=g_yb-uO>}cr%Y2yS>Wat7@wZgja
zk%Uq1o70s?78L%$I)B;WZaw?bJBs}v>x>B<+IGn9PF)ipH=jZ($AdZuEf>oQt?27j
z@4pzoOmWQDJj#^I>76ro4=Kb7))-dmE87uXQ~a*tQ<dkQc%3-U<8i{Z`QzUpZ`Rkk
zOC2QYv}Cn<bciD%Sb?0q3Cr@t5$D6g41wN>opw(eZrYTmImuftLoT46iypewNVkp4
zi3$71WKGgcJ*YM3Br-d`LZb^K+>$ENgn8Jm^u{EZMMl#PAC!8ST`LN5V{+ehAqMK+
zllxSGR{SGGsooeYlC+;RRJ7bblQjHDpmt-cQp(fV7pI)lSAJ){j+VcXtNrHe%2N5O
zI};EW+i2ko5s_MvkTyy&-Ri#bngldyV^L`0moL?w&aLu0XXrk^e?Goln&+v1*}}%h
zWa;PlB{qI_Z%!Xf^;>@+`-RgkU%g#yx8{q&6+Fd^#Nd-Uo2z2?L^*7_Hh-sZNc)YM
zw)eaaE+RghX}O}Gp*_B#pC7Zwp=C*<!ycEb_NOME?U;EkEKt&><fYo^r%BF)&k0$2
z(+j=kRx-!mNZD-p5++&^n}l_u+=v|ylWq?&&(+=Ps~y(Klyg^0OO+FnMbouLo%ZMQ
zb;_tuOEaBojq>tTu0*B0(s^N~>7zlYTgti5a(MAE=cK}(Q__R9M(gKjf+s#Up3ah~
z2o7*vQzBYfVKQET8!j@^xw?RVDaqAg!kqbL!fiC0yS|H$1WjJc!fy=?4*!();FDEM
zj&t+9tviFf6-E!)*fhjvpVy#|?ZKmcKGT)sygjl;t$Wk3V!V^*wX6GfljAZD)>TA}
z<W8q=aJ+9Ze!+Glw;{=g@|H0#ZjhmLUK>fx@?EYWBU0v>vv7fpS(@J1!X?JJ&nJ0W
zZ><V-a@|mIU<95s)zk7m<ZTuy&tr?8)E351n2{Kl7<3l~=;$0?Q|&Y(YWJe2Ge@r(
zR>25C)AdAyp`_?kNfnLThO!@bG|Z9CK`W^~gV2wftUP@96FRTh<9$l<_zQ+kPLt9{
zs*V|hn453(eGGH0UJr`+$Jlzkh~80dn=-VsTo?+7-xk1R-Jtlo?&%r>4Wb8jHe_UB
z-G-TC;?B+AQn^xPGO-{<_7zP5&Cw$X8c6Nq!wI)YvGD{xHx%blSoeh5(8We|KwD3D
zA+qgn=Scqv6#C!z_`^8E;)qQjKbuy#PG35MzlHp{)L$et?n1nl@htOwCc_VP&J$*X
zW<()lgUoe6^p3=~W(|zRtTf~v1r}Zpy2qHKaE1CS|KyKhV9A~aPo2EyCQonLzuRwL
zxEx?a&#$rl0J3on!!Cp@13&i3I;4%%)Vvz^27W(vvS<$ep@>vlN)|%9aLS^9f6j+j
zT*C>=F0U=VlXms(kvx-+Pw^1NUe~Cd7)UAtWSY#(u^gkXY>(Y~h~~x7JJZoqM9J&^
z05ZD&nGfd_x{lQ!|NIjAiNc~8M~$bmI|#x?7@hmhdVSHw0Nwj<(;qH;N_u-k;-`2H
zg)lhHIJlD@u_ymuc*{2#-fwt*X?pDS9K2~i8sTB!Y72x_@FV&cJWi_8<Uxb#y7(vI
z=ZDSt;&X*6dAXi6kU3)#A*}|v)165X|5MlQZ{Cs>M^a2et~<|GO<5cCR4>^w!lE4H
z{!`x@V^GAo9!{7bVQ|3`hNxBujj@aW`5#7T={=dzW^$#Hz&_?2oTpINQC5!T(1^lE
zqD;h#FLAs7=4tl*P*G2@?29#&src+Ap2;^4Y+4+I$2Tfu@r-d>m}El5-`j*B3RTpx
z(FY<+oWq9M>Z`wg`)G!CSwTw7OAQ4yyPe$eCtA10S7&iAPbJ~PXzTC#>g4^+PpgGQ
zi!prqE~<TJ@{8=+fWTgR3*^lB3Nk|Ia#<ZS{XaX@-~OWi&ZnQ_QcMsHhea(DCkCB1
zd{}i|;+lJ|d4@-t0YUEiMC+g9i~6{@x$ho0nx>5{*lU+R>=X5yZHQHvhUk=f7gCO2
zk*e9`V0GW*h4qRHf(f69-_}c5X9_)&$Qer(e5e1(Sn;n}SW8k&*L+sG|7Z5optHDn
zy?H_<lRsj0I|Un@FLMs<ark}0Ph|8g<QP@b-O5YsY<u0{fACWU)qR4v09)<3pA9$7
z*hLEQ&^?NaL%J?Qm<K}f0EFMx%Od|i>;CT;UGKkj<JvwLz1Delsi_pFY@(nlMfJru
z_?d%=9iQnTh<o<G=4$;%_W?RL#lT@+*qUfLowB)wy35w2^8?!N3mZEWKD#HhSbLKN
z8)WjB60Uxu#2WJ0BQNzG{Tm+^d1;;As-u^?Xo+*}tMowxgN_)^E9MADLY4skum)e1
zsEVl7(x+@IH?F$zdAgtB=5G~`FA-W8FBRHn9g93es?h+hRtnN&1H_&2-?&y8aE7CY
zTF=c_z8>V?{*c~-$7taq$U>Dt!`n*yw*7z7ME|3U{sXPa-!m;5{yP53WgMV~hm8pZ
zPbGEww}lflyccf@`VU~i_*Vc6tb-rotDf)s&tT<$f;#`6j~~q_a5Xah^yK^D6>qY#
z1(x?MH&&YQuuvR+4&bB-rF-w5GmqXg*mdaAJkEVd-$mI1#tvN@M<}&h{!<|Q_xz;)
z1Z79g4}J7>ueH$`+xu(2c)QcqMW-yrR|<kL7jmt>@<It4?i*hJhfv?&JnXt7-dhd7
z-<bOb^?bMA$DL*vQx4Oc%)1a*Qu{Q0qMn4;?Binp*$BT^c46wQv3Wfu?z-E^b6M##
z7!e-k&W)tQ>3GV$e-F6QG+DgqqTeitOW|Xlk=};1M{k;4rhYTWd*prshqfc1UZYOw
zAPI&?+0F*5ql%=^Xv97AANLv_NBy3JjlX$b_IcT=VLGE%$(B6!a1)(qMjJ1ob<<<I
zr(F&>^)bdNlguFm<B7r8kf;dBcMqj~f0py56Ie{qi34akNiZyCdJ;McoZ`}23`)V_
z0ag9Gte{L>V;hX867UmSL%><}%CjVODsv63)@>~D)uxO`A0>1l$LJTtN0)fhTkZA?
zBr-(ES5xU9qKX5_Jb^#ffgh1_v4SfYku78vJa63G^SV1r{@qa*6ZMA5?pP+Imb|iR
zNQOE$aWnb~+b2C@6Z~-y19L^OVa2LRkI<Yojd&s*Kj|2P$OBLC>pz8rNHk|L?0l5;
z-D&KL_E!E-M{-NH=smKlZ#B%4atx$A$!JStn?i=f8+oaWpSwzZc9#Iahf>FOA)$1k
zJ*3{2d-R7Kw1v4S!FBXT&<5lOxy7fsN8uzWn8nz2A*+^1{rT%4Wx4P*FzV+BTIUMi
zqaq48fqjIOhB2mJQE+ltw%`^CPbqoY=(4p7*&GQGEwM!D_t}ZhQsEJNN6scD)>^;K
zwdb|IqG%&H@>%fxPNMfNAs>17@KuoTZ)fnsYJ1=NXmR6tjur79S~vTd8YP|18Sh`m
z6bug^xw+)hm5%$pl~GA3$Uc-Cv08D&Jfu*@^ThQ#$*m;ZZ;tetzB-cVS7?ot40e?A
zZEHw^hRAm+{G9iqhtgyF(9trHO!C8+cbW9VWeqUt;bl>8bfEBa&a2EFn82_y5>H&l
zQTS-7U5WCCQU!I8W*Y#J)hm>H*z|sn*!{lzyhvOZOud}+_GLhG@4)=aYPzqhCSAaT
z-H0FVJc@?vo{h7>Xz#?JITqCoqzg~NN*JUXwL=}m91D})H>AjYGfvtAXNneCYS<W_
zB2{zAf?CV~v2&zy#BDn<L869J<YH2dxc|^*GQU{1mkdmevL1}v2QnMDStYZa>!}<<
z5!wjMh!&F3_X<n8_3Z{nJBb`snh**ccVv+&$u%o}twz}<yh`k7R@!Az9@{yO)L7d8
zfYRhu@a6IJBA~2hN4wv^SD~ySc)aC6TjQkieqGhZNfKVmTvr=k$)$JXlG?{ucH5ft
zK<(O+PU~Og!gARf0Xl?=z{%>Xvz2E(Wr?iLUUYun^p0?CR~9Sc%!fsbI~1}&+{xk%
z2YOm<Wf$@qGX1R7<xmd3467uRRfea!ntdDPV+8Yv9lDSa=@lxNd^E68(y-2<vDDp5
z%sqy{0<YMQR+B&2$V*|mX?oWmI$YH%2}I>0A{!PCk8YXj%+8%y_cq;+@Gw#?{u>>t
z=r!Hk`KL*QZ?^a0A#2W6c^yjuGJ>sxn~-?ZdjztDI$GU8>}2qZJtQ2h?|dgRNB(6O
z;wiAOag=!7+f<9eRP}&ha}hAqY%IZ28z`NmXk@2jFRI1TRf9k5E6T7GJGM=|-GPKl
z1*?)IM@FveD|>EkCDm94V@3EbSb4u+`z{aoxudqmaP%D$i-j3wQO{$Ry&-ctLD&Qj
zesXMjh{TM@B(pI^%!Mcz6T|JrGOYN0fMcYZEyR{!Ob))F2I>wQ&_q(Z)&Q|i(w!b~
zA~qa>?%+<V^xo$dzl6+}1sWs;Q=8xOdd?ttoFS=ZCcWc5dOhZFB1s<2G3s8Pi?@I4
zeR+G5T3E(BH+)q-22|_HB~k$%1)FZiUFoG{wxodQT>2{tvW_3&TEfpFha_E7w|nYw
zr%_`i^Syp?tzFWWT27*R``k^^7>~W9L_tjnse1Uc&e*QeD`%?p(g32Xu-iY$);&aM
zOV~1^Q!S(lz!EfL$#>x+9e>?FRK5$zj;0H<`IcjB=mBVT2@|UpMV;w=SswsqAQOe@
z{36ZYKGt`MW*%Smi?<<1;j&d}t?1MNy?cDw+}?kf7%YosH_<!t1OeX~AF?7vKJUOs
zb|DYLN5TUwS6PxeKGJV(h!RAKPR5{(_fV}ny(6lmC(V#V>l3a>K!)zCTlNLbc}#DK
z#80Z7k#w>8Qw;hF=6(UzC+o^ee7^Z(o`(AUj^|WI=8R7eP_e<qv~g&T8II6P9uKh1
zcSXa`YpH8z3Vzs|h8;G;2AjW)1Q?^oz+b80eyTr=3>|2=!N0?$idg-9qh%vSBu~Zv
zpr2-Tv^_@WIC0kfXrj1F_eD7r>L|2nlnjTqNy+K0-U#wOE8=V7=?zY$<<tnH`KO$w
ztA8yzsA#EQ@$m3s-}L2mPSDJHGV(RGefd6=x9|ikzi@&n+g@-3?YzO4ty2B|s=5pD
z_qcLMogUb}W5Xpk$I)_h*5d88qFH+2jPHn_z?gQ9C^AQ!K3b?Ga0Z(`&(?9tdVVjX
z?Bm7M`jN}0o?of(qsv29eEB8(=wE<UUg|Hv`hOX)TBtkUPkw2<?~EL_xUj@mjUmCi
zY@d=kll8|UoyG1+2Itr2T38fK8?v^-Z^S_Zr@GrKc%G5lbz*sR1J&F(=c1`PHN;j~
z<ke#|2YS&Hm$f}sSl5|}OTA1Aa~JKr>wNLu$c6>>upCo1*Lwt??s)fX=Uw+e<Dvrr
zS4WrG8Z0?ECwJP0%g4)oXG=m{2`>#9!~-;k1X&GaG<E$_h(Sr7!F&Zv41jC?DacEH
zJMe+hk`()vttcbT^{ll=-j}Ch348woj`ALXuPTA83)1DV_$mq?5aDY@L9xw%oAYIK
ze<dkT9GDye6xU*m#EJH+Es%7|ltKnEQ=Bnv?+$5%0VkUL8&dke$9d~jVbN&Kbsr$V
zxpM#+hp#5fF}{%FX`;&-pW(q}nZro!!<;HhyOS9_R0L3GW@8$OixiUDZ#f-(n%rkY
z4@>Zu!~KgP`2ROU(7dMcT4-^RPn;AAfb>Fib`d&lo91_%gxy8$yO5#o^i8b1K#_y;
z^(R{s)ia;PCSJ(;GoXTUe)i^&sVp-q05FwLNPFmlIfu}SR7iWy<;x^*v?YLmXs1xU
z3;DP@5ay$H(nWjYMI@n-&j&qK2IMRw>y&lQw{9*zFg^ao$*o@>U2?cKeSYnkIqya<
zTC4oPSNC+l?eE0cFYhR~%B>ah)>l-8hYyO6Bk^U+Ff)!OAZ?=PlrtX5pM>ChM6O-P
zjD$0!rWtN9G&)SoxAonEs5R&DRPsbOuik)e2n=AAfn|m(w^lT+7B@9Efq~OLpcT3M
zP@3x}N$|XA&*RH9kRV^W1vw({zf&<o{1vS&e1H_Og;Zls?gXE^N}TmVWoG*0E+mmE
z7#%J*UGkXX*DpG#uXcjG<|AZRzgXjbhtzD>;E%Az%y2(P##wreN*98V8_kaluu*?X
z$4_jp{7e#VZ!V<Q5S<{?xGdJk9R2}h2kHj?XQS^7uN}OP70lyQ0^E9eUg=wb;6vxe
z@%zF&G78RG&7C>q<nuL4o6nt>b2+c}Esy7pHgy55u&VPc+FA^4$}{t2%C=Z$&{+W3
z#s2zKD-f8>C=zaT4_77%Nf4e=s>s3)u-DurQkiHP#IG7-9&lx3tw_#k`pl?yCbtrL
z?f`&;N$ZUTdhAU}=h86SM=nqJ3vCB)W8_?%=3KSZ_1lGur*-V;3jB*dq`rrIO6Z*R
zbU3h3eky#;G@8Fti$*B}Hwx|kiMvh`1%owN&E45EYK)Hp?VxJ~RfsrH1*Y*y<VEha
z-1|IBmZBMp?5DFc!by=qsZ3egu<V(!+qqRaALrQ3`}m?M%Y{=t*k!wnNQcBl0rq=l
z&^sF-SY6sD-TYXsiroK#0PRX3U~p;bY}p!^AQsLy3=uj8d@NSK6PLVi$A>@XPEO=3
z-d2Q||KhD(ZJz*Fon1s~1^XQl#t@yFPb$YlLalmU_D(u`<Sp@U8Rk&a65PP@8ucew
zg^Erdpf^Urm#=wb{?!;?wT#?R41;}UY<c*b{uV=_=oDbn!a{}5x{$?yfqMAteFI^j
zJiY?>5?9hSpyWM&zv)>#a_=N*m8c+As3--RJ4jeq64Xik*p?#H08Bj_Kk1YrNx6i}
z?fSX=<jT!@gcnH?j7u*PQ_MH=iQE_1W04C>I>UAX7~QEQ-ZM6x*i=x`U|6&>%+hk~
zqMhqq)(onEhjF)(XQAo2sSCk4qAP<%gA?IaJZ3;*83LpmP-&=8ihwD1@MHCf;9*!C
zcrr=5n!6noZs*has=>Tt%d$uUg=F}9-%zqBgnSzw8q$FzL_exew0Ov#xrn^Y`ioZ9
z((q#&7QU`McYBeEAz91I9D9t9^O7xYry{&qqY?^bTEYGY+vZ*RTE)4(?&0ih_net?
z^WFut5&Bz>cqxUC)mLh)GVA|g@z-Uz>=2gXrXuGfbvmRfgR2c&&W<o}h+F)0gXw&u
zEVCt2ora<rr8H3+v9kb1GT9a`u~}7$PyDVwM$YI9ffDT_>+_Y)an)NYaG(ngyr5m&
zCoOo&I1*LX_)+uuX?<V&gH!yxP3+J&x{uL3IB9PU9)t7ImI}|~G$#3?O*;zc?J-&i
ze(u3(<6~A6*C@#i>wFkxXbGX|R-W*CY>$QgMgIS)%1bhA3VZW#y$ZOl=q-z=r6{{o
zrdv+OauYw|`tofy6Z5i0>t||mzIhu8fd$UZb8p2{)#>e*Ch;Laa4DY)apTr>s&K*)
zIyU!w^4=60_kh7=t}pLMbiR||kjSEP7`{acBmCC1iQaFdUGX|5JR%yyvJZ-H9o9HW
z@yu-cM@r1RTKDPk3zs}yyYcP#mlmaKm*>bR85~CYd+~6>izsV|$rqx<?EMU}LNYxa
z#FM}$Dkd9uAwEyK5dQ3xWfiT>*IAxb2^m4e*Lo8^#=o;M+p?IysN8V>Q;o}mFE4GI
z((>i;=mn8Ck3x5qTt5)-Uxfvo8lrI@cbJtDni*1Oo_~xV?rR$<CEvld<K@dVr6{m9
z`uk~Mar|-c*zFqbp*Xe;N{}Xqcf5i40R#Q~G?(H>j;nn`4^Ev{BvP+v*4X+irSJ&f
zb*N{1!rF6(k`Gl(cyOWR>r@@z1%xG1En<dGK_Y={gks=tR5DO8@K6u*5WCXg-?`;L
zT<jC<9YMf757o1u6_$QSl%m~1bF7#RX!|&eWI#&ONIc+?t)d3piq!sMWjlf-N-3(h
z#r(REEU2LhhDCL=$5LGbbXflra7Y!(n-eU$(Rmt%Oz#ftLX74|u@BH4NELTU97;JP
zcF!(pU^5eGnTg#|K&V{>kIBq_O%d?fDtM@-mmS&O%Nmx>avsEXvL2+u1>b4s_H`?w
z2SNhgnL-G&xr@ZXxoN`*Xd616s`MjuPUu1owDO4}Fz;O!7XhlySF4g0qTCkX`yoR|
zxw9OIf<;xSQ-AE<uvrYqk9Q%5pc=Kc;~5n2^t+Ika0Q=uAgY-(k@sp`0HrnB??Ouc
zkoQjbs!(Dpy6#=H$?unaxM&)1(M8M_)asklW1ynFP4L=|@R910ar|*h1l;J!Gy&8!
z1!Ep>=|%VVDwY6u3?OsjBQ%NFwnXY|-^W_oq|}N6*kT;~ClKpBZ#Kw``9(VHsnp<m
zLgObY)I4h5N2uAW60KRz7*a*!;eq(8U~E{v%)(zskgP=8BvU!6gib@;ALL4sj#2#D
zXA97&>mWd*EE4AHj|Y2*KGO;)yt!miFb;R+>1fDTTFEz-h|gE56P_<T0yp-7jA0iS
z{)tzsd@Brw_z3+VbXW+_JmH_x5Y{(DAs_UmYoaIPbZ2_|E{3iR;UvAr@gu#FTh@iF
zQAfNN3y*?L6!FUHp=4E`Sej};i=iK2^0lawy_YbtW!9ECafsz(uMubR*a?+QQ!8?-
zlxDefB%QGgVMKeua=X;Xs>QQRi11*h>EQ%s@Hmk*Z}B~MKPZGSbiD7@Gq}&6n^APX
zILtHEih8tSu!n`BXnw0NwS7}NK{P5@<iwuAjRY$W5I@aerzaJYO&;n?x<}^OFr`R!
z5p_h~WFoKRg+JS{qV;UNr;SW{x!%Wkr8OO*m+BQshF1+6<F`z?oQW<CY4>kBH8K`2
zxGl9d)MeD_L(;-W<6?=g2kn#Ye6#h>P04@M+?hdIcm74o5pScGJ+2A6o;0njzN0C_
zo&N~L_|1Y1Lv17qN<G^xUYvbEE_2*yqBCps&|JNWe3^OwFPEC0K%SHRg$P*vadA&b
zM||Y)Qk3ITt_{pzAO)LwI(0kkVHe^|;;(2`0;VoA0l&1GJRc?!eHCh=xoKWnVAWi3
zMfXaQ*Q6wI9l-|tc#s|MY$icRbTnd#-G%j#Dp~{f><`In!F9DPr_c0#biM|C483V5
z#N+k{>)2t4S|jO`J#I>U4XDGBD4@KeUGBFDdOiM2LAQm(Q)v~2Dy_ydF^G^3jPwv#
zY~wm{<MV)#cj4fC;_9R;--=5b*cial`_bGTY)dR~8ChWNM7~ulm;uf6P~N_jRz(wz
zKR|DP3{Ty#s_Rw?tOuw0e@udq-+va#Uy<^!NGaLU@c&k%>=zShvuIX4abRXt=3(c;
zNcdF)eI=D<vo1tMfjt6$(+N4Y`zX)OzFQE54q}rv?4?^_Cg8sU|7R5S?X4h~t1ix#
zSSpy;7YNuhi==YAHuHkna%EL+fz8p$8y_2tw8{!UbJkPq@#8Pc5rl`xAqr{0D!u0c
z*S30ZbQf|QT$;@`9#Av0HfN#SNku3)(zXVqg;cp$1O(NcSPKAs04V0CzQRr?AF8%-
z;)bwFZav|h*g%iFq@*9WOwq-7vO?UIb5YNrrTtJzsvPY(X$Q*JB8~u8uVmPTto<HN
zDBD~Fl8gtJdWsTax7jSx1;p)1>jd8qt#Q318a<rKpkKtKgbrhu#TPmUIe{lF<MgLF
zm3xaYsyDaNiY)h3F<&`%p!NDw%zx34rz8|^+gky1)JziV8m8GOAPV1aPX!m9k}bx;
zD;4Tpd%c&J3i=J}s;ImUz|8Jlh-#Hu0^3aF0dOiA4LKy=gGb3!DC0C!*ufQ@S=#i$
zJk{@SmFADoHVWArXGHdT)<U_~R+4HD>DfvAV;9nf0sg8Od_+p9ZLuE)Ot%m~e#EHC
z6X~dMgP||6c^QzbGh$+2lM_OlWvU2D?k+2fC`Y()Me%Cd4t}4qp4f%$S(BzDH#{wz
zpoqo6dDSMnL0&MPDKHx4^@_Z)Uw955G(gyO5At+t0?JHf2&)lKFGcZSD}*LA0F3C(
z3xX#L$^ll3SY&yQ5$p32<&b@5pr6tKKGm-esa}FumY`4bM7Kzg8I%{1+L<oo+5I-@
z6SdtNaYM>z^wXMF2=tnV<BR)j0KoYLWtlPg^R@i3KtU*8Nb^_g8=3QE(!2A|0Rk>v
zJF49fAN~yN`$Q(0JMSx6Zyt^K#D9cYPm)}@I71-S5CNqyG(m?rK#Gtp3W&lR(7IKi
zed=M+d(^AH@Ahx;A825H)Is|nZ)hr5+cYN>E?2&Z6+ABPB;4^4m}AU<oA*iI7BiLb
zpLEHRUe(Tl)XX3Mio<_ptQ%zh%2<D8tpEBrTumR3);xC4!_%#T@hiEu-nJ*$_V~KU
zbHKJ}f{y_gSzClyekMO%6zfx@JVAB%h@p2)wF%GOt`M1cIgud&N+Y2)-(kE<8#vEf
zb<*7YdckQk!IJJaXBEkY2nA6`u5i6i{;oCBRkL3~CiJeIE;3(?PqO2JQ@MLK2CX>4
z1Z@_&3d&MFZRwjPkN7qutW;lS<`rmau8r)wzju^ynAa{|R5kOQ)$Oy@<nfMHU;Xx3
zUY@7uw#Z~c$e!coZ=TJVx)2eDA=uKN-U*3erZYJ*P+T|$3E3FXOZRT*>#vwTK;|PW
zK?He&Z!ydH9bkcBq{diOwVw`Ks=kxap!S?i`SL;l)Vrph05dpht1i6Cw1n2nOdvwK
z$&fl*MF`ps-4gIqN6bI)dgSboY&ujC_#?icuhI4~wxN>m^d}Vk6eu3W=R%W#HQE`>
z5%*lk_vkHuSDlFGh^!4Le+A@)X%G_wXlJ6DnY=k9(M@q$xI>(=Le+K)Vj1qG0=Zvg
z1yeY(o+;R@bPV&5d4uv*eMUjH`vIuNh=utw&-#VpzcS_j^O<tNKQls0ugWt{p|830
zSz7g`vVJW)ipo0Qz--GwnZ~##yR{kc|Lim<ReUmq;>Bv(n=k}_K&n(Krq^D89->gu
zJ`jVt0qDOCye*kc`*u43!5~ra3o13+!y1=<Vf%ar>~Vg^xf3M?jRkc1s1|}en(OPX
zjc02IV=Av`{&n3*rKvR#<=7LY=#8}*RKfOQc98ciHU(c_U9k-F66YVrVW|s$aL6y|
zPGtUUG9)2bz-Y8?G$TE}Q?c0zO6u>#mF8!w2-u)WyrbQ@gB}x0q60Nb#Lik7;^$t{
z&Zsp*fi(#??CvuC6mlP}dl$g-`ynHEWa5IalEE?nW*OKq@|crA@J={MlEJ5z^9Y$?
zYZBZZZ8MNgzPhn&$&kH4HV1d#nml#PO_~)aSX_1#c31c5t?wZ{nYBTtOko!T*Mmh4
zWh*aO=Hi+O3Z?C37$1Im&Erq3&gKpl(T36fer)zi_7pd*Q$<VJOD<XswhVbQcmMXw
zzGlk@SH~mXkD}3>P1W5UMbR;crnRrGVZ)H!`BIX+bWgj07!?0%Bgh{*K0=c7Hg|*G
ztDZ)!UL|5Ridf>jHv(@${}i{mFgYWO=wE~RVg0--5>wuW8$bs0)$F@ikNx{ikG4<x
zxnw1~+G|!9qJeUumb5qKt*>lFvUwLm{D?9ZLdH0$)k(6>y-)huFYuah88=vo*TKzb
zgl~~3bN6J0YPmj3MczCYA!#e`aCP=1pGQ9UOhq3v?%_9;dO3Ck`unT%?B8#<)0I1v
z<s~E5Rp`OtQmQDzHhV}+wn%X^71$(k>C3xYqQLC@4nOh|#9VRdiw>2q8@LEqOO83>
zPJjmsi_JzF7Q-;Ku9NmY41=82&yN>q#`bx|dY2jY$KiClXm{Ik?pGDrSzBXl?(0_?
z&)M-rddJ1vCmFR7-Q2x*Hb{rW>^BqAwofVWPxCt7;nZq4(s4AGQiwL$7NXn;uNC@3
zj_83T<YE-in6xqqHX%OGay8TyCdE+%*b5-_1NDjP@!o}Lr$$;?0&W_a#WaDe&!nW7
zqmC#vk`a1MqG$yyjmz-+3v!}zok~)r&|FuZ3*=?9AxVsY;Xg*g&2ULFy&&*EP07Pg
z*-G1D{Wj?jffDO@*e}n09G-drZNKEmKb(3G3q&Dl91cn2GLpbZ6LD|Nr1Dj7FNQ;v
z-0M#JsFvZPZE)6)s!768jwHe6S;t@^E|(@|NKKtq!zt>*KSt%Ls<e2tf=cJOSP`*f
z4Mc%yyA=Bfy^-ADsC9|lLF>>|LKHe%#m^8negTeVS92?VK7r;N)ym|lD2hUzt?|q<
zCZ<HuC(_z~hH1&RI55~=sAHfOf*xd%(bF>5zl}gw5>|v{^=_K;UH<r2@4$#RU_hh9
z?GP7Y0{S(>zt;goGtD!YN5K8m3K}Kf&XJz<OXb1LB(WPQI2$(?cOOY`2;F&Jw+CgP
zz-0-Le$RM_e$qAr`!HuKcqf6e$pGzhv53^96bb5<$GiYv*>(b8J<b-fCUoa5ZK$`{
z(MU9X9{kZPvotz)Vj;c#ffm9gm6_qDp4n&z+Zi5AeSz^Vq=VQD2T{?jhHgAC!Gg6;
z-`*l>W%Ru?{&Ed}GFQ0}kemnO3?sqkJ(+WmEJttB6@P0-xGd1$WKk&d43oQ{0fpP|
z{@!z0GvKdTomI=Qb6Qj}@Lde%WiK2F=2gfeV$EO>CY_p`J^i*Vk!njgPp>)3mcdRQ
zfNylMgnp?$8{kmcK&WG2bJcJ~=#S85t)J`62fpACTo7^*oqG%MS|N6%<hvx6Nict&
zvMMZ-y^U~9qMh0j>3^UVVHMPK%BrvPM4P*rdVON_ku9m~nO`V5A8X!YS)w4sz}mxJ
zbfSX@6FWXU_qtAh$-EeSOKX`4Y2%msdo>#tmisQ)_p`4@EW{E2r<U*%mbN+i)eIXZ
zE+YP>3wrvmx|<)wHuS5J1>{CQE8D&ce;w!7{}}`ILYoh?5pns2UOln^RsJquhnR`|
z|L+!;?qB)O?Gu##MlCymo16f*@bs2H?oU3YDU22#M01}yk+|cYiv2$8!H+0j!8R0K
zlQ}_ea)HZM;2FmAN$tPs$F3iE2>@9Ru_oiGUPj=0-jgRVJ~P2()qK#CMj3f$N$szq
zswZ`ypD>Uc5B9M~thW1fM+QGW%1fI(fIRL;Wqd&3klGJm{9s<131vdu)Fp%1(FhBm
zex57a+xw0XMRPy6pTJX&0Ct&mK2;-pP*U(|5@(Yb`|orjU~|@0HWs9g%v8$GA`D7f
z<bU48bG?PkouLg23=q^vt5b9;MlVXNW6z<T)?4ONzKlN7N0k@+dLwUA;bxmj6|3g&
zPIfOfyJ$V@kKJ-F8*Smz64Z-GbvHhv6<$=4qc6$yA?;5x@7MvrC4ZEIqm!-y%FvwY
zrG}**NS!WP%VHj%uOn^0M>)lN$PLqkOWJE@?@02y_Ii2V`T#`835fLwRH|Wzzj<;F
zX-A43e;D)cl6#Cr;_pg4@}`JhGs6|qH8KDr&2=%q)ivU%AMraw>Yj5hfj58$>yg;?
z_qqyYzh^2zU%~I!Jvxx1wh%RccY<%HoTPti#XmzTUHo~2)ON)HdbL19@SW5?!=|}+
zlQ`lc|0}QKpg9joUx@-G_=;SK2Jzp`o-*HCMe^zRN|LNk7!yBbPrec_08<*W0M?M%
z8>+ae?#?#GH}xMOKU!!5!ntUc3Z9{^1_@sP??Gw_76UeE!5arLY<vi$-kDyTf9f+K
zo!6(=B-GQn8V+>frqn4ku=<K{GRL}|DO8MOVvopj8>IRa<Xwt;73}k63P}kR(Jix}
z)G!KuBRZ<aO3=s!6TbY2olt1fg%rBRdj;wb@nF2x;LnE{$xXX~OKyLz%CYL8`I0#{
zK2u;#<U;VS-@Qeuzz3KD&LQ$>dXt6=oSLOCgWhjbz7$Z!nGm_l2SYf&LB(c<{ix*b
zVnQW%5XyL;hhH+$SJv)FZWPy%`Y<5iy}{1K-urS`eizc1w*l6@8!aHJnJg}|-YoG4
zrLl)UemMp4ve6JQ1Yzgzhhk2ci6Q(Y0P(?9u&$e<&`(L8P$!OVtWXtp;=?l`+>}{>
zgFo5{{IPYuOyloo==A0ufSQN#fyZ0u<tR3|N7wNia$6sU`#FdpXkxy7rq;tsY#2JL
zPd_5TES0{|CN$9JTS757qeG7*iHUGI)t~X5u!=<&oiKx`DK%<$W85SGgRg9x*=RJ!
zwi=ffFZ}kzUT=}C-)LE)mH>c9qOHDrjfw+>Xa|7I6A=7z8i-suai});G9W_$O#=v0
zQ{zey$~b^FMm2Rh5d6}WBvA-yC>Vh~1criZjd_jvdSI-%n^_d>^pRE&()sEJ{}&Z_
zt)HWC(-D;n%s&o1pK^SaW-~L^Qgl->ohBrRzf*VLAbLMEVEcZ6@&Q=cHawlgIU*Lf
zc!F6JdcXI&-hUh7)m%;ED<#9J?VL=<rWAwTx6cMR)S(~L*-N+}8VzKj>Cd$W_;Wj4
z4#f@)EK;8+Cnk(6ztEq4fA+}T(}EX8t81I=ge`t8h<KkIK6K&x5r5Qu+OaYnIk-~!
zZt5f&oTD8%i5-k*jIswMK)#FyG!7&QH$p^f)4-)*2xb7OG64k?th@)xWojDuLbB>c
zeWE!?1P|a)Q_}&*>}x?5?Mmfgr&F~=mb^@j^UbL<om!&}B1)}vpYEIL|IWc`Lv826
z2Z!Zd^VW&B(_n$}W}C#deH9rmoVJr|vSK;jrkN#EeL|lY5oZzkORBOF3j~pBlFt>X
zPy<JFR|B=Jg&DD0?>g!{G0c72Xjc`bx<Ns|j97Aj0W66g$THB9_|>}wTE1VM0Qvpa
z!z6*C=u|Ww2~{#prwk?s49Zud<8k6ZTf^l%el3r!y+!|zAUl=l>H6U||G6Ewe}NSM
zD=FF)RMAz66#5A@2A}Y<@lU71)ZyS0_OcW)|8_d220mdjHx0b@Rdy$aYT#$jr#jIh
zNc)NW6%-=M;6N^Lt?_#zqV^TtOX4pl`=i{+&mmXwOIA%`aP8m^Px0Jf#U^QB6Xr+@
z#ubsOap+$ZkCM5yUGhzLU{&a$I{t4@OprdyyUsSQWTeqY)#I~x{%$T#&X2aoKOx(n
z{BwOk1Yi2X0)l2Gll6rZOTdP7n28XNwqV|zY|+I$8T6B*8K&YPIuuU$O6DCPe)h-x
z1d%wROO)JQqzVt!Ges-H%jdDWkn_o7Sd`3Eg{ww!w@lR!Si6OcJYea1jen)RW|An2
zmO$E<$6reQoQEg7=3z=s<T2fBluRizo+%1o4nmt&f#iKFEx@UnQre0eP>z$h(;Ucx
zd6t>~CK=-Y+M-S=U0kkS5lz68I|y)_c#l&7DENs})c5<)j@pN4y>`A#W<eJcjXrwV
zV1$;?i;9jQmP|U!6o6Z40v4D+7ig8I4x-V~ikkr3;weH!3fk_Pg80zIb?ffn^?!Pr
z#oh?FJ9V^?THjwqIy}#t^!DOc8^aM-_WP!7KXJ})apBYwH_^=7;HK^*7(lo`NrLj0
z<Hl_$>yo<)e)_8-bWe@$LfBi-H%1jy!Q>-k(S+4qNahBa>PV415jQmikn!{=kh*uH
zs#Jw;xa{bK!nF-E3QeE>AXyNFIwBms<`Hihp04GFU&*#bZI834_&Vwdv}ti9TEG5?
zR1^AwD!SQu9<Xdx84z95XrXEi{RoyTh-F{qgf@nD01KnnRHQ1QpFG99PNTj8JoKlp
zn2}2<)qpOC42p1qD+!x>DS5B2tRRm&5nO5{izF=d$8WSa)x7PyX`gg0MG4Jmt|ADB
zeFg9QU7W)8+2AV+ik{(ss%K@xrFe#J0*SCI>sx<*1>qeQN9@!>EXb!dizFPaNyQ?`
z<1I_kXSA=pY{+zm#&*jftGZ>7!bq+SJ+%NweDV`dUw4@mQ8a1qrH}7M#2ub-EL10I
z?qFhsmpYl^^B(0KpvPHae(h9jqKiIJ9W+TFdDo#Hae($ylSE|2`OHT1%OGi}5dSX4
z?eO%UuV~-he4^1bU6E0h#u$+<`%CJy;`#B0r}11nr$;q6!{=n2nr|w_rDY{zUa$mR
zZWt3z+3FT`XqS18Ff3}Gv3UF2XKS8Zto^?C^~7?d_Od=PTpmDLAJZcCH7lt{27vN<
z+m=Oq>{jw8rOj)LoHe!0FTY)6wLbQt*8$aSixx@AOw60^J6VDf`rob~DgI~Fr=+j*
zj#me1R0gM0#)#GwAE_<gnQ!N37MrbdCGY#&>96;X)??|MFJ{$>bS}Rpx0}*PtX*``
zfz)f*BYMpzz3HILNulB)+2p(k#OD<N>s*~7KzTH@!D?578B$PI@W13f0zX?^i;uK~
zt}jLn%r+HNv$Lzu2eVtb%h@Uc?KWd=WYvhHw|!o+=!44qoEewhcSIdOyD->Mbe{H7
z%-PHyQcV#~flecncrV=_F!Z4NJ$$`%8k4)?3tF%H-it);x8=u<1GQi{^)G|{U=Rjs
zz@YUngvE8&wYgAJfVDgWur<&g>k3HidosONk}3FCCWiz(b4E0Lv}U_y#*Z>M*S{I^
zY?T#N1VKk3B!ySCc&VxR7@b~IE9Rzc6H)w9Hd!%W%Jvq^3^!ipOK&1Y057oJ^}A+0
z6LS5USh$es0y11~edQn_WSP0|+=Vzj6fSr+S(8yfeL)c2ObDho^3nG7u5C7mLJK{X
zD7vYUP8W{)jMk;uD%ds(p&FF?Mx+hS_G>794hD4%Zy>T^8rClc$u!2X_xm4ppq@d5
zP9ZVB2Ne1)h?g9b6_qI2;Q5=z=}zda#WNl?H9GC^z0fiQ$?nf8SpV#0$k9$JM=cw4
zT?@t2-G069sUawZO0Z3qlO?)9pu#WF=)v%mB%<`EWd6*!lq5Pmj`omH@=~l|k-Qzg
zwiYN*O%-50e3mns-1=_%0m90tarUAcW=*wl>$Ga|^YOGZjm>$bi*b`kE?i5R68vLm
z#mG4F!&VZS0dG%Fcuf_Kj-z4Y)vCgT*4!i@r>!f@urpScc%TCN$4EV~U;V3c^-ri=
z(ReCbxJOA7MTU<Z7h|^z%~&Z|MuEb-@k?%(9eQ2sYg92V6Y)vgXbW+F)HdOKzyJpR
zAnK%$c*R6kG4**XCchPB4Q&HAwm~NG<3Rg)5PW@9y=L;F=HGK-)n*#SY_C6fwl+vT
z(%*4msob7Frrcis9yuB!F+lSFT%}B!vl*8a*SOF(RPMY|@L=ZeecUYTU}*tXgNA5Q
z)3$g>Yoo$MEkBe7P~xfxr*aN;!i<$sh24Ca5!qSHmyv3!Tfw(Zj_fX}flm3al3M?g
znggCicdI61XC~!=5I0b*ne>U#B7!9Une;Y*kVmB0d&GXvJ4L7ou0Ij~Nal_xK->4c
z_gXyw>KiRiW<cI62LUI;2V(dqPaq*aUM&_QQC2Iy-mr=e{30Ej==_VaYjzWjW8x-x
z5Z|BO27kD=*@?Y|?nJM+4Q%!S4?1?PlO!;7(kI&Y8gB8wy2EsU=idMK`(~um>5dfs
zQo<Q}N2|S<r)IR(lk>^VP$!t3##Y2rn!xyDv~ceq#jO@`zLR_0y$sJDLBYm3lL?kz
z#N6=ZLF@$VJm|(}m!>NJ6zk{!?KxYjqE$Ero+E$zdn%Fkvfrj3yT+yhu+P2e1)MXD
zIu@;WvCPusx@FS&i-Ub(a9x!dHs6t#^`px7_Y6N8s2i1*Fi_v>kD}meE)J)16N<|o
z^*~J<D|wO0`+S&8j*D32PqTqyp^+UeunwV_@>IZ`+;~mq5+Ol{3)D8Lz10upY)b~L
zxf9cXDspu=VGZ^at@)Ph4eE$@kVM+i!W!$%%4Bd;>uhF+h%T6|#gBnpO_96d542Rf
zl9PB^+Wbt#IgZp+Tqca)km`?JMr*2ZB3S$;QcaN*S1}`xH%BFtB;&P~O+I-92G8(3
zJZ%UTsm}t8qFbyi&zHKL-S+m&oO!GK)*|*#V`kvWW^7OVlo?dkD>NA(lw#!2mFYFF
zL<X=hf?;(O<T~ZnNw*vK-|+?h0ew-MVRnu<F?j{6Hp6yGzVi}nXUreZOUpjQYFu`n
z*BF=Fk{q{V%A;ysJ}pO9z~GLDEHU2jcv?r1*|*5;p!j@yL28CngHz<M&Rm*bJOR$&
z^@z#B8*|%<W8<!JlJ{Fo){VG$X8HU4>gm*-fHahOLn7A}Sl21fdUsAKe;XA`3@60*
znte(O+zWZySv)WH&$iU#Rn7$q?as_&UjlqXGH)OcBY`gV1ap$k>WA(w=~||*oLnLu
z7?Z999aDNf?ev$2dp1wuDAl^Zo+|fhy04z^-PAc^)sIge+K#YEn7GHK<}s<Vk{Qdy
z5FH2`KJ%bqU)5KSG2otDsvjDc01fboS<#j7>+2IUL&`&Tb$%%Iv%bGap?JZ_&FhTb
z$%T)!4@0;O-DUmXJ;s_a`0O_X?O)qhOGpMdW=rba%ntw-zUNKPOyQ>FSrQ7XmrRMt
zt>J?tr?V!N_7-0u-+w6xBqy}0NlLH62`ABtdqj?*juII4BhVK&va{%R^nTlZtjsUi
zJyK06YEKfd8KidCXfH$Z0iTaljx#Biin%vZ1zIJkGkz@icyPshg;dBOKAnm{>z<M2
z8T}Z<AEkF_;h3W+x{58sGsW#Jj+WicmHj%I5w$y@I}5T}hWV7Cb(t|9iNxV(PVqgI
z(VT#vvqA*+{VJej8=u}1s(nbmt}9z(+jtgqP19h6#Ka{iHZQ<e%dC`&o2peQVy9Zr
zc2(A-&elE~5TJKHffmd2HMOG+O!EF^OR;{YPh-|Y>1?rG<!)h9vD5KHF>fPCk%8u&
z5HnzIJ~@H&gznWmnE2X971hBuAE-haP6&k5%nIh@Z7jO<gEIs_^#a|Q$Dg7HS>LC?
zCxBSqdGNwgKV$x+nuEMxU}ffmvENg#ldf_U8zu&cn4cq>!>EP{L4(ubqo_75`g%~p
zP&t}B7*&+E0FKdsV8*?rRQ(0qsvWm)xuMCW53p-_Q(XnclUT&kFBgSsqE$13L)XKp
z@#aX^Q+bB26n+({HU1tx=BqXC9xQBj;yac@C(sfK+6u$4BG7KZ=@}55mPFY&C~`L*
zV~jbq{`jre9G?);t5(<v12)L-=p~;a?o<^Ldjl1l<>#wcDXxw7D7ZjB^S@x|x!IVU
zHLJ$zT}j<S|3Z1|t4)&cQH|T=1t^4-e3_&2T~{w8^BC}Khv=$^=mMPEWiS#lD-sJ_
zN*9>B{A?E<GH0;Y4%S0b`)@Sco!QEip{xyN{N=#d35SY9m}=@w+CW$q%Kv>e<_s#M
zuM!>c96<F2?SzF8B_)TiAPRkpx=tBReYIivL(slFWy%e1AXMvx?8$hR5;)p*yJ-`F
zM2y8wqc!y+0?xhDg**_3@fY>mgz(F<$sBD^)Vx`S3RXlx{@?pO90rWIn@#^M`0`rO
zuj=coH$`kJeG9jY%n&ZM$*I!^7R0CHuyC5S5Z?1{hO-<CgZS_j-A5A+sV4~*eNUZ6
zOQ7$i^OuoJ)$lYAGyL4EaKf2=VyRaP$sgtHB0=)jqzj9)?wzAphbAs_&7Q#~L#rGa
zKldV>;4k*XYTP7zcq@ET8wKB)C-xgh<9YYUke9H`_Uy%Ob{q@`Qy8wBXC`Qi4=2Pr
zNmOMmCw?;qL0VM&OVBu++-50rw5Xb1E-#h7anny*BG29rkMZ*!g%}tZb?^I-6M<Lz
zY>`q@PzllKoXmOfTnqa)L+`HdWU(mj_dL#jTTS;Lny2!_+IY&i;j_-Hzek$8d}ur=
zu8CBqM8w+BH@-_n5I&}sQ9FY{ZFz>a`s0Bex8EL{max^_=<SIM?lE>k({%j!N0`KB
zD4HW(t5JFb1Zq_pnCIti!b>Eu?2w_Ax!Y%IjcPruv-<8z?+C|f7FFeIX3ZRyR<vM(
z&H|Ze_gc|nTGKZmznI1%C>y|h<XWEq{0yn`7HY2{m@OcK=1OEvWPPB=hGOHqH8?1)
z>PuRn17NZjp-3*Up}%3CD7j5YWDQC1dt%4xXntF6vU}Mvvx9fGtX<>sa>zzT{9$UJ
z1Kkp+2>_^`cExVe>u%JeU@uVINY}zYklRh#>A4^jc#MeX{Umkth#S<CM41CV%BAjF
z)oE83xqFoGV|G5(PQLPT_l`YNY7dPxwCzamdnDYgj#hU9pUB`I26Rh_z+nkzLfjhJ
zC|rSlTn4}x&a3L7G$*0o6q*9Pd#rmECA7+0Gtv@6{pKB5VlyPtX|l!FZ)G<^+h)yK
zd@<#`nM&CAK35Jwwc9P*ftu{|6@AZHihn`~p|H=~8d|+Z26==J!Tm+LVWEy;Mbu;=
znNT)Dd)T*m7Uf^{2Q5jPumNej9rAX-6YboJg`)PO)MvQzjy*8?K6ZuWglv)QQ+n@v
zi};Xz`jV{**eqUBrF~kfQu($9<@qf0^N$YgbW*xPSXyP5aPryOtYQ2u&$3Hinept&
zpWoqUQ$zWrE<{0}_yJSEM)&zKeXod*CEhPVX%wtw2fv}7@F6iCuu12<#@IzUhRO!c
z$`A%V&$_c^?64^aQuPY$|8*X2X(ax<@<G7F;-THyU`uEplaAI^tk8I^q_XvC@e!JT
zsdIGm6}HLXT>6p2S(O*F-xqHkG)J0!qMs6=o|r`*>?6awoFO><s){_QUT{0|*ld2w
z9%iaV=}_yza=x+y-0Y5aQqv7EBO(jH$D50=N@$;{y??f~trAb2wfK$4SL#_E?Fx0h
z^i72|W7Bl3)3d7Y&7IlC+aOcXnoZ-+N5`N|cEymaJ-6a`;OAV?<0#V6Y}IK~qDORK
zSPIT@wtrbKd}p-ePn)k#Jz~e2*!Q{e2=3O{82o87ZW@(F>>Mi=(!KYI>$6jxXcjBD
zBH_5#N{1@iaQ8{+#>G$E_{$cwj8zE>np|<oaL%6urBMi~h0g+LA8Zb3<tzA2JD;q>
z&)Q>Dl6~Oynpf1y57#(vGe4j4x@vdHr_;qs%dja^xZKf2$As0mK6qE{)D^9G2ZFB9
z>{M(bKq9&sT9)TFPz6&I1c9!?W%vh+&PA{FrZ@zBOPYUBR8cJ~E^uF6Nl9x(8Wbnd
zPKXrjB5b}|=Q_;0&@z%`?8m-&Io)|8JpQfC!2tcEw|ffieYwR(>2n@UTfX;MVX|}e
zSBiO3K-Qb=^a9O{wxu~$=hmMu&{iIF_s%@CmE`c={wa|*N;HqBT{*(W_FcNI?6blO
zMznGkfBKgpnQ^}UXE~-@{9For7o(M~UNYcMLLr$F5da?q+`|<uU!#J#K19$=@#wtA
z*?rf0pPPVu0#2`(MZ^Qh{t!O(G;q~KOs0zb=9XD{B2K%{D_fgV7#DY(7PB)cs4{Nb
zwv&7J%E(UfQ`B9nS^NL7_vV36_3i)oAWOtlh-_0Siljt~ZL}ass2CB`CL|`wlI<v=
zWGN*@p(M#NWhV^TE4xfl#+scO>zL*Ad+V-l)boAr_47Qx&*%O_W=`YGIp=-7uj_Ta
zu5Iq^zrgW<iO5@HT~CxmEDLITAu1<ts_WD!chmWGobR17lr(zfaV<4#HqQCii2mU=
zlICqV?Rh&ZhBt(2j%5KjA&v@x{9KU8*pZk@p1P7J-YKrogCS$CB&qJOvszWU`Q4i|
z8I?u8#~*n+RLfRDb5)QV=2D&<*f8zY_njV5QN_!0)#3)PDb^L5$X3L%1UpuU8E}LN
z<}l-rfp%UX8a5IO3qUsSpju@#9lZ5L9IG>nZgxhE^M$}Pbi2-+53*dH)%VI+dh)(M
zlPwm&^AIXV-_en^SLbX~f8BM$iyAg*BA9nS8t?&bcm~oJ&qGtdwsKeJW&3q#kTq5y
zj{0t|$wwRl=<Cmy45hsPP;|zm#y(5*c~H49z)!zky@u$yUo(1pagGS^V48Vz?~u_>
zQR<V*aK@WCx9;TzI9@GuS1>43D?R`+Fo>N&ke-(kx!l`pRp0IAtwD^<wn>K%zq0aM
z_gYYgOUC4cSlbN?n;$F^+cfDR;URMF=J{d8k*^6qhG(3ZYiuDNbv|my&N@}4WU5rz
z*gkTHu0%9Ws^j+S?FUzAaFul=-n&_!gPnW&-)JyQVInjt>#-0pKs<o_t~*|^4&R3@
zLva@HQbQXwvMAC%>z&r7-&dDa5e#>hOFkjI=OE|slfH<K`L^dbTVvm!#g7P&2*M0G
zsv3pOD8!5seksYxw)cUqQ-o*fWy7B1XSLa+1%G0R2j)2XcoM)gbFAh;Gp`y#T2)*L
z#+k&IUh&B{tj>>-xUNF-)1WomsQ!;MK)v)euO*CS6w;Lqc}5*0Z0}A-WDM6Rf3hpn
z9=&m5-*t_40haO|CQaxxtHI6-F5=Irj*r(~@7(5vrt~9VC)pngsxZUViF5_HTG_55
z3+r!s()-Do>zyf3^JB2NSYcLck&N3fVj1cU$@1gxJ94t_ZuJkgHc?jYnlMIyOx=CV
z1mItE6s&~aS@eMPc{~fgx3dJruFg)4H_fLAT~TF=C!IFE9Hdk-ij!X*^!CoB?Mfj(
zlfRh);W-J|SUh14V9a93lU>@vFHH>S^Ha?VGeztZj=rfqqM4^^dF#Q7P;Hq#eXZW;
zFZ9<Onu74KO1n*d@G~Ouw2)~^EO|^X`$c}(rwA=lPS=fVn<d;rb(Hky!9rkL2tSmo
z4@}6<`xpjC9&4K_B5=AsCLJN;eAb`0JD_-D=_lTnx8{3}D(vYxYt1&D^h>@swr(Nw
zKEX;s2hBVFCfW*U%Y+{VbYik*a65>I`<FoUB~BoV1cwEWWzJ*Bjpp`{rEYI?j~u93
zNR=wz7kME4c3Jy_4~sGz1Qm9_m+n3ui{6WX?VkK5v^!10Hz29f3uw;a2_AbX9BH0>
z(eWlW#;qP>&-Qd^-FWw6Q=81??NS9m6U{;DMsKOmjE&~*i>MB2A7M9n_ukcy#SRj1
zzw}|}=F1*8Z;VOY8PHfBP0lS~1N&ANoB-x51(+MN<FQvQ7IEgx;|%Dk2-|X=#hQ$5
zT`bIgU_$5G2Kzoj>d0U_WGPn%Q0&UU8$l_K-pHvZtnxpao=SoZUYLpkUyvGJay3t|
zLj}3G{ETA8-6X5SC7Sp)8L@<Iay}t(Zpe9tEG2#)XcFzHO|HhbecW8MWeaUX-Z9&O
zLm#_8h&q#&JKVdUq%c#~vZe7iVR^;~agfB|PlNYht^!Z`7@DTb?#VM>;u9P4gxU!#
z43&C&MfnBP(=FQ1awCV9?qWF_w4pY#+OhgA8##uj(YR-p$ugX4+WcoF_lM7N2W9u`
z4(xbTx+!(1-vQQ>ZDLL_d^_jawb`EL#@^(bAr&NzQyWwOqsWiu7(h@{n>FjhA+tUh
zvq)Lian{$a<A^_2!_fW3x*BjOdrr^-2J_6O<O=XiS#c*7)<Rz8O*K=%^6GZ*+>yJ@
zrk4A`UJe+*J`D+XP5aq)?PSd=-OV~;$Jf{gD03tH0Fm+M>JHdC^jrKpvq~YKkTzYM
zU-OuCUH@$&{@gMysWB1hD=V4fxIhlIO~eSQJS^%y`3Pj-eZn$Wp99V2#!RGydl!f>
zv_kYH(BzEAc{EdHv;4A-6NkAraB%e>vg$j6o+nT-$ch@O-AMCI*#9m><}L@w8R$~K
zR%Qaa_X6-!5J}Lq1mUC-?!C+8*@k1mMqfNMv7?EB4)DaW`QVUI%FLZG-(^lg7pplM
zhYtxghg?!jyxO&5!}O_87w4DW940rt5#xA2Tbd3<|0UihO371MUjS%^19no?vOEVj
zbapitM{ihDgpwKFMk9jQ)jW6fw;F#u2!$~2f$$L12gGuig4m^)@vu&L-)K&+6$7{q
z^m13Ja^uEFyMyAR9nRedTw5XOt&EZV+1LG0!^v!$uU(+MRCYSrCii|+Y`lkBgWd4Z
zsbCXJb9(6RqpylT^>cx0m(J)Pi#*{_29A8NEhIqoAcS{~Eu0idNpODE@S%$E=K52a
zOJ_COHBi1kVIn5ySANhf=U;dtW*2oOS+dx)D*TF~Ji?~QH2A)PK2FuWxO~0Uxzteo
z6@kDYZv0tKN19?2%;~{Fk$CQoAApeK168{WBXqA_ebj29V|l!K=+gGpp|<ml%NHB;
zLkWo&;78|DB)xgkWdAn&#k{#b+~Sd?+Mgt~-mSjzY7=L$pEo)RnwKs~czBHf=%Put
zg4_t&NPE-W^;3NHdwRSMf9SqCof~?3_y&CwTrF~MgvUd<L9+QpDq9rwiyvq7L&L~{
ziDe^`_RM?9DVA9%KE<*n&yTcE1HK^_#Qp$g0;=EG)3K<4`*}UC%Xg^RnulYPPEcQ^
zs@S61-{{48#?wWZq(EqLg)N%CpKx%V7vUr!Oa%o$MTF)P#S#mjHxM$<bT<q$l8dV|
zA|+1hg?bqmRsj|M+vpge!dpmbSVkm%ND%Zs$(NqVLL>GI&D-ZGoyBM_b-rFNaf-y%
z2ULZ!a=oVZt4chsQ)hDbycWPz_QnNo8qh#hsr`Fi1X7upLKPFHH=zQ+ibw!5YTPoz
z3xxmYu66%l!7_$KL*AMv)TOVBj5qd*q}P2gKb(M*u{kbdc*klxyVmw0@Qrc5=^JU{
z<dzO7xL_$<A|xSAb$`0$M#Ae(zf7CkYvZJZo@rUOC@7B_?ste%Cic!JR#IX-)={^p
zig_SZ`A>XCa8dH$TDnkGw`ZbN_@~b8AJc;K0kd-R2Ym_(iU_CmzZMjsU~N1RJr@z1
zrFy+4q+POMI|!V+3y+s-s5lcGgI_#hqY!>L87;A#kWMO>Lror_o}lpSQFaw~$Uz2g
zR~EdWD<y2!#;GQV+}b2y{xrtp>B92u_|*AQ1GsJAy7Wg0G^iq>w#4P*aw0mGY3X*Y
zhmf0>-MD47{dh@~do5=tlCfoKNCzbHz9CTA!R=OzL&Z3+#^|pVP-sR7HU*jk(|#Z#
zZbXw?NEy;hqyh+T3jzo%NgN3GR+sAqd$)Lm-{U0Z@%?yP<;f_bl<oBKjIFDmt$rGG
zZIr{nIq>{M+<f);c7QZP4aV-NJ1-6u8E<Wp5Wm{i(0<cF*E-t1d*2do=ir>apMftx
zc=9B+p_u{%z{Iq9U>Qc>cIFD%d{}x0df86}pVpKw2JQ<=2#|di&pj`kc=E1KIXK)=
zd?@VgJ%u9GeG=P`JLHH9pEbI5iL0O26`HqQ@SY~)=4^ZG&)cc&3y>96*>)_`SBGcO
zrYF0D+{jN60$?ocY)4*u7V<IR5TC~Y_#AvDe!jPc!8OAG9RGcPh+vD?g4zIY&Z@@q
z(^*~lGH|usPT|hfJ$J7bv+9VUPDdZf&&Vnm{gkHDVy$v@1#>N1;2dD|toOIJvHM9o
z*`y#_{Yb-%Ju76AQba#h2qS_EX-)Y8EQKH&TkZ)|vxl$+cnj7bRAOcT8ra<M0-0SU
zL{Ck`Sv2|DY1CM~d=-Hjfo@;ci0*lWjf5ylBts2hH?6yBzK@#56mDM+-EuQnOXcE`
z5ymReDQo=N7!H<t(Z81M;@ndL#)B*K!DocpDn`p^#MKm90GOch>sb|{Syy9Q@X&i3
zimI7`obD6)L}KWnW?jn2lNdfuc!}LE*iCaf>obC1d)^n{u;U>2*_@O>Z?crXckhJo
zI2;!JKlbJS{;rFOij@;<J4}`zda!zUwG)l)KlkJg>}rpsehStKT`$Xe)(jw0=NC`c
zvIRA&s=2$k%a!1SG253*qhA3h{4ZOafB$@DONCO#L+dHiGov~xlAK3)A4S1!7#&T7
z!&WcRf~bC*x=tR+v_&`X@kbB8g=R^MOQQiN+uRigLG@n+#(d&WioY0gmdK^cDT-FC
zA0GPHDYwlaTbj+LMipcJ|K2`x{yyYyz4G4p2I+7uz3qoRO1TwuT8e5OLwi5`ncc;<
zt3E}s23=!`+ibrj)Xl09X-b+2+1_nRw@u#LEp_pR(LvmsJ=-hQE>5IG{E1DzXiI~%
z9Y0-BVLF>_yNc@>n$>uU%ip(*zjqlbvAtWv4Vxd~gg5S9lB#dmS8Dawsj%O<OC0q<
zOf+r~vsBg6psVd3=&Cyil%(0=x(ySe&sF&X9_(b4_*~bIQ3{*8EB;0$Jns&g{K8^d
zO-R~~o6C$GuRrR!5#09Qv`NyF72HJ^CR<v{uRP9uGx!i+n{oGFsUv=`aLqN&wmZf}
z%Nit8krLzAbGC0QLhT*+cQ$a|<Nr&I??MmxCmahZg*e{=U_c@&z6jDgwpamw+f3!E
zxI-^5&QRk|>9?D)CROpx!csDCOtM~wZgl8pw*?JJ1S?KQuy+rQXBny<L)o+OM+xv1
z(lFRRu4g$UG9K$cDH1rVkg)4!_KlWzB0aO<$PCE&Q+F|U1^*qq@~{wYT)NwFwaEdb
z@et_AmoglH=**BuwzgWlcH}P;tXQ$su|dKw;XkHf(jHn9P3?+bt4MN~$J(s!Suw?$
zYr2*#AfMblQTm}FwrrW!dWHRG_30UA|1lMv%%jFJ+S)WM<obGl%!Y;KoCHDQ|L7{t
z1IcsbCwJZXN|L6CwdI(G;l2BLTRb<@{;O2XE`wU!hTHG7C?QgUfn8CFedYgIC4<%@
zGl%YvV-)f%4eojGzp%iKJJO_fc!H-UeP?fCI`Ymv$~Z?ZK&yXJ)5$jPt~D)YYfHC&
z>d>&_c+%J2N6_b=bOe7G(37W&kN01Hq9Bxi+l2-0yX^WGR-g4;%l6=5UM#%qK7q+b
zf?*^q6+$(21|$m-qEpe-2*iY&>??G;4e)zv)j*6H(2+9^mv1}{1f8coBM3ggA&Nm_
z{yE7%&*UFD`DgF^WvU<t%|m^P=J$+$owd<3^3=JOG=j^szRcWLpu{H|zm!@n&u5`+
zwI*@@$v~a!?GB+@%Ba1FA#42F`O0aIa`RTR2mrl)XKT3m%8l)9?>{G*r||!ICNnhR
zTqF3;Gx_J4Nc}c_l7!+LNazq(B|VZ_EcUqG<vgNin?y!F=puoM8i-Q&-~Cp)L82;d
z<IuFv#rv;BA0v|yZi0f{pd|`$sCp$nBc{j@1+dHwfHID82dV8vfak8|Osun*30mwY
z!52q9BOuN%e6pMhMB^kd87PJx!Y%)wlgx1J|2&g_<m8{d^Oxw81222ACM{U@cN-*i
zw)AXkWhsM1vc9%&{K=RL{B|mcU#;MQXw&*kmVvs&YA=v8P{B%q3qTZG!wo@>m?>EN
zFZfY&Z^tl7(L-QrfJMH=E^;D#JbDs@b*=CgF|2u8JNy}bLxh13P_Lauw}mKW7re`v
zGyw|Yx5OZozIJe(iFE!af7nRsBH-cXk|sWaYCK{LCYqrHQmmW^11~^r0bT>t(w(F9
zFoIl2ldnP0TFz4xyWMpR)_O5AlhgC?w)gh`hJ~292K6}5yw_WiM4R*nQX+^yECGz*
zb&g2T5T_PcMAnR7qRKz(ZvHn(oNNcK1>1ox9r0-uA`?rZ{hA+S)ekBO`k16`A@9pt
zT;&&-_F6R0_Lg;})<JsFhf|;dlZWOb>doQ^C}2FNxG(p966X4=J&0n3g@RXm)eqj(
z5Iv6p@|6me60Ty!@+>Jr<;strVhgumhK@Zr%;UQFY|9%zqrkmyo?kfWcW(2_&Wj=0
zGvY<noouD{brLQbx3V_niF(=D;holT#B!G|@au10B6`P1Pjv7D5AW+so147VhaMKN
zQFlD_YH6+s8JE9;Udqru7Ej$}L~^+Bu<XWkX6BHn>c(3eS(CG$-sazY@qWpTfNn`@
zDSd}mqdJ<Hho-CaPe$>PxB8v3wV8PT!0ayW<aR`fxUk2&4^Jca>|9P-pQwR)G>-X{
zl1R~=3})W)FnR#JoFvNSxs%5%vUMe&mfL4eZCScy{}HVK9d4d5J9^k2;{`{j(3w5u
zA>tlO>+?7}P%Ni+^4RS%gfh!+6s>)c5oE}CHd-!KRd{{*#tT&l&rVIR3)*djmiY2!
z1obMaL>9$$daJd4lE=2&hKdGdo#)TisKkm>3ssu-@mGokOrYOyo%C-~SF+CS@nc*#
z9)S&36?K@@+rNN)>Au^Vw^mj+wFZZ-UCn_A<Uk-iov<6c>PhK=Xx71WPkxPhvy!OI
zl-&hd;g1uJ3U^80m8#V-AAI^`<B@x;h}E9j7EUP)De8p|oEV_NmMVP|Q6Vg~=1DE1
zL}}DMY1QCPTJp^N%7^#`Jx>H!D5xv&X7~e^TX7c!)uDgsJg<JSG5sVVz2T8lZu*t$
z3qMGOi@HBsda<#MRr>=-tAkfkT{^NZo)5VqA$TKNuW9Aj;%&uculDMQUc4@{)!4X5
zQu2}(uZJxC0Q{cBca%)J1}}8oA@w%z;FD|ltd25r8DUAG!<qLVMI3+XZ23&1<;Lx!
z+wqSfOO)SM!a{X+SKLeGxJFEP>FT`uZP#_V+_Mx0gBrsujgD?zxAMyBvziD`q1i)e
ze&cISyobzZ#GQENGca;&fS6d}B?!Ga2)}~{Ck7f37(WKc&ba<Di|HQ1<W2mLSnRbH
zs<3Z?^UGnsMwU<MUwO-Kz1x4&Z%k2vBK0+8b2~J2qrHZe8f`+?E+&YzBsniu%g?=L
zn}uDw_U?(pCd>yIzjLf_)Yv^`DLozguv`U;sj=nVVfKozwdrfmEG9ZG5zJmOAz{10
zGfhL{g{H?CeSJR(Y!kA+1_fIhWSC*@VqsjcsCE~wyZH@NSG#^m*(2e?o2#uhXq{#E
z<ekCr-Jt#(*(}-#QjtJs1MyvL#Fx|ci~$j+jDj%B)yqDq*tT5Oc8|Ya{z;)!_{Otp
z11C!vt`(>!0+?bF(n&sqz@jSI8G`a^EZC4v;RuQ|Ep|A0JCa9n*hKLd<BIU+?ZJu%
zxVy6p#iqkA#xc_gHoT(nvQv?rDri41u||{1I2(%Tnv&GrmwjEV^ty#Z@~Zs0FCW|6
z#WtQBK;6}rRgMd26{8l*54O{_J#TlGY%V04YNee?6fnJfV!gma8L7iYlh+eo3CNf&
z7(->04+&Dt+RDpttkzy^xFH^Ki~3zsH@kOwwK%N~TEuN>-E1`diX50zrmz__23huy
zpq#g(O}?dJ54kIPi5$G<Fm~YfbJ4pVD!i9heRvo&wBX#Cl3<QQXBSFdRTiObo;*e!
zv!U<w*yK|>WPe*Z<6f0jg4xP7k~x7KA0>vJ_ar6dR4ZUsC#+`cjo7KNT$M!^i}ZR+
z;Lp=Z>UYBqmXd>yP+yl7+fBUW58AAIW;%6G@o>0CYdcHfhl)L%%iBj<b{SGH(#>qJ
zoF3e_Z&KH__BYv<X<ZmTm6;h~94f>oc)ax@rd#FKvTETI;|B<mX9`4IaqY;VcCve+
zgy1#H+jrZE2Uj9=RI4~1os8J2oEH@8&gX+*!KNJg(usAZ8|&}*VHt_$g}Tgfv|&)k
zYFjH?1JAM11}t-Ja6u6MUfu<e17|+POr0qX&=B+e*3A3&{1znRifaN1{EG4|Y^4wj
zENzQwHr0w_Yjr)JpP#peU&2jGM7YzcL~8f(?%>|+emys6=prTZ09BG~0h~(<v7BYC
zs~WBRiuawnD)qiw<^JS0OS>DAH^L;HSwQBdVVwm;eu@+^V{9Y$Gqeiqv`a=T1tJ2g
z#Lh;i@7<bv#@7%Lm|5hzoabhxQt!+I^V-JPPW?zf;o(l8po;bV-MaJhhr^y6(Q`^R
zFUGN&zCWpnJN;PJxp?#Xpcg0hI20q2uOr{}%XTK~cf<#Sxl@Gq>8^y%PY;k+RV^2#
znw5&JS*3fo>xOUMj(WufT;Xq=-|?}X-Lrf&WRdHx$zf`AObcUYJ~iY-+g&FCA!B7_
z6Rkdh13?vOw}alD^%}Tq0xc6=f$$18#0q;=q0>2oJj^->iy3^>_m+J3DLgMvi+B0(
z2r>!F!wz{o@IH?VX9+8oN#R1N9Pz3}rIW5g)*~nnE~<k&&Oz29-PE({V;(E7NcG1w
z)uBwe`q1S@9xoa9*jEQ4ypjyDccPank($RQF_#)Ttgfn_z)`loT;<yN*6w6_a9!Kj
zowaKXbu`yIC;BxDr}S1R(UjWd1#Bd?kPB{h^0FS>azEL$*u}o&SfJwyrw=L{ITtu@
z7htI+_CTfj%~0nAp?t`t4PO||WfQeFTqEt$Cn){uf*4f?)j;gf#`27P71;wfr@gLV
zK;a}ZM5Of^*nLMSA(!%apD;8)Arh`tlxuzJ^Bu>;OT6<&pYAa@c8q6qXxnzn&K0Qr
z?>vAwT=IDh*@1-Lmue!iRyP);tX^*NnzsZhDIDc|j<0j0B0<DBlR=@5zHZMdk7l*-
z*#1^&%kK2-`^~yG5?akKvdWk5mK(jw;#AWrqo7T?^?UZux6iUn$omtzSRRA6pkPWS
zBH68CE=2t5gEaow%JlC+gsPn#Q^f=F0nv-{bjUQiwfX7lJtt0eX>Xfuy9u+%Bzg%>
zo3*2wkq5~kgK`}N_S5^DU|I6Sc$8qkmTh56sV^2y@rvzi7ZP8zlW@wG(txGP$5P8-
z?&VJHK})gdU4FJ$ep~0)(K0GmB$wJb^_J<JJVIOe&aIZSH+ry&T<Ecqx~3iC&=U%K
zvgTP1p))$u+-uUn(|Orw(YT=nWv0{GCvAp6K$B<-<m-%)Ef5VlZSv96s6->Ag4mbu
z;X}tT5R~xpgKduOqOW=Msaj4_c{%j+q2&deyD6wJfgIc|=1^>_xB5Mb2WHGw*UW$K
zw*1ht?xX1O#*!j5Wf0Se9yB8pJ6(cck^Zb`Zgm-XpEE-fI!iwl9$r0oYGcXn(yP_t
zXJa!B=)|G7UhY)?NfT-gxuIi!GG*%|hN5;mIBKPHxM7`>Chqx*P_rp_Nrx=QO&U9e
zM|T!9q$4hrm!aNAcS^lzH%TW25z_hYMJxz6yaU<~4;{<6{E;2EetpntNd4nGL{}ii
z5N?^wjNrWFwZLS+rLKQVt`jreWvk6a0ZQ>ARvk<zYc{*Tv`-_^&VjxW%$pnQs6riP
zCV1$^BYAs67e1^VKT4N!oVpxn<UHKA?4zXJtCTmxDUL9=(c>19HMA(*k^W?Z(uJeb
z2vNg&rD^RQ-mV?qX#KZ7h*#&3D}pDcgXjuN61-YSVoIK<p8AJ6ZDH4kcb&7371v%a
zCzj2yO~CnLv#fm69mvF@D3#igJe8%QwtA*R;z}y+hNDljhNF9o+Lq)6F51>n5WobE
zj#V<xq#OA>O)0MBbaNAaZO&Vz<wCYc_OWn3RoD}x;CxBv^=AItg$SBNUOa|MBzNwl
zMn7VRfsN-=>ei_4nf(0H(GZFE!YCE1u$o(KyzeW@6g1s3JYwm(@S8r$3%V%i4*#~s
zFVCsnWvNZ>l|_eqahr~Z<KD1Vj>!hzT2o+LDZnzQj;8J+6FcNVXCpq>tet5Q_0wzQ
z<!_4U351@!8ymh{hh;j;=iWq(5|~}WuUp~<$B~OmTEz%C?jL1$yj|etk)*>NTAsV7
zLDhEwC);z!0b{2cKWtWciRV&%e|RIs#b42USs#1;1oKkmWMZVV@O8v<`Bw2m7gpw;
zWVIbcT_SLL$dWe~Zt-Fz$5{Ds*raSYiO_X&l-l}QYFlNY0oqr?qLF1l@ee>qx2c!-
z=>Vt&$&AW(%k!V*{c8Z6T#B6LZGnhjU`B0o{;N6uzlMbEZ^Xslr_F%RbH2oq4dne@
zyrij0$dL3EmC01rA*HpJA<W|~&ksf045eSb6}@NRWr4cK?wgY_nN)uAwc^n;5^AGL
z$>zsZiv?CbytT+_$*tUNYMt2PiLK`K=d0XzH!7-~R*KYbwCm2w(V{0WdU8O+GayuU
z)E!><nw+^#^+d<W)u40Dw{Nz83g200k5VzWIH*{!h5R_pEry)#|2PJQp=BC7=Q`aJ
z@eI^u&q%TrFnsdz;fm&`oH1;yNp&E3Y5Im-k`Rh~@QP)ai&AF5$w1nsXX~Qr-B<7p
zZqo}qfDU17VPVF2y*~yL^!ggS%DXOvGPB;$(P8(aFyUD)OvP#y=NLW+s5U`HbedFZ
zlhqP!&-je;JPS9F0C9sGbDgo8DofWNAn@d^Npe-`jCf}q?q(kOzWMQeQ6ATQx|w1<
zH-)E>9zGFFVT%9dad~m7*&(va)#TEq@FcVRj1sE^9S()~v~}J31{#t#un6jjHt8Wx
zGPi6rkH=z)R}h}6`l`mtHd!`lvf$YIcig_(($Z%lO}aB#_rxrUgx-_NCZkavyd8u9
zlIVE8RY~*eT0NWi55<BR6IC%`S=x(^8=c~Fd9u=sQ@tY@ZrI${M9Rb$COGX}h78+r
z95H6vetUyHQ&H9;oUA`-PKgSq?8JpYD26sNK-JzbHa+0&p0rfsgAxW|A*&??jkuoX
zoSMEfj6!>TQ0K#50chf-KD%h&`Qjuq=6(AI%R@PA{aRK928L}@({6N%K{g4U4-(N4
zmU*%|^ejSLM`;?QLN4e$=(x2s|8ya0(Dvi$CDz*OM(=`l7pNPZOj&me>OuMiDt<TQ
zXG#oC=St_f1J^hv7KX|(CQIxrkFGkHA3kE&a_{;{P`z42T}OtBQEcic6Dr%wmN>r;
z=~+Tc5o>aDZc3cQ2P>QvK(-qs-VD5s+(on>NmA6?N!QJyn|EjArS9NcIA*zf4L@7-
zU2ic=C%!0J(5nH(<Du6PjT(R!DB%QM38iUwH)L!tsZHHcw0QR=p?BWF-ksdnycc5z
znF1_SqYhHwXN2$svU%zyYKarlm*{M<z9GHo60!Kr=<<|@o6Hl!OiF|is~umWvM-D?
z?^A6%&`Vrj6gF#LVH|!?*`9AfTQwze{!w`5*%G@oVfq@#6;Ajl4>K}bnd#J!OFJ^8
z$>sLK_8~Ss+wE<kI+~q(>dYUX*s&xq#mXX;&D*sv8vo>vP<H<rstvuBfYakKAR8x>
zrzXS6jhcAI@+ZZ17qWcTBogilJHFq_8AzLcdy}GKkj4bXnzl9mC}es#-O5*0^qA0<
zkt^L|AJ)5U@)zgMjkN4Z527;DDD+}joT6Q%FKXQ?8)lcDqnMbm=C0pvsRas;vWHKc
z$FSuHQbHK~*mkhl$Wta%d@@JqR7rhYSY(Cn#%B%ww_fOUt-CHzzHfInu!y=}T40Sy
z!V4>HS&w)e$u}ygA22AZ>VJOsRY+kDV_>|sU<dOijBtJ3wuh2Uaq9LZF;?e>wwihR
z>70BoOe%>wV|c9K*pv$raKl*YItC_$+)@-e@;*K#_4TH*!97hQ-S=ExUfnigo*!R`
zYwquF5{;B^W<Qw~CfIqhK~kooSmuyV;F9`+Olm$ku^1(y%9h}vPC5Q8jlf;0=sITi
zDR4O0q;dhr!Rbqf$a)^;?F3&`meor;5j2Y{Pid<-k)dY2Xqk{m=$;LBM=cf$Zo0>g
za&p-#aCRF4!O2)fMdGOkCnNR6shu5gct?$pS+frNWJH3Pa^Uq!T;{5}^^x9&&r=VE
zu|-XU)0JxCrH8U8)y)Uk$6rJ&-cxziN`EU`Lf2#47U&cpO8{Zeu0=@pFreN{0r=X8
z!c%hbc0Xmu;v&)1wDM(bJoc7HGT*Wo85wG9r&{f#;yP<2yxwD%WuSfGHP}V2iNzQN
zi^F;{xXeh81cB=x5W<5__Fl$)Q~CxhG(ozhg-sJSO+EyB&~i!T$r~$<bDRpbZd$Rq
zab!hDBz?Il_Py5|3l{$a3-&ght9apu43jIt3$=Hrb~YDyGyx2P2JQtMA8&(AcL^BJ
zM;A4p98kP1viR&&@ICuAH#91$m*36oYG=XMsmbqa7&qV2pjCfV*G!b_#3e4!ZKC7q
z&Q)s%1vOZ`*ViaI*XlTIaNZr)f_VNi@E-eun}Zy-rRtp2hBm8^^n5IqDlUE{U1%py
z-}*sbA06eGvnMeD(oh)nI9ESQ8!;QxWAT74=V49VPT?`3^b96;$f>{HM`2qouPA49
z#LTbq=<_hlqNi_4w+CGpe{qLjV<Bi&vGG~$DeCgL)5@1XgOkiO*2CM`-Y^FbOjjVy
z;|?toS4fPyR6wHINF~S@hj716v7&2^M=r~$cI>&cxboiER2p_!A2zk840+8$y<YVK
z-Pkp>{Lr&nVZCDyY*zZl+&p_+s`bj2&{$W|_hkavhCZ|<Uz`y3-{Yr)&!P@c-C+$8
z9MgRl2=8(elR}VC1)Cn|C}a(dvzU(0ct9+le=?)tZ*pwRItiE5*SJm<UGGD4dzg5L
zr|fXO$xO0uS-ooel6Je8drSM@7Zn*E(I5Os-@1dmfn?T{QDu(>%D!Z5p_BLlOPMzU
zp_1l<LqTs2-^o1ExdMTpc?BRSBY_eE)SQlF^nRaq`Jh;`UgOdi=DKSxZ++?d%KOQ(
zqdVKiZZzcy!2(l5_d$4KxCL~{8<2-r6|{hWz-o}zqei!a>fAw@?@Kojzqo>bT7c?P
zm#<}@;%zKci9#h26S>*BjY}6~9Desg;fBCbnW3GDg+0qh%!6hK2;dOCj!rKh9P^~m
z&|Go1GbDllV5yG5nZ3DO)x(z6Eo|BBWsyg0JU@51z#qYS|Koh9hc8`|!9%U0xF5;7
zR!CU7njGS1pQL#^?&L$QvHjCLdmb!z%omefXSrZ)rP44c&Ryo}(pFwPB-=jXw^c1A
z!r!Bs^R9!c|M}Dcl{EhHW4>Z67I!rfu0o{%YIrzPj1!JIlEXU`Ex6MzEz`ZTx?x4Z
z=DlVsi}yWB5enp!t$^C~z21|?Ertl>C_r0v$PVSM1<eJh2T7SOOraMzhIA5v8bBA)
z$as6^yqxlD^E{E@IGo)g7tD?YJ1^8BLJh9H964%c-Dbs_@#ITyS=NFh)Oflo$wQOc
z-G;nYYiszfs4X>aNMz}VL*_y0SMA;w*<B*E;9sWke(QnzM}AM$_z^}oWHDW!u8dy6
z9?+qG7GDQ?O3vluuv)}w3srzyQ|K@ld=WR(Ljp|9`JF#VKQCpkH&`K9kxDYS!qAV@
zre1i;y#KgfQrj6Twe4*k=3K07s?hXRmSE=yI{!;`!8%pmc2D^tSkz~*`BVr_JR|Qy
z<&v^&_7G<^lZn-#ZbVH@gvZ@ZeLqyjetbu={z0-!nXV_#?(94Hao4PGhjX+_)-6i$
zjh1-y33QB`w=VVph-smh5D(gxSH6U15!-<eT1Kl5v2zp%zpeYAS;PLQF?+|-RUCT7
zl4rY|@F`Ee#%!k@+x@|YBGWqwAzqtmlC|AA1B(}=ne0^+eRNpehG#L!MGVXot@}fI
zakd`)Z94L2oz>B0O=n3qhMMZmSHrTm_1Gc@%zYg#K)B+%2Q<+L^~z~TfbwOo$@{&e
zDKLv6#HJOf1<#rMb`W~}+2JiW+B}!R^|n_7AH=TaP1;mgbJiB<tIO{D$8q%jeEni*
zlozyua|LnARoMXu5aRp#Um;ZO^bsx4V=M98Uv&6MD}TM#?eO9^t@XFchBpP$j9nO#
z%`UMJ$*6_ExCSa`h#*s_)^@4)Mu0OU8uCS98mJ~5QmG@1JO!0>fA(mcxHYS}V<Bxr
zq?jH{*tJl;Pn_|!+@Kcam#@Mfq*&1+p{dAe?vAx=$Zuba{`%j^ul~@%+bG_?R6VgX
zs6~ZhTN9Gqp4)y#pg@4BfKctNpoHYHaSl&y(auY<Vw0;GeTx0IqL^dJP{n~x=1AY+
z^Z#mT|HX>`+rxI&QM`2%>$-v+S6x05l*9?0wR!vNV8dEYHkbDM3Oorh8r>(H4R7}#
zI_nc>$kceZt;`AY{h)+>r6``(58C`Ld3xxtzBcP-SACtR>fv@qVN?6w_H32l?H$QY
ztw3T|ROvE7gT1>ke3F0B()@H~d*jLK*OmoF1l;u~<34L**TvjyK*WI%$ErMNIDSlk
zr91lD8i6dOuRG2j)M-OMJuEJL@jTrYndNzJ3E@@semDEi2$6B_PLDOS+NXnwyC_i`
z=d9*i3e|kwmsxy*w<@_g&6^c{q37#f0b}ESSqigT1{|oL99&S<<|$wu-K_EYQ=6E*
z=!G+GRuePjoWJa&iS6jFL#W{pCg0z@+Y_L6HK`^QOvoE4SD#F?obiq{!q=^o(rE%3
zhR1{bp4%jj07T5!JJSB94staA;86@{Xv1UU;_}=~CvUw!CvVMY;zYtjkY#d!1h$ff
zaBj_I#USOu9mqSc4sCt&e*0<~%VG4lg~LD{A=3-xnU2^`w)bKdp}#4>C*O5q*4jUO
zI8l2qIqIgTr+9k7QRC{pS*nYs1no0trcbT3_5wOQCISH(-OW>aAJuIBkka{fqB%ON
zHyzK`o7Kc`zSk1D{3z?Qa&Dvp5bgGnX||6LX!1|vib1abeye}I`v29mq+?&Y?d`kK
z8nW5F!XW6wQDmIiuU*qM#J-8azm@c#ivQ0K%1~YO*0Ufh;R)P86j$B((#wUGxQ&@E
z8`U=TvMZ_6n3~QOa`f9I419~FoAga0e$x&w_pw`<S}!8JIjibKWzA~t0onVrhHHat
zET~jjDXb~q$zh}2P};U_y#ccCWUEwnfR%R-0n^bgf#Sa3_1Gg#fr^UXA#i=UNlwv+
z!pTY29MU&gO>AU<Of?DMb&R08kNnYX^lb%<=4?d$YsC?^6<2n(UqQJyWuCT$CMcf~
zdFcBH%YF;N`lJ!d%xf18tzb2ax;rQn@@*@>w~ZbZ%%+8O$x8ic13nu9B;V&R%2kg$
zOVR(d*672QXPn#FvK49ls_TGx1(2SI?qADxIiFk>Zs(%NC6;vNkY!FGdT;V~mF?sf
z&$yfd4a7pq5B`<%{ues^-ySv&ioQG^dCe?`^oEz!@5ok8X#&{m<2BQq{pg7*=sv;$
z^?M}j-+te}5L1k%{KfG@by|-W71(e)WZQK=?~87FkkZ--YZ`q<IHAW^YDgUZ{y@H8
z(*FZY{#$7AFHU0JFTr1LXZ)`G6G~a|diRU0cyXDJb_3{h#c+P?-(t{NgCpC`WkzYT
zHXem3`}f$pE|?Njf8Ne3sh>wzRX?$z*6UPHbc|sB1E|83)DMmG8|UeP4T+|LCe)^g
zwr`f|-@{L}gE~RS>XRPo=Uy+1?G@mzeL6E*We6Gwg?d+|{4yTl^B?LeAF!4!i%eqp
z*h8&o8W)%m1@Qtywxx#&GIwGEI)2FAWU~W(?kqy!0wnBX5(YHDsJu290mKPLpC|48
zGHK|WySYx=bgyBPw+e0<lmK${FA%QJoPE}txhbx*AZy~=Ygj7HpE&`aXK@6r&acOn
z(@FKI0|$S7{Uz7&J4v1LeL!N4>92zmt=V`EfIki*GMykM*IStv2~dUBz%OIB!@dll
z{FYTEm=zCIYz7~8Mt`zhnmIl)&2bt#T+XNh&4T7-wqSlmX@VuQ-UzUieElfPdskf;
zBnS*e!&T;C4h2g;Bf<%EtIvoujHL$qT)^d$QRJEkqXBkEIRW~iDzd!!c9ND#00v_|
zBhEv;PT+f_838`bI?!s6_#>Yt8c|NWkB0M`o((w8hCJWyP9SFK*>&ssriegU>C*YZ
zu8ekOqRa4orp6CA9xO;LG!jhJ7ju}v1KZi+Sk|ggkG3H#w`ty=4_VqSZphrl-uS%i
zMLOMio6%7Yoz<N=RUYvG07-VBn7kr*m{T8g>8#}(r*!ZtgB%lG+(M^!99qg<P12K6
z0Me@u(V!qm5lc>0UkUuKx~HI#6u?gmb2t5>=qnPcFQ`RrW@h-Mrw{Jh^5U^%Z0o*P
z{Y1k{Njyg{ZCe3O%f4QJQIJ5aZ9<WR-(6E>0^{;9L{hr{tM_b2b%A%*GDU#|f{{aB
z;$X?s(ZHpw)k7SQO1~AKyk_oN|8E`!_<cvrWL+l!J47sp$*>cz9u6)*Px+6<@bWss
zJF`%R4BwMYnZP4oX%_MX-kEAW=MPh2Uq(~t0HVSVbKK?{JJCbn@rKnf&*ORy$-oKV
z1hUbzUQ+dwx%{A+n-#-n0!8~os0PDF_kq9hiv5f@LY)5T^ikBP44w|o^?(};g31Oo
zk`eS#pUGvwxnw@i_@5u$@so~Re|}KioTm?#?jI%l-4*`-jglRw4@?NWa+_&ni(n`-
zUqD^`9(Lmm(D+9?>?$G=n}uWC*AFU)z`>EpiCv}$*#?*kWItMwk<3653>Y$Oam*r`
ztPSg;y`jOQ^#m|?4*`c{K+_9B5_EyBa=O#^m%kd$7V-$g_W^Ayd<mGXIc{F5n$0?W
z)43K|HVdSOT^uH~Ld*r2CFTT>JBD<E$7V=5?`;9|>rGn1VU5KQ@f2)ybyn}-9|yU8
zx#jP@dxP`~2!`h2*Sxy3-hv?;2!}&o(@@-md;xM0nZd#<%>^9`gn&}80d_Bl;d}lN
z0uuqo9qpV#kx}5%;)t`}8E`0ljU9;-0-(+&60?($cqN5Yilz+UU<zzhGFSB+Gy_4R
z5N120UnGLW^ngZ?JsKh+Jq%xa_1J0d%4ECm3b1x|Dhi9awlBJK5=-l0YMz^DbUg^8
zEy2}6jRqWBW~n(N=0SKT20c+nxNk6*;GTuNj~ln%1ymAK8_vOx#Vp!8|H%#wqq~f-
zQ^1AVIV0W<z|sS&m>W%79K*M2K53pg5uF3b3yZp@HQ%x952sWbm!IY_W9q|~P7|*Z
zx}-tdVC;a=`JWur5S#e=I3=C~(4$U-bEM%*GTjHk+&&aw4UgOec2w}(7ZF4MtZBx7
zLfdsQ+$oanAup3R?`wqDxL!TKcEUMKF3U_`ebQgFeAtNMUHnAc6%VLezcl7;sP{oR
z$j=-Yfj487z{mp~Aa}Y{fLpQ3@uPl()lb-HEI)iwo1qjvm<B>@d}xLc7NSJJ(32Qs
zFR2s~h1sc+dbLiFfNSX)J5_z_tmgB6ulgADQwV?7ZSU}lwjD)gIH1W*P-+cRU>j(_
zZte+2vD=RMOlE+RBTG6O%)ENQ318&(P|-lueoCgF@_KFhP|Lskd3kpz)cKOygPtrj
zmj3Al^a0-8x9WayT-*_6%4@^Sfc}Nfw@3oT=fsmMUbT(+^sfu^Z6i%fTPy@)Z8(6`
zJsL1e|CSz#F#3k0oh*_7b}ny3j{t$>i6An3H=tYLuyE5~U;YZHX5_=kLf|tZB!Tb&
z&Ui&MYS*XIBAGP<8p5Wt2aX$k)P;9$#ZG{YJ}!A^Zq7(@z6!h}BAovGGh#CsQTHpp
zwq{#8U&LRy*Gvlnf5QoVXJ$z4_cy=p2ENxXv7=L;5obWV+tY)IL0}$YJ^-s|X2c5_
zgzO~>h5=~;*)4}J+vcb6lwv7NA*LgWwj_q{_}4GvYt^3?<fH=>vI~62g<16EH-F{Z
z?+;DrM)24thBN%lz9Z|e{zic$^C@%&c?HGlB~{!8E$lTFAG&6fNj#-HaG@*m&7JIH
zcjAK_Mgu)^rtp#I_Ab(LLNA6)fLYCfqhyZo%4%pB!8c;Ma17PR&aF&t7-+#lT>`9|
ztcQfTH&d=1Xc}iKLyM@ECvvj`q-1T*3(IC7(6YZ$jTAHB4D^~i)X{bT-vVYa#+Q-w
zwthzFHN72`LkNOPUjTKbQ90o<UfG#Zt@w?2eX1lVc8e}VPj0lq(lFu(NW1wPkp2f!
z0}0`qWlk`-j%eyc)|Y6N%sdD9A4m<HEZY*&DF0`KxBSG@Tp@;UG7P-I*+*me+`eu)
zq{&>XlF&yQMFORmqBBD-<a|*%sLwwsk^N5?83q{`4PgB$!pSXgfs=kWua^#H05>y6
zeK*06G%@xRT5CJH_lspz|2>nM;cy0UPFD{Up(hayV9>0s^Rv8p2Yy9#|Ii?R2Myw-
z3>9gjyV_O&E4d~F_LcJx$GQS&Fz=WK`+0Rh*W5@1^Z~vF#}v+}!R%_6pB72S(U^ei
zs=KCv+6Vumgi8tMT0m&F-h12Auh>?N=OwUkyfeV~dP+bJi3jW>ML<7j!8>z-g`ORV
zkgXq>{dO&z$Tt)HU#Ns1rez8pxYN92>C?1j_TFnNrMzY8EC7_`DQ^>E3gmNwQD*!M
zQ1t5C2}UUgSonNKT)YY^vh*opBB7TQD*H?ybP$6nECCVOepT7dXul|x7dZ`iOFDTs
zTjZx-OgVX}+Ud6DRjvbF>)PxsmrtznRu-9grY#NVE)<9f5|v;lFi875o${mqHJdU4
zUR{1UBOnp(hkAGZ%%^Me>h@>4YXd}<#C(mUO$PsXK4LAQOAa@IrY%ggkOEFxG=l|7
zHu~8?gk$43=dSENaCqogT30lUD=|!eKlnROJoB7BbAi#TFRbv-%F_RrYxz9RlIV2(
zBK)Q1tv;&aoh>cBPHgQTx44$(Kfk+=W3xzTG4-0(d%;Sm_?*`p5FB_*AhQ$7ywwvQ
z766vbuZwvHc;?cHP#X^vgd%^4P6NL21dhRiqr@Pe_9oK8lAK-a7VJB$abV#<MHO&I
zf5rGuCyo<g^#{Ot0CAurO@yv>4)D}?Qf}j~FXwRevQNJau}?A1#dAaDjih176!Ef=
z&?V{g8DWkEuBy`Blz!~!l3ZZE@qnuniJR*&*a6$VsFCn7WICc`Hu%>NzBR89MX;q6
zs=%Y@35-&QS+5Ze-w11Gi`ks{5!_`$_XzwRj7NI=-hkzZPynHmCQti>9+xy9noFl6
z9KWAp^w-IFbEx#WWC7q8IX*py;{k<@cS=&81^s`)Cc)8ULojlh3q82-1h{8iEGMQF
zJ^zpai#hT#q3d)VVPLWa4F8aGd|OM^g|(7JVS*iD{CF*A19~3&LmT*ERvVaF4HcoL
z*TcX!eeSatXxHq4H$RX>8mPKS@bp)zL{C*&83%uV`Ri@XVZe_Yll-ABIb<DSoL@s2
zP~qzabmlIbGd$BpLXAkT`4aD$O%eT!e?15EIxCVkNP6{L)*d^J0J_aQa@v4n%~w+4
zTYegHE)C?60HP1FsTACYOd4T4hEL`jK-@Vih@Ijx+lGLA75&9fg+(XKHgjd;xpUqk
z%4Qhu8LBi3R9U%!$DRj;0coPf!)ic1x1wKx#FK9x-OO*9(io2Z#qL0QmGQ>FjqDso
zP60Za^Cdw7ZJ(^&u3w5@%ux3)TITl9dbH%yC*Am<uCbLm3Wu~b$_AZ=EVMr(kaBMc
zz4zWhZD*nxrwdLn(*dv0ljj0>g=Hj~4zEX!1_~fy@+g@0eznRU=BNZLKw=iv1|(f0
z8dcpX{Ke{A*<{|Jd!}%l&Cv$A55aZGZxWZEeY|m@Xj&k#OV54lAn@MSOtsfA1C_Dx
zt#Yv1ytLIW#3f~0o>mmHykj$xnT!Lg|4U}no3|ogZc+9L*t}k^f&cp%@g&6}Hw5h8
zLBd20AR^1&qPx~Of;AGIZjgiCwn9X3(KMF${^-%AIbX=PzT1TU5%fSa#6;9`j*iR_
z?%0QDvR=sadI=dIiKhmD51DK`mP^<TwMC*vO;nA*<h?2)9A;X}Ibc4s>~?+8qsT^R
zAObvFu=O{?J7^&wy`|@8JO2Yz@suze4_}tOGABeVra4c5*uRloNqbwq8m%ACXWNS%
zU6w_QW-fS(=Z7lvVFaP+=}(S%Z}0zxQ%=k=P4TlSurFaE=<`piW&dtNRlA<QCPwR-
zRy|7f-rCijI1Aa$bQO@k7kwBS@dJ}Z5pkO4a5Wg%(S)V96JCZ44()<H0X<nD3dVVl
zg}~fEf33n6<3ODY`j*o48f*%Rd?35azlg-!8OFUar*TyHXz<EG+fnu3@4e8W$zcGA
z0fvDl*Vv$F3LOP2@W4B^ADKom0RA~{4E`^6gKP))7h#0slxn8AP*cD`+#wiZ28h?o
zNcz>89bEp8feQMw25K7cUxCc03^g(*hCJmz5yD-C*GaH>eLl6+S?(pGd&_mhw2Erm
z9yIlW{6&Bs?g8GZtpcJvMKBjVB!CR<_O1f3W)B^J*78{i^b<gs{-IP9e6^GWFY*T<
zH3o?QZg2opSqN17QC$?1+mxzYFbIzM@6SR@ByTWuUY{%I<6C`J_?1Hhb-y=v?@XGZ
zq;U|?gPA1SCfleKe?ZnLS(7H{KoiF<^H#t6$D$#4vWQZ6XLxb;Jxgqfn}+1OR!yJ~
zGr!wBOc`1bjYD%WRH@fGPzx-mOM48WHS<E>4=-bArQX`Mp8xf>n#Z=_In2Ms+E=n;
zHL!$k8$^>9f~X28#_4>vcFa&+8z}(2L{Z+q30{!X&_&$9p8G+KdgH`u{?La?((!;0
zPFtIhVt69^wxY0+Gg~k3E};GX2|xdwD@)u0QKmTQ1C0KR!)bs_&PJ1fO+fh&GNT#y
zrh-2s20DP*)>(-Um+eQl=)?Qn%R#Oe5Sd5UiLY-VFj-zz(CAq#A7M}}9*FV;ehJ>5
z8A0f3D-eg?xdV+?2YPOpc^qeE7pS{#?I;iQSo08TA{8BbdaB8}f70*;SIpfTY@537
zvD|FU>x3nc^gSRcU>DH;g<Od*(bj}%dBPIH;IT%)UO(PTi9pa@=6&cR=&*rn5$cRC
z)elw|A`4v|S#McCp|inbBDi|nhomj_RmJivy4k&qh)vM58peGz5Vh7u^f<{;1C=GD
zzLu=$qVT9-L){4{=S9g+G?wpCy~J}?+riXLI=Dc>hKO7bZ>-yMvUyAYMu~ICsZvPE
zOBra|mJ-zDk{96is*WThAU$HiMlw5S$sZr6sF%9^*nO=<ZP8OtTf>v7SVQ_55%d+s
zs{IZ&Di@5yZ|F<aUS8IG`QD;Y+i5>M2&A_~6Go3XA?ZRB^2~S*6yHp838oiRE}npO
zIRglThXEX_Pc&r=<Lgjh1SI1&QV@EV1^|DYaXJf)I0FV|xt{;aBKbcl9{U^Ye7Af{
zNV$aqnX{0|W=z(~a(Q4;gD+oj^YEM6J$0hCerpeNt<hb(ZGx2`4YYF{WGzoyn>zw?
zR?UpQ?ZYz!v|&AZF*K>52{uN!5j9Igf&{8@Wa&GgI!z)(R>#3IAygY|GoT+f0I9h+
z22qSUKxzF)EGkd8110SB3d@$K*^#W}X?EyE4h4DN$x%{QG0MC>$VFx8{<j9P?^v+h
z#P_6S0L@mbvq8j~tLX8Z#Li!^CH>luK`g9;R}*1$AHnJ~f-V3Uwc3?<|1Sw7Y(pR>
z^8(fe11~_xE9?=}`#OLMstK7d2wf0}>|Qn1h6A(lSAGHRv|B`U*9zzZ;0TP>Jl4?-
zyqd8~;6)zC->R~-SMlu1f0lZwY4`2P)(g^7*2Y@wu~ABe(ri5S1)e~m-}E>M{4s1X
zin<VpYvp+sT8pt9)Y*>$pq|);nF+5*gYb&*On3zk>!yShk;sO~YXG8NOIPC_DYg&|
z=G{wcZq^EJ>Q#PJ|8&y<`zs6B%69Z*AiVIl*iiwnd)#Tp=C%WQr-{NKlZ7j3u0QdA
z*kk7G)@KOdXO<awX79M5-YI-l1QSq2m%@N|2(%7FXF)P0CZ^MVaiT$dyGVo3mDNX>
zl(!+!4-|-f(M>VfQL}LpT_~NzOu+K}Wll7)P=0y?fe#uySqBt*s?4bA1l}7vYVIWl
zF19lEGotV0A(nbCp!G)`u!{+U^1wZMfbk|0emGRig4Ob>M2()a0FdZeD6m1i*zLr+
z=u~aSy>M^z;A#Sm&46>Hs)ylO1JfbpWMdq<a~PUxnSOg3A&V;}j4t!<CD9fF59WKC
zhBceO#G^Qna6c#ICxui@8s!Aa_yV}(B}X*pI^~Ft*^cbY>shY1@fthe`K{nW52Qgd
zU;y*nZksGKS%2UC?v!~c=%wG14SOwwW@Td-Ra3F>j(9kC>!@Iz0gHuA2XdO{15^l&
zFR=)p{Z2#p8~^7FuLBm|44MiVok{W(0I}v|0=;}1Gb>wBFoSt9p$#Pm?$Hh^1DXF5
z=wW&0B%xD7IO9|etlsv8t+EDn8t8=22nT2+1HjRUmYt0!vIDd<U#8w1!)th@vru{I
zHekosxQ@UDa)j|vk?C^<u|M&}+1*_R-+o3&vy%tl;${dAF{OyipWB<xf}OkuV)S&1
zC|o8EAjRI3_~rOhKN-qvjld^;MjVTtR)&Dk=Rbk^3O*I=`|+5!x-*5y2yO*x@mUZF
z4nEy&`*hv`aH{#H8Ow+H0Fm`c(;Jlinb&CHa6=5z|I0#;qu_-NPi%6xmJmd3bF0eV
zpHT$<V5^?Px&tq_oyg9*fa2y6fBLA6RsS@DickNHIDvgZ34u#^fpWH~<?QokD1{7y
z2l*wffdtYT(CtZ(C7!kxY|6F*`e#HSacmllE7yR?P3TnOFZl~tpkd2w-+GBK%&Gd6
zox~$qRJ#sMVq8Oe(6EtRyw7Wl+_O(7_jM)oMTIdMi2d3Oeiaz44%`4@m_O{VhVY8`
zDp}t%5$0sM%mn#ErUfxfunB8@f93tYk#~aM$Z3E$4%2U=CWgd+NW^8q|1_yO@ux=-
zJx{={nR%cSK_D$*k}y$;9$$iyiMl_FwZ8RBP5IC2C2dQwqsS0AH-yQa7&jXvJYn<x
z&O`mKEb0!3x;BInIR(7?!IMA)bv7hY><KpLwuk6Z(0dFJFQM|x)w%f6+%K0+SLq-z
z?v@t<51%3phbwuUz*>2jeMa2eM&$*5*GwgN{U?0jFIIS1zL^N?gae-qc;}Z<v;|7!
zA<t5}DcCg%Owf6F(7*>Yko7AN|B}Fj{HZVhGq3-bkEf$?Bh%Y!KxSA9IM<`+#L|gL
z{~;#WY1=!Hw220w!EP&43%t39)q(Y`)FDSReUs4R4`lsQI@?1Sx1tvUJ#1PG_S#Z!
zbX>kWypxDB0&<{v(zqZWa9g$^AQ!54Rq8wq`F}dY8x@8oA0Pm}1qpoemB6WPC2X-B
zkA%4~!1D%}53K0|^te0mQ`Hv^%d;F%j-bRKl=K)p>wNIf&|d1ff=*zeFG7{L258Wt
z0h)~S*#4KrI|r}`LRT(hJCNI(02kC(Bv#rs6|g{+PE=rCPe_FTm*R#2rw;uf5Q$Qc
zoj^P0*TB9OY!=KP4r#D^Zl#NwX*<#NT)EddX5iM`3^5t%cA#CvDqtF+=Q!X!UPKRk
z>H)h(Dh64H=Z9MX6j4>{&@<B@)T(N(X}as8mXs{!?6)O!`EubDdzR)s+bTH46Oph4
zip&Q4l0bNYl!%6P@Ra}$qDPHXfwnxB3@#k#fAtt-Gm7d86(VUu9=e}FtHRHSX{mf{
zTl7#6Ii#bf#BOV0lt5j^3%&hJY3r^$T}yAS<2Ggv!7lbjdxQ^Gd9e}Sqn@D|_mRH9
z#SM-Hd&JRzb10!7<YONEKS;hajsmgs0n1*|kDBD0Q4dBQnpk^)33mXGqYHSaoWP#b
zWC~WrumTbW%?Lwt0I9VCetP|#ftrC@b$;;2U|KuV@B9dJU9h)W$Rs$+%_s`U?lgn?
zRx6EmdusoH3B0-I8~cn7r6E!phH+1*7*9vee5aZ<D`5yApEXk*TLI%94d1G?*lbZn
z2d-A7Cp8ENE?+zlTm?)ZfP#nNV)W>Fm=pK~fSGOZtn#?~<1vKCP$zy`JP+L3F;N0j
z>@7&1n|A=5Y6cLap{{K3K;-+e^!;A}@z0#~X6QNoBefhTa*j06<Zqzk6<vQ)iB15j
zbx@-Tn!JzjIQ)gPPe76PjZhiXaF8l$jG~E6OyjWgkBzg?Y-T<@`)5|XI6GigV(PE;
zEZ=RoackPtdJrm@A^EL_O+YN+Jj>PD=K`-e3V_tFw1pCYM%Oc?Y-L;qID?0II_0N_
zt=Xc!na?K0k!j9mfFOV=tJOU6KVIL&R^_#YVT5K^=f^U3`)gk_jcOWh{jg)b-<nm<
z&HK9-xzaF21Q6TN2UMv)%&7wq(m)@s#Rt~&ISjPfeSmwgzKSd~po@`m0N3>1GlAO9
zM>Fh0s2A*k6z%RP21@D?2s}AWLz)P$2VM;Xp;MswD~gIKzuC9R8oxE*4azEDmrRIM
zQ<wdRB7O8;Mm|1U)$lVyiluFYE(ZGQM|~<l{bGgF3gm#0-tz!k1s}Z>>T|@DDNVYR
z5~eqYc!N}>4t?f^)6}xRo$aUZnuNh#yhq_=y7hq_r7P}%#aerjNTZHX00QI9(LhO`
zz+dY{q>Qy6Q8?nEPxj4|EzofaQai-gowAtah`8JT$KIC*LfL-(j}&E_WG6<^qoR~0
zSx1W`NhPVoB*~T}$u``UELn?CT1KfTF_fh&GuczttP#T?J2RFs%+l{tPtRlN`~5!8
z`+S%8{k`uWHSXIzxBI@XbDi@!=X}n&Tnmviavi{2r89x>K5+syKhIv+dIXd`QK~Jf
z4BNM|JmCdl*vyFO)i#L*8N$|@@(#6+Utu=*wF9%(!BevOCmfII@~cGaio!SX2pIdY
z4xrS57g0sM3CN!2slYx?Bj{!@NFULXn?$GO4Cqp}0{H}mNhT8j^K+^Ybs5|qIZXD?
z5ZPq}dgWSzkQhWr&{-`ZqQRjrtzPG=q(e+A?u#GvUl()R5B@L#-k!5Byj4ZmN}9J{
zXGfYHg8UWc{3A^cLgI7Drr+z(>|@e?LoYttvgBu0wnNRpa@U3*P05|lr(L)*P3KDd
zl;x)_ndbox2I}y}`6tX_sPd;JG?;POazOgn4t&SyA3tmVmQ)juo0md25d(}SPtcqe
zqA6bm%=;BYER`9`d=tr7+3Yz#yRN;q^w*@Ecs^!)GkjW?83gYCduIK62O$9USswDl
zC7<5+Y7eJ^%I@t)p|gXSiBZB6@5_L7MK>fG^4KCd=bCLarcL5Axvi^ca^t)3mZu!r
z)5E%DL<XW!w^5{5-qKKQGhjQ+p%)=-#K-a<vvh<J`W&js^9H`g(rHC3`Jx-puKG8L
z)PT<0J;LW|NxBdD3cT%2+VI(18mp$k+U@A5oE8$2SWzz>r6-)1_i|ZfrMvV4J&kA-
z@3$EMasawg&TN@=WQ4XlG8?c#vZF}qMj?7ELxov7q5}Gvn-qfp^+IOY>8VBBu}qf9
znvphui;cP+pWgindpcw{BZ$jglnbe4G0y_jA$@Eyh9;r}G2}w>V+-s%h3IjpAS6Wr
zxWlt21z=Y&SxEXllM;CvC#VMLH||s(BP=?j1X}`*zD3qs07KaGEoN#02<YSJ!|V}s
zps%8ybt7rqcF0+y4enDO@KVgo<_C&k1vUY=0_+_IB!SO)zXc(ZD_TMzO{^Blyjg`3
z`#)eapH=XGz-A%R`4<73L_~UDcqC|!jui7=r+X&;R;zeK%;rL=TPImOZyZZs?&CJn
zidP*P8T*7~8Z=WvfE1fY&t;@-SX=O2y}1|$Unrc;15Sn4Ew;^g`ThiW1AK{vrpOF`
zqG{ttxVw;R>|MRZ{ddLfeUsz6S5c|0<ZXuZ%e3ubCKJ;rEQ(|>HY0;vjT~NQ2~yn3
zFmK-<fsg+=0gr!aMI&o3W)VhGrlry@K^5utND5y%FoHv*r{7^pCs~(%o82Ej><q$V
zA3fu#-ngZwzF%5R%>k&$i6q9vp#c(mQiT%!Ph>Dg;B7SKA?Fgr)DS>;19C28FRzC*
z2&$+C#cTyvzQPzAB5?A+tx@m9&|@80pcV)?lykf;`+D3efZxuOgcw1Sc!n(SYQ!jo
z07=D?rx}nu^Lhl%9{U7&0WqWMh=7{MfzH1@^d%Q2+YX&a&g?`5K@_o<AVuhxzqz29
z0%R{bfa(f-jy!pW@J7~Iz6NABv+sMQ`~B1N55C0*1b*+s)GTXe0_xrd9jB40Fbg}>
ze14ya%Y3M^xZ3Tv8hKFdHWx<w(Ku;IZ5qeO7_!GClF`6C3p|Y9e*T@EsxH3@`aod3
z0<Aw&UZ$zcuk=ZV&9CN3qr8?EfXXLPQ<YHvJ|ESPUr9ac^U8TfdO!;GFqDg&RQtVl
zgLT4{$zcH=5)*(la0Fb&SJ?Nb7$6|#cN=;q(`8t13Iy@R(0J^3@oHkE0oU|6l@GXG
zf-!(DQHnM~*?Tub<FaTlZ5f<^omY1jx9K|+2h!TPFqbhG%v}f|Ho_K>BHj!xqD|Pa
z4qz4+^LV+?t^z4(ptcQw7|uX$D^3Rp(_(PozQtp4&I2}pw1Z}LVuMi>p*P^wQr_AV
z@QcoXFmy6_Kf;Xa*fOZOs)iY5fZDnQLL|V&sbFF@7?nX3(FO=4$|nXV5ldqx%Wzc;
zAEGVN7UU*#yLn9q<-9THVxT+Nu#^!LRc4HSuPF$4rXE4e=mHo>s_;?8WSE;7V1(JW
zLF8i?QZ!YF8TjMam|p#XnZ<mXm;fKR%UlBi=b)DG2<j|@GdE{uei(xfMQf%7VF%)4
zN}MuN1yVmd@8|W%ZjkOg=RAh0)byf(8Ym(_IU^}hOP(h%2C4n5Q_NpEhs*LH&`VTe
zh|ve`sx!wK1f52wQDext9y!5_5-70WTLuBnv^?Dhcr|HN^yDQ)Z0tz{RC3yfAI$jI
z4WJxY;FDX<sHqof_AtA^uYNN7L$}{^D<<k)&<xvL{ro!<9qkqFNx{e-d8cpL(^?%5
zvxUbn%OBaA1tiCjg`IeQgwVaWK}+H**c_-c0bshzAj#klQUwfwIlyb~OTsV@VEmzB
z#UyCppiAFWKLQAO&oinDvsle_j)f>uz9hWz|8dG?1liI&N(VYMgB2j<Ru{6BR}xK>
zK?s8=lJ-d2As;QdIU&L3;Lqb{R8Xz5%t8-CX4o5Su>H?K2=N=8pWy+NjU!Z{N(BJ|
z3r+-sC1qR?Z;tQ0OBg@oEeW;k240W09gMN;hXBr}N6vY5zc)$vNj~sLngr<LUt=^&
zQh}_f-N1#+!q3Gfeo1w^%iIl-Rxq<nes(sKrRImQ4T;WcG-IqeFMIMmi~p}G5q=GY
znrk-T1nJ^r*Ah*^>Za+#b!W5PEc5rd=|#B&kFI>zE<9;bEjuC4lrSI%F#}LsQ0JCr
zpsEhD4_OJ!$N;SO3$+;NdTbz)Bsc`L0Z?~<Hvr7QLleLo;B*11^%x69WdzvWUkX8#
zx8)GHd=IuLo^gO0Nv<iuBamk?{8VhInoHna7fYhPzO;dss#JKQaiDgkbkv?a1>Sqz
z!bJR*U(+sdeyGC7O`Tb;9?6h-gAE!QU3{kh8)JEREEF2IMzG+A`bsdfryI%PUiZn4
z%!A#XDhxC=`YTL79=H8h+k&@8&HRD_8kiBX8tPx&T%0v6G-3~`$v#jZAfdbAjAf$K
ztKZGm5}t+Z$j|r~-!HCZ;PE|ubN#{0b0KC_IZCLx6_GPdzt<Ox9BKw(k&h>ah_j$e
z!6KlNU$z*{^t0CC7Xy`9qs3?j7_P-YMYbqqF`V&1KGfX9M2<p3DL_p3J!`C*e^mP|
za``)8xt#Sokf~)8eDWJJ8uW}bc0CHj{kPsf-?C_P3Bb93^P?%Cs{Z%Fg(Zsymew)B
z)BzENcEc}ojOE`Bezz2?FXtB{FYdY6nGbSXh&rfwI}kh3G5h(Y$~LUGe?&iXqM16o
znEk*Y4S~NhuTuHjWX;8skSb8;Q^fcxs?lT<h@s1>(Vt)&vD8gLbU$x?hBBVPQH|g9
zT@XhcTgR9>j=o6BI#BH^XHu$6$r%CYaDC2O3~hBLmLVuhjQ^fi1k~XbV9DSOfGl}}
z|5{^6y+6+${j+$UOR_>HQO&7nVBn*MNPWyQ0|XHN^$9%Dlq3LZL#QBZY~hJvaOATU
z(3YI?;h2r2YysCjAjvS$ATTl@-S{mT0JuzmAi&4OU4$m1>|kWV62OApwFK6{WS1Zn
zv@y!=d#wHo*k5S^Q2Xxpvl0iOCVWGKi!^=+HE5>>lN}>aWk9^f!IW9rMqr3LqF7A&
zh3I^jS6NfXQi1~yP_Is6%7~wI2SM1k;D{-P6fsDccn1;mnPB^B^%SUP9bi7lQ)}78
z1pruEIhy(vrV+%L#CIPtcxZQc?}WbaN`L85arSO)k@;W+$m0Uq$nW94A4M1|=21gR
zeg`s3dGwm|TNB2VErc@v_98!4vOkjp4gzOV$xw?rP(^jdWC5={kke!bbA7V`IV=u@
zqyITI#D`-N4iXIG$L0!bE$*v>1fC@piPffpFKsX^#8X7-@sm&0V3Z~N(wHg42)K!i
zW;cG8nhmtwe*50{FxfWzdF;$i`5+WUq8_X=kLKWJKtn~%r=ccrfIJL}9!Cf!P__+1
zAcE#r<*BkHjRdlo)-*04hxJqhV`L8~&!1Kvw_0*eiDe=u0^f5L1kkU6oY<tZ32?|3
zodCBg2Y?o&*rU3^h~S?=BWk~<bG^YIZ~_B@YQWE>f>h^FCd!sshIj2oFoH#EnDy9D
z*+qcn3Jck4F0SiUKvM=^&t=x3D4ZXmNvkUK9+aauAl+WXXPPklCNT^VXKlplx;fJ5
zkjyY2(Q%a<duLDyLWD{%7CjUxkz&pND)Z<8DRlv(-Y`Y+D#^7xYRx_EUge}Q=0RN|
zo)MIZXNWj}%}|k~MzE`EAA^5Co`eE`e2#Y>h-Jhf`yc55U_>J@l~FbA{49)RP_YT~
z17`Bbda&-i#sC8ddy3>3uE9>0DLzWg>H6|I_sd=G`jh!7%hLny964uI@;tRD_g?qf
zsUcJqH*nDkH34@0JO$B2Mxn_Fu-j|({}dqgFDZj(Eymk9&g_5(<EO@~=dvr!69cGP
z)Ckb^zZ8;%m<50(0y+&by}qNl(K+Z};p^7MEa!A&@HwF1CZFf>uhGlMIgAKU=JZP|
zPU2T+vLrM0|4*_Cdx$CnG|1hzC<+2(n!o@}h)$0ALE|U$fw{C`$OA>wodDm*NQZiQ
z^SuBA=L3imSYqRc1`ONSnTJfv2~%|Z>@R9FX+9j&s87Lsv^{7Im^D-8D)bn{Y;Zx0
zXsD-~35;FKQDjd8kb6||Cfu8K!v0bet*6gt6d@<P6e=7e1dvrLvKSK7u<vT;E4$n}
z&}yFZ>{Q#{Ee^uETtW)EqE-*(t|qfeM-SRpgA5DMWOnki5+GXeR$ZRCT^o_hJV^NL
z_^z&}icv-n_vVL+9heQMNlh^SK37rnv$*SwTSz;wF%vJ+2F~H8Qf4j}eY9>$FrTk2
zE9->?Jw^42Vf@aRK=ETFUBd$~hTiX~tT%V&>|nFO|BB`?^!BB}BK&duVo3IVX=LH;
z@iPLr<zRLLo<RK>DE{VFLKcA@LRbb3?*^^CPGCcR^%<7rm101yk{{%QyzheG3<dm`
zCEEoI(?b9<!%wL&yxXAu(ln1;#46}rV<OZ%lEVx;1)H^AQcj98^Iq7e0K|Jv&DEoC
zXr3!Cj7KtKb<Fh_=bn7j3|leGL$X96MgY*Cn@OV6V~b9}K()rgoRY~0U`>Zj=d5X`
z;p_~+Fdy}jS<AiUfGRmg#xTG>d5ld_`LhM=#2R{xp$Hfj`y=3@JF+#&Fl-hySEb;5
z3c+w6M9dzVA{%=F>_2DiqT_1;gQf<sEk3#?z(b#ao~fXr_mPIEudr0;S#m!Nn+;ji
z%&gUt0(T4r4qX2S4{AK_6C()ObBP)W#&0WdoD#PH$YwVeDP#Z)?q>(RYwiZb_2t0V
zOH7$i!yiO;-s1zyDv>N(qE<Dj=6&DdiK+WUb)nRCQQfn{<v!0>_kcZp4=w~-=3pc_
zb2{Zg4**t>*M9^I6`>w!gYpsFVRQ;#;<X|yAb9<(9G4HWKwDf>^$1~f(*+N%W-y;c
zBUH}0t^nF#jrQZq-oV2!B9ra%6n8QqLW|R_?EP1maUJyPZEk&sEcR9Aw9~#lvg`+Q
z6<KZ$>MF5}ng9Lu97%P+w}vu<p5aM6KqrMAW*@~zt4xcHD>X&z6|s`^454oZrnile
zjPLgVPXimqQ%k+$y~{8HOi4iZ+=I2p^SWGZa(pqB=)j-Ke?;J#X-SvU8cC!gUzB1E
z5Vdc;U9x+DbF~~X^$8zy_V{P~ILx1!r!u~@)tLmw%llEDyddL)Bbd|!-9~fWkF1D<
zHx5(Kfc9GBU-Qcn0rLW3r^f<-3IEY34kqwrj!AF>v^8L#D=8dsS#ZqH68<BywxDf~
z?EJD|eiq5VqYAviT4iw+cWL1>qV#D7NFdliVVIRnDL3b00uO!vtQN*4p{-qSO`+pP
z+j#fdY^(6L(MMsd5o#{jFyx=ttV?GC*xw}x76-f*KmRp^%)o^5XZ4vA;7wDAz^7Qu
zddg=m;YRhdwg4tI62x!C0d1x<hZ(i!TL5=Sc@>CNEiRlgmojgsDMJ;7Xf=>8qhP>V
zX!4O|)<NSaZz-sy_!;n}4#spV4-*zcU6{ZWW>^udb;C|*It$KF`P^I)v-ngkdvyo6
z@q$t&7xTquU;;qb@l7-JBOZda%6u}%Z|Ztai)5~%0X0<))_n#9ocPu&ETrCVfOSC_
z4kx#i?p%5cv>FhgOb>%N&@V$Y9jHZK54aC;%4F`PXcvaw^9?RL&Gk);=A~kRWgh{S
zeV?gGG0?)r_?hj95KwO^F^2%GvVNfJECR_8no-e2G0T8sDwTyT6=lRR0?>M0hAWws
zxeuOoZ@E~%D<is8g*UadCxz?$9&3kdTqWEkn<PD0wGo$_=lA==0>?L?jU)tGhxh`X
zN*xsLSV7xOhk32FO(bghtz998YZu(9%AA5-nKZe#M0w<m$QqHru1}fz&>A-rldPu6
z4@p-dh#z;!9jv?mf_3YzLSd~*ma{(s-*cb;(RPCMOE)5(&!+x4w>qIA%RsES1JQ&w
zBCa|YWKcW&rQo$qoM&*t(b3)0>bu^Kq2!_BOUz2-q);Q0#L_=khQtjjQGRP;E&F8%
z-quI4t{>^ii8N}sy<bnd1?d#>4k*Gd#%-uy{yEgiTFdT8GXi-@v<~1cc_Kkh4v#s&
z$f0JD%e#Wg96$LJ@pReHycF!Pr<ckI$~I@415Kx8oYI$u&aO$4$fDl-UhQ&;YREju
zSw^7jG==)_+EG1lT_g<swoo{J_!59;bvqhKQ{P#-g<vCyZE$v|r4PjXA5SZj7+y<H
zeZO>Rd=UE+2AI+}2K&zfgt(vdtNIpAV{iKk%W<Vg&NW%ckljDGSOmx`yzn$v2n<gr
zlC~x@k|6>LWxg|Gei9=583hdVvRKUOYQEi^ga)2OR8^-u4L$(Pb)oZ68Sw|2WJ~wh
zEEC1x6Gu0r4tTd9$1itifjQ^}53{|w>I%#Rz<`&4^$P)$P2<p|$K%|vKx3RE%<Kda
zU9dv=BWILhmk|>9o;P$bdl``?05VkAR}3|2X+jmw)KI|WlmN^7NmO$*U@4e-7bU2S
z7<KyWb8m?3O*-9Np*&SSya_ysac3m$Iy8{`1=K7zLY<FK1q42GKcltw)BY__qcqEA
zwM{JrYG%`@&nckKgHibTy3jVX+RL!?!u8BUnCU^h(i~w}N&=WVlrw-M#V2mD%m&Y0
z%6&-JR+{Q`uCgS`n;;_%!m^vEF`w|1(!1R?C5}{WtcHkv;=T)MTXA*b^_6dngg7%F
zML0JrQv*;Sja;aK?CA<(_>O_$JIp?O|5vN|EzC17dI@NmzxP~R8swSRYUMxDcEt(F
zr?m;KMpfx5K$hTFkB&N|X3558g$|SMT)%k?FL{}b<x9Q_Q!I#L0k{-6V1j($+K6LE
zP_kcP^OyIcRqF@F=OD6&kCZf!3da1n%ZxrM2}wHaDCh>2o3Es*5XU$^NVgtJ63p3>
z_dZZXHn$S}-s3X2x8<)Qy%?F>FNvT2x~_f#%z?358bE~!ENK7IthuN?49rV$NnRJ5
zhQCOdQNxINDG?YdrY7$)VJ1t$qSy^fAiTD?%!nb7TmdpCvuGMQE8xcMIlcG{u-S|8
zc}nm5C9%^6mTtTt)bre}hFN?na*hn`W$sV5W*$fi18igE<ZqauuK{M#il3`hQcZ<3
z^QYhp84v1n90>3D%>|>XY@n$=P#vXS%U%lX<{VBR)cY1M1?#vCnchc_K=6XQ=`?|W
zQvuolb%9xo8Q+Q41<O_lq}%FZCt1kK->iK(4ItGD5ynGk*fetH^f#kC43?I(H6;LC
zeG%E~^rNJ?4Z#leR53LG4k`x<VKT#>=s~mW9HZaCmt$tBz(g`*x4_Vb-|YP>h;*oC
zW}UeVxCsaz4#){mD^N>L^0Q2H&ppQftPm>exY4&}ARC|}3`MRLyq$5Cx9eQVH|Sz`
z0ALN|@X0Ko98==yqu+g1V3Il$Fyl_sK#y|QgRwJ4>!BRLn%afz=>`gCSS;yVtTpQY
zVy!?L|1GiB&kzrAfJ(f9-fQvHpIZ}+b6J_iLDlco{H?&Lt?&gX*9us0ry|@|Y^LR0
zye^vgo}Jkp07IL76Z(8q28td}2t*D|9|7|pge`1k5E%g8_B7GK9Z<Poy=%g}l)qi9
z>Rk_IzG7D2aMRnOA@7LI2|k<WYMW%kb^g;MR<R?m@+v81vi+@eV3+;k7=%E?)Yx+%
zBv%_)jF=LpKu^@EWB~7W21x~*2adPvGOrgzGTAFY*f(XI5&~oF24*{U5IJcu4C>bn
zLFo6Xn#^^vGII`6T$seC3vFg~!`Lv6(pV{Hjt}YES=YIqS<BTtA%gm7T{Jec{ws_N
zh~<fo$jK-**zk{>-jC0JQTMw4LQ^1Cc$v2-j*Kae6rrl?#|k@UwM&e@Z%;aVbJfN5
z=M08f)?Id&<cnyWQTYnv>81kEx(fv+zVix!9*If^75u;lE>!W>+Rv%`%+1SZP-nOg
z_<Ms$sw<-0ah79_@5KrNeLWtG1sk~dXf3B&&_vP4NKj#9s1~fzg<E7wUQFllhEN?B
zsohV@+|NZ<S+115(S1?irSm<L8Wd?189|s70P@9d0ATYJjRhpCkih7^W~xKACdGKF
zICw;<y1((5MQTFB?&Lj(7_s@fBs;G?x@w~-*y3lU>)ZD`T;nb0SY6W-2)|toJF4c4
z+Og!W&BH6ep66d-OK+}ASC8z@VXQy{_sq8*8rSrGdqf+t${v+bQ@Ujj#RO>e!gunM
zzNrHb*2f)--Isgg)NP4PH&#Sfd`2no?C=enK<$7Qy+iXbz`O4X5DYYe!IrU+l?*I{
zcsaR^>4yVO^xFaaj6~$p@$~_F?XDmbhHc4)C74%Z#qdBu(ZI0W9RDwkJ^AiiAD<9L
zoxbyE{N8ST3&X0I(-L>E51_8kkT||ywu?aLxeGkuA<8naEdT@r--&(|z#D83L{f8P
z>LxD!^Y}9x6X4fyYRqyBzm~L{O4oVc1Upd;S1qpE_4V=wr47O1YI0|8hZL1Y_iEfo
zYQ<exa!+gv2iSK9Dft*I&x>?)D4Usn@xX&S1^1wlEZwE!jLK%pfw-C@a{^ii_VPln
zqZu&EA$~1+YErRiO+Xg6gs>9C(_(AHnet4^QuEdO>SJ<SIV{RnNj`AH-o^i=7xl&4
zp<0@Wm@-sYE;cfMQ?gC%Ku-FpeW9){($1`0vSL=pg@oVjoXIt*LHt26ZMG(LQ&_0=
zU8rDhA@aT_&-&$U69;2WqUB7Yl>*h>KkBaJf(foW#3J=zYWD7LO(SaAzm%5)h3icG
z3M;{fd2b>&`6l<@7?>@67tjADFILqrxY66?gfQ#PmU!~mQW?25ue0J$<{4{36`^sV
zQB-HLYY|ShIcp1#0dik~{`IDgQIlJb{CG-j`1xus%T7K8V)B}bp|PT^$w4J}L2BNC
zhDrXHO;et<enHcU<GTc^ir@Rb_3>#tac<cq&-kuCIssT$#vS}wJW%;7aVl<(L_~<P
zxS)fg9?kk*cFtwJ?Gq!v_+Cxf?L+^_)f*%-YtnDP&oU$R0xDGZRz(d?H`fm05WJra
z<w8U$XRhY0QDjCo$0uL_0Xp(>)cOXxR*#2yHALyaqwIg{C7A#(1-58;7nSrl#qNQ`
z@=Lo4zO97;$4^L#HuXNqkocO64V~hau!pCcnL1?VM(Md$<v6}4C6uq7pqg~FE~t)S
zBT)Fi!X`kfln&y;M;HM}I(^qy7=)Z{ysiTSn8gAhVdQFvP{YhZv;YUJrUWrP0YK;p
zu!rTJ_^#c_fup6Co~z>5Tdib)-`M6!F1<W|hbS8uvk1fV22n{ohGB<H2LK5Zj2r{n
z`BcWn1aADCE7?bc(`LTHnjQmqyOXvTF-@5Jt<9p}Cek>d8LB^%2~DM}34ygOyl>wC
z<!{|K?a`MC!fyTT>)BZL5ef3_P5I;V`{r{Y=Rd;>TAs=)C$U^ri+c}0Z+`NvYUeL+
zQpFO`{TjFFvm0+0RmJ`3Nw>(6oA+mvws~P}JGSvw6R&nho|>7U$ul`t0Zx?=aMH$N
z=WNxiAJ|nnjwZ$A5bJdW`w{vV;!tVLKq)__CPOd3VbRI$Zt~0-pc}DSJ@dG{tl%I5
za3j6d(O0OOx(vD9;*4rBWuoR!G#iIADoLL*4@&R9=6O~+{kYkuBh~KMP>>J%D@xcO
zw}n}b;zv7?T)Sj^m4IP0&@B{fy`@s6+z`sA{P{+vxBki#>B|)J);X1*nkoLNCI>2I
ziLOBY?6Iiw-GpI#vJr9Lb<zjJTE|JZ$E3J`;VYr9oLtt+el|_!sS^muS*N+Gkb(t)
zz$LeEJ~3~zQ|~`vy`e1KVNdd&Zk3Wyq@bJO;kvGu{s&&Yw93nkQq^ct8I_dVx`STK
zjtoFI|HO98;goK$o7W;!ZP$+?8NwsfF6JZ+s)!~9RUXe-{l3fbyo}vWOJ9mp^$PE-
zI&B07k=Z<al8howmyPk#_mFUjkBXn#;Vs8w2S@jRk<fX(ZAb5k)9`r3$lK(|CvE^X
z7Vgy2P+wtSR9tDOEq`7H6c-Ln&1NZ~W;jeiR}Ka+=pg&l36zXPqWU?Yl#*&Ys*#kA
zyvP7Tx=I<uMn6c344pDF+~v6iZjz3>e>Okn@pjYG(rK%OVOz3*O^|twZHnZF-syd7
zXQ_C)7v^n4cVGl)ZKLm|W)bmX{T{JF5|~s}mGiP1_ss`*tF`Bh!sIXcIe&QmDcDbK
z6e(&B3EPbfM1vYb-Mw)JAAgDYVru{9-0dL0{bh|d$0Sa;Sd@B)0>l%e1p?=O-e`L=
zBAmX5D8_B1ZE9w8Bri8u&r{)bg!`R_ZGt_lehLciW2do;C*m7siUc93MHp|=dN_q9
zN<T(bA;%pg_l005-6v`Kp2_#+MjywXt>63Z_3;1}5sitDxXDyJK&9_aArhhfUwn`v
z80M-9@~8W^?Qjx3(SX}Fhs;!#zWdzXLqD^FiO_j`OE)pH$P6lN#asuv|3p>|*P!U$
zWle*6!l%uY4h9tOzOB8wC+uc98*Ty+-w;crz5i#B=cix}cZ<4?CP6<;btg>|a{{v3
z_vl7Vl;x_OXt}-hv0Ibznx6Nk1UBP$_OOkWdVTsS>}|LXOjtX0W_?`bQ4(z<fZ;vR
zfU-BvWZ|B+B};MRtYQ~%RGi5A#fVp9B_M1i8u&0V8{56=O42L0(Z@FYCqJ(`H*va1
z<Z@zfQMTa|uYsRFz~92%KUsEOsU*`*SD<rzfudA~*r@8qYAwkDqxp{Av9{*++s{)-
zCGSfgvRAf?ylo3}#Xbie4dXy?b)nu({2OW;`pJ2duF?DC<0zJ0j${wo!8c^l_We~`
z9~@{CcA}*#T%X{f8|VJEkJK?3w7^0e{g*8?qUG;>*CdLLN@qNNFtI<f^I}RDO;$Md
zKnJ&x+=$jF2d@}C&--38sK7?!8Q|4dtQaB6uK?{Jw#<1xO1Zjp{_WfQsQ!g15AMbQ
z=uqILwlIYWqo(|17_!GkI&^#*%{R%HzF6P;@GI=NH7Bj8U}yV1r;pD-6bVF<{%Ves
zEhg>Gme1yP`fIqld0LP$C+CBOl1Dz8aPya*AKZ>e?`?ICyTEflC2sx&pXYT+=4T-X
zjX<=GXkF?^_XxbCgA4!BBC|fz=7O|PoWkBOC%(k*J|=y1+3Oo;dYm7~_WT6x0*~mA
zuggwCYZ673nn=bFN17whGRZcifr*o?Z&LuszdRbSBiq=*DQxl!&gl;?>)g4#Ss_wu
zhns6zhXi>A<ynGq_MPNbvjb8EckVq_lubYE0D8sw-mLOYIo1TgB3LVKkQ$8MGuPVb
ztb8bjAE)GoDaVO7Tsd)N#N5W5KNmoJpZ!u{W5xkKy@q57+0vzE+VnI?&!(0)K<iE_
z{oH3Yze)0>@KDP;4MQHRAwPKatl!Y$k=@mt=oczc`A>JA<0<5Sn~=>8yE%oe2feE&
z*olN;A#z;VR~Uc3S@#x%QEgQD=|)G#GbgbdPA)g$QwkB^%f)(ToJC_AtNDkAr=Xr<
z#|Xj6p{1#9UD#lAnT9JTuT)95i^MjKp8Q42<wb2;qs?K&owCctVGP+x1>kbTgH{cE
zHQnA=qNa&)R-=a2x`Mn1SGYa5oq4(c#5qpqYPd!d*h51Fr+L6hFsW-=D!l-pkdXp!
zT=Nk?0Aj>3f#C(lDwKOZ;CEr8%;C7lcq^!7f*4eYxvpZqg?#>DVwKg77}s+twl309
z?yrVh_uCwHm^pL^;-F0&14A71eSgWHYV@UQmhp389O}n>nxO2UJ|X3!cS)>oP_l)(
zWJuW$6HhVX2X)(=U0zv4Uy;2gc4&WvL?CYZ&j|ESq6WAqswQnM{U}ML$S~A<Rg?VW
zb$PnIm8IjkTC@FRsWy$uFXy*CJG|F_IolY@b;15{@G`?c1`&T|P9M`Z%-Od(<axuj
z^mX|SN>`icI?qp<o2r@ZcW6>~4Y<Ak%j&}PAlRmnDzCP`Y+qL6Gq9!GX<Lwmuaa#D
zZK*`e2u5+_sMp<L!nxI!P;{R6XxsJ(M#>lv?w^bj;DH(tB7HULpShuBrg3drNaQP%
zq22|>E%g0M<-9)fQ(KozFjD|6P=`af#g4u=I8;O&Jt^9HmK5{gxJ$9ZIvI9MY{B{^
zK2*}*@y)Q2$P0#_63g)ty8i5bW0|j!>FW#hBeOqioK)dnWpGz28aY_)9ML!h8XWwI
z2-Q!})-0owz-DK@mg?F`;22a91RI&vW}0%YdcXgm!<NidVz;E19gEp1%@xrw_*=7c
z_CN0IO4jxv$AMcs4IIuFr#7-W{_a*L&o=_VO6&l1k9lKq!^Zh0CUaK}4Pw}pKJrm3
zkkyNCI~;gEUXa6(V@4CcD<mV$HmaTA{!DKF4&R^)i1+C~+i6gH^#bP`Y?}NnZSC=o
zcLn)L6V~J>fg;W4$S*>8F1_W&ls$Rm1N>=gTj2Q*2{B0cE?vr14CvmG0TDQv0+&Nz
zNsRm8jcQ(?ri(y=1M#U9qR*Cog`wrC#_u;S6He_6IIu^wEVW@H4DtRUv$`HQ=S}jo
zE~qH&h&G5PyvhXqou)#WR1_(ZHpe!{oWRd9kT}4u^hWb92VK{ZOs}(`>$(tqb1y%O
z;exA?^X`*!>+lNXysJ0PzYX}S0=Uximg5I`XU_CptP7hZ-dXSlr~lduW9AJHKQwRL
zOg1KPK=VfD$l|}cF1CZyXF2(W&y~+C<$Z@24EjI(@?V*w1&_?FRk+_OJDB&~y6?km
zJbdAavve#uPvUWcH1if=-njm?S@<(^|0JZTY1yfjM-|*Q9E0~7)oBZe7L~T-we71c
zm%g<;3gR&Qk380c?Bg~st}nO4PWv2YDNjB5pOTz~lW!#ECc2Jw-_z}MY3_W_E6<Yu
zf!r<p_TuFoC}!`8p5c}A%pEzb82Q8iR3jL>J8ZE&Gf@<^P58q`8ASMrzxLjUS8vbK
zRKwBO(K8NH10VmJ=!ggJtF<5PmpbFm8N<$6>vV9D4*4_E`(M{wn<^}=_^wOkgbQC;
z2a}piId*sj7<(B9F!m}3U-~&w8|G`2IeoieE&nO?^Y_JIp_bQDm)&_<kBc{vC*T<C
zjPG%qza(M{@4eW9{NXn`R@YX@AxsQEB9?p@hV=Q1a{o_yD5I*FJ7Igzge9nQW}ZQu
zKKr8XKV<a22`u?t);CN0>}gKU>$eX!L^~S1=RvB)|065@_vT<FF;H0O+{O&->5VXJ
z(z7|Hcj|_id)4P+SGLz2eZLBMUEus*y1!M9-z7n}CT%q~SocIS^d}7cKW~w;VUqk-
z0;|ZJ!#DsTYz7Nc_RfZ!@Rq3!yJX0vDht5KzYe-Ae4}S)v$A|ccX8}9T>jk0iF$1E
z@Cx1ZjGLha`4_4xHn2!Dk74Ekj(?we(7f=X?&O-4ji^8tZT6wk|2}q+%Y0Y(LDc8B
zZ8Lw>EdI09L&y7S1Kc^@?7~c|U2nyM#5Y&{6=b_mJpLS>KCUji@t&SQzD@B%4i-ej
zrjMUm54C;qmA-h#YB@9O>|a^JY|o!>+{<FT54IMzLhaXm!@mBc(etB_3Cwj+PYTNc
z{L@>Y8e76gQ#53TCj>S-{=em891(z;0P~oQplN!{F_q25$a|Q3t`ep|1Iir<@XL>I
zdjKCI|9{IzK1e@c>7l8hy(3{}i^mEuA}Ce23Ff}MF`5;q`iu3^3jg0S7QfE?OAQc<
zl;l=n7Jx{1?F#$vO!BYIOWFNcRVwh+B{`wl8AArkP+dy3w+slhFC#E<AcC*$0VBoF
z+3SDWzo&VRZ9eWV<YkeL_JIxRC9oec*B2_s`YGCTc2Ut@5A~AOm!B{>k|S4rXc?Cl
zCtq0yu5swtZyj!bG;jL@pDaTJ4*VB@SuY?d9#eyhR*gUxo*&tx0yqigAoRS)ZhA9`
z5P}#N%LaM{gt;GP!Q4S@)xw<^|I)F^?RNSFs-lI9a`<%Q6;AuR);7H%21gquRQ3cO
zH<DcED=azA4fAjj=tYtW=9(DS?E!o*?iV0EJ>v08WU38wUHzG+E1dnxnp*vWAzzw|
z<3(W7YogX|e!E(iJr!pA2Zm(vFtUtcxS<|1$JtRpQM9j>tzn9Sxz8m`wF4e|a{{PT
zU$BmF;n;im?to}!Hq8JTsXBe_wE=%=Ocsx?i4<`RK^1!|c$47PT_PyDoNtYk{m|_Z
zdQtw32bRl9JhmFZTpCQoyWqj*tW;lzEq%WEdn~Y}Y|EdPeNm5;%hB!<z9qAktuXC-
zW8FV1wpvd}({vvt!Ym~S$i4vfq=W<rcC&$OwK-r?_Akro|JEMn>rWc;{cF~dMh;P^
z3=@dq$kK%Ck(n8cXeXlHK|!(9QR#@20yp<7=!iW!^X|>o_uHIn*JvKVsW5VxLYV6a
zn6l@GjrmnZMPbUGXW6S>!gPxI(zxVCB8F2;A&>yDF+d(++;YI}98$QR^zq7d?!jpP
z4UF5{ULy7=J>L<j%g-{bB}y$O^_5BNZ1!I3W`CUK8myrNr-X4j?%cofOu)e^;efkN
zWsR5JG|)*BcTj9|*&aG4^n@<gz}!TQRIn%Tx~y^$dEyjg!`p7Tg*h26=ijgsn-A(q
z)yRFNkubDEZtaVF$LG-bDx&^Ft0e4Ar4E<E&CAa7U%HpxOHv(bDiFkPMRq0$Hk-r?
z6?^KhIsV1#XzzN-&Y*K)TUQQL!Mq$m_O`XTQzitruAEd{+#yZ7nxmBJC=}zUZu&vN
z`m%J_jvdRQ^VC=Nm*pCU;4@K$c)#&dYcjmV=>+Pu%pLK^wi`-6$e-U10*!{}y%Hw;
z3ttep>0DG<lI8A9vQQ9R*22-NOLo&^1KoFq+1c+ztlVYFDAs15D&AJSU51y1&Z8ob
zp!Mr&Y6`J$1irl?-!*LzF3qMdlo!&m>|tohj+cpw+jq4e;or)l3S*q48q*G<1*kQk
zRkaUVSG*%h)IHo<>9Xl*2k`@s-{a08N)D%slg;8HadLDd)k5pKL-&0n+73frGfr8S
z9s`y8mk#*c30NyHr;3Jq(ES+Cn5xWb+*U_8NE8aF4`5c)k;%nsH3DI=el|kRh;1>)
zuIh5^x_BVg=!O&Mn(o!EPZsMMEA;%ZY<*m*kW&Tk=^Lq@%QrYxyKM#K0F4heFhI*J
z3<ot3#cd}c$M?DIU7g{E)YQ@vUbTCxts)ds{o6S?W{rx?Wk}Ound*jXs3wPlj6aUW
zp;ZJ&B~(NT3&jhUuXbO)dY`=yDd#Odz?+{6!He~gJSPnwy;vsl(T2CRXQWt9kvI0L
zbW}+XUl~k*EHr#-8p*}fq@QLyaIlVY;MBh}#ku@OPsy{*rPR2H6aEstEK>>a`gr;P
zBb}~G#0F#d^Vw+|smP;o2SQ7II_wpT_B)^D48J3OOj=QWU01NSTQ#JULQbJv&!vXS
zkMvq4(v@=!kR@u_ckQ2k>e#R16{Vhh+CVL$Y-i;)7pY{CQkWNpDBog9@o1p4ApJ8T
z?$RXGmb?wwPt(4T0lDU7#X2Oi>cb{6ryH<tal&v`XH5tJuElqfDogCG6pD|HRxGua
z<`V(7_Ug0i;-8L<?mnsAx02sFea628lw#PVUa^nl0~gB#n-ZnB-$`uPnr8Ch^_yL<
zml+?oDNcWDsVGo=X!vXahO&v;O@xP|IQkq#a_$ffxy`aM?%O$@cCLW0eeoQ2Tr%bY
z&q;mPw=f#d{F=c(_%qy`YWx+p-H>JhqUz09lN)oT#<!Cg>7cRWP%~JX+*E)U*5x_Q
zP9tW;Ox~CT(m{~<Mtcs!Gm^ob%vxphXyG1PKRDfg?{mBj8II(ZXUdq-#d4aE#Pp2N
zn%L5jQ7J9+(r8)Pvh~o-VvE!FtZTdLlr;~O@2kH!6@Jtz`(;L;USAAu*lF*R`;qB~
z^{`cuZHLObyM)T*1Cd!gkzy+|X_@uPj&het^h$Q|gEy20qWd2e7*a;4@n}s@u*o~d
ztxvgcNaU{_h&{{q=5fwlxZ*Eb<GE*%G@FgAgNA_wJ~}VixfmkFp|u*sj6NQ$xhP}r
z4#ZPJWrgXr@L&<vOTyLrF7v1ljstoRo+s6t551XGQ#^DgFVj;=?~{k-2j+NJ`Ktq!
z#Y5%~wO6<`Uq{*?{X5{~eW{%UK5B%f`tjOfC=<IrR-{PB?rImkP9*&L)`v$<<gI}n
zb;DX=-=RW~Jeg1kL^|S$TuF`Mb57K=ba3`uc{WAqYW8`naB`!lv3Bw6DwuZwdOhP2
zlMj9Qb;AYbTDzC1&$heFiK<(k`n<S!AklyQ#&y9v`?L~FM7l9;Dimj`XPQA*fpe+7
zR-(O3oW|AT6J~M-pOtP(>BjWG7f#pUjqgO-sXMM>R^wldg}D%}wQh(Ea4}pRYM}h2
ze&l0I`prtK4B;aM7PeQ@d$zYX;j9_=v^2fTQ0z?Mfm2(bYK-x=n@YzQROX(JuGmoN
zub}gVUo}GEz(&MYHh1YLmRToKH!-KUCNy5w&A+SZ4O^C#*o7PAZ=|p1$=%#9xW8hM
zb>q99bH;}JbW5s3Cq~#-f|YzcGyZB$TB$VUMg@PK#72Lit9K5U3Lk!<=W4#f+sI9k
zNGLLsr__^hc$#KqllNNk^QZS;ob<Xsl0Y*ke(AsVApb`B-kq)6gjLS3FOa1)la!Ea
zK)Vg(HLc}&H;GwIXU4aRY9WS!`8$8tcKD!$j}prit9Lu}3f8>ung%M$)VfvYNzJD{
z*1{9$d`rd)hp{|?sR&H+iNU5Fvk&5&#M|$GwwF`fTi6|?B657+X}`y7{dOvU5uJWp
z6JRb$h2)E6G2EZ_BxJI!d=#?Wi2XxA)GuYnbtG5MRPWQ%Xnuzv9fJ7mk)$BE`Y!Vf
z_0y58c4e|x9XmsM-ia!U-_qxNEs<u!I(Z(qw+U|%`6;DDw2Fy;PQB>v=puXlqP0`9
zS+ec<%cECcByeSDu=t|PnmNhp<)@%x`5>GmnVFo|S^J{*&8K_r5h=9Sg~R7AKi343
zgQR;f9x!=2&}T_w!@VSSZF`8XeSh@eY*?JD%JvphfsSK|?>^i_j3K0Uwmf%LEgzOC
zhFEMcoOI4dRMz6V)tE=7!BV!SS0XN-Y>!NNBj@C^>MiVyNeLhjj==_n%*Dy@5$<k5
z6>}-7ue-W*wU$*@#PJD^z8f3vsfh1*bQZt&8*EyzpW2Teiht*mm7ui%jhbNS%7AU^
zf*;BQ#phV<gk`wP<UV@1D}nL>v}@SnCAJW)`l$X{_&0aGMou9l>Me@iMFzb?B3BI|
z?3*>VOX(!H78I1`y{%Yx9pAjq_pOWMEjcx3KgA|$d6|S5i9OH-E=<KuT^T{>M^?SL
zAM<&A&&HX<irb}qmig)JT!vTE+F1UrDu(9lrcb<r>Ssz&@4_h;6N-CxU22oO><L2*
zPi)sb)IDhboKcX~u3u`~iTqSvyvEXqJo=3DvqPWN<E;Yvk)tAFl{Jj*N-WbTe+|nn
zLa>BzJ>9BHJvRTN@?M3ugLV2WH(3SRdn=PW*YGGLN&!N5+JlUI)r5fu0wm_0u5eNL
zUyAjz13$JS^Y~ZtvUkE#<6*L242&oxjQg1w9y7N`-K7?(TjbBxhiplGcYl<4cx<hc
z(t%<AZJ#?Yh`_S<W1kL@Ur_|9+dCtJJJhMsL=3`qwSBrxTv5mSR<fX{RqPkNQ?^`N
zIbH=sY+J?d=T^NE#xSP3lhq4R{PcD3h<t%$uU9&^-#u~-?`*AGyI;i0Sju+L!?W4n
z9-o68F70AiW$I&uLhr11Ui-4O_4H@$05_Sc2!%Aw?ya)LUtx}3=Rb6g4Np8lD^gt?
zczmByLl~*%9gP~9rrRWlSG4h0G+yLoH0sOmIRW+<v;hSJhF#^p!*uwQlC4|1eJyjN
zyu^E>#t(9jAF2}BdV0eDZ9fd`-^~yrSx0DW09(LQn43t4QrI${n0ZxnIuuJiPv3dY
zqWuJ$c&Yo-Cp3LJPaRg{NYWs4Q^<RWEm!xsT<vF>Xz4Ix-c^x&imNYWIBM{Pne<_X
z?~}#e#-m1Dh{%;_kvBKcTIBmFpG{l!#md&@<pdwxEq8k}|MeKp3zE*6X9-1^Yu;ki
z`!pB&kp{Y%O?<@NF3&d?KQ<W?xN?rFJ^n>ULm{D~aDSnsb(kLu1@8l0fIjsR4TB$%
zut6G!`HsGUwsxq${v}I@>kurwSLbM@_h6LpB&+fy1Fb^4OkYcat}|B*nQzUhKa<&<
zFC%Jxvb|8Wd~>;2a@6|wEE_GQeQUQZzk+pS+(A1(L+__@cjASI6hx*tKk$l|q<+v>
zPRYu9*D+cbtM$c4vk5PaV|POoBQu*v<jM9W1TiVKN+y3kP2&FJ&v`2!`50=*-fjpB
zKEC0o)_|+r+YC1RsVA{JE#04<xmm`k$mav=iSk<Ey>q}EL3SOnBG;HxpV2e|sMb$y
z*JM4Z3AT)NdEL-+Ms?f1B8AEX#jWh`kG-ob;eho+He<`&P!tPlHz^0Kr{xcsMYv{c
zeABdk_WA9ni(c*y6I_2tQ0GRXe0zJ=Ew<&&Ljm5()Wjlh6-KO<lAC8YVuVY#)NIY*
zxvZQfbG^#h(?hYxO<gU8t<0*gPrz^8vT}-NglFaE;IaZK)jIK<*G@{gsG8`C9A|$m
z*>iaL<gc`eA3c2k$fuWn*Vqiq7C$utXZpyR|Bc8@-M=d`v+i|{-)M)aesa@^^T&1f
zi5x#JEa~6CTZnxdx!M-&f*bLQh-tCZJ-Jt6J=yD8)6sh`LsuP3mmkSAJ+Qyi-SL4v
zC9Yej%#jVvMJm9D(yv}LIrJt^U)kO!|I3<+wNYD-@%rpNXWcB3Mb;fA#vwATM>Y`+
zTVsdxW6O)xlwW_GWVz&fF3RA6X~oSO2OdUWT?Kp3&zD}y-k%es#I0C_L)2?*Pt@sY
zvE5PIYML~C)_0jC9VYO~+?Z_CZC#A?i`^j_S5k0#Kydb6q(C00`>0ATcjflwmwA0W
z?s1R}sg8!9jF>^kfjfFW5yUbkStu*KZNMfc%;_Q&jR()H!-ArQz)407vfss#E@+3O
zN179!Ad8?#34uBd2Z_5D!~)TM?vn#!FZSLQUt3lqYPL^e$LHJCwQk}WfFk<bwrLi3
z$Nh18SfMA^js2QObv6?oVk?lJLd(s08Mgubdydq0E!+P<toW`mrS_wvZ7u~R7gI!v
zV>VbiiRL>@Up70D0)r2=J2B;{UY&-BHeKuPNcahRQgPk~p{Bae?pC?C6gF)<q$=wV
zis{g2z-IE2xzj;731(&M*6?nE;E<4oG~%4%$#NbZ-8&+oTMr4RpA8NvrJ2Y8wk+ax
zqY{b}Eti+Qzxs*Av%E)d!=w~T*PdHhzrk-5v9a2z)H}dUzRRQ-6{KbF##bzQH9&e$
zP;bkY#1K9H%dGDbo#AH<(16jsLJZIaNh!=?Ed3FPDc9jI!at2pza1d9TbAfOA?Wa2
zymIFVlORwO`9L7_;_1=lYzX<?`$xJV;7kP4b-NL3={k%g4V^QyR`~FX=3BOIV&Q74
zl8+x&Y-DS>dE!zj>LU(-xD|9^vS7hi*s6VbonE3eW9tctFZRObdI;6_b+QxLp4kQ-
zVMlb9>n&fIQ7#%_e0bS&UmuVpVBe*^#^j6jNGJ#a@Y3a}ZJqM1IokEVG%6LqHayJ`
ze?60UU8Ib%Q+@BvgXB@<#?1-cmud#?>u9Vt+qEIJ(q2h%lddjHOiUn0y2fhV<)W||
zcnKj#W_Tiz%F|sYioE`?F^wK;@32GJB`?+sFXt$>aeEl4y5+FKE=@S_^On=8wnRfV
zx@@C{C2`0#vjOes?bsy3pBQR&dhc-3jdSO(f`Z9aMjg25On|;twAkwt#pGc&m+W}c
zuiTWAG)q%cNH%+jZ)Me*8vfS3U~z2jA07Wk!kZGCXnX0~4ME`!8~RA6ahPUrdy2l*
zgXY-2D*p#<_jgA#KH86WQ_eMH)wJpp`-<9SN-picQsXZwtNAwW#`NJsM_a-sFK}M)
zQ$WXIYw>`Hmz7hD{Y06hLM7SFIUc>GKDbwYD}6)Gj7i(1ZaTx)7dF~?W2K}-Ofj<v
z&#8=Mqwj2BN>a-j+T7xZ0z-DLD=T_)P1lsI*hE(k&G)&=!vYI|(Y!;*@*xn2_l)`F
zhQ~&eWzE%eniBa8mvQRf4JW5K$ZXS*TuYG_9k2PhzJSl@0Y5lhi211*<e7>+yx(Ic
z)pD4e$_T)feK=_V@9GQ2UeJ=Ynbo{D8~e<bKYKQ(rB2?w$)<2upTa7?TT;E!XD;d5
z!N#H1`mU=fWY11yFnvRUWSogk2&Gy<;aq%U$%9ki?boyEjKqo%ahDi(k+vA$HFU*~
z=yl`+`fyISS4EVHB=h2DZ7+T?vca8tCtkne(=kV*Ao$I&GnkJY6L3K~KXp&{hwkUp
z^1|FhSah4~>CY!i>a**7M9uPAVB7tiv=JMq?3AEr;`9g~!jHL%YORkW@r19AHfsJQ
z<4TOQS&wpFVA#g1tv3UAmjpVEbC3ybk!2<rIAn{thEvJN^qf&Lg)+19wc}!?-&k(m
zs(yPK=INqgGN=-mzo{#d&DQpjo337F-4$wuRn9BlEkrJl`x!n8o)4bl)$z79Gn$6J
z$T3~e*JX>gL^}~`i=%kd<*!f)Q;J+u`n)4^#YpD6#?qeXn3=Ot3W^9BuBrgmoibr$
zLOQ%yehpQ<EA3S`R>ZX6L2CDhv7F#@M<cS6;kPPN-^W?se2aX0Y#RIYA(fkSlu*K2
zy3I{UC(`)%#EPO8ZDGB8W3a8PefL}1$IQVV*WMs#U3dU(vLlqVzPvcLa*tc+@tnZJ
z@D9IriOX>8*cvqkSDIx>$4ZiQxMg-w*(o**cEjB}lBTp1Y1a<9$@v*Oq>id4aMc7#
z2u2nnxasm;D-(Gk%YoR~7YUYqAsb9&KBr61Mw>M*+PiPASpN_FW5=NvxHecM)gE+%
z9)g-(nfm6w+AGk#-8D0!WB8FcX2_}89Sa>F*yTThLl$3ycBMI^w>bub4s$M^A69;_
z>)^9^aVmyJb10l>;H}vfcZU7AF*Y7_@wsn!obed4*`iDFBP(%9k^-+=JUPYQ!#U$X
zIrPrM(zR7gTsy|O%F|>-m1Y5&$CPMoqUt|BNkF-cj0D`zl)Wel-@#mA^SDE=gZ}_e
z?c-P6uUX@#96upC>A=k3R@&f18zsk;^BAoWj|-Z<^da&b-4a)cjKoO0B*33;#NN&~
zqn-1v_Fh5Wn%D*Br@r`P*Lv(J=WXvouU{&Jk_l_Mf(BO!R6`5bX#zZ5c47~6Rn@e&
z9Ms~1q@;R@bVau9p*WBugEI*nZmy+I$Bs3QN@ZmQ(X?if3brRUZ|(CnzHE40lIKgo
zbnycYlA}Ma8C2s&_**<=;0$;44e!V>b^qu*QPI13?4se8=qec6pXP;@Aq$mY{V&<!
zcyl}pOp8oXl7rXpp3ry9NW8T#G3rB1w0O~Kg-aW$*0fXJmFC+>22bA*J=<l9JxUH^
z+lDW8p6H2iVXfpvV+wY=@4h4L0_%alk$sm1+Up>fqm7>L$RWa)+2k~jt48VdWCT^l
zw59AX+;tgh1%JMqrcQTm_TEW9P2N)OkKUAYg04{8n3&ZVU)6<s(9Q8sVEWq3_(y^9
z@#~pdA_3@Uoi*XD@`dtkuEdCkM6$Hj+NQMLS9y8+TXiy{E06n~T;_LmjfkPh%u6+1
zRC@fr@W*b@A%ET43uo#tY%w{vgHrcmTNy4Scv&)K8D2??#b#DF7<qvZc}lCmlI@9w
z%i2)of=YcQ1^L#qJFi?a4)o;J7R6-cjHq`a)}nXZ?Hntt+J&|vLZTWwoJx~q53Uc+
z&9m|;bY(qrBuYX543BpsRe(f61)x@HY$7oSVn#DV`StyRWGQ!6#d;;>&P23m-A;wG
z(<P|JB<8G{VWHNVdL{jem!h7gCx~(OE{8iWax&jZdYU{wyn@@rF!cDxF{NiMFI>|r
zI0ZTbZXSrzB-xJ#QT9<^B``nabjUIu;M6L>6jx1(-X)yss0^o^EV;f`X3dJO6ML~@
z>8-mYV<cH;+fJK=;lw~lK4|%IrW)xB2K!iq*DJeKwRKB7I{VT|MS?9T#m=Ew(9e@{
zMph;DBVP5r9f&J!{epc~?4(3VD9Fm)E3(h3Edu+d{c>`MtAzdJSW)hJs(rDjZV+!}
z%9E3At?U}HtM_H}Va&3EI%@)vJUDS(Iq&79ruzvqopPzx4r}El&9!efu%WMhHgNr~
zWkjdsgYZ2ppe02!XqFC3^?0{1-Sxda+Si}7IbM?8@}~w_IeR@S<bEFLbm_&eGx-;s
z8iU3AS^H*XoteB<D5e{h!2sqxByqt}0+-r9{C#3*k@>p&TsK-BjR%cueZj8v#}K3l
zJHRe4g~o;<x-rb6Zmvc9yfgp)--CM1&>_Ed+jAsNs=K=1?AE%|eh?H(JDYrkJ&*+z
z@0MR-#AH^Ukk3m_E;;cZxvKvQDK6N-M@8_#iHmpdvq9uR|AiC9qib9@8Gf}YlX0<(
zdv{~-FS5b0Co05OnJo8?x!lpw@o{-})^&?ONdaE|&B@|0Rv%eVx3%c~mz-dY%dgL5
z>@t=gdGzL(>XFT|K88;_?Skl1=4(O64dWyRHIgn15Z6Dv53pL-?29pz9GRfRpBE7B
z?AqGeT6bqtGUT?)wC_FqT(Jh`WF|zHipSH*C>jm;GXlB`j<kiZ9Pe9j=75IEiwutR
zXEHt-Eg<Rs99Hx^?zo^zn^v2nczNt)+*$TkuYJ1;)1De~**mF5=wEtKuq$etE>qKB
z$pimKrgVv@J-0z`!hX&;u?1bEy6)T_kn%aquC8+tADETglKA%d`%Al`H2jdAU~8xO
zYw6#=x-`-9X=)nRFNzq61xIOcP1aOt3-d3UoPRWV|LWfx<y_0(ogNyPS)xk~eX2U$
zuD!PR+g=b~uwX9;6DAXcqe_neLo)yP@UQzt%s;o}8uo|i`AN8x4@@!J=6Hw_*;*@e
zr>C1Mwcgu%Kr3sIAGR<Xm-0EK+<U#PN%L$$es;3RaTt%8$^9w@aU*MmO==BQR~eci
z;-@J{=H|%x2R_^S|Nd{&BALAMtAo8iIv`ybBLg}0X^U0EAJE}{)du{d?|aeZ)AsUL
zT;$*R`AyXJn}b?*)(h71($)R17LY@c#iF0nq#`~ms%T4xMFXpKn2lSYwyMkFYCHYS
z$Tf$suE$|(A%gKjrKG*qY&s~D`zVEQv|+N%J}(bB#b>~wCyg3aUuHsDpZ)i(CC|}`
z10Sc@Oz^J)dvBZ(u{4qN;C|bPr0&aC1^ks1LmC<ET@B)03#4+X1Of%RvN;IZFJSu>
z<}C#Zt6CzIt_yCwwO*3t7#ECP8#ZDJ?A?4N@sA9ymOSTB^y>(*OQRkv(&qqm<`|2w
z@rLw_*RLZ+J>7a9Al^txZpDm7DIhrp0DEt3BPjU_0X5#=6fbhoF|7De&f;^6kN=g|
zWC&eD?@<%6$m|Gh2-?jCG$u>ce;$D>)Y=+U)M?&T;jhOl-a0C1Sl6p3vo8B@3qHPC
zsuLP3e^r~uCfxK(1X}<WF&_n3#OW(+Spv`Hf8x`(EJ`Huc5<qRim!V3<}kKT_3fF@
ztJR;a&+~xE_Q^|t-t%67F^Hc$+FX~_$yBU!5yP+PIK4|W)3LWRZr4X;aZXlk30Iwq
zP?2g1sA&`c_#@*hjC^eIF-y=Ny5<VV2bSy@6!kQeQ`ueZkEBlRTC`mMkP-PKhUUL`
z&0zqG*1!VvkdsZomdu%t8IA3!MPF1MiWH{U_7@_kQ;G|r<G<ZLJhLFB6<jKn%$^YD
zy76*dH*+?D$0MWz;<i8XM%UN)R;RzTJ3Dh0_&tjzjinSHl(|aTs=24$;jWQR%hs@s
z)|~y)Wy7B5|B>e1;lR54O-$13`qU>e)h8eDpO<NfQm~W%v#M7_C}>j5{ivH#n@!Jd
zj+aHgzTzt-&xtnweH*>-w+|i&HA|aZ$>B;?DAALEP0xU}|6Iqyaz#npB~#F*%WFme
zcJ(XF-KT<XpR@$4E;;ev1iTiKB7=`3GQ|$+02C78HU0nC`|^Mo_qP8@h|ow9QG+5R
z9h69#=p=-R)KF>;p-l#fXt`4qktm5Q6G_rCmJ)4K+QgyKGPF&b_GQ|pX_mX+HI|x|
z^FGh>p67Xg&-1?L52u;?)ID?G_xF2UpX+mduIn-s)A+k~)IV!y{<qH4r==RB7yo<w
zq@?kCb7sC>H$m_${)v|CpBqpz1eoQcf=}7npO|ImAYP$QQT*+j<Ieq)zRp!@Xacl*
zb%mS^!E3#@fz$QN^;g~H^`#?zFc1H&?(@*dU`Tt-M1jC{u(kZ6;kpx1duVfJ;fJPR
zQP16hp!e=n1%{d=2HicQwhWm<xLDtG3-oDROP+y5Kva5~b~R>xA|mu=s4o5)aJD$m
z%camhMiyNnu{qxS9`uBW1MvJ;!?L-IPtTi;244@=#Kc6tp>*_Ua4E&FO`DQHw2m`E
zUh}O*C+&nk9+;7s7=gUq*!+EJbl|rF{-3O_aQ8dAX)n76W<O(~)c;#Qxq*BVkqm73
z{3N6;yh}Pm=*@JIj32-Hr(4enK1-w&c(wT<{g@L-gCl&763@qzpAX7<y5tB|{p{du
z&Rsm3T>P|oNSCyqS10%>9RUQx+no%4d*dF*!Ph!6yd5A}TmVCV?RXy?0`u9S=mvv+
z%wBq5aQ)C{FWe^UnKC(@);Ytwfx(#=GTEru|9a2EZ|#=H{FGAuUzkDs?Y^qRe7$~5
z1@7bhh-g83Zj=d&_I?cV)}*Ehe>Ez&n2xViKGuVKYqCvf$(feS6f=$&yGQ6$6A3Wa
zhY95@5~rT8=eCF)%Ie`i=%z~9at_`L-~V(^#a`W+tET!-TQiHOG>54?8Ri}%HpY!l
zK4ZDw>F9J3uH_Tt?G_k@uF>^Gn*9n!AX8%R*EYSxu~k8V-AMhSz=~ym=s;>zC>}}n
zt~STQa;;9SR5ZEvrnK7cIPcv`Xr;PASedF!1}^TZ8|43C6$r+MQaZ7P-`d4#jy6BE
z2Guj}$2j7<p}FdiMkm$7Wv$BpAE)^KH~JfYD^~c+HvVNB|FVss%-8$J=Cl8|ZKHHF
zXb0Y!Fq)8SU_A2xY$%g^@Ne_Px>ReZeR4?<;O(1D<IgF2M>igJ07jo2LDIC5+S*tK
zR}6-&ElE+J#^Wm~Fmc6{%=d&@4$?Zf*dcB){t@yP4=xbgUo$xV+&-};e9s5u?r4tx
z7<l9-{g?y_XDT2WXTLsaBEJxB&3Vi*;d6ew^!O<Ldn|`_1w7>*?hRZ|tK5h)ics$(
zZ%yGsyQNH$fCZn2XODp<x{;$E#neCkQ9RJ&D>S2g$-TQ!e#j+pYAKuvk=MT=hyGD|
z4*l&ZM1KwkChTCOcbwaq-6|H6%PhuU)8KWH-{y_owJYUUM%M?Lg8sk}qY@25N|-~_
z@~z(i!*$1pVRF`@FFUL_w~^l4ed5(H+HBzUPzVw^gVxc5<u>!0l6OOX*Y7dHAl2*0
zs8jpI_ae!B<z;b$FNcE5K~QKHI0zQqe~DA=n^2C>A@7vuyE8IjNIT{s(gU+IFeYAI
zwbv3o1dnt(ILJd`XbbXt-uMN>Yxqv~QaY8h(I#ZMz#;y8JvhBsF6bQXYdiV{dS@1W
zA<nK+?Q@1|WQ41Jau<d;uA0h5FQKpf_!H>Pdu4oRH5^q=9ll#ZXHMpi50XowbS}a6
z>Z9Ip{!ZlWV|+P|BX?{lOcg-L*HK6Wbc1Rxx08rh_v_I__|DP$zSXi0T$*XHD(}1s
z;1@J7Xz=xQ5hN4WIgP}rqy0`_V~+E|*DBIrSKNv@KJ>PhGTV9Q?{hRz4?xX=F=O+r
z-FzVz6^zMTXR}z)B2^dzD91y>^lTfg0^yG(N}AIkn~HTryC6)O@4nh+c#Z5pTTSDO
zX2-!KjKOkR81LMd3!hQ&`iyH2$EPL9tXCz4%2)O!q|kPl2VHXe8V{^!80OfnpS`+d
zpy$=V)IEn38F++1*v9zs&Yyt$9_-|aZs^4^aq!i`=tA!d$A6pF0{32y=7ZD}Gx#;U
zNSI8WiSKA8qaCPm#!!ul5s<x9KYkUjNj8|lcR|{*^*XY$^}O>(v_t#EbE54X&<*jp
z9lY}g1WzJl+R^cwL&iP`>s$z_Py0Sut&6n!+in}(=SSOP(G3t@8S#9rPD>%Vt9ZV)
z4cE3LguOoSn~2s`*C`leg_|ZQ`b#r#EXhQuC;hNDdbAmIta_eQJ-XchEX-3ku&obZ
z{vOu>oe1|`X#tQrpYafM<?j;Vt{0Ztn4PiWdG_F71$>0{E(8ciUGTWhB~)$%L??pX
z`wG64F;Y(8PcRgd6h5cmwb_~6GLSe@t-;#%P+sqY+R*%wr<CQ$@Xkl0K)0+{2Z3~_
z6o@%gd!8hWc2q!C!Q40ahuvL3tHnVWxi<j*Q2KZT48F&Loru{COVa1{k0^PRwg~?M
z>sUuFU0EIE2Q^PJc|FkZFF3KG<a}|$jptjgWzVxVEjhLLqH%VG-WpLYVx1UoRd_2N
z%Bah;QBhN*BbX23M-NES=FptTXx1iLw|^6DMgtUGvF!;Z2<n~a#~Tl++Zr__-PWxD
zX4<SPLn_tbZq1_JXb8bF+C6Bp-5+#Wj!aeGlvPgeK4lvhGj`%^X-=&l(&CeE%$>f-
zxKGk>4t8soIfbW6aL0G=j8)}LB6Lu5axWpBRyYw_bA1Mdv$v?j7lUksl4@=EO8Hyz
zXiOD+mLJ=Kf<YR>tV!${lp4}UN0@89uC&Z1_{uHI=!PI>lj+H@slgT1<ICq<yLUOd
z3}<?@*}{y8MYnZTQTb;>P~#@uXw1oqrGrx$P7WE-FrAo##4=iso#JY<o3}D8C{?q!
z*{G`Bc3mGHeCnd7!O~}eB~Q{t-+gqbNO6Fhd&3Ujy&Y}{Z*ovS#y3x_X`FN)(vQI|
zEk+vd!X{_WQtijg4n^hnM8Q>P!X0C-AfXyMxX05?VqQSMgB2S*)JnodL*9;m{uuL@
zml3q{{ik>t(06S?KWiW&+Vk})&GBihGa2^br4_U4UOC-9k(p-x<nroLx_VbTPwT}V
zuuP7ygewowct_mvhKs20;lZ6g^r9d0Ycm06m*gPqa?@w9G5qahPQ7GBGrI^o)GnOk
zKey1Bx)P0X$8U=?(gUL&Mgy)W$ka>Ob-VR_!wZQ=#kTK~-6n%>xB)%-ku4ZnwO`*v
zRlWnM&hAc(-=Lq<y6{Oe%X4W42CBYR`4piUMM!-)jnEOwl`@8=i6{B^EcX?Uqe(5>
zh^T96Hw?tZKT{|5^^yyC)}l*%9F!HN@AES9SeLoVUHmV<_+R?PGO1%z<qTB2LZ(a^
z@P_nlk&UM3vqIJKQuoZfEP2IAzUMCgBEaq3L^n!}&MaeRa9~$`seDI5-4bN*A;j%_
zXuHFr^u6NbC;p3-u3+${7#@-Rg=>c9Ys;`RYmg}o+AHrBZ2Q!6n`>vSmmE2v%Inmx
zhNkng#W6XG^qV70Xp9)^67+B2cFabz<Y3`#IY6Bif{nme#o@t}aOB(YOj)oXm+AS)
zbE{U}j+$zj)^SyMG4}h+7q+I_A#@MtlOE-weI^d?ZXLL|dtW|5EI!Q5npliH&E?(2
zp9Qd0i0@`N=-}2mHKY3EsmSMz?fn>i_2Y3x!w$s4NI43VK<QhfqJgGyyip(Oh33{=
zwo)$d&W+psnBVvyOVKkFYZYwRfq%k;4U9}&iLU4P!E&!3ZDL{>^7aX>EgU_kd?BV{
zkjWVYtAl54#ve~dlS8An40Ldvd3{(xk6i<)b4EOe!XH~il|tSUPz<g3$tcI%*EIb=
zd+RI(P}-HsKThC}HGUT!j=a^Tqpy`vk0oEf(PhYJU041E{<<1rFY@Ev?)3sxegK~T
z=4Fu`M`$+q$_MyPa>Z;IrRT58S4^O0CVYdm+_lvV-mUCZGN-RY2@O(ru_Rf!0P+i1
z&7J~mxV9q9AgiNZw*JN;ag^CWk1y<F($UmBWJ%AwVw#5b61RM4img4~MS(r<5l=he
z<x5XvW^`1+#a9307ONi^u0L`|%0dL3*f{<}T7#0nGN#~|A=q+1AqJf$lgpcqV_hdF
zKiQ4r9`)ue##dF+VuSc&!9MY11oU7LlD3If`^p-}TKWZP+*ah~N-U=_4dph}6WO7a
z<tE);uacQwx3tbp9@Dw)_y>2*$@c=8-`R0wXK+359F}`g0H3m>ePOSP@uAOuo}Oi%
z$iQHsqBBCwz56mdQs;C)fX*Bfb6cU;Lfj{i(M3*1s5W3v2M%4?oCxe^6S)^3joD{Q
z>Xb|Xz(xfqDH}#~6pqe@^2<UT{eF`QK@j}GfDihITO((5i?_DpYl(jRLUjKtpTXCI
zA;((PiO72fA3V*M$mcN{Z%CcJki>?0fwnD=F)@=eNgaF{z5|XU8N+&|Dve}!lX`wf
z8aBGG`5CjUJ&mv3g=_QUOTY)z3e9vGkvtLR=a1OonofX4=`~IN4qvkg>_y?cJ7aOI
zvEb+O!l<5+(9dPqp~skL+LR6P+6tedKVMytcy_bg(N7|~1BXoH%oA&;Wcro7s3fJ6
za838bI|A!MXFLl(Eq^A>_Q2)Sc@HDvUgW-Qrno74#PT6@IAmvV?zI7bcdXUi*;0P>
zq&uW|GVd<IHzk;i-j0r}HO27D3-M&~m(U$|_8Pc8i4d0=lTqWxmbPIVu-@_SChe$B
zmPyxCK_7FxsLu9QhZSEe@v^%(V|Dr}r`b0Tu%y{Ii#X;a)PR~!;m%S+_)C4<*R5sU
z?qK-BIxP{)erX~<uh2faswig0>gvQ>lGEkKJb8+77H3k~yZbRyAwsl<$;Yb@bXl@0
z3${Xl4;!MneK4O|JZZf9TKw?eO7Y+aYB|){8?Y`g7hDOgp0cs2#9q(1`bL|T{_NRa
zCi^FF=sCtPBjlPO2&%+#C#s_Eg(BQ8$nD@9AP2?x7-@Vvm(8W{S*WAwUk-hgK-UCz
zNKG5Im)=)o`@#zD+ff8kckXye<S=12yTE||kVoG4$c}S9EJ|ZW)qJm}-8F{x`;=@W
zF>BUT5`PB_{tTT5%RC>3Q^Tov=FZ-&@Mx9R)j&oC0^d)Y?7qIkw1B?Jk_T??e&{X$
zfk2wD6IkRg^z3_UQrKNg^kZ*^sc&Wy=c~u&IYADZnz8RD99+6y?3H?#y7Q?Ppa7J{
zF`LnJh?@LNF;p%9#2i;PP!yID@L6Fih?nUv+=fquSHp~O)})0a4*li7!H54H{(F>e
zPyb>5yG^x(j&2Tur3kY%&fz#?)tRlXnOsOl=d9pU05BMOf?m)oMee&9kMGAEY45}F
z#NfA0z`%qqQ?3^0A}r_<2Jcc4)&d(GXGL|&-0#2MWV9C`5WvD?_-@JFJ<lL@sMC*m
z1$=JQ8O)MQD$i9H<@94TsRJH2Z`=nvbVCphqM7&6+Xh|^b&v^EaU{Et(T13Q6&hGK
zS;YX|ph4<^jnS>J>q%P}P<FGaoF;HXl23+KMGWz(fck)2nH7N~;=0sCE)HjP%#-VR
zJ{<yT{C6Ps2Hy-4Kw<}lcuwOiwgGAZ<25Mg2lGITbedBniiiAkjAcJ&gL(yBhy=|O
zTmIw2#DsI;=0Y6k5NIJF--R4&^ao4+c6dY<&AU&0Bgff|P&73iJczp|ntB7znI6Y<
zbgdbA?B8(4^)Po;)OFW&7FQ9;$#D`z-xvT!Q^&aRDv)f@2e!v|Jt;cWg=*LLV_a!>
zO+v;DVP`XhbNTMefv1HO{!nJOmA4Ksl{2*$Zr*x<&;e8^qUA^uER+c$67zqi`G5X=
z`4!giSX|E+THt<@GFm6b6HV0U9Eru`hYlO6F>0q45)W20UfcH|jXHCTf5b6=c9xKU
z-xecRQ`kMQ8J+quO9V*;0`E*|KPCwY97dHJ-^R1Ko**`HG>roPl3j*tr*?&4(-@@$
zbh~pl($irN<NL1C-v{>KM)6R}rL;rTzSXMUB$hnHh23Z97wGYBA4r`?NfNY{b>L<i
zdP2neF|j0G5aR-a^-B}NF<RjC!C9kK20F3yR%`GtYf=Q$$XDu&bG7152D4JY#_?Oh
z>22#(?>so-n3_me{1sa5NtlRn?g)IM0}(*<umf<a*HCBO2P@sb0BONJMjD>5ucW}*
z@Nj_-$g!cZun>kCW8Cr}RWP7#5gyzk*z@>W++$ERsDq|qlue(_^NEB!N5ckyy>0kf
zV^XgXY~Vp9jL>Oc^e)9WmuNyfYSp$Ip~sFq-}n@Obo1<x2*3o}OkrHlIS6?s@oi&9
zJ!~y{??>|+(PFjsKCu$`mJoF1>Fs?Y$kenV&HhNk{FratyjcGz{UQGHL4Wz6{}~@d
zm1TEu2`(CJsR(OLtnZ0v6SRKEQ?r}FX}=xVHQ8uN-TPS|*EdYaKEp)VPTr{m&J5mW
zs3d5`_bqhSE`N|R9~kZnE@J@#lQSs2ZD3>ktw%W|`gKIC7}>^PFLKE*b8(J)bz!Q1
zgURHO9+BXkE89sE4Mohc=V-OI`Lt#hSjCMh4n@O4$O8PZAdu6{Y6Wu=(Gg#;tshh1
z80T_HXWSWjs%qadXcg3QARkQ8^Cg|8Jyds6r@7l4NG;`r_G20k>=OCpRtQ4DW2C9H
z7F7tZ?i1bVv=N6OeZSEy2iJMr0&vz-f3TZVNNec%l;`@qJ3;RF7DXCI)KIMbLp{$m
z7v&<F(I#YkO&ii%3IPa4NRS40YalIYEjqpDVc0sGp?TyOMmnQY42Xu)@RcPYi;4;w
zm9C?6a|KKXW~6ZG2&oUV9vURy?lW0n54`l;3Hat0$dVCqD3QiqdXw_=4szK`4y`*z
zU7N$#J87Z9y}=-&*F`daSXz+=3$XkH*+=7#Y!bRY90XjQ()orW7B<z`zDZ?BD$*@!
z5hVOhXJ+9K!tErw7}DQnNUvX~KvM~LCOewO&F&)(69?{j0l)SN4_R{KCm$$0*R*d)
z62p_@$3il+KIs&2RWlfz%~M=nlXjpbNT)-^T1d`aBJ&l!Quc<7Nls+h!Xx)!J3_QA
zNKxrWJNf>{#Z^Xuvp!Cpp<R|IY8~P7J-_NVpWIi+xE3-YQaHC+O!zKSDE@eN&c*Pi
zwS_U37NJm<JN_rk`T0{={BcaWbbv)&Bzob8R~le`emXqIivy|82GMChPxRyG6xT`U
zhGIyE!TJl1t0wnNgg`-tQ)E8s;XJW*KQVEX&5#u5E>Nxtxx()O#%LAEe|f(@_I@4l
ztvIfoEpD8fauW#~oiy7fO5buylt-OekZ<Y7^9RxlFOOe&C9aCZ>?|jt7e4|*KKmT$
zNWk#52U3U4o%N`^StMi|*MPldt|FisW~dZ^uS>SjYtY0ugqF^oAhR<${^bgl+H)bn
zdfQS@yIi}IC)Ri`4p5MRUoS1}VDZZYOi|48CWs67s10B#fe|4Z!%RZXZGLXEAm<O;
z-!FnzhQ^AV7gji%7t(4KVB?yT1X0zNq4<6<oKuH|IiPJBDX&nD2h4xX0^e)Y=L0Bu
z@Xer^Y)M+sk5NTfU_B4TynDqEVpBuL2s5UMC6!6Wz!c>p^yLDE7gP6B%puI0D#r()
zdh5{!Ky%F(h6c5;U9j&c?8PLWpP<M0Q(y-3+ex?8IpC1SDSwZzfwvul<E(4$I`r*|
zwIR6kIE&<v0R>wWAgnN#HKb$E&qRk-8{bG{soK!EJ5%O_j7h*vMBWyKBJ|dH^s;Gv
z<b7rk&(~Cg?^cR5WI`hi<OrirlQP>?jltJ2BfgF6p|^tHR5Gd|Kf2wIdDLv%H^W_g
zgzGb|jcX=hXtfEvHIPCG&<9%U?PJl0d;qz%)hY9k&lV7&ZJ>%!-9{XO291~b{`lJ3
zB*+5ngFBojo8mwXr&-Pnz0JDAl-+Xwy7~6c+7b53i+K~qyW>x{4|qe~*A5DAx`8p1
zep#I%p_I;@ef|@*EqZM9>G`o96HV7;O!zo4>)2ifPb;1XX;%k`Q7R8p1Zh`7<e;#d
z`i{&#eFUiZU+W11>J85>clK-f%C&F8&vIL4gTo%tnq~8FVW?ce?VB+c`h@vnK_Xig
zttq{d+#uRgBhZAaT_3t?g@Z_e-@+nq<(v6%2S8moiiA#;=ROxmfe*aMX+rVk2pt`!
zo7>zY{lzK1&CHd=A<7O7F`<#b58+(5>Lid6sX?Hy5#5lmq~1jp969Vx^Vxmk4bZ(;
ztwW4=z71cA=S-j)2)_%NZ2Kc!wN}{GnhWWHhAz{=FcPBhTvWRfqHvHDx=D<1u~dkz
z?D2hT+U4MDWyYiax};?Q0m`}RNKTrICM<Cln3vGvq>ZDvVhm9*?X@<Aw-MHsH~5ZE
zx0JK&zR;{`eX2A`TFbsdfH#k}bU=^FX8dO>bUP&#r~z9Xhy0ovFyc@M_HGhb3x)U#
z_!b2K2BE=%T8iOEFv@KNw$j5fgpwtbf{w|W+7LTmuUpl+3xA^pm1dT^W<$`)2ZUPl
z82L<#Kw331d{{i^zXJR3V*-)fSsPRhdDF=r2Z>AN4X~^Lh{&Q2)*;x2_oz%zsw!u(
zsSG+UoLxw#$Y@P`?1anKfYzT$R^hgs3co7;M$YnsE4C>voM50Bc6uhUv{2H2CF$cj
z*Kn>ZpK%#s&9%dGx4W-VW!sZWJW+3mftfRMAzKUJbV@YW5XtMsqhr26$zgCG3{B$P
zMrWI`LckWaA-r)ffY4U&rME2acwbG;>~lG7-+?0qnEGvvec}9Dhc52Mr`Y>%JFl(0
z8X9wkSj~5Ph@`Eg)pS*p_@{Rx4Gv(u!-Vs&g4e==_<w(C)c;HY(Z4d)`md#GWAWtn
zzBRd15qb>l%7@SBnLRA;R#@&zYpSuF9<U|dH#FHP?&59QGl68;%%t$oN8!8gD|^S(
zRQd6)QvlX|Hcbaj!8CV#Sh6#^fsO-zn5TEZBqdoue)xgjp<({>VbRL}>Eb3_AfCCC
z26Q?W^bU|O3ay~6vg^K$ii77;0DA)T`&fLpGrjJ^Hy&7EfX0mh1H=$W8<XKbty|c~
zQ$~USopC3$p}qCBNHCt%QZj#LTrN=;IBF<W=I|l?lz&MFw-qg?bvkOV;@HqQM_y8|
z*)>y3<I(NmS&VI9l(rQYb*f;bNBH;19J}Ha5hTZv?ErbEg?^A{x@IW0nF|IwC({LN
z-qYk<^t^?r1%JdLVD0p|GvN1foutr|BS^QuZQt>fQ2r$Zf;Ct?Bn>=gvpUA9J(JS6
z<lag+RdMLIG!}&z!rKE^9Z`elPO-%U!Faj{BhAupfxD)pTZM3Hz|l0$De6dey~Y)V
z56FX<^Hmd`qYtEU$-PH^Icwm2|2H~oJX@)q7bK^PX7UaCF*fw{z+N0rsuwl5av#!D
zXc+L)4Wd==1o(zaV5xIyV7OXziV@abmpvpKG>dks2FJf3n~N5NU@r;Kx^j;KI`J-m
zcna*E=6C5JTfIk9X&&V$DDr;7`=39B#leC^(Ex!wD8>-~0ky<`%8Ygg{;Z`K8Jy7b
zXtgR(9>SRzX|WI}6$(!Ti8dbXAjaV*ck*`(ZT^=Qi-~%$6FYP_u~nJxdaIX3ho;2d
zh&MFrll)M-yz66GTe^gHU<bm}3QEIuZ(l}Z9>TMWobjz`ARtPM0+Y5~aL1`{f-!tD
zkbwSd1@vMj9i0qjPZk8O29<ssB^eTXC5~rHQ-o-#2v<U@uDA<L$2T+Muf!%V=-$1s
zV~4w0j>f&zbi3n|$R)|H%SnMny<#BpPt}qGi*+{}4Jf0hcMGt)2<ucPt#75DJENzo
zn!sOX%g##q$c~3EbBs9~O>x-_*5?Ch=C`5(u`*H|l^382=nBo8fbX7QS5?Bj#EJv6
zbdZaOuYO+)Jv+r^?ROnE87*gY;@Ma<BU%CY$GyZ1X%-mk@^4n%9Twr1Px$VGnJ(+U
z|GKpsgVE`5Z!Q>Xb2!&#OX{WOR%|I3-R_Wq>uC-L>$cRf!xbZ4<^rLme?S$UJ83u4
zxPgc@&rZknE!LI-V;Sd-USEJg49@iO&c|p&bROdKHriFl4G)#_4=R%c(!8IFrM&So
zVAl+0O^r|Ogq1f?WIVLuL8wwhP)W(?oLopN_v;HD!5m^@D!NcTNL7^+>I7Q6fj5Jc
zlIU{?NdYF5;`zaeOKMye2nF5l;zJ|rb~}#_A`j66Eqpe5&RX>5&t!>kY#mcU=#v)X
zhQfl1nLj`esOTSbUNUih{I$aFd7H&LGV$EChM<4lgS2}=ZyPp_g|Y)X#H#87K+ZOm
zAW3S#m5A=<>VJ-r$7;N@GrtYR2uE01XAW@ab@e>8@YZpF^kpE~n|6)?&F2tZ7^v&b
z<GhCf<{E4m(ft_7jiU9YBj*t<3iA^sR)#;_+!-3=N+Hk{Bz=op9;thWV>@pqMEIG8
zVssL7XVC$+XXUx0#z7k}o@ryZ_w-Ro$Ive&_wGmA;Kr%@TGSc5Z*CuJ^P-HPEaSfi
zdXu+h`~RNdPcR}$d`}0JAxuNnJ{1m`f2o9w7E-;H+wkl~GQ1mn4K%YAJi9G1{W@kC
z8ra)23@^lrbmxws08$e8iAZgei+y(i(p_#(nx7{!T=|rABiXJ}h8INg#<NyfLq2tn
zG|WJX!}IPCGgP^g$^ZcxSE{!&h8+P>N+I%$%)8Tj7T;V~4_ja%HAuM${}}*@b1J2$
z3&6(107J1WBXGbO><F}AC_=J9((pC<{BX7L`vT-*Lf*Hy{a({2MVdh`p)m=JzBSQU
zGzkjkl+Iu{K$lcLyf=Y(5CXYlCyo+$83Jnq&?-I0MH1+N>SlOemY}>R);^>)lbUQH
z=o?|2s3-Uo#EvA+Lps9hJx=sv#Bv`P*oPC4`W)x_h^a|7@z#3HbyjOyNuGhd?L4VV
zsKLu<a8a%8=#}^oES>YyFZ|!gOp6?~DLGqf`hA;#fw8j1bv3K<nT#q@091?hW3rG-
zZWGx(&LM2wEZG>}N7?%hG?Q&kEnPm|sWW5A)fEKOJ}Fezh~)tU&m}5S`GySbXYluF
zEv>q!D?CYP$hpMXT$_L7{-KxGVU*w^SGPbONK}SxIifwYfOUc09^tZ)8FTkB7$~z>
zaS26H`)@2(tvldUal*u8+|%`AJr;@Jhw~V)USCpRy<P#FPz_w!=6TehMHuQ4elV@J
zGDQyg>`VZd-E5&SlJ0O<Mu&^C)LR(CGw2xU9!d*<U?oB&(i?~6Yn0J*0*4>U6Z=>J
zJj&lJlNtewnycP%{m_YA$kAtm0eptQQz$th!u%?T>xc3UDNCt9V_1?BJN(d$DOP~9
z^29O);%I5kHQ+-&0zaF35$XB54~T|b(v6-`j_TkZ3I1TZ&_AKP!oRON`(IjXD*BaD
z>WOaigvjRUU2oKzbY7hs%{_|bJhj8Kq@D^0V&<doEN^9pKJTWOC#}Wd74Trr!ECpb
zk%zwj>L~by{)GWiSM5h_QG}j*5#RkfvPSm1&U+O6LOSMFOzYfAgXheh4uB+Y9@YE1
z(7P3!b0VtWj}Za16ZjBmc<;mZNMZ03X|)Jc7G5;fM=!skUicQ?i;jX`T<>!cB<CZP
zA|``@<S)3Aq8c(LvHkq2&T`%krel=CifvJi4WW<r#7xB&zmn~4_dq)LB!Pq7$5#bX
zp(X^|Xle3E^@!ud@Qb@bB`-gfb8oO)%N5&+Rzr6<J5p|Mt0!z1m6~=AO5#h&xzKfR
zg;bVA7(WvXLh*HL*=wsO(0WVgodp)nT)<H|Zv;i1Ps@YQ%TVJrZu4wCg^ZfX=OQ>)
z76&@OrtsrZerU%|d}Dh559TGX9~kV#-aHSt>3nq>U;HC&v0aZHDh^wHF6BIa8~DKu
zFR1TKzpo}URw9@pvq}R0f+-QbB;B#iHJWP<taLO}3C&PZ2B_r2{;{HK{&Vvd!wHLT
z+2u?cWS2q3K)p@jD+ar3ovO~|1rlb#PO;nZjk?Z_%yU}`Yi?Huh3^%I<1hA4;y_eN
z{APod)b9gBeh>chu$?4L905|;38WrskGkAuRSumKd<Fil7o7;JdO@iP*zS|Pkh(Ub
zryBQ$^Q3YOU*xAObD~<%^!*uInD;_YG|Ou|&s_KHx7TwI#2oWq)2@SiHl3J3VRm4l
zGIE(y0Lljo)VU2e`~b!^UQZALVW+Z1;W^h|#J3j#F{ns??IOt!O_{|*l8U=?$J+!)
zpU=8v8^1!Gllu5wYwWnru&V+0UW+#C$QKN$f(99tu<UL$A=QAk4pa_7s18t(htORk
zv{|AB!n(S>w!$USi$r~X1_OCXU^WI)Adz>OG@d7ce}Y`A{2*jZQb7i;QwFRl1Y-^J
zla~#B@4ozJTBVg|)Q^z|MADA?Dv(kv2D9)(nJ^Gj;Ai41L*N!3*YxIlf~(|9lO&0y
zP=<o^tpRQ#4dmHS+5|Zm9X4Qs1%xyjYgPfR`;lfBryC18u+Ck)*<j`Ff-%}+8G^O^
zkC!h%2m&B`8QhLboFKH|dk8XwsmT*d|Art~loq>QH?YN@&qDa$s>yzyQo#3hYtO=S
zr>ax<;X1e(-!j5Y`aG$%C|-L3Y?DCnmFwKh=p&9{+=72-PzDlnMDnr!khT{8qQdQH
zvpwb(N4RQ-w%z==)+)>t%`C!0;pZ02`v-COXO0A@@ciW0Vs0cT`V7HdLylnkR!y-&
z-<*V|3vSctZN04pHV!u2<+hE-y;hXmJ&WO-4rJ?oXDyC8g`!(U(-^EZWk}<?4=#Ap
zh$G*BakSzwyO#D5wb%lDII8rV`if>p>RU*90h`NbD$+2Oo*w(1z!{ZPA&sP8rhotS
zQSuond3+kJwq9#J=%^DA)IXaf17Yo5CIL0jaHqB4!+?C*>sY7W$N=+R`@(4wn$97i
zf!R9pzTDrypuczGqO>3g6thX{^iPzRSl*p@KU#Ai(!JHO%N!%!Mtgzd-BC@aai@Gi
z8qzGJM*|-dn{XXaF!oc<usz0h==<-s=KoTk50O3q6XSjX@`><<@(4J&`+MkLRotSH
zw5|ABDQ0*hgwG8oj(rB-mzbR!9qCkdfh@waNn0TB+MlwR8b`CE^+~6|WBYtQ3~6{q
zn*M{y`->qFA%XHO4Ug2;!LWxekV@Zeu6{BK7}3GW7@%?JqtFPjyrv-G+5tMo!)Q9C
zbF~XZ0?%XsI^^Hz=}>ol!IePXR(0Ue408w_Hk##$Wfa18mgWF;GPV$KAvGJcD-*o|
zsu1fO-EP9itK)0!VAb<I08Zu?ig1Gdf(w7ag?|}bXvl6^Q)9_2sLIV^Tv)z)Vc05f
z%djm^H)YPyv8a_^Dl%7o?a2^q@fPK-Ng!@}g5<A;5InUJbU`?f!Wls#!W$I>IGoiW
z+RgIey%BWgXrgp)Uy5gHf&h-RAN@dyolJ?<ILF%@UCur^WxV;cx{9$=e=i_QjmhwT
z_bO|-j(o=tX`_fxcIe-|ynsG({N$e(Sc+!RY+UsXKOu*5FKBuNIjY<(T*W%L@O;Ud
zska}emAqR&XU<VOCW6i&Wgr~c9x_h?LXS(-aze>%piHV&`OaUy`DGplN@4mjlT@qm
z{3&PRQ87T+nfb#H&?Ny_w1(Lw)L+2=MwwOKT;lsNOv`OqoSuK3)vcnjtLm+-6wmr^
zU)Nk2%=LbQbpB?s3as0kQgG{~#Pq^%;9+DEsb}P&?(aC`KP|5`9Nbt~4o5&IL>d^w
zu#M~D5da8~({O4<y2`j)qvehWKNQ8)qaN3gg`kF-G=yGp>|57CWHqp%8W!&({|ob_
zpX+ZoWLLd-yvNM@neS&^^3jfvLs|iG8wqDBX;qbw)%E6&Rcq&AeBAL?kEvNWwiCW}
z9+YxTfe;EapX$uN2{UkgDbrC+C(BfeWr@#kZ!VH;+5T)#wiIRIC!3(jYj7)Itgk}M
zp!nn>Qj9O86MiaBb!M6hw@WiF7p^RPJfMFT#G28`(6TN#i2JM`vl7y|(Bxqykf|r}
zb=D)4&|?yvbNi%$`*6hH6g1fzcr#dZCh=alLO;@1I4g!x7S4OwB@0*Yhe+)IA-9$g
zPJAT_74HK}(`m12@chz_;y5#vB^6f7Lo4<wirjQ?Ct>b5xZ5e9(<oz*yegm}DzmUa
zsH-<b`LVYMK$Wf+IBw6Gc=i)~)ncX+UqTj#P6qS~bBwwQM)E;?pd~y4b8Q?S@RVLX
zP4Hq>=oP=)CVFxHv4q?}$4b92w_V?%SP3JJ!Si8^L0UF^qIK`4`q;ISOWILA@>x7<
zo(+k!l$fdZee>aX)1k!P?KUJ$Sn!`X<!=PG_&5ltX^=g8CTO5i63G1eZ&=#cgKIGZ
zVsc$2fz$UI$e_)UU*B>3SSnC^a1ZXK3naxF6$?dbXfu%7Cr-I&x~ZR_V9wRTv%?DA
z5`G!d#p(iq%htg9Ld*zqi2Wp`2=2BUgd!iTk)Ha;NaI={=s)#iqLGNlO`DNoT4CV{
z=(U1>TF<+JlZ3PB6(ijTlU=rrL~`9C9p4(ljl^d~eRoNZ@XT$5K3N_DVx^l%w<E?a
zjQ+n`_3CGolBcHOCFg8WaN>2i@4-!NZ8c|qx4$}5`Rb;(3hP!(IfT41g2y^X4bPjI
zCdlEAPnNz0vt41dWy8@tkY`$oeVZ<|KpUcv`{}xXdN1Q3QS1hWDs4T)leLamu9ywv
zNtdL|r^X|d_eF8BR~~5AYzOd>Z`>h`k#4}j{;Q2(YoO_xr_k%!7Kp`JCI#I>@kCly
zh9iwzPl&XPOhrNNba;U$Bz=chZc-&jB&>9#`!ORVwk&dqDr(T4LFUeM#0#iPP@i*h
zS+igThcofq5^wZ&n_ca@;fJh6>&bG++tz9<cLGQs8tP(95~#xt0GBx;59P@GI*Qbm
zeOzGRhVnzC8N?Fg8MT}Z!NvaZKo5Z4r~13z$C+gZU(b>MfYu`O5I+2UFxxAof*%Yw
zvmCLea>q5|IV<YiwdV<JIJL;{GdU||8halTTP@?4WTVAwG~vWuU#TY97o)Lc{yu#r
zm4(<iHzRr(A^~T}a052`v$KI9`Y9CYMR*!YEt%kewh_Sr&E`Sk+pCs;lh&{pB#)J%
zh_Bm>d>vc@iCIQ@A;TZ1PU2z69aI<gu?~e)l1vg#!`BVBx2$ZpSw)=Z?VCwAraiP#
z*hg4hI=`!jS_eo8_M7qt)ouLSab~`jJK_rFB2r~%Rm$FvJ(!Yhf4;1CO7_^ZtvWbc
zx-i{ET@K?L!c!^TtF00Ca!;7I7juP#<7v81wMoz9XatV|!(zSu%ItNgfOo`%>m3-j
zmQ4ytmWfX<?J7`>j+?#aR^Y<8z(bb<E8X8;ORMbCnf!xh{pJe)4a$K2C5;$F(`mNk
zzE#)@w4Qb=5Ca{fcQumoNUh*&7N&K`pvf&DijC1kUB4&2+5}rQ=d4pHnmvikm$D~e
ziDd$5PQaLsc>WbAMO#>sBLBT?>6O+FIEP4OJ9kbSzWZKX3XVGR(Dz?`kpi9~=Z?#G
z$SzQ(A&m~iEcKDXrSI9lYJN0Qd&?~sLhVIhUP5$967*4oOW*S~qcm-x0?{(Ey~Fok
z*JYt_DR-bQ`%EJkECV?g5@{6SQUiv%ovj3J`F1Lag#$MPH>hj0Y~DuxQ>6B5k}C*5
z-$NSTk{aCHNGKIUx0_`%dU!O>Le0JyFf%w7-0c8AHaMMqo8T2OSGpVD0uVq6#{@0b
z8!)#ELnL%-=)55O1sZF4lR$K$a(dmF>}+b^GDH$-u>}9v{oJ$@rua0gekP;aZ>IvE
z&j61)F4N^UX^&ev(s`l+3^n-Be$1w2i$r@hzvWvB24qq6mESTijej(K!TSD<Ltn(z
zd>cXnIL3E{u%fZ)>F5*Xrgi1NA5XVD-tg|I@r@Ph#R_FHaYEuCoRDaU*!2O#{Hk1d
zdilT75vDdM8~oTdNC_78S(pft;4Y93iSQ>2izc?&O{(45Za@FYWq(6=3{SbQ0qjo@
zKLo7a13MZUrvW~QbP7#n%t%pk(dlTvbRu_-Q}_<Ly-N3*g;$i{-&*)<z{iWH79H1F
zt!R~(C)Us^Pyq-Dx}dOq5aSN-w6<Gbts~yuO_ptoo8h8Im(eXavaI;s8q<kZQ`RbM
zio9Ta^Wcsvo3iE37>?%;b_^!L5=r0f_8#fpF&})PYdiG!jDAf!O!;!$T;&9VO?WVS
zgD*Vt6jbWoJ^h{#ti^+**{Ezf3!v14i!$h>c7*-G8Sky7SA~~lo2F;on^?5*S!k2W
zl61{X#d(*PSt2)dqSF`8dEA&iT_jA<GT~1)>DccSpNURjJ6CyiAbvez`D)LO-bk_2
zRZM+X`TX*)vtDCZ*>ZMok*X;CK^kY#HKa177fLho#EPd4bTO8$RpA~$-}{0;ItT-Y
zTxE1L_+H!>R^iw9cA#e&dT|&hB5%4Sx~U%W{_09-K(gCvdh@B)D#&(EbZW1}p5+lL
zUANkD)>Xu>_W8Qse`?n1F#XvJjQ4r%TqE75@SPVu;+9Y8w8y`o@SU0|9L?fyJ;SYY
z`&LoD<f2KAxqLZ0lH^}l94zd=FBZS-mL{+Cj4AKk;hEiAyv;3Pyf67^USyyBeC4ez
zXP@e*JazE&;nh<in|YAbr+|-D<zG71k5PsSptS<gEEz<z2pf9BsuQDWm5<PAvFBSW
z*{4rFOWBMoS#;*|DKFYw5Yy6e?9-}eX$^NEWnT&XNQ<16&e0B&Cpb~7WUgp0)vrvv
z<m4i~z++uRav2?W<mRv9uXJRN96v5$$9$6VnGJ|`A-SU-P0_-Grxer|e3Q^k5lHA@
zbB_nClm$@?h<axOm&QQ9b+aYbLVH~f1jN_{Obx!&B^poA%vP2fgNnKw_(ZcNjIq^T
zo|m*%`x2k>cs5q0PGiBl%sF9a#Y2wGotfQ5t>2!4vh8bd(oVBk(>oR>R+gLA>lNzz
z`Y%6bqpZ6u`$c?N;&o5YD|#3=fM<g;Gkw;{`~4X0=O7%Jf=ctckaPny4@{9dgNFsz
z0+WelNi2X&4=%>Q&2agy?ZgY2X_LUhmAdzJ0u(k@I2=v&EU2K}B@DCD#@vbAx$N{z
z?bZCdQE`l!O?1tV4@#oT4pq0A1n<1qOLUlabgiLS{&!W>-XOmeMV>0#$TG}=x20o#
zgRW`U`NwuP^<7h5?^vi6=wf-MC);A>2aIDZ(GwYNIpO;bGak+#q$?TBV%&2EfBGgG
zljIDy->w{_T=sSw3$*mBO7kDv1ZO6gn{Ag2#ZLEJR7#q>Voj}RtIiCm|C#K1><x|Q
zbDdgi9oJ3Ru%5pd+7k{cXNmQ+*1itiYSr`{H(YupX&R|vc|{3Bt@+?l=EN;p>kSSo
ztj@`Dwd+b8<vDHER*b*rY&~N_$%dJ=i<IU#6LCVm7SUob0hGkz_|vxm@D;5SkQgJ~
zr4DWflD5S-pay$>N802R*_UJGC(LhZqOG-S!<Jq_dv|q+y=f)vxZ8H5jW^}t2es$Z
zauxlxT7J*yG?=zoJC4c!UatdiY8*s<0e&<P8Xcj)f{Lu(!X=~P02GyQ2;lm1*e!a*
z3<%CbT)-vzhgUMVlN%WvBe~Q3Gh%O$z9nw3g`R%A3!7?rp4hYh6-Ac+{JM?S_AgX)
z8=d6?_sSwb(3Z@uT9!L92E?$vzD1EWY<Tz!6*@@JJ&%N5e)&zK;Ei<d$3y(vvMx*z
zFxp~%gq-<)=vHw8cM*<zMY_Y$H5A9YIi12X#P`lwBwF1hknZxU`S8iL#XN|)NC`mS
ztaFo~x6)d2JEW>;Z*sq^Oyy8Akg3bO?VniK?iaH=9bVSxxJdbe@!QnBwP)#2iJ&ra
zyIRAte9FHeKyiA&@NgOEN*(zw6aOsMu4(#Yg@}d?)3A*s?GeU+dIx_5@9U<N7fbMx
zZBOgR%p~n@l<j$RfAzhrJ@(_?DX;(fLg`)Ek=11@R#=O66NgCP@h{@AeLX=ao6Mxv
z-yX5Oa3eR5C}$S>u@GNhOya_|#;V61n6acXQYB#wcC2Ix?JjwQSuCoQGcYW6o9?1r
ztZwI>n3BS53)1S^lKJq)9{X!|%l#JIx~O>Xk-O%SpQ7j=Kh?Wc`zD$338;rH{UWUv
zb_<5oY<yyrgF171e$=YUKND`cYc#Zl{<oM3|F`=mMcuU`1nrF5J^#0`y{pVtr;a_z
z$rHPoZ?4aqA5ZHl#?yI=Ak;SLMbBh&Kt+ymC@$a!@6)*9Z;4mFOw}%m;J-xb-<6{L
zeoRkO`yc#~eM68})4S6yH>*oFy;&wwqvi7zq|mv%J~M<tLz7(!pvonYNodKfF`r!(
zeo4(<e;2JoQL-#u{ld9)*$*qt+|k1qzDz!MM>IRxRUZb0Mz`~N{Cc3O0h3jDB+AiO
z(^Y}^=`cBJnY)x(F6PFpF?UlnR~bLmQBa6JhGSCQrbEP`&19hB(KNPC50<S*M(U#w
zDjNb-p>uUYk4hp<9BH69k-@l`FnD@@{~uRXGkZPe<u7CwYER9#{IqO=RL+5yQ!iQF
z#EkWwYJNTyb3{Q)JR5GIGcz2$C}>yAG`8=@WDp-CZ<N7<t+V7KZ~O+nnW0z^py*__
z5nv2%I>KJ^5{x-Kiog5MmqW5{qUt%7?$a&`q1r*x`TlEXJUv(PFl_sQ-CJ5D)+-#H
zm8Ko{9vS`_%JkCpiJG>b+qb=+=TciUGfUHW@P7U2Z#BuWiZ-UA(Zp1`-m}}lS7Y0X
z*=Bpx*cKBeM<?h`Ke*))>+8$20Xb2W;Sb`B_aCgS(+E$QaUS1INYVg95=lnq`~BU2
zdz3@*PLhl^`&?oN4~<+|Fn_A4{;_?NBgj*;6fT-?(wm7}Qh97QJgPrlLPm<r%-S8X
zk9NP+A-FNlV~z(i-jL799yq2Zk70JnD9V)>Q^^k<K}tU2P6IVy1<XDpjTHAXEXuk<
zJilj(JV_!`I^)CR2_6X_Gp8$E@0{Ziz9!QZLS~_bTGd0rIw@p0q&rmOCRistEY6Eg
zA~^1w+&X8m|Fa0k{f}R--jSM<8ioAbcj8ao#RP`8)|6^<y&go8jsLZX940Pb-A?d2
zeRI(pk51*~n8^}O(%`qmd#eb1sdjMrl}}3hF&C(iC>yjaIHpep^o)m_O@e0GfRu-?
zbmMZ3RNCZL_1R{Ls$-noN~&k=sJ@}HqFm9p=qz#L1o7wh*3a>@@D{#N{<z5&`b#z)
zyxj8JnS0sLrVxM8MVi^XgHBeBL&i0j`qt*&TDI&&!@^))_Z4k#4dxv4x5J{+?QbX?
zB|+=(G-Cick{0nuRcT`4>mw$c3O6};5ZK0q|9YD)Vcn|2d7n|?$;S9q-W~sj^(8)Y
zUCQmTJG0W+*Yb52_l{4Vgi4w*`kHag=yaj&@2*wdn?{=AsGQRyr=OaFOGrq#78Y!C
z;{7TA&2!qUP96}r5C7wYg^?q-Z6Ur&I=LTn?-@#~5;ESRu*j+`cz{74xK`14(q>49
zba*J6l&URXc)wT;JKp-tL4Hd6l4?4^YgWMVK>HY<nFlwEiCx_k(>wZiTr6krxA5b4
zi{Fgs6izA6h^W<*DEeq={;;d!1v2<GEGBkvS)R<?mfInX_mn#x-98;z+I(u^JH-Ot
z^M)p`Z6E4lEobgMy7>dfIkPQhf6_htbF-J(`)+Iv@GQL3^CaACf!vh9XM3hcVpY9N
zDm`=OUU~3LOX>_oN=xoOPw&&Nk}pmV-Z6|k;Npt2rEB*eAeO#9g-M+=ZDk()pJ>@f
zaSLQ?S^PPz#mCec=aZB}8w$#l0u^tV>@Z8r+O&37sQ8vzQP0eiB%u`@I(RQd47i73
zp$EqExlmt||M&RMooFe(>nnH*VDUIT5mphm4wDaZ)h2x%yLwmZVM?`J)1Zkq=9-Uh
zn`97qykT3ap#)*}cGpFs*gyA)-IP-OHC(yxU%04!VDS&V$6-QO(lF!;pBlV);l;if
zv8VKPf?uN{{4+Upg*+bNw_wu66LK0R&#$ask#@>9Amh3FuGO1G7yUN#a?NVRQ!##{
zpGk{0V$+&~%b%DsRWEemgA@$2_h#C5i9oyf0KdNa;AYz35+yBjau7g|%Wj$#ZtquV
zt_sCoP-|>!y~kRQ{rn|L!Dg?A;);D9Q)x;TuR^fkjE7q#;tqR>+u=4hILj_RDzVXY
zjc|}lzN0*L(cZSh^Y?AstXO$`vwES>Rr?LxJwq{7Jx5`yC%i`NhGfdce*N^QK&yRC
zQPPK3rIf1N$Gy=_BNFCzd-9K{D_;**G8Vt>c9;|J*-%X6k3ZlY$<(?e>qd#{g%g)l
zY%?E9oz}=(kl|eT+p6qw$HjCnckQJQ_p8aequ>2CGuYdh?$(qN-Bfw`uA>t8=g}uS
zKZ7Yc9dl~LbFtWC$2Dg!goZ6$ZaL2yN<7%UR{o`}8M-W$ywcfk9O}Spgz4w>CchSR
z*cp6KI&-v8d8S_RCHt!PlnK@^Y*rolly-<UH7<6iw!w;DiTj`B<Y1%%{%c=7bit0Y
zum1tK*Unp8=}II#{p5(BU>$Tr(=_2rn)<mFdWWA+-Mis(nB`=hw}vav-V5~m(~b|m
z^*`V`B}f@=lFS*Q<cHajPP%%9J3oC|ZC_mXW|m2XH22xBv$nsbKff%pmp0<*y7jJ4
zOfWb=)az83aGbEZ{gW^-bsuGm>5n+YA1sj7uH^KJ2_5RyP4Ws%uS-wL744?pXnQlO
zRx-R?=aaHrsMh346a3$Mwo$&xPQ6Oi^LbkWbStqE`iciuJa05f3-q4rAvM>hJ9l8i
z4So8RJ(S5#lQPa+FUh^&GW&j0Q15J)_m`rzXVtm-NT1r&^!w2ek$_api6<N+opg?V
z>Y3SQCO7@;fz^sOCZNyCQ_dH1xO;IK?b3UAl1%w$9mI^OIj-SOC+8hgoAGL;=8Gt&
zZC{CN_7(bIPQG9!R>|KzIoG=i$ih8ay!v7z_tpk9${9Up64T9BbuxAoBu%Q~nkDX;
zvHQ@at?`=`u1z?2;-HT4b@%mRb32vqv86jG8~70}vs5!EVxP+zSFP&xZ_Jv__b^gi
z<Ud2+NV9eRjw=pn)yb5>ttwI<9OQ}0*~xS@1=Bp#CN-BnQeE2B_|#zOl`}6A7avk8
z3<gTzA6Q|dfisgU0aYY#&iGIX>U}5G4F%NJi_7<#+2rY}9q@P<`M_(|*eh6FAYcCo
z#K+OkqO&Z@-k&~QiN_L-t&R2HyfElUTU++h-!msF2KW;c3;+JTIm*U9fnu3{?y;_C
zW|xb;kzIDp=}Fa*rBU<ZZ>=}FDmQ)VgvzeBG2sLEOd8pmx2#2VT4VXlWlm#zXF~bF
zJUli<dfH~TYt@T>%uzv@5iruW0q;79(uJ0Xjy!4P9gcRzrKgB7yW^!A<gxi`Im_-l
z+XNTCnQ6L7G0o}D1CO_#XBnlXzKy9L_*z4u;z*zOcW=m1wvd@D8A72@Q_^l5`KkGb
z?c-MSB433(^;*=XFspBYv7y{)tzQ%Z%(MUK0gi~m5>!;HG~P^Vo){Cx6!Z4nxEqsT
z9JAWcIotfyz&Vt^=3XYnF+30}(uH6pU~{Hhms|;pU<XffzPW5^#=;53ZKvW4(E}I0
zOp#O)cT$+sp(7tY%AIs3NwDR!w<PC9CCz(a)|hZj*U?Vy+~Toy3a=+WI}iO)lwZw}
z*E@y(13(63#P9$QKO8eR{gJ00<&WtE_&V#R5vJ%@O_5Z)ct{3&VClRWA6AR}>ZOu6
zsZ!+C<w}v@HX)$=U*mOn?8A5f<$nO|H<%i~yi`S-l5SVkIwO5+?8?_Oii;a3i(L46
z&KgtZe8+((1c!(WokV-hwJMFT6;<!tIj((O#QRU7y0ngQeqJtn<YFxrGb;k4!tO!;
z4f5fdSCXf1GM;XleWYb_&8*biEs)n9^z{GAn^9qEyNcr)m~wmHqc5eVrt>o=FMl&_
zb?A=syYy0xMME$mrj$QeBRkX|?8ZJTb;Nb%3Obo6T_A4@PW^THhopE;X8C2yFo*cy
zDa90<b-d>a3!bY*jX%C&N`{1H2T8*%mllV3B6gRek7TE5>Y-QAS(_B^G>w_7D5Ll0
zT&KnMdwV_}D}-DLt<(y2iKYpX_ZH8?Z|B~6^f@Aty=$L-9#QX)?8=;E=|2FM<LT#t
zw3S7gLy>vr(g&?s(`<^pDoiG4UOITj!Oyg4WSQV=6+(J1k^aeCfBuq`GSNFsOgBrQ
z-E0yvWWvwpa`6ddlGBZ~yLm=c$*Ox-ZFxV5^d<XvzSrERMsw|7r+!?1M;w&oBLK57
ztPn=S3eheL+3Uk~r(bmvP5=GC>b0?vB!YD_BnL*E;)csuehg2FYto!RJMS_p)K*h3
z+dIVU);-<^$s)#k5v_<h^;(@`(ml8VQ?p3atBALz($zJZV$O8C7@>ZBqP@uLr(Twm
zCF>9DXg7+JQ4;@UypurT_9tE8KM2=m?`5{1<!y5-p<dLQ_rR7O^~B3Lae~$LCzltg
z&VTWngr}CtpRwyEQ7#jiS;ef1x`I?U`S^RYBy0TFEi0Fk+qC3e+KLHU9-6Y<<N=6e
zRKrYe+7xr}a_iilbu@8IHE%~{u6OXg<WsIu?3KEO?F$aB*lTW+w@vNbB1~e!jQtqQ
zfo_b6*a+<AV94w#|5_WQJb%$~iHX`byPqIJ8*6la=L0^CCZTq_L4BQSDrel?uaQw#
zKS<mjdpxV@c+{20;!m|aB=pWG4L(s}_6TztgfO1kvzvlmG?`M)CgD$?KGC|gVA~_J
z97X>%@7E^ITbj|idCkHnf6A7ACr;#mIB~&Oe!y;}#g~8GYud<6UvXq-^vhoj)0b>o
zt>vM2iY9ce!ckqFgA=yOH6!=fu}Yl8`=wEP%3~8x9y+u}(f<}?1>#PS&Qm7zD0}-L
z-l&{m<SYA6(GGXcR%o^gn@0=jj;<PpH~s)fX>3sSj(c}cMV2(tEeTiF;oq-V>ZNhp
z;dJ!O(hF}Z>lKgb$T_yBBj;Fj2OSDFX|>6!B`d?stUw9$n%GSM3Nc;7_py@$0HVMX
zZ}VqGFP-P1JDXptA{$Y1?Cr-XX3QDU*i7;`Ys$3P19k>DDO0`2>!)OCeXe<B*hGDC
zybaUkp`Yw1e6^!K*?(+Dj0!JP0;+OFt?_nfI&9u>*A%gruehryleJGwb7epL9D01!
z8l$VuKL1lbwb1(BG^X4Yzo^fRDU$h7(W;i0&DW1Rw><OslkG*fq;;+?+b$}4nPMtz
zI*F|(CM7!CJ)U%b<zvI+PH9lwadpIE{6`4>QSt-I`>|N3`SgvinP212sqH>-o4!Zs
zg~0@q3gT92s|(Ar_MP=uuP`QY(CQ8iiAKTAe+qQlA#c`H?t6f|QPaXbKn~183s|Zx
zY_|{ft)+E+W+%y|SuIv`4kc0t%+KHaIwyAW7P57M7ZrNDYptuB8jky7l6d+slB-3B
z*kTZR_c?45_=jq&pFjF;M#8H5xC<fosy{PUd$;ICRoaEmy*T;il5^OAn3dUTB2CC}
z@b~!@q;gc%oC#QSwyqv!%Tp_7XZ@$=qpNHtF3Zm#?w_IPuQN-r?tkn)9=6q%XKth(
z-1n%d0@4)rYO%{pbl06*p?}2pa@3}G8${1u4Urgk3ik(AXB33`r{HaOUz>gGZ57R|
zH*boSt)utvez4mkd)lHm4{cwG`(Q8<m@r|>OiIaH)|UASXodZ|KW?t2y8T&9xjuR-
zzUCwFOQm=;*+TfEgl|BII7tXk`L#mSrftILZ3pm1936R&31UH9Pf;dwd!Z8lp=>?|
z@ZPb03}}5gJ<g*CVYo2cM>U3CxaVo|)ous2{!-b!CQ`QUp*=Z?`a)Oroy37{AmmxW
zj;vXWb4M{nsEX5-iRx3<rsvEFSa0LG_F~v34B?k2%ui#Er(Z4>1L@|a5~bF4m456!
zDU#1$eTvd9PzeffeY!0L6L!0IYjbS-%z9X`6rn47rVKRMA7oJ^EdCI@Jsp}Y4vD)M
z66ZC3&GH8k3XgO$3Eu^Fqs<=>l17U%9Z>PJg6a=uE}C`{;T#R7IW^tQx*J~p_Jw&#
zrrO0fFCIJ-rH=hE!F^yN?T`HRkoUwy=svNBR!0760(z}CB(@@!cZDVif(6_@kUSiq
zmOlx%3%6k{2!2<re4HmGyTnTVmGtrN#pA=0>EgogdH#?VOhj@PGVy$`S}0Mmq867#
zJ07%hxv%=$m8xB}6Q{2hD?H?`MPNaY9$?Sj19JNmVC^0rBPJ9mTv=_o`+?yF_i@HE
zF%s?~c}5LG3Z#T?$G~xFTCOLqGms&Pd^Unpe%ce+dnMx0-SF(=<ekL}UaeWE<1;6K
zUYB(`WRjp!#CMI<(4TG+4Y<xA8>B_)G_<JSqulEg;k%X~Lpn0!*`zBeM=}|AEZZ&D
zt&Y4dC?NVT{}FR&@%p@As|-J`7-XNrB#Fhw)AQ1gN$DPweLtIN?0yh$Xf4`BKqqbn
z(Z~mA{+;a#*UpOLSQ@uRs~oQBJY2PD>YY~K4VO8U#y;QV(m$Jmj`mYouqgClQ3Nt+
zXxbTl^zISC<(Ju%xRyn(D~dAObW{AJ&+nKuGZhaw=*WzAis8qyI&i#8CE@%t5P5~j
z@(A!ewUM_-wKw0{2G5cE;8r_zLRzhC#JA<)8yIWywIlJp-qxB&M;DFx*7IwW^@j7L
zFgL6AErIkN(jtRJuhZM*j9wPRau-^?i1QimHqlplGhFYxzkR)(eyBKz{1owTaIED@
z=(RFp(^r#(=P#1i%e6>;^f^A(XO3vIr;hwM?gAv|8BbME&jpu|uT0W4n(wnnVfViE
zbF6=P5{eP|EJ7ICjs8c`+Zf@WG%;<te$2bQaGsXMQK0jFbcSE^1TMYNkyR9TVecHu
zFZ0`TX5v?^{N%puZ!He^{~{grt&oRWcwQj+hx2tOR3t&s%0V~;JMtK05BcL|`Le+i
zmhHD*rm)A-?MdYd%401f`)GSu8}{=yDSSNn;QXw0%td3)2yZ)hjqe|&E&0z4LtVnA
zcUbDO7cdL<u>Gj>?y0Q!Wq!PgI9+DiFPk3hbnyLs<93On_rFyqo6uH3#`v-6`M2zC
zRXFUMuTned*Y04eJEOm1aEEj5oT>TbzJyKK%%{1naN3FWVOgvzaFGp;+TUf6uXpF*
zWC;!Zx|O=hi>@6d{%#{NL5aM+8`~1b-N5#&ZVkVnHKjv&yE(2}>}=hPgon!KH=~Ev
zuUMog`g+3AhcQRroqQ$QET{RVbW1_&iT8?<rC&0xW`(*($e!Q979riTm}yxBO{D65
zRF)#AAaSI(a&Dy*YwhXH*%3~sTsMbc3uTAhlOd)_@c($V{4h6Fw>(wuMz&d{QVTb_
zq>)&1<2pTlisRU6>2-%r<DELwSI(8x&C1XDO}YCBO{NE#<zoJdpX4^z$+HUQsqYil
zY_KwHcXa8@!!5<rJr*xJcyhVNtt)zy>K0b>rX77KT8^&*O`gXc1gf^GXxuI`EFcbl
z8neu_Ok3_#RmakZV^R8cV=8iHipbY)92*rfZni<~Ws&TCpJ=Pp8UL@nFOP=u@B1Iw
zx5g4eOob#QA&Lyio+io?G0JX8B1?vmHS5^7Fl4JNlRc5KOA;Z=*i9%)X2>+m((md%
z&;2{kxzG2$@8{3^Iltdu*SW59&77If^|{{L>-B!U@d@M_I;PPKYC$hW@>T#L(T!&V
zh^?aq#MXyL)C*2JgN%%YxEGfwY6sRu8+ov}pQd2`Z|cS&<?9a&{k*l@<fJWLIN#l4
z3s-ybUjrTf@eTiTex{60PtJDHE_*p^=;7g+Bwc@PS%aW2>KbtGQ20gZe6D7PM<k~u
zbSy)e!frwFTYl-5VDiOw>XUcAbeVUJ)Eh3*%y1qju30x00y{8O<2=d(GK2TF>&=sW
z0(INEZnWt*e%#*hcCvEQV)hI<NUImxx^8>zP!BKWv<%K+V@IC6Lh)FHM@xb!hD*@c
z1FcKixrf^~EsmQStvC%1N(t;-{PD@SkL>}_Q8moJ4lWG}x%H)FX|vib-?z}|N;kSI
z9IEvQpv40c%Ru)LCQ<Sv<t|liJ*!ZN*dGahKfd>U$uu@E^kcNs;X2;3%&5CZ1o&D~
z4hcUiqB;I)XmA53XiHN3s7C8q^GZZ);)<QUx<iEOlj<7oGv3*rvY@xyhdG%d<x{E}
zwkb072yldh^kmuowi<-*lXS|?$67nP%z^D)O-vsape|hZ|8_qN<Y8N=>|3!x#)_lc
zm^&XNGX1>BzRn5?OEkZOd$MP|hhzOF_GMsBje+rau=b@a?hfYTB+wu67Z&qx*Meca
zxFi%WgCposn;{8oWy&$7o_Zg5`O6rVsK>FN^K)Ry`l2gy-kv8>=~&nhluKI%d1H(r
zAGCYj+sAyg>TYdQ!yB*Z<1rqt0;<Qs%5fPH$Lbux*;*s)1DM)hTWWwM#j_KJlQDNP
zH`~;O^V7fMt&UW`igVJxN7YlQ%deGXsqJ6}h0c)>y<3am=y^p_!!-eKut{6%(-aOB
zeQs4J_Q-6I<H2MHrvAJ%s@bsLh;ZXotp<n(#M)y6vTg*HMh6fh6d~Uw36-U34pqb4
z?s3W8do`v|Cz#x)kt4XR15rkv`GQ(3fs4>s%V_$<;bc|2P$g@D9O0db3Av_T`7YBl
zQ*S_Z0^0Wtz>fi1964+RiU^hJvCUX$P>=Jk_Av~xZ%Umq6u5L}KR!^|)aLNqv)G%%
z_xE<tU06ZJoE3%>DUTplF3qFMD1L@xiX~0xTN&bWdiRjKLLN&~bxroSUhy5ho8yQ9
zVrjq@<Ev=i?GswegaQ!;Km4hr`0$L^;T2;`h%;JXV7EAlC~-8cH6zCHI!L?!?-us&
zM{EiX0mjSPQPe8jfxR7A(O(q=5T=_?j+r#4+&9|!5MzC(#@Nj=QQ%0ebQ)je-c}f#
zb`?rY1LSDhCBIwue{A~f@0!Z@^KjO$X08SMmFuCgWgye#0x*M%nOM?&9%=QK^rL{l
zTxLjq2h+1N%H(Zx@~;%NaZ;?+Rb0vNVs7^Bfcnz-D2>l)4=8RWNTZSE3$^P;T?kjl
zEgxUOBUz`UM7zNmi}x+WnKiIC-$aPNkZx?zmA)xwXeG*1bxv0@%6b(>6BJ$#fwLfW
zh%X;b99X>-I~O8)kH2~CYdnaq`S(Bj&yHBYkv;tl5{7!BpF$c%60~{3h8?78N4=0r
zPAoh0run(&b9I7B+Ya+?bF0^nD8Vugtk+q?Bx$_BP$||INCc&O85JSP8EN;0X8cM;
zyZBXY%owaeMDTnu56LIc%B)CR=i27od{C%f1x1O$Ws^-FenI)wjrFW6&b1Op=5+IR
zpnBRK8id5tPhdX|g7E}GbEl8QcS$~aHB!c<-2GtVX?0`uIgH%lMbWR?&b4pL7wrxD
z_4k{aL5!S|(ie3Gou507N2ch0(=B5ESB~9Uh_L;%os(j}G(fve*mB?%zt+4spz?4l
zF)E48!BAuAhOl-feliAmhhC^Hj`RhHy_Gb>rf24{T?>c{<a;+I+J*wZ{!n4<ehrs)
zWCMNu!J|<Gej3IbHYIr$7>MXF<$j5$-opWO!Mf7N>10s|_LQtkOU`1-$L?q=d=b;Y
zENtC=Xy{!*Lv!{+g>yPd(#dO{&rc63%Z7u#=F#3Yj``CeKwo`@hpkzam6cVYB)=ZZ
z>Du072<J{qejP5*B1jqT&KNsr?2Dsn()7m_@ARMktmI@Rp_`{7Y=|4`ECMHKX_7R_
z&*?QteXQpKi<Yi8ad&Ax;o(HUWV)U7FZI#Wyy@>GBE3&MwTOQbk+#WMv%R?84^D!B
z+tu`_8eyY5cI$6r>ILt(Ki;b=HJ>|E2~q5l_6+oLVR`k&w;|?mwf(P6)-t>Am^y6a
zO7xY1u!HEHQk3Ac=@dbLnyU|q4uaB-$R)=Twx5Kl*3svF6jXl@>bDR^7~hhluwK_L
z4+p~Pf8p)qzryj><TfIEhuPT;!-m4kK7Bj=31(*m+xn%@sXqY1b){}fK>EA>xuY_%
zla@Qzi>WF|84AjcB()q?P%geLgdifUci{8q@{GhtE}5fq4+LbyKSYQGJBU6l;=xYc
zT!tmJpP})k_l7~gK&-!rTeZc0N;T(55<dUp(3x;HLb4pD8PbOnMG8{P%hz#WCD!bx
zkCJS;PeTThA4(|?NarM5KDsznIR;f1>H&<VD(>qHnjVA2{p6?|z%3TrGHPXe(UD=|
zpMPqi`I-s>@VT$U)epRqv`mjMMKp19*!x9>_bi8B+9LwJUP~E*gzdv}QpGVdQafWC
zo0uoIK79r9oothYPnB*}LC>}<UKjS7-=ivXuha;fHdqXF`${uIx-4T7;lvw0hdi<&
zd`Ls$uI{sLa=h!?*NuK^L+{=ms~3)6nM%P0XIy$QC<Hndnn@HP#75)X!Ld#4TEwo#
zS~vHzn_nGA@pt+oKGz=O)8(mt(qg2+ejF5{|BpTnsZKQu8$t2?RFb3wl0zu$-mLv-
z+O<OeY-=V5MeV7W6Cawa!^S;0A9R|B5k&5U4#0sD90!;tjU!qp#XoK&?b;cPipe8?
zg9gt>^ev!a8MxI@_d`(ebzD>bX8&FGUYJNw)8MjOw4w`|dau))&>ZBH8sTGLV;&Kq
z!&3oDu|40?|7|To3tj?P=q@q!q_)c^(yvb{-fIodzt4Fl2H)Yv(}M4=WwAa%TDaCc
z5ylDJXWDY(^Q`-&yxr;cp%WIDJAqO}yB84d`2kHi*_+*^_*rF_u}9og_uXlp_5kQb
zUQp<lzf1M}lOv0W^|{-&L&5&b8jc33Q$H=u(_kZ9v<QkGiJeTR7dKpn^`j1J-7Wo&
zKW3XqPSp=ZAjBi$)9(7Z$+=jjTKjsXfy=_;7^936Eks-g7~)Ab*?y>+@Z!Wyr>nr6
zY)TZ<@P|=miuG#$CaxxIY68ll>dn0v_(ZkUUSRr1c__>#gHw(3Yi>@ClgB-L{2R24
zjX-HE@Yq75iFV!4CUP!;s)<W#SNz=Q!&!Bcb8E9C+SP-LpPyA>Ndp`a=G$K#S`Fv?
zgS}znX!azr-9_tuG;jYR5S!s+-K|azdI#19X6Of|#44|hRK7JTN-+e{Xa0HM^JlaC
z#{@skB<yluEYWnSyq*Mc(9^PYcGSOLB~)`(+A;NN>6&;>V!lLWR%b6#iK=qakm`^0
zPgdiq6buJ-fbQsrvjPQ<h+<C>sU`I&B$PYD%+$f)0W{%5a|S51{*><q+5}X&bXk^6
zrwl9^Y0FOs{W2qZde$ZkDd`LQW}MTQgX9eHR6W=4ge#Y794~msJ_4=F-r`;cCo~lv
zC#v^UU-U;km1NJcT~DbIlHJc`d0&)r`0I(Tw3co;{35u@e}mgw9@aRfkYecLCZH#l
ze5lk@$Z(N0%q8e9Iu^B9QFywcmOiU}{PU5Ik>RkZtrc0xvZcAbQ~%DS?rxCl=}ky^
zVx_@2jh8sFuZEU1L31LIriZgTc@?TZh~6xy-HrU1!_0fY4bKOn0i3H(6b^)P@p)Rg
z=zXre!nZ8y&G?TgvygMUH&WoBlj&caEN_MzNxOz?XVWujx0jWUF^1=PNNx#A7>h5B
z;<qK2mvxFu@2bXga)G{q7OkTh(>T#E#UPQM1zzoA$Rp#+V@Q1h;=7|;)=^ip)}<Il
zhW|rWi=8x^vpP2&DyYJ!ff^bVs7cB}e?GURpvlij@MH*iofu4%jV>d(RxH^o3w-E4
zb=tW#S2n|BC$-T+>4N%0lf+ZpN;jC+;l!vvkv{Yrinu`EZ~hHh2@Rbl){QL-#fGaQ
zVDXbp2j~Q;xM+Is3(FW|^;(N37RpJ_3^^`h!?mobx?!_$mMtgJWbXp#+go&$l@SB4
z+V6AW^PT9vy+Zr^d>?=RPi#x6@FE<q_Q@c`GVVzO`l%#;{!lrBFYd+#mekjp&ha{&
zOoRuP;nL0zj5>=Bum87N1^D-0jAQwGfQJ2~M3KYAU>`xB<Qc|Q-Dsvs;l~9&8eWZA
zzIbvLwe>OyB}SL13F&P=0<^*>Bh_98>GfJ6P^If5jqg5!QzX@oEhY*{y?TB8B`5-z
zqA0q!E$A_?N-RV#Z=KZ78@$~R@#ECZ6bt9WQ2u=WDH1#%#U-kJ64hy-Efq9q?4xNy
ztjSZzbJ<Vd;yfZ;ez`j_`*!VSO!F4*9xBm+yS|ke+XYJ&L&wr2f~6A`TpG`tdZiu9
zOum&FeDP~hf^7PL+8%rp5E9p>D3ODRArz;sYMGTUn-L*YYGUv&Kc;ZIV`q*_p9H_<
zl+IG<VV?U}x$mF+SAK0c8rP8hq+J1dg@o!>6dMHwRZVJ%%J)7JI?#g@teA|vAK_R@
zsgrs!;MjJFG!)7X<R`9lv{zd&fx<Y4R3_OkbPpeUaR@P}L3@8cpk{X9XwdtBJMt2}
zw+%#(?&^rao5L!KJQnJ<6P+e=RUS8u*h;0g{7CBYIeKl+1%HP?u|%7!qnw6RtVf})
zKCu(F%XM=ox8Nd;GhL!(!FP{pr=BO^+YGPM<7mbNxS&U&3{bm}o=568kJ_|}-H6=G
zvZKQh`H|?HGbg-2HZ{6*Q2gnoq!ZCBzI}&Yz38~10b&~A0)_sg=lzFsNDhW*fu`*I
z>3zAo$}pe&?pj&VU+EE;cc<SgFlB(7f6Ax~wu{Y2dBCQ-8De5!E<J$k3nsJEq(oM!
zP-iL}lTzY#@U1Cy9;ZIqUQh)7xIj4-@Z2B8Oxay>cy;yCT1=a+LN=e^Zc&zWIBD{b
zX(0OuUEdZkj5+Je+e_#uZSn6I+yu=Gd&RptJ6GE1UYbhu#>e`rB^wUKBqy@Aj{wpD
zm55=ZIXxn-^{Jb5?I42Ms3BcCqCvmPN2kOYJGD&l_XoumxPb8jtwSQq*C3+W7bv4^
z%Ss7eAmLN)ZO`nk;2vdiYp-50zUE~q-84Go$V7(4He`<Y^Q!0>%)fn9b)oaE^OYwm
zl;x&>Q$>O*LtQcZ+PGxg-X|FSbbR{&g-ZI>wNQ_X+R}0%@T!}Sd->d-7(mq%;th=L
zCG{6p2l=?Js2?v3>V1geg}Q+$)+mX#KoXWA0NDal%q+7vA<jGPLvOF9(*yGvZgc$y
zjT;UFB&nt0D276iiUkpp<gGnj=IPPcq3N77z9OB}g8$Ybr2S4;l!f`lsG2v_VbrT}
z=+vW@+VK;~%bYh>oZ~GICxVMv#Axq`lhZWdiLq)<>_OO5Wy({_FRhPWQ9qHXkj7>u
z6>epfBo6H<eN;m*rYC7@QqFg_i4b=Q6D6gT%Lz}n=7&GNX^KP7#_OIZU{9&=eKInK
zUt<g})T03`!mawyYIk?_*|I9nXy@@p)0+Edzgif*AK-|$*)lI|xi^%q1JV7*?8$oN
zSKltA8%H^9-Gzp%CmtU~>N@)W23;fyZXIqGj`m{=6bitomIsjzCrBHMkO-7Jj1AQ#
zOJEI>Y4TG`+B_<Vhc(~d8|Eo}@Ga<Wa-vW;`>UYr9x=Bl+U1v&659;jyQMQ`fts}-
zg&V~I7qz=^mr;wkwug(^yP$$W!#5Iv$~@<(X8GRKaMDcQD|?vlstWdtnIcf*H6=ak
z6BeFlr=*jH%_nI5o}Yizo5Ig72N60g1B36l7_nX}bZ2q9kJ#R3qd>a>u6tAN;xOlQ
zbxdmMmDhG%Et;0W(&>XL0=IK^J{oJ~fxiDcLlhr;?VR!XzR{fV*H`;iJ3fg>^Atu(
zhax)LnZiELv(TS66u>Qs_U~2_+_SIXmuGGmmAnxjId5DiC&P5~hk4c~%Qx6CY-HOb
zs_=X-=`eK}+X0rCLtdCvv)B&tIdU>YKGv`0sm?%yic7AIQEQ8(EeL>R7AwSJVlan(
zCW-BO<u#I<nQg9K%-A<KUVWQfcIC~@<R9(N+t(<FiuF8LH-_u`<5g5c)z8YK_XBy<
zF~>OSoC7qUVAE;LR8ccX54Tc0MiS{0NK)w$bM325URFB!Y4oc1he63-bB!%G4xI`<
zbDe{l_iwQysPb9piY$47%mh8S`i1<fe}Q|BGPanV+@@4foV9!7GuEkD_sbez{Z$-M
z7G#YXplSqT8d=~CXqvRf&Zc%ZJ)3+T*K%%~YFMzHAM7x_4ZjYkgQJMow}osFzd`kq
zaiw-2uzpK3l^gO=?5kF3OFwQ(jg$^XF0hYGrZQ$W3pFWJqU_SCwE0K-<9XzDwf#K#
zV7G8i4v7do6^IMgkObSbtNB)_`6cW`*e4X<{V?%4MAW;zh)#0CMQ2xq+qw_<9~yI*
zccldgIk5W?Mab7)Q!JP1ddn5bNK-$;FvQd3J-z_%Wy2pc!GFTdvG+lA!h;geXqE^j
z@;|Kt{;}7-5JBkIXA~@>Q+?s!#71)!OwfISm`d@nBK|5Mqqa<qw~%wK<X}?9T(gwM
zC*M?zEr?>r-~{Sfr)C$(V(kEh`XsgpX-_o{67PhF(KLxFJ;K>UPljxqOk<lI$7}lc
zRHvIKGK}AqKJ5dcw}sZ&RdMyO4h+Q&6|=UqO<to(i@qz|-GaJOWe*l3-Lh8)$9(K(
z*UGHu#lSpkENj?jF`0s<RxP|ibE%4Pg=bb#FnM!I#^#t$VhZP-$P})+{4}QRw`D&H
z$~~GNCu0=?$QND@DE6eKNExM?YF^Z0<>CaU8s$QW*AVjQOAB#gTluiVPf%4?v$#H7
zi4@S(1|y|YnZL(x0AP$0Jh@PHEW7hJNRXkmehC*f@_Ycni^*0SH_@F@NRk^MR28}p
zON&u4OMo!HmMjt1bqV!^CV1~{Kx%oh&2yGonK;)#*8w+I>3YF5Gf=>-e}D#s_CpuP
zAl+z+oLfSbnk%0k;ho5%NNrH&yk>ARbTMG;^q$lN?gnBNgO?&Z6ER!NZsqf{@ZqI}
zBj3z$W)q5RuFaZ}TP{4$#?Q^-&et`c=~m!iH>6&pyz1`60q{6_@7)#4D3p_f-BlaH
zquyBqj$4mc4b2_Qs7w@x-EK%ddyE*J=ceb5{kr&3$ERz#QHHZxo;BbSuHB2MdoZB7
z#@A?e0fOgV-usnEQ!6!*Nid5C*K>Y&z=K`Az$t2_d7EftZQ%zmZW0!@wTy1cF7L>i
z$nq^c*DDrnL`b^w{PZzU5s3Y+j_|jCQ91b&EA2uPAxD0~A{M}VN>3Tmb#1sY)e??8
zO_pTijy(N18-+B3H#`KWI&<C^iVuO#zMTEUAO2KZ)b-N+mix{83eS-52bKsaaxP%;
zLy1hs&uQh&lfbb*;ll3L3^9_>bNKhXd?g=0mPi1);5m=!e)Cn*P;DPTVi~<qp?0z1
z;Bq88<@1Xmr*8F~+`||@&$Ef*?YmOjmWn@Kiu1AAq~?K^03T6Li3QXUh><45VgbaZ
zTUK?RW_+`sJCBXd9!Zw~r?#{(L~a6bVF9v{$V|2-Hh0kVEJ&f`N<TmbM;xKqyV{IO
zRgrq_YBW;6$)lrSs-nC;)d8{io1_iXKS|yHTxVaE-5diT*h$hPqh;U^=xg<hc4VR3
z`YonoNUCSzv>k(g%ahYp>baipvDEue@^~LEDU=DA7xe%jTLzM5`cYy4Qgng=;6+hi
zZijV}Ve*?ZrN+?7;NjKKSyZJ%<94AUu_Bh{w^VJ~C3&XajxK8zUH!um8BTh$#^U%=
z!tx2L5euJ=DrlHJ=-G8rMbuA}h=+i}gzZqNi0}FQi1g#Wl*Gm4{&*UJa$|rb$ly{Y
ziTBx>2IH-x{XaX8DSgDa8x=Jw|2XFSb_ngPhV!Nh((V!M3l^fZ1&|NM0sh^_fb?A_
z#g}Pz&M9B39+p-P49l?G)+#i{zWQ5R?ewosRDFNsLbs%lhtgTBv*5zQtkZC74$q;f
zC3B_|e4!)8(zUL~CAa>?LNdWs7}Jm)FE}4=6F^-XE!uOiDwF}@219tE8_Y-3V2m(6
zFy@=7P<$H}`e8q_IXHwcH+FPZyMbv1S_`M;ZFd9gksY8;WFk{dS0SenFO-osfZ=&x
z8*sIxQy8Xvvxg~VW&J-0ISM>=_4P+FsMBrokF8yZ)=StwO`Fm45DJ1`La(MrbL)|@
zvYU+i-^H%&hkTv+Rh!gb7R<iIPL&1rcuaLUWx;r&rcIm>dh`q)AuOfTDE^T1hN{;L
zr>3a=ZNKs4)6C-x?e)iOsYJv6!1>e8_R9WtUo(t!4!cB`Toq-=25^<M@+Tu!eJH2M
zxs~4m_hi;^kXP4@dbM!JmW>5fLQDPX_Ts#BtXiBORhR~&B>6dcP<4<_IFe7S*Qwjk
zx~=N=_u1mw-_vwYF>PM&bL7M?=1(p>J{(;_ani3PZpUs(y&m(QvY5wn#0Z=&VY1XA
zGjD9@h>ia@Q-}Kf7lWz(o;3dd<KM*i{x8$yKX^TufB(hQkN@D``#)Gy8VJY_Q+k(u
zrnAYN6c?%$<(?gSGs)I202gsld@yb8n!H4h*ZCvd3F^0)Tdbiv;DrNuO`|UjTOr1E
zlm5XI>)qxNn$LpMGt)#fE?uIYPtgzKTUdXL*#Z(5);nmu*R6Rl7TWRt!qoR&B8<M!
zQJ)(D&ABh$OI`f*E&I*UBJljdX{-Weh6KMt5hp(Dg$g5$%IzRL)5iYyZZ=kAN5u$f
zXn44|BnoyYruTv#C%o+|n~k;<l_r8DhOFyxwc~fG1{hz7kGK0M`Ipw#OXC@S?Hshj
zlt0>ltcK%{P>FC!T3~PyQoFu2YqT3yd&gjuM@8R*IqbVs+-=JWU#V8!3SS=UVtNs3
z5rg8#-NW$Ohkb_Dx%JpPE5(3X#;ev`?20|B)J=2vPrQjcZ);QbR4I$wFXA-wf!OnQ
zvp-5Q9tbGDVo^JtBuX&#4)`*2eWIGc>XcyKZ0W({xc2KIfOVY@h{YS9LG16>V+fNy
z&>2$#YIC3Hb35Ivm6Ehv5P1FFyWmoDw=(?t&EW^PUVf0f#niMjNUi<#jrjoE^sfG$
zb;w0RfqA@gQY5JRg~lh2pOtF+W>mnbb*eQ&rZox7NE6Jm6eaJI*%`!5H{Pne$MhO2
z9Wi~ukLd<+tKPg3M`QX75PN_WtUPg-!3|!79ig~rONNa(u(rLaeMh9@Jp!@JuetP|
zRdSj{unFjA8ecXo`iY%{M*sn!DlQSn*(O7R8RDDfPmOtR2gnU)Dm>Miirb4fdmlPJ
z!d_txkH9rwZ>$a*!!&1iL#AcBCU?6_DWT6p9cN5FXCfNyT1QRz(}fa^>t$nI9XQv1
zrOk7AJ0?#|v)t$^=0D<i=0ykSF_VtqpIp`wkuala5O=mBSva@JKq4aBM}PR{ru>d6
zgByxMRfK)i7RJC@(k2W@(Ro>MACzxath0++96ryAv<G~?AZ3FZMv;X9C4-yG0znQR
zmOpzTFFKOZVZvR#`fYAL7`|wUl6yw8pQcmLInK-nhOb2OYH};ud>7{RyhihnXX#(*
zpHdd3$&&yi97ow(#!dN<-6_7*i?j<9u3y4r5f}Jv&OV>{;TC^1+%zF4sx^7v&FAd2
zou-J!sv1_C2&VBbD@jniwVbFXG{Z?HWN=l_*|S!oHs!*j@0o-BVE*avx{GB)<aG0+
zrFlOxQHv*2^--4{gh_^dP>Rgtmdp}18p@BfgT}XW292b8tM%<<rHb&2IvFZ{HTB#~
zx@2HtEXdZkQw%kNE>>7uTftwi^|U^_Sjd^;PIb5d$SS_e(U4+!;Hv`(8dS9+HH^LO
z7vrf{1~IK0e>ca^;TLq8FWi*rQTpX@;c&$_16}OuAo(&qw_P!89?tu-n3T|m&x$3+
zc06gQKHt^ZG=%cUt_>+|>-CxBK5%`Mvi8dzWYup-HDG)|0r=tKGgw8~XH?xnvYIKH
zM>m+az4C5Ymsz_LKLkG^6fd6Oaac+-)m|dw<sh35_}hng(snrHs45B#)SKi;A$kd$
zWQKJ8?LvXTU=fZ!`KNvhgSiqa3!=Rtyno3i{40jxf9f^V@4uMI@W0t_U}FF#34{f;
z6dd-!L1DDYn?^@;pH|yaWy;#E63V+}KHiYEYBPSgs;hmqysq-4AyZ}y(3S&s(t#vl
z>e#np*<=6g7_nbywUFrSI}|++-;$H>{Z>t*?WOrp;rpL~N*d_Cydi^QoA`L68^?y!
zCyn)(jBY^r_t<w_vsY8AyzIxqU>CDi$`$F0FM^kGC)+P06)9+90y%=Twh3R9QT~SF
z*D@|OyXzC#<Sy|wRE_C-6SUDgg|MwZ<BhU`@@la@rzrsPptu-@g#BBem<Fo1Hglq|
zT~L>g7fLX-1^v*t=<qU51G-qn=V`&{X;%t4wtsH^yV4`{K7sr#1NHSv<^j~fo`hX8
zjAdLz%8{U5K1YPjG<%wSyhddgVy?x;xJYp8pQ#KFl6eqR1kQ#o=4_N>H#LdFRFw?n
zbypZ%lX`vQ<1md|gWMYmBUO0-YI5j5ve?IWi###zzd<!nm-LUnLH&O4C#d?#Eftaz
zgEO;o9f)IrL0WQ))dqm|A9XN&R6aiy-fK^kG?mRw6Wv+?U4$u%sVlbf+&1)~8Ugp{
zBMkwC0;_GZA3arzvu56&(x_iv9PG=b_`V^4iIA)zlPQ&Ko|)xQYH2fgh%N9m`*Tz$
zpWC?I61Iox0XtxM+?ddF`BX@t03dQuDG}VfDPCba4}9^hcIchiv)@(sk$1HWsWtdb
zzI0%gwox|;JC2&NCOf5*Eh(H86tH!cCR)JLul;3}v50cQz-Pe$1BP^r>Q(j<zK@`}
zcv>Gl9Je9|Z-9^W0?9?H?Fjo%=090qY}hD<eUot+$wjK@3MqrDJB)obHAJ;aY}#av
z>TRY!S0=MquT;<k0?AJriNhU`I4CdUqzTz@i6`5(wX{v^l2?LsvvTz1r9!#e?^X`s
zeB}25etpE#AnV?Rdi5<`5;ueI@<bk_U-4L5gqrjvhuK~kntZZ)<50^z<u23Y{UR{1
z){7aE{1ah{v|E&Bs%sElViLdd^B^?NC)dKRg>`Q6+E4xy5z<l=QzZ%cUncF|^i*x`
zekzplsgRT6Bz;BEpsUwyymowD#Cg~+eZec)9QMo7(DT6LZ`C>|1h?3T*+CQn>K?Tp
z08Wn?K00)bD*1ilEqt?t_!vJFV89UgEGtECP0%R!qqL^O*-)Jb=+{XtAenN?%w{}m
z<f1dt({W!Xv;iTasobS)y1K3oBo;%wMTLqZpv~D`b0Xb=DISkAKV?7HFR{O~>&-j+
zw(LW{!n85lmuIXyi5H$^O2r=pGLE2W!g`DZSS)l0{2O!zF@{@w0jZNs2-6HQy+Uau
z#^VuKNDyDs3?g9W^4t?2)u$I(hFKy%CRy`z=JPal0zI)M3~mU(X$`su;bx3Fa8EV>
ziNE`M#ck)Y>LuqlS;v`zPrS*STg)7uNAncz9A$0*1-tcOrkELA##FuT(oRzLGB}!M
zlqYy?!ohinDQY%KU~%SqZF=kAwaC;CEDkU|S@?Gtnjo%kqVTwX`mKn;?(YGlwD*(T
zuWaH>(wNTJX-G3Cf7xFF3eXi>qCgm^L4S_CScTMTm}r+ko+Q@iEZm4$MtJGi+kVem
z7I-Q1wbN4I^dLt#OFo<5ff?`t?YQIgC~X-%$pC}A^NDHK*|5&+vgv!OOnwYbv}6I~
zZ||_`ikPBV>17(><y>&?!37=ip=Rh}iR^VqEvy%|1>J)>HqQVP;@YR#fp;SH!X#mM
zQ<LgJ876B`3`=h#-G~7%E0MCK#9x7L*{Um^t^~;@2Q<HJ=9VgTK~Cj<boo4<ws`(M
z#|F-Q0bjWc5Ki(!)yWU&S#3fuXDF_SH}@YQ=(<*y=gm(+8&wq*E|5Rx8}DjrCBG2T
z1JyA7!qEgy!_ia2M(Vjc$=Fw+7)DEQ33OjDG%SBI70F7`dro7eTwa1l*|Nlhz<8&t
z*zOfy+iew<&3W;C_2Uo9nrI(l(*1vsL;OeQZ_K~{LTUbgweJNslRIt(x`^IveEB`k
zu=m+BDx$!3!md);!*%kcOLW&IHj6`Vq(hh4z_tfnZGp?pn{*<afF`48idYVtpn`^K
zfPWeCW+3Qlv}x3%Go3cCeux613`O96?9{8_JM;#W_qGTF?;th0T1U)|Hk?+~xPPwm
zKF`Vd`4eAE^_hapW>V_jmp)>{1#jdNvm<G8i%Q_z6d2G6ULIXyBWD_-dD?<QiskFp
zRgiO<@nvJ_{4_+#-`W7UE@fqxBZSd+EAe`vh*RqNS_lR2<XkHj$cIXI24n<z`M$Rn
zw0y=R<(QyEl^oYGR9!gNBtR+uFR)PiF})b&u^kPqXG<uh#4ZmF%@xEn`pQ(dWlaqG
zQNIy(FTB2Pu)@xMl#p*tz(i`9%F>QhFu0hCfPBzo^hXiJl2~~3{t^2Zl6RX^%VURb
z+-AwP*7TD!K6lmcLn{0YAlCp?z|T>>6!k7);@idRNf&Z`&b^5jZ<v+Jku&?&5o3k#
zaWT?3g{5gy;Kc1E+1TtZ3=d7yX{lQJV#>@^*Xzpn9fyA0a+wWha+O|W%4gC6fFgaM
z<%Ly}9^XJ6*i_P%we7#$kaQ&8tC6{JWb1CKcQOaI8L<Z3<LoniZ$0>IO&A~L`t9<2
zKenDvXSq7187n)d1)mDnie*UNr1Alc7Ulruw?hJml2Ts)sw)|c9_Y_o-Y6pb4WK{X
za}Pst0_1rO8Bc#jp@_x+)%K5&a}1Q8f%=9H0{U23vfwn&a7|R?Z_qJqVI-3HWjIlr
zAMu%zU|8Q8!g*gR-u=pK|FMS+A#O%fDQU&;kFM4Dp{;7<J21Qq2=eUcsS$1QAjEY5
zmh;uLH%64(_IhxCT7NNGclA}-hvzTvsYElsu9*nuT5I&-pxlJFB=6dP_G%P=>RV^}
z%)Q*d->hogsK3!)^#;p>Zr(Oi*0&1gMP0ALkAQX954E!c?)f=tAf9G+07O6sL3LQ1
zd_Kts*k*(U@Eir!cW}8J_Z;p;)zAFN>F2GUv2oHTUgyJ%y=9Cc3p1cP&;@`SX6o0!
z0BXIs*u1bxgwR&iP*rTa|7|HpwUGJ`ua7fM;&lAqL5xNK^@yvrk+4RGt5zfhb2Kn$
zxL!+YahsFE+)Zv5Bx)zQEzyMsQK>F>582u2ueKUdPbQwJS#x!IG9X=O=Kt?omHrWy
Y{SQ26`YS=z|IHTi|Iy#h{CoQU0MBWA6aWAK

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/ppc-pvalue-nb-pois-mean.jpg b/docs/2_39/stan-users-guide/img/ppc-pvalue-nb-pois-mean.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e46ba0cc41773ec935ea6f45d50226df9398951f
GIT binary patch
literal 125174
zcmeFYcUV(fvpBpV^eP}AAOtBQU6CR+DAMJCC>@DNjg-(qL4v4sq$wyCP(V<DC{<~p
zqlki3sUe^O(h|V1C53O}d){;2_dfUj?sxBB-|u<84SQ#@*6dkpX3bhNYu04<%kC_&
z&%(sq1YlqQ05j+h*d+jl#!)D50I;$G6aWBV0~iuf022g3D1az*0{~+o<3CGS*$QF*
zgqdFLRqoEQE}9z|A#H4}P0Y_2|0d1)#XTs<pHUJ30s_NAY)wx|IyyN^vW^3cfC#_|
zQRQ^^2n{-EV{>MY!XED4=f7X%(ce7)hLr#F`hUm%58pUFy+S=87TBP2v}dSC00c(?
z0P{nSppY;CVBHUupNI?#+JpD{us;OqAOv6BgMI!1Lv_IZOaFkAe&d|CHG*&+LNJeq
zue&D%mqW19rGF{+`4{*feSk0^I;LKMULhzC$vs)vgrL1q{$9T;{_*&4s{aM^{{#m3
zheJI7zTnW4E7b3tCG>d&;{B$l`{_Mc1%k^x+>OscunYuGM4;^d_!FiFg`Kj6;KR^>
zONIN{orGXWBN)xRLrm--7^1@%;u~)EPdNXo?|BEPTnH*}54`k;^`Gy4@d!17WC=(?
z<#b;!^F8@W002x96=r@Of+0FEgAlas9zN6un760bX=4ahg5VTCwB=qOAeykc(1>$;
z<$L}4a@E&pub+G6eV5%WY#<oYP1u~5zv&)6)E0~y6lU{hADAQp{r}i&7ix>?v{&eF
zejz$c5@EjQ&7g9qT_*pqkn?}O%arMjGBN)XzsfhnY_ELp-H9Oo-#QBQjhQhd+;)$5
zs4Zp{_Yh-K2!?1f+js@q?a6ZwjsWZ#+yO5D4P1hr9zY;K10(@kz!W$IeFgy`P>nZ$
z0{o#8FR0cFD)EB8{S!y|_j{p$A5@M4B7WC}LQk9DnE!b5pSY5M4^$cmEdIvz0E{F`
zfN+SCFR%iY`u;^z5-@_kg+paN{{@|Y^kwN!I%p<grlU;9p<9>f43j?72_|hok_o|N
z$Yj8D8Y<Oh(*29J4fM`GIoj*b64c5+=mbM8di}<;hj?#;cnkyFp(g>tKm(zFQTR8h
zWQv9#`MV}Q1c~zK&mjJbtWeOKet%&o|C>au|2vs8e`Q`|9)X_Y%u|2SkYt`@o`U|y
zA=&)R3;OT+Kc&1kQYdId(f@|`AIIXKSSKM){eQn107(G_@$B_4JS+bzjh>jM1An(R
z!t};<h^r^YJ1|r9@6tUfcqRTp{sV|t`n&Y#*wNLaBS)VcEk3#o%YoIxnqhBY4X{2y
z64nZ9gW+J^Ff6R`FZ}<L4*fOuz+TVy^lGo)d)oC+4*ylwkjC<y;t}JK<~hv+=aJ<(
z^B0yRk1)?M9y1;}s7CCsar;}pCI6LkNG?uLYyXtbf8N&KTtWRs2^s?iZb$=Af4BH=
zI$#I&7zIQ^G(#YH@9A40a0SBs*O~VxM@$k-$C%9jgLa?T)A@hWW-(?t&2kcuWKm$z
zW;w=Uu?PQiF0;r%b=uI=_+R@5>BnCf|Hi3Tm{-)EJpqi+L03XhKE7d+YDbT1NuGpE
zpO>V0pofZ*q`SYr<nO9b$q=tluaF2YPZeO#ZtlGY0GqbI?ID9u|KDX{Rsf*C2CaPq
zf0sFj13-Bw00^u9U3LUoD+LPx;Ekq7cu2%Q=s|}1BLIYYA6NVRoq_8Vw0_O(?ruWs
z7;82F(6V-SDNlEIX+;qIDFA5o-)#l>*qC0ivcVXn07gCr7$3uKCjglm24;re*Pny|
z21XbYGYcylI|nC(P`eLcWPrgKnPAM!OiU2=HK-n7;$!ANbnGOHz*%=zso?!;*R!6m
zNuPS%CTRPaD0BQ$NIW}-knjN!QQ5<CN8}aMH8izOXzLiAHa0OeGq*T*-p<|uV$Z|V
z%iG5nbvZOFJR&mc%GDb;ZzUuqC8uQP+_{^Zckh1w)1u;%XQgG&UsP4s)YjF%dE3z5
z(b?7A^X`4`;Lz~M=-8L>3Bug`*Kgk!7MFgIHa53@lDENMJ9~UF0I+|s1wH@4*?-`R
z58{iFi3!HUy2lp-W8@y=d`!%Tj<N8cJj?1HEFh(Joo)ZAtS7JA*rku#5(O`XeC806
zQ76cf_E`JP+5bDn;{R8i{l(bdd`&{efbsVQV`PLe!(cFG7G~&TVPo05*f`jJUmSm4
zT)!`#y^Hs+YZoHI0MTG#Vq%5<_i?gw?)x8KyVH;x8N53IaKRWLWrFbm2!KX?nsNmA
z|L^~+WdPe9_)o_XmW76cd<Tla94q^C@!Ve};}Nsx3y7~`n0EmqmHWHE^~_y>6iy9`
z?T1%i+XYt7QBGaPuy3tKAb36=8Acn7C~*fBy0loVI2e6edpn0S#}ImARd#`=wH~`b
z{C&(XsW?O-P1k8qBiv^YUD#~*<KkUU*m6z&AolyYIV~!`961d%dg4ko1kcSYq6IGN
zsA=?!Z;^sae;DuM$9uP3Tm#-&A9HF@aRF!!y8u%at$`X$@^ERO6&R>rMpyV9b#9(X
zec^g=+SQbGhvJf`*ZZTi&s6fYUC>HU4Fl!6@qQZy=w=b$PNr%+bE|yt{PJjS9=hKD
zxb?vh@gB}w-b`P&=JThpJ-m~=WOi_zv_P&Vu6M69yZH`)7>iUhbK8v{zLn;$-=MrE
zl3~w#3-T63^n;PFw}7+8`~NpLLjqlnIDEB$ER$dR^oMNxwR`;s9)xmT|F#ferVSH(
zi|iZA*af~FC*#QKi};<BXylMGu>Y(vAei^UFaL0&s8z)ep`EK15k;;RSuD~I8NYgt
zmc$;~1=`SvuXm^#yTGfFF?x;->uO2KbOo&@GWWP_S5vQuO!|!?`9-^glM-{~4!xLH
z{nUKcb_7MZ9Jj@>Qx(T?<mrEZD-wIitF($(4KS44PDHFM6O?@@tnJnq!mo9MPA-Q?
zoL_?#$;!Zz?Um+=)syRowiyB1r$R1WgaGJwwk1H-dZa=HEfBFN^;qkUKcOn6D?yjk
z73eUZxbl+1MD+)iDV|X`<`+`2>Zlm47nT)G=BRM>Jj2Usd8uImj7M4#TAzX^H7IaH
z=KW{??G{{HTUAmnZfd!&TtN_fY<@sW;r2=oz*}){@aIQi*Yxk7H8`)GI`L`D?~1KW
zX(juqAiwYM$$I*k?YS{poC+=z+usjLsZz6flkjBm`dwhqu8?=q84`-xPb;+Xz?}Kd
zmy~M;+G3Y$JD>Acdz@f?$24E-Qr%QDtX$Om>XAyocd=}<dpbvd+PlUZ-2Jt%3owh*
z?i$F_N8vnriUdT0gJ^j3TIx%Ul?B_!X41zFNGQqQmNyURJ0Tf7(~?iiixmMAXYnhP
zS@bQsQufRRWJi98^HlB2noJSFdymq3Z{EIo7j6M;H~d!v^<hM#S=cC9ZWOz^1%3^9
zF~(nP@iA>ySW@8j$aCZ)B=IbM{M;OJ)txxs-XZ{c1(&+aEH+fTeF~TV8h@Pc`ds_H
zxjVU29r3K<i?A)VPa8uvb+^gwVQ=g21)57;T<jQe3Srx^%O_foM-mE>V!7%4Ei52%
z?!EH25vq2nY{duXMstQNr{phvWo8S{&u(F(3mH?66mpD3=H_3k$*bQiscn)DeXq!5
zEK&Y3j&<|}>gr5t<>}4Edh*WA;JQ(0Xzc8}gDm(NdM^KRA^+M1a<steUErb<W=kD=
z6Em9(-#AT|+y%aL!0FkhhVa@jXj<$|4e!hOqZv=y^K?e2qA`C=3a-hT-Kjes=Nk71
zOR#<Uy8llx$MvvZ^9}WPykwJa6!F~Wddo`0_P{~qJnDn~c+3h4nsR1=FINgcxjgFq
z$aXA+8ArckX$bfEYnCGu5LNJ1KN6OF16thTqHc7=YLCWXxqXF1r-XF$?;RaCCHvib
zYJNfOOw}suLgs6cLz>Sgk#C3pH;csoSKndS{rsPsl3zZMQ0y`9x7{#b5H7pGz$or{
z3SFUvIcj~rrvsaDwICVcBd)OSNJ@EhN~XSg=)w6rlX?Zs=<26tW(Pz9nj98YsTw_0
z!DV{#E|7#FCKc=gS!J*t)261Hp;b8?4w~2jStYqn2c%3zx%>KhplCq0HQ=yOO5s$-
zrtbo?&+1}jiDvs2agzqJ!-OLS=U?S=zCHlEeZT^u$XMCYmAeZN9pDo=))-h6){G?E
zl6p15IzS^Gm#8eEMaYemzwDK9@J>W9{rw5-s46JevvLT&l2dp&R*qCLQddEObp{t}
z@lP@iI8Mk<lN~*S%D+vq-X)jG+8MS7UgX8L=u;y>4CMq>eYoWa)wID^ee$5`{$s$H
zaHJaS+hps_xIqJ6TfnBTQz09{TY*9If=7H-edaFM^Nwq^Ej?X~IP-cferEJr;q-Rw
z=WAux1HK<!xcp)^jFtH|k}5PdC-FR*gdpq!?Y<?!rKo(PWmVCFT+7tuWaZ(i+vUmr
z5etQ(x1Zs{5k9z^5^#{0=+}<p)YA?;bi=qRUCtlaoPL83K0g2Qy}Y-<{nl9qL-?(R
z{?!s_sX9u_3V&p<OwK3%@SEWNI%7i#uXq@2S>su(wPV9qUL2y=e{^2X9hfIm$U(HH
zEfVx0Jl6_cma2dMrrNRijZglZ55#JKCIMy{Peu{PgKC)_m8^awGy!v~<v3M~#M{0L
zh_uQL#u%Yw%M*d7GacKZrjN5vgJJsjjWK$7qA5aTxrKvP65~OP-7yhz9?c6HGAg)(
zTsl`~+P0Pc{oL)hB1?vFXZi%ZUbqJv53k)$pmRs@Y_!9<F9-Xst66*JSXixgn5xLg
zNh_S#!jALufr-Y%yu_af!KmA-O6re9C;UD%CLga7iw;vWm-dk3cwyd!(S#Fi@xRPU
z<9C7O#6-l_>*eh+ay}ScL3Mz9j6*Tn#7X9nTAI+t#DEY7Ht2KCLI0k(%2s)~LhVA{
z?tpW24rWRqeFVX$QNT&>oDp!*XSKK_{npO~RT=Y4##DK_i*oHYOl8nkk5OB!fN7L9
ztpx#@>I*n8ntn_)DaSg)iP*^#ANjUsa9vq?rj=L6N~U-qzVnDrxN9bZOv<lyGC%Fj
z7+DfWLP9c3nc-=EGx(E}r2M7cF7@22rt2~H`JO!NmT^7u%W2>VQ`(voE>Q0nC79lY
zsHwm86AoO6n)vidmEYN_bb6`bOMsd~$`Tj)K=!p))oQO{AF^&=4K)<sUh$(iM|qH&
zKv%l@;L7dvilyVJ`Yu*#R|<R{mW4;*ff_aJkxp1<Z^-$aje^xlNMrVm#WI7VA7<rR
ztt3?WW6YX2?7x2G%47H5cDHKcJ1JtH|J^S9;@gSBN~7F%S_MTv_5+*|=lxROs)b{5
z5LMJ^8Hhd{z5D=ItK725c?soqto~!ne!EX=p<S^|^ojQ*VX`^2q$Ffa$kznFKsDUj
z68)4X+lugRJQ?`Pby5U4B$Kv&b|iCu%*7O7hsDZ;&UqsowF|KOM|@;%X**Ko7^5k(
zL288i1bI(SqG$^jkGu6ygfMMq^iU+>xf*0OZdJ9g;|W;z9pzu4`t;Vy`<6uQnx?wU
z0u_4&JEw?v@KoHxu?ngJ)Qh+DUI_z7<(LcMWgw?bCh+0;r21%Mzvr9UO%pu1*o@YS
zMGfr&-ncLLd3={6=JEvwf9i!fvo-(9x#5~-H~aeAYps&?nI*-&k1t*E%bZ<6_vs9<
zjyJSaJo6Q`TWhSCHuWe)4ZIHx4L;74a=_ho!)ztW&W<TN3UHXM??QkI=<fA2B;Odp
zVa15R(&QnNEVxC+`8PK^g2zn*zog7CY}Z@8nP+Zlv2SUHdi;iZdYH;iIpl_NKL3&F
zvD(N=jnf65QHrioeMs8vh=xgj_i4u)ZIwtm8;yuDBu1|E;RHSExhtdS<fZR#E^1Xg
zE6~e*)m~a=NP4$0T(8Fv$D2;IBR!yV9wHt;lQ(DDc}@Q(FRwFpLZv9X!?dYB#o)Ww
z5b!2>|F!wIh6i(o$CgxAv?<r35ah-ol_tb<C%J^CYsIGyHx8YCKzMuOkX2WaR-er=
zhR-d&A9x(aM|L<Cv!zX?66~4wnVe)T)1bbJIrTXJ=Ya<m>>uheigp<70yl4*d3$?*
z`-`QWxo%bE;joQiD<xlDX195Dh8@!HWvsMCf&Kx<O_iQQ-x3z88=L4ZCr7fZO$VOu
ze5Du1eCcQ;!^aq(mzXv)%w_m(QA|RMXzV*TzJSQCpv9|Ubg9wNcV}Mzu$?p+vt+WJ
zn_(+@*nlnwY2=ccWx<$2a-G7w7Y3tu0S_kJj=AhL$kH)&f#FX;PO>)9YVCo479K2-
z8^M0(fb}bF>U2B))IH=lfN6aEJq{sH^`kgq+Uzl2jgxSJnDAP^bfGnyCmQ0qRu8PR
z`VG#$9US2bY&|HKUmAbvp(w4BmV-g5^2UxJS;Fus`La)jRGgiZev0wWokLw?;mS$`
z0`2Xc_GPl0xW<JjWUST_J|6EvR3W-P>e^rTNx*kF-`!vZZsoVLG~DWrF3*x@9d`I`
zP2IOfp0j49It|@_AJ0iUZ0AapaI6$ek|;KLyTbv{#-YtVjD#WMXwCE^^iTLYp9BMu
z$|4^x;&h%#!zkxC`<l?;l`n2;Cw^^ylb-<O^tG`PBN)GRP#(VO5PEISdR)3qAj+>b
z(^4j)9_jP?O66I0rsmOSF?l(X-;~F`BILcLgMT&BIXa5W=Ma+_R>UpqZ!OPDd`Dk=
z`81?gC3#{!ZETA`fO9MwFca%?X8SpvWfQWVOmz657`c>!TCekA4!vKyqRv3w1*kmG
znB>O}!r2Vejfkn_;Tmoa*}jPSF$epzj>si%SEe^o<I)K(K|J53)SP&>Lenc2TeN9!
zc7XuIL}Zt9O9q7n)bx#kKm8PdtM{y}zY;j*^L<l;%elS$=m)-YM_iBdHTsfxLFKx{
z(g~_37(I7D>U`sHn0Dd)R-j?h^>s$@dWJ^gATAu^4Nq)PZO4zV6Sm{k70=hicNkZr
zPGLS+KA&@{^pa~dKi-)0<Ju7z5!;R++FDO)kpsZzUb6)g>*m+aS!0e&PtG{Dd0+&~
zGwN40t0tt3kGg8Ai6c=LdSl&b;waI&2iPS|=e6{pysNQAOp%(KCL!4!YKiH?+g!zb
ztH?w+(N$c`8p9f+L}I@cJIzfmxh&iB=~C{IhtKs}J(+l>=AUC*TDYi2EktXoKu-rk
zz_WAA?En}lQFx!OIaD)9y79R9*Tv|3;1`+<t!D90onu5u0o!fJxSi8(@}#cv_|Y|T
zz~H#kg<}%|^9gDaa|COy!WTt^Kge7+zQ5P=gJxjuaLrHB$GW^wlzZG<mw%XSp1as<
zvbkMl^D7h5KxqCZ9+b{&r|Hf)^dQ7yP^7qkbGF3rCe;9$T>wEnqmlR^{DaVYHKr6p
z-j}&Q7_dujaR?!t2L@DddqQIE$;C4vP>-bGT;Piz^DFlGcg{AKtGTfpQ4U=FCewCY
ztGAqIZ~=815~?#j6!Hmc@N6~KTu}?H7_laO_UhrLGkKb#Ib9p91=f7}HRA_=mB@I?
zQC^pp#NLXrp%pe6LN4MNG3;%(XqD#jD4kxBKv+O{%WzYLfm7z~G}*h4uJIg!eYKu2
zqI0B@j5>FL{rFkriqq3)*ccV(_u=U3YYlT7-!%`>I~e-hEN1Cs<BPqVzzdj!CMKhR
z=ztiX88<i*<3}7PY265@>o0Tp7*L|q#o*ply<}(=Xt=n>aX6$m$T$*CJdL|`RW!<|
z2Pe9w+}Y5yy<F&G(Hi@GhV4?ry+g~C?tFdxnu_kiwwvSA$8zyu{HBr$y&Vd@QIa>>
zm~2=#GG_6qGZNrWXh=+n8C`fzmwwbUFsD;T?{h~8kbn60rvt?%6K1-eCL*j%T(@g!
z7T{2~n+SM?pzqQnp@z$_{H$(yMWu4h>xAQ6n&(3vbA3TW*<|SKQIOiaxC?NEVd61F
zdn-5rm*{a6r4yODvQV9RfX@YgIxp)qZl$}L$qj_*jV(q}g^K1(N(uOQ10FD-v~Qqm
zUfkJfjr>@)e(<yFY5dBsq0_s-`TY#Du}2`C%ZeS_1#0k}iv5iAE{T2D4BE#+O@qm1
zXw`{*L5JUFFNZ&>YB^!g$~52Sk@A6rSQUdf^rPj@X7s?tYHp3`sd9F!aF0B@AL<)#
zG-$q+UUka+myzx*?l^{btfSCg5;?I>7NO<u0zUe@?QW8!nybh(4NYik;5J$OqgwaE
zyA?iB#|rTSMr@}U4pg3$&Xu7hzkocpGq{W5p=TODhZiIT&wN345?xPfCcbP6=~B2G
z7Ov^2&}%07b3el}zDjO2cvf{{nJ^Dlx05vUDp3BJmv_!oO1Nl0^s;yb>q|HTSFtLJ
zV3SnKMCd^(T3@hX5i1I9YA-ME0_HF0;Naoy+gpf4NE2;zGuUFZh96T}=_(JL2P<Ty
zP74d|OAcJ|^8Id?X&`xWvn=@S^=>n4J?=k{DT8epkwu^lv>^yttxG@i=+}|<|3%bC
zdtK;<mJZ^UMupqE>%ksgp?Y2khEYY{uD9o<aE~)?Kzh5LGWwXEbO*<;q}+62-X*za
z63L0Gi%6Kz%nlAEKM&+`@;R*t@KuVB$5%3VBNyf7gsDjevQQ|>Yk{78mF|a|P1^-{
zY7m0-&yeWTe!lc<$Z6YAwDJ@;y5bNSDkJ4yFz47b0$=wjE97+^i#>vAyR4zFgHDQ=
z5GQ99zxXxRJ!Ua@ChLjFL|tfc{l^~SD{r{-jt`xE6B>UhNDpB15uiA!-nzDi<k6<y
z5b?w}VOmx6i*P=B;iXe&%0$}E$lG1CGZVvk`2D&8-YA}k5~FiU`*;z+u7~Hqp3{$V
zkJSi|1gt)Lqvx?+v*t2(I1W*Q6pi}Of#B7Xdl@y~?Lb<(dN>&^J2Kr@%6af?TF&+N
za!<>cUP}Vzj5t`#*}aa?u)6>Xm*nu6nU<m_xpVDS<@ek3E15IM8p&C!P}$LCU9Xs+
zavt`cdp&~d1w`kJWYL1eST<0XP*-r=NH{GV<<mG3m5aAD4i>;vFIWlCY#anBjPx-m
zUbiwv(O!(@C$Xj%4N~2o*VONjr{(PKPpp|<Od9O+U2H#aN|LQBg#F-a8Q6P0Yl;b^
zwa&YM*Bfb4PxtDf@P5vAQYz#lKXaXs%-qVpCheb@9c%N6U~~}(VW3^3Wg+}B7-Gke
zyzs8a!pSwwewLm$90t9DHpctr73|_9Jk>7#3JDN<b&}!E^{+Zu>i!R8hB5UI&B9J#
zqWJ19sx@<`fbZ4D9wpo74yNb4;>(NUdCwwmV~AGk;)@twNYx7?CH8^n=la{%sTVYB
zOPnoEC#fhM;-CJ~`?>F>WQ^gB1)KR$SX!1L{4;u$leF1dP>pO~wVo^O6hW?N8oS5O
z928ksDe1q9W|LuDjW?I$cyf>U49B&lI(eMVfudYkc23`3`CMrF#8)VUEvc%u%J-1m
zcXnm^3l6=;AV<^cH1QkKa#W$7Ii8o#spr9HV#4^naa6wLkMHVo{H{NQ`CXoUTYHy_
zS1rBwWf|E7Id^nHFnTtxy?`r5nb^e03Fmb(MrX1td3P^83ptjTd+Jlq@Z(_y@NO}t
z?bI&dpD~{2lGlo?E4cZk?xjyi*s7zmOP9KXLYj1@7-CDe%--s$aT=>lq}Ln!9UPtg
z7R80?L3<GpD{Vyz{PF2qliH5OI4yf5pf0s7%xk*l(fGzWiHAF5gSdKZyFTl@@XB)%
z%xq-!=2*Va{qTyVDnRcYHtdUA6j@-XkK?n}i$>n7)B$K26~*)<cB6Nep<n@xg&t>s
zr!`{G{k*a7>~$~+VajR<{-|x2n^P|19}ZdiS^D1*7M3*#5gC4A@<z2b*$eQ2W|bv)
z3)1Kku&2a$9?gnMH2jPmunypuSj>;|bbo7M`aCDR_Ihvb3F(Jpi}-Hz7tUGLF6;DH
zv4a@C@N_#3aTPzYx2s+abr%jj89Eh^;`lADs=ZzBeX^nUrK@94!D{jfErYt>^8$LV
zJ5ky=^MiX(_+h4SG5=hTg3AE)auviydh9f&YP*xunu9(<KPdXK{HFSVo6|}6aM>0^
zlu~8+UFDL=Aze?~Slg3dV>iJE-4kP}28=)O`{*opMt-KdbmWQVC2Z`g=o%hPSI$)U
zs`~Meq0v5Fer@c+GMyuu;tlz?HDfoZ$^_)i!I>WuxWrm*l|PC!vg+z6G6PkXq9Q5%
zZ@=<sa-B4D&B;_Go6^~b^zz`Z9opZ@%HBf_$r&U^s$B<xBd}kuXixQwDCaNZYBdup
zVlIDvIz(6OQOWkN;8PwE8MN`g?pW;G;?t^)u2kGnpryn(Q39iKDP1oqZc(w~JTo)8
z(na@*YBTyGj$bQ;SARGh#Ds|!k6K4&<A^r!U*=<b@ri|Y+>UK6P0faNu72e%B{=3P
z*haYwrp}_%Rh49kbq`b;GJCl5jgJZm>`zp?krSu=vkVsqpY$MS?QJDK@*1c3DfTB^
z?TgS!3#!f5=x{EMSEK|RoZM7P>kZyW7O03xg;XAjyWYtMd+3c|=`LU)iaU*)t;8{5
z2wwD$mL%RY^_$@tiNdEqL6OkX>41b))2admd;MH#L-=zY?-KX&rMmo<Je{d=r13$b
zMWUjNZJB4dL0T~%Qa*MHQ=>n&@}8mtSvtNg)1;^bC&{6$ut$+aZd3J4&%4)EWV2l(
z$5WuK94L$J&H!C7RS~yJko^mO60oc-u}tee_j-nvN%08Uqrs!m_a=!{NKxY(V@{67
zD#cs@uTYetBp)=s6ky73O-$Cf;5Dk~I~Cd&w0iN{VQ7blgqT|caipR7g*MKNRD@Vy
zg_lmt6Lp<a&Jz_?wA*9bwiXezxy@wZ$q(CDX4)If?6V*$E9pFrna;Xo6IImn`A*8l
zm5vW6cl9^pj9q-Z`fW^JM$D%SbsvPN;`1ZYpV0eo)mH|$PITr+cA&W}mb&q}6wQ(1
zNO#(j6Q@?ADBT+|TYwNYfPsqS^ijG_?C^}V(AdDq;K)*I{YTTwRxj~Z_X8~^j5Njf
zy^o*e{v;CX#RObSWlh=h6m%*dw&>JWHN<xt%32<-8j!kUg4e`XSrniSEG0->-1**e
zg9h2z+0v`nPE681Dl_ewx^I)}ywifFectPC@~e+g#n^F`if^)okM0H7@{D0}E^7|p
zEp~y9&p3C)j(ou{lTpMk4x9jD_8AUxAAESz@B&S6k{>ygBsu&OUv0@!t)%Ht$CK@>
z7vQC>b=0Wkl*m{}cB@I;EOR06kKUAxjO!BGZ^1`I>sBP6DrSDUGhgUYnROf>;5u2H
zoyKz_#8mLd7Q+U5FNkzFn&d}s6-K^u*lebV(#K~sX^-ea%Xp3!?kJ}wD-yg%XUgg#
zR-m8jSd!3D`vdQ;JoA2co?yaxO!^z*Gw;8X_=ctOX%TpCgrethx1DdjDX^<~tk2PI
z0wnPjxuib5PuJq+{~EHN?{hj3BB8K%uE2aBv@WtFo$Dssek3fP*V3{P;}-gH!u?#%
zf|2L5@9JkiXr90R;b@!2jy8pbK9euEdadY(1bnRK3l%Xx&A9*Z#-NaLK)8SBp|^E2
zmM;<yxoIldR8c=RW(>^17|YdonMCQXwU#Gvo@h{!8`{ZQjf#rnleWN^o9*4|&=lxl
zf4r}mD_AU-jn`V)+J+m5t--WeLLLM@1y{A*Va*N}5sbSkYRRu();?-$sCT)0G&M34
zC^zBxc5rMMQ&rH37Km}BJ&Iu_DR$c$tJKbkYp7IC9OyiIU*w*6*@?3~7Su1$1c5~T
zo*%?OQ47vpU`eKS8@&Pnhtb3Wdh2V<BH9rC2+jxTO-aE%I#=vuivk!-Vp$t+5dytk
zMxvvmWwi$H42k+V+WIeq<(a6zZ;y+;Bz*V{I|hXY6}DFu#lVGGxf`Cuf*X~b&Dm{!
z7xfkyemT1Sz^gAD6ux!#zygPkL*FiNAO==Uh1{h0DmRc5ndFMPTA#YL)pRSt-o-OT
z<Zv2ANB-PT?2r5d?sW;e(NUvEp4ucOF}`?_1^}~w4hKMwie-a*8tYxacm3urAY2j4
zA3N@BlRIHqR)G$zRLZd$u30}3ms15LttbIUy*n@7`IXggLQlHR6v!&MnNN>|czhFk
za7bCth4`Q=v-M@!<p}5Ihl6oidiJ(1=KMKR@^5qC=a5g3;&gex-!9AjZ0gCOH3@y^
zg*Ip9sy`}{?Jqbh`W;uOw^?4uV+vvW2*f<1SVB=9Jxi?&{493m3So<6P)o+W=DtAi
z8)<mM$}LggCqi3%C;c(R?v=$8SM7~vj~>XmcgC}!)=8XBLxFOGPwn$vPOPZwK7DPX
zwQqE^cg)a3-vASPcM8)cK<8a#?J&>+*P{aukBU<0d=%}2u&+k@kGW1ApnX&8KCp9`
z)xLjS7KlAZZ!aL)Hp9ueqXGnnE^!@v2Vx(W|7b;EpHetNKGN*9(yhnxcNk~RdUF-t
z4b;qljyW&GlO&|Ua}@okU0@Dz)9u910$!AsaBjL9y9a-_T1h6cV*>XW2g)zR&(j{_
z{ArwXV;x7s8Y?2itogCIn#ETuYCja0$3<jzW}jI5M7yoHElutnnI*BV7?Tu{wao6r
zP0AX*u=kwm7PGM?afl;yacB&k47C059PY%&8H6uQp`_|goso|%1!zAH$-26sY|e-j
zBQyiI9`;bd<(I${OYI59dPR}|IS$gDR;s|@E<o^&ITIhP%H6#<q`1_e|FPKY<1MGI
zr|JXG5h@Fw+LD3q1sCze)7FVsl{@kLbOkq;;)aqZuLu3Vd3`Y2|J1zlRKbp~z0!93
zjsUW57Z6j|M|9#@sNsZFd)fS<vjabm$eJ=-cTz59y|FYLJ;3_)*#l_3xG`939d|Y2
zWeI3=86z4l);%~!q*Z;CefdeyVy1F>@ZP~U=i`puEjS_u>%$i4MUy7!oMN-hQVVna
z$y<#T!X5UD+X5$ra!RbY%Q>`;<?hR#?oQHHm463(YZ&8(tdlz!{S<sazDbfx|LJ%B
zZ9%(^yq^^HmzG0SQp^Z^W?%9-{Tq)#>{5Hwp@sxYzk8#OUbeSy!xR;{_MM!KKOs55
zy1+{n=qE4bk>3!%V+reGPA;fe9b$vAZ0|(wKBTkxM(VxK$;a+ua!*gS3fZ!8*E&Ib
zsnSYgp|B~QTTgB-<Ik{Z<!B_!4=8Nft~_JN`1*kEols<60ny?BBzrMhgI)k(`uKK)
zWzn+z1Kle@$jLjfrB|^cSZMn*3(0fibPF8X(Lx6nJLbwjY!1B_!w0SBq2zF)a3@y8
ze1F4OQrK`qqpt^_iKq9)h(dy?xZ);;Tmud8qbOVlT(Fe(80UrL?wKnPa&d(W(G=2r
ze_k9OmRMI^d-H~Lx5-$ftPzVCn^?va?$y!$l>p*HSO%Eg)lcPK?llmmcR`Nh3YrAB
zp*=xEVS7|)ICEG1M9kT;@rrW(={HxrZri6Y%7)v!R==L;(;Z-K6VAgEU7eH>1R1^$
zT#hQrPg6S%8=pF!A(@c56CjDJ4M$dKQTMOU{oL+O*CJzRFSdLGRDW7_4GGw2mRdE}
z49onwl(O-Ka7>!Oa6aw%MKi^rE$OZT;@N^P^=)lqHei8!w3H9~vyRZRK!fWQm$DH2
zt<htfJrvIv4CqBU6+3oSs|$Tob31pG33b$3`71kGX*;(%YeAvt?c}#B7Mz3@*Nc#+
zxuZ>M!F>HlSAyg#grspR65ybeDK!g`gdiV<)|KL{v1xpLKlMJYtAN7zVHbE(5+e<%
zmf0?lEk|OdcS9q%m8}ENUhV~YXRRj-$jii@6%1igT&KHB@8o=QtY)6|(4&;L{T!24
zABR-?q^Iv1yQ-(LF`ugD?Zb0Wy`VvQNA>wU4;B&0?UqQMY5BLcDZ1REqasf{da@)6
zx%<jYrvpA_y}M*4_fGGEJDOs@w-pW46Q5`HYZjX9tVU({|M6ARO)SAk=7c`~xy#>L
z=@h>AA2Daml0}t*%j}?J@{54Zo?mT7tHSw16O<YS%9T(Bz<B~H<`~Je%}v^T{jR&O
z<YVUAH>ttN?|n<SrM#|~GU@E&`AmUANK6}0@IwW(gCaiADmR$yw^>i;^daV)G7uuF
z_9&V$=`=<5bZkR$b9|e++Vsh$ozBk|!sDN=a?l`47=I;=-i70cJFyESyFe}`nty<`
z3sYN4-M`#LKLA!BR(?_xz(B-)uxes~VlcO3R@}q_9@MXWR4yYTp{CjRR^h<WFNRYw
z&KnduXR2%|RVH5BJ(anu=4_PbyE2pDulok!R>i_(Y%~3Ts;gs>bN%3v{uP(m+2yhr
z>sQTL$8Qd~za7?s8;-s<D@eYmzX;pJlcC*z{3EJc3;6jNy?+;|bxY?ip+ApmxOACi
zc&WZF|7L;GSvQx1hh<_U6|(Lsu`^$d&5a#5kffFAX%qV`n`t>UNwrv^_b)rdr__r1
z6_&bZLZ#>C%*LeL_r23zcy>JF`z}C!2SS$Y1NkPcatob;XA4+YE@vCO_}Jwg|BTbR
zvXAs9JJZnm-i<rL9WifO0f(QfhVX9uO7k3iqR50+96s$b%abltLQ-?AJ#@`a%B4X{
z??>P}r(!d@(@_?&r7YwyqE+=E)o}>n<ER%+aW-a6XtFNpD!Z_(XME6}TsPSJ{?NYT
zg>WA`k9h0(uGoW^w&PIll$4v+E`UWt1G(fMPUE0@gVqH6Z6sGM{+5AKLsRu<su90Y
zsO-&`-2J)PN{wfyWq@ObVhSO}x3?I;<T+JPUT!s%awf(QM-*PMoGqySGO;EfE?eUe
z*)2Y4AD#FX;ZbY9QYpqXA1$E>>zdJpWIaqhJh(urPxW})JXk*V>8+#E7PqmTa4udj
zMX&dTwznZ5BKtZAS&C$7IY{qC60LdJe>~=U(35-h7>OBAzL^u1ake_u=8+v!aQ72?
zMp930%B_s}Ptl{M>4@2~g4f6!ga;o;9bo+TqnOWIjhxTNX}J!!Paw3Jfgp$)0z4v!
zU%fO@A7^lAsjk5~eM{G?$)Q@QZsJy<`@w9PKl(nu7gIxD=;I+e0Q(R&%fBu_oeuPy
zkf?<WfRN7Ejn%KO2Er3b2S@EU6uL=H9}W(-i9HoER%krQVa60>_8s>^0XXtB<^TG7
zckq9b!4im@g#wl^1I;J_@<1cWI-L%WO6-_WbX;artrb7KHaJ{U-U#IT-c`f5AMd|R
z&+^^NmWBO#x~CGL-?>4AuQe=cP2<?a<;UWdj>SDLj|?`h$h0lBPZ@V%jFCJ|0Yx$9
zMB&6H3}N|Z!%&lvJNg<^*2Cvc^Sl{0=O6C7!nE%SOUo5nHO`Bdmp%q(mmwN|>s`7^
zzc1hPxmL*b13F9?We$ApP@q<_N`DJtH{_j5?BGp-;s#hy?x7Z(D||h%-aw8R#$WAX
zs$bEydQ<mHV{Q7+Clm3fENjpY`K@!yT_oXy=;RpFry!!Ep)^&`JR(=;k90qcZ_}NL
z?E8flJ+u0&$`0f5GeDWZN6*1cau3O&^UB4;GyP`6Sh+h}=07>_jqEE}KZ;@_pL0P9
z(4qO)W`saD>&FU$Mug_ETQRaRKLWdIYr7sbcSc`7d!@`x&`5MXGEZWfHbMkw@w7yY
zM>8+^B<N*ubNlQMx(bOYYs$|#&B9(=;^=9cQ)1sY+_@$#0jK*9Z%qjrBE&GWCAf_&
zYvx^GjF$(q<T1A8PJ+)_#~bj0S~lPflJI#e;f@JpZeZY7a~ZRyaUq%B&Px38s<KwS
zEF=`uw#kSu5Ya<XqCnGb2_B*Z4(qEqu*i7fREK3lLrq#o*s3IIF1_eNO2~L1>tS33
zU4g0vC5G}QLhiErT$e{NnA__o-xtMiBI~M@mCeR0=01>Vcu|cTh^d02zsZ={!WN+h
z60PNy`@D;b-*fe`k?fgyE1Xl|>3{@(m#MpUSsaPIvs+Rv;vi(Una+clGDo+|NLY*)
zw9B+eBYyE?P+64;GH~o9mD_ffQu?a07X7~A*g#b>wCKj8qBfmxw6^Ro=rQ0J`7Zmh
zws<1pQB;<vsjlqv<4*mrX4e!SC-&K$b9*v}29MxZ$4IHBZ;9}B>l+R_JlL`#m*&++
z>9lXdV%Hs_QlIv4auL_WeB5~0fC{KH;ov-pzh^El3Fp1eKhqygSln8EK(V~?uveYZ
zs9kjX{LP0ew~aG%Z6elf44Lm0j9NL6Z_zSiPf(5C&~n39E&m{4lD<7&o4LlnZS~mK
ze%{&vHQYpGV|vXLqS->Yk@5>eXD_0#gL!1%ni0A@)tRqn9?S2d{4nC|yK0P&W1f?z
zxvZ_-vs~j-6mvT^;F8m8$PBL%C|uOQ&n?=jt7wY6aqyLwO{5=MqxNUw$)ogXyV%bG
zd_9kIuHTnpxc3@;_wzd&-r~JOrM5!|FT}z7mDGdzrE%+?A5_WLKEJ7*iLMd8WZy%P
zmtvTNrBFS<3gYHGoGJh#XAi73$<IA`ekIemY|?cyu|D<eYqgq?i5ZvNAciAtd7;<u
zhF@$AprmxrIqp(?qN1%64M~ZuIl*(xhL8QeT(<W%apTl@ZLh80UX`{51iPX;3dW7d
z<+LWcfn!WCtz131Ns%CS9EkQSlX(BFs@_A&(aGuEnfLqCdSa>~A})TKvSA&^5FJ)|
zmb(jH&!o@M)Py#|0-~*kOW$eVp7=2n?fxc$_N~(+f6E$_>CaHd3dMHQC5WpPgH{*<
z=d`JHV-o!FapmE>;6RB4?J?FTZ6fb9wtG4ei7kR)$*ew_P@fjSJacP~(_E?Y5-V!s
z>w#W4A?$>w`?e#aj>G1iSZ4YZo@iB&yi<_U!oE~5ae{c^7o(GMfJ{qub6v*aV;CQX
z&B9#e9)5Wn_CqoXeZ{by%h+!{<f7q1a22vB=ZEyP38tbY%>_y7+y`o6$|wvs@jeqG
zOl`O+jg=Ybv^PG5j)#lXp!8tG0>b_s7HnDRcnNN3YkB!WnN!Q=slf~de8ya+W?L>7
z8<)Xj=!I}-DE0JAiAJ5Oq|NyuQ>_dY={u;Ms7zPQhk-^)oBcYuPAZ=E@*BD5?ft8d
z$HX1PQ-$W~X1ESK<p6XTHE6E5Zv91$erAE7MT>(%;{*9ECn376R1Gklbete}Jq({f
z7aWOkh(uMIs}C$`9><&S_Z}bCEoS@XASWU%GS@N!o$Da5QRm3Epa%iJA62hoP~l$G
zA^7ppS5<T{)Aa}PGPcZN?u)IWz#_sA`AY+xh9S#EK`EnFn+{*ssr%;X#_OrOz|WtM
zg;*30-}y$dUJ|XuQ8cIy12tmuKb0-d5}&;d@be20v(nrA;Gt%^|7@Ys+RCXHWir`-
z2(}LkN@uj=pp-@~b=8z0ss%CjH9=m3^U<M-Jcpq*C9XM{1|AP#wg>HN7bZi~(x5|0
z3Y5cY!@8*o4a;FDwwoTMN`DKYq1f&O?pX@`3Szbp!EB%w)lc3g=B35JXZ0}`29%44
zVBXnRgKL%L^GBzTn>R!i`F6K`@M<~x9=jMAhj@qsm7(;DI~eGYfXt5D1wJn!2Y5R&
ze#t!+hK~L0^5{txROnzgzJn?bz9nh(;CavnVo}nrB(LV%_oKx-^Z$4d9-G<Ncgx<S
zKD|Ag#qflT!y0c~K&29uo$x{T-DCFIA^-Vo?fox{$`wSMtf6yl))F8V3Y)c|abNGp
zs6nlCt<!TVu0<1Z?RhggP+J~Lv0Q`mYIr{kcS&{K$@i#*p+`~&;Hz0keN3Bs%K@qd
z_<*=cn9l1!v;Uwgy{Q@r{Q4%<{M8#=)Y(Ft3aQDWyG%s(*266{T4tEOUk9RUy&caB
zW>~)&thGXPjP9G7xUhHdt8f}7lVv9O`3})84*H410nBXDW8J~6ele;Q7#tmvh4ia)
zj(CC+I4HvMc(ca<Z_C`yuQSWf!qBtjgzQJJdQcdkOi?+~3+o9qS<Ujn&FHrp0`AXd
ze~5bZjc?{25R>{oVcVxyVt>f7+vCXbvyhT9o6+i@4&ev~3T9($!RIXf>PuIjQhPPC
zN8Rh!`sMf;9%~M;^izc>Jq_n?5E47YTuO4z-o*bH3xaJL3nb0fMUXSgXEATCy<Ebk
zVAwTAA#oXF<Cq^De)%j)*L?VU91Wc4HPcYMB~9rq`z2AB0#e9VXxDMR=!wvIwb%iC
zU6rd7DnQS{7PXVae(gEiuawD^_Zd~rm4N5Piu!No#Zs|bT9F=J7%$BS(TuLnt|kK?
zxKc#5(z{1Z&b8U{u<7v`!c!5%)BWP8$O2ZqNaEB~Dfw|uW%{{ya=c`}WnXt5)^J|`
z*7+yWC*Jk1-oj{M1a|>X_(UF=ADj*)eIT3GG?0XEe!n7eu_hTS8$XJIwG>?QF8-hg
zgYg%h;5b+l0{B%xx)^^Q({|E82US7rA(*xc6QP9ST0x3NcL~ZP%Ck->=u<}6X`!Zr
z!uRT?zN*!+0UH@2QT=B~V^krN4#&@x{69X8bve5|#ESIZS6_Av@D^E5_6RNrJb^94
z%+}Kdam_^5cD$Ieh@ZZ+?XbVB`$NSa`#O}a?4+7tGg`P}yAkkM*4QaLPe3@9b6JP{
zp!mmCyYEHViKY6GB}<J?cKvt4El%mVBH~3TPOhqKLp|Qbm-E|ga5nm91pCkRn=R^G
zxjKDDAKs+Flq>o?^|bgF;Oc=4-9a4+vG(c=zY;0Jv7H64c7Y!J_{%l~UoD2A>-loM
zP=WVIiIvNF6Z$}{$Nq2gw^wa7L!Mf_cyV;LjKlaElupUhM7xDqyw6KLco~|!50IHH
z`-#v&E-RhA1j@>oY?~2%&~3nuyYpUkeD&~fSZW`4a!~tJa>d~ZKsUt)fFC=RzlA(V
zpDC#CUlk_2ZxN%)fTq!e=izrX^R#vX(znJGs<<opzuUZaf#uv6`+o9WxU@P08M(4N
zyAH}`$6S34#qNA4z8Rf3o@qDb7bV@zhu;U-o__SXve?hbMlwdz$*j%RT<{x$=+gg7
z7Qu~K(STO{;*nK2(FzK9z|Ye1R?zgrF}A+N;0>(HgOh%o<pWE(#v>D|$3n2`X(^sw
zvW5)j3+~lJ@h9e+pce^B1njU*B*VF>&>?>1_YV$lbE9e=h@~z`rB{#>!s|^Hiizxi
zcbvf&S_0<lfC02={L&1CI0_v`?rxf{1Qaem7N+GIKuLd{6XBn7pwvUbVlB<QpyuXI
zeg*3L->d!9Q6`74h*cD+gWLfIQ$kFiE?tG<{O|;w`lK*L@e>cONDZ##ItCaF2{x@u
zyjNR2Osq81O*~M@yJ!IYK!V5UKcq`s!_6ikHqIG9tMwNtD1-wY!WFKIyG&gG<vKNU
zv($$^mR=|hewrH@aV!2s*;gn@G#uNJ0ZR9`;Zt#3Er+P5o57sfYoYBC+YJ{UMr>T3
z@EVCfv#cxRI5DTrW%%`aHZu?lj%})vhUn*XpFypR5maf1USXG>;vQS4;UH`9qkw4t
z2r*}Je^x(sz1@5c$@yubQ|H2Peo(^9r!MczX(jJ{Sye|Prf(^nY-5Sj*QIk}2rd*n
z5Ti`0Ih$KOsG)4qZZFc_@`OvRr=YCn@-pVq`j;Fs9HdY*qU7c@v9l*?^&{qnRwinm
zzr37;pAag)^sp&-`i04-^X9d7CI<x>%&<fK8F;1`Y07D;+*k`gIH2aZe2iCR?kHDX
z{jp5ql<C8m&u>^oI=G(yava;7Bv^Am*1-@no9}jPq@qcj@Vv=m_Q9_+*|f)l0j6e6
ztzYuEl;7nK1{A{HJPTsj+%I3qJB9_r7giK^fz$6Wv-1wSz}E>9a(t`}PqdT)e|nVA
z`^lf-$+6-MO)hqp=~u0?CyJAGi!PXrRuvk8{#9A=#o3DU*8jsPp+R1!3Sh{r!x%4$
z7!+bE*q<+NAh|sHeZaYWSzX5~D?B2#&-LX;X{q}n<~D$x+;=60Uld58ONp`AC)60C
zYG)Z=QPs$nY;}sw3E`^&!&N8x%Pf<U1Zn$XG)XgwBt)x(1ei>CdF$<vp0Nk>77uRv
zJ}f}GT3*&s;v|!Vaa8?kFy&~J0~re%Um-3aA5aiyuXoMll--;mN0t>U@{JS)S(x1y
zIhU2dVLB(*F^!;eWKlTu$2xWa7Eouo+fCR!-!e+FD$i}`569eW{7QNKKBb#$o(4{G
zN;b!!+!-_m3fm<)n|8p*edf@~wSFbFYQ7GmLN|X#PlC=9<NL#9K1eNy12RN*Y%dl(
zR7#DC@!kb0_X5WQ=;dy@GF6p^L--+psG_-oRXvwhHx?pG@(VxviF0k_W5JO=eoC*O
zI<BECtAPZU8ynJaRhi6dFX?OnXsJc$tZ6OjmrTJa!)Py|tpXaG4eKXdooW|kf9kNS
zPZ3_KoR>?@h@WyQJ`_$&tOW-SjlUMot2ayBE>Do>fNh^Ov=MpqV9Tc!PBcZ2TXqPM
zSLa2-u_6~<WSp3qxpsScvgO^XkUXFDle<ZD681JC0RIK@gA%}#lrxqT7y8GY3o|&s
z@G{f#sxL<5myO(~3-B)|3RTtlAHMdo2Xqx@fsfNmaN_y?iFxN*Yc3SBx_o;^X+m0=
zOEkxL!Q*-LzWYGG?wKjxxpl*X4phO=+0vD53W6#^6uv<h0FBNNM_;~Nww<cUG;JaK
zxmfy`^1JrBMH+7QbZCpTEhEbi#6MWLDFQJ{()5_X#Bb3-illyW0xcaA2)`W*jO`^q
z`)NR#SOFg&iu&9P`{3Q!c)Q&xzN(A)bxyBi(?`1VuWY{NZ-=h;FGo@Z3{*vZcY!MF
z1ggONF2H%YMPRf^F&yI|+k*AdPpj&Y?U?RTM+GH)lz87B2&BL3&^`dbv++b56~t_*
z0jJz`at1Utmn_D%-Xr7SL_5rQ;(YFF(#sBf4YG~KLg^H$BMW7^V!Tg{IYc;&3Yg2B
zJHY4E=U;yBV!2>e{j6aB7@TOmKH*0$2fxi(^DSa(Oyj=~)X;RT6a2W1D5C3Dx37#|
z$S!(+>{v~jh@(_MtG%Sh34l5SIRi1fK(AGd$1ZSzdTtje;YJSnBvZAEqdGdS*57P8
z2{a7;Y&2h~B<Rb;e?NXOCH6(9!z}B{apJjoxtqAzCo_tq`V=ZFcCkhL?Ssw)Ws7ag
zhPtXuYuy7WQ|6-GF>Sr>E$%}aH0}Y`%>ZccyiD@yU&*3eim{?SYB@BMYNayA2^QyC
z-xa;aWggDlUuEvIaAM6!b+8!f+m%I~f>k6XGHQ}sNU~})C?G6VYB-Ks-}h+1siI2_
zG0V3a-ky8RW%;vm9?F=eD8yKg(vK~|MGZJW2S*aujO<YHtC6A8Rd1T|<yX(WcRDPz
z*e4W?Vsa&(J}vPEQHUX)=^v-qg6tHvrPtugGY>3C4yl!1BR_Sc@8v5dY0W7{$(S+>
z7au7<8>g@~oAy1<paH}aqVWe~gh|`+TO;^LEQqliA|RbqDrAyP=Nuh24?$eEfLati
zYB*RjhfTmqG$ADFuN;0P{yj6>ZF*eieZV1?djrlNdB~nwjxPfxP0U)kks~yW!7fmz
zN)=e1M|L362BDu2y&kmI%XTw(u2N8XS9m9U<>r02(Zh35vPzwug;{T&b!QZ^F8WY)
zHp!49_|y|Y41NinEd|6ZlS;{usN*eS`$b1H2<vHX@WhU>ofTVaKe5Xahu$jhKeMlq
zmOqD7wmc|*?MafqK+XJuxGJ3^of3euhqX8C0<DB=bNB9?mJK+5u0A4*WpHqUE4y=}
zO@t)`fF1Z|z!TdErJ=5`mXaD&GZ1y+6*D&obz#{q(VZ08>As@}yF3EC@lAF0U)Ox>
zWP}voet84X<6;FDXILO-vjfQm%@!{A4LxaesSCHP)|oF=z8xn&wy70l-k@|ej^(ev
zIPo-PMkb!gaXuZ|*fH1{wgNl=tnfLMT#WzsU!`I|DXrA4W$aBnpB`g1Z~aokl+J_G
zQ3h5&t4@{c$NJgwmWupP+=?dm4`|@`(|d6OR9!;4&Vs{8I5+=yML)|EZgX>24sgi?
zn}|$jTrA9|M`L_=$A`#^ATJ>SI{4!Sk@iDW<KoL&qHaZm3R9~$@jtfDYy^CFGJe>u
ztx}k6>gm3ZU)n@e&FoSfd?XyBiWJ2j`kh%EJXC|{uxPmTC@)3bQi>GcYWlR88#9E>
z9q8}3&a2eMI+36w6Ev9!Ruo$iCkgfKiUnNY4IAs~kCE=BmxZ-RLX?k;@19&d@owtJ
zHx@^=%mdf<rVySMr?DO<9ElWb(PY9TiI%H!mt<!C5%$jeyHBp25;j!p)cMTRthcGJ
zY<N`ZoX}QQ#uD1j)6$?VZgKY5H13rubpF+jp)htq=U>e+8Ib9W1N}&1b3X07u%!ZS
zs^SZ<a%1Sim1l|OuOII?KGm#J**4?rtu*_+$?vBn$LM3hb7ULPi_pJ+DQ|C+Uu^o<
zCjZObY<b28t!ls*MkH_3&*|XE8pcr5ipU}8$832R$bZtJ3O<jOQH}e#Nl*pNJqjon
zQIW{;8MA5*Pl`j{*Olq+)i3?xemi1?ybIfO&Ik1RSuCYRntnC>S}?^j%9hfNAX=yo
z)qLs86N=>|juyqScV4|ZK7A%?D@99V7CLotLfpg)8XU1CCQ;Rifwgm=pQT5yBL5%u
z-aH=azug}n5wbMdcO!%>A*4ly6iJ%WV$D<(k|qfwV@4$V9-=ITl%=sNWU{wdvy*KI
zVVFT<n8o+ieV=pxKIgta=X}rkJih1m``+jG`D1#FX_oi$x~|uCJ+J5WyoMyb+qWP?
zjOAiKENW;f;3@UMC}^vQc|@=_<UxCeF0M|%xoP+_u7hZMixNUfO<49`I-jQ{U-v-{
zfzQ>&AE!Z<mICPK%d&3P<D3Sb<lfQUeY(-Ubi}VL&3?i4+9metaE(Ffdx|a+63im}
z%oieV(S>e}Z5WrBrFGI3jQxoS-Wqt4%wtP`FOi%WnL^FQf&B>XVK_RF*>~3|Zrc$x
z&@Hk~`a*2qTdO=V9#r5$^y_UHIIwvs_a3DtI-g=!cdR<Y?1{6(PpkbZ?4Jn}C9<la
zt>YE25JO2cs92X*@bj8gmM3$+^X}D?3XQ+A1qpbeRmI@5Cc#L!?AV>GMwozs;oFwf
z$wL~Zhp{svXW*&~3Nw^2kpUCgSEt@gKZhOQn{{}!IDg6R^0ybnl>nn+^s=VJjtc_%
z?b5cRwR^8TI8Ze&MoP_Es8eAku~g9hjfcHE@cHwSGQ9`-lkdE5Zd%z-j&Bxs@i`h4
z<jTH^V{rpiMiI0|F+(*7Yfr2fVNIPI%CaE1EVf){h23YA0iSi0p+k!&f0+!8T?3Jy
ztZ-lLUT?ob`!5`J6;S{SkQ@ow7kMYl)_5Y0yB+%)bFA{+doncq)*{hwNx4T;A?9Fj
zSxxyqk&CA0iq9a2K%kYmOal&?@faQfr}FfG;w~nMDLycf#}F8wSa0rpWO#%Y>qk+p
zmz;e#QP*?JsVebO&+S4%7xNtTl8Po*@jzoa7ZLh^5v*;1A;iV9;-8A$q<}(kiU|^w
zi0XkoPKRUQTaXff^sFV=QD8J>L_IPpgPZvd<3t>nL_?8Xx@rYSJ+BdZb!!Uj%_ZGG
zrfDTwNVYWh3TElcIeQ`j=coRPp(MkUmP!u(u<6o><C(8ht1TaJbbJ2wmP=i-N_1i5
z`>z~#3lca_wxF}1&q3awDL%ybRao<<=@U7#r7YbhW4XWGOq747(|TRHtTgDp;I0SA
zP&gY_9D@LF%Vl-dDNyvA38IL+xnW%kU$jTPcf}*$y{PL~g_2|Bj=@sP8j?Mh%;9yg
zQs2g!A>6f04u&QzB=_=<^N#4dmip&PYeGnGURXR(ysOZv`p9&6R07AtIP}Y*BSzq?
z9qE*^ixlXC+TEU02;Q5#d&(R}hAsWBz8X|~n(bECo>KAFF&oQMIbqPGe1~g6EQm0X
zh$~5XSX<Rwd-N4L-<51KU;X%7{I}fal~GKdcoH%Sumey~hBn!Zl$wo(O*%&A(Qp>z
z(U@otPq#dyxZ_otYTruk1m2EoRllBE+8ULiwqyec7SG445UspO+Hn>;gCoVPZy6Fc
z@73((&0pdft7Yyf^S=DJ?1i?rVChHr5V#wHraVhRko(sgSV!NN^<h=eRV`58R||+2
zgO_jOuGX0gP}|i{F3uP$`6iuEe^U@w6UMbBK?hOOr>6{HYsKn>O-XCevUEIlLmsya
zw|Et}Q*#CUTR<Mlr6@Ne2NPTF0^>+abjK=@zj>KtGZUO%D-Q5VKN>9x@~gY)T%gqc
zA-EZn4Lm#UmlvMQY=BAs$P_dHUf617`Lg-*&`Tv5E#^+5C2!WuyLtB+OR!521L2on
zfpR0uu~yRy_Cb@TF7mxIQt|SEudB_QPf>A1ZD{!JCnMTNw`Ep<-<RoN!_SKnl;IOO
z@M5I`y19u9#eI^e8P+5tIKH@QcSKz+e`NQlWVlU!4n=T8;v-+4yVmeqT5CNhlo3W2
zChEe)8sXuYJKoe<7XA3_QMBx>ahRnxbn4E_8<ErRK1>|mzE7)pFrkV42T%<NjavHx
zePGB#7I<s?dO>!#*uu`cGs7+Y=;=F0AnO8A^r`;3=geq~lM8v7wPSn|8YUU#p_li_
zK>O8SSFrPd(7&wU2wb`TBAxDy?KTw2q-M|>C!_5aY-{T(dvr7T2HmOzM@N1vC4OFs
zRC$;%V&62M&zJQ{!Teq4s3yIPF3UJbKn;f0XYyj@f9R(jU#}MFPk%Kyc8of3Zck5J
zOssIegn0Her{e^;oLty#3@utF*b6O~&oO31R3uhC6VYAnn!H3F30iRHTe2Ga+@P!>
zZS)YX;}W7i@+|@9WB?*Bj;GA1+CAUvawkb9&F9U6U>AE|PRD%L;4gQXyAXSmk^V<a
z#J4XVcm)%}?aUZa0y&()Mcc72wqT4+pm#Q6W?t&rl<buR;t_w>FoV~rsho%SjbQw^
z`V+Nqa%(-T@<#LoF692c%t@wZqt9WzisqN;+l&Rj)#C3qH?UQkNS7qv_#(R>jbGSF
zNJI^vXYJUapJ(X3Clg{(w|*icWHl(kqw|M*<|@VPG6zu!#{<eN1dkkwGf)41&O$AJ
zdfycDJQ|KnX?7Is#vnZNzWorK@4vK;y6z`b?{%xE<;~T!L))s&C63tIY~z*#YZL{l
z#DIHnoN*h<wa8KcrG~);DwM?|LN`T+Qm-?g=|LsY`q$90wza{gJudMkMdj<}iqVEW
zjk=yEb{pN*C#myH!<q=;<2&H)%Y$Xq8sNzQ4k$?ITSWboIW>ERP}%dncX6M(o`fVq
ze&W~;v$OTKHrcl=za-VeUnbC$LFGzVfhyEn4v)I80Pq#{`;P|>W2_H|N#5i4boRB^
zPx7iZ=IcJymCANY(8k5hij53i1YVtkRKa}1*yBq#V)#>hiuNSq{S7X4+qD%1+^3RC
zZI|}Fu|hq5t0KJDsXV>$0kX)D6Zl$JY%>9xan#s=U9`f>V^yN!>FedVkR|d-0PZOa
zXj@OiKEQ#KsKhgbR$8rSW<=1r>pQ&uh*;xCeB|#R7}vQ^|EUH2UNTq2Y3D%c+ta%&
z*w^;eQ1C1sWxBW9d5kV-&6r>)lW&y1<VRT7+K2HDe@cos9Jo`Gax9N8xhoYaS8UY~
zXbAGTAk6&zokd_#Yi3>~H{EH>Mvl>5xQmmz%y>r3Wxh0&ya*K8=H$u_xudq3il6bN
z`Y=;iI;@W%f;t<^Z!`v_KqG2lIb%<`DEt}GKKC}R^xZlwKhZ8}E~vQw4BxxBhlBO8
z90P03|3+@+@8xIyS^kD?_53dT;lEAd3*R;#4Irbs1K6*~O69;Jj7gT#19I>L^V?zK
zP0Kr_h85Vc3!kJtqxL^UH{IfHcoiR+Kq<ne{#$jQf4#i0$2fqQgu%)Q_4pkiE|>)*
zb4S<eGpR@WGTv&uh#<<{#$>zccDGB17w<p$G-5(0e^#AsoSBXT6q+(zEK7P0PdVRQ
zf=09?_ZK(k8{_!mPiTCVvsDVR>1uF_e{kL5kpoGsE3-$&1X0{q=+dKS?}}Gnihq%|
zBi=dU<lxn$?gy8w4$4JI(!^T860#r+5?Y3nTHauUH%0n-mJ{i!RARq$=92xgs{(BC
zg$gwb8U+e$NU{|!xK5dNq=+Fx<IS&%&_jtfspGx5uO_HjSm?{OMSOjc>YT)tylwB<
z!d9ONVJ1N9Io>ccFqULBs(0dfO7l~<%`Hd_9#`Sme{rwg`zPT}6)_hTRHr%PUVlIS
z;3OY6-i2mDhp;|@4^})eu$&lYH})*DtNz5ll&fC-u(<jze%IR@G5TC_a}gib@ujc+
z+wM6a!2D1CZuYI-|95TZfB5zM|7xwUZ}t7&eOy7t)~WR~c94~c@E8L<t*Sl%sJwSN
zSDF;>Y_sH{{yj}eWugEwzmr>T&EPCo(K%#CRXu#9lRD4LWgK-OqbB-_iq@MXl;2hN
zVJy7dJVf4P<}o6-<#TB5<`Yjrk3NZe^)PC{AF!3G7$Uq>MGMU^g1C3I0x{*m>F(A5
zDyXzy*<Tu1YAI8c+r&4$82u3)d__btznE+Jy2vzv-B1T@LJg$ZW-)BY22uCg@5-<)
zJoM{d-Mv>=Fhr3ny8e93&i(tj^)@2u79hkjW@)(NB=eZbglnBprT0X-{iE{QzMN<m
zn>V}Rgc$D3UHL|)LoUxAb7=7fN(W*;pF;wcBuFGDN^#ux-0GCa(FGQiqKY)Br|t3`
zyhmPhSgeN*eeD>1%ATN+qY;z7u#6gXr*?D^Zoxta(w_C7o6FE7ueR<i$}{~w#hg<*
z@<bmqcDlv}GUR++)m+Lp&~V}fK?d%M8&RM>rUlc@k=RejlG~YXoK?>iWbEx~WA<WF
z!uCGN%d8Wh485OS*0QuB^C~-<b&(<p+V)6v9rU$j;&OPc+Y8#dQTG7JE-AfM%b`$f
z@0U;NIH$5Fri7L&xTL$SJUbw;T9J)J54sC&wFx9!YPg{zr+Yl^b@I%)`BE2x=?gU*
z)|UEit6IE#u5AR>pr$t#q&@C^b)2lPjw;L|g7c_75z)Zyu?7o6-gw`WFy<p@Z|5F&
zT+*OCqW*;k<)i-I2GFmt6UT|s&t(n$3PX7MbV)Fquji+VS6SWY1t`@Ir1`FzAzDwi
zAX6g{?DDezsu8STPh=JM5QCaR1&^2AMANB-eI8OxhLR=KH3PQ`gSe4jl1&5~UX{G|
ziT!foY_?b2;S8KILyc|<Zm1dI_GPCFpn!kl$s&M5wlz!^5o@b$PSx=3N^ofBZ%Swr
z35E|5FF3MI2YrCm30qIJWYaGJvQ6=N<eCl6eL)QgLvz_aF=oDYlHQ$dki5&mSem{4
z;`SzMfnfO{k(BKzKU=4`n)-VPH*ivzYc%)QCyorGEvyD~{d+GaE4eN@zpH$r7F76-
zXYgv!LHWR=!?jaWqUlro$VpiAT4rd%C$0$my`!i2<xl2{t|&d<Qy6*c(_U4Dt9z}a
z<Ahh^BUm^LJN-U(3|<}@wAT4>e4EEpYg$<8x3Pu8ZK6JBxz#NSeY&1M(I_-|?I(Tv
zR{nDiJ!~XKn372<nphsvnSw>b-99Dsw2TS%R76Je=Tw&IK#PBL-bwuEq2r+oL_Zo<
zOR7XaW<k#~AJmG_Eoowt`>JxWeb(z~u@N`=XVSh}kWT0w;>_?U>JvVda7qNz_js|x
zSaB)?JB9B+(v%P+e#&`jA`M1_NirOIiGAK4@TT)W$e%>Y4?OpnDyvAcc2><lv*9ou
zwHnh<D@IKLfoL$puAilcple{8Gh{9Oo2ga&Zhi&kk{u-u$8b1l0mI!F0;5Yi52<ip
z5rZ6pFdx4l@L+5~k7d0xYu8*422n@3lrm-SuB)$~cH)_+uN8O7GPEo=z2*RyqWwvG
znIi*Z<HZA{^yk<?I25mUs4%<R0694uBV&Ii`b#Ul>$I=d?Ty-0pJKMskDo6j94LbX
z)Sti#u)YH2KV&o#0>1?!C;lw-&rV)f>ZNYi&HY17j+Up^Pgf-KugZrHZEtE@e7^Ge
zfabZFajHune4=1IQ`$f+fYPE=LBruK)vFUgE_&mLJ9^4*8#jsjU6qfjLM)bZW_#n6
zE-sRWu=OW+6`~m|nO)4JS`@`(vYqJ0lJ7QsO0}JSbg0j;`_+j3rzhX}qMuA({Wx_<
z8t66+gSAj6gdKoY0d`;fBvbhVp=_|U3w2Kl;HeK&nyaW}Grp9Bk8LF_2C3JgkNX>)
zoOC<Gn_gg-p~B1qffcSeB}&&EchjeYmPDr8Jh6XDS9x(c$GOzlQ~E$*6`S#$6z#Zq
zZY!;}EZF;UYrTnaP{}89JzT>{qtB2~YaTxRQX}R)IdL7&yQW?pGzk89h%5dw5ZttG
zMpn(nu=Fu%%=8LG1Z<o4g&uDlpTo$vakb$_AElmWPhPSqnL%1Q5=6xt)U40bg&FtX
zG6Bfp0_DM+>dQ6v9*AD7z-e5B5Xw>eE)5b?H?q|0%^(LMJ0G+Cye{&Wg02knHE0{w
z#PC-28*KL#$&0Erd)4UlvNm3kb-|OhIfGI)Iv03CU}Yy?gnKEuwG}3e5oboOjmN^v
z`!;&L=>hUb&!0VcC1TiIx$aJXvbpfhSEX30in+q0FWN14*r*{QG;NTb6K5GBQq=yQ
zCG2j7B%vECPLs9${-}NCWRmhqO#cq`%5~rUkKt+`U&W{J312^jmzr80wrBC2r}NQb
zkM|N>-P7T%b<jTLl}pAlHTG?v_05HSFOlmFbZ%`=>^PRk5pt&x!bQ>~!HX7TC74gK
zs_59pfzlttFH<H1Zp`=3kG6Rk8yvV>d)IU!sIZXjSKl1wI*3zKXsHxlB7SikPm3fX
z!<HF{d}Ct47n>z-snB<+au#aJHjN+6c_aoO^r$25Zz8CVC_dp+fUgis)Q_pv-<xSp
zfd*DwKGyxQ-gRI3r#(^egQTnRgA}ABu3i*A-B$}XK&R++*3-332-EKcLBRv(4vhvJ
zl=|~I%qKA0hYSO&54Hg@P@PyPvlf?gYZW&&HA5dYbbMyLkOUUNG-`2(ZiBf^14+zw
zc3Li3s5z}o-;tMGKd?O{f2dxll`1v&JXA-1ht$nOTwc;Y4F=u<zgEf2TuhuSQZpmB
zJQWKwltbs?s2)+dQCdw8_r6V6kT<oOR&>oh9bLnBAPvIY_JY8F56=aT%B9vL#|&Vv
zLY8Wc=g+)wy?5kc@uSIe?=ENuMA-CBN!Y#|(|J_?QFe+hf*sA=1(+UZdci&b8xA!_
z@L|q=y=1}>2^;FCeRw_{KKbe5)6dKmU6(J(TDHdm^#iem-6#A)Tl{em`bgf|6Dabf
znc>csWmL4Pd+|$-%=;^oeFIXy-sb^J3s<?`0vkRuWlbu2$!gBzz*u`ujGPqh-|JlQ
zF1fAKfa-KE%{@Ge`6EW{uJ*2pBxR1`);a6N%XBTw73Q<^&E^z)V)Sy8bN`8iK{HQJ
zU&TxE(1Y!zM^c?3pD!PBaF%Pw3hYav-vH2R>zOBd(aJ2bb@-2(tE`SKNSx?$&GZJ6
zG6_0(^64zbz;fpU)*j3yCP1!ky!Rrykje^e9Tx03$~tw2?M{?h_^KoqA1}#Ky`Eq?
znDyWbU$Jx`w9FM>59h{qkCUyCBf;l-YmMppBW~k-3@(qol`%^{cHg@pb79oMzEiO@
z5bK5F$-lF6$40??W(9$?PKVN3%>XJ=PDjb2Vb&?1C8KlBs(B_fjLnQ46zll4PilYf
zl@^h>4q>jncsI_AE-<DLLdppeG^?L_{4l$Uh?Nea&g0}Aafh`wgAN+Rgr>1jEKzWO
z&2b)xNLjJ>v~v^Uppj0M_L`{H$Le@j)k9JcmBa{$rZ?QOv8ZvHeIk6~Qs<sKY)tKQ
z$Y=24MAlgRJWKN{?$#yP&?e>pVaP+*N5i1!bcaK$XI`epRP^~7X!5s(Ac#D{3?z1h
zy`X^`SrOWdAM(FM13?4-PuI0}c$2SW9<};?j+mvosHIc6^0UGVA!&$e)4Mud)^Xn$
z0KN%&itX=er|3}87McL>=U=jYdLr`dfr5ZnBd;6UNb35~6a|HC2aE5#G}4cQj-;t=
zEYko8v;8EEM8=(^<TXHt6C3ZMpSGrnyL;@>$ch-7pSrjzs#yM3^viVlZP<NFLFfF5
z9lWn)LSw+)4{F1Q-ezgP$-SCAV&AYH+E{(>@Vx<1>U_=6iQ3|}>sJULZX`I6IhRz}
z*MlM$8{iYZtikF$5{le<DYNB3oE^HcL4ViCP55=IIENE=qWaf)aw=PJ`%c{BVoOwD
z3yR^Q&M|8-7UUz%)%qVa1Je|-%AjVwCoJTi`@>Q(%TkVAyE!ewu`30;@@s5^IkyGE
z&#B-W!)y7$NxkSP4+hSLc1HJUO|!`nS5@7aw;!aU9Fc?lgPD^XE*<yvO_0{u`td4}
zS<->D^~Tun;lpSrA})B9O4qXv*>}xsB`0b3<5YRHbAzRnP(|uVxb_7IQ{n|dhk!zk
z;O&`DD-ckOJk8{V84(d)x9_+=`np+~noV9@J==<tYLBJv^JJBJ_V0U2c?vMg$_>jj
z^_lm1vBC(7vdP!mqnP@f0$atB1fB21BRe!c$s9@zrJEH&u;03o(eM&wQV*r42@cZ)
z#gygY#xp@9Em}nlc4O+=BMwqEHjLS-bh#~vnKzo<T)1Srm9P`qTVW1C+nYk#PO$*m
zWc<%*3zM?BvPp%Gw1j~($+jNPtIvCAq))fiC|x>h(GV)<5_Jz0ppL)7H;l4VqrlAx
zWeweT5X&t6o}L-R*qv*#J-46mD!0_)^-X~$!cFH@rzfg@sj9^$D}0=o^x)90U+NUr
z)Kj!kP`!PGtbBZFxHVKSG{m`1YeXS8r_}Gp((p{X)ggn#ws3hdxhso`dJvrP#SIbq
zNrpKENjgfmry;58w2Osfb1WHhVL>5n{CrPd!kCjZPZ9NantWkCN4SJs{OqZBu<Aoh
zb)fwq_=^#W6T={Ccc{xse0}qCR}|LPZcGV11rWiEsBN)c$j3m9G<N&^{n<x#O4zS;
znoq`Pp{;g3!%xGxI=*miTDAplS{`jx=-+JwD=Rh@DHj+qs!<3(rm}WXoPFiB`-%mN
z$DDqO)eQ(Ugh)dedUmqq^$xO9am&cLlH`uD%L6{8J8b!u9v!Ej=DX-qC2KDbjLer6
zr6|QuY4ZJ$)g&{Bvlasya&_g8?<Mep9^VnlCC)&7cTh-Hfi)7sGUR_zuX*(+s1*U(
z=#j~&@sa*(9c~KtHU5~R?X@L)M^pxbPMBW$_Nh$Vr!$Z>3a?0M0TLEnI6mBs%%PeK
z!3c_G_^?v%_2jQoAA(rkK|?1~ZRDb@Y@O$PL=1Kk3N3C)w%Ly957)<jiThYUAJb}n
z3?b;>;XGPm1Um|v*UArs&=Cy5t~%AZxyk`w9uMy$Cl1k-vLV~h?>>!uM1PuY`Fv)F
zd|foY9!FDddF@9(g#88=BRET?f>^}(m8G|LZp(DSo-%p&yIE`d2Tz{Pc&WFjgM9yo
z1l9j<)k^+9eOzo?{r`wp;I}agaMJ^Z+PJ}|L>N;a;YfGf{;S{i(;Lh+P_)%=f^Xi=
z2!tEJCn9Bcebwz88fbRZjR3n(g?FdG7nJUo_4bOSa@o_T14O<jfd>2Agp)vZlF!X>
zKQj}nh<3M+Ipm>Pd-QdIU$$yYWe4TcqLa7v`N2Sc>q#z#pcl23=1X>LY=Un$+;=k`
z?Q1>&+)ah<`!jPJK9N3MIr>_i+8a)M@y9>5hj3_J<K2fPf&OI_E+|8QCPqX-7&??2
z>BMDmbW4lJzA+7@nIm&WS85QdezR$!&c{lQ__@C=JA_Pey#_I`W^w<RI(qX9a7_hH
zvOkFArJQwYnCaw9yECsm^E8j=d_emh0{M96&cXT3@#f`lyJGwI-ncULuag_83N`NG
zOLqhIpoz-8xdFc#t_aLbZ~Oy7nan#%dZO(@oqkQgjhW%=@+UREN6S|}I&pA#%*rn+
z=lbao`}hYNxpZwb3rbUpCHYT{2W3i1EHWgB_}y3gj-URh@W>o?rdH1RNtBnl^oWUY
zV(_|TAUvm<<wtN{uz-gEoxIL4);4mv8Y+7}&!K9Uag$aPU2nV5dq{MboJ{pjwV%F3
zebP8xg#p@ZTBDY*imt=N9MhBP_x$HHm*3dG$hmWD@k8>IW`%Wn8$@pd?93c*$;vT?
zI;`$Y0k{W3(ru>oOz$)y=Ho=+-SK&y9ZEJ{U(7tXwG-zjoRh9VB<SX>HgI2Bfri+5
zbTGAq`M7WoWlw1M^^B#Z;eG)>PbtU!pL{($U09{A&PkXiSL5qbOQi*Ru{&_}PH#M%
z-;(u7CggBZ?X-J<k=uqttK5*{UAOAci`NzRmouWGB!9l=V)5*t2u(sssl@Rhc*XnX
zjF5>wH)1z_qQq7Hgqz8bLmNHEOPGA!_JL}G_lKR<^RQv-so6ys5&CemxwRQ3L_@St
zIXh5gL&Kv(^L7v4eDvYf3&}}7#Ct`(pj_c<q>2P!2Ebin!;tJ)WyZmB3hD!UUS8-v
zo6w+!u_IuPk1#62|K7!w@8_Z(Nk{EEm847=_>j5s6t!y$GBVD3l3arTWU64y)m$``
zXdPW~FgLidnmLn>%0qD|_6!!a4(CM4<{Qj)6;G5Vb)0=b%uod4N?;5IoNPnFwTufb
zw}fZi9@#S;Mv}g^ZW1kjV^Ey5uihN;pm$=M5!WD+Y}+a1=nDLPf8=vJoqYGPJ!OyX
zzT@tyM|m0cw{q>5*W#kTBO$f09mpy%5<7qjeS@*mprOxhpdwwtyob2rEiTG*s#`77
zxvZIzS55TslH2{{Y8_<(iN=1b<LH5Vzy?1r+tc>EY9wCGbBOjg_dk0Xm1(In;$R{M
zSyt&$<->G5_n$_GGK8|JG};k=@_8eg9>uFE*>NWY5kwj|6Y|qM`D_2mXK2D`cE&XY
zTLEn;(>C#>Z&^Xd0x>=Gv)F!Qc`WG=8EH?07qULH6uxriQ@=K1D>p+Lhub|4pHe*8
zk&&t{k`33|uvtZ3WUy6IDigtid_w6AVc@cFN{SX_(d=6eWV-Z{ax31hoYp_zgn5@_
zf5rCL>lUrIGHOWbC<qlE)08I>4=2d@#qUH*{pKymt_q8=i;5+CU)v5Dz2Qxj-QQhX
z>h$GgZ?c5?$jQrNtHk#R;u3%fI4G2{ca(l!f4J05M|R5gc4kF6=cq<or9Qu~rIb!m
zyn?Et#N;|3-&wAY<K*Lbemgq`FJ(DO#F}K664o4CX^7g_(y`;L(2u7tKWq;CK$1WJ
zt5i=3W;HGE!u_be&j59nAXVI_LRtZ<^~f?)->q+xmwo}b)M5-E8kHPrg8jv%>cJ;s
z?<aLcf-1aezj{wwMr+gk$CNiZ?}%R)Q%#RdN36hU@`%t|-T^4$dQ*{&##l9BDB0}}
zqb+{4qG|n*VM48^(7e_AaTa%exbkA-6!-Mm&tv;)IAo2v5G5b=DC>*OQ#ANQ8oXjT
z7R!TibtxmeMAU|aUKG%gi;++aI81(!a80{i(aVF+>J0STO$aMtVPl4wSftJZRr`8-
zLj@XdLswB<Y<~LPZ^m4^8if{nMYgSN(>DtAnLCK@8xI?=!YyW#r|NxWb>7k9v)v=&
zCFDxFm$L|U?`}uL*YlQbvyZClbM)1M49tiz1gAb^90Uw6Q_OK@g@Ie%Q<)>@6Ie$S
zf=a&NUbQKvUhUM7DVJ+O6g)_{7zmlhSWgxpfZ+C>-Cyw+9S?f*8`3x>q}D)I(Tsj}
zKlGu%GP3(94*q2evbT?cU_JlIgmkeX*}##XgHJbOm6*CXrh%>XcoG@L5D{XXwkD!_
z%(fss%}Ed!T)l?D?i{o|RTZc%XY#pu6D9QcPf?FA=(hW*tF#4N63>ohv0oNr2p2R;
zA^<JxT4&fozL}I^3d#d_Yw1vgzyFgk=cB2Q%Q4=3W+q5(@T=+s4?>8n1nsB?AX+;N
zkPP;`LSu<Dmz5UIC3zGsExws?8#3(PEl^m}sTj;>eK%<&p^%fn5FVg{M(-(D2#S6H
zv$NlW;Yox`jM|x?H%)j4uU17|Pr4doW1*fRbUG|Z_uH|h(ID$PT%L4Z#(e?}0hV+7
zJ6d-BkIpx(WR{2J?s6SJPV|YcIJw6;PmOfqcJ>5nN3l6-Z$Y4+dkAyU0M7{i2^d+D
zOjBl*;r4F~HfAbDhFsI6y}NIs%D>2Rzs!m6i_r!piCr7^O@7icsk21^+RSj!Y}gq>
zU~VGFS<pC;M{CBWagj#I>`iv8s!5Y9H_7z&ElB8_we=dYj9dI3uqR_KNjvomzk60i
zxhNZ8y>azw4C6uuAB{=zrRq_hH6dKQ^?p8DPIAtuZxT8TRG3|>T(2Xk3hmgXk>L?L
z<s^5(V-II<s+`Mo;9-C?BPeWWjP(Uq32(rUm^D@+^kT9W+zVQvpTB-_%{ZQKLzW~q
zm`J)rHJNAe#SVu<NwFXS(<GZMZlGAHlbWH%@Cso}x6A{Sw{*tXwZa2=-Obouj9X6~
z)DffCH%nV^d3l+pb~=*en|&#KPeu6E??`v{)!+sXsmNVoiVu*F3FMQ=P&-1n7!gTw
zn{r<DCdkZt4UL>`+@8IP?F*1k?7RN<9@1%t2%zQLS|CKfu|D6wSG~<#DG5b9vR4nK
zVLpPI?kLfHct>G%Sj&rXU4^|}9-J4R!A8S*c#x~8uC()o=#*1mg&VJ;Zbs&flrCvJ
zapE<Yw9r0;PAWZfsqQd<nA~u~p&+NMKLVrK5IDZmJSKPpZuX;5dfVTMp#rAQ6z6!{
zsieneyX{b8lMa~dVr79h!wq$?pX#KUDYdX%np3kZjNDP+eice(zW+X25xsEpW|l+y
z!KC<6$J0-~efRM=i~d04W${$dk&L6X5dcF?wAL5Y6Vvq@tApGn`|-|v$=WBwsQu)e
z4<2_;FWS}*)m$sk5fo-$(=@p@PB#C<E4>A2&aZ(bC&}UWfpb*HeJVhwuv(>nGyL%r
zXpp`S+Jdll>@}b+$yiTfa|Pt4Z-*k%4#3WSHeSaY{QW1^*j6iI8F_}Nf-T6rGO|kq
zcL-282@&$lWKd-)I;`J{lWA1MiuDV2kP(t8BW{Mt_uRv^vXZr`YfAYlzg@vJ!izEv
zGV2J=ICk1Tl;jtRrdfa7G3Q6Er)nbB&!(3oMcE&3Iex?7#@UFxWf2L`A{R`T*oB)Y
z9G5O(E(_y+bnuXo!#zYq7;rx{)qu_jT|DO~!?=cQu<6>frovIAS6|+fFfD9DOl@e@
zDc^#qy96R6)DIJAO7O4>&`tvGf^n_?6iV2}Ja-<e{8g{$n!DSDg5%z5BlfxCSMqKT
z8J~ZURF4M^3Xnk^wV26p4?G8U<gmtr-ZlO+?iXvgc}VZdt3<s<j>HIey0BiyTET*Z
z;-JpJ?w+vvkZW#+3XBR()}?_aoB+dvejIa;SzIT9?V6o{h1UvrPL7O9sx?O<2ygAf
z7MD*KsV7LC^n8-}`PTHUvpPU^3t698xdmCwoWkumF(m^hJ_^nou03X}@l%baj=J*G
z+#}CBPRrm<jy65_0S7~35;-z81tUq&U4MXF5`IIx=r&7iuw3Chp|;pp!;L+;=KW4v
z$=!1DYIjEMuc{@hLf|gH=y(hWV&lwf2JR;m5Bht8`rU`qLl@U4NZa~SHS~}I+1~`Z
zL*5sD7l?g+V>I?6R|Pr^_t_f(R{X1gfOTNf+O!=+Xe^<Wn)i}ss^I+weP&4m+dp)n
zJVm~LuTz?E>yftoHS&QL`|}?pa6{H(hoB9I;SYd{?Q%$lZ1?UMaiPXEkEV>X8_sWQ
zsw^VI1KJW#E+JmyRlvc-(acOKzKho>vPT`oN7o`xR;bx3N;iAD?cZN2+xDdzerNig
z-YT3S)I;5%X}@7W$pa0s5g97Q<j_crn9I(`oUixWjyT%cLsk1G(=OS32}=^_D)#tL
z%pnk~55E5i>l*-0&yh1`sou2RMI*F<CRy!_Vcwzgmtx6eyPO8ZzC$n66A-##VFLPB
zEnM7G#M$QvG{uy#WxesiM)TsDZ`Q_r-WK8((la_L@1_QIr?DdB6vnRA#!D&K{j`#)
z9U_(;DOYoCTADERnz?V2$o|_rtl=TpVpao^y#hD1O1(_Z48&N94Ka-65wGSwonfl;
zKilfgwMxB+%JB#?{jMl&J)_Ck)rxQFqyy_(-(v(L^nwXV*loCaFYIhR0v`rml|t4y
z3Un1>G$(;b#$ji7{*3tR9xmHd8gedC3SG1AL$Y`nd%zjbFgt@Xkpf$gr2~Xt%>1b_
z&2{Sy&`aJuH4b%4Yh{RzTa?v?I4bzP9E&rOYj?}ap;SWeQ1*7d>a=5px^J*3a7G^2
zsm&2!`3KxL*#O%ecn<E;7b8TjRw)(3lfrIS<bJFm?u*4<Iiohcyc2VW(i%(v1!Cbk
zZ5nbCm|n96>U3YloJaY>_4K^*@Y{yXGL3#(4u``gw`o>G#yHI&Oc$_?g2*f(iPFiM
zBy67Za{%gh2y1Wx{$S9<vY{?5;<G&g!Z?+%GH*L!@^*Osu6jyX_CX!{%iH5#-u+Zg
zcDTYmH%{}5C0*Qtm?je8U#x(PkO!uOta+C9eF}hL@KAzlwAn;wSS@=lErImBH!Hw!
zZTQ6Ok9HTu(lgYqSJsI_+E*ahyY~l3cd^5nrB1Zdv|fxYd6s%5?9_#M0p41X(W*s1
zIwCc8!{s{)?7?+1Ob|DF{9?*gq>#=LkHELv*xO+Qx=)uHR2?uQnRCZ}sYD!gU_08B
z-LB&6VKfoY<;p%63;bB<;#Il<E!zZi*4G%7^cwCq<=8YXFSUrii6|w<;2-mZoe_8$
zZ+(czoCBH&V1S?s+HC7mKLG?+F44H=sOds4ztWFvKV6G7To}0f+6zb(R!<dHhz#mP
z2|f1AmmUu@p>4yrN~B&z4ppR3(J`Y#;#z2h_gozluho=H_p7a|H4!RO7HcLnd0>J!
zae|CPvPr0&a|xYitKKCXKi5#dy~ZFxns56FT{fB;8Q+{K%5Vqfe}#;TG}K2s!Z>M-
zR&S%1%0IHR@tK(3tNJEsZtrijtJ3}u-1|P1fSag*uiRt-w_EbL6=)}HhG6ev)M)+V
zK=&m-Lzvd*nb(>9BV%vhhaQFM%Td&GgP-z-3D5WYnfQD)UHQDD#eEIM;{6ra<0oV3
z(uKIS@mFWLMgS-px`@%ED`4b%)@c1IdF1}|`}M*q`^S$}S(UKA^Rr3(wu*d$BRkmJ
zNHu_AGMct*X-C?x>D{_3GF73jvvdJ_akGN10CItR2G}G^ju-*1C%5|w5El#Gg2=mo
zo)hkygiY14K*M5i*&p2;xCA=!GKJa&oHS-KvyQca`74$FtBcm)TxJ=9jfkpJW~d>o
zll$!Y*9D=IX1<c?y-tSob7qs5;UupK_vJ7`v00RX_Di$b%TPYM@s=l_yk2snDqk%%
z8HdPpsI|f;;vBWhDX_%=mr2^wb(V0yx0J+%46_X4&of84-+ox<3^WV`qdY)>+px(@
zW?Y|CQ^S?JA9jX~NS<Zaq_!JKY?EWHb#KJQZvH?X2HTtx_K-awBwR8S8qgF$u_V5I
zankwtL8K>#lh|*n92oKPWN20d;ez|ddD;NpeZ1JE6+f&ni3ru04mF`Yf8M_HaS0>7
zne(;kC*g?u#NoaBNWC$>&=$~F#tib<2fx`lf+wbs{0f`C&W*_=oV<tA5~x_}9kzXt
zp(7G?T6$;H{I?qfOaYL!Qvye$Nwb2NhK8u4iWt%E&8NJXCJOvB$oz7Z##h=S>Zd-e
zUFlo*hTsedB2Um>EZ%dImS9T4x>d?)!Nmnue6EzKGHk8Qi-hdDh=a3)s}iTo7^1QE
z1G*ur<(=Ck(tcDgN;;;c`QxS+193kWHX(nRSho!{W&ip3$7=xH8u(8kkPf~84|F34
zV%UhNP`kDs`yZ^mJ(~9lQawJ0XFuBH3NnqVE)C_GeywWi8D}QU_DgA;qn??8g<`vL
zC9@F*8WjJ=xP|1(F`f4b!^b-L+smW#Y_@fE3aN)@=tD;A+3(0%&-pfh)&}dvB(h7>
zCDuN2XRzt>!lAbb!?PFPCes&}^6u7pp2FmMA5^95@0H+?T+1ACq1u!67n3Gr%qC|8
z^^TfdDDM4vNY8dp@0f&=ZSEs+p4UMt);9((1YUTS*rCPsbq!)2jk(UuVDYiK>=f;w
z0Hxp=rbUbmmQ*EJb~}tDw=63WmCUn8AG}oy9F*V^Su8)xHEZ?=M72rC5>%*n*iT<_
zbbsNsJ3qElv%NIlDr9LFWO**qA1ra>yN(JS@2=I$v7M8ZDGol}qf1Mca<_Nw7vCMu
zDWVV!xp?+4MBDE1S*{SwEoLf<6We`Ymkr|zIq6Mpf3U~<Pnz4OdDBOn-aSaTe#}n=
z=62K=pIi>3?SU^^Kc8PHFrQR=f7*29>+q^ob;37+9+fw_p=HTjZ{-Q5ud?{kd5yam
zn%P4%|3<rjx}@A6l)?l3l_lz0CX#C>r;J1{+-E}`#h{rnV8e(VKC}77hE_Fsq4(!y
zL%5REpmN67X_JI8dxM=B*S49ToeWCwnNo&Zb{omMks~Lefri^@pw5DC)Hqa*hq|?^
z)!$wha!7KU-v80JV`&SbQ*z4%k^#YI0MxqP5O*CKz4@~>f`ObUFCWkpf8V^6YEUV7
zhV;eYH7|FLkMf=&UNB8+_`+GPmT^v?=LjD{Ox&p9il`Nuc=B?<<5ao+@Z+!J?Yyr)
zqA&Tl660iJCDr@~`V2M5Z-S@rJPU?9C>QY~_mI16zqzgV)5kOq_s<`Qa9wZtyzovu
z`g8rGyW9o?#wQ?X;BF9snLEHKqj)c))h1@ZvgBq(`rqGt*cFgJv+Ug)-zsEYH*f23
zUEa7{%W*%axpG(S#mQ)B%a~ZdfdtVXRXI1zOmE5b=Uk6%udI%Kk*fdOU+D*8JX3`2
zhHRKU%=B7qzlrhSG%<$4=27O<cstkdo7?N(YCW4e|K%3~>(L$i&DCjN=tmh6%&hX0
z<*-PWNCidYeTfrF@#%|qRT|M_mQAM<&1%^1ShE%H->Ej_ng#8M1+~k1Rl>FLoeKi<
z%ac+?HcgXXc}b=-_IJ#ToUDzehSu4?2_Un;FpF%o`^mB{gCkctQ(wAK)WH(yF4Qs+
z<Wn`1gyd)4gIudn$%1CKE>@8hG2Wm=mn<u>1&IWW8^*}WF<fWnAbe`1ldjM3$1XFi
z=}7+>eAch&q%_d_05BINHH!X^Y|!^j{9MPt%g~tQ^Nq-IM1!UO#BAj9zD3>GT>(!!
zGn%t49@912tj&u*C&+zi%K8rFnBt0K+eoF0vpyl&aBi`KCX*<J27zBGJ0K54OnWLA
zg68gif-;We$3n+1l)5M?ES)u%+vbJ5joHb31it|vTChe6VGffYIypuO$mZhIQ>xB6
zdl*`rZYxgM!O@a`Qc);Itq<<99AqfLe8v)-15=Y=etxgqBzIJ6bz-~Ue__6vu}>Mh
zcCu^oyWCYSh>Nn)`{cjQ$6diiWwO_aV!I3l5!)EZu9PN8MQiObMScV7_kHr0F2p-m
zChTS7Sv+`a*U3X73qW}Z`m6Od2)m1>Y(euP&stoTYOGc@@JS9^Ii+ALlY^-(T+w#%
zxn>)#_$+Pi=P2)SieaqZ<g%cRasH3{QI`DeMZFffLdU205=<uw_R1KG7ul(>ujZWv
z0mcPZ)8?bM-iZ?<Ptl`8V;wS9KQ*3|d}2FZ(|e!8n>|asIe_1Tz8l+~41{G6L87wk
zC1vCrZ?C%34@$Z^SG)U(x@NfN#5vStzfx_v+$i305T65XxfoD{T_=au6S!`(*w9`k
z?npMI-Pq8VmlfZ?eHfWP!s?OiRPQ{zgD(ba%VyQ|hTjeZc~X;Ri<C)w&{ZeX1;`Cx
zhKHtN@182UW&7Ads#A<oz`E7vsl9^y&nqc$`^^pC1B+X3!3Dy&0C)t4noY(=vLq@g
z{OL71BjrztZ>xMDI>rX46I#x2k_y?caB}cjw-KQ5iBwq0Kx3vrJ@F}&%do`Xm3UI@
zVoll#clsIQ6`6;Kn6YaFyd8Zd0zxM6v)Xan4R%w4ZE5S1<2wTqTd{KC6Lr1V0Uy?_
zyMF#DAN|rsy}m2&6<3h<KJ-&l6m8ZpW<UlCJPfy-;f1sVFETDSr%ziL2)?<iuMo78
zn?p7$(3G3AS`}P>+&OUl&tjVoKYHVRlCjs{Md-kkJLhX3jgI>ds!i`44UY05*0s`1
z$Uvd@0}@*G2q+Ue_B8tJ_s|g+E??(M9q#Ge?e_9&Vz5F9WWv+2?60G2F)a9X!UKgV
zJVUpb+{ti=-Q;0buCg$!`Ynj;nC}o_YGfTZPZ*|;>ji+()@-0r2tXPjtWBj9IKHJI
zBQWBXlA}|p`!sc5e=l|$edm&J@X4{Irti;MYcmSnU(oi@`B-i6GAbSxZYV?BFFbIq
zGv(E|g~rcbTPgGKyc;&*96?SO6$N)R2cS+sc0;0Me@?*I;nzDKz{`--amHQ*ABZIZ
zg0n+xsBWC~)X#4=AhCS~H?`zfNm|XJKk4x6;4}>U>ojQt_NmOp8qmV<SW=EQFqug^
z308&-9We;+L8&t`D>kscjQf^5Nn@u17*o<t%7%=5fn#_G9}igJvvm9rYU0FlD>uQ7
zH<J%Vi{u<l%Ns#)9<UFH!exgySM9=N&cA1^>#c1ER@Mp93xE^+I?VpuHw!Q>h7R+A
zFA=FpH)f6adC&T(FFw1BP&q9j&UoOJai8zLSW{+5=1!It%@M_kL6Dhw;ks@golw5w
zcVgJKfmV;HxNS!FR2!8!Yv#f2_JIFt=R7ta{YpOsHMY&1zPk@^@SxM^c#s2b<P;r0
zIotSaX44Lg4p%T%a!wR0-Cfl2y|OIqc<x6?P5i1s6y)2I#eqXgHEAmBK+hd4O%8%<
zyVCecTK%MZfcLHN0|VPLcxwV$6;F8bD&`j&eS2P6$vNSPJV?`I?ZDARX`1I_tr6@p
zH)kX`;-jsoGfN`%AJ4O&+#Mf~W7>J2$@U`9@GfNlOa?iMIZ64&UO-nh?1hJ8WMpy<
zpBSwe{T@4{Q1Ybn)@zZR?&HU;wAJ42JbaH>y&R2X!wRvwWko8;h!zbfOORjmGewWP
zh}GleRylugIvg2LSaPW0)Rfz`+Ucl3Gkmv^^!h{0-Ym3GL$!VoX0JO{My!$Qvl$Rw
zx1RE;vLfn)ASym}<fe*LYNTvVV!NMpD{LrxB89a3dJce_Bn^J89+A3G#nK`C$Uc%m
z-ZRyYUT(U3N`(U!AH--ol1@Ne>T})tH!s|mzO%PDLg1GCt?CWcnr$*eiXIcQhnA6H
zgpz>t)<rk^U5p7cp<>7m#)qEKeoeb+8{SlRX2YY;wz|ydYqJ7QxER89iuoB2GHtiz
zAvoB7BGDF(mv2lp$ZC@fpUbck-`E~c_-1r2_({Y&wz%r-sLZ647+y}uCeoTB7RoS`
zKTOSYe;K~+cVosm6kDG>QX55*es)iUww-R3xt%o%BLhDiisNEz>mMF_Flpf&dAXOw
zop4BFBK6I(ywJXj@j|<@w;83GN)OBr0NKjjG#Bc7vS|3kxVU?jokmB9FLe;5{vx{S
z)HT<1yF>bz=$eCk+;Vqs<}Ieu1sDp<cwN#7hUog?)-b&!n`LkJl}A1yJHJ2LbK+uz
z#<=#mg!Z7ekqatZgYm7>+J)y-jvImAa0q)KL<WM_q>WO}AEzmhQ<{+?wZ}ed80cTs
z>oWDIyZ6B(2!eaBo2!8I5t;X1=_HxcwT7_0oBEGxWr#ST%X2K@uigiL%2*)Eqg*bm
z6bp(yDb~)~zNv<3lgihCU;s?QkxJ8i^Ml1ru_hT)AO;yKG`zohS8cHSQ<2o^gj;uv
z-@SFg9SfIYe}o^XexCc|zEAi|JJ;%?<vy20ceSEW9IAo`pNAilmyZ#-Xav+RM6mz+
z`SbtY==1*{ybj;$`p4Ys&(5HKx=u6p-<zPGzk2e=p6b^heC6oTQMe}sB383ZZ#?rJ
zd{bkN{c(Rqz_FgQXpQfKf%`c7j+(tK#M4gd;Wm~*KcFV<LI!}N;tDfQur7ueNx{i+
zK-zD0d<!CtBxeIxj%4`*+!iDb(I+4Pv2QR2MVL+;!xPsK`<pMhq$kW)2Anb{ehYGR
z3qtvo>B1-F`nDjm1lB>qq)qQxpf0H>NnijFzg3I@;P!{X0AK~OtTf^lB*f|5pFjOy
zzTgA4^P5rC-fJrsDWP8meehB-(3_wC?!p>2vWz8DzG^bL`(^&TfB#&oc6ed)-4#CO
z7NnXmd09UUwX|r(?yl7Ez3juGmZK-`>{uU|S1Y5SHfvA?xKohEzq|9W-&%6vO%-;R
z-&nIS?l*>9@+ylTMAm=v2p}=}H-_BuBJcq2H^yxC8$(u;jDQUN`O6`H{Ayo%8C`f&
z|9;e22a9)7KBj##!Wj+kRENpyjF1Vy9mCBj1+o+Qk;=ev{>4=)|Mu!j|I~f_Ki2VL
zPS9Pj-7xVw;hR{|Dl%6XyfnM_d5!x6s=e)cxuy7d`dp5JRP1Rm<6_a1ZjN!Iy`GZs
zKScpWj|z=jg`%#Bi^*}MxX%U88hM236*%G@Z9$GF2zI;9Ao#)<5X3ZjRE(R*7+*R2
zO9JRh#a~pYQ+rznSff<K?v+tDUsPyi>-tM*@Ji?;Z`;_;%>F})%sfTEWl|d9=)20E
zfO0sJ@b6m&Q;Ws>f2rlHX;xjvH5{dqwjhKS-z#8Pq;^cwm*0AYD#FZ#El42X1pI9J
zbMyY^3Ryff!}RVNt^P|O_O)Lx+H~MtGWVY0$SU@m8Tn98hkWMb&!X5}xQ~uFfVKvy
zu>Ce#`_sYy(`;%Yz3BU_r440$MD25jJ6sN%)|u8GlklI`@QbW+wnt-fef0_e*#aT=
z-rwF~TmJRf{-=M&-2dg6`@c2Gxj*_iR_~eM$Ge@9C5AasX=NAImN&SD!H60~AZmQV
z;V(sH=<f|6%1?nW6yX-0gB=WYX+F1ea{;+Ihg-An;^HELfvlg_kl$@C*8J`1g+JO`
z!uIa!7DO)E;)fXPq5IA%b@;Tv>i~IhaM4%00GfREk385Dm+`&UF2wPlKI4~INiE9b
z$X9PVn+fxzx+10ud&D0FDT?q=;NGA#F;<YfXk{%(OLwE;6UfaSS$s~tU#&7;Cj5th
z5gcd_zd)`?kj<F><G8sy{a`ZU2<sFS6s8tM37bdN{8>DEz;|YH;0E5WvqJd+gQ)%}
zVLFSJ$*KojmF%yCy|1?*d&JstKa@8$9)Njre<36=4LkeoR9+A+!8_u9=hWD%wjinJ
zS?pVovH^k11jgq~GK`t=_K(lay8CZEy_yV`yxHFx2A|f1GeGIYSePyH58gPo%T*7z
zAbk!@>&*e7|L1uZm*_Z}=~g~NZ(XqbLp%tmO%Guj2Ckt4fL)D6?!Nvpz5)f&*&B19
zs~_?1FG^a!t3>su8T=!j^@q)%$Wn4x>iw~U`^4^Q-HM9u5*iNNKZoxGCG?;~w4M!~
zp=KW(9J|4TqoyHfe(+%2T$%%Jq6biOySM=5?|&C)J078*W_8~;BXI>}Il382CpI=7
zk0I*Y?yKvWN&URbc<`yu;(CYs5vn_FBNqdqC>7!7tJ!G}^o|(cDlzk-I!OqsEsa*J
z_K4#@-^6`7v5MR@XGLDYO@AVgEw&(tvh`~b3;=P3Cerj-;km6WemLC{Mqa^z6X@1`
z3yd`nVEXvJ*wq~7t`{Y>dq6=@77cB5Jkqnc^C>J$Jz;r~wEJ=Ej8}Obr{}^0d)>cx
znrwiT*al3T4G88gguNMOT}Oaes?*`Gg*airn*6Um4*i{Nio_7m0q#ysuobb)UoIgl
z0@D0DKI1>7fb?dh0XP2!TflS=|Jzen_DEDMwtUztVCv_~Kjgr<3f=yVZT|j_^*e%B
zf12w5Ia3`WFEc`(Z|Z}MPO$Qq=<RM-Y&2LcX%Ln@*rM#VAm`8ZCTu~9aIas@Z5ypJ
zW$e9;o!k6se-G7=Ji8QIggt+rbxJc0a1jw}_F$bUSpp+RZU#J_dlA93*@7HK%#}@O
zGM+9sBI(<}s#pEV1Jo4)2`eZ97_q{OViffL!`b|~kpUvae`+K99T~Adt>Yio6XcI>
zWPe)6KXsV*J5B3<TF0N(@n1`b{>LUkzeA}0(>ngNjz6tK_?!{!_b4xaTF3ub>&UCB
zcw6lyQ!OFR7k5;3mU;l>=7#M?5|1xs{>yW-e|=3x`ES758no2m7a^63P~RBwtRjrJ
z-$f1%=P~GdkQ?XhaDRPjl`-b-W>D}@h1S6jG5G^{89UJ8EV_jbW*q}nnGd)%0h@*O
z(EdsTW00)*fqoD8CU+JefCc|UNRsE@`+ricGAn;}p3wg`?Qc&54_5vH;PUjx;C3GW
zWhCMka&f;U9e}-b<mPndCU-#+;tZ&q2H=@@5loJYpCKxsoA-CQNPj}n|A5E*$EnMO
zNU1M!G1~8*Ot5`Rd(Vso%eJ$Ob%E{a2{)Y6JZ^Py3z7sI+{|Rq;oprU$}go*4FPmz
z*tv1;H}7?{rIbzKf(_YdNK=Yf#9GbhIq%3+kN8`|S{Jh~523UlJIN+368`0DkeP6@
z)j_2y+SLhk-mw@xiwtP3E}IKWR)dUWg~)iZ=%5*V>M^ddxsDg0``K@B<38hzy9B9M
zS<oRE4P8}_+Z6Whq%Q$5AO`S7pnzSx1vwV;4h&Kn!BIXlZodeGJL4va<A75=cWwZ_
zT!UgAa;yhT?q)E;C6cI@@&Kbh$vlrBR7dz(xxaka#G##X#pj50oYq17hk3d+YX^M`
zqOA8E6hAV+KRL@rFfj{O1?y$@>wkqby0+Kcj;2|CV|M?{shk@R(dZn7*)d+`F&-oJ
zz$Y^2D!Evvp8}>SVU51G`-I453F}^Fx623QhW_{?0(E;8>*Bn~cYxpFrpK|8KowMU
ze8c+H-)b0KdQ<d*r?3#*F3npn{rw|xJ8zbnDc(BQ{dMSbX>mP=K%AM}z$z%N@!At(
znL7`6viSDlXy0+i;DWg6c#J7%?{onax3AXtnYzFufW2WiE{vP=?8lL>dVz`@pumL`
zQOsDOGA>}iff$@|_brbKJK-N*js3gNsR3&IKYp9q=8=kI^YY7rZqGg`rg>-UXn(qP
zjUxeJ<-QSu|1s9cGLjTy1_J`$npGD^`Jd3_pT)F)3<LfnWclCsU_8|>jE5w`!V3Ts
zGI(j<fplw3It5SU%*>-T+E!LnQg#-A-Q<;nqGwCWUDX$=Z+Y?|5Vk<I^`6sULrJe<
z-b2nJJA7NmH};K#y1}AzKYXp!4)O(9LNmpS6&e7nh0_|?M=Z?&DF&bO9e|Us|FR&p
z4uG(k{n5A8<<q_txy#BiBMqVR!#~$Hlh%eS@iPc7HA^*8hWF);>5>l+z6?k+Mh1yu
zfW0qKnx?-6x#|JT^^|fJH0;*~`0JrYLei9%8PEDdw;+!9Ls^OMF&B}5L|YHXP0y6W
zH&=V%Yw2_B#2>hRLT|6PB4GhJfAcV;%A=dWW1m@VSe`s8Mb3Rk&t5=2?gc6Yku!v~
zM5gEm&?KISn@C}0UO)WTa$(}8It_SZB|3S4^#S<j&+m`b68;XB{mT*kI;v?Fs2P8f
z|0O(ptpg9-A^v}{ckO{trTu@TQc2T=gqUtpBwa}5a#SnHN-8O0x(QjS5HTE*l-m@Q
zQW>QpVv<{;8FEV!av7Hx*W5EB#+b|Td$g}hcz4_V?YsNFyYt8HIi@o+XP)ys-{*Vz
zd_G@nz6-n)Gl6Xyb}Nv39Z)0ReYkuBuxgs}@LCbBN;&+7e=F$XWlcc5jZim#2eVl~
zp$21AP8Z;KHh;wvwJeFHjB$oE7W^KQH0Yy<R5|K#4GK!(=rc;gr>Ha5gYbr3Uuzq~
zBQo^7zUSEMr-E!@kb85Xo;ilHA*21E>w{$=zKQza@Is66Hs$;524YJBy<Kz9G4Vie
zdwDGAvKv6@IrbQSv#Sbusqlr^tkTfM39`8idL*y2!mtj?qs~i%0PkyC9~UxR1^Lmo
zmdf=^AaGZNkQNb>5E(evA148(ZDJhwuz7_7nS+z%>x5x|u2Ox-E~{+=D54tppLPvN
zmY4w763E@QMi}n-HVkW54<Fs`=Vcmz?IrdqNurF09SofzpJU>(<MVsM`q(H;vo?1=
z4){KSE=(%`Ei{IkOz)dwX`+a3UmcO_ZVg;IC(mK8{46|vS&|>2uRz^9NK4`JxF|q^
zm7{#Yj`Wa%cQG*Cs6ujw1TaSrZjx_#k6FAZ9bS|&6s7RWkK-beiS~NlTIH?Pp$x-t
z8sX2*DDaA*ytO(JoQJos4!h|?Yjo)Nm#tM^X_mKEf9qPcrOkTV3nX%=m(t}Mf3mol
z<0*4#_m<X=3-Kp=p^naNh>O_(iX_2VG3kbg16u`<;|T}hu1kY_(zpb4e{T@Nw4sI0
z*k?c;eKGRB1ZpnBvg^pN-8R~c^P90B$eHar5$dyXn&;~rf!wa4nAsxp4q`hte*~mA
zHrPdu%CQ`tffU75PvF5<L0sW@KKY@j<2M{<VN=x;V(ZcUf*`5}SiwTgW5lr$vD;dU
ze`OQCCpsN{ja(v)z=|S70+?65)yO%jzN!loz2WXxB<@(?;`jW`*DmhB@#uux1K^e=
zF$QX*;^D7R9{)sIUc3mH7Xj-H@6;@ay_K_vd5y-dGf(ru@;lzyU3eswb|dQDO~Z8>
z&)4P5o^oMM0AGVrH_TBac7kYGQ`7`h$NI_4$UdmD3egS+r4Nw5av>w`LO2x-DL_jX
zl4Eayi!>=PP2C2O#WO9jm4b-(>J&hg`~uSr>Vc9AK+tFGL<+Gz`u0=~`py8#I}8yw
zWLyA&vQG@Cwc)5714mC`K9&Mo5qFSzkIui1ne*6@nzi|6ilt4h$x(~?SIGf(+8d_S
z+|NbTUL$P7-V<REJ3!7+3KEAqwtxhs2+=!Y*!Yi0uv5Ow!4|*r;<za)u%`lYY|Eq2
z>Az=r|7}4l6DU#eEC<hWj9897(N2&dTUN|~K@m+Ita^=0BuU~6vXxE$astTcaj!52
zP^`&Hw9eranoATL2&(4bvkC0k_IU0TtNKW`A!E}33J<z{G8VOCSTiVDBfKZWo~|RH
zxGf}zq$k3xl~(<i$1GxAm25QStZ2M;3~$BqRxEGD4*LKTB~*ubHh%%-<^ylVPV`8|
zzV^?}dH*$gEbDbyW6S0Rvq^@#;su&i=ge&sM_G#;>j=Vr$r?a)1!`5SCO7*Gps35;
z<cjvI=h_dTRIy0EqDw$K;z;IpTTFU2>S|pptBZRI=E?du?GWY@`p%gA?;4L6^H%)W
z)&0%i$z(~#_H;;)Bcbe%b9PXAd|N2<Tvq7icU4Vnvg{XAz<U!)W-Gvj4gk=Wjw9D;
zVD$B|t3fb%$~BMLBUu8|unJ7@Z{x1oPw^r*DhO}@IAe=aH-@vJx#JZ$t=0l8SiqC>
zG>+v0VB84_iH9$BZFV2MUEA8-a1Fl`#UlI?0*3pPCevxfR%24KjkG**N>8zRk!?ZK
zg<fi#EwlKbtZ~4XuM6u9W8lUV9QdZ#KtR;{kZZx8JCq)h>l{E`p>`i50LE*F{8Lav
z{-KvXWo*6NO?}bOFpVvP_tFhqK<Rlh7u2zXfhKQz0Z7Mva0o^FJo;l;g%f%7E?I12
zI9GA7U2}&X2;~bG6TK{kwpsp8Y<V-H>Z%CPdLB>V7CEelVI?`XfaA!q1VbPIhM;gi
zNp^&FPTo!aiHavLKo;VR<7v(A90P~2Rzi8%yz{fKSA5#&lK*Apri~)~&e$_>x&~;B
zl>!jI0=p1B4XK0RK|8Q_!r*sDAZLJ-iZGmEf_xzKKyA=nn4^iS(?q#oe%D1JEIIZx
z&LL*tpt`0vb#^b??{urzPPQ^3sn7IG@|O*r{9QjC;>yuakqyb9e_Qt?W_WKAe6Jei
ziU;76?@8aGn%Eib#!`zw`Z9F@1tbI}@8A5b5Woul`M(3qGimTWDK^tVZ|!6ju{UP<
z$a&~2c@mvfY?=R6TI4G><2x<#(6v&r;qV)E3X&{?T##+r`>hDTS7bCaz^mmCt%Zq>
z{7smuFhmz{eUXg$5L?N00p+tpXAuC6<$_T9h!2JvOrANinpzYp^pIW3b$wjVP$c0U
z&IQ2FI_VfBTld%iN=G_|z}O3xz8@33e#8Qbw#D#_%`X`nQazweyAJ}HKnJjAv+i+1
zzGf|IL*Z_dXsClf90s^Suc1_+T(2`SJq#ov6ue;*(f(>mlaMPBPFI8*r_l8V;gl*b
zQCz7Z+x#m>zPzz?f~WRlEx_)nzo&fhb#b1nnKytku<gaRAJQE;*H~ygC>&6HXRy@p
zyO;}V$G#yIgpXeTUsF5q01pr_(Lx`3jkn<cy=K9G-n-x}_<svZ7y>7ZIey22=b;LQ
z7WYI)-h$^X`13y#2LCQQJdeXZexVbZ$bEWETA&4IHFN{D&Kv&uOv-owWryrTy<Lg?
ze0JW3TMQa%=2DZiICrE$W$e-_m~oP`miQ6$tbZuNy{+(k?Eq@t&yN3a!GfRX`2V3<
z2A&`AAMgWsD1mRx!jmN+=SI-iFS!Os>j4^W9=S4(5z7r#m*BY24KJjM))PBAMRl##
z+So88RwkZLJPmpQs6JD<n-bfEKQwENTr-h*idY$rczJ**wOVBWbuCh%DiGOqCkNA2
z4!2Hug}bAH0ahi$-nA6!QN?iPw^BKNS{!A=PLKuImJX7lr|u1)-jsF&{!=e`-0LOq
zx9`L*0}>1mO=X!0bb$uZ4&{1<xj3y-IH6YpD{5S+sc>xyseFEo*rZPPrF$=(+q6`*
zH(Tz@IZ`!fA`7D8EG$DM1JKqId+8!0MgA%5l_0#%<^Z<Pvv^EdG9FLvHX$tq*}&YV
zaO?4eV&pB{eTIxCcgPF$#vJ?KV-5WYRhf@i+HR<Sbs2%dS4<A1WF+*7a)i-{PC_`>
z&91AlhmY1dl(!q3kKo)VmUc0h<2i1mm4mcR2MtVE$#;^8|E}7GJZQu(nyC!T7#o8<
z==zb+ZCA5#0Obl6_zwB@#>!!iCfgx{&<#j{B5-*Z8VRVPWEwEKyr3owA`8;{H5jgR
z67h1x&=>ODm_N2y@;ls^`yhV_4d*Xi*lby5Lnc)S0KCcmt{Z6f<*{Xo1szk9Ze<5t
zTz5!E0v8~+_-*V$kN!e|j;ZKJ3)F8guYej#D|-)ISdBphpTf*646TNcAj1^`Y3tgT
zIRhx)-2$C8f55eNAwt|_$THeEizRgfu7A;(_==Qx$p9&MZc>yh8#VJXpO`?zCLnNe
zQqB<UgIA+-*ctsZ-i{QL#?(vxp2TJzaAe55d_9hJyLk5Of2a8S8|_&q2Gnl4Ph#RR
zY?34>bmtI#qQ<=jir*1K!Vj@bkck1#=Ows9#%xgW{Y6`cs}%eO^e82;r&qu&J)!7U
zk5Q&1+QbIgnuO^>H_FNW+j_~uv6|00;|L_Eqeb6Y1|C~39&QeikVzTo6OolPC=#`4
zuzWjQ9Q<Y#{7UtSzfAF)Mg@^hqqrkq1$6{?f(gUf7oJ`D9|x82?8-O@%3o(!cz}>^
zEH%$x`QHVr@$AZfkzL{8PIz`@Pz&mB0D}Ds?8;|ke&wC6E2KHkg`R!1nm^}r_8C&$
zRM-fG+(;nz&-e(ZLRX(g)~8z0Q);~~sqT%iP?w?SUkZve+IX<@5tXz-b^*Ve)>d@G
zAu>nNyrT(lgO>uZ^cw<47v<C`(uk!M5#oZl!hY)qP)lICBAl_^^*y#{k75A_<dZTB
z1!9U!H9=xYjJ*ji+EEO3%YZ7V)<ho_SO*I`D&#m#CM~A{G%L}PcIb}Q&5x2%>mDR;
zHD2gx4_%HvIh>#z#8&@pdAU5~4bSB9Ox|~;em<zrf^$u15ZBADfGg{tTJvgYZ8PTv
zl^>KM9e|<w5YfV43jIXv-$ayxDIc*cJrP_Ec9&}fw%gWJ6-W=hAoqmG_WKWKfuNJ1
z6EWO<>SZ(6PZ`6~w8UCcvIsOis9)K{7RkPY=PrpS#<z@~pnqS?FAotk`Z=R`F&!_a
z8=aRPorC5XqW>iMfQLci86uuX#Pf)L7G54<KK<z|sDCKRD8@NUY7^74*15%wDDIfI
zvyd3>xq5z0chlmlnT?*OpKeK)-R`m_<`NkYU2DTY_4ZJY4O?^-xyL0R<$4FS^=W4U
zEQ+QA02(nRMFFlzi?f>SM(x?dfq-CG5fBXHGy;@j5qZqZ``~f!#J&fxN3O&KpJ`%q
zN%deYv{2JU#+|eGG1ILz((iQQ`lp}PE?%r`EhCY5*E~iDUT@F34Y<^W?Le;lRXmwH
ztKJRzafqrdyMtU7Ub4@A8Mb|I+X{YK4x<scJdf!)YzBt4#Rj{aQie^E;oMX9hp0f3
z@PnXJ{|MnueFDg&OG;fqKVXG>lRF%|fXKOZd%f&%5`!i9E4)staPD<LhHh}hz~x5h
z_Cq;in=^Fp9OCt`;d%$ot>$zxThIz(>8s(k{I_&V_^dEqLJoVzwT#-W3z-j~*z*8A
zzGs?M-6-_gIKe1`UzB}&K^{%*BzjqWR$#8caDuG#*r--&ZVP{1<?t|R)xw-$uTBA^
z%M0#_ZLe=*fBOn(5YaYtaSz^u&~1oA1J-Rd$ZzcoBPtVjFgJl_@(yW00$;Ki>WSQp
z>eZ`~13LCOgx)?;BrprRmQqYeH00c+;Gy=<3S1d)(y?La61tMWlp%2L#3TcOecc@7
znvr3pJ>8r$uK>}d90Jg~#|;3A7cJMhd3ciitutN!P>>Vx4a5X}hd4oM5=d5_0vBV8
zCEn;405TOo$RMCO3@%R|#Iby~;QxXQf!)Y^q1s?cR~eX+7&L&o|Ix7No+}_Hrz;cf
zqL=D2qodnjdXj>|jf`8>?ej}~gx5}WzPmnUcI3LX0Z!L6Fpql-fh>lgb1lpjE)ari
z%c=%YK4HlGJ}{G6D?>orrSxc^@*t`qSt{}x7}pvmxE0Oy8Ca$gXmL{C0^~DZ(ICXz
z{1TaFWesrQg#z|_G?4`L2QM&8<H3AO%VS3?&><#~$}u?*&m_`}+CEEIVXsIn47N1U
zU#h#S`gVo?qzH6?cI^xl#~dI48(UM)AY3oxsxc)up*N^A7c`pEZNT?+Ks^^~NV+47
zCJ6`D92hUUA=jS<<TK=7;WV+Qn6`*MxHBSKvJO~cV<)Efa1uxT!Y|v+)&Z0m)W4DH
zMP<zQU{Ti6`w*R^OqeaEgy*ih>82VmH0UHM6F}{kZ=%^CuPY0X>w4JPhG5|4ksEW=
z!G>0Dd=Px~6<{-lc^4CO%=)U0LDQr5Z;D1Esl8~fj1!9a^5=#Y+6pWRE7;#nzG|ok
zxb#q4a7vRB)cByE$OR2#G@n7;$Ei%|0aSR!F7Fk@Qv8>PY##=AG69xsu!WWrY(Ss!
zZP-~L#g&l63Dm0q$^)5fm^(?Fd2VD_O1b`ULr#Mp_wwc(<U+!rtjIXEzG1^7_pKFe
zud8JHATz)um98vC(l$GEgI{rOc%X^~=gac@>@5SRlsYPRl@s$NY20hzXAiw$T%N8y
zzkN&$?WVSu0k{)^!SrUXOBUcD&<)?&_cV9j7;H$^P{3mQfDN^tcxyNr$Il2yrZ<Dz
zXpe!*i_d^}?7zXdhXc=O@)z(q_7teY@&$Ncay!|(h6iQv$?>3Gg#rTgQ2&hF9&FPA
zgOuUZ5uwJh0JdNGn=knYKqUL=w2C4Zt!5>yf*R*MaRJPkLcb%0e7~b3%HVtW1G|VN
zhUWFu-gDIb$c=0nDmEgfAdB4ANs8vK;P)N^$Ne7bN~+}E1)f$FY(ZoTw2)bwKar$~
ziDYxAe*?8r4oPPzApXgr=qnaOn8O$*3^0eyKoGk_8h9~b0m<U+B(MbD^w*Ku)4`N$
zX$(c*hfeABZ?;(olfh^_oZs7R?PA703+EO)HgMT!?xG}O;4sdQzmqx(ehYC7fM<{j
zv2<*(a}29U3kOiwVP+JP;|TZMtRXEKF^Zw&Uoc@W;a?OK2}Lg(9;|UqdMKG93{bl3
z>R(&n0W;>8!95SaO206)a7m?rw&V)fKlv7LZ*v6PszS*E?%CL8K5pu(gSb{)U9CMo
zcMA*^JU7%{OM9r}Rzkn7UjGUb*fq4FqMy*5+sepvxRtT@fwx@mF;1&`JvW-%80-q<
zx*0OS>bA`ON|KpCl5}KL65wM5CY@Of)QV@t0+Sd$*aHm7f#?nNa3oXm7?6W9Vn&WE
z{;?*Fd2W1>)=gb<c(meOfdR~oa4m#CGaZ07ixg=!aolU*w-3KzoWJ<RX`EXbir&Y4
zPHy~^dVT-}1dv))4ZcXaY+{?ix%rM?kii@J;28C3#3kfD0gnJ&oBOhzz)<`Mw~9P+
z3CGTd(?LO_u^P046h8&^cz+-j50A?v@bvzkF!_Hd?2KpQZt`s0-`d9A;<_gk%yNVo
z)@5C(`q2kk_M2xpOk1rac)9cb>YuEqX>`w4o;{s^Q!S;8K#Nxwf{S+2a*$c_p!KUO
z1RVVC!@)t^>%e0!{>gxT4+U$Fy$UntUGN6G-2y$gL6EeqmckV#&V+00DL=Q>9rKFe
z7jsP%tP^bL?<(3}=|{rfXZB@O`f;m%x>QrOH!0W>?AkMyY&xNyW~jaD#7fiUjrV2G
zbY{8oyXKO+fm;d^3LPsk$c0ii7g%vjAn2{T9kJt^^sO+QoTMv_RhDxZYE*zYVAt6q
z%;)nvJtsbsq>BTn(nDNB7G>VZd7;4kuYD;#4K}F$X0I6BFTUqL^nq{1;hwZoyU_^)
zs2FXq_j)IPP1j=nsnStA8_l!Pe``OKy<QK5)ssue9rqQGJ4cDW?H&0L^5eWc1E}X0
z$t-Jc()wX5g1$e1I!^5eY-2Cr&I{v#Jv+%oQbAvUTp(x~25;u@aRcGT1_Z=V<$ECR
zl&C(|7X*?4w#1LqGHJ?K5xu{OjOITS>c=yUJk!WCjsJ48c&71NC-Q4{2KuM=HRG8^
zo@wNnMxJT>CGqutC>)Y!8hNIXXBv5?@s~{FKNO(JGmSjc$TN*R)A&oKkq5T?Zm=Z}
z4g`dK6XOO@kLK~<Kx29d{nx>Pcx9NMX+cA3$zzV+)6MK3+Ov`8HS)Yhp4a%@UgI?G
zHTj1xxeqyf!UsOSyzJ_-)UTSXVb%AUR_+qxQh6g-vhyls&V^Y{b3%kr`^Kq-noTWs
z7hvzn89+?}I8^b%(wLs8{H?~9?RDhwNn**qcDvOq_+`BVH^ba1t6<u4h#od_8al1z
z$@*{rpdNmIu<_o7b4-@S=FhB{M&D?!;;8Dsxp0cn5P9i)kDaM9NMv&xx$H&o@TuFz
z#6d_<xbmPJ>9Fd*@OLWA4_@b#4R{la*jQF0fi*4ahTb>r<-Sr3{^lJd-&R!SIjlVJ
zY=cn8f!ZmOyL9bda?1!2%a&BK%CF1cediIQ*20f}5X}u%=;r{$6uA{qbQyKdx5qn)
zVL@*0rXYay!7LheUN~hZ#{(dM0fr*iwuszu==lJus85NF9(4_+p@1Xod5&XWL|N33
zadVr$76*5zP@uVYXmfN)U~dBNPeh8a40I9PW1o-H!CJe%hr6xG@rGQF9xPYVRs%D4
zNZ{7c1Tb$9t%5<;%g%xUl$P{aB?=A#-7CIiz>K-TRV>QYz-)CLrzy|M8_S^_CIASG
z$F~zF;`$74@0cyFztmU44&Y7*oVzv|U2MTET&r75HXy51)k9q0;2h+pj)s)>5Pog~
zZY1&Y1bupY_~=7TQKwAQf|uwW)rn=wbaC#M0a%NvlNS>B3WYQInbh;}X8^p${LYuk
zbvm|vT4e!MVXneSke!QFcQBz_A4fp~;0NWH{&hB-xd5ox5w0n0*p}1(VInzsKx!P%
zqZ}ET@gR57IJVF5N*4LH)kH6?bC6#%1(3J`IxPoKhb2||)lfK<;nA|-paIZlH#Zct
zL85gbH`Q<^*m19c-#zpO(`z(aWM`K<z&WZshFka9f9h?;2tbW7Obn9o(j2*=r6KiX
z9KM}pJC5#SHzY*J$2dU%#GdI-;EJtS?uIO^7^NZ@=G<ET2vE7E!#!0ZIP;Og&~wcO
z`aU7x_Yj|udxDL+ZG17^`%?jJ?MS{3AaWij9_0Fx8=F-H0N{Qv+^U$+NL>ImMmts^
z8Norw^;``p`=Q^0B@tq&oZxVGhzi=VkeoPzM5<nS7oxj2jnLIbU}4ozE^~%1{@&vV
z29t-!{D1R;rYieDjT-bIu3G|zJG+$g-moX9zMy}$su`Hd@>&C^AC%zr*N0{j%J3tU
z6wbL`AOM#W9>FcPl#J3obZ09t73<Tf44GCeYa!^p!ZVa#X<9r(Il3ng&rtre4CUVp
zoyjwlL+tSpSApkIhIb60KoY=(jJONoR7OUI=TY)J$`zsC$ME?Z2t^7M5aRKj3gO6m
zeWuYS+#=PT$P(|it=Y_qe(VX0>@|z;oa<U&wV*3R-4t*bZdt-TJAn>|0Ej8x*y|uO
zO!`P-K$-(&Xzo{0njWECpTZgM6v-@U5N~(@orcEEP~9hSEO#ylSkkNDZb{d@MH-kW
z4DE}M0&)c~CTcs${j>1Q^OwM<{_q5*ouhKRu-AJ)&n=q_bR{dVf$N3mNZ8a~tILj~
zJ&uAI?h$uhy;j<h`s&iOSc0-jlGLr7O&D(jMpDN%z)aW6EMU)0Aacd5vBJXvHy`#i
zQbXw8s|u9Fj=J6e_ga8OkM)NO1u}+u35qZ%9nB!@s$y%x<&Y+b@hM7l`SGa(6i#O-
z@-cD97(n%7RFy%jB?oaA-grY9mpMSWOdrF7w`w2P0&^(t#!Mje!D~(lrrok)1`HH~
z)Tsj~jY}j+pk!-h)4!YmGBMmMfWXh1tVHV^j`$}K`e2V!SsGD$F#HY8DP6*VE{yIr
zhQ|QPA%o01&`Y-;`w~c_*bgP1oV3yl`sGL<@_zOikS!nSQ|?e6;s9eZAf*;&&uS&G
z3?xxb;p0?3+FJ|8=qwFSYk_g80PG2n#u&o@9ZFxisIkT1-1m-Xg!C9N>n*KvNo(L-
zBYG0oOG$ydGT@atV;#tL+@)c-F1kRr;R7WQcNpAWLW2!^)=4aDuQcj>>CmrnWx3|S
z-6umP51`<E;M4lsX{*VHTro(|Icyxc!-p1vv;#RETEGD6lx}pkXy@aCw`;KPY_AEY
zB%R7$v(~?dk0q+<Y6EBF^}sA?Y`-zY@zUa`l0n$|z+eD%13b<*c6_XDzvmj$3O3|g
z5v-L1Uztxui>6`APkgGWhvNrOmUr%vww}rfVmQWy+^)_-xmXRY-(NcZwONb50q9lW
zlLJRMr3(k<7tvY|H1ka&=I!Ky9pXZeOV=fAZ|4)Ba&fM=-*UtYvvSRZx)nJ<$~soi
z2pHg-rF31FX!dL{EsE*ukxY4@p0VB7FiVk_-?$8jh3l*^tb_8rMDS>}6J8?tk4Xeu
zaKHi#RAkNpNn=x>W%Qx>YhywmZ1)MmB&fYT6U#a@dZ%+q0>V@M;4J~gPzNRgZVANB
zHLQYZdW8N>^6P=7kQZ`Ak&nIYmrU-^zx}nQ_iq5I<(b}ZbWBHlCteMXSA!eTnEEfR
z!Tobh@81Bl%QL+^)60WD{I9|vezWNX`M=3MNx^XC)~tGy*OD8!gln&Fy>9+;Sgx*(
zjYp<4cQ9L>v?g1K<B{S}bd^Gf5EgL(;=WA)69+w0g8TahP&q*ytNsfEsEa;WAS88<
z!bM<u7&2ELMlLvD`+!fK?FNNY|NM78ZjX|Sl9_IIN9>Zu8&y}4{V%W=&*mzpqc6q%
z_(>a+5o~$lSb%oYj{$mnO-hH8Z$p`<!J|l(qaN3wpcIZiqcnVqI%D03FT1nLCe<~5
zL<W6Gl>U2;y?!dl76!RD7wVZ~C>t^weD@Xk!1q7d_uI%wGu6<@cg-&4rJvut-@H;S
zYD1zTx?d28fu@mr=4oK=8F_JpkEJ=eaPvKIYN@u?a+wEC&R8Go-OR<}4D^`iW!&JF
zdo>`to&__jUH35Y$acxS2nnR=3-gZ_3bcoFR0GqXo?Trq?I@XZ3gq9veTODU(#H^J
z-Fl!zR;2O@<nFz}RDB86HHPqk44s2B$nJ{_pc?G^MKw{bS6ypJYvP&ibr=uR4bS4N
z)u~3sHlG8ksqZW<`kgv=Gwr?F8SEXP6W<Bad&|}{96=Iv3gt@@(QA|ptWCKWvYl{+
zz*uaBWv#0utt2Lpk7D}8U6;e<p3!hi-bv=;Q39r3j2@7jyj|PHp4UU<&T?Wpk?Odw
zpvK3JDo7?*4kVdE+|;Fp^+?BR+7ho=brLIp{CpAv{IX}RK0wPh#)7|o7)Lr*)i!4P
z%PudFTvc}jn%Mhc)}|HwEqYaf$gVp%n67fTb;>K;9qhsJIKW|DX|_WKp&MWqMd0!-
zG!jrn$uwYec|lDWL>45uYA{^sB;w_Wp?){~EegLVWgA2Il^zNf{JBF5Xaay=FMWy&
z8FYg9(YBV#^-LgeSAfx4G(1`fa4mt{ZEJ+#p69in0Y#$^IbTyA0ND4Z^}$kqR}89W
z=}HRh6(CXixIC(1^EkrGhyr+PS9u<?%Qg+}?Nvqsbxd{X9>`9LAGu@G6S(!q8|_eR
z5a({Q8;~4pv*9kknR=%|fOCT=0hil?65NN|EBN0>3@JQ!V!p}jd@XCxt&ks!Yqhv8
zj$obWfT=Pf^b49$fT?n*i`>EY1YYs8TKe=s0P1@-k|rM1?LUe)1m+h%3u3Ab8i6_j
z*T=~m34G*`kdBiAOIeIofMmT|4@8kaq)faS$R*MUtSCa{0LrM|YUCVMUp3f+>+V-g
z$KiDFuLyOIzK*Y5+=1iK3AYEpElXky)JAP7D+bpi|NkUr+gMzH<Jk$e04!_zQ>XGm
zpr>gAAoySh&_%&(NJ{~P*$b3zh#>dgv;IIr0uVo6V6Vqw0leupC7+Q@{2A90xO*F9
zj%TAxU=^o9rD$#hn3|Vd3os1SOXM6thYfG}6Q$^tc%|7X?3Dm?p3PAQTTMKsEE$g{
zcbg18g7Vy_aO?4eV&s?b-Kg`@Lv^>YMw_&VI8MmCP#|;2^AZ^^P)?F1j<N0_cWXy(
z$F`l^9Po2<29$AP3)NG;pc-SObIOmo=OeeMm-zN1oU8c-%hE=C4#SPPWE!wade83(
z>tmxZ&DvmV=apu$Ye&+He_d(TC2Wkf{wmcj59bEsrh(;7oCCMo77ApQ{EO8xh>4#E
z`heW)<GLv+G)dxAfE0T=1UDvA;d1^xh(5qY#E)XYz`%tQAU9sZ0ZScgGuUd@E{2he
zxxg~(El49@@|&!gMym-5(Q-b%lzgA~1Cmh+H-AKqaV+?kdzX#;HFEX~uYU;CD!K3X
zC}2{gTfo_k{Pd?I`@&zh7nE{2Lw?fGF)|2hkA~ZU)Mt^+{<cR2tadD`1@1BaOS?-L
zM)%~kM{uefbsnF&=3ly<v<lQ)-K?D6d(n=)?h}{hW)oHgx9)d7Z@S!!tr$M}{$yYO
z>wJA$jO9@SDD}>MJUxs(50v5w=Que`$``O{3kv28?iD9trUZbwh-z_GDPRTz3b2FJ
z{sh|Jpx_^&@?au|o%9LH#&_r%A}&>5Fb&vuePgf0(4M8jV4v9#4%(o_!E`%5?iv8i
zIZ6KVnR*Mv!^K5C(Oao|E?M4R;gcfc-=-A}uZ73F$G+FRL#shSw2d6!ccDW`_bjWs
z4BIJ5*R@@BEwNg?>5{=Mj|+SY?S=X1U7$qTc%&3+Z-cor%jB&czYT|m8gmGMCNdb#
z)q;u!Buy0izUC{q<RsL;m=Kr4bz4oo_i8jwkXNA|)s|?)UEl#rdBD;?o2vAib9GrD
zSH}f0N3#o{$_VN7Ykln?HQ1|awF8xbDj{FD<D+#BIbGn0aX?C?meJ??)M4DOT2wwN
zy?EAf-p&)o=WC=-bf@J`(~uFNH*;i2fLnMU4??Q6WHR$5D~gc<ylmrI(0Izo5b+VA
z>!R!lv=C-9a!C!iPm@0V++pluUk=o>whhY^9W<+l?g{gi!LOYg2&@IY3LL>ZUmKRq
zqXMW{Va^k{_7JeNXMtpYD`=!PFFezAaarn_h676X_Ou}zdvxC>%}b4pZVr7jYlnac
z%NctEqUqZ7V{kxz|1iXjxxtDaUaC|6!hWKEJj1}^@wF#Ae3lr;?nim#EGnF8FY<)z
z6A2fjxL${kwh_R_@(5*CWJD-`84fn6RxQSAux+V#vUX238M#tuo&nVT`J)Sz?Y6EZ
zTCc-oFWAC*!q}o9P$HZI&6ZYsmmpcCP;)hTClqV1*WYLJ;oo1md3oe4_4ic3Vja2@
zz97w$LwAGu*PUpH+3gP>x*PrBE1NOG>dPGA+5uFCi&nkz<@o+y6|3pB7c)+^1z^-(
zPZB(QdY3K^%wM!PQ+oil)fBY-f$PA8iYkY!**bUhDx5Q|Y{h&NKHuFAx7W$hOx>X4
z0KIV`04IBCx^;276DgOT(pLzhmog*C996428SZqa9Iz@tmdl%%<FtaGu@sR~q6H!s
zUmZihZGh%?`V)K|y~Z2GeMZyjY^Eh;BZ`HpqULo%`)gd_-3i4FH?<$Ez}w4aK26(j
zTJU(oT?71OD9P|O_>!896dc;6PJT~Y!`9T`_u5B;7x@>)?s+Q~=i$E|dmRP<Ku!=I
z^qFF;YXPXvgCZx9VodjGcn;K|0$hi%V7EbFZe0U*I@F?526b^j3n>hZPP|Cyk#5KI
zAt0?rmTU<{m*bg^l-GK!xxILhl>mHB(1Z2+eyrg<_7R+3k@^}Wk2H<m?A0Mt{1Xwk
zZQUx-nZzZxj)tt$KIFf%(Xh$dC8vSdzlLy$%2<ybKy~J$aGQqkE7(J>uL<4uroi)l
z5ro`{v1)W;H86w$I3|*&#myx5R0Fc{ns3vyWt4B-m66~&iwzz??o4h7)WK)Z>HL~(
zVT29lPS~n&`K0##DoRHDP-cT~g0P!XNsUEw{GfSI$JR1(k68k4?}#~Y$zVIdxj)hz
zxK~()h?g2pZ9HsR1A`t)tQb~d<3LMKn@*4iKFk#b)aXN$%!KjCX+=J|{E#KJr-BF2
zh@(-t7oP%G@~q*~vAF5$d_6K_J0cRFlE?NF8yS-c8qJQ+v<@SD`Fcb<fX^u-L1QDY
z3t^*A>L4V~J;wxj*<;JW3VIYk(fj3CJ)W=e`cUXJu3s4Q!nn?f8kpI`BUVZNwjClf
za(q`Y5So1$Kt;n(BHFa!GIB|Y9p9-T`n09|5TlyB5?C*3(R{d7h9wWul7Y)Xie7=M
z{?<-k6F~6q$xP4>5#(*g^%Wa6TZkRD11Kp-9PZds2K9)1rFfS&BdV^7AWh-%6mF5j
zitHY8mH7sPL=EsPQ;gtpfnA8+&l~7&sDE`Cfx%Zy4jdIWiZJkL035G`u0(UGvp^%u
zW0b7Yaj%VzNyo6JO>M?9tL^yO#x|QJhC@FQ`!^A#V9*Sbr6+>R8J^mh@vvx@+h7{x
zUO0(m><$`0>Bbsn+YK+Ap1uw<Fh@4kLM#^8eYGDEmowABa9d?l84F9uJ&$+x5`HP&
zdY!L%-|!T{XA%3?>$zikd+QartAB0}nE<#2mb9e^DHWKoM+F)f(cviq0O+H<5c{7Q
z>fib*8kku{n!>|(%Sh^U7(gvaN+Ymm#}m2p<?4gk#^W6i;eXwc>lN2D^+m<0V8IGv
zbByk#2JN<xl4AdO?FS79?+uochN{OqQ1eVd;&9x?(uA*&L$nDwgUyRSLp{@@>e^6T
zV5_|RaG+P`eT4%+&H*sZ==<P|Y|>n~Fz6{vWx~uzj>7{t=qU7iY&AAqT=-mznU&ef
zg;(ae?b=b3Ol(BRuE<)XGGsDSpdlH-oUD;7u@{W8va+06)-~Jr>9r<}nzY7Q@q9fJ
zE)BhpDe29~3?&-wE@<v50gMi{9AbhxRs$peFvq%Oa90NgM6ND%nm7>4qQ#gH*h^1=
z%?CdGvvcO<%K`szU-=;TanA-L)jm%V8q85O2Rk!&`pW_?*8@0c#7F8=agH>bPX17)
zharuGKmYLFRAa}xCNJ08o_TR0WqONnvm#&nkmV58-=%4O`0~1Urv;Kjr~aB(GUPG*
zTmR*3gWjw7Ez+6Wm1Ix^qwwK*Kg1>Z9x13ln#yD-(LlWvGLwmi%_7_TUN+s!x)EvQ
zS!}=Q=>u8c{N#Y;sKf)DoV$n5#k}=K(IKWe<`Sl70o@!ye+JFc!4tacx&SUbmPKGf
zz=^JJuL7K`c>}0RXP8Go4;B1;xEAU9>KTjW)Om5Is)Fa^!6qZCUp=4b<2qH9XyIug
zqD4yFso9{*<Gd~PE9OqGX*LKse{5g1$7Fx@1Djh-L4i4s9E5iZTmjzBMh?hGgy#4_
zG#D(etfB$b%m!Kp;vognDCt}(aMVpJG*ONO&U$<Xnl;lAOc{Ybm}7#Wy@zH(Eo<uw
zIGAd%80<T*Pxyhwfec&bQU4evyhy-9Df{;4phb*3B|%?fA==+{4pShfMmm|kD>}mQ
zQrV}L?a`7qqt95Px(gRo37--<J9XFQyTOg{A3H`94*7-`Olcb6an+SQqLB@)&G#1B
z8%M-6u}#C|GwXLG-k841_n_g5Sw%Tmqi!_wNLPgy;w~PJ2)uFu_6C=O?vQ4$)3Z8b
z_j3Q^`P5TdRaBWXIU;(<0%rzzjG^1XF6Q2aJ$e9svvdy&7{mMG8KLMK;Oo4|CortZ
zCkafc9ykN-`VQ)uRtnM7+S@=mrddxA0Y)f}a1p2TkQulAL&CBWMdKF=))id`OLVQ5
z<cY~TE?Vd7->VF&Rmb59YBsX66nBJZ>2y8VKh9RZG()<5$!)Wv@3#7*a$B)mL9JU3
zJn@ll0S@OSDef(}Wlkhq>r5#HQ{pWkvt&@?(A-H%dK|P9H0lO>do_jmy2+e1eI39@
z*6`aqfa0%tC9#KlS6$#Fd1ep(BE`wRpN3pHFW(z~wB^d=4_zOBytkvj%9|m(g1-qK
zJ!+WIU%~<<xqj@oW2N=&*N`F}xTza{d%L!@D#l@^AT28Ur2{^7>EpK9ZI#%2jj2i2
zvS@V~&u%Y|ZWR}_QfJR@fxJFA7V5}%wpVIo?RtFtUgh?zFYd|ydYHW+@%L*#5kLCI
zzdmjvSR`s}ux<*Mf`BQlFYMcU$L`bDkWy(-sK}^w@aHJkVP`Nsc98=z(#wdJ;gKf7
zX5PBDXJrIIiyA4wT90}jU!VH_f1rTdU(iuXQ>b%c5MAojHXvdN`5PLvlh;)%hE`S!
zuS3mRlNp*}FOqu_eg<_EJYHqJ`t}p(_SLnnT+##5bv+VU--O@;msYtLvvvz)Vn;u#
zU!4%YK3c@>%-igZI3Z?SRJm2inquOom&HFVD!#wgL{Y=I>gI}RlDiHy_7RVcd*Xw(
zY4orYBY~~_wMF#pa5~9=MtLWw?jbX2LEapZ2bM{#j~(C2IoeNge(YeEU~a)!+qC9T
zv!qqUZ~o|CozjN!<nk1T)dwXxU86khlCt`@7R|nQnir}5Bz9J%Tdp;71%C!<DUwI!
zQ(ws}-b3&5QOw_JZM<)-SfyOz6?EX-tuMBu_3wr^*Z%<zVWKI!$&#&%oB5o1#F^$T
zw1MO3G=S1K*>fPjSWNfrgYDA-BP4d8F}45yKLq1g?nUwmo%ZVqx^jE><GO9A3ujV_
z6B6pW`UQdi`E%(OmtRY)SkG{#3HT_I#T*@UN{*GppD(Mgewb=Gd$XOi-ehM3g?T^Q
zN1s%kC`aOf4+%2lvY6&qFOI)DzE|m;-_omLZh~vmI<ebB9tVsyAAiJTS{=()<_J;L
z7g}Hh6?)nkg*YX(tgK-9;#EG<d2w%yc0HYL>*iz_`yVm9f^~2Q1uSy3uYBs6+UgC6
z<esuk$F>J|KGuj^y#4K6{&eCAu!IiSl*5dpM-Q}$Kh}fjNDpE3Y96$_97}F2GJ~VU
zRDQxVt*Xm^l>gM}p|alT)#mGOS_H4TV_SS_(^En@yG|g3>Z`6mv4bz)OcLFHbAz1f
zlFDO((`QB|FSB1D<ITV_R|Ih`;pPCLtUfAhp%U;HCxgf_z@r{!Paz`sot(!&mj|l^
zIG2Q$L;>VS5OYi>_psGpePh=rcre6L{FjGx>O$A|7=F@XODWMJRqb$Wu2r4Hl;Wq8
z9k#DaY!OB^x-@t$4E{p+J=Xfy_eEc(0K<wYN~?ipXVv*_K&0s#ZmVBA|H&l&PV}kM
z3s<|gFFc*`=zQ0&Zd()lyioKVa-BQ*1l&RG1iO^ZiS<dHT_G=?errm}1E_W3gN;JX
zW0s7a%#<+P?!j#sXG&k<NX&M!=TPWXnik>`k3De_w(-`I{i>8$klp>`w|{{QLY!V2
zQP5E)3qtqYQ{Jt+G^@T48)g!x;^Mz;neoEzTSsmE6FR^q{99d%Z-4%vn>BXdjke}z
zY~<cmRgP(`SyHSRt!<;2a_dy6Jnr3t0}FRt^e8#Szx0)hG;m)bkoYjOal_2T9eKiR
z$Yh-eTm2+z$Mao9HHsDo^>-i9;`bB*uGW8GFE&-E2&?@Ont6h`Bly()0o1e}R%n?x
zl#q0r5j)Mp;Wp(|7DnO6FQ)MS->#0^foOOyAel!_Ltg^htg3CD%X)pEeIHC7x)0jW
z^-CAO4MESeLYs2^z@9jOn$sSPzih82x5p}f&nK%V-qdwZ0vuL1@)aHrk!#%yJUDmu
zEacq}4eX`+nh`=xG}iSs94GBpTuHj|ByRa8(Jue(R;|5b8#(S4$RUq5IlsDhy%IRY
z4JJ)-^h%QwXIee6eCc695_V2rrGKj8`2JT{^Hw)6Ogumjx$&QwQouRE0_S8Fb6dT(
zszGh_W<;Xj+ARLW&6anhedRS9Le{5!dgc9l{M88;{;zp2**JT)5=W4E#jGw0J-0Qf
zJF8B2ozU`3ExV!{2}uhyGF?99>{qiYfd4~IY%L6&SeuOMrCH(_FU59dC(h(>r9CsC
z^`ZQ!I`?UM)f<n!A%bGi|2K^I^|ekkqW5(~(X<+Zyg#Sf)VE8tDk=LfFbR8Uas8=I
z!s)1Mg|?>mMrf1(9HhmN<lb`6VQhbr!-ZbSjIqp-KZCJ<N8FK&lV)nuh_#RJL|K|n
ziA}vAWLXzbt9SZ@^!+zK&TsAWJ6E!MD)nCI0IDFB+E*C%z7TG%v0=CDmlozrYPNvB
zbSZOtR(y%a^!MFCSRz+6QP%yHXbwAn0Oc|R=U^nfXIYfZ>fIOAkCf~e66X7)z5;}%
zU$RJx$n@p+pD=F<-rKr&)fKC!AJZoVRxH(;5mvq*JvKUbHWTKnhsT8IcJSC}pOOe7
z6&YZ8#)SM@6Mf!|IvXJi^=Y&m&JqMe*q!Y5F(EflKgBsWX|v}d?Zxs*?(T)Q?MCtX
zzbh~EH=zFyCyhZ>gg>y0SYl{iPwhQN-H+VJmZ4%JVhXazZJnfOP+#*Ng4F!hV>A&B
zzcEEKMUe|u^Al6m22hI~p+V?l=*(j!ANCA7x-))38W>s))jq&>i-DqyGZk(*U54h&
zt3r07K_;kUKAEK^=Zwsd2A@fo3{zh9AznftJD9FUZVY7-kaR0hj{5NLGV%pw08&og
z552!S!B65(8bICNi|~VF24#*3ne&za)Hgz2cn^cdhalYFdp!Gh?WO)}RP6R_k{SnF
zq|VP#A~U1vRpk7(=@pqJ%}uwAdk@FIk*C<)dYh`~cK2Eu&>wcrgu6uoTgY#pMFld~
z)7NWrAF56<r(<6y#<d2Y3tLwBP&DO-w`K2-9>qOxzhJmD<VG8T=3N5}7f5kqX`!=B
zz1yHV5#XPP#3JN6EzXi8@|QA*!dbx|^x!3)sc8c@KZS`*ec;wacX4|0y#uHjsHT>f
zFl3G~wZ9$HjtIlA<(vmlYR9<nYqw2KauI;^v%e2<%3%$lJTbg4`}e~b4iblp{B&AH
zk&9NdQk<=)eioFA+S8(&{cd5%hPx~S^ziWS7EsW%A3c_%hMzC5$Yj)W8$J4{vpt62
zJh@SiT4UiWk&;o5Zsb|>z6?zNZ-&u!Wgn$YE{<-)O0kK8cjmpMaZCN;DbIKK-fa0W
zdwsy{;|=(e{X~Wf_bE<{ZCqnz<ZvP0Wt$c(q%}ECQsalo7g}CM=t(y=o;Tl=1NP%)
z-oyN6eE-~O+WSo!%5?OdV2kYpQzXhi-z&Oc?xuHI>k;aV%#(*P51IS>p!Pun7dT?}
z{@2%CBOu=CAluZZ0VuU|7k8F8NX%VuhI6;jA?^hV06{!*A+Yt3Tqy-+CtSyVC1~`l
zI7rL8ZsDOT+P3Q#OPpKT8sDsRkG&4bab3rhs5ikLqA2AFZCBe~Xe?_yeVg2Lk#2|3
z1!=_h&4Kl%4!t3IXW6ZsL@1FKvi!I{?Mb78FexEdI@i+lbKFfvQbtU3$-lQ&*aw(Q
zum2Vs2qR{COf*t(NYil%LUEmLgQwDR!?u?|bxXhGCsoCw%aa6c63x}kZ~GAxKC){*
zQc7UnKKSqV!9UfV_+;3Hj{eEf1KIio6n0=hw$1Uh*&>kt)w4-?Vj=24d2S;2z$<5R
zNE%x2B&yzm=j)92JpOE#F~w&304g$>MeL1)MlB-ZEwTlsvLK!CT$ADV*pZ@DqI5-7
zdtvcbJNleCK1U<dInDUM)MH|yXos(9fR<>Eii3lV$cG*7Ck#%{I?F%vOO`Irsr@tE
zbrqpebVOOf1Hpm^4j&5?mkro_eCEEMK@<?*zyx9`L{C7<c!=#bN7fIZVk)u7-9^Ng
zvJyVd&Jn(%`m2f(*Tvkw-f-=zV<`Fpxbik&wpywKh%vER4>XDhA+aK?JEJ?7%r?rr
zn!Q<eg@G7j;l7d~%+a5D<hv0AH2>!92Df$|?%~H?28DPnYQLDe3`}_(1S0R%a(#+!
z+tG_kRZLX$VpshX{UAx9FExJO`_1N)+*S_(tN!F8A5#Uxc~lONbGqc7w^n`L0O(+D
zIe<FJcBnYY*1{?u?RvkqFE-?^!Q0;*>wj-fV4~~fXeo#w-=JP@h90XMVA_t0u*3F!
zj_*`wK6#bsT^nA#ZV739>otuz7GP#fe-7r*9DtWv5^ur3aK=|s>=04?w;uV&Dl)`4
zXNAZX;?Q-`I$6ky6k|!b9qyl3FL(96yzcbon`$Cj%dQG6e<{wXWQIl5Amtc|;#!$Y
z=Jrgh+Jhh3vkafb3Z8e6ki5C0A!6o?UUoW_Ais_;-!+$<fXOVtW8%BK6}hKA=_pK3
z#Gi2vw-j1<a*<|8>EUxODHWA6t8&ilIdntfTtlxcc{OR~frFMXZ4c(k%M*7d8C4oN
zp4)iDTWQ9%oP|b5BW3+3MW9fdXK*z`r=2=w-O>3dHRalo3zIFEIj9)S*?#bfDoS7(
z<rBusRmQahn^15$q0_k5>#{sa3a3(Z%IJM%x7G2GPb^Ko=MG`dtRJqOp}%|09us#Z
z2DTH5I%GkJBWIdyEq!;^Z!SaWNQqe3+T+fgkK*0ba>1QR#gG|RAa90e0Jm50$__+E
zIc|{*g7SuDplV%baZWRfXzYf4P@Gq%f;1^A!fLHjwinS@R#WmrWNi1xCELT#W$)e@
z<t4G>a9b%0jSZsGc0b;X!IP#iKjif}ygZk*!ccO@!D#8-&b1;NS;5>bp_BgA8|+Ux
zl9wT?kP6aj3Vaehg|m^VAE`vRC~wb}m)o7}t!P*BDWFX_S@T*c*+D!p{n-U=`G6U7
zdV)Hw>CTxq^B*Gb!G2`?p6S^zfchxlcX{)<kp8(JQ)XV5-e!F&GV=2p1KV!H9Qb00
z1amPR%{a((uJf!G@=uIDv&n>LZ`W6O{-#5pZQ=3G-NL7*pj>kRw34qsm%Wo?$_>Rb
z{AfvOdKG=f2_J|F@+*m_#M`Dxd%b(t?MiRDbV6f^mtR;w+I|PlVu-dE&%$t&2T-oo
z0P2c06^&x%)pzD}HqiMqG1EBWss==(gZE^!PL*evpLq6gj<&5_tgrC9vn#M`l|^d<
zJ=MuE2$(hw9o$GuR#4~CTI<bSuf;Fc=wK}}yF9C|-cN0vw`kJV9T8Xobb2WI11yvk
zol#f48Z-B8ka*^~XYw7_+~$1#;qcD4DE)ZSj<kj6B;`&%x0AHhrmJ5$Qm;0(Rl%-$
z*R;>p;l54FybWghs)v>(45O%rz@h)CqsSLxCzcK!U_vy}Ok@+S`fX+VhA11M>%Lc(
zSH#Hpuj;tEF0u5)PqVc|dJJ5F$w}8!y3@<lsl!ierRl71-~2Xj(G{^=|1uegjRH5%
z@0EJEd#b?29x8Js))yxq#!T8pFY}o4ILUlU$Jzp$p8{7I&bcys$Afe?>!<71opvNX
zBb!G{Fs>GLC_UspNT^mh^z!C`T_KfD+b?W<ofM48yD&?;%GZNwn0FSp6~uEF$}M&?
zeKIgJ?Q~T%p1G^%hv~>vD)Cu~3%S4kn4P~!-!C$vzag(cq$j*1gRNtMiFfc^-BlO2
zS^2Wv!TFMUyBunco)Gxy+TBUV&Muhgwr%C!!nu4s)i_V~0oS6<)bwZ-4H}&BN?q+(
z4x*GiKiDAE<E+-red}H}oz>a=^zhY3SC(D9h#^{a1K{Ig5NQV@Gt6jxJ4xzhi9TV9
zm%T#ml3u;o&+3x3?ND1W2`k(J({^J#rNjd36DWAQt_6~^P(quNc;ycf)(sMm4!q3}
zv-X~DvnuScWQso@8(-^{6)oD}m7yk4zujw20)Ef-dB*Dnteh|3dKkNQ!TSZcaQ?Gb
z%f+&s?q$xd-Xb0ScKVXrJ2yXjV|L`iw4Np8b#U#iq`j5QrS$@yIyaL(Bor2Tl;_R%
zBSXsoE2(-2>;ESlXPX>KH?vY|D!yFjF0OT~$Y8rRq#@TWSclHURP`|~74|}soQ?EL
zf#DWsPOh)i6zPqAdTNHOwrAGeM>qCsnq)oSp?-rU!`5M5Y6hGXl&VWqh1{z*A6AEY
z<xRPzV;8b@wg4oz16_x`gL#f5vyUC%%r1&oaIcyCq0>&s{N?RlwWGV2mB<8~IC{kI
zk<Urldes|;cLpq6=*GtoO}CkeR8^YeJ3b>Dm`U49Q<um7D3kQuU*z_~6+iqS9Cxu+
zeR_Y6R01ObC(Twz%1APFhfJ#icdV4EGRA`gpOn0Rz(+ozCaXqYq-N{FoS>~StxGO!
zS;*gKjXK|zMHEePQx|ni$}U=?9j_+a{dtMTqI}_1MZQwA4)J-kV}r3X5&-gqk~M(h
zvxMj1);IXgr|OhuKAw`Akl<|{yOS31HgPp-)kXZBYaTVEdtb@HeMSl$8TE2H%N*_)
zg}l!c-jikU_~3CFN#8STWNJvZOoMv3Dzdmv!PDAR1Z-<P<XILhv-opGZ2=z65kKjL
z9F100nZG|T^2S#Fnd@iH<?99ZEY!7(rFA)BCE3DeZmUs^KO@C@=`;UjcFA9aKOepM
zOfp4zR(MEtd;1J#0|U#AEXc0vJeH=usR*tU0g6JY+{Coc`)=qncF<Q5HZP#^6<v`_
z-f?1bLx<DvcAUn@dGNnO)*W>GvWID0Z>>`28)E2v>ghx8^uDKNW`Ru~CEpkvet0$K
zemXH9>p@1tHU#&OIZ{G284uMBH;IZt>vhkl?m2WwcX3nG8M~CrU>8{v>Kx2H2O#fL
zRp&&xmK7lsWCMxz*(WTIuCBXAXiu3THdf0jvEY+$f9<I>FT_YHIw$<IzZ)wvc>W!k
z%lb=7jVkAzyT;9Wu|l;lm2h4COm4#~9K8;z$Ux@6mB+H+ctg%I2zZkhL()J&;)F|}
z`8Cf(38bib1AtC5TALJV+H=$tIz?zl*$w23qFQSxGr)1b{&^uv)sl<v%P+548Eba2
zF3G;S>VBvY8BY>Tp@*j{NIGKmw#k3GA7+|!$|LIi;uBZk+FdAO2vmEoM)n!VZJkh?
zw<GC>X>&%BVVU7H^t)*3jE|4C?rR4LkkI#u(H#D|E4X>J*jbJnoXbj@3hk;?rly#!
zpa@yb$TX8Z_%=xHg<rmJgG@Beyq-{=WKyK+5Xsrfy^AdR$WeYtneN;^#iMin**EVF
zJik!8q%mN2nCG$%*;EvIBR8o+)q&nutM^)0I<@`F6Q#})t1pWCy9-crXRA!UbV5B3
z8zDjmi3x1Qx{NwQpI)?g$R@gc0bbSRiQLkY4wkWpw@V!O!SkMn$m{v?ul*v0=c3Dp
z85Dm;G{e^~c%QR^`%)E6hqIyis%mwwL=TkVbQCWu*vs8A3y{-j&hwrfm(%n(OJn=y
zCH`w{>#^~tqS?kKoc&BK<t#r@tvLEJ@muPW-uWdNN|Z`#|B5&F(y(iO*ko24;*76x
z>i22z>Zk*xjyv@-88#UB^)@cR_RL^#ur7Bx2o~%UGD&yLZ#Qc-!4XyWh<$PXYcy2A
znpY->DWuMoVYVJK3uh8K*LjGz3!y8rsjHffUp=MT{D~t+i4w>%l4AyDi0616P2Qjv
zdGMpr(>jL*zVzn;GiEf#*!uc5L=Z)}!LC!<FmZkEHXMlv<^$~fJm)7muB)5eqbug`
zIO6kmM)9s4Q<qeHRA;Qs-ocN|M?RQNWyaD4fqj}8Z>7UkK14fC6e%%mo$2auLvGtr
zjiooYzVQ!}zCdfwQF`rK%?fc3cg-WM#hV__Z>!p3;lI84rShEABkN0pEN$_1`_<%>
z8<abo%^C_YIdqRYy!U5B%CSjfp6S7yPupx0&VAaEk-hDC-GR@tZ`ZuZ32spADko%y
zC;_-yJ+{)YUbi{Z*hShW{n0e_Tr+YhBl*q7{tFN0r%my0oQ*!k#Vh7xcW`f$FbA!?
z^X$3zKDpIKO*?U;>EdD54|({Lto^rz^)tk`&6w8UJH<fXE%Y>sKUA3YyDij2A=@v#
zCf8ixhl$6#lO|i$%q!$ro4XojQKr@izL~q!>Xd$iws>95KC^vuqa-7#t6kP&w@~o7
zQ%p=Pj1QR+m~N`<nYpvx$|_<;bk@_hX>+%~TX1&z%xR0=8_g%9oKcH>{kIU`Qp>3w
ztmaO8dO;Rx?qbgB?t%pYi*FqdvC400%eWwscOyvhj#8Btd!cKg*EUeXgrQFU2>A7)
z*z=R5V<F2mcHKV`EUM#d^6+b7Hdgt|b?A9OQ&NE$O~1;NOS>5@^1^0A{dr?t-`*QD
zKD_-nDHDZ4$@-%vxv8BXE0C0LA(Fr;+)5I596;?mGBvPXkUzY?z>Q#G{~}(F!1$3C
z09O*jT^AznHxt}Q>P&KjS&r|CQfr*PP>J?KwKy%w`!7PoJXSq+5PJ55;pHHexa>8~
zSUeU`sM%W`*CTVF1#;=~akrn^98L@Q;F97H<7l?jYg@)lyTeaWYI`$Di;7ic>N@qZ
z#7iCs3N5d+O0K@lchplbHkTpN-wyDaZwu|Gsn22F$#%MU0%B?_C7Z7D74WZ7+SBHY
zUwh9r($WU#TVQ=_);vq3L74=9C;y~cpZ$3IV55GAc{x%uqEg;A$ZE(Q|Dv%}#TE-Y
zDyJioGy%`}f~k(X@_$nFTOx9yMkKc~DtB4P!CL|Ea&8{gZY&hg9pah(2}gn3oWsoU
z2e|o6_z6ynE`6g3x&3It4@Kq6&hDE0Vb#v`;{i%(=3@OX7M2k+U`z`noGSQ$vxSzo
zw%9(;{>7E!RiC!&p%=RAtZ*zBK0KYzy|JwCG7fscjN3}4Q7%W%9YFn9Qs?Y^Ynhn#
zlB-9qBa%6bx;GIExXO%`%xGGfbN>2TDBEvahLjlj`Wrj;9eumQ)qS@yPETy?ca*!W
z%k(W25M*z4A!I_d-Fg?r4Lzp?9{lhiA#jhyrBD9TKHaQ+c&i2HmtR188B$N)*WN^Q
z!#pQ@k_6hsgM7LYj0vL5s)IEON-3Nxw<SH+ZTBP`KFW9W;BEZ=Y!udq5a%CGhZ;oP
zNwP+c+19hD>BUAm%Y$#`%bNOF-9Kyaa9;L2M~O{-v`x*AQ0K51S9PW_!+~}$+&A^e
z<F?Iyew~#S3Jlzt9l3I0Z&ElFNBMSaZ=6@KfOzDCk~L)d;irWB<@U+3C5!#*r|)h$
z?rkv37kz)ScoVV=w~0o`!cRSE%1@6DuTqTG-S7g07BBBUyq2V5vS3m;C1U9!j><zM
zn;^*%J6t2kJWp@e|A3Rq$~@O){^^kM4>_MT7u`NX2^70*zHCM4EKCO0oQ~!QcNv%9
zq<**=y!}>tl=ifbm-ap1A$Gu@@1T+5PXnmA*UG>UHB&3tj59Q-qc`+WxugA~k_5lq
zPlInhvCq%?=&da#ws?+?WaN6vs)Nt1T4aB>u^X5`QD*|I%x_IJ{TE&jYA1pm1Lm0q
zn|W4<5DkC%X|JEv>*v$#jg1ff@ciJ;tn16F9Jhb;-4J;9iKHLOX^6i!;ZdX+S<j53
zHPqES32h;5ri+AlI@denpIct5e!GNilHG`vUH5T*UPQL-8=}Uao6-4$vq$KG3FIRF
zS7%7%>64pXN;4C0QuK9HG3`sW_S+4h&Tn}GgnzekFZK{CmohwqGO2Oq8Po;3b<j*J
zA*twkljCC2sX8j~#~kbFHQ{{)GAqO!+#EJ!nc3jDF;MMY1vZAW9$P6>i^ummyrLzP
z602rrElR#}%Iv^g=kuMM=MQ(bJesyi=pJ?DWPX;8!;Yed*Dfsa4cLX^PgW9&u71=b
zXlu6a&{nS0hFRfK7mpj(Dx_1ru<>Y)i1O<~$AShg|J68YT0_Q@SvO`89%hRqxNNUn
zrfe7xHAm{8zr(t9Po$S>R%UXz^td{OOa-R2Ru07P$gcD9UENEV&J56&(m8L-m+bv^
z2U2Q&exZ5ZHKYG4@5-a0Z2$gD@+eD|EDvIor9y@>iH0FbLX^jn!4%n(qOy${5!tuU
zVi}2K-wk2L5-pao3)#2Q$P5{`X7S$7`Mu})>p8!F-t&9U`<(aRxzBxH*Id_qzTfZt
z{VbQZ<$EP4zX@*vR1P*QW7E`JTluiux6m}uw&X*bXsAhR1(aP#_K@LW+yx}&1>>c{
zOo=rQt;~rEYfYqv9iLYpR&Ka7epr-q*0EiHw+enPoF}tHc`kyS+jw>f?yy(Cr_`6h
zY^C=8-1L{m$~Plj1-VP5BUC-AL_VwC*{23&3rv2QbUH^APtB*Jl)tv6vG>gmkFZv;
z_t?)!AZXHdiqyt5!1d-J$s<^4-vE9oQYOAXsJBjUL==5JXa50tc%4GR_$9vpICVLR
zY2@jo!G31ECJ!p0jhn_6-VP`YuW6QeOotSe&)eA4@QKt&pd&R-DJQGj-Q<P*Qudsy
zk11?OaZb>7^Aj^3&QY#QEPCw~J?KfiGgVA8JE<BLdg%#<f^VgK-d+G%2fvy7FTmD<
z7R>PmLsvPbXaUJ1-kwl>$sTow5+k3zuHkaRr_S=I8b}j(7&TyJ?`HwN6v#$)9*8^;
zqJyWJbHrbKnKDAgiLDyl)zPnRyYErS{wNc0I4NC`D9|7W8(0$CNTM3i3(@RclpB06
z%Zz?O+My}j8@&VJRcfqyzLs<A7%%^{U2+(NJDO_`nTvI{lrx}roc)$_nspF3Skay5
zh0~?$kqf81d=xTkr%umYeG#-2^RUO*i<`gU_PN9Stzm+zmYC0(HHC4}z5R||Cl-46
zz7xxJS4VI{z&dANhd@On9BBxy;XJPJO(17lTg@f)mqH?6cdi!qbi#7GW3|@PCz6sf
z=D}ku5jZ22zS2pC1`_9d-?dy?xi*5`k0_aNa?(zn9e0coGEhi(>d&8C&zJh3!C_7x
z0<*(`B)-}XPB%!nTrQcP5<f4iJeuWcf297vWaCKOW!tQ1d7CFGwGBo{gb@>pg<`^H
zYP#ZNE6d<<&CU`|W96>hm$w6oWt>woRn0Wg4O0z5se1H^KEcV?-|!xA;Mhvr`gqs3
z7c2TS=4WZ<?PAJ<U5HQfLSe7FsG)9S#gwl^po=NJA`(IP?YmcRp?5}rxEA!l=*Ddq
zj~P$ugQ_-aQ0{@)Se6jDgVzEVI#K)Z9=@^U>Z9+xzqz+Wc^1~3YEt)ZQ8VP{*vwy6
zlXFdbFfNWGm(SFK=HSx!Fao66hi-3c-7uiJBalioCJ}@t4$HJ<59y6-Lf18I7cE#I
zkTaLqn2!}6)6)d%m*>~5fQUEh^&!t?FEF_ykT<$EE}nGI>?Ajf2_kL3|EozXW-mQ?
zHl~0W9=MpXXK!>V-(g)Y=$|nwja}00ODBLOfB1{Q?GDV|b)}?Ey<<sc{r9M5A5P3z
z4XWk^S+3l=Roru)(2B#AU~b`h$+SzpB3OQ41Qcn#rF0%FGtciu^cA@Kcq3j}s9a!*
zI_&%y>~d8eR81-0+AG`#TG9BuL3_P*T->l9pEgAHvwtQUxnZ`$6q35gP?9><1Q_3i
zaEq~v?kA0x%o1YVXykUv{+*&CjJ~5wgQ=Ua#oaRrDTRU>8*6i_!eJtZusx)tc>*90
zi(-LZ0A*Mq;AxpBNqeZj+f#S*d<NxC^BY0$L;eOinH?AZ*o+tqL*4PH_>=@FC9Qcb
z5aR0fjskr2PeDpR3XMMH<Q>M{*_(BeFKF-LouNtd=tckVx&G`PnWT3hSZR%lo<0kB
zGJUaHbmNY1i#mOob7ea7pCD3p)9XxAOS{c|XOe$(iAFR*4kd#Axbh9d_n0Qm;Xjs-
zxn|8H0<7KjV)uV!DKFua0B_EIVNbeBNAH5~g!jBfGy)T2zvvv&=;Iz`8glLCPV;Lm
z;Y*vh`VNdH1V=(3-o0cu!PproytD7b{3JsS7@yydHJ_yjedH#^1e`s)&z|bsg|<==
zPn5hVd%kk9*fjO=lJhLA2vbj_ZhJCKa1_#E8d5_LxTtEm8hof@y|c((Mbp5-Qjz<m
z%eX`GN$wy5>);bcPBXgz2Ss<!7qMm}0UhhXI=}bZPu9#Prz)v#`i8O_3XdS_JTr9w
zAM><B1$717EEgb&=JxT$qyc@rdBf+uQ0Y@4wVGS{hj43V$%VOoI{iHBf1Nh?ztBql
z+Y;-4FfMHOZwwXwp*0KF4k%X)KZHI%QP>qyO>Rll<re*=QnzyhNEiDBid?u)f(A4)
zz{_2TgXTJ0dtz(_!xEb0=y4Eiq%LbqiY*G>{tUERNo0XAez{ZtCu_}vD;50lN}DLA
zN+NtQ;au%la8o1fU5911AD)|wRsl+{W*b`Z?OHTYm^?HMCbYaY73w;mh+=_HdPx4h
z0(L%!5<0Oi;P{}ULyGsD1UvLQ>X+JJnC?Lk+g5u+sM-#5se^sJ8W>?Np>Y4hP<ao?
z+ONKBvc0DnKMNkm2`E2!NyD^BJ#P1n%aIwvN(a0eH_x9Srjz)zkJ!X&^*ud29X7W?
z4dBYq_1-$gNMRE+cOeu043{Rr_U^H|VKr5s@}Esn3aU~GwKsCzyKon@6M&;G8i(e|
zNAPa=N%DJ4B`F#2MxLOwzU_IC153T~+7X?H7`8l{>TLadA$I<YhFh%NlHBnLF14v#
z+2?d=hAFr#<yd=~Qvq#r{%9wL|GS>-s5s(nZAvO4&FIJ{f@ZM&nTfBZyp`HwjBP4q
zE*rx*$WAD%X1_$JplSkTw)n;shh$BkaeDo=GtLb4Br2V3u1>d}_-*;nmi%S}BZ2)E
zC(rpZa1BSnR8bxwzRC@&U4MFY_~stIyp#|8UcJ5wDHD_XApsdC7uIrbTrdpXqQ^1J
zfroFvnh04ZI8HGgQ$t!UpiTL_3*R<2X4nkF%gP6TdDxi5ILBMR3e$i%BNuoeu)~4=
zj3MyfJm!pUzl|zj9%mtkSh{o4+VOihn;+BDe<hD;B^^3>W-r(ubbc;fy$`XBsRPR$
z;Fu1skE0GeVVbdAY8z;rCt@Ytx!)cYJ3cR;i<HkkQMPuyy{ROA=aupq%8@ZDxMJJ~
zfl#Ij>!`)}!otj>tujGdTEW!v#}lQ)AKI7Q8*aXha*wiGHuifJ#eI6^wss4M!wsS&
z@xJg!V0;559@{kLAA=-Oz-_|E*I{(a)lO<h^PSe7X`zDtHDv{1FeCN|RVEmGfo>PF
zL;^sCZaZd+U4s91tByvb9{2S12d7{I>_RlV8`oUv*c8H##<A=11m+T$BEjwupuiJ*
zuf?BnuD-u?SukL(*ZoR;V-k?u^WE>UY<x8>_@?l%)Xq_cz%C^14W}Q}7)b0wRt<dg
zVf_Ml6q#l$B(Y%e-pC1pcG7zqfM7#yH`!p%2^=+bgg*2QM4Sr(G@)oU0?nav;=N?r
zj?*(ev3Hmu4mYOX8pp-?Qtw0K<M0~O{27nZcI>rc3M$c!gh)eAxvG}7Z;`bbN-K>B
zCS4y!0ZrFLElfykkXYk`iJYm2J(hDF+6sE<X!YsPN8K<xHOrL#FLoE@F;^3V%4ayd
zLv)xWkRC*f1A+CsXIfh5OH1{|?WcEKO&>FrvQu@serz;SxSWZ%^}v`P9{Z;VI&y`1
z93Wn-B5rO~2y`LVG@%)rKeyacMDL$aY|KEwrf*<UQ8d#E0ic8t&n}n+y6I5T<PBOx
zN4=C|oVudTO(Ih))ql8D1)0Aka-%60!imC31@ajIXxerM{s2;^>E4WhCu0J8=z2A}
z>tKirkd@FAl4}0o=f4zKFiIpN+q)Rw=5I4W59oT*aR#<F30!eA1a!DC{fbPd6OrAX
zs}=X;{Y9$UpP$aoC73NB01@&=HU`Y+t1wKz00ggyqJc!Iy|~YMTBEFxyUY`TV`j9(
zGOv2fBx)iAIT@M0qCuT;zu&GyV)-mR)n0(ThSBozv(>Pl_E<1xCgXRkFuK?)G}|r|
zpyaz|RmutlKrv&IFiW0pb!qHQmqVMVmYQPfV2~(kd|I-Rm_8h3re^nCoD@mNu*3<o
z)SQFK*wZxQXar1R>Tu%VAV(#zOni>?$EHH~{56>Xp>z2+2(K-(`hQEJx-f-s!(?uh
zmo9o?rAs%!VC3n$`WNxRrS%dbnK3`$`6fL%w&QZqw_m9DhU@BAD6pGe4ZQ|;>@}xv
zV6{+u8O9&V2Oi}+9dw;3F(FtNvcfuOIQw`_xX{Z<nVpC7$hNYnBhepbQl%7Go3CeT
zdsn_VnY&R%l-!^<uD(py%HnpV{8<dyw4CaEF}Bu5>$pIZygTpB=DI)}?D12*3||^q
zPo2CKf9~bxrMVc0tVyomx0B?Y89=#ylwCvOE>9q=QEI+T-W}ty_O1E3vc6v7(@&nQ
zoEE0JJ%mT;p|Cpi199}D=V5Ncs%3IvXp{S}PraEjc^N`8fkg@RD@*RbAMGZ)k|Qya
zc%P(+QNc;ff>Q)MiY1wA!v9MwtF86<+Hbe=>O~+F{>x??mqk^&lOvX|hrviAtY54f
zYT)GU;XG<7z3trvL7??Hb^a#EOGB`^ZD)<CD>r9qkFvxn=8Zc)TV&>AQ6Wo-*2~9b
zN4W=lTIHGFN!GfIC;Bkb&;K^z`cLsJ{*%we@BWPx@gG{z{M8#QZNV50APu}3G9ebk
z;j$Wn^0LD7Q{U^_B+D)x<2q|P@H1Ov_?kz(r5(EB;@W~;GQQOkYOi@k|My@*6cSP9
z&=FMXdHU(*WU@<YpnQ6NV&t{7WL$bZ#8T`6WEN^_^cTejs{!T+UiJ$`_G^@Tg2j0n
z`%MEqAZmtbYs3Xz%uzgozK;LaY?Q|TaCzTHW?caj%EHGn!ax;2s%Z68O<-aB*SAWI
zqRnQ+3=g48vzttS@!_&YL-m#w#%t_ZAbj>c$Y6Nvbes-PKnl(QnvO1A2l*15y!}?;
zN;d8yCbH{r+=_q0F>SXkP3R7q=?G@?t0bXGmO7o#0eeKA2*4qSMb^)=mrBsag&&s6
zR?<GaU%n^ZT=}q?X(Gpku?BBAgL03X8DC+z5B!`F9?wLH1{Pl!t+SiDs{p@KT&X|4
z_<Qaj{DnV0V5VET-ZL(<Gc{*ms7`6rV=3jp*QOcGT3sX8I)}9VbxK^t9T)3qFKhbz
zG3WYuVp(qNJRCoV@*o(TR~KBQbFKkVq!>Gp9~IO*_DJQ-o_&k*(8mt<=bHsNvQ~c{
z13hr14g`E<zz@%ql2r^0s{~-^HhgC>F(i?}X4?Z<eT<LA*9-$IYr}XnV$#DiAbX4*
zcsEf1STBwnbE3B&;b|tW;8<_CK9c8_P*s9#F=-47I*i(pB4Xj-UhQQkfk3t6(Ke1G
zkmM6u>P@_3cMV;Zz45&KcD`L+{He*}@V-{pXf9;3N;BAXLb22EUy}lj!K{#hVR_Zo
z;>R%zcN(L@rQ$1!w;5jO6?~KKXg`Ji%1ANaOREfAN*GoD^tJV5rCc~v6uN#7j17mO
z4{aoQ!Ug<;10s;HSJQg0@;z+Cwfy{s47^3JYK^2DptA4PT0$|DZI-AY=V~NVhmBO+
zg%k;l;PnwJvI>AM?NyiFi6LajJcY+oiV|}VbCjx+z7&6I+Mb<brKxAu7dh0O9#=ki
zx=O)cU*>Yb9ERUpt8&$zl;R#5a9cxkX!5;kN)p2{EoH2#>_-;wpMF0mANa(c_Y5C)
z4%U2O7lJe@`*LpLK^%<!v1m#~mD~8PxRe`g>cV;~y#i1oP${8IBnujMg?|4fkzSFw
z=xA##k~Cu|xz`@8jr@^#T5Wct`r>M^?PrV-j-RCtloSPq*Z?te0OETRuVX0DjA`;9
yaeCp{w0p+y@>zN8t=rGOUJljyw*uaOC+_{fa1FKlHwxB&di4B%IDf9)@&5uG0Bn2!

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/ppc-pvalue-nb-pois-sd.jpg b/docs/2_39/stan-users-guide/img/ppc-pvalue-nb-pois-sd.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..f77ed895ec95966fb81ef65f32069457239b903e
GIT binary patch
literal 108780
zcmeFYcUV(fvpBpV^eP}AAOtBQU6CR+DAMJCC>@DNjg-(qL4v4sq$wyCP(V<DC{<~p
zqlki3sUe^O(h|V1C53O}d){;2_dfUj?sxBB-|u<84SQ#@*6dkpX3bhNYu04<%kC_&
z&%(sq1YlqQ05j+h*d+jl#!)D50I;$G6aWBV0~iuf022g3D1az*0{~+o<3CGS*$QF*
zgqdFLRqoEQE}9z|A#H4}P0Y_2|0d1)#XTs<pHUJ30s_NAY)wx|IyyN^vW^3cfC#_|
zQRQ^^2n{-EV{>MY!XED4=f7X%(ce7)hLr#F`hUm%58pUFy+S=87TBP2v}dSC00c(?
z0P{nSppY;CVBHUupNI?#+JpD{us;OqAOv6BgMI!1Lv_IZOaFkAe&d|CHG*&+LNJeq
zue&D%mqW19rGF{+`4{*feSk0^I;LKMULhzC$vs)vgrL1q{$9T;{_*&4s{aM^{{#m3
zheJI7zTnW4E7b3tCG>d&;{B$l`{_Mc1%k^x+>OscunYuGM4;^d_!FiFg`Kj6;KR^>
zONIN{orGXWBN)xRLrm--7^1@%;u~)EPdNXo?|BEPTnH*}54`k;^`Gy4@d!17WC=(?
z<#b;!^F8@W002x96=r@Of+0FEgAlas9zN6un760bX=4ahg5VTCwB=qOAeykc(1>$;
z<$L}4a@E&pub+G6eV5%WY#<oYP1u~5zv&)6)E0~y6lU{hADAQp{r}i&7ix>?v{&eF
zejz$c5@EjQ&7g9qT_*pqkn?}O%arMjGBN)XzsfhnY_ELp-H9Oo-#QBQjhQhd+;)$5
zs4Zp{_Yh-K2!?1f+js@q?a6ZwjsWZ#+yO5D4P1hr9zY;K10(@kz!W$IeFgy`P>nZ$
z0{o#8FR0cFD)EB8{S!y|_j{p$A5@M4B7WC}LQk9DnE!b5pSY5M4^$cmEdIvz0E{F`
zfN+SCFR%iY`u;^z5-@_kg+paN{{@|Y^kwN!I%p<grlU;9p<9>f43j?72_|hok_o|N
z$Yj8D8Y<Oh(*29J4fM`GIoj*b64c5+=mbM8di}<;hj?#;cnkyFp(g>tKm(zFQTR8h
zWQv9#`MV}Q1c~zK&mjJbtWeOKet%&o|C>au|2vs8e`Q`|9)X_Y%u|2SkYt`@o`U|y
zA=&)R3;OT+Kc&1kQYdId(f@|`AIIXKSSKM){eQn107(G_@$B_4JS+bzjh>jM1An(R
z!t};<h^r^YJ1|r9@6tUfcqRTp{sV|t`n&Y#*wNLaBS)VcEk3#o%YoIxnqhBY4X{2y
z64nZ9gW+J^Ff6R`FZ}<L4*fOuz+TVy^lGo)d)oC+4*ylwkjC<y;t}JK<~hv+=aJ<(
z^B0yRk1)?M9y1;}s7CCsar;}pCI6LkNG?uLYyXtbf8N&KTtWRs2^s?iZb$=Af4BH=
zI$#I&7zIQ^G(#YH@9A40a0SBs*O~VxM@$k-$C%9jgLa?T)A@hWW-(?t&2kcuWKm$z
zW;w=Uu?PQiF0;r%b=uI=_+R@5>BnCf|Hi3Tm{-)EJpqi+L03XhKE7d+YDbT1NuGpE
zpO>V0pofZ*q`SYr<nO9b$q=tluaF2YPZeO#ZtlGY0GqbI?ID9u|KDX{Rsf*C2CaPq
zf0sFj13-Bw00^u9U3LUoD+LPx;Ekq7cu2%Q=s|}1BLIYYA6NVRoq_8Vw0_O(?ruWs
z7;82F(6V-SDNlEIX+;qIDFA5o-)#l>*qC0ivcVXn07gCr7$3uKCjglm24;re*Pny|
z21XbYGYcylI|nC(P`eLcWPrgKnPAM!OiU2=HK-n7;$!ANbnGOHz*%=zso?!;*R!6m
zNuPS%CTRPaD0BQ$NIW}-knjN!QQ5<CN8}aMH8izOXzLiAHa0OeGq*T*-p<|uV$Z|V
z%iG5nbvZOFJR&mc%GDb;ZzUuqC8uQP+_{^Zckh1w)1u;%XQgG&UsP4s)YjF%dE3z5
z(b?7A^X`4`;Lz~M=-8L>3Bug`*Kgk!7MFgIHa53@lDENMJ9~UF0I+|s1wH@4*?-`R
z58{iFi3!HUy2lp-W8@y=d`!%Tj<N8cJj?1HEFh(Joo)ZAtS7JA*rku#5(O`XeC806
zQ76cf_E`JP+5bDn;{R8i{l(bdd`&{efbsVQV`PLe!(cFG7G~&TVPo05*f`jJUmSm4
zT)!`#y^Hs+YZoHI0MTG#Vq%5<_i?gw?)x8KyVH;x8N53IaKRWLWrFbm2!KX?nsNmA
z|L^~+WdPe9_)o_XmW76cd<Tla94q^C@!Ve};}Nsx3y7~`n0EmqmHWHE^~_y>6iy9`
z?T1%i+XYt7QBGaPuy3tKAb36=8Acn7C~*fBy0loVI2e6edpn0S#}ImARd#`=wH~`b
z{C&(XsW?O-P1k8qBiv^YUD#~*<KkUU*m6z&AolyYIV~!`961d%dg4ko1kcSYq6IGN
zsA=?!Z;^sae;DuM$9uP3Tm#-&A9HF@aRF!!y8u%at$`X$@^ERO6&R>rMpyV9b#9(X
zec^g=+SQbGhvJf`*ZZTi&s6fYUC>HU4Fl!6@qQZy=w=b$PNr%+bE|yt{PJjS9=hKD
zxb?vh@gB}w-b`P&=JThpJ-m~=WOi_zv_P&Vu6M69yZH`)7>iUhbK8v{zLn;$-=MrE
zl3~w#3-T63^n;PFw}7+8`~NpLLjqlnIDEB$ER$dR^oMNxwR`;s9)xmT|F#ferVSH(
zi|iZA*af~FC*#QKi};<BXylMGu>Y(vAei^UFaL0&s8z)ep`EK15k;;RSuD~I8NYgt
zmc$;~1=`SvuXm^#yTGfFF?x;->uO2KbOo&@GWWP_S5vQuO!|!?`9-^glM-{~4!xLH
z{nUKcb_7MZ9Jj@>Qx(T?<mrEZD-wIitF($(4KS44PDHFM6O?@@tnJnq!mo9MPA-Q?
zoL_?#$;!Zz?Um+=)syRowiyB1r$R1WgaGJwwk1H-dZa=HEfBFN^;qkUKcOn6D?yjk
z73eUZxbl+1MD+)iDV|X`<`+`2>Zlm47nT)G=BRM>Jj2Usd8uImj7M4#TAzX^H7IaH
z=KW{??G{{HTUAmnZfd!&TtN_fY<@sW;r2=oz*}){@aIQi*Yxk7H8`)GI`L`D?~1KW
zX(juqAiwYM$$I*k?YS{poC+=z+usjLsZz6flkjBm`dwhqu8?=q84`-xPb;+Xz?}Kd
zmy~M;+G3Y$JD>Acdz@f?$24E-Qr%QDtX$Om>XAyocd=}<dpbvd+PlUZ-2Jt%3owh*
z?i$F_N8vnriUdT0gJ^j3TIx%Ul?B_!X41zFNGQqQmNyURJ0Tf7(~?iiixmMAXYnhP
zS@bQsQufRRWJi98^HlB2noJSFdymq3Z{EIo7j6M;H~d!v^<hM#S=cC9ZWOz^1%3^9
zF~(nP@iA>ySW@8j$aCZ)B=IbM{M;OJ)txxs-XZ{c1(&+aEH+fTeF~TV8h@Pc`ds_H
zxjVU29r3K<i?A)VPa8uvb+^gwVQ=g21)57;T<jQe3Srx^%O_foM-mE>V!7%4Ei52%
z?!EH25vq2nY{duXMstQNr{phvWo8S{&u(F(3mH?66mpD3=H_3k$*bQiscn)DeXq!5
zEK&Y3j&<|}>gr5t<>}4Edh*WA;JQ(0Xzc8}gDm(NdM^KRA^+M1a<steUErb<W=kD=
z6Em9(-#AT|+y%aL!0FkhhVa@jXj<$|4e!hOqZv=y^K?e2qA`C=3a-hT-Kjes=Nk71
zOR#<Uy8llx$MvvZ^9}WPykwJa6!F~Wddo`0_P{~qJnDn~c+3h4nsR1=FINgcxjgFq
z$aXA+8ArckX$bfEYnCGu5LNJ1KN6OF16thTqHc7=YLCWXxqXF1r-XF$?;RaCCHvib
zYJNfOOw}suLgs6cLz>Sgk#C3pH;csoSKndS{rsPsl3zZMQ0y`9x7{#b5H7pGz$or{
z3SFUvIcj~rrvsaDwICVcBd)OSNJ@EhN~XSg=)w6rlX?Zs=<26tW(Pz9nj98YsTw_0
z!DV{#E|7#FCKc=gS!J*t)261Hp;b8?4w~2jStYqn2c%3zx%>KhplCq0HQ=yOO5s$-
zrtbo?&+1}jiDvs2agzqJ!-OLS=U?S=zCHlEeZT^u$XMCYmAeZN9pDo=))-h6){G?E
zl6p15IzS^Gm#8eEMaYemzwDK9@J>W9{rw5-s46JevvLT&l2dp&R*qCLQddEObp{t}
z@lP@iI8Mk<lN~*S%D+vq-X)jG+8MS7UgX8L=u;y>4CMq>eYoWa)wID^ee$5`{$s$H
zaHJaS+hps_xIqJ6TfnBTQz09{TY*9If=7H-edaFM^Nwq^Ej?X~IP-cferEJr;q-Rw
z=WAux1HK<!xcp)^jFtH|k}5PdC-FR*gdpq!?Y<?!rKo(PWmVCFT+7tuWaZ(i+vUmr
z5etQ(x1Zs{5k9z^5^#{0=+}<p)YA?;bi=qRUCtlaoPL83K0g2Qy}Y-<{nl9qL-?(R
z{?!s_sX9u_3V&p<OwK3%@SEWNI%7i#uXq@2S>su(wPV9qUL2y=e{^2X9hfIm$U(HH
zEfVx0Jl6_cma2dMrrNRijZglZ55#JKCIMy{Peu{PgKC)_m8^awGy!v~<v3M~#M{0L
zh_uQL#u%Yw%M*d7GacKZrjN5vgJJsjjWK$7qA5aTxrKvP65~OP-7yhz9?c6HGAg)(
zTsl`~+P0Pc{oL)hB1?vFXZi%ZUbqJv53k)$pmRs@Y_!9<F9-Xst66*JSXixgn5xLg
zNh_S#!jALufr-Y%yu_af!KmA-O6re9C;UD%CLga7iw;vWm-dk3cwyd!(S#Fi@xRPU
z<9C7O#6-l_>*eh+ay}ScL3Mz9j6*Tn#7X9nTAI+t#DEY7Ht2KCLI0k(%2s)~LhVA{
z?tpW24rWRqeFVX$QNT&>oDp!*XSKK_{npO~RT=Y4##DK_i*oHYOl8nkk5OB!fN7L9
ztpx#@>I*n8ntn_)DaSg)iP*^#ANjUsa9vq?rj=L6N~U-qzVnDrxN9bZOv<lyGC%Fj
z7+DfWLP9c3nc-=EGx(E}r2M7cF7@22rt2~H`JO!NmT^7u%W2>VQ`(voE>Q0nC79lY
zsHwm86AoO6n)vidmEYN_bb6`bOMsd~$`Tj)K=!p))oQO{AF^&=4K)<sUh$(iM|qH&
zKv%l@;L7dvilyVJ`Yu*#R|<R{mW4;*ff_aJkxp1<Z^-$aje^xlNMrVm#WI7VA7<rR
ztt3?WW6YX2?7x2G%47H5cDHKcJ1JtH|J^S9;@gSBN~7F%S_MTv_5+*|=lxROs)b{5
z5LMJ^8Hhd{z5D=ItK725c?soqto~!ne!EX=p<S^|^ojQ*VX`^2q$Ffa$kznFKsDUj
z68)4X+lugRJQ?`Pby5U4B$Kv&b|iCu%*7O7hsDZ;&UqsowF|KOM|@;%X**Ko7^5k(
zL288i1bI(SqG$^jkGu6ygfMMq^iU+>xf*0OZdJ9g;|W;z9pzu4`t;Vy`<6uQnx?wU
z0u_4&JEw?v@KoHxu?ngJ)Qh+DUI_z7<(LcMWgw?bCh+0;r21%Mzvr9UO%pu1*o@YS
zMGfr&-ncLLd3={6=JEvwf9i!fvo-(9x#5~-H~aeAYps&?nI*-&k1t*E%bZ<6_vs9<
zjyJSaJo6Q`TWhSCHuWe)4ZIHx4L;74a=_ho!)ztW&W<TN3UHXM??QkI=<fA2B;Odp
zVa15R(&QnNEVxC+`8PK^g2zn*zog7CY}Z@8nP+Zlv2SUHdi;iZdYH;iIpl_NKL3&F
zvD(N=jnf65QHrioeMs8vh=xgj_i4u)ZIwtm8;yuDBu1|E;RHSExhtdS<fZR#E^1Xg
zE6~e*)m~a=NP4$0T(8Fv$D2;IBR!yV9wHt;lQ(DDc}@Q(FRwFpLZv9X!?dYB#o)Ww
z5b!2>|F!wIh6i(o$CgxAv?<r35ah-ol_tb<C%J^CYsIGyHx8YCKzMuOkX2WaR-er=
zhR-d&A9x(aM|L<Cv!zX?66~4wnVe)T)1bbJIrTXJ=Ya<m>>uheigp<70yl4*d3$?*
z`-`QWxo%bE;joQiD<xlDX195Dh8@!HWvsMCf&Kx<O_iQQ-x3z88=L4ZCr7fZO$VOu
ze5Du1eCcQ;!^aq(mzXv)%w_m(QA|RMXzV*TzJSQCpv9|Ubg9wNcV}Mzu$?p+vt+WJ
zn_(+@*nlnwY2=ccWx<$2a-G7w7Y3tu0S_kJj=AhL$kH)&f#FX;PO>)9YVCo479K2-
z8^M0(fb}bF>U2B))IH=lfN6aEJq{sH^`kgq+Uzl2jgxSJnDAP^bfGnyCmQ0qRu8PR
z`VG#$9US2bY&|HKUmAbvp(w4BmV-g5^2UxJS;Fus`La)jRGgiZev0wWokLw?;mS$`
z0`2Xc_GPl0xW<JjWUST_J|6EvR3W-P>e^rTNx*kF-`!vZZsoVLG~DWrF3*x@9d`I`
zP2IOfp0j49It|@_AJ0iUZ0AapaI6$ek|;KLyTbv{#-YtVjD#WMXwCE^^iTLYp9BMu
z$|4^x;&h%#!zkxC`<l?;l`n2;Cw^^ylb-<O^tG`PBN)GRP#(VO5PEISdR)3qAj+>b
z(^4j)9_jP?O66I0rsmOSF?l(X-;~F`BILcLgMT&BIXa5W=Ma+_R>UpqZ!OPDd`Dk=
z`81?gC3#{!ZETA`fO9MwFca%?X8SpvWfQWVOmz657`c>!TCekA4!vKyqRv3w1*kmG
znB>O}!r2Vejfkn_;Tmoa*}jPSF$epzj>si%SEe^o<I)K(K|J53)SP&>Lenc2TeN9!
zc7XuIL}Zt9O9q7n)bx#kKm8PdtM{y}zY;j*^L<l;%elS$=m)-YM_iBdHTsfxLFKx{
z(g~_37(I7D>U`sHn0Dd)R-j?h^>s$@dWJ^gATAu^4Nq)PZO4zV6Sm{k70=hicNkZr
zPGLS+KA&@{^pa~dKi-)0<Ju7z5!;R++FDO)kpsZzUb6)g>*m+aS!0e&PtG{Dd0+&~
zGwN40t0tt3kGg8Ai6c=LdSl&b;waI&2iPS|=e6{pysNQAOp%(KCL!4!YKiH?+g!zb
ztH?w+(N$c`8p9f+L}I@cJIzfmxh&iB=~C{IhtKs}J(+l>=AUC*TDYi2EktXoKu-rk
zz_WAA?En}lQFx!OIaD)9y79R9*Tv|3;1`+<t!D90onu5u0o!fJxSi8(@}#cv_|Y|T
zz~H#kg<}%|^9gDaa|COy!WTt^Kge7+zQ5P=gJxjuaLrHB$GW^wlzZG<mw%XSp1as<
zvbkMl^D7h5KxqCZ9+b{&r|Hf)^dQ7yP^7qkbGF3rCe;9$T>wEnqmlR^{DaVYHKr6p
z-j}&Q7_dujaR?!t2L@DddqQIE$;C4vP>-bGT;Piz^DFlGcg{AKtGTfpQ4U=FCewCY
ztGAqIZ~=815~?#j6!Hmc@N6~KTu}?H7_laO_UhrLGkKb#Ib9p91=f7}HRA_=mB@I?
zQC^pp#NLXrp%pe6LN4MNG3;%(XqD#jD4kxBKv+O{%WzYLfm7z~G}*h4uJIg!eYKu2
zqI0B@j5>FL{rFkriqq3)*ccV(_u=U3YYlT7-!%`>I~e-hEN1Cs<BPqVzzdj!CMKhR
z=ztiX88<i*<3}7PY265@>o0Tp7*L|q#o*ply<}(=Xt=n>aX6$m$T$*CJdL|`RW!<|
z2Pe9w+}Y5yy<F&G(Hi@GhV4?ry+g~C?tFdxnu_kiwwvSA$8zyu{HBr$y&Vd@QIa>>
zm~2=#GG_6qGZNrWXh=+n8C`fzmwwbUFsD;T?{h~8kbn60rvt?%6K1-eCL*j%T(@g!
z7T{2~n+SM?pzqQnp@z$_{H$(yMWu4h>xAQ6n&(3vbA3TW*<|SKQIOiaxC?NEVd61F
zdn-5rm*{a6r4yODvQV9RfX@YgIxp)qZl$}L$qj_*jV(q}g^K1(N(uOQ10FD-v~Qqm
zUfkJfjr>@)e(<yFY5dBsq0_s-`TY#Du}2`C%ZeS_1#0k}iv5iAE{T2D4BE#+O@qm1
zXw`{*L5JUFFNZ&>YB^!g$~52Sk@A6rSQUdf^rPj@X7s?tYHp3`sd9F!aF0B@AL<)#
zG-$q+UUka+myzx*?l^{btfSCg5;?I>7NO<u0zUe@?QW8!nybh(4NYik;5J$OqgwaE
zyA?iB#|rTSMr@}U4pg3$&Xu7hzkocpGq{W5p=TODhZiIT&wN345?xPfCcbP6=~B2G
z7Ov^2&}%07b3el}zDjO2cvf{{nJ^Dlx05vUDp3BJmv_!oO1Nl0^s;yb>q|HTSFtLJ
zV3SnKMCd^(T3@hX5i1I9YA-ME0_HF0;Naoy+gpf4NE2;zGuUFZh96T}=_(JL2P<Ty
zP74d|OAcJ|^8Id?X&`xWvn=@S^=>n4J?=k{DT8epkwu^lv>^yttxG@i=+}|<|3%bC
zdtK;<mJZ^UMupqE>%ksgp?Y2khEYY{uD9o<aE~)?Kzh5LGWwXEbO*<;q}+62-X*za
z63L0Gi%6Kz%nlAEKM&+`@;R*t@KuVB$5%3VBNyf7gsDjevQQ|>Yk{78mF|a|P1^-{
zY7m0-&yeWTe!lc<$Z6YAwDJ@;y5bNSDkJ4yFz47b0$=wjE97+^i#>vAyR4zFgHDQ=
z5GQ99zxXxRJ!Ua@ChLjFL|tfc{l^~SD{r{-jt`xE6B>UhNDpB15uiA!-nzDi<k6<y
z5b?w}VOmx6i*P=B;iXe&%0$}E$lG1CGZVvk`2D&8-YA}k5~FiU`*;z+u7~Hqp3{$V
zkJSi|1gt)Lqvx?+v*t2(I1W*Q6pi}Of#B7Xdl@y~?Lb<(dN>&^J2Kr@%6af?TF&+N
za!<>cUP}Vzj5t`#*}aa?u)6>Xm*nu6nU<m_xpVDS<@ek3E15IM8p&C!P}$LCU9Xs+
zavt`cdp&~d1w`kJWYL1eST<0XP*-r=NH{GV<<mG3m5aAD4i>;vFIWlCY#anBjPx-m
zUbiwv(O!(@C$Xj%4N~2o*VONjr{(PKPpp|<Od9O+U2H#aN|LQBg#F-a8Q6P0Yl;b^
zwa&YM*Bfb4PxtDf@P5vAQYz#lKXaXs%-qVpCheb@9c%N6U~~}(VW3^3Wg+}B7-Gke
zyzs8a!pSwwewLm$90t9DHpctr73|_9Jk>7#3JDN<b&}!E^{+Zu>i!R8hB5UI&B9J#
zqWJ19sx@<`fbZ4D9wpo74yNb4;>(NUdCwwmV~AGk;)@twNYx7?CH8^n=la{%sTVYB
zOPnoEC#fhM;-CJ~`?>F>WQ^gB1)KR$SX!1L{4;u$leF1dP>pO~wVo^O6hW?N8oS5O
z928ksDe1q9W|LuDjW?I$cyf>U49B&lI(eMVfudYkc23`3`CMrF#8)VUEvc%u%J-1m
zcXnm^3l6=;AV<^cH1QkKa#W$7Ii8o#spr9HV#4^naa6wLkMHVo{H{NQ`CXoUTYHy_
zS1rBwWf|E7Id^nHFnTtxy?`r5nb^e03Fmb(MrX1td3P^83ptjTd+Jlq@Z(_y@NO}t
z?bI&dpD~{2lGlo?E4cZk?xjyi*s7zmOP9KXLYj1@7-CDe%--s$aT=>lq}Ln!9UPtg
z7R80?L3<GpD{Vyz{PF2qliH5OI4yf5pf0s7%xk*l(fGzWiHAF5gSdKZyFTl@@XB)%
z%xq-!=2*Va{qTyVDnRcYHtdUA6j@-XkK?n}i$>n7)B$K26~*)<cB6Nep<n@xg&t>s
zr!`{G{k*a7>~$~+VajR<{-|x2n^P|19}ZdiS^D1*7M3*#5gC4A@<z2b*$eQ2W|bv)
z3)1Kku&2a$9?gnMH2jPmunypuSj>;|bbo7M`aCDR_Ihvb3F(Jpi}-Hz7tUGLF6;DH
zv4a@C@N_#3aTPzYx2s+abr%jj89Eh^;`lADs=ZzBeX^nUrK@94!D{jfErYt>^8$LV
zJ5ky=^MiX(_+h4SG5=hTg3AE)auviydh9f&YP*xunu9(<KPdXK{HFSVo6|}6aM>0^
zlu~8+UFDL=Aze?~Slg3dV>iJE-4kP}28=)O`{*opMt-KdbmWQVC2Z`g=o%hPSI$)U
zs`~Meq0v5Fer@c+GMyuu;tlz?HDfoZ$^_)i!I>WuxWrm*l|PC!vg+z6G6PkXq9Q5%
zZ@=<sa-B4D&B;_Go6^~b^zz`Z9opZ@%HBf_$r&U^s$B<xBd}kuXixQwDCaNZYBdup
zVlIDvIz(6OQOWkN;8PwE8MN`g?pW;G;?t^)u2kGnpryn(Q39iKDP1oqZc(w~JTo)8
z(na@*YBTyGj$bQ;SARGh#Ds|!k6K4&<A^r!U*=<b@ri|Y+>UK6P0faNu72e%B{=3P
z*haYwrp}_%Rh49kbq`b;GJCl5jgJZm>`zp?krSu=vkVsqpY$MS?QJDK@*1c3DfTB^
z?TgS!3#!f5=x{EMSEK|RoZM7P>kZyW7O03xg;XAjyWYtMd+3c|=`LU)iaU*)t;8{5
z2wwD$mL%RY^_$@tiNdEqL6OkX>41b))2admd;MH#L-=zY?-KX&rMmo<Je{d=r13$b
zMWUjNZJB4dL0T~%Qa*MHQ=>n&@}8mtSvtNg)1;^bC&{6$ut$+aZd3J4&%4)EWV2l(
z$5WuK94L$J&H!C7RS~yJko^mO60oc-u}tee_j-nvN%08Uqrs!m_a=!{NKxY(V@{67
zD#cs@uTYetBp)=s6ky73O-$Cf;5Dk~I~Cd&w0iN{VQ7blgqT|caipR7g*MKNRD@Vy
zg_lmt6Lp<a&Jz_?wA*9bwiXezxy@wZ$q(CDX4)If?6V*$E9pFrna;Xo6IImn`A*8l
zm5vW6cl9^pj9q-Z`fW^JM$D%SbsvPN;`1ZYpV0eo)mH|$PITr+cA&W}mb&q}6wQ(1
zNO#(j6Q@?ADBT+|TYwNYfPsqS^ijG_?C^}V(AdDq;K)*I{YTTwRxj~Z_X8~^j5Njf
zy^o*e{v;CX#RObSWlh=h6m%*dw&>JWHN<xt%32<-8j!kUg4e`XSrniSEG0->-1**e
zg9h2z+0v`nPE681Dl_ewx^I)}ywifFectPC@~e+g#n^F`if^)okM0H7@{D0}E^7|p
zEp~y9&p3C)j(ou{lTpMk4x9jD_8AUxAAESz@B&S6k{>ygBsu&OUv0@!t)%Ht$CK@>
z7vQC>b=0Wkl*m{}cB@I;EOR06kKUAxjO!BGZ^1`I>sBP6DrSDUGhgUYnROf>;5u2H
zoyKz_#8mLd7Q+U5FNkzFn&d}s6-K^u*lebV(#K~sX^-ea%Xp3!?kJ}wD-yg%XUgg#
zR-m8jSd!3D`vdQ;JoA2co?yaxO!^z*Gw;8X_=ctOX%TpCgrethx1DdjDX^<~tk2PI
z0wnPjxuib5PuJq+{~EHN?{hj3BB8K%uE2aBv@WtFo$Dssek3fP*V3{P;}-gH!u?#%
zf|2L5@9JkiXr90R;b@!2jy8pbK9euEdadY(1bnRK3l%Xx&A9*Z#-NaLK)8SBp|^E2
zmM;<yxoIldR8c=RW(>^17|YdonMCQXwU#Gvo@h{!8`{ZQjf#rnleWN^o9*4|&=lxl
zf4r}mD_AU-jn`V)+J+m5t--WeLLLM@1y{A*Va*N}5sbSkYRRu();?-$sCT)0G&M34
zC^zBxc5rMMQ&rH37Km}BJ&Iu_DR$c$tJKbkYp7IC9OyiIU*w*6*@?3~7Su1$1c5~T
zo*%?OQ47vpU`eKS8@&Pnhtb3Wdh2V<BH9rC2+jxTO-aE%I#=vuivk!-Vp$t+5dytk
zMxvvmWwi$H42k+V+WIeq<(a6zZ;y+;Bz*V{I|hXY6}DFu#lVGGxf`Cuf*X~b&Dm{!
z7xfkyemT1Sz^gAD6ux!#zygPkL*FiNAO==Uh1{h0DmRc5ndFMPTA#YL)pRSt-o-OT
z<Zv2ANB-PT?2r5d?sW;e(NUvEp4ucOF}`?_1^}~w4hKMwie-a*8tYxacm3urAY2j4
zA3N@BlRIHqR)G$zRLZd$u30}3ms15LttbIUy*n@7`IXggLQlHR6v!&MnNN>|czhFk
za7bCth4`Q=v-M@!<p}5Ihl6oidiJ(1=KMKR@^5qC=a5g3;&gex-!9AjZ0gCOH3@y^
zg*Ip9sy`}{?Jqbh`W;uOw^?4uV+vvW2*f<1SVB=9Jxi?&{493m3So<6P)o+W=DtAi
z8)<mM$}LggCqi3%C;c(R?v=$8SM7~vj~>XmcgC}!)=8XBLxFOGPwn$vPOPZwK7DPX
zwQqE^cg)a3-vASPcM8)cK<8a#?J&>+*P{aukBU<0d=%}2u&+k@kGW1ApnX&8KCp9`
z)xLjS7KlAZZ!aL)Hp9ueqXGnnE^!@v2Vx(W|7b;EpHetNKGN*9(yhnxcNk~RdUF-t
z4b;qljyW&GlO&|Ua}@okU0@Dz)9u910$!AsaBjL9y9a-_T1h6cV*>XW2g)zR&(j{_
z{ArwXV;x7s8Y?2itogCIn#ETuYCja0$3<jzW}jI5M7yoHElutnnI*BV7?Tu{wao6r
zP0AX*u=kwm7PGM?afl;yacB&k47C059PY%&8H6uQp`_|goso|%1!zAH$-26sY|e-j
zBQyiI9`;bd<(I${OYI59dPR}|IS$gDR;s|@E<o^&ITIhP%H6#<q`1_e|FPKY<1MGI
zr|JXG5h@Fw+LD3q1sCze)7FVsl{@kLbOkq;;)aqZuLu3Vd3`Y2|J1zlRKbp~z0!93
zjsUW57Z6j|M|9#@sNsZFd)fS<vjabm$eJ=-cTz59y|FYLJ;3_)*#l_3xG`939d|Y2
zWeI3=86z4l);%~!q*Z;CefdeyVy1F>@ZP~U=i`puEjS_u>%$i4MUy7!oMN-hQVVna
z$y<#T!X5UD+X5$ra!RbY%Q>`;<?hR#?oQHHm463(YZ&8(tdlz!{S<sazDbfx|LJ%B
zZ9%(^yq^^HmzG0SQp^Z^W?%9-{Tq)#>{5Hwp@sxYzk8#OUbeSy!xR;{_MM!KKOs55
zy1+{n=qE4bk>3!%V+reGPA;fe9b$vAZ0|(wKBTkxM(VxK$;a+ua!*gS3fZ!8*E&Ib
zsnSYgp|B~QTTgB-<Ik{Z<!B_!4=8Nft~_JN`1*kEols<60ny?BBzrMhgI)k(`uKK)
zWzn+z1Kle@$jLjfrB|^cSZMn*3(0fibPF8X(Lx6nJLbwjY!1B_!w0SBq2zF)a3@y8
ze1F4OQrK`qqpt^_iKq9)h(dy?xZ);;Tmud8qbOVlT(Fe(80UrL?wKnPa&d(W(G=2r
ze_k9OmRMI^d-H~Lx5-$ftPzVCn^?va?$y!$l>p*HSO%Eg)lcPK?llmmcR`Nh3YrAB
zp*=xEVS7|)ICEG1M9kT;@rrW(={HxrZri6Y%7)v!R==L;(;Z-K6VAgEU7eH>1R1^$
zT#hQrPg6S%8=pF!A(@c56CjDJ4M$dKQTMOU{oL+O*CJzRFSdLGRDW7_4GGw2mRdE}
z49onwl(O-Ka7>!Oa6aw%MKi^rE$OZT;@N^P^=)lqHei8!w3H9~vyRZRK!fWQm$DH2
zt<htfJrvIv4CqBU6+3oSs|$Tob31pG33b$3`71kGX*;(%YeAvt?c}#B7Mz3@*Nc#+
zxuZ>M!F>HlSAyg#grspR65ybeDK!g`gdiV<)|KL{v1xpLKlMJYtAN7zVHbE(5+e<%
zmf0?lEk|OdcS9q%m8}ENUhV~YXRRj-$jii@6%1igT&KHB@8o=QtY)6|(4&;L{T!24
zABR-?q^Iv1yQ-(LF`ugD?Zb0Wy`VvQNA>wU4;B&0?UqQMY5BLcDZ1REqasf{da@)6
zx%<jYrvpA_y}M*4_fGGEJDOs@w-pW46Q5`HYZjX9tVU({|M6ARO)SAk=7c`~xy#>L
z=@h>AA2Daml0}t*%j}?J@{54Zo?mT7tHSw16O<YS%9T(Bz<B~H<`~Je%}v^T{jR&O
z<YVUAH>ttN?|n<SrM#|~GU@E&`AmUANK6}0@IwW(gCaiADmR$yw^>i;^daV)G7uuF
z_9&V$=`=<5bZkR$b9|e++Vsh$ozBk|!sDN=a?l`47=I;=-i70cJFyESyFe}`nty<`
z3sYN4-M`#LKLA!BR(?_xz(B-)uxes~VlcO3R@}q_9@MXWR4yYTp{CjRR^h<WFNRYw
z&KnduXR2%|RVH5BJ(anu=4_PbyE2pDulok!R>i_(Y%~3Ts;gs>bN%3v{uP(m+2yhr
z>sQTL$8Qd~za7?s8;-s<D@eYmzX;pJlcC*z{3EJc3;6jNy?+;|bxY?ip+ApmxOACi
zc&WZF|7L;GSvQx1hh<_U6|(Lsu`^$d&5a#5kffFAX%qV`n`t>UNwrv^_b)rdr__r1
z6_&bZLZ#>C%*LeL_r23zcy>JF`z}C!2SS$Y1NkPcatob;XA4+YE@vCO_}Jwg|BTbR
zvXAs9JJZnm-i<rL9WifO0f(QfhVX9uO7k3iqR50+96s$b%abltLQ-?AJ#@`a%B4X{
z??>P}r(!d@(@_?&r7YwyqE+=E)o}>n<ER%+aW-a6XtFNpD!Z_(XME6}TsPSJ{?NYT
zg>WA`k9h0(uGoW^w&PIll$4v+E`UWt1G(fMPUE0@gVqH6Z6sGM{+5AKLsRu<su90Y
zsO-&`-2J)PN{wfyWq@ObVhSO}x3?I;<T+JPUT!s%awf(QM-*PMoGqySGO;EfE?eUe
z*)2Y4AD#FX;ZbY9QYpqXA1$E>>zdJpWIaqhJh(urPxW})JXk*V>8+#E7PqmTa4udj
zMX&dTwznZ5BKtZAS&C$7IY{qC60LdJe>~=U(35-h7>OBAzL^u1ake_u=8+v!aQ72?
zMp930%B_s}Ptl{M>4@2~g4f6!ga;o;9bo+TqnOWIjhxTNX}J!!Paw3Jfgp$)0z4v!
zU%fO@A7^lAsjk5~eM{G?$)Q@QZsJy<`@w9PKl(nu7gIxD=;I+e0Q(R&%fBu_oeuPy
zkf?<WfRN7Ejn%KO2Er3b2S@EU6uL=H9}W(-i9HoER%krQVa60>_8s>^0XXtB<^TG7
zckq9b!4im@g#wl^1I;J_@<1cWI-L%WO6-_WbX;artrb7KHaJ{U-U#IT-c`f5AMd|R
z&+^^NmWBO#x~CGL-?>4AuQe=cP2<?a<;UWdj>SDLj|?`h$h0lBPZ@V%jFCJ|0Yx$9
zMB&6H3}N|Z!%&lvJNg<^*2Cvc^Sl{0=O6C7!nE%SOUo5nHO`Bdmp%q(mmwN|>s`7^
zzc1hPxmL*b13F9?We$ApP@q<_N`DJtH{_j5?BGp-;s#hy?x7Z(D||h%-aw8R#$WAX
zs$bEydQ<mHV{Q7+Clm3fENjpY`K@!yT_oXy=;RpFry!!Ep)^&`JR(=;k90qcZ_}NL
z?E8flJ+u0&$`0f5GeDWZN6*1cau3O&^UB4;GyP`6Sh+h}=07>_jqEE}KZ;@_pL0P9
z(4qO)W`saD>&FU$Mug_ETQRaRKLWdIYr7sbcSc`7d!@`x&`5MXGEZWfHbMkw@w7yY
zM>8+^B<N*ubNlQMx(bOYYs$|#&B9(=;^=9cQ)1sY+_@$#0jK*9Z%qjrBE&GWCAf_&
zYvx^GjF$(q<T1A8PJ+)_#~bj0S~lPflJI#e;f@JpZeZY7a~ZRyaUq%B&Px38s<KwS
zEF=`uw#kSu5Ya<XqCnGb2_B*Z4(qEqu*i7fREK3lLrq#o*s3IIF1_eNO2~L1>tS33
zU4g0vC5G}QLhiErT$e{NnA__o-xtMiBI~M@mCeR0=01>Vcu|cTh^d02zsZ={!WN+h
z60PNy`@D;b-*fe`k?fgyE1Xl|>3{@(m#MpUSsaPIvs+Rv;vi(Una+clGDo+|NLY*)
zw9B+eBYyE?P+64;GH~o9mD_ffQu?a07X7~A*g#b>wCKj8qBfmxw6^Ro=rQ0J`7Zmh
zws<1pQB;<vsjlqv<4*mrX4e!SC-&K$b9*v}29MxZ$4IHBZ;9}B>l+R_JlL`#m*&++
z>9lXdV%Hs_QlIv4auL_WeB5~0fC{KH;ov-pzh^El3Fp1eKhqygSln8EK(V~?uveYZ
zs9kjX{LP0ew~aG%Z6elf44Lm0j9NL6Z_zSiPf(5C&~n39E&m{4lD<7&o4LlnZS~mK
ze%{&vHQYpGV|vXLqS->Yk@5>eXD_0#gL!1%ni0A@)tRqn9?S2d{4nC|yK0P&W1f?z
zxvZ_-vs~j-6mvT^;F8m8$PBL%C|uOQ&n?=jt7wY6aqyLwO{5=MqxNUw$)ogXyV%bG
zd_9kIuHTnpxc3@;_wzd&-r~JOrM5!|FT}z7mDGdzrE%+?A5_WLKEJ7*iLMd8WZy%P
zmtvTNrBFS<3gYHGoGJh#XAi73$<IA`ekIemY|?cyu|D<eYqgq?i5ZvNAciAtd7;<u
zhF@$AprmxrIqp(?qN1%64M~ZuIl*(xhL8QeT(<W%apTl@ZLh80UX`{51iPX;3dW7d
z<+LWcfn!WCtz131Ns%CS9EkQSlX(BFs@_A&(aGuEnfLqCdSa>~A})TKvSA&^5FJ)|
zmb(jH&!o@M)Py#|0-~*kOW$eVp7=2n?fxc$_N~(+f6E$_>CaHd3dMHQC5WpPgH{*<
z=d`JHV-o!FapmE>;6RB4?J?FTZ6fb9wtG4ei7kR)$*ew_P@fjSJacP~(_E?Y5-V!s
z>w#W4A?$>w`?e#aj>G1iSZ4YZo@iB&yi<_U!oE~5ae{c^7o(GMfJ{qub6v*aV;CQX
z&B9#e9)5Wn_CqoXeZ{by%h+!{<f7q1a22vB=ZEyP38tbY%>_y7+y`o6$|wvs@jeqG
zOl`O+jg=Ybv^PG5j)#lXp!8tG0>b_s7HnDRcnNN3YkB!WnN!Q=slf~de8ya+W?L>7
z8<)Xj=!I}-DE0JAiAJ5Oq|NyuQ>_dY={u;Ms7zPQhk-^)oBcYuPAZ=E@*BD5?ft8d
z$HX1PQ-$W~X1ESK<p6XTHE6E5Zv91$erAE7MT>(%;{*9ECn376R1Gklbete}Jq({f
z7aWOkh(uMIs}C$`9><&S_Z}bCEoS@XASWU%GS@N!o$Da5QRm3Epa%iJA62hoP~l$G
zA^7ppS5<T{)Aa}PGPcZN?u)IWz#_sA`AY+xh9S#EK`EnFn+{*ssr%;X#_OrOz|WtM
zg;*30-}y$dUJ|XuQ8cIy12tmuKb0-d5}&;d@be20v(nrA;Gt%^|7@Ys+RCXHWir`-
z2(}LkN@uj=pp-@~b=8z0ss%CjH9=m3^U<M-Jcpq*C9XM{1|AP#wg>HN7bZi~(x5|0
z3Y5cY!@8*o4a;FDwwoTMN`DKYq1f&O?pX@`3Szbp!EB%w)lc3g=B35JXZ0}`29%44
zVBXnRgKL%L^GBzTn>R!i`F6K`@M<~x9=jMAhj@qsm7(;DI~eGYfXt5D1wJn!2Y5R&
ze#t!+hK~L0^5{txROnzgzJn?bz9nh(;CavnVo}nrB(LV%_oKx-^Z$4d9-G<Ncgx<S
zKD|Ag#qflT!y0c~K&29uo$x{T-DCFIA^-Vo?fox{$`wSMtf6yl))F8V3Y)c|abNGp
zs6nlCt<!TVu0<1Z?RhggP+J~Lv0Q`mYIr{kcS&{K$@i#*p+`~&;Hz0keN3Bs%K@qd
z_<*=cn9l1!v;Uwgy{Q@r{Q4%<{M8#=)Y(Ft3aQDWyG%s(*266{T4tEOUk9RUy&caB
zW>~)&thGXPjP9G7xUhHdt8f}7lVv9O`3})84*H410nBXDW8J~6ele;Q7#tmvh4ia)
zj(CC+I4HvMc(ca<Z_C`yuQSWf!qBtjgzQJJdQcdkOi?+~3+o9qS<Ujn&FHrp0`AXd
ze~5bZjc?{25R>{oVcVxyVt>f7+vCXbvyhT9o6+i@4&ev~3T9($!RIXf>PuIjQhPPC
zN8Rh!`sMf;9%~M;^izc>Jq_n?5E47YTuO4z-o*bH3xaJL3nb0fMUXSgXEATCy<Ebk
zVAwTAA#oXF<Cq^De)%j)*L?VU91Wc4HPcYMB~9rq`z2AB0#e9VXxDMR=!wvIwb%iC
zU6rd7DnQS{7PXVae(gEiuawD^_Zd~rm4N5Piu!No#Zs|bT9F=J7%$BS(TuLnt|kK?
zxKc#5(z{1Z&b8U{u<7v`!c!5%)BWP8$O2ZqNaEB~Dfw|uW%{{ya=c`}WnXt5)^J|`
z*7+yWC*Jk1-oj{M1a|>X_(UF=ADj*)eIT3GG?0XEe!n7eu_hTS8$XJIwG>?QF8-hg
zgYg%h;5b+l0{B%xx)^^Q({|E82US7rA(*xc6QP9ST0x3NcL~ZP%Ck->=u<}6X`!Zr
z!uRT?zN*!+0UH@2QT=B~V^krN4#&@x{69X8bve5|#ESIZS6_Av@D^E5_6RNrJb^94
z%+}Kdam_^5cD$Ieh@ZZ+?XbVB`$NSa`#O}a?4+7tGg`P}yAkkM*4QaLPe3@9b6JP{
zp!mmCyYEHViKY6GB}<J?cKvt4El%mVBH~3TPOhqKLp|Qbm-E|ga5nm91pCkRn=R^G
zxjKDDAKs+Flq>o?^|bgF;Oc=4-9a4+vG(c=zY;0Jv7H64c7Y!J_{%l~UoD2A>-loM
zP=WVIiIvNF6Z$}{$Nq2gw^wa7L!Mf_cyV;LjKlaElupUhM7xDqyw6KLco~|!50IHH
z`-#v&E-RhA1j@>oY?~2%&~3nuyYpUkeD&~fSZW`4a!~tJa>d~ZKsUt)fFC=RzlA(V
zpDC#CUlk_2ZxN%)fTq!e=izrX^R#vX(znJGs<<opzuUZaf#uv6`+o9WxU@P08M(4N
zyAH}`$6S34#qNA4z8Rf3o@qDb7bV@zhu;U-o__SXve?hbMlwdz$*j%RT<{x$=+gg7
z7Qu~K(STO{;*nK2(FzK9z|Ye1R?zgrF}A+N;0>(HgOh%o<pWE(#v>D|$3n2`X(^sw
zvW5)j3+~lJ@h9e+pce^B1njU*B*VF>&>?>1_YV$lbE9e=h@~z`rB{#>!s|^Hiizxi
zcbvf&S_0<lfC02={L&1CI0_v`?rxf{1Qaem7N+GIKuLd{6XBn7pwvUbVlB<QpyuXI
zeg*3L->d!9Q6`74h*cD+gWLfIQ$kFiE?tG<{O|;w`lK*L@e>cONDZ##ItCaF2{x@u
zyjNR2Osq81O*~M@yJ!IYK!V5UKcq`s!_6ikHqIG9tMwNtD1-wY!WFKIyG&gG<vKNU
zv($$^mR=|hewrH@aV!2s*;gn@G#uNJ0ZR9`;Zt#3Er+P5o57sfYoYBC+YJ{UMr>T3
z@EVCfv#cxRI5DTrW%%`aHZu?lj%})vhUn*XpFypR5maf1USXG>;vQS4;UH`9qkw4t
z2r*}Je^x(sz1@5c$@yubQ|H2Peo(^9r!MczX(jJ{Sye|Prf(^nY-5Sj*QIk}2rd*n
z5Ti`0Ih$KOsG)4qZZFc_@`OvRr=YCn@-pVq`j;Fs9HdY*qU7c@v9l*?^&{qnRwinm
zzr37;pAag)^sp&-`i04-^X9d7CI<x>%&<fK8F;1`Y07D;+*k`gIH2aZe2iCR?kHDX
z{jp5ql<C8m&u>^oI=G(yava;7Bv^Am*1-@no9}jPq@qcj@Vv=m_Q9_+*|f)l0j6e6
ztzYuEl;7nK1{A{HJPTsj+%I3qJB9_r7giK^fz$6Wv-1wSz}E>9a(t`}PqdT)e|nVA
z`^lf-$+6-MO)hqp=~u0?CyJAGi!PXrRuvk8{#9A=#o3DU*8jsPp+R1!3Sh{r!x%4$
z7!+bE*q<+NAh|sHeZaYWSzX5~D?B2#&-LX;X{q}n<~D$x+;=60Uld58ONp`AC)60C
zYG)Z=QPs$nY;}sw3E`^&!&N8x%Pf<U1Zn$XG)XgwBt)x(1ei>CdF$<vp0Nk>77uRv
zJ}f}GT3*&s;v|!Vaa8?kFy&~J0~re%Um-3aA5aiyuXoMll--;mN0t>U@{JS)S(x1y
zIhU2dVLB(*F^!;eWKlTu$2xWa7Eouo+fCR!-!e+FD$i}`569eW{7QNKKBb#$o(4{G
zN;b!!+!-_m3fm<)n|8p*edf@~wSFbFYQ7GmLN|X#PlC=9<NL#9K1eNy12RN*Y%dl(
zR7#DC@!kb0_X5WQ=;dy@GF6p^L--+psG_-oRXvwhHx?pG@(VxviF0k_W5JO=eoC*O
zI<BECtAPZU8ynJaRhi6dFX?OnXsJc$tZ6OjmrTJa!)Py|tpXaG4eKXdooW|kf9kNS
zPZ3_KoR>?@h@WyQJ`_$&tOW-SjlUMot2ayBE>Do>fNh^Ov=MpqV9Tc!PBcZ2TXqPM
zSLa2-u_6~<WSp3qxpsScvgO^XkUXFDle<ZD681JC0RIK@gA%}#lrxqT7y8GY3o|&s
z@G{f#sxL<5myO(~3-B)|3RTtlAHMdo2Xqx@fsfNmaN_y?iFxN*Yc3SBx_o;^X+m0=
zOEkxL!Q*-LzWYGG?wKjxxpl*X4phO=+0vD53W6#^6uv<h0FBNNM_;~Nww<cUG;JaK
zxmfy`^1JrBMH+7QbZCpTEhEbi#6MWLDFQJ{()5_X#Bb3-illyW0xcaA2)`W*jO`^q
z`)NR#SOFg&iu&9P`{3Q!c)Q&xzN(A)bxyBi(?`1VuWY{NZ-=h;FGo@Z3{*vZcY!MF
z1ggONF2H%YMPRf^F&yI|+k*AdPpj&Y?U?RTM+GH)lz87B2&BL3&^`dbv++b56~t_*
z0jJz`at1Utmn_D%-Xr7SL_5rQ;(YFF(#sBf4YG~KLg^H$BMW7^V!Tg{IYc;&3Yg2B
zJHY4E=U;yBV!2>e{j6aB7@TOmKH*0$2fxi(^DSa(Oyj=~)X;RT6a2W1D5C3Dx37#|
z$S!(+>{v~jh@(_MtG%Sh34l5SIRi1fK(AGd$1ZSzdTtje;YJSnBvZAEqdGdS*57P8
z2{a7;Y&2h~B<Rb;e?NXOCH6(9!z}B{apJjoxtqAzCo_tq`V=ZFcCkhL?Ssw)Ws7ag
zhPtXuYuy7WQ|6-GF>Sr>E$%}aH0}Y`%>ZccyiD@yU&*3eim{?SYB@BMYNayA2^QyC
z-xa;aWggDlUuEvIaAM6!b+8!f+m%I~f>k6XGHQ}sNU~})C?G6VYB-Ks-}h+1siI2_
zG0V3a-ky8RW%;vm9?F=eD8yKg(vK~|MGZJW2S*aujO<YHtC6A8Rd1T|<yX(WcRDPz
z*e4W?Vsa&(J}vPEQHUX)=^v-qg6tHvrPtugGY>3C4yl!1BR_Sc@8v5dY0W7{$(S+>
z7au7<8>g@~oAy1<paH}aqVWe~gh|`+TO;^LEQqliA|RbqDrAyP=Nuh24?$eEfLati
zYB*RjhfTmqG$ADFuN;0P{yj6>ZF*eieZV1?djrlNdB~nwjxPfxP0U)kks~yW!7fmz
zN)=e1M|L362BDu2y&kmI%XTw(u2N8XS9m9U<>r02(Zh35vPzwug;{T&b!QZ^F8WY)
zHp!49_|y|Y41NinEd|6ZlS;{usN*eS`$b1H2<vHX@WhU>ofTVaKe5Xahu$jhKeMlq
zmOqD7wmc|*?MafqK+XJuxGJ3^of3euhqX8C0<DB=bNB9?mJK+5u0A4*WpHqUE4y=}
zO@t)`fF1Z|z!TdErJ=5`mXaD&GZ1y+6*D&obz#{q(VZ08>As@}yF3EC@lAF0U)Ox>
zWP}voet84X<6;FDXILO-vjfQm%@!{A4LxaesSCHP)|oF=z8xn&wy70l-k@|ej^(ev
zIPo-PMkb!gaXuZ|*fH1{wgNl=tnfLMT#WzsU!`I|DXrA4W$aBnpB`g1Z~aokl+J_G
zQ3h5&t4@{c$NJgwmWupP+=?dm4`|@`(|d6OR9!;4&Vs{8I5+=yML)|EZgX>24sgi?
zn}|$jTrA9|M`L_=$A`#^ATJ>SI{4!Sk@iDW<KoL&qHaZm3R9~$@jtfDYy^CFGJe>u
ztx}k6>gm3ZU)n@e&FoSfd?XyBiWJ2j`kh%EJXC|{uxPmTC@)3bQi>GcYWlR88#9E>
z9q8}3&a2eMI+36w6Ev9!Ruo$iCkgfKiUnNY4IAs~kCE=BmxZ-RLX?k;@19&d@owtJ
zHx@^=%mdf<rVySMr?DO<9ElWb(PY9TiI%H!mt<!C5%$jeyHBp25;j!p)cMTRthcGJ
zY<N`ZoX}QQ#uD1j)6$?VZgKY5H13rubpF+jp)htq=U>e+8Ib9W1N}&1b3X07u%!ZS
zs^SZ<a%1Sim1l|OuOII?KGm#J**4?rtu*_+$?vBn$LM3hb7ULPi_pJ+DQ|C+Uu^o<
zCjZObY<b28t!ls*MkH_3&*|XE8pcr5ipU}8$832R$bZtJ3O<jOQH}e#Nl*pNJqjon
zQIW{;8MA5*Pl`j{*Olq+)i3?xemi1?ybIfO&Ik1RSuCYRntnC>S}?^j%9hfNAX=yo
z)qLs86N=>|juyqScV4|ZK7A%?D@99V7CLotLfpg)8XU1CCQ;Rifwgm=pQT5yBL5%u
zzB{hTY~4BtN;lG^69p6%kS?M~L=;5SsHik0DkvZ#qM|`Uf*`#KNE0avh)M|^DH7?T
zROuZ8N{56J0x5jkId|qdb7$^3GvB@U%$$=yFhBE4-Xwd!>)CrhYd!1f7ISV`gLKp8
zvRp8Z-6L+X<UReMtRmPOZ=%D6a`Ks87=m+9a3^iMP!>kn`0Ntjq@7eArxQFad)fFt
z7H54J1*fzqAU~gzvNIXnpzTC#>)+b09(13I_;yOOn6kZcp0(Idu}k8vtc|EBJp=dk
z69GGILp8xv42q6Xny50mt^@>k2|Pmb$qCmxLdn6B=}Oa`stArAILe*ge%tEty8ZH?
zT4Z_V6QN;esoc<@N16@6rLCfK&+7S<yQGq!Hzf12L&fn1(bgB1&ZzES{fHmRmfGc0
zJD3mi(h)-e#yT&bm)oEqFOs9kxlKddC-l-9#O;N0A&tjm1~!9D3fai0gz;(Xyse2D
z)>72phxsaW626N@qWj>7;$cF&%M_}qr!k#8V;8)qC(oN-?0Z3&cRQGcno|<xJj<uq
zAYs;Dy7SV*J%y9PGck!%Wjp8*j2$S~%6-m_xHpqxl5KlBB5%E`u9{aRhE|K%+&kdm
zVavLJWpIG3j4UXPqWdV~7o$z4VO32la#CP(nRQ~B;roD=3u4v*ni?gP_-WWDWD)HA
zqy*bbcRIVITYO?O&kqE-0I^=L-2u0J&2)zzb2MOHqYo7ny(2>XZcGzgX60IyWP<m$
z<(A~_7CNVIDEkbe1vXj<a}*GeX;0w(a57gbAa~Ibbdk=XR2tvl&~kN?x6Xb_hzm)s
zTx{&oP+98@tHSW}tvA#8Z48qYvh%BKMcj3_*a%TO55np@X#(twGrS`qt0X`aN9iE}
zMN|&(IO>OnuR*dwre`tCoCKrDA<B_~@f`Fbv=w1c3<X6ttIMYyaJqtTQ!h!gFch=@
z@Kib6NUWy1jXzOyi?tIHXnx8s>4?$vDKSLP_p3IQSgy%3`O>^jOS|WvZ`hPY?g&Z`
zc=ws@c3Rkmqcx~R=yP!HPY~&*eIwRfDVl^0N#ZB$hC?{|u7^uMQ>(oyk(=Z3fPd3N
zqz{}0BZ5YNm!&Y8%VbEJ)%eYb)bSp5qx&d_a_4;S`#T-4UKWT9-f{>QlUot#Flz`e
zgXP?>Ea}ExNnoQXQM^(vP8eIFZlBORom1j9llj8vq3ms$+FjoIJ^i9sE}GW2fR5;e
zGdSZ&x#vjGhoxJcMiAWBxwqwB=<PY-a`{!4Y;2NUSwmF*TgxO2SHX~WmE0}%DPa%%
zP&hU_>QQN7Tj_yUs5iDmy~*MyeW86R&kF+ST%j|_K%fqAgwj-q1~V~9DA=%NKq>`m
zMC=a^a&WRsJ^1)=p^|)G_AU3Dk82gK#^lrn#>>x|0)fTz!9qkWH<EIg!AfI`GH9QJ
z_)gxbzT?cB?G&PH=p^|r@37R_`g%_ZZ=4t04ndI{n?VpemMa(s-sQGqcAyGtp!Z)*
zAzpM{ypFwGX2?fwP&hjMRd@USh$9M_X^%^M*%w8rVAoU^O9NQTQoyf@nShd|!yzlu
z*iG2!%OIQ@%HZAt@lXm$t{T}DUUM6O<7v1(W;?O(j$RTy%<8plC%1%me};!k*>&r*
z?G5ietI<gyvSYtR;o*b|n8ZW^zcz@%XXfWF8b0?qpB-OgXtg;zbJ4Jcd$+DAt0=)8
ze(@Ec8yS|RN_SxIl_&~AUz!4BFYfugP`w)K_?Vyq_1hZVt8!pnLO%F>$wn63<Yv4a
zd?*>7wLOh$sAog6ALgotRY~#>PA{16SCD_xyR~1;&-6_)iN9C$15c{Ga?e{zZ8;#y
z@V!$*Wq!hyaK8l3%+eDX6CWKi=A0GxG32|C-FkT~V6^D{(7yG%m8-kLs#qt0Yk*JS
z;wR`s9WEl@Tiw@FQd@<mHs1c)Q`3POy|o{*%oj)<=_q?n4?<hn5a$@2gTqi?u|Nln
zRBvsSSAV;Mtvdz&;|})6=4nn-sm_=d9iaqrJf(6t$b8DIw5*_2J%Ojou8_aKcVagD
z<9xu5M`67dRg-Uc66<6Pi<<hCsJT=r+Frb4mrr>DH%59w^XcK`Vxf+>S6u^#$epLR
zwbloR2)+>&NgCZ?IRu829kY(6Oi2K1p%MK#+JN8~fKf<5wB*@F&Jue)rtEoU&kTI5
zkW-X6_z13M<E7Bs7lyUc20JgdSbAXTwl8HV!!vr-#s(g+b8Rn2gTE;9=Gly&4Y^K>
z`y&v!-yRC0f*$^6Lhxoh(T~PX;oKdPHo(HyHs=3zQvAw@)TJ=Oepg#x?bk6e8?<;2
z!g#UeM@r$u+HzRIwV)xa*Mr>&!*r#}d;2u<t6#>g)8+3g#oey1U@6v<$d0`BNlF!k
zo7#vEckD4{aIR2|X&Uc{_z=e%OGtkyMUrR#<i6JN0%7xnF2}ILZn;MM`?Ve$#(g<$
zB>!e~_Xyn>1xH3zTk^M{5l*Rn6a14M=a(I?x(JltxzSpadHJc<x?)4o{br`?IJSU$
z)Df`6KzneQb`#1z&5!|9L)R1;%HR^B>Z5$fSLx3*pkgS^E2t2&QcwL>n^3)syk$e#
zAf47qb*Cd+58l?CQQ#VdRpI>xIpOwmUAg2E5XnFukRRVZ?YJa)Z0sbyp!G}J^lk+W
zQ89${&>>EPQ{`r+NjFb?iYSM_45P>a=1Ne8EYOw*4}2g4@+%4to^<X*o9q!5yUXij
zecwVe;!d$HPs_397?vCSrZ#qGScuSR5Y^dc^68nhZTG1MX|AMu8QUUpuG;5Y%<D7U
z9*}2p%w~6Io^gEgc8B0jtGu|%hsX?_4ItLqVyf}b_yfAytef-iI4p?f$G)C>>@`a~
z3X*#=ohl}yu=j8fBnogef%)1q6axaN+`Zqk>`f?fA$YsGx<78(tr=@Xy&K8yf85$#
z;^yckBi6;;B_te!OOEPnXN*<{r5QssIpVdPm%Ip*QVU=1p1O!ooz7d?QHN4_BAa8N
zTe8knxa)xPoCkXH!Nv@5Q!6E$BUat2OhOJ&Uf7GwG#haB8cKeu$bR82w9d+w6>>{{
zH3s+fKKUL!ilN5%0CrHPLU<1jKuJ*lQdsgpEIWxeKIq=vl}qh6j!O?Uh#B(Bs-EO2
zdi<!XJcO-tvHE9rGrzQ-`K$aNw$}Qq>WBX|i!WULU=T<dmF>ZNMi$7qXQ1^Gw?8C$
z4$=Ge5w4%Om7|l788}-f;S{L)2vv21qvBO)Ko}_l6Z3O-pC5-8_5=%ZCcdydd^wI2
z><h+#$lTJl_+-q1_V~AoFZ>BxZlaUy)LR-P{IXP!#`+Jby%|$r8KlQyfrKU(8^Vwn
z$B~Syvr&ke$d0V)H@aA!&?AbUx0r4BFm0}|3VnF>g7<|P`R0UHNj*eXd%8`lhJ`Io
zVK(%|Q_fIp|D#=(BU&DwKeKmBpcqBC2HZkM_%3`+Pee^7T5wgUy>%{}x{FNckVu%d
zn6s6EO~2SdPJ{*mf(=PLgY_(vqwLS12~oJ;6#8p8Zmv?mIdfbNlQ%Nb+*0rV`9+L%
z1bgJVou|@k?+ppkLAIVPlOB&gL6j#uhZ~csW9?SgAi+3nzGcU`of_|={jBnX&&li>
z-SGJJm%|T_@^Ij6D0)-~qYk_=KU(|5&>*X>Q-*nQ9q(+4Lh+-l;@h}QZ?6SwvOga8
zf4_{&dG)_G=Rg4SuY7IRwYL9T4E>9b=RfOSVO?wgwRxOJhLp*7&^U>51bDEvhH_y$
zNT|HCI-L^{YHd2}pz!7C_8miMkjaf4TNbrXv1gn{HWrq{dz;9U^c30w8=~V-dq&1`
zm8e`%aXZ@Rj-7*0W<n}0VBH%w<*hs-QK<gt$FCj*cDe$!QXx%<n=EUj<ck+^4w4~6
zJv`o0>qZ7l`z6&J_nZ@w87WmfqtihjP@b2BWZz`5&s`N7#k1<Dq4db^6thH{DN#G{
zZo_R!#@R<M9Sd7`s`GovvImtL2X9nW<<M9Opc;WKmM%ll9xIkgkHlYTf^L6Dpjvq6
zm9{4b*_dW-h2w)c5;na#sNZe#><ODPx4VQp=HqE3P)UM=<WLTl<E~wq+#o8=h?J54
zR6Mppx{-VTYc`{0pYG3%{jscJiph$>aZ_`SUH0V0X8aA9Pv_HT9jC|Rm52+q8#7Y%
zzl_kwx9^YEL=PM<F@<znU)^OWZsx8t^a3vlx5f6#ke^UIsfI{Q9Wwi7g58F~=Q5HO
z=B2?q(NVrTqf-;gM23AHB<0r3&P%>Z3SyiiZ3bm~GgLLyl{vzkU#Z;-%JRXM&KdKF
zxKiZ{>B>7l)hS@Da-;R}HJ7jvw`;jLA+S=Rm2d~D4P~JU9Jb`K-TRNXI^1pI8n?So
zp7Nwlm8_VY(6n1n=H{`j$M5o}$~2nUX5U`6f$`aL2R#FC-EToaRB$*f!hDg}-nE7e
z+~c<}w|{(COgqoN{DlMQgXYc(P_M8Ny8*44!s!0C4dH56pMjZvHXg}ZV6>p7pkx<t
z%(qqYQjT7OjPyb<b91f>2Vor=LJQbOXvYyp&%xa5C@ML<-9fxcM=ZOzr1NIF2M6*~
zq#l38tL)eJLOvZim2~IvzId!0O`fU`Ce#3b^P<&RK;U1Co(AcVbrr)IgwkTuV<kMB
z!Y(xMR)y6IdBVF1XDwMqJ>J901Wkr&lBnl_*d}W^V9^w7KP8WZq1er$b?L9IB(^4L
zM{c^HE5X`uZhe&rpQp5kP}KUUrP>kps*YCtHLN)L3dR2Qk^Ma=qcffAu5ITcx7$uz
z7ZpUyd!!d}bzSz@EA4)ur*vdwbKD3oau^o0nBWsu$L^22d*B$a^wE^f^V^?qOAokF
zw{w@w<(+3F9t$pn`ZKU-R_X)H06fpfW3lPc;5vs`6N+z6-@w$q`px%FaVQw2-)nvz
zt(dO&+C}2#jW^HPG%x{ZK~lm@#?V~1+6XKNZdVu9S~I}knja9vn_Q5q2F;pix)uJx
zLCrxOn0{0&WEXD!kO)0Re^@F+JwXv3-d&i2X*XGZ8sdMg<LlGUMl(k=v^K;$WV8z&
z3p*x+Zhtb}s4F`Xj~T%=A}QMuGrS~YaySJ>fQivAv=Q2!9pF{Q6T~{9ygknyMso8b
zOssdkIk|FSG;krfqEwh11va9dH1iII27;=Hsg9RA;aW{D<h652GZbsgzHkVOmEhCa
zde%KCr%7uE$0cEi7KHxf1)d9S25KzJtr?rfThWLz()p+n3ww1<&8J76ZN7hoBWezs
zn-W*Dhh5g<sD<SI&Vj+K&Y8I9m@YUJr=gXe)S`_X9t)PVI2rV*mfC#$zVgkL(wKW$
zEIA)Oo(<cR3vnwyg5hI)2F`!TAS48S1MHl56H!Y|+_vQNEw-zw-Bp$+jx8V0kK|pD
z_Um3>RX_cF{^K5{)4_ven|An6+H!(~w!9muW_vybi#wrE90ty!*Y?|^MqJi$%y1MH
z-rFS;g4ytOedVQQ_8CP%lR9`Iq8i*Xo9GdxjwHR|27(<!x<x;hY&Pnx)vnX>s@I|}
z`U_7`^yuXeBU%!`x2Xr*3uOYBPFNvO_k|AA<=*3SyK<Tx?}~$b>Z7RYLNd{SCo1ei
zeRhp@%$1<St_P0}+nwZ&OEZt(K~Du6D{NL&pt>RU`n_yQ1d(bQZ4pb|@#12#b&jr+
z#Gdp*7TsG>Dvu{Q&M4O>!rtYXXbcSkCZE9mXb~%cIyvLG=y0z?yw|Wb`Eld(=8;!>
zwLL#*v4>s+hMOkU$ilH;h9+8`9+!{shplry+v<$vxzO7;DBn|gZ+q*r=$9<p4Inj*
zVVgxN<V}pJg0u&3=^)8r1LZ;=YfmwB?hKmF$10wK;PV`JpYOu&T1iwWH-PMgY<$A9
zbXDjFLRXUh8k7wyp}7k?wAbGkN)0SEcvWfjvNTkdan^~k`qgpQ!PD+X_~tkA_}k|Y
zYinVGXc2nA;$R3ouYIM>nd&Be!1&bBOa48Ea%HzVA`JzvzskXocNofee^RNr#X|NH
zqNsq=Ibn<@Bu?&VoyBaWiQ!u?A`~gJFWwDbk4DJN2X}BP6fECYeFB&N@G3NlNAT(~
zocPFGj|GFvn94&5IoyV~wU2|>mO<O)=FjU&mRQt()HD>lf1X&bt#)I5c;lf|Hm_Uh
z5cV0R8F<E&lqfwGvkMhc*_kuZ^D=7K?b>9=WPkk~UF|)$OK<B>d8DVae5;#7Uj_Tr
zC`t^8n}C}h#8CnWNZ&ac;*BmL?UU)Ov$#)D%oZbgIn&AyhFqdu4_g%w4^|OmZ_+)#
z5ujHHA!r7dYVJ%hBthNtFCJ?7P;R?BuWnmlXxGf;&@K{E3|qb#KH6Ri7C@_@Wk&4c
zDunLx0x#o$wTA`)4NAFjGCdk*_K2o^=DnF)FjOrBN-xDG-&nwojC`f`>sUTBnTh~6
z!KmYOH`Nq<lL8Jg>scu&M1kt3^_rI4#PZJdUT?a~1!~FS<IjE6q&dZ}Yq8&vSkmr%
z3*uS<Jz+Y0I78lmSQ9JktFr}_iX}S)rUWWiJ=*y;PDWb)%&4qw%JHBQo;^<?^mQ-r
zymxWz;9J?0IwTu_+^fI|`AXxHFKqAbf0X4te7fkYikrV_+lZ*y%K<g-@()rYR3S`%
z!X}{fIN1i)0a%ZZE`kSr>hpO$wg6an2j%_qQNQ83bFm-k^XfLAB9+Y!xof&((p!$W
z_|&*!{WX!?rAHiz=fCze)yz2t?Xu5$$(Hb7ez?6;{Jyg>P-$TcmTO==`$tS>GF~#O
z6WGuuPD8y%H+Sr`&M%6rZ_*}PoqlTXmq?!omcOmCX(&RDEvt6iWcngi8GVWV%(&W+
zWI+g;tFrDm64qtl<aA&5yfk!gL(cvfYskloS{JOhG+_94M^UeV)N1X?XpJB_hVU|c
zqU188aSif#^IXa33X(JoDtO*d8MMy1riYAe=<{@txw`i54#8$7H{I;QlsQ{&lX{Q&
zksPD@J0i0VXD9Ev8m8Zs`0x`?mV`Sr*A`a}=fJcK63-xeJx{ll>QXg(?FM;h><&8%
zf@deT-n}Myw*P`flWdMV=8hxRn_C+>SJEaE^6@jvR4Ap^0Axk-sE$%7m`T(}v7qEr
z@=jqD178RB3O9aRCsn?*NeGEvh0qsY6b*8t(sW69uROdc<;;=>FU&mOf4-5R_9T+?
z@jm5MkG<N#K2I5rjLl$v4Y3Z004d>jl+!~Zppb4C<uyURozYlS*g7M#Sq#m$Ij-V{
zsZoj5ShV1g^G(}svCvgcBcH*u!WjdhlMJQL*c<0z-K*$5_-+UFdy3kv#~UxyI;AEk
zjszKhg+}&Gc|fG`2H;?a-wSG%5_$~)iJ5zwVloWOBg75>S|&txevvY!r=YC=nd5Y6
zzO~)9$n%_BV>k6SH?AwJ&!qu%rAIUGeultSlFT93QC&ImE^Zc;EjAx}>G6t2*o|l!
zws625?(hm%4`_g~nL5H?&;GL{&2pe5=+QP;s=~6%7-@!zA1qX^+)<cN|FBL4!IpCV
z;KtU{%O-3l@>EC60OJ6tE>=R&!G7qPq^PzYNzHgZl5j6zM!L}`qF-=R<>L&Uifz}z
zmz|!k_q<vFYNAaDj?)RG1VX1je&{7kgdt7|uMx{k9lj{T;D=>T+!wz6$;OfFrq4l`
zAHKGugwH>d9r9RypCbhgE5Ep(vANSf-%Xw%6?H|Y^)>2&U|!hh`)Jkm7Otiu$`?Y(
zE)vT<{ZLR^TCTy^(t`#WZx4T(Atmmz?W!BzKd<4Av5|}E{)$2qa?bQMOU{d3AEr9j
zxx<qJdV5doX>`+PSQqB!2LWSJj<dn);xW<j5D35Fxcoele3n!g_)P`V6IGp%m;N+}
z9#WCu^9A@(*`=vx^$zr_hAnGudd)GLjen~5de_xQtkgG)!04*B#{OG*b53MRl5Osa
zSZc`4k>-de!yO%GU2b-{s-KRIWfVW?%*T{tIfp1(R5{uKVpHU1>Z#UNld`>)8jY!p
zu<~~!CZ=V+J6NZ=<VhV$J!2!a6@!!=!}MomyQgjq&~!+ObpiSfaitsmUK@P18#NM>
zFEYfpm}YBf$;q-Jv(G|IMmuNH2v{jh9bAegK)p!02eJ>Rh?{ezZKZa%JeoJWAyhk?
z`P?h^$}y{xBI~8Bk8#qcuBDdadrC&Ipq_!ULpY#Jm?CH94Sl4ufojr2usA9?t5fx&
zH6ZfP8l<fB%2dAh731KqWrK7aMxc_EOSY#xB+ev3o!wFz4n!Il^=e<}z0-Ehv`qrb
zDbwSAVBt`wdV+xMQ>>r5c=9`tm_cRE*aUvr2E4EIC*4Gk%1oK(uYAjzcv2eNNHDV9
zTcT7xJmD33KYkA%|7}2ucFK>L&%&n*W5rgbcg!Hu?*3U_cK?jui)F3jzn+$H>*#6N
z(M}x|Y*#D+M%P4GQZ3hitM;tRL|*}vp=K3)^=7;~TpK<VAhqeUdQ*32wWYd0h<*9a
zO?;mm)jusa<VWO59Y5|S^d$md@FxB!@I2(P(@~`-U}RDD7QtE$N~H&0r@18U3NC0Q
z)lFMDn;3VwyP6EM)A;X@Ybp1MmX%fT^*Xz+hobHqb}E5OtG_!rzH%?%UURaha+Au6
z6;J5l4-H;y%2&8|V`f10tRvPVo{u6-aD>p*NY~;Bb0Vmk8i(Bjirc^LAJ4c{g4pFU
z_H?uLq3r!G_HT2wkWsc*AlfDjj){>2*FOQ%zBDVT2?B98d7131Km4`9np>KFoXcs_
zt>G4)csOBW*W~J8^_-u1mc=_~Y_7uR;guMf5_`Yd+iu%X1i7{pw{PV>czS{}?jb%`
zvWOHHZZ=h>S>kr>YtL2bql#aGqzk-{?Clvi;}V#B^|+Ts=);v1stSq$rECwGaUB`-
zNDvd9ril`8TQ9dCKK?<*+YolLbc=O#;2lGWUOmBZ&t)-pcyckr1#dlN1or|iZB4$6
zb;LY*$E3;B3x%6>tCXv#8tV_fg9J5iku2UQzjU9VIWtJzK?6k|wSlu3S=$~$vi?zp
zyRPF(bD0(|l5ZWFejhoaly4GO57AfwJTTi^qFk_!8lxqF5AJ{vv-?_mvTYO}{9!2l
z_TZ!%=XTROp9~y0RKh2RtRpT#M5%_1dN41wz`@EG<w?$_KS|$4+UDbTHGX!sr-RSM
zN!(Jk?!JSQ4I{_aIs#o~t9x~1HYZIZgcDnCmFZypmZ&+SNA#O19kq8mXt#2qc1ySH
zZM$Nhb5~_m^JsyAVoUGX8C;wsfnn%O3}Mg%p8u{o-fO7cj?jV|%C?n0VyD-Ap`Mz2
zN04~c?BT93=l2^;CSg4$BV*Gr0`$>pb#1ky00mJ)-q7fn+uhUOJ!$?Z^TR!E4X@-#
zE$&MSX?Zf219pf4UK(x_;)`U($kF!ZksRN%rl$JrHudqS7}yWG$Ot<6yWTxF|K)U`
zw?yEkV-a$s&i4uPv5uS8AiaZ(=*SWTkn(t<FQ=fO1e2iry(ylR#q_Unj;W4pvaMYi
zwLQs!Qg5`!o3n;;A{tLUB*e=CYah^r0ZufX!PX24%+>fM-t65r>N_KGWmzvs`dXLB
z%<ghS$iudwbqCqIkVLa40ZSk}%IOGrZm#ymeqdX!_w6E%=5oh7Ivus_s?y5r)S?-r
zGNut(C_KXo5>@v-F%qDh%)B!{PR6j6Fz+m~nQm3KkgIl4DJr#?aPQ8Rt<jguNK;4@
zrmu{x7487*`Z#An+4icEa5?otkgK8Vsf&&YC)9c`=m|sScC_x|K{q~k9Yy)j1d_-U
z%6?a(@j;3P=}uLo<wg?1W2W<@*OFo6=Z^W0(6Iib_$xAId@ADl^&$~{i5`dC(XG@|
zm=0uK$cz>dX+eRfGd?n8K5uwK{#=PESoNyxX>izgO!jDFe2jun5?p!3bOCve#!^5k
z2nRQ0G^xpphE2LIwmB`4Vo|dv!KQ5{CI9XGaZTeYbWw!GC9^}XYn0zg$|K4BV1<7|
zksC%l8Y1GRzYtDnR<A)e<s12)lg-}w+N}FvCU=aKYD;O3)u*Fvk)jH{M=uU65Z)mO
zv*3`*Mj{Oqkvh0t;o{dVSxM^~==pgY`W5R7G<gM2h^s|}%IuOA9bV?)ImP~Aka!r!
zYi>^CCd~y3nar3+`Br-t=s51KY2-X5FcJIm{c7g~at3f%+sjFwjH<a!*ooo?G_cNk
z?84TiQ_>i<`{(GIcI~U&)UzO@W}$)1C3}Ar?1zx50Uru^5YZR_IAHw_h1UA`+NuXn
zNSSIygwNA4`uj)y=iwA-gwG9UH%G#9RfehJKry~M((V?mKD0l-YFSGstkg+h^31zI
z2FDvex#`Lgj?q&e2X>dRN$Ij9vOj2$mZz&nDDa`D@cg+D3>VtgCYNaAU+U#^j!$h%
zu&AutKH|f$D=H1LcN}=moP_pWhcLpXR=(21GZYwrGcLE#k)d!`G-s5ByotNrr^~*n
zQee7GXx-vE&4cdu#`ogd2Ym+%vC~P!k@EXeYDJXLBzyl*(Jk36bBXw}qMQDq<=naJ
zECS2gE$=HsI=>3j_($Hy?*;lIee_{^zP4RztmJ;<FvflvkL*v_SM{<nmzxwN^S0C=
z(jJDLbBBzgO@`AD!0!3h*3Y<emV2FfbtoG|#TVBgyw#{@4}2c+%^_P3VBw$EAUoS>
z2*&dzI;5Ee$pXIlY4~V0W;<OSOV>6t8H^ypXhH&v<6sloYOn@rt&V`$V9OP?w<e=3
z$h&~&Zvu~<GeJ~?_Za!`lzM}U!j5{k^CC$h4AzUnG{Lk=F$53=U1{>2dSf826XocD
zy)moh@9G-uYkeT*Ngmpn$3PFs0X|g$?|}D`5~Un)05abToip0oUZF6AuNSvZosMuw
zpPkP9YS*pPvXw7AyGhoQ$K-ZIZ&>;UJWa5Z49bj0U|x>YJ?M=c4m2kMT(sX@54Ebt
z-F3Op|7yhLU{fQ7D1qa?9_oFEX8S!%Zm~O2xoHpZ6a=`P8;U4NZzh^DYl#ep6I=7t
zTsEMNG(X<*Bsun=+T)F_?ka&SmMUuy=*L$4cn~;EdoBSzkQiN`9;mavkH$ieK}!-#
zs#LbN4;8v*oKTfK@?|<mJ3G92#iGhZLNaD7!%c<m2TI+Vyzul@1Thi%7;O8MLY}hE
z^b%RC%o&BldWm*oaSdw_pUlPO65;q8ybiGF0h@@8n(1Gh3IlEAv@y=ua(S9=I*o@y
zC*3D&ke*c`Y@9WgyyqgU<IAfA_5lxI+d|u`NU}@=X12Gd*IX{bp8wFJvFkfdCtTp7
zZ2^Gt?H*u!!WO_QaJ>eV1qhAc#3}o<TIk2mpKRj~zcD3>5i0b=Yyztc6L~^<{Gc;{
zy<V?GG?-p-WW)?B`RKy)1<@^%4;`H);s>sz?@4W`#<ZdBTFao8XpQzU%9PE^ixm0e
z0W<fL&iif4_q+TBY0tXgS>Yfaut`|<A@T{HcogYlj`tG={Wt7JtQVZ|l9PA3dyiMH
zPg=mVyGe()UwwNQX~ii7MB!#eh|PVBj}Pt^uQS{p;YirOQ^Qfwu-9?4F<br7Et!QL
z<vX;iJ6Ic}!SBM+SSVNr2Vx=FmSUWaiaPdLu=29w^?=mgoLR+aD{k#!BNZ)FM9#_c
zW&1#a!w%cy1JWnu`(YGQJlhwF!w|RI^$wJy#`?Fyj&7qTvyvTd6;MNy>^S9FB;1CZ
z8HwP<a2+*FU6}+usuY$&v8tAW5gXI&FGI=ncVGJRgQl)uPrT5uHzKs(@_2OLmwOJU
zQ12<+46b}Cl6HX73z8s1wdHB$ggDL0Vh?+<4xBYlq{>lWatAT_y~C~J(`Mz}C0Eka
z_yt)PmGrI*5)JFPCDtI-Z%Sa15nFJ(!EaQ+)}^7M7`5Uc7)~q!2h4XKYY+zKPHpn6
zq{%QQg>TE~O&`S5J+M<Bb(eA4zkJ6c%R+t#Emg-cZ4ELhhin$YY5@Tu-d~y?2{?|7
zeVVmc$x2y_a0h=Q5g``UYo`-=*WOP#F;cm>B!{P<?-IHao)N#7UWT{EvQl<CihUv}
z8FV~8WbIvhti*r$R9tpMpvB>u!`HO0o$^ne6A}dlTAozVP1vFIL5XbooFI0hk&B4z
zX(b?hLHMEUa%=Ke$8G4R>6Q?+S8ZDtcd%t>l;^dEjS89)Vk&CY^41^<Htq;fg?)I+
zcDQdoC}se&pj+xX;wWfpm@<ix`>c_1#oq30+F@t;UW*iwOQ|=zb&VfJl;c2v0#Zav
zWqKss0mp{v-KY4lZISn+{kakju9>2|!p(Pj_Xi6$*)U#*oPl`=Jce2W+(k3mk&AXZ
zGPHap#`(@FEFaB*dKi6|o>eA_X&xJb`IYiH4fpno$yWy;@NX@Ar{|7mD1?b0b&5{-
zcw_X&DK+5vge*@eSc6O_j9@vBj7Y)>-kzyFrH9NEm*go5j+d4U9a5biD{J3MR-vXm
zWTS}=BYQ_iU^Ajr+X+xk0KX=Zag!n3l_z72FU`8IXvZ2^@@}J<*w!u5^0)deF3U&m
zg1~LQiNR<P#LCInH0%--2hP<V9ritOKGVxXGwa%86g7~1Nqu}RUhmSs@P$0T)*o_?
zJs<TH`_UNz?)X=3Zp#4DDilrvGz6bRPJKzymvPQSot#y~babE1i<0ixY4udFc7MHU
ziL|?s>YImQ9FXOZZfM0m_(R~KI_DKH)v`5MM4<AiLsk5#73;Spg+>8>ZuQ|uXA!S)
zJHVHDOfk?W-JiZnk~&~1(!UsRG+*9KR-)R;PE|EWs{T_k{MP7QjRiPOpq0ErQOTr1
ziJcW8{_#7qh&};mVVjLlHoR6f>$NnufbMD^etO>YlWzoHbC$#VEH=InO>q5aMjuGq
zohHVQk)0`9GkPhVRZ=SPJ>1=SFNGtC=E)U^-C8dc!Vv1dzI>XOjcn|8h_H_1DY8+%
za~gwPm4;a*eI~l?&PF0e5?|GJ6peJLk79&~QM649mFJ@{s+8;zPN5TxQI}IpYpT%Y
zN-1w6h_34#OyFLaEJg*AH6Po(K)y&!a7UjI?xyKVBVJ89S;KZsF4dQvt`&a~nC#%8
z|3y~9<f{^GQ!TEhi3-rQyzL+i-v;{gV2)zTkKwIw0Z7sOL3ZF<$p%6;Db+*9^u*JT
zkNdQ5oxGWmn(xAu8q1SDn#A{(ukh=|jQIgD^6QcGmkh3fUJ8<&LOJ%iX|s%7y{4Gt
zbSz3$Fh<tn%mc=Zm0s_<vMhQr+<pxb0?)zLD*1F~n_2XsVI4t5ZL+~l^(A)FF<xJC
z>gTcyM;*epK2o@?=4FMvfS)m;iqmw7M%Nz{R)ug(?pxny@ui>*mm%aI8W+~xulyKJ
zg-VS~9F%}hhf_~7J|XjhIx;e8TUr&-@R99<PH#;6%q+|%b2v>!<!pTJ_wXgl+<n+J
zTSH9o#;Rh65}IpC;tZ~fq@>z`?dOW{pM89HD{8k;Kgfok@ozu&z~$aF$yNT!TU^)3
zVPaGWe(2Rxd@xuZgET;#@akc`F1-74(DQr?yJ^c#u1Ap2CKSV3WfNE8htppkmsl#c
zE5V22fZ^K-3xbdTjqUJ_m~F8W=CdL_$IY*YWmLaE?(Aw&b*3e-@Cgg+?mQN{Xy?Kd
z$ZZ-NBg)Z)3`Uf^)OT1Fw5BUMlN@-Elf&LdR5N}JNe}t;tOY8_*8^(wQn>Zt2GmAJ
zSRF}TzkMcBKw-}Vhfz&)uSzT5uy*l=jfRl9bKK90PKP6dK%EIp+#bmfLTaL=4#?)q
z8()2eN;T+kHoR26PN`tf{P?Sqtb1O!1mn{@4@#winXPTJQ6xGZIdevM-}E$s%hb(F
zn6a`RzmHCARd85*x2N@Z-l@mtM<i4pzULJ4It6Dz`eTU}SXSh9JXe_{TCX6)4e#S@
zU&8r8bce>?)9oQIX7hRwuTKx)FJLyJYFY`f3gmjqe05<Zfmf_z#ib5rTWH9kGik^U
z>=i~sXqyd&_aw%P5|w6DhLt;3Ydw{W5hZVZF{CXqv?(pW>8(X)lIX9!++sld=$SHV
z#b4$GlGlsd&w4MMCDLx2Nk8*l3dmB7+yW2yP;x~C>nuWUaN9=e9JY)hl?*KhFa)nt
z&%^j9TFsLKRkyyCaW3sz^1o`xUfenF`U2HTu&IK3OYs6^yKPJFSWTBY@GJ}SSiQs{
zasy=ShFYDoLd<&!scs0Y02HBtSIwh@$zaolOFyiiO_E;86dIfZTvNBOH_uo?ys>%=
zW7}1orPa>|W!N9vM-vf`)R8_6-ewy2@Q2bM`O1zxt8Q=Zn>r+!hU=tqsk+8rcNC_?
zB8bRGOL&pVnS$Cw*+uxbS~b%7ekpenmd`e`58-Kib>P6Evls11jEJrdJHjH48C%^p
zN&?>|_=D5TQ-((x(TBkiH>`N(Yb~L(Izt!klr~^G(OgHLNaXM93#4)m)25am{QbUi
zztz(0-|%_(TIWBMXS;A8WWDV0{Ad<}qmOxgtHlIkXRFfPv>1nveo5Y|>>m1o#W_A)
zqpx@AJ3TfKWclXi%vMg1$3QVH*z7TXZAFr6<>RTyf&sO6VLgYMcpLJ9Qcc%2Hwh^C
z#cM)(Em&`DF&V#K0p#N*(-B0Qs`HH9#3oPu=jq*V!+OTfy^W+!&!*lkbvlMlao)R&
zs<~5?O>8lt+lFjG)SQkOk~A0|bJsXva5k%LNlU|QTibx>cC!?35w6!BJ4~*1opnF^
zEWA;f{qrKkBnW+#9?#%mG@HwsL%~^`tB2x0(3M>ncB17%Z)DBfcEWbUr2dC*<=wkP
z*@dR_PO*;}c!NL_fy{RFarRxhPYmivzjAA0Jvr%);#-+Sm9#{sY3jpS4qWpAfx|7O
z8p&qkQj%Go$6M7Yk>d6iwjCl{{Wb{61VPT7+6PfFe{zc53w?tg!`Ogn*|W)%c8M5~
zS=!<0@UBj2{U~=_uT{~*u&ak$cEIco=;9*tV3cj}X_M!ZQ)!07^6!r8_kQkKI8z+f
z$Je?eGsP!2lKrhTUjJ1hPaL;yGfgR}o8nq&?p79&GC@k;(@~JEpsXjhcy#2T(Aftp
z$OC8;9fW+Fke-vPpG+x*!)M!;F6zLyi+9Pze;(Bf8?eyc7=LA*;i+MduzMqN@DnWu
zrR<0S!%%mf^%VZvVO*u6Rvr#&*C1bhb6Ma*gx#p>hx?7QYY?^U8#a)52reGvX3KT3
zSD`_xOSS$q<WOE-r;^CK>e(3W0{)XTpR`|db0puB+t$quI`;IOJ;h!#xB+<P!MhPd
z*Gky^O9h6aUv@ei%hT+6@_Dd<`}GIZ`Fpm6$5J6;@~)ljI*P<h&k-Eglny899Ip2+
za+6t~-Es$YKxtpc<Q{+9<(iLEMIu2T%e`-NXm{!!fuO)n7U7-%0;!E7_eGTa&{uF<
zaxem1AFMuVc6;-6&bckLR=}`q((J-jY27?!OVtgAa?PdZhJ&Cr1Hv8Jq6AmRg7F@D
zTup-OhUJikg5sbTF`D0hN(V4J8NsjDVL@-B$CawM3=Mid6{g9o9-xm5Hn8_(-dyf0
zb*gImk56zfM>VP%Do{RA570#EiFrr!U;zxFe3H<+Y%53E*cU~Gia`S>s*Z;nl(611
zVaZb6DBo?H2yF}jal3pM{>tFSDZa_MVet&ps^QPvGx}dGZW$c3GC4TXz0BIjhfD+w
z1CmhoM{}FC_g~%+^U{u_0B(VnbY(q$o?QlFkT;2UAy@KuBtjEvrwfU)Xy-MEcchf)
z8YBRS<8_e*1K6g7F8D}q6IGMuikYLEP?4@*af#o;A_d6H{XptSo}@XE1ZryuAIoSs
zNeZ24T#3v>RGe@f8Vi`)J*^(H$t|`izB<Y1301M)#4zh)*p^RKiC>^>BkYe^R${0k
zj5;I>)-I$=Z`hHhi09p&<OUXJhi&;ZenWc~eo0H>6M@5Lb8KW~W=|PzS$79{6TOk{
z4ZjBOo-#oSp!X3USy=}1Nu^*Fq6$x2JLni4ug?nOWUF~|R8}BZz8!8e=b<A?f5zY+
z2c2nPE-tU^#5jwUn=mc!KG8G3T0{+8Iodq@Wy@uDh>hI#caeWvE9S9*39MzCG0i&s
zh;=k%b5xa>tcl8iEUz~C%WmoOXG5(E!gjK7P4B(2>8O^_6mYkJek(8Zz-*$(8By*K
z$BZtDR~GNmz8C2`e@w<qG8tW(KCfbP?~0k9?6ariOa0u3Njf1e!*l$mx^E^P1fJk+
z$Y?WC7dSk^6Q)0uwo_79B*T0M>q6=&u<)K`RIPfybq*iujYai$4>U@iSyGJ7u46e|
z()NJOnKe<Q+KtzNx)sw93CueX9-F0DvvY~p-d?t+?iI7OF1B~sY#VQ%{P;pi(yLuH
z7b`^?_u`VllnVpb&1IrbIiCF{g9UX*&mPHwG#}{x^fLcT-}~Ol{ft(zCWWSboIJr8
zGnO+|nY`u{d7}8R!8B>u0#q(ZHUk0~^hr>XZQE8zJ<%{Q6=M~=-EH~I$B~`<yw3|r
zk5vtI-T}nT)nMIW>|nnQhZ+or1TaJkNW5_+YQ1@}gtvwFAZohW$HQt)ZkS1Dy|jUi
z$D|$)g%8ERvO6mi_{s^fP<EYcS6jkS;d3QV=Q-j|>ds3(LIe+7!Q;%S^ZpPbo|n;p
zU9Y{B<Y`7(9v<B27G8_l;x|;*hUvV=xN+OXHR^-Q)BZbO<aUb4NH}XPDQ!j>R19bn
zfy)WaF1csw>7Ezy7pvn&jkNhQZ)?hUY~)~*N_5xf*igI+j6W7sj(MEIRPXc7bUsSk
z>1rdeXT*NP>wAig4;~il-#XwKc#lw4OVJ|&7u5GiXrVW7f3s|@yr=4*CL)-(%oEeo
z+O*Z~<>hcsnQX|AlV$GT+J&MS@T>TTG9x&edKR&Xb|GYyi(at6Kr;Z3Ej4hz8$Z&!
zjGe^yPzN>KK>J`8P>cZ%?s7iUF{Cuhw<M%LTBJa%e<WtNlKSf}S$6$JHeq|C56xD6
zc~)B*pKkwxvW?2asE6m0aWFp}Ns6jq=ar_YSA#~1OKoQ2hJLBnO#RqAtj@{ub5^@K
z9)WCy1WGLp!C2u}n;yb*k>o+zP6Q9=@(A)2oWk8@gN#@gmoc~ShI#DBtV_YnLNYbF
z(WP;Nj{Dy__U>(hq4yJ&s^L6+KoY;(%Z{M9d2WM|MOUQhGS7^2)2^w6h&FHI4_i9R
z3z0-aeAmm7Q^MIO&`NRBZHJ_G9FIX)RA@3S9&ZRDn<~YaXfy2i9^^@(9H0WDJ=Ht!
zUGKpDtNXuTV!uXwGL#<)`m845^z2*$h(4ScakMQluZ)|LGCYOk8*f#{N>8YQ3QxAs
zeUC{;S+?v`&0=$R(HI9y>{uZnwTD3W1fNHxMo_;*;*H{Gk&Uw6ou|s)7U$_3helr{
ze}H#KJ%}IbY#G=@kH%~OL`h>V=~IQJh!{2~pOWHDrr(_Hw+_A*6nCVgzGIxzPiNFV
z@X{p#%Vlt*%gun2*-El6MSw8(C@L_kA;Vn2M(wWmYX75-9Dc2yZAS)!k8PHm0)BY1
z%hhoPmK3=(+6GdzJz_I$dz($`%e+{>&y8l1Njy+3|I6yK&|*GzC*+%md1Fjn@x~bH
z*&|eAirmH6%Hp`C08{&7>r?VQ*T+BQ8Sd;$597@)(ATwX_pNhyFGVHb{qP%0dF|X>
zN0bC6iWT-*W1Q<sO8f=*9Td3bm1VYA^H`G`FjC)=YuYj|87=ri;c9HpZ0~UkP&&Tc
z>2s7m@P7GesO{%NP+H#R=L4tP3{?}3QLV^t03PgVTJ$$%J1uC&ouwwP$Aw=sa?ATF
zN_!mr65LmG-dOsJgipELt@jcpU0}zj!$CVhk1vx1+7}(su2@Cs$~(D72d-Av&s){U
zyC`dZpos2;K!gsxItP&|L_P)H4A8&@v=%+qhT=RtW`{ma(jUINf4{A+-9WpqW}vUu
zr4Q%Vv1>(N!5dAD;x>Vifd;{^ACMzQlR8Rt3I>F8EF8EGql9iOuK2+1`rpQE8)zJg
z5V_*RFw210V`^b6()3uoO#&B9C3<K3i$??38<RJ!1`W>1L4vtzj`|^-j~uX{Sl#hM
z=iqg~9EbV$_s0PG8x7w7S;~6m=l-z@7}j6wt_(~b=D&Vy*p}}WXw7w!WrS^c_57DR
zmb-vb4vk%E!;UK+lxU*Kis5M>!&xg#1G&gYo+fJ$;Fw4ufIO1b>4V~M;^XUUkcGiD
zhy;?D#Ng*eGF*Wi9c*=jjzB;rwL%vp$D6P;Cu~Is<){XB1*8^&l_=*m1X*%Hl;40`
zgB)0ckiMPG;KN(m*C1nf#$NoeY1=8_Irs1I(N700+bvNosUNOEXlcy}DVNR#AStUp
z^d-h!mL7`F>C$Z5_35#i-c67(ktHF(W_>rJ`812=nX(YNd`Kd1!gr$!_BKBo*_s#b
zd$8Y)uEBRBD=+2`>HfvU!oHu>^y@oVZN8gWLF{+mAo~^blYj4MPkG~B0r9XnUURt&
zQ$K?&DcoD5Vut*F%u_>Fgb{R5C_f$H!^GHMDY6WpTu+ylg<UNUM;+uC+E}zG<Z~x;
zGQ01S7{bfs(uVeN=w}=XL3e1PrqLSim?V%^;PktWV+X~4hB;;+{ZlU=d1YJL>sXt9
zSQO<EW*r*hC-hAfzw`tYvR;pW19iIaf<C;I5d!2{A1tvTo#L^B<@?2r8GipVOO8-m
z|KP<V&kax^e$iVM9}eO~s$FL8f$h+|f*xb~E{o?cR@(2U`QK*`^H=@V85h+|+HP*1
zZrDAyi`{wU>}4Ol_nG>I>W)QN(6#sTDc0|Q7!$SspQzm^JOY>{5tQDrra$_`aKGO4
z)5y51vdu<cj%=?LDeoEoh<U_^d}1j`2f`G)cShCwbmCk`;Hm?Y^8-7Dsc?jN=;*o2
zp*J`_90;~wOeKebzC(N?j#PFdS~W(PBFr#wgn|MteNbR*pop=liC$gIzMFI`$=^Q7
z{xyAU(R7HGq&!?-Q3hu}nLe^7{F%)5sN9ovb$jwVPo^${gUHYJ%->JFX`9J}5ApKv
zIk3nZ7<M?zXB-ldi@mL-C4TlU8(HXMM;VA0TkWxF*zxV|d-5ABPUV&<_ZnsI@eF$7
z0zYjEY#+WmO}(a_0IybK5IW)iUwgo*duOyeZW(R;q<y_=HB9w-gK;Z<6b2*>7s3G*
zAawhO#5nxdnwMnSS0m@4A&lY>`uen#@BjCT)&2K)2c&Jrla`>=Dca;aQ!u$9udsv2
zv4fHZ0T_B=3vkYKTIxa-&h@&%R!v}k*bPP8lW_#hJrYjccDZ9{4H9}6bns|m)S3R_
zY7CanU)ic%R07$?j=nWWF}}W2!(|Nu1H;&<4~}c*e=w=Ut~3BM?;=5IUSn$zcVONx
zo!vPH&A5T`VH{cjhLYGtzJnrpK<py;2mXY5!P)#nHk6v($#?M`uyYFl+maGVh|?dr
z*thox6(zZS0`1qxVOdjm^r=yx<G{=|$mNzJg8yi72@3(97e#v<B986ahx{r}#J|VS
z??+)76?+m^wd|Mm{tmK@Yq4{AZo*%&F9r|8R<<oDt*$81H3e36Uiix|S<xoAzL8j+
z&Dc>z$0bKPTgMUmarpe^!qq~ChwZv1Km6!+Y@hkqR3b@}_Na*(aDo2QxIU<fTZ3$&
zB3DNfRyool5GO$<#0^Kkji9rg`}jk9qF*z)`5qVlM3e|!`=}0xlw?tp(6b?@PV%BP
zNiyUmVn`((hVp!$Lxab!EQ)te2Tr|qxnj)vq))KQ>6s0X6q5UeR=_O+l_NUR1&l&y
zVp#H1`0ywL@Q^R;oPaJD!Du}Qdm(5thNWpIl;amq?1mHVfN;eYDEz`Hnhd{`S_Uti
zdIEy+J&gndNSwz4qPIo;El5HG!p{zuTLwj+`;6IxRxP0TrHtdAr+0?;L`z}wV13K|
z`372xiO@s85lvWcVLudOrvNUj)A#kk=65Hou7G+g`Hg>A`LS;fP{bKUYY=B_Oy>0k
z23!tULkH2OQiutwy$Q=ED^I`T;3M<c)$zebcX`m?;vc=MfA+dB)XFIfsjcAtDFp)p
zqqCjg_sb&z>*hZgF4mQR4BN_PaPfQl2=aw^5b3VKmaP&uO8?Od{<PycFL~2;5EQgA
z-X$mU?E4)W8Eu2V1KM%m(lmca+W{y5Ca}@CiZ#eO5SC_!*C14m9_;uR+(*<LogDIl
z*0g9b4pvFova2AJZ(D=BBw!8NP9MRfW8=1hW+N=-s~}EhFfxFT{?iW`ZtRRh9cz&D
zz}fo*#Ox<T-<8@9hCXHu;$)xjQ=T7c`tx<~^~B&j6s$&>hTvNMML(=NuwQfF`EvkV
z`0m^ahdEek&VnE7g#~Klyjt2bSQUc05x{r+D{kECTfFkZ(VY==4wHo@>Tzsc`L|9f
zSu<cpgRet~lRwTVy%7-YquuIY^tlJI$7m|JzPT<d#))tQc=fww0LZj&qS#++%Jw}7
z3)yJ_sf;o#`B>#oqx^eWm5n78*r}V?e&kpCrvNJ<b)X%25_aDNT8>dLMsH_aD@tpk
z&=MHMnhS|s;70f<SpS~&OAA9>?Wpy|jw-=v?B6y$!FZILy;T)_UO0il`fd6l)xQye
z{x0K?37o(gM|`pU!EgO%=0FXDsSmwtoYnV_UN56Td=?3&IWNM;;;CjIfH%%=2kXY)
zfo&)sJsCgx2-V9d|GWlKL3%OPgBa0U34G-F@<3$%ge->rHw*B5JTOOfOB%<k(KoMd
z0p&qTG1$&J?8MuImL<W{$TZm4(Qje#{CAt#9C@TJxcBn$?f6B3gPYMOe%!cTWdCu!
z9shA4Sv%C{R@i&MFO&s0@{tQ~9^9vS&S0m)u;bF1rvNE_iv8E@lb9g<4v<U`{(=T$
zf^dBM|1=1Hr7D{VLM8}*+kue@!vA>?{)%Lq2|^|anIQbX+>SQ?*SZNN2$>*cg7E(m
zge^=EG7~euC#k~Rj{eVYM?W(`$OIu1g#X;^=&v-|_-AfMFP?a5S8UzWSopffZqFvw
zq$XF@a{k<BJ9bk}hrh2I?bHCuisg#ARTxbRu!$CqjNR`0%;lc~%`tWOJ{`)C06bwv
z;TlBPfF?J{jUUGn#~8d0cuFy3!Zy>MAu#7PxCp1*0lhv725A>z#02zi(0dOUye3dI
z3V;LgIBO0dcD4dR!$spY$RQ!x0;3f+e+$b8pWcsMgCsr0yDehHX?siB(=%?M9TPA1
zHE0_ODo>s=RLHDf_hQS9T$Le~I&G67gBg`Ih#WgffKds)&Z^cc*2*7Q5c%c(WRW#U
zxOFSg<SMz*<%u=;JNU~XD|=4B85PTepWSsbe*+dw3iDUWOqdkrzYB#4Q(*wEXWl*@
zusygLz$_W0V4$6I>@z@cwY`XdSY@>07ay1meW$`0n_h#gLAH$y<LQCLa7N@2bNtV#
z&U9Qi;M|OAE3s73yI8t{e?BNPUfB!xO(07GzO1F;pp@h{uK3#@vS=`?mKM7`VGVNG
zb#4ObtwD{W@?h(7fP{f8T*d&7>CS?`-V=x^DEc*yl_@Ct9~Bh+N~f5A-eS%K>vw`5
zruy$!)E58V>c2k~CCUUV6RiKe9O~a2tbZ!%lnGWQSean`ZC(AJiehDgl?hfRSbrO=
ze<~`L305XpnPB~Gu>PqiTP9eUU}b{!x54_SqIQ{JWrCFn*53we+n-4c#RMx8tW2=}
zHdvWFD^nD~bXoa}xU4XFRwh`PVEv1LmC3U*!O8^dUj(d7p7p;6>vJl<MsN&O$#z++
z;JJNjwey?7QpY>06&XI=fkmuZXPhPu!p@&Qi!Y!7-!Q#Bp`eR8XqQ7ei*=wN7~T#t
zYY<jYnm~B9Zbrz6>R5qYRI`IoB%x~%Ap`OtFb*8Q_8Csx2V?AV-`(<Gu5#{X#oVFx
zn<H7R8vT3bzh0R37k@h%n#H;Ioa{4*Avux#^s)D8LI(^4pZsVx#_|?EdJ(<^Tmo}|
zcNg?K*YNefA!`16&r9A1-||=N4V^zze#N$0=*I$yU$ct(@1DE;3p+G<$zNDB39<zI
zDB=FSkKfrGI{l|wxtX2s_g?77*Z<_hB){<|tF%)aB!9;UOUY0Tv2`4zLb_)P;|iXe
z#=QoSDf9uwBr-sUo)n3{3%65R*($XM_;20C*P9GuA7V%E5!WE}M-K4C+nryu?=z?d
zEf||s#YNz>6oym<YI<OR+Uf*P0<K>8^YF&U-|A0(W{76}y+xC|0wtDl$cA2>i7AE(
zU_TsRgS^deYW_hZ4}K?my4}y*(9SOY%%tfDdoSp(HIeu?7AreZrWnWNh70L|oxn)#
z4sxSOA7&pY65^x)n>#B|_QdFGX0LqmA1Yd*rG#nloE%~ZyPOZDYxZ^4Mf~!qX7<?|
z_$lP%^?hN;CESvR<WFvA-(%b2JEPN|Ch*6J{6`<zFAl?(3gNT31&szvZ0#tp@od4z
z-7OPXgIL^QOoBdt7uv^Y7+?r@QGBI&{q-NM<9_F(#4nB3YLTTn`Pl^63U|{Qgyow9
z801RJA9PtXaRl{U(pB);+6za%Gv%)OE))1;p!|m(&wmYCRs#NWJ;y44a$$SI;`3t|
z_y-=@eHnkp=dd5zW5hUW5QLwWR9X<CAcv4ElBIA^oFa~(OXRO0B-bEiEsMw)Mu6KU
z>}XUyergoI*oPQj_1}SO4_Qe57F54?TQd{FB?2c!^TEl%RnA0io&PjG%=D}poEL!4
zD&ATA@z=9ViG63bcJ9D_hAtVuR0m(=Gx@>Hfc4Mi0{l<UX)_W2w?=S3SnB=;@|g(#
zg_G%@9N|og`TG+96X8sR|6*PL%dg=~UYd#UzmMwg*L3C@{?}T=nPfH-;Y@`8V%`2e
z^YNEo!<h^?6X8sR|0^S$Nt-hf&P4dXGQyc$I}_ndg#RleoLLCXL^u=S|H=qwR!=h#
z&P4dXGQydBI1}Mag#RleoXLkX5za*TzcRv^d^i)~Ooaa{Bb>>HGZD^2_`fp3|M&Rt
z(;e=TU+U<sD`&~s_@2uvh(XJNr4>KwMNB(ViXlRK&`A>=auk$Q)6em*d+s@X{*Hrc
zPpu4X;n3{udW3>gum4IKLp+uI8nm4nT7wLgoLqy9UWd?^N#?l+3YrnTt;9@&PvO+m
zd0eF<<z_-pD#OQf4dU^w+fC@JHHh~F{S9^;1*hz8#jl9D!*8RD;KkTo+CThD10{by
ze?iE9xdiJXj@)$~*?p%f_91Md=RR_EHiB{hnew!-dd@ci^rgByCpk69ui-coFr~Dq
zt5;7hlui7~hJNzQLmA@*6|F7t4?4fb{Zvc#;D5Hk4pTz^_X87CVEfk+*#2)(_bHNg
zI}OYsG(-A35JSdGLBqNWcjxHR-er4n*=rC-?8-hR!8J&yBnTZxpW}jmYB9_@3unXD
zdD4!oL9$=eu2`5<Ml8cT;G^XX=`~1Kixn*+ZTKVP$BOHvroYo5%M6*j2BAfTc!MTb
zs^g&9ote<}Uv$lL*WTy<AD!FEMo68sXhfwWRTKx_@U=_=_m<`i`gaKsdl@yL@gBV>
z5$f~fhs#gFyX+Y!6W*^u=B~q59gA?`|K+JfXkRzB1K-x>Ec>^5*Pp(6A8dYi!s^Nz
zq#G1lv(o}n8D&`VvC1C`kk|eAg?`85#*zx`)J<$Z^6UR=@5=+BUcdj}qLihn>|0Y&
zMABj@WVE1Ci3%kqDMU=NXLyV3%aj(SELkGkq%2u7c9ja**RdqB#8`*feoy!F%T>3#
zbU)quxu5UHKfTAh_wu~Y^PKZM&+EL-IhJ903<&oE#COKPiXX7z^vqsb9o;lLsgg{M
zq~|D%MX}X>qvFlCYG8eRZu?ik>pa0<hrL3jzfv9$`n~h>Wo&<Z8QZ_gZ$Gcz|I+w*
z_5NM8{;c-%>iy%@`>%4^&#U*mdV?e@pH=sd9Ou>h$E){W<*}bv?|Jo}S8uTM{LhxR
z|Et{f^XfgX-t+4HA6M_c%2)rtq28Fr<GbDJR9IOXv{ARhFIv6}o9Uheh~f3b5e-@i
zVZ6;&73oG~K!vKX*<GoDfbHWLt~;x)b)U|hbkSVDs#{@qpnnnrLa@=o;Xd{*9CguS
z07T|ctw~5~pkh4D6(oGSA66QLXeN-I^ackQkSdxR?vM**Kx?bNao^Esca#MK>g+^6
z!PA(q+6>6O9!(9^pejpE0)#M1#ve5zSOGE?4v=*GA$kN>v}=gDoB?sj!op6crEc4`
zVfdECHL8Vcyt;iMupd#;gsYD<GN8!9FOsn!EgD{n_E7rL%~{{Y*T~!N{cYd=&4hLh
zn%~NP2kHKiCH*(9VXBz&sn1m2&45-r0|yamW^uATrFg(#<!T0W6`+nINE*xGMbh7q
zF!GBuowK>^>`Q)5d8aC@_Gh1GZBD!hvJ*awO;&8J&hG*Et;;@De!K4gKb>$xYZ;K3
zXc=*GY3Q^+Rc1tn8r+Tq*(-zM>A@#xRkT>byDLsD^uo6cXfxW2z7V*=T8aU9ygZ!&
zRac;IkcgBaq}p_3Usg3MeIK@hp0CM(4wley3DN#EeFmhs>l4VEF%A+18qf?F&`!B)
zCuR^-uM~O;7*=7>lP=8klE^9E?3q60gFx<E>0J#d)z4yEnHCQ}S81SUi_<+Mdydo)
z-XSdk)P5a2hv%aEr_?7X#$QD#_}?%@xW*&&LrI0msTCj#sr_p#4t>(CD00-^Xfa4E
z8cM&LYR>!~xg2CBEKY<QYJrDAAo)b|S4L*R|HttWeMyt%z=r9QVFqLrNzpkmAvsEH
zilHAg=>&uG5<IlevNqwn_T6)olsR=Ia;7D6N^d&sJs#CP3eWTr-{hc|Frbud24oN4
zNwpZGqd+>})F5i#f{Z_@R7-!6*U|*v{#9CR_s!-Yd*f$6$@d(kvIszdP9vvGQrG0V
zFraB(U|Ad#p|0<F$lPnCY}F~ircVs5HLjn{dI=TJ5!>&Ho4Ex3&6KqteSXtKP~(Fc
z5Yc=_<`n}{%lAc%YQ9HLjG||(#m8OMxb^zWr{8|E?Q^qRCi^#3!SAFo6Brsx)xyZx
zB(@DZXKlf(a3_N1<hGQI2xma6DCn7=%ESDWJei5{SMd$z)qh_7Kc%&uSO4EyA^gYH
zf8N6%_HTIj=VyQI*|p}UHRt^7|5d{JfBNh{&j6TL|M|uL|Lo#_o*^)={`2bpKdb&f
zmO=2Gq~$kGV0IEMF;c9TKJzTIYD4=_r=Lgem45DsEv|jcM*d_``pWGi@$@S}0Lokd
zup2>CXLlqU+_$%mFtP?oib6c=!+U&rF6Qf7AWHiUd;Dd9w$HuwA$XccAXf6$8DA%H
z-Gz-C-Kr6`xDyb0YH|u-YSAPunoSL1++UE~m0$i#+VL#07V(q{;JS4P5+A)R0NG>p
zQ#l&|=3Qqc2Y_-mYPiu%yw5E8g}DJ`24E-|Bqyk0bQX99#ve=8(H@}Qb<o$qjSdWm
z0cpscbpthCMvYX207bbXh5>yzb1aE+0De(O-2;%~Apk)KTR<cIp9@P+aOLPxV-zQ0
z&=jCuEn(z~UlWhrRfInyRQObsvDUOAo&|3G1ROWWjqY%@p|HU(>YPOBT>wV{l;qo3
z4NzujDTv!Z1EO`GX_}>F2|T)~KXPIMMfbMk{Rs&E2@H<Le}y(CRlugWX*rw^z{ox@
zpp3P6omokJhAsc-ah`(wmD_c;xXw+V+0t4P%uBz6)s0yGdk%&iWO^v|DozM)R7Ag*
z8X*)AMwMo3^o>R+{a<1^;5U*1VSe2-Ns3-j!+<0w1V@K)k)xQ%6@)=f)Myl<SLd^A
zQ;vgLPsoXCIQ&?WE}AsWO0z@vsv?c(?LbrQjMf7bT+xw!Z8ut2uaN;|5vbvO{_sFB
zz)xcjH<WYF783ep@Bsx^h#Wn|fR@99=4o`@x{;n)r`>-z;HZ2Me3tmJ4jjb~{*d!N
zU*l`rA<Hy|@C9yA<2uwpKTPAOm$q`k(bYwc(DUN~l=u4dY?Uq6@nc`RCY>>Z!2GRC
z1I60t&rp-G8U05;OX+)$Obt%Itg&Y(^X-Csc1$Necj|W@S!KR<Or$joBT~*g`OtTb
zF`xo?e7P!r`geE(=Q?43F1XjQd`=-g$Y$<E-}9R%JI1vF_>U$-9u3SoEg<9Xs2Rr8
zh58L!FtvL!1o)?ycVj55n$%J^J~AI&1A_olmO9WS4QaO{<?wL2E8lxKnz#o!y>?7y
zW?F`(z%`>1?JqT9LjCYlA%&kk6ZLAAeW+31teGsjhxvj^Kf=3paO;WQfhdv!^-3ir
z;3Vy5y(pXOm$<0~;tH?2;%OJPXe`J670ko<k!*>1Gx3|{`Tklnf&DNN--R!#1j_ux
ztcWJlh4M40A|9(zsXl+k<D7CqN_N3=scuYIOvqMhCZY2o!4xqgk;#C_3mq9ySs!z&
zIesaK6n#Lujp~bT0VD$+A_EeNp$p!f1V%SH00Wp8dN9@6eMg!9$S~Og)8t9_mnP7e
z3_*F)9MrcSn95av=F8`xYS*R8B7E9q0*=?Fq{QYrt!Q?7?!6`P^&K7!j()}7h?{M*
z_HS;E{a^b%Us-T=nw*^g<|_;SP`bsuCHvbfSt~jNNxfRx)J-qoL%(Uy7r{RzSfHj?
zM?NX1xo;$dKf@}`dP0y<s1d-$ncjG*ffnFMn9;UE<L#%HFd!CSt3rBH;KnQ$1{B9z
zH#0SR&u7NYZyo1r5+DABv7OyG{jb*~{&#sCsaFS7OF3*>PqsK~CoW%h<>4xOTGwWp
z<=tp^MdT2ee4~PfX=7Gk5>B8uz%z@%Mu{j-q*`S6OEBO6;tcMKtCJ#r@qi@0Qr;7Y
zAgojpS)ml+?D}9pcgRZdy+lg|{4678&#Fix3pNSn9++%A+~h;QEErE`eTEz<CO$#X
z7RYH)5tn8FeP|j@=XaSj2JKRR2&#_{Yfv{$4FYDJY7oFUnA`IcqTN;Wz)E2y?Ka%E
zJQ412sp+Qq!lv6|k8|y<Q&`Oc#}l;CI@PKxrAP{srsU-$%QC%uC6f2ugoBC1WKAhG
zhw9bK%F;G2yP$H6r$~ds>Gn$Jd0LS8(S^^EBsy)N|3n(#(jgCnLbZf9?kXmL+;u9l
z>F$Bees<6HQD1fQ4Yv#&hvzn%Vp2+QYw2S^Z=cTmk@hRQ9HQM*VLb<PWMHmjhS`%x
z=uezQXeFk7cQ{DroA-pPPV@PX;pU3T6>L|qc)%;&LD>YhY@$cOK4vQBOG%-RYO5lp
z7*H|7?G<W#aUE)82WZnAK{eEYLfCI2_-Hr=WQJNy4&6sG4T`hamtdZC!Kmd)`;&yQ
zX0=;ej*h6F6%gsi@X?=x#T+4sdL)rtZ1i?wy_|^YdxIoz&3d&hFH3KG2{T=PoA33D
z?f|>RFItIDl!!vy1gM@5!4x^eVi8GRIS$Wh<3PXv*Ms7cd>myPc4?iPzz5`#)bW9_
zPTGo5g8}o>>J>}kh)?e*UK`8e89aKJ^(Ek*&KW<bIq#U5P~&DAC44!0eqW}ky`5Qu
zg(G~OdO>!>NGFYhwX5sW2}&=Vt(%Xo#A>nkAXwiy%&fF*>5D>yeq=zW;ij<3e#;l*
z;RjwQr3VG{GTRDAs60seq8q;PiGOC?E60?fpNJPCS2`&aHH-VoQa;EM7ZptkUed68
zJ|KLV9b-I2?|9b``h|1{cY+$y9{|%hk{0r)?+0j~^4(5_!KslW9a(-Gm5+5;DbBD{
zlJ;!~e~`0P=DL*VBeqsEiM)m!=Vk<z@*2QUY$eg|^3)+#y6bzOQeG<^JQ<M67);!i
zCK1I+Py)@1bS4bn*9ebdK$UsKsjB4t63m!4Rd8=RhSI0ABz@)ND6mPp(#-$B<=VaG
zsIg%M+PKsTf2d_F9XUO9mp;QC;owy!4T!+`UZ<;7h1_PFfSY~2{*A+nh_g+I`TTm_
zwcJcupPwo5`FT8kYLJy*y09%qhBp4?+3DHOR-XOj)7q*c!#+eU=XDfri~wZaY8#Q^
zXDjc!@$8D7>#c@D_)WIIPyGfoqAIB3;1x-p3PldHq5qAR-boj>`NJ*!1Ja2dOW(kN
z%JZHwASYevUlq`=h~f{t;E#y^*Pg!1iv3;-090JR5?ROHozF`@4X+gG(}rv9m#Ecl
zQ6_BM|A_fxpx(M)q_g;0+`lp|e`TP3@2KCqt}6Fmtu6g`wB!f1=F4w2KY0I?Y5PeA
zl(!x`cP#(u4wOpBby<GD?5+BH;u0^`_vO}>l|K`Iqc~8rL;TBj%-Vu4zOs3%x&F4S
zP}IMt8efbe6loaf^Ou(6yF1jhD%aqr;rF9EG!)pe?`_SAu=8P>(!{3DYpRzkdM>}5
zVs6i7`#C0pb1Z%1=1-R42X*jzmS#Q|XL$I&e){)X(;wauUyh+)8_O%TC<ioSknV2V
zm+h2%Zr9`xQs913YQ(a@MI#uic=9ZqZnYJswcT<G^(Rr(<Dch@FZzz4i+U3>)ZMsU
z>rol0TPf>aZw=3W;6+MtWu@<`XF$4Ojd$>jZa$bQS6jkKaNj!j4dlD@U1qe>RJR<&
zEH!)<xMEE{M;t|D0Qbf|uyXd#J~}=;&DsnyEs1^N>Yn_`w})W-Riw~QPRk@-0%-z4
z8KCw*(+BOqCBQ`Bc@fbFaHC~`C1IS+M59yIaEmK)Mi2R2yW_b#XCY6z8bGblC<^fH
zcYu%qJMf8>^cVVevrv?_+Ara$-#9n#llX=s<bT^I@w?o~<Zv^sP@N3@HK*cE${TDe
zU~!nKV!d5$s6kzmMHs6`HBcoO&=uV&`la$22IK`@9M{ukK#Nm`sS~&f2E>Ny<ax$`
zBGTxDPF3M&JhgC<6u^ykn9$%N99azm_a?)5@?lmPHGAZUKiH2~f_O3XNDO_s0daZ&
zIY8%Ts`xK|nXWY4-H)7Vm~H^e|DOf(oU|}GVnvm`43E(@x%?n2xj{c9@l4ikQORCr
zv5kxN3Pf=-AYK{;1IQNpy9dg^HkR0oDS|Uf^Rt~h!X+shXpS9uwn3_UIJ^h98%Aip
zsCl{ljbfl)b>tjDMlCh+@1N5LbGnT=G&<+!F8HF;pq{?eIbATP&)m0Pl1iA{;d6U!
z{$=DZ#RxbwhNubPhkdLo;?TV_m;Op#sKzGr)tf_`15M%p>N2l5{|uAB@++am0d7^O
zbi5bpOn`{cQBrECOR`4SzIEO_HPpw4TNawi^eSUTy7?<FO%gfE+T}*tZ+1(gmYo==
zzltsuCJd}lN+?OhXNg|mU@8>d(YN5{|DyT(hZy!(zlh+(<Jj6g5O={G!{ee;wQ(4K
zCPzaJ5xiuF`^p<kCED{d+<$WJ{y9v3_v})DXS5f1+${+#WJ<Sp3}w@~9QhawzHhAE
zV4uq-pTYCZ4&AGD9t&)0In)~8z|AylH@_1353aFhom!H;vr9!uTCk_1c@|T+P^DiC
zzqm)+%wg8${}2Q5->}R?%F^GWM|UX$57wy<@NIk%ARvRV{Gm<;L|NPn_x0fW?gC<p
zp;<WtS_(F%i}Z~QT25L_;umOc9QA!2Y+!L67w6{e<U(JPOA=Ra7=LYq;W@WO!^ECr
zFfmKOn9F<BS;c#Ly;$`1E*t*DV-iIlq#ZuORy0BEf#;8nkmbf(59q|}*jlfDsFb=V
zN4#P8_^n@AUP_JcCSU!nW#k-|6>fHbNWGzNaB42@Ko$?7yL$bkdunBgh-58U-n$1r
zir<lLFuSeobdG7eae*vTyVGZk;Ax}>igz5{Y+EouLKW#c3L1L0YMQRjZ(K*j1eH%6
z81T#uwlxB~ke?Ude|mvi`~|Peqkf%7JwN}uGXL1%JIp*geqw{cjBiAwe5N}8(IZ}o
z3>qwUw~R2!QcD;Q+dsbFQZnbp1p)n2p3hfJ$}*8lfh!*Dq^>HI%FK}8qQ==1X_1My
zGI`}!d)>@)=p^M<sNN|*&~YqhHjaL7&VaxAHx&#~j{^94%18~JU@t@CTj+KlivY|4
z=?(;iZyZk(3w|cpUsgtzY;P&Q!1d8@nab7=VUK*TMVAYD9fP1n5OdT5cah&HlJE;J
z`j5I6?HT!Gm>Q>4hMwdmiI3hQIQ55}YsTuW8&%Q|>3n?PGWT}zR0Dzc?i+stYWcGe
z&F|i)R~9*P7&E;U6bqOr1Fqsjymga?YRbz_mQ<ZAdtBCbtl;dn-m6Ygh&PqwwdK^X
zSI9|$TGV8?Ti{XN$;^72Y5A+M%X^iycLhcoADM;JJ`0nT@r`hbKRRYVUx$AK`=Uhh
zm9Qtt-APmy+MV9?s{1347?9taXU=S|dtL@*?oqS925~qr^@N(~u1b7pLE}#q0>+Gk
z%G{A+d7Omwl9jTOIYQ$0Fx=(DW!d^%fM^#HZ<{8btz5<0>r;Z@b8CVhyJkd|Ace@$
z<z0>*x=x+7YvdnV?I7GMSj2HHDOKxb;PKkcOjpn7ebDk!J{adzSu~>&FutB6$06(7
z!l5Udn#+W)tXS9C*y*idQtIa9Mnv&;l9u8uDI2KDR|fL8kX-AcTU&N|%U6|4Rtjdn
z;nlFyEu^G*n-)lFwE<uQ%Lo4H?TqxwTl98GY?W+_!FI8eLm~Pm$F7w*7jF}DHD@M#
zmdM|AOcm-f!;NcsMJ#AgS?v^H^QdYltZUB!%+)|i_3UoD;A48qXZ$}}qK>_lO>rnP
zlrnhidHC^DXy3vTOW8wA&pCqG?kz)zra*3HR8-_-2Q4;FMMd)5%561t;XcIxr4J|U
z9L{Zas`mGBaB7=OLptC7SZFNxP+$UXx;}Y@9MpyI&0j~0a8qhAh!4(_76`{kbBift
zcAM%aFzx2KDZ-ZMt=_}^u<{|+n%odif1}9j^T5fr;F6E;t&dVy(jQk_6}2>%nu>O5
zN$aZ#j6sG8msW1W<4~s!Dr5o)B6+qHHK!*h(r~3&)pE&3NcX1^FB_e0oI{rII4%qp
z3*>Iyvd=YnR~2i2-JeYSG8caR?%!?)=&g~N=m2G|uD$t!Sl-&^qA~Sr2FGjf2d!PA
z>|drNw?!&y^-kw)Z&VMfH}FjI4ILr1_jdGpN~hx%tI_Uv6Ztf0kx@;n5UV40Eqik$
zr%eCUN`aRTJ3>&~VN1kiUE%`5StR?p&`}zpRFUX+n$RHA80r$xB)u_!v-kZO+}fi+
zM>_aqnJ6~auO>-K!nclz_p2Q|d;<6{lTjaKMj!V5ULu$u{{ML39`Xe^7RmzWi_G8)
zP~mW^G!jjFdmbxe!ZJ$BY;TKhd(0&>Y}9Vn_WlsJjMj}Az5a!6E2*5Mtf7SoSW9wo
z7woZeiKjswbzkCEmHUZl-Zfh3d|X@ZbC58Ox(B0FR1XXq--BFt$CG8~Z2F;-l?mjj
zV!XxU`*(DOu(yjj3*eHxi|Q`^wAJ`m6^Q2U?yN$L$cL57XerpWXCIi)4EUyGq-=^@
zzqCW(oywWCoCh6->g8(FnOPA1c1&!-WA(!7f>oYZclPa!V*SeEGcmrA;PQKw{Zka!
zxyWXd(iu3Q1Eg{@OWK-*Rj<3TVTz4e<AKJSkCm@Sr!OxZtsGT<Y7%lmg^aJbgX##U
zqi|&>%5k!h!6&@0r&wM;DE<0%M7VTbW$%mkThg!fvz0oUc$PB<FN9o0C|l?+FR>OR
z9j9vJ+u)*oBxZe%nPB|pb8PAHw&hIl?e&=a-rlRDjr`gZrEcbc&%uGUWk4~vNk|$6
z<UH#McO4#SDayfLrY?ylYo2UaM_Nk{wSM&a;mJi@hTFuqY;(N%*tV@@Kvm?~->rJi
zxH6RgC5=LiPgs0OkXzBPcFlOC)Eoa&6X!J9q!pQBc0#+iwc1|YUWh-G90c#IrXGn#
zJK2`VUn719yC;LOWr<lYro)*W>*6^flE55Ew(1^2bm1<`qI^|^PutNhHf<EuU6ptC
zsZ-=RQLHta{1Tzd_}6IvTbHb_F`$KFWGr!M(aywv-n4xwaRjD-l=6FL*{N0*iLVyF
zJ8vB?_dy<+hAzc<*R-FDWK!8mB%2f~jNVxqZhwAbQ2ub9iEF1O^003SQ?{P~{yZ&`
za8`x`yR1$QDcFChv>_0SNnT{|%#`BCmb>VzANzL063$m|UfI+xm0<3F`8n9`Z#<$T
z-9)jXv`0QeJCd>1Bf6L4EUTN7j`v|6wiZ=4pRi)%I6Pf_!#4B<if9seui*`x>77D3
z*^gX0ZcZ>4;Zv`%ldIECiLD72)Xg1T$7FDRl$vieR&|k8j#7%=>Q+m55>J)Q@~|VY
z!bQ8mioJKzIy#FEH>orc&TmnBVivv>Rd1$*J799pol~Z1;4tm}1Sj$+rkVXfja^G@
z$nyB0?Kf|BN;-zNY{>Hrw8P);-*)|_pXeJ0XzMu7+SzE0?+IG@cdk`ae!wqw$K>1r
z3;i`Fwx`V4Ll$Z63{lDGAVLK5;RSBhCl28g$s5~8OVc&)TiIA8h-tTS_6dkJafi4E
zI!aWSNw(?{^a|S*X#tqiWFJ?OUI%ON$9y5nu!oM%viz6txLKUtTC(a)v%uq5$zq1B
zi7C)Uc;raX?%i6KFR2GPXC6F3F}R$(M))N*A$H7AL+Ad<_2rIeZL}>_u4dw(cxR=T
z-7{NMv+kv1feK3ALY2WR$G0en6g0e`GtmMc;JOxumTKWt6mrRTx;VF*=$G5@T{Ank
zM!x4P&gElsYXTFO@=b9T*Vn&raw5MK?ee{r@gIAABAVFvL03UABHl(MJp3L;cI;Kx
zwrKxN$J^R`*%G5d_7w<+6^=v(V|lt;1ng1FPM%sMo_j0!&qN-|_5LV!11c5hG-Y~J
z&~dNBv)x@WOya`}Jc<l@x3jX;MQBK?6YrZ$-DmdTWJY+&<d!AT!JUw4U=B6T{nTET
zk)0nUn5%ws|At4^xk*yDRZ(5yup8^KYGx)~uN`=wKB_Ujv9`2UlVH0iUWn9_Z&SJG
z*@D@)*`Gq6{zgmkBj5MU1d#vO(eqUV?XR?xDfNS(fY_<T7_sZ_(8q(v9@pB@_HUIB
zx8<ol)scQEb?bzV<@9s1ZD%D9ND+GsWe)r9BPn%pPWfOtJ9_nM(uH`N%bNF{@#f;*
zax2JCcSi%4taWLd9@BoWaa1tik{*x|pubrutQ7#tP(=pjX*qXU2|6jimOE}lcZ$8L
zZqzo|?E~M^=kx16Yd*EY%`lmX1|1O#*>)}NrpFqNTjCH?YuGfU11>>}7}DS5$dqU`
zGw;7%l)tuS{6&02RK#W>^?Id}>o<(JHHF<SXix37>gLB6M?z$?ZZwrAscYC@d3U!A
zce|5jC(>K|s7Db;x{V+EU8mDNDn~<=_^@O7UD*(49FHo>0cGw|p#;iE5xRn22a;NM
zwevJ=B0aK4^=-8iT|-bnb@&lSt=r|mha{qDtJCppwEM6z$`_rc1vZu3)MK>URF{YS
zMk1kyvLBJ;{T^Nkt*pVAP-LGlv1tf<uj=vJoKzN-LwLD!3Qb<)NWGgC1P@w-g0fv!
z5+FFfBOXpSC5+%ZO>d64bfyPl?MX%sZ&R?hN~)SR>-Rd5UcQjpd$r7o^6{nPT30`c
z9Cc|GY=-h^7a$+J&te;E;OTrJ25FHW+M9i^H!@<EgXhbrV|~#Cq-c&-vnHKg;9dV(
zysV7Z<@&|AHR9FA1AX%SI-yiAw=%bCA{(L8fx46wS-d!Ili@%Z_Mw@?X>X~<H0h*|
z{Hr!R_sV{E#xV+UKAONw=m<s$WEff&<jaxSyO;DRHh5y_ZB`-`o9|dzzX*1E)JlKf
zx#ax#mdj|cBPiUPFP~M2Y}#F_*KGeHtDbt4PqoHBvtb2VYDxIscifrBkL%l=dM3d<
z%qG==Ag}W&GSukA490cl4lm1lY#AXZjc*fZ3a?kn(6+d9dhdcJlhe=c+-=VXjGP6Z
zVpx+5jGEe$p4}t8NhswnD|9!m6*gnJq&Irs3{{Tzg)>8eL%V6Lqy5RcyN)EGQpJzq
zE|<pkMptTP)W%12JUntkCxO|jCv?hWxS=06cpZRUeVE3YF8kp!Y1=OJ<okeg(I-<%
zmRPO}7q-PFzkS6rwxL>M+0~$eWhL@kwyGkNU%m7{?hA2*Tw*$!oUubeXQ|!5Q|l{c
zoIQK-=`$j+U4-!0VptGq=kXYOo^&wJNGfl{u@J#_VEL*mn~tw>FkX<(bkj8V4o|GZ
z)yG9kp|Uzyn+zg|S;FevILW7byWqk5Brc;YQlyWuWqQ`ro1$TZyW-lFoQGDd2$5IW
zV}I6UG!6O)A@jKPdZ(ydNt5Y$v_v=d3RFOx2C4kYSxxCv(TnRl&vMi^80+<*w;%Il
zy8K@BYGRz4o5Y6(29%BpWI)#VAQX$rA}4PvUa*O47bh(WcZA>X=M27!d3aEHRP^C1
z)eWas+&1l!TlEZrkDg2TFxvL)ir(fOuS}$u*>91Ukh<rejrMJq1mC&YNwK&HlZ4$u
z#-;8i305C0;cKW|(e&DZuxa{KkeR2gosDc@>^tu=L^K#4n_x?nFU%cUihNRDzWsd;
zslk1>tfK<9t@0d4bhA_H?GPi&bN&x)D}00=zdY`~!<NpbH*y*@62<IOhGcm@1LIT>
zfD*lA=}i|_E-~JoGr6j*Bg<KeJ;(O2?y~GdHgQc}PNO|=Ti}H0iq#ZWy!lctH3=gy
zeX7<f_(ko~y?R}HSVERJh|7)Z#y@Br5VWtqmbfcR-mzMltE5{>Oy{AV&JMN^PPb)n
zp+rV*r57sQtQ^sFdZtSwv&P!mT3Tm!LFnE*`K_CVa`fN%UT;-bVuNUdzhZ2E_`dfq
zd=K-t(w+W=@B15Mlj3YCG7&p@gjfIL@|e?s+66YA88^I-RI)770AU_}gmf9|ikLDC
zNkNBPLs{$8;iRNryK!xvBKC=v$)xl9wQ=%}nYrOSjraCEU(m}`)FE>PO|OPWK;TGY
zGZRco+J)gG?|*SqEVp*GKfP6SSyblzQ$_kseoP08k+(0WkH><wS8jJGl}n}@6F^;0
z+4bQBF&N9-<#I(n=fy2ezqO+Z{AL+pomUPGaovnqbd7)0k*5o%X)=>w9`%SLb9HAG
zJ#*=R2WcJXWtRDCNp<&f_VQ(3J1vpgP;s@CGreKOLHW14AZ{t+Ea6k`u0Gvb&AW>U
z846S`LKm)7%jeO-Q^TD@MsMgZPYW6-`C4V2EHZSm=~0<Dwdmd1wMRo@DhM3`kxgLI
zeo0?NSq~;AKpn}P8WI@D9n>{|Id4;KjjAb-+so#xLYb!hy`Ax;md6xBgU#{?b%X4S
zcuB;Ta?43w<3eK=&G#nmgD;mrT+`P^rb=NxxFaM|Kv}aBy^?HADMc;D>RrHFB^WFf
zTh6;oi~ZiNEQ6&lIvjR2vdGt|pF*_bnioi`!0yBqtLlPV1eE2Hr3=S>oAx6akjQH{
z6JR&E2=ChREJ2c)qlB5==Ztn!cEN?a><NILS#W|_fNN4Hyi`SRO1m>LWS|qgZQ-q(
zE1W|`muYT^-s@kT`@CCn|C+6e*9M3qdxobKFoK{1>tyIxn-8;kxfQ6@(Afe=R0}G4
z6UlUk0w}R*z*|mb-8PBulO!+=oo|^0n1#plu90a)RO#AVdt=?eHRlUTm|DWsS9rD%
z{b^*7KBJ~weu%;kTB;`t^B1U;cMJzubG*Z*dDWv|WOKBk?A?+s6hG#fsh=8}mLm>n
zQ|z!(w5xW0BI7{|qo39*Ki;XHr)}?0F4FbLZ$B4RXL*_Pj1rw~Y(!Vf7x^R(Un(NF
z#vbpTmE`;kRS%oB-SmF9ar^Yi-RpPUl9%g5kL)9|lkavTN{0PWnR+EyLyP8gFcZp$
z3>8hky%I4PwS&dRS9I6b(OV}cK<T{-O-)O7{qyIe^`(cZO@d{ajxArrb95;LonlM-
z7x|NkN_H#Eu?K`Z9?kBF*F_(%YlW6^mB2Yesf9~WQzd~YL0TX!xu<S5E!1tJW^`b?
zjF(*W?S9tAxK=${RJPQxKlL)rozSx$osO;nUZt69CmM#WvtrKWi&kT(J5p3P+R=2{
z4GeDkt=h-Yr=uozs`JLRQV8CJe8g7_w<MJk*e7UulSjy2F-}Kz9yIPGWaN+1tEK`l
z>`PQ!gswj-d&$?!ygL8zMXff%NG#m(7&(3dyB)Z$;OQf1X(vqsGMFb=ZjrmY3>DBL
zbyNQNoKA_I*t<`k?pvdddGcDgaWp0`h(GpGK5=WpYKI-JoFp?kJ9{V0M^?eGy1k-v
zkT|Dsa^%-3*V|imElhsGncL-GZ|Ns~$63OCVVlQ3=pi(TCX0&?8j>m}wMb^c87RMI
z?*RLFr_}l?$q$ZP`vC%%aZ7z+h^Esqz4Q1$vl@Q<UEpIi9|vJP9uf4nTMofBIvDQh
zRpX8JWNh&SQeuV+Z3YGrSaX_H1`rm~xrt<-Zf|@-tJC$;$Gc@Oot<cHKsMd7k#aVF
zA-bhpIj$*={W%j){@GK&S1BWaH6?@F7qXKpNLn44o-+P2$|79|ZVScCyUVv{R}(S>
z+du42HT1Z1&AaZj*B-C^_oEO4yo3m1lcyzIc*;#egn3QXNszi9ZP>pSL_BGM5k1j>
zCzsRNQ6s~LM*{Szt4DZqUArsBN3En~y4+8wT?E-=Lag409Dj7^7_Xd6;EFrX$F<1o
z_y+HgSUZod6sO%+*;Xu<g;QCzXo;0Qhzh)L{Ix`OnC4aI<Ef_2uLXJ*a5kDsMj=xs
zs5A6eYl(DrX41j*2&<C()f9~?Wn;pqywnU+US3AcU@sy6=BdogptxHa+m7xLx^sIH
z;s~P(09k6AH$;N{3V;}ckr{I1R>F#bQ}UCY-aK*j7|!u1my)Yxn;Z8^Fi+)wXCjoJ
z7M=gFd%j6R|L>luB^ErZi;ElQv~*!Rxr@4G@BLcbg^QeL%-gv~2tG#ply}55eF#z@
z%4d>_(Bi1gY689WS$$wPV065xpZD!?Vxh?EycmRVq6q&bp7Wl%=20BsNY$!5rSs?W
zQ1J%mZVfu}4;k-0oRwM&nP!0SSJVbn$3ytZdjg5MGZ;rAb|!CBwi$`&rl>2pyV1tS
zjW@*EOog>oo1GljUbo1Abr6KnMy9iRtH@HJ>U_m>cjU7aZCi;3MG7x4_7?F&4xP3k
zFV;Ld*?pn8jbD$jCF!9C(@pQi${q+4Bjb^V4tx>Ovspf_v(np5B&$j`#pu0eZqi=%
z+bf-qiarWzy6GtCwOW-QGP$gZ^pDJJpmIBQPL&T|c}+plUpve^$Vw%D>}r15rJG@S
z@lkpL(aev3IOKTihBFIW9%YHIAsC|0B0c0kMVf+!b_{#1TAlxPh$Sr0&qJ(s<G!#D
zp7632+mE7JkBKD0K7ckFKiEBjFDWmB5pE)VGb^*l&gXhGVH>l8Qr_Ka>p61<DPz4;
z^uF9UCxj)DaFeODNZ5>KQC>{oqZ%W8<V5xphjd3?U~D&si1a9Uvgk04t>+h8si=1q
zE`AB0M$Sme6UfX7@Ti=tcz5Mx;v%>8{k+XtB?M;k2+3zp0-k%j9?{Cm3)z;%+ZeQE
zGyBu)m&*=WdVxS&6zvAwhjY^SWVq2(3wPza^p!^_%RoI4?F=y0ivaBQT2eOIl7gXE
zp!lek@mNK&O?RUlez`@vW0Td2x1;i*&;5F@b8cAddEVy9#o`C7+jy$rVkG?~YFr~v
z6qpK48M>Sy@(RHeOpj9qDyz~-W|!$gPL8VegO0R_$zsin%<O>I>VA`&!WI?T0yjEF
zV?7I_5Eeke76=8nxI7ih$ni7vJ|@}MFK}Gz@Xij;5BfQ%D^%&uUT0ZO!HNUU&bF2q
z;kAAf5;1aBAO|#zB0j=U<1{9HX2>9EoesMN-QDtyvJ%+zYh1yiA5v9EvOLp+L`#|*
z!Y79gy)EO%Sm)QQ3d1PcsSdy2d!J~dbbEnw)Y1nHMfrSbb!+Pf(_NHY)h`|h-P}@k
zJ$r}aUhd6rT*BHx;5W4oeBf=WRSwx^WC7aSX!+R2JR8z!G+AT(uI7ak&TF1ZhMrH{
zzCEZ!RCxI2=Jsb5ld`mk6xbBqL@9<jMMUY%jhKu(Cae7HrJ@8ew_SSz7kUZhsK=*V
z%Fi7!>-6!Xi<(FB=I!nz%OQBl*6Vx@=HJfENR+2e+uRy0#tw~Hwbeu{+M)P=1pfO*
X7{h<Lo&6hjU9$-w|0(ZcW;Fdj;@Vxw

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/s-discrete-posterior.png b/docs/2_39/stan-users-guide/img/s-discrete-posterior.png
new file mode 100644
index 0000000000000000000000000000000000000000..5a6d6e422ce40d7ac68a0cfe6c76ebeb35909b1f
GIT binary patch
literal 27054
zcmeFZWmJ{j*FFjeiXe(o3M$(Gq!kGPB@|G)Ya=POX-Vl2`$3TukY>{@-3B5c-EGk&
zog2=)ec1SW-|?PtKAiuDbH;c-$nb_8_r2C!b6)ehuC<;iDcq(YqbDOEAfS-GBcVb-
zK%|BIBRvSesWN>tNI*cyg1L1|$@13iTeeoV_NsP<_e{i%ZH-J+Zr{8jD0q#4;L_`R
z1_mm3I4*rLJYit)rHP%D%-%)i>C=By47@+oe5k3dX{`BL^Y+c%yWdVye)~v3e3_oc
zqx^FXX#)4O-`5Wzb>t-~DldY4ml;vb3S5`i3=Isv66A+A%94rAoJHweWM%lOM_@!p
zc!t#2mN2j5+zgcgALZ(q+{i$b8EWf5U{g4;=~dsEAF@w!NyMyDD^aM4H$kx>=>&_F
zkK?h}N6gQ>P^2;{>@s2n4qoCW%BF5o&-^Z2PEw({z7(|LNd3=|e~z3yDfa!SGER{4
zbj&|SW|T&xK9o@x-q(Dq4)OIJJ$CbJQn;_7@6iJXcGdjfM4mhF&`0FJ$11^ktfYz3
zy^|6DoH|LmK!~lBp;JD#SPCO|*zS(D0|5cu3FIH4S!_EDtQdi`gt*!R!kM8%Z>c4R
zcNa#29vVw0#GU-EAf`-xkwy8vPoL_QSeCK!j`H#4R}0Ncf;Uz!8<tT~oj*wV?kTCM
zB=yCoK6jc0xn^Q*hdO@F%+1VwZ6CVYv6R+h84)oz6`AhR#8I7<o!!jd8~Tqw0TC(6
z2l??WN{nPgCILlEDMmo}=Ld_JPwgSXfB#QR`CHqEpwUd}`K`Ub5|iexANlvK@O}|;
zN`s9fGWYjB%!h!GHWvR*u^KZ%B92&b)}#2}plW_x{qv>Bk97_)pV6zAYR}<+L#*X-
zaQ`M^Nl0@&4q~t2Z_9_^Ma$oNen&tU))Ghk1pgaKgN0N0+b8y*Mb#{x3i+}BSwtKn
z2k<xO(?Ue7CGt?^H2(SFVa@Q5_WzHERr5pPh;)jq@<Q1)H#J^I@>%wGXX$W?ijEZX
zTrG^7>dMf;ZLa9*gmdXergbeB>P_i;nmW()ezz=*(9#N^=gZWqbSv`Qa)!^)X4Yt1
z6Kboj3N}^qUAA2xYRlAq_j|;#N0`Sil-7BedT;6yV6tdNhq2as;}}RwNR(NSyOg|l
zO=$wxm)5wf_rk=tfRxez`zTsK^53!by##?GS?Ype*c<mGNgo1s&Y23<y=w^wGxml`
z*@3-TLvg}9VGPIRkN-Qbs0O%Jp407ryf^2V<ER=La$(!&d)LY^>(A}y+xhn*=FSTO
zLJf4UtLonM(|ouNaSt!ri#IVeL~{Fw=BV@k#-CUSL|npPc>%@VBe$Oy^Z9={1Zpyk
z<%`G=iG`6-Q&acO7LVAxGizC`B!Q_bVz4QBk1n>Eb?>8oN<j3TjMCuIbW`~`crp3c
zXyKm66b!Ul>@b^VyW8WU2iRl%j*FpA!!X;ub)_SweEL+(%gZa1X?IC<_a42|;)TkU
zZk^RWvsi794X+T{Q>3WCv+&~OlBH}?6eXc=_4X*YMxqpjK04pDiJs12L*?WnxUqhc
zH@{+5x#IB4fszm2l)7?hR(x;b3Si=}mE@`Tr6wT{6J>Ej#vMOVv&clv6S3i!u`(4*
z1o~C`S^Pwl!@&2K`;hG=4Y5ub_=N7nf<yR$=7NF#-&^2PVx8rSBPwzjSIE$Qmu!22
zhD|LuT|J|WafGa?)sIaxd&VLxJbXOE@G*anx-3p73tcFeonY&<w5e&mx+I)0Oalvh
zbMCa>Sl%W_N3{u$iS@y%YV83hR?W8r>?S<B$6C2Nvvi6#HFCWK$H&J9UY-&5V85zL
z9u%!2F7BK2IHF&#OBPoSnSt(&nf(o1st)t&e3A9QwmJHl$;3Ok-LNmgE5hqLrk-@n
zLj**bNojtKsk;)?lpCST<mmUIu_l}90@aPrj-7XiuUuT(6sVf@DmRsJr5{pPiMopQ
z{%F_Z?!O>1IX5mhGZH7izH^ZCt>Ea@L+lQlev$a-kbvko-~O8Sz+QCl+=h7j-xm=7
z?>PLMPyTlt{%^;D&BQ^N-@4y;X{?F;-C?A}95{CUa+vMT_C}H6$H%?X>nxH>Zd?4e
z!&V*1^6&U8yR|+(rpS_fL90=0qiZYh3J;RR<WLz0GDjQ32c90kU=+YQDGj-1=9u^T
zEvvQOynEL5K}>l)3yM>nY4>{cy*;;?w-)LchqkPS$_iUoo1=wwVnsa3tbdZ=i%MV@
zVVGZhwr+V??~BGT&UdG-nrgvKl8<T^T7>dh_jmReU{uY^_0-^NT8j!Td-|7JL~RF(
zZHob^^qqTh+Nhu3Yx_aewfhO`IeJ6X=CAj$og&;RM|6T}Yw#sjl-qLK?JVU)Ddz^(
zlu!8+o!=mN&G;NRnyU~kR9h6jY}}qC%~di5Xo=&EYD55oKvDdyC;k41sb*3WU-Cqw
zFYaT}6HjL@uDOhO_VB7WZBC}pk$Dx!V_I&kIleN%U3Hne8;<qdT78t?ovk~7F0#T1
zSzmKsow>R7z;~}qmU0GmUC?53iYAI_5!w10poA%470J3=+Fq&VgH?Sh8_rcI6UwF?
z%q)+9u?6j{xVRb;3<W*!<>;3jk&aXMH<u@i9xNDHcBQMg4b80p46trXkUYHR%D{Ib
zMK!>S$(FateK})Ljw1y=C>!nS>RLr_YNS&s<cceU5@M@r%vf2Zk*aj=-e&OxFH!YE
zSGPr1`YGIWv6X4axM%X@$k&NiD9hue1hSf2m77x;2sr4#R3_9jyr;k6RzRNpW3AYF
zpnG#=di-UVb5mhMON)SfkYb2McjhF&?O@53Hk)F3s$X^goKpSaH9OJw&g?^3m>Z=q
z`C+w8t+uJ29DQ>J?t}}0(m00Yu8*4!xhK}X6evnDuh=H{NyLkk_E*0q&lTC-cF?Kz
zs*JY#aCk1KCV^3byQV+i<i*O|(15&iPdj;}C7($HHJus^&6j1(B!PbQ_KBe4$^rse
z@7b`GMip1;FB_q}k%>viOHx@1)%EoXo$6kmQMS#tjO~9*?wRq~)M$5eLsx#;x9f{k
z>-_5p&yGA!sx1NYLsiw{<aFHn-UUe)u%o&w<PzfY+fP3N{2jd0@|=cUZYf$pWW(5H
zii?I?-fDNbLvd)b;=;>o=fB6Jt7R@ImMtnSrB)=dlQDX&8%7md>Sm>>rj)TCsnoaM
zJa%1qan(H~?Zl|ya_cR5Q7jXQyO8>%rpgwtz}Jje{ZUgRK{ERlK^42v6hr2su8iPK
z&De;$q?h<kWT~!ga!iMKF<+cNmJ>IXP&BoM!ZC!{xiyY%F;scKGj00d-Jv^w(#>b(
z+rwl1yIbD7ExNDQ8rXA6q^4!A2VAS=-f@LxFp#R8Xq8jBDnmauom0)vDbio*?jk@G
zI#F}<jL^k?E3zO*zHd>sRT$CT?RVt4%aTg%m^umRd2G@y@f~~~@L{2xQupVsT2Ih*
z9SyblQKInPb+T+a+u?oa;{}BqD{93uWl{;*rDOEmvJtQ7X*w_>6xC;)dksy-6hq#V
zr}7HL{c5UgH`ht{6-_V4ODLl3wGY2-NViXn5)bNS-EgXB3NLXfT^n|volJ>!ZZT_q
zi=lT_sljfja#t<~xo0^gFW=RiWD2I`(E2u}2q4;vsoAxi_3`(QPX-P#dIaX)iB48)
z&$0`lzBak+uW&e1r?}br#co|Aqc=m$I_~=IJt#X!P0CrLg2L91M`~psXSXf5lkzJ>
z$a&@@gKPU+`gvO3vR!>1O!Heot^x5wXVi?G<@tNcG#I_N@8|hQrw|rSl}ptWpv%J@
zP2$^y=PQ0%+rKJ{jrSAcoDZCdL_Z??hEI4B8mIz6ExkCk<>jHR`RaMQt<_l_Qf<PX
zt1Z`m5+&RemU$3AN*S^`<f6a!W`eQlUHjB5qv!JL8>_wdl3n%av)a+>u^x-hbiV5@
z<^JetZnu<o+np<)9}Lxj=gk$aP3w4Fwy8)S5rof!z7Kp`i8$tY>emndjGRt7JSTI*
zZTV?VIJsNFxgDA6o2P#0O*|J2V8##@uYbG5B|DR5la-i3#P^&=lnrZ5w)lKl<Th1r
zq18hLPPW$eS3y!Ad9TE_Wt`+x-1~-ayTmuib1PhB%B1td9Yf>B_%YeRQo%oh!xf7Y
zS=licF<%1DnV2|iIgu&7e=GIHQM;Ns-4g4JdmwJm=12XD;_UU_rUeyL%7!7ahGgi2
zfW1T+BW6w*)*aD4`My4wrJRYL`qv<&-R!gIrG|1(pE3R$uz*&F-RDb{6B5=4mf5Px
zUzO23>i#6d#7g<=`FpibUtWl<6O|>bd+&^U6P4ah^-{dTiP{z!J#+kv6f!0Or!f?n
zKVJ$rH~F?-zr+-9ddRUuer_X(NtTS{hWFOM2pOl;6wksjs8veOeYlL<wwvODcCKAO
zN`c8?e_DgI^Fo{{Os5Lu9@@wzGSTe(ej?gSc<A)EsmY<8TU7NlccbkaSWUWOaY&zc
z&Uc-rb-v{1(c-6e@kd5I;<v&T-d?-HqM^mja{LMOc5lT}3B!gp9nv2kJwvbgNIz!k
zxk#2EjW>C;{_5fpz`S?l#biIMJEc5&L{3*mz+Oln$UsPJuZ}8k%5~H^MWQk6rv9vk
zum0^fw%14Yo8>5EatbbfRM*rsfq<(nwCHN^+T9w7>aJ-go4>f%iHV_Ph~i1K&@%Cs
zJ=r!Q{u**U*EzDLvPX{|JbK#vmFP|hC%WL0Oi=;J9`?3ZL*lrMulWh1m${I7Dmy8k
z8=uYKwPLG}>^T)io5~hF+0kdbw-;=VD`q8RlZZ*<Glv*IGIXw&BJJE}^6JVMt#zQ%
zW0aLi?rv_3McEFv=<jZ73NXqB5C509oHd}SZA)^FK0nj<E@fr9rzOU7Roi``mTt&%
zm)c{$>ib$z-O=$Wbe=`KFy0K%o4kEN6Ybk4+Y}`rFeDJT&utk;A;Cqv>%P<^pnDnX
zBUAkK*|AjKAnd@n`HylU_W9vQeFtSc_ZS9(KhdH^aq=>XIlELVTG-_!iCW~KUi?jS
zI=;_S@2T*kSYr#X%v^syUFEM^=6K&Bn|E!h_0};y>pwB`X9ScfA=5zb7KZ!DxN9fN
zXCCnwvep2OR9V1dZn&~sSTeQ4A6l`Oew@~Oyp`{18st59OY?Oqb=gZS=R$2(GfTD5
z43;1K5A$1eURq1^G9N6-BQN1CbLuk*A2si3wc5L-FB-lDa&GN+Gd>;_{_1$4TT|b&
zYv`M{=J8?gU7YieXDt0t(5yZ^J-#|Hvcs>Vn<lV_%zovHnFCBYWjj(eRJHztT;Msj
zEmJy}IU@VoUo6zeM(O5*jS}$&Ae5*2&@Fpl{DJfu$4p1ct8(4u`*ml=t;de<MGWN^
zdSq|_T9suSgwo<LP-Lxh4#mlqZXmOU>ZI2yvpsis|Fs&(Yhf7rKRu(04rZ=kbohED
zPdOc8qo;4`cIE!W7$IRB%BG(Gf~)KciMkMpR=M-6QK;boJiw$RK_-txP4W7IM7ycJ
zH?B}VioZ!gc$PN<;8lCw>;3z?Q!IuK!`w-3+ud2Kp50z*5g<YTT_3QH7~U@`opqnN
z;tpJj_2++1J+fJyneDj;4YCfoY>Sinf80K^uhM}`NXV_(6^ji?9*(vjR22kkIqyVv
z5*pvcbBa6?r~*}g^z~(1@2x0H=key~S!}po{`-fA%6rCR+_NiTJ=nbL);5bCEyB&Y
zR67(mr)lLGJyuetzy};k(6?18*OC1#qy#kJXL$j4jdtt%dJn#nM2UEo>0FQ<aKy&j
ztX?<%LRrk*UE_P8Fyikb2tcNooR-7x(ivgrf4-aF$UJcLT1^#hRi}f~p~4Z?hRzFG
z&cRBuO6%);To?gS-+4;34>oi#balYC`X>L@TzL+8_E<}-zPUTligE+bZMZ|8(e1DH
zgJiJUF%mMEv9A%2PFDEHu3h{`k*p%bamnlievSg=ky*4JD00duuHIg3T>Z>o`{|#9
zM^1i=sRsC&XYIDQIh|A9gC>mmd-K7tg1ccl48!%kN+&HoP$^t{e^hkaZoD&1l|}V8
zHiHZxf{Oc8T1w{aS7-TM7mYh;{BNJWFTwAsqKO~B8c2@DxSr~ruB?aZ@i$4fHpXM`
zZ4dWrAt|r?x!v~qg3ZWEkG{TDv&IZ6W}n<X#YSo2gFT++jhoglvKbO$5O%iu=U0#{
z*Ojc_!`ZsR9d~--pPZdq#Np<>UyXi0y#J(}MDg-={iO~m@U%^(otfGlLXHzkU+##9
zdF^8>u_AJRem|_BMpuSrv5?c0arvp0J)lBJEQ92NcCtAc`dW0wgZpWrMlAau-h>Qe
z8GWv1j$ZRrJxlCY)2mZ2@$6g(8!~U`f9Uxd3g)^viwBKdhFw#vdvoJE8}g|37amkb
z=F;f~aS}}u5|X*`<R}}SA%|S;cc#hiC&EHt5>?x|iZ0+8y;H34p`o#{wt!KT&k>X8
z49mxn&+elgRAB9&vN9s0dEzf1NhODnP)}2-1)SKK>fSDbpAFQY&geIM`6vD5CuVR#
zYrb+VhTE&seQgzcnrx4GBqaWDf|R3XYM?l$9Bt>qknrsHe=bxa^4z0>zXJUUyBk8;
ztv41&o9{GUg*+=2K;M^d(rESl!y|;_E}sAW@mMjle57$OlhYm<O-TG!%z&_aZNyvu
zp9gb;rQL;=nwG6`zC&))S#x8}>KPgVqXW={+`uh&*u>JFGpSuWNr4AKs29XHb?b2u
z1^3s^d2Eb^TmSy{o|f@$eI~ll!Wj~C#F#CoZdvC`0COA(N$L1CQsYfR#PFVtEVC^Q
z5^Ue=+h-@G@}AGV@*YZ8XBVDK3LZ4?hN?FU?dsE!Go|CgE^`I+d=~57*<>_qCEwI#
z?@8Qnwf@@JxYTtyc&M{_^WH_eF%~ym20zQznb-1xX)0I)Z|f;w=??ZioaBs}w{^pP
z^jI_SR{d1?tC7390P<{^pyvH3S&uSkNlQEdU`qcwz=6BDG8Pj`8ua0)r*ENs6;UzL
zMoGVa9Q!z8CJb1{Jfqr>EzMXQO0{)_-80N#bW0~@ha%th<0bftlMN9IXCWNj=PQdX
z^N2WcM9NCa=uKG1Lvsl!$7O|?b=Ee{7)<8uK<+8gZ|}1Rva0-fo64T4MrC(O)L`X)
zXRu3byBg7i)v?^plx15-(LN9F#DQW-9oqiWnVxGyJ@f)LR-(A&V2V>bN#-}mD$nv+
zgjC{I2W)gw6r#yy|AS<hM555|H&F(ptf2yhZU);5Lx-`iSDIST1?H~<7)6J<O1=>Z
zZ_T_Ln#>x3Tss!vVIp*KWnbC4?%S$xe2RkRGA1hn@O{Uy=bG-?PxYjUdpUZQGgVu2
z>}3%>`1ut3Oztn_#}ptuSMrhfnB}(`-!XZ+e>D4{;?9<i$2hzXcOg;U_KMkGMG;(-
zQ>0-}+NwO=JFc_K%ftxS#$X8Bp}!vRJ#^OGX{vJ}e*uJmf4{&OD~2ZGP?XUmWle2A
zwqG)zJ{OxvAx$K@ExeEXgMcuV2{IJMCkQ14VDRlf+&{Vhiy$E2rFpFOvr7Z3n2)-`
zwWIqFE^`7^!zTTBcK_J~K?`X(TxYod3u&VGtcQhGeLM_;_DS*)JhjUnlkM0d*CkVA
z;qY1YwhlS`?!EV_&}OLY6njZ{r!M~Me61)+uh+oZ2#xmM*;wt?vF!uiS%weB$URdd
z3++6K4ey6U^&=+SjX)SA4t5epmM6d_gln%M6bVotL-Q}8oQFo}zo2uOXpR5sVT@oV
zKUz}W8g#z#vz5t?x@3juss4EVkj>HJ9zFL`ou3nbD3j5K&?2K+vaKI4xUE1FbA~+A
zKhc(G{o~WK?vndI<0vjM5)$8}UWbA(8)=z+c;*b0g!P(+WCO@nC~aS{1DU}JLiHq?
z0E}~QpLPA@wmPFdHCgR6-Q5Rk%1U6+?%~$->RX5AJjf=ZeMN!C2#NE>#8eMdbXI%s
z`U`ljt)?ul54nu=y%lt@g4&&ST1pT3Y)ao6B93VK&Zi+`TJka1`+(rlG(7-#+M4dJ
zoi9*e%`XMVPBIACy!C_%J?9j0C>SK!d<6qQaeri-DbLx$o_^fg$h@pJwI77dOtzST
z=5d8c-g+`JvP`HB(j=M-6v{r>`Z*c0=D&4^|1dGFK@$i}g63-J*VBM5o6DUYC=MUD
zd%2%^+nFh~e7f>YWP4Jfe$*{4EvZ_>`K4dE_qpVrkjvcL_B>-LJiuw^L|zN*LUC2s
z_Qq0{ezn(}#d~A=a)b9o4A<=9Fh|3s@<8W`{%ZJZ{|k7eP`T1`?lXB$bZo>e4JKv8
zIGf@!NNKEQE7yi?A`Z%XjFdTLWa*Z7k2OV^MHXXM?iZ-tg0^M`sdy(Zu^~_Vr%+#H
zhp)$yy>*`}wb6c@kzLl1HM|4R;*D2>4BAR<+5kiSSM7$5z#xV6#?!0K*FN%o_k>{L
zH>mMBk#ni@sphM*H)fu(#9WSMzF4wI>?*NG8dy=0IubX2Y!woCc<7k3D7s8MquBCg
zC@n4I(`=orO4FuDJ?FFgP&Jd5l&R)9T|;A|!%_~AV2srf<NK?~7-`eOICV@f+<sZW
zs+wG~p#@AJ-w@9&_DTU*L-HP?zyJ&pC90>ZJv(xl0<XahQo#~xPZQIXUdOK=nkra7
za#Vu|{3#y|(-~Y+IcD$|B={2f9@LTz<H!^7+$>=b$(Th@1~5M*Obm|Cm2&tPE%%m&
z%lu8wHal1sWf`O_3U4RI%uM1K)c4V=3^gn=A*fDw?Ya^q$=!dwzC9?Ug^k|nNLm9(
z)@M`l{b1A$_o8TrW<eW@tP_?kNB8Ue0TLpeWbu@P)ktGeCnI3hO_Ycw(adl2B<F@i
z+%fCqIQkt{;SA7F^<w<#`=~F31DR6wzQT*QAo0rF-AYfgKvo10f?gzM{|y8bl%xhp
zVH<)S8E-NTt^Wt^4Tc)80V^*0p~Bke_;`wOLY&3dH&^D~070GN$PuwWWsc;qASN{g
z&?6SWl-ultX4A?sPx1>MICRvJz0#6816F(eX(_y(DTca7OjQ0{X^pnio*aE(8kV4o
z?A7I&z77%{qz>vdRYK2E<TS0JQ|_DvJ!uQCtz~Dbax#`4Izm%LAQeB0DIeb-CT%1W
zsDB%DXmQd!s^`?DZnLr~F#O=-vZ_YAQDWllS7VP`OvLjbQT<8zJSwPMRewFpmTNzc
zboj3!1hx<10hx*fr3-q%`Q)<1Jow-ksg!c^KN@Q?Vj;g#Uvv2kr0Zbh9E#h}LbJ${
zzW^e<C#zfUovA#x@L%#@gWUgwf%Ma;8)hrx{tbKeON@-j_mI}$jNqeT6#L%Nlemd$
zC3q2a<MxlMH!tkfBa|d0O%9vYw??*sXZCLS{5i^Ab$@v8iP&TGc!qqG62RRL)0WW*
zilg{Aze5T$asS=;9S;$6;=L~kxrNF_z9c+R;2QDY57hFhZHl?Dun~8BZ*?gh^{vT1
z|LcGFK<Ov{(=B#M3b@aWR96E!Rb7QHX8?N8p-R#h-KrP#Ak!z2H~eWIb<3UGIGCzp
zM7^s3MW6e=185Ve*IHI1keRGOxF`ZS-4XzHp?TYfA-X)R&$|PG7CS)=`xRM2AN89+
z)?a*N#&xDQFIoL36&CP*FVb+B_dvZdzaaHbE@Tth{^3#wv+0f$#dvGC*A_cQB)NDO
zj=c;jgn(U-mz%o<a>~3{#jLRN%w<lw^tNQEtgcbPUY&VRA`3zNp%zXCp8ru1k~?F(
zcU%XmJj<aLHUe0t>?m6JSy$dx2*oCfh?5%|IU@Y>_Ff=<(e-UTkSWnQ;<*-m1}I#c
z<%wvQVd)U;Cu<?T66*ne;cq?%GF4L)n8ucz%?Z^=G>6KZ-hrl5)&!Q4{LcXl1FiA5
ztbwd>?o_^AYM>~r+B9Ez3?Ej3<wU-m9h9_bQaK^35bH_9?K%X7h={Ui*&N<I*N*uJ
zHNS2Z8!76H+Q-^v2xfNM#jOpo78@H^Zm0sD%hR!~ENne@>+uoR<+6zvAaj+;ll{r6
z|4mhSlut)ZU84FLQB|6n*dikZY;9n0i)&xd>11eRadd4<Hg-%SrF?O;k!tq2bqDco
zLJH44;l>4tS<NAKjZCeViSMQ<qct?z)67IEu4;?pbg(is129X>n(sc9rOk;alhOF$
z%CAf;u4p%p?Z{(oSF|;LY(+gsul{PP-3geFt|7_<V!LGySo8HYwxv?$;1T;q&PR)(
zGzP90v|x2RzgORW4}D^x7jCmudbK}$Lv3SYgQ<Bb9ZUei9zQ<^%w6f)o{Kxb`!$lU
z5eNxYW6SU1H{4epQ#i<NN7dPg$e4rA)XHJ+E^cziiTeA=i681056=;$=T+nj58Amf
zw#6`Xwxw%_0(z--?2D+5BtezUX$h)2+08~L07a_+e#%ErycF4-2oW*S^C>iGq@5lr
zFOHhY#?tY+STuyNa;4Jqng>qiDA*RrQuR?W*_P88z(grWY#Knb#MzeJ%RgG*-S$-L
zLi@UeXU1`1Tp*7=YNExhHTL1<hGYGgf<>4-ysxGSZ0E=_FXv1JzNFUjaoS#}8vyyx
z+AP+y$aAGj(<V9{{@Wex#@^p*Teb07-rg4IMa)IE(4>_N-o+#WF0WY*UFCIX=msiW
zm#jd3oFlsxpvM~OIQ4a4p2Tl>+IWd2y|_Oqwmr!2R&>}hd;Kkk`ijjhbsOfe#5gVQ
z*=ZTI=(Ej_l6Jbxq^z4it97$34Os_U!I*{Tys+bK2+zVrJ+BP=-KKUFmrdqod^OaK
zBtl@Ix;m}bSZ0^svYR8+Vk=CbHCKV$!)vjgWwlOpr;S<yW3+MW*o|L0qv}dej-HWw
zf7infJBuhRaqTUlyEA83-fOfc#6N3KmX8`V^}H0K(lmec2P1yrxgr^J(n91tz(i}b
zS3E#y{s+GegunY!-8^Oeo9N;^w}Nc|XY@_EkP6j@!^G=qODt@``%Wx(Br%^&8@uhf
zHCsGaBxqZ)@L1Sn4SF#5^2U|abE9~f<o5&l%bNenB$#VxhYUw2{kC7DYc^TI7CQ-}
z3-o)*)m{|^>RLsajD%(<5Ffkt!7Yc5oHWq9d;gJZ%9zS5QNUA7bUH$vXw@Wj_*^yp
zYDYCw_^&LY(ncchB>ItroVH~2%#R#XX};<;5j?9THb+LxUth`F4%~4&r)=Ee-;qy6
z?0<b}&krwAT*@qP+q%u@Y5eNO>YGV2i2#i(y~_S39MIR1Pq3l=l}SC3%h_E&o%mYc
zA7Y&8{@kwqq-=W=WVyBcvO2XTI-xgsIw~-j2y;01qUS{XlYTmV55wwR+`Mj}x}N8d
zqLStYhML_=O=e&d6q%V-!N{6y69z{MW#2Hb5MKJ^Qp1Hr&Hfk*uQRvyAdv4MEl|x&
z&bmf98DuTjm*hHFg@$qkyOQ{d+?FQ{p{;eK8s}5iWiOn*JuS;5Q}#nCc(yPwRe^j;
z#9#72#`;f{CPr_LmOZqAS|dJj8+A^2;&WqpRKc4RM99FLD0itf;HiTZ6~R;ga1YqF
zQIR?Yxi7DqAK&le?)ng9lpNOC@BD_KZR_V0Il2$UpMzJ6M!=99|BhT8@Me|Q*1g|H
zp!cCS{K0IskHej1eF!Xz*RP%2n|#VkT!CRux%^hW29Rp+7pK;@v~cTXlyH_w^X@mH
zU;YrENJmf831jO$05~Gpz8y#OACTj)KfS@^Ss!2;tiGn)d_<Paz#qJS<ChNm^B=N=
z`hipeL17ZKzUN9P)Y5veCfuD;49UgdAK}+yguNt7bu7z)Op~nY{}=tG6U7G@%np`z
zoC1}fKRv8~9RCb<2oZqPa7=TT9|?Kfe&c(d6NoJJa%bylDE#toueE)Cd4>U_MvwQA
zmLh*=8iNI+Gu@pvoA81w9|EYYslLJ53;k+uVw9;#t%J6pZJ&jkhmIs2p_^vDUTrCm
z<tqIIl#Ma`lC8~o@mD?5+`U8ixd?oQ+!f+!fG&!fN2wLdxKE8zG9g4bV&we!O7tv~
zaR1-#29Ye&(S_-{9t*XH+22d;!$y#D6zpG}IBmT@m;XZ-Pb4h)0IG3cf1%~9A30a9
zC~i_QIdrg4;#N8aN=e$;n_l}Lm__7Eq9wZh`ysOMIP2qRA%L%%HOpDvb9LB%(GdWv
znTO6kr~y9IqIzc~M_1S}>U8$`BZ;!E$M#+Y6uSY)E;Ip(U`n_vdGgo?{QI9~5(E!k
z7%1dLgf66qZ)XDQAD8Pe-QM+icCeZDWn=NgLm(H#t7H9A)m{KoYQyyxXHJSe>cZFY
zFt`(CW7hCC(=P}SJ=>f2k>6(Ur1l+pyjzs?0W$h!)(YUbm;u(m7-?&%?fxoCmykCe
zwgo>7zsI_rFtlOT&=eI-XO}lu0Xu!Etetu9@xB-$lLaF>06E_nf!rXj^n;#=FhuXb
zA@BhpnF&I>LX$1<q}(qV524kN_K!_H-MD&dXx?hD<bIKP+j;l(?@zLzTj%PHymId`
zB2E_V?riB4+YBMCZUKm3w`<=x<K1yD5m~La9Q2?sT-Dq5&~L0+E&Qlmok)Czo*%B%
z&E+%vK<v`;0V++uB%Q8ttk))HJ07oq6B6_2p>m^cuna7MyR_dVTzAY1P>vyY^>+S{
z^MG}c-S?Xamp<5H_~eW!5Gh&Fo~wNtrS_x7FPIr2k^@7D=(~2o?BkhJ?&W<J<Qiab
z_(mF&#fT{{aRjas$aeJs?JW-^gr3_Z69aH{<k=M#Q*<AG)dyR*#)&&mXBGFqYZ0-8
zR9{dWg@8|#${#Z`G^$!)XhqGIEw{^BgP80aA|FqR`u#$zACw(+p);v*GKm^O>J!+5
z6!g6jW-%B<Jwe(+sh6iQz`j&_;#PVP=T9$qI1py(-?rTM3#rvW5hv|kOuB@gUJSQ&
zhl(Kjg&h#aRUx;}f#_R?D6-vE^rBuK9I0UD(av<H&lg$vM5U|Lm7`yMGli(ox!cxj
zi9vWOHE}TisA{rY9<-2kFGmr=pHk8naH|eLU=K`Wvi8I=by(H@7a)%S6(-~YLb<kU
z8sxujAV}!cqTQy_lCh^{WYGQbfcdy;+SQ-_&1kA55Ffm;HRRdQR{Zo!hn~6)l0sNP
z)AuKgbAGw2vjZ%iM^5cGuDO3+uZ?@mU}NXo1yTe$+xh6(>m_&V?o$#eQ#}B^Zs@K=
z^#FE0O`%@Ec6e|;ffe{E$}g;lS|%?80`YE)Y6&l`FNzqUS)l8@M=4p()03%vZ+E>;
zl<1G=n85cgqGm2$=A(m_(IS7VOH+S#PIh9p!(xtIGuu&<4DPFE;Ohk<(ocscc)Yly
z?|(j(Tz+x=wWEG(#8opi-jD1f#znTi6eTEKI@ys@&Q9WF@=tp$o&Md4FMt<%y?2%v
z|IiB`5|3T~?V}D>Fr~A4D4!Z`BJ@8!Eg4^_4gOuN&ibfMozaQaYr`gWW!#1c3tT8h
z{tXjqhy{~dAAl%bb5ZX7S_AZ>HSsi1!cK7WtOsw{f{p45BZ(#zi$XzL;_dPd|J|+X
zUCSO_m+0BNy0f*5QkYvH_8~sAVrI<73s|qzXxXpDAFgJi3RR_FT-n6DF>K;cJF{XN
zC@q^nj5)QOS}?_EOjhPmO-svJkDA!wob;r!?vbRC*_1o95h1%gJlUUIG$kB2vZbt8
z+FQNgNe>#0yCo|kH5=N|=(BQOOZNb;PUK|456}fpRSO5T;3_yDd-+-@Yi5NpuKdE8
zrRZyT(VOP$v{t`)Sk!5D^$Yc5#iAZj)hhPk?e}q_4&tkt=_cdt^W%4CM%G(yf$p6x
zr;-ls+V)oX`ZkCPc3Y0MYMsFM*Oo!ob%Il6C|>ytjwOt>T0%9;n{A_#JNM;6$K_BV
z|7d{;f}6gsfnnm_Tn~3Xo}XBJxO1Iks5-54gyDOunb4P(yKGKmLOEWd)54U?jP&Bg
z7a|rN^1OfQ`rYWETvI?l(#1e>yRReb1%0ksu>-V}oPtz%!#IKO5o%Jtn)WUQA@v92
z(Jo<bJ4j#5MX9=+p+53oAcXUCB>@L7z;QEBye?<Ctbhjw4c<bzjI!z5AG?ue5RyPM
z&b(y3Xu#%P#AB;XUS+)O_Xh>I6^lPS*jIq*n2B+p%R45z-HHq8JeCscRYv|vroY8`
zFQ<#4VC@M@P7s<~XMhi7;4sTuB`aSE`BfX}Kmtx({jBG)An7t$>2ez1&-#cJkplle
z8gH)G1qQ1*Jubb9Uf^>#waV--Yj)pU;CO&%5GM{N0VQZ=)nAY{`DCBul8{7k5_Urx
z9!v#~Uq-{-j5L?|VdcE!Gkdiz@B&f%K`pcvqA#t^d+!cF$&IAG-}-`$S^Y52zEhEv
z6hT^SAE0&O*SIks<C;fteIF8>N9`q|;;PWE)aJU-X02xE<DNTZpZ|R>;qW68M}~hy
zx4$1-{di22dieWC_96O-59Hj-@fV-}u_chIKrb+zkyB-#QCtk3lIZhC@|8mZ#Dqv3
zcfW?~xv}Q5@mlp>&dq2210ZV{k;F28Fm3q8VO*LB78U49G;F5Jw>2)hizzG3V3Za|
z3|rsk+`ip*4f(?aQNBeKUWj^3q(h;>d%th!TlJ*IZ)%`6X;_I5Eo$8F&eZ0rVVhq#
z{RgE{f?<fg)bd>04sHvfY~%$TnD@s*8Ies-eEa?VV-~da&TW!(r8?hpiEpWiYV4sy
zfb*oY)ZhP~&|4mz<H*f@c=oSy4baU+;Qq@x<^LndZS9+5+3(SC{i3C2;g&MzSql)Y
z{mVC&5TphTP5}+O#sh6U35^i{sDk4hil!h{J0pHn@`NI*KBFjp>wKh{+F1;*?xj+^
zZkQ6|mhN)$Z;W;$F^YY3;?U946MpQvf)Bt85=o0%2^zypY6#KS3(XdE%z)Q+&G|o+
zrvNC3-QZEM$k5Cw3RV>9Bhf~ZlhZ`&xyRVrJJnlr!do4&JHLu>8!g^1XMg?|-J3*9
z2)72nMGG9pL+2BjhGMgfYM(p9I`2m?1<)B#{kJ%DIMVjX>Td`v62UF2cKlieX$TF>
zwCrfl4RZJ8iSLVYu_Be76*GHl2>^^dL?k=#V1>F%nL*M(YrFLEZba`dD`Uhl5T>xf
z>Io)DAYr?C0xAyAL9OMy%6)3TD9D{aU^Ni<!Z)tj4qq>yO3xUa{1AkK<!CxzwUs@2
zk|;AMq!%x%N3c;&32A+Na;zBE$spnm>3Ey@u34zjv?XTNN@V-@tv^8Q*O%A)_kTzf
z&q6roA&o%#nU#Ge{RA1T)7k3X9WI|T@DL2zA3+DbmqohjwN81z9-~CF7*|s(20Zme
z;A?dpA~pU{d$Fy4l%PZMkk{6%&FePEs3p3H6$Y%`t<AszZvl6Sz@_1H_&%F}NSg_f
z8d(zAiEYa!nb#7np-%I1@}RpHDTWvRzAS<xFpMDM(w&C~DW2Q>;M?z13K6Uf#(-Lz
z-*5jx10)Wv|1az}Oqb)a4<NK-2kl5q|KgWFE;JR%wn{RWxxt7j!(aQ2S34KvA*|#r
zPyomMDVkeaiotx2;50c)lggdVY3FunmOelLvHlx<e<jd-WGZdGRRDn&ZqWvYi(<Kt
z2uA`qB@xkBSHX1^>UIP~!#%A=j*8UtOr|jbqDojoy46|m1eLP>^v(a|?s<4>f)o%F
zOUhBe4v|;1)bO-2<sB*1y#txxhcDf9pfThbx`Th;my5owV+KqE<^kwGd@&&4@>`A$
z=Kn24fUT}@Zm2xm;@!Qvqgw#l5XFxxb`iYL)t&cxpW!tt<4EMm{8E7BVxC-endxoQ
zgUp<?6ufU!`a+E~84yFgpn+?emLI$JjX?Ov(+ipRe||O@W7;=9{o7SS0eNScUCK;F
zkxGT1#cvs6pY<dsEZJG{%@!r8G9-~q(E0|z&o?29m{JS?K(Ass{(3I9X<=oc!I6!q
zs^Od7;DSm<z4*Jl65!4?YM;|&0o`2WwlZ~7z%v^>Yf-J)k0fE%PhCZtRm`wPLQ?);
z$uSveuDa4x>mX)K>dv}nvZj_n-pc~pSvR;JGE}h7b+*7)uKkCIZxR@pv01*3X}@R*
zSWU$lTJd)XLj3TEoD&xQ8SihthtN%td@s(3dhd7)<Qm?}o9&M)n=DDdz@`GAz<G7v
znFb)5bXNv&X_D{usV&M}C{Nn4PTku4X3eJ&adidMV9IK#4Bije*q9sck=^fwfOZQi
zql+hccl?4>i96kAy7X@rpzbGQ8Gm%_ZGKpYGHRhA)$_)!<n_2MLeE~d2>lw`+#TFq
zp2D-*sRjUC|4g<Rh$-NVdfk>&wb|gclA%vSp8cGx$aZA-kGw1DRe>o;b6pyfeD_Ju
zvUBPYot|By2H6*ALnfsnm=$AY{N5o9M0VAt#lXE+H;n4QWzrDIZ$0H#+WUFeJwY{i
zL7Kf4jCM=O(wmiTEBUvM3Vte(clw{VN&v5SK~tuFSV!qa0Nc|I=!0b`a!tW*jb_?O
z3RZX?S_yEfOgRHM_5L5VFJ?jF{%0upcFIB62r#zlGSyk{aA`p~g_Y(mB-SRVT2qUB
z(1|uNFbFxO%D=h%{-w~=B^QglPd9t<GLIk=WxtRXFq5$l!L(r;Ow=9fC-ZDfsg^G8
zI2ZQeQK3s-@7#yMB)C>$)tBEZ0(34bsVp<1Pn&e4qCMJqj?;5?{3{O3{}9}Z?q4Mc
zEflGP?Qn#4?0)gr($#)T*(C3u0Rl5nvF@kfc!Si@?P(&oPq9SHi--=8vLGHQRTx5T
z!1^0q=9mTwYkedy#t5Vlhph^4#|ol+=qulhnbOD$ISG1hIts`n*Ka;+2VG-`wNYZL
z!di$kt$xLeN!KKeACvFhsk%*r^`f7988!1zPJGSs5gCmK?mBHDY5(n&V>~3fUPWuR
zY^%d-n%2!B^R{d&+=_bhYYwfkiHn_0^Y`olEmJ(i#iHh-i9Hch0>B+pqy?#jW#L>)
zqgkTrL>9;!E~7b#xzY}+wU>JE5@`pGI~EHv0n=cXsxR2m^9Ja=)#C*G_}rGw>F(At
zW*+AbP1#~*!(^*_c~wDDU>DJuDWxl6eagnlO2BZ{R8G2l$XR}>hxLxi2FY@4k;0Ja
zO>~}-WLy`ikXRcDi#Yn=tg7Q#>`3%+$Bn+>>d}Sfq<LG{so1G2Ot{z}3&Fl1(Voqo
z>2O#^mEQp(Oo6}n1O=qA<gEoZmE}B8U{;H(aaj{%NqW;ELx4?wgc&KdQhRT<lIJ3b
z1uS09iTY?aafdA4vg9}mpahbuxhmSY<c$?WDrT`HTF+kI!|xH6&_Q()$c7XTIY%Kp
z#2@Yeu(r*@`)7}zt$}&Id@?y|2Jouf26cG%2R9hVyKzYb(up^R%3ZiSAEGs7k$oE6
zJI=yFkJ4VhgLPJ#acl*hs3|ji*(~1Ax$^g_<(ZtIAcgy3PLZnbW4h-T+0#5D8OFL=
za_q0>ohB&-`o+b%YjRlbj(E>4I$h!jNH4#p$Mei}AFj4VI*BL8HMXqo$k*zbk@NE1
zvVIxM9wl?CFHEyKc12j9&P?)#$-lAzXJ`B&Eeq@zI|f6-gv;-eu&0Qa12DjpJv_$~
zzW6H4r#1CWRjT>>)?hd?;T&=nK%|G);QczL>rKk-oX*r2CoRN67XYf2Go7wb^<l!@
z?DCe78D_0jNw4F=GgIe$3y8HAs=$5gEPM8ot?Wwam?~?HN+1Fm2LM&RwK&dIO{X-O
zL(VjRy;R{Gv&`^>PQu~e&lca|cRkzRLiILU%Utff)PWdW0Jp{H=*XnB7AS;`iQZWM
z_AspBVsB3MPCtoeWQg#C-``XovQvm3V-y(_fj|25bv*wqGyKVoM1j7l`z}Yc7e{2p
z_CoN)i?Q!`eDNzrYy#-PV?&lXl{1~HiiH4tK7;slfewY~ypu8rW?C5+llVEEH*)P{
z3z6m6M^cs2<4k8#+z(ATCS`MR&|jY9{r1nPt9P~0SwphJXmr`s6%h=MQ+FeE^h}Vh
zD*Mg@yz5HLk>utL$<*VFLkxDX(&v_==5D8kb|a1imP}R0$XG0v0+D$;j1rleBH9Tf
zKV{W_j}%?m*_sdUl|}zj%gAMq)v(o6S_5xWU)ycQS*Z*)#;kHI%`o3F_MH<>l`%q2
z=}@9wY$i#gy*fjmmQ5*zTTX{u+$YPQ(jm<*4+WX&pY23T#w6G2^A?FXbh}c-q`;x=
z)T6r!?NMoG4F;+m*B6zz(V^!oRriM-Y;8jIroH^i*pw2r77|tZy8#K(`MB0iD*xA=
z8jka~q;gIl+QpX~q@xN`*u{k0dGUPe;8&)s{tU6()5eHt4p`A^;PF;Yl8|Okvn+k0
z7No#p@hkIQ-OD*LDS5RXnb}QNjISt(SL6s*&`5|g_1OloMp|m{zx=H4AOf2a&1L8W
z8JU#~S@w0eEJG~H9H{O&M?-1Cu<+{Zuh3a&OCxL$?_@OlLDl0Ixc^PrrL2_w7hM8Q
zv)0CDvj+@D<<4<ND53RM_j|X9ivSUwdJUhHI$QdcYMkO`_XM!W^}>ZlUwfkuPrj2H
zKS!R9`O)-poQ<SIm%3(Y$0Me5M;Q3OB5pl_*KY|jE;8Y*t6~5=_z%evl$}6apf0i(
z%@q)%wMp5PcSus5n7eSz{?{F(Rt~Iwx$w%4fu7~PD8Bs$yYfM;vYt4iRD{>(tcznh
zRpAJ!`GMll#dTyiz4KUv+2^8#ef0Q+7V!_dq0p8V;GlJenm2&3Ipu7MvKJ@9uA05Q
z{}_?z_`#(;Px&8?p%bJ|#ErpiG=69~HBp(ep4je4Hp`8J^AG;MMh-z`h@rTgCg5_x
zL3NMz^Ra(%HycrW^mz~P(G51e(N4X}#(1V5Uw)Yoz8HLa?OoymYTXnuf9Vx9ggktC
z?FXq#qKUZ3#;;)Il_cH0XZP`WL~Bs*2BsCTM-lw77g0omK}n2mFUAOnUQ;0|NSV|0
zT|0FpC4<d28h?Qi4`feiW@pBs@vjk7f{yff*w}{LKC*vs=*eXf)>d?`f%rBvevLxA
z!i}&LfD7MbU;Ml6K$(v8Vhy^=qjU(Gt8PGf<<KZZ$I~9iEjDt`#yvcmPtIL&c`CaA
zp)`>Wz+MX#C_RUcz-5=u<s+W$D`-yMbZc-|JhlGYqyk|pCY1|NL^MMw-Vd?oAPGkp
zwi#uC-LTmf_O{MFZmBnl@!FiMH)TRt?>|s3@Hmk#u@;C>g$VNiun}p~IXYVr8yABR
zUB6ZXxx7EiemM7Cq`c$9<O2;kkS3j>55Zz@FMAdIM;`&_Vvw(8zA%M9yMi?7A~>GZ
ztGzrKc#Q94@_@M`7x6wKn+39pho4)I@Auf0e24;*pajwi5`SpeU-4q@Z|_F=FSYCD
z{rwT$`A4_!?RUHqM7|syffE6}`-;e~=U@xXUgn*}^Hi;1B)yq97O?jolqOu;V^c8r
zM%|wo#2I=%Tl}n+A_ru!-;l@mUnpJ>XIfKTFW}1&4P4)y#Ps9)p7_K5MXl!MoIl_z
z42nsR1q<G>Rhw|t0w1u|=m4GZ+-3}^6O@98&f~;NanoRp7)q`&9c*@(qfu#S!)>i`
zn{b!IJ^Gr%5hId`j2IH4s>8p=$Lkspd&4BZrq{)Qr<`|7TB&Og!q_(zIeDb=35JoV
z{M>J_&uIf!@0{qbf3y-X<1A$QU6TVX?aaTS6!if?_-$)T6I|;628|KaI&Y$iBI<)D
z54FA_g=`h_9ir?lxt}=XvD8$qd~@<iwtn^S95kctSM%t6PgA058hjO!pFz<0u;fA)
zIG=_a6R6IU3EkeX{RZW`7`SglFTFRqMC^N)_V;*HXro^N;*|QL;tTxsh8vj{AuUvo
z9rkZ6GwQtf-j<Nq2(;WQWPf)j?7-u1{Jg`8U$bIza8QEOe{54?gwuDtPqX4Tb5TA8
zt*rLagMDpr{Xz{7#m+SqG9<*sB#8A~M`@z2n!TLQYLi(jYcY!!a`NDgBD$&wpS!f_
zPyZ*jlcWv6G4BkCmt+;8{|kUwzmdDhdd&f%g(rnO^-rXVMUwoml|t}~Hln%kLB|pa
zW_rbB9h%qnx8U;xHX}6tY&J={^t<0OM;kaF)$45cSHYl$$Z8=ESadG1)?8e><IEm_
zSK=MXF8hTl;C9{T{rrf)z!Pc9*A`f`GqpI)oL!~P;p10P&DCWpO5u_Me5%G!^v4b_
zE9g`YWE~hi)SzSf-Ef^{OZV%i`5fkiL($7;H!B%eLZ0Cmk_8Dh-t=QDdO*~X?Tgyv
z@^r;<U0p#Fd;71+vIX}?&CbPOlqzp;h}Sb~+ML{}_d4`rsa8}mo9X>8vc$6y%Uf(G
zYj?xwZBOl9|FZTqDgSUTPn-bnW<IFWBNhqO^x&QydS&~_)LeYQJ75b_BHT1o<+(L?
zI%f6W!yU;x8<Vp?E^i0pxZDzF$gTz5U^c&jn>!*j>GBE4Fh8%Tg78?lOJ*{M+z*{&
zmH5TUf-KI2D9pvZ{NxXwZklBN`A2&kgWx#`S*jtq3wr^Cl7kg>xzLpxzc>%W_1g(k
zdi>%vM9!Z{YjWSqU}6doUZWBM9e4KLK&lGUCTBg2SJ~}v!Zk}-gw&sDgaCR2@o8wW
zq=(O!RB$cW*2H^aFFk*tM%9q1My=te^DVMgp4lY7-g^UOGUTH-{S|b3PxQ+lt`!`w
z2H^A4WiX}xpB(~U8NoK0e@B2avIf~T0uaokN*@ILOy}ADFxc_}amIGl8^RM|l;QXl
zO;bG{RG<wEp5q$9JOWE3?ifH4cK&*$t`|hK1DGzpp1b$Fw|_DoV2!$Y{I{x)s=t?o
z3N1WU;7>&9Jj_vYVpdTU7xME1uc!`8V?H(|70I!ICPY9<6hLt6d-2*!-b6(qe@rV*
z#rrLdAHj%dI8ue;N5l#YAPvdG32?&Czcq>m#$3BiUJXCy`O+{+QEo5F@ROtgvuYwK
zpe&6K17>&<t0~U?QlyQ<4b6}Q6FzSKKXManCaNZBdjo-NjRC*p=22znT}K*|Y6fXw
z9EPe|Cc}_M1R-V1#gQq4cC;#7))!P{eP4w3Qoeo0<OLn~YV<|mk&tccBaPUaL5RzS
zCa;!B4P=u)g2e?d9|%>vN#JXMlpg9kcG!MeIJ^l&B)_z4*s-@T3C&J|6fZ>!?LMC}
z)56N^cu;^xfE`#`Q&bAq1RpWwdXA4DiW_n6CoA6*ZAZ?1;;#7ZN1nfZ=@k#<?{RP1
z)5js-Bj~`QGxfXdED-GI{WF;&h~Vi~fuIH}eZ>6jgTyMwzp;H#EBNwaZ*h}eL}rQj
zg!W!GAn;X#P*G_+zX!G<mXO7B;*RmY5~PA0mnHe4VlPsB#7+?1m3%`Sly5a$(U<iK
zITH$RVukZDqJFitH6)uHZ5A9qjqsQrwr{vftKfjCEbrZ&s&IK2QXjE063A?`pQscI
zfQx8bOIL%kIR{&xr+adcvS}~{!6^M^^2KGNPfv}e8Jy+@``}cIV2Q(R4Lm^H!9JW8
zBFj#wAJ&kr<t?yIB*th{!@8Geki$xNrjxe85yX0x5n=5?pzhAmx77y~4iTtVwt@-R
zFAs;1G1ss?w4CiPTm>pC%BH0?Lz7$hSKuwVfzPiPZ4rg{kJB~luFL{C4$6K56oxU`
zr6<7hBZ`LgU`ZQS#qXQ?4RUIviT9qVA>hc6d}1wHpZts{AVPlU87*2FigGGhS5cx&
z6}tjU`~f0SPP6@8qc)wXuVFuQFYM*EpkkJ@TdQ5P<onK(9=vQ5FmZ`;S(wH;Wk+Sv
zGQJtHo2YBvn$T<a#IdRUHcOA2yHNsei;gUQgeHZtd?`G4ZG=`v-A=NB1J)$Z<~^vh
zT({xm0P@jH-SQ#?#3w7p&MBB6o}<<s$EsFJM<sz<>g7NVPPIm^0kqkmt-)YdU<0^l
zLD`ZN^0`99XW{sQKULi}OHhwa*YPzU7y;Ju6SYu;8myJ+s)mr^5b8?{%Kh`v&L?t&
zVRKV@Dt~OCE)5yI@9#6PX=IvjF0f-ln4bzaxv#vaK9w%@$T70>RT4RAh~sT{ZsjT#
z!^bV$w9jI`2j77mamp(T)6`J{KlGQ^UX6_IwqF?AxwcmB(Ii<3I~9J6eC`=9`;rE`
z4W&u!HofJ=`=Ii^Xfjk#V;!x8jiJ`y_f;$qo-R6W5ZZ3C`C-x#F$0_!LkzJ4*@Cy2
zuyE2w|C=lKWcp`|*1&h~NY<W{6nFUxwdi;{C_PoxG^61P^t<Gzc7JH}?QFz*zgWX;
zhpq2)Rg&Tq#KRdyo6bMotP7tn4ru5&6;wV~#{M)xHzLx05nMC}HN4)CY+Ws7`bz&~
zT@Ud$%xkoh1|TPnuz#BRJZur`xta%iRD@Ui=UISeO4(LXT3{tH>*4bF&8i!>T!|Qh
zz<m=IgArJrGGcDvKBY7=g7SQ>P{&%z98t|;+k?sV`<i3wW^v77Ob&LfjL*Q%)WT?J
zH)`1yU^7AaC<E{F%XZ&tjJ68JgPWX4j4<`{hbK45#ipHIQ;22U=hfCGa}w3UBuUFW
zNJ`PK<Sqm>hI4Z-&GKaJm}r#TIoRcmoK!=#QmINWnLpFrD7Un*sW}o4l43|-kz9s%
z1Pxb^?+z!C$}f{vT!yOJ=2xcj8QIrUp&KfrmJQFi3J(9+VvxeAE=aFzO?OqxeWAWO
zQ`n=&diJ}bv%>MP9~03;YPms|Ts0FzooRG=mNIx18|SZSY7S%0dr?X|HH9pKnSh9+
ziQ*2ki3%<_%{JkB#E;li8xw2UTKy4Dl2)=@u<U7QwApyGjk2g4yhiM|#fjc8;_51o
z^ovd`PL7l`%l%f{QI>vto&(ELS*pntK<t`wkV@W%8>{7vhbW%_5Xej#kSGAPbG4}!
z_895ry@eA}><5ls{|&>}i{HRn$_^Xeq(<B|F|IER-1?l#-+-Q=1y-=Zw<m4??tCC{
z$Zc3ZV*Xx%v+y_QOr@?W$Y~oo46^JM(_XI6n5)B;FP7u$;t$AfCXTtxeMjMBn#ac&
z=ZH8-bYvnQxPLgza@kG&G2`AVzevJrzUZc_`4c$;LsuiC!>{aEJXh%7R`D6+?h|Pt
zx$?oorAE2Lr}^=IF*zUUx&On6ACBC7bpgBI#b1GYJ4v#yITCuqt!&><qTWLXVko5L
zOZ>LqeSUid2XBhL4B77?KEWG|sQ(%cwxP@{*1cBQFHCR5^xkJvyVnL6kfxG5&karv
zxV!5{!ZCI_5fSKC*e^DVG)1si_9JNW;eU>+EbXiY9p@dG(%K=@{n3kZUPho#;-b=S
z{8X%*(Dy#A%8CV5In8*D)0PPiFO9P&zeFk1b0}^wv^nqqsBN|*-n*6H-eUZCWRglW
zBM+=tJ=GKp{7K$8|E9qFiIF3dPkqu~PvaU{Klq4fQR53M5-6eNo{Ek5Sw_JC`53U$
z;wc7Az*8<;F=6;A5`!ope6PZR$E?{<cIQ@aaPBFezCJKMM1EK;JlxVHq@@S=|4O0}
z{8(0?7dly9#AIDmDIl)c$x=%TLS)T0i_=+GHU;__{qD}|6J?rvS7FZVeSVTSph=9_
zI4A!8`r-TA=Je5c2P+0Zwaft_TRf_7Lr%qX-fb7!yMa3o;RfkjU1j&KpB9N>%>`a{
zo8;Gk(lHC#{5x>YX@ht#n>hu(#+f=s?*b^#$Tm#fVUTOSEG{lSKNBo7ys<Ps0CM^B
z(0B^`Nn_td_?cNKC3d-B`?J^1hM7*KTOklaUy*ZeGGL$KGn#gjAMnu#OY|me!MrWu
zQ5PVcPunwh^iV^%#x;SBDmKMt<1K;EdE@jEaCL{LM2nxj<wKLiWfIPYtjRv)Boeq6
z_A51;kgfliC>_R|Z<<Vs_8@jS@NK^8T~%A0ip#rKhnO;r>gq(HT)fl;v2G#iK}>>A
zEKWyB>6-}q&$?FvH`apuW<aqmXS9u4OO5k@&6UW;E!$xCnZeR`Ac3Z|^6MpW1<vh^
zZbiApV(lfgaZyt)JYDnm(l=BwI$ExWb3X))c^VOsg(0w$D5du^v7*=d_d|#w;Lr9}
z<SZs&imc!ShFm11Z&WQdusaXCj+b6gt)H$8mHNbM-<>SaV5>Tl2jQMKeqCA~^>haZ
zr=-k0JSJk7s0e#B-XXi%k)x^1#0R24wR&>e{!V5B+5_3UQ4Gjd2lmMgidd`qApQe^
z6X{|vLdwKor-en+n!rkx?AX))7*m#xY{y^cYGj_cy{t;v`EiQh>dhoCC~yM}YOQoy
zX%Lg$@wX1G-A!>_9r5-?tUHr_ycFlA3k=y@)dLkM(vKcGZRZ2qmFq*BM9Wzyjdv}m
zRbhvqLO$$lEE@+~cmaG^Kll?2K=e!>dhjzKuj4Z$Tl@Fwppp)b-#&Edk|msELUBf@
zlUeSK!R<3nA;L5>WItnhNO;n`H+I7MbXm>r6nd*}C?s!s%JF$O1W-&8Fqb|dZW`Fu
zjDyZ>PV`Oy|3s$&M%ugf5>0e0_BdxWeWud)@K?5QlgT@q>OWT5U%oH)_M6aOnFW?|
z@Gg5(>2t+MZ#e9zfUl)N7J|k}5iC9PaEOy0&aq!9F*T`^_`0ZQb@k|^epPkaicWrM
z@a%}%7RR5-4z||s-SFCAWSemr!`Ul{EtxcnY|Zp4Y6pL*tdXJm{<8xVi9zp@$lS~}
zcC6YW+b%F#E8mDRO&Y@@#u&iHG9(}4EWFNQ;v2W=0#lmJB}8$OQO%uZS>yneZ<na$
z@RFSaHn5u-!=>BSS>ftXi~(n_q{~7XwXfWyzWVEVTh=q`&AMsJwrnHyYSpjO#<<NK
zZVCRQ>K}i3iFx#R>o1e0QkCC{yGLjO5&p|@C2&Wz%ZnQuq*?dCn-KQ5_B+!2UqxMe
zJd^t$XIY_}V(C!XNs28gDWTXpj*wdta$UKEP>0zLs|;OSLWiarQZbjgwnMZjbaNsV
zD&#V!8fGrJm&^P<k8*zhJ=^oVzTfY@d%n;2^Lc;X@AqSIuE}>>!<fckf9bc`{aP6W
z&iS5ra(_RuA+HH@d&Dd!jqxs#%e4!41&2`HL)ysq%3fHmWa|)yg2p}=yTcw36uDHX
z9Sy?q&C;3Hf|>QKVJ%h1m*<%OB(Ia4JJiafp3(KYJ8x|C{foS!OP)3?h4NJE&&Rei
zD)nE?pdW$0XlfI4yAi2cBHz90F`_1SJ)iyZc2>7k(%^1EvotF*f4FM?vCXAYsVG#S
z+L@Ot&yb%gx8cauMT%V%KtHv`k5{q7j~#woC7wLSX0b-oEhc7e`ibpCo4hZWEPw>I
zYrjD@hoGlHV}&~GtbFORf+t`aV(G9FYE79)jTLc}fibZ!T50v)xLCBMxpIk?F!l*j
zZG}X>w`*G?=CYgHCaN`-VTcF@4gg4=jX#93^VP(kUplT*7^xM1HcT~P`t<QU*o|cH
zeS<B%mle_oHgow(V{<Eu_(__2ZNphp=MPUJqw^G{CY+OU6x#G2s8@K~_t%GiAZ1EI
zi2e6vMwmvKc-aivkRHPI0WyfgkY9=CSbyDawPE(Ndzi*>E|)t1%n0&wr?n&uX0$g1
zH<xeN94!D_8Rr%)rbIrTu;*hdl!(eroX*i%)yK```SL7aX*_IjBc!OWZl_pd{}CT+
z_$D55oB#F)ne+1=n#$B9GKh4{_3eV86uS87yYP$Chu<6URouIaFHGPS!PW)Os2uUS
zKHf1CDH{J(V|~Qv_^lMjitfne9S1Y@UHjq*@f-A-js-zv{6`|U&Q3$jK3M+mym-WF
z<yNMIN=9O@Kv9|#UeCtx`J9Wh?HRAAO$KTi15>-q`R+ZW?^DM!x2T}WYB5QVud^C1
zHQmr<RN{_0tXu`TyrxTD%kRnSO6;1L&RHZWTjAe`>;!#k&oXN>2ZPWW2C7gmxTm7|
z(X>}_R<#-NVNZ}x=#_gfMSM=tNzJeg#mPw5-^E+`tBC>&fx)f*DNxf9%Ni4dWJ+dj
zvR3fXWI@c1mn|5z>D8q^O^i$CO4(Lw#R-kG8`Hg~zkN=On_E`UZ0@x5;CxJN@rt6)
zrec|_rc+GbF7TS6L3y+z*ZZ6%;+<4J7_FKRt*OnPWB#3!<?$(b>oZZEo#yvyC7%&(
zvEgcD7}M<z0u9ag|LtxoPi#0klH#gyRrk{YId$9}vpOdVG2&`@0^`OZcam#D@yrq5
zXT(fLeXksu*NCu-7i2MZAFM?jb*=Gjm9lAe4@?XdJoEO?l@)TzzGm4=wP<W<c#7C^
zIc(fi|7*9yJo`<h*fv9`<)W#;?l#Xi6B%_xELp~V6X9SP@9mp#dx!m7Ozw1?vnUNA
zV(s9=j-ryPFud`phaXDVzwjhH8XA1#LltiKf4g#dJW-`~L&2R(8dV*r+6}>1l^JgX
zEGyf!|Gcy@tnV%baNp-9nzk?VwaMZ7bN4=|+s~?BdEom=r~J)GRus*G^@d9penPpU
zk0FYx0$bTvh?*G^Q78Y@+din7JH)7tJd$e&TSX1+u`TT^D$~JAwKY6)0ESaD>8j1j
zSzPg+{izB2LwVw`dBKNqVXowG4B;1)QsG-qifdA_OaW7w0Zw1BW@^Qwr@xplxAS{v
zjt>_da@djdMBbG3MrWIK@L4j>;XLc;GcUDt?+?Djxnr+<YdEE|d==06(YcALDqGE6
zi~m=S(>!)eo`lXj+y=Qdg2jN;%p&N`%fOt*pI^vjL40+MdwtR0Mh{?FMslZB=3>BR
z3IQcHL<2nC7DD78TLeYrrJF1Ul-D4jBlg?wT?{CZ_N*ud@jo?Q5m?B1BHyw8bN(`e
z_R<C@N{by5WzSk)`_aKcMMvOq@zxXuhT4H&*2Uorz%zkN6+A;8^XZJzHWW^(ByY2(
z=%S3^iI~jYH4gEdKgc~No2`u=I_nGf8gftk*~pnN1blP1AhGx}DE@sB(si?yK?BaN
z`wah^xLXg^re0mXPnj04uSzGxpr{>xT;W;&3w4z<P{uO7ISi!w;84)s4h<a=dQAg!
zt`I3VyDdmqb<r&=@9{!J-jKD8axh`<XdMtP01<4r2;#4W$_YY1Maq=G1-pds4=#<0
zo<r{6F*7yH=aY-ex*pMxGOI7J2(p(yj@?+j6iO-jFe?QjZQNA}6ia5o8e$4@Km+!y
z%-;#{unJHCV+!SfrZ@zbLHwj<V6zASo_ZNV$1Qc&jlR9}MC9_ywG1lavH^kTVBmVl
z&~w@?bvwUA7C@x)5$I@xV7sY<CdVN5*|ZGMpzsHaY+2^_{QZ9e{-#3f1bQLM`)jQa
zM7=dd{8}{3VCa1c_yd)aT6_)4=K&d{%;*Muv@1b8c>;h^?OOwlpR66!Pj@IeZCp3l
zj`7FzmEEAE_uyMY^xq7E2~|P4-ekpT<N84nu!Uj8RDxnAkk@}-q;sOC=ty}a5EwCa
zPnW|y1WKBeaCYIcUxUttP1lW`fYPa=fn{F+09i2x=-fzvA2uO&Sc5>N49OX1+ZQU~
ztAYe|drv--&IkWGUptGOsY)H-)FE3nq6?>hY?qemk*L&Z%LKsF-+5L(-K`REodK=z
z?{Q$}t8+mS)Q;GCxKTNkOY#g+(h7?E5Yrvw5r~D_K=@;)9z_lN3Iy6%OM(hm&DLiI
z$mYgwP3MXlGq1GI!lXW<fYD1$ik5pIc)5l6r%rhS4mJs4bj1w-f5-PKhJg_hLV3)F
z_#9v^9jS`gol0}2c#soW5gfjT40m-xK68p^Rr|2_V<D*zc)Mi1M@M!%`nt8`bkX~w
zqalIejp6IG+y?dvaW%$5;P6p0ogpYJ1Cmf1_%H<kejLMj@xb@wEbx8T&>F!0FFFAh
zBY8S01$s=Qqtt=xtoE&X$?A$Vn(>Ch(hkI^7=m%?LyVNr+>qrO%;`v^9eHBJ>p6}2
zI0kzOqLA-gyRvIleB>Z=nRo3*YTALZSG&7<PA0~f)W>?A1YV~n3p%l<C^JkXA~;fr
z$tv$tEZ#!Z8X{;8WL7t+VLLC3o1gabJ%B5WpMd}Eq1rFzcpiw(?sv)-Q7mJr#bp;5
zDW~3Yk2{HpHO(NG7l74_EL7GBflvYz*oQYeNd=h@BAUVO#0PPAHLz?2vY75R^BMqr
z)8vC~u%-M9GYi$wb?O12)AV<rl?S(h=R`2FSkpdG_y^+Iy2T~-ZdP;reH2ns{0l=$
zqIvRCF8@lW>)HY<P@T%Q$yv51iABd53ULIrh8FYd<TR6Pom2hrl7e`z#{GEKf-UoN
zqfp?_bk@5nrq-V60fyffI?TW-vbLo$)&UKuKri@IMrI%^Fpu>*J~q2n4K|ym6z$uv
z*WdB(92aF1cf|^#YqImk!UscYS~fITrjIP|*CH-Zh(+EMQv(0ui9Gybc+{TaGB0nw
zMnlOmE9s_0dti}T<`*hjF5o<I=Iw|ds1yJ3c~_eW&_;E2j`jeTAn3||P=f_}QHQ_P
z@8G>U@LqP<2`z^F>;XU*{P=6&O!cfmvRT8dFe3y4w{0%3Yw&TPY>U20pcE&Y55l8A
zq6tm0JWI{b9-e;hdKPN)Da_E8=%Y17wAC+0MeFAVK*&@B3J&{$R@Y2tG*UhCz(1vZ
zr&<PSwT3t~3HTwrbjMJh3AbbK8MP@_Rnqf1Pq<l(f5Tcgv`!q3Kx*>Fwj{t!hTh#?
zC!YP1Y~A6)Zc<`<t<Sc&_2~ln`JJt`Dk1NVur{7_n&}5>7ujL6cz$xI^)hjNG%D9w
zHq7`XYDM*Ri)v%(nApq2zRk|}GAkX|g!-y)e*LxLQ@yq~GkbL1zvmCSRh0!ebBpp{
z1?{m6o)?(XtK-m81&T&4<tB83dFXjNH!4GsV>DRtI$A;!#g{8tDpF?qt;K$oM}<!M
z-0lt1oT5sGe|q8~C5b9l+!-}7eQ02*%xngsE5pyVrF594x9aJ&gwKD@zV4nER(PPK
z{G?lbN-uMENz3Kpu1)1QrLeBHg}Aoya8*b8;N<`jPAKmZKQrYk5)po=-q4g=`HzR2
z_a)>7wEKA|BSSb&sBT&&B`t>~z{MoVxQ-|yLlurCQBu;<j$uCz<gjwvu_Sa94q8M`
z$jOzA|2V)Bu#SXk*`>!|^&=;+Sd#F^0f9~^q0=!d*zi`4h4!JNCI9b862IO~@5c!`
zoz7bJV+4|@8WhTYXKv8X3ne8FuU;4$v=IptNwN_vUWiKBgp3(!Txz+rW6hnqH9tQa
ZQYUMy9@XE;fwu`snEh^LTy&5W`9Ix2W(xoS

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-mu.png b/docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-mu.png
new file mode 100644
index 0000000000000000000000000000000000000000..bffabef603f24528bc116887176fc2e9a1df26cc
GIT binary patch
literal 48757
zcmeFaWmuGL^e-w1QW8q3q=9rODF}>Wf`lO5A|=w@pdbi>3estT(n@!OqJ+{dgM_3=
z_g)Wg`SSSxaK7ww&ff0_u4}G&X70IT<!}Ag%>xCw%cR6di1+N-Lwfbf1;ssk@Ur&o
z+3Q2N58ioe{(cSKIBY5<rC=s?S<3Q`rHzu6zJcLoLkmM2Qv=1z=lARp2?|oyHlbIh
zk|?${K7OJ}>5hP}-D^ow7Sd1^foC~Ip8fqo7U5TgdXA0XTAFE!{Lm!AU%dHD{-T20
zsfie>n(}-vX0}E0UWpvuFk+UTHfA=7oRxbD<AF(&QDXd2rbTZD3}@Wt<7VSsX9|sH
z@y82O)WV-o;m|~9aMPZ~+WI~67w3HJ<CjRg@0sd#22Sla*Kmd8NgQ)E2c@c9b-Ua9
zw-VCxFELOvTQl9v9(_&X{|M(tvhqg<9GA5xOp+5Cj7vAi^G{)FcwGvH3_7cAB|L&m
zjlOYVTmA8sm2Ue-P(1onyK-s&`l+lA-9*{jlPTjP&u_-q_b0x!n$_P*Ft0T6oj0#e
zZjMTj@R`-uTIUpG85_D@Ba;%6aA|`OYi91ANB^}qJm$#($$gLe2BI^Q4$NNAO6w&^
zDNA^no*Qt~W6Vcro(s$E-z+@Vpj6qFni5(YTj@}fFR1bJrx;$vgCfo7xq6#*bEWv7
zV<QuS(H~u%FKn!pIA2+Ew@ElN{oQSR-Y1-!VB=N9(|go4_xwGdT{^ckhTVGGCrz>U
z@a-cbS3P+`UXW?>4r`kyjW^5erxk8vlA)Pg^HxmUzK|Vg<CUqtFZKGvAo(8d!imY%
z>$Od^ZV7I_LaE-KWu)0e_-v7Uex<~+YI_gdV#Dkkr0FtwbB}D_X{vOp!vVU_jA-e`
zD$f2C&53HPEUS9D_5Qu?z14!I9FALl5_f6d9oYy@qiHpqOZC+<pp^9B62_H0d6@ZN
z_q}GF(^r?CSf9o@<{WTVb}5wp9p$arwJzy97j@1GcWjnC<X23Jx!8Xur#Mk>o1f#F
zmQrUL7g^r@;~&}A@wH1g>U7^~ediLso`*dgSESmss=GS5ae9e!<Mzw1p5IjG@@?uh
zJ{7;aRy}n`&DK%*i21%@njfE|Y9G`x%v;ZWQzgIT5&j`+EOSBothyx+*`>4y{I%Pw
zDu?BQE~G8oTky-_DrI$N+pH^XbZ>BPbZ>I6)W&@x_Azd_Y>3B-$yT>+>Gu2)jX_r$
z=R^6%-^0z)q>moIL(NEIQzbNXUH|k?*8PMA=dX8pho?`j(<Jc5i!+cJFpVa!YnAb+
z;5=A=i@PAZrL278yVX;61{F(Lk+O4?4^K$ldiYf6#&Ut`{Ig?W@*|3-6iQ-BlyBa)
zog??lPI5>z$6eoGXn$kzPI2Ok!GJridaJ0#BxxhHqs*7kn@2K*IKrbR#{FL(y8r%K
zu!)jikGTdN{_McQqlK8$#pINlJeOnF_g9oj&P}cJV1sk^gwuO@*B`ey)Z9QYfG;xd
z)AO#klHth}H<GVy%Tr6`Yp0z&Hru}U)!3$uaeL(oi^Y`>&f0ZIV762D7zcUcBqq2i
z@6{ri1zRCvsCLyzR(1~uye8a(<7c`D4_@KG-x2uRvj;a3_pf*GvJ&_H^?I)l@{oft
z`1zham_1i7oV($KGu2P@R7*KZ<VWvSV_oLQ_#u06x*y=+EpA+V$QJBxOK~IiuoXpZ
zrEj}FfnY=Qak(4&s4i5DC+^2R7IU2W9tP9xc!lqA7T!CI$~(uO%i&5FmA~|V&uDX)
z+#EO)=Q8O&xaqE3>gGIK7M|-FbN9g>99%p?3TBMg-!Cxth!;#+3Vbg9^VVO_m{}9O
zMuL9%?4J)>T5yTgNaGkQ5BxgN&O<gmXYzjyzw@p@ju)<qRPtr3UB3+viPZ-7qTeOu
zPV{PT`F?;NHC-><@Z~?J@yo<8QY378f_5%=mr<*LC5Zp>$v?mQ-xB_{QUB8^wDh!m
zCdzn*r{lS~ct5~cs&Zlc^YMp=PCdVszdAqWH}WweA}#xIQbo9?MHGqKOkbh?;zYZt
z!}&IR?|5<7`$N@!wBf^_G%ZW1RnpY1ziT@0g3Ha#tslsp_a1*{K3Zq1C$JzlS8zX_
zP};557|XK5Dj8bpJ0Ghp7i)~-XU)mYebq^HAcAmVqMdU;gJ{mJ{u1Z;subl^=??WP
z@`)?+qOMC*Sw@Y~3!+D{2fYVMUHuw`2TN{g2b7DAY|C+ezv4@y<*j`5N+Vh<Z(PRU
zq>5%}u%P7Q^<xlNUuHq6f9(<?rwR->_-PXbx5H3En0Ow9nfh+qew55&?D43wQifBM
zckA9BZ(<@gJ!L{gezXd-A5f@B2~JfZLd~1wM55RKVGC+`w`8g0+w<jGC`p=5sc0c%
zUS3{!O<zX`P05JNl@RUsV?wpdhVRCe$<8LUZkW31%pd$-K59ETAML?2FaCZ^pnPgR
zI^xQ<V1i~s=Ynn2>^X_#Rl~wYnU~wq&FQg82_N0af1YXNO^)ofUw=F+!=gi-#qsPl
zDf&QCh@R@iE9SPE6dcAIZB9tIA?&g+en6=ER%l3M1ef|jm2?ed5YT8`sr&ms;Fp<v
zeonDHHuyBptJ$n)wO3(bwRy72m!vvc&`c<&c$$=_CEK`}&*{h45H?y~9eTwLU2KS)
z8X=+tt+K+6S651e>gk@BHm458MIX-xq32f1s?a_u`?$kt<V)hYd(ti;*fGIvSKX_Z
zIdYI=f6pb5?Y8>!Ws&-ZtT>Ysm(Ht>M{ZL(y|ij?Z~96`_atAvZN7o;+HX425O#F5
z^JyMt<DQ6QZ7he)H^DFN=Cd7Mk-aO1(wfX$Gfxw+%evd=o_1bY&Q;CSnZ#$3Ox==l
z6c7KVLbaVvDVe!-kFAg@gVt(xptMtS%rZ_f#L+`KEF=E{N-HplM?u7o93#Dp@(2{;
zS3t{-AD%=D@{#lK<meY>Li8~N!*HS@Z{yKg_Wy|=O3Zyh+Clm2S}{GinWak_D?{X-
zhp=X`VD1idq^XDANZv=zNKYm<mBnY#SMU-`!-azzMh%*smfZ2HA8w(*O;t%J9<F~x
z-kL0L>F*?T(%&qzRA}S7yeF}yd2V~1z_1!F-a#5G68)+Vl%l`d7}0}~LgpGuRGBxo
z!+CWXX9~ySgNts9<KP~qP^f6YgxEj%XBWMvh-#k8^UKaHU8cmr?WTs|9Bebkelb6{
zuIaVCzW&T|u<YtW(P0bsYq%PZ@eYD9l0I?0>clUWvZ9Wd-gWA&euTC|cb8F$$Hqi^
zYPGTCdUr}fJrN<rmCF^}KHtAdzx-wF9aH9gjhIc(@bt8(>~1l-0`{0QPV2$bf9Ctg
zIv-yJqkAn$I~gTI-wE0DUWIZ#Ir!^YAYVVtp6C_!A!kzH*PTLkoEb6Tb~Y@EyM+#O
z5}zvTbd*Ef$zK+^Gn1FcVJx|f3(~tM{Qf1liIc4vt*ABHf&6~?5OxJM;V3FNlUqIZ
z(Wo^B!&st)2iZ{*4h6pwb*)(LU-E{0{R?s`tg6RQ{_B<^Y-FfurzvWUqA=EfhqBwH
z{&y(<r)Wxg)RcXgn7HLY$&UukOrQKg53xJB?JD|+Nz2VWJ0WP+;d91z)JIj<U64z+
z>?}l{Tn`WPpIrFfn%q#@a(1`!C7!}-NKFyAG(}2XJDFL|$WuJ^j?b{+sn)YMA_WlQ
zjt0nWFIjHWEOi?vU}YJaGPMd%_gPhw4n>r27^>x%T>s#)xxP%kK3nc7G8$=cK2{i&
zGexU|7sTx@#RhOzwX5nfK!nk5VP<Y_uD`bMeQ2Roo}6i=f}rY6a{LjYLAQB_<si;6
z2uwKTYEuh(V#0Z}k609aPx?_s!s|oATf%qe-G`IY(?$%am9T?J5;`3uJm_Y-vfAva
zV`BKV!h7fi)22z6L$A55>gC4RGyHvygYHMm>b`HS&yo+WHaS*3@>qD$@}hoC>M^%w
z9#^bQ1ACz+#qieLM7mRZM}hUQ&AQ~;_=}-R0;c_Q#OY)4t~`iS7$T5dId@3Z&ZMs&
z6)c#Df&mV?@?D0v^eN|skr=1#^+Bd4k>Q7n85wIoBg*Md%ExQ5*@<th3=c({Wf1u>
zIH+y4R81<YYEvi4#YJn*^cOpdtWRajScsinXpxdSE1^?M9jMK3^ocM+%jRk%pZ@u0
zq^OA6+vuR$_Q6D>3GJ};pD}|QC)3p49&-$L!oR^Kl<R1wM_;k|`bBU<kY*S;J55eU
zhIWkZr&1oB5|KWq;UHi8<TwY7t8*OX`j5ny2t!ZGHt^mro-GqAS?tjCHT>O=DOB|M
z->Kt#_OtcLWyZovoc8jSTHaAda%=k3m0Oiq_FyyW%#;8nHy3^S*QGB^r@Gz_RWof_
zS<HhQl`0AQjNZ3Puy3ccMM1W!M3mn?U~ufSzQ-NJC@#7%-J74KOYc+a<nrywW#T~V
zzSO0<_Z(9{@}uU4>uG5W=FA$S1%Eru1AO~bqN1iC<`9`JnLp2Ut2XnLLEW9KS1V7e
zu&mT|LO+#C;`qoK;u8ku<Cnh~i@62m9~?~v75bJj-s3WvKA&2G^1abWT=z+U?1%uH
z+qe|dM91gWBv}%WXkP<|0q3!*jgP$L^x<6UifS{`M>&FvsWZnl%G{i13PciJRf9QW
zYfvc^GY+i%$jR>$shK>n=3QC(EPKnE<Hd`!(qFw|LGsb8zV2&%mZP7d3bk(@NE2%f
z<Sjdd4W%kRU7K3#&~M+FG1bGjp5sVjM6Xc3(vW|jNQK#uq$b&ZX{tMWYM7zx5bm_Q
zjw>XuhbkZ3e1V5;&e`~eWEAVUrL%=|^BS>}Ij)iY<ALS0_06e6>|Z>2YX;q8On(+{
zh;?&qmDH76y6FtQy=tGa6Us|xcWVuPmR|2rkRcaeP>4KjD19_(FPTUpACvKb$NF@Y
zO-WH|sj#KK-}5M0;r<+j3XZb7sU{I*#+HR6;Z`45BJ2uEouu4~u$#++x+NSRR!z)q
z^LWb&b-yeWeJPBOlhBEKvGo%q53(GclkdM)-s{vfzSx61=XgzN4R<dA*|l=Yi{gj)
z^e|m_vKs?BB7<-xG8@+=eiSR2T99IOCa#CQaxxqAA)+KUufN$c<j16=7R+CZpEjd)
zl#27+D!V7AzDGAF%7h#S6Os~JD^jp_l{JovhF2$6OsQmQYbv{$Y|fKBAQ_vQ+`z2|
zh9a%t>O3kYN`i1S<>nFN#%O9eks%+7K(*EHZA$W_!*xL)d1-Wpl?dmR9P1vl4dgpX
z42G+sVwVZrx{4IVeF^SM-BvxO$-ej`G*nGiYu}Uo_L85?%F;SZnQ$^P0V3}w3H@Iu
z8^SnA*3(>-RrRF^_;pvRXsf_R^;LF!X`gF&i#@OPy`xSmU@u)At00X$2`$h5{yA=m
z4mSg${Y(;PX<Gn`Q2%MyTo<4ypiS*RXn>t=uT|REQr{5IaEe*A!^3)J)DT3}d>9jq
zE3v0Bbxr6*@DSxx6?=zBb0R{j?HLdsOjQxZ&5i#2s3a#<ncJSBH$IjFUST<K_FjQ1
zNzwi3I)dl>51)Sd?V)AqqKaCni~Wb9OS<p3<5Yq~sIyWMT?XBkeP`Xq#C*?R-MBdW
zkiUtLX?vw!?5yVY)+R}B(L90XTv<Q2UM<d05kFfWxF;R|Q@&By)SjPnb;-5>os7_y
z3VD0Q3Y|MBHh#CyVNYn+1UB{Fh``%#_7oP@a`VOS&2^4VcVY+Qu(L8A#?oR5r`E8s
zNrN$uIlnmRZi$E176zT1ThC%Q_f`{DbNpFaaN`0_YMa4}2#bQD`#->5W&Nl=Mw8Se
zHd~UlJmBis<1j~g9=pv|V~MGKbZL70D|TD?y9G|_3!!w@-BGnb2!gS^JHYuBvDz!M
z+gr2e1Lu}<cbr)H=8~^HLnGq5eP!P^9O0$bT3*VpBD=^B`7{sLh~;OtG(HqMV}gqV
zTg&|5?N=(x!(2Q%wa?q-J=a=!s55U8zPJ=bJQv@&vCw0dIfB@+VOcKiqMR)EPHji-
zPccVP(+l*%?YLxgdSHUq@<7RCMlmt>X8HE=wvDQR{aC<jr*gBoFb(;$a~GbK1{=k@
z{rpgsB+}JOqj~4Lnh<(oG>Fgp^o?3o>%dIGaPWCf>%=7zo&shn@&=u*0`DC=z2v{Q
z($b^|_K~D`>TNxvuk-)tSDGSE9oKSSZ1<&+Sa?S6o82VJnX|m$RV5eqX$oQ<-}Gkj
zWi7<)=sWjEMd-M&_c@Qn`L-vl_l*(H8~vJi?ku|z5T)D$-LV3uibHc%a}Jf6ZX=w9
zm7Nf7G`56g8h#89i*^6@{>f0qUUIIq7;vyoKPm}GKCVX5>bjkZb;3UIwXILhtP5s{
z8Jt2PV=Xsu<Ik%ZiGu7(-mhnLO|x3)E3~cNu{tvipE}Eya;>`b{lcc!+A}=_&;F>`
zdz7rD>}GAij&474pEA&&EnOyaUt1u0boS?!nef`$K-#fjIsp>}k=1XZGCf(l&yK?6
z<c_a&-rih}aK1I>S6kV|TfXu0N28?&|Jm)0dHwbj<)OS@i;wQ<uD1Z*sIGZHSuNCi
zCwu1PN3`8K09kU>n}-Gp%4SDS-MlcvexHEJ&B|X@EnO@(C&x!bZ6niTF%?Ph?S(Z_
z*`q+17iU((wXrZ&9re*C+G1Qf+V)?rvW}DiXo&<gN01Qi(j!KOyaAU<ip;jts0i^L
zGvuF`@ZqmBb01M*!TaZ4nuOVr_U@=3J^`m1rtr=Oowhy)&7;ZAXhtp72Mqi{>2L*9
z+NzGEtv5Rsqflupf|Q>BaFAoSvtSN{R8##^kpwDjJ$Mk(RyOtX=(P1Zq^*JKoyI8B
zk3g$LEHbZA6Xrs&gj17i*LJ7oR{(Y5e0GZnC5?|Sf@p9VX(yo8NCa3c%TumrsCecb
z8*F54+9W?}jYnZDpAeQCs0q^mmhkSsL)j%Q{~gMInzBoUG5=}G@6A`D;4?K}4P$Tb
zy#yU+?#{%Oy?SL>x_519nodKrbh`82Ri}#(1|o%qgEQVOFEE!AI(152aBd%i<Y4rQ
z(Idvm?QP?7kInPo-eO+dJ3Mo7ZW+7ctX=G2eqS*9#U!{>)8OC@UEZ6Xit8~iHHg@@
zO2`iJRgC3ov<TkGH*SufjBA{gdv#(fgm*Ek?Lh@pBSgES>ylPT^4=8UVae|(*%H?<
zbtR1+{NZe#;YPDuH<qjR;>;*DQS|XgjZ4Lr8_nZJD={y>%<?33N(Tzbxq3KqH2UJI
zyzzS_hX+XYz=w{yJ}S@m)bgCF8FsmHr><1Zul8lAy{+07VTrAG8=vz$(7CC5GoiCY
zZY@(H^ZQs<0f9tj?U6>|+4&x=GeMFvF{T>1x0SEVS%%ymVcvE&ROBA&$?F|x?xfo8
z_)yN9`YhQ?nY8_+x$V}*>eWTA0c_Tc-`I5i;F1*Y8!iX>DWq=MTL49zESJTJV8<L5
zB5$V;3qOXcWo~M}yLHuHGDp^wq(-+cuj^)@7OR_1Ym=eGniJ<%mVWimz`{<5L;Q@^
zr3zo(JFL#O)o(gf<@+o>o%=Z8PZm57T)q*dT@URUJkc&)QIw!6b^C@)k;8PI=3M6{
zc|Vn>V*`Zxq=kzFr(V~o=Um<G!I&|L*N`Y?i2f4lkpdk&=@Ahl+3k_MFj3$t4Xe&P
z++})UG$DABYwbX@+ga`LJ`LWg!N>LxdNgwg2yPEgaqZGt7~Bv%Zy3=SA9}PNz~tNO
zy46u(CSC(X5ge3F`smSek!oj9cJ^*$7LV{L0Qt%%W2@1l-3C}*na<c46}nTfDFFa+
zO~elsNfEYPf+s>vP3O^DngoGNloi==R51BE2%eb7#`&VR^uGi8?|_iX(|;NS>Z$Ob
z2I2m1G;xW!pN8o>E_&1II34{2cmTkl#~#uNX9QmE4nkQfW?o2PB3@TbKED`Pg@AX!
zZD3mJwJhhM&Dtp+%qMJ@XdWn?b)O|kPVjKa4}OczcJRUJa2UP;4|gqA<dka3#0-19
z%fuT#$C*A0hCHjGDvN1LbneE13$We!B&dc{Hc4)8SV1ly-tJN|-GWMGnSVIqsOl-N
z#8G6%t_>`g(Qa*wNk+8)1h$Wjfzb$gu&OmFpmXJ2Lz7-Odqw+<_U?9A%!V*UVnfU6
zp4_3)7)vq)SyC&qGwn{V!x<c&_OqVo%+!TyiUfs*hDL)$v-fUK<lP3YkmGTsdbZI>
zU9k(mFvRJLGyP%@>Ce3Tb~~l2PR=MvHb$kZs8vpkU58rW({6{QPR+LuU?T|~XZxB1
zmwm9<_Q+EPRR{|816GvZo4&o6>B&t`NuqXgx97xk-~&_1+p!{mn$sqIVXB*--mc}`
zP^3XH5d>@OriH2bI5gy6GrW|~n|SX!;rFk>N8ul2B2guvx5CUwr#bcN6)KfPr<QVN
zA*2AKrr5(e9;+jKx86U_swNZjL9pb$R*~Rc<1jydiojk;oI2|Eje?y%+z|O3*CZZV
z?9?T<EZ=m<h}+s+_ftD01|SN+e?R;X>Q%_na3~SE<i-!H_ecE>v8G<A!`*c-Eq5UH
zpnh+n8lXBikwPk4n@^qrJRjA>2FWK@{RU&7q9{Fn++*?q)-!9Yb=_h88vg^6Gw&)#
zKJpZE)nsN|13cC_<8eorJDqca(@1n`=aDl~{g1fIgk>Aqv%V+A2pQL(?&E~WSJtEC
zd|#Uuu9P2{c?Owbj_^GKa*LFB@*@JT)5=|=iJ!$aA#Ag_Tou84UW0Mc*{{8nC$4wC
z!s2!SVr=^sA-Fky{L1J$^+eNSu9mjCRbG-IoLe(o(*tTNRde-BOi-Vqhd8`9ePy^y
zdzG_H$WHH$SY60k+kW>e;sjyAx^kS{JtET|%!j`{SA6o`^wkNvOol@9lGr<UF4z>k
zy&GihmOXN6DY<*?#sf#<P-!+HBRU(+Xj=R<#YaUv)U;VQnb&8EEoUZDRwPMlTcJX+
z15Z_xOU7!7n{}l9$WGYgLaJ=F#ZhQf)fclC(I#;5$kxDv&o3_f^*i)(-^uGXDm*vT
zd}b^*%VLgW=CCC#0`+ydFZYv_mM`^~Rp*$t5`X*+E_m!KaoWCyZ`Sn;og*gLuXC(%
zc8V)l+VVqK5q0Z_=>UWAO#z&#Z85w#>**Om;PvQU1Fud=!AEo9s^HHiP+(+vRs&S*
zm77Tm_&;m=o#>{D$j@9UckG@#2blRQnTXM{azptSg)UmLz}y;rZ(b6oU*Uzr5;y0}
zJ#@vPa!jzh&Hj*+N7oZF$18WF+(caw96hjLWxnXvhmgL(d+nkT4ra$Gn6Nb4Gd%H%
z8~yCtjvCgI9D<o%%IqIRbF!!NdPnQ(UA9)oY>g8n++FgY`{Sprw70WMu$+3uzIC_M
z0*7;}B~TZc;NYQQCfR_H?_>O16&ADeN9rf|UvVpJkmLcmtMJq1!JeKX9r*N<Sf+CR
zav+-F1$a6Jcna7r@I6~_3GerhnI1EllIQA(Ja563wJ|HlPIKm)l{-R8%v7zAo^wpz
z7&M(+wi@m%@U<Eyo$RRs$WcjT8!DkUB87hs8rSfvZ>TjI6<-jUYa>k9rLBL`6gk_k
zm6UQz7-pp^&6&H`0zkd!w2q2>k7!pMQX>s{hLkO1LD7zSCahU{uJ`iUjkl0q61E-v
zR3!mSO?&+TRQWAY9Z$=<ojIz{Jini~HucaK-L0Pyq?d&3C)yOw&a`*EH4t`s$bgE_
z-vFQ{c%|+>DIHDS!^2AcqfovhiwiQ%b(rqqGyU?iYFN<`fL}h3jg_jo-$@^kfH6^L
zH-1!;d@lLfz^RX|c`!}$7W=}0^;#rFJk)`zwsg1GI;Grh6rz~TU~=jzZYMruw5q@v
z0<|U2A8;|_NhzKw;ygS5iGQdyVau+|qIkLr)LyE@{0BHga?6srOGke;dv47mh%*n@
zZ>j>JU}x_c&eIL9TtXbn%Ha0qpw_Lq2R@RkALM(2-pMCO4DD2EHpegb_FJ&*GK84K
z!+3q?yY$o_UVM|?EFq?4_xYm!`gFdfaM8JY`+S$d0S|qV;Z-iVt@>VimxD-*f>Thg
zy~F6aF?ZA>jDH%4UPCFV8N`9Kykz{wP5S}jwi)lux9kG2zxw@?;~!1GoI*Lc6X4*E
zoHd`jo%+_Gezfe!*&9PBQAs)Wm8L2s>o$srg-KKdLmn2=Mz(uhUKnMhI9Qc(eb=lZ
zIF?38uF{HPyXS&~tBj<5#Hp81SS%g~3nX{!SEqOTS3JTS7a@h?NIQapk11Y*?^1bn
zA{V7e%nH;H7yR#8kV(39>YsBnBrE}1hCaxC;c;4S4Jw2?dWmyU>{d(SxuO1956AKL
zRMpM9Rz|pZ(HeS{ckzdSMUvjIT+b-(;&W8${<jSzN`t^j$@*hQM@JSoL!U;W-ZZSa
z*<-{KL~FuM6waqQ!p2^?^z{>e^L|<_t8=I49H_=YvS}<z=IZ$LD!hUnc8HXQL<n7C
z=tng<xlA}0`{c@{Gp}{aJ@kKm`!IBmX+!UY&j~W*@UCEhxkq&3sz1Coaq#vfp}qfr
zl2z!UaDww_)Q{f5tvvR%=ahJa0^t=GX|<oa&2a?q05{}Xw7{FURu^|RpA^E)_8;sA
znUPuQ>J5}Q>qAH;+xoeS3229S9Nb9EC@rj>zvPyXMpsG+H3DXJIHFrhtX8|_2P1^{
z&^s*X>?EQ~A(SFOH<6;3=kWdhEmS_E!K;CNLHi*dVVk@j@Gprk(_3HI#2v4T$}Aj>
zhDs8--7sPC4xsO?2;(b~%OFD7!i-z(M@|m)9@%>rTAdF89XD=P52<&jFt?O@Zr>`C
zC!-h1BF2qI2#?-eL<PQlGTj~MlJs5B2BMR)?~y^_lQ6&^FW@UsmtgN*L)BicC_r7J
z@+A?vkir7<<1*CLMd<<&AT2DB0!LAWly^vh=hdA_RH2^v9E^n*dh*s^^Z9FHm~c?i
zcTdgTQOa`ysd4Ci@j(?*-Xmq&@OSB%s5Rn&B{_I4`Y~$4>QLZmP|RCI*Ip7~th-k!
zPyaQazjlmS1~yW?A+HQodpQV@ku;y_c~o`l)>YWZ{|;rh;QhaSD2d;R)lyY6$(kj%
z-3sRF=(3+)MhlxjI!2pB?J}eDEhGlD)y?m-`FQDgf^E~8XP(<z2&0(DkKX2rT7lKj
zrTjO$g#|QnjX0-ce!+8dQnxx(ULuM_AMjl|25A`?88yIzpT$M!dYma)@1Omh3#Wj#
zkqk&PqbH$Jgv<12Zl?~Xdfu<v`^OJHj>oi8W!-#xyWcu<9~ph5&ZP?|<zQAgrjV;i
z*qNbC7c7GH{{gN$EH=w&2*`RITu2Y+(7TU3oLr9fkR=J~e+Xe2@}<=w{20N@d*%*g
z8Y~}acAX#m^cTw48aSN6m+i-&`Q4XypHR|Pi~@eQziwDG_UjLWx`%IW*!G|?&)1<C
zWy_;y#}n^WNp1~Gc9fVQIa9&L&yQvvI;@Bt`}aP+F2S^CR5bXl0RxZ?p$AO%4@3;=
zeCR|x4OHoUK7_LRX*>4khRB*tTH=bkE?sMQcKOlbWVW^T?aeF1rDT<LZgraB0FdV_
zJIfYcp5wSRXtot!Y4Rk&jl4<b<z~;on6iagXe+i^$-GIeKYK+&U3;+0CA-wvEz@~S
zTxp|J#3MYKyK{CyK+LxGOyhFTps4eM@^Y0@uUtr5eEk?**ghplmM?tr)D%U?lOFVG
z2-CrBW0*<TGJ!N5ai8OZ<(>%9>ChRQTjQOXgZIhamR|FQ#v`cv57Xy;I2j`#$9L?F
ze}z$uxylm<0YzPJ;0?72dl4syOI&z)xXf53lBI_8CFGGvZag}rc5jhmYoM1)Y2%Uc
zAZuzuqtSZ5BRK2<;pr0d`(z9~O&Y1u=PCw8U#X|L;BdxU0CVq^@n-p}%6kN~9kV&D
za{9pA;r>nRNWn~je_OHQd_O%d5@;$iJc!WCSeJC2ldhoVR5{U3UYW<6RZFA!P3TjM
zFfa7M6m7<($zJ-Jo<AUcvLgep$Y~DyxM%rO#zJbWJC>XD_@#&M8AQR6wsdr;$ngi8
zGJ89Xd@9=RWs<Chiprftwiy1WKuHpmt;aT7AZuv?U#_x!Lhx)gz(6IITg}BQmT?Fj
z7HF|;e3Wv^q~1<p(?zPO$t(GCTeCU054God*^h*&-baYEotfr&-AP9_TaEDX73qv$
zzi<-M&^^Z?+pfSosw8JdsPjrm(10xFt$f%O$p~q-j{Kk{n?s$s`DJBKga*4~8VkDD
zTK#7WjQE2$P4gte*2~x(zA9$i9A;w8e|09s49BdV!I3Md?%==WbO&Cw$LsLC)+;Ke
zhL&Sy1$l6GE)!zHsP9$m!=G>=(Rck^i!+`ipTLOK>$T=<2|Aqivqg&Be3q&M75EaK
zIMG6ZpC-cBjbIxH5znXBqB#AgyaF3z7tga;LX=CxBwk6J!B2B~)s90zPr>N+qw;yy
zCzKZ>q*y7&^w(F2u6(^momoPMa+!FdRIewg)yh{JEvv+r`?7VcN?jI-fov>u==7~C
zhk)ub{`tj>ecs~DoV&Xed8e{t&#ec#YR^)7$JCSvShyA|b;4x_))psK-!<+AV=!JA
zA^ZuA3YUe?mq{k(Qc_bh%S4@Q^t{o$ORtJSya|;G4InZ{LhgK&2!M%9tn#_){oWKh
z&Ld$vJUrF}w|6(?N7Hi65GKva+;BBMtro@F@~rM;s~8h<DmSiv1&Y}I<X9VNr`3@N
zTVu6sBiXQ~8>>dWJv~+L3_sdPY7RHQDRW;}**c+o+j<ULt9)BMdne*ZoWqA1xL_Z!
zR`ukXk(_5*eSdj}!$eq!ZSQv+j5~Xge$oQ?-9x}7lr7^G-LZ5>RX0{Y8B0)NJuf9m
za0Ai1h8GDu!P>U0;M&tMltO)zMHEQH@*m#e<Eqp8tLg}TT!0Y-WmFgL!45~!0&aH+
znrcobT-mcT5$03gMbueX>ch3IKwk=^ZKQl(f|1TQxWlI2*cL&$*zN%13lw|Jes7G|
z-~6MD6vVEOuSs|r$UT=H9y(yu93L0z41~oX@Ak;q1J?!pD_~pbfi~64{MnAN4i26Y
zD6Flka(RdZVM1_gr7+*bemo<0XG3U-<y|eAlM?u@eFfI~peuLABj+nOm-5R=t@r=A
zy7$j9{lsc>G*zCf5uQYn8$;x_Go4ztvLcgd+2`r6wO;V5Si^nmv<Uofk4;k20;8TB
zQ)uOlq&*6%ka!Sj&YTpS|NO{9V5)3=W-{HfhI*1FYpKt+8G-f$ZoZzZwVQ7g=G041
zzKM_Q=dFv07WovHz79W<l-Lh^Ut@;tuW^7asq`&|qUAIY7v$8ucsg(Bc_0smkEni?
z&kiMPhltmNf*rgtdH53!B>M52?~gzFBwn@xP!kDQw*6yqjyk}m_BaqMu{KUYqJ;$V
zhxm=Ib=kGb&imP|+5_;UJ<u7vJ7mSdg#+ot?SM3k0ilwIKDd}tzO|THxT>n-%;Gkw
zIoM<Jc4^<?{_#tKD2K_cpyXXdlVt(aM5F^C+M(;iZ=CnR*OQ0TOs_u(N3Wg{*S-`t
z71~{)mo=4f(?M=W9)JBE?+bXKmqnmtl==rxqU;%_6tDv|Ugt4X^iDC3MClQcQ)o<G
zffq54AH;ak!$m+cP*ih^6+I6ZqBeVq<u^*n_gQ4ZZ=}z<PUot&_jK@~$1q1`EIRS}
zMb%SwrIe1=T(eG<baPTvt^$=;$SVJzm!H9a4?%kZEe(%WS9&~>XF7hmi%(JoTPPFE
z08U>7z#J(vGwovKgQ587l>||~HhVQjTfTC!`*VJ7uJ2_gcL7ifPeLf0=pzjpk%DF_
zE0B!SW&kvVbZWlqbznxXbC~diD~UdE&Z}Jq*9ZO~5`r$CX&0L*I8JX}b0FwA*TLUR
z;oy}#FXg^T?)<*fjw9_!F&5F944G9$fM}8Le>C~mT>ycOr{s4BY0QsbW0a_}o^K*W
zgwL0mw69bzmv3)Yl^jO@j2(V-k{7j-g4^U*+h%cEM(1x&?D&t{pg$OkQKHUr0N@S!
zI>fZ0fRp?Zm5pMUt-T#n#2+}GxDTjj^m+qKTvfNR*##|nS8gN<elF-GT?c|A5ojmZ
z9Wzr?_P-gm(elh&eV(IH##LWl$%Ib!ff27p`2P@6A42+BZebslTSg(8lvs!lrHc6V
z)g!t}vDi5EUh{nIL%_ET7z!-h>z%}*r)}S^%+mi*o>kya$KA;O1FAgt?~zN0TuYEB
zJtOv7^$(vxhMR5FNSv_NmRtp0lq!DNsxwa^g55tiWU;*<zkQ?dQtk~^@1#OstA?lS
zLrA<V%{REcwH(2v`DY#<efcY(l`18-<LxGh5OEOjuYd(c0otLM2XZvuB$j#ZhVOME
z?^~G_8sN^<vh>b5Aq9g>NK34~zlsR&lgJtl7P}9jZ2G}VJ%5gSL_x19_6+Hvu*W;b
z1KLJ`&@5NH_u}Vf2>NO?mq>YU_=Nnp1NjlsH01Qvo1mzccF@hbBv4Iu`i>|_yIAiX
zI<$ZuLu@Nh!Bb9+-{yMA0Mhv2B_{F@3y5LCyTAYN!Lr44OXcTpQu3{l(hh+4S>Mhg
zu>ul91NB@Bc)YAt-#&!Wof0^O>@bEJ?@}=RVL6G_ZyY&Vt0%V?k|liyH)kBfEY$)@
zkAfvtd4;c`saY+#Jum48=+q~cyJ2<hj60JvvGQJV#kgS=h+Ba<)InkkHOfNPgBkhH
zDpdfNq9dhwVzqPX(YR3Ex;i>G36&OYSz_}}b>>xEmBBLiY9Q#wxW20Zxo0@*oYgeT
zmW(QQ+)zu*z)wSt4!L>2(4KvZ*DMWh8CrzWh8fVt@bP^l@toF{NZ1#{h7Q7Rv$VWW
zVYOjI=?o6;2#{LBUW^_>J@SqQsHgp5&jYkGXgGj40|xvO$o}{MU*`k7{Kv+8^O#Pu
zYz&Rd($rxD8oy7@6u-Xql|b?k(nQws!cSD|a`?ZSixFR<NNZAn+krFySbaKG|1Cg8
zyRzTrBX_m2#C()h1ADR$qzOfweqDYj$HRQo`+lL__aNZ`LZp|f%X53PSgRzUDYJab
zrC<dcSPjLl0~zI8Yd+(?D7=cfh1&Z*n;x{#XzN1{6fxp1ix1isuR~@g>9&-eKsuJ-
z=~3L9j*iT)C?(G1Q`A{!+#E9f{QM%zU8L(Bi0S(g#op;T)3x?H%y0Cqu0Fuv7Sivf
zzxs28rQRL<F!3yySsPWGusdjSH}m6{n0|l{80>HOtIAW$C#=-4Rv{s$10-J9_Q^)1
zq%{<4+e8Fj-4Fo;EL`dLY$3QOLW@UQVmid?h|wpRgNJw<Fw+N<u0<(QOhS6O0(C7W
zz*6%_X)>|rkRAeH)0@s*^_}BL-w-fvisf_u`EBDl3K`tv?Sq8aMH4Bh(&J^@o3nmM
z`_C<HRT`}MeI&|1549-bVqGZ7`mVY(Avp0+u-FWtyu|7)b^RL72JD_?sHj>M%$BX&
z?Cr}`d;2^sj~*q^SFR)B7g9<^GI;!KwE|JtMv{;Haou3|Z1O3Q`h20(Ft?Aq_~H$x
z<>JBq{v?h(Px{Wum!WbE{l^LiJ=T4#hwJPMr^7f^@6%Zn9NmB(JS|JNEGYDj)vfF#
z-oNcyOtetdD_!^^!{Qrtk4)6$MO>%30dy?V^&1tELhp$$V8$83WvZ$@fd+ROz!?pN
z$ndbAYG5k1`~JlWE<f<4ls|ilORIpdyaxqfCbC1|hp$4PoZ{pHlAu|l*0BNGa|oL0
zwb<5bSgAV<0`fQGJhxVd23%*Sm%c}EYtqP_8F`9yeSQm`?#?#uGEQ)}QMFuzdhXd@
zIuxD4x}mj#Zl~J(Ov=w;U5o5b%lwc2zGPO&$L;tz@=?YO;pB&m*w6r7pb9duYN*-7
zh;1OiQF2s}s1e+FG6aIEvjX%9Efq0qtt^O{MuVsS>`=7DPjfQ>7{kPAW3FDRMENvL
zvC*f<C&h|pRX#+k+P~?XlSa<7F7;JFvJJr*UW?;2wFuonEw|e!hI%wrJR2%!edg{7
z+Z$sad33n{>P}RU7n{xr!76*~sPakK=%>2-p^?+kX)-<kw{_hU$O2U~&8fi)t7W+X
z(nLHz(p1A43JBAxDRMnav&y19R33(WDCKrSCI4ZiT6P_AL70D^o9gz1VU60<6fBkd
z(xB%y899TP!=05Mev+}hjO)-dBJ5a@NxD1u_I`>aasehtQd>DQCb{kGugYxx2?}7e
zVf2jk!6{2rnI)oF|Kn8#kHl3(l)1Dv!QYVfGjvWu2PJ#X0fR>dI`%?AOp;|;XK2c?
zU7BbHMK+P0&e1nEmdlVBRvpHsdqTKpr(ami50Ikz3NSCUMegMHJABlxy`gr!HVR;u
zZyJyGW-!^jKRN_g7pOaR77r-f)QL!_8Y^-6)Sl~}Z*_&n$BjFFnR6;t8&m|fo{R0O
z^MN`j-ywS(Ni3p_;}3(URjeo(jW(--vN>%L|H>NhSE?R)ZVdNvyZPSV60!w1OT(*T
zFF}(b0Y$_6x|<WKLr5J_3T#4=dl>>=a;d+2X{DvQFZYo=;}n$ih9X50i+@++L=vj9
z)zQM_De4zGZgaKzV5>;RVr}J|0Jae=;pUN;o<*YHF4N?=s#Ep!ls{F^)#hQ-A0X(f
zb+&TMBUP2!s^0I(35?+D)t8(hN!1n+$~$wj#=pwOn7?+t_E_Zaei&#mEjz6{kkI>K
zlPP#Ka@gF~dk34`!4Y#ha_Z5ma_DpmfZL<U6lfq1FSFQ)x%2L7!PeR&gzM8q9buK&
zVeY$evIwn2e;#6lBKmD4Hs*goK&xA+Y?2llwX#qpx%pi_ZfdS6N#yI6YWZ5bqsj@s
z&H?k7&9%jkq*)1=m1m_ZO8&*Ahla<je|)W+Ht!j?pDQ)p$PY3~bKMKQ*O{*S!}c?8
zm85jImW5c4_7*|MZZ(o=!&O63y2_tBSF+E_%{@C#x8~dxN4Y)1>-oKDW{Cxp=#7U}
z%Vc*?xWimz;j8GDe<`2rPD`ADC*WrBvDk8I?BJuYjn(Mw!u~|ddlr6+akFm8z4dLw
zUXiVC496wfra5(2rDrM~w>q3z>2gzlme@9h6gsW<L*MQJGU#$rIP351k?a)s6?#=8
z*Bde?#|3o?1;L$!R1S&t<bUp<rOhg`CAsO=qwj&V$9m++WWF^yWa&E9K$RIT_WfmO
z(K|D#+0ZZKb66&6(x4vSuUKI$&fa}?`pq-52LnTZ(Gbo3XwKC(cVIrOX$k2)E87_1
zRV`AVMzVb+Z?y{Pk@Z{245mMV*f(c-sT)=tLzWxY6vMV2RXjz)q_zXRkyot)99_i0
zg3`GaXa=}*3Xc{IoR1`TS@|(^RELy`^Wlh_8kRt4z-9&Nk3QsJSXYfgUAVUr!!#>2
z^6scBx+8a<K!Hy8D|Ftadr}<x%ii9=Bt|jbS*oFw_gTpjYZb#Ky<QP8RMYFl18xP0
zll;IC2u%T9Y4$>Qlh(%Iw8&OFZP^dszSi|JaG~M*O$M5dl9r~dgf!1+IV+V|&bdiB
zs_a}c*20I&&>u3$hePc1XFSw$(OX389KWoGJZGa(zvWRbQr2(#thSrCH=6q<)WqEZ
zma&y*c_@GKVLly_;}>E(fSnPlsk6LtBO;J@mYeahhtwuJW9G}^h}OEfD(KiMeWvs+
zuRi%=(n=$<C%JW5D6}|Jmr9-;v=Gz@42*6!DGxWVThHU%-1;yk^eVRF%h_C8<+aW0
zvs?Ach8*r1%dWB|#3I3&kyQfF)J>5ti%0m9m7;w4<xlt1;{A@tEQE%idX$xg`c5>C
zcrs<!#mAU`dW9uOQ0Aqh8?@Qz^?<4e3u$eiifB0t>+7oXTkh(<NmA9+Spi~)139(1
zv#6LNux@kiA8RkZAzggsG&n<(g>5&ROHFNsX3>FmZMvJ~B6(FsgK5~kxJ6>nf-0Il
z!xXj{@Z#WD&}{3<w-#wRO^ZUqj6&c}a28rlgxQ&d=OLLmm-dIn<6D;MuLkGaaRUrw
zY2<;9bQya1`2$o1r#2q2#vd};n6Bgj3;l)C-`D>}nS4_6x?D|d_0y7Uq5i@yrrDVC
zu}napo!(w%36y)YQQFq-Xj9V4>02CztreSVdWSUy6!1go*t};l_3{hFc~V=Fp(2JJ
zg)nPyONEt#otw3;{SJ!i!TAcK$tx$PCxjERl50hp6%7f^=y>7@{*e@~MdXfz75|T*
zfv0LNyEqEfUN+vYmfVo?-&?UgS-xdf0BxS4y`@9>we`1PP)WtPl#q;TNXO}v(4DTV
z-=y7`Y(jwxIGh6<)RwCLzNxbw%f&}&-T}7M^JWgw>)tY`>;KT*UewixUc_c?i_Uh=
z2vPGAGP=t_N0fi!)phE)n%}!#vu+SbVBZTt_WwYnX@i=`o$hx(3Y*<|fPDG3hDRbO
zD;u&*9gc4LK)oG6tcF$Mz|HHNx$Ke{cZr#m_n$~*k#JH!LgfA0|Jax6T|}FOBzRU3
z;ir!kKEtUVghqd#(xo2Gn##$5WT1$O*9a-a);(g<C0qwkkn@wHfGx0NXoPCdIfBIQ
zDtsGJv88Xd75L44I0t^r`GY#*img6!=lei8s_@ok5sB{FmxwALv(aM@DIl&-Mo1U(
z`XR)}t4CB-L3_L^;XF`VNR*PUv2$gLQD$O_A=@TN9ag@B_W1n7ZMk05^~D?>0bjx2
zM$kRn3{BX$WnykV9kSf=OsZPm<7cSV=3*x;W6<IHN>K=CitOBuPYc+5dv68FHO=|x
zU)otkwp88hoam>-<>n@K_%Xt+D8NG?4AYr%L3!dQSdMKf4TwKWtC3qW&L2q0dKSo^
zu(gm}ef=^6wFe|eBf<WvytU_=hPxElW#?|cBVcr5rd<bqXBFabC6?Yk@*Rh}7ZUoy
z4B<A1%W5S~))|`j_K`|2g@1VZ^O(x693{~M*S;8c5*k>@kP6bXG2=LUmG0V9x8s9b
zOZoHJg`E)e=G{8u>#JrJn6ErjGT&5aHB5aQLLN=z#tbQeqM`eAz~BN#+StDGL8yr-
zjz=Ft1v7?3ZRuWukd}vdEd%z81i*Pcw^KtnGq56x>~c_e+PP^%1gaTQB~`ck-@q*l
zNL7v;9CT(C+zQgse+LCAK^w4ru`eH4na<f<)z|C;1MRCoP#m);4_vn#zj|NWJiBi*
zSaNHHqP!sr+pdgI4QGJusIEWJ_D28vmsdxZEFqlbVQ&j-Ei>39b~w0i2$3n@ZhL+3
zMLm<}S+{S;nb_4k`+c2?`w?=Z-!v5Fd{>_uWB-RHT1>yiQ7y;5BqS_)v6<kp_ML)2
zqFLS|^P`LtN(5eVBJneDkgJLZ7BlW@(a*C>2wE8=fyNhDIYHA^72Nox@S8qx=Eq40
zK>uV?Mt4o|oH>B~LxH**H<1WcRoh;<03bmUzuMGHzy{R+hpz*E@m8<BXRV1(s~jra
zHu3|mfZ!FS6@sC$0tCSi!Jm8$uu5wP+q&xAZ;`@*H;nDi$PwF)R0gUXVWg^Vg9DQg
ze-R`3N=77gLIG_#mU&d%Wr5Iktcf(UZ0+{Uq&ER0q2$)wBcy&vTwfnWDJP=;g20E@
zwB<KNrhWt(4k(RB^lN3vuNnXf4U8J}pA|rUZmYW*Hs?^`l&dnwJiTgOn+fQvS;5!4
zBmt^uQY*z3fO(+IkndDHh+L5)0YRZ-Lsv&fUM04@9@K~;^Xmje$;sy3SI`h@FoG65
zxg9DiMAWJ4=?*s+&{e&@{-P+IHbLl+rC7=4Qg1b|{|2Xr8GZ%h|A5$93UKYqeX}A*
zA4`!D5ir;(>}2Gsikn-@<>b7j3yL4YRObTH5lJRieyav#0@smvovvcXS!~iLGzu>8
z03$`6_1u(x^7h8cTw&(b(32yPv^-jeKolrdGIbd8KvnYzg+9c{3!`LQ`v5>dDz4}1
zHTMrkpki_ivnwv?seLL-U5k+@3@9-b2sYYb?{>;@|1#MaZM={I-z=Wfa)AAK<KZJP
zxE0`-eu&S1GY$p1kO0*z#LhqF_C`1XoEZhjxYV+k-vrm6@<AMc339oC(TA5F@DXS4
zzitH32gXBG{+>T}l!N65u(j^<H;KGZY&6}0s2Zq{uKJyRE6fWbgpwJc8Wc`7hfEIh
z6Zfr`;W7BG_ZnR6+Ac|e1dW)BczyR(4_pe`nZNfz995wRD>dXnKT;PFc3}mQ5M~iy
zV(x*w4^m6qzCh=EAD{p~1W7%vX)UsIq_2Z2&Ji$BAu1LVYF?v`P~#ch7d{7>R6avi
zLl3K|``~*8R;*y{Ek!OafbgfOE%U#S*6C)zy^-lD;)q>-9bN%f7%&3lp_D$#*)bF6
zIPk_x`IIc_0cHhDoP3H7vAs=bK)>Y;6l#*s3ZYBHZ3vrh;INg)E~5{dAO}n<(UG^m
z+~fC$aAFb%;8+)%yynr8X2$^(JoZOH$jSX%tQTp3bm_P+b~l?C$q8WXa?W?qSl;V=
zn06`u*A4V=aX?SAZ!kFi=al}@KTIi1dm;{J429ZKjNgE@Dlz`14n6@l|0>V!7m-8{
zmkeu-``LxoaM*4raT!;vq8u5XDBQI;@wEODdQV*t!TLYzh%jcQdlB&g#w9`~<@H>Z
zxhN(I^lmNJ`LSklBs(z|6Y6+z`OzOOAaL{0UsTe+g`|iT#73Ogm6Evy-y6B#cWSZA
z;E}Hs*_oHOw24U|?IuCux|R2rB|^>@-27EiFs;O3NKBPd4e%yd`2ZVTiz~TntOGe<
zq=@xRe9bnhf1T;m<8ac!(csF*Z|DJL5N9E8I!tF-@<g%V=C2+XqXYsbXsBTyu-mG&
zoW%u-cf8gFk@tP-9nv$zUF}(O)C3L?DIoWA6+4AF|Lf>qXM-HaQGQ*(d$+^%?4#UN
z9G`rAR7#Oh30S26sLK1NaIFS^m_^*LQ`wooOW2_>ohCFH48DS^lJq7058lE*da%cP
z0Sc{OK1TVY<TDCkO5P~4M^%d-$HQ1IXDCnoYs+AI%u1jIQPTNMC<F3^5}1;_#>dc1
zNoin8Dt+%z|JP0+cGCsM`aW{4=GTQI#=`p=QI1hczG(4vT!`9NFEI=f5|Go2@a;du
zmw4cOx88n0^6C2O$=YcF6TN(8q*T(RC2@OjAg<!$^|l9lGao;;^4)XPhx5QWiW8T}
zF5(}cQoHU*&Fet@sC{wIMSOC+5nkfEuZt^5=Jtds&(&$GX^~T%ZNYzV*G9_AigE_e
zCZuG0!^QFbcjwNMl9}Zs<Fduo6RpPf?fFRqF{`B+6F<*sJ>&L<rlj)IvoTjg-Kb1b
zuS2;9^9+}gRW9vQ^M@2S2d*4%QGslez!WzgI+-funyr?PvQejBwQStVTf&{&Y<Jxv
zl1$J@$jTCUSOHZpzcbJ09g#334(?#$e6~l$SU@>PN_k)%_dwiS`~$@cnm0z=DbHyj
zj@N+u>++;ppkw`F%EG!mS8d?drvo@fZ}{=B>t>&)dk+h07Aq3!rN?Q?$1ehR;YP~9
zT&2lykZ@DjDxl9Yk0iI8EHsld($uq)lma`HMgkmOH@JF2kSpFU6sOe<dZ#+-t`wH=
zAg*tq;gVOydwhruw4)M2KZbMDIpfD&55VMHRF;Sc!)<wLI`0?<jp)f`Wo1dFFEq6Z
zZAQjVTRG^%5_50Qu9~(cQLAisJjBb=xR^%Hp^(d4Z>@UHY!9c$p?X@G2bf~qIsHVA
z*-gMgb2Pc@!??0;x2H6eSmJ$@@!)?SzsDN~AJ*6&v5FI8r6qI;Ho;cC8`C&we=JM&
z+pMi;A4l<GPjh*|?CR;gYIUSJ+XL1ob_=ang$N;`UeMkm8I;r#=I}CmwU>n*N9hJ$
z-8sx2Gx`>7xDtPt<R-e~;G`|3)1!Ol`N(i_)rP%hLUxNZ4(>Dwp<b@#$dz4VfwWsw
zP*fZtu#5QBnEmmaCc2nCsqUPY(bIl-&udQyV6y17V%Bj(dnDVQ;H9u;LCS6irSBx5
z9NzNP1rSdIZuw&2IA|CF@xy(LuC8uIB9b$Gff9CPZ^s)>xQT44?KwN`-z)aM@3lwe
zUe$SF;td3;2;dDt_;C$XCZrFWp#oPc`AZz=eG_m0QLeh{JPd3Uj*^5r_IlN92fcZy
z5VjR<hjiFK=FKh`MjgxX4WZ)7f^y7V+o!wli{xX(y-yMBS7a6(Q_=QDsTJ5Ab<=Cd
z3xB|kbdn*}3^HCA0-`%vmHXJYl#i~eEBUG@{^MkExMscM4+0_f_!mfzwI7Hg0DBl)
z^7(_zExOEWv(Nk(dps`_SU69c>BgKDc)pkJ(v!a@X9GVxDw6h(DMl<3vLANQFzMu6
z|DWhb-|+CVZ0J?~9^>y9qOx#+tr{1c|9<E17d9X%XUI6wf)aNUB+``o`C0U%Gaz^L
ze$;3YlB1-kn8=~IjxHU0$HJ_GtBL<U(Z60`l8FfQ>@@<}$p3ohZ#l=vVfJv=>;LA$
zfmT?kX!^yczp3)FGJAy$XQ3}PXfcI_#&sNPK)1y_hFLdQtD(un;{vcyIkN*J=tu8i
z)+7IS9M83h99;HaPp6-yrL!|4wFOGVAxjVz_2tQFt3wx9GEMqe!{QrxcMkWtK;5(p
zrJZo~H*!59(yAo@WHAA#{8xPxx5n0RHA7XIHhLKVFf9|0)5jH_f_PjH`T5I6>tVdX
z3vuR)mst%0lc&&Jg>XM>7PLE3ZKs4E^GD(_<@BxwnZ#jG_o;%D*O~v3$9@(H$LCk@
zE*(=?(7t%QFwcO9V*DDMWz!wnS4ckfhp4p(;-*Ux7AM{0P>l9kzGj+p`lNz7juath
zuN|@6UmwCGM&m*~SYPz3*Ith6D{#MIGF<bQeb8`v>3kFXYX)-w$-`ccdTw@lzKSGB
zjR%rL7*c^>yM4J_$7#s-L!Tu>{{?YP5uCJma~JA=k^nvlT%;1@@=<f(v;-~FkTR3Y
z_)(m+1JPmjE4FY&-NKiMvYfOjlE&p^$xTxpuUuf!sEmqx0@_KeN4;k8Sp7Pf2pSf4
zrhnG`r>c4@f__NrhoLUf_QG5TEx35K0j)TRRIopaYKCa)!*9;!&&^@G2xf&!B?|2q
zs92c2l%KwI``gU?{X+TxMFoeiauV9ATm+S({{P={B5z1%oMs5C0OmmiQjZB{dxXHo
z3{VO<E?Qr?d!61K5Chojac)jqddeWEu)p*Y#6}Otod{6bYsxV_LVbd~HsGvudB0LT
zvHPGaRtO9Bnz{86lukf#^$Q?NJzbsvgC#+~`t!FET~)wj5b5ISMFkaMG+?K*MgzaL
z1A#DGF&$VhrT}KkQJ%auo~6ivYHQS+0v)3lo6O6wt0ol|2zH%NO5kv6Z$YtA7E(Zu
z{c<-T&TM15j+cd+GA0w^5Bn~$L}y$_oHQ~s`r2Pqen+h8E>_U14OdD0xVXNy_66gp
zhwzrUfgk}4GILX4OvnPy&>&jKyr&|n3ni`Nr$LO#Znq%qz2ab;)E@|S>D?0mlVb<$
zpqB_`z!V^v!jL-X)rbE_un?eHax;DGLlH2*6gSLR)4}GL0tz)v-*c+|LKF+`3u=Eq
zwi_d*RFeF^E~ZF|Sz=Iir1-s>KZ#zDuT4y+38Tcy%J3s&oRl?6BF$`sR4EE@>ldWC
z*sjmIK3&=BJ8~bi=;lJJW;UZ0;L6{>lY6dId##Jw4+ASuEZ3w#`KY&jhdL`#%MN1$
zdYGlw0~?NCgR5$inz8j1H`>ExLN6rG?=f4r?>o{$U!hTGV;FAf<{W(V9#p<H^E`lk
zker%P#DQE_t(hR{>EY7sIuP0<u|AapTn=g#n!|sR2mg^zQAC@SG@h#jkM2>SG6Ngd
zIFk`AYJWT9DO|>BDZ1FE)H)KO8*g3?^`S3N<BtkH`}2LDH9FqtXa36W!_XE0O#|jV
zP|axtggu;0V&$tH-1zva<Et{2@aXesVM`sCI{x_2Z0`FBN&t;KGKp>j>Oc0C436^y
z0?nF7oIDxs0oP<F3~Y=FCr7e!#_KtC8%Df}bLe?9eU?mop=F9E(?QPv!SGD^_Lc+L
zbff{J$I4JJd*Iz0s?3au3SrAz*xJBUlJsx?teWr}9(LYcxh$}aL}xTs8&}PKgt!bL
zBs7iI7tIt+zWc%poW={eUBEI^+AtHYRXUjZ6lr07h2=2wBfel&z%l!(l5wr&Sx&$&
z|2(RK0(Wk;jAQWP{nwZDiy<C;XybO;<x6OMTfq<Tn_tw>tybeSAi4qh-{2B<I>FbH
zQCV**QE+l12@?6F9f$yiD3<-+bkeB4d!wU4u4~!w-xl_-7jZm#uy>}<0FD=N$5JL-
z;Z#X`FTDVDp%c0MtKIPbS?B~K#a$f?y3s$QiPz`dew4cqPFm=$>h+h<n{U?@q7DE5
z=#BG0D)h?icMabA*AM-@naW7`_n0OIS_XQ6Eq3G;T%BGPdOy0ABmSWf)OM*dpE!Uf
z<N(`Buqpp<Km6B=oehA)hKo7Uu16y!w>aQ!?cMhl!6oggh*)2mWfUe^Z-~K_>;gcV
zuPZ_^oiX`@k$Y3`Y5e?pk6cbw%lcvwVCnA^H`hAL<>0b=dzTST-KaeH{T!Q-2CC|P
zo)D<Kga)<b>pwez)Av29+0DqHAH5$jJKHY!2#EGMkSQQ+9YdBt)`RhIOUK@D8=%jA
z2Ufle+&WU>wQi2epzQ8>9ra=82srRmbgqW76<9y33r~~J8diLfjT04|2DtqDN8PP=
zU)#OTqae}#6Nn0!H-f7WcORx3yaHXzS0O>VhLp23;LhYP&^-~0Fto_Hb8n)wI)E1Z
z?LbZo{jddcZnw6H8`X^dKCwSvcn1(3I<NNRzT&mrE(ViHOsKbb`(rAquPE^tgc8)I
zY_+J>?l^(Wyb>3bL*qCGVZnbt^WUcZCu#o|=uO2-o%9042qZFiBX{H=c5a$pbfz8}
zo$UHPCjQY-_E#<;1<}WTW#@y;3hFzFpXw5{W4=UsO7a{34I4a62Wx6DIei<zkb(**
zO|=cN+&KHI<ywe`FmyJz@CYR{pAX<mga50&w~UIi3*W^NDFqc3B^49{1i?{48s!BQ
zDQRgz1R1&;$3jVil2$+@hwf5BC8VTFK!z?U1%Y#K(6{*gKm5-+>wGxt^%Kh(m_5(4
zpS|yW-B(b%O#XH5gZ&0KOMT|PrSNC%=R~+_3Ykb@fkR+2705OXebor_;)Pb}0F@(u
zFPqQogU$J=u|N<JE^@y*?1Rl+y^A^0pI5NDE1^WH$2Fw?{&u<q7$NHcn=L^qcgjLw
zJ@sYJ_1>~jQ}`$Y&=;%k_tz2e_r^%41RstWvd8@WJ8r-~xTomPDc8>adqoFu_7uw<
z-q+n+TWvGXD5CG;C9pMCOizZCB&{l-t8&ZMP9lZ=oFeA^mNlKiCjl_P5bMINzL{m0
zNl*CBle}L_EhA$zRY<7DLpL?>6#m^m4~)111UxIXlz6t)=u{;UsnREEhYkZ@;gm`y
z5D{V(hg2Z+>+eHSplyeAnScKVSlDo=WF`jYD(3CH`seu_{Sdb^StI+D%A|`~H?OMd
zzOqwSb13KBfcJ7rg~LHlB{84opRI>b5>#?1XHWM-=$MMqE-!7K$1drS)>~aM+0<d$
z>M3+=y`3~1u-7HWcmI9dn>TbPoW4%cDck?!D{Z_~33v0rI6MD8zb%@tcj3nDBn$jE
zJ_+56bl8R~ovXK+-9_ucZN7WKaBCQ-LNYua|9zeR?#_Rc?Z0Q|zop>+@Irm!KzP)|
z&V#7kgd7730Jm)d6-_e)tTh3I6^v9>Gl4g^EX%(T>R@y<wT1Y~*A^fFL-gN#MRVvd
zje?Dn%(Qb`p(5vod(-dYV53uQjsU%~RyUBaWPJGy%t%qaQ3%usMUgbN&~bfT?@Bm#
z8)WX=4;yWvs6;bvoNf<*dMHTr``bOQoo7tIW9~LY3|#@zn=EkS<d#9&6OLlvCU&QQ
zriu?qnMQ$3coo^Jc((juwPl;D1f280pt?F&ADA#SL6bUah#zSEwjleswJmv>RA(-<
zEf%q;4%yUV*zam1V>*P5HSahXt{_&imAsrpBK#ojUZ@TD$BZJaD#P*R22ardN}U#m
z3D8L1279fA7F*zK#Uo8;y7|A&L0KAPVXHj!ITV9e=~g*#8%UKMe@xyxU;#?h+lZEI
z546j{QDEFNPbmC<V5ZT8Fu>Accreo{wNAXsouJxaC}jPf4XFSBj7Cr$Ix-rhp&EVI
z&CQMdgdsj2K*^Yr^CMWQ<H_-o-Us8Lz>O&x+hV8zy$Kj=Qvf>TBZmscxEBbjaY}pO
z`0Wb>j<vNP+F$TL%!5q?rWgP_#^W!5au_OLbIZW(%>%;VQHm2=ij)Fby^Cfe8oIeE
ziCzo&Bd@%hE{QESCXWFB#0nuNorzUINPdwDwu2Mx0LyS50zAzuf|h)BrF7k8xiz<>
zrFAyUyx_JVXl*(LbCh;_-RT9==M#=kC&Emh6=F~4W|pg>C|0e<d=LnT8bGL9^!SeH
zoPQ9|=;D!)o04}JM;nA!Jwcg&?FVfd&C)9m`Yy=nID*$G!=Afug3ncpcDp(+7`&ge
ztMocB3XRk114rDaWL}&-o|~eirm-?E#b_2VpnE_O>#=Jt4S%s<eQiZa_(N=DVz2I?
zzvAqrWI1{88<Iz2!qgCzDg109ulsw|BiU=okA-FjGi47nfdx}A&<lFl#vla;J**7l
z5k=I~OOAc~@qpN6IAt3I2g4(U8d@BE0Ma!VjHiZmwLa#Uy`J&I|L3);?~z$=$rx?*
ztyTgPs#kem$KQ_Bb!v*e!Y;M0e57*t<9Dr&@WduvHBj6jhZMrAc!1S-vd<MOOvPI1
zJYKqbJ)tPP=9E=1VFj97ye53lxJ~!cy@_wtO<tJ1S6szCXg}TQ^)2v+6jQQKOoym~
zlct!Z|B3Z<QU1>8A8uz`+nDPX20X3Cxn~={@&^r9aZe`(iG&YqzpSE!6;G00cyy${
zuKZn+^5XXfwU#B<mKE$cZT6{Ri^OhS2z7o2*H6iHG0DWkS5H-?=J0gjsbG;cb+s0n
zw|T91v{c3vht?Q8I1|6QlvEq~PiJj+!>3fQl#Ez2TvX2*(IB+_lr%Eit`6J`&6*^K
z;oB-d#$_DFk`DZ%uj?aK=FnLbNu=6q(A^s9sp)(3yZBssUCp5C)7=G@-GqX?=HW;p
zB7wpye_T+8o9LkMnC4#%U@C58n+I!;&@K<XJ%?^k-0<YFpKB)3d_S>UrGhS0MV>XH
z^oU>uoXk3xM<q(Cw766lOAc-;o~suVbNf-FD~G}#RZH_$vjx=8-;8_jWE3z5os!EK
zd?OHGbzfKw+vU(jGUSe%L+%z@CKl=*6t@nWv&G07MEU_;_;{GwO>iDbt_bI`Yo{aI
zKOIJ=Qv||q<*}ea$9i2zr0NKll6UEB&iH|c8*VV7k>}|#-xi`9D~-YtI)qt~!)PaL
z29(^+FIIUUeymb*G}!%m1g|2Ep*qpaKTD6a45`W<zDut2ZH~Tc3}QBkpLId1ylfgW
ztrYn!pt|;ls)xn<>G<y=2i$m<O&b$>b;E}5agH&}=Cg=0SJZ!Kb&~cjkh=b(VmKWh
zXAJ7`T=(!5C`x=XFlXti>b1t;W(%QTb%fGHX%F9*uN!p-T`vFh!B>Y{EG$)lxr8EB
zhtJ8{V9Xw+7pn_Xh0}82e&n)v=$L4iGJ9p(B0To&VSQ)ZOI2@(w$J#kX)trqi4FP#
z-N958|6N-%j;ATF;_}#uJG}c+_W7HY@TLi{@pc~%^FK7oczjBZ;o~9cS&3w6c6Wu!
z`*|qalcCa&`4(#gN3vDM1HXNb%G6`OQ!_Ky#+p!8(BfJ@elyrt^OR&h>2yPJoc|t!
zg*}tYScfbd*F5tGSD}(<UXK%_exZvxr^liqUJ>>(nMc|dFYy$_RPFs@3~xAm=bF@=
z455ldC2ztSgAJSN=?~Wqjiz(l{)u_B;2*8>b7Et-_jes(daU~RJ)c<0skOU(Q+uMn
zV--Z;iA&He-?Ag=C8N`_hU=4Y-aYAFrS{pPqBc*Ot^I?{R<R>`!|i|dkbfOU-#kOz
zv8)Y6mG31y?J#W5$=th_8Or-%v+itXwo$nJ??WNZd~-?n{h_OWw9y&aWsG~@D*^qv
zhuMVpN9?$b4-cQlZ*s_=oDf?ad=xqF0O&~C*y^S|)>o>&=}{u8JVPPOyWA@9`}x;L
ziSr{Sh61+a3n{m800CrbvdB2xpxExn;g_yp*~Z5QE$WxPhLr#MMIxN}J}5t;Tf1Rn
zYQC!T?U~eL6m4h=%^BX-aZ2>KA#W50e?GrLJT%vo`X;w-cq+ut3ljOX6<z=*KzMc5
z-$Mp-0ck{<fGQTYTW}<L9-NHM0d?(o9WZ`YL7puGNccw+A=p#qaeG{&DNRy$ZW$R4
zZV{P1eQpYdeg)|?Z9#$@4_yte6aWN}vbJt5`=AMUf1(UVS$XKtXgFNog234NBK*ER
z0PJODkT1O8*C!H+-8Wi7PsoHk(V2fm8waL+C+i}Fa_w^Mwp>@W9eLZ6B*M;3r8=YT
z)oLN?pP_;XRcOF;@8WS;oS<<qxw2*SIzY)v6z6#Vb!t}rANUnoQ|Dj7po-ha#<uwJ
zfgv)H>H~(W*Zz)RZPbg!Y(VPqPD%Z=<)e1O%~yGoPP6sKcta)f-^cvlxM*h*XgN|o
zVNa&DsA{J4JRY=dg+eZD5jsEJEn-0Dj|GhecW#|@HYw=OuTAb`b30)PuE|T|c;xYd
zNM^Ecyh-NUri^QWY8R_k@+~mG8o5)*LW|4DJS@<J(iBRomee~<M#RVv%ej(6=V>z%
zEH;G;*jRTZD{Jb(^SSv3)ILKK-I@XEs~~DL!n(`ZNs=fjShKWdyab|xvLG@@>on>`
zYUW<NBW^??3<dP+G723Pim{Zwa7SED4<ZKPe~_2|dtg{apCcg~!@HN-OLei4;#6@X
zVJUsOF?+7cL;vaf{RhYY{p=9ocGIAfjnuWjH{UnU8n}~iY>7-u+hPa7(ttz7b32>Y
z)|Y1P16aBLV9=K13W=ZsAbHyvCI6z}{rR3qcnqE2z1wQZCr%E4ot<6`+t!!L_QM$D
z)z!B=w$vlyxGRU$Pe>kBc(lPO6zZMz)6#)<r<-{X0NoKs(!j29ejE>i%PF6C>|ih^
zfCH<AB3+LNFlm~i1FoL!!?M~J=T;WlP+@v*XFLrFw6ReO(0!HHeOOwaLS`f>?xgDJ
zEd@WY%sGMKB$Ol)@bFKb-R~cw5dna-VoQc0+*!0UsJgK2`bRB?HXCtoSZXLAscgM*
zpU|F=yr{si^(EBkK1iMP?==X%JJ5z&u1~(?+4|CH;^!KzDf>ot;!s=Q*U5<$aYO1L
z^#dH)?QcUDR{NS*Aj_Fv;-x+hj{<mFnn61qgeX~{T?qq2w&9=Bv*+}R`i`Xbr4Sp=
zr=0yZq3W@46n<lQ+%fmgxPNvTkZnRx2LNgId7#kTt6MT6s$}ba?d92TOc>>`leV97
zefcOS&JauNNQRvA2za(eHT#xTH9)pF)v$}No#hbt6^&|7JF9saJ|@CF-O&wai&3P5
zFZlZGYDSsmTnF&8YwuQAxO8<j0VSik2wOG;;!bY2xv)Ma%RW8ECD?s!A$DVTj!@T>
zQP~q=`^87%^s7leB`Q*pB25*E*R60I_@c8LS*%hj-W$zfkRDD4U!q?X<9b!P7?bmo
z9C+AYSHKkr{I2)J(H4CnWd-I~kCE3W?ohmsaE-erg<jxk&UMWGN6egi!@UYGl1X0;
z=8~PgD)MS13XB7r0D~tJtmPvnA)`hLxh98_xri(vMsHhuorRu%5Lj@QWx-)H-oQ`b
zge(|0lo=QdA{C}B&_{L^ctjt9tkMtSVT@frsf-oWN33wKR@CU-l6-_+iaJzEFA~od
z@+_AbD~>apj5{6ZS!dN<%}yQ94kj7MoCNpS0#&!{jf>ytXZTg%;I#i^=A^jvyeC-5
zw$kg!FdQ9;@!zQC?O)_E-h_{oU7y*r{v>iPd^i@Vh<kuI*bzKIw><AD_Vb9y`j2{t
zld3O#k4~rVJrB_x%^*LG?9pKg>@<$?Z2}46z(gM<ge0T=c*5_!^BwlPAD~3mMt+se
z5ED&i`yGe`q62$KtRnKft)sshm5HCqSx6Lr`s>aVceajc%Wkx<Rw1;7S*0H>%(oow
zXWde~x$a-I(58B0yqJYrVhp*oAQk~{;>y>PVoqhEZRfwNAM~0!Cpk=8$;n!;>Y34c
za=u0J`iq+2GW8v<qgQpMk>^xqjVv^)BEX|+^&v*vs`5aG_L0ULtd+F=AVNH%xd_U(
z?Udjne470sDP+hU75dUq?`|&$X}ilk+6%x>ibsC?{G2n7Sp6>JU3$is4VgTgM8PjN
z<Bs)&PpLUNd$0Ood(wcB^BS2c+1cd447|h1w-_zn_<IJne@@i8fulQM3C9$_H)L~!
zx~oXJX)o8^?~0h-uwDrVUus+xJKLsfQ=#fC&LTzOy4)Nk2NqfD&y&okj`&Q6l*heV
zk!k!P`J?88&H3P0i<F4|7iG0y9Wv*fTat7FC7;DmW8%yDR(tw(sYiaVWY$e&t-Rli
zq?d?N#P=l~*b+}NbiY`UtHyqhW-6!f43*#_4CaB4Q0CxZUexM@(dyJYuNy@cPBoet
z2Ccdts}uN!dp!w(hj?l|+{vi6-*k}QLDNxMJ)&v&wL@uFa)t=|3}VPg5^*THf`XN{
z<Oytu&GoN7tTDeB$FDvz8_~z*;p0WeSnk%QBt#IYHoCnY`T0r?e%(Zcuq<eJi_3!8
z;0X1cM$2@fSNzLed=+ff`${KZMOsf*w|akh)!QlOtp}8E`yUC>j*mvI|A_LW(y6qv
zaGwhs_l*BO#4Jy2Cd=P#DPf>Qg`<<5Id%5ix#+jIapI>kJOUeYt3N1Z?jN~@qkG1@
zEy)?x)nxGNYe!c+zQ&h6y!)Aw7?X?w6fF4UtwkIU#5fPIFS1^dTzc_p&buJs>F{MA
zDVJoTR%6l#u_<B@L|FtJiFD8TBq&juvILt@)odfqz$X(_Bp$dB?$jSIDZo@mY5g9U
zSOBx^HY|)-Q{PguT+qlCB&4$wtH;Q?lliDN{OCERaC7}sj78v6BmV5D0?<P~>R-Y1
zNK)<c$pUiS>)_I#Lv@~CgS;o!pQf*c4rVvS5nBznK!ilqJiHcnt(*d~oRf5A>z!LO
z8M->JvlZ0SMCg8g=_A~g&OR3MvUHTz?!XUfwks^|U$nrA(YJdEU=nn!L6kRxHW1^I
z6-mTe-616w;Y8k)w%<OUG}?37v7E8Dp;h<gfQ0?(J^Kd9!zl>@Z+JV2f}*yi6tuc%
zN55H9k?K3eTCZbau`q3pLxer{DXPabdyb)cUF_xq@3D2Oy{mSYO@s<%qD9-V!W0t+
z@Q#vv@4FG!l{93@O1Amr2C}o(Gd0mA%P`!-=8@o<NRVgNJS<Xt-~8hPb_MGnj>?m`
z3HDwcFA?0iNR!?FDKknOze5ISKrR}?m&Xp3(+;GUFJ!@P&_<it_PtV|LHALN7#Hv#
zC8K)5*s~c7phCb6DM-j6`heJWGR5>1HBzq$-ObIcB}a97iuTc!UW49`&dy*A9;Kh?
z^H%Zcs#xmF(%mP_T=;kX@#R!x<RJ(BU5jPW{bSm;QS5@HY##JZyN4C$-M1ybk{sfH
zO^gVCgU)*oME!yf&vTV<Uh?we1W~7t_?+h?_?6AOX!OIj69Kkq=@a^xk1noykKWU~
zAFR;yocl$toZ!B3Z?&bS{*Hd9ELYEwKj()-R=5WB2sg5ENVU3TLsjoj)T)pGPgq-{
zu4hvYK8$4G9AT+ssW@vWNYgJ-V{`IW2R<Es)6hC_Q-Y$6xQ;!7Rbl-o({Zmt$$^iY
zld&kt$x40hke076&<N?$%a8+sEm#-xeDW^6nf;p3ziXD|0foZqlE=D3=@*r7S_xO=
z7iCIeFvs0Z5MHG+ZcXbRCzf2qpIug^cUp7~OY(5osRj1Fqu^R(5MyoO+#IGT!zQA9
zDr5b<-F(~2tjTdcoc&!jtKdO}gv}aZ5o(0$SRc1Zw=V9%>Tkys?ugr*WHL^Tc+1uJ
z^GLL4fMuG2*-3FdM;x6O<NmtVX220MzbLyEuXX<1<}y?r*g<fdxkyEyTrcnNi^7>W
zYugxN`b}ioNgg@%+U&1*|IwZ7vtjYR%W%myX0H6CTr?wT==3eWei;yFH>9<2!VoLx
z+irY3B+hbDVe=lF(mN5TtMnv^uvAa9UkT#~h4HO?8-5z!cCr}}i?C2TNLN;5Y#gc?
z%ZVY>y&?-XfyE@ZIr542e6V194Xq5W!h0D%zT^##q4P4*K;{UXNEZR!ax);s9j@x;
zyM@KJTl1(oD@4QSs;1kZ#Cm&6b%=}hk=OuFpW6Wm^1-Ds%Z<C;r0cmqH%eS5dh_CS
zp8ZzK5m0)I;>M4vRE1aKb1C8X>Yko!7z*<@(>Xk%-Uatni-q_qq_eqX_PZ~{_z4g~
zP>BYbd|twA`){VHY>*T1OcyA1DtBPydy1Uqn$AVe`c(Z8gWiUs%)Y*Gs71Ek>!lsw
zl_%gUU8(hi+kMuXaa`gwO~?A6;PSXa=i=TG?q65Y3Z_Odn}fGdTJ=Jp%SYl_p1014
zVry9?qMxyDBZ^Gt8nll}L||bbUgxj`gUQ%EZX*(-AdxnIg2Jjsk8?XANM$4VWc>Ao
zHr6~Q2m4~+jem+Xr7N}Nh87EgHhDkIo{PS4<DZNE2^><mQF0fhU5i79JG1&}4Ja#!
zNASM4(!`@rLve-n)8o}2GEavS@3AIV#f4-rxZoVUu8Z)J$BlK7xtNKn5@F9A@7h7d
zH3OxgXjpLK;Ucg8UEmhs>I_AO;P)%o^x=>4?$&%_s;i%oNjz<)8r)Hk75QA&`@K}|
zRcU`qBB=ERc)8k(<YCU3Jz(Xutly89SXj>sS<^V5PgSV3d0XG=N<|v<BonI}oSt%i
zu+ucFQOftH{u-QRT6tPMH8wE*Leew_nu8Qs$1PFc3z4D|+}jL(p8eSdx(a|-ir=Kn
zU2<-C9yuRd#SSqER}BOX{-pW<nw>NHxVOuj73C$Z8z-qp^l;u>1#hku(0MVJkJiZC
zgYmpx(|BD}RCsN3JU4XEc!GY9r#6%~>1iL(;G02w)~?QMa^%FL)Oc9mKGPQOL+l}{
z{uPM&3Xy}gqKa-A6*duIi-DM;1ZpEcb2Z_{v`7u+f-gqm+EpC5?EQ4fo4d5lq{$>F
zal}b#k4GtDdB>#NK0PPDr0M;3+2d<csh?`wkKz-XQ_9#M7L9THgJld@t2Pd0y)d(!
zD}WH0v(_V_NxE?RJD@lPYygC-S@9w%>83a$9FFv;_|J5IH%>R$qAZwZfK7UQWA~gD
ze3s?<<J)JZKMqfIBkix%<vCq%gP5poyBH`*&f;GEx>~$KIP<-aPVT9i?!1iQTqosr
zeETR#Re9Z?avaY}aJcACNwaxUi1!hGU(i7cKxg;D&I6BhXX<n#@z4;uF}FUhyWV;J
z7NX=M4~-4p>?*0g@zlqYQQ_{e`S9C|F6GzPO%(yy>jRDmR$f1Z2k_N+_l2Zp5YY8~
zuz2$Ty!iR-!BDs`or%(&?CbgD`Lw^v%<iv=y@tefQSyUak3qqitzG_V#Y0Qo!z`>1
zhrY|kbUy;+V{(`-5jMF}8*hn&Qy>1RkqoDZme-E6<G0~pn8Xc!y?WF*l$wBWL(>W0
z{0+}J!il@46HNLypvE_=4Ljhe?K~PM!L(Xhh-^SV3s1KID6$+eXe2JSj}g{0pI>#t
zVN-LJsf#$r3=5G;l5y56hV$!rC2NAuR4qAi11mE0ZfS7IGA7}YDRyN`?=>!|T>b^7
zevEE=%>Ay;|D-4!$;d@oA!tYuIk7dPgB4jN^jLq+KlOJ#wn8aqU=T)WmD>tfZQFUl
zZ)OqAiRqa+)HddIVf=VZUn{BtR?cXPwbtJ$xZkF8kS%6&)@25FY4;RLusmGj6*F`I
zaESR@W%bDVHKJ9!KkB~d5kFXZadugc-pS9f8FUMAIoQCXX)u8J;G(a`eT(G(C)9S1
zvZ6<Adr#dpv0i|{*(q6ChQB`A^I_WYy-<edad_^xCq#IpY)eG6?YGYGk?~6MB6ejg
ztug+hp|804*Cm(16wU}43=RHVz8kg8ZU?@YOTqS6v53ztTvpTl0u7iCB3E2uw<S}v
z6|7&1T*k6pCh4U8RRqRD5lG9h`gFgENB<O~8=p1HnG#|QW+kk5{xz`5_4SuRj7|c{
z9Er?4)_$5&C=2%=Z$gSqdZX~C8Yl06vS0QGLH8Fh&r{*ES}{J4f=cEK+KRb?B}r1a
zz;dN3$5G}x9MY?4nA=f{PQU#N+6QdGES1O436)Z$^$d&}u<jUdOYh7imR$N(D45<$
za#W+JII{{)p6S<u5fZ$Vl~(0<1CDd+RsySbbhd?s8eib%x-Ns|c1R0mO3P#_{+>$F
z=C(HlkAH2<$Rr4}ysIs`Nd^*bq170P4<2%s12e7}ptm5xrwF01R`&Zw>~PQJ1lcK%
z`BvSCCTay>E#pc*cfrW38<~DY^T_*Y_xUe(Zul>p(^Ef6K7Z#g)S#g#u|19$Y`d2-
z$YcIk7_c!R$!fiB^mV%-njj}ud1+#GXw<`qRL^f<1bCmbel05o`jhD+%#2Ci`MV+A
zmk$ogwli<Z<Iz3@tJAr{=P^H=G4pIEqC|%Rl$xL%ZYZib3ujP@aLA?Wuzb7CKD&7I
zdcCu%Go1b+h1{VX+QGC9)+Lt$khhQ*Np&BmFLy;JIwJ=~!dbeTmGyTn=$THj<IetC
zs9X|Xs^lb*!0D(j7Aabil)8Og!G=FsWfhH^detr^wET)T3-unRM0umVd(@00#Z2Ce
zZQ>l&u-HqypvkU3gj8?wgCuZKQ%%{LKyK5Fe3ela0bh@8vmZx1@m{#Jbe)9|;A#sE
zk~nxOxF^9ns^k05#)9wkXEZJhJKX#8YW6g^kBRH?0EV>A=D#Kws%mA<y>oNL?1af$
zV~pLFd{%jAt2A37r0fH^8m;~9tpXe<Z*+hH>d1QO&o<7VKlsETq4hu^&lUu(YzBbF
zaz=rzff#yBs6Vy0FK$a4faD>l0vLKf?A?m94MK;2*Y7)*>^}&U=zV93U<fl(ep`A?
z)O42s(6g#=xV;tX>xUsc<hLf-I_&d^xyo8{bt|Ei3bgRNMNe7(#9#mXq3Rma!p~}l
z{)R079HgEj458nb>B67?^ZSQ0esFyTA5AxH{jTMpMcZZZm166#tpIgqxa1nV#YN(O
z_ZB^WC|2>;)mAx(LZO<$f$!<T(;5Mg$`XkH$}C9fd=1vQs`(n#znX{hsFkyaHz#p}
zGK<sMj|_dAAS;BSr36f?n?WMhtv<c@03BLKm7<oFk=HU=@))w<n=cv`u_$xnv(FXB
zFZoWHYi9$0Ror1G0*9xPLp6S(!^K=_jZ%zp5O$RO7G=f7B2GqnZyhCCg#>?meeRj8
z&+SCDqw$`QOTNXto=p-w;3NeXL8hnO=yU4#6?eS&1r=^-Z<A7TIuniVI>~?d4Spw{
z0v)4@rd<B-iFl6!btNWqhrwsm$+@3WHw1a$+X4wE9=TccVfcPed_fh!q?I(K@_)|(
zEc*=TVc8uB;D;VTs0$13x)qm8Y7f4B-9Ie>R-G`T&PM$^oFDBOa@F_zPl<n#jj<dJ
zNpnB5T314`saK%NuIW50=>KaTotYvglrb!wP$2p>6~wKGXI&!+kH=Yu!8wW6VVl5j
zDE;4d3<^PIP(0dP$3t~-8^n>dW|_am=Wjyq2&5&p>Q`@)k)_xzLGQWId~_l1fqL@u
z;>}(wqk1$84g+F5p-Upo*KzYXXVghZ(M%W-usyn-*9(41FZ6%V*E>peIu80?y1w#B
z6e=7Gc?1w`sDHrI@F^^1g;}t)+1tM)XYWBxQvk$+UddgHf&}L{$SXP_DQ@;=9?27j
zk(9Dh;RF0#0T3VAy-{*D=?mq>X1J#Vnu}yrJfKX>g4NFmKe+8>#CwSzUbt-yVek|c
zcTAaVtP9$l)Z6Yr1y_2lJh*tma1_FsuR#2!9I6SLkohT%?p;>NhEZO;lk5gbYop~X
zu<+o+cwyhFx!`osTob~(5K7$QFr>p;HUo2&ySRUs#pU(y{_E#WpB$C*J1e#J6g7U4
z^y_8~D*KlKRE^RGaALYHr_y8A&<&j2FC))QX8?{O0t?eKqaRp0A*Ew#-hh{uA7;mL
zlFmY6cg|7OqjZh{2gM_g{Obx<z#+>wWz};43ixY&Fd`qFUSFHGSqW7eNVo?KDE|!E
zy~d_;*W!)C2=mW^i;lT{FK0II<rx?vEMh-#-;KfCa4oVx9C59Zy~Y9%&Dsp#Yl`!G
zN71>vf!F-ece{Qk%SktU2{=-`9CP3s5$Z4$R6Y%HEmj7aB%ipz>%3~&lG4I;kg*vK
z72W42Ak|A61)VXD4`x$KT#Vr~igdbl@QY9MHEifPBe_=pcYMUv<#8i6z_7tGEu!Ss
zG9s&!IW{p38^X!xToMyz>tB7cXm1j5_NCwJQ&pN<JZa{Z&0XeJ?>1$_*cH0;-um?U
zCiyQg%dYgnR1zBzNOAhyhLz_ZFN6%~<)OBMm-a^2Sf|^Hr|deW&<NeK_0$bqMCx%m
zNI<(J0(f{z2cZ=!F>LMBLM)ZPTE~%HShQ@iEw6jlss~bVN`9nW48r0HOSoP8Dcul5
ztaN;ra$9>&z39q(%Uy}>I_n^MHWiBcFpdCF@>-)~jA32)$Ud}=m1s1WwP#eg#2!pD
z(V7B2-ZUiJ)IAiz+w-gTp`_a?rxeOrcjLt_y?eVf_Rw1wA0(-c@BjzqJx)K>Cs3|Q
zdys5#ZzC77&tD!8;ZrurwXoF}+M~j~g4U5}7t)6%o`s8H3ctE@iQF?N?FX?An6i!{
zpvQVePHc64!Y_tUc+HxeD%~0kuj4f=*`eX&`Y2|3dKu0U;XSuMJJTwzv#vW_J+1x>
z30##|LipN`m|6G>uMO$09Y-Gi)e&=#<z0_Pfc7=tEJph}P(tl#uXnT0K~p9>qUU(9
z)-#g}ALijKRWNB&^{otn)}xZ#&e!bY`z+ghmo9hH(@j|ot8=#9l`oG7>WMq(5YsZv
z!nAnbi<#O%<S;~-n00(kDuA}k#O}Pjuw<Xuh#G~!lW51yVm)(2-)53&<X2!rGz0N0
zsXeeAeBX(VU$2*jh~KD8r`Rz@plHm^Lo|Sq#CH>iZE+#ANuoUC9&5uMsHq%5NmUmv
zaE@{LK)Ij(d9_LN7Zc%hQ%le<IT``_#{@g|W`st}Lk3f^`$u>DwJJ7jY`pLg`)I4C
zg&7sQPk#Q|Xw({zhTg!rI+mjMk1qrS91Mm#KLo$;^$608vGF=Q##3=_{}25SIswH3
zm6u+2YMXRWrw}GL8d2n;GdO~5jE>(CJIpaG82;uLy(-Br3{lK#ffTAtfoK5x#0(|)
zS@jyi_0?}sHFudLr7(mIwznT-Cs2B`$ct1q)IjH>uq3p&N}<+xurZdf*ag+HQtfq>
z!WHhC4EtPluV~}XSa2Uouy0d7sQmbgwvE@zq->$a3<LcY;L!x@fMx5gxrhfXU@2BJ
zkLGQMOUjQT<TRNt*t#Sso|f;1b<i8?EOqbJ!=teKW)(H<mUivJ?Hnkj*moga>rsCB
zKr!`dR%9gHW+MqLgFK5Y52|gumnaDCAgZq>FeWiG;<JUi$PPJo%Jn@G7EF(b(b|V@
z=6tmHpMv^m7ZMFm>Q>QSo`;f$>^@&}L(cB19U3!l@O;y<eZeL3aau#@%JKR6+Zg1v
zvHA-?6MAhIw|Kn30^a&DLXA>qzA~-ot>RomtTR1UH)sWEo)=s0-q;?g8djQwKltRM
zwo!Dy|H5-m#uF-q&*Sk2Z`KUa{%9k7Ni@1Pf;HQn(VhLO7w<NST`_Tp*yS97mKV1a
zRjW^W?48+D^!V&6&E!J1%3Yc@G;LmsLckCutL1-4)k?Tk9hjo?iEd1-aJxm>oXLfz
z5)JBe4mw4b$fbLD?w?sU>^Z4p5MoxOOMW%`j;`+Mv_K&v*eDK7%#`#mORZMOnGQR5
zY@`N=TkNRNRr|#nfF+NNQcp7aR-P@Gz7tA_<ROX|znGQTFITgu3#Rb-HiK6kCQ}nH
z)E^r0p1mac?db)7{PY`H=g2}i5@VC+#cdA5I-V=f)<Cs<E?A8OZ{pxL^o`BrIFsf$
zlZ{&Ta}$RbL)G*-*&jcW3)=~ASI2mEC?xRlWb2B&;>s@Uc_^4NjST&#ib8};qd&`T
zOGdVdYQry=6?={nE>qSomyVR)BfaD~&bT0~*_Zir$uZZa*mT219~SRgoXW~+K-f|7
zVor-fxtmH?yGnOy*3qT5gb$2bXFu)^<KI=lG^c~ee6Z-s@2#Dr(=%M0=r9wZ=Z>UM
zUs3G&aD*U{Nb1QjKFL+lOonyH=h=-XeWy6-{o_gbVMcO%<vVnmWL%?@*d_+XyV-$C
zwF@mjZHv!uG^{Vxksc`T(HK_5rf8jJYc61C6lE~OC?oR|MJn>-c)C;V!da;$#bG)I
z>c%#SZj!fk4p|CBhR{t1B1`g)wDyp4vsjzPT_IM@Qj^{HZ)}(~Rv~4ZOun|yM!DtV
zFHZ7G|5I4MV(kwa8zM;-EHJkd>@R(`jOST=uXQ#1h`#TFx^V<8M{cI!al%WJ-HKTg
z=E+h|lKHgK(+^X!zMvBE^Jh`7EmoG#7E)=-EnS_wvHi@#U>hOZp7z>QR$3&4^qF@l
zaAdy>WcL{#T^`5R6T}kn{!3Vyie9Y}T=jk%JhV2Q^R@SW_Hqh(D2-`rMlR<<#|XOP
z1RB-4(Eug6k97uZMR+Fx!u^#NynA2?_*1<OF-Yc(A{eyYx-NwG4Yk?WODN#rag=6g
zvkVs0VmZ-aep1r%4d1ZJqH-R^w+SSA3~xRET6C|gn}nO<^zyO-KIJX`rJwKP;jre*
zsdI1lxSw%@a@Z{Vu!9YP^W?Lg;uyNTbl%jh2cYn3tsKnlmCG9-b{2%94a+@dQH_ta
zKlz{!`n)!h=D9tZ+tVcbt_N@^lP706yMvw7Rg8eUS#QB!(udxu(%5{m8eO41Ck%wp
zGLoDE$GFaFfz8@i`7%r7H#OVDI{tux#}BOHFOhf*BpAc`rfVI1+o)Y>_5Mkb(JKp;
zH+176DNgA18c%t5)izatpYrDM>&y6jk?j0CO*YVR@6LIp78+$r>vX9hdBN|MeQJkb
zc;?#IxAvb-uY^Dunr-(uuPa3$yEVZ{cg`AX3DprlR~z9{&8|RSK^;KeI+Pd&uXtOj
z8N%IMC5|!FG#<`)Ftp(vdT+;9u?tl80h60&IPo~r&f@LZXz5<)?_ZA|gNQ0Q{vKN;
zWb2&afjX_?p7@2Fkocq?nmLseSuZ_JtXF=3i9ERsis>0oeb>qi1-F*YUJ~^%!K~ht
z&sQAuWW)GQtG%~Z%W2p5w%;?EFnDxC%jK-WaxjDlx-aN}zmeC=yR*}%v)XyX^H7FB
zlg`?=V52$RlscEBK1s5U=m>IL`i9@LG}7Ehw3BX}S<1dsW|4_@vNswAYX_OgdtQZR
z=v0Wz!4hEkWGN+I^m{B69Xj2)jIO3FI2Q-=k*i{fu*cpO5O(I~7?A82I$jw&kA40y
zj2~<EZ6e5~_ghfzO&j%=$e<U<+WJD$fjyY-2%=u~7CO=gr2N;&hdW`xJomymy0^P0
zg^4*m(HRJA0kKQNH5a{E&TXD@qD^sx_{CigDd5EvAELg3m;zAyy7QRMoVtsdB^8<?
zdjX6`%MSmQwGcK@V$~G+3L)q>Kcw2{^ZE;Bhm(y)SHES7#97Tbvyv)xZU@EdRRvRq
zI@~8@{e_rIp$Tm$PshZ?W`~YF2(+qv&IkeU%8rd$HvO?@k#7Kwj}(k1P6U*1Tq<!q
z1rrDiMuDtxqjaqGq+^tjgSnRnCc&|&%2jxQ#zw}@+ABZvsN({h1H?GiAw@Vup9_B(
z6EFjAIxa}DuoHiAoarXmBWNXIvm*H+Cpf!0UQNzq*Y{m&(}p!cyBt2`P57VW2mQ#J
z!J-AIYevEN!EPD8d(gngAXpT}?-(w_UkgKVkMm)DKTl)})wkcF|7EEx_($0x&}C42
z{d(Rnd_f`@RH>)3p-7iSLjhu8*!iQHt{X)J))i&~hkY1Ygo;ef;Zh9Z#-h1CL(Vg&
zzC}V~)oYnA-tk=*f61&bLG}S>?xzc_43A8fj;0L15Hm5(e<~bt*4o5veRssl_X*!4
zt|$wdPplOvU(fW9t~(|YD0OD;x|4dv>$-!x$4G7u*~V>Ekc0*4R8HvG4PXWLj6^>~
z(Z-8KMfmm%swzW;ovi0lw*`+ooNIBg*C0x=djzHu(1UmISaMhxxP?bqF<#tjlCK@#
z>1XyK{uFS(OZMu5*^F^UYrh?a$sP=JB%<a$Tn5{bJU7KaXf!%rbRproxWH7kSG;!4
zr{{bPJpmEQ?uMJ@K9d0TWOEFi-xy!R5HWJF0Omn`%QtA?%TCI5GR@6b%)BOiHlK<9
z5uUEIR;^a@4CzIx3|Sw|=HwQqr0Y~%tY!Snp#DgpT<V$rzG);Ja;OH6{-SPYJf4at
zSG~|dWfcNF<&dl7CF-jEq6UswnOb*{ki14<h*=;OvE2_#HMq_IFkPVfwuvCjxp|G`
zfZsQEUsYlA0yb<Twu)EjexN9igilD~GnQ0dviAMG*px%<Wlzh68a*n$ievLA$8|~?
zQ>s7}$az*|SfKu+;S%b|EFG)p{<KY1Uw;Y=Ecq8JnhEe!M^zO(2<RC)I48*)k0Jn^
z>>7qb<P(*eJbu3Sh=t@RwqoFB%WaU0fsWb9g$z_$r`tzrHflfWHwAb4ZmC#Y7~13}
zlkPA@aW~-uQwbX;X=Z7G$Xrog6x*ZwfJik!V9MHQ$|fp|V-4Wm5?cV}hjV;y4ME6=
z%q&TTv)2!A;6meD)I42%S!$8<NjC-$l$&Q|fA}5Kyj|R!5>XgD@X-xbkjqAVa+*4`
zUTU#~{wDzNFgNkGX*H|fhQ+V+tDm<gCu9za{Fdmyc?EFQ0ZrK#8%U4%4r0ttV}NGr
zhJM1W5zQ5Ckd`@$I`B*E1UnO1jBCP9cH#>dC$~9a6%13xhnM$(^eC4if7^%UI@JxP
zQ<Z6-ATW%@A-5Gp;vj9>Zaa<oDew3N-gs1mk3yK%?cM-QzxeNmKvA>$u#~WyVK9*5
zykuCLQa3&Qg_M56hz2a(h$b@@Z;0PJ60o__@FN#4368|=XiKDK6KA>0pCW&DrF&R3
zm{eP1ILE7|GF%3eJc*#|#ba4GFNBOyW#zNxicH>?zl!2_tqr|z7*6&c4-~R`5=bG4
zjNp8W!DHvLC$qbwbI-U=uxyYn^HI9zeLy=}=+M4^kvVghvyQ!*W*a7<w5$tga@{>w
z_kMm%qykEdD>H+i-sPQ}L3<f9U?xrQ)J3iV-Ru^<8IW#P&oRZH#;j}2X{#gT$BV^=
z<g3QLN{mAr94x~0LrmrlsqcN_Cn&wKLvASg3SJ&)+0?+%9HYiP1%kw};4UE5g(5!W
zuJo07kLCCY^Q3&6;n$8Cy-4__Tv}t|FS2+UBa<qi4Vi3pNry_A94=>zCd_PS{w`hs
z@Z?T*7n!r}Frnk2A2;;P`-VB@z7<3hTz_)36KPiD-ZR<Ac(JI7R7U7SBC@r$nyhy0
zI<|J#!RclK&2%FQfMO>IX~GrB^l1EqNE%9R&z3@L#>$!3Eaz?y78JpX8Bf8&A5X!+
z4DpS(V|@9f5%0cn_~;Y*IrmDJ0l)|}2WNKdKM{*?mfL^<8JlLpW&*b|krPDsL#;|E
zgcyv=)5Klr{fMrTD26I^;|@G80vSY~Jgt`Hg+<AH(SDciAuEv9#iQ;+N6#N{K!`+o
zNvS(YB`ME!W4JI*E_2<A*Oqc2G4jCyp-3h=RkXyH3J=PKhKw6W3-zK}fyUv!o#K-?
zT~fiD5>z#)9Y)S6+wIn}Y+H*cPor>QuTA;IZ<9Yf=sWE?EYv*vp1IG$*n-9Y(f~Vp
zR59|J#oVplv=TBg%<m)IHXrv5D<t#VR~9zGcG4Y16NcPZ`FkA=yk4pU%E`pjMqw|h
zck*EY^aHuW#7wGu=sTVyFJHm9jw4*hEqfmxrPZR?xG;|KjRPxGvAnrvJREzAyl!Bv
zz&)Rm@l<7cKfJANH$wpZ*iW5q+UG;-P)U545=V4zvf~t86zWw<>wLYAgBSc6>knaM
z_zyp$-os$*_D&z0k09%bV#e}aeKQaFv!-%3A?s78&%ES&e^ECM8$I)@T)Rz-@l+wn
zG%px^O(S3DP1FL|)x2ql_V}5+rk#!3>ECPfU>d(e8r5^8i<?NW!u)aWp~r(<9zk5l
z^jJm>S50IhJ+m9Ma=U8npdZJiU1i{EU+Z>Y^~}Sps!~U3<N0zrIl8xbbc*e+<X`D%
z>xVX<@&o`a<Gxjg|AKip=LmX9FAMp)0&8t^gnxfM4mPX6Y7U!?8T8k0K&{e<Iu`%G
ze@BMFjO^22ijsypVuVU<z@M0e0WegXx^max=;YCj3Cjz6cVc$Pw_lhw37Bh?ffTC=
z!ETu1*S}puoPfhUcer)qOa6WAB!V(}7j;kDslTB4%i&S{Jq4PTw}EzMpcn~Y@GY6t
z!Htg-k2Zk|q=Ll^ZL#;B4*_wBNb8rxe;R4PAOY1TQf<+YarzDB|9J^TkYlXDuGE4*
z|L4~axI`o_NvRA7{>NZp_zeayfd4+`|DlUEv%amiow`7L@`3{q{JSD4_ebhQy}SPl
DG#)q&

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-tau.png b/docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-tau.png
new file mode 100644
index 0000000000000000000000000000000000000000..5c17b348db12d80db435e4d1fb5e21ef35581120
GIT binary patch
literal 47009
zcmeFaby!qw`!)&)NJuH6AY~vaA`(g}CW!RVrP3gXbdOS^2r8i<VbI-3ODaf6Hv-b#
z4f|U7>ift%zx~&@zrFYI`G?0b&J1ha>y9hV^SrKkEPLx32@x$378VxCjq8`?v9NGc
zv9R{J6X3%;X=b0-;f*6k5)!f|64xZmP0cM8EOhVaUDLa#XK8dt{@NuhEJ0siB~8Pl
zN|d7c4-Af<s8cXK=V28qMskwmCF<PU%skhg9{zhlH~2e`jow|Ju6y;RPLMZ$>#fWc
zStp+HP|EV+95*J`C9*EjOzuFUlby{>tfHB#k7P%mL{GdH;(cwDm)5H{?X(a!6Ye<u
z(O?E|^kckA&~r*`>X2kEnzI8By!^aHIQ-qcB5Cm7D$CJxXvW>d;g=z{%~J1^D0S5C
zXzAIGNXofNPsMbf@pi^=EU~vA_H2w&jSaTL`g2CHaW#hJ+oL%=;^o{9x&3$A${vWi
z_!{Z|;2dc3##2%-_6{cZ`&zMj^}q&CYO8jnwDCmz=+L{{q1HW-X%;iO+Yx4;3_TXi
z%3|tYM~J%5=xS_m@SPm#mn)Zwe-UwYlVHHa%sKn$_pYGO=M-Z2{@uMHDbW-&mo*Z*
z_Qw}Rd`QajVRjjD=U?C);PS2)7^znH)SeLkvf|AroAMk!wfT7=+>*z6>LGJgmK$aY
z;qxOy<9s1Cj`o)~*9+{gFFRXCoSzzV8eMP?;@ZC%6&&!0s{E0+>)Wdrmq!M+)4C<e
z_Z=~&9lGIq`o()vb?!k;v*^)!sRJ|u&5Tmi6YFUC$eqg>Pb}S1RPht!p8Cpka(x`1
zSd*)$qj8FG^59QEyB3jT5aO}EI^$JHB(1WK;x4N=ejjzaVca89{Iirvlt+BD-|ExQ
zjg$z_U&wr2`>CijVEglD?MG|5b(!pUy+j{Ur_*jeOQdenn@jM}xN}I%g;M}WjO7T^
z!H!4uT4!%8Kfiw#`<T6tu=MiFqv?n4&aAgfnqJWo7HHinc*-ju9eSnbd}e;6&JHj8
zO$~*%L{8Fd&*L?08+e+9o0ZyW8e^OSa@hk%!t#{s*0k3qHqS0|Y#M*~?)pP{F2}M;
z?Q4Ge&9X^Tl?S#;v}X8&)U)4SS3ItuU${T_Lz(QVOVF3sBPolT!m8${Nv|db<E<O7
zp^n`0y_~rCXwfT^v+$HN>sDo9t#h?=t#h68CrzB^LN#H7Mg6BO7$0a?E*me<s`WWi
z+aJy`7z;8<lw>|`O2t5JS<2rpr+aq()B%D!m*m>fK}i!E)Dhg_BJ`wp7>8pvG>T56
zupe)v;VepTD=8_CSp=}rqs*lRi!L5|dP3sv(*S<Om0ab8x5olyhUAUN6@(NH#icc0
zB=gFMwuv;u+1R9SiMy9BKi+t!*BM8(>B7AUl3FTTsm7PLX;b*wgF?nfy<-o1e!lt4
zP{FIyOpOk2rgzbAG4yOc*&+4Q*WPR#C@B(~o7^}(@GKK6=%^dI>iE6G_0{`(@dOv#
zJJY*9(LcZLMEt#ZWpcTA{j8nKR`XbQ`Gdp}F1IWJp|Ik<8LL)N@tp)L17BC{$OtE;
zeHz3wU@HXmRBq@?OJlLaYXU56FC#2mc!drBXyG3g7EUD2uXk`$BlrD!z0V!_ke%S!
zJ1i`5tQ(gvD%xRB_7DafwGP`qyK~~D5VJcLA^CNk^NQrM;>zw7`|vt6FWX$Pt+4xk
zM~QUqZ8Cw+L;T+45}(WW=g{fL61$mLBVX+wKt~d=2K!qOurTlcc&h5_x_M65^o_H;
zBAR@fJJT|S0<Ytf<6XD5BV1QLj0h>U(2KZ+KYWaZje|=-&Lr;k`-S*LJWs03(|Q#D
zeDmk0tb_!q!W4gg9r=Wai>ZWtahCPj?~D9;A?`MG{vX%=&)c$w*x0DIUf1sLxix%@
z!&#3pA}sVJVuDm65n-R>7}p}#u_ygw_`A2DR1<eo8ts=mig7JBteOY^eb!&YjBLQc
zNxVenH-PzUxGLfMFo)?uO|C=bU`Kfk<62B5^GE+><}%%|5*o_h3+|n5ENtwckzd33
z@5RL5;^C@zFfdeLt^zE<e#~M1Z&(5=>bwgTRFdC4XtY=rMYgDTP1{q?=v4=V)CO^?
z9&AliRakiCaSkuL>t4QmKQYbe0~2j2G8u>BlL-zmiVB<NPh|%V31-;M4K%d0D8W}3
zaW7oBP}&@)Am!m$7%kuu8lRhy!E2gP%g>iU&80@!nyjh5`?dH!Aug3vCb%7Q;jP%*
zt2*W06uJ{hImz~u6*SrhCR&pc?7v=ez-dWoTU!|Mnx39syzq8_7XAGAHP0_EPqhgL
z4LHWk&lvI1e=+=YMCHguV^543XJUjcDVnTy5@UybzAo+-Nbx}*bI#`}$aRJW<CQSh
z6#Wzz_wQ}~PTK!hcUmOxdz*6-k^VSOsJVz}G%POKjmO1EE+uK^^TQ!~>z{QmIrjcw
zrN?_e(F+CB)+t@xEDWdqJQ!SH@{Py(TRt9tCU~sdai;K#gogc4V1KK+ISKsFcHa5h
z9@{IM;wD8fL6zZr76*)j>El`Uhqku1UP3bkyhj{84*p>?(VSrZRE>=jWN|b#H9g{E
zzTF1<(qwpOdCxFz5aqg``J$_^sDl@MD~xtPBuf6(a9x=Cg#~sGwU+DuHK@pepQnWL
z8kaCtO8E87dETRT-_afkCJK1;j+1G+6+Tp<YC}AK9_0oU&S_@Tg#K35Ki?O36L$cI
zu~>PtVy|mJXCEZj$!f8f;N0{0;!Gl7VI1y6-^RF>I=IzN<6P^Yy;jLT+zpGPkVk=M
z@3kV6z`&UCbzhMB^9c#augie7>U|aWYVWn+S#T1wY43$$Shx?Tm`Wtgzbf9`do4o_
z7_mA@PTil21A}QiM6MGhpwG_o=cd3V$P5HOm*lB#K)BZ_kHJGW((zIq`eUKtTAh;O
zZY`*?qzsIyT!e=l^L_1uaV@ixu=)Nj%im@BW6S(KE&tPLY3M)|wsd&nUVV17qhVX3
z@PYBt@(2>XH5faOr_}FL6}w&TO>TD6h}hKlfRK=ogp7<#C%uC6=;&xHt6d?A^+os*
zQK!|bOI2+NY|iN3ls0K!>-9p@6b94rtRt?}%6`_4k27zXC-ezb@GOP1ZKd^ePh>eH
zHqGeW5t+UBo~Jmw({cXCmm5pAL)C#dTDr8;?o>S^iSxba>Rf$$D@gs`>npsStdS+u
z>ZWof@d3@=Zfz#V9V&Wl2g)B;oR+n=CcYVRE^gX+(%%z4p&e9|2~;=BQW6R6;%CO5
z2{Ya=<)$>D8%KtQ*iW>T+FS`9@YLa*_{QDmb57#kIr;F0wxm%Kho8|$<2&j>*LNK4
zO?{2l@2E~+w7*$XQ<=^Ae0ycSowjMQD!u#B{Hvf~Uo+KXTQfsxe9>cmdLQ=>@o;g+
z2xb><$WBX9_u;9ivS#0)TNtUAiH`v<DjDCq?c-|@5XFzbwG{V!smm-!(`ZGo4nkn_
z4{RIM;!Il6=~LNVt%~VZV~xK9-~l~$g_v7hRMSf#y{l&N19p$g_`6IMGK{~xIH^0;
znZ@U)dR>NTyO9UNZ25lfqSf2&sk#ylVY2sq#S1b`cVl4Fxq%AZo;>T_??$dK<yrJ!
zavi;8P9P93*n2v9g-U>}$d-L4JX>x*uU(1nowMT2xO@GHo45;jhfUKg3?IPcj|LyK
z@3Qn>dyY%+1N=sHJ*wu9L%8b@jt~%}{=LoLN&CCg{+^q^#^$f-{6E0ogst%<4YAIa
zAChsmUl{gaQ;2piT<N7F!t}IZ;At;>8t-uGNZ0FcR*)f)3gRR*cN(PIe^jT(s!6(E
zy>A^Q;zg8VUu#ktBepe`x15HN05ai0YIf^=rf!+1#q`IKpwa{fvbhEcG7k@8n%1l?
zDegkWwq<^d^z6g2?+hyEpDIG){ff^7mAf&WzrV$yV6x%z-rSFAmIm8EM{o0{x52Hg
zD%Z_xs}<ZXJh4G-N$SdZ1}OUxJL)z-`qZ_$CszO}FtZ7x+8YO;10ab%;;RM0WbWE<
zXxU^(Mp?X40)CHiLBC_c=9yb}DxVVWXRy<OF!QjLqs@uE$>fLA5Mb!wM!0UTHEt!l
zEQIzcs?nXdyxZ=wF=bL09iKRPXLzoPF(t9*j-QCTG8GB$o*{_m6XotlJvz=<Ul4Ns
zfqpD$ivHJE6x$ocJ3>b-O0cbSw!OqQ-=@nPNWHn|FR{=zG#2<M{Cc<K<Yak{GA}Z+
zYYg^x1^e9(mBz}wH&$a9)pKhRu%OXtAy>o5XcZv`|5Q@-eSEfzj63Wx6*t|{Ny_@+
z&gLU3CasBtGGbd+cEhg?N<Z9qS$Z#jikjut8^QLncE7zjyMF*=)1Uf%3{aZZElRrM
z&2c2zl%>Sn1>~q?4OL@<$@Vm6{5u)?wM02RHtM6()5%6+E@%54dl$lIpV=0aIj0-n
z<=h*%C9xp0dU^jrt%{6<_0V{3MvY<@XSPUo*JHXf&NQ4V<fn7LKeTAmwDoAZCB${}
z_1UNumaWq}KTe4eJ{bNwW}Bb;X?1R}?3&p6H*7QY0N)@kbyR@-*|_5_XOIB?`3?1?
z=Ic)PmPzf@<6Ng2gQYd~=^t(O2H(kfa*bmt!@+dHuCNy{qJB@hd$F6?tbF;S857C*
zo`cVXk{8Fy*qXIPqK6y?C#)AozYSgKJ46-B&_#HNWpyXZ_~_uqB1dxWF;3O=598kk
z>MMDsCO2zRoP;J*C01v>#Ew?_(ixaO5wqGtwQnq4@b44YFl)w_raH`TI;b8nGS?dy
z7iK#99VZbVJEObsLWx?wjal5->WOE;2d499ilT;M!X4O!H)qRqm&RM7hBju3#ojp0
zf9db)+qUU3Kg#GZ`nIv-&c3m-A75UUof2F7f;LrqWmft0h{7wibk-C2Dsv=Waj3|a
z%8%K$Q-wc*`&;RD617u!3LI+Gdk^xjRD}oNx~?}xc!ru6osW89VqEKBYC@aSK#6^4
zU3u9qu$-=U)^oN|ihFD6)8&!d1S{Vkk$F-azj~r`pPsoS(>BGY<`5hg>b7SISP8A~
z<D^=!QJg>EJEy}msdjKsBDE1>VsC&l?l0Py<`CK3T&VXlhjY^pW+}1vtMINdGBiR@
zq!bsOEKLx+p401KGUlhy8=JkTfEqbSl?!X2cC5f>(@%8WaJ6T~MR{$|?w!OtQ-Em{
zbp{_acT&D1=n3aHQ)_=vFH(k5H&e=E+w3ripm*?6POG!~_<e08f^iw%knHH&=v^rt
z^b^xk<(JsCE^3;QPJN$IqXq?(7hDXVJjOx~5GJZ+#oJxsla)(YioqtzKgi_&%8fUb
z=$y`}FsnwlgvV{d5f0;V{rebK#m7?W=R!v8*eYP9M8ETAcU#t^!UiQV+*lp+nk&*^
zla6cYiZI(8if&nN){fXv(=LjTD6sm>x8r;xDkE7qddTNjN>}S})3I;ow3ZE{*pCw)
zY`sL0D?``k44l}-Mr!V;W!E{F)=aI7F5(QONe_J#q`Wk@*5gR~S;WF|rs?UB^+c=3
zl_%6KM=wN0@bP)5w_h)rus;6c2O}%wmdDw)Ij4KPEVa{DPnT3*O1oh;S+GzSwQ;9H
z0X?|*<E2ci(wDEVc_lV!M=4WwWQ-)ISVFftDvN`-G;&jmYdiI~Y`V;p$0r6Vp1fT1
zNHN1VAgA|@$m<lc^DgeTwYDaC$CpoegK7Iy)aDbd)|9sSEvA|+-xZUB)qy89Pdv7_
z2N`z+ot6@`lj7!p8@V|!qUv-fua>{7xw!}uPuU_@gIXfu8-(TLxtpKF=a%x~zO%%U
z>M2h8P;u}Zk8x<PNgiZ8cQB=4wK+yqNx^l}ZzX{5#EsL6#aef+cWn4wxM34t`{db*
z@Ut1c_?T58Gh=4d+DhiQ%U!;igQbZlLYwOr$C@~<g$vo<)9XF3*ZV~(0i^fX@2>Tw
zt_|*Qk8P#bFzoINFnKyM?sogyR@oly3ybHTND!qniQMSIAQxq=uvbdhKNHlx<~`*!
z_e49fd*-Cexm7@*d(5p78><(J?)5KBe{=)`amB%@gTG{&`Zy9EILjWiSfjn2-dD{4
zK(MTHUqaz>XF{K|>0Wn4eq9<^MSqGX5yQOTx+M9OHSS*-od`#Xh=}yVQ6rmkd;_bO
zWAT~UOL1A?=cfM$Yh*|TJ{x^?US!5zCMv`r!kHgAs-Xwc2Cnsm+Yi6)Un$;k*4yxN
zS$xmzqWBSmnr$$~nMNx`Qiy<kD%(dUOn&d<P>{TmZLhrna#U%aiLQ^P)|}3F&oXKA
zwCngjmD9(ty%-~=#4fhney41EV@CVeQT*qqPQs`@U0r~UW2cg3thD+?#zj7!)8$xy
zTk-aiH0r!@bF4+DQH+N`@mg)D#>_>G$>bo&bx+lKe^??eqwWHShF!mhE;x^#yVi$T
zZW8Q9F|ZRk3pzje@85&uqEQFI$v%5;KmX;F*DBwuF{Km-rQFH%n$r29nzAMtF?wWP
z&NwWN&ZY+!a+(%zEk8196vfyxCm_(kW6ZY8Zqag_Qd(!Xi>t<9ihmsYey~(Sg-;SA
z((wn(nuFuU)Lusk2*)vSl^#m(_udd2M-|ZQ^cT&fdpEe7_-P35{@&()-K2>dph%3o
zu+UC;s1z-FYHqDJGZ)X^b8e26)9uc^e<UaV{2v?APyn`bzHja}%kfkQjziSAKmK?@
zUMwHF+~>MuF%}i<nUvFeA*-?FPv!)V!h%ruBL1Z#xDLct*49fdtY3y8IOaxn^NrSI
z!$0o9>IpXL@L(hn4jrMDt*!Z@N<5kH+K-oULCN?2{GuCt5$jPT?jP?W;Uc=0ICp>R
zuvsFrEGv2+sUQ<VvvDX&hjI+BPo$95i*8H~qa5G-84xxD?GgW^rV=;g)EVdX#o1_q
z!Z>@p=KDX+KNu!R9iL5S`(ryUa>0%-a!$k`e8>&cVO(GKeV@hL*v_D|SYJ-VprXVh
zzre<3A#*FkAaDp4feRe(s#3#zN*1_Uh%gqZ491~r$bli4Gn(wN2=D+QoIvi(V+sC@
zR{j}Vi}%40{yxy(6~g_$-Evm^Ypa+$?(=1x>F$sHc@xRRH9=f89}3?Z-vuyT_v<U3
zsIn;teETIB*BfN%FT5TkNPTgHeGemzjT3svZhLK{OxKH^8kMM$_EGwU{bqjO*1g=J
zAoUk<wyZ?fg#EItLim52L2@Slgpx4Z-j0L)N|*C?wl~TYWJCfJl#P9fQ|0M+miAJ3
zbVx*O{Ju)wC9-9`X;cL~tK7wny+_W2@|Cc?CG*6%W^<nP#EX1$<r&c*g`6eT>M9$1
zj<mP}ZqZt;x#zQI&!Ug!hd&a9;GzF1Z(XcGFt8CP-OU7kT)+VRuS-D!+iDkwVFem&
z-~?vVS<+)73t*chO$rqMG^;QeLm@E8T_hh#_dWz}BQWJrd|L4s_9-(AY{QhNh8Ko$
zFhWRVTJDy;(}vr53rJ)eiR7|9kB^1Ugox9~H`o~CT4q3B80Ry$>FzaK;%e}anUzvm
z7}v@M+B>Vqf8xxZH5O;$Lf2(hE1B`)u}FtBe=Tcso*b$PZY%VopQkv?bNlk=iM5Oh
zYIXPU@Ng%hwsC(#D=RBgtHim%DlcHd?k(o-Z>0LXzH|X%Eh`>=xiy3l>#DDoK6<Z#
z%xrw=f^MmMLr;&$DF2j-F#7SKQ|miRJ6=&O+uS}|UnZPY=fb~C$;mkthHM{eFivR$
zJjT8dCGrtZMd`5q2K8YP2dh3hN>elACF-Z#O_rWyj12R%ey9yReudR?Is{P}5Rr8;
zM?<C>7M8Tc&$9jI>x6;cOxkCMjtM@KZ^JpkOA}dZBD79;b7c3E&Q-LT#EY4z868cZ
zd@{G05ZyMR&82ofIjeUqYk+_%R)zzW$STXX6`XF|62C2wUg1xWs9h|2cWd*p!ALpV
zvy9?(vM*<b9sKRCUfU#ASju!*8b^HfxYyTH?6NL}$r>NeHt}8n`B!M{=jfVeVy6`B
z21S6a^sRfg=+LwHDtg&)seS-P8>gz+maktMGG6gx<c@2)i*}H(<U{&d_)es?wY8bU
zm3MxXJK744g+YBQLl+?-lgPT#b#v-P!D?lV;sNww9lfnhwjG_PgGajrqPWQG8@m+U
zT8zyvH8MnfQey1W35L|&qHcaE?_N*dCn}|e!U4Kw$~yKZe9J|bk3Ta@!1KX#4*8*m
zI&i8+?8)Fx%C`3`%`|GGm=#||wV1nR>2chgG!djugFB4dksCD8r~_|)4#sT$ao6`p
z_Xc{IB6~14Chj~H=u{>WtoLqAEHpj@x+FNAVSmO25V$gF!zn8&QrjbsLC&8noQ%J>
z`8#QUciP``^Z&fDiOl%4MVzy-zBEzBxIODNJNn+w^Zm06@fJ&L-n&+dQu@1r*oJEE
zNVsDu1TU6sW}B;XQ8vm+YE*go`Mg4v&MQ5^s_L}HjOf*)1#^{jWdQC6jV;_~#k=Om
zK;OGLEa2q~2qP=mgvzdij*P>RCQ0qx*<SJLgAY%-aXXJm^)qhwGt!m{ZcJuKY_b<)
zYv)-Ry%M~CTf(gK_3kh-(@V12wCHYK82f;lepf^hOCdP;l(#geXelYAN?t>7reLA8
zMOmAkRCu<8vPPMA*W<^kN5%z{SMo+NIL&@)Y<qE1YW#H33MoB3J)lU*jsWR9nBV?+
zee&3Y-q=Bp#f1np9@O%2ed__&Z&5}fka=0#>f!DP4qT3m!)_HPN;C*X;6&5yDzWV}
z{SKU;v4t!LIrjuFh2bf5Acu1HO<Uzem2YRAzftV^goX%l1#afalw60_s?_)u?YGq{
z-LWzBUAtis+J<^`a`FHOY@{DDZO?}fiDbPP%{<0SbC}Oq(OaACX7~cz(-*Xe+hO8H
zcjF}5sVx8`S=RmWfq?pk1eu7U>I%a303+2G%SmR%Qgq*U=e5{e`61of5I(+q=}`Wp
z-@|OkDFHHu=9bV@W|08<=T3!N=TCkO6-Z%fVLh&dwk_W1>8GEuyLNDB7smZ^D&pol
z)RmvZOvJM!P94~a?Om(0^~@>Wn%Wn?25b)597h713wzDeBfROxw}ke_VAwd)2R|<{
zs32@iSp=ux;b2gBdr6)Cc=t?xrx%7_jMTyTh_8~CqP<qFTsr#gz1N4XxgvVN^<8s;
zawx^#fAq|WDl2kKPwRwdneU#ev3xpz@}LXgsU9b#f&v)hoy33vy0j*$5)<L#AUK&?
z#y)|6ffv+MjkWq!zH~I%eJ&ekfMh~X8w+7x6c^W$EcM+d0I9NW_;6*nznfnL%#ngr
zboG06ItEG!4<pC4B1OQ15jlfbzpQh;uB1}~9jxVGr1H^H&u9s7SAm?@RhZ#KrwX#s
z05$KeT+F+Lk%^xn#||bZz5Ip$U#1%3T|VDlH0Cw&k2N$)5eO)Ke~3q1`cg*p$ZiQh
zvL;<?l6p>DE&gJ=P~Tb|wwb|k%vl&9qC8eIJ~l>(kXElm9Bp!!TGb_hl4QA^rg>;@
zXzeE6i6c#xI`A6h#pq&}yVC2}6r?DE`K{A9(#88*<*w16EU4Bq#Dw__Nzl>ZXzQgG
zWv~3{Jd@*JfL3u2F5Vo{ZTxTp;%h?P_SEPRQ@WN>jB8_|-%=PFyII%rwKXrMzkh!m
zL7NXS?yOdoz2doJ+^YIdkmBbSG?YO{<EsFGn!+(YmzaQXR;xcsh#q4uGFpuI+*D_Q
zUQLk4sR*a^vjqDYol<Ax-|1r3C`58LWMWOgQ861G8)Ivu=v|B!P#Uhq#4X8L;hG0a
z`^&t7O<0{>w<ff;GrHM{WrXK0zY;LldUaM$vK<JxiD?wfx=O0fMLXb-OGEffN<n>g
z2;oSt`qC=ds@q^Z*aub^f{7QsP3)Xnu7z%`B?o7D%?TNirRLPTUtY{s3KXR!W0`<*
zZ8y>0SrFSfbX%yUsXgsZ*;_9L+FL-Nl8;F5-DzkPwPEK;_oirRj<7(*O6S#nubK?5
zIp;D@lK$=o^;{mxOE#?dS~IRY8wiUmK$LP@Jpq+OB+z4}AUG7DP>S%~<_p;v(eH_q
zFkB9;ORL;F^W0R43CVfv<r&Y^bbW%iz?N}zgRil+H2|)!|7FDHT|00qQX5xeiZ}9$
ziBE;w-4+B;uk7M}dP)#Zs;aGgKJFL5;0nCTgy=K+e?aIzXDN>099M~tIFJA9GrE_)
zj3cIOB*N8&{a$DiuNKkT%+B^q|8kdk-z2|t(k{jgeht-m0$)jcy)jr)d&Z$HUoxW6
zuI~JZ`$C-%H`u@EOzDDRl(0e#20L%4glsR6&OgYVEhXm2HVC)ZNBoy$@Mk54n|E1U
z-rX)poN2$~M^X&DJ-*Tsf~S)Rbjx-K;OmCwecS?mQ2o#{!klnw738KzVC+kA4#IO&
zEYo`Bb}2EdV-7&a$&Vc@wX6I#6s*ngOl*6Vd|&cdQ%ta^wRsI!4(H6-?=yo{evGr(
zzNV|dg#=IAE`|1OMmOFRb~N6Y%Hej#(H;8aqJJBUnw$A)(=(xIl{~qPVVe$<l)_Sk
z^dv7!Sej__LU_@F8Rwa<?%}-^2e<?$I2xL<CK5C1;x$eS00RSr@8(<FBQraT9rZ5R
zy3KYd#a;#BN`zliR#qO8P2-N(*&He3%ITI@cC+5uT2b!wHqauWYh59<^g3#Jovp-2
zsA6`aaK_oZctep?2jj%~dy6aVOSFrK1)k;<MSN|rDrw39Nt>VW{AY_&!c*>Y@2|E@
zHhlXgm(lIWjITQvJvlVilHy`*Z<EnC809`>^ywj9KX6RM;2ss$nENp7C>Jw4QNl^w
zo$kknHRxSd`=fZ%2J{|1dX4ZNskbY9s+0E-zRT`)m_SI7#*?q^SJ}H3z1<LBPB>bQ
z!LBi3BfBCjEKC;)qzKQL6_wlRw<sxVQbB1%-kL9ds{Z!oT42)cP=)B5ahxlBhjJX(
z$6};H%gGbN*}|7v0AYyKTo-y|D(>6<Ggu>UPb~2yp}VxnzE0V9C}85LYdP<)iZstl
z;8sut*!CPZ_YtP7=QFld2Xb&wBBT#bH{EAS^LyH)sqyvSX+IZfZ;dM>2X}wE5dvJT
z!gBFPnmtMX1_7Xtyku7!MLlIV2q7%*x9f+RSsE$7ucc`>PDl#TzD%b`E@o8@$+~jU
zd|(CZhiB{q_1{%9^gYO&M<uK$Te_w1V72vaqz|J6XW#oB;bqC$(FVWLg{Em7PzPB>
zT9m_KSJ%$~l1N=V+58$EzUeO;jIU{&P^fiEI^@Ta+LGz$#_4_~x{zAs$O|Ro&O3=k
z&os84T-!W%?LWRq)(Jh%oAN|^YM4U<Y~JeUYoZ6YSF6PCxg0n7Wd(jJUcbJ3Jf0DQ
zPp6ZlRe0fveCEZU&h`gqDpb9_@jGKxo1$G@G#}R}VXJuU*iU}3^Czj8A{tm2z9Fq-
zt}deOe~<0aPdC!;F21z05vmf_Es@k3@dID{)6bBX1Q*}teH6c&p$*NT$ir#{*zWe(
zeKRI)$w{WX_qubdH^ucy&;RfzPeC(&rkrpD-UC0~NH<%x$jM&U*l6X_z}Dw<m+gQQ
zb3X&_Aaj=P={mL*SAavAdv{VCe1cI`yh1nEh|bit<%G8)M82m(U1@U66lacRrlDN!
z_WFcL$WsHf!PbMKVXPykGv(W<`}XUPG3*@91DGoyb8sUT?b0seS*GwgT5XEmODO@R
zw_d>ZBQoov98~Fnv&;HgxJ65!hjJ{eTq#rCeq-Kj%d`>fDns`<d+A;i9P_j)2I7|B
zx)yaO*RGkIK9?;s`BzI2)Fw5{(@QEbbbr0Ml)Z$n0s+2^xAwj;Eg)*p_td<nx(_9i
zm&vjp!@oIKD<7n=w*StiOf~m+)oT8jq}3bxh5Hf@54Rl}6^NN%^c$yX9(dZ1Ai9-)
zjM{G=@+*8PuwURON*&mEp6FU<K0M2ieDL%7PTOjgyQLe|j_cC?cf~r&#eK;GGB&Lf
zd>wBS*a}uAT6RWU`{x{z>_g7!z1?(FW7uyF(L(qADi6eN__+jj-W=mLZ82A=ah>xK
z3jobcQ@g{y#O==P@Vx=6EG~Nc(2HY=9r%2cfJp~xRTkTjEDk5PMBJtntM0^f1bl$|
zT)b)p9+}|y)u*wFy#9XgwbJk2s2y8Rv8&)<-w0*i%VTHyAlXB=Ua;nP5wmImgP0b7
z4rl{>ryVo4mb<*_orZbeW@N<m&^A9f-n*j*TrLS%yz53gecQ`9TfGtEpY}5*9yLM2
zCQu{=>g`osVWCkFE;T+_8ww`t<k8Y!nb6+3&<)IZzZe7r5}Q9fz%Pn++YS2mhnOVC
zb-P(M>^CfZHm87HgWW0v@|ZIqGZIZe#AzH8D?PS-)dm0No=!Lt>I&Z(bK6HsVDRSt
zRlCJ+*UCUvq)8hrG@bjt?|V$_zEvb1kyksAzMl7K3-X&Xgj=s(t~&bQ=D&#(0&!oE
zq{i>pg-TV9F#9aP^@A9m8R%*{j77DCp5S5RMhg#kA7{_nCF|i<Nu0q3;UG`3sAh{%
z6zvoYcZQVGF;IB}Bgi&nLkAJ04tN9EaQvw2=3L;XwfW#;jn^_p!;mIhc&}vmAZK_L
z*m`;h0N#+-bLJ!Y2_B6G`7PnO^6D7MTd&SAI}IF)xE}Y!7y~hHKG|><8#QUYB=UaX
zq}@uNtE=YWEA4`w%E20zH|#oUv{38@P$IBfoo5hcjS}2iY$~&E)1(RF*49!Ju>A2E
zQ4xBAL#KIb#fK44T2dIsVVQtgB`)*a%90`zS(H6(XfJR9suIpPE>BuC%U$!Z^HiFN
zgV^|}O}lPC!ix;`Y8U3DVQ>qEROo0IBHdPR@Kek#P;!{vya-yN;3ej9;~5ahkj|N$
zQ4r1_2W2FebA^nu(_~szDaZy(GmV-`A<tC&_|@W!Sr<mC=l?=n3Qq-yQ`3_LK99)j
z6^%jI(=Foi`j?6R9du5y&bCcVsYR{?Yuczva}UHE3c>geZX>nfYpe-qzd<+>W~)_e
zSg@RxlBAlw_bdUw6E2<}Ol|{}j8Rg-o``;Iw+?dP3HRxq<MG0{AS-i3#8Nd+iZJuI
z6YNMt(;yKOIPv;dF90$Uv7NR0iFXqa<M)ghLk#RZz4E%=DY||Mg~59ed_W=y6C^P=
zkk-3G{V44<uYrEc8yn0g3Sn(HyHeug3G?2MM1}&rc6Vx=aC8qv(2+{-%?erF#TDv7
zz@!5jJYARpQY1Vrg_xGnX<mmqz7r!cpuC18nPe`rk5XN>)-M{BXCy%mK?Q;H`Pi)~
z47<hT?}>y~AnW_^_Jbs5ju=gt*({P85<Ob-d)a=!_y{DB<@**462?uO8i*TC%B=sm
z-G6Lj|4?y-8Fbc1Eo)dONZl_8gCeH?+cicmXE=>jWwZTdNa4+)Y=|5h+mMJ~YiGU9
zOSkNYMQFyN;NFao$r@dXry{mJ=T}B}>WwT>68^({_)8Dh^%z&8h|M3T%3!0O^Vc-R
z&;T*{`-|(|(gX?MHs$t`_72}skQDACO?kI~s5(#QjlN5rDcbOasF<JGAqsQ30W>1V
zZUsW|2t0tF#ZaI!>4#zIdl^XbRll=YuM1_e|0dGGLZ8D6aUs+#s8>1=jtDl7w}1Oa
zp|upR?4fDXadKCrmXiAkyr1MsmCwp&hwte9NJ415yfJAYMyIt9(iTwnBVJV(z1L%j
zGu=Q(?^mg!x7F}AkWrE`ckJpa1rv4%Q2~y7354u`vgmLf2pV?P7oeUCY^|azDIO7-
zAw=8?66f)ERv|qWMrD}rhr~Af$;u#-+~~^mpBQAsik69B=mp2)RaY0i^NKO`TW6Mu
zl)=E&9l<2EEOw+^`CLS`7!q|}J9_`}E089#pG<fOqm*FTRbzuQIG(6g=<xa3>k72X
zMpvKyx0|}1xdT2NmkMxLqKu%zte<gbj<GC&O+lVrXcHcnxOlxKr4%4c-V~vtJ~}ZM
zr??&}4iv)Pef6X0yH*6YfW&np$2HWxi}u8s!|3R3{e`U=S24&{MFR7#Jv~fuFK;Z`
z0@7Gzh6xjW#5vPwJKf*a)Lj@f1|B-977^MOZB3bb`MGWmI#1oBhfT(R&<>PPWgq}=
z{n4vo-uzrlj9elY0k;$qh}PPlUt>4|QF8XBR+Nk!VwSuauOBOqzgC+Kg|~SnkH|z-
zh?uHft?n&tQ%AYZ`E3-&%($%N7cxrTUIW$e0nn?;l!9{gW$X%|Gq=W8nosyl#e80<
z7ZWSk=$V-|;7B0+!QIg8r~`XKk56P1e3bC_*?5O4yVv~)7>b|*_JUE`o$Oa2tcV=L
z5$JPzAmy%1F%N_#;T*Rm+fAs5Gl$Y{HE!)9;mz;+#jc|@ePL-2pWPLmT2_?%(>J3_
zOFf9yul6GX#qqCcJ6nA_W=A$q<5D#TBtvaJ7-6G?edj@5EaxE-Zfy-Sq8v04)?r?}
zB^i7uXyr*Tx#(hp1So?cojSvWR>X<GwjQphi?Td&Vg>yyb_z_Nf3$eQ)*Bf;%DlWh
zc{Xct>q@bMQD>tRO<be3inj<=#T0R*^WD#AO)EPOF58V=f3IMAT<exf_&}6+iqjCs
zPg|`ci}M~J=Vlj)LWKftMGnmn9&nUrOjVk;>*U*Jm$t>9=h<W<()><YrrzS1_3_8Y
z4wXBeyEiUhaICFc*=*Z&O6pXxH6k9L3p&qsGimva%qI=seTNLy+^A+0ltk!3gx>Px
zQAi))5M@a1BVpyhDJsMpHF%2`7#SACXc09*07O5q6|AP`<Lck;y0i6iw2r6h=%Im7
zk=^{{z)c;<ugNkD>KWQArF+@8`)GIk%mT-VwO23Mt<`92txlJ7*bKz9<WyUMF!7vK
z#W&dtn{BRJ%CC40>SP%j(+Cb+k2G@aoZf0153}hs631J~lG~KJwIuz3hD)uvZMgMU
z#`j;Ja)JrLRl>QHL@H!hjI4a#wj)AcBmIP=?D(&k5hwX|g-9dpR>=AfBZ0Qu1|-Q!
zuNf!5A}Yp#xac7x2qcG8G4o^GgH*^|6~?C&Znx$XDXovrT|LTOFk2eZqYVbSjZ+=N
z5kUn!;)x(e)u4D1yiJcpTSE>8MRUI9?i?uv>FHhI)Q=3PYO!E?olLwCS{>(M-Uz~%
zEYk4=S|Gab|M2n_(NedQh|As0HaPgRX@1U$jkn&M2iVCti0dCd=0Bmq2R6hZL$PZ=
z0#rQDAAR?v6akSD3phKOQSS*Qsv%25NTG&j4R^6Hgf|R;Dc>T1Mh8PmQBsHcfXPq}
z`WIC114l#0WTV*pb3b=52`3a?iTcy^YDz~$S3Wkjc~cyh492O0`0^<&AG<Kbce=d=
z_PQ<cN}-(2h>hzsNg)Xflb-2hnOI=j>(Jc9opGdP>=9+aG^8CC<rLT45s>`-_~6Nh
zfbJ=dnY<4yWbih)=L{A!`Qv$~6lqRO?{tC%L4|8j&;iB_cGs=d9I%q)^||r3v_dwf
zV=v13wx(_SBOj5*+uC+vMnjY;;M`D)N5J6bf}Xhy+E36T(QhhTZ|Ms^9KOb_l-`x1
z6%RoFPrSKbc|;e&6X)qWCo!ML0pMWwQ~}k31G5R*ThOTj_@>i{vOh}Z<kN_ohq(p1
zA^w%>Zw1&~s-|!y2P05rYEZ-RAWn_9kC6_g9tg&e<BCT*;}Ja|)SY2gf%=IVFXN!y
zLY3OSLFrx$gcBPli9<FPYvS>#aDrV>3gCDKM87NA;tGe<6XBhhd$G0=EGie^zm4wR
ze~F6D;So{jEG;g&gYG?vQp35QSet<7vjc3QbxLQXg~vqjX-(Q9=o{w_JWi(7@fD;V
zdL`oprqkv^oOXh83MrBd39fWMF!?bML$S<P8zz*aZ++lX(P9+0M^53C8MCL_znqj-
zazh8VrN=T(#}Imp&Zr1rNIDVJRtZe@cq)(}9=c`5`4(WAyxODxW$O5FjhiBvr<)+)
zC_NdhA^)HmyqBQU&PE>@;!|n|3u#1GKaqJN8qBXy)ZnyQ#+a<?+@Tlj0sFFen1}dx
zIQaX85z#rm5}%d)_-)EhFK=Clv*KMY-q|Q7gMfG6!5)W+q%uU&Y%!J9hNz($Mdw6S
z7&qFh@H2~j|7+QP9cUjeBr*qpQTK`gActfKfrNilL&;`y=QYw6<*04<{Xnl70QV5!
z``_3bB;umhNdhs;=9r9Yd~BG@`dA=uEIV22`Gt@5&dK_I$Vum`p~k#<BjO>JkZJ+L
z!f~aPcvXh4rHJtaagk!j5`+P9|D%yb6|6ZdodZ5&yy9ZL3{Z9H=B^QURXQ3l&_XwH
zU1We$sb!>mtYpHo;<XUj50LaTg2k*(1}Xftf60;E#6xhTiBlmc(>iVg^U}sirr+Q1
z#YhU)D^L^2nVEk{)^~$XY8_uXfzz9{eVfu(%$6rRNL-D-U!l=dG%|;zSQm231kV4-
zaiD&zLnNUhhoF^Z`$cEb)s}_4Nb;7u*eJ!3Uy+dlhU<5k$v>wP;sD4p4y#z(*r0yg
z12%a#RrO+Uh@Dq)<&_FPO@d^@tE1P%)H*UDdHou~mym9Xfdh*(X`+jxuSt)V0g6F$
zOKfX#)`Q#Wt>xn3+q)HeVKYt!Y+penA9iW86Zp_FkTW4MP8r05%IoWE^{xy^%;*8Q
zm(sEyFS6@OLxH^ssluqv!X?;sARI5kkct3+>4|f?Ev2rc#>AX*nsMyoa1LaA8oaME
zBP-s#QA<mCYl|wy?uy{Zn+NfXXJy<<$cd4Nc-R{d6her6@6_J$JzdVrHAv~Ib@x0Z
zKL#CdvSET6sBz}V{DvZGuFP1p?wC^+7LWfF%qJadSj6He>i7upvE)iw-gMps{{*co
z^YoU~{TbiB`JkMDy3e#AD4S6VA|yY22@MUm+KzH<r#JLX<mm6;J(aR5Xz48;xmE4<
ze)+?@mf7$KI=XjXLZaaPf%4t+cX)T(4SDU~6mjX)DVUrEVuTE6xe2$G3}$ScHG)Ld
z43-GbtX;9bG>r;?j!h=A|FL2XE;!O+`w~Jn5oTF%{`2uN1T-R_@mPAqxXJVt4A^k`
zC%ej401PBADji6GYZo`h&_E`dp2xQE+g$v9!F3EQGe=0Jqb9xOXXdCBh=+^IH=zj2
z1CN;69bqLP-#VXrIrU1obKW<8oH>cZq#}zAC)Vj^szxRsp74}>j!SUvhK|CDzfaD6
zZ^pWprB;A_#mGkQlI$S%)<PVKLyNN7)8NCE+z_@I5#0>Cz<1Pn@!JhbMzJ16x4?2)
zwij4v7DCf;{i*%*R<i9-9}=P{=j=tL(=wuGvKy<&Qlc+)?@M4kJdPRlutKY#1iL%$
zj$uYM>;Mr`d8%T5FaU|Pj{pn%e~{>Z9GqB?W-95v!oZox|9^U$(Zmn|sYn^LZdqfs
zkU~O^j1vdp0ePk(Qh7TrkYXN4_p0dTT0Vuha(g_^2Mg_ho4%>4u$3X0-PP2jAj6h@
z(m!8Za;F$o9B%1#<l(-%YZ)x{mI`qXBJtToM)r5cT+KeD!}qRDf8*BcvnS&ipit|i
zQp#<@AIzyxdgV^3mCs$cD9_Mit`-U;nM|n&jxTIRqg$jm-Nf(U0F(7)%AgjhTT;1S
zxXb{dQpz>XgR$B=Pq?x_Po39x<#V3-*3ogH*x2td&#@ybfSN3doOztoV_C!XGL4a_
zGxJv55h9AJf7ulQ;uGbi9=FYEZx7c;L?~84U%dduLZ6fVUL|0kUgidvLH7`Qt70Rw
zWR<jIALe-<0Cy|>z)}YXm-diKp@Vg_n8`pCI}*^v#X=6tgP4~zoP=d9C>3gXp^Ee9
zkV>|xYIObDCncWL!qyaRu~d>n?ZyB$e#u30K<sbm;9NkL?@rbfRxgZ`VWU(#R$TBv
z>a)=keVy(D%KZ6cbK?b^Icxt?leaC?83#9BK6IQAZd!5|-ynjCj$ZHOiQC-yR&=~C
zTtsGUrw!$3{&J{3L^z@_uKB#|qk!;*eNzg@yFR+7QiyC|#-u-xBI2-i?0@X;|I8CD
zwD044XP}LFFuFlz`SFktM%3gcejE^{tBaEuDrYADd?+Hay}ciE4Gb@X0Mk3%8H4=8
zwWI-^=dDSs-<Ih23twoULi;U)8#At*h5>Xxrr(R#F^NY~z%f7Wu#9<TKWl=TTPSVc
zuh8bdGjZ-%L&xX3-9JI!Cf0zYUl+t!6{qLxTezUYc9xIb|Ch~uhBR$EK%;3TcTlq!
z{QS#Uh3Ahz9zs%As|CYmeNzHl9)J88;0)cPJY@g9SNoe1$Wn|^NC!v6y&|Do2s2cs
z=*J4N!1Gy>BjN)~L#3C$pML7Z8Q1MioAKl#mJ^^VH%tuxKY%X6g=BtR%?trbem(-)
z1rIh{-wZ!L6NERx<opsnJP0Om83#a9Vo+sCSY>`<jLy#&T3lQ_P=xLHdw2X`IpefC
z+ud20-cLfue;1r05mZI9x4W?-xUwpd6V#d?KspQsNe?%+rhtONqE*#ks4@HM?^VR-
zPK7w#i}-Y|&Me7+{_r_1&3ivY-M>pXv4Wh2C$|hUYi<U`d3=3M>+c8s{Xzn2UTAMy
zcwtnnk;g(mH>HT+q~G`by?UN1;%<E9-$^lBkjh4YFA(Kc$IP$}q`6_4cz>eDsHGH-
z6ou{Mtd)%+>?c2b8ymZo{E-l5FI7kyaT3=HOfk@7rZeU62{XPeX1i1LI_UA2>^nj*
z4wsHMoH~gE(-yyH<@bvva8%<9>3qLe{?`kLcL-8F4|n(fo|WG(#$j{*_MHEO?a|E8
zyOrec0{!n=AS=?jxuZWfT;>~bXa*r2vD;wa+ZCD?VN|mRpGhla);VhmrUoe;IJbO$
z-A)U&h*BoSKJgo*k=yetQ;^194UO+8PDn`jB|VH)hJ%svYHIgjh;ySMPHPiR2cbTB
zG1AQ8v%C&?Hon>laTCBmpf}b}!<@tpmIp<zx`o6%Xy!ws&iJkV^zCR+)Nx=N@ceGv
zxCjBz5>DEl29mJ3WI4gmP@4!V?n3)qOtaO>MQ100_q(z9V>}^+w2&Kp^PL4q9LW*j
z8c~m)I8n5-HRA>KAgAOKjv>R1d<CcD_OmDP$Yy2Ig*pWK2VZ$^EPjWE2;@2^dE|Xg
zq^6Ysg+ST6lx%BO@XpN-0V{A4ah^SS0e|EF0y=3b&~jz{wr9W0Z6_l3_k5I$(US`u
z0LPBaH%JY`j+n7R8BE#5Hoad~_1D=FmxNIUT$M`teGLEe0?iM7G_(36C(i!*>aVlR
zgwzsCzK=p-T#JqfT3lK*SpP>H`(G~!&w|Af^-Z2*@3kUHz#Nzfj^<$iGuLl`KWj9M
z3;cE4U(T?>6GqH`)JhY>h<!l(h*4%UM)9wq1dN!DH19Nqt&~Tqlg_E_wDmfd)F=f$
zDbsrEd64Y}0oPlB$5yfzadC0eWEB;~DXv_>#g_A#r-^z@!_Df$^w5ofot0ICvn0yf
zn~903(EXtCDDI_$)oa5&E1|mjx9U3{)J=BOuW!BRu%5~a=bL!*dSx;zI~`*wC0Uu=
zluEBEW3J0{T-;QjX0>XJhhcr|hJ_-e(+tE|<SPfsOW5U2*f3|=9cLjRwa!d?=*j6k
zRwU<q#6EJB&MW{?k6a8JggEfGMcsWNP62oCOby&#h}ijnh+t(;I)Wl@zjeVQ%)3&Y
zaZe>tABQZ~q)eXe;NB(Hxr^J4<~uuGsI>$KXqQ$t+W|N<G)U$4<4y;2Vq&gLS@V2T
z3|Swv_r^-74dH&MY+f*T6%~(1t-@pm*}V}|1)HHHD=;Jv2O`{<-OCI_A~||M_n0*Z
zjVt&*C4_k5?RmV&L>wFyv(|ZpK~K6&QG%SdfBFt=96cKH65|Y8@h#kG<eP#%dnvxB
z-frhpueukGJh8F1jtZYHK|0Sk02S19+)wxgfc$Bh)$9o$u28Ff7`bVR5v^Ihf;FV$
z8$Bc={9Fqf>Ly-BB8(6ak~9J8|8jf1tvDJGvW&MM=I)k4l|3I)x;XsB*gx&kuVe6^
z0b7M2SIRQnn*G$K?J7bns}ym)UE`}!Q*_j!TY}6$J-JE&5@Hwva>WJat>iw)CdMEM
znOeyeWD57rh0Nu^XuHjML%>+4<SB>#F%df6IB-el<!jr_`n;Sm#Q^@5S<=&tnWCZ3
z4|hc^SK|&Lflk&pp|_%Ld>{}=ybB<qGXGtRFzePYtuZ%Wzqkjb{EJ{O^(PPw=lZyE
zaY)^TiNtg5Zqm5Lh$tVM@F~3z{}eRO*mogYlhgtBSXUHl^*^ZH5^}uLMXLjpJ121T
zC{zsJdfa4GY-`6+u|9mfP=&F?C*2e{xQ-qrNUw}oJCU->XClk(R)f9*`dtXdWd(Rm
z6jTz#^b2zQbCa{a6aPxa3DH}mR*3?^1m9IVdw?CKO_D$At2xb8G(p#+&}X0kY(84x
zbF9+h6EGj5NzgJP>tF=5Vi-eoAxVGG65?`r4xy4!u}gAL9qxJhJ89xuy5XQQI2A(Q
z3xHA!M7%_16@%l4)qV!ox4F<4r!hVu`NDVMqZs~&^&u=Og<4$j+2qtxqK-`!+dgMo
zn{wb$8fu=4m_Kvt5ci_LTEZFrLgmBaVv`I_T<$;Zgg7G(>?DIuU=h~^-#Ob<Y<`;4
zIsLdq7B?K4ER{NwX&c8ydCCuo)>D;?zl_<kJq_qn9Bqli?5F+vmA(^5R*3xp=1!d^
z@6#M$R)(WJMA@)G+&evIRbm`3`Z_gl8q+)z?j9~5pq*~c1$;tBuj@MB7YM@828^D=
zmLTZ*a3v*gJl+bVB~c*v(>s|5o2)7K{@^-SPZ^n*hLAq!kFxYIU|{qu3XA=y(99no
zEm9@cy7D}^czfMN%BEgq<-@vt3X4p5_sn_Hf9&O-r-o7*EQ++6DWpF}fZe174!?O&
zcf+~XsHLSP5Rto5raVWf>IG(Y=!dq{<@{+AVE2qtVq+)r8-IZlnS~08Q;3NMb#HbR
zQ!7wBZ%D;9XRUy*ZGF}U)<;AxH5!9PA3cHnKqpJi1fh4wfK0uC2zOBNh;F>+eic0+
zi#C$*pW7iIm-o?qgsA=<fEYD7*=Ls1(}+)JDy#l4gF1l>%A^+QFHj52kS5p@)h`2&
z-9cz>=@<0|mb?j6*&#?P9fp-IjLkNRi!4MU)p1s9AcQ`EN#aQ=n6DNIx(AT0(R*Kh
z$(T2mKHru6F7fYmUtuR;oAaC1Lf9e@21Jd~`r_Ca;Ka8O${)cG)tr}FngC<#iH<^X
z?y<sd;@hxO)<TCv?sS{++0Xx&)J{H9^K`GfmnDbc5jV>E+}LCJ$XOHMl&Gp}6=Jv~
zJPL4x*Cp7H%<~W1zH9mQ%<PviVh760ur-)SqF(*^2M|7{6A|bHW(~E3{>#X)6Ca)n
zxV?8;aI4wDtG5OzoA0^3n>a)isYHJt=I_e<-DZ2N>)$i;*P5yPHO>Fa{LC=h_48C2
zB)HJFun9lvupI@pn&DeC<K*=}eNw|!ME-Q^$_7KPq$qYGYTQPp7etB{Z;dm*KP0A(
zFCf~3g2&?!bEb;3cdpDDcc{;?yR{Itkk$eb`Pw61^;@+%tA@uP*$z60=xex$5Y?2U
zSvOy6{0X-f_72joE6_lYTQLbCm}BSJ_xy-PED?ks9C3zyWx;C3amI5GT=^b5PJcLy
z(K(d-vw}Dlhm5p6CI}0B0HEP<2QWed<mW+3sudIvID1cg_>q_ZfO>X`KPL#x6j~S$
zH+8-_fM#zv;U*41y~@Xz)xQAr|JpPr8Hmb+`G&6Sjaqg&1k~Th`MWxQH{0K{13v$+
zDfnv&{@*qQI^n{+FO|1(%!}6aLqJ^b0A*H9P+{8yEufp`Yq%qZLG`%F&2nw8Zxjc}
zUnfiV{pIKS_2FR%<6{LC0WpAF#JvqbFwIU7{x>33os=~g$Z+IK@T;Av)`+5#Q;^Sk
zyqQ4?sra=5(kKei1k!{2&gl6S>ajUNIt))9#>C{-qMx)1_+m8DSUrxmU>87T2{nDu
zs%DaOu1(`X6jEv2106X!=+$g{td%=KZ_^0H9RAQjx=o)??kL9dh$5JT;BZjQ5ra3+
zfwB(gEJn~y`upV+s#pmQXhg@q3*8{{9Uq-GP+$fiNTwB120%84f!2?*V#_d_z8`)g
z>g=!%Rba9l-$1n7O<oW`tpcjFMkuz?)|6J(v{rI}yBSYl>IB11z;j+rI<lZvp{ev7
znux!WZthi|1B~kpf1lHW=?e6D4o5K3UeYX#mQl!WZ~|gg6nZZ&4i)lhg?+eSoWm@;
zLqDsBV$)%vR>0)n?s%-h*8PKmv8tN`bjgf*6;K!gh;iZg?UWRr04Udy4h>=eQPX(5
zGRE{(5x{o1p&$x+tud$}PY&D#@P>M+2fjRlI5T<b8DY;!jK7LOaO4|LM3<TsvR)v{
zX6aXF-e&i<UD~=&f%_68Pzz%@6Mxn!DWQp7seB!&!x}?6ccT^&A?w%+snbVf`H->{
zEBN6Y{VI@~B^#v_F4H#H-*4as87{Ix$wpV9f|WB2VA5US*~hI{k|U{E_6=wiGy9XX
zJj2Yg=Rujt9{`9mwIl*Ax3?y9Q1~b9d^-!19tA}W1)5K~WRsBwzW@uHWSQ~M^c<Ap
zlv6h!@c;=;+A@>3@%i_YTeUIDA~q+q4QfLI5%ksbSb2RIM8Czl>iNm%R@&faHU%q+
zk9`>4rXAMRc3=F;TmQk%{{|mbVR0p6)PSGLRJeZKo7YI;tLJU%kHwWUt*i9v?#-JS
zEYszF_54?^BR!BE0Pi-wf;<j#VofHd3dm<x?E2bbKy{n$l4TZb@b%SmEvRtxoc9I5
zy%u^3kaTt4KPf8Jy7OGD`f%S%Y7L7|Hk|iz&QifW7SAm$aMbhG3$tC_p=7QY`Le2^
z8sXEZZ2)H+`K{{pcQLT(L`m*Cy!Cu})$U|3camgM%}1q^(+y7AupVl)_zNCfIQv?f
zH=ktX73Q?9>IpaGXTNhwayq_@s(foY6IW2nnCtao$0CBxOm;`~$n*r+ki%SA?ppAw
z>^xoFgTY{1xf+IBbJ!nZ<IYDaeQ+rEpIz7Ozryb5G|DmVSTwqowf02qd9T2g>6E4z
z8JW2hZ#K~ls;uIxXA)xd{8l8N*D$yyav1yQXgm;4X0>LsboCXZrdkux{*hi~+ZkTe
zMm^*lVSAp0`K!7W=|aNJJKh-VL}hJfYtoe7*D+K>DZF8EDX2e^gfo=beMWRkxT&-v
zS51>#a$9sf$(k9}i-m7f8NDa2GD`+yV^0@&UGC2fE53tcl9Yu~gEE5+vDC}g4qs|#
z@zvZp_i30X-<gWXQ(vh`XT;o_bS_TDIo`n`B84Yd>YHcE74aFUF1!WUt7Hgt@<0{m
zZe@w8v>QZSvoi=-0RBOmC*_+DN3zPR`y)!(hr?-eCQfHH%OjOeEwvvDQ6u*ZuY^_#
z)02s=>hPz0W=}|z_dahFpUQK9rB{Hbz|NPim9?3xYA(DOy1}QP_eH9&3>LJhqKYdU
zvMk7UMpJl-uRm`D6#h*uZ#qbDW1y6nf8||HdU55M4m%?Aub(KlvaYjvek$K!5#5(q
zv_9@Jo-!Biw@sV!O?%Mb*7NlVk(KV3Bnt>2P0Tx@zEntG95Jw>xY#n9VIO1cniN5H
zHpQfT<l(9lx!Cd05w_Y1+NA1JqJB^d<7|Y+qyD^Z^0i;doif5}FhhQ^ZS+0D&hYOj
z$*r3l;Pc(4?UNWVaFQE#N+=RwfS-A!l;9sETu6H56&JX*1QPt~@S|I`MUTo2LXSlq
zZgl}2v%vyg<xJbLulwnWzG)|v;SCyWBE4ZuMN?Ka)jynT&n?#(XK~{@oJ7r_?rNgr
zEE``e)ygQ;E+E+*cJ7r>z<`&^bNS6kAHw@c&zCkIda<wgLdq&rcZtnTlB3hE&oPUN
zzPeFC?ahr-`*Nu2M9?F`Q5yBg2U3iTmGO+7OKo+ABdZoY!fQ587ZY}FHXh;Q(&mbf
z2zxh7aWpx?s?doxOm2|C;+#woJz0t26}NQWavy~!>)W&sSgE-_%N-r-=P=sjiCy!H
zys|mDJ6v}X=A%zh#C}PxYD$}Uu90GH11T3<nyUpQmuR&xTxJ;;lN(?6K%qv8i?`Bd
zR>Ioc>%up!vxXhR#h%MH&!+y^?nvT$Y}YWGaAkZw+hMn8d>`qZzUu~#-UD1AYl~;U
z$G;If;Bt(<z1lg+axf_x97k4O`@{q=7^p6qmu;w#!*OPr|4o$tlOK=5Pb7n*DhEC?
z5fxGlT8b?28p$7rLpqbKTUBGSO)+W`8*U&>Q*#KAtyhxHA?`g746(^XE%6+|5HK<Y
z+4exkBy;^i)9r%zRv<?J(sUS@i7@1cjMX0Rk~h#hrMWeM82V=1K+0mKzE$2|X@N?*
zg#1Dbi2s75PWZEU>HUiO;9Df2dS26quoE%z2&JA0t&U<y_}iIk#dbEwqC@XVB0BaE
z2mw%CJa;NTAW}1W*#}B{00?M!%L%?dqsH6J_e;Agknb^Nq(YoMP&YP)Q0K^RN~F8i
zdW1lj*0{@{tKa)Ltl5gO&h*E12z2TAhujGdHQqQC768TCUX->fd@<0<BnrGg3*sq^
zlCoNjd(ALHh6XvD=@R3n;ivft2xU3h2h?Xc5(%9HH<$@++oNy1WQodEkirlY2V4XW
z1uN<3%lxW82}ERH$zBB>r>1>;97E~B>JO`drj?Szcn&?lYS`~2B=GJ9ULwU|>G6y?
zul#{EA>Rs+gE2f!O0L7$%Q&!?sbbS8F&Il28Fu2??7RyY>JTCby_6y=<rOg^mLm|k
zR{tOUFt1{s2BJpJA5sA>d3ds%;=)8+Ot>V(nMP$Teiz1L5e}8lG(+AL4gGIv@^K_1
z2tf^>r4GMMzfv5*4`u0>=uboz+g!aPI$N?2p+1dczmmDGmSsZU5rN%5JQ<|&uANGD
z{pR;Np_!XVp~Is}XfF2#p?6bjCDZU&ZoZP8Bw}@Dv4HMGALwn`-(DO9I$Q=iKpB^1
zpG^Vc?&FeB!8D>DLDCutE`z!^9jW&`u+Yooc3Vqr`Acmb83sAm`U4(9+f_95bx2Zo
z4TIlID&zUT+WXRQD%bbzS}7_@C8UG~nTrgWqAe*3$&@*f5Ryz~TG=UtCPU_mmMkpE
zJS`=WIV|%Kq0HklESB|NH?{Zg|LZ;85ASij-`@RcA6sqrb3gZU-Pd(q=XqYbKo;AM
zjiumw2xDI*RZqoV=B8YmnQ!@7+F*$BV=lJ5@;`dytH%wonV_J#$nxipz6rx2%Mn71
zgVl&9#)I!)w@&QGxzne#|IHUN5w~ZrWWa&nf9&5M6&~jHA&1@nBY9qjh+4qyI&<;R
zf1lEI*c@m#c4wsj5u>>giAvwDl`r7WA5Dh&tZO!84g2?3{{~9iHO*org`OSQQK4)G
zz2PA>Q;L?732IoE=I;i5tdbcNZez`+V7b$@|Nf;dz%Hjlg;pvNS@iU>0|6uRK?ww<
zd~yp)Oy#%WSU^+3AXGHl&_%Ibt_|(p$xhNa=a6dfM{EH#iwSIdw%%PnrvHuxK8nv^
z+jc$t-;8LV14@()VNR3d&%NP=7sAVo|L-`l9y10zO)?pKDgNh=F2Jr`;MV;kP|>xk
zLD(s3SF-+d*COc#tqpQi;}6nx1LaZNu03-W>@weggTov6GH_5OokZ6J=S783?qdy@
zI=PwikuuIUcR=;u@X3Eyqga*Dv%wcQPp`y<J;xd*m|G^avAVOVy8Uf$SB90%5;o4Z
zJh5Nj!ex&|B{N7?l&rz$4n=p4*@2ny<L*D$cH1b&`ucHv+dMX7chzU33p>!P|ESk{
z>%oWQm|Yf0cXdEafF+FX`LG#;nY)INl_$De-<RB^wt?EomG#idN4Da}l5G~xD8}+4
zrckjDLFF>)XzH+6s{Q(Fq(Pp|mhBj~QjC*x2|FFXn~LA<iQnxWmf)}R{u>5OB%*lJ
zCTL7=S-}Cto-*!A><|voou%bITxkz#_uI+vY5oiFU5?4qmS?O%A;GS_PH^JnGgq&#
z9;p@YNe)|~SFyFqyOpkwUX2ddm-aZe0ld_-mbF6oC_tqMXT5CkBzD{Tpmo^&bV$l|
z8w^(x(D(hIh`$K@Mu)KU`ztW#_c|0h%KVJ?Z2<4OyK<6}D?b!y=TJpOSi0Lj1qt|E
zW?ptpoe#(T`%ky_cM%P~<`_J&`VwVK-g=0csP=k0<6#Nf%j`(KZmPmG$rPV&Qgw(q
z&MpKvNYcZ2_xTrvGl4EQlw(V$Rw9;UL&f*8-D1L|vbcdkVI$a5K7Io%DP!L8O>Hm|
zt`T&Iy|ZbqMS`e48-hqzU2IWaj|0SkYX~Hx1le~6ZG6+??AXAg#V25L(z&*=vWss~
z1?l;116Ku-l}@OErnz$5EoD>0frlPotAm5QOEQQ<B3b2u>2e>N?}{RGi+=&%T~)75
zvkB&gmp4w2nx_6Vggs5MOYJUWaRZ}_Mx?st+3f{Kp=zALy7o}r3YYr`q2^<A6*sfe
zaIo}X#BDg8&B+BHw+PO*{;;86rfpq&R`cuwb=Mx~riKhBC+J#=m|{u~Pl5}EI29Oe
zwIV=in@NOG@~=9My@Ax_aDm_<nHvcU2Rr~gv8q}*s2FdB1!KfxW$7h|4MblE8rSoP
zy_QpeH}R&DZv#DFf^jQyEQkgfL%98(a?@BiOCv(@`((cp4Aq3*ozD3lDJ7Iy|Ccqm
zVcB;_r!MWOKNuKvn5Ck?JL{E0)`7a6l34<5`rKzWB6kqq<PU(r#37>EXmoDrVrHYd
z+V{^vwiO2_R=8e_5STvcs&X?Qq+1ywt_V_r4S6HUj*;ilL<BXlaOr&aqVl5Q6Dw~K
z_c6ATN;R#M2YJ!Rj@hpvUgCj=RCvtQu`un2Dj#;lYpOha_^fE}>;$+QW`gW_-Yo?j
z6B0(*+&Zzl`A_ndANyE5_&9L!ga~nYdUWB$I6xI@IR)wF>=Q`(JsN@7k4u8L9b9cJ
z>B%#3aqFq;+KdITa!Y!G6kWk-A)O8Q8~htZKsC>@KIQDxA{mTk9V5YqO)5}s$=f20
z=yEXorSHbR;@IINUj{KL92R<7tI@$3qC<^ts!<$1^hH%v$xVh&g}S7I<UZ@`MNFar
z8J~_W)kyWo3F>x=YEjK&?IyX_W4h@2YRhaW;G)rCH7O9>0W_%O2e^d+E)9&GI_O3m
z>@I=vy6U8W*!1!E&(>JQ%J*x#uTi58sGs>+FS?XuYpBR3MBR2T&5Y8X#I0~_(b_|2
zu0H|F%J7<JMocDUNwIDLR5Pt^Pu_{;S0bquPRGtD$?LAOO}bah<%nH{wEWyCZAMOH
zYIgUH_cb=<D)hmkwSY5rZ8R^=r93#00_3u;?KbThC~f`vTBRlXv#D=XBXz#Vv;(&3
zC9lE;KY^di>N_ibt*=hqM!}g!^C)xa{MGzGb`ExH$PzR_TIfmTVXAVUqgCj8+kMMw
zc!_cRL9#kI^8&rl0^L#uqZqJ=vW5yG#cprKUu#R=2219abmwfGm!cM?9pJ(U?|jI0
z0u?I4^H3oWvO7CI0v+;FhXh;-`zM&+a++Zk{Wl`4^)>*oR@=`at0?(+;$LMCMu+Lg
z;sfO(FL`R(Cao7z*bwQQ8g9%@SRn-zlLF+by1R#a+Azq{M5}YpPX7I8Z-ckX+~wXv
zinq~e@3Cj?wcB<j)si{3V>%!_zLy@<HLy<fGnkVy=Uxq~7hk}I;v>!|IKx;fVRTQi
zif+Gq|C3q3ep{wPaL>u%=KJGvl`Npo+x8MUL(Q>dt<=7cQ*1*G=4bM$PRGc>y<4YR
zdu<Ts3m@2KyWBTAboA3#E-o-UWbwT7u8dU;pOT?!VDUtmqjt=?tj;G^`-UK1G4}V>
zkd;a*nAvFlbUwm0BNC0+V!32F8#X;kb7Z1sNm+NuUgXB;P0fzx?!xELZzAhW{Eshi
z0=_^)ZLd;+N$`{+rLkR{0s&o9^OC{C?|oIb5`I>|u~=ki@^0M^#sfLH-6zq-7OH3R
zMTGTPVL82@+j&Ph^+NX(4stKhRK%K<oy{GlMazs8xUT+&Js9^n!rG;~M;OOSm%nux
z)WpeMU-y;H9P}aSRlcW{JI}ThXO4{=!Y({o1XE02VHioIl&rgU6mG#$ID6gcW(LpS
zPCkRr=XNA{?Fimk)^%XYA`F{=P)8SiDyLCa5a7S;nI#BbxD68a=Gs7_g#?b$%QVG(
zpFxuf8sl-+!hjj`hZ6|JX1bC>yKfkWi(60T2&)o#Y(=wC?t!n_(!q&5aUKh$w**iX
zowF*(fQHJ?`+Qt*ppEujhGJ{lKIa@+qIgfwG5o42|JnhVg98wLPH>KCw8Kj<O-KVH
z&CulfF?O(hXly5$3zMw0`L1lU3x<xu_V8t{=+_$(2ua+kuG4}ikh#(kG-wx0fArJi
z55vp1Q3(xhes{-)UWvB#>%p;H4c+INxsELD(F`T4awq-{TutM2G8f}l><8IvQQ16i
zSOek8=)Ct>c;|Y^AJ;<EeLLk?CWg}V13nUr6(V0{u#^Japtuhiu>b5dpIf||7;Nq(
z0p75f)e@VX)eSKj@3R^t8IFx8n3dOFkm?fW1P`V*&gW$Ma!x&<;oiuHEZ}{OsQ(z$
z9$^SCrj&9FmdqTF$9`7bym~DzP}>9K)LN^PJ9{xxD$$ezg85IgWs|X^XcrAAw%&bn
zyRO5<Qr912`aFBLyQDy-R}J&W=Dh`fZQVHRFfeKt933lS>V1QHOQXE~jE~|3W`}+|
zINW6PdXZjlgS;aX8WlQvdrhwH50PjwIxY+-ew)P$<a(X+O(&XprU@CFSi?a`i}CJu
z|B)DRsQBdK9!<iPywQJ7<mEX@JzDLqRd<KbfIM!6N$hrX?koN|lLzaKe|_YA@Q#gN
zWf^v2Nd~5QIr`kOy!sbN<#D<#OLtPY83$awldqOQa2h+P_>D$44z%psKncdNnYg_?
z1ET2%2{79}ExWrAxmQU#3_Hu;t8%nz0Z{0PeA_Xvp>AljSc4U5YRq>l#h{NE?DO9T
zGBQT-$mT)-5r-h45Oh$r$1%B_2Jw{7XT`vF(^k4ohaS7eeUT+5ofe$l$E+SqpSM9!
z<@{wPqM}?`O60Jw7Rj!tsF_R9hZOa6jP4HG!cq?{wnj|{u8B!Mgpznp`f`m@6Qo<;
zLsDWfb5%#f6Cz-8U>iPeJ<~p221eER-@%ez=L-F<>rPmJ6M2_>FJwg^4WP5=d`mF|
z%k$nfF-CrzB5TX}<^Z`{R|-bIRCxSO-WDOaUbB|XS7H!39~u-NArsT@5cbh|ovvu6
zh+#9jd5SV9MSPL$D*yRtDHm8HSx4Dr@2Ow(mA?))Ao!SR7~|O-cJ)J@&!%mHben|h
z4S#*IzG>;Z7&70LKt#GMXv|m^(mR`azKBfAeY4W8>hA9PN^v!id@S7UzRlHeiy2c+
z?#O_^F_79Dc%kxkt$XAt{E6->N!)9BVfYm^XDG}Mc|4FG6zX|t?p0C%(@;nok0b&%
z6(4f^LAGG*b^Qx9+l?oW#|wN9Es5Sd+q{<I{)i^EU#4r<;9g-i^t^fQxjnbW{(UyW
zcSdMW2rDw$#y`R@tR1PZ`ysOBc3fU(n_FglR1a$zc9|EYuUT?jfEBjia@bm@K=X(c
zy?VxO9=bGh?HQoPxWoEz-IOahn9Ra|%62KK6eL8egS#{f!S@{EjgwFHuhe?>`5&=)
z&_v5RAxYVz`TCtA_pS>3zgfg3I%p|P6-a6^R`r+%3U4ybLRX!>H>Wspm+G&#cRg0O
zU%InK1c9Du;dsU6dj#6F2egIeqG$+@l)Z*-CFHzi*9DIJkTqN%2fMhKG%yZpuSLH0
zZ83VjdA?IX)K?XaQ(4woyvntte+NbY*esH>CSmYzi;G62_}lD#TfH`k-_+-Sr`Y}b
zI?Mf{Q&dVftPuCvcxRwcMum@Ugexs%>j+`Pper}_{I4&vVs(W_*H&0i0>-$yRmXY5
z#*dcKy&z?RT0)x*z{mQ)k|A7NEI15hw0B304BzcpGATX%I4j0eC<P$WEcf?)>1f<*
z$ZMKI3qsw8e-xqIFUM(_@#h(HrOg`+zW%CRHb)tT%4%}Qjv8JN{Xj9(5`<kd_2xV;
zcp^-3#q-T>&vZCpRvlZ-m~h#)S*PPySVzFJLC_f7?7>X8n4v`>)(v3Qak9QF1M}G-
zV3w<bFNJzgA2PL}fPGUabt5j@(Pyh;PBVP$rm`~AcCn4g+3z!$o$OIK%UaAv9DBd*
zjijqo#w1lY;rEJylwVn_Qvv?26~m``A_X^!-hywkJ}i($(VJ8I_=yO`@qZTB_<Mn7
zkmsvIvS2Jc0lf-2)a=b74}S!ow+1u|Gs={Eym)xXK6rQv@bI&j*S(cFqT->r(hy$!
zTwbUb!v2K~i!J-&+kwMwn*tq!3+QeLIh7J_>Lsmgnnnwpg<eYx@<fv*=6|zN)cI)J
zwsCs=SJ5;&mbOyHhWzUH#wN(OuG3}c@1h;lph~bD!faq$yBeAjU`LIv|5OQ1UcO&o
z`?IYATT-#*)2C5n8X3@XR`lK7>i4@e&O{-mKHOB#nXmyz3Quz!zjj^*tjRCU{9WIs
zPTRr#c=IyYi#td<E@_@9ULH7#1OK)}o<Ifz<6B=2XW30fOoFi&e;NH8Kw3C<NjKP-
zH^^+>r5f}GVH@jg_C~TUGhWBMGcdpw{w^EreDQ8744;lsQCX%N4#j;=ul1EIPO1<*
z$oE?mTZNx4u9+p!PKl@mlp-bA@rRbD!dPnGTD^BG=RjSz+;q2yPiRik@|pA8u{H>7
zPJ~A*s1vJ(FgXOSS7%WJEUvumT>K~oU2d#rIRcb4)`R`WHOproQOA&ajdT4^=%tI>
zL71fOAF)y-p5DfX{r6y;nn^Z)umvW65Wnnyw38HA_jX4;h(A7Te{QWz3lSS4yyOnO
zXr_+Q?|r^{kLW&k%<blK(%T=n64RR>F~WR3u>@4>(a8G|?xCZJ@S24Rp1oeVBRb2+
z8+URr;%Wo5fVGKLV(T&T4AIbYXuJ7gzYCLFBj`0Fx~GC&#rG)yIAh^NmjWNe5M)#%
zQyh-r7K}4scSVfqI@k92)KNhwt1(xd-qQW^y}jiyu#{E(2APUF%7}DvxqIEgDlp<T
zCF>PU<<xA8w9U37&OVBAH<M&yl;w<GdaWcCCx@>-1m9i3>Cvh~1kdr+>eBAW<iht5
zE2Ssk-Tx#E-oChXLqcY&d0>{Y^zMg>Z4kcDHxP|7cX4nM&!i0Ft-jAZ81Azw(cBo*
zvl)~6^VhlE)vm$Gz^Dwkj6&sW;<iMcXu~sbC{RTkm3g2}H#2LUuAuh7ADRp9$5AEU
z==FZMSnv50qGbMQ3MiQOL6zTt;lE(mqj&%n8Ttyu|HSTSfT>aUO0SJe`Om*;Mh9r>
z8RnkEKjAhC=<|kqDt7#d&>aLprNq4jkw4Yy^+W&(X=j4}Xu2@7p8_aw$0t|oKS9D3
zSOP`r7Tq7qE@1%TOAvRl+8<2=hS{Tl9}=G&{P*d9Kh+sc#dWrHQ^tSF;^Vzx2`fSS
z{zLipOKx)P9_-J-hZ7BdJ}VUnt7M0MhCi3R10Z3JQy$@e$Xxz!U(w90_rrf5C_v~Y
zm>DLjaCwTuoETwL2%r|BHB!()lZgLT*7IL&CB7%H>HZvsI;-yxszn4DB1vF#vL|GU
ziHNe1MOHUm0I$GMq@cz@tzA@}YywikAv8icWa-A}4Q$4IP_4aj78oZeT5Wj**ivmu
zq`|KPXvwAz<cLOa{nNUNW;-+jYG-0N(SILhg`B`s@zfPP8nP;Ug#bGQ=6vHbcU-bV
z6kjZzS<1Zk6uq1hGrDT51iNtg?>^ra?nz+!fWI_JOHEj!->e(mCkUU*1oLbh8s4!^
zEZ~=vibbb{`QxkX=v%*!!krxpz1zK8?KJ3KARoI#nADdDvm1DY(@QX2!F#Ix*u#+g
z4?)1&2F3sRor7oLbOg6X|AyqRn_|&W52pLF-+VS8qEJt|U6LS5r<5jPgy2P`nuf(v
zzVyLJTgmF}Qef`3{X5AxVD;EAn!x{R@>=chv!0n%0&|wTG>eaV^loJu(Pc0KG>v)n
zfGWO(N7qRj)YAWTj9~uilt(PZf8_g5X!MaipBDW(d00uLik;b{CHL!3!mf3LzqeoM
z4~WD7xow4h&u@JyUnFR8m2E`_);e@bEN+m_H9%)SMjmSqZ=Q3~qTA7`?sdq?$XJkP
zj+)Y`vKrt1DipU{SD$A<{(*WwR|_PN<yu0}Uu_?<1H)Nn+n#Qm1~@7qb+0w(RjvIG
zhn<B+uSX<EGCK<Dm4v!(5($3#ojVk$09jR2$^b!rY~KA2rZ2up3^$Rv%jy3SO$3n#
zdJ=zk!BlY!42v?j_QFz&PJKz0KwP=w(v1QrE6{CF@Wes0cGfw-6K#OzE%-v864z_T
zp&gdd;4WsjFJ2tX3h?Yd$aM~}!GK;94r#?~;%!u_E$rOKe@`8ak7)$LS2Q4(G40Z%
z`xPT#Pv;B{z<~|mwHg!oLpOOpo2gvSOzeCDGPoMr6}NDjXlnAkknd_gmo?<7>S!WP
z_A(R>Z;&>xU>u2D?E-0iRTQ|eo+D<~EZ9{F3A)-@sc-F_hA&A1c?K~_e?Ih+i~3V#
zZ}z~uogYeIt7WL*Ut4!C02nh46_OYiRQI$4cGvLW<m^j7PqN@t(+6bU*!G5jfrLxC
zlzG7uiAWpJP;knH4F<vA+E>BIRyPF_4*KS)y)y9G-IAH9K0dPjt#Hn;=WZ{qg%Wz3
z7Jr{gXl%e|=+JvSZt*-~u#Ba{5tL8OZnw!IKQVa7^YUUC7#s?3y?LfSN%}&pob-qa
zG%y?jd~SL3Q#$d4VguPx*pW*i>-#8P2DD;-ZZ|o0Zm*Mt%hOd^9_Bn48%Rm@0y0iN
zE;y;IDbr1u>V7!y?HMpL2L`>PMbgkXXqm`Z>o#j8dsIAGi^te)JJB?v)5mOR0=wK?
zb&A)y7J)5pnjOA}t19%{St%^YAJT2hcI%CjqX*@ZM?!H?&RyWh+BI$4t(o^eDQq6n
z@TX%ZVGv8cz=!nqIh$DdqT>0Yi7`}zMZUMwCosm?R^X@a=Uwrpst$+{(p1-_^)BR&
z!w3RhZ+(%f&~@r9h)0+P9JG`U@||Sc8yyzUF60ZG^4(7$g1I^4iMWc9_i@x4A&av)
zQ-e_s`zM_RT&u|u#)3V1Q(j5)hBrTKyep*SuvhR@yx$Def1kUzab}O9Ni;1G{<<~O
zC^C{qRfv@Kmg45!6Yq)}6qn9;Dx?-W0LzEytaA@f4deG>)yC*%nHz!Jc|)?=fPG4u
zs$!-~EYxSPQ_L98o({vsAiiHwwv257{$@-Xq&_Bs8StW{MrDT*o@K}e4^N!2J`o=h
ziY%CQ?x|S=MCiGX<k`ZHi+<W=-Q>ZhHGh!ipf;d(yB$Ei?;(ywoMPX`o|HqUWf5`l
zq@taW{1<fd8CN<q^$@$TOPSRbCkHQZL+v;?nqCOyw#Wyox=kMH1)L{#PT*aRv!8eG
zODjjpcpZ$>hLeDhm{g5F3Mlwu@MHzL0||)aQ1JRaarwK|sm}ErRKaoPM+4ZV!{g%t
z1>5kh0duaLOXy09--mfEK-BbGc}v)*YqOT_Y~f)xHTwZ%E`!Tboi|Ih^PG62RFd-@
zJ98a}pxIL4H>op@Q58L^>s$dT@nT&j07+7Ge$b45%(OY+ii&78?v~0~7j?$@0?0HQ
zQO7#+fdm!6ay8U~!H18|aacaeI1|u&4I!y^XRlP(kiTeniIt@%p^hL!_vvpC)YIO^
zT&kI@^Imactw>I5A%5<ZA#3yp@~B;$qp)z@7jS)RsgyaK!v!4KA(*{zhx&%+Xt^ey
z{i!n)5fc5o$*G-PPhX`XD(nsWcyyDr)z>hvH{sZI*-a;rw5EOVgk-jIz~aZgDsPFh
z`l|P}GWh(+qR*Pu*S~*w!*J3F>WVgNxL7nKm)}LIG`+l)W8f@Nwv_DE`u%1h;Q+UO
zPjRShI!8NEH`jy}rN{~B_Rh|^MXzfB7mV#lX&$qaYS~>LinT-$^m3~GazEefxlUup
z>~lJ+d~6;t952c3I^<Tj)Xu4<`c34eO$}0g3ly(>Fi)^(!FLY4u++&DJf=J;+wKK8
zGr!-H&*>i-8k%wJuZ;4(ACu+Hb^tEx5I6z?$IW=wze=P)h&mPGUG4RzDQ-T{boo_p
zkrA;DQhH#9XmVrp0R%>;yIvYoqr@P{<zS=aiAKh!?X}P4!QVLS)}-@d{>=zZfoiQZ
zAk}K-+dk!J&$t2k66_7OHhJseKW=a}NP?pl@9}ml;EH=uBIV^dl9+C>kgF&7HjX6I
zmEMs{74z29JIvj9GbYwgL04nkAwqj#TS%Jf<5V7E1yPJ%Pa@FBRM6?MgpuGd6_Nh%
zw^}X}S{9*e)I_?m`^Rk9D`R}@l);Ph1vyF&GUffAkW6u6b-HnjOyx_-&-|Fr5L4!Z
zu4GSnjS1^uP8o-m=$1+&;)TNlS8|I~@)1em{Hya{H9!pCIn2XE5Vc5$U#cXXOn%4{
z6DljbH<*y7eXrLz)XcfmeJXY>E-LD9ujC8-`q()S)f=u8#7IyR5qqn5Yrot~b)$}Z
z`(<yO<q{I?`#F7AvmHoFHmMsM3Y<c=wk9Ej@uc&2x^KoX)NAwJC5+p<jZLUgpmy$h
ztVc%>a1EB|f6}6QLb64AFpcAEa;IgpW^H=vdG+q^Z^zh6IMU{Ij%-YKft%A$asq!;
zr-wmQ#Xv%zh49t;?$+Qrcr15GqSe8k2z(krdyhc7X0Xi*0auARXM+~ht@N8Q4?eNR
zy3pPT`OIR_+OuHQ+4Q{@2O7O;^WFe0vwyx6*Z1~nsDI~(6ghw2j0sA~n3hK34z#Af
zeh>=z4%5#sZd0dAD#M2jwux$}A3mXN$o2CtIlJ^*;c8=(s@y_4#S!z;qwicM_X$PA
zcqpb)uB&|C5SF!mfW^s^q+onhodRw_S`pIs&P;nsaDS7O<MX@UjH`SwPa*9O|L*TU
zW#SIQcEXtggQPn*$+8)T>}j!kbR`HxbfmFbR>2QBgBRM4_5O77OoW+M5ESZc0(f4M
zmH6--t28u4>5YT@R#~LABI3CkH|C<~j|<-XgC5A5G{Z-e?!I|&6!uMo?OU<id|kkS
z%eoZ#qbzys=M|p8m&vs0_EH|W;GgRF30|8S(+=5FZSm7lo$iCVdaO-y=i8@V;8gsk
zvc7QmO{b}#Yq)s-o@N-xW7pc;qH1~FGGH-18T%;0S71<K%Z!W0!Np=5*QoAFwwb1~
zeNLp?)uQF#0#a6}=b0F>D=n0t*F2=S8!qXP{qp%dZmg$LBQ)>?7H31qxf+1hrY9@J
z*bGs;J%9Dq;Uhq`;ND?Uvl^ODV?-CH6}9Ld2soKo(l*`lm=PuUf5Ug+pULY1Ud?%)
zo0pe?py8WWD+Rvavp}v-ce&G9gC>&kZtWyV4i;*@U`y&%bGM+iRVIx2o`ZP{E_2E!
zdHV@(7|fYIr^0C^2KO)7<c>tPkj<Y`DFx|XK)SO!)#3K)dgt4w&X(zXd+$ZY+tW0l
ztBBn_rX)ec7HdRp7<2DAr7Ik=N`D8hOrF2XY>yvatF?StAL5n=V*31VPg&$arY2F~
z#2*_lhq{ElVx)rm9FlVx8Q{b~&cH1VLV78FBjsdG`jxBO1yP@y@?pDIeL7BzO%n!j
zbd%DBCqgMlI+VM2hzH7E*c27{z)3ew?nJ9c`{F`jkKUM3OL3*{3$D~FNjLW0g+`^7
z#zf!|;m^jQp%0jDZ+$cuYPEj}1Oq4HZ}g>orZg>gr}pR(<+QMQ6fu3YT<jdn_2rs|
zSW9hzwUiz#Up5qNz~idq0C>6_KMZ!nXOT-zJ^)<ry%9A<<3Zh2b_C8go_`P@`Hba5
z_8^PXsCMkO2&6z5G|o#^fLTpU?#j$&>cjOOY-*iYohEUk3H&R{>H56sPsOv)!YVf0
zjSvj;B(r6>FQ^E2d0z7*8!9UxAlH1ki|<gr)0SRwAaFJVH&KI{`aSfmYcf;{I6Dl}
zSAfL|isG9wHP{er(-0suXG3v*iAZpw&$|`TQRuZ&jDr1T<A*2ODRH#LIMjb4%Sw}@
zO$w!bBqGd=sS;5gqWAoj)k>+w0+AY??h5mHb%_xg$<vwDvoLjoG-0D3yg2MN{VdLc
z)~8+p!Sr~UX6ax-JcoK@X{g&mNp!->U`^yRxtLXRpn;tc?-fkLs?OT2mAbtfhvV8?
zOf6A`#Ij3-h^PzzC{IwBOq7rD7MqoKaX&kvF@AJuM-SJoGL@3##dFq(bsBf--b<E3
z_r5V{PF|ASq4&}~FQ307{cxcyCF^*JTHonhWS_|_+qG$FtExO|VZyk=>cKCF-8Nn(
z&~R5|zOrgnje17Agj4@u<wPIaDaCjy9q@Muw^tT_xC-h!9~jSSHV`Z$Mrg~KTs;|U
z6CnMjUAnCE@{YngaE(QU(g3;D5dEQ*z^)3D1-(QO!kDM`(tTTo+nsOG34=3W@UL*B
z?NNpF!Kq@KZ3|?*Dd5g$5tYtZc*}jAKsC(V1XNtAdPa`QbEhlaG6n$LYr6UZrJsr9
z7E=Xd=n`AU&uEY%o&XO5t?&Y}oJ!zfSsRL$>05$2oM_~%->Z~&pE>xm9RRGihLX#+
zwWV<9sd$Qb>$<<Fd8zton+OLcpxJ9dZikR!8;-`VJg<d8iO6lI35uV3Ir_^^+&1wo
z*H04oX0`-1E!iB?z&(<@?nT>{Y3w(X%l<J3zyf&9`J&s~O(B7t1@eR_;;l0&`{PXa
zJ2zb%Vk0gnm%6NZ2*p1HDjMrM((_+ahR<^Q@sn_fB@tEP2@RhiDYxA}bSADP4}P^@
zpEAf39Jg3))d1PfijCv@7G%^?#zvF7BzNWA$%T1neymz-)ra%hKo_N3ucmawytNN0
zm$-kNm;<HBzWR4#x9dSj6?UP2zpG_gqe#?`b0b6cl`aCr3<O`S!(suAV$!a3T-(ah
zji9WKMbk<A4E96g&N}{?WVxKYXyMzFOyX{u-nBQ6(oi~zoVUg@c6?d@GMA=d-Bq4J
zCYx<SpQ5k$o)h(-huf#&X64=PnOHxvi_xCic^wU>K&IKWWVEy*U}@T&V%J!B)SD_Q
zx&&hno_Rrvxjo4Uk@V+|-J|%L!o@4SC9L_!m0#QB3(;LR^Pbq+k|miJpagi)C-U3|
zlHI%p)cP<Un<oQf#xv*YCfUF7hMP*$d3awc05;2!#-fygcwByVO2+OeHa;#xm#*It
z9?8+sD~EXpW5IySPD9Sl{@De|x2fCb|C;e#KinfrwnPi>9AKSMCgJ!my=fxM!M%Sj
zOAU#!9;2Cynb6F915Wb%F&@)jHi)zXBcpMw_2)0S8ehIo&oyO-Qv+-EWYU`-?UhR7
z!R6jXqdy&5n>#8I#vs3#;_kfzqknNQw`!nk!pl`@iU?NViC_^Vd1G&7<qnm7Px>}L
zB+a5K9!&-ykxrT(WN4HN`3e(IzKTD^`cv{x_g>Yv(6G+!(W`OIXX+nMbi369T#c4X
zQz{l65Q$qSGe4|y!GI0BPxJB$)bdjM*tTD02P@%H$$w5JMql}cgyq_8B+UG<SEa_4
z6b<ML)no|95vX%f>L1);H2XA}cp!m+uB@L-p3yG5St0jer+TC~zl1`f=sgS3zuys9
zN@c(ms>-h*3%Qxi203`VCNvrp6TDh2jYe1HC)MwyndX-SVw?5i)L~G@Dqq0?f+lXk
z4*ZR}{D_kK#H$5&+zgN;yvZ#PzGpzK`6_`p<g}F~J&h`pgZ2EY`6gDK8G`!Xy4v)s
zQT?&6BAr9uIbZS+8V<vH+d8&GAlTnbnbEcbdOMfs-rZ3}^S?BrI&O8Kj^b>pgD3vw
z4bhYv@vi42qFbv}>0)$^9Lw6fp}mj@;E~{1MwLjL#UhF4p6H^doUzFGmVhhwc38aC
zxcG+QVUUFZ05^NbcLrpXM;nQ5_L5r^89xyy-U(T#*Hj?WKtQ}YU6Oglwe$D5Tv<QI
zV4>bpky-fW+X%w4DCFdnl5!E}a!!<HL`|68!Tre0jIKX&!X+@T0LdX)O0`EON&<Yq
z;%bf4%E_a-=yT|wL|b$5R=3XL<AzDXbfTxCM2g3U2?Myk_SOa!3hc@Qcw(Qc&ba*$
zvc_*E2fG%&#*2x%mpTG2WTJQY`q%U^G}4yPlIoFXe<E9~&6Q9tIe}(ZHryPC%&*1c
zY_>_BT2gPWs$y#S_vF2$Y10WG{hVHn_deCjhaxi{@g)6hwL9=T(|;z@HHb|NXH!}#
zW6X^K+`S#A><*PR$8@lhL8v5DLw>fTaEW%Q%AM9Hdh{HNpp}x^a!UteV|+!+v{^r`
zTAHNuLcHk8Qjk<Pds~oxa<w-}GnOKD^xdbO#-ZN~TY%15KrTCB58$OSY8nzOxvuv_
zG<xNO--FciqUpasnN(>J4|>%$wY(Se$neUsL??pposVp~0FmeL>hVC#gub*`s8t1n
zV0bR}(0EC}mt1ccSn>wVOQAwcd3#}DzpMKvV<{xdaFMz=5a=}4i2&f;mh4wpWQKP>
z%PD{S(Iy}t4R#=rc!-~kRj*`y=CHk8B{-hopBH>5|9k4r^s|mCtqA$D{}I&W=SG;7
zBx)UY3+?;z{E6R^=k8Lm9MHDzzuY&DAH-cn1i~R`7#o*zYRJdRRqus_Y2r3?!kqWE
z?=3^~s$6n(svDc-uwIO!LD~MYOKWWHdVY+9r_9K6j<@t{mFclV*vhU8sJo;l%7VWY
zu0wEF);xfb6>S=W-cOxhKoFP0$-Pf?<oW#^i~;v1k^U;Q&-{AE8Wv#7Hv$^1Fw{wh
zFnQVkn+~I4;7cD-p%em3x!(lj*(6)dH-s6(yrX``E0Bn(ql`Et{4JM(O!Ndk!fXW^
zV;9hrVI)?&D`a?sv4cVX^Cg;qYdiT)=h^XxSMsMG|Ls<<IJ1j2sw%5mtjc1di`{+!
zCST;c=QyvvAI<C7i;nY`U)sh`u3knyKqWKZhG)1|79&2@RLG#I<G1}y<*@?oJ)2rG
zM#6-@=LYifa(IKyStB$b!@EeC5Ooqmbq)AuMDD3nH<E!Yea*E~nfov%x~ND6ZT*b#
zs-*Xl#P`<|@M%ZoA@?`s_|#C*!Qud~1;~itD(8QBMGgjJ2(jz!`5ZMo;gb?`eN91h
z+Xt>8dhu5RtJl!luEP}y{~~i4fP8g(W9L5%n*UFK2U=dw4E4HD4)Z^}J~)v+d3{jQ
zmbGp8)y_eewK{7A?wKIY{@)+?2~1`YJcUtGq*QZF@kb9Njj@MLaP!lnKmlzNP@6RO
z4gX`sa*Pk?%Wm9VKq&W>DPJ2G>!Vl<S;^oCztq}y@*j)AMsGahtfN)yUiq&jv|>4f
z9Hi<9t}E4w1KP$l{j~1~R+kii9a>KRV)y?pli~S!*rBHqxx>HvB>y~h053>oGFdN~
z{P!<c`~r(tZT^=0&*Fb?6SN4{>3%gh@4lKa|GPiyF^6}7?&q>EUgSUD&T5kQe^?p3
z2MeaO{Qp?(TAuL=>3&kno`?VEZ$N)aQw$;c|NmkBpTC)Jv<(?N&EF&v4QcD(pL5FU
Lr_=s2yZ65Ul0#1y

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-theta1.png b/docs/2_39/stan-users-guide/img/sbc-ctr-8-schools-theta1.png
new file mode 100644
index 0000000000000000000000000000000000000000..63ce49d7ec091b694ff47a64eacecada44dbc542
GIT binary patch
literal 50644
zcmeFabyQUCzc&nsh$5ksBB|0PDj}_iD5)UbARsN>bxR2XDiYF4=g?gO11b`Nw6xR^
z!_Y9qb8XML&-vYKfA9PJan`%uXFdL*i^XPU_UwIK-}uCLzj&x9Lwttj3=R$s@q_z!
zm2q(J3vqBx_!AO<-xOJY+5^A1Xf7r7&{9f9%GS!(Ugepgv5Aa{jfuUvv9in^92}vr
zFjd{Bw5k;1H4bK1u6C+e-3)Ndk|1U#j#Rt(zO>qBbX345`hmdk<@qODE1j{QJB4o4
z9K2VOe(1@Eil=C4sPJWE+aw(kFXfFo!#w<*kxjgG=lR3=(2T`*A~)WdR~L<$taz>`
zqLVyUs?5+Q=c{rwqF+<sQpFW;Q(vEQ2#N?6<9y{GltNALUi}drr*5`9o`4dOOPTh#
zRFjAPP~Yfba(=}<I!Z=62BVVMETZ5D+|^9gc4u7oz1Iv9C@uOeqxlLx$rfJs%1Psa
zW(RTaFmuzNTvNTlCskD}f@8=cz9M(-oyGDM_Uor8SS;qu&wMb7cN$G8dWJSUOt${`
zG+^DjIkW3svbaCmPzTG&&pbEzs6{U4P4c~c!YNB@uX5Unk?8o>=OhST{TPcwWSm3a
z)yW$<l~bRbmR}aa<UQvvu+BBb9o!{2_eJI7U~Wz%GU20hO9j8y+L{P{<BMwTI82*8
z)><WLZEgm|AJ^{Tc6WcT*6se5mwocB<pt0Ab^mDYQ~T*LZ=X}PJP-DHfA99z+|*&w
z4_UGk7cFRJ9{BLQNhQ(doz}I^nD3H1OD*`FL5^y1&rdn!=x#}<y)QzOAotPBFr{Jc
zD%9fcBV;GFXR>F2K(3!pJ#oqDlWeiqg6hsFXq-6rgiVrQoNDlC_Hz<~>lFDE7en;l
zn^IqzYZP4*E`8VWvA*f;;ipgf&vz?3OF5ndiT_1aOtT-JN7ZYB$qmpkCYSK$62y~W
zxyX2a=y{jk^#@z8?XKfqb_)?z*overCVzt78<e$@))N)%Kd60qLpdW}dh}LlO^U(M
z4Gwu7m4Q4ilJdYS?d;f-x^?@l`b9bmT!N3vr!FQ|t9S0|?=J3N-{Rc2NSpBasg9|z
zZ`1l(Q!L-SWToNYqDo^;Fio}k?H%$3l5X7&^HZJlo_F-;cXNmhT~SS29+G=`F(>yd
zcGWH_hTYBEc)k%-%2mhe#dgqI*WvZWtHZ0)>!U8-YmxTE>H0~YXABPdty>oBG+N^x
zRBjh4%od_8^JJN>SW(hb**6JHJ~F(%#(I{}_|Bt2zv%o$ELAdZk{BI{F~e*oR;Qjv
z4fh4M2ya8-P*wHu!n3#RbZWK=LiM-FUtX1Z^75^~<LyfI_4k*flxCF8$y7vC$g_*S
z-zE(z$#71w#>4K@^<~=>E2Dah$Gq?~dxdQli90A=<a#2FXb=J%(Q&Bx;H(RQpX9@z
zsss&NYh5~t9@~i6h`(M#O0LZ#lYl+jSTBKD!tzXom*Paz`ueq9vANLo<<!_op>_Y^
z;*pPZukU*jO?=;8+G^Ol?&^K;ec?xoL*5*>Z<(M-V#7Gvv0q&BC>O^p%m+6m*;Dm|
z4iOr}3Lz7X2c`-NI2_<-LLA&6a~yo|6E65i1ODOQ;HBXG{TuwkloNmde8L|($UzwX
z0S8AC=fU0Ek6m$>Mo+(`Q}vUW7<phe{=n#t^DF7-7oR@N@m~ILjqyesE~P1j%p+4W
zlU_sWn>6^hQ20*4f3cmTd;7?c3_nOZ@XB2Z;$rjrWOp?1{F$#ZVYjx$Gda+8jxKuJ
z9({-$Y_5Z_lVy$5V$CQo!bjIhzdgD^{^MO=99%qnLNZ3l|N4P36VJ<jn5X0NfB)+5
zV-L@h8SLKsm%shb1E0!#aX1G?Q|<owPS8<2uJIG_-&RqP8O&6B+zf|(x}@)n$KN0K
z-zUI0fQOf-;}X$&4))=Mg`TAUdiwwI;-6ZQzEO)S--M2RIdBv=8h!jL`{MZN5D^yI
zkBZUAz&@O@aZT*M&*bmJuPkwK)e354rvGv6ejUZvs6Pe&tvxqm;}@sxr?jvS_r+=7
zhFu3=$%2#k8dWRwVOL-u4sHTH>^hMBe{mD?t85ZmH=p{c>$x%KX;&0n8xXSp`RUrz
z?v&EH#Q_8<Z^OQQb>2f!zarzd!2bUJLaU+TLg(e-DP3Qx7<>W}+6(u>XjI=9IWF`R
zn)haeI8lXMxn?z3c+GLXyQj-!fA=P(y21Vq=6!lPbDb|Gr<%WNu3CEg8RYr-0fhdA
z8>aHv-@3%Mb&6;ZWt4bh18UDUf={9?J?;tyCau;xx6!$L&Aai#FY01`$Z6Q3pkZQc
zwDA8R48J$m$;k{%?&TW67K`6KLc&6t_gN(HBVY0HHQs&8iG|&{lyhLM-eb#o4m-kl
z1FRX-QVtf_9ewKWi<5gU1ofM{Di0R&pRe=p?oO$|hx1O4&(v=8$W1O}CX?i<<)4XU
zRU#p0RXm%YubVnn>*k<Q?P%^g`_8OsxXiM7u*kT%)V#OJ7xx5*((sFPF4R)X!m{k&
z+xqW2Gyhg{Ksib=@q^3K{tt()0M3HS%U=^kDkyG7={t?rd&#};Z8%(P2s}Y5TIi`#
zGqbrk5VVMB5Z|9TB|*rkmX|eXyE^f4vRjIjm^IncK6pkelM9u%`C|*Vn;lgpefw0a
zil^(bqv$}(f+`lUAKqIn_fKUg#5YYxNYHnRZU0#9uc+rw@Y)(O+?fBS?UKd8tzDi+
zp@^)h9F9{%=+W8KVp>-RjUxgOo->?=Ey3rkm(_}RLWMaM4g8L;!&k4tVpAN9x2b0L
zM0<?OSzZ)&eCj$-Ge+q=Y@XRO`Y4o|mt^~ggHrIi(81n714;7H{#^4w!{KIkW>>k?
zt9;E8mISAM&2$;E=|jv)zUx5$#zL<gcHHOaV6q;&LWHr&*<GItQc=9ctmD3rZum8p
z&u&`)TmY-}yK3i;ukq_-WCneH)la{DP$r9djESzOWlnGza~t`gyUBaB#yg41MJPTm
zI7RQusN=C*Hu;=k-@yK<aPkc^MU6s(+vZh1M~8;9U*Azi(9t`5<zu2}Ku?8y{2r+!
ze$limiO%$E?CbtVx}QFDq8v0}8}A8o;WD?RW@7fKer#m&{#7WZIhtTdvMvaeZ9X-i
zZ9MQNS7Zy*j=I(&Y1tml{n^O8-hSrGvu0w!w8hR_a|t1<97Ba&;#hyp5>w{T$%2}d
zGl<$X*=ij<@QG`CQLrea3GxhzEZpS|b1$Zip<xuaPh#&|1WQejP`K*;laFic;g*!>
zoDh}$jV7Uo9v;189%#BTOt{Ni0oxyyd@V8;N8*aJO7hn=V2SFJELcd=Los^p>~xtN
zYDFV9HP#gw4<cCt{7#(8z#?xCwr9Gn&!&gk{m3nE{UqzMBR4tmoK%t8_+c6Nz2hzR
zCy&gOD{dXnPbnh!Fa7$YXK+st_=oaTv$ZgCdMm^WSl93)mDstC(wa16t|Qw{Pp2$>
zE=AX4S&+_im`6=dw7{-g2rpn)$+(&;8^I<}-C1gX@^~1P6g*8>SWPGfmi+2gMMIs6
z46jb*HLKxLcClGvTURM9Z&Y?>lO>`qnkVMeqlGUycW7NW$G(+;l+qI|l;W{#r}<s8
z4?pyHh2i)0r0AX4TbO=w9fuD7g(k4;ISD*{CIhH@thu?l4S!VHE!fKhtQ{B`sEV6B
zGIKll0(*-UvS+8%T-c6gXdA%JZ2D2<&|5-h*K*GM7L{7QRuoTLP0bW5b^rA++FRGT
z^lB?hEqCd3?4+rjoWEzN>wA@LJa|dBtWv*xf?-E8`1v<#6BI~cJLscBbdT>Yrl4YH
z&}^5^NxCvzNNprLGl?dUltIsb)^x1(NkSEqRN&*6N-?`H5>wveJ&=f2r)g-*l~^eA
ztg){@`R3@v^rFqaPv>_Bd;($`3ZsU&5cY?M?u%w6v0kyI)q8W->Q_*`cQa$xf6nz{
zhmE%Mdr)Z%kuyTu_0GBXtl2;JA4F9%k})jD^BLcB|2h9Hwb`zXE{3P=j@#g4aRD`?
zqBYkBdR#&`UQ{+V4=m)gmGW84zTmj0gLeF|j?187P{7W;a<<;h0eyzC?fPxX^s43Z
z8-dvsWA2tUBOcR%Yssl&t_Nb6xORzasjPbXwpAJRURWyB(TlMj-c22|{q(9m>N$6m
zgVj2I$Oe6Zn=JB{!+yGEGF4`SutoNE8QlVfL;QA`Hn!u6zc%-EcI3ytE}`j{g<$h3
zA7DCmzP`I5CEwbfxmJ@|O=AEeQbz5U5WX@RRWsyi_n{^Q!O`zm3U<OpR_?bcES{tO
zdHEknKcx-q!s|;@%yQBGy-E^dEo7YUI>oU|*Sz*Np715cNTANQHRd>Rj91jsMs$7X
z)Eem{-_EUX&d(aC1Y4!h`(W>bfPT$CHg^j38CU;nsftIv$%hnBXdr6~kj-%D5)-6#
z`}z8_5wdo_ywPc~U)5^y>3h-OtomHeP+<`ZInS#pH;lEAThSE_snzR++yoLcDz#!R
zYd=5F<l1Y7yBtLYWfi^tR(F3(^J6P|=)%4Yhy#zeZPn%;+fJV}Vo;NK^AlTx=^WKG
zuGPNfjZALeSvTh>kX#dv*YO(|x%(k@eQiJ+QG#4*PV4);XjQzlDe#(YMbxcsQ&{_l
z_)BhHpG1X%w5Ny)P4QNaM%j8!8li@{B#)*!iKE?D_>V9W!RwtOytLjslgmjaAgL{s
zcX~dLMr+d<fQ>W!N#Kl~6=JcVrXY;l8sR<@rQX!e+wfznm9`+CL?}c4-TC^xKE!Nc
zdXz%-!PZFg(ohMf+C41sC4t9zi(y2VUdic{biSX9>@I_wt==yU53(Mma`1>9Vp5(m
z6hE}Xt@C8^j{V&IUP|sZw%DZIz3N^OuMemYDX9zHH&G#W5}X2BpXwy{Pr89DlkQBl
zAf~fo7~kC}WZ<zItIoI4ovAP%Opw%r9VRdmD9ZV}N|b--+h;H=)V)1Wzg4z@sk?Al
z*YTUQ;nBf1dQ~11=6)TipJA-g)6YtF6?=7kragugY2DiQ<wNJjlQXbLVRQZ(f!gzw
zX;+mPU1!ZoP>b}gavTP5jO`ZJt#Syw-+xr<Iv`mql@YxVK&^M*B{bV(_pNl|%qA!K
z@nl#n1%f!v%mpE&48hT2!y#wBHF(d{<A+-nQV9ds6%vGsjLycGC3`v71R>ivrt)__
zmx7zcg6t9feX{Q+2GywW>*-W)%(;&u=-M0k3r7bHN9xt<IeXz*jzQq|CFHFYE`B`4
zU}cQ>qDm^Z@v!=G_>r<$=UpnN4A#fT&ydhD5-K0&(carNOeAj-<?T|oc)b?n-h1L~
zJ{j4XJ;mVptXhti(A%ST)MY$zr5I$#w3)R#R8G+Y2>Ii6)+f-XsRpe=*}9-=Hm)?V
zqTyh9a(}&x*K5P2x|lpbisaVSQIEs@UBi_hRSWh`hTlF&PeS`zoxx+N$h@ld)$cE4
zG3YS+m_#l6^EQl!r>Fm7INE8GdyZ&}NXehE?1<y1G)ul_ct>dM)0JMQ8>U|gw?|#{
zs}m@aT&G^FW-EDdh`H2aEzot_iUR6iOR`F)@TvY0L7?w(k*L&E$0s8HFs5POWT%<o
z@Ri<pC~y4*aCViBRnbkX;+Pk!A4$CnjRqnWh1q|2pzD}Oy6bI5tDmh@%>_1-h~#ZN
z2s}Y>mT-NxgHx=y5(Kv%yXHV*BC{knQ)#+BB9Pp?NZMiREf2IFx^E9Q5k8~HD<a}t
zB(2_H-vtu!R5FV0&8HfIx0sC9wqf^DWf*GkIePY)ImSO*B~#5dN2BvYX9&3l?4CYh
zn>g3RGu6tMRUFqEc1eAYqB%8On4?7`FDElPI-2ZM=+Pvp&xH)7WWBnoj+uW`$<c64
zkmT|FrRUVAsw-Mt-aIv`YL;MkZ^82O8)n0gf1NbkMD-c=q&+bF@|K-xOto)_9Ynwp
z0xWD72y-$vqVhDzw<1mywIS{)YJMO3w3QtSD`W{l#C?5@_3xtnzrNO|bbv?1jDC1@
z{L<m#y#Po4yyVXn`)8H=XTkj0FMq<tpJ@4C8RS2H_#aIcl9oK$>5uB}qyTKQw47>A
z`&xNLru?tMzKYdrxWZPSUX#g1s?KZA3R@97nEBwyS&BU2!$-Ua$yCcnF9>dId_ozG
z?;T}U>R!^<j(i}&avqtWe#Ee_Ssr_!F@{Q0szY$Av+Lb*BYiw(kG8C^pkNgn95zj-
zzkJQ0(f69g_pBxzh28k_rkmA?f#DZ#u$*NOuT9Lav*iy>8kkF!iCA%??ju659ZK*y
zFAeIs&WLTl2_m7RnTpBY!Sfaf9S@pO??&iLASz}UPx5%^Yfp4qW-g2k*aQr1b@gjk
zu*1Q5vZvL)B872dC+jM3Ee8sS+hcfn&o+VzgQ&NEnSWz<E}0Y{cEe*fV%{G6i*}y)
zMnlCWOx_Zpp}qEOqS1O~RtLZcgQeywe7yH_n+pw_WnC6p{<Fnt+`&w9($DD@83i2q
zgp4FIjir;Lr^0(d&!s-W`EBF%6`8=>&k54W{uP<VMO%grm|2%qveqI-LDxG{T^g9@
za7s<;g6f^)#VO;fC{QhMYHq?XE5e(1L7h%^FA4@S6Fy-EIKQk~9egtppCdEi_kY9;
zKcWdxkm5J0y#Gu}U=2!2fl`my@bPK*5j9ZP=RN6@gPn9Y89<r;e4Rgc=g(sMvv>Z4
zf<J-!|0(a3nyMDLA94JmjsQ`0{pBS=9{FT(F>0=OL0b~J7#`v-(RF!^8fWVZBF<K^
z0@hj=t{F)33}>D@p2hs^!CI@j_q-jjUxZNGuTgHLDfIJ|lm$uQFLtU9?S)XucUzAi
zr-c|Tk_gQPZ?DS3y~Soy@73pIBWmKr^rAwv?(^>_e?}^i6j{Lp2tTC(7y3vEq6c*B
zKQpJV=9FZ*?o0+X1Jq?OyE&Rye{AxQ?{4VicL3^+&U)vnBS`p!LS0^Z&2Z|UzH;y7
zQwIsoe}J}7@*IVTc}Y?64=xxabx;kuzDBOBjV$|hZydj6lcrf>RG_*Sc8eGgPzDR(
zh2ma&{#bN_#1EHoFQ&2g{62^K&%Ur{`pfaw-F$iRM#k%oht$WDb&4Fw)jUOIJ`Zah
zQ0@AM_B$Q6-o1y4bXG%!hEnEV^Pa@GQV%TZZC5`j8r^-xBsDxLVAh#XXz=l`o~^?J
zY|y(KX3=<gKYfkEZ?TB3HlCPVD(NEOTM99D?nn|70l?>#A!NdFbfo(xxFOHyE3Agj
z+f*-PD6f(;v@N0lu(Pvcm;mtxV|)D-fdJhpP@`5W(79++IT^UI<Is64^Kz}M0QdxP
z!~Q(YN%G{KyXJkmE~Au+Vw>L!0cv%%CKiU+#9i*rWH(zjWYsNx@<8ZdqgT#YhIvx8
zwwK+zSl=u&fQULHXbEtGP66sgAA|~2YjZgi0Z+?A{9RA$>2z}4_LxZRG##4FdZ^OQ
znEjQdv0yF?ZyV)cY|M!i8FSkNSd`E{`0u`D$j3eiF?`mDdn#?wf_s0r%SX{A3|FnL
zx7KaFhhwTz-R^5{c*b*5aiLM#W`s~u%nR6clJ^VmeQ8>rCpo8AVWXQ}X7_9(H9}n2
zb53OZG{Wa#k+UGBzYVAT)H8%&W5}cy?OU;TTL9Wj8(a|;myMPGdy^DDRXkpcJq_{Y
zE2L?pf1Z0KT<C)%Yrp?#@@gI9vkAl1CoGqT%M3A7$VtHcoJqYGcBW>f+VTRQQHygI
z`L@q;*+9<$K+oSqk8rjq(bEb&GYFo0sX`l>8$CUo+vN@;(&`|X_vCb<>VTey^#uuU
z&qV%!RfxF;0*WZqMN!vh!5;^a)cUkGb({Ux^fRH!@~%JZ+M-r=EYK?j)OxN=)+#^D
z{*xE0)Ia-mibCodIcz1r10O>$KX6_TS(95X7$qTo;nbrDT-CJ~{(Wt&ZCIYVjh@w9
z{R0k&!L<%ld2oR$f=<9n<EeFievM2x?K=XE8JB0O74O||5ts%sc)M<*oB?_|!|NiH
z<ymt|z+CDqg~u0)6Wx=|LECTJ7uKAvPTKA#A!f`<+ga@mDu`NY9w;Cw?|TcJI(wme
z_Q)1lcb%I7(AUBG8AfNWO2ij!mK_?F#;OmmYwM21U@Wkki<wG9W6KYc-SxFs!>v+x
zLBArM&HZh#x<NS*Zw9x2Z*x)3EaZ7Xn+!o@QT8wV_tTY#v}CRer_3dmOM9F8)b(}U
zHxRXMGq`H|%qjis-hUZgXO1|}(oWZ5jPuHK%1{Q&JYpnNQFnVL(rf*i(xJ*HDQ%Rs
zlq2sOueyPnx@qI_t+d0@vCXZ(<@yGXd%W18rHcm|-jv<~?ffCS{V~Xu22{UCT|O22
zN^w1x&3M^>J)%O}1Ybj&J5G;-w9v?yM0h@>XH*rllD)=0w%`Z>y5mQO+v$CB(Om4r
zv_2pLy+y0>9jdCI$m_&crev*vA$(6?`6@A4kv+MR^X{g(@*v<;n@B{K9@0=8HWBf1
ziS4QDy40F@Ks!Fn*3-Cpp-tk@PNVXfVKpt$OvGKM*_AhIUhVCaQQF6PWPFGa5Gkxa
zT^-$<;Eu{JTT&ZlC6I$Mz|Opkgcc1eXP_#Ct`AyQjP?jmWu9i~t#(>`Gvz$@c{<dt
z4N#p{E=Ny-gfgVf3YZ9-In@i~NxBX^0VZB$Yn)H~a;51r*&9;1U>BNJyE%CzMesD1
znC52I2aMf*{Tf+=QEI>sq86E<>BQzs0#^4WC<V8}`WppKPI)|EJF=bX#xK%W056It
zx{4+a1yp0~@KCw%VG>zeoot3sX&b69H&S{X?~JRycB7(Y>SwV@FY8V6C)PTqG3uua
ztK|pwPq;v;h#idx@x7<nd=V0duF|Pfp&Hktf|Ng3Z_UnNS!|VuM=CtMJR`LZE*S4_
zq1HGNwaS=S<ISyF?(H-j&hUs?P1LJEdJJ_==gaNr3UtTVSkewlt#D(AZPN&r?B&>s
zs{dkyt*4PMCARPe1B<f1<)XE99(|!$eH$A?YDt9~#euu=#OqqYjS|hJv6l^3)lX^c
zl@5Bv)l#q3mk%Bcl8+6TUNms?FB}{nH$;8UUZCXCDrMttyR#eLx}#pKbs`pUSectw
zZIyc{e?wpavc&0zYpRl<`^J3f7ZmOQt5Q;eO{fnt+(oqSI`Z~Zr1{cd(SvXBBtrp`
zjLP7Mgn8{uiKwK7PgjGD8(=qjbibfmLpWI#6Ey5BS3xd`Um4Le1kF8VR|C3gE@^q$
zLJSsMv_Z2w+EjZG`NL^I_m_4ih|XEV^$0u8pa*m$%M2SaVpP0(2`hpoU*49|5<zLX
zW~o@NdIhjo^6kr%^q_cf>;(*B6MzPKP2h5a8{iW<9@<VP040e!+fiC0M66jasa9D4
zdg8t3k6e(5YV3m=xad4rKQ2@Q<uxmJg?c4+rKVQ_yHuj2r3PF5H{}2lz_pvR-ddCa
z6g$xcb+d7ncQ#|y7Fru9rrMgj;`liaVs}6SExVj2yCnuHqBlmd5<cFsA}%)7I=aDz
zgLE!GUKfaO4I1|r<H0V9CWyGEpDXOfro%;ebg>(~N<nfwRcRe$oZc<VpyFssKHSK3
zJW}#ri-c5jx{7DWS>y?2=-r+aYY4A^lq(k#*+i}+dQN;ZTPdAEgaKV|u^3+P#K!@a
zLtIXmIuE=^FvH@>(AwQ~o$x!=)FC*oGrZd=f<EiBKh4nWF^(N81axo^q<7jGoB&&z
zD#_<?x3r1WYhDVXPnBYFiC7aIl`FT$Jfvt-#KbdP!E%Uh9S&(bd*K@6*LXqRAKuty
z!<jEp=~6+&m#8{L*T14HxPDVn0JYv34=n4!My7=B9!2BP!T1sPGKIwJRqgRTwfIvl
zs|&U3okVE&ftukTNqH}y7wNmrL{0uh#6t*e)j?2NRnA0f%j$m?|6B`1Lg^w6ITGKl
z4cWlb^t=sj1etTij&moqR5qx!0&P8~FHt5lYItkZN})il=1{r6)?RT*kGU-j)W=^z
z%|+|<{Sgzrz8JENlvHAWbrrCG*FfD|=U!q0yR5zVK#|CFV4j&sHKneeAy~JdVAz>(
zE4^(GEN0py5HlgHYc)`?mZ`-qzxp~zt(`#Y&@=FcSxz?MN`YRTuzcK2cKLVLnOj84
z<Y9$(EkKXf`md-izi^R5e*At(EXDoo$7&jXRsL_6k1r=i74TI}zvS8<KT8}xcc3@j
zJR_WRygrfyufbKzsO5n3Y+34ny+x!<!)j{$>y$uI`R8N)T(mzc?a#jX6Kwv3&i|o&
zlTtQzhE^#h7cW0Q|Gj>Lx2{<2NW@}q=G@OZslc<RrGiM7ottYDuPyWf(QDw&!BdV5
zNo_Ad*VTzallEvMy;o{M08ojc*xOoOEXxV<l4z;7HaA}|RPXcPVcqq)p;xqh4!N?8
zIufZHqpFcGUW`mu0*ujUOrJzN#sW}A!<pX3BNp$!ci>@yZX8~tGLb!0VUzU`v6uto
z)Sdn72dW0$DYx0&V}14W-0uaoO4w5@kkW}siH+n2H)tSvRWK)OBiOtdqV;c|E!;4#
zx^LG$Hb7N@9=5Cka>rji-`i?U1JL~#j#fL3diED2^ZW}>=z<5NvP>IkhaF!y5u2^g
zs)T>89wZKo<>c#PsIzxu!hte302C?x&adk1*<aOmvV?^mT*UT0miKer*0B#t8yGU$
zGVDR^&|K2DjAXAT4M7LVgN9nq7LbQF46V`I{@Z0hlJqH${FlAK)dLS=&p=c5W6Y*%
zXe<&h99+lU^i|$*P2;w3?t?e`!<LAEGRuM4Sq}v%VOBEq>;-;{$H?1&>f4sH54c}M
zHKmJ#L4;uxTs`Zq|A)PUGEBeD-ASX&LKP`Yggh7Q!}ZJB@ez_v+tC!Hj}X2DqT09P
z+2qY8Ue&}Z@6|wimbGG0Q3TqvWuUJbT-ofac|Ug+rGoJPP4!{V4j?^~il-FE%O79K
z%Z!b(SHHpwmOn4~bH)D07EJdvULM$lfietUH-7-Lz-RtASbsxEp^=h}+Yd|t3}ZX3
zg3<NfhwI4$9j-v59{BP3sUWef!7H?08{Z6f*Jhjsvhm;nboNZC!1{&{cv-U)sn}zS
zCZLZ+*~rSt$w9r3R+oyB80F!t?IDXUylW+6OOjMUH(J~r0BT*GM>C*9*DkVxSw$pM
zT=B$M8bb_KH!IaaaOKvEc*+OgQ~{Ev5o9()#mt^S21yc1#hoK$xZBiBDv`kBjS1C1
zs}6MW)jF1jAgWE44d@X?h;3;=oC4A?l0_j$qZ|fSF#-$OSG+tS`aZrkTJpn5hQs$P
zQ2;~^-090=Cn1rr9xlD7nn@48CM>6)>=90SjIAfFZu=Z<k9wN}T>}ZQtC>_J51fW?
zdm7$X=Dw8iSfHg8J4-86h~c(PZH0_$M~B!T{R5Qxr0)rmr^cjkG#WpI<<-y?|I;m?
zIi>|#*zo`m7@2c>c0aQw)4s7Q?36g#e+I~j%n=Ljjl}`pvw++<d-3{J`PgerX90fI
zFSQJ_a4<f^@%wrr?`a`uo+}>kdG4$_9<0=$=^^;fQ4@Z5d?hJ9?Vr2z^>|AN?hxwS
ze6&d6u;qlRx-KQ~E2s{OI-$OAIhVluyA*1ooBWN#pqF>&tCRapsUqEY#lMVgC<xXW
zrkDe8BIBFLZ<Yi%o`RQuVLkzmZuL#k!<}ge`!PJ&+l0iaz?;-t4Y-VqjDsb(0ri!>
zpALXN%dJjBc7j+j4`-;2Z<_6hzjT>uksmHHq0mlQz*Vd5MFn38|E1$CVlIP!OGa|P
zj0iu_w#%GGTwl<Nf@wP31EC$U_gTsNy!U3VzV($MgnmO|8@-4Bqttr=;JgFGz2cHe
ztb6Zq!l()SY8hxVEBWm8hU8kAo1fO-pQ3KglsH&`ypbW+aPcs{d#0qyi9Co(QjKuY
z)de<{kF2LMuEM}MTs)t1Plrp~AX*IqABRt$qH}z;%F*`_21SwG$B&~j5Zjq}ZV0?e
z4@8<o#&<hQ^WeKnlJVSAE#JNh+wnBw^j*JG^cuh0`(Sl}zv8!>Y6+ot1>6{CG&fxC
zLKcjtKvc-05Jv%2JuzENS0Y5;L{q{q5aCtm@20N!Gwd{;dgJU5@MnQM{RZ*h#U@6w
zli+(GBa?w2%@D-0q6<_!Iu}ek;#4`X@I?r6diQr6%gjXQGvgqTzA5*=c*E2m5<}+!
z<U}_U{`I?SPHM$_?Y1%actTM~X!E;@Tm+dOFaS(q$B(>)FfHU#XDc%l6KMdTM|Cfp
zc4#4kgig4$w@v6c!^0@asE?;WTxi;vaFL2j<Kg)iOns1DPI(t{BPJg#nJq>S7n{h9
zI%UCMno<S5)TU;+{F=+kkJ&RqBJj{!a57GqK&`;!H&$szYCMIy6S-w&{7`(MCZo9x
z`-oRcei;NBE~8HB;;N+~u>H#@h)2)#RYCnJA^d~2!PYK+oCotOl6?vTq9{S#_e}7p
zV-1j~VdWyYs~dcf0$qz209m2<-+tD(;b5yh&xR776&)j>Z9%Rm#02b?f+d`z;NqD&
z<^INI>wzCfW8l=qEU?W5<rCMn&#cXrcGFG3izqT0+Yv7qlr9$&2&D8ZdAc<MFX==C
z<l_an0SKAW>%@HO-+0}=fp2{;G`g?C2pRIM^`e<S2<ngrAx-5MQ`8JOo&a0T;IE;y
zdlPgO&*t9;1p@cNtrq)y6)LBGcN;7$Kvi=<>vJyAg#^^Z=*J@id73K1vw&72z9eXq
zC*<7Ze|qvpa`T%o=D%kF?8-m2wnmvm4FjiY1|;8q+mst}ufLG&1mc3C7+_beSL%TK
z<=J$YV4$nZfO`E})HS1)V1aE;{dfPy4srgeMbG*H+NYs?fbztLUktYUVtzUq>sWf}
zlh1EfOOg+`_5spL+#E!TwjbT3zrgUVW@jq288Drr3pqxTF8!Kj>65C!``QczjB1Zu
zrDXAR%X(n>yHgu)0UR2WZ&ege1N)&lRLs}R0ke#=lIe-#D=XzAwCGV)(HE9z0d1R(
z5W28`PKNdzgxx>GE>{GR`s_4M241qs%G~2&z)pqo2J-sswoeK+>?*{dVu9^5IJxse
z=6}=TX~qAf#bZOJE&q!aH;H$na<Xwa4-eTFPj7tA8nelVOygXmZv4Q<-s3i{tD&yz
zq*_tV<UM8RnTM8Q*n1e$oEycdu7+<rm*Ai%?HZ-i4V+@+AR0w+1p>~Nvv%zZ=YqW(
zQ>bwJqB`7R%H4ehFE5xv;vKq$TuHtTazkaz<Pl-MA%RGtZiCI3Hkm^a5PO3eo$`D*
zp)RUObKw0-_ZuD&@do}A60~cZh;00{hJoOyqe^^@=_9jv*5WtExALpkk`bp1KPY)R
zUVAsS^AHqTa#T0s+e8pkZbz#@5@~b6|3Q@ZOouIep*b_N95$_o<U<I@k%;e1I69oo
znHtpY0}EI!x%v9>JH{yI7f$|$t1IF5D8x}q&#$jGlbSt?w$9Z7e7U&^ZOr0)Y6J<$
zH7-qy`bT>qblO^49MQSdMr2e{6;FnNQX<l*W_^dv@H#hC4Xl5?0erMc(utI;g|Of}
zPvF-?GBp;hU}<R?tdAkzn|e$)U+N@(r5`!LL2{V^-4X|UJ25$)e;Y&TMIbWhhaQvn
z+C-z@iDqE(maA#O9c_<j|IoQLYy_j37&-muDGi`qWbR7&vD{@q<;uh9TLpDn!`Hm`
zcdBzdAeZR0apH7qm|TnqN6P^~>2-#g3Djo5^@$$P*A4$Td9XdMUspXkBeZh}tZ5^o
zNfpGSn1k=S1GH^VL-p&tWcO|ojEja~N6D}VcgXh{?4xsyQiXA$W{0gVxdXJ|9{J&3
zxk@9wpO5x{J?|j)=n$0^&?gn;TDN>Z_rgdmg8QC?H6m&TQS!NdY=GrFlNGkWXR*8~
z3Rs<0zRg~E<=FOE&Xtm3Zt<xW=%cIMK%Pt&`P@TEM={09%Yv&0%&3@Gz}=q0p00Ff
z|C5Y(&o>^*5#V-11h@c0E9K%^?W_GeQF8+Q!{v0&gG`VZxPR1R-y37r`Mm@m%xUN=
z;mL~8QvlXF;5o0`KzYoXkb(lX1;Q})t*9DE+YyY1J$Xv)(jnSg_2rdbAhOpXMmc(o
zTN~p5xwQjT;Q8uXF+JL2YtbPyqP%Z|6aco6Rh2#Cb0AC4EKdULNNF~|Bcdp_Nb+`G
zcp&QcEPx}6wJE}TBlX$tOiWN;u6ibvF|{EEyJx1B`U^S<W#L-}_G3u<*WbIF9G^ou
z1@uTaArDq!_17IXJz7f;S{3N%x8HSnW;8Fmjc6PO8?No!b6a!$a(#9H*XQsPhT|QE
z<du^k>mX&uzM$hQz6T%;xV$eLyw2D$|Lb!pFm|O?j=HVA0IW8Oj~UpsQwflBv$jB2
z6YiWnNR`~>4OLP7Gj4_+@6D4&^hV1ev$o?bHg9hyshBpIHNsdtPowBB89xzzaal1C
z(L*?0!Ize*RT%&P&d$-$#K-4y5;2&$-p4wAJ1;)Z_!xw`1gVen>HWxE^x}><GGj3=
zq5u|1^kVk1zMUt7-eFBsAn_$xr3$N~`yM}48?nQ+Tk><meSj|xrxPi6pdNiu7Z1|t
zheZBmnj@8iOrO2XXGlu*$F-U{3ll78it#K~6*so1JWE@>dRr&oDCxxIvMU7<$>C6C
z>{u#?EC<b1`os6Pcj>Gtsg)Zc@F{S|WG`3pEUOy@un5*~sVfyJc_`28czKC=-8#)M
zy;zXkk<>R*WV>l@hb<L-`8@);+%~v|+Ah<x4INFM$=|Lo*GY9XL0gKflvs~%y43ot
zl(#m}E|vBwPBI)}7@A+OCS9mrj68LTm=6pNDg6TCbMy*r)}iZ&V^DF^+Dv<IItmBa
zYkMQvVHIoj1E^Bx1$`;c5hV{xq+Q|qn$qMVONqSV4Y#$QrDL&iZ8eco*4rcTdY~D1
zj6rE{7H8cMdk3dx<Dm)!X@*enp^O!>Gg>%TyR+|>kOZbQEE8ye3Ox6Mlogm(vTtc}
zv~3GwYeFOZJuczzIX2%)_(0$X4AOzR&O^+j>;KATtI(F(0=w&iJr}50kf*ue-{&^q
zpNT(}B~3wQewRAn2JNx;2mm@Zk<rrsSMOz1*)vv@1mL85>gKwV8D={Z<9XDnskpT$
zRw}2%9XE^G=;Nc?AtRz*13WQU`elUs^^GBs5gK~6u6EEo2G@;mGIHynRgb2B5qcu3
zVZ{l-=~cIEZA%0R=(1*d_l8B(llqlF7YzBLfB6*oEn8n+oCKMk#MXN~o-=M2WJ-Ui
zms*f%X{$kOff)!PUT*@pQ1U1NSbORiB)Ni@r+>H>1IY$?0@TWImcXe`dZ9h^g%<Nl
zn%=UjSlqfgfK71hc`ymyirCI~3;3`Dep$1f2MrAYYt8G0ngW8crsaxpdZ2~VtPU3<
zJA7xBG&IQ%OwK{9rLT`C+-^y~ANU$w5=Wjp6+YOC)l`OqER`5AHKQw_ou`Hi8aprv
zAX=^f&BPo=Gm*BBgVRnr&pw_^uj!h~Jl`r({HCV(OZ>>c*fXOMV;Kq3i6aRuXJAq1
z9$wXj6N?ZH5ALf+W^=B6q-$FbXjTq;uDHxBhHB6Y+UP*7vJfgO^-gVbj=KeDLLq_G
z&i|HgvwK6Na7sv>&sD3)NE(@rd^M|J>#j2BHka6}D6;%`FGNIi_h)1}rfBMdCaLJE
z<S+2X4r*`YjPKEb%8ha{zUUdy$pq~5MX`_RI)T#pZGAMCrm}mQ{eQYX(`A^abyX7+
zL0l}cE|fT4Wu%ngDUcQ_3QhX=xGK#M`v63Ru6`q(H$bo+6E>4u!7c#=7?VJ&bQY@6
zf+L2J>?AloK*AQV!|pAzRM;%gMqwvS8!YC{Zx2e(d*w>H#~T#CyD8ZbWQY565~L1u
zoeIO}f%c0U0;t}$dJOBiQj*4jT?|qEa>mlgn$kDs3q9$-#*a{Sz>%XfTu@aHczjAn
z2q$U=HSD9Jg6M#q*hxr>(UZBsjBWGlRyckVxOhlv&{mh(O}9P+rDxh-e3dAa>!YGK
zz#Wg+5HeIY^z>hy3}R6H?T2!7m*Q9g3pC}4iiSBcbas4UbvWo_YgdH97#Y|a)$z+X
zsbQCpv$g<GEH?cmWw(k^XChGHD@BAQKYj`m*pji8OZMXB0W@Fjlft{}=|)wYM9{bl
z;MrYx56yP^jsAs<EdzRuYsKT;4KVH|iq|=tR;maRe73CswW#@TV5^kaUzmYX5^W#}
zU0lw=ic5oByDm4RY-J)LNtT~(i%<%gg<rUnl;FZq)&ZeR#M@;cMbXo!vNx&z5!D1O
zk&tE*@CG3!MDW)k$)8&II(6SZm2QOj)&R6He7Y-ElOdc1gwX7jp5`q;VUVHMuERPv
zLI5p+WML-0{tmKFJzfQWNSvAEZZZFLViBm#2*>wl1)D*qpVUuIl)U#nljC;~355#u
zCeV}(Ftvj7Cl8P;{nF47@z!1w1!X7%i&Bn622ksJ@_7QW6DhREg}{DmhIDcY=`!R1
z_u2U!{+CjTm)~Kt2N}(uK@RXWB_<uPfp>n)iL38HW#_`9Mo{JiK%-lbq0~W;!W)1h
z)g2iT7x*sg09@dGS=mBNzbQ8FDs1`8CrI|HBtcne#I`{U3f6{Tn$k$AxcKEA5TM`I
zmk()_T^_4)+U!$TKTD!t7Ak!_ZsGX-b^7M!xuzDB6T(Bc5=@C`0<uC38M+d+1a;3n
zSAu@Yiblq^$~)<8eBiIACHi@}F+?0d&%`nM-fmEKX=cw%IPW~_G(a-6w~f}PIYM1w
z*u4)aswyY`B5DFXfQV@8J@{KqoxXW+uF0Z!=fZ@~ew)u*H3`s&Oum+(XZPYT+<(%v
zQ#lE4jq;fa0Q_ia6^M_)@|~|_*ZEJnzN(!f<~?a-4KO7RE?%8FqhJJ?n#)H3hUKcO
za`9>wr`_D~p#3=t$2q$lPTGS$1gz%C_2eT@NFE8vt$w`Uv-QD_tU_6Zi3I1U>uA4g
z5=8!`XFiwUQ7DA{RVWTLnP?Jdp*T;f>+dvC-#J?hps4Qm^I)Yv8a@Q`lLEne9GwlK
zcYkWmxDAjI7uSTa-)xifPu+hN($AKP+5#bw*&QBl8bg`h57MC=0&QeVA3}M~Kw}@I
z&canlxOiex0EX=cN%NxjUT?BZ>MqEI)+Gg=NT5Nxv<22scc#(|^AG0{Yo>m4c<7nG
zh!?e>B&gr}YZe4(szXZX9lLK&f-xj+l?xg19R2FNo=e6IL=8s==w`5Y$P|Sa(n2~M
zX1|^W8z@)9433XOfrhwn%2j<P`LZi$7^4Y{k7^pQHQmF*T9Naye8x=`^_%(6z*-6b
zI`|$wJK8*`DbJW*hTpDNwh)wdv@UTJiCStnI-CTpu6xHvRriS_$}#NNb|8dV>52g$
z4;p&++uiyr>;WvE5|Ye4UItSxl+Y=%>dwWhRb5^MQ^5c(R-$bG4OSe&3c~&(po$RP
z?!ED2kBfdp1$>TV4FbS)dQfMwgakJytS%%3VaFU-jn=cylLjcTca37&0ofuOF*?S2
zF++p{{`WuCz~A?`yddX%NDs7sAu|PBEx41^JUX%T)j$B5371pi_!*tK$=w(y>|_xz
z2+Ajp=$hhHAfEOI<yCr+BCwJC+w<xNE{9)W!DH*Kr6DIDiwTn$-ROQnfR-16Fp<&z
z8yp-7!O@ge;#)-#=@!t8GKPbBnckUBD$o*>7gVEiHVR$0A%N;P%Fkb?%HJ5`rwVWd
zj%^6-f+T#CP@)~MMyx`c1XBJ%se7RS<UFVxRgC!!kR8HV$S-Em#Q&?v1h<eB+HyT5
zZDfnBmu%0r(K^Vpe~r6tA{(p^=`g`Sq0_weJ4Vp_Iaz%`>)7th3N~C+%hPyG&p?!!
z?CtIfYzCS|ov_ZpSCS8EHKmIoG#eCp+mo~Lwu&bqJBK%rV#d_JiNe@5x>l&|GA@5E
zU)u#^f-ge#QWQcN211ds6-ufwFuk<F*fl^C=4!X=(95svKm$BNi~uU|LNfFO4X_vl
zISt*1RS^$q{W4BzytV*C<tD2a3y22veY~J{Cv$Tuleo2w#1?epg`nnj+;ctkn^mQr
zvJgQQ-3w8flVWH$rApAC$hs&WOi?!wLQG-hI>%TMKM1JG4=I~-eicCr0SDM^UWDws
z;{}amN)X{sj>eD-yS!Xr>8RY`2lHLduY-oKxhGw&83bLZT>W|OBmG*(+;tcUQAYTa
zr_mz!8|GZ|j>&H#kQY>C=WZ9({ALiO7Eni)$qq%pPl+AkzzRj!j~BGQlKJ3>R{9d0
z=Ij0s)pX)p_`9hAt!2J4d`uu2@RH!<vM(vZMinwX%mqi(*x|r4;Z0ze>Z5RGWrij9
zEKn}0=Dt0ET_gNBppdXQQxgf}#|Xbe{u|XV-tf$!06CijMjycQlLgR$_Hmd<z|(OC
zn5-Gsw2}sYx3V*Yg%(l84Dbu80_58YhAztRcl*ES_H+-wEfHq-YaoP1w4DXC)HNy)
zWfE##fLNXo)b_kDz0`=8M=ETyQ4nbf{1s1}Xf{o`x_f$xJ1zn?tExkoGvvyveXG_Z
z)_W23x(R`n3(hAANa<<8fY!><gC)Qk0ZdXNes?KVOFIf^h_Dsq<?eL~*Uktl?S01%
z|K&V!#9y=TBv5r8ZAt!u3^J+!(GnxU9ZXE*2YE5}#*h{1;;uc9tM%qLHo0)e$Hh`5
zLbs0Yc0KIc;;0=nT^va)^GIA)Twdy$*%a`YUJ-siFtqY!rGDMDZq#DAqCB5Uc9aVU
zssal=ch_vhEeDh;Kpk7akWg|Z6&ua1ZQq}>Vyn1Mr0UsR4u}x41_sFEVf#G?ouUoK
z0iZEP&Su$t7?MO#`M&k$neNKWuEfLGjITAYAi(rWKs?r0Qh#sPrbn;dUrFNa@u<`r
zv$obdOu7A><1n@g^k#(?eK`W#WyKf3lv%3O<nn<p(C9D_14nAU^4L-vIKAui+dR`E
z`4p(MHr0Bq+qTmhSr73h5jWmbxVPYNdNUO0)|8Lb$0TGG;qfhHkrW#FY^8Nq6&dX!
zxDVgZp2!;v!pqy=R+r(<^VShNERsxiz8a0T1{As639f@!bp5d2Vh(y9<kgboj^+0=
zo*tgOKfSCwMY02DH-Sh?lR!hv>|ruZ{lz8t;DU#uAdF!4<U?UQvJRROnOi-AgzfJT
zR>npSy<~G3nV+H#XgzmTml}=Z`%V%AHY9TK1~)9@Y;b@RK%i`>Ve<*tl0dX_^~yjV
z-}rudU~&RVb#&PXcG5^NpO+HRRSIJPyK}(_)aBIZavcvAK)8aJH?{eW3Pv%ygP{(&
z>P{7*u&79(1cG9%crPscOCrGF>&=`M=c0d}^u$qx3+t-A>gHsl65~}XCM%}AbFb*0
zkZ1js_ZMS-NU+AobPeje58vu8F!<jw|CXYbb2-HGVX4tKqetf`$jGE0322&(H=iH6
zL-l4mdE9!=-E8b@t=qeIf;VjI=hofc)qteF{q^1UY>Y!Evt-H<K7QdIqVp~6GF*Fz
zho^Bub<`&MctYv+0Y?th&17Md*DkQ(;^z6~*a*Vz(7XgVa+rIA8Fp9CzV*fFmsG8Y
zhh4gjWM3S$qrP|NVV9dgRuacDC1<1qw&3=ZByn<&Q1UNe_xl7hqc8u%zGYL`kposn
z-##CwjFx}i2r$6g?>t!}O%}?z1a{<*oU9SmAgltrt&NhW2@QxJJJo2y7WNT1vZ<fQ
z4%-X{Dc~j$<W!HsSKOc5gZJn5kp0;`zLNik?w-aLziMs4_|uSe5{!uC6a&*C&B547
zbD+a82Oxn0aICYykb(JIAY3NpzIn@2_4Ca}zO9$#W9AXVAi5sVuaB=LfJxYyU_OD6
zE&%@*z`!Gfkn@m9i~{sfm%YI@@n|DWSSPXpN_{oN_q*?DAq5=phh?78zNhC2d~`JQ
zH$AqNxP&$V13w^Y1JGy9i6xLE9zYW#p+3B0+6so`y#!BZ=+_7pw5ub$Ty~=l;M)ta
z#^DcUCi>pU@w|g(ewPA)T?TWs#^nRZR4@4BnN$e9mdg)28R7la=3Zz9f|{;VpEH$|
z9xS*bLD<#eD(_4z1=9*P8c&e^qz2x#8$A}e`O%00Z4H~M8Pu{Y5Wnbw-m#|m0Ky<L
z6YOQF8aa|zJ6-B;j5_uzEC57T;oMTlVlv>)7SbXWIT)s*V5y8BT<3ys1NG1FBY++(
zc3H>tok<T6gU7PS$3DbPu$0%YHW4kzNbEiG{s|^Ksh_LmKo0_uOD}+3d%>^2S@i-j
zWK_7&RB$QKY{ZaaV7=6QNaK0ks+`b%O33rM?sdtw)=utY?k6@D{|G(R*FXZ3X!T*s
zD!o-g+p0(!l1vthL8A3NOoWflG+R8ZoigSGnXG*5ir#XU&!L02Jbs~;g5a2|agy{`
z4@w6;epyISQ7`hBK4^u%48Zku)lA(&ao@tu6Yubcl2=yHG}6$}d|@%+jf6Px8Duqr
z?`>fMq(z|z`0QGaIxLZ?<A3$*qud&%4)ayXr`LdZ4^r_R13=jX^~iGOzZC%b<eNYl
zI{c2JycGjM74;^7joY=a3vgYV=#k@1oQivGq*!3unH!V547>emC&4$Ow3^<9q=~?z
z8&gaR=G6|B{(K{s0Zmp0KbtfMt|22&gHN5<fM$HJh-xZ{?>^D1jwSTo+gR9jPYn>_
z!83h*nUDnrz*n<_bgyubV@VxE+IkgrE`>wlbs?&oq8ZyMyqLEEO&Umi)^wWxxyp=z
zMgpV+xoKzQ&{sDV!a0kZXGB_ny8al^MotQi7*Mw@s5u1S%Ps4|j|5KJ+r#0nXGs^!
z)&w}Z*4lZK>sV>%ZV+O>IMFEW>sVx`6HG?8&4}GAY!K$I_+iiPUM73Meht7Bc@r29
zPH&$F(=rA?r*oyV?Uwh7^Zw`q%(~jh5_&KHP>;+hR}ovIc%VBnzNyaRg>fja?HDgJ
zdGR(b^y&Vls~jB#8$sc0S8}p0yLhnCCj&Q|9I)7UM*%Xr8+p?B%~k{e#TNjy)D?J0
z8LEEP0Lk#OBV&y^NT^=)ZQ(nIY4>q3xg+k@O%axrY)^fK<nPE6?F|&w8vf<F!gv~(
zD+dV4v++@_F_QZqV)pbI>kW4EP)F`uW4HIkuE#Yz2$|Db;l{NtsX(Rn^|Zv8It}L-
z*(RXT!(&-N&e2WdWj=irjeEJ$12G}Ed2C++_MEU;kaIZx>H>``;PuiVFiP_sG=A3+
zgrp20^RAtu52?$%x87#N=er3mrut9wfaBZuT0dUW9ijdy;(`jmw`2mOTQnb_yyjrb
zKY%Eej)Gb~nLcQ%s%cwC&20g#n#8Er_UNuV7e!E43#~>+##9k<#<a@?_5_ubL_Bwc
zGL;f!Ewr~Y9D*Br9?8%}k=*+E$|@<3P;C@Q@Xb@h1iq}ZIWr^*0wsf2TEN;WQpo1!
zC}rxRT?si5bbRB?8>7`PY8X)FV;r<Jxn%9uRF}LY<Kfdm&(M4)m780uH0$!j$kJ~3
zBAp*<7q7O#Bh!W|^XfOARIk=Kn^L80pOasbOJ+n?-L#|U#|PHt^1Fr(^PAb+b{)z+
zsl{#E&mUm54r@1WjWjrE@Ib$<9kWfY=$qjkml4rBSvdTDFkQ4?+qTZiI}aMv%=r<R
z(X}7FaeZTWi}JbhjP<5oXj?doFFz2!TPg3l@2c7QYz>)}e4fv(*haZGcs;vm52>Q=
z@i%Cu*6z9K*P0oJCjzuBT!umY3I3wGmU()J2n0mpn{(dXvt$nJY}4VwB=eS-b*&(p
zIIRO30Sc;DMBjUfVs1Wf&^&$a<dh$4Zb7w$TBmEutM3hXnC=YMV1--U8ikBVv?BPz
z`?}!EsB`VQ?$kJk@%OWhDQm|NGg)!Y4VrD`sMm2c%mbxVGH=@*yUVSvcMM97Lh1Bp
z+s$v{=B?e*gBAD76l9GSnWofhS|(AdYq@GlmGxko|HOXSez@Uk!GgSsnW1>?2_N91
z01|@@YM^op%;;C)^L;%1DntQ_NA-G3VU(8!v@G)}bnDs|zR56f^Y(j%y{7@3!RKw;
z@*&gRV)%>zK~UgVzt@|NgLvj)pc3fS^H`1|5t{Jb0OtDvA&Z?1Fys2#xR7z{OV3+W
zMr)<S0)vd>&g&h5bM2c{AmK%Os()oy5p$8Q?}fp-1~0+ew-UOiwOiHi&JH~(o0;~;
zj*LfBFV>HNsR!9Y=sqfKE54;cFff3B+!@S)nFVhO&Bpc|0E;7$_c!x5<Uk*#y{0-4
zR+Z5YUgo*zQoo+~SS(YuzNC~aqB`-GgXtE{2uSPAkI$+_ML@f%J#+zN&iezpF1q!4
zbz~n4p$D9ddjiY<>~9E&QIO<O(Vk(8^k6K)Za>f!2qoIJUgUqSfgwaev#m#grGFO#
zO>2PEdG&gF8bJTh3y2?Wy^hWrpypiLTx$X!8xt<N_G!Op{Tnt%Euvrz178Ei%-8Jl
zczz0BA%Z3l0p7<(7Lt@YuR<bFpiezuUv4-$G(XrJP+FhsQsT2(g$N>dpb=X;JHDQ@
zowi&)!lRGS69Qy3^n8M3qrbMGr(S>usffp0a(2EGb+fO=v|j3+iP2Aj1U^gO!6Pct
zz`&ns;Y*9K{8m6nXy6o3cPK5f`&r3U2^SK3mOP(&;jh2@McpEa>|-WG`_y5=Z1bub
zYWY}<kH7;{k`zTTEsmh($v8<wm6p8!lXXaGH+F3??LlO5l764kO)m5{tn|2Gi7Q3e
zrse6sDE$xxl(Y}@_j+PDk{C5ax!vzC115a|WII&w$J{ly-b^JWXaxMQZNq_bf_|dW
z4y0>I<f&b4f+{&RbUA#)L7XqWhHk2@=cIM{5KAJM&NEb44=rOcaGlIS-d2paDt-{+
zmQTehjn08)e`DPM3s^iE4ED+cAhmOj1J8;GNzytG16l^af_Bfpm-B_ZF5T$*rn6EO
z{p~^|vA-(iHxAezr{V>YIloo^4Y{9xhR~plW8U+41%u-k4vskedCi||@Mj(T*$V*a
z`x6iTTmwL8{{Lbf$XoHBm?M8GZSaD5k2O%(u*qM|rYy6Qc-y-eO`%t%I2|RE`Bo&^
zwu(MfT{%-a>`<GhKf8R$XLBH{W!Use;@Ehm^_Djszjl+6ZtJaK(-QkS51lH-J`;5R
zqZ`be1I8E)r#*FI+`I-Y4H{#=^cX#1sB1Fz>_ayE0Ps!K4(s;dyWIS8EiJ<UO830S
zqnw>U*fv-fT4OJR5A19GuRTqK%keNCm%Y<4SFV8$nL<MOhzOchbY$B^_Y+T!yX>)Y
z<}De0-kq9qL@vT~@<11>9(}lQxHfQI8-`7hU0}o&D<sG&_E=V%>xhrh^TZ(aIMqhi
zxO6HDSL(e-FyB>_(yJWiwzP{z^u|=xG?}HlszDdRgTy{UDft1}N+$Fy5{g40n^S}Y
z0E4FKKKS5qB~=s>P%kBWz)5ywsY6S0u3>xb>$~0t!J8Az$v|%~=HU&d)14^zHeSIX
zh6xqZq2|(f1s?G>IxfgRk-qJ+yyJS}Wl)2qaj8<~8yvOFCo>`+8oRq6qB2t+zICZh
zdxmDkDptiT-o&F56;A)L>uGs1=ezmbsEAvD{m4?qIC<jDi@a`@@dvt*$Xg5hV@v32
zE}zb5bkm%*<_91Bb|;iWxzFM?*%kC~>!ul@p)rF8tb@U?<^q7l%$a9y@G_<qFfn0X
zOEBt$Y`z%4eKDMkGw`>gAtE&3-zthZ-b_oTw1aokEx?I^>7mJ<LZ>j;rUyR*9`J5X
zRd_mZb8JcxIE6o7=+7O(!~e77Koj_94*@0DpP&F$8GphyXyR#z<gIE)#xQJ!;4!T?
zp@*tLGjc(e&j7NVbbwn4DkO=ziH;R`94{F5bO1Qj#rf<eG8rhetAV4q_awE>D@Xww
z0zEEo2zq>7MugXrF?%%g@suTr0PxGEtb$`DEdVJy|55?77N2!+cD%d6stX2yFZ^#Z
zUoMmGka<CPHr81AmTMWrO(Z%Gvs(kC&in;)^nGABX6PXn{PF5TP(L(_lLSn3bcF2A
zfG#Q{N_TnyJ1_8jF}3uzN*#dHuJ-`QEoyWGfIB@vnf~Tv;TJ*dF=ERd;A%ZXX32?=
zWO<DVj(X{muTnG>e2eG^&<Bvz1~3z`H4Gl-u-bDMa&ik~fv3x=-UcRi?&gM2(9aRy
zoQ(By=T#}q6P@}$vsfTp7P!j?F9TG}W@rUi^_UNK+9V2Gvfxw>ACO7V+(519%mX3o
zS6<q6J@9y$upaP8+c<Ri!~hV1uQ|cY6@E5^H^WUGQ|}fG3HZH#|K3V@%RUf#@~m6Y
zise~mu1xS;+;lLEG_4i|C-f*GLa#8QLz)55Q5!&TJ4+2ck+fdNmMKXTdW_;+jqx3q
ze~0(F><mLZl6Syqa+b6OFam5bj4Gr}noR!q5+rm~F)$M3T^UZKbOlJ)8WA)}Qz2nc
z#|cq66|w-_c3(TWw<f;a$Sqlgir42P!dabTlkt!oI}?J+cEagLi@~Z<d&}jX`OEDz
z;%=8N`ogEk3V4#bg4f?uMxguGpeMxM=n{?WK$Zh)Egb-i^~=&PCTz}DSeu0eF+x@k
zb6}G<2g^<_H?LS$qy(*@H%X?%5Y`nA9@Co+Z6(yUR~ye*l@w*%-?IRWS}~fNU`95x
zwAVeQ9#B44#7++;h}(|UF8#0ezB4MSElN`n6a$J96$y$;RKx(1RYVcVNRAdtMnohh
z%K%a+h=3p&0m(TjS)o8da?V;v4n@vk_5t6U>8AVdtXXUN{bI?}x^?fl=j{FMFNhR9
z`uz<J@3Ow_<9c2FQnNtHkH4o+VI8>Ce2(7k&O%CT9R_|Dz_3;){#gvACr%l-^5@L-
z(6Xf(1t`o*PJPJud7+A1mQUYB*6CpRpq<mX@)e<;!L_%TgLUk-s|P7Y&odh{)$4po
zUF3cpD_=2aTTl_8VG^H1o7~mXxt4!1sZ>~TwZ^+QeC;ivkkNHlh3RVOkq1O~>N&a|
zgtFh%EhE;r*n5K+&df0Q{#Nz$)$^_N+ZGj@R^&XGo$1Z)uWi5FqQ)9(Kk#8VJJor-
zjW9Ra43o0iNeXriXDnui=_Tf|DSUPa;dxFwwXV_4Mt!d7i2$C1_vF`EI2_BEIpu_>
zZrEq1>JR}KF5<MH=5wH?b)=cEq9~*DSRBqpgK6%!%@+M$)AOTU&QoOUF3V@}0k7a!
z)L8#%*Opn$b~Ma#;d-pnL+(&6R~)k_{F>+ITHwtYTXmjp-Qsm;nvK2GVr%aSB23q6
zU8L8Rg?f9l^;aH_wicU$*@svx(3W{bAbQc6N^jyx^|9-1#=p-E=#GhGf8Lp-QD~cI
zcSpPM$?S!<|8YKCqDFq>YBBS!zR--rB_n=jhyG}eb!c}x#wkqUDrf^3LSY_><__0B
z%PjmyUbYAeGUC7As6TsE!Qmsh8|D0s%TGR0;$!!g3!kq3M6aSC8}-4J#euyLqKnUk
z3lc_Qw9}(1?uVpKv!1%EC)Yq%B8U6!%`0~EK=7D5^~|R(z~|eYjzBE9aqmx4YGNl^
z|Lp}w5HQ?1xCaN^w&*twub0hKalfX}o>#3(CAXnvnw;I?Z(T9t5qX<8iF07G&93i_
zJ4NWY-A>v*5(RhO`^zwB`0jSPc;IsCaPY&H`WTU(zA2qvqo_B_1`~d9pODEi!$Ddi
z(ID)xP*B1fdo|2@n5<JxdD<KLn~YZC<%9Q%e=Z#%P9Yf*7ttWtyoKLNPh#cug5B&s
z?}bHBS$4J5cGef9-gGo8kR!Mj_5J1>O)XwGk|TCZ@k(EG3oYw?5NuIoujx8?%lUn*
z>#N#Sn@>Y?%)}|9g()WH>fVTP?h2zB+;llHoc<YbE=Dh})kW9q#CBIy7>>bwGi0qy
z%b4Ne=X<?t+6=wcgu;di`<@PS?V4IN*G+dEhjmaDp2@qX1yq1O)yk@4As1|(dD_Ov
zy14AuJF62qi&w{=u)s$&NxtM^5(?tim9B9BzY=dh=OVGZ>+_8vM|i)iF*%O;td+^U
zP_oacU;V^A>wndrWRQIFjX_avQQnW8>}IPwHARdklGHPo2M$f`jQgn2iy;lbT=-)Y
z(C<-oE7XI^;X2EDCg(Gze(K@{Ppc>Ia~;PIVach>?)UmIxvoyFvu7!`P`@fR^*r=^
z($-w&Q&Gndv7;y<CGREUA=g%~Gj8&9A8QVZlaH3XcCW^><={8=UG6N6zH-!pt=FmV
zYn#Y_=J$uv!C<JsvNa_Q4eDCGPAFE(uB$jTXx^CKewgzES0FFG_qS<9!6`iN@2SOV
zAH@UB$kD?^?KM}xe|$Hf<L!oV;K(Nz_f$sTiw|LDzoh&SMrKZ5dcMI#{v8n~@BlNh
z@4q`8yuW6HysdSW6XhYmiSj$hH06h_IfNhlS2Mp{H1j(xcC!`D#%8$itS+fQ%y24n
zIKf2c408BC!Jv9AacbW-&Wiw>9JTIu%u%CgL}%RVyilEL({P(D23E@nT^=436EY07
z(oZnnE=<j7!+1_VOx-?Z+ABTQ;jeA?lk_0WW1hfpD>g8~8Ps;ei+$d5$g8{q_#O>^
ze)u-N;qM7NSG+ns3rg7g50KDA5cZvlB&T!Rh)G7L00@vY3>owM(N?@Zml}(dkKLf@
zH3lD^VK%X)+w)4!o`{fx%=#Ug6Lpjes|2@=Y`D<K=B4u(k08*@pTNX<bGMT?lsIw;
z;I1stw8SDMBdVsF^OLvnKl9E+>``eQs#yP4pAX_>uSyqK=jl&i@)#pvH4qJUKtk0J
zU?L%RcKf3=g_~Qb!m58aEU-(&<<zz^EOkHq^|sN=(N5d$=uTLe$0qroc5cH*X7doa
zs`c&kHnPcXC{;~MdUrew|Icts-6*;c9r4O-QV#k@P{ZtW-Sye|zqmN4T^%Smw&@u3
zO5jD<y-M+GejDIZcn%)f|9qj67m<o1bJsbUu{xBCV}@eEDNv{tu%n|%5L|dR>(D2+
ze!-b0%NAyb!Q)-Qk_pv-s%DwjpO&=kAA@q&`1iM$OD_+Bx@i^!#zMLfiE-7I6c`JP
zwyLMkzq^`TY|4{2^f7CBB+>iH$knm%E0ngwXZB`Y`vs%sI}rY(66VWOA@GunM4KiM
zNFx%RufL)3ul8AAPb!wlViHt+YGfHH)Jy?QsvnKbhr*r~o8}u5D@l*-1vJ%1R?3Z%
z@BbJ0jw;{DIe@K>p9E1GH4H<or>9ejmPa<X+dQ^32zrY3cKX9gLly{|7Cw)Lgfx$p
z(%cHf>`Qob=5Fq;-BfTeEsSY@xXOyvaD{W_+rBrlN8_BS+@y>nCfc@>$jN8@pPqQ(
ztgO9n8;_kHL=Z_|6Rm8wao>ly;5S7H=U#3DV-J8jq2*We7olz3clewgD!R%6-?s7C
zcLR^zbaJGNaogUv4BYp`)Py_tw!P@j+0I)|K5S>CqAx6g&Z+?p^#){}hYUDSiUi_u
zTmn0idc%POB)iT{0`nvm(Jmlg=ML)ZZbR9n5_Bcq)`iiSxNCj4^@=%_k*%hQt&4fS
z!qLO9Ety-vR<`TW3RqIeZT6qwi`ED*yTFLhLt6%8TNg|5(C=1F@-iWHLcR1s>`CI}
zud_>5mu4CvB`D+^XsKs`f~KnFvW=}8QB(qKiZEm)6a$>jbNfl?yip!k=F>@c(b-6F
zX%V8aZPqG+j;R6ZYlBus%)9;)Vs$NSbF>{TzUvQOp&kmzi8l8t<qy|9cIFYVXOQ=d
z>#-c?OANrw37&1Qb;%jv2sI*h7|56~RD8JVv_}{YkcP4L%tkQ96iwKnlHVzrKjWD&
zaw9C%31Qocsy)tUDVY@yhv33aW;cWgSShaZQXXD{uI!>rs9<)>y`01Jc32eeFbv*^
zvKnAmBIAj?3cweLfB*S$Fb>(|qN4V!T3^2}IkzVW1vKQXn2w_y3lsC2Tw|iZ%a=Qk
z47tPAAFN||Tw-zA<yQGhq5s{r@ML^JM|V$l^~aR{DNah)s4!s3csBEOcHis>)ooH|
zn;x5Wy!XTRPJAPsj~Ld43c33<!b$w&2~-XbB)pkU0-|8$Fsj)b1ZiNcF){BoI7PsY
zkVCWybEOG|QDuSBa>%3Ef+o3uk!}%ary(_9N0y}dSj!ClEjeVMk{OP?W6_~H+O2l*
zyhO<&Q23-#d<i1wR8Kpj2(~%qBbK2<yy{%W#O%9f45U~ci#r_yECu_!_0@lY>&qR$
zX4LJi({+HY+Lz=mne5xy4qr^;Th1GPEWM)b@^EY{#N1pc&Yo-0Ont~ub%^t9R{onT
zp<;GZwrTETPfYB;$#Yn9S}h$JQpIj8dT&Vfg3&`JK@OrULLXOhl)CXD>aV(-kI||R
zp=#<jTd3bzuy@4zO+X(<T}pzcOO~tEy{j^Uzh6+f&80Yi#X$owsC9X{k==?Bn4&&+
z<1OEiCqfaSK6%Bad;^!9(_vETmDrQk^|&hLa;sLRgH=vSVe9>L%aHxE;+hS76-&W&
z<I3bWYS+P>Hw)7#$#v1NMjC7}G;P#owM5}<X#|i(a4gFv*Vf#MKAoaZn#EyxqqBHT
za~YCAlq?N-Uz{(Z^YZK1e5QWIRQ6tf8obs>)W+rgDn1`U{=E%CJyTxXoQ(XmGlvU?
zWh@fxy{B%8JJowp`^s;$x~{2eKv0P(lPN3q8r3V^)%smrimTTcOUW6k>F8!`v@G#7
z?NgRrwJ!6S_X^W6-zWBnPGZLbAG}vTv(h`_Ad;Q&{N~~<N9LX1CoH+@M$K<`9?!K9
z;cq#EWnQ0j=O9t@Az0Lb^_Hs~A}LpQZ9cf5GCnUV>`<o?lL+Ot{*fl>9f{rM(cMdU
z5$lt+CW*4iJoWLf`|m{Pl3{cbAEzeRUDqhQcgxas%KaOZp6|O&W>0dss@L)@x|clN
ze}f1Ac-h@`(cSH9;+2WF;s-gH!wBAPb|Z$?d5AYH!+ZCm|1ahH(S?H{w7aDvT07t6
z2<TZz<_d9Q=}nk-CO#GfYW?W$R6?Llj;(oS(u8%v-m>#Oa!v%Q_O9~&U&^n`JlJZz
z&tz||sLuhie*}(2sd&+pm`K!q=~MiL5Go>9*sGkfBX5F#Ac%^YnRz#_j>gIJLnOTX
zT2nj88O*clU!6kCB@N&u&-4*oUne#qYfdL6M8fWHOng>DRNY+m2E|30e))|gyxd}+
zl3v}z`sl25w0_D3*CGm@&YJlYbH3=DOgXZTgUz7xFPa4l3*O@A<XWMSm1hi&g|)et
zGwUY!8YWf=OoBd!7x~7oe465i2YHQ&bXk%`v;bKt6_g=V3>uMpqsh57wjeJ^G)VPX
z-Lao^_YFiO9)@kRipgyfJ|H)KqYx)nUgi3BkJ2B7P~5d-m0fBbj@c=5*NzPJqg|&#
zsJ_rk<yU9#x(#e}K7`2Yw-~LPUa4m#7McdLDhfq-lT}*fl$YlBK)jKmRb?TV4yJsL
z@^M(t-MXP4WdEpRn|}LyFxKeKlP*jBigag^Lga6q0lUZTk{ko`9y+YR6|X+%4A5l;
zTjHB1VQZx2nI&MRl)+}@HNd3*N?f=-GlZ}GzV9?<{#fxN(>0#QS%rvFCb6Pz9OK^r
z7LI|)O2hJDGR=75dz|@mmaWD5!Zy-T>o_*>kH~;8ckDvJpa{pOd?UXOw7=YJn4inP
zN}598Xsf46?0N2bGq1uX*T-g%uAg`#jP*;FRq+lLq09Zv3!CMa%b~;f!bP1Oih7Jr
zczN>F`4GgfoxG|yOM(}xj0G$FG2@s264PJnzgt)oXJxWD?#-CWl*b@<Mi`uOGdHd!
zE)+lLSQ#4qezZ?~YpII%7*2d#BS9v<GhvmyW1=K#tTAY%TAeaqSy5(LS6PMU6riH(
zJoo-#<kovs4e!zWx|{YHCtgI6S}iWpvbzd+<-cY+HFY<@paEv`fy^pnU;QnzP^vQ)
z?S;djh^X@R*w~pUi$6g*Rri8i9%UUd(iY%mFjt63c0qBEd;R0Nf%rnY%iM?FT}{nB
zZEp3p=gma6Iu-y80d3c{xuxGk*Nqh=>GL9rCWel!e772Mio?l*_UcG@m*0%vyf%q;
zAs<6lUmK$7M5ldzDyJBU2&cSRUG_k4yy_T%nX*6f#hh1m=nkwDmfnVes^NVjXJXPj
zve?Se{#F9>Z{S(h@>Ph_BC7g1^nU-Ps^8FmSNRWBJvER;RJNt3Dnn<F%+@_LB1$fF
zzP_3s@c4iQfKG)$0%x-`VR_2<2pK&k+uA+v%CJozui~UDId;dqb;KTvA;#k=^pHYD
z#5H`MWP~c|$Jo2lS65K4D}lR*MZyscAlSOmq%YpG8|#8zws9|8$lJO1+)T$sQc!Br
zESO@6b{_ZbhS|WEVUWr2FsbnNhR*9;(|-U*fP?0AzFkQ1&hDN#GJ=>Pt}qodq_<jR
zc4qV1s$U5iy#0G^O`F#y$p7LIceZxoAQ`Uv`e=2q(&!lk&6zTqGVCPT;G)FFZ0=c)
ziZgy7tF#<AcBW16rau?bmj;4k{Z(}Yg741&sEi&e6-G7>JSIND8O13EK0g~SxZ4EJ
z+m1XYYD}m8p1*|*-KG~wAdc~#i`vgA<P{;H%WNDOI<DP${C4O1H9Dis+D$<WYB%mt
z?L$fFO*YA4bgyFfzE@gONa&xtncDwb+EO|>jP*@Tws#53K9sgK3>(lL;Lqw#cY>2p
zA0p-kr>gQCw=6r{96XY8i$%b}hVk+D6YahfeF1Z562jME6t#MMTqHK_ydFtm&-<+E
zosQyp-EJhv@hlJI71R3jy{;Rd294hMZoaq$$G9ouIJUOIk9DEnGtb;~=NrK`pHOR`
zkG(3~o)m(+7C)Wikc!tQOB>;6m;L2RhCDaTbP^s<a=Kj$0h`C+vljMuG(G-u;2(cN
zb{Y1(j?Fz>SCS~>xX!o0QJ17iV{;&)1dev!wOi6lEpTAhPGfHAFl56y!usa|<1Lc$
zhpgi@JaIxkgK8>~pnNjPSE=344Va1jQGMAk2~XO7m=vJp;gq{WMfM)H<UAG$w4A}U
zMz){HYwFIdRrrcl2b6^FjGbN-sep`3o1vlv*xp$S`wnuE9^3c&0n`!<DlbEKok7k^
z1GdM-$Mgf8VwDQvxUkXCSzkN1buYf<LV@t0aIfB*m>ko&f)bFe_y$-)YdQ?`pz2nJ
zvt`ls{g;5!Km8HJJL_^RZj;XNpk$$cQW2eZX;xO1*`)fyAuKpR4b<V-kv-f-#0N)H
zNpI59SD^}8i`k3*0CusqX?oXff>*Hfv|gg%Tt|LRdqUDoMgG;zB6<G?9HH;$IG<jA
z-;vThLxS1|!0}_IABzHfgw#ImnTD|d$o$FY?=;KY?8P`3mLCB~8H>z=ehTol={<i^
zBQO|k$0&o`Jvn#3z9qN-8n@a*9P8eX&6;|wj_u#YURQ<ckrMuB%@htwIi}Ye@^7e1
z#{mUf2&#CKm9D+HkUO?5<Q=A4ZyCY+Ie~5jfwg59dLLJvUE<|R&QjEHo@<_g;nQwy
z7PsGFpjaD*fye{dDd1X*j@n-i>cJtdKF(f@;2M7DFHwJR>9U61<kvkj-4_jheh<*U
z=9u~`?E(wQS{+)<_80T250l|s^>3zn03FH8PIl@ks`xVFOh1%7!fqWorzdWNm7W&N
zmDrJJmGgK)2@6fxWgdHJJ`qLSkIPlYaFU<naR6cVsY%GpWYu<A>i;6~(|V3)W15GC
zN^(Gmfg6$hmkQJ^+mNTz?pqr~s8#Z~Xa);AFrH`KLZb=~v8LaHed*3-Y06MPe)%+|
z+}PRJKrQF2cq(+W%+FqNsUzLmFPm<`ewpY%TXKxcxOJETi&+>Gn)u8Lmt}2T#iZ-y
z>CGn!CTgQhrhA~YL;-#cND8o;V+&2(EI6H>SqTq!()Zn8uW1V4XtlSOQ{rG&rR{rK
zOieOuO#?IL=HOOJQ=C*)VM2VcH}(0+Iv(^q{cNLGlGrFq*Ku;S=+GHu)Ub7AgEgie
z(0P#L!`2$}yl!){Ha)>*XHxn*o6w@e|7ZY$0~SEiCxv8%pbYp7hS3e+y7i$}hO#i#
zWNwn3ndn~<vG>M~cwyUHtB2Qoyq!)tp8tj>X+cZz2jx)|qrh*3!lV5?=vor;Nt4Jw
zyIwm7P33zK|Gf>@)v4>LI-;}Zb_8<=l$w%o_GsMs1zlo_-LZKBY1+x&&4;UE-hT5}
z>fa4Qs>4HggR`zZW^c?!4AaQi=|8fdu*YEL7r<ePLx&Y#Ho?0fD}2@*RcEN+bd$h0
z-a84tc&AaG^zCNf8-_w_;(9yCkcD7$;O9dB*?Mo6ON`)JMWcSBZUj)F>RfzZ*#c~o
z)UIu{;M0Up`-)1?VK6ZL?rmS|*yBr9J58ayzr&H{RWp`ytEVi}te0sy{Ob1tI_`=V
z;P6Ybb5JXa@_4!VCH6jXf6aj9aUQ-i42E2l6KoFP>-(TE*lg=$n1Ir*HY&6e1V!;2
zsg|&u1cji3Y=l8(`(zOq_4^^|9MTE!3D?MyJU{gaG(!g1XqAaCDmzR8b8e5GggJ5l
zEA?AxOE0KoB~iD0OIXE`+2{MtKIsTubc1Vp>>kD%V~Jq0*}hAT@M2N(B8kQw?sU87
zhJy7_#(hd`z&f8xyRtLsbN}YM)ar4;OG24PeVKXMw}Nmz<yu59>3dz5rA>#@!Ejem
zG(A%%dJ#m*9z_SfT?V{EE)`-JJoCA!pktq@6%%7P*S4<=e8j?aW`bJp-sgo}0<*4l
zzD_e=Xw1WcT3<2w@J1tNC$kpQjFZm=xjGyf>4PhhuNG%P)zno)*4;)<qN?j`od?74
zjHr(RQ*bi!e>*y7Yx@!ElDWe1qq-zhzaplysHe`qB&;gDnaKJN<<nT>u0%%MG(9;M
ztZ%s)!Y)BZcNWsJfkRWYzlw%2%S5-#ks)O)efbJ_v3>%-HTJUr*ckiD@ck&C%hk{s
zFI2uZxDdQ~MVY-l>Xn_IB%1&SPfON?xp#lJ>O|Q;?Gu6^h02Xv-8K{<?chwa+ZErb
zLHc1EfK5&r$HVeexwrhk2u5L|gkcI+N}lvT<J3E_%K&cAnU9^=W)d{C6JRo?I8ur2
zNfZC`LFk_R-@J8Hqr{AV!!IIeuxM<$DNqMetg*kc$|fSH0kfh;=+QKOw0!wH6TxB;
zuj4bPZRzG8M9h!~jb|x209Z7$A4J{C#xH49zPDDad_y_vv#A@)slx!CC)y{rYFWuR
zlrr?lTj}QWBP-8gC})3ULd|B{m{D~uI8~E{l4PyNU;zplZh5U#J~r5i9~MlNJLEGw
z)P>~e4&VhfA_R@m7A$sAG5lph07W${1DevAx1wJG9~4|8ve4yL(RLg8j`_n%^k-_>
zTJ%^x@-C;DYG7>WR9bQFmY;sp_vAP7(1)ZCjnaImQmT(lSQreR4$yjc4uJe8U&-h$
z39ll>*nV2>@80SF%@=jm?yv4~73=L&dtcbwIfk-wlXCS%GRgJMJ`@E1$40oufN;@E
zm;88IHIRqoav$@CnyG4$4A0MyZ)n^XM0<XZO5@#w|H$eu(JrE}Ak3r*2p)~v9HXXa
z6!XxT2C1|yvpKniXC>s`Uc2+$W#hR@Z+B7ekO-Re+UC=z$f2)z8rahX6#u-|=10Ih
zuIM~>EcvBz=pT$xxhE&K_Cy1u?QV>XV{Y{CJ_ENP7MNCij+LJUAJ!3FwUiCqEe619
zRR)ztcT_!eRC%c%!Nwg`X^4`aVb6}c!uG5hqIwrNbJWpmjjabelQ%V0_5ZAiJQA>g
zCvwg^Py2>T0|1uP)>2ssz$(`>)#L-7CNamm1#Rny#l91$62li*1JV3G8&B(00VL}V
z&5o!RJEW)1AQ}M@<I#vm23K>F7$P@2{BN-7{IWyk<txDV$zSULGCpfi^Uwa#M2%b+
zIdfcx#F-P+a<50Ik|FWqyh(j|mKJYuvl!Sb8kX*^sB*H@-#3JTs~w5})P0ZT1DP#b
zx&vNUou<ypYBAf2g8K67&PTDN%UV$8FiG;xPQs!boe@$%3bNO7JAFTU*E(3MZhhkt
z`u&KXPQbw#bpv&fn>Qw6ffQ_`i=$C~gn7n~Go{Z|5&Rki5<w!3h6g;P^G`upOraN(
z{{U^9^p05Y+7(74)b_?f)FLU!{9zPvMB4k=pBHQIiydZH2lOtp0JShpHD63l5<j<q
zv@i%`4-|bVuB5bYZO`BQreStj{QU$n(PKCH`~X>Qx=R2Pzlr{RM1BBnK~=b&rzeV6
zU=_5h{w<(Cy#&Pl92#ynm2yWDc;VX}T|qj}(Uaf(gsIA=neY3tqxkc=MwzH%!U@!?
zs6X(}f*5Lp)XN8ni=P4~R}sEbpYl;AIkQHv-I0?d?@a@sx9>t1&_M-ECDHw7vS&op
zfG{Bnbv{K#ZOe$h=(LY_G{O;UMNMxB`8ZnZw4gaggf*NBEfoD*$AzNP60$*f>;bl#
z#dqLfG;>&A8dP%HyS%<o99}sSf1_kLl;~tA2q0Jk+?LRAC9PMhMX>rl!J;~)4V8Vo
z!^4>A^eHkTg@v8C1JSWJxMwcKHpZvu)Xk%;L`gLM2HKo&M!((lmH@Cg91B8|x16)<
z9z12dho{jdf><(1T}w55=U!AJhgkrY`IY}1^NlF(V<g74wTwuSlE}fn_XPB`2G{V=
zUq_y-B5xaq!eJQT%O7D^Rnjv}PKOCDK4)HK={%<U!jofyW8_odJj9>O4$HabeKs!Z
zCr@$ywTt<?m*;{KZ#Oh2G_9|A2uO>Mb8pgi0SRiuQNs`)f3atG@YKYpjB5p{w1ydr
zc3}JOsr$316q$MOu9yk*<XeD?XM?-g>hCe@ww)Wfk)aE{@ixk-W1$^P;>yJ!CXGhs
z<ok<S_NADm<7w3y#0algGxUB-KJci04lkI1*cck$SO2n}x){Ca;<dwBIE_U#7ilqs
z!_{6r)i_Y?nQ>#XF!1Ce?Kpc+ZatqTN6pHwJz{~`A7tmA;cPUWyXlu#fNphm98|V=
zf*KlkVIS`I6V3X|+lY|Bf^*5O*^?U_GCd-9m){A@qHe2KrHi>LrXsVoU<Bo~M1sP$
zc*4GU$*=W}++l`gFtXLUSUX*==p7*33p_gZPNyFGK??t8dza=PyJYL2J|q(Q*_;sV
z?$(C7FSu1HUCL1=&5_~DclVVjA7^eAg<5qqEc1$x)K||)MGIVH1JyS-?ZlSC@*Lo7
z8pqAz3nuyzLG#lLZgn#l?xD_{%_{ho7%d-*VX`XC;ApY;i=<9t;3N}slfveGns%<z
z?74un+lQ)h>)mRdC#+<OV?RpATy2r(c&l)+N`+V)V$9a5ymU-|%AsC@jshucWSX_x
zZmzkGD1{%xEx1NCnFXIM3a=y%BCl_m!f?XH1TcSJpYjgQ3uve-id0O?QJ8X<D%{N_
zeRG+c+mJR~>^HR-7;G{u1K{5pskkGhK*;gz9611feK$LqPWqxgfX2SeFn*J_5k+Ox
z6_7i-_w3)I&Xa9&W4(9d4l>*9_U#47g+raTz`ZaO@xn;98a1MW?C!FnAH^=y`7Tr_
ztII`==t$1h@Hi_}JDm1g?~1ei9+z`V6ir?N(KHV?L*URuMho~H6&1{o#_UN?CKh1T
zJ`_wuLp=XyZ#SFj#=+TTYz)uP2YYJ^n79<dyisYEoqK^L$yUou7v^QSg*gz?hXy&z
z;q7C17KV|tGpeXD+2h7ZoywgSzcwzs?g<1S_z@?WPjJzVrrWcx9pEr(ofU`@m^2hy
zQAf~9X3r<qP`Yr1DIJ^*vk0C3mk*8}J=*oG86OsYNj`|nb_G{yIdqSdqzOzShSl1t
zcYg4}Cj3NK?|NB&p6-FuTZhWpSxVg_Y|z#8ZuSOJBDXT^qbnHz4``<LC9639Ns%YQ
ztR|9>7KpsMjlpsKXHl1bt;d&Vm?4=1M33K>V#4U!?Ro$36wSJBwsocMn4vodsuKJ^
zzE)_F<n7`vN|UfZiSN=vTVXh6NFko>I_{SY=|jZ$Sq&L0X%It9IfjF&9s*`Kxj&vK
z1Qq5nII(L|BF!0*^7wJ&%|p9`<DuyT0k@0zv{*2eNk19<xmWSfJ8J7}Xv7R=h^s(R
zsB$a72TWsd_g+q&c0g_l4`nT#%)F_s-0%Tok8$^Gp-*zLzwG$tcc4N=)a@xUv1gJ#
z696KcE$T0=es))s%bqrwFXEgVt7_GcR2!rblH!_TdL{TOGHEKT0BF`34&Dp$U2?qm
zM<w(F?d%fj7&`-AyyKN;B*42gZ8U|(>h4J4yD!=>o*EABT{Xb3h&Y}6u3$!&+_RyU
zl}K2#F%cN>!z~y)x%oBr!ffuo^^rL`Vl{o(aVtrQ2mum*OIcoj2jON$+D_@<Hw8Fi
zEP#pU^L66)nAm(!^u}{PJxZ)Xv>$rHv9ForEObf-!o9x45CSEU4-V(tY}G)Q=R7J-
zV>i}`8#SV%xIz3%Oj|i_1(T9Y@1-Ui!=%<eIXLYQ)f%{Rc3?Ci0Q1?@Qtf_$yM9V~
zee~y8`U{$}Q{fbc-}<2%OOKl5aLK#B^<)caf;;7V7n;FR31mYMi6~n6S2^r1-3`(-
z7!q6W6LZsk1hzU=wH`D2pR3UpJocI3))3GfDZ}5LMJQ+psEDMeRpO#?-|(waTfQzh
zi;y1in5$~ZN^Vj!wy@9f(zM3$ucn85l`Qp<65()@x|V|z?mHG^|G82<YdrhX9(BAm
zp$MV>hT_8`+R@ZWKiqWu-&77Z35n+1>ZbMPsb*~+n#6a|iD%<l*{jnpV??t@gez@e
z-g-zvq)7_7U%N}`)^Vj|0|T^4@KUeJ$mJ_{_X#V9P?g?gpkO{*J;jGO!8j{Ip&HNG
znl*2U`~=k=n}^=DuQA>3=7r0iqvhv6E9xZ@LK>*%Mre(r`7}k-TUIzk#gT)+^a+|&
z-ur2n?zu_`R4$~CWmFyXcfhI~*VW-58_qo8+^^qvc_c$eqwB<rxoNJ4{4dht?jfVz
zZmL%sbN}3GTCx$cw3CvJ(s=-ba~1uP3Ath!H+whTK$a1FniDx|-taiYgehOHO`hwY
zxfs4AU9IoQaHr6OGv)Bkw+;7+cyZ46x0WwBwcT@9{zLaQQ86_fC)|;K#I1SZ_>9VB
zU#Y^MoR#z`C|9&wX@%n{&CWa@#p6l`m~Mpf2dxZ_Tev#nr+8D9FUB|ZnXr?v^qWeg
zp{aJF{fN;nHT~uk6WW|=2E^&>ic?_V(cv4g;L0yIbxJ8aZj{|1X`bUGYwd`0(q}5X
z8Gq?T=Evjx7yKPG`2@{}pvX+T!ZVPX7(kjf@}mum7O+#SdIyslEabQNYfrB0A!uzt
z+4u~R{+yEZ)RWYP8xwdL>i~WBN+4S+r57)r`if&z=9<b%HVDJf$?f8Hwydu8`eG#Y
zHPXrX(th3p$xXqh#A2xFZ^!VISi6f1>^WEd4aPyrysNx!Bw<d|KV<r;<yahHj<01o
zRsn|CUe%>amXQZj${uAZQd*R8Ie~;AIdFfRLg)tgEKA1RZkQ3iQ<&gJK~Zh1a@Gv^
z74g({KXc3FlaYOeuEm2$&G7{P3#~%@Zi%!?!=sXx*#pynd4*8Sb?bYJWJG@C&DA;(
zU}}iF_a=K_hD#@k+KDU0po-VYI{HLBLWq>&15{id{?t`He*FV+E7oh*3Glw6Y#uwa
zSv4|{9Daebe-Lz=ci385k5TuteodfaYMg;=m6ykt_fq0dIK5KC=n^r8_f8Lpak^aN
z9!pSry+hqN4uj9Bmv++;<t4etyC5Ur@9-iTfvK!6JI^tTb+jrT^1>;7v%h53>gSdj
zGYQb?bIt+!k-;Y=JNHDBAM^4%sdw~R8zw<@P{AEfP4y<gp^TdUvR2B$`(;`(<C5Jb
zs^jdA)=a$aI=udFPY(viCVeh0I6l)X@pLN0s^&KD!d+|8{yy4;%V#h=IX)UoWtS5N
z191^5KVzHO@gX`0NXMVx-3Z@2#-DSXw>)prslCYr!q)6BBdw_ZL{8~HP*CX-Bob?<
zW!W&=rW1Dei=A|G|44~yvVLVJ<{c4ZGJmY!%1dA%aP}pSQ)EHV+|V)_FrpEQlco?I
z=K5fuUqK+IVZJ^ZO^?hotNGertoF~Ge0BL)?zDa3v_{aL_~DN?c5t0h=81KXd!1<G
z#m4V-=A|Jit$xOaq-E%0eAuGvCHwMp_`*?V4KG$y{Mo?ks6kv?Yv%2SKI7PWrTFEG
z5g&v*Eu}LLoqol(h8Jh8jBn=yR?Cj@iEAq7yf@9ohMb}L?@47njrw7%D9x)A7eXbc
z@aurJ))W<#`*ulL>BRsk?;}C6`T!4(N@p%XRXL>9=yX__vUGp!QQF*QO|KL8>UTv4
zN?A=rw-!!V<7LdXy_n4L)FQY9M=R=uR!y(Skf|F#9^+yUmQQ|jkvS}B#D|wJ!6eWL
z#~sv5MxqVC0eQj%36F)bX~#=&&sl5jDtpvmM4Kq<C*bB>Ue}4qUx8g9Sfr50L807t
zwX#&_7L~9?cGA>rv4P@j4)-SmcS&2W^Qzp5o2}gt1-7#HRH4F|n60v+?|MWtkI!L0
z#QE@@d+KrP+pRiI7sBC*aj9nUvSa<7KhrWD237Z@5<Ul3eVTny(v<QW)7*M)w>U=E
zsa_|<F>Zq6h)CM9fbv4L>70&~T%~lTL3P6S0pt&eUnV3yfl&<YxQXS2ZM4-%799tS
z?s5ghQrvT|0dW&v{_EmW8JnojNqYJj#6z&S!@JQ;C?L9PN7uNGuo#Axx|!>kXE(c<
zXKo3BE`dXjNwoeJ*VO1*gy3YonF>POIYmObw4pCx3_maIQ2F`+PDdWkbZRrk03{TA
z{lqFWmK>8wOEFT^_C9=LZ1AqgZ`UYC<4<h5q^d$b369p;4sX)URFmIl_bj38)(n9p
zOv4Hy&-C1D-$h#l&T6rX>iTakH|!Bx{7NlOk?44>xmu4}i13x%)Awc&zp3T)lY@v=
z{MrKRV>o%dcY!#uc_Lq}AjQdzn+}}ZPj^n5H{J3;@bcF~F6S|moFo^fB={V(QSVc&
z^)a!c?0hV_nzESB7%-{L79TiS?q{UBO4mp@D3NfGaCcC%pq%AgZssk$`c%Ok>RtRn
z{ZlO|n1)&MW96%sy!RhPM>HYAnykoN6ZnCRgguV3OtMXymzOKIL2bQJE_maN)B11S
zJT+-7N{Z8lG}zLbb8<Hry%W8|1J#6eJgKxQjw=<0VAhk!b;jWwP+vm&A+vWA>{bU(
zuR_XV$T>iJb9%1N9l!6nGFvh-gY@?;kLMi?nyrpXq!ta0H<_5I5=*AqQV1R{M2FV=
zjBtc~uK@@lEw<8Lt0-)pc}2k?{U8b;j%fvk-7w;~He0I**KDPc3@O9wr;~iNABmj6
z^A;ssH9!}uX<3WdvAA)_j#+u-9qi2_&#Sy}jB{eN{IP%?O=}pCBy4mMQ()#J_z{WS
zMvjbvkoz@lo_y|Z?KKs~S0;QLBQrm{l{o0cxufXn5IivcY5^l%R}4sk3RBz5?5UoV
zXP4Y2AtcMCT`sBOxcT9q%S5Qts1Mn*f3mv(Mocqev50|&uYSMDc)Y4h4g_tw=IFNw
zhB@CuhPfioyD3E@oq!%-0o;tP@PtLENEgzUBrp*cixvub3V|@iNh`34OihvHYjcF9
zT>;xY;e3>p2TxFG$suz;DCuKP+9`KO4FUl{zt6Yah-hb_noW<$YGx-FhWu&d2uWe3
z$tpax8cS_AH^CzZ86ab}n%uD8EDV2<Zl=3If~ravnqr2}CB?O=8J->}^|*Hfp9L~U
zf{ENY*Kt$bDu)rj{d97fD|POocXiB{E)NXEuTL}0ynJ&8f1@3xa^g8G>`ECpTau%*
z>TfbApVxv38O>;Q(s8iRL~fRq4g`tX$dL%Mi}uvsN9D#j2abB)%jM-*f%J(m1clIA
zWKT>(yhE7CsYtsL7)F0q!**eq{@%nbL$ZmWO(-jZWfCfL_~gcZYVU$xfG?{OS&P<D
zlbNOurh;N-nyKcmZVv&9IE%yJ>tfO2Ow*~MZr?qsqiVSf>Ev2d)MdEP%PN@ZQNR@k
z90?u|5hPqB+Oog-&~El3i}E~f{3x=V!Z$;pgtPHosu^g+#N;;SEjHfE3iSI?%XM@@
zSzr<-7PP`FH$H+qHxb990m)wwA92<{)hXs%etjh%J|bqePUVtY7vD>|IzN(oOli}c
zfZmn_nakltB#7|<93^2i3;khL@H>7}n_G7w*OT}V#s6b7P?UiBa497I#OL6(J)TMm
zucd{ciTo@;v4sU4lXECzhEKH)5$~F+c{cEl9^Wd`zUxTxu4Y~tRGxLgIhdV6wsmWb
zVrxqHNM2$m>UMDE=@+)X!DhwSxElzTy0M@$at@SBrf4n|==*&LQ9+Z%&9`!pMa}|0
zzp**$(dK$4C*l1Wn`d}?R_`bW0tJUH;0~GY=EVfci$#;~#{Zeae@Vbxxu5TLqz2fS
z%3_hcBwV~s=?b_Lq3E%2IhZDAHAo>hR^ulc6$|)3LwE!WwvNT$6*l004@l3_>}E8K
zB0>NfQQr+Q=&I?3pO1X)K*XuiFpWhIIj4_yYDL1EUxbrHcFKi2cF3OYS~9&&Cp(hL
zG6FnF<9>ruWlI>j#tz@XJNwgSYW(>W(vx;nBC@z)Kyxv)cJX1IoL(OGA23FpXw-J{
zl`@wIz~tF{3x9H|`8v6s2QG%kvv-2Pu;qzdJdVmV=|)}rl^=GbcQ=(Qkg@fE3Q{`B
z%gYv~NOKYImf_4IxWeB3Lo1SdlBK1Z8aB{|*~hqyiV&^(U7TU(=tOGr{Cl7wiUl!n
z3}G7e{Fw{JO@jbO&p03cj5CK3w?6KwDla3%1rMY1(^6^k+giYjgy(8;Cgow3Txd>p
zu~wrL3M03NjLlg3>zZR&8tinS>0dRdGl%@LzRcDQAl0B|r)X#3Tst8@rL_F6=SfNo
zTQ>zND`*C9xgd~Me)!H%T|S_Yk+}u^XT@XgDT~Ma_`&7X2^O91Qdm!%Il=;TXF6hd
z(r7CD-5i>QQ(DRq0-AUwm)b7QS}iBJN$C@zP4Q~-M0z^ws+1gO(iOG|5v>j%gc?L5
z2b4d$DXEI`dU4b@s*<WYaj7vn7;zYFbiBf|Q@l+4E><&o1F3#EQ<S((n=JgI0yW}i
z6y*~L8VJbnR6nG7j!YT)%Oac=uAfWf=wl-oGalt63ov0)3K}$yqt-kA-omofXu>tA
zLpj|kRO?)|t8(Mn*Q%Q4*^yuZXJ7<0@}u)d53j)7=KL&Rw6PDTg3L2LS6n&(0p_gK
zak!4@&Y-p}8`PZ=xqD1hy(^R_6{bJ%g_$1u5KXNkn@ZD85*wnIqiQlP)oMyG#pn^^
zW*(uT$!_hdhoB|8@wX+?WO0}e0mZj*r5^{4ga?xv@6kEe^+DH9cX$dfb%RCk(hFh?
z&pY+nU^{~5EIX(3wd&eJppPpSA7Fiby`r$}&p3W5hR_2A#-V|QrpoKHiu+=rCx4>G
zSFd?`fGVN6GntzDyTd-j;bF4T$>l3siS9XmlQ;aXuz!Z8aVZ7=Pq$qv3uT-EvDEUF
z4iqZ2^|Zup#ElpR81^3iQNy%7zO}w<!_HI{xflup3h7QEa4;O?%MS{ZG3%d4eyr&)
zntql@72o<!vxzX6C~gUPj7t5I&s6MlkE9UQei_vPQiVw)J-SXD-X*dJY=e>g;!9WX
zkdAL!IVr|AP~@R^RV>Eto%USUEqKN4k));LF~i}4Jivl7B0YF#6BKazrSr(35osx+
z#4DV14}4h#7&-`M1&KkeDf~%U178{2fU`9F1c0(w@{i?I3=D+<tdD5KcUu7G#Ekvk
z+;P=x9Dm0OGrT8mhbl@L2s_E)em^<;(Ww)S6|&>$A1*?d!QcB=Og<Y`@?z*&i#M67
z0V<1F&WDj<4knTtNLU2fJ=P(OBgC7<+_Q<lMEy{>cR8?IbN22RK%W{wRpx^-Pm@%Y
zj9+OwK;@!6v)4{vB~3w#ePGz(kI$`)9-G5SuYMV%7K&je5*sLMl0sr4Rn<!_51jH0
zPxI=z<{5m`aicF;8?VIPH}e~26_qZ8;TVOD!#7^Dm>JtyO4R<uTTi&pNQtv$E~tG;
znrc5vOx3x4__ZsfVfoRqIdI&JP~EJE20&UBl>7J;syF|N6-Cf~_uT&@I)aQcK`dgv
z+n@3O`2E)Zc-0a9X2&Q8|Ns7rfByw6A4m(bkKVQU_h0<?hjE&~ww|#+TlK}(=V2%I
zAuw7JRr~B^o2S>DZ1T@wejfas=VsfIqOl3K<6$7Lao|$y->W1gKaf%qD>@g?>-ek-
z)oED!cX~p-3Me&I!Um$GBsaa1TH^iD2nq=mLMwHO;7aJ#pd!R3JMjPO`IvTv1m-}*
zG$uYzUvI0wmo-2JKF|gJtlN|*q<Jjo@5TA$J<<;NXT(X#QWro{D?B3&Jg<pBKj-A^
zmSeu7vnvu|K+?+s&AD4HugZG`FLt>oq*2KNM2AnXOcdR@IOszXNYb~tTe8$?f{bup
z#hIOA`>jXH8k$$i+wXi2@*PcHmOr-rSCB7~#7UX>@1ywlheF7XXx5yIy!7v%{QHA4
zjF**JRJPv&2jrldJ0foU&zkx7uaG8yGBn|8(l*f#`RseRYKvy~hJPR6zdziCX{$0;
zZJW4=atL`e^duaL+0M22Z*N?=u}IO59Xk)+zH#j?+(ghJDl<!oZ;eC${X|L@dHCAO
zWSDIx3s)nMw$&j~)X(ids~WFHAbHZhvTZJ%Pp?{__>2E%AJ(spUBA9Rul9b|aD4~-
OcUwaCM#^=)Xa5T)fO~BK

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/sbc-normal-normal.png b/docs/2_39/stan-users-guide/img/sbc-normal-normal.png
new file mode 100644
index 0000000000000000000000000000000000000000..56cc31f183ba1d62c9cd504c22204428b610dd45
GIT binary patch
literal 286125
zcmeFaXIzt8);_E#77(#Pl&T_v(nLB)QA7l!2vVd)>Am+VdH|&=2uLp~NbkJ_rAzN6
z^iV^Iv`|9Af9K3IXDE)(%slhHpFE%PBk{iPz4qE`UF%xcy3ZRq8F3OK3Zg@Y4v|Pa
zyf1&~&>4e6hluD-90Px`BJA=#bm-_MLs3yVV^MKYa}#qb1&im}y5hQKx>knT^5XXn
z9pd%(SJW_|QoJBkYOPOqtwq6v)6?dQFbO?Luo7o{UWxnA5Vu*V1oyzz86EUu%ZG**
zUXD_1ysVg<3p+CMLN%=L#bw4dQn*kaTL=;TK-XnPp}dV(ax=b(bKwFU;f5vI!@7$u
zE77PJ=fz@u6yZ#9ic09a3x~-gGFTwD$F02sUkft5{lhyBax7l?F*TD$@*@InSz`Nq
zwGq*8&d~nuAzW;F;R9;2%a*iSxl><=Uk4ukk)+sYci3s`9j!1@l?JUfQ^+n-&E`}z
zrrld*E#&5Js5ikp-uaqPQNieS*qOklnvDmiFzi`9&^T$Mxs;j71g%Khp}1@d)N@>{
z=~n~K71OGu*6>)NKTyxrF-%<a(_@dTrBZ@oA8el(H#T)Gpc;jTM!q{MeC+Mua71R}
zS=4>?H2Cq9ir7!-`94?NrvKnxVIF6B-O4llUEynAYD#cT)K|OeLN3+iWr3rWZ%Wi6
zmg=l9rV263)00T9h(>3}``cS(jt|kUR<V2wvo13$e}u9e-~Jr-{uNpEtJm)F5ALF;
z$8p(%l4p)wGNPE2aAytrcv_7Op<$Xh(<*fe!qY`7MLxIXAs@GMKiAjlMds6EsgM2q
zWd~S_k#n1mYg!;Ku`Zt6sUGeXB)KOE89&_eE+>*!IdWEqQRLVNd7nY@tJBABUr4`j
z$p;#*2cev<ytB-o7vB7};@f*%eLeKmW>HJtO&xEcm*hDV+W~3how`e@p6c4?h25BW
z2!yX+x_qwxRjcN03G_S5+lQ|@`rMI52UF#o*FkOdNt%dh-r?!NmickWCq{}5@#U4q
zJ=@{9`AA)%H;wspffrrl4Gf`1`F1TdTYZ+9=W)UKrRWmnmQCp9-1coW)3(v4QTGYu
zr9!JZ)uz&%M^*DCD%SRj6sE@z<UiWOYu?mQuUIZkD3d;L3vCFW&Ro^F^VFR6^n<i8
z!Y!jsrAspY_tREit$OD%movCBVr$ErUBA0FySBJ~)gX8$&=`%V7-O}dwT9NBjaDdB
zN1VwWFBa<0h8m|yUZFD~qanBY#y$4<`R!$fQzx|VJ?`@eO`pS%$Fju;QlHkQol3%}
zSFkD_euK#-Se3>pDn6ODcz=Uh$y}PZ;_i9BYoa=S@426>7b&mAUk#C+ls7!1AfRwQ
zIlJpFsdsLoU7RTaW}CV@*(^sM*`Yn`O7OIk-)xSgnap0QBUp<fllx|91aju}my2HY
zj{*!7ya!BGDG5=-tAVSLw@XRStFeklVNO+62rtcJSjPkM4uw*^@Tj9RyV&~u_%I>w
z${zzc@UPVG9=Z^ZcCF8&VOzHy+^}7<gVok)(=0FYc?6<iBPg35A(5TbL;C*ihvQ;h
z6pyGAqYfPsIV5ra?h}W@^Ft@!Kj{uX{)1G#?=Ib<#Qm@ImUq>I7|q&GQaliAyYDBX
zb^9iT_ucdC+Fi^SwY8|$J->*nw_RqvO!V4E;O@zUqF2l>Ule7(PB_t_kV*AORCzre
z$r?Aijf~<~w1r_7OV?6;r?0KA?Qo@hQBgljbM(ZS%OWoh9VYlce&{CR%qU3sZ(Irg
zAa5aO+MZqc?>yPja|8qw=S05$9~b!de}0&x=l^c4KQDIq%H_B#hd({~FIXR34;YN(
zx554mTmClKzj^e38|=5i_UDP;G3?*Wp1%$D+h9M#55Hl>&#dwbRQlUszYVrOl>A2e
zzc?^I`|7vBerA<lz`WlE`)#nFRN22vqhFkvpMCY)U_Z0U{(QFUH@^+G%NW0~(_dfx
zHrQ{2{bnov4L<a*le6o?Z?<BWG5!@>{x;ZegZ->7|IJqX%qqXAC4L+1x54&@lHbMm
zFAmJlzWQyjpIPM>xcIlhejDsJz4Lde%fIOhyUx|`;(M1d_NTF*ef8U5KeNg&Ece@B
zzYX@A-uYK3iC_HYH(T+G2=lYAejDs(R{8B=zuAi4Y{mYQ@w=A%H|vMr2K)cbU?N!Z
zjfr4c?d1s`t%e}_HRl99@qJM|trLt*^vuS#PMrKnMZDkH_kz(%SVm<t&j@+60PVSj
zfra<(Lwb?X<CRBBs|D8w*E4;`BhT-<K&o`yW3ifXpKk6FkAMGuAo%E#e-L)R8|AV*
zyHtH%R#9O2ojA@}-MU$~J<qVy2*Y?_4U%UK)Ft&@|D&53tBoqd_zP<-yiDAxyhs9l
zSxa<GWxF@`UCiUywYDz_T9#r*wuCB(_M_E(_CD}{z@%+22)#(%Z^_(QdY@47{Nn?E
zt*$T6+mjXQ61L{<@BjB-B;qWk#kl@WANq6SL|o5Z<Ti0=?@a#k<@H&*2V_G~@%`5k
zkp})?)bW_^KbY}BhEz7P!Bjh3G~G5sg0|)kjJ$l25fKr!y*;qcfd7igJe(MAnGcr*
zO>w&8UjFy5owd?3!fYrbfZF!||KQ@in$eOGotn-IkLBH?4=ld>Bqx4I>#@dTfq%sk
z`+pNBbdp&$kMT5boK#Z;mjGY7Of-LO{X0?Z^0gi*bhpR;l>PRcK@>7hX1`zUJE=b<
zhrX~8q^supq-IudMsOO+Bi^ePno;*^Iu~1bQG{IBw=fU+qenU4Cwykx|G@vkowIrf
z`R#`pXk2aX+sqfKeJ!T4X*;~sr9dj~eR_eohwI$_r!}Jc+YvN9ZFA~BxSz-~QzB-y
z0_NJl*fVV9tCCG0*f}8eSsL>N@Qmy_n$pZ>Lq48;*KHFaJj%)0M%&YS^*=Z(dm|}+
zkG?as6}gCBdb5AO{<Dj>-NHXz(@1yWf9dHMFUQ6EeenA4T#X{}#S1U5&C)HUpZeMU
z-92cmy?+t81Qz@LY^U6R=W10K&a}1Na#tSl`LFB+<43UA+gcR=ovRsq00Ej}3WZcV
z`|q0Rr>oHgi<NuuNc6vRH4+67;j6YaT$%PKmY?4If7`D(KZ3b1OXmOR;Yc2hA_5_G
zgT<(73<^do9m(y_ZK_)LfHMD0Pr8~oqIQw;X-lm0*O!Ejb9&7YV(<dfk7B-4lT^bh
z`yNY#8n{>2*|xELe+Nq<fj$Gkie0Pv%f;I|Dyzt~*!JQTAu;#aK%p&0Fp9tVfzMhA
zgyp#%sD9lyPhvzWbcnwKx0YGM4R;aT8MLgUPH@GT`qGqk1(1=E6|ca^%q#flcA|u^
zzP<Ltjy$RbCYLD92d}$($R^29Yop57{p=PBZmeA}E_3Hzhnr#4PSFb8ZsAKZx0$u7
zr--deVQ6BNp}e!PSTvV$r9FQxehIo`i%ijQx&DhlBF9FUBUrP9V5r8f2fhx2x^dB7
zcuU9u^2(P{mtLqlc93S;F1u4k;C{hKc==-aI!$&Bx!Q$!*mcS}Kga846`KY1Qw~ct
z?yr^7Rnr?f8CuuIec9?h1`5$|=r#HE=T1+cgQu&b_^hwO&C|9G2J$3K%GZ>#TZKjf
zu<Iji>w`SpLwTl?tZQ54ShN#%<w^J;rTL>b+1hnA0N82cwl|BXU)J9fwc0VP$CeJe
zgiMoXH?oe@6Bgtp-wrso+u1mE4j(2V+3Z~#+8ZSB4v4&VOZR(T(OM5&yDhGC8A7e|
zO4oxZ`GuEmV`xp*bGmVJcS<VlT0j5KQeos#b;g;}DIFCfZkXESq?q@l6b4;i9)~!Y
zQsfW^dYvv8eh@%K_B8HpcHn#|Zi1n%ZmDkFPnvsEGTS@-om+p+Gc0E0;`*@jNLWRn
zFm6JX#GsxvqGZhL@Mh&}QsM1esJB=zfW7O3EY9tI{=Ep%LzPG;{?yZ%?X2fjhzbl%
zz30&~jn#+!!!w^Gf>*#my6sGx5Xk-3i^Wi!t+Xed&hvBq_C=GSa5{|W3R&Q*(*wc*
zg6I!qw7uWwmM0q#v^x&05TTuCxNy%o$jZiQCq26@a&_B!wDRtz=xP3$7lT4NnY=$L
zvA&nI(pR%bG~L(X${J?^?<}hfx?$qfZ&l?keLcRS#L~OrSVh9)S9Kbri5uztdYtwH
z>HXxTLt?wv!g<K!0U4REHJ{y2C(v~ur2dra>V7m>l{?GzAD57+naX$hR+@&F=Ai58
z9Tf@B$EzU-P{9PwcIu}Yq;${O!Sj6;WHwuQ$Nq?E2NqEE$JYThgplnU*JWIimFhBi
z8B605>`@8`2y6pYKb+|=ZkH`@mlWeoLqYxVf#A7Jm?qcY)iropfTRA33VV`?TmKy<
z;R?*q5v?Di9wCus48J-j=0{_kj2=a~q8-_Fzjt+WCxOWEV0}@%HqhP0WxiNHDJPbJ
z*uE8ghSudryu+9$ggC$*;otk9(99c>mcCd!A2^Jvpk^*B`1`v6oDJOEP!$->r?Q1}
z83N)X75dfdFn5Eeb?ME|&~W0^Zf`959!H}Y=ewaZW60DURg0bSHumHKX&zI`I;u`V
z8R2qelZs8f<VED}=%k_#LRMO?aP*t~CD_puU5b}WHzqfyq4h%~Q=T7gjrmR|i1`{(
zp$lPH+eMjcN#ye!Ruf-lx>N04mK%}?=kg5)^<&Lz(hG+#Qkv#HmhgtUEPY-IgQ;tF
zuqU?&%xAf7uVpINu$H=TbmzCMZLUBjSYskSz+NCXrx|K<TXA$#&5=+<5k#YlO@@EA
zngY}Hd1I;07x6hvv!(S$X_oO|(b7_nTCNdQj(DJ%V=jfjjROd8*I`zvA^>sFZF`Cb
zGV$_=*M&PPN@}K;s09`&1Qv@#7mHnWoIL?f_46c#JSYg`G46@lz>LJcVb`v$g`2ms
z@IHvIg>EO<M_J3Pd2DzFBoCt7#c8JrkkBUYQsMQ%mE!W@>5m`UUU?ib_Yoz<7Je5^
z?^y@Yc{6Su)HnRXy<H7GYUbq|+bgXKLn-O@D7X9Do2`)ATFS73O!b8qVi)EMAs$bX
z8zTnIGi?TLn&-Nr)XXkj1o1$n12%N?NOxOSzg}eRS}*i`w19h(m!{;Ny|X9mf!ZgT
zP3`{U|Kxg~^YyG}oYuN~$NhN{t0x+ucNnB2sH~?e6b284IOUrSrF0~@S5Wg>$Zao&
z?VN86W82Z!zBKGlEx=l0w{R9@6l0eKW2kh%hoQ-FKaQ4wudh%NXqNn{i3anDMH6e2
zQ<tptpWdSo!p!G>VAs}r6e~nG>|*_3ajkjV7){w6wW*Zc$<$c$pkGWjSFgFHl`gU~
z&k)*Vi0qo(TvI@z;EB?6K``$lySD29Oi7dR0Oq6HS}SFwEyzMTf?DF-)7SS<!$~#A
zC{41A96z`2EYa3Y3I|eSOmKna;sGv<InQ23&@_gULGHa>SkfDtRY;O)MD>HDF1YUI
z<R+WlC0J@A)EvQ;Zm6o8F8$%hy(rrrzT!=oyu~25>A?LkR_|&MqAr%AhS|A?Y=Y|t
zawrcsMGNp0vhi`zI6S)#vgl2r?H^SywbKT8b?ULjm45n&Z^Z5pxEY*vU4XfDhrBUN
zd15^)*R`!x7#ice<A}3H3*d`}-U8-OR*j`?<{ZJn3=LA>0#lyN<qK#y?9-RdNGnuf
zuikAud$7`qf>=+LX+&1}3SB2g&9r~ik8)n@-`ol1H0t^?qiuU_SKz5ExqLb9xyD9F
z)4Kz}DNcp}yM<-XIpk?|N3|f0qRnD5q_-A?z+|c_2o)W1=mu^*&<7j^K0CN+*h|-(
z+8STsH?i0>We{i`@?DXtDp}8-Q@O5MvfIpzZq)MFrFB|_sauJu+x08jQhZ$gny#vd
zHNDyE>$o%einQ(xLyY{R-h{0@GnTk`M#ku0Y;Di$N=0RjFj9R+n;h&qiV9ryu}Psu
z>AguKIJ$ot{!WlV#^naBQEa)=x<Pw_qwVLAfn1TTl@<(E($sddEpHY{YOt|^sIhDp
zbAL0J2}^z^Cx#x96<C<q*ld-3bI&>Q>FeqOctN9Z?jdjCoy~7Ws~t>;2xD9X@ne;U
z?JZQ<idmFRM{6)vo;nBhuGzY#$_+#GrUVvh4J&`&|0w)6OAFsL&(a1$Hq=TbdQ0!_
z&3@Yph+09&kz=c}$11vJbx!XFN|hAkXWFv7zD5Te*oJ3A<Oajrv5-X_d-u%+^PMxj
z8unF2JM#TCoZ{5BCVtcy6?|Y;Y$>!D`?k>@=oV30^xP#xwU`n_!&n%+#&w+}fZDFF
zVz8>$sf?UOwQH-d+xA&w=$g4q>Rc&B*nvC#RXE4F5f56owi#7j3gAsJ9<))|;dk{2
zptkpa{L!msb9&vE((Iy~McP!eFR=lmg9SB(vb{)E6rZ_xYm9I}uV&;Ei_nDu(_rEW
z+B0zTk@#oja@yp57T~>6b)@cnbTG*FR3rzHEt9btxyVR4=AABe?lMS5bUhPp&UV4x
z^NA<9JY94=n4<+y=SQ}BG_yf4tEr_lztGPQJnL>#j37f!ezagLVCfuyYu?<65-X?W
zLXN5bXq+s5@w!56t#?nAJNFV4NVa3jw-2yUhdi9<vc|lmm#v=`m^@vb?c~W3C`{W@
zF)v@!8gYf4r{*_-sF~*L;x=cKq^_1eyY4I#vHIXq%yw;xH@z?$phaj;w8M5nnATj^
zht|iB3-bzTx~{DOrz7pxP~o>Rpk08QM>-d@4w>pkZWGdWCc!TCbCinAEEG;F@v(OQ
zz+q^s+r@VnV;ra6Z1c%ud+(Qv`RW;ubJ9nD1elS2tvCFiZ_|rP<QWHKGJAXJIvern
z^2t!O>Kz;^mZ(S4v<WS_vr#u9PF`(ML)QC*&&8CGlhZKgc}Vg5)GP3$=5Lo$;P%)l
z9($iFqKkf+F1iTS=-b4$4x=8<;~%sG0(+rsREHj?dJt1}WsYnDFdhl{@;<4-4Fd@<
z)`DTP>tbGWv~DX3={)a-W-6vCS#B@Y)yleA-NTi5Db6top#W4NGhfL>)XBb9Gb<MK
zI!#U9&vm9oD5Ge6fOkHN6;D5Ek+ui!Ufr7N-`HCxLDAr!@Pl&V9>vzvQ;UmPw}d6d
z7I#P*zmb8!Nbz*^Yg2gcc!yM`-_u+@h;3-R=<zxl=Xn~2XdBf#t8FtO0S&A&GyN+c
zc*kG62Lc5BT{2%$*zeNHBV1X(#&t>8Q`AoSxYOkQmK^#OA*E-g#f!5%Lr|eX8lO4~
zQJ31&;bZMZmIx!+Y?saDq8;4X%~HLtDKiKRGgNH3HCvuS4NLr{X5Re0RTv{EwBzFV
zWWp_K-qnsF;D$nCO|<>6#)8-*&sx{5S#hm8YJ0{fH%nMb)g;JOvrYOt>P<q~GV5V(
z8)M#^Tvh;gke&MWW$~!8x~0UT8Mkm+Mq2c48|};4l|~SiB8%r{fw|T?&YGSy>%*AB
zKKGp#Psx0<M)W|54c}FOTMa+6Z>_^_LXM7@elEC2E3`H95l?fQ(JrlQYmNueOIf*}
z!5LPVqUZT65&lP(F;1QVEM44YgTckOFm|4jwnmNjE}iE4WLyTWPgrNzo*?E&BY2he
z3p7jgq(808Ej9q`+dc@aN1qG0Suy8^S%@SKy7*-B5N2Tw#KrHV`I1$JNo#LPQ8YP{
z#?^qRJj*lU;`(R!GOFP&TWKj>%8iYtK-BuPq2JbLJa<@aKZ|GF69U`!<fk=7Q2x7O
zb(?$-*yNf2Ngc>F+eMd$I*_xUXkHx~iXc_`;u3$?ZtNS%-QDdm<Xk#G4i6UHwmLPt
zHM4S!cIK_g@BQf0rqf1t83K;+ywwVBnvfS9SxWtA+Xt&N?di48yzU`{HbzC;lVnHg
z?-BX9nv@vA9TtmW;<<X0u-95GQH2Q5k#M<`@h+lsV@b9SUw3z>%dE{OWM*iT)=9CI
z%Xjhe3|dH2*>;^3TkGBvnfC7{@=OsjWi9~<^Cdd%T76ankr=rJN$bwH{4wfm^&F+2
z!P#V5ICRmQx<~hO!GMdy4wiKMN0svg`cV?ZcEgt~z?Y?PFyTu@josFKZmZQ8wyye*
zlWb3%>vh94s2HN!atEA}J(FT_&Sq>IMJ^kz$#p$l?%1{7K!#Ws#vSoht~I77xg$7F
zNmsZzs&hK62odmA(JjZDwnEr$q+B}$El-&fsRdvKN~`h(+(#%+HntN{w&JE}<4(Ps
zo|S_VrUWjhW^!<|?Jn2;r}We%H|R{1W{n#Y5c7B1n$=IUfxMm%M8!HzEvb5P=rC-f
z3)cfMi<IgeLpQEVM??`n=VfrumYuZlq7e-6ONyn5CN`bXwBi-z4x7LC#f0_g%so&e
z%iJm)yq(pzNWTU*N@bBMQD}`Q>U>}qU^aA`ZOb_=VALl;HiK>-$+{a7!OBF?;$1w0
zx8hX2%h!i<2UInP`zurnAZnZKWsWP~X^pWJrf});B7~dD>Jp_{(erj!-vx(-{NCP+
zEpG|$Q8~}I*PK><0>h%=Joq{!VM{NXwPJ(x-OwMG*QkWHhw-bM&Kili(FJ}}>z|4%
zZd<Rgw|M&PQVy#0>oLlSo?g#(5OMnJ67UYJh_-l_Ad_}}_cMQImRobT4B0&D^^*vh
z#d-qJ9z9~xE3ZhWT>IIG@6C+td^hONnG7?{cg)42sja@AuotsC`xtL82r)VHc8)X1
z*{|m^N^b)oj5!pXdXw@v?t0{qbk0eM%N~DReMk}w>o|H}MCASO(U8Q4idmOm$6fVN
zk=wrTmh{QduP<NP+`25LdgAp(etCj>mq||0yYgUSOE+`ouPe07x0qay9C3|tw3~`r
zkHQR(48=9@tqjjMTp!|l5U9zP0mHEszdXxCFKa#K36l(9;knhwur`ydbz*X&>%179
zJU9Xe!2MtJ1+y~&Kt$SQ=z&i!Ar$~3T$rCT6W7+(Hpyvt5S^I(#KX&^BfnQOwR5mW
zh?85X01r{DeaI{7t@yTPTJv4sl&ffJ%wonNf&L{WC7<H0r<(i&-E6#rGD%&Y<$l^E
zB7ABNy{sTLx8S<9`gGoo+I_p-H{Z5!EEOt<2l<FkhSz0d&H*OQK|byBZa?qJc|kyu
z?_xjk;4et^_YZB_1d|uU>w^Xyr>PMyiQPiWmOgiRrselwtYt$tB$Knsu>A_GZcDzg
z_bV8lO;F1vu*2(ZTKM!sUOQ!Cx|N|!CCPK;^~)paS>Ki$UOA;)yvH*2-tXGm0|D{x
zcNBRRey(T>X-uqXjwW36=UG%2iGOAR5db?-*jb;Lm?#rlx)L8{Te`KiwHRfTo$bEa
z$c1MgY6b+jb&S7v^D1dNKAF$&We7^7;g~{6SPqfo4C7*+rKc4Z78F#3Jy{&`qRgO$
zn-)qZ6-c_&(_FP2cAUNrp(_SF7?Y){^<P^CK0JFv#MqW~e=|DjzyrrA9R)BUB1#hd
zDU25P!<mffb=Hq+Zz?N`v=Zb^V&Uu9CT{6y%UZ*Cs9A`_`3oloObazY7mD9J^yZm?
zp6%7n)(dlsRwGu8Br+6{8t$T^y0K0QGu6p_x(MeKgZuI7`3K30zn+a5PQX@nEB>;G
zhK2?w7YTD8^i9Cyj}Po3^hPTWM+@0cS7kk)H7*NbZhPyxgbjn06&WB!y-(jTF3+FQ
zapWiEpQ*f)?#qL(g!NjY@*TuI)zS*>kEMIAp*L5i?In=_R?V*#2hsLG-OD#N>g=T>
zmWQ2DQa<N*w_rU~$>rkWLs0>P2df~bM{)62uy1bE(O@=JT-VkW?-Zt0Rw~1sw4Sce
zIE=a}_V)#o-Pz~}3y^KmU7onnkUe=ckzCZkWIlZS)YC;^hHNZ`M^Bh)U)I^P5l-rH
zpFh*%FeA!DF@?p&e2FsAN?EXT)N~I{p*Wi>JwljHyi;WsRKv`}3$38hMGFN2<Nf-v
zb2b{jZhbnz{!E%WFa%%0;PNYp6@A!|^9(Zcym>cK>4o-scvAoXm(LJFcB<cgq<lbF
z8PO~wu<(|1VWL4Hc>L(@!yXq(cym4TRK81y)lUMQ+`Y@r{Ev`;-*ry(3Z*WrqoHLp
zZb}lyHsp*qo@~j;&1O%p6v^i{`)rnCWHJwJ`E1L?&6PBNE!k~Ivi{lbv-+0<{Lcr+
zqW~~*<y<{g6W~pYQ_fyATlc@np3q4pEiH|A15sp&hO4VnHtxMktzU7vPT!BQ#xTEG
znaa_xue3obm5Lb8@4Zf6W6MzQgc)~|aw^AGpAQUFkj1K+<Tfi=&cw&I2&%`G7$XH1
zR^Qc}UCt}poQiZ*<=VaaXS3(z;-1fzKy&DKWvIVx)R!O;WMWi^k}5m*fWfPqx@Xx<
z`r7&|s@8(1T4R=TKp?zcV@J=(A<;b9%#&n4d4ax;lauqkUjqRJW#@R%b*8gjZEuzG
zu0|%h@2vKnOX`;XD^eW%5XbD1CL)@Q@L|eeChJJYEfkJ$$2g3Sp?fPmj+Agg>%Jx%
z|AF2544Fgmy2r1dzh3s_OKQ1m{o5-GsMIyRDw58g^i)L?7{d0cIqFOHC^ksbyp_lS
zl+bT1hiu*7n7d;reHm27Lcvk19Wc&RwCb7d9P#Pt=?StV*nI@#8VQIvyOxLEQSQ!$
zfBr%w@Z=}6@5y?WUY?Lr+}1}(1<Mu(e6+2LGt1zSy#P<&WtOmT4t=}zCysSI+=*no
z5B(dRgKkqQ@HpvqhAK40H*x;)i@<P(@n+VbRU_F#D?ca;QGhyzs@~2>J=`DG1wi_J
zOls=N1Z5amo-E(is)VmpIFqmA?B|`(Q`Ed!C38hj2^!gi({yN_R{j`cy7s46IJ=e8
zmk-AXry}$9oG(gJuS~ZFragT1YWF#m$`#Sw`TX24^zFf(edTQyIptoq*)8z^^&2-b
z1nwGVY~ntRrDl{!%B$PIl;@VySy4_E5k&?oQBWXDt;2Qw(^FG@g<Q9;`5PJ-Jlq>`
z_M0=VgW&1&Ki(tqI$K^FXA}$3=Ot=wq0^<lGvvP&{jSs8+`ROi>AaLz1#XLyUb}?y
zmQMZKPdAjHSXYo3PUZ~9K+PpG3#VICoYF<T-m?+~M`J++l&pD9Z#VBDLVAL+t?fv?
z%-sV%Cx;<eTQel2P*OFlNr5({q-I3f1jN+qPSkhE7S#tua~a#ow@e=6!tdlvZ{{=+
zr%vzN2xTNnv6yvyHwfK-a^9a)dG0x>UHJm{`CCGMK3ByL&&+DzfTi_BXP8;UjmA8?
zir&w;ZB8m{fqoFZXwsu_`P+%5hi^c=S~JQSh<1`Pe9_eug=^?VoSv<#%XgFxw=>Mm
z5W`Ut^XA>|p!y{P<p1t9uAl6k{MEN5ho5~8QA*h&V}WgFeX>49bB7VXi*ygt-g1_y
zQs&DOvwbq#kSO`^N8D<eQ-Q_(d&qmckHWrng`V}8Z`_2N?~Hv)jpi}ejI#R9qYXE;
z_wi-jk_Lnnm*HU0DP(7oY+x*|TbY4A+Io2c2{8cd>w^aMobgl1#$)shOy}M@ot|6R
zJe!(79J6E6ZLDEdJQ;L7GqW%t<|zH^f4HVc1Q|$BdglJs2Or5v{<3Y=L-hM_Bibsj
z&*TOd)fb*<wZ-LKJ;TDnqS){m3M={KIEU;2K}hO^WOHheEJEU;)QSBbxXU9vWC%(Z
z3Z#-+KMR%p_~yIRY#sx;m1Sb{<?9THC~F+UVkyewgP^lra(v_cowj(<pp2EtFoxRg
z&6QxSQi|27$WE!#L$<wvp?=BtU#y|GeLy@GbXN*7HZ?W%2Tauk&_8nryd>FDJsl$W
zI%@X#t!wnV*9#<JCsvv0lo$^_rH3Zd+mh#>+;?hIC_kBX7#LBga1`g|J<+#*Y~=AO
z#*JS_Qd2Nmq1*jZ)UGA%zs|Gd^L>Ck8dn?HGAjcwR@R-2A%g&t;rM`Nf2Eg~SD(wu
z4g_wJYYR7CX%@Z+Xhb&Smf<sZjJoPSH9BAx3K7tqE&G=FBN9s*CnayRS!|5O&VKLG
z4a^>niDkB3fT@CQ%|)utF=ak7!9=@lFEJoMZ^pnGkl0#HQGBWMC+FrUztd>{V~@cj
z8cB$Fc&V09T3y-aZ70sCXc5@j+Dc5DnWVXScy%R-b#*K5#I7GBF_N?j7lzT_kU6t|
zjeC|a$Z!iJEZ(JR#5kx?XL}^3TcRVGdTwPbkS(oz$;=GIqwfMO61X|c)Fiv+_(z5w
z`BK}nWEab}m#YSR6m@7m2f)or%Tov`KqhKPxM7mtP0sGL>LjLpv)iy-5H`_KGAv{|
zFo)k#xaJOS{blS^wa)=vf@eecV;y(9t(>6;5;dA(Q#<t&jFEMekb)uw8h2%F_A1D5
zWFa%Bl?$?OPS(f7A^v$)0#HO8`Y7m#eks~s7On;T9Qo`CK*&M>2i%3;>ThlF#bGJ6
z=h6#(sO<-Z?1pToRh$>NnCvoCBMqW$dwdvT98z>t3y18)T@aKEVSaOxm4KH-fOZzd
zfOq>xtBo!)B>=y-_8`o{K=D(As>rqkywS9q&>6Oh4SC#ssXRSTepOXT;lV4mwD;b0
z&hC!9mFLa_Xz`|9fc?NORdz78_4UcE<tgyceImD-k@+TQ8L&8^&wSvQyM3jG@Ytyt
zTKA1AbHw@BJ~jaTqsPB$%ts44%M9A~4w+_hwD2<Zs9XE@X$Q(yS6~*&K$Gt-L_fXx
zdB5%F3kDAJ*$m^L|NREEt1`zSR;_~T;>O0tUmbZsAEi$}R*)FYmJDK(lw`$<WA0TV
z?YaP8*bcU(2#YQqpG}ILrF#3ru<Oga?4?@Dn#4%6_n6}6paiTf-f7w9bK6E(`v$c5
z@B$kq8TiTUMt)r-_$tlKdt$$4^~_~C1%l+7^CiJHNWDKy5cl!}V|c~$Grw-{4)Xhb
zAk@~Q{bR3hBioKGm0jQCB{bYzIU3;!(e#6EkB6CN&NsrvX|ct;GjT_pY9P5F@T~vI
z+Uq#fUK;TQy#}98eO517i$)V(j!8=&L5%yh2F9qsSwGzO0O_y6$@%cM`idy2AeH-8
zr=n%2bYV5<Up(fuco$RQikTw1WpV*ufti#pcx&Z0bH5p@QdYTkGYIFx!`a2gEpue|
z9)@xsCZq^05V#KAZI3?zum}Ckw<X~p8G@M6_uorKLp{8<gI1=vrcrH@drt9QOT;}O
z%wbd7LC|UgcBJA0>75_X0(Ce0v}LPLGDKZa)^xOuvLBY!?p9Qu&9+uX+@DL!wKK_U
zKlW#zBnsCefC>Q8Nb&h1gt%mF;<0Tk%D%RWR9ODjwzu%MW?JzS1tBI($9oM;|4~@4
zX#@d?c5qI{lP8Y$Xga6&>4rrQ5Z%VZYj%epqZ8RR1R$jO*v{=f)s^yB+U%3-<%9M<
z`QH}B405Q%#ef>Zsg~$chmYh3%(BaRT&!=&#ETY7gzGk^V<|t1XZC*V2ZbDVX^a$g
zZhb^JL%dbZKRDWPiYGV;i!N;Sxx}G&y$(Vg#k<tK@m*Qw#>JcwH|ILgEmX;ZinkIm
z&#s|)h8;Lg^G(zb&RI*;T*BAeE@Q-X)gVHg)2<u7G=g0$9M`6W7i-mAx{cj(Y7JlL
zAGxQ$Yk6@L1a`{GqTNwHTivRW2cPxSDss-+@@HTFPvKLpd>PZRarZd&8T%1e<89nJ
z%)cBxAQ%x5;WY7b1f(e5<$7&8CeAEQS`eBws6fRj5%vQV-PJgZ`^`kz*q94YMuYF~
zlZ&ls&IMnWZlMes>I9uP!(k!+#+XW9wu%d}8`VPWdQGPEY;L|RLKl10SCTf?VYIJN
zF4L=h4hRydfLWu;AXe4(?lS*fi+dp=4*Zqh$U*ji3uZMD<>lpF({PrOvc`ko4r__V
zVu9jM0j7rGu@og;b<Lm$J4>OernaxX+<zsNa|D3pd81*EHNf}%Ypq)a$pLGt2&CZ3
zT5wb03;y7Lup(tRkk^a&Q56{AxH4rsM<wN?vF1F2u}jPA@R8{K0lt_ZdQNpQm_b&;
z2X(P)aF7ZDb&+tH5fuO`1)<*TR4f(*taH<`?q;AQ%PU-)h+K7kUM@J|n!M_S0df<q
zKtXiibZqQfl+A*vaay*Mx>dtYZ8Hz_JpOF!>_}Vu&u0B?A4Lp3+&+um-}AS>h2|M7
zf%SBvDNe$FE}*Z>(J)e89#Il3h3s22i;)zRWvK@6tlp}b`)W7%@7{pb{$REzOUlJT
zVQ$%LJ!%EzIdLY^^Y6rK>xwK%CpCBcj?=m@Cd#^}j&4{90F4T#S;?$9zA6tD$P*N@
z_0K8hGi51q9k1qLO3QK?bU<`lj(T`k17CPz)ukXS7>qf$Esm(CBMoeI$K|rSY!K1c
zfzG4rlD~P>+15KNv1NG50^fU<&Df|~bi_EPu-j4#a%BiOCS`m~s<mYZ>&i}sHQrd!
zwWfBQSfcm!Fe&K2ZrQB}3%nxg_306l*X~qxlpkMp9IjG3aK1fsl`q>W$k6jVxX>Zf
zs%sX1!>O!XmDPHxtA&g8J_|$AydgIcC-eK2;62ydZ6=XthGz{73|^-f3?wXsOb!kV
z3@BY2^*B!BhgX?_i3M7qDVXb-RCk&s6O@mOkAPM~wf(!TJ+I>^j#Jwh+`(-G(+lQY
zg$QG3*{sCKSbdJ1?gO$l8$|sSp6(Ib!C?`=eHzMfo#++NH1}CT&@%B(vzE?Z(2ufI
zO)DHOA=GwCp>y>w<*&KTuc`0Enx9Lq(XfW5#m)?>-wkoMyY_S*yWB|X%amEBZs}E{
z!@1jaE=vQ<$I0;cV2mhNzq}?v9Wr?F@$0<e;?Hu*NoXUpYpM<jp`l&sDY@_TdeWW}
zE8GB8_n96Aadci>7hP$XzkhQ3_#Ol8J9Ax$BC$fQTB|c_1DfJgmIi7;vO=3SgO1a}
z<DqJ1rkHyU#m08)Lp<C-=je$CNu@UJ$;O4r#xWX_S9uI-$i%BZu=m#KhGB(T1lJ+#
z0P@i%MF7~LZdnsMph^D<a<bls-(^}LsSXGgav!z^VGIB~bVZ^xkti!%ZdotC<MR8W
z8PIO&l^Y!j*LUN{xk_?+k8jLGh>T>AaQsACD}p-r(ADPwvNmXO=F^mXvNT1Fl{Cu(
zxpZ|v|2<_cM^{WuO`+Gj-8h`l^Aie9v9U^{R~_$p9JKB~{$h0h*(;I<b&v{+NBFa{
zH1iQL{$8Pi;>hvSj7+6qyIGhjv@MXTdfbVsbK%4*pqhXEF<{FCG&#$c)-q!+Fs0{>
zf6MUMZX(h#DW05Kg6&{@0Yyh?oWd(==aFnY*aKhL;eUc$UnyH*^-J3j&(^MfJ!V;X
z7(-6nz8413AH86}c=%J~0e^qif9B!P3HfCIvo%kqo28hz+s+p^g>w|Dq%625^A(al
zbm1`WN&8f96fHc`(AYoIJqQ>iAyLzREc1X<luJFEsDEd<+yDh_pHe!zzUP<;U+2k?
zHx1do$3hjV%aeM}A15*L$#h+iDt<{uS_^3XOWw3@51By)Ev6v5r4EJC;_0-FSy}fp
zKn&ShG@A&k#bHqApJsch3ZxZI!{`Kyhw(t88i>N)3BNm?{?yXT4ZCG)uoSvyiQmb7
z6sB0!R3Bu7RwX^UNcbjQ_+Y7W$(|5-o?;uWKofQ3T1#Y@QhofOdIvG*lj-fvW$B^&
z!TzR(kK?Yy#&hp+r#)fTWB5iEG9|SdaNTf#3aiF}pEEhWJI_#|dOJ~0gm3P1bMuN-
z=P5=`#TqyUy+8@nxbX$OCmE<Ho#(UNA4PB$WAA{@jf_dzl9ws;eP$*KVrz>{$j-i^
z^i+4!VFaUO?3`Xag`s2G$l{~|TR-wSM%^3+x~<wRQGA={XWN`cJutQ#6D+&n&CwG!
zVjvD=e7PjL@4SLRqPe5;#HT)y%g&sZLk4S-n_NY<gi8iS3ly#%3or@oBH{Vm+Q+BW
zlhz+iN#x=LiX7tACl!vMyV?+(INUgKlJHluJ4Vq_JX$UsaZ5F=9{t)_qiS3Z^J~{2
zP6tI{BWo$I&}uw2m^Hg0#^AWpJS&FrwGGYHR83!N(%RH@<r!Kn1l+ewK;f{e1!+!!
zhW!v9XufTOviyVVJQ=`_+lAQ-y>Ro=g@A1d8v{HgM#7^G6ZI|J<C$$_9u*pwbbd1B
z;W2#uZOIh|kbH!aryu&y$$<3GBu$~2{*5eOC2kC`;8M36UZzY2MZFOD#jfkBj#7?n
zD9z_I#+YUX%kbydF&`tPp%SNYbZA%sRCAfmqxctqCnrOyrj;l4F6gz!P$xxMIZ;_w
zw@3SM0A1De_;REBP^u4yPG7QdZ?dKDP~~k3`PO$7%C*T5;sp@!0YJ0PUSS)!Wi=t$
zUYMDAmIXP^v3rw&m-D3l!poCoaWkjojaVn|oJ~yKY>JLlihB-xMnh0ctTsO{uNsgr
zPu~?wN?RfP3*QqNqS*8mR4|v#51i~>rhW1-4&)BkM4-{ghoRo^#Nt-((#vuH9wUg|
zli>U2OIK@a>v+SP!?@U;JCvo1CHkNb9A+PZ=~94^kL=*=ohBg@3cW?3m9Q0Ju+uiQ
zh6b)802&lMLeO}g)U+I{c-DnoHw0yz7pU-bpPf8NqR*5Q%u+r`fmt%xTFccnKj|*v
z0CGc_2>xz1!<^H1UmYh!{z%?{hiPt^FTEgI9d=pyaFJ&*ej;3+SpAwT<1e~xFGMa8
zQ2SP_+jw~4IeGw}ee~eb1SHg%=`#{m(SnL9SRh082b`#2j1w|k<_5K8b-ZPV1cRN?
zoHVUGC?Id7v@6EyT9&(hf8PP{errhy($$mE`O7y@BU&h%NA3afDk(=()(cZV``pdh
zv-Ce{o5m86Ew6Wuib;~KpSNnR%Exv+PA8FWWZ~iM$Kc&HU_o;9{S5`{$V3?y<1fHe
zmM3>p1$*M+i8EiaMC#b8VSx7~iz#2V!5h3iPMT+^$iO|j#j1y~@y3+ge3a)_PQoD|
zF^J|x_np^nzq$dMWN$!-EA{k3`=D33%m5u_=D-JQ+^m(23|0E`+_dtP!A<crIGfv2
zswHN@-qxA*{BCIS%<v2#EewmKtaghS|Mq-eN-8^wo_`vMe0_VhKUbfjKWU3zy61|)
zkI_{E5<nKr-O9-^&z%!MenDsh_}&<y;^CE3bo5|+|LcZNRN{cTJ4q&*ykB9_dD&ro
z$euXH44;|#=h8H?U{AzBY^J7`Y*sUKpDICB9IvZ^K!faOP79dO0&2_=<0Pexbh0#3
zw^x^MecXbQRbH7*UPA*0g5mRrbbA;;(h}Lpchl3;m%20E1T#vI5nWgIK&7lMV$db+
z8o38KK)+p<JLwT18*qagYNVdcC%N`HhK?(vYtv_~HAXN=#XhNGKKWLN5D+YPPuwss
z8Vh|m<Pdacc8AanL8%X>FT$XQ?s<4weXjY{`sJSumlZE`>Ow!cO@-^jh1aVi(;V)`
zYlABAW>-j;W{Z;{G7KFoQ}(dU6N0-5nvY+_R7Wr$$P#82#F6D7HU=mo`TndpWYY^t
zY0A{CB6c=gtsAm$cqGMarEg$SnU7AA?M^Trnsjj@qM5}ZtbgotJ3y!j05*ouY5WES
zHID>YX<}EAmi2pNBnLyeh$t6#mfuD8JNS}0nG?R1%iy{_uRqOD6MEjJ=c%v{NE->R
z4;2&?x_WyH-C5CzQ6xMyQ~RLK-h&gzMBtmTJ_IUyjqpS%?HG5rdlzo{$tf#C0NH9E
zHl@L9usG;QJQBM7wD}tn)QHyYE9*9o{Sq3UBzHPFyY-3fY)=NNwhctDp~tGi*tl(z
zD)3Z%QmY2`$6_hRIG{!Rq*ymKl#_1lQGH6KgJ>nSkRHstJ5v7{Qbp(qiA+zau2K#4
zU&x+O_Y`UkUl~>~A*#_?NVcq*V(eCK@p=~;Iwjv}o9x~x%_Bpz4%$={014?>xnx()
z4*+Dn3KXtv2Bef+Mml(KSxi8J9f13khXeLJ+st*Z4fbaKl$*DVm8REHVwl^06vBp7
ze+&fhP1WP^1yf&a$K#eQoqQtBJ4@9HOJFF6OnSX!&dhITd%8b?UCKEL<X!u$q#k&#
zHx1z67~IeK=IOx#LFDeenm+03vf5sU_(+R&P<LI+UpQhvXo&$Nr|5vN^B(6IW(Lk-
zUQwFif515|KYcdU6fThsl({l^24EO46`#g{&@$vgZ*-g(j|o}iT(o`1GAp-uMTKc0
zR<2^I%(TBYMR5Jj<;9Y@#WHNNKcY@J6fT$J3`(=f1yOuk^&-dAR=rzdNg2@cEFj7c
zvf%$fa&4Ui9C|jI$IEwcmhFS;(ck7ks0ZyO{IL#@UO?rt1;<)_SB4AQuSp{*%(>n3
zdv4<!iS(VVW$OcTm0*^Jy}hGD{<fRpsfXo|9ysyH(BgK$X_`B{+;EfNZ^3LGD{GnV
z9K#NXW-(PM_b6d^H+7dK`fRwFa~&~TSs!<t`ACK3fFVO&h_WUPi0(xb*OHAF=o!xK
z4jvx#jBn8&3h6(@Dg5N<&z$s7hDN`)i2#bl>!3?%0AQ9xmGjjwtC$&}1$loFc#Ztx
z<WXvi$_h&m6U31U$~goOQ2skaIUyyM`deuzM7>T)2x*`{1roDuD}tH%wP<efg|d}^
zrBo>PHfRs{faok<5YIg60MOfw=UTr63+k6Kx7QM&uiXKN9LE<cnJ$xm_{3%+{s7!>
z2%a~zGcj`CsiJk`iHJB>;J%2Ad~+TuMYFXvXMU|N^t?l_#`-l+{b}L(tdCmwQV~p4
zT$;-i=FhoD)<6CA)$z%mE62sOwMADK)F9Zc?)-G|mdM^HrbKqObt6Me@u+oAPn4Q~
z#t-YS7{|xp@B@i`bGcW|awFSDe_!8+_iFSnL0O?Z571V$-PS`eU$n9PhPh*_Lu$w0
zw4k58g@4Ykz6T*ijj9l!!)??FPxmyjy*YVa(}<Lo&+4=KSFaU>DD6N>I?JS!e$$6`
zp<&A1ppiKp^t+kU?PZ1!o=h?_AbVC0n!;B|_@*nEZ?Nk4VZxzU;E@tVa8%;F{Rzo_
zk?dRBupL)=?jcTWG8J)M$X4+;qDH(fHk2vCLNvJzjhk=a0Jtg;F=*|67K|{;0!j~;
zJr$8U&CQ$s0p8VBC}_4gSiFRC7p7PTzX8L98cU<Ec1%!e8+srW_(5^Y)5ZRB#5lhE
zC<do8w5sY@py%}RFDQ~1BA@?9UP0nZGaa=eyp$cdaE{XCmQ#C%k4wz0v6_wHb+@<n
z8F;yb%Qk|x5w2{8Xk<%DU(#DqG9Fq4uVc_q8hCmZhu7{eDE3sc9wnvXz05tb{R*6i
z(!6qf;$v+J;;sgfI0;lc!W*MU5Yxgtx@w05xBG3(nti;e><7cfW!+0sOa<LQ$K{P}
zuZF+_zl*~Kq+p&((+{Hg;mpofFU-2;5-vw2J1p(;{epm)^l7#`AkLKjF#Q_aUv&n^
z5--q+E8)gS31w6EbneOU0-kSiiM#pBzXdIot)l>}lFQy%dLlm#h`#$^_ysSFj<^6d
zlT%W8Hs5F2P_<ZYZ!%wjF>6?w1eLGfC}5TY1=qLl7biXR5#F50tOhT{pqMmy)^NRL
z88+@CoL`>P+8?a_z@MTYy&7)VIcxw|3Cf4?lbdMqB`G}i256g%2%3fkLUWZg={;S5
zedo_3=1hHsdNk?-=pnl-`9i@wQ$`@rgt=VNFl+R=K3|IZ3OX$2tlV|kb#6j<ePhi)
zX~$dF1_8C$M}jH&X1!fqT?N;}BgAEEo1^%c7Ar8r1-*tz(Px2F(#>ThGN~4n-zx2#
zX@!s}{ITKcQ4s{Bw}R#D=P-y2qim}%&Qew<(^Gs0=g;npj`F01V8R9aMprv!->|ui
zeV73~>?B;VHF4nFMTZergcYf3U#WKR`tAb);P9~XlR^2A0-c{o0iMW=>I)07<&fOD
zA&&kd9FWr^@7q{w@QmCNXcY(7X<B?0duJM#(!;LbGCz@=S+P|DrcT3x5RE6*^`rBF
zC_;u+|KV_7+M!y|PW1y!ty~Uo)md>W3fM-CEI*lk{q<<jS0653;Q}ISM6thV;pAM$
zK^5RX0&`pnfhSNfFf<c&x^E*e*C9D<Av1%;)}Z~v*5JD7zWxEU4QjptPEfYi9)?K=
zHXpFqoOJhV2u`|2>JIqD#+x&5`Ims7yrU?oM&x*tyC4<3>%gpCdHtgDXV5tFq<01)
zq={e{6`+UY$No6y<U^*9XA~!C(k!GiDjTzX!GkBpa4;;aBLScP8U(3F^$+&+<&6~L
zWR2&V@}j5n!(nOrJO>X~-d_NsXb6x#Y5JxW-IJ0+8#R<iAV853D1a&xSS+5d2(o6T
z$m&*7yI;_+SO0c0AwK>J1tldQ-?j$N>uuElod<<+X8BZt=z!49I*d&mt@^g=rFu?T
zPV@2p#JEVaEwy3G8j4(o=Ppwm$7|}}F-<NTm^3mtcSwgeS~apjXtzgqn7GRfM%e3I
z*l}CEbH24U^HiJ$a#Suq4rjMNpnL!bRO{ld!gAys<-l^kRbw;UYM%G{n(Nf+XSQg3
zTMa7jZNs++M9(*i?#{>VY%81P4hS_^uP?xP(0vF*cGYQk-HXHF5~7h3XiU&%TInLc
z={oEGLh7A6n!`2@br{R^dC=3{T?6PF2pCMzf^TWIkKT!XdSx4^Y_7BjVd|DZ>d;`m
zss_8W(qvxyPMoGf1)A*zL<;YjN^tiuCCHcg!=Vm%c?yUWme@5OGRm03OguFY2=`im
z6{rDESf38H(`+1k_G(Rg;|JL?kn!_cPo?#l7MA0CtT!&m2QNgL7D%W&FT{RKHhk8w
zomR-EF}DB^5IOiCySq0<gy3#@1;LIMzW~UG$!$597Em#-h`ne5`_SM_lViKm%(LU3
zJz$hN=_8qTktu^WVIjW(jA`S^no&NRzltdEA&wkyh+jNQFCX8X+5Y^6M+%Y4BvVo~
zneujW%JdBH=9o8a5d4FMU0ou2T|d@UMo0F{Y_fJUSMO^}#Ny)e>}i4d70X7JLOaLp
zgd;mx)N!p4MJ3HVzD?_buX6$;n{Hqp5yB97gW(qZj}XD)>6KXFgwZU4#SL-rYL9qA
zOydGCVBl-wz1>m<pZ`)Obb*X_wXL;o1Q%M+6XyEZ9W;Nslx()?X!|qL8C{jC;}GxC
zyQL#I`7|A#-&0%q0|bFKrBgd#BxdGf4^D{b2`&|?aR%hHl5&={W-$o%dc|flhe8$8
zx%J2q^^496`Of{;<tr@$rp36WK=(II28Mh>>%~>$;==r?Iwd{uu!=P_hy^jc$S-Ww
zk>)aNj_y)~I;5VUTU-G-PL*Yz+_tfr6!;qCyN^2R*)Q<1aWLjZr?2PC?xa(B4l+|%
z&GB=pf{W>veoIuu_jK7<Aye7D1|nH7&lsq+wOV*HbOf*F-B}S94ru0v=gmZ!mn#-q
zlMBAv(>&k3&-iL#=!}7h$wL}J?b;^%_&Z#a`1l%X1jtoRy#_36emUq5(g21<vrps4
zG*H4-1NS~#N9%6GtU3`t8uz(K`^V_lrIpy7YdLJ8VEoq{k&=p|2-i=i4O3q(1<HZI
zI&a}J)AXd0uBpOLJ#gXWccfsHX0Su)(-{7shG7ffW|6!*+t$If-@wuaSz{abkZ{b>
zcswO-FO{)7UHlEjYn+>b7n<wHx?JM57&8`uKS;n~G1G47z<GTjCkLIn?U7U=BocX*
zGK?DN-`LGHt@HAjH~7H&f(EA@i>0|Z2Dp2eH;PBGrKpyAe~nwL&>U9`+E75h;N1;p
zmU=@G=hMX2s&A!S)r8j8h?g6t8e@uXyW_SCqCmfyvj3R3i9%i4(_F2orl+RDxy?dg
zVB2hwi>~l-iV&2d!L;9&hl?v8qpSIJ_-RbPS6UMb{B6inJPkKkykeDE43gzMItsKM
z`t?F4m6Jh(lJt!$x}U=hox6hQg;nG=$1z#!_hx}GQ8VW4ejT8tBN7TavXu?oK`a)A
z$W<YXVB-KK1rolfFje?NNO#@ftE=j0HuxmQ@JqGuO%{uGvFS-7dMX;cECW#T*FYDL
z8OY*;Tq5vqW$3|n5SL_vIB7Yb9UBM+#9#%s)f49@Pd4?m<>c3U&LvUz+U6FR2y&^-
zHTK9uDz?_tvT7sKHS~;A`cpsax|x~Uo5A_@3TjJ7BlHT>+4SbZ`c<dCag{A*z;%0f
zf@7no|Ll&3s?}EWR4gsvYYWGW47-Sm?Ma34c<)6+C}nF{_!`>frlF~nRk<tBAiRk*
zueh#kGl>E^^&Lte%y}Pb!)gQ?Axr7z(G;MjdG>x~`YN2`md(st6`+H&%KU1YdH2;Z
z?1+#bA%;CiIEi?jOviY(Kej4uLkex-H9!5$m7<}6Xg4`kdG5?i=Kz1M?a8R#8$ZD!
zfSo++fhREWhz~E6u>=!@wP0E&-?ZQ+=O)N+H8bsVHFZJ-)&Q4RumhRpz4%|g*=qHs
z2dLg^!CW3gjGNQfx#H=VK!{tt&9%S33jpeJL(|z*UM+jWiq#He2d&ifg4bZ`{`UI7
zRs7I$PDPVOOL1soU)qUU&|kb@Ke!ZF)1hfub2fYV`^T%84kx8&SD859kyW?S5VRGd
zUJ{(AhMj?BwcR&1R<cOQPk-yKWOLQINCZv4p7z0ANoRaA3_5GY4#rC`Qh0TS?_^56
z{mDl_c$T5k_-d&Vh+F_dFvqwy5M9-QiQ^TJ&i=3QcpN>ravlgO%;9L~{?eoBB)b`r
zn+$*qm}G9=t(*V*_aY=G@(KzbDCLd$^9-B7p9Be>OiE6^u}*QZ|9Up4`g%n|<^NH-
z{{x2oo$!zoAm|UChq^gxf%Xbs{2t*f?F$4P)${$i)-{Kw>VU@3Dl;B0xh{x(`goPA
zSN)?MYuvHhhK7c3MsVomjjyC`9R-^G?4SL)2QJI-j&1ZgqlT<&E+YBu6esW$H+aj1
zhGheN*d)uT$Me_3sn_~>G4mYzf;YZm_@~Y4zc2Ox7xMR2h`bQtBfv|@Ku{+|G%&<2
za;9Zxv$yaf^6GsquXb6kG;5D<AYg+^&*&zjzzgC0mJoGfpcyOwt!8gv_t4b@%^+Qz
z@}Xbc?4N!mmr)7c_u>nXSrnKoJr1Gd2Qod&rJCuwt+~uX6=*&s7y(KIx)MZ?>`I&H
zdNqY{S_~E2zLdq>NZ=Ya>1)52+dsH>KmCh`)loATHlxqF^@tX#oU}<TYAmZt=zQU@
zQ`NZ}EpIjIz%U|B9hJ4tGQD)n9eO^c%X_V>(nGk#GIuB=_Ma~Dr?>svMdT#o01+g#
zmRv>0)FwB>I~|3p1I2>sm(sd{R<Id(>E53bs5?mnB=m2N?z6p}N5ulF>{r~=KfiZ9
z(~r^-Y=wcP7ifePUkAOShTh)ZX&(uf+bCP+fiEIe30H;o5<wo{K#IIj(7u~_+iQ$p
zoSfHbufK-t<Kn(lzhR(J&VwZh02NW;Ck4<n;y9=XJrghTYJ}hbD*PAjCUVvTPo{#A
zQ2|0aQ1xB{<Jn?garEuk-)}P20c7$1&Kh`?jb!bK{SQkHv>&592VPhA&rSG`3&@F#
z4%zEh1EoQQ2<qh&p6_YL(RYWF88X^_x?su3<?Xh2Em>*cW?5;2hP@L1IFtXf2fb(c
z0caL{BeZm;DB5ly-e4yPsqQ%Wm|nKLmLfIG@9h4EW(Mt%ujT~*g8dFWMBIx?kTH@(
zTAiCCO>)|uX^mzyMk;Dx+_yo-Q31oDMvU6Y`y6wCoDs*fFwml5cMwed?|%W<{|;u}
zntmE;SwkjO%lzgAzK&ka1Mlq6{fSHy+<R0I=)Z<@f9#>l|MwgH%^YnKqv8573#*Rp
zq+o=H-D32GSY+77G#JXN1KF0mS-V$5STzsIu+MmHPeI^op%L0Wmh&&~<p3eRIKW_B
z%mFz$Wvn@Fnpt+69q>mAw*u3$s=<J?Mi%t=zB^_W0sh}N(%?Tj@OdY}8%D~$ROcCH
z)&cdqFEDG1H_euEd*>o-M~K}(PNDr`^gfe`=;LK|&Ko~VqzC+D|Jn^8cu5X?WZF5H
zO#cJFt!hOE{eWG&7>v~c2gWIR7mHdU^S%4WJ0>7<aaUlzN_&sXes|vMZhtqENLs$q
z5&_Aj_F!2dMFn|0;1s911ih66UJ+%JR*qIB#Mtitmz#ta4C+w{A^+=g{&Cp<dYHFN
zI&vi<RRBoHE6D#p_P#PK%B}lb5mZnNLIfpLkj9{U6ct4YMLG=x=@4mf3{+ZDdMrYa
z5b0E;Q;?L9PC>fiT{j*J!2`;N|9d^>d^p!Q+_Uc;YpuQVw@4KA4PINW$0s8S_8Y9Y
zIm5HS2l6+<lc@!ZxA$`s<h~zxu3s){V-J}_j}P%zK=$$zBwZ3m+@#$*7!htnzwRgz
z;%|&{r1*m!RV5|a$!&NU@sAuQR<@aOOc-xIZI?ZKtOM%tnxitE7)+l-G=;Tl&U=vs
zH)C1JE2ziq*!u4DLfTklGr#TX*AKGSfk&FMs?B?xL49nBRApml&sw*&j(DGN#H(W%
zz`Y{3=(kcu{OTzkDn>=@&5+gL1X_68r>c_+Ug*A~Q-S9g_E$n#$ziM35a-(Mf_f5)
z<7=-JKHN_0r+=={morjBk)0>z&34E#KKHtLGh-wl1NiV8>5C=sEscIkV#zp6gF$5;
zHcnRheVY6N-uq{ckZl5}cif&T*0S#&nbaGb?6#TqLg46*w=mR9z>0GKW~?G%w075^
zSTZ<WvLR)KSndJtLfKJ@NXVoPb+$j)gQ(3LQ|4y0-g?N)vVgeSky;xraa8*0g=`0J
zn8iD8pup%N4oXKzKj=;=cce&FRxq#~OVx`_nIT(X*q{fItg({Qe>4pTJjmS)GN@5Y
zw*$BHV=NgJD;>+>CzQ@e;u~32MKJ&c?t{qVZLU_AeUJx~XBN!ed{3waB0Kk_q~*Bv
z(oZa40*obE;Kd78g#yYee@Gle3GF^TAMmJO#qF3AIgxHX)65#k&E+I8850e`!)u$g
z;HVB=@4Yopm99)OUycsJeX|Y)>xm11XfeYjFmT^_5rVy@grrz~ApH~VfawVqc<acZ
z>L<P#yk@5esxRFAbO{PT?lH$ttA@uE*8)~Wplge~y!^%K@tnPk8x20ia=wM^;!RGZ
z2QfAPQ*hSc4Q@_0;@8XWTF9$OEMQ$#eRBBGmR&AGd0s$6<>okG01|*vUNeBi;!bZS
z=b`4Px!kf-!17LRgL>rTr0v7QY9dg#-&&`2E@Efq)4Gn(2mVj0-=*7J0%sk1AU4v9
z{)CfasBDP9!UOdM74;fnss4iR7s!K!x4=B_sn+Z;MS_V?QE+v-X{$^&fX0|(UwY!m
zaZhN=#QEL&83KrrO~D=P_Ve%2YRGA+&2VOn{H!o9kJpH|KV(%GtF!2@hM;mB;vBow
z{8onXupY-Lpr?zF2oE4hb4^GQ!$k<)M^xACUkHL*23&R0h~W5gmbvg5y3-#Aasms9
zwZ<(V!<Z1slARm#*#QA^scQijGHtWvwO){d<6XKh_6n3_*dlq}CnU==9=BZkNzMTc
zR3N2)NoiMphhX?sye}lHjn@b?bakD8g`a$42L#NWT>7dC19t#^pspmSu4Ao#v(>_%
zXbQo$*CuRzx$qAQuHNMx1Gc6tMi0MZ{~sZn|H=R;QPA)u2;+|sEH2oW+=riO>+9T6
z%#iRSQD7phyE?=DH@iP3n*UCAaT5KN2Y&tL|NPz!huvqQ{>LbJAB753)SY2;Q8ChW
zG9!MNLxk{NsS(ylqNG%Fq+^WH(5q3!+TwZhh<Mvh2{d<mNBcubwAq>Y5%_5(ym+OA
zaKtz{IfYN-v>(DjxOvzQb*b-sy;YjfXun`h`Z{hzT4Ba;&o?*Py-eSC*+yo*h9{$J
zpUVqkb*=1r{+F>gwTm+5hdk|!ic@F%z+3B|X>MRL>;Q?Y8Lr!-lv6I(`6D(=ER||%
z!j8V1yFFRZfP*HR6d@hADgt8-nguD1!C@2nGvT`6S7^1<$d%b+oh<(cH}%3R4EApm
za__G#+O{gAsn#Iq8Z}9c5kIH?FIdD^+59V+FZibOCK&EYDAr#y7k^#Jzkeee2e-X!
zTwi?HuW8GFBGi(oK%pZk-TX&xg<nL0Co|f*pt3FLpK}EbyRyzDv2Ck@_Aw6U`9v5X
z&3474Ax4*%bNC@^9@E3FmS$qjZYj<DC<Y}aXtM3OvvW2yvz!<L8|RSi1Xtl5s^0J=
zdF~g+!~<UG$F0P}>(T^G1N8~-RqMI1w>IgFhlD>d<T=3Rr%I|S_opz)%qv7R*x@8z
zYsvt{1g*^EGDD6LKaeKibpkUuZ=X(jmE<S;`~)KR#|+bX9Lvh$U2Fo;s!-t#loS#k
z#|t~Pz=&El-Tg!j=mu&eM>5Gb;?4r{{OPe}4<2F%L@jlJoqpNB_}zO-ePOMgq9*0B
zzfja4ah=q5P(>NS_pc6%n7hGRQ@eDNep~pvY`VcPfV4V)r{d|dfAKr1fYJkL^@y`x
zwCd>Z9$WUIpGV4jo=5q(FZ&n2JF-qBu09-;+$H}<=eTf((9aRInpoOxq3yBo!R1k9
zDgtBCaE()G)AFAGm$yg|gO}y`8_jd8A^<R)xNznA_62FJj%a?q7X*Nmtmy9w^_NdU
z!~0>a=H>O*nEz!{mS^ewC+wEL^$Yp!KVkRB{Pv%)`(v^EpRoI#Li1ax_n)x)|3lc#
z3R;+8(VECnY?-cM%+r0jrw0G^?c2V$pWIM=Lr;fA+(r<+m-*<~M<_JjV`2rI<9E!D
zoJHZxN;4g2J{N6gheuF+@}Ncnfe^}&w)J2Ypyn3Y*l6A2Zs%mpD=sW9wC)%Jyr=1g
zj<J-IDJxo|Pkn?d7G!jGH`e1Hzu>Vt+FHo|wfJPuuX4Uv9UI}l$^EHx{cq>~+qr+W
zzy9OaKP8g?cJ9BO`*#)OKOy|5RI=JTho01wcUZP-5H*0qp<TVrZ9hMmWUv6Cj*gbS
z;02E&06TFS7>%c-gz$3cOAom%OZt2;NjPj8A$>EkJoEaP1Qq&d%H_`P2%3p`yi;`k
z_K;63>=WIeo;``j^_xeXqamStYjFd=UdqHX#gGv3PNDR$HIu|6dPHx|@p76?iwA5_
zM79%Si0zs)*J_60B<Wvk{-i?@Uf~{}Q5i^F<jZ-WVJLvL<v#6xFOx1cWhEgYE6YJQ
znGs75v(Wr~D2xgoQFlj{0Ash838nph6y^<%rS$oxT_Kh*hIa6cKoUJ#zjAnq1P<06
z)2%KC$CJ?v39Pb51F+b4afA=-GBmr0`=tbEMv`g*qb9M|Gaw~bS>*=bImD-c>cS>V
z3a*{>Vu`n!ZM7*HtTg&i?*~qIbQ+NcQ%38lTXDuXx%=kBSngs}6Bl}JV4bOI@E39m
zz(R>2&k^8l?hOcwcjR#^v6=4gqN!t(g#1X%Z7g3K>Gw4?qyYEs5Gon$8D-lY&<!A7
z4$#sZlSFdW7<=P+3~Df@xm7(oHI-VB@pE9(u=KkLa4pfjKn?OfN!4lZ>d1Ol&l-b_
zw++i$S#jfWBy=X>jePc=3hGWH>inc9JFNDp{WlmTKbdSWT}Zn~^<gEdp7PNWAW>49
zHBHP2FIzgL({3;3h|pPCnq|6`Q6spJQ_)fpA0nU(J#p+WwHJC#Cg7F-CQYPuAG|?8
z<hRA|Kl*>DpyBE9M;oye?Qby9EKSfGvB=nn)=G)2*uMB_!~Pi`lY|ZI5i#?MY;}fP
z?a6hAX>m(0J4@Y)UVk8~-+5+?veApiy`m0*OTJk9s!L7E?RTdTCyu&h`^m^?$C8`A
z;DWiljjPR_u=y=vqYr>WQ%p{jP*!3d+32Q9ElCbibP@`8v|duL$~NCX9RaCEM;Cl~
zbsV<rcp9Y=j>(#$t25kxlUqP9KXUo+{`BLq-jhu<R124l#+sMW%N=1+Lqm#MLglzt
zG!n}k_C}StG(knC^IZxbBFUgLj%(MEHAAht(^zUh|GV1SGIe7e9)mJ2sMb=sM|0Zq
zZQ$&yy`Ibp4MVAMaU8Z=DR(e4XIi8=?>w{@QdR>Yw<{wkI9cp>t6FqUKT{e;Ws^o}
zX=HpOl<RgnqRQMals(Nc_c_crGtBy~re?xCDK4&fyw9ZnJ_z~w#;;KSIrKQuEVrlS
zS;9-mZF|TwVlv>{c&6(r88l}tJn7VC1SOuTyA5-X7>qX&CZ}(4eD_4S&B$lx-*TW7
zBZ&11-4}PZQ%NwPpEkW7OB=goI@uLkFG`wZct(=&v?r6JaZja3_z@u)85zFm&rb%R
zr=v{Da9Q~75j6_2ZmZ_|jl-75>vkhhn1Gf%1!zk->H@FN{`J&YU}_$EQEf0aE;Ims
z)Qq=trV~T+x>w<+HsF)JFE9Q0LXSVG!vcz3E60y+0ptQgK$C2wfFk7Hqd``wa+w$I
ziaHFH2y$jSQ_7T1yDoi!E>EGPGXXbLzSZ6L!Z%QXn~)x`y2+meF9PO_oYs6w?^8E1
zhs>>kJE5SxkW8WTwUbjZk~rBs)tJty)!p8{2Lb!Y2JrtKdZwAVvad7@px0{pl$J7+
zY}7%hK6n`*Xf`Mbz%}ItuW=~v#S~BahJ=RZk?AccEn6`Mp<*|JRAlD8N{+peNbw`o
zH{OTj&Fg1;hGi@3?(QDU?z6?|$O7O=U-k&uTlOUBJ7iR55+8DWN_3EX57qz4jaHf|
z{Z{?=AM7(J-08O4k5kVY5K2vQ#BC<+J9~T2IQI1RcI&P;uqd$vv}ECZS0_LIZP*u6
zi)g!}<Z+Nq8z-llWk(F?;VRA>D8{9<tE2XeOhcuAaPXl_--c`YnrEUnw3n!CBT*V|
zl6`skYIdj--(b{uj!g>H^%^zp>;ciOP89qMvlD@l_l!zrlE=Gc(gR`+WkDzQ?Ow{}
zYS1V}=5c7~)$;QZOWt}vbNP}&B7_I=RqCE*zPAS5u<?xlWHO||+t_E>tDF{%=?F*$
z`0Y6>q^Vx$p=Uc~ldj1rT2JJ#a;S#@uwz(c8UxamT!*C%U#~j!!$&vhYsi^1H(fG!
zg@5ju*OjbbfsvYo*-5P@X=yB4;Ya?K$vl{V$!Pimpg@=IAxc<S7{GG;4XTtH?eTYU
zTrN}qd{PqS@z<vP7LDZW?ChkMM2z(-o>xEtluQ;BamV?*1VI{+*iJZ58BdW~A)B_r
zbvJaExyr`5^*#ci(Q_3*rSI?39hZlU_&Im>Vyfk;B|nomdX_2*+QmXfvhKC(*GKQR
z?a+2aYV`R>s}U#~`6xevdFDnR1=%a1adrY=|G3btMa98_R!S-3%?|HtO@Ize%5Q~1
zIpvs$BYj&2v|XAo?Rq9h8VqG!tQHl17s{Uk#I-Usmtdr@u^bG#8=o`_7+=(oHlGXV
zUiTFoxqllzS^93k2yTSbZPw(uTh*sq071QPYXNgU8C0&fgvT`O0tnDHNQNi3JAC&`
z>IH@qD>jw8CJ`tu<QjUaf6mPE<2n*Gonrlb`u=;+!8lRdhn2c$TY%+YsKf9RVP`v|
z;)!I3OCKyg=QGaH19qcfz-<4Vwb85i_znlcKb>?fwz8G+bs*IbZ0enD*%2y@&?tLC
z1$w5^W+fT-n0ps>&GQ82Gor;eoQE<71{9T<f%Qm5pgq)18@`HtQ6y^swZoR9F@bi@
zR=tBe0oNq^;W^5A`SD4s@udGK<=ntD+`Opn@zKoyq<;^9jkmV(85jGRA(iE=e=~ik
zU@dvuoE0hc;pL<@zhuyK+>~JbjI)4m<J<X(19J2&TQ<@h+-)K9Xm+jIX4pL%EzTqm
zEVQ#Nf@u_bW(B?l`AtbZY3{WW7vG0)zV&UiynMEsf1;J}(p_@f_QILKxrutvZlsF2
z)-F>(EHqu7pZb_1ID3_yIhd((&@FH}oT*#lOW8;Dd6=IlyCN+L(~Tn?sZ35ekt;P&
z$eJ#$q*dnE#Hf<Ua>A+Z5PQ&p^#EN^rZo2IeW<wLjC;rn=Hg6qD=RC2j8c&7XyA&v
z{;3u)hAFE}%T4}JWJn!=@>E+ywpK*Oyx*h<cUNdI8|6GZ=h92~rMDJwzgPo<yDO`<
z<$FL8{G1W+)ljILGiADo<wQYM78GtsUYgYW_@Ob4(YPC^6q>0Jp3B*E=redf0%d#v
zy6`Nu9)r@t%UI&d6}_)m0Lz^H@P@TEr^_j?*L<n;$q0n2L|+_Cbd`-vDzm`}y)T`!
z><<*5RO(MKu*ibqBuZ%dq+(<n92h`W9>nh@Kb1D+<lK~b_n8yPMOEl}c|ZD+NwK#k
zASfOJ@NK`Jn|%EVgPnJU0GY=t1x*^1YBDzT1Kxq$Lxp}5=ej2V(MDL@$QH`!-aky+
zozN4}ukJq!X!9jJ$|0`-A!$72oT$&RhO)uUg<G}E?fE8L&)cD_o)JAqAX2oQGwg^N
zn_W)}5X7jLoZ`2tQp(A8<eXK2LKrq6u??U&?+_Q{XV=2LODS-u$Mn&9ajqsEfIV$2
z_U_|MH{(%vK+gZ$*em!ja~u<GpqkP`o-^vz)~?u58{5+i$Y1%Sv|p}ScVF17wzz%W
zdr%k*78GL$Rkm-&$HyP&5}(uByem%VNC)#}X!mN(1WMF7$;zs#8~~*b;8D(n-X_9X
zpPMo-LNighau&q|Iz*-ch;ene!TLW;U>x3n-o8&Bx@M1Gv60nO)>4d8J?lWmz9&-l
z$QOC&hpkx<x)bpWVqCC5h_#ruKFE0)l=AWBkVs`6mDV1)^_1os=RDuttH1t%-E}|y
zvm}RI@g~ZW(03!ffbm}%5&ioXqeZrrVjdA(KjjF=8j}I2`DAd%jD&ZP8|Ln6DP%3l
zB`dGGRp2#85B<Bx<O-^_BKcoBk(ldGkEafbOymfr`N;TD{oz@Do1O<1qFeO30t;5}
zM~{w<LQPtU@F4#usL=j6?qxHf0l;fd@3?0T0C183qWxB$>t2uFlKsg)Ho)-~5MhmJ
z#h;3+Z?hmi0JF%_dxW!%D(YDVpoUN!NM+M1$n>qB%@t;`2Yt4#SC+fM6fpFRJlU@l
zcKRK<vw0_{Ze&W?)lwG{x4+y5G~RYNilZDN-q;&kX+Hy+*?o(?i3^n<?-Mo>0g#Y(
z<wHYK%ceV|+^0zoJ!V%wvuE82M{^<2pYps7U*9|)4JbsbA#H9V(kAM(CN+xc0J<Vf
zCZ|JS?3rJ5N5E7(_oaNAk69$a9Zb(+V_)Hux!xRYu!!^nL}24A=pSs9jl1y1S}Bi9
z;DUm1{tZj$gEi<BZ0qq%%ACQrf|zUo;8n?j&MhwNL5l0$+SfD6MC8khL5s6juQr#E
zNjkhf`=r4|rWOEs+KFj@Xy*1uIwUF)Q4RZepMp6tNT8vo=L}8f=kiD`in(okS%Va@
z+6(jQ>r}@-UwFWFm2Tu&EPHy;HXuTJ<Ex9c*;<jzmmory%}ME<_1pt#$umlT<~OCr
zB=?eyF>`#tA_uJ0Trx?qThQ0IP_Fpi_7~ij?6#MdmAT5)TKb%6^*H(a#$?|}eaR1K
zfgM1E&YT&;;51RjgoPm>hf~~y{zk=eaQgJBv(t6>WPs*gM@thATd&n{+<r&L>TbCq
z9^yOze8QA)14qYO&G+_gDIk;7MHZbacTQ6BQt`u!*nol4>Jj%sM`snTHuq%Y&PWsL
z-iM+J&QLP3I-p*Nn2Lal)#|_)d}CsWq@4)?*-<}p2uohxw!`E4CwY9b!J7nsge&l!
zM<hnEM$}XuVZrU<a$FS}3mvkWq<nd7+N=}STtq6VyViaObMf~zzk1c`alObVE~Po2
zH6((s;GC5Ari8|)dJ^iB66utu%!4Y1E$$SqQL4Yod6}YnOxSJe4E^LM;_ftWJ^@e7
z18!!nQ%<L^tSyo9R&sq35LTwcC-dxsX@p384Og>)`jfXRX%-Qa_F0})A|qt#1!jy*
zaZ{qBN`ta)eJ$)_Nsd)qT^!0SPv?Izpy;9N=5YRvk<=c+4fzq#6CoYXi}>sIqA-@&
z3{BC=maopAqeHh^%e)xl?`JLRr)f}!9|k^vAGtRVJeKGu6q(_67$fZ(^FLRoSEhcT
zfBTs4@C^M(deg8-zTR6czg`PhK4;Uqn%Z0~zZaq${LKO}q1o+K{4zRy(Ot}AzM||L
z>3!BQi8(>>b|YlL68%p&UQ~<>ZvV_JHmPjt=mu{Z4(@oaqZHQhymyTJNgHCoa`!vc
zZrqkjuRm7bSVHQC4SmesrZjWlclY)>v=z(|X;nphnw2L3vbzXMJ5oT*iO*O&Ld+=L
zUReMBy*q2r&b6m5-Gd_ik?P9@fz?udYN$5Ph(Q6Olgjcml$8nE%ux|j5`*jLy<XtB
za$=M{^**`omTR1;cwiRNNGVF8#n6@b{LyV*Ps@|#%5NqZbe@YOHM-Zvx%%1@e_)v4
z>}40b($>l@Hh2ris}Rd$6ar_&cLaS>(gtd|VzdT@&cNa6xK5oJeAwbDmoc+H=Ct8^
zEx&8Nt*PU2vl3kV3dWAq{h`xq_=W=Zk9nU<%PCSkZbmUeMr)-qS>DsEW5w^45;$n;
zX5Dz`=nYgOo@wB__}Llb(bm{2u?=IEo<7!*qXPTtv!@)c`HENsp3NUC%To`9!)r8o
zK4P}tDr>@-I=!c}!`eF)9<HI`?v8nX&G+(U1wqFAfEVT`QnODr7Km6lx}~|kaUN%4
zPrf>6&T3X@P$Qo)e~c9Lm0G*g316K};7PZL5NVW2Pxsd?4^$@6>yd}wrZ?^k#a?TJ
zD~+4{6pX~fc+^FNL1(;u47AH{K}!&Y-dE+hqFdEc8Y(Lhm|0)b!MtF=^dkj;xH{Q%
zfr-r0q62Un@jH?s=FS#@aSW8w-)hcSxAorBt4AKO<fTA=qS<FtPp=+RQL^2lQ(S}D
zbIAB_v(qK;ky|U1MSdOHn&C5`pP!JC>FYBDEmK4K5u(mOLlWrdHhv*oK9)yI`*1Y`
z5>I=v^h}r!WRZgHI!A7796ONBNwE|38|gC-_4SK)Ct^Wd%!Bc%dP;P(zUZDoAAL6r
zdYzF;xvth}9+spPsLYEB`w&a;Ju%p9eQ^VoPK+p>gYr{g;5^F~4V=9@-m5;xtZdd+
z@&2LZdLG^{O$aA=KHivV0azj@P)OWMLP;-4M-)aL5rS^w*+Zudq5$o*gY6Wn{YX!h
zus}4o+h6MfrFXGnt+?VtK{`g&V*zJ60<yl2a)&c%j8S$1;t^*_`5b6o?{}e~6)Z6l
z@Fqzq+RVt9`w(h#2b6rW+{O03a<o3vG)<HQT2&3oz5Ht@xEiyBN@WE-g3?2eRl>*0
zT>IgJ<OxneKF|h>wgA9kP_Wq>p{=5|y{3=!rdw$R?Li}X3wL-ZU;|WhJS&MmB4lM%
z&(X$w`SXpn>jp*Qb(O5A++_}bF$b3P6n?)lmQp4rPh$FKdqxE0qppIFVhbAFQ->n&
zP&;miyL#HZ*MXF^<rZZMmmTcZJivI2!R&<BTZrg#QsnnoC@3f>8yFi~Rbd{%F-DSU
zX=)DszKQa3ri$^y=`Kj*iUofgl_*u?di+b-y>)IE0cYHxgkR7>K{t_^hll5I5U6qY
zm9CTS%C!Bo4VutJBknt16Kip!I}s5P4jrvV@WpY1BKAOMz{oxKzN({2;?a{QtkCO2
z;e(0Q_4S)`feh0ecMr-({iP1R7R!nnZVAfQzMkNc#YH;un$hg!z21R%UoB^cR1i>N
zZR29ZFAv063^=NsBW&u`445fYS?PtzVZpblW3zM=mY<v6uixrjTyLsf`>Lbaw{hAA
z0fp1!b04u#S8v}>pb&6g#!DshE^`Uv+fTN(Yt5eJ#nT^G&ikoD{Brhq;WM54fUhSv
zHoO9RE<FPruNSlIJj`g20D_-IA5OI8L?DNKqs^QZK(kLyJv`X-vB*h1=`46}PgxFn
zu*m{o8)XitK!QNotapMblBfFS`n>x~g=n8<YwtekbF|f{t6)9;y7g?Wk{7A1Sr!#8
z|4y#o>;eWQOk5cW#u^KIGBAU3j*GYdxVZOW<Cyl%NFVNZ&emMIsUz7bKCIf@XSK9W
zgLkt70K#o(Tzfv4PgKUwvPMb*4Gr0aLGXhyV&{<w4&{)ra?GQhvxa|JVC60iF10dq
zU&l3Qtc?(L(NBJfI0FDX*|+7rWFu*+jyB+pwlwc*OrRgFGaA!*-iCldcLZ4M>EQPF
zpJ3O}&{GLyl^MX|X!S?}CwwH=UvqiB9L&kEfgD)M<*cd&85;m+s$_wwy6wq{)IFDq
zKXE6Ni|uPc@={XA8jO6!<`A%X*0d*2_NkvCC~bgABwWO9D?fJe^T(izXP6g4f-NOo
zz8%+s4`^4!QZhu+g-R9BIy$8;Z0ZS3wmp|ROCP2Bn}-eg`jE9;qD<z~)YOb4JI#wU
zkj#FmR#0wh;nYz+O@*r%AHlB`aB<7^>(|@PG}hJmv7T7kH@=Y87bmlyTpIf!7Vx-8
zS3MaT9PENFt#X@p9;8x`jm#dcQCIY{o}@SH=<M#cyILMlK-O4V>SAo!qH1BG!lBVw
zCwroM4B~L1m)0*>^oniD9~^iXVO0o<n4R`B$WFcpU3Ut2tu59aNcfndp`(*GHLQqF
z2F}jG{8(KjGfs`2hmYN|P2o8Amqs0`205frg{LTpjuFWwU;pHJ_naPd?BJEm!Q(y%
z{bT{QF};Xat#3WPqgk7#o{E=@;-Q<BkwCpVMltL0xgRXRZCv>d+(ua)KOcG$aUSk<
z(Ys}voqZ&_faMYgcT9qeGk1Fk5KUwod){!D&Aq3uj|d3~WshfPPm{5QD^vvxOrP7W
ztyNG@cd0sxT;#0CgZ56|4||zIysmB-9-N*vG#9p!>%1Nwqs3d?bs+fK^#ktW?e*=h
zpqu(syEi93kFW@64U6}B8eMMNus?5T`dpElI=32$wB3|w4YT}_)Ey^h@vDY;P8;w4
z{Lsxx#jS%e@<BYuAep+Bo1?I+jbuO=ho<&yeM?b0g+ha2dd<1FpDeLmJ3?Cd6ALP_
z0>a`rYD}pUOH9u+ghWr6u=rYNn}?kbY|-ZF7*!OvzEbgF%sOv|%H&r6r_f?<ke76+
zH^xYPeW%O@+9|S<JoucYd#p-&vW6`j!5re~<Lu&iW*+!pno|GHIQu2R17eK@=bX>?
z44v7to4@@j<0KP%eYSdN7f0T`61&4wrYU*7uDh8lwETwg9L_(#QSz+Zr1+9pl4!JY
zi>Z~1t}^~UF!W5`wTNE#6)_bIuw3tMx#PXB{n(n+cSgExxJ1ov<@pM+cAzw9pkuYF
z+N9<{4HVYV(No#k$}S~d3QmUTM(e4!LnY}Chn_k)IU!deqJ}#}i;7NgXylgcZJ|!(
zdnkU(7CdCVZA?~XNh-y&_SSkfdxj2pKcDlN5*_I`I3(S9E8${@v`ukqru{j-cB1ZB
zcNHeaz^TjheVlDE@*YGY8Aei%sHfB^oOx%A8skkWo|!wwq>gY$TH^3ky7xD@8>uO`
zxDWNY>*)l9<(uf`r@j;c+tu0_vrTYo$n99qshQ)@<j4K)jhvXl>btYj#|_hs8hiAT
zXFCKHU-i%EHExP-23r$vfMSoF)*!b-$wj>Ap45`FKnML{?vLuQr>B_X$wU3jEd%7a
zEu_Re%p(|r>qGqH>+~4ZBlAPcE~aMFOE(z4>?w&gwJHxZNjK4oyxsBq=uD^?O?a$H
za+fyNE*!j(Dvn)46ObB$e!S^bd`A_#rZfL_MssEbG5x|r!a8PjBcu5qVdruNbxNZy
z=e>NtJBva?ORIqUxcoXhKU%!Sh1&8tPT~sKBJMS|ba&UF>#&e}dP}#HhNfmdV}#2Q
zVM?pZ4P7reHF6_IA<LzeU**xdA+6FV$w$E+&VMO*QGpxZkLAr5tfLNKBMtvPeJU`z
z(rLdQKe_iX2N+vUh#4KO*Oov0klbYVc8C&3L!c<!3>sYb1FIvpkxt%6G2TxPSx=y7
zHd2peI$;sjO5ND~CAae`j%!ZeQN9iCahgaD<qgyxY`lT-z!`p>J<r|eff?|aB|=Kj
zd>e?&)mtG;6tZYtOVE~1DwgW3h==_gPTX{OuF;8SeNU4Rqe2Jk&ZJp%iCtWg5(4Lz
ze*_b=+fHJ>gNVCgMv47tG2NpevD_U|gA^I+b80c<c%d5+v+1HQ(Wx65L0Hfa$Vsih
zkW)iLd-toZt2D_CWrz79*W-stzYuv53Ys)xgzjKbbSHD$Oufb5-yd!vFZPJiTUwQ`
zj5+f(N_^trKXrV$Tbrd_GA8k!ONAo(6eO{<-r6Ck&+Hd`2L2sFy|#ZX!`1lrQ1*;|
zgG()_F_KyN#w;J4PiPT>&lDKf>!)AL%ashz62b;fck8W%fFFoX+xA)RtV)P4(ygX#
zPgI-5jlmER;`tY7!-#gwOoGU&WmF4J6tyXY=5KVjN9iOIJx2P&wlW?Bq_C8uaDNZk
zF<vW$xX6%5P4&anG-&?OQCMD4;UN=a=XXEukm|%>9sc2Lr+V)0o{_b+C*+%uH>^~V
z_<_FF7(fH})^&Kfaob6foqVSHlOOl*16uK|W2mq2^n^<Re1DgjH9P!`I5kcZ0Lq$c
zxS8Fd1Z_FEzo7Sk`k^4t-!<r24w(ZjLt++fFQ6|#gk8wCwF+}#q?<j}`$}>Nxqzjt
za)-3dEnEIU!)<@*0YqU4kWAFuO6T^i#qz~%{eZau`hl;%%=2BDcej>kJqL{-+kw<y
zCII}bZNxIU7@SQz+Rwyy1dy?{9{Z{N)el>^+!K&Eg!nBArRm+*pdGP*ORu+7gg44W
z_W7>I*Q$zBd%;M$n=gSLFhSkfb%#4)>yeBH(APE>`|}=9uQg~H%_rTrJhg1;>7kw>
z8AtTI&mgv_Oj4~LI1s5)a$T)0U#iB|o+IvB*Pve_7^p_sy-7XQexkcwsadZS7J7%3
zt{rM;&h;?_@oHaL=^1|LPos3-$n|GM4MJRxgM#)4XZO|^8D<mhQ>}RkzHfx3JKiG^
zMex_Su<BNZ%RENZYsEe5+KYSx+3UeaU&y2T*J)VKaA8xf@&+kmVMDqR{q=fdZL#}u
zG0MfHk`6c23Ow7+ShPO4+vaczYrmMKBqWdZROJ*LNOxt8kXV~%ndqak+r0DkQBm5h
zOK<#gIAIO?<+hLm_6$Twa=Z=^aykR~jFDIPwSXVMQh!{1HTZ#C3m$*MwX37+@qtgF
z4?<2WyN&>ahmOHPA%+NJq+cq&2$>2b<OBuH=O;oRuF#g{D%KJ#SV$FwN&QHDK%;QM
zsKwc~b9&qr?@CKcXDqZVvY`zCIm7~$#GZvNUiEpL5P}*SuLuaKB+j6%#>hzor)mxu
ziVX}FM7qB^A|x!V&ZOX(dL$D9OxyOIc-0=xk6UduretwZ#X{#)<#<=QdiE(=HGKMN
z)x$Dz1&{8E-CMhE{pC%Ee%JQs2d+R1kU(NrPczaL6@S-cin|@l$4x5ZB6a3Hq63e%
z`pdlQk>FeOm<G3p6w#hUI`!s@euSnMMPX0X117AL4esFiK*k+nV6xX6<fO9qjC<aK
zw#ruvgcRxn1gx^9bL@<%suY9$r5}rJ&yvTaEKekyPtC9zo6&>*N>Ea8@rm19btmO)
zlk0eqUI4=}o_V0JAJ+9{$`X}`!Z9*R=d=dRHXlU4jeGYge!BJa>kYK2)~Q4?(2?za
zcsV2oW-Bxdwuc{5*dpfl(89k&aPA_LVuD72q`l!(Xt1@fk4cBWaL%K<?YeXid@`0r
zoiGIvD*a4+73SAWGwOB`5%`e#7e5?`qp&Ed&kSPITPKg~5W@`OG>k;J8L?cPoyK{<
zc7U#eM&_1n=2k&ANcN2rZKLjB<KSrL>+njqq*Fhmw+@i4ciq2QOH<<p2HL9&kO;J_
zJ5FK;7Puy5b{L<R+<qo(iLzug8z>cJmv=+*PA`rVPxn2<7?kU{wN=26HE7$~$TS{a
zc)gd;^N0u*mZVOh(3Y#~ii(7q9azOq@3Nj5U$=F-d!#2TX|UBz{8<piLCl_piQ;rI
za?WMm!f3<ovfzkJbU(*#eDH<Oi&u++ki7&c?Shj7#%LROK~<o|)9|R@ae?8|r`1sx
zgo+ay)&r7pi=1Ri2Y<gEK=RXQovB4|$+zEHk1w*Tk6S>ilJO99Y6?G;N<KYNV&m=8
z47u0)koIsqy1!7>GN=BPYfv<ZjhmMod>sqBS*f3g1{!b0h%m{&;<BFU5q_CHd|iK}
zyJDcef0l|}^mM^zuKtu5$s8kmBqL-@Uc8ui2edc4#eFG`Q?)YOt0jv>{W4!_Y%EQ~
z+-%HYC+>xC%*#9c?`y!2A-BA((yR4YGqS2V<ic-hyrzZ3(9mR>TRK9~n^o6hUE0Wt
zIfy}d1qWC+-Jx`VRvy7l9jyA5ayEpJznDul323`}AtvC@e#K%#e`3eo&|rX<zWX?o
z-r|YFdpG=*&S~M6hpa(Q;)R^O50WC6LBBh+-{F)Sg?*k|popUa;{AjiynA60Nr#cw
zLuQQFWi4g>>I+CYh<J6waAaT==)SJ8S9}@eI+)`vSp`O;3seLnx5*$`G<7y}C9FZU
zdg)fppY#M-S{~k7o;=z{CyYkYOPRwG&Gh-kKbpj34+p=Fi;FcVJY-WI=&6n9y}vSI
zE*+?lCw6QAqm~WrY%^^K4vwKqpiqQ91J~y|zy{W{$0giy%CRN@?W>!_&cn;g8jw__
zY>}-lFcf)KQqs-9Qeo?CeWO1qKtdmgi5Rk(8w(zYi%*TU@ZX@PHR&*rw{`DaN_NDH
zY+qmB77*+V)st~P8DGiq-((lj5vr>GYZVn0#&cULj~suPcvvz*IW_3;CK>ichrnZ!
zQ21zxlyzamu}?T?I5I{guyxd_jK6V|^WD{YCh8Ola_%j=PR~)e(F(@qy-Qal=Qhk4
zZzd7k1uX$=xT_1a@^}P?yv0u_DUIIkhm^~6kWU`z6*Dt2jHGvukp*jegn==V>}hi!
z5}|b1+y8A1i|qSPCxO-^%$+4ktB8~Dfn2QTI8K7I)d_?71ur8EK;(iu+$#sd4>8$o
zWQyt5R?tngqa|(1+V#+@M6;77L$)QtWfG4{z2KOfP9I36)2h8}1^LBIkbTo7*J6l2
z3u)O-y#3RW4i`vNEUiuMjti{@pVlIL8b=;Wfi<J}5vdOmw;l|962UUl^a;SjcOpsS
z4-dD+e|R=ht982h66I8tY3HL;q$Q0O{roL)AJTRt_or!P4;<ai*Wys#3~pj?_6kT~
zX?a;7K$bo`vQMt%F#h#x*Rna&C+3>V1JhW__5C2V0%<sVdAHGDD~;@U?`2vZhu>d=
zo@M@AWn-ndnT(AtB5l&>-Me=WwqJVpILlz7jna9aGasM8_-mt!bqT45vvtfrBU%3a
z)?)!k4;#s${i1{{$-T{Aid=TIO~PNx{Nwjq4?hWf+OCp(UB@KTP75jC$Vu#&)hRKD
zXwkWr<fJ6%zh}LxakA2+Ik7{znjs=BEiIWl>uw8q|9H=f?A<?6$D(Au7%zYSl|&I@
ztqk?a??PKMA-<F>ne@qLK;wE2pckL=FokU6eG_Et-S`H>?}W1<t&aj4kRP%fkCKhk
zHUo0!gj<w^RlV3QRMN)FC-|9~siI-t9Rw*nD4WQf>Z<4Gch0kt&@2EwaOXhXP6$hd
zN<s)|czir3(y890k-1?!d)>vthyE66@(Kz`Rf$YHn9Z3H`39mXGum`Oi*%?rDzkLF
zt|P+f))pSBmW1WXx$mp|ere1L=;w&XU$a3;YhP&z&TMz&s68Zgv-hQrrS<ijWc#pd
zojZ<HGeFTxDxwy^U26v=O%7Z=;bSKtRYqR+tTUy}bo3$X>4foC+DJz)WgidY2rm_D
z8rMjjM5JJ&D>#q<#GATs+0)mHtIvzd*hOcIvLOOZTd*&Qov>3FYh=WDpkXYHw_!9c
z{;f{D6(a?tGD4>Jr`U{-uSE@@fFRqf!=D^9?ol2UJwgZaWFyO#U7T4nV{?2EVWZ{V
z#oo%>i0F(7pWkfFet3wt!@`d1Z<7tB=;3mZ(hWPl9n$<zWyhMJFQ)^@Q|Ret%|Ky2
z{?4>Fn)V`CHwv~v4Z0AT1YW(z)&B5Fvx`u}&XJL}k!?KCVufpFSkgQVhz#jkXibnl
zS&<p7Tc7lvU58sjL17nkF|;Y+afW(;YO{`a3bqM=-uVYK|D773blSa2g=DONIZ6YR
zUV0a{Lnq0Ys8xEZBAtQ!?^h`+5I_mmOnM|ueGczMNKSiKb~vTO!jS)JaTL@Noq&6Q
zzM#G@akSB<@njlH>hNQvq#;`4U2{O>olS22_U4s!kY<@0(Ny1~-6hu}#_Q++()a#_
z%<I*CX8wPz0?8vaI+gyhELhlvc#vB4782UVEZa6ml-}4Qv_1!HFNsGLjJ|)jOE*vx
z8{d11(=+^SO2<DN?;eb#h9w?&z7Vkf_oC|ivHg7N`6q#!57%yf_HUQx@<^o(A!V~Z
z`{#c)n+3l@A0`0ib>I-9Up~YC=Cn{|NO^6*)MJLf<=FY@U>vZ!3qp4MuyfTS7;wA*
z$3A$-J@iZ#S`A3IY<arrJ*0F)u^m|s#gr$4!EcV)zJdv$NH_A0ymue3?+?nVjnyi^
z_~g?+g36`qwO21gjYKr`#f!?PSA6z;ix?!}O!4a2Xug(%P1<MZ+3>O#^9{|^7m;2W
z@uDG7^XKj7<4GIx=QSr|CsM;&5g=~rwugc?)=*Q|0WB_^NA}uF`ug5|3x=tcpUvzB
z)gMSnf*M`Bx6R`_<})f18|l9ZEni-?Z%#coJ(OqEQEcxfZf(LhH@yibcmGPNY%%&+
zh_E3P=My!u#d3*IOk`k5_~;7#0B3{O)I9y+LZEU3uJJy<`&+ZtP9yXItFFdNU1e%k
z%EGZkQCCD2ph2&3#mn72Km6`zW?4X<z!J}oeY9vyrnR1%QV+Y#-WKg3ZrR*DNOx@B
zsr`lLo+GEhP<Q_HSB5=*D+6{-I`eS8CcZP@ps8HNla-_a-Do6bp0-kjPBBPY${SkG
zAApw8rI^?L-{y>A!U0oG&eK@R@GPTrcu4DD#R}?!&p*1+sK-<s6d~JWVqm|Ik@2ep
zK-rK0SEe_v3;&}AeZNSeTtsGw?OpqwfPdZRWq-5!lamN!J^Ed!pbry|7pe;o;NvTo
z>DC1xeWWX2Xf>{^HVGxQQPH%iCb}nrS0+myFB)kp%zLHCJa$`X>kq`Ix(jET{R78M
z+;7c4=qS;Qk<)h}xN}|rXT*R8y5B%F+$v>5(6>i^`29CR`1u3g0sYaM;GDC<qD+>}
zctztG>)hZTEKQZ<_;$i=o8C*$5`_Q!yJerj_ucw3poAiLlk9pJdB2Xo@PS`Q;&+Px
zAQn)%U+<Uyqt<)BEa|_={UEMab`bv^>+k6FH+%UX&HX9OeV(gHCZysnIJ`Ea@yz}W
zccQjwtv&FZ>zLdorSPnao7nH1J$uYv^#RuAn>ix%RQH6BJl}DQFODvsT~RLHVg1X8
zrtVkRlPR#9?=T6s?r_;HeM5DNt@j?W^+CZF!Sy(I_j`A>^v{jY28Yol2dDSXG>n<$
z%&I53an=(b#5_t)d;C?S`eRu*PZirY?kwGQgwW$RB(j1(U6aYUVcR!pw}QL<H`{-V
zN);u9EqZ=+q-&`Xz?vlWF!HR5=--CEu@-+j+gaSzQ58&ZHfwZ7M0T#uXxLcm*^ZSS
zqxsF=uHaAg#GO65rtiK4b<kYc<^&8+JSBEtKnRbDd~M`h7Z)$$8!;o&$Irba(~6{C
zT*G!vSZ}(O&R>}u`O_&p>VZj>k{oubyoh_ywgWxAk=oclSN!;4#1uvA;;|`nP6w2V
z>O{j^O3BgLRfE^X4e&K&ilNGE<7HZxx9k}`HrHp24#VNq5}#}yNw<#ELM4+&MVi7|
zi6Yexe?bMK-}K%lH^3vq+hbto6Wq7kZzA`58DYp={=>JCj{8~+Qfge48~u<eQe(T0
zlVf-Sa#>GN?^}-$#nb%<b*7(^;`)vDnmR;dH*cJ8NYaJ-US-aLmoH*=R(!^aiW9My
z?ulatQXf+b@ZJ+YbsRY}36hXG<#(p_AvV$ybA=g>$ZPHP9W>&r5SzQ{Sz>QruICLY
zw;#FJ29gNUHrc>3B?REJS7Zl6-+4^A;DvWx6Qirc(%6OvfN%N~sE@3`v_mGmxFn&{
z0={|KcVlRoQN7=b3v}J4ouw{QJ5K$rJNL_kbLBzcg!6txAo5l~(n~x{H3r1gns%Jv
zPr4ue`u>+YAllfog=>{;lT?s6iFyyPSK}P;I-5<uVgal{hlBLVO<GsH0?qx`)qeZt
zKl1xM3oOlP|1;$GKSpxHZ305LOIKVF5QJ1{HbJ-4jZpWtX%~m~W{-T?qYp0Oma|^n
z1MnkFE<#&y)dS3~fLpy>NW7%AhZev6)W1l)+*s5M>*%x7Uba0E$t>D+vU3iphmWVl
z(g!$?btQT1`a*>gTR^zyzf+f5Vs_}VP4DLpzU>OO;i}WmG!fBzhno>3>iY&_D{s0h
z!sZ6=>TkZ<d8`}~vA*&DA|F~MF|(voSA+d`NQcWlD8W|y+RHTeB!=^e5&6<Bq)$qz
zf?v#*m6|S1yXK|DT*>^^y2vZOO8;g;09U!+w0R&_YW`-1Ojj(B>eM&rmh1uSVuPN%
zu|EDX4rCeV(d?37`~70J<=(kba?9x5`cwbBlbZ<S{9Vxp1WR$YOG?J%$S$|7@`Lx_
zbm<ar4E|n*ZwrPOwDgAk<GMx{Bidn|1ol$%!mPmp@HU*>m$LNI3?3J0m=u<J@*kgr
zhDb)oYtch;|0HO-OTaBkz@e1)+e7@i!@t@5`9GnwV836rt1TX_|F>T+R750>0)>iH
zU;E}kUg-co3*#5K!?AI7noI5v$`5jqMr=jeoDFtVU6D#n-C!&KqNpft9u{k<^934v
zg$1$Gjr;cXJbn>b@rJ=e<e|XKN9lOEdGufai<7{_-{3y&y>N9uPj0!}@A4c}flxzU
zq#53Ff3x(P=pMsp$9sMJeovH35Ac)g0ev2b^`ys#3x7W`7#Fo2Mx(n})q3U^Z`m^E
z^V1GJWdP0idEJ(8zW1M|{?nt&pmsYt7;Ss<e)Uy21TdwffY8siUEjIVX)MYR631XP
zyJ~js)v+l;@jz>5h|gT~T`N5a^j;WE;%!9U>afE}DUe?06UmpEaQZLXOLP`S+k0KK
zcy;LaImp7N`JWE13W;Fu!DvL$T87Jo!2bPO&w(dtkCg7+f01p=6f(!*7JPJ1$&K~*
zFI^@DxjaXyJP+pnNy2dxK>_YmCW5W{x$}wh9nDyu%*`<(p#tT4gNfFVPQ5YG>XTOH
zO`J_{0D?16>TxlEf+=EVCu0rUoZlR<9_Ts9#g+RM7oAi?IhR1M{UJ`r(zs;2r`ozH
zW1R}IxrK#KHTjflXd_3RNuyO}U7}GYF0J1-bMuE;v*eEuPsB2W!nbFRKAaSnm;4%O
zLw<>RixoIGO_E8H_PI?W?6~4(yk|=0UC%B6Bxloji?etWn=p7k=t>(KDI2C5#_l%G
zj>csN^XVS>*mQTU^o{1}9xQ-7!OMY^;c<Nzj|+gnXuDE$P;>~+{<&a@CoD<^KP<;x
z2`W|kZBHLy_uF&7|5k9IHIJX=VBUq6Y{yf#32&`I@4zYMY6;w|uJNvLuob&9T#m4q
zrzgA>%WkD59%w&Al9XqkwEs?4=)O$=-YJu8bMn5}P6~=GGAdTa_Zg%a(3TtdhD$S{
z=p(RoAE}(XGy-0Fp*%?R*$b7G7Sgy20b3gm%X$2;YPOB%!z%O}t;RSlD>B7!0L-UP
zOsHD?2Fe}_&R@@-mx&g9QGg;B%!@w0OicK1t3rukgHe5I-_v;~(wDcQlCYjUd2&HJ
z+^6pm4VO+wu|G}7x%=3IG#}?oCc1^aV+K-X+xx=&zYVgS4=8SU5uI?L)TvduBCjD%
z*A1p|OBe(zWbU?YAQX06NiU6EzmiFpqERXvEOdT63e<b712z-07iwNc9#^`=n59!c
zd6{eEy^(y)Mdr0w+}mZNRF%g|@3wbP`}USYrc~r@ND(tBJ66m;R;b^6KSwkv=!`gv
zatf2>+|-;>kZZ+C+h?LlaCeq|-b<kJqbiXxhYN)s;h*E;$#1pH;c)1f`#~}R4l!3J
zM#hZaw=~%+?b_&V;p8-n>I)lKeNi!*6{3U`iA$S7KMww#DA0y@!=5qPOC!wxK<vv9
zT|OjmSZ0ijtD^_HBA0ww_FpG0eOa!CGVpKvi}2O)%!wd+;Hh1a+AEal--P>5A24R1
zYx%^#bp3uZD`b-&&xtM+)NvY}l$Wb>4DFymYc%AqQfr8(pn$l)8pvNA1>}E;;7Vo6
zl05#&b9_ZDp>;Jj3psU(E9gn~OyehojccLZIi?WHv1(1#t;05)A^vO7hjH1`W~;PJ
z@af7;Zph<qyLSpxUezx)mUB52X;uApf9W_`<2YBGdBuD?j8B{1M{bWG{4`YqR{=U@
zqbYOAY1&q#Hf<P(EOM=*b9qTFQo%swVvCv?Y0$DJD=WSy418s39!fCJ38ggX$*tA<
zl%V_-q2RZ)666Q+jQ6FbJx5>zi#9qq9Wcb#IFXNZ26;>SaYQgJ0RDm=#O0GrbCsI6
z(U>=~%@zt)zEM7n&@V<2Yu4{<;^Yfx7~f3%S@~>AsuU><y`L~7y`#k{Fz+@W(iGkl
zC%`eD&0~9~D@xC+8E>U>a0za^cLLU?As@PE#m-M0yqrr2qkv^qkH^Ha@MRc6H2kMu
zz#MN&8`#@VeP-A~FR{($apYO_k8J%-g5OgDl#9UI07q@rs}`rEqNk^4T2ruPW6tg&
z=L?`&>EuNJkVduOZKa5C@W-qD$F1;9s8h4(XA!Gx@=`RE+ey*jZK}O?3F(1qz-CEo
zwKqFW?={gI-`HrIWlwU)O^oM|*TgBirR3K=6%6Mrfz-k*SE5s@?;eL`stYEU>f7#R
z7mZkSy;M;OXwBH94;_5(d@?Fue0Y;bYeQ9ct=y-QLw@PJ-;*cd>~2%tS(vKIL$Z^U
z6c?%*<T}vDzU}FP58T@!O<Ayw+KsC$L|Uuq&89hSr;F!*P{e*-D3KVv>rL#LH$Tsx
zN*OR8`tgK>#B&rd`HRyF<{H+A_%W}Dc3l$74*&E`yw`g>M-MeGZFDPBtD-XPPg<gX
zp2D_V5mkNtgJDnnm!kcoy$5;ZyF}_<ZXn;ela3cXxx@jlL2pJB&o%WAeqPOTNPLlJ
zP|=8XDV=_Ksd+8FFN=&czu?P=cDcUOX?3n~JDO6N@M~%dr}mafBCH~asXuq03Q5CE
z-=$zh8BZL59oRC~FeM?EFSR(Z&tszT&v!fSuG><5AiuVjUbCyL|4xx_{HvG>ih0?%
zn3n@3gLq~cB0nzJYZN0EKR>@Di(q#M=WR&tG?I*9EVxM<P_2m)lJj|i;6#68-LD@W
z6LSH6{+nHmrpI25#L8ETa?WJmu$6unzGPJH-1bh<Hln*<2l#c!6@7!p1V@ARr8Qn`
zaw{4Mc^DD{){&JNT*q4=z!uxZ1+1`unUSrMP=(R(>i9yJW~`OWunUr2T%2MgSMr-h
zPVCFi(O3}CmiX9yYGr2IQj0!`T*%k_{8wt&FTT%2vM?GJr+D|0(_j2|i_;W}2k~aL
z^den@7Jp)?RV{L+C`<<8%}Ovkw<LkD=p|*_!J9>{9{l$cLxdL&#O@zu6uGTVwvdYx
zT-*Po$^8A+(M7>%IJyDvm%0L0a;8YG4mj*{YF4SR0gX89^@pCWP72u5eWhN-hr||Y
z0rP}`GQ*jin##=0J!!VKHY6oDn6@y0pBF$lVjSz^Rz?LU5$m#5^!TFJ5(<gVLLC&o
zjN(8n;&>#yDujvDlQ*qOZs67(_xS8nwY%~Jw!(_2bt5%xsjnj4gT;UJksEDPl1c@A
zpw#izO}9LCoO^LT`^|GHL0n-<ep)D~6Vf>rwX%9Nlp$~%wL1OcLCV>^`X6-S?`lO%
z@HpPaUcY{QCavH^*X_r+%kHx`Pu3WIkkhqBF+Vu>exSu<Vx%0xD89t(F+Y?R%M-$T
zYCmX`{P4)XGY1+QA=fo$H!doJ&Jb0nOx(Y}wO$;Bk=}Hn!1D#}F@pIdu*Ele5j_QE
z0UPzI<I3XGy3QyOCQhh-S`j1vVn2l>pvEnGV2~<B;1ekgcQ;<qkCu8EbroSXogU)V
zVKp|!(yHF0J&P}giW`l<KlH{{1?1r{nLfQsAx~**dz~l+uWLnD@>@j7D6E(IC;i}O
zZ6J|YHf6lvLb2ZyBVU!;`>8GD!FMXcY}`hpi&qw~D}9C1D`#Y0l$%#_Lw%tR_wyhe
zTsf3$mO6+P4T%~8@?z6dTOE(6#5rI%ky=x=6`jJ8!m$WcnbH)GAD!e&KKrk`fjWiI
z=hc?4UXMQ?X?}y*kgUcbY3bE*Ow%A))K48{z7?9R=nxk^T$%zma*=}E>bT(lBe;e8
z`?H)_it_#=xW$F{BA49H8|;1_6ZzEmk398sk20X4(Hog`bG7ld12%-~H*7dfS1OkW
z2;?RYSv!iFZf+7F-NN<pM)A#>@7^9N95i5Q>$iDS*#^$^W=B~~xzYRWtlAE-Hg(f-
z-`PEuljR+Egtl;VMKqrm!r&uX1uAA`3eT4%(gkG9j0tXQFXk7}u1|Vln%6<s@<A`1
zl}Txm5>V^r@FHzywsYs7Vfao|%5+g?L+t}=FZFNksnjuTBBY3=hR$C~^BH=KGS-|a
zGoSRdu102IUnJ~wd_gtxM}mzE=Xr%Dv<%)#R$RKVyuC%|L)=`Jk+ybt(Z(N=jXKjZ
zgarl-GD!(8&JG3M%?u3=*lufUOJGO08GR86uoR}{`TTsC%bJbGk&l%MdYiWLeOBGX
z%j7Te5xVJZiLW#4IbXo{-M2_Zh=$%eeAC0hoL3C5hu_t0oV_fYIG7Xd8O(wu3URMe
zBDa5%>{c`f)B_v}v&%{k%1d<mY{oP9B^HTZ{I>sXI{<Niq-}lYxe@F5Zi4WrJ7Cy7
zI{3nsZbf?HmxoIuGCDgok`+Gr9>bgqJu?^{{GCNE(|$C$(N}Hs$tf%(3wZ}us_UY_
zSl_0$+z&1f;5s0E&HKI9ctzW@;9)w%c+BeZII}wDo&)yeVV<h!<*wyN(Ti5ZWl}U1
zgVusEMjn!$c@VX4)~`{Mx#E2@=KJKXNvXcRX?9_t`GCoY!(N=ZViW7#0+a7bi0Rc5
z2~H}3GW9->e4#ISIaJ$v7D|n~nrIHxusVmTxt14tvmWNIdh~rIWdsieQcL>R5JssN
z`EWNqNXFjSD}phRQ5j}mnV$ZA$nWGbKSOjnFx_XFO=Anb{{zyT-`HfL&JS%eR$m?d
zpHomJ<Eh>yf|_&1)#H`#M&m!ss}wlfLrT{ARk-x9wm56ys|?K-Pc#!g6i_aEtA#S2
z9*7z62;GaUvHDKN+p4+A@)7u-@z**OiMidFSu+%E{N$A2$QZf~nLL#MIy-P8I7{ZS
z-TXiIpeOr@9joyBiD_;#BA|MPET!XS(PMUo4F|g*Lr9ebYh=u_JF-J}eseIS=t&iE
zJxfZBr{0d0+wL8OW2LJN`C^ro>Y&BuB(B5{5zd!B4DORizmlNa8cd0l`=ETj#JJc-
z{E<qaUTiXzxgPI?^_Q!*(t;L?AzT~T=-*ibPF-{YgsrwaI=4h9&s$8TcBmP$CZ2b8
z4`X2t_k=M6WFIMw?gz&Mpale)90f}_1z5anH4#!&J-jx3-I<`4#BrSNxmTf!+8axr
z^3tc7K@HmbnVT*2t3yvH;D+^$8~$J-3uFZ1UqjP_W&98;w|sSmgFDx9>vGs~8P5bk
z8ps7aD)H5^Qec^%`G1s4OP|*w+w}9p4-vwpzrpuiH~Cj>>38crh<QEsukT#U8T#v6
zRq@~ig4PTo=*z1hZ4g3Rw;BVjI-#Kjuz#4ByO8STR&V`c?m!0NXwmcB92}Kz8BTxM
z0Q40cPbN76>ya=OQCa0(=_(C9pU6$Z{Z*TE&tZ8fjUKo#+H2*8M<VL?;K73^-TFz$
z_ahpuz9h{x{_)lufvH|XQL9rs4zsbjU*?L=#sT+JwoMgZ@axC3lPc+35-M=66h7R(
zg}1l#pv~-5j!Nbgp+y~!iyZJGK>&yWL9$nqk5*bk{tm6L*Q4mB>FEx^ts`@F8?Q+s
zNq5&-E!m>=Qb$HLG|!%##j(A6H^-(bW&hm$cB;%fn-1U>c&U&$@%E8(=bpg(VVizO
zgjA(YC_P@^-XA7UWrO7*Vm5cDd)ip&#QZQ+ED)6?lR5G{?^#AniGXiG3iWYJFIKF4
z4UTKrTxjt6H>YU*2`VEST@)pj@KXLtv%lsrqx*2jH|f;J5P!+_g+#+lTTwMy2^%~t
zXa3P(A~6G`EaN3HM6F48FemN*K<&3|ca{#~t>Ey7?wntDW02+`N4rW~v5fHLFOJLR
z?bR>UEcmiQsN=DOgm?{oec!^vE>AORW7AI`_LKeo({E!e`oRUsk*B}N`#nUC;>6L6
zmFLW~|3`AT%YuUGo?!kgm`36S8JDMLtPx;EN~@Zhs>k7SEju<;i^Lstcy@iIm*q}y
zQE`ZV34tyNPqDDSh5Y_P@fr#sE91JfSI4iGzqeGz^Jv9)mkCf`%$1_eK=4Y}Z994>
zArq8xJj1L+@cP{^@9#YGku`8a7J(b%*+lTJFQz!@pf=?D9V2dg;vN<;dDOIxQN`2w
zvY|`X^aSPBCr><1u_n$=+5qtRi0i2;_pUrQu}N9*=BqIz_QV2+TH|d!>mj>{KuPy?
zwzdIV)cp*OoGtTMJUqMzS-e8>|Ls*hzu#(f&ohR*teV~uIx}ic-tXWCNn?8EODYD+
z8$qwdx2}Bn8UJ&rxDpSjr>AfW{VmQwPjnet;)u@%P*}y$cEffoMP2tHzeV7iDxdRf
z`S~wcqJ+-9dmFk5G76a@=Lpk?<MjFiKj$hf_gc6ois6jA`F>@d{9k$dpKneg6}L~1
zXRXX5yC``A<3A>Zpk0zD!{2r#{X!uB&24$A#lpk=Z{Pmgx8G}-|F`+H(>`J^&dRG;
z&;+=>b&0G1wOI=_7b=gRD;yY!Nbb98<wg$IE?r25BWJ^IXs$elQ<OG!b!>FhRwkvn
zttofkoWT43Wau&`jt&kW+sG(0OBa~eMjhtOs%~4A+SmUEy&;W(mh=?Tc4Yu`lx;T7
zzvRw+)mRp}R{r_Z13V_>>Hen2_AFa&-4c2r9m0eNd=24`y)h+v>Iz+>(8J*Mg*;cA
zO?p;6zNcl9Z{iXiQJtB@ztE5(Y3HKZ&Ry&~miF0EPjT0kk^p}*79?<j>cz^o$ZEJS
z^22Urv3%b66$@Y~IsNlAQSMkHO`8M9-%i-F9;;CTuuzoh7*>U*@tC-I5DnV6hvRmO
zlZ9RaR30nTsviQ<8r=UmlX6yLdZgFt(Yer8X7&kKB^3`?IFGeLZ~i>`7g_#Wj-<Md
zMl^i=CYork(D}j~@Ou!7Yu9JVScNLV-_+k6CsC^FyFLX&yX-;(7P1m_(IQUF6W$A(
zcn_%<ljRKqw9OU1`kpjq)+-VWhbJ$@=PMkeEYP#ok#osaSNpwv^=6NUNQZsWiL+mW
zj>GQGJX_~VBAYtd-g?V!-rscq%-|-R-4y?R_zUNSp58P(Hr63{igf-IWw4>bRHoUJ
z8=RXoz17iA3b8e5U6L|NOUpw8<!skYO$9Otc|(`-8>k>~SX|EU*I)5cza2oXJVrW9
z&Hv_i01!MRi*KzD_3=L+^xNn_1`0q1FGoyPC)Gv*di!9t(*4QQFvrpRoJ>U_Y|**H
zBrAD3dzTG4g|(?r3eIz+4vi3d2h&3m=#AHpt3p3kuV-VS*4a<V#CfwDmq<c2eMXtA
zy+rWQ4Q~_Lwof0XGVO>3`^5CmB8oibDy+~a=;a?^o)8ItWY#m2Oz3`IVCSaq#X9qG
zLkTD5Q9<?_LMj(ZwEY1A!Ia604Q;fK99R=4+8fDAT@1GF7RQjI*KM`D6R1v`QE$<I
z3w&0e^}S#Xa=61+jH8TJ#kyFFbsX`U)A_qDTVwt$;k$T)jphB0_u(=Gz(Nh?2Xgx7
zPjq%L+VL)TzExQ(x)7XBxeg7l7c%DWugxW-3+gP4Mkf;7`rY{X_XYjB#IN66-Vf|}
zWq8Z#6lK`$h0)IP(1@=NbF%{JAAVwT@Q*fXAtgu<!Dx16^k%D*1mF!JEEm$$TdPAN
zj}c?rT`X^PjP2Yra7Pt(lCMtTJJE3%?eR5{fYqVjE6AmCqF;4YNCZO%qY*um99$j0
z!ePSBg5Jv;2&q1&_ttN6QB?ET@R`*u=k^_{tvCei+qFD!m@UH(-78U|yrUHz7INqP
zBwaj5v$Rgs$@{7|kDkAF>MVu{%YJ-KnB|e}caEag%CucLDtiotw`b=%Gj#&jQ%7@$
z3wy6gS03~(JR?5#$s*mtM#<x1*-U*uZ-<HP8S|te${)5xgtoM*iUg9ug(hpc$EVPH
z#-zsLp)2?yv%%JGtWLj$6uy`OhPs0dwaJ${iYF~X>PBfwKPdb^_P#qF>-~-UB+AGr
ztB?`NCS+vY9Yi9tlwHW)*{i!k$&7~VvSnw_$SiyB$liO8=ks-b&4ZNt`Qv##=P&om
z&G-8m*S_A@b+tDOPW%rwON0jk8bvw5+3zI8-rALVoIE{pksdUh5QYIRrVhpWZY`+3
zFwjgT+yu-qcoyQLugEnL^&E7#uXdM}TyNILPTpfv)J#;UJF!WZ_2b@+RY$^u4tWwx
zY{vg&XQN&fR6AI-e_GANPPjE{10)ix9r=8_#gqF#i9&cM5@<NAZ8zf+Squg-KipL_
zd_5E?R=DtECO2${siNujZgef0R_)a9;~MZ6WnTm-OEQcv*>bS#GVlK?I^8aRdUBb`
z(?@axmNP6->N#pd_?;2_&npxt6JyV{w`>6o#OPS&=6e+`tu_!qvWuO$Mj)fOCV|*N
z$!a#<0?1w%I#TX^VXSo`ufIE68#3&cKPRcZ<8IYAlm~yVL0!X{F<T?>Vcnje&5#&`
zh{&zsD|#q``G!>8zK2lVZiJY9{0dQ{W02Y`5u<mw`a)_Ha3~GDfUVl<wF7~A&cMsj
z<d&v?u`T=%vcfdcT}~a2zYL+A^)cMzE9R$?W}a@<LKqNCnZK_5<N;zdunWk}8_TNW
zllY7(cJ>Xr^NvGfpiu(60sPuJNMU{Cw3Gv!Q8DdV{-u-Dr&#9nE~W+_q+(^)Op^ix
zgmKp3OQ#-4*}z#b^xNj-{MtXnrJo5_dfMK)N1YRriUjH}49A@<zJ!g|ZU9$Bf&rom
z(7gtU2%QD_hjxVkA+oWtfmb0Oe=)w?U3_nPcY*fPI&3v;`hQGS6IM`gchQXV(vT^*
zk+(u%Lzj;mAO=|PdjmYpFEVzk9pD#5yf-5LO3mLsgq%jgRZM;?@gtKIgcLZeniqfc
zW4<-51rOv2euPgS1l9BHL6^dWAr$Y$@i`}EUfzf+d}<BmS8q(UsMEi?HX19Iz4DK&
z^+O-a-*lMnC6uh5bMxhxoaiC}4T{8~HByg}d)>Wz*BRM4Oz3+F5_W0|%$7Mdku_Qx
z?dxwH_~BV6S<3Ce7t7pOdpm@X<47k@o_tMUI$f;#(C3QHV$^6%GDz!x?tOITntZsj
z{g!5ji=%X{G=1(ZAx@}(7PjfRxnyIvkTMu4mkjD4MPVUqaH6Mwf+5)Eips8ntBha6
zLQENg3=U3>SD|kMViwm&M;<~U@8I$+#Jsm--kwHiQBk!;Z;C;F;a+Iy7k^`l4|YnQ
zUxXOz2@e^WJ&fh_ImB3Yndqq6Wly_%5M%D30|6!T;t3IAcB+Je%-**zkNi6tAV)a>
z5>w)rzDt7rC`5j7i9h}^EFSFCwBM3z4-)3r5B9nTBnVza*v0Q9$Tkf4-F9}NOLBRz
z)65y3J(*0uT%70y!1U;{&}3Nz(1=$o*s0R;$vvkocYS9VIbeDx+UPH+p#5>{f8y;&
z8vi9}{Er;}6K{XS8Swwa+g>T`2fF-EyzNa5{H!7V6K_9|(a*=T<Dvd1-u@GByV_!2
z_X^(QxytQkAi-Ht->BqL4^(g3olq!$;}RbqZ&gh1;ejh@C)pri(Vku)h`r8lnAzVn
zG%?`;x}?}M&y}eltxo1UetbEfK3!#*$9Mkx`30K}DB-#zseb*+85gI9CBPL*>g{L~
zX^<y6tUE7N$TO9x4bB)?_j$9aPxq3Q#HyI<FTd1FehtD3FL*3flq_2D@4q`DVR&u&
zgI0CWxh3m``!6^Q6pBy3{Z;_6GdHmiz^jrNlAU*#_2YO;pUF1y&IC+JCV`p~3sV30
zH!ui;2%oFa1%%=kDJ1s=SxwqXde{s(?Fpd_uLWL`8+D(_uCxm#R%cr%E149hn{sDI
zxlFVqQG!`yvBHeZ<yp^|lnFSy*HWnT<o5zBh1mjU<!y7j&Mx<J8owp!^Y9akXGV&b
zpLO^c9YEn(2jwK_!Fq9IuGJ|`{%w0OWTh?w>s~V;CTN$tZDX36tLJHVxQ&n1k!Np<
z-$r8Q*`7Xmd{zOYl<Cu~{@9+117jG-LqaR+oA(&)+-2j?S1i<E-RifJ5d&a~N(O!e
zouumIFW$N%_5Z*a1uu}PJ#?P#UZKOz$f8~{w^aAMO4}wlS%0d}M-NPW7uM8?WU<eg
zc29T=g2RJ_4L+6>`OB4?@57%va89xY?1e?7kigNLKK_bemn*Hq;p+3c-%Ckm?MPId
z7Fyv#Y9j<1#8iQvDMx~}Y%{LBWWNL}A|>VDOUUAQK5TG_*&H#IwXhH(;0wo`)lx=G
zL_z|@UbK;_!x3_`uya5T$f;Ll?d4c@mX%+q<t2z&)gT9Lb?n+%<n4{<{(|d*oko=e
zm?I{+$xZ>-6^-{fVs4=k>{QB~dlWIP`n3O6yM^1F<@=8kh;l3FBe>HhkRcL>vf8p^
z(*f<cg0Oev=U;Z1s!gqQQ7;;ELm>z}S+$|jVh3uevx8yKOi;WFAsacjhML`1W*@A~
zqHKkkY0_cq!4YH2=L!~GEBK5$kCIDWfv+sut;rUdR$Y?^lvVb(TH&V%TmHjQ!AbOw
z?~h#y@j}{Yk}FH1`3Ja+cPV6}P-mUFva&0urtMA0=nfGr7zffkr!vQ?r$O#8zkh4X
zQ?Nr}oZ~AtCMl|Ht~EYH5KEI2!+W~4<Tl|<U7Wt|>B`_)QdDrTUcO+S2FIx9TN*9w
zG)qVK;OJuJR;~r!gY3+IcxpYk$G1aX<v_x-r<Q5wzoZ1dP~$G*VwbNJr~sV^d!>B`
zdfx{rSI~o}<)k!1z3O&(=^nqTIRzvk<Gr_-2zi>63V2$7U|iK6-@ZHF!U@oTU^*IR
zfS9Mvpo=pY>Fs`}<mT{+2?>+c@y$RRV#e4_(0A9URfPkLhOKxiDYrJOUq@b2CeYn;
z_aN3obx%Hy2ywp|LN2m@W$XUEVmfMYZMP0K|Eo#gcccLKMTRDXTdlTlQiMNgd>K-3
zu9?AOg*n|DCE`jT5e5On!)m($HZT<a?946CJ%}AK8kQN)0-Bpe3g0(--rg`IA{-hX
zew0#mp1ul{qm(dPfVQx|BCrr-poqj+gcyIpAO%xMJeK@yYX>JzrWd>tetfIH;_2tS
zE~j5TALRAqfRz_^p4)u4G=r#;+UJtWN0uRLc*vyobDp%mesb=q3XEW!a9Wt|wXKGc
zyRR?=A8bS+R2zHOEfW!dk`2A0=8U_r*GQm6x6$0O(Ngk6lg}k*-y}cS{#d`K1Suwn
zJzho{e*ep#d4P(Ng@wgSJPut#zF_c1R*LMF5dOvTwZZaTIgb8Wz4yOX;{==;sSIx#
z#8{3rFALtk!VlZIRo&LX@zS%{uk=iNSmq43c3<1b&)-~souuD7NjF)edsp4V`bN%#
zd1;zWyT+pJQjf^zAX~ej(>ySz-N@n}>%tERkutZYl3i(yoqno9o)lkR*<3oM-FILG
zgL%#QuJe(I<q6AtM&O~Cofr!kct)tY#&`fz$Ky)lt`CFm3RZ#?cw*AXc2|1(*TL`f
zSlA4>-zQ|DhxYs*yAdN8s)fw1!Km+ukj0T(1EMbXltN4eAFK)NG?^ex;Mc<EJHffr
zAA98hr=?FMYNzji=Dppy42+vQfO!L>9Wg`GL{&hVXNIz2x0~Agy1i1tPTx3<@5TH1
z^{%OyL7HdzsO}z)_}34^DN_JDZ7N(qOcwhIA;|5np`P2ri+}B2plUFJo$B3O{k0s~
zi8sH$zj6$a#ddf=w1*e(PHI~RpG*LDnwq%2gC+d^i+{a1cpqSTDuKT-_weF9NSPa8
zr)3H)8^3<>9%vYdH(Z}d*iRCK{3e`ibIEyVXessqkJK&@Y}ip`Q156rAN%}0XZ~^^
zXt3Ka9XpR7%5V-RDyP61leo9TtoreaUXfnYn@4bXh0kAK?g#JMr;z&*_-PzYUoAL1
z5kRaaw?qHwI=wd{gB2lxS!gD8m&ja+4VQF^nzMl6GKM??KT=AzX_<7ZxMB<|>Fe+j
z5c@bprRiQ?zSR{0S_-@t<-CaMtvwiA-xv}mjgq%M9_z%0XU;Ve^R^(zS_@q(YEJ9!
zz(3>yKZT~>*(Mui+VB-s<dq0y!~>VeJ3q*MnzVv}N<~-Ykda-kU5sKXcXB-u&tCf;
z%n#XOqT@uf@}duGY8c`Fo&~EzQcMHHZ*$l63!CGKf|G<hhyK`-ljeCrmN{`nqt|Xm
z5ISw5X+N&ho6FM3F$_?(bW3V%dIhr0*GG$mq^Dntvcda`u=9!sd%L3*3BjqpQ8-No
z%yAR;j$V#*XDWo9KnA@UKh!vrqWX#zY(5VsioE^LBjCqSk67`jqD4fZeA(q$5Zu61
z$BOwvK;K^VC%-|5;(Y~0(nLG2?l+iR>InQ%F^wSp_XhwVY%h^yBkmN3x{%6M4!8{4
zfWASs^(po*L^{5Q&;7K38aGfZ>LMZ{!p7q}+IBUz>1ixQClxJ=^-P%6I}EH#$qNjw
zU3u%nRwL1Il!oKy76xbZTkCM!)5fhP0v$lxbm)0*ifV0di(aMF<3?&7;4`whRtU|O
z-QuWffX|sy!p{H6zc#?<>Uoe}(pxN_HvjPyS(<T=u*pOwA%3P%>R0~!*NH#DyPnh=
z|AJjd=1HoS^c|5>H-@y-GjM0*+tm&dYa{Z_9CsxA2vx277uBl**hKWF`6icFGZxx5
ziA*G(k}4&tN!u-dcrF*MnY-ms0ely!!_tCP*DTh@E#TQ$z=?wg57It;bL|TeC<551
zsjJ|3boM8g|K;D&Ae~`lCH953|Fyv{fB56Qev<@0kVqRPASu48?c0+6@Rxge>@i3K
zO-=bC20Zc>60Np~>K{x6n1`aLhj!2P>$`4P+Y`u-g@2|+@9Cg-eVXWXu+zdK*<F;@
zZ`?%<DX=@o5!JAIdoA#%6db|_?DaN8^<5^v33iHxEjhgZD*zjU*#c?(7Gdp!3&Bp0
zkp|U$5n}sVBYXi0x_ct3(f0$$xZ~vwM~oReeiQ`VNfFhr`ac2pXCnPS0k=2n`)5h&
zm00lpfu8*TWN!@t9R5z$M$uTt5n~~7M-v+uPkHZH7wpzUHf-)~r3ZXv)v5(1M^D;}
zrw5od;RB0Cfr6!D)s0tP+vyD-5dTtk2fFF$Hr(XC;i*HOA_X0MD^ssB41!LgMz`ty
zse9u_Z7}7k-*M^%%mg$7<7yr~TpZ{L4fCiGgNHhHVQ6cI$LB)tVS?KnF9PKq{UcJG
zkkCnFhOY=z&74zHJ^!WD2KqDp0||uBjUvwL3WcP27dP!8JPiKpmq#2!0}KjR+2*hA
z%g@LC%O`yQmKp%vcXrNTKukU0e{1{xZ1;W<|4*J9ZDjQGFTDVEFn)JGjsIwG?-cU=
zN&f$IZY)Wz5~E*c|9U-#feBb0HlyRyQ$?Qd<CU{#=qMkVaT>G*2nD+SIlkd%7y=#)
zls%HVE#`0fTFuq=qeNXq+mCGtqXdR3QEl><7OyN<Ku&W4jD)!2ma5n6Zg%+j&d&7r
z_r3$QSfk$=u+I3}RQ2l7jJZ1c2W)bhE&r?B`jfT8Wk6L7P5<?C{|idsNqnl}nPgJ9
zG_tRxIT3^&`RmyiWD9&69h$70^E}IAKU$0iDnCOya$#S$^mPHSBw)Ma+49Hy6b$*3
zV+2MDI;hYH{YwxaoPLOe{aVp)d-YQKAg075xz*zo&;xt~l|hM%8H;c3(+E4SSY+@@
z19xq6vp@wM_C7q1fB8bM(e1|6$sYQ#sj`n`#V~NK&p83S6RZ9|6?lKLG4ccODr8-`
zQ-AWm|5i5t0pkCRoc;sEZ_8vm2-<G~;6Fh8AIY-+>V5t#y7uy$|HRvW;_a(>vh9ce
zs<r>fkN|Bt{;SshgOmHOT6+(Hx@}C0Fmjf&PNy0{d=mZ&{f%NezhTbwK*n%HaOnrt
zjppfL$zX!RWOl|BAI1Ul6MW*z%Fb2?gXus)kZks5duuEA_1Oz1WfX@eHzsF;H`hRW
z<NQbQ@#Riqg3{ltlWktShvm*d%_v;5*g(q+(8O=_G0$WUk6}hnWYV4C+_Nk!iP^!x
zSuGY3uo1h7L^NXFY1zGg{~Xg}%?}fg1=f0!HWSR>9=Ky}UaY6O8pAb_apQiz3R!)m
zB=Pt$&waQeTf|8NVX0(1<;_567MYN2TiLa~Gn^&!&E3c6JzcUT{8%+JvfFJ`9UUFl
zW}aKVzbbFf{EXSrPDx8(_9L6gV&`ScP1hxxoF2>sPfy1+N|U)-5qCX-f4B7c&4UwF
z21&&cTsZ)ZGdP;DJfQe_ZI;<hA&#Th+WIg%XcgLNvfM9U61*|@{-HU?%BPf@m?>J-
zuH0t)1R^KrXD+5E25l?M_kNQA<N;uaq`|#tINB}+CGMTmsR{P>4aUwkjH~0E;U`az
zn_rA;A_VR3-Ir=T<J(7*_$FVRpTBEWe))EZw@4n92Ul<6^?V1xHwvk@r<zBgS}@Q%
zax&8{!Sgvo=9y=V*_`~-j~pM#URhSVbLY-jYRTTJ?{+!K3MhW<CfkU60ErPlW2$h|
z4U9AL)TySGZ6m8myjfJjH5f443*s+CX(ed5)bB5ySy#6k5+r+Agk}3KQE$XD9+VO0
zt#(e9f~m*6D=RAtnehn;jx0xdD>?@&gB>q62$FTgkh6rMNMIeAIBSY3ih-51W10bE
zBW%1DYnrha)`?^l#Y|Blg43X4nX9%H_((yi-43Gm090MXCmEHm8nux}@uj+AP|a2_
zm}f~vI#Vm2=ZiHA*Tt~Hk{Vwg&D(IiyOp#U6Z}M&zjSqNrP*@)bM0VPXKU*TpEHsU
z6M-i-4J%)!p4F<#pJWT@S^BJ1+9<DjG%|3(h|-y@;X?qwy+VNXKn@s#nTKm+{ef5{
zZVmMI_sBF8I0+_)A2*LfiN<J9<%)AO<XqJo;_Q;Ol->!W8I=rMAG|QO-1aVvPJIZ<
z&z~l2nXO)3DOFizTrNlj_z*ZXva@Zlw^}eYOc^x6f<HiZc1S3zJ|>}UC3gC&Kg5Cs
z^`)}SbNDphYeJqy^({NBL7!+o+61QkTrROI^QDz7$w|Y<0D(D@$Y6(ScPx$C^s=GL
zB+N6m2yomWR4;KOoSa`0wInfl-!omz?7-Bun1?H(khYrC|AEZ2+HvEubnK;!llUKy
zzMSnJs3|Bw5{B~r7&RY%F=&AqUQkS^O6t!kaBhGyD{JJD<aj0^J-&g;ReeDJHVFrL
zcz8Z{m<^V;cP$L$k`j-eoCja}6U#=D;>fG6uODSrOx9?fAb-qOwbbx%l=oYjX%{0t
z<SLQ~+4{?<{L<p_O=KIJw9p-G*)}LrgjwbR%4PWAHbVuB=^brSJwDVDy4v<^j|rL9
zOc_)+mJ${1mWs}B;#VaOMwkuQ@siq(O@r>ar)|4(n@gKj9o2J|D@#jFN-Nt&&sy$5
z?t?zTsj&6K0U6r_A(vDaCXmIdV#JS6jmL~kg!y=d;imm1K#rP`$aT!^?s<>%EJtEd
zN#0-ZzA1E=5a$HnrZI_@IvxqW^o0|6$nU*-v>u*9jmr3lgY*y?`4kAyGEiv@r)Fix
zY{r+ojShR3aQ<rJyg8rw+7<i7uEp%d-0I3`FS!+{ga`=1sPwxtwetk&04s*cT+`l8
z$do(x0e{d>>@>Z@ZC>m?SBehH?%sQzRsQLETtL#Jd$_017GemHf;!;}{*9*zc8o`k
z9PxF>{Yt;PeF*Ud$V7iJOyZrxe&R3gB4OfQG`fj|Omy{AOzZ|FC~1!9oqJL)S6nT(
zQt+5L{)oJi$<^A@jRm7>wDEBR;nGsJ#LpgPMJ@ts&6;HP8_R@6i+Pr2ty~B8B<01>
z-Y941H6)(hkqZhQpwZN@+?*qzhJ_pjqtg0njXyA^_lKvcy<4kkYkS!&ALT99m<KC5
z{alFvNJ|VxGK)_0=f-{BE`%iQT~(lH=KRNiOu<&2`23BCn#M+Djw^f8O7L4q!*j4C
ziuD~g0!PD_3c30BGUJuIFiJFV6%Qc3wx+Foah7kbvs)P_xm8@nG4vwcQ-%iR)I8zG
zq^Q^eIblO~X9DjiiBjp#CI!*+g93y#o!V4RY@nB10Q~ffzkk=B03UJ;EWEr*(`=+W
z@*8S{U<ijmxuWOCIM65!PKRMP$kXgH7^LXUIl4R;OgK;#h|B7y4=pkN;RQz3W))k(
z+F?Zng~D2q{MaF@bBDl`hCuI#%>HNX_n6BmrO|#})U%6X1ihxc0ezi4&k%i#T#5(u
z3%5DnVfEQ@fN?4<{p5+R`FH_h+A6Q|QRb#7O}O)6vNeh8$$d$;F1s}6+}>odU3l~A
zbK`NVaXoBQ+p#3Y)@T*0p5jF9+Jl;(k@qAAc0MSahdwZa&B?CMR21?^$S<#v%iryJ
zaygGn<{<^oZ5Dvyohp}f(0|Lv%gfuOIuh{=g~%|2BH!wLlQoIM(!IEB-8%k}C`;$q
zguJw|j{K$Kgu;6$vmZ*(c6sBsuUJUoI`t9GLWP?mjkmMo<AW;GcccI~5jstcl%SF<
zE;$Z1m>ZqN4Td+$sU#CB8LNOf8BN;V&zy*7?xpEA#K4MH=0}U}V{hD!ZJhC+r*oc)
zw_LvwhJ?A;y{YDKPhm&0{$3Hq&`iC@6~NA(%+tfya$A&(i?ekJ4HeJfq}Zv@1n1y^
z{@3cpqB2yl5DO&wYdJXanFColC|xtpC=1)WH=2_2^f}rHpQpx@IZo#0W3}pAFa*aL
znL1HG-K68P_3DB+jhd4R@xuwO=h{eEO=RMS|LwMo?EZFuyGgzOX|9J)0p%r~XDSDj
zCjsy2>6u!t-n2b_{5Xlf1vE#`CwOh>I*+An3|HYZc)5n@hue)4jteQOz5#j(xeG@!
zPhrjjSVng&CF6XJV7@!vnf+EO!AOwtv>z7HmV5sNe~G#_<zo4ud&A;VSI1!BVknE&
z6)hFjIV8-+P0`N8o9b4e0BL5XH7jW{FQF|fHe!5zv757Qt(t#TDU9pFo-T41zU(EJ
zS`Vi7D2neFy$h?yc%z!nZ}*wFyTXiXERoY$-GiGhfw}oOQH6fJd=#a3)P~5&YtIcP
zUwRzrNhHkG@x_t9+q)x-id_fOhwJy&ka1`EOCFEZxT({dAy(%97d+zF(ujNly0UVo
zx@{1vrr1(LB>&l_u9*lW_=CzMa7(iWja_7pYmbu6#`9KllFv3CX)(&Tr<90<2~tt-
z_gu~%l?tJtK73jN2|4CWqSf^Mp7?Bsmz+JVTEQnQS~UlA@+#e<P>7B_8gUHc-&m@v
z)WNe4=CQm|o1}1LENhhMCZH?LNR?DwiQ^<B48&)4v7z?#6`+`y&HDEyEvY2H8&ZJZ
zu^|V2JNm6xyTP7|Xd?rBo6|KZb>Yq$teO4lS2G<!7s(G*Y+6_BG1&W#+04f-GWyF`
zjTaNjKii|lu*<T@v5?T&-}m_T0g`D%27@A;b6y>IBy>vYPVI82oQab-HLlLOMsgNe
zbY|}%aDk_rtNOcJxwAFI84p<k^(<N+e|`FI1Q0Kga!8hdM8{C}Q?dg{5{!v<%PBq1
z#2akYpL}Jnps!ADj^yOU5d$GJ8h>wGDd~Z3UXWv+oyq;y6fGy+1Y%8WvjHJ`bWkxd
zD;<IR72LlEiL#ye0Rs(UR_otm<}0HeBd>XbR8&;dQ+GL^gw^lk@pmA-j0Xx`KPFiu
z`7;L768MZtQu~BE(~~Nu>YBG|0c9-95kYVuFc4^5cNvXSbwa;C3*x666HkQfIlI4}
zoq>h$BViD3a53DNx!!dPOsnX2j+Tuan8(-ENNN4MvwiOLu|~$4L{rUWB+U8<iFRsF
zn>%+bsyyC)e0h#p*cDgkrziiez745G!PC~B!+ip7shtsYrC9+3kL3p|&E6lyC<Kl=
zcS6E-N}os9LMZ&+Oo)ucT#a=NXK(7RmCh@8f87}0XyD_p8@r;asv|;3hP-hIt-=#X
zp>}#>5W#SckDo}=X=`sU0#*@c6O*M+<weFCF9y!5!k_8o<C%!NYip(dxtkP9iS}-?
zK_pB@?}ZkU3k-BTgSS+QSLpqfcj_Rctppd_V6w|FNvB`~M(zVK{%Yg@X|r$?)BJ>{
z(Zy?yRW$Yl7lG9$Wdyge-)Oc~ZTikg4@rcDJ06;{<aRnaUvdEkcjgLs0z|;^^n%`E
zZ0GFBjT>FFT7mO<iL+Ha{qs+Cg}~ZHsl+5?r(3|~J++T`t*+bpLIL@Xg~i2}Y3b?A
zGD_-)By>K$)oA>LG)}pUzsf+<)cr~E$xbyZ52?IXOF5p}<v#&GOqdo<q6y<1KuW@o
z>YHUq^Y{IyL1vT>Ofj&joY7}hKwYfagzqx6Ntg+|Bi~PdKtOR>I<PmN!|f~Gh+jF-
zO(<DqmKu8t$w4OZg_{Es@-1esE7%V{FlF-^k=c%?YH^Lf^_g9%ILGQIp6OLPtxKrg
z@-{klEDJQv)l}C#>P3I#tfts<KKuCOqUVTikI|JKors+k&UbVTU%>#jm7^Iwuo2F{
zf!^oOS?aCGX=KWoQm1$t^X8OJ0}^Jr#yr2w;lqao&nxFWS!{^ze@xYPKrSh%2#c+v
zs;RkIp5x-4UiQ6_^^<OfbD_#jMn3wSsK^bZ2s_e;=K{DOj(XU7l@Bs9?_=|`BruCn
z`d3~=!zkBal~M8y0dO0dn>R-qqP=@{k2Z95U9)J*%iBmki}cf@cX{jfU%mfEZ&9LP
zejk^T&9MC6)IKL{4cp#&OnG|-ur<Z^Q3I_b^vZ@TT3NXkon0=4M(8wkW=m63Q?h4Y
zm71|}qJF?%`@5S1NbXEhxxv(ibAf<y;Soo#XH1oXfj^})3$_cblh}cQDYGlDI9?HI
zi(U*&)>i^I)yRM@%@G}k^a@YAjIXY_Ila;6<-Sd&Ga)d+wJLE?+wF_f0xYwPnTpM_
z$1PgH@=lidt4ld~Vnl+1exf~FWb&FoTQodI9`95gbvml9o?iQ_t7{yAq%{26dnn!C
zRRzJskz*F?-=A$wWJkk!(Sg*xRMHq48%uBG133bs)TOFr<o=ZdmGW<D>*{3mqk1>;
z^72Hh^{ewvF-o0^N=;3DKSbsdGdxpz19VQ=^;N&5=BNaqWaj77Kuy^>Jjh`)ULCg_
zh%=t6okTx;a&85j39&Q=JJqqqRne0~6+&VuFZC`e7(Wh4OVdUndPQK`a!WUTzZXMv
zCBz1p0mh0xKr&TQd6jDvg0N&`4x{CrI}flOml9@IO6tpQkE$|I_?&R{)n)Z_!;^m;
z($thDS|1^U`&>N?`45I>{m3&9$pLX7Q&E8a_<2xliy+9s=7!2ENHo{9a`Pk@@w$ss
z^8}TqjrxU#mN3`UG&K?Fct*?7$pnz^n>!f5yD5V`z&PuNRG@zGL(fQTLgnL5vueIV
z+b*+PD&1DZtz$xf6@<jn!^2O@C^6o9MF1!TUpvYvpL<UMhp)2xUmr&3!Q1U~<1XHX
zVIt>EyBj%mb#>+V32aPzK8Qj=o*`wZDbsj`f0<@#C?oq-JU0sq*TlKB&38SbvSBji
zq9*$`bl8&t4Z7Q%H>HGyyeW8c+S3VScg`?RFdA|5ehl6$$;lhn0fx@FwVhHm7AB^!
zjiDO(N!oaV3(~Hl;{Hw2kHahX+o36B0$9WifgSgdQROa^%X9UBjF&>!ow&HTo}5{J
zV4LT<i$`n#g2xjm8>bH>ZRr|!UtOkQV!QqE_~IJrpMyL9A;KC_@Ot(-oieAt;I7C7
zOgVf|V4#aFGQd>k1#qar2^jRDUual;O3!|@{~Qa8HwqD}rfk4jb(_}XdyCcWqDy)q
z(kHoc$paZxpE40dGo~ty9C#8Oo8H{$`N$?&JsNwF(w|ZFVF=NvJ~ycHEPYn_YH|3E
zI-&Lper34xD;KTr^ze^dwKwTRehn%5x+eV%z(;ErK7l#I#2f}A6|*4Kg8cw2r+@zH
zYwvP#A-Hq2w+z{FQ4!Zz;hImM)LM14K0Y)9CBw<9dhvg4mg9`)q)^_>jIGhjkc!-z
zrMAuMSuP=)I%2DAPBv1b1sQ<q31g%@<|=_>-Gqa-+FMqSCQ_!rnI%iMr+$w>-9do{
zj`l)6J>7!Qr87mXZ3n;Z5B@h&<%ZVXzo}QKQ|=m7Q_l{!5yLOhO@X0yPKrj#FVeGf
z7rW?5Y{zP??HhG9Sm!q}lA~o`<7<mX9*#+^bR8TX9c5nq_~r)NhZ1%}*fJ8HyT5HJ
zo8Ae(8b?+kS;qO%h>hUWwWb6+72d|2mKvEXk4<&926rokeIQtCxdLouu?C9s!7GqW
zojljF{VF~Rfdr=-D{7wuSFGQre*a=#`BZER=~sH7!`B`VI4GxkntuZ6QHBu=)abJZ
z7vRe);ioaZG@i)RauIsPL8jBHzrK0z&1=X-oqMp3^iU?`W@Mm6K;jz1e&_mtlH=56
zodMn3+Rzj`eF`Wj*ps(PU{s4+NZ~?q97|wL0vw?}yD8imWO3xpV(}{*AzRI1Fwl~Y
zm~x-QcNe(kGA5!jfckOD&@@>6jEnK*4ey_|T^KVSPvtbMSB>+4f;(Z+6ZxC(YC1bl
zIy*aGG_CZ=%Qp>10|O01^YaDfz(_@4vw6oF&1J34d?@V5G)wc#fBjnBKkx68qfK%2
z4a(2+fWowhzsy9iFN`vD*Lnv~h;$kCfzcg9J@uNb^JtLuK&HXKYHPK_QPXCjd;&%V
z5>xrp2@4Bp_j-6HxH#Xweft1}&F+2n=V^`)%s|5I?Bw&Wr&-AZP6xqBw*8q9AR%-G
z=@HjydMNlRACi#3tYlVcdFEt6L-a|M#fG*vB8#?Y;Iz+=ay}5JmNoWJuvl1FaDv0(
zPcBm(dn9U0!~N*OepM*uDImk{lRRJtGiMm5)+aaRucuG+*IhVET#3&j8vaB$rOSTN
zTDX)08)y<Q`Z}q}W@pU)<_>kw&Wj7rV=Q&Q_u2l4qNH%4TPf6)0k=ZC<s*M~7l<10
z#XKt?pHVn9Wn8Od(H>_9ckU?CYmTQsFA{g?6`oUzWw*U<YL4|jbvSB+GH@-kd=$O@
zf{USW=(pg3WD1=<1Hm*XMi-uSbA&s~<9GUh6<&Y(!2}nC)eO&2t0GCzBrbN@I`b{G
z7|2ZIBqZby<ymywtJEej9tL_NMKgtI5e&*<$_7Cb;Fl3x`lThhjjaBTE@Aq}GD+of
zuYhn$El}*X5duRnH%Zt;seMA@;>`PWG2-I<yJac7MLm=lL&Hy>T?rEH>%yfRf0ne^
zv0k=$D*sx71MM57%ozQ(9P4c7K#P|=M4Y@Ng*7!&Wg4uMlbL!nPznba+O>LWYR2qL
zfma9vExU36&j#7(-l*l)JvUW|!Tc=HCVzc_5dlI+35p<pxK!=~^T7x(Z}RJLW4l63
zU*<)hC6&+&z$19^StBo#KLpy%p!GD!b}Mr*5qi;gKzYXV=OYN*yCK^UWU?<}@Zez*
zow%Mg7cEnC?u6}haUBZ1%6v?2VE}jFxf|_a`?bxf{|<2w=;1w@okD#8$>(B7ZZ6};
z7iTHorl-5u5NJi=tFii#v{uh~*)lKa8W>=@keI3^^(1X*eh5Mn6ckh+x9NcV@FyR@
zq^jA=9>n{#lyErVavjH>F#`gAc}Z<!jnjE`LQ%=|!~OgBZ!XVh6~?OO8kJ}0Vh%JZ
zCiN~qsXuAa5_-B6^Eay-S-7FjdR<<Xx6@n1C-U+(58qK_;O|NI<W1HnH2Z1zWy2M~
zdTp&Qp*o;;Mlbproc`21!?Nb|HHrGIGr*Z~jcDxZBE>pCWG-Ba`7iVPt0La^1B*4X
zVb*YjS9Z>iU&Gi?sb1dnS%AP=<3~j!>%B><6IO_Zy^}B4PUdQXzR*`%GtLCt#psKt
zB2~nhx7>7bPD@MEWz_mbW3^we50e6jT0xTgR3JC60cwSUM29t5{X|Wk9;yV!da@_)
za$g;Y=A+Z3H=&(b(@cHFiv6472A2l(b9srT)LHi<^@dCXrqRw}^F9i4?$kB)^~AKV
z^*M-X<Lg3}XE(#fDDzi4ac|C#tgQ79g6JB4rB-_9fKw3`y{go!A>`lfhNgWSt`%gR
zchl5+3wHDEV6>5eq$ldZ>jNP&L5w?=i$WC%*HDNE7`Pb31J9@GV*&74O7huu;wgZ*
zdhnE99zSs4J+3DvBO~MQ695hwlq=vbulQeY)-PhR0gCOMCs(}p7u*VCZcd24#_LL?
zUUKaTwg?v=N&e8W4uR!zK27yB&QHLWA`E;L7k7@t>|-$DNzU)Y>K{pE@+v5OusqiX
zGUv+3uL!h<Mn@BF@B?$u^aV)tlrKMLw58^qZ%BS}ww=cc7-@uoy$!OGSUh8ig0&_;
zwQc%+UE2$J%K?&-6Ue9y!Xeq&XX?)9=GuGrmf*7MHGMSheN_6)ehy@N6gY0@9Qg2z
z5nGZs>FuY_W&?Me^H*Z)F+7I%D}-21Aj|mN#9e^wYz6eK!-t>=AJ_|rBZH5DF7aug
zC!^GzK%R*Vy|+?@bHnwnC2p3V2K<c}SB5eNH_V5IUMSq@_}#<zk^S@lcY_y5QWx>6
zN_Tbn=c*gyiMoxKKU>vN{=g|XhXIB58jG1y!PE+oon2f6EiK<7VFJ4&PA3RN+lr=3
z$#BK+#jKY5=X*v~H*W)I)NJg_4`$VIsOjAfTJN}3SKAI2N{i&a{`BH^1GCiW1ptA-
z7+^c0Dls8(Eirr2(J>SW6W<E$8&nH$P6zTalgrohz`CO%3sS1%QGvSfex{TXv2gWD
zHP`32P&4l*%M40OpQj2A=F^^#Emw;dE$)BrNU6cvYC!6LR{i4An3e#pf`UTYiG8Oo
z)mR~*ir<xb6|kPH0)RUqwXQNBRl!PSs$;R72$=HRv@#K;zU*kHYFe(@<3y<97GN7)
z05Ag_3<`4v92Ad)Zc#noBp#@2H}^lgZx|xAKWsR7Do{T=6&k@QM`Ju1he2=CYumkg
zL3sKz8vk4k=L1iNxuVl#mH4h)iye{=O356PHIy>n*-k85XG~-fV>sLQQ4T%;V+(xl
zsh{7kc`42T2#UJ>Q5z8DSC2xbwQ;MaDjLLDua*t-mYb%EnN>)G95<X?gJRj2`g|l&
z!GgrgOMxkh!@rOu=>+h=&rg3w$Hc<aEac3$mRRZ-WTIs!sp6fTGj7v4Lr7B3eFDW<
zgyU&+;Zy(8$2;-H3lgMOZv*}H^>=+Oe#($cE@(nhbnF^*z@GXC-{ORfzRp9rc{|F>
z_g^spSPn|sMqZ2M;c0%!R(U$Fn7ISffB+xH797N0qXsZLx+7RvPNAWpfPZo7hUd!8
z;Q%{YP*PHIXk>)w(G6Sx+`Zk=o-^M>=G<`(U|JGaC$gQCEn2z95{uj*ie)iqdkcjI
zWuhB~<|7?fp4EWvyYZ=20-HJBg9?QP>FZHLv+L{YJ-v@NX>$h7t-63Z-rSz4BEkpH
zo;`at-Kx*Qpt2^lw2+=p3}NRCdQGiW@6?_tR9!0$He!oVPsTVRzJS>VGI2j}{8yPS
z<qex~F}HV=I_d`vwS9k0ZM52Uvy3%Ogch2$(4a&v?{;&vn^QM03gt(W0y^>$rcbNA
zz>P=@g*0u{XRjj%kW%pg2Hd|0^x@UZ6zb|2Z8k#9)PsWeE9v?qcWOpUmBC;?h924A
z6)USEVk~skrNZYTuHthB+~!^DpFIUm@UIV7J8uO;Ov~V}ohVI<DM3b%q};^*ddh7B
z2qV%HDUi#VGgsG^qs7U2yQ+g??L)pD2Z`Mj+&?>KzxZrzq8y#S%)j&k0CfCvAj-h^
zFL*iK6g0L2nj5)w8@%gt$xqsq^49#kvgxc(`;;5kT`G0Lw)<pHAs_Z`M~eKSkiXdn
z+x`Z|2CzRhZjpx7$S>u{y|vL~Bd-7q-uUZgU)qG6!8Q<T42@&7Om**yl!>0IcuHb#
zq@#ldlJb3BocYPE`o~F0vHZlTk6>cDEBSZ36c!Y~05XBZnry9TwLn|-Jl__*&WqX3
zDz&fXm6B#V#l~mxb~vmC_T+D_ifCe-6H=ZuFydhaa=G>9VyETfm4{qA(9l<6`qc*?
z8DNB5vVOxMd>2LmXSVAJ#xR$BnFrzy(Al}Gc=9b+WsIkhIoVERTD~P0YLFOkl_IhF
zG`f{F&go}#ap|=1C!Ciy*`Anx_5N@FMV$&_$UZAp1|EOGZ{u5aj;i3AHj4<X^ngJd
z9|}Px)m<f4ez2?g0Vor2^6>Cr(W)A&0Tm0Tl~^FyEXy@lBY>82yHfue$LfBYjY!FX
zLwUsxnVUF{@ZtJMCn)K+L7a4OzsHTs;Pv9uJR@S7w&G}8L1yg6L#jO?g@0+MiNA&5
z3ie1%i0Mp6qykRSS>P1FV(gs?{w>hH{05gX1Xd&ze0P4onsAHF;m(Aj=j@!1{iY!q
z5{d#w*%tapG8qh!jQVxVR_9Aq_rKXLzzYim0yqlyt>%{vd?WSUej|Pj-v1#O^Dv2V
znMP-j{kGXbH97Kk@SXJQZyt6Cbm`&tZiu-g<a<b%4o80>NcXz(3RDnxI#kB8lcL;~
zs^GT?hQ`PB2FBZYw*9{na4W}O8~-NM2$(1UOr&V3v#lx%p@q;j73%3a`;`w{G!k+Q
zur*!nv90;pd!w3NPMkd)7DH=kQ2(0`l9LE|tuAIY>)(Anb-9<nc4eNpND!NkXwjf-
zyLQ;{IB>2l=t4ewz1)9z03<mUTGUlCRwpK}F4RzYY!ow3U>Mo9NBda0q=DJsKVM-8
zH_EZfOoQzGO8Ebgju0UThGsn#Ma;Ebf~cVGe!2ld9!H)97_&A`2r-N)K?huJAJQCy
zZ3+0_2mOoRKyb2$hNlrUx<&>v>E=Ee`2Xv^f3b{hY5E}(*y<hn9lNjn)=OYMkkJ=*
z*r<Kmcm1ZYy$ibow)#r((Qm4q9gg88WT#N`S+ygE9bN(a6;7uZ)z3_~!`pY$VjwFU
z1|`IY-?eif<Z(Dqc4X+<Da0s$YEU>5dI`HL4Ka@!0$UYIKek^8!$0TD9u4?j!e%N>
z;*ZY#XQlZ6fQXAB;AUsX0}#8_9?N`N-Xqx9@&+on+gd)qeI%(zMH_D`q&zzY@+*rV
zLsLbB>32@=_VvV1iN;Gk7$xxlhWpGWC$t1qLKjIom|BAAkZ2%id{MBU`UE26xD+Gc
z_?Xi--tn3ZR-Jg<1%cK+dt_fsP#OfG1acC~4pk3#&KU_%T>OG_@+v@)@qkn>i)Mwd
z@ksO;PSXm2srl_)|65jlJ|^lmN{emqyx^lB+hhJ<(f5O*rYI*seQxDz+Y_W%JqO9S
z&U83#_P(;+cQScvxxI8a!McB$;4^ty+S=M`>(H{W)Zs@WVP>8raa>dNFbD%RwPE&R
zQ0ze6ez(@}yNj9c&IaH-KsngGD<&1bv?%gdsC`gYiBo4mT&6EP&1U|kafitnXewxq
z112Kw6-Nu)EA`MEfGuOh2Z+8O6;;_>X?BvM%a3|b>B&oUMPPlM!lxXs<Mr=GU$@;b
z$eLMQJ^0Yy@R416TN}DZNgPa&)$a(1Pk-jHU9R<<Kn3yMr|om2zpIh9)rKZ`TPGO@
zfQp|gOh96y{XIxVqAZ#MXd%H^GeQJiFE#MzkOQyl>BWF|Fx1p^b*UIL=dblA6oJUd
zyR(7+CgR`!4Zt;^vy0d0fQ{fQSzB3Us5w$vM#wllmH>DUC*j8goTej)VK8rC|2}*#
zw;pm!7_;|!$j#Q|eqs1T=Af2v>5*Nm^F4nB&j8*K%Y`$IZEXt1{TcY$*E3But%k!>
zQ5K(!Dgoh!$oJH!oxx8>!o=pAGrsiBotX0}2`55A3=mEY+_?<34{e4ihd*OWcT^yl
z0J|v`k#@7l`N`))h|tGdrCl#VWa+Lw486XI9Dc*J6@ZSM$kskRdvcbT(~yqa@TdIm
zo-g4CQ)UaBjdng9Ld5iE;Aw*ZGkPhBnGBe!#+cz*8HmoV|7;2ojr2_jM2oqX75and
zpMY`+RSY(*sw1-l<3vyuRAT)V9_LK6L1%#Wj$u%I0~1Wzh+$9&zXV2HfE-aiB-0OS
zYKezG(Ny1el_n%x;1_|<6k33$Y}KPx_&jf}5s8ul3Y`Opt!>}P5fFhR$-KSZV?R*V
zg7x}yW+4^?u*{4Y%#{r0XA}?S=K~4Wb9nUcz38!OqY|`&?U|G;J288Bx|a>Fr;YiE
z>I~TZlqla;Lj6({z&e17LSuaIF`<A|$l}v@3ID9WCu|1%23<>hiAAY#L$6af%Szg-
zY8x400~CLlOtgSfd0bm`Tw8dFs^1yMJoz9a1A~)(o1WFw#~#(x)v><Z01zNhvWN=G
zl3OxN)R)#LgVFO>8aE$!Pq?aNo+42B<)Zem6><p2ipTZNaW&l$ON4aU1BtS6xI!aY
z_{_+pq)W%S8!CrM5;rwkUG4vjYJd1I7%m3YORGuYbcgpxjj&2dR|!FSTZqtVRMFFo
zlmQ{$9eWfoXN9o_R5$0%D7>|4Sz~xyVIrjsTakOQsN;yx#Wgk1nQ;dl#J01CW?0ih
zFDcOCpN$5!@jMsN@P9Mfv;8#@DF7GbSgBc$Q!ve0-6-FJF5AfyOHYrnZWE{YZHvw1
zm}oI=tIeRz)Vsq@%d_3b378!cyDpbH<brw}Dsey7(+jJKTjGa?5t=_}yguKJarqf9
z0pRn?hdGyr>cS;CGw+V?!@UP61Q2f<SgYb5WI!RRIBvilAo})!<C050_N2q&BSnjB
zC@u`YL^GAYnN1U5Tu9p!{=Df*VQ1^HALQZ>6t?|Q{$S#etCD+nH?%2ynyZ(@1RZ_a
zvlO%M7L2u~6c##d;xpwfv7uZvhrEFTjbk9){k+JeH1G{qk;ZvB+kN3M#h_B|CuiL7
zv|f;!3KQjsVL<7so726jPN3*E>rRL?M4JzzVIM@h9CM&5Y6#eWw)64ZowO9_mbx~b
zeP2FLB&zDoZD?`ui?~zP-P&hPK}q244u&NO<UTFE^lXo6lfva@_089K_yz+~3e)rE
z$5{*%L|$;44GaMT+O|wtHsY-dkn&z);eP|fRxU^PnPUkJ`Pv}Il%PbpNu}AB2Wrb0
zl&n^laCPsVk&V`90XU+I7_0%N71>2SF3iU}6%U3{Z%sCe%7RS@ct!SAuK0KJzHDyV
z(P(@r5n`@Ur0C$t!do|C#DTmD001x_RQ+Y^;Pzxk(<Z9JM*iDx3>NYbHj#agF<)TH
z-sC7vr;l7gy5><Vi;XqGqEDY9YEK<GT1e>xdM05M7GVn{ON;^F1*^=F-qH5C<`{J~
zHS)0f;hOH?>=i*()$SJict>MnmcCnu4!eNrCp4h_&MUmQ>B(n7#&}UWHs|A3sMMX&
zYjwSFe`&rNambhF9hZyAk|R{Ua{-7GfyBy@O!Ei{|1=55xbq{K*Wwn`RCc(|@7$4J
zQ@)fQTqur^#qlOwR+5i<EVX%)mN)A8um5ikFVUvi@=my(lxt*sy60(-Vxy|M$zgwf
zTN>QM^L`)@_#=E#iLsf=+?B%uOB+**j+>c@I<n2jg(M}bTXKoFjs`~yP9MJ%X<Tq%
z^}RL5)ynmZ&-};nZpF|=m+xZ@FlC_Bqv$O+Zh<0Mj7i(>VF>#&&;nG9I^LEONj`uy
zJ4i+XkV;-JCSTbMN6KuZ#M->Py`Rn*^Ms-ej+f{SX{SQ519LUm0A8#w_1r$T@$HJ?
zkAO9G^IkQWN{4j>gMftNnXsfM*vxT+c*L;#z-OrW>+m$E<LtX6hOBIS?c+tioBH^+
z*zcu0kq}UZhqxZrU$K=0GRmb^+{Ok(zwc6dBcu;?8Lnu5Dm%4M$m`dXpdo_k-%V*W
z-@pEk4_^vOO3qv~e+1-a^oR5NM)Lo|J9LDiavKlN{PC)O<iN_)Ab@QanBIw)(En!~
zQTNX$JaK5~jDhKb1=^1k3|Z;v>17NCkrBEhW~^o}1m0J=LPu@Qzx2tdklArr;B~77
zC}6+TRELmQcR0$ia2NMuQR3u*S3LPx%m%R37ZJlF0Z=MqcKDpy#PD3f^=rVnBFVNy
zh{5<jm}9~LGs(z$^&6D|nMGg3<Bu1<g9hE%h};T3_1f8BzOZ$GJNse22=zq-?ZNti
z@{=<hcytU&kpPs2*(*E}75weRZL>FWe(>P;gKvoV>mgVCoSq8Nhm^QDgRw4zjG=#+
z0u+D=f9|0V>lO3~SX=jq&h93iu`m1ZpA&_@zzDk)ls`udgUYWvEZVV)p6A;Mt8k#R
z_nByKXVPu6|DA6Vm{V>L0k-yNrnG|E0Ad>(o5D|@K9!xrgNl?U%_cLamEJx0TfQOA
z(1FAMBxxXX`jzv6q`5Y>tA!=*#3Jz$H~TBDZXxFr9M|)2c4j?K_satn75@DD#xO!B
zK<)vqj(sRPP9^vJ+FazKhmrs`w!oU`b<i~fv~x5Vnv~?!E~(p(e*H>eKJt)KIo}S^
z2(!a9{mq`j_9D>iCLYd8tV#z8yQlevXc1v~)u=(6Du6>pU(EKyu{{{7ON+pUkw2&*
z7l?Q0@ZnfC?;D8Vr~+i*Q?a}|Yz(HSJ^(nu004`RL8}T|`LMXSDDNo4h6oW4M}q?2
zNHv_y)&Q#Qge%E=Zf-G>(^4X45mdumN+t701Ro6Gi?UBPov|3S?3ng3E<}qIZ|gn6
z-PxrA!r%+?evLahMZcl8LlhwSb2<5*HL2C7cK`<->3qi~<X&ViVOxDYZhn~NehU@1
z_z)d}mHZX{Oc`nqlQYZoVy^gvsWTs6ZfY1i=WncbN2ZSQ2ja;5wS#8pXnZyWgTmX4
zs@N+ZO*=a~aZ?hnBH*0v!k$8QFspX-Fo!`K9zbV4zU(9(lLWFGh`VhN<@3;T91d4S
z2UTG4pIvdY<t#0$-&3oZ{~-r%2Ium?ld~DH4{da`vayv@X{e%7Jc1p8C*te7!9-!A
z8ieS@>KMY&4Rm|~cCL1Eim4@k-#|$klBY|JyH4;_hCtd|80F}_mPE7*$H>v1_=&Tw
ze&n=#pf{oKcZqtjYxAQ~)pGEv;pU=sS2vHA$Dq>ekD^$RN{UrzwdM9-NZo4#>4y?y
zCQoA!eDpkst%L(Wz>?xYytLF;0&qT;dU{q=NC!%4Hp{B05C{tkODidL%!jo{%f(De
zWLLgTOS^da@?~~TP7zQZ1?J1d&FT!Tyh%v#yi@DMsF<d|c!SK`T=9Hx7HP1%yF{l&
z^Y^Iv?ak>u$eMDHlc5k<1aKM8(zew6lXu>}dL<WNH}fdTv|rd{smGChXfgN44SyA^
zzxiPdf}Ug>a)q2U%$-j=5ZCTd%RRVu3ADR9JsodZ+39c{F-VQFB=8<?Br<)tbN<Gt
zXEA6{8}xeY0Rp5IVB-~Bnh~SAjvCZAH)i#7)@JArc8#sgZt{YvJhH<%U4Q=}75{jY
zYKtI<lxS<8ZK0ZN@u~O(h@DCHf(kYcnP@pIR@{4keB0l@=J-Q!Ob4agiqDNU=W~Rd
zEO9$=vRT<u6AwT0*#Y>C{q~01l4GV+A;(}?A<c1$0wFYH>qFMkva)O;sRW40!e>H7
zr-z#k($+{_yG9he)~DJ}PE4@}qW}IFeuzPQmDBKd;09Q(Ck0;5*QxjCu^Sa6JI@0J
zYl_SX5MZ5w3(SN-VuuGlmrf=X(P#ylmtZwtM+~3!i2!?tBtz1l!`wysc6FDniyp)F
zyZpM{J${&>1pXLSN92{jIwRjFPkLL-s!y!PpG8RG7PU_jT6POqc4{}xaPN`Rr%xmL
z+0{Y0mif{PUAEWqXQ>gPG5><E0Xp$IxM@AR@E9Pv=TGQ#WSH1j`61}MnZbE`Onh3O
zRBh4<>)u#s<R27A4EZfX<R@ZTqY8I^uBzhyngK!b(c&Vw#X#xyN0!f+l#dk`=b9O{
zB%UQABC@z8pMaR9XW|Uy-hwfm@=QFgUh3hl`N3m&wn48suWm2^4I$!}JRca0eKnf{
z{+rylD_^$iy$59q3C+`Cm>lfvLiYCd1D6ywW)Px&b3t`kjemt9kv|wb-(lW7E-lLz
zA@U%>VLpOX{SpaRzb~8m+|VWE>ZR#=A1h?;eI#PO@;ICnY;TUdHTNJ$z~n4}S{MA&
zr;BgEo8DM6In2F2oKo_nMaR&vDoj~fIZ#Ir5tjz10uGCpOS@v$ol8oN>y3P;x{ZsP
zZQM>9gn+h1KBiZ%YDPfajK8k&k3T$tia>)5C)Z}H-JjAkSGYQ?j(59*f@eBC2A#lx
z0Rdp)#>Zb#SV7SJ0rsz<zifCYe=fYbG&(ZK7p<wTjtF#z>S;+cP`u5c<32DjAYO6x
z>Q%qALw~%h?{u0d)D+1&Cs&RRKfPnd9dvUQWW5255hc*1vqD@z$2}q<LcF4}5uIIw
z1vI_bw;&>~g_vP^<)S+ge)sxB+*z{$H*snJ3*zim>|UuB7r5jOLJHzpSFa-Ujd%s~
zU_!>}iLB9u<7~I@Q2S%s4|Oc{66OaCp9!5VAvSuSWj@LleeLSi+igU@*5|)IdGd0w
zPrS16@%+s-P-SQIsJFLU^^&cvh}{5Nusd57*O@fKP6hHC3&AC+&0RQ*jB$R#;o;$k
zyj7GtwxuW28RBynf|FUZ=9}7YEsR-~hCg@UaF~zPdIf^kdi>piAX#TM6e`kt4}+f+
zDyd7Uqot)S9~s^MRfEK082S?79Ct=0X#Ri0Zu+sR2j~rOoYPRhp`k&@-2AHC)vLu(
z(yz}4yGBJtJ)D`DIWFK}M`F=JG2Nn`U#0=zHYM@Bhli0+hZh0L=TnPbf*+ulptCx4
z@8d1&SACy97p8k!k<*;(%`{Qd(NVX|Zpp{UIQh1FS>I#4v`Z&az_Hu#(VYv<cUCJo
zyWPNC07@0x@p)~1{SnfbSXBjuw$Y3Wh>-DK>>!h6YOD?wHQQ~jT4G-cOYdqUS?pXa
z@ES|a=A@;a9!<@5A4fz<RpwEe3W*oqC;Q)|R2}#J3xp8M^enZJke%aTB7@~w0!0gn
zSUV>B$D~457Y_~RZDo3o2O(_?U{~9k)9r8ui77~(`8+MH_BkTteh}Y&>cTp&eUm}q
zju6%<{A;KvMChaZ8Xyb}4<r=b#2zRylw7Sc8U)Rdx0e)Z$5G!}1l?Sn#D3}g&W~iP
zI>?5wWn<DltFvj9_4N+_cDDfa=+K6}$;#?M0pyDwkUN$&3d)k1P4S|Q3-Jh7=jj64
zz0iKml)uynW3G1{FXYqP+_zf1PsV!=xUn7NU7{Ox3k!3OCY`_AUt-(sh~{mrzsS|q
zm1}j3%?vM&r4V#EA-s4hdEden+Fv#eK(8?&VS%>2^d4j%XWl0m=;>WvJ3;cx&F_Wd
z!$3%<tY`8aDA{MGqZjFCx#b|3`u1%f=Yd*87{CQMP<gzN*TL3^oE+X~4BWcpmtBg9
zP>7t1&z(m`zy;rd?CZ~}_d&fdN9oa7!Ygp+wbsXoh}a3zOR>IfZ#57>dQ8aNR{I3x
z8!tkL7${g{J9d_-qN3teSQwSq<;&e048u*E>sFU8nE*f+|KjA((8FiXnzRiKC)V^^
zCsIz=J8H$Jn-3?=*$zjhrExUHDUyLSgnh~7$$l3E*fgj|ptcRBc8_NV^Lb)oV_)St
z=y{fbK?^d#mb!Y@dZv%m%!vzBmlJ@l_6lo5A_9*T4VM=|CT=yQPG46hgaShd5$iZc
zFXh|v5%;B)w6rvdY1sw7<X{d1UC>otRYKgvdG3Jg<Huk^&Iyw$ZmnW4XzPUI#vob6
zd3kkp^=qn#ct0xDRp4Ckc69W(kTp?`tahc-=<4d)xSi1?e=1h2C19&_ElWZ|;)ean
zo#ff?u#}e^IP@7~vs`@b7c4jbCS^U*=EbHpaf0o(c(s08OUnxco5mSa1*Fm<p79fB
zpS_3sv1w`48S3g*y+sHj0QLkD14ZLtbd$WCoUfKJ6FvQhku#~XN=gBjazjHY>URd~
zY_A!%Kgml3z^lYg2k*dxVxQOgo;-Q-#zBS$ot<iapbrtHC+m>I6kKH(;LmB){id^C
znAo|fdSOsoTbl$+#PragU;H(X<mNBE0KUm}oTArsFeGNvuAKGc<cXo7vj1C$CYj!x
z3mlia<1>lvgxbL1A(>e>KnOZ|hC=FzXUMIKU<%X5_1u-g6Rxzf)LLzq3_FaA&LwGn
zct%Bb0kq87i!#~$Z$eyvqT~}bWs|yb;{*o5Sg^kuw;QWolM1&RH68*~0Zy4aOxm#U
z;osZS-tTTfG#)IsJeS805kB<+B<^E0<<cZKFFu4RDk_4Za~f9^WqRcq?uR16t$l<~
zBc?xaj9C;^aZ5g`t2?VKd9`ur-bZPf-`({$(*71~R7(V=;*o0~R~%44(;U{>*CL*X
zlslp`<M?V5t)N~BgGG=Q5usoBJ1GBY#bz`|3A~6WesS?&!~lX$8j@hO&GJ^|oSd9m
zPk`<eM92-ILGiX8!t2_@oafGI%EY`Od5TE=(brC2&_PT@#Ciin8Z;5;5mPNMpffB2
z$KNu^VO`Jf4X_~*|M&#E{t|7V>@==QOM4^iSbV9tm#O+UdLQQxi&=JLB%L+v$Pf*e
zxg?g~trQG85-l}5uBR^*5?<4cqi1HWMwkZ;4A%SptFif6G<h>miP^keU1iSN%nd<d
zjApGY<*QduG&VL$-{7XZ=x#F@UG}!V7>kasPEt6mT~kvtP*TU&*H_5I!~_o?e>`zW
z5cE}ko0=*NkX6m~O-&mMgJdOjX28m+{l|SytzVCb7y$ZWY{aNNjj_u{&rH8<?)Dqi
zO@*B466cStX+|KGl<SO$dReL)P=vmtYpAyIhICsy&#Sbw_|fv0$8QJT=|L4hh$P}d
z#lv(F#9KMRzt&@h&R5*|MAaloHK7wpPxjAM`&XG)mI=JY;Da;k3+<d9d<!o~gy`Pj
zR8}%NJ@wK@fC9W*a0PfPi9W{n?Z;m|=I3vbKb8-m68zIz#`_+W0G+NEB%4I?4?{sB
z&`*q0q!yi0+<v{OP*jKjA*W9Wti%@K0(f4KSa@%TnE-E1f6*;vd;9f7E*HVwmH!@f
zGH_#-c&6ak0bWJg%h=&C!Hr4SAq1?Qg&k61?w{{kw1URDw<bDT4=V7G5HvO$oy00N
z-iKNEm3GK+KsaUTxmvL+t|`e`$DWbV(OE(z4FpgnIm-iR@e~8FG=-$zB{%~yf!@ey
zI@V(|E<||H0L4R5ntxD)WLiX8&bPv(VeLU|LPE8qnlxep02sbUrXAAS;V__iKKiGt
z^3HzTT*xwgcW>DQG#7mj;_t+tOWfU))b(CPL{QVw)4$HiVW6X<d-dkcn-t})<r_pG
zj4CZ71Cm9>xf9KxEi@$ET3Q<Bmsc&+wX&DTr|<CaHItZ|n@?BJ=evW3ann{)?_6Gm
zho_bXK}ltkR?T1!gLb0x!3IVGm`MQV6%`lvxYD1~l0P+Yysaf!8177Z_jlWTe!KZ^
zmJ+!)6jbYJ!veug6%2N}m3qQb!n~<WpjNEQ=y>`rGv|^JD!*l&$eCTefy>#oG|#p4
z$!Qu4RC#_b9wF#Y=M-42F~3r15TkNP=&Gbd<%^85<jd%<xCGCq=Pm}}0~po#(n_<@
z!<d+gIjMm`Fj(}&VuzV(L;@o0P-N$9U^H82^mL)KLlFQF2=I2%U%|GX$RzUsWo0+P
zU@5kg#MD$q{-t*U?v1g*Rc<<n5FGFxD6d0mxCk2ZJ(Q#4nR<ZsIpr{Vlx4vIjNYVD
zQrE@2MLxebY5LKgz(RIU)l)4;)Qh1OC!~fJ4N=l(K@Yz-Wsr3|La3ls_y7$I>9#Df
z?V1(wYf!N_R|aiVr-7BU^6>>iGCjv3jAY;NwZW>uKz#0}Vw&?9NwQ$-+SRa|8*A2n
zb1HVz;Z>(F&%AfWBxY4dND@RX1Gv{-k;H*)XRJ|(>g3LyJ2wqxKJbAc7^|sX6E7i{
z>i!Ti2K~+=q}-h)0Ms{^)f-qa*GS3zANJlmuIB##AHPIa5=zmgR4OG+Z9+ndkhV%H
zY47czqBM*UE$v;>-ZZ62LtCZ2_c)E;<K=o^<9)pym+JM$_xAgo>o4zj=bYE;`FK9|
zxIdoyMho-?IvMwc29!Ql%AS`uhS~a4P=u^Y+m1%~xMw@ThsdpmsgqKLBxV4eDm1jI
zoZMmJ@!2ayux#P&F8hTs=8ghV8BeIN#2BNKLqdRA#?9`+#=2pElDo@Y2?J>3Pd`*R
z9XfdMfs9?pds@!r*<zAHSaVyaxnkhg9~#cniQ-a^pU9XDBiTZLihRvTw-2_|RdyII
zPfuKsPdzX|{anrAGPU)$j;lEfvuK_45+*2wiDP7e=pn6$;CZ2;Z?q50%uXoVjfDqi
zXJ;4tqH%-tUMGN_kFE3wpH6TUQ@7~xAw0P}rDi!PsEx|>HKR1RxFIc9F<qJ0KRmL(
zk#w_mH(Ko+QoFzsOQl}EoqisMm2Z^2Xw|y`TV**r=X%t7%`529pwQ{Zpr55OVhPI4
z$G=J0->_)Xx12tzjY`KD<4J@z&6n2o9EEzG=-W)_L&f@XqhePcBa<E%4We#O#6(5)
zb7#W-^{~pLq(5Zv9~tJKxS#X~L5(tYg>4n6oEoT~@oDKUPp)pW3*daY!XNtaGYA>s
z&DW`%x+rH2yAQF_i|IY8y_T)0z#Xw5>SODo^H$+psKZ(fCD+fFC&Itlte>DknZPRW
z_gN;XSVtGYmhe~C)wLQvo0?L>`thUE(zO*qS2~{inxJw7Ag&wKou{Uzgp@Vcp~2`Z
z*AddpS!@+fg$4@%qIGOge-?S^CR)92PVBN5OFn#vO(4vf@N3e{cXk&#XR<IWVv$xv
zfcX3@W*Aq4Em31N#3yjgRz5t8nOttU2E|iBL98@GS)MUC+=nl2NLjWG5H%e;eKgG!
zXP<ud*4dmidGfmqNITU|(^?7JDsXEHOT7d3R)s+;F<q6Q8xpYw$`W4mdk`8h<d`v&
z8t@q>VwZESxM3UJwq-e?eL($FVd66*E}1iuVXm-p`Y-qPZyb(!#0H4Jk8Eeo`;(6b
z<WAcj*1<p*a>004coB47Y?>UN9kbo0<98AjMwCSuvF6pmsw@N-8UW+308ISn^%b%h
z=XG)lXr#dj5jPH_dhT^-G7xe$GZ(7?kVRs{nIo8W;ryV_UI?1^->z3e<5?TFUI+}?
z@r2UgsK|V8Wd}5?^b6J%JRsHh6e0yls4o$bkq`R&qbkA$Pq|0khdN_L<pq@?s4vG^
zSX*0H90l+n&Bghtu?{!R!k4_2MKTWKM$NV(BpRJX(swJg`aT6Sc%LDj7vV(3vp$ZP
zrA?{EQ?>PlP_c6yx)frjf7k4d^W+nbI@t>sTBj5g6)7P#^=OHmgTpl_DCpQo07pSp
zV<R~%E>Nj_c1THMqPsC7I@-;{<CtOn0MW;1G@sGx(_jQKH!)IT_D|CYC*lbqI>5uj
zV<B?%w0VgHyHd*al9?ogq1T+eDM=Q;5Vzi=zzV6Vp7Rw@FQ8cj;D86v)9IOJDzq}N
z&>z3hHnHzV3E#bYxA8kd$47ThJCOz@B{43w(=Ejr?RtzxJH(DJgzylHb%uxI9R@Y&
z9wa7ydYARMHyQRvI+XbdF;{DQ3IF)*R>n$JNQ<#0qs2rIteWY7U2tvX7q)6*p$pc%
z(th~U`_re;Nl0x*U{whUFb2l+ec{{8bT#US(t9S;W?DxI(~LWJ*=~G*Lw&92kQf9U
z#f+lCeKGXx^Ajge&KW%PYKtxQYzuxXkUxWzwCAKxn4!RB^oI-JN!z$|9Gfm*xl&Xu
zN!QT!xS)c@yr#oyVa!DnRtQNC?qx&+-Dkr*i1lkupBA1a$hIZCvZ$y`0c7z9dU5H(
zFb3tQQ5Tav{^VZ7+C>MJ$ng_B9qeAzZ<D^(kf@pA&KR!*{rCV{WM&7e28vvFGoO%p
zpceeCCFmgvDW_|P36``LuO&HO8-i_W8j#o@t5bS~ZHWyE5-#$fq2lG$%n)T`lRRl{
zJ=Y|o@YH*;U5er~3oEPZCDz&z|B-VR5SPGC2^g_C1TB4aQ$7ZUO?6&-Wk%HG<VfMR
zhwX|Ao*sPC`SmVE5$u^G!;XmMvMmW==TI3r&PSXg;bx<Zl6Q+U%L$U-04lJ+Ti)R)
ztg}l%8s^q>q*ZA$ta%&Cca@mu1qtcHA-?OI2CF1UhLgqfvu7MQr`sHt^7!7kqS2bN
zqr^x2LdE_qPj`OrL>;oEgyAQ7-*%d90cfyRU>=en$ajKd(>sx%3g2{XPcC#wBd!hu
zdtNTK2+5IR4BI$3G}ouh*KA4<e+u({Cm2;3OIQ}zs_+QLy57zUqCt+^;_b*=hi4*t
zoUVKe+AsfzFbI{r;?~pRmv0NJ6!{*t)joMX^dd@4G~$6q-!nWa;A{!8nMg@Vq0|br
zgM}I_-MASA;9q`lRaR7dSzC|Jgx6Ndz!0O0dlh`B9cI;yd;HoesBJ2I8&GQWrI)?>
zJ*y`QR)rM2n*B1>PN+;0Y}m*k%k!eJ4rfTQGX6Ui1c?NOu{u9wd)dmq`XdS2^AvvW
zZb#w^bbjpt(C}e_lI0B)UJdM*EWv@hiu%UKh@h{t_?2PKUcmFS!8YuHiUbU-eV%*R
zH%~3*;0cY;%EtmEOx8{g2;8bbE?APN_FfZ33LAOr#awMi7LN%}Rxymt4;Y&5*48l`
z`O>}}kgME~QBoob+NwF~J(4uZ0??qCmOq2eNEZXrQMaLG0XBo}Zo{@ZT9Rv}i8;)F
z4v>V!=I6C?4O-?;i!Cp<&6s7YDJzo!HjZs4;up)EKgtz$-Y-KAYx4k&E8QlK32L`V
zXg7;v_Mg>?xoUh9Hptj^Bz{~HrM2m;_T#9idUaMw$#!AXg`^0UERg~JpF8P!fpqR7
zhU`PUINPUJ;?|xD0Ujl^8v^h>ZsYz)$6YS887we>;Arw#7v=z(?Ps#zSdD!W7%$N>
zt`wQ+VvM6lg%iAx;5V<SA<Noyp)X)bf!Zif-58l#@$gt+ocns0q!`=gc#&OUd}+4a
zP_<NMR}dO9jv;`;;OpNw)UjkR_`?U@hy7Trp}mly)83Ry=9Z;)61s<kMBCkATMQor
zq?s}UsJPnd$Vl5dl`$2|mX^!D?T{>E35;<FC{58PHm*Lo%-o7?Jun7a&dHS>cA<j)
zkSPZo-V!NvKU+RFWB+;KvB`+&rX%-LO>!rzB^_xcsf{nis4kif-*rPJ^dCpkf&O1p
zNw5!Ol&J&eYnI)or#*&Th*Gjamab-5=_1gXgs*}NexxRUoQuJ&fb(P#|L*Bap*Qu?
z{ENA1-7g0$JIWVwE<&3Byg>Um*s&ut{_1`&8hlE>2{By1rN{38pbi~nt9EAWOODww
zXoQA++@_t{Vp!{UJ*>yj(8vjF@Tfu8R*o!eQ(F!tk`LnHPaS;EG3S{4`ZX6^;@z?F
zk9rdU54WL0MT1=|4bs@g6go^%QIWvhH|`F1e!j`KH%H|zU-srQ?zjgH!0(#^Ix`zd
znL@<cteKGhoSNENb5W%u0RRF#)@)i?{6r?sqbI;|A2h~n^v#2soaWrPTt~RrqA^rN
zZ1P4>Nn>sJo+zg^=+79)p_CCG`^F&&t-S6h-C9XuwG)qrkby2>^<9HDb(L02*e^W>
z`xT1;H|aVoe6sAsY;qu@5sIiE9Z-vZ;LRpR%HM2M&QY9-PDb|oNR1r>BzfOJ)kKKK
zJixx&f?V@zox;=h7XtIW48c)RH^XQ82MQ<s9T$ov%|s|vt&YX0V9zw$O+X^a&|$(v
zExyp5zm*ICTyL<eW<^x;GNEz7;=2f4AHOhrY?=cCqWL+Df0Ow=K)Wo`7JYj})uzv}
zvq4+Fn73YeC)*X@1SMDQ$!A|wZOb1b7@4Y8?LOb@>CnGmqeM<av$H2)`3*Fy^_#cS
z;}*<cw~42t!0H^$N^fjzOj{U_-bigcKdEb6xLeItUe%gIba}24s*|?Fvn?%e4nvoF
zdXIyE!{PvQN63>alaw6|+5w&bH8ZZ-vkeua5Q2<b<iE-sFV*XeP30RotYK_?VWGUu
zP8f_t;>1LJa7KoW%g8sAcQui6fP$Js12bBe&FXITjy)08hFY!X+V^-OH%=dkD4X7H
zK|#R|XR0M@muhkjVR~A}h4!=#*UN<RFJ9Y@JI+0)djPG2QN#Hcl$7cfvfA}TidMQu
z_s7Za?_>ymWjp>MTmw@7i>HM8LUxZ;Nr>HdA|e^IRM{sT=EkFJ-4#W9P{q8~9J*Ct
z@aWg?{3mMuG~#z3%}7R^!Q4lpOJlhYn+&j?5l|I_$DVf>e?sT^wo;UVEt$Xd%MSiF
zyR&sKP^pgp&W9Q4fbN-Fc^cW-s%$^i1sy}p1GC|din}$7>#qBaJyTi|iH=7l;(W~N
z)$ID)mR)TyCo8-#8k~K@4qzRN%lVg0L4B$AJUJo&lPy=1`zI9oM8`{HI;ZN@Dk(ps
z668rCs3)6Jr#tDE8DGA9Y3pv*VlICQ8kpiurKp5Qvc%k^o-p0`WMDXQUubw%6{Y5x
z$4CugyTLGElJ@bI2Az~F=qPlv@jKgb4a|wJMdJsB$To{6Ka@i6@3nFhBRz;^26;=v
zMa_-lJEdr$JL~h?PmW0cRB|$}-Z^yUa*(Pyb@P`Z_g6M26cVmEu&X)Pd5(pUfDUIt
zI)*)}Yig(~MX}0gQ4(h6fReF*CeJ3nBn|vReWFtTKGY@<c)2T8ki{27i%nsn06Z!N
z9tAstKOFnEaPI|7Jze=4?yu;FQqDo@JCF%~3fQC=mS9Qw;%^_Gye)E;(*{$ZP;5Ql
zr=5sS{)k=|QLA5z_zdI9U5~)}+L$Gy0c2Q#Mi;CiN=d3e9#Z5H`@wZQ&XRb9fEbvt
zR1lEQEikSzQs}hS{Pw+^B1Ou4z6?3aC`VNSj338LK=+OoI%z!dPe|_SoxAkQr_gL?
zE4Jz2H}--Q{U(j_m)dVTYU}E5p!^(PfqiVH;5ojX&yR!^Bc7*BP0!Xndim;AFf#j%
zlB`O7m(?rK0TrLSbZUE0H*<mg%wnMA*Da(>Cs@7vxeWeR241~~!+0`~@!$idqm>Ux
znQw%ODHHMV@bMMpE_^(spe=35B;#MBS@zvxGD-yob2o$;Y)LgA7{KYoFq{oco=9tQ
zbo5#hefNE?cbg1mXt_09rWf1MZe4l)N6QkC-`~huwTSJ&<Y?1i+V>%%JZ#)uCh|52
zg1+cp805Uf>dosvS&_Tg<XC{{FKP*B7G*@5MZLr^(;K?kwQxl=*Jiea5FMFax;8V{
zh-Ar}7^@aXvV(-3%!EvES*dgiz^xuGKgrIB2|9k}Mq0N?%>_3hH?D<tv-q404u>%}
z=B2}jw<$skEaF6G`(H18`XSqc3YF4>6q1zQ#Lo+6RRDHpC+=^7py2Ji)X2hO+p>i4
z@`777ttPY$q53>Ye)GXM#n(?Fcm}X*{*WGfq#vkGn&>Mk)!OrY1iKwj@$kP7I93nQ
zST-ldoURmG)@ZislvdD`8U&nQ(et8n4|Vj1GQSjvV#}mEs=uKzKfK`x=+f&f7KWPD
zK{F?`x*}repR3NNe@W7ntGoJG&@g-I3bpOvrjD-XYJJ>8(2iTddlu`<k50Q3Y&?<h
zFo^tp=9-nQQOQu>!;N#vtnTVXL4d;9b@us^qs8!s(*BIW^|Rb%2_^dn=%cq%azs3!
z3Vd^b#Z~v^g=aobXKxN)Z+yk37#V+d*p`Xo#hFucVH<BM+z+Csyykj5H1R|K?GnL>
zigSGfhq2jZT^s?n70b=Gj@Xe2BLRQYE}{M1WplRflA;~=^EESvSU1Ym2pnTxy~_A2
zS$A4u-8$lf?4@%(0ZO{5#+|et&{-XH(N>UOc<virkW7HXSfgHiy-gX6g{RbUZT#z-
zKXI-9<xdzhiXBy6@3#_D^3ZP=F*6FKiO(AEM^>h@C5J2JyU;o85u1H?)VPdG&C{Dr
zzSbrSuz`y1@66Yu(MHGhw?qydSLH3QdZo)lY`$8iBgl=zjhKq>QbX<`s`?lOQqAHe
zc8i8k%Vy(Z5onoD{k)CN;cC*YhD3wA?x<wTblK?x_Z1fQwdEQ@gJ}EZ;rWRGH7e{-
zy%;n&+juWV8rV&g=N6MVTB%R=zL8XRn5zV9=s941?``GMvFsu$R3z)uWpVL*zKF8j
z^2!&7;!L|2MCNDcs3FVb02|huYCZHKJW|{IsQ|n~esX}H#b%~e%pk4(+0~R=*4e4p
zZhlYCeG(gMZT+{PlJ}w^+`x0|ZWOW)oa<(dWmJpxM0ZEpTz8YEr^tN4yG<k>01w3p
z?K@F40XmXU;({i4A%Hma*mHEeptnZ1=ZpP}eJ3o5bJP8-Ds9%FQyZm)fFNzD<bQ+)
z<Al5tKbAqmB41&O9pWz>Y%nZ}j<1LqE9K0dJo8ipw$7<e4rgmm_DvLif!UZdZ8PrK
z(~aDXpr<+)Iy#^U*v^_=C1X&m^3!a6ZmVbR(p2HNcA={`Z-q?A#P~-}4-S<K#v`@V
zxa{Cc+y$>^1O${R7ZhId>XGLz4#jugUY|n-_zRI!WK?{&9{O;%u{er~35exhfX-Xi
zl(w9~hMa}$z`(C>#hrJvU4gvYhn{er-k$84VH)j##k(6@@;gwu<XU*Ii-(rd7c(q|
zki^=+9^u&p7NTpXza?9n8pkM3U6^}PvJiVT{Ot1z$Ovj)Am7oB%jH;+K}(bhyuhcz
zL^|>**Rd^aar)JfvEJPk%{lg&FL?Ss^ftVebh1!e(J$iousGl5%{r&=vN|kdnTur{
zZ4|zBg}zhYC2&0Y%AwA_aOP7Zt+q^uo5)bP<|QEKr;t;y>q7n9i^HOdNjm;a%u!8T
zim9!l@B&+mzhXAk=-;`3%H6F80S&oO?IAQA&1K#npb_qu5nde`C02TD6U`28?Hua6
zn@D0Dyl8*<WI}SL+<#yJ{4hcEyN&ZAL{1rVA3ONgGHbs7>7K)1zw!t*EYmS_a&`a!
z*@J^}{%*7)7E8<918`i*isWDX*;cs0{P*PK<VB{HA~fJ`v{gr4{+J}F9<Q2NBqt;N
zP%l24m6G4U_lmYDIy;bL;fR!H_&%6_d{0v3k<Hj$>BwCnw(R@GAoGS)+>7Q(8p}cQ
znO0ldGM@=W#2T5n`JjRGu;Iu#adC;46fqJm+l_~k<qM%YhBTVFF-cQ~PZnTD+L#S;
z<}#oU;tJq5F-smhcI?bd>Rwcwk4@{A8%1fNWL>z;hMSKXq<1;Pwi7Nc_4lM>#bTmV
zj&mQM&VGAhZvH}LW}oycz6W4z3aceOSsX!kD9(I1g)@j*0k>n=cY`;g?ljbzt>kOj
z+R#-AUALtYluNV49{^6v6go>@P31_tk(URX_MEQ<-7zlioKHeW9;=ufH6zG`ibr&d
z9Nytwt$tmM0)Tu~jfW3X4%EkcLYLe#(T}A8Bx3B?g^>{e>a33ZyggKXo8R(8Yvh8;
z9*K?1>x|H7m^<xr`?mAb3;YYGfX?m)Eg9Zo#JR6%P!Xo#%{--k(HRHPU>QhQbAj{r
zAl(NQo3Qzt7o3}#H6<DiR$52!U<>GtvU2ucR)9ZIdgAmh8#eHKlYOJ&y&~m>>%*rJ
zIoB}QQfV{Sr91Xft(8tL?3_EW7*p1H!E@{|CRI%5xW6t=<<<6LV^jut0j8gjI3>}E
zZ3`I%MR9Mnb^-WI4t4b@`mX>Dz`|X{qHkyBaK>41a7_o(G-qH5nyFcTl0=ez)AZXn
zeUq>Z!QA-$mNxeXqyg+AhN)@o&UzhsYC?Q|9=+@<5OJK0cP$$3>&wlj$()lV44#)a
zYqPupH0~&?F7LiBnagZEx}OBL5)lP!7Qg>Av%UJY1QuEn+Q)_r9p6T^AJk)GVY%P;
zS&gl2Bzs2mey8WM#Wt$xuC%UaR2r!*Hlc&me5gbUkvnvrx=Ig@8DidOI$kk!g)pV)
z4k3|dcNstQx%X%+%ocZ~_bgo+F;AJkmw4M5HcE!RPkh2GMrS`i=~n1QA+Iyzgo`Qu
zB~<$_%3aR@Ss}YT8u~n}P-vv=Y_&mTLwAi_E(NHJ$b<&wfEYI_!>r`vuQ<`T8~V?P
zvk`H}Q+WKUtHqfXJ|`!qF?U01HQjunkt}|Q0$7vvR`U8d6}1nmtXJfw$|O^n7o$yr
zbm>x0ir#>sEsV>Yeh8Yd2$oUppb;WJg`FM>QpvP<^Z9(sK-Lurl7;wX!2`5b*X*41
zMdu?}N{>@=q+L6QmJ=C<xFE;6$T^hvxRD=<`+R==CfZ#v1}hI!$d!WbZFP6&XGWB0
z_m92XM8rzn1+x$CT$&XgcL*E{So3DlfbQ9Qu!@C2#!ubFVR0(Awyz-H+5YI#LXXs5
z(h6A%$cSXkHrD4T_Bm)_$T+h`>(2~PqGB$Xz8}1{)k^b9Z04<0mA7dV6-uS+Np8~k
z-dG>CRfk2W!_a}7tNeVw04jX;yFIDZs~d>Vy)ZBfwP+Tb>A?#FX=c<_VC{A4Vst};
zrG00wgA$X!S^qPQ{mTK`5oyrHTUq(EE86s$Lsc?UjO!G_>jjwngow6o)rilwkr(U;
zI_aj_U-O&|EiPw*gahGjr8gQ{?wQai!TOTlP_^@g*~sAM9=FT1V@?!Hidvbc(6C%y
zQb>#^3w`22H!Tmclz0NDYi#4vxxU`o0iw&aQ-+SOE~EJ<$}~j)Df93YQ&k$`kZ|7S
zI^2?Zxhdcv4UCb;jYq&v+sH%}hX2xKIF-TAS*nB)oDX8<*xG3M`n3GlF54sDh{*{?
z6Y0DVngIlR2m3JP{G%we-ou6vAYs(?1{-%e`%9NPHPn{I11YiGp>H82a+%f~@*-9P
zv8mC6pHb<hI&u=&alJgNSguV6pT1(T19nRHD{07$x!UOZOPSS)S|6i_KBJOPJ#s8?
ze)9G?0sD=<ZxRv`8s&yll;TS&E8|ndo(NuvRlGc8><0NA)SmYJT#-<7L&7&!%Y*EL
zvRbxGd~sLbBsOZst7eBm{PFWE|1}dsPYS+TNZ^@TLN2r3DpA7B@EdMNO<Q9uf)OFW
znyr%G{_EkGb2}hPxWl@eocv1qSw2FZvKL%$ONE~C#-V=>8lfvRRkE_OUUyok9Nds{
zFEIE`{+lpZmyFIDN^vG6OLEM6`_T|b?@acq2??ds=kylZmhA;y-OfgdSg5PJpiDUT
zJzQC(Dy@#ocUhabV~rPPCG~_&T(%GC4c4PFhWkn7yH-!CSIrK#or&|dnE!g0(7W+S
zta{(4P>3bp^3p1aOSlUQqM#w@*CPxUaaxl0Xx69mMC9x-HP3X3mnX#dk>c7-VGEb-
zG&j*amn*B;WHkfp(Buc^0f7CnWKGsSlmFg{s57N$IBF`*$%%mS=E)<!Yyp4bDtuA|
zGP;&>+s&$1zwN2;w#Zyw+QgQubtb=%!qYv`>%mLaSS3R`B>vXDFN%u6nnn!Pr67P_
zFC?UR7@5?RW=vAtUH%FR61Mr8o{LO?3*MN2U?U0bi(z?VRGujt4c@fkgOt|W7lO~b
z`dRd=s=sq0qA299JQr2<Y*?+ka#`{r?c!ZQ8q?mYedd(fJJ2v-r?rvNN6h&rOw!Qv
z%@LK<Ih5L_5_w343btV{3M`DqpSV!nFGyqjB|&0<r-B|8>Q*4sRrY&we%@j*y=Qp0
zg?)ddSdlb;i{k4;G`U&W;Vek#Rawon%$W<SrW&YdLikjx8*>a5#q5WK5ov;Rb8|3w
zD}SYN%0+C!SQtAv;obN-^!4Da^&7Xj&$MC#D<mle%})CjQ=(!~KPh}KtW?P79NF)@
zZSPg}p7K|2n%_!DJ6kM!F6*?_sfo^gdjkEu6P*e99<GIM&uaz*n0*f?s&*3NoGt%}
z%>2)v`dH}$)u}Fb&kw|VWM#R&idRl&g@%Wn&`%%kO4IaJ3CbtJOS7bJMsn;s0W`q#
zQ~Z?8)QF?3ZbW{SO*t1T^CKY$kLzFZ5w)Lo2QH6F!ip`=scxeY+@FJOk<(~M)bgIE
z%6QHGqTMprHc|M6IuJ+`bPH*t$>5i>yIyLqZ8Q0Ql+kHP;Q6KmR(TcklSu~I_TJ1^
zUom5VWn>RCjV{P4;>#WlitbPrxv%Z-Lk5T#7G?t}dUjLMZTE8*^)23iq%%34yN7SE
z>R!*Q@7&SHQk6C-8{jzpY0Rqv=l-UUbAf}7{Id&&o3sM5kqe?1B1~Dpwl14I^E0<#
z!y#I-@|cBB7^Ni#+4}UNoh5eba-ZqvQ(6pu;exDiS|ce8V+%u>y{7kzak0Qy!G*Ej
zVA76YPthm~JA1@~+TB?!M#&)QzlTQ9%#Fcg_GYtnPxc$5_Mmxoxvnp7G&%q9g|A16
zE9{YYNA9_NucYRtAg<MUUjzxXmh;Pd9K~3$!{3u`4F^Z(az{RjVzUK=1_rO-kkY_7
z0<GNLc1ttsb1CXCSELQ@V8MPKNhT(&5T2L>STRYBqK7`U{u;3Tatu0fo{70rQ4|Co
zTdz^#b?keVvz^%K!qqRYCl9hOH-iT{Zhj?Bc`sD39vnC;%I>I^Wi_T9&;v^HoZr03
zt;VGEMiJ5>f#EdGjw%TkM!^<g9Q0jL{Puel%~)Z4VxseTZWhCKfv;YIgiN=Z(+>LE
zPs>1(NYW7ELnOTg!WG?bP?drd%sPu&J_p)f^+hA=a*GIas!2o6_6iv+<(PSJmUB_6
zuxLYWtWv7Ot9#stv*|FCSsV=T!dv{Cc~CiY4(RJoA$&+pX&dEaD9=&pMUM-zUUZ&f
zOGZepZjp%+oyx-rF!V3UqAmoV>e{zS#CF*Cd+BO?e7`~FP|+S?t7|Pxdr-;q9+8D7
zQSOdSNI1+Svm^m!T1iU3sQKU|-OqylcH{goh13?mlET$bxaf~!m5A;1ecjJXyVREF
zjdd*&U5lZGJX>Ral0viKJ_D=8EGnZ^iQf-Ko_x}c95#cH6!O+D@QhD<jz7^hUsqz)
z7b;fFkvVxzLPQdkRHTv`{o>Npa$T-wQJmE~LQReS8nMEilGM-0EqhyRJ@<$$Em_UB
zV}l_`)iq{v`LF0@wcbl467+2CVw?uZvh|&j;Y>x7eR3TyG(B9ivKYzF8uc$>XC5&7
zTQNeD2rXF9BkwMsWBAGS{PLfC*dSpdaWi#!30qk3z_V~5Ud9ck_qYVj8($xjX$_I;
z+yf2+wE2#l$=|2?5PB5Cc&Ig~LL_*44;?}US)iXuRyI}rRH$2?D0YwrrsZ0?ihV?E
z&m-OoEF+j0d!Bpl0k59u`PmnU@<fWTP=IB3+%B4?#$(tj+&*;<m3FoUMxeyq=yW8m
z89X}E{1)(0AlX=?hnN-UYB&itDQnW+dd{|?2O4%xLB2A`9%~UQgI&&A#+p9o@C!C<
zTTgi8`e*>5B=av{01m?fL*LF&va3qSIjEEkjl9+IYHd%4v>8xQeP;>WP5Kgt&!7><
zY~FDiQE>ZjcB2vf1Mv9)xQ6i3p+ua7S&58^?*}cv>jF0_ed#+&4uX$$^s>QobVjUq
z?F6B}|MDn0Cp`~Uo{~tDO!k$~2dC}!x(`n6iJCH8ou?~%XOlC)BImuEv{1+ifMlG2
ztN>%s4O9kI=)L5CMNFcaInJUY4vhPOe|Pel{!gs3GK&5Fb&T&WU=g`43k5XfUYvd>
z@b7~NA5;czHh+5xEaJ<W`Vpn7GLk7fQI1Mqm^ghnQdPc`uJQ~ghp$9K<5k>hxEB-=
zUd63W8{<{nO2qSTE;wGrt<nv{t2n%h`xje@_u^Km%<*2_>RfBQ7q?1f{(mY^6+gEB
z$scj5iBw?7WFl2+n$Z|}n?24j-Q?^w(?0Xmg+=<*4b4c1kj{|M)dpCG6@HDGxLSB4
z&g6|0UXIlHNCQz+x||;W?>{fK`BtNbfr8Fm3S8R3D-((f0+t`!NKEAO%OM2zY9UZs
zvgZukmpL$Gfy3+s@~wlC0^q}B5>jf-Of`G_i*o+Oc+)Ejg0^qOv8!~i{vpTL)8W_H
zITHrZSi`oC-~$nafUQ+qaggL65C0e05Pt^0Rx(bINsH@VRzw3l;r`}jujD51gu@dK
zFRFj4G`x;mWhxU-I6UG0#ggOQkze!z|LOVQ3HSd`!jWB{TU7Rh$iaJqtDBqVyE`XC
zEAPzCb|!~b7I%$k4GlFWy8{9d4E-!Sz`j@MZ#_NoiVUPwSK?J^-fAz4JmXZOh1PsB
zHu*SKM?=_I*XP%?&@-$b1Y@*sdi*mbp{?x5*vpEw!ZZXPgK`be=IE1KcARF$xKr^1
zoHci6A%z+MGdM6qlc+Y}uH>ATEc^J;tu9tc66TW&y!Nfv{F*1BfgMVX|J<xE5)l%@
zc%#COLDgEz=3Ia%cXx$7e`#a?_)UN&7)))W2v*Boe>UK55KP>g``p;}FRJ8UJeJ=u
zF~~q|a;P~M!3NlpYal$|YY2@Zx^FCo8t;jh5)dP8{*)$%gBQ|=GhS&dFJhm%@NU~<
zbXwO+V%~`t>7^p#Jof-9`^8z-6QmiI3bzXZl;q;liDvReEFD?!DSPJWFPf?M>bTd6
zh%6q41VWQr1;r3PXm!B)!Nv=0CjiOB);zzJ-(6l9;KU0E5{JgJN9>t%reYsG3J3@o
z{G_C{^z<NSp`T?d)$)8jR!J*chGLE<<~4LVXioNy6d7br%_n^nS}P%ak=7SdEek1v
z^nJ3*%G4wxlRC|oZD}k-7j=5;!=ztz3}-s-)_T2}Ld{OyY@{{$b!v>oB242I13C|>
zr^xh)WX)1R#pIiNVPZ}W>D)5wui+A1Fw%u~`-n+}JxDj*k^GmcpTn{#KZjS|my=mk
z^)-55&%ntUAJ#s<O3D9|L9V+)3><SX)NMzJKI~JMFZl9CzOc;4Foe3JoWIQ#Izrt(
z%ZMESIDy`$F60ccEYJSH0sxo{c8<(JZ7#h!U)B--fB-dzI6y6(?fvp5p_EI@orr?N
zbrYQ(qrZcViEdtD_swE&CpExo$?n1~eq$D6C9#jNNZ7tgui#IIRAGM=%h};}SoruY
zArIgh=>V*;f^UoV>{EN#8f4t*o>~YHG84LTXId!^DN+YZ1}NwY<iFcA<GzVDO{hW?
zJ4u<pw{&|`=-HmM(f3?q18Qw_FvWLqU@rWe^}!0kfrYD?7W%Q8#H$R0{<KwIeNd=B
zl6|K_hQ62(`k>?}D~t>P5|Hr_Ba@7SD6QS-qo=-FfHY#<YAxmgH*u+vQ%p=OF0G>j
zkc`<hi)k|D`{N^D%Z1aNxy!`y06N~!NavSgp#z={un#^1)RHM0ulWIa&2*cJ^T=z0
zLaGp%d4DzA##~VfaD5k%HB9`yK6Np30!0t!My@7YQxCPL<h@DuTy5#eo+H<<h2~`N
zJ&onMQKn%0Ns75fP^&SiZ_G`tS#dyM{hE0-QT!MaU2DS@gjW-H+=D6G8Y1<X7Tc3E
z!!1GjO?&|&JFVYn6uN%W)?!!7*$-4E1(?G4`iuZNW!R}JuWW4IY+hgfW}-sKAxAIT
zo>pkOhs&^afwf*}tiZ6~>$~+)gVBbaz16`T&;nju6Z!e3?>x3`dEdru0s`|>LJ392
zYn}_a9t8q-F4otSlzs;?G<ta--17X47z+&9sP);l>9pmj#F~~8khj`R#Li9xEK@#V
z5|O8EmK_K=v@u2{lh0wH^Q>lbT4oSTOesw=%_OsLYe%%9$s-OG!D1<9`^?gu6I6iu
z<qNIHazTUVd9w{N`cl3a=4J*1kS8zkXUUZ$K>lRlzy3kb07<=Xku)#CG=~zLeP(wx
zI3H-&@ov)yAf!`LNz#~~@O!_vlpHa5t=Tj8VYZOB$TOD{){OU#r%>`o2bsRR<Ak)w
zU7loYO@a=!w5htxD3z}dxlcXcc>Gq&?R)RvZ%5V~UCQhUr=hb;I81AO{e8Gv_F7@r
zOIRWMsc`xH^L<`Q1PJ-Z-@^dngC>K_eQ{aik9%9Q^}IFv8^6F{=N|6vvb0VQDoqy}
z!8-Fs?M|k1FJ>EhV#nH}qT}3TL>1(_F8T>sx&q?xS61v}ec}Fy_UeqO>12+HDv4YV
z7|W{>tXT7^b01_R^fPOB)`rbbi2xp2ifngmTu3#H=L6`Ho@RjkbX3jra}oeoQ{c|8
zN(wGfZ2ggm?nMSw+o36&@<@YG=${3&SnbR8G{v-7Z=IS5^(9447>k!6p-~=c4X}t&
zO633@{Rs`~QevFvo@XiKckbV?IV3vo@qx2Z!i^^>RCb8i3|`VwOLPac0C%;8j(&iB
zx}5#&<zaP=>>_!7E9mi$9)M=itUaQ3SCbc&f;&g^UFvjwVE=*r)2B~sP#84%K0g_D
zsT4*(Bj^*X#$0yiJh$(w)vNDqnz?;uNa^is?Mhl$w<{J`|5z?8iwt3odf!8vDqnDM
z-zc9Kn;s@SQ<;%NI_nts;oK3gvU?{?4yJnT5m2t#z_RJePQ~+Yl!KMeQKr`%xHxvI
zlM=K3BqrEJF*GwtF}9AQ$>6yS89B@86E!VS%^Ti%u``L6=s$k3Lx;(<&H8SEk9C{R
zj1Tif*7KgZte!0G)*i=Y`z7DVo~$Q^0d%cd%hs?!T)n^MAU5F1tWg=4A&F^s8EeKQ
zOp(2>beKL-ZM<gCw+JQ>j9N3LiZi$K95BZ*n*d;JdTEJK%-5teXmpd~Xlsem<jjl<
zOkOaJ&mLbtl7m&-$~$5xG*u%v?Rb7XA3CRZ>(0Rja^^=xdvVTN{ONc6{2zge;89=t
zD18(F^dvT<80gcDE6dAM!B~QbMB`(-wCTpK*Kgi%jgvbO90)q!nr^CX1?be0H2y^H
zEQM@04;NNCE?qz9p-Hk<_u>HQjw;zBH>q~gAND86Urz#i1ZEFwmA>mtFO1NW_7gnd
zYhJfE*kf<=(S@wLSXX<8!8gmK^YinksIRE0(Rm2=-`|NXehWAd5)yIK8u7}vi`e3C
zvGAP(XZBAQW;Or7L_znZEU$c5;D*iH$V&)++vecsFduoB?(FznkXwa&R%Mv9(tb}G
z(NWmzz>}q0GSTirM{cr@IMk;6m8DrotiMga&*{!>Yi&7~B?fmPs{NVoqu(&XnDSRH
z8)*T%d8Ef(Dg)We{W{tFW{7^utp-Zn=FwxB7TM$g>Z$F*Ph#yaG*hy#QcA2WK#czr
zIcM&|nGoZRYpugXY($vY=xoL5+j%=M6%`c;raf_s`Ytql%vnSF+|y#u$McPgX{dBF
zhng(!4G#~W*4#v5Gn}sGl#`S5PTD%)EKSv6!`5nRZI&9q&y99^nY`v#Is>nC*;;?Q
zG<_3+QfdmxI#{W2Qo{f2T*9??3#Xf=W@ade6<Lz1`fEAKZqk0@vz;#cPM25wKo<ha
z{akAmCfNhX6@3e$e)1cpZ41C7PFa$y4>&Js-R{+MQ8Xzz`BY=v>(~1LnQY&uFYoR`
z@GLa@=4+4c=f<rm%(Yg%uMLR~9eU!%k*jsqws&}#0|ld2OE0TN>|%xh8EhkeA!2*w
zYWjG_qBOIw(ZW(?=u*|EPrGzKet5!sl9J!9iik62Z|l>U=DI75x1F2Ud5GE=SoKG$
zee=^Yl2B7qL;POYw$?_~JjlIfp&|pJ@~-c$+K_nYuxa<}eKT?Q>S0&uewtvw2mh4z
z?ygDuMgR&f`Wlv139Bv;_Exq1MC?L4pK@C1)|Z3};u4Gp@2=6$WmO4G`9#ud2YMmm
zI;JI4&C@IkhR0Zu=L7I$=hxyp0paKUb|KR2keqyy$mvT$d+kNKa);GulE2ez*G`tj
z+<xhhZEnyM8z@W=pPeG~sc53R+?~f@G!Qo4Zg*gxwQfvGxicii5383?HSRxCaw5cO
z?P}x(NQqX?Z%w51vMmIe7DKL=@)qVKB`Nu>svyE&Pst?xlE>=I2hr4&^X^IVw%8Oj
zbXOn2l(&O;4@{znA0UcRTQ;fgq%o5{E71Ld@MJMDT>{K&wM$!9mX}Lg>xb&-=;^s9
zNlO;D!lR=$VlvcSV)jlC;smWMc_u!bd8fUNZt{kkcrx5f$-}(SwwWBxin3a^r#Q;t
zWMmOEIy6g*m=QOki@N7RAyl}e8XW{1mc8>J`iknk0^#k=h<LgsF+SJxM1ZJqNH8Kj
zJ-zn@<HjU}G=r_#pfiqLUI=WatLND+q&4h#!WO3vYuH_~nEvkT2+8~=|HXIG7R$D>
zN7oVbk?&c+ipuwWS>Cw6MKlW6ppF7*iM{;cS?u~Y3hklX+cEWQf%^uQ2_pi2!+aqa
z8frQ|tTJd`v^SiB>SBUeX(LdPu!pJTOJgIYuA7Lh;C}b$vSXyI6IXG9RRjB9#k{f0
zeT-f27v~yV@D6C0k$2T0K&$1<B2HGv_~u>BRqHD_6?hY-iZ2b-9u-(e&_!<5?bG8*
zt3E#|@&3Kj2&-UR)oqY0&Klv6*z|Pk64OXplXp7+K7E@l%m$i&)L0BY&ta>r-yrk#
zH{6fv-x;x2;N~#RlsA#iD40_5P-Ii}MZ=<WTq*!_h72&T545^+A7K94RPgOq?77?G
zrG!^vZIX1HAVqW`B0io?&|vm97^kWR1>tp3k-8nDkq2x=djp>Z*_xY~k<n=4V4k15
zRDX;4FgGD-B}<(8?cs*amwfXo{Dksl9Oox4R00@wh|zF*&qKi*Ar<_9?h_KXZwS!0
z5p#7|{kUfJkf7i+IqENvPqIuuIX&%e6X(XMO8xRxCm~DLl1<WW=ZuVu3zc5iwP8VN
zb(gKJt%RkeB^ti28E6}C2Z;*F+v0^=5U&<H;;2(&TK<xk+ivtOg{R<+q>6}j#L3BB
zSYFx#f&1h&B_)e>Li=%v0!|M3yK)7Vkljttg?*lV`CX?Mq7W#oCadY3rUukXNA}d1
zw3ydBZy}}Jl>Bf=v*xz*{CvinL&6@xR8+yW^$iWY9+`@3-BBuT4PN!r;+8nL_DFFV
z8AI6Rlgr!RS1C9kqFx<j=`I*{Ra;V0^36W%{6K#xMHT$FqwgUvFE1J{{bNKN$A<){
z))DMx9)(SAE|9<Ru!`&ZOtK|lwm?#H(6jLo(VhF_QL%^G;>U`Lbk-4HxJw;WqqpX4
zx5<{LBWlynqBXE!*XYB<;f%B0{L97#4X~f7-83MRi`s7V7=y=y#_FafQe7Qb-4kLg
zwe6;k&cmcz_V#lXd-`s#**U#W50N~<<9*1@Z-F#ao0EQ1is${YYEM-fkE7i~4AY*W
z!QtWK>nSOzsP=n5C5l5-%9><db_j9(7B~8VPJX-@pah<@XSOZ2F$_<&&g2NOtJ*Ey
zU(_7x?dx;r2@MNl`jp1F@g%WoL^NrY*S%ccHBLL7f!WO0Nv#$P;(=QUTdg|9-VdcN
zcim?W;5RgwEz=g|)o}FnXl;dCR#9m&enuC5bn#JLWPH3E3Xm4wfZYB1v0-ro*pST_
zSoxKU5Sv2y<HNn5^C@y?=c;OIE+vQ=Xv)aQghZ;S_&q8*z{PUrenE&-xzsIv>9qxC
z>j(}2db+HM3DquoS!sfqmYHkdMoVs`O%K&Ffs|6yaU~_acT4MaAMSGSQp$b%c0fGk
zXkdDFw*AST)TAVoHT=_)Um-RpQk(yzM$w!Jo_@1MQ+fMnzGM8gWei_;o#%LeKdZL5
zgkd#9_i^?tT$bw}sthM*h8#o2xD754Z`k}WGp{aosNI>$V2g-|2w{EB(){lB0#Z-T
zFZ{Ib<>lp%p7jmJ4BmGB@Pv@(M8@eg`^~C>IUO>+w?V*I-iT{!YtIg98*Ue}9A-_M
zaNSKuqU}HDf5*PF_Dgd!nQmlUT*c|m$9rm8T2#y$o;|y{U8k2*Q7vU-&Gl&ITAVXs
znq+$hEZH<^#ELri!2B>~{jO1=n=An%ZZ0-fIHlEj>#TuRgiCl5(jbT>j`sWw!}Z<l
zG#M^`Ot2VoBkZw<1@&8c{AP0C1slH9kR1wq{r0V%a4OAbMH(7gH;;fLU0;JVSuMrK
z=2>S3JM9yj>6*?_kDQFfnad41pEtu6nR$_LXGzTMHf@P`(Z;sJ+x$9l*5gW^-5!xo
zG;eE&OW|wkiuMQ1BSDxmUu$^pa<qd$(paj@$LDIhMMp72MAPn;nx>{nVg|i~gVZCj
z4Dl2P0uk40f4nyRukr~FZv}G(V0L9cq;}qdgy*HNYRgl$?ZcVNp<n&wRnyto6-=bw
z<uFnPZ{!*L@`n4t5JPN3b@c_aW{_()53aRoOYq-#VTv!pS;+(<(_Iu+gKq-v!AiBb
zJ6LuV8(rA77D|YHeDjupfk)D$2aV8v_p=(Z;HB(ES!3uDR0>agH2Qcb)n#dEDJxjT
zqG8kfT6r~}dVjP0eJ<ySz#yMYZOF;CTSsh|HAZ)`l3%lE-^t=VIAszCt^avdfs{zr
zE&tF_z~Ota1p=ar5SNEl)z<o@7C$~f;i4BeXQXO-kP`lTKiOB9Pi!&VyxYds_N;Np
zKHI^D&s^8)ctiAOCG_?6LnxgqC)Qk7--Wmg9kBwWpiy^z$w(GawrtVP2MowZscnxB
zJipiVahWGR`zi5T?Mn6x6-Wk@IPpB4Pu7zk6X_}Q+1)=daMqp&i!=QBMYrwu*`gTo
z1LA?72vu$MXT$~aOA`S9HY=Dz#ikGb*JbOWP=gn|(N^LT!{?e(M_8t+_d6Xg^RvD5
zp;Aj?t;R%Z%dDfm|I-+A$Xs8=0A$4=k+#)b7qPXra>`)hXgi78qk2^j=YA3$fkf(C
z(`$tj_<eTz>Y+18I5XIi$qx&NC9b~hUhbkXZb&5Aevez(zg{)Yx#6w$sm6@#Y>`iC
zW7J7nAM1iX(wINJ?HqJ^t;y(1NKHB_opsXz@vq{95Cpt0_2zeiHJBWRksmqLcEx4%
zOIQze-Fk79TTwjUjWdpIj*H)*`I1>BfLfpup6rZ>eRfw6+1|4KL|}M0bF)$Yp>{y9
zs8H&5WqWo-SwLdg3&|>;^A+G4$;zjyi?cVjlIy5^#izg5+>^mmSQzqZ9)hRVy8i`p
z6DFu-rBl0NHf%1~)jKpq3`-7`W@j9ity>q^;+4rwM$Xvr^6=avl@X5Xt*(o;J@}1)
z<;)y3jZ8!|`Pq#JUBH^-G-kiZ{}FshR@FkY;_U6b7aJhC{a!z{WhbreAhY#ET*gfW
zL@`<qdDkT9RA;)<hG%99A*!R=xRaMYwauA0-Xo(~iun+AvfA3Ifn?+L`H^DS3G&{N
z5faPhG4IjQ{OIxSZ+G@`<Yp!2?=^Ap*j434l?rAxo$i1&VDN}SZ0V6!e{T!v`>|p=
zA&uNMHyzZnfO|=4SnjiB$9+bl9zRFY;mnbDS#QohSEY>0GEq>qJ-|n8T%fKFsj?tR
zrMF>b>xc`d1^Ysrplb1^*JsVm1mIPIW8atJ>-`(1guoG-+O~P7d7>q23_1api~vKe
zJ6YZ&Kj`ELt4$WKUL83Nx9WP=A@ilV5vusun6(bj2WlX8qU^H3=5+*}2S#HJ7iWxS
zT9&yD(`Ry*?~H0YP6tdhvSlM7CAmmPyu;LS(B0X+X=8gvXFFxw)5>BLxbvl$cauEb
zW4vz7(4MT9z~OWPaF?C*J7m5FEcH1$8QOI=SwDdixP5ykT`p66HYMfxqg^40(?1yj
zVAVR+T$=>H*2?ocm%mDVm{vwev0xk#8M*h3+NOWn#1*;}n48d8B^%4%(+mwMTkLv-
zW}GvJ=QFdWxhrO6+T{W=_~Nq_DHRo&TngPPIc=I<=u|J8m0`Be_Q}xE89yj0ayF_s
z>L`E)k0K$^OT$X&miWSHw7f{-lsg~2oUuF?OF~4E!wnn39JeGV*DekWKR!$D0(<Fl
zE?;>UFi@9GrfO%b@-AoWrX9pAr<D6U+$db`ZWq)ka9HzIMJWX4yd)VJ{2l@ekCz0u
z6RFK!qy?z-iTv2qRx)rb#^)dFKtM_iUhjK<5t{WQ{Eil83&W;yvB8=nOF3>Ff?{Pp
z+-#pNgq=GopltMx&c$)5$MzJ>^5V!cJy_gt5t<zqaUk?rqwgE<d~Rw~%nw-jphbw~
zS*sd6eHW|}Pje9E=Id9#7K<(AwnKBqanM4Qayo`><Y1H-tcN-qs+yYkcVvv-*eoG0
zKl9>Q1>xxv#5yI1>>9M_Hn+`HWWSD%)+?}I8fweQ<z!KZ1#^&gXfm!8X`?ajAMbLc
zgawDE2J6=f<Qv$CxK3DNK{3oWqVIu!*ee6#bgz5I<7tE(<Ra2F8Wj~4i_D(Y)kU^v
zO;l_NCOojv%`cR%VrhTKZvJxs_2<UIjk)8c&afMgWH+@1QglWZQNI&)|3$NbvtjDf
ze%V*gj@KN!EuOLg5ZclcOuEWnXJ%&JBr??X5$=A=2~Iu>9~78rx54hA$rtssy2l+I
z9RXrI_u!s;kBEHI`og?lhlSk;>1$kLj$(7jaZnztaTj)ONIZ~mqr5ReqmURPueVV|
zoa9Ne5ou5bdw(dkc~5`;spTnq?2zR>Xz$#us8Ob`wmdI(FJN)-L6$XCg4S!Ni<bFX
z>!=2-c`Ji1>7GTk*|1o}DM%Pj5u{gP#Fi%p^8H0@qjH|Xe^+_gm1jMShLY*_nj1hR
zB*AWu`F($_-YFXS7}MG)I%h*itRoq<ML93sc98V9*8<iI7_JL;*DHnUG6{x^2r3Lh
z;1Sw?NghbPbg7}fo+8r&bQwvHpJheFgym4<ZqeyCyW0W5b1b5BtNHZgq;tIe&`7@8
zLgA28Mt-8!bEmT$3d#vGYX$qp@NBIKf5XDNlio`d)7siP>QmWF#&_$>i<ccH$4Ebi
zB^So^JPBCn&1-&X1lu1x(~4TNZS&pv9XQyp9u~9fm7v=On{m$|df(WNgQsJxIACkr
zLbsT6Pk@>qe_T$EC?W@&9VaCnU(|<V)|7~Y(DEe+WXPOF?J4?`1h&&=_W&(?j_fko
zdstFrAwKo({jMu9vsPBcLLMa0AM;!S63+JrVpX$msO2H(ZM5fw$I=elPRH1fq@>hA
zS-YaDHAf^1k<Yty!sO5et!m_Vmqe7M*=9hx{T`_Y+n0Epr74_}`QN#pzh3#a_({0;
z#rZVfk(jp?n<siIiTGP*(?<u?;;GEbA%VR$edemFxw(1!d!kC>=W01T-;PtRIg@n#
z0>~!0p-TCNs7rxx9pG%Me-+{4;g#M-%5-j<yE`!?vu5H*(wCf_ixvY+-w{krnYWh*
zOyw@lb(`wz59RbUHa45;mx%PVLdcA5HQWj5UZ>PnNVL69O|>k^3=h|CGle|3)r4tf
zUS(@lOA8G|8xGis@=7QJfelKxEY1md!YpH>_iFvpQSFMD(c1Xdy|h+eUyo*ww5ah!
z?CW<PahTCnpBs;UKiKd#=see2Ctbx^5k?tT+<lVXo&XB(`B3`x%6VV(CJ2fH`A!u6
znInq}6VdtEGp%$`HexrkusGFvfa0PtwAR925$vK7w9PpyyuESCUvJIH7#<{JH8B+B
z{u`!jD=hH5<7KO6J0$r&-Z7gaDb=uPrZt0i=B>!Bwj2f?etwsfl*re~$>v{0cmeiY
z1k_Xo(ktU*E(ayJHCYYJBk(vMG6wsB6dv61LfHDEK>NMUi?R=P{Wr_5Tu7hF2c~2-
zH6cH|rZ@ck>#}p;m$NQ=_<Eu8dn|9k<v5~tduXj@?t20VQ;KAz=PbY7L*b;L(R_eW
zpUx57l3AgT{9{UA%7L?%<}xs^HGlDY7OWI;f6FMY*`jf+;(_CcJGDe}bl;C({udK$
ze-KLcuXcyE=IoH4^yY-Kj*paVL5EF_fY7r{5Pm^2z-#A}2s@<M>L__070m9<PXV-B
z!iO_&dhKrpW>CBf&Kj{GWO5vZlS0c(8Hj(W-MSPM05pO2tZH!9uC_;{YYkm~H~3oQ
z5u6iCf4m#3YM|rSd)fKtee@=9RxT3Rs`zE+M@L8!qhFf<i9(dF&5mH3mx*zgCKK7d
zVmrV|EHwpMELi@BzfbJwrQ|E3Yu>hg3>P09Ve-NjXXNLKmF&To04ICx#kA*+*sJ;a
zM9Z__;6xS{c-S3wN^S<(S3YNUqXY?h{~%OqnS6E8VKECZUDtc!E~E2u1<<0lgR}w_
z5NQBAhdW9%t!KpvHL@b}ssUS}?r+hi#Bo`_=yv``>v$nJB@O9%-smtt7bx(8d))4*
z=(G|KuvLT2$!E|}xG&%qG~WB50ye$*x+EwKi($o3bh!5%+yYlSI%v%s5*a|Sge+q(
zqM~i-!{8QP7=^91zVka@HXv}eewkNr;peQ3dw0meEu0hhydwF*^<m5)lqE|lRHAXU
z4X(`nz^gZ0g#NkW!mGDci7f`N-d1IF@ak=4D))~_3a{Sq>g}f#z<X~$B@a%i`-k^d
zbE3884=jK`5*vTWAkN?T@sshYZiNK<554iK?#KM}Z|sQo<nW%{Pbu&ZRrhcF^e<e9
zS8sUr_EQS{znzD}%nv?|`cKM+-e0^tFX=&kSRJwLG_FkU=Idtz)dQQujvbgGv1=9T
zZIDES?X+j!*S4ATsD%#-Hp5pA;7Cu=d)yvz>x}SuLW^mfd>XXtr(*gqa`%_miXVX#
z+hgNo)}KfCzxaDz`ynZIFymb4%JMF_Sl|P)I;N;L3I|30I>oPqk>ak9ZVu@^;(dH&
z-^20Sj7Y|AuY5E*N!WwPKA@nFm$ssF4G$szcU?Iaoy2;`X~@{Qa>yM-=Nj+~w>sg#
zGaL>{{1<ZKHQX;M?|+7<c!t9>94^cJyD7k%)K!WW-lVQlwD3OMzgRmQbj33q4#NBk
zIsflzxMeY0%LInPog3EG<S8yK$0r~DIyI?#DwaI`d?Lw-9Fd`O`rh}Shpvm<XM|47
zxPvh;aG%bLpntiEU%bDW%`7@;#kG;>#4295?>GmlyP^{@HZueUw$sXoqZ2V6BiKrL
z$N3;S0i%&89Mo3X51oK<!xk2u&6Hs*sN6xG2pn|S*c{JnD-sYqv#lmShG({)JN!75
z@c$uZ%Ocd;>GV4p(oTL|H12N;f8yw?YAb~+sW`Z6cKW`pgG0huaYbOs!JPt*t&VM^
zDmXj+z8=q;=pZ69JhW>zizPY&6(#x`<{2n>@>kNR{8}E<Izr-CVTagINmf?vlC7%J
zc#9K#WknqM1)K@>&q-S7wCyT_=~2AJhfW*is|{p4w_c_IUcrhK`Hw#=epdXLgj+e&
zYLwjn2=TQiz#P(_pLze!UvM$?k6-Hduj_#I*Ll;hR-ohGAAJj$BSE(@5gniD5b*6G
zzE@~mz;767U=BWMr7NghLEwSFz}JO00L}#k_DGNtKb9Y+`w<;)E(!QH>B}vglMpMS
zCLb`z7an~DblSu5O6{LH+n)*Me-tTRsjWiI;g#Ad6cb*ltw=Zi{$aeQwjzfQuhdo{
znE!F>@JekJw8JYkyi&s%7U9FsRY(**{9KJh;ls~W#6I}&a}}}A|I<pXmCwohKY6R(
ztqbOj5uJS-Iu77Sqg-Vg?RqpbHTMqNIVJXSTLEMV^T;vvl~{&}G27dmxOCUG_`I>o
z&y|%VIm2CEc7Bj7GiRW9@tu+jY8V+$;-2Hqog!+uGB<gun9hcG0~2?H1K&{~Z+3la
zmP0Gu77@-0-?x`=eCum4L64(hxAmB0*MhQh-dvx;Db|Cl&=WtR#9hc-ZRDakiPMa&
z!~?&+l|`q-HyYPDR`!g%Kp^h1bCT#}Tz%l2M|xOy|A+=S$d3DU53@jKdB^7qEB@k-
z)RcG_oU<(b3_4e^4X$8Mff*Sp=kz)V<#^M$i|AYdp4C=yDR@@<g*4{>=t4ZJ;aLr*
zcZ1hzs}U)@0bPYi;aLsOYJWM<(T+Xte_#Rp(-Xw|ZL4sm@gc%L+2~iq4bN(LR>PIv
zxQ&lSS0hsR#K$T`3eReIR{L3wZIsXIS@7?nJ?1oKMlvcs{JPlr_-vfe*A`ps91BFm
z#J(J$KLpPD^k!eN5fRQ-Zd)lFEoOP%w|ffZSzDW68biG@=D_crqPg^=Bb<ZDI6ELq
zNCmd~i-?uHiar5cpSBB{c<FLC&Ejl|o8D@rnq@itn2sQT>sRdauY6`f8QS_P?PJjC
zh1{?tTlIC<uRQk`KC=}XD+#@{a9S9garuv=;r5-4mbiK3N8iBt^Rh4v#B!UR^e=yW
zMPL8Nhsk)-)~`b6{ow5|$Dw_9-F2TI{Am>tXJFtv8gx#$nG90qTj@Wdb4p-}VRTL?
z9=JocO7=GerhnYX|0Dr_?ZsJij5e+EtN#=RCrIQIlpI;bt^MFA;>Yd^{owyzjWDo4
z*Djg(GaO2ArMI$`vKgHd-WJ~njn2O%u5|(Ce@Fr<Z2LcN<kwi>f4+rRVJl+_yb4>P
z%Eha&6{=jk3R|H+g;!yC74`!&jrYP<=2PLluoWt5ycf1YMU4+6SEi`(f#eE(O}q-j
ztFV8HnC#}P-2;Cn1V}JqGFt<Fo>`?{rYnH9)tTafk<85@NV6>E_h#9oWm28;UP4?m
z_Rk5_Xh-}5=<G5VqzreRCX=o561eR1{X+60@-7q^Eiv59HR3)qco>J;9BtY{?bzNu
zHHW$5+-7jk;T?qZvk7?vJ&SKXOmvWZo;!x)i#rY)3A0(*I~WObUa8t~^yX_0Lg7Q1
z*T<NC;1%KciJ!R&U!gU!UAJ9m6<_^93MVfVQb%XW4N*$x`a|Z<pFD@*X9xJfTl|wt
z{_}NZXemv@6AlM;{wI0#m$q3A;qn{@Tef&H%3ITatuLM|OX?m>C)q2tdCP}=`5L-A
zIyP#mlQZ6{Bk$PZeDV~*UE*~&$~lUZH?mP<hJMSxqfxHzLBB!u<G#cE8p-R9Q*7Kq
z6TO$Qy!Nx{%{YJmK27XYRh)lcOzB``e^hB1Cxf6MTU)fH9cLz+TFziD?)>2Hb?Ycu
z$*2E1*-sAr?|*tgK)}P8D1Z|e``>@z?`ge=&6}%a-y}%>IzRvH_fM1U2;BFm5SLO}
zp+JeB6u<54>Lh}Wk`NEN{TuNE_U$Ms30Am<OYDQ_^mcF9MYuud^mSu&Hu}=b?%Wah
zT0`A)<udo9%k^TR7nj)YcK7dWvqC58Z`_4(g=^SNWQ)?TodDU|r9h3dsJSwsyR>10
zPQ=wOC~1Dit(&X7N;>G#F`0q;;Tj6x)1vch^cO(3PLOZHt+iHkL3kCnI$6W3xK(Nr
zyoy_yX8TjI<5k=$-3h#k!>hP|p#a{CTb*-)!FzFk^N?3SL%bLFr(IjwKk$L-iro9Z
zK8#m!D-h4W2{q~}E`u%e&y123foA6>7nw7AMEa7S|8-aY?IDV%Z#Y$yDP+D#AIJO!
z`BUHv0y)RFVp>H`j&6YgT%-LZtUKzrhQ-lc<*Z<3R^x&InWadepYS^sUT9%7nY~AX
z<`{Qy;s_+gKWyk&&(9oB_w=E-C45Gf67$5dji^fY<?H$jI6N^gmL2Qo1F{^vUQ{0=
za3bqvzaDSq`Y)#Uzt4*E!&TytFQB|1$%5n2E6Y#;1qpb_OqHk5c{`mzNR1E|TtWwZ
zK=vFQ)%9qc4LZmOJhG55*r>sZ&QV#y;HZVv^4U1szJKvTf4?wFBt^BJYd1PaC6j}r
zYIsDeqLbVshNS3g%nMg^j_P_4j>=VhIetY1!u4UyM#xcBiJv=(&QS}IqrT{jlE<Zl
za0=#+1W$ZDWJPu8!-8<V{fbmE1{U&G$(9+{vY{jU7?6FQ9dan4m~athxHTO5E)JL{
zgxjeFb_{3FEG5{^43ExsV)I+Gu}nFZ78cH-XP@88$q6NGd(NpwTOqdmL4CN{|B2h^
z%N0@fe?Kg)3$$h#su~bjM{ts24;2;BO!LG@d$?FOce8bmNc(FS+U;XkpXm=br?>0M
z*q%w7?Qko~<-3=xSAX65mGD~cdn+4owYtld;>|l)qVsxtb<(Eln%?(_&V(iycZSJn
z?4hLGDz-Ex!<=(`_JmatH{ZSb+@<99tcePd(DAIb0-lSv!r@D>n{SE%M}=)}wrKLq
z4;MApw3@6gm-DF071|rq-Y1jGqHH@f@IJlAU(WIMnN{Yk{^WBn22jr?JXc==v^3@>
zL4H0%`;&l$bF=;OxvGag7K}L&QQYIMusSX<QO5ngzm_dOFRvoOd|-MkUjKEjqvReM
zn%KFeCc|6?xE_^u&!t}C_Wo!`F*7r>nljEu|9inJ`hUbPAa=g=bpNJx1fjdXq?^iH
zwP8i69md?SyNEL~GTyVP8Y@_Ju`Qpynr?DZ!{5rNDP^dgNNv_+wwphf^wg<SQQcH~
z_tKDTH10fMKVn^(d3@XawY8?zWz8@ZL6)-1pbliOWBxz(-ZLz!EomE81SLvRkc@(W
z0)k|TtsX_misURgHaQ3TD4;}@BpF3Pa%ggvoIyzf(nyX?&P~I&HtNhgGjpzYW;ox!
z^FuCdx_j?ktLm=1!m5QH#?%!K9N6`JCyvry+gRD{@)4~-TMrBncajD7XX{X+uuCq_
zay+MyV>i0oOs)=F7h^>#HnWzA{YjY=Iyg`LSLpGtZ$?xQf%N`1d9ODPt_)|cX%}i`
ztN5@Om~4XiX&oJ%u~SStOk+8Rdr!xNy(c!z=L*ke>Q-L$TzR@oqNcr?uQ|x}Mabsc
zxr1>Y$#Z`G(ht{L|E3)O<1frz0Fx%&HVWcTAifaNi@S1EOfdK`7TJslbtEB?{QFaS
zRqW!2yIs+dk()5C35I1?%aZAs4CPUK4BEvn)N64fZP)17nr#X+drjPZNZ|=U71ZHD
zH6fj_jm+YqrLKdDD{4I<{*6%2(*a$~o^8uzMMbaS4&hM>@3k<ohwCl>>4W<rG1k9k
z0sIl1Gv`3kaGETLg{09-=$Gq^j<M|!%(3~{AvU_p{<gSSxUjt%Q?B?dSvoXjIt=k&
z*~q`X8O8{~$izo;{v)d?vi8{7X*x9RJE)%G9`Gx5@JLTdl&K9k-&VIdP<?$fORLC=
zNbg`x{?CxPG$eBoiy=8k=8|i#stKMxm8QT{n{>>tWSZ){exthlDXMCt!C-a#PFll$
zkCfMCdJ$V*!9?QYLiZXQfNs=XIS&Fdj;tDV@hFY=7E<pK`tZD31(&09XJ^jl($dmO
zEwyBh#Yh-|bO=4c=>BvuyWXC{Kg37>YG42FSCV*fkn2$uzGI)sfesXe)&PIekRCiY
zJB!Z)pej8@_|n)HSs!-g>w1TK9R^lbT*|sXnrHrX?f&CeCZVEqMNqgi1+$U`qI{{0
zeSe&KTSZW0xx%6D%Lmz`tM0ES(kn=L=~z6_Dz+gi9`!c0kfb>-i<!I+t}GrLSp=@E
zi2zTSA&BvAxlBDE2Y)pFA?iRO<(>CN@-W|2*s=YgE`a~$wtMF6M~YnT0xwy}_ORGG
z91rA4k4Am0r>>mEiw@YEQ(SHhS3LH~8E&9leUXRdLp&~ixa#+QtwT8E-G&Y!wW()M
z&=9PAMPNCQO&}X}?LN1CQwU<^t6{A2e-0LZ2oO9FKp4L-2!jA(|N9xK^ucY&=RUtl
z8&eqX&>6<3;<P<-m~bUg?$}ns46KCl?tQu=E2;7d^-shu&In?Bwgyf6<lv6ufL6q(
zHA^#jaVbk4UrBF5suJ{C=txyOma)c=L<bK|&X|eJ#Ig5Tbs3FMcWAM>c(MPdi2DaZ
zeCYwv7^!EXuqUurcuUoXh~A|vy=oyrMHc=zht9$10Lr9`lwrq}-wjAb*iU_k1}dWb
z{Q$!xgWg%2YJ0dYz*iR|m{C{jOY|^sahVCfDTG0`gZ<dPT^zK<y!3zY<u2sgS87g?
zrfhi)=1-qCdrsp1C9G4ydH`_LuK?ly47>h^#gai>`1ZX9@l+6l8XTnA+l%aP=cjvN
z*`L|@Xb~(J|DRyn<$$(yFR{G$f~-xyEpp@gX<%3c*YDj%78|$a<N1EAaFa#Grbzz5
zivPz7gS^zeUp*fmVZ#v)(6xbbc6s#GP})>m)Q9QlU}7&Xujqgd3ZqNPH~Eeme46Ef
zYyvmqi9WD9IVq-Q3q5y|RzVD1U;$6;oAC0I<&Ti!&E@|S?5dy~TnH_BYe*V$vwIkf
zG6pp)CMDQqWu#t|$)lVHobI22zyEiQFk1|1Udw2-d<Q=rq!T7rx0=aM((>JGIp)bg
z-CH;#)}yybK=%ynKHllm&M8|J#V+0!34_hu0|QZMU+etO@K0HG$vF%mHLO3U8(M!R
z4E{rUsCk9#km)yxg)4>=%S<-LXL{v9WJ8Oyx$m9v!0O`bOls=K7o2|o14r9?xm)mH
z$$daM^G2^AF6!Gu04tiUIrcyaA5nj;K7?)7>!ogNZKXiEbDg)o!fn0w6^pb5Rbr&>
zdIz`S26?nN?yA9k>gpDukUwbgLhupPpRt+OP$W-!D`y&5MbVrHlc*kD8yJo*{OZ2i
zUNQMco1*WWdwvqTprN=bY;i}Nv$Z#Mhw><zK?qTbI4`P$=v2nKZ-00F;d;BO-<?3J
z6e<Z3kvB2Yr;lqxZe#%WNR!_O17lVA{{7R_=sj_tgMGPO6PBKDr4nfkXU{U;?_l{8
z@Xm4O?W-ZsREWqYpci?`4ZE$Trq*UQ-5wpLx!v4+4|>ez*TRLoH1_PdZtF$e1I4T^
z=!C4(p&=nDRM=(~twtXcKjrm^%XMGBs#~ZD+jt7J#|jGrg4u*O?zXd17y5mtum*AJ
z%tP++nF~n5o=EOtPM-8dh8ukKRskI$4CN^mGAIex6<*{OfpVAS__eEUE{6xF+EcoJ
z3oWqjO7Ln=VP+6@bZyO;9Elfm;g7vV=~JC*jm3&Z%inCZw`P*k(y@HhJ?{0a0G-!V
zRNBRhccrbuyeCEa_C9?s19dZR--g4Dx-er`*EfUbkkU*eqEjksOWoZMOq9)v;7o78
z1-DnmZi4?gqlfsx-)yu^ZQ~#04Hm<FrtU;kzVX+<WAA)nbwmY{7!g@o7_TCQlkN7S
zEZKQHngzD4<*k)sLXpP7UJlLn%lTZdHI_YGL)bAi2Eo|hGIYwn46KGQ7B)(%ShqzU
zr14O0Pe|S5W6J6#CN|PAl&a6X6=_cLeYGhoPw^rVW;<0X*)rHGT^)aw(RxUTS0DWT
zK%3~^@Ho~k@#OgrCDY+W;8PDc+K?@wl4X}c;2*csNKxHd)Oet-USfOmMTANJrrErU
zMgv#da{kyDxyku_vP3+Y;Sl=k@ED75<6sF%mN`@8yU3}8yK(Hi8E+b9Zd&8w_fpMX
zpc9Kbh=Lcky*xViiiE0Uxe8O~G=RNtkp}m{F!`%hjOX`yB43RK!#!nPS3Vs^X>M>A
ze-k)3@#$T!CpyZ&8zY#;t$B`zE+i<?C5QNMze~9IPN2XaSwERsRSk*oH!{>_woBiG
z)m1@%QNk7III#A2Pd3NrOP-DmS&q`%RIDQVkoPYBkq08=URiRf7W3RlX39zq?GIe2
zS??Djo3(l>WQGGp6LG9$he(yktnzYB*&>aWo}Qktv~OJG`uyivk9%4ef(pu`-ttuW
z;eoz?*Drie8RFtpcnww;cf&!!Otz4kfq|j#HT|&(7ac^n_#e^AJb^{yg1iBrQ!x_h
zRNd&|o{wkNOxt`>+28oMspjrWh<>awJWw<fI%sk7roNGph%$e9SsC9{Y_Oce^oG&Z
zqVjSl>-Jt)w8DM;rEgXfJ~F-@e<(%ZXe1>135UF23P3HcNY?NC`jPtQW&a=mw7;aE
z#SF1MB#15RGCyC1CWNP6;}!pQaFW*DecEVV-^Ilx#glJJ_It&8LR`*iOK4)vzYVau
zl@RM)Qq^y#t^`4@XN|@7V7EVTonKS)kIr<X<<YM!Sxu2?j{?qqmzF2%GAxM|O3+2S
z4}Jdp>dL5<Q7f(M+YH$!euFD;r)aoqo#0#GHZx-1f>hAIy$M(n9?q&kEOo_qJnoMg
z5zw2u&*J2`_o=+ift!m7+@bsT9|LgAMil40H#fG{#sjy{N@wZT1u;v(fU6k^co;U^
zyLZ8&i@0y76E!sQ^^J^-M1WXjmqNYT+iT}blqB7BS7KC--ycgyDXON-X`xnLc;sAJ
zS?0+;ytxi=bIo}*Ad+p8BC0w^#jW{fLf1_6kHpKXi~x+zKgSDgIpL!`&2-q=t8nm`
zV5fMxfmX5dA}om+NUkp0-F{<MDFq<L6<M+uJqio)!RZl$Y&cPR2l=+kl`g)DUi188
zGg8&U#3Y4Fo>r8vS54}_sxb)0I?Q&z&9xpbvfWu3<2CJ!BV%XdRL#U^U??0_Lw0Mk
zhJ+OJY3WCWl_Px)5#6KJEW~He`XwYJ<l0XK<XUNrwhY?ui-vd^PNblRi+L8m=CRP<
zuxD+L7ZV1_I8iQ*d{+Ovsotpe0wnM`1gB1k)qPHuX16&9{+iywbk%C#jh_;kmzS3d
z0?=HCnNC~q72U(bbRfY<2T{!EbNsSy6U#BKO9cZ|9nOmv@K?112Og}7UWpNu`uf28
z!1ggf=KXCSpRY($#pf6zDlYt0QrP%%;ox9C-u>A4GCE}o@!}%u1=3Xq{r&LR9SM-G
zi+N!gvS)Fxx-l(hzrRxK;B4T431)ja(X4*FA&}i$sJtAB;<H4Vsw{*nWZv*rwHYas
z0O=fN#~&2}Ic-Zvr+80^o}sREO0AJ0wbdA|1k*F5jBz`sigrdXt7U24SW?@#QIl4}
zBn2P$yb{x}s~jiH4H8nGjJf%^>nWNhg*kSh3UQ$bzRZ*;%s`VU+(DCQdW|U&el0^q
z|I-%}@8I>FiBLstOS#V2@f-aEYVS*@C&`4DXid8kDk;}f-o8yXKvjqgXrHIz(G}2m
zaJIVZK(KSN;)=I-qgzjmpe0+j<uFr*;DCr7>fFpU__6$oDbX^U<`AVow1;*Irks&k
zWmJk2ZrFu^AEnb3Zdv%5!!he7DN>4`%>gF{7n!9#{3sC9(A70KJhV`emYPUJeT}%w
zbDF;{7PXNQhk|FOZljJ+hMfq?dy-V|PmLVq*-J1>iSn6K@o}l1IXyEpb{!6DRno-H
z=axM(H4D>mw<Q#??yKKF-KN}@ILpLHU#-@QUBEE5HFX5o`w1SnUq`KnSaL-OZbeYe
zXdcunU@M}{^|qISEf-Hwc2(8iP^s`QH)<J=rs}ER50-+P4DF1RyM%U4oH;NImOyPk
zckt_n7cPG&O7~i=YU+$HZyXqO3C+oC!Z5Bk%_Lj0^#{ier$mp+TH9c*7B;wfh&PRw
zI(r6V3trsVU&O9TPb_^<b^rXW(u``^V-bAy<6wycxyE%XPz(~Kd+i;pY>LP2;ZZ?a
z>V?^Q;c+LAJdlbzMx<+e?@>{fTcXO~5I?3gBqwkCu!+{glg+v#rq5tc%E1zA+P7$F
zlyOl~i*Duwdv;z9t)7^qTd;(dTSKSQe#MIg7m&}2SLX>?Pbu-Eo{?Xr@m;flG=)J_
zd=<0U{dNNULDh4Aa5DdS$3F5ZxTmfsEpEyi0bP#NaLaFkognj}*dEI^@O7f%illJ(
zsP*(rha84>U}ap;dY|jLLx%`FV1N!?UARQc4isx(64|mUYWV4QXICLN``)1i+L;|%
zOiexih#g8`SKMVEwV$<Rs<J*DrfnD}ytz#pw_;@v)MZBB#HFpvvm)~Lz7~RnPUzV}
zPPO;AX0ocTIc_WS%T4ojKJ{4^0BS>bnRPFi4vyD#XARxGJN`antDiB@rf`^}P!fUc
z6y^VBTc_8jcJD2Uf!)}+>ZG<Mui=T%jZwNn84epw?QRE+1o5h-Mp9EE&G7w$OH9GS
z=JPr*o^-{E5Q`DPhnS3_9wwo}*na(T!H>a16SS#&`X!BNHsF^s`HHF+12>B1aO@-n
zhXZwbEIXQBi6l~7$lFsd{q>eO<@1IoE8LkB1uu7|<&Pvw_R+3ks$5Cg7wQIg(Nk%K
z-vkw$ga<_WtlvZZPoR$VSZRdRYnEpn6@0pAw6b^p4N#Qg>s=ph3=<xVG>Rve1$(6R
zSw9W{KfOYN*c+rc^5v;*d7R2xCds!SvR_W|aS7%hF<N_W8F&B1+^EikoN%-_g;esm
zX#}?h^Xcpm9|7ywYu{;2p$`m*gi-WVmp4Y+@VV84H30wd@b&<Id9_2+7vPi6t5LlS
zI9wzw?$eKslMdYcyQ9H}Zdoc?k<1jpSj|+{d;f?9{E<uiXIGT;7Pv%#aAgDFxy*uK
za5y~WcR#-&F<9O2csK$^wD*piP8-q%*|Z|2OCU~83c{(ZT<3Sf%zl1*6olm&Ha!f~
zDN%aj_lWhl8D9H~-SL5{7Oda_W!QLqyo*kd+`cWX#l|n<%38Q|S(Ykw+I*CF4r@7D
zWT~-$tGUfTSSD$j<;G>%NHI;vxa_bGttARtF+RSk38clty6tFu5xB{_%ggVd{8&&g
zRe5MZpYD}5M)!)Wmak9$7MhlGw@Gg&Gv?}Jsf;J<Y#HI6%6Q|ZD(+-MY0Dg4RbA=x
zh739I@wjhy_1SQ;?ZDUZK$at2BcANlcuA4lj;C<XEZX$Zk)E-9D0|{|*NN^*A)fZc
zqwQ4<m61lEheDmP^FrBn>BINZfcB)81&0seWMNXH3Wxj7rb%X`hyHFE_v1|oquDm#
zOJMuad`vk@EWR*rsH(F3)%?au96YL7bO&wHTPI=4d|WfA_821F@}95h=|F9Y7+mvt
z;zWA9@D0WaD&djRp7E2-=9`_dac8vyW6>4nt0yK)+-DR*&gA&YyPyilq?%O3;P&FK
z+*W1lDZV>5a)wI7)=yeJSdSOx{#^~?lk;RN<>hB9D60G1l*d2iAQw90MNi7^InT)7
zKU39dBL<(x<(eCg7rxA}bJ}sZwC6-U1=U!^m9Zq@q_)o3)69{%_%reZXDV;%I4=p{
z=NmPMz-5HKmuXQTI$v)n3xjT`mz;>qleEsusNsBXlCx)1`YSE^T6@L${<~(zf6d4=
zKLym)?ZRbips|iXV=JTgDnKg%q_N6Ct1?WIEsXylN%GIu^#>Fkm|7XkrN{EoMd$<X
z2{YC6!~xNH2B=bun+w`P=ZW9bGo*c1T5{Tj7I!QdY1S@vKYydJ>Y44OFmva&7}N^J
zhzMEDZp?8n{_IY)azvfMcd|^M$onjLDn`=rDwLMe!I?hcwCV;bYPj~?`e$D*jP<yJ
z3$~X%>3X5Tu6h!@KJ>7r@?8-<RWOKpV9L1dCu3?eDZqOQ;$g|tCk||q(0-~<0hqkJ
z3O+BxdI}K+sv9B%^|H`=p4NcythP0Rvo?g`lwd5<Wx0x%kB<V>xGknF9mmJ&o{r6a
zP8JMAFD}N--tLg2aJWPUFKD-hY56y`iS$3QKX((7PzU96<;J^mbfG~}EcX=G8|9(n
z;mw{wPjg+QPKkv!nHXG+%q#|`ZG|?6S+AnT?wx9VfC~vvKvBQbRb^Ask#hO49s?7C
z-pG*vV_7=iaN`zwhB#m*MH;-+Tn_VO<L*2^KQgX~c+$Cf)g9_qh~Wh7dE?0BO)C)Z
zeqPswp6PClyPf5?KB%$4nJicL+u7tKl3j|A+XR80o2r(O+OZ%u7Y0&_oE;qn>X}J8
zP+Mr(ob`>K0pF#iyZbjykshhX&i?iQt9@fLjt?!+E&Y+QjT0n!Z_Az0HtrrsTwf8)
zBD+wf4z$EjRQ8`j{lC1KR|{Z&dt|0Pq)(}^gHr-=DydJ3k2|H(j0ptUL#A8>&%oWW
z<7etsZ#^66(<LS*ek2=HI014_58J%QDe=ti-Fbb?sxh-LRD?45C=d>%-ZFJ!xnXs6
zq;Lz-j{dCShYR8}aY*I$=3|Np#{7?^E&}l}f>eLE_2~~=7|_?F(9ttXJY2Lnz_pf%
z?8I-2-0W*-Id%t00#rimF*D}`B}9rh{H6-7^U6H;(CaDkq0h#~#(F)2+d*8n71JPr
zoulrPi#qOpW@b1fN9+26s1t~X1lo?D;i+5OU-Jyd)T-TC1@)5o>i2};sE5cB61r=d
z(btasd?@JohM$`QSNMUVnIh@p@8TQXb@C2_1#@Z!Mn>-~yg;-_QoX5Fts$s$lzut}
zs3XKddg*1;gB}w=4OncaqZRSnqIhJ{%TGy#cwo0Rb?Ah&FSxtAKOE*dekzy%Tq^bR
z7m$9FxnleIPIdgy@);&i4nf6azW04jyHkgy4^*!BAIoe3P!EF&Ter~dw-<r&-=iv-
zeV2NfUGBzSo1P>Yh@#)v8*Mo+mrN|B`{rmB3t4@A$*A<s;v~iY3=}^Ns2~kEMr%>t
z7|<z-xQ__rL6LzJiPUcX{Pwc<_8FBP+K61^w*Fi%51+nq-Gj9#zmSlS^toeScTxrF
z8VHGVri416q24aeoC@*6>d}L(?T72Vv19B~rO6iL5)uxzvbt;KDm|Bvp)E&%+KhxK
z0HcI8J1j{6M@`4c_>C}Jx8uNOXycyi&X~L)xav+8+^g7v+-K#C8dSAVk&iky&U8Um
z__G5EAEZ}AD)y&%<k8D6RqNPZ@zfZDo)kUbGmNert7rf;z5`WT^YsG9-i3S)DxqEq
zdM63I03jkH=K$N?)1!>cIoyJe5`er%di2aAN_g3vkDAwl(?I3*jjhg{m}|#GZYbOW
z81SQQ@B{hmf)?$y^2N)7m~yBWjP!72F2+sa@@ig_kU!jWLhA)w;&7kcUzW$q;Bm~c
zH`50|yTKpaeJBOEf^SWRo|g)g^Y5rSSg7_s>ev)qZ)ddwEie1nQ6Ijbpdba0VNfKm
z!=Scj-X<p-YQ`LU{&D~-xP@1DmG!_aRMdcKeC-*@6;yYh3T9wiT5?U;V4>fdWVK|q
zvC9|qDLQCs#K*@^Snn<ypoau|zR>?GYyIzDQ4&B5`HX25pyo+|!V@yvj`<#50sVag
z1B)%7SJHOG$%hv7BYcY4{X$W_UCHFv<#QnWVi2`IZCU3_Ey;bo<K{7l$C07jp#<&D
zW-RGapgMQ!dn+doX_U^^>lm+*Vz{2`@G$DU_fFlRGQZx#@s*W(I~%EoYOaSEsn_B3
z*`?F-wtXsUY$-R6>-UnL<1iQ+`Zoo9z3v4O;Lmt|{{D4Ml3Ooq*E@>eJ8yy>S&|%&
zk&N2QQQGvPqN3^3ObiSpF5m22cA9+lZ$YJ<)TxD~rL#U;rarXZ%iG>AKsoMEJ)dc>
zIPGZbS`lPO!XZr#akOr6#El6RC*h<rmI>oG?m{OXLNA-WtJy!@D=4blsPuPAn`Bv$
zo^bk+i(^I5tJC4h_>id05^BD!rq5#!&#oRQR*)J&{h)|(0)cVe2XQR*rN0M5K4p!C
zY#?AF#;#IJoAbHl=_5AW918O^DcD<Y9NaUg6p0PfhQ5sb<U#{lfOhB>_>yTtbsn4S
z;-LVj%ZB{!116x7g*7hKn5VXNBtCSf&zw<a=oDWMTIo&ICF{@86J9N}($M<~8k;U$
z_<Ekv!^&ajU>i;%=CUNiWePgv#_E{7^;}z(uicY#$!!LZcFxUl6S}=ZVDTT(EYNWK
zJ5*-%x3?KfuB_3#gX7|#Q>i&6e1|~;3oSy~lQTA7f+wMrOf(*?UCzBHetd~HLBktN
z7lMq1W^-ZisPoi7eY|f&P|qBS1Bwzy{d*@6jribcf|4J$LIxInS4$7{3}x1U_K!M`
zQOqZ(A=+;{FC&9f+peLlmg!*3@)q$Iea>Wn_r~;2UZ93uV+ZJK908CLM-GgIv5Y$D
zZN*Gz0wwcHy_2NhCS4W_;=luJ61sj*(Y?70DpIh#pLh8<k;NVKHe3gH$uV6r4{Zv{
zhgxCo+-C$Ucuo;S7ORbqLZN6%1^ljhWfxx#P!rxOV`AW~_Sbj?90fcl=tc<-*LaEr
zxRmM~5;2*;6b}UfhrTu=8l#s8ZgX?j4SN1HF7SK_(u%_nc~AU#qgB^eY`~=oZUJ#-
zDXD=DG~A)v;CJ-(vzj2(LO7o{tFOES*twHqVIX*lK4?X3SbU)sf#f;MlIa5;{2o=V
zZrtKH2l#j^GpOefCufERp9g=>OOOG22h+m;<}!*Mn~sGOjABq(24-d)4zZ$QVs&`b
z`zEga$OQ5P?4oRKnUf`dwOZpP&@-bn`MFOR+*vOb_?Dv3N02-;lE_I%z-FdN{@+G{
zu+zb^2Z-kHLFB>_Ncufp3v|nTL_oz~fz^?Gx&b<KD<Cd6@Tv!T%w8TK0<0g@O%A$8
zGpT3ndJCXY_QB*JZo44;Asp)Y;|ngT0*zdv2oO>G2Ga@qA`tPEdScG?%cl$mvPYmk
zRxYg~fsM`0U>|Oh4Am@RYTccJT#$eMa%0Xm1_<2z))z>|+CVJ=#ERVSf}0XR+!c2O
z+%YEK5fXPIxb-2pT)}g2@eYcW8hPaVzT4n!o`gT{eV5)Niw&0a4cJ1s$Npx*34}c)
zlcm)G4<KPqAmIwboEaeaT-`-_W>6J!-R=dK0V$jF>bc(wR>D)YmAX-LRA4=h!{HFq
z1dc*ER|SvTG4#{btU;S;B%9<jfOL*aeSomopd#>ZRw{qa-R(ldloBUW`i>>BUjm0w
z0z)yoRFr-k@&EcVNH2gAciHt=9B(~;eqj<c)JlL{OXbgbR7nVAXG+Wk|8~E}`K24$
zogVW%>*}BKsLY8FLw5L9aNHd`z_x(2%%G`O^7X2Jw0-lxQ|p(C;8E9Z*l2>*t>cL7
zr!5IULx$KMXZ$&P2AfmCqwdlR104D@?k)z}Gf0`I^5^UsoP_oadWOWR9yj&$W5ML1
zjdU-*rXRPF<A<dFg1P@uoBx8j{|-O?ZkT_;-2dul|AM*y)zALFV6NuSbXnp16bE7b
zs@m&q-LtRBC^~J4><?C_$(}Vl-NYq5dy@<g*RfB+QRj~KO&2_!n^YVA?~<O3zX&+P
z$U%51;Q1wg{N}gx7sq6nU#?%?O%0Q0Tc4hH>lx|c-}FlC!c;`3&(EjBoz$FMD%1Y3
zP4V9<hEmcKJX*r~qCaVKvM?T3cXup0rT5PvR0td@GY+!h?0<Xfe>eL7%q=5O=W=1U
z;YG0z{~TY5N2s3k7G_oc=lDu@<r*B4%0U~MKZgRH<KR{LUbXm>0nOe=oadCS<a+v=
zr;C|X0}47|c{uk+;~Oq0S*4+m5+M0x&F>CfNMGKOZGH;eRs*5YCOt<>6O`n_--v|S
zV^QXZBXe7oms1MB;y#zHLB1;)(y|5-OzvIgjNQlEgqFWy!9Z!)GCu3}2}4n??1Spt
zuvn-umFVbF4$OuyYTB-A;1k_%VP6hihJ0Mgy!=&iZy<ihYXpVdng@EkWg~x15L0=X
zx2QNE1G?4!)s_5%Rw2fLUMp3UN*uR%{o`IF*z^&8jF!9&KwSSLL-<FQ96$~9dUbm1
zxGB9K_p<Q@AcA}5r*lz%!mWYigf7tQhS*`K>FM}u(QOD2VK(oyvi6_xZLlQQlSBgN
z;5MOu!mR<#4<!JI&_pX|_veK5e<|JnsxW^k-T#U!e<9s}3<YNOa;JXH0{CZf|D}Nc
zERp}z3;(5n|FHu8vxxsvx__3)UrP6{?DsEz_phW|=}FCq%P1YFU=w`})A?7K{hwFB
z_P2KGs5HTFUziU8@rIgHuip=*+^nU4u)jM721!sbdMxIx)Ke;g=_@l?dt%OuUvod5
z?y23Fqs!bYAM>O0({UcW62v4G?(%9<5maz);YEVET8y46ZzHNn{ZbQkbJw=#RD8l#
z`n<lMHfRib&!auU6hbGQ-Mu!bF@w`s)=4r22&WH0eJ@Yrqk#ianLa<5DB=uL-1){J
z8p;&;SSmL8m^U^oY46Co52Zi1U!{qz2{_ML3-|Hx+F6ULE?RTjvgX7bY}S2<()|+F
zm0>ua{5Z{$YcnWw_<;`21A^J6cNR7gU<+EvlihJq4lgRB%mw6W6$Tt!QL5)@dO>N~
z3``gJn7fm4rh_~R#_X>Z9Y~q#4^qzl4&Hw)$uzf3eQ!d2KKj$e2YXx~c-muo2s`5J
z-8zWL#>qxHIN^Yv&+QT*jJkLi)cM!B0#XxAp`=|NC}f?k2fAv_C6w1;+QeKa3=;1L
z3iUEa^PM^-8{z?751o>UkZM1VrClxc3Rk;yJ~pNUO&{!L-B~d!YTf;DY^^s2O_`z)
zA7e3GL80R^vI+LF`IN1lG7nSr;;^pV>?c~t5;pH!u78nn$hI6tnn^-T>}P~{fk4=0
z9AtY-D4iE_w#q-z5b$!J<;(o&d4s`x<BN7$+b%7yFB^P**KoL9ZZ<V~Cxf1vOZ@^B
zmwN7+%UF8Dwb#{P?2}YBt=Jmhh3K3-oA=MRamOjTdHRc7<gkzBoU*LVFM=0JzW*Lr
z>r1VhU-YG=5FLIcObwe|%9bEtUa2zhsiknG{zc33GhTz*XAw-(4ApDPJ3tWzT9&)J
zzo(P#o+Tmi{q*URO7*Ez895PP>;!UcsKh>4yHt`<*CV*X=)Qerh0&Ct-CDUTdUUev
zhFkUCklUt{!;}+66$h7g35QiVx_>NOSwT}P(S#?+sZez9P?Pc^?%o~dIq95|@z#~i
z96j2DgKf-!swJv|T~(q>ola?-`ugUByk5ab_sE@shXOze&UJ@6;hQ<&;JWM9c;@Yn
zo1Doz47NyN34y`!UHx>26|Rq*s(&;v6iIc9SrVZ>6BO)J*i?2<-Txat!ke5|@0(WQ
zSilFVkPiviFQEm>q1QQ8x$bQfv^m>HME)?RSvSPw?IJZj>;mWp)FW{j^-^TpbQ>eu
z#&4oupUz$H6f=39C<j7>m#IhNf2vl4;r?L6k4pv>S@UDPiN;_O(0b#b#@(L_dQCF5
zE8gxj&?-povVhjiGNHiv$;hT*)K>pWWO2on;{5?5t4uAHF&;z--bP@=R_Kw6M^J4A
zz`o%m?oS^2E&Ciwdp*q^wkh{>=#t=6@d|P}&xZABsO7}LvVmcF4t2&+SZ2@daw+2V
z>5hyv-8H%C^bGHZl)5QT-t2u;pQn|TXmez|LUa;mkZ`sse>0r!Bu*u}epQKwwF&sK
zwI#=V_heT$XBK0R1PmxkK`t0)STAa7m6<`0**gsIRm0trukp0QQgJVsmOP#dHB2+b
zCi8@JBSg~EB7d+3m>5p=!6v+JM$*N*PdKO$>6KhijYV%>+#PY~8mU`ts=mBfGI2Ke
zH6he}F@U>NGLbX3zs^HEptFr>Z_BkBT(iY)`lxSuOh+f4ZDPbsucQbwsLl$DC?r(q
zPH(n~)|vOL>6O%~M>%D*gu<2yXQ|bSb&j+fJ;JZtLSV!LY3?FF`W1b!QwE*P;V8?Q
z%*PaF%HKQ1mI+7q`d#Y4s<iQni+{6Sc>8vhWTKUfT$$;-tPXe_nocHT(3HY-600tm
zZ4bal7(lH{bv*QgB>V|X4kzJljt%<F(#gl^mU1)sWfb>V$QI)A@@zIgzdclRh`e_@
zX&Zb*7>?xg_T!8^5Q%gblET4A_2ZCWM;k4EWJn$o5tH%|fjKzTwZhfA@hQV{(R??o
zkRAO7U{)JJtWB96XjGK0eM%}>u;!xrM#4kkp6af})_VCURub&XVyY!zr>6@M8Smyb
zX@6tV(f~46No0uk%X_VARpeDW4S~1vU*1AVwfLWwJly5+ot(zpW^(uWYTwqej+#o}
zXID(nZuS-(Z0Swg<?hzcTzGG6C7UD#)Nb|LsSBs}*6CF_aln)hbfWdpE<3HEj<})D
z(E3eLoXTr2wFj_ZP~X`oU3ZO=sYt2R^v<2tDea_-DVDuQ3Z@+IN<q%t9fq@)mtZC2
z%LZd+Il10IbJOHy@a{M<neojhFbkk#6K;M>-t3#^7cU~J&OdkR{Hjjfr9M#1o)bWR
z;Uc(5G|y6Hmu-W+1f#5kbX?@;szTW90{NWrBPC1H{QgRskt8*4Ezj?;Th!)T$%>ya
z{kgMJ`!91>8|hTw6`kc8NJc_1Vr-x`)Mp<V!DOpjq#(E1Hd)sNq`sT6wf%{-iib5&
zat9=<Va2$>TduCW3XNj~JMB|y18u=F_#1|}dAK%-p^DNIJ3KzetWNk7<`|ZX#R*ws
zorUj5pwy!R?q9<BOfXA5QF4a&^lvT}8N3Ku1Oa91?x$}tzo!t3%zvy?P9grJPTiy)
zwfE%ZOBg8;wExx<w8EVQ{c=xNW|Kn;Rl?rB{Z#aY`jyBp5eAA#Ej~SgSi+xeDx0v$
z0WEg7fzjO<GLXO0PMz~=1@w7e!WF+AIzdy0iNDmzhGACn{(*0HK~7Urry^d$W^kqh
z7wEQXft8jrhJHxc46PTr-2^h6`gh_~)GU9buJnUzZj*NC?tc32L(=lhp)#>@8^vF-
z9?$+Pt7s@+*Ks6w5IOTq(Pz(rQ!S;u$Y#F`F>JjZWV=+FNEpT@uPOW$`l54}Y)8Ge
z{kN;OpI)3RF2g$1oa7;3I#@Yl&)w)t?R{C>_ESUYQgTl_PsK__?uhfSbBui7R%6-K
zvv7Peb!k}j*Td}-jZd7;y7CbG#<c<(bA`e9BvZ_-?+rXC9%!V2j1IM-&S9ui8tG{Q
zf{mmj7C|DsZ#Pv-EJZk=y&jtqh$EbViqx_puiES55t4mqrY$${@yJ<Fxxf2La}viF
z;2q&q5kIuQ4B(v+Nq)Y_bYR;6o6K)XiFo3-mk&*6mt6LCiB*Mh&;&=2w{V!3UCDpb
z0;lpa_{6qU-TAgnSFX*)EJoJDFCk)^y<v(SKEhi+sa&`Um|rpD=W{T#?R&$4G5gAR
ziQR;%ck+JA7}zXKu#hu~Dvnl65Wm67bisA}1$!+R7pD=nU3RiX#@PqEpfh%vXHY7L
zLCkrCBM0Q}US`yA7w_NLU#PB<5<hUop2;m;s!*8T{^kI`Ku?d~<-KL&^cYD{hAtXb
z^UA##`)#ddYNd{pKsi;Rh0DQE$0VHZQg;;2Uc<Zzy4d4{A<PNmF3oI$5L?=@ELkdB
z@JP_NwdDo->!oXf_g5ZhP*;=hgB$8Np(|LrrEppv*Mk0R<FSGsNt$bw8y;hDpK>17
zLr?yp#9(0*N5zs;4h?!b+J?GE(1Ob;_(MW;0zMxLEa?%j`iFx4wV>lF7m%>uKPfMD
zdfLfww17=Kzt!+>STuTj0J~2u^QvSaD~Dw|G&Yvavs90QpF?iaNj?BLx0`<HEVif4
zNR5(6z{vCp`w!mcfTJj}>9`Z+vm3gyT|KI%yk+$EnUS5J(JSB22SGIaY!xfM9@+b<
z$T8a<`C_4t=}%`&eF6XV_1YTDFc*CVz@u5_J4F1zV9xsnW3|2-#?kjBefB5h4g47r
zXo`j`0_E1|qVMyytN^I>HnL4au`|7SBa~I?i=b7hQ6IQ>7fgw@#N7zAWUWna__8<0
zSaU(iTnmgO#^<E=s}S3t79Q9XZy=YZ|FVSy!1-ZNVmV#i^$yrQ8A#1@79x&2`!m(&
zKuhLg1d$x-!gQC<)v$LqMT040f;n!+Y;I1k@Bh+WXZvr%kx;wYY*XuqqzvdztO91e
zzKLdaS#OA+9Rz=K2D+Q`d~=xJaa$~+2u$Nf^x@=n($FnjHHB3NJ4M8}-AR&yq&o07
zseW!V(j(G)5MP-_f0QJKb>VwXvZVT!FIT17xP^+&2OAt>x3lO5Y=A~HQEs6;x*}t}
z4lz+V`$BE@K4vAxmqB-Ck}2+MpB=q?qx)J@Bj%|LwI|P1u+lq!rK^3x&PXS@g4oX7
z_kxRzJu|Gm=wD&ecjkl8H`}{fZXt4Uo^+GaQhTc^qr^A(t_5!2zI|K6D9n-(T_lUF
zy|557_nZaBOy13-8uHY;9N_Asp?**rymhY%FL}TR3M)V#Gs{r6UtadDexR`*OmxX-
zbr42l?Evl7K>O;!as&|vJ+0)%+fYBt>{92cchzllsuunXMWoA~GX!H`^ru#x01Hd&
z7wSgGU|PuQ&on!<f@Z!wX_!>Mlyw>#=nCWkHt&UeSH!EQ9bSM128ka&Rr&zg-_0Y`
zV@^NgothVX0Vj{}CplW34&<cLCFckPN-rb3p~dtAD>CWRbh9{lbC>8Kn+8aMaK(+`
z_Je|oF#!GT=O#uEv3j)+q${}rgHJNAGEKxD)~+XP?<Tz2G1DXm3LCEb{jI-AF_+;Q
z?7)}j8_OcHOPBL-i3a$6lzprb!WXbX32mo7!Lb0&Y~oNU$>J#?7D`sHDbSHV23YHi
z?<$CCf|h2zV*JsDjX@flCyU-6^i><GRVt2$xwA>Nj0#q_V8A4coPDx}{6maR)tVPz
z>WR~!8+ePV(=0qSFZ)%HEe>w~`EH)<{#gi3YRaTb2XOIOaon)Tj+z?SJ@2MqI=a{J
z;n@xLQYxE`&`7w)@Mi-M<yd-9H<tAoVTJk&y_ml~0W<M!N~gNU7WAqe`+lzs-m7#9
zjN3qBma9k8@9?_}8mAUM7~N?sR+jLfk*dNFss3$+hA2CUzzQjV6>^bJzXew43Rt1*
zZ+6EUg*qPZ!yoZ8*nw`$?=*Y!MRUavn74X#UZFwQ{8%C_uinjJ^??qDmMQ|Hi{O)*
zd82?4XEy=LxA}Q~qzn{w)Ya;KI7YG;HJKWuzF?X{g<ql9j994NF5<h@p@CC!u#om<
zu!(7cn?fCyjhs}Rm(vq%8Rb`8F(ohG^8(tnnTq2uE&fd)5W77p9A{gbXVAti43Duf
zGFt1p3PxtGr;ip_Z8qDY-1G<`22e;u8zeV6i+xJkNhpOG<WcqZnB=6pOrzpjdEoWs
z{rxe_=wahf!-dzPldDL_xi2sG*_cFS57?M$?&-1<Jp@|A%bJnI7AVF#`fPA6G%s8E
z<}l@R;8aV>vM>-hk#+56q~H#b-=jL4vCS_aaAE7~1l|};*H<B{vN<+`p@Pl2-ZbUf
z`WKXHShiJ5DsHVxvbvH8B0kpYQ_)<_y!moS(=LtOv592MC8nYg=hz3MbQcy>Ik$V%
zh1W_}6JR`qadwS^V+kGy{rR)^eLTM-a)|vI_x4n<8_gvBdOp~2fZ*jPrkm{x+?js4
zi>wgqRZ!E-e&U_!GOW--K-R1P7*4|%@b6ui+JSI7rNl7BY{C=M*$gyh4x%|r@rqk=
zNsB+gO_eDcE6^W-)mE0oF$d_T#<>aj`;mLjlzdo8cLfwTUtK~S71<H^xC&C(pa`bl
zbsiO`#99}byM7Mb%BCAI|8A+f9sp+IQ2VhbxaOOy>g~1;)X|}AmEQ0f+xIwL0w&x2
z0+tN_En3K0bLVT5Bz?kW=3!3QqMCKAXF1vfwVu5K@)!bf_N`5~GJMu(O6z2KDy@2%
zYH6bfVzyh(wBJ3DOO&*psukb$Q8EDjmbJP?YG)E_nAxVKegMHu^@Us2BM{@B*?&Zd
zqd`baI&iH!cNa|YV3%C|oV(etWn7vOvQ{>qZfQYfdXLX@pj<({P^sRGPq=?`R9!e`
zqDBa}mF=ac*@r<K5;_|Z!@hUKXBal-dEM6BNdgnMhO@tTian}}HzFhdC3_nBgM5JB
z?}4j+Efn;aoLT83KrO=J-SL1ssX4%sQhzUESE>NA$8HmkU=>zpUP(~Qp8r8GvEaI`
zH&WwpV+h{xWY6v#b_Co3oZXvkqAEG8e?HN=X-iS17uq`teC<6J;6+Es4mMJ?XxY55
z3#9E+bj)E4Y@^q9zeFvK+KyUY0z-DrdGAkf_E&5|#olNlNTbDfC--AKg}>@+f{~qy
zCix<DWzoxJ3!P&=huf3}dmAMEBsRNyOXbTXz;h>RrdYDytR!W%%Dw3^Wnt<wa?dhz
z!LQ7VVpiuQ_Ax%zkyM{}cEsU93&A8ei21aUffFHOWsd^+#W#C<#>NAyRh2s{w}vDZ
z^O}e(YI{y!SlF}7lpS9BLe5cM!xn4xu=qK^SoxjrXAUJi*JG*ug)8TBQv&%}X@cmY
zY|6@Y_1;+N)hqsT+EjMlU%~9~5ps~7Xd2N48yKR1c;~l(BSC*|5j&*ody!yG;3%Dh
za$vzecD8=tSUu~_US1SNiSRym6bPACl`wUgI?a25xG6$t<1qGopn~psx8*S);>h_N
z>mF>Aj8C72^-*aO){8@}<9NSrFBGCb<LA!^1a(~+pW^Y+NVlq5k6#|ZHa0(^lLg}v
z5mW%0+S6jkC&La|pcIVLgeb|k$AKMn8m3<2AOQW4l?k{WNC_3oh&x-Mt@ZPH??_>R
zIMM|Bd|?NGOdga0e>~~a`Slb+_q=+Mm4^K1VnCT<a&1m)d#<Y0?eF#xwOnQr_nr|`
zT>PlsSav|WAS!ynN-pU)oLFnsN?4suIht1TU^Ub)@R}%3#ahJS1!#v#gpgZG{Iu!*
z&Rc;N`}>r?G_^F2j1NTM=O9uk@bnfiM`Iw8J`w=mL&-fZ045-`<4^^}^1vOUm0#DJ
zQ1}}$KX@g$wmZ5)Sr;c5AVf~&T=B1Pk}kd)WK0SZ+tcdZezw4FMAtAK*cw6st#D=^
zNE2LGOK==9j*71ZK?N$8*zyvir>*Ke)${An&H!;-lP>QPwqXY6uab+UGqcqZMQnAV
zC%ar5iy;01i+|I2^C$-bqAQy?=BIBeFbBi|Z$ya<bu6wD5RjQ&O476Xoj>5?yQPys
zCzH+SW67T1OF~Hvx5t%gnB$ssxN~?eXpLrF>AHEQk`1H<R0YT4fnJK9&qP&y%*%Gq
zL6Of|BS4VG$9hP%^E<V;Fc|z}dXyA`_`yZ=??jdsi9%kLaDUA(=>#9cvb(__vh#ZR
zyVh@^aPb}0fEU&A@c{cH#Q~wr7_LRWvj|lfbyMw4OF|Pk>cOhtSks;B<0%_$DH@r)
zo5Y0IPNSyTr>>UQJg)M<P>0Gm;tV1m{|%4s2nBY7Ci{mU)z}NjkE+8VF@|59`Z4gH
zhEUtSML2f4XdC4r7~FN04=uA=z1Nq69Qxo1v{&h}68zIswaWP^74tQB<bZCu|EaUC
z_vaVYGS$hUK)iQOI?Q%$vL$&iGko~Mg3KTn;hm%zz`h5CHOQf$adB8a1b6SgzN>3V
zx@Z`C4t0uL^uzQ09m0vb?K8J4`9N47>u@D9@0v&drN**FS#D{Z5z4RdYOmV!YMwlC
zPW|k)$i&Lwk3_GX5q=O+a%YOSy6{W?;_0`Gfc||0`e$Dp`UN0m957`I%dUWQiG0@?
zjR#Z#Ys}pe0A;qe?ra1ux#YuzZ&1t%Biz^Xj!1aER9WDKg?yeudICqq8qZY=jQ1X~
z8rQ%?UN6$4@-`d|K)bR}SLVkYn7iD(E=%8DX7=_ivhQ}=l1rIOz3~L8FphU3TOe!l
z3}P1ADhqZKjeb#8j)&>F;-eLq5jZK@)4{OPhpaW<v&Fw>Z>&jufHdplS=`o(NqX&7
z3~N=uW{pNo>{+|vI4p;QCe!U5GRh6mkrHDumf1kbv{3d&Z$MHbW`zqU$Vr1kb>`K!
zx>}|1GX91?>{TFJ!8$z~=g6Dv2Zm5~zI~qlCFsbSqU;Wh$dM0!4)PR`D=6B|{5ZrC
z=#p^W_$>!1ka9PV2|UpS!ZwUh7z@BQUQ=yz6+@kdOBrfzp-vhR%96OPtol_V&%Aek
z$wUtO+hE&cgkOH~%eA8>EJP4)cGcccuIKKWUn$6tKLhb@k%4C<d5b@@5bzIeTww&O
zn5FV+V^IR2P~F)z&3L3>`;zI@)~8f=W7)l$sLI8L&3E&D!Cdf>-xzSSi*<dtpp4gd
z=G-eMaZ~`PT@8^#O?m+*(fmHz=vGn=*hLtn(wD<UB*5Bb(RpJAsbVaPVx1D(`99G9
z%=El|W}8g@Yekfyj<F&*CMP$$=FN4(ey`z`&F*gfx{``ZkAE3kj>D}gU~GE8*swxM
zIAF673$Oh_AC#)j3{~g~$ft^6LOccmTo+1g+50nAVau9L8VAdv-&yh{y|FCdj%v;t
z<3j$`R8RbkSfF^0+%YacS&p2`qx%u~>PesMexx5Rk?#P=?<+_dwn94HqcZ21HwE!*
zw1pZJ+NOQE55d1P?p<$|)+)@>Nau9)d#^gWvvToWnHostR0jE4Zw-<`KqvZ(!En7z
zU|^e%L`mD$0e|dv!!#pc)kNW&!%E&OPnn=FvDohr3ZgF1w=dV#H<r<^$>FTgFYsMS
zlEeRvi&sG4kt_Na;viST8qNYAC<l{Ly->&i5`(88#Gz|}5GNDmrWzjc-V^^E4^Y)6
z@h(URlGbc$Ae{{f;T<G|$Q<z9`CO?L`1Kr+wuobX()J*)db0&9gx}}Z!H^@qvGKB|
z!cFm+6n1Vp<ub|}V=55bC2HTqP^=g+HG1f+m+lA>ldvQP#H-QgYwxHlbK^Vs@0SBI
z@NBL}_+G@o;-JIB4|LY=ywxFD4D}Z?{DUkOj$|<mWI>G67RjBF4(7wgs#Y#H#-=8?
zi^lmJ+z~8CISs_u9blHVC=-mI<Xw~PkkNGmF}Q^5ohLQ-5Cx;p`Mf>zhy?2Y9(A_=
z->B1w1#)el|D9}dj%0(r3}o|n1p%_51+q!5_@N*h+%amu6vXGZ`_ccXAY@0!%7CrP
zvHb<Jc(h-#tDFz9qypT<_W)BX;LLSh9S7)BQ5XwqQ;`Ne^1>RJs%GkPbdhEHh{WEk
zp~2jj4{2XQ*gn6%PbrPmNGrPDACZ3JoFA-<XqI~2ytV>zWxEAY+5{W1Fq&$Tsam3H
ziq|dWucV-$o8>8=4?MFi&QZk=NMtu$H5(M6Kq8B)p!T{>a^u4R^?8li7mrOrHt}i_
z6KD8vCurL%1v)=;dM)!~KWiWvFc+opGf+;Hdfl!oL(j1~Owpqy0wzWHhCjO?Jumz5
ztEr5N#xkxp4jvGfzu`H1F-P)h<ie`@JdGvO^&xkyB!dF`q91-qf^{XT`Wq_0K+Ebn
z(l~~Taxu4VD=i-hyM2}I1=v}1TfZ~TKjYzY>CBSKa-b&?6navIMnLgW^IAdJ)#TGJ
z8bg^znNwVT4|b~TQI(NNq6Q|jTkud9@=2aUve}?alC(9fB(z|0ez|-ZVLSIBsxRCC
zq)7?PsqHfcOM*`r>d>iMvvyc!S|;Parg*Nvj23=JOhR(r;&1H7H-W?X?i1Ks;C*hD
zqJvH#UK0VFwPx-Lu#nHkU(Mtoy5mMR8}h{aI@5N=-koi+AJR{La$`a!gLdI6l}1&o
zWoBD`5STv%2S(RZY^Lk==*ao-W63WYU(!dMZi1!=fmT}BcOIWoq%pTJ$kDV3Xs~^?
zLdw^H#GsbR4Uet5Qg2JoHk^APx*a8GPmoAWR{(X--_y76R7GDmbWb0-oZ&pZ3-hdf
z<_!=@J!)S6HSh``71;1d&YHaySPoWUSl-|i&I7_sj5UX0QwT|{Rlc1XgCusY%kS0U
z2?3jLU%?47OltQ^gPeAGX9WAFlr>8!p35f9!hd@)M=Kav*7&fi#Qk*PWwNLz-dx#R
zs#;+F4<wGw)&F1A!xkJ0Q^17|qG`C+{qo#wbl0~k<5H+T4!GIJAHL?Ls(v815Gi~7
z2CA?c<YSRvY+~HXt|+IxZ*^Fk+<f>!Hq5LcVCa$CwF9N{V$WOP$x2!y0D9~-Jl47q
z3gu+;tj*xK?ka%23=X}6SFGMO(AU>TW8Ud$L-pda1#dm%@F`MHm9G{V+!nQBB)t7j
zSHGgR+=86Y5iqd*aGn6v?;r7q6zBfUIiugdynYf#hZ+PQl?3Mi>Qg|@*_wF0p^#0(
z*YyY7PO^WearZ9TwC_v?cBHJRc>w~t$MO6^{e5DzYea1zK+P$zXCn&*7T9MUFZ~l#
zf>PC*Q&L7Sr1e&<Jk*fl066u9<(GWg-&GWpkMMMu1nmU^@?p|O`8ZlJiI`;{UaQJY
zRrA*}5N(dc;WxFRdOgaw|6a$}b#G%X`n=0z@dg`>;v3$J$o=wHuky0#=C|&HH1I}X
zLy>VmWKEBUy8xxnui;ND)p4Cqt5j1lh_x=;4!XFj8tKrvt4%TmQWv$?n<`4#!~=m<
zQ3cGcH@j!bR-0Y{M4Jt?px283f|$WYdJ2$4g@b&mdpR~7?Brd<1IYs_KAmLCD%arO
zwtOTwTTzcfaPz5Fh$SpPPwyI!01j-v#0@g77C!>uzNVX~<$T74wQD*11ofBnt6ZIA
zO-H1E6sip8Nb@sO@o4jV!>Nx9i{mXgKtcR#{v$doUT)Svfpb`X!@Fg7E<&e8ybCme
zwrg|z_PvFDCUr=W@A}I=MXW!10@+%{xqD%utkxzKckeE&THQ+71mY(gMcBn)7!MJ>
z`|QH)wPAE3$To2GdUV?DjQLL&xu)c0e^u>r&*UuH*t(nKg9idDqKv=SL7ZQIB?om7
z@a%-_#kaxYy{F>Ui;GD&zh$VS%m^}#1!OD4m3T^xpEaaz3!gUn2FqvZp81h)36RJ|
zGA;ygS5}hG>VXg}V_0b6C%6t)OTfGZIS<0#?DpO1JW`1#Fa`uD9Q6`pogK2Rqv<<|
zS1F4LUNOIBLUc8M1n3O2G6G~=_*A^b7IVGQ9nC4Yb;=6Z2D_EuJb;>Im_IxT?gp?|
zt5cj$-)oG5{0N`fIk06w=}pbIQinQlm`_9u_A-$vqxa55!7&XyK(SxTMJe_h>XcQ=
z+<5}yL|i%?WBXN2(~6fKCm@580qV;|@vkYFyAJ^xBFzmAb%a9fFOp?|Oh-eJXMb0B
z@X3av#7e+XN@2p&pj#lmSWE1NkwkwX5^fHZi3==tp?v>O3`MTeu%^7KfhFUQ{d{U4
zeDMvX1w7h*EqG=2BWIAOc!=H}^Jlsrp>PSvdm{$Wll6z^-I^E@qp#}3{3aX8l_nzu
z61lv*yxCW|(6P`s;obE_LTJoyPfAvzGqSyhW`U1A;v(WBWD-Iye>VxTS-J0E!oXwo
zT>YE!Y*0G_ulbkyn3YV3!rxD6V_(R-Hlq9C5-cgA=DWUK&-mYrU|yU;5p0nF1u^$x
zba>ne!~r!>1bI5Tg?zhkB7{01q&!7aNN+!z9~yT8(TP`f9rOS!RE!L@TwKV@j`xz4
zeF8It6u~-wDfvf*F}REQdN6JPMlt;EWwM{=8%<v2_1IdB+@0CI_{2mk5jYFvY@r<(
zeyNEW0@+;rjj$SU)L*5s6KIs7a2_<CiJbewre0uXr}9OBG%k``i|Sx+pu7&0oC@Z*
z5oZ{QTZ&6FHR3L|*qWK~T7kwLgNa~aO-y+MRjfaDYgi0QVpt?<RA28m=DsS0V+TsJ
zf!52J%Yd5vSXo|o1e7n|pOu!%YS9O&S<*9%JLw?sIR&I<QPjx^o^F%z15nZ6ni9KY
ztvEh;6id(F37g^I|0Qf2i2xz3hI-Ybf|IrJ>#qV_&Erno<aY$%jy46eF&x>tvu*%Q
zGe!3=^=pJWp_laB0dND?rHg*;^eboSvxRRBZOAYY9siNkVZwi}k!A_<zdC8=Lwy5O
z-1*>C0@`K;kSNsw#{|JGOGBwE&6at2Uyk_WQ(e<mFYH1fKBaFjRJpBu(ipSU!>H~I
zc6n=T$9fOQgK@lHcrE*HXLt{gNUgU!)S^6=X>?ZuPSrw9PWjgcNBw9Q`p<92K;@*3
zKy-jpq~rqRpG?sPAR!(=uAcY!j>tl0SMSImDDFhfZF*P6Us#w=o1~81RFBfTKPAU}
z4RPJDkxF&!mm}g@WX)p%vuS`MI<H{k4=m&!H(>9ADm~8{%TmNaZq@7}5EfBVQ1X$c
zHJ2ca05sd+jKl#iG}WK!#NX+jnNd!O08juv=_;u4H^W0yww=C+QEY_$+`A%kc?F!Y
zP{7p%sHs*;3(KdCxgKTXa{Zf|{x6iRSpn+d%3OQ7(Ph3jOI__cDrd>JW`OywkG39z
zj_d*zW4K>Nb{QKh<*E36E}b#~ZLiH7&<|C8(7OWCluTzOcbp6|vWjL|htsWlL}(Y%
zn=3BD^chZK$3-vTRkZJc$Uq5}JO~a3W+2k`-c`J1A1twE8l~e*c(_8!LrW94>tU!f
z{a&8G-96K?j2|lMJY@cbPO@&du7Fc-qd~2o&mf!t+$$gTy}cO$uY%q2{ta{b4(ej(
z@ZVvXpzVZuc<l(9&kcKT-F;Fsi&u67SS6iTd9X=aW>eYc@J~WeZs%`bUG6cxF^#g6
zdeIZRsq}MNK@y@SOtBMArc3~uy&g`<BXVB1bP-UpJp)6q1V=10c2ej9f1HV8Q#RKh
zd3o4%BDj`FcV$0vZs<O*{pVN&qSbl~ocB>@D(RrR^EouzjJ>VbJkfQ$WhH*uQ)++3
zUy5Maec;8|W{wo^W>E(P?T?CHQp6MQya=VD)gwI5w`M9OXMR~(1Bip~dMD=bNYLS2
zVn_~Mf>To7wIk0f(Jwd}s}6!@4zgeQ7-R9o@E0c#R=696gCeca+!GC8NJ4rQOV_7-
zlKcBsV!)O%Ag6_RV1e6WKTTc~VbcW)qIU;ul1HvC<k=KV<$}O&g!1_>V!0GASXn-k
zSE=QA`R5S@ngpz)npGeqQf+}MBbq5KoS#6IMAJmVeyTH)4lv;p2u8A!wN||%t|<fm
z%8SIrL6*_k@whCQUZ^uJRu3x3eN0A04v!NrMB0I#y)os)J;$BU!yTnJVAf9A1M6x~
z5|cMMg@bz+gf!Am&aDV#KsI+W*q9ouBG*w}!fid_V~c5`*A^jE@E06&!@aPurevL9
zs%ogSUilp4rh;&O;kriSI*9nln(xA-A|WsWB}@M&yN}l!78uo7634RWX248{UBV#o
z!+j?CD%AZY4T;+>)nPtbgJ5pnAikpqXu*k+7b^;r1z#?P35N>L_`i_P^hReAiqjYD
z%F5{ohmItG=mVf*sSiKIi~N7>eR(*Q>(~ED6iGA+g^r@gl(7uk$w7r^FprVUQ`+Vs
z4OBuSGM7TejN42JA#>(=j%^;caofN3P@VUCz3158^S=N6oIlRh6+O>$uY28Vt$VG{
z`Yda`m|G4<=~1rH2I*kiO<qYuu7iB$nMCsOkLX6N&eShRW^R}0gpz#SW7eA=?~o}4
z9iyfapa1B455{8<cKD(?n)N7@=%VbWTRXLxe8d+&PXu)?b4d*^o_~y<)w;Kb`H42<
zBs<t+U#KDFl4O(J3$Q-_0tB`R%wwJc&Ax|41cU8+tU*W+4@^BX54MDL^ol(IBL29)
zgw$YjA(NcuyC&phGG8D1A*7!A7^vUM_!GQWOZ7$oPx3lIeIypCB0jPdaRMq?7xz<N
zL5W%{F&8mkOyaPqhteT?YVZEb-mRDo`HHEBzecrO;#{-gTJ8(eb)3~a7MkgumD(5{
zDw4)H;dW+6rHN#N^0QN0^ng2y<~yxBN5Z@m<kB7LhM|1^@mlTM5!!0C58;kAK_!e%
zqM9}pY4YXI4EvC=KTr$|LvW;LV%iU4rHgKWg>8UJdBWXG)#Y)>1)NME00Jx?cIdTK
zzQ<Q3%63vm+o_#Urm~{nWZN1q^K#rs@Kl4hbbN|=GNk4??#Dx9<RS61c46D=oTPwQ
ze6ZTRX58?7Mk-L3llls&OG*NR(rMP5-xX6KUXt?id{V!9#xNn!kysrnp@sxZqiGrk
z-lsrH;<%S6M7GMmOp%w^l|@5K=$i{mhdLJ$FDbG-!JXw<7$~;DzjgTS=?ilJW3;os
z!9#l^gok~vonni~1v{N@*w+Q^D!-L)dJ6j%b(rsl98#P6qTe?EEvk|N!OfUHdbOCC
z#3k_Z9?;3sKz06hDX;MAN%HYN3f}9?fsEy;uzTS3b6LY)zkY4iW-ZfrZ;%escvmp4
z9ctd_SsPkppa|K|yj{xI!I#Mp1ipgeBp*rxiiphA5bDe-RnRVVMlZ^ED3UBt+IxiN
zPks@*J_{AH%5!Stkm3qjlR*&^9Xzzizl{+@d-azDM8t6V?XJuXk75!2C%AI|!904u
ztDqnw!TKnZ+_~u#`^jhgrEd{5+7OxRLb=X;m|dq)cQbD;cYIfGNfeeT=*Cqx<nW-+
zn%;#p_a0ok-8jsa9NR6}e7%=Pzj%hnWpU=oW4clll^7Jzo;U}UcCOg!WUMTr?lD-*
zO9lv-+L}wBkn}sLhJYgoJ7_hnECNIwa$jiN*Nu~vKL&ipR1juT>E7H~ISfwii9Smy
z!~{O1q0UpK6PZ<7ibB4HyPPDOj3TJ8#r^w@;>u;yt@B{D-gC?zz~Q>%m$^`@I?)fI
zrwzp1gB~{l`5@5%F*nvLkbBohf^<Df@ZO#;8c?>$xo<jMHH_4E)ir<)^My>f90CdI
z@%4&>|H~p>k+;}?+5Zs6r6dbWcMnw!?OX&;!Mwvgl|$Qyp)nUpXwkfKHqB;2EKgR!
z#6&V;Kn3HA%eB+pnq1}vh2-nb_kFtW6=<;nsXrb-Fm3l3L=jGN(CSCKl?Tj}ViQs5
z`f-S@_OEm^X{WMRL84R088(74f4-<b_JFlfR>Ya7DL2IS!SIXc9BvBndo@sBR8=~q
zfN5(8;-;ba<O{Fr#cuO5#~yeJ8FDpBXI#plCsJExZGC?w>K==)v%b82$Cl!()n%K+
z#a5^RNw5L!FHe*lrcow?c=j3n2u#W`P?RYez$t4&&KQu*uJ;}sd*TbHZy`glTwv8Y
z1)lH%)Kx2qK_Em;``!4j3pDK4xrrF8E5A(B$V`(jdMPPMS#worw;=^l`q?5dEJ;7x
z?mT2`10o+B8NK`H>Z{c(F&D-1e2L0ExVnGn?hxv*kC=ctz6l<aYJc)Op5rkndoj^X
z9kvfD>MYl)#B{2ytUYcRG+AVu7ru{j<#F!kpuGpk2aH;+hcyDsLy+7pJg(R8ts&7N
z(FnIXmAL7dRWp9ZlJYc|@!U128$=H8gC=Wa@=wSC(6>E9bgMVoVYs16BsM9*b?hvh
zu#{hh%5Nd(GVtS_fg%KYzz{)gT!$brBP|+&Ch5!Y%y<w}vtwi_=c<@$jbtTnO9+vc
z2|~ccv2q+NI7eO|bSnc~<EQ^@9!g@u;@X$^Z^ZQsuXVuh*avo5xW1#<o57!5i}B&M
z?VgtnM0l!mPerna|A|<~BTq#8R%-q8+IZ>#1~U($s`{|9c!eRd^mjctsshcTgQv6a
zdvUKg3wPb7ZT#%bstb{%s*?X)KB4ndwPKd5Ydanodym0z+9K>W0cZ0#({n6$esg~T
zX`&a3bgbp82k~0O7eIE&kv^ly1fw2hp$!P(^Y)Ozk74k)gMi7hyJZ>7u8GZov$rLh
z|FUEwgMLs);c?u8Z~8uJb{?XZI7oOz(F*7UcU-st@}q?-zfGa7&6m|PXxsk|oEo|N
zDFi_F$Y^2M1@uX2a2Ud+T;^Y=FxSfUWzu6{MbM8uzBcyg%Q;>9jAcsxw^~O|Ep#QX
zp26gT%t6;9qElZYsA4A)>_QI0^)3q3ND3MNqS<4`aQ3dky^rfYc`#H)@2Kjj04zos
z9N936AWQL-_>oY@;h_mH9nBCE27Rryr0i<Z^!b&m%1Bx$nEqTkDfwgdu3FiAhOxY*
zj;vNDw`Gk&7(dmXk%MRsW#d2?R4jv1Xds5<`t|h&D9j7r-`r~;4<c=8DEmWmgmXc2
zW|S|3E6W7U31LVNl2op$ZXZwO;1&aVJ@OaU5FV-gS*|k90_u21WgJ{L*<VNO{&ehP
zgH#DgIt{6{kqZ*iB&F=Z#;(u`^PE+U%ea;+hsPqAjQ*zaZ?AyR?m+=y`8}<g4Nzto
ziT*i#OrJh(&xgX6^6t{DcVN(cLWRY6?%d9(u$8AguFLaj*VOXCf17`T(d>>;s#Dic
za~pT^kGfzPCtHMEaDLeBU2AsRMzR!gTj`_&Umb)I^er0GY>{nv52>yG5>6FrW5)`x
zuT9R@f%>Hc>35f&c}S!~fNh9<C5VX&eVyHTo`VaNLyBTeJXf(Q<7^uhPr2VaNXeHz
zEAvG2wV)ZA<IcUb)z#JfQKT^=DaAMeJoOcoS6bn&R!?f@E-|X!l*b1}Y|W4;8Uqa#
zPM-d`W`v0(-;g~X9v;bUQtdf6D%(yvFZVJTo9ei&OzV6SoBKw<b))U{;xtylT`hoN
zOXu)YS~NqVCABn9C_{fa+clDcVgMw{RiW6u7lm1GnOe)~YqhhUVCD%*O1OQe1#Wh9
zl)rpianzj=2yUK2c<`t9>*J*4e{#UhmUOx(biY*VmT3DS1Wv5H*PsqkheOJM=>kqe
z(lTMNJf`2&9Xt*B8v3;^EC3^A?k{t8!%gt>(+D#99wWmvgcsiZ9N9A^?z&}!MOZ#r
zv-WMKs+MK$`OFcGNU%D`&vpS2hHzho(IIs#0rCD5D8h8-wVuxp(Y-0V<QUVL?7KJ$
zjiW;7FP`qddt{V7SM7y<qn6#SahLJz*0N!z&g>imM|~~X-=S!_21a<Uzjm$;F`RFJ
z8}~7*(X__<fSybr6GUo2O-|!t*y8=Wjp8`YX4*1=%TGH^36b^4FY~E}vyojOr#YB=
z+tqBwHXXS`Duc^ckqzXwaQ$<kGkQQ=kD~7w-1)VWE2b+}?ae@?9$0nd@KqSFp6^`&
zr*e}xjpt$c;KO}>2?tlHrkwB$LTk;@*ZfH6hy;lILUX0NXt}kY4Y{_Knl&f)b3Rvq
z<c|S8QU+F(c4u1dD(Q|w;T_uF$U|!dK-7$(hiwSmyTm)OKT((&0ad(@we5eYk|qZ=
znDa-W5OVAQgpU@%{D^%icGUwxU(~I6BXwkfZAltc!sI3w5q<m<1Uo^eD^wfpO?e?+
zQTi3%l{;QM0Xg3guq$VTULaKDXQ0JMGagU^&~!8|%At-Z;3VAj7%=ESq?o6Vf+y_b
zfI?>yCMh9FJ}@a^vH^K(?60N!%?7hOWkTy3?!e`>u>{rb1z_T0;{8t!4UkZktNZaT
z(4?#y+`s}DCQmOu`(B}u)3z2BX4~@KoKsm*D1wnvPFdd-tKoY=YZytpBCsd<ItFau
z)B}hU`oPf?Q@t2~c(1xiVJEPzP-xnBAG?MdIgPf1QBhUYRW4&G!^K5=jN(k&YK@Dt
ziZ#55+^KlfAzn_q8M?ndFpgp}&dB>-{%RANpv5>p{>@UznsAl7&e{}Hee^~HN9RV@
z{`UOy%jUWE)|s?M0ipn&cu#*BQmYqOU_M%5w<Ob|(LqMW1dYh&3cz7oY16YD-r%K^
zqtoPw<-}%vqcoejh7mCKMw=d8hO4s*#B8$~Ton+ph3ADR!nJ2_##~3fXQCUv)<Z-T
zkjwPNMf<MB#dfy9suwkOz`yzWm%a~7UuXs=0EO*_S7C6CVqRUfcfTk{jAgIdB^TP)
zq<U~4X|U{HUGt(X62T}J1Y4T|XxSx#UtbH)c^_QCjbC2m<g|m*K(bo+>#y+<KSpgz
zx}Sf}8d(Nrvgmzn|5GC<!g8jhoxQ$#a*sbPfSG88+fnu=kW<h-EBNXs7l3yA<=k->
z!XE9#T0FM6h$bNJK~AUZ(%XdL;M24D-uCj_in9dvtbn(Rh3#MD^hdT|vGXECvPI<3
zB=S1~=|nmX(RKlv(_^W1yU@LQl>80#mXm+~-S6OgZ?T3mVZT;^`Jg%I4QbH<6HtIT
zsL$njDU$Gl;$ty4R~6qy<TWThn{K6WNLo@t-WAyV6~F%$-?7^?gW?Y~Tw3Kp2ELwx
z&ku(Hq?==Z-qVR5KY(N!GAC1wu^2fPq>2f%twgO*_!&N=ms1HN44H*-as&#QWz9<_
zK9^Z_QLDW<g~=kYLACLP3``mlgMcE=JbLxGs|&&HV;|c@FO$x0i@b$~ERjQS*F=b{
z)rGhkB=oh+%cJP9CYU|M;;R|}MO};<P`ecW+)yfw^(^`4qsUuES9M_REpXz6bK%4{
zuve(6WWW4x;YpdgI3@p!G=moi$1eo7%uj6JC0w6ffs3+QIFj%lR-PO|z>?+^!FC;F
z-DMn!G1(QZovxa~Gcdc@+l$i*({4)z&4;o|qn?a`?ffB02`7A^&hPjq6q8yNjS=6M
zW^oowjWpDr!gdaZF5UMKSt@aJHb$}p48<h}LG8_FVXsG(kzdRH%Oc_15KAP@9nQp%
z8^VdGJv|)up>Lq}XrxpRy_ErQiA;+Ri9~9PSk`+$bWQq|4r6E@7dv}_uc6-J82nZ8
zU;ebYDn8k-3SeXAbZ}=x`seA$-{#Udk0fQ%jhNq3gMjfvzXwt=F6ZBaR2*ALkl6u1
z>g}9j!RsT2`)<|u)H05NR=?{W#okW<-V2BshW<y>z{-EzX%^^VTDzK!Rc_n1?Nd@o
zUZ#cSSNGzqmIj5N-&nKbXs4dWI>^okS0FjEXbZ@vhO(45NDHgpYOAg7s+z|K!PY+7
z%VXFU3TocK@pHGOAzaA#-M{{4L&i&%&#|_oKw9SGx5gC}UP20;(=TE^`-<UIk<{7+
zJL~-OjM;t&BlM}g8H>X@o+(x07z<aK!)SZqfq)1w&xbOGB8L`nP=Dq(IMpD~zx(`c
z32LZ+#4?yZmsj`QTYTGu4}w<MaUg!Cg6WgbJw!w%3RLdD;uA7G$YgUbDIsgN?lHPv
z!yfYRV#sLAM#?dD)nBQh(YHzZz;g`={r-z)+u@L_B5hIltseoc!iON@a#(00Ajg|k
zUxz^9Vk&arW-7eJrZ^%+ut6#N9UV3XGvV`_yCaeac+XE=KFp_?r0H97xoSYhyGa`G
z_eMuq5OQvC{_X|R4CV(RcCJfi^%ZY{>pgNG@au2A{@hzV@+qPt?R#Ls8lPlPj*iio
z<N;{7p}#0P*<`Ul{@aq48~*+KG9mMp)X1U7XnBV}!O%&9p&Kr~U;}m3Xu)atj%5Vr
zb)T<NII7zR;2cQS@qx!=lFUJDjPGsG{MiE?0Qgg2#-rNeqA?Bw*M43UUoDg}OCnAO
zfGYA=j>8S!1*r_)^2UOz4>FXhXYAPh;1Zm)_F-cLmQjBLdC+5kWz>i;U}HvWe+e1M
zDqRiqG?@K~pq!47He&%kDE?==*D7d>5=fhtS&lWfh-#oRgceF-%J$w!%Qs|kYH#HO
zmO@_1Gx8O9^4k&@N#;71FjlqaqvA%&{TfEPzpzkK5@r-i8t+l6a`K>uUtHbCU{>wy
zz74Ka(tvaMz6GF*@1E9nKuVPz+GyY_3!oi!^hw)R{>9UMPZ~X_go101pINlh*XpM^
zC(@vlJ^}YoX=JXc_nVkcSk5@Lm&_4Dza&$0f|&s%a~vddE#mkdkPPUE{Y=G>+NGk(
zls8EKupR_i%IYX{00u_s4FGp<5k&wWV=kN!J%8J{ShH^UD+I7B^z1!TX#dYMGV6Qr
zbv$xLE}RoZG=%26wL52eKy&i8(irQ~->IYsUxOMQ!!#^19S$Bmcs$DpCj(KmAA*=X
z)Unh-?Ez)gnDb6xCw9UV=T?(XWLHU{=BVya*QXRqM*mFS8!i8v6eRulKus=ONn2Hu
z3_u4%$5rVpfi{1>-$Ue;u=QEto(F~W#a`qeRaH*!XraThYaqWQ=KfB$jtlD7Sm#ea
z(};Xwd}k9NMEnDgZyw5uFb{q45||6a_+b!6)qfR6eh3l{7FBRSx`!6b#jB*Ggr3Gq
zRh8y4_^V+Az!6m|d=mQ$zsz)6j~?OqT5ZdLc(<C3B*?PE(cR;1jIa2jYPZ0}l;fXq
zt(GeGnIp*1;ds(esL#Cg;tSr#=+IXa_!jKOd|45lPb>wkA}2IU&kx_;HV)a=lx@5(
zbpfIY@Muy;6;NU0Y+J-x>=%Ltw=XJPaVg?^VWw%D6&s9IaKDD1t>Zk~!1G+(m#0CA
zJ_FoL@=ItK|F-uaqqaJCM(nR6(_Fu|q!MbEN>H2NNc>7c^WZ9oVvuSaL{a^}7sdRU
zVd+z!kc{AMi7@h+mHS<Mi*Juzg%(=Z-B2I@7*j-tunCyP+?uJ8x!3E9ie%N4h)2MJ
z-?sUXH|yRtM%u3FcB?NnsH&L$4(_P~;H%0lW=o@W2>O8C0?x*hu!2Naw0Q;&CUK7G
z0%`ssKp#-`O2r+3#rxc|EzT-aMvjnzZt1gIl>87y<#Y^W-Yfb6_A}&SyC(+`;C;9@
z?yCr=0W0?usU6eT!TKW%rS?-Ml7Bu4u|le=2ofVd3Ia*?A-4k?9oUf%e+d_E6$<aH
zRkO(Gmk9&s{+K(=B9)`V0@dTyhz_|uitlgI^Lq8(1P2|VQ5XCT-T)dS^(#Q{YjxT%
zU*(YM%&p@ejB4IY<juJpZ?i7ZF}-<a@w-sGW^i}BKsU)Sr6_a>&9XF5xO}7xr3E{j
za`7DAAA(>m2-CiEB32`}B@^k*Ys>i^n&sVLtd9rcl8gr+;);Pp^fvjr2hjNZpNAxs
zV3p#$@W?S}aL<9CHNF3i@n=KO8K&o|9<lp{?pHdQ^E*unxRu()QN*qM*a|SFW@g^#
z{Uv{i9TZCJW$`sNH)y&?P=C$yUv}4>1Li;P>-W|gfy8$QoaSpMp|Wd7D|8FELZO^<
zE)$XqKLW@{G^)_eC_MD`$|tx<si$PGHfpt5TfV2DtyV?4dj)-#^8?!L+_cO2&y)N}
zO94s|K2Bm3>6>5sINt&?TFpa%O?m3^iW){TBWd44``!Duz&5`5Ya7pnHojd@%X3x>
zy#DI#FOZwu#ou1R1@;0Y&QAnMy#_<~2Y88Q!9dUu1rP}|89q=$4kq3f@hxiyQvq%j
zg6RO3)fWL&G6$U1XoS=u^g{24xSwsyX3l*NjF0l9U%F?7*iQ(fP^9S{j^zkdg5S%`
z>B?-Va-R7Jc;KKL3y%%W_96|2{N4yUbl(wZn>n=FPsw1Bk#xhZ<dDsaB07_+>f1$E
z=6gB!j^-^*Yeb&ka}ax(7m~;lrEs%~Wr%J^>Y1g~W<@O3y(A!XyNsFIal!baSRMS-
zhi(tPz{uU1>IJHpme=h`m6MR5wQ(=<q7_n@a7%)2#B(pXy#D;<S&AmPsUgNc0iMM$
z`#AUrq$yi?2{#pIl=j|Gs+max@a@QDrk=CW0`V;>zV`jB)gnL*lVZxD-)>TFTaq=R
zE^(1`5eeFyMgwPm`j*kd@|W2l{IEKN7!Wu;wCD0lCr?*3#~W}00@U)zD@Cp;!-hD|
zz%h#|F*{3o?q^_ArO09M#K~aCVdHL0er_HJh0*EOIahot_e7I^1+CFJXxCc$Lr7n1
zX(b(m?TmbydF>c-*>o;R-m@DDe+b-1!F6f`p}g==q8GBzL<j>IioK%2q{QQl?BZgT
zXCP8RBauq{NR)p(!?y`^Z6fXLFN~;E2n!AbHwhTgo{G_@faTc(SRMr@N9EL(`jj`$
zKHQmmzDog8VOZG|1MzsBmrVAlz26T*Sk|2WGRY#%Fk4?MJsu|U;a8(M+}FjQQ5nwn
zxCX#Ec~>KH(%)&$O;h#B`g2hsF~%qUXZUtE!dCe?ObH{m(V9dYD34M%DTs_dwwkTa
zPiU^UE;)SI=k!2y;KV7ixA-S#YYZ-GGP%qoMP`c)M+*shGtn8xF`Vdu^eO^4uV=^K
zL}D`wB(?fQx+}Td^&nU*T;!`<XllAAFW8)69QcOC8(rUnKh>brSSA9sPx6!6IYt%b
zer{esKygCuXRMko`sygi+z`qrVy(A!!0Z`+NP7>~{RUOu1n3&l>)yXgppd9M=8x2P
zV_NIsX~}!P7a>p`^wM0ps)`l|e1|IFu>lvDm47}g;8$DvkPqwo27p#yL(&?nC&`zr
zR+F5yJ|7Ozjw>01p-w@r+0FBly@e^EX{Oiu5g-it<9(|rn%>7Lh5J=BPoFZW|FqN6
z#nge;(0*|!st{6DnG<7n1gCSum(s%^y?^YaX4AanKA5vEY2Ps&TBn@kjW)MFFUNca
ztZy&txF>SBdU5+Dp<G0Ycy*+u)GTW<!&<+vMuw?|U;gP+$knvqc^U)~LJpExQ-Z9H
z3P5GdYB*Kx=;NQ@hm$^T4q$LDg26f6D)k<22pDin%Pl2P^3@<PyzJy<H6&qylqS=O
z%MP8?NKX7}OC7`qAg(k-T)pYzo=FMTqn~LWXsn*BuwSj_qlc9)Yy1}#LN1quknlHB
zs}mVJZsG2symUeN0TVsFiP{Y$ZwCea(}T`}%U#=138gF15xc07LnUvoMgqNlD^{O(
zm)x^-ofhf#(^xr_kTVsNywBYHwED{vycToN^pX%TL)rt?a6wC~a$|o5#)3P&K2O>S
zl*%aa@LQPRl2!nXZmiH?7;gLt+Cq+9gHD*h$P?_fs*Rdi_49SkSe%~5sA;Ocmeac@
zffAx5kYbS2YJNj&^s7MkjSZ<Y&jf=L9}JXp2Y>F4NQ?d>BCV=NQzFHEa=Kc8A84$P
zc4;5Zv>iPVL2aCtSc3!KCEdvidC~AB3GNE0uh3)k_tcpz8|R^&lHzd!EkHJL`J6m6
zBbfWqH1toTc1mu<1rths;cbYb-Ih2?VyukKf|do!I-|5i7$H}G|7DR6{EX)^paHX)
zH64cxSnd`A=x`5m8$ynrS#l}sH3VV&`?1dD?Ca+jEnjG6U%SYSpzDy{8jnm%OgrBv
zq%VXgPYtetuY80_CV^BZ2;&lqof^=<(6*ZBgdqN{#gy_rf`dd3_%5WpS4^P{!2uuZ
zA(xrm@^$;*K$|AC|5{UhscNJ4`M0bzo)S(3t1Ax5mtS1`c|;GOeVU*~S_J9RM?x4G
zB-{k$aW#kDM?JVlArYv<4ttLdO-#Mb=+AFJ0)QXiz*TLE6p9B%oB!r@uqC*b(^Wqa
zsvS@F+kp%}x38E0(b*P~-4r9>u+$rH>8%#jNVoPWtpgOUz{w)`^~Q9Cp$Af%0*4J}
z`^%Lsq>X_qvyVX_TNo}x_hFddy}*fNG5}uh)1HlS)6|GVpsreGdggcT=(O4QR$S}I
zaX`-@oHSNbr3A)epTq6((4N2*D?tK9C)yw;l2t=z)PW?w&#_H>yQ1LP7o@0;&zUb;
zx{naAC?$XD@kUp?Uowa@)s{^n?b!!G?6u)X)YQ~b7lV{O{luL7_fD*v(DzV@(SmkI
z+-&}(1xCZ0Y#$P!o`p5}mbYf=Oj?!U!W!|z)J><4`6nefa2D-ZO=UQZh7SJa71gYx
z7~7+?djzil<`Gm3D1Dqp(_{hw?^B*k01^wcSCMt3)S?r*_TOJOQ>8exW3^L;TaPoa
z(jcRsf=TmmX_ogXxiT_X=dRbcxGng*#ZYjY7C=v9R*Vs?&T>7{?iTaGA~$oYtho0p
zZ^@zkp$lI}WwiM9+oOlf&V`Rd98nRJ!Gv5vDq2-fLY<uyGI`U5`o3ZvrzV3sZW)#H
zOR<54qeX0x0~N9k-R@6P?>X|Wmp=PuoO0n;FP~b-)+tC^gC9<sN_z;6K#JmU(iSie
z0pVV}BE(M6uV5sPL-+%-=+hFHx)Dr5i|ziih<SBH=!qJ0a+j{bhDahC0!}useR}0b
z-x?qO>9KcY^oQ2w3Nr6;KXHZC^Deckd3lSrb<s=Cx?6!=fadSy@dBow7)w{n07Yc{
z@u>szm3Dbdm11+00-3CxK1-o<IoYf^U$<zOr%Bn5*@We$KTtsBGDDWnzaekD!3+2U
zM3(#5P>I<~A+{|?4mtF7qkC(&(PL{boR?u12gfiixaJt91rGp1*VH}8GEcx6Duo!i
z`QSlO4up7%vu=L~5SeJ^Fv0N=*eA<;(H%ON*wIC_++42i9zT^!&PDG$)W@rUh0ige
zbYH-4z>p3;l-dF?&;Uj#BdDfq|LGU-c}z$#@V&(q@R<igB)Ad0T_DB*^%z}YdfZPU
z7$W*Qj8g~Dt@S&K`Z-3TlHaCE+|^&>%KY=_M6nR#Ay&D7>eS-7wrPwNFoE&Q*BSRI
zxTzBgGtcLF9=5d#ID<Bc%KiY+0%^-01k&S?Qn@HsA;@k2arzUm?kp8HCFR;-UpV&`
zXQ5DwH423Q?##4Me4XgerA5E8luuyWNlo22Vh1<3VCN~fdqA`&U_A<5wJrOfhk~((
zGKhIQ_iF9O{3M`R21rWNh9QZpR<ReaGXnXgKz1#mJ#<+5SO^aqGd%RjKc9d}^^0Wm
zekV_W;NB|zy7^aXOT|-sJ`&F46%kVZ>^FBDNDgD!Fm?dq551r92e?$rxOQ&Dq4Hbk
zLn$crz<1A8hiZ!)N^7bjnjiT^?w+49*^G=z+X<)fE7H~&|CY5&jSp$0yawvA^ZaG*
zLWECj_6{FKHiRGyt`hc8@anFx6#xA>{tMEPZ%pP2HBk=*=bkG9rFI<-C99=VXcI8)
znIeZP`vI5-%Y@X7+nmNbKZOyN{gDxZJCTOnLWomS5P&B)($M>fHgDpc6|Q;E1A&Pc
z^Fpd2j0du9X~xRXxQmpaJP|$-DvqflBy(al*FM)Ua`S5dFOdF;)1)`whz-oB0Vtz%
z#0>@#Yp4taVQf%{g9*eW0pgRMa<!QpTK3#}NeQ&iHB<6{)+bf2F`g7ub_$7T5?`y5
zOi~=Jx>UPzQa$y!d97UCuZwmh+nuiv#sJ<l?rOzr9!SCb39f~g>KbTn3c;Iwz&?c*
zLQde#iMmXI^+W<d_R%4@QjAj06atZ3f<yLPAslxPJ&Mn8jS|)hqWzVi%ZiYV*+RM+
z_1%`xBq<dVgI4!W>WVzxlMBpN$%>!)O0fn$Hqh=R@O66B(oG|P-&pU5FH!FbG<Dqz
znuyx{tcX&2MRa+Br8}dygZGqh-5j(OQbV}0lM<*Lo4L8oImWMVdB}-r%jp)}x^-)J
zeWL3^jMU+VqcA0SjC?iy{n|6nRbKIkn`Q27`ULzuWh}1)KZXGuywj!jj2f^C(j0~&
zBzVqnoyWaMb+2<Usb%*U$*vA30%P9POUt)qiwI*>DT%%Tp0}Y0`uS$(uW*sH)7URh
zX1=#u0@^BYH|W3I&7$TLCTRQxaNx@*&KHm+83Xh!S0oj*&;;WxuOdEM3V?BpB?vH3
zDohP019!|t1F)UIf7Zxk7@`h=J#599lHY&cO6BbC6@{3XwbiSv`52n80_Y&RcXyYg
zQr#mqA$F)tEJPYh?{z>J#+dNmqtHUDUHRoYE45kHB#`oX;<*N;C_BMKS^7K0tm`V1
zTfx?pn|K#+OW=*!TC{_YoEU#`;>z&;Z1y;9H^O~xXXAd%)=rg|Tvb8{g~X@xY$1F6
zxD(N?y$sZq21(@lTA`fmmL$0#JhL{}HHY0+48_^#{pS)cp(Y<H+=oftEnjvYIZwC=
zd}ok5%ELgoZC=X_5i$LvBQ#+WQzNXSRV==ueuBmWL%(dp^z%PFKqUdCs~i}q2o-`X
zNFHYdY=f+hH7@@Wk_NgZ3&^<QDf|?|zR@dq04{hg?yf;dcb4Vg(NrCHkH9`ju=mqN
z@vS-Ef2IXBn66#^FBi6b!c$*_5|(wx`yvGhAE{jrS!8q<M;f<0l*<cQ3APc}n7dnD
z>iJmVbpPASZj=ThXdu0GpQ3DYo{s0Br4kkZBz)@feRIW6%6R+aW@vW3kHkevyFg2K
zAmEik#6ZMMh<2=6gnWGKc4f}9yr&3|K%PJVKf<SkQ|5}Zz8*LDGdvM4c-F`PUmQ+V
zq9e)Np=sMpeOZT=$$0|%sk&M*p}+KZAT2ut?OP>}M9F9FN=#K9HFiU6k$J6V?XRQ#
z-*RZTAf*cJ)O{1v`IodrDMkY&{C^SSafRsG%7)93Et>z(z=)_Iy}e&FK?7}SEnI%4
ziI$J=6&ZsdOng65w1+r{#=ii{%%6Nj#%K_Gd9UwHc1;zupzPD0c|+^*{)&d1tW$(&
z!@V%*`RxWGtD!)QDZvNe#|ZF5O(`_ZT^nL`W*OjG%izly6l1#&*!unq|K$|%o(Gbj
zkcwqn4HWS+fYc?XwjvCLLYf8&BYedfk%Y`k{PMS$`HP)!Phnn7FSK^9Xt!Grw2!+w
z2}6<&ofQ<>x>miXNrT)I0qJhJ6xPG8z?>#A^+)@2g%zX<ZjmXU%mbjj(0xd+>M3(N
z$j&Uf#c>o`NA4ue;zWGxJxcDJqi9}6mzkQ0md3m@a>2s9vw~`@f87nMwmLwd#07W5
z%oa&wFsjNRJaX`2CBbW0Y32iP*I6PU!IEm_joG8nIhBuq$KXxa<?jSSb2cq=b{V@)
zSFA|;`7e@F63o1}|FAsrW~Z<4)(4Ib4_?+J?MQrcXK%5}pE6x0<iqHI6G;P?)miTL
z=lV&|u-s+2TfAil8S9R34^EvjSom!}$(A=1Eaz_@t@eF#QDI!%<92(9GhQ^x37<85
z)owf&r;pD|nH%efs>+XdFPNvDaGdNY`Jg@CT{bms?<YRDB3G_;B>c$SJT?^kA?Y5<
zs}!&J3@aZ4)YDMnzG0{_mclxN1u8*b>?ghE7(Q57^5qfmB0XL@%Qy24-0+REy{UB<
zhwMThf}$KWcAv!xeRde+9ACKmJW9r~UOz7nlAy8Q<;kW8t9+GWEgp=vhK;89&UNNT
z3jNvbpP!RxsC7mE<DG>|WP>|rd*Pb)&@?eOfz5ufHo5^{!c-r`%$4<h@1!5*<Vvqv
z#8j?nd-MnO6YkdBT{!{0cE2^=wswm)xm^&Ac*tXu(Dib_e9+rc^mI&#{eo3|xph1P
zca_Z2d0l?ebj!{ToAGYSOYMUv@X>c330bu5^Ez;NuhPUtG4WIQ$ARw}Pr5Yp+9%1A
zkQPf9IGC9Qo8N~4yn8@qlqH4OqTMTk!}qHwI(L6z?qYQGar*AsS!%_PeOfR<#yV_T
zoPa)^fG$Ll#DSdmugW>0<1aZsF4q7#k2hi03VNS8bQT^p<v7^nTfS*8x`Ovya9hH(
z_IBN+{Vc2F_$8rKD$H)=JM>rQ9ne{<RauPJN$zWpSYi6M)cyfP{dQjyOrwP!`zdfu
z-g{w2P~Nng)XSU<jIW>H`J`)w`jgORxl2{O0u0^QW4LKp%It4I$ggo8C}>#b2n-C|
zrkRlSQR&?|fI;M!yz_TGk&w|d3=^PrWMpIv!wh0fD-HZzxO_yXlw1b92S^*UYQCje
zkn~<DaNw;P|1Dul@ASd6M$+Q5;aOFe$h=;<p470_9?2(o$(Cca2i5%t)m7%$f(|qD
zNRudKe1f}mZ;ec2Q9LHW2Ux_OMX#`w>7p7ET#f>qyHyW2(k$kyCN`&pqZ?uO&<Yze
z{B|<;#*j^uf-48)t*(&;m(bf9jX+m!1asG{ULq~^OzITo($>;|i@JI{GCV^E8M8X=
zSO<9HI7nEI@jTVxvyCseiD!7<`_Yk{KkIg*SGi2|(~iQI{ezPyUHnj#HcEnMI}Ccz
zL=CeRyA({50U`Kq0zm0dK(sP2s4zB%Eu==BtQ24R3Z#9#xOtEE5D~j#vAJb_$Pxs0
z2o0}HmZ__&_k5xx*qzjNZegmLLV|H=sc@Lb40o729TV=Cnk5b^JI>bYJoL|Bu71Nw
zn2ADRDWWnyMB^5umPS$(zP`HRRCji^7^=>O(zBkj^=`q-_;I1)E5PfxJms2iVQ%gX
z2v?Ss;lz%;_1*+W=ZuDW$#y{s{`A|^dZC?LG`G?(sixYI4j+Mwvqld$VQQ-ws@&p4
z%Q67HdEmxb?q$O8{!1EHRa(N+P&?fH0xMGG_^czYbJ<jrGj&xnqM<(Kt?9G(PMKkI
zsV{}3#5#$v;4ihO0D|V?NcOmw<Cw$O2XuTlW;_J1&Pm3UypfVANh)C!s}1s&Lq*&e
z5|9mJaph4(u<|Tjq76Kf*JQvHNNq`^b;@n(l=kk~a-*KK`1Nh4?+#<Hms|`FF^1c|
zDmX}f-M*p;`6`Q@yXb<Ld00ubxyQmBa`HVn%By<S7Dotbfle{EJ=%H`*CL^b;DX_L
zOvxb1O)V9JK?#|Mb06SxHa>*@77GT%e?~k8rROqL%I37ffsa(hjA3#K_j4VmE!>#u
znX*^#odm7r(y*MY{N_mDywJT$nr`lNW@WjmQ;;2-(0GA6d&H*5a!>Rfn+{w$pZ@d!
zTk$PBAl`efKU6zc44sxdk*$Y~^b^pC&M}99?9IROh0yZN297^y;C25<0enLQA?3up
z2>aK-uiGY!pUFR-VEN)~V^+&KVS-Q@U#%-mQ?|tZ(CIHMSe&-7$9}^Qv^4O#ZkIOq
zdR2Kz&9S0(F#7O5Wx3lD?V#_&;-t2aG`O^H!ZE4QD>lmCwEZ~ab!B<oV2TV+ZH5nW
zgGF+=J}HGYV|${#Bs>})Yhg_kI0p$?hx7z5=$-><cJQ@+y5GAbrm0q0qPEyJog&bv
zxrHjTSYu7)b$v;Ig?cgL`<tp_N$U7^{dUD07X$g<&rwm;yD27~rFRCWt$UJ;wTGz<
zjBDrT^K7|!0h2MY%DCmhPE#+er+QOu?H&M!nf89=t9nqlYl%rxt0J0nP+iC*Ras8A
zjY62a!Q*6|1Mii@-m%-Mg`~yjy4X+)RjTPDgngFg!{6RdBKUA#=+fvhpnqSTmLN!9
zN${SDOXVircHqnWcA=Njkv3QTv{PnIPt=(0+CtK+1$H^dHyqC<q00;!CX=CZ)yX!a
zf$G+azuT3Y?sAEA*$wrfw3USZ63+`HvmnjM&u-Kze$&>_(&><A^wylx$mng&2c)$J
zEX@|uGa{H0yL=~8AeF9QIq(f=l!Ke5!-_kYC=cYk-Et<C5&%g1qF1^G;w{BkpkiJ-
z9YDV{!Y2R?DD=<SFw&-`vqlfuRm0QJQ#Lv(A5U>|K7WJD?57Sh?GzFcnse=J({$-P
zR6kcA@&5h$xuo)p3r9L;zYn+2X=!P3+mAFQMa(T77Rh>poAn8;(wYSJs$$$|D`6rN
zhX%u(ic#^!byPv5Y1k`}c-pox@{xF=+O#gvuyy)C5bD!0dVuq)>aB^7+l!!KEK)UR
zJ{8EamAfy`p0J6J_05joWAx{i!J?Mv>u0!c45#h(J-zg2cn68O`Dl{!LR2p(`Ur(q
zz=68R(Yon?28M#_wzhB6eDf(a&CPVJ>E>SX8knL|HbPbC;xsn5wXh}6RLi?+!A(_x
zg>vt~z3#_k{fZZ78A9z6Z#0SpH4Mr-b>{{Rhh7|Yy*Qe8aTbkcipoNzEV;^!ip!36
z$fAAwn>&^`)10`|5@U_)T>`XxwdM>uCVeoo9gU3Ux8(+-SN%Y$FBKMm_^u?C`m-q*
z12D1*1=u11jk!lAyAse5c5!Na?vt!tIjRgvYv%&fTeGs~Acwoz+WxCm67<pcns>Rs
zrnJu$E^*384)z75nqhJ6Ie8-dTF>Km7m@A1UGRVsM!_fc0wM001Iu_wuBGDfc8OWm
zewF;i2f~w|dwbD|B1g2Ro-Mrdib0rD+<9r5auyzDOjk1D#F5~5=AOXFp&8}5R!fWV
zOxKz>`i^&g_j%|rz3b4-8TxBm8ndJik}XMuuaH!FWtT9oo{emoCa7#*-)|M)DmGO~
z$mVng07Wmw^)oUCh;j9{<8Ov~iSz|f{kKQsJt0o>KGvmOGd2nTz`UD-<P}*kt(~uH
z<g{L-ww><>D0Y5f?^4G&^@R}ZH0A-GQoQ%eEuXSlQ$KNraU=R5u|3s1f|coKSCdV}
zX$5r(sJ+x;se+7d&idoMbNPwlD8?W`^D(@_bDP0RpVKx2Gom+2wrn4PXbf#w(PF1L
zXEumjG^N59sXzLPCUZe)x2EYscao8*sTU6C^Bw-fbH7L<uxl<=eOekP^a#&)ky%&&
zLFrgU*jxG~&Ej|ONNw_+3Iz^H1O7uf0q+~T4<GhpmC$`d4&F_=cERn9XvFLv;M#u*
zVej;VZ;+69d3!ow5WzOGCfs(CN=rh=2W3=QhwMHW2L@Sw&D7i#WBj(4nTMB56Km2t
zyo}-v>&*I{=Mw3ukci}&z;{&-Z|?NN9SMJZ^#TC(*PK5R`mXN(zxl%x<?z$ygY-Wd
zW&ibK)~br+`*x5U9lv*3dMy)(=;XiorXT;RZ$C}-6T1=g5AuQk<<mAl?cZ$1rqnj2
z_Rpty^Mw8z?bwvsrqtGPW}7DW->Ao?)HbEI>8LgX@PF2UO_NK$8F~C0Be^NHO{r}f
zk4@w8uUyNf@z^vT|7+usbETH*Cl|m5bG@5-x2bpkYrWglwf`(%Y$h50&7s<q+NRY0
zId<A~RR2ahHl?;HwN3ZFc_;m6ySk2cY#NVE<FSr<Y{qN<Mm;vAwkfrLw#J(#_upv8
zrqnj2w&|!gquhVCtLtdTrt#P`9_y&bX1w-q)MHa>n^OB{YrJW4|BZJ1e<ZckeGMnp
zMs^eFSmGHV$zSfz<#m|qg9%|5sv>X>&~TaCsSwC-+_Y5J-hLSVgRZ{jrOb#3T;{uX
z$Fx_b{Hcub3_=!<k*NqbDuZ|P-T1~>Q6Wh{vDb^IV=yur8XCxhDubJo2rEm}#)Ol9
z66sHp+-o#EW`SjFdE>W6Wlrp~vq<Wqsz#}feYfOz&MHun@WF$<;M(vv_Rp_z(ehz!
z*-Z)6_t_u3#x?CPuHM~>LA7?q^xoUMa&jySi*p?uHSYl`o}!GdI|{ksW-X~MD>Mx}
zxk6TN($WNCWcHKRw6xSNWkP>;s6*tAYfZD!_j#`_wLPzyxa0ehBk}Pu*-FQc{GIDK
zJ>*9P-*==w9vhAi`On(s-~X$E@8q}8qa#%O(PydX@e%RyJ2Z@ZOPD%^f6sROyxY?k
zMl&N+MC*yAj4ZHRRte_75s@|W3`I-#$jAqU)NWF_c7?cA2H<~u`jd;{$Fxf5SvT<9
zgF+->%Y9_PDqz~PvefuFcdmnWbo4Mw)_6sQJg;R(P5{bH`;dLtmZVNs8z-hj$NJrp
z9hFBs5A6eH<?C@e`9}oJTlTfLOKK~HkquUe@Bo>Jf__V?vCF_KVy}<jsQ%X8E_jD~
zi|<&ay2$#nTJ67+&~{HqNC4hR!eea0k}b?E^Dn`TR_&tGpX{mYNeRQZM&CII4ZElM
zW=G6FR#sN3X5+e5QwBcH7jchyXOAsf$7$p5LMyJRrI)x6AuUmi%a5C&3gjcDNVD=_
z<kFpg6zbY)ij^fK3Y#2T={NAT1?1#ywynd^F>WWVp|KR+h=ZQvgP2ZNQ+@#fD#zxh
zs$DXuERl<+4Xkh7x;18vM$SdqOOBqnl#aBLGkoiBvV^V~X<*3>z90RbMEVnT`$%D!
z+c)5)t)eo%??s#c-lBVF{jM;pPu5~vsMto)orG>CuHBU%RF09Yr?S8RAhKl(mA;|5
z?E1g9rqnFQCq7c=acfTp>Tl02mJbr7a(wkvTNezLlV(3c*)kYLs1a)jfPr1>3RIti
zIsBSguDSR6zqa~Gy7x!}gMxaW$#OByw5;@^#^`2FvE&v63Q+ILC<DH_l74?ezLALu
z{fFY6>mE;7s}r(TOAD_Jt@Wr?*$K8HAoml9PQ!nfu`3y<2-=?Z_1pmzu1Vh-rd&~F
zlo^y)`btu333oa35!tbDaS{^x;a-c_4ZgekJ`>Zoz^p9WV&HT>6>T7$JTKCj$!j|{
zGG_O3JEOgX^YE4Or4Ym4*A+BynyW}iNV?$7H(~^J14w{mekjSR`^a7O@)#wdoiVOQ
zz49l$vD-A;qfjo_fC){%VByB!ThL#7DS;;?<^NXI;YUQ{SCvFsLKmh*TX#4x(R~ZF
z$m~xC#uRM0AKk;AG*c!YAP&s#R66WDaYr?429K}gd;04wfB%PJZszdUEBd<BM7Cq~
z-5w`^M?B;(5a}G(mNgv2!C;2*1C72{y9FAZ^$V1=Zq-ETP4$&-trWWZ_iy|4l3`Hj
z*Lf@H8|mQXAtB**tnqrY!}#FWlG(;Ydbx!UrrFuqy;z<6^v>BsH^2O0jcXqNLDsXL
zH>o}dIz_L4Xaweb5ml=K9x|9P!cx=Lb~VQj<_&PBTePcdqS`)*j4#iDVSVbVGF_^m
zO%op<zX3VZOM+VIKfas0-uM0V7zZN@UVCyqF7oPxS65b!TGz>UawafYbgSfNC3h-)
z4MTCmtb=14d94h*w)gRUF&H;WWJ}<<M;5@8jmq!_EBtKiHYL=z9$-Z%26FTeS6ZF)
zk!2*_%Ab})=9aA^KlkF@tD%uaaC;IPQ7I!&^5Ck5uno8$ug9rPE7n{94|a$!tK<6_
zILaRld#c}B=;!DGMiFevd|ReXPmRpD?@(RT+$%$pbv^_65x<{=jsB=l?nV~R%ySDO
zjL7*fGJXdV!&7XWH0|0kkAMs$L$9cDQER5n26Zy{wGQ3h^$%&aJ&AmKyLr1eMg#X?
zJu&_K^nz;AbhG9?(}PU(Kvp5eAv~6W@@68;`@Cs+y$?TPi1G!#u)L}_bx{glXe{Fc
zgtlM<D-V^h=^cG0wz1gsmb=SbyAY5H@;y=k4oB3(&gFrXp2d#`v>p2E@my1{sdIIN
zCG$_Qt{<y~^OBMB^ABV+0crHFlm1VCaKA>5ki=|Dn<;xq;E!Ic?ZM%2Fmu5Ycf!)#
z{GqSQNSk<_JdxAlUJOhC)fxKwRsZywpJHCsGh`G(Qj$>(Z;`ByO$Dl?GDO4+D@&P}
z@2Uio<<b-uoI0k*fP(jrupw4rHoF%H4UF27k_?G2bqh%$%{&Y}Rn3_J4w0VpR%O~o
ze?+z_OD_Zkq)zIrByD7=2Vkj9z52ksN<=EKW{=~*%$Lh@qoM_4Kg5ynogq>&#|m{H
zVfgIzL@ir~owjtrYV=GL>eLu!Q4m=|brWJXot#HDLM%~0Ll!9?bU@ZGSSe>lUa0*L
zgP_^*gieia33uwS`OnF@oF?M`byNOkHT>bXZ+mPzv@e_3B@Fp)o>L^U{&|V^QX;mu
zij-91PbPF;xdx0f&UkDi?gUU3UwqK(OGw#>fN4R%^ZM7R*FWKF*3aWOu#_i0-m4&*
zXFs%`Bf$qav<@^<U$qC**#o1thWVC0QKC8w1BnTjM;x60V_)lJe<HiE`d-O>^VOc>
ziaC8*=zK32^4QR`=)NN{qbEi@#Yebj!hSl4cJ7r0F}wJf8sfTM{ooN~@bKSkz<($h
zY-q;{u1+5=$3?Pk{Tv*ROqV#8$8lqVU&^LLf17ZrB#C%)o6q}{v)c*jxXKr&<V&O#
zPyD{8^(ZRtQHJwzcBYSXhYSxzy^?e5Qz#+bas17Zz`)lZhvkBd$cw+dNRr0b?s<A~
zANw}`%P96Y+ifYA1gqVxL=$qdQ*x$)=N3PXIE>9XbPVPtE-kl)1o`W9&<6<-GzudO
z%KfC@T{eC|`5&MA-)*i0J;^ryw>4~P>szlir#7S5_2SCKhYbyCnF^?dzT&1#8<jCU
zQ)kAFin~nIMb)n*dls>Tm?K>4J3%0@CO=sH<YMCW3+p<|HThRd0)KO;|5D$_yCF%o
zqjopPQMUT=%3hvKu6<3_-BmD+;ZdH;lHg}DBAdm0c*|D$+M$;)e}=e?a6AqZXES*X
z%4luEU`C2(Pl5Z^%sHm!3|G8IQ_A37mhwW>_Fac)ggQkrTsMH}dtGH2qJZ^kS6!DQ
z@_UHBRd02Nfa=8Dd!f!{^W_Oog}x5A>%A^>Vpe^<HDQzozItuhy6uqG5-}MY@3@*A
z=lAYaR@_F`OWsEd1U?V06&4q@TIN5;R;L>KmlS&wf=$O?%X8<=Gx1vY6rq=vX0Z&3
zxwtUi>tCK^We|%L-?Ayna~sgDQZ%nCA6WPIJc)c)yk2E6!%e2P)S;*(^HyBjl1gG;
z{<U#*qn5s~<lRT>3z@WbA5Zv^iaeOsPV8ks`n8=zsGVPk{iXBvl#_ZDf!lAIyAtqo
zy-S@0&#YX}F|~9vW38{mZv3_T;f?JbrEC0->#J3G5D5$;^nQPZyWGX3J!`yZ1;6B*
z?be><so$Bm=mLWleJS=<qQ=x1i0LsEqQ~E#pWMj4I-b}4c*h~La|UKd6j8fQ_=aVn
z%9B-d7gs#Q?AolIB#Elz|7eZdNmO*i__k4ti_DcO#Wpw(cd&)dTbs4wbkI&P&F|oI
zHn$|=$HkZuj*Bhf`s$nlA~v{M-W^VVaa(4JO_<}#Lb{RZ(j3=ZqH0oBH%#u7WzZvL
z#Gj-~94j|bYdQ91l?|1M@NY}Zk=l;fixvBbq0Wd*<A>x-v2j2=xZ)wJ?do6(bEcUZ
zi)4Hm=@Rp@jsa2RaL)lE3xM|vjLsNrWUUh6OC%A}-j6vF=BL}XHzsz{jOR>WEL?FK
zQ>&rT$h!5kdnXk;k@cR_DY)%1;}ngIzgk~mf#{5p{G=)m_uPB@Qmg)G3mUy-H{Nj=
z$iMQqY>VllWK6MCXkJPp4>H;X>KYi~8$;aaf)$7?`1*=Z?|b!;Ooe!D7q>&gR$bp$
zUUG~h6k`!sNhXUNYvr8>dBTY+QN<Bm2}zroOZ$nh6Xq!HGcq=QxG=3@6V{T3^#+1f
zw=+3zop~;hnlZ&zDzoqQI?#TZHigI<+y{2^guixhe7}c?P>|=iTkL>wee|513xs<E
zKyRE~DC9hs-RWi5>5@2>Zd?~xJU=<uP;`>WHibKoP7!P~9ZZxdh?_llYavOYC$$$H
zU*uqeJ+4{jvrQHL_q^G70@F9Ek&i`QmSVO;d_*3^|7!KfJg<LyXanoqLXt`5IupUB
zKkzkipabPN^Ti&<YGZM}3_`3B=f8`-%8s@sszr?MCett8Y(%cGo3t5IWgL2+A6(0%
zc6Amme0>+_x_N?+V8XrESB7&};BfFSv6^mRS?=0<qLpJI`(Wd=oL;9e&SW$2b!^gM
zI}xkQ{6DeY7sCsJPYeAiD_gbl;s$O6tKU`N&c-8>NF;mj{``+{!+a%ix>No?*u2*d
z7GNm{&)hQ7bm-sT?yB)_Pm=v)e7pD{F*o{_4NjP&*mGVY>VlY=0t%V{RDN2kWyqP?
zWa06g$^ZN31Cuz1{zO+w_UynhSqp4~YY?%m&_W#PXvkM0zqcB+IbI4^2otYqIm0o>
zP~vm=F!<}L+)imCigHc^6e@LzZH>*A*e>jYg386K><ii%R^4}^mGKMm;+)y*AO9zL
zt4DFsAbJC#C5%oQCx;3MXQ;!C>8QLoH$q*Hw@`Ws#I`7OnCJRugSlFiM0#FGfm>;f
zL~{M_s-`3n>JY|eSiTZlax8=yo7ahGVxVmJGlR6a%nO&PHlR+wxrb7Yv}X%#qh{Qn
z){@7ilZ?iCO)gZ0HTK&O%hrZ3kj7M~JbSeM0)Fh(jAqwkzRCyOZVSU)Ixsx;od@5o
z_q_-k<vRa7Pz82<U^KpLo!|NOBe-ce;byf$HFM@0xWXLP*O4tG6&#AC$?6R`{qG+W
zJD<q8lb9nrBO;<dKVj2TA>^iEXa6<e(&TWg#vAon!kcd&Ro4#z!bwS7uV~(3<sg<R
zM^+Q=MB=%PM=p>*sk_=7!^so;`VT20O4&~neO--{L*{)9k-X`eyq_rO&W*M6>V3|v
zshP}|p`ICcQy*zb<J?9?e~W*Bh=P&7kO@~0y{xp6JEsJ5SV8VwLOesI)%S04YK`=j
zPkgJm*4=Y*{&%Y<@rrWf@%4Z4hsecfj4UmEpR$ycjI|TA>>SJe(gS$9m$L$h-Kzd4
z>*$&%!8TeN9jvr!Hh4~DC@dawcs<1>r!g`?qP48tM;pN#d`~Et>W`)OIjMG2gZM?F
z>=>fC$_dYuWQnN7>L+$RM+Ws&0G1&&%M+qrm?D6YhFL^s-``DIKMa#7C0%aFxK5?Q
z_17_-m#@+L?YDz>&;R+y)qmU`iAE9_*kfW3=9klhk3ag?8U6b|z!<HUCu0rzIW&sa
z7wV7C9R{{}xH)(S(QK!-WpuG>F^(!u&}!`gB5hInX?G&Wx$J}@r(tsp5!b)AVnyx`
zH~VwZg><{bLIkktg>=KO;{JXOKX*b=-Ti`JeR8dG;*Y_oKY#EK_9%bvJdDlJZ~9=J
zWLUKw&B&cQa~3W?_coxuSHjgRla=I&ykJcS=18c;X6C30>9}%d_7u{&D)jCBYft}f
zA1`o2Ofr;D^4t2$U0qfGfiKA?VOCT7?GF&}P!SDdjwjyULe)Q-TegvS!If>I?>R&7
zOO^C`wa14}izehb-XNY(DC30B&s<7${JXXN=aM9b$i^}arS5)wBZZNcm6es$Bqq_a
zRo9{1F{QMT%A(>8_o%QHe;6`GhbD~*>p<C#ISYWLy-JH&|M>mpev|TQxmMorM#NOc
z{=`77&0(;es=q44DPjW`SY};`yW>R8%BrL!PLK(j*`I7_(T<DFRgeGNn3w*6__dLA
z7v$=?oc=-W|Kpr1o^f|s!WC)R;n9-qItyhw#R0q#3_V1*uHg2z&h@o;ZKvI9DPStK
zR+7t-DH^T9z8fxvhQw8f{1DthyC2*O`;Tw^hxNRpvF~}vR1%|PN^gR%^-p-@#g+Ae
zc>WfIgd`VV6H$b<MLu~{0P_Ma!2E`i`?M?9Fjri=@23Zd<GFis=T@S+%`{pE0)!`h
z>4^)j>xsnA@kIZyo*!#=ze<W5?__$;idWOTRTK6(t+CPvX2Ku08-@uRw_woFV}(PN
zRuuv|0&ppOOXN~`JPJAXBiS(<(C70z%-S+e;KuM1R%6&k6ur*;r#}0ES8n)k>K%^Q
z_yE4afdE1Cftr*O#_x@ZmcZceW@GlnkeJF;Ggn6)*-*JsqU-2>gq1z(D-=%YuS9C@
z_Sd{={WbVB3ydh;k6)UDF+81n^A{>An;^+LnGz<p7o1T;h={HA(x{jI%=b$r9?nF4
z4ArXRqq?*aE#5-H_DaIZ$tf7!@L;~{Zf>q$^Xb9I-a0E&3&OMCt~iZ`XcH@CRh&>X
zuS~v5ESf|%zuJrhu#Cg?-8=D%v)*Xzm{9QR=Dg#>>sPr8DE7daA^qscxq%f)jPDR#
z{N9_3`(%=~hZ?o43xhJka4jOa#7f$M8f#Y^*{d6=6-o?~09ezyL5)SPDwNvs-3!Qq
zJ-kn8IrP%bfmo*Jg=Dz8&nYEh$SFi%Pm&8@s^Qf{D#Q6=p)tYWBmLZW+t%0g2|9|I
zg~wy)jR@;E_cjs+K~X;^OBDvsU5EEji+(s0sSwssf54ik(ys;7{t99nGBXo&l(%ec
zi@QEX^o(LUhpbss40UMB{R!qf>G`8e#;u+Z@fl^Dikx-}F&m-u7j7ntsW(|3R;em!
z%y(4@cEPD-*5TU*SD5`yW^%T)2$o1vi|GnWTAXHBUy>4C&q>+Ul*`43++oZSaf9vt
zTs}M^f(2%kO+xN&P_M+Nk($s7RX9kAGJ8-n(HDA9i7Zdo%CeQr4As4Xgaw-h1q^&$
zH!^VllWeEaxm4>~qQ{>22f`vA?TtQQ_cCbBg>7U}>;8`M4k!iO=)At6aB<ybBCpwZ
z9KtL2ri6`|Qeuz$pVXU(JqxDFDMHeKBuQ%{R&M<bMOOOu1gtdB<+;GR(%sO8!cHoN
zg9{|n%O+NV+t{Z6PqM7UKR;XoZmc^#+rp%fjV@;$@jZ!zl^WyZ8~zWT1=$yOsIKpO
zFiNZ_vmpZ~!7PvDD@TGhR?uBPvjc?y5;*q9X<C*3Z&n(J;0T(9)HYP?CAwZRIQELn
z>9HGAC?bz@=SK`&ie>Wuje*+%W#-osW-2gIjQFW0`a7Ejj+mnU{kh&WaQ{yGHVqt+
zIPf1#j>M*c`*)6F)4**Sxb-FQ|5F6#PS`qn%|I^HW$f^8@ZWi<i+?7Z)xGoo01*y0
AQvd(}

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/sbc-student-t-normal.png b/docs/2_39/stan-users-guide/img/sbc-student-t-normal.png
new file mode 100644
index 0000000000000000000000000000000000000000..ad859e2f338b13e51180fc4b5545774746d4ac8d
GIT binary patch
literal 295585
zcmeFaXIN9)_C2g9Dk7qwAWc-7Qbn556+sZ`Ae|^hI!Lb}Vgc#WrK8fT^qz<`k={E|
zdhdi1NZyrmkM|;n>+jz4zt8*0Px9>K*;#w8S;ic5tgV;4>^-8BlqU}zIz%KTc~|k!
zp)*E@4xOYsaSU9s#_#kwbm%CBv6z^=sn|U+OEXJrrKgYe4DK0N7+4$YDc-wt=+I3+
zKNW2wYL#=sWj2O%^vz0U+^=j?M2HxOf**3f%P)lu4e?rpN%8hynbJosG{3EHzR6X#
z`%X?=-j#DE>RdIV=-H*K>%;@X`5d7q8TvafT@}vXd?7#OlQbJ4z!hO!nlo&$;JO;K
z80)f7Vz`JuRg$I_7I^M3S!5O)`Hc}9&w$rLEN@<V#*-g=r>a21qMiBxk5`V+p+FNU
z_Spr}*ENKR%Pf+hIe*EDRyS`Vh46L2;iY7i2K&R#+kvzqGa46Bx>H4*qSYME#ozUM
zs%(Va{EVNBv5s`S##d1?c^!Txps{9C;uM<mV>cvT#$+~aYW%%!l-*E#&eO%mm^kxF
zqgSiuRmm+8al$VbA8Vmmco-(XD^yFT1;t71oER}RhZa(Q8wiUEJS%eS&ERlkcGB6!
zyIL6o$J5|(37G}nm)#~`@~*Ouu)S{KoBXO&*_)mgToYYsUtPqbv9cm?^pjVqX5@06
zHQHP$c4cyWh9|PY#p&+OcDa)z3Thp9Ykt~wYV~Cp+wq;m@Q@eht6#i^y_2|&njFF8
z4BkI;gu;Y!Tnfe>^x?E7$Edb>(o~D|DRRC}T4}P`ZFj|Z?A<&c>u1^O$I=yi{p9-D
zN@ixa6l$8uUE^F|@us`O;6!-@_*dUv_pCT6qju!1{#DUqNU~m|)EB3Z-8h$dj=~%A
z?g=^7<R`%u{``ohO8Dmx%$F~a7hA>6`ONyB!q3TaDR=xc$T|#`(_d-nk&3vn^5Kav
zQd}bGd(onELkbmWb>r|ACvQO+R4{cess7@2?|n0I9YMbC-Ev<p#iS_lp<DT7@sF@v
z%n!7bdNNo~7kbb&Ttnk)SM1b6a<ryd`4kFAC}K)go3|iavpY9XEITF%-(X{^%SG08
z8jWSS531(O)NCA7D9w+Jk}b7G)Ogj<tXeIPsS-=Lh1Ew)X0K@rs#~(3mdFUl-!|EL
zNFnQYH)HL^nrA+11rzk@Zf!*q^eeOp+6=AK#tRf^h#7@{XMaj-1F1!stWs(qUC5j+
z6d6v3nP%L-OlNlfBANAP-tP*JZ>%t#I-z$*q1Qbua~4e&#}O+;b6SsfA{ng(XMcFu
z3!Q_vCWBE?QJ#Jpa*gJprOZwEZBk!)F@4_<UgeEq)zx=bLgmI4jn60vD3PY-blxWR
z%uBM5H^)Qo&~&9*<SNdz>kUKk)I0bsW{H~4J4m+&>r!U(GKWRZOubII;PK^wzmbw>
zzqtk#{^Iajz*^LeGGbCq_IuIjQ=i}>%X4V<5&!%{Vbssu>*y>lw0u22jDK_WWq<BK
zB~75DE8(}!jX4xz`-Y?2Zs+u1wN1t(+p_{bffxjG(Y9My6q|m?&<}PvKF(F;h!)}E
zp+lmFr0(8Uc04>cbUZ|*tK|3+G3zsKzeh4xPKH%KR|~pYn9O11c;z{x?ivFlv3S+!
zJFmBfo#%WxIfY#=&yWyOR^L$&o%4%7Dqnp}@#x_kg}Vv~jb)RRjn{1qQ8AA%d}&<F
zu<NOCfT8DXninm!So#bx$+xBIbAvvfy(IeV&|$ozC(iu8{h@rd_2Xx6y#K+Sa0?OD
z5fy#m^}ku!8OakTj3f_TBmIASz#sNG%Ki4p|FgCJxS9LQXU{y2zPa(gu^BMft=|Uw
zFKqeSVE<)6{5IHcgYD0JzjN4sSv`Ln?6<)VLR-IK#X(m21uFe*u-^vTpGtls{a*yk
zK_C4#*g;nL1<d<xu-^tdpv(T<8vP<>4*KY~!49&@{(83OJHHLK#~8n`(?37@ZLr@4
z`^{GT7kubHBWKSazuAgC#`sTc`P*Q>4R%mp{+q2h$SS|+C4L+1x54(OlHbkuF9PPE
zkA55MAglZW7ymZcZ-f1&cYa7+{!3rj6IZ{R?>)xYU&apl=(oWRvdS+k_uF8<4fdPf
z`A;Z`Uwr2`Tk(qwbI?b>4R(-KetXz&w&FKiu|G0?_mcl*|M1&j|Nj}xPW4-doQRKc
z$FT2sLwL`SkFCKzG$rZ>j!mSTm3Z-Yx<TxHyrLm{L)$q=<$UA3dl+bP4C-|i@#47+
zq{t3r`bIR%zM3KLgGY4#epqPd$TGF}-#L%+vGOPxn+A1q+^|5b+jif{mhZ)*7`iM3
z=Bv?QvDGW+xqY^E-)3t4h|$hvVLYJU-yh2Rq4rGrCk~j??Zoe=A{t(wr5kOR=E%;z
zn-#Th;VGi=e;l$oX%b2BcdTe9@o4Oe=V*RM^VJz=N^YYwdbPf<Pt)|1cR`r;<=glL
zLhcN7_|(b2^+RXrBz##T-z6l_OZZVL*}`b{-Nn6><5K*)1#E5Bn|+VC-!l8G{KqbT
zAt+z@YN}%GI3B8+-U%_>zjO;i;AjJg>~y-nu@#uhAD@2z%&g9;kzag&*KYrA&J=<(
zHl@4hQvQ9v`hRUP?m@oM;<+qrH8HEtl}z%b{qI-)^wVLQG<~a2htVpji01wAH$-M6
zub7*WU;%mmw=8qwE~|Da_i2F)rN$Tmj{3HDV(1G(&T3}4^^Em!{{8;AE*AfIxA=jo
z)PG`_ec$PHaGz+3s<Z3SKI=!xlNRz`+=tAnib$2fFhyA^@Mn+d6ghmDXnm2-p=RIn
z{`|H5%lqTY@xv!UMv)g>u1+yRy2U*|c8}UhO%Ofzp)E^$;JKaOzB`KUp8tmhF*<M-
zroZ7^Gao*E4Q0{${8Eb3;A=Xk3o2UcQWbkni?B<b8+u-o_Ldgc{xpU=(skvH?d8Am
zYkv)$nIh4uZRl23?0;aJeOq#n{oatBX??@n{^Eafx0FddJj$7^qffW?S^e+bT~tF<
zRIFc;_-{n%U!Oca7MD>bNk5JMZ_fM~1D7?X6Z-20{f#T`&g>wS#>yQEm;39&|HhRB
z{G;3nerx|p9{#_-GgAWG*ID?mmhe8m{(J8HzjnB&?$ud`zvFM8oX>2I1smi%sw`Ws
zJWhLewASx~K?7ONco^FUkgM80zMGWk=H42q@mU>~iRAWrn5iM&k)+wVcG>sd&wDac
zfwPbZ{-M7HSSN!!fKqiS3VOOzt!LZPdtl;q{>RTlooHb@Ru7^#bsHnN_$HcT&;c7x
zIr{ZMn=`5ozS*!{)0_@jNDIOq2`w@ju%CQimt|5H6cjYLjPx%Wj*NjdMXuMN%NF6}
zO_4l4nS~}1h<7Pso+ztt?&y0_yaFogX9|Wz1U3W~cT9b$Z}B&aARx9gw#u1tVtq%R
z@KqC0SVIO~7F(wVx2851vBG&~<j8&H{in^Q$Io4_ARuL|c|q*&t*Yb1MoaPrp9_?d
z{BQ6mi&A5Y3ajv5y?~)Bnl6pQ#DY%|y%uHjCsEn+wK;lm0Rt9gT)C^|bGgWSs}oJ8
z=yS|M!wnn=5;?(T<;GYcCQv8hG?6j1tCKAOX1m#I-NjZPdadecM&e(<X*@4FOD@<w
z&9@jzvkmv#&Rbvf+n&7UU~LJ%B~<H2MayN_><zYST*rrwkf?rl^74kI%fOB7dEM*^
zcp#y<vAP8{(cWy-+Y5rjY!2~e^#Eic|MJt?G}oxVz=C(>i^V47*qiwveC7l^%7<B6
z6aWu-BQSQ7zBB^t4>QyhmW;9y!UbCB%bwKKhOvEIwA`)nw_jMV&v5JW@e@&@9{FAr
zV$2nNp<6Vk0oxf>8VONyuI|=^CH4-PBtq<u^glaXRDFc_6t;aMH958~{xDIWj0PW4
z@H*{<<+gjY<21QQc~P;<2L#9eedXvRaAys<O!$nJQ&;~KEk>UG(zlrR*({pL(foEA
z7!~Ja`}zLb$DVeL6{qVFJx{LRA-fu=T{mBZJl{tfnce4qbENg=bOLFQ>(Hgebk%Hs
z1Qx2xehTw_hlQ|=;|Y2$Z=6%FHe<)MHNJ978+`lYVCkdmn>$NlJ`}ny)^acv%UQ7S
zm@@%FDE&dZo^M*$s|c6BdlEK;vFyUPCZg&%>s<M;i?d(oh+B@oS1x^VWv^TDlDHpZ
zU}&&bx8*%3#bZK`=2C^#2E<(m5W(mRqIyY}`%NuF1a@tjr%4gpDhTU&Q%XiTw_CCz
z6%LJ&JT5Xy1;Y+~i89evwXmAhWYlW_TvbecljGcY+2NZ3oK7ntCi93V$#cj2g`v&9
zao8Q$6?FOK!nui*xoOb=!4*eeidD7pWr?k)KAo0%!p#B#0X&l-F%bLM#3C5Fs7m+C
zo7;&|uKV%i-i$9V|E<rkuGVQ%`$9)@%T={(uUbYK%6ml?bu|HQw)f`{Fwe*fxq}zf
zdz}lDHkPUas--z9w6bD(2CW;2+6rcK@J(mSIME+A&<j*yB8ZLQdzWX$f4mF81tCFP
zv@z@=fp9ad+kiXR`uS_dRA#`)9}ZmTMi9oRqrQ309mem`-Uxp_`~4M4s-VDPh*?%J
zM{{jqh^27IK1$iaZm+4$<Utjq3PDyL5d7zu-VB+oj`(nl$BM2hK8ReaF`aHt3|-iX
zLhdkcj0JNMq9}x%AFR!EF$wRk<vj;}f6=zJXvAN@!i%kFF!z2L#J=zR@}vkvB@wC^
z<1k>}v)KU2@~c2?KS`VA?;&T?h%Qq%wZ?d4l0%}YIm(wd)QX=fB#!Dh8)K*XR`c(p
zGo+V;<-%%0u4Rw4Tq_GS+u0;oMh)1o(B!&oWI5&C7q~|9@9S5msrv>Jd&zBg*2m`E
zVKGAfxsO$-Hu&yv4tk+%T&GN;(4wc@DNij|UrK-0#AdQN0I@s2bwR*ht9-Tj`E$K<
zCOca#=<2OWRccv?QCzM$NZC^C*D>R+z1!7Z924C-5MSLp#K(wAdHjX9kqxJdhT-8M
zA!9FkvP)-5cS@2@=hrhvRi~+>D-JPdILvl4VAhhQb<;D`3Y#^*KNCL(_RU>6P1+np
z5WiVTKp+3wd=att;UgBQ?Fgjx5ta6jO}v?-Nl{q={xk?BDGr98c23P*hHvgWP0Rt%
z%Do-0Pev-c0_9x3HL>=VciyL`Gqppv^0|Yw1Ivt!4_Ol-ii+Fl>COl_mI_2YyIQuz
zXtCAk2b;+Ss}>=L>K-_%YwUWT5>K-*y8Ob8F^oPk%^uvRX|Tgq*882<Vrt+Uj}{TE
z;FFeE(!3{4f7B!w7z;LbSHfnZT)V3yJuoUNH!1^Ews4Vvx#%a&J0sCg8sCP-iKy&C
zH4DuY#%Dhs*_#XY#t?Yb!|vJQQU9_JFs$_IV?u-MQiTr^fv>KvP1`&)E9`q>&|nJ1
zwqONsVWES?<SVc#A?G!^z*i*7g_&CFu7_?HPDH+nXnD)Yh4A3DnaCKNRNvx(r)9k}
znCwiD8vQDpDPbDSSaEJL*2QgM2f}vKq5;%=rQ{DIAWsEAH_T}>H7#k>HOU$gk1C%J
z@_+YroO-9gX+8FKbLvCDNq>bxtEtxGHt#4nsu{v}c<qNvWEP6lVz5FFv)SvjuNlq;
zr9aGUhU$a@dyuiHW;GUM`krU$E*B0klPR#7)O)Xzo{12AxxUD8t$SKbCL5+YsW0Js
zfr4DSQ)HqAmdsy9ULi*|+*aWdQ2g!r>IzoK7*v)v^o-;l+~M9&a;8;Yv)x~8&&5A2
z8h^qavx_~W{_Kv7SyN;vu~G3vB-O}>4_1w|ikzB`Fl=*p!#@DP2R2lnVfJ%qZmmIs
zuwNV|UsVvZsdOHt-t&2+MtbBWsjUCwDvxSDcn+GM$YE<-wJuaur%YBaWVpm>bp;_a
z?aUcR>$cf;b3UKTds;&|j*RUn_c{DT(to?N@fS~Oxy<`*J@vQ~?l$#~W^|>2!%y4Z
zSSdy8@xbllP=VZ@v3jMI!>fHR;+g<$DJwJg#Nt;&9KSDqEl@3bXEnM`+qzWG)G*R6
zjiu%rvNf&z=smaH!auvV+1Jwi`i`le`s?aufuedu?jcW+-RYp>5pUCx?OALPp@Ldq
z`I-X~%HfRKQO|{>iP_WiIc4=_z_J6?HZ~%)uvuUG9>fWAHH%<-mOGVNlnsjPW}@&@
z^*4BYX+=1Xou+<s3aXy`qAy}^%ypds+d#x!_E+q8W>fV{udWYl_-crd)dU}63p(gH
z+BCb{nHS~YFM>(9z_(D}Yd>KAmBz9@h+(;k7DMmARrAuiAkqSt2r<)?Ri4CNL-qp&
zNuJp#N0}(zOcZ?9wC+h0FWg20(qp2vHzEA|`|cz73kaCntM8|@OKtsuH_Pul!gnsT
zL^$|Ta9v?9O@R%%u71K3x>DHte)E~f^CKBzquZyA-ilyjrrJjDEpN{%5ylz-Bsq?^
za=bRHtdK>Gqw_63uh!E_r{x#4wk_FB@b&JwJ3}V<Y=kctyXT83Y4X+{btC9ZY<2})
zrtYB5y!d)YI9EQ*pN81m5<2^=_eqpFNPdOBW*1p?%5MV9N82~ptkWX6s#U%{Gn)u&
zelHwAv1-J=GoGd;dKdA2d-m=g&;F&mhvD-g%M~!ML9F@0p=fbb!rR;d#watA<Fq?5
zMm-kcy^lTTSF>A$H81uN@%Nd|bn@x-=kqlzgzdm>r`ugGRX;o2lhL(Nu`wL52*ZXr
z&KK}?o3z07<)FqrS?2lXWi|_4yc@RRF)Ms;8za{r>Sxb+BsdJK%Ud9ntB%vUUSNt|
zs&d3^C9l>=u4nr>cO=P}j9ypamjKD}0FG$ybCq%|&E@HRkc)$yD{lqBj2*Lx#i)IJ
z*i~ybv_cHq)E0jIeI`AdqX_MB%E+kGXw`aBNr|Yh)R?SUzKws14^eXO6pgc->o|+1
z%Iq8UlqxaeAzG$b$NK=sd_VNsgh=aAGG>><d%kd>P&DyLWCjfOU5#oMDJQ&T8*S6f
zv3i-%09JvcYOW<mpA*3>r*5H7eeLdXMZ&PBO@$|rOb_0kD?+-do}%KtlIud^J(LCC
z_P9CmRu)Gd={mFp?2U$5?SHVF>u%?cA5@y-ue#$Cl8wbO=FlUH=r^!~2H#JzJj{5_
zy7Z_yy2PsHd{6aY$*MP{Y469&IY>9OL`$q;4PAgByj!@h>5Pzb^abtGoqJ{#>#E&p
zp}RNi=T^5ypU4%Y%<!7^3T-hx(MLB$tlBTs#?elEChK<Nm$p}u2Vm)5Pgb2ol?M@v
zVD(H_`e^0xd*gBK3Ngv1JE$V^yNkmW%+!e7&jXe%O>r=kONK@P|9xu_S=23;l{#EM
z0A7=pzLb=DXH<A&P!9(m&NH7M3KODXAk0)Nu@=|oO;-&cLyBxqeb}LP`sV%}|HXA?
z{!d-1kXRSzW@2V$(_auxDTa8`VXXC;eD=O(yQ!_HPe)*WV#M=mDa!MOW<%*p$?Q1X
zT2Oo8`nP_qt1dkS{mk1d%XPGMfKaoUbxvO~b!iCaXxY8JX}DGBv^r78R&=T{J5MDa
z)`waijjMmvP;_meycwA%@A1KZhI$c!y<M3T$!ks-AR^36`xNp~>^!B&&eO-#R@H5|
zURb3!eKX6G+WINGr0&xanhcm_^el4RE#(H0VN4g_LYbveMEC0>r+Ya-hnMt6`s(sF
zcib4-7)cfp;sR)0evUnGNNK5>^uv3r(OU7v4|IF0L3c9fmDf{E%?`+2Jgjeqj&|jY
zDg+MYzv(|l45y>bkl(KLn3M2A?k>E4Z>hr!^(EKYblDkWs$ed^onnjRiOW0loGFf)
zhU-aV<14P{NP$Jz@KQ(Ne6tmJv)#Ro1#K3hCu&>|r#C!l-RQSb^W^H1w{tXSyEAo1
zsuAMBj6AT8+MTV%>hGQo+c&i6>|{w91<3{G*VdJb?9vSRBJ36h*^(YO6{J{Mz_+$<
zVm8^u*Soy-y8Pbe@{%av+;Ii24S`L>=g5~1Z?>AvJaV$TyTIzp8?;U3>|_Q{V~oQv
zNgZ{Eac1F++t#-i#G}1>Ws}|44bSsbH*f?5fFzKW0<W>(-dSz2dA!Y3n;^xZQrrMa
zFIjSI|A)`cDugXU8%2ZtbM;%-x6>9h)v+>sL)@`W6M;LlZK~bb@=e<t<jhn9+YU(h
zWbKdx@wJU$q3x*`Ig9Z5h62@s%Px{N_~bjh$4-~$O=+W1^EAMy4e!yjKjDk@Yw}jq
z$Zf&2oCBJJGbv%B05sVgCPH2_K@L~6=**^)_R%6muscd!Ck?-x-ki6jy#VT0YJ11^
zB-xlS!pAp7U^WWVPdyff0(=1S5+v_$&HVcPR9}fcAH-lX=oS09G6T8Tl3mp88vATT
z1dV+!{Jz}rG(RR6{kp0C_1^c{deEb)=b%sb4<h!u7caB>9C3;z-09J)Nn1dMsH$o9
z+$pRL&GZQyE(FE69nv#xlm1Y%uAT`SKgG$5bX|RqCL37xpt4Bgj=oIz64woT(`2Q%
z;X+_-&PhW7nkO5sjY5C9c=C<2AN7`K!DotGJvD6eIbQhsq6bj}Hnq(G;iB*Nl1sm*
ztnW$G_(TswTQ;(<ujhRsAQ3W`p9a1^o$(VSI(Z@l3z5O+H@<QJhxTanEROu};g28|
z(b_aXu%3AQ>6z(!^}HvLLY2uXP>UPz=?1C#WTa~>`ht1LICNGFHZBUmY59g7W%T6~
z)x+npePzVgOTOkl{v=n(*!S|y8Ob|mU$)L>x4Yb%KJonis#wBFMREDVR4eaNE?s$j
z0?+&86>VpZcanq^FUb_I63{1@74Eq3Lb_x0zeWs2WNv8nXmL$K<BEn1b6L@`PR%Ci
znUs8kO95do&wfyJu=)N9CP|}d_UJ}jen8fJ5BbLE(ksSyHE6AV4vjrk?=apF%hTO;
zzM2H7U>jN8iuLY#^`1WO{{DV7!<d2_rPmCednAnJ=&6345<%WJH#a8b2J;F&%M~Ae
zcI0+9Hs8%l2wBdr=&;u2dC$PWAT7yevb(Nuz(QqdtE43}JPo0B8-<u`SWDPlJbdIB
zt97Zr;EJYCh+W~g&CviaFq9Bb;ZsU_;Aa>|6UY4U%w8Az%693tSjnsP1E|>eYP=fy
zdy*c*7G;^E&k5Z^x9>#V>CiS#Y$+V9A1S``Sba9Q%p}@jnB&R-4PP$rRr$ACH|?5l
ziCjCgj-1&TjtbehlZ%xg!&-PTNF-Qvs??ux9*I_@BcMOP5B@aQ6tBB!_O5DkPB4dM
zc7^>DyjRhcG#P5<1u8c*y1Tjzs2x!8Z=v}!Gc(Pn!`m>l*kbl1S*XYdv+pnT1B$2v
z62mW&md&FQm9>%f12L&1{8ai_G~&6C0k*89WD5~&;#4#dnO+CbS&+g*?-yf9rzhiJ
z7Bl+vxSD^>qwDxy+wkLTlT2WDz;feF@yQ^r%biCU9m2P^T3}kaFHaC28-`(bb^8p$
z)t|nd;VF@yUF(qZGjkY_?7HnAA7kHt`Jpz?Dqbe_yUZGniZyHa%;_Gs!roS8^WgQK
z&JYy4K*=##f#d|JDudq_9@X(*_><7RZ+?7?W6=+QH*|IpHLF9sy*c$H3~jj_m*Ra#
zn=hWCjdg1&2qB;h;(%ip9LOhQFe5(1EGAekOG9<$jUk6P)<hpNp_MPiZ+T2n@6R)^
z=U`(9`bklUsL040jCKG*>QjfWJX-Hk_2Y1}@>JFq-uxz(I&yUH%M#-U@kqEsG}W*N
ztn@ysyizdcG@B<arb*3ht0lzoSX<;HoIhSmLqbt~>8#`=3<SjfLZ~mDHW4=rOlEp@
z@j4jAQZnZw$4=D`b~In}qPh8z*J=46VD-aWCh`3I`~XR^IH7E0Wb_)xfEYrrF^6ec
z>r3#xI3CV>`ELK%c)gCkxcD^O9^@H^y7X*#SWe+9;%;@r`5>kWOOe`RRIb7**!syw
zBh_B>QDNg1yNeIHV?ScGAtKr4ic+872yF~z&KFZ#f?hmbc}O;5#VzgT?Y+-@wMcXt
zqbR=+F%x${t#p6rKIYGDOzSe2W7^;Pe$>8i+(*1%tG@@XsO-vFS<i23+QD$)YEO->
zF~>++CXHHSJOBA8d)fL+#=HDHm;DYZNPil|x%^YL*P?jP%ge}7kmG~1OFQhO*-P{#
zXzlFmDC%N_B+CLGYR3pWxkcUrtrZP^m4rXF!u2Gs!(yJN{QzX+Q6Udzl#xB(6LATR
z^qH~Sz-(~VQBx#GTkC=;sRtO$LQJJlkYS+(^TIxBg_o=h0PHs?roG<#oZ1{9PI{uK
z_RTqQJ(_6p(RQ7q+$NohIo%qjwk8>_exAe-!EP;^b=S-0cJ<l@-I|s|g`<iHsR!-N
z2L1Wdi5KAoh{(uD<=yV?Bp>N6b@%wlNLwkEAd6$CAJA$(QtR4l`F+k&p>|2CRes89
zJQT)qu03@dgco-#3{zh?`K|H=e;4wzU&lh9ib|qfEMn<GpFh(BM;1*hiN24It8Vx=
zSbdy`oJIvUgzmt1Bj;d6J#Vq_vJFJ=(ThycPakN(c11=!MRtbYld@XXzvcX(H`g0_
z62%YQu4@|{)VeOa*CI!MomP!G@R$v)J@BP9pzB>^<xS{FEqMi1zMGa03|V^A=!UeO
zG(~pxy~{1k&XfuNcFA0;`?fdh1DD6@?dF{I)ZWn^PY=M8f11TzhyEid{pX|S0kWb9
z*RocxEcQP}50QtMq%sV|(#$idmeE(bM%Fz$JU!CB7Fu1t-aUXXcAS`f5pHtB{Vbga
zdGtM!>(*l=AbXPPV+G1|92Q}QaVlH)gH2DK6argwMXLSV?LGC)eSz%iJcO{+g{Sra
z=-*0wl$j-6G1Gc&N&$Dv{1NXsnJ!n5CZ&Me#dDLJ(eH+{+!pH>DC_K0c6bsYUXYoQ
z;R^~`f{BunQZTdZPX%iqb%rF9B%w=iiguovwi>eIOCXrH;yiyoG<70`H&86HnUa!n
zWXOSu?9?tGpHJ4>@;>VQ_&Bu`myr^@XCJ)TU7T|ss=&rTAC9xo@Mb*KeXGAl!Sy7>
zlEis0?=5=&Faf3F9fAXfD6L9I)6zQ_j4tq)q~^*wfya{Nkx*?kSeyFn&?(6%07mA@
zHtf79g}mC&U!<jFNw>bqxIACp?v~|kclLm)`@^oSWk<u*bnl*`<o@_AAu1|rH06W8
z51@Ej0_ka|WCHTN$lGfZ)zJcr<y0~<GWcT2vN0OhlkyH5!%xL!RIrdNCl4jhc?%h$
zkukgacID8>7X5Y?6_wIi27X~uw!Jw;UQP75$7hKvKU?+>cV4^do=;Bg3lt4=s~0(S
z9n&~HPSXT2XK~#IS>|rq#CYjVR`Zcp`Q#T8H6D1!udUAXSD(F>#QyT9VEF0A@l;2H
zI3_G}Y9>-8V)-m}qFq-u^g3jUynNWV^&Z4*yQSF=(QrTd^o%|3v&*84EYC#9wQ;%D
zrRn52vS{g`=*#oNww28*M-ynOFb?+B*7Gn!+Dz9(AGSiZwC42mdDG<QBzve7QFVNJ
zqpK|0&3m8t&w?JJb7>x*xjx4H>+S_%LG$Mx@5W>ApP~q2C%CJpr>8!lsG<^)hWz@~
z)+&`>`fP~3PGUJ)&fA#a-w&wHOjY6u%;!TD@?1v!_(su45w30>m)&0IMt1eB$sPWS
z&Qon~xeR3$cGt$g)sWMMnK&($b8&*0QmnD(HUjAnKWgD)RUy9K{j-li=;{6=^rr6v
z2heLzs9y&CQ{8wtghEbe{XtrD*|x`tR7Qq9$=KbENxp0Gp;_etNz%DQw4YG<*y#*5
z^!*N<bBd~b#dVE|IXQylT!}JITe}=arMR-QVwL(tP958qSNGm+D&$kMz~VWP&B_4X
ziN*-=^<a83$Ib8eG~=YIz=LdlCdWWP?U2Z9;c)%F%4yX4>HYQ`OC=C)$rgkF9)=#X
z8kej&C>+iVr@L)T!cA@4+9S|z*JeHkGh~u3BZp*xpv;?5n=V_FHQRj<K#RRoj@hQl
zY#|^W_62<b`OsrKJDE0f?!9XiO~q|<K?>$4U;9Tag{a5p=HBQ!QBVN<=nZP&TxjSx
zR%z_bWxG_Z$Vf`18iIowlYhn`1m6QxEdN^*uW_8~^tszmC{#Mv!z#1oHS#Vfv#tv^
zvjvQ%Am_^H?XHW^YUgD4rHY57EWbWmrAhjbp7GU1he0b<by99Y><*z*^2hHUvvr`S
zh~!IB4n-mIq*2|wi^Rg4*JGV$BDyC#<XT`dU+=F=y7=8mt?eM*3rKSbLa=F5W*?Xj
z$U~y>b9g*Fe>|ZDql(IO(0$uu3?*+_n&?VQ1Ac<Fe=*S({%Mh&RzMviYC$@Q68ZSL
zA5beb#a01D!m#!9RlRSY`g@H&5b4ck7QF_YDtoQJcFyZzXUNo7mh5w2GNN1Qg|w<0
zTCP}(QNEr<ceo>2E?{|dmC5MlRD8BW!fb9KK$kH^pqMhrL{bwAtw{=Qeyd{34ycW?
zgoi@k<KGXyOXKJF%Chhcs`MFNRq9W<H~A>N(bT!P+PzNr=iWPKiJF_s^f85_Mm+Pq
zsQJ0RXI=Qfwb~K(@AJ<yQA?5w+(wUf#yMc#98lxzw*WY+CkY!i4;F6>qc!7_<1oQ$
zfSc0|*M^@1|6_n=MnN~F0Y#pGfJ}E7<LxS&79m;=_@;FRd^?RHZs7>AAazqy(=D{N
z!;qbBfRKHYI#T;Q3Q;C4Y%@sj^_ecf*9TaYX`ys>*l{{Boa=}pwI2l+KgbK2Y7RpM
z8ah3`cHQb_<AzPhQ5be(usk!E{`8S!r@l~q2-^D?+Lr;!QoTxbfOtui??i%3Y->bn
zV#Vg=68ZOo0l*A{rmZ4-ooC;+yT{Uh>^DiTJ+G=oa{QvBk?Tr5lR;cRm1Q}_WXwic
zx8X|7JYce6x=8580vF`#FmxkkbS@7T@pXBv14a$QqC-=4eF#dwJ@>Kp=;<Qz_u;^p
z+t%pVG)s-8OnKkx!#2Ak6?ZmA<!S-WB;`Z8n!184QF4SlEj1;@Nb1u^+;EFAXF7?h
z&UrSIl<ppTj=XZhHlL1?QiY@d{@&f5It|Kiso2oly%+y<fl|>+^EGtZ*F(!49$o9&
zB;nQ%kLJt1?D9(S-(#f1c-KG?Qb9T(&8@1>*42Ip5Dt4UZhQwYpIe)#BDvRKHrjs^
zy924sZD3c5^e@W^4^_QLWtv$a;S17HrTU;v%PyE5Zqxwus(zi<{1YI#NI`B>Ze9X&
z_PU%AZ<eu)#paV}<j@D$fVo4hx?v0{E>LC&d}+!Q%@Di1nFJKPefm`w_-&153b*jm
z(@j@YeiAu~h7%~;_DLQ)Fgb}*;0Fm<I<hj$Lm&vFcO<`hR`C<`^=E+$cyW0E8A}C*
z5+-Na1ZcN$rDP(?l9pIRFl#kt{n^E4!Bups;UAM3;a%H@Gn=mNq)c#f&@@>$zo|g=
z3!w%q<D%D@OCQFANEX>P(wIYFFSE5^r9z`0Pbbz@E8L?Y0~04x!IdvU%O7ZWFWC=T
zl?5>vO0K1+4}B;!6?rAGCvja0f5sAKQ-AdU$O!rd+B0tj94u(nUNQVU>iNl|ZHq)#
zUM)j6h6EN6SXA*$nn(@cX{kX{*1nM$!DUzr#3~o1g4~BOAUikd#z9f{3w*<MoNvK*
z#<OeJraP2sCQnB4ukVI&ZUoc^C+ibQHP3EMVsDtne8w;>0qfL+!}ISA;yNdC1ZF!f
zU^c6aABd~lZjSnP=*}1Pu=XZ>I6M@S|Ng^jxBtJQ)g8_Rf)y!{{BREx`Sat}I($kh
zDy7ViIXRbAv<#oOw=fJv3vh+!n^RFmwR@npi?_G8l}(K-EM(6!?J4a)Jvq)OpPn5)
zqU%Xh=m*9apTQ7ka-^0j+bdX>pPhHeb~Gh3lZ~nsH?28NvzGZ5Ag$!OsRd{=Eq2qB
z6ZB)L-OFRaas~8Ny-Z0DA|jdLP<b&iF)i2S#GK_?O<bGY4NsfjHBOG92GLNWprq9C
zushVUJ&q8K>}!@z1-V=<mbz?iW!!OhouJRdltuILt8t(MhtU=dT5IECJ8(y9>8BzM
zJDW1YL0BMrfVV=$<Ke<X`1g!Q`A2V9>XiCCYOl3>!freeHu5|(U(e1#tN$oby<jk5
z95V|`nakIMI*lJZ&6GHvii%23n*>$9aIi5GS_y}ie5p1Z8&<s>=W^=ww$}or3*Q^Y
zI`IOU8<|sgvCKZLCEvw^h-e!1#3!baMB4W?5PI=~c`?>;wW;VNFKoFcToZm2VNPxR
z<@R!ur5l}uzt;OlQ6O_(g>O`Ji~>+rPbu*9kvW(dW@`qC%taT#(@%CRm?rzpHkSYS
zuU31Z3|d{b1Jd3DJX8HT#TQ)W>(R1N0pliCnIhf2Il6TcSNCj^JD=~p`j2PaE*0(f
zd3N9FvIKj_noVj;mw0%1c!R;hU~FD3eo9T_wGtQqqeO&b<-bsj*xrp#N@C<4e7f6h
z*bXl#DdD;O2_!<Y0KpsV7ahNScf{@z^S9e+Mm4aP<u(AKXVqtuCF7&3bgmby7LLAn
zX0#5`S_s#&47y8>*co?XlRD4t4Q7g5-P#WGlH0fztNN4sk<r9(9SbLT`!*o>Icm9D
z1jEzQd)?6HBRFzQMOB-@@PVpI0@8+EF4q?|`#>Z-Hjhv1>+WJ+i*OM`>Q~ZvDKaM!
z-Rh;y0c6&j=*MR%_Ev^9CQul*JNgPA{>7!-g)hHBHD5|BKSM{ZB_^hqmFBx*ty7%;
zD89Bf-{xA&D}g-279CJFvg^f-wAUpU<U>$_(GC6m`E4VAHTb`-oFlt%{gG%!VSid?
z%L`)R_<~A8j`DLGgEo*s=h^v8$T1Fk?A$xgi(WlCE_67cv~<+^Zo-hXOxf`Wy|>S6
zq}dAeK|&$20s@1YlF3+dTGyq=l|Z=B=*N;>Ub?P+-SOQs`qU~4!)UT7^TPT&JVOpf
zpcR1jx9Ie@Sfdm$n*0V|JBz2@OJg+Oj1EN@x`_nQ6J6X}8mC(D&hI)ccgy|&u|FaM
zOl<g0CRCW->nBv_*L!2no>+R79;wyANO!ddM>>RBI-A=O5fPa$HtarWj=q)ZX=HLg
z{M!{fPrL*Ar+;Q<{j+L%eqAFeA5Qz=!X_4KuLat=WllgwvIds;fLTr7VCXfI_60UZ
z?NC#gRVBBniOEQ-7~(9@x;=8;C}XNlQIKX!VXij~qSt#t1V4ko?m7Xwj&D5$kb!s|
z3%3Ta%^0hzU6ZT8i+t&~?!oJVr2Tzy?gn}N)4_%{LYzy~7#B+Z)g|EcZ&06u6}4Xn
zwpa|8+twc|)nx}%`ec-ergs0WA|A=DC#bo;)GG}<ZiSUTB_A{1-G1(2=JEAkyZ_He
z{_)3Grt8jYQ(%s=SPe$eG@9lG;xq2UPzT+aUNiR^dX>er-HOX%Gb7cc0p?N~%W6Bb
z5SY)4)BPeq0uyV#jo*-#o~{S>*%yLY$+#eb@haY$fYpX0%J0$FQ#$v99v8*c9*RbS
ziQj7wh&Rx8`$HM)Dbmu?`a>P1_w27b=Gr%`@Ajlqk;z{lT!-}y3NKz49*tD+$1w*w
z${lS^^z-dm4!{gv)ZQ`D+eeAUDMSWbJ9P4c^Ucjd;x9+NyYZJB{j&zSuMo19&D~fI
zN9<ff4Q+gK{O0a?^7Kus5f^;D75sB~u)fyo*@@}5w@O)zwZ+B7%}S?hGg~ZS=%rqR
zaE>~ldg-rKGYAdU12xoa_r0r~or?0qhvERtv*tCnFGsYn_H1ie)wD<iExqFt1f8LX
zv~~#=3LsbAV2MqhT!QwrFnX?Q;fv<pHLgK$d%svxWWD$5A6w<F!~BM~QnY`JAm*{8
zi($kTIh);+gwN))vxJcAL2E%MytvPFPb6>x-MmY6ZFRJ;pbtrFHMjxtjBH12LnqD2
zUyUsa@GQ%hp8yP9m$^#IZ=0hArX*odwyv%&5a;A|wN#cgU;xu3Fr07PvD!_Z#`*-C
ztA+M?H37;C)?Y6`=$bIb5#uB|;?VoJtax(-DWN^|?M{=63?QTDlG9i=5R1j53v<Cn
zsv7F*6sB1v&SfN)dqc7nhR?jc*q#~vFA(x$vqTrqNxB@C?=Nn?nm|Kmu4OoJ!@&Xy
z$ea6GZKR~!K#yXY`XDIYx+Ed?ndn~(_v0>y+%M8CRV?maDnz>VbWaj6MYkKQQUdgf
ztBCk@n%4D!xUKK(X=zZ6QRwMH8-bMO6)Y#rc3OB-lcJsizMhHoV#z8_kOPP?zGHzQ
z*{|WIlla9y))ZJ}_|jn<vjLi@U{3qO7i}NziHpC=zGRT6An`9J7sZk;!S3xYMr(5Z
zb6G$B5al5M<0%FFY(41awY*D?$lRM$g2b`kt^yt-z<*b&pNE&X_8zSaixlJ)f5~)`
zCgH36E^?c*8k%&LzliYd1W`~v21<eYZwLiv8FrmFMUznjd0*QfuD*(o_VDm<)eg;H
z05fYp&>r&u<s-|DZK3fH(ovui_#o$YTX;S;Z+Fo*;O@X}Ljaw(#)5IxwST)|{YivD
zShGsvPusRD3kW$khp%jJ|FW6!MT7|m(Zn6{lZh1<Qg`vS##iEmTx8R-I<^yMGPGM;
z---3pjKk2NFPS91dV3y7Q^`hag<?yS^9k4$9j^Z3@LGJ&$$Ys_!(9ZF5`G%YV8XW2
zZ^wIE481;1laiR5P7(ugnMdfB&D;BvLtJm^-3QQ8H2UdhP6`9)^rd->Ij0m4|4aw%
z1|nIOEjs%0az+L#vL_%YiW(eN3Z@Y<@Be9Z)PVnW111SN>>~$2lqp)gX%6`aRr@z<
zHrHUeuk#&_UF*SD_F?1YF4({gaQF-9+?A5yNbA7vYvMei@$as(AH;b5V^`gU*;+9T
zOF+7(5T0#^v$w!R>1<X^5nSGY)5W8P-sUQ|Md3F6wNC)-o(0N@2dvQ{Zp$^EH7zjA
z2eYQ_2kP569<@&v2mr^;=0!FyX+QT+g+PTgS7pO_cz9}new`L{|CY(<_QOB?5w~@v
z_rXSH1@WHVd-2ajkr1ra5-NX%wkh5bFl>%i{%pLfXxGhdG}XFBKzY$gU#l!E5gL`T
z*eVv_3)rELDgF#bSG}dL|MgaXUWu0@%m?Cn^djs`J<USd3MYua=&H<|T9cfWuX;A!
z$*EUC(y0)Sg!_0&nE}9p0C`3_k9WW%o$Wm!$V{~%*B+F?qIk2o*p2J1PONscZP|8V
zHgIi{3}+t<17>$hHAzG!78d@a0ayzyn<hKssMmWsx@V#RXWqoWyBB>dcrWb#(}~~c
z#P1H7tQ6d*3eOI>V`p3^Nu_F3?z#?^*#t61_~5Ig&VcbSO<ucjFxxjxsAy%dOU9CK
z0#fNB&<42Togkp}4ki2dl=)N{-s(hG5gYp!$vQZ2(QiL^-ZH2fzvLp0#K}tQ9Iu&!
zB<;4l%~THH>o|U=9Rb^Bsv5M#D<#p^Lu=e34l<-;;y*HEy9yd9=E)xB9GP9dqA}CU
z?VcRH@&a_2h(v<(do`%;7a2gO7r;-$xcBOhE%^uv4xYAEk3Wk1$#|4gf3Yj{Tw0U>
zAN0A0wbu2#u^>7<&^Vuqd}0c<HVlO9&J^J|mF;@DW`FPmxtUVh+S<yhszo%DE=rB~
z!du%4`wrvm&0DQ#e6nE|m!(S2GBX8mSmUxagLn#<q=44FqQfRNJ1)F5z?zXC0ouEj
z<`3zldSfvUBA5Yx?y0?re1uBbN{s2lUc6;}7oNp6!(KWSKv72{^!C<j2RH$thf|;q
zwz22`wB1D~C0koFca9y<!poP7X88n?Eu?ahKV}_N%^EGS!MaRFYjH~bX$6)EywKWO
zk2^zlDQ-h{J+ea!LDK9#%cBrUnmCT@>S!sZ7&63G6TbOlz#RJnSXsz$TTDe$9nJ?F
zrd~C3Pq)24B;sia8e?Nh96Cb@C^5T>12*?*?RzidKt*_-u(>!V`qlXT)+Jm!f}<ZD
zzV<+1O5l8Mx|5!>OCqSZh9A>Wh~TI*r!HK-gr+BBmJtpB(Y<&lv4#eKj!hr~<ESm@
z!K7z?2Wj@~ahiPCKn~we_<I@*#@E)k^xgJe{WB)WNWjRw`wl}Nv0>xez>4=T6TzXL
z2VDbe)3nK8B-Q$v+-rQ+wAR;lwYleBVJQ$h-PE+|&K<jx1&X7<Ld6CqW8IiA&S|0~
zj8M(qAArL5ne`f`nP<Uqb20jCV{KtjUJ6meKM?9(7x}XfShIAO6KNU@285hdTdRC5
z{nUISZp0L6nq>;<L<w(Q*{WxvMcNNW@ELDt1#^BaC~`#4JCJGJ(S;9KpsO>BhRML-
z9e>nCPqg3WF;-xsZ@$sccB=JV&am_B*n;c3$VImeG&tAOLB`DH+S}QA4%7_ex%`fW
zWdMZ=c~eqSUW2au8U|IwqMK9WK{iJ}KkQKHe{}Dn<Ad?0En>RsZ*w2uU!w80c#9Q$
zw^R&49f-qCqyWB>=_+TLP;8{R3p7}ixPb%DhVFCJuuIF$)y6pS{<3m<0(Kx4h`$L+
z^9U94<|Pjzsxar`vF0m466RUFP)c=!y|y1Gcfm<fPf;k617GtRyHdy<VBC?EOh9@-
zjQ>C<fYu4(P~_M#*aSVzO`xj&d~>1LGTd%mReQY%0Q*-NdOJIlBHpE?mFlzJY|%HZ
z6lWN5H3=ttSghD^SN>kVi6dfJ3p#^v+(k9e!IDYkdXG6ywO(dJKNovnSjfoC%={rO
zE!PhAbhz9j>?P%@Myy|E_G&<Iu18o!<5u=6vc>1?*RL0ZU6;cqmo9MXVoc4=&C^Ok
zuY3X1vYM<x?C_H)woz~bt6n(p)qNyrS$p+^Ge?<3ZV3hLF5d}=;1cmeVAdDeTzbQ6
zUlLdryb8D|l*aJ5%2E$Zw9MF~^M>U_cJnhUes*|S96_su>^;EW`7bRklr1KitS^+2
zD%IFA8m)|7OO0eAf672^em?lwC6&7Q=j~b_b1v<GfwiAqj|`0b4gUPe3ih1p*PvO5
z(FkH3RJ)p=DO}RgBnR6@0IB0-4X#0&%WDbOt2rNTSrTuO8s?M2+&?liJ39&ng?_+}
zHy7aOVZin<_0N!%_t~^)k2)aGG7Ya(SFvlFz(p8k?uT#cRjTRiMbZE2lmFP(?WBLm
zD-s@mufvV+0H2A*b-H9(%=OE_IJE&V*N5G3*<0h)n<L~7jT$|HpB;v!iY6o1w6mc3
z$+6C*DXC0S%a+C8C%O$4hy>0Ef|eLe$b@NeQ^HxBlJT&4QI6@zUTpJ1C+PCf^F?wy
zKs;|M^DPj{8<8m0{)W!~CNIz+M|STn1AKYwp}K(?YsuJ`?peHHps#lv^&?o0!wL=s
zMSrLDp?8%cxkHccug7dZ2Yu2y;qZj3jD_rVr<Jh_LZxebdS7m(oZO4>_0PVEzf9<T
zW$(2=Lr_TN+wO+QYaox}>{`PS<8Otb&aCjwk<<|%qWdI-faERiY`W;W6xi!D8KYQ7
zPNnwX#&ZA)Z`u#7OV^wZWX2=}`}=5Gln8m3JYt-_8G!}sAKHH4A1u%XWcj7*7JPvE
zP?&wiD&>?_x8w7qxn6SSx(jMf>$^{g$j;ISpO9+@z=$7PGzG(}Cr{540$OhJd$dq3
z9H*k7tsTRbP-P(X`6bZoa0qW3^<RNfyKVA~z9jWmPbmL@Umso`y!$ynz<T}z=*<ci
z&L+j&Q>_KZN^uNwW%ncqjElJL$`y-Ft_G(SwA+or-Zb@ry}@x`fbdow&27l~BJ_@x
z2dIEW>09sv+>e;*0Xa9}X>unS3+|`=#!PiMCLA}yNy*M`8Onlw6C?IBEm^KynD;5)
zo6AJQ33oP7rrN^!hY#|?s}H}aBKSpu2tC>bgUb4&VVK2^m-bgDzyf@6`N+bh5}#b$
zy6s`fMR1@?YG}QnM~BuDz6GdzT|97Z)eNX2q-wb9DWm8=0nM@(IOjlmlF{(y=C>mw
z<*ROifzg+z0ShP6)bCLE8E6BM)|^8$eQ6<kf%=d1_M>vpxoZ(nl02uE+95|28_p!G
zyz0dT9XL2=zBPhv<^&9Y&r8_6Da4C*cj~xLh(pphZx#$CE;$d?0DAGE@*U57nngI8
z-B_$lh`q$?lPbblerL&3#D6S0bOzLtB7N0GSe$LUgKlSvLJ3pIdq#-0$9OpQL4)7v
zEK{i!>&+|m6hT>&%`zJ${PX!8;N=8-N{tdH_M!m5%EXsAJuFQxY2Z69bkGs_B^Vw!
zGA8jLlKWcF*hxRX`n;U+zBAeIZD-tUXZh0+;`8POU{ZU}+N$N_V;^y#vFPmXzN?~Q
z*Pv~lh7<Yu0l{Wt|Kxxrg0$WSLp{;L5TR<FBW8WJy@?MiU|Qz-sy>Dfj#-E50pSbV
zYKnP`oV7!YK08jS;EGwrazIwg?L2AhZFg4V_{RK<A-D9uyy>+>ROn?Ebo;s=IbPV9
z2<Sj<?GFApBHqnAv(y7cQfOf-epGMp$V@smuu#f#6nCUGnmkcxxf-Uy{$_yA4tf6k
z`J4uJh;L?L|CM7z_aB&O$ON@WkK}!2W+{p-`#)$5JG}bvC2%&87@X(?JOdJgI&1Zr
zpe}!*V4SE~F@io30czx^p$Ho32xh_LUXmyhaHIuo_2u$$HbvExM?j*$Ce?~G8kmem
zZ1i7PEMK+2<%#0&9#MpLIg(I7%LD#BE2o~3ss<>sG*F~m8)Q9ddg~W%S$m21kFx-1
zWDj71#WHZHNy1qxJMc5IIL<5p)J^=!C)hHFd6>v*qfO3i7R0|VuLDW_?we;k!y_OT
z^?~e+WZ>p$%}XzcF%M@<&jdl|fCd3uW8J{k6Qr!0ss<!bnsKukdG;Ck?S?k~@1E~%
zuoFo@^P(KIeSo|}9?u;dlfIQzzA{6y7LtV{@DSs7UdI^_VSpyUYs{ZPd;O1@@nA7p
zKyc9gxEq&6#X9PA5v2G#2r$Y9*1WaPHK_r?K3|(}jwUacA+@7RWjNjI#Ub@W&*585
zFKl?r&<kiVY$g7p4l%<K!!*aKm%4PLy!IxHy~_oq77RMTmzp*=5FPVn43KuDZHJ#G
z=`+ic@h6!CBf&?H&C}My`vBQC0@R}UjIy{%O{Uj$9LVOV$L5Q{3EyFxRroozl<5Vb
z=T|4*`(Q30v6zS&?rGaOTCC&axx#^F@eMA3q`FHm>)F`X)u}~-m`T`a-@F!#RONI(
zb^(e;;_@K1^OORodH>q;$s&1C>tGM?xz$|7M(nFQqoxo6fP_XDo?C6<`SZ_w3=`V!
zSb3H5QzIdDH0^%7kL&?Z^~|N#COnYc<|iohPm2IqP<Pojy%&?)N^l--jKg(=-zp!c
z=ml>PK;XdV#QgYZfHdEeVRz*Nh=m{ZWR1eTzdy3mbZf0c-3lPv0%3A)N}x79MnriI
z_-JO!bVqXCT%Qe@@}w&baLt=zZ(fXkp+lS}WODk_mlCdeiiV%tZ8ZvVcco?p48D~>
zGb}>KIX@jB&9Q4Um)k(Dx7x%vKdm%husRF$EP#d)1bC#oo1O@~1|aS<8mK85apTyi
z?W7Zcfp&0YCE8m7TDw34VOvnVe1Jv9&f?JX7(3UC;~b8sO)4CXzkPCN%RWNql$n|7
z2NXD1mxnl&IQ@LkJuuO0jOsb#=eG)(Y>Mi!7R(;M5-X?4F&2U4$L=;p#1-bpAlU~(
zn3<2|`$uRAh6+x$7|Rb(lQkM+w9+&um9&Lm!cwq$8Xa(7jY(w)gti|4e7!h5n4S0o
zElv@fk(|6K`c~ffmT^OkdRJdkU@5D*1L#_#AS0oT0uhGRXVpf+w1AWikZ%&u-SRL)
zD0m4(vbiicD`rw~e?m2_?)LHdKeVwsGpYzbpexJhil>7znd08CfO{vxMabsoNkzNB
zK`W*bM{~Heb}f3$CM=|<Q@JQMx!SVQ>3(zD&MG%j5BwD*>$tar*k8QN-F?{k5ey2S
zZ!dqL8y%@<it`3TKz7{x)42^mee=E-(s5eQP?d-NRm9jQ;FOt5!2(dEUcL7yH#u~U
zh`qPP4ABx9!hnvdAJ)t*D*2|a!UJpJ>63ndx*2S!r>E4QV+x&x>*h}K!?bNQHqb%4
z)%uoJm9ZLz8*L;nK!P&-V=xYqLm@bo>UIIBX<KqW$13c^EJjm<#zAyI$F5KvXeW|_
zvN8lV5aHVP5yj(Svl=AXgA|i-STs0BxojJ3;iUoiR&X>ingj<K&#f+A>8k>I^v5}1
z&|M(tSt#szied%UBi@J*zOv;|`Y2AMZ$K79!$|NbD2jnTn=mU%Qb^r>FIu09|MYBP
zLcNK`ES+L+DgBoRp1TZ=qQ+H&p%)1_l@A+)VP9}Q{;Ziv?l3;13u;^rPEG>O!;^y-
zDH!4gnyM*(<>B|-+JN*e7`;1!IoL>I_=E_c7uNwBUZe|(NT%0-55?T*v7#UE0&pDr
z!k}A4)Dz&x+aEt(;+9diz%LFBEPI>dtZ5c>V)U(^>}zns*D^4b1J^#Swd-BkYIhEL
zbVZ6gkUm8&Xhl`IQSN#hhH#mXM24G>iGV3mlXR9ziYZ7RTOU3cJhm+r<mE8zR8&=e
z2}mVqaTSXFhAKW(`VoK^nk0pSWC6nb>^ji(_3jqwpfi`ZsM$)!NZ48!^g>l?!I4q7
zYv6QEM|X<>lH?=hUO24}IFdKh-wz(-YcsYE4pK|Cdjtqab~~Syww{bBEiDb3{3gl%
zfC^Ny@TMBky_@tSfBkTB@AHxiy4Yp`Qa#XAEo0D&MS>!)L4U>#gn1o7C*rSxLkemQ
zHp{qvu)Bx;_LetGmi*jxs4zSyhZ9Frk7VSzwUxqp-70b77FYDZy;{($4F>tX%Y~{)
z4fAnxR|D383TP&rx~tD@9lN#GYFXXlX8#0!#_s2Z#kUXuGH6mZ75^9T@yELV<y3!O
zskr?OyoSU-Zf9(A`7<Gh!Gog0LS9*s-A>7%OSWC3VVV{`bOAfEra^w(dj3E?73=gw
z3rEk$6kphDY)~zja8q;A<hFz4tpE1s|9lkp;VVl^RinTt&0e6VFblkv4d%R7%QXm`
zR^BeK)S7(oX;l+|JNFVUPJm;Hug^bJ6SgbgH#KP;z-M+w<24?17+{b8|7^b>xvG`w
zC{8NYYu8hu2P$iQWc2q>M-Ugm`4qh(dPB(_(3DKxock!Is3aS4gU2po|8J8YP22u_
zCBuCmRM_u1f7t=tYFd@>BLNMR4iUN<bd;#%;$$=+|3J53t>U1*1^Q}AoOJIG1r#SB
zqb2Eotoa|!!S-8GaazrQ@&C)p2b#b+<Wn5hRb_bWl)&6a?EuN<H}NWf_@ocsyw5BG
zwAY+Ji!|MZ=>hFZ%O4ST=jTNn_*_4F#vevlV|AzZs*Es<4Ruhgqqey_yq%Rbi0fjB
zZkcvxC|m7QT!=zIoBS?2T59&k4gQ-)bdLTcPQJ?t21bE4Iyf$qh)+wz5bB_en{<NL
zERf)PUKDn9th{jeXNQRAzy(xzb5g;7^J_nzUr6H7M57#d;Xw9RfjEFRZdgJcR2Zcw
zPYSiuTV3t+blw*>QMj;?F>KJ<3q$tX?cXfKU59R|S5K~*Db6*>0X|@sr!u^~kf!B4
z&57@+14KcDwI}u&4k`r(BXC>Z%vXQIO{X~VI1oGcD@#C_YT)aY(#5~Of>o6TTpk&i
z4VgGRXxgU}j2}3<f<;;v$uxZW*~I^u>$nfIEdeg>yw)d0X*QMXNdkB}FzB7($9(-p
zFK}9r;R~TZ-!?uQw{tzsDeM2nU<XFPKX>lVHy|cz5&#`f=><VY`Qf(<DP|@n-a!7`
zD6n=~z$~grhvDZtO!?sW=9)YcnV%Q(A3dLF5n=gq?aOkTlys`-B6t}y5~vq&=MN{B
zq;$P-x<A8;pGYH+CscvjdMe$pyY$DR|BK&sw>S^FJ^kTj5ugD!GBNo|)j<;qpif+R
zmD^T>OEu7CJL&BM@9yxc@?GTQ2Ds;&Vs^i#lvImrvH^_p2P%Rbs}Zld>ZR{hWhR^N
zYa@W37vv~iKyG_3FW1FYgOR%?goT+|W1`&A_X216I?oo}5~XPdSwR7h_&QSQIl6)0
z#k<JqzuK%ZGVt!48l0fl<`{iF1vvTl0(Sx=nsZ3Os!cq6jJgnv)IfJ0b!%4T=Nr4j
zSwZT82$ywBJQGzv3Qnku;^aa=c*}6nWv(9R{w!z=K{*F+H-pS`f)jjBLhGGeY&Nb}
zVjV|P!5c(-`X@9_B!T)Uyz&3B_myE)r_bMtpo9ntihzWY(n?Dm7XhUe6_65<kVd*W
zDwu!@(xoUNB_f>yf;32plz?;`y5m1zcYnLCyXq>uc&_W=h1YdFhjYGn%zS3%bI;73
zX{{6K*<@yKN-V=E5~b*48i(N=9^t#V?cXSmZ)&rZ%Ef1o&a=IrbtlstIsN<T<FNJH
zOhy+N!d;{;_1xEa1=ClMOpEEkQ~@k(>BHk)H)-C?G<P20Z!@=~$E1ir_mF^u44I0D
zY*}E#E~>jMXm!Kq1GVoVl=GCLwvV47w>WN`GqHWb)9e5-qP8~|S*0RZUshjzKj`x_
zpb%k%UE;+a644bEZ0ZRI6EBSzeJ}YEaAXMRH{3rc-u=nPyGAwXZ-G8G?zA;}YgJWB
zU*}?OQc}YrNO3J58d9381(rrEBOgCLXvPUydAjL(NzUW1rMl<x^XVbB49%GuG3fQ6
zOr(J}ZHn<(Lo8IIPv}P%D~`!rdq(niK>g-oe-;L`DM^5^{Z%L(2rkL7pSoJD4JES0
z^Aid{FwJ#dp90h52`wxb`qf<*DpI<^0SHUc#Q93;hp`^%hgf_dcXocz1jmBtYFsBc
z`<rOTtkQz}r}iTF`mV(bKw9m)$7+qFFwh4Y)(Bw==OHEgs*RY{z<s|@_^hRplqSSL
z(yFpS6RRU|gwr&yAQrl}s%pV<7CQE6LgSSDNXQd}>OD^c?5dN?PuFQRsqPYT{!V=T
zcZo;FZIxR9yKyLU`y4b3>p$QGc~%gsh$9sf;oMQIED(cHKmgDP+4y=*U9)YL&GV~s
z3S}3fzsgvcCi}c|ldhEM`6pGF;HR(R0?kQ|Fn0YhRcv{~iNP0q$S(W&%?%*@b-lvc
z72&zIDb*d%<KvAN#0O2G(S~CwAbB7!<6$RE?kPz1clA^=%cX;?T+0&nnD@pho;?jp
z3Wle~xh@|+t*Qm^P;n>5@~8HfgAAB?el^n<X}TT3u|PEg_t`-iDdjyhxxlck8nTR_
z*<rZdP^|t-uYAZBN4)GkrwZ}D>i$_9jsR_qe&+`~VrFJ@B>f3arhKkhoI*iTRhw;R
zzS0{%76XWwAdkKCHj&sevyp$W@k{Lk)(z?1+~KMmM$T!4kl754z1Bp5;(l!0NUqD8
z@aod?m~DRvP;lJkGS0QIpN9M^KjpVi??j6eiLd}GL^&m|2TA3(AO}<dA>jEF8y-PH
z#6*G_Rj8#4;1XkE_L$iN`YYpKIRPsg-~_C$;=OQzO~gm)50ckbI}v8P|K1g?2lR(C
zM$=qVIa!96AUfIyprH-|*>R``62(deqg^@EIvnmw480@efeef!SLfpuG-syPb_)y|
z6%pnWi>)nq!%5juxn_ydX@Bc|gpq&C%=nSrLM81E;OeS+e<iw|E|ZV16Ed`HJQ4C>
zrq5!bTl$>8I`;~am;jiOs+z5CnyIU{hkrRChO$I=^yI;>6qW9Oih7yd59%h-5AmTD
zjoGE;d9pKef|pNfIL*W@J@vxKegs0MB)K%~fEjRyhuY7cJzESw(A4#5=c$`;nsCdL
zzk>^ev^!A0g!8E6(lz?yCmt4Gs+Z$8)KY&qx1t-$@cJFz-=F=}l7D5UQQmuz$nd?>
zlw5cCT8=23rTANhL2bG;ec1t6-inym*lKd+J9nU`B^{*sMIhPJDK_a4=%-cACH>2u
zV)qf8bweJCmcS|TbUXy<_|+1ulfGw_{pW}xMsJG3LF}91r2M+Pn9&u8S1BQ3(!SL~
zzXJ$F{!2b`d~EOE5c0PjfE`~UO2MJ*#~a`rt`v{>sfOqg-njAU9a)fh=X&e6W~$G5
zORXmP@2^It>Q}s+M6_g=&`2vt+s$I%!%8o`EXH5&D2CI={lL>&H95HdeVHQiku(Rb
znjC8q+WSfuo8nt6ss;0_Rm0<+fPg8VJ5@2oAg%)5E@H7c2koF8k?~Gy;f#ynNdCfU
zEiElB4il8Cs{RF{@AZiI`$G}eEm!Me1v|et0mu9~B-X`TLi|o($Em^Ji7tU`;UIoK
zIbEoc6`rn*1fIV0E-c;A)rFej6h2C*)->94+p>D$3Y3s?8QuHs@A&m4vb<$6uaj^!
zF2fe$HPXl{EH8iV)W{I)2-pY|rKWF6Ql!^09z!Fet4+2~FO@M!mF$*sO1W5%G#6W4
zY&OC#d4}wgtj<A?pEBHLc+gbb6v^@}=@(mDo8*Xp{P;10YT+ZO32H5YObZ$~!?HGd
z>CxBMmovKWEFYxrxvoxGEq~l0cE`H8oon`e4=d7faWC{LoVa#Bt(l&@$fDF)2>hhi
z^Y8p^j9KTt9?@IU{k3hvtnF)o`e~L>n}HTR3=OGzkmA}Jl~zg01RA7MRAn$fXQWft
zbs#$2{lru>bIdJHA^j3bx*-Uz_|Om4-#tTdoOkcsdDq`@YBeLflzJ(3`N*tKybdSx
z0Ri)iGJNg%#gL1)kJNe~H9tN@T!7-|&=n6&UstHLw&*=y=k2oZbD97zd*1vDq+y5-
zaAAk+!-b(_=+aO_?2#+)eEs@AA^t?g#L{Wzz@h?lDm1@YD!$D`jhGy1ksW_y`XC8B
zAK!P)2D5UmDGQHyz#402X2xQ+CBHnt_MySY>`oEU*^Z@Q^4^wR&!4lT?DdZ~Zp&c0
z%BX*GJ93U*0X4XT+3*Q;*o^c0M8HpfQ;-J9lK>kkk~jG%_i730bC3tNPZ2-WwL8L$
z0Fm;%+0-OxX@}Vl2+rBV>AIi>%}%pFK8aRX6F|;rkz)kJ8F}xzmXxN>lq-cWjao5(
z%K-=XBO(+vXqM8EXo+OW)&^s9WRH|1IV?w`oZ2j;()IjIjE&2gYi#b*A7SqQ945Co
zPnNbomkv>+g9K8!bhe%T)B$|m4^_>lL3u^DeMNZvAh=qQqCIuryh&9I=^%`p;MaV1
zbrFO)we_J<DCD?Z_X}<gB?i{f63!g+d+v+xdKmKfY5B1)k4CnjS7{YH%t+>T95<pz
zMJV9r%zhYVC)ZyKOZJ0P1vbdW4K^?~u4JyUR;<P*L~76wcPG=s3+e8FRIVb^X;RTi
zRRJT_9cs(iH)`5lJa-AU^b*7lHI)7dFZ1KMTR%MdISDm`6O<2-3YBmHSm}H*PsrT5
z1!f|pcV=N8O3T(T3&&-{S1mMSK4o=3I)C8^ltRSzW?q6~${_aUSyN(b>+4f#N?-Td
zC*223%ZmrNSlEjhsE2M{Ku@+jy!ANN<k?-$JUb-aG@%rn3zOc34qxu$Mdy@t8NF5P
zsKMo_E0oX`?64>u?S?!u&i8ZR{JnpSO$b&IXDiM>{n78dGwp~@ZMqQ2kXcA^SAYi?
zP)G@FSX_paa-H#yuM^71AUF1c=I{lt9qjKUzxFB=DbRP`Sw)fo+D>btA+TZ8wC_#5
zhjYMnx$ePbFFEQ_68_8B>{g12*n~j--v!RZ@NwNyA|Ml1Dz6=-GJ(>-SO*u#TTyzG
zN;JedY@Dx$`8iI8l=Z%Px-imm0}l4yEDpPhBJDO*fQZon@RqJtWQ(%<yD?CF>V{1y
zaSe|N6=8-4{a}kTrr(L_vl+q}77|iDNr;b)O&)d5<KMftcj18p98yYv)v12$BlaQ@
zIY@2te|ZV=7+*0YzHP>UMePAwT1NKrq=u_%{~-e9@5e$vbXaOxvCF4Sx;2xnd_V*8
z33&MVOI!5ADQ9M;J}^T+f<4fVa5SywlU&{>+|g~UtgNjxej_W801}>_MBi%i@af|?
z>bUrqSD!zI%5(iQZ`g5RS=C=?CjHam{?t3UGT^^<Sm#yxZ9(tBZ?me;?*VtMVqjt*
zlISjj-~I=u{X~z854t{;2R#smrorS>s`D2|-tQ&jU>Tlx9_J<3xCWRR%EEyM1Ck{?
z^i`<y9TEN4k0MI(d{<kGMNDrqB9=`8CrU1OAClO1?+)Y-I5`c<2<iDU>>wocyz8sW
z8WY33;@yB~?QeL8vW^`uVu_0nF<H3ln>$sVeb={_1#!GmUB1_a=)bI%h*rdOC*#<H
z5uZ579E2fSq)^IY?6V)NWC$dAeDG?`KkQ4ZecV`AxA$5)gQGK4+TqWpj=Yp1yLVW(
zfF<spT%!iiAP{9Mz5Y81{6oq4hu{r7K!~!p^_R|I<N7Ln1a|12&{&3$&Mom>f?Sf`
z)Q?W5LA#Y}9$?tlPQpl(_sP=r55IjN^C_NH>J8HL1#<ClvcrV!4Z%l9x~RSfVy8fS
z@$Ymf`7bUJ%A1sVCH=PJr=xO+qdSGuSDoMuhlM#iz0RE{r$o-vBvqHLzrv${;>T8g
z*0LoykjcC#Cc{>j3Iq{eLeAv-Yg5u~9gb#aUT4rp0G{vlp#8g!|6?t_9~wZp$6+L~
zS(u$s!VQ&)y}H~Vcl6<QBqo8>P_l)GtUP>tQ9<&$x>sDEGL{hny&Y&fa(R_ba%T;t
zl2<v9aM44vU~cx^8wOth_uvoA<&R`3LW;n57KGa&mQl(L<w^$8Spt%dt<ME%fMLl)
z&sp_-*_~<|8eW;7Zt9^|{p~OfCQnTu^YwVwU8&eoxfBluzmxu?U;sPV-vch`AIJTQ
zZ~y55id=|z1)^Oi!z&2pDg|X}X$f&%Qe<*+vZr=K%(HGV=*B|TGf~dLNZK<T$>$9{
z8D>-nJM*FnM2hs@VW|SRI_ZpcFyq2+Fd`PN<3C!3AB<7tJRlOaqoM4*E$B1&(0?~!
z0z3<StA$X^fxXwa6P7bq*uWC`CfGfA2_nvD5Wo{_e$4}jW{JMKy*RLU3cQsKu^6%q
z+cOB9A{D&LcGD{2qam<=gK>2|C8-SA+^3;HxP)9ca~B0AG|!2NXB+iyDDz?J4?3f^
z@bR@fiR>#J|M4m)#68*A3W*r_B{LAj-yr^3DgFu~Y9n!^q4bnG_pA>c;srzVIt(1i
z1e#PNLGA>k_+*V{)CNaIA(}=2C21>9PbuM-5Fx>wtRNOrviH1I&=Q}?0Eu^S82|lH
z_djo9L@EB+nfHN#`(c#QwBtyt1Q1W3x_W3q?Rt+Q^{k|7-e;v}kq(~H8)}%Fl8Z|<
z9shV7N2h!@*QMfHXWOg6H@>Kji1RH|f*4cs%vWU+Stf@=sqIkFj<*bq2P5;aSEi>q
z9%>z2p$J@I<9$WSq~I-q^KebRiS0uCQ;CSoUP1p8@@TZ6(VaNvH#G~N)MuVbbzN2w
zxss+K<o~H(?ux4z=EH)&<tuykP^SD8=sEDwHbg>hpD%rQ;HQ*&C7Low_1YuLp{kRd
zU%{~EeE7|Kh5#4u;X;hSg2Jg-m(nsZX-vGF3;)=aeA^k)vejiF`NJ;Zn71_xVOPh5
zxV=A_xt4Wq5!ttQU@*Uj?O0DD>x7WJi_w}`SS!;)dj_3YiqJc~F4gJjPMT$}lyP!P
zSvM3P+H3VdRg<zAC+izm%7^(=)1UIMJQ_OCGF8KV1;dx_xu#rc;?wpvT1<WJff}u=
zT<mJ1>YeZ@k<qoS%lVhaYP>ovuC~&PnZ0XAvDwl%r<Py@GHZ+#)~J{kL?TI>Z!xTi
z-pNgSJ#!^fj31_8;?v+eWE-*^IbD4rzvq*v*{rE$>(kglm;}$t?DYQT=w4BA<h^++
zrgYEYdiZT#suike&ee83?Zb}g1a}3-0^H1+tSjYMv9KM`1(2OPaaSwE|42P@gbdtD
z1-Y&}q6{nOqy{_+(S^&Bk7JP>If4>jI@(h^kC=Zf2;gQ~qtU=`WU@pcwbY(jBmCKX
zZ>CjBos!w^@ys%psEo_Ds*2_M=o5WL3fboLmNxCK4L+j$B3C%tUG*GRcZMW|xq3Nt
z?{qeG=vEzUv6L3&hEJCc%zSED7@3CMwCwM*thcO@QwR&Jk(*5^V#uzzc1Dzd{;1Xp
z6;r)VSL7$ldX4bVF$!DlZKWXS1MzO*`k{MA!bhD=q4pFr^4!_<80-7B4@}{KVbhs|
zrz{6u!#%o{N1XPWMM=~w2KG2jQ*RYBdl}Bk(W0@>cT$zMD|J8mlVwke(!GH9I9Q_m
z1oS!k=ssHdXjF8V*!bf-YhTD5tWGA2?_`Cd(~Qu|3!a|PAA%nb4iz3vbU6APl@yyO
z)&VLeqm2hTXP!OpHzqX%r23_?wnE!?)I2efE%8THj#)uhG;xBoU=c@~jsQ4k$iZd#
zIk);BK9_?#YA{c(jn&WSAvpVFXrw<8kf1c=73~m)T!NyOYGxTr1aBGHz-}wOd4?Ax
zXsH+UVn2_8;`5BF*o{q1q$4)7BZBri;N46eiuq23=?|>+XPg_@i+DG8>Mz51Na>Kw
z5;(v|qC;b&^Uz%mnu6A*^<3(BDp#v``g9geEVtfMnOep2r^Igp=#zn?#tEkHzu1vu
zFW3XavBZk`FVFnW-wg0#JzL_lNK$=ydk<}GGmdG_M3WsP?{oXvL%!%6Xph8Q{*L;o
zLakz8TE#VsD<ovTM-QKPeTtp*KI3umcbDOafFN}lqYXiXtMUldS5W?wmH=h87b45_
zFu^HC6m1;bdqY!-aa}9DmxaWpvmcDo@4dXV3(n${+?5dlGW<}C1APKz@E;gH7`<j_
zXb4r6F3^y~xw;%=Cu9rDj|k)+xP1<upJm6$&-e+*jxHK{2+$)L0?EZa-eSjCScSsg
zM#RP*j?7ozf#i{&J7={p(vqT-I#`)?hVVX9Ojg!0g;CeI$xPt3SCxx|*6HhYuz;^;
zCSutq=T>kC3yv9_;DQzt@?NA$0XbzS*L_3%r`SLM5gVmieng4zJx5FwBh*4V-97*k
z)0%AmWkLEc>{qP}&U1&s9696`N%GIFJu9@r!cX(rjeivAzGbH8YyAmUW!Q%Se9xyr
zhTu^lYIIJy`RP#FbC$USa8jdm9B&AXHFcq{LH$$C9Z$}ej$L2a<IB4Z3=hZEoY6Es
z{qFasSkv<8PncMD7ca8sm{3-JqrVn%MZBQHWKE#v{GS{J6xfZl&#qb;AT1~W`EpyU
z@eTi|N1R4l6giaIAC<wv#>Yicx{HD4ez1GS(`<wWaf+FT-yYY|Y1YbUt*xCR;el3%
z(cSUS%U8mKK%mYlUUP^*O`7EoudU3XaFzb*`BO`P{f<F-1c%^=1TwUN`>09&18eL9
z#G#UZ&C9_Z#-eEvB>PvxDAcfA@^Ws+4kQm~MGcvW=-T<;$GYf^l2}uH{s@#(8AlWf
zEiyhmtl;)bU!JUZ)8UADaJs4PdQ?>6+BF+-@hma<UHiErA}fZhvYl>|eX!M34dgs7
zzNVt@Nc5Am#6R&f8d!(ishmLGtyrVj9FTQ>1hQ{z7<QLZJ6?V&M|;1Hs{)oRiylig
z$R>k%GynqrW5SaClzh=c2=lIw_&xGl_WrdFzoeA8lHfFJT0WQc-GVm$UG{BSHOS9F
zZQ!Z+{MCN$@?!{yHw5-k8N%M(f3D%Y(stIO7gyRx{c)nHdB<ZAyNw0!%OofFFfA`O
zlU+UwNl>Lwr|NgQspUH48A1a)#)acA6EiJ`g|c&Yr7QFtg)`fnl{tUwJ_*G?CG{u4
zaSM8_rX#dK<@_VOEuyJb=WOe6+IjBI9*WpbdhICw02fCJF%^}h;vsBu#hw5zMn*;*
zS;i8BfSB7yEUxXO!+yQD<TS~Rf?IC)d@iCCdkT-qrIfvsBa<5#N=$U_U}qcYO-LV$
zbF2~SB;ZWy$kkt1!{oQQtOTw8o@sv?*PxbIA`ttO8%hLCX_|2ujx+k4<#X3oFoeU+
z`A*?FGR{7{*Hp%&nhc(VmNuLXB&5(&^AI?el{FY}cGCFHNBUpDnlF|JMT$kibK*4P
z@jbLJ@(Tcpi4!xrhn#mMsirWh+jZG59boOQYEvJU#FiLInWLbiqjStRnV<S}IhB$p
zUb8*txHyYFA6o>;It8%~cc&ytEDF4m7ybq3l%0ig#tW-+y>5)d5dbrENj&~}93#f$
zgcRpf_bxUxTvqGJI%449kk$J#GBSaJ=jj7|nV2&>Dw}P>S$VO7OIbc<#cE}abRPCD
z!AcBX<b!`4ccnD=@9xnQ+C5Glo$Ti*Z%H{9vM7pt(sFt(xR1}X=9rI$&}#ctDl3+c
zVfC>A6L+yhd)TB$mWD)gl1f)G>ikQ0?K-_sHto}?+_j&xFJo9ga=U8yxnCE}=1rey
zNyzgaj>?m<X;_KMl<P^k5^mA{RAcSgWy}ZFJDORR8vKJzF0p}2#_C#JOSa{a$!^0@
z$)4PvIO_bKCT*5Mtr<2$OBtp^*=*F-?Vlf<)zYl9W}6tiBC$H%kvhdhAjU3%y6*98
zAW+5)br=h-W3xkQH)mUx0(87~>4!|D!%7WRjf_mX(yFFD&n?06@lld^)-rL7gv9Ye
zwQ5S={V1I?WhsY$8)VWq+<jSr2rCpf!}M6)QJd4WOEa%>N`AY0+RpQIfi;OvW4U*x
zHFxr)4(ah3?3aB`6`c`D4!aHw$<i)8=}0;k8c4qK#BI3b(S4}`{SaZpzWIAS6mRK1
za?4ZmS{_ADuD0Dln(#S@7dXG-AIfM`olj})-G}j9S#B{gYe>Gw=9HxB;jrwuc#p)D
zu;MNJ#?gkJ&U*cjc_Wqc+eLeP8EEG6R!8{879Z`+70YIq5gl`OF1elUqAQ8vmJeK<
zl^2f?^W@P5-QmQ8jJsa#wJ1C?jp+xqot>d-AKKJswQcNa=;%iI445%RXM_kTASY=&
z$~^W{3+d}qe;K=95nVJhcGgsSVB|@S`@rlmo5MwHH=xs@V|cB?vB6@fgERZkukyg`
zd0T|Gv9Z*HSesW(#l<HoCg3FhMEwyM(JhghC|voUN+bD)5y3rikZZq$?=^6pQF0e;
zZ1`Xins>cW$+j_>@8eGr{AhQ22NY^_rlvM<x=d7XiHcI|>FF6JSF`8j%}*HzgoWWj
z$su00EoK)8hhfca_kIR{C&2?Y@3Xcw(U6l1epJcb2Hjw1c5zlD&r<LlXyNhOV9dFX
z2TYB35fsO{Y$+K{Uo(3frYi<Hz+?ACgzR?`5j9e|Kq{MiUS1xTn3&ewa`)@e)(WFq
zF3~)rBlkR#U^(rIEL(1G(5O~UpEwMylOB=<1O(th?eb`trAn&dQ&V}kxCU41KwVDe
zaXyPuul}Nu5#x5=EPM00vD;j{yp2|OS65g41fM?CS7K&bXJqw+xHjV8ouhaUx1cf3
zn48+hBbi~>-csz}xoa0s<&~b28NwR&uz=uTHG_9A=3O#1!(CSw{M1Z7yn5`{d9lI*
zgnLinmURlZG=V{_;@~8%*A}$w#ZoV-t{|hlPRsqOmhKA+3m*0+?`yM;i}Tqp%_e&o
zSq&F^Q6*49E1kw9L!%|w34)^a(y!(C@cfo{xIpDeprH!)pscJcTi$4Fb8Kv^$3k&u
z#>Yp5maR)CSnS6_Q^MD|;N2v!;CVr}15j77AR+qf`&U=Q75YSuBu8s&Qt~I7*_mQU
zTGLEgmquSMFIRG_sf|rNS59q?IQU3d?Hc(Prt}A4&%ZioidWFDrKOl4lMr|L^hXo3
zrMu&h!`3WAy=~hzZ36>|Q9)9;fYga<=gP{qRDG?1;bASiQA~~Ik`ZP})1k^GaixA%
zXShkf2rBLBuP!gna1z0-*zXH778Zy>*9$>VM|mBqZ-_X~-ne}m6uhBfx-g@xtjq&y
z2s<^edsGFd^QcFvF@-rz#`?)XEmEACSJ`%uWVQX<f9g>5pXhY<JQ%8SKgo;OxFhW2
zT|`>@p))|6KgH3@a%6tP6QP)8{fy<wxC^GX&}-1cvf5s&!lK13-lAx%okaWc<#R<7
zQdFsj<JB{-K28=dI-W{GB=Rb=WiT<saVWNrkX3V`JhCXu`es_HEsK4cyY_m|A#`#4
zFlGYEekd4kF)f#{FsY=JRMGgjxk`24&i0q&`}fE6rX0JR<VlU70aK{erO0|ts?L|a
zbzI!sQ}32~=ewYS+BB;wxWh`ie#$XimZ2usa`x*x!hcd=P)^{V6)SV^hWwlR1&5<^
zl6BMS>+8E7NKu;zCdO=Sae<<`n^6^^4nt*fH5k{UkA&xHS{>I30Cw0s2DhTmA7Vh!
z9!J|ykX>ejwYP7vN;9^nudJ@-YcsR66UZ5}X5tan!hjk!RkR;j=OYD>M@Gs&fn&}S
zXb{jeoABUZBO`UD>(G8W#{9kZ@CAAM@z=dZwY7#x&_6?PCm~_#J$ua7*iV_Z)h=Il
zP5uQ%yd?v&nlZ$10oZ8b{WP78O^GQfDJGu7R<+2Q_xeI{e@t({lkKlQ3|DfmaWtQg
z{S-)0O25uYzr*1%V9v$Hf}D*ZtZbW6Q`TZ=MwwP(eQdy2siXF4x$~R|ahhp+(^d7|
z+n|V2nb03DY;YEsX=oy37-GLK9X4~-_TDed_Y3MOVi6?lY`)|nSozE7i3tr8iQ0ti
zgk~;NvD%H#&#~)dz$FQ4UaCWGV#{4+<J0vuJnDTwjNn2qaKajXp$F1oRXaoembdZq
zAxTW4+jhdn1kbm)#4)!D5;aD74FW<!Mx=`)BjbwTz`Nz@DAY%#`RVu_`v?g#rWwVx
zwRgtHZ1W?#7S1z)Kl|wV)UWt)UyQVov^L^#koE4pTe0@;F-SjNR9qaEDB%p3i`{(0
zs?4Y1THz;E)K{ez1WQo{0di{Bp8gO^8>kVwP8nv_4_c6>qrLNsAN+;!pi)5#V$O@4
z5O+i2;FOfwrtWP7G&J$7;5MiV5-)^S%5_(V*wdF^^#<f@L$pzNyg=MKI!Vpe0J%VP
zC~WxCMHYLTOPAQzIS^iCAd*xkj=Q1Uaco8nla3cvhxTYxLGy$T8ff|!G&a#g)4At^
z33Hv*<6}&&<JX6W4meNLLb;F5esa~uaVI8ZTS!QB)@h#0$zWTSDQ>~3K0ZNZO-*|H
z_(izLOrz>kt$4G?Tuy(x+#E2gR9(;+!r&NVO&PQ=JOl{|Ro2neuFFP+LY)iz)-}C?
z0n%ai#+Cn4-~Kaz%&Y(~F*z~{r!d$^>~$g_Fz^cNJCoKh6Qi~>+e2MfGqtU(tm_#}
zGrIDNMCO(uN|%?{lVjn?{3Hg(xNU1aVpO1s0O#YRB#fJ-j3DT+%#ox1>(k+j_5dU*
znr7L%X^8_1pfhNvO|gznjk_uT6{wf9?&`iPMzVc-P-kBL{2R`jkazQZ^nEZP^nEh(
z$R5k;cRRSUvXZJIO4XF)y0Eq1&w=)lu>a6y(e=qt!6Y8f8f9Ju+09V!eH2fjp@H<{
z$c)6?*(EB^8t(Z|K5J}F%ROriR)$T?-B0d{?nZLlqIxcb=8UGT;dL%H@mk+>9?_v8
zWC~yQ$m+i%0DCB*qWUjQ;D2WIB9^eWg)g;5i#4y9jKJojGq6ia!U&++SORV$781f=
zh6ca5LN+6~Np^2a%ppD2mM7ikbt%nS5EmccKIr<gMj_vtqOSn8$*p9U1mP}73a>cZ
zOL}^I>h`_-h06JhsSeRa?qp%gjxG^nnTLp?9_l~PYx;}V{27V+6Q~EgP@T)mV+&(!
zCg;HFoR#?qcN$Ycd)YuP5fKu^*8^G=Ff(2xx)6}-g;8_j3n)1G5MYA%3L00(;SD@W
zep6gd$Cd}FdW4aPpu!ym*9!qjxda3<#+NJdos9OAlb59``xPnYO=7whhs}%y*HV*2
zIGo0>)8Iyj&i2(dH#eIv!=*nNOJiE@aY?H-$jt#n`~Ban#40!+${E(c2z*pe-vu6I
z`f~g_IdR2%U=`j7TyWol#>Ez0C{wrz(wcqdfu`eVBZ1rjzda3*p+%iH{meCvH!#2;
zvl8xaqLPOI++8H4Z})H$6@|^n``zHXQv(#-9<L8yWfNW+navu8zMZeTroz{Z?N){r
z;vMr^cPN#mw0bptr{Bb%8}XeNZ<)9P%z{)A;8)y>hVqLz^EQi8j~b^wt)e0#vnv!$
zaKS0XOMND9G#WkLF}LbGUwd=);AJ_AatBy2)x24TUy_YKhh)GJaa`PzMjtX;529CB
z^IjM>mV+moGp?RF`KEbf<<4H#!dk`I0~hA)&CR)T<`1aYjY1b#(Efvt8cvl}RY-ux
zB_u?N@B}q)ngb!a-5V7hZIo2+>J)lGfKO0xFC!C^Q4LuqXJTqYGu%{s1Df^GBY4{-
zC6<+}8|h&)=eX3`xd)o)(P{UT5RS?Z4rQF39ASC#W}TMcD&BNe4JNFHxBj!+6%&sD
z9Su~hmzcMjh^ZOi)|p1AlH4It5I&gG%kE0uvDj2!=s`H#8hW7|M!<&C9Q6w(`AcY7
z0=D)A(M_cN*8n@>Ds^?;Ouqr>uVH7}9+YcJ(3<Y;{@}rb7pCnW%m*IdYzi|f0_EE+
zFS<ae4fgwq_(_JE0erfPKjk(y9(6SQUoU{Kzlb8jGpMZ<f8&e4KCJ#@mbFyq)iE|B
z1YrId&;9f}C=9Gutq@$rw=sib*h)x|g>9s|A)|}11rKKcMB_X6HT@vv=*vYKcnDCX
zaV>S7e%Jr@K?r>EX?b?tU#al^=RR4T0=I?Z^@Z<U4fCa(e>t%~vE|Pf15~gN4$*!`
z*Xd0C@q=0PfLZ<*+1F_+{s7KtFW84%C7Hl~uEMXpqp=6PPnj2f>%`?hrs^OC`_K{l
zOm-u*YYUnK(R~J$igmKcf2>O`9qa?9;SA&VcmGc@`2Gu&KcZfH;&s;v`oFIfx??x&
zgRSA=pQ<$dChA1UfO=WsHND)JscypD_hsRm&{DezbH9iL8xC{(N8V-q1j@PN$|@+_
zgxYC#q!p63q2crUyh<vUffa;06{DiT=r!y8`}dzbs4MtwaLQr<z)`km|Kp~PVmig*
z(kGc0vRdjtM|jVfK8kAP2HipuO(pQA8~3qvi?Pkw=f&Jnxe0H*!6&s24^MFC)iO3X
zkFZZ@EbO*^0@N#PP<sc(j|G@%SMWoj8=EM<amTVbQD3(3VO=>2m-!+isFb717N-1w
zbl|YYxw9zZhM69oJO-Ddqu=4&$*NhB&!0Ynk63?eoVNbqmjcKqVY>&CbRUZNFOJ*0
z#9r$wJy^>xmfz~?xlbzL#2?8AR&umUFAa3J#_8wZ9(Of7yye{<Kp19~h?u2iS@y)i
zYdR?h5Kv;%pk3OA?s~?VV(witHuL%Z5=Hi{a|?!z;ylPVF3=jK@trMr(qcQ+Vmtl$
zbK?90bmyQq+!RM5FL2PfJqudxOILr4iR5#KzIFcEGeeCGQkQ`EeYba9Lm?~GUvMkd
z72NqmCeLFPaF%2R6RA(gt;8cfDNbt?mh;5zER&JZe0b-+Emv;eAt0@*e%gH)-y5d<
zpD^(fX$Ew9<I!i1F-K^S`Nih^)3t_v-}VTe9)`^hvi663m;dwt7S3^q&vGV@l#qU;
zvTY+*%DD7x8+=wC>}vU=6R7|6)1Y$q@kZJ*Nlu`#dU|?r67>#$b4(})uoO>~Ey;hX
zsl+!u4JtU;sy!I+Wm?uZ|MrJ-+3>OLv=Qj|&h5Hi?VKVx8oK}WBe70-vC?^3D8G}1
z_P!kKMvhD0R{Dp?@#gKLGJj=jeP85q51YTn!4OE%<i_&*4AaJSu8e{Ad2IZs?>E19
zUn)YZhTifQk@f5SI1EnK{WGb((mxRleA^s;0)O%#S@5YJZtgc{_XR<Kl17B{tV!m#
zm-}ZHArm~dw{P_Je6`PCJn+8^>W{y<kHkZV&#5JD%uG38)2}86{_{q;Um4v~B;t^6
z^kdzanIcC=>c))gJT3i0?l#5km-K8?+<r~ZHpOjI+`hJWf0Ly9o9Om4HuTT6+cc)X
zrdXTewkdAkTY*jA?U$U^%~0XHR_>co#HP4yirY7Z0}(0IW@P$pw&OcG_x}+QQgBha
zto+d*yojRK#XK9ksMT06uWD<a51%suROEsB?QKXS16}h|j)<w5YB}~hCoF<~GB%Z$
z9;<00fwZ-e1tK}8W-5^h_{wR&g4NE!?3%gr&bp8-_yJ2|g}T62<cgNmOK4@3g(cD6
zzd{|6#k&t~G`K>LQ4_7<+CoBGyM)y;$G+q$7S3c8-ai!+ND>Y_%|xSY@$a51N*OyU
z`{i>9<kT(#AFIV*cx>ui%#HnEl5448lw%nekeQs@Z62N^%}|VVbR5IIy!OC5U!{>s
zEe?n%+Vg0K{2k*)67h|s{4k+iq%pSPV&=Yd$ODMu+%OK85<)WI_iMx4mYf{rC4o1g
zJr6&v!pbyu6??IwrD<b0qb=yavY3C#7yU}_rIDjxdb`$*A=@U9!as$C6B|P-`XY?h
zsqwULQ{>-K#eaO%4Tr4<!!F$ZHpTkY(tOWhBW6aD4<3Fr*_iYFV&s%sICanAU$RSI
z6PSt!OUdxMbz>4EZXw6}c7_aRzImMR_4B{CjuG_0c}ibPXWx*S5_=<h-XQ(-&j@ur
zFHom|dG^IN``ZoV|B$2aip3XJgO(a*+8z7o<i_L*JOmz_8LuO4Jrn$Mqfo*y(_$so
z2OFdI3~-vDxF^SCV`jPuaQ{`r{%!y^0qzU!`}@*-cR@D+?k5)OizVI!xL>njn;zUR
z+OSQ4`$GGELT5Jt?k5)OizVI!xJ`ij!3u1KsNW0rzO8nfA?ly8*iXIK3{iiPh1d*H
ze~rl4Xn@;)kLm26tV7n}TIJgADVziya%5YX&lfo7sVKw&cN$@c1r%-EEjx(Z6ys65
z=j=OMNYzL_e{>RP)kC}q62Q`S5Win*_E_nhlk|HDh8qP0aa*%<v(MNJ9o+8poD*`b
zS14{HUD#!S+ts=gmI;rE$peaZ#)!V~Ik1(AkECz?QF_Y^t)d8v%duv|6B3S`D>&wd
zxB}=)Na}P=LtzjH9}n_)Hx#trOXc+-P+`h0)t__te7{qSAN*wM&_U#$SW<Q_MK$vn
zH;EKXUkYYG1c^OqUC<7lgZlZL<DY-CgBX&T%^FVcgV7k!f6NY(Jn99hxDou2>+Kln
z^?53E=#ix0Bm=9L*L!?YlP#(4@%y4euvVv8s<Unf{S;<=>yK3=fuQo+-Fp5P-SNK%
z)i5}qlvMdoZ%j^c1#`slqNA5uf9jf`Z`zw%O2A<X3^N-3B6|FNg)WjKb?K7EY8#Ur
zX#wfe#QN<g<i0t4L}-5Ci2SMDY>_~TR@^dyro?~Wpnv%V^qM}V`ua!=fw0@!Q9<&P
zCmFXsL<!uu5yvdE`?VV?;ZWgKws^bv#E;VIb?21vr7x@Ysz^wy6A7c#WfV`ClkH87
zxw`g|3k$9Omi^{k^eJ4to3~HV1_p-83l0iqY2^pHB=uV_PCbu`>cq?qcTOybI4tG{
zI(<8z{(rM5KTI3(IAlq$>Bn>U59@+LV_{L?7xR(+hrj)oRr=~%it<~x685(7r>wI%
z%O9JHPfA>?NSpTfhnw})F#c5}&{y$s4Ve4$kOrI^uyl$MZf+h}$%Fv48}NI+N+^`L
z;9OPw#?ZuAIM_U-w@jEfVxp*s5D<~_Of$WWQ9@E?kYALs_{IQ4FCrq<&fH(LvR%jK
zMxa1`aV}AA46-sugNR%m3s&72*vCQSm$+4N9rOE3R#Ac=BEFWLl^bI(nH52P(eFLh
z+4ABKz2AhmzaZreX$JkbdI5Y7c{bthU#-Aj3;Qi>+=RRTc^W@yb~oYf7m;}r?tZM5
ze_5`-KyMS`HX-guD!1v;{VJModUQW&|2BixA1UL%f%u=$kKPPgzq>wvF4WgA{vU_9
zd#vVce+K&(&PGJrYSt-tNYO367245{M8>F>D7wTt&jC7n8)=08piOc!veG<`d=`<4
zLtznWc7i4$;j#@#7c=sa{aiIPwV<tKL%$ZI+%9I;l(C3p!J;o=t2f`Z;6wVNNs7~E
zm1I|Yd*mE}zf=ry8+Nj2l7!$CRfjd!e|I0hW+oE;I<8gzacQ$bKtYtp%u_lfC@|nf
z4-_40UO3;ncM9k@R1bERWA@chvykX8K!jRDti-8=G`c=<cw`B_)r<|9v;Aj`MT?@J
z)=Wpv;e5{Ie&LKeAFsdkUMdcmVi1fV=-yTMd1D$Z(fUr&**fG>rTdH@kA*WvhM*pz
z)XqFU`mK=UC*=aI0>QX(*E;X(``<(F17CCSfyBYD#qNKf|C|p(yN|e*SWRGKcmiB{
zphQkyJ`K%UHez(QKqiX!dHovf|KnS?fi>HsO1d#wq}Z2$m`Zm4ET4banSVU+zoRee
z2@q2!_PPJ}Z+vfkH^uGOc4SlBehKe3#cfmEzH4(f(e2klzD;BLOHRb5xNVBt_XTIu
zcl)0E_|3K33>ALO#orXSO>z6CQX!G@W@P$He8DCW@@ss-rnqg2+jlM5MiU`Ddt~<h
zk9K2W;)>nMIyN+_W4Udy4hO+Sp7R%F&1^|1Yo84}Z6mfuI$AK|o*;ikLQ#wH>kxgy
z`8jRbfE_KV(H~i3{KWD1JZ5+!Ik%p?)9r9(HUE~=WC&Bd8)1A`#BN=<Y$=OmN)r+r
zH0X<n)|(aR4P1l&%fbJxJM=~5-XG9_5k+hT?$>O~6AoC8+Cr>4F@CdRF+taa^zgBL
zn$fYl4ep`nR*kMy%1;097Kr`8MSD9vNy_n-oPJ=ts~Wkk27P&J^T1`YXE1>Gn%L_7
z>HeXwcEP;p33$4Xb>0$vs2}~gkHaK>hGhR<E96e6G_l{g^3K)s#uD&7$jKgdNS$pp
z#DkY8D^J_dq^2-6=sVCzVjh~*5vFzoUhaXHO)hFC)LSRI?WaRFrklgr-+-HDNNYK?
zF^P`CK(bB_2(I%otv~cL<_6HB(#9zg{}Z0?O6?C>7<+7No=<g_9N)FY{}oad@qhq+
z^j_~ST>T4z^<TgCtv{Z`#Wi>v*{{CtA3~NWVh`*BOLEn|U%rw3Zxi~oo4EGAR1H#&
z@WVC!;oqPbU>D9_3TFSW?We#0i*M=RFvL5|a$}@d8g?P3m``n;H@g2J5BH@&wWvk<
zH%59N5kq3_ZId*(9eD%xfz}&TOW2`$W7^fQ9E23&J*~zMHezUQC~8oxO@R9rRD65)
zHv#U~0BRH9eqPoPP2U8#U$gw10QZYFY!l!%0q(orHd-3@UR*KDXXkPSJ1aZ82(J6w
zcT7mxgV(6ZBh^T&ZPt18L#lRVKpL0*q`Y>yZ&|LEQsML<RoBPsL_2RDxN1Hy5YUtq
zeU&OvEuBMnx;8R*Sc6j2zPG09nZIVoVs1hLyKFF@_3(!$7;e2v+#OU_r~m<r-!k{j
zH@BSE4Sm*jz-N86cD_im&~5&j1_%YUE>`#@zA$c?G%uwhC^B<h%1-FQo(ZQ)Z`1K^
z>8cg)B$gLj@g2`<9wkwJ_3^ZVf&wOp26wJKU4Tt9hi$owNWmi}KCHC(Hc1g>=lr61
z@$Q>5F1do0VIpy1^^(uz&NT_LWNcrj5BQ#uHbR@yeH%ozQfMg}XW`}L{`?U_-3+F9
zSi-qI;m&E?C$!+&<D??(n^Y7XQ(^f;H)&R*+E(~pULPA9Gw%;2D6(!|5fGkfsHYoR
zY*f}FoNKe}Y)T@|+uJee$A6Q^d4Q$}y<<%5$!d_rnR87oJB6Ir=~VC3LZOPrt3u#%
zZ9Hr~<M+ggnSD!La4WR#z~JDO@_U$k*HgFh`f9|4on}+|-D8z(P0Iy}2CcdZ9Sa+k
zT{(rPD(`fK__d^Q#&?Yu5fKpV3n8EFbk*z=tTI-7WX)Hv*tt$$xQK+fmP?p30gsp&
zjhgnJ%b$E?b_1k%-_h1=j3#OIvElF{2{LEKoIyD==F63@uN$|#+nYBPYE)Fk-+45{
zuwbgp#}?laldJq*KRmanGBDetF-{A&MkM{^TpV}&g&C)Tu`%jemvJA7yJJLm$vE+7
z7T$`v&*ft7ap_eCyp{`E%wZWeW_eP*&e%$l9^~VMvHT%a#5o+(oQdeJ*3P`N@xcU*
zrAq>HZ3Kl2{cA=73!g~vsn*=dnr_s%t`c;;-a#XOz;@()a@Tad0v^U0lU<mYJ@)WD
zBg@E$Iyu}RMp<N%zG5~oG$d`(+}PMoxqTuxp-026YvCZiGjsRpN7nnauk8FqGxNvJ
zp}PoP5w4NyBM~((&XbECb-4Ni^&lU3%9;x^h}bUO30igTEkCFI`kL!0Lj3@-wQ=*6
znNGs+<xhgVMdVLzSBHtPI`wh)d%!z8x&|V>D?^qF-REv7ICbZ|&@S?*OtYsvs~c(`
z8lv!BT>D!*{%<y*Rv2QYs+30VtynmCR@T<;oGugP<~^Qb>;}ct+6WLb_^VU<acG2f
z#2n}4%$q+mKON19Yt)=4@U%2pr_@7+GnZN2;*G9$jY$8_*ran@mVKshq5G;xh|Pzx
z)F6`t$D!EhqJGgi8g0O&$}~C?5Oc4FQu}p^`5WAOta|$+ku|o*;%}c8>q={NQJULQ
zA8mW1XqOPrNLp*6cFDa#wfqksJ`}=D$8Ae9(stdqqYJy=RHTU}W!Jk9@|$(EQJR}e
zWOR98ZqCk9EjIFunXfG7->|G%F@~Gzs}F`s8fDIR6BH$v=2@TA$i+EL(;v2U8!-XM
z4cqbdAD&3uErM&qtISsy!k@R8(my(<+JDkG1aeA$xb)xgofYFD%D$BADd~nH!5+&t
zv(QVTqN*>PD9sl+U|D&(E8_YtT%D48vG(!+-FC|pOoiUAPL>%fv)P7ea?jOM^i0fN
zye905q!AH{J!ad&;O}$<0m<7dP9L{n%Jx>*=GWs8h?0gmO%9sRwclMNBNEJ;Y0|ox
zo@ebM(0!}OqoMb3O!={?{p%&<BKD)xE_ob5Qh9^u#y=P(x$mHHqEDTgot-tGZP9lR
zHqGL28qpyr1icL)uO_zMha+&nydcMX@EO_6c6KM*Vf)zYPtU?7)$H1D?)0BAl+rxl
zez46EvvQjXtPv|*ZjMoir@Yc--|x(F!b{aV$#q5T;0JF>k>@X~CG7{1>ju{$Xc3t}
zhy@MD*~0MI#>b1iqdc(SmPDbn?c5>uePTCG`7nCgF)ody*k%kpXDy12ULAe0`2MA?
z%zR^#DP@zOs;3aR5N0k@K@EB|YxCE*EV^$rEIv337x3LPuNKtT4_h2b3GFJOToX#s
z5FvyyJFva_ap!<<NAfUN147W~gk&PU<8B4GIDTxZqMs@)dP>}<s+TTjj2BK7HpB`M
z#nPO5JwMUoxU_s$U^sq`Xn0y^Y<~2`2YueW<w@3}63^g0!Bs)L+y|^1mW#7;+?hhB
zt6xl2D;>>VsPlH&)0ChV-og=g{9IFDZ^R#&*S|{fMOXu|_o`Ym0Ir&*BUZ_>x`-<J
zR3_wZ=G0Ts)u5z6aonsW+G)N>-@`Vi$E1aD?ZcB4OI^@preRa9xGqEiCwqya_yk8^
z`G~8jS#k`Y9(dE*VPib>-g$2@p9y|?`)qJtG~J>m;0|u-dhc(GWULKwEfX@VIKl<Z
zwD-J!p{-;0=|SmswgnB$Vz&9)gFA?rji!AFI0X1d@|H%jvZ;hivlI?AjAsp-?mxRn
zjEY-N)~;X3`z}F{xz+N7SyEgx+{MW|Jw4sf$QOGWR`ivP?m9#k0iFa_pw+DBN-A-Q
z-xANSU}*3)G&EW-cv1=Rg*XhIY7OHbzPLO;(G_A@otV7@zClwvUqaVIhKj=LM4}<*
zsdfpQuW^}tXlPiJJbS>ZcB^)ENIQ470|4R+gSzfYn^TCSK22ps?(STNei@h`lA0@~
zC*_7BL=VtR&AA?Mo_?UcIPJA6Atxl-7Z6$4mbui_=t_r0VVcp^7o<}a?s}qXQSuzM
z>vHSx&`@IbzNSV~^T_A%mg89>$eGG|hU1J3CtphMq#No^!MkS_I~qc+mfzkM#=lgS
zTuQCcOg!9_kQAqJA<|#d;qcB~k^9?bZ_(|dRB!O@FQuB}x1MqAbe0)1g!?qX)pV`$
zTBrY04ixocVMpC4=on9;l5VPjSg*84l_8mG0@Jnp#*UiBuTM)+oo?o;Ynv1v%bBu$
zoSiL{dKXc4=ec&K)Vv3KQ}k=_>!akVMkW{fm2ZlzT#?YrUPX3q)yPJ6U8_Z_h_J1_
zEv(!OsEF`_m%2nZ!otE#7?{HBoc2S6lh!YBU?g`=^Kk<E($Qo3)nl1amP@0yjlyC<
zQl#$`8+1ByGOY<ytp>MQrw$vPePlhG&<rs9U1<9O#GtR@U0$o`zM4>K3^CBH0ubmO
z@zXa*yeP+WZZVw}C<5fu-^Ht6O&w_Rjv@zuPJ3apGy0qvd#71`b7Q=jRqyH&Mp@av
zMtCYZ>w|<RtW(x;o^>A77zm=fSN(_C)2&FrldZ%G4P`BgA!2C%UoU`oRqk|-F1Kxc
z?t$F4l#hC41CKJ~%y-N1mR6J0j@OEN-+)*rXoZZbmJ8gKpsB$)_2llByrN3XIaPhe
z8{5~lWLvS=5v<TnFDZt@M27KRPS?lht+lNw`HW)(n>ey7aqF5-(s+w5s`SM=jvNt6
zk)xxm3{Kb5XeVTfc+4^!Pawa}QU<gisew_MY9P|LM*7hBWJOX#$RN3X$oUk}xXC)H
zwfl|ql>tXYK0g=?l5ALxlAeFu`Gj4;{ASh0WW3;}3>Vg}vqUC>0V+mbeM$>}QRot@
z<gaper8aBIgc>?JItujqk|+%i$J&h~mvRTog=-q4#CaZPTFzC3DBIfF4veS=ws07N
zQ`_YJc=tNs2ZS3X17$D<WhO`9R@-t~*n6Vp5IxfB7d>7=O~y)bNAdXq2AmTZ9|M_S
zWhEQe5=H(fMXxG8u5~YKScS3jBWu0i4AuvNw6*WBdM-HFFiOGs^uB#6S<~o-md}R6
zqU1{K%*<KgC1ab6*RQT9rF-x+)N9eT%7$xdF&E$2Tcfv2?ehS*%CSUpd(0?$1^0h;
z4T6|zV}_j?ymugr^C}Ewi}9tf5W^hnbZ|711nIp6N8h=-K=EHV6Ijh~)T}Je=9ZpE
zp5KK}?;J&w%Q=oHK6+HVKwqS_RP+LJ`2P;usmTWlt^n{`Z@LXH&nA1HM-l*d(CoQ;
z>rzEAMV*)W_FB0Jf^*@k6J81TDDNZ9nCn<l(w`Z#4X3tgxAt#XbYMB3;2@gPa*w>8
zyOk&MxRjKXMtc)U9<rM_QFY7&cH6khDKFZ=z<E>)tn<nWvcSj`FcV}{drxlvJUzL6
z=AmXsUyK3R!^9J(B6GHSz~7(<@OQ4ZaLwBj$hY_+pmuRTR^o6`!8_zcpMlgol(Q=;
z@F+~g<KV|*VmsO7;QNZdZEarIV;XeW`yr~8+0#e`9_xGIV%bC8E^pW6{9@k9&}=i6
z7a!w`v${wV)dQ(&5y{vUUEz^BIuLMPSu0LC|9(p%%W&n5+Vx>OD&iTU5Yw{(aOrgd
zi}_SlI=#F}DOGFH8#1l^CEhgQDq<mp=i<)m@|l@+9!!zrrjQNqIIWyc$!}V|VRDE?
zalye+)@uv;JgG|B2pWl0JY-D@Z%lOEVdWrj923+(*?toLe4>-+)GG)mD*37mYU_?9
zB%1P@l<zV?fB~z1y}+Q0Q<6GTDSHogjh&eoXP48U%$ne3Jrfg?u7@%l_z-R!6d#hj
z@WNq+ATT;D9djRalJMBEV?#qWMIN4kqhLa0Z=E*cO0&IeFITp#=5W{XVH}IrkhRb1
zo`JE1A?c(<tEzyGM#rTG%X@Am^49T|(y3l_x|;k>6J14m2ItAJoK^e8eiq`DkUK7#
z_B;398oL|0bin&D)|iDy(Db8>(m|j7i|r!Siax6hV-|}>JB4k$xoacWxG|+mfBk1p
zV1;)(b#^KFz>)dD@N?HCy8B(uzu9-)JI8eU-Qt>+>Vs+KC!n>UTMci$iBm_gDfh#m
zw<-8y>AA|XuGV%*0j52cRaU_<QTWYXAFu{fJ6NMyGfGdTPiP)#FX`_Z<QubiXR<uT
z*IptbpCxy;hnlIq6kd5chFDR~2@LXect^L+x10R+#%bgu8ssBG4?5!Z-3|yjfeOV^
z*07cXNEM8jxq(Mg_{@J*O<?kE`cCGWrB5}kn*J>kfXI5mdOWwT9b}XXB7IVD=Yn_q
z;y~N;_!PhTfhGyF$@X3JnB&uT?+p+Jmnv>Go4g6Dbb^LOGKmTnO_N_-Zc+LP(cZ{r
zFR7Pq;Ywv3r(=n3!q^YdwKHb*+0heZ@XOoWmTvEP`uhY4Ucj^&7mEr9W_pgCEub;l
z$ven;Ta&-eAAWsK0QtG)_#xN>k;R>*{Q}||Pxd4QU-pmT$ts<25&nG@YS!MFG{6-)
z1Sgv72Dn*>u&VaOl<PmP%MG5pTb9H)w~UNp?)D&`CjTJcY0tSkgQCcKJ~8sjSCE;F
zEnktI95V9HDm^tfk;2km(i#1%^#t##DEpsl96S)zlY8!^w_^xAxdHhc*&GXRA5)S>
zcr`aMcjfa!e@u5Ms(a_R1uadW?mXk_u4c9}v*mi357W&g=g~VJH@S^Fas+c0hvK%C
zNISeN+U4``4#DBzqq<l71VFRnTcW_?-;P*k21W!a)+I&hWI4uDZUqMT2<K#!v~nCc
znBeWQI@KW}g}we-2WCGQ)oktcw0r5YvY3_C0t7HP*7qbynB?D%*z}ah@wNi}#k>3L
zf1q<$!*hkUKFrR-V6XOJd-6Q2b!W#@ZSy+b{SZ59i`qx)eOyw%vF{&8r<5Eo-lb>M
zbH|5)zL*q`&!TcH=)Ingz_9f*<7Ns<F5F^Wx|^h=daw5%ynbUVdW^hUTYhv|mrf>V
zY^`qSylu^r)j5BMRm`00+MFn6B0NlTYVUXa0dy23X4H~aT)vNu+aU9?j42$Szs+1{
zzLzt+T^CozKWxwHw8EMUFqI#A?NamXMpK=(5(4ftW?k-b0faePbE=Si`;HyrnIBa8
z0`Z=$7xB9Q@p+c_kkf9(a#Z{HNLiDRRe4p-?j!n9SG1V3L1nMlOsv8UZ<p!9Bc*Zn
zv}WeVjf{+VIn`p4u0AHnwtd&!oL&l4qG(OKoYOiR%+)#}w2N8Qy?`xfd>r7W*C@v_
zdd9dcE;Y~UD=RAxjEo#0Br+g6MO)aEAkrGnnU}$F@7_J%+;aZBiJC4TeDQA5xC{(T
zy%=?+wt0V0yVH4&-BG^IBWEk&=%|TAcH#x+OaA?Jonne<Y<~8BIJN6_ns|@_>&5Q$
z-GUa`3w(f75qFJnc69p<lh!o%{xC<ECCi0L$F#ZTrlx)Jq6@_c`Qrh`oAK4FS0i=S
zi;Ig_L&Me6HNCyOywv^WQ*P;oWk!&$v$G!0D_{{lG^wxCp(28?z-La>v(2976&5})
zmcPc-8gAKFxl2FP+5yOPBKO-QOk-#K=k7j!+@=03At51i$nbPSOCV>&LH4%?{p_0x
zwZhjKX*eoJxa7BF+x8C{V&k6RcyXn^ueQEETbZ;e!SV2Nk=Sz5_zqWddzIY9z&nY8
z!ot48IcaHW>ei{uLjwbqjV6wc3r9+h)~GAqSx=A%8cA^>BKNVBzB$Y%=3&9}b8#}W
zJlcgfknoDjsHu_*nAJKl{Tr|qwXI9#11bFRTKNZsSLc1SFJ6rJv~XpImcw8#oUZLf
z{177jaA&;>g8^xeAD0jrfMo9H9y_z+Lu-X=+-=>kXX*#mOOkM)C-Zq*)u*^FfIIjU
zUwCDj)Ap4@SR0kRkjWJ|{+RQK^w!J}xvwpybDL~sWn1JyLWLW9S;q)$J2DR;S`owI
zY;f@+m(#>uExh3-25rJ{HjT601GTm0w<C3kQeVBgCz9ds^%RG5MEFKp+GMBUrNI7b
zYE7i019&~Zy6HU8sD;iTlj?>2NSjV+eA(bMEtQv_GR*q|*DQDfx{50mJwk<A^uw!P
z*iBfphLdS6cY&X|c3><UQ!3<tId63~J)gu&P)}Q+>jpz$>Z#Y`=6P-&^jn^|c$;K$
zw0CrzPvh7i-A$0=KC=x@GYhD3msc_2RKPW5nGeNcSRwXWV(705ri4?ze0SEQWh(;{
zb{m4+*$<rV32Kfun3tcL#l^)LyfH537b9Hf#2rV6g}d-cm~5VTG5Og?9Wq!id4P|;
zl&{f7s&o`d-)qpM#dDhPCd(&2CM0xKD?UA)U){Fd|7v4Dp4iD_Ll~O3jauwr#-)yC
zkB*JWCx8IP;!>=Wh6+LweU;dJj_;wZy@e&?$A+)0J|X4qF$>}8O|6xog-Xj~$0l*a
zcJ6H8j?fmcq;7BbIb(DwP@m<R<S$DP;tg*3(n9h)F92EDEjo^V=I&uzN@c6xWNU4$
z+eDAEvf>HU$dxk=C~>x_%G7}MqXW!WoYVXDZra+NSjy@a_Mp_2A`RJNSxlf#PvN<8
zCrw<lEiOi^jru#T)VzNEy3)wTRv(9__(+W`GtRGS%;z~n>N#g^$2c7i3f7J%qo5&H
zsc<lQ?dJG(nzk^br@&|L5?g9nBY+i<ZCS)4nZzNRo+I)djcg)xb~e-1Jd}LOvG`~W
z&-KN0LeE_X?H%szea|<0pDus*$=OOS&xcsEa;N-8clek;8NpEX)->-)a&soZe87}d
z`fw$Wppr#0f13&$c09L`Iz=6w<LRu_QQ{af$#>x$&JbBvplG6`fpyK{eEg0c-}Gd&
z@>crruKCR8Io^Knu*zJOX63F>O4Z}!lpe<PIG(eoWXHx`J|E$-wzih8W6+>H7TKr~
zsNy33!mJz<tTKzT-^%_7cV0Dl4{tv`7W`u8jy&Z@>8C`A=H@75mfrI}(|8!U+Gzs6
zGl(Thq%c=^;X;99$UBa*)17hUt@BG)dQC=xC;5=E)IH?*C?ao9Glz`z4rJppKY51S
z45*7;Ew+}^ww3S{nNjm93va~ZI94;1_{!=NwJ<ji)f7Fy-AglF;6L3_oRc(T8P3C&
z6WZb!eZ@a9aa+3CP<iJovUIZ`o57)D=1WnD%W}>Zj?uCHNv7A~L&?qtvNW?GhlL?N
zyQL`8X*pMI$LK--7aSxn%!Ud(rpZ#xf?V*x6+#~)$nb8m-+CxJXJuc@HUEG_5k;&w
zk7c(mG5VZ5j<<Cw3j2<DCiY0Ey6hbr__EEm30)5Okj&>vJN8I$JIX0gt)Qb!?wL8$
z)+Ifrot%~g@A)%hN%zDkMx+)_g-9YeEreGY%+FsLY+E2N^-jyQE#q)1_-L>F*hOMj
z8eS$RC2^)Kt;~f#x<FDPthAfwKlshxv$8zy;95kfFkYueMUZ3fV-q+u#|2S0zQI6w
zW3wN-+SSI!C(K%?qSaXcv7tLU8w^7`u_Ul4_>4~y83YJgN*Y|bQWi4XS2aUmV7D^E
zf3!4xZIxST;}Ph3!Sce?Ric#qe6iHD$Vj@%phwiE_pQK(q4c<UQ)%hBL{G5UtUDTw
zZb(#9Q{(O@hLwK2u`A7}5EkMXpcExr79T2PYok~7_U+r*`yQ2rEM(KOvwE|31zRnD
zLqcQ?z``4f>$2cb+ENQ{Ero;qPT*klfUXGhb<9wT*4x2T8R$b*dtcprs*gQ0NiCr;
zENsNhLwuK@KyxnuKgUj_G-W;R10h06Yx|cJz;PxY`jpS)4j$QXQFcp*DjIcj#l8bR
z*Fa?n#Meu=79n)&`)5bE;CNT=V?ugIxyUCvZzl!4(YDMKZlj0SJSqA6p2z9cEzVFL
z%@&&f2oYFvB{zxe)?fFFKJy&n1n}U2-Xj9=;8PdtqpC7*W}E*Xd+!|;WzwyW4u}B}
zF`(q2pd`srVk?p)N)VB#sN^I$iwY`Pa&8b2B#PvWARrmZIW^FdX;PE>*2{eJoppY{
z`Och~?mzCm3)j-sHqF~ryLRpH?5Ao^oR2QZGeQC_o{8p<cGoB~7%7e9DHC1zBFO{B
zg8LZ*k{El5kEqos-lMAh&E`z^zOu8Yb^+bs*~nZ^`;%VvvhkG2-c$$Yj$3QB>$6>|
zW}}w=PZO-#`>;&~TA{1&Z}zpO5#C{2S%YJ)?k$cT=sNAp_7EFEd4Z0h@M1n1nLhvP
z*C`pg3i>>U))-_>M~BKYU49}x5Pf$q1xk3>nC`X(<hIx%Z@S_v;e>_ZGcYg!Z0~^U
zhW6%w^^k?5&$0(P!pF8h-$Z^U>Y=uFt(H-If)?80+hkos#s1d?_^Vg19xV<cn+i3{
zu9S=tah697E{0t5Qyp@&0^%xbAL_$$JG4TTK<C~jS;9N)!30!xj*iP*0Lwmk8}Rl5
zbEqcjO6}))-k?U&9E7|^Z9g(y%jlJ@Z>g5KZ17saRo5i?GPNKR)vO0{l}$n-GTM8E
z73v^by20<*fZawVh>--*l5J+_y$keQtk!!dKL?>RXV3Z!S(X_#^lf->^YM|=x~+`y
zys`Nyg*572(6Ju(T0RYG;A7h=jeW`z6=;0Quh|7p8b5Rl3}`)%n?4Ityyt6<5U`Jh
z)Dlq;nsGo$0d6}3p^pd36WPFCG<-QE(MQ@zbZ;ej0<Z<;X_`7yQeS%$1u0~)^b~_M
zB<1Y#^M%<|&E-{l_NpJHy*H6@T0f@^qm87501XYBbm&}K8St2;oL+c<{;oL=V&}7J
zsY)aFDs2#-R`!?2N)huemBqVDu4<fP0|VzH*ms|pDwfpO`314}<s-H@J#tyi`@h&U
z^cA?he|;~^wI!NY<c3xV4fDOZy$3T|rb~g?f(4aeeD(Sa42-aGL40<0_D5*_Nz9BO
zMj0Saee<T>V`Fq+B~w2e7H>_Gym72tqxsA~O{pyONG^)&(e08Mng-3xVylMvhZRn!
zI{S&~$iB$MgJD|<(h&Rfsq@XEY`HdPv0Bp?uzj$o-JZn$Af&>FGLgX)i1bM`w$od+
z^xWLaGxu@bep1eI*4Id}onOZ%5Oh^(=C1c89nl}HN1c&R!o8swKJS?_5v|LwiLbhL
z4i2w*+qMh_HZy&Ib)8O($5QFIA*GusB?qwIQ-aIv!p{_%9jkWtS6X<@>Z!-5PP(|b
z)G7+PCoiOYQi`X{uH0$A4FZLM7*S#}9c#UQ$1!&``Myl&?3uI8kvYsFK5Oj~oC*n|
zCFZUP&;T256JserLqgzpGB>zD3m+Z=K-Q!MOpgN&4Pd~Mubk@hsjRHj4O|Z3tlXQ%
zoCn<Ho{-49w<*9l*RD{kC5VGTUuY9UOGb|jyQ3?j`NJ<~YUK%JJ}fAmn(azeCZgN_
z<$Z0)ml_y7DTQuqBTOGiA0M1d1_Zws)qsD&|7m9%secM!q=+a)f#?c?PJ$f53a)Yf
z?Cqrl33_|&2dRcI^PH)1Z?c0_U{0c4=HeYkY4kkz1G;1GQdOZVw3`xK2A;yt=;~f<
zDkL)jC>!<z?FNis$^E|SQ{YxhX6vc<%Ph|ItJCP$08yutFVxr_2f>5%u`36zjh0}s
zYxVc)khLtSaq3?yI$w2tS|n=|-}UeADzaQ81j%%tni;E!0zQ|=dh+x&W%qX1Q5p}l
zqcYpT{l;%#o4*UN#nN8SF#%WNv6-##0_Ke`M-MA$D5~7Z0c6H1YH4Zdrtk$6s(}vU
zT)f_?{D$$&BeaZ>k<nIj)m|QihYVGGX4ThsLmf{lb5%9fFB)}`0azzIrI*xXxX?no
z7*!uzv$FD_3nFK4^uS4g6_F}!$qaa!5Pj}6mIJ4h24b$g;q3+4PlL?`N1w*fvgQaT
zyyF0lU^?`_z3PM3yNk^-y9YYI7TupJkS#kk987$QJrKX4>(UfeXi5{sZRk=w6O#e9
z-Ol1}>3k)GJ$`15tk|Bh@&T~DO#OP#;GnQ@gak>Wi~26GS{aL<B(QNrz*G$qe#H1#
zEoNsJfd?(TuLiY@5SIhPAhGk}sCUgZnw6_Pfc7x2X12yB4oI34MEb?zd#z)`=X`v8
z>f^o||Ky&SiNzTGpzMpf$R-r5GtUdxig1|$9KD*Sk?sPFh|$mnG4wFq!L)v*?pHrR
zb>+M9(9XQE&=lH<JzqKx3-A9t9)~SLW_Psz5`as+s&Ad7i5i$+S&`iuMOzXg;h5Z?
zuvZvL%d*A5mAy62*YDrIZ?O%1sj2U7zq=i&v-)O24ZfrBpwg`_YOS=iv?g5h;2x;;
z%-4jmgf`^&%~d-i<1ZfWfhCZ))Y4&4@YlP`7&%kj&-DO{UfIaO7;a@$m1e8kZuK{C
zoPjJQm>MlKHNGw~EDfRs!Jt`JmL61m*Y*;{BNho72d0Lkk=I!vit7rs%dc;bdNZRR
zb$)#JX!85Jq`W*{^N~`1^TB-QhP)zlczhZ8n#*jbn)I=A5#AKMpMp$=s7_{=uXZIz
znh;9tO)Ng;ME7c2d1qwr%&&-4**a})HE^B=6sg-FX=Txdzt%jAMc@aso^P1oZl(&l
zKn7dGNO3A~KI4_)ps77QfZm&wtFQ=#ps1ziyzxNcVyi{Z>|IXqsD4+d8gMpJFNT10
zVZY&?40}JV0|djmVkWO;7eB6Z2L=HnzP_;$ZalKl;dQc&1d762;zf*hZFkm9l2TKd
z%!i9_iemPNCPrQ9`e{{i=e4R1Ou@032|)5a+qgH60>*XYi@oAsZJv%ae5QCi(+4H1
zN!=4!x@9~qa8H=2;+X(t(OlLWI#CBfmW&wYh9D0+`7_ir|M`O0w0AI9MVB5Hpb(}v
ztXICB^m*DWP65=iZKZ~Yvv<q1YU$wn#UK)JSm90W%i}T&;q%iKKTe8&;iLB0dNh%q
zaUd+ZyAade)tl|k!PtH^2-F@{T&o=K7f*!*y=MHFq!$DXFsFVWmZx51$-trbrR4D6
zyiQ#D1=1TbUAu!#Bq9hSI_xq}Gr=UGMKS7xXiX2?3YI*3cG(jn_XsFhO}TTxE+;#n
zM#rrb<ctXIa8lnOk=XjS7=4S_`TZ!mG8EW?eI%FTQe1gzcRm2ddr9vhvDBJNg!$$}
zn=^&JZFU!cSSb|@jE#jvx2HI3BuT`mY2DXzzj*BXK=?<YMYCh9nS}+h+;7!MZ)mAU
z_m^sGK}D1NS-XePt5%*dYd|&syi!XyO1o5o^y(NJgD-#lnf~42+e?NTIi?E7-92PW
zbo}nH3n>WyUHVI}|JA-WCh!VX66Vj;*;biT>CM)qm}$2f_H<&fU(qmkn|!D7XrhMX
z(<aC$>8STl!Yt*oe*S0Gp;yrZzU2_E^grgsn186=qvcN<YysgD;#w)h32`gLgo!di
zd;HC-voUmlyAHSEnjce|*ytNO)h=Fk9qb2_UE5lpsrG;&cF54J&+dAgbbSgCM-%NX
zORzyEIZD{R$x6gHL^HP~PcEXC>#vY`SPyjbL@H@V@YvmJ|4|}|s$QFSmF;N32V!m|
zR#dRJ#N9O9&7-K-UgW-xM=C>8kT^IZpyLU?bYh$sJ3`F4_b{_Nw)`4@wX6Skxe`nY
z;7hEJZ6WQ#GCTmkel8S5j8XT8bpA@L|Enn9%NM$D;`<CEE`|o9zYfT<+*ySsSVqPF
z^5U039G~{K6<aR*??Q(ufK_<>(83l=%l0q8kM#n$Z{mDxDE12qm4KqaDtyl$oVklt
z^8E)EsSb{mC0!L@!ZKI>*D7#>Rd{Ps&>H@~QmDVu@ibXrgruP3Zn&80W3US6x=dr)
z{_1Q05Y7J>AF8~Exk?$`eD|-3axD=EXK*FI|0(X1-Vp>{f&@z}#b-|*u~$!Op7<#q
zl6n|&@<9z_4YS_-gnmAV-&5Mo1Rr38k2pc!1v4aM#C!rYroorchcciK87LFn-9Ixv
zMn8D*9(4OjVlKhm5hNQ$62?*cT?~}^sJP<6H-3*r=~a|Lk6f3zG666}Q%;qv-M#T&
zDn&gW#o4gLHvFEg*t7NtFPr};V{fX{Z^NrBTnFE+LG~dr>3F&gTv?p?LPtkOvpL`b
z++kJ8La%w42zC-b83eVrav1;e+R_Gb)$j>$AmBAL{N0x8HLZqC1gHGXyhbFtlq>ZQ
zr&7IjF0HryH*q8d-R`#=r_Pak9)38}Xm632_bSZ5?No7lP>?+z5iliT7iIA&9WZW{
zIu?T{Mjd^aWK3`&>Pqk)83NE(ak5y!SHaab$7L?!<Y6#+aEXrr9kgON7#$DT%};(a
z!S0~*zsGXQ_{&M0SQ$Kg3Sa_xUhw^)ELc93fUo};4jdX}|IPrvK>>@wiA%!+lU%@!
zR=j}u0B3uKz7;pq1e4svfqFs9Mh^kfCxV~h#BINL@#2IlE)*Y(EEwE%bmf>VIB1It
znZaPM#Bj1Dp(A1-Bf<L^jhh9aPy9xU8)I?m33RvFZ=G14Q~E<nz>Jan+jL;};>e8s
z%{_5s#{RCt9htGeYbB4&*x&UmM`rBEjQyzuj-s)@8NnPyV}I9LA4Ow-*IFMXm49lh
z{)cSp{~Kn^as?6a$6WwerC>)1=Kr7zfBK`N1oQ7I90mR&#p1uwsPm`Hsp1g^mu7CN
zdW+i~;+H*S*6^Il&$Ixr>X5~6+_hdtQkh8kdxJ0Pxs=wsyFD>?R^AHo7An{9mXR3*
zpQ*so_%nE^N*vq5d_fp{DSH3r_3Jo<``$uTB!RO5AazRI#226cT#?HLy-vViTHEM$
zv;e=?t(W-Fv*`NvUc;;EEl^nHI_T>^^E+1NrGqb7l74uE4wiSixLt6*i9!(A+*QKj
zcBHMU6u{P=k;)#VX6?H1X2)>@<c}T>;A(Lu4KTR3iUw#n-rm!;euo#^fZ3REeD4fb
zf@a?^827oCyXCqNG>2jz7ni8)>stVCj*DB7oMXP)@$lmVaQXh<b%uWyZO}_Y6{GCZ
zo4COwy#ew{I;n~qEaf}YNw!`?5F)hv-Ng3)HsXoF07|^A<BJoF(~BICRlXO`&ERIH
zq);*Itvu+WfP<NyO9m4q*Bpt%%}j3tIQO&-3vR&l$E2WMgh-FY4Lqp=)v~T|)ZsvA
zpi<-y)hBiAZE-Wx)1Yd{{5>V=@0z|poy*&MV50sRmTiCf(?8qS|F{1C*E_H)V9$FO
zlj3H}bA!_jZ*6*7aRcl!L#5Mmk&kh)5nxBg?H`Hp$hiG8cXMRi{wRX{EsJzy-2Rb|
zI5KWW#_hj+`k$zf*HPs5kFtiN$n77Q@uSG?PxjznCv-<?=|3~$N5<{QxcyI>bCj0;
ztswl55C6X#a~m{_%l|i0S${e|Kumq>N8wLi>C^CO0!SPBsQ_y0DmlOpGB_0pK-}FK
zD1)v)HD&MMaq%*p3i$8TV1U>5x{yQ1T`8pu(CscE%Ve|=#KI*`0FYSm-k*SGORu6-
z?Oqw@j{rX)rvr~i7}B&t&041khLO#ULQ08Y0B=)e{o`cxdqPHuJ9hUUwbdrOpHWYu
z(}nga%>H<~J)Dk`CW1a;E<0cpdMvTmRHQ=&%EQAtpANlkn)y~7dRN!Wv;wZYwAT1X
z8`&F&Z-~b#izlO@Rq~o4W8UC}@!uYb(VxNOISLv$^lZded-6J74}Eeyx|}FC|MiG3
zy=)fI9hHadmb%(39|&HlytY(9-@5P2%-U19*`sko<qq@e1H9(m#}ir)NT1BkQNlu0
zI#iqd9$z>u_d-3y*>lX5^I6Olw9>Qdpw~@Xe7k$X+OTaqCJuu^WA>al8dqA#M4)*Y
zo>JdQfH$`7Wg1;KO#nRwV5|5Zy`kl~;20l$@TV8c+k6{Y(B(8!$L~%wX7N9GmMHWF
z^X@J&z2bWtJj@0wf>;$gQUX5xkD`wM`mJE-Wcj*lB0er22BSwnRNYbfY20x5XTePz
zvQ;Np|7*C|%$A1(0OLo3`yaEPKh>!t!TrOHb|knX!TpuRI<j&9(DNJ#?nrQd5&s;8
zBY$Om{)2r!65NsC{>r)>We)!^Ejbe0k>LI+2|LOh{)1;a65PK9a})som5cfhPW(u4
z|GyL*On&ED+8_2IxUl*`?u+-7q}&f*-BdVaKR2G*N>Hxv$pjnd$4gk&fvA2=gCA6V
z{K_)kzBaJnmkuGxM0lPFIo=;`g4Z2R_&IwbFW^IbdO?eCfAbn(8(;nn+$`px^rCv<
z+=`fl$c}j=!KoHKys@?a^X&qNo607)_Ah|;CK-8)^L`W51k-B{KZWk{c0$gQ;cM{i
z&SWcMaP--9ICvdO6o${e@9goMAsHHE9>9S`9hDcsYgy^U&UP1qX(mI9`(W`tUQwBK
z6%P>3l!N9ACZ?s#ll=@_5d*tV;MQ)Ga&Zp0SQrm}%F2}H31+6e5G_K!moQrf^BUT}
zNSXeS-+$OT|295yfWIiK+Bf{~H?VmDlLUv!jo%6k68tIt_;-gK{1IT<jP|Wq-aGs+
z5%9eRPxCQOR>T@>KYZYk;QpcUITGBF;Qqoq9BtLVYbuXy+&|1kj|6ulxc}v5I|@hs
zp{g7S?*G4nD{309t*y0LYt;{Gpxuomr=rSu9@&gd5&>m$hdUDFKo7sF{xDBRfJ<(D
z(N&n?)d({7IscxFAis-B55zn`FD6bw>XDHV)S?vz{%dLJ-Ppk*8u`i68y6RVdd-4l
zCMxiLz>;pE`RtYgXGK2SQt5oc>b+8@-NmBsWuqHZ2!xjS?DBF@I2(yycJ`97sc)2s
zqh+SnTEdOUOo4wG=lwU~aWQz!P>4+@9lHP;q-E~7+1h$MR6YGf2=!es9L*ze8RbQi
z=|@Z@aDOEB54rtYfdAx0u%_4EN4zPG;@cyj4eHNs^+$<LEXsJqDKv4<o2>k;KIY_9
zZEdgjmu^I(%#{9@mh)du_RkY97z2>A>}Nk)WLsA8^CXotC$S7N(D7p9b*ICEllbUH
zHu{dHX5Z_1@jw6lzc))SQKH}U7cTyB7Xb9lu3(~<G%YRLH01+@=j?xW@^Gt9zL9X_
z>&+Hpl9rZcUHU17i=_+}^m0qVn^JTPaRK)*tJ&~bfpdtruS9U_SGMadm9J$N@64H@
zOP26)v5-HYW3XYRy@P`fZKlhQ^MwiI+RcXex^B-1E@wN9;AleNBqfV|W<f#9WRvq5
zV`DS3jN-^;_vK1*YU)g(nX5RN2z23aSSi0zTbwUjP(Xl2dBU_GJ9~KTbr;v&$!{kG
zpOx4Z;DV;+#X<=7`vtW{tMtpJ!3{5bQsD=38i}sU`&AkZXK<o><%gQC1{%L!(rnX)
z)fQR|O?s2*Wfn(v)`6Sp)90daAu;-3GO#Wb^XjAclm{p}x@dLlzU-0EKq@TTWd82N
zoS};hRU@H&ku54L{1D}GpVI?DRiyZEP@c1bZ?^arPWH;lL)H5H@SJ49JWKe^*{>pw
zOa9uHCH9@J4vT|gxcwTFCr>^C4JI@OeV~0o)18r@U0vEXbgmqyjr$oUuxwa!h{Ubl
z=m<ehhNf4LhO9(-I@^P;IQ328CmU{^tYE#L`G6KQ3G?gi?Txvi&7PPvxDj*XVY1xL
z?k*xsZIp^iz<hAopPk)%e-v>7C&IM{!o4JHhpL17gGLtVHL~Apyonw+hSb#=%H4R_
zttEPSf1x5zsPK!Z-f4#iUkq-cenqW}{`^TxLqo%DHCmo_`Mb*LH$Udgnj^Ao5CNVS
zK~oh))OuHJs`|6W-?3P1J;RhPIP>j&7PMIWT+<WDxi~s<DaCMYCQD;^U<{VjLEZC`
zyf^bE+u-2f2NmW!^775?fi-O6Mdg!gPY|5#yf&4Qa4QQ7_BPCNyMUmejGCHS$zE=5
zE-*~~US8k&3(Pb{H@br!@9*p;J2^khQGM%;WtWTJ-OvxH0o7@my#*r-<LKUueg>Cn
ztULolvIY}q9Q(HOCv#3;<>rd?fUi3a3`+Hl%f<<Ht~sR}7nR-A_GYXOBGVBkK5<xg
z%2_4sv@YRfzf)WS0C)<w=cW<--OS7<)?^N4DT9MOXPebVtN5m@B|v)qdmi}DOkDJ3
z=u5<8=-k!goM$LBq;(T~tC=LW%W~beS3eKW(c(nyO@P{m<|8NlB!nB&k{Su_XYF{8
z3q}WX<KXE(AU7pfYTv@06J!|Vj$m{oPrq(G@YItQifaTGg#T>X|1Hw{U&!%CIQWwN
ze6x{KTeR()iHhSfv9Xgu2fw;(22E0z+y`#r<Vy+;zr<>BkpJh;pOx^kGm!+BIrT^#
zhf!$Mtg`2YQ@HpRV`i`zyCz05GBRPeDa)^u3C%<$8zSmAP;HKthO1NohatRI%W1w#
zY$6>Dd9iBb(AQG0IJ(V_o!5gT7p&`bY(KAjd3#~={!{P<GvQVOx*h5oNvU0pb~8lx
zErCyQUG%z(v5aaIh3NUed6}Pv82d{F=1U6D9W?YR7owINT~c=Cpznh@9A|a6Qbs>?
za7^Wal;ZcFv0R9v!Dr7ij6GS{Y(L)?!0EZ2lU(8M0gQOW*buNlb}jKkmw{LR+1Yu)
z`*xDs!c8knkDHhepFgwciTg@eGAE!nH|uB|-PWpo?{3dVMBL7NXm5Nocw)Uv8A@ez
zN41-tY9$=*qZXCKb+VTwylxK54br}JC(xDShi&?JuBc=zZ%<5G`tam3NI}-4dK5ne
z{;goq{gDqcWqGU9vEtB#J&^~KFWgB2?Z-}aKTK4KvsNO$&c1J6xz6jt2~Fxp6)Xil
zyX$#wx+=#3l8G2NvydlKBLYcIt-9TXGB3vgy1GXUzY>X-bppjMnBnO#kCF0%6WUbF
zZJ-k8tZ<8d2pkdk)B!gP+E?c0%P*s@;qo3zwQHst{US{3JoRRGxXfX3wvOv9=c29H
zXCl!3jGK$=-2Q$_vc%5LPJO$4-y0vF4|eUKLpo^q{2e&M$X66+Z%!eb9+ArRJk^6n
z`VZ6QfA6H=Ye31=7Nq6nEk1LUOnneqYav<4b1{kE+}w1}X0lI8FuPW8at$Zh;>Y7L
zp?cMO;m3P3os~gPqjpVNJKXdv_yL`;2^+k}Ogj0&k7L+qqq6eAJ-g)Jv&8DfN6-ZJ
ziWFw8O@!$(hYnZL%Y=jkLbyA&lK*{7olqN|$0UP8Lx{}0UZpe`td`&`9ro)A6!JQ_
zYU$&!4~fT$<8}3pkrCdBJQ5se40!Sm&}-GDZ}a+|cA0wa!%mmP#PBKWx#O_bMj#U}
ze~<zEfH6S*vQez1Wyksa4TW&xK`B57fnYD`IGcLc<UHF)B_Rp4AO|phcL5jDt|st`
zAcVWgLXLkM)djbde`?`}i(P|<zYA34i)V7_@XzjU^+@|0w{Z|n5S&ubPjuc1a)QpO
z;Xo<4%7C)9WP1$Z=GHN;V_p#yqKNbO8pk;*w=U*Bs{OR;WM*bI?7Cc(K5B*&0Zd7O
z@L5kdJXDwtbm0VD<d!%!%bqk&#;Ip1;6x-Pu(hdLX+i~M1Mp|htF6r&fy50QlmPA)
zn<{ao$aFU_ur5z;YimQ<GT~qqVf5`_XRi)doaq(k;^NZNFhbVg{>j{lNxZKvvHsD-
z1uNUE0GeLbuD6PEzpEa_Bc`EQ3>`MXuCvs?hBGAtw8O(CeEY8B(lAB41g6F*-gc&i
zLc$Rz+i&#-@G0#h%?}p)^jzJ~oH-Myc`fl#Tn!-(cH@*z;x*(3o5V4ELqkKG-9?N@
zcJZX&#q{$XTwb*&aiH2ia)Gyv<F~T2<8yI!O$JT)K!-@*2J15<B$4g%%PA;TgWD;F
z>DnV}3%`CnaIWkWK6dQbbaGJ<C%`2%L>(FcsDogV7@$UvpkWKc+g7^@5=Kh%;(a<e
z@u|+>3c28)*^B+T3<nc*mD-|}7lTAsYx^tRWy-fE{eF%~1po962hkh$vZmAoTH@nU
z1`?R98nSGEk)^F?qC39~DbmfKrJNxNYFW6Yj1%C^9{_JZV3I)Gh}GsC-*)#U&d2Db
z>@(`>u@`{wO?lsI;lhi-njp_^Sq<8FQX%8=IKV<BJik_;Us+XEwHX@o9tR>j^^qQ^
z{_%A&`B8VYn9ba0vkz6C2JI<6F3a=tW@-EVXRu!uz*F;@87SMZbM)xb9Sa=W$?BpK
zh)eo7G{ly8WyG^7Qwax)|2yri(WE8Mq)QT-$75iODV3*}dqI4-cPR@eH`J-z0t$9K
zjFB@SAfWCznNIi0#F=Zg2??W&EV$u(&0PoA=~;Gja@s*46^VFpZd=BuJRL_zM`0pu
zoCN3H55N=xZZXALqL)Tc@08tKzG_pNW@cn$ByAe<;=pThxk7zmbJ1|j-gK_}ZlG{>
z)ROARh}i691A81;J9-rcuL#Ok!x)xR!**S*YQ-5l>@sxwrBF7GSgF;&nB&sp3;N1x
z-Ow%Nn`wt@iq1rWrkyu`4i8g9nQmBHY)nkL@{aHOiHQkvN=n~v-@g5tpAQ-x9i9BC
z>9M@D6sY>^HYnfdEsx~7I%=Z&E81N>G1Y9(&^Xy|Q))o|@jHc}!F&JL>1dOj-r-@=
z)C8BiI?o#$Cbtklv72i+>XdU<2Y6#IDkT#i2!_6kUUcu$GOFP8z@Rw;K)-2BG5RUC
zy+D807g#*B<b-aec*JCSTVs<=Y@8Wt_7ONq69}$fG)^$@cwj+5>$&%?zub8~j7hs_
zX}EaOw_SPLLl8GXeIrCF?JW}BjbK+JF%p{3MAsGc>0MG(RJ2cPxqyRR0l*J+5R)rN
z&HlFRyi{VL76V$59!J%2>e05g4)R>%62ysP5+;<zPdm1r*DfxYq9MNZZR{P0Mp;JY
zuj69p!4f@yKbkWT3xDBr)Ak`EnXV%7Y|jJS<SbFyAc~pwZ5=-#(;qGwO9j23TOO1=
zpY~X`J%$4p1aLD?0Ik?_>6h9ydT45uIkb8#+u-KO!GU1t!o@i{&MO|8Hy6HW>{c`o
z<3wj*!jQnnJwV@L#nQX-@>d|Z>3T&2Jx(MNoB)lx9~$?^jIbJthYug(`q^|2{poU=
zS_N|*at*||*{ZW(tERoeoR=vV(A%F2*U)@e?qt)$jhoJThgiy7_bbP8#G;0#%S^Oq
zmhdwiL?=KSWI=8u8SlcS1*@HD)t*!Leu|3_@i(t%i1hJp@JzzZ?WrQEB$ephZKVF&
z%F0TLmbvq|+4i*f;TMn1<*we)HvIWP3bk2;dDx?CHSo0nN9A7_y)Lvj*Q+LcI2)4Q
z9Jtk8TCUtp@a<c3$j!;&EG;W*2rBQ%!3n~uA_tb?TKMbitOvPQ0f(XR*Md%Q`)QE(
z?Th!VI!%l$i!)rl@!+0#1c$plZBZglItG3T0z1uJHl)IO>b=Z&!Acw3{Cv{uH`cch
z%Jrani@C1TjCcbA5!6=^Vd47rc$^@2qD1*qC0pga9-!{!$LT)XBn0}hUpJqKG0XPT
z-|q<pUpn+ADJcmRGBkt?f{R=UzUBq$Cr+CK7Ilb$U$e99v|`if&W&sGmm4PQ{0wJ1
zBge=<S;f2NJ1qRh;#1)QJA3=O<e&rjh2aujkW11??9a+iqi1GjCKJf^&6oWi4}5)U
zGxa_v0K{Ki(c--5wg~rUCv>i0WgES;phl)C(mC3WdV9ZQBmzP9wTe0oTDosKO}(f4
z&bi;ojp$o-;1SwdhA;>kXEyx`Vpf&YXPti~y+Ks&R?;bpqSQtA*JsGd$)h?gK|J^&
zFOOGmEe3PW%6(qWrc;io9+a0;*Ubk7p+LWu1Vb-i-cK4UAV#e$zMagh8{3ot`hR<D
zM+UtE+%9(7hB{CUP@c|JUy5emRXE=F0jn9};X{`h)w9fbt`4*x$Z>z1m@VGu#c1fc
zfkx|q;*+}NH-&E4+g_X3S0TPn4gSDWDk|Lfh6%m+peU(vYo`w*zEp+z_G@u5{vKh9
z{(<LgN~G_VlO*9m%^YPsqg(ZYVYQT_s}UFdCWr@!%*bhI0+u}--G{AfTTuu)39)4N
zNc<bLP6m?OZ;ORT*FPa9!EI8341q#!(redH9p(c@7aV4=tky<B7eLL%_AI>n^y$;q
zU|;2hs#P}3O#w@{G*PQrvL=bmT=~hIgE>Eb?bL*e;U1q|+cT0SVYReG%Nv)RCf}t_
z?#x$Fj$sb!V>m^5c!X(tSG+lOE+OL`9Y9GohMDv<zrNY^i`n`Zn2zh>M()#QUS8+&
zIbPQq%0P*O{s%=%aYB15ed-lOw8!$E!Z*I;pTN$oR(C@YUMov(*-UFJS<#q>JhkVZ
zT(13mRYYIOQmLwOt+sE~y1Vq_`Rn{P`ubyA-HVHdKz+BAft_7Gxy$OtR+FaXTh-v(
zojD*oDPAhA3RHbiohrV85}FFuWU6@5v;F3c%w&vNCABBovFLt9Dj(oF$bjMv=l67t
zH}$tpX1ztiJ+CCVJ7}YvI0~)C&R?r-)B|}Ww}6EUGX&y4DBE0TKHl2B#;?7prL1O#
zW!R~p2y%?>xC<5YiLcK3iY%A2D~X~kpv1{J3q+xu0#>6JBK2J4KA81AU!@Uo2=i=o
zZI4@ATtmmXFa;M^x5Z;F!?#=D4bUb-73h^z8wQEdA`^C#wpQL?%;FpdS?o#JohH1=
z5F`iV$_V(uEL>mf;kIp(3Ul8^CAvZokWutd37vJwgrUoVBkK|f?Xl8Sa%*nYjY`$6
zI{rQ#eZA#QdCwy19hHq4%CjvI?!Cb+>R4`2ep2J?37t>Ezj-AGy6B=XZkPgQQa#H?
z0I=<iF7H3e)Uv(5zP?Uf-Lm1jo}Q8YE-7gObzECpy8#$?SyhAiUZY7#x}D|EuXz^+
z`l_p8IyQAZs0PnHiA|mK=3M~v7><+@rTg&4S=XQdVWqFmo6N(h)7nQhmrawFnhvui
zg(~vg?Gc*&coD6Q+1ViZE+hWjW_@#W2|aqO;!)wk8Z2H^8N0|OI1X&kDf3jZwH(aO
z5F$O};Bk#@(eQIKYKQ*!sL#s#j{vwP(a_)D{~DDq0pP}loeeaZ=gk8K6lS1|M5rR2
z$7Zj+n=$mvXU2~m3{oFY-oi`q#qYdy{_^=7%pYn_OPrIctC@UBpYbKLlq15vPr{Gh
z??w{g%NO(-A7$^+i(Ls<fN!_cv;;>NwLO_(U-MW-+Da@}Amd!yY%7s%B7#<{;(#7l
zHNq4JXf@&h&Oh^}I#gM_#~tG?4vK?()FULt?L(GjydFDU`e(LEzv-+Dnl*#cUYp<3
zW<B|m@=u?xbCmv)w`shlAJi%^#Frwj{q1_J!n_h6YFcRL7uv|IR}Ec5#90r-&oDS7
z^1P77{MFrPq=CDR40gg*A;%j_<b1KVkCq@G+vzPzPEKaHt`v6;(GV15gQ&0X&owk`
ziQ*pF{Y5V3V&_lE&w5=sf!cK;*Hv`AgU*#fyWB}mdG!?|-tFi6aW~8AW@4>+*HY#`
zs|hZZ2+SIFr|qPrr_WVnV{U~EClwTAxmqm_3=Q_u<hWV{Y&7@MFy&ZL7ukCfmQ^5~
zfSVT6f2(|;xjQoxaUi)7*VogN!mcT6kVZi3zK#M|D#x!9_?z_hq+iIB5?et_-&wWm
zM_rfsNk>u^4O?US5DR(jLC$M4sok2Inhly&J9Yh5RqkqYtL(BnLc0sR$+<Q<ds9pU
z7nMY}#%pc1r^ozk+qQKoN01}gsx#4oZ+)=bk5U(d6O(>huiEJeKviK}n}hxH-L`#;
z7Yjw+e9W0x`dHYz$Uj^kK8gV#C)fDYnIVfJMf1|x%!DUQay6k(g3^3J0GQ>f4`^)_
zmw}I02k4P`N4XqNm><VZYVk}Ao1q`O0F|A@0LLEiRF^o*R~`>uXNtGmJ8{kEK~-0f
z=Uy^uZ;i8l*%LFqQ^(R!b+FNzyIi^KU$HsJkYXu$n)PV{)#Is$*ahTXHK(>txZ}ks
z3r)O2wPW0*@T6I@t}q%iXKj$b`+4KR7cOMJsj2BC&9am0LIqlHB*(RWsxkDrUMS;(
ztu;6t(9Jg;xgD2i#NJUo*Xz#tHdtVWC|~37H_NHGE*(JGljBgU6#C>8iRg;TpwhFJ
zC|W@Bd^}kC?Xt%N5!^3SY3Bi>ZBwB;x@;_H#9_B6EhneF_r`MR-DsmxddkbNM6*-%
z<i6rQ^k;s*MUxV5`KO5r)NRRtfY1p}cgxR&6|A%@!??RRvF}wN41apO<uwLlZqVKd
z4_}h81LG)(S-V0hEeI)u3M1a~ywpp<8^AB0W*m2~gmw4;Pzm;)puC-oaP$lK40G@c
z^fvx2s%HeJWII$|NDbTf>AF->dyKi0kQ(H7Qrv#&)2Ze6p0e?SlwY2bMb-XhfU?_U
zcfVa%wARo@PG)+#FO5TgrIEyR*hNH)nFqDtu(cZFJE{|3TVn+!Kv3BRVgxkt)YnW4
zWV`;kvE5OAiXl+5ZYaBW`;6kt*Y>Z>%TZxl-KpZ2BXu37OF!i2<Wvevc%b~|CmHCr
ztp|6!Nbfg^o_zda(VB=+Vyv}R?Ck;6=qJ$_e-!B_ZDjz2Kzb)J&y}=&?Rb&^L{;*=
z3Z}}uZ~{72{f$7lROYS&)nR1vfGK#1cYmt=_#siBoC0*eqM(>Jz6|8>gK<1BL4l_U
zJerFKh$qP3o{tDT`pE_3<!o8#_qR`}BfDf{iOII6m?R1(;={B^`1vh!c(lKMC35`s
z(%Zx^f$Pz2H_YzXyByE`=3`229aXith7I)tMxNKRD|dXO1T0K3b8*N%T}R0_fnz4t
zsJcjLHMJeicm;T+9IyZ)Cp&dVEl4ExD!4uNq;C#k0H^R$hGtIV0EZ?pnnoNaH=tF;
zqQaQMFiy38{GlI-+jvwIpUGn{X|{kz!Xlymn*#W?3||~9)G{9y2&b&QyU6rCT=T$o
zA-`)#bxz=;!5f;wAhzt{C)E-SVC$-iuKqBj;;~yUAQ9iP7TxR*tw8RR1d8wM-_sQn
zwVE#|D6kp#+TEbznJzCWxoKJQ)bj2&4BEoSxPL}qS9M9;!mnLZ9Gz^>n7RsCcDJqC
zhDaTqS_%~-Bc!!M&hlxp2l>YL9(oaR2niMrKVQqYr+O%4*4t9Nx9aIzUtf=BZ$Dx$
zcqYyLQ+=P_?wW$eJq?XZfWrQbV*bsC$zaK|O&PKCIT(98dt&ih@+mF6oC*QhNuA)X
zH0-4hK1bceo2r}_$(KCS0s{jlPhNR>p6!eE#AH~BMPBPdzis<XmT5_?FFp6QfwxQ$
zOypRuKpy}u?!=4i*|Du4PdB~<x7E)Fbrr}Vt11hHIH4r8o6y<Oo!-UAU5Z9C3ofJO
z+o<tZBuMkf>|0>mBg5fX2B%?#F9?W_>%jLnT{Zdm{8`4|*3qN`)>|<ix%(ZWkMd5|
zAR{Dw(&oBYAVcl8g6AC6)6;V)NPNd>bg_$nZ@b+yI=i@NA%1x{1MMV40>HP#^3p0~
zg6o=TU)8;omOM;(JeEzF!YU7kn0zl;x(@H;f@bx3Kans3;<^pG*<{X#`Ni#|W~S+g
z>SO46HGw!LqXcJ{?!G>x*qzeX7q<1v0oQa3K}Xj*C@M#J?=Ab-cT&%-TDp_eN+~V2
z)m$9dY&FQX65<h3o-RML>^kOpTJ$l>Cp0n1X)~PbIp+0A5b3KaL~&IbsKJG7Zd2QJ
z95@H{NsN|rE?7E&I<l0Jkx{+8shOGKd{$A*1BtLpyw4PZo-CCVGJ4em26sq?7JLVo
z1h||CmQ?~~DzCw>@9bU^lrC4gBI6pL#kTq?3Jfp`j5^&_h~eX&o1bUc#&^)w9U0R%
z+l{Zfr>CubQA0z6Vs{x(=L-urjV&w|JuI}B3$oB|y1Jzfa$3)Izvk$a@PH(aZgG)@
zqCtM=>v?hxFGjrTogn}p*zR||q^7fN*+mFemGI}}&a~@CX_naqKu4|Y>MEXUo98;W
zyWY0?#zkkI*QyOpe9$%BY+?u!rH!>HG--2~r&Mne?Q890!!Fx|ITHepH@-3W=uJ_5
zv5g@Vhp9=6&qOhRNIm<lN!O>8?(oOA09_Tt1@s=fLt}OA9-DpuNf_FFJ<xgemL%Od
zCP`M1KaRaVxx26ycx6!n-c~r`5Szfrnp06IYP9A}L^f?6n*Gi95*s47AwgENM_a&k
z^L|Bxw@|;s(17zb<E^1Fy*81!-~B(9_-<Gn%)U{IujW{Kc6Q*1;~4+b2uNu>F=$Wr
z*r_v(1R%0`+?%Nd;WOXai-yVR_J2vNM{C;J+QI?f#^-KgO3F2GHtFZk2ndELW6f3L
zzF!wvDwedO9ylApp?eveni`1tn6=vlKcJ3}k2jg`y>walR;|EL`O}dOEq#6G>7LAu
zc}?K1``~$oSVn5A7eGXMYQ6wii;w<0*^Y}eAhK6x;f#Elo}Runr&-m_G(0804*b@Y
zFUCJ7#YaBvM$dH_kSiCCc?^}i0Of3oHol@Oqb#wnP&z68H1ApKFI(|prXh1)k(?_h
zLK2fe;7k@LjipDu#zKEf@_fX#k;ukWg=p2+vOW$!phW0dYs|oq2@vga?lF8x7GUQ;
zh>t9mAmhZ5-lEZ>&Z>H~z*g3!-Ai6IE%QiDg?m`;cjY&zeNs@;7T59aO0~KW5P*2@
zPz4+WIAzs5z+s}overT*$$+I?LrBxu8`GqnRLT~<+BTh8+XnwiU3>6l;91^ek0x^C
zPH$)D!<Xkzn=qbYd<O%F2c2u`S3nlspe5T$%+2M_xzZ3o6{}fx5Rl+JGjJfqf(&IM
z6SyULp;~*CM+ESIbHg6=a)P6a{@ZJ0-+>|(de{sifPxP;vSA5$F%toM>ib0jG7$c>
zVyrQ-+AcCVIXTN>9LQHOF)^bn0E%=8qB%!?*NcMplg!u1nuxs{$UMauJ{I^h_d&)+
z;`=#zU8$KEIeUA1VBB&&QD5N7Ao$WVEMILZk0Yc6A&Rq?$sH_H)35kF*Qf?goxQPr
zd!um3vdhS)pn!ov(}#fAJ)%eM_$esJ%F(KvX4gjcSnf-MfY%CI{EJLVZ>PjkI)<hi
zx376SO^0jR`rHSLPM8Ey+_8v3U~)vH2ig_aTJIH)L)%0A)g=T`uRA{ZhtLfqpka#M
zshIPD&7RKGDPheSFC2L0ptsm-l@D}wL%EAqIhyCD#kMt>5)fm~l~uXQLq$};S`jaS
z4K6F6x4${GK~Vzu6Yb6=3Ekb@>pr94U)O~f)v}i!<UW?tQyX#^Dw=`>MRa*ese{X5
zp?`yur62WC?&2Dy8EW#{%^{5@QQO<bSOmt9bX7X>FYu9?IPWkSGpmwIiZARCQi9_a
zZ^}JsE->?ZMe6Nnw*?Cfrg1|dmu=zvn88Gi_r9NM8x>{!$FNHu3|6P)oQb=-!*<;h
zA=?s#wQXTj(f;~AV5R!3$U+6``$f-R@zW5v0;zjOP9+aeX5BWmb0S`NgnQtaM%#b~
z+<laypn(78(1z6)gKr<8-APPL%<|Nm8H)DxYNtZeQlOzdxJRjB9~84tsP5I)ef63r
znf2I{%1foC86hJPdN7aTbq=}D=(rYvrP$y9dT2f^v>AJH_!s3LI&l3ns*E)+=>Pl(
z^k%^B^~1l4st8AoXLMB=JVV7+zyD3U0vlJvYqzv|C`9Sub$6>`F<N5RU7ZjObMRU(
zSc+Zen7=zBIM1l6FSeE%lgVZ?mKiGRZa?-ByhfRqD_z~_Sx#3E-H<OwX6B$~a(8DG
z{p@6xV(nP1!nSq82kMv8|ND^TO!u;z77$`g{~lF9j2Fmrxz1%>F^_fo?9xCpCh%G;
zLRu3tDJBG0Pvajy_W%GxjIP-hn0<6>52J;<xVs>qzntsh9~+=I+oEXzLR~I?FeVO>
z!3#PzyRBX7CuWygBp1@wosyHj)wyZuRp_fJOA@??=c&Gv?#rn-Y8HZ~)^@=|E4oB}
z<U{A(pD~khQCxL39hd8rE^VS(Gh6hSP4oYJ)+r*vh9}gyzR?^(!IY=CN|xFhoxxW5
zsZyJM+s{!u9rk>T5go*?QO&EA-dK8}T?mp_=$J;MYaqJ@y)%c>5)Js!sz?*OK}|OV
z{CrloH{7LOi|PH&90gO)9YD$a3>rwMd2&V1g@kSK+POr2i?|N=zF)JZw~8)oG>QdE
zF-5y#RHq*K@jLudR~ci<L1q-R<c>wr?jh~`Vu+7u)N)rR^aEYPXH@M%(H46n7SE_K
z;_P{fs%nIu;!PaECB3{#&#2&tCxy`OS?ulRw+I}#Km7gUqTp>gQ}FNR;Samd+^J^L
zJNB~^$_?iUMGv;%y(03SLiTg@+(vCnvuw6u*}!DV_`PKukmoLB<o^EaXR__tk>R*q
zKk{&%_P7U>Sf#}Tr^50<8gZ>c^d$o<G3@wKF(Nx|!hPjc!yOBYkZDT;06c5w=T9BL
z^Bepd;~5_rFd)ZVk^?Ah1uRXQ68Q5VZMb_3D;xcrR|r`e`}V2h5ANfGqT5dp9Z=a?
zWNBxT)qZ<>0T8mBMOUinzP6bR7w4i5&~z~cil3VJ&3d1t&#8$6qzQ^EDEQOGk=BZe
zty)7MM9#aKhPA`>n)?dMDNh-!`ya{3=#EH>?e^z}Hea{&x<H;*JzU}_EhA&p+YE>@
zhRI2ftO-f)=d^L&k_5D$UUHwjui4ogMrgUth{D1`|3%<hDe%;>U8exBPEYgHr_cY|
z(f@8~Zhj*uQA*C>1-`DXF1V7S=vjp+U^SyANIbdBV$H+1uXChjbBqo-i4#(W+!WyF
ze;RkY<V(gMcLA7!EScfOX)M?6>v`r;LDyy<*~DY#f5j#~8isN7M8F140Vquj1u7@4
zl*<wah-hw8b@yI@0oFA|9Tjcu(5%=;Ss6L)Rvzr^Srmx0Fz06kDu<UH{&PpKA^;p%
zSgUHdOk!4+MTJ)&71J$APM=rj89GKs&r9r0cs_}5zjbnv3_`(s1xi>!xb&QV_Hzz6
zwD!`cw6wJRF~IYh`XS?X9(Ekdm}39}6b0lpXuZJlZ=PTUL-Brfp3%CQ-4+q9IrB8`
zXZL5S2BZWg+J9sFKI1$3)q(U3KeT|JK3xB$kNG=lpfKB+DFXjo9*6Nj3cE_V)n-JY
z->(xa9i5$>u1_Uvqecx+5=g&<fiuE;TMZJeLgAqN0gAp=)QASFuw2E;OMrUs%%Ao7
zb7RJcHD5Z2O)|i#zZPzg+N{0o#FCMp`0+SEoM)$5s>VY&zvK+n2hx-+b+CDzzhd2^
zA@TN_e8jW3W+U;nMyGl0qUHw@!M>ZDoyr|5`udT5VIkiyMHhl-`qMTWbR{JgYs5?k
zqUk21th?f!0J~=gS$GGP(Z6KVq&p61B;<m_D^(P!Hr05`fuO3$W4I4v12vr0(vRus
ztN`i$$*eu;z0r{IxiKO!iD#yy@(tEWA1R0lMf<WClMfEiIqxDOdPn=EgJ?%UGzhy%
z1a4Vz(Q!jPrAYnKDv$?o_qMhN5>dW^fqNIo@(T-zKkZ&A{lu2z0bmoJ!VDbksA!D#
z+SlmY&ICK4jI~K1Y%}>fO33Ch>6&K+ooQ~>-t_iXqp}C>*xt-F_flGZQ*|hpZqgX{
zWvir<kGP>ob&zkOh8P+<koKqGy>78@Kr}Qov{ss5P;l)~gz%};h7AHwXiCARyL9Us
z77Gf{Dwx(GivM({#i-2}lu}|w-vj0S0Ma%Px8@@GZz@%O7X>Hmjg85>)})rFnnw-m
z{g=d)LTLPH;}?<{Zf1jQYIOcXVW9v33qF*V3LAC4$<EG}tyGD3yCOUQ*fp4goo8CA
zqu|^turVX9X6-i5!G9=QTU*7DW_{TK)6JUSJ5!n;AM72^1$p^sNtP%5t=b*Rz-Eo{
zPqCKow5=+{EVe*yJJ~5;bt#ld`L1@=74L0E3E3TG4*|b|Ys*(k3a9qjhoH>oSz;hI
zDh{dyq6A>CRSRF!pVaf5Us&)BXVctRn2!v~M=;vF4gg9eiuw+p6dBu&%vVdx$nXOM
zmW<rG%dsh0V;wagF>Zy7@9*Xox`nW<dt&D4%y5HQ!eRE?-Z~!VJ*OC$!u2~O&)Fyj
z44IuGx4*~eCFOUyA+pl$UVc$gYQXqQqPBOs4g)L}BPyf<e0)<qit_R%OvdDb*;X%O
z4notzoHyY`wpco_)r)EDRl&`ilZ%Bz0;6jQ#|)Q-MVdIbh~E|$6j&U~c3|<EB+Spx
z|4vr5)dM0CsX_X-n@C#-;!nw<O0V(p;dZv~85|7JESoUy?v2<rD$nKCeCg8^g|yxx
z(S^s4=uYc3CsF0Zxy-Sq8W2%u>|W<}dTc1q8N8?9T-9hES}AREb-dDs?u>L|^6u5;
z#@XrOkB)M8#Sl-58g;Y>_!DecUH1i_Y|e(hw3H2EKtA-OZ7+uc<sm01P<GUCh;1nd
z*F{u*Uf3859&oL1%tV6dIRKo99wn6>mB|<twZCJ~Pry2zr{a0P#f?a#Ni0~4-ZlVP
zxFB$_B#;*r(}I8axOOlaH`+F)iXfvnkOi4MZ{N}lA8lw?bBVN1S&Fs4pm*}>ICyl#
zn8tubIr1agX6(O4no+;2IOAHqX7h}S|JeE9#7FcGsX8`6wLG(!fY_`zbKp1b$P3JY
zFi_-$>=+7gf>Mdc^WgjH)HdHAnaulqwQh-mj9GsE+37FG<}yBizG<{ec(LBdLJjN&
zUzPQyDNY`9f)lL4Zof@R279HO!1t~QY{Len4HEH}V!eD#;HRAGAvsp_wi$HmASfEd
zxE|oO3%v@@SeO?@sBi9&IBX`^oM1x=S7i(zdv)G=?2L3PKk%_mYSp}$*vtDtdxnh%
zSE(G9c(aJts0cO7PD<CXUN$<Lu<eJ?TS|7RXF*@Op|mI7cO%fXBxsHzby)L@6Uw!o
zCBgk>&;mtjoMw(w?tlAlG4$W;s-JAP5=$>!*E(%0Iu-oN5L~Fm?yYkAa%~KM4w?Cl
zW9P|Pz09(=?+x2DSamr#?#!mvofNmW8!S!YFgN}ocuTTbg(Tk5bDPmPO4uHexero)
zTm-lfLZ7by@N978g^s1eIEA|<hTrTi!?}CseyELm9w6(tpfX0`#^=-oYNiAyx53g0
z<`zMlZ(pA_zcJh@-!V5eZ5pr~%>RtDm>U{Wu>!E(Nyn+Pwj|{`RQCAK-8#T`9CNpF
zLzii$RE&QN6(<GgtsGF9j4w$^S8}o=*|j5}?H2}OCljqV*}(frsT|%h{m<hkmCFzy
zVjT@(+EHSz8Tf$@hLgZy_X<$|iH9ZN2T2?}(Xsnk2XdA0=XHckZuH{YjcRI5Z^{f1
z0jYO@P}?3waIk~e(#@kjRdGmY9*XI-&Z}b!bbCWJAH7etBog`64418nK^jDAWzBFA
ztXN9q0VNWS2?tdP`I5=+7bVEy2kQ-LO+@!pG&RZ8QlBYX27d>4jQIZg^^5tAtb6@~
zvck_pnFY#Pn9#Yoxj0)8iEkF!V0ofS#RHO@Q|noYJ@@0ZxlXlNAap^U5ajU^CAiG~
zkQ%cUqz`VjQwx0;=QN43*_owlPL7-~<1y>?j|22OhfPM49|P8JsWJh_)+i{}QkWOS
zG`3JoGxNFwtEdmS#KZ7&Wp$yTZz#F6G`%|cvJsG=YSp@4oUH#{dJTi=CXJCsinbsS
ziGET~u<LwN8p<0Np{n=mWi+oXld<gMhoyE}Ns`TMEZ-Lfw2_xIp(5!Stc#@g>Cv7R
zdS8<wjSLJM0s|I{#yApO=4SIR_S-bjJPs0b2!i|*#PPX!-9HG0XpoVVl=P?6pi7ns
z3IsPyEFtcV<k;T>M~~Jdy8F7jKQ87}REPmm{)dv16y!z#VrB`^TNTOFAg5~v5Li|{
zkL_?!?V}BJT`Ca-34q!Tpx7w;5pjx20E}mDwC}*(%C}G7Q!IqDJm*2~2Uq<8_i_Nb
zMS$~uD!1fLF`_L);wG#-U9VkB0Hmoa;s>apPhn)qiF+5x%)`mFaU>N`I=*_N6;PB?
z6g#XBw)Njyd90a%nzhF3w4JltAioUAYZDf<EJu0ns_WRc20bnsaRA5CCiBIol}?5E
z!9Wcn&bBRba=-`^a$&WR$+KAGW-SZLbRBCHznQe2B&S}vP>zG-fbm3))XlOXm-)R5
zd;mSLeQAPSegWaBJ=bVPP7!m+b#!=ANq8g|uheix+<ET9SP05bS1pStB2me7x-=mE
zWt6HIqA+q3x2)JG9JM`3&?J=za7<8rV1sbayM)_eYDzagPo0r_!TW3~GI0V>^cAAp
zt;xWJkP^sqP))}u_9ImE#)TdpN?fs&fVgbA(hL+DN>BZKLbuwjCUKz(v#$$9hZ_Rs
zR?BYzi{kqdXNkms_RI^|XCe@>9Y$`Vdn?H-NgnITOm!xy?u;qr{l{aiDoym3iQ2``
zZr>#1B|IPXau*Gn`bB{{P_Mu|E8EkA`I&hvk0&OILTP19>0<es(mjnH$8i?5>u7f7
zq`Y{CL2g=?U!(m-aL0A&0+)>#p@lDolR$AxidyizR4)Nj_%_F5L)!+k?<q@&>Q++@
ze8O3m1<)IZfu{{|Nfld@`V4Zgdp|5dP_Ol7$hKYVYj2Le7R#EBXImgd#=*iQH7P^l
z&K!v1Hhf&~PvI^NK%E6!JyRx73VR36-Eohi0*lZ-n{yd!lhwz@$kveU%j~wW#BKs%
z5fO;1%C}UxI~vxZB(gZQ^-D(iE(>SMOR>#<F(V;RuRO$G1}xJk84(q$O&MBcuDe6X
zCt1tT;1?vUusyyv%>b%@?|V%*7qWKzbu=`3kmaPx^#N3Up&uUs%FHtHGPvw-PN~z{
z_=aA2@u9TOLpRRJbM9NV#JkGem6dz!qo%~N2sFE6n~_6y5FAxrYm~oLNT%<`A}lP-
z6UcAcoz`!XO2gBengM*%r`WQ?dy4+K%)M66L4}w3{Egn}$iyp2loB3pd+VtZzNMvu
zdL@7%!Esi6YsG1?pwFe5`W*9dFSeX=Vaw9lh%gl;;aTWtFjHz+<l}D<+Lp9njoK&%
z^V%<S+>*@B9X;5*c~4&-&O5U~1(oP=R0Nit?~%AAX>Y%Mqb|E_>4N@Fr?dN1g43k3
z&5tWGfVQ2n6DD#+<%XB%eji<(ecwnh&)yqxI~}*x(9}<1k<Y);x-KN}mK(v{IgDbH
zOoPob0hPd9MG^#%C%q96hYkw`_WD2+rhNa!2Q^}g-}W!^b@%iTFaB0LJVaLIWNDsT
z_sm_|d22#u@-0K4AGm=CvdE+l#{n8>gj$1R9Gkv0&^W3CHXM^`lYE5%aB<BPli=L1
z!i|UdLPljt#w#6?K~EN*L6GKV5mJfyeKy1cD=lGTz-G$<W{NvhNzIqGavjn(Y>OLl
zL1z^Xevo~C#<Cwm!iJG;vL#zfWy=Aa9;lkI=f??7Ut96h-}40KdY|!WF%V$pZo<CZ
zUFVymP#6IbZEs3e>~EGmFC6GN0qU8Y;0RDuZftU(uqY_k^Yj`)y8Db*1zXZ@>0yb^
z!l8d63)LZs-dr+@-x#<m#k2KDS!6M+-#EEG>?}vcx^MCJjJ9girE}z#UJ#c{$G(R|
z)n~vH<5)Lf1dmqL<2Vi>5dDThEfh-t#-1|IOCqLq8c*`q1@xf0q&JmJhVI$EK-M_m
z+(8ThlWcjdJ3WBuy`O?gR(-rU=L<-|Z9(90X_)YF0w4KtP$&;*=EUgTMcd9ST-HBW
z#rVz5t$LJ+9BdgHuK)VUvZh@%UPp$Z1b25#?np^`xB+62(Zbp*I?+NAlV)j(`L)kL
z#50CTP<MAMR2pVN*d1|S=mU$O*D;b|36DJl{L}0G7T)$c%w9V&xO|v8R5kTQwF=R&
z$F45meUHh_I)Gd*07NBhAohw>6{J26SU*3z^92s0s4*uCQ<pU%7Bf@Rr7RJ^t}%qj
zYL}W`6Cx04$!CKwJ2~@T3#$Or*brFJuZ0xm(|(+lq>b$kr52!|iXy>S-!4tZ3x~Gr
z>kr&mPEHQO$1XIhRNkVtt}f#};n{dCQm!NTo=lwz;og#CRgje%A_Q3IyR*+hVHsSq
z2fxcg*Yq+lux;rhkrm@3kXBHs^vU7BbGyjLNCUU7ISCHYc+vn)97V%+6{akU>0S-s
z67hm*9GW6jXO}YO0pQLxLO3oUzQYvq0WB39P>UI2yw&Ny`Wk4kvp&lWn2k;wJ<cp^
z5Roka!XurUwV$PSv(R&q;**H2H-IBCl@xLzukD7uKFvvC<B0wWks#}MOgg^{O}>?6
z(EjkcPfwdxQ+LD4W~s!8IBi)cKpzT?Tdgb=hZeG|MmPX*uR7oU6-#EA^V#}(2rc;M
z?gtOQ?qV9pXooZ{3x3xpMS#7}=TvbU#-GNgy1D{lJ>Ru<j441zEo23yDhhLLhF>&D
zpuQs*`s@Wq?Bt=?>e!M<*Kw%c0|*kSim%@@MuRD~1pzyw5J{3=t~zs*nVJks<|C;T
zh)O<u3Cg^csx%^mcx_p^9pP`!dg{#mabQ3o^1UhmY0eb-g%opWe(<)tr36S4;3P!z
zJ+;vN#ob9i&H#wcKmWM;DX6;?K(x$M9Zf)@qiH_QZCJk^R!YncwyR}Y=scG3(`lyQ
zeX2@tzgpR+Hy~<+|4)hSDK&5x@HujSScu&?k$y3KZ?>*TRI~hC3|RSFb-YGx*)}uL
z@FthV$vXMT^)3lIK$0@=btqw7E=ClgLg9NxgU_I}8FCGsy~I+F03DJ!a~;d_6xSXw
z-e;;%Z|NR^OT{RV#gFTne%yznZeJqkgeWnr091S_yOi<%i@k%xSr^I%TGuB47faI0
z(FZgTx6M4{cPr1`vdqO!jlv?w?^XLlRSkWW4mPE8Lfcapp{pMhwFSexK|MH$uM5k&
zU8`44b)Qu(a>-)!iq!YupSOd%E^UJlRk~Mzm2m0;6fv*0?cRl0LpqAhpTz$k_O3dv
zs<c}xqJl^$AkruyiW1Tw;h-WQB?{6aBHi6GDj}&TDvb!Jl!Sn^v`Qmg64G6VK6h=L
zx!=ru6OQNHzb^mylT&-|_g(MGXFW@bhfD9{;Mq#&7S*WH!mJZ+8{=WzO9*EMsPa}M
z+2tSTGFS>!ms8{rx>0r>lVSv){K4{UpgWbk$fAvO9vEH$R99W)S`RAnQEuM}5D%g!
zQEnKDhbIC(j<HN?(|X8J!vR3X@pY9Hjc0GQm_1x@K|$Y#289z*@p?_SB;5HP-ozgc
zpMTe8+Zxf|*;{xNFjc2(H%1epCZ#OWzJ{omvFA-31|Z!;`!V@QvQJMq0?U1$AaK-!
z_C1dFCsosVOvRvrMh3+Yl^f7Ucu8GM1KE$hxh4+G6nAkw0^8sTAn*J{z>yE_mJzDM
zMYn3kRmZO$e0jy=YN9gnhtI@9257)ln)ZLv-gIIT3vSN>%#JbBns^pG2-1yIc|0MU
zOzNTM4O6(O8!J>#Fww&d;tYA9iQ8R250T_7bwqtNIFJUYzi?9?hM+)kh&>>3eC<q6
z3l)|CKqpJOx@vxIW_C7ZTC}xYEU1*a;nC3%>M-zjX*$(B95TWv_~+Pg^MmkX&LAOm
zh3t{w9bC&k_TeJL0jm>}lanIDU|44>ES*utlAs~a8yk=JYWE75nVAK|$KNy`bEM@r
zo5<DSTebxgTLGPWltn8w2wgM(@TJc|)3>MHP@<4)N+?fK<(@CNc@2S0d66q(8U)<|
zq^6F73A;4zLyjmkbR9k*Wn3xA^Dp+?$HN!kn>V#u<7vnJJe3~M2}qBC$eNP82rbg!
z19=^LNf{ZLo_V>4S0-`>h|=mrr@dWKuKLyCf{fF~1kG<Q52>=s4`5_(aPbC@feYp)
zrxterT0*tULF+YsUN{vB(RVM7J}%z=JV02P(0_0PrFfr0%OP!8pLfMkGt)o{HQRp;
za$P(;JY|C)NOc=uNOiB3A$=Ux&oa~jzU|gFK>d0rBsC?z#zkr1#9H%K#TLfq$*R*$
z@*ucukCJ^%QUezV-v>>0yE5hZBl0FK6P+s@zIp`j!`eCt!x+N;QOdx1WqSUDKHmp@
z{;k#&)>9s#Qx;>M3{2DW61QI>jVO=2#dGTxg2gD>5bJyWCf@#LeG$<ndP>wCRp?on
z!}T0<?a>!uobKTSWZnhO+jR?)6+)p%75_B;{^m6oW7Lj_XL}Uou9qQ4=|)KU8`5X|
z=33#-Q@b4rg&gMw^E&|eSX@uwN6pV?*W*|jGo4fB1SUV^@ERV$u@y<t4}~$E3JYm5
zTsd3BwT%S(%bgb#79NNS#UU=in;4v)SpF;ZNRtYysYeX-E8Q2BR+q9t{?>+0rmuN+
zN1<q4Y=OUk?yg6><Ww|osRgnN#cNJF;r>8H=Wzp%X?hkKdybeRtc7H9#cM8oN&W!H
z^!SOuA(A6AXZEAIxYO*MR?+iwMNW%jNGjt7G(I@`i#g7{*m?xa4)vCACVdsO=4H)i
z2mJwGSZ+CycC2h5Q0$rg-Ge-XvT;W+(`8%t_vj_)4(43WxI-=A|IkrrE+n@(jnnpL
zbxu_a=R+4_z-Uv#aJQBOM$Y^%OQCQd=WX2K&li11H`nO>pvt0Cm$`-Sj}YShP0ys=
z^uUekrwtAD_4OM<@xu-fc8FBQUhqF5APMaV)lp(%S`K}24nxoM2keXTIEHEq7(-5~
z3Vpdo1d12BC8=g?F>RUn=2yls!Fk=OJdoGBp|y^3N7qGPkhY>``atziM5=|;Y2jsL
zE1*@Nc`k79ZG5~Cl(_eHxj*!dDJWnBE9$h=oTyR>n0wvk*H;^wdB03GYFPjSOgwaC
zV&r-M;`Dx$0^z_(YB?ZDBRYahDE0hqeEd*k!HyWB6_UO5F2zBpSJ>G#1COzvh_Cq!
z&{|hNNSm!tI$<^j3eyI3V@CgwX$gjIIP!H+8%TLPG=NS-NKR}+clY5}Hl9!h_lFSd
zr47zOZ16XdwC#Xo1K$pFh9}TP9Ob>bn(Lu)xy2NG&H*mxr4~Jrz3(w>0@^YU9=70K
zCIufJYDL;tfT18qhc1pMx7TgWbii19?Xn>*BZf$?Z^xb!7ja{E-YC?hvF~bF;XBo7
zhZJbwG#rPsbS=x8L;wr=fn<aO&Vc&^oik!1(YtOej3|~0{A8II2{y~#$`kFcnsJMq
zpClG53wPuY7;x{i8MdwU^(1T4=qVG~k6C6&tEU%zAqN21JJTcX6l3U!ZpDn^EFcP}
zKkQdpS?hHl;at=RH%pN3WlG7PYs{IiMf;x0YReZmdl<t^__ss8KSO+y6P3E=h69K4
zdeOr<;pEg*UnHj?I_skZRCmPw9ctN%E6LBjvG8cu8E^_!#Q;}qAPh3i1t=i|DACdb
z=p}MemP_8V1N>u-rUr*eL7%R!Y0)n(nomjnd>2ZNpg2Ev4Cj6Xn2mvv(fzl^trg>`
zsgsVf0j*GlIwikK4+&b6OhsK(p`uCphs-w9Y|R^}ZhQai0#mQZ+1ABjtSpzG4U!Uk
zQzG^74zw4t=?#uH@%unV%5LJ_FkRZpP}tW9iz9K2yRo2kzy%1}2R_J)@D*D={aQ0a
zQXWW#E9Fm`3J<pWC}ZUc(D?hBfbXG))9$4<T~yD2?Oy0)x-+<coHHN8>hm{ca}a5m
z$0JBzo63vx%Wc0E7XRotCP|V1JhZd~n3tnTrH~OkL`bN+>URbl8-$wyLnrv#eK!Iu
zCFY`W8g!X1QN9D2t9wU^4HrW^+6FNtjcV1A1-4NS5))6IhxWMtJk^iiee_-2o4Z*X
zZ<rJ;hPYlx0a6C*DP;UM=D*EKi->&YA%LV^jd}ol+-u>Oc2oW$0@9Mm9wwJVc^rl9
zRW_12_UPmNn@~&irZ!*Kn|_hq<CKU1T7P_r|Kk-w_7@VQFBTz_?hjQ9B96upPYm;w
zzbwg~qY5HH6&=_(e6aT?6k&m-HEgX}FiWQXdUN1#mfp_c!OpK?{pJ*eqg$w+WQv0q
zeo{T_BskrxC|4=yk}6xC9mqrZJ82>L7-%0mIqHaqA!72IEIbN@@FXtlsAENjQS)oC
z&Ak;+CMko4bc7e?b8vV}Mh2TcF{Bc`BkX&)piT_aXoeBe5?x3))H+7&!G<_SUq(vW
zv5(y+(kchy!`zx89`!9beDM%F=j>b={XB}_+>L3?+|Am!ZjqO_c@~;PNf8H`#&zKU
zm+R`n*~M`T%X`0ge3y5>mK-a)(3lVHjSawOp?d>(m=6F|>tYV7XKimp_>l-1XoNrj
z|CSY$`jB7A7=X&%7~KGBH1q_9wYqO2N{&>K@Hy5XEqjH=y4veTZFs@JP#8Lu;enl<
zUHPf(YJZ?+VeQ$Nx--zCZq8Pfvj@Y*@Nd_RJBeI3r53^|KvT*40GNQ$Z81?>Qbon`
z=X~Iq>MxiUIZY(+rC=74-sH#dVzF_;Jq{u@YTQFQ8ze}VDR3j$mjf1$!#Cuq&W6Lo
zfxM~QV|HDZa=`jjbPi4h+6dDm)|R+lFCzf$_mhkZbdoGE={rol8<t^x&Hs~XUe@ti
zp>cG%=o$@4YT>HUhw@_#STH~goZ(0;kubGmwQ%s^r>!zElFIz@x`e*~y(w0v<vd9N
zWP3OJ_VQz8v?A+6IM(NX)^=|L(mw*r9!Kqf+S-&wUzsHb$e{q=&fao>U8$yX?eoKi
zh20+qmy>uU=w0XrK&#5fDq6hWY`)GPpk|MpkNc4Mh{^FdjE!-7cx2*8RL-DJ)nB+a
z706w?>=Hlrki*>AI51dFh?mY~hM1W?aR(MW{oBQ!kU)EL#EIHs5}*-Cnc^qtJ}?#M
zC@;zcz;1v*|E}5AqRr&x_ul_8tQaik{$L1X785&9;sD6r^0OknA7I4w4S8o?r|DAP
z0HkHvqvJ}On?mn~9X>u9wCxsUMjHJg_tCuuAT|KJS6J6=Farw0XZe$MB-ET!@^>Lk
zj=)0JFk8IKUJ~Xi^t2jcCpUuB&5kC0Mirt`XR8<kO=qeW7j4r}J=5N({NZR|;k$DK
zczyz)SVuU8T5N?G!u0N^fg~$!xjEI1*1l(@bL8WL)V?}*H<nrv)Xy*=m5txQg%&Hm
z?`6biFJ?=JEPW0fw>=_be4W)(aupU9_Ln_(>7qA!0g4ttYoq;){nDz-urO+@y-r|Y
z0pVx}Fizc2t=sJPb+55e1(Y4}Kq<Yb=u)@+w4W)aJ#F6{b1Sl!?3K}<1cyMTP5|2i
zU-WUHcsatJ^QFsSG%jHP%8uUf)=7Yr*6T>s2C=`X;qN`ko<HLYyjk+)fo?XokdPMD
z%oqV|0W>wyo;r2PaGV?~+Sm$d)xOP^ZT0nI&O#&G9GxW~QY_o`&}1A%>l!Wt)Y|?K
zxu2Of_KMYk+6BP;CHr9qnF7wt&16cPS)0)A2j0QPOxu0fIDg{rAb3<Lew+t{K0eTH
zGDwpKNkD|}8d8&$x!PjX0~P+C9A%GCUy2eHK~RR4%~w^<+1GkLI`wQAbPoay`Ni?W
z#Ao){S{89OAbTTM(cZWd9y^)-Np5zg6jiSGMDwzh59R7!^ecJSqlM@VwpEJCkB<{`
zgZbChrQhLT!*(d7fbAeN_zDMm7CFu>4ajnXLn4fvu-cB4M<;@r2%&l@C4EsvW46G+
z%uMdM3Rc|ZkBIk;liUF=;Z54nYHl&|R&BI47gV2<CZ*Hn=A*>8Mc4a?m+(y+o|jur
zz7-o+8p2A*_9zhy%5JrH@2t)11RB#S>BxIGrs5PeH48t@MoeoQ1YqzqcV*PZ1chsD
z1eW%gwbyTl9#bLk^X34la?^$UCOLCoPzPtPT)i5X%<-fIm}1JApSZ-z=Kwr(SDHm%
ziKre{qIGdWM9^<v&0Cx5Juf3;-0=YTIzQ=7w51u2e@vdqY?AXq=n4t*z!wKGk4J#B
zm<aSQF?F;C;&m5)6agH8p=7)Bbis{YSCpK(O29o{US24MNG{@Utd5&cU?~oN6pL)?
zwYLB_e?j{-f4(p%D?gk7nhSD7`+BYG3L0RvL_#%IQqipZ5b^JOa|3D2qtt^~!Ml{m
zSzFri9mPc6bWcM)XS_WX^yAI~8QWr{Fcx&gPMoi-TeXZ+nm>C2ZdMo!-#$(!9j}+b
z(esR3e}QsgwlWH9D;c)6l4&;hRDj{UjpL^gegs;>d-T76m3@w-Opk&YqIa&BLloMm
z|EQwElFPIsQ&f*}^d=Yu*ZBFtVEIKs?2MKAD2wQsq&5gX*8EOGUe&c$vO@}L<x$5^
zY33$j`GHt29q!eCl%214^O(W4^qYoJD=8Z7{AbTcN&hjARgA9dDY!i<*t4<=&HRjJ
zu>Ni})7hWzUhXF&(~&y5_ei2zvq|!%x^g@F*p<XQ8N=6_Nu_ZHYMrE$&&LmEN{6f2
zCe0s_rpdZ`l<CrGsSjswZO+R%8){ou1y>4BpQ$?&^)3bVF&|~!Tep!XR`ibUgG}KM
zoohju+^ukr1*ilbPo|e2>K2aftKp}39AQ7&=$Ve{!HR;+hfz*1f2#B(WFt0M{<3!B
zk;2<kJ&wZ>NbvM&Nj=xbu-)1GhwlClx=g$W*rBh)=-S%8)yvCY8G4?M>K?6upqC_3
zIhkF!5ez$wUwvg#P5s^0)y8hj<yS6%zcGwK$ADe`>PXApT~vqIU1Ima?2ALdPOYEl
z4>?0<KA}slAjHmE;(Z>Bt3T7K3L>uTW>08<@ZV_IwWA+Jyj{4pJfo9b@og-<hY3Cp
z?~{yl9OlPU1HMRXk$&upU{pcmOQbjbF`0o$;j`X4mCK(CVC92Oc}R;4%b;erW`wnq
zUTf00%<LvM4WI9MjeOp4cgO0bUTe|hcw1QWsz~uT(ec^dk(;U6Cyqi-u~f0Xh#L|7
zP<_GfGsAiI?h#0eDQ?(>Su`-UOvL7n_!5k-xe_B8W^($q$rG5x!vc7l=ddn%1CH+2
z<>X?%5U^wmdEWL^X4->@0z)v)=8m8bS*-Wrhsa51!cz~&Q3GNRUXPrAX5#xCKBU04
zvu@?SV(tNWxYNhs;ePqJ=82j2?tIHKIhq|rVNm{x?#H2bsa0{=#gC9(c(9Ytgil{~
z+j^%7ZdW*>4M|0foc3tcpA`+9KJ7+5EElamb)r`1J?XKjW}1qb))H;k#kUR%4V9Zh
z^@N4$@2vEoVP?fN@@AQLpJ@O!h9nx_tcW8_>r}WWM@TPv#;#3gEj<D;*y$)(A1J4n
z1E%To5SJ%0ealeJ+3yzLzY|#sL}bT4XBq~7vdbL`9S3$hUOin$S42IR3l4|(nytj$
z-lzSCGX~r7?_b%PviSOQfh#IPEvpG0g$!U<chb|&6siI_9>DOq2MSOf#9?h#pH9s=
z@bF@#W{Cok%a?C>*9+yT<&j?Zf^=CqVN+$K&BElQyg9HI>UL!A>PHp1$91Rg-9OTM
zMqfIwRTC>F-4Ri%aqNqeo%~Df`ZwCm78hmu2ZPZCCMP+}H)p(5ZhU-1$z|R%Ep*Zm
zsiH2nuo1_PSa|sU`?dY=1Omeym~HU7?GbdG#Wb6Ndw)LMZIxfmU*8ch<n9LsDkkXy
zMV89cJ^@p-e6no-K2|O|zID+UeA1=+AA*l%b<6@FJYc>y?@hBq%bp&{mjS$P0BAPF
zNMS8n)U=A+TmknsY8;`D^CKel$H%V91t#?q_9Jjr1M5m?^gw$9hKx};S*>|!F@;70
z*m0^u7n9pu=F?`or`^OtLmyt{c3sUpEMRdx{lhH`!N6O(-D5ui7&<HKioo7eOTF&B
zeD)i4=rjCw9fmFdN@g_GN?rPrQMbunIQKZUVx&R65=jAFD_oZ!zfuA?w*t9&S*%nS
zRLM!|>vu=2RPYzJOr^{ZZLn7aMjZzK&gu-b4o%jMClS<%iHY5+E99~3b^XbYHqTna
z;(d9V!>syvez(4uvjq^}=|r@u{*Ctd{<qRuVcs4=E<z9p?6&A4JA18+crNx_13M_!
z<v@Rh&+NuDnw5%*sv{z5=_epufw>{vNZWw_Gj1^o#tSBg=-1i;EUUSWRY(wvgfC@r
z8PpfUpp^`z)a;-iQb&J~otv`JEXKRZ<09KLifHT3%$pz(a$33-kG*cxajn40d04m7
zj1-Oy2^5H3=Tn6%0PsL6;kkd*!W9x@`HL;;2#-Zk{-JXE&6!F-#ZMVK7HrtLdW|<}
ztgp7p`}*^Bu-05QMhx2@uKllH5oZClW}~RSx;Rb&<SQ9%dQXac?-BG~GbRToCjfRi
z{p(_Ou=C;=<()l*_XAyCEUaGqlnP<l^xo6?z@*pE*0zu>Beb|SnMF;F)It)b`i8Kf
z0irt)4RC6cf6X)>AgCLQV!l443FnXBK2p;f%IlwY^Fw2ZO6cNajf{p1)D_`^Dw4+;
zN-zXrf7|{MIUJF^CY?Elkl6uyU)aTn%EnJdbV4?Fo(iLe4s+kKY0ueWJI8eOqlLFd
zbyEZi^uJji;;-|j6FIGL>Hfhzgj9#2uOj@guw58g;wh|@Roam`=?~~4oW@iJI*e2O
zcJ5l$fARdL-hR6E2RWnh#sdTNz0B=68CVf2A`q&ZFAew7E_}#AZ{|S*j-=kp*Gn?(
z!1dlgTS@QdLy9tUf5?UvIiW@t@fs=k$9WFRea^b61xr=e^xRJ+NCtX?%@Um|K-B_o
zcSe!c3@tWH!$>>`=IORJ00csrQoCX_0qwk>iicB)-0a=gLzcicg^b?0-&5eM|C+D-
zU!TZ7Q9&NH2SEj^rqj<A#?9PO%_uwsSx}R3m$})r9;Xe_L0Yk{IaxkRO0nhl6Hoc4
z)1HdDc6jAIYsSX%GUy{u+w<mIvp`0F+>@d5v`V+*&<Uj3d>>Hw8bHY4$mPD-WLb8=
z>pT##T}5S6Vq=}Fe!g8~8g-AI>7`C7A-^901R3N6zrFzE`}hEWR=trX`W{#N^FKF#
zNLZNcJjNm$N)okP<`>dhL2VxHlJ*Q6j~^Wd!|SS4btV0&D1nSa<CAgueezRB2%+WO
z86ZUZXXdLj#$7_;G*x<12LR~G-WdO8n|<#quqx5Eg@IT>FGUN3Cs1W+zXnPgI+zR{
z9|xpCW*lLipEtJrG+x6m(~Dkl1zYELm@8k<Pp7nKUFi8&6`lQpn>!VRu&^0dU}%s;
z^Gjj(*h1H}imC^U>RJ&+VYFCL7M8e9hlXFuNn>itz8##4$6(f2MA%hqZ2oO=56A$J
zy=)J_&dt3b1sz;lbK(a=p8sB)XmL20`#zgUVh1hL2w9bj!y!)1cZfyFpo2`$5{3a(
z!VHefy~hX$fVU53SfvEQc|S(*^`*z3D5-xfWj3MWvI*Vo{_6NVe0&uTB<V=OS(;dD
zZP9kW=&}&24ph#80Z5hSLSWBHzV@3}Ofnlhxo(urTyx)8?2QVGGlxM^2fS$J?mgum
z9vX6ztk_KvP6}{^g!C)V)vzL(<wyd^&uWrCckFfIrxHqOwz2UiNKe?sPl?IG3!_cK
z0H?gs_9k~o+c-AM&S0Vly&$kqUzf#OV5AE^5>k+SJAwMhmRbRv@>EI*Q>4&lf7aw}
zbI8@i(hJy-(P%0hk;F6~k<F8w#x2<`&!7z3JlkPR0A@DI@16CuAJR+@Qxm6b3+3Y|
z9^p_GF2r$!bT2J+PT#j2unPpn_z-_*iT{?bfB(uHJGkc3v@Tyfy!wHK;B})}bNkTC
z*v;dsOOr}j?RnEF-Vm$5_&52ozwxkw?6(pQU#>!vV_d}RY*}c~TrU-)=`t#X*&2J1
zo&uvTY{)4|Rie187D1hQH?S5;^gmo%7^Q`2M5fSf|Mak+)xcp<7l-}S$`le_W+c=m
zZ_dJ$qL*c;u?Zh&U)J`ES|9wFZW_XF>V5fQDRhHf;hVy+U;5iKRV5+;s*TzEpLfr$
zjuVhhW}<qWE1)QAXi~WH^xS-&Q{(b%-rKtu^H;}=_2M)f`j~-+zw*4SKYiNi!`R4=
zGCaW%u~xOMz3&E-{SIC|V>>1~o2Xq=BP$>=svzRD;4SaI8k}6KvhT(6Vyb%+wl*Vn
z>$xXC9E;K#XCF^)<$w~EuR0EKi;$o$X~#YT*2fOrw$`VWUs+*A1qgwZjdrydHj~Pq
zdDpU{A8u&`w2Ds2^XURQ>&xBGwTlXqrNcl08JL)u@S6SmQ}ZWxIfz(Wav_++H!akq
zcZaYgki&gNTPQsEt1l>(w8<=8^~cMw=>yPQgXL?2URb1^vKoBgC5hhL5DpJs^o2;Z
z{+17P&zX17@M8tDF~<aoVqdqBfq{Yjlh9Bt2*$OIP)=IxChxvVcAi9-&&;g582+_I
z`4eBeDj38?ZmI<+*81h$of5h=5jPPa@g~Ui_EVnDskt2G=A;!ium>whfk?0zHdfgo
zr_@{eh=6NyYz8iqMN!Y`zDa=iZ22@25m-_CvLKhH=~v7wpt3?_o>eDGDwH4j#CKy2
zWuEbw*O+SH-EgG&eCSzh_!>b-Za1Rql~zH=F}l+?EnrAfj$BH#P_}lKiDMr1DXn$o
zyjBg&lM$fV97znsL>P#hhodVXzOqZ~GD<uZtCFfUsa4l>59+m<>Sn;fg}tjmT=hY7
ztTWpjBDgm-nqy-|oiud=UpmcOv(=Y?aj&yr2s`=2J6me&RRM!%M@P)nwgMsb>M)c^
z9yKoyPo{ItZmew6Uc@4fTn0lJmS>Z2+f$O>HNS>CWDijF^-S?n{E4RdlN0mz+bB-A
zRTGmb^?XZPuK*Fpy>4N5Ur$fZr()=PcKl;a3G&$79EE%AFv4V+&)#wJ?QYY6Vbd9)
zZ`K2U@vS-x2r|X;rD~pX71U^ME?2665OjWld1U5Z%&jd%lQX&4$)s;3wd-R{vuUYg
zFPL^2C8%h-CO2Q}*#i`PWY)ty4La>OR6>K653u03STP{pm~4kTdNkS236?LpVvcV{
z8*O_wNozyPORu%1#m=GoJXQqH77@I$`eUM8vCVSRg?UGZ>fB?R9kliUDGy2SN^P9+
zsq7Rv@lv%1m;PS7zU#R?E+L8W%g*ugQHOLZOj?YxKe4YFCZ^@@hvBnHtoJV2Ebuza
zNZzQAK1(ZWxzz%Vsp#L-0*m}u?Q@2Hoxs;e!D^FsD_>R0z^FhO#P!QqVyZ7a+F`Qq
zAQbo_7Gqxqe%EvSgMCZ73WIvpd`|H9&6JAKL&x9i2~UDU`E=PmLfIxK$IXFF5Zw;}
zhTN5*N5_K?)^Ph^<El4R5#J*Z&T*8$fS;23q=561dp6alv0}E=b*`mu)NKQUYauVo
zUMU?r7OH>nAD-d6^h;Gbq8sgp0ma3W)jp)Fm)B$U$#izoT>CO#NI5@8;qqnRd>V9s
z>4l_^`K<;kkX|B|P-ARfJOONsCDhikIu_ZqSF&>k0`g#M^`ftIYB|gLsAxIzT1o-;
znDm*}b67bsdT?Si>mzm%Jv|W2GO3w04=HUf>!IQPatiGe)YD`9fc!FpVbmW0>(X0U
za?d<TNzjYtv~|Q#*d=WEnCT}#JNOD&Z?5P=+5Br*YbyspyY$PTw?fk`hM))hOk@%>
zJ(MWaA;C9c;Fc$+Jy1E_K{Ljw7t!(*n>WF2))ip6+!tm3b`3aar7t}@%~%M7!7E^d
zy}IV~gvVvhQh)xQu7lN_us}fbO`&8eXx!3zIXMGEUi@zhc~dc(#rHgUTzp3D>IDt^
z?(l25nK8`|+~<PX27q$T7b;C`L?R0rib_q6$!c$nu0q#EYUmwGH#!(z?R)Lbmqq2p
zq))|>&Qo+s336Ibn1{9?upayI7E;6s{cBLylswjQwuT#AQI3|JpFA<Ef1+;X<5GDy
zs3nEtfI5(t$|Z)Alzv>{)v(JbcUa@}s4G%J6wktyM$7wXk+XiVF2wOR13nxNKfKIE
z{T*oH=YBkrl;BFfye+O?x5AUXsDGQss09vO^gMz>=xrl{H%Fv*MvdCX6&8SVTfc(y
z*r~RtRl`VW_=)je9L<voK1DH&<&ss?OQoKKzN#5=zNB&jaWxVr!Jz}JwxT5jc4eDu
z)6{!uI2l3PQr+20Y-YWgyPsV({RQ(5=m-ge){o%|@RM5g9KEddYgW~~S~b7H7urc*
z4i=HIZV+7Gs`&?rcu23$W-sVG7<O55N<`XZu8g7P6BUugf-2OB#c+-dU+tZQ&rH$)
zjlr5D9v3+uIlCKu*2)D3H#eequUHeu#{9Pmj}5Vh^adrMM7jrt-yes^bi;`&s$YEH
zTbCdhSLf9DQWX*!9<>xNKTi*f3=0>j*dZ68f-kuhF7$R17+h@HX*|cQ;3Kv%G+suT
z2pyYfX6O}9>~D~yH_uLAU1hc`Cr5YAjdka-uv}_mLxKjz3;P3gO2?k3q-3pJZJpaF
z_SmI9rKGJ*Pe4(4z&wBU)~8eg3U(OB`b2xP2+AIN0fcY@P2o!<G*XQ4ZszG*VkK`j
zs0bRG!af-^)@&)g8$Yx{)t&RePgE(Qu<x?8hs;zXB#Np1{r%I7t}dbj-8)VB`Phrr
z*GwPlVWr7pP%T<NC2^XieELo4641M#6(2MKZ*aOy38j&-SA{>n7U{0NL`hYGU08wO
z-jsspy^squklOzEq?_!F%}MH%yu1ZSLeje-mR~*l-Yu*kJi?Je(#2S(rHS@~R-Cu0
zZ&AKtRizYhItR#nii$EH=VR*E4`IdN#el)v&x`P#Leo11z?fSoU%%eC?75$khQHTE
zlg9S*fk}BQWWX}Uv7*f`ur!p${ZF}K2V%UYeCk|=qW!KaYt=4wno|d!vC^)7nx?(7
zdm2*@39<Cz{A7^eejiVCn3DnK54-n=TNhFmIO(OH2l{F50Vsfzhnb7A+;Nfylo#d?
z_n4xpr*|`TozmL5Qb|tm_*4;k3K{|pp9z$<rKuy$KZjH_EL9h#d&Nk3k0Z1<7_a6P
z36MqJM`C4kKr%F&^EE^eQoN`NkG-3&3nQ~f%tEY2YlQ=Sq0Q(-q}%EP=stONeEJ4<
z%3-M>G~o+oz;bn8&^jdms!^;krp+43h|4AC%SZJ9A=nET7D^pfdSuF1-gT^a$ikcb
zqSwf0rCdQQJUhun7KLDwP*v3fZEkCm{gq;y(;-%oj<>XcKwr~xZPLCRsAQ=>JwYG0
zeOBPyT82=kf<N3n*P$EO9qW$Xd^N3)4I&#7sL$h2S#{MQVN_fzu)qt#Tsm490(|Y{
zwHYVCPjn*KlbWK!<~ifhHQ?z!%z0gWKW*;WdE*gnw~cuewP~i#L{?|z?$d+I`e#Qm
zmI|<~&|yI()S7K%Ki!>YKLNhm)y_KX<o7S)i1^xE)5lJg2k!--0)sBpz?cdw9QbGj
zqy*J$=7pV`YXtp!G4>W#(&G#Ytf2Sq8D*DlIb#@$zagYOUMOaGVnd0Lus(vL{L|Y`
z9c-i=cX;r?E_`!&Fn&xRz>)?H7t=BM1{(qbJ{RnQ;pZf*Kv;ki7=jjsNxN%UkzNLW
z*oAK{GZu`|J%DH^ytETOiG@7?1oZ{j1u9;wK)is+3~IZqSMq3)VTYJ+>qSPWu=|Jd
zV*X2r%Mb$xi|lnbkqj300C!ux{owv?tG6FI8QbdZ$Nczh_4ZwfQP7XS*Vx<M+m9KY
zZS}UT-u`w4w!OE%U7o)>b=%(CA2sy%>Gl6t!6~odAtW8YC14|3II=aBs+t~*|7jyd
zY96{v0r6YRSq6t@!cn@gsvc9Pz6kaftc`DxIlJ~;_Ng0oH5<@Jik#grZdz=YL#=((
zrV!Gl-{Dl7pg3FiOE3rZ50{pvm5QyLzsf%N*l+6}PejqP=cCu4f;_S*V)xTGf0^My
zv9r&uXF&Q42o&h&tTP_|EDQJWmf(?SH7Z;?FBqw`0OW4iMN2rr48Ft!J*4tOH<5Jk
zeU_Q`bx!M*;pf-zEwrFkk#SpWm(B2$5H-BHNt1$xvV`>(H}@3d7u0-r!V@`!cN$7N
zI|<{SaNyuR#9O#@`UA}my6%5Co8k|k_k!%*S}leH1%DB2CJ1YTygx8u<>sQRAX9O_
z%T*u4LDIkYMF$bca@kF+lz|x(ZbGKw{8V8EHcAOQaeN@GWx8=1D<ydnR-{0F)V{YJ
zJNcq>unPydzDfHZr04hU=JzNwG6XO1Aa8mC<6qjUARL8V@OgR(EA;`~`zIk^RMl;7
zhZX5%cmTU_?h*?&dOLUGoIhlQ$3Ofl7Xajr2B(CDw{+xxdHvtB4ZrzJ&_&phqb$K#
zNbrK1ItrSS)e04X70NV{i-7z*av)d=JHoQ9>i!TP{zeM^#xdDeb>Fv6e_X<CRrh1Z
z_WYr3_4WfRx9!RO@c@3`vE5d0-?vVGT-|N;_5+6$9OiBH_Cs##wtD;H0sOvW`~N__
zp`!|%Jsu_|k>YMHCWi9h5AHs)lMMIHm$SILjy@ltwsbur%<x>G_Ws@qy$d2jl{!z}
zSLOuJ*=y47eKhO0_rQ6t^JU~@XX$>%!?ip_QYBTwIEXK`_u%M4?CWgva8D-9t0=qS
zmSpF(&9QKyX4fKj*IxA61_ysO+o;$tD!}6$4i&`j2e&dpKc*Ib^Q}j!;&(hxtMv9@
zD}LX;w;u>@xA*ji^z@GZPwN()R~?pZ|4X{4cpgWy&Ry_p>V?W4)v&p|f~~eJwc|k*
zwxYQelC4g}$Oo&UZjZZc8Xbihh<GoU>$&(8sWvLQ>k_UWm<NT<pt-Yk#YUv&0;Ez#
zkG>h#7ioy1phI*6rz@^qL9M)=zUcU~`=KcKZ1f_I=(u4)8acdMF9+YV8&eMh2P2u{
z;<%8usq1THhM-+L38}0UaF73DD<%(lyLrES5p-@rqD*1m{L?LvYa=;)UhH&8i$(6M
zXIt%0mr1<c4Eg=nRz|4kHOEDe!^5DBW7u7nS38`Tk`Bq~;v8``)O>^f{ANsD&TAET
z7Bcndp1K`|Jf>_}_a{Jc!TmO_BAEN8;qW&w!6*KSxLT`Y)Z(So`4TfclBc3LDf3MU
zA$CY^+X+WesV35CSMV7=fkV^oZo*PLeY?qKvQ>7U41I(hC1|pn$DpdE)95#9_LnvJ
zfmg(L!mqJjZ+L-~widjbY|&>WvU-%UGKT|qA1nOYPG1(iU&Qj?=V*VsFbwJ-LFtQG
zR6nK+{_rgYarm`;*LaR&Cu`om&!9diIlK7>wEG`!416mO9QZYb;^$J&Fk7)-a%@$~
zV8l%dlHnx(_qG1@D`X1rYdD8wKmLHU{`d98v%p=sIc~~}6%mQP4ZlYA#^5qm=mT-n
z$?Kz!sGBEZCvU!+tR5x<oDV+zub2F{r{LE;3`)Mehp^pA9V<<uB&^Z!YuhUB*Twz*
z_-0$h{g^awtGMr%vESL0Z58)t+VDFTb6dszpuyf&aoZ~HAI{vi7x#nEhixzJcgF4e
z{jcpn^#=o{?LhUvgV67JWn0B<tGIu%g8!d#aUN49PcnW@#W4ip#Og|2Rl1lIQbH4t
z2#QAw-V4i!z|H)bt-sdsU)YLG3q=)md_<UU25_c^E18k$^4C}KbE2Z$c;9X%p^v;7
zuiGJ-EhW)R1I}ud*X^jCbBcRaVcnyz;uvZUt3Qr%M3S)AQMiq{wQEWGsIf2I@JWNm
zkkMUH{;IEqvvLq}Hrk6pzw~0)^LfN%C7b9V+S1^yIFPuEyCgoYk%v26!q2JAuIoYa
zSXDnU7<rGni4qtZBi9mvgz%(g_6hL&=1wY1XVCf2&1-PIgEY`}E1boY`QWg#Jcv&?
zs0vAC$CQ9BFk)}k;7fVt279(%XE2J)H{SL2%rF~2&LsJW$@H8uc96q(gW8aOTW`JS
z370>2z5KppNmbQFx%3Bne#}Yz&9_wd+33G}`WBu=fybkvp}_)XBHROhva`SlBX+kv
z14RU6QjQ~&$z-N%I-f#L={h@bnsnuGAyX=1irnXyT5D@-IrVENIPIsxCsXH#>xd<3
zW*`1E#dShjbxAdcRG|cx`UkZLB2qsSKvr}HM>-0TuKk3B_kgT|3&t8h?n|pw7Rjy}
zs*(3fNl6(ltedA$wDxhH-auKjYRZ&0#-Tj*K81LVTze>KZf-6x^fa~vsQiVO&aj_U
z%jEpf*>O-Q#5HSo>vN3mZzoP2zCryA%{>lyN|KUz*^R!jM0YB;^#L?B5Tc&(c#)XM
zZDT*2cUB@<nIOB%N*xA-PJ5BFDa*A<Jlu8X&Yivh;i-<z{e4lTiCVTA%k^HOBrwAC
z4U>5tQ~b%bYTTxy7~WtSq({pR;)96q$AKLrMlYF*5%G6fC>^GxG|c%b9hcpAjc$)w
z-ekZ`jk%lfnWeVZSDGlzTV_70&{4zG&^vu^N)IInp>G~z`y_g7{AlF&SkAxxOtdZB
zPiAdPHOR8Y^_wrK%nBB>EWSQg@mia5W>Cy3C|EJ6Cz<@(?~7($ACXky?XFBx<t{zD
zw}6wzW#+673=ATq<~u_PBpSj?8l)}B>hwl$)tXSfj>FwLqm9lblA?1@`uw;z&)C@5
z@c7F+zd5;fsk6of!&S`Vh(237MIGt^B8k6%LU|ddykpmJU6e0Xk~a;%vh-NG?nQqF
zp4;4qgpzv{ZZSH+J|LDek*Iv`d{hBi5auU}5-O=>+%nCXNT<I$p0cL*QaO=u;B|Ay
zHEwTnbM5K%c`@?m@x-cUKBm{5JedsSS-&}_|N8DZ9X$dX8uTTPq4}oa_=%_CUXYZ8
zQ&B{1Ty0^xMRA<D^K3>0t;3g{eQUkw)>}uO1te?cv4=ZPRhCrvhmEB;m*A)Fr{5Tn
zEFsXcxZXYMw6_%&dvi3ZkP`N6$*EMBR(LRo;+`-2s&-6js(HeIp*x(Z_{m$Mpb!7{
z&@0$@kjlGA<ZLYU@^x1*26oXA39+se`5OqSdhDIfn~s@De^L-PN+c(`a;gvJf0$9}
z{>`wP=rnQtfg_A?P1Zv3&pBVxKe>_gto0Ck%DePWm(=j1oj%r39`%PYnO)H)9iQn%
zH)rKrO>666F0Cl)9hwd}#_Qx?3rb|Z+gMU35k;>MlpIgUXS7RYQ%>6Njz*zLE?jt>
zmZn6=`bjilY{2y&9`2v=Qtfz9HuBH73MNS5x`hR27tjax!R}Mbi7dMIJ(MXVZAnsn
zBE6@e;PI8A_HL;xvyFizPtrt{)MuPmVVL`JwSFyes{4lVF+XMzlA2i``EUxDzV6qS
z-Q_skI0JK{8Trgk?9^)?zG9N^g^?7AGpPTs%~xaO<m4oATkFg#mF4fIn#vhSd1Km{
z)pqX?hbT|}N{u<*TeVXfF0<LX1F^TIndu3sMAzEU2Y~fnm46roRP!^rtp)K@Z8wP{
zU6;H{=DxC(n!8!+H)U&C*VL65Ygzv!u(_-oRv~K#CpJ9E*WsUp=<nCrV*)9;_H^Bc
zIyb9DuPgCgwv-$zXKT-xn3%R_dAUNBp%nRiK*8x;azHYF@ZB;@FGAmy4O@*`D{P=B
zD3~;_C*hUw4psc@j2Mh?+m>gGp7Aad!BOjBGn*)P0T|nyAv?U>(Yh%}V_nVsiZ>0!
zBkVa9hU4(6{KfUnPCvkd!ynHu^ElPzRF~8E<dGD?QXulq;@8XRzb{j&FKWFhS3~(+
z4_MTv&n2RkmRRl^U#$m*&r<%v*<{Zq_k~z}ugtKEFsOCiV(J~eS<cXMF{ykKJ`7!X
zJkUGUVywFKPW@sJh85(XS5QKFW~J`8!y|L|^P!zJbl?GW_Go;0omfa;D?A<krq_*N
z%+uc1wppZb{JnHs`gqKxB6L4R(ZM}?vYIum`${%Se2zsb*qhiB7j8_N(;DAMQ<>oD
zUF#tScOyB?(h}rYKm3j*hQhFpoVh5qs=GTpj^mxYrcl&Z0wWH{YQ@mOeUf6UW)8h;
zas=G^Q|&fODJeIfl-ltbz_gsWK2%XrYM&?!a-ogM;BFVCl54ykotGz^KkJ`r%(=SW
z2M6HZcuL;p$}Xkl*C4hy{@H)4`2JpI5rv<r7|mPsp|gTeHFSrE75+($9P2xtjDaqO
z%a!iy8sI&c-wT#^lNlKumBMXoXn4SylAZmKHHG}_6D=plQp4Ew-qjX<ej0h*FE4_7
z9vOrtrsT5t1q86%Po}sM;@Z-gr;DNbT0sh=MQ<MfO&C6|*38elBtR!!OCKDW@i$+u
zvuHT4O_=JoID5pf^xiD<!wQuXlZN#cbs{KFu<zH;bmOOb(YsxH%B{b2cfgQNpx^t+
z-Ce|8Vs;;!6I<^JPis(7t8hf0TuaKful#!(@{db7cM6aQh2xCEIJh@(wbnXRyv9?}
zQp;m1MPva&V^vQ_DoUlOc@~zI_#$82V+0z+*!YYni;`K{$|kBaqc8ZGL&sXw&1J~4
zjTPXqj2F5>0D*}~4sr+d!JIZD<v|7##vr|>yV7w7gJoAsH#Ro<fR58tXcipQPMb5y
z@A->X2z}yn$5WgSmQNmrZJWx`*TxXDKqd(>ME_^$cn~p#=-61shKa}0^3QX4?!X+V
zC56OhD!KOVZdxM2!7rWzhD3<YtA55UZ1-_-od0emy*4>@wyrp=!$71f^qOL_4hl4|
z4)>MVo*e{$kcV{KL&rn>-8RNeyu?By_l_7f1~u!>_EQ{F8?!H3L!Gl6I183+*{uLW
zW-Kxpc>WvBGFkBa7q_0jx}cz)&!8=TWpoA{prZJ9v`!<wp#cdZ4sAMKETMBSK3$%S
zV)tXcD4(;q7+!r)4zAwy_}ouO02!T-aI9$o`2UDwloNS1xySbq`cf3-7>W+uNmLf$
z+SHo>?Mz8<1hZupXUGOmdoYXVAfq^gj7LB~z`0LXPdq+nX}DWt`HMuF60Jq?^R}lv
zyAb=`U=wUuuMpzO@-r!hAkrWi*j+2tLsADAf>b;rBO@;<InTV;OB=6=664@wskwcR
z07Z~-tHv+FsymlZ$H75>k9{9lPEs8_4BIHH;=mZaJL&rZ9IHfeb93gV&vZ^BeQk?F
zHJ`H1N<6Cf)pjv3(E=|eNi~D-|K_Up_V%}4ov$IxGHLIyDL`Ulh>PR7k`HU-SQMRW
zN>O_&#Fuh%&F>z`KdP_4*M339$TDlveSl@=N0zzzvQxj$^3WdZp=u(<V9zA_`5|WG
zeQ8F`q-n1}gwsfH{{9F5@qVmM1#Orw4uwisfdMgK57N8iX*=G+FMrF#W^>&!jVl@N
zsj$7-bHiXBZDcz=@G;#^1nlzelz$m)XRNrqy!|c1WKHVcXdSR6)hDB9DME$qYXf@u
z)}2Z{k5LEs6F<_68dzJzaQ2nLB}6%?@~e_E)U$X%2v@DFm2IXuq2}t!eM4960H=9R
z*SRL5{U*Nt5Dx1e!Nt(r52}G;(0C#Vi#jM99m|;aX{h7vxNdfDYbynH)BW4;tcRIG
z?K-oT_Yi_3bd>WfhSvKX`ADtlEl>%!#Xs!kGjTYr9Gz-@!Q>Jf&QNVcm_dhr6s`BS
zC+j#=&z)tZ(y3gBDv{6~ZJvv`(V6aRcBK14%P>8qFpY=^%f5560Vha_g-^3^3N6tv
zH7HyU^km%AAExuTl{L5LW~)Z>wc1kMQDxu9R=3)7D-AEvMsPR08{krm)}2t8E-MZ2
z!_TQSbfbcQCP=%fC5Hnm&+UvFxWq##)=#IeaiiaOp<UZ%%Ha~3MId*>&2?|*a7&-n
zR5!ZR>4UWXQ~t5@mAcPwJBRzQ1OVbwv3NL?RJI~;d0tQBXhiGC{QPMbzP@QQ>L?C3
zWfK$94dDkPy}k~gDOgM2#%|UQ1g@&4r9Frur|w)v8Ty;9cbn=`qXghTYnc^%Q<hYn
z=NRGNHMlvG0qgK@FKzt+KiRqUpWX!CET0Qq13XK!_SycXewQv3o43ebkU?1WPLnwd
zJv)__xVX4jUXgi}!Dv6&6X(08ouTd{2l?r~Dns_^Inij%D1F8YebxA9;rxq<HF1TX
zmK8+@4hk2oe$9vwvD>J1%t-W}4&EnV|4`beQ?b}FBi1uD81LM$-#m-Yqo+<dwGmDy
z0(eO+Y7FCblb+okmvyCOU(~D!oTGfww$eN-AnZJ`8X=ZwM?Nb8H^p-7i;$2YM}rcD
zMT%^<nNUHK2-@cLEk0+<$*Ky++?4)MZcVn%T-SF)nYDJgbrViXMI{<zd4BX?x!L<{
zPvdMR86|x!dR4*KfEP6+dSRSEz<exkIj?h-eLymTf8d}Nex`IWS}y%+`Vf$87MtPi
zCnolK|9%-AyyQnYIezPHt5(JhdjU?)dLBq_ncTnrH&OlfSifiz1pD{=(nBWkXcHXu
zY%|%+uaw-knC=^&R{#TG*Wuw|Dys_#m3TZ%2j=WM0^a12k4#cHxF1>G=*gt-A%w{#
zrm>@=qduB>YRR5`eSMyma7HC|?85Mj-a7!GLC&~`tUGb^4qoiYiO$GyeKN~LNcB9F
zaQ}XulbWha?Igt*`CxGTb#OaSVFmY&AeX^Sr$DZgm&dj?xT#wXKAFMf`}Z+_^*kpF
z^gVv%0{EAn;P0O8@ZrN3GrLn#mUs=2c;=C=B;=@a(kC*G-2Rt8`^OdcD1m@a;pt_m
zM>{-D<NNttAvMk3T19DQ)jOWI{AoZY&Xtp4>gyPLEN$E}Pf=x(L~0BBt|T)092^-N
zlO}7u9~~Jf{oF7wk9znT;#OhWoAVocub}Ml;K4e-!OJU72OlTo<>l#UU`ZeVRetFe
zlk{!zXSk5@^2zGliZ98})=+Uw<Y=j=Wb7gQ+BO_OSBN3hmNgzsdwEh=5<bb(A)IET
z^ordQ5~Ze(4$EfR+1so9--v)8Q^!<&{q^TL%DBviDQ|`$j(Nfk8JnYrG0fPo-i4>R
z#~RFpPr@1|{%~mTgAZT6D0CXUZdRM<08iJ53bLYn=Q%MPMy7xVp2<E%7O8^_3Vevi
zcTSy-%1Hl?r;$kjT*c)7;wpuOg>V&`YWq8$&$z?*4G_nUT!{p;!zrG!JJ&iQVi5dM
zh07j=g_J&aqC7lLksL;8X=z$hZ76wbbPNqR$wtoN$&-{9qrGuz!&a|SC%d;bph+xx
zK8ulqwFDz4WM09XFL!}qb^7BwCGBe7ghCmc&yG!9@v{c+B*Xdr$ETf|+E}Q3Tn_W{
zpB7Rq=GA}h!}tNWi@Md`iZ*5XwA*LLZ+30Uk|mZI;hQDh-8C=IWBGy?M^CV|Yo1@!
z{i78T&mn^|{rQGK=Tu%HdVz`{XNV_@tMOW~V^dqjMeCjJ%FQUtK&4*2^_TOvG)l&A
z;;ySUU;D(HkS5F1AF9+_>G<n=H-=S)YijB8+=lbpuMVzEM+n!cFB{q8@))$yicEvW
z6!*owReiA~2hzeX&r_Dk3v<3X#b>ZIQ4m%E-mUC`4$;HC2C<XJr5U_%7<C2D%M;@S
zAMkc7HrLjT`Ec_DpE)d4{?!+US8^(kUx=6s(X%}nGctPi`e~eoy^EVkp?*)>EO!pq
z<lJrjfR|JG%r6A~L<xZ1KF0WkPXDyp4SVpk5EldKpm7A{wt*hBu34qsOuL!ezuCjC
zS$DK{h=%n>^W1|B=S`Q-F$St|CAIRo2dVB+A%Q|<rwC92=M%5F&DjP9N2!q=Cxt&>
zE4x|!?$1Kx&s3+XTSEJ6<SJ#`NkN|Ov)8vi$vi8uW)|`gmczc+H8kWomR$q}4xHH7
z_H}K_hkWd9IcY_vJiB)7N;BrT_Qt&FDOjE3KcB@g&G-{yb-YhYJ>21;Oz_$0tuhDK
zs}<mMKUi)KMN>r$wx1vJ8eFTl8O8q9n9S=rJI1bL)fLr=&P#wHoluh(X)fGDsGhya
z#aFbOV95~c8lv6^EYX8=GJXNKnxbPMt1*#hK0mZav1DT-OyaGY>?7yUvvs1vBrtF-
zRIa_foifi0dGJSAUFQA7$4u+%$1X)ip>OXxB!ZT`lWO#~V~K^3&{N&cVScD4&6?if
z%VrFuZLuGEKQ6ASvQU;`*3w6B3{dl&3l>@YTv28nPu|u`-y7XmFF`Do3D)@;o&=^!
z(+nwRm>M!&xGf%nBO&pDipMXY<+M!VvOj^=qbkTfU%Gn)!vRCTPZ+%*q%wNjP~9l>
z<@R}>mX0?%5=gQTskaJuAMs)`M$$zUKSN#ldmQz@{7ltbB+6$G1Zz+wecix-wc+wx
zHC{yxs<Xa_n2qb-xU78S2?_g51Jy4df9?F``nIR6kRHbN!@ht2{@kZ#qgbE*Mm(oJ
z0^<G76KRqYSD;=V=r1LFy(ZLFHo>LAmuMfzlPMLqDEM>%)2(b68bU6lkMCCZP1)Px
z_i-2*n-!9kNo%epHXkgHJMg)wiBxNIj!`#3?cH1h=jqc2sh`{`h6+!Z911G)oi;9&
z@6#u&D@E2qF^dGE9gr9*_10+?aRxn)_wP-8^sCb$UG_PxvhXNhHiR`ZFt)9tB95FS
zIUCYRvI#I{i&?hPp7!J?QTy)fX$(6g?=wPRWog8$goCTb#3>-~$YBVSN?DpAE{mZC
zNpgls`y5Y*3{U09$ANd@U~^_=XY=Pz=y8||XqjWDP;eg!?YPF6WZ({1p~xBJu#`m2
z184(+()E@38}rV~eG`D8NVe^=l=&&{vQLd;<!$UN{WWCi6SdNr#iMuQ1jCUsJ-eqS
z0U#JEJj92oU6=dx(zpzFmnAQz*l#2pKXo<4RSv&048ufT2Rl4&rwjuKZ^x;biHu!7
z2oyw$CQSJYq$o&*o1qe_aIMHq!bF2=r+x26o@WzG3V&|VWjRa8JSGdr9vf983aBEX
zc^|@vl~acfA8reAxCez<*!A6X?(3F16@HXyP2CoyvT0m-)H7wIiKUPgqmrASvM&-r
za2+(}d~hFp+?*KpHwI~YOs&VWaD&~y$@m8HnPmioM7JQx*pl?D@3T@bIgD#&3mtOY
zSmgKQt~*m;SUp9wxs+|sAO8IED89LKr#e+YB_a4`mP8eu%p|;r+8OVaC+b??bdR`r
z#8L^8gV#d|Uf4#^Q(99+4k>3LapKW^=R^#gl`fw7O;`34=*lh}J3<Zj@#0iHFP{$c
zJ*n8bq1d{St@h{WWt>IPrvF9OjUM@u5jm<TnTH@d7<Wc3@;tN*2QAPfQEcnx^tpi@
zB&7Z#!>a9k<b^xI0-^$k)Wgo@E8vuqd5exNAYXd5JIh3>CtAngrAU#3>*{ENr^u4Q
z=BEKR6;Y}X4(;|vwatDCex7i2W|JJD^W<BV(pjcmklZ*(!1c#a`(Nt7-(dAjN!rr&
zWcsgWn|6i>@YNI}aI<MQRHtsj`d1Eja%%48+mG4k@rgw;lcqX{Yr?2SXFRya0%V7-
zt1byrlBQvLCnO+(^tM4T>&}#{E%TMw>{u9|sWeTP-4{yzb;p-iPRJ%~tkk(v0sup|
z>FFyXEyu6TguPC+g7id?4W<f^zXkdy+|LKY0UsVqfN`teq^bhh;3F{UdYV645-e|K
zqMl_vT#G;LGTa%OLeCqPGxA&}wBnN9%xtB~J>%4k^WE8WmSr*}QpXC?rrNF$h3G{o
zrOnb|=)o=loULBenT?boh}7fdig;t?>LT?+&n%{?YHoRAB+e8<!DMR9Yq<+O@m#@_
zG_<sY+UV8t7`0?shfXN00QzZDb7W-XDQA}OUdXa^bIk$EEG8<;<3G*wgO{$a%stlm
z`caZp-xj?=e0OtmvuxIIZ!Of8T&Aok?xv&+J(=ckn`P5k9OHd;5)D&1siyL+NAJE9
z5*DtsS~1sYcz%zx!VZO$GV|nmFfHjuh|l-V)UXq?0Zx>`+5=RW#cVnHAYiRXQ#F%Y
zN?MEDEZcGy1ATgf?Y=x@mh^)0jM<}^2!O3ta9FnwKnP}36#NbY{`_7b`fIkOn`5c9
zL^2^XO7<uH@4$(pQXaz4#WE-W7b}@R3noJg9`L5(GU-{jaRAfOH}}c?u$=H@5&$Mq
zmxS`D&2j|89f!0%qtNS~Z!8DmrHaJ-`NN&Y@=AnzSF+-jOndSL*B7<jPjtD=r-oHD
zQ`1ns@|0m;X`N}Wb4%;75ExJ5E*VXTQnNo<gWP)Ptsx9+KnDKe_QvtNh@aKLXwiIC
z3}60C8jDHyYGvAR%NxLV&MY*kmw;l<C=f&f4(2of@RU*%&U~G;UR+qX>o`B;S%B(c
zs}AMmt_~OEURbQb<G!^%`tIh3hCcFC0inmX_ThsA^ypOGUc0--DJ$&(UiPa_omr-R
zKNE=3BS6fV_r!%9;Q$MWiH+4wcuHIJs@bS-D$kvwa6aNw$u8nEIRo`esgQ*$?cJQL
zEctM(NU!8tA`zH`<T-ARf2kb*a8AU9z(05~U~mWr*XrmD+$PVN)scZ`x46%op#ev{
zXhYS!uc1LrhKlE0%|X${S4H@e^c(3jUpdgz;KnGJuS{0bxy<y<jLK^tCK6qao|$_>
zPe~+j9X(n1GVPqZyL;Q4zE?BL1BoT6ZniQGfL$D`CFPA?E}>xGFG*{AvwKz<aq`p=
zjP)OG?ce#F9yXBUxLDb65`M)XtT?P^Vb!8d$G3Dhg^MUMi_wrS@XVZncVw`mjjioH
zB{@gQW#1wPckfW1%yC61lzeK<#pwDwacaQTHki_dbnwsk;6Qjv$DLl9z!wn_7K;*V
z5#dv&bhh)Fni?68L_2IWp&HU&9zP2)^vh<W&ZuT1UZjli;K5j%wE#en&C#Mle4Dll
zLOeVRkS)@c(EXYAl**Zz;*HYm!o@|4$+ty#Vk;l&Zcy_w$&S=*PLx^H3Zj{5EQ&X!
zdK(*&{C1o-3x?3}rw`#!E8Uo90l4~!neJ(=hpnI0B2}R15pMNSI7<vd-uD(YykZ3W
zJq|L5X=pCx+W|~RV6N(kCm=7cI@B1;;jK*<vCcYY?1L@Nn9C%>P*zVvdh(&551e@5
z+1FZaER|R|Go5c;dY_yP=C#V3xu<B;7ntG6i@fDpSX{gZNAoTeHv>e5??P^Z5Hud?
zt-78(PUlmfL<y8gd@uuC#HYdXH09)EAxt*UxKW5W4`zBn{xZS!h$&`OryyOATAE=N
z?66nd7H5Z}QFWe@2}8i|FJbr~Lf&;oMbZqz^YZQM+c)>4*^TX$+g`bUUV-h?`lD)U
zd*%M`SFRYXKW!F0c;XZ8PWRE2yXV|+h;%Nf;GR=B80aa0lgn7;p;Pw2`r2U{kJ1xH
zx9%{c`zcmat|(R<u;;o#!nxn#(sP06*pmMAQth9Dj*$Fxj+FE4!yPOyBz}4*m(b*X
zwe#yb(_J}x+vWnz-U8GsYpaQrHP>lWVO?iJzNSsl_n=Hk?%m!=sF->7;EyiG@6cP~
z)i`^2%DQNgroX>TSbzUpig<_XNzPMa$*`B5ye=+L-TeR=(}GDf{q5`93-|r|_`Bua
zUbyXr`={7<`>6h)g4$lV?S=a%%e!q=f6)Fb;yrp^_0zxbl*C`-KzpHfn{n#wtH=jm
zznHrWP?sK02q#O)pPMex@uCg=dU^0K=eE6I0)IvNP04rXaXg;lo_H)hGBxFMJnp5H
z<I2}Xo!W4E&%CMJW7Sb;bcu7%{M5_2;QfE)|J1we+(i}IyNmJ{pH7zI<$`{F`@+d@
z@IFI)hX{EYWer%pvLgyp@}(OtGzCcQ2e*-AMkvbj^;59(-jdO&V&tg#*#!v}<j=F$
zqfrZqX!pLe*Kh8Ie*S}^D<6vR<wX|Ts^}(oaNbeOXvt5hCyC?W?qmN_5c#wFPF5f)
z=pt~4sB`rtzj#~x{cfkV@zl+E=)OPX5u)~=(SqM$Ma$UgdwLF-TTxb<j5a2e&V8-B
z$!Y!9)U}5D4%Y|8r?mafUHQAybR8GBCAF6mCxL9-o7Y&2Rr&pEfS~=kaq}eOjMljb
z)3tu*F#q{yw>~4@h4a;lc?y^NsiwBJ_MVK8zhpoSiU%LwaP0o~`1^l#d;htu!Jmq6
zdun7V9{BEM_=_k0_7tHwaXfc?|HZA99pSuHJ;Ld>UTDeGG<dS3Q@A96R#UdNWcIIa
zNW2Aizd@B+<>2?vlpiAz7=3}Ja+~!Pw4JE#4IyBEdcZ8#P-}Ca(t*D^9zn0gWnMSz
z#xU*ttJn^T!!pzGDB}OsQ~ix-{P$bpxe$207-tjutDFDlegBPI0_*rLDy#Z)%)fc<
zU;V)EKRk+@tCZ9c>^vzEve{=zAOHRuKl<A**g(nV7Cy(YH}sc<5+vb_KC)N#3KR-|
z^oD-3<UuKLu18nz|D}ca&1e64HMRxrzmLuLydu6WaQ{lq{`bS%0{1=Q_1~Y_7P$X7
zBmXN*-}9esf%|V^`kq&|1#Vm5{>gvX_Tm0%h5p>(+xFpp&=cMkxNU*^d$qhBrT*a2
z+Ky6x@MvK>E%K<5kV9za;GKz|f8_%B%N-R@#^EShx4M=&;#1#KnA~#eo_1?T2Nhsa
zPxa*t1mMR#L@!)j=D^$W=Vs`?9{8eSvHYtzZ?-K)-amD&A>9fs(wwfVeilgYMq#1V
zDfOgw*)Ks7L`2!$e8S-l{rGvBv&X+X)p7AcH1jjg=YM*!Zz;sjY%}lbG^tYZrt&<w
zKNGYSnb(9;luC+PIRRK+h45V3<L_~2w>~qW8l5>Tlk)4u?lZkn!6{cp=jzGhSl&f)
za2_{zUs&{_4i>K}+&o->Uv^+?l-@&f$Ydua7~taZNL>H=akylnJ)37>q$}c`uB6KH
zN0k&J=&NltdwX^IYa(i4Ypx&36Y)iJX+Jpz=G1w7hJ64+^tgz-$0@b4m*M^k!A7{2
zI1WzX-itdF@nCn)o%~`e+COv=89;@Pm$_7Vv0vptqL{h%#yjV&@q*qm#p5m-8$q%Y
zI8t5Y<yx`pziL4VyesZi$9(9=UhTIG_=N0p{TUA?t>KNm9z%<DAK-<<ml$5HITmFK
zKTS7=HW)Te_s!Cne$#tpLG(@rZth9lMSEXvqVMfRwyf+}(B};om&YHrS{|tmScvs{
zM;5@3cxzurwXQgwUeb~mF<OEJWN_U%z4*d>0vEo;qh7q>k|F%X1^)ly*cUfZ5%mZ2
ztdK&Z%X<w#+Fx%K{_|4m3(yztZ`F2py_RL73RMZZ8-cw~*NCIJ{)fG{4y$r|0!EKp
zP(l=tP#UGA6(p3{iVBK^Al;&Xv?5(@1yov8KspRiLb^i`0g;ks)7>E5-1VZA2NdMo
z=icx6et-D5+3&m8nl&?PX4b3;_0M*fpSfZXD&0E=1-vh|-(E^{%0BMKLzMa5Wr>u4
zMHx5IpgFS$pZp+=IQ@$pWRr>P^e5U9tiXkVCA7%Ms@v(yn1@p=!F)Ll3T>YrmsJC?
z#yx|BSXGvo%V=D2v#v0}I>^=@xxcy}cvV<Jl5qp%1<u*Gyp#x{Qo4M;b`UFs9O9oZ
z5?`2l&7xG#|G1^E_M472pHbI5g`e$Wv)$b~aW@O1?;X#??EA9jhHdFU8PcT!cLxjX
zLC|G!+w8gwCgM7Zyu=fFsG7(+a39O6w&vJ7uu)Y8g7A82+2j!~^ovC{-U&K24!0}#
zG`R#|k?Ek&9mJ|s*q0$D`|YCt4VAky@G};2If71(>!v=6>&<DP>=5h1Uu8a~@5Wz<
z^JRpVe@3onpyA~Cx1CGgDMq&5OUI;kUX<Y_4&5yp$zS4=5q2ojk@7FZsnp)V-b+XB
z`#uZ<8LMhfgR=XoSg4MO&@HdaT@Ue0Y%y-+t?lfPF5y%0NzNuNPTjh10pjdx+g>(&
z5ID;#^>0zHTP-Labs;D^nuYeor)&F4{jGcY8HVRS31VSR*K2!={{u~4@HG4OTYnPf
z%xOvPcVUYXHshgb2eiThOGUSfaxHk!YeYukwlDNH!1phrVo-v8M_C^swB6~~82{fq
z01v`JNF4Va{N?`29KTjGkkihSzzD0T^KNaTo~1If;d$z37bJr-8E@(d&3u!d>PX}b
z;mdqmuw4s3YJEM29ig(>cYgk<kD>^P@Y5vs%FYE}A5{-KD`mAs0hx=~BPC`ysS%ep
zEbd5Rc3a+*ooRaeh}v)w(gB|#VAyn8yy=U!Hm^~)GGPhN*m03!O3&PUXLAKm5XXg@
z)cjv~`@tuO8Z@if3c~a*qrh}&)j;dXgWdk7qRCx8?{hQfhUV&>__pfi$&Ekq7H(88
zdut~_Zr^Zg6C;l|n6($Temcxg0&=M;&Rl}7Qa1w+`vnw~mmqLz^T(S|?>@X9M$DP%
zGl^sWWC;>RAnDB^r}Dk{<}=VGf(i15vii22c?@qwk&d>BF+BZNZUkhn$@sM9ZB@di
z99)wZUO2V|>=b)p{WZ~tIGe2$fKu-*tE+p7aBdg6Gu~6)`uO<4L*0Q(&AFeR2<AO~
z&iD&*5IGq4?wP637IsA|Brz8n-q?Zw2rbkDT8|S?Lp9pgOSGR5A!<<w7-f%$@-L{X
zjdd}_y!L!)rBULCAC+aJ;mx6^t#Q}$7ZD6LPIe3I#g$tls8L&NUW6Z7e8;PUw>Bdu
zfC6zHl|!!{uY{|;-GPlgtf5BcAlMO?%?4sFVqJ>$-vr97_C9MY5MIs^Nfo~!=jZzr
zp|G*8MludPi_@WX`7caT6w%`}8jr|GKuv$QJ<S11aiX@=WgBC&U%|7o4Q%541RlV6
z4u8UP{AF07i!EWHO=KWT%8EWFj$pi8Fz~DVm(gD&%PM<>gCI|LjQ2sSUu^gQN;85I
zg6S_RD=o}8UNpCEcnC}@ELih=m-9%uTYzHj7b}+V>-52|JKe1a!_REa`4r-zaAB8z
z#?4t{E2M9RgymJ^NrPM5oJ8ec;raz7zm>deWC2D#P)zi<NOug!W^5-ei;z%bkm1^5
zFd<ph$pHg9b=Q}EhLr>M{R0g+O-qXvabC&J+m;GonGXT^SkYXCe<92Y0#vzRpjxqz
zS*az>K}})i8k2jvrMHwEcaYWS$W$gVFe<gnJku2l9+3wG3wDXvCPh}3u|}f}-(WC{
z)0-`PSsxMrA3_18h)qFWYWE`<l}5j``aTw|ah=2{xwx*=fpc~ne!ux1-ej_Daljb|
zs5>^?@1HtW25MX~r`)tiZgm*8znkLmG~)H;O@6(NIVp(21UEy)Vt>T?{GevGJB&?3
zkt`}}@eO5-+zJYJin{C(RC$N2yn`Dy<YN)liZ>!^U0>y5UQ?5Ac(ncTQ*gj_J7?r|
zv~?_gkw7-x>@v{(Os5d_J7MH?4E%~7SXS@SAFK}wP`!y#k3<S~QM}!3uWbW2bYVc}
zqUxo`{v}Wwr$y8Q9T(R4<;q{8Zry6Zz(GNP`grl*zig$`H@^NS7SkrAtIDk02wFjG
zHaaZ{F-Xo{JXJ(Z<vB-~H9b*QmvTJD?Bbz#-8#0`^f5lEdk*}SS!UmH4}u8`BC<hp
zFy2r~chd*8T!*jK?j~(n^PTFq$tn5P5&Vy!!9Rh;U1%T)UvDl@Tt*Z=qlG$e-1MMi
zBkmt9ias4g`Ee*xrJ10sc<$N*9D7JcA)ev^RzmOg8!Lpehwo9j858Ve?tc(p(c^D2
ztD~=Hr8V@yZ$KMNpG)9|4{SYjYj13^bwLlM`wl#&GI|?VA9M~>gUj4^ymmQc<Rs=j
z6B7rJ6mNO*Egk3180A+ZE(P1H)t(U45YZ7xQ7pX3dz<j2c4Zi7PB7<SnmI-qoRDyO
zu>BJdgxMcZtD|ji)mrKT&w&!6ujRRB;DR0~Q=yZ({}#<`*y7dv13r4f!op@_zRe^}
zv(-rlQ==Xe%0$Q>J#*F+FXrCq%~7=lBCwA@rfe=v9Af1KRSE>PZ~#Wd8zqtev$jG6
zv!y;7f`767Ayi60YP#|*-^ou`yuc;|mXwlG<8jL%U`idi-cqfQMXvER`B;QXGe^%E
z!u<rGqd^m)eroR<)>~hmBM9+E>OeJWm~`aLondkUJ-#xjn0f6-XHh+w6v#xmTJds^
zns^+T3+-c(vjImDJGnUhBa_O+Ka>LE2HN%E!|$drfhVkam^Q#HWGQpP?t^Y!?CitZ
z9SV848W0wAy8ZYokwikGs<Y!q)sTs<9WVl}Z(dne=C$eOMhMjBKi@b&C7!__)<kQ@
z&J49E0baoI&wHP2L&~ie-r&ptrd2G$U+Iw{Bn=gAZ!m4%Tey$OFy7U5Q0?ObdsG7P
zpK6WO9X_!bWD(f%)H&OFs*&*BRLl=Y&mZ^0ns7Yg3ct<e6uD+?en^8)M00}%AA^hW
zrNr1~oiALl1`;7pVj;$L`WKXY$*jj;jYJwK@pYsgWAr{V;nbSaU5yu9i4%SYQc<HF
z>Vh2+wTJ4G???`u(4fx(U2w!R-27S9&c7MemyXRRx4oL&_byAX`ujD1%M$KD%>L?x
zi!Ywt1Y9NCk}+&;aOAXhB{SjssRV&$<WMYS61gieC-c>pr?{;a#|S}Z3Z7$0#`MlB
z{QHNs-iFNOAF;MSaZQ9B*69^YjAj58d2pZ}saMe^HtXNB-}w;vY}wrYOxX{bQ+1ld
z52PAoQv?=lm5tFVL3@e*&j(|Cv&m2#W@nMPiRbxv)pE$q{g<Xn+=w|7-BtEF(zstl
zybZKD<8<3iw#1Eh(kywRdic@VA#XxZ@q$HgKym*zB^E_Lj7g3KzPM3TLCdVeqw`w8
z@7FaEtzLc@lmbI0nR0L<{QNS3zm;pMVN5_lK|z#WV!M6Xz7Yb8z*?PljUQGR@&JW6
zv)6vJYRR8qAU>jk(9(S>E!xGf!9irsU{dq7F$>IXDreAsNuqpD<*HHYo|(NkH%z!V
zU&`;jc<v~dT3sx@8Yr}Kw17Pd&(hK|7xHcf*xE>i*$jNQ00PQ0bMbO11SLrIrxE2j
zcGc(J&0y^&$(X6eBv##~bYn3u-sRDc687|dw^s9DQjBR!p~~%~srH~2vsYQNkOO6M
zzlez3COoC_)qAAGl<cT0@1!>?kjn9HMY{7Z7y>cf_%pfg-8;fDW*`!=DDvc^tu7>|
zNi2GTTc_LI4uuN3mFNyruLy+|O#K^o9RjK&f%ggoUEZy}qZOMC;@3$)52MdCnmGC2
zWt2{{DVleF<?<*LB5)$$B`e{bnjcAPj{G9@`P&bhvKVK&HlL3uD!I*;qkE)4)c;gx
z<w&Ju(XpZ{XCR`u`@yMG#E;nz4e4W~q0KK&WgP;l<H-lZ*3dMb{@hkHXv@r3@E2nl
z*v_@))_Kk)R3Q#XzO~_Ku$l8ui9Q_(JH*}L4*CT&fRZCb`}{m(lK;(fzkHKj2CL#1
zY+YYE06~WeQHrbcdBp+E$FXNUemShnKn03tbLX(ZNxr%Ukka9q=$X?%Fu#t0b0d{V
z;o4jR>LN;U8X{a1m6Ew}rF=+PinMJo78cmrW&Z?3t^@HE!^okfk*u8R$iZH+V;ii0
zNKFK(sDcQ8yUwD8Hd9=c11#wy65uFMciHaCMdy4n1w5O@v888D@;*uzfO{P3+fqNk
z?zhUcTZ(f$SzukjI8}PrAh|&(`$1K0;KUcVU*uH5llxPY9(q)z{sCj%mTo$4FSV01
z$$kj>PI2ttjP$Q>`tazgQc5}2lq0xSU<PGrsw+zkmnZtfVAz7#Z>JexPS=DY$Qp;q
zGM@%oEH1VUp?REL_B@<o{-EF3CQd(7s(OfLL}+?CpM@yiH>qKC!4Rouv`<O$xs$bZ
zXUOw2$BZphG-=1GBb(am!^@`yd5AgjR;`9DrEZu?w%D6015Mh0nSG$*AQKZ)Yxdq|
z?>6*0bY7DI<+{t#pgRm$B?S!8$7{C-i*g{TIa9zSR3h9iGxLjkKjZ2K`ojkVwmikZ
zu=T?-_!l6*P~=}DA7EC6=eF=|<+@rPui#a-FiT|7%_fg+9s%GHu-CGOpAoKy#sOFB
z_ECJ0-105h6TfXXW2)qk8{h0rN^L2s^{=5-(v($CUeUvhd?(lP0J8+sz@p;Z>FCn_
z943WC2c75)?&!|bucO#XrVYJ$%q1jCTnEdtmgeMw-@iYb7~dhVC9PIhKK&3Fg_+pv
z@Kq22*U`}dH+X!3b@R`zTJ=|(>?QI*!FO7ZJZ@~@_U{gIOuholfC0nt@?l4YaGINA
z--P=QXbIIWj8`lj^D}LFi7r^Ik*pK#uLl?4P@vUW&%K+`OdJj2dH#C%f8(dW58n%p
zGwk;@cUd~~=eorYAS6zHxxVo#<kv-{VLB5h$sFOg$!ZK_?&#Lji?+L)B2iF}zrC~$
zxE_GGvg;=P1}4}z38DcV^i8zI<ZPZVZm5zHJ7+seE15Gp5nUNqwatob7{|&a+|Rt&
zp>4dL{%hh6++QjH8JOCuyWStN)w`>Fx8(=NW5A<u?M&phuZp%DX<6*+BP1Hn>g@<L
zx^%b#(!sFpiS0GBN4$GLH_gn}z2scXSA8d6epmv=%k7ulo9igxu?Mc9pvw|FneP)l
zXbwr*RjP7C$KqOt@l<0L8svvw>{J;#_>4xx`55ixjSR5D8=HEIev6G|ao}+LgUMj3
zVb@E|N5DRB)s2H{k*}xc&moTC(p`M+bbUoKh~S2e2O?i&*Hv+?KWW?)2%6>1j%wbY
z99Vj(`Q{o+LD_7LrE{rZG`J18<@aale#`R$tRzUKB`HhkC}>IbaisN7pak5lPiHo2
zxVT2R#qz#I=K$0$CKYDA(3(;5R)-N%<US|B1$_FHg-_i@O>pE@ErnIh%}^<aK6@O~
z&8S3)?F2+v%)RqJ!Gb?Swc1<Ov5(7!5RvEPW%s>3e|Yrj9sY)x<4=}UWfb{e{5&>4
zM8BT3f5bXl^?cNteg0S`JHskH1L`=3;(=YeVBEygN)t>Etm-@Hj4F$|=k~K-pZhQe
z0uM5tF1LVYr*ARNymoF5a0GTVe1ex~xZ)F%{3RGM=-Qs>F&B(_6-JOF<XB};wR$0P
z6W(IkoJM_pX{#%@Re~TiFt5|w?4j)?yJu=jODe&$8KuDSi4x(7b0+`fCS+Wp-wqxj
zzCvk8b=8-lI-kqGqi_OOQCS{&@trFN)IamtFB2njFprdpPqXY<i@P~~;)8aT({XdZ
zxL@8~UH+dxP*5Ty&fl?Hxw7eyXjz;iKftD0f+5Dg;QK*6-&L`?R_j{l-IQ`{`}*Pj
zF!4cYsAUeA5j!UPnjk5uy!_YbLC{zwSy<xdRYxr22ea=UtZF$Q<(MgB8qS_cq$Loh
zSi^h_MtZNvetT5RA?MQ>IK}<@`KND#bFIN4i>{i?Nf%I*tp+RgO&H<(;l`{^rX_07
z&+{ngwaSbEpkbo7Pu)y-JiZQus1WgSrXSM;&A`|+ZD<uOdT$krUD&sdcvQ?EZdbg2
zl8?7qSWNXTXh3EFoz~9=j;zngOCg|pA)kvQNaEQ~q{F12W5-W)w|8<7DKHV3pAFN~
zBhK-l7S1iTF)=k8`M5;fG~&Q{AiXSZ-DGQmN9)0Tew-SVdxAfi_&X+}LL!e<E2fGo
z|25EMaMeN{IfGzQs8^Ztquqk2*+OX7_g9-!9si(Jez?_<A=C3J>$Oeab$16F8~q%{
zPcx=NShXn~2|3kduDx3!IOywujS%tWAbcfpYXu6Q<1Jn;KJ7!-*GPsEOZnk$XP+*W
z&r`<kuQO_wPQ9SnjJYo2(BZ!pH(nVD$V%<1zE;L0Pk0E7HgI0ERcY8lFyv$$(EzeJ
zllpylk?)T!q7?!M?br^BnXW{Mfq)xgPNK&zmYCnK5(+p;El@f1trD6dC$T8XsuL+e
zbwE2%Gt%<@`+#4LPCtl{T>v%>>u8tgPla-dq6~@jLNLDemO*4K_E+S9=By^lMa0rd
zHA<#Bpo(m6z52<Y>&hN{7`?JYAW%DFr^G7o3D#YpA{a5pCtk$)**s4HKA%f3y<Tlb
z1-dtI;gdzur&oqr5ea`bE!TKrl{w%+CO{TyC&hZ~ZkV5w(}lsqgNzw|(tx7uLv~O3
z>#}b*nxs+`5AyP6U|C|UO|43W3MxI5UUDHi?IC!{TpIMj?&<G$I>iaP2QI%=M-bMe
z?Q)cj26>U{xiuhX*mYbCn)Ny0DcD+idLV$1yfu;|&U5%B6s|#f)}$k3>N4R;lYv^0
ziaY%(*AWz6{uj2gDPT>Fq$v5+KY21+6$YZE+Ciaf$$bhEk99y<(NkCb%p}&I0D#_9
zwz_kTt0!phn54M>i}`l`CbBo83ljcmdMmfG19C03>TU;@W~2F=#~#3xX0rE-4kyV<
z1UbG~*PRYripi+(@+z9t*$Xf+e44oYOcbNYE1D#}8gzS|3aW9k`CcZ_Oq4ldFAQqx
ziZYzBpWQ@AIX}M7Kf`nE>%q0nw}#$L*1UoRRT<EEGq(RNUzC9AL?2<!MM5Pa5F2FB
zU#o}tFGPVK<>G?`@Z=lSJ9BSC<nJn0Aa<q9&O-m{W2|mA|DD!wO@ghlqkAlnD<}41
z{`q*n;n&&ysEPWGle4tEl1bTOHluN=Q}vsLbZy6o-)!0F7@$8a%@L6>5U+M_4cavn
zQ``cZM`9mAzzSl2gY$Kd5H-Mth_)?Zo4|Zt)H%zETr<+v-hzujCy!4z*Xxf>6~ShJ
ztqt=NGVz;`$<tt+utO5At#!kYr_rBl%ir2DXqRMFM?0PXN1m~rn9FeT<fNWjo}<)v
zYMSHg6vK+lLp{tI6s(FR!@L=3w!{P?0_|&{tsr#Zi+JTLA8?32oemW6l?%@!^xlL@
zh<X%xZ3T_r9q)E9eNZx>8D7#9{rX$=O=jKZklLW`LhvQJq7{s6x#;bThiC&fq<BNq
z`&Aoq14VZ_cGxvw9uWoqN5LqLP%8T!%jmnC@WD<Y19tu;NT=DwZuef|_cPEzsqA@!
z8DBOLz#GKTwg`<Nf-b*tzMT4RR{H-A!!7b#n!;{fmLtcQ@7MDTIO^>$2S2wDjpv!y
z5xEB;C==mg9}94vTzP!t38pHnT%F8^=eW6r;*GD|bo-vsdDwko+CvyH^5AU$Ke3^)
zSZoy)6>N>X!eYZk{e3U_3JixM`-W1dq^lOHxXT))RIHnqmgcmFBHt-282mR#xr2(Q
zrPYw)#l4+<@yB=E#o-7CJ8_@2Bk8sz?iP~Mg&_zo&|Z@9^}pdJI~WE#mEE`hhSRka
z_n;CHnGW6h^R>!$7@rDHJcyF=xyJ9e*EW#}Z^xi3D@!(FG5Sw<gd+Gp3=2J&eZ^|q
z&U&_fhe!sn$Y+1cm;a4+MMsrCoLI3nP2GL6TUu0;y8|pG`&mj!_3C0zm$o{sJ4pP2
zr`ess;<4DOaMx~{H*#woPHp^cqv^qBo69CJT=YK-w;nGPmx8!8(u)5RlK+EnAQTS@
zZauWlL$RBI;eKG*cg`pNH$=i>DXSlW{j-&B22r!KbNeL*#Wa5d@(g}*rzg1`i|tH3
zD8w&jS6&`7Pb1M@fJ72HfEc5Tz(>(U959uBT$?Agb;C?S&{{fgMr>(lXi&4BJbCi{
z!oU_m>JU`oj0pavu|to<tlHS}uDo-on{V0Ae^6yQhu^M0dP$zE-MK=CD>nF-iQz^m
zSUWultgx52OZG&Kt?(U(x@^*TF_q>_P#oU?GtFf1jFuDm28kF%1|(ahKqFL`Ynvis
zP5&Zbn2N?#gqVIxdXoLUy;v#>eVp27!^$@d74!<OQ<xyV;#t`6Y}qt=NB6<(jwQf?
z*Y!Jg^Y^Uh2B#dNqWovjWnXOx6aobXu8qb}BH1s~kMx9Q3DsXeyFp-ug#f}yeX82>
z_%1H5x-fgXb+&&gX8iL$g}Kz43|_0Lv?O-f>Tg@^SI|I_m4`=S98brmd1S;;B%kfb
zhCTb7K#Mt}UVZ~0J4Wj2h=|)jO}wGD+7g<7qw;s(JjVf>{D5=SmN;FVcsm~uw-MF7
z27e2iI40uF8*C1P%s2vpTQ<(a*29ll-4=WHzI2nkzX3k%2QVyG*k2(O;e#2WlJ%xb
zUE@wRK7*&mNyd{KOba+Vq6@b_VpI3i8jEiS7t8NV;{$5hpynoRfPZB}?wAV@60peE
z((;>mRQ>?mC?nue5)hD!Yy}=+w1*ELnv1@E6O)vXV6X1ROXFufNt{FZh(w|y2J2sa
zy$Vvnu8|9*{JsHGf0%HY>9U<s$jm#+4Xfo0FhC^GC+EqE>7$54RxP6=!%H@Lkc<nI
z>)p6V0aD-(G1)oI7QbyNU$^3D1PpxunZ%MlnP4F9r{-#uYNTmumP=fMF2*-gNn#m$
z_E*9MaREVxr^$S4V+z}!sRguS@?^a&7HC_gZ}y1&pV`?-`%;_9FYKw`ma`<+i%rlf
z9ZYhAh<bgn$ghxWeja?9?Qg1&@-*b1QoNJK7k_!DLm3!84PwE6hW6LVRON%RvTSD+
zR3x!mUcy44re*7S614E911-G6pRbRP-uWym;EFSWP4JHTy~<zh*k54}VueN8ucZkg
zxY6TD<4r5h@y|WoiX{sivOi9Z=hNg8m0r>cftHHk#<&nzK{hYsn{|#|@Os*sRTB0T
z9zHfhgr#=`f;J2RzvuRb+WyWZ+Bt8bU~(l0QNVqy++8tIsZfr3aFP9e&&+CUi}?AV
z%11D=(Gz=aZROMmi-U(-&N{MHnoAr29yi2u)Ppv!m@Ed9R!H!TaJoQ{87rUE^xiCK
zcSL?>rgdu$Cck%Fn7ywEq9;YTd#uin_!>?vi(xKS^TDN=z-A18e`M|_I~8{^3q^1b
z!R~kGb8qYX7!;7~>vb;)mdIdiYr}E-($sipP7&y61VW@ep;gvhIFj-CY^k@4pY<rU
z;YgK1W(XnEulQb};(7+z_>OmQHiOE{w-FC7Kd#&i$C~uFnqDxhJYVUfKwz}Uf%clm
zTN{j)DCYBR8%81~Y=oE%8;w;kWt3;^4J$;NF+jSeXlAJADudbMjl2!U<c|C`erolF
zKYa))N9CI>cRNwyo^#Hy+4_)OQr3c-WY1;7)gPJz+m>>ejn3w;S(1c`m4kzTZz?5=
zUn`Palfe-55wp)Hv$RS~L#~E81yR5vQ40l#R_EQ>2aFRVP)Z)L{<I0Dk*|(~h=_<<
zu}^M}lk&Ji@axyNq8jnFmq3)QeNz+NKUu>owE&dtzl1jx@{E_|g6Q|qeo}|;ka4H#
z3Zj;kS~KU!&_M5Hqk(X`G{J817q*@0k&$|%yoLf?UEb=B9`QVAU*&^FvpnTYbp-zw
zJKwOULAyx8kR`@W)Pww1*3<Ld{KO!~(zuiObggG_aO=LJ<k{X+k(n_#K=g=<AV4!Y
z!$N8=d&n4P<>R|n9E^dS0!g#W$<&6G*97r71fSqMKCI=cqxvr){$~#TDc!(LCWsIl
zvCNw4G8`*jdeO<6_fy$bUTYCCu+W(X>A{X|2aLQQk}pl?w~kkc9pdHXHR{QWLK$TO
zu7k{OvGJXea7D6m6C5EUE4(vahrv55T<fJY`B4{@n&=09DGtLx#CUP*XJUgd7Z}YC
zwSh)%2LormYK{gtOrw952U1?HqHas+wn21n+?n7}b_0sdOc8B&H6kOBdZgvyp+_JX
zT{CiQ@6R_|x_p-~&pX0EUZs&^Mz1406GeUuKT2!Ed0^WZM#BSTd<|6YfM8A<aN`8H
z_4LsvG)T4d>AWnQ%<<P*g(mK&9fu?XXU~I>s?2Ml<p(MrL%I0qfzzLw?a1d_1l4My
z8eN1~-~0Lc`sSn>_vh4UW>KaCV-O3IfYDou-{cs!Ke_C;!yvulax@z+PQgoHopAfI
zwr+ivSRuj|q@%z>Ty!FE0KMyy*;Bt*lQYOLms%ghEr`knm%%D#c?^r1BbaS*zM;jw
zNe-X`FN4)W{}PUVO6O;AG2T{Ze3NCFj_wcOMi|yT%jF>cFyn9V*+PN%{{8z%byB9%
z9fyaC8e{EBrj4Y0<H2rUHDu<Nl?0*^1y4)JMM7|k_55b^b&n95DEe6oC7{DFzpaM-
za2FNmS++zH5<<AFASlFH5Y-k~x?5!$Z><f41Y1CB#e?K?wX_wzjNKVL^CO|$rSNHU
z?Aqk-Lx%m4NsoX^Ge=9>calTI_4|Ozwch^z_aBq<j@-C$jL!{Fq1LeeHf=tq?tOyh
z+3I4oBF~b21=k%>MkzmNmIgp_*5Pe{1--M2OCAdEnH1iSPRdtV%#l{Z8B?819ar<y
zET;m>8XHNWssO!Qs9s`-nyf{9kRx2($u^A_>Z)8CojI!_IdDw|#gTes;C=IrP8z8D
z>xu&2)(3XG?1d@*BS6BTpk0?yof>KcX-G{tnG1u&@6Wn75nl;9@y-AzE`7f9tsO)3
ze80D}OU%soe9jYPWk)2Lib&8DZ5Jk5Nr%k@0UsBhi$Blkk=>(peIWNdg*9R-WB9?y
zJDCi|1y`f$_Li0)h{4%ftEGV?sJ9@7N`UMTT<n7IY)}p3+;oZQqJd~AA;<CvLSm}z
zpIuu0j7rdfj_x&nNBeVJP}tAM2s;#er$V6Y11vY!rZ-ZJR;BGFZt_@38lGlBr8AlX
z1aaR9!#sQ$x)SkHbk-m9Q;2gSs#9HbRd>9c$M|M_;l(j}?BKp(7M9dxVL7q?<d5RH
z1y8aglbU><3E9hCKs2UJ+1X<f5}Zun?q;`LwBj(5uG|A)p~nOUZe70gPA9IEQdL#`
z5Y<lX_YYzLW^Fwu{*!C=qs-3E%D$olF<dIRAX0triAnF~DA6;$6Rn^sO|*CR_xvU4
zyg`@DJGl;ocET(fROx`q+V?PKfaw`T4=a}Hm3q5VqW&uRH7Fpr)5B^XPh5~7P7B>E
zpOFh{6}zIB=n7Gwi!KPBOR{HhmmX$TQ&8Zi(P;R>AxW5w(nKDV;P!ku(c%>&P?BGQ
zQg`Xt^-h5s3`e~jzUdYVQ<<yhEYkmxrKLezb|$TH;Ma4|pEs@?4%hcj&1i&*BV7@8
zk~LgFoQl~uDcc%zMqW0X?JaJNG+4-uPc1U$Kg2g}s5@_9UL@BF`Y$``)u3T9r}nVq
zeEM|Z_$&HlgNTk2gzDssgo{~+ee(RQU_6-6d?2aOM68{fugLk)BSDRnw~bfObsOui
zX(t#%BxT(X&OcrEHDz|nZ8y)gc&oc&44aqE$f3O+I%zfE9w-w;nI~79pktamlp6iI
zRy`)oVx~w8eb}PLJWuqLnhBxGycf8Yh+n>yxQOS?ZyPXHvG=egXf(@ECjRLm4=uNg
zn_D-Pgc{J76#TH_8FZ<wJ!;Eh;#W4?N<boJp<^StY3N^T;&ti?bj-!oD!vwR{%;&z
z61Y9X<U--Jjes>M6kwqAk^vC2#nR%&k2U(MUC%gcLowCYRa%nUm9gvV=Pd@Zv&BJx
zgh<tX&SfdWBb<G<ERJh_Yc;neQiudpwX~8?;=TZ((O9&Gi!;Nz$$r!zP!MqmcY1MZ
zi8n&YQug?3#Zxsd#KV{%KTo4jZ`ltDpFr_>Uy~59Q=z*GqkBL$U{eeK-Y4r71RJ;r
zQgcxCSv%rx3Hqjl6(>iOltj%Q&|M!thISQP2(_^8Y@}5{;S_|hvnJym$3)MqV-@kM
zi0T_}s2O&;O8|{kqhO%@?)=DW7Hf;BM!XwK{0kLX7$6HwmGbUW6a){Zv$Jy-w^htK
z9Jm{SO?j6Ok+Wkb(;iqAu57%P8rNX_v-efjQL{>k17(VN48Qgw2a3FFJgxHhj`q-&
zF9*p&P9z<^TieKD*algyp9@OBu<rZ&z%yeUvq2rPLyyQ(Uoop|-Yy$NnmP>IE+sPB
zKcABY$*)Tw9c!qe`{Wo(NY>$n0hNoOUAezG=i~}(c<`1w&5Ni)LKuIm_1L4b-|{cC
z8s?zqb1-Kkz8a)-zQtgFWrYU<!;}v%TV_8sg9`)pi>TP2fBfxqt3SVv4P53hSCPLT
zNZblh?0eQR@YR9Uz1{^hXRaT*@!B5QAPcuMmpK7QmR1PJC0soq!(G5ZAq*#e`xUO%
z3y$SEDSTl90>|qPT)n;u!b@D>JCUrQTy~JHkaHC~JN*D+sWI4Z-jV+Fo2#-NAm!#2
zU?X1@d8E3jEZf0A%j-ZeV#I5lq}zB!@Gl(wmQn~tC_m+8YmMzfGNb+VBw4=sR<RmA
z-Mxn?+!b3OZw#T*_}fBLI>REQep+aD-DQ`55b{8}iVN$h@7UqK3LBgmwS;f(bZfBz
zADdUqThe$K)@`RgH@GoMfYoUD$V{n<l7Gnrh>1k(!`IR;OMkt+jGWc+xB4IiS%5d_
zu~r%TFX>RJqI&b@tHpDVA>?8KyH#MXXzKG_jylQ!n%0xfU@3P1W-f?<{00F$!RwU5
zPWc0p*6ab<Wo3r#)J*E&kDGA7eh02TSf4@m_ge&^3yvx7gEKh)O<f=NJwWB1Vnsjy
z_YkdDPq9p8K7%6{{^BxfBTr;sMHS_n;mqyyQsJp+#Hd?r9M0s?MOI4{G*Df<57z_n
zL}8Ih8?TRlyXDpcxi-~xCDp<Ds@90OMZv24!rXssCpB&TAF<|vR#AU>+TDU%jPZef
zh??K`7ly^Ej9vRMN&$GEOyjci+E;69+NrR;4_AV)$&mGNmpkbP2nZfv@Vh^f!~Jh;
z9B~=IpM75p4FAFc{5L~T90H1^L(TKkW<kkYiLb49Mx_+k*{P%8;J-P0?Ti!!4}tJ8
z_ohC%&i>tkowh^w|FPWGg!?UF{(sJLK?zZzs~qQ!j(Ac?cj=Jh&H13fg)O<dQYD*r
zVYU>32KK199^xT?nv3!r(^gyrPei$5ZPFiLlh`H1UIi?>VENeDpGWSyG^}^U)0xo@
z!oF{{!3!QkH>+&8o~)2Wpg=|R7}Y;i>Bg@S+atYp+qQ9XidzVYsc}iN^QI`L=*#@d
zlkIly(AhZR&&^)|Mb^E1iGlbn$+x}D<bWKfe9v}#T?L*^r`;awQLsUFt_D5IP9qIA
z7dJN{aTHPFoV*KVuqMe6Um-#MG93HAx$t+F0)jmD3|rfO6Gn6DI8cEaZ_ZQf0{v_H
z0WJ+xU_sy4#~U*E|9a^EG2I$!*BEMZrro$N|1V=Y`8Xo)AG<YM7iDs&k(kVC<SBGd
zWis4!*uv6O#9B{Z|6q|P6IPKIi|=ge-Oew-wUI^8RT{|Yt2VGpw@;ZW5I^j-cEVnO
zNEZ)#dwZin9cIB5jf-0QXUkj`oA0uZtEuI{Y~;%0AxmVU+?}5amqao8t0P5XeL4Pm
zfBcY+>rZgvD9bmTx#Wvfiblm*4xkR!4e1kDSXlPp0BqEkh0RB3(`_n>0lfA5p>jjM
zQnn_$6eJ*v1O%k`D=g=se}M7N6zYLg-RValcev}iksoul;b<YP11J%b3o;P%-_&Q7
z>{6_bLWu@8-J;MzQ;Q1$LW>Qnq*M+-p|A{HnKujD3_G5d9ug!qz9}20VcgSKj{|b)
zIA&A?L6LN_&-wWbL$)9Q$R2oz58BS>52Jw#3GU$V@bDX!#JjZI9a#Wm-@9cU0AzG_
z*{-B#bC(vImuPuLvIVHeZr{F*t{G@8J-yIuD=C{{Ayvpx{>9Z*ueDw$8k4ih%;*R)
zF#}RxSJx``$S3}*E~Yb`<Y?E*_iviGxP6`h`33Bp`hBOMBNHMW4IKWCGN+HU#pWFq
z<(=plc@eEh+BB6pb)B&UyE9F{u^yCZQYKg&Y|bQ6o{aVSggf6sn`AG{Y;M~#JnW|T
z9=hS%1p(nu3Jcv(!xtzCz#bdi9AJ4V&vpl=rFC=9#K*_mx!H{8wjK^lKE^X_F36*4
z`5IrRnuGw{`my8?H=Aj^){sP!+0H*1@v@z%QM-4KF)?y|%KAnrxIIwwfD=F1dyF1(
zj^mw&LM)O8Vwdm}^#{#F`HG8+TaS~c@XQYlXAW>9J)$1Zf~UecGu>I#CSGTUYL6^V
z8bmwmz1=NdPC|I;O8)|tPqN3JexEff3uTpF>AKrw-tAB!#vj>{_%bSeBnLAm{A~b9
z8WM}siWbz^Qi>S_7qx$@Czdfp7Xa;GUr8vW48&8uaHNb-Na6sU6p#|dV++Ge&p^Xw
zGGur2LSuIh2*TDqi?s^|fb^InHP|OHU6PRGqqBnx<|i6<zOYr{@N!~KW&Ik+A3ZQx
zrz@MFeG615ZaSLQ@5w~kn%bX9h7{eqE9UzLaRVJV{~~6dL-jN;C`Z1#*`7IY;DG1X
zTZ8qdTVJ#ACUhjGM1c?#mBR;l$K7l77iR*Om<Ae=gu9=*4Z6VdOL-0@onTOca#_Tn
z9>OHNPO)kFGn;^sk#SR|yn;e5gYq4Vx|p||2M!)|AT~(9Ju9fPOQ}Qz4?2_8TK|vX
za;+Fd9~z_?o)0F6ho4CAIvq`)#_NIwY7a6THXbWBXu_tUp}|e4T5dq{_hbe&34nvZ
z_TVi@=;WX^SWjiNBKDKf`<8ysCnp4@Hswt$iHJeD&6*q>!3eiOZE$l4p+p^ddw0A{
zA$|bZ$Brl6V!@_>Q=)*{HktA#-6X|$U$;%hb7am|MPIF~0z&55g+caRYQBOjuy1Ap
z1)$dKVBch9WV8;Y4M%zxfo}4$cP!<JRVLoC;Tju*9tO19Z+Dziyb=5GMaZ;XBKZPs
zbs!*G57`m{WmeSn4RjoG<(mzQMOHL`qHFb_8tJ!M<|Kn*P?jWvJJEl);&BEF=Ug(P
zgSH;t`IfonrE7USt~{eNl=R=k-*<wN0GHV-d8Zn@!u&7oMl*nA%@Ba8n#X7@mZwq$
z!D_9V{ApP?n29;kg8=zB$xIs@E-2~OkKYS1A^Lj<J|u(CIuyVu&#E-A2W&4;3~WES
z6_>M{o6OhO)u=>J(MHGbR}C7{&5Op9GI!}n!jlCv!Nj5}0Y%*j5rSmq>>QQ3hg1lW
zG{yV!$<ipVnS!W-8{c-f=iDs;<KN}Chkyml0&)WgV=hN1rW{KY@xiK&QabF(C`W?4
zrKu@%<mk~SyH{A55I;`MKY>aF*khZGmj~u|#oV~4hsl0n52?{thuUPoo&rF&*P0Oh
zjapvEOY2GHZixs>{@9dZjX$OZ+4Np4?`z)>5!Xa1JqtsEAVQkm?=%}|Wy1|mAa|ra
zc5N(mX7b0_Tfzd@Uj<heIpP;Oo(Hu+UsneoU5!QlJM|m%8J4i5S3^Xr#RaLgf<(Ax
z_hjuqGEs^+$Tvh-o-8T>VMwZrG`PmEb}KwIpd}4dqCr-d3*`qr1uKLDh|8=(Ri7EX
z_2lzlU;WenVEaN&oh&0-DAyerQgLx?9Bz}7_Az?z%2o3mJD;SI9yxNgQgpMD2<wRb
zCtg9!g9|+O=9I%w5;O)_3v#lS2ULMqM_v#8INDi+YkZa~3@&ht&Sj$jUEcJ68XOV~
zKi}ad&?>$&aDLfrhN0av9(-IJl3}f;>#NN!n`vbpqCYpcA2*>$Bh<6-L#57_u7=0I
z_S{%C%jlTOq%Dj!YFtGxaXhW-LAi7vE{zhBpX?Khd-Zc$i$@GP&cSsvuYv$Rc;h`D
zYVID4L*lRrQa}k$5^9ypX{%qQRZ9ahV$u?mlJXP7FFg!n{qz78GY6#4<_)zc$~2@5
z5r?)N5Z&L$hXq>&<kt_l7#9xBbPUG1gux%$0|ecHIritH*KHZYcJnetbYLTWRN(7S
zy6BkcXLaWO+;^ZLyMLfW1T`%!I*w-`B=A#Wxb#DTgL1gYiR^5+F7a;d#r5&w1qpB!
z6_t=_LS{wpFs0K(#ew`RiAV(EV=c$-#Z1uY=2oZWnkS=-A50BmVOXG;sIhQdWF(Wx
zKuk;w>#0+?9`P=NS05^(gW9r8Jd?GODUK;bANA|wPbVUSgVm(>KcAgzcS;5P(t&Sz
zeb4|z4SwvqFJ&-1B%7d=HaT)82rt7np9$nPp-=D}dg2r<jSQViOQxom?c}|aB{3dA
zm;{b!c?<`8vnP9+->K{He-7aNUL)4nn2~;v<ZL<j`G(PsXTAcsHOV%knY?GFoE1QR
z6nEl5ii<T%cZ#Yb{nKflnhXuNgkt389*;OkH;Aku?@v${<c!{X<+5^jsL|tUY4V2?
z;{%Soi6K)0-HLNBF9`~&XK)w{p$U63y6VY{?-S#1#PMxwX>nKYzb|nUCrP(?D4-?2
ze0q9%MH?bM!NhnIsGVIuH_wSRsEHDrnCQE?q-WV?o7eAo-P$^BIZH~4G#9a;<`#5X
z#7fO<ur<6mb-mrw|E#4MDg4Ua=<X-90VceBy;Yzjp{bUAhGk<#;3e>V{+P;KLP+=V
zHiA-H-s5=wVtGvESu(>;6<59oHpbQ)R5g;?irG#LWw=J}eI)JCofTqPH`bN66qDH-
znh>TETKb`V`IqeMc(!8AV=AR)dDfhKL)^``L#eX(4_jyUTi?l&y*KcQsT|BSnHT7r
zH}Pz0u)N#(j<AYKmU{m`V9YjGPt8PHb1|zo3B%wP8eZ|~g6({oRC`Lna;+6@ijgt_
zH#^g;Ik6jpspH?Aa_ck$@sUwPSWPn}PC2Evi$%3jS!4duj|2{KQza@*5?w&0;|WvN
zR&E-9!|7mP)y^-h7rPucx*~4YZ~Cxjpre)(t!=B8f&Znz!4KD3mEAPHN|)N85DBU-
z-h5Z=@@(k9DeGQgF)aRtUOl#20m(JFwSFM5GeHX~4qddY`CIeDjZ$Y_Urr3iAzzx`
zRuavAYw<J98u1R3D*T@#sO9zb5)-mtu-xOdwLP+)E9xKU-{sLG7IKJ8b_>}7py(p9
zfiMVZkeBb12?_=~>Y6jixp`85dzZl}?qWbHj&d7>fEu~L;J+l8kK0+^pqQ2P-h9jh
zUrR{7fbv?-ZjHs8qCqxU-+j0-_CQuAJQDTu=SzpN>b`yRfMifYLa}hAo{rAKjjMP;
zm)HnEVCbUrRLn{m$iJ)uaU}$R&Dun=k#2lh)+T^H26{{hE?ZsJc%V-YaJ+mbz1~&P
zabEQv`y2><E!GjfSyx~00~GiS4%gzeN|D%yV_QXbVjTmHPg-`NH{YN-Zy=}muqc-e
zNLJj6&<JNVF!k}nl+LNp`IXS*;j(>6#d_w9Ct3(NLS9SLX1^kD7X11m!a;kwCax5d
znTwxMzEZgq+j+~l7ZNJN?|Fpe@t@rwZUuGQO*8uaLG`59J-1i${eC?Mc3f!*abo@=
zmHvR*{S7#H-PZ0~zGct=$zq-n@Z6D+0_Rzgr0*9(hIkHmJU$`gQXQ>A4g$n|K(X=%
zig`@G#t^QMMf^HqDr;#0q|uc!X<Lj!heE_D7>iY_EvCwRTQ~k~x_0%glG?*w!^iWz
zfzMM6a?xaX<`Ox<-=NH=4F3$1DwBnUhK4@DB_kudQW-_$;(_;6Ha;waNT8o>FHAFF
zCKbq$DZBSksb?S`p_-u~>)va(DDB=9Qe3H-sgF??tW7b}1ldgo2L}h$2I~{lzPIJc
zp`~6r=`7~@EaF2^L4i<DPb~rD=JxUN$^G{Fk%Mu6wVkJD$l^rcXugO?VO@r$Y0tob
z->fN@?c%&)t=c2|TScDKb*K01g{ae<G5IF<`Sa(u5x_}q4{~EZ+${9?V6#xqYS5H^
zZ(o=>OVt8r3~2ia^_#%91}u)rAQN=}w1V`RH$6%kC#T|ppLT`8>B*BPp#Eh^Va?1C
zP?1C-MnRyk*i~y09p~}cCrxkr&OR!)Y^{z|$Zg871g?&;N}Xz-({g71VvEWznyfxe
zRU(nZpt~rJgM&lv1JtoxpVN=*1rG62+2mhW_&S(DUnNd|k3cCDs$r;<W+IYsro&x;
zrj)W;B*Q_fR?pQp*1Uo0^1JA0gHhv5w(+S$qM3xEStZdzu{t>%GhF%}REIN4z$$n_
zWz$DJTnh18<&`lBhOIeNX<k~-hPXwa?CDF@HZ;`gYrX+WH-sBacNTex_=j?E(poQt
zFIh@4?Fnsv_x`<3pZv))Z$fnPdUz7XC6Gpb-%eH<=-msz_Yb^O<5+&TC%8PJt$uHo
zU8O8Qy<BaaW@+U_DG+;={qucQ?{gjTFQ!OeJ?8$Zcya1o&n#<d)KG8l%T#ISp_&V*
zS3-3LJ}Z66lDU{)3A#6e<;}iT?glMbPR^VO^wP<T*I=)kFuQjUr}2>U5ay++G)G&p
z5A%1PUEre^B5geEtegb_7N9zwS^G_5)<(yJFRtp%nPI7N-U`s?;1np4ZB8~)SZ(J7
zC5)WLH=9E)QTW{hs;f$7Nz$z2so>kFlR89jjr|PPw)AJjoTBnVjS_m4#qI?VGzvOr
za+G+?XLgl|-;35ncJMA_hL6QV4+GDf(PZe7m`%RyqjOL1D5=U*Brrp1BCD@9<_e|S
zCL)Um{gzr9LNr4|EGt1djDY(1rGUlyLd#im>wBDa32h$_Ubc1RlW`hy8O*9-Gw3ry
z_KYE|7!xw0x%o1e0&K*@h5`24^5%wn1q0)58HAT4v|Sx_b`9?tOXx|t-$P_HGupX-
zsJCX$m~Y0i>guS!4zlOv{hqq}J$#qO^Ibx^VV;cP6JshymeD@%&wT@#MN3sDjiYJn
zHG99kmYlMXS9lK)T-;9cL~guJ5S+jYyR#f<hDAy-%Xt?7AJEjDR`azY;k@keWT;ng
z_(^#jUPY<x#qP1Vtha%HcRWb6>u6HFn#*x`3Cs9rUds1(XkKKXaf|K=u=z&!&eHW6
zyQ6$Oa%sM1vD+%N<^u#Yneq4<0}U<$8%8Iuw>!&IRB34->#fR?k>70Lb29U#F%{$?
zMQ~qHV%qyno_|T|OyWR9<B~Fd<NKF3{%(s(5Z7rYm9s^-q&?@)e-cn!w=^%a^t}9?
zRJl>6Ip8(*OV0w^FXT(6mfF6?u{}NKjpQx?(<R#X9L??O>XE-Q8=%_u?6EeD*ZJlu
z?))buNv_;o-i9Ah3GEJx@tm~hGCoI<1*=M*Esb_8fOJF<4XoV$mIw7Lb)*GTIZMiC
zrcT>n3Vku;7zLC{Dk^O}CrfrdQq8&i#LeT)|Gc|EiXU(^yfO%pbdfVOZkhs4@P^7F
zI$Y&s2qW5cSGB|}Jo%KG8@i-)mx!bakI@TB_!B!xp(nlZztmk_SdyzlBu%r=$W_vY
z$gll*(j8aHCYsrQu0Bm0<-?@m#P7^aAC)W77rxhUA>XYt1y*t=K<8A(XYP{tPexgn
zy7SDLYNr}>zbPH|C#)H6GD=N()u=E|hMyrL7r@X$(w#z+l?q;t)}CHA`M(izwc54k
zYVyy8kDo0ub}R2rB-_KtYY+Y}Tr?EwU2saTPPj2sXB;Wp^f}>%xv!Q}seaMig;K}L
zH0<*;J$-4^7SdFUGoNNZ%SM;`w?FqX1@L~>OzPC*36yT4sc)N!ALT2#^99M3f5M<W
z4mrvZAjV+O4!-4B`S!3LSR<4>WkhJ0uQTLe0iD<oau`&RBZN|c5j7sK9$y%IFHj?#
z`{ynizdP_EK0|?Yce<6kfC3IngIii!h8%Hlh*b)LG|ZFHvfcTj;DBj>|JQWr0x)k7
z4VdvsELWcc4J{sGQl-TX4pGQK4{8kdPseuY?JY=w>y)$;E_WX5fLb!&L?r?`q~EhM
zJ6m<2R@jS-2uYKMb%|EMX)%sRMzt;FeFQfJHYuAr`xOi>{VTCoBS`}1ZRd5gXBPBs
zEAGh`X)jq&Jmu!7n@3phQ-n6tSYs(QlL8p9>9gl3;IA*pS_!!zn>;#({1^6-P4B;y
zJ}~kd<#6>K<tXPRsgdfm);+<5z2JMd;dx?AV(fkJkPp0ttUZ4}n+-NGr#S!eNbIbC
zP5CHWJKNHaA4cI9fl-yS0ZS1p&|?DT7;evam+rDMFkc#zfx7UmJyW?1bG7pF+1#SH
zy|f~y-u+-7bz$>8-CUh|y5Wp1ByZE*pzjmEUH~4)ms+=lAsZ(gBR+|isZ2a;|Evd|
z7Ig9WjyYlQ>*ZT|8BXjz>TiYO<kA#LI1}QSzX`ee9xK)uynCmivLsdOIJb!-M`%J}
z{vAo;?hu!@<P-JI)imV!B6@yKT#o8*AIHfr_>pi{aWc1|fghH<^8>u}`hHEhyf&dG
zW4_3puGs3ATqfE~%Ff?OIP<*o<uzNA!xPL-el}{rQzz?xYOns5tZ#<A;?A=r?g+ts
zuD%vwgz8uGMNf9cRLgPcul`U_?xtK2chScu%h{QZOz!b;0aH_hMv8s=_T9+m8~9MQ
zoUOQG(EVyTZp-OYkOVpN1Nh5<JA<MIbRQag(l$IaB)V+1;cIb40GsXEKYL=c(-Kx{
z0c>^V+728Pph5Ui$BT53{R`p;fJJoxsz7k#M#zn>v!>P2Z~<V5IL2HgK6F-B3V2XP
zDcmnLzUz^Vvf}AJcOD8@rF(O~X68o8`v@=sD^rWHjeHK21cxUHTGgNImXZdqJVx<3
zIL|4AHPzisuVo7!-W1M04NvS6U<VI{6u^u1A+sQ#5Cp2FmuTwnT>ya)2zJzl3NcAZ
z74S+y09-zLti4WH<O0e&C{`9<+<3X7YavQ~eSHovX>uk9mGDh_1{?8#6?q;skjpc<
zZ7?uQj+UF7i=PI9E&{#;-#AWeM1iez;j0*BNlYOhJvrNn10k3Y^1Nf9@k5^-cl&B~
z?<_<ncmit3eyeX>fo{s}p*0rESKk2WeydfWDJ=X0z0SWa&zR{U%Yb}VD5ESb0XB=>
z;?PA$7*&h17&R#p;Kbb)(Zj7WWiiU{b_yDqd(YH(QK2kU7vu)oh3+xE6R%FO*F8Fh
zbpqaZtQ}VV^6BzuXq6%@Be^wsuDpDFWFg0}5|m8_9m&T)hV!Fp=?PbDUe3?E%fPIJ
zY7WRd5idKB1E@dPSOZ_I)+7J3X@LnUOg2IicsD?02^}2AlVA~l#mYZ_)d!^u_|s;O
z0K913L(CIunVkz23!dNVsJDd&@)sDeD8Z-#SL3L_Fz!kSaE;+bmkhcAUl{zRTUnRF
zo0t~rCLKgcaCvuEweUS4y@oEE>I&2h$echniCGTe07`&_blnDbTsUXDV)FGL>xJ(2
z)_wh@S_*;L;+Pv6sD$vixRo%=j4@H99_!4RGYlp(4+u$)N=L{l$GErvWd4GY9etAr
zB{&@!ql5UrLplh!LFT&*7A<t=#dyA7KSE0CqUWXY5Cnzbr=6YYILx4*wB6DaOMw|@
z_dI(=hmA_YR?%oQ`IYryL(Ss{*#e54nV%j?Sv|mT<aEU7dzZ$iG`#mtV9Su>{?s78
z>)8sLG8fKW#KF6Uf+B~)je_T`w$J^j_no13V-s+QH70#oMJgxQrP5rZBBOEP(xrJE
z{HHIj;cw-ra6GPxaA>Q1?E#P!q=|ZvtL1jMxk{Re9;oTa|Ms&JJ|%ZEXd>9mp0f8)
zL)PtNI3x!C9-oLdG})kFF5_@2FiejeC^OGaWuMmW%AXMF3=&htWV39nbH_vic({3H
z3g_4l5bkC_0YJqW_h7Q)z2|!YouPP7fKzgmT7P8so34dPBr*vuP@i&;$8d?2mDSxx
z6RN(lo9P@W0V;1)hDt!F?EZnuTgeNPXC>M{Il7xf5vI4;=AU_SZ<o_pz_VdkyyBN*
zLC`be6dH+Pwxpz!IjAIIpXWq=)rIhQj&V(68S6_@(*{ZsGu7D#xLAj=ce5@4Z<!H{
zK&vN({+SveKb~X@sGI^?#)*nBHGrZZ7q$JqUrJtDL<Io@1gEcd*|S*Lc!R!<xrTdo
z#VdPYd51rM@T|V;PaqJU!5%(j1Fo=iiqpehN+#yZvv&FP6VaSxEh7h|DYFcYFmFeb
z|K^<_Z?H}}jSPCIV3mbL3kWcw1pV?}-)<VtoGWb=vrSFA?e~4(Vh1h1j`t_v561>A
z@9S6<A+5FxvcgZIGM0&dwUD9*!;HVm9;l1U0TFrgmFY|4Z~4Vs*s9T>&5=Bqc(#v>
z&veLd1T>*b*Z20{&7=V4JOY5<n0}ewA0!6#hJ<r~%eTS`)cfHFk(W!6xBX)Fl?2*e
z%N1w_esH_n6>m6pzb6%;J(;Bc0=nA-ds4GRx;42ypoV?ejVG20)ZZZV{d`oo1@yik
zWX>3;H9pj6IZSKc&C@MsoI9|S?n#?$XS&S$AlO7D{|dBu0uGyXGN=&k+po!9Jax!s
zB-C(Wf|J9g;!4YB7h>pwfVeukS=^`=T+xB_<|%vh!!(t4bvv(Hw}DFH2f^x}hX%ZW
zvW<DEKTK2`<qGo82D9Fl7*p%}%EpgFz$km<wJwN3IX69M5FLG4r|r`d0rMF+@B|L(
z<P7XeMgc3-4L)X&>6w4-^0IfJ2xYQ)X*dVjG}-kg#bV^+qa5F+@yU^V{-!2&W%(A6
zh6V*65*vpu<nsIBf_KCJ10uKqB7BjoxdLTe1{?;5ctX2&S4-szD;A|xc95`~$pD&Q
zN{O^rVJar+TTizUW|4^UFMPH5*!t?3GMV!1x5Is)J>zfo@ircbGb7N6IiqT2qmD5w
zmw6la$c73mz2%!AZ*6m8xm-M{Y&m#TdCW(~4tO@Xu6nZutIcB(bJ+DqrBPY|#=!#J
z5%Yb_tgWqE8;5NyLB+7FX!N+yFxX|wm$<=({g1xM9;PI{3~?b`5h1%UpBk(i>Ee5O
zxq~(w1#uIrz7Xfz<=N?1(Nhpn5kt6TrKU301yHQp4a}=VR}3l6hZJVabdY5%mr)$g
zyl*+1q4MYo^s)wQXo9^02bAnWG!Ys`zn$tAM~AlYRXBh1RY9oes)b6BK^WQsj$@1k
zoeLJuH<0^31Z{ZJ8n*`MQY3o`ZGJ#g#?E#wN$k-J2AMZLUwxAl%)UrYVx1j~Zy#xH
zNHK!C0OdS9#78oeF_k$o7Pxc_s-$LI@$$5e2Ixa7ia9r!WYB1M)`=dwUw{wDPHoI<
zAVtJNi9(v+0O6o)fIcdAAASwh3RLVn{az95gd;d*r8?-~16|_%Lq-H?HZP!CHFX?A
zF<VA$SI|8KK8%)JQ*qXq8ma(u#Ob~zulhA3sMCstAXwVk_ozglL%O_2F@QahZX6B8
z7WZ%?=7tooK|K>&bkjy-)A4jMANCXJJs0yGK<_0fO1W1bLv>*>d>!fthu8#?PHK3N
z49bm(F`(LB5P6BgP`=Q#<oexue+PP^3!9yZtx6unr4Z-XL*B)x;KINHfcOR-wm5Mg
z0+=%x!O?}KCLkR;fGTBY1l&6godC*;z%P}fc+Ei|0z4IP`Zt$P6yi7l@p(CHkecY(
zgQ$961IpekTm~gDSe_(`=T-#Tn3+}hDwCVk-_9q)cF(gyxnW(956-uhZr8y+WmN|J
zlMYs_aM%?ZmJi1F^`+Z;=usIHm`*<&eFkE)_1_o4hI*&&4i-WQ@Br%m%bP(B=MU7U
zf~wV6$N{S&L(y`y$p*-$_~wSAU~D*rM>bQ^{7FTt@6@fMMYcH7(aQeZRZkAnIsluH
zmc(Y75>%2phy}`k33RDnWPAsYI5Y$35yo^O1+3Hf2hbK7)CK$?&K`e6tt6y6;x6Dl
z)eCU`XlgJ)P%T)A+`z2V8Gsq*iy&DmCX5@h)iRC)<PKN9@B_D3s*Hfwoi7$cEzEjI
z_bgbgeNUiu#K}kf3j?scMyoSCpvVSR7*JeH4ZsL$1*Nx$78TODW|maVrdzWdfcmme
zF-pQP>iT+g7D=z0Hprp2v=B(7Q5qoSJ<1?-03l008Uq(_u6KuU(uVn|ubZh1J4^o+
zq%MvCP>%{+gX$OdC<7>_s$KlRS|}U<AAc5KuKD9g4NkT(eUi5UAF#>;mMA}#2=EB3
z&?yT?slY4HalIVSYxV0N#wm>fq65GEhhnjiSN-hVtCqDeIoNxufCqp9#`;_H4Q`4_
zC0uUt_X7Y*tec-5;^M<3=hl6-i=h!00L#_y@<6x(AjL&6*a`Rph-<eY{r~h1fmYeR
zUEqBMDhQGjz@!w~gZtOb@itHZ`vd$T$De@TCV7M)YGSzWyx!QR7zTnaej})xzP27S
zTfqr;1cOQ3M*+L}KOiLvV_6W^f5R0h5@0ZQ7}aL&d;VhWs)&7h|HT3Uqq-jh5)hB#
z<5no$4%*kfgUPP7_rDnmfd|ydtAEh9*$;Ug4C@8YD*0rC&Jq7NT!Hoqcp@lKg;@WB
zaGS6tB+C8-{OT%x|KWe4Y*W(z4r%`rWjp4*|B13qcJXhe@&82GjxD19iL(EXMH!rY
zwKE99MF6G$IO`{&<L5c3md_f-C^yE-iThsZ5JEceo-#%Cgq4Tuva8`<AUwjRX~Jr%
zDv|^`<KR55DoqQh2>=`b9cIRzeja9+P}BF<O5{SFu<*p-{3_{Zq>h2lx+j7Hn^(T3
zn9&t+o*V1ZcdFrd9u|;o=Rz!_H8^n->gK&;v9$>c&bwt_!2NHml|8Ux;E6435TyHW
ztO5#PuN?oMy|N?X{O{<sQ-=B9(Q8Kx2EF{hqt}iF-2aYVJJweJ6J`GsWt)hm{~f({
zYWV%{=(UO6`#XmSFxmf)cZkd5B=<*ILS1j<xi9>w^~z&L)gg{kVIAPgkf5i1mA}N9
zKy`)Kkw^UVC+8=kZLvT{Oj+63o4q4T&?Q!6q7P};*E;R#rD_EO(#fW|%W9(?h&~iu
zaO4RRMB&h_#U@}hQ|&Q0H;rtTKCrN$H*Fhh`zjhq|JC1j-2C`L)2qwREMDD=eTjCY
zG#<pyeW4fkT#RPyWYjq;J#-hHQT78XxHqQ!8o+z@sxL=^qP<ML1GxTiU<dih3E5DT
zfq{*Y-ZY5j;J)WJQyrXXE$z*})X!^qD237b>0O_>o<7%$U1->b(F(xr8N?0)hbX10
zTjtHz2IQ}UE3?nXe#}QAOl)`np^VFFfD}%sYD(|xM)nWtqTE3DP8%#9=6><Ilu4Ld
zJtH9PG$|(gJ)Po$<r5#cJmf%hxYW3CHsM8OOwK~#51^&bw4~7FE%U2T4Aw;WoISYx
zRqYwh{*ZeRsmXnn1Ue~R0L5byT1~UkPAXbYoNk>&4j(XWUg-W{G0}I`{BkBMhfa0G
z2T&*BktR=j0VwYD$o?gprp6f#H=!CuBfaDfkH7=g2)+6dui7_afn?wx$2_NTU%}GN
zvojAAR8(w*2)ugwec9)2JIqz8GM^lNmPgCs(D6|0xGiz+>_or7jl|AopVK-y@1))@
zbsJQbElK4%Fi1xn0RN86DsZQlT^wEVR$sI+vnmLkpZLZeP$e1T2LFySZku2%zf7Qr
zNHt3|JcSlvl~f8)H>ZY^@KlO4DT&|7`J`SV)B5<{+?s^Kgu%-2!sh3!x4t|NW|<Qu
zo8>p3wjJe}AwJSWzKA;lB4t2@CGy4oe&Uo#f37?K4{2{6PxZq6kCwy9kdT=SB_t%t
zJeEWQN*OjGN!zT<^KPG0RHh<i$dDwNVq<3(A>)>?%`#7!=lQPBR?qW%pYOT9dtdkd
z(fQ-Nu=i(u*6?0yz1Lb5&3X1C(Rjy_BKt+j*=Why<=N;&o1S5Z<<Z?*m-{)2j-m>V
zT(>;pews4tW!_>QZS0neb?w@eicI;&IE2aw^et`w!cA6k{P|KCLU3AZ$B+7#ZH<&R
zqD%{yz71^B#ke`Ihvh7()kXEuH~PHge0uJoiZ7YPM9L!L%QMh3J$8Ghy~i#of6}w~
zMc_7hp>}!6OqZNc{jxWz4rNriLVqGt1vO?D#vvKkmT$A|o$*wW-1Aa-j~(02!RuMx
zgj5k9dsJQL6(6<0Vy%u(57hD<c&cS$&i-~j#>by~?_Y&;z`CD){N?bdXr@s9a?>H%
z;d0m*mUYN$IuHGF={T%(gjOls08w9C!S70JZ4k@2OfK+Vj1CjfJgDu})lVzde*X)D
zKdoqAb6*vQ(j4}Cr5MJ;CGgE~L^WsMT-RJQvG^E%Nt)kQq5JrNGS-oDeD~z?t-bfw
znvN!~<A;NpI}<RG7|HUkd2^;nTV^a)J2|4pUGHcML0?3HuQ`AW;S$|){zg{;tFkIT
zI#0AjA3MIPomsq4Be_{*r;ge<Hgd$PUSd_;RbKrt5MvWWww+xW#Hr{{e0W&U-PD>6
z|M?^@jD17%iD;ZUkaLR=nH(D)y&S0GL0%{45Yzf`KE0)Bd6AihK)Lh=ex*9EJ;=$&
zV#29jr>Qz4tK+Co)v34^@FXPII7zbwR!$YayXB!<c27_Fo{bk~F~&x9!L9k+fk!vA
zJTyC;2#SVRRBXY(E-)dq!Ag&@*UGsS^DYcGxGq_5Z_vWtF^u$<TRSSAHp{Cw4cR<&
zr`wm=5&Fg7>Y3y}cj6XJb1U3girU|*Cu~wIVfb8blBD)?M?db!;jh}a>s|`2^Y>K=
zaS|?`R9Eh=3cA@2eGu647GD9Q(Is=Zcw+{b`XioRDK8)PGwVHkEuvIIwuU&^;1ft7
zd?TlPOY<^0UMrIR?|s-~G~irpr=yFe)5~y-AT_UEK1rj9sr8(fxc6E5BgdW#Tx<8}
z{!-;G)grgFXq9z%fV@y2l+Cl(x%-wWYNIFTr9=q@i?kmwvW)PTe;=P?FaXYv%xcdr
zYw6<a-4Be0osu##d_VKMGtK&ndn;VK?HFNMVB*KypZ`$B&{V!r3sB8@hQif2V7!b!
z?QHZvzM<`7?7m)adRXA<t7ehH(}+NsH8$R7`Al~Y)SH&_9;Xo+-qt96GFK($%YR&&
zL22#t>Z8|ged4n>4rx+1eBC}Wr1-fZYF)JXwgRWav`n1JEcN(q-2=s}_}O0K`&dRx
z-2_XXrP_@Ft|63cHBQfzZjf=@6+6r^9a*-e3m*IpT-3wgt(zIH?JZJKMP<r{;|f!w
zH{Y*(gdN^v6m>1yZYG9@fJg8a6h~`PzsYVU|IPsWezJM;o~ctcCHAvMgC+OgUPy?c
zAnDrE`Sc@0DCx4D;h=H@@=%+zPWX~l#gv>|Mxd1D9@8w-Y``SQxESugS5UJ&0b{Ck
z{@=AW4JzpfTEK!ig4ud7U?Jx9pyOUljpSAIK*fOy{j!b5xg?%#=kWl-6le97QYN37
zf-&0o+3i`3zJ}7;I8+I-#S}XmMD+@|!pCl3JpZm(TQ_V6PD-)BmU2y4sP9{O9}_(`
zcs5Lst~Iu-+Z95caHI%3!O<-I1U8PLd6#+D(?q#nLQI{VpI!|VaAtdO#(;VaIq3D>
ztQt2Mf{x<&OM~3=!tL5nejr|9nBc_Ffq?=4-7cNX1AlH9J?qc>YEdwHv*`tk#Y~cM
zmdMyql>}XE&eHdc!Y4p3$|i4sdqX~;hjE3^8OhWR-_I&qTP-Yc5NZ+DLs6b(Juy{m
z_vVg*@Q!<X@iGm6vk^p5uT?KCi-P=9S7g1?#=eT-8KvxRWs8>$C;x8Kj~{~VR|aR7
z9uMIC?=BSe4S_tVVew$98Mx+5)?04<pm@mW*?MC5YGx*9=}1$r8^#-PL<r5quAm#K
ziC$p-wA!~>A>Y5g^L$DBweJz1C5cJ^BYizN4|di~X-qd-9v39@@C_^ycF%YCta^xt
zcE$S5+e$nZB)=@2uq(W>T<lw#D`rO_mR*im=>3(>F_!Z`*a<RcyqA)A8g8--p|Wdm
zSA$fJlNq1-_+Y3S3E~Gu8}3_un*3lYZTudOy6)QNkun$L;9Y=vneQJy5`9BfJvyQw
zB){xS?DKutKqP=jI6i?LC+8?Yl=@X5%FK=DoMdVYAj%GsDC>lEWhw76l0<2)AiUxp
zG0n`&0(`R5Zrz*xPABX0Tp)1tDqi$*lAhq>Sua=b^16)vUEs~t|M5aS;t?!wbVEqW
z>9P+N>aVB7?_lrO2?_e6aE6=7i|i8j9u3zY0DxwR@La81`W)4BboyVr?FcC>MyWpr
zyk^fnjI5A(L)qiKC^|6hfYF92*bV7`q}aEQFU*u~#x?i%v#xYI^s@xlaW4a;<J@_3
zVTG@YXAx2SSF%p0yPgK2zn$Q}kdOl2v8_%2?Ucv1w@Yu=v1=^YZig}nShSX>Xxf7u
z;2e!g@&EjpJ*w`GLfXBPE$({8XN<@_wRurLU><ex&HZZjUG7oBuJ77n{48zl0pIEk
z#Ihd4#lCfOMTLzAB$cHP6#ZtZbojUm?;91=#>A7maK$y@03v16<fx+QQn?p=WNLu!
ziV9f*oRup!BPI*;)DjMoS-i$;PPV)2^;QI&jnj>6V8Qk=o^~?|1SmJsRVXs@vn1|#
z{}H2vqo$^*zgQAnEC!`p4g>K!6TJgsm;$_jVC{LT+QEah$7sxb5D<FA@#{I&vV{Co
zJ<i{Gq6KibKUcu;cnb`eACL|bd=8EygSy9kNE5lkXKGO3Xp!bW%CYOmX;1C-bN<W6
z`cv|1%Gzg+E_C;suqyVxq7pDIf2CNISB2qt-S6J6QSD1M!U#xgvRH$-xUqh@#PcF>
z=wOCfe{DG{YhwKlAQ>_HiJfQj={h579E6j}3xO&QROu!qawHHLV7hf#W-6U<P?XZc
z+#ui9<K=nVvTfIab%L_MWy2xOw<5&YI(9$Sr&6opefohzRP<l3ATis-7jDrfNFz=F
ztq3x0@_ZpoR&wfxiCglY0nS<7n-Qe1a&=Fp9qxKo6aW77S4E=j5Q;K0AZGVXYIcPg
zDfjnqqtQS^L>#Hy@%tMSXawgJ&2++UxJbXGV7WK;25IuzvLnOy#}f}EbGrfxIk<i1
zo*O1nCXfCYU}_h#+DOXwbkuj9(zvC!iWT3~l_g@`2j>UC1dZKiz52zMR-5imrV9E4
zl6qWen%U(!RUn!yE9baC6UMVWUgM9kb9C--#~=9f&+3>9Gc|Za!NxOp6>Q$|1*#MP
zhz;L3)q?m1ovb*96EP|aQ=BC`uN1vp-Ivo$`*g$u(af30g3#eQ9v1-xgsPp&`ZO3h
z02^k@RB28WANQX7n}=54*k!r-d@(1-J>L+;IQqVYh#QqL{ab#14=x^XzJbw?u+bVh
z__^cEM!03nKCJ!W6IYabHAffDKUCQS?al93sAq5Ny?6bv|30i9Md3uBc2e=wE=TRz
zzFD6u=A1{#pNSx`|E(5$jfm+TPlWk6%>){pVf?oAy$G@0LI~}0B`=(E$?SK9h)Fl0
z^A4r_&yPMb6L0k|if(|FDbxDQ&C{AyJ!97VzTJS{;e&+4^ygDfLrt1v;JGpx+<#!z
zQavsS(33zYXfoZl@`;Am@`B5+?qVMsIh2SZMB;&*`crfW9$bF)jUiNNzGQa$J^)y;
zN3~&sA&0mZT_Y}Gf!qD)ZBHJC(SMbH;1YY7`U2*=X-)~;3{DiYNZ*YwqgCw=AG_op
zQ8U?=AsYSdgys;6kA0Ks<qefxX`NZM!Z;!!axpKxm0XHyc==p>X=&;6EzRh&57zW^
zrC&Rw?}piQU*s9|uH@9B{+^=kAY?VvNvZ#$8W+Lw@GH682?Q`aa`3U)nLAb(w2>!@
zi(E>Ajkv>eWRYI8+ix4s_3x};?iFgE@o;n#IgR@Desb0=CSkdh@ri2@zJ)Xs7EkwK
zr2v*?(Am%AmtN^IiC4SKcvvd|M>gXUvc@{$`)F14wU4H(JXVYPrAwcbxO7}lZZYV}
zYj06mX{;U?QM)yBs@R@8M|B%cUqaYZT)!#}kC_ZU&0*&X2tJ8soruc#m7X}VTA3Z<
z)ze;C7U$ZILQ5++9KOH1O;agdCaa7rtF+|M<^@a2p%yXOy8Zn<JkkZ3<yPWjfLqyC
z#ow^C{&!b8(OCm!z|ajx)(~I*H2#Li)}-~e^VY>~80|U;3v&qmzAqMDUA%7CXPQM9
zFJ))=X>HKK!Q}bTS`2pxEUxey7F!QBTu@#=ufCvw3cOSGEc_^{Rua{7blswHUMUbg
z#O$nq9i$rHy*QbB0&!1+B^Se=sHBnAp1-}!{l_2M1Lc3L(uH8Qia++ccKdbW(2%hn
zsBBHW;lCR?r%Cf7NaV|DE6a0!!&zc=e;rI(goPA?iDH8BHy2HA11ha{e(j*Z$#AO9
zU8=C8L!h~l3@-ShD|+2_WhDbHVZ453*F1_W9#*T|o*Vz;USo9l1FWBWfNpS;YWsA~
zEvJ)X8Ud&C@A_CgQGcjXgup|5ciTF^L#_Qzc)sO3Gg4ZcOQ^Rp!8%8uM<O^oN3DQo
zz~_od7SqJxaawFnQ@22#xdNwEk{Ua~bBI|mlKdj_F7qF~SyJ1BIW|Tg6l36Je!NtH
z%$33$O}K{aGkKaQ_BHc^(YtAU|0&?}Cc0Es7cH(6F!nOD#fz0mAxiCMQvo;|A-kt(
zU;rn&ni1i`ijW2QKDJo-**Rd&vw7crA#7Ja;%;>HC0~YB2p=}?O-^ylTJqE8N2mW*
zOJa-xL@e&}z{XiFCO=H<ZIEw6l!TXM4)3m*($eaxKfJwb*Zb&(P<<6Gg3VrJ8)fv*
zoDr~HLQFWq_(b%hz40^|iDbM6hodYzT!_YZ;_VwIjKQ`g10;?=Em_V;IV8sP&AOr*
zcP9v)(O$MjE+_QdJ3G0PjMm6O(art3-%#*ozZqmow7BTy$lRQ*$(xm77pPCvOQtw-
z7#%CQdtCLQiWUNeoCd>0U@}AC_KfS>iiX~5n@>ssOaz_FK*@(pFnyDu5C6sYu|^04
zFT{pY`2m4d<N7>DYVA6~H|pW{#RsIXu+{4o%_h3jiANZoY!_z?&^?m;E2vODnu8F^
z<GZ^V!>NOi(?=I<LTjnQOh0(H<ASZ-XEgW2a51&=#y5)?scm3;m%pNbQAT!dFI!aO
zypf%xe5}t(nK~-gtp3X?n9kgU_O8t<33QsMTZ&Q{QZb{I-cph@X{TH+{qeC)FJnY#
zsz8pJ;NC9eF~;BAOA@amBHkHdbPt1NBr^73t+l*CjBnKqkPtE5!QUTv+@D@M2J~rq
zWIQf<8|BJIcqS2Nk?Qb((mAX=zuVqOT7GeKnx$`d2Xm-@z3K4)bCyi@#Nfe8WMKjp
zG54~F{rolfsq2WJI%0KQ%_FNdljy$)(NPIO?$eAe0%YWcD98qrT|s6d)te3I7uCY!
z?qus!twpOzy8&apO*@3m2=`?HezKC2q-atahz8`o>>}J>B6fQwzwl7N86rfgDM&aA
z!@D>{pHGMJC57btUSc$U2TOBm&9kyc6VQ*sYJs}+!4QVI*f9n98)tTxO<9TC%e7Yz
z1i&^c>@I^rvHJt#06Om<y^v)ZPx%G8je-8MDJ`dl01W&b5gH9T95E;zpzz?%ba&_7
z>(7oU2BOtIXj&Yl*H+opk42DVG4b(B%B7rgs8QMElmTLD`#Cin44c$18sw*YUjH;4
zPzxypXHBf=5b8lSPF$Moz~S-d<tyF#V$C06!t4=vprnDnXiDi}smH7;27`GQ%fA|h
zQxMSR{(4vxFS7?K8+?XW&V&MSk%1gsg?d6L`F=7AdWf#67o70S0c-?s79qbkXDPqH
zTu6}&g5RI+ZahDRH6h`59c{?E0?{Qag<+<U2?^UJXh7M5lnV)GwBkI;^7<`PS?iVd
zyu?G55X!9?H)+2Zhtgy-`_(QR!t>SRbA4XQLO+WagJ~B(3f=zv*W;ro$wp)*Vu-;Y
zcl>j7<!Z+e)j-1E)hk!>C7()m%&*lDXn1z!yf`1V_iv|}ZVSNnq&FK_hLzFYGWD-_
zMvn)gqf##%SGldBvdfBR&mDrKkq6)O_F_UL#iUO&+Mg?bHu{+8tu>Xk8MiWh*pU+K
z<=bJ0H@030LHp0!DtPAA<eMv`DW)ls7n<IN9F2P?VPI!-*LbaGtciLob}<!zE@#5|
z)!(}S4$xr=O2T{4#tDgiSz39Har0k95oZvbd)SF*$RkfCVjorkam8a+C+uK8;tHax
zr9D%r24CDqO|%cHS<zdDJlqk>bdujthIG9>s+Dr-!$Xhm$!Li<U9Uj5G_Ije_eId?
z*T}26E>`y&XkMmSPJaPCdyd);m~BV;uuE^t;VSp*ck4<^&40)E)lfGSI&Q->UYbG5
zOV~5X&-@<$<MU$NYHYsc5%0PE1LH~YQR}xjrR@*G7Wx~bYZ(_~#mz6K`8(Tr521*9
zL4ge8m-la*ZduoEB62KidaIG3PXUD;{qSXWiFb%XfAD$T1EMKqn;e;U$?|epvSlZ{
z3G*wi!L7+g9K&u=z{22<_d6~6>Ebt9%lu$m)~!LXp-`oAxHhcDJ9wVkV9&ILMwO~J
zpFMy%1rI7H4K>wk7sB(zL0RMS>jL79>?g|&C2<C!Ek~bcfv`#)`nKa&MdzokJ|p+M
zp3oO-iIgKZB;u+Zsysr_^G+gaWa|ikVPmz(%-X|=$&WqTHGp0lboVE++7smEsfE-E
zj-sBXck->Dc_7b?D*Na?omXa~?ive)DKZ)aYv@i0XhyTeEvF}aSQ#mp&&QVt<u2Gf
zUsig_B#fzz?0hzNI*vJ!B^;~b1-amS$zm}exSA|6p~K@A*l#>boiJ@z^iD>)XKoX&
z73dR($0V~nEC6QfHSWyMoQ@09|G2WIP=o8jb*d(~k#`kXo)=!ddWSj!$*+Yos(aK)
z<35GzNRqPpq#j^eFE4m49h#Ztn<94<ow5WY-a1nbLzW*f8-v~*^SQFM5ei1ieA>at
z*r^(49l~bh5sVgJ5Tqr0efYGS{4vu<1*Ll>pVyO&p3}uJx0J^LM!Xl!6U)VXAGmaU
zG*hjd1N8ZEb7qmlujZge>9y=OJ=~43s{&(i6uz|89dZ<vgcs%G6m(FyuPyLKefYK>
zRyxy3@V%JyBYHb&(TLIPi2OTkv1&5fYTSdfWd4Uwq{7Sdts3)yun}Tvs&gEMoZuEl
zR-G>%2=P6pvYS7h(iS3YkFU$aFkkD>knBo#7S(LIAP~@`MCY+<N(jKddvd8<1@q25
zn{TfOVHjKx(vihG!N5yTO65evCV*ac5c|z35&V3aNf4Se?3U0A@U@1OuxNt!Rr2+@
z^g<<z`V~GSQ+5cIryiIAAHlXB_9xZ^Y7sFdD+@mZ>KyHiBOLEJ=#B(iVf`G5Wi6x}
z8GD}U)`Kdsm<Ti!ry*r$?p}sn*n;=u+GwSFUJAu&fE?cop5D+m2$mUa_Wb79>$Wvt
zb59F1P@xtzl+TfUB29{uV3^@CPetFnJM#U{1qAeu-7wYRZ%!lY-MAv6y!JU6xp1Cp
z=Cc^nPrtFbxSo?epYA7L7hGJ8I}sH|i9JjS>+7B0zTB&5jY36>PfS9PUxoK?q(E?x
z=?zHW7@)rJI@hd}I}D3aUW($9$!fejV9!dK9?O_M{4nVg1IR}0(rg;t^`;qYgtO*F
z^mg8=CX9naSGShQ`kiBa+x**L1gD}ZatS0$G9E5w1L!5p!BTLy?+OY7R3Q!=-17Gc
zit*2dY9T}KyE=aX_UWl8V!xGDIIa!&w+2K^mBO-czJ@5A?mH(D!(5pg#n~55R(xoz
z#$~q_n^XRh#V3+1^g$M&8pN!ux}8Jvw)%O!PlHD4cUhX`Xa9=6F^eOZ47^VYe&5Ms
z<3BGFao0u8BM{w&{-~A^ONPQcwpbWm5o^ANc$W0yl?I=iP|8<H@Is=TTgazaKo6ow
z#F88^(XKi+DzWN$$NF*r483iXdEZ(b-z|2iP04(F8C+dMw;nNeKD~>tU5|SRm0o<w
zUquYuie1La!uT)s$ZvmB$JI?nJCcb>-t#Mu8P+O!>|w&w*3vM}6{8_ORp+=0MZNFd
zG%$6n!Ut1(`u_Y^BJ2<Kxe#AhLLs`yqv*BvO0Qe|)f6vez6?^4d(XP9-$DT}-&2EF
zB^~Et9R8Ia8;8i~hNj0Y!qnJes&FdeV(hp0BcSSMs3E|lu8bs8>a))Z?Gys)3S_et
zs&S1^y&&2CpaK}cI<|wM&ap8yHax<7CW%xg?&U)gKCdq$k03cm!izEDRlL4*IsWo0
zf(g6zHBuo6g7Tic{DIt4x4hWSrIDTRLGg~=h<IF8U5qp3%X8&D(;A{<-&psmN+PNl
z`m{h6%P0IoT)&)35*cP|$S!rP-ztR*)))jrS5N!v%UHOaDgN}rLlrJ8QqJr0f*!_A
zJG_&5R>?|(b72st^t)5<!-Nv$UI=K%-9kG!OrMEc*7K<q#o9wfc;p7-|BytN>7CBf
zq^bg_lvZx7v$hQ*GleuLR*N)QnV~~NsEhAelHe=ny`uO=#iYYKmGu!!D7R#bWI~s}
zpIN=0Jr@YmvkXX)LUGU$<R(oCF(i{>)f0K#0wUuT9Xav^s7{TT$EYKoR>_557f|Kx
zzWl;bP(beN`~^0`oe@nrWDc^&&L%HX-Sp-jvdm-Ao#}wVR<~_oMI=_!p9n1H*>zX9
zM&<W;+FQrIV#x3JEjkS?3dvfv0Pcoq;_waaQ*bw3QS>7ui66FjoSS!ayQh0z$-8O2
z`_7*hGU~GETleyq2y_?fTc3luC0qsoS;b(-7c%8BYxG7p<Pjy)#oP6!ccNq--01*4
zlQYbR%M(4PH!9VUCYpoi_bDer<h%&aAZUuW<4IkzDyH!-Ss!1LVow&k6A#jrq95l*
zuy!xsDM9Gj{tY2{(hvrP!6+iU6tj;ftMV!%P9@)CiR4re(@i5em8LzX5&}-851h(D
zl2efl`R!ELH~u!TfQ3%~=~SR(+kJze>}K!9YN>H%>5HJ!QAtx|k1Y+W(sK@Nau8e~
zO9v=X<!0Y%QzK?`dUg2Q^{#wH8JSlTw<PXeB`}GJwp0h@i7yzgC9YgupB9fBK?_?z
z4Z3+Z4=-SD3Z~O-glW2%z)B84rc5@b+jj9&Gm(h*DC*VDEEw&&-zZ6Ji(Tn3k7sr5
zv`C5G9_lnx;0!UDs;TBC7!t62lrQyfLwi4;wG=5OqO1%@l}@t6G4M7kB57Zd`YkOG
z*`MHh+wvezckzvam=NE4+Z!pL5VG2m%vV8DxrToXGM>mMHBt>F<MQx!HQa=*7%wzm
zb2s9hmRXB_7f`H(;*zPHXTp4~WEp?2=>B@*xrjZG5*FTNhnXhELf-c7!2;z~y~z57
zPqU>IW<+DR+QYg?o86WLnfO;-Akv1U3kwjUBm^(Js_%Yq2pxYZbmPt2-qZ_$=+~ic
zCS)c23A;89K2->U!Q*FQgH*C873QB7`H$4oaW&1@7rD%osw@PYM#pcjbL6MqNO=^d
z?d$R{7?U?-9*oX-je^eOMpyBi7O7pe(DhGtg(uk+6^Mvi9N#nbG*w8HY>4rGmY270
z1~(UOcd;~C*JqxJ$uuE-G5=B{2cfIG6#n>hV*!p~+y@D3a)F({yiI!UVL}cvgB1`e
zLqaeQRrG0kFuQ-XZrSCv{e13yExR|QzzHWx{Qx%5;N@+a;J7aKJz0oizQ_&F+6xy0
z)weHUYsEun3^=NB!QbS?v2lkHWhuO4LQ)ntP!@f1lZ8;$$HQ6QL)I6?sF!Cni+@^(
z)mnVMGKBK7Q5bb^?~9e^gb8}+<U80L-E*_|AS(a2o83>A2QSw~)3DcE!AAG0H`xdi
zBv&QN+z2WuEpZvMCHFi>dg@*_c<S^2=cmryHF}0}<o~@xK7j%YyZuwD`=Ah<`e5P7
z^X14v+^zJn7s^&J{>&Ru4t1Cap;W~c_w(=EtzN)w2qQ*|+X;A9%hPdfH&eODDO<rz
z+RoKJ&~HHU&|qeMD<FsPBg<joUX*u+K+rHvyqH&KfXzxxhnYe^;w#0VS3{^3+_Os$
zRgN{!(5?S+9b*6BI_TdsLxizunS@Z9b_aAn1xirU8LO{W3zrwCGvuMo!yBU_IBFLb
z%TCB?(}o48H8=y0&_mc9L|s~)8-9C^<Z$;0wFI#%#~=J=6!ePBmH#n8bM|VyQVE|L
zQpLDB<^x+O&Qviz1*qS966xGuMRhw71n)hIWU*?DmejoteO>yE4FTQh`oe4|&?3f!
zbiPh|t)!>>tA(Hy0<=z^BYVt*D1)K-zm(yjiYpdb^o`V$B@e`UFQirGTHW|(iS+mP
zO{^i6IrPZ#|2J)bZrS*kV!Kd@AyodxQ)b~&cNSsomZR5L5G#ld{-0KmyUKVJ<#_yG
z`k;pBLo)XZA}e)^mI5q%$vsgy0E>L<n6!_BXK+p{S6toVm$Z>UAM%4eDZ-K_5tL9g
z%6t|v@~#uA50y|}Wk%Yy17(}Dru3nvA)UE0s-Zsses_3IWmHOaE7&jgBLxmf^O_-e
z8Mrica8~0w-n8&ty9!akYf5qyX)%x#8y@A;a(j10a^%iifdznaY;bPHLvqu6`gB}X
zXhVcn!@Nj0d(|<O+hOHzFPR%%b^Jb`u_K0L(|U$}1tHHKqv1P;AOLrUA(9aupp_aF
zL(<A6WIkP2d>)GEKvY?2_iFI!-B@2~e~n|{eSWGr)ua13x*7)yiP`x;dJ?8B$F3pn
zIo$BSr1xJw<Sy!|ZC9=pmyl|E_=sLcj*6=K$*-{O<cR?Az<Q#i+Xwu4I5sv0y~WK&
z>cVx$rlzK{y80Z=p^PyAJrMh~3GJ-+oAu>J)yBAcc1X#4piFc%-le^B0BHgfd}h84
zzyc(;lj>Xe@6|ZbDn&uarylNPy|}8hNm94PRKC>1yr*5UfQ0<CeI}h;7~O|BO?jrf
zO8Te*(aSd%x>=nR>ct{;1g>30GNb&w+k0pR_YYooAJ!0_I!_>D20+kB0VS*>^(SI)
zpJ#|i7PPDS3Z10L$fIeT9s8teqvq2+w!a(-l8visidmv$GDeCPe&#E`ion7dUvMI1
z^JUjVN74s>B?RQhs1??JsHUd6OzOWKZHNks#viYE#@uFlw#oT9k`j#{#=hlgYGG9#
zVpqRVKR<p6yR*}U$CC&?)zYPmK&kog{OX;+v254rt~qGqS+6;?r$-H-2y!Q5?^n=N
zo+|ozHtud(aN`|Ig|uV1a|tg2u&?NOkzAG6R8uFcHe>}-7?1yOw8xV9Y)L#*+|E*^
z`fM->4@X5CfPO9CMOtguu_T*Jjw;Z<`QM;#AJ%&Bi^ddXxN??+S2ZuMSxR=3*9k%|
z&}&Lzw8Vbfd+*~cPc_%>Dc9&Sqx@M=E_sI*@M6G@LZ})o_C3NM3|@|)rlk$)bR0*1
z?Hr=!o9i&I*QswwHGOgU)W?*oQHS+14BO6zSPf{>u^?6PLo`TlRmbULqf`jZ=Im~6
zt2>G>(4^yCR@TCbYiEwdfK&YD`A<tN^*I7Wonmz!ACWC0#u)1El|26V(&eB5H8HvT
z<RMNqsW{2d?D&0HQK(HASr6|jAl9g~yW#jJ^ipU<jPez{p;?mYRj#~MDde(R(_^I&
zkQx=I2a?*hy#tcE^h;9VQJe1#+@ZI>N(tqe8SV$N$cmM`hXqDX_!z@SgkyH0bMZhT
zn{gRK8;Om8PcLyu6}08q4`r+IrD8jsdq|*E`V}aV9`RnJ5UuR7t9m0s5Qs=lsQa~w
zhv^3Iy4oUSNhvJ$ldVD!$)25d8oh1ZlBCb0si=a0K<5|buu|wnbPOyjR$5xtyc`rM
zSuyU5D=WjSH4A4IpDdo#b@*!cz)y4$4}FrkT5Vsc;>}*3gK+4g^UV8^F=&|89#Kii
zMN~-NClHBtn^TRjti<xveYm^lpL~TU<Eb-r0{x+6g};aS$%S~eiHRyULJQ#-Z1s4}
zM5_7g;m<hnv~lNyeitK-sp(2RvWbENvP4O@qEwf_(gS#_vR!h@FN|#4a*Lhe?)gDF
z&mJKC<cozXjJ(ZMSS>=gfQ3|CUfHO5AbOgRCvlj!`PrTZ73zY2TRBk~0eg<F)z^V;
zz~818`RKjAa5Xtb&c*tjyT^rS1rJ`yPFTE?T+M(iP3#3^cojjhQjJ}K?B6T1Ym~CA
zV)g!jjpEl`G#vjP+n!Zc^d48;J<RQK*E@j!CV#BikaSsr6f6bjtjxk}cOT52QWuYF
zbIXoIe8#qE#-%>5JEWvDlYauf^Rsd;w=E%r6=!n@$t@74WQtP`0U0UWj7eb!Xyf~5
zBD^KT1QJB<ZGjxzs{=BWMFb+ti-{gUY9vs6mm&v0oIilmKoX5%ey}i;oS^SPB>7aA
zo`?OFSt`;E-2(g|AADaJ4#eLTZ-YLrlEA#Dm@a96DW<RciNZ3h#E9^{rHM=iw>xue
z0S!bm&FXg>jr8Q6am5WGR{=uBSiO+46_Fgg$(R|C;WTkRgHHuvBiZ1om$*A>%N0_X
z?J2&24{4e9(IE<dM4mF{=v$1VgF~3>wXd|GImQ_CH|0wmm0jYFbqGS`a=?_x!X@uz
z3Mu;@wkEl$pHMlQmfRAr+46ZM?Lgc(aIw!uA>d&2(E_$1l;8I!kX9a`AOXa-poSjx
zJbzIj2#6th_+pDA)K;(?NS<`7p*L7&O+;s#__T!l4mfJA8xlr_;xjH@{h^J9BC!0F
zc{~C-<1Zm`n<_M!6tBj4PIN)r59|}k-qN~b3%adC&e)<75)bC3BYR`vvCGL&V&h?^
z*q^O@^_nk_I{*~;d6VTea=zC5u!aCt3(N722Q^$3zz7_C)#BjV?7q=}bhLrMax^4+
zAS_-%zQ<!MKH8$h{?&Z?sQ_3U_X{z4=#y-yDg3wx8TlESTaL9Ml|b*(W5nFP&0D;)
zEZFvwfYfTlI;$tD*$5mdr${A-UpADlFyo*bpEw_potzy+PiDaaDY)^doSrK*cMz(d
zUGJ-+Wz^d)Ju^D&C!~tE+z@IQh_uk|zF-0FKruz3MW2*V%Dl*ISEi<NwYy%-2Sw1o
zuwQ1>7DX~6JyUk-)ICu=gnOhpCe5qMnv-nj7x?`}LSUED-KPQwbSk{~<C(B8J8;+t
znv07~Zl$7E<6<*-Y@?gsmop3N1$N!QIKjb&BM-hL4m?*qbPz7=Y9d)Vz)mfV>c_*X
zmyz7!cv-^2m~uZSFrwWCHl{GFET?YWRL(x=MNCI=0#y)01*|Puq?O1X%=Yw-Lj6_#
zW7rm&@!hU|a+0xin>sXUz{!zv^AJTArQLP0wm%O}%g%opsvP3RyYzlkaOEH@T(gV2
z7dW53&i)9zhSe{dxFWH7ecCKwAGQghS<V#a9zkBd9tbkyX+bx=H%n^1afd4JJHX%K
zlDK5%Art(nXlFVgFJ5J=eKcP4vXw%8Rwifi5wPT!`P(2#|6h_U*^}gdIi%+<ZB3WL
z<X{nIyY8tlQ)grk86=NzO;ica53FnAHh-ddn31KRnnQt-_ebh<HR%+$=2T%jp}Vbn
z*Y~(HxooQ2_|i*d@;2=a17*f;z5EK>6aI859z^5SM{uMvW?)Qtw`%U#Ax-U@f^fj$
zKm|K`YqPmueG)oB77A&zcVMel9(Z%Zu;SG(q#~m6l@>u2-hlFgu{c(4pQY^R_Xy0@
zFI`BM++6H2Q>edz_w53^*%0c!@=yi;%WjNf3qq{vjNp+?kc#_Q@(AK|+rTE%6!Uw|
zxJ}_SjPef<sJ^h=&yvr#^z_`b$r%xd3G39(-yziL453Pwjm4J1-b<XP_bQJ_-L1{8
zsmCr)ojqS5BJuezXHLna&l`2OY2lSjwo{}cFbGgNY&m_lwJ%>GnCUC(hq`j1NRM0p
zUw-ILgcpV{+s>e^tl?}P_iKWEf@^dB{j5z2i%a*L&uvQIicRcXhYgP=YhTs-cuH6D
z#2zC=BUG!OnZEJz)Le!`qXV0gT)DfH;TI{m2L~LWpt}p*JV=);-gH|dY9H2?OuFCu
z3UXxfmh{3U!IEPQuf$924NT;N&>po?TGG|-?J_P$K{`-pCecyz02ur?9hLW*C{p<_
zUy3N`CatJ?f_Qe3+Mg3&J`tyVuDCh?Kl+JSvmGS&R6*$ZDOu4f5LTD$BM2_2UrH=5
zL2~RFO+GiW4;KCw*YGL0Qc`$zZ2aftyMIJ2bP%P5ZYaC}=LX6@?6(DPx6Q9}d}@kj
zcR`GY8h5YJmJMCJ9|-;LFAWN5cS_%KA7HHnCRLJvt%y8Q-x+W$9#h;EJb3W#R6SmS
zK*8#*2wXDHdSlvnP69^@C9lfumk)Owm^3)0HaBrW`x`UImKW{bpGX0VN>n++?g?xa
zB5l8YkzFNRIMtUHA-skt=rWbU1#QImoDvXKz%Nl%{iCQDAb@3Ct~B)6NpP7#c}o{^
znC!Zs?N4k7hGH)&<D;Tyi&v2bZU+CExS%YP8b{~Ad0UY&7g%OD&G^fM3L&)=iw;Oq
z*sW2doof$Ou+XZT+M(dc$#jhJPV6M@n&Z9SB)cnyEZ?KWZM5S{#ebHphCzRa5}10r
zx@Qh>ujS4?)sAIAijr+RY@CqjiVqYoDzUouJn)YQHA1+7V2Nt)0INj3f}$hRX(jI5
zqepb8f5+45A{qHfvc)ls7;H{c4CgXZkRN$T^aCcO=t}Nb>U_FllTjRhwR^sWXB@n2
zA2L$DGMP`TrCVn+`Y=_&vY(6(0SRr=h$lvd3i7+4A=7vXVxA$Ch>9e$*s{%gLgq^O
zFKTK)n=YHuc?gW|&`_s=wUc+v7dAq5<R9vcywCB!#LoZEE-PIidPLl;;*rQV$ERT~
z1CK9-K?~{%g~m&yI0Rm&=>9h~W!qPy67lOxa5_oMl8u{>w;&1!*-7IiF3<6E0}T?|
zbYX(peZ^ucVIGUFH1S)*QD(qp5#`_Bo;3{>DxDW$4AUULES9i4b(Qf6iT8cAS5|mG
z*-@mkxYiPRD*a0K=1d5kI?r}Pp_#|b*|?F^ZkcJYg`w8RpwqO!#M6FnCaJ3uMDv;z
zDNcJ>$%2C|e?mgvR#_B#^sL)Ik@{k<y@Gl*VrBL2*5%oVJ|%EMZ!`zQVwf}}f2!JD
zs`Q9VG!irSK!|AVo>PL&v3`bn@cO`*uHD8iPvJzHKi<<E2`HW(>v%Qs(#I$4S2gAJ
zVGUp1;RXqUwfhEAOG3Da-%cy&m4A84SXtD+je*m?M>yiPe<oGbD7hN4v9Ub~_pWqy
z?J(Pf!^`n~Ed}-kBO!1ycWT#gK0@$JoL%Yw5OROHqdw&w1daE(9QOenP31i*G3KY6
zrq@QRwA>n|Dk$)05ZWaziB`TwIWF&LnAW+6<uc^I;MA9Yf>ZPjz0pOD3vzR)w<4DZ
z>Dgi+>X|!n?p_}%L!`83bB>yvGWS$-5I7^DKy(JtB9^aOIj*YK|0XzKXec5@-rm1}
zUE=l9te%Y4;=P5%UM5GuqM<j_uz(z!kFYHGQoHcZd}GO~AlB#&$nc2}La0%9A5Zk3
z5-5POHnJacm-C(0YeQ*;(WbjPKcHzyg^rznalgxImJX%Q82rNDXQflPc=5G^jZo?=
zW;e8Ttr~ap7z50rK|#{x9XlzmxxJf-4Oi);$SmE6CUul8I>6;uB>HpTlopu<=TrQ6
zDuw6&o%5mp3*bOQuM~zKpr#W~B6+p&50h1QiW*n%K(F3Dd|+eWbx<MW1S3$P!|a4=
zldK`xGoV6oi}Jv0*!rbHcK@tGBGM;92eV8{-kDQR5gjxXS0P|oAA!d~(n_AKAL;{w
z%Im0Jk%NbtD1qn3ZZ^{+>rxqzzgR#}l!HNyQ&9mf>m&N_swLx+4?hJ&!3q;R9Qx<G
zkcv|FYkB~|3~(&GZO(s*Guam?;L_A5w|v2NPHKVc<`Mc&gemn8S(*haxL&LMNFV*8
zvQrj1_lX@@6-Zh#laPyaysBLx0eK<@-JJq(T-n-P?(J(D|1<XeQ+5XI@%U6{n|5bM
ztlx5~$JNrFpXUJ9+cU4hw2|r>?0hzo_PV6(UCqv8W%c>5LaOEFrn_z9$ZF9jr>*g%
zkjK#JY(HB%pDq*vYtjoZfP?*XtObcXLDP?5M_Wkw`WQwLTBx7w2M-*X%bjz>tPU5N
zS#($_*xahO0s-*-KRJ(t1_aR9^q0h0hloN2z7tupQU#qY$;|Kn`Rw;gJ-Xk$uf|m?
z{L`P!?@Bso1U)TPQDMk{K<Q6>EYG$2ekME<C;TE=togJ#%;cQWeG5<~*ZB7Q;DO2m
z%4^@?R&`<F*iQL@KIXWEmkT}xQ^**?&%&uD7tP6O+oh3Dnh%1|F~_%{9M%nWi)EV=
z(-2%t`x{M=Hx3Ks%MgaEZ{-(l#buZ-vY>+Mb=E}geEQDy6+pqFznmru`hPpkgmifB
zDPpmg;<7PtZ~!%8j{%)?l6cP$>F4=Kdm+H&cAr!Xyz$Z-UG93jC*XgLv-|=j0ehh2
zzq8G*+|;PcP-vLEI=}EF7a~OK-3-U5%g3R8$3+i^S6tk8H<b^=9cIJTpR=As<;Q5n
zsrQ9uYW-2^J>D?aXmM(Euxc4o;(mjy8mD=92O{G%VA&1Hm#&auX3gFMa7)_rk5BtR
zM^So4P=nX(ncrzDid;Gha>I<S6j6i3!E(04$Ntte3@Ej3(^<l{CSId{<HPiUGOtQ6
zZcc^g1U2=c4ATL#bJspUKmC;{lwRX&<9&yPnNoahpCCHkX{BH_&}(~UwRaX0mafW;
z44?Kcbe7jn$}(Auw8J4*Q*RU-M;#qYR8J+h`;VOr<@@o|-`pemZwFQ5wlao6?!si*
zJx(9cBg`Q8mde9Wh+usJ*+*%v1oR#*BBe*8UR<0vC&6>Uc=+&%Frb2zNpg~y2t~EP
zy@$q#U4R$lG#MO0=yfCHf5BSag0Y)~I$;3L$?6BMke&U~TsOH_2DvP9+WcbnKHA>E
z3sR>)W}4A7MCX}To}orkoiMUm^3O0-gC9urXBp#O`GaR|=su8Vhcg5C7oV$|7=QWz
zNBh$4Q9~$WZee)8;!x#q`TJ~PJdvOuQwxJZie&v8?QRvYo;JQ{5lMkpi^p<%T*U8j
z5Iz`{d~??;c?`WZ&r7@!URulP#gkCr@HBt|$0k`ev`Qmg;U>u!pnm2*2=HpZVK<D@
z&U<=<%mP{ox_Ij(#JN~Y$bwavKs(*(9guXgE)JoVSsbB^#E&(Yk!`~*d;nFOV5kh;
zM=PG2BM1E(@F9n3n1Vh<c!#OreV>tABD}RC7tdfg_Jh5JMXFvszMEVHQ&5Cx0K_ue
zV9a`B^xT2%d#}z~=_H<1UT*#FQqNA9TrhxcO$ey{2hO+Ow!xQnU4^tagVlZDF>A$0
zBqn<_I7Du8wL)q0r>>KatuS=9Mf;%U+IJp)tmHU-y<&=|o{0;zJrzZu?dgLxfP(xL
zd6-+i1JfnlpQ;q6Ql%EsQ@So3!k&SNw+}{n*AeT=ivp1u3WGoO&<4&8Drr95lLn-)
z6sjV!8as*dK+^Q3CuWx(`mq_JxJWBlRi08j>ub(3n7>72f}=ki++Izwy&MtOk)-VP
zVVKj4smLRmM0$OaXfj}85s?vcGq#NaVmw`zd)5BOA4;s-J`|1wqFp@Fp=MMA2~Qhs
z%eXIi6FUK@8?@8h@?DRtSR|A^FB$?CzrRG2jHH`!dQ>%TGQt5yW*m;fPhBLd#rL>Q
z3huq?H~;jikHf1ruz^=S{eSza_25-6#=)z`{eIP!$ZzSqXO4GzHqs{-0Wv)r@r;tP
z`Pzj`>?9b5T+#rJY|;6#eBh#=AL+XOzJoJTc3LLa2lOGq3M)Umb^X)NoO&m`(Wkj#
z7&4)8B34DOw?wi`m7$wrQITKW^&)F7gQ8voYCz`+r_oj(x~0W%X!FopmXw2b^VpZJ
zcT2i$<>R@yMw@(4jmz1Yr{i?<z2W8L<nog}pn)Dp>e#+LJG?^ZQBU^ghoB@z4>;7v
z3Jp<N1YBy2jTOQ3WQoF5>e9yx5L`CO>e}sEcRl)bNW%o}K(yj2N`Xn&OfVf(*J11@
zpg0F-Y#YtXrWnj*n5T|)WWN1`Kb&xMW<}qN^zhBAWH2yug>(VJQ^d5dWFM_fdvV9y
z7EIdbnXywKFv1=SLk%dH!65h*QzkBs#nbn#mP&N_Q2XF_;ci7@vn(N6GR+<sUVCPK
z4$RylY0u0(QDClla125frUjKvol?HP)1(PtY0N568fC7=)wLt<KO+AEuRV5fcNkY-
zer?aa@k1Z)f4MiLpVWxYK`Vyei(5uK!L@tHM@>mh0-GpxDF-+%Q{)ZZ!2}opn;?QM
ziP$|==>fsjwC`lB#oiEEZO?9Xvenn=F`{awciT+c<A{uVVTEqId(`V{bEM}-!I<9T
z#Kgp4qT6&A_9HYb3u|Cn{7#}=Ywy!=2)=A7<t@n}kDQuh9X=7d4DwPaXtWMu76dUj
z!^QX>ak;)iJ%Q|GV?ReA`Y{_bJiI@cgOxc`0XV%B6CK1G{V1Huq^pI2?)ae)j8}(6
z<ZGLZV7Y$x2r0~qc@a@jwTI<^;f_8k=6&^8>WFIMo~If-y?TAZ3}!}VXubf$%Cw6e
zG6+|Yh`Eg!!Ba;X{?Q1Li350R(L2=wN)hUL3H*lP{>A*E#j|nAzG~ID++3#ph#r;w
zX}J7LCD4odp`zrL_kS<l^$~-!D?NTL8gB;qci(qR#OtPU@HmAix*|!~%u=c>H(-!_
zYK6syoVfwIT?>f2jqSyvPeBaS=-10}q5(Vhe6W5%-Zm_(tmA}=YEJmPNWa!8@M%H`
zyH~oC$H5mWNrr)BVDzeF3ubV&CEaAqtCvn=cAiQSYf3BJBTV+I)!>}ralkNo49=>k
zqF<9tW2x}W>I*{+cr1A&0+jI0Tm?8317w+O{Cl)w>TCG-m<h{<h+LR;8noh4<6AJo
zAYl5=%X1KPD&GvD3gIUZ!;tFV$=Ld(6u9*?p&1avlcMc5(zQa$m*;P<_HrJu)Hqq_
zz5Xnih}XFG1Zh$Xjg346HRw1tx3wsC@BmsyP+{}uf+)&p_tY)-h5PsYB=D!M+@ygP
zd4`;XAbPdeG$PC>3TkIyvJX`ppz_$VBIK!@6|YVDX0OYkPiu@E(rM%>9pZAJk)JUT
zd2i78`lb8hc@evwe|_IS4@j>Z|K-8gp~OM@bE_10fD0oweGsIO1X7sK`v4M%|Nkci
zZ(XOJ>8uraNbr74WZRge%>|v@zb}I)z>wJxGr6dm<^VtIzutPIwphut{cS>b{me-U
zy$sLNGkj>Pm7gfHc%PLf)Q$go>Q0BUJF76w(|Yw8#01K%9|x2VDeuf{ThG`|wx$jY
zhlOMn^gW{wL{I2mb5^*5e}0S}U+za+MRwM<uQ)a7mt(MKE85TTtVI7qdJuTXIv}Ge
z8~f$!9M~g$Pp@@X;^JDH;otdZDhI(s=7tBX01xQxJutmM)AHi!HQF#mHiAygza&f2
z1o#xDn5h6#gpk5WR>AEv+6VZuu01ZQ4rN1(H2#`mgs73;`&+^^irY&Pv~Kv>(yG2s
zXM0{Cp49UE?cPF{0hp$Bcc1%gs%w{L`Aw+uGrLw=UF3Dqk9Vr-CL8P=56S;*J`Lp%
zJI6HEAat7m0!56$hjmG*%|hmp8U^+_#0t4QC9+eZmy65m=Qp#I*Y}8`jGLE-P@WV_
zT!a?Ux0GK*uPw!UOWAexTz|jk9&(@;{lDCU0Jw(?D@||>>i@$%bRq5`8r;KaC6arn
zx3^QcBKa&BJ;9<wK8TD7dvD?woV2ogjB0cUqgn4h;pzV)Xk4Sl7B~e^Si%^_rJAYD
zFO1q!3Zpxlv&4^qBvB;8YvZht9gAh1DVp8+&>QF4^%soj&Sus{z7YH9IO4P}e5Syk
z$Jg>+sw=qa?rIKrdoY&~nSbfqCR8P?1;7QaQDhcQ3Y&H2plxO>VHJYe=fA?NGx2wB
zmV<ww`j>w=PSAU{W!rW3=39gMx-X7|)VT1-z=Dv=bXTAzXvzx>pUsvgfNlu8#m@MO
zXM30-@=t@|$vNZ$9ls$kk))FIR1iMnO<iq<P#>=URzNVS(eASF-{66y8dYCDrTwyx
zhbrvkD_u4rr+qe-{B%9W6RT+Ja|_GEw3UqAB4H+=&kQcZd@!m8a6p->a~t{@-djVg
zk3cF`ZT8yAK8ZgL>YiMdfCVF{vhumEN;-^Y<(W(Shfqfa9)oHy?BDLUR6Xki?qc>u
z0MUy61Qt%$)A>XDaJMz1R=lQ|A8<*NOT7X6U-Pz0CoS(OIQ7CYgOgCc-8jz_BV~8$
zW~T1rU+^AS?*9u`u*2k8em=9}6#Md0dP*KKyo6!=0P?BAlHWl|TIlriosKteLegN=
zeX$yMop7%j*Pj0ne#(>-jE#kr$C&C%Ttt>_cdVaz(d_3Kw8!M_QFN2(HA<NmfP7(X
z5Xpaqo2qu$RS*;By^vbyznRf5S3KhhNda&3Wh}(o`#bCLRJ%K~yVvu&5s(>bs@zBg
ztyE<d)x)ug;|GQU(T$zLnh#anAXiP=Y3u>gSdmwwoE*GZz)nJ2js3S+{QRWa(pzB5
zPesLJLE<e3wUe+CA`gUeWf@FH7??9>d+6wx$uu^mwRZM&ML+x>bj24+7Tp^xOX=v=
zR$N@WsnLo?0ulLK(blU7#LhBpYYI`BrFn(7^WkJCT-1zVCm8lq!%yvkID<194cGq-
zxZb-XfU7TGA6YQ~A^RluSie>&Fra`e#mANA=2%?Av+L43#P#5dnG4Yfy7O2)s^g4g
z@#_s$15NUnn|bvTgj-Bn%06bgzFBvfwwqN`2-4!lrQT}2l@N1udY8L1&&Jsa--!aO
z1CVe<tP)8#ax_IDP8OF3N$>Q}!7Z4a89&V20d+rFC)YHicVWUX?Av9*MmTA6_o<3E
zLIM}QHK7IZPD=zq;8(>|&97J=g#{S1lmItYgFi_?zSfUuOKDX@@4I(9tU`Iuk2PQ3
zPbd_Yte;bWario^JP2`4WZN>ZaHZK@ue0gje;+`H!{Xano8__CI|caQ<hVvt-Ff??
zDyqeHS;&!G2sJm!)a;$ZsFjzR2ZP(ZYD1G<c<yS(NQnZ;psX`{@c}Q~_+aiR_HcrE
z()tRwQYSCw^M_vz<bL;*kf{t1D`hL4dn{bhiQe_lm~6ID27)r~^yzDxs_X~2KVf8$
zSsgjMA=Hgq)Ql%GfdAwAePt7olWVsj3=^7T(ecLIz>cYRp@9t?;v$A;%j~1HGGH4~
z#LVh^x@%+lQF?!v72~AVcfNlghHdni14rU+C<<q>nP2%|W~B_2DLd`7s3I<x<uZ-w
zg3_Oo_8=)6Jf=(;it5RGAL@hkjlpYLLlv?2-L_I(TV|9b`5+L)yS?S{L;v=^AKVe|
z`=1(wR4@_4r=42*@%L+*=H_vOOJ6pCD=MXcMj$&}pn;acEX?(;XME@tx#jLCGZd(%
zyI%5q9_pd~>x<fSz+6>b>m+bf;`@Eo#bWiUWY5Z_e8G3}454s$>KISNgNS@%qzY>?
z%UqmnIJs;@kk~!ir2<^VlN503Yqp{*&qi1RkJ{Mo$2h%`U$y8ScJ+nCqrdnY+YQXr
zXuFhLyFoQ>v<ZG?gb`xeiCd}m0I^6?6RW(q4eNL2MJ44TvMszz^95k317x^4xsh-H
zr5OdzWqMx9O@4CHw^)%FSpp?fl-qW`!%|?vPJ3g9opX9NLWDj`{PQw2A*h$2Bk0%;
z0C##pC#d>Wv0ZgEaooThYN?*5^7%+KK3MT0zww5Y1_9>PEjFTTrg!L0Lv~aEdU~z$
z!tySi(%Q9}$=-1T2h658mptnMXw{lvg)={$^xmG5*4rQJy)ZxEGb^8TvP}OBu&)^Z
zJm8-nr?sH@nNx~s*RL*vGhayj>{+In;>8-t)zaBgF**nL$|>v2EPRE#o_38K*$68z
zKOKfHu68Wr?d|Q7%!8JFp`&w4{)Fbk^7rMiKjXyKod2#~q<O_7Dzk5I)?MnIt&^ch
z=A0z{?)QTSqZ?m8Zh^&0%{|KNZwXRzsib1TJAK`x9VJoOotsJK>_*iNwshom#}IfF
zv7$o<><2!`gB6=b?QiI>4k0L6kJ7!}*fjf_#9tdm()j#u+0_UvXd!(U6}FxD&pXhg
z?rMPIT%Fz7>el|kg1n$$$@cm50^GG|SZ#pdrnu?JfZz6vdVtt->~v=Po;|-LLqdF|
z=l0fE*+GH=F5Rp^j|o{4XFYRowRg;qFW6nPaaj*BeP%M^twjQCOe)jMOM|3H9V&VK
zM&bQe*dbk{1?E+2`wdT6?%?=`rmbl6z!z<drzzB=b!V;%YDP0PDXtPKzdSYfs<%_9
zw+O$)%K?aL``sy$8}#Kgg1&QJd9<XM->{JZ$9D0Aro%z6nF95sJ6tFSg*3t~zxi}u
z+6?|{_Sig|1JwHW_spAxWZvJ?syjmeJL4F4LDbXcs1rUZe}^^IhxkgdW=}uGDi<kd
zjVAHgdE}Hgsh5zJ*&jnF3wisZn||}Q-Et@a`Y;4i7!?`+8&X(Q6;yR7i5^Y~HzdJL
zm_|4?*Injw;T6mYR^p1M^Jk?+B>Kq8c9t0jQd8nG3u4@ghv-xn8Y6tc*qcvbG3$}L
zTnJp!Ez&;r(|7J~j81re06QFXy;^1EgIqr&D6Za8Gr_aimzAs-I6?`d(qXIG*IY?*
zsPol?tp+TsgF>ig=PY968`Cy$W37!$|JHmm3Ai$M563#zQFRYHyCL>n)?b6$$iqdb
z6j1#6XFGG|4?ug=k#-iJf0fk4^Z!?sZV4G^Sji79=)P{c8_?P})zCM@JeOYN+r7HE
zC$X$OiNW?C9U73Bk$q9?>`-U^a`shT1jRPyxyt_wY0T6;!{_d!ozE|c-oxiDR1L8O
zJ~A_(@Y3l-Br;T<84{ffi|^5OVuj0>scU*`(G(d-H;;|N{?h;_oX09N$a7f;D&*N{
z?gYOWojn$}SaH<9*mG*4WNonP{z$mKgYk^so%uokN_?4*(aZoH%vQr1tJ;F8XHcGe
zc3DV0p%@XGUH5+oEu0LQLHacDE=E|wDdLdQn(n~EKUN5+?<FEbD4$JV9>OHC$&hS>
z1=?Af1Js$O{jc}Y?o~XJEp3sq4seV9oz(3qnjNBOrJ!gf#Hf-QQN$6h%&v|BFQ?VP
z)P{;)gJQ!2M%$5mvA+0bSqR$0B~+akp)NP(IsYG|rb7)}v<GGzOA{_r`FCsi;5I$K
zOMRf*NCtisb6>i1_NgE@EA$in=eaA8bnq+lr}a^zN+e-zD})UHtoI-z4-KQ>3bDI<
z*hncy8Q<e*vo<2eQ+LnhYzZKoxm0aqrauoHNDG#CnQXm3Q@}$DMT8HAL|ZU9l=ukM
zO91`)MVj^Z0pMo!mbOFNh5Gnv9ILUjRrGa1Nz4wFOS`K>_<VZ44|_GPD-Y)fI;WR)
ziaP-cGc)9t8iu6G@KsZX>Ig{q;J6+GQVi#MfCm0o7VJ~Ba|n1<r0KIQrUO_efr~88
zg{UV=vQcY%mLz0^)Gy6Ibg;2&h1nRsl<Rc*(QEc3$vG7LTi+(1;hZr(X9Sz1bZkf{
zP9=c~i^934dp>-LUq@Cw75|JF7DeUcj{5t<7M&DW7_(^PctjrPo8|eiUT<X^rt~9i
zW4KJ_pZ!shqU>Yx>WszZwnEpt2M-;3jEt(G<4`OSr+kjXsI?fZqUdJ?E9Q9&fv9Kz
z+3G_y+026tXeM01s9Ts@8`5r72$C+HcdJN8;qo*45mYoBD)tEQl3nA}-NAa(os_w`
z$}4GyWFgxGt|f+HrO%7-!Nt55hGEqjr677F?56d{ZTljk?=Ni!2J}f}`an`)V|<8K
zHq@&bT4hiCMhdT&yX%>j{5E+HicmDS8U~q8&b6$sui8tvg0qwVyI7aN3MH2OGF`yK
z9Gj+ZS4?e!RmyQQ8$^{o1mwGE2!5H;;uXXga4Ngj2+H!EPUpX70^+ag&S(7XAX&-^
z2SsIN`#$`N)x6jAu){`t4q(TLefThb9(pY@e%L;A#jt5^Iji5mDITdq!D<i*NEiP?
zNO#P@wAmH)p1i^8cl_b!eC7HGMNXjegy0+PCl;EU-@!#;dKlA?@7cAtweG*P@2TQS
zhoV~yQmQr&_NdoT@fW4=oewxrxlHV5(vlf?ZImOou)4E};V#{1n{o_sfu$(4+P909
z+<FZpg3#vSI4;s2#5xQJSFFZHSClgM%I%33^-5TiL!_SGfm#RMrSgv^TjFuy7=@{-
zsU}$lOMH1<IDuK|OI^4yO{ASV$(u@)csBUFg`}KvDOd%FQnOFwfsyYCqkwjsMxT4^
ztbbx&AZ5Sl!{AJ}2tZ}ABBs^&QZ1FKpkTx!;xy^+!@>g&dP@`dr3W}IpC+LQuPK8`
z)E0FBZupB#TE<?yey9>f0S66TWmY<pru4);_*~aVcCX<Z4D^&%I)(-!jO-hGm%_6Y
zP8W}emPHn62>@`a=DSZc-W3D(S&Hi6<40%Fu*+qDSmv9PQ3hQ@XM?=S^>=?u93PX1
z8-B&#`=TR>DVX^HeL76coVihbuKWXn;x2oW>{#(fs~dUqmjt<p5w{ttajy6e+l(-Q
zv~4;qb&?dt!}nQ!-M3F=-=z!ZZn|z7s`A4AK5*DHVpM<s{Wq;pXzB1*zW$z%x1?#<
z2-Z`PW(sM>@-{F+#_Q3fvozL;)@Ks^zuLR<aH!Y5Uk;~iQ7S4j9YhKhLSd%kkwmE!
zg(xZ_d-kn4g%cr_l!MV?3n6RvB1^Jl-(}0b@5{`*-`})6=RMD9W{#&n-s^hDUvpi4
zbI<*|zx#53zD*CTNQ%}GI$)3^g9(*LcEGM61i9>u<U9z-on(Yrk=#)xgY4i75giYb
znl;liSzWN}cq+7liuf3&#P{vJZJ1(W$cmi+KVPISHRM|!v0=kT^BoWbn>tB;34%AC
zN%;7zgKf!7FU_}o=~EhlGm~ug=(5K4b}iz31<_?>)w6qSyg0iy8M#HD2W2(tXOUdi
z$DG9x-Zk77u7K=mU><UY!>~FkHx#^VxNoVeo>+v%+8}5yI2hyXdcwcm*WDx@uQiv1
zUi9`~vA%hPSfj4C#%&>Tw0U3VrBdj_KGC!mn*)`x03zaW28taOJQ_AkC)GAr_Gb1~
z5znmeZcx#wxN7?~=Vf;iq4!`gKb3yr1ZuuF;0YS>-i$t^26!XBd$me&i_(9DhT4!<
zE^E!;9j%M0uw+cu16iLy56<L-<Ky|1P?hF%mmt(x@-gh;tymbg`Q*%a8)-Q;cx~qV
zg&!N|guT-Wo=sMuXP4|_z66EWF=t7&8H>D?i=NG9i@Jx~j@K;A_p6PvLJq_Pj8F}6
zW=5;oa58r)+>He>K>JMi<t#f3#sGY##D6kcYX=GGf*q1_o^L$muN?!x%2w%_D*Vyj
z3bZ^@&1~g(1iCoA%a>Q<{#mFZlZ@O%&(3_9D7&X6aP)YhYZf!)Jz0Bk#778VUyrYV
zA<jhuuruzDZ;cq4DNM?c)mZoYuVBkYko47hF@pg9%KMm!gKgFo*(lKP`&vX9-$RO{
z7n#W{xvO~?3eZ-5hOId{+&GLN!$vXPb;za2H=a5<&JOmA&3=sHm){Fz9gSfwBq>-;
zxO@K}87(v5J?MV#S+GXuD8A8y7z=bj3!(#dw3}%7tiG*;+;F=Ewyg!QEnYY^uj`XS
zRI66(mT#&xv#N8Kz@MCU6cd*JT`vIs<g2+_X_hVIfG^a6T4Q%pVU5FS1j@(wfk|>F
zAQ6!AlV*vtxx{R^Bkn3@7orM=@%HJDFlL*DcP>xq%|NY5VTN~T<4EhMZsphwChOjE
zxVdkk4wwf9wGQaHkMW0o6M~VpY%HUOj?lwxcKElh+tY@iC85^nYAT&xF0id}HWgjs
zDJ(yJ%g-I|PNbP-J{JLIPEBaqfz5$Sbf1}7N=!-cNEja-r_NX-Bvqce``UE-o3AM`
zM=x<zyhDdh=iBsg1HYZm_h8MhU?S|}uej(%;D3Ds=VI$1J(QrJu<lHxHu|%6aDTLT
z7!s^7x)cdV+t%7`hdkm*VAbcQ&__Sp)jy3=1gN}Fn?T}O=;yuURQsF_o5+lS3#}z-
z^DT=3<|=e}B-%QoZ8P~bjMvxHaSn&*H@>X)^FYFX35>nd+{y)DLH>&?n?=}5q$=EI
zGuzpmMr6dGT~Js1B{QCfo0|#xmnGr$Nct~nNWQ(9Fg);T)dMIkM>FU`DS7NQn=Ah=
z?U}1P1{|uCi`@BtyO@mw!{siBa~6gM#Ll3%cc<GI+d?{!qLXrhN5D?p7RDvy_8pRk
z4r&%)>%aSZeLIoBmM;+Kylp8Tt-}MY2b>pLGp0{B^LM@LmAAWKUmHWv25;puTnvad
zX{UG82H=+>Bzuyk=UMKHiKM@4$e|pT@B|I8F{|E~We#g4+iOA<ypyXU!_i+x8oO^b
z|LYB*ZUlFvbU?y-bDp?|V=YNbkXTFt-?Kxzec>43b<B*7#&#H@#DiKDRo)dL%-`Rm
zKf(v8CJf+qK()h3spS#GJdD7}^Y1>HfoDeu*cOO!W45p3DM|t;)`dlzfg#4^U&G!5
z^TVxZ-|^TD@eGToN)^yhi{I#>3!of5I@_VNU~3&8e0vh9{3EG=WCbKktK-;DWNM!)
zEMv7JdTDaAx8X<B-wf)`Hrt@zXMzla*_*u?6HkcrXB#*BT-5e+opQSg*Mg!<HP6EQ
z#JqHR4-XF{9W{S=#c!6wQJ(v976x{{aRpzoV^gbjh1;4{5pzcQR?QocFb{59e_%Vv
z&wCSqFE1Y^oDiP>xzsS!VPC9PigEMN=MRwkOUGn-3ZRXRS!uh_bEa#keEYhU7d=sx
zPaz7C9)aqC+ydP-Er3BxFs_~Ks8MehcQ#*ee#{3sc=?VXlNY;ArD`&fNaEPzXuXx*
zi>bvT|0rUv_r{)`>>L{nN&rakjxd%OrO++>E>;uV5w++NaSynixML{2YfXNIP%0bX
zZsHGDdUVe#PB=kiaI-$^(;W?7+ly`_>6z{U%mZjh0E5bbIO3rRe9&a0474IqN>I<z
zR}s)3We0t`5aG!lceLI}pRtWSKg+b!Sk~J>cCgT-cn{-kC6G7MY}e2)1(~{>BV9Ut
ze^ygRuwV>a0er5)9gBqquy#XSc0M3rtp}KHmu{(f&&J+RC$qWcfk00Pf}Oj!G~HvH
z`Gn^uo>}Zy!*S~32X0Jglmg-@kO{^lbNI?Hoa<^*iZ@+)RdMc2-;Gtu7QuGegZGi}
z*H@zhoD1%@#kT5v6}|nDPmAJgc6Xf>*)&i84)9>Mm^PnT$OV6@O*1xVE!Q7;&~Eq0
zqS@4LecPM@@BRZwr{PYW98RXq6^FDBqT87@pIRef&TGr94&8|oPgk8}-WZHrX^}IP
zyLV>cm+9D?+cS+t&}8WajT{*QNDF|7UJ>-^h?7k0=8FB3TGs2KMdahzStFBzD&pId
z6}pZ-GMwPW{0`B~o_(P{x>sP#bZ#eDmv9KYt=z!I%1c4-LylwDxj$#4*~9a?{LpJb
z!R9itp|hU#OyeZ3R26&j-8I)#`%PjFxqfzl{@t9KJSf-ku?VT`ME3gAn)t3UDF!|5
zs3Ci%qkuf0HSUwkLE_H5mAQLD&!t?_?mDY|X81lv4eFhT{t$dX5PRS_Gv!@gvbHZQ
z50GVB5roUqH%h_sE{^Hrg>H;u4kao~=G8kHd<>t8JjRU%?5uKNU%jo)Z9=(89$>BI
ziik&~X7{)%I6=B76>)`4SDJv9wj=S@J(C9ey3zQHy=q|&kY-gX)9)whM;lJY(`&N}
z;V`PubPDZnhvF~bo#%B+jcan@Buj~qqhDw_GtiiueRQx~=ekGT*(MHBHV{=Q2nliX
z)rdQR#PZr3qdK7bZ#uu-*!D7*s#*CH7D3>W$6zMZ<bl8RAIiHljg3z2MylARO;g}X
z0S^6<i32kHuNfj#A7w~}EOAT&7So%)_7dW>-8F~Kf?Is09D>_RITT(7FCBySi*7q_
z^>uq3N=K1+L_*@!_yK8Y>6fz7((4aZ-^h3Ke1ps%fJ6&naMZ{VDvcM;qvm<I=AAw$
zx@}m4UXs*aQ#&>q@0udLG?=T)YnF@m^*7kp5_XFYNp-=r`!tenLiGOj)FI~+U1xjC
zc>FPGq^btH3XxMqE)$<&<xH22O0XTqz}1FwRyMh0d+EM6wihyoFS8KMq}E-(&{@Ek
zKU+=0_0J!ujmc*gi(O;xI_btMN_?#=J-sL7#`x9yN73kr(xOB3TB3ld;gk=T2Ol}S
zc-b@m?!KySkM)D@`O?0mzz>GImG?V*x+|-qj0?Ff&#}qn*j-t+M(K&G7pG)OjP{qP
zgg&}D-#Wq4bCaV@aw7IsPe57Y1(>to_({U`^)72(hR5sqFz0oUK7g)YU+#gV*-qr9
zTZQ8xG3a}$$Waa9!lpv|)uAf}BxY#V-EFHHrbe$1erOQDoI6?$^e%QQ4oIjVH{^UP
zWZqWw7M`rAvQuRhhPH*DP%)&|s1;Z%W>QQpm18+M=IK1jJ)fZPp=mp>u6$ouI}6DR
z1swbgKyb45Hha!ZvLT1hBY_YUZzLw10LF=Rl}SU*RH$9!rCxu+5;RBftXXGiWZzh-
z+cYKbcmXnmV#&W*qbg+};xXnS5u^G+03@G(hDu=5J1ChWxs4IVf-25vL%a#xXLI3&
zypK^XA_t2CJji{-5YDk!K}o65a0nLl0l_9bj+=^wOkHrgXLe!~NWcHt0%nLNTrTql
zNbkU#mr6r<=3PEL>K=Uwmm$6U&Ew!+S|ik>Z8sA_kXale{Jh;Ek|sqjaA}_taVN0d
zTikg;uOWU|M~NVg32~f%`e_gu400%^)@n-occt80_6VjE$<awH%JP@bQS~TH+M5!Q
z!7j8N6snC^IMF*i^z94RaLCNxyFPw3S#o;vQpWRIcMOpUIfTepegv0*{nm7rNO7v@
zLy$2zM7QZer*EfbjPELp(F-wvMEl!XVt=oJ+uIm4#A#d=hnh9edTk~;Wvi#|VGu?F
z<m|}$OEWqQM#aX5zhquXsq4?403@dW`3Ecd?Vq|!0F;U3Oe|{&q!{+)>&4KgcIWk-
zFISrX^p9^({F4vejIdhdFMdB{Yp6SGp(u4@+n3wK|K9UI+0nnhRm=gasVO4OMgQ$<
z!lEb`<EZen!rdSArh>3qsF&;y+1ho?A<T6T77zX(^rmXCn(PLLA8~3>dA@yp1?wh&
z1^pq*VM1Xw2jPhyKu|;;HmzK_c`w)9AF?$-i-grCPIZ5s`0-!vaW5U6#KqnGpDo6J
zO<4z6tvkB-D<J>BrDR#Mb&;OlsQCP0Qg8Ery*2s?WHohvK==9Y*MgD)O~Ebe9BB$}
zdE7@+aLZJmGzGUT-Ahw&UtQUMcziSkM^kX*S%>DqEmM`#Jh*?Ti@pZqv=DW9zA7z5
zT~<6zQ*bl|N8to$Ik;tMBw7w`dCny*2S-63zw5Ema&X^08nW3<nu4P#IP%^)t;nz}
z=MpAB(TWV@IqN%;N2{VPQ<c-IsLS%YXbO&|;3%8`t%|z5YBa6Luq=&4YZIX$kKgqx
zXjRnj9u3(B9GZfoDLAru0t|{)WLTC`q7@mInbM<GQJ1O8|KC(m^N8l7ZBJu&jXGh8
zGh2VYa}aY+n9d7zYqRWjpC)_d-#FJWovcVKley)O3Zv7-#6>9wrN)%;zDUwAj!`;6
zuTVMgTHhu_ODxJKWqM;zx5Qf04Wo8^n_o+(<z*FZT2r0f>U6m++u=g0lYA`Pm~3yK
zbPAnykRM%=O3GQkdbN>Ip2ZJp&K43xg*^`T+Ppk(@vq+s!(cE9D?L<}rLX?F*maD|
z%9X4=r)tRh57e_N10CIFrrkO}s9E+MBak8T$)V+0@4tRah@HLi`1Ot-)YX3c7?43`
zgX9k)1NQoL4?32>WopyEP!DSbWOx@H^|#&GVm<<9D^|$aAy@!^JtE44W}>H;SbskK
z2VJSS&jBPkrs(sXTp@X%<v%um90hW=B46M?n0#5whXn&ktb~n!5J~7j&KS5ve^85i
z50K%$`iURZ;!aa{%ar2(+v<+%l*KT0+79c-5b9{`!m?FX+Jj}&#JFMeY~^Ex%#kgd
zH`_J{Dc?0RHT6kNO?@$1S7+LO<7ZYc<$G(QjPWqiL#oY<SP$c2@(t$3OmWUR3&h-T
zYd;YYkq<2`OUOE*P9tlbhEp_8FNz~WZpla!&K9e4S3(dsqtJjR`{0)~u)7$7g5;|=
z964@Q=HcOdmCdA&#e!>pvUtITW<zZki~8i@$IKxIS(@dW^SQL1X5-U(9{4`>BOH;P
zD86y2XhE3Jqo=z`<$-g2lxagkE5YrIg1Y9%e;F7UTyx!j1GRa}DjOH`)(lDcg(&fx
z7Zwe*xfFUs(d9-$3u(r+6A$GUr=A3DA@@S`t6A1#5B}5>R!)^8fnvtSX&=a-W8JU7
zYK+&jwl-||Rb^>FbM*4d+LC*0<rjIp$t%Q<)k1fO5@t=I)G3*U6uMVB1#s&-*@SoC
z3K_z{zoNEb4@xF;YQu$Q@<c?IInF`Xjvu*eKk|xod?;y)K7JwcX?VC~wM)<Bs|JAn
z(0~8XMSrR<yDJHYSFP{8qp)2QA3WYQ>m%mUUuoavBW2K4=;P0STI}>*OPtH0B7Y@<
zF3XmEu@_a5o0?=c`ja9mkiTpbLD)7dK}L}5qn}sF!>nz{urZOH@6Ldszq#JwDuIel
zi{>%%0P0oU>4{#d?qkvC+I#VE?ckJTVqBbvuJd?!Lqf+qz;x9Y@%hPaWRo4MZg_dg
zL{B}l$WP`|F~CT9>krg)J&lfz&KyDDlgRD&GzOE8ahWVXO3kSz0u;U+dt~TBnUR@U
zwEf8a9eO^icy+eC(s|MO>a42$)Q6(3n;a2JCMMePrIHfe2DQ&OST6Lp52RF}$ZmuF
z=OB!07*B%(ReW=5C*0AlN$>O`o~VUWQX5LM@F-{Hb(!+#T_wR?-`nOk^|aJod@A}G
zS!!XiLa49{s2m$ZKzks6OmKoxS|5j!df&pxCXMkW7&NRQD>%rg;JIelrX*V3&zo6P
zI=6o74Kha!|Kf@QujuE2KkUrN(d5e-*yr>U^Ug4MLGP68tRdRzWv9$pHH7k81G)EJ
zwAiI>YtwD@YO50qd67JiVqR^;RczkgNX^+kUnkvhvkO5VJE>?5CpvBqx1WWB?p>PH
zHqYKLSdVV)x(hQ-4yp!AKQ6NTiNfplA+Nijc72kXySb+XEc(OUPJPG%71a46b0p5_
z*n={=mhssSAN=p1vqGfRsXg=A(8FKIv(oCT;L?2*Buo0QD_}Oz0Y*qP%r~3!y4>cv
zNPH26F8!o!_9~^QhcM?S=F^eyIV){Oi93Fwmzdi@*>gn{lRHL1L1CA!vzekeoNNw)
zhO21((#UoEZD6Or(pIs|8rMZ1@A=7!T?ZmQ4{-U5XZ>~Sz_7$^sIU(7E}g%1vj4j0
zsKe*mvc&agTRkH~tLy5v0?3`+03mj>c}D+(t$_>o(DS-reA6is6_s!lgSml2rVEp2
zJY)l>nQn9K`DMStX5o1l`!gJnJZfQnm){Z;Y~VS=r8$?b+v1l4c+@HQ@>qTFT;FaT
zq2dPtw39!<e*BexMeoUqSAZa_Kl!OFdeJU<{IMrN0CZ>1%TnjrOqC~ziHVdsX+4OJ
z(2NCDS0%%K!QSE!ESTySjQG>}@|Z4-`)LN8BN?2Lmc~m(>nVs~*eQtWqHi(%iO%0W
zX{;B0Ioj=lBHpB~MfZeTN*RnG@`XVtPlu)eP8kJ`)Ja>_&$E8NPpcygR?1bu_aB_5
z>Uh}?@4UmW%x(YiapHx1=I98PO?#z5aC*2TjZZx%82<qP`)ipwMMWPG9#sr&iqD$3
zDj_KuaC^u1f1mWT7zeEL{7YeKUV0_n6$YzUI>AmOOUc6@fRoeO9JL1s##=OF=ad_x
zy*GyN1wW3xO+iaIZY&m1wyv5?SW6Wn(^ru|$_sBl;wmc>xzf~C%;zSy)oBQKa$2K$
zDxSFDJ=fYU_xz%<dUb;RbZxIbz_sq$q)tI6U_%)u<k&V*<<ekKae2l5d|k=a1gEAM
z`=xQg&UPZcu?8QoPgF1G<<NZw=JNIB$2AU7pm#R#bYE+c5H&}Qt)XM)3=POgOR}te
zo*Ny7Hnmn!QE73W9Jo7ESa{I#gQEA50T>HNmbjH^5XI*BTqFC?R|9-lz+$e$^KFNx
zo6Ko*!|J0_`1roQzQe7!RLioUW`pX8GJ)Zw%OM-VJ=+<*lfsgicCbGS`>J599Obt~
zrR2f9gOH3_AE~z6iz7}b8@Hyyo>DiRs-RLunA5B%V8cLxRSrXzZCl7aQ3{WJW{PO~
zoas2?#;1pTRBM|Xui>Y|FEB+523~Rg^vKN{aF=5*cHdXOTnNtbsgs(N_~;tEXq8Mq
zRj3trj+_Ssw>?fX{n3sG{s5%w7xt57QtGp_wGHYY^DhOSYpFlgMTw@~KvP`{7V0d|
zq4R_#hv-Pt#+Q<2=@(895ZX3tYioBEZZQ|VGT&71>-*Gq;fV4%N*22cIQOu5_&q9!
zc^xB+ExDJ=u7@8ufxIPPv~A+y;fY>!GS*4@G&46Hj~3dLGgXBaj;5%YF2UIfPgYUm
zAY)JgdA~6jh7Sy0b)2rP?+SIQix?S=9$y$V7upmgJO5F7zPq0?A4REx3Am)2LBUdd
z=IUqb<_e{b`N++-@F5aoQy+J@Rr|$we<bb}iHnR%?lSWQmIl;pATOqvu)Sbq104FP
zafVUAM17Ty_ZXJPp2`k6I@^-853m-eYw)<MI;le*Y7g{>8ulQbzc6JX?Lp30N*n=s
zq`-9mWhIOS*GRRtzA@RxL}`~Ep{oY(?o(D)hPV`*4>DeeWsNNvUA?;NhyUzPY)j7E
zPDAx4r)~GOTI4^6h)ZRFFzgef8Gj2fd(Cr)7q-e+{p&pr9Ut|{gTlK!=Y*tVcE}@&
z?~AOnu>MT?*pR$as#F$$H;=OsEoNt$YufA2r9F9O_xd++gM#1JQo&Qlkib;gM!k$G
zC4mVO!_Lgia7me+%vu~N%+bm$j68Z^TiuInafHiYHi27LLcZ^g0bbekJ-Uwl?cm_-
zQM8ew@J7e8F*ys-m#It6R|e*Ne0{@Uq-AufwocBKh$nJYpozcEy6ADL1nTNlO$?j6
zsY@%|@`CY^<bKx{#3|o*{t;CY9Hxd%stDzLrGqm)$Ev(8@oJ;jsWM*Yv4bsErE!Yc
zAr)s|cE6#RbzyI?1YtU0W<<$kd{*q)OsYZGFsdPr&hc{YH+E{at&fX~a}R%64<ETi
zMBciHUk+Y>Ptfi0)m#d0&FWKPkb!nQAWog`K%i~k*JMK^;<_~4PB<5jSH4?N6<!}W
zMY4F3l(zrRUekpMK<M9d$vKY7FQ!WASeLyV4!o2*nzPtuPnhTQF7%b|OnA7r$3#(*
z5+B|10sC{6JMLF%EPI$Hwy>}eP~U8k(^ZpA6s?O}M7xJS8H{<MrFE^{0pKg$1t`92
z@gb4~?Nd4HPC>jsD_U$Dy{_wgfb_@fV*gT7%0M?h5;xR{)p2ksxvPf8t8!ZvcPLIX
zaugPMuns<j0Xk75l+?z{wZ(=>x2g2g`&7;o@kK?q1PQ%4O&LX+fRVUG%4{g7-PQ8#
z22Io4AwtgBn@M-^YbNRBy$j3?DCn0`DU>*m<cwjrGQf1nq@qN%HOZ-|ZvlkY``u{P
z<T1o&e9|<M$>t|xouhurT7d%L8ktrb71vc$SNd?=0GOMHZu1<7s1JSfI$R~*PeeQQ
zS|Ch*&M-1D8O<5Zg5fD=_vjb=PFZk^HNf_!hH6k%*Qks^qWNH(x?o)=rd4`wcpm^E
z&K(s_Nlq?EsTURF_SF&XT$<$ESldF$RsBbYtjka?Ykf>Y;u*0zB$qHqY)okbv|sfs
zh<6oD)c)8fxkh$uk_sux#w@f?Tqa)KNtHt8x<j`(azOO*d+8n1jgT8M9G~k#d*?31
zm>y(x=in6CA30VhNnRDN9^xlu*biJWEv4!=QI$9H&e^qch#L|vussd$LF7e6Ec>Ly
z?kiOVklu~_C9rfuElkY%8bJK*2Vmj(p}=ISICv$U3!^H|Revb0JslAo^rrGk8NlCX
zzA1i<`PJVp|4kj_m2Y6LOt5x|J4S)MN#-gn79)@I18bvcyX&as+*riJ19rAS8gk@p
z8K{;!NTHU6U4*KrCTueddduZtWv#_CRtz;p7C{IZ;}8qeWXT>Gq^WFjTjM$pMv#5%
z+o2!dwcB>6;h{Xfpi|!~Q%6+)!xQV1*NiCyw%AyxAzl=?!b8@4f8FWn)rg>Z86E{Q
z$$hO7yP-myQpE6&{4CouAG_~G7WI8_(o8PzeA}y06aVNNW&i$ztq$T@Y|)~$57H}O
zYUz-=6`%u$hq~ph<j^m?eedL~38mcERpgAij5bhrM$B+VEyX=mN718M4E`01Nw{lT
z8DCHE?}>gV1S6^(bZ_WE63n9h`$xXfne?G{H8{e__I-~%okE&wBzCpgce!)PSY3>F
z`I&;(fpS`TVxjR5Eea`$&zr!!r85lP6H_6RWQY?Deyk?IN`7+<GIX!_uN<>2uhJ0s
z-y8nU_aLDH5fax^rBPnU>Bt>(`1A-@1o3uFS30GGXmFsL4@pZ7pzyw5w7}t=fGwI?
z1cQI73>Ismix>_2P89t;M|~reZ4O9zi1`rXcQ*gO70gM-_n|J1elC#ws_uG7arg6|
zPl{k5OBLs?z~<2J-@jkK-385Qo-?m;VzA)WBI362;bw`6>tCmnm-9T)^<!I_&;34p
zf~t3M<mh8LFWvcj{t>NmIXkR{$GQy^+|VH{l2OT=@eif0Q3oS~h4i-|px{q_Xd^U_
zbVNj4fFQl1DC)%13=`3BMELs;=j)lf?eKhZRB<t&&(hWFW@ZTmKTa~Ni><yXen0;t
zdFLWel$4u%>5c2Rr0T?0JO3I6mBB(k^zJr?UQI<lm<8w!>5t=-{j5w~RPG>e|1Bzi
z=j8e60ZD-lmLguXqIACC;}-IToiH3IuPKhKBC2e?(E9X=cP32cnJ+lS<kOx-P!eK1
z-yM~^pPspG9ge1M``4J3Cedha^$69SZyMZ)_)d4u&|A)J#MdY3yD0eHc>a%RX?Xh4
zM5=x%m08E3N3vNgGq>V^`C|(YGgeOW-|KTDaWFoY^#FMf<BQhVy_*4vEnr~1)z#Lf
z53IYdHkgqZ7*f(7&RJ7(4e&efjJ6j!?Wd$DkiB5BA{LO#+6a+?@%tc~APWyH-k@WO
z?EBP&kxa_wmOz+bNr{ZGQj`-G2CGQzF>86&kbK4Z!&#i;K#6qsof67EY-KLpJi#R^
zk~|}R(Ux54bTJzEqsr3xmPIb-r1SIe{uWkbZJzNN2nSemxwXlQ>}Rq>>0$YmJD<lq
zc))D|4$u)}im7oTgjNego`#iFP$F-|`t;t=l<^kp3eFH`q<9ATVJadEQ|Kg(@|itP
zfdxO)=_l2XUOQh1%Z$(W;=pTUrJPv$$E<5X@48U!=EZUB;fZS>JM#bNOsU^j0E*XZ
z!0hk!?Ke{wwvT=yC6P6g8kR!s-c5g(|5mB!;=z&&y=4a!s7MqUD*>^GvzU>gUlhzU
z)bdx*LDg;RU>!cI@7cVwsA&4fEXv_!lnCU7Ql{mHDCirKc!l{gXlZG6F18b^?{W}c
z^Dfmtl(zhpa-#eny+@QaRz-B)HEZmRielvZo?l>US*e#XMN-UwX0K^GkjO_?d_EHt
zyMKK;R8+cxOdbRT56!v}yXQJN*+eMkXiXWr?>>`>juUBQr9wbfV4FDXXBvB-fGu1j
zH-$c(^-2(@!zmatrPqm{bvw7$t~J9GvX(hIlnW|N>x%2^Y^kEsb<8j$<T?u1DmhE$
z%)N~-xDr|B>nFb;G`*eI{$r%&8%FVW9^%fEf4B(+Z%cW0i2Q*~#e?5Rn4Y~x{xMQm
zhE>DnHKoiJdGZn>Od?}6+czq0l@ZwL*r(^%IuCO`cwl0W-<9giwiP15yX`FupqGS=
zS>*)<r!~xj;pf`1wyRXN#A1%Fjy-Gb^g+HZ-fg?D+)TmTLcQND$aEEei>RcouI@bc
zIsyu!^__%8;-0O}V~-m$Ql1*=c9bb{ig+<{%!uuJMWNt;-Ar<aPUbHrQP=8>7p|<U
z+Xx+hdRN<)E?i!mYw2C{yWUdgtp?My*1g)*zmm0G?5G55hKU1q4lz`jl(}onvFDuJ
zB05-Ztdc5x7f{b^ICqVBra_J@!~S#}0>>53{l2P{6%Fc+?#I;Tk;CmRN`~?{?=9w8
z+k9m{Mal1HrLaLv+-LDgIoilZb~j(n1FOYuWNz-4J<}uxbLoRiY$1W>B|Vzu{wVem
zr7Age^39)Hlvbn0Orpm6*QSr=tlA4lHHjvIhp$!<lcr9d;qR|z5snt3E*nuP00R()
z5A51wJM{2OqW%>al9|k-TT1>1U?h<nDJ)tT9S^03ESRu;3`kTo)raqDwUg4we%+Xq
z%cH3~{}kzks5N;=seLD9+nsnH(*_2rXafqVcXGT`Pkyc2jA6T9_H5hE=})e`cJf=i
zIdNMC>*Tm&G~(FH?`R$4Bfp-pHYts5YW{(`ewwLxjFXd7i)}N{I1GL&DizDEnsyjY
zsYl9AiK96N1}*89=h};Sqy&!bzoNymSt80!YN(%rh~0SwTJF8F2&E&Uu*?@TUGoPf
zr)^EQEOk#);vsA-17uu8jO8ft3l&ZRjJg>?3}x$18yJ30!M&kE6<|;%;6I%8*7`9c
zAfgVyYBsY+cz&no?oi`%6>EY2RQIHgDkVVi^ahaYEv?xnVTLy~CqwC(mmAqD@1#Uc
z{{SLDP?UjO8?lS8!Ty-#Xe>ubD!-%DXe_r($)#zz<w`D%<!CJT9c}c@NYV`Ja(RSi
qSeI!DX)H%$If@|$$^=VX6?`!{a5&<z><ai(Ry_8v7r&nW?SBB}i@z-Z

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/img/sho-ode-trajectory.png b/docs/2_39/stan-users-guide/img/sho-ode-trajectory.png
new file mode 100644
index 0000000000000000000000000000000000000000..6a6a12862f7b14eb4c6c02888f87e8dd2e1c15c1
GIT binary patch
literal 322150
zcmeFaXIxZg*F8MO7?W7g*rFn6jJ+TrQj}sWXjBjzA}T6P5vkJKB*xf5Md<>HT@e8l
z5fChZ(p8EmO{oIXdzpFHo;hbA?YZ;q{qPU>?>2_OnKRef*WPQdwe~f~)@!H_`*y;&
z91dsL>Q&1(a5y9O(*M8y3g5|I`1vya@ztI+>dQHN`d@I|HBWr!8;eys)*R0EmGu7s
zqt7i{if;}+xO&}+!DWNL9xg3?>*&1I9L{9U>g7u}*}tnTnBxA$($Onk`Dsji`qkIn
zdg4!ZP5%D%;gtz;o>M#nrrs)C<8NM<8DKpem-)k2(YCdJ_;>tEgRk1aKYdS)?)i^j
zUw<*|kKf>MmIe>_)3+vnz3@-p;e7cEz488E+{sz}|CRne;>!tt{?@Pm-4y4)l|oef
z7gJb-|8^f1;lGr?`7b3<o%k;${6CNq?)q`K@wU<1lX_j{mn(=5*VieEI``nf#A|O}
zuID%xM$FZZFOH8bG>Zy9;BU#X&>g^i|1BqK4Ci4eXV*!~B@(KxZc8LeomD1&%MH`b
zv#GU>@s6r#RjLR`8jx^cP<%^RQ~bQ?6ixlVI@bP&X*mC(KZL}8kQ9sXA9%+i{MY7i
z{!0l|B>qbY|D^<qga1cULV{<UlgEQc^P30rRK|V1e*Nha6CC_DnXEK8yJYNlRtMY0
zt=Kqa-Hb%5D_-fZx8&ar@bk*qx^A4Y$?a?zId#*0M#rywQGB7m@X9XtlF?VowT3Pq
z8U6Lwl}+8oTWazS<|}-v2zXZ0Dp^|@U|e6>`LrUktv&Wn{L=NMe<g*6eeoCS$YjJ{
z2&VOixPPIcCVcr9vMPJ>U*KQxR|Eb+e93+N7kcc!TNCaK=Fk6sR?Y7edlOu~p1QjF
zme6hIb8czRSpC4P&~M7J-wmp=E-JQUEwiq=c;ssMEB=r#gg>7UxS}txbYg4|xb~Dv
zR?blETrJ^HcBb}ic&TP?U0i#SuY%>C429B@YyB6cH!v^tc4}WRygj7c_<U5S2fxFk
zz@hTOr`p<rD`EK&hHEO)d_5JdE6P5ekjhZ34Db|l>v%R^de3jaFgK88EH{CT^@Kp<
z?zYdJwZ^Lc4<0-yaP9saBc9O^f1vP@TrsaLN_O|t-?rNo+I*_3TeO|Ihpr#{V${h;
zujk{q8=GcFx4ql{CMT!$rgP?j_YcY=Qs<}cyLQ2=GSDgGH1k?j)_q%`cInvp+kHO0
zxmGD3-<iI@sn@M7GN`=FQ!}H~L)Gjp^GYqV`nFr!h!f+4ZztVfKK`mppljEs*zUr5
z4JW6ViwvWjW!!Id2HxIwe|BY%o2!zNQuS-|-mX@~?%K%6-Os*yHduK5V9ma*o0a^z
zAX=8+AjUt9>*w9CH+OCJ-|qP7%2?)J)PnkUYp_RmyBoi~r^!3MU|wAP<L<fwo4g%f
z>`#Wx5bf(=G2KX`b$6mG{0h6o=g0Rn)fPm*kzszYDyMHjT6P!rRLDo_C%fYZUAx~G
zSLEuN-yh2Sq@!HllDlMk&C|6?ofp+|w9@MH{k&~oGd~pPC(1fp6_bnr#r(%EP9FC2
zJyd!2yzY4Zd#D3~ud4g~URNv9II~VG6A!8`d^Ga|j-f_VnZI(XbRU@JpB*%0tU*z%
zgL1Wla!<oEFJ`^f!lw0mr}kZ+u0`&8^6NQ6-@`-386~?<Qm#|Hvo-wbCFWXxc-^0N
z?AW0bZDp7evG)G9^y6dZye=#(3<+v(&QuM$u-NXZUcBpA=AEg9P3?Ob@3*lI))kpS
zK_>V8y6a-3q+%O>3UeAZQQj=A;ls0eqlB-;-Kj3xV`^e;2&}7fQ*mQKy|u=ls;a7T
z!YJ&$NH@WkTG)?$FEAMY{ly_M6(77n#N6iRqcmG?<V9twbZ=#!Tin^cg)THLO**A>
zZ?6{KhmKeNLreW2=L_rBt;=_+kDDhhygYa3YSGOkm}M0e1xC!>7`i3$;HTFI4<2mk
z%yVvdGWgoXi%*)^opQeK0c9^WHa7N>(hJ1xL|a!@XvcTuNXQ?0CZ<|&Sa^Bde&1s#
zk&D><?Dx4k5kDI07}XU-=il46vCvUWQe_Di%Ur^(r%^)I<C{^_AI@Zcx&K3nJ^t?f
z`;A_PH*-w9=I`)xcXxl(Dj{pM_V}e$S2nEPzkk1kW7Wj~tCU&Fd*-L7r~4||+#K>H
z`ySWyy+^&^kdUcj$}Q`0A*0N|3o`=kFPX&+9xm=P)nkdw`JboNJLbhi@Oj<*qSDf2
zkIae19Z5?%%RFOTG*)OC6uY$BJYnuGYyUT*;%9$&r~QY+1#bM_S*bGC6=^Rk(*0|4
zom#41?c2Y9&6N!=%U*iEiFf0x>b%d&I_=?HHZ{8J>>AtdwE2%a_Kah-ss6iU$6b@0
zo11Rs%-wcx<{Iz$>yX5EbS+x6$OuUoIc?VQAC5@6+E17;!M3wTbMfbU+a1c!J2btV
zVfVh)eR18xp3lX+&=X9Ma(AxlTL8(%rc|lQUj|F8((&?DcCrn0sPM&7Dp>5At`XpK
zX{L-}%DI!`icS2UI6r^?=A1BBIpYA!m5*X#>{91WoH!Ag^n6m%%5iM+iSz1<IDWWm
zZ&#YXyc8brlV1nrV>=W&N*>Hws#xKpzv0Fv8Hc8_mil7;=I$bONtc5e%IyUO7SD#o
z7~aV-kqhFrTx(90a(%Ue0Vw@r=WR$RQYL;1SMF>o|5ROlb4~Hpt5?Ns-|iMJ2V&i6
zUAGHzb-KQXve&dVUbS9|#rCDA_U+sE@`y^ylyQ@o2iI?p?DB|k>26!nQE~{+t-Zin
zGxxU6#;Br0Ukyunr7n3vv>Fj+asK}P)`d2C<tJq`ULaH=bdB6P%e{@`d)rb!T??xF
zi3zsgypbZz)#8w;%6F!O5HVNQyeKxd<@Fkyn!?ww&#v~Ier3ar%b}t3ko(^su>O9D
zy@eSr<y^H)GR396P`jOnl+)RftQt6S>PmyAvX}E&wbF5nDBr3D-4g~0vWH=+*Oi!<
znE5XLett5pJsqQ0oWYA27Cf|6HknWFr~PqzoO9{K;um@GyjG<8$UD7@m6V#Eh`BC^
zDt2{Ba-UT5(;(pr1?Tk@pO$alKRA#YVVJtX%digB<o1r32WD~Yg?7cU%@^v|YK=xD
zqIPL|1a98AaidDtE02obMrph-^xi>_pl3yKSa`Vfj38$#Y~SXFXJUD`iva_u_ETMQ
zcs8-9<RkmC+WlU3+b!*|jGoSBY<VijlNQ@%huW=Kvu5`BpO0U9@!L0}UKbQ}n?18n
z)h#+J*;6vOGe6VSYyX|->CF}CsTs)9oi&5SBm(T>y9+xL2CJN2HeB(?iHl5M+}ro2
zsiV{*ujidjO>TKTpI0xtKlk=fNtJ>`rlP5N^wr2Jxkry4MOc-mV1<$7#?0ArnHD*&
zZ;@whyjqsEOzg9)wF@d^U{rkXXUc0885jKVNDLE?kI{*9v{`iE-8sAkg9Pq?@k$5x
z4p(l;iv4`RJoWNwPZ^u)Z0rfu{dsrRogF$!X<<}>)fx?z&a&G&kynk1v2-%^Zv1ZR
z_Xna3vx1k-w_u{!abaH^csu~GnY%Mq0UN}+G9&$wO-*&K6EGg(G%QzRvSiScLxbBt
z>4=S&`SDtitD}Z$Z%c7+5wE*G=+2!x3+4HZ@7Ln3y4L3n7Pn9{$_VfU05X1ab)B4(
z3aV;!!_y(wy{&otmlACD8`ZICBJ=ku%RtBK8*{fmT!7vA;pyT0aDDYc$EbVvE+fWg
z9$)LTVCi19rsKv9UQw~r^7PzXGL&j=?|4_J>DKkexydsq4TWR#I929q%3h0^yv5-b
z-ieHq^-$^BD=fv>cN?OLklqzP9-X>j^oscX$`?5tp%*|y;lpp@s8(Z}X8SDI{nKM!
zYgpx_6Z3!8*49q%q7_tZNEp~0?%|Tg0@kWCeWffWF*j5<(%|J;q};R%i{)!_LvQKU
zqg<6gEOlR_HQINHgVk4J5~fqae%Yw0EQgSJS?;a(culMJcFPsN51-^`ovEBBy0B&6
zg>BxTrmi?K_R8GUxm)i}UTjx)Ten_XKMCuyY4&+>pz3|5ruR-OJ;Vw}+J1hw-RblF
zm`!f^hTewR*R`mOePAkYR!AR$+$nqE!Udhj&JL(zW9Mx<r!L`e*(^@4{ZU@|^$_=;
zQE-mU2=o~!wj}E(ThC1A3fcU-yO+6kJd><@Eoc2|#kg`Hxb!pX68r3#MQ{0D1ZTP0
zIVYqZ`y`$icTIpxPEC~04Q&Vd_C{V^v_gxr77#?#;TF!YVZ)+79-k9oR`>`{|E^JQ
z+voSzb<veZNTZ5Ks>4PIFVB_e>sK4b?Kd~q)bYBp<(AH)gZl^p=Q^eNNF$G+`i6#v
z?%TWf?rDkZ09l<Gc8@*ZXe^#b5s=<BY~mt=q?6-)<@Rq`=`OyH%iMacZNFQ;jd)4*
zmeK@nD~PskzNOcD6V-j__os7nbFtn52VbvNEa48Ed)Zr%avy!VN|=GRXnNPRK*!nj
zF7-~<$&>61UYwrt6%&RDW_=ju?KDavXQ!<}I;38P0C{LT(A}HdT(3(rt!l%Qjl$yE
zB2Mo#p~t6aZFF^=zt=cD=WwH5DYwql0J|cG=*|j%fC0CGOv%_4)mJilR)7Bd8MO!-
z`Oj6~cj3Mb$bQ>&b>C!V?V959{NfP_KVMnn3gvnSO(cLVkvweYOm)?+SI*7;#cTET
zLk2Ki)m`1b&~d%?PXOrHs2Nvh9!KHO3R$VMO49cENcFPnTVWZUl|k`qo=8R-rKi;B
z@w<AORr!%@Q1-hrSGiz_RP1R2zWvL;hlH}yfddCnq-hUzwbz?ZaXNizX47FQ=PBMh
z9Y4*_(X+|fJYLbtVE)6BFP<^*J-AM^g%Tw8=G@$jaKW<Bqs4~)dn7d1Kx~nI(#a*w
zX|mBxDRb;@@bq})dR7So2AA!qcl2g&dA+tMUUrE!^ns!1_9Ca|_r*Mvx52|EoD1<#
zQ3B#iIX4UUcKL9FlS<<;O@qQmHqqr?x+b}{Ha6u*FDWYBwVo)n<Jr7u)7KAlU2m<)
zGe&EI8*i@6%tUWgo!cFJXn2Zrvbg8ZSI(lVZK=u2{HeneIIih@n6n_b0u-%Ha_iCY
zjoWUq)U_o<vfP|kVLp`IN{{-Aq|1w&n{RCh-GUm4YkSqz)jeuCU}BQ8`^Y3OTpsAf
ztUw{PWQ^Xj;;!5%s|sIZ!cKBE^5#VW<7JJ??!?rkRJ{rj?b5se5tOS+d|+N2I9Spd
zDZK*y*GoUS{n)1{K#E;eAsa`pa98DZocvz;=}#RehKegJDDEz5mpSn6-V(I^ay3CN
zZTSJ#8H&-R?u#SPgHD*t?A5SYqN0_Rx`f};?5pbLni7!~>(JcvdW}?0zY@Q=sL@@p
zI`*5f@#o4H9^6P<2~|7iwi~vY3?6lX?fV1c6)b|;IB)Rl+eymX5hz>kZ1-i7ukM4n
zfK)c03bV7%HI^n{6GS=^@z$@yqP@Jl3V;+9dt2Vv&El3-s`4FFdOnC%KB~F1X5&WB
zzRUNmy>b?|T!ek&M-&Wz1LGi<Kf)qk8N;7{A3A=|l<dNXmZetDhRK)JMrLLJne5N?
zHuTkWsovO$9^I5pz_HUs0k0+&?$}zL7Nn~BPvvw<db-)HUG4s;?vLB<hn0g((Xg|#
zOYcHDFhbK2-SP3nFfrkpbG0wT%wm>*mpp)TO3L{;!bWP#?XBT!RDa`{>_yOYG?p@t
zRov>ul(9c~EOhB=xxF(^Nw4d*x9tsHsvge@Y1daabIBXF(AjU6JD69KF`UyO>A&cJ
z%x^mu*f*3cT)433Wint=q(igOp)Z1s#pm<$oV-G&b27GYN=yg!?9+`hOaFDy5E-Yo
z+O$CB#Xvrx!dmO2@BZ?e83Cs~HI!y8`+aEm2*>pN>LA{yAgk6K?fMM|;$3@*T>z}u
zI++`J{vs!9YPQlLI~yB7t$AM!;zo|)DCmX{-hQjaEPW+MEzZWy?pk>GtI0bDxz(*y
z;50}K;uJ;EKU&0*W}^|AR`qf16@ay=?voZ{bDlDH)+Zpb_JWd}=k#?yTSw{+s=83G
z%kK)~V+Z7xWh}10trG)`g^D#*wezLs>v!)=mU4<hI0@ZjIrbi3;2#OMp6(dOta!uh
zn%delPzZt-Dsrk%M|ZqEs%YR>Lj@ofRl`$LMe&HZ)zQOQGZrpdRI#|YQv+DUySqL<
z-l!KrW|cJF`10H}Zfuumzgd*bx!O+%wt(26b+jvXvBSS=M-@FgBBlupkbPt0Ml@aa
z4ZDV1)SR$jmzrzCA;~g==LmuQxnX+oD)p}IG3fVYt;$xh<)}pJbryEYfuWr2Hv|OS
z-G?8#zWeRn-2f^*o*r23^26hHE||%&<v)qIIHk9{ogY{H9*|F5et^Y;FF3|?-j2E#
z6=e*T$9J)v8FlLd^xp0d8c7^oK5Y$DF@9~l%f+w@p}-2BqcoRD1MnAJ>56*X8XDc4
zu{bhsr<2Z1dZM*s@pkx;9(K02M$dj9_SK*Sv%I%&eb8wt>JC6Wg?+gHbA7Jqmz-=}
z?13HT8nvluBe{XW;o*&9W(VI#TN?w|4jqj9;#fTRRppupA*|E!azG+MS_3TKci_Of
zKG@V?=kZOlk>+CE&*+UhGx<H4YlBpjLFC-l;iHK$CKv)5Q?g+DaJt2oFY7|T48FG4
z%xr8H2j55t>Xwfj$%H^cP#gBAQRBx%)B!^0NMm@-j)o7!4QP$VpQ#9p15+AQi@>v0
zcGZMs>KI=*@2%$=ZPfcQ@r3ouY2IZibHdgxTXsm4ijIrO2QL3gT%59@&ed(9MxS0^
z0Y!AZAHux%;6c|?s}5OYQ#(O5A2e?J>5PmFzgFM`4;janDq<3UXK2TSIn!CTiL1P`
zkKG~47EMH$ZvB9PUtPX`U#=ffqWg917s--t&MN%Q%u3Ioo-$N)kP+imob9n&marAQ
zWQ-^`CrIqWgVc#ISdG+=au?yyTmcF~+411Jckg8Oy}m5&R2!bZKP+zUm%K0Por=4g
zTHa_@-n@0^-IcCP*S5%f@GRD^SI-a?-I%o^e#7ZxTU(p3Tf4Hx`)Wu?$b1*D7P7Xr
zZ-M^+Zm|EB0Jo0)BJ2_+VF$ofEA&=5w?88_CXUk!K*L=I5G}KWd49IXL{+itHXe$i
z*pdb{0^I<?^5~_*K=hO({32|%R%wkMFK2S@aKirU7cWks<^aVr8oAufo9AkeD(MY!
zW~y6np4$N~bNy@)!N*lG1RdZjZ@v@jf&?jC7gzMir(b%`xVGU25doiy<KqvjO_GzB
zuLM}JrA7{E{E|^GC?{Fe#2%0a8K8Ieve)m?5#gPxRuTdP<YA<9XVXC&o181=`uh54
z{{8y^1Gmr=SurDBzBqt-9gyJWCb?ldViw67rM;MBS8xW^J(<IFB^{okx|LO9TKrg9
zOr>KNZXY6nKe6;SHOhOlg2i20UIE~5yt?+~i!)Q_ed*}a+xA{czS!AXL$UlE0Rg!h
zZQL3*zEm-v*WMo<mw0@a#VnOtSVK>9rZKOlvp=MYxpHM1C2R12y1Keutu@9|vYy?)
zcdsmIqIsHAQFEYa;iLOTy|1sVX+GoO{G;{`o6Qp@P86XYRS|j4mDt~hG`;4np>)!<
zNKUT&?z7<_@mAy6%9{lks!6rvMm?6x+ity1rQsK`%ZM?fUJtjP(xJd9q&bL6Z~sf{
zgVc2ZnTpcek4MQj4qEHCa2C3%na6<-RK|Ycl2F`RSX{nFuWJLZQIc=zU2PuQm?RDo
zjhb-LrHe&4lUr!gTAf2cih>ebsr-uf&YQaR05!sh2c>#BIzPE8Yg}A=Tz&ex`}?gE
z$7t7X=mpN{1oN>3guo*f9XBklo69`f;5+}4{J1~j?K0<DeaTd;y)}?d_U+Ggbs5R3
zyhb3zA3SGhE@Jo6SyAZ<9*)xHMNQ_mQ1IVNz#$=24pwEY4Nz#byEs%({@*Thsk07B
z#k%-o7hN+t02WlG<B3>B+k5lOx8eHiYH&oTGIkM(FVt!OAJ|DnV(?jThXOVXeY-bc
zE_zpzE^>8@%Lx^O^<Bx<!?(|(2Qmy5S9=8mDg`WgGl)R2Qr6bi?g}4|BDajvP?$m&
zJIry*<_;K0sL$Zty?qWnjmg4t3z%qYl<s%_!?Pn3Ckul*L832t@r?-@QFn=wY*$xs
zTQ};CS=^Cf6TO7pNEw)`%#tUE=6(5g+Oi|V)8|DMoF(`GI^u>7A4)<+Z@rtFS<L4L
zLnkbJ!EVvE^`Zn|zXCK|<la}8kReGosm!>5RX0lWVd%oAzkRdMtSwx=!Oa?NlL0zh
zUlkWeuveQRc^{5QW}xu<)P%(s6)Q74(rISUSf*ubFt)L^tpII`x{7WM%!WD;l}(t_
z4tJ&M>j*^xRbg(uHS@l-&xtLcXa2G^PcJAkLut-p+uG|o{5L_og{U4gh;q(h5oVj3
z=-xX9N`WjNE44EKi|=K)$gE&%R8jZmC_Q)Owzs%{bW^p+I%B_xHJZNG=QO=ibk(I3
zuXA!XNAkLh`L4qiO1RstUQT0+-Q6Riillb-^8Z~!uR;R(iTg;gHG3oaV889Io#io`
z_~!gJ^YM#Jw}Qyt&eplNheR>{b{STLS}rUHK_S=HTY5R6?T8GZMn5E6!nxqvxQ970
zIkdF*_vcl-+aFeVr7LxQ%-oQbr&0jn*dX5}HifOtyTtxUNkONLJew@xWD3z!YRCY`
zbN$($+}odlMpum6aZcAfPL4<;cH7qu5T)@J%Rf^Pg^nKMWv3%YfzCti7gW!dt^Y*+
zRG{joO*taIhs(r>i>R&Ss-PhMz4`+3fj9r$Zhmij)QDj1PkCW_kHIgO-Fu^virC*4
z=?cUE(%4!Cu?~6SH5=Sr3q*V0B|*SH)W=ibL>i+&e)t>yK`m?%`jmo$Z`PssY>Ys&
z<=3jI+`6d<O>70!6L3Gv*nT^>uli)IVKj&nW9Oxyn!I`SO3VB4sZ*!MFSa!m^3vEN
zfxrL$)sO*Q=Rl%>C@_0@<658}c!0dy{BkKbUn+1Ydy&D6SA(naGYehRr2d^U^ZF;Q
zN_`lwI?Sr{lzaur5di)BbBh6rI;+%G{YV`KR9~njTz;;~O3^EbJ?PVCP((?P|Fb@0
zz@NJ0uDTNVRjZuM+T|#>nn=3oT{@9Q<p~4D(ypoY?qfr}O202?7Tnqr`t$v5*!wyU
zOm(0Z)x3_nabwnzNw!;sOM@<W?!d{Yp`B01JwClgD(;|nb=c#s>XQ@XW`pVYg^k3P
zDWV;38~OQtROOSwl8`T>p#n!>=K&`!+`HagrK<{C9^|`0*^4t{zi3dWek^lI^V&PH
zO({EW@(U_eyotR<{KzuJhpX}vyATOx{c!{Y1KiH0sG@oE<}oB2aOmqWJ=Gxt-VO@t
zcs_#&&H{9va{JNNHaB{>Et#tBo(A%MDAQyIi;0+l^)vo{B_Jkh3e*x{wD6%^{-NK0
z|J}KKj^4}PhfS10v*gzbZa9t5`FrztWvrVIi}hQqL@Tm$DycE(3DC2VObf#JQEL$z
zzh+RSvJ71%V{r5^C-z%EKfk6>bC)^BfeuCxX?I4tLK>khWR<tW&JZjC%Wr0#5cLxF
zV?xpKgRBBUtMkxeH&C;dU|ob7+o<VlS6N4T{5D|ckb#4Tr;(%t2gp3lC5_9|yHMK}
z;p!T7V9aY{>K}_Pzb1Mj(_WNFWkEeWKSvuCAG`^0>koh8WT-z>EIYky?w7$DSQsyV
zvq!U_LZm{`Fv(SId#mp&t-n%G;KUWO*c{+EMU(@=`q47r@$I}_7qsWd^udZlgT)?<
z;1*s6WRx$m*C+H$s+?Sn8?QO2@|h&h0_FeFLOHqdlZD$yO-cl`c8yzt^iRtIeEan9
z5OPF3YBk%pZ?ym$`a*T^^LNP}lC|;VMEUvh1s2OPzz&ksX5^yd)tP-w<vdZqZ28Gb
zeYl@p7mq6BW#l=i`gp8=!4)?)UaMGV?&x`#)3R&S42`rrus++J8xvuDxuc^By*U<b
zNT#|<(@4?QTp`+;JszRDQARJ%PGhPNLlOKwEY&i3k~c^~h_XW^mc0Fm<9LQdAArfu
z*WQost!dTkOwpDBj~xJsSajEGMCGk*FA9awUm)mK(A)OY<X=Kw@P$x3WI$CSRPK)i
z@;mG7;&me*#KbJ7T_v;U`B8DZ0!z`QhxA=~W}m)+#fI>(C8{XQsj$W>TT30vX&I~!
zY1HIQ;kcw3t2FyUESpaJNBTQB1Dru~q=iNm`9;NjJ|MdKGf|Xx#e$HRhG$0u5;*EI
z;hVpQ5?M#&K*{J9cWnXOAV~!B<%0vdBwVNY$;~l$eB&=F6oW^xZA!~uNc_L6pmzrg
zsvLYmMqMqRI-z`PMe=|GUYr~+E2wf{S49y%lZ|k-psf6S-||#%;ASdvge}p4_AwZ%
zXIwZpaJWLkPn<RjM0-Y>udH>-tZQYy#(_6<iXld|$MbumKR<XM0ft2s<zk{HwCesM
z)3>Lg-CmxXir<~#FK>QFx1Ld<h1lY@<kdkXVzu#nUi7D{{xaYRO$#1E;Gnj)e4u29
zKtDvK>ASjU4^?gci;bh@fT?L3z`QIPf1pGV76P!v>4^Nl>5vu`26$#Y6UB|=Jv{y;
zFt#Dq1$;gPi$_9x#~;*<uQ_3FiD2Gtoi3+{z<iURuM^4Rhud9y;+Sjbmk<_da9FMF
z@U$ZjKFMyIGJfr-AHSJ0BfI;@AGUlkOkwzsU#y%sEBuw&wh75gKbzbeeI<F&mGMeH
z=!Wb{PWbxA%v*~2N6#ldh&br5JxS%aX+u_XPxsDkzH-9v)4uL6*X@kxaSaN)8B_f@
zUM-LxUF6o2?3Om(IN%H*Sx%EM>2TbC_?86-AsL`1n@T153P3g3#rgO5Yo{9pVM{_B
zG(l~>s&T=uD9#C@X?zs9)<`WNMq|<j-hRFAZ6r>@oU@G)C=#@OTFR`O#iqR=poBFW
zRzEa*CNfcvAFEIj7uIY%m5tikm6JK*wvrtQgCq96xk}Q~lFL2Ds@?g3VKPu~Zj>|q
zqWGZEtdG?E7-&IbG}GwQHC1~L?nJWkORp=AcbmS>?;LUtVum#LrJU=>1S|}!6wPt<
zE@8>0Xs}e;!aWS3UOVK)6d5`&OJXiPh_Rolr4!E}+`gdg($SFYjLZ*F_sAe*3s{ZL
zx1~BK6&H`jn<4{Z{v(_Hpd6jj>Wkx55%&kklwlWJI*IjhHH_WB?p|>%lB)!A6R_rL
zkO1o{7A;(80E8iOnph9zxPJ@e*BKMv9YfX|op@IlkIo%2wuWuB`Q*c~ez)(M4pT!C
zbeD_Pd@CT06sS(59*621*95AEX&Zq=0y9vXsKBA}RbbK2lMP@A<ad%hf$EA|e2Rf-
z$+pzlTv?gDQS-i<aCNv*++P{jOi4Q$>j#X>3Z!FxoReL;k-VH78BS&%-)`|_(5IYR
z-iJ6AYrmH<crkI>FhGhL>0&U^<lzE>)d*5?nL9#w;ozjcXSxlKhH?X0P3ERj#zpa#
z^DTs<NhGLLR2U5&P~wvVkp%>{Dv@`?Z<OpUPdQJjXLGu_t88m)E4%9Jmx;oBS%h7E
z(OkX9ON1UD@cDB7$`&OF+u^H1)_HSHGjrCpfR9)ed4X*ReS0*os|FbSl|Jv~jG(k(
z(+*AHaD-j}w8+h!nQlewb~0JTu786rM5<Y9yymV3z=8ZcTwGy&WIjxB48WP8u~t7M
z)1M<APc@4yDgw0@Z&aU0%=}6Zi3?=k@?t9Y`l&k?FyU`r1p*OI4$3f8d9dqltK5Ec
zR3J|OLcOywY#2!FR&Dvl*r_wl&AIdNhfK7Mk9*3cBscJA)8c8I-VQBPbnSuc>XdvL
zWt<sEa0#nJ-3KB5MNSPRPk>inPrNb|a)QMOj>cXAScRj>8D5k+>Gy?&7j(^{r=fL%
zbLNC-pYHCvZcBl6rXg5k>{BetOPECVFNy(f2fqt6`SJO&eWoq>-*I^6LEQzFZ~=J{
z1X4DFb4Va*QCkf!k9aDRQWSoTJt)+ixueuTA5+y78VZ4+kpS=*GgSFKNb}T)lioJz
zOAc=!TFK6?y!b!_t|8!ua)q8-Z|ioK)*OeA$7T>IBiczLF9kEp?Y<G+g9+)L<g+2y
z2?}$%5gA+n7|F7M^!U}#?@K0g8m6~=elJgKq&0jM<>WT;=%m)dZ~vW_Bt#voTF;sV
z6~15tKEYDN2~>~+dbq@rt&#aKpCBBuJe3#hc2?Ka;Qn33$X`Y1jnsHh$hFdZDSnd_
z`Hq|4>U&&q<MY^vxA2+8m<GA}Td_d7cXs#-&73FndM+77f+X}Oz77zFWPOdrpBD}u
zz<p*4GXar@*2t5U_w#nvNb@6udgn@ZzEGEev25WY#|ey?z(S$(+abceo9v^#(_tD(
z=7{xsdtXiI4X^k>Soj{CIWv#r&oWR(wj($SPm8cNq*O<DRo}7(nJv*w>Ht-da&+_e
z*beHMXdj6W6xDIkSW$KT(#_1wG_*#O&<`YBU0p5yCC9`RA=D|!FLf)s>TgC~9FI=W
z4nGpL5ybBd(q*gC?!M>x)1*n0&~2_GYYUuTdcD=!a{h%;=Ed+4O(ER@77fy?ur479
z$U?-D_q6KqTc?!ALqs>gu86LCR5O=p%}N%mXEVF{F+6l*M7sm*z|BaX<ygvF`L(hc
z3R9`~gDyq89&W7SrJRzP2Z_Ez&qP2%Y?d3SCgbNvN0M<<bnW)bY&aJE1Tc)Mp%ZIw
zNkK^~k`@{kmPsymA-9fZDmy1f@oSFwK=S&*$T^k#&mQ@ZV~C6(bnFj;Ad^j7u43#)
z4gS5YzBp*}tu08r<Zp2JaCn@_q8NB&$W#FjEU0O?=nrucq_5w(GhdI_<PADCckYfy
z_mX=ayL4NTb8)b0$5M18rfjHpOcKT7R8s^V-8#TF=*I|1vVQ3*cp+K1(-`V5&Z(VZ
zWjIRf9IS}>orfQf6Yfjyfv7Eu4whI~NOe(cO)CC{tw{=BprIBT6)520IP*(8uy^kZ
zA8B%h!pN0$YGUR09E$)FDJ5XyB5(~wxmuc!{8S##%j5e|#XGlABb>du{AxC*i+_ci
zPp!_m1y)dCy@&dZh=JW!NUa+HUb~K&$kCWQb>+zy%sw^3?<iML*JlhXvO~v*BX45t
z%-N-}A0^5H^;Ix>X~19%3mX31iOhshhhTng)<)d?Yq&Xg#yR4t1_YC>Z3$dv(bYF^
zMId2IG{c&0s={mZs3bxe*<v13R!c^RmR|lM{8YlCm+mJ=9o@?UPy?tH(}hVG7Ebp8
zoY)UoiFBGsAdl$J;AA{H!QcuAGU8-J8F)DxGbO5b&5IwCf)5MF5!IgEi_D9*N9I%G
zPF&-qnOJuXuwH`)a9nS}x&+r#8Ej$7mwQNqCFWj|*KlMWyNyQ7WD}QLDI7tXSPP1C
zNq$Fiex$J`AgN8RjzfD)@dcp7@?VF{U!@<S>BvdA=ns=I;Xh}vpSXCXN_#Ze>6yq)
zSJ;ZwlCXciFlTAl!+TF=HKE}Tg2J1^!X2LZs1V~-U7ZCC9VJHI8@Dda)n^N1?5^Fq
zm1@c1%73Pm8%Y%k1{Gx^5H;4Y3K(F=VqYI;U#Vv>`(Q18dWQ$jxYywkF@)t0kON}X
z+Q21>D1L|<cuRI^D1G`-eE?_p2|!8bPiyp25bco`AD&7y!(W<)b^YMX?=H?nD6^QU
zB#O;A3E$|W7BNC67(+^}g5V*o3Ka_ejR_N7ZVu$Q2jGzx!1&uK#V?OH`26m^a)sUJ
zcjrlWg_G|E`h(rS2v>;1A0GBt_*Vrp;+J83M)gJnk!LZ#HYq8IapnphhHB3<)q^8A
z!B?@A<S9_~-eIfb+FPU7?n6dg@~~lN8RN#-64hYNKIxK&6vh7lJ2nUbjm)kD4kD;%
z1NEBA#FWn=bx(TjHm>*MxcC>f`H_MG*l%E>!A*-v#|hG5;M@W9)}TB=3;xIYs&Mlp
zJY>ai16q}z*EVXeD^#xFe+uHwA_j-YtLLi_Bh8rg8p1>mI1OV*X~-->vkTE4k@@I5
zULtyfkdfESk0vYZqeC2ydYolxQaQ<~G!C<DcI50-nV_!hYfWTFXR|TyJe$tkzX^+N
z`k6vUphWBhQ1MHbkTF<A2Bk|gDJ4La$zRIh>Sw_C-yG%E9o6|ZS%n;063sMxK)h2=
zeo(J38<(O6K|NfgUlUJ8+F~TE=dk9WB_mzq(IX|Sx4Q$BXa^cPaQ-kvPvLTp_v4l;
zq~(#*mEE%nAJ{BYa{U!p1IP<x5_olKMJKpu9jr!oTRN=NNMlx?u6P>(WDNg;BkpYm
zTcqjx2lL?a*7L~6O&2)Ugp!_5(+X^&(b&r-8v8XgpEPI7ZA=sp9kde`R;&kn&fB(a
zLz1CYfzt63>oMvnku}7)b$$?Yd#nEda)4XcH8(%<p>&q2^7KWQ-^yP8di1|q5e)@2
z7SV7YHp8<>qKk3`BE`SvPR#rvW9PnjU(~jU&F&hj*lR0%g#WaOlqc*TqK|23<eA#S
zXhgIaC}#Akd!B!d6<_rlU0VS-rcO`QPERsHT6=VoryC8Q&fQ&0f&7h_Et>F+I$N96
zpHue^WfzPm8Kh%uKi?Dq4uVPmsgCV+pC@xTe0O*&%D{G~fp%$rIz+0)$K2dJb1*dK
zCA^MO#ZQ+#*$ks_QM5(xl8NMB_3)e{yTFw@VB*h;P;id6-e>7}z;XdwMmrZ}jK$d&
zPem!L-jvVl)wyr78PEXUFY8tDkDMREm?s5Y7jTdC!QaIs>+lBRLn<mNX!=82&&39K
zrC8q;kD(JQ)$IIvzuv}^jEWktfZd`ESN}&R590GG`a(pdg%5SEqyjWu6mubZ5J>>~
zxHJu?n8mej5e3fI?8B|nf*k-T5;qR-7G}71ct%ql!PJmQk`N!rDf!wD?}1qBo*cd2
z9H2;8>WF>-HF$^JpUKKB`_SVCle0I5DdCQ$X2=0M2T%+cpR@xeDVn}rfRMR<5&7Ey
z6@bE0Y!)9Q3X4cMfIe@+&>9Jr>)23PeU{86FhC=l(>!5aVT`aBgKL5qX0&ins>-p@
z<`>_a#mplmfEaj*NR&)LdbzmlNCi1vMK?W<z3CZup^gfjig2;f48wHmi4!I;gCa;-
z1qHR=BM+YWw|t4XH~S0&)Y`DkKuE@GrQAjN#m-07oFCip6uXD`D1bK05gcyFRv1lq
zb@jaYsQiX#j}abucXkLyknBGm3xI6T=5TR4_Tu7|fo{A`q6D#hc05{$Ij!sYwb1T?
zNpX4N^uk|N&O)Mn2@W6tBT))uQzi*aq8ZF32vTbqH(pNw<`YC$*N){1+f{`OIA9FX
z9W(VaFHDoz9wT|MJv|+b=<~?&0*eN1&*X=gn~l|aeHTrtL^CKNVvZ$TJ3h30dXxNx
zuqox1{?610ZfOEIIWpN`)oF$c!hUFe?b1VE8N$P3eokiS_X`r1ayag?;2xqeZq|$^
zDHUT^Und=pagze)=Jd`A`FP*1E=M+Nxes9j+g*#812Wb{%%w>^aadqcCH(w^V4CK*
zl2!&?z}4-rU(Dce3fvCZO2dWH<PtD~jh9roU7{d-vPO|ZUtIm=|J=2&^_7|Cz1?bc
zQQLzk;N668s%VOLhh5=$fAhzRph%P}B<)Lnp<<2793qOsC2SO`hbud^8DJ!a`XtCG
zAgZ=R<`Zxr-xV?~nhj=MUX0P+eg-!$Xl2lRktH8nl8`mqd?{RItD8a!`3<3F{(;4a
zjs}XOMmn}RV~vYT9O`@nzw`MFl{3`(Q?VD;Ht2e!;iFcGu583!+5Q9}EWj}#9~j?}
zD22VQ8Kgp_F}#&cf$gy~u3fP8!^JID0oi6^2h1a4&oRyR;K8!rCz`#$)No+9elknI
zTC8Q6#)h%>z%vnu5x6mcK$&h7yZ9>PDcn2$#Q0AH-S}0xg$v6-W;^bL^8nMsH<Non
zdj`_L6ID0Ko+px95v25Gk;Wpd^;1Pk{v7Q)Ye~m7l9!i<^_f~abkjCAHb|`r^`U4K
z$aXeArZ%FdT06d4Iv&t~$Z;ZI`qLxsvqV8iE<#L$EkflSGH(8o%c@%!Z|~}(FTm!S
z!!P`4GKV8R@-GAl2@x>(E%H!q+e$M5Y~_?*#-zi9-S?m;5%y$^JObKV3=XS}*EH6l
z?kqa6ji5QwZ1D7zrzV^yuetye0eyhJ=+4h<BDWX%_ci(<Gu8^0hLaHfByXMIuiWB|
zPEJn3E(2~@((*l{fgngAWqMMW;)GzUWL{%iRJ<}Sm5Qo)0UcC}5C4^i>p4op80fer
z3iU79P&a?g>+O!jblzO}&WQo%wexw!=t^O_74{ckV8}m@9H3wV&yzIq9L>1sLS%Dn
zB-m$&Uw3O9=eC>Ordpcp;!FWEuTHc1C_+T_GO!|Zi7c&-(E3aLELNxFx*Q3F1i5PB
zc{R&Yia~0Q2UvyBsl*mdcN6xpx}FeK3cy8)*HRqG;s3NOK9>KJAyF`4w^E08cYA#%
z4J`a~mkx~ikap7A7}x_zUM1iIwsG-d>o@D<%i%2|xGLJFv)OBF?3Ie|u>!#lsE<FJ
zVWG`4ZNq2-%{me9jhz(d*lLCD+kGN}9=hh^v!sfG3j$0;J;!EJQH8UN_*q?qEPYvI
zvl}VEqL06TTN#M>Z}|(3cpi*})?_I^KIhI0v>p%~0F4QgH&fmPz!p`Z&1{WJj>BFf
z+7sCe-J}?7^A#Xp04-!Oic3kCs?got&%mc~N2S0$)Kqdvj7F?z(9ok363i?N%G0nU
z34Nk8kj+9kac{EP2^BpW5=51`)31mQng4Jyvp5=i|20G6{w)(R+|r*IT|h|mLiamp
zXP0MSyKLFABoE0Tq1e;&O0xG19&JLrC3);1-M-kYz2iLUL5q(Y)?8q!z^JwS20cAd
zGSJmv;@u*xU$2gu_Xn6E|MPg5NJk^oxe@`e33kGqc^*;*^-jq>pm(hx#gqf%-($C_
zxgxVo6Qm9QMSpIBE6d+PA!!ccJFCF+FR_6~lJLrt-=7h~`4uc*AA2R1VPV}({+`8v
zV&qV{xY!qK0vAXCe6&cvoFh-*CG{M)(|!u&0N}*vOc%k(?s@H<``Ep&yOLdtY@%B*
zx^V`)aUX91LlfApPp3BgjxIt0&9L|Yw@z<2UqpF@B=3VH>XDsYB^i=1tVY{HhoN2@
zyhL#_hP=uIvxG{cA_fQgDMyr~PMuP@h=+J)i?+76lxyROe7NgrfLut1(4<BMNdmCP
z>oXZ$E!|;<ur-gJrnGo*BpPj@JDUc1Ol=`GS@Ak6;cW)3I{zh2Qcp&?8@m_mAZCm&
zz_99~194KQt&!eMiU1}e*f3wek&Qz?2Mmcsn&wAP%0)#=L)4ora)OzQY6XBfymYcc
zwt^d?R{Yq-8fLRfIt_%Zf)dR*OyVU27ctS4#X>qiHVKGpiK;0LeprEL?#Fp6*2gv@
zcI$EX7NaL%oSV)gAQ}_E@3S-??EVk5Noh14JB01C){kQE{@G2;CD05oCKoTQIMIZ0
zn}|Qi0T_=;zpfR6B*1YWNn$64Vb7*Y#V<kb2T!WWZ_$iyOqytoBMPpBzmlCv8H~dz
zdC_e6o#1aBad8-mqnT0kL-YzT8s?a`WnEnM27Cn~62cgbM^Z#4Ungy}69;p2(_jfu
zca^lBmm%-HTeo>9h@HJ`RF1sH@ZTL50Krg<$knqwS_Tw8YF$AAKpl#y3{GkhGT$8W
z8Phcb1LAw%$EUB8FFg0pgoye%21%Ek#z7QiL~;IaL4d<8eM^()I9f$TujA>s;j{Qx
zq+Di!l+L8E6ji(L+5K;RpNb5;$H=`DGIw^wm@Mjsr*4&S)O`M<s!N3LQKd`1v_cIs
zMAt=gL&EKZX$OO^Fe`|bax1f)ES9saIZsj)FAVxxSCN~S2YitsGz^h#P$-jdG<F4b
zrWp&Aj4*y{JgyEaZHQ#=-p8<qRM0UxB2K54sT&F3nEjzKrv!2l5d)f!h6ONojfoY!
zb2}8hl3^&}ok{YK&oGKmy>vX<YC1iKs0#x0Y(!dYWr`|yR2CZDNKCk4zoK{;pjngi
zccCG|>g*eTq%IQ7(GUMya%6W16Ehkk`h<1_KB~yy3hkVdb>mrCAV5);4ps5!1SRBL
zGEh<gu8CV>ck46O-5`GLf3)d%tg_A>7APAWS1Ibsi2@<pEy;==ok&mRG>L<8k$`^0
zNQ2WxhQ3Tb^S|2l6RKS7rQtDuVMWjZ3HlJqJ2!6Kic}M9P)<oZD}uRO_N`2R3xcpI
zr6~%yDM4Lu6h2~T_E#ta%AHw9RL%jBOElA{Da<~;{rn`{;uFknQb?*5`45PK^PvG|
z)3$_P23kW>5|003=pzRK&-L^F-@Hh?EVGp?hT0>Kzk#nA^%+UH7Go-Z?MX2f$BinY
zkspWz#CIP#$EWk1&P-LWK!Ur^2#qCn9|uhnEQ<JHFcM^T78VT{X|EA^5!pvx5XIoZ
ztMZ`4s4%+_6QTJJwniUE)C;`;2XxU{*H3!QH+G>B1?0QtJ9crY46DnPXZaWLX$+4+
z#wt*xWEp{+45b1@mutwm4?-RTcp8D~y|_MB7MXn=R1l9&cwH)yP1AwEn%(VUpRvIx
zzk;dRv4`mZ2nK=u6OCbv=UMXh4My8&gjz86M4%<P2T6%WC26#l6mwQXQZU%LO`x@_
z4g1$J0H@?(7CI0*sOFEJ5=IB;?<pxc&15p}RMX5rWu5RnD<}hF#srum*R%~L3$kp1
zhC%wHlP#!nh<jt`^(@Q7paMIlMcuz&en_%s7o8y1pDU4I6V24%f2RX*6sEwKSxz>L
zJCCYw))vLZB==~#HhI$QZf6j`*^P{2?BerJvmrl}a0i^S#KZs^2xSr#m##80H=|dg
z5upIvcl+r61Mc#g-FTVAGg{%i3(U#`L&EFX{7KVIiAoCFd~}Ghw_p5fPFU<&a^wk)
zO91(K-zf43RskV?rN7M1;kwy0CXTD50TUXo5*9xUlaZ-pqIoQ<G;oW5X5;3Y{v~9|
zWj%f6?KqU@&td?lW1c{ag|h|SXyughdXSlPln6$h&a<GGKcCsD4U%Lw#S{=3h%pL6
zM}^(auZ2o%gz*m~_Gt8qvL@LxDK*-UBKeXIq#}Ocp9qr<J{u}-`WPvpQ2E~=qrs_z
zwmuS2CG1bt((hlrIy2Gi;S|!uF_5;O#Ta~+68LeS=->z(&IRHgR18b<7|^ZXx@C=R
z&wX|oqjxx{Ag>w2{bWL8{3M_=sM_&L#CQp8l*ltTUh-QuUY?F10@DLPYurT6!T#kG
zaSjtCA?t*Z*>EEe(M^1VYiFunn-Acm6KO}ZNpZr<^+AH2$>FSKs<p)zbSO$75I+(b
z9YH3nrZn1%-V{PF_3iaVaj=tcf`2DW;ns1)?FA_TAlcdkN&w6X{+R5Ti*&<k>}pJ)
zBuDk{+OOWab!)nHO-&6(@GEI(160GL#fyY&55V;VIg?GKveOic1CD8POrs>63jgpK
zdRM1T1#ZKm$5+C!4DDrtzq!x(WB0zSYjgIqD<c2Fi+<HqAtko=vw0>)`e|yr5~T$?
zI~`;2;JsPu1-l})rLp*xDWwV0<|6k&pTth%paj4E;64D2T=4wZD8Z>zlD$>xIE*Ss
zp9iNgjTJVJ16#s2V>1>o;V$m^?|OGmaxhKLBYerJZL_KqKJYojO=3#PoQXw?Um5N$
zL3%fyI2{{;I7huaPd-RQOM86~9qlypIE^EKSQf-%X#zbt>c(DTKFMiSCynF?4!npi
z`;Nh++)-@0a^Kd+Wzet_ByI>&I9o}mS<rxh&~Ly=u&kmvI1-pJ7uN1Wvpoc=akSP#
zIQ8!`#Z=)58{^ks!{jHN{=_7Opi4Isb?Ff5kXj~D5&CjHDqI+b6Q~A)W`3vdig!zP
zyJjTv$#m8M6W<Bnu)?MJ4fsPyb_2JK@_)rB5?zB2267e%rF5FKp%Pv4Wi~1tjn0w`
z06SrpzAM(;0UCM`opl6fP=uNw<TwgTnHvwT_!laG>UrGyY0$@$X#$l_Fk0uosMW^}
z@{Hi1fgn}A50b(Q2TYDt5@=B+M{EA%*$WbIJ7(xUh36q)H0}(>58ne1aXRdVL13k`
zXz1!7k=uPR^6VBUhOv5@c6$xNjUnr~r2$O!aX;FR^M*Kz130&V^M+vz%5d>a!|xaD
z;vT}qEmr;2J%Y9&jixARP;nI)Y<AHt*D~AwNIH!#(wwu9281I4nfy=ki0Us`n1&cd
z9B<Z4^B|yA=`=ZkeZ&BaRA)WEH=Z?Oi6_Tu2$#k#7m6{?{$WNnFo}rxZu#ViLl9L7
zfkFy6Oa6MQnhCp!3&&ch&;TFkdgv20nA6#CMAektI|1z8@rSE`zq?-JS?K&1agD`J
zyNpsNPoDX!k>!VxKd+v<pipi9(qATjb5;B93X7*-uNx?p5}=f*HB@WqMy-=iro5Rt
zd-&ZgGG7ngxIk?0OD$P(&iKRgN&|W3f!At^9d6!VxV83vdu4M*s#HOIZ(RGtJf1_n
za`>La<EyT{35RXtEuB7Q6lGSZ1VhBQmj*IFuUQjv`SSCh@C+0TK<SLc0Y67+1m{){
zHq&T~d>zNM<{ZWbQ$2aG(3t&VFGjV`{Cwqy3mVDj!o|mo8KdZ5h)1z%y62fyKYoO}
z?9|hzKMg@5ooo!<euSi3kL8l47gNVa%E`$|qa)sn5<rs`DR?~i;wmLE9}vZp+`M@+
z9RY${#w&bu<5jv@EsDWFnI)L46+fIg`pdw<NgzWHOu_GoFB&vxkR_}xq%?n|`2nA>
z>(`HBqG@3ubTtLY#-oJ<49Di^K%s?t`SR!3Yu~{d=&@<`7#K>A!?|2;Y75KNC5Rhm
z)=m$>KVHFWO!pt0J9GBz<CG5J1&rMaDLBd5!NK9&`SXcyG*u@-1K%{6-T^+<YvgdD
z13X*5CNGSx{pp|*{DI@opFiJ!buPhHqP2egF1r8hZ3r@VV(;NppMcidEU5>_35cU5
zI|*lsI`ot=?EQ0@`)^%^6(VQT>rV$sRR`#~5Rh&Lr&LRAzU+6$mIuI#^a>?eu>4{7
z@U$@<KaQqvO^ad`^?5kf!5uwElDS*=`@OF(A7kLa<V;0jceg(IGu(L~*+v2n!G-vJ
zd3ibB0PyV0dGk&g$czVNdkQ&JhwjWF6vev{$3i6AVP^3*Z(=uC!DUojEx|CUTZZS|
zr^P`a1a%%0q#(6P_do*S;L0-^*Yx3y*3!}O^^87z`0y)0>s3E~M-RZMy-u`5vv5k~
z9&(mRTa~>~uqaVWa-MzuIO!*5(dA<qgr_E9Eo`iVhrh+?Qka9A(}<ob5gS&Ld~Vi@
zVo;kbVSQAW@jZ$6pgzgw++p0-`pOFsU!}qSfB<(Lf$$vP46WmYzPsXd;L2~=>jzp1
zuRmGy0Q~o4>B2h*hyFNIN-8tImQDx5W}13t)p8uUxDc70_SAJddaRNzEm|ad-0FdA
z?h=B{63rsJr<zVPkJW*WSHhA^J^yne?0g|mUuIYj88RdUofh`g558+x!qO2E9E{->
zT)7<t`#6Za?qqAye{l8cbV6WsSQM&&8k2TT8;kG7597xZK?sba<{pTMdR2XzDD^MI
zfq+%0u|ImgtipK0z5RI^bn%k<haY~QxCiWTI8IkNnSb-<&F|&Sw&Q$`_+w$1<7GGP
zl8?ffLbZA1x7-RF6RuAGaK<|0>M$@Uz(z@n0;=;8^xv^x;PT@~j~xq=RQ4l$LdVcL
z7^l@2#>kUs?t?l00E?&Je2ZF%FYb~L><iq5S3{T)_TVjQ;anom=oKqgSmGcI$_aR&
zWc!>)*+k_C5ajdT_F_qz<><k{VwwP<(v}5U83PG>^JKa_hf_9yNod0lA*K^=g*~2t
zkSCz}7U}{r<=$66&`Hmf))18-oCR-S@Y2<*R})yop?Ko-bd!3JPfKC@JBrP_1b&OF
zYJ%%?u9q_p;KiAFx(}Yi1xJq;BaHfZmRKP$^-RQThwaqbUR$~wzo#&oX2yNca!?y%
zTUX#K9gpqyJuo3n_q+;TI*=)bYV#_FZKqe_F1w6)DMBmA>?^3Yf<?;5BE14(ig)k|
zp3n}=wi!_3IX!j9ql3p_S4$)p1#Ow69vsy35$B8?2dSXGbm=clS_t-HBEMuZ=@2xf
zfW(6b-2vQ%Z3SO0Z7Y83%x5Fj#dp)iEqt8eB=P{CK7%a9*KgiT@o~dgNC@~w>3y$%
zWcSp(a^d4mh_AxAzDH3%R$^yE-$+D`N76^JW+FdqXWXTSZ<ERTw_cYHj0F9zXW(*h
zH++h(keD7~*8Lnam2m&Gyv+doAV(zx1S6J29**@DN*CNA>8*s}K*MO<xN*str}z#9
z5N1#FM4Y-L$j0kWAXO)#fB&9Dr`c}!)s+C}`fv=eA-|2OQ{1_*_syUji6jqNU^MwH
z@@wI<Q%55+)pu<Yq^e6yX`SprUR*d(#Yyo^GDR<PQc_g~ha-&A>wcFfx^cEb@e{DA
zC6W72porikyi>=I9}h^cL-b&yA{=EeG0{TL{)IM`z?#sgBsfy{VCH3fGmy9k3T7hq
z*LXVOSM($sR!Y&rB=EI1Zysa*xYG!ak)Q#h3l#)#h?oE>;A9XGw@#B5guBfuOG`@}
zRU=D$Y!VW$m$cr_of1@|U?F^0Dcr1_^=mMcN?+OgG0c0ClsVAfc!jsnNivK`)P054
z5s|31ZQF!t)20b8z7?mJe8jP=HlzqO;w&J;G;{RwDJdx*;c7XFwVjE+mcE9|3U@2l
zu`<(*XH8b{M!2$QSA|4G$QVb)9Q;Jjlu95hFIro-VnPxD<3R^aHsYMpM7UWdk<#I#
z7iG4COvYqk*_(a+46{Nd3z%5H-i(~L2gQjJAf2CGisL$w5ae;DoMT(<_+!g|K$00f
zN$KESy1DC>N_KVPkT~1{N#$+pi!Mm>(S9exJS_op_{mF`eqiD2ZYF20pFu?yHcAOt
z7>+_mnvQWNPdo)=UG$SP&b~CDxeeqJBYG1F#T}4szoS!n$zn2l&K%i_aVX8Iy`MDU
z94u93gXqXHVLeg1K*tgHRau?h%wdBOOFO8LAY%q9JFUbfrxM8FZ-V9TBRbR*Aai6)
zOiW}dD${&N(LoAi@6~45eetgtfa7or%S}c{#7-l;de1TN?2kbrzlF!>%$z?#5RMAY
zcK<yQAZJ8Wk>ma?onYwIp(K=ox!ufMy3B{U>z#BqdQYBPuQrtu*ksOmdsmv;Cc94}
zP=?UFbno8635!exPc|XGmaN_6)<guKm)705aTI#LqeuuVQE=ELQ7jXtne{aQ81s?5
z$PW+>qtW%-x6@u0Jbn7K3j6lQ?~YNDL?btI=1jp;EeVc+rbH&@M%=1l+LkR_k|3~*
zC64A`ebElK$WCRJ<$eZi52es8k73&qQ}5jJN*zfaqw)%%E|C;gLbM1001!mJxD|4<
zG(5R)(P$E28Q&b_+O?0o=X9d10@QyN`ze+%?_dtSS2V`rC{yT7CkT;QzdG9wd~gHy
z0~6Bm&e9lrvBw05Y(><8_hX@QAH}{^Li!0&qsNbm^PDMD!~IdnDOchhK-6$Y0u4x;
z2LMb5tx&{6D?d(i++LJnv%f`AJ4qug;61;ivq618in!3B26R&XN~|3F4vsJ|M=gUW
zcrpzmG-+8Afn!ibP>Cox(7B{TF~syP%3KocM)<xr(@)SVnVcU!N$EKbH7z9qwD#To
zIr=<M6CSV@C1DB%wPDU~dLyVk*E{x!3;wc-J}bd%uFiLaVzFdU{*j@jU>T+icRMoU
zIK<-Rz(MmUfxQ@l=W>cIhys||L%v)e_!@~M+{A8zADE{-7;voy4UjrY7HZ%Msw@>_
z$BbElGdHluKyC$$1X<<5^g*1a4JcwCalpn=sK6^w0+M<lVt$7da+Js%w!kb)5H><3
zdJiB{QNMwWU=Nc3#UX+TV`wl6JRXTWhZKRqrl*A1xJvqvda)Yr9WOOq&u-??W=2pa
z!v=Dmsf&Ysg?lWavI%pJNtkp&T};IFK7ytC8!Qr{feQ!M5nqNeJ-1^-%-`x6CSNcD
z*m)CF;`MBEJHlcAcv|uoKjUrTyrzjvYY3V(1i*Ag<cZ_QFR6P-5Qc(Pv=Or8OTk?`
zlFH-(G9j6KV$!;(N;oNk4s>1#(p_}pOiD{oy+=y|5%=<&Pg_AJP&;EBse9jK4C2`l
zwGn)R+PQP*wo+rV-h9mH(a#}!>!XkoP(Rs87!M1gkJCRIY2y4Bx@m;W2&QoLoN6z2
zQ$-maOWO<o0E_b?zZNqO>Ozc3(f@Ie<ZeQ-J$i|WiIxzY6X8zzcuKzLg-<b9&Hz|8
zsVXf$GJF!-dh&lTmnUSZrmX!pb$9;!`R*Gv6sO^FI@jnHAICXLPiS>GykqFwk`eUu
zY#ZT?Hb^rscqk2D*?3KrP6Khv=ObC5?DmWB*{&D+{7<k1DV|b45(K+K3C3AaSL~6b
zO5xFVM{+bui*L=<l?Lx)kC~tlq;2*EGG7YA^7;&zS0$z5S~gIA#tiAEO`FP0ZTsp9
z6@YJt3=!<95<QCo%j85LCCc@uhzm*!gFsGaF;fHJyhzynbAmjX+gC?xOWn^@5DRGH
zM`H&0WqvIjZi)ENlCe~Ehl_s?itnDFZiyS9k+&kx-tiOjAI%gL#UP|kcqNySyvy#X
z%qPsvFQXbO2q54dv~K{I0qH8>B|_i<{8QtPq=-UBeH%4ax0|T+-ZL*&nGfEQ0k8?i
zOl2vHhBilu;GohZ;;pKqrb)m`a_|`X65KQ)27oj{)<3OJ`4@+}K1UlKxS5B^z~eZV
zE)m8QHr|7nhk4qVKnT21)F?xgYXsqmT8NRC5Kst6VtW8gUtL@_m>PY~^+ErNXqn42
z`3}2KQw7uz)&bOzu#JUMN39wK3P{foq(i|?uTLF~y@Ug|zQbu%N6D<C{}`BNeq76I
zuq-n^UiQ@3FquM@!!cmj`U(XK_7UnWnIiCTtYxC^-d%vKIhJhesJB%x^WegQ?W#6U
zd()=Tcxc*mN_05PgC{$inwn<LnR9}Ql@d<LQWLxY=LM4?uTR4@OTjLmz^YTZjYWp~
zW4T4EAL6c*BG^q$>#=a9*hMGdF*R!KeEknR3;#D^ELwd3w~an{Cc+j!{ISvJRC6iY
zg7*JNTM4p`aT=x^-xFnFk7G73K@MZLk{fgX)u(`Qw*rVdW3N#tzJ;}Ep?T4gZ4XSx
z)0vWV91K9L{fi*|zfuEwZo7ByK3J1G4?#-pI}F!Y%8n#jqw9y1Y^q0lENNOr4dktm
zVED8oKLU_?43@B;eEdY*mal|M>E;o1nPO-CH5v>jFVj|NL^Rxv(eMVbNX+D&08mVV
z+NqBrDI^?q@pHy3!9w&L#)&)-%w@GT0PXu)>IxSVa7BZL++XQWc``V#3V$av5ctJL
zP1SMT@dybz>797HnGCha8!p85*FB3Mj6=a&ya!ntZOZwaAR6K)pC{FF0p6uXo6JA&
z9vtZQtu*fkiSQGIp)ww~X<SfBgb{$YA9}=5!zL_j7$$hDnyt7=Oy!Iv$KxfGCw*7U
z{#vF<CqIYn`XZwnbm$2YQLsjSl(`-`(i?FA)N&h})=HiewDO5G1WG4_kPkH(^KX}e
zgG+*VzWw0Sme7>y!|jJIUcWvQUfs*WKEIT#P3D4*iD!a5)d(gt5oCB|iYB~{Sg3KZ
z?>uJo`4WDzFmY9%r1xzJ+W9%e1G_9<24HFu0FWfjSQg-W)C9-#V*0ZfXuscnJ6mow
zha>a?;3IG{9_Qknuk7nkc@@zV2_<eg-E}*Zi)lpe3ve_Y@Q4rCn2jOwV0KMZ*)`dU
zQ?BUbRIQymU03mV#W=K58Xi0!SSD`cTwSz%WVNI|C^C#98@Y-g5q<auMht&Rq_ltf
zKG{w&BmX^lxP3xz-aHOqlt|MH9mwc6nHT+<iHm{MDj6CYf-Q7dRY@PrL#Q^uWk(;u
zK?xikju+9q36)#6ZlyN<C`#pzG_58u{m=&uQ2Dof`ppcct>ijp3KzNQ9a7Ur2#zXc
z$>XiAfQD$qiO5r#c*Y@zU$3EUQxd!nX_rg{mJk@9(+KEEnx|LS(b1t}tMLr4sp04F
zRV^pH`4_q%$QyAwp=wHMD*U-ssOe0*0gyd&ep|TU(^-yaFtie(X-U_P-2z%kA8mOf
zs1|k1b})cKEv|f>IlVwr?t!SnvE~>6DpI!7($d1YR&>9l13iXYf7^8WAw?S;8<fam
zXJ_y*<C{@ekAW6ShV5rGw#NF&^n)Bu>sev$E4u<(Itl0ec*<wKN5oX&L^|9s(Z7Hs
zx<Am*9@&hd{}Eo+qcoEq4qSJ#6CV-s915jAQUU1OU^V(VEW;#b8P?GAfPp-_+L)wb
zYKV}bskI=)Ku0`tx0RF6pWIte!6%{x7~hN!6)_J%VE1E=*4GOWn_5vP4#zFU9tyR|
z%z;6WAv3YpXhUxmd`!>%qTr>6KH|(4ot~C!s_C;;(Q@rYb)&;I7=#eK(^A0ha${+@
z^(}UMs6*iLA4bvu$vH2tm8(}D8$`+Dv>n_ElOPlvC)daMcud+~0-Y$mH)}Ej3D%pB
zzDe(y>r6vz**7acAu$sn?=DN<KAndnU`EsF30v~fTM?@YLT$Sx{WtVe6NSCSX*+7J
zx3w+_i|0MMLjUD-`?ueIy9ZO()Fj<*LO5B%Opy%9eg^x(Qp}R<B;C^V-Mt%Il;0p0
z=TXdZl`cMfsDQmaUfHqb#cY<nftt69!SLA*%uCZDC+lkh_rWNFfZDA{ub%x4we|o5
zD<D=zrVu?t&a1cg_j@y$>iQdDYM30Bo$DQo!#np0e{nok=hP7i$C>U>e{MHXq!R6d
z>cS?K>bXpCkj5T%KJz{xRw}J^u!bT!V12k=UxB5*-2_~95+|*w+t}P^U(}O*QB=*?
zEh^g3a^_OA!&Rz;;oO~c*5fBmM5ah#Hvdjs+qOiyE~?!OAqG<0xesGAp-xvXUAnZ*
zNr8$LhqoEZ)(IMIz=B7lWFo`kz*;l*H!ImiR>a6$1n+=Lv=uUmL9dk({c;&0=0t3O
z$P7*3MhT}{Y4#VdvA?(v+ZoNsl`YB}K>;&Qryd7fpTu$e2IoU>+&G58F6z!k3;Q;;
z3k(VFa1Gth{b~njV}^n=?X!o_3^08F!7Wo{#_hKGhT*oK``#tHM`+$5)WPfH$B!MD
z<T$|mE<-*je7^T4BE=09mo!c!vxj}3p-t3^go+9`_6_)B1-xIA#7H2Gz0kJ>F}dD+
zA=of@r$(W3PM;l*NBzC@p5>?_E0~}vWAoo)q9}3fZuW6w46?@x!29}o^dZj)F;#_l
zWXwi;(j~~wX#AL6IoX_jP9S`?(qU;wTE_$K(7}P>LutXQuduYh(W&k)X7kZaCSy1R
zY~Tj=7N)Vehq^Veq#(VX?Dj1N#zBH~-7Pq$JbORW`>y~S5ShZqQ4}Yj1t@_D6D<at
zAaXz>RAt|Iv)k7N^B>CrqUTXmaUH|J$>K~oyXCug+lGbECqPQuRBvLQ|7klWRLy(w
z0flz4n0ZqST8#6@v0}dsqPLR3m&OS=9VDsSxFUZxV$5MSr9@mimf;K?YAXEA-Ep*G
zK!!Mok{<Y0HmcB1^hm)uaGd;`HG&2m3Ut6n`VZ*I*_`6-3Mwxd?Y@;#Gu-CKz}YSS
zc8JSu0Tgu0#9YicvI@^;n{ZgUHZystY>@d`LCN+e|8k_xylwZ6k_XT%wsB?hczTtR
zAdqH9!9!7hxrJ@COJR7?VtAALfl)7Mp<XOTPk@HU=0zpRK$tihg|l~#VdKma{A9%X
z?c2p^_L{Kmtsatf;Z!4x$v{Q!wsfXm{|w4SFr00TShhq`70_!?nS%DMgoIZbQ`~Dt
zIOM8kkOcIQjRHjB{1w2t2XvRD@MH=9&8}z)w6<&+3m^Fg_D;8i3qF4E#SA(YH5Ml~
zjkm_ic+lQPtG@@zbvT@3M9t6y7=1z%G2Cc`m*XkGTt2puKHlY#UA+Xc`<Fn8y}GV7
z)sR*&AwC$P)`*Bt5^&Kt@Yu%<QMm_}84^*|5r7P3rWRT%B>fyhZr+@YBJ3rdheGxc
zL?OQ`Q(fG0IZeD{`j{sljYa|}Vx-_vh|fk4ZYSqD9Ti1AMN$uDyh@2FC1Vast>_g#
z&e%#%uj~}LW=Kaq866Q3LGUCgpj5EKEUGR5F*M?%gl5-dmJ%;XA$muFehsu7TiJ(@
z<Uqaw{W}10(#Q*k>0ztQJBuzAGoLz@NGwGUA#k%<EtHMv-N^9aSlg4)bS43azpw>#
z>Fb7MN1)@%b35t7AqfJ_VfMqeXTk_DiGy~TjDjLl@$=6=@4=Ze-_jy;x0%2YCylck
zZ$~h80{~WEW<y*b!=Tt)?11h)sB`133BZN`@G-w$Tg>k*E~P1#E%}gM9)d0|g=cY-
zINcbK&jle$$~GaLgM3qIlb|qwHu;`;*P>Py>!nkknwlDN`}SOwj*@}p!U|-umC0<O
z^qD;rJxK*<K1L!mwZEsADEkGpXfj!(@p&Ahke_kTA*sg6fcZ>HO4!ZB$NgMjGHPto
zd=}%A20x-ZS}(lF86p;tVM%k=Dd&U5d=8AXr!T{n+{MSo;N`g6YQQackcla62M*zS
zg)>NmZzb>(7&LNjXl>X4T9MWEfx>Sv0NpBsc^Fq<J(^5{%E@T3c`9@)G;@t=rU9&(
zdK+q4ze;@CT8KK<&*{a@oy1z7jHanhRP8Z>yyS2jHIKyknRC<X&Og9#SR&5gzsQ82
zRSNUc7jQCxJe_JMod=Omq2_kvXu+@BB!CIhe|>qiP9>m6E>T@$n5vq=RL4;3=5%>=
zl+UCBa?{>+3lXG&_1!?g^TG9;6~vHbx?J9KIAPh}*06UX$j;qnh!(jK3CcLF73Pb_
z__To8!pcM^g{zRO>2?G-E$oPgTjLz>Gcl-8#>C(jI$(`BQvEs{$E78}tJzE8g#Z{h
zNlITt3agFVBPlWrQNk2PIK7`l({)6{C2c2#9g8IlVWpc$gVGp%?4aW}Za2M;GWQ_M
z2O>pu4K{p|;7X7&B7BKmlR_wnt90k`<sb30<6v(|n~h>spG<)Rz`~T~lGAM9Oc7du
z0;J_}UY!DbZo-vsP0&)GGzP8TDe$O9=k;);l|Z+eJ(~^RrIo_;c4SBYG!Adjm%^w~
zhOFo9gE4a)^7i8`kvQ$=H-X(m>xG*rVRt*4fMj@x&q&*$C0mZZgt>{e;Nm~l#nih4
z;oWXRCul$i86T%4L5-YH?BhNI%{H5k)jb&vjPwvw*37SOf?QKNF1}+$ZYMPHWLOHq
zn8+JGK?o}pZqvzx&ru9LrRBl(AazDQrHcnk%mH&npUsHS!-*oq;4{IKu#^FGN7kWk
zQL7_Nnyx+ef_KhAqjOmS6IBr@m^4equ)`iW3)!32t6-Kbow9>HJT|fw<P=3m6DN3O
zE>&V2r??y~PGkx)(R0j~gGwruWS@NebK$lbegcf_a~ffn*8cwpJMX|MtE}r^h7nW*
z1f+%DtJ0+^QHme}F@#=1l>kx#k|6aejuIdcq)Cs|B+^1rAV5$<6)BQHKq5?N8bSau
z5KKb8-+Iou!I^nyKK}8Y7w)~!bILAjuf4XfZ&CQXd(Ah*)&`Qsu957#w&TRorAw9D
zs+i^3{RuWS{C+UuUo?hu?p|sS?a_WVE0+~YniD77w}5}fn>`*wMr_Adpa$H{#${7q
zvH8XXJnw8p>O;1IohYlR<6NMOm_FZ(DJLclCXWMT3$`e0Y_o>vbz2TTwpZ+|udet8
z6JfUH6UKC8wLt-S9tfin=N%^)@%%#rM93f`cpUv2TUWRq&ru&jOKIr|<ni;VZJK>3
zJ&iy!L@WlAMIdDG7u$7jnYAEYl5m{SAhL7!s*rOQoW*>;w<K;Q5p7Zl1}pw0l?e01
z#>)z~&Z@6|Ib$|eV>c!w;~ft3$}f;fI<J!HQ7A?m?eW`Too90?lsUC35rfz5+0$!~
z4&R~SLG;!i0hQWu;(UC(laeEzYL8IGDmEkzVFJeK`%<DSJwX95`iFMyFG~cc6jI|o
zE)rVoN!{_oU;-%55nN;M&B{G8UM$y;G>DpkFO~Q+(FvR~vWPLRL6@7W@r7Syf!}%w
z^ae8PYPUbD%`YFw5MK?`817bJ>!l|M%a_2apX0pl)l>CS2;07(JIlLIr(moE1qS0B
zoBn0Q7$bT$Yz5kN@$vCPPxdEm7VzLi5Ld%%)~XdEmjOPosyq2bdLV5+N01ir+=Lt(
zNEC`xYnTYf4SK>&zz_Ct9%7u&#9z-;Vw`CiLf-!~zsM?R4Z-@2?=1Vg`xq4e924p)
z6hdv~2tz1eBy`1wf-4=Nf4&?6iRTt=^F05DoO&_gN>2bBnM*CGo!Q1p=YWUyAd0tc
zd;ORTH^pfDf|;1f+G7~Gb?i@9B~KAHP<jH6BM{oT*~MnpMNiTmla6u<A-%tG*u)<%
zTB_o>S(2@jDxG6F<51g*aMrG34QejMrIvO6vDy;ojcsl}YfI7j{c&p=YB6Lh(85-j
zW#-T~pNQI?su?>^^q`glkysV+cTMQz{hBrU`I9)!CC0{P2OYUQt4Ni#IEv>Tt2{m`
zJiH>JzFr$PZ0LQ1=Jb*{!VoF~Z`qTRZkYmQ{RDy#^uKqq3GEs*TbsD?>h_u#FV!kR
z^Okg<PJ-qHPAC*v&*Rh8`i(c5&{*`;f`Z1*H1_ZNjhvbWpcu{2TnF!<=*Bfr;BU7s
z;xoH-;c7=#R6S&?J`$AL%dEm@+^0LqoV!peDnl20Lj1t1hw|JoTM*yK7{GC71SRI$
z&?C_KBb_Y>@1dM~Fak?Y@ek26qQ3gd9$%mN#sVFBBbwgXJzTFwH!+&oC^-)w)aU%8
za-YrESt%ys+4GfOY?5zj+xjCbCw$@Ex&NHNkcuz-@a(fY8r84)Q<R_gpilDT>HYW8
z!^7g<Z2W4akqHqKzwA-4Psyliv-hkGZ<c!U<j)KH?fv{nMynIA{xGe=jClt?JDomb
z##jkO4%9jH4*dz^FS^?Gn>RP6NCicPp`Z!0avySl7BqwENYNOg{dO4T7n~R^zL{)~
zx;E2ponBJ`Fu1<wuvkAiMWfsU(e1AmM2&JDfLa;HMMOlautv`Hc2ZZ`vjKU>82jj+
zY)dQ;dx5A5){ZU%c}}cM`^2wO;{27MDC*UP$4S@+X~_AfCM<}!X7+aAeASacxK}3%
zlOmlPWA{ZSS-tj=mpB!r(EHhsaH!|g-+MkS$4=kX;>1t@ukjNm6f|%9eDmhb8=x22
z42gqo>73yOYt~eNFWl5*pWfwcXy^a46WY;CH24AqD;O#(zSU~G`)pwOq8ZlPOk>5<
zML&d$@)|&TfN_lO7RQW#Klrr<olqnz79+|*(AQnx)i-$GnQ*MqXKf)d3*YaaL*y^c
z+7K-45xWFK_Rn^#^QaGL1CwP6>CZj7M)Ph1$1eJe+CBQ`S5+Y=W9gazv7rUFW@XBu
zm-PcK*Y3|uhBWnM+6P4<i5o-r=7PnGhuI~(O<XQvAg{&&VNo+^A4mqeX=?Yqob_)0
z7hzMC6lHF3FuDncH5#h4+4*APTM120P5q$jT~PHQ1R|a-sA4Z?ofYVdMQx3UIMz!K
z<-KaWFjIhzrTf;<6i++c$#MD%musTsR!^;_UsFqTND^IeWj3uGKijN&3(4h%Kp0s+
zveB}4G~xi4UFmy!m&e;+rcd>SA;J!$e>in&9>kBC&fiZwj|f=nqtcy>QkkPtn@{8D
z4H;+;YoC*5yL>a#x6LSr#qW|zBC;k<tVFY1r%v#7Sem1nj2A^*H$~Ji%=Rd|<9%!*
z<QvJ?6T`;6g3=96*q%AI`a{)t<MWSdI{K2s;lU}DGNrfh64ssjslp;s;NS}Yeh~Cm
znOC3uCCD%Lpgd_<$Anni(b-L1*T@myH>)?Yv~ffoc0t?zqWOgP1_z&cBmZISy1n6b
zbZr3^Kuy|3J%JQA@8gd@PCPK>8_yRMv0vZ^5<^1XDOK$WGafXZF~lsuFM?aw?%&@E
zGRmSTft9L?EI2-Nt~D~2dU$9lCBeRz8P^=rA6nsHR0{GrM>=JT>Ve=IRch3z(S3$k
z8&tCgdv@SEdtu^1Zw3Rwgd~BqtR=3spWdx1kHYf!Yz4K1HjV9Bec8Ih8QqEHMG1~>
z2}O4Z4WZ_vvu0&wWt9THOmQmxYRvN-(|_7yx@|d0&Q$V)kR)z}YVPKv88cN6lN40c
zmaSXo>C`~37-pZ@%buEpXFv*uK>G~o&&sVM1GJwmM!*dM8J{7oo~SF-P*s+tu+Ze1
zJ+(gO)ZY6C<hkO|B0Gv`Q9RVENyAdw*h4uBSF9+5Fn}VGYTf<ceYeQt$B)I(ENxFt
zK6`Rv@A3aUEX9gM*@awF>5g3s4OMJU@TCu;p3qJ{m-em0gxjHZ<G!~p%%5Pjy?9;G
zWe^6qe&dF@=vO<HB{J6GgOxie>IpQPSvos=Cv5VZofpm7$!MhFLKZ4Nqa1^|M^j)O
zlN<IXzF4%u<&8&t9_0^gCk0k=KF{H4W1nHoGVBFMuzC=CxnR|u2Sv3E-hC99YHKm#
z-O`EH$4O7jvx}AXhdZWnKKVnj*gq4t6^#M2Ji<i0A?Ef^Z6_iK^`f@$2cr%G1YxR%
zcy`e`swV+}WbgLxzbcV*Y+_`E!p^C^cS`^S{U<m$s>5jq9u2S-r|dwpcP|zxl7sql
zarI>lexwTHDRdmo52t=a>kdTjH{|5vfW@yd#M0W>LwPK7rAQH-5x*qjv``0!-d<8_
z%z@uNQ*!-CQrK(gNzThbX7p8a;1A`o##wmWQ6v!;CT4%%I;+i8v~Qc3V>2LZ)v9v1
z_u2Hvyw*LITO^*xeuza!jC*P3GABRcR>NiQyi~Geji8GcFA6=g&)p8=G*|Gn#w`0~
zbH2rQxzE#=F$Zs8Ym<M5ukK4PY!+~Rr&JNmg_RkTvwbN_e|7N-*O*fGPQo^LpJD;f
zO3}pl*+bLK0(iehlv<qHKuCWSyfs^Epf=r=c?_yU5n!P}D$Scx{3&)TAL4X)d7i;k
z1j5Vr#Yc<u7SbPl<Qh|ajgLI^dOUtxG_}T*2fOE>g)Z`06<BJXZ!j22`0A+5o9BN1
z`R8$IjK`i$pKU3}$qj5R0BA7hYQd^iX&+vJp1*d_o<!@iGA&6c-H&_DsJvIAZxhdm
z>1^(QVppL83dFiSf@@9q^m@tCe|We6!KDq9I<SIW$?})rt_d`b9tB(`@iY51r?qZ$
z?7f|IW!(1#Va>j6k6>yWa|Gkt5S<Sfhb1IQ2@aId#^xa8sZXPql1Ja&g23KS+&&34
zK8Dfi&vqNU`kV$hkJDhy#Bz*YZa{yJIR<4b@?wFAu#2_5DoWfsk&sZ^8)3cPG;?{}
zlQp(%7EFk`(Pb!gH6C^T<Ur>)^82u-%5cQL@XJ;nO8s*dB)=<@oPXe&VmXoN#{k#|
z<DNqL17fU&4Q5_HBG2+yM|<fQX!K4>Ysj?6PM@BCx^>QF&o?YEp-%cSN=k#F8iphR
zmC5eeZ>ZD<exelF<WD01<XpDpPGahQ@O*=XA=2a2dBAUlBndm?N$Wf7=w0v}?dht$
za}zeDXuP=UY<@j^dagq~&q3P5WShZ6Y9UFO)5<g^nRRI5!~bsxESCj{=|;`C`cZyC
zeV>}!m{`Ex#{9n^bgL|)Sj^aIxBK2l-zCm|z89&_5HifjOyqr`2z~TVvwRONt2g}+
zCEB$h0mgb?ziv)&_@^;cD~7;Ih|E+`^Y<G+*sNfCe*2@>WHuQjn6o5Ng@!;5h?jHv
z!<W>oTX!xmzEJ^}N=5JlMl^GN*S*cK5#24u@KDBVlKJPxm+n3q-VjJZV}op%s7v>$
zu$D4s{^}^ZT!s45!Ysk;kR&z1sQ=3ZT<$*A2@{w!Qf|}g$s?mgb^HCngCPJ2N6eaK
z%(jpJy~_E^-l-m4hO%a~^Kc_0Ni|i!O~w>-Oc<NE-6;h@uxV^fPfrswc$Z6OkW-vv
zLs!wP9+}hIX;5@8|1GRdp%BP#>gI}a+V#kJ1pQl&6=rD~6XRL-qANCU-dw6>&}Q@V
zDQ^DV5z2gPNk}9AM4IL(>@855pX_m~WxSnF{w-o+Vz#RV!7iJ_5bCpLw<n6as=Gj=
z(M-Q=k(43jdV=%k?X7PBH$&w0xK~&z;sz7hyIX7|m8Vuv^y0<eRzLd+avxM??3E9*
z_qD}MVXFJ-s5T}O%V=DOce!{X#XYbpd>>(wqu;;S&>EYm&lxYD5=P1;nvn@H>3cUm
z-siLan5+f^54rtD6N4*oqN6Cl9+y}X-?>{1rSKk2Ct2K9;-o;*4tv(-#nA7TOut~~
z1c4SzKoqgf<10(r^D<9`ny0}fc!9m*`H5|v_6g1N>>djXTZ9>kgXis<=gc=oVqzQt
zW-?FKAyq`yTvO@&+JdD^ODfdT5Y_Inumyi*=dhUH+HaWQd4+HeBY@P3#By+;cwv{Z
zmhp|m)zg`QCLZ;K!fS}Yg4(rf+cR@7zumYzNaV~%l(BJi8APhf@-6l)J`>=I`C;eh
z_+?YRD6i_<UksH=$Gx|F!{Z5+Di%Y1sU3xT<8yuHW5oj%s}VxJoq)*qIo05f4Q*&|
zP0DL#6G!e<rI1G}8beKD<MeUP$PVwM1nS3P{#ABjbdc?%Ph<3O%SH$<kxB7Xn0}X3
zb;+q5YL0+++qUGV&Ht|`-DjIYe0=6~R#KjtxWGxuSuOMAhbr8-b?Yd#UEfDdlRoGs
zLUBVA?)X>=f={OdusQE*>`1gzWCPB7wUl$;XAA<T6uDvjNR%grh|@>=v-PhLORb^U
z0E8^4G&+X4C{>y^y<qNI)<3wsV0t5=gjDMuOa7!j%>Dcu&zji^>&P#09C7Nrh)Yuo
zgh;Yx->z;u@#Bx7OhbsHrS0=5wLg<`h{$1gKE96K`8U;mgWISPq#;m4hYd3^w!5B~
zQ}QoNhiK6~hvtCxWb}hU(mfAhLC%rOWKPgUi>(pFH19{^y4qEU{Qt!1?#YH1?C$d(
zkuI5A(HIW$5GND4x0uG$SJaA5i*DqfG*cW_`bF2EIsD$M3~K28^6$`kju2L^O3co_
zk~2-sNlncz_+O-1tSrjW^Y-uG@0U%cQTBrWj046`xmy^DAASL5MM!4?aU?;`_j2}c
z>S)b~6k5YA_D$yk0MC@Qc4y*pb8&aaiI+GD{GeEjW|qyB?3UTW)CH<rQ@Zqhgfc(~
zrHl(EMDt!{pd9a)efsw=$MB!7spH3wAFSDvju#oYtrV*Ant2p?>7DAuuo|J7((|ct
zF)3B;W%)=$weO>_R4qso)ioX7^La_$sq3^uf?AVkI83J0V>449OI7muzB_+MM_I8L
zR)$RUi*KFRPW)=e&Yjm(^)+ES1Ums@e+ko5Ub0WHM<o#G%}W#0fgQ-OM>$!!>*2nc
zg<90qJa6?XbxX&^%@tc1?(+L4L<>J<O=rHJF`eQ7M+?=b!^cJzsUEY{j1qIVbTZgE
zXcNWpV>UsIf6gA>f!bSDn4`<k<WZI~vt5NX8rLdppb!Oizif345U$QXapLW3&gpXP
zUTyZUNjzVxC@7@A*dZh~_UvXJ*36Ip5vfNc*jxOiYNBhV{@!tLQ^-2;Iet!p>6-Gr
znthb=07spEu_s-jxG%erBW<B)WH?o@!_<(C*`)W*Zp7Vx3Dlz}-5+cZ-|i*$!Os9X
ztmC>wm*K!bx=J)Buk4HX-w+DD8`3l-z1QCH8|-=SW8)*Qn_TJHN~NMl;TO;cIA=fV
zqdqxtsJn~r1LQFWuHz|ta4qx1yI>=ArvWz&?g^O|o~Q{(UHM6adXTss{V$Pucn_N!
z<)bndoXn*44+jTRK0JzS#$a=HMt*Yn@@0|DrMejTr~DCYs3Lb`a}M5j&ZpawJltNo
zv(Z?_xHK+7=;h1(W;*-oGo<n8OO1&aSBuOU^+X-S7`_aO@t(8eiB(<{-Bo?^(V0w&
zEhdJwS^J2<LM>CbtA4(VSfjhTtW`Y)khTy<#C^N-l~j`;<O%5yEApx7^or!(2;hJ%
z&?TWYz@ds%epik;kLW7oxx{W6PO0f7N^Rgi;)ch}M#KkDVxLQ>){ayq>WS#IOu({{
zYQ$Ol>v<?H6;+s7r*`dz4oSeZ`_2D5CU<-N;DA}~yzdh|$`w@$5&*%Nr|<1gdP#gn
z)%c061C!%uhuLe7jh8((dEduIJ7)5w;>All-r3z}6TKT!xQ_a9XhXUZN>NQ;@A;ge
zsk`$sUZMPKwTDE<K(xL_7xCA2p%<geUW&soAnI$ppn@N!h|KnU{;+*yHL>F;i^Z&5
zxsveT<dLr3J-T%1qTG~$J@J0i0B5e_u7L<W`&`P{h_V&Q>>TGAl0>`ixeW9QM%d%F
z+Y^)%Y<A8K%DKuR_PBXj|CBeJ=Obn<>X^&dr|z*YP^=J>dDV?V%(Efo3J~M3m~`4P
ze>6_&P#)u?4&~WS!J!wbt%ENBlLq|!^%naLjj)_$)YOd;*^~lXup9zQ^PbwN{EN(W
z&%8*(naa^2{i%B>pfHv${M2v%j+ze!RkvoQP)%d4AMQkHFOI{$CNUknuMlvRy)QK;
zrd@8%n^kVvGO61UHQPgLwmXp5w6(jl12DlV)WoTtrCOpXwYXaln2%{NJu4%A8qMHO
zq4Ps|OjLMidSfOHWZ<8R#)!ewXIH==v-BzZsVg7FIr;A91e3y+#EOU8-_17UA$Rt5
zypV#JszDl>&4dk`y2h8$C}H&)L|y{2V`yUs&C<0m{m_;`A1-V@;bvP7#$b^!0w6!z
zHE;U=iOdg8H!PS#(~15*m2}163*x&oBhr3DFj3-Q2J8jW=d>~9KjHJldghKM7N(?B
z1P|IZRUGlB#xmU($FLUI4>PtfY6f>et&rkY8y`7vje%eWR_IIbtTg(fTauGrBO5VU
zkt@BAkX^okMO_djDvg2$YJRU=%j4L}OwY+dwc?yuIw480Mk?n<nekIZ38a(yEsABn
z=~85uygJ-_d-$i>?Vt9m<aUzr=$%ppzBOa1O?2&=apqs5b0O$}u{OgRxW*X#ffbeu
zLRE;VKF-wl*Nk)?@M>2jxF$4*t~Z#WW1by+Zrc;R#KQ&B3lI$=E{(C}bI2SGhTuBr
zJ$?FgQ5`3?=J<(1Xw%8miRRj;R#DvbPW6R}7~xYUsJmB<fGimuazZ0GGzyKLf}h)a
znsi`$VP*-`rg#2HmFU~BjaCL8DV@)WzTEY|5{e^i6UuX}Phl8Uai}tLp~NQG+djaa
zo7d|zDJQNvC<EmdgP^)<Q3pMO^meics1AAnWxF^JzSS=G0h`PnNF~;uXYA$Gj|;3N
z1hl`uh(gCp;>A7<$yo*ZT`$_-w{kZF>qSu-a4J3|O;VmVmo8&=q1XO6Lr>p)oKCCm
zIY-Kjbx^0vd(U~k!0uSRKk4l)EoGR&!A!<E%#gLt#$X(}p!QJZ1%MpuQ+jUcQ(S=+
zTq%3({8t!*HKil)o{HenFdgUQ;+L+6hyGVgzRl2h-3Nw3!iTqhDejflPWQ}k^9qS=
z{v$4*^6>rd&58G)V}JTv*-8L)V=EK453&n*K#JNS(7xic)wT$#we6qIQFL!Jg%kII
za7qHdP<xL{ljsR%-K0aVhiiqKL;dRJ7%bXQ)PZ7wtEtuOq5Bepcw}Ys2}NTF8kDb?
zRMM4sNWxhlDxLn1pghc`6H@}d{rH*ttWMP+r_YKm?t#wJZ7)1@`0%3yiR*w-BL7rd
z5y8zhrdA)YXUET)o4nHgK})f^V=zHR?R9zhr92&E+cGs)ec&4<e)CysiV|zF?y{<>
z+6*i#mx?IK%0wx<fpd<u3x9NZ)3N0xj|F#OeaeL`eM>6N<b5slo#AWNU}|Tng^7F>
zo%u7W)->OI^R4q516wyIo*#nL?|*~OGrltKk+EFNT|`z4Y)UX|H!%BGo_M~#yiki@
z!~L?w7ojXU@15#HjNQCiz}(_{P1JIxI4~|vg9ar%WxpfuRm7}kiO3cdwC~pE4C$H$
zbYWBTj5qCD6cp5?_>xL-&D=u0EB`SDD!)V(@$ID3XrDk1dcT?Ta#wAq&AsAVpns>u
z<~)hp+UB!D;uaY=$yI2(@H)-}cmf?7oBQqkX`J)XL)K1St+VgOlD-BoWw77S;h9-~
zbN)e-ZyE8RPR*_k5T(v?UJRW#{kiwn^h(Zui|>cbc7Mf{tm@G%Y2R22`@_DmS7j^_
z5tPIBjuQFhEQ+_2pt#!2Z)>|-{g3Nf{A)-VUw`l3y$$5XX%kr$l|yzfS9Kryr$w4g
zHdk>=PS)w1xhTt2j?P3zxfDb@n@YudljMXdwqQ%UrB{G*q*#rJdyLt&h+AggJB&a_
z+hpRU8y)H8yr0j^azTlM)T&R9$*`#DAX<>QUZi^0&>*MrL(G<~zeiOkj%gZ2V<gB&
zv+GcEbR+&L;RvWhknpbgWuskKM#Qs}l##|{dB0{;@{dWlNr*Z*89AoQE?*y+;081k
z9sBXe&IzrhIR+oq*qo=IRTJ$1#$Lyz;iE$VtWFIy&p+pbJC~Y>2ez*sRulcUG6r!@
zsbpfagKaUCLK>of0I~u5vgc*wm%{2Kgu5?iDN>nFL-O#|#!O%Bl$CX>^?YdQIh|9%
zq9r2YdFH5p56oNL|HRn3@bdQ4+i2D5NxMzeAeGiAOL=~HuV8tSw53s7s6k0tk0-HZ
z@rWAsiq-*|xrRldEdS_k1v(EtF}FN?$q&bmhlrOzbN-U3D~JpJYha#kX%+qh#omMT
z61SD=Hl3maXXeq0d6$D|{{0@N)FuCyusnR8368E7kkZ}jzzlI?DCz@I7VvxpMWfH!
zQYv~%W`PMkIG4=gZei_x&MnT$d^V*<5DjSo3Zd)@wuqp039xxb0N_B>bQ<o}%!)XS
z1?=PBHODTiHH7LA7%w$~)DJ?OXqPc1LTyOErjj^|{8NG`@X^K^a}lx5E@LrS%foXh
zcdB`Vl`^5VD`k%aCLk&e9lmMG_`Mu|c#93pm5+bJ{&Wad><Gwy(HP`v0ur}>X}6%f
zq-!K5HL<l=@-M&K#D1`W{ouD`|7E0IF!+(WS*QPbZjI-u7kC2mM~qL%dFg3b9$v_v
zo5GlqA>^4MNtBPDqLx0VFH;K)BwW}q@+%|?Bu_!w?K$nXWw)|8qY}ud79^1q6dG*K
z>r)tztsQ9u?C91V67HgdL}FP6D`tzc@p;U{w~gEK%`aYPa)rPt>|gTY#n&U9i{?FT
zvEXualc-uW=$^yi3<t3U#yDc)w?3z~J;y(FPrfU*39}@t2Pw}cp7Ff=Ufi}g$wbW6
zyXU~@Xbahyc;Pz}ANkBN-X(sDK+~ZOm&K(?s)QiioIBT9=!HrL(Wk;kQpBRg?P*Bv
zZKch8w$Xz#LACN4K@wp6Ano$SfzCS|s_?I9Ibv2q%l$rE=Uff<a+6bs4{%6KW9V%;
zVqC<NH#c=z<><(eB>QDc=Z^T-b6%`dOFvKG=%wcH!NIhRyv0Xsd@}V*BJmnTbc^<_
zX*9eG*XoS%O+G)E!ccrCyaH&)^2A|;B2=tR0+3Z<O0ZY|Vbiu1%=o6!n_mnq@visT
z+3gmb@O^hwdce!YyRI62Hs826$GM8vI{)R-OD)R0`$>si#oHcgvFrSoUABA|7!k4i
zr5b;K`B-@6V`VEoIF)&0^Uf|;mZ!d4JuG!v|51+*=iHh8YH<0m&#E?crF&7AP?zYP
zEKOO-KnH%kxk>@;UvLyQd;f-J_H1Ihdc)ErH3~G}b&xq3Z!7PBDzkoZ=jscchItIy
zTKn<p_T!sqxX?7whEUr}V)46|ay}fkgqB}zx0KID+|bEpfG)(THw|agyU{?WTjp2`
zNh^pCUuyyawe!VJFzi*=PfPel9&D^yu(wR58BQ=61d_GJ{RJf3N7z6ZE#!33UhXr>
ze!Cpzsmr}>GgV?KXC8&Av$%(2jOcSu+!MIY(4NDeYjj3WJ9IIWk~la4zBf6%`^68m
zUvcLr4Ez^OM)lyZHQ$n@qB7x0D6ocojbg*Xe}AGLH=_%n^LES&gE>be%#*vh5FnD3
z4UyI5--&jiYr>@jd@NprLSw%V&bC8baGiankL@d6`zAJfG`g52YfX?e9D<XEhwRF@
zBsv#p$$ap+UB63?^P0L{jHPP)Mm~wkx;JkJao7V2>3g%q2fw~3kj}nPz6t0~#1R-F
zy0L!o8VF{nvh)trD!9(Xm2wm=cYLRI2$pHciOZbXb%Y^yOT6MaEM6Rot}SOsD3*k1
zvpo-QnkO$kmT+CQ%td>}*~TTAE(TZQ*|g!r$vyT!Vf<BWdIiGr?)JHhnCG4w3^VmG
zTu9%WYFURUZo02rX9A}4rbnUih4uIvU`zdM)ZTj9kLR%;e@Ge-2Y0~trn1B~pT6*}
zIgMu`R4ssv5eTZbadACt4e?A%RKKh=LgI1z_@gQ2OGLFH>U4XK{qRFK`vt#}lER)Y
z9SvHw!ResTVJzuC)rwpshS)C{MOpMd{ehK+ZAD~Fnwqto8oVoU5THdd0o#z~H!b!@
zdP$i$GR~f7=xD#<Hp8<+A$l&_t59bA_^bx9ztB8}+}U?~t2&XS_}l04H&>>&q*}cA
z{H@l`-@8A()lzgA_uN0|U~`J$4KE;6Dekk~h18ee;f{6sC<+uf=d@&Np?B}zZP?iE
zawnaZn6L%vT0q0C)&vsuiB*d}XOD2p9}&0`eL?hx9!?LNs<^X_tMGg}@E1frzQ81&
zxGW{xW>ghC5AcEVUeVz1WLb3s&u5RBUk+6F(5~^CjI_vMQMnqI7N#x*@H<)ZU0^P|
zlASjmn7@I!z>C1F)c@>z(}7xk4WH>bcD0lg3XK>sY#|&+kXoLt^I^v^94dqn$|wmG
zE!bHF6BP!@`4sZZ=ASJ82a@k9qx#;|0rE6O@)s*HZ=wQWxHc_Toe;?DFj0%;#fR4I
z`b=2T>tNWz!H^|r1*K|EB09+N=g&XsyEQz#A}TzkXt|tEdLzm(VzlACvwKnYyvD7%
z9C`OkuE$v_Jt8wV{#?2}CF=GHYZ%N($Rh$4?1psq9WY=3icqgfIP@q1%8G{mybO8R
zeHHjvOsk(7P&J=d#RA<yaF{M~jbLClFS|bd3iXF0g#VI)z0bjG2ZOsTwDUknNJkmt
z_*32Kmq`NdlO?3aKN_d^J@7uKY%0wcP1MNvZ-)QO`rU{mi|?VFc`Ch@wH`nNnvESj
zqP$bh!xf#)Yy5H9;@&~M_fHCRRHr1{cqj1Fdfnhgia9H_D0PeQGo7Ct<t5W75V>D?
z$o!?(UkB7%v{wuPSnG!3?+QPDo4`tWI62Is9@B|y#ad}MY}f{C?TeHOaMwlHL{sPI
zU&&4jR;z5jDM?(lNr<~Albg+llT$tEBSGlL`iW=-sBhd#qxa@o6Z)QflzIw9Calw%
z3`b3^2yoOkTr{sQKFW9OZ)nPCPKl`e>-vlK?Z}%E=PrN0u&h<Z2|tnvOH8F<FbQ{i
zgv32ZwvMFk)c?&(_j!!NYxuNP$~!MEtM^q=BE|XM@;+i}{ev!6V_MBO*y^{GHDRjG
z{W%r`R;n;FL6GWga=g4ui}!n==6xlfI&3S2VM3&u_7MH}6Aj#DnqHYY;nHKdLISOI
z)~<VhSlfN_A()qhDf^BUnTJAWtVkKDq|+po_r_`-_z|%&XzZO!On>3AP}Vup?}b8n
zvigN~)AAN1rV|^Y{fxrD&Fjwr_>J$&986wXYl0f?;S^g#8Su4QJym<+;-z({lY`7p
zL){kcz&^W`dDzMqzNQT+gTM40yAjC2kuRoQG37@{n6(Z9GaViDS-BFDSCb$lBb+lg
zTqI?1jz8qUk-pR7o1Uo;%>^=I*a~)h`7)mUC_TE}9Ir{X>+wFRp-jCXIVh>3$cguv
zsnida0=M&4vy_N+_RjlvoA-Rh39kY3!lLSb>oJ$~$7aYT`_5~W<qsO*$RZJ5m*@YU
zcvGs^5fKpq&QqKr2WyI(Xxv8dE70r4ll{jZuXc80=Q4JOO8-+lAE{W47$zGb*(PG6
zuEulUYf_r^*o_QlFRs(7TLM_sF9zQJ_3d=vgECm%s0UNo6;;2RM0QCj6F^bugL9mU
zMmu5gd;y97L_d|m!2PVZZ`htXny#U$!MlEk)f}Q*Oar`an(?2>N%x@u{MCcaSTvQC
zNcOY)TxfPZtc@CIC$7%9Gj_%tiVfAzb>s}-(8oDH*@$~30sRw~CF=@z$26+B!rld!
zFQV{<9J%F|EF@c%5lM>UhB4Ve4bQkr2k@KH4QO#13h_O8!<qd?p(KPh(!oEdBaOZf
zspZvW1DsQTGlP~^;hME3h_OgOThn;tV;$6@C`A>uH}#kfP|{#(#=XR^N3Sr)Y+y?r
zvp=v42%TsBw3Cpl8I1HYsePydMh1VBB#(8bl(gbVtO#K7S(RJ2_8;WfH}4}U92phr
zxppghOb4bEuGKzQbmeLtz00p^g?+GDGWjsg@*NleT6()xU%~^j=nWnJ7_2zC;rC$b
zf-a|>P;+uJroD%WOv3JqR4S1NaLe_JXYdQjl;eAt$Y^RijQ;mh>*n(Gx%HS=xpMVF
z7NVE$-n!CZ1ppwujGfn(5ds`m+VQ+k+HaCHOqY4B|8!o%wc~q`_3~1ayR@c)rfgp(
zinC68+o9lE-k<hH!V14O{c+cfl`BQIn(5yN{F#QY`7KUU`nb00mrmRnJk9wJ2lVRo
z>rLihzK15q>yCEnu&wAF-Df<bC-3N?3H8ZDwH;gf!+xVR#mBtO`s3bthkk*|`GpTz
z*>BpHsVp}1s^Js@fBx+NHJJ;|`*8%YG9|{~H3RZJ4PszlJq==d52xcn#pPYjcwfPS
z-Rv)Xt6zYUy8qNpCV*`@yS49?QMP9-Xk<MkVkfRf3Nzyf%{c-2V3@=n2DkRTsUoZh
zC#GH2HFOI&%qv}>ZS9h3f}<})|Ih|wQ#&|6!ZPt1($S7dW02Fnq`RZuiWnRIr5>J#
z;o<4B-!-1_YxfRkBix&p@NP7d6H(n=zqlR+2`7lH+uJ5Y5ub4#E8?sX;plPeLjcP%
zcP@csgm7+KoF=c{*)7J2mwx=2?hUawYE4iyMkkKvMogwFArQKg6bsLt5Z@BiBKXe>
zlby@w-R-^bJPkFD_V#Oc)n73{)180L{>7^=mz#!3XVt$OIcgMxFscp{n^82|`sVR-
zFlQ^&(vnK9HsXeu9jZ#831uDU!M3Y0ppsiUwUh9QCm#`i8<pL!Po@8{vDF!j<eV0#
zX~DY2YtI3E6H>*|Zik<y$w7h3E;=TZ@UWbt0EJ?nW2SQvX<JKrdeL5~V?C7>a9z#u
zsOfb<%W~Jjd7~6F;Ea!mBN86&{gLzF_h!fnXry9`(~=XU=`3_|HW^&Phdr)Liko+@
zdp<OE5?>S-klulsV&UTbZ`{6h{d!vph(myFLUC2?uuDl{7z`IZrrO70uds)NShhvl
zH8lh}at$?*NYjxw+i)@#_FCZM8%vL(d+e2c`;_xQ;mz5_((4a)T}jJZ6Y{|_0Ud5F
zoE%zf)kmL-{Gq5x^)6!9ZGZg9qiAr3B%f5+RS5(#Z3ro4(~v2WLy^*b^$FXUa$E(~
zqmr#hbVKs1)@f?gTWY$N$K~uim^${<<EYq8AJ=O{lV=9fD&1~s76aIA_;;?;JzSH%
z{+s#F@9%g24329<sl*V0icQ=w2f*en$@PNy^elg<P}stS*DgNquKplrJ?C=9<6v&`
z!}p(Y53l3S8N~7GFs;h793@t)Ugq`pTHGD2TTzLlq?xi(p!d37_yvU7)KRELEP3E$
zbtXK1rc>w^GV=jwk?}sO$DgQ0KvcN(h}exfO-w5IyQ}k0^D?q)EZfFwQ!DYkDQ_cg
z+m$kTR&R0Ld;b|1|2ed?`}3{;JUd@Q8zefp>vtJH4Fu6<-N@5;$I6vub6?wgLdNd)
z&xAEkDN`x2Na63(`>)%-f9j^K)y`k|5gCOd$c{~(85@hxkn$mo-bf$7=#Gr>!V#Sj
zxoPkp=e_nYv5j#Xt`W$A6)@&+22{v6(5EP4IG)X0Fzj8NX-cafPM;o2VLv#OJ&uTs
zyf`8W=6`(VFJ9-Q#3}#DLNrC4YJDUy8?!x@EZU37JW6YdHBtA9ONn_di9^#Z?^C^|
z?#a59e97r3@IL=Tw>(4V{h9B@@#DQ_%&(Tvgd<zI-q>4NJGu=R@MLk$3K&O{$rLOi
zPCj%2P>f;%KE!ILi{pJ7eYafb11nr1`6*JNRp&7xjcXLE(Pgeq>990P%Co2mTibKB
zj+ExGJx3%RNcVJyW!u2f6~+vp4AqR4Gg>>|=YQ}2UDz8OfmOph4+|eOykl^9r3D@0
zuqtKpm;1oykCcQus05_^DD#DoNd;vSTAmfzxJ?=RIjXsQu=8(-2ve4Oi9CABiGHH>
zKTYrw3lXJBH|j`mY-W<8_!C)DHUKbJ1l0~8$39Ju6YsKA>_dyw=xMF<Z=KL2Wy*}u
zm${5g9*cuk?MMy_Fuu{OP209aW7lwup;DMC)a<Hz(PLn;IzSOA?>z&^Dj_efbhn#Z
ziCV>YF<FP}Yx1IZ)mrNaDQ?LE0WAUNrOWLJ%^jl<1_0q*J^PbWk(rx5Xt?bCKW=Z@
zNBox7<@UydAAJhEDHI?XR&R*9qGAT9mq=59oS7AiE=rwm)4cA4gWbNSeyq0K6aC0Z
zq~T~IjYAg*jW<brV;%7Fh#Qj3ud#C6o5Wk4O+)$6RjrD9-;5aRs`PVyFBQ#Oyj27P
zw#^9%`rX2C%qJq9ggv``5^q$UI$&9#qQG;1qe1Q#ImTQL&!WAOD0<h=Z@u*!`L;u`
zbp`~f^GzMd!yLVG$o`Fxg^OPG=R|XJwy#qZ?0cogK2EDw_MhbSKW9Jj`dK8?)H9<n
zuEXy$>|p1qllbD@ZSs4ODA5&kDxl$`k?e}bxJ>_P<90TtZRG_K5B397(Fq(;(-pK2
z1A<M;`TZ_T>y)y`KHev`gMi(J7bv8v|2TVfQOpNc%vuqzGk}i81@}~Va_=(-Qp!<6
zPVQL&zMxNeZ!hX={&zn_u>~$d?{{6PAK+P(Pu0uWx_^Ja2AI{4^X>0B;ukhH-{YJL
z?woBOygumR#Dw5b4Ck@q$B)mR--h&$Ea?^GTzzlV2v)2mq5?pA>KFd04}MqoP3Mj4
zjI&&}W>+n-gKw1F_gfM7#)>s++`4jP4~<>QT|e7)oeg;BcCinsa?e2-Cbs0?TJ_Cy
z_2Tm`{^K70RlXp)w~m4wigi1pU|*-zo-pSKY3JhnVQ)r_J;?%icY}TZXzZ07xa(uG
z3gH&cxF6O2v9D4ic!4`k?SxFy9&*kFvgZ>~^V<=-YZ}4C)fW)g0D`@|i{fl+Ri`G=
zQ{a2^A+3;4rEpwlIxwPrQ7H<&mmaGl8;ztvhp{V!uf?@akz@Zp=&jAKW9_$`+WF^^
zDh<06nG`94kZt4&$bL_bq7+@8gHdY&C}z$#6N8+~<MKLBlTY&g2d6sK>D#w&83ErU
zcSbYLJ@-~iuZlC7{j~4YPPG|uJ!9HWyZdrC)Nf;cd4ODX%!oX$ij_6GL^5Y5E(7tF
zBC_%|A^M?0_ZaxzstYZ3yLRx~iaeRlbI@Q|#{vk^XR<F_Pwm7@kbU$euG+P0*QuTO
z=;296*P-R2)S$)brel9e-H4HxrB*aHjVzJM8xgNX4z}mLe*JnJp+YhPIy%?!rvq5U
zHA=`P!g9|YCI=f;<P#1=<COH<Q?lPL8!&4YTM_kOGi@!M3%I<`_k~HiAGE5Lqo9F1
z_qC3xr4J#NxG4evcKxowTzC!sWj&Jtd$(6Uvz~G);9%cdj}@8RxkMZh%B^p48oBE)
z7FSrvD#d@;qg%IYyh+8|NqBTp^A!z);m6f+=#j30HPm>a#EI_L)jg)8*yH47`zC&6
z^0H&iv}7=Y+AH60B^KAZ{+W~{w>{d=&-H%XAd=MjL0y`*^hQ0V2;D*4y3;a<_6JMm
z&G&*$><@L>p%aWUii*G^ZitFl3XOhPS7xS^doJl$!*`~&c@_+7<gVN2A6?QMH_7PS
zI0}ty^QoO5fBd(mbzQFBcUYCK%ea;5n5pG?lb=!3awJ@P_4=a6bmZ9-&Al~x^cMRj
zrIep3AfC8-8fvnjQ)O=KBcCCohAz<iaWvjN693lstvX;vm6MTS4w3#)tqFqHXkG7V
zH>dNVe*tE}b2^RMRmM5p%L@=U43Z&kaa!&DAbka={|M?O_4ZY_P-r=IV!o3g-mNIE
zLDF_gA{#i1)J{7tQ+XY7b3f1T9D|FI{Armn9a@U<3U096p5teQ6z!1pRL>I)UjCDx
zfI}+;F7wCo9BS@Fn^C78Y^;CAhsG9^U;kV>gIwVn0PCn-qqg>nUi-@fF4^8ed6<5n
zmsBfutqDpJ&KEl9Tyj_AVZwr?jf>LOPbd8GWoIQ*Mz^Xxggcd^_$bUpE9LP|J5}0J
zctvzkFz5ijKIYU;?(JK=XmB5AjF$rsz5DL{4YxQMvnW3|KC$}>WTwfd*}UMUEPU@x
z)Et~Dm&n^1(P_Q%@(2+$Q4c7iuHh33q=-4+LfZ8jeP>p$@y8xmzXxT<9m+Ft^=QH(
z(3F|^V$Opzr(~}f%bfLZy2U6HS#|A&BC2YPU38nUs4*Ev_gf#bdNKfGdbnIWPLY=k
z;WpHAG!)5CJ9{?84?jTHYcp`sbB<0Gg%>XZL}&j8z6EAjIs!RoPQ{e?57C&;zM_+M
zy-)gq9%y)WsSqF+O%j#x;S;=o#;(mAK&#*p^2FOea>_1*K96Ong=^7N(v!OpyP!#U
z@IUOP1R{zMh)`yH=F;c0{gr_tJV|rqGIPoUmSCDcZIN~Fdw>ltt9o+$Grcs8$=Qzw
z8Xm%-4<1>V{4<(TYy5`;pwi2jf-Ah5*n|ZwLk6??!T#{@ikdUEw<Cw7g<ubP;$dmR
z!=6Hj;iUs8Ln-il2sFOy_cb$~d?@*Lk+52sJE2hJy%Zv4A*MaH4anYbON<20UpvtF
zZ*HeIC8d+rdrGT%HETu$S6%x$c;y%oW+$$Oi}S71#hN`Cg2GV#SN@5sXMmVH6MRrE
zeYAUus(TVMNq<jl=T)!wW35{DVP)T;{d^e3u^{}wvz`^KqLRAk9|XRo@CBWh`P1#J
z&Ljn)?Q=(n1-v@m&$X-;;Vv9Z|63n&-~g~9e)%UM3CbZ#yp*^#nm~_jJFeK*w+}EU
zLnXD8tW9gzCf7Oe3n)(s3KLxqEQ@`TzRi)t9yaVo(1BYFSfD4$-x(F9cy@xY1ML4#
z1-1FhV58DRS7}(9*nq@{7Oo9n_8ZSFw}5B=>BMQ8`>B(W0bAMF#${xzA%fI_@xl|6
zt+(PNI6dzG0=@?akUZp)u;9@38$RZnq@XzvXkLwc<zP<s-M4T%2bQTT3vu^r$Va?&
zsZBxBT8V6a(dxtTS!jK1X}W?zGz3^(9>0#EJSQHT=H_fXIgxHK#J|dVR=V52pzAcH
z)_#hvpLzd!q^sf2G<jYFD+s}`oL{C%&jbc7qY06v_AXg;xnlyb5Z+kq^v%7ZD=U^I
z9Zw&y4FHF7)9|){rwdcQq$Do7^AWPRK4Zs#TOwxLH4X%$4Bbt^smzuUj4YBC_+4Ab
z9E8&^YrGlvVGh%gn$e{B)8xWO%T9QQi2y%$evYl!OX{mhEVU*`Rf@36<Dq{{6psJR
z<?)@Qs<I`E%D*OLlEikn;at@-*A*i7gb6l$)x;j6Pr*mJeX`oZ<4&Eb_l}`4+ZH3D
zL+R8(2#&vE*(9Mn<VI&|6NJhkaJUPOzf)yiv&v{ucW7D_3prd=A8yd1z1Z3+Aoz9d
zf)8~oP{)`Rr=#($K94n9@GK%n8slVeWbcJb*8T0+R1sS!$OtAWbH|HtJCNzw;-W17
zproectVFVF!NQ$Gn3K)YOVKBgi70!E$Xn>E!T&a$l+q-eE}P_rRZ?|RqYL|7<V5d*
z#hkoh=>T5?_H|wkURG4fcF@OOJ>h;6D>Fna1@et1b6ycwQfedQxO?B-c8=Vq89BQe
zpvQP)jGdVmQm{DDk5h5YAhHi$YU{`Gb9HR{2LC0S`gAX?xyl;<FWA0uZ?481k*$!P
zE$_@wpf2N|nNIpD7FQ^p@9m(QEbm@Y({fY+b?yV1;LL+TtdqXj=&;4yO@ppean#~P
zix{jh@x<k>G;KVsUweOg95;^bK8#0R*<(6~N0R?)bFwD-^Qn*W4qZl`BOeG_br>Q<
zG1mBs!UaWW4@ar5QmnNrf1dwF4fsEPJ%_wfqD|_2b*3-qLsC)l>)w6ZjSIEuIv0C0
zPyVtms_uOKdSWbratvjFnycwhU$j?0gdsC>V^sGN!1&aXyi!8B<Z6wEiMc2>XU#@g
zp5(8|&n)XrRl#b5F*8!(MLjWbc^sEg^&~YVD~bkgZr@f+Z3a74R{cmjjbutYNRVFX
z2g-eL=8eEJZ{)wT{S}C+&X`(yOelH7y6}&KGErV$<9+5{5i_Y{4CE2oR#PzsA-CeM
zrRX_V4uc#<gf3CmbFMAvJ0=8<Q^QAaZ5kI1`NT#F4SOl^S5J?GETfspvgk2uD0Me!
z?uYUNW~|+rp|CL4;_{=>3awMQGKFW6Qk^JA7|7pTuAc{q@{A__`jh2g`O6boV_dO-
zq@D#KBDeImW;zX1-9c<9Rd}x7cs#zR7SOf))3U|&--5qsXbyK{(Oyu08-Q1wqp5iq
z=Mog{M%?hHROKjn8f<7GC>@^+sIT$jvb~*`Wg;6=2E9F7#qnO!Jazl`$)MV^SFZfy
z;3w%Ey15*UZl+q*_&Gz@$TxvGYMV|n3_*jo))GYX$6a?KoQcCtF7#U%bP$jfPVn@@
zj_ulhuCzl>@p7yuaZ=Z1EitA;Aov)eUzc}@G6e4$`gSL4M)rjnhcUBX8R#siwnK71
zbYTfK4>%sRCWxP#t1;(#5#mYzmGqO17&dC;$Yi-VV3|(lfm0is2i|rE^EwBj_{Nin
zQ?{N>jjHb;=c!#F1`5D3=!sj{;~*WBLcFl4A?A7|)-v1<|4v-acfP0(J0v{3;$N`e
zd$F9WYe@fuN~U4)6-0wYFrofs!rDWHhC8dC(Ff_lZxy(oxXjVfa9>-kyXr6co#qly
zSe??xnIEB{EpXd=7|L(UMNVtjz?kl+gh*kip|q9ZuteK4n&C@S;>>O1$v*!-O6ZA$
z%xnBCY~DG;dudT<i`}#a*LxKp%Nh};)C9OCivgD<<fE!C>}VX&_BMuN^z{$Gkwf4u
z*P1}CF8;0&>U6mmIb%eL?$O++w<QZ!Vi5|{JCmJtPOM}fU>F}&qr;-TbLt&z#$~is
z2G>vHH71L8nlFN$XcDFd?OJwJ=qZO`a&Pe#f`cvf#yfo~fXbc)tWyiMLkj!rSLGmQ
zUjzfX#=@cNNNo8tb(kdB9E^(A*v5E&n|L4U`6~T1YN+aSktk|uzwsTFRG1=J_*wJH
z2R|pqK8lKESi^6yh7us#U?48n*PS>%5<6oo;G!M}L8nD<Lc_pL$_xF@KDmw#0Ri+V
zdcU}w_R$B%+@MWH<AgP}t<Pf~QOb?96}C0VT;pcW@%O%@DXUMN)}n4*EJMoW-xr^7
zO%w<$HuNNbTPI=$05cJ@mpMHZM++{+zHd0$AKo@-k7wn*oV<@Z>sbJYUB7GS3P%{Z
zTTd2H-`t{IA}*Atl5NaC!ItOn%uKP7B#`pr9#0hMX|Tvsew^1Q&IzvgG67;=<|i12
z)KeoHFpjiyj%M!C;nU}FEVK3;4Nr$U?#^cnSBDX@o;1=nAC@L@5JhJ87B^n5m?n*I
zZ`WJj566nqvS5P+{}tw^Hz1nRoUFCmP03^$o8|IK(g<9&g8n-yb#J%X{FHRfQ(zFn
z@`l08usQ-YVWZvrpjnz811`Ac|6s=b{G+q;KHFA|*;i{=b>e;?{KbYBb*PY|5Bgt7
zaX%k6pWK8id;Q6iN3okuX2TJlMY-7)s9mW(5>S*vGF9z!Czets#-<E`!FIyob6u@t
z_>og1In=P7{@rM9?<Or7HK8$|t5Y3~ozStXcQE5kbL^&8Gh1FSog$E0ak|}f=G9>v
z*LmKaW|>Rk?5)L=jkvKC-8^Ad6oo8iwaZxjC!MjJfaB{=N?!uB&U5v;eDfa=j^Z@1
zhPTZ_tTzjBZ>$7b4x^hx^@?x{E`UXWT%EsD?EuVa4g}I&zpH@@O@mk2WgDQrZC+yr
zfmadg7w;^icX&2Z1#La&rk2EJB@|82-{&z|c3rZpSB>8a``kG>SN#=k69+GD=+KD&
zVl)>vVv>`_g#UDkT49oyLbc41oorwJf~+O1yZ^A3)kneC2t3=a<c3v<)6OXWlp7bZ
z(ERL-&Irp&mbPIq|Lic9eXU*k<-0VPLo>N|?I4{}t0kq9jmFYu_?mZ~HH#l;BP-f*
znFv<W1+i8q{CKCTMWs(8T)xj^%%Q?%CBydkTrp?Aa%GJzfMUnlhv(i77K-P_?+g>i
zmr9t>kr6jKp<96hF2|#haToXM8GEm(%2KsEaT#KVN{gOH=sn^;5`FniVi)T-bByU0
z<yDQN!^HjuWDobO!73@trBd&uLy?*WS)4kni5>0nn^pgeOn>jOP`EVrc*Ds%Ac?A+
z*pSQp-dF$52V+*&JGmau@SJxf33w*=g;0r^0BXto_!j675{}!|`Vn=6u7~-08?-p@
z?)_?CQREoC{?0OR*K7$f+PAM*MG6+Q)qDcDHR1+IcoE*haB97mn7i9sDdrpnq(WFl
z>u9g{AxW8_0o9J!%BomN_NXz=?EQ`}?r!E^xa+2aX&wbvUDGgF;D3XoWj)un;VBV!
z{!3=%y~Hef?-b2gae_;jtiHXS-*pY$EIEH^oH|rqO+S*tigC4J|1d9IVlrx}Lhw2+
zb2MVMe_zN_ab2bB5QRg!e~Br`<N7>S<c}*Yj{EB7#5vwe3@NntuY|U5FY8EEkuqvc
zgUGt)Ua%=fD$(1B8|ugt?|cZnM00A7Rdx=3#UHY`PxE5IAHZ5EbD{X?v71Bx0WyEI
z4fDP@eAqqe8j>G*S_tcH{PoZUf3=&QK^~MUx>MpTq1q5Y^2WvfAPk!Go2y0~_KaPG
z4@r8GG1x@9WXnT9)I9FGH@V8|&M=3XHKE9X!5`_~v%evjxa-<8*H+U207+_15Um3c
z+!JGMo|ri3v%?eui#Kon+qB3#P;pABDCOkE-k;i&lT7Kee_e(J4jV(`{8`kQ|KSAU
zuCN4&!0iFWJ9p{|XqPzlW}ZmJnU2A7*Y)|UrNqBk`ns#~;Ooa080w89wwN>jK+Djr
zJGMV)_9V7r;wSEp*9H`?)&1Q~OU|D^dp7#~zFmp?=9bT2bwT$fMTa$Thre~b?4YXN
zFaEvQ#Y#n@$9&&wSgX0`&(_{Oazl$Am*=$W-8N|G@mI%AzWKp>mtLQL;`0?#U*9%k
z+rhi5r`&xIOA}^E$Eb#XWSTa;J_4yZ488L6J(M#1nMXRn8ikC761`FiGE!r(!KWm7
z_N<fsoN<|EmV%do@FXZ05~IR8p!nY%pXqY9hG#4ZaY<!wx^$NrvXdXBmE(l3p5$DB
zf^(f%?98Db!6$Kki}sXZ0FhK!J~qpHDCaCSoG)NnDItk`DB>NCNqr0U8_aO~Gj2lX
zmkGEQ?jb?!uVR4b1?nhEC8a$0aQC(imuA4)We?^fKN4fV!L%-(af^h?AzZRY0#+aq
zZ+pI>%sfc8j+zG*!vytCsQZy|Zfoo}n0~Y~-a@p+Oy1K{y0Y81``B;j@FliLB)w+j
zg1&c@M9MvY=Q?`GuCN<UFPA%CiiS$SgszdMpqFNY;x8CO;rl22s&eW8!CY>*RF8Ik
zlrL(UL%PH?vw4N<_DaGzE=eO1_uO1Q%CiymeigF<36S-_D71m7^@+1Z#|AqsQQmnQ
z4?D4l83u1(Mr5mYdaUKUxpMjf8c4Y_@{tOGAiQmx>51HRO~+&7`beuxcjQr%ot_zL
z9`AV8?Rn|G@-|&}Z~5x0ISLk610(#?d7~Xh;_u9%%h1?Gj@jb~v|i6}@>>4M!P)du
zidN2D-FQjBtwu!4DVs&_eWY(G{@EF~1mB{kKK?z<QQOn3uV?J(%dUuZbT~wmO`@{#
zZtO{Y{C^!|97~vXG_No?Qgx^S6%FQKlrYB8^>7{4In8EqTBaYW$xa84!khhl(`+A5
zaR4F`8K<F9#3dD;es88|uAWuT-pr{~pai+H>uyJ*Aw0Yy^DAGr4`3Q!y=KuaS$&LA
zEuCWmT37xKVqw6-62R)02@dzNboZgqB|BFm-psi1&bxi^;n|az%TP2J3p|IL_&oew
z{mFku@<D26!$YjUv<+ky2(XShGgdKw=-z$%_6aynmb5bjH_zJ5ysFX4v^BS@zOdtN
zu}>gFn``eg-(36NV?b)}<JPLDq0^tz3_?ckr?0E`oqM+Fm!*(6u&MT_+D6%-(INqX
z!X%6LMfIn;_L4tkhpMw#ioywBZ{7Ne4$$dV59W?@{xaR1nXtneQI8QBki3=s1kS4I
z`SjmZuos&oL7g!I#0X%c?5B;b3EQACtEy%tON#_@MODwID=P&+(?>&6eDA>7Ek`|2
z;!xhL^!x6~X28iYUui~>P*x3eT1wDM_iT8Z2yjJ5;a@WBcPLVqB{RJb)7e~fpRT4E
zuD5u)`l-17QlTi{s+YYp5&31UAS^-m-s6`kRZnTA@vrT!=Qa1W#Bvx^Q8P}I{#SG9
z{hFB7QY;YwlJG=cOPbyk8{?cy(Janbnz}n-E6S`UZ{>($&RT!#EHPts;S_Y!kleE*
zB7T|v*(p8cqawcgs+hy+zxtwa89Po=Dweo=NR09^($s2p9pl@mB~h{<(jLlI#~}Ck
zJLlAvHI0PcB`(rZrLm@g!Fu39gT5SSe&1)l*xnkzu<@7E6jC4>)o@Pip-M&}c<2Jn
z%%N&oELbFcMl|;Gtlh1xu}}WYxOjT{lF2hrCZV{{s*?Szk;WXVGKMrXiw3W1F>yAC
zOb`3nV?`-Wbj~j44>$KRb@=<VU_a+Lxp86kw9Ai0!y$=Fg2z;)UcIilAEmL80d{3e
z_8a#`R5p;oIz)KXKN4cT;6?idSBYNcD<2BH04k92Th|%2?TQUuhK8CXNRny3x0VV+
z0TZ*>eP>w;6?HnjD;7}40<lrCV2!U>S$2M=S+%fGG0()v{{X^&0?B8(THM(kyJ;2g
z9g$UtpCAsNs8b;^>PgewExYj*=dd=fNOnLhLf<=%6p=a`l45$A<Db`bJV)G8z)1!g
z5+l|LRkwlmA#Mm+mf9pGwvCrGkzLeC8mi3ME$eO0?rRSvF|ZmA5+<9VGUKxu8TmcO
z?;}-zb#`d@h{|T*S9$P_72}*=$Zzhy&wP;m^_N1+HutrdL%_gS_TT$Xk|M~pjEu$6
z!u&B2&Q+H&^0W%kHU#@~*&cGPzEi1(m9W0mD_ptXn1hdR#%F@eHj)DWKb;>e<Z_lh
zygW-jM@q^Pb0()MU@dIHMB$D^pOIg)9xC)wuib1OHq%G|`xJf+{?2`WBD|28ueR>1
z4QJ=2+?Pe(b*fNmW3CF7{UW91Th1^4WL$d6GEom+e!25y=+;uc_ZXV?a%~c)$biMM
zbYv_IQqng~W<zIAa=szc$p8ABb~^ku$Sm-}&f%wjO^jYQ(D?`Ohcp$Oc;&?_8AqD>
z-jU4Mub@t_jyswi&o;`c5pQ8X1_N${#Bd#t0w{V6ZsUgPITv+Pv;x1()>;27?8t)9
zv2xfkI2lIL);)PBu1s~2YH4s`N>bo@(X;$tLNjno7R^!#(<!kq_clCY>(&y+-qbNp
zG9?5mdMnM5-UH)vK0W51T?p+3^~JGR7(KsC?K96>#oT-=iW^^y6$%yCGO#J%t}eXX
zrA!kO5TLl_`0+nwSDi!`0Qq@s^YNud&O1IY3QY4sD(eMaAVfq=GuAF5@}TN}k)ZU3
zU%CP>lz63|@pXrm8lw}RS7DoqCxOA|IyY?xd&d4W6f6-{5cQ9IBqWjExPHyFH|~07
z)JC&7(*K>3vrA4xo>ZeB?9p-O+#(;HFUCc}*R&l2I3$xH7IC8bs|C!G6^hnd1*KhH
zwkJj-m2h_-=P`Sq(-k>e%6Q`7tr;0h)~*r9`*~7xSJ#{OF8e0YwC2gj^;dWJ8fH3K
z7w>u!h0XdOfw|QK{>+Cmf9lWQ)C=s%^?)1Ud4d_;@O;wfrslO~WDX~z9i8%$Y%xzz
z)xN_5{BK3IkHz(8o26k=(jJ@+H?$@a1)+-Ndq>J=*u72@`#Hf?cms2V3N64Tx8r&y
zr>Pu;&Aexpz<V-EBZKQQ$Wvl{ewl#(4TwZa<l3vO;=iP6Gum8LhoE}T0yPG3&#!Ce
zcsB6hc3Dxf=Q6xm<U{iAIfQc0J9htGPvh%9I|Sb1oBiOVv<?;?f7aTLskQjQK!Mol
zR=nSySbuYD$_}MxYHae-*sNj6X<VoI(33qm&DA=!kQ1Z<Y2z>R(P$9y&ls#Hg&l)+
zwPGFUO1sN6AHTkrdD%nL#oiW8IchW=S;B?DGU}N_Zq;K7tySNlV<IzwSjt{`d@9)U
zq7Uc%?`V96an5C{$_~;AnYDfO!eVrdc`kY4HVTUppHm@w<0WY`sEZh+!R)rl!N#3I
zlQeb~Az+ojM_o~m5VrK3U9i+(YNfNNIJE~ZB({Q+LSa05;_V)ek94;;Vyw-ch4DIA
zDlm@H9&>O9+9k+yj;o9OhFZusUDCi2`AFvO8bcEhZ$hJ)S5=vG3?_`K3+)KQJ!ig9
zgmjd}Z3Jm~XZH-?FF{a%@>QHOcP7BN`ux2>w*+wps?%f-bw$rQ7MhJ<Vinn3MqhW_
z?GHN~c(sRf>%-5Q$g0qGwdMgNF4`mQeQMKh*ihqYcypEADLA2Y>w8BU<XF1zoI4$A
z#Jk72rjculeGu25<2Zy)Fwc$4p9CyYT#Sg6X~=?Lwf*ex8W&k89Ad)}swu^S)e0lF
zqvv<SpheDplAXEf()ol0jT^W|In5ZG0HYk`&O4N=>&SAXNsthEy_uecuheA93omg%
z4HtlBa!0&>v}BF3efMe6GB>8NmlGfy%Hs=L6Y5Uq%dkk=O`m=+_lp)UT<Elovv>6p
zro#t2%l1JbD;(!I8?R@~@pow+lsrO64AEx`!otblnnU28{G}&CN8!l#KO^+Ulk79p
zmqsZ)$H`+d?)?+rRUr2ps!(f(`<x6{uhVWvHjlX-(&#YfFrN`1OFS*&F5JrwmsgGZ
z?zyuJO3D3(Tk1N&syd?DjwCkGcqpYV`wd2FGx0Qplp(P9LZoNip8AXD6s2rX`;Peh
z)c8B?&Pmudy|@I4$a(BH7{$%R`Qq12f{axxSYn+*9_^Nf4`_*Np+BRQ#*f^AsBjjF
zqutUla}%;7C^|dQrn*wGoHxuOdS?e8ueMY$vhQ61tuT3tIgTe_-2UhO#=prtPGnfT
zYpeThlegkPq?h3_dqYDdOhI)pB^#lV1u2=`^qubajr~Y<S|UJ6<D5wYPp#$9#ga?_
z4qh=*q+}(D`{^_`N59~w4vMtBf<v7;+30lDG?ko{#(7q3M;H8^rX4<0vR~N9SaLhD
z{G>{3>`%du?N=zkP7p9YIFEVR-gU@Ofn^R#_H)@tsjy4S6*^L{R{9qeZ<{;se^x5m
zut+f1(r<$4)6K4R#jivXRA|kF?`}~zsoU#FT(L&g9kUT;bb#I4!by3h^OWVndV=hG
zZf-%1nsIuv*Uf8DJHe6e7hME%ZlbImM$-3QafS9OM#m))4V2!rDrosY^BUeWMKtQQ
zXWSrFqZ6hB;fwVWv1LnfC(JtbwmJ7p{XqHXyzQun6EU6W+|QrG_lZ!C+xN><KQm2v
zJ^rb0cW*Y)!qkouFcCp4B!-%jhM0RSr`LKhb~w*?!1b5v;dD9vpE#Fx$2@20?^T0o
zd{m|QJ><o`DY*K1j_MmRa0jmeG;}$zrv9KeTT!7i8{gXqyk->TI*RTR5`)UKO0S-6
z_u4PbJ-S4FN8e6KVa!Q0(vIHtp^d`m)Hi74k_=MyG<CGmN8-r~0vb!1L$LO44N8K9
z2Oo65$tc6jv}bp{02oox>_uD>p`50|ykw)3fW};`Ycy(ws5d`hI!+7N__SKa%$`}<
z`l|(q3*Mr`ty<bV!cvOXZqJghJBN$|?D*-9nDIj(@80UX0w=BNoEwk+yuUZye@=iY
z5(D_Q=iG$;q-itid{=QK@}cyKVf`GY4El=s2CG9~Yc%v2V*90gMqzL08tUX(`*$@`
zjWDe(SyQ(=9<fH0&y1m&Yz$3@5%{5KLhQICja>vx&V7Ud|2;VOsb{WEEin|O+aJua
zn#$C<{`S}1t(jChVK8023l=U++w_Av`3PE#w}@|P(#hwgr%9K0)wK5W%9h5h7O^{U
z^|UnmOk}=v$&w>rdH--uTuO11O@Av5o7U6{iv^1!B<U{CV`n|1DyHVL{h~w|d~v)v
zHIvO~&kjD+`siN>i4iZY%e=Qn>mqO!VsHLOTuybUgkv5fO73qW^El)Cl!ULG0l>*=
zng%Wz+T0&5Xk?PS*rVxD+323HCBM{kEQ2PLH8bH3U8xV_oZH*^vgQCjO~PHzNm!QY
zj?%!`bZIOtLUaz4XZH*e&}_Z2?Xk&Huu`rR7z|H^L95qVdw|0*-@k_DvCz6?@xU<W
zIDu{1hQ?I*E6^L+6yi{{^X_1_;c~Y^=LqcZjSwdwUe_bus^J{BaUD!ZJmXuDKB<il
zPIefyf3cPBdjABsnX3p0Uci!(PB{k#_u8+&?%~+mtKlZDe==XlB@<eIZ!@t=m%k+C
z_lk?gjCm1hCc)`s4@_=rc5kN91<q(K%H1Twx1zxkzgKO+vjJ0)XZ+x7$rk&32to*>
zj7w}5VV13Rw_NiG%Y^LTW8f0IYG@>aNSzN<AgLFv_q#k=mi(c{ybUjpi=70nRWw+w
z_h_?u^4IMZfHE2)Q@&ti-<~~7O0ER#)u2yhUovaGM#ho!wrDVBF7wTtOFBpV@opjT
z2|Pwkn!b0WEv1kOTWJ8tvg16TN7@7iRncIvClMd<0AKFB{~IJ9=WXt25lteNj8f5H
zV&Aza2Y5XCI9C(F2z0<fD(te`p=&{f0Aw+Cu23m++bUyUN>W|wdbuDXRwB!OYu*7}
zOI%@l_`cMgUQcj_jMzbblk_h<TF{}t2*Iat9VjhjaRfYOzF7w~6p6M=r!%2F)i_6L
z&x0W(-gR~_{4BnFK965DjeL16<rUASmu;(wJQ~IZ1!MWhSRnO1#0JJ(JZ$N_KtQ~x
z$k@2fmo>N(qg17od4WTf#;H@9?Wrv9){~_MEjQlHN=XOAOUAinaN0YU={J(%a;4l4
zkJnITWHUdXtQLPL)x{lakvP^q^d(3KSm3H@((@6@ll$6ndJRhE4*Voy<0N1%`@03q
zL+_HnJRYv&<>C;K{t`n*<QH#BG7Y|?q4_ytd%2PL)M3uCIa70*i2TIY)zdT?lXCrr
zf##iE>20N+PRUF%VSbsQ2a@i&2Gswe6HsKlZ65#K<=jz8H>auf78-6bYo5c)!?x*o
za#hQwT~HhK%S3QQQ{|>&x00Qk<8M@L8@E7bb{^yAXLP!}3O4zDXzw1l+&#JG{|?6M
znFzsal}bF2Sc(NJ|DhD^IeRIQL*_NF4xd%hUXc$4kBab_3z#3QZ`RRItnHlp6>=8r
z$;@O_Nd_#UdUgxFUkj#RNn$FRvKaZW-u`=UO3#XtQk>&(4w_+t0e;l%zDRlNRbt5E
zw#-zq(@KC)O}BxB8Tn9iLSgopLd(=jW)}i9$XXoNs%aWWmjT1Hj_0RKFv+iz#%@JE
z(%e!Y9%DGxn|bb`N9YqM))XUwbF5_~{^5p-jm#Csxlytf3*Y?UdJ~0*eg`wikzGr6
z8k5mb&-KW4n{bd-?(Q~v6#YtSRHX-~V>6wKie_e6<4Hg$qPXl)zZ3zO$Iyixro<%?
z74aPWGAXWz(w+>uQq&xo73Ro%JB1>xqwtZu6(*FreA{>;g1XYy6P9;(>aCWM4{4;+
z;6}+18*B5Vw%M;U<Spw^Vs<icimhLG&TYZAXe&y2{Vb7$A00;*)Kek8f^jePoFv?x
zZ!|cds5Ws)=@<Mf0NZcQ{e}+z8I8kKJ(3%ZKdE4zVY@0#>`I5ryNS6!eW7$^&#vW-
zS1?XjixZ4H<V=b{%|I^9fq7EWeuHv6e0QhINho7R=Dci+gYMJ^TgFbSqkdI6xhK*s
zgfrRM*@mk#oP6G$Zv<r{PS)x@qX=SA4)pdcV!kvElNyW}q7?a1)hsGb|I`+`b4*lR
z&=)`pAh=ebLP{vEo}enc$R2`mF_Y4#bLvC`G6&8o05|rQfHdj~Rnw%IOJ3_KVTTtf
zt$C$)Xa9R7Zb0^cN#g6WpB*k2q&?{`7Sg-Q1gbf0>A?1NFawQU<J_4e8@keQuZ?SZ
zR8Z4Y=RZtQxrS%4t1M#PsEAc7C!X)kkqJ(lt8p>}e>I(`q_L@ip9x2PLsQ&~1>=Gl
zj@`{;BfKI7drK;8{lYG9?{)-?@rKcq*qkUhe1SbP*9d=RL&d2dcmXIyJaLby4!Ex1
z9xbu_jwG!E8>l(|^_;8;<~xem?~q&w*2k$uGjynx=ONUL_uo-+FEcAQ;6>WE%7ObB
zQ+akv-=?0=XH*aZfKKgOko&sbp$qsoxLKyMdghoDmwC^uk4ZK`CE{>@2`aPdiQXhc
zOBi|FWbOtuEkEbTgcmW#ZSyzeosv69PTRcm)a(MU?D3p8*ORpx2_r6Pyl2wiIGwgL
zCLZ)Yzk(k7_=JRNX%u{`Su5q5(M_dK2OU>g7p<EDPW%~a=oUkFi_Iy83kvUar~Rey
z?y7iJ6f?=)MU95|!?=;dJV9Lt?rdyAqOb&1FN!*r@h`s#yK|Z5X04p@<Z<LfWrms?
zM0=76U>lqHsA0RGpqTucnyRX&N&5DPW=MO0O7D@}od$e1{z}vTh`+teJdI;-V<W|M
zb4p6k^xw31({*q=9qqtIv#5Q&o-m!8>s0c-8C%k^kzN%Y%W&*%d5SEYP}j@-nNcXk
zFVF8ETKW^_=wJCN-;BMI%U3yb$n5s{NU~atVD&U6*cN3%*}{R^6#h)D_OH-5LbsBy
z%#FbVyd6Vb@wPNP93$CZhWwLaT|9U%f};RGi0_oh1;!ztm@8Y*P<Dnc#sN4Jq=Ge@
zgFVW9f$0-O&qNck+g+sphtMkQ#M0h*jdAKd4A!zUZ?8WBQP(2Y%`0kS3U_vt2#ECN
z#b?sPaTNJLyXh{+m4-EeCtsXM(=e4t**)?5Rk!H6pcd-|aY{v4PS^ab;n%n~%S+Qo
z`1WN)8CuWbyFHa>SD{(J;{;c;E6@xP`4IG1GXtfM?sz44UE_^=58`S%g4Hw^+)-lZ
z2|&kri?t3|Y=%4h2UlfiEwk;<HyVBlz&NOCno1S)U+2DgUa=XZ?ei;*s!`ee^;$~E
zNNU!0jNDM86&~>j8HZAMz_%|?yEDpv4jtPb1JNu%xGmycXE{B7nF7G+z0%kZCmi!$
zVy^k*ZFtp$!rlIjpyZyJ!uHQ%NK9Lj$Ea*|J7^A#2PvpH#~iu?4@or}kRmSMllkWb
zW&xt<G|)M8kUxrgqVV(z1=L8Gi!kMW`c=ka7@86WicHsIU$2J!S7h9KiBm~(Pri*!
znCH~rB^)W0!Ka^g28z<S8L6SRNX7d)rmbQD`_SD9%(<|p;yr$Rt7Myg`Lt%{SXCOq
z);Ot~U{YS;ZO08q+}eMDf98GQ59W2ek}$ukBjv|1(2CYExzbM{C!buS_2Z6?mC{!Y
zn3mq%SXDEL!#(*o|0g&6*8&U%Hbtb+lW9i;ON0}?n=_5{5|MXHhgrv*t4Uf-pRP{Y
z=6&G7ZU3AW(5@cGB{3+z)UA~duB0Z7Sk&_$({k?Lt(!V0;@QD|bNyz=^>{P%-EKqT
z10st)s@-C4-frh#IF{^ohvX|zvr4DL4-%J_E*u{^a#8zrpAOkjYRlrm-7kL}@Y3JU
zZh!vN@0qQCJ3b;OWN*)Z^!DF&^zK`0hCj+myF0}*!O!KI=*!J;27hwl{jw>f6;*5V
zw9KGud3aRsat1w|c#1}sf15h$az&hzR4iDeqxIs9eHIu&Twz9OrI_q8qMMV^!xbwU
zLmYP0KRdzS&fRdiTWllX*5y^P+KjG13Kkn%-?<v`Cd~F3?#Zz=CQ7mrc0>(ZsE#wG
zm&^dJ+g*T&lo7}x><bvPH1P-}lcHkG4}MfQ=U(B!6SX$k=^1c`P^LaLTOrr+;~f*Q
zmMuukMw)|C#e}_GIM}lVL<Rw3h0sj&^uxc2y(byvv|A^a6(e9XW3%Ippt(;CMnO0&
zz;S5`$l<-59jH^lpZLD4wPO&oPfvfia}pw?X5`3j!{!*7L?zu!g~B$)#uun<KIii3
zss2&=fiL-url_ZUhEauA%m}fOCen}Jhf_XEAoWF}a(d*(2726B$^jH1_E0guM(wg|
zLWNe7>Z_BJSXv^kJeOg#M0BTWGzZI7El9mjJWCmmDbJ@4L%|AQd+#KOJF5O$+RZ;Q
z0aFQcH@EbmJT7fjGv*${YUKA_uJAWTP`A<KBMF^tOhG=agbp5+8nbT&9VE%!XD6;o
zMRsSH9NB;s!Kbe;Ow3P*nBxrHjV12XjI#WN%Nq^<c;k&Dny6o1yv-n$VSqlKzQtE0
zE21Zwp}UmZ=To<B-@W@{^A?}lx6n8|XGl{ff`mnQ%-py5R(xlrGAh)htEiH7y+&&z
z%pPwN%ZTD~3uaRFYm+}<fL;7RyN!*X7cEIG>CPBR%_^bI>!=fc`Q$YR@5lk#D;dFk
zA})>A-Fc7-j^uv4wlt4H*EXeXVpbebvfD9FSL>U#^I~$<E)}^_)m=;qoBP;jOE(`s
zl%5zfXPJ64ydHD%KC_1>f{y+g#H*_kS4~cd6wlL&=)2tQZx3&s*}ET1u*gdwd8J^v
zsfp}UWz+8O4GnK(Jkw}P<7z{as6<t6`SXxE_MJm}3Yu8fsD>0*{lv&pmIQiZ#z!`L
zA?qA@l-uipSULA4scHuJu~Hv>YriAl4^69)EJL?0%^9J}E`nXC3q8>>u;XSK0~@-R
zmK$K7szCq?b3Eqx>I{hsNo$VGrC*eKj1|RYKDuj8zoo(XRMC}E59f-+RoLi2kXX;j
zaXndzh=Vgq042Zfv9yD$Y(_VR8JAP2h!YV+6j8C}eKuD^+<V!lzgI)7B9xGdf@&~v
zL^E^Y+<z~7>mky8mZT=RjQKek^Wz*8mf*)@aa;!v(Yshqw0uTSZu?kVF7y<}IQFX9
z-<@M~b^D2-4Fi##s1XEj67tjgp1a>+oF+cO+xB%Z2^bn)4x)9iy7S1*HjB<Ni584v
zF=Afr3%X-R1i4%tz9eW-=iffssD2&-*lG5Bn;D~5Xa%W}bQ1?JC8t9U-j4DkWPMJ3
zaQ&FpYzDf~gCOA^`)Na`DYQZ}U}+p#l}O{1lu0}m-sS%$I+4n695gGNT91RPP5;5k
zIG%jrB$;JPaPjA^LB7Qq^(lE+&vAEe6O5|5J~g*KQBO2>PMHIyz)Z;TH|6=VC3FMn
zI;?Nzjpj~W<vebSpwn3)vsx^orjz!cRoH%D$~TUwxZ26YF>S_BqR}v!xUFn1MLeIG
z+E(rCg?MaBuiYcN3rKawJ=X;2fj4EQ%7@u!7%xt+g2*5V_!AL(Jf*A9M+~5?g%Y!L
zXx~YI<^H;gDik`7I9Kl_v-4rx<(^`1Mz@sWU<Gio+?D*y2O3YZx_8mVGh@E#6gB>8
zy{IkkZQp#`!v%1;K8sh+96&`#k_KGS>t;v(5DpF&o$Tp+_&wvkbZk$>kIZf~gx)BL
zy|^^#17u}Q&0U8<p92zcI^Id@y0TdVx`f{_Cx4A`)SsOMs!itBbxIZMkMS&fJ=gwf
z{o14fYRc)p;`mTc!O(o#%<FcyQ5VX593R7q=>@wc5#ds116*a_qY|gIPSs^+L>p)M
zC2^3=7Iw8;s3x#Lm?LqjG@}q@C*Kn8<Pis%&7~8qk#D*ZP}0GSq`T9HJMdISj^i&5
zU3dt;^e_*yCXqw)GO^!i&+UtN10+>KS5k!XO_v0LD8SRvFEHblEVz_$i6(`4@L}Fb
zm1^(Bi{;zda}>U@+{E~I6)-{5>yQu&67HPD*cq|rHD+Y=l$ZAZ0SH+g`<+vz!x~3`
z8Ib!FP70~Gg!qm5gh)MU<YeUQmz<>vHG4SqWqiPJjX}{u*hwTVcOS0JH%`38DAw*d
z>I4HMG|`fo0+ZpT#NQc*`DkW^tcp@K@oc|$xpDl@I2=F~7&6`cc@GM&<^+bCp}kXw
z5$2moq$Nw1u~4ilC9KZYJ(Vf?uGv?wR;>_>g)`I3-QF-d4m!&7=ePN<Nfbx<K61`K
z540Qab5s}u`DE^JHyrgR&T$RcWiCyp9~f{-#Vqj#TBsD(Jx4Bl-o86u|161U$8w2j
z1EPA|@z9M;RsP<FZ1JLUp4`pplDc(YN)#m7ubzJX#HmyBMBTaa<j$3iDGrSj;;!oO
z&aDkUYcjWb8V&+_yC6}yCbPNi=oTXMe*ZH<KNDdN-O}+iHywrAvDlf|Ohq8q<VWPU
zVLIJ;M>nN>^Zv!65Z|FozkW@J>P&zMb4Ehw5i;V80Y%8n&JZODc<m<;Q4PMq9BoLE
zV7s(wX3u`!e7>grJdSjZRIL8fdQRR*6sP8w0N4Xj!c9~V@l1&i|JKASeh=?nsQ_Q4
zQ>v0(^~>ZMuJm}xTZ%+X`0tlZiB>)Njswa4w8BwAl&R{Pr5R>@?k1Fv!P^YsT1RGT
z{sJX%XPl(>Oq2PC(DYGGIt{6cDB0tN9+T17l^dV~cwEaJCt1)?I;s+e3`o7SyH0c+
z4AWmCOPALyb!Gu{GGq%NhZ#eYs6yFaH7V7M^O8U%i0PclAb^*k%jNp3d{wy{AYjQ(
zcn-)Lr(u@4;dwzr%E;=i^0>o6N}}0%9fkd;-%$O3#^(V(=GYftxIpY2QD>cDn6xkD
zdK<3@wb;<!;)oR`jHQE%?gtHqGUxVNV=e17{(pRZcR*BU)BUDtVq(QYu#u=>>_&`O
zS$joMv5Rc5Ay!Nf3(Dp-DyS&YSYll}U_%Xx0)h%CNmNu&ET95O6p>~SLjBH}yN%?1
zzy0I=eZR=vz4y8E%rkT5%o$LzsDW=GgmZvZ`eDxq&|_pdQPpD=N0~<nG`~r&rHM<q
zGwnVU*eB1n9PLv9NL1|4^h9Sy2D(#BFW_B@p}?OH1~Yd|Ax=vlcKajH*vP|Q9i<H5
z^}Wj;K^Yo_tR3NIMl1dTK6VXJMqdu_m>TgR6;Q46+m8@kv7EFwwK3s9RY9d_hBN*y
zJRT2j=wSN&FaLLho-|jv|G{`u5@ez?aY*?=eHB2J7&>)jVRC2e>7;Tb&twR76O6M^
zF`lyKEJc+8XBnAIwVdlkmy3EE-DHce(AR+gVMOw*FMnd6?g^KO`{H(zhATho%O>-z
z%fB@;cy;aCx#_X~<Z~VNAo~zU7~26QDby}kMoU#Zqc9)pDz-~}D~o)aBYiVQX*gvb
z692M?4)xeV)OG^ECc-F?b8@Z(O%Ri@{`_;)iSW}c9fr)BZ5kDYWi0E9z03KJK(;S9
zm!pWz$bLz7iZb*ztEL2)etv$0d6ZdtY(xU6Zw#x89Soid;uuH7Q&A`Z-bPMG?Q$j3
zJP2Qjh<Bg(N)R56ZlY``TSpLYLy&fU$SO$iQIf|v)y+#eU}B+dQD-3tQf}VYYIdV4
zi17=24gf!hLleqDbXNtJkb=Q>b12M-$tKN_wYOVLSIG_`&bIReaoo6VhI~uH2Cza-
zTyFcScJIHO5V@u*Z_aQv!%T6MZ4m(?kuY8TJ9G&*fCPCY2QT7NI0t34o0noT0V$w)
zP6w&U-v!%&Rbj)1S6%)Ql{ThgSrz2Cpf^c*?>&SAq#ZO5*)~v}x3LRw6+pxl21`@9
zKB5E@`w_>45gWy1fu)cZl5|kLS4Q5dTZF+JNOp4bhK3K1QqzjHVpIhuAdOjt)oV5J
z!%bdCAU|J%qOAJz+O2|t(@UI#L3&U|1W(GC&lLYc{xM)<GK!C6F(KldZ6B1k6<>2p
ze2uQ|BTzos=X-_?{_F%YJ^p!kUr2aN+5<Xga0v`>5RlI7w5Z3*d;3UvZ~d4w0qC+>
zf@Z=^aB#+Zuy08{cC6}p5|Cx)GdyxWAOXelCX1s{w|GOPNc@hQH(H+K;oxT*w%@Zr
z^crt<gY2O+1(HZO9X-TopHvpYW1B?A0ly)p9<y>BhxDL**24H6)=m9)Y;NQDrMQf+
zU#O9w9hAv&TfIOV^>?A>G5Ge~3`f<Xl(M{a4%Buyobx9NlSm^vDc8!JXYHWFs9m9y
z4nMu%Vym;~dVi^hNmc76pq+<(-Ck_WI$UBe9;`{Niup(@5{IybQyFGQYK&>3AG~h>
zVR_h{^L!9EDa;MqSL`N0PITQ5<N%Rhumrx*59|=*({}x1f5prF2hkT0{D^`KcJ59T
zAyoYGAW7d7KV*`aPs->La9bAm<gR$y#qhXJhlsIaGyo;+4rdi*KeWJ6Z5&&wwbC1I
zC&c^CoMfn<(+qHHJqw07*bZWQ1(#3>pOYz2c~oGduF^ZIt(HSo9+KbAyh6SJZr?o(
zoE$&ulbm!*dM<1xFvJ&WHdOen5xp!Fldq+4QmjJt14mcc|8{E=I&2?m<Z^z22ttA8
z$x#E&<tTqOVYX+O3%_WhvH;qX5TAx*e_#qFY{*hJuAf&w7=Rl}eW-=rb1ok{s&@nE
zKplV1U$8$RuG(FPW6}){)t{YFPcZ4qxLvY-<Ul~ybRRx^Ic){p^kz6RC7`~h*h{r6
zieaIPugjSTSO6PsO_%SHSUjNdj!V%99;mz~FYZ~WvQgg&0q9gX&<_=Q_C*;(*`!Hm
ze(b#HS&SGAV#;`+p<mnco@;ei+=LF!n*&$e*u4ox?p8w(r|p42e8C3bMaMXQ@*t$m
z(A?YV8Y9?6LzMN3X=1;~8gF+;3OT*#Z!<y{x?8{M<8ZWP?N-Lzx<;?c2clrf(w(-y
z{;6kyd~gP6r!sQX+)ozvV;oLyf)ObvF&xZ0#Y6Ph+szC}u9IMq(iA`nx}~g*@2qeb
z3RDMSHslI9m%|}a8p#-k=Y{FtI;Ujib9~ep)XgX2{l;r>i~Ojg?)`s}cwU%9;)&|)
zCl|W_VA_)6^&;^zK)a?lFa;Pa#sZ#28{WWw1>(=zD%<uuwhp1Uu-lo&KrFL%SSY`)
zQe}m(IFG4I7RZEyOQ;hC(zI;(m<bY0qvaW^2P*QYY|_5q$2S|_V72E}k@AM^74PP%
zbi>t*Vdjss#h?80gE$3^24}Sqi8j0PvsNL*p*NDbquP#={ArhqUS-UCwOcO1BwgQ6
zNrf`~>g*@so}OqH!Q4{NPGC3S{dO0-$vYzJ<#d~BgeW|^fiIbI<JMct5A_dh*9)$m
z$}vLtybPfNX<LqZyGvrCAvYbq!Cz*b0oD@--rn|<y{O#hd;1o7n{NFCpcYcQ$qpXJ
zX099PkVD^8jGRmgs|Wi2Z1F#JYGgKP&G=E%^*=KXC+CNvq5q1*^|YSCB*=?dqJ{2T
zH}CtCX2-$Gjj)AoG@C}n?z;c6)480}9;jNG_pEOQcTe7Iu{)<x?A*)(+@NWve|+8O
zm#9~ADjbEbLI8sP$r%Bo$BmmQj_$J3^I4b($SH$jy%rHvIji_%VTH+L4}AmD*kCl8
zK?nP2)_!n1$=*|JrjhU}bo-#{LyeHk>X);#ai$nirbKs_MOQvx!^{o%#$>7^!1$(%
z$Eav29wX}~&ccHS-IL-1MsYr<H@|B1_pd&*BzfF~I9St|A`Yx7&5eF4kyx36&u0BV
z#1ftEZadJsf&!s2?pF_>Iv={A6vzNNu}EN*SgdlLoa2QzP-AWa$%9#L-jIM93_QoU
z%kmt+C8PZEoRIS%?@$`xbY#K$=e(`r_mtaai-~}x>_W0cCzQcpfh}ieXG`<mvOj(b
zE`hh<fHi0&J`{`5b9o8}kpu(=@tL+Lxj|cBKad(fQMsQd95m_=qE9Ed1Q`^+0P)5(
z`2xk}n~Sf6CWrV+9Koi$|3Dvv2RmrbK>ATes6D=LTQ1g9dJ1FG<2Fcxc`6h|4#E){
z4Ys%o(^>2$PfYMUoW)i!s<ax;Ha(#y$&>IIW5P2%(eLSoRw&~)3^{=NKC?7G-5oa)
zQo)P`Y6XLBwsvN9O5H-#E!0gXV0<8$GIQLzA!OwQ3gZfn>4El{>EK!auy=1xatSAA
z831zaASIOQkxk)Tq|C~oVl)XDgYaj=LDyYnngWk{xS(6wL4s6kd(6W(p8+Hky8z>v
z*nMR5@ILV8ACTReR8EcyG6X3Fsaq(~NYg@gQcCoA^&{I@B=idn(WCbC^J5O>bE;zG
zkjq$0XxDv!9OyxZD4UE3dq_6KAP$*3$YLw_baL^f<eXKzi9P%8@{tP2;iUUPIhbob
z!eqy?Hx8BGxUtPW9Az70AiwBCY|!}seDHDl&-X?SPxi&Gy#w(c_Cy3Q*G~LUF`Vjf
zcI_pQl<`|5HyM|`)(`)05MEPZgL(aim?vF;Ne-R@7`wyt4QD{4;*fV*RB#D?iPMsN
z;CJF>9b%Mqm=2!pDaBiYCTs33?LVR74`nAw-Wx}yH|(teAw_-g1gfC@%Juy@28p8G
zZbblZyD|GVrl8L{Wx?2tfC284a<mc}=>>UowSGcWE4U~~b_QG&=y0QHOTVUQwwHr%
zMHy<A-#zD#>PbOaeh^;ZY`z^Sps){Q*lJ8ZbZddU!~yghQ$&Ga2KdT**C_Yff^4YW
zsM(Sfc7|WdkH<t1VMyU88lFxCEBu0j5$IVwUNK;mvzK6Gc8BP)g5vuZF3&U?+~h?h
z?xEN{CJev{+qzlx5x<<MR$le5;Aygn(cc9%J7fsIXyk|SKVdB~=m=h>5hp-}tU}21
z1Jsxyl5k26Mk|Fj=<$#~lC)j72Q;M((Z!~{l63H2aX3a0A4I6WPQD{ci3mML!<Dc^
zdWM}x8=G{$YrF}sm>&i=`xNwpJE*KUmm}pAO%Rxl&<}J9OGDQXeM>9p$dm$n6<6sM
zOC590oC3@JXqGE*Yo?xtLY{7)azJedXw``VM*nh_PS;t?qo_8F6~#Qvl1qE<9js8$
zb>-c)B7+n8sLc{KtpEu#o{TE5M0C(tusX~--k%*OhFYUrU<OY&?Q+G}Ob&&gp~iqf
zqUiiIols4eaOe_%zp+KEuAx^tMb5TDMM@=1hpIH&%Tvf{2f^_}qnAqZ6eTfGHG>}^
zMIjT2VUyyjdJD~1U9%M726-X=E@;_;G*6;&*$ZRmUr23;V`*gac)juuItK*XQumf3
zj=JV<D4;;B&h|`W>^|rkUo~AlV9AmtB$nU8N$y+@lNI6BNKiUtj|Wv()~zpVBjg#K
z>4{d&3=DGUWQm30N8S4#>X(eE(2mr5TqR)H>rl(KLO-rlaMewfWWDAtDDm-_6L=P0
z$56sn>6&?fXbB~Y>uhj0WpFvEa~arZdkFtRYgK^WLJpm+7lA<svFO^HW86N=Pms5Q
zf*`wTk~w-0(E_Bblv(m#KT43$bp}o_&eTasUHof%^IDR2Ih!w8#!-rSQ!a>}jsa-$
zQX;Fa8r&vF+9I%pPH<BQw)z{LrIp`8-_pmV1{i58@QWR;(fX453gD5X1%(iU9KGDU
zF$R0U`0?9KR|isx2U~E*rTsm<7oh9mhM13+Vm|90f~1x{A-ZIon>V_`<Ds7Pp3#8p
z)U+A6|6@6}2y{?)4}Z=Vlcg{oYMrCPB&3PRJD6(-CbiY(0fE1Rv`#oHEXg3xnJw1d
zL&=@4y<nBZ4dnrUAw<DA&YNyH6b_IW6sfu6Tp`><J#9!Y-ITz@OCg@F8;$O1x2@Z@
zHNflXYGA9XTO`6_mSU)OI+#Aggs?&#V>Cnv%}Cu|5}#+C{{FqA@23z_sQf8Kl#?;l
z>!yZ}#!moeKS*mBo0!H$Bhh&Qn?nKMSGk2SWzSyIq19J`Co$#59G*<P!P;*{p=)JV
zsIlw}@dXEw&8D@Q^>P=xn=G3z=Ic<51(yz}2vGIDqAH1L?fZeFwYV4j%pHjHIhR8z
zKveJniDt-~e^=tmT4xwvLZRI40iK*)?#B2o(T|-PVmCK$!eyZ=)$`TK3qk|XDa4&c
zekV~JJU{Q@mK0#c9Y=!KV!DCF>{0?1Y7naOAtH}(bGPyutt17E`RrRj&(v0SdjOcl
zyq=<}Ti25jIh_31B32hUoJo$lAjM#t^dqa|zq`w_D=b1ItP!t8jU5obl)F}lI*a_M
zWsw8Lhgb?HAr8|R?>cYr{|?DL^rkZeNs5q_Tp$b91mSBn1|9r7+dO0T?CWr789<}J
zcS35+PCnq>;<N036Q5=Oejh-W^h<6JK+~iUiP3tC`?2uLc`#@P*cI{=NG8vGG)+k%
zW09rgV2eJaXGD0pHHB?w$hwT@wW{I-Fw+4D`kNp|2rhx!pk_RtSP9J6DM_^JJ>Zv|
zz2ct=*VY_=E)P-Be1*5wUaysZ`WAm<`}J1G0d*%NPL8|(LCAQ|;3?h5p9~w*``uPr
zYs(*xw5-$q%fahHtlyu}@`GjL4P1Z*@AU20ukZ2XF~3fn>-lq=U+f$9+%zV*hTG`T
z&P%>;SyVZ+@|#OR(al$Ef3^I>+`*@>eYC1oku6s)a2Ti(7(mnxrLuh<m1ttZjF7(Q
znDS#x8QMq!jnLi9dX1t2fKCm}UzK$jYL0P8k;eEwO+e1U?Q-_v3E6Ke6y`_7*AV?9
zxi&VHQzek@7@CAt+P(LIm+Z8gGbS_haOsCJ<NzAvuCl)t7n!}5I27JPJ?TEeOEEB=
zRWThdau5TDJ1f1-AXH!F9?MpYHTxh7aQP5AvB}}IFDfS(SgR_{1t$E!UdH4u#GrWe
zll&n5b%jE(j&MgzgG72;I7^2kYz*ycDpkaMjCP<J^sz_SVDZ31D0Hm+?mNmY|A&J7
zbpEla0EAzg$?B<8n$0Myf=?7~j|nhiUz1mdqk0o8Pg!xLjVI*;(qZR)wVOY0P}Ngu
zd$$1;$YwYWZ3-zz`NN8M5i)G$;>Kd<DMQ&EN(#1r1n}V2W+HDY)e4rY*$B9)M^=z?
zDP*r4ST_XbpDs)Hxwxk#EZA)eN8S%^Qagj2R;_SkU4xYX$haPR)Dg%7nX_=RLa^89
z&BcP`{&*I!Y;&Y^mSDTToh8kUpB3X!{2?cgQwyC-!FrB`b~kmF#mDZI@~e6nY8|Zj
zOCJ<~9pg?-v1M-zC-%S=pyoI=yc6iu(pereQo=#&BT~VFutaPs#5XNmD@z_24OX-N
zB8$z)41<d>U<k?xpFHPXEm1@azJTddoS~jKjCKPkm4^N+RZJgfyk20*FRD4%dB!GH
zMeRpj0|n~MoWSV$0JRGaP%`~wG<gZ=9Y6AD*13oLzj1JiY>8Ik1P#eU9fPA;>v>mi
zvPf;#ZKP7YJr4bfc!W>8Jf`4GWw{M!sx&KU-xqUh$$_BQfi|83bl?q{j6C`mbdlk-
zkA;ffK96V=y@_rC7nE+~Jf#u8xCYyfnCmN&D`w51D&1svFtx}BR3t!8yF`3->D!m}
z3_a1$hKq;DYrxdExo??gI;SvuqF2Yz5Uo;&Axzys^#im49ua9X^y}7z8fjfZyU!G!
zcwQ@zLz*Ml9w7~1Fst@Pi%GtPG>;5`%|NUb0Z+~kL6J{vw#-6o_A<x%(2)Hj@ZQYZ
zHSrW4D2#IwT-Jlr5KJN|2K~otczzVy$}0MNhtnz^+Me~i$#h09tntmO!<MjUNTQ^z
zZR+(8nOC7+Qir^}XSvGre*pTxJ_uaS6<Ni?QN}H!Eq{q3@5cQVe;pHkd}rSZA2GNJ
z1=03Bi&(<BGSr7^?#UBeO;6o59%5mZ$d(+#Zq7RAiCE;iL6lF)|MEH_?{tGugntBz
zJ7(RXplAM0FkTue&Lewd3Dnrc<MgMTWqIBe7^QX{U01upS?<U%Q-{JA*ltaTCBq&}
znBieh&(EJS6@orSVLpAKwQMh(Gn<b^=;ep48AGKug>_NibVlk*0L>?(@zWt;oLNa_
zc1~bt`l#*k>oI@)zZHMrnFG)<^SP%tQtntZkQ*c=RjRI&5ni*7gNElDN>>F1j~jPe
zP@k3)v{l?)@ou6|U+s2$l8PFN8@(7mPHmfDrU^3u%oo;#2&{@SsK22VKdHOC_W`jO
z!B8Lc-+AfM*oTwGWVt8F$Qg`7+79gl$$&ZbZ)WFb$=9v<3t`HUUe^;dKUz_Nl)4*+
za9ZWDmsklP^Up!`NUq?-58S78VI_SbKkb#43n_VyF}3iP0Zb<4V4w#_P<W05OZTVc
z4#1wn!7>(e&H9AKuI<r&_sw{OX%^m`KLQ@3(LV+y)1NbWEF$Nr8KZxmZ0fzDjpcL<
zqLk6=0R~q(K5+p9f{$o2^^@MTGes)~lhYW&qqaBF9$%olFF|<FIe<$U!03lQfgkAz
z@YLkmz51chNU<5b#veuQB^<S>3;=0zJ_IRREm<cxgNyT<M8OkFG^i8Tlb@`OR7OKy
z=<{mI(_vvxp>&C@cpWZ$f|$)@k@wM!ECHp3^H=;MkO##h@HAPC*GQb52XVeAvU-$2
z6BUOCr<I%7CRUSmn5-}9>c<uf0Ptg4guNK)Q+*&qxy3g=3w}!w88fqb8u|^jlZC9I
zyY&!g-3Z9mc|O21ptHizz=hXy5^w{mOjhUv7#Ze^dOa(UShzU491qdZ(Ds-O$U>3z
zTM(Ha&&kBbGedWaTpMDfqp^>iOR1n{=;+Z-md0$|$cJcdC-maUI35~w7c^W#e)77p
zBD)d#W-!Z;m4M)IpW~FbC{SY%@b;hTDT)GCcCEn+GW6GiD6$vakih`2vdV#Zr9;uH
z9$j6Ka$00<q@zTvZNZ7#n?MgSE}Y6$1Du$sr1Rr>Uo|F=03pg``Pq2Ic)Bpmmq?P0
zs+f0fP(uU@*&hf|ZvtV?kAZhHbt7+GBCD>_&aU2@2X1T}<~hBIO@)ZO1)3BCG<;ru
z@*1$@j2xEa!k)>(0Y>*bWg}u`)!QjM=W8JCLbY(gBbvvPs*kMf-0Pr_8UR1rS=tBL
zFK-XhSqm{fD<Ny>?yo>PmW~ZQ{0WvqM$4I&;ES%HU81)*1Gwy{Fjw%2MRsOkT04T{
zP8o7sBrPYu0S_6QiespS?e#7n$s^!@mJDJ*4xd*gAP>oXNxv2}(x(8mub^a{CJu3&
zM+~(luwBCu(ilxg>jI92t&G9|ysES)LpSQitaEZ&f)rj8-r(d{IE0-m{ic}5ciC&<
z{^Ajc@Sq#qfK+pXuvJiJ*)|?0Q8xKvA1WWQsgQ%JLkK`)xcAWAr!X1~#Mqu=a&1A2
z1#?uyTY~E9uZ3%9`7cMA(x2)d0Os{Vj+1{LA{O&Vv6yZ9(BMGEi{d2B6x~#+l-2B{
zQ(9GV3VqVVz7wheB0uSLQck&~9uMbzh4f*!#F4-ujm}-QVj`Rshh6^HK-_tX_~@`(
zP`=54KOFwV3p2`58D47iw3?3WYJjwJ>`>diKYCOm#u|h$gG@D9>13gdt51%ZH0e_a
zfR7R#(!kI98^beg6eW-IIn)=7%H1CD@wuEhgBHxbjD>ZNgnSw_t{H}56r<My3iBwi
zS8<przfyS#neGOFCHBc2a!Xw|gG9SJOdt!VxiI<`5Zc`n&OskoD1*E@7I84?oQQ=z
zW8SF_eJK!lSP?_~tvrR^UPwJ#im=<lo2RI`zf_evaV8mNpKK6)E65DSgI4f8H(iLZ
z%CwWEc~e?;d88vm{Ey+Bj#;$`FZVjp^<2#DFpBaeOF*tZ*$LIfV&ri3ykR1&=9t#s
z;U#W^Z8B+_k&PRa9JS>lRHKW34EHEPg5GfFaX5qwniI$nXM^cLu~l(Y_`mZZdMDCI
z*%RcwcTGd#d^)0S|A@Y;s|`fXWO{2J8t{e`Y4!0CcR~?^LxDxbO|LuR@w3rKqPXR5
z9!#o|IopM02v%2|OL0Ju3WcGYlljUu(TaP%LtPq#$dDd^j2}FyX?uGdMmviT0l`5Z
z^BNY{nBD0(=S5UavtQXaXW%NMIB-5ahl+UsstmF<a3E_T5l|O9<_^G9=TfvN4?>Ax
zY`LZpZb92yadYuj-4;xp_w;ga&u-Xs=HaC(B$!%7Mh!jJ^YbQAb{sNP;%VPXL%r7O
zC7VWh&ikWd_fZ@%O5@AZ?E=2~KSFZD^>Z*GJowbpr^IHrX1t_5$*1cTHrNQAv$q3H
z-_RbNZPDil2Ky!1bLN(y;}L*cu2ad|^uyII;#8RqVZti$xHd{}HQ3K1W}Jh(GCi|N
zL=VyE2sBI|f{1nSdH57IKX8_ySovEK&d^%T1QL+{*S%3ld&%iX`M`u5BssARP=<ML
z!@^ENNRz_f(^maG^En<#8zsqrZV-Pw1n1J3iTX!XIjx=vf+^kvXSfExW!natraGi8
zf7M>jq$q)Ab@3B1+=bjbPZ^D4CgX-!{uFXg7U>lgzDi`{9Hwj(5M3dKKXpLh%~{hC
zP9$swhZ}jWL5vOYia)^lK?ow6;6Mm|%lWlLSkVnFA`qd26<7NqOmJx#)ci-KpTa51
zaiu128+HS6gcVGVx#3ZP4oK@VKnE$)<%kH0V#gYuR<!Mb*qh9L@(a*PHUrjkNA4bC
z#8u~gB?AdU2g||gydt_dt|JtP8u&+G?ALYhw;G$?{Pry9v5i1^RFQ2&{}5VHe`IhR
z(@Y8@PA6Z}O+siUd@drHDsD39ch=1bie{D5Hz0Pq`VKCQ5RQv(@@}B?kziYij$SPE
zw+yoUu*Y{uQ%40asg%h)#0fey!Ihy?lxIE(tdL`<>v^+x)vT*rrK`nJa>SUE&jUav
zxdqda_sAW0P`3GbN+YtDBb}1~aTa`GX)v9n&mdic@}LDoxv*k5+S4&SZ1V$GUl(yT
zJCr3GPW2Dy!rSI>QYiy=)1vo4+Ae^TbY_)<hs{ZWMn$!3I|->g=Aev{k3HEcr&_Hh
zmI!?fx+MtcgCI-6nHh?9M<v1{YJeiy@5q2CbgZ%ld(nv#CqX40CicBSv<A*9gZD-J
zpnt9IEt4y5c6%z*Kn+5Cfl6!`)&=v6QlLB{8F-b*vJSW_{%Ce`2h%8|oJ<@XB`EuR
z!8-_CFY!=V0sJLnenr&($nZu+yU|vDgPwO8?u3mL{B~5Izh^vI{_3Q>{$OPh+j7DO
zQcK$$%C_TcjOUxGQl>$$7|&@C9nq!%Kt|LYS_)hh^i>*E@|I@)Fu0yKid7rTghq*u
zFzf)GU~DetxHkn~IZ)pI1pTEsICVuQmANmRWOt*J^ts4`OXM5RJ)HLmij4a}#t?7Q
z@plVHgn(grXn1C~1%CJOh0Vw}jnJy9<t4Nb5)#1t5$I`9q6hc&In6e2jq=LYP!<XV
zoqFCJv}O*G2DV1^nX3LAxgJuELt0e*!FCYXsZ8XxwB-xnElDGZO~q`XMi5Eaq`}wA
zN9|G5%FYk3%pS<Z*w>t{gbv}&AcGE}={{>NtDa3I|Dj>#5UDC<4$LnAZo(caa*F68
zNc0<l?km47@<Q5j4~&FYroDjHAH~YMVFCkea`@y63!1WS4GBr-QZ(^)1x}0}k$(Jb
z>(Ax!-yTAS18J4vtHRNeG<x(88o6=&Pt1bum?F=SK6k*qaV~`%A61fOoSQ_4EtPj|
z`JdgQ|8`{0vZpr-F5NEY*EHE%s-WVKbs;J~B#x*!Rj-(YNpz6nfJ~BFH=Y!9LiA!X
zwiUlMm@u`j;-%0M!vvCMsxP^Qx)QcL1OFK7g436>U1pa$gWycvi$bBOgL@<5L|%g8
z$Mm^?TiA(vN<Rg=V(rFszftIgsOx#N>x0Apgv8ZNg$xvW8z5913$@;ykJ@BO%GYMU
zizz9L!=Y>S@fZQb`8Oyn%W?;;R+c*$6{9%#T?=nSOq)PqH>F(J<FMa^;*L_;SVf8-
zj5k1{gZRRNl%wTsFlu7)5Eu#!kKmXx2(uaM5UP|{lyPAwL7u@QxYSY-Cg2%GkWDF^
zYFEy}3s+>C89yO9H0ZG*ik_{H!-tozp2;RqqWT7`$x9@<WBPo9!7K4dnJ^{^^EJ_G
zZjfn3;J4ocXFFAtFy$C3yJ*H@y8s4c`m++Fluu*&P`QQNv-DHwu;VB#(_1JfDTjAA
z5NxuLa^x)>Lan@gfqp_qQJ_~QMX|8~e%Z9&v%67Ioh<Xh%YaUR=E@P3q$p;tNF|28
z6DM4H4isb7quOpzFhndZvvZJPCYA1oxJy1fyr5iRis4N2#*r0^FwDZ6fB?`=)2}*z
z{Le5P^*eEmVN#%vnN|n}aBRm_`5R(WF`u%|1Q)Mu8Z3u=)L~yjmpvR(#$)GWkm5lS
zhS5+eRvRV-`j}}{ToeCD1;3G-Y-<=Uc8VgMij6-TyWAQQ_7n#PfhKgW#leMYC&wS#
z&$w(B@+o20iy{~6`x?VHsKb(=>m#yMikdYJVUD(jboF%Zbe%FHDP^v(p*S9@99cW2
zY?xv~r}<%+DUFclAl_#1Tk>2;;Kc|K#?jiz65DA`7oV;yIbB?y>$^X^=c^Ov(m_Br
z9k4-f*~Z1fI|VjGvy(-9JhYgu$2)RFN4X)g3tj?G-N?|fsS4_hUDAjx7wCmJ_Q(N%
z)8$cIAdipan%+4Mm2Vs}(3v{qi42x86O>0#(;9}ngc|&iwOZM3I=KFrAo%`sKKWXS
zO;Qe>eK=2Akp=;nK?AF4&v(5`iDlf)sHo;Wm&}9NOY+rq;=cD)*>e%L0Ew}VZKiAC
zgs}t<qLFszZ}yKMEXBz7cq`uvmFIBNKz$ojS1oL-$~*sV%;OrW$dfAT-=9uT8{hTb
zs8t19Ctg_K!f-FSg1m)v@H94pR57BLZt^b)34dR`&_K0x$~$GR=Q`ex?xhOdjB&md
zA6Fi$S;?&2wK(9tO4qM;2f#;u^nr6*I4}``zU=e&gd7|%od3T5v~T&VgYT%)P={6i
zCu|zh+vtKKYKg*|^h<VPssOo7N@Rqr1Taa4I(7UhF`70$>59Ygq<MXs;+<r722!a>
z_gw4e2QdNrTlh-llT-uiL9~wiH91Y$HXUH77<LsxNK7*z@tZ6`2Two6^w1m;)=oYI
zAc$NRD)oklhz#5I&`Iwk4zh1SOO?hK0MX|!K1}+OlE>^l_I!X2dJg;Wsd*eZ$j0bH
zTxPB|mRVWtte=sl#`ebmL-ut64z|sq$}wBdGuX<9a?M8{VvV&I8aP>0tmyu#ZB12a
z3d@#&02_SX7f@FVTiZW}`$i+Wh1~jk2eckv!wq_(2f5`c7Z1d1m`(FBtBo7RZ^*<!
z3kO{J`~k&8iBG7W&HsQ4ul$5+K<x8%%`3J=lYD~gvO&o9r9iXUu(+Q39?2C*@;5;*
zWgt9v+<R|1;{eIT9fz_<$Zvd78sOPdR{M?*ez>?LmVODDzhhRud5zg=&+LtHmDBO2
zm^qBqtmxQ$QN@>{SNZ9evFrBZnAKHN&gb?hQEWtqTB{tOc+KFJ&9-YH8<}7w8`$Ds
z%1R}5+nV49ot3F%5i=Q015C1!Xc}me>68hw_^(Nbx?zw_n1Kt>RahROzhVN?R<b;l
zqXJO9>8R+@3L!X^B9=SAgHqRK4gHOWaBnp`kcKG7L}Q3%p!6}s2+!+oF(HBb$4L)v
z8QWetHHRUoQ%%VNv$qg%hf6~pt*MtY^2Bkbt4HcpFRmNw1A_7*<w(cssk;gdH*`3~
zRbd6gMa&Q$*`P*<ADNA?EDjx7LV}g@06As|-Fu76wGzNAWe(KqM^ftCKY~Li(DmKM
z?BP#Cc3SWgkaQ3(AR=N>6c#c<c7u%s97`Udy28<io?u9a-ne<w^a`y%|93)y{s+T&
zJExjTaQY+ledr-#Q^6^*dRm%R^uE!GihD9AK?!0IEPCu~SM(2sLm&s!%f|2~Qf@0Z
z@e5!vc2jx^Kg?#UK}ixJ@cuoEfB7OGO%(&c_ZjorK-`xXSuXrI>oD>Va(ydHV^cZ7
z-Fzkq>Ynu|OUM#IVePA5m}MoUA~B=(U=V2lLgEa~XsRKJxM}nNjErgyXG1D1BaE0;
z0y75w0ySz2Zi)0$P!^|*wkb>tyh<v%419u9L`stZ8y9@|t#IzO&Vtjj*I-^sr!!au
z<+J1}x!OBiS1k*fk8pu{4Cqd6@ErfDX)XuN5HJ`I76Yo^8BQ&-42;M+s?lGqQuX}o
z|IJ293u<=66XcC!H&$73zJCNZCe&j5Oy3ZJ8?VM{4wd~v%29)4DU+2OiOoudG7>YP
zgygh^k@<ttPk|-ni56K?q!YRU;>cDoUSZuoIz2h84N-3t-erKwTjiAQI1ZjmpyIA*
zOu`Aa-xOwr?q*g(%<jKsB`Lp(!CX$lKf0mcONI#09d<Cn0Wr;_E`TKa5FY2Bb%B^^
z!n?>@gDTe>8ESY6YylRI%^srqUZ{Yw69oIgG0Yb!E88T+01TJI<u~A1%k)z~$~<p;
zw$AJc)=3n>CdM>si_E&cuRZ-<b}NBgE8Q4fBSMCzV%ND1t3=nDjxSL>CB|4-!OAwp
zLzJJD$TqJd=?82!NM8WWORUg0#{I5(dyu2nPF?(Cm-A!1V@vx*8pHUOlxwnT6<4UM
z#tCqWuZ-|sXm<Ho{q&wIA2d4rh&?vAlg*&8UtEfN53Md<uWU8vW591={0pd;RYL2~
z+GLVJ7+E@mj6Mtbp$xvs{!DrLKf%S`h`nW#!)o`7`zLps46xLRy>0BK#K6D9#`FZ`
zJ1J?tRg%ieF00ClQ&6f;UjU4Mbo8o~W>0YjL1jFQgL%)$Fjp|}Db;*_;ZqrAkS*#$
zT0cL|;AjV;vu9D*>6Ps{WZeYeh3-e?!3J|EGPLj28WQFTO*abfVofDlO_f(sr*2&^
zxIobRAw1&u?Ngv5Z<zIm1ExGkpQ0S8Sar&*OrX_Yi`lxHIW*Bfg2GW9qW&6VdV-Me
z*t>+4-L~8nTM;a98B(`I@m5bT?_B6w@}uUkt&<}e;Hb5Y$@P7EZhst>S!6A<g@~!!
z1dx4F8Dtiq`iiO+hwYkNmeWmC7pSv5W<D?!QexrXWYko)bT{cXMNq+N_f4A|_%oA}
z3_42XA89ZWmj#aIv=y(rK8L|*Y~^(DN@-q^w{9tZ&BFUBR30+Gdo{>WB97Q);O$@v
z=?@;!2qnyUJ70*+d8MFEE+VWbf!i1fC+e;b{@=93TIT%dHZ2D{%duxepOinV?%)UQ
zqM^ovT5F|4`cKZDGIVzW#JGr}0H_qA*+<+q+8)r{YXh6#A+<Hrd^J3!N8q(uJ=>uS
zs)%}1c?2o@;DH0zaMLKfBW;ft?A`zz+1t|_;{@U%r;U@o%;9ZLuS>sNy;WJ<Eup4$
z9kplwy`P7TKQVYikBtYLUcWPW%89V`J)Ew7?$_?!8sAO4QzOUf>gON+=f|7xUkI7L
zVeXU*S$DD~ZrC*L=Yt17`CO-M_s-e!ik(k?y7w}v&*mBlH*dZwJhtjt+M87s9|ih!
z#Le(y*O@dr>ia_7bIfZn_nyXs?(;o^qG%af;XaF#u`tEbKY}xO02^!MO8$>y5>rzc
z<mR%nZyoxaMe2^GwOtH~^191P&HCe1M9zAYS(K2@we-;gQz+nmmEx8toVFb!imgDV
z#I-$@{bbjIGTrP6_k4hZr%I()b>khT)FMV+Nm*4ylOyc$v&C)cQ`B9~;teXbGq!j^
zD}Au(6*uXOAsF@`2`XabZ2iO;J4r78Sh*bY!S0h>mX1MfORA(4s>MS9GLWDzrO#+=
zYx-3KXj5FNJhmIuXO#gjagAJ?01YvWpF^yA9a}4&C%XtsismYWu?+);1u#i$M?rg$
z>q-UA*%<v5rafXkn3XYnnt-hl{1)1Ved*DV8c3!6rf5kdenOx&xWaQDAY~A;=bHS<
z;uN5+jhNQ4JEfGt8jg84s7<85gKnx5vd6mT0IcUY*%qEGA`XOlm=PRTxmT;z`>#X}
zMZ^5Y779HG7488T$V@A@lM1qW({4y21IB5U1Gbj=bQk=E1qPLU`((-E`7)9Wk8p)S
zAVWCaLjiafrvR|3L;daa9Eg!F$F#-n|H17kAG<Du<)?z^7+i!}J5w8iAL?^*lxK$G
zOd2f1``Egc_3;bCb*jwmAdE3t%qk~A(ts-4$y6BirpY)RM8zP2Ba;xPm}&~@F5}^R
z*<2!Xsaei|7MRRMz9~R@;32{plw4*1UlkzSZaV(IBxE50TpqHL!kAEPO<S5#ld-$|
zS7I|w&=_{MxbU>^-tf|6z9q+sANZC}FugSj)-$P5X7Z4?h$n$++csHo#cpd|VkCX}
zav$D88>3@XRiXv>tsfj)+-K{WZ1G@U%3u_48FuiM2fk5m;0;^5!Ed@4oBGAEJ@uff
zVC3~c;(g}nsJA#w)~5jdFb#Uu!*uGA5|1wa4Jl(_cmmh~ckhldMY6@aIS7I+WIRZd
zE8u-8RXSk){FA)>Od+UPuj|F3r|TTbKJzTP*1V#V`Lqvg?Ws(w+JW`T?$_5yFI>q?
zq^q)48il)a_2GIgSTF~L)<o=4`J+=HmJPK+o#m_zFgs~(*0zedr;DfnM+hSN|1%CH
zJHwb=++VOvTR`jlayCX>M}LqUh24b>b>kgO*Cq%X>jJQw>XRlr&0Dh(io+lk?D3f5
z>e?JsgM2VB#Z*U1!nQbu`aqCf8S<nMF&?R=rl>Qf(N|Db7fm<gunkZu2@|Rn<&j<C
z5af$RR#;CcNia%N&=Rv_3xHfUae<c7A^EbtjK)?tS@Zb)0j=QWkIo0R;0(Y?Md}xI
zywe&yvhwkngM&}T>L8#7OP)jzaEDGzxiFH++OA%N%XA+-hJX2OU+<otfYu`{dwd(D
z(O(a4(@Z!jRGdClxQM@OYs=|<hWU%`cg6B-<{9T<q4s<TiW8<K3>}^gP}(#O9sF}j
zC>o|GI*I^I;(+-np`R^)`eUpMXrbow3t81@3fP@zx;sXXhm@ng_ER=#l$0!&?YBj_
z2M@%hXW&R;Q`xcw(A|{xYF%<zg2k+a_T<gKZVv1dnZj8fbaq4K$gubsg?`RvNYBx&
zuC+cr-OaRhQ|6QLz!@x{++&;xJ22%^A!4d60Ehpq$Wu4%`IXrS1AKeLIF#O%lD&oP
z$c@mVnrM&zyi2rebpC$>a?ZdB3tet(hs`L`nXE^>nkTD5V?BBN_y*+uabSG5Wa8l{
z)Rx3poMVcgq`HqeKC7z9IY?x4vAI9&>6z)0gQWSqag;Tyr>t2fXPlDmKI%!+RmIP%
z!15}0C%|`q+}hh1l4mnQ%0V9H00>C8${hPiGI-OyAFE{|k_{@ufK`_}78#wq`%=7w
z0BVC%lJGS^DGQ`V8vsi(`Baa8V=)EC)c~rq<U`A`RSwCH0Q%Nz@Px|8ZWn1?P)%|}
zcBKUW4BXLuC>792!l`use8HucM}tmd#*oXBhcn%bH<0*(UW;v#DCA*45CFN(;Ew5s
zn(OD+*}cP_1AQC~U;;T(ZPTm2Q3E|=8=&pRuCp-LOW)3!3c%wkej-rVm|B}>XdoQ&
z1a4{I(4qbl4L7cuu?@;8U3p_ktfO6r+<~~OR}w@|%3&3agA4s5I2si3k*VmengkwV
zAJ-TC{iwrC9b)m3$CWEj;CK)^6ua}0=$aYKeJdc+Q=w}Re7>--km?e*d%cla4>N?!
zWWAxhU`O;uhw_)B{3FmJ*9}pCiGHiP;X&13lKy14FneZ~LV!SSyRGu$A?MG31qJI}
z;Vi=p?{IN|F)CA?ord-sP6onT1YsWl4%_=FwUmz4N_-&(*ABWy-rzX2t>;ZsFdK&j
zgV>sgIvbmu6ymijsuCox8Vc?Kg7MwxXNoHE=T)Oh-Oz`4tPJ#4ZbD7?h@_j8<VR|`
zn?BRdivG^}<Mx<DND1O_!i02BCU=#X$yw^GQX+B;mf5~Qr3X4X2FMrGkT3X^WA}g;
zpvMM15%_nrm#{S%3@U==fObAl7F{@KNXUd|t2Hdx2hOCQn}{SgeF4f`2;vN5&~b<E
zciCZJh{*p~3<@YGOwj;zSLCFdGy@Q<M3(>dlz|pi19F@UG}T0%viOU^{)`7;tV|g}
zz0h>nsBcjOg;7mm1yN^&cQPi{N-+AsR)}sJW5)Z4Cr#lxQKj)U+alfR*=SD5=V2bq
zA3H{uDZW)?JFSP4%O>@9x^1YAn2;6qEyfxwdPCw0HPnSqo3Wc&Um}QEApN#8zN2(5
z2Z<=tpz(j0fMxBTanR=GAZ<3*1-~UHgppOm5(!PR$BTP!NEQYsGr6D@&ukEsN_Arn
z4>+J#)ti3n6jIKpl{7&EsJf)F9FaeqM8h!2j8cBiK#X%hW8CX@AD6zq;ELAnL8$SJ
zdA$QQRAMQpRIpYOvp^U+c4x_(LM~R%o03&!KCV~sQk74lk1+mm%7T~-)-{sr3{Df(
zPj*fDz3f}?+w2aaf2KVL%Eh{X1(ktF#O#Qh^(W_IW01R$H7bedVP5JXo>SRU3stv(
zGunOd-_hmG`*V)%ZHR1kWmY*bHrB&vHKD`B-U2d_0d3@rXiu4Ug^bZS%0Q*MIT_&u
zW2C~0U2{+Qg90NI*d@ZF)BV5!mkaQA-B1s4!*?nl^#H0MV0yt=&gXOkC(%G+jOn1P
z)qkE=yuBbwMKZp=N>h?bTbx_F4|eDV$SwCyzKlh}kfnNnFV}L5DnrcF?;SsVdNa}p
z+Z>$u{CU$Bk8$NJ0mxFtfeF3aBN{OS<;GCR<hc|xK>%U9P{KL+pdH!fpr*5j3^Y?R
zp(@+P%n7Dke_fNsLqf`_xJi$O62H9j9a74$(_yGB2g{^)KvN1p4^!@}y?q966Q(yX
z8&hUM)sHS;0>SmcXxx*hkVXsjdzZk5Pe(6>GRO@n-Umjb(MD4`OVmu@5kktz8nf)<
zvoz`ceYu53$}I%K>e<aA03pOtTbp*S^%`bLR5xYs>swA9=S(ra{Jb^t@@vk33&}@l
z-7<&FTFBf)YxFM>j_fWa4p<Np@+`6^=jN9{VpRDG4bxpk${&`w73c1`fwpM&3L1%;
z;mjQ_+pC7;!OG8SV1qL1g<QtzD8rj5eX7gfGZI2=cjv!i@;2P-F$>W1v_B39u7x*d
z<V--oU>F>Sdn>L%9>0i7a7K_!-_T?7LnQgkp2fN{f)tD;0&l5G9I3keACybDn$jw3
zocIEAaU9;uv)sy`#-iSkvQpe-!*)ZSyvE^j=d`PXY0))}7{$r}dK7S+OChaf2MBID
z;*4}I{6+EAsqP%DiXV@4LEHJy9?8Q@CvD0=tmi?<|J1@D9^t^oqnzc2lB9e=H(A1n
z1e}hPaq}020W<)aIK--KH@rz-Gz<;^{TVeJG}BNg4|;l}8pSBnz_M+qYy>A}lfx$3
zpGKH^T~sMGYxT?yl~ay$g)7iPwt)>*D_2I*ddoU0FCvdB2hhX}41a}!{T`r{W+t%U
zTU76^Q}KZe`nN3J!d)=8im8+>m|YFiiN+=+C1C&_dMj?+E}GfhXnffVn2WQuOBsO&
z2#nl~n997Z8u+dah@9q%zCjKfuP~k^1pit8NmR254>KFkeTxoWLqlYpx^mYYkX<%J
z{Nx`&Ejl<avrSvg4pcqSqZH2tK4(7sH1AnDy+hkZfz!bhu~iP{w^X!(QrY{aSFigH
zWfU^?_CTvjnPNvpO9=QenpJm^!!ElYDkZ^7P)jlK9aL6B!88KI9uHj-OrzF&EPXd*
z;zs3--$3(6U8RHtF)&X|(=*}0s(>IE6M17V?j9K)QjR(WWzO|Oh1Fez?3Dd3!$xR1
zlzp&t)hbS?HyPcHcaf=XWXmf`mU}%aDMXIu*XerH$*#Y$cA?;rFle;!W{s{1athNM
zS8soTO-ai~JqTZcX>=4ZI0SJrB%|6{!Xz$Uk9m!hVD^td^@8NFPYkb3b#?@$34zQB
z=!B%gih68|@b5PzOv!7I3g&qJpAMDXg#v(|JEt*T9k^y>@4_T{W5EzOjT%f?0s-nS
zVsAblK=B?h*v42Fz`ks>HM{qQ)M}(KZkj7>j$|di^~4bp7rS0T135JxFvSk(2`b&D
zrY_Zti|i|t3FO2OP-CF-Btd}0Pn4_gA{0x<g39E3>Kd9^oJ~49+c$w*bM(>uq$IYd
z)raiOB|da_;e8ixSyl07|8-b-J#R`q0^Bf6AjjXhf6n6%&neo1rR3|9U$PXE-t<^E
zq*Xm;M_v_=ITd0xW{+-huS#4l4Us630Unk7d$ACn@>&r<0I_;!cw<b-F`Mg!yKqbd
zD)lMEa=aN_Cs$BPA5)fN@!YYz3+Z}~Y4qeotJ?_ARYM2#b^Z%n?^rOwEW8<8p?>CS
zvPJi+<Ps171zh1=%33k!lpQ@v)||G9)Z%^mGKj|Ww#-3swz1jFEZonoa9kJ<MJl42
zpv6Lkf<vRP!vAK<WT9st1DP9#h|VUb7T6tCI@DBwEzD!|P2Vs^2fvSb(hVtz#WbQg
zXbO}V($OyO32QZbp-mQ3Mh^g*(jIAO-|sGrsUa^RDHH(b?BMwTX@osEQ|6c2p`C%_
zK(flmO>k)dPHB+XK@2rqhrFav)2>|@U5R5ABX2Gz;lC;$?bNR3kU7<3^TCj%!%U3*
zJGe8heok^S68)B*Y!<hF{qsHZAKnhOCQHDWH+SR_-hFSfr9oBLL~rjXbh*H*fDY>b
z8ugDL5`Y1e8vU;$Qpf=0mqGd97MtupPo{h_4#j9gjP58F6!?t9B(@8B7J-U?yDAG`
zbrV$8(d=>WTfHe-mZ+8siZJInaKyFlBSD?(;RHGIXxsMf>kaL^@D0BaE*Im=PC;`V
zcySa`!|ORFcdxh(?_MCCDAQJ@@gGfkCbpZ(u0I=9$cM!#JNET{MCfGi8C;i*%%&-B
z0{7V&Ew7m!L*Bdr%hfD9c8?dX8~Bfq_i`>}Yl$Z$qQwh*FJ54O1|Ygku;o@cxB8T3
zS0(nU?7lGp=IqA6(Be_{Fm@X;o8rcBnryxE+=1~GRel(<+9WJ&8Qx)NR5Xy540LL-
z{HXg#2tO5T0T^~<KecTRyBf?jE)Xm=CpnzTsE_VHcrX}|NH6H$ahuE#4XtOVIXsI)
zjKk+XJo|g2ON)@->OXe1{~Kn`_CH41fpHS%M2|t@1@a)~BcqqIs^84>tzW-Bn*kO6
z{Zo|2sY!-b{8l_Ix62E!xC$Ra|K=Zovn&>Fz+|?nJ`b9$$gYdR*m{M3&x#26WO1nc
ziSz3xXO=4&h!8>lO87D<I6*$ZPS$@a2*Du9Z{r4!58?ov5kWA0<@kooM31SwfCE{|
zO$&&zqpD>}a8-xz;@oDBi-MeQ_AQ&vsakkpau1(>iU!JO(Kn6GvpeP|DMJw_9X@*0
z2JJE@>5Dj(6_45wr+5Hog5va}%<vEUb7x`u@{_of!WhI)NHl@KnnjTzqGt;}BfG#>
zvg&@tl;DCGhJOUO@*NQjF*cm>9rZ2|9b^DJU(Zw%Fx%f3A5$EL=_U@XjW~w}p(E;j
zM)~+yJuzee2+0SoT)8s+vOX+DiRNeSu1VsX{QTH!h&p~f^4|TCI<ZN@eE{XkSs-L$
z%>kr^b1}KliejU^pm*bm3E6}6@aY-k;~(lNuy#S%^gUq*4HzFvk8n2f|NMoOZx^P+
zHto*(U`^d~DqclbwB@vu*whCA=JifIaR{gb&A>`n4;;BWdr*?De_Amaq&zC*(6e)3
zLt!v73He0aWh*gal@Vblky3)$zZ7Gi>FnKS(0&rLkt=YiehP1l5%8!D5>*L!Mt{mK
zc|tWn93Queo{is_>Xkf>-FqN-?!BS>fO_>Vkkj`-IVt!&DiNL>f{7A)_2>9!E8z&(
z%?0{7tqj9JhCJ~?`_ihijOGmO0-muyjI|B+wY=W`w34Yh3jJMB0%>{%OM#EJ+E$bi
znA0e;lW`LiN@hbO4bc6B!>>cJoxXu2Ohf#<e*~Toq4w`xu#i6^`|JR=Y9o}Z(oUX!
z0}mBXWjXBF`c38ARqmZA2gydzTK_Gbv3u9y>uRX$72<G;2i@|=a5&H@6x=3)W0eAe
zO9Nqt9`cB|N-hiPu>sJP^P@}xK6X3Ny}&{xvWQN}s~~JGc-|$i<Fu}t!WP?gvhjlO
z&yeQU^Cs5{(H-Hqnn3^t*kp@)Uy%4<C6+K9DES-tS}sc%sPd@m4^+m_Q_O*FCthA*
z^39^Swd)ARWL5gfndHrE#ryJ`rdQxRzjH+MJh<{{txuN&!oQB0T5W49>kZhPgH1_A
z0k8fik4kW!DNBX2-ZX|M)g;!uc^CGv$D<o~$7lF`0CfhF**9oqZS1_0qju@P%<2in
zmN;jQYUrgX!jPf(Jzt!)Vtu_pepb#c7Afq(421jAR$D@>5BO^1!rti6>;aLFX0`Qh
z&3zFWed$oT93HhFf(01q<Z915ld}EIVc5s0E7<xT*t5SmDvZjD!sbm!Hg{`%(U%L;
zdezoD0QdTw*MdqQcu}BUHd|Hv^>Zxnk}d?z+{d#Gdm>mW={Xqx=}BP(R>7@O;E0={
zfA04Bq5`typ&R`Cfd0J%5P=|<1xHSDGNmiiVbt|;Vrjb3<7;Of0>~&uLR2SEn1;XC
z4$4FQU}kbDer|BD`pP%yM#B)|AgrQ{Z)OGL6Pz*VSenlY$P$rWasP@Az`628B07)l
zK?&sA4a!52ZbowHB)q<LditQ<E@UI@vOr7Am&vQJEEEfv7`0riB|_#J*?nv*BvQ@|
zM=nW^R#&Lk4=(Ja!hIDGNWEuJ4zr?~&~yr|Lon636wVj(fjr^ohtf=36w_XVRG-YI
z^aaqx@I;$4PnH1d#tv6z0&c<xfY)4aegZ;}EHS~;*LQJlunhw&xBl{dV$h4z(9Vz{
zsYVzIxs)HW#3tK9l5yLoJ4S8Oo88&I$m^y9`w<!WQtZ&rI$-3yzH*{z#M-BJPZ760
zs%~4u8~6CuJsa|9$Qslw67G*$@bL1=h;@2<o3VzYFVIMbStjVxcDZDSX8RgqC!&^o
zV-Bm~EY5jefWwK`E2-EW#CbRC_fn6im!phdC3E>ufna?;_$>?4r~?nC^}5JOv-lAK
z1(29WJaO^Kk3+!7Ow8^ej?5XUE84VfedE!!y2^s7-yowq4V3@{MkwvUyB4ouIPEGv
zH}eF688iy-4}Xl<k+LN6a6MrpXs(@+yrM^DVE-|`S^D%;-QNlwu*TvGb+VR%0>X2m
zxCT#M>)AIUctPd6@W%K{h~_ktFGS!p(+Beae+w!47UsKLpXR9VU*g1dB=T}9fWwk#
zD8>{^!PDB78vck(>Um@8XeB{wn%mh}0JMSQa*v_uu9^P83=I2!lfYsr23p%XqO}`m
z<Gye#-&FlMHYt6uL&@m-kvm}i4#ih$8+MoVaSGuRI(%9j-)+omYW<NBjj;ye<vJ?n
zb_>x|BMc;2n4S}Qt=0yd+f(p-H$K1K<eL#qaOh9UpEk1R>z~nKvD&cHW<()_?ZWOr
zh-|$Fq5_2cEYWi`@9}v@8gfVFJ>SC4K7{zlvm2;`dIgiSr1lVuZkzTi97nG@%8S+|
zK<<Am(Jq88falv!X@~@5cwz4y`1i7IbX7OLAb}72&?e_{`Rj$(i=rz@LA3rDT3T(H
z0m3f}+3oSuiu3FAo5b!5a<8BWGeEdk`_WV#j*4F8ZQs+u!TzkWvH#{ewnbycpbWG{
zUX1$Lz&z%`?@#(=2+rOP7eV_NTYjW+6uQ6l?doTpfcTJ_3FNLh@`{=jCFEUWMqS!!
z<*`(|4x`;77Sti;bHSI7(U)@QOPq9&pj1g7Is1y@a1J;WQR>BpC06z}vdRZY&K4`0
z*e-K8ZsHUjHAof~{t@IGK<&6CS75_laBmqf3|jWu-@-N?*;nRg`WtXWCcC!*FqoNH
z97IHFac2-7s0&n?!*&p4lJNlPn*#G7|1*}99i0Cbh~IB4<M9>P<y3S<iSzzbJ@p6S
zN0ijtgO#jVvWo+f5Xw&QF}y~4nW>OC4lY8P>6+z*DBKOi1F~8f!-wSlj$_F^M0T4k
zhJbd-ht`g54lS(sx^LNFJdf@pir7I#nacH6nT4Dc$_X{!BRb}1aXd3?O3<v1M4v#F
zc&!+#Bsy){J)JO+I@Xca*4$xQJN#%r_-PJe+6?;V%nO%qK`g@!9gGOiPv2C`Mf>Cj
zB+-OaL0AmFGVP6dtj1{PM)%A2>La7e1;*#Ev<;oR)uSy1c6sCebAS*-Otm$^Sj1cr
ze-&{gBvo|K%93el^9bAu-JB=Lcx#sQ{;@gEV-l|*$vI4Z;)~0V`4gBN)<G%bq7VYy
zmJzF?GwQcfm%lv1T6`Sh`r>WA(LlTH0pz8FH?oQ31iiD%1V!;+HHWFvkdK?7cA0b}
zcj4OWhOR)Td^!T05l;$1Y^&8hvyMCgLEtCw2%ZR_eS=-0&GncPCK+flpzLQn_I%qm
z@)|&CLIRV)aj;3r$z5N(x(p}ciTR?-@XzX7H@IuYjvG7H0$hDZkzK>_R1ZI5y>7;r
z3BFOO0CwP1`umVo!6TbYIQ!p!9D5K6_PhEs{pK(1h6fxFQ1A!^#zke6m!k4mGni3g
zCP7v8)-ar?$^i`8i{Sw7J_7-O|87K^-6>J1Cu$|0s2au1FC#q_>{3Qc?sp5o*3o?d
zn_aJPBJuQ=-4`F$U>8hBj`f#!r7sK$(de|+${Oe%Mqr)+5!Qf`W++!T1)FX&osXUD
zoy3;yVGZcn%Mj+z3yW{wQCU}?|0Fy=x1thXJ_t=0`UAlIKF*zbgh0ke?m(`(%$koT
zaI@nr@H%7?bOcqulmuGa<YyGMV;v+2C#|*ju$)VmD~nfn&py|4_q<uvI!{|a^>B)N
z`=iU}h5vkH(E93=Td&zZrnPsS?^}NV<ip~@-FshG4fXqMfOWk@zv^pR49T~QTfKBn
z`=n;`LmCGA)$izM`BUy6XCsEbEht_6bj~iDhqq&LvTxlg$}jd#oAGDlANwE9{PQAK
zVObH1nuC0Oef#&?!TJ`V^<|@R3Ji7BAp#jX8$nP)?dbytNM4ni%5j+PU*a6v{M~oo
z)m`{LT)TWezg+fv$8Je)9%3qGBS1qtllV_NogriKJIF}=3l^Z<!pw_W{kxpnUpYVQ
zh@1X3SQlL`P0jrZ7h__X`Na#MzDd%0!2+bxquVv4;Jstev17K&x~weuwm)7JKNru#
z<fsDFIIH_gOyexNivPHEml}b=*7%OTfB%{SrLcmW#QwDduv4#VV0wqowi}-I<F7M~
zU-1^{r`?lp9$|b9Cr@p_wIe3K3#?^?6|~ExMw820<m=b3A<C+?8FzC0)^4b02Oi)Z
zV7aD};nly->3#c}bwnuLY0gJ*zMa0H(q{g{+pE4>=8e!J9=m)27tn3e_Ermj`TeZ%
z>f^s}c60o`qGFZ##V61SQmtCGE#Blv_dCED;O=g`Dfxi-?Kt<wpX1tWi$ru5bGQ|E
z)3p@{i8A2dhNd;eUZEeLJpQh!qehJ?i0}B9{@P^^{Jt&)mtbwp!T2vp{rIZadQ%B7
zG0~}D0RpBwckgl<p!;Ipsa1ZX0R5Hm_mHACThs5rfqnPzGcUb+(@Mlv)}DvqFGbk3
z#hZ*%bXQ%`VC%hSy5puzo93Ut5OosQ=9Z`BF8kVOXKT-6l(%KGW+T(uyIe||h|4U(
z-nVMAZM|WIp1{3iSY`c9a_2N8Aeu22bcA28L}0gZ;mTK25+YEX>Ab1?AC?IJ)gzsc
zCucfdocb01l%K#AME9=s+R^Vo9VDA)O;+YUk?2_3&p-bh3l`Kym~QtiGaveGZnPop
z#ful6jMo!spm5hhzds8|K6NYQqT6!O`<L9^ak)ODm#yCXnMeq{a5P`$kT!|W8ZB4u
zKs60ZS`WwvzJLi{#}8p2FKH!-i`K4P>xMWeAp)-i@X~=RaI=E}czjx3Ufv?@L;Q_=
zve_IOkM-|!;P&%ai%}cBqh7g8!|T=DmBn~2;RtJExt!x-v#RuhOi~SOk?hfRSqt2%
zd_o5waPp(B&+y^De(zHbneq6+eG2M!q))ETUF+WZz4zWD;W5)?#flXGLT_uZf?a3{
zrZ<9FCyZh`pb7J%zaNp|)*OX;zeexi`Wk%Yc0V#6+Xo;wQxw6$TnOTW%??TSC^$N`
zLu0+`d{&7b*v9Dy<eEP?y^PolC*U9?(aqtLCwAgOpk>oB4KYa^f(Pm<-#F0j*Dc+Q
zmQ~uOehrSyzK*Ve-W^<Xu?7LY#p`@KG5T_rXR;(yx(w{B5ozSG<)d{tjgQqIb~fzD
z7QLOR-X4mG=t;DUXroK?Eh7D-zWjW?<V8Jv*;Jn$@8OLDZW#Oo(i901P)E6rF%8_-
zH-h)=+m{(Y$HPy|cf4U+hrbgza`)#87dF3)c4@QC4MHA7ru~5bs2mf#dtj-)@0lCf
zg_%oaL{`XgpA;<3;uG7|{f~_4tT(`z4?%PwqL>ff3AM(GM^~;mdGaKOQN%uot|+ue
zG*%O~VAr0t3Y#Y!Oolqz1jX^vJK`S~pO}2o#?0}kJSd?hed>fHIi$MAnybf4a*sLg
zZa90d*La`Tb6qzuaYV2D*Uoq+!3{>>$p)ZD*9G}P`(@n_>AlFx;%tM=iQ8u;H0H1K
zqn$7t;Teb0E``;mkE+NVH7K(Ewqc$u&X5FyD&w6ycfMA`B)Z92;S^nxd^&yX{u1Yc
zoWTCI{6TnnpurpZ#`u~Gy8ztGyg7Ps(+1t1ELsg*M?2M{D=f}p4z|yo2M=~cP65Fb
z3R-dhxNvN{hFiOZm8IEYyPm5D&!M?m#*TXyf|eCQ;-f3H@;)JOkf835?SV~UWw6h#
zoP_stSXjVP*+Q8=@daDV0#bGc@pxt`=g~Q}=0uPo)!?#*!IfAVuWi$CLyJCXj~~x%
ziRZi^&sj&{SG1NUL{KWrAQ9iWQ>QvOSDWXX?F~)+G2g8PpQr_H>>?d}%=Z(|xDEWI
zX3bW)P_xTA{S=W_-MT-_ob*ZU+MAX8!`iDl^^i5IZo^B$9|PF;`Ms^{(6sU^&CB{m
z=B~`;Fupo<Prq+9hPNc2LADLtQ!s!0lUlX-=r>yNO~$7}Xm)4W@e6h90t=@of1ZQ!
zimZ+e-2MDsyn59IMjAMrKe_QZp}Y#H@ONKUR8bV|0HMu^T-M6|I{Ys`x(fM%S|jiu
zw4Pz``<adqcSGMD{eH|ekB{tr9M-D8{;xOwhpQd)ogAkNNX;s~<$ZZ;q>P*0o9lDz
z3XC8ssh)f59lx&Chn<!m@Yc2O(J5-CVzDSMvc-G;{P|}VJ@3c6DSh$s<xrDj-MGo@
zM>+swd%hH^K3lxG1DarDH+9(GVv!3586HCOt=U6dYHVpKDEqPpX=%;i;yKwd)V0b1
zrW(|j*FKBkbwLrgeg5c@&#_wx5g&I5w<b+g?rYp|-2szp)QCG)DNA+f5z3vcqAYum
zoZJA!Bu+<Pq^{!HJN4MRcdxPst?_d4OUIe23VNA~o;ffFUdiC1XWFyv8t3QVzkeTt
z1}+^?-i&I{fr;&GB&O3wa+By5V5nMMy1#-2s9kv??W7m69Ni|d6VMR%`GwOLDwk^S
z)WP4Y>^nsV*HEcbIeSrYXXv>-ga^V^_PPTH4xkbGm0!2Zr}aN*8FlK^DaB(3PN4lH
z4GfzcfC^4VLd1ip94|w_Xn6<Ww%ejbi(*aw<xbm%8xnRSWTL5DYza$OyMT7z0e`tY
zati37kaxJLf$jJ)*mG{vz-Urq+RLUgUKg`-#mgf**z1uH!8%q&PHx35T&W%tHrdfU
zOx(D|%ufD5!r*eRc3LAqG)Ou-<81`~2pc)-+~!mL_SgIfWB%*yTeJO?bbHm?bH!-L
z+8wz9Xss2Xf}>Ypth@gD>#vsPo_SC({C|7j8<2m{eJa*}bUa$^fTE5}a{=pWbnE&Y
z3z5|F*Ry6v&I|sfs>x2?=k!0KQJ_f&gy>U{1KraV><XtTd(bVY>@<<MUf}}Z|EucM
z$IPTVpMwyPrIppNG#9iG-FcLCvNbyFdm5d4{%8D(MVda)*ofH)2)}sqQUL5*z2RF{
z(6#e-rZq7AJ)-wRM~=)?-k)NdmKLB-PKV>N%|#7x@^rk$(-O7wZeohv=gjfnv}u!)
zV<`5hN!kZqPcg%<%v8N@7<WNlh&(p0+qfSVRkwK~T4TX5Y&q@dDQH<LIC|trD+KT}
z+F=iNt;>Jq$I>Cr%`^7;2iPh4UvR!$u<dj#st|iKQ}>rY+|T(1T<1ueVD&ZNI|rkp
zQk|5GRnv9b)z|X!LJ#2TC;ZN5T^n0jtq0AqRW46!bnD47oO6~<jglEZnvI9wt$FR*
zwKL!RIZ{lJr%KG+w_L;g{TN`95D{H+JbFEtgTH9_FJ}Qe!!bYbGsQd8;j7_!U%&23
z&)btbC(>s1>ecy}er6|L2t@NR+SP-0RqP`rxK(1}g9o$mHm$w+zLz@odGpilSm1`c
zBN2qWNe7C^!y#{1GJNW`VMd?YLQa%Y2D~xH&%^z26yR2G7Z(@Q-+Haa%R6-U-o34n
z!Om^|$2FkRgZ9#Bi#Mclh5}W&#-|JVN1z^KDK96*fu=KU8=H5o_j!9J)5Y7{d#Yz#
z11{*G_IM`8VCOIK@chJBzRb!RrnvKEbObTS3-UMn_A+{)@(<7lp&5ZwBClKzL&c<g
z0rw%^#Co%M%1KC(o`5~C_dNQD$?{z8K+_o+KXUf$*~%$!qXqBK1Lf39yZTsmMHs}!
zPA++8o)kyN1RH_?&Bpp0;`B~#usFF)@-RmEq+29JJS{o-bUg;aJk{W~cjE0hb=Xl;
zPp#-QNOt7gQ8eRSFvO!rn;&-l=4Mw=AY`h{^rH;t<d3)x@gw2b($c0GtxA2D-iG}R
zTSU3OtkAVmtfKR_Du-D%R^~-+Za1SbwFH}|&63=-y^&LZa@Mc$0pIWRHVxbv3OPnM
zL8`!ydcMcszM-gQ@r^GY%-gKn8~g?8JlW=QMy+qfpS}!@HNPEsx8}BCaYK{~Rnv#y
zk5!sV-n-=;4u?U?kq`mFkZW)@uL0zAI*;FP8Wj~~mH=%8tY$ba2KVU;*YghC&C?AX
zSH2vc5J9yzRQQ_6y`K6<pd0V>k2hEA#nIgh)<1v#JTnyvtsHfv>maCrS?qDZ?aa<d
zvBv`5uNHbknB@dG{_+ozD^TKVgr?dPeC3yds=m0jW-qRMG9os`|8$%-kH53P^v3q9
zVb3uYbYG#HDfmJMmnqGKC@3?kb~5ey3>t!d(NccR%UeV#o@Ed2-*1QxoUiz|a)!zh
zXPkCwJz&L(QpIVgo5?TiApK}n-s<Kpa|z2Xef&6W7G93-l<|ga2A!GMB;S*!t!#_p
z=5Q-YE{L!gP(0(@Ak#tMF9$(rM_y5J9kjFN1Xj0tgR(V^T5jC0StEJh{Ak=Hnav)~
z)%5dYxCyhiS@z)J!%y`QO{WfKH-k6>Q8CUu=JA1Txp@W*hy}YuMmZJakirjrJQ*e)
zKM>Y$0+qC#QQh1Y`BER9D%UPtxX>qidh_9FUf$l<*Y)}9GuldqO#}DV^=m-8!j3zK
zpN(fsZ=NxqHXObD=}5D@3Uc~4FL45NBv)epZ7N<G?7V*a_J+o5J990xE4kvCPf&}e
zRB%GX>RLTo6GD~iW-VIO35B3O=>dLClwVY@G_8#FpSY?~@7}!=B0##mj#Gyx(8k@!
zE^#DV<(owqx>{xt29H~96UH*>)ZkBx$`4q$@y)HCw>7Y_@Q_wE6HLCvsXD^bl9G~7
z;A=A1!wS`#Qc=+CTCVY7QS#wKj!V-_ojGZ{xknFr&+OTfD7J2LE_<N=!M7dn@{gz~
z={tPAD&{iZ>tYWGkeop_pP34@MV~fVdJ|eOOTx2s=gxIe01NS(>RRGe{=V(oXIVES
ztoEqL6AK*S2aN9b#Sfj7HN;dtzH(*5^0is7qZF^DrR9(`G?9A2v$Ssk&(hgsVI$8>
z9HBorGVUz__VXqxcV$O_#Xm;wMk{e7#!E7o@AUQZ<;%Bpg99`ElyxG0RP0{6=OQNH
zO<%gyep%7;+P?OaCr{@48nKfQh56k%A-Io@eu}S9pW@rjyulGb#XutVMfQX)WIC8M
zJbODGAVgx!HFSOc^^PaOr8s&I-`=h>GO|w3>9@puxDH8!&BEKyahHJ>Y_y>0){wNl
z6b34dt}Nx7vYFZ@F%h5?c${?o{HU9nnPK&{>()(2FWJj(>rS3Dq)s(IYvBGOCubAz
zNCXl5XnRQPpR$8iRLkf2$F6?O0n8^z-(4EoOc?#|T!fJ4&b@nG(#=pg^<LAj=@M$z
zsud4Kr>)*=`hAAFr+X)Orq3uc4GeUU@>cnCLCsAzXm^o(4NA^urUEr;3|`32E4ZVz
zf#&C%Zq=nrmlU)}E$d`=28yBAQ7hT%4M@@uztl9Qs2Vp`Y7xTee^%aw&L7Bs=T%UT
zw#j9we(Am4X)}0_UmKTxM5Nmp`}7RDDg6%xr=_J)YOdfEV5L)$1;e=gYYsH}!*mH;
zvx%SThO$F<6%sVtT(a?loZIjvS#fj9Cnk3I2Ug2htO&sAt87_|M*y9fYb_HQb3_%K
zny!1!n92x4saN9YMO}9Nl`DPeqb|~8Cw|Wsxm^{HvZ4EOFe2iy2Dg0eWtFTrX+oRC
zI(31nNVF&z&92`Tu6nb(IM%ig4!~=T)9OHI_p&Uw^YCwtUS20#UG<h}M&b<67bxXC
zDy=;l(|h#n$&G8*lJAfotv@45RcJT*E}H@*E4})lu4|i%s+^xaVhQ>46WHhGn~y=B
z;TjIf;XzWc0P(gyE3)GSN5&*L@h9-{GQS{y>;AHqg!nR3L2byu$Xsr^#Ke%~EWgd(
zUS0zS4_<dbxlWa(#3>8;N*!fTE6N@U<P(<sSdGO#Gjq~c=^6MfLulhA*;Qz_{lC4Z
za(|b4chiB?Cq%Fv3VTq@`aoe((RI{}2S<Kkf-Qo&h+#+3B<#MU=mf#|9O^RTcoD;4
z;A?uZ<z7pc=+pIHbb328rL+$2&%Vgg42Oygdz7wZ{UZOde&Y#;iJ9)B@s3ywo?y_+
z1)$2rBa*XO;}N#&f8E&#H%fW4jGYiceqS13Z`t7Wt9l_NvPq+>wFyR;Mck^Srt93f
z&?n$P(wM`e;SNo?vhEhWTu~8W;Ttt+4MVdwxR$y3prSJ>Z7*=n{7v4G*Nj)1P!ioO
z+1`dQl)G3{&#Fn2jgpAfSkuq=i~meeg3g(a`2liS$&)|(G(u1Bh=LrHb^W1K^PBwQ
z7}k5teYyT6%YDuQd6Q!pZE}0`=;2mT_P03Ky;TtELNhZT`UgSIUMS2a$k6q)zm9fe
zS!|~QDyi_X>TedxbHjG>2aXUA^kj=SDzX7+oVhADCWg%`5-quYg@wc(^y@bRMPqi)
zw94h|4U(_q&>R;0Pl!`@OHM@Du$_jo^%gC``y+e4(5)t;PtK${RRj18-Nbe*An4;u
z#f<8aE>zHeSlzk^<^pnZ7SL$7-i?V-gCsRQcK70o^9*N(Cl4MtUww4Hj0aVxk09L@
z?GJek+hs*-yG4F>%c0UozGD3LnVoy}>XmM^{g`5;Z!P+CpdZNehlh-5xrlr`kgv9S
zbLdCGx@Y*OfU4PVv65TTLUdQk!AJ5#t-RYZZwwcWM}G!un@U{?m!gc(4#_RR+XoLC
z;p~zWhZvsj%1O}upk=Tva*BfMi}@!Rg9@wRjG^;k^JKuolW3>%fOOZO5-=+LoBgFz
z3l@M+O$9u;@-MS1%{^`-{&%YskXhU`8tY{j(D(zy?LMlDJIksB2%0kG7=1n#Rg;=o
zXr2AqCx3(Mh%|8G+LC6%HFP|{x~Z`U)M=GjP(i#DbWN7v1oC05q}2a(U(7%RD9gqr
zv2-wt8i~xR-z&M`TY7;qkz2jVIVyYJyn^*B?NLb#cv0d+#kyn>)ZL!?eB<biUf3wY
z_eaOT37DnIdz2@UBT7T9k2qBIeuylMsotkf?y}n}7+BeD`-gk7w52Okt>CqGf%G+?
zbC)<yg|ES6WerZYdkJUT{ax<l?R^2SLq72|8fg3T%9?EWrm#h3DhjmCa+hB3DZjJV
zS@$(yPlSzwAL}p6kfU2uJ36k0D$%-H_3G0R&n85mst5uS)3wBJS<D0PutgQ%P{0Ch
zZ+WA!Xk)T;z@T~n%#g`t3l|;6u3P`cIP;15t3ubwN+XW-1BbCCy$pMnnDHh4_)l=t
z{Fh?LjJl0YR3P)qponl&e(7JhrrAX{6F~+bj=HQHT5Q6B-^vMcvWhp$U;?Mo)TZ5h
zbRJ9wmfyDuIhdp~`H}5hV1CpYrfsKr7xen}q>N{}92w(}u^+4Dk0-<3jR~G%^bX2-
z&F7AJ{}Sqq9wn-~Xl+ArP9l}D$A2$9WR~lBn*aSWz|>SI7@0;sV2AK$?~47O2A*mv
z2*=nH;8vx`Y4IPz>LlhxP64^!(IEv^wlmwbEjRN~6kD-)MpluyjiT)rB3q%8WFEi%
zqT~a7+{C3ObKj}D@MQ6DoNBhY+uKd5r?3Z9i=`C6zt!c}udyaG_goEFu4<Y1`CF71
zVnuF?cl4W{(H0!<&0Ng78#b(YVFYKFA=aEm%inkoHQjjl+5(E_6Gegh2yBitvPGXh
zV76tt&cuE4qsf-2GxDl!1=6)U<d3-)Uj;CiyZk{!`g_fbwO8o0g2x|k(PvtR9gH{m
z>GX~~*fhD#O&m-1Pj6w}jr9Vg>&^0w*^9;va3{f?%0B^RCw{{3h))0|CgZqkxkrHv
zpm8WU<iO1n*p`W)aoueOnYlxq)XFz*-RzUu0%yUV$OuTo1~7Y-;tg+5ogwdZ=^C1i
zJz<)<l?5Fuw?y`Y^g(*>++~~jTu+SN%zY{xydDFrE%$W=>O1!q{x*5abi|Sjg%%Tq
zH><+F-`a>8$`x7j;5zbiF-;GsZSjU^zo+G?PYFo}4rgIzY@2ONu=z1xJ*UA0RWSWK
z(`Cw(DN}*yaT6?VOIXv)MBkUZN*+hu6mnwy^A9yIZ@sLD9@i6e63#@wak6U7ITlD%
zBFc@}Ny>^PCBP(n{yNkg_49APjRgyz$DH+pD$&}AjR+9+0o1O2SU&Y$A0#@P5Z)uV
zfVyQYbW>-EosX~Tu;2aG6oc}1gH~5mRD{1i|5%Arj-R?j5L=xyj(iL=$o5|g`3yUD
z6NKhFFDtrIdsQb;LS+K<#RmkH&7$O;DslFw`D2#T*#J)owq+CAY%_!^>H#wtk|Vh2
zT7I>*EIz=?)5nju)k6xzBVsHgfQ%dhN|;Ll9>njsP1O&z>;Ubw(C?EGfs_AJ0tyC?
zep!_m%jNJb<l-enu-^teYG#010kROEZCW}u*xC5P`<enW_vO6l>Fy}~bG%$;Dmo)Q
zO8R!i$~5BvYyN!JX3y6`O5$ieaZHQ)z5POeJ%346=}L@6&PqPth&_0HfJ)776e3Df
z4o0K6eF0N6+!rn+iV;(}!#pk4q!ad4@UcV*LgHgkb8kek(}El-qQf(!a#<=jV8!I4
zp3b7GO4(wTJZC@o<dc@U=v`>Irr#9}{;6_!RagEIBRf@I4IUMvm;YCG(z8m!r{*hX
z=4gho1xYD*<44-JZ_lk#>_BH|;%|)zM5V}m?%c+T-?>4UN;^6Qb#G2vPl$kM$91r_
zHGEm+va4`UDt8}LOASG!7)9%L7+Va~W2?8m+8y(bagpsVn^l(7IRBn?EkU^Y>C=+?
z7q)m){<q0Dzjdxw>YJUqV&`_Ry$*S|_tft6+b1;|v~Jev$or{wwRP|OG-sJd``VK}
zoi+FMFYZ0&ElXbDH)h$38doCU-kp4+{yeW=pCr}zVpOYLU$pt-Wt%HQ-^+Z~>srij
zs@h$meb2v|(6~!c@84c7+qW#LBI;?EqM~I_XLRXwdj+24(hswA2hN`#58|GD`a>)$
z<oc^>@U`0KF*NsPyw{(0egQQdj!gIkCjFu>%I<DQ(KTbdto0n-*AOfKK-Xa)Ho*rs
z<<epRZ_UV(`i*H==Vb*m*5Duivfi5T*U~6Vgn%GTF#HWa+RuKpBWQ5v&H?-vo$gau
z%ZnqoZr!4if!is}C9!JUxB-sWKplpYUVjB$D|8SXHeHS9K${wFNZq&XI1ssVsot6o
z1xn~-fwpdsl%zf`Ph1UVcpPBT6~BSL`h}t1>UZQ*cNqJG`YLs8z;JE|fH76}5iDZ^
z;*Nn}Gf|fk{oj;C=?qTGPBeyP1OKnFy2%W@Qe7jAC17Jw=LwL+$KchUuHvEokFhV0
z%Q0=ke$vQjvX+sgvL`c?5K=uRCSsCA42q%{rX)nE9<LD*WiYa&$C6~rGM11UQc2X<
z5-p~pA|={X>N}3BMdp3K@Aut*z3<?;@B6y1>s*fWIFGX{7#-tLJ_zbTbs=$?r($KK
zr7(U86w#(AVK0zyqBs#@Ge<pFBm$RDxCT-f3hZ7~ZK6y+nxVFA?CmV>#2{$#FTYr(
zFiHpIIEtI|E417gIN{I5HNTMSH_?T89OeE%`OpC78dg>PfqXSK`7jx$Mg!aT;>lzu
z@Xj~d;@{imR2zwGCE^+_7Fr$8pj`{P%BOiBwFfH0AEl=cZdte)=)`tbK^So>gT8so
z;4_;kT|)A>={CdS`{*DN!xXPq#gtMFH4$sSyQ4Yxmedk2yp$oMfhXx(P+{RG3jSf=
zdVk!K8;O_7IjRE%;$+Uh@-w<MDe%E=8q5|@^onAk4}4}#lyhoxp-CC}enGH!d7=4_
z?#h9$pzqr)tf8(zY#$pB4-dy<t+)Is$oGk*Q+NG&&YZjpVV{<<-w^-1T}3uEix^1-
z2%_4%X<xr+5ZmiKq#;HV?hRSGc=5Z0-dGxY@Ye(Oe(~UF2on#ku1+Q<O0gxC&q&Ni
zH<G1B8-9^I^q_Ipx=PV#RG~^KBd5kL;y`tSn;X_E@QdsG#a2AB1XVpy(cSjV(CgL<
zur4w4n?JXX8MQ$972@KmlkIMB25ZjSi)fYjq@>{=hc2f)PANEc?AU=(smP1Eu;&nh
z5l1ziw5gnX4&TyiH8`9TmM&fDSn`cS^$#av*Wg3EXjr;r$)pr@Y$b82doA&`x^(1-
z`7}vdDs)tyza91SkY_d$Nh23P(cB4d`DxgF>qeE1b0{Aszy<Y!rm9cPI<eu<B7hP>
zQSZMbdOW06P`$fHr{C2Bh#d=?Vb<_B(xI3QAr?;DS9WMm^XX1Y1(zZ6H%(|S9zcuN
zswgv8$g7jBnuu<TJhH!V?2SPAuOD!M=>=;V9Ft7G7UYsI+Gpiaj*^0%Am5dY`%7W~
zH<Px3y&^c}jwKEa6%Odvo-0<EM_XJvW=DH02CdUtO;Ub=^Jm+S&PT7-HV&tWsR`jU
z^zJv_ZLCg1ZOC(jN<%te3nyR~)T|I+q!YptcG+=E`^bt<9N$5`3$jwR)Kz4T-Jk`P
z&4J!m5M8KxP`3-+qSN^l-aISflKM_GObdS1{fL4j<^mcA`sZ)1c$1#ULe`^?kVC0v
zfXAuMc{udTb4T4OIe-=s%AY6Lfu5lI{r5jr7!c8gYF<20L=%urY3i?*-E;Fbd~DRm
zZm#=;*KW&kv+6mM6g$A`P?o6WAR54o=~u(j5rW#1)P<VEHhwYF!S;vS+eR&IwEfkr
z%z?@m2MP(6y$~Pzz+M+9aWvm?bskoKI{_O5&!-DE-G(UYVVV~ccmeC_)1AaDZajug
zJG8M6<j$5h)L>YRHUMF}@g3JdM`(IXx^ADt=&Yx4Lood&-H_H21C5_N+RzD#Xo0XU
zcW-U*<r$i9M1xdVTs$RZ9>@T69pZMV`*cHSVB-Ax&6{pb0e7$>d|+2YER5XG1kKKt
zNnIF9tX#D!JlP(k7Dmt=lSab{ba4QI*Jcqy@jO_cgPYoYN}=Ctpk|g3*9kGMDLlfd
ze#-cQPt8Go+_r4wbEjkIF=}?5qxp4JOcgssABv0|gm~x=J#T0^+y<;<A}@%>5Z*j6
zD<<!b!6|Wsi`D8=-WQ)C?(Cj>&cbPc!+x7er$QNg!nHq_dSKm(?P=x`S~iZWK6d#A
zH&72PlN2bs2+tZn7oB?KY4&{g0s2;nB!k*!pk?|tlrG-({=tk#R9_PlY)ngl2e4qr
zDFB=aWA9Vf&h*`S?BEl9a6)R}(fbJ}Ahv0~|8T_)AkA%QBlA5^@&;6Ax*5b{-@w?J
zM4oS&5#eine^q=J3yw1v!xb~gWksXZ*iNc70krnB)>4raRYiABdx9=VW%UO(86_jH
zEaW0IULP^wVi<LoPNJ84insT*!xDKIyVUWj5a*q>Ec^Os13Oz=H7Hj!j6LpB3rU5r
zU*!L=w*@9S;fj0SSnI-exAi0v{|yKSIp35!iM`xaOAhqzlWQ3EaS9NjD~Yl5VHV{s
zJTu~Y&xzJ(Un>ux75U`SxV{*0)UZ=e@b)oxOaHiwH=t;3r2d|cXOK)_G41h+lR$Wr
z{ZIv%^){tr&@E7OIW+@3KtiB2SFQ-f)eW5k4`YSB?zDVL-$?YS;4xjJdM=j8%GrGa
zPOJIlbauJfh80khhHbyetk+`>)hP3igUBalLxXJwjEYBEsvQI<Ik$&0(E%Csjb#Yc
zvduU8(x>PyhJMN7btN8QTfy^F_*jQxqgx(EkOom?2yJFZm($&GONlGk0SD$V!m{TK
z9Cc!vwoq0<>ILX7Kg#gDp4UrB!|NYP6JusxMDoFv4<?zfNa*BKXdbDg!73aBt=F%A
z>GW;Q@mXyp(nB>cNUBI8>t)*v>u#g9PK(_?M<OF972%exJYOt4U&Wf3BS-oI%U5eF
zQTvT#h#4iAVG@X~*WpeNyi1|(;K7qq7G21oK7fgBVLLpg>_6q`C5|`y^w4a=$o;)`
z)&7cYVhW(O^2*mpkK=MI4b!aW%sJT{VQTIeo|Xo-0S%Cv#Jcm$Q?qvHvGcrBQMwBe
zMw>+sJ6cIh7#&zICiR@d&;Go0spD%O>Z{VUt}8IkDeH_EumMMHQtJRz$3ob-XV0H=
zTFY!nunWQuCvcNT`y^nGR<7}Z{%;vP&MI#obYV|(a0sjM#i^KkV(zo`zpa0Nu<H6i
zGC<-p@qzO?jVV{e@sR#C8^JTBOwCu6a!0WS3EKC8f5zb%`iXEQyUqy?S6qr1Fr{bk
zFgv>>$4fz{)2GVIvD;94>Q=t=I(haeYfRdGZ{aYCzu9J>|FKiin=!x^roXgbS2n>f
z?U1(GNr^<aAG!#)A~;$*F;ySjCL-CJuz`bh?n%udVt%;kW-G=&W2&`o&*~KUsZ-yv
zk+ZkS(r4>8|MABk6rb7jjgRe#HWGPo=E{dZ(d6_txrMrG9vdmA`n;a5^)<qTPW$%n
z_r7jFZypCLZb4?kE2#@DcLU{5Ajy6H`BDsqv4@Nrjs{0eeDatHpFC}y<&t(9129iK
zJlwo?^;>EfK#b)G=mU@6U`FwQb}ByNC~&U)HcU!uNDjELuIgIO2yXgdIpn5A4r1AS
z)80#zG*I`&4jeS);Mmcu5@y+x?9RCNr;4>i@$d;^f2hsC=+!^;z>Tf#@NxFt<m6m3
zEQuo;T=S~<rBnB&6TqI&l`B`~$Iq^>PBc4ap9~M8z8wf|*Vs)kbko(u%c5>R$uKP~
zJNt~TL<a&}2|XiML=W4(W3_Idw{O~9b>V4N=lI$cug99zW~8!7Smp|$n^w_PrX&8A
z`j7`5dj$408Zz*{G1_)*qXD#M-P&k64DYKJ2!Qa_qqPTpDra|sla%-TYyg8qSU%A>
zO&|X7z^GEhSdXKkz9~G2|1Q^|Gzu$1(_Orv8(@N+CUYHbWF#DV<MZc#OJ}wmZ=i*y
zgvj%>l2!*szqKBJCZ<98{WM`D&mt8&U<@C>pHljeL3^cuEmfzDHHQ!&8K%5sntT!E
zbh7+eF<DU)ILYa2ggdXxTpkKRzA7k;7F{-cX=&v5=^5<B;p4}Xw7s613v+N5j`YA-
zJEXSv*B!xI22V!2<`<d5=y(`mwKMq4RK4`52?X)?O_>+wxj`1A{G;g--y3g-LPm7O
z;r)LWu(Hg8n9KbRAB9cp{q%;9x7ERuM~>XOg{ObIZFqxhhMiqR%hUhvEdMJtwT<NC
z86(9Gkibg{Fsyo|8MD?WF2|5x;WucIzsirUyJUxV&H6GlPx_ww@A-4YF!>jUo8?q?
zkA8pfShmXW@>dlKe#|+oJFh7F$r$avqm4Cjd;LcbzFQ6l(s(ugbFqE%FaEh<`t94d
zAuPUE)urT~bVF3S+>Hrd+Xav&qsq#GF%K~Y^avJDyhw-7o_HV*`7@*8Z6x_8mW|x6
zw`X$cBB+7BsY-8G+_)jA5{3Btz;u^=!xn96hdrIv>Ypd#IrTLWHus$|WRH>tVo4hp
z&yvid-saj&LX@eui;avrW`#8#{OSUHZ*l1~@X!4>@TTeer~S76$^LU+D|SdT5?_|B
zF8Pm)MsPZci8uiLIt+&#74_V3V8s<|E^^p$&lEPy)mRLF@%f?WQ@!>aqUwV9*`{c;
zulb2Abd!l1iy>@nhu$7u((Q}@RN5hAj$8qB7!fu?EMGsLxDsZg%CGl_A+;Mm@IFNB
z_pH&uzsN%VN*PHl!yile3%FK@(=-Z#xKnpO$Nm8wdjGx3+Bb@;#TT>&pEc+<TA9s`
zx^RXqyUMEc@$atwotb&yil%GJa#T^p6|=Pk@F4BI?%vKH-G1z%^t0Rj-^(jB(+q`k
z@KG#h*pJA`KffOPZ8PVJvB6f_i95el@Cf(*y6h@Wv`$WVFa4zO_5J`+Dq5avIS%Hx
zXbJx-II|W$H4v&diRO5$!u9wk;*!-Z!n5hFgqe<9k+JNO-L0<`BaIw3%&@gxla$uz
zj)y~yXB4mvrK)dEB7GRX<oEaZx?Nkc&m;$a%pZ&}T<%c@`rQtU@<6`&dKvVeH#HgS
zi7?AlUFCoorMCwG(z#=`<AFGs;4x%ct$qvH4g3P!Gyo_ax5w$}5%(*d)fedCgZwJB
z*M3(ecKk`m7%&VZ2B^QdYS}NpY&jWq<L8g~HIvWbPwekX8h!dSoxaoAD(d;kL%rVK
zv*rwj=0{r$0mCvRFlIMkjPYqQ<e46*wz%U+4c4YfMHUj!qe0SCG6~+#`)@pB6g$5V
zLh#7}hHfyWGtS_7Fe}MqVPT_xA3SJK0H#lV2Vl-29;0rIbgS5@rU}%-+7v;Iu=G>h
z>er)!!b|oSAtK$Cm>Fql;&>JT5?1k9mBCxY_gebpJ7~<Gvd^D<X3-aiQ7mduy^N`;
z=2#BSoWo9;nBt0H34q*@lP4#a;ddK0{rgKZb)+B3Sl4rLi2j2Az@O5e&%ZMIRftn=
zh|~4}3Le7kcAjcdAt(*-U2fxHeTy@r4WXTJ7+Nwk+9VDWA9cUke&BfgYF(9IgXdf{
z;ssQ#MfsgNs??+z4AG+(QApBxCUM4jNjsiEqDA~dR`(F0o3Up}N|E8}l`G5FnC!pD
zlK-_B`=5W-fKiMrxd<F=JTh6w5}-*b0j_jgdY6&&)ZFUmLmWW;5^*GU_K>jXS9NdJ
zk|k9R1460sgo>0NbEJ50b~|&J8SufwH?e=~?F8P`I^pAlBY`))Ph~+DSd3m_`WiI$
z#Y4bpryleceq%FiR}Uzb{cLsp640TezVoCx46k0eRSduXt~-Yr(^(s^Z>U3u4m<8S
zcf%)s7mI@%l)e>#3=|hzGR!1lR*&`?mb7ubKk+iKL2MRPJ$Exhw81}&(}Kj;Ha5c|
zG2Rt259mDP3??qY8a51GLVy^f^*&Xg>cC4ngh7EyI|y}I?w)hO`6pz?9g1c%$i<0^
zJkN$#levX`&HhnF1)4+DC>L>}7!6_S0E2CjJn0vMfy9xtUS9^mEnwnD*Z28%B3H0V
zL$wZbuP3Z1&7dkfJN?&ONwZ)KQuBhfz!MX$V{SH<fwM2dvFNv+;itRqx!M6|NL=}6
zV8Y_3%$c+QF#0thg?WX8YECR;?0g=~Qrl<<U=4>Q9E<0sJM*!I`W>Y~&YwRYzYN=g
zN=#UfxU+xUS=iMOsvj-k=CseAHwIC0L^9B34|A{k)a9O0LKAI$NVu=UO>UO}1?<$j
z2Jn(7&2Csnl{xt5z&Bg`&=KK>IF)+(OXL^^3tg7|`s+_7ojLk%?eVlu{NwTss7plD
zAA57xp?(D>Nsbdw_6SZu<e}n~gRcRpMF6R*EJ^YvIBN5fx@<5r$h?>%k%$s97?9AX
z?14J}_)SInTX>D~?8q4oAVK>ZagNh!2rK>l^r;YWhxXkdRK?V1{#RUd0-z@o2Lz!>
zhN2b>xaVGx;eb!eW)uD{eKY!fQeW>?u^iw<o%Ni-k8mET?0}-Ic8_u^2zDK8%$SpQ
z^!+q$lgLBmVVnef=0?qHinpj<z(^r$bo&iAt%#gD2y7PhnGa6=_Bn=`2T_m!BJII`
z6ym^!@wEzbhsvjkeJWfn&_Lg74iHMD<UO^e;)^L~Q>fShV+$s(y0m$7r*$tB>x0eB
z2F(OC*?`07k&C2v>h2h{*0Rmo65)#f?kNLyhD5e9GTX&z7`ViuyOvksAzL%;^e~R<
zMl>4`4J{5KZ_<Ql$b464txE4d<o$&4XbWXCYv29%^swF3M?R=iv~8Up?V>3j^${`^
z>Xd$qhWcXy^pi#&ci}KPo0n*he5!8ddwo;;PmD^XW+(qICphlx=O`(bRgR@gkvmv-
z<^U3OJO8wM9w2q>6P4uXg8&eyj5K{w>X<j*b?RC}e;`v}$DWv@z6e6@2~gY!e*jOi
z;4j=(TvFf3%_)d_zYQb4);};0?K+Vw_K*B^3xxHT#HG^sV#@Z66UBh|b5jgc%R3*^
zW>}=ILx!=_gQJrIqQ5-CZ5i=2{XC=J=x|hb+A9`Qi~;2f(;1bZDVniPqv&;zea6__
zyY-Oa_>P407(2dD*axMF@EBBP0SMXk0z}0P{OZ@yLbU$*lgE#5#;ABLf%ug+`0am!
zn#G~B<CcS(5|R8Xc*LByp-}6!80eS#;~940$<+Hr|H0U?g)THhN@<ilI(N=s#%j2u
zbL&2?5cl{VrnAvuw?K=x5Dt4^XP{Q&3>D+Z3qxbdf4`~p7pKpnBgNsh%L?3;4A0sM
z6+Evy`ko+?Yi#opgB#x4f12Iq0pm>lIMDeWA;lbdo&t18KXkE+bY~%|0o2oyl)B?Y
znAjWPANWkih^0}R%ezC%TXs|F4yyT~HQhU#yo*zGN0oj&BB|KS+GgDw1WbGM8qh9l
zXo?l}0`rtsN2R>`+}pP~?B_UenARbHJDkE_8K4yB_PNp$2xKz(RLX`dQ>AS*^;lq&
zk6~3YcR7&~@nafb8xyHg2@dJgJnxcc7Y9>?&5&BqYf!kE?lT3tvLrb~CiatMi}(JT
zDLZGOoG9jmXrRtBu>RJ|zYJJQPM$eaQdR>8?*!NcFr+D?h}XsJ$*1t<u906;btDmA
zZF)h@n!Ih~3+G{W#a8yY0H-_=wDi|sw-Jvd%1ap~CM6Xn_*xGLdTwf~7+>CA5~m}K
z1Pclz_853$bKFLbk_+_77X2C4`dl}0srd^JWMN@}%FYk7;TVM|W9hhn%}=2nasW^`
z!-i2{^E>9o2Ve%1D#DdO3@t!dbiE52Teb7}dB`3BpQ+pRO+3JV%yA7Mmln-r`)cyo
z<jiJDjwutWIzVTfn6p88m_m&Rnt}O<DPUArAh9;;#~$^L7!A$#r%uXB!94KeR>DZ#
z#*Cb9fO6g#uuX`v9DGB|$v$`BwA2Xjq5|9E1Kk^aApEUVn<a-b4Ej86r_Eka-X&-+
zj;wvO&!2F5)THei_A$l9%j9rcW~L^@7aTI7fAh)r3?`)CW&<UJ-^zmLOS&Q6ng0$m
z6NDV}@#DG%*Wju-%#MyzEf7di{1nLAw?ImO!eBI0d#y<7?fz(*#h^YG_gD}BjC$bX
z?Nb~gLgfWtrA|srOl0FmwiveVP-tliO{_ZuR8lqkueTG)u}^t)s*hFs(Zxr1T5aJ3
zSU5BPa!pQ8t1GlH+kz1Ax%v8WM2_Zvy{_V<!#2EA|5w;i!-o%N9>OQjj}b1U6^66D
z#efzb;gC|uVhx{=f0q8J4CTo0vOnJU?0Dys3&!A*{^y7uNG&Cmt_)Pn<rpVTff)qp
z0U4<9L2l~??eq2Y2M>O46%N$G@~AczHRhrY#wohGeKv7sjF1Rrk2ms8?LWOSh<i8+
zN^awqEQ#DC*&yP>&U?qMT`_#*$YaSvW}oi|_jU&RS|{quDqW@>CS)`M;oO1HwLHkP
zGHgZNn<LcgXL~%wCfY3<z5rgC8nOutfKkh{9jl@s-sbqBL%l1El3E$HBoQFBs?Ox$
zOvSy>fRZ_rlyjKU{1g*9DhoZ3^Cm3V)azT!e*Eo+c4?HBX5EJKqqZg_B#;46Q#X3>
z;K6ynsakEO<SKILsqqD-Rnf@zr;k+GVS_51O6dPb43uPi@7}#HYQ0eTVr`6q&X~Ho
z1t_c``FPdDv9YrPfr(E*CIB%H3Y&5u4Ommq@gIqvwC=3-VI@YPn?LRNnKLu<b$Vc2
z6nYE>B3@MG^k+=(?RyTpp987Yqi)pvLSzjHR((kx@<AuO|5o;i5k@68<%A;*04LH`
zfT0rqr<mrK%i%#-hj?K*Nw{A3erFiXRIum2%qJbLc?38YqL(6fbKhM%26A#obv{H%
zj7r<RE-nrvEdqrv+C}JrIUz9jSJ&yp9A$oCDBheb<%s0_`;CU8LUEegr4*=3-J+N#
zby`{GA1LaIVzMNQLtbE{*jM;1#9!RVL-6(O`2M9|EW*;H4pcJaROo<c*f3TW7haf6
z7rVAUTr7AflbWpTTB6LN0t9@YsjiL%MF)94u<8LBAokxz8giVE<HkoX+vL_d_--fc
z^{w4B51PYZ9zxK|b)p!OWsWTEAR>AHgUIJvabMPo@~r+`#xqF<EZ+RDGk+Zx6Rx?%
zVk$Cjko}X4C`Y<I5)lX{7Zb5@yo4$sv@N6fhy^@gQ-C5go4EHfJQ<Q`hMd70NDSGw
zZCm7d^m7HE?PD<PLDCWe39VTO`Ql{7Ehhq_V=)4Dh`Gh00_-fp1bB?#3;&Mj*UA~u
z3p$DJcs}Whif=A9rUo4P;S9}TP2BnK$B)BD;W-B)t~?&K>e_XHX^mWf&pEgj$L}?A
z2$V%UsW}n}?iGa8HjA_oFoVouyo*-x=>KbljFD5Z;b!vgP*mq2oBK0T9_f9?1mdzD
z1@i_(@^~V7A`*FL@ii1l=;t95POLFc78Y2I#ztHlsTq!l(W~hZ-`Dnr5y-y*<-$bH
zIN0Ogw~si;sD_AL9MJ+I>3%1|D>nqx00FI+bMNE_&_N?Kv{^)XhbEsAZ1_12_~Yn$
z|H{hM{TYN3@04ooHTHC!0*PJ&RCJHP{=u1_;MX5WloB=#8Bt68I1o@1Y5*v8PDXk!
z9mfR-w!rf{{&RC9=Yxd&3XBRg8?hJK!%CUhgTS~IPGSP?$SXepPXKHZa5O5VI44BS
zbyA^y(;tY=^hJA6set6B9lpB$QFtzZ*cTJ0w8~B1PkbOX9}l)ftAF0<%Y9|!Dzf~A
zpi;3a+xSs<b6FklhC+r|ti0~-{|YE?l!p%;gm0sc9XY^YdaL@|SQ|V-qd!yA;OATW
z3kqSB@V3ohFCSO{T($WVL2&B#MLFLBk(Vgv13eTrA;<^ORM0~!<L_sz<}@ReU1&@;
zU=kGmWvg7=qR||PZ9CTU-iY*g&^lB-P}=@GEv<9LvYE`g<-fSPTBRUN(F2c$mrjt_
zBK`tMSnG-*!DxZJsjH`FnPQ4<T!wnj9Ptm6|HT)|-+?0_>gE89x6lT)(I<ZddPAXl
z*YFz@MB-Ud5ep!0wlM<5@21=vv^m~cw=)A)O+-N=sbcH~V=Y!yO`wb|oBv~YGZZD!
z9nNE}OoZ)!<x3kN5d5!IZM8OShGDeA;$_RG?2dtIJ#`ytDwXE`8~R$yR22w652vm^
z&Z7Z31#C0OMu8g?aL94SMlHu0m1T(K8-kHb!`ZkdKh(B~x-#n9?`CyoVw{13a^jvF
z<^A~u74_f_aGn9wh3kstk;ul{Fe+IhxP209Yd*D_9Ahxz(Z2`_g!re~0;z=HV89C@
z{G6uN_^+i};!#vYGO7FsQrTskzhSc|>iYX+N=!4-@};Z}CcYOLOm9`r87r)3<3?9I
zyYx)-0OWD|Ly%}~MWY9>e=bqY4E+DxDvj27k|ASh5p7D;V5j18@OOd5Xi$k6iQae*
zj}K=XzUdMw`N;Xx|8r|eW5w2b_Uz}DL%$bW3ta?a87(7~=dYQA8IGLq`S+toYETvC
z820gYcIi1UkykizRzg``Vj{{sPRvH^?a%C7@SuHg4$~pBl=9k{RgZIA4WnZ~9k5jI
z>?3dlat8vitpeMFEz37wJsB7bnClcsas)3i($pE_WLZpRn=7T#Iq45mSl2*)=d_Nm
z%zmKs&^2_Cu>W{7c69%;*iGEa>~Cb@26`Ojdui->$!-d&7MD9t{{svAB&#lVbuD>V
z0Ksc-ju7G9SzvHPYYrxG9^Toe=quf^Ag3{g90k4z^*l|S_wU=sRy=o1kJkgpRLEjU
z<F!$cF!VK%nD}pk1!al0?%gJ{;t!qwa=7b!fkUn0NxED7>#vU|9rSFYc8SC=Fr5zS
zd1E|9A}AP>xlesaNfMV_9}`i2T(TrJ*7ym}hV^2;0C@(KVB^qc#nnNaX8-L_drchP
zF?SwX$^Q8V0_FiuK_(+-Tmt9-IFkX5e5yE+{8_a#uwq)|hv3j`Td$<91TCkqcqqFa
z(3=Ay3<43VU;~%`;riQO-L?YwwAJ2!XJPK8=dO3R#!CNgufHs|dhn3#RPQHVcY&J^
zG5inox?!I^;9TJWt1gt|KOY)2LqYA5m<|YS->Q&7i0P;^=uUnYcrcw&{PEiZna}BD
zB$Q#!1p(lzpXTzn5ip|gTsk&1f7-yfcf$}N2o??^V1sMmkno6t8(95YE<Ux1a3Oq8
z`EptkiGNSU4%TsE90wG8g7$LfLYQ!eI)dn+XyjzZl)WR}j$w2eW}c5h=)8Kr4>AzO
zbtvyy6!z;hM!`RV3Q}djm~MkbxF;lr+Yg$%+fb*lCdd&b?#El}&TXt9!9ni&!|NMW
zd^+_lGb9_S84fi5T1-9C4r5B>^y}EA?jrCdruMna4e-BRJ;=`|oomdU4fvpMFrjf$
zMGj>g7cX91SQZ<GTV%DELF+56DoheJ6}+yTOtdw4fvYh$Kqa_Mts4{jLlko4T;Y<+
z8a9QY2Y<|+GTh+Tf13(-@IC1g3P*h2-h&Z53y2&+0YF3LN66Z<-(DoK5XmjeBtBOu
zl-LXTN^xZkaemS}tX5o6l`-rAi#SZrn}rOdKS#JF?F2!?)G>ZDa?6&oj25Uhjg$q6
ze7^c(S(`tp!-@upYccFMLM`I$9`8`FS+eS1RbWZhpuV%}_amW_I*ty^!w_gol6ja1
zP<otu@)w*95$P}#<0U=F7+Io%3b0%fgYm+kokl-jiY%FO*X7S+V2z*W?f3Fn6et}i
z$pZC-SmmI9<AjmKX@Mhor<A8J9w2@hfr7v$^)q*!kSM2ujhpuNB0~P~ns0)$Ast!0
zsT%Cdv(J|TzC%Ic36+GcR-96m#g2rRB$Azn&g`iv7<cfNgOn0r-;UFGTKre!J_z3?
z&Khg%0<p|*XsHp<v1Y7}>^D$jgdWErw$ofLRqI~ct11qLL(=wGOpFPK26=k>poH?t
zmo=tQcr7v~{5(&C;6gIYD5wao`rSNKQhdMvp1EDg_Ywl*8AuSTA4mCiBS!|QVcO|`
z!`%m)Yig33TO<~fkV9@(8KCk+melTTS2yUxc{envdAp`t1z-Zrpw5LxuN>rY*D?DK
zS;NKJhQGi#L@7I`wCX#|3&nA}{QM0#aDPwQzN$@HmY=E{7_wNpg^nyifU;>vgtKtk
zm(=ilaN$k}DvNw5UyGBU@;Gc2uOWb^Je4k8SVV$f;Yc}EU<2dl;xC?a&--yvn@=~T
zifSPMO)wlNuUF}2*g=KpQhb+m7=L^3p9YS8(q8IG$Pv#ldH1{~d(YT5?ewcxuUe64
z>Slm1Ap)0TijTGC5GInjSoi^{zreoOg^h7lp0#)73J<VJ+d4z$CCeT@I8&B&IU+Um
z&O-n4p|&gAq#2Wvf57WCQC_5#r~`5ho;H4x)2C-N2Pp@!2$m_;8EYl-%l4277u*1J
z5V)etpz?0PGB>v%4k)CFfTHn{-`AVK|0t(`q?PgZ;^mmPzbdZIp`eKOx*?&G{Da^p
zjr<?6O1%}<z9BI?X4C9>X|=fxDJi+bZoNp+pBq1Z;&(&4>FwI_@<{&a^1p^X-}j%Z
zhC_R)I$vMkO<%fs-`u*Q`Fnd``E~!*50^&l{=sl+k%8`=x7ok^GJkL80GF?wucYkT
zo1<Ou=hRgPD(1B@yzpdS`kS?NM{c-1sj9smtDo}oskKqb$(Ajlj5rn9l_(>`qw6+%
zk+yCP91dfZyoz)OyOC$-s0^1X5ROjhJ_PaOwc<<;E;fbov)4M1h{oL6VI?_lusBVA
zg;RG0u!xD#EF9y*zH1;J+lu{ApA4KLJbc15%taS<cIt(I+4c6phdPLb1rCL9Nf~D7
z==EBY${SCr8GjSwYE~6}gi%>t`)p;KU~g7yMj`gUkURvl4ei(b)0Sa?-@yIpX2W89
z5!&*Sy54n`FF;U)AgyZ*LNkK@KaL44`QyVUurs&A#!PT>I`V#f6~bt>;3;7fXTg`_
z)ruc{D1{DNsMecEgElzQd#M<K@r|k;MJva!7aGXZ1xBh3f7~;hiyCW7rZRPrVcJ=r
z+SORw9v3fNN`MfE24C&T2khCADBYGQNlL-R)ztxc)M1RtOjOHn-=0@%;TtDY2-Uhf
ze3x8RmN^6el@Nrj-tG_|0QVeI%;>g$&Pc`GOe@6oeh-4R`EHo1Hx0q_V3QRk+LR4v
z3#4nCV7MV`>9Cl^$YGznEJz*xW7_ng#|yujpX!4GVE`hRu_ueka&F3~JCo$scqz{t
zS$elBd#yxgC&}=byOgCB<IiF?qUP7mlG&3YCCV1Zx_=LAmHx7z^!epd*LnM`u96|0
z3G9FL2k#1UTW@tT%hY~mVmo)PCStX)K80fLGMVli3b`<)sR^<dXVR<m967~3ZEFY3
zQP%AxYg{Lg=#^J?!>5)*7|@iIdRp=>Q&}?havOR2<ZzpVa}?Gtff)+eyM22{hdlge
znbo~8Gc%N6A0J77hv|EX?8<<=T|mz^{q$xksG;PudZpJX<Akq=g@XLpi+xc$HL)I-
zZ!Gz42xDNEirm-W-Hh!G%%x8s4kEs@W?h<moMqE)N^&A1ycYr>__(FN0LlkH#4|ts
zA_<kQUk7*Nf`epgZ1M!-a)y}s9rHjHa6J4!1}Uh@vMMOTim3Czj7kVcZ^r-*7An2Y
zff0<b`hEy3uT8T7(!$2=U6iw3L6-f7fBEh6%LVZl`IXC1{(16~Lf0Pf^1l~t`m}ZH
zwfZ3M2cx|l>2|%-*_KcZJ%x}NmryAKd2j>{hID8JyvlggN)o1E=+oh*5B+}DNW~8D
zkvJaNkKxrqrl)^Eq2<aTX&=4|_&3H4x3VWEe!a5I*(x7UZAB}HK{|>Is0H7MS_P{;
zIR@<&0ef*in_Ithf+|XL%s@38CJ5}IYY}XCy!P@1cpu&L%fo#7EI@LyC(Nu!b5dBy
zaaCEJKE@QzLa;g0!6FQ?>lI9?VPYP`0*bg~*@w{Jmgi0nevLtIUC~s(<jAEYHs>Ws
z0$?e6e_B7Jq%LSB+D;mO`eeu$g(Pk0;I3fl;GP}|K$+;2+>N6v-?IeibqdJNEzh0W
zL0w;`=>{JG3x{}cjL0$Y4)^R=T!?f?dEznc%F@$LUyV^H9P`b$-)&|jV33YflIj6X
zkyF~%I@lJg6bxXR*i0B%d*<__F9Dr#p48q_oFVN>fL(9ie~5Tkly$>2bmeJ4<p*lU
zC)2b}ZrE7oa@E*R+uQ{Qtve^u<kAykzXWEWjjS*0__VQ_YR6F)kFCu?R&S{g=W){h
z3_ve~>42;KS{<A?kHM*@Wn(e%_m8q_90x~K@=RZ(c?_0R8%K}H{S#3Tpz_rl3g4$I
zkQyeJVIaxnhfd^szW?Ad#sVWKF-LdSK$Az4K>Gx2nV(Yn7;ynzpxK}8E79Z-7;AxD
zQ!z2<IVuPw9I-b&zL;~0!k|@@i#ma|rdkNz5@T-Saz6PX{Q4U-2Th#p*SDE)#7~(t
zDWF~fCmPSoGcU7W05JaUnz5l)i8!8WZS%viI{lp4=+sb9yJ`hGySIQt8C%Oi5Y9&x
zWHCdEZtd{4T)On6U+v+$4()VekzXn&nwcup&%q8T7BBOl&oL{>`*a|;PUv?Eh(?G(
zg#{<XC9t$IdpS^Jv4kzK$DVuMd6(UCK*@Mfnx9*nh#ih*naq0%!x(e)d$?62tIQEX
zER!P=hjr**_?1p->>*edG)66|`07u55m<j#c?#RMq?P3zGp-wDBTao5L@E>nJ0s34
zL1F>YQ2r}PkH$0J&gt^6&uPL}Za?_*wg>H{RgB`$Yq@CmJd^_20m3J4!|JfO<ep^G
z6^rqYKo$NvaP=5l^<7nhQF>AfMoeZEglgObJnEi<ez0SqT5zlnCT&NAG=uZsDe871
z8C*$2^rW4{BkRu3P#M0)r-@3Z)4-%{bvhE4P*{wBsq1bw8>DjoN^FQBCVs~fOCyb*
zzxhh;P1RaXbc&yg`(e}2AoBc4`^S{nplhNv(cA+Q)~2Sctk^Udq}?K*(|7kSofK}v
zu?@{Ok3%$(H-&2XA(p#ZoXPkd_S`DS&d%;lKMR~p%MD5$ws3qYER16%l>*sb2~(?V
z3))da9{b|dx_bvSxjUU400sk`(VNLH+h73Q1SSJkbMLZ}j={UG%E3w391h7a5P4zU
zqxYygt;&ctTJ|)!3}fCA^Y;yvth|VHddPsO2M^3x%^~s_VZp#s8ND;!^khy6W(%7>
zj+zn;SP1~;El@Up6@QTu=^Yf#BL!EUa!k9%A=cMtPpHvDr%o_NC=qFg=35Vx%;OXr
z<R+mZlOrB-HcV3hBA21+?iWy5SsD1bBX%Q+BxeUgMm>;k5@df1WXsM7*sttYSa_QO
zI>aGkA3C|kRLsonndg}f_SzVe)nI`Y_DjS9{sPh``XRKj9ZyE>;fOwdFWK-BBl<fd
zFm5JRSqlz>cOR&Mp#t>OsX$wQzAS91LbfWL$9yskt%Ba7A_U%FY<lWmtB5@C0E^4l
zj0}^XQI*{!5dewlJ-ynYkcI%1WJyQ0J_)fQ8aA3OXIq{9f?)^JRiW+3($R190JZcm
z>!<MobQf70*tj0}+^@^45zt&uPyL8Q&;kfxM54DB{h^6V@~%NdX5#0<`ORJXayP_f
zoOvS!qN|@fF?Ch2!=Z3DZyLPIFQ6&_@vlmXjwPw#=%k&QU%fB^c+j?xl9LgM%n8uD
z`1E8!EwJb3Md#3x;4^K;jPRYB#b3#dkJ?0DVbeca=?tC&*M_Pe17S7UvL`7m47$dR
zaIiYN>PB~{Futw)V0B=W0TdA4AVLHTrz_R|m}D`=1k4yUK~*ufb8>Pbz!<8fjV30m
z6@RMAN``^F3w-dd-ZelIS1`+oPIK)LEZUYW%7|0Mr6v)3?w7@jsf`}~Jk^`c!C=86
z^z=>9SW~+0rL2vZw`|}zJ|Z#9x26cK+c|>^lFMg7BjMg&2Epa0AbBv5RFYHYQD83K
z@zPu(nO}P6U*C(hyR+BW8wi2%rI$maP$;4FBd~=_?n?77e|)Xb_BGsCN%>P)^IX(G
zv+_LZ5E=y$=)$XE9^1^Je-?fvuNu2V3q!CR&p<DlBY#0x0x%IAK`r4gf+dAUbH`@1
zKn}Vv*kc)9Sq)~*?XrzS*Y!S#*TGRR+|{Xg$1Y{LKG_Tg;P}C9<z3VABYKL1Y=CK~
zSzdLpUL+D4n`Fb(2X6R!Zykqeotky-@WVXq+NX(}&-->ps6_2Qh<Mn4=>$vwvE&Z+
zn$R1A8lrI<`%CK=5ng;w4dfD{e5)-$WuK{CG8*l+qX;cAt7K~(8vW0lN(C;g#WpfA
z>;d)XiA8bt(@wR~oq{R?t>C4HdBB`S_Z~KU-9c?{bWEHBo=0GIxCcnC332NcJ>NV-
zA(@ICq4k%M=-Q{ZmmU&8Egp9EB62f^UVrx=oNAp;fc!<DbJI={*2%j9A1DESWlxBZ
zd)hg0>UQFIyb)H-iKLG*U>YL`N~q%-aR_gpKA5emeQQ&vAfknAz{_c<ugW9E2FAFj
zi=^@dR)6_6fGjM6&jL&LnwA?*8fh^6F9c7~&9HFYbyD*n#H>yAx1$e>tUNdtozuq#
zu7&&aMEz?kn|`yAYXnBG9McQx0JG~*Ne{rBbzLCL?>B!A8gJ3N+LHo}X8JudF2Ro4
zL^jkONrPo~$LQGGc$X?q{tC?9gQAt3NqI-Qhc3N@Pzx0VKn)vnkLtMg7_+#Bt%sPx
zw*8R#Cg99V*yss%8S9MbhA@i<om~S_M#`ES^`B}JMK>HG8mspRa!YKH*P$|DfC1%!
zc_X`YVP_Y?0hXr$SS9rZ1j@4dt&qmgA<5rFx-2WOFQPbtiEM?C5Gs^%bm(l8S2^29
zNmh<<aF@qb;2Y3+6GLZ(@!?!JA4Fe01HzmSPU86R$qqF(h<A(nrNYR&0*?wnRYa&k
zR65kZbl}{VAkvC}%LcP7in_(84eE_gB3)mY6YB@UDkK>#BIs+R<Krn$!1<af;Wm+|
z>;G|6cEA>!bsZfrvVx&xI1Dkii1q!H%Wip~Rz>}$8spxa3<s)w|5wz{qQC)aPV~_4
zIX8KjR@A9ewq2X=OC|#bSuYZ{x=$a@s1DS=T4zTJG^hwaFOa&LwzE+v?`ZC?5-@p4
z^wG%)ZTwYwbnQNG<upwcqsH9%U?sWAAd#cU@J^~1SSLe(O*f#$p;qWAbZ+GezJvek
z5k#z?cAn-y)dRZ)46-BoIxcWK_6w1$lOr$6Ir!h+b!#}wmo&;7fT0s!{Iy|~jSk|1
z2y*<fbIvE9nRTuY2h;MV+m}*o;xp*~9a3*uKS~I|k;i9Za$uf`v(##Gq{M3vB71e1
z#K262f2wd^C6mUFv+ZFhcA}8FfG|N2V%JmJPM85l+&vz%p!Qbx*<vM`I1B>{g<=SW
zpj9uxY6l899@f@@A6YP?#X6Up!u>H@oi=U9d1!BVRsXXo59ssM6d(4ya+c)+sTEi(
zF}B!AjvyP+sdDWoHUZTJHA*jFtU!Nc;N7}(;c8&}TfkYX^cdEU!7Z+NUcVw#+lF}U
z^;h!l(NK+|X4^UBSb@>to{=)o!ZQME)}ls}C?zgw=}xqYf|!qU6^Nf5{EftGFKPF6
zA3q>Y#W-=BL1r~7`6C^|Wq`$ls9nA#t-Um;$MD70Xfqy#$V?N-QI884g1Vh8w+Q<&
zE<rL{7mgAY+ZIO|jDqFn%}9Ij^=0;j#yKCYu0vcuAex*&4p7;sFggnTw37V)h=}i}
zeY<9`W_Ql{o(8{c$40AfYm0bATyoh$V+7;nLnwPeew#r$zyXkU{cp;ylhr0Wz`;Bu
z&pQr5hQs0Au;numF60wOji0h;QFwE{u1vf3A`8rUGw=YqdT^LDx45$6`~T45t)M+Z
zuxo+QHz{ZEVVY<!Nd_pUwvwID6&ud499k~qi-m;r#PD2%=DI3Lq7gA^P?K>ub>1_3
ztqV(i%rG<`JbJY2b1B(+7O{OnfU80<v>6x~WnCOsll{2-g*04YEsd{abaM=L>fl)i
z;Lgnv*!GrwpR`)AGKA)dw6yHShE53>64TX#lcm=cO?I$R54N+>lN-Yc_hA>PRgfRr
zMrU-Dyz`5l^iz1Qplo7X5~qzchg(5VqB7?eS3HmUX)9nwfhC~2$4Gy35h|wY83Q3_
zd0HdbPGxw00GddR345bq_E)g;egvWg`%Z!xNTz*Ep&*+G7!b4=-GNaJ6qo@~`t-v*
zo?pM=2*qFpXQJBry@3eFFvpSHH5<VG<!2FDeEBgaV%&Qpf6$XtsR(CSrcf1{^%Q&x
zV<n9hw(#SWmWLwO4Nhjn1Ncp>8-Z%?R?ZV}hlM5KH{^m7tY^kP^cmHmQV~QPFd%~!
zB(`o+7Kgjt1T5o0d1WxB6JU~C!gwk8ZmNa?29mY8j6XePns#k!m?;>G>&zEO?e%7k
zb49hEDZL5laMdX;rmAGS&15Zrm2MW0?+gsn)UJ+emy5;-`=bw`Q7T#c7Gt-!!c7i%
zMcN9`0QeimrzVRbYtkbK?8ZUun~q7`!e-o5x*p|A`<X|k5WMSD_|^MA*86L|XH!4l
zg)QK8YTo6u$l{xe=0A7xpE39H@LW*P(Epi;sEl7oL}Xam0A??n7y*=Z7y&!Ziajq~
zeSS8FhMIISB;l1(CoN5(GbFpdNZ#8*&`%l4u!%3G<_fCNr6f#kWgVP*c!Hv2Z4v6I
zTLB{@6^C3Q4E>J?QPUK`dFX10O77=d!MuVU2b~vdIeeZYl$aSbhG2|AtQG)gTaDn$
z*ncvrb#p8inT>lw^EhO+@AgrB(9qul)0NjQ;T$gns_}Ejgo56uG59X;29lAg;TJE)
zY~5`js@CjLKiTOyYwDVuZv&%)czu%zPIOr+0%l+@*)I#c5MSjz469$sL|IFcWc)b+
zG^oWfNd&Wl-!1DDK@ma2-~!6uq2*HqkJ1ItKnciPeyhW5L4!kD@n?@Dw;Nk3<V29c
zCJ9HE#G6YVmr(rU^juiBIY0KsD5qo4a~Y4i-b|rYfrk4@G^0kM)X00J8H?;3h>3K;
zRcO2Cy}g`cmGToZC85)#QwkB8ap1zQ1s%Bjslyr;crI^8W>VaNWg)nl0nI^1;knYw
zcxWS59}XY>p(Vp)TV}Zx@T%1uB^wZp6L1^)oIh<;XxlsdKL^V=(;g-7+JnA7F^Yol
z-(ngAZWmTTm?F}JQFk*BiSTCl@Kf5QLiGkB3cxYX=nlE`SV5A1$LMUNEtWWuEr1Bl
z?}q$}jEqz39)wTk0nK8v+o~|5rj9r(&G;U)lTOCw4TaLVx`E?wH)a2sk8emX@P?DM
zdy<^Qz8=Ij#`h??paq{$4JWS+cX$6aRf9?xy-~k82<e!Sk|gPyz82d9S=9i-+4aKI
zv1Q}Muv%<A_5ode%66~++QA@#!LKPNK^i^};LlGf9pKx+bGE@!9hucOs}zpcybjIv
zRw>}se1~SME*!8n1z#&~)GddG->>HXKADOqWy~)KO`SSZ0ABrO#LUmODM;QYBh}7D
z4-Jr7cp$14@a%3KF>>SpXV4LwNqg4wVbzmfJ-nzOt7BMzD5LQ-Q3Ot$CiV%G%V0E(
z%Z7$B91snK5a2zUH71cgm#jI6fw{T4`OmL|35Ga${eDGB{%(wy7~<ct*S-++gEK&8
zdk?(vN;IAvYgL}S$z$NPhxsO_!iUr*S(tWpIp7Yz4*uVvc4I*rhKYkY0<Gj*l$7JN
zLL&g~pkJo08m4~V<4!}Y;3<4j+rByuBK5#ti4{&DYPA}@Q4r+n%CpeD)mcECJu%#-
zhw!=jX!}p>V3ZjgG8kAA7+(u3dgn_%#@k)gs@_^)7h5wUAqDgBz!d1>{q}&M7$Kkw
zV#Aoz8Ej0^uR(2#9~C5Poyhz{w0P<~WPVb9In)JJQ4?4Z<YO(7j07jqc`Q3^ly6ZV
z4BbeUDL&LRVNj{8sJwl>ziTDbDF=um2trQ3u>-6QTuDzg{^ikeH0k<wVFo?y6{S3&
z_~gaXwd=Q%U?XsA9D#ZUFbhLIYX730L^cG^{zqUUR2-ob*DQ`EmjHQ`-K>JHa&M~U
z$Ipci%>(`9Dpq}={Lp@I?kD#MazM-IywxSZs^~L0Hm-*U1}a&x@>_;N$dv^WsXZ9#
zTQPM5>lu(iJ`j?b2%@uXK*Pm{1D_nRa7*OHGKB6rfWupnoDO(ZpV&~X%?|d#D!Ldo
zc$E?qPm0a~n2v|=(u44V93-ehCW}+d%EqJufB=VS)kCxqP}c#T_>P`8j8y^x%Shbz
z=~m{R5ae|_)*Agf0VR*TGAv8c%AcnJ`(VQ%6?hz7b9}0{q5s`>f%LCoX=(QWvgH(=
zqq#Hfk=2v)q(Lm7lv|TVNm+We{YsMjZ5%Kv2$u+k$p9E{ZoUC0Z2nMiY$hZ8U)U-s
znF}Wnuos~)7XQ690=txbBL<h5ICkkm?)cT0E{hbrw&06jLLz4s)N5`2j;z(a*z_KT
zRGhWoz-Ue&b4U5|nmg7T3r$cFaBn~c_<9_6CvJ^;Nb6i^E}XJ_`6oE6KuVAr2+<!X
zfwhH=0h`7Mr?RO0DH5z}6+e^RUyV6iYDnFkN|;ANR^#eg+>8i06^Bk1vkZuim@sj#
z&k^S}v5{EwXp}=lynu>zHw1c?ipnU<yIvBYGJN=|+6o$1*Q-Ux0{4PD!-k7IkQqD_
z74R)F8+{ZcZO<p)hrn?YTK?1=C#O7qshkcB*UF$}N*cp_-VKrjgAr47>?w*@jjC|m
znm9NI5#ay|f6%w!)3fv&-!jHYm;$hmhV!J+Y*;0TlztnJ8Q-cNz`Li-nbQ-bR)*vz
zX-}Vqgn;c188vZ9*TcP0I1)WtW8b5Qo&evXN?bh%j%Kh*VK?M?7vSPnc@?9=S#<8r
z31C)E+eZ&@dW9CYZJ6igw`Jv+w?DR#jE(|jy@lxi_@rGT=EN`z@YdPyp5EF`>B3D}
zuc3G=YDJbQVyG*$_ogn8M!{3<M^X3eLnkJP-o;lz6;u^1H?T`RJlM#AUHk~b#_wT*
zR2|W{2!IMiZ^Y%(V0Md-fkxR%2}^AV?==g;DH-Uk9N5)^fD^L*@wnFl-b>r``a>s^
zPm1f!H+9^GLASbP?aL4|Py=d}A-<Tkon3b%ne&-#SwQKadp)(x{)K{xn~Z|-XP`EF
z5wN-*sQp?hdO=zvYZ)5AO3R-jM{z_Q*~H6-IW4NW175+7<)ZZLX|5nU00Z)EO9>xx
zU3qBWWoKQnKQ=>a6kX2P=ASEYBvf#-N<o9#5Vk9AREXj6syx)m=eQ`sZJ`Y}@!iH+
zBYXG)qN7k#r5a`0Pi-YDvXNA>fpPYtJ!qDN2EZ<C`4=XPc3N8$_D87spPq9$fn<CS
zY<L3{MZu-E(MUU~zPQktgr0;vYQtO2EkJ^uhgy9K>ApdPbP!r77=0^AT8kB9siYBC
z$7FHvx8<;7YdSIEYP}!nzBVpERfMwc@tNJBof99nw=KIMu4X{o?~po@&xk}jOjM*k
zdXR$)JfJ`r?@aX_=ES6!0*{>Dk%b690zQy1D>!uV%goKq(R(}(G50$yaQg|xnbsB`
zX)f8dVTZ>19ME!lfPX(Vh4tMVkOsvu*v(@AY=s=cVa}LtP<LR>&5T&ThBc5t$OR!T
znUxj@tmbpbX`YgzWc05TDg`(9)Oisup~H?R@VovUhO20f(kE!qq<ms6K50Ft78Rof
z)4&t;n%fvQuclOP3DvH+y)Fm$mQ@oc6oOWlFIQDx?5H(%l$Kvz*;D+i;VaZo^N{gy
zwnU$SXL?HHhJ;{Y$_!2xqhZl8>*^}r`zx67vOTeN{l9b?l+4S&>`?!H4+*;QjB2|B
z4cl>3W`9>|-ig&cr*|*@>R~8K%52hvS<{NmX1^09uhfSvoane3vfDxh(!WE`%*B`$
z{XJ&JG{!})uG?`m4pbk6s@IAP^x*XkwQ^})NterLUF#mR&(Zl_vfdm!yFHI{Y9$+!
zar;8(eN71+Gs2)M3Mx>PwBRU0ueyDmjEm1QGSUEi2>1e7FXq&%(su9ffMymPE;3yJ
zLH2jHgBt`edlEZbXTyIb)*<lKNe-m@k}wev9pqbn#2qldo=V|k80KJ5h{q#jbaX|O
zRdz1O=eq$v?Ng&@20UQ-(-GR}suP20P@`+)>0Y=$hQk|<lAqqQqa^<T+3pBc&nFLo
zKrKuJshNdB>*BM<%L4X%eO!V}myGiGxzM#^|Jlh-cfZnIh;1+t76k*nhn`%99G}v<
z?vOr9tVKD9gm_0}dL}EffbMUlmOM=Vh2#c~Ne=QUc7h^{qfswO(rCPeJM`%Xyh88*
z{HEGM0nhOpp&)Syd?tHqdr7nltCEO(`n}?`Ie_hbu^GMa6luWcH50Y{>bz@45K3gb
zF3UAq*Gjl-LU$1ioLv=Bpj_+}FAqeTcm;}F@x|1UYbp@FnAZYm%E5LJY2u2z{0rjj
z?S|Sf1Vy_kb;eyU)#c`LjvEUvr61FYZkTwzC8$nZA5{fiqpp&qr9`0+vN-Ra2Vw{b
z?cDT!M9ms6xPo9mk`TelGB5%c6t#%sHzhZv=0QD8*R3Jhn_|4zBOFk!J*yR4+1uPd
zp129MM0&P|uoDt>a+)0(5OdNKxWeq;s{1UVRQ@J?Xfu~ZT2dX>@WIcyqW2%<h*CLG
zS+UxTi95JXW0~(@?<Uuz7vqISyT8R3CiQs&N<QF|zWqv)AA#YwC;*_$bFex2Bko=A
z-}22w86~W2&E&~je}`Bn${%Fq&dk_r=<*{_n_Wup{Hg<LG<j~SQTpux7q@ukM@TWe
zyzh3nQ6!YSQ-gKY%Z*eeNt2o90SH>S+XwD47nvZFnL2GmS&i4`%@8rYw`Lte2Fo@w
z&E<?q2S;81J){<+rMjRNwdE!s&!62do8sa0sO`QLjO)M*>{+9NGYx3@45XghmT!h!
ziMt0!{(yWws&LN~5EQD*j2d=iNzErr*w_+LAK)qV0WcYW^3cFn=<{^KNy2(2Mg{35
zeXppbA_h5AjgfMSKy(V{c|?y&>s%IV*47P5KcbO$``Gsrjnn{HP@YskO&~}gX@Q!g
zQJ0ZMe2KfweIUJsf=x_SSBckCB<$)4_yeK=aT~+l9MR_n>uu2sP`ZF3X{HIBBX24$
z-QEdlZA4T9Q6C0G?_b_^g6M#&E|M7bY~qxKc3kx7G`k{EUS<1~SByY$(K6HW=6C?}
z@$rBl;V?|{5lU{oR&sT)0d=mS>9G5zBpa$SnpNvNt<zyl)qoM+JqRB&nmL2NS`AIy
zUA3{%TejwbrrA&$do~e<5}a4>^afbuJ8Vsbw%{GTYfu(IooyS;WZ~oApFC=(qTU%T
zm@U0H>q(Dnj1{$pA;_5+@h~8o%s|-r6Z$gG1@hiOAm)9Ce%{pEs7k1E*Jsc5L@fon
zEQxM@wa<}FBwnj;Ckb8(J?maAffYI|OW~CZALD?1S<e~D<GUgq(;f0$%xY&ly27hc
zFhN&EuF6gMVje)H325aN_62Fxw@vn?a5q3ia+XB;+qikXTFPUEzOVsUt^K0H2cd$K
z!(~=9NHQx<bzxi7sq>VKjyULLi6<)`1JFrmn>7AXj5(ll!BJfqiyd<t5&=6xYEe_{
z0ruGWo04H7b?5;j+t{(j1PZ+ZKY~8W)t%8o>3L@-I4roetmH9D_iUIJYU{95*j!)n
z9d$OeG6Px22B>ia+a|ZWgJ=!Z<7R9sM%waJQS2N!S+zD4Rd{#AD0(V7>(~2_9S(Z`
z8p!7!Ms>Ug5TyAGQeZb5gOs}Gv7bn`DID={jIgjR;xuSxVlS-RNhNG*Qw@Bj)KOC%
za8me9$$6ZDP9WJjxj$PHL0z|8bW{ZJZ(1ry3LU{Ta6NzseD)&BKX4ir&bTxx_}+Q1
zR)^ecJ%ZgvW=-J$wi>M*6B@!6B&Vd{-pYze|G^tvLNaoNa+m=Aw=)r;)(f(nEx1h|
zq7kp#ZUQr$#UUJPAR^Z97-FJC*qrh-ZmC2}XUZ)wYgpU3rKn(~qBB`ZLs>K~Ra3-^
zxdOTi^wEAgIvk<+B1nj_RY`Obp23`5NF7t|$oie&7_@YYq#2ik(P(hNDY4@O1!;FG
zSs<2Ov>Q->D@xyXOC+<0rCHBMo2dp>n~wyUj)Ao}%gl0F=9Os%NSW?LKy_&p1aQ8H
z8I*sfKGx@$q0}0pA|Yt6LL;br>-W$N@C9Z!@D=rWZiB=l*OyhrNzE0!Iai0Wv>@`B
zS($w*Z{7mD)nbGWpQC~&Pk#1-aHNp`7hC-WsNYRxo2BeA#l+uTA(TA>gruwT!USl-
zH}E<l4kp|U<=c3keXxK_)xho8kAHp`Y6LcswvLrFUO1y{+f@4w?zr))o6)juPt!=S
zX(OH?efLe8D>lQvp^-Btg}n~`JzWKfayvxKShxnJH5er>lQHO~Z~}D)$qYtKG$k`c
zTAJ=qJs3AjyMUqU^@Ayk5I+}*42M!HD?aUj0OOoPU2b=pm3UCJ@<7K5#D=vRB$1Sk
zL*}dru%eOiho<Sn=gv*z;~zlj4}4>^jPGr>)MEQ0$-w}H!af`>fm8OqfhW#=f_&Tt
zHKnsX;rb6j5V$W4=GAm9H`MdbVKx{A1M;Y=11ln$$zTk=0!Bf}Qh|R^gap|(+n<%K
z;0;jY`@QMSyaI4!VCmo+=DncKI#sU>#864VcTxJXtc5t3_Yg_4-Gwa~>p&#WKIyjH
z7tclI5yDc5I0zrPDf)z3LC8(6`Z)|hl$n=P|2Bu`^)Fyj+i<u7_qHm6mUjhw(MW|l
zyN4WhQDwyl-wArVZ8YLshouEaAR@}8Vh0Dlpj%Fz!czy3lh5|?@)(I=bS%E#e;JM~
z?Idfu|8K)FPO!0x@2#KRv<xfuBRnYVLE(O(<m!o<SUk?ql8p+*XuoeOcgwjUZUEA0
zFV>*QBH>8@(A{9awhs7nJiyV;?X=e|U-x=?*cKS|{=Rxz5~<q-gySN`ITIZ{`ZKX8
zUpQ*f?9OM9WwUs>G8Rx3(#@%Y;$Y~q{idWcc)fp{>B#Q!VVps7Q$*)W0-4L}UKxf3
z*y>%h(`rCYwg~EtZG~Vz%7p5OLQg$Vkn}kWNQILjU3OCvj_Sn(Io7sFGfDhTB6bEt
zQ*fu_vR6?!1{s#B2iZ-Uc$>3$f)VN7kY8-%0`O@r;pfG~2~Gq+q}%%j&T0bK>D_PN
z_V#@a`slZ`OH<LaNY|0Tpa*MaJ4TXrG8DpO?%<6q+)b@pC&)j+5$Swx%X{J04F%Ag
z8$O0&8bbC5R^mi2A61^v5Q`78t-wUDuopo`woy{SpTf$wIZHe*F(_wn&?okfqSD|<
z?HtZ37EKB}tRi-Il$2Xin}l+~279a`7`pMq;bfrj*PO`ILwaL&0?VxPOC!Pl5i|+>
z-eLI41rl#KLi@!?C8+U331{vY21@M7s;YK>*tj3dpfoe&$3XQipyLHW@o2_Ba9B2t
zlNEMYhIhR{b-*T7y8Q1y8?z%I7Y8PSw-=W!>J|+(4h@41M0f^X!6|)NOuW>5s1Y#k
zE3ft~rGQo$UUxi#e#Xq+bypa>(F{@{CO!hToXwXUOsCvV;<b^P19T^TpHaz5NK<%{
z-@n!UVGvL=ghBA0kCrnw=$|?dXL|p&xvoOe_Vr5j&!Ly3N%u-oDD=W?<e(FDv;7jJ
zUB#`q?&V;IfPZdnN@Q~YmHu~kkf>cp4nGRro2nj+WuUXfRv5$zNyh)e{ku=n;3StI
zD3pRKUl~U-HvoTaRgZxOmj9rMN-&J>R+ZuJ&XC1*g{%$?S7O<00%9x`ZPRr@S{`^?
zc1y$qmA)f*ZxCUk7L#62iSEQE(}<8CI+&R33t~I<1A;vpseY|p{qLM<gTYN`u!(bh
zhN(iC;EV@EBc1}%V51&q?<`+G9-A(Ry@%wJvpYH$M2GFb4@|sL*!E0uRTM`3)&Y?W
zve5P!j+NMj71>3y4s!Gh$)?C0%TVZ}U}OI7FAG1}{ZXV7O(1cuS5d)mg3%9%M((3S
z<z(yzz%LmM&Vz<*CrZG@;`irmFrcm&Rfp$q6Q@KxWX)O7hmb#%op$pXJaXAn`7*R`
zQn-rJoC|k<;@C9&Iy1JleZOR?l=t~;q<ccQt-h}i;G2t-QxG}i->5I1e;F|}(H@TA
z+fJXuejh4DfwTLz(&+}6Ee60@N4=0KI)WKkhT6@Ke?tR<I16bzqSFIR@xPxuImQF|
z12E}?Jg~3XzTOP7*;9gGWhnZ8u}P+;%mXq>p&WhLkoofUCNq)L5W51zFc$Eo+3kj^
zh6|ome^2{%^|R_&^B*U+**B?!&BbnKpLL%z_{`!r&n{26G(YR-r9MYoAA2pE*CD>U
zy_#zKfX!cRRN8v!+ww<Qu5Z26-emlcK5y$=y^_PLmOh_nd%Sf1IPZM{)4J;48S7W)
zJ2mX0Wl7KW!FSe~He{zqkM_8y={*0{jgtP8In~=cy}%nK3GnF3wDYK%8lXBb0hoYc
ze?`fJUumD98*?n#6p?RPXz4l&h6WqgndTIf0kl;g7Vl@?L`vF4wWzb+RNCTTqxekL
zGogxxe0jBMY+$%UujO$H(||SyAw^vWy~wRla&X&ozl!=(sEeudL(p0%D@nFIp}($v
zZ5O+*vE3I?A5EarpQz7+TT^K0w7yB@M_$g<qMix~9|}^ZiWpRA_Y@CM(kL{-=&?ri
zsT<X~)Mv`&v^W;pKCLKW{wN|zVFm!Yanpxmck@81dSqo~31&IKI+y|Vvj8EFT)uoc
z^|MQS0J`;FK>#v_=lW=c;MTyvh!|<Hi`hwpCSYt`J8G>Pzi-6S!ysA;w#doYnv>ON
z-R?<+2iK1uh?20qHSgGdR~u7_^-2_WosiB){-H}!y;oLO&;tC|nSvw*08L!hUZm}@
z>Zub$bqWZc{T}ihP!Zx#`vsC~U`uE3rUECWQ4wNJ6dqqho`};;-n+!YwYg((jgZ?#
z00`t#ew@XZM!5_!TbeYLlqi~mw4jB#VF2D*qXuR+;-IF(jvNBDCu_JqXC;k^Z^3;D
zL@5#M7t_u#a&t*Jqa^Cxg42<a0lm#v*N3k`z3%ov)3N}zUYm@TNT)@!Zy9xPtCVia
z8%YS824N=LK5V*2^nvCG)RK!|XeN5w(Rv;bjeMFdx!sh`wo_j$pNe;3fi1DREH7_!
z-|d^hT2BO$aqZ2}>%GlIgvPjtNkk&(yxoCxFf%B5v=-K;GtD<qwS(VD<ck+x$hZQ2
zg%jLpKn9hBVF;lg5+~c4TA=y`6Tkp+8V@qtYDz5U9kSkx?W3!v9FRdS0njC+-6o)b
zyFGmVyeFsEF(~9M6TrC3D@)Wa-Q%GR4wpNnU-(2PXP({yVZ0IO`lq*KErKfq2U)Ti
zaaFi!0q_XTfuSpHz&*abvz<h38-;|BknGf}1!ZM;$&Ff9_zHbFgffy|uK4`j@RYDg
z@S5$p<3!Q4gi9p2j1q59>FV==#0Mb2Bm0gs2I`B#>PaKJbhqGpRENp=BQ!~T8pAk7
zcO0D`?3vQB*O=j)e^yQSnzMyCl_;`4kaD^r0Db$mVtgCPguJB=wLi0hH=~l78vfWG
zwnRl8{Oy3&;r6Z&>^R8$QdAMTzwF96QG8&u$Kg|_rYzxe`gg5@vyAWMnb{?*HnSiX
zjw6U91?;e=LkmpvgUuaFGNa4bvC9Z=)XQDdZDQ}Z4BIO-7(sTh{PfQYHW9Le^Pnwu
zKKCak#qE9V@tgvVm`jS7Z@v_xkgVs;YZ3qNN`}MgEYuG<HNBAn$${{LRpxg>x_ej)
zS=1GD(dVIMFc5TYGxqY1KV<#%<%d5Y1&=y-S21Q8=^54H*5m-&K9g#MRv^q;+rsHz
zK`f3N<NzgMJYn*eMEdtrzlKlKqkb0VH>=baskKDqYwYrXy>o}O2>Ula**)^wuE2+>
zicwc72TkT677W39c~5r*cZ0MKI#S_-5#zuDV(VCdS#=#VfZTZ?Nm?`aFVDIAhCL{G
zV9o}BZgn`7%_1NLd!1L19VUquX5I2cF$!jsP9Z?-J<k-}*cQNB*97z{VI|J8fov<v
zYXe?U;}qcXDFp2LiXAp&vJ3@b&dsqO_s0fM)PzCTZ4P!ET5#X_0xJ|GKC&fZ8`W((
zjwzI9RnDRj70Eb;7QHd_gnhvpzsrZW3g@0g2|RwTr>79`rg|8>Mx|a$@OXVEatXm&
zS>wXwfZi$~nsE*puu5mUt1nhR>RbCfCjqrbEebk{SP><MAE!>88hgMSdz%5LAU9$1
z*tVS@TKcPG7!D1ux?W&%4oxSf>1qz3#S@G!ngD0H8~DJit1Fklpu`uWctq(JZm%dy
zjl~elWS2~H=pephb|^+RBL$eWO7RV5l<J|4&>>u*xUG_g(hk(6*hu4u4!ttcNkA;g
zBgY627UokKK*1>x*X#Rp;0=p{$Rd_UjlB;t;IK3BHl~J<VL0CXD>xZ;Dm?n;5AZv^
zo3ze~M}TZk>GcuPrZgk&1$9ueLrEK=a&LSl8~j=C$4c!{U#z==-c?ksP-~!fC8v>9
zJ741lF}=ym&-Xg4v*^!2eFzdy#TAJ@BwGCR)4h&C4Ujym$dyR8Jm6|0BwXou2Hq0*
zJmEWR_2O=LCU_sqHjR!?cO0x3)ys4x)ed8fjh#R>jy#X3+F;@4yBicLzx7k>9?6mk
ziy*9)fQopVb(=y&yL&u3(NuwXBBq!<J-fghXT%s>>=AOU1|oW3T4nuHhE(YB)~_VG
zmXPmf=|>%<6UWuSN>S{b0@Hdn5slj8vBko^ID@<HIYVo?a&J*_F_5^rp#(aqLlY29
zf@B$rw>CJAJJ$R6S&8&Ao;s31a)0Ied|I8&tcQ3m3S92cUcU{kfxbKMs65qTovoJ^
z^fc42TT5hzM}X~!qRhnIFg$)5*u9q#n>;=8{^Pn<uvyUw6bC*~D7V_r9<^v9Nsg{~
zVGvG$m>?oI?^46LkQ;wE<ZyokEYPSor!M3Ye`f0qV_*_Srj;*8cI*WVZNn4tuu&y)
zJi!`sY)4NJv4lFP_%7vWqLUm&8}xzKF=dCh4%x!fj3nr;%J5ysIfKcak}Yg+vM{Te
zM)nBA*QS~HqXFXWJ3Q8{v3C#v+{NS{j8?GCn+@YcsoKJ#A}WM<e=gwFa6>-$$T|;A
zX;%-zID+jUQIx^Z4MVa|+cdWN#fCC`&D-2c@W_j`r#f`knK@&|+MBRomjMV_2)9Bi
z8X;UdcC7{YZE?wNpmx5})6)V&*$+!}mtdDup?u>Hy#<{b%!H^15+(GI^?%V<`{P8a
znUV3&ihQ02T#|gvT%XhZw`+_@R_!EFQHF~B7L-_zCJ#Zy80V-x_t0~;jEYM>6nngj
zk3;u1lJFPE(k)kYh=bwn5fQq2-%W`UhKy{m#TYddFh;|l)b#=S(Ve57ienq<H@r4*
z1yyNNA}Td`Fg7Y00E!!F2w$4O)hFRpTN7}nH5=$%=h-@qt=H89f_Mn09FsclQB+h>
zo83)ih=@yj=2ZX4By8Hec_zoTm%6#R5$G;mQ;hzYDo4a<+nOvK$?024N=i&GFvCNo
z(-c34&V@KvD7POBxzj2dnbsUd?mExJ)E<q9@tLH>qAA++kiJYIUV2OCqs1GHlIoIE
z_sWQZs=9j<KkZFjH-MCjp`bzb<9vc+R-qacLeHQ=H?^b_Q0)=-+`%}l1t7JgHpJ6&
zq)>s%`A5iLv_(MT9===F<M4lSR{<JKAU=a8vK7$5@6~_AeF!dXVGWe%rK$(MKBdXv
zdTG2bw=k!q4k&fRZATbBl!W_`YoK&eXgdtw4s_KCh3xcU71{U0{v2XHb!j~$^6~$-
z@Y0pza<5(mxvlZ)uXj&+3!l7OTj`y(V7HPj8W3FtHjz8{wRn69LXY3jki~8#yRRhD
z_8JJ@K>j_N%=B^_Y&WF<%PIU6AGn|iPLw^^joVzfFnyY2*Dj<~A%-qi8228(40R?b
z$Le+9&*A|lylM&BE}@?b?LSn(ON)ZJ2}=ac(h-NZCBmGB#eVU&1pEnV((V+K3G*;N
z0M#NTGNL5bm<lz8r-iXr$Xm2XARuiz<evsCk@wznw}t=y8|vd?h?BJ?b*QMxtjY~a
z!@hS4Smlb;ub1fl0N%~t&>_~VC55)B##nP3pMtWGt#h@RO}UzM&=P>W#61i=S<1o3
zj=+W^NkpO+NrwSKHbyS~{4)x~BDlGBiZJ(Qn)ja=DkCUZ)_Me&8mEf_$7TUc4;zX^
z2g!s}h(?6S&6|>v?Hr*}W|fErjDBIM1Dyd4ws3cEa{n|uM9>E!$kdddKm|dFGk>Fk
z%5v^r`DVzeB!Fp#`1d=5Ybb0JQKG>bvSjJf$tn3OMo(z6Tf5HUkh_j)D;-3nI~Qd6
zsVS6Sc1L&XAQk!7qWsrKAjgPyrY~|s16`BE;F@%LkYlY%ZT-IkyU@iY;;=;hk(Za3
zBf8%s&r|JG?2o1&8}~v?<3)YPz)n}q5mqR&#0xB#(xcFbYQ18RZ$<m-N&XlKzGQjs
z4Ke1R7W5QOQ<v@#=V^l!ldixVOF)rA^#V)b$fBSjCiCpiRa?<251a)~;{E$p6BP6p
zQ#J+qxsCQ~|9-el-461J;$_EC$Qn*BOiu`V!*X-stYE{YreJ(q3yt<SSY!F4?~w+u
z*(E*`sy|xr#M9NKT}0BoA&Dh?h>}g&O^P}OVNPOnvWtgm%8)%Rqte>r(@O_=BR|Tm
z&1vv;bVc@vcBW+;71TsKy8dPy0isEPnxlsqJpcReE~IKdd`M|So4W-cS`}c!Ip8{c
z8J3-6+2b?u#aLQbR@5EFZ81MM5~(zr#B4JNr8O1mg7@)Ii0gNJrvZeh_0>%P`<5Xo
z>Gz5uG{~wqU}5_RIYXsG-0+C7L-n3Nc+iz|qOn3!y%xJUkuH_PstmFJWq%!uh!`j|
zTzg_P51Iy+NG6D`G!+}hGL1sGSy`e16}cNf-FSF61XBK;v|A9euBhy17f`rFWn{y<
zP7Pu}k7M(g-N7pIiFA&#aVH==iU?P=`|-Ygz~|M5!U%JX0JLJ_0P#CFH<iWr&+3+m
z_1Z8Ayj)rs1EYmW!RQ;S>^WH0M`$4`on!Bc*!ScEq7#5UWyX*2v{AA3Uaey}_VHWW
zP!F*Y)@5O%k(sLN*tquB(4<N2k!GyoM~JwIeYz{P;E1(u`Q8?&Mmqkz^cJmAM1;jO
ziCPS!4Kbz3zz9rqDr|I9vd+g1A@vV+UIIU$s=SLYhL<AQ63*sDua_S_erDQB`uyHb
zLwVy5Twi6`;dB@dAXNd;u%!ep)tY`z+8&cVr=YtcBAJ!krdG>$BN0HiVp~dcDYPq&
zx$ZBVeu89RjDaVCd6a36>&wOo?de^cflU}$fk<PU0bqv$CkQ*&h`K{-0Uey;IC(mj
z6u?TOwV;mC63GUkDy`cE2As=^sBKCLrX<lFa$(%Tz?lTr*B1r|2Y7zkH2@#EIE_{*
zy^vP0eSO>l$p#S+>h}I$roUd;*H)H7Jkk?1MX+Ak)q_p3qOd!^v-n2+ox!kW+gOAM
z&*}^O>Ird<k8h|7cl;dTj_1=XpM#zzeChtLL{lrYTHRr#`v@!j)jWjgBJvV>EqiSM
z_`z9r{<zNcvn7-voRg^7Gd^N#Q$RGU$$|!^pCl|+S6e7JAfORS%k!XpfsyZ0JkU|h
zf`e|mTGQ#qJ=On2banXJHDd!$RKF9Ue~D`NI8i&_bsT%$0-?%@Vlr|X2g~Skgu=fU
zrW|G=zOm&IVFOEa%8Kg0aXU)e?`P43BcN#7;ddj4Bq8#7hYI23&zZhXpRDHi<nKZ_
zx>v7W*;3+<%rgYZy@lVu2{bDc@-&S*8;u1VSd5*myI|6WdHf=%pV>7>JmdclUFRN`
z^ZowuPx|6J-<*=k`AnKoVsc6jIV1_?Fv;yUnK7p}=fiErl$^tDP7R^vY(pyD+meka
zr(r^?X-0FFGr#Bc{@m4_?bkp1K7PM^`W)Wx>wR6X>-BnFB)i6K%6{5uu^^FB%%)_D
zWe1rwEsZ+DV%2;z!3d@A9yFPhNGePmMxwjIKFy=3<@If|R@EA;)_U#?h+z&}L}#x}
zGGewtt~BkLF~VXJcBEEE5G--E>p$}L?b{=W{c?NmQs2T;e!SKucdb)Tg6d(uOP%|r
zT34(HhrwTF4E5#CRHRwx*Qk;VKn7K46xt4P8$D<}#+w|^B653LHA8X@_47v3(o%~=
z6bAczt(oV2CC4aVK?Mb^Vnu1<Ax4bdFBXb|o74XAey8Vw*GwfXz^3r$+AJ>35~Gc7
zV1|4Oe>I0%*JK!$shs5HFQO+xf$6Ku^M)MqXK6K5UHtw<ID}kHy))nw7ch4b%?Xy(
zAeTF<1R-wYkD9OA(OK{FF{{A$?wux6Y*-qz+LX`&WKvxvkNDui5!79~9w|7k4N0J?
zx~VpWnZ8U4E+!7^-+$kVZ{v7C$d{1(<)i-Rk7Dufb#oS#Q-X*r!7kUCU!33WRhhFn
z(Ahg&r4fmzfPbZN8Z;TBXsh|Q!=%D9?NZf36rt2ssoIbhtw;%`rHZ8siPD#{6SZq+
z*q0+vc8A9U$Uu4X`}~Jmns)4D;m2!Nzc58w_AV5C%GU6B36!FX<ruSGr!PqjYi{fF
zFFpbbvw{c?>LgKX;8InJq2<4P^jvn8q0Wbx6^qu=H;sA?s3e%fS^-1arEf&6M6Ta&
zq!50=&NvRO+d-~m+ThNq$1DAKlVlqXT`2}xcq`4alR>BMl4!-g8qytFw)Zf|uZjAC
zXV}9nYed)gtGcVaogigiqWU*P52iOSzOUi&as&L!_Vewc>d4mi-Xtl#6+SIU=>x}n
zLn>WG{{Q-=gbCxc5(>Qs0_;t$S16{HwRuC8J1pCQF&hPFzmbsHstYO#6+;h)>Yp1%
zkpbsVOT<{T%=|A%AuEvW{5q8r%`vP}Bfo{T6(P%YRRwjwwN9bK=CGfYF97minWPsh
zBCbd{HyqK13UqH#4R)n0^C~D^vid-6e%}5X7`;AxdB#<l<SD3~vDvy)-B1Ow;`KoB
zJsp%|N;AZ-E~QfnwA4>N1Jj$AoxBlc^@Oqv0Ym886@N_EGNz<Eh(^VgASDydO6}Uu
zGrr=Omc(DvVtKf3IP@s#=*7Y;$5Q7WAa{dn`=pzw|I9hf|2&qJ{N_b+j^G4oHzvg&
z&f9;IOJ%D|wHpa8h!|#_Yr@QD52hjmCYKBN<L%5oaXv98q7%K#NBQ}Uazhbw0I`u+
zPLv&D+8{$~#A2ZUi|aQugBsp);jJ1z>wU(*VXP6<SIcXq*cd|0B&#Re%o+bCe|+Uq
z`Yee*anconbX;@)518t}PQu>QyJ9bo!y~J*q2$NZaZT=&1q$>x>=74*VA$%qC+XM^
zFx*m)clWelW_VS^?w2kJ>tBLH5BjkUZA*ix|C6=<5p?)5f~|$eC&P+bl8!L1jJ=B8
znwF3yB!7_)=)DN(yK|}kN}lCs!%WpL6)4kCGAEsUUzy!g0xkYgdy1zMDO&=w`iuwz
zh|wzxR=-}*dM{^cGV6hUi~6w)k<A@_1o`V`xYh1cgo_16YB%5RUSsv%hxRK7^f=!v
zhMvP}js6)=SDRVJ_x_Z#r-B-sEr!<aYfj$ChJX^}tc_Rj)uoZrTo!(Q&_T>601nXb
znd+?&EX1aZm-W6}Cxen7o>3$KKuTWFz@ax%TWN`yc~f)_6DGc@UaufzX7oTey$c%A
zl?r1R?Hu)^-%iHAwS1@Im`!*{-oaET4Em@bdZq0imh6SAKC&27JzD6iiyE4iP0T0C
zTF^(!YXVJVBJv;xyKHFJdjI8}XSIS$RqZ%?CU?SwpkYVbV|>tw?8<3wvBs#n&R%9T
z`q=I5E^MN3K5e61vAYl!PUTXTb_&;c^sT6j4^PZ-wBkC;Zy7t%DA>I<eJnbu-jBDi
z6V#7yudnZ}XxFY?<&u_hhuHc8zdNe_3K)xvgQ=$Vdd!wX-PH4O%n3<v^u*mcORrQ`
z6~u%qY7H<yX||Z4xlf@^6!{^yRZc~(=T~Kg!-tREFG4U{mD2CAKToG*?bBmh<;$p_
zYZ#_#rM{5KcG7W32?hZV#+s#!eRLnSmab;g{aUIP_+;3R;zi8Ma`0A-BO7Yly$|^H
zOIN5SIQe|O3Fwg~NSv^=T+5og)ICjJWKCztHdDsZ2f}F2r%#`y!1krg^73s`qpVLk
z!4>FsX~SDh#?x(k4c8i|mzDQHLE?@&?l-XMwbT-Vm06!IaYz9;nAeQe5IZN`3obS*
zh{Ok1OMWZi0&~?whebI~W6K)ZwZfvyl0b@RN*YZuB}(*^OzCThTC8#@y|gu5WxVV(
zcL2SVOdO>pSD<{7W5-*a{c<N+SD7b((fSK@EKhmLXx-WpRE*=K)dCIv!{ORimGE8q
zo9b4^F%~B6A`IuOF{yM!E@s({2JGg;r)<bo|M5;1M~Ap4=0(kpxLi2(U%RX~`irSZ
z&wgyt5}*BaxO_EDKB@Y-B<&EQq;HAnVI834NYAceE@)a+>J4H*JHg@UPssd9#$Bv2
z-~Rh=5yiH=aLfc;5n^*8`R*(E?jphAKSs4o^zXhwKz8lY2$J@u%;LI#a`i&-BRY<o
z18->W8nrCsehdQ}yb9e*1#%=I0emu9Y4*U(dZeMrsl5muXqVwy9+(u*U+Z&B`h~_G
zmnT$rT4#v=EFlVnkZYNSF3$zamA0>8jWF-@Aiz>NwybWmeI*TlPQ&g;%RelQRs6D-
zxwgY5#+tU|ui%f^hPRsBqoL1W&TqQp4c(uhwRZQSVc_aunqqxN3_g1L>!G8SS&QM9
zIhrwzTeNXwtz&jtUP&FYfEowURsXpr6I$}W6h4`Pk15~=i8%z<6P!ldI}nzU%odJA
z;WnoKt5I;zU+rN@r00;uk=zxY1MQG7E1x~XK>l>h^p4LGl3yf<iPj!B?V9q+w}g0z
zjFvuIe}N&b>b6VsfZ?+7pT6s1o2yU8>$ILq{D<BLY;7@Mcko(sb-FjfG$G5_8|c%?
zdyA8_g^I~Bta#c|>jCwR514*20A#WxW(ocIKa1&$#Dx!FMk_jlHd%rX?aOMk#pnsE
zky4WpTY$oDCbD->1-RO#^6D}q4b1abvnD97!sDeEsmnR{(jMYBW?U7|2Fy)*>eZ^w
zxV&Ufc3}h2nQCI`Z;kBd1Q8Bwt)+d`g0bW@tPgK}*EICctAttS3IftozV%9boS)x~
z@DaDDq@8xD)mtrdZSoE=Bh;(H$-jv=xr{n+3>)_R>>8m(^>^09yPa7wb%5%j1^0gJ
zlU5YkFP%CYY2?cmc;%XE(sJveBE_gAy6NH!ml8}Rt9RiasngJ6nuv{5_qR8}xG21h
zfxfL^mO4@_cP7yBO9eP`43k%rh;&?TNtO3*(aNcTymzZId(y{qn64J3iU|UpsYS8j
zz7V@)uv<zdPBhmJb8|$&WEF56TmWRmkrrsdUjP{)X+C=N2+ue}YsFsM`Yh9W=y*$0
zu}7MYIt5OjF=Gq(cF;iz@CyMIv^*w8u**#z!un+Q_E)*QKIF+VZkePXHsxSmKmmHY
z>;z?RRgp|MhRwNOWllv=Y0CYndexzaWHZ7i51RbL7bO|x(w2bu=z!{F6rP6C6g0hl
zK#INnNJT@Vm4fedZL|_1@hi!gH8%q^zt4*Wd)_MqKGSUAI-4HR%|%@Ec`uT2T_-8+
zblu}>uVil=Qad4yik)U*vQA+RP;j$55rbC=pRJ=0u~UBbb}1|}pFe_dFRTl_?Sz_)
zPH>(ay_(7HZ<+o<3$sDcWIB0P->gR~A>RWHb^Tynfd{|Vj%A{^Y#ssqHbF(Q(TwH9
zhc%!NvCj{w)6ZclHljilYM3jZ2ngza=gyk80!kI%lgMF%89XQWnudOl%wM;36S0ym
z;WHf4m&a6qTL9QCvQ-iYAmVAqSv~oTs2BkqXW8c6H&^oZCX*WUQDYoKWUA`5DI5~&
z)kXXW<EB<&Y9oeBytqpxi=$%=#K@^`m*+>nltDP{9-!rR69sNcru*?4m2~mlrN33!
z@DZDf>C#IM%9pu=*wB(NXAM<%E+4>qj<730nn%f|xx1CD8}WXWxjJfIWEI`j3Q5=?
z%Lukti?y<Bx5!XBf#o6iSWkn*{}rB^H=ENmomGozx0ZH;#bh4u(eQW(D{|Owchm9(
z+@q0Q6_Z)<eUIHQnLIcJlE$o|g}H$D9Q+CN_WbKi<Tn+>J4&?*CPIK|;$Z5ac6q>O
zlD6WY(k(6zyMuyLpr_64u<GCuFS>Ru#D^+EXye#ZG?sI;+LmsM`#Ebf6jw3Y<^S6N
z8#0Qe;ojA_B8^VP6Z7ucRa6UaHx9IX)ywnl`~SrM4H(dDEXwTFB`a1`2YVe3Fw`aP
zYNpVI84lW`#_ENgX$!eD2TroT8h~MSjcQUH^*-Ap?0%l!2z~kL`kPro)C-RnOrPX#
zCof;dWAgIXwFwK=2iH^0(_M*e*784^0<mRLXs;R<U^c$u2Px1+L9bwmUn6wO1=nHP
zlFY<X@|2?NXIW<=*R5Nh6zp<wkk&>^$Kf?LKaFxvx4pg7^VW_nPNgg%svvWjE(51@
z7#;=pY<OL)hB}?Nj7=V{F4L1~H`k9rYg=xvj;+zUe7n5^RjyTPzF2O#GqNb~vM_=m
z8*C+dc>zT~o|ee*538sU?R4ay)ai+g3Iao8<-lOFFBia4P=%c#fK4$GTc&Oz_&&&<
z5b`%LX9Sy(S_)b(ys`%BRnd|nrg#sNq2sg((TC$nBNHRYqLRcmL38_9M#PLmtrmFs
z`!N?&0h>V_?v*3FzRSW?JtO^6rtjFad+QT3U{aV}#RQnGPC`FJCShuZfHs(hSZATr
z-pt{f0uop@sTQXSAab2LqC&{~S=c2qMZ8iNEb5sI#~3Ehr*^U-V=dJctdB<ShiIE2
zlIxw=4nK+5Uudqawf%6u0yDGGy#Ep_2TIhG%w=`ENSl*j>a8}bLt7EZ6;!WYFk`Ga
z&KpLBzwf@XsJi!dEht%HMhPWp_P3A34%Q3LA5ug+HOSn}rRI#hebTk3)iC5>Q(Y!+
zdNZsTb51?t%XpFNl%IWXUP>msBpD~hP08>bC>|e*`qnnWUzop8(_-qJ`V;j>p>yLh
zneO*GlW9g<inj=~Qu<LHy^2|o0=9|)DD}va1q(`XxC&5P#4J=R!15+vx&$IqNy)_R
zFXxN@HRFZb%U|HjDZB?pr~fKU!Di&p71?)IGp)ENVIbC~Wf2P@Rj@KA?{^9t1u`@S
zDgUZc7LzJEoF6q~4!juO#^o`)UU-(A626>4LyV1VQlvmtXR4s$VkaCnMEoA%@s<2N
z>w3L;mM-sHCSPi9H%aqVK63TOGtuTz4U+8c$>s)E6+8QTd}~dk4fp8776*h)c4{nJ
zfTP%Uw$7Z;MJ4&xA%wm~s+{L?0@kYs#Ifwwp`Jy30nL7eysGGr){C}CO|6v-RtA&-
zJGt0IbgAA4(ex2*V0b+I_~xK<v2I!^r6u(4`S#*o{OOyol4g`$a(nmP<8RI!+rPEz
zudczyg&zM|aH)a!%DJiKDs8(L_Eot{_Bo!zTSQfj4qtCJ|7@tzv)sXtn(VLnQDe`D
z4Jj42f7v7|x0HV)zqJRyOKDKf{<~m@gF`;7WB8wA#N>wEr|fxMe)g+Fuli=RIr#bI
zp-CZqlJ9oETRNfbZ(-L)ebjZdW?7FJX#ia-n}ODKR)H2!v9D|C;pw@<#pHO98r`{e
z?LKLrqQ=qz<bb{E91lBzM{lKsDR-9$GcynQG(9aSD_r*EjYb{D!*xrX$lFstujdet
zVNA9f9m@Z`uh#I^x5Hk2BH1uP3)~0p|4@E*z;Mke|M6n7X;XZz&=ZJE+fK}Wes0qe
zOtzTnYFY2@<yEUO=JYLB+7xD^$*R^qqKoW3nhy7(-Me=OUS<SeN~hYI`BRmvwTUI=
z6-H1>Fl{7yHr+|js%dVGsSwQkv@E2En9cz}7{(oeXqI#R&8zE`!h~y_v-Hi=r6jjM
zsN^=SC)vcl8Bgx0<QoN7D&+;?Gwq$x3u4u)kw>0zggU4}64bx!2P8ag#Xe7oJMrcK
ze%h?eAIt}uA#V`sZ~u|W%r`wIA}2O;!y+Bsxw`OcHl-iGd2`2R&9?{aFb4;R2Dx7M
z=9W*n9rC2@e7gWA{d@dIiCJ4LDprXSyGF;^!J{>Az)bxMXqL|;sHQoTE!(6_bXR2x
zDg4Ipc<Br89S%_LQcZtbyCAHH`z?rV_~qBHUkl55rWd}>(H|8-2yLmG6_+Mz2`}68
z^IL(&al(PbiJDAkvpPe~55<p0hLpt#YG3`ule^f*2khrB>Qh(Mt$7aR&*m;cp56lU
z7_)iC-(!YVmnJK3yst`*D^$EY{bE8k1#j<FSAjKaw3)>E_3nA%#v~i!#d<M23j68O
zr9p6g>jxc#$wIVQRdA7eo5+r`N$CC37B5*MLT}?ovJJd{`6cwgKQDuVHd~y>jN_yg
zHS^gdUi4?`MZet6g#<EzM^VpRh{c3OI0*rh3O~MEO+LOuxpJ;$q8G}9xqYM3QT8Hf
z8hb5qFw0yQ#l887CQ(IF&EBhCLHVkQ#o%`mfmcv3eDh7Gg1+kd)qo_BHzP{87$s!t
z-~Z7!<&Qxwa^mq+=6r7mei1njPB!n8M*bzS2!fF;r7e5*G<Pq)U!-W{Y<3Cxs(kGw
zBBpU)!-Pj<Vt9_Yd?<@Ro861M&AXP=?!wYeWEobg+lCNj{ilZ@g(3?nY{Gn3uoRs3
zup55nz>cSzEDj{jnvd$bS?*5Y_>`$9%vI$u6%qT+mM{VTm=xbLy$S<cZ`^qOJ^Um2
zvHTS)3sr8=Ukr-cGv^{aeo^zIp4aeU=ZM&x+w4Pv#C0XQCQILbVjLsSm(=daP?G2!
zJpIkPcQ+Le5#Q1>R#)_@U%$SL$qT#;wY?j1<+Ga0C;AgF#SexNIRn9m8-%`i@nQ?b
z+Jk3_966g?xw%!??!Vz{n?T_K-*5QTse}4zKB0(8l~&39$&x9F^{-+7FL$ZNm>3+<
zOUxF})*vwvUyP>$X{5`C6lm@%LlX&h>Sy*4^2%?%sq16sv>Y?!y7b8O1ta_v2h$1_
zE$aM`aP9H#!)T!TkHpl!R`(?}sFeev)OLbfSGZsphNQ)eB9U`<b6aLkki4wbo@sPH
zT;5TS+K=NrLi_{VTeodnO}yKeFR#d2@oBv7l%gK#H#MPA;VSOIcI>FMf9_9KtVwJD
zoqVa>m)hoa=CdK_o`e@11Ts*N<39pPr}g=^^(uV6x#8jAqQVpi`W3l%Eu9_IP*lnq
z7QX0vRp7sK<xq1b9*mazYDn%sYgW=`>(x9Ztot6HYhPNa!i?kS<lI+2ZCaEQ&#r0P
zzj-_2L-yibwd9Ee|F^z((>ZY9=#~!U%VU7nuHVQ*7#|SaeeI*Ss@&DH?7xS4r^Rf^
zf4uSEMOPqt{qXoHM;~sufACOJ@8Mch4gHS(6743%Zj6%TP0X^`$&;xJQ$5zX7{F|S
zexmojVtYiYHWAa*+c!UAZw<Vj_L9UjrHc}T|7<Jkiq9$v%17Xiv8Zosl{WHxmiedd
zo}Om&kOQH;6R~q2;H))N8Ih$bXZ#{`us*~U<H6+QjvV{-Hl@E;PILJf_~X{?+XXa7
zOVz^T_4ntHZ2o+DcFoZ!^9(4%)#zT(xzG;<&n~jisa>|8KplEu-<&tIdA^eGuQ}N;
zw8h%U!cSM9PZx16C8aCpU<&^+O1le^e^e>&UsNEQ@UC<qgq0ST2gIMljC0mfnE@ZH
z+wkZmT3m6gD1_WRkkPu4sWdQ^cmy7E?9kE2yHYYaZ=cKDErO$d{!Li@|HS@IJn6be
z?O~+7qpvNK$7-%fnvs)}L%^-`(@#H%0-!~Zlw35Mxy^!0a?^028eiyGX=%-N@8alA
z;S(@HCM!5sufEPER2-ngT%n>n)Ls`;iTxY}p0i(0HI8Gv?ETe;5`9=A?bQ<LZ=mlX
z(^q3osKt+0vb@5E#W&>#oH9;pX#HI}G=7*-+94Bf?>a<j*G8=+ixZX=;X+$=UjwI&
z7zTBSQ$@zMS|9N*S@ef9CttCy4lKg0@R7f&RNQ<xH%C0Va~0zF1~W3R#mMMri?}$e
z=pk1zkCV@MT$cG%d>AT)HJ@?$q#zxXJE!%F%ooC#SgWp@_2^9Z9Hx25XOuFWg2l&r
zFB4MbX}*$>!1QbWlaj|Fv3b(?vO}pF6&HZRYP6PCRsHF0m&6J|<v8HNi;+hYOwFGI
z%!aRh{(O4TRUxz}d@%Gnu(hLb_sHa?{!~4i@iu#Y)Xe?WQSG<S8uH^>`#pK`<T&NH
zz6s1hDn8pJnN<r;5c5!R+|@G-mc)sZ9k!LRWM;1`Pf1m6{hZg&&Ro5^!Sw}}<Ms3J
zUf<JFGYp{JZ^HsSjYXE}$3rXwnWnnRJOl8crfyYZ4%2UIB`tEXDp~28KrQS!>5qkc
zz`-j|h1y`Ap4I7_sI=w4pJ-0*3&GyKwRrL>BEFW}K&`2B9}@(`8ioq+=EpMQpj)?Y
zbc50&h+zB_9^#RB{4@wHRl0QLl5V|F>GvypL)-_NCEn{i-_EVEiZNyZAH-ireB_P!
zbU^owxhPz@1NGht&?5*UXWKhFSkAvdDpa$N*Rzj|cSbg!F<97SvPRTacRRJZ+0bgD
zw~5PxNmhk1DUm#B)ivPiMI=#^hpbmO&h^4ILc*OxC?JOqQ&pU)Y6c`&nVw>7wkf?k
zAc~&+c*xXuit7uDTj*<FyZ2)2v#Zn8x$Tkn%~!P|#M(7HzV*@ji-m8*qi)w9)$Z-c
zPy0yT+YSI;7I4FY1w25`NXe-Ur8cym)<T+9jTA`@)(eYcHH#y6T?H1$0)eb3i2IMM
z&4%!tkY$obb>{${H*elbA^hq;6Al(G-T`2;GG#J6UVOO3Tf3BHy`EN-!b)0T8yZbW
z$)HWKxwZ}cU|nx|D^&~k!2p@iEn14u)1pxo9jkFSQ0e1Ex1dey588=DZy%<jf#aj^
z+`YS$zI!*XO#^FaZoA$}(}!@KGAG3cfE+43p5AsFaM`VHnnF)Ck8uDfYVps&lkrmW
zoMfJFG+>48qE**8j=fp-d&8dj8s(qdXA;1m;o7}T>J6lQ@o#H?bWwq}b{Tosgb4wC
zwR27R)T2gGTZkN);)dn$QUR0Nfsg)u)%z`C)q`ahuKJx|8_w{uoGdM0{8;5^h3jOP
zM2hL>)@6f?a9+MFuL4hKPZ})ZZHT4ka&w2td4p&d_M#;H4y`Fa(FOg+(<cyDZ%Lf_
z;P_ge$8QgPkW_^XzQ=(}`$O-B7hLrW-!uohkl(5#TDaM$cuDBoZm^_^eMR9s9yv1o
zETFLIFC3I+xo#xTsQD_DP!!GoQOi1{@)h$)m}B>`;`b^Z{F^A|rK~U0TG%Qcf5pp{
z9*TfpEY?$k+1KJ&<e<$it-{8Eo$MM22rCblsa5d|fGnJTxwRN>P1>gg`O^WmrXgRB
za-xgzj!dw?K*F2P*-qV%izb9hkT{*~Uap2GBof03(*6{tM;G|FX`^->JUIF6!{y7D
zQ$0kKFdM~fT_#=jPD{$Jyk{X&BN(6eX6V<f07;pO>_OmVF3eean1$AahJPgw+km!{
zfaW*d%iX;qXE@jFOad=sX#zzQ-eC3sdrL4k54z(`Q|K6^mHtT^j*tn73s9v>l_Ko5
z(;q5FqmiPaDq1ly({4^t7)Jk_4>!G;NiEvI{Rw{UZLU4s_{7CQYord^Xj4-}V;F?5
zb+y?J2Y$VHvGrfmIIsAdg-MKhiJ^}<fOZc}a0~+}VIuioUUjt|&rb((hu*^n<BlW*
z@1V)7>xqwU_9f<y%7YU>5x0w;9yXb%-#dIikJ}(Mkzf{YdZY=xc_KVxzOAGF`n;u)
zj6pjDgznu(aAv!=Xh8!wE0V9m>$eF-Zb1^<q>MF`c`2eKY<t|?r=WuD%1p{1wK^r$
zVkrp_(hT|f9jFpL(rR^>-V*cx{ccvINM{RO7qo4@1x0}0Ts{P}%id{fhA-5BVZ*g$
zYV?b##P=9a>8;4SqFK*fJBqlw{$TO*%HYexn5&9|clO=ackVid?LqJ>u0Lh$O;IY8
zIPIkh6_0^RVKl2G!3Bsct9ic9*<_QEnjP8U+pc5Jxf#SDXfowctk*g_(;HQ}dfR==
z{P(08PjuY<(4p4ivBLeVK6><Mil0qUV!h6vcno?@f5c`ek3?bwUI*OJIaR+!TSTa$
z9|1_tC2;~a+PYd+?%SX6+<R+I9#hO>5xjx0=z(Lsj5R}}6iqh<d%2BD{Be9~ZD17z
z9P39&Xg_HZfRS6)DjvQJfiMaHwTQgXZ@4z$YyAbymGno@nXl?o?+SU^-QE2gPpwYh
zW@~Jy;)b|_0)PElHJBNLJ31y1uMuhxBiG|&B{yXxnyZuK^1RaJQyFi>)9K71xh;zy
zXTIAk85%NkW_x#$KQw1_bmdaoDAnchuJ~`Od?g8bv0IVPUaNHZT9JEZ{XHf;eqhqA
zAUYE}+%(TFDpE(JQ8b8uWq3TJ6a>2BRk2-?2g>27c7WWCAx<tdoh<(hc(-c;v_0ru
zN6%_m4(eERX~8RmgIh`Z1DAzV%9Hu4>kt%R91Fd_py2ikXK|`1<5^bdYB^*^;qiwO
zZ@hnh_mm{~Qxo3RU-&oT$cSl`*%gY|lT?Nl(74K)f}mVjEH+N$ZVe04iaIgD&8mR1
z`llhR>kdbhVBb`~#eM{PF#~Co3sY2hV5-Kw$6PgYkC)YLAtXPk2m60ZLqX2;F!4sO
zHs@aHInlf}oCSFw5zJEy%Db8%?w^$ofcU&So-J?#x)zwL*XTwz=~$Eg+=Vm>mz@@d
zW~z)_)0cy=dVbUlWz<`CF&2e>RFtU8C-VB<G(0uwXb~*Xj}lTfbJ{Xny!Xsby}G7|
zRDCK=aJ$_!a|!+4f}y;=6vJ9w-AB4hwo|WOy<&xt8&24o-{Ry~SCSXcVc4Nx8cSnO
z1RZajRor~^4-j9F@PE}<Ry?gNeZ?2`89jE>G>7fnakT88KT?k$&pCKyT(4Ec$uTlx
zxvaMB*4N#H9C~cLZNkGbTAH{}w7>?iX*hpWaoSVt)$R+oHauP$fv^AQ^SUc{HDBFT
zl(_w;GK-}D$OIrtIyJubrZ$QKAxl*|00caoBd4d93~;?z_gH0mENv8GOB4MR`x2wO
z4tk%wX3fRI(63${#3OCEdWCWsGk${F=}bK2H{LqBNqTYjk6cXh=QfV%x3SC?#u3vA
zV#D^cnxR~;hh09kjdcC!HWN~*B%Dg#8={O0Uf<gD+SUCKKCMr|bmZHWQh(T2jqV^U
z$L^=eG`>dL&#<C$>n<gcZ357rM7?f3qdHfUM-qZrbsNoBQ2MWDa2HQGT92Ur`To1_
zQW!=5wV7tm#FOtI+B)5{lIGGu%C({#F_u@7_2|{9>o0rTv>5Q8@4>_N)*m(+G1aAn
zDdV9kH4dertWV{1D~;e$kV_|dWBl}|UuyZQs!d<P+<N>*Udq?{BJV4UTvziIE^<qa
zmBkB~CPCI6bMLyCTRjC(SYe)Vd&u=_vQX)Po4e^RYY+=P@$s(ly%jwYX#`bw?lMlA
zp6QTRw%DjV{bGgIDGA}5+cFtA^Ye>C!{Y(q@K^$N_4806IcDaWfg!bo!xh(hn6<(+
z8jYPN@(DPbd#0azS=L&}XxlmOyr<LXRiL37m$_%gRWKCooRZjjhl`3z0X((ZPjwS9
zcZ{8-hFi;)PSgXoSwl_U@YaXG$u373mcldvggNXqzG7AN6%FMEG0h|x!Dj4!(rP`Q
z@eZePZv(^oeFeUfl9rVfz1h@e55<)KhPJEBFCKP$^U#wKDBPBZ+O(D$be@)oG}Acl
zv(H+J+>n>Z269|o97gwxmp7J{k+F;mDyJh(sYbaWFMhk2s`I%m4yxPWW@XCuVP;H&
z(DJptYCv~IDWFukw&!#k-tkKGl#3v5Y(H`E(4p=MPc|%2OT*lSit$|>8G5ct!YeYH
zcB$Oeg^L&06tOHUp8=M*#Z%=U>oe#1xs*(iebI1`ccmznh9v^$2o+Q^`W#8H>GHPJ
zTsp~;?~g4+LFKV67)ZnzDy6Muy&^7c>@dSo<~<5{Ki|&cpjtt0$BAOct|iUTDKn(@
z{4=?+s8Ge-(U<)>h6FWo_Yla+;Q!mc5#3eO;dNg6>!x^rZf3>maV0le=2I*BqyQ~A
zvyhBOr%+WD+eE}N%iYOnYNx`ZQ*v;=6SPS)<|zEZ+#Lwd4r5_z6W@_s$(#D+qZS3J
zx8VfG-qWi{l<FfY&ZXUvG_#}@s#uMt`01Snh19|TD7IE&o?gbLrlfqCtt<9zlhApx
zXa0umx={Q^+z;VgVI^^z8|rtn7vr>UG}_|wFm$S1zWqHa^=J_i>OY2qQZ&}65$NPL
zDl#%M1=2*nkyr2-V>$(~t(S@~X%mGA4%xVE(7|uNeQ!yv4tQw0*r`7tmoZBg3XtGb
zGOa3r$}pLB*C#D0r{>*-!fE9D4_4pbL@Lo;7sd)ykG4vsGv3R4Rr{Do6e8{t#b3~P
zCgOIM8Cc=*Af-MkV=$~%8ABZ_WU1x>Nbg6S5X)*2;^U1}W`*?u&F^xXubFi)P<?e7
zfjsrpn2;@euaW8j7A{;!BICn8i_0Smtn^j6n(Pm!SiEh-U#IaYo9fj~Da}}gDnf+O
z=e;*?Dt-u#Va=!g(#zU((nktwS;_EHQwzfT88Sq;hoiNCFUeX6TSh5>t@pUCKSnfL
z|14F_)UP!q9w4RIMjVjzbOek;7IX;}YfNwY1qD@xaU3q;BoB`~vV7)j1bhigy2yD&
zI|nZAcCn`CggQszG&af=hg&3jFNiZGf0WLI_D1xFI#U?Wml~zZ)j-6htHaejrO|b<
z1b@)JeT{aL^Lps{f;cj0Wpo>q*J}zF`8qmvj1QU5vRY`ftdKxPE?#`>>C?lTO)EY_
z7`gSxn|!SMM!h^!`_=LvnOv@ZIcv_armNR<2OEkj?IBr=lyE9<cX5mlZf0{1{uNaS
zJV*cWQZ*GYQFEuMka#I^+5xU;(XTk=!-CNM`lpUByUe8LPD9fv{2w(A+(%RVC;!q{
zP!rxz1Bb}62d#=Zfxk&Zu?*l1M0iU*VY?_kWs`)Rk(t8B#5^P=laan>N!FphIVpXb
z=64HHre3Bl`06Y1*o!!GhPu}qulHi-1c;9%*n>4dI>c*OZWd($7#F{4r%*<VOxzHp
zUM+h;csMX>$_i0XntVLdWikfym|D`Bgj)_M3p1fMc0L<;N8bA0Il~+;MS{!kzv+3P
z@K)p!Zyz%7g)PA?SOhXWs{cq-r2Ucxfl=2~8>dGMZ5!syomE5x@i*eI1rc!xP!Kje
zw!&x*>&wtAUWVxa?uExohB)NfgLmiiR4R^pyOOX=Us+Ts#(eP$hq~;;%8uTDIdAKq
z!nYVWu)t|d+tyo72U6OAGVn`B(6LnEL7DrzX)emKLhJGAFDn}ArW|7Gse3oNc$4j^
zQ4e>n<Ab4o{5HvMVmC%RNgE;7E1ii^)zxSg)>QmY@A68|?H7XTTS2`9f~FA$gW;@g
z@^>yy4ja-pm&w<kot~17yB?u1Z)EdDA`o%VB9t6W^S|49e`0&6i2XVoUVHL}I4)V2
zeT7%AUQJ|TP|qUU7->-Gp@HX)y|YCq5eQ%;+F7r51s61-SOw+dEEkddbWyVG=Y?ZG
zSZ5w>lK`F*9~YH{Y<~Q6{iP{Wun$kktbXm@f~mxZZ+UluA849q{lb+$`SZ_@VbDhP
zTe1$gxDC#wE_UzU-p!`XYG$o!I<iJKCuVNhs0B2gIfh6xAyp>AL8tm3Hc51=v6Z*x
z^85d>9|shiB2F<KX5^>H93|-DAW^zOf##!A%SY`5p~=5<b=@c5Kf$6y3~b3PLkGXT
zEqAdkV{R2|CkQ!(8r_Gb!&8nL3D;SS8n>UgL$Qd3AcRJ<&p&bz$br#;_<L-9{_;a<
zFX6VrkycNo=%xdr>K{Kd2V)qj2r+Q9x3_0A4StLG;P)+9i&|xuX5#Fo7$WY6Fufjh
zXq;@!|Ei6dweu7MBqgs_d>pSQnPK}zo_v?ww5cfhY<f7~#aUSZ5j~*+0me$9oF?1>
z5Um=-i;8-@h&qfFP`~oc6^oXkAo8l_7M?t1_@B||Up^l8=5rBs_ex7*zHB5zG(gxQ
zl(&txEe&=__NU`#42c^{ZnWNWB7PfgKdHn(_D{fn<vssFaV>bcS)gG-c$z=?m{F(i
zUC-uM{I2?hPMelT5Bx%x4Mg=qRMMIk%4JNQ-RSqEl#VJbT^pzQ8#itg%gIV%qO!z|
z#g!XU+E;DcdEoXFqR-TG8UZO-o*qiZE0@ydM??Cd_HTSjH8WhMcEplNccNhuE!277
zzUZIT3pbqD1r*5uzg4a?u6%mPy3{=y&}w0o(K;+_9e+pDOfj<p94IZJ+ca>WA|{`h
zzHqMk=oRicnKZUkY=_-jw`k#5nBrI*dMKYg-xBHuHA6Sfs)(!q);kC&p<b=q`$VN|
zu?T+f;E!JEa6nn-nb^;oLm`d%)$mcDSFUF96EShss^QR@T%zj;Wh5`pZ`D<7m{D<Y
zV(G$mZQH4Wpg;NUwWNtRGwQOeS|TsADOA#;=m^odbgorvfK$M-qxa@FPm$s}C3D@i
zN81K{{|R$Ytl6XCL(I{dG-Ld!ivPkDwOB|l4PTQ{V*59+pz8Ffck0R?UrBc8wEo{a
z2vK^|#Se5+m9a8J)%^uMF-iF3if=>gVyC7D*{++OpWmL1o>})emhp+e*1q_N@@BSO
z6~`h`h~!ImjNv!3bjwPzw9ST-l*XU;c7J|3efGQL$KHBV>D2+v`@U-aS|SC&&4J$U
zYg5T?8#HX#Fyl6?ZVP}j{W7;Olme?u$;1xVu}biSxxBCxm6#Ypn|Yr+xrMVEwYu>G
zAW;bWBlH4xzT8uzQ&U>$%R}=6jP~9)4VGq(a0`uhD({)<llj>>C<-P=J(sR3j`?t3
zXi1BiHti2os)jPf`?H~fQvdKnCm*|#zk{$Vc<>5r`&%B6BfM0sH}}PA@bjIxvjpV>
z4lQEd6KiH$kKzd9wyo2qP7_4KyVFy$``XZTdZs0jT`rL`MD9v;WlU8Mk&paKxc8g;
z7UbPcVHGU(`O)+#q_?2l&<erfC`Its(hg(4(3SKsYE9!|t8019^u*c==+feJQ?OLq
zG_1b0KWOT##wCr%>MSJ_x@~pza`AkHFZB&SAQ*9};(sRyd*QR2%fLySq7JvIt3o3P
zi_ZVnC5TSZuk!2Ie_i}pO%dG7^0o2SvP80soaPatoon9QU;bJF7od2<z^a~i8^azj
zvOi@aZAzXrI`xwFz_OOAH&?)l5B1~A{{H)LAG-s&;*F=|2*x|B_RJh9(gm;~tcY1n
zh5_#*pQOH|RSSh)#xq8NQX7C5?(eE+g8-z7<89C6&+YxzpyTubMT!%pL;qRZ?6hM#
zh4a82@rM`8PwBeJy@<QpV}+A<&Zce1t$npth4HiqQM`kRupfjTC}t7PT3SVY;k{aP
z?!f~YI3@-v?(4{c3L#9-veI*(Sw)U%ek_5Ues1mC?{qPwir)bKzLivK36TsPUyj>0
zdR<fUciAY&!yqOh&5&79EN1^MZ|4Rh3J_`YMA5zA?U&p$P4p?^wM~|mTR}liPI43;
z$Nsh7NYvLkI+z(%yx8)f*a}!El8WuqmtPWxE|6?CL>oQHC^?e+TxOm$$OU3FB2jez
z-~(h|J?a)`8JOM4@j9R9fq?PD<H0-&aN`l`JO~JY(Om&dpl`^=nZHa2evr4o`)@+H
zwZ$hqBrpCLqs%?W#4a->^V^WeaR}@KdGVdpi+Ayap)oR(x!|$;rI<u(PAcihb&zEg
z*59ctpEAEZeL7XWO(}o?lA(Z@=jr#eO#Ck-xo|9=`1F)zXL_?VX-4_|&xH}*_GHMV
zvUzQ)kEv!!IbU^tpkvLNy9}L^(mrbR;R?SmK5SlTWkA_?w$3ft?@nGVV~L)F4ewN|
ze!N!+_sCxlRy*&$%Wg>7EseGqu59&n__)>XA2-fFRqN+k`7d8TZS(5aPx2qyt-cyI
zx%SeI(>lJHetKhA{;7=*+sl&ij;V@!M0V-*jEMN@2GMb>&dz6ITk5KTA<)^Kx9AVD
zTp&HFz}R2sW6kOWg_YjBgT9{RN~?D4+cz;Ij9O>{p4w_&rbykzu$Eo6K*fV-qd4Y4
z0QuTA0s6ty69bRZg`E11df#%99Sqhq`A(owB4CX1j(zQ?-nv-c<gO4&t@7W@g<T??
zFLd`95|e%q=4z|LU#?*CCGI<q?ueRW1k=I7<7JRR_Azvn+6bB2rfIG(vOnx!Net^G
z@9!|_j~+5)lJ*x<%-?he79o-#RV(D;I(*AvFDEiUf>X)$qE7AWy2>#O!LZJJ)>4(6
zNefIAe>Q30xxSF02r_w$74v#2Hk+N&!{UQ&!|+b*2I|#|IJSAS>42KGd^G_HP{^|r
z17*Sqy)C^xap89u(2EtO^Z|@;^lQxBdAh4#tP}m3dA5>1=gGXf<^5b&L1zM){Qi3c
z+(`%k)ea352K;fn5*@=D`>$UpVAmWWYt*SHGOyrf<-djyUAU$%@v*v{kfBd9Z^O!m
zhn-JQs-Op@9rf6~bgZkkYW2&-%Me`g8u*IKy?!%yK?57{s6#u}Lk~~p`N=`x-Wp$d
z+RbI`wbx&%ZxMHb0(Z5UQB*n`2=4EfXEo=CqH^CB6XR2mH0akkESM_MB|KhyYH^)q
z&aC!Ekw~?mQM6q}u)`d%Ql&NvRkCI~0!Zq2#B19-N+4#)8D?;Kb#czCf$f<hqPWYr
zo`nAh!&tTQO>w}iwOQ?p!0S@-NI;E@1@W~v_+b#~pT^7rA1j9LmZzqZ@$SDNjeIwz
zUlaF>NYS{{-8NTi$p|-bsENyibRgsB**hZ!h*sOi64EU%b!lxiIly6+h{zO!JkzJj
zP$wv$aP>{s^;#<Kb4jQ%1L&`&<pQV6jTI)k-Z7ty9jjF93p&r4Ge_hpvHK~yNR&C!
zUe?kV#NC*K+Zy4@47<Bg=+YW(9L08LQv6O&Ep<Ibh0Q77P565pJwA4$^5&MVwNShn
zn6wAm@EVmn1;<lssJp-@(x~aDrlxP~AmeI4@gmzGP;@_s*t`JaSTK}}8TajQ$%>TI
zB%DMhML=w<Q>dk;_EeEz$8@2B>dgXfTxSHkZ4;4P@fFYMCxloohVfxtJ~)=HGQtHD
zFV<LyQu1is=KuTJcgCTpvke?rUyho6g94o<{HA7-sALP^2_{`V*4aG0P~4G$iXHQ)
zzWrCuyl^#n&K%Kplr_26r<t|#m#$%92MTDf&XWm$QZ!iBvgU=gk&;QGoh>9bXU$Dd
z+#;(9gDnmFl%Rar6pQraJMtay7j~HYxJ=J~F0^xSrFTDN@m}QA!dFB*MhS0v+ugEb
zKQ1Gp9Iq5ci7m27PFA{tTRw~rUP9Jd?rR$p5Q9M6r&lBRkOV_FbwV2CYp;nkk+dj~
zv}GxzVnrIxr=OvZQ|b0S`KwQCU^N@c>vB|Ko<pQ4*X^F}PCec(`|t5{5?Hx`>I`wa
zC9`u?_<td|3bW=w$h+<8{LM0(_DBypGV9AG<8CR(O?W(jfL>9V&g)lky(;!}C};yO
zqXg)#%x;{e6RiIP{6g09=<(wwWjq_=xjhAJwyFCRWR$vXywwN9bY|M`yJBVi{Y+_h
z8JEKp0f&N{y;siESLxBz)mDkKsw|3SzX^2r6vEgt^T`{NCcgMv7Obb<K5+hCrts3j
zq-T6!R3CA)W=oz|ttGjPcSHPb7x5sSK&wpsW#o#pvvZHMtDrI>Q*_g^qVzO0kWTMy
z;i37j`V_)EZkE{RMiyCo?)VHM)Vz&2eVu*I>FJ0h3BFWrt3BZYva{&@nJ2<u?3Pq)
z(Tg?T^B0gaEIW!g!M<Rj&Sm60kh#3N0GgRtF?QN^{jA**!n^_2HwN$zT1|cP`V|!$
zrFFNW0keT^S>6F6N!JHrsyX(DlK%cw?nD&&nwI*dgbJB%={iHk7U&3LH=RZ#FDLK_
z=PN7FWtDG4iIB4r`J|t`(Jb#w$5<5kAHxU_p^3}fY2!Ep=4-GZX_dg-zoizdTrB!5
zh<ewqPK$he?3({2l&EVMCS{6S^q%P#z6F9<FG>kbi>wO1DXI{XLI_N4KXXweAMXo=
z;wkSK{^$B|4!7evUVSnofB&Gv6TV?Rdng5@$$caUr05T!x=AZ|)+5mB0OmByt?(1@
z7yonnx53ShGgQXNExRW`bP@(lHElvSeKqHbdlSmRwaN&LN|m-s5VVaM84v06G$!jR
z#8wu{&e*M{+}}li93BsWDOJ2{2~VT+Pd4Iii{Sp1fW?IycYyLUT+aIV-~lzXpxEu(
zM`b*0x#uhW=ClbP@Y?*3Bg>W<ODs6(*!uS`xEOYuD}~{$C%`bl0!C|r8;tv$Sv1Zq
zZ8Wu(g+#~oOP1ons;gquoIxrqUbutM(|F;7!PbZBL7mpkeH?*^n{oYIoNRPq&O~%m
z|Hm>8L<pWK!5nTogX6Jauq;!%$znbTHfqK;-gpPhQnq1tmC5X=#y`cFuF$(h-BFyZ
z>@(6-eDQO4FUL6#8qZ;~8I*meX7C%sG;krmHd@&;?ap<0fJ&?#*NrpEMpxSngfA;l
zH0<|<-xVI;@%fgH5yFFk2<5ib`Wc7f-7gbhl*%ZTW-BVRAZ6{ScDB581GR&8)^*KE
z`c_1m#-VR=$G>jGIHesBB$bb@W?HEF<OmAF<LAzuD?M)F7p|;<i4rAkv$U>qvF9(a
zn0w0H?KQfF&f33nkUq6Ww)XFf{1^^iwti>wuC`9ZkWMlz)z}m-MNWYQddmECm#j1n
zFQi<;VuRez)A`)~>o29Sl4mP;I-H}&ynHg}4aqwrP^Mt%+=Y)fSdTid*U+9kRy=cw
zHsW`Y;>XD7!h^Ffi=aXB%NxyeR&ILvkr8!gGdBJ=M9H0cNPY8S^Z&QXE60k2`t7CJ
zU@%6FaevO0KTkphfw}Nf$7R-qq|A3NA83qx9=I=H`CFL*wn1)Y5i1_$rpVIqJvk2N
zmqT}bkjhxH%8M&|VaCEFn@ue4h$4E*Cln}Cn{Xgw8uGQvnv>y|h4ah`rZg?MM#G(W
zuE{h#^Ly_u+6S!$O&cV?$_;gt6)Hoq#hn<XW>w6q{Sp)v8$j-W36G`&I>egzB}_=j
zStD@ENXdhkr|Yw$H-o&0<5hw7rQ(|0lO-KmS}r{l-xJ1~_E&xFZK=l_+L;cdk7uTX
z=C)AgDytBlKzjxX4Ko#X{FhvpCT*!xwkd43H5)eM#S6Q6VDj@I;~la55M~^Q!UFL#
z)U+OZC*3$<#t=mXawtqv&TwR=e9k0?_~2SLX@xI%&a$Vco|LL)82b!wsEPtU#{Ice
zECn4jXXVU-M+42Yu~!bhkKv}{0eaIB2&-c3+_H;*$W?V#{*pp(t@J79k4^hx;C^1n
znJsD>fCrn4%}05DjGD3kNCumw3qv=ml!|b&%sI_ljUGF;#-l_gYxZ>OHWN3fK99Z9
z#Oq2*f5OWkbsU+Z#>L-QZN}mvE>9q3Tyi<ei>%WE#W#jjxPh}h6^jg0qXw;X0Hw=P
z0zbyamc4pTKf3uDeDf)`q~$V$8Oa<%!{a(u&@7jW%HB^<MNP;~2`8@zaEhFSG$~RT
zjN$QOSF@Ypkem@e=}Tk%7u2afI2q{43SObS`68}E^w%trsFOE@Wxd(pB(`L_ec>-3
zu}`}9Uc9t>3GJu!@pX6=6C%XppTXZjabxMP*q1KHvL9a5W0kwAP_b{*cPp3NG6>#2
zAW6gA-SN3!Pi4RYBq-~jHJxLw$O3cwX_Jh*2Q`(Mty=<Jn{cHf4*MsZxTyz)WGX`_
zQ!@FABj`)9q`K)00Z&PoNl?VM+C5>;o7b1V8@-ZRazGKd(8!vQyG=(3+RT5w=W=rb
zt*#uvF!=PQiu!i-KPLljfG2vd`a3T#5*{%Adc$qD@>yIlBwyWfj$EeVFvvuXVIqVR
z+ex_}y&>djFU>(#QbQ?7jj&VU)MJmiz=zXCjbN0uM2=8Vc3(@>rj$C$9~lw^9X5l+
zckm3404h-qbUM3h%pE?v8KCUTO~Yv;b-4KqTSViJU%pdNx<}WOh5eTd%z|IkFqf8P
zYf%-!<1nOHulMM;i$_1ulkjKhbG&rk_UR*?zj3=cjePI?`KQf|>dt^lAkH|?kmfOs
z*YJ4E1<7&o7=*!UNSXHJeSXbh?vQcu4Keas?wUX|TF+lbow8WGIk(1U*ck3>l}B*G
zgs+uS@&}jwyqUne+Y=?-Wx7og?>|Cc%?q->7z#dfXx!ogd`7r1CvNP|);xt&Q;jG(
zuSh!+&{Kt+W5&LH-F$1nQj;-X+oz9{Ii+MDfyzNcbL%P@x_ZNi-0m~qnB{d~bAU^O
zP@AN|`xqGXh+fERWXe`}JfM0Cpoz_8zWkT<UUyA#Pp}9MyVNBz!H+KC%=+%F8Nwyw
z={PO|oz%JPmLeTcFeN<c7afysev~i6I^4Em&SxW?+0me&z}+#4ZdRny$nG9|pwWde
zd(ARIj}BdAb_EP;<~214eRB7DZS;4fkX_>B*@`mJo@ZzNH(^jy*D`UL)XCXU>N*1K
z(2yO0<bze~stb2L|0kXahyU8PPGPwtfY=ERKa;pnCKawSBdhGpU67cGAa#v$=_tqv
z`h0U!&Gb903YNv`6(V24GTyPE3e^LA54lp&Y?)|vr9Fu1megpo!n}?{G?Bt@ptBaR
zT;026a6gv6fOT{eQ}TY8S9*)$@;HPGNMV~b>_Dk#_>4&PXEt+vf{k~eNrj6utAn;<
zkbX!v{TgrOrVJ(z%9kCYm{2B<U!@;wQxog=^eOT9KJwj0x#O6;@suH@BZ#1tbwla(
zBM)Ux>P2pY*BL=bA6<63hu}Gd$zJpdc}cfrT6MP2r5(c%c@LlYXiIIX|9~_4(_KC~
z^Wymw`UizD%%XSSk5Dj-zYLMqf8rp9IB)RZ*uJVV4v^`~B4{qF?f6(dn8Dfa>z3%&
zXU65xp%>WG*FNhz3L(z^z@Iu<ui_R=pt<=7tUnbPaf5lrxpmi^#AHF1x2{HY^DA23
zbT(>uPQcE7{c2uyr}V~t<_b2hA<;%LZlXF-XX&l8(B_q;Hd%@&y;Y2e&elWD@b7G+
zJN{UkRdnw^z{ROlhx8khd`sxJ7@dt9S-I)sqV3{N%6m)2l^y2#vo&+7T4||)dfDg}
zNvB3)QtW_@PyW7A1xp+%s_k2mYEb3M9a*0_+7aKzvg(RzzM%LlS5ijQdarzkvcJ#_
zX1?}&<Bt5bP0!w$coN%E>-QeBA%#V+{{(J^%+)EqY)kLQ@`U|jqs$4ozT{v}jN4y8
zEB%Z97px-4Z`}v)yAO0-l`Wk@9KT}hY?K;tL1(>x`ekL~oMTmn1v&*DmJZx~hVXc}
zKcS*;A*^cVLVE}4sc!vf;w7Vm9nmI)i^!tej2Q`MwI0eDQ{C7F#iQOVd)swTY+wXl
z;T}PgnIj+r#BfIt)dS~9#TMj<%Gx<kxNJJVH5{l!vcWR?Bbp*sLZBtD=^OBan$Lt<
z0Y`FD80iKQCF1lb;oM2gRo`0wr7ozAWD_Q`+3YUwE{gb~Ob}BE@KaYKK(VL#o@gJZ
z`m<WM2gVRpu*Q11sN9B%C+g(LV|MpeNbY89BqZr{+!e@&D{r0jQD!L62IJhcXjgri
zRr~-aPN25f{o+sqLHE6~rZwlqQ(b?RJfn9*g64(WG7UoHyzZ^>=^B(cb)Vdio@=;T
z7+;yKcd_}(dPdW6NAX~av#s}IB%6z55sX2VOpmD#-|T`Ok}InY9XFE_yl+q>hsf6F
zU1&cnE!xm<Wp?<>SWmIr(7!v@j@^hbLCT_33tVWMzGF-xua8~*gjf3U=+T8Ly2?OR
zftx^2)N@N~!urwmS5l@0WaRU=GvqX-eu1m{atYmt6h9lULj4j#MZS+?OQYOt%-%><
zR`!||STU73fG4F&*LMfH9uqmJiso5!@Sk*;ehq^#WZ8vG26e$TLd8I{TPcifx(Ln%
z8AzZa3TTyK71Xs|($BY))RGy+!leMi3R<_Wz_&Y^LG9^N19_SI6S({?7pe@j(*o_)
zo(z~P6LZDLdDT@2v${Al1<w%k({52V<-590Ep*0yg;@XjyI=iB+o7u!qB~z@(zZJT
z2m>z*!&SkFsfX2DvMEY;^e9y=dOEomsHbn*_i$F}VL}BnoYx;})%4MOV#D!zkZFKA
z@`B#ZN01kbGmJr7)6fW(e%wz`_#f-c!-Sig7iBM~K92s1n{~IAnM_HN0;Q<m$SHTP
z)fA`uY8Gb#rVUuKBamBqAR0o|b4nO@^$3z8px(lZJlXgpv2kenMg|5!zf$y7i?dBm
z{p<PjQL;Je-+l|&zrSRP)I*=EnC4v>ub?h6W7d6O(|3hxOaCjXlv^ou6Si4FJ@m<9
zC1Uv*!ozF<p%i#au;e^<vu;w?Wem<?Nno?<cz}~?5q^E|l{Glv7;A$_cnfc5)u%MN
zCw?);@uLDR0DCqiQ!J108YAWwUeldbuFD6-%>(86*0Wlw`5zgAe8NeuKul(cIg2|Z
zd@xlD0-o5RzmvG1rH8%Cca3k1%n93k=n$6PBgB8HcpT#Mxn9INI0rUYcsvy0rzI+~
zx^?6*XzN?6Tcp)|yW&9CjzHo$_W$iZ-X`JJEltdq#QDZ}2Zw;qgRvZwQbJh_YXKu=
zrK39Ww^`{UaCcvT;l7I=x;57UBSz#z|4AO)(=C!yP>mf}MUL}n_RnQB@9(X5_c0Sq
z^~GL@ou``An<^TOrP7?h)iO$CBMDVXCiAjSPlb+$#>a48ztDxL-`K8qVU`!JcNd1x
zl<I(^@?shHP<XcRl3kAryep(M<8CT+{Ez7{8~pSw6m$!-x`hA?cwrD0F-_gw($sgi
zf<h6**^n>-cNRg$UkslBnf^G65XUg4BaWcOAfME%7ZY)pgc)h>Fk^Du*usovOCZN2
zxNs+1pWDFi074swN=G)U>5DT<XFCl8?RTVY+YD~zWJ2=@TpW1WJ0{6QAP(@xxe>>V
zXaX(M+gn^NkO1A7G%cfuuI@S%PXg&Lt)E5Owiq-oB~#`gV?)cQP1Wm?fE&;xWE8oO
zohC!k)p?AOouzSBRCD@XwAP%XE@BsULqtr*JNWzbo_TLxZ72(F+w?=euK}k*8Q{6~
z#N^47Q_(WS@CtQ`g{+e;5E-2Z3*;|~vLF`S_bfuHQiGp<Ms7(#lP3MM@OXO6*z_z^
z=}Tn2;^?Zk4+!Ffmrp19{4jVDbLWpUg#Lg|UKu~*$`vxo$ToUv8U?<-5cRdV6&v%j
zewYXLxvG>~glgt7GN1tT+aVq_`pXwWqu%-<=eK4H?#jbPi>!BUd!MD~Pk`$<zg6vO
z>Z{5)n`uevT0m1yB}e^mb;|XcqK+`$5jwd@N9E<~Mj37=goC8lu4v0u?P_t!D4sFI
z@T36(V00Nv2jqTC^I1*YpX^_g)RZYQwCN3L`kBAw@BPF2OA72hF*_wC5AR`R*eF$N
zTyXKi;Oyes9o3kr1p{cwwHos)j6iSWWQ<^N$ll4fD7a(>W4)ZEX3d8ljp58>X62m_
zv6^iHunTdBtds@6cBE~Ys&kbgvE|@_pG;okt|t_c{lxV`ioMwV!0);jQQ$tB<4kuc
zXtM`?XKju<w?$wqoB^Fl@ad<!(AyVW$ym7lMWa!090Z6@fDIHFaHaqXyg3ur&(ckA
zUj7*#kDqc25YQNTaUKIpS{A~k@y)2OZF&0GW#unZH^fz7zfsL*bN6!qp*pEEf-aq8
zY1!J8Z(Y$h1TcK#^4MR;v@|7D)(#J`%NkcwyE8zm%)t64?qRIT^mQRyoJ*v4d^2-q
zcBM@{(y&cAMreQ8R0}zz31y{zG|Whx>7y5VBsxIL=ujeH-`SUQBY%I0c=tW82g{eH
zTz_@-dc)8|WZue%uJ`!qE7jjWckY~r0!knObaaxx%b%AU39NH>*MFYQ5sh?;Uu@um
z#5R$_F>V;yPx!4u<IV)w%30B?Szl-VNrj57T-lM-(bA<7Sw!!&RlYI8g<#Z+YuV@s
z9zUfxRCHgx($QW4<s9xHR0$C~)gEa{fYKsV;l$25q*o!%ZnIxLefme^j?(1{zYMD1
zOk!<Ilj|f{W^~$Y0hF@TY?4-FrSTV~{UCo>tnqOTA3R#qlJ(Oub>YRNJD-bLdf?^s
znYkYKIjDk0Ye8zPSy8C^Btt?~6G{6$_JX46v5L7Ti>)2Xr^7x8(jbKZFFHDkmWfm#
z<UwL^v)dI-5pcIRfAh`nQ(pf~8WlTWRn^&2X}m-N7$ze^dZwLb$FhjkWGnd|WuDM1
zY8*-=nbE*zIn@6MEOg$+gduh>JHk=9-W{>l?P(92=0iulJFZnk&5pC%OTjFtCE2qe
z?KSu)U7jBB{hR~NvbVak^Mp3F>rMYpv?HS4m)ELa^^NI=Hp*pt=T8aOyANkacuvl{
z6df;lu4Kdv<jaU2Zz(<DP_=2wBi#c$^3HPG1+rsOeBL<Jvr@_l43@rbRL@WUzB?y#
zXL9m&@1I;4I95>*=SpGH-pi3Ccf1;Rb#Ko39TL4$SBrt$<L8famUgU#;wYe+_k5{Q
z*V<;gAtOeNFm97-(Ele9+si6gYyOrtzK)n@mDf+(5gB4JUwAT#<DRem4nrYY>ij-z
zl-jLY&71QFv3LDOyxVTw_drH;m~52*e@`b)n%qATIwBBPN-*zM0;+G!inE%lj)KZ<
zcay0k6bQ8i)%U+jaa|Q=I{y7~dpW-j84n<0>1DmEK-A7mcpxl<Uk1YTVccQyi><U)
z;g;B{&dv*oo?T)0r0pKGub$pWGp*52{?xt9BRz(9)H0>?p9=dos#L&Sf|Lz=C!jLv
zYR2>ID#^|rpZ-wZF-(pZHFA*QQ~CiGIRrc9f`VS6oq};qWrzbSi@`>&z~(q`wg>HX
zD%m7aEFfuB240@&#ar~6^k0U1Cch6hW^ukq!DI_*iJFUd?^gWG{{#!MfM#n%3l#B%
z6CK*4x_MLr`kL%?beVjFWjaf&g50)pT%qbc&Y<jVElp)aSz$w=3Y<^Qv~ZztZ+hI5
z+0@DfM&annutMvh`_fgYSW%Assve;X!N$9SGnn^Q7i6+@hmQE<ZpNIe@Vo$yacfEA
zykCa_)OuBPBDdC0|CU|5LLBHVMlq@+k`$49r1=$LzeAhfd3eS{)@?zx7e{jC8Fn3|
zmX*paLve<TPa>Dony6w{dwpEaLk=&->veoRm13aFXs!NEc(r1{Ce#X9ZT#$LB*Bc$
znnK&X->i4ep|H&%(%@#}?W=yvbX~jyn-`Ww>0U3Og?9@meo%jn!9qhRXnI}2G2`XE
zuR8jr4P|uCM8-Y9d&+Xsai-bnP+}`HYA?jNo138tR6C|$Pj0Pf#>1B0vMb5MUJj>n
zmH~(ja>@8$JxM>sxrDa9OmYXoO_To=yF1C3R8JZ$2plc)rOs~wHF03f{=@#Iqhji>
z<0?<*am-eh77)gf9AbSsJdWZCz27co;@mkYgzfcomN;zmnRTmwY^@O|SHDkvWf`r1
z$i%IZq$n_#o}UnGRr&86Dhn2ivI2sF_XP;F!aOrwrw%*fIWfDdaUAu={fL2yo%xR2
z^%Zt+1zkG9GxV0m)O!Xz%3zEABwQPtDkMlo_Aw85bJiS)W=AU(by}cI0Jmt{^`y{k
zhxE9V$BI~F0;`zMY76?phBr1^cgnS({)k)7hLE^8dkjpB$5oo{g3Qm+7gegj97PM4
zXJaTsk5xkxr%t3<)ugP!{1)OOcl8scLgkXlk3}&DHsh%K%zZs(cVAvwRsE$kKzS5S
zBr|(XG#c|!U@3?CNrNY9KSjUQqV~Rkqm;@qmG4sZ*5mxr@WMv&G?d&V#;(9^#oJXN
zWERc#6JAUdy3e1dbas|u;6-Ie9OCF6f~d1`9m9lkVMWv}RrO4DYrZV~*V)Zq*9$s`
zH~Pj|`GaHZ*6T7~**?xFHND%)YLH_-wpM?*-(UTxX$y1}T~?HsVhQ!NVXkcgIN@Co
zXN;n-Vq005)q}6r*#M_sG{Tj_4qcr*#BXg2><_grdgPkmEBO-2)G*bj76%GsIf2zI
zgURy%1fu>UcLZ=JaOf{bU;RD)xcIiEWXf<%-Mv;@pvzdYqFyo6ZrV)rr_{;&NA?oN
zhOfQn|D+g}$kEc4h!}@<!8wyEPW*%5KB3uvY%ZbIsj09f%V-Xd-v5mHb^p;U0IxCq
z@He#xEXBVN3F1z~nR-xO*D_TuW_<@%pF(t$0Uz#S3?))F7U-bj!R9wEv}Q8d*<nmG
zDusbtQz@zW*qT9Cy=nee8m<X;gm)afU;LaHKm{?ROOc>!%Ng2%n~%PJL)&J{&fT|P
z-A@j^6rAtb(*FGU`%z}|*(SeyG%{q#kAtTAe&-(^^TpS_4y+C?RqFFJL$igZM)o%&
zk2|i+NNjL0ZDo@PL)75d@X%VlbN26SS@oj|Ypae-&%0jx`rRilZ>`ThP_@~CCuuqN
z8^4IZ{^G#KP0!}Fnvot>{?bSy+^ir%9b;;$hsVn>m?iBW^q_Xm)0da2LyM6oqC6QP
z5q1Lieh+30?rdpFDhiZIL2Wl&yOC3>+Ra5Ge<Kd{N~<}+V#m-vdYGnJ`l{L6Jz~+K
zEfhYb%GX+NmT!$rMhw1dSVD?g;@9x;csZV)g%}j6q1-vT6{ME0yBAq%wYwQTk9@Q4
z_B_ho97Jg#HE&;GmsBqI;VbGRxvlBX-lcl*&ySESW+K@|ZqzQ4ENUk9WceE{G6HPI
z+LrO}+4GlXweQkT#@nQO-$(DzrG}6@>BMOo52ec0fB|HhrN4%~7bS@j;c^&wt~>8b
zY9#wh_OKy)rA4h-Q|QF@#Y+F}l{`2QBR2OLbadu4$vMK~`%v>tJkex3<$$5KzUA{;
zAzoN4whegM8T0X+3m1=^(b|k?h|)GVLuU?;*TVNariQd1J9g|0?2B`MGS2I!m%cMX
zT2U~20?US}^gt))ekN}E`p@+D?91Mp6O08JYhZkX+|vjOo?bfcwM{;sPEy-@Q3^6l
zUX?9`Oime}^q65aN9wMbyP$5D1%;5&7b{PCcR}2?`R0WX&Ruy%<L>XiYl)n?pDe1<
z`eB|lF3U1^AEpjBuP4tjP6iyD$b2YDsOkqfhE5)T@%VO?n3$Nt_GQ)|Ei#yVvfE|)
z_t3i~)r?0@gp(NntYx;ofG#y&;+ee=dFkmUO7Td~gq3>r{BOT;!IPNt9lo?@uAJEX
zxU^@z-X5?gq^5+Cw6e9m+Up*g8a9P%|Hp6SdGz1r7fxdK!S+1pPa_nV|1;;`KrDP_
zty%lv!6YB+#R8T&w9HsT_?Jr1|1!yyvzr0Ipq8)3SZWrp{=vaDV1c?%y_cm%k7w^c
z*{pvx?<F$H1C%AfrUw~NwpvRUzVF;?x5P#S>ZH3n(fs(aQC<(L${Xyizrk6zqUV*1
zAeyh=FN4KJ5aDb8b0-XFYzWfx<F;;Xpf0w9RfAi?3Nw}MFlj}y)H>?8U|yI^U|&D|
zR&1M*w{P7NC6BLtk7Xyle_~=s0e{K6lkfI((pr1ejS-!_^#O~&4jy$Z`Q^!E+ly6s
z_-Fc3v6v2~C(p^z5EHkv+T^-xT4$d;d9pC?=+q*^d)OnGkX}dMyt>;s*SA+MnOY6f
zb5pn9FugL~%=ku@h>NSMZ<-fMewkUvPOU=TP{{*zRN?BVQ}yHM=AV50@y~yqT#P;~
z?D>!6sQK2_Kb!2}AFKbRzBToLz}Jor;1YYL;d>%f|K4gr;EF$!7yd|BOj%~5u&pA~
zW{l0CB!MIA{sd3ksoO!wU;K71lYg>q8gCzqX@`Te8*N#wgEd?XD_g_UO}r~$s`gRR
z=G}=8d}Hih)>Hw*p=yh+1^K!YxYwhGVvRe3?-CwQ9qFZyLo(IU)5c48W!W=(uABd5
zvhP+%enAl542Sm-{*P5U5G<e}mcQ%OI^h1gbFg>N_jpEqnQSC868$d|=)`;CtoGXX
ztrbOto^=d<N~Q5fpM@I7!8dhBk0bBNO+_2;r7q2yEurFbZa;N`aWcn&Ql6E~+Eeth
zTYK>F@6oDH)xGmjR+A56wWUqdPm7EZ<=WniMr)hi<ORf!W4K`fEWddVTHTHvE*NbV
zElbxXO@z@Hoq=T!L#KMCoubFTpj=J5jHdQ_V7S@$Ul#bnKjtntd-iOPn0`gyE#}-`
zl9rXZY+t^81!>%;)nxkDeWZty|4b7YijjLcQZ9($T)c?W;kEiN`dv#JN{g8NQkebx
zaEr*6U8NGY7VtN{i`_?V#jW2j`m{%PuMcA}6kd*-ttS3d3<XcgqkUV663o~B`FApG
zYRQFSr^jml?Ngo|<_<3iJsME2N7{x7KfW{ZY|qA!ChAxZxCcM78bq{ks!0N#!ZOQe
zRjEgV$wF*;<4(fi2GDQVcLH>pY-|U6rN#7%6@9a_=DktJmj76M`M{{-NQ`mv`ly9F
z{ZZK4U!661>1jM~Lz~qJ{S^)F6^QSU+ls##oZ9$5Q#mZ4&nz3I;4N)h{SE%&_{+n0
zbWeauB-RJqRnNi_@aF%}SE1mjDqRDoNLlg|$%Pv#o8BWwkfdqT^u%e|z<*is(dqgv
zb4%XC+&<2TPVwWeHoG{BL*V7hYt0EL`iws>t4oOaSvMgxC3+z8@ys=km~>*mz=6{H
z<c^xeJG8X*{rrNLE({+RCo$%2PZqC>nI`g(l$3S)=%wxh|A`YP3hjwVR_TCX1z&$%
zEyiw^FY_*JgF`fd=8)Jq3W<FCSfc;deobQRv_G2Y+b4vDKTmX{+tnn`CGGA6RLdA8
zL`mC2e_+3UD`*3H*OeZcPW6&Hu<zi(bZ1QMbw<z6Lq=`*e1IO<*~Jn-SXW<C(c2Us
zWrnuh;GZUvn25xKkZqy5`^!Rjiq=P6OQ#0@Jm%lxx+$b+1j?*VJgH<h_4bI>gBqaM
z4qN0LY96(u^wZ{#?~!JuLe<>DPIi9*r@SlG$5~TI^S3jJ(CE3HNm|m6(Gb9X0fSB4
zZV5C_x<JnRT6K?5KS3kgjI073Rn6rvg?##txR>G3GJr<GE$E!@xex5$Kj);UR$%uN
zT?YwRDnp<g!$|KWuTdaf#u(bgcOKmTtZhx`1c3#D|B|0(M0W~b%dYWRzDB`2H=57k
z)X&@dQ#OrwxaHn+^-|L5Oa3Wn;&R^@#{<y2A{fQwTK4AmEO916ttu#ye(1%ax;8(5
zvd{7<*Eqq&E6V>eW#D?RDy22=+186i)?eaTPnsV6^TFO0!_#U~_U7mE5AcywDZK1N
zev)VwmAQZ#(RmLCit!F3ke1M9W0_iAiL^2&00pwmBbM|`Lm5u8v_y4tZw8TfeQm?K
z7;oV1Ka%045DbmGA!oYJ$R8_jU#G)*cWOeIm;NEnXDZA~zTYx>;Am!frWHHE5?8EP
zk>ZEiaP!aa$s(dF(jZp`QF;EctP6-(fI$xtCI2kUXoQE^?#|&ZqrLa+qq~^mH+b;c
zf&$fL>R+F=bw4M`;Dc15>^;Sr+H97QvEuy5qiXt&EOU17s121tG1Hrnid5$p*h%lH
zozy=hWOqRqw#-J}Sx+!b@-~)<Z(7LY%OjN|C_%cU9&rNp3d~>|^Hr5!a*yQV{cWoT
zx!5mE>Z&eks7hgyFRY)ID|wAefS$O0(dx6C8KM$!s$98Y=*3+_s;GEhCgiO^kwrnm
z{J?E4(}R&}(zL9zeJqq|Jz6;FS~TxY)uC*@y$8M^tvr~LW}ChJaB^hyH_-6<J~~`|
z+~K<VG(E)^!eWh{n5l#wGF+xoe7PG_7N!I<>62PpCT>KPa@q;~woxccw=jWY%93~8
zslaO4(!_EMBb8YyyeN56-KWU(#Tx(ZB`bYiqJRq^BTf6o2C8(QGa@*~uH?7Hi<xU2
z|MLxZGUL_!az(O$EFVp2UPhNSL=Pm_sc%e;tlDNI709n5GdpYgLeGTQ0H&ixwrJnL
z!0%3SkWW7=ccfS6g3pgY`y&}9?0L0YQf>Ew(7t8x&uH%-Y)%p`<utEU+25p2?!b(P
zZWMUpmz*bzW4K+A*u0xt(~MTu5?7r7^o;c#kp;-SwO_1fmdU;_YuU5RdXP^)A<k^#
zqiwuXzxJ~pGa1-{yOwuzz(Q{EW{`}&<^*dqa^UXZs)Fk9v@%WOqo0?N7xycOOi7(p
zwc0wvyN@v4#_W!7q!Q0Zw=ec#87JxHX3-bh&@&KDZ<~>FAkYTzwJ+0|FD)&bVl_?1
zn62mj8tQe8{)TiL^9Hg=@1rk95e)h$Z~%vU#v(nZn;s)x7BX#MH|P|F+2#WW`j7U$
z2G6SZ%K&BlC+}OEF8p~$xzlrvFeXoXd|_bDmFur38^>8pc2<Lv=r#Jwf8;Cmuf!Ks
z|BCq&{l6HyX==-P7%V2~{o-Jyev&$qgmmpH2mYn;lo2b>rDqCR#}iqf+~-N_eeNO{
zrh-FFI`IL<{7RZ#XM7T)dc>$aC6j9JU1Wuk$i^iztxAf!?Sb`AxC|mM5F&TPt8BTO
zH!suQU|#!Yhr8hrn6Spzl63SadZ4J_D%S6&@M5WeXgYn<vd*5D#Hod%Q=r1A+4}m-
zF#q(+u%OyMzqv)Ekuuj(<|B#^<Vg06`~*yxMDG|jj?;@^Do@U;ZsJ}N!powTG>|qt
z>a0hVkoWccij|g<M}Z)&F8!)4t3qQ|l8MwMHLD9R>oYh8x7TCq{Zs}%NUkrV9@@<6
zq5nNidw|;G0o{f7L_`)IuguSe|M(vQ|D%4^zcl{*j=sW2S-)R|5Bg!$_YF=<l|^T)
zJmAp(Bux^(Ryke&+P<ll&^=pN|3*nL<!1&0iwI`c@0Yo%*~U2sOKTZJo^7hdFJ&|q
z7;j0NTysh=aT#NS=Q$&zkISG7<=evmko;esB{M1~`@3`gi*P|1eA#;KKRVc#{bL?T
zi5G&OWCpN_8bk(!$paYm2k`pl@L^BdvRi;9+^;Z^0^rWqK0iS~qlxr?Q?RCf^4Dz{
zQhZ+~n~TKkGwbyqd`(pOO@33f;l^<U8)1xhk<V|9Q0dN0?#9IIM$}Lmv!dnZ;_IZ2
z2VwXT=;FnV$3msUH#U-}>RMaBv?+09^xTaD0ME~*Us6OfOKC5hpG_gU1SDho`dRsx
z1^QpV|G=p9CD;<Vl)dz)B+MeO6i_1h=p9}S4nnB~1lT&`$u=8zU^4i9bXhpy3yw^d
zgO0=GWi7S-l%u5(yMl45(A>R!WON11%Z&PaW(i#c0Qt{G*0->14S+{~(iA#N&BQb$
zUyNF!qM{%;M_ZefkJ=R5z)T#CR2%q38rBHag!Z#E<R|<ps-k=O%)@%&pW6Y;G*Q#c
z)x^oZTe=;?g3(95o^E4c@)}JSNt&1I_k*akNjg(1qo%|rTJn!>wt9&^VhKc<lc%g#
ztaFzFtZ^cmk%^~TU2KpGGGXL==^LDv{z%iFmwEG0YP#n;Is7FZQE4D$@U6wchCMTw
z*ptn5(-w`1hhCq_d=T~Xz@MX<b0EUF=ZF$RTx33?BB@2z>L7K>EFIpCPNz5xhZRbZ
z8GgAV`(FpSE`3X9zbZj++Qj#pC%>E|!$Zbddmh@fhUCFBr`_FfZ8)e{n^|X$MR%H)
z&BQp#p=;|+`<v%vR3;t7^d<=eE+O2UH6`W#P`Y;7x3?K>GCO;l^;VY+EC?VoPBE-B
zhp2`f486Deub(92=C94t+Yr&|>)L+)3Af)?ICg@>2o%TydyfP3`X3h_)9*Rm+fKh~
z9ZGq_oagL~<>#sdzdYCRRsUcvDJQG$E1=B^aa|RGJiS>tj=hvT_G2wX^J~wVI{eCh
z+k>_#Wav3^w;`>q7^ErD7(Mq@3<kmqVwC{b=GE8|TE+`KaxYvsue_Hh-1V>%UYK-G
zq;=<7_ELRjrfzz8=W7aB8D=)Fob{=vVKDLHq%gOHP@)Fw-k;xmW+!=Gy~&LqGYR_=
z8ryAD4Tjf6&(Yq8lF)oin>G{_CM?W)t)Huxs&=V1)l9AHuxuT}IOPCXEASs*o>zBT
z(O#|+&8{4PcXl1R7D13#U2xO5X{cQvJ~%=Y&J<am{5<H!>bZ71w9oq#i`=<m7;`hH
z3*kn6VLcDa+T&2OZA&o<KlkG^X}?oT_}U+dA!Ie*^X<-^X{QRl#dwbR=A|5*_n!1q
zu<Jn~v|`b<!c1*p#o)d45Mp`b#!)2Qt^Z@}%j0@V-}es&Gsqe$DY9g(i717#By09i
zDvFdyln^P#mMA6>N-4xxin5i|kg|lxQjw)dQHUf;{jTer^EPHa^ZEWh{W0^J*U<ZY
zp7T8SbKlo}UDs`qgYV1XiaQ#VocZwK!-d2{Xnemcb(0Gd!F!qr>Z6<+9r*~$E^QoI
z!8C=IkH>Mb7)*tl06>r`diKNrj*5BSz-ClGGNnh%31{?AE>4wyvF`;So8B)^rfc|%
z6AhQak3CJ2cM4KcRy_#mJ$iggz7`Zi=$kfmn$1D|S!bI7F>8uK>q5dI)UPV2{2u$%
zesfhEN%4&jMROGwPT`I)kO%W`cf{L<8VXD9>obdY7Gj%pXF>Tr+&RT{z}s^LAKvb<
zg~c^n1o@B!_`^YEylYGGMFAMoDQ;Fke<U26iOPvDsSV)VKRpAl@*>01_$*W}rm_q#
z`yd6k5ZJVGhv*L|q=XnRserGB&Z3xw_y3rxhZ2rLK-3NfgqZQ;tAXjR&0D~}>_oRE
z74XSPgRj@#ke;Zmz-btcF*bmaD@|(6IblRdG1)o#VE)%PF@Zp?QfXQUqDwcWt2q<R
z+qZ_++wF<19WGr+GScS38IS~u$G?i7gWPNQs8OQ`4JWb-^91wLQy;6&3vGe5IDq`R
za*^~tTQ&g{4xYUG^%Iz<jpzoio&nu^^x&R;`EBW7gsL)X&}im`O50VjtE<oj;<(BV
z)6fgu0#<gHLZ1C|)gNvk0ZVx*RkGWDV`>F@1bRv6qCip=q2EE9`(GVE+X|mBEFBF4
zXon3?3{aI_nGo=&3W-U@A%tmSC-$z$<dGOcU@X)~98X{M-Yk&}9^;R2qBAg`Bl;ac
zd*001ut?*lG=G=~eB&?%g|N%D8Y0ldHPY`EO5AbCtyd$Oq{ATl_YgLj@+19qkctbn
zk|~0(x{~5VYECry@*+rN{&mn!Z@+9NW}4U)Z+|!#DSpDI3z(Wc5E)oobPW743G~r8
zdGF5(zlI_VLo+9qOhU@dD*sfR<IT+&kO(U;-&#o@lH#g@l5T9Mp<Z+e>@6gs$r4Jc
z@O!IQd*7sLbsyZ$=tUQKZuw^)n@}*YP+D;u96F<}TJl3x4VM8K`&}49q9IW#TZLB+
z4_$J-JA(DIcT%e<b}tUNL+Yj$qFfMuwNI++FZmrCjkZm=flGHiG{#<MZaUG@vJLY!
zs!K-%sOKDyLvo}X)OFlu$SHJjkfs{6l0MxxX+E(M)$lvBr?Ja{yV~!3ettf=r5MP1
zu#WKlbM5WJ%yWJ7&=lhcJl2L4sH*}*u;l7BP)`fMX7&rRFmevRcniwp`<R8^F7jMu
zZEfo^X3@k6&<|p5m9X?J_=1tAPW9omGBDX)6$SoEGihting&sAF4gY#Zvi`Pq0s=B
zLGHL^Dn*Uhx*7XA^b}ZDpYykyYfMzz76kBO`k4hYXlZpP*{3$i%M)WoW+Fv_Y@^no
zph@j`Y%!9&=&cK|;oSolVxYLj@^?(O@B?~re!pUu>D5yz@!MtV)ty6OLVEV@y%m=6
zQ)Z*t1mOf%!gUV>){0x0*6(7swSzX0JOARW=yY6yD%=Jd;nUDj;t)%=E4HGL8KhUs
z+zyle-lamiERr$L+J1!M!9UBwxHt5E_Mq;&lE28GJWB~cOfk8O%Z&QlX)xf%4;Ew=
zvOE46kIB5mkxKuqo}G1ei~4{*a4ykIW$WJ%Cm{;xP~xX*#3W9gR7OXQy~@>RO@{vp
z_b6Hq;8DEJ+8JvRf}=Rk5E&_l?tFhodA~~_YvoD}xD`xCC*R%`PH>B#uWt+R$`>AE
z{6-ZiAZ0`Wev`1#6Nzspsl`6xFt;?CQi2k+pcgy=k;8`?!X&Byb_Sn)nrZXyK!Quu
z%a7i>C*kNpn=H<sSzNS#qi5CWtYVK0G`C5xQW@msarV}7^qMfqp~o%5f-TE2m?>`p
zzW{zb$KX`4fX!b7{sWzpmt4n<l;&Hf9%w11kvqght|9{jsshx@&b+4la!73Bh7|`U
zH!L}`Ea{Dn9+XU&#eV5&^-6?;Df6G!Bz(TCS>q-SedY-R=O!OmUUq3;On|ma<;;{T
zP@t*9#qiUga^fT``5g^c7QU73sBn~2un^Mf4F0(LHX3ujp~^@3y=UDfD*K;#{=<?t
zi#OCp1B?Z^wmJP|?a|XAg7Rn{jbuO#;9PYW;8+=8dLn;(hbv0q+6OQzqMj!MO>|vk
zjp+CbOQaNLmaYmQIow5s<>}WsT9Bj5l)6m^7`_;le>!T+{MgG)-#kzyIXUi@QBPX?
znL)raiFX}-DmN;U*V<Arb^M4y72Yy>bF31)E|f5TebSL*$5^x%?y$l2DpHe7hZqCb
z&!=b<0is*??!^yO=`dHYu9d|mC<#YEH1z#*?3(Du8uLO(mn>lsc;sf}ENNn!knR7^
z7^%{r5tCe=e01De<HpJ}D!6IzbPM9TiplC#H02x30W|n|wg?IHGIU)<iWnPIcn`=O
zQd}jlBepF%PaJ|YrAKgxH+Gu?N|e#{yw-!0E;3coMe|2bD%N@)NuRXp(!m{FfLwEu
z;zYNY9kJ#y2(81)ilPF$77{mt!u=wy_H1#yI!@0i1amXQcZ(bn%gK-Erc~6ASJlyJ
zF#+`Ug_*?JN4Ia!=|l7AbU=eDG`3U((gtc*uh9(cde4}Q?vD^R>NRKrH*O(O;;W|&
z{uRc#gfUvmDNq#1jDT?2_NLi}I9{hL(C32nJn)}={19t7BqSeI;6(W`Of~xh+Cja<
zmiV*8r{7%Zu%WjML%7nxlwrQ5cq0vQU2vQd@n=q_MTpzcpxj9F%T@Dy`f|~M4`h?a
z$NTdApjo?>IWN9{uy{fNGnix5Z3eqW>|gW<d=w6T)I)PTS#sqNiX_(|q3SAv4q5mE
z=4v8Iq3aeu2Sf{2q0^z(f%V{i>mLtG8HU_;Z!Nw1h-Q5<Q%=luzuB)0_i5KcirRrO
zK>;G<$p=MqkG}Jy&{ZF<4VXI*3_uyEo;qbpD$U2xX9yV9Y*4)QJT3+8Ow92P=^}qb
z`UJ7utl3V40T!|_XX6F6<XRVoa=9azozAOI9Q+1+_eyzjR+Na)#VjSW0KZ)s^bS3&
zeBSESKwM~a@{4w8b541tTO)62{FK*$#~EIj1rk`ZPeZ^{ACOmti%!MJ10%_WYz_^^
zPsK@Shs_W-M1VUS&n4O?WgcbM8@5TRydVG|dA!*|n=*93uz(#w2sQiuK}!I+j?SRN
z+#i7}P$JOE&+j}iwbW@BKeR^q5t}aEQ6&v=@TlJiJw-P>+_vb9d}=zl^4U2GeTgn!
z07QqwFTzFahcJai6H{^cy^eSog`x40d6o(@sY{C|L*oYFJVirlTE&I~(7Doaqx$dc
zgHD_}^`(+S#BD>c)g186=tVL-dG$BXsxP=nT`ztqRMaPp<N#2F5mN6>@?dk!wMl)<
z#nle;_?@fPQ+^WdE?v4Hrwq+%Sq1Nw(2ohmri<QF+kmjzimgX=+>D4Z_ovxy_18)0
z24LbsWmb62fhk;cL}gpB3w(~_B;a05NS%q`gkif}tA|nzogFBm28YJo=uY2<<uA6I
zV<UGrS6@6_ymA;Pd-pDVph{zMydOB@f#Rq7{iW1J)MuaYkHG1%vxZ2GUQi0j3Sx?H
zD`69}Okj4sIV!-Yhp5&=4ucw(pMzqM0N7s#2Cp~$!itYHcwQh9|K!E$=7yD!x?-Q|
zz=Qc+enXSWA@(O=eU1W(DSt%|u~#Zo&A>xK{}Osp_2TykFiKR-U_?S0sAIYm9b=;*
z6b>d9C?1`>gsF<7Q@=-;g*N!Y<x1b%#27{BAhn5S#YhQqW<5qzfNr<CS3!$k7+={m
z)7l%;Hv-Hxz4cM*i3!lRY9QP6Hw8UmIwnDav#wY=_RNM4XC_cqSrDjQIJkU&vnxna
zcw1J$kBzTy&>T`JkS|F^0ZL!=3*#CnsVVoJcT05ZrG1s>n=R<q@hL(Vr>T}fqwzh?
z_%@M<L;IJJ`uvaXa_mE?Css6s$At78dP;n+lOo0&1N<V^wFRn_R9L$tEKQ?ifrzdH
z>~_jPwCkyeXaZUT=pb)+QVNVHqj-!f>|Ms2VX6|hv$nLI%On041W>O?LEsf~%q#1D
zX_b!xb;eA$KkHANd7&!#^{qL5u=mwj_c`8FQ>@#I^D`T$;k+u}AG7eug#ppGEyz%S
zU>3o%N&>EYH7sY$Pz9En(LGU~IdE6dio{+Kj}0Jiq^qZ_=h4iRp_iQR3twL!Rb0(i
zCo`8}pxLzt%~xqxL(P_so-l9d!P2ho(CnW9;N~3P<RwUL)2|2ji~=8#)vjyUNvN(5
zBN<_a0Ud-Q$b1+pCylSyQ=ZA~Vb2M6xB<lvk>olZngK%PsS`nb1Oau}^aPWGBPSZ)
zOAJ+#Ns(w6zT7dHK>hcp>4c}Lyc_i72*Ni0e@l{5_2yh{&M+Vo6|S~oox^f|hYboF
zy8SxCk`*P-#Y%FvnQLHxywBSQsvy134E4-)28`5-dS=BXt$}=sq?;n|e{F&bRT32L
zJCIX-%-b)WVz0CJCVF;#Bbg8gKD^5FnH-Y?3W?sbsxP<&7IRF&KXMM!X6%rIc9t{-
zO;Bl5e{X`u-j+7?)lP^lfQ%yl2IyBdy!(G8_23-WL?}Y1hJD2Tcjp0l79bZHl@W$J
z3t){3SItfeo>%)qH53Ro%&Q&VaHJ?eNLmA0+w<=UJI@WwHWQbo(c_C{kpssj<4ARr
z&kgx@nP)Gc+eEo%=fba_--UmicO0qtA{iao$Q=r$^^NYvxFcZ&7KM)><LUpKnk!CB
ziMgDb|FMf6HKJp}GV0QCcfm$asKDeKMw=D$@#VxR$&b*28XG=%kSJQ%7J*H9;^NoW
z+@b28*|YPE*d$y`q8~lCDPxp3LCR_{1jWL`YYj(qaflGoPNjBjfJ+0(6sq8ufKT!r
z*bh#xP#I(tTR#aWZh;bhM}uuq4gmwYYOJn&4p~-Dd$2)qwbA4u6nrKYXl(B(K5DQ&
z`8>Dps5o_71%!wYv6p5IIF$Zym%(g#-)~^0;j>IN0F7>myFJ)Ax`h7t0I0~aM50J3
zJao3L{?9Z$+nn72z!pqCyRG6R#A-}ww#aCYuelf#Zdr<cldx3Nebi?SYdWB_-mYtN
z#6U`?P2ap<S5>~J6y2U9Sdh9PVsm-uENzCK0Xq?j)0&MPTCSsMkU1e605OG1AYi_E
z(<yi~JgEQrpq^+V@REBnVHL)R(&%E6nyCT3T*zOAfM*CG;6-P{!>`Q|Asbj+>Ct>|
z0$(&<KvIV3uy+p-U19a8*~&dSbjj#Yya%L*5$4k7?j~qfWvAOiLyWFE9MHsuf{-v9
z$gSSITcMnU*pf55u(|EfS0jWb;yo)~w*J2Hx*#`1Squ}))>D0L!6ha$sg+0z=y~EC
zwIt;UuUydR=^%_6K4gk)s#ZS|HFki}>%)ppUmBcsf*WLx)BILsUhto3JTC}@d5*kP
z#MX)lV(a_1V;sFs57*U)qa#MpkIAs{xMNm^NoEuH*^1B83Lb3+8rqJXJG(mCL9$Pu
z_Z<yl8=`gZfD4#Duc*m~p1s93I#`T`#Rr0D_B~m#7d1vIC_Ak~p284W@>6;OwpuzQ
zyhwyh1Es8Cf~lA_INrQ2W({dzF&)?U9W(_aK~XXLK^$h>6!jYJ3_$Od=vJ!ZRu{G8
za=PvD8?IPw1kRG2IMHkfy`m3bRmK7mE`s4AUv8req!F#ekWocL8Zht=gPs2jR$3y!
zP8=G|^~CSVNh=G(js(aMZ@+ItOdRzJC*Xj!mu)vO=W7>>tw1UNo!|@Uv<j|nzw==4
zq$=tMVdbAGYwEF7rbi+{9qbUGeQXOdWa(b4j8wd)B|Btdcq6pwSF#=&Ft?qWTBKd_
z>e1MV=~5`~W{Pm*pC+2@pGb(!YEYx3{_8K1OO>WC?Y#>`?xwR&8|7|paQWb7tA|cW
z6Cbbn<I$#3t(q%^gdBF7Fsbj*qsl$bEI2(T{*98=xZj2z|LsiEgq;&-m`_~QIzDp5
z$kbN1s?r~=`266jvHJNJAHG!IaKGrV@8jM}<tgrJ?K<h|>WYN4RCK*N9jsF_lpRd2
z3(G37Ev^@*(G(Uwo4gQV&@xqVA+BA3Jmt}&*7wJehb<R+FkICRbz6(1sHmj^QA5>8
zOv^DC+!brq4p0ePfeKQE6?pf_hhCJAYmfd17`o~#2Xb@&P@LlNvm^GuT8@q+=QsPE
zhdpGv<Ps+&NWKI@UN)wSw#79yJ_Cg<ioHz%0Zs83_UJ*`LFJQ96CsqI8O3RA$Pu*#
z!Drs5(hl57_(|HnX2ZTHumz5OevO7>Ac%U5csi0@yQzWY^yw-ja1o`jNnY*!7)=Cx
z|8XO>sI+J{xN;U#BplhjSz9q`RlETa(zaVfD+=!Sb}wGNVz%w!%}`VTUg_uz(%ne`
zhc&aWjs$_58W=2jah2vZmN_&%0Kq!ALF`^+MImTXa-a(kN%sUaO*9Bb6W=ex)iF_#
z)cZ`FKAm=dnWw4pq173U*1*bez%|3oHb#P7GV7r$kSflSAlr!mD~5q08nkQgkvW8e
z4Mj6*T!RDhsO$<|`e4RHTidrjXJvp&D>5RY18AB1UBLI8L4JbXvU?NFUshlFSrB54
z<&?V)?jPzyDi8ruOAd|O%iDtLkkO+@bBbCZ3WlPGF;6X$2a9&(?#gZ3nj;IeeA+>d
z1eudnor<1-p{J&#q%?zfMAUfA+HzbHC(^^k1Lm~}B8*f7in!}FxUvH1q6SUN)!Vm0
zn*75Fen(_N4ai%L{5KMJJSAMylUFCnr_<NA``$GhuwBfSPDIOMrD*-XjJH0%AAEp>
z1j&9u7<3iv8GED^GeoEYo!mWlg^F#&sYrT5Bc69&p+^BJGJwtr`>X<Zqlh(L!8j?1
z5Qh%y6u}iH5ReS!Vt@PyfS8SQj~P-}^9n(yhK=u*?({*zwJb@@G=AT;i@K1LUY}l*
zd0k!{_p#Q%+bAr61SAk}NzeWMSz8sSvVI;$g3R!4@h3@JKDD2H#7F8S7hV}~(ouol
z=w=VHQB&cF?0{1<8(u0tJT)Lbw59lYNdjwkHVW-w3||H!rlRyJD#Ml@PrIUCb{7Zg
zeP(}*9CgG`_zZHdL^M>x%nr_1n;~7WiJq`$s8~y!mJN1VZ9pT?A-WxDV}SPQDR=9Z
zh2nM{WVw3vYBcYLJv;><>o9CuFqtnTsc?@=WNa)Po&^tBaqtLx9VYw;0*bv3_&{h#
zvtU=6r3*w&j{^<hbQTKL+kiaWkaRa^6LZU3{!RQ0vK%sGi1;Z>IQJYQdUl1hLu<Dh
zR(eqkV;|(&Zo>NY>p8|p<TcQ>obxYm?G$|suU*w^<~dQ@fc~zNg@G-e84!A3dt4GU
zqL(x72=nf}dQo>;WHcE3r*kiNTDNpxv0?@EHEq4WRz|zxk-Tf}HNv;b{B0{54mdVD
zr&a*j*!<vuC0@^50(!9C*Fx7T3U`R{x39dJ?j9WD06`vmb8_H}NblCI8%_R-*j1J+
z&6pvASZ#>>KC`!!0u?=HhXBy21$CSugT1|out9D$18c%ryo9UKK)6<%Mq!|QgqB0J
zBmw^TzAJNNe55hWm&9qXyTZdY!HS1@8sJ2E@!U|XkWC8kR2pkXP($BWC!h1M2$3@I
zG*ovpwJ;DNY5Ikt0zc_p<*C>Xd6r=N5OfIN;|y}xl(uf@R?PdPx)jKtI%Xj;wR<Gh
z6q${|#ig5U?lE!rN@1bn!8#r-xYSlwY9hJ<9Y(-~3m3vJ(u$QTCAO)-t5n1W*P{^;
z&nElinO=VKnSveUmfJ&&!K^EgSSdmtgqe6zzP9Wf=z(Hz>=fX{GLVP6D@8Q^^7z%@
zwuoD_p<tR5nza>6(vJk-b0w#4<RdC7?$OLra3w*|9XD#_M5&kW7|dS2a+@~nkh;p(
zgKi&q8mQ<yO|W-GhL0+SLq$i|E3&sRe7=rTYa@X=p}!Lz>dN7OR7X%8XAjDmA~qy!
z1rrgx>n^2c$(VjEk=<gvOI=v%fJ5x-3YRuKld0|!G890aiMvh-Qdxm!?R00sW&$d5
z45_E}2BP&Vj{@ASTX>$nm1h;8>83~*GkfESK)Oo<r>ynqwyOXba6(rIq)`5Bt$};7
zgP692GDPfB{g*J2aG0^3I~$<Z;;S7uzVIl(T)03;Mf&1$%g0OxMWvK^Q1sF#o-tn+
zH5b7m?Lzj$RQk=|!Ckmh-i1Sfcj=*2Nu7z{$@q$a<Z7mE`{V8Lwm>Q}9D8ip*H3ab
zf-w{U*gNax8_8zpnhhtMSsFs3G51r=&Cmgp8NaRNFtq>DrAsV8MI-}V1qU>b@rLrD
zn^d8}Iao`f$<A}YaKeSE0bA7oFv5n(7r5yzUa5Z|%25JSh@;<TqxzCf{Q76)uRn=2
zO<)*vgTdGEQG>Jvzr6tqiREbMQ6Gv=3K0_#%8_`Rz)UYv5q)lSlcG;{+&+~0(x78}
z#7Jpykq}`mrM;Q{1mV%Q3lrrp;E!aVn(GhM<C?+VUsKtNBUFmMXQ(Td+?=g4sOZXo
zGWU>c{-TzR8a1L*OGPLgaSk6ja;=9o<PP_%8VZ7qH21;RwpZ}zE3nwKN22bRL;4Po
zk{t~)w>HrjXvmnwN)pO6>@uJee(&I_!bKzZNs(EVz|%ZoCqSn(n(2iYgP!XEHx)4s
z$d7VE>0jbNUqn!QajFY*kf+JD-*XT)p;;$Dj3d956kBAVZz`eYpaICZj8DZ`+Ni{}
zbalxv_E;&s`Asyc2Ghin5;D@m@gO&m_7SqgSb#nw0yV<2wgS#JVJ86Q6c!f%xGIii
z#kuP0YC?;HM3M@|W2ZjGX=(?=id2k2YsEk#Xyw;>&oUw=@E9da6!-)pdXcF_iUwpk
z`_7>}rM@J(9eWLn3)0MxnLx9sz<pCRCQq2KEhOj6!o886aTs0GThvCj`*oHn9pSld
z=~DbO9X?zgO<`(kr-~|EQI>#Q8~t4eICfI(TlDDcJ1kt7MLK0IP+THmZOy9-KP4@2
zD~RXOAQ8w<0k~YFGVqRjqcz1Ark9+&SNniF5Xf%gQS>3)rP81%4X|r6Bw2yb5g3AS
zu{&$)Lpz^Ago(s6^K{i@#oKY68!`JG!1C3*?r*DY3!dybtgBd#ii0n}FFAy&3?=D4
zkr;-ciE15(rtp8w-b>nG7oUKv1sx=0f4BFwM=LT=tVN|m`Ti;5G@ZZ!U_jUqW2cGi
ze4?6Zvu&df+H>j#uu(J^ZWhB2btR>dm<YD$3hv^g^0Az?t0#|T<)!U;uLuuIy=>ag
zEzWh0W>nUYqOwqX15Jkl%(uGQqY5n$eX&Bw9H##7X*={6JTS=772s9u4P|BJ&W<eE
zPPao-ub~)}-vGIy8-lIvK0elB`LU*bd)Clo6QV4*krLEXQ%A?N+2l!+TGG^HilyZU
zHh;c;Lh}bEw^3jh<6ravkb^*FMic##L8RDQ`j{8dB?X-<w&pct8J*8mTqks{28lqD
z3&G0;&S4v`{jd4^?Nr>(B9MS#pF2Yyp>X*&NDJEwZeiSp)#HY3XyOO>a$c|vpgpk_
zkE9fdB4^Nx8HdL5Sr!KS5WfKK`hk3w3Zls|`OS3D#xCZ?JTY-PQdQ#rGm%^(K#}-f
zE}1n8l~X9%;{iBYf>H$Yd!FD<X$A%3KXZ0XJ^TuGk#KpYqt9?cu>+##0wV<(?y${a
zG;~0d-5t`d&zzC614X%Ho-B;Kmz}L5wg}nHgos1b8wi8$)l|LH#Pzb6Q)^n7t8dTR
z%IU3~$Hn7<H!wzV#XP_(nSr|v9<FJ}P5~tcxyR=q^1&#nYU^-^Re@M;g=7EFnjDer
zGB~A+*<c^>F9~~}x6ByuGv|uw<$b3CQ!NKxhw*KQv7AoFm0B)E$*b4Ffh&M_;?*9Y
zKR-rK4CR2=xhXwbgWDmdwH7Ad%5kf}ycD1zG5FQ*q+c$y`~(T*q$BsDQAn4lQD_Ld
zRM*0<6{YR)=nP%c<k0nioCV~fe&=y&0p5tSfP8JAp@$W~UO2nUP}klpCeOS1ywJrX
z$bmr?qeyA1$9?(oCH(6f=rXr5TX`Y|WEyi|hSf9%ml#R<;(M&Mbo5&#ukWBm1^MEv
zaR$i#+Ny%gHW0uNl!?3VziVBa#kLcC^Yp-=WJgHm7QCO<T3LD3>zH;<l_j(WqU*6e
zJpv>}qu?bEFe=$WN;s@;w@>vf%fCW^MBmr;0LpNbxnK;oUYwzq{=$RYOCEW#+XEiI
z;(n>wi@ap~x25%z!t<zM#s<)vhW-(SJh)$yBr2icnM>m{Llh>W-2tsU89dZXb=}c4
zO`|%Pm2OI%hDgQ8Y{w0A1+7d{tP~%GJ4z|tXAd%%dHr-!?9L+Nsl;flz2bhB+w&3A
zBp@QiV$8uwR71cw0Jeh2iqL*-a<ESMP&rR0&kbqD)f+b+6;nNfaHRqTh9AWja*)}?
zn^RQTEqRZxBw+y$?%y{<e<hf2O1=g{Sr5Ev<}eL3OOaGubdXI(JkS4-pqJa%uU}tk
z&Ayxk_Eec>ZJDOud79EA@1M?k>Wx%Sb(lwd2?)S3QLVu2&RIyxX5*VC%4qzg53VW+
z%G%1C&BCbo02OichmsFyW@AKx*oBY!o0T4*!CDy_Ex+nLgj<lLP!^IT%ABxdC&CM<
zmP3AP(8cjlZf-v4DO*rGU#U}Wk5UVXjs45(!6>Rxcr-Jq{*}a;@~rWJaM-g@eAtaS
za@YwYdW_NrzGD3TF<^FzCrW~JQ|3*A983xdsG_ByV9Et$szg+mEcNJiWSvv>N0?C4
zyL`J`Nk#0s3O}Q$s9_V0pTh@xGftC%-r3rkDkMk2%x+K1H_mR_O|n2bb_%$}Rtyoq
zJc~`6NDO5Tg@H3IVYHT4{4g3(6t+i9)^O!0a*iZZP*<qL&HaUDLSnRrT|#x<({C8a
z{{3TO#}M8?ueYT;&H>8bRbw!l@hK2pVt}NKBJBd^iTs<8k!bpHp=^WVVH)E7xEzOx
z8)uA%7^BdVj1qB-Ad+FQ^7O-L1Fh-$>qobl4jANlV7y1=qW?8bFIv;6abrraPkrQ$
z#&r@h6_We@la9a_ODMW8tyb&}b@vQC%E#u29#HXT&S14_acBo~xFm<%vN0Il4lBZQ
z)2i#(O&rpuc=wlOP;$;Z_Ukamt!p9oBPbkOxLBX$PZ=}S(lYHoOMyx|By#HV=`qA2
zxF>K1K^m&h_F%e|y)vntq4=a>V$#t>zw=7i-_p7UK<%U>=RmRsR3yw6ks@CkRs%?)
z2-{25fXILsY7VilBZ5VEmypr>MBAa`L*g%T>h`4Sz`<X>M~e!1H&A1S8>CAiUmYQ=
zA3i2;7Th%;=`A@Xs<=d0;Jv|1DN)A~lF5!o!G+4Pm9%0badnj9QMRfgP=gy_a)qpf
zu7zTIOJV>#)0J1unj;Q?ah|$<Yu_MaFwT_@)_H$~9YQM6D2xKja7WfwI+X$B;N){a
zoPQ`DVM7!BRi3!tMe+V#paAMH6~bZkO_r{=6xZ$la`H1?53>^}aX??&NIWQCORCNP
z9aAsD4Z5m-AncAvKk++Hm32*Vc+Du>)bHiK@2s;Vzk?cc>=ad+lANnUPX(k7ws_0`
zfYj6XF1mO`@fclUvFeI@fsM3dr*H%&%bLu;K0yt1!BC%?FEiK>$tQZCCd?r4#;Jrv
z#H!vVF0wEHk4B^v6Y4Yyvy^)e9uQ(ISq2t*s!L4L&X9KaT@+$90WrRzl>3!<qGs@J
z(yo6mJqU8eYRSlGM+4Ib?2AItL3$i8)d6`ldH%P&yy%jA4r|WB3P~G76*b`_28SNz
z_qaBzTT59@b}9ZGiQLFV_KG}4iUBd%#JDs9UNz`pp2L$Afb@BznBtRsvao@Gu&7o<
zmBu|Jw65|g`5VZ6Fb+vKN&jN`L0uF8LQNz2nOdZ)yN{iTQ$J&FTQo~V0H6aYRAktg
z^Ms%$`u&7Zn2Un~E1K`0qn8=`k>I50cYs;m;r<x^jVTVodjshPn-xe_>5pmu#q;XE
zp+6iE+M0#BqBkVg<G4>qR<u<d5i=F1>=j5XAvq05eavhFpq?*6$}_p$pP~_j!ieVH
zAl`|Mfn#O$-tVRt(dsWZ>_(tX$<h;`J#-wA$f~&_Ab+2fpJA5n4U`8!?#5$!M&)>R
zU@*R~tKMgDD+lOYthC2<oDwY`77eP-L;AT`T5<v>LG5ULVQ>)sHcGOs90rc5Ig#VT
z!)ljzMj^|IAE;wAe5x1Apks<H^LXZid&II?D-wov_(nlfE8rdCKc&Y(eK2;F<(+`m
zZQi>yGy(QZ^+ezO<l=%6IQ+XylWS|ho7Y6*u{gDbITAHONO#QIUV)#^k95|OE=V+x
zc~XG1c@t{xPLg7UX*1vzCTpD)Pqqu+i?Ai}fCV%iLUj#1pmoC(oyFvh;uWe9?ZPS}
zqmGYfp8)9W)pxSTPHbBhr;!y1$9!rC*}=6(p|tWaTGfr^$LR@#jG)rd5>V~i;}F2%
zG<&XmekjHu?uU*P#t9>0v0A+XGhPKcvvLB|$)K-*FK(N6ht4gc&CfVUgHeya-CtMm
z?IrKI7}bA?1&Q}4rXq2RBon%^WNoDkiQIIm)~QOCv%m^p;F|DmMUI$sCY-D{PIpb6
z36K=MuH_MUrZMs}9X*j;^OY2BMXMmD=9#*0-aI5>$k)dY2bK)<E^PI;f6v%U9MSM^
z(E=|h>C|3)BA#2^=_LTKNaxT%W`9-A(JKBxA|U$nFM4{>F9(9%0cFP;e!!Ozxd32H
z0+f!m#`QQHFX|X_3W|S2T_MLG0M7x=Yw?G3j)L$SyB9!1S4Xcmrk6?Qr@u`rcj%br
zkWZ%36(qD@<Y#;nFsLHCThE?1de8_5z$S(`C*FjzLzN@xz#hI*r=j|`EDqY}qmb+%
zV^=WUJWsA&yEb*~B6TOgqnpG_fMbp=?f$5XsHZK3B9C7HZ7l|{t4)7UPFR7u7c<g5
z_sWMWxaYJHASu6i`LeW}M!Cdgk>As{U9m$Fx2@Tb?22wF4oX_;W9R6wHUL~JI^Y0X
zNU5jkE}|&kg9Tur_7XWPyY$EO9r_F0sKnRdURC{1e7&S@1GSp7D1&=K`4*5m6R-uK
z^4wt%C=HNKlj$zKe~A+nn@AhL{qz3?)_rexKqC>f-cNs6h^6)FLWXZ#`h^{_&DEuC
z9x>+2+hxh^qW5{tw-lujSo`1xDH<T%lDS89EE?ql@(4Ck?!jW16N(je>459VqhV<J
zu$=Qx(kP7RA}V{qVA1G+eYagoKofT06>zx~wE-zE6owfJ6(K)@>Og>iuKpKLz2tyc
zT0u$;@ivWQ#o?n=4eGf)Gi}yf5MD?wA%cws&KYR_bAr9GC&~XBU2j$(Ejs>}wB2|=
z%2>TDK<H%Ektb{aR<q{0CUuD)-$VU3PcF7~oZ;ZG7$n7c!C`Qms|Es&OLpbs|0{Ec
zDsPIMG0geX^s-OQ7O~@L4H-mM<v`qe;ZoL?IoAfP7(Gz7QXivdL5xkJ5yVbYUwUsM
z$FOm*G$l1+vy!%jsbuj2*i@ivt2X1f<JhP;i6m@@BF1KngP;xVnV*_Lqq}N^bUcer
zAzvtfsPe}A^R?7{A#Z|vc}YJ2?KD*|{GOS}WE?Vc=%O$|ev)b>u^io3Ggq_rRc&=R
z#g(F0%|siU(*1P=UmHo|F4*LP!qW~}0kQ>H1&P_htQa*olraWSF9myLdvq%SJ&4W-
zOx=NYD8g#ZMyZv&(xV}BDNB<Ac9H?9m1W|+=qTPx49HC2SO<&U0@UZrA)Lp)wq2I*
zr=7-Y2=pb2$ih~8gKq#~IN3!+L*S>b*xNpL9yL^>KT8^D`8z_|q0Pm4m7i1iL2?bR
zqx)mW01bO%v4<!mtiw(vJBt!o$e1h)o`7@AD}oj=KeC0j5~WjmHQXCrG1RF+_peqV
zdk`ZxF4t@`c{}u`4jC+>xa&ITZR0XoeXvgETqcwM3(j6X9|;5Tf(nf<ak|!LdV#^Z
zH^dkck$ilkvRFQo<lW-y>VU2~O*tKdkHeKn<UjlB%taUD$7a7~7J);QykyF{xEvi?
z2X!V(TCvSCXmJE>?<iiG7Yu99+r{pqm~Va5Ucf(<u=T%&gM=NB3+PhljLX3fL(D*7
zAU^@!3^6sq9#$NmB#TAu#f?pn$?rUcO96m^n>7O1FF8yIMV5cd7jSb=Q4?(Lf>NB1
zo16-vYyh;#7a)cw9gh=9y0zd}MDaWt*A3NYabFXE#p-b@cf?lCleco-E#^QN?_yyI
zi@{fth}(46j(=e6q+wC9Ns$?~Zt_iI8nX=!&B#RpzwTZHCT1nF<yP+WGEpZl;@#gq
z=`O6;n<GIRld%wrMgS>Zj~l)f%6m0z?=gYr2A$jX8{2zMw=h5#rY$}8kRpl{8mt9=
zOamcTj#RX%-|3mWqq`kUL&j%#qj?qh&Zo9Y1K~4!oMN*PMWi!7(2a2!7oi>ItHew&
zofxFQycT$XAkq1LfGD&8m<{7UMeres;m_zAFR+~<^Du`V-0{;FJERrJ8N3WZwl$D2
zH`B9A(J3B*EXOj3HZ#~vcfoMLJ|ZO%4TS=FT38rBfvHZ(8n$~HY97Tp2Hs<w2IqzM
zQcL$vBerZpC*#p?*pG>d+vhYeN%UR8euiNr(xUutpPynnaNZ>diV83kW-b>IFf;m;
zxC(oqROTBi1La0eKzub08lXf)Zn5}A@3Wl?Q7e&J_qDw2_7yzIvJcweNnl@}25<}r
z!cX&@`S1(a$P#*(xezG_<xz4e{Ck$&A((g)XJk(l@cvA>Rw-8*<zvURA9>K-<34Gq
z)adVQ|CSyJK)!Dc<6}wfN+BT$z3zeZVV~;x<-^Oxh;evBu5<D4KzIS5o(f<g9Zkcn
zO%$q#z!BVH<+7!y;?$R+G^}0vD0~ri*GXh%sWZWEVdqT)#Hs9h@^<J%)~>~%PxnMV
zfmB0emV(Aw_TM1*A2>6a3#C@VlZ8GP@wKuqdao2EB;Z_5XoiHRM6wwbidt}0A&JM4
zZ5fAxlO7#k&5sBkV}Ksg5fhXgQX>9B;{p9q4k;H#nY}-~ShL|NZ1Bt8*wq^8EkiPq
zj_cv=Y)~R($4_C<2t{UK18plPG9c}ESaV!aJ`kc6RIR>RbomRWaDV<6woF3sKBLS`
zj1xwZEw$c2z|jC^ls!ydswhgRDE^%zrUaq{Qt)t5kYsRnL3}MOf3vJ4y9J@5J5lDK
zn8hi=XMeBPrkBZEKUQOLLoZ#u6^sLSHpEM>=pZGz<%R~2-G+%6!dV<nq}2W*xiIrM
zaQpu$Y4=!x@9ekxZGW1YNpf&RWA-+IEjB|wT{1+@3Kq**fj@lOLDI#Lc8+4j^zuNr
z{}p|YxFNy_9%t8Vpr1Z0l?qcIlPL>C^8+GIiiX6?^3^DYRBtPo9LG+Pa)neK*r7&8
zFhM~vbRceQWwj^EpioxA;t+QQ>H4?~?j%T=;W9NRGJYO|e(84}6$38ZL+I}d6?Nss
zn~n-$hWBJsnPdoN8KyjXF)Jk(xfkw5;u|kLL~R%K{W^&yr<MaGR79Iyli02<wy3AX
zG7B#t^)an;$cp&;2Vu*CF>vbZuQjkt4IrZ7ER-99hB}mmFQOn$^k5(gG`c+yWU(2M
z?+P8><bb^^GpK>u-zgERFrWToA%@r5%E^aDDYEiF2%6^xd*glN10qCF8)WepBZmG0
ze<kl^w*nBqARqtiH>VJpfi&%^9e^r53aH=F1qfc2CI;Y4u9ORebh>rs-<v%CW)Eu{
zoA%6J@fmkvX_wE*toCHWNJ%^3GJpCx{DTRza$Nsm^--MK)QC<^QY)Em&iFb-TN{EE
zFX!xoZ1Dq?G7qRd&Gli$d!1}Ubz#s0+#uWp|Iy?p1`&YH#q|A*zu3F2CGM9;arkqz
z05;KwSMOo;_;84o{IS0mA;T^)llOP)@5uq!#Q89%^25b*=LcpC?+m1FU=p){Bim4n
zeK;fdi8Ytw+CiTU;HNJKQxCJOp=!Dn{)SOn8uux9agbXoFxFjD&j7f<1r5B%>jHIo
z8eoXgh5^14g&vUTcCi@natEC!L50zl+gGy!MTU&W%QYKNcz5~EjzEJjs~u8!8)>kB
z;1x_eK{(j%33-GF3*S^T7(L>7%D{1LMPsVJ(RB&aAR+ltd*hiX*4WcDoCt(#L5^3*
zH$a}d0n9}bB(qPUya)Jln;y9)_WlkoM(y7}1iM=e&AR~HSafMv1)V(T&8Qtzp4L>H
zB~1@m{2)0}$3F6SwPpjbE_0aQ*}rXUS2<MN-H!DH)3!4FZu<&B+zZb!H`oRpscO`z
zKx31xQ<oGt_^T7U{ZuR|y8YrQqT1`P3?y;pvi43;2`tWA`f#quU*n;Bf)un!OAgmP
z5fH8%o(Q2d>_#`?0+K2P)0p2u7{veM;0XW#w~EscD>NwpME<S~L-j6-M;Rlv$Ij?{
ze)<gp%3iEoQYFtjP&zN%0dI1>e3qtR3=%s0JMK69v<PGIW-h%yb#%RmX=;Fe;I2A|
zUP|2j$l@T7;pk695V&L<RUFtZr@pMz<WYK^{DqqvyDoKR_+W4P%Hga3?Xw;lp#|cA
zUo>N`CgiFde*;|5>o{#+IHvCUj_!A&Z{PAs-C+^!fCetvf`nz{HgNb1tZfuPqVPS%
z^Xn<m&;YE#E_$Mb3F-5NYk00Jisup?eF3=4m>Yxn^x2|Zs_3+WG6O)UvqYI82Y|F?
z(srT2mbqEJx6w8RuQ92ol0KEybbrw<BuN|W&jf?E*n5i0Z$8?vVgvHW9D(}2d~~7y
zoj*fi9`KyiizHENhk1GMu14+#O`81)5SGw2D!Ex7ly^A{CE5O%AXLB5Ho@qdBsvVx
zR_J>V3&5d2xfxVYqQx{-E`*JbCpOWqhNa(D++a{)HBoH|_xiH5y4cu~wouUG(0Psk
znH#(ihhJ1mgGlwi0`pg`;BO5;0J?Gam>PpaS;f@%#9hZc(&2b#Nhk4Q%p~fBZ!!vt
zc64v|I}b3;94WTQLKbQWV*hVhydR8mwY0#``TW|afU|mZF+afmEpW*#h-=Fy*AXHb
zOT;$8hrsUN!9^Hn>}Vj=n#t5%(Ihs_phkD)XAPUB-me<h#sw*LoV6!&e#<qM#+!HT
zGkO_+Fd%AehkN@UL<MK}J8rc6&q=e_9oe#Evg^wpMOp`q=Fe7&oA0u4*y$d*!@74E
z8GI)ChN^iB>%h4i#{9gd^G_p3Ub^)${o~c&+GL-1yY}*vl}k)&l1bIxn^k9RN_2Co
zz9n8Tb3shGf;Lw^hZ@p}VyLt;o}J6}8Zr#f1gJ34VK*`|w!rVuC<4ow*4>E#-?tMz
zv(e+rJCH;C7g*Z527)GE<rUUbK+4Me#;=MrbD^{lV$W?P-K!5=KH)I`x*(`ewbpea
z850Qbfv_t0I@2-mWIf#Z3q#k6o!ZJxRD1?Nh;%RvR0N(%s92TNgjK_}w!^O{lJbYg
zOK$YuKC<(kq!9psMAu3n?xe0vz`cW?%Ek-NM?A;**-EXL^Q$FGme>QRY6?I$zFs4d
zp8l-U@7Lx4dp|rXIYk$J2ShhJC&8@=&KLRQoEAZnd~Yj`GZH;0aOla~RV-HQnVFft
z0v^KsHPUz)S?cikeK`yJ+&Cga{2Bp~2p_z#yWZr4F^<aFUngZTnVlz>yu41Id=SK=
z%v^Xe5jtht9zvgABm~{JzjbCT{v4UzcW;Fu>{6CGU>n|tMK|evIEeB-XyeB134?J1
zX@z(O^=otbAly(qKmWtilbTbJwt&dtkFH#Q_CA5~^T%V!!jlpE5joi)>F?|bRqUP9
zF4G#e6olMg>Wf@0w<dnWwWzlmw6bEXG1DKU@k{$J5}el3<68_w(5bXZTKU=ajv8(F
zX-C9+6?V>Tq=;aueZNg(HL?<FKiAegjdus;z5-Cry&*pdzMjK}h)gAQa~u%VBC(_5
zXxfF%^|O+&#|a0w3>=D7ES1*ovHh!RkW`WvMsf;Zm!Flz2Z?Ak-=)41$9BWjDc*he
z?%gmk9orDeb*`%P9NW@3knsxo9rq68kbw)(y0^aI6j@SI5;Abbk_)Ks=_PjYud~6s
z_X2d6^6M@do@+JaGH@{Em7m{^GhPPu6V2jY2z=jB()->s3)UR0&JTsrcNL}YaLm5U
z6e3h<fRl=BS`#Pa4%GW?j#I!PD(%1naf4b(e&-RboW|5wf<<XSI07#=uDd%ua+|_>
zEPj}@aU*^KU#SI4gp7=g__%i$t6U-J!QRari<gCK!5+?AXt5~!4Uf!mq3bC-!^}(|
zk7z7yt%%=T@G+eGhE77fJLRQx{af}P`jGaMGR0LLcMtF&p}Qc*U6>Dn61usX5T8OL
zYo2r7^vHi6MfM}GSBbgERZJEEyL|tQ`J;ynTez@vl#e>CB~b|Kls*Qqt%*TSd+VWC
zDBU3)d8ZBk<;cfp&fR6xhQwy5!^|x$E&U&G6cqUgC+A%I9hwit+nKdzQl=YPJwabp
z$TY)yz=?ort8>t(hR6|fyG#2Zvk8s{tPGVQR3Bq(4cP(czsYz*!9Je@P*5@LKrdse
z{{upoAa!2>(?bP5BLQ+?G1_)WSszY!q4(0?Bk^WV6z=(!G#|)+jd3V51LRQY_!=^G
zkOdEpilY+)DL!~bqe(LsEiNF7x`JynFR~xppr49kEJ(ORL8aIV^$hds)32X{N8;V_
z#la(a>str=En3=dR?m=NEc&3T#JUFJq}FOes6e40X$<fQRK-cSdC><UvX@PQKC^m!
zljhyTw`*#=ETOR)$L>KN;rxo3Lt7h<?e!3YOXW!Hv}U+g3%2_94O)39j7TUgi=o8(
zsiB~4YGGkPiXJEHKrv?L|A5U~>fd15{j=F<@k4~t?BZTdR0@g```q)7A3vtBzhD-g
zqM;lmtq!BtfEbf~AcrQ4U;^S!(ascI1EEi3qiK6kP*$W!4&ck7kOG}sU8(Z*^*n?V
z={(QEa>!-SFZXy{3P5K+W|Vdv#Ngb)@_TRO^Ld$1TT&i^9ADv~k$s-Y6rTNb{p!_C
z)~@Hf5H$pgw6_(DK>B{M8(9rw(zbFCGyWaI4$s0l%KEI|C@47?_6BeGF~+|)MHupE
zE}za|z)|hlfX_Wo{S!bz-hYV7YuT<xzqcAxm91#h$C$ZL+vbW4MFmm|uvE?ETe8#?
z1MsxAk5pe)IJ~VsrHr6q@6|VkrP0W)L4W_W{}Kz_exC`jg0~DU$D=Sn1RDXO00hlc
zmg^YkHwBeuBRL!>jp^L-4m_<2ZiN=D-2galU`_QaljtxKUg70&bqE?}P32G>D$rjY
zZK1?baU6ST58!NQJ>Q@M*n#u}hGKgu1Y{9rE+|makR95?J&jJ%^0gB`J0~Y6J?Y!<
zHu@M7+f!p@$w_H<!|L{#fPEoHQ;ThQkGjA)cX{fd$AJ1b>eQ)8Sjd*5-wB)=jl_00
zq%hX;X=vLO+_0-zLKaEIyQhlH(Z8F`kjtnpEwXR#K`H{SUvyX^?Re10a(6?t`)#5W
z?X*XNkwNekOMmwkM5AdyNOkoE!}=Ft?o7C~fR-d%wiMQS9YYn%`;12oC0t|q7TP*$
zTIK|}`=hcvUNfNfOL(m+NFdIyP?+S0gUq9jW=4^(vObSS$}ji8AEa(D$dU9%*jt2R
zKPq$9RT~yr^gVh7DidxBM2d5i=)x9}$}7b(wkU3-M(b$M(C(Ye1F0P?%e%M60er&;
z{{6E&!4XP_lNk08{E2{g?B!>Dc>}_fI4&DEKP+74#XE7H!xi-IL@U3)GLdKcW2V6x
z6up_OKf+PhR%=1r9pd?x6W1x`Y>NENV}hJl{e-<lVSEvo3%QhDoWp6@<KbCfc$SLo
zixE9nnZm)5>9o^UjI>A%9pG*c(5h^sN%au<1VrDtrsG6yA=(PaOi;jYAKSG(*10G`
zUcXTjCQN9KQKL?GX)#VY682!$`_D_sr~ryZC%!XITiIY1$ptuJ$T@$O_WLt^!<_m`
zWXx9s=3JJ0o+kB(k43&6oPR}pl(+!gi-)90Yw<g-YdL*jr$b5~43#)9#|5$u(kBz(
z6j4$YJxL_3<2Epd^yIH2_DuQplcbfoyLYzHhv1(bxILaH0DqRH)N5vljgk4Md~XCf
z{WFP3#V=dmK!_d!!}Kj+V~UI%6gt-NY#Gpe!8M`yP$lGG-KL0P*LQMfIK`HZ9fJUW
zE393^rs_ygQCg?jeY<Zr;BOyx*gQ~!sQ1;3jIqWzZ|^9Y;D><-W1~zLL(<~42bcM^
zInJPW1+W)81J`$npCnkU37VSup48hi2rdVfKE`x7)!Op$wJ%JueA;q|w>U+OIMQ*{
z4YVCZ^3wpBu=9?Qog*QQVen;r-MOWoV+<XQiEhwB2rasv@*@QmA@pK7IVS=5@xibp
zkX1WrYp1V4Sz=)-Ws#wE0aV)2g}9WSQ@E~1fCz}au1}HMmyfT~Im`7@W8DRJM7BDb
zyaRai0r_Yy#rc#`nFxdNeX(Xv5BBi+S;~$x{=ED4s;)IP?`oe6;pYuie4dm}+&f_K
z6W=P>*d35ncEW_YyVT>6fq?*hN4mS1f{Cn|VH?-sz~INN?p##vVMmS}asJ>1I8_f~
z!mc;r81~!t-MfoBa0U0v8*n=i=rk~s_Uc3Ga+`u!ZGIAJ7?=uuaJw^fOi5m%0rhF5
zg>ZtRoEjMpvG5pjnH^s;eT+Y2f>alHG(W?~n`%eOzkNSwO;Oi0Ux$LN7_>U4nGk9e
z{`FpTK91yUo+hu)Z;_}xfj>WyF_!P>1@@L-jiu4{`bq?kcyE->+@)i)XH_JUu84vO
zQf!T$fXWRl{}H$M4a&O3@kX39ab(J9H&m8v>e5Vz9P&ARa>NrvmKh~z6DLj5nt(5g
zCAV}ElgWia4}er!svEj>D18jnDj$-Y`1dHmgXB0KJ)MQb^w@|W3yIu_w|R-_<aM#N
ze;aZ&T=ZafkY~f3pc+nq%JN0u*yG)kL8W5>1Ly9hPr4ZH<b7QkSlewmg3vlS2s!@>
zasu2K=HC^CQmC;Ig!;ek8~$Z1jnt{0M&X3b7IX3Id@=+vYUHuu^7j#PHo2#xcuo`6
zits5wWJ=9Stw{F{4ORaWSfKU+`3(gTY5~B!*AoK`bP%u$F7FXCH*{&~htJ>4LDm3m
z!~;a8H>ujmU-l;Z;eW_6FT&4a_>jxo|43;^Cf=&8)U54v5tCsWw-*E>IX`yWgWcMR
z2{KYg=XnNbIAWsTHYCVfjo%!LBM)zNMO?TU^q9DF5%t=#7enX{5(gc5Kj?!`xzn?7
zd3_~1KB1QG`t2A*M8ijH2%J+#Fr7CS=sJF)CnAOwAmq5-EdgVL)|)Ia>cO3T1`A^L
z>JZ=l%3GjFBG`^=PH}viAd_&O*A%-U;}Lx-kdKW|OcTv{fmZ*7Js1`ul^BLMNn~N7
zntL7{N;adeUmv}{j;Zmsv0WW-PU#v#2UV;15db~O+aqU)9tDaIQy$md3988fuRq~q
z0!h;<{T6rXR6e?<&HJ;OmhV6PF<pZgQi~AQa&zw9os?YC(ATq&v}HQUP}*RKGmQ1)
z9GG(u&Wnn)MNbFIH@o`&e)W}*0}L!&Xs(8mYSo9JaOEF9Awmf%zeQ+Bo*xA;w`##a
zaG0_0JRLctkdJ}nvQyu_aXFtC0bQdK6E`|7ZtF@gPi3wC!HE55&Q$20s4@XrC6z+W
zii;f31@^ub&aAIomx<gg4$lj6)P~Lv0i+e)P~QOq25?FUB}Md1!C3;ST~yE!A_40`
z4n0Om@a(fm@EGtG$LZ$A8e^n!Fs2bV`B#p##3pATl{4q#BY#XB04by)=YxRnm!?AQ
z{@d7hWr*CVxSy=69WOFb0UN!wSUx{g)U`$TPQ}FquB<X|tj2wMrQM<n+dcC*;2G>j
zomqm=K}$NirhC!GlJf^TkN|mUfk&IHs;bKVb}F_oZl=R#E~xcVSwADm^=D&M@o2{j
zmGw)WCYzvwdU#;zyo`_*s;Yz6Vu7oZOR(<{{kE#2ga|S@FaT#hJKi}6hp0VSvKXYA
zPLUHTy@hwwIFO5d3x*<4;9E#hD6<$*=-auonlmzEId-yV&oPOGAc1It_<qC<6pYLA
zW3<yp`cVmpcDEufw8H3Rpk?vyNUgSzeMg1n-_Fs0=Kn?d;<BworAQry&$ov@S8ybE
zdnRE5DoH2x;Ss-Tle{Z_Bbo-l9Cy@iPfsvU$i(_ZNigyyngST=Le1AVL_*sfhh$2X
zA_u^BoZ%>iJztBCHbXEX2O&0NPrsw1v@em19R-mD*Ke>~X2eev$2eiA*0??1=mwFV
zi(TGSw8jU)qw5dJKi?qTpg2@%@w1dV@<D2EHMx<R?TVMd+^KhOk<kITZ9VSZ>iTkB
zxB*FtcNv6={l$!dg65OU-q|8BH2i>uK|lU&j-OK{nDYHRF$n_5Qdc@pM#q2&u<k%M
zo^@*+lm-dQ-Z2<H)0JVm{>PZ<eS7C)B8Em!93;Sk%KVfdFou!FZlV=8l?%v;F~2>P
z9?~Ro0JN6jlp;4k9GD7>$m&ybyH*wEI&G1vx=oQ!&sJ8WQc$2siOPQ*Ru5CY%=T0P
zF&QCE7ZnB6uPTOavk06nF3RkvSgsLRuDhc;x1>D>&LST#U^EXXQB$Z5%^|WpH!6<T
z6F3um;}JOEl)nSQH^cqo|G*G5MDq42t$-!BrW`yyenzVC+{LcSic_Up0)}-PD}V3y
z|NYb(9Y#Nn{18+?X6W)*#iya0*_yOBX1$OJ$-h3{<#=x=##?yRPuzy;YO$P5d(ssL
z4i}jRawt|t(bhv>D0(KQ?GI0E{V!>|pDCaNL_yD^r|T=#;?h?^Fly5oUwWqEOY>}_
zg6Ze%iDdYsb5>Wr%Uxk%?J1q+R^+&zli$%qX1SE8$k2|{GONj%ZlFS?BZXP_vmhKe
zC|&cpyXnr2KJ2x4ck;x+^<cVy(9ud93&9REFwRPF@}5NY6Rpi)AW269+hP350B=US
zG*Y8?c+r!j#O8<;@p7av4hjltMK^h^zJ2?amLP69peckCWdWpt$paftu5et#`?zU~
zhGs_%@A5kg{xB1=JS_fl)5OfXYd7ewdfu4xZyPrk1gDnrL2Bxa7rUA54IEc+3}sZn
zqBHml98${pOo)21W-j7fod3+cGY_3U5Ych_$W&|}>nUD<pFJ$()ecKQu1{wb=U3pl
z^VNgR1du2!1JX4$3KE%K1mf<BXM4&XcsSufLyw|ua>VI>q)UAz;5*~I-eFUxG*qlf
zFfM;+2MK+|$1#O%sIlA>Fx@D-zW<-JIybbijsBc@^U?_u7$V%yQ2fpZEbI{lgB)p#
zkPa$KV{8s>_>y|}pfh0_IG&Nq-aYAUC3<voGz|^P0K*NH;z{ln{OT?3jr+q{sLS7e
zHH*0ZUlwF@UXy$RPjZAM2>#4r5Rm^{{v0tb8=~T%d9D10iyjL^ml^1uW2cXki`?Av
z#QPA7K!%Br1E?mR4=CK8Vf(#7^-2yDM`A#h;BrR|eIk0qR&MmE+*lb?`&CyCBhgow
z7c#@bVLg5AvolD~H0B{!kfuCD4PkB9Wh`#v-tz9LL%BXiOrW}f>MachBd9@BIj<hR
zZT2WZ%~3iaW1;|`ZYv@LIBlA?lyN@WL8yS9S<DEy(*RwZb_zrD7<ve8PPq1GU);eV
z$_*#NK?b<n(_EIf{HQp3cfkn(A#D@Z02_rwPbSvGweFPkqfC=rOw{4&0D^dd;3sqn
z?`~4Jsn7)$tv%-^K9Rm)O`x0MBhJAca9G|K4g27xE&}8(;ey7V$Kqzz6E-S9L7eD-
z@jhR(uq&{m0x4rH3wyUp8(=^u`6<@{V6$0O>SE^)(7N`(v1AbRS&rr7?*+x;!II^2
z+h|*8XegPfa~vHHg?R(~DS%R=rJC@*Bbf`(aI%u@)U)SNTxobabk?P4ksjX`*)v3s
zenAjt(ExXXAJ*3cHKDaAeglwC=|NJU<$(tqrg*TA2X(!>gKK>MKn`7ys0&2%rg9tZ
z1{QS$qw!wuNA{yG8m#xls6f)-7%e|n#8mPcd5fSIG5Nzs&v#G6k#vM>GzGzMtaJl;
zm@<+0TU+ptKln^-=_ylrCvdeg?P8G;CnCmxE8$O2hom_u-df3N)dI-UDfdQqYVbjm
z&s=HraB;?jl-Y8ysW<w^^qoddAR2Vj7oLSEz#e?{zJ2=k-RoV8FU;9%WT4|*bd*Zf
zh#AC>SvgqTxNQrE1D1kK(Q<3vLjH)$VryWJ{*h5}*sPx&>k7}#F{grrsSK73>4d(K
zJE1$lIZvV$@?UwtNEh@VsVobBkp3PX8wTiiw1*r`hCn>#KoT@hx~REeQkXD2frYz5
z8(*5{VRrz{0b1GzKtOY3(6I)6rIw)PB;c3S@sqM5ibs(!MV&bDHh%CSe1&v0c?+Il
zwYEdaut2$<5IqTJ1u62+T3f-hzk1csbJrvQFDQD3)N#w5J7*4~@?eiVz&}jn8%Atp
zmw<No@XE@%#q6pK0%&us?XhR;3M-FxO}Zue>P<fxdBtTWcqA+z!A!jP%n#PI^@Xq{
zi|U)c%7jB<!ILKdiV%a=)-^hYiVl4es_Y>i<i8Ng4Xd9fdX{d6DuE~&+F)m{nOtV~
ziQ{NDnpgt4LDFmDkq2Pql6j^Odv<NB&od8(B|EkeZ-0*kAPJJ7N|*h+w@%=81oi0d
zff<4?9(B2#k<xdcyr(@$)FzAbkZ3^1VV=O_&dm~kRYVxMg+Ky^<NLDr4(I>c{7IIU
zZOEa(Ci46??;C&&WpZLx74{8Z{0xXjtI4Ru@Dr8-qxoP4BS_fe<5=4FjJeQ9C!t8|
z&pN`mC79758b6~p2OX`Z3--g_rL|zj&^&`@9s5@W*<#>ut&Y>h-)J}vC3ACn*P$<P
zxl3bB<zLlT<e|%BG*t)WhToaz7rl^=x3(VQ7D?a^DD2Y1j4|tx^*(1J73H2j`b_Zc
zTFlIgq)!aZ5Gm>~ymhHi!z<whu^eu-oO4H&NMmVArbPr6Lp{!8Kk7z^s&H4dXWBly
zJqxyuV8s6Q>1h45Vy}~tAC-TAaTV2Z?88Df<%>^uJdfc+qJ=km8_QrLmYuY6S=_@)
zc#yWD(+K#b>1-Y#da}MiUgto{Szh9@=-xKnFv=MYU29HXLPpa4Z=bv*XYQ^mXOq8`
z(v#`#t-(;DiNiwBl0?Lan9<U2_+z-}iy|dRf(Jr&2TgbOv3_^~G+v>*b_o6lfXtDx
z;<|hEKy5Z~e(|M-a~o^!&_$~>deFpr+9N<Ypw|n4vvt@nVN)Vco`1KluCL`E@Xf^J
z?jZ+YBB_GPH2uOLSjIX#ABhkQOZ)Wc6AAEEGr<)w%IfD0GdE!Bd^)X$F^twSH;1-R
zNQSfkkv?f8lpMDz!P?kxm=o9<VQpea38T9cGUSw24N~xDj9xxz<XdrkT8wRaQWEk%
zFm->b9|or8r{_!7GY|qqYqm>Aea6LDWDc`W58;)dL9YGAZy&n^UAubgR!!zoMBB!i
zL+ardT0C|S_N>`61FH&r`#j+BE(B><NdpSiSmuchMRr6b?7p?JqF3NTtyx>;Q~M5G
zrPa+Th6b_|k>diZ{>F5}J%?U*J_nxxto~CBSecHbORiY)#coG0Hj_Al5@(FFp<reL
z7c{FiryT-ti+H%ovA0zMei`j`If5;o@i&lHc<ooa+A(f|sXv}WR0dN>iZN!tf+Lhu
z(vRM?9P<UA9r`=mhDcZ1OV1Ac5ffwLnse!dO(LcJ*v;eu&N{wr`}P)G8quf|uMbTj
z<`n%u8^TVreJ^bBRK)@7CwD}Go@8JI&pL*?<}5hWh0Gq5AF+blN|zJ%3Q}>__9mtI
zRHyr-ia-~TUI)$?i<>uIUXxx}+%0%{a3+4WQHRT6W#v`*25+wbuHbcD*6Y|L3`af{
z6&w32{4VJ=3#oyYiAnAIr4TXQ0g+;4*V@`ZBN*MEzu8(JSp=Fxq0b{gb<mb`MWZ8f
zZ&-}wJ$eKf<8M7-udx6#Rs*(dX@bo1cz+1^=&4QDVt?Qs=I~38dm9cV2H014khTPY
zz(A8S$L2u5&3juLo?5pbqqFb;>+X+6=-$3}@5*y?o?kt+zLPF7D*q=gZs1sqaCCIs
z+v6&ywxED9ti%@cldC?9W2pK_EtNtyJjBI0OI>f)74*&ICJJo`MGY5vyo)NnC$qGd
z<{JiA!%jN1?I%o~dVk`LHb`EY`XmFi$v(R%Jl3jv&z^5W<~GK9Z3j;hw%VSnN*pwe
zj|6t=)F}#!Mq8DA9q^^DDZbR(o_1eGGc)AGEl<{sV$KEYwqAz9vvub=_K)=)HPJYi
z_IN`#Z3}Rq{kg`Rm%-W+sI~N(;}@eqMA=rh|I+%vojSZ$Xm>1bz8zSyd#BbnyRHKM
z2!R#Ez`*`x%MoEHKVI15@(rEDj8?w@_&v|5N>M6;VKB;R_#TXC?u=Y~#E21TdW*&%
zoP@2@qnlunh{k7f<@(3p@A39<^$*~|!CS;V4x&^U92n2~Hye~|U1!#!`!($SqHA4X
zZt`_veEGE9?1&Q>V3+M$eca`~Nx@I8>tuE<e|`VWxw8fsNCaD+h$4ET1kN}!fm_yP
z#ftL}U5?+wB?h}0<C&-4|M{?4BbCA3<}#1Qwmh@f<RP~;9fHwl?Ef(Y{g7%<QC323
zvb5;c6!dg~Nj0-UNp|sl3w?yaXQ_jxIlqe8Z6-$K7cX9<(kKn(dDHer9*=WiQkc$B
zcOZwZttVKVLn%bM8tsL7)l-Uss|96@It{luoP!w(2+GPh(roLx%_+OI9pQcESzOdi
zc5sp#9SsJpKVZfgw-;SZOiXZbY^?fdg84`b{|N1<MRsEQi=l}8BgadrZu1r`XnJjr
zL6IEab1=*s+HnWY_dV-Vb(Y_2GnK5QFEZuj^h&h6pO%OW33yR$E-Oe+2kHLFiCax@
zl>@iv0ZhC7=>^ZP_Mspas>$DCt%{>?^M88>nO+f^*jNdV+$|(}Q^Wk|aX|v!*uXWY
zDn=s2_TOB`ck&UOR8ZT0+CF0T?AdeX%u)YXfy4`G#M-*g;%|URp)ts7RZ#+9JM$jF
zEt{GePl3E%>_(xAaGnRKx}@?`j8FUX=g(JIBAr`0X(0%;#41P`@PA1EbJ~kSRb{fK
zZZ`K(eIa^gjBQfDN<x#^;=gQe^%{L_z=|uR{Y_~EdQ(KP)|?)S{T&H6IGAyKZacRz
z@;9E%?;$PBrC%89s*?aU`+o`{Gqtv2QSF27AI5hL+8!46ZT}nanc5Cs+rT%eb?CWQ
zJxZ(l#Q~f_#?XwH7p<Y`Q4<O1{G5aq;DW%YyLUjS4UdRr2o#|z9PeX~4RjJq?HEIv
zZyKRn!g*w-Ob%bsgum43DYv@SnOJNTFdPT2%EKM9Dk6b<Ma#h8v)eUpBy(F%87pWr
z^R5SZr+a(OLq=|%v2V~LB!7aI+=zVNg5LYQ+`$n-aG@CL69${%a{*2~JJ8O^!$~BB
zpf9l;n`y82h_iC>!poJnFcQp`VwIHX2#;BAmf!kpK=zDr-Nw*bTTMTX^<SECYCSMH
zdO71{z|FS#gYJCGsGptP16?R-jvlYs5R10yXQ{MLnFt`I4`V%&b_-+j4=VH9CX0xg
zy5M$qDY|X!^stY_91;>46aAHx!H^Wg(uw(J<-O>^UBu+|*#SA1R@a#hj@H)+&o3dH
zW-|}buMPZ=>be{R%P4AR?^P4hw(Z^9jyEVa=Fdl!ypXfAxL!0luoy{XATqujE@IO$
z|GvYIsjPGWMj`qF{O4}3^tYS-77YO4BW^<Od01kd!#x9vN}tja5n+c#OXrrqtn7)(
z!e}m_VF$bvCLI7AobITBfacH4l$}tUM9<beu;&CE3%5ys4~V(jVS?tcvp3qy8Z}mM
z{Qfku=hD*Le&BdC-PKEd#+mnje*SBRr^hGtKt_ux2U<v30*!a=-mQv4)b;{4p7u9A
zu$PxIqO||~G|eay^*Q=yNFdS8d)~=O$hQ8|Z{>-*x9s}*bAJkkdr;`m8INir=XtbE
zqFxJ%)JE9va}maXfL>u&4j=_>R2`EC#6lu3mcN{7$fO_K)kb2{Z*&q5g4d!x;}U|q
z8sMeYmOYW2cpsicu{kcPW<IrlB0eVH4D03F3Z3v3&No!Gv^b-YCOTWx`t^X`$@{Lx
zZIZCTl`HjliGyV7<u~HfTp`pRu5xI7cG{#L>i4s6#iiYw1fz>2tfA$Zb&E?zs@ge_
zw@Qx-DNkO<(-D*|!SriCY4&+MaBrB<viw|?-zvO$H^mFNWO(CEF-B4@Ma1cJL*)#{
zF9SumH2+|;(QN`O&%7V<d?-Q`@@(_z!-uC5pk{^(1nQu->%#2Vj$RcdsvDcO-^{)b
ztbZ+vYR@79ma%k4Q|gIxp!t8frJf=|bi3Z&$w_UIN4<!o_ah8^17c9DFmR!m8E3o_
z^tdz@^~jQ>h-wAJKZxlATd*w#-WlN57lnms^g{=Jzp<}F9XYIOuWQyKOkGYC@m8?C
z*!N@>!5e+``@Q7>mglx1eEu_v23?Ar0g`Dx5|c5;254E4f{YR$s7Nak7@=^8_cGyF
z83HFNkPU}@U5-8o3xXQ=;EHj$<Dx^?Za7pn!a&Gvj7KX{UtRU+DRmz~PvZ4~&VUxR
z8K)uJI?R$-1h}8+Ksc>0Q0>AAP;_beMY-r+$#I;my1oxSc8n}Ma{Txp$DE<g&c)AT
zQ1^dW^V|{{<Yt5d6u9Qtb}I&eD2BG^<XNF^1QofTN@rs0J764gdM_d8++qh>wzGqZ
z5VPpRZ4-Ef=QC+THr5p$7ONU|#v8$<6rF#5bQKa|-^mt3hGi{}YA2MgH%O@Cp}`Tl
zK3{hEINyD$L=X1r(x)}Yr6=(eZv~F^K(*T@&t&)5G&W1Yr0`k1MKEfNIDyBB__S(M
zQ7vA?(sQ$BV~jX2y2`3%I)tWQzn(vS4<E`1&k2$Pyux4)Y73{5pZ&9FsLFwH@GlyZ
zn?ngN_~bi0UEe)PA_=e(y(dl5v%wH23<teTja$cS<O5wH%8KrjGcA4*cGb6Z`oG`S
zKxA6^paN~B*uS)d?qbQH5p*&zcqkT{7K!)+3>zelAgj8U|FAU~ZD5~UMwmuBF!gIK
z#M_$JG@poi0KCUnhnX<p$?SoLfdBLCTIOgRk@Viyyl4PCr=$Gn*emw}$Eh*jig#yW
zk6g&d<^JvYDCXz%>0aIyY!hDG)wsC#0>ZEuP+w_W!B$rxzU`Bu2ASs>18F7!tazNU
zHTcnBF^i|>zT*XmIyt!yS)PH_g*3MJ5Yl*+t*INfK?=;7UZrKjhNDEm$UTMv1pLc(
zpcQ@del!{ZoO*twfm@JHUA9As*UZNb4Q3>N6?NQ=w%`w3;iUkOKRt~60F;8PT-JGR
z{%f(n*UF}oZQvfmF9i9IS8fh)ANFBp!VLq-U+y~gvzc;Ng0`ZLJ%`PBJ8A~}^zn2Z
z-(u_q-T9EQn28lkd{3#as&r~G4Ca2{OBXq_*x6~Qxjv3@*Kfw(U@5wviKx@$mRn!n
z5Dc$;uL%rWwZJCt+oF!gU{tbY&1&B_<1@s^X;xuc4^9xO#zo>?H-9o;B$f}w^4<J?
zFn`Jy1%2SrD0j=;pChGp$CvfE^}z}&iG-qAqqe-et3JP5W@=S@H6P_nZ~E^*haKu&
zkyCau(0lB3H)WAfiO@Qx%4!fwRt)TJ4&Y>{@p|O^XvFvwC7?K49~e<56T#4}$G)TS
zA0K9Di8tAvPUtJaQ7IqIg#diLf?KE>4AKHMHEE=VTN=gc!V&m9lHDF+H;Y|t6tg#N
z9YE?4gBd4pBEKoan$Hy1HFSbhfGuVwdQlfLWI44?vFeWW4SU*rr(XHnJmcR-X<qSt
zz>6hl^F2_t4~t9eUM=D5ng4<_qD%uGt_%R6P(kD6=;8uc-meB7xZj%^yaa*~q{h=1
z0-g}ZLfDq@jYoz~l?s))K)uX);5n`1I}Cs%GxX$z|41?ie#`&|P+28(j%Y#;?Va;0
z+CEmM8C>l6EBaeEN9liks3usPMDPa<rp0hI-VEEfFV1*0`N$O9AB`IMX6c~%MF1~D
zCXIsRzy^0AH<?pwC}@igdxI!Wr3K?R_JMEG)V-hR>PJBwdP==*3l`NBE)uaq%nAEh
zx#(9p!XM%sM;tgY;mf&%h|sKUA9w)wW*<-l3DSF6U%h#AP__5!)fXpCb?SuF+5vKa
zlW0uG$5G<M{~Btn^y6OX-lT5CIMhB57cU}F2ASKCJB8stY#XcjDALO?xU+xhUiypM
zqsK<)5;Jt+fjssv24?F)eViW$p3>nDkSA((WZfKCBL;YwhTH)jDc0=q?noR?qCi2q
zzXhR|0FB)baj!$HrO~OG&!Z3xm35E?_C!-)UH#GwxDr1(uw(=J86yUu8=vXeMeY8s
zI&Pcc2oDm))(&^wx@zGnou^fI7Tz1}Zm;0u@7|)K0lD=8lu;Fy3vF?&uy0m|UR_M@
z!A+?|0RN&V<bS;;MA*cgt&1z#H{gf*N(^z!-qrYIC)O4ner;S^5HqT)YQvwG#m;dI
z88Y$V!T5U`Hm9#P+SK{p&_Aa9tno+bq}I--wKN6<|McZU-;UKMI&{l+7#G;{jB9h7
zKTkBDFvP02>%;j&O2(hwJ*Ur4jvK~?jOsJ$qR*=2YhAa^`F(85KG!}jz7=}FApG&Z
zi|5u?SrlFz9Bo{l_-lOcE?vxU@`JL9cjUS5y6NcB!`f+Jp_}0g&&0X?<C=BJ)LfBd
z3tOLwxHA2}eCE@8+@CtFMXs5B|GjfhTUo0&UOzM+^}ngzvnA(qzj^7a1>b(vsayFM
zm)5_mzCBh%C~<o7fma)1Hnklu?~K#<=N7ErF#>RG#AZgTZrBgv5vzx378U^(`v&F|
zyRQPS{S_!u3{Xip5WK&!OpH0cwn&hzdRsf$UTXgHPs^R6+fDD)Ms>rTDW@_oEZG00
z=b%AJc^^!KZ%;YLX=mS&ci~*rZ65z>X7p~g!@Bi^@;=E=W84hKHrI<q^zAT_|1|v_
z=NnntL9~EVXBtCl(H`o84YPl>hCkPW;O3Ec4#TqGH2Agahwj%2BWU1-$nhyk^V`U4
z*RHK=)L}pWXvlFv@HYEZWx?qM9aWn2_qMgN%6%FB{F%>;UOW5WXrtbAV~J-Xn6dk@
zk}cc#h7A5%)Cm#f$^`?>!K=<a@&0UsvLDMk=by{l+{Iy?lm2K`&6fFAX7*(AXTIs?
zRHT9%SQx2%;f|7!S;TfFnk9!oG;m+&>3IeX#Y+o-%*Ra~TIQ8xb!NeCl;ICwx}ybd
zi3Y{}<mwyANXOUxi9a;!OkJV;06_6C-)h>dZ~SVPPUWh>D+X$qMNO>hA2eI#_|D$?
zx(;sVXO12<s$(x9k9Zi8k9(xrg}9)Ei9tBpn5ol;aIiPjVMV$Rd$g?Hw_B|9vG^Vv
zQI+*XGEdRY`7?(Pj|A|Z0OOIL!RqSVBs%K|us%!BgEP^jo{B53*j?RZ^~}DG8JU2$
zPA_>m8dCu!Y*_d<sdG!`xCMW{`PAS0*x|zwW3?^X_?vB*tsmS=U)S2rP<`CNCW!M>
zmMrPBrcTjRT*Ac@23!WJJN4%7z8h1x7s^*0YJi(qNXmHkxu<&5R>PZ=4F3r$$$vVX
zz{8P~26$!Bm6^sKMH8?YV$H4B?#AoQuq^8S+-*88xOgb$y5E2Bz_VE0Gtp-Jz_N8{
zeEp(q8lEOEsQ$wX%lEIy8Cr1i!-plXwK;O?)bFwRb52gr?Pcv$bLvqk%;bJ*5HY@H
zFEFsqD+cZhaq)pG<b1@obq^Yb+;p6A{6U-aq5HQj{|WE6ZK_xD?EiAW68x{txeKaW
z@b>NdO*>@1&4m{3HNfF4;905d(ooG^I%##>D&A>3aEue62Xeq=*x-$~Uu6N6`-GO|
zB<={z9Udp4v(F-bY4Dr$(%&y1xOqg>y4KO4>N`U#vE=+rH9h-vU?O$JTSM>Opta7q
znL6CKZG6|A!Ts1YxG%;{hAj~>vR^LWGxb&#B6fZGwtD^L*=skg3ELzHerx&9<2k^5
z!D4y3|1{Vp_i$2qfHEvIEPpwQckurhI}^AX^Y;HA`)=%O(HLY+kxB;H_w1pPh?Evm
zLPX41vyIGHl9FvKWos3(XNw4_WGPD|YqpgB@9RG2n9TBf{^#|2=9w3B&VAqC`}@6?
z&-J-J7tr@@Mz7AKerpi)6Sj&XYYM3zD_Fj^MX+pc_O0Jp`Y^+I>Zz$c9P0m)l5-2j
z!YHPW+I#ft-~T$m(xfA)kEWY#w&N{X?j9kC|Kl04UX-C)Hf#2H(W@yNMym=3gObAN
z+W}~FuHef%#;)HWu<Y;T`)jJ}AGfBud9<ljaXIBomG8pKkEedjxnk9C{r5)hs15Fm
zi5GiBP2oveLBTwFM!Hj}{}B_n7CLMI+vE35oAxYPRPs3Ka?t#MiL^12_lC1(&hXxj
zG9szGx{?>svc~cV>!_i2cHumk*9+*KKM$&}?|-hz8N)|j=7E>$yzrjuU!&~OGu!g=
z^1LtS&NeJ{(n3gYN49idCv=$MS7o2;ZQI!3)AEM@zWi>epN@peAgV;+6r%QveR%HP
z{fd=J<=DFEX6O*($rtjeB*)ELjTwf{V`;fNz@mh$m-@d^Wp_Dcf3P=OytT&M?H$^p
z*BFn(W7@d=(bi07^{)11&lnzk9KU9qU5Uy~iwf@-g~2I0+ol3@3BK7o>&C_0$@;D(
zRQaVHTjy=~etCU*Ak!k^!L+~ZgR7kpnXsmD3cE03(#0Bt@2U#ZnA(>XzaU-=_54B~
z$@cPfNJfrwjw79}H_3oMGr-caxq^48YK+9i7tCL<Z*rwo?SilAuSwXBfwlK&UwM1t
z;WBX<24@V7DIirE;L8+usXYM+=OmpeU_sV1;u2KZOkHT=U$c#f50dFp2$6E0t;`t9
z#&+C-ngfKXYWnk$)cthFI#cShKI*kclDB7`Te{4(qD(e=*s$N=>F*P1Uwd){eyt~k
zm%z!D><&afbLicBF?G)hZK~4Da5&X<aI;E@m+Lg>Hp@(BMqEbwz&ZXut853!K|!jd
zo@WM=?mAL{99LS-p5DWa4a|r4FQ>9Fefy=b;?gsvtN)VxY~=9a!vn`3IB;M?N=~yD
zs*}AptY7bSD!c~kbu6~DDrpDB3V;uDkIY_IA^}&EIO3D*l|52h<(Il5b?^92(;xUz
zu@5DRG4PE~b-hhF;(b7H!LJ!-Re9!)t#LTD{T4P?_HWX$U$N59$whyy&E~xkYJckC
z^b3Y7cU@~Zp3~Sopyjrr=%cpb-tmoA^?Cg6b4&|W(}?M2`-#10W>0x2&qZ%>FDKu%
ze>;U&tSml`+t=^)Tt{^#f_D!XK4L^9eO&O32LIM?M@A%$Rh%Afu*Y-)?eUw;XYLgp
z(k=|{O=CPfk*E3l!s|{WrAh3=%Zz<Ak+kUCIZ(}Q%idvccn6d6uKf~d7nu1PR$Cej
ztYh8U#;J`hj<D9Sk`~?GAzw4|&n*unmKD%UrK<7Ao`w1C87lQov*M*XS1+NeQAYl0
z{u~C%CKxT(w<$3Y3Y}He>vd#QyjQ^|kH?Q5Wq*t>^4)Na1YjG=CgW%}EL|}q!AKoj
z<V_~gtysDwhI#gJyE4CY|2(hj@4UUEhSlp%4V!+-J^J!9uS@)sMb7oU?CRL~)Vo8M
zTl;+tkBAs4^wQ5=J^asw!er@7H$eSu?}s@a8EdsK^4y`JEt)n0Kq7|8m=)IK)T>R)
z<ze_(Oe4$ON@-hH##tJr7}y;f1EyT6TL?R>@CK>SdoYV<*vjjSPkZ$7OK0EJRH2`P
zrc)?f8exB`<j!5YYA98lknuE9*InP1(Bk(E8(LQ@Z>-yS=JQ%VR62ig;vv};>lQzm
zTq$upq>9ms?^anyLaD9lmidv<0Bbe=B=P3SAx%pto_kjI(rKl1i!Jb(ocFDM`l;ta
z3%w_4=#e~!$4F)}k>JMAX50O@pXXh=c1<c<aRr7K;ZE&@L*<0GWr_Q0!Z)wo>eud2
zzb92Pa@vt&$M*lrpjA!H^(GMMq*L|u^D-~qRt)JT<Kpv#QTF6oY;tn)%tsfhiz~El
z?D$s<w|=68hJNnHtbTj&@k%fC%Bn)5&4Mh$kdqefQ0xcB-Q?3MA1<xW4_Ns4$<gVN
zlP}jHX6Y+dnBv&oXPyNNm`E1J0sdkjCMkKtiM(d&xF~Ngb@?EKscs!!e15~VrvV9B
z!;Bj=J7`99?4G)a2tPgdHW*C#N~)wRj7lWXa@8)!iU2l;s-49{O|uWDWi+Sf8J(Fo
zp|4B~PrQ6NS#n`o@2jEMlbu42e;_k+V(#785wP`!ld<nU5s#6Cwq%Zdp!w8D7~r`t
zy?;=4QD4g1Wqvtjb-B(lyJ^V@D^^s5RZ(+{4rg{U-r38XhYjr7V9E8o*fS5;rsw{V
zdd22s36*bx6y&pI@>=Z6%`V~B_m4eNUe1}t=?)#ecF+yw=gyw1U3F=NCS`Xgr@joF
z!sN`y?|mv%R87AwU7~PS+pfm6y?N45Wf2EEN<1*Y&qkMcST)Yzcto@llnpY$;u9>f
zNZiSBI$6hcRYo3CBOUHlCd5X)J^a4vD|T)A{OUao>owhPrjF}3$KOa!ZfW!4sjb22
zxbbWHY-gvHE~B@Nf{!yuq*2jVBE7WZbE65-`Z`rO>1uL$YGjmIor;UkQXjzAhI|Ph
z)$~)B4%IbVBijv4zcd7TXnGE#-Pt>ZRC~cqiSPmg;ZvyWEf57*k0MZSU9;&HY!99+
zC$AGVrCkUp4JL}YMg6ti?0)clv5nnWd-9J4mFoW7NT!9BZ*7_AU-zmKEE9(k(oyjH
zsSTShlFYtp7gZ;QlU%AV-_2|I-;wgN1O$H~0bZ*X&V5I**ab-ibzo4r1?l)L?spU%
zjU#vV9-kXZf=6-R-=c)q=U3I$W^ei%IBpkSeT=q_O8ON0{f<>kZ`tVdwnsMkzD?>R
zXD$9SV;79)yq!dJ7ajL<-9UrupyS&qoYb&hWEhxu*gCB3>cu^Nd{e#wzN%vOiSHu_
zgFh%0<bXGcYm3W~6M%J6=XtWL>lwmCgxX`}49a%t;>GK@)H}~Sv9yx9(v*Smc&d(%
zli1I{)hUCE9!>o7?A`e_yDft<C&aKHishm%G-yk}VF({U4`|J?(=ScC(1ciED@rLA
z@tPtd=#qc)jvYtQXOIHH4Sk8Uam13lVCOq?!chfKmE9sB#ew**N1^NN#4#|b!dnS$
z%zg6yY&7q6_bzWQ{3N-V65M2$;I709v|pmmJZ3sa^&D8oe<zX68N<z8qcy1qfk7wl
z7aK))ePhlR%iVB0ek@5SrAj)WX)%P48rq<XPXq*`{pb#zw*}zc0ShI=BSBYvV=zuY
zkMAd&#(y3R>2@Dwxa6nhm!3(ZFJ*pV&UhClARk36NFCUkheSnLF(zS{UbUDI*KKd`
z`H$YruC=O9g*KLnmuqj{(B_V7)UIn8?$K0ZdhjrLhx7A&=-e1Y2@>x!_HZu>@lA|n
zXBJjemqPhhgQJ939domFF0<=rQX{$Wct&-3rEAHzMXQ?`!vn>LqCS!av_Qo&AH9U_
zbOkt5K0L}CwF(k=pY$84%_SD4<hbR}&d+nk<yJ3dsM@tChLOCh1G1367nr|4Ltuto
zA@Z6PL=Mp3o*z@?>yGHzw^a`Ivd{eZ@eikO)SEcnJcwv}>bo2etG7PKOP5i_WMhqt
zXTN?>CxL@W(RYX$jj$x_NGJ@|tF|oGo8krl-v$KOl^CpIDQiRp2-PDQopbBvO|NG<
zCV_)_IAIn{yf?l?y^ZQDs^85ydm;}ne$uS+LJNzo!E2P%b5gxNBQExM#~&Y%#Nk!z
zg<lP<KF>C8@c30zm9JfkS~q0s-B({Gq~;;Nk&QxX*=^1Aj-<BPcgyEga2If3ab+*#
zzISEqcra^*4NFUcXD(FD_qZ!t5{Pb8g8fS5%2_3IGe$q=@slUd@pVC6`|4Fo;&nw%
zxz-4L$B7Qf(v?`XkZTo{)U)(2r*>T%IB;O?ii<H<>dm|K%#U#y;fz)j1)E>Xs7|V6
z5~Fwa!UPnR)EQ66{q`M)jxVdqo~&JSd;KOVi%{jCv*x0T$OX4<%)#@#XE#`rpsCG&
ziFIg7MIg8nW_R40qUSjCrRw1J_ke)y=s40srnXu>Od6ET!T<E)cLVFr>l%zr4ZYrC
z;kk>3D{*j}HyBAeykkeTI!UYUJg#jK`USVr5wpe1icp&^7U;2eT!psXH+`SvpL8wb
zbw6@N>G&zl)x=@)&HmlHFBH4ZHt|m3C!Ek6Pm7z?Zdwhn9mwFD)&~WJp&&LDMBN9>
ztaG8BwQ+-6a7lOLp+Qca;Y_Gk_VRa$MVa0WBvB*b*)s{4V$qq0POT=_>ojA>x`2QR
zI9abWY5D*{E1EW<ti~Bk`EO^CUEx)H(Fu*i+UU}F{-+7N>h?<H>i15kv-cqS8_ih;
zqFZ$q?Cq`E_%SnvL`C5Olr>YY@721M%i_XoS1xO_zO8W`L-?Hg_U|7AfTAQ3GrFgq
z{=BnN)ula^3}X37zpg9EE}I?}=Fnz{p;u-~*1{5ZbV}0eG0^^h$GGD=2tFBTIz-5g
zu^(XC-G<2Dael#UNi3b7oNqYoCai`%r%(5w^4xvfP)r>CQ@7D*XxN2juY6eYXASnT
zc`dr7j5#)aJ_+&Vr|&Liy=l>46?wzv4Xw9~k4hR714=3tVeHz2P>2M{w5*s%c%u%k
zPARiL4ytl<LkX47ru(l(o9Q%PR`1U8@~X8h<ey0+iX)UcGh#>D+O|1ks6Ofzt&iCq
zKrm_-NW#p|I(Xz;bi{d)307}mcv2m^)UcU3GqU7D3rTn<SH^Sg9ujPY@>Jd7I%I^#
zu6GPmU85>ps)HZEXaZH<b&tN{_+d;bM#If1v-VFaV`WHC&f3m`w46hkI^g1@CK08l
z{1xU35=`=!<O}rI5Zml$*SCAm1nTP?zXucy{kGDoRoI_>I)aYsSd-43ZKy+QPG2r9
zZhkb9?PZ@EjNd2VBG?bXVfkA1^wdx3%h#MET}g~=7g~foyThiRy|z_x?9b`zfFVPs
zeLPX&<jM0rd#P#+L#9CbG~d0_54gC4s^N2zm1Yt1l}23d^xgL$W!@<Ge7D6WQQO_<
z;=f|Wdh#%O_<wo8yG=8k`Fx~cI>oWS=U)goDH!Y^X~{uBa;Ueri9)-~Xj89p6ux2}
zq1V6IvFjtx+t({QL5!m_#7JGAlSZVT>J+Ecl7G)3<BpY7`B?k<9tRNH&YhnBhE497
zvoW<IKy{gjT>49gfuW1ofkCinTv;RE7EYgh$ujvy$^jcyz)m7+wgE(+M+T{edA}NF
z;TwP*I8{p~I6b=XTV30BWYvlV`&`_Dxf4^H&v>*<r)s%v?`!^loUT9CuipW5q|VGj
zGQ|Gg!Wl%kq!=(y^RyhX1i1LMk3;m4C>a9Y9^SuKz78z$X%es_G2w#$z*>nR^D9An
zeuVYW2LuoAeFD2$jT|}jblu%IZ;oUKtCPr8!JT!g#%8#0N<7vD$F655gygYT_4-5L
zFjQyiWoykZsYJ)(ay<X;zIfmqD!xCNak+Z`#P05u*he5E5gTm|5srxsGEmM_?(Ns&
zpfBYUhFkxkcwXyTZOnA4l8fcd$-3+wJ>m*ak~DC2qb_?zti|@}jMzMQ$kpnx)BH*-
zR`&$+?-hQeNtuj&H$iDlPd5gaetkc_XcRD<>`C*Jjz^r~JRPMsT{0ILN)gmG0W5tN
zefc)hZbGWPJ1@^mazu;Zniy57ot%d&B<bSm6(bxj_He-@EUsj*VIWxtUR9N}lxnCm
zGl_Mr<K}&yF*eMqheU;jA>X1xy$Mv3v&sHjwQT7;YaEsZrKvx$JJ0a+0(kmWMLQ)h
zVnNY-+JmQZhvu_h+U~w4y>Yr-=Z&HD;0ElOeUBbZQf_R`-tcfM;sK6yt@5Sit=f^j
z0d<f`aS2p0bQhlcQB$r;bc_peJ@p;Z@ucluZg6=vgDaEZwK#;w5pq(`>21#+tIFz(
zfah4sh+(M#TFZ)=!_uS`_;8+%*|qClf*_JuwhFDd=MEtcFZpRzRY>1s^hk<VS$ePE
z@!4-ChK-B+Lw(e|PBD8Yoc_1xu)oO`8UM7MY*Dc=L-iKEp=y8f9JQ*EV7Nr#a?p<3
z31G8fgX!-EWEDZ?GY1^2-M(wF+J)0sa8a{nt_W=!Y}l7^pzro!d}C3`>B9hf#Pkq4
zv)sdnud`cFICK7o^dWx*5Y;^EDp}9@@3Wp77g3}%o^}rnnb;0b(x4G#+uYCpgZkSb
z@<>&X@zk3wj>A=yv|R)1Sq5iO8odoYhf#^Pg(6f6vrI~@7JthtidQtKyO2tPQO+>a
zkxit6=H5wC9>F8k{;l4@|L|{*YpbRkkqb&gzP{y_Ww9R^N|J=GcR9vAuh*>ib;`PV
zgnLsd!{a=2Zh*opm0N!C!nI_X8kOT|b@CErK&s2hq>sg1@R%)?&p1W93P6_Y?E<@)
z8}u!WJ<ZOcr#QX^<@G0p`Cubnz&AiVJ#uzmaXhl$%EjMd+t7YLa~LT!9DQZod(r(4
zH5HJS9o{Qo^X9#96P%$huzlTU<;t-`leQNXd`XUydXuZ%p!v_3Y+8%$?Y>4MRXD0G
z&$|`9>n4X2rr8#Iezb-~(;CB++Tpj#)+7T<HSRmZJcz{|+_=gar~JDbbni)SFsqzx
zN9p9n&`oLZFw}KU!;t@|b6#$(dN469)^j&QS8i>K6y}r{35yzB^fP0}=lq4^#d)oU
z57SFzN@(;B!MOG)lW!(sw}omq{IH<O5_`jbT5QaC!Oq>gqrXfuQ0W~K%<;qvf_%{i
zyD}4D%2*iCZY0M&auWW1VgFIn{FzEz%tn&*j#IAc=yvfjev(E3>e&wqKK+7pdV6)K
zm{f)En_deX@Xg1bB72M|d=lDdkQ~l@f?(A258ym_T@DWyfrgD5ReJ(adgV?OCe}QS
zC(N-m*Z__VB6gB+A%DM<@m_&X^oMLE3>2`Ia+#%5p^hY)pfpbpe23EX8;KN^pfvOh
zVYSl6<$U}ylNkv)KxOL*j(%p7n6aCnWA~OhdbnlTZTL0Xx$hhETz?1goYbjBQJKk=
zx&{wX(1oG=OQSV?uRE<jnR(Otp}=_Ra{V`)%#gja_7lqf<GudeG69dTK5TwXN0frV
z>1X#u3y^s?DXD5&kzh`hSB7??WF$X2M@YAj^+Ed*>fAEB4|q6H@J$NIq{10HJ?kym
zme6SKAQ6Cy>(;nj(id6Qye<8j-f&D+=g7}*w)AebD&XA2-zL;k*C#f|!@vy3wgV=3
zb>@Oeb+KO=$Q)6J$PFJjV89LBW*(&Vpro=YCAUn=9MhuKM1wL41U>tE2s-rgrAs$v
zop`4!0fJFVSJ;G`f1YzVJoZTgiO;i}6N3mBg$!XQow7Zo-Ea4`>XplaaL%B;atK7;
z1|1wLW5N7>m9zp6*v5etHRb|u0ywLnq?f&)Edn2fJSb@7kyMI?>$h$-pb!*rv6npu
z&7uQJI?E)6kP+A|N&I=eMc4klx=BI;5rF8EtpUI>G5k0U#d!#oUvxF;piYSzp{rSf
z3~BHF{rd;rpJo!+5o6od?%1bAMblK-z6Fr}z6Pri-98H~9xVC?sQcs6GuJc0O}W|4
zU1!B*G-yAmJa19&$AfbGo*>6`gmzkotL-55T?Zta6rEnd&F_EXWck0)aK%WD5tpuB
zUHy4Uju6BI+$Rs~D=wppJ$0-GrtVu1tNW9b2!Lef^Do=4|4C<FCs}UTQ__S<80^<*
z#NY9hXO{O<LUt;#nn|ait0Wu^)?mohvD$F~K(qTR7A^KJ;#vcQn4?k|J4_8ZQCXFA
zxnO?$>4%2h&z`jGg#kV#|2*LcO%AIzW$D){opEAlIQ3M6s^z4=Lkr|R_N1|;Rkv>A
zVFolE`k={!k5;N0qtpq-oC-bls@a=Hy4xQbv$<j&e8sS+q=MhKZoT(1z9oRB-&e^n
z^O{ji+{`qtD>IofaRaToFtLgaE&e|(n!?gZ`G=Pi7EDVTS{_K{+u{fM-csO_NfRU0
z2ZNw9QzM@<QRpUOIic&<*XmKmr#g-<DFt{0u%W^DQ}#nK4<OuB=%1W}xk1@gsuZj0
z8TEiHJsD_QuWX|xC6h+?+CGeVguQ9_9PbcbD5VB$-KwOr^@Ov*rLT?uzMvLNC2%Ut
z(ak1qoAI~l<G#R2eO6QuT(ljPrghw}%4aB{h;(W^3%p;jOJj(idH#B40?vOdguerk
z?iuDXZj9L2q8pgYPx1Hfi*J5f%&d`jkd!7~u4ijd_Z+z~|79NW_bdu)G<ob2TAxZ~
zJ>mb@0#hL<Q7am?^!7{UOACQ^n*vkW&3OHwd)skGM~(+6-*Dpor1s88(ORFF7mJkK
z@RDVG>amv9*8I(wWv(o4%yg>u9nta(&g}OsTdqJkuxRH1VZgUL(W;b0gl=&ZQ+FJ$
zYyPFtCwe%QDpssCiZ1nUm7i`t4;Ge_SmUpvn&_x>t47q2-AtG<{F)Rq2`>ht_7#->
zb;**lfKq#6V#a8oK9C<URX@90|KDgy&|IZhnrn^T!Ic?h`yYk5tFdDy8~>9U719pT
z<z(`G$;shHK~(|L^*24FJel1G3YO!zB{IT(8?FgSZcj*v1=(ryp&{j7&uwjQ@7%`D
zEdD#Jm+Hgi+EgFZl~kqlfIqkBcGBR0${?y+;j<x|VRc@aW6-wk;J&MMhWYrIZ>uu&
ze`lHt0M#NrpIRm|jQB|!8Tr102Zus8Z3hUc$_u5vd${yrql6GfYt}y0lailGm$@;C
zePU4O(EwBpjGe0CCz1yJJ6Q66Cn33WWvBXF!&sV+jEaVHpoYytbpIPR>-THwb2QX$
z!_=wedHmeO>$GZblR8@?iRNR<FFF-s;+D39T;s&$lp=I(#4@xTv8!5(q{JhmZ1Bd@
z-%yNDfJej9B};f$$|?I@Mxh`guLcPSRMr4rJl#mP>QQ=rCx$nqh$NAjlxWjb#VVHn
z6UyneF^oGq!3obn=4{#LF5|zXr9_b33Gk_>+<HGjE#Vpk;j1q{g#|mZd-oF4d{jq0
zqNY~j+#MbN9Z~Mzm?P!=iT(XzURSH`=f9JLo4(fTlG8$zBHU@wp+i7YS+oBzU-uyu
z%D}S}pge~Y^}noHP_SKAk~@;9A*LTfbbVwgTD56&6CTOYyw}bi9*0C`Bby=Mx7N(h
z`>C0}e$6(5)$Q5y+x=J5z@EOVSS*<kF*eE`4_t2vy6|&Oohgv_pLAc8<8)h1p%1pC
z8f!=@@qcW99jQ~nHmH|)F8%5|7uf*;9($CQNT}{6=uCiw5%gJt<}kOFvl|p1_tLy%
znWGb;%p1o?pSGwge8_RHKhL9}Ft~B+R;^;<{Q(R2x|-Mk^=BoVFS+FWaCIf^xc|m5
z!P0yyk^98dgNANA;dE(=Nl|eIojQ$$DxQKQeEBNv|7S6+tY{<ti8YeChf+*?PXm-Q
zog{Lc1px5B@XYHu9^)1=!4Y21h6_<4Uq?tgLWc*hpUT)B(nUPOvORO&NKN9r18_A@
zVLtam8v^x-hgjaE0ts=jbqc-0+zGy_(dGJ0pTXJgs#)Cb|HtA+q&|m*BLyhxY7^%d
z(c`pJbXiJQkd!We8DtX_kFC*dUJti9&Z{c*1&JZ);3{45tdUK}gjZo}ofe(4S9>TP
zF0No&m9L|Vp(g%|)Q79I1iI;#Tg<kJTsI!mzf;_COsrB}s<N)>UvX}Yf3Kq0>(z&Z
zCBvsgM#v~drsHj<XUCm*Rta=V?YG*8h~y3kZbGGOOx<tywTJIMM@)a%16Eu7a7qON
z^(`vnH$Q~lLQv9MhszvD*=KLqBKVr%9UFhx72?~nWlJW*Y;-kAC(nQp2jT~s?jCL9
z*kRfbiMRaxC%=DwOLy}^CTaFkM_^uy{vDH4$QK?_61Biln@L$_<5R=RJ&BYe>MIVS
z>#n~u`OZu{lr6ns-_lC+PBtBb4`IoC1~jb^3$*j}-JW6?@11i--($SO9aCx)Qfv0A
zA!+>JX~*6yk(SquIkEs0b=NNH5@G)rHeK+7Y7y0ytjHB-`@nPwNg0-uWF#Tk+7{Aj
zq&I9zYbanMhoS4UaoS%sH=<bTU0s`3E03PKg>*P=xUb`bs*jVFu~#emk!v&H9<RZQ
zJ==C@r6APsCy4C7Q3TY~UssbrP}2zu#i}(rOUmgp5WK=2VTnRIi6DUpe#MmT@>SXj
z)W#0=`T9NRun=O83n^PZ;tq9{N*fjSKVi0Qc)*LxuNRoOiweHN+yHpUqm)6NIN0~Y
zCi++0eMvj5KoDUoQ0_dcsl__C8dJXPFKsK<MP#9PJr?UmU49HWwZ^(V=L*Eaxg6?T
z2(=UnHPM5C;eVIhLa$lTN^^r=`v&i9aZK`}Bt$=`WYlY*mWutAY6WeLFss=?!c%4L
zqMcaZR(e1VLWUe0^+1F5PA{tQA6V~;#8TqsYE+OdQH5}6v*CnABzWoLs8U3NMfyUo
zZEhUl<2$Apv-V$#Zl%Vmi*C)=BqYB)N{3im#`)1&IcmUwib^ceonQpmKAAlqiEX3D
zosy8*+z<N^A&}WWz=T$`XtzC8GCP{0n;4IQ^{CNY-P&c~Ux*1|IskY{CB^Zg&evU{
zvV253EHsC%`nL*q*$61(Q+e@9g)1oUE5cS&xdZECvyGHHnjkp|N^l{d$xg_*wrblp
zH?I@9vP0rVqn<(t0n-adTA*4RzhB9X74)C19Q2>$p}G|a*;*=q`FFq=E<YX);5R{u
zyM1ob_5L1m$1JBm5U%Shl|kkvX+$ge*HZVO=<T5X{)~OGDpo`oHyb;zgCR6TvPIv!
zVi%b4nR8J;_$DC%TJb4+UO|%aL2W`JhPD*-F}J@~d=LuQg`bUAOu$9BT=gnn!WQh<
zu_JlG{f9lmY8%|i_3jj8$+~?@84D}Q{Un*TVRr44U8UFMA>rhL?76~#RE?WN8=I3}
zS$J&!g95n&nq#z?6`~Z|#(yZGO4{JK4?6td;I89>=e_NJ<((Z=JlLh(;>Ifmq4bP@
z@bzH)j`L2PT)Lv;J`=~*UUq+x)PJ7&eEsaeA@kbQ@h7)zHP*=vx7Kmqr|Gi8>#_e-
z7SV8Ua_4txu!;w+UdzuQHEz}cMb*ys<27ctwlXl-HN9GHF~2<DL}5Q4ee<1tQ1xR8
z(@yxtNh1IRNAHyR4E7Kj<ox{P;2KaOEEdtlwjYVZj>q3#YHfXJ;$lkKyRUhWD1ow#
zEQa?jDwDJe^laoo4GWpNMe#bU;6GqcO5;etnTL99<ULr#OTtgjUe&O1hbER$JLqcq
zyIRR#^7&MBqTq469Oj><$gHq%UkP`R?94zQK_Jt$dN?U5Xu|+)NwN-~459#h2XqzR
z(`B|78;tyx-WYooovF8oBBT}MuPT<(tdRRhTlNWvLY4aMf5p!5DRrA}_V<K<KRr1L
zk$LdepEsS9Ki9L|MZbR8iKGY3wOt{w@~AyTetr-01jv3-NtLY^K%b0~_IKc{;Oq2t
zR2Dt1T0!Cf;ap7(r@^~|B@?odP=Fd=dV=EkO6S1}wIZd+hhBeU=PT1qgJ?<lL#njx
z5OtJoQkwd!nz^nZX=8>BI~9M;W5D3Szb87<lygiNG8+|~&#~_iQgoHy{Bw_IM1Fbi
zWZWwFTHm2a3vswMn|Xw4jT{d6$HL@pg*CVR*!+LS8>*t#EnD`d1Un28IF;czmtRhW
z2z8JAj%Lc`uUViPR|ubGvN`@I)vxQ>sv#eH7t2KM{h3mwO`U-wPA!^HukEEe3Q6vB
z98-{RBJMH;PpImoK3z3Q()`;sv4m<aK;Yl=(aOI{!$6_Jrp?cB71Br-nyz&7UtNTo
zNGWOH192#N3+Zf}Vtshha#3S42$l)IOHko4cDz!rWSJwQeyF5wYdAsjqNc)qDz$RN
zyUfM!{?VPIuszaFk&+%wM}d96=pY`uADQYZlq*-0hV=0KJMdGiMxe6IhC?}$v_e$>
zHoBkz6^1!|r^nnOi;{6j_(%~{D17XbaXxnnV0H%4nQtQ5sDIsdCmuYsQ04CXEBo;L
zo~aJU>&fKG6z&M=l$wjmqH%=KI)DF-dV`wPsml1vRy+0lcfa=d6ayUmN}G}gGb_eY
zW^qoKu<d(-bZ+`?4TW#0(F5M|-|24?>z*i1e}m|4{Cj&uDu&Y^r8dSEi+s=q$dlGY
zTT`DrCtn8iG&7X_klklj+4)DR6eH~vCrXV=J6=iU_mVQQ*D3e4oIu^FQ_89Isypr@
zw;w{Q%gdyAH)O4r!BQK<rUj`FmZ{f5+z}N%b@E9EBS~8SknCAUey0eC+?Qo=*RTNn
zAtt}KHU7h>7|Wy{RP}$bOi0J1P?E2=;qtAQDWIM6x;E}!Vi}=H_8&Ix!!Qt6B59Dn
zfx>VPr|epP9llW8+S<soIIqVkUEHQjM?;nH+Qje-riVxa3frTmnCWiQ6Fa741G|ji
zO@OrC*}3$JzrX*?yzF7ZA@pG`>-u%;dK}DtM{?qScrhNHc8)=Wop}CH0NcVC{WB_x
z;`7a8whs%)u1=TPaCs-#?rAl*Xx$Ekw08A;T$#C14K=A$slONKjtbbN@=?xR+0(A0
zMvU0bLCAq+usrqyQ9%mQMRR7Cl_$*W0H%j70a@kqx1G?|JK!XhX+*0l)H)unq$3vO
zcWv#u1|Qh7Qu+e&o+Xne{dCX_4=?j|)m1KObD$ud@Qs3ERYK0<NeksHN`nuP=S-qS
z?-alW;>Dgc=<h3ECFno!i(*}9A-@0W9>D|A%2ev1h08MBQ?Suq3A(eKAm-#lIW_lP
zJ*4aBUeb#O|CsYNV;qtfrTt^Rvv$jFW9^FhcY%-H$Lh)UR#}V=8+)X=6hmPdtx>Z)
zvVZ)0+L{qbg-+FFl~}q`&Qa2%Kr)!|Fj4@`1?82E-5BXq@$YnY??M$7eAw{_ZPBAa
z9M-@P+?H8VU82Qt=gnb{_t;iVn@K;IZbk{9BfX7MR6Qsd4Z|tr{l0<2w1#SV?RQ_j
z;jH?|SYl{OPT4j7+Q!9m);l!U)9X>Of%y_jqZ)G0LSq{$h%OXgF_2DIA|XZYsoLo0
z7I7I@8<$op)4RXY&e@uq1=D->wiDA;nrn%F=Bi>$U~2wvm8P@Ob@a~|ii2<CKu!s8
zf9%11d!ZPyN(+}RQ=OH*9N|U~R@U*%MMHKby&TmV99=2&iGT8YLL7)LBH+}cpxFX?
zPTU823$0_OZ-Z5PZuJFTRN+l3pnHVRDZMA~-@MR(qaKcfkyu|!G#FNfB6ipPdXH)$
zRG^VUR4Drk{Dv)XR-2H33C>6p*yOmyle4?higN2l8O-*9l=Qqqp`nY)^~zemd2_44
z#M`v2emZo>Z`+~u6QJ8@n8MsdQac)Ec5^VLRZYsDN`~OqXwu&h@T3g#9Ef2;4>PPb
zWgpk<jm~g(k~E(cJ}B4i!4C{|=3shKB!H%W1NF80kBeTdZqRW6=T!Ql(rvTe`@GN3
zIp6{lUUtDqW)VF}BNomP(i>t|;T(4t4p~9>nZxmuG-#|!kVNsl&&&@m#-3gQR8BAE
zw$DOoSyWC^R#nrcf`p7h{+gbXbqQHmiVg7vJ89AH=d{fK-#d9+rJ2e{w@ek?o@D2G
z!-O?<(jq8KD9K35xfrETJwA>opA_;I=*d~ISA9b4W%u2?w#5;~?*68(1ymMqaftII
zVx(6r`Dui8SUw_OreZn)L>s?eV5vSQxKDL;)!*Fm>4r92ih^^CSA_}9vr}d-D{nE{
z?QYC4L8cwTr5cGG?m+uxLUI#tBXwaDfgz8tUbJNOwvZAUP{BUrsug6XQAp}7^=`=k
zTbDeRx^=T=i@Vj5{b{J8CL{b9y7VFcKEKh40j9GWHJa@EYE@B(fZv-{Xwkpl^lig`
zx3YB}H8g2&gh6kc-Y!Fiv~~Ys)WIh1dKH_;T(5X(_}sB;`sUm_*C=H~^ol6!h$fLq
zRyRWzRjO9K+S;|%n#XsWKg2g}YWq@?9Nzkjt6TSvSAn1J9eY~!@z%TVHc@GkX%G(V
zzc=(Uw*ZQ`UFUP>Z3q&{v`HCMPx?A()GInXM8V~nPSoo)6DkXvN~Cc1EwO!FgkVa-
z>dr4^mZiueW}^U#WGJ$8V@0KbKM94fF0-Pk5>za8GP1VnSJUbxY^K035|K}tt9~R_
z1mPeH%zNn>JfcDR6~D47Tbu4V!TVhAS9U@qp?=cZ`lTAE?%2g#ui|%gp4)pFSwTX%
z<$BR8+3CxG&u<2l^mA#Q-HQ~wku*E?ZS3o{uklYP!@(l3DY#SBJUc|j%rjLZ;><@_
zG3R-jgXpb}0ie*2|3H;3ru*!MGVQ*9|BL!P>wJClj%c_#Wo4PS0Gtqf^!8id*4$YM
z3!i(vO1rKSr77C})|!~2C=cAb=(JPyvS=EoRkSX$o!q&|^oWw3s-mv)<2txoN13qd
ze1w=~HTnGcknX?KDu*n(ER-Z@=kcrcWsawas@N3N3o}B<E)yxkmC#ueM4u_RsNiY(
z&R`}vWp`NGw!op@8Hx=PSxVH`%=ME^;6Hqtbf*W>oxA4?{|TL7mjBpA&)C?9N&|hw
zF@C=jEd!?_L|ltOdMi(3bJCWTP?O=(ABe0sRB{6eSsgu@3`!eEUD}&PmZ~L`sG1_u
zuS0j2qKJS;oCu>+H#0^eZZp(evmP#OyqbHB?A^$dzmcg3L&~<92X;`@e)<IFrN|^5
zsxEejDw%bPB6PYqrK-)R1j8j*=R3~u*WUNf<d#%kuE=qSJRqvbg93HC%|G@E+n|d0
zk)0r)sP0IM6o|sN!|G1WiB>l!BEe}4wMhpZ!_nku`pNLKVqiCA0Zy+Xv`V1I*!--A
zFEnsP2E5VyPY<){C=G}u?ltM;I)QG0;DTe&KM@|^`r7T7Otwgny)fs_>26=+zU^f9
zFLDPL`Ur_8Q#yqM8!|g}%Zy7cD{_lYJx`2K8lMiX`hcphsAFaA-4{L}P|B{2`s#S}
znnx8@xx>80?dJU`$YqR(@k2Zla?Z<yJn20LP|?Nf6-?sVKTMF(xy%f>9T2^WjAk<@
z@me=d7w5BuU@{>Adr@nLGqtq-WOO~4>)MJp8d9ZxlH92(sia6<;;+0IzO<lUV36U2
z9y(kWP1jx0anJ%O7swBpe{NsBLI&E3-i@eHr#L7j@1V{p0Y_@nyTEQL2ng=x@}LKy
zrtAo4jL4(V^iUuE_x!ZGG}vU(6K7SAu%}@MY$2bJGELEc#d>G)^8>y-J1qh$fK`Ul
z0vWuLhP37H8IUBZ;`&>r_Yj5ZH;6&B-uVjxmHu|eeR|`bIJbEL-79ffjUu%uswx_b
zPmaEqkk;>yH>8v2(LqR#h(%5hd9zSuZ!e5fD^XbYW7>rxRs&tKotqH7R|glqn8NJ5
ze&<RsYDh+jaV{GqC6#6D+}lk(>wYzIeb~6$$LF^Oj+F<Ah!xRLnbK!sI`!tCci8f^
z2{n?cVx_*K|JV~j>^`V3xx#%zN>0XP(6Y0afL`TSc?+ufZ|E33hk(~c6Pfyp67F<H
zKacl-h<8V3A=X6=v`G|5VE4alXTb=0%_2%H%rOOKF&Dh;9Bpx1Oh$v$aazNb2VGvG
zO43D{;3l(8#~hy2pJQH5Bz}351tSeGh2D2P5=7-whPiRUy2JtLw}K<LZTk^M;d}2c
zG`rqFUUOz9@6F;5+GcV@I`R$7m5zZjd{>m;xULQoK`TnBvUfCEQ1E7D*V+d<2oco#
zP7ksA&yw|-)S|xAC>sn2C$eDE$G1gs^kkw(_S+4XBTeqLLPvDfpdpF+7*-+R`l~%C
zddmn`ie7TzVTe6p3a5#P76Lej#udb-gxRFL4F1`RBb<PxdD)bA>(^#9Gr@SQH6w;}
z>6lx^SjMYL!cGI|UfJCuhLAII!95v#0@vc+@wt5wm!kmO1hu8w5W`pV+(XK?6Lpw%
z7kAW@{o=R(qiDH{n0iF4@5c;NOY>(<7qc!AKb2|_5-^8cRuSN(lsqjtOU8-HP;8Mk
z)tPbOM{#WzE?#_=w7@&qTz)QMC+gb9qB<EvpZ^U<0s~zL8mnk6QLO4_zGzj@iK8u3
zEI5{R2rPpOW?{<X3%Kcyvz4Rt^#Jn~%_spLjgYw#<gXvnFozbMIcnz15e_9)nRF$n
zb9C><(Y#fLW2eprea%`w!V*c~(5mSR6+>@f6UQhoMHrW1Mv<v9*AI_)4w1X-+s2Pa
zC;f4lZESmd>Ah`h-UOFxSMcTFVs<nPO=$5e=zjR{tDL!$bvAeMbR>A+kr*`nbOn#3
z?&VryjY5JjJ3$uw==g%nx-z!QDm{zQxy#Y<>uA}nFb@&+XKTCEQ*2X2B<fy@=vIP`
z=O7Vc3S_#Kvy8<Mxe9rIkI<KVwKhSND%wjZA?lNNZKI`UOJ(6A!vYk?>Iw6bxTUnb
zB`;O~+7=(a1Vdk1xwHcf2O{t(^109lMNo*Xwf6d}aYq~nb0l<!+1RB2IM+}yS57}5
zW->*LQ2oiutP$z+*ol2;f=EX0+Js05ac6l0Q_V2-yx7ov;YVBL9xPh6?A3yzFNcyX
zJ1a|2YPx=zNwFefBvW`@n<GtE_r&YPcVLNTdIo>7C2m^K>Ikl8OoY?00<v}256uq|
zRT+vm_{4a`jE6^^+%0;Fepa12QTaxb%wNcoUoZgbUm<018G9FzDzarVgc!ppN3wML
z*w071)SHj+016u}<jHS{>vtu-C&pmJd>DX>U)9roWiLz=BMg)0a4KpYzL~$)wCp$c
z%mR)@CUd7`Tb?7zZ5WR_E0e|EbU1S0_y=nyVt1=*_Q0>|t-hY7zg1yCgNW02a^_Ng
zR?HH0fL`MPG;yBb>!zzW%q$r+FUsFYg|+xWrFR)&(&IfFjqcxnxjuL=yq^AS5`6c{
z!!mt^cvSIPzx3OyI-Ekqq7i+6gTtlTzcP79H(&p*tBmyZ^;er^E~{rWu@~yP#+$y-
z5}p3AYwqsY5><HrmN*_&$S5eGXQ`yMzl1#Q5nk}D7F7scWDhJAR>GF5oy}x1RfkWB
ziX&-Lzrb_`Zf6ORNo=WkwNefA1|G*z$b8O@u_qW;aK68OP%WZZoA;mQ=0=W|nISGS
z3c1VGGSN_XZ@`nX#_;0J)XOpJMt}Wt74)~P9TtUKm%6PnhAIWk<gl+>$GJR1H3x~b
zA<BD-SR|ud5V#69+0d`4$Y<K^V{>KOS9bDiL)Lk>c?#^RELyc!a_??wS+okQWcEbI
z*slm@odfMRAEcjH#R;iAFR`vFw@gu5N34EFBK$m=^4s}l4(LL}F-#!AP#5!?M!7us
zR#`2TQ!~?hvmMl@7$@ZcJE@hN;$MtHN*qq2`!KN3>l{g;%wF|>@jmi^Z`j4<TZ9h@
zk=c8r9o~<R`J&9sOp(@W8t`oV)^Rb@C$O=`u^3~eWkmP+Y_t;}^%pDVXLXQ4LNa|f
z`3Z6zSv<oii5JREvM4oaoAAb%Nzm1^Elsl$wbx0f@X)3!<F`y>-7VzMo!qS}0VpF7
z9M{uf;f=u?Ef;zL?DxtrVI2as*pij{!b4YCcLy_M62R`nRVa61KDH}o-~wl$hp^!S
zZFBC&FF@FDXYY@&`U*GO+Tc2w+jOF5-bCecNx#`r*?A+OnwtzwCV7|$wl$!X>ca4H
zI@2xkpgs1~xhx{axaqAVcdY3I8Pt_YsbfeWU9roCZSSL)g<smUri0wv+>U`>4i5i;
zk<RBn-sO!nY1i&mhZH6;wXr)E)6>gAb7e`dC|qQr3fxbzi?%o<<;tijFd?V0*D{Ar
zXdeFjqg*xVv`l>={kha)E=B5Ia0!2x0Hr(jc<dAM`PEFYxb}6?kmfdz17=~+C!%BP
zKBEx9o2eh|WbW~a6DKadA9Sz0yr-laO8AO|dizWFWTM-JpVn}W^acib-RK|o_V}^3
z9yf>eJx8dRb5XADSH~Lw<}5MNW)oBRt!7YlmH23(T;7qg<lt})W<;9aB1sS~3bGZK
zF;_G~1Ab{O%C~jun9q?P@>a&Wp`o^An752Y1$UIwg93ZQZnWDG^hR9%$}8~aT4`!Z
zep@v8W*RZfTXBr+`rHo<WKA~7Ccc7PZJ;gN$TiY8aPAApD|8CD#eF!8(^JmY@ny1C
zA_7tGh3B?XfTC3JZaF!DI=w2WPJVD>6U&rk@6@?(A6*nPKSwEMge;~?*hLzyB4TMx
zJxeKTx3a;mvccM;cjYugDq^=j<YvTuv}!!sMvOY3ZcVX&PNT^yMJIaYVVOB8u1I;e
zf*(m3;db;XQEH?6pRdkZuVfT$U0FpBW)JO7q|o1%5mzB{v0T3rRo3<vJ!n)Z2Zz6k
z$yyy2H4E)m6x%z-+JWkXXQ;icmzRE!*YGVkYb8$*!PyZFR2FqEB__&bsgBf7f=yDV
zfdkTY6E>TBR5_KAd^$frKOz=pTEz2m?J=E6GsHF>1{82inRREvQmc%3LK+)i*RU`Q
zF_dddiX5tto-#@)6s>$`nY<>$N)oVE>&vRvE*804iUDhWT5T3O>SHJDbk?pw_Z5?F
zgeiVC>BiOtUq8#t+8a-(Z)Gu`SCWd8-c3fAS=)sKczg2{wODfG^REz2*dy;Yx#-T3
z#hnn~LUHbmlp3frGRxaUlncp5nR?KWdy^5eonIX6DN#0uV9T%E>PeH9%*1m}-la?@
zk=(~5j(OY6k#CquSJN=^$vV%mg?D+uMu=xiK4ar$hAZnoXjA)U5*_xI8@FgJa`uCm
zSGfHJ+_UXqDZEnE9ZS}K>?CSygTrTHYi@P>yi4SU!#kS(9LihuTS6e2Krk$WWSLL0
znGDpqF@m;z@XixCRx1hR^jG-%2d$NxwW`kyd6|MCb2WD)UPW<G1dQo887?;p?RmMx
z>M+@K<>kR2oFCYz9WjXMi8rfcctj}Gf>ZpGDrvw=rnXBEl6Dt0k)38FWwo#w2X*D(
z^y8QWxFEFZ_P#K%VN%lFLy?OIl{FUQWkffUjKfMB@@K&;!h8V$WIj`-<W+!nt;J1C
zcD>Zh6c5kHu&vpKPoGnJmWca9@4!AB<G6d<`~J|oMADCJ)WXX}S1(>1R#6om7Ig}^
zRg|vZ2z01~r~U><TZlX>it2BFEfo`0^6*VTRRoz(^vvnxmhnLtu&!D6?{fvw*sNaH
zZuSHQ%`~dGGa20GLE0tor}ThH?VqJpJbYF(0`Uj+eKYP)KT}2b$yU!#O9`rR@!?^r
z>276`N)rV58=IEi77`M2>*KX+*Y59iaH)i&W#YOhextE*BIevL^}7xvioqA2m$1z2
z#lbTZX)<!2HfE|D@r#Ubn6=S!CYn3TGrZFi>*kt3baHDyuZ$|{d^`>I^1Oqv)L9r(
ze<=<po(QjAT{7SVrsC9L*!n*GS^#HjAItnlk-?VXU!)VYsQRpu`P;ZHUAE+M&Zwy6
zBHD7ERJf7MDwSyxMhQCaE+);$9+YAw;FMo(|Mr`_+{%nqMz~4H>E;ds=z(BQtIF#C
z<jz&utvg}^-=|Ns=;=Av^=^z2-XYy>Xm`(yh@19H7DSVR%1nbST*nGwyt(_x?vJ`4
z?-z&b!En%BiC<+9ybi<>4_-}cL60N^+v1((?2+cYyYTV>CREIOJ|a;zqF>Yz5``H0
z25&qO+W=xRTsiP9_LRKP2mpKPjIg%5YMVs&b#QPX!;Z~3x@X@$mk)QWlQCSiDP!4?
zew#~#;n#KsVyNe2!qdvbq{r@vCJ!P&2-^5iuo6|$2|%rs7ruo>hR6Y<Sus?((1ncp
zuZIXGRJE{qzKr%_!f&AMG<Q{x_Nn>r#@xs~#6x?>wRA=A%cfv1I-+9Pq&rSrk$|Z(
z|B)9Yvk)XJ0Uc|;Jpq<Q@4IA>f-{(&7spzk`nm+AXNgX0qTugt5=SCH^ZlfsKiqNn
ztzXmiN^s1LE<t5*j|^Q3Fky_TuYm|=OM2tF``EyGUoXNIt}a>0!Gsi=4;@p`j16$n
zoovx_?nN&Pgv6F1gx^6TcA0HVC}A2no}myD#tVa}&$w~pj-=w)O#wO&eacUXf9WqU
z;DtMuopWnZ@LGnKYwPxble<hfi5^gPS}6f5mVK3oqT?PrToUMjSxSEb;rr!|9FBEn
zs|4D*_zxL8D7vp_LP}cu`bh_wjdB+LawZ7KsxBYy{FFO=$+Bg#Q<(u+pVrY;grl9^
z<_p~Lr2MH$aQie|fNADql~?*!wB0S~Oj&uDtjeQ}bwod2sq%4qB%ULlz|q+g*_PP?
zwNw<NyPCiwHLpyNYyLp<-v>nIe|`3WtlT-;eRIPk8Bq=T?7=4POI8y7{7szKvl`+Z
z=e^%IesG{Kb|Y=(7>NrG`sT|B8*=7>X?Hn<nVs){wj`u%tp-h1A-|&<VNNkK`3Ztk
zqA0Cg{pw@Vn7*E+s+oWBTk=a)^FJ?cE06BJjl_$+Kh%d&BIgPX&HEiUQJGvWlrg3?
zz1ktyY=0XTy~M~_`^FbeO?$XTbfRE>OIibJtbCGINPD|zEu`JuG-Ydj4*W=ECrj+n
zB(aKxxT1Zp?k6Q3yv1vfqHM%7I|1qnARJC>-+_4mCb~&OQR&$o<ZZpQ(`c4p-F5%9
zLF1T{^V6n5Cb3%w?3@T2bVbRK4BD=-E(IILrxUQHnXU$J-`6%8xGACNbGIUW_Y<F=
zM5M}iKndEJC%Lv}34!}D)$)H4M<D(lN_|fWGy%g~t4#VPX?ocH>OO@0E{GzR{5bT;
zINr{3%2!ySn5acoKaWm5!5b`^1UBisv~mQJ7MYGIMTI3l1ygK{&VSP}TEyCqn7)`4
zfZ-O?h{%uW=aS060C)@qH%))F8E9!~`GNsgy1TaHWBWq6jQ6s(7&k7chRSIc1M7$Y
zy1t)wW@M|KI)M9;iv%c^+?OupsU{nEezZwMXrvD#QMz2mTC3!F_`i@+F0QT^()4X0
zTQaHc#uE{D2Hbcv^Q!lc$W@PMoHTay=qbeV{O4caOmnP#vgimW!8Ik}KG<~gl(fvo
zwG(r%b9gOh8qf~f7YcP2LY#$Hl!ZbKmDJ$F>&5cIU4Q!qlr);_nwByTy@Uo5DawXh
zA=a0VQ+LSBBV{I#n>*c|t2$gY>p6ixAKOH%XkwM_2({zOO<nIKJ+k&Z^V8u`TrqIm
z0ucEv5&U;2t)7t$+2UyHPv+0=%@(Es;edRSdZ3ZQ)Dm+gvZ1sxUJl?`K@*x#_L(-r
zjY?M7i3Of>jc2){6&OE1G5mma<fZ-s61zrvUhSm+w~$_bm9$}cx?A))gcp*ZRPFR=
z?WeSq6T2j;ZA1Ds={I8W<|&fakZ?NPDii?lku}@1bH+O~pg(4_Mgch(!I?KCqqk{G
zki+8#`VP)+447V9Ow0N6ea>M+<m38_I8S1%jV2oyf`_M)qk5rzS!0Vj{Y^_xV6TTA
zUEsOqe*JRBy?S{OuU(9U38vqc?;ad1%I(VY%l?qkvyyG|S5c>EbFPaI?jQICX`!0q
znKvd;)Y{1sCZn)X1B%aJS&hB7K9ClzalnVD1NES{om#eO<2u6)KBuKMzhxgt_g0hD
znCO*<$zuR>CjvhyAJ@flW?;wOi-_ec$n;|6DvRYRkfLjX0u2tQ%Pb=89#y(&-U4!t
z=&DcbJvf|hLvv>_rFCQ!HEP*|0{2lfz#;c#bkVmbNur*y>H0^T1YD$Zw@1}5#Oad1
z?TYPQ9ziM>C4<7j_4e@a*gdZ+k#(Rg6WC`dyppz|F)?Pqc8$nR-MrDN5tVmxM&)B)
zNk&1uA(97=%1t+w)-g$n^#&^WAtq?*j9i4{R*h_RtJRki$gm9#&!rv}^fp!*5ml;z
zc}|j))Iw%fuHU}tRn*m29+q^CD6hNL7u07e{0Qj<R`Q#98ygdn*8v?pd<ACJ?MG>N
zeg_I>ZBI-AuWvK<@Q#b`arHAVJ<ViQUA*=Bu(g5?auExEsC5&;=esFMf0??r)pAk}
zj8+$)R2B`25Mrn;(;w3AG6%LnU)`DO1vxsrkyLClXmxeTFamN8KuGt0^_f1YF7%(=
zeB{%&UeZC*&M`y<GSXe7&(DEdE6DA>mom`LF@-jcE$Vwfb!-hh$3B?3v=?mkEM%ER
zYVV<g46DLV_Y1s?-2`Tm&|p%mEy!moRh&#NS~w-V@Nv5Z%V{8NPrBrv`|)YFqS%X(
zU_&ykE<H>s_RTN)a=$MPzf;10&@nrhy0A`qCR})zBRQROBLV^-T&SEp1oIM9)B|SH
zjzmfV$KU*ZU|ut7zh>IQHE)R&pd|W<h?UR*XLdk2$QwqMm=w(V#$Ak*<0o1B#JS=#
zW!QG}l#e8x-nmMp@?qub)$w%Ge)qjk!|<3W*RcUP0{0@@*jb4KtUMSAb(Yd;`%9}E
zck3#$I2!ZxdK%Gp{5c}64JM;Gah$LbctZ<)dx4ci8J-L!&~^c^Q4@lJ3wDWJtUy=e
zesEHvqbE;HV!y)oaWqc!8n|Ql9w|s4*+e39dAaPlh|ORFl!}#kp$ax2JAx9W*7APR
zm$`emx7~^3FrSCcRP8L4xsaN`%7<@44h4X~$qQDPek=h&q56U@2ii7fX1t(J<K{}W
zi|v2>;(Js~rM`1Fbee_S1_@f{!QoGVf=Fp!HI?wddKBbX$kbET`|><8%~rcBP8Pk7
zFZvSyChmTrf!j-jxaQ46p-sIKzwIk|!IDv8Yw&q&z;jc=0ppXw31s?o`rZVI4)214
zDi<^D+%&Y4lv3b79=rIy9qQB`Wcrf}X3LxqsT*7bvRu$bLezQ;jxhPyl_?^#2@*Fa
zN^LQ>Ym8XrY#k)etBaOP7UyI@U>Gy*dRFZ;Y2WyE^4EMW)FHZGA(4aHv6!Z$d-BKT
z{yQq_btdXCd$Q4Cvn_M;Jns8`?ASRbq@~D+)4tkPn!%fQSgPwSW8>X!9h#OF(en0z
z@%JCCK{m0i&oL@zci0bsiIO79pvPXmzJ*r0o}_C)-|c0Zp=A2)M>DvDhw^~!3wY`%
zLn-0VnT%#`FHzWPIKo;B9F#`er;aiV^e&1Knah4!Ey;t0zWAy`^tGP{A4qwWPc)Fp
zNt^TY>yxG4kcRK{J;p5wWu<QP-5CR^<VIbTE)gASbh>-}X?8Z3vTourXxT<+mVaUj
z!umHMODO5c@Nm-T1mW+vnoy{f96lmJTtd=15CIufNzQP?v95t_J=9sXrrgicg&r&U
zv4~hE_Aur_@+k}b05*X;MGRM=8OR))<{e~uZ<{d(^}3Q|^p2lj2s+S)5@-#gBU1N?
z{LCS?sLXZ>g;0}PD5xzF(P}}`Z=DhP2YX~*0rCc}CeaieXH$605>3E)Z1FX`1u45x
zotNn&#LWrB&=Ya3N{%bZ8>m%%-80qu>c~Fy9Im_tcXDVCMr58P!`y7r83~1YU>-{w
zo~*!w%d^*!X_I5<W=*?Gf35=<i3GnE`aq|XVkJ~sQ}k(#%u(9x`&$cOTLKJ+H<{;b
zj*rHeZei0h-Sb-2)N_)t(sRDDljoBiiMP2s$>Kjj-@e^sZeywCO}bD`zo}f&-5|c;
z?QB3S8E>EX1h95xLsNsTWE<J{j>lyxw?(-si=#4-!dHC1Kq%G%#3*oj%>kqX_p5e(
zQB8tvhN48LH3E!!Ljj|6K8k+Z;siR#Bsf?z>9E_^d`S|aO~`P(Ez9_Ai#Txx*!y0J
zGoZmIbno5)8oy_;yZxNEj|4RG++aJS|I$MHo_xi617+6!h>o2n^|GGq$0R3bnf`K!
z!{b(WoX+%<B_t7N`mOW6m&s(JLix$0e(zmU-hu9E=8$D#6R)<IE66UpbT3(+d_qP@
zWSB|E5|tB<SeEFSC}3}YoOw{cX!nfFx%a==$!Jd6d!iE8Y=HBPJj$eb_?k5o_1#8~
zBM~x{0e@zodD{u$+UDN@$@3ubI?tch6z5Y@V9)HK@?Xh_`aY>d=Y3J&(}XL2`>jTg
zt~+|%%u#((WqvaQ&lJGBe{7RqEt_*zNZ}o5fLdOrQk}Ch)Qbql9}IkDV;OD-#1el8
zb0;8Ox?5d#dsTqkPi$iy8kPFUi4b*6B8ucX-K?Z6Pu=_juI~=|dOvM4E)vg^E#om|
z<j3lt-(L7cv2Al=xOHSJ$pmClst`03=oULS!1Vq+Pci1so|3Cz;cX}I3qfN-JqF(U
zy~?DWQU($*ho<j0RJYG_P8cfHwMi4oKQdTW>i2S#urqxC9&dY7D9MCRuuR6Af^my-
z&uX|VlG$_4`jYtotn~(5+K#%Bd>gQ18OEPH%cK2dC!$?SpZk^>`;gnbIJgFDBckNB
zRJmw#_2C_UeXIy>u&*wW%#AAdfn>B5AVk5+!(<oZ=8?P3>`_8V@qTwvxKSsqXnb?^
zmQly}=2~TP*-c%S2@J(H*qlfZdY)3AF6VqigRyOu9O+3c5!mZG#zOkK9JJ5%X_2E<
zi6mnCdWQCXO)`s$n!7<bfTEoMLbi{9RHf%4A+iiOI}Z-F-oyrN7uXx#+eJZw&)S*v
z0X~m7!=$^bW?7Z*cfe)pty8z5WN(3JvibV@Crr+R|1=03LmQx8KqA**C9LdjZsAmN
z1g>r^fRToAWWG;Z;i4lgRd!Sr)4L&>1EJTNd57mgd>P5UGf_elV!&<km+JDED*K8c
zv{SZt^y!mIM0Z}gwKU(gkNe0fod-?aXc8^uWd>xsNi0`Q_+;|&@)2Ixw_sE%P^{np
zP0|--L#`{Mz)ZF)QxZWi<*)RbXfK_|{sADZ&kWBECUF$s#2IL!$b%UQxIss_d)i$A
zjB%$MPjk&rziLQfXT=JH`WsMG`kNK7k2^xYff}=8S4~7sThgfq<JH9ml|`>!Td1ZR
zL7827SO)FVRHgN#>A#D|CvI44GVAu(Ddz~!CF@gtU97omuU@^F;fF|TZ?<~c3+T9+
ziz*o`E0)ZqFBVSb8-zT~-NNav(7rIgAz;_CY<@#VGR=qVr%x*MyXMp)478WmwF5>(
zqIt6MuV2CUWni-W)$5j;6@T4M!WMCBeNtg(f^;9!+MKGnOtYGK4B1#`Pa!yDfoVwr
zQx$yCPaTVPom@PGVC9z=3n6r6R)-9}u0L5H_=dz#Ue<#S*iHGI+%5C6FgaL`9f=0m
zAZb<qcH1{#2MZy2U?9qbSpqc6iih;q-u5PBj2qC@?9(Ter8|es+ppPWAw9<uPTwW0
zcJ|~vdB_M1cDJ9S=E4>M;@+}J|I=*Ajkb{A9%7Qhw=17VGrcxRc)vh|14d%IM~m??
zzN~rX%_Cv!2xv&PD<Ty&{a(yP`?h2cA8#1HT1J1np+;}aR+K8N>IT(Nd1gPyD)kKY
z431`pt~?BveaO!^1X&dMq>bW5Nj(KrG1*an(8g*+zVerPZ8hhzUk)rgp)5>ZuiI_@
zJkeOuWSmZ>e`DyO{VU6*_mgJ1)>^T}B)=-Y#V%wAC>3okZk@&s(2;Sw$}Z^F{P=O+
zu-}!KMMCP~cZ43wa}N6e{<d6nhPD1UFp$FZ{xOG|xVckI^B^D+Q{uN$yV&O~yMBHo
z719ouXgiePL>!lVA~uO~u6yj`KfdF!-D#q^=Yze4DaVDKJ&Bu);GWXtI8<|*lbQoV
zO2x~~ok7Qk@HNV1+Y+8CKW?)_67<1=b(94C)S@yT6r6-~As?riU=aQ|>88qCy4vNV
z?PA`~!E+irV((AyfUyvGRvwmA2%9uk`@H-5?;LJK2tqtPDDXJF>x_R^09scg&1<^T
z>YEK?#?FtVEXT$m8m!$SR<Jw@nyC3!?C;K&RCbiBkJ4U!AMA>==+aRtZ|CMsAyi6~
z+Rc4V<bDL$Q0l+nI=>C_=`7D>lTf>v@^m(CC;duzFx}H}y)@=$Ve3X|S1nW+jCfsV
zPhlYHcqMh0ZDPMqdnQflck%J6+`zcS_7`--bvSCIX7;M|YRd3d8~4~|ZsGW6O+8i>
z1-(tVrYp6k`Z9%={AB01xPUbpDWpI?2NPw&85YXfQwkL?B9bkYD+g*U=t+}wvR0X#
zU4L@$oF|(kq|he3t^u?vv=qsT9SXmPAW27Au|&;^h4kLShL-`XD-W|G#gQbNW<<n{
zhzmV*$jHrG=BY)xjFzK*@Y|Qq4GJ>vNa;2CiBwOS(#8<UZyw8E=BXUAKeZ?#wp3L!
zy%V1kr%90OS7(A|qqW^d#8G^sQ9G_I7m$z5B)pdgO0v5X$?rBAI}h)3Kb{liH%RL^
z{2MS_dCUC#XZ0sjXRO~YzTmKSqpZvzCKJA($t;SB=cy1*mZwgdsD0{lGC1RU{DPv0
zSlT;fB4lxAd4vAI{&#(H!kmRKCUq+HjZfCzcz8&N9*3$rHq!aXz(3k7rvqGj<IXH&
zZy26wcX=yy0Valt-EVtIV*{;aaBn8;^eZjq(nga|JYepBjoOU>V9uUW^hbD6x$8{r
zUEf;SsEfJG7cEA4P!{XK2!+(4KZGb;NFB`MmK|W>uc6oaSp`taAPF2ca^!lGFA_Hr
zAMj#_K`fQe$@m!dR;J8%3Tz14paW}DHd)9h?P~8!qtd8GHn(*0%!J*hjSgNfhi7mO
zZ*=2{s97>tsmX7OtqBA6Res$%qvh@`HIUo6)<VKC`WCG;r;3T|LdyK1>nA-ocOd*j
z_ve36JdTx`-9oB8@xT3~g?_ZKgg^n4pAb6Mq48HTZB>|dvC|;x_a78!(v}-bdw~Gr
zMBK^%wVCs?bcaPXY&?uAA?J(Q+xOqSLgSe9J*c-Ah?5rB^BPW!WBW4#XwIJ4SPuvh
ziUqY26kk4W&rzpbM&sM=MB%TZm{M$a`B^$*Ld6cQJPbK=H94>NBEP@BO%L<;yNjBR
z)?Jxz!X|crE4JR`<IZ-zEd^UsHc4~sCP`HMB-6!L9%kzF=$|3ne|zFgcH0UF3_BB>
z1zh|Sl%-b_CH_p)#&HqKY-|NOv-7&%Tg|Fbo+-PuoKzTr0|LL?jy_Qv^LRi)K&K8G
zpOa)jk^`Ts3zgX$j@=4gpOHc0M7-L1;Nd$mGwoCAwfTRTq!7Bx*m;4bEdsSJ-dhSE
z>R60S`z%10)^=Z*$f?4(tTp-gOXC7P;DWz>bLOR%yU)TbSG!8l;?`LG3q3$MUuD5K
zunbHGfzR51<f?!iDmN-%%9EvKz$DmCaOdVu%4I}MzOt*4rhIm1zSJ-Y*{NY-F`$Gh
z<}GcD#&}1>QF~ArZQ6Jr=c6^gV;_+{u91?Rn>!6*lC&yz#9M0sbx;4-7Mrw?296d)
zb>|~uaSsqQlzYxSueoRQl`?7^@HE)uV-cFUr$*KB{+bZ4zh&9B?&(95(<&XZA05~F
z&Z($U>ER)|M{`3na^3d$rFnSmt+8%xi(wTy_17yKR-t<h^OdVxubAze5)|c7uGMpI
z*WJ0xN=79I*6H`E*{Oj)zVSZ&x$sTS3$M7aYr|r<c)s;1Ec~2Pc)!EbPWSZ#6Z^pY
z&|wOqW~E!q5RTx_uymHkN#+C1yA0>y!A1{F1h>eDW{fcqMaa|i-OH6NE2vSZ+TL4K
zp9_yi8by}J@phf2POTxn!r{_F^-HRFiBZJW!BXeAnYyLb!eE4G0<fVw($dp^3QTPN
zQi@$vLl*Ii(+71@hsqYGkO5dlnEN==ZR*q&$e%A)|DdY3BSI9yS8Mu`EK~?Q)s-t=
z)n3t3sOounXK5y_8IAJM8bATYMx55%@NX9!k6gcg{kIg8fQx4-)$5B%OW0CM?bU1T
zKxO!8VUOFO>NPlr8?<-Zlq4qV)46u@e~bC@dTBQuv>S@r$bKwyN_*kk0?sCUei|H{
z5_SA|7mA;4HRiIGSZM>Mfd$5p&e=vpM^};qbZWYGg}tezC~Z$M4eQrmPNlQ2NnmaD
z9jPkbf$5YL`*iETyQq~Iq7U=2!r#7)v+Ofu$Pl3Iv!WATA28yG_QpeGmTUrNGAO3=
z!(VGkmqZ;|utk{~t5@a~Wt-JxsAdNM?p(6q1XR;rdZ>+!4T70x$Eh<a<+|Oq`yw<c
zEltTsG7bK@sq~;gt(t}&>hib|mYtftD)o}Jnr^Ipy~FBtn?GS!(Gmh}Wk$d}U{e#?
z4{Ju3Dpd;Pa=>-vjH^N{1U8jcqhnFJy1E|V9I_ECPJO!x0TO$0dv@spt%FFk&@Q^4
zu-<qJK`~!0^zB-sabb2}+1Md^-)`c+_E<tvQqoYlBM(#eyD?7Frqxwe?6GFWdhY~b
z^6~MJq<DEcpu6x^9?j_VP<_u2>cvV&4EWUYIN>%CRePP{uUd?$sy=c<y=a`+@^OP_
zqdAX$qyD9RSJNviO;4oN`W_Mh^>J|6ZqFBM8JAN(Q;PvY6(I6&cfgG|^Wxrd8_gj$
zyJT)|KGDm|*n<*bEsU<5rlCnX>Rk}xL>Ste|7E&Qn$(n-k8`(zWt30fi!{1b!)!Ta
zz7z%r8)1wja22;|k3^Ux$sB)Wd%=atM7*(%W{0d^M47r7`1jj1$M&20?M@cEF!Z(d
zx_<@%NzPd?c#1&6kerukFM3CCVyaZj<G|ZYB^^;#sG@d-@~mxavY1ae<Sr_x6t-7^
z0e!pU3%s4Y_wSp%yz$J7p1<<4A+BrBQm84`DmyoK6{yA8+><9Q_S?n|*RE0z?le<*
zG419(tRw)bvZcR$d7U9NQ)2ilE^)Y9!MYXZAIIg)i@x3Rhuwuzvpf#?I}xX*C&$0{
zInD7OAZAs-!UJw2MvQoYT&nh>D;W}{WA4?$W7hjyqZ4fIy{)4>Yb))uw#RUW?%u6W
zAMM7NjQ9l$aK3#EHlR|je5(F3ZC+MGg*NN)k<)KV6wy?<bb|KMmxPXp$vKcKW-HM2
z+v~X(B6);)=g#laq9P-En+M5z^9#+=Y~*HBz7~A;fZm6db#d{h((kW%LDNnO8D=!2
zRWsco>%j=rQEqyG_NLddk`h*Wn6__UgJku$L`UsXrA(MU-H=*MaLmBn$=B1;S}8Ys
zS$ng?*`0YGKJ;nz`+&PI(;{2LZB=YA4|}X?P*9Kulc&#SG}~ZXgQ}Zi+e5VnubB&C
z`5p-0JdMod4U3+eW|3!CQABkXe0{p0`AebBki6BFLmT=;`>bJ+kxGZk*>NU;t3U|N
zX1~_@qP0s=C`GY}0gmfqM2t9#rU@~E#lG%FC(HHp^oDepEKgs!aN(HP89OyLcJ0yw
zzPT=MB*O9dIBUVjSspTuTiKPK8VkP;@AvN;YxEmX_&%dYkG4yXii*;npghe9vL7Uy
z|C=8?b*j6@Fg}>3fZ4)~syR#glN?wY<<Qso_EqWW*}8Qr?Q#<{y&;1fa!qZo{rZ7V
z@Pryu_8xLq$SDH{`rGpY)>qb8ntB~DQ9A7CId1Xq!={~s{-%As@zcrf5-wcmm-&(c
z_o|P_o_0~L+y2{iKNmdh))>C1on1Q81vK+%t;N?#39EHIP}0_ac`yHVG!1pfm8;g$
zUe)tIca{B>%a3PVzjv=utgV4Z`rW%abd}VSH?&Uc#~dvGdt*7=qtB=@e48<uFJBdX
zedWNuS8jU7x0@FIEn8yeIc{_3&V^c{E0&Cpp`#3S6|%#lNY-N&1%AZ?UDSFYqt*lx
zbhrKybJ_8T^yk%U)JXHdenzytwUacy=}T6yZc*NlBIW2dyQn!0=l2UI3AekQ4pfiM
zhKALsSIKkn@iTfV0|8y{9Q@H^w4GfenL%aGpRaGDT`t0|z&h;oM3=cN_xWPZ&*2h7
zRva+)9<xwlN+6GGBB!vK_7t8!28?$W7USDAla3wFJ$Rr!3M(tnhBjNeSclw|W&lAJ
zl#}nQIr)Z9$enr<Z4BCCX5<mKpy(^5{7ssZu>KeaDJ(4Px66-fL<_cN7az?sS{P~0
zfvU+wDE8NH-~J=z1JequAbXD0oZtazSesjnc%PE_a%G3B%dYvp^i*EaTg@x->-Jy!
z%5R<V&p+!>7@6aio}Rva|CRA0dDa%bTjwjU@WC~&|MGqo7oX+i_)Fn_c^pO9ZeT?%
zWt|)rX`J8dDLB8-=x9?af|=e_TcjOZbB0>9R7&5d#Pq+rimCuTbq30pmuSCSm#Im*
zb;o%PvrE5s?;e!OrP_5``QgI{s04Lu%o616*q4k~N>FBWMzb$IQ3oGWefL(MsS;eB
zmjdVP*<&<Iwc;K%#jJ(p8ue3Nkq6L$Seug@wMU@<7^3#|l$5&vVytRK(wx{E6(}U~
zV;D!c9EO{C)?(kT!oKo{#P%do+N2zaw>4*BXawZr6UsYsXE?%}XG0cPORR8SN3$az
zjDmO5lvUi65(WHJww%^~hGdg*7`nr`KRb>R{p!`LTeN<+p_Y8N(U3FzANYJL{NO~y
z7o|Y2_JgLT8h>PQeJs^roN7Cdj|YnK54_p-gKt0O%1t#_);lFWFf6R?fm|Gl1F5?9
zo{i@t4tO5<o^>&u!DzDR`cgcDk@hTHI3s+If3a77p$Dm<&jv@TF-zEZ$8yHKQqr5h
zzB&TK4RoUavQM8rp<uIzQ}=6Lajl3~%+GrR!MB6QN7yS}!ecTk%1Nr1tliK*W3YNp
zn>I}wScM)qpbZ)--lknSbo=(74&+k&bD&d39>;K(CUjLDI3>jbfumgF=+r2-wQB9)
z6?NVUI(RmNq&9OzaQ=FGF1vD$<^Z2`xk=nUVz=dBiLI!9cvSx32kkm{?=<_h{avqQ
z(m5)Ia*_w^>_yOQe?@zO?Y09HWWRbPM3etV+Lyr9n78jgmd1=_jIzsQ$(}V!iOQNR
z#UQd&mWn7XrbOy6V<;3Mp_DbI5~YRWoFTHbk)%yYDqB*P()PdZ@AEvTZ123k|J%ps
zH@|nx>74U@?(e-^_jO-4pI<GCoGut<%R5B8&%25oDaEBo{Vfy^XVX*qQ8bU=4~@27
zq=s4ku-D*>wu5CU3BGRYF!(-VdKZ*Zn;fIUl$mjEqDjXY5MBp7zbMjr?b=-pfl#9e
zb>vg@APjOKIH>v<12rwbcqPafq;WyKxSY~Fsm|deVld11{Z#5C?MCAoIsQ1Zw<z%D
z&?l}`7f-9(FoM3<thu`;7I+y=JPLs@-?wx3?z^!<D8rGo59ZA2NlAK57xBBcxtzbX
z70o(3Q?9-O$*y3Lwhw4?%3G{fhIgj^SU;UAeX@-Fp&F_92Y$ByL(?taOt13KR6SLx
z?>r`V{oI0cs6&wVgY!foiF2#UQDYvDi?f4iAzL|;vD1onpP|z%I4un;DqN5Z;1Rig
zh<FR-9#i&(2Z>gBeI+GX@J&ULG(-z+;z`2e^^8H@^9U5-g3?**HqKq*kThy5-cOd7
zFXULtQ(-ZL)UzB5Wo$xO*?HkUD9^#)(Zoc*QWPLht=DZjVusILHb(xP=b-!Z?&C+j
zvHk`*KeuijkG3G09)9w7;Hh3!g1OAdx$?C8n;pQLX40qL&_R4m%Rs@qNX&ebY(m<?
zItiSO{K&4tQ$oEz8C$pM<CF~)zijWobLWPTzlx#s0;SU*fBYdnriHW&^DWpqBkewJ
zY3tGh5xgJycU7U(;5;Ruo|Qa*?heNl{~ATfNV&t3Evtaiy&l5kmk&ul<(?vb365<q
zU^*q1Vxb-16_rJ_cvKy8b93Kf**LNW>|G=^FEhMgPYcnse*B(kmHjmitgj2mQR0`v
z>S;KFj_KAuAtKrSDbU%KoUj@PAgaoV6DNf4TXs1zIvTjctQ>$b&)#d%40c()=y3EF
z!lG#Efdb(N3_p4`*wlulzhV$%0|xh%oX}|@;GJE;ur1=-wkO&$p`RI1D0DD6%<c-_
zAWR?0xOj(vGi@L<%^xK{VQ(amqN}1z_{o!Hf+63^#?a%hq_L*xL>((bmjWebjcW=;
zYwSZ&FfUK#`TDVT$U+}4jI=r+9C6_@;51$c%JEA97hQ~s;y>^o@(&C!VK9KH!x<mY
z&DsUIes;r<${fdJ#&DXxkIyAw8sI#8*O&4;V2(O_I5<u5ONSiXdHVF}&dSPSLMMHG
zHt47aes!NxbYD)KJ*&@**k80pFjp9`a6CMGgq_yNk?oU`#4uJi21V8sdOBEpFcO_W
z1zZhDhL}fc`~Wbl*S_0@kr-H6gQltIY|7l$W=9RKtCrSl*)4#*fkNbNMo;z!@!loe
zqVK`{)B@xjAeFTfd~!XCI@m{V1~w==<=By>$hyhD=_JsKJi#K1&qXyr<gb#FGs5{!
zo*Zu-o!e0g#7xnNsH0C;s8sv%Z!DTr!tC#x)Q)cfX54nULG8l!>~9&AtaCJkABd@_
z=tPVU@Q^pN=d(=w)a@nTHL06t0-(ZUhxSgtQUi@Af5xB^WXF6xUP#OOb3$SznpDZF
zSHi<J+Xc6_6GG527#-LJ#xC>d!-Oh<Ye~2&WX*c=k&on9g`--L2t}&`kkT9rN=eK?
z1Yk6FiDuC>9PJW~mmv{~PN;s+EMH-3E1<WM;UrSyo}8Tg89v5s-z!-1Y+ZmPA&aAj
zz~Gdk(fq%clcUC?NwUs%FGHhmf`5w&vg(#z0feH?&?B<-6(5lT?6)uA3zKZ@`MZgc
zqu%^Wb=|vl?dpNqhQ6gFB7|a`9dbMQAqNtY3{9T$taE6TBcO;c{3-UE*jWCL;+1dt
zpmFHKl-=;LV+RWF`qL+<DJT<-prDjS=0O-?r~IS*ltKHDdnrfkLc;YQD!%wdNy)93
z3qKOq0sj^MfyMNq{dwwKV0mGAmZOsssy{6u^z`r9c3hb*u@(SiLW2ZMhF<)8+ViD7
z*K!GxY3#S_TMB_g%&FaMh<SbT>EXN#lWK=q<3CzImI1=b6NI3pyM^Y*F_#@@awZ#7
z7V8}yDHF%f?51GP`~qQWq$Pj@X*`QLC_P{Cu{I|vK2FEknnr>*V8T!Qz{?TBkozBG
z=*#{4tnO1*5Z`hysmIWGsXKOLbs`-m4GjS?)!n-YBw=L@Tc8*;5sy)^k?gDX7oUyC
z<gxyq$bzOIwiGz7TGhWHlj=W_#<vSV<3fl=%b1w8{ENkB<1ua4tXc51MPIt8_CQ{g
zw*TCHdJgtMt359FiKc3X({mTU>q_WsbVf#x6b!0fk0LC+gsU^905Y&siFRn1a|Kr&
z%MLs7=Oj1(KcTeP9RbZfcJJQ3U{mb4cW;gmL0aa}ZVn($C>yh~difi$Bv`NEr6~QV
z9zhDM7>y}Auev%{`AfeEkLC9lvt*A4=(aA$l!UW&Y;5cbtg#{AiJ$go{`u!0kRvu1
zoI}MAl#@A_6*AGmEJs4{oeWB06=XNQLdQ7ICPG>exC0~jlTA%cPo_POe+?;|rx2y<
z#QX9<Sq-`ti~wghhLpEft@;J<80XneeDN(Bc88!^MHV3M;Uv!|e%Hx}W8xDc?J&D*
zKlR%|PbguB_=K9$7%)0?({rBLUjlyGC_is0%Yc+fB}#!kF`#{>5z~u*>z?9UH^&Di
zwl73n`r2&&>#yUNE(xfBWW_5INdq%EE(!;;*C2?JZ^?>V;&HV@z5$5wa0abAv%^@E
zaDi-Fhnh;YX-|2-Qf(-vCSVR>FePuwE5w`dlQD!W5|4}lmX2-ID5v!JG*_}BAJY|^
zT~IX|G{(kebD}?1OiNGi@NJv&42LW3+@*_t4oQZ;J^C?Dk3xKZPJCDQL!d;UJvEu&
zPn*1}<3Ckb(|5v;h6fU(b~Jo)AZF9c#d%$kNRXG(-6Gz<>8V7XMO;f<lbxH}9htol
zeDeFF!}SexjBEL)_=t`J@}!=;dn`1LX@(xfBYTSQ_kZSsDobO}eM;+|sMPgD);W|8
ztNkzHBbwaUw20-2j<O<eTKPY21S0$NuWmNC)&S=C6~h^<sB0;=j%(zr@RsNS+T?U@
z6#De0!CJB1S*o9d>58GK(fY8r?JT}+JWlxrs#tm8TeWJM@RY8<`X)FwTOxl-2MvNw
z0Dk*M;9C|&L%HlJp*w7c9yIef3ZhwcZiZ~?IPl^yNRBTq04-Q(#^n!l1{dimXo(vO
zN>A=H@IZ~zimrl!*zoM!Tkfo~@84H!$fok!+`%J4<xK!4W@9pLqCv^5xv2<m@9GH*
z3(O~f)&&H?1qG(~#%(IG2c#>W_{<!Cw7lM8WY~aNmYHB#vd$xLuc;cDH}F7A?$ynk
zP)iAAec9MHZQJTGj2XBSke4yUTkgN|Xby8qX1ick4c@Ba({UyIHGb+B(DNZ#xtz`A
z`vQng$K0!e*lg|m2a*F4Nw7PJR1p;cEt3}fhRGkqwkH(>)lw`CP!ZwF+0Df#G<+3s
z4KYElMz=@L(lSVJ?O_uD!<y)w2R*2K1GvcS^0pUUo!PF#kfTA$SsZx<G!hy=5v;LC
zTk_N`e%;W+ZUKbY0Nb20!#A0&Du2@us`wIp_p*xb(WB1Di3Hyq=Rwfenl)<*?Jva>
z?s6hLoVAYnE6NYqhc~LQPW5;`jU3eiay~cF9aD57^o+FX*xhX>hQFIJkba=iFq3SA
zxp)xk`w3~7DKz^3A<+pX5?eq$e>@~HZ!A7^pzj86p3txyQwbyr6&E#l@L<}tE1YrN
zHJVvLo--q-Hd|V|*rjBW3L#XS$1dUCgQvUR$w`m({x)pTz?dhxg!6<uq*H|fvM!Jz
zSi_?(pzTg&Mm16RzFfO@q|+!`6%9@wmr6~=mm6QQHz%^Jdn_^ZAHP56#6X)V;yucD
z^TTN91&D3LMWLj<fE8Z+2bT=T;8L}T3y|@dIae@1<ON>DME2IdiEmv`QuUzFsIt5Y
z)dgtPJ@_LVW`V^V{c%#|SwpZM0^B_I7wrcKenV1k{$0zG`9_k!0aOc=oQTtExcS9g
z{z-|kyHbbz6;90A8lpiTsIC<}^i4msA*VVyITb)Ghz#ZAZv*HH1?~_N=9>BNbKt}6
zV$~lo)SDo`@Fp$uB~6t_(d$eLO4S3ZX1YM)r#*XCBKbgOdf<C5YnVlAAR;KKd>DvH
zbB^po|0ITn$|G=D$c{QVTSu?WR74FEAiixXoHDMFeSr;#m2Uy0%w-4NL_Fv)T2u86
z>ymhXpuzkoo7}@OT2+Pno`S`B+zelLb}!*Faf-ix&$a?pcYAtH?Yr;x8st`8Q8m#W
z({&d=?zEAgo7KZ*dros4&jm;z))oEd`04W;j5QCEC~%LBNFF<gR1*HdzYqLVEME(s
z!NzC^q%iUd6egB8q*{HRL0)Nv%kvqg({re2p{J*3k`RX9D=By)Km>cD(R1@1VLqKv
z7~l!OziK7E^$9=~M?uVt&&9^r%nisgSy0?BMkh%eHaGV~{O;{<(9b3SV@D0uVE><x
zyzj=mM&HuM?@o`ofhE$tDYWhHA)08$O5*h`Sa2}o1JR8jZXm$>fhE|AwE<llB`1{p
zZ=<5G%wo?=tQxp5W6%^q+U#w2@a{9LRv7>^!dSRh$KU(Eap?yUai5q?30jZaL7L={
z8QypHaq*FvM64g%9lwA_*<p<|Q7e4qDXYOCVC5c;1@k0~2Qq!X;7VjKvFA|<4v04~
z)2cfQgqp)6nb2Hf?4y#bl(i@@t$SwLoH8{<vHe<hnMNmhD<hR*5Xpp8t^ICj*k}Wl
zRCGdF0cuXWv-23B+<%#9q3fpM&M}z%ukG8Zh@PsaCtOL7iTEpw3#+@wl3oh9+N^=I
zUqd|SF)KkAyqA`?FQd~j62l;3GJi94|JekE2pZRB+$Cyb(NPVPBO`Ik5Ecz+Rw(Q>
zME%=0<Xin$q2vWRqT~cOfr!nb(=Pdq+CBw{uRJW`^O|5yrSOEdzAL-lHi0BQV`yj;
zCCEH_yq$ZC59sHeKPmN<r|~pDj1on7KnkF5-2^Vmm(_@>thn9PPCTgXNhB7*Tvz>5
zox0v(Lk;<Rq&y3XT$^fMS?VBC;BQs{Eel|2_)Rg#lU)BB0bCv#frUc2O6X?jP9R8y
z`e1eoAYh5+6#rWJRk%LT-#`z^0!!gPzRsq>EdguOoltrI;MM&ZcF!BJ%Ot~Z3MqX4
zc{&CkAD_4=Y%FmPb9pvM__7J0qm1u>0n+I?*km^qv|#@6h`v^tX#hbXZVcZffz$_k
zWm)~B*zEj)t<=4)f9%;}ko%&&lV3+IHJKW${<`Iv^;X%>FPcqx{m9P6&3KMq*8HZ!
zCn_HJb5HYB)oHCeHlMySRNe9Aj-hK>*k-S`iZ)j>>wa-?aQ?Yrfm;{4|DgJ(s-<+j
z<7Mmjnk~1e7GKt`?^k>%ui|y0pB86};e-d1kz*B<g50?SG%RC!uD9gZ_KF2K`9$Q6
zL55<A$vp%T0r8c!`-o}YaoDqaDL$8Uc8Os_p}}6fd%9sH&OCC|s41*&KuZ`4(tGjI
z(3Yr|H&AbeXFYuIfHZ*^*izVEyaFazJXlF`UEx{D5Qu@wtx)mX#?u%B4-S5W_}^d>
z>mc{qW4n0B?=B;4>5Qq6f^)&agRmwKfS}M12+V+gx^Zn=D<x=|I>VJQbi7c0^Gs+i
z5K<c+>&N=X<c6EiK(QtCw*oVu@}Y-IQc#K-)k{5P9E(Y}>_kJ>i-(-h3iKF_t*Iay
zg+f`l@EenW!#~Zdd!47~M2vor5?KZ=#sg)|Q>RXmIA#1&<PS6(F2@UW!_L6eP)x?>
z0*rf#{GH`WVocJo2UG@)J`IfPDn?`HM|s{EoCUm?2`L-g40C-S9G^>C%Fv`?uiPOv
zwj{mRVibhL!$P7+t_h<hOZX%y|2%cn$Vh{&k=;Y@mbv*J@(}8|wVO>$&#TCREF(ih
zLUsXY6C9&*NBG*eIG=u{w2w0Z>(9xja&=y!L!Lu4eaL1_&w=jDP<lAAu>vD3e-BL~
zQtB1I6rfF3)*^bYl6sLdlz#?;-3!4qvWU!uP&?YTlZ6kw2h`cgNWKe*XH80sEak7!
zkw2OwYM0WCpiwrtTmXQ=qtV_=6g<e9_{3Dd3(`SMu-k5ica`PIkN9tZq$@DK;vS2~
zK=Y2Co<9qY(F^0_LoUYxj6@i)gT^MyC2QJ<R(t|Wo^{YeXwR@$Gzf`QDFh&GACmN=
zC4cd`B&!GFEqg$fv7%S?ybLvr1hC!186w$WPvKZ4{7*<*UG(R)f=|#Sg077S?Ug9u
zOldo(dn^DDe5?ytPb7TY$uu5<hXn)_#Z4aQEbQTE%WrJJ9)lp7N+vzpm|V<Yz*2>J
z%;hAgRtae8D>>oI0A}J}>MiE%0W{d*J8F%QQ8tMxfXX4B@B_@xbVKZMj|ERP0o>Z2
zlvR}N6~E@E{bWrv#u*Yt`WX$@RI{6C4sX$11-K1%InnPG$ea}AS6veT-FF;Ry*esc
zlmYrV00QDcfBy2NH1&##C?tx+L%}T(poN+MTFUtr==NwHR6HlG+sG_2M7*%?^hqEV
z!!JXEUz21s;;|3_2<!44v(CgQ8$<GF3o=&CEf4uA&8H#$sk^7=!2i*p^ZNDcg}2qT
zf$YKsu!iGvF>v|15)cDR3-*i>tx0<jZ%zBt-OgpfH#M|ObbrpivdM{f40dBSl;wkN
z&s=fkLirhABvS(UJmPaxk|>j)E7+ojIk_0cqd5@vO}#{lU|>q@Q%|ks-?XP52{mZN
zdXaX49VC3t<8Ba3fvbEZf7>`DZkWOFgwSjJYgC$0)exM_t)75TX*vuUtKDKyh2R*s
zQj#M}5c99pp){|J0^~rtBQVVdifs?Q8Z?JCK^0Zo&WVQdp&L1f8I`;I33a16CqIZb
z8e2nwemUw*{t3yr45VvN3|wcFO~Qse{gVq2ksn>@eWVd+>$=ARcko5;AGN4(2bw(s
zT?s4eUQNqc3fk4%-RoQ?QNYYs66>&%1prw)0^>He#)<qwj6NQ5`r(YfmeWm4$tyaM
zu=|}m)7d#s62bh`H^faM4S|p-Alzt&PUSDH8Fv`wMB3@@u{7C*y;008Y^TW&`4kvv
z-wt=3b&<v0AlXVqCyW=(`Rz9bt=ryecrBy}NN$nguLK++pbNdUA9lcjz)bhvUa1(3
z@k>jZ)M1CXJyA%60`t0J5{)Laf?s3z1K_9myhrZL5u1b7X%v*A1Y5oY80^+2II8X2
z#H%26B+WUnK$M(d3B;q3(+zZW9^x6XQ$)eqd+1OWg7eRdF1aM*i>T*OUT)7HSD;EY
zVB244V7{P~gwbellz->l@^>BsSB075J66O)X&^<Wwxt#5U>t_W=RzT6MN#4*{*?!W
zSIQ(%W*iewboq4BQI1-dH-yh;-l?N|>JD0UzfxLT0zK$I&B&IQi8$G?9$?HYJa0Au
zqj!C5i1LMjG(?VV3`do?77#MX*8~trf}}aSz#6Z_2ld0QYuBy;0l#Gxi}7@j2F~&Y
zL26q9X7QA;f<2z(in=3?$FX~~tXQZRhXsDewgSBt0}V0Cd+vsU(}Pw8ku?V{5HC?{
z4g6w~nT=nHB$3iTJw1UAma0eL8`yh9)<g7<0w=f*q{sX2XNCgd|7&IdbqzI;|G~LY
zjHWahBYF#X>nJ%vF%j9ybSM$V=Iy2gnw|6&F~c^~9ZT>YuwP?q71BkqwiYhZCtpko
zsDrn5G^YANkah<|42DRQ%ol@wvxmr(z8wJr+Y)_V{+KF3`egNSYOwsAL%!>=N_vyx
zCmQnedDeMB0b?0TOd8I)4g3{CBjPn!2arP=z*~p1`S|hUQF3*Mq1(|Yp5^Z$U2sPE
zv&A^pdc*hJ)tq-E?TM0We@u0d+ozxyNaeH8>nTs<nheGCBpTw!&z$LRM{qX8`tRlD
z4(5kCl@+M0$3O{Vv5~77dCRaj-d)oBdh{>&C$a~F{t)V9p%Cm^A~Ffg0lj~SXb(gQ
zeO@BYlIjk7P<nE#E}5vhO8^cVTcf|aTpkH!?Jk0Wuxj;c{haP7VeOE7vE)d3j`*B3
zPa%1NJHr&|ix(c)c~DYPB2003?z!Ypd=tyM6`e@NNaR&=dh$%g;5O7p&^jY2sscyo
zXf$f1=*hr~R6(QPAjcmFhaL3X__vm)>VzF9Qv=i@ax>bH4SuU+3Oy+x9PEBWr+_6f
zp7Q#@ZzQHH5dqc~3;{I<G>nEGOi3RdG;MfR7mX0eDJ9Ax;fK>4;BSzChApk)u#1Jn
zh=^BU-PAv5eyu>Y#Yk-Ab$JpS=h;#W|9)7PwgZH)M5|ch3t-rS&m_U~C*@)k%UV2h
z7Cx1wI-XtSFC7i&N&~7x5m`pqIshsr`(79kSP6m>q?bJ6bD<vu&%*%C{4-)Ud22P`
zT`b3@DR=j50^%VdQ(9{CgxO90Fe*;aJ&LEfwx<yRwIOEs3)vje>+Pw4HV=Ux!L$&K
z5WwT)B{*5nMHtTDOZ&}5>H+R<RR`UOH_vIj5O3m+9i(Fj5Ncm~6u}W2pZDQie7@3)
z(o#>PS13L(R|qi=mO&~1E*e%%CN(lKPWg4lp0Bh4A~sK@buXogPK1;!vA^AiS@dR+
zKFhiJpauksh2$w>a-mrX%w9+URMH^Fp@M-H!<;r4zeWd9<gvZ`N&fEJXgq{eJ<_92
zo$4dc+9h7cfO|a#h#sHYVZoAi09@HOE>Rc1apfQyP=!EG-%<ct*qC$!ilo9bR)e7U
zDA#u<RQi;n)T5bLRxFgvZFn)#a&oH>@o~?FMlzllBaVTDwGAH>IN!{iS%}mIn$Oz*
zgOiqMi#eV|x&rgC1g@n!LW8Bz(2L3wl1*abK{>V8b!sNjW()qfJcXFZsnCW1TF?~|
z1;i4cUMP<8-;~H76RbphOz$+sgL2scx~7=ETd!N^h4Gasz!z-ELb$R8G-<Tq*?|R$
zPH-waMT$wFp=g`=y`j)P6#%GFHYwF$ON($;(lT_vreLC_fMy{fDbr0r=Newy;~#Xg
zo*rIv?zUAn`YDjKR}pT6&SC^<(xP&Yg)rP$Y^h*CS*oSzs`qRo+66S8l$=_${dE_d
zNJCUCjH3}u{xJbjiJ|*VR$U_UXSrJw5f;I@Bc;#+M-T)+XcO2TnqeU0!aplI5$;8^
z^88637EjuEENCPrpqLO$-CxiW_@Y}k*~;2F=|yZg?d8Ix-FW`#rU*Hj#A_KQkx7_=
zp$iDdV21J(R*4^Vg`wM=WTgcNfRxSH5q=scI1Wzxug2jJdyB#O7ZP~;WZJz$f`SF5
zBnnG|2<!vb%5RcN8jy$ONTZCcp@-5P>ezqsSxezAfVHb^Yz){n5__nr>%_#P?57+|
z)qlgNZv8~bF|^&|iI3XJ(*JiKGn}ukR7A&R{q{^hMF2AxzlzbQI!S~dhS-r!CP|uD
zBUI21lD>p^#b8|eFaGsn9xm%Q5Bjq#^~6pL6{<gZ!!Xk&%uf!)KyQp?XeA583#ieY
zYC^J^VCxmogwDFC&u2S7oFt0YTU2X(O8;uB@+&5>`IkD$Upkqzv-kGa>syMsCwju@
z6Y<7uru!Vh5mpfrMKs04;R(#yz0(qJk@9h5@Hk3O2%W^g2F`?*rit*z`#TIWil?bh
z)lyoGaswa;8vsikDjKtGXHvw00=?ho&dN-eB;VRY`opbY2m~(RK$=sg`f0Fe89(oX
z_7}~YE3)+c?p+YsXYud*=?b!q4H%hGa-ug^7RjKxQZX*bmU#nc0M!I)W_#jae`Z&o
zB1x3j(6xMDd@eSY`~nDrH!Jx)*nSVAD;>dy63j}w>LhPg?-moaB0c>;%n`I&{V)ph
z<tsCxwb~i8`?st}!mWXV{!WWF9@L;;Jw1oSoM;~?yGE|^Yjj-={WCa~Zzk}8c^J+5
z8Zrya6AQz@PI6~O5(aJEDkt&LtYi%B2n++nP*6%e`kB7NCQue`rYlxEfd^Uo4hbo~
zG+dq5S9oa!Q@O{26h*>h08_EB|AROX#cjNobIE~p1>9p1q)}90hC%SoC26Fdi}^c7
zt<gX;;>*|0`L3hd{E&vXHlVSp=tNKnsXefRZYJU!+?7`(f&+o75bFNE#i}G>p7GO3
z!bPYZNknj{1F=cFVS(b!w{1>cA~>vlF?Uyvpv_a@&>Uw;Luyb6l+2Ezr2iZ+>uf8(
zM@^|NM9l$hEAN<G%;i9EiHBEo16aoz@EZVSz-ITKUt~p+M4T&>wX-MfIarlZHUV=0
z!-ed9^EDDa23N;D7F~|`=&SDsGt+1y5?Lj^OR%7H7yfc(q(3%L0W1;VX3xJHgJxiL
z@P{l8LvfygLI_;A&^LxcAV&NfoYjdU#IfB-y5e&&C;~0KWEM<1pO=5%Yx498niR{i
zFOWYe+j&HQ*nvuL92?5L3ED9m^BEQ>zxeG*RoO0m{sQwzYf#L9g7P`PtlgMz!Dc^*
z^Sqp=H|6Fm{-Mc3%PTPCEWB@g4KUQ*#G!DHWz(t{yJJXq>kV>$)dvX*2%WUG%{?1e
zT@Mp={Pk6C*Wzd89q4qMN75dhlWD8T&LvIcSMWs1cdTk-E<GY(JnpfrhQu)a7lr^A
zHW%HMXZUCC3e&4{ceODIbke6--!L<$4OY8}(P)o6FVnd5eX|NXq~&DuDmQkFof9^j
zB|zAa`AK_mo{HNnBWH!o2+j%u(9px)FML}^rWS$4p?KeJqtyW0GV8FHU!u+N)WKuk
zis`q=8J_v)Ar46OB#IUHY(#Vrf0bCgX!K&RwLCzDMP)ZLCkiZi$ddrcYQ@MIvSn{U
zT}ePruH-}-a8$ps8>}f}?Go*iP!NKTKrIU<F3pMYwe0o~u&epHAH5?H;FKqI?}pQo
z3+m7E^&p5Pf^AHrJ)iKndIhCWZ}Xr5Vg4=U*GPL6jqReyiC=1IX$jpibiW0-!KD|C
zgaD&eErn>eGLDmftvC-Q(LRHk3e--3MMLdkVMD|gWYC+$a3#t3<N2zVl8hA=8=Pb@
zKUqv91HHOJ;edLdEV056y%8VCw0&q}8Y#-gFU1}gY?A$vh*O2~yYT5enl&Wb_iUU7
z7FzLKfP!;?G-*4f+`_^kR|2WhA;{_r7&jlnh%83;2ompO>h@D06oTDHTGDp0$Oq)L
zp(KQaRW2;M{#Sf1b~4@&Dc+pDIPT7E&iRg!xtxT{r{QobG_Jbg%CU!`DZf#!`LwG5
z^PHBEq(ap&Lt1u>);YyvZ~)>FG{}Nkv!iruoS13pN|9+5P~{0^DxM}#Y16Jef3MjM
zj?N$;ias&juv;L}Pm{IFBs%hl7|;MZJPh6G<I|CkYL(r-38isvA}p+{V~<H6(tkUU
z9)<QVrf5`o+pK59r(dCH4KqrqDXgnizwlvZW*3NiqH`}yb)eg>1mYgUI;jaEJ|T%z
z<cI6(%!Q;}<GK({Cn30CUVj8F;*6<A+qnILC3f4`QMt)0EZ0>zF?O_CiN!-~!Cb_D
zS^Gy2&m>XJV1zmA)?EvW$-p?bq%nQzpI8Du4<G~BgnjKY@CgandHAp|`?k8Ci-6m{
z!ZbSA|KPmPJ_Tl;Wlmx+O*Mp3QErI-*67cTsz-|}SU-FLdWrG5*hte&gwgUR$zSZ{
z55X(GDg=p_90p$n?^!{#9X5_dY8A8!jrsVpo?=`oDW*wmC@iK}Z36CdL+-e4wHq2&
zHZcTJ+jdR>rEg+;1ap^_KZ@bu9WD8#iOtSX+$tbV61quf5o*wBNSe^yLf4@8zT2uc
z5Q!|sCNg%-^LvY*dl*NXZ%LflJaQ=*x=>r-pVBZHHhruhh4Vh8Z1tsxq)ihav5cb`
zvYRA1|G8z!xJb<6;2zD)IrZCF8tR$wM@<}kq8{$+X45}8?uR|_IyX@8C^`vo5z&ly
zL1x@WBxwbR!Uj1@H$Ff078EwtX_u(UKeYsMa+4h#9H5Ze$7T!`44?+#J!k7-z-Vwk
zeE26t3NL2#hQ^}*GLv)>{TG|Ogm8uM${uu;u!lpnlyuz!s&k6jao-ZVc3iXB-!m^6
zSB7aYT8^RUL^HiPj@j1C=eo2OpH*vJWS#K2w9t4~4t%#dHWIKf#FZ3-978TI*curX
zcD*puDH|^Ty1L&nu2ipDTB3{zV@*3Js9n+4RF)#j62uuP88;f^jSypTG0KL<;UI|O
zw&8Enr(>Y#+yoM;NUTKEok~LCFY%HGnI=f%NtL}5c$2dzFj9jRQWko=fw>laj-{c%
z;Vo2TfO4JlfuQWC)v89>l*)s1Ifa4vC3@g65zk~=P8&FJoCW{NKgDZts|n3j)5<(!
z>z6NI(h4yCIlWV&x+OF7Pz;zOi(EcL1h76@zLNNad*~7A8xo*<V_v>`wF`N*hHbut
z&m%VHkOB_qUd94P52P>Be&O<e5l?$hZ(v3se$pCX#b{`4(4>QKWampuOVKY(wxS)t
zP#tH@pGGg8M8h&P<Z6^nvc7mP;oF{I19S|Z`3y)??j`12_A?Q1zU{kIDsarSU`)vg
zTf(Tmg{Ut(4CUh&nV2QVUFrjV4sHY`s&LeA_hzGoHWlWCWjp<Q^}=gNL~eI?EysAY
z7CNc8lue+4FFTUqBMENQJ;PV+>Gp)~3U?(`4!*JLeb0pct+yC{_xwRRk^omp;k6GV
z#*x4?&7qMP7r?WcRBK2slfw3pONHJsv_+c)H@eZU$sd-D3AS<2A#;yKr!xjD2A0Zc
z%&e*IMgIYZKC9ejApz2hItgCml7DG;VO`!;;wh0@9I3lu@4-^i$XUx?b^Q$HX@CB8
zoG&vB@@wo0D)$wyVBr}Q5NIfvw(dUm;)1$52*DA`O8O`_phb)^m8t95Z<_QD%S8LS
zLe)~WwC=4+{qU=hDr0WjwzK>Yu8Wo(=6VyqiH3Peg!R#tP0=OvTeyJPsN52XbwJk|
zyRrfpg*kti+)xv!F-p_@?>-gN2p1g|DYZ{#d1qzwZ-aW0>taWdBXulytlW4R5P@De
zbvu^B{T*jF%%wF5%O+qH^(lokVS7+{^a{01@wp^R#4=9Ta}yM6NY56b-1Am3_<e33
zG|fcwT-x*<bbTniY`~!@Ib|$ym<6FjCI}!g5yOD)&sr$vDIdg`wpU5<RV>D^L;kao
zm7$z5eV+m>ib7`t_J&2+thUk}D$=HVY{zxJfp(hxJnc;sIx!Zzf!B%u2}Sw|kdo}2
zur)COxj)n6YV=Y))`En*{fSSl$$6t$ypdW(9H&C$exS)pPEa@WqCtIT3-21P+XHGX
zL_}1a(Y{U#9UvXzdyswPGAB}#K!RbEjqa2j<|~;ZH={suoP@b(zf}X(@#94HmUW(>
zcxupAjQCtqx*)4(hFxOb@Jh(q*Y7}<iC5BAUf^E0XRvpzY=w73rRNvZ4Oa;=Lo*21
zkn|Skh9t`GKvDW|w%GW}#>j(B=^e{D(WD^LSR!%-qO{sS^zS`6xlc(25d0|Us3kCA
z6DE5q3n29y0n@{p8mKGs?oU2p*CT=PijY%lzn@uO6;R&=nI6+=2HJK-mENF75dPH^
zITP;MwTb-u?vXpAXJD<ra!7fA85T@~%pUhHb<*B?>8o73J^ty=Ds^v4P4=8K3g5#S
zn`AG^Xmjn4)SI2R{8Bc(sbkOgA>-}*G*(;e%UZqf^8E3;mu6V6Zv9BZ@%8G}S@XI$
zjUC{h9Ho@f-@0Gs!$`*_`eA$bDC~-zxNEQ0fKJUi?tHT7{N0w%OLOy7{9jo1nJ_3j
zD>yg3xT4rOrF8s9O}(s-mp+b9F^*D$S5Kzh1Ypk;Asvj?XE$a@XK>##@o+<G>+YCV
z4XWCmSJ|cs{ne`83VhPc4H*Uy_eI>VdAKG=Nff!QcBPqjA#^2gA$<==7mrYX;|t5T
z=`+vV@e~luB5aM$F@BGQ^*To&7&a}HPBCkmh&i`PMCEWH@Fo=cQY<iZ2b3NW`4aK@
z5HqQlaplUD=dWIkJWH>W9migZwuLEmuTm^tFK9mnJ0K1C52@bpAz_xL=U1#7eMHkX
zU-xHajXl&gHbys=R(-XfvX?e+pdRi^JP>Z-9&+0?cipOnTE5NzB1p|=8U0OivxZ-Z
z!Q$J;ormt8+WY$T>s@~R)xi6KRV02D#v4HbK((nh{0Y5AwzZ6w%g`}})IuPnf@Y*-
z^f!%v+nbFUv6wW!zjkz|+)cq5FVAMY?9#RC9Uv19q%GUu1HT!C{>c+3t}|ojiu{pS
z4Rm{_=QeVUG&=6t;3}XegU#so7P~j)c_UmWUwn?0n2>fx!rbwF^*9Wi-wB^-?4$1u
zp4Cdg+0Wm+F`UyLa}B3%&`Y_*R`vNRHGf15{JZ+B;pTzpgLY&WuVS+N>7)W!xLDF^
zWR3IFr%yA1lR1{=90ImYjG8HPPWJbJ$c^R!kFaF=eUcB=>*`-i%|TUq=t^I*{GONV
zJQ@8}@(F{U^iPh9DWll{zcDN5G1u2st7~ZZMqhaiF{Y&F&!6wwz1tiEs#OO}s}2wq
zq^g_S({(Fh7f_*%Q(&iY@dny`<%gib5%P)F`(DB}spfPB`Qz+c|H9rwrceT}XG$|0
z&%hbhe6s>yJ!NQp`OxBnXdv!WeZO%$Bm%m0@1BI#u|1UTG*zw_7Oq5p#?b$_1;OdR
zwV&c~{1&pbaF;53zL?P{c75~Fb!c8~sgvbvCk*}+VmF%-e_kOgEG#UhcIH+EO+B#}
z>BAP^8%TfBGBc}AXDkji^89uDz}cK1W&6)dOSMC-?d?Z|8qKif_AG9a=wMo%l~IoK
ze-Y>P)UC#=es4f+W<c$XFf_cLq2QWf>@(o*2=oc8Rq_N22^a{z%=N!;%(p`EJ8p+^
zzLteXT%skc+U=ONi_17*kay8wORkJKFuhM6R*IIBA@*qE>wy2AZuS0Rl)Jn8qpd&4
zT#`av=S5^DR#wL74u6v|uvN`b;~&`>azkd)XnRE%_VOfy!|L$(W@<*p$I~~uc%NJp
zdUW8fJ*tF+e<;m860R13q9+c{)j|6vin)M+zwA{d=^qr_`;lru&h8n4^#x!0#S)cl
zo!X%K^$!bL>zBVof7L^s{&k{KclA35o1|LcjsZJwI=ofCc1_OD)^PukyVOB;4FsC2
zE`_~SG>7(k3~;I#Gf{W69>4VrC@BsK3UUBHb?yzN4{!1UEFv;L_?4hvr#XM}a&*mL
zWjgoX>|NZXM|4w3fVSSg#Z88p)>iFPCwU`^4Tp{%)qa{1in`ouT=_JtUhA`iyEx_4
zZ~Bpo#)n!jU*0P!u*onLhhUws%(WISE=vM^Kdw2T;F_W)B%htm9kIDQfb<tD;F>Ph
zM!tcXWk8I^ByWV$Qy`VYe`R7|t?TZol+x@sKeYYVjr~#GUh|V~Hwr#*;lc&M=wTbl
z)zHAB<0^J7Eq49;Bg=)<Or)598o$QwFIBqtp&I0sW#jG-KjG4avwpa0<Ot+BzeLz7
zw&PlD+qZ+PO3fw7LS6bPSb7+$=qIhisxEkKR-@%^Mejzvc)#&Zcp^j%@PFZZ?|pdm
zkomIX{mJ8kGd<BpL6XywE83CMQd0Fw&oCLi0ZZ5C2VGY`47dcrO7*d04^O{wX%}}9
zyNy0Orab##>CDE~bOg@nqJPsimI;5@aBA&;UEfUCM&Z0h+=r|+n90}n#+TQ03vKI?
z->m8noytGd>r^(MgBJcc^h@EmVE34uwyK`A5>`usTNp=~+kOh|UO;=|QU}g{5T~KX
zEWdu;ikol?;h;R#ys591TGk#qcI+6*Wp8bvyLdn<C!2UH^zp8yZPfNh-f1ee?8rM-
z4nkCWb?Tn#B<Iy-yJ;a_r}|w<URPJNt_ZoG4IQ~LWLG?nPn*+T-}~*`w*=tAbKsM~
zOC6~gKJrKsX+P2D^`>p>B#w!XNV|zi_bTd_CU{n6d#gLu7nu6ZxBClQQkL2gawBm*
zvtnx12ret+-l~;WR?~Y}JW=h%b!-W%hovrDaylre0@TIosw;WbSNz5~m;2Q|J}_uO
z5s-|#+tF-vt~%OQMcXf2ZLW{GX!kjP!`x`#jYedU*lG4NJi4E^d=>TR@+%#<iun+2
zA*LT?+!Up{R2vOEK+q74G{~ByI)3RNZignHY-IS`W+60W8EpA0;^~Aj__%E^vyba{
zwZi{RA&x8=8ja<1i(AuLodD9{L6wxWb;Fx|`}SpGFaQp=0uv&|AjYk$?o5~;rih9#
zB9wvzc`wG~t{ra<3*jvH{_*KCtnY7tIMTP>q<b;*X!h``wCafT5C!N|PGe~MRh-$?
zVk67ry3JadhlZcWL)Oh6kZS8VuH*;V#3%T8u#BDAaPZ~nVXmXFFxb}M=s4$xI?=$g
z;!xD2fZbA|u2Z4jYU~dP2<FD8)$XFA(o#}VvR0V{2R*0#0FPdfPf7w~QO$sUU0HD@
zeJq;WPDHtDSsT0lcAtZVsF*CsbFAY0I<*|h4GS2MDqC(GtX8(kLWp|#Nl14GplH8?
zzVBg>F37dtg$!0X8Wqm~c56D9uf)`YH$7uyWlm(|Xtg{;khu#oXvZn7ts)oU50VN8
zh&96K%I5v^7odsAXUi^bymAzOVDOlZrKkzrw;673Zo);`NsI8fq6;)updEaCLO@lb
zWqV8dcfJW@E%py>7MKL*gbqOS<sA)rH8nd%p2U8Tq}Aex%$cX`pFC{t2GJXP@RfJ%
z+O-lxRMhLv*8U4rvyn(94d%3`=t_!rt(4%bxlXEi%*oyksL`)W&a2Xj_<%!P3K6j(
zB`HgqXw+lM<L~qrtF3X+7Zln^sse+8ig#rkNwT<^kf7k<EHyHIgohjWy8S8*Ew~J`
zV`YKe=&n~vaVyBz!-*p1K%EcQFUo5Lavnal$;1xDAy+Gx?`sAJCTX?x4I-l#twN$a
zG7rAh;&pXEU7Fh+T#)8%Bo)(P%tgo?7pi^c(#z7oGn)?WvW=TIQEfTXz`z2!Uz9GO
z0aSq}7HO`m=c#-fjredN;q!viX(KN3A*oacp;vtl9d7DW)7lpcQvTMs-a@;q$qiPU
zb(DB^d3Kju6(Qyc<gXX@_nX|DJZ`eV6%#||_8{n!93MY9J7Kn%i$M1+V)>1hO6;3W
zx&9I1ehhZ~B;gK?(A0Fk=}OCm*`NjVs}b0p7o45<Td0xYnj^lxI|i@f5?dfSu#9`W
zcC@;-_K?|$)t|XR-%XK=2ICLZHCB&4;tXJ91e|=hnnOwIv8ZtmjKVVJhGnRa9J#+C
z^~3Nv<etYpEm}-pHb0>P%I~XLP<J<U=|!W-vnJEuw9-x3Rc_8*G(K_f)`JK8brY}Y
zPev$!q0vRx!B#WpKC}&;w2W!Sq4?RGG>D`)q|F+0s=o)Vt#{7Oer&Ov!rJ*jcKr%L
z(RT)TrSmfH7#78hqelU7tpgl_31E)1M!`Xmvu4hW$7-gcKcEq+Zh~UKeo)8wMdr)q
z1~!pI%?*6=YC~!p$+l|<OUR+}bBc^YjRqmfs>1g5y{LkUuw^3&QwZ7Grx%6#AkD|T
zxg}t@ODJq-wU^{fw^5L7y&YPo0#mouPE1TJ?Bx2SQ+epR&1)WBt^a^%QuA<Ktq+bM
zv0G!0uonL#)v@JBzwYpy=V#+X(2Ys7wt0_Hm|`eTr64e46WswS+pSehBNrvX&mf{8
z1g%>arr(UEdkbgdMHkcDYQySBxxIN)d&Rh#6d|acI`{0nwYbBNXXATT&D`=s$GhMA
z(t%el%@^$rjojPV!}RdOTN6Msz=dWY{e@{kt|ab;YSUYx&09(o4lHgWTcU}of`(?%
zAGdjUm6*A3G>S)daK1X^khSLqZn+U|WSsvoaUwl^{It@8D-AWeo1brXILU%~WP%%^
z*naoA9c7VVl#ckOb@gTF+rq@U{I+%ZAs*^913q@XA4OHkwiEv$f=a6PEv9uYnNl}5
zKHZ_P7bQK2mi;qKpuuZh!j3(^A2~`46%OLN8NR?+7Cb&E7N5FQFU5cEj6rG#0GCIt
z9PSL_D!$4J6F;=xn&YwU^Y6R!I*{nY5R=1Db0PQ)_D9hj+2!ep=t4?6aP)<KUdIsi
zF|m$nQr{Z}cD(<6h}vLu9$H9!`Vv|pnex0961EVe+l3I*D_qSxkcArm!DDj^J6}qt
z+MN3N=EfFKNX6AyU0uBdISgh~KB2C}J7s6DDQ|*cc0Le3b%Q!xK5a;tByYT22@oG7
z?!D+zm0xrWifcN|6cGjPEiJ2W-@eORb>%817#gN@3A-D*u>WT+^L!;f{HuA({v@1T
z9!xVM`T^<5^7DFv?K43p>Q&OVC)S}}uC1*d;88Gi+<wNjYV}WI;!J7|1*G3h9}47A
zy9H&q3uQ1U_9JOETW&%^1Q1NfP`hg7VgujqDk>?U21Nk4LDFrG3F{z_X+mYHPPXD(
zUkC6>18JIHWFCg|AG?4c*?|>r3Lz_s*4|m)RHper%%n*V9Dq;WyN$u=cQ6j4&{9%y
zE+Rr5Q~q!)1_n{)z1x!8*t0lkH@Pp8C@}Uxco;v_Z76#82PpTNGN<t2;&n+F4svXz
zK(4OU!04v=*K85I2{J%ETNAPwWs^T}n7UILxEd7w+T)#mJ4*%D!_mWsw_BXb`L;!a
zBzeP9n;R29%P2Tk(vngiHf#?}O;!Jj33(3WR}P5NeqdTBC#2&0j2q|o<jE6NeU{Ax
zwH#ejHERK#sjYTR9khatJ6KlUFs->UW-u$@W!LWghoVBZ?ZD!%i_+^~6{QQOQLrvF
zx*FXyFqF_^AR5L2?s0IP6ufH^Q)&{Ae6aiGjlztA-`mSy-6_9NtH_Oi51Z&sn;bHs
zCWR{v8gvEp{*$XZV{4u3-@22%mjKD36aZyH8m`lfA(!GO)6*YT%v;x%`9w!bStz)V
zNdn5j<@a+04!sh>-iUvw2a~~%L|To6r5qDw(a%n1k<hq+`~doc{I}cRJK{m1n|ro$
zQG2|D#Nj3R?tr@I0m)f!paYqbR~zfb<47HUGVDd40zbtxLW`s#4dRuUiWuVg(;1Nq
zP>g27ZM1lQ^D(FOt~9Fep#7JCh^$?!MzchK)Ui^>4;3}F8{TY4(3(ka&gmRL9ZABd
z>gwtdAFcop4M(alfn6e1@e*-G@t@lZGL0@!i81zZ#wx8Mm)1zpDZ?o#S>=7wC>XI9
z-U5)GImkl1M^d?qJ63_UpPL@K5+3gQwv*c&8GV3`?L??d5_Kls>TJ4I7iHx{kOdqt
ze#nwF_0gjr`V4>Y@;&~vZqo+6(LzYo&O+=!pur^6{d@P89j9NlVJ<#Lm1X^_0RQ=U
zDOfcO?63Vu{DowSh)N+dB=N%I{ejYWuLRnGv4teQo(1NCZ-UwvH$Cy6KU(Zlg+_oe
ztt`Qp0gbaa$g9IycIH=<S98Q|A7xDsT~%uGTSR~@$DqW8E5`!OvG}uz#L;SG*F%R7
zEAB7`)8oUg0}=m%<w$Z~^cyjLygy*~Aho;r-k2W=%3K?@+2GuzqQi5isL}@^_G`yM
zJ)*r?VV2+Y1Eh)woQu$IMoYA*0D0q5tYsMap00wW;1{}l!!HyXsrh`C#hGb?0@^5V
z9ocm^c6(B>O2rTH;WSky!KFQVv=F>Y%3PtLsYb)DpFVxEC*H;f+wt@eI{k$r>to2z
zXFj3DT3XSY9sl3*XzQ2WsePxhjqn*b!>mw~FnsB`3f=nlL)P>c3u<^klq-N->f4!I
zAlyR73usgnO;Y)k8WOZU3|lA=r;u+1MQT0kAlY{OOW*hPckBu4`rN?qp3t^`;M1b{
z%l8|BZXlTex@wQ!%W34}Q|7G1_~!GHl40z*@0^SSfYzUbQGizMKDTnix{0hF=Q_5P
z+pxkK3|Jx&m@-FIAJYV<NHtD?3qi-cjza{-0qY{`R=rKpp4Mmg_dt$Vj0`Z?y(tcC
zp^TWc&sdJOMhY>`nrlOAwo1qos4efV{#rx(2hMe}m=x1mVOY9L+Dqi2$m=`w=aPwv
z39KvN;b=|GLq@@UG$`ii+Wc-Z%h0eGgY<Z=I6TlRkg{8=B!Jj=KQAxfcz=7k&tOiC
z(4?XQfnMzwR9g5xfN(8eVW@F$eYt7f4AQHXzS~Ig-0f=S+T>4NxzwNe@_%T}_={AS
z>Dt`*yo}{Ko<IeF1qY{rG+@#{1)wRa=!59GlHt+{s>L_$kU=PG+N;84-2&I%9t3Q7
z6H1^_s4-V!_G|Tm`WL4D=|xA|ZZ4U$ZV~U1Z~Ui(j=PjTaX#u0A9MJAlws<_hu`}!
zoQL7%(ABkgvN2Q&!wlhQ$V>J!@Y7f!g;ovJ%EvoHHRSE-&3@%AWfSLoSs$T5K(4$c
zPra4kt!;b$EjKlyH?7<Jaxp?Ewq2YV+_vk(#=Z=3$N&Nn2t1Gt&hWG2S)C3qfz}*}
z1QfV7?`rKXA%hzy_sBywpmCtjPzL>STChNiO7QJ4{+eZfZ7Ph6zp3_MNEd^RLgD}@
z1#nrz!3$DSfGMT+Me=bMdp+Ny1i)>g0H?ru`c01sLw%eqzyUim&CSeuVEs%L$yLxG
zE`vK75>i@pnOs(83k{Nes;C4ZqC!<YH2CcyfOe!mxd6>nt4Mp$W#F0fs{HeA<Kgna
zd!zU20-Z-P4gvKvajSh#%i|DfZsTi}L$Ne$@Y6FHFRL&la=iw6Y|C)S#E4?<9I<g0
zg*2IFc|d)ITk@kvk3wo`DbTZGY`+k`*!|&%bG=)07w7)}+@1+|bN=R#LBL65k+X9A
zd*OAl-eERWXaqxYZTiK<K`s>sDaP@^dD&@c4mV#WSvWk*XfAQ&t4_BkZ%(jM`zuqD
z8J<H(H3~LIj)CYzDK}Cz*fQ_9I?5f_eRQBOqe9N){d;r3)0R04skb4ke*+$bPH@sV
z0_9UKGYpv9xm|4jaK)fS+{RwtJv1W!nKdCYQt5*Z$Gdmy;rB<39_<4v#7YeCT!N+$
ze6$UOpyFOzuj6qGH+Nu33~zHyw;DxYQFGp)PRb2R`4YPABly&lqy&>8n&%H6fRLxB
zvUwz6N56{UwA%=+GBPTbPyjGM-Y-Hl%zF0}g14Lm+}7gtM?wwOVWpDrkRSXn5BWxy
zICSDk%??M*s%h7cvB31csHWYi3W86K=`kK*k^t!uYox#n2;@bRB;iwSBh1967pL)N
z;DI})m0DxO%>1k9B>8~()y%WkexxmT3}^?84IaB8wZ`!TprUd%W9HZWI2JKFUO~Et
zIoDjv^pBX3d*!uMaEtOpRpK1@nHi0LKMzSNK{dG1;e&(fJiWXup%F!mI<Q^{4M)~N
zIpg#j3lhnU3JAoZ{37B5Dz%s8kp9+{;V-x$X5DeC)UR>6TBEw|o;hpQPw{HB(+e?Z
z%$3Ku_wP#PY-ySw2UgKcC#R3At&yf#XO3*jr}|6!Qy29W<mVBQ)~|mTmvfyc__Q}1
z`5h^H9X<N1QZxHOJZRrBZh`6MQS`U8H`Ta4x#oF5-Rj2x=;|hisb^abz#T!b>f8Lk
z$$BJJugN!J8PX3w{NSx>s8hRMxLaubV$vqDsx#nC_%(GDL#jv}5DDSk2;>0Zx}363
zi9#`MR{(kHn)^d*?;9@p>f{xQ{tiBOhp41T+%2uF2GYJ2)?V$>t=pQro$;{w)FT{8
z+!A;_mOnI3J!EkoK<_Bjso7pXY>gCTsv2tdF>~j{)Pw67&KEl6Zms&&NSC*v)RrcN
zidSer=0>ocLda20L+VUf8{lVOa?WOtFM*cpfCVJR)te!IbPJ_CySI&g_dX<L_ydjB
zneq$*c@5o@G^<D?&Ny@7#(rfz>39!W)qC3bfUbXjNG&!#crc8Mr8Ilw$+=`oFlZ~(
zw153Ni4gWXgTht<J*6xZbOEXdQ>r?nj7Le}gZ>X@AbO*)&cyg6!e<03jyQ<mvt&(o
z@u(w7KwHP-LO8h=J3%R0kySN4<V|m`)2{U31(tkl)Axqwf)XykmLlxI(zGjx_w!nY
zW-F)-C~am5O?zYuco#L(t2wMXba<gtmr(C3TnhM&3R)W=!}sMhPW}Hkb|pC;3TW(n
zZYSt@0ON;?egg1bXz>>cS(=|ZT1x0E=Xe8CFn=f=go3jQo#0G|^2?3=&{pP$K1a~M
zPSgU09xb))Q7^+ypt1zLc69Ly;NfEoI_#L{P1q?ZPyq^qbO!|dx-zd^Gq0{nr@lC%
zqvbSjwBb!PEyc{`KVeUM7GJ0o64M}$e&h<0EFxale_8tUZ2_MP$SpsF2Kr~2n&yQB
ztiiJYUUf8W`CbEO1+!ycVz|vNb=ZjxkL=B%ZS+*|BT@fur31<hapwM|(kX(HToCiR
z$t8@S(!sx`+O7UgKw1F>J5KG_|Im&`%m`%F4p69*YIjl8b^E$uG_y5Bh|rwZct+jo
zu-dG!q^vOH03S<p2}VwN{_?-9c{taua(JHbSK|4x#vjWbwvU!)58o|&Y`0jAK2mk9
z(^Xre%|wQFF!|EKamj~}B%fwgziJoYX@FPMM~v`B4*)mxi;@RhcYTl%3mp&2i&@sT
zMEBPX`CDj71ELF3i*g5@j}A?#+bKKp!gqDToIYGxZ+WW;Q|+@XEsKwstczTvx$)f@
z-6pn=e@F5d8(V!b>h|s17Z(lwrwLy*Y9t|b`T5Isr`tN95J_iyDbk8hk}lte0K@8?
z->5_7fV*_^-YUNfrT|^)itB3vE`gy?yi&RGWxDp?U8e5vhrmsbpo~K*KOV8~i*9B>
z#qf)ZLaC9@V>OKmA8d8(PHl5`L~6ZAS<qr|@*Y*H4Bu-G;(@n<{{+0jK_nl5Yh%xx
z`XnH4M0p((K^g#v7SP7*`@^LJsP_<ITRlKyifTcN2sDPkGrom~i9YOM&2BsAnlxG1
z|H}s4*Y3ki`zstrEVH-kd}ePTQ3mm;9WahH$uauf@w|OawLe};Nm}IBQ(wF4wqf$>
zFKcYh*Qo)+DMSf<2YuQRus39pB<79$5xqU(JZZRe$dz~Z3DHH94etiq8Aav%FhkVM
zgU~HI8fl6aLxT2);S9sc59yzZzos8q94!eGBE;&*MTF_|MI?x;DKADQ20VlxB$O1J
zg8J#B&vp<18{i`9zG8dbX!bsB(SjcVCTJd*Y&pf+{ep;@kspGdGyfyodLOK8m<MWO
z3zTGs{QMG?496iD?Zx;~-Yo>N3?DnHim6!zKaAd71U8w-5qU~O^2b$utgp)}=IL!T
zCKd!eS{c2A#+v_odWJ8%mD+{461FBNhxx84Mxu#+!5o&>{E%hnR}y2-eFBfqWaU#K
z$fA(x=J&g5DG|VFV-Mb&)puhU6n%y90dkr*wS7GQi|)Xd9FH%$KeFY&p@kWoJ_0l^
zi-<!+X+|_gGQhe&lTM;nfIndU!J*&5n3+)D;%L-u9O;l<?d3NH=&ZGPU*y#-x6|l*
zl@F_<H{SWL${yRF$!}Gi(D`%%l{DBuG_T5%Gl>eZSoohM!-mAZ3P)pSB{B}vPxlX&
zW4s%Uo&Rg$qq7A7+Zd`IT}l2mV&i)&-fj~f{x5ZpLaRHC+CNz6R)qyn*&l{CG^kNm
ze)w>>vk-&TXcb;c-lxTGYaY6~y$`EPC${R@d0CqtE@OS3F}Q<XUPq<@jGBJR8>#7{
zY_z4<PH1MPjd;KCVPn1+*TeX~seNciG>Rm^9+ZL8LIjq-Ryu&3aVW1gH*b9KRTACG
z-pP1j=QmWY<QdV>22_-XLF!V2<|ud!2FV0%8;|Ay5h<~(=(GHF?IijQs#Q+)rFqW~
z2kO<=NCA_P+6&`=>Rj+V<T!44Q%>i&NXEh7RThwptIyv9T0f!Z)v`TTOtXd%VnR#Z
z&u?aMx}q}0njX<xr@W*q18)Mig32T1?&fDuF}26_4aiz@KP_#=asW~c@Lgg{vvP%4
zsz=>c$D6Kz+eCI=aojG4s5s$n{rO+rEl%esaWH`Dbm`oAnl~9RrMbpFM0^GAHN1Gm
zs*znA*Wf&?0JPaSe5E_3KW6u;%MVJ&F>V>8DBo8YeU%tg4p@pI%}X!-4h|<Sgq)Q~
zkPhNZ)`-6ln(zno6Iy<jVd}P-@uf1M=>>HH)}q*X2Q8Hfh;Ujzbj1$H#31eW19*cE
zIxV6KoZloA-Cws5i$e_F*Q^FBtmeU*zPC}<0Cw^Lnh1;peaick!=KO3g@Z-s=4fOd
ztFK7FpSpXTs}&mV{Etkz0nU@B6O`gmw=~pfO>)98ZSA9&9I-$JEqEit4z$A9VPFR@
z(v&}czJ~lW5l@V(m!fj5iZrbr#eWaKvOfurqo0fVpLKZ}cXlw?N(()ZEuaFkM5+MF
zK_(_$VO9|s3L^knY#LH3E{k(SLFa;5jB&UWISyXe{-ihXQ`cTi5Ho%x6@=s9@!vxI
z>wVHB%o0ssIM@aY@I@p-PX~s{a6O*13t$v1=`j@A5(g1Y1!MgwiCJ%mi3M)>Kr{N?
ze(L{I%#j%``$qYL#tmFlU8C~nRw$3cXsjcw9RU$czt_8Jk)sfD6MXKnmC3HyMhqbc
zMi(#roOp;!=n(xG>42*+1b=gW=brks&58gW029#Abw;{U?9w{HOGRgXk6!vTv_ixz
z08`*h-NjyBh6k()rZzMyfV?)pT2XOYfZO!`6P3k&P++aloo>WMt&bIYKRgPf*-pTx
zok;esCQ|w#!=hGN3wG>nTD%O<#Om>`+yP-=$b8PkqZ}~zGkBU)Q^N=ls_a={DuEw+
z6r3WkJyp$(i>Tg!p!_i0|HwtaC+-4AhG^_wfDV*H5eye)y%_XaC+2&T&qCUxIHcv9
zniFpCyo%<3m8_X<Sq%C-;1$CBR6$oq{N`DZkk7mM?=^OrQuRKas(pku^IS03_Y2(S
z%gD%ll^QN+uetJ60W9KTDC;qxwg_~eL-?uqFZzcXgGkWtDmHT+ou7wq)$f2uLLuS#
zzTLE)#XB=)Z9{(tO?M(M^Oc-3^t#YDUptyQJH+eMw+C|m44MryKd+HxM?OIT<S>zR
zQ0h?eY@I!tIRdlf^@$gsE^N(JwiB7mJF!aYG^TB4g+(t4RUW@_ULcAN@C(v%b2l%_
zMh69$Og`1b$lZsfe5<)ruT#b|nhWqz5~p-py3(GH;63z#%iSYB0ZJ0fS5Eh)bYjJ3
zENJS^Rirzfyg>{p&Av+4><4K$ea7P8Y?nd5<$viQ%IoSF$!16e9%hhQlC)|}v9(}T
zUIGIS-AD6{8yEGkxL}q3H>m7!&)?%2kL=n8C}T}OpYGky&NB$B&saNp{@ow=D$uOm
z1_<_sBY}mMT0b~B%gCNZOEzW1K0ZJt5E5Bp^ib<tsW&@JnZk0oSkg#Z6#5X19`JLk
z9!6h10FGsH_Q!B&qp6z&pxuwI@yxMY#fa#!&gCa%6w)1Hm^UG^mJLvu8Ccz+n@F-5
z5!xl-W%Qg4i^4_bR_#1{_AHroW^Mfir}_)O`+wYAnHLx5vwq559aZCHh4Im|=4kC!
zJLhkHJKRAvv#0;b6$i~{#Qtt}K_Pc*;N<h6lQI$)7d+B`IzvI~#BsbXfNHna&9_K;
z{1JW3Y|HZbJI^i4o^sOKxL^3mJ0rXY4LUQw$KBrxd{3_!9yM)l)X<?)xAmqyt=25<
z({aGwjaTZ%6*s9Yt5v>R+}ppn;#KxR0_5|shNjczh!w!UF`VRw4M4TTp9_$F{8;x`
zGp}mSVqB3K88tO7tk;5z*FAwI027I6Jmb^H=MQYp=h~R=zqte7dxm?K$FuT~@QbU#
zsy?0;_@t&1EH+Np8UgEII0|=am(s5CqqI02{qsfdH2HCnmB>h)LB2T%Wt7m4Q|Se2
zN^WH>-1Z>zFnyx|%6n-s@l2(^-~V{(X5U_Tx5y4h=lis1)0JDezUxRo6_v~Qq-R2r
zxSaJpa)f3a%AwIj?98f98PcPaeB3LhH4y4=-0G_%P`MC;kd3_XUo!R8tF=f|z2BfC
z&^-Yhm|Cd=ZO1K_?qhsKt3l&})!KF)$(=7DA-SG@+Yy2Oe!ATTOD4htxfs=&q@o$G
z9T)efPl4~e<R-WRfY$0#>Z^2|u}vdo&7`DlM#1TXF%q$wc7c@R1EpQV3uc4EIN7Dt
zA>eK?3S`fxVDlma30HLu0?7Nx?_>6BM+5({=5L$eKa^y7+|4&OU!4Xf-;^<<Mjgn^
zG@h_NH9NaHx^DSPNT{PXwY6#f@!)v_(Kpb1nU$Eh%WY2+dJa~DB<dgzFnF4P3WTLj
zn}0@~#iIYKoJ(o$eXIs?4~)1a$3g5u3y5%d2hZ37k3n>vMWi2jF#b&>$}oq=z-Lg-
zz5{9K-u&QnbX+6O!AT{6UHQPs_6Z(OrB=j`cvSQw*)~tk6-DMujYc4laO1c|i}$zP
z1d=a47|_W3xw-zRu!Y7jGScF;x_G^;0R1NIv<sS(ylLFSctjH8%`Vg)xr%;0{dAlK
zGbTDLrWI6gw?#&c^uIqJ-gnv`Ps+(z0*ncN2q}T<C3?9|+WBMA(HRqAJWHArX^iM#
zF;8RAYAdGQjWYVhWMTAA`stK}T9ihU0G5x#%o!CMc*<zb5@#D+&7qabRImW90NUJF
zAJ{+8Yhq&&Wd>+WN09i;wj`yb6e|T(PA)1g*7Brz^`jWY`jM(h$&{c8M9`rhTYVs&
z#Mwp*8GQ;u%nLz9q`*xiC!&MTDZ00IbVcf<w&dQY1L{Dt5|0guwy9~rHUzeXoLl8~
zH(wG<>7D>3vjNibLDF6<2OiQA*N6X+%(xE{P|C=uF_^qcTRWNuLLmC!gFU2!E0rxQ
zEkVvwz4YX-ZbT13GKNlJAvz!7<ncNMe)vQfqREOkU>lJ}4ErWKC;mYgAN`o=qq~ay
z09Dv@@GR(_6NUqh4q@BSAKRMp1Or*OiK_Mu8^Ev*om+mwJCKP75HU2t4n;r4XJTda
z>tj}3F6#93Wpe~q%m^Lw_wO*d1v(AIHIlGg8_VU(C(h=UtViBN2;b5phYsn-=0eU5
zmX7PBc9<ku_`wo7Stx%>OB{&mN9|91Q(`WH(L9fCUQ5Z(w6CE3lzVSuuYtZ&NOWMs
z+6RV%Z|6nK-TlZa5~S<rTR7Z4qhPd@@NP$sdZPDxCN%%r_+YiSP8RLx1Jln8upV9#
z`~g#PsjCP`p=K#YNq`xEzW?lfhN)?pCc#VcF2&3D_&5Nes5OFW=^~xI+x1$&U?$tO
zH-ovjH2mvp;p0zpNVGY*kqP1Ap~9d>$nnHunQKo2xj}t^>gWzlfU=f5nn-Hc*=w7r
zwej7mtDn-t9F@_$%y>4&_34=A47+V)spmtmV)xU`6@OA{g(e+OB7j(iCiWuZ*+Ng(
z($&Hn65l<E{ehYU?LW-kTOWhP-<=+tn~ZhPw-ZxRoRl|#EV2?}-4p2+<RgyYZnB@4
z5IIfE+{R`IQuP#~K_nzpU}eUsTpQdxP>dpkG=N%1&}BH(R}pj+!^ju=Kq)nI&PL1i
zDpa@ED-()V47$-C8U6AY)%8gMhIi+1F%rx>jL^~=wBu)R4nUfIT;_^T7a#v8e0&LN
z4@#Xbj|QKvzsiMXFbD)9u2U99ihaRf0^$Q$#B~4RCm^vU(dY``1VoV{IEDqxe$0%K
z%mCwXuB)P(Pp}f$DAu>@lZiB6TnpcL`?h}D7F?&Lk&9-98fm@ELQM)_1@603sj<1c
zRDCBq4g8<{Fr+s_a1TJLo9QjUqPSAq+o~i3Mu^IH@I8F^j#%R{FYmYdQQ4R=N8uOL
zMiESv9IQ!G2wTkW@#X!q6KhPWM{pf8A3T_cJFjqW%RbB@45T>(d{xASgF7=bGf7Q|
zW#Fgpty24>j0u+m&~bS`^ZH-JcM1-k!H1eYPw7e%y8|>B#}Ky<AOw*x{oR6(jv#r?
znDVdRFfalPjMKv>mS-ICpkXF{xFZhq?N-{`A$1zC8UuyIQ=`Eq&|RrcMnZf67em(P
zHSL>9D8e)$rcKvpj%1T9W>u+jgUatmV%#)D3YhT_vIY<n;rKv>pyYXqW@z=cPA0eb
z^FvNfVab1WFA~_F7p&a|W;vugfwnZ3`MYukx&YYIyoIdQ_V(MbGVv*``wjF=TqYuN
z15Bn1?WI$P3EV_lheLGve~O2V8FLU-8va1-4ReeAh(d)+LWM5Pz>@QeS(WwLu@svS
z*e$76sekI02%o);UBy0na#t&9wHDYn@htQ6&Ac(AMMk-M_bK$F(B{Co!2n^TmM|3k
z-w>18L@OSWJ^1X4`s0$^BL=YRRjf00RL;PMMl-tpf{eaSj4Dz);pe<e-r2T?dLJR!
z?9zi<IOPy{a5U}*Zw?9TL$K1Ls~?>QjsX(3F}=kE>N87g4%XkdTLMgk<P@Mnag3dv
z{(boaGHVyQDDnloiYme80Me*69IjS@wmxDKZRJ#N)607$Xx)_HBkYx;g#{n;55@CN
zi6nJ}5gkY6j}UYphEbyq4Anara;~bX%6T|<-na2xPsvn>``2JmwgPH3WG5*j4N0#N
z^~3Rcx96A}&`z6N&`hOL7nk8wR8!nmnIX-`QDzG+QL?5DWE?W3Zf-=kH$QY8!y75i
z(sV2T1xl^Bg{wfQiCpwLWKCLH8YZ9MQ<lV79_Uk-tcIJ)ek3y{eP%`kx1p7dTS(k6
z!ZH*BCdK9&oVYg)<R{SJ@SFA#BLT!BT256ySLgj)G9!(31Ym28Mh_R(XrBGOdr$qK
zM)zLrC_wsfHQ@o^J`ee%OgI5eHUTzpMmBo{dKJG}D^QS>ra}GcDDA^4)hMOe7rMOY
z#}Ca`BaYo~b3%<!(!-CAAKx8R_SuX7*|y2ZxZq79W%lf6Vb^b)7ggeOw7oIVBK4>7
zeKgIk0)IYC*%jb9kSl2N4D#EQ``ovQu7G;Sr?{JSLG|wfe13Oe`G`+jdEUeM1Y+)k
z=QOK2L%PWxM06oDp<ZDpm!HAAgbdRyw4uCJp>3W*q?P0S3GkqSdpgn~bEO|QNpXSi
z+2!QbSG(!dm#EbCW}e_Y;>ysQ_kKcuv$3uHhq$uOF&47x@U($ZP4jy-cqRA?FiSzl
z=(zb;>T3dYA%<4$>!>A|%<0$yHX}1P+qiKoc&+yK<$h0rbQ1w6puQ@=^ur;{p}qC{
z2%d669A`$jKhkyS`+CRkoW2QwN21DOzzN+iOACv><g*A9p?PaO?8W<?F-*`4EJ`vq
z^eJBCK;i+%573!rT?d1$r#6=uA<ou2s`}Lz1}y0)<|dbmB8N6J+l!F_b4N#`U%G36
z;d_dvzxc{Yc3+-=g7*&I^pz3^;^q<Wb|s!MCP#N~qjKi_%on?9ViLPMw2ET@mVj1m
zI=bKVvGn_nFA6<v6zu=IHI4+6+JRA9TYU@(kIy?aMKm#6XC$P2PI`xiFQt_`r5b7H
zpJ3>6XZeNS<t_~9FbM{nf~C`!gXmdGrY&IKnDtnq81Ui|^)f$P@X8FKUo7)JAy40B
z=4&x?$38ds!r&e#YrANO^~fDe@J;7$D^`N|qGmBidl3IIL5{CqM<RrsZ%!36m_DyH
zY{2*<9vYvx`}m!!x}i}H6gysw(-YRMd!uQnU=^Kx$k+Fm_DCOqgrZeroPjbF_h1{#
zRbW<XcCQ^7AD4#CDGq=qT>QIY={85OKP=F_!XYCeq1N9GI%HdSvOzRB{iQW`rLQzT
z(Bu*geLoMsLu2yAae1)GQxksW)}wX#FSk1X4KWc&BZ1KDb@d<Ck!cYppTuEmXfTup
zhSlzjLziMNI9e;gi9j0751wJx8pTiqPR919&Rx1_h84Ar&at0?T3LYwLGK5n!Zvr|
zcCH1y91#(*wM0gEX$P7pWC;m3Brod)&D{{$5mzJO*G+0c2L6lQRi(dtub$MEkWdu$
zcaAqX6N;1C`HI>%l*B=;#IYvgQ!q5DU4dWJh2Oo!e)GS2ZO&DZrZLKyRs#5ALr<k!
zJ%eTsWECg@07}vE*@@TitoVbDFe)dSCFG3&-}$a)w~rLbZo)#qwEzjN99k9TnS#%r
zB#-&r<BRZt0@$VRO=An1>q}<iEs!A-zh5EX`nSqosD3!7Y<j$p;V$IK&J)`S`Zg7A
zi2aBl*d_V2DJ9}&`VEP=q?HXqu8?6cX2<&nb6Gw}h^#ZDO5irZ^^~_4kb1(ZsTo1a
z8NlIM0pVG9zn}J7_lk|u`bP=cKHxn4NCX>6pU^ds4K``bshtLdmYAH}^7tQUpJUu?
zOkAE#G}zre`*TAsqw9#SPQ<yS#Ke=iHqazR8qRQ(T7QBpfToV*Z(3C_<PN6PU%`kp
z!8`KOW;Deql{8{V*QB|9GSu>nrQ#Ou4b7wS@FXkIwyB^yMsUi%7If!*v37m)>EYKf
zMzD{COJtL@ebI01?^a0iw6r!AS@yjReizlyXcj{zg{HD8h9n<@Wq|ZLKI{VUqWa>y
zmo*s*aCiF;R7<m;CG&8Q_=9W}EqsPpl1PM?SMk}Pql@<MOX>Ih@<+WJwkpk$R>w+@
z@b|csZs5?mf}t(hv6~<JzaG2Vofc^nr08zll}6DzY+<SgHSEsc5|)D5O7t;G%Pzk>
zhX8c%54ho`F(F`V%%H!4JVpB3lp>wlG#%o5nFkmLjXj=jWo~bO|GI)~%7TA)C%;A7
z<X&$5^UoH@a`elgFG*P~`WzU7pscH-%g=xV<iks#X7S?&z#ItqLyC|eUT#|d((N@l
z@w!Z%YRkiXb2(&{JkgbT+hOGR8B|X1fU%6Cs3Km;5bQ3L=~GW!=<l#<^bs_jsp(10
z-It-!JXeaZgzh2}J-@a#-L!VT>qjTI=%+DAl1511FP8QhM$gozLqjUE5P3}2p_@P6
zHXUP7S$|{{QRM(GmVaJ8lk5ObX|+eo3UDLma{vU-p`k{>Ph;FF*453zbbpVpe|blF
zJ~aZx2HGeX%0Z&@40zJ&=xme)0BC_&fuuX~%hK5a%pIR|B;<u$v&8wzk@{e80(cA{
z{Hdu6rS0^y(DAdZp%{|W0@eCg*UBg3(5J-PT)K1#B?A7?TnO+Ol0ZQ&BUZ`L@a}3|
zkaI~Zw4@FuD}V|Z5{ZeS0cM%^<h6mseBI||js+lnK#H0!m9|8xhR4J7C^8m|wI3}E
zGwd*g<qycmiE~Bi$-$8Nw>outG==N=lu_N$wozgw!rkA1dDp~s<`&Luj3)txTDL*^
zuSG&&)FQ<%?F5Jt7A0ZmY2)c-*D<bvwF%FfGza{&MA5bKx(btaK!%vCc`i0qiz)vQ
zM{Z(uH;h1RNUp|#ppnB+-G6mMwwl2Jp1Sr{y-h<nNW6(m+h9vTUB;)O#fb$VmuV0N
zISmbTU`@)3V^n2cE!#G9{aWzQNG9fQ!3!O3@Iqz1Nj`$cbE;CeYwySNR7MMaX2qX`
zf23PcVTM#kFePAaU~^Xhg&_e*P=m?!%Z_W<6-*}8q_C$kz&#5oX&%a(?5KwOklzOJ
zkSJszWfNBcsSb&Dq5iwo+pDlTufA?;%CTIVpnI#p=A%RbDsaD(J~N>Jm?v5H7*zsY
zkdBGq&O4dRy5m|g#!Vfo`hMXb@}mFoA%9>{I)zvST4a!yzs^%zbeg^RY$!Km2T2E9
z$B2gWm3^gHdqulsv${Qsv3sid2F*WIsV!SdcAop!oR!l#Bqvq)X>T42`~icSh`l_F
zPFfi}jE-EaqLs9)6%jzdr~y|2^?(|zUMC8@knvOlrmT5|a|z;~@Ii$}2j7%dR0-NP
za0_HSKnw9ML*cq{=m=Gi$JDehoYuJb#w}*D0C0J>CUw;y%~qjG%e5no!@XO!G~uBU
z<C4RsyGGak{}n{K);H?a<?ag}Sg^TIS_$Z1xtw-q%fd-6rhw4hI;Or&9i0k{-Uc`<
zK?zN|R8B}TC%!am6yXO@PJt8(aMQy`bWQ$Wdv6|>W8S`x-xy<`WGRFUS%zduD=L&^
zXDpQxp%B_7+Qy7EiY(D0TXrRt7O7COv`|QCQ|T^AD%$J*9_MvkclVv~%*^wA{`tLL
zzqwxXdS)K#Uasr?KHukgoX2sTu)sooppX;#w2!N~lrX^Y<Zqx;4gpReRwY>2KvJGi
zp^F2LdVP*?-$dqv{2Coz{Bsv(9YAOpz9}3laI=ZFZGau1<y8z3|4as)vWTm-U{4(@
zgEF7Y*q&}W%EG2jf!uK3NwWUIxk(w$=5gV#e9`)3;rYL)jtHqccR9dC9X@Cd04dax
zjFD(CtuS;xx>rnNYOcSpk$BQ8VFdvVIjB<$12p_c=z0J`v;qLF)4+>u1CEghldT(3
zOM$`>l>iI`q%{STKf}Q6+B#b}RHfuMFJLRc$d97amFIor9~9v1{Un(H(J%pPum|~N
zwsj*vi|Hz&`g0guN4`A`-hyfqHm`Z+A&Oj9-ALNPA?{&sE2g!jt!QlW?W3!Fzk=6E
zCm;}aJuf0MBRC=(M?Lh6tbY3O;``K*9|wsJQY}7$t#lPopbo?NtS+)lWY0ll%Lm#u
zZlIqzMjbC~qK53u*poyLaG&gn`lc8|gkY}d(3YJNu0ZBn*hKO(1Ql^D5zQ1Gr+SP#
zIGlW&<LNCT<?YcxN^|evmIy1;w*H^*+`#I+)G9>l?gvVsU?m7O+Uszv)3CQ&-%+;4
zt|BP|RJTmrAhilQi{XC`U3)svB8sG1!V!%8+^Uuk!<!`*CQXKU4XF;@E0V&*nt=VN
z;{01Umv?esTl4}edRjyAF)+%ZH&;XHb>3-C{uX7F{zSbZLJE3StaA=nfokRtQE@AH
z&>T1ZtDJI^W_Q<fT;vivGS&>dw&HpgoODTDj6+{XgRO$lEt7u#!_{`V2U@@>@Y6nX
zehhiwjV7`>vVwS>?J#0!rh-k~<1PTxn#$JPI5sQr7JH|H$+sY);!;nEX`i*ozXSn^
zwZ<Wp8FC%ejcuz5UmTqqVu*6pHL_3r^+aeaDJfC+Xu6r-c$0=GFx6JU?MrPa0?)_{
zZ3hNGKBdz!gU4gJ01TK+QaVuNH*)GdjL9+kyN00t8nkd-eGoW-i-u}8g)&?<Ha2I8
zYCN8TEo>h(G6uYZ0BX^<s|B3}fB)$<ipNNC^q5%7{d9!YGghvV7WuepG-aio$$UeE
z)>qdSrBM)#&-ZGJrRHtu%Rlc3vhe$MnIAgv@-EUSL9E2s4C|a-c{OKp9Qppt&2y78
zu5qsl4*Z{wn8{xd21;0#5w<L~YS1MopSe$wMb!J`I)Y&naYhs*!I5DCDmfH+AN+~{
z77?&QUKWR*=UrS_dE>uH-FSQpGzGcJn!G9anCqO%f(1mBL{|y<6(Na$FVJEpdk@s`
zl)vF2+_rf?7l(`0PXRDhWe{FcrL$*Cyr-z|!K*)zYnQQrM4sdWB5LUTT}zOtYH(!7
zdwdp9@FW3ak~LXAv{IyE@4oCg(bBb=0hm$ouq0SyXV9Nw_D2o8-u*LwNwIugmT{Sy
z%ESky#e<Ur_ZE3nHkOcgIC%-9G8vN8Qfn#L4zk=JW|6l0BKp1CDq;~*%Jpw*vIo?g
z!IXvc-Sk0;N{RV!iVNh36Sw)*=w-b4_eG0Zix&~*iwXdYwdNYv)-(o8;<^6rYWxEQ
zTe_qubNC;7IAr-px<4F;`giCICHFhNQtht+jmKxNpjZFw@HxVa{)4Xg^Q+K$CjJLH
z!*Wzvq9=_NKq?nng1`}Qnjqsq-9MV>1uUl)leU9ms_F5|%b2uQdlNYT)h3yLVD3pj
zCgE)!5)HlqKam%2in+y4KYqM1OuZB|!N8@C51Wv+{Fyg0KfB_N6*bAR&~b7!Yk#M#
zA^%)hPAdYzVV8${C(ukrZh}K5wR&NJh2#SeHFkw=wTDB@tc}udgxwF6`b()b|Mx&z
z!BS`iQm4$G<AoqMs;RCLrv|lCNM4FkkCxIDDA&mInNDj|$eiY0e++NU6R3~6s8h>>
z_(Yiud<MkjZ(5xM0?J=v>l<=SLVXDX2IDVvm)O&(Y<}MZl!_SIulNbjlfnXIDS`s{
z=6bjerHMDxH2^F5k2}17*z@3>JlH<H#6v>cft+7(^N1e|LO^m#N?=>XgG}`ax~fDk
zAs~FK<t|?wD+=9eYHFfzw4_}(NO7%~5X@lv9;I3$z;MlzXox*HB8!|J6X3DZU{^#;
zSY)xtSxD`Ll}GGh<XhOA^d8`#fOinDQ<db}oYdeB(A=Yl@zY)D`gJ9{b%uzgck(iA
zJ5<fN(avMA8kA#5Uk4q2B+fj*0zk16N0vo+9}#xa;}e(z4G`8i01Bu&u;?P@cx<F?
z^87bZUKf8m3{*k<-uW?JawNgRx<C-XeIw#9`eI<5fRIiGR&q~3-?0V3A1xnB$Ps&f
z#~qVOApbt{ASjB2>yb$^qA7!K@c92VF4i87yg$-{<tniLeMB?4{>t!6T|BIy8nA%K
zbDE~TBz8dm#upSqd<h)t!G5+m81Z?ONP2{ijSuU5=f6+4B6I#z1x;M8Eb{>4&uYkn
zMR*L3d`9AfOoi+WIFl+~v$jUf=`;oukVO}Q?WDWO&dQ9b-N2K-EyQU%k@L$ka^@9+
zu~%>EdoV@vw2WJ8(TI4#Uo7l6lPwp7r9CKEcrcvjhghmp&)D2UUPbbgI**V!|8(Q1
zuKG_^O4f0(cDbYy^_|1La$`Cz5g$Z{ncoH%7ma}Wgwd(x#_x+{1(79;)bsN?Q74*L
zz_1)xYoTJI1>iWQW+(zN$<mB)BMon|H*#8i!S4p}wWMRFU8C@^Z0kGO5}Ew^*-6&q
z_!->%G?)Rsc`oyjvb=aYq<urj60QO=d6s|M(z$YTO_pyPec|sKh2bZ%yGJGiSSL9H
zm8?@ea&bK3(<+4N{0}*O!D!+}kQ3t9uNz=phrmIoAHi;5O1vjiwe!hEsB^$JV$2-T
z2j)ZR0ul^$1Tao<<2fo6!R#uF@lixdy_u0d@-)u{IaWMFlH3NtkqPNlG9hJ|TW9Yq
zU3Fi^LfMz;JrLIuN{`iF)Ibwrg28RrzTGegee?izkG>fR%*_Fl42erw6=JGGWR#uv
z@RduBuSDSe``SPLK|2Z?ru&`7!$(R3EPwpdlD*${akynh-txA}Ydi2x?nNg$gKuw@
zuB+L&gsp@N7KmkKYa~gI1w&FmybHWgc4<Barw~yLkj33Gmv>@t6dU=AF^Mc0KH|2R
z03luo$M>-XZ^h~Bb(0EJg4~~3Hb1gt7g*j8B6x?Vup9J7A_jmu<cO28n)p);p#2|t
zXMog5DJd$1jUIrFLK03h7nj(gH$?k_g_b14CyePP;!E<3$fKD|V<4jdelBdzAk`%K
zc`Uq;6`oE*j39F}Mr8J<ZlBcw#siw<lBKGjK($ESr}5=yHW~E2C`vZtN18$M0^l>p
zRY2B9Y)6XhbTMHUtpKzHiG!8{(ho^%KtMy3x54~EP*NiIBRW+#z*j8hQ67FBNH2pp
zyQ%1j&P=y?Mr~on@a75%T1dixql1(nk0TcqA0;Q`2nGK%WQ=5U4qru}A}~RG^Q9YF
zit|hwmr_H{qX?`k391FzthtBS+3YqFK7T}8r%KkVm^_8i`e&I*sTLPXe3iuMLgs+M
zqgra;g2*H6v&`%e-*G1>@tEi36j&5V&X!;;N-{~H(J--xBTp(%GT=nMS%Yt{K_(f%
z%<*23kRK=3^nJb+q(&T^%Mpgij{&QmR91Ma&R^iUKMFME%}Lg>WOsfi;@*NabVytM
zCYA}@I+P<;bdWtFvMFfND8Gf%i`0gnz#GIrxP8n_2IQ2?E(UMP{3x$rfhyoejCW|g
zQ51=OAXo`Km?*6<Ol)PJyo%rwu<4K$ID{<o3Hwvn<^`Ej!?`gzZN3*$)%A!3EH+Bw
zZxsH1m$%>br&<*ca!67hWMec`<T-tff{x+`!Q>_4L}r2wxD1^T0Gy`CMhSQSn00LW
zu3F?PQxWRbGqg?7-M_lR`Sp)H%i=m$L0t_TGeV0Z5jSb<0h^#3tue0QXK3{2DV!Ca
zsejY^L=(<W*o50WuHb=V`z~B{el5rmzGQF*+sV|amC0b0TCPPx->eCO5TGrjOZ>q!
z-U38K&8I-u_H6H#C4{9tEPC+V&oQ8M%gwsp-{*?()`&`0qMQA>4+G#HDl>A2XCy)l
z#esGpxM7>l@glc4Ls2j-c)=)wnv@upo?GYiZ(I3f4lLHnf4O4wIQ`cDT#m2@07p3+
z{BlM72GCrIjWtAO6i8)bQ9-2PUQ7s^cZgv+p>9X_%^(wu_;s)|-8?SRDsAk%W`BdW
z8j$A)v)ZW{1AcBOUjWWm=w9HBd@ux7i`Lb5tB;d6G4VZ)AW$?hUbD4_)Dcnk%4^pD
zyVe{AV*bpLfF0@tNqH?vuO0pCItaXCZbMHJ7x?_(G54(pat&Vl|N87?zoru#_Jx)I
zvGI9l#pesN47Y5Gu<mhijZUU~NTJ=;^8Qv!`VES{G5fc}U6qr|T~3u59<!FacgID4
z#dBl5Z8z@B3$9S@zSVt&cH`XY_wQTZ%w0B5eyqow1L-s|#xr4Gn%brOebIci>15!(
zrSAdwv|n0URi(;P_z9-JPbtMoRjUL@nx@&PE12_uHqe@(4bteta93V9I`?f<UTQ#Q
zp?XEyk_lrx&=2xY*O3iO#>Oe>8>f(3f~51n>M2`5=gbzIO%!!ffhc}(T7JDz;zGI!
z_&kqfI^`ut#qO=2BB2xF)ij(e(XgQ|0GRD5yLwWC7!e46CUUn#{lQ1W0KG{0?j$m=
z2Evo9R!J34O_iss5&Nai*n@nKJf{<{2i3YTQXO7`K^WdB;T)Xu%Ir{XNFE?#q)(p~
zf}FEy9o`Ak?!1IV9F&X;LFcN*j6Bv965Zs2Nu-p;NT{SO5BHKf?E3X*g;ZOrVxMBS
zqQ7bGN4phdksQo0@Y>#C)xDix2L?pEtx(Gx#FLuI=pS2R>-y+2w7RIb*X64n%zhEx
zd)}I#kz`~&>Z&&x`59?ifp()|4$f1}(1-A~HkZ+sd-C+exKL!_*hAD=1}hvaf1}>S
zWC&GXRt-b0nQTF+Kg?J$+q=W9cmU4no5STyf7*r&EySv?c!t)2|A17rL)53dowgQT
zI_fsJZgT+%C|UU^=9SDMvn6X_%{_v!Lou0yDB}b9P?V4!<>n!Gw^g7b%5&Bf>u@8i
zVJ;Wm-|z0dr^8*%PlD&iU8xb`3u^-&3MmD!u!vX>V=4Q*b)|Odo2CrHKcsWSNZK=0
ze}S7q17~Vt>+Yq^>~L!sz+-iDYFUumXhXavy}bO{pn;T(t^r5_@<pFXT~tu)=ts6I
zcHH)$*udGz+Dt(P5>?P<L}-H+q&ew7AaAltB}Ka9=lfmwei5L~;fvNZZHp{<P<(PF
zPs$xIzH_DYUwlyoJ67)wVEhRUw*YB{g`Odi>Z~>f^Mk130kj8}3^i)mgU$fB659mS
z^upeuWXoKs4ziJSE-0N6+c@%Lp(J4k9WYA;JY@w6LD(u}Vq7^1C*4OhGI6TS^D;tb
zf>1o9R4A-~#nxeix@;t$q7s!jHGT}S+pMe>v-{>Z`xJTYw^qFRac#?n^21M`T$1?y
z3rC}+d6Y~p2@0vLJ)=y`u!WaLsB!E-AW>`XlNJ(GBR22n*3{4>loF>{0D%s87-mc9
z;$F|jN&n2)A(zzpsSP!sh<-&<=1uO8N%B%pq+g70IWIDN`KvXiAO+4s8*=>=M;mdp
z(P1b1mtf%WyA!MdClb?-l4V3CA=-dWTmy1?6g%}AG_OL9Sz=RN`IQ^ldWWx<d5G%i
zTX1oK$IXFLzr#IUUp}XedryE=yeXb8mKatw0Y7SL|6z$46@uE+N@Unnj23Zz<On&8
zBdH`ezZ$EzYJCn^xTc)br3`Ef4vk<WDmfL)r5dJ&qD{<QR##cO7b4)1tcmFAch)`x
zDkW7MPpA=C-pJ~)IjPR!ssxboo(+^K;#pZYHuE(#neR#SY1n>^pRQN}YNcl=8lwVu
zb3E9N$O)NARmGRWk&>~rc*@}j{bUUxDWk^3+a8^*j*fda89D10znVNjYaO33)jya<
zgX2r3(B2iX9SVX^ojNt=`YW&fYN${Bt7acsc=;H*m~+Ut0Vz5}9*bmreSGF#&NwQh
z-4eBQqrJL634Ph&A6cRN8{3pMGYKPRd0Ql0goW_wM5qU)kK%LUkk=$RCEu?Jvsu+-
zo8BbJZ9Qs;=<K&^?aw~5V3Ge1LYOLjrnYpT8XT<+;Aawp>DNWmPtfS38HK7~xDXnQ
zlQ9r=0<tfh4Cs`t4qi3JP3IG}3td;F3?xPixtX6}h!v`7(Pm*_-TsF82Nz#$)=L<S
zbro;=4q8ftUKGe$TR~e_&=G2PNMZzqjmhT$NQ<_F?j_S3%yT0)gPTq<j*BRas00~r
z5<e{>*yZc>K{fI=+u+d^!g+KZ2J-$$l2HPFsvYZ&L<%G~?kd>F>bc3U0bzvk!C^@h
z%(^O=+qSMCwt@SY4-O(AQW#D1=}@ojJXHY{NG|JAN!gJBaExnkbQY=2Ky{M@VURIc
z0?*7#4#omIDI*7g9a$(t2pCGPm#iL;#3L3#QKQH~6IGN(^-<4aBrrJsrH>H!#$c#x
zQyuB{GY)8d7#g{BSFy^YU~sgkw@l{BYETETI!k~nc{HuU^BrmTz!QTkg2^>lLa{-d
zi!;n)CSx@ZosI#ZI3I0Oo5*|}qW`%}$SmmiLJ6?Sh+aJTi^AYA3z|T*H1ew)FK3Y5
zhFo`eBBI*BKEHez)N0Iao+FgF;qev-YPiLOpotOk%UZuI64>2ecV)$O!m5}i0>T==
zxXxc-$1$Fn9-+p6wK*Jt&G=V;cQOh8Z-b@&0GLSxY(4ZjE6qjuo0Sz%HvqhZk1)MN
zdAb3tS4!{5RThrx`eWQ2iMvmb(2Wue98|(*JX=(a^HdkHlW&ODi47=Bl5s*2QNGNs
zWcpqrW<aY%b4$<(KL)N5<s6aL>&Et%Rr*17B{R8gLTr7ZUyTOX{`}D{g`?%zvT~v^
z?zmz)=0o|21&~0PkPQiNKx|O6(WWDV7;kUi?#g}x2hZJ3`n_p4Az>rp^rT&2RNqEG
z!`X}jqNcj07P^U4e}}x&P=&!qo8SuEaECH*SCZ+3!n0LOEeEDA+wwy8y=7}k+%KD?
zk4Cl1W`p>O=`2eyD~@x22mvogl_9C5=C0mwG)|G7n91xgqa8sQuFyt{Bu)c*7c3o?
zlB<}ciNKk0!Q4Pj4>u7{-xX7VfL6fj&2Wrc=ojF_safDex|Eo^%x06lNG*~0hsSe|
z#{j_Nq%36FsMI87j-~k*Ug#eg06+)-KyVle4+2tJGa+g{pb8kMAe+>Tr*<abi4Cca
z<U?kzQ8Ta`5nkPT92d(%ZarqR)hB|jexJQrhBl~+X}fwtcyOx4M-Ru@H6jYIJPm*f
zAa&Ne65SAiMUC%Vq8LE5OzxR37CU+3htoJe$gvTGNR7m%6G(BWd!X*0ikpNCnQW=Z
zl$Q%cS)=+urZ+#ZaD;n&U>R7qPjLvReK2nzXRf&{z7pSTL3|MQGNRukl@&WB2Pp{E
zpxA;8#SdPC=6UN*ojJqFb`XfjOb#U#5%Iy7TtwEVhGPo6K0~X_rw&Wn=1(LWd#3Pf
zN!1b|7vS#vbb7mjj$*6<o>2lBj0jf&vIYu$L<NNB30UtwD6-H;X1zA$1a&9t8E~6W
zaoC>a%<Lc?Nd7lCOazr`{9H8Gu|bb8XGvuI|GDysj1LF@L(p8qq;Mo7+|hHMw*pBf
zC%wJ3mR%=HS9NgD0~v<tx6_oSRMlW;<0yK$d`xngpRQm^9EGA9?Eb1MYbv&&d8a8q
ze|h?(FT1(ITn6+=cqAQ12?Tu{5^&qpS&94y-&8<!)+0*O475ERr2<`BzzZ>G8QfAl
zTJtKK&B*upCnb=P9~r=4W>Qv3H?YDM-pyV8d1o0w@Y7p|hoG+m@mqL*tNXhMy@8K*
zN`q|;dO)z=nRW~Z)9iGGe|)yO2~04_tu2F?jE!y{;0!bU5@HgdxvpM0$xhv$?Nku`
z6X!((ra4y;X_CU94O3nh?EbQ6g44SjKG{y#LaaCx(;UPOJU2CBlzk&JWQN&5;b?<Z
zjhu>M7!jJyOc_B4pe2orYkWVeycY6BpusHIr3Ia2AB%R7{BW^t%F{$&h0JMzpfY5s
z2+RB}HGg0t{dyp_dFwd#H72jZ><xPAY8nFzu=7cN<0{u@!#ni&1ddOtckGwIP@lUf
zLza!zNM&FG_XGK*ZCn(drWe8hCj+ClC%RL#U09P#M^Dg9AuDER1jH<YPly?22Sfqy
zA=~o}koHM+Kjpym-AV11>Ik$AWPUdF(qA8KCUss%iPzj;SUm0n?1$D|Tv5$G%0_uK
zdxY%;Anzw}-?j+pm`~B~Pp&faokOB(I2MPf#;Y;IqCo6@N(x2QU<}jnY%TSyUZ+{R
z&TD%&0GcdR0MvRO<-e?1($+0-!u4PzQnnE*4T>w1mvYey|4R05S5re-Lnsfjwb<6L
z!F-5mYGaa~m)QZS;lLWQY$dKQYC-f@y~&0_s(by>md0Q~qhAV!7F7yb;)1!&v-JC$
z=h-;14o`>A7x0kTQV-Q8l%PhGl+vLE?;QXXH2$ndr{=|kSi_!t7=TT4KL@B;sl%B9
zy7Tx=-UA|HI^NRfW)?kyOuVQofqbeCyH4mpWeGV7cf~ibwuE#nkiZS^=)3ie6;oLM
z@KU_6rZSjQwa`r0SFbjlb5qR55@fz5^ikeipHyeQSkh9k`xZ__2AxD&3YHWhQDhbY
zDn4VuG5~!QHZ3K*g&1LgW7n}KG)eONS@}@DO%s`sv|v0_38p|@WQ(5R;_=^Vh9MZ|
z9Q1`{#>&%?i{wILoVyKYC^;hGBCk9-((Dn903OIderRH)RCk57I42l?TK`u$5El{7
zx0UXbesc3((DAq{vcMiZZEyr&G6pR*U@pS*(10gfIAW;v=#zxgo<{soz(K1YO$R^=
zl<+a#zk)vC`qxZl$OLNquLwlgz(<**M!z^3+uZD82e&;_5Xvy@LejPrh8S=fpQ=%5
zrw$JCer@uH#V3e}s0#}LZf>VO<O%0?LfP1@HH-y0PCXJ;L(Pq+FFmbo4hczyG28UP
zgLs{fz{>Ht;G|m*{Auh7#`uxu4A03PxrJIwPI>=T9(L;u(&rqfnv{m9fTk!8Fd>r>
znw5+QCjWTtdA4f^B`yXWk?R41Bo-A?^4x*GLp!*qG48-jCI%Cng<K{!$1Y`~$&c|J
zF~W~6Z2?2MB|aD*+=h|Nh5mTY>Q(o^LnQ7Kq6X?m`_<cC+=Ulw^*aeyRYa%}vJG<}
zN8aiFqx&1A%rvGpxuMAy61cGhoCQUH7^`<TAqr~BL~j0yGA51;{13<?Pr;0o_}qdc
z!!g!V13Q<d$N`0zUO9<$SzLdNOt(3}fMm6>4mLZtM$)nL+`iOnQ6)5!*@DGIq}v@k
z^Jf2jg^YMXE$A8yX}bJvg&gVD!g%$;+Y4z8EUz0*2Q+ZGr~7`f+%tZp1^GH1mL$_K
z7-S$#z~&*_6U;2D846N~4-(L7c!F!PUfDFkfK@n1?<Z<6OxM(|N6~QIrB~}Cm1(ya
z&2N4r+BgLQNE<nIBX+$|Hi2B7s6)V(W0F2Y>P}5{KxxGF0iB2fKBQE?S*m)1gzPe!
z6heMApW+0q#Ao3u)6+qx_tblL{NEuqp2D^457?X}?qvZ@#%<<)TpFbi4tb8|i0|+s
z@I}9Em6QH3(jh=g2=!B{swI#&$HubDHJO4&6YnWzDPL=%|0|5mEsC^bLVcn0mlrf4
z49FwMNz^k2U0JDe(){`J*Tb>|i%B>r51<fGiVd)Elj>gbWQ6Cm$qe$;1)MAnVUzwv
zEP>(W?P)Vc*}E6;HHXBb)k*rX9(_0y1HfBIK2Fm5Wn7Xl@=2Ltx2OIFY${*Sk{@83
zXP;^8O~j$&Y=S&Tyj1!;0;FWDBjUsUc^y5ZDHm~OUNjbgNFe}FaceR3Yty*`<5LhA
zpeIMgtY6c-xG(camV1!_xOC=<ohoj&H`w_P0BuFN0$^yeqzqLrZMiTANe%f9Q$Yg+
zvt9FM(o$6!@SR&TNup(&dKU`k8Uj|h#WX3-KvdA2(rt%-VhA32X|WY9YU1_ZlryAJ
z<D>6LHhGBZ)vyuEoOxfu3X|1IF#-I!CP3hX8cB{x$`nk+Q$P8$gvyBfn?Rq2-9fVt
z(9%bJkAj&7{$U;H(Xrhx5p9#aywyEA>k+0ycKNA})RLwdUTk^kZj_0QbqF?IN4|$X
zo6p#`i02lC+{`t9UnmNs^fEg+k+>8*$PfDvx4v!KU|JU3<eEfHn^8qx&@nrWrY8z(
z6#GYx$k3iJVpFnqlsoJYMlN$2Aon2M^d`KU_0i3KgXZzz8D#gu-xhrPasKH*;W>)}
z?*Zl={<(Bfq}m8~13@QQ86-$bPf4H)=?}_!3gbNzN4^8|Ae~7~Xkrf6W}T<~eaN}G
zI}tt^D1-SQsYh%J=Qs<Xq4PPRzK-<4;)qJ!9_uP*Olkmx+<rBeN~s;1HmznpW@-Pt
z$?xbtZR5E0Xjj6J7Yq5NwJTvzv2(gFpzJhI0*q2hU#N2jO8TwE@;iavGyr~fU9qpX
z>-&{1I^5ZbzNqMi-s)<MxlDgq#Bf8=ylD0jv;Rrm0wgeMkLins34-@=#h!wCPxNts
zQxfxw-8A04#UqVE$M5%R(BhPm<A_5@0g#^37*j3;2$Ralvs5axo8{d;*@X@c5o`n)
zsVyD>+>?>Af`*Sw(F^&#Imlg{=P$U`$kDe&kaYvKZ1iaG)&|rsRFLI5n82}Q!@0<U
zW};b>r-~ZC7ZkN<xN=|RR1owd12>^{*oGY=5lww#d@rL6l*$o(I>$R^f&vCsM#iJ>
zZkdi#F#HbOB9>&t=ARSlcjm$gAasEo$q7f!JE=c7^*Wk#O4u?!H8ooa4FEr<8cn%H
z>8ck0L!l3Qo_4Vc2G!(~f>=f4gh`u*grr?9478H5_W;;T9HO6MZCA4%yU+Gvh_B22
zp;RZh&08kgG@*^UH|F_&r%!`SOhJKxJX4jopcXd#PR;T#CL36jMr)O4nxo<uG+lo^
z894okL<;gad8vPZ^YL7Pb<prAM13OH0gF<3d+Qx+VS3n1yv~ZVkUQX&Fja3KDT|QZ
zlYce2@T8tbO|TEFs~feoan0U>rNaFoessl4g#CTBg17ZXraw}TP)Fn@nEpU6=#_95
zkngAIC1?&1*pjW@j#NjoYk=r~(Hs^kdq9Qp;Yt68!g%v__cv{!35g0>`)$8_ufRG7
znH>c{Xdi_wW>pK6CRDSd^{v5&Q%g&J8rZ-i4?AoSCiiKZWjn?@{R0q*SHBix7>OAD
z&x#)5%3}GTjKIr!c3jKbIOi2zRA>cox$Z@a*<`=<25S|Vhj#LPcu&epp{q_Rd$1mf
z_JaRu9wtQ7xv3L9hPKV6C&eEk?%K7dzEbi?4RuTw9>@oPLc>BIV;0!3O)(b#v{}Pb
z%@o1x5HywG9B-?L3tqKV#DEFFwH(D&#F!Ic^Jw-mqvqCEag8f&iF)ta3>ztyS*z7Y
zJUX5-$6jB+O}j(ig^(5$4S@Mx?#|hwiqP>_JwW?zWMoNWpx2~2qL17_UNh|aIwBq*
zaZ@QtgV!D)(Guk5PgqVJN~1H=cU1#l{iY&DV$M-GG1fKUnZn)VnZ_5@{TC7dgn+2V
zhm0d2rHytn=e&rCh=Hz4{!dlP75B$<yCw<B0Xd=zQ~tI2<)Rk0waE(=SzaO~*d5~3
zPz^$dNpqnzy8diyn23##u>rVxwK>HlMUF9bWnDE1EhobH9>a?$gllBts$^_a%Q0ht
zVC7Grw&^jp6K3tAKxcSgaB&9RDdvm^kc>du2E*+r3-sn@3V&P3`p;^1e{9+E+7fP$
zPN|Ntq0&47LfR~*v*nf{ejLZ2UkrPgq#+K;L;awb$s{j7ybItFCnym8zo7+T#*0z|
zjRXN>{-~#u9N3EY_C#x8G*t7a%GOVQHI%lyR!wN#!=NL)d|k2g68s@bPuS!#`GThg
z;{pj*>R*+sY#);DAXaSnV(I<l22LFnzna!E&6dmLf;Da}r5EvaBBufz8SW3nx+PR4
z-9}!{elsuycFZtsgE3QaircWfUcB4Rd)tnt9wXU>gy6FT-_HRzg--+J*@i~_Z|#=C
z`A~K#ko@b<OW_?T*r6thdgRtebrNIsQ_RRYdXR6F#Pl!E4N_X5weWub%zGZg<o(>K
z-8#_0lp@hOAiFHmp<yXdr`&%bdKm-$qRe=a*YFt~pOLQBqg^NSV~BY}Qz=I>{k9#t
z|4YAZM>jS}WlV-Qk4xPHXDpIkFJx$?Qzula5bcK)Sh$%)?nD}KXCgly@F;1)kS7IL
zuZ95~P%v`tm^-M(G)QpiZAGnLQ$Hb@YWiQ|qw#?s$?B1%&>oXRSmszVkeu3nczM;)
zb1isCVvxr-YpC*pWowD$nOFA@EpjUC%eFarwjX?6KnLEL<RA>>MT{!<JxDWzQY>)z
zL4lGEg<T{+z*R%&yw}-tEtnb{8yICImqS{YfR{vb1k_}<prJ;*s+7IS<Nhfo%KQ8~
z*jR`$PXyDy2&fGQ{E9Tppo|HtR}4lW6o=DY5vlr$>>A*FOz0=#yT%j-`0x`mJVEr5
z2SSE=ij2wu+j5>o!05=fV&P@Q%1JbmhU=OCsR#FfP%0F;EQ!hqW?Z*Js5W*BEsO}J
zUC3Ua6%dj{M>SZ-DX=>Pub!F1J2k(7dVOuD46!OOrW9?yWKJX^gQ;v@tGwG$rsjql
zhM#8Nt6kjp>-2x%8^)Dpu0_7avPggcMrVb0q_3bV^Iz|cW$OMO#Ui~oIqV7k0okzp
z{L>nkQrQ5=BVML$Yw`|EMY6@#$o7-pPu%J%N4TNZ*GCfQ$_@GXD>$7uG49%{4_n~i
z_?@oGI~c@-l7S;!(uf7YwG6-q7~eQe6XvnZ;^`vmw3y9J3--NMx&VW(9$SQb%L33>
zq<Vmf*QRV|OcLr}a2?&yl)l1@P|&M=xgmdTS}TmM9LZ5r&6eCd!QC%8zj4m}!gm}C
z4}Ls5|FRzErNSj%L}mS-7>AX+(~6|wxe~g0OBK8d4Y{vsT=vgGKyol+G{rTmz|`+7
zn*T-OPc+w6EgLpQyd`EhH5G}wLnlx*=Cw~1(6Dng!ACMvp-hsFwGjQq|2Q`+c!0pb
zHPN|l2|zj=p#U)Tk36(NIK)2yFQKPk++tXI9?NXCSRL02;tEYlBJ*LWC(&-$xzaTa
z7U*iMKE};qf%uuNFGoi@E5f4r?le}KW0pgX05UH`tZ3LggE58|V+%{AC7W{BVNezS
zqeTlna+mT5mvMrMxRwYk6TuwUOtJy+bop{Ynek@+93+0|@!Vg6K){(sh=>AUQ?R^4
z#6Zj~VWvwSI#j&U<^4+Xw-ZrYt7i%P$t4l(uV8*vlI)_(8ET`g^R~a;s?4g)MwC%H
zdGtq~RJt(hKL>jem@rk=$9%uSPR;7Fo{l^Yq&YiuUm0lXtS`Ww-3~N`l<U%w2902_
z`|I>$ogB$666X(*dWmb#Z&swU*lQ7d49QK*KDCIpA*1Ccf2>{)b`yV%VHf{p%P<FN
z$eR0`o?C5qdfp3-h%UL`98~h8?O+b6|5&eBXV;(?UqaxNEv2sR?{L5{uICs<8-Hod
zeO2PCk+yvExKn~}slU9)deKQj)^pHrwAI3;AQqFuHCqeQbXm@l9gF5Yx@YUND{5mN
zeFo#3Z;55$PgWf?5tunXQpK!~plCT@@F%`v3eGEZyQ)c<#4JJ9t%RI4K)03pXjVd5
ziECYDYc4rnklti#S9<LNF_$s?U++i)?pRqCh~nQNkgYxEqu3lsl9Pt_Fju7s%XjDy
z!im@J6q<M)M6fp9Z3948Z#XC8+Y+mfNUAzD_s~N$NfFe`3PxZ6i{_?x%weY1Zso^Z
zpqX66WULPHbe7kvHg~Nx&w^hUsr3oPf!ouE%1WZ8Q16Ig+`?uK2i6Gj`pzx17LRZk
z<3cMTc&9lZCx{A6xLH$TvbY3P=l-FIgvT`jtr@OQvCHKZ=!L9ii{rt(%DHIrlba9;
zTJ)+RaaW;=761gw9*?2ZqsPR0(tY-4+c@!l;^{Fj^#BC{VqMo1{4Q|`TT3`dh8Y2m
zq<g~SNC6Z^@)Ke{&jg><%$<2mG3~#NEwSv{ZZ3`8#JO>r)-{Co$>5aO%}8wOg6S39
zM>iLfs9`W+C2I4d)TSoW5HH3D6JFi5u5Y4zpGS6A@_1M&ckIxe1$~d$T`=9K6@9g{
z%KaZMN@*^>vVXje)rB732YrK9{jk#bYQfAgkqTD|uE{R8Kezkk($bmB`n_B_cl_}m
z66E|<9c<sNuH0H($xoi-cxJ=s(`q$mG7^0iYb=thu*p*On6%>cl=UKU3r|ba7+Zdc
z_1<gOYWQhWC(-SX_xV9MAOQc*EjD^bm1&Vj8IUUbifpCPOjkhvgcegV($7Pdy|PD-
z9*3vFVA=8ee?4#BJp6THqR#jK8jSW$#ny*CJ*TSC%f(-7{{cY7+n6({mxP3h@?d7+
zPx-z5y942QwJk!%em=5E1O&gq^eStXx6SFp|I;!7wRa~5a^R=%#OzGzN8FUUO`(5t
zm$S;$sc)|oIerXD@?*%*YZkP7iTvrHE*(}ox&5v&a?l&cnEQD{ug#neFH?FK+U+pF
zp@XdDiLnIiRF#ooe0x(Z!VSM-de4+X{nf55_M@3cvcKKe{_PDA4p2IKeHlQSLnQUy
z$Cz)gE&ie6wa{T!he!@^@ozD$P|B3!T-!dJzLJ~Fa52=I*E1QzPVS*I-^sB8u<WZ&
z64MIgn6Iv)E&d$_TzR-pA``};CMGrDQQQRVo%NXs6f&dQz236Bxk2Ezy++e5-8=Iw
znsiX_t@@G4_;dRG#?WP%*XOM%K&bqHdixsGaXGIt$)@oGCKv>t`&l-S`MGT#abubL
z$^OFOTPV~TKR=Wx$8J@cJlRG!V3(6qOdgt@>Q42y*a2Kt^O`p>XE*yh8xH|Z-wA{N
za<pn8mUMWVrZQ#9G7vpYoxxKKx6Ie*#J<<*cFRW<(c}uk<SQ`Wj7SQ3_|T!8mz9=T
zw*%okw>ZGL)&SY!u5iqBn#E8?w#|FXgb4O+-7euIzly`$P+$KFpt9Ge70o|1H`Lg_
zd$7kRARrixvgMaEu7Z@BjUZZ)6BP<-!)BmeugdIt_NLc}T-WZwKY9xRXhnWp9Lga<
zglpyYyRWu2=`#z}s;<{SW(}}!{dv(ywpsqgUHls{f2lQ0{8=%7)Vr(rYG?lF2l2JU
zzUXiNz7%f6zxRq+4gW3%x()v>hJP0Wa|ixi3~jgJ-^lRqV)%D4wB3P!7em{v_-BzJ
zY$yC>2k=TSzkmPU(7<5Q{sMHW!o$MCJ{0*cyo};{;O`yJO13(LZwWWt+tDqzJMZHP
z-kY1ecnqc&)I!z})KCvX`g91re(bC@!eLxff8O?Sk!r})gMYrUZtDo%*?gdcz25!U
z3_|Ev<n}ij>#00~jnhQ3CAixk6%t;SclI5~z;~5o1N{7MfUTB#7JvD1h1SD%sW$WN
z;<LpQb>@$F9mVS$=8xR8#H%UhkB0T$$o{nGf4lzsQn(ra-YaHBbj_a4<E23HDDkSW
zu<)8^K7h=<zy#JrpeDXW<5-(J0c&zjNg6$BR8Ner$1Z5@#BCwbnqvNIg^D=9!~Ug3
z-+$L1?!^B$_v-FD05KLkuh?FmrYX3=hQq7SX@0O%4+n$wA)fw6{DfO>(x-K!@bBq_
zlQfN*XocE+;xK6iqbqy%>?zl%1-$wC&6_vTE&}NZt8WCjG+u>ErgtM?7K^oO&+WXI
zCuqq5XpNyh&h~<qx=b*r;HZ#UT6(1i;0)!nhr@L3?&davZTREPp75bi!U&WHK>6-^
zp2i#2?*Gx&1S_v}bpAe@{O(sae+{bBl;uTQob>iCN9g_v*eR${5nl02E6{Apy;rv_
zVilMiHx<oOUxHd-S*8XW4J3ZoVc%n6YWkovu$1zg*|{TvR^KzMM6+JYlZ-;bchzl^
z{+Y*n<D85Qa7%Z}siwJr?=<T*jyJ7{cMUtt<2CX@`nU;6#YOyT^t-$!jKVa5Y(1``
zE;cM|872=b@Sncx4jh(0p!E`r`FzsSc)<~`Ukwe3j4Y+wcQ-9MpL>HGjk*)Kt9MiC
z&gDdOY+Z=p-MD7MhCrCM8k;zOUH*0Fy<T7Q1fVFv>UWtlxEKy!&<@?^jMmy$j1A3%
ze0}qlEy;U;E!FO~57n(7hRGRkoRuGNkE^gYDi~iSXMC*0``G?g&a<%B9WtT0{eS^K
zSHR@AnWZoSb`^Jvo;`L29-aRIO;{yO%}P_;{9OsrStFc|{Dd{2A+xq_Qs36Zyn9^?
zjPvV?Tr09dpS6s7U~_EOnqG;$PY)ZGFt-2urCXI&ubT8J+P5n0j==s>#e9pTg@Q7Z
z1&U=o<vdDfDr`KZJtm^7RR7_fOcxJViR!vVvTMY>_%k&v4^3(;hVyM}ZBlPP(-^AQ
zFe28Z$x(K7fJCZeqXs{m#~Y?==Vv<-=j8w=*|~W?*eK#o9jrHV0vET6=8ainAmGrf
z-$4#ez-gOvE!)e>YrwP}R)+>DjeX=XM5cfL{x879Skdyap@v@s4<(Pidw-M|%Hutj
zRMGMHJVTwwJL0@^X6=i<c-|hRs(gCxKR0L`kLgzI{kkJ>ObnPok06F9#QK_krwg>y
zt4Y=xGU~+KMBKKaaM016e`?~-PoF)DTV`95e7YVdKY+a<)_~&tG8^H!Qv~-X&&y|a
zKJA1hZ=)TNCaEWTaomInp5ztYq7+tp1b=?S`3s)ckrCSlGaum`mR?M^u!VZczA2kK
z@%p7gCb$kn+7~%Fexa8@ueC-e+pnA?KFe0D*j}zxODzhJ4NiudfyG<}6f+6Zf>rh@
zCr_T-fhMde?%2^UwKTi;;MB^PjNjSMk}3%t{6IwB&hlsreEI`u7sSuP61cr6B28l_
zR_$ZSA$%ij!7)GN@AFdm?e)A(9eDj7CAv!&K{#+AK+w!Tuy*~^<G(EV^;g?p>r6og
zKd`o###a%;|K?94Bqaj_1Fa!lOr*(AW5$n9F^B-nx^eU7YORhux4|!N+G?j;s((0?
zZ-ejun0?O`xg&H{SK{5(AJm;IUU0Ue=SH4e$2S0biJf&AT(IIBdY3aoc2pJafc)cv
zNt}zy)(N1ZNs-xBl*YE4$Q@r^=#KDT7*XB`!T&tJQfDY^-e8FDJ`9>I-oMR%gqOyg
zOpFPXfEwVjhKf%66|jq00b!mUdhr(X;5K+*@7}$B<ycHhR;`kO8&!1Kt}18^ijC+o
zk^srNaFs=}kJ7kN3w(~-I)f2y{q@6g2}I7vlDW#d^$T>dl$*Z%@$tfFZ2Q44tW#<G
z*H$}^aqPQMw$mWq+5L0p&MkzV@*w<@2WU9eFA(tWS-qR-1aXmCCSFhG_aA$WeY~1l
zImeG5H*%;zYb3;>a<#-T;d@X7z&F{&eUmXg*@eV>6FFwzTm%)YO4_wbI%MOE^K&j|
zOfO7pE;kXVPd=S&pZWCZQ+ofgZ&m||_k=o`ZzMg3*%Ya254x1$ziOkhLOstQklARj
zWu8v=C-iiNY0~HKdTIXYKGe-Bl!0sDk>aHW1_PZNzx-0_qG*=<IEs?zM+iUD4l#TB
z+2D7-+tDq_`Ix1<2rsF2exlzv96Eann-*7(danNC6rNUx6DLk27ACqE;debC75sWd
za~CWCsQb@vB;SJmBkS=a#t!3;JxKt?a{BRaUOQR%+E&bK4|&hLwlcl;Z_Ab~(?+8M
z&+NKxYlG)twLV!}Y8x$8mtGy9>Eg7cV*w3cv~9s28aki$Wwf!!W3y1&ezx~k2Cp89
z5OTyh+P?kG)@7HCKeKEQI6T;gxx=&KJ|<-N*lq=hU1w_PK$UGqrgN~oi}!yzIbx8E
zjP1~RJXCI#*fnvhWXS+#m3Vv>trDJ;g3*97yJ+3If|pl+Rm+@xc+m1N=VL+6t<4P^
zH*Wmw#%?JaZO{u-!>g;S$Nch3)n_cVj~O@HKRa$`Ib&z>x);LwSS@qx*s%xiRPL7?
zk=+u_Jd$xG`!7nQH;iW%Z%$9)-DIECV_((Vo2+ohd3xG2mTB>6g_YiNm!2?cp_z=j
z-NqNSLnp4$I)38Bd!yjsV29tCrL>}LDV-L6oOjGpI->GhsyUhf&)4$P0u~v?L2rC8
zT5#m&S=lYw%xAg&o<0lM)6X8V;VG31PkV|eJZa_qpl)K=gXiYa2)2o@?ahl9<$mlr
zVDZZq@Iy}7PWIEn?Y7Z&cqUoVvOpL9rd3evCPvxv&!0Q@^PoY43gJ0&bm8R-c(jus
z!HvZGn<ng+Ja8-mi<14~S=V+IccneW!vbxf`=<W!!JZy!nK**_UR`$WN>d$pX~t{u
zX|P;OJvKh;QiiB44SvKy#A&$?<>e!eJt@BDGWNtlh?$b`0O1((X+Mtgox?g{d%ZYj
z6@(agdwJb6NqX`;M82AV=&0D~W7(&MgKHw?EV!uU^@Y79cs#G2j(kEM6arp8&4^te
znx$;4iQNEZ6&+Y1AmW`;-fEbG^lQ@k$Gw?=Bn+`**bpnc^NU|<rQ@VLUum*>wZajK
zQq~#8bk75qUe$+hOALzKt*yM=uj?nDrbtZs4l>(6rRnN|#k)g``3iK5-udvO+Jh*>
zS+|&Ob{Yr{Za>XaFD^Tz<rSujb-6y&pvw%b8(ll!!AD>E&UAMBB!L<{0d;Z_wi{~e
zMa`d7#^S!;W}?mQ6SjeZXY_i*Ec2Qj`_4I;-jR3q0@5%a+nYCT+$e1tU2!zU>_T8*
z;sQKmwexR__oEq_khx+$##4J>bo!}S1bOboma9A#-Q2)%o?Gf2=oA!MWQ#{H@ac!V
z(e6U_u(DpRW0yBOE^esZxM|Y?yX?2v?dz`x9lu$;Kga8m4S%Vj;UF|2J=Q+ls}twa
zavtaLi}dt9D}#&<U;{kD(=9?X?q)R29ax8hx(aJ@73|~Oe)8gu`GZf|Zjh^Dhf=!$
zr)oW;NUL+tAlG+piq_kD{K@)gCoBtI)-{U_<J5*nymmPfghhR6q5mKfMIMsX)IiSB
zw>*toVvvi<wJHbh-w)T4Bxu7;p-ybQjfc{OJ5(hTi-kKS00ByC--ggEKZ}tgN9yEj
z<`2~iMhg||*3e^PN|!Q9o$yUOBuC5+YmA<?x95S*J){y~T~UNd$2v>`)QUNp%gF)w
zFo>Bu*N8p+-J+)-dt8dg8{#tS=){omn|zGU6~Da@7WT63F|n&F5{`I9SV<Nc`$}7b
zVVs(PrtP2}(hA9|?!q5$-!we=rc!LHVXN$441Op466|)LyYt}$tXAypw>e<lt7*n4
zW7Isw@ENtIr#ft#qOAM~x<=dAFEcZfFxUCQ?c3?IdWe>T^f!C0pVATFLH(QsXrXqK
z!+YxHOi?R@!$QT&dm6CPQ?)A4O7$^m`LI%B4p<4oRVAG)T2<JxT9=!^ayn7htN8E~
z{DR8W!+2J<L1++5V+T*iSv&m1+$cR9*ei_XS3Qw5o~5U!M?xU$&jt7W;*86%5njA`
zGXT>2T4_9(>^a;)6KTMN%i;x&ed~AQDcVJ2t8ad=;L@T14Yf=hHwR-~(}zMMV~>D)
z<YsYKwc#^#baap-J_1R=F83?aK~;=#nSXk6q6u!SFzwuF<-r7>$G+90darfrjzFwu
z*(ft;(7{*%xI)?ePf`?~=}|v(!-frojd<kK!rC5-F1Mu|f6e=2XqwxPyt5A*>na0m
zE7Ak!VR3EpyqJkP<*}qqf;8nvkuJ3#4J5`>w!&tZHh)lxua$RP<y;bbjLi7`$M;4S
zn>OkBm9FTD$}%}h8R1I>IV*i5Pdvxb3v*(v-tFFp^|fAhI`#?vf(bgqET8&5FJi-H
ztV#<~r*R)z^)lI9=Ew-c%h;^AQ@*~_+kF8^W?daodB31Ei;3-9o}Znu#L&>kw&i11
zsLtCa$S?3jPT4xcTV-2s=IdyW`|>GwD&KJ4Jok2=WB70;->AM6(`~vhv(-+VJoyNZ
z)HVoH&ZW0j!Vh$05OR+6UKFrqdTiW-faPaVW|u8mnl^t7mSooy-PSWHbX$jc<<$pF
zbG*H=X_WC@B)vGQv{bMP@!ik&$0=qLbi&(v*P6Dy=y7DkaM4W(rJK?(;nbvcHbFx)
zU8YSwJ%35<bkI0b!?B2>!NVDbCwcREdb(Hr4E24}#h-86C;5R%7)+t6g$t;RjQeDz
zPhg2uAw?@O3YfCVO9rPiKKhI9U?R?$HfPY}_4~X0#WU)c0Io$5>}j4vwZ71m?eCO^
z<URrP-_D5lsD`c$rfqO}QGf?=l|x|S@&+Ci@F6*OW!F$B{H>jlQVrD?E#zo@b|DY5
z?C3w?W@nz(x%s}z<Dt>ro*OQq**sB7UcWbDUT27K&P+!F<vDa~9~F9<xnp%>;B6fO
zRyK;0=ueHwrxp83>{c&?l-9nm5qE9c4bdI3JK7eAw)s(z+cO#2uAc?6mkFgWBk?&C
zu}E!$@~xAGY{Rl;ULQGUadU5D?F7_m2X9xxW~C6kWVKB2#D_rWT6YU4gs?);VK)WO
zv2Ua`bIZ+OYGI9W67P)!aQz#2dj6$hyG9yckBKoFhSiT#@BMyg4Rqlr71fGxm536p
zN3{8<V+Wwdl6#Gtkn%flXEt8hXiGm8#;bxylg@VFjk$tcsKn@caIhx^jX3P^1X(8$
zY9*VXG(R;(WH?1;loGtSX`^jmV1Q}pCQqVVkd5iX=@VJU?usZ%LAHO?PyLFlv%WDi
z?$gh}T7)TMkN3jN$*Slg%KDv`5ZNBlfQ%qln62-Jsr%)WOq}M24jw$HiaTkasbZNo
z819LO^p1@RFSiev>Y#urJpsi=6DCYp)KdrHA~VC^R{LD&E1a}NI53_@wfg2akAiLE
z06;UY`<h}A!FK@K=xYJJ?E83Zll8KczPW9A)j9M811gNR+1A;v^DpChZfDAn09xb7
z^b=TY-n==}e%{=<rZ3LxsAVdip8xXItC2%{zPG_`(1)MDQLK}ic?u@)_U-gzPpjWM
z@XpGVrK+8Op0)Yql_dw_D#i>ij1x4)F=apSpzZfc=8x)&S&;ilEps4FX;Zvghe@uj
zP7-4%1TvXk*0mmH<CuDFQE^W9((dVoJYIzZd;}vgXzA_)Eia&S(>!D~^GeE5s^5G8
z;fYM1H=cLWp@E8~K(N#V%`neBI8f0n0lXn97U#yS21Sjswh)*RMQ*P!OpRW0VRj;-
z+4NJO&-HpDDTo8(eInS5K06kNLF%rG5oS5iu1wmR9X4jdgtT3`_~5cG^=q4}z2gp{
zo*DJ}<niMP0WDbpI>D>&P4g4&hd*osjVS18)fF*C)+8#jx{!cb14e_pO&ow-R5V7>
zK;T2DEgcTz$*+!4#Wxyip6d6+AQbNoMYDwIgFm`W+LV3~74Qz!B%?9I8)N4Z!EP!9
zJ{5xcHKy;!opqmv5aspw$&))L;_mx;jfkCSSNePh3_S{I;(=-Io*vSL@RZmO2Cm|$
zMP~+In0@#lC2Ij1&gWxdrXl(lp?)TahP>X%LKV@|@av#YAhMrA6eE;?J685v1szYI
zWX(=9R$`=O$06%brY!CU$g~PmfRiw+uMjn=S~;RAX_I~$I_=G;dsIxr{q@k#v-WPp
zDa(LOl!*JO*^VL;!H7FiQEDKQN`pf_?r|OoDU$;<^E)BQ5$;gcm2(Rz)j6W#ziP&!
z<COF2)5uZ%J6jBT4pzOjao3j*uf~xRJd&N5y`3mw|MflXq|r3sU{0VqU$f}a(yKeb
zLM_5%wYp)V=z_g(V@<WjW8di`EIVdW<+XSj9+5i@<<{xell~LQ#3@zl^RwAkF4+&%
zND*{dDg52I4;p}sES4=_&MYUlKd%E`wOQIz-*|6i*t{`!P`oFGrfJAfDFoeo5}?_s
zcJ~k{sCY>BpQ<u<-aI7;b&L1=s_Rl_O9myVUvz0kMUYX{(pwf1gH<P@LXbv=emoev
zpUa03R-(9!h`%CUnmM_3kV_>aYS9QaT2zHVqdh^Ao2~4-^{die3E825B=(as`^Du&
zvI`KV)Xpar?}sxQmF$)<39Z7jOTgaf&7|cp=TFH*cvv~#XUZ=~w1(pNKD=z~`?Wag
zxB&McD&4y+F^f!%lexXMAp)PWgK@OoFPKEs=j+$6RSo=rB;-=(@#u%gp6Z^7O*{aF
za!(m$`A43zve>ifQ|Qnu-Y-~$xD9kuZ~$+W2m^{jHv1R=!~zh+U_fe1R;-Xhns*qP
zQX&im*WnmkvSi8HMrDdAY+99if=NRj`)W-cfRn%dbwqbAvf%cY7bOVouOs`AzkG8!
ze&^Nx{jv5p>Hp{Vjm~)P%EnwXwbs^3fd)A$hmP`Jn1mpE-GKXRm2LnCxa({$oOpf7
z$AhUbgV0wA<^F2haGcWI<<@LyXjmBNz4{=$W!?{A*{Enwri@0u$6ZH0!kZujV!s`H
ze}^g9sC1p(Hy7wNcNN#dicogax<@@94&?r5XJi;MIrfV8_6N>*?fUg~03Lm~-_p8>
znt{7q@<FDT%DB8ph4nN?Vv(<B(f0CA?{Lso!OfrCY?ZhV;FeE^2X|>FkfwJBSi(g#
zzT5&HR)gW#=?-PororsbXh|GJ+~=;HxA8ZLG*DRo_IkbRdlj8(eDe=sQtdDRcdX<Q
zU$F~ibL{0dlxW!=!Oj&HRV|gSnwDE3)sr@HnuHUGhGhGea6kEF;-<Endv>%Tz#e&n
z+j2P2^0JYACBlbk5HRm_RpHBy4Q#v5U$mj5+Wg9DFIhiA0JoK8Nh9=0Q2T_t>Td&{
zO{#;z#t!)73Lghg^yCX9QUWwJX=2ZSsoQTL)nu;Tl9FOieZs|zvU(bf@aFag_jin<
z2|v|*M)b@+$H7D+@yhk<^KQ`aTd8pvRQPjnB@K!w9Ie@)XPQE2G7aW?>r{af_7QS`
zby#vkaS}3DYyzq*?s74|zq@mlgKM3`0VSN$?ljBvAbx542MZLY2S}8<#Ka^(zYf+1
z;SJfeYMYzl1iN|;m>55cdyRPez_tbBCQw4Eb4tT`r|AM^-@~Me@^8&1CtL^BhwwI8
zS5=sJGq&XHjNc1EfImzGcq`$D=_3lv@s;qMd$Otq6TT<n30;m2g7POu3BDZe>FMb!
zj4iXnHie)Cy5qyu%Q<%3=cpe0#4OZctMIE8xcY0lY5+U#->X-z1NJ~IJ&`bNe`>LL
z+cpnMz>%Dq=AwRChsKV7wGU-bSe?mHtATrAiPmUw3nxczDH2uoiWeyD_*}39I)>`k
zH}yl#XROX3lgyvYPpMe0zTC^4`?0fpu?0He>dl)et7_2ZOhWmg{{Aa6?e{mY1_f;|
z;(mP0AkiYJ_ad1gjGIsDXP_t~OT!l?^8k&KPgaN-vg#@f{EV{Zezo6W(a~kJ5+`S#
z5AqJ_7JK?}1>@-B`sCWC3;Zqz$1Zand?b4QK^D+AQ=*;VgL*;}=83CbW-eK|(mUc(
zbaZwvZX2JyLMaD(6-4jtIk1nXM&ri3XnLm#B_Mh~=fRaP!3PQ9DMfP+8SaT{EfGCY
zt%c#?;f@xx_#=0I-MpmMB-CsHV5?EDm#tiRcM136{ivzHUBzE8(MNIYxN&_gQvKBW
zqRturwAJ3;{y;gzT*P8hL8V8R+j)Xh*vm~@NlD4k0+T3Ke|FY{sDuh>(-c577tWu5
zK#&|=XG=xjXV?qM*&xfCdxMO-VYYhvY*TA}lBNgVapY@ce^E2o-{RSI7d@sqQz#dQ
zCtV*4iqq)1DwM`DpN`5o9e1hk{?_B`ckU=#rXZc}ekUIl!a69r4=~HxPQ9Xpdz<IK
zh<&<X&DJ7Kk6(VhB=?%e#zTL&`bf));MGG;&mR?X>DslYhq<qQHba!KXkCYUcJWOE
zNEBz(BD&8r3c&Kr`wd(N=2|=lg$s8*9}CaXhzqC}LdZXxI<7xe*do!|0WYJz7T2y_
z%K;9=T<Rx8PbcgtW<CsJ=7j6pVpG?@@N{{3&iTWK4}CJxo|N^}Y>kxRZnxIu$)X3e
z;yUdHc#IzWsp6=m)<c-6N#gLaRpj2JbeY&^+l<1_lIn218#KEMrN=Gf%zT*q;Mf@*
zLQC{8_iJY}#MavN6DZ(7APCSumSVOJLqaN|TXBw^j9iU=9U8#aBZO~eC-!EeNd9`<
z+k5nDnm~AH{-}%+i(t6OS>MBL7B0(zy%R;>MR}m|7X4``{SGLBB;koE8DVA7*i0Vy
z_Nx4yIEa83L<!esPH(X{Gh?;`v}-ReS-eK&9W()l5gl=|>_>Bxc_7-Q*;^2$Cjs$I
z@Bu$#I{Gb{A1s2Hj}{$mSj4?S<XX|XIh#VAR^;M~Zfx4T`5;^y!k1_6!V@h*5C#@=
zXn>@<)cEur+<n+KK(w+qdgGao88>d;4VdHUTn}1~Xb{=@F*I$qy8mlotZn}bh^F;q
zaK{D_4R~uM_@Jmpqh;ZZtJkijTM6H0jM&>mWM^l`DB<*)Z{&*BJqr?5f-rb$@`i(E
zT;#EuCH7eFjyH;&IBNc>Bl<MBd1Md<_lyjJ+k}O1F(XM4J||w-Dilb09lG!PX~;on
z^j157AY)OL5zsn?p9cAl=1|vW>%&~2D%dL)L4Lv~J_s8t{ZM$Jm$%knwqFuTbDP#2
zZmDLh7k(X8(bp+G0_pW5#F@Spv^&veohNX@UOJ$;a{2PW``mBXbrV|xYq10ll!HX!
z4lqiPr+L?@!3Og%WYk<Q-M&4!B-Q_V)OUZ4y5omrR1`wY;FBCzC&;{E&=*aaCk7^X
zrx4+<3>1GS$i)!``dYX&7s=|5I@XXYFoBnCYBsE0ai(VIbH5bbQ}l(#?8javCPS19
zRPb-G`?R&Gy8vT2+t&)*V(riVqMO1)(PHgx3V<jC;?)EnI1lJ(Ys1{p=<fFM=ElpH
zpV-TVgy(@`(dz!o*RMwf2{6*+Fo1h~`eZ)n$s{0;Q9`SYE;oflv3D>7X2qg}Xu5i$
z_mw~fJ+I~A$dq0dX7=6pxVZG{m7t)7H!vxt!8hUSwQFM!a|zMeXJUu-?rT9o9>_8L
z%0d5n9JL)=hI*JvT<COD5KF)Nr2{W5#PG;N#1PrqnuU#XJ}wrZ-<|UCmUEDOjt(*p
z&8qiyZSel7ViIg{$^BvaGWIL+jI}u43-|x`d<hL`kteTx<(B!FQbGyyi#S)o`Pe&@
zM^)w6rrEn|DCplBh8m;vH||@Ob`)hNTIa^&!2s>ey%a!(t?nAateuzrOzi%l=9q%d
zn*Bxb)vH&7fEm)I<-I|aw(C@Z(xW@}RkY{j`LGA2#kh(6F`k`)Xk{(#2NKX>&}q&t
z52u$B9{k>BVlVaUs#U%bR1v|7X9`!2yD1RV&s_)207*{5d79t@JXcpoXO9^d(w+yN
z0PcDN<vx2Y?g<q6aJRmrv_~BSo`#;Leze_H(NBvdqNa5E6<X~@?UQg`$9;-0Sx}(X
z4R;+Rl!E4%j(&{;Q+v}U(M!D-i#tn@uqEQOwoU8P2W_gZvcv@l@;-4wX8~qZ3qR_3
zu@$|f`E6Xws=~%~p3|j}54;2h%3RT%#WontkxYaiY+s93n0Xss;C`l5dG773Eh`w|
zH~rC=<A7Hq0ooS<0a3Fc@a|36rekeidaI~hbDP;ndjEuXJtZYwn<8AV3-c<waU-*~
zC|_?2j{P10=jw|E>WK~Krvy(kcXR|c3G2e#gQ7UvF9Ai56rldY5O7Wj*#?HyB+I;M
zZK>fZJyEqe#2$5qex%jfymvi4r&aH7t6%KqwSWJxOt8X=U>bG+jS#ulUtzRII3YX|
zGla1`+MgN@?Dc%8_XHEBsDrZH-$8Yjb^>(a>xESqFLy=zo#eU^c@^4PgNPW1w#uBq
zrKn<L=_nwq^fdw|s+3Ky!U;o!v7O2kVIbQkwoQV{X_&n!m{mh!Q?VC`??(Ir!qVtE
z`&O*bPMnPSboDT$n9x>em)44%$72vTrD4T*5Q}%CZ~^9?5xdVhfK3ZQAw3LdTw`H2
z6#0wTs=EKBJaYn#;C_Q)Ylj9*nmLR8AiT&;A}GSmW00cR{Ao4GidAGJFlo!1iM_Zb
zT5#!CuN1RT;#DwLyli%qtvM;}3A|(4Pe=t{ynNYDQWDSQ5R6up2nobAvX#wo+drM{
zx@N$SYu-#=_-@h7o*!Hn?Owyn%e6Gq=FaK@12MN}?k(PVbN|uyuY%OJ{7Lv{!v+W+
zy6{hUKZx6uFn{W{=1;I)@1AX-itrbAi<`zVf2cLFt#lFoP}e_y&ins<DBiz!3Jc`l
mHHDk-@9x7*_+LYU6>p%O;`W)RB;qLE=;<uf4w|`U_x}R~y89LY

literal 0
HcmV?d00001

diff --git a/docs/2_39/stan-users-guide/index.html b/docs/2_39/stan-users-guide/index.html
new file mode 100644
index 000000000..89ccd8743
--- /dev/null
+++ b/docs/2_39/stan-users-guide/index.html
@@ -0,0 +1,1175 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Stan User’s Guide</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/regression.html" rel="next">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan User’s Guide – Stan Docs">
+<meta property="og:description" content="Version 2.39">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed slimcontent">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/index.html">Stan User’s Guide</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar zindex-bottom">
+    </div>
+<!-- main -->
+<main class="content page-columns page-full" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">Stan User’s Guide</h1>
+<p class="subtitle lead">Version 2.39</p>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    
+  
+    
+  </div>
+  
+
+
+</header>
+
+
+<div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
+<p>This is the official user’s guide for <a href="https://mc-stan.org/">Stan</a>. It provides example models and programming techniques for coding statistical models in Stan.</p>
+<ul>
+<li><p>Part 1 gives Stan code and discussions for several important classes of models.</p></li>
+<li><p>Part 2 discusses various general Stan programming techniques that are not tied to any particular model.</p></li>
+<li><p>Part 3 introduces algorithms for calibration and model checking that require multiple runs of Stan.</p></li>
+<li><p>The appendices provide an introduction to the stanc3 compiler used in the various interfaces to Stan, a style guide, and advice for users of BUGS and JAGS.</p></li>
+</ul>
+<p>We recommend working through this guide using the textbooks <em>Bayesian Data Analysis</em> and <em>Statistical Rethinking: A Bayesian Course with Examples in R and Stan</em> as references on the concepts, and using the <a href="https://mc-stan.org/docs/reference-manual/index.html"><em>Stan Reference Manual</em></a> when necessary to clarify programming issues.</p>
+<p><a href="https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf">Download the pdf version of this manual</a>.</p>
+<section id="copyright-and-trademark" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
+<ul>
+<li><p>Copyright 2011–2025, Stan Development Team and their assignees.</p></li>
+<li><p>The Stan name and logo are registered trademarks of NumFOCUS.</p></li>
+</ul>
+</section>
+<section id="licensing" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
+<ul>
+<li><p><em>Text content:</em> <a href="https://creativecommons.org/licenses/by-nd/4.0/legalcode">CC-BY ND 4.0 license</a></p></li>
+<li><p><em>Computer code:</em> <a href="https://opensource.org/licenses/BSD-3-Clause">BSD 3-clause license</a></p></li>
+<li><p><em>Logo:</em> <a href="https://mc-stan.org/about/#copyright-and-trademark">Stan logo usage guidelines</a></p></li>
+</ul>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/regression.html" class="pagination-link" aria-label="Regression Models">
+        <span class="nav-page-text">Regression Models</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/latent-discrete.html b/docs/2_39/stan-users-guide/latent-discrete.html
new file mode 100644
index 000000000..49bf88b9b
--- /dev/null
+++ b/docs/2_39/stan-users-guide/latent-discrete.html
@@ -0,0 +1,1933 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Latent Discrete Parameters</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/sparse-ragged.html" rel="next">
+<link href="../stan-users-guide/measurement-error.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Latent Discrete Parameters">
+<meta property="og:description" content="">
+<meta property="og:image" content="./img/change-point-posterior.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/latent-discrete.html">Latent Discrete Parameters</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#latent-discrete.chapter" id="toc-latent-discrete.chapter" class="nav-link active" data-scroll-target="#latent-discrete.chapter">Latent Discrete Parameters</a>
+  <ul class="collapse">
+  <li><a href="#rao-blackwell.section" id="toc-rao-blackwell.section" class="nav-link" data-scroll-target="#rao-blackwell.section">The benefits of marginalization</a></li>
+  <li><a href="#change-point.section" id="toc-change-point.section" class="nav-link" data-scroll-target="#change-point.section">Change point models</a>
+  <ul class="collapse">
+  <li><a href="#model-with-latent-discrete-parameter" id="toc-model-with-latent-discrete-parameter" class="nav-link" data-scroll-target="#model-with-latent-discrete-parameter">Model with latent discrete parameter</a></li>
+  <li><a href="#marginalizing-out-the-discrete-parameter" id="toc-marginalizing-out-the-discrete-parameter" class="nav-link" data-scroll-target="#marginalizing-out-the-discrete-parameter">Marginalizing out the discrete parameter</a></li>
+  <li><a href="#coding-the-model-in-stan" id="toc-coding-the-model-in-stan" class="nav-link" data-scroll-target="#coding-the-model-in-stan">Coding the model in Stan</a></li>
+  <li><a href="#fitting-the-model-with-mcmc" id="toc-fitting-the-model-with-mcmc" class="nav-link" data-scroll-target="#fitting-the-model-with-mcmc">Fitting the model with MCMC</a></li>
+  <li><a href="#posterior-distribution-of-the-discrete-change-point" id="toc-posterior-distribution-of-the-discrete-change-point" class="nav-link" data-scroll-target="#posterior-distribution-of-the-discrete-change-point">Posterior distribution of the discrete change point</a></li>
+  <li><a href="#discrete-sampling" id="toc-discrete-sampling" class="nav-link" data-scroll-target="#discrete-sampling">Discrete sampling</a></li>
+  <li><a href="#posterior-covariance" id="toc-posterior-covariance" class="nav-link" data-scroll-target="#posterior-covariance">Posterior covariance</a></li>
+  <li><a href="#multiple-change-points" id="toc-multiple-change-points" class="nav-link" data-scroll-target="#multiple-change-points">Multiple change points</a></li>
+  </ul></li>
+  <li><a href="#mark-recapture-models" id="toc-mark-recapture-models" class="nav-link" data-scroll-target="#mark-recapture-models">Mark-recapture models</a>
+  <ul class="collapse">
+  <li><a href="#simple-mark-recapture-model" id="toc-simple-mark-recapture-model" class="nav-link" data-scroll-target="#simple-mark-recapture-model">Simple mark-recapture model</a></li>
+  <li><a href="#cormack-jolly-seber-with-discrete-parameter" id="toc-cormack-jolly-seber-with-discrete-parameter" class="nav-link" data-scroll-target="#cormack-jolly-seber-with-discrete-parameter">Cormack-Jolly-Seber with discrete parameter</a></li>
+  <li><a href="#collective-cormack-jolly-seber-model" id="toc-collective-cormack-jolly-seber-model" class="nav-link" data-scroll-target="#collective-cormack-jolly-seber-model">Collective Cormack-Jolly-Seber model</a></li>
+  <li><a href="#individual-cormack-jolly-seber-model" id="toc-individual-cormack-jolly-seber-model" class="nav-link" data-scroll-target="#individual-cormack-jolly-seber-model">Individual Cormack-Jolly-Seber model</a></li>
+  </ul></li>
+  <li><a href="#data-coding-models.section" id="toc-data-coding-models.section" class="nav-link" data-scroll-target="#data-coding-models.section">Data coding and diagnostic accuracy models</a>
+  <ul class="collapse">
+  <li><a href="#diagnostic-accuracy" id="toc-diagnostic-accuracy" class="nav-link" data-scroll-target="#diagnostic-accuracy">Diagnostic accuracy</a></li>
+  <li><a href="#data-coding" id="toc-data-coding" class="nav-link" data-scroll-target="#data-coding">Data coding</a></li>
+  <li><a href="#noisy-categorical-measurement-model" id="toc-noisy-categorical-measurement-model" class="nav-link" data-scroll-target="#noisy-categorical-measurement-model">Noisy categorical measurement model</a></li>
+  <li><a href="#model-parameters" id="toc-model-parameters" class="nav-link" data-scroll-target="#model-parameters">Model parameters</a></li>
+  <li><a href="#noisy-measurement-model" id="toc-noisy-measurement-model" class="nav-link" data-scroll-target="#noisy-measurement-model">Noisy measurement model</a></li>
+  <li><a href="#stan-implementation" id="toc-stan-implementation" class="nav-link" data-scroll-target="#stan-implementation">Stan implementation</a></li>
+  </ul></li>
+  <li><a href="#marginalization-mathematics.section" id="toc-marginalization-mathematics.section" class="nav-link" data-scroll-target="#marginalization-mathematics.section">The mathematics of recovering marginalized parameters</a>
+  <ul class="collapse">
+  <li><a href="#introduction" id="toc-introduction" class="nav-link" data-scroll-target="#introduction">Introduction</a></li>
+  <li><a href="#estimating-expectations" id="toc-estimating-expectations" class="nav-link" data-scroll-target="#estimating-expectations">Estimating expectations</a></li>
+  <li><a href="#evaluating-the-conditional-expectation" id="toc-evaluating-the-conditional-expectation" class="nav-link" data-scroll-target="#evaluating-the-conditional-expectation">Evaluating the conditional expectation</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/latent-discrete.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/latent-discrete.html">Latent Discrete Parameters</a></li></ol></nav></header>
+
+
+
+
+<section id="latent-discrete.chapter" class="level1">
+<h1>Latent Discrete Parameters</h1>
+<p>Stan does not support sampling discrete parameters. So it is not possible to directly translate BUGS or JAGS models with discrete parameters (i.e., discrete stochastic nodes). Nevertheless, it is possible to code many models that involve bounded discrete parameters by marginalizing out the discrete parameters.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>This chapter shows how to code several widely-used models involving latent discrete parameters. The next chapter, the <a href="../stan-users-guide/clustering.html">clustering chapter</a>, on clustering models, considers further models involving latent discrete parameters.</p>
+<section id="rao-blackwell.section" class="level2">
+<h2 class="anchored" data-anchor-id="rao-blackwell.section">The benefits of marginalization</h2>
+<p>Although it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.</p>
+<p>Standard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.</p>
+</section>
+<section id="change-point.section" class="level2">
+<h2 class="anchored" data-anchor-id="change-point.section">Change point models</h2>
+<p>The first example is a model of coal mining disasters in the U.K. for the years 1851–1962.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<section id="model-with-latent-discrete-parameter" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="model-with-latent-discrete-parameter">Model with latent discrete parameter</h3>
+<p><span class="citation" data-cites="PyMC:2014">Fonnesbeck et al. (<a href="#ref-PyMC:2014" role="doc-biblioref">2013, sec. 3.1</a>)</span> provides a Poisson model of disaster <span class="math inline">\(D_t\)</span> in year <span class="math inline">\(t\)</span> with two rate parameters, an early rate (<span class="math inline">\(e\)</span>) and late rate (<span class="math inline">\(l\)</span>), that change at a given point in time <span class="math inline">\(s\)</span>. The full model expressed using a latent discrete parameter <span class="math inline">\(s\)</span> is <span class="math display">\[\begin{align*}
+e   &amp;\sim  \textsf{exponential}(r_e) \\
+l   &amp;\sim  \textsf{exponential}(r_l) \\
+s   &amp;\sim  \textsf{uniform}(1, T) \\
+D_t &amp;\sim  \textsf{Poisson}(t &lt; s \; ? \; e \: : \: l)
+\end{align*}\]</span></p>
+<p>The last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>It uses a compact notation involving separating its three arguments by a question mark (<code>?</code>) and a colon (<code>:</code>). The conditional operator is defined by <span class="math display">\[
+c \; ? \; x_1 \: : \: x_2
+=
+\begin{cases}
+\ x_1 &amp; \quad\text{if } c \text{ is true (i.e., non-zero), and} \\
+\ x_2 &amp; \quad\text{if } c \text{ is false (i.e., zero).}
+\end{cases}
+\]</span></p>
+</section>
+<section id="marginalizing-out-the-discrete-parameter" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="marginalizing-out-the-discrete-parameter">Marginalizing out the discrete parameter</h3>
+<p>To code this model in Stan, the discrete parameter <span class="math inline">\(s\)</span> must be marginalized out to produce a model defining the log of the probability function <span class="math inline">\(p(e,l,D_t)\)</span>. The full joint probability factors as <span class="math display">\[\begin{align*}
+p(e,l,s,D) &amp;=  p(e) \, p(l) \, p(s) \, p(D \mid s, e, l) \\
+&amp;= \textsf{exponential}(e \mid r_e) \ \textsf{exponential}(l \mid r_l) \,
+    \textsf{uniform}(s \mid 1, T) \\
+&amp; \qquad \prod_{t=1}^T \textsf{Poisson}(D_t \mid t &lt; s \; ? \; e \: : \: l).
+\end{align*}\]</span></p>
+<p>To marginalize, an alternative factorization into prior and likelihood is used, <span class="math display">\[
+p(e,l,D) = p(e,l) \, p(D \mid e,l),
+\]</span></p>
+<p>where the likelihood is defined by marginalizing <span class="math inline">\(s\)</span> as <span class="math display">\[\begin{align*}
+p(D \mid e,l) &amp;= \sum_{s=1}^T p(s, D \mid e,l) \\
+&amp;= \sum_{s=1}^T p(s) \, p(D \mid s,e,l) \\
+&amp;= \sum_{s=1}^T \textsf{uniform}(s \mid 1,T) \,
+    \prod_{t=1}^T \textsf{Poisson}(D_t \mid t &lt; s \; ? \; e \: : \: l).
+\end{align*}\]</span></p>
+<p>Stan operates on the log scale and thus requires the log likelihood, <span class="math display">\[\begin{align*}
+\log p(D \mid e,l)
+&amp;= \texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}_{s=1}^T
+    \left( \log \textsf{uniform}(s \mid 1, T) \vphantom{\sum_{t=1}^T}\right. \\
+&amp;\qquad \left.
+    + \sum_{t=1}^T \log \textsf{Poisson}(D_t \mid t &lt; s \; ? \; e \: : \: l)
+\right),
+\end{align*}\]</span> where the log sum of exponents function is defined by <span class="math display">\[
+\texttt{log}\mathtt{\_}\texttt{sum}\mathtt{\_}\texttt{exp}_{n=1}^N \, \alpha_n =
+\log \sum_{n=1}^N \exp(\alpha_n).
+\]</span></p>
+<p>The log sum of exponents function allows the model to be coded directly in Stan using the built-in function <code>log_sum_exp</code>, which provides both arithmetic stability and efficiency for mixture model calculations.</p>
+</section>
+<section id="coding-the-model-in-stan" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="coding-the-model-in-stan">Coding the model in Stan</h3>
+<p>The Stan program for the change point model is shown in the figure below. The transformed parameter <code>lp[s]</code> stores the quantity <span class="math inline">\(\log p(s, D \mid e, l)\)</span>.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; r_e;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; r_l;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; D;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_unif;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  log_unif = -log(T);</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; e;</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; l;</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] lp;</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>  lp = rep_vector(log_unif, T);</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (s <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-23"><a href="#cb1-23" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-24"><a href="#cb1-24" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-25"><a href="#cb1-25" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-26"><a href="#cb1-26" aria-hidden="true" tabindex="-1"></a>  e ~ exponential(r_e);</span>
+<span id="cb1-27"><a href="#cb1-27" aria-hidden="true" tabindex="-1"></a>  l ~ exponential(r_l);</span>
+<span id="cb1-28"><a href="#cb1-28" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> log_sum_exp(lp);</span>
+<span id="cb1-29"><a href="#cb1-29" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A change point model in which disaster rates <code>D[t]</code> have one rate, <code>e</code>, before the change point and a different rate, <code>l</code>, after the change point. The change point itself, <code>s</code>, is marginalized out as described in the text.</p>
+<p>Although the change-point model is coded directly, the doubly nested loop used for <code>s</code> and <code>t</code> is quadratic in <code>T</code>. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[T] lp;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>    {</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[T + <span class="dv">1</span>] lp_e;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[T + <span class="dv">1</span>] lp_l;</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>      lp_e[<span class="dv">1</span>] = <span class="dv">0</span>;</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>      lp_l[<span class="dv">1</span>] = <span class="dv">0</span>;</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>        lp_e[t + <span class="dv">1</span>] = lp_e[t] + poisson_lpmf(D[t] | e);</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>        lp_l[t + <span class="dv">1</span>] = lp_l[t] + poisson_lpmf(D[t] | l);</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>      lp = rep_vector(log_unif + lp_l[T + <span class="dv">1</span>], T)</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>           + head(lp_e, T) - head(lp_l, T);</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As should be obvious from looking at it, it has linear complexity in <code>T</code> rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger <code>T</code>.</p>
+<p>The key to understanding Wiklendt’s dynamic programming version is to see that <code>head(lp_e)</code> holds the forward values, whereas <code>lp_l[T + 1] - head(lp_l, T)</code> holds the backward values; the clever use of subtraction allows <code>lp_l</code> to be accumulated naturally in the forward direction.</p>
+</section>
+<section id="fitting-the-model-with-mcmc" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="fitting-the-model-with-mcmc">Fitting the model with MCMC</h3>
+<p>This model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.</p>
+<p>The value of <code>lp</code> for each iteration for each change point is available because it is declared as a transformed parameter. If the value of <code>lp</code> were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.</p>
+</section>
+<section id="posterior-distribution-of-the-discrete-change-point" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="posterior-distribution-of-the-discrete-change-point">Posterior distribution of the discrete change point</h3>
+<p>The value of <code>lp[s]</code> in a given iteration is given by <span class="math inline">\(\log
+p(s,D \mid e,l)\)</span> for the values of the early and late rates, <span class="math inline">\(e\)</span> and <span class="math inline">\(l\)</span>, in the iteration. In each iteration after convergence, the early and late disaster rates, <span class="math inline">\(e\)</span> and <span class="math inline">\(l\)</span>, are drawn from the posterior <span class="math inline">\(p(e,l \mid D)\)</span> by MCMC sampling and the associated <code>lp</code> calculated. The value of <code>lp</code> may be normalized to calculate <span class="math inline">\(p(s \mid e,l,D)\)</span> in each iteration, based on on the current values of <span class="math inline">\(e\)</span> and <span class="math inline">\(l\)</span>. Averaging over iterations provides an unnormalized probability estimate of the change point being <span class="math inline">\(s\)</span> (see below for the normalizing constant), <span class="math display">\[\begin{align*}
+p(s \mid D) &amp;\propto q(s \mid D) \\
+&amp;= \frac{1}{M} \sum_{m=1}^{M} \exp(\texttt{lp}[m,s]).
+\end{align*}\]</span> where <span class="math inline">\(\texttt{lp}[m,s]\)</span> represents the value of <code>lp</code> in posterior draw <span class="math inline">\(m\)</span> for change point <span class="math inline">\(s\)</span>. By averaging over draws, <span class="math inline">\(e\)</span> and <span class="math inline">\(l\)</span> are themselves marginalized out, and the result has no dependence on a given iteration’s value for <span class="math inline">\(e\)</span> and <span class="math inline">\(l\)</span>. A final normalization then produces the quantity of interest, the posterior probability of the change point being <span class="math inline">\(s\)</span> conditioned on the data <span class="math inline">\(D\)</span>, <span class="math display">\[
+p(s \mid D) = \frac{q(s \mid D)}{\sum_{s'=1}^T q(s' \mid D)}.
+\]</span></p>
+<p>A plot of the values of <span class="math inline">\(\log p(s \mid D)\)</span> computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.</p>
+<p>Log probability of change point being in year, calculated analytically.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><img src="./img/change-point-posterior.png" class="img-fluid figure-img" style="width:50.0%"></p>
+<figcaption>Analytical change-point posterior</figcaption>
+</figure>
+</div>
+<p>The frequency of change points generated by sampling the discrete change points.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><img src="./img/s-discrete-posterior.png" class="img-fluid figure-img" style="width:50.0%"></p>
+<figcaption>Sampled change-point posterior</figcaption>
+</figure>
+</div>
+<p>In order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of <span class="math inline">\(s\)</span> is roughly 1891.</p>
+</section>
+<section id="discrete-sampling" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="discrete-sampling">Discrete sampling</h3>
+<p>The generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with <code>lp</code> defined as above, the following program draws a random value for <code>s</code> at every iteration.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=T&gt; s;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  s = categorical_logit_rng(lp);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A posterior histogram of draws for <span class="math inline">\(s\)</span> is shown on the second change point posterior figure above.</p>
+<p>Compared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for <code>s</code> given by <code>softmax(lp)</code>.</p>
+</section>
+<section id="posterior-covariance" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="posterior-covariance">Posterior covariance</h3>
+<p>The discrete sample generated for <span class="math inline">\(s\)</span> can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using <code>lp</code>.</p>
+</section>
+<section id="multiple-change-points" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multiple-change-points">Multiple change points</h3>
+<p>There is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, <code>e</code>, <code>m</code>, and <code>l</code>, and two loops for the change point and then one over time, with log densities being stored in a matrix.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[T, T] lp;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>lp = rep_matrix(log_unif, T);</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (s1 <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (s2 <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>      lp[s1,s2] = lp[s1,s2]</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The matrix can then be converted back to a vector using <code>to_vector</code> before being passed to <code>log_sum_exp</code>.</p>
+</section>
+</section>
+<section id="mark-recapture-models" class="level2">
+<h2 class="anchored" data-anchor-id="mark-recapture-models">Mark-recapture models</h2>
+<p>A widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.</p>
+<p>The first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.</p>
+<section id="simple-mark-recapture-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="simple-mark-recapture-model">Simple mark-recapture model</h3>
+<p>In the simplest case, a one-stage mark-recapture study produces the following data</p>
+<ul>
+<li><span class="math inline">\(M\)</span> : number of animals marked in first capture,</li>
+<li><span class="math inline">\(C\)</span> : number animals in second capture, and</li>
+<li><span class="math inline">\(R\)</span> : number of marked animals in second capture.</li>
+</ul>
+<p>The estimand of interest is</p>
+<ul>
+<li><span class="math inline">\(N\)</span> : number of animals in the population.</li>
+</ul>
+<p>Despite the notation, the model will take <span class="math inline">\(N\)</span> to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.</p>
+<p>The Lincoln-Petersen <span class="citation" data-cites="Lincoln:1930 Petersen:1896">(<a href="#ref-Lincoln:1930" role="doc-biblioref">Lincoln 1930</a>; <a href="#ref-Petersen:1896" role="doc-biblioref">Petersen 1896</a>)</span> method for estimating population size is <span class="math display">\[
+\hat{N} = \frac{M C}{R}.
+\]</span></p>
+<p>This population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, <span class="math display">\[
+R \sim \textsf{binomial}(C, M / N)
+\]</span> given the total number of animals captured in the second round (<span class="math inline">\(C\)</span>) with a recapture probability of <span class="math inline">\(M/N\)</span>, the fraction of the total population <span class="math inline">\(N\)</span> marked in the first round.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; M;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; C;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=min(M, C)&gt; R;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=(C - R + M)&gt; N;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  R ~ binomial(C, M / N);</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on <span class="math inline">\(N\)</span> is necessary to efficiently eliminate impossible values.</p>
+<p>The probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.</p>
+<p>To ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for <span class="math inline">\(N\)</span> is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.</p>
+<p>The one tricky part of the model is the lower bound <span class="math inline">\(C - R + M\)</span> placed on the population size <span class="math inline">\(N\)</span>. Values below this bound are impossible because it is otherwise not possible to draw <span class="math inline">\(R\)</span> samples out of the <span class="math inline">\(C\)</span> animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for <span class="math inline">\(C\)</span> implies a variable transform <span class="math inline">\(f : (C-R+M,\infty) \rightarrow (-\infty,+\infty)\)</span> defined by <span class="math inline">\(f(N) = \log(N - (C - R + M))\)</span>; the reference manual contains full details of all constrained parameter transforms.</p>
+</section>
+<section id="cormack-jolly-seber-with-discrete-parameter" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="cormack-jolly-seber-with-discrete-parameter">Cormack-Jolly-Seber with discrete parameter</h3>
+<p>The Cormack-Jolly-Seber (CJS) model <span class="citation" data-cites="Cormack:1964 Jolly:1965 Seber:1965">(<a href="#ref-Cormack:1964" role="doc-biblioref">Cormack 1964</a>; <a href="#ref-Jolly:1965" role="doc-biblioref">Jolly 1965</a>; <a href="#ref-Seber:1965" role="doc-biblioref">Seber 1965</a>)</span> is an open-population model in which the population may change over time due to death; the presentation here draws heavily on <span class="citation" data-cites="Schofield:2007">Schofield (<a href="#ref-Schofield:2007" role="doc-biblioref">2007</a>)</span>.</p>
+<p>The basic data are</p>
+<ul>
+<li><span class="math inline">\(I\)</span>: number of individuals,</li>
+<li><span class="math inline">\(T\)</span>: number of capture periods, and</li>
+<li><span class="math inline">\(y_{i,t}\)</span>: Boolean indicating if individual <span class="math inline">\(i\)</span> was captured at time <span class="math inline">\(t\)</span>.</li>
+</ul>
+<p>Each individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.</p>
+<p>There are two Bernoulli parameters in the model,</p>
+<ul>
+<li><span class="math inline">\(\phi_t\)</span> : probability that animal alive at time <span class="math inline">\(t\)</span> survives until <span class="math inline">\(t + 1\)</span> and</li>
+<li><span class="math inline">\(p_t\)</span> : probability that animal alive at time <span class="math inline">\(t\)</span> is captured at time <span class="math inline">\(t\)</span>.</li>
+</ul>
+<p>These parameters will both be given uniform priors, but information should be used to tighten these priors in practice.</p>
+<p>The CJS model also employs a latent discrete parameter <span class="math inline">\(z_{i,t}\)</span> indicating for each individual <span class="math inline">\(i\)</span> whether it is alive at time <span class="math inline">\(t\)</span>, distributed as <span class="math display">\[
+z_{i,t} \sim \mathsf{Bernoulli}(z_{i,t-1} \; ? \; 0 \: : \: \phi_{t-1}).
+\]</span></p>
+<p>The conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on <span class="math inline">\(z\)</span> as <span class="math display">\[
+y_{i,t} \sim \mathsf{Bernoulli}(z_{i,t} \; ? \; 0 \: : \: p_t).
+\]</span></p>
+<p>The conditional enforces the constraint that dead animals cannot be captured.</p>
+</section>
+<section id="collective-cormack-jolly-seber-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="collective-cormack-jolly-seber-model">Collective Cormack-Jolly-Seber model</h3>
+<p>This subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.</p>
+<p>In order to ease the marginalization of the latent discrete parameter <span class="math inline">\(z_{i,t}\)</span>, the Stan models rely on a derived quantity <span class="math inline">\(\chi_t\)</span> for the probability that an individual is never captured again if it is alive at time <span class="math inline">\(t\)</span> (if it is dead, the recapture probability is zero). this quantity is defined recursively by <span class="math display">\[
+\chi_t
+=
+\begin{cases}
+1 &amp; \quad\text{if } t = T \\
+(1 - \phi_t) + \phi_t (1 - p_{t+1}) \chi_{t+1}
+  &amp; \quad\text{if } t &lt; T
+\end{cases}
+\]</span></p>
+<p>The base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining <span class="math inline">\(\chi_{t}\)</span> in terms of <span class="math inline">\(\chi_{t+1}\)</span> involves two possibilities: (1) not surviving to the next time period, with probability <span class="math inline">\((1 - \phi_t)\)</span>, or (2) surviving to the next time period with probability <span class="math inline">\(\phi_t\)</span>, not being captured in the next time period with probability <span class="math inline">\((1 - p_{t+1})\)</span>, and not being captured again after being alive in period <span class="math inline">\(t+1\)</span> with probability <span class="math inline">\(\chi_{t+1}\)</span>.</p>
+<p>With three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.</p>
+<p><span class="math display">\[
+\begin{array}{crclc}
+\hline
+&amp; \qquad\qquad &amp; captures &amp; \qquad\qquad &amp;  \\
+\mathrm{profile} &amp; 1 &amp; 2 &amp; 3 &amp; \mathrm{probability} \\
+\hline
+0 &amp; - &amp; - &amp; - &amp; n/a \\
+1 &amp; - &amp; - &amp; + &amp; n/a \\
+2 &amp; - &amp; + &amp; - &amp; \chi_2 \\
+3 &amp; - &amp; + &amp; + &amp; \phi_2 \, p_3 \\
+4 &amp; + &amp; - &amp; - &amp; \chi_1 \\
+5 &amp; + &amp; - &amp; + &amp; \phi_1 \, (1 - p_2) \, \phi_2 \, p_3 \\
+6 &amp; + &amp; + &amp; - &amp; \phi_1 \, p_2 \, \chi_2 \\
+7 &amp; + &amp; + &amp; + &amp; \phi_1 \, p_2 \, \phi_2 \, p_3 \\
+\hline
+\end{array}
+\]</span></p>
+<p>History 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.</p>
+<p>By defining these probabilities in terms of <span class="math inline">\(\chi\)</span> directly, there is no need for a latent binary parameter indicating whether an animal is alive at time <span class="math inline">\(t\)</span> or not. The definition of <span class="math inline">\(\chi\)</span> is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model <span class="citation" data-cites="Schofield:2007">(<a href="#ref-Schofield:2007" role="doc-biblioref">Schofield 2007</a>)</span>.</p>
+<p>The Stan model defines <span class="math inline">\(\chi\)</span> as a transformed parameter based on parameters <span class="math inline">\(\phi\)</span> and <span class="math inline">\(p\)</span>. In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using <code>target +=</code> rather than being part of a built-in probability function.</p>
+<p>The following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">7</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; history;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; phi;</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; p;</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; chi;</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>  chi[<span class="dv">2</span>] = (<span class="dv">1</span> - phi[<span class="dv">2</span>]) + phi[<span class="dv">2</span>] * (<span class="dv">1</span> - p[<span class="dv">3</span>]);</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  chi[<span class="dv">1</span>] = (<span class="dv">1</span> - phi[<span class="dv">1</span>]) + phi[<span class="dv">1</span>] * (<span class="dv">1</span> - p[<span class="dv">2</span>]) * chi[<span class="dv">2</span>];</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> history[<span class="dv">2</span>] * log(chi[<span class="dv">2</span>]);</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> history[<span class="dv">3</span>] * (log(phi[<span class="dv">2</span>]) + log(p[<span class="dv">3</span>]));</span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> history[<span class="dv">4</span>] * (log(chi[<span class="dv">1</span>]));</span>
+<span id="cb6-17"><a href="#cb6-17" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> history[<span class="dv">5</span>] * (log(phi[<span class="dv">1</span>]) + log1m(p[<span class="dv">2</span>])</span>
+<span id="cb6-18"><a href="#cb6-18" aria-hidden="true" tabindex="-1"></a>                            + log(phi[<span class="dv">2</span>]) + log(p[<span class="dv">3</span>]));</span>
+<span id="cb6-19"><a href="#cb6-19" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> history[<span class="dv">6</span>] * (log(phi[<span class="dv">1</span>]) + log(p[<span class="dv">2</span>])</span>
+<span id="cb6-20"><a href="#cb6-20" aria-hidden="true" tabindex="-1"></a>                            + log(chi[<span class="dv">2</span>]));</span>
+<span id="cb6-21"><a href="#cb6-21" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> history[<span class="dv">7</span>] * (log(phi[<span class="dv">1</span>]) + log(p[<span class="dv">2</span>])</span>
+<span id="cb6-22"><a href="#cb6-22" aria-hidden="true" tabindex="-1"></a>                            + log(phi[<span class="dv">2</span>]) + log(p[<span class="dv">3</span>]));</span>
+<span id="cb6-23"><a href="#cb6-23" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-24"><a href="#cb6-24" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb6-25"><a href="#cb6-25" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; beta3;</span>
+<span id="cb6-26"><a href="#cb6-26" aria-hidden="true" tabindex="-1"></a>  beta3 = phi[<span class="dv">2</span>] * p[<span class="dv">3</span>];</span>
+<span id="cb6-27"><a href="#cb6-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="identifiability" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="identifiability">Identifiability</h4>
+<p>The parameters <span class="math inline">\(\phi_2\)</span> and <span class="math inline">\(p_3\)</span>, the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, <span class="math inline">\(\beta_3 = \phi_2 \, p_3\)</span>, is identified. The Stan model defines <code>beta3</code> as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the <span class="math inline">\(p\)</span> and <span class="math inline">\(\phi\)</span> parameters.</p>
+</section>
+</section>
+<section id="individual-cormack-jolly-seber-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="individual-cormack-jolly-seber-model">Individual Cormack-Jolly-Seber model</h3>
+<p>This section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number <span class="math inline">\(T\)</span> of capture events, the number <span class="math inline">\(I\)</span> of individuals, and a boolean flag <span class="math inline">\(y_{i,t}\)</span> indicating if individual <span class="math inline">\(i\)</span> was observed at time <span class="math inline">\(t\)</span>. In Stan,</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; T;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; I;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[I, T] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding <span class="math inline">\(2^T\)</span> observation histories for <span class="math inline">\(T\)</span> capture times.</p>
+<section id="utility-functions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="utility-functions">Utility functions</h4>
+<p>The individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> first_capture(<span class="dt">array</span>[] <span class="dt">int</span> y_i) {</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:size(y_i)) {</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (y_i[k]) {</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>        <span class="cf">return</span> k;</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> <span class="dv">0</span>;</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> last_capture(<span class="dt">array</span>[] <span class="dt">int</span> y_i) {</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k_rev <span class="cf">in</span> <span class="dv">0</span>:(size(y_i) - <span class="dv">1</span>)) {</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> k;</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>      k = size(y_i) - k_rev;</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (y_i[k]) {</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>        <span class="cf">return</span> k;</span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> <span class="dv">0</span>;</span>
+<span id="cb8-19"><a href="#cb8-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-20"><a href="#cb8-20" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-21"><a href="#cb8-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These two functions are used to define the first and last capture time for each individual in the transformed data block.<a href="#fn5" class="footnote-ref" id="fnref5" role="doc-noteref"><sup>5</sup></a></p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[I] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=T&gt; first;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[I] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=T&gt; last;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=I&gt;[T] n_captured;</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>    first[i] = first_capture(y[i]);</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>    last[i] = last_capture(y[i]);</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>  n_captured = rep_vector(<span class="dv">0</span>, T);</span>
+<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb9-13"><a href="#cb9-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb9-14"><a href="#cb9-14" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (y[i, t]) {</span>
+<span id="cb9-15"><a href="#cb9-15" aria-hidden="true" tabindex="-1"></a>        n_captured[t] = n_captured[t] + <span class="dv">1</span>;</span>
+<span id="cb9-16"><a href="#cb9-16" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb9-17"><a href="#cb9-17" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb9-18"><a href="#cb9-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-19"><a href="#cb9-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The transformed data block also defines <code>n_captured[t]</code>, which is the total number of captures at time <code>t</code>. The variable <code>n_captured</code> is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.</p>
+<p>The parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector <code>chi</code>, the probability that if an individual is alive at <code>t</code> that it will never be captured again.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[T - <span class="dv">1</span>] phi;</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[T] p;</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[T] chi;</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>  chi = prob_uncaptured(T, p, phi);</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The definition of <code>prob_uncaptured</code>, from the functions block, is</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> prob_uncaptured(<span class="dt">int</span> T, <span class="dt">vector</span> p, <span class="dt">vector</span> phi) {</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[T] chi;</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>    chi[T] = <span class="fl">1.0</span>;</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:(T - <span class="dv">1</span>)) {</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> t_curr;</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> t_next;</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>      t_curr = T - t;</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>      t_next = t_curr + <span class="dv">1</span>;</span>
+<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a>      chi[t_curr] = (<span class="dv">1</span> - phi[t_curr])</span>
+<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a>                     + phi[t_curr]</span>
+<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a>                       * (<span class="dv">1</span> - p[t_next])</span>
+<span id="cb11-14"><a href="#cb11-14" aria-hidden="true" tabindex="-1"></a>                       * chi[t_next];</span>
+<span id="cb11-15"><a href="#cb11-15" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb11-16"><a href="#cb11-16" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> chi;</span>
+<span id="cb11-17"><a href="#cb11-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-18"><a href="#cb11-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function definition directly follows the mathematical definition of <span class="math inline">\(\chi_t\)</span>, unrolling the recursion into an iteration and defining the elements of <code>chi</code> from <code>T</code> down to 1.</p>
+</section>
+<section id="the-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="the-model">The model</h4>
+<p>Given the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations <code>q</code> given the parameters <code>p</code> and <code>phi</code> as well as the transformed parameter <code>chi</code> defined in terms of <code>p</code> and <code>phi</code>.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (first[i] &gt; <span class="dv">0</span>) {</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (t <span class="cf">in</span> (first[i]+<span class="dv">1</span>):last[i]) {</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>        <span class="dv">1</span> ~ bernoulli(phi[t - <span class="dv">1</span>]);</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>        y[i, t] ~ bernoulli(p[t]);</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>      <span class="dv">1</span> ~ bernoulli(chi[last[i]]);</span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The outer loop is over individuals, conditional skipping individuals <code>i</code> which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for <code>first</code> if an individual is never captured.</p>
+<p>The inner loop for individual <code>i</code> first increments the log probability based on the survival of the individual with probability <code>phi[t - 1]</code>. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.</p>
+<p>In the inner loop, the observed capture status <code>y[i, t]</code> for individual <code>i</code> at time <code>t</code> has a Bernoulli distribution based on the capture probability <code>p[t]</code> at time <code>t</code>.</p>
+<p>After the inner loop, the probability of an animal never being seen again after being observed at time <code>last[i]</code> is included, because <code>last[i]</code> was defined to be the last time period in which animal <code>i</code> was observed.</p>
+</section>
+<section id="identified-parameters" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="identified-parameters">Identified parameters</h4>
+<p>As with the collective model described in the previous subsection, this model does not identify <code>phi[T - 1]</code> and <code>p[T]</code>, but does identify their product, <code>beta</code>. Thus <code>beta</code> is defined as a generated quantity to monitor convergence and report.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  beta = phi[T - <span class="dv">1</span>] * p[T];</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameter <code>p[1]</code> is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case <code>p[1]</code> would be an unknown initial condition and both <code>phi[T - 1]</code> and <code>p[T]</code> could be identified.</p>
+</section>
+<section id="population-size-estimates" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="population-size-estimates">Population size estimates</h4>
+<p>The generated quantities also calculates an estimate of the population mean at each time <code>t</code> in the same way as in the simple mark-recapture model as the number of individuals captured at time <code>t</code> divided by the probability of capture at time <code>t</code>. This is done with the elementwise division operation for vectors (<code>./</code>) in the generated quantities block.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[T] pop;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>  pop = n_captured ./ p;</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>  pop[<span class="dv">1</span>] = -<span class="dv">1</span>;</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="generalizing-to-individual-effects" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="generalizing-to-individual-effects">Generalizing to individual effects</h4>
+<p>All individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.</p>
+</section>
+</section>
+</section>
+<section id="data-coding-models.section" class="level2">
+<h2 class="anchored" data-anchor-id="data-coding-models.section">Data coding and diagnostic accuracy models</h2>
+<p>Although seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.</p>
+<section id="diagnostic-accuracy" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="diagnostic-accuracy">Diagnostic accuracy</h3>
+<p>Suppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.</p>
+<p>There are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.</p>
+</section>
+<section id="data-coding" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-coding">Data coding</h3>
+<p>Humans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.</p>
+</section>
+<section id="noisy-categorical-measurement-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="noisy-categorical-measurement-model">Noisy categorical measurement model</h3>
+<p>In this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.</p>
+<p><span class="citation" data-cites="DawidSkene:1979">Dawid and Skene (<a href="#ref-DawidSkene:1979" role="doc-biblioref">1979</a>)</span> introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.</p>
+<section id="data" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="data">Data</h4>
+<p>The data for the model consists of <span class="math inline">\(J\)</span> raters (diagnostic tests), <span class="math inline">\(I\)</span> items (patients), and <span class="math inline">\(K\)</span> categories (condition statuses) to annotate, with <span class="math inline">\(y_{i, j} \in \{1, \dotsc, K\}\)</span> being the rating provided by rater <span class="math inline">\(j\)</span> for item <span class="math inline">\(i\)</span>. In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often <span class="math inline">\(K=2\)</span>, with values signaling the presence or absence of the condition.<a href="#fn6" class="footnote-ref" id="fnref6" role="doc-noteref"><sup>6</sup></a></p>
+<p>It is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.</p>
+</section>
+</section>
+<section id="model-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="model-parameters">Model parameters</h3>
+<p>The model is based on three parameters, the first of which is discrete:</p>
+<ul>
+<li><span class="math inline">\(z_i\)</span> : a value in <span class="math inline">\(\{1, \dotsc, K\}\)</span> indicating the true category of item <span class="math inline">\(i\)</span>,</li>
+<li><span class="math inline">\(\pi\)</span> : a <span class="math inline">\(K\)</span>-simplex for the prevalence of the <span class="math inline">\(K\)</span> categories in the population, and</li>
+<li><span class="math inline">\(\theta_{j,k}\)</span> : a <span class="math inline">\(K\)</span>-simplex for the response of annotator <span class="math inline">\(j\)</span> to an item of true category <span class="math inline">\(k\)</span>.</li>
+</ul>
+</section>
+<section id="noisy-measurement-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="noisy-measurement-model">Noisy measurement model</h3>
+<p>The true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, <span class="math display">\[
+z_i \sim \textsf{categorical}(\pi).
+\]</span></p>
+<p>The rating <span class="math inline">\(y_{i, j}\)</span> provided for item <span class="math inline">\(i\)</span> by rater <span class="math inline">\(j\)</span> is modeled as a categorical response of rater <span class="math inline">\(i\)</span> to an item of category <span class="math inline">\(z_i\)</span>,<a href="#fn7" class="footnote-ref" id="fnref7" role="doc-noteref"><sup>7</sup></a> <span class="math display">\[
+y_{i, j} \sim \textsf{categorical}(\theta_{j,\pi_{z[i]}}).
+\]</span></p>
+<section id="priors-and-hierarchical-modeling" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="priors-and-hierarchical-modeling">Priors and hierarchical modeling</h4>
+<p>Dawid and Skene provided maximum likelihood estimates for <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\pi\)</span>, which allows them to generate probability estimates for each <span class="math inline">\(z_i\)</span>.</p>
+<p>To mimic Dawid and Skene’s maximum likelihood model, the parameters <span class="math inline">\(\theta_{j,k}\)</span> and <span class="math inline">\(\pi\)</span> can be given uniform priors over <span class="math inline">\(K\)</span>-simplexes. It is straightforward to generalize to Dirichlet priors, <span class="math display">\[
+\pi \sim \textsf{Dirichlet}(\alpha)
+\]</span> and <span class="math display">\[
+\theta_{j,k} \sim \textsf{Dirichlet}(\beta_k)
+\]</span> with fixed hyperparameters <span class="math inline">\(\alpha\)</span> (a vector) and <span class="math inline">\(\beta\)</span> (a matrix or array of vectors). The prior for <span class="math inline">\(\theta_{j,k}\)</span> must be allowed to vary in <span class="math inline">\(k\)</span>, so that, for instance, <span class="math inline">\(\beta_{k,k}\)</span> is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.</p>
+<p>Because there are <span class="math inline">\(J\)</span> coders, it would be natural to extend the model to include a hierarchical prior for <span class="math inline">\(\beta\)</span> and to partially pool the estimates of coder accuracy and bias.</p>
+</section>
+<section id="marginalizing-out-the-true-category" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="marginalizing-out-the-true-category">Marginalizing out the true category</h4>
+<p>Because the true category parameter <span class="math inline">\(z\)</span> is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as <span class="math display">\[
+p(y, \theta, \pi) = p(y \mid \theta,\pi) \, p(\pi) \, p(\theta),
+\]</span> where <span class="math inline">\(p(y \mid \theta,\pi)\)</span> is derived by marginalizing <span class="math inline">\(z\)</span> out of <span class="math display">\[
+p(z, y \mid \theta, \pi) =
+\prod_{i=1}^I \left( \textsf{categorical}(z_i \mid \pi)
+                     \prod_{j=1}^J
+                     \textsf{categorical}(y_{i, j} \mid \theta_{j, z[i]})
+              \right).
+\]</span></p>
+<p>This can be done item by item, with <span class="math display">\[
+p(y \mid \theta, \pi) =
+\prod_{i=1}^I \sum_{k=1}^K
+  \left( \textsf{categorical}(k \mid \pi)
+         \prod_{j=1}^J
+         \textsf{categorical}(y_{i, j} \mid \theta_{j, k})
+  \right).
+\]</span></p>
+<p>In the missing data model, only the observed labels would be used in the inner product.</p>
+<p><span class="citation" data-cites="DawidSkene:1979">Dawid and Skene (<a href="#ref-DawidSkene:1979" role="doc-biblioref">1979</a>)</span> derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, <span class="math display">\[\begin{align*}
+\log p(y \mid \theta, \pi)
+&amp;= \sum_{i=1}^I \log \left( \sum_{k=1}^K \exp
+    \left(\log \textsf{categorical}(k \mid \pi) \vphantom{\sum_{j=1}^J}\right.\right.
+    \left.\left. + \ \sum_{j=1}^J
+           \log \textsf{categorical}(y_{i, j} \mid \theta_{j, k})
+    \right) \right),
+\end{align*}\]</span> which can be directly coded using Stan’s built-in <code>log_sum_exp</code> function.</p>
+</section>
+</section>
+<section id="stan-implementation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-implementation">Stan implementation</h3>
+<p>The Stan program for the Dawid and Skene model is provided below <span class="citation" data-cites="DawidSkene:1979">(<a href="#ref-DawidSkene:1979" role="doc-biblioref">Dawid and Skene 1979</a>)</span>.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; I;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[I, J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; y;</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] alpha;</span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] beta[K];</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-11"><a href="#cb15-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb15-12"><a href="#cb15-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] pi;</span>
+<span id="cb15-13"><a href="#cb15-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J, K] <span class="dt">simplex</span>[K] theta;</span>
+<span id="cb15-14"><a href="#cb15-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-15"><a href="#cb15-15" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb15-16"><a href="#cb15-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[I] <span class="dt">vector</span>[K] log_q_z;</span>
+<span id="cb15-17"><a href="#cb15-17" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb15-18"><a href="#cb15-18" aria-hidden="true" tabindex="-1"></a>    log_q_z[i] = log(pi);</span>
+<span id="cb15-19"><a href="#cb15-19" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb15-20"><a href="#cb15-20" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb15-21"><a href="#cb15-21" aria-hidden="true" tabindex="-1"></a>        log_q_z[i, k] = log_q_z[i, k]</span>
+<span id="cb15-22"><a href="#cb15-22" aria-hidden="true" tabindex="-1"></a>                         + log(theta[j, k, y[i, j]]);</span>
+<span id="cb15-23"><a href="#cb15-23" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb15-24"><a href="#cb15-24" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-25"><a href="#cb15-25" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-26"><a href="#cb15-26" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-27"><a href="#cb15-27" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-28"><a href="#cb15-28" aria-hidden="true" tabindex="-1"></a>  pi ~ dirichlet(alpha);</span>
+<span id="cb15-29"><a href="#cb15-29" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb15-30"><a href="#cb15-30" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb15-31"><a href="#cb15-31" aria-hidden="true" tabindex="-1"></a>      theta[j, k] ~ dirichlet(beta[k]);</span>
+<span id="cb15-32"><a href="#cb15-32" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb15-33"><a href="#cb15-33" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-34"><a href="#cb15-34" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-35"><a href="#cb15-35" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb15-36"><a href="#cb15-36" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_sum_exp(log_q_z[i]);</span>
+<span id="cb15-37"><a href="#cb15-37" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-38"><a href="#cb15-38" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The model marginalizes out the discrete parameter <span class="math inline">\(z\)</span>, storing the unnormalized conditional probability <span class="math inline">\(\log q(z_i=k|\theta,\pi)\)</span> in <code>log_q_z[i, k]</code>.</p>
+<p>The Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are <span class="math inline">\(\alpha_k = 3\)</span> and <span class="math inline">\(\beta_{k,k} = 2.5 K\)</span> and <span class="math inline">\(\beta_{k,k'} = 1\)</span> if <span class="math inline">\(k \neq k'\)</span>. Taking <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta_k\)</span> to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of <span class="citation" data-cites="DawidSkene:1979">Dawid and Skene (<a href="#ref-DawidSkene:1979" role="doc-biblioref">1979</a>)</span>.</p>
+<section id="inference-for-the-true-category" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="inference-for-the-true-category">Inference for the true category</h4>
+<p>The quantity <code>log_q_z[i]</code> is defined as a transformed parameter. It encodes the (unnormalized) log of <span class="math inline">\(p(z_i \mid \theta,
+\pi)\)</span>. Each iteration provides a value conditioned on that iteration’s values for <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\pi\)</span>. Applying the softmax function to <code>log_q_z[i]</code> provides a simplex corresponding to the probability mass function of <span class="math inline">\(z_i\)</span> in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each <span class="math inline">\(z_i\)</span>.</p>
+</section>
+</section>
+</section>
+<section id="marginalization-mathematics.section" class="level2">
+<h2 class="anchored" data-anchor-id="marginalization-mathematics.section">The mathematics of recovering marginalized parameters</h2>
+<section id="introduction" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="introduction">Introduction</h3>
+<p>This section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is <em>not</em> necessary to fit models with discrete parameters using Stan.</p>
+<p>Throughout, the model under consideration consists of both continuous parameters, <span class="math inline">\(\Theta\)</span>, and discrete parameters, <span class="math inline">\(Z\)</span>. It is also assumed that <span class="math inline">\(Z\)</span> can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.</p>
+</section>
+<section id="estimating-expectations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimating-expectations">Estimating expectations</h3>
+<p>When performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating <span class="math inline">\(\mu\)</span>, say, where <span class="math inline">\(\mu = \mathbb{E}[g(\Theta, Z)]\)</span> and <span class="math inline">\(g(\cdot)\)</span> is an arbitrary function. An example of such a quantity is <span class="math inline">\(\mathbb{E}[\Theta]\)</span>, the posterior mean of the continuous parameters, where we would take <span class="math inline">\(g(\theta, z) = \theta\)</span>. To estimate <span class="math inline">\(\mu\)</span> the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.</p>
+<p>If both <span class="math inline">\(\Theta\)</span> and <span class="math inline">\(Z\)</span> were continuous, Stan could be used to sample <span class="math inline">\(M\)</span> draws from the joint posterior <span class="math inline">\(p_{\Theta, Z}(\theta, z)\)</span> and then estimate <span class="math inline">\(\mu\)</span> with <span class="math display">\[
+\hat{\mu} = \frac{1}{M} \sum_{i = 1}^M {g(\theta^{(i)}, z^{(i)})}.
+\]</span> Given <span class="math inline">\(Z\)</span> is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out <span class="math inline">\(Z\)</span> from the joint posterior to give the marginalized posterior <span class="math inline">\(p_\Theta(\theta)\)</span>. This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws <span class="math inline">\(\{\theta^{(i)}\}_{i = 1}^M\)</span> from the marginalized posterior.</p>
+<p>Using only these draws, how can we estimate <span class="math inline">\(\mathbb{E}[g(\Theta, Z)]\)</span>? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function <span class="math inline">\(g\)</span> of interest, we compute <span class="math display">\[
+h(\Theta) = \mathbb{E}[g(\Theta, Z) \mid \Theta]
+\]</span> and then estimate <span class="math inline">\(\mathbb{E}[g(\Theta, Z)]\)</span> with <span class="math display">\[
+\hat{\mu} = \frac{1}{M} \sum_{i = 1}^M h(\theta^{(i)}).
+\]</span> This estimator is justified by the law of iterated expectation, the fact that <span class="math display">\[
+\mathbb{E}[h(\Theta)] = \mathbb{E}[\mathbb{E}[g(\Theta, Z)] \mid \Theta] = \mathbb{E}[g(\Theta, Z)] = \mu.
+\]</span> Using this marginalized estimator provides a way to estimate the expectation of any function <span class="math inline">\(g(\cdot)\)</span> for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation <span class="math inline">\(\mathbb{E}[g(\Theta, Z) \mid \Theta]\)</span>.</p>
+</section>
+<section id="evaluating-the-conditional-expectation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="evaluating-the-conditional-expectation">Evaluating the conditional expectation</h3>
+<p>Fortunately, the discrete nature of <span class="math inline">\(Z\)</span> makes evaluating <span class="math inline">\(\mathbb{E}[g(\Theta, Z) \mid \Theta]\)</span> easy. The function <span class="math inline">\(h(\Theta)\)</span> can be written as: <span class="math display">\[
+h(\Theta)
+= \mathbb{E}[g(\Theta, Z) \mid \Theta]
+= \sum_{k} g(\Theta, k) \Pr[Z = k \mid \Theta],
+\]</span> where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, <span class="math inline">\(\Pr[Z = k \mid \Theta]\)</span>. More detail on how this quantity can be calculated is included below. Note that if <span class="math inline">\(Z\)</span> takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.</p>
+<p>When <span class="math inline">\(g(\theta, z)\)</span> is a function of either <span class="math inline">\(\theta\)</span> or <span class="math inline">\(z\)</span> only, the above formula simplifies further.</p>
+<p>In the first case, where <span class="math inline">\(g(\theta, z) = g(\theta)\)</span>, we have: <span class="math display">\[\begin{align*}
+h(\Theta)
+&amp;= \sum_{k} g(\Theta) \Pr[Z = k \mid \Theta] \\
+&amp;= g(\Theta) \sum_{k} \Pr[Z = k \mid \Theta] \\
+&amp;= g(\Theta).
+\end{align*}\]</span> This means that we can estimate <span class="math inline">\(\mathbb{E}[g(\Theta)]\)</span> with the standard, seemingly unconditional, estimator: <span class="math display">\[
+\frac{1}{M} \sum_{i = 1}^M g(\theta^{(i)}).
+\]</span> Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.</p>
+<p>In the second case, where <span class="math inline">\(g(\theta, z) = g(z)\)</span>, the conditional expectation instead simplifies as follows: <span class="math display">\[
+h(\Theta) = \sum_{k} g(k) \Pr[Z = k \mid \Theta].
+\]</span> An important special case of this result is when <span class="math inline">\(g(\theta, z) = \textrm{I}(z = k)\)</span>, where <span class="math inline">\(\textrm{I}\)</span> is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable <span class="math inline">\(Z\)</span>, since <span class="math inline">\(\mathbb{E}[\textrm{I}(Z = k)] = \Pr[Z = k]\)</span>. In this case, <span class="math display">\[
+h(\Theta)
+= \sum_{k} \textrm{I}(z = k) \Pr[Z = k \mid \Theta]
+= \Pr[Z = k \mid \Theta].
+\]</span> The quantity <span class="math inline">\(\Pr[Z = k]\)</span> can therefore be estimated with: <span class="math display">\[
+\frac{1}{M} \sum_{i = 1}^M \Pr[Z = k \mid \Theta = \theta^{(i)}].
+\]</span> When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, <span class="math inline">\(Y\)</span>. That is, we are really estimating <span class="math inline">\(\Pr[Z = k \mid Y]\)</span> with <span class="math display">\[
+\frac{1}{M} \sum_{i = 1}^M \Pr[Z = k \mid \Theta = \theta^{(i)}, Y].
+\]</span> This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us <span class="math display">\[
+\Pr[Z = k \mid \Theta = \theta^{(i)}, Y]
+= \frac{\Pr[Y \mid Z = k, \Theta = \theta^{(i)}]
+\Pr[Z = k \mid \Theta = \theta^{(i)}]}
+{\sum_{k = 1}^K \Pr[Y \mid Z = k, \Theta = \theta^{(i)}]
+\Pr[Z = k \mid \Theta = \theta^{(i)}]}.
+\]</span> Here, <span class="math inline">\(\Pr[Y \mid \Theta = \theta^{(i)}, Z = k]\)</span> is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.</p>
+<p>Other than the use of Bayes’s theorem, <span class="math inline">\(\Pr[Z = k \mid \theta = \theta^{(i)}, Y]\)</span> can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the <a href="#data-coding-models.section">Dawid–Skene model</a>).</p>
+<p>For a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see <span class="citation" data-cites="PullinEtAl:2021">Pullin, Gurrin, and Vukcevic (<a href="#ref-PullinEtAl:2021" role="doc-biblioref">2021</a>)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Cormack:1964" class="csl-entry" role="listitem">
+Cormack, R. M. 1964. <span>“Estimates of Survival from the Sighting of Marked Animals.”</span> <em>Biometrika</em> 51 (3/4): 429–38.
+</div>
+<div id="ref-DawidSkene:1979" class="csl-entry" role="listitem">
+Dawid, A. P., and A. M. Skene. 1979. <span>“Maximum Likelihood Estimation of Observer Error-Rates Using the <span>EM</span> Algorithm.”</span> <em>Journal of the Royal Statistical Society. Series C (Applied Statistics)</em> 28 (1): 20–28.
+</div>
+<div id="ref-dempster-et-al:1977" class="csl-entry" role="listitem">
+Dempster, A. P., N. M. Laird, and D. B. Rubin. 1977. <span>“Maximum Likelihood from Incomplete Data via the <span>EM</span> Algorithm.”</span> <em>Journal of the Royal Statistical Society. Series B (Methodological)</em> 39 (1): 1–38.
+</div>
+<div id="ref-PyMC:2014" class="csl-entry" role="listitem">
+Fonnesbeck, Chris, Anand Patil, David Huard, and John Salvatier. 2013. <em>PyMC User’s Guide</em>.
+</div>
+<div id="ref-Jarret:1979" class="csl-entry" role="listitem">
+Jarrett, R. G. 1979. <span>“A Note on the Intervals Between Coal-Mining Disasters.”</span> <em>Biometrika</em> 66 (1): 191–93.
+</div>
+<div id="ref-Jolly:1965" class="csl-entry" role="listitem">
+Jolly, G. M. 1965. <span>“Explicit Estimates from Capture-Recapture Data with Both Death and Immigration-Stochastic Model.”</span> <em>Biometrika</em> 52 (1/2): 225–47.
+</div>
+<div id="ref-Lincoln:1930" class="csl-entry" role="listitem">
+Lincoln, F. C. 1930. <span>“Calculating Waterfowl Abundance on the Basis of Banding Returns.”</span> <em>United States Department of Agriculture Circular</em> 118: 1–4.
+</div>
+<div id="ref-Petersen:1896" class="csl-entry" role="listitem">
+Petersen, C. G. J. 1896. <span>“The Yearly Immigration of Young Plaice into the <span>L</span>imfjord from the <span>G</span>erman <span>S</span>ea.”</span> <em>Report of the Danish Biological Station</em> 6: 5–84.
+</div>
+<div id="ref-PullinEtAl:2021" class="csl-entry" role="listitem">
+Pullin, Jeffrey, Lyle Gurrin, and Damjan Vukcevic. 2021. <span>“Statistical Models of Repeated Categorical Ratings: The r Package Rater.”</span> <em>arXiv</em> 2010.09335. <a href="https://arxiv.org/abs/2010.09335">https://arxiv.org/abs/2010.09335</a>.
+</div>
+<div id="ref-Schofield:2007" class="csl-entry" role="listitem">
+Schofield, Matthew R. 2007. <span>“Hierarchical Capture-Recapture Models.”</span> PhD thesis, Department of Statistics, University of Otago, Dunedin.
+</div>
+<div id="ref-Seber:1965" class="csl-entry" role="listitem">
+Seber, G. A. F. 1965. <span>“A Note on the Multiple-Recapture Census.”</span> <em>Biometrika</em> 52 (1/2): 249–59.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The computations are similar to those involved in expectation maximization (EM) algorithms <span class="citation" data-cites="dempster-et-al:1977">(<a href="#ref-dempster-et-al:1977" role="doc-biblioref">Dempster, Laird, and Rubin 1977</a>)</span>.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>The source of the data is <span class="citation" data-cites="Jarret:1979">(<a href="#ref-Jarret:1979" role="doc-biblioref">Jarrett 1979</a>)</span>, which itself is a note correcting an earlier data collection.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>The R counterpart, <code>ifelse</code>, is slightly different in that it is typically used in a vectorized situation. The conditional operator is not (yet) vectorized in Stan.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn4"><p>An alternative would be to compute this on the outside and feed it into the Stan model as preprocessed data. Yet another alternative encoding would be a sparse one recording only the capture events along with their time and identifying the individual captured.<a href="#fnref4" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn5"><p>Both functions return 0 if the individual represented by the input array was never captured. Individuals with no captures are not relevant for estimating the model because all probability statements are conditional on earlier captures. Typically they would be removed from the data, but the program allows them to be included even though they make not contribution to the log probability function.<a href="#fnref5" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn6"><p>Diagnostic procedures are often ordinal, as in stages of cancer in oncological diagnosis or the severity of a cavity in dental diagnosis. Dawid and Skene’s model may be used as is or naturally generalized for ordinal ratings using a latent continuous rating and cutpoints as in ordinal logistic regression.<a href="#fnref6" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn7"><p>In the subscript, <span class="math inline">\(z_i\)</span> is written as <span class="math inline">\(z[i]\)</span> to improve legibility.<a href="#fnref7" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/measurement-error.html" class="pagination-link" aria-label="Measurement Error and Meta-Analysis">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Measurement Error and Meta-Analysis</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/sparse-ragged.html" class="pagination-link" aria-label="Sparse and Ragged Data Structures">
+        <span class="nav-page-text">Sparse and Ragged Data Structures</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/latent-discrete.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/matrices-arrays.html b/docs/2_39/stan-users-guide/matrices-arrays.html
new file mode 100644
index 000000000..b638c4842
--- /dev/null
+++ b/docs/2_39/stan-users-guide/matrices-arrays.html
@@ -0,0 +1,1434 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Matrices, Vectors, Arrays, and Tuples</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/multi-indexing.html" rel="next">
+<link href="../stan-users-guide/floating-point.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Matrices, Vectors, Arrays, and Tuples">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/matrices-arrays.html">Matrices, Vectors, Arrays, and Tuples</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#matrices-vectors-arrays-and-tuples" id="toc-matrices-vectors-arrays-and-tuples" class="nav-link active" data-scroll-target="#matrices-vectors-arrays-and-tuples">Matrices, Vectors, Arrays, and Tuples</a>
+  <ul class="collapse">
+  <li><a href="#basic-motivation" id="toc-basic-motivation" class="nav-link" data-scroll-target="#basic-motivation">Basic motivation</a></li>
+  <li><a href="#tuple-types" id="toc-tuple-types" class="nav-link" data-scroll-target="#tuple-types">Tuple types</a>
+  <ul class="collapse">
+  <li><a href="#tuple-syntax" id="toc-tuple-syntax" class="nav-link" data-scroll-target="#tuple-syntax">Tuple syntax</a></li>
+  <li><a href="#applications-of-tuples" id="toc-applications-of-tuples" class="nav-link" data-scroll-target="#applications-of-tuples">Applications of tuples</a></li>
+  </ul></li>
+  <li><a href="#fixed-sizes-and-indexing-out-of-bounds" id="toc-fixed-sizes-and-indexing-out-of-bounds" class="nav-link" data-scroll-target="#fixed-sizes-and-indexing-out-of-bounds">Fixed sizes and indexing out of bounds</a></li>
+  <li><a href="#indexing-efficiency.section" id="toc-indexing-efficiency.section" class="nav-link" data-scroll-target="#indexing-efficiency.section">Data type and indexing efficiency</a>
+  <ul class="collapse">
+  <li><a href="#matrices-vs.-two-dimensional-arrays" id="toc-matrices-vs.-two-dimensional-arrays" class="nav-link" data-scroll-target="#matrices-vs.-two-dimensional-arrays">Matrices vs.&nbsp;two-dimensional arrays</a></li>
+  <li><a href="#row-vectors-vs.-one-dimensional-arrays" id="toc-row-vectors-vs.-one-dimensional-arrays" class="nav-link" data-scroll-target="#row-vectors-vs.-one-dimensional-arrays">(Row) vectors vs.&nbsp;one-dimensional arrays</a></li>
+  </ul></li>
+  <li><a href="#memory-locality" id="toc-memory-locality" class="nav-link" data-scroll-target="#memory-locality">Memory locality</a>
+  <ul class="collapse">
+  <li><a href="#memory-locality-1" id="toc-memory-locality-1" class="nav-link" data-scroll-target="#memory-locality-1">Memory locality</a></li>
+  <li><a href="#matrices" id="toc-matrices" class="nav-link" data-scroll-target="#matrices">Matrices</a></li>
+  <li><a href="#arrays" id="toc-arrays" class="nav-link" data-scroll-target="#arrays">Arrays</a></li>
+  </ul></li>
+  <li><a href="#converting-among-matrix-vector-and-array-types" id="toc-converting-among-matrix-vector-and-array-types" class="nav-link" data-scroll-target="#converting-among-matrix-vector-and-array-types">Converting among matrix, vector, and array types</a></li>
+  <li><a href="#aliasing-in-stan-containers" id="toc-aliasing-in-stan-containers" class="nav-link" data-scroll-target="#aliasing-in-stan-containers">Aliasing in Stan containers</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/matrices-arrays.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/matrices-arrays.html">Matrices, Vectors, Arrays, and Tuples</a></li></ol></nav></header>
+
+
+
+
+<section id="matrices-vectors-arrays-and-tuples" class="level1">
+<h1>Matrices, Vectors, Arrays, and Tuples</h1>
+<p>This chapter provides pointers as to how to choose among the various container types (matrix, vector, array, and tuple) provided by Stan.</p>
+<section id="basic-motivation" class="level2">
+<h2 class="anchored" data-anchor-id="basic-motivation">Basic motivation</h2>
+<p>Stan provides three basic scalar types, <code>int</code>, <code>real</code>, and <code>complex</code>, as well as three basic linear algebra types, <code>vector</code>, <code>row_vector</code>, and <code>matrix</code>. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).</p>
+<p>This leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[N] <span class="dt">real</span> a;</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] a;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[N] a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.</p>
+<p>The following code fragment shows all four ways to declare a two-dimensional container of size <span class="math inline">\(M \times N\)</span>.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M, N] <span class="dt">real</span> b;          <span class="co">// b[m] : array[] real     (efficient)</span></span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M] <span class="dt">vector</span>[N] b;        <span class="co">// b[m] : vector     (efficient)</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M] <span class="dt">row_vector</span>[N] b;    <span class="co">// b[m] : row_vector (efficient)</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] b;              <span class="co">// b[m] : row_vector (inefficient)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The main differences among these choices involve efficiency for various purposes and the type of <code>b[m]</code>, which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.</p>
+<p>The inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).</p>
+</section>
+<section id="tuple-types" class="level2">
+<h2 class="anchored" data-anchor-id="tuple-types">Tuple types</h2>
+<p>Arrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single <code>complex</code> typed entry, all the entries are <code>complex</code>).</p>
+<p>With arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a <code>complex_vector[3]</code> holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.</p>
+<p>Tuples provide a way to represent a sequence of values of heterogeneous types. For example, <code>tuple(int, real)</code> is the type of a pair consisting of an integer and a real number and <code>tuple(array[5] int, vector[6])</code> is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.</p>
+<section id="tuple-syntax" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="tuple-syntax">Tuple syntax</h3>
+<p>Tuples are declared using the keyword <code>tuple</code> followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">vector</span>[<span class="dv">3</span>]) ny = (<span class="dv">5</span>, [<span class="dv">3</span>, <span class="fl">2.9</span>, <span class="fl">1.8</span>]');</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n = ny<span class="fl">.1</span>;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">3</span>] y = ny<span class="fl">.2</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>We can also assign into the elements of a tuple.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">vector</span>[<span class="dv">3</span>], <span class="dt">complex</span>) abc;</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>abc<span class="fl">.1</span> = <span class="dv">5</span>;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>abc<span class="fl">.2</span>[<span class="dv">1</span>] = <span class="dv">3</span>;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>abc<span class="fl">.2</span>[<span class="dv">2</span>] = <span class="fl">2.9</span>;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>abc<span class="fl">.2</span>[<span class="dv">3</span>] = <span class="fl">1.4798</span>;</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>abc<span class="fl">.3</span> = <span class="dv">2</span> + <span class="fl">1.9</span>j;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As the cascaded indexing example shows, the result of <code>abc.1</code> is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., <code>abc.2[1]</code> pulls out the first element of the vector value of the second element of the tuple.)</p>
+<p>There are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> a = <span class="dv">5</span>;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">100</span>] b = ...;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">matrix</span>[<span class="dv">10</span>, <span class="dv">100</span>]) ab = (a, b);  <span class="co">// COPIES b</span></span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>b[<span class="dv">1</span>,<span class="dv">1</span>] = <span class="fl">10.3</span>;  <span class="co">// does NOT change ab</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="applications-of-tuples" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="applications-of-tuples">Applications of tuples</h3>
+<p>Tuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, <code>array[N] tuple(int, real, vector[5])</code> is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as <code>tuple(array[N] int, array[N] real, array[N] vector[5])</code>.</p>
+<p>The second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix <span class="math inline">\(A = Q \cdot R\)</span>, where <span class="math inline">\(Q\)</span> is orthonormal and <span class="math inline">\(R\)</span> is upper triangular. In the past, this required two function calls.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] A = ...;</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, M] Q = qr_Q(A);</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] R = qr_R(A);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With tuples, this can be simplified to the following,</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">matrix</span>[M, M], <span class="dt">matrix</span>[M, N]) QR = qr(A);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with <code>QR.1</code> being <code>Q</code> and <code>QR.2</code> giving <code>R</code>.</p>
+</section>
+</section>
+<section id="fixed-sizes-and-indexing-out-of-bounds" class="level2">
+<h2 class="anchored" data-anchor-id="fixed-sizes-and-indexing-out-of-bounds">Fixed sizes and indexing out of bounds</h2>
+<p>Stan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.</p>
+<p>When an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.</p>
+</section>
+<section id="indexing-efficiency.section" class="level2">
+<h2 class="anchored" data-anchor-id="indexing-efficiency.section">Data type and indexing efficiency</h2>
+<p>The underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.</p>
+<p>Arrays, on the other hand, are implemented as instances of the C++<br>
+<code>std::vector</code> class (not to be confused with Eigen’s <code>Eigen::Vector</code> class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.</p>
+<section id="matrices-vs.-two-dimensional-arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrices-vs.-two-dimensional-arrays">Matrices vs.&nbsp;two-dimensional arrays</h3>
+<p>In Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.</p>
+<p>First, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.</p>
+<p>Second, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).</p>
+<p>Third, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] a;</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a><span class="co">//...</span></span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ... do something with a[m, n] ...</span></span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Arrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M, N] <span class="dt">real</span> a;</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>    <span class="co">// ... do something with a[m, n] ...</span></span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The first use of <code>a[m ,n]</code> should bring <code>a[m]</code> into memory. Overall, traversing matrices is more efficient than traversing arrays.</p>
+<p>This is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[I, J] <span class="dt">matrix</span>[M, N] b;</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:I) {</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>        <span class="co">// ... do something with b[i, j, m, n] ...</span></span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If <code>a</code> is a matrix, the notation <code>a[m]</code> picks out row <code>m</code> of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M] <span class="dt">row_vector</span>[N] b;</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... do something with row vector b[m] ...</span></span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is much more efficient than the pure matrix version</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] b;</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ... do something with row vector b[m] ...</span></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Similarly, indexing an array of column vectors is more efficient than using the <code>col</code> function to pick out a column of a matrix.</p>
+<p>In contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[N, k] x;    <span class="co">// predictors (aka covariates)</span></span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[K] beta;   <span class="co">// coeffs</span></span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] y_hat;  <span class="co">// linear prediction</span></span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>y_hat = x * beta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>than it is to do this</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[N] <span class="dt">row_vector</span>[K] x;    <span class="co">// predictors (aka covariates)</span></span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[K] beta;   <span class="co">// coeffs</span></span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] y_hat;  <span class="co">// linear prediction</span></span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>  y_hat[n] = x[n] * beta;</span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="row-vectors-vs.-one-dimensional-arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="row-vectors-vs.-one-dimensional-arrays">(Row) vectors vs.&nbsp;one-dimensional arrays</h3>
+<p>For use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The <code>Eigen::Vector</code> template specialization and the <code>std::vector</code> template class are implemented similarly as containers of <code>double</code> values (the type <code>real</code> in Stan). Only arrays in Stan are allowed to store integer values.</p>
+</section>
+</section>
+<section id="memory-locality" class="level2">
+<h2 class="anchored" data-anchor-id="memory-locality">Memory locality</h2>
+<p>The key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.</p>
+<section id="memory-locality-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="memory-locality-1">Memory locality</h3>
+<p>CPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is <em>much</em> slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.</p>
+</section>
+<section id="matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrices">Matrices</h3>
+<p>Matrices are stored internally in column-major order. That is, an <span class="math inline">\(M
+\times N\)</span> matrix stores its elements in the order <span class="math display">\[
+(1,1), (2, 1), \dotsc, (M, 1), (1, 2), \dotsc, (M, 2), \dotsc, (1, N),
+\dotsc, (M, N).
+\]</span></p>
+<p>This means that it’s much more efficient to write loops over matrices column by column, as in the following example.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[M, N] a;</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>     <span class="co">// ... do something with a[m, n] ...</span></span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M] <span class="dt">row_vector</span>[N] a;</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... do something with row vector a[m] ...</span></span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Even if what is done involves a function call, the row vector <code>a[m]</code> will not have to be copied.</p>
+</section>
+<section id="arrays" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="arrays">Arrays</h3>
+<p>Arrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[M, N] <span class="dt">real</span> a;</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... do something with a[m] ...</span></span>
+<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A difference with matrices is that the entries <code>a[m]</code> in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”</p>
+</section>
+</section>
+<section id="converting-among-matrix-vector-and-array-types" class="level2">
+<h2 class="anchored" data-anchor-id="converting-among-matrix-vector-and-array-types">Converting among matrix, vector, and array types</h2>
+<p>There is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the <a href="../stan-users-guide/multi-indexing.html">multi-indexing chapter</a> for some reshaping operations involving multiple indexing and range indexing.</p>
+</section>
+<section id="aliasing-in-stan-containers" class="level2">
+<h2 class="anchored" data-anchor-id="aliasing-in-stan-containers">Aliasing in Stan containers</h2>
+<p>Stan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to <code>u</code> and the compound slicing assignment to <code>x</code>, where <code>u</code> and <code>x</code> start with the same values.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">4</span>] x = [ <span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span>, <span class="dv">4</span> ]';</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">4</span>] u = [ <span class="dv">1</span>, <span class="dv">2</span>, <span class="dv">3</span>, <span class="dv">4</span> ]';</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:<span class="dv">4</span>) {</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>    u[t] = u[t - <span class="dv">1</span>] * <span class="dv">3</span>;</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a>  x[<span class="dv">2</span>:<span class="dv">4</span>] = x[<span class="dv">1</span>:<span class="dv">3</span>] * <span class="dv">3</span>;</span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>  <span class="kw">print</span>(<span class="st">"u = "</span>, u);</span>
+<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a>  <span class="kw">print</span>(<span class="st">"x = "</span>, x);</span>
+<span id="cb19-13"><a href="#cb19-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The output it produces is,</p>
+<pre><code>u = [1, 3, 9, 27]
+x = [1, 3, 6, 9]</code></pre>
+<p>In the loop version assigning to <code>u</code>, the values are updated before being used to define subsequent values; in the sliced expression assigning to <code>x</code>, the entire right-hand side is evaluated before assigning to the left-hand side.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/floating-point.html" class="pagination-link" aria-label="Floating Point Arithmetic">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Floating Point Arithmetic</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/multi-indexing.html" class="pagination-link" aria-label="Multiple Indexing and Range Indexing">
+        <span class="nav-page-text">Multiple Indexing and Range Indexing</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/matrices-arrays.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/measurement-error.html b/docs/2_39/stan-users-guide/measurement-error.html
new file mode 100644
index 000000000..3fb2fee0e
--- /dev/null
+++ b/docs/2_39/stan-users-guide/measurement-error.html
@@ -0,0 +1,1467 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Measurement Error and Meta-Analysis</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/latent-discrete.html" rel="next">
+<link href="../stan-users-guide/finite-mixtures.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Measurement Error and Meta-Analysis">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/measurement-error.html">Measurement Error and Meta-Analysis</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#measurement-error-and-meta-analysis" id="toc-measurement-error-and-meta-analysis" class="nav-link active" data-scroll-target="#measurement-error-and-meta-analysis">Measurement Error and Meta-Analysis</a>
+  <ul class="collapse">
+  <li><a href="#bayesian-measurement-error-model" id="toc-bayesian-measurement-error-model" class="nav-link" data-scroll-target="#bayesian-measurement-error-model">Bayesian measurement error model</a>
+  <ul class="collapse">
+  <li><a href="#regression-with-measurement-error" id="toc-regression-with-measurement-error" class="nav-link" data-scroll-target="#regression-with-measurement-error">Regression with measurement error</a></li>
+  <li><a href="#rounding" id="toc-rounding" class="nav-link" data-scroll-target="#rounding">Rounding</a></li>
+  </ul></li>
+  <li><a href="#meta-analysis" id="toc-meta-analysis" class="nav-link" data-scroll-target="#meta-analysis">Meta-analysis</a>
+  <ul class="collapse">
+  <li><a href="#treatment-effects-in-controlled-studies" id="toc-treatment-effects-in-controlled-studies" class="nav-link" data-scroll-target="#treatment-effects-in-controlled-studies">Treatment effects in controlled studies</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/measurement-error.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/measurement-error.html">Measurement Error and Meta-Analysis</a></li></ol></nav></header>
+
+
+
+
+<section id="measurement-error-and-meta-analysis" class="level1">
+<h1>Measurement Error and Meta-Analysis</h1>
+<p>Most quantities used in statistical models arise from measurements. Most of these measurements are taken with some error. When the measurement error is small relative to the quantity being measured, its effect on a model is usually small. When measurement error is large relative to the quantity being measured, or when precise relations can be estimated being measured quantities, it is useful to introduce an explicit model of measurement error. One kind of measurement error is rounding.</p>
+<p>Meta-analysis plays out statistically much like measurement error models, where the inferences drawn from multiple data sets are combined to do inference over all of them. Inferences for each data set are treated as providing a kind of measurement error with respect to true parameter values.</p>
+<section id="bayesian-measurement-error-model" class="level2">
+<h2 class="anchored" data-anchor-id="bayesian-measurement-error-model">Bayesian measurement error model</h2>
+<p>A Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data <span class="citation" data-cites="Clayton:1992 RichardsonGilks:1993">(<a href="#ref-Clayton:1992" role="doc-biblioref">Clayton 1992</a>; <a href="#ref-RichardsonGilks:1993" role="doc-biblioref">Richardson and Gilks 1993</a>)</span>. This requires a model of how the measurements are derived from the true values.</p>
+<section id="regression-with-measurement-error" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="regression-with-measurement-error">Regression with measurement error</h3>
+<p>Before considering regression with measurement error, first consider a linear regression model where the observed data for <span class="math inline">\(N\)</span> cases includes a predictor <span class="math inline">\(x_n\)</span> and outcome <span class="math inline">\(y_n\)</span>. In Stan, a linear regression for <span class="math inline">\(y\)</span> based on <span class="math inline">\(x\)</span> with a slope and intercept is modeled as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;       <span class="co">// number of cases</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;          <span class="co">// predictor (covariate)</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;          <span class="co">// outcome (variate)</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;           <span class="co">// intercept</span></span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;            <span class="co">// slope</span></span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;  <span class="co">// outcome noise</span></span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now suppose that the true values of the predictors <span class="math inline">\(x_n\)</span> are not known, but for each <span class="math inline">\(n\)</span>, a measurement <span class="math inline">\(x^{\textrm{meas}}_n\)</span> of <span class="math inline">\(x_n\)</span> is available. If the error in measurement can be modeled, the measured value <span class="math inline">\(x^{\textrm{meas}}_n\)</span> can be modeled in terms of the true value <span class="math inline">\(x_n\)</span> plus measurement noise. The true value <span class="math inline">\(x_n\)</span> is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation <span class="math inline">\(\tau\)</span>. This leads to the following regression model with constant measurement error.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x_meas;   <span class="co">// measurement of x</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; tau;     <span class="co">// measurement noise</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;    <span class="co">// unknown true value</span></span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_x;          <span class="co">// prior location</span></span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma_x;       <span class="co">// prior scale</span></span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>  x ~ normal(mu_x, sigma_x);  <span class="co">// prior</span></span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>  x_meas ~ normal(x, tau);    <span class="co">// measurement model</span></span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The regression coefficients <code>alpha</code> and <code>beta</code> and regression noise scale <code>sigma</code> are the same as before, but now <code>x</code> is declared as a parameter rather than as data. The data are now <code>x_meas</code>, which is a measurement of the true <code>x</code> value with noise scale <code>tau</code>. The model then specifies that the measurement error for <code>x_meas[n]</code> given true value <code>x[n]</code> is normal with deviation <code>tau</code>. Furthermore, the true values <code>x</code> are given a hierarchical prior here.</p>
+<p>In cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, <span class="citation" data-cites="Clayton:1992">Clayton (<a href="#ref-Clayton:1992" role="doc-biblioref">1992</a>)</span> introduces an exposure model for the unknown (but noisily measured) risk factors <span class="math inline">\(x\)</span> in terms of known (without measurement error) risk factors <span class="math inline">\(c\)</span>. A simple model would regress <span class="math inline">\(x_n\)</span> on the covariates <span class="math inline">\(c_n\)</span> with noise term <span class="math inline">\(\upsilon\)</span>, <span class="math display">\[
+x_n \sim \textsf{normal}(\gamma^{\top}c, \upsilon).
+\]</span> This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.</p>
+</section>
+<section id="rounding" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="rounding">Rounding</h3>
+<p>A common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.</p>
+<p>Exercise 3.5(b) by <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> provides an example.</p>
+<blockquote class="blockquote">
+<p>3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\sigma^2\)</span>.</p>
+<ol start="2" type="a">
+<li>Give the correct posterior distribution for <span class="math inline">\((\mu, \sigma^2)\)</span>, treating the measurements as rounded.</li>
+</ol>
+</blockquote>
+<p>Letting <span class="math inline">\(z_n\)</span> be the unrounded measurement for <span class="math inline">\(y_n\)</span>, the problem as stated assumes <span class="math display">\[
+z_n \sim \textsf{normal}(\mu, \sigma).
+\]</span></p>
+<p>The rounding process entails that <span class="math inline">\(z_n \in (y_n - 0.5, y_n + 0.5)\)</span><a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a>. The probability mass function for the discrete observation <span class="math inline">\(y\)</span> is then given by marginalizing out the unrounded measurement, producing the likelihood <span class="math display">\[\begin{align*}
+p(y_n \mid \mu, \sigma)
+&amp;= \int_{y_n - 0.5}^{y_n + 0.5} \textsf{normal}(z_n \mid \mu, \sigma) \,\textsf{d}z_n \\
+&amp;= \Phi\!\left(\frac{y_n + 0.5 - \mu}{\sigma}\right)
+   -\Phi\!\left(\frac{y_n - 0.5 - \mu}{\sigma}\right).
+\end{align*}\]</span> Gelman’s answer for this problem took the noninformative prior to be uniform in the variance <span class="math inline">\(\sigma^2\)</span> on the log scale, but we replace it with more recently recommended half-normal prior on <span class="math inline">\(\sigma\)</span> <span class="math display">\[
+\sigma \sim \textsf{normal}^+(0, 1).
+\]</span> The posterior after observing <span class="math inline">\(y = (10, 10, 12, 11, 9)\)</span> can be calculated by Bayes’s rule as <span class="math display">\[\begin{align*}
+p(\mu, \sigma \mid y)
+&amp;\propto p(\mu, \sigma) \ p(y \mid \mu, \sigma) \\
+&amp;\propto \textsf{normal}^+(\sigma \mid 0, 1)\prod_{n=1}^5
+  \left( \Phi\!\left(\frac{y_n + 0.5 - \mu}{\sigma}\right)
+        -\Phi\!\left(\frac{y_n - 0.5 - \mu}{\sigma}\right)
+  \right).
+\end{align*}\]</span></p>
+<p>The Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_diff_exp(normal_lcdf(y[n] + <span class="fl">0.5</span> | mu, sigma),</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>                           normal_lcdf(y[n] - <span class="fl">0.5</span> | mu, sigma));</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where <code>normal_lcdf(y[n]+0.5 | mu, sigma)</code> is equal to <code>log(Phi((y[n] + 0.5 - mu) / sigma))</code>, and <code>log_diff_exp(a, b)</code> computes <code>log(exp(a) - exp(b))</code> in numerically more stable way.</p>
+<p>Alternatively, the model may be defined with latent parameters for the unrounded measurements <span class="math inline">\(z_n\)</span>. The Stan code in this case uses a distribution statement for <span class="math inline">\(z_n\)</span> directly while respecting the constraint <span class="math inline">\(z_n \in (y_n - 0.5, y_n + 0.5)\)</span>.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=y<span class="fl">-0.5</span>, <span class="kw">upper</span>=y<span class="fl">+0.5</span>&gt;[N] z;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  z ~ normal(mu, sigma);</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This explicit model for the unrounded measurements <span class="math inline">\(z\)</span> produces the same posterior for <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\sigma\)</span> as the previous model that marginalizes <span class="math inline">\(z\)</span> out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.</p>
+</section>
+</section>
+<section id="meta-analysis" class="level2">
+<h2 class="anchored" data-anchor-id="meta-analysis">Meta-analysis</h2>
+<p>Meta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.</p>
+<p>The Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.</p>
+<section id="treatment-effects-in-controlled-studies" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="treatment-effects-in-controlled-studies">Treatment effects in controlled studies</h3>
+<p>Suppose the data in question arise from a total of <span class="math inline">\(M\)</span> studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen <span class="citation" data-cites="WarnThompsonSpiegelhalter:2002">(<a href="#ref-WarnThompsonSpiegelhalter:2002" role="doc-biblioref">Warn, Thompson, and Spiegelhalter 2002</a>)</span> or mortality after myocardial infarction under a treatment of beta blockers <span class="citation" data-cites="GelmanEtAl:2013">(<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">Gelman et al. 2013, sec. 5.6</a>)</span>.</p>
+<section id="data" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="data">Data</h4>
+<p>The clinical data consists of <span class="math inline">\(J\)</span> trials, each with <span class="math inline">\(n^t\)</span> treatment cases, <span class="math inline">\(n^c\)</span> control cases, <span class="math inline">\(r^t\)</span> successful outcomes among those treated and <span class="math inline">\(r^c\)</span> successful outcomes among those in the control group. This data can be declared in Stan as follows.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; n_t;  <span class="co">// num cases, treatment</span></span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; r_t;  <span class="co">// num successes, treatment</span></span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; n_c;  <span class="co">// num cases, control</span></span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; r_c;  <span class="co">// num successes, control</span></span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="converting-to-log-odds-and-standard-error" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="converting-to-log-odds-and-standard-error">Converting to log odds and standard error</h4>
+<p>Although the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio <span class="math display">\[\begin{align*}
+y_j &amp;= \log \left( \frac{r^t_j / (n^t_j - r^t_j)}
+                        {r^c_j / (n^c_j - r^c_j)}
+        \right) \\
+&amp;= \log \left( \frac{r^t_j}{n^t_j - r^t_j} \right)
+   -\log \left( \frac{r^c_j}{n^c_j - r^c_j} \right)
+\end{align*}\]</span> and corresponding standard errors <span class="math display">\[
+\sigma_j = \sqrt{
+  \frac{1}{r^T_j}
++ \frac{1}{n^T_j - r^T_j}
++ \frac{1}{r^C_j}
++ \frac{1}{n^C_j - r^C_j}
+}.
+\]</span></p>
+<p>The log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> y;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>            - (log(r_c[j]) - log(n_c[j] - r_c[j]));</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>    sigma[j] = sqrt(<span class="dv">1</span> / r_t[j] + <span class="dv">1</span> / (n_t[j] - r_t[j])</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>                     + <span class="dv">1</span> / r_c[j] + <span class="dv">1</span> / (n_c[j] - r_c[j]));</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.</p>
+</section>
+<section id="non-hierarchical-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="non-hierarchical-model">Non-hierarchical model</h4>
+<p>With the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;  <span class="co">// global treatment effect, log odds</span></span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>  y ~ normal(theta, sigma);</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The distribution statement for <code>y</code> is vectorized; it has the same effect as the following.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>    y[j] ~ normal(theta, sigma[j]);</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is common to include a prior for <code>theta</code> in this model, but it is not strictly necessary for the model to be proper because <code>y</code> is fixed and <span class="math inline">\(\textsf{normal}(y \mid \mu,\sigma) =
+\textsf{normal}(\mu \mid y,\sigma)\)</span>.</p>
+</section>
+<section id="hierarchical-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="hierarchical-model">Hierarchical model</h4>
+<p>To model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> theta;  <span class="co">// per-trial treatment effect</span></span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;              <span class="co">// mean treatment effect</span></span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; tau;    <span class="co">// deviation of treatment effects</span></span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  y ~ normal(theta, sigma);</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(mu, tau);</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>  tau ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although the vectorized distribution statement for <code>y</code> appears unchanged, the parameter <code>theta</code> is now a vector. The distribution statement for <code>theta</code> is also vectorized, with the hyperparameters <code>mu</code> and <code>tau</code> themselves being given wide priors compared to the scale of the data.</p>
+<p><span class="citation" data-cites="Rubin:1981">Rubin (<a href="#ref-Rubin:1981" role="doc-biblioref">1981</a>)</span> provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.</p>
+</section>
+<section id="extensions-and-alternatives" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="extensions-and-alternatives">Extensions and alternatives</h4>
+<p><span class="citation" data-cites="SmithSpiegelhalterThomas:1995">Smith, Spiegelhalter, and Thomas (<a href="#ref-SmithSpiegelhalterThomas:1995" role="doc-biblioref">1995</a>)</span> and <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013, sec. 19.4</a>)</span> provide meta-analyses based directly on binomial data. <span class="citation" data-cites="WarnThompsonSpiegelhalter:2002">Warn, Thompson, and Spiegelhalter (<a href="#ref-WarnThompsonSpiegelhalter:2002" role="doc-biblioref">2002</a>)</span> consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.</p>
+<p>If trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects <span class="math inline">\(\theta_j\)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Clayton:1992" class="csl-entry" role="listitem">
+Clayton, D. G. 1992. <span>“Models for the Analysis of Cohort and Case-Control Studies with Inaccurately Measured Exposures.”</span> In <em>Statistical Models for Longitudinal Studies of Exposure and Health</em>, edited by James H. Dwyer, Manning Feinleib, Peter Lippert, and Hans Hoffmeister, 301–31. New York: Oxford University Press.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-RichardsonGilks:1993" class="csl-entry" role="listitem">
+Richardson, Sylvia, and Walter R. Gilks. 1993. <span>“A <span>B</span>ayesian Approach to Measurement Error Problems in Epidemiology Using Conditional Independence Models.”</span> <em>American Journal of Epidemiology</em> 138 (6): 430–42.
+</div>
+<div id="ref-Rubin:1981" class="csl-entry" role="listitem">
+Rubin, Donald B. 1981. <span>“Estimation in Parallel Randomized Experiments.”</span> <em>Journal of Educational Statistics</em> 6: 377–401.
+</div>
+<div id="ref-SmithSpiegelhalterThomas:1995" class="csl-entry" role="listitem">
+Smith, Teresa C., David J. Spiegelhalter, and Andrew Thomas. 1995. <span>“<span>B</span>ayesian Approaches to Random-Effects Meta-Analysis: A Comparative Study.”</span> <em>Statistics in Medicine</em> 14 (24): 2685–99.
+</div>
+<div id="ref-WarnThompsonSpiegelhalter:2002" class="csl-entry" role="listitem">
+Warn, David E., S. G. Thompson, and David J. Spiegelhalter. 2002. <span>“<span>B</span>ayesian Random Effects Meta-Analysis of Trials with Binary Outcomes: Methods for the Absolute Risk Difference and Relative Risk Scales.”</span> <em>Statistics in Medicine</em> 21: 1601–23.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>There are several different rounding rules (see, e.g., <a href="https://en.wikipedia.org/wiki/Rounding">Wikipedia: Rounding</a>), which affect which interval ends are open and which are closed, but these do not matter here as for continuous <span class="math inline">\(z_n\)</span> <span class="math inline">\(p(z_n=y_n-0.5)=p(z_n=y_n+0.5)=0\)</span>.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Stan’s integer constraints are not powerful enough to express the constraint that <span class="math inline">\(\texttt{r}\mathtt{\_}\texttt{t[j]} \leq \texttt{n}\mathtt{\_}\texttt{t[j]}\)</span>, but this constraint could be checked in the transformed data block.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>When dividing two integers, the result type is an integer and rounding will ensue if the result is not exact. See the discussion of primitive arithmetic types in the reference manual for more information.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/finite-mixtures.html" class="pagination-link" aria-label="Finite Mixtures">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Finite Mixtures</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/latent-discrete.html" class="pagination-link" aria-label="Latent Discrete Parameters">
+        <span class="nav-page-text">Latent Discrete Parameters</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/measurement-error.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/missing-data.html b/docs/2_39/stan-users-guide/missing-data.html
new file mode 100644
index 000000000..38af5f613
--- /dev/null
+++ b/docs/2_39/stan-users-guide/missing-data.html
@@ -0,0 +1,1415 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Missing Data and Partially Known Parameters</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/truncation-censoring.html" rel="next">
+<link href="../stan-users-guide/time-series.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Missing Data and Partially Known Parameters">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/missing-data.html">Missing Data and Partially Known Parameters</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#missing-data-and-partially-known-parameters" id="toc-missing-data-and-partially-known-parameters" class="nav-link active" data-scroll-target="#missing-data-and-partially-known-parameters">Missing Data and Partially Known Parameters</a>
+  <ul class="collapse">
+  <li><a href="#missing-data" id="toc-missing-data" class="nav-link" data-scroll-target="#missing-data">Missing data</a></li>
+  <li><a href="#partially-known-parameters.section" id="toc-partially-known-parameters.section" class="nav-link" data-scroll-target="#partially-known-parameters.section">Partially known parameters</a></li>
+  <li><a href="#sliced-missing-data" id="toc-sliced-missing-data" class="nav-link" data-scroll-target="#sliced-missing-data">Sliced missing data</a></li>
+  <li><a href="#loading-matrix-for-factor-analysis" id="toc-loading-matrix-for-factor-analysis" class="nav-link" data-scroll-target="#loading-matrix-for-factor-analysis">Loading matrix for factor analysis</a></li>
+  <li><a href="#missing-multivariate-data" id="toc-missing-multivariate-data" class="nav-link" data-scroll-target="#missing-multivariate-data">Missing multivariate data</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/missing-data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/missing-data.html">Missing Data and Partially Known Parameters</a></li></ol></nav></header>
+
+
+
+
+<section id="missing-data-and-partially-known-parameters" class="level1">
+<h1>Missing Data and Partially Known Parameters</h1>
+<p>Bayesian inference supports a general approach to missing data in which any missing data item is represented as a parameter that is estimated in the posterior <span class="citation" data-cites="GelmanEtAl:2013">(<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">Gelman et al. 2013</a>)</span>. If the missing data are not explicitly modeled, as in the predictors for most regression models, then the result is an improper prior on the parameter representing the missing predictor.</p>
+<p>Mixing arrays of observed and missing data can be difficult to include in Stan, partly because it can be tricky to model discrete unknowns in Stan and partly because unlike some other statistical languages (for example, R and Bugs), Stan requires observed and unknown quantities to be defined in separate places in the model. Thus it can be necessary to include code in a Stan program to splice together observed and missing parts of a data structure. Examples are provided later in the chapter.</p>
+<section id="missing-data" class="level2">
+<h2 class="anchored" data-anchor-id="missing-data">Missing data</h2>
+<p>Stan treats variables declared in the <code>data</code> and <code>transformed data</code> blocks as known and the variables in the <code>parameters</code> block as unknown.</p>
+<p>An example involving missing normal observations could be coded as follows.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_mis;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">real</span> y_obs;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">real</span> y_mis;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  y_obs ~ normal(mu, sigma);</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  y_mis ~ normal(mu, sigma);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The number of observed and missing data points are coded as data with non-negative integer variables <code>N_obs</code> and <code>N_mis</code>. The observed data are provided as an array data variable <code>y_obs</code>. The missing data are coded as an array parameter, <code>y_mis</code>. The ordinary parameters being estimated, the location <code>mu</code> and scale <code>sigma</code>, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.</p>
+</section>
+<section id="partially-known-parameters.section" class="level2">
+<h2 class="anchored" data-anchor-id="partially-known-parameters.section">Partially known parameters</h2>
+<p>In some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the <code>transformed parameters</code> block and assigning to it.</p>
+<p>The following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[<span class="dv">2</span>] y;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; var1;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; var2;</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; max_cov = sqrt(var1 * var2);</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt; min_cov = -max_cov;</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] mu;</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=min_cov, <span class="kw">upper</span>=max_cov&gt; cov;</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">2</span>, <span class="dv">2</span>] Sigma;</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>  Sigma[<span class="dv">1</span>, <span class="dv">1</span>] = var1;</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>  Sigma[<span class="dv">1</span>, <span class="dv">2</span>] = cov;</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>  Sigma[<span class="dv">2</span>, <span class="dv">1</span>] = cov;</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>  Sigma[<span class="dv">2</span>, <span class="dv">2</span>] = var2;</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, Sigma);</span>
+<span id="cb2-24"><a href="#cb2-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The variances are defined as data in variables <code>var1</code> and <code>var2</code>, whereas the covariance is defined as a parameter in variable <code>cov</code>. The <span class="math inline">\(2 \times 2\)</span> covariance matrix <code>Sigma</code> is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.</p>
+<p>The constraint on the covariance declaration ensures that the resulting covariance matrix <code>sigma</code> is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.</p>
+<p>The vectorization of the multivariate normal is critical for efficiency here. The transformed parameter <code>Sigma</code> could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.</p>
+</section>
+<section id="sliced-missing-data" class="level2">
+<h2 class="anchored" data-anchor-id="sliced-missing-data">Sliced missing data</h2>
+<p>If the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_mis;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N_obs + N_mis&gt; ii_obs;</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N_obs + N_mis&gt; ii_mis;</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">real</span> y_obs;</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N = N_obs + N_mis;</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">real</span> y_mis;</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb3-17"><a href="#cb3-17" aria-hidden="true" tabindex="-1"></a>  y[ii_obs] = y_obs;</span>
+<span id="cb3-18"><a href="#cb3-18" aria-hidden="true" tabindex="-1"></a>  y[ii_mis] = y_mis;</span>
+<span id="cb3-19"><a href="#cb3-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-20"><a href="#cb3-20" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-21"><a href="#cb3-21" aria-hidden="true" tabindex="-1"></a>  sigma ~ gamma(<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb3-22"><a href="#cb3-22" aria-hidden="true" tabindex="-1"></a>  y[<span class="dv">1</span>] ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
+<span id="cb3-23"><a href="#cb3-23" aria-hidden="true" tabindex="-1"></a>  y[<span class="dv">2</span>:N] ~ normal(y[<span class="dv">1</span>:(N - <span class="dv">1</span>)], sigma);</span>
+<span id="cb3-24"><a href="#cb3-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The index arrays <code>ii_obs</code> and <code>ii_mis</code> contain the indexes into the final array <code>y</code> of the observed data (coded as a data vector <code>y_obs</code>) and the missing data (coded as a parameter vector <code>y_mis</code>). See the <a href="../stan-users-guide/time-series.html">time series chapter</a> for further discussion of time-series model and specifically the <a href="../stan-users-guide/time-series.html#autoregressive">autoregression section</a> for an explanation of the vectorization for <code>y</code> as well as an explanation of how to convert this example to a full AR(1) model. To ensure <code>y[1]</code> has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.</p>
+<p>Another potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>x[N_obs_2, <span class="dv">2</span>] = x_obs_2;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>x[N_mis_2, <span class="dv">2</span>] = x_mis_2;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the relevant variables are all hard coded with index <code>2</code> because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the <a href="../stan-users-guide/sparse-ragged.html#ragged-data-structs.section">ragged data structures section</a> for a general discussion of coding ragged data structures in Stan).</p>
+</section>
+<section id="loading-matrix-for-factor-analysis" class="level2">
+<h2 class="anchored" data-anchor-id="loading-matrix-for-factor-analysis">Loading matrix for factor analysis</h2>
+<p>Rick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis <span class="citation" data-cites="aguilar-west:2000">(<a href="#ref-aguilar-west:2000" role="doc-biblioref">Aguilar and West 2000</a>)</span>. This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K_choose_2;</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  K_choose_2 = (K * (K - <span class="dv">1</span>)) / <span class="dv">2</span>;</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K_choose_2] L_lower;</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_cov</span>[K] L;</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>    L[k, k] = <span class="dv">1</span>;</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> i;</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">2</span>:K) {</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:(m - <span class="dv">1</span>)) {</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>        L[m, n] = L_lower[i];</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>        L[n, m] = <span class="dv">0</span>;</span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>        i += <span class="dv">1</span>;</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is most convenient to place a prior directly on <code>L_lower</code>. An alternative would be a prior for the full Cholesky factor <code>L</code>, because the transform from <code>L_lower</code> to <code>L</code> is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix <code>L   * L'</code>, because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.</p>
+</section>
+<section id="missing-multivariate-data" class="level2">
+<h2 class="anchored" data-anchor-id="missing-multivariate-data">Missing multivariate data</h2>
+<p>It’s often the case that one or more components of a multivariate outcome are missing.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>As an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations <code>y</code> and a boolean array <code>y_observed</code> to indicate which values were observed (others can have dummy values in the input).</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[N] <span class="dt">vector</span>[<span class="dv">2</span>] y;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[N, <span class="dv">2</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y_observed;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (y_observed[n, <span class="dv">1</span>] &amp;&amp; y_observed[n, <span class="dv">2</span>]) {</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>    y[n] ~ multi_normal(mu, Sigma);</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> <span class="cf">if</span> (y_observed[n, <span class="dv">1</span>]) {</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>    y[n, <span class="dv">1</span>] ~ normal(mu[<span class="dv">1</span>], sqrt(Sigma[<span class="dv">1</span>, <span class="dv">1</span>]));</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> <span class="cf">if</span> (y_observed[n, <span class="dv">2</span>]) {</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>    y[n, <span class="dv">2</span>] ~ normal(mu[<span class="dv">2</span>], sqrt(Sigma[<span class="dv">2</span>, <span class="dv">2</span>]));</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[observed_12(y_observed)] <span class="dt">int</span> ns12;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[observed_1(y_observed)] <span class="dt">int</span> ns1;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[observed_2(y_observed)] <span class="dt">int</span> ns2;</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>You will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n12 = <span class="dv">1</span>;</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n1 = <span class="dv">1</span>;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> n2 = <span class="dv">1</span>;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (y_observed[n, <span class="dv">1</span>] &amp;&amp; y_observed[n, <span class="dv">2</span>]) {</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>    ns12[n12] = n;</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>    n12 += <span class="dv">1</span>;</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> <span class="cf">if</span> (y_observed[n, <span class="dv">1</span>]) {</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>    ns1[n1] = n;</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>    n1 += <span class="dv">1</span>;</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> <span class="cf">if</span> (y_observed[n, <span class="dv">2</span>]) {</span>
+<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>    ns2[n2] = n;</span>
+<span id="cb9-13"><a href="#cb9-13" aria-hidden="true" tabindex="-1"></a>    n2 += <span class="dv">1</span>;</span>
+<span id="cb9-14"><a href="#cb9-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-15"><a href="#cb9-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then, in the model block, everything is vectorizable using those indexes constructed once in transformed data:</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>y[ns12] ~ multi_normal(mu, Sigma);</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>y[ns1] ~ normal(mu[<span class="dv">1</span>], sqrt(Sigma[<span class="dv">1</span>, <span class="dv">1</span>]));</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>y[ns2] ~ normal(mu[<span class="dv">2</span>], sqrt(Sigma[<span class="dv">2</span>, <span class="dv">2</span>]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.</p>
+<p>This approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-aguilar-west:2000" class="csl-entry" role="listitem">
+Aguilar, Omar, and Mike West. 2000. <span>“Bayesian Dynamic Factor Models and Portfolio Allocation.”</span> <em>Journal of Business &amp; Economic Statistics</em> 18 (3): 338–57.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>A more meaningful estimation example would involve a regression of the observed and missing observations using predictors that were known for each and specified in the <code>data</code> block.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>This is not the same as missing components of a multivariate predictor in a regression problem; in that case, you will need to represent the missing data as a parameter and impute missing values in order to feed them into the regression.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/time-series.html" class="pagination-link" aria-label="Time-Series Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Time-Series Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/truncation-censoring.html" class="pagination-link" aria-label="Truncated or Censored Data">
+        <span class="nav-page-text">Truncated or Censored Data</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/missing-data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/multi-indexing.html b/docs/2_39/stan-users-guide/multi-indexing.html
new file mode 100644
index 000000000..ffaf5a741
--- /dev/null
+++ b/docs/2_39/stan-users-guide/multi-indexing.html
@@ -0,0 +1,1460 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Multiple Indexing and Range Indexing</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/user-functions.html" rel="next">
+<link href="../stan-users-guide/matrices-arrays.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Multiple Indexing and Range Indexing">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multi-indexing.html">Multiple Indexing and Range Indexing</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#multi-indexing.chapter" id="toc-multi-indexing.chapter" class="nav-link active" data-scroll-target="#multi-indexing.chapter">Multiple Indexing and Range Indexing</a>
+  <ul class="collapse">
+  <li><a href="#multiple-indexing" id="toc-multiple-indexing" class="nav-link" data-scroll-target="#multiple-indexing">Multiple indexing</a></li>
+  <li><a href="#slicing-with-range-indexes" id="toc-slicing-with-range-indexes" class="nav-link" data-scroll-target="#slicing-with-range-indexes">Slicing with range indexes</a>
+  <ul class="collapse">
+  <li><a href="#lower-and-upper-bound-indexes" id="toc-lower-and-upper-bound-indexes" class="nav-link" data-scroll-target="#lower-and-upper-bound-indexes">Lower and upper bound indexes</a></li>
+  <li><a href="#lower-or-upper-bound-indexes" id="toc-lower-or-upper-bound-indexes" class="nav-link" data-scroll-target="#lower-or-upper-bound-indexes">Lower or upper bound indexes</a></li>
+  <li><a href="#full-range-indexes" id="toc-full-range-indexes" class="nav-link" data-scroll-target="#full-range-indexes">Full range indexes</a></li>
+  <li><a href="#slicing-functions" id="toc-slicing-functions" class="nav-link" data-scroll-target="#slicing-functions">Slicing functions</a></li>
+  </ul></li>
+  <li><a href="#multiple-indexing-on-the-left-of-assignments" id="toc-multiple-indexing-on-the-left-of-assignments" class="nav-link" data-scroll-target="#multiple-indexing-on-the-left-of-assignments">Multiple indexing on the left of assignments</a>
+  <ul class="collapse">
+  <li><a href="#assign-by-value-and-aliasing" id="toc-assign-by-value-and-aliasing" class="nav-link" data-scroll-target="#assign-by-value-and-aliasing">Assign-by-value and aliasing</a></li>
+  </ul></li>
+  <li><a href="#multiple-indexes-with-vectors-and-matrices" id="toc-multiple-indexes-with-vectors-and-matrices" class="nav-link" data-scroll-target="#multiple-indexes-with-vectors-and-matrices">Multiple indexes with vectors and matrices</a>
+  <ul class="collapse">
+  <li><a href="#vectors" id="toc-vectors" class="nav-link" data-scroll-target="#vectors">Vectors</a></li>
+  <li><a href="#matrices" id="toc-matrices" class="nav-link" data-scroll-target="#matrices">Matrices</a></li>
+  <li><a href="#matrices-with-one-multiple-index" id="toc-matrices-with-one-multiple-index" class="nav-link" data-scroll-target="#matrices-with-one-multiple-index">Matrices with one multiple index</a></li>
+  <li><a href="#arrays-of-vectors-or-matrices" id="toc-arrays-of-vectors-or-matrices" class="nav-link" data-scroll-target="#arrays-of-vectors-or-matrices">Arrays of vectors or matrices</a></li>
+  <li><a href="#block-row-and-column-extraction-for-matrices" id="toc-block-row-and-column-extraction-for-matrices" class="nav-link" data-scroll-target="#block-row-and-column-extraction-for-matrices">Block, row, and column extraction for matrices</a></li>
+  </ul></li>
+  <li><a href="#matrices-with-parameters-and-constants" id="toc-matrices-with-parameters-and-constants" class="nav-link" data-scroll-target="#matrices-with-parameters-and-constants">Matrices with parameters and constants</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multi-indexing.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multi-indexing.html">Multiple Indexing and Range Indexing</a></li></ol></nav></header>
+
+
+
+
+<section id="multi-indexing.chapter" class="level1">
+<h1>Multiple Indexing and Range Indexing</h1>
+<p>Stan allows multiple indexes to be provided for containers (i.e., arrays, vectors, and matrices) in a single position, using either an array of integer indexes or range bounds. In many cases, there are functions that provide similar behavior.</p>
+<p>Allowing multiple indexes supports inline vectorization of models. For instance, consider the data model for a varying-slope, varying-intercept hierarchical linear regression, which could be coded as</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(alpha[ii[n]] + beta[ii[n]] * x[n], sigma);</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With multiple indexing, this can be coded in one line, leading to more efficient vectorized code.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(alpha[ii] + rows_dot_product(beta[ii], x), sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This latter version is faster than the loop version; it is equivalent in speed to the clunky assignment to a local variable.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>    mu[n] = alpha[ii[n]] + beta[ii[n]] * x[n];</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The boost in speed compared to the original version is because the single call to the normal log density in the distribution statement will be much more memory efficient than the original version.</p>
+<section id="multiple-indexing" class="level2">
+<h2 class="anchored" data-anchor-id="multiple-indexing">Multiple indexing</h2>
+<p>The following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> c;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: c == (5, 9, 7)</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">int</span> idxs;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: idxs == (3, 3, 1, 2)</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">int</span> d;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>d = c[idxs];    <span class="co">// result: d == (7, 7, 5, 9)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In general, the multiple indexed expression <code>c[idxs]</code> is defined as follows, assuming <code>idxs</code> is of size <code>K</code>.</p>
+<pre><code>c[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )</code></pre>
+<p>Thus <code>c[idxs]</code> is of the same size as <code>idxs</code>, which is <code>K</code> in this example.</p>
+<p>Multiple indexing can also be used with multi-dimensional arrays. For example, consider the following.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>] <span class="dt">int</span> c;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: c = ((1, 3, 5), (7, 11, 13))</span></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">int</span> idxs;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: idxs = (2, 2, 1, 2)</span></span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">3</span>] <span class="dt">int</span> d</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>d = c[idxs];    <span class="co">// result: d = ((7, 11, 13), (7, 11, 13),</span></span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>                <span class="co">//              (1, 3, 5), (7, 11, 13))</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>That is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by <code>c[idxs][j] ==   c[idxs[j]]</code>.</p>
+<p>Multiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">int</span> e;</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>e = c[<span class="dv">2</span>, idxs]; <span class="co">// result:  c[2] = (7, 11, 13)</span></span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>                <span class="co">// result:  e = (11, 11, 7, 11)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, <code>c[2,idxs]</code> evaluates to the same value as <code>c[2][idxs]</code>.</p>
+<p>Multiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">3</span>] <span class="dt">int</span> c;</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: c = ((1, 3, 5), (7, 11, 13))</span></span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> idxs1;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: idxs1 = (2, 2, 1)</span></span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span> idxs2;</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: idxs2 = (1, 3)</span></span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">2</span>] <span class="dt">int</span> d;</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>d = c[idxs1, idxs2];  <span class="co">// result: d = ((7, 13), (7, 13), (1, 5))</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With multiple indexes, we no longer have <code>c[idxs1, idxs2]</code> being the same as <code>c[idxs1][idxs2]</code>. Rather, the entry <code>d[i, j]</code> after executing the above is given by</p>
+<pre><code>d[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]</code></pre>
+<p>This example illustrates the operation of multiple indexing in the general case: a multiple index like <code>idxs1</code> converts an index <code>i</code> used on the result (here, <code>c[idxs1, idxs2]</code>) to index <code>idxs1[i]</code> in the variable being indexed (here, <code>c</code>). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.</p>
+</section>
+<section id="slicing-with-range-indexes" class="level2">
+<h2 class="anchored" data-anchor-id="slicing-with-range-indexes">Slicing with range indexes</h2>
+<p>Slicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.</p>
+<section id="lower-and-upper-bound-indexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lower-and-upper-bound-indexes">Lower and upper bound indexes</h3>
+<p>For instance, consider supplying an upper and lower bound for an index.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">7</span>] <span class="dt">int</span> c;</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">4</span>] <span class="dt">int</span> d;</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>d = c[<span class="dv">3</span>:<span class="dv">6</span>];  <span class="co">// result: d == (c[3], c[4], c[5], c[6])</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The range index <code>3:6</code> behaves semantically just like the multiple index <code>(3, 4, 5, 6)</code>. In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.</p>
+</section>
+<section id="lower-or-upper-bound-indexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="lower-or-upper-bound-indexes">Lower or upper bound indexes</h3>
+<p>It is also possible to supply just a lower bound, or just an upper bound. Writing <code>c[3:]</code> is just shorthand for <code>c[3:size(c)]</code>. Writing <code>c[:5]</code> is just shorthand for <code>c[1:5]</code>.</p>
+</section>
+<section id="full-range-indexes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="full-range-indexes">Full range indexes</h3>
+<p>Finally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (<code>:</code>) as the index or leaving the index position empty. In both cases, <code>c[]</code> and <code>c[:]</code> are equal to <code>c[1:size(c)]</code>, which in turn is just equal to <code>c</code>.</p>
+</section>
+<section id="slicing-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="slicing-functions">Slicing functions</h3>
+<p>Stan provides <code>head</code> and <code>tail</code> functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,</p>
+<pre><code>vector[M] a = ...;
+vector[N] b = head(a, N);</code></pre>
+<p>assigns <code>b</code> to be a vector equivalent to the first <code>N</code> elements of the vector <code>a</code>. The function <code>tail</code> works the same way for suffixes, with</p>
+<pre><code>array[M] a = ...;
+array[N] b = tail(a, N);</code></pre>
+<p>Finally, there is a segment function, which specifies a first element and number of elements. For example,</p>
+<pre><code>array[15] a = ...;
+array[3] b = segment(a, 5, 3);</code></pre>
+<p>will set <code>b</code> to be equal to <code>{ a[5], a[6], a[7] }</code>, so that it starts at element 5 of <code>a</code> and includes a total of 3 elements.</p>
+</section>
+</section>
+<section id="multiple-indexing-on-the-left-of-assignments" class="level2">
+<h2 class="anchored" data-anchor-id="multiple-indexing-on-the-left-of-assignments">Multiple indexing on the left of assignments</h2>
+<p>Multiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> a;</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span> c;</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span> idxs;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: a == (1, 2, 3);  c == (5, 9)</span></span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>               <span class="co">//         idxs = (3,2)</span></span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>a[idxs] = c;   <span class="co">// result: a == (1, 9, 5)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The result above can be worked out by noting that the assignment sets <code>a[idxs[1]]</code> (<code>a[3]</code>) to <code>c[1]</code> (<code>5</code>) and <code>a[idxs[2]]</code> (<code>a[2]</code>) to <code>c[2]</code> (<code>9</code>).</p>
+<p>The same principle applies when there are many multiple indexes, as in the following example.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">7</span>] <span class="dt">int</span> a;</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>, <span class="dv">2</span>] <span class="dt">int</span> c;</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">2</span>:<span class="dv">3</span>, <span class="dv">5</span>:<span class="dv">6</span>] = c;  <span class="co">// result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]</span></span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>                  <span class="co">//         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.</p>
+<p>Usage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">10</span>, <span class="dv">13</span>] <span class="dt">int</span> a;</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span> c;</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">4</span>, <span class="dv">2</span>:<span class="dv">3</span>] = c;  <span class="co">// result:  a[4, 2] == c[1];  a[4, 3] == c[2]</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="assign-by-value-and-aliasing" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="assign-by-value-and-aliasing">Assign-by-value and aliasing</h3>
+<p>Aliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array <code>a</code> in the following code fragment.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> a;</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: a == (5, 6, 7)</span></span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">2</span>:<span class="dv">3</span>] = a[<span class="dv">1</span>:<span class="dv">2</span>];</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ... result: a == (5, 5, 6)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The reason the value of <code>a</code> after the assignment is <span class="math inline">\((5,5,6)\)</span> rather than <span class="math inline">\((5,5,5)\)</span> is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> a;</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> idxs;</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define idxs = (2, 1, 3)</span></span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>a[idxs] = a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case, it is evident why the right-hand side needs to be copied before the assignment.</p>
+<p>It is tempting (but wrong) to think of the assignment <code>a[2:3] =   a[1:2]</code> as executing the following assignments.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ... define: a = (5, 6, 7)</span></span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">2</span>] = a[<span class="dv">1</span>];      <span class="co">// result: a = (5, 5, 7)</span></span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>a[<span class="dv">3</span>] = a[<span class="dv">2</span>];      <span class="co">// result: a = (5, 5, 5)!</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This produces a different result than executing the assignment because <code>a[2]</code>’s value changes before it is used.</p>
+</section>
+</section>
+<section id="multiple-indexes-with-vectors-and-matrices" class="level2">
+<h2 class="anchored" data-anchor-id="multiple-indexes-with-vectors-and-matrices">Multiple indexes with vectors and matrices</h2>
+<p>Multiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.</p>
+<section id="vectors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectors">Vectors</h3>
+<p>Vectors and row vectors behave exactly the same way as arrays with multiple indexes. If <code>v</code> is a vector, then <code>v[3]</code> is a scalar real value, whereas <code>v[2:4]</code> is a vector of size 3 containing the elements <code>v[2]</code>, <code>v[3]</code>, and <code>v[4]</code>.</p>
+<p>The only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">vector</span>[<span class="dv">5</span>] v;</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">7</span>] <span class="dt">int</span> idxs;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">7</span>] u;</span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>u = v[<span class="dv">2</span>, idxs];</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">7</span>] <span class="dt">real</span> w;</span>
+<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a>w = v[idxs, <span class="dv">2</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because <code>v</code> is an array of vectors, <code>v[2, idxs]</code> reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, <code>v[idxs, 2]</code> does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for <code>w</code> is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.</p>
+</section>
+<section id="matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrices">Matrices</h3>
+<p>Matrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">5</span>, <span class="dv">7</span>] m;</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a><span class="dt">row_vector</span>[<span class="dv">3</span>] rv;</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a>rv = m[<span class="dv">4</span>, <span class="dv">3</span>:<span class="dv">5</span>];    <span class="co">// result is 1 x 3</span></span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">4</span>] v;</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a>v = m[<span class="dv">2</span>:<span class="dv">5</span>, <span class="dv">3</span>];     <span class="co">// result is 3 x 1</span></span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] m2;</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a>m2 = m[<span class="dv">1</span>:<span class="dv">3</span>, <span class="dv">2</span>:<span class="dv">5</span>];  <span class="co">// result is 3 x 4</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.</p>
+</section>
+<section id="matrices-with-one-multiple-index" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="matrices-with-one-multiple-index">Matrices with one multiple index</h3>
+<p>If matrices receive a single multiple index, the result is a matrix. So if <code>m</code> is a matrix, so is <code>m[2:4]</code>. In contrast, supplying a single index, <code>m[3]</code>, produces a row vector result. That is, <code>m[3]</code> produces the same result as <code>m[3, ]</code> or <code>m[3, 1:cols(m)]</code>.</p>
+</section>
+<section id="arrays-of-vectors-or-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="arrays-of-vectors-or-matrices">Arrays of vectors or matrices</h3>
+<p>With arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>, <span class="dv">7</span>] <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] m;</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] a;</span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a>a = m[<span class="dv">1</span>, <span class="dv">2</span>:<span class="dv">3</span>];  <span class="co">// knock off first array dimension</span></span>
+<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a>a = m[<span class="dv">3</span>:<span class="dv">4</span>, <span class="dv">5</span>];  <span class="co">// knock off second array dimension</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, <code>a[i] ==   m[1, i + 1]</code>, whereas in the second case, <code>a[i] == m[i + 2,   5]</code>.</p>
+<p>Continuing the previous example, consider the following.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[<span class="dv">2</span>] b;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>b = a[<span class="dv">1</span>, <span class="dv">3</span>, <span class="dv">2</span>:<span class="dv">3</span>, <span class="dv">2</span>];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, <code>b[j] == a[1, 3, 1 + j, 2]</code>.</p>
+<p>This last example illustrates an important point: if there is a lower-bounded index, such as <code>2:3</code>, with lower bound 2, then the lower bound minus one is added to the index, as seen in the <code>1 + j</code> expression above.</p>
+<p>Continuing further, consider continuing with the following.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">row_vector</span>[<span class="dv">3</span>] c;</span>
+<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>c = a[<span class="dv">4</span>:<span class="dv">5</span>, <span class="dv">3</span>, <span class="dv">1</span>, <span class="dv">2</span>: ];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by <code>c[i, j] == a[i + 3, 3, 1, j + 1]</code></p>
+</section>
+<section id="block-row-and-column-extraction-for-matrices" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="block-row-and-column-extraction-for-matrices">Block, row, and column extraction for matrices</h3>
+<p>Matrix slicing can also be performed using the <code>block</code> function. For example,</p>
+<pre><code>matrix[20, 20] a = ...;
+matrix[3, 2] b = block(a, 5, 9, 3, 2);</code></pre>
+<p>will set <code>b</code> equal to the submatrix of <code>a</code> starting at index [5, 9] and extending 3 rows and 2 columns. Thus <code>block(a, 5, 9, 3, 2)</code> is equivalent to <code>b[5:7, 9:10]</code>.</p>
+<p>The <code>sub_col</code> function extracts a slice of a column of a matrix as a vector. For example,</p>
+<pre><code>matrix[10, 10] a = ...;
+vector b = sub_col(a, 2, 3, 5);</code></pre>
+<p>will set <code>b</code> equal to the vector <code>a[2:6, 3]</code>, taking the element starting at [2, 3], then extending for a total of 5 rows. The function <code>sub_row</code> works the same way for extracting a slice of a row as a row vector. For example, <code>sub_row(a, 2, 3, 5)</code> is equal to the row vector <code>a[2, 3:7]</code>, which also starts at position [2, 3] then extends for a total of 5 columns.</p>
+</section>
+</section>
+<section id="matrices-with-parameters-and-constants" class="level2">
+<h2 class="anchored" data-anchor-id="matrices-with-parameters-and-constants">Matrices with parameters and constants</h2>
+<p>Suppose you have a <span class="math inline">\(3 \times 3\)</span> matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.</p>
+<p>Suppose a <span class="math inline">\(3 \times 3\)</span> matrix is known to be zero at indexes <span class="math inline">\([1,2]\)</span> and <span class="math inline">\([1,3]\)</span>. The indexes for parameters are included in a “melted” data-frame or database format.</p>
+<pre><code>transformed data {
+  array[7, 2] int&lt;lower=1, upper=3&gt; idxs
+    = { {1, 1},
+        {2, 1}, {2, 2}, {2, 3},
+        {3, 1}, {3, 2}, {3, 3} };
+  // ...</code></pre>
+<p>The seven remaining parameters are declared as a vector.</p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">7</span>] A_raw;</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb28-4"><a href="#cb28-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then the full matrix <code>A</code> is constructed in the model block as a local variable.</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">3</span>] A;</span>
+<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">7</span>) {</span>
+<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a>    A[idxs[i, <span class="dv">1</span>], idxs[i, <span class="dv">2</span>]] = A_raw[i];</span>
+<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a>  A[<span class="dv">1</span>, <span class="dv">2</span>] = <span class="dv">0</span>;</span>
+<span id="cb29-7"><a href="#cb29-7" aria-hidden="true" tabindex="-1"></a>  A[<span class="dv">1</span>, <span class="dv">3</span>] = <span class="dv">0</span>;</span>
+<span id="cb29-8"><a href="#cb29-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb29-9"><a href="#cb29-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the <code>idxs</code> array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/matrices-arrays.html" class="pagination-link" aria-label="Matrices, Vectors, Arrays, and Tuples">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Matrices, Vectors, Arrays, and Tuples</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/user-functions.html" class="pagination-link" aria-label="User-Defined Functions">
+        <span class="nav-page-text">User-Defined Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multi-indexing.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/multiple-imputation.html b/docs/2_39/stan-users-guide/multiple-imputation.html
new file mode 100644
index 000000000..5e4b37331
--- /dev/null
+++ b/docs/2_39/stan-users-guide/multiple-imputation.html
@@ -0,0 +1,1450 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Multiple Imputation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/using-stanc.html" rel="next">
+<link href="../stan-users-guide/bootstrap.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Multiple Imputation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multiple-imputation.html">Multiple imputation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#multiple-imputation" id="toc-multiple-imputation" class="nav-link active" data-scroll-target="#multiple-imputation">Multiple imputation</a>
+  <ul class="collapse">
+  <li><a href="#motivation-and-outline" id="toc-motivation-and-outline" class="nav-link" data-scroll-target="#motivation-and-outline">Motivation and outline</a>
+  <ul class="collapse">
+  <li><a href="#pseudocode-for-outline" id="toc-pseudocode-for-outline" class="nav-link" data-scroll-target="#pseudocode-for-outline">Pseudocode for outline</a></li>
+  </ul></li>
+  <li><a href="#imputing-one-variable-in-stan" id="toc-imputing-one-variable-in-stan" class="nav-link" data-scroll-target="#imputing-one-variable-in-stan">Imputing one variable in Stan</a></li>
+  <li><a href="#imputing-two-or-more-variables" id="toc-imputing-two-or-more-variables" class="nav-link" data-scroll-target="#imputing-two-or-more-variables">Imputing two or more variables</a>
+  <ul class="collapse">
+  <li><a href="#imputing-two-variables-in-stan" id="toc-imputing-two-variables-in-stan" class="nav-link" data-scroll-target="#imputing-two-variables-in-stan">Imputing two variables in Stan</a></li>
+  </ul></li>
+  <li><a href="#combining-posterior-draws" id="toc-combining-posterior-draws" class="nav-link" data-scroll-target="#combining-posterior-draws">Combining posterior draws</a></li>
+  <li><a href="#cut-models" id="toc-cut-models" class="nav-link" data-scroll-target="#cut-models">Cut models</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multiple-imputation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multiple-imputation.html">Multiple imputation</a></li></ol></nav></header>
+
+
+
+
+<section id="multiple-imputation" class="level1">
+<h1>Multiple imputation</h1>
+<p>Missing data is common in applied data analysis. Ignoring this missingness can distort posterior inferences and reduce their precision, and modeling it can improve inference quality. There are several ways to model missing data. Chapter 18 in <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> offers an approximately Bayesian perspective, and the chapter <a href="https://mc-stan.org/docs/stan-users-guide/missing-data.html"><em>Missing Data and Partially Known Parameters</em></a> in the <em>Reference Manual</em> shows an approach that handles missing data as parameters.</p>
+<p>Another way to model missing data is through multiple imputation, which replaces missing values with sampled values to obtain different versions of a complete data set. Then, the model of interest can be fitted to each of these complete versions separately. Combining the resulting draws from these different fits produces a sample that accounts for the uncertainty in the missing values.</p>
+<section id="motivation-and-outline" class="level2">
+<h2 class="anchored" data-anchor-id="motivation-and-outline">Motivation and outline</h2>
+<p>Suppose that we have a data set <span class="math inline">\(x\)</span> with columns <span class="math inline">\(x_{\cdot, 1}, \ldots, x_{\cdot, K}\)</span> that make up covariates in a regression with quantities of interest <span class="math inline">\(\theta\)</span>. With the completely observed data, <span class="math inline">\(x^\text{comp}\)</span>, we could estimate the posterior distribution of <span class="math inline">\(\theta\)</span> as <span class="math inline">\(p(\theta \mid x^\text{comp})\)</span>. But with missing data, our matrix is split into <span class="math inline">\(x^{\text{obs}}\)</span> (the observed values of <span class="math inline">\(x\)</span>) and <span class="math inline">\(x^{\text{mis}}\)</span> (the missing values of <span class="math inline">\(x\)</span>).</p>
+<p>Fortunately, we can treat <span class="math inline">\(x^{\text{mis}}\)</span> as additional, nuisance parameters that we can estimate along with <span class="math inline">\(\theta\)</span> as <span class="math inline">\(p(\theta, x^{\text{mis}} \mid x^{\text{obs}})\)</span>. So, we can express the marginal distribution of <span class="math inline">\(\theta\)</span> given only the observed values <span class="math inline">\(x^{\text{obs}}\)</span> as <span class="math display">\[\begin{align*}
+p(\theta \mid x^{\text{obs}}) =&amp;
+\int
+    p(\theta, x^{\text{mis}} \mid x^{\text{obs}})
+    \mathrm{d}x^{\text{mis}} \\
+=&amp;
+\int
+    p(\theta \mid x^{\text{obs}}, x^{\text{mis}})
+    p(x^{\text{mis}} \mid x^{\text{obs}})
+    \mathrm{d}x^{\text{mis}} \\
+=&amp; \int
+    p(\theta \mid x^{\text{imp}})
+    p(x^{\text{mis}} \mid x^{\text{obs}})
+    \mathrm{d}x^{\text{mis}},
+\end{align*}\]</span> where <span class="math inline">\(x^{\text{imp}}\)</span> is a data set that includes imputed values of <span class="math inline">\(x^{\text{mis}}\)</span>.</p>
+<p>The equations above show that we do not need to describe <span class="math inline">\(p(\theta \mid x^{\text{obs}})\)</span> directly. Instead, we can first find a way to sample from <span class="math inline">\(x^{\text{mis}}\)</span> based on <span class="math inline">\(x^{\text{obs}}\)</span>, and then use these samples to fit the model <span class="math inline">\(p(\theta \mid x^\text{imp})\)</span>, treating <span class="math inline">\(x^\text{imp}\)</span> as if it was <span class="math inline">\(x^\text{comp}\)</span>.</p>
+<p>Note that the model for <span class="math inline">\(p(x^{\text{mis}} \mid x^{\text{obs}})\)</span> needs to be explicit. In a typical regression setting with covariates <span class="math inline">\(x\)</span> and outcome <span class="math inline">\(y\)</span>, this model for <span class="math inline">\(x\)</span> is not needed because inferences for the regression parameters are independent of the model of <span class="math inline">\(x\)</span> when <span class="math inline">\(x\)</span> is fully observed. But with missing data and multiple imputation, we need to generate values for <span class="math inline">\(x^{\text{mis}}\)</span> based on <span class="math inline">\(x^{\text{obs}}\)</span>, which in turn needs an explicit model for <span class="math inline">\(x^{\text{mis}}\)</span>.</p>
+<p>Thus, the general outline for multiple imputation with a number <span class="math inline">\(M\)</span> of imputations is:</p>
+<ol type="1">
+<li>Draw <span class="math inline">\(M\)</span> random samples <span class="math inline">\(x^{\text{mis}}_1, \ldots,
+x^{\text{mis}}_M\)</span> from the posterior predictive distribution <span class="math inline">\(p(x^{\text{mis}} \mid x^{\text{obs}})\)</span>.</li>
+<li>Use these samples to get <span class="math inline">\(M\)</span> different complete data sets <span class="math inline">\(x_1^\text{imp}, \ldots, x_M^\text{imp}\)</span>.</li>
+<li>For each of the <span class="math inline">\(M\)</span> imputed data sets, sample from the model of interest <span class="math inline">\(p(\theta \mid x_m^\text{imp})\)</span>, each time treating <span class="math inline">\(x_m^\text{imp}\)</span> as if it was <span class="math inline">\(x^\text{comp}\)</span>.</li>
+<li>Combine the draws for <span class="math inline">\(\theta\)</span> from all <span class="math inline">\(M\)</span> fits.</li>
+</ol>
+<section id="pseudocode-for-outline" class="level3">
+<h3 class="anchored" data-anchor-id="pseudocode-for-outline">Pseudocode for outline</h3>
+<p>Following the notation above, we can express the outline with the pseudocode below:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>theta_estimates = []</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  x_mis &lt;- get_imputations(x_obs)</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  x_imp &lt;- add_imputations(x_obs, x_mis)</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  imputation_fits &lt;- my_model.fit(x_imp).draws()</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  theta_estimates &lt;- theta_estimates.append(imputation_fits)</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Each of the functions in this pseudocode can encapsulate simple or complex procedures. <code>get_imputations()</code>, for example, may impute missing data by taking values directly from <code>x_obs</code>; or it may model a regression where an <span class="math inline">\(x_{\cdot, k}\)</span> with missing values is the outcome and the <span class="math inline">\(x_{\cdot, j}\)</span> without missing observations are predictors. Below we illustrate how to implement this second approach in Stan.</p>
+</section>
+</section>
+<section id="imputing-one-variable-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="imputing-one-variable-in-stan">Imputing one variable in Stan</h2>
+<p>Imagine our data set is composed of two numerical, continuous variables <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span>, and one binary variable <span class="math inline">\(z\)</span>. We want to estimate the parameters that govern the conditional association <span class="math inline">\(p(y \mid x, z)\)</span>, but several values of <span class="math inline">\(x\)</span> are missing. So, before we fit this model for <span class="math inline">\(y\)</span>, we impute the missing values in <span class="math inline">\(x\)</span>.</p>
+<p>We first find all the fully observed data points and identify their values as <span class="math inline">\(x^{\text{obs}}\)</span>, <span class="math inline">\(y^{\text{obs}}\)</span>, and <span class="math inline">\(z^{\text{obs}}\)</span>. Then we use these data to fit the model <span class="math display">\[
+x^{\text{obs}} \sim \text{normal}(\gamma_0 + \gamma_1 y^{\text{obs}} +
+    \gamma_2 z^{\text{obs}}, \lambda).
+\]</span></p>
+<p>This model can then give us samples for the missing values <span class="math inline">\(x^{\text{mis}}\)</span> conditional on the corresponding values of <span class="math inline">\(y\)</span> and <span class="math inline">\(z\)</span>, which we call <span class="math inline">\(y^{\text{aux}}\)</span> and <span class="math inline">\(z^{\text{aux}}\)</span>. The Stan code for this imputation is:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] x_obs;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] y_obs;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z_obs;  </span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_mis;   </span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_mis] y_aux;  </span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z_aux; </span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] gamma;</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  gamma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  lambda ~ exponential(<span class="dv">1</span>);</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>  x_obs ~ normal(gamma[<span class="dv">1</span>] + gamma[<span class="dv">2</span>] * y_obs + gamma[<span class="dv">3</span>] * z_obs,</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>          lambda);</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] x_imp = normal_rng(gamma[<span class="dv">1</span>] + gamma[<span class="dv">2</span>] * y_aux[n]</span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>                      + gamma[<span class="dv">3</span>] * z_aux[n], lambda);</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>generated quantities</code> block automatically samples <span class="math inline">\(\gamma_0\)</span>, <span class="math inline">\(\gamma_1\)</span>, <span class="math inline">\(\gamma_2\)</span>, and <span class="math inline">\(\lambda\)</span> from their posterior distributions. So, the random draws of <code>x_imp</code> incorporate uncertainty from the estimated parameters and from the sampling variation in <span class="math inline">\(x^{\text{mis}}\)</span>.</p>
+<p>Multiple posterior draws of <span class="math inline">\(x^{\text{mis}}\)</span> give us multiple imputed data sets. With these data sets we can model <span class="math inline">\(p(y \mid \beta, \sigma, x, z)\)</span> as <span class="math display">\[
+y \sim \text{normal}(\beta_0 + \beta_1 x^\text{comp} + \beta_2 z,
+        \sigma),
+\]</span> where <span class="math inline">\(x^\text{comp}\)</span> contains observed and imputed values. The Stan code for this model is:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] z;</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] beta;</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  sigma ~ exponential(<span class="dv">1</span>);</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  y ~ normal(beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x + beta[<span class="dv">3</span>] * z, sigma);</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With multiple imputation, the model for <span class="math inline">\(p(y \mid \beta, \sigma, x, z)\)</span> does not need to distinguish between observed and imputed values of <span class="math inline">\(x\)</span>. Instead, this model can treat all imputed data sets as complete because we can combine all the posterior draws from multiple fits.</p>
+</section>
+<section id="imputing-two-or-more-variables" class="level2">
+<h2 class="anchored" data-anchor-id="imputing-two-or-more-variables">Imputing two or more variables</h2>
+<p>A more general scenario involves an outcome variable <span class="math inline">\(y\)</span> and several explanatory variables <span class="math inline">\(x_1, \ldots, x_K\)</span>, each of which can have missing values that we want to impute.</p>
+<p>One solution is to do multiple imputation through chained equations, a procedure often called “MICE”<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a>. The MICE procedure in this scenario is:</p>
+<ol start="0" type="1">
+<li><strong>Initialize missing values in all <span class="math inline">\(x_i\)</span></strong>. For each variable <span class="math inline">\(x_i\)</span> with missing entries, fill its missing values with random samples from its observed values (or use another simple initialization rule).</li>
+<li><strong>Update each <span class="math inline">\(x_i\)</span> given <span class="math inline">\(y\)</span> and the other <span class="math inline">\(x\)</span>’s</strong>. For <span class="math inline">\(i=1, \ldots, K\)</span>, fit a model for the <em>observed</em> <span class="math inline">\(x_i\)</span> conditional on the current versions (with observed and imputed values) of <span class="math inline">\(y\)</span> and <span class="math inline">\(x_{-k} = x_1, \ldots x_{k-1}, x_{k+1}, \ldots,
+x_{K}\)</span>. Use this model to draw impuations from the predictive distribution of the <em>missing</em> <span class="math inline">\(x_i\)</span>. Completing this step for all <span class="math inline">\(i = 1, \ldots, K\)</span> constitutes a single imputation cycle.</li>
+<li><strong>Warmup period</strong>. Repeat the imputation cycle in step 1 several times as warmup to let the imputations stabilize.</li>
+<li><strong>Create M imputed datasets</strong>. After the warmup, record the current complete dataset as one imputed dataset. Then either <em>restart</em> from step 0 and repeat steps 1–3 until you have <span class="math inline">\(M\)</span> imputed datasets; or do a <em>single long run</em>, i.e., continue iterating steps 1–2 and save the imputed dataset at <span class="math inline">\(M\)</span> well-spaced iterations (e.g., every <span class="math inline">\(S\)</span> cycles) to obtain <span class="math inline">\(M\)</span> imputed datasets, without restarting from step 0 each time.</li>
+<li><strong>Fit the target model</strong> <span class="math inline">\(p(y \mid \beta, \sigma, x_1, \ldots,
+x_K)\)</span> separately to each of the <span class="math inline">\(M\)</span> imputed datasets and save the posterior draws for the parameters of interest.</li>
+<li><strong>Combine the draws</strong> (or other summaries) from all <span class="math inline">\(M\)</span> fits.</li>
+</ol>
+<p>Note that, as described here, the MICE procedure does not guarantee that the conditional distributions of all variables will be compatible. <em>Compatibility</em> means that there is a joint distribution for all the variables used in the imputation that can be decomposed as the conditional distributions we used. An incompatible imputation model will technically not sample from any well-defined probability distribution. But the consequences of this are not always serious. See section 6.4 in <span class="citation" data-cites="carpenter-etal:2023">Carpenter et al. (<a href="#ref-carpenter-etal:2023" role="doc-biblioref">2023</a>)</span> for a more detailed explanation of compatibility and of the related concept of <em>congeniality</em>.</p>
+<section id="imputing-two-variables-in-stan" class="level3">
+<h3 class="anchored" data-anchor-id="imputing-two-variables-in-stan">Imputing two variables in Stan</h3>
+<p>Imagine that we again want to use numerical variables <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span>, and dichotomic variable <span class="math inline">\(z\)</span> to estimate the parameters in <span class="math inline">\(p(y \mid \beta, \sigma, x, z)\)</span>. But now both <span class="math inline">\(x\)</span> and <span class="math inline">\(z\)</span> have missing values that we want to impute.</p>
+<p>To apply the MICE procedure in this example, we can reuse the models for <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span> that we defined above. We also need a new model to impute <span class="math inline">\(z^{\text{mis}}\)</span>, so we use the logistic regression <span class="math display">\[
+z^{\text{obs}} \sim \text{Bernoulli}(
+  \text{logit}^{-1}(
+    \alpha_0 + \alpha_1 y^{\text{obs}} + \alpha_2 x^{\text{obs}})
+  ).
+\]</span> Here, <span class="math inline">\(z^{\text{obs}}\)</span> contains only the completely observed values from our original data set, while <span class="math inline">\(x^{\text{obs}}\)</span> and <span class="math inline">\(y^{\text{obs}}\)</span> contain all the values that correspond to <span class="math inline">\(z^{\text{obs}}\)</span>. Thus, <span class="math inline">\(x^{\text{obs}}\)</span> can include observed and imputed values, and <span class="math inline">\(y^{\text{obs}}\)</span> need not be the same as in the model for <span class="math inline">\(x^{\text{obs}}\)</span>.</p>
+<p>With this model we can sample values for <span class="math inline">\(z^{\text{mis}}\)</span> conditional on its corresponding values <span class="math inline">\(y^{\text{aux}}\)</span> and <span class="math inline">\(x^{\text{aux}}\)</span>.</p>
+<p>The Stan code to impute <span class="math inline">\(z^{\text{mis}}\)</span> is:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z_obs;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] y_obs;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] x_obs;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_mis;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_mis] y_aux;</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_mis] x_aux;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] gamma;</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>  gamma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>  z_obs ~ bernoulli_logit(gamma[<span class="dv">1</span>] + gamma[<span class="dv">2</span>] * y_obs </span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>                          + gamma[<span class="dv">3</span>] * x_obs);</span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">int</span> z_imp = bernoulli_logit_rng(gamma[<span class="dv">1</span>]</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>                      + gamma[<span class="dv">2</span>] * y_aux</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>                      + gamma[<span class="dv">3</span>] * x_aux);</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The pseudocode for this example of MICE, restarting the imputations after each imputation cycle, is shown below. Note that we do not need to initialize <span class="math inline">\(x^{\text{mis}}\)</span> in step 0 because we can impute it directly with the model for <span class="math inline">\(x^{\text{obs}}\)</span>.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a>completed_datasets &lt;- []</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  data_m &lt;- copy(data_orig)</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  data_m.z[missing_idz] &lt;- random_sample(data_m.z[observed_idz])</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (cycle <span class="cf">in</span> <span class="dv">1</span>:n_warm) {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>    stanmod_x &lt;- build_model(<span class="st">"model_for_x.stan"</span>)</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>    fit_x &lt;- stanmod_x.fit(</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>      obs_data=data_m[observed_idx],</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>      mis_data=data_m[missing_idx]</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>    )</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>    data_m.x[missing_idx] &lt;- get_imputations(fit_x, <span class="st">"x_imp"</span>)</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>    stanmod_z &lt;- build_model(<span class="st">"model_for_z.stan"</span>)</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>    fit_z &lt;- stanmod_z.fit(</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>      obs_data=data_m[observed_idz],</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>      mis_data=data_m[missing_idz]</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>    )</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>    data_m.z[missing_idz] &lt;- get_imputations(fit_z, <span class="st">"z_imp"</span>)</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>  completed_datasets.append(data_m)</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>all_draws &lt;- []</span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (dataset <span class="cf">in</span> completed_datasets) {</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>  stanmod_y &lt;- build_model(<span class="st">"model_for_y.stan"</span>)</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>  fit_y &lt;- stanmod_y.fit(dataset)</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a>  all_draws.append(extract_draws(model_results))</span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-27"><a href="#cb5-27" aria-hidden="true" tabindex="-1"></a>all_results &lt;- combine_results(all_draws)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="combining-posterior-draws" class="level2">
+<h2 class="anchored" data-anchor-id="combining-posterior-draws">Combining posterior draws</h2>
+<p>With Stan’s MCMC sampler, we can treat posterior draw chains from imputed data sets as if they were chains based on complete data. There is one important difference. Multiple imputation expresses uncertainty in the missing values as consistent differences in the estimates obtained from different imputated data sets. This means that chains obtained from the same imputed data set should converge, but chains obtained from different data sets do not have to. So, we need not worry if diagnostics<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> signal that the chains from different imputed data sets are not converging properly. See an example in <span class="citation" data-cites="burkner:2025">Bürkner (<a href="#ref-burkner:2025" role="doc-biblioref">2025</a>)</span>.</p>
+</section>
+<section id="cut-models" class="level2">
+<h2 class="anchored" data-anchor-id="cut-models">Cut models</h2>
+<p>A full Bayesian probability model includes a feedback flow of information between all parameters and all data. Cut models separate some parts of this feedback flow so that different subsets of data influence only some parameters in the model (see <span class="citation" data-cites="plummer:2015">Plummer (<a href="#ref-plummer:2015" role="doc-biblioref">2015</a>)</span>).</p>
+<p>Multiple imputation interrupts the flow of information from data to parameters. In our regression above, for example, the imputations influence the distribution of the parameters in <span class="math inline">\(p(y \mid x_1, \ldots, x_K)\)</span>, but these parameters do not influence the imputations. So, we could use multiple imputation to implement a cut model.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-burkner:2025" class="csl-entry" role="listitem">
+Bürkner, Paul. 2025. <span>“Handle Missing Values with Brms.”</span> <a href="https://cran.r-project.org/web/packages/brms/vignettes/brms_missings.html">https://cran.r-project.org/web/packages/brms/vignettes/brms_missings.html</a>.
+</div>
+<div id="ref-carpenter-etal:2023" class="csl-entry" role="listitem">
+Carpenter, James R., Jonathan W. Bartlett, Tim P. Morris, Angela M. Wood, Matteo Quartagno, and Michael G. Kenward. 2023. <em>Multiple Imputation and Its Application</em>. 2nd ed. John Wiley &amp; Sons, Ltd. <a href="https://doi.org/10.1002/9781119756118">https://doi.org/10.1002/9781119756118</a>.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-plummer:2015" class="csl-entry" role="listitem">
+Plummer, Martyn. 2015. <span>“Cuts in <span>B</span>ayesian Graphical Models.”</span> <em>Statistics and Computing</em> 25: 37–43. <a href="https://doi.org/10.1007/s11222-014-9503-z">https://doi.org/10.1007/s11222-014-9503-z</a>.
+</div>
+<div id="ref-vanbuuren:2018" class="csl-entry" role="listitem">
+van Buuren, Stef. 2018. <em>Flexible Imputation of Missing Data</em>. 2nd ed. Chapman; Hall/CRC. <a href="https://doi.org/10.1201/9780429492259">https://doi.org/10.1201/9780429492259</a>.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Section 5 of chapter 4 in <span class="citation" data-cites="vanbuuren:2018">van Buuren (<a href="#ref-vanbuuren:2018" role="doc-biblioref">2018</a>)</span> details the MICE procedure in a frequentist context.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Such as <span class="math inline">\(\hat{R}\)</span>. See <a href="https://mc-stan.org/docs/reference-manual/analysis.html#split-r-hat-for-detecting-non-stationarity"><em>Split R-hat for detecting non-stationarity</em></a> in the <em>Reference Manual</em>.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/bootstrap.html" class="pagination-link" aria-label="The Bootstrap and Bagging">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">The Bootstrap and Bagging</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/using-stanc.html" class="pagination-link" aria-label="Using the Stan Compiler">
+        <span class="nav-page-text">Using the Stan Compiler</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multiple-imputation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/odes.html b/docs/2_39/stan-users-guide/odes.html
new file mode 100644
index 000000000..164f13882
--- /dev/null
+++ b/docs/2_39/stan-users-guide/odes.html
@@ -0,0 +1,1543 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Ordinary Differential Equations</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/one-dimensional-integrals.html" rel="next">
+<link href="../stan-users-guide/algebraic-equations.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Ordinary Differential Equations">
+<meta property="og:description" content="">
+<meta property="og:image" content="./img/sho-ode-trajectory.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/odes.html">Ordinary Differential Equations</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#ode-solver.chapter" id="toc-ode-solver.chapter" class="nav-link active" data-scroll-target="#ode-solver.chapter">Ordinary Differential Equations</a>
+  <ul class="collapse">
+  <li><a href="#notation" id="toc-notation" class="nav-link" data-scroll-target="#notation">Notation</a></li>
+  <li><a href="#example-simple-harmonic-oscillator" id="toc-example-simple-harmonic-oscillator" class="nav-link" data-scroll-target="#example-simple-harmonic-oscillator">Example: simple harmonic oscillator</a></li>
+  <li><a href="#coding-the-ode-system-function" id="toc-coding-the-ode-system-function" class="nav-link" data-scroll-target="#coding-the-ode-system-function">Coding the ODE system function</a>
+  <ul class="collapse">
+  <li><a href="#strict-signature" id="toc-strict-signature" class="nav-link" data-scroll-target="#strict-signature">Strict signature</a></li>
+  </ul></li>
+  <li><a href="#measurement-error-models" id="toc-measurement-error-models" class="nav-link" data-scroll-target="#measurement-error-models">Measurement error models</a>
+  <ul class="collapse">
+  <li><a href="#simulating-noisy-measurements" id="toc-simulating-noisy-measurements" class="nav-link" data-scroll-target="#simulating-noisy-measurements">Simulating noisy measurements</a></li>
+  <li><a href="#estimating-system-parameters-and-initial-state" id="toc-estimating-system-parameters-and-initial-state" class="nav-link" data-scroll-target="#estimating-system-parameters-and-initial-state">Estimating system parameters and initial state</a></li>
+  </ul></li>
+  <li><a href="#stiff-ode.section" id="toc-stiff-ode.section" class="nav-link" data-scroll-target="#stiff-ode.section">Stiff ODEs</a></li>
+  <li><a href="#control-ode.section" id="toc-control-ode.section" class="nav-link" data-scroll-target="#control-ode.section">Control parameters for ODE solving</a>
+  <ul class="collapse">
+  <li><a href="#discontinuous-ode-system-function" id="toc-discontinuous-ode-system-function" class="nav-link" data-scroll-target="#discontinuous-ode-system-function">Discontinuous ODE system function</a></li>
+  <li><a href="#tolerance" id="toc-tolerance" class="nav-link" data-scroll-target="#tolerance">Tolerance</a></li>
+  <li><a href="#maximum-number-of-steps" id="toc-maximum-number-of-steps" class="nav-link" data-scroll-target="#maximum-number-of-steps">Maximum number of steps</a></li>
+  </ul></li>
+  <li><a href="#adjoint-ode.section" id="toc-adjoint-ode.section" class="nav-link" data-scroll-target="#adjoint-ode.section">Adjoint ODE solver</a></li>
+  <li><a href="#solving-a-system-of-linear-odes-using-a-matrix-exponential" id="toc-solving-a-system-of-linear-odes-using-a-matrix-exponential" class="nav-link" data-scroll-target="#solving-a-system-of-linear-odes-using-a-matrix-exponential">Solving a system of linear ODEs using a matrix exponential</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/odes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/odes.html">Ordinary Differential Equations</a></li></ol></nav></header>
+
+
+
+
+<section id="ode-solver.chapter" class="level1">
+<h1>Ordinary Differential Equations</h1>
+<p>Stan provides a number of different methods for solving systems of ordinary differential equations (ODEs). All of these methods adaptively refine their solutions in order to satisfy given tolerances, but internally they handle calculations quite a bit differently.</p>
+<p>Because Stan’s algorithms requires gradients of the log density, the ODE solvers must not only provide the solution to the ODE itself, but also the gradient of the ODE solution with respect to parameters (the sensitivities). Two fundamentally different approaches are available in Stan to solve this problem, each having very different computational cost depending on the number of ODE states <span class="math inline">\(N\)</span> and the number of parameters <span class="math inline">\(M\)</span> being used:</p>
+<ul>
+<li><p>A <em>forward sensitivity</em> solver expands the base ODE system with additional ODE equations for the gradients of the solution. For each parameter, an additional full set of <span class="math inline">\(N\)</span> sensitivity states are added meaning that the full ODE solved has <span class="math inline">\(N \, + N \cdot M\)</span> states.</p></li>
+<li><p>An <em>adjoint sensitivity</em> solver starts by solving the base ODE system forward in time to get the ODE solution and then solves another ODE system (the adjoint) backward in time to get the gradients. The forward and reverse solves both have <span class="math inline">\(N\)</span> states each. There is additionally one quadrature problem solved for every parameter.</p></li>
+</ul>
+<p>The adjoint sensitivity approach scales much better than the forward sensitivity approach. Whereas the computational cost of the forward approach scales multiplicatively in the number of ODE states <span class="math inline">\(N\)</span> and parameters <span class="math inline">\(M\)</span>, the adjoint sensitivity approach scales linear in states <span class="math inline">\(N\)</span> and parameters <span class="math inline">\(M\)</span>. However, the adjoint problem is harder to configure and the overhead for small problems actually makes it slower than solving the full forward sensitivity system. With that in mind, the rest of this introduction focuses on the forward sensitivity interfaces. For information on the adjoint sensitivity interface see the <a href="#adjoint-ode.section">Adjoint ODE solver</a></p>
+<p>Two interfaces are provided for each forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. For instance, if a solution component slowly varies between 3.0 and 5.0 and measurements of the ODE state are noisy, then perhaps the tolerances do not need to be as tight as for a situation where the solutions vary between 3.0 and 3.1 and very high precision measurements of the ODE state are available. It is also often useful to reduce the absolute tolerance when a component of the solution is expected to approach zero. For information on choosing tolerances, see the <a href="#control-ode.section">control parameters section</a>.</p>
+<p>The advantage of adaptive solvers is that as long as reasonable tolerances are provided and an ODE solver well-suited to the problem is chosen the technical details of solving the ODE can be abstracted away. The catch is that it is not always clear from the outset what reasonable tolerances are or which ODE solver is best suited to a problem. In addition, as changes are made to an ODE model, the optimal solver and tolerances may change.</p>
+<p>With this in mind, the four forward solvers are <code>rk45</code>, <code>bdf</code>, <code>adams</code>, and <code>ckrk</code>. If no other information about the ODE is available, start with the <code>rk45</code> solver. The list below has information on when each solver is useful.</p>
+<p>If there is any uncertainty about which solver is the best, it can be useful to measure the performance of all the interesting solvers using <code>profile</code> statements. It is difficult to always know exactly what solver is the best in all situations, but a <code>profile</code> can provide a quick check.</p>
+<ul>
+<li><p><code>rk45</code>: a fourth and fifth order Runge-Kutta method for non-stiff systems <span class="citation" data-cites="DormandPrince:1980 AhnertMulansky:2011">(<a href="#ref-DormandPrince:1980" role="doc-biblioref">Dormand and Prince 1980</a>; <a href="#ref-AhnertMulansky:2011" role="doc-biblioref">Ahnert and Mulansky 2011</a>)</span>. <code>rk45</code> is the most generic solver and should be tried first.</p></li>
+<li><p><code>bdf</code>: a variable-step, variable-order, backward-differentiation formula implementation for stiff systems <span class="citation" data-cites="CohenHindmarsh:1996 SerbanHindmarsh:2005">(<a href="#ref-CohenHindmarsh:1996" role="doc-biblioref">Cohen and Hindmarsh 1996</a>; <a href="#ref-SerbanHindmarsh:2005" role="doc-biblioref">Serban and Hindmarsh 2005</a>)</span>. <code>bdf</code> is often useful for ODEs modeling chemical reactions.</p></li>
+<li><p><code>adams</code>: a variable-step, variable-order, Adams-Moulton formula implementation for non-stiff systems <span class="citation" data-cites="CohenHindmarsh:1996 SerbanHindmarsh:2005">(<a href="#ref-CohenHindmarsh:1996" role="doc-biblioref">Cohen and Hindmarsh 1996</a>; <a href="#ref-SerbanHindmarsh:2005" role="doc-biblioref">Serban and Hindmarsh 2005</a>)</span>. The method has order up to 12, hence is commonly used when high-accuracy is desired for a very smooth solution, such as in modeling celestial mechanics and orbital dynamics <span class="citation" data-cites="montenbrucksatellite:2000">(<a href="#ref-montenbrucksatellite:2000" role="doc-biblioref">Montenbruck and Gill 2000</a>)</span>.</p></li>
+<li><p><code>ckrk</code>: a fourth and fifth order explicit Runge-Kutta method for non-stiff and semi-stiff systems <span class="citation" data-cites="cashvariable:1990 mazziatest:2012">(<a href="#ref-cashvariable:1990" role="doc-biblioref">Cash and Karp 1990</a>; <a href="#ref-mazziatest:2012" role="doc-biblioref">Mazzia, Cash, and Soetaert 2012</a>)</span>. The difference between <code>ckrk</code> and <code>rk45</code> is that <code>ckrk</code> should perform better for systems that exhibit rapidly varying solutions. Often in those situations the derivatives become large or even nearly discontinuous, and <code>ckrk</code> is designed to address such problems.</p></li>
+</ul>
+<p>For a discussion of stiff ODE systems, see the <a href="#stiff-ode.section">stiff ODE section</a>. For information on the adjoint sensitivity interface see the <a href="#adjoint-ode.section">Adjoint ODE solver section</a>. The function signatures for Stan’s ODE solvers can be found in the function reference manual section on ODE solvers.</p>
+<section id="notation" class="level2">
+<h2 class="anchored" data-anchor-id="notation">Notation</h2>
+<p>An ODE is defined by a set of differential equations, <span class="math inline">\(y(t, \theta)' = f(t, y, \theta)\)</span>, and initial conditions, <span class="math inline">\(y(t_0, \theta) = y_0\)</span>. The function <span class="math inline">\(f(t, y, \theta)\)</span> is called the system function. The <span class="math inline">\(\theta\)</span> dependence is included in the notation for <span class="math inline">\(y(t, \theta)\)</span> and <span class="math inline">\(f(t, y, \theta)\)</span> as a reminder that the solution is a function of any parameters used in the computation.</p>
+</section>
+<section id="example-simple-harmonic-oscillator" class="level2">
+<h2 class="anchored" data-anchor-id="example-simple-harmonic-oscillator">Example: simple harmonic oscillator</h2>
+<p>As an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair <span class="math inline">\(y = (y_1, y_2)\)</span> representing position and speed. The change in the system with respect to time is given by the following differential equations.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p><span class="math display">\[\begin{align*}
+&amp;\frac{d}{dt} y_1 = y_2  \\
+&amp;\frac{d}{dt} y_2 = -y_1 - \theta y_2
+\end{align*}\]</span></p>
+<p>The state equations implicitly defines the state at future times as a function of an initial state and the system parameters.</p>
+</section>
+<section id="coding-the-ode-system-function" class="level2">
+<h2 class="anchored" data-anchor-id="coding-the-ode-system-function">Coding the ODE system function</h2>
+<p>The first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.</p>
+<p>The first argument to the system function is time, passed as a <code>real</code>; the second argument to the system function is the system state, passed as a <code>vector</code>, and the return value from the system function are the current time derivatives of the state defined as a <code>vector</code>. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.</p>
+<p>The simple harmonic oscillator can be coded using the following function in Stan (see the <a href="../stan-users-guide/user-functions.html">user-defined functions chapter</a> for more information on coding user-defined functions).</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> sho(<span class="dt">real</span> t,        <span class="co">// time</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>           <span class="dt">vector</span> y,      <span class="co">// state</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>           <span class="dt">real</span> theta) {  <span class="co">// friction parameter</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] dydt;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  dydt[<span class="dv">1</span>] = y[<span class="dv">2</span>];</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  dydt[<span class="dv">2</span>] = -y[<span class="dv">1</span>] - theta * y[<span class="dv">2</span>];</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> dydt;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function takes in a time <code>t</code> (a <code>real</code>), the system state <code>y</code> (a <code>vector</code>), and the parameter <code>theta</code> (a <code>real</code>). The function returns a <code>vector</code> of time derivatives of the system state at time <code>t</code>, state <code>y</code>, and parameter <code>theta</code>. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, <code>t</code> does not show up in the definition of <code>dydt</code>, however it is still required.</p>
+<section id="strict-signature" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="strict-signature">Strict signature</h3>
+<p>The types in the ODE system function are strict. The first argument is the time passed as a <code>real</code>, the second argument is the state passed as a <code>vector</code>, and the return type is a <code>vector</code>. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.</p>
+<p>All of these are possible ODE signatures:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> myode1(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">real</span> a0);</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> myode2(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">array</span>[] <span class="dt">int</span> a0, <span class="dt">vector</span> a1);</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> myode3(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">matrix</span> a0, <span class="dt">array</span>[] <span class="dt">real</span> a1, <span class="dt">row_vector</span> a2);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>but these are not allowed:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> myode1(<span class="dt">real</span> t, <span class="dt">array</span>[] <span class="dt">real</span> y, <span class="dt">real</span> a0);</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="co">// Second argument is not a vector</span></span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[] <span class="dt">real</span> myode2(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">real</span> a0);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a><span class="co">// Return type is not a vector</span></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> myode3(<span class="dt">vector</span> y, <span class="dt">real</span> a0);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a><span class="co">// First argument is not a real and second is not a vector</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="measurement-error-models" class="level2">
+<h2 class="anchored" data-anchor-id="measurement-error-models">Measurement error models</h2>
+<p>Noisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.</p>
+<section id="simulating-noisy-measurements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="simulating-noisy-measurements">Simulating noisy measurements</h3>
+<p>As an example, suppose the simple harmonic oscillator has a parameter value of <span class="math inline">\(\theta = 0.15\)</span> and an initial state <span class="math inline">\(y(t = 0, \theta = 0.15) = (1, 0)\)</span>. Assume the system is measured at 10 time points, <span class="math inline">\(t = 1, 2, \cdots, 10\)</span>, where each measurement of <span class="math inline">\(y(t, \theta)\)</span> has independent <span class="math inline">\(\textsf{normal}(0, 0.1)\)</span> error in both dimensions (<span class="math inline">\(y_1(t, \theta)\)</span> and <span class="math inline">\(y_2(t, \theta)\)</span>).</p>
+<p>The following model can be used to generate data like this:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> sho(<span class="dt">real</span> t,</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>             <span class="dt">vector</span> y,</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>             <span class="dt">real</span> theta) {</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[<span class="dv">2</span>] dydt;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>    dydt[<span class="dv">1</span>] = y[<span class="dv">2</span>];</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>    dydt[<span class="dv">2</span>] = -y[<span class="dv">1</span>] - theta * y[<span class="dv">2</span>];</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> dydt;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] y0;</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> t0;</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span> ts;</span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;</span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] y_sim = ode_rk45(sho, y0, t0, ts, theta);</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>  <span class="co">// add measurement error</span></span>
+<span id="cb4-23"><a href="#cb4-23" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb4-24"><a href="#cb4-24" aria-hidden="true" tabindex="-1"></a>    y_sim[t, <span class="dv">1</span>] += normal_rng(<span class="dv">0</span>, <span class="fl">0.1</span>);</span>
+<span id="cb4-25"><a href="#cb4-25" aria-hidden="true" tabindex="-1"></a>    y_sim[t, <span class="dv">2</span>] += normal_rng(<span class="dv">0</span>, <span class="fl">0.1</span>);</span>
+<span id="cb4-26"><a href="#cb4-26" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-27"><a href="#cb4-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The system parameters <code>theta</code> and initial state <code>y0</code> are read in as data along with the initial time <code>t0</code> and observation times <code>ts</code>. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations <code>y_sim</code>. Because the system is not stiff, the <code>ode_rk45</code> solver is used.</p>
+<p>This program illustrates the way in which the ODE solver is called in a Stan program,</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] y_sim = ode_rk45(sho, y0, t0, ts, theta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>this returns the solution of the ODE initial value problem defined by system function <code>sho</code>, initial state <code>y0</code>, initial time <code>t0</code>, and parameter <code>theta</code> at the times <code>ts</code>. The call explicitly specifies the non-stiff RK45 solver.</p>
+<p>The parameter <code>theta</code> is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\beta\)</span>, the system function definition would look like:</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> sho(<span class="dt">real</span> t, <span class="dt">vector</span> y, <span class="dt">real</span> theta, <span class="dt">real</span> beta) { ... }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and the appropriate ODE solver call would be:</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a>ode_rk45(sho, y0, t0, ts, theta, beta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Any number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).</p>
+<p>Because all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the <code>T</code> outputs with <code>normal_rng</code>.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><img src="./img/sho-ode-trajectory.png" class="img-fluid figure-img" style="width:50.0%"></p>
+<figcaption>Typical realization of harmonic oscillator trajectory.</figcaption>
+</figure>
+</div>
+</section>
+<section id="estimating-system-parameters-and-initial-state" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimating-system-parameters-and-initial-state">Estimating system parameters and initial state</h3>
+<p>These ten noisy observations of the state can be used to estimate the friction parameter, <span class="math inline">\(\theta\)</span>, the initial conditions, <span class="math inline">\(y(t_0, \theta)\)</span>, and the scale of the noise in the problem. The full Stan model is:</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> sho(<span class="dt">real</span> t,</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>             <span class="dt">vector</span> y,</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>             <span class="dt">real</span> theta) {</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[<span class="dv">2</span>] dydt;</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>    dydt[<span class="dv">1</span>] = y[<span class="dv">2</span>];</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>    dydt[<span class="dv">2</span>] = -y[<span class="dv">1</span>] - theta * y[<span class="dv">2</span>];</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> dydt;</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] y;</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> t0;</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span> ts;</span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] y0;</span>
+<span id="cb8-19"><a href="#cb8-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[<span class="dv">2</span>] sigma;</span>
+<span id="cb8-20"><a href="#cb8-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;</span>
+<span id="cb8-21"><a href="#cb8-21" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-22"><a href="#cb8-22" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-23"><a href="#cb8-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] mu = ode_rk45(sho, y0, t0, ts, theta);</span>
+<span id="cb8-24"><a href="#cb8-24" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb8-25"><a href="#cb8-25" aria-hidden="true" tabindex="-1"></a>  theta ~ std_normal();</span>
+<span id="cb8-26"><a href="#cb8-26" aria-hidden="true" tabindex="-1"></a>  y0 ~ std_normal();</span>
+<span id="cb8-27"><a href="#cb8-27" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb8-28"><a href="#cb8-28" aria-hidden="true" tabindex="-1"></a>    y[t] ~ normal(mu[t], sigma);</span>
+<span id="cb8-29"><a href="#cb8-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-30"><a href="#cb8-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because the solves are now a function of model parameters, the <code>ode_rk45</code> call is now made in the model block. There are half-normal priors on the measurement error scales <code>sigma</code>, and standard normal priors on <code>theta</code> and the initial state vector <code>y0</code>. The solutions to the ODE are assigned to <code>mu</code>, which is used as the location for the normal observation model.</p>
+<p>As with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).</p>
+</section>
+</section>
+<section id="stiff-ode.section" class="level2">
+<h2 class="anchored" data-anchor-id="stiff-ode.section">Stiff ODEs</h2>
+<p>Stiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.</p>
+<p>Stan provides a specialized solver for stiff ODEs <span class="citation" data-cites="CohenHindmarsh:1996 SerbanHindmarsh:2005">(<a href="#ref-CohenHindmarsh:1996" role="doc-biblioref">Cohen and Hindmarsh 1996</a>; <a href="#ref-SerbanHindmarsh:2005" role="doc-biblioref">Serban and Hindmarsh 2005</a>)</span>. An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the <code>rk45</code> suffix is replaced with <code>bdf</code>, as in</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a>ode_bdf(sho, y0, t0, ts, theta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Using the stiff (<code>bdf</code>) solver on a system that is not stiff may be much slower than using the non-stiff (<code>rk45</code>) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.</p>
+<p>If it is not known for sure that an ODE system is stiff, run the model with both the <code>rk45</code> and <code>bdf</code> solvers and see which is faster. If the <code>rk45</code> solver is faster, then the problem is probably non-stiff, and then it makes sense to try the <code>adams</code> solver as well. The <code>adams</code> solver uses higher order methods which can take larger timesteps than the <code>rk45</code> solver, though similar to the <code>bdf</code> solver each of these steps is more expensive to compute.</p>
+</section>
+<section id="control-ode.section" class="level2">
+<h2 class="anchored" data-anchor-id="control-ode.section">Control parameters for ODE solving</h2>
+<p>For additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the <code>relative_tolerance</code>, <code>absolute_tolerance</code>, and <code>max_num_steps</code> parameters. These are the same as the regular function names but with <code>_tol</code> appended to the end. All three control arguments must be supplied with this signature (there are no defaults).</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] y_sim = ode_bdf_tol(sho, y0, t0, ts,</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>                                 relative_tolerance,</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>                                 absolute_tolerance,</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>                                 max_num_steps,</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>                                 theta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><code>relative_tolerance</code> and <code>absolute_tolerance</code> control accuracy the solver tries to achieve, and <code>max_num_steps</code> specifies the maximum number of steps the solver will take between output time points before throwing an error.</p>
+<p>The control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the <code>data</code> qualifier.</p>
+<p>For the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both <span class="math inline">\(10^{-6}\)</span> and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are <span class="math inline">\(10^{-10}\)</span> and the maximum number of steps between outputs is one hundred million.</p>
+<section id="discontinuous-ode-system-function" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="discontinuous-ode-system-function">Discontinuous ODE system function</h3>
+<p>If there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.</p>
+<p>Nonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times <span class="math inline">\(0 &lt; t &lt; t'\)</span> and then positive for <span class="math inline">\(t \geq t'\)</span>. In this example example, we would use code in the system such as</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (t &lt; t_lag) {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> [<span class="dv">0</span>, <span class="dv">0</span>]';</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... return non-zero vector...</span></span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In general it is better to integrate up to <code>t_lag</code> in one solve and then integrate from <code>t_lag</code> onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.</p>
+<p>If the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.</p>
+</section>
+<section id="tolerance" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="tolerance">Tolerance</h3>
+<p>The relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns <span class="math inline">\(y=(y_1,\dots,y_n)^T\)</span>, at every step the solver controls estimated local error <span class="math inline">\(e=(e_1,\dots,e_n)^T\)</span> through its weighted root-mean-square norm (<span class="citation" data-cites="SerbanHindmarsh:2005">Serban and Hindmarsh (<a href="#ref-SerbanHindmarsh:2005" role="doc-biblioref">2005</a>)</span>, <span class="citation" data-cites="hairer:1993">Hairer, Nørsett, and Wanner (<a href="#ref-hairer:1993" role="doc-biblioref">1993</a>)</span>)</p>
+<p><span class="math display">\[\begin{equation*}
+\sqrt{\sum_{i=1}^n{\frac{1}{n}\frac{e_i^2}{(\text{RTOL}\times y_i + \text{ATOL})^2}}} &lt; 1
+\end{equation*}\]</span> by reducing the stepsize when the inequality is not satisfied.</p>
+<p>To understand the roles of the two tolerances it helps to assume <span class="math inline">\(y\)</span> at opposite scales in the above expression: on one hand the absolute tolerance has little effect when <span class="math inline">\(y_i \gg 1\)</span>, on the other the relative tolerance can not affect the norm when <span class="math inline">\(y_i = 0\)</span>. Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow <span class="citation" data-cites="Brenan:1996">Brenan, Campbell, and Petzold (<a href="#ref-Brenan:1996" role="doc-biblioref">1995</a>)</span> for a rule of thumb: let <span class="math inline">\(m\)</span> be the number of significant digits required for <span class="math inline">\(y\)</span>, set <span class="math inline">\(\text{RTOL}=10^{-(m+1)}\)</span>, and set ATOL at which <span class="math inline">\(y\)</span> becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in <code>bdf</code> and <code>adams</code> solvers, and the same tolerances are used to control forward sensitivity calculation. See <span class="citation" data-cites="SerbanHindmarsh:2005">Serban and Hindmarsh (<a href="#ref-SerbanHindmarsh:2005" role="doc-biblioref">2005</a>)</span> for details.</p>
+</section>
+<section id="maximum-number-of-steps" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="maximum-number-of-steps">Maximum number of steps</h3>
+<p>The maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.</p>
+</section>
+</section>
+<section id="adjoint-ode.section" class="level2">
+<h2 class="anchored" data-anchor-id="adjoint-ode.section">Adjoint ODE solver</h2>
+<p>The adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with <span class="math inline">\(N\)</span> additional states for each parameter for which gradients are needed. If there are <span class="math inline">\(M\)</span> parameters for which sensitivities are required, then the augmented ODE system has a total of <span class="math inline">\(N \cdot (M +
+1)\)</span> states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.</p>
+<p>In contrast, the adjoint ODE solver integrates forward in time a system of <span class="math inline">\(N\)</span> equations to compute the ODE solution and then integrates backwards in time another system of <span class="math inline">\(N\)</span> equations to get the sensitivities. Additionally, for <span class="math inline">\(M\)</span> parameters there are <span class="math inline">\(M\)</span> additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the <code>bdf</code> and <code>adams</code> forward sensitivity interfaces).</p>
+<p>The solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.</p>
+<p>The solver type (either <code>bdf</code> or <code>adams</code>) can be individually set for both the forward and backward solves.</p>
+<p>The tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] y_sim</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>    = ode_adjoint_tol_ctl(sho, y0, t0, ts,</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>                          relative_tolerance/<span class="fl">9.0</span>,                <span class="co">// forward tolerance</span></span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>                          rep_vector(absolute_tolerance/<span class="fl">9.0</span>, <span class="dv">2</span>), <span class="co">// forward tolerance</span></span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>                          relative_tolerance/<span class="fl">3.0</span>,                <span class="co">// backward tolerance</span></span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>                          rep_vector(absolute_tolerance/<span class="fl">3.0</span>, <span class="dv">2</span>), <span class="co">// backward tolerance</span></span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>                          relative_tolerance,                    <span class="co">// quadrature tolerance</span></span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>                          absolute_tolerance,                    <span class="co">// quadrature tolerance</span></span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>                          max_num_steps,</span>
+<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a>                          <span class="dv">150</span>,                                   <span class="co">// number of steps between checkpoints</span></span>
+<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a>                          <span class="dv">1</span>,                                     <span class="co">// interpolation polynomial: 1=Hermite, 2=polynomial</span></span>
+<span id="cb12-12"><a href="#cb12-12" aria-hidden="true" tabindex="-1"></a>                          <span class="dv">2</span>,                                     <span class="co">// solver for forward phase: 1=Adams, 2=BDF</span></span>
+<span id="cb12-13"><a href="#cb12-13" aria-hidden="true" tabindex="-1"></a>                          <span class="dv">2</span>,                                     <span class="co">// solver for backward phase: 1=Adams, 2=BDF</span></span>
+<span id="cb12-14"><a href="#cb12-14" aria-hidden="true" tabindex="-1"></a>                          theta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For a detailed information on each argument please see the Stan function reference manual.</p>
+</section>
+<section id="solving-a-system-of-linear-odes-using-a-matrix-exponential" class="level2">
+<h2 class="anchored" data-anchor-id="solving-a-system-of-linear-odes-using-a-matrix-exponential">Solving a system of linear ODEs using a matrix exponential</h2>
+<p>Linear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.</p>
+<p>The solution to <span class="math inline">\(\frac{d}{dt} y = ay\)</span> is <span class="math inline">\(y = y_0e^{at}\)</span>, where the constant <span class="math inline">\(y_0\)</span> is determined by boundary conditions. We can extend this solution to the vector case: <span class="math display">\[
+\frac{d}{dt}y = A \, y
+\]</span> <a name="id:ode.linODEs"></a> where <span class="math inline">\(y\)</span> is now a vector of length <span class="math inline">\(n\)</span> and <span class="math inline">\(A\)</span> is an <span class="math inline">\(n\)</span> by <span class="math inline">\(n\)</span> matrix. The solution is then given by: <span class="math display">\[
+y = e^{tA} \, y_0
+\]</span> <a name="id:ode.linOEs.sln"></a> where the matrix exponential is formally defined by the convergent power series: <span class="math display">\[
+e^{tA} = \sum_{n=0}^{\infty} \dfrac{tA^n}{n!} = I + tA + \frac{t^2A^2}{2!} + \dotsb
+\]</span> <a name="id:ode.matrix_exp.def"></a></p>
+<p>We can apply this technique to the simple harmonic oscillator example, by setting <span class="math display">\[
+y = \begin{bmatrix} y_1 \\ y_2 \end{bmatrix} \qquad
+A = \begin{bmatrix} 0 &amp; 1 \\ -1 &amp; -\theta \end{bmatrix}
+\]</span> <a name="id:ode.sho_matrix"></a></p>
+<p>The Stan model to simulate noisy observations using a matrix exponential function is given below.</p>
+<p>In general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] y0;</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span> ts;</span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">1</span>] <span class="dt">real</span> theta;</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-9"><a href="#cb13-9" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb13-10"><a href="#cb13-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">vector</span>[<span class="dv">2</span>] y_sim;</span>
+<span id="cb13-11"><a href="#cb13-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">2</span>, <span class="dv">2</span>] A = [[ <span class="dv">0</span>,  <span class="dv">1</span>],</span>
+<span id="cb13-12"><a href="#cb13-12" aria-hidden="true" tabindex="-1"></a>                    [-<span class="dv">1</span>, -theta[<span class="dv">1</span>]]]</span>
+<span id="cb13-13"><a href="#cb13-13" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb13-14"><a href="#cb13-14" aria-hidden="true" tabindex="-1"></a>    y_sim[t] = matrix_exp((t - <span class="dv">1</span>) * A) * y0;</span>
+<span id="cb13-15"><a href="#cb13-15" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb13-16"><a href="#cb13-16" aria-hidden="true" tabindex="-1"></a>  <span class="co">// add measurement error</span></span>
+<span id="cb13-17"><a href="#cb13-17" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb13-18"><a href="#cb13-18" aria-hidden="true" tabindex="-1"></a>    y_sim[t, <span class="dv">1</span>] += normal_rng(<span class="dv">0</span>, <span class="fl">0.1</span>);</span>
+<span id="cb13-19"><a href="#cb13-19" aria-hidden="true" tabindex="-1"></a>    y_sim[t, <span class="dv">2</span>] += normal_rng(<span class="dv">0</span>, <span class="fl">0.1</span>);</span>
+<span id="cb13-20"><a href="#cb13-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb13-21"><a href="#cb13-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters <code>theta</code> and initial state <code>y0</code> are read in as data along observation times <code>ts</code>. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations <code>y_sim</code>. Because the ODEs are linear, we can use the <code>matrix_exp</code> function to solve the system.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-AhnertMulansky:2011" class="csl-entry" role="listitem">
+Ahnert, Karsten, and Mario Mulansky. 2011. <span>“Odeint—Solving Ordinary Differential Equations in <span>C</span>++.”</span> <em>arXiv</em> 1110.3397.
+</div>
+<div id="ref-Brenan:1996" class="csl-entry" role="listitem">
+Brenan, K. E., S. L. Campbell, and L. R. Petzold. 1995. <em><span>N</span>umerical <span>S</span>olution of <span>Initial-Value</span> <span>P</span>roblems in <span>Differential-Algebraic</span> <span>E</span>quations</em>. SIAM Classics in Applied Mathematics 14. Philadelphia: Society for Industrial; Applied Mathematics.
+</div>
+<div id="ref-cashvariable:1990" class="csl-entry" role="listitem">
+Cash, J. R., and Alan H. Karp. 1990. <span>“A Variable Order <span>Runge</span>-<span>Kutta</span> Method for Initial Value Problems with Rapidly Varying Right-Hand Sides.”</span> <em>ACM Transactions on Mathematical Software</em> 16 (3): 201–22. <a href="https://doi.org/10.1145/79505.79507">https://doi.org/10.1145/79505.79507</a>.
+</div>
+<div id="ref-CohenHindmarsh:1996" class="csl-entry" role="listitem">
+Cohen, Scott D, and Alan C Hindmarsh. 1996. <span>“<span>CVODE</span>, a Stiff/Nonstiff <span>ODE</span> Solver in <span>C</span>.”</span> <em>Computers in Physics</em> 10 (2): 138–43.
+</div>
+<div id="ref-DormandPrince:1980" class="csl-entry" role="listitem">
+Dormand, John R, and Peter J Prince. 1980. <span>“A Family of Embedded <span>R</span>unge-<span>K</span>utta Formulae.”</span> <em>Journal of Computational and Applied Mathematics</em> 6 (1): 19–26.
+</div>
+<div id="ref-hairer:1993" class="csl-entry" role="listitem">
+Hairer, Ernst, Syvert P. Nørsett, and Gerhard Wanner. 1993. <em>Solving <span>Ordinary</span> <span>Differential</span> <span>Equations</span> <span>I</span>: <span>Nonstiff</span> <span>Problems</span></em>. 2nd ed. Springer <span>Series</span> in <span>Computational</span> <span>Mathematics</span>, <span>Springer</span> <span>Ser</span>.<span>Comp</span>.<span>Mathem</span>. <span>Hairer</span>,<span>E</span>.:<span>Solving</span> <span>Ordinary</span> <span>Diff</span>. Berlin Heidelberg: Springer-Verlag.
+</div>
+<div id="ref-mazziatest:2012" class="csl-entry" role="listitem">
+Mazzia, F., J. R. Cash, and K. Soetaert. 2012. <span>“A <span>Test</span> <span>Set</span> for Stiff <span>Initial</span> <span>Value</span> <span>Problem</span> <span>Solvers</span> in the Open Source Software <span>R</span>: <span>Package</span> <span class="nocase">deTestSet</span>.”</span> <em>Journal of Computational and Applied Mathematics</em> 236 (16): 4119–31.
+</div>
+<div id="ref-montenbrucksatellite:2000" class="csl-entry" role="listitem">
+Montenbruck, Oliver, and Eberhard Gill. 2000. <em>Satellite <span>Orbits</span>: <span>Models</span>, <span>Methods</span> and <span>Applications</span></em>. Berlin Heidelberg: Springer-Verlag. <a href="https://doi.org/10.1007/978-3-642-58351-3">https://doi.org/10.1007/978-3-642-58351-3</a>.
+</div>
+<div id="ref-SerbanHindmarsh:2005" class="csl-entry" role="listitem">
+Serban, Radu, and Alan C Hindmarsh. 2005. <span>“<span>CVODES</span>: The Sensitivity-Enabled <span>ODE</span> Solver in <span>SUNDIALS</span>.”</span> In <em>ASME 2005 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference</em>, 257–69. American Society of Mechanical Engineers.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>This example is drawn from the documentation for the Boost Numeric Odeint library <span class="citation" data-cites="AhnertMulansky:2011">(<a href="#ref-AhnertMulansky:2011" role="doc-biblioref">Ahnert and Mulansky 2011</a>)</span>, which Stan uses to implement the <code>rk45</code> and <code>ckrk</code> solver.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/algebraic-equations.html" class="pagination-link" aria-label="Solving Algebraic Equations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Solving Algebraic Equations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/one-dimensional-integrals.html" class="pagination-link" aria-label="Computing One Dimensional Integrals">
+        <span class="nav-page-text">Computing One Dimensional Integrals</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/odes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/one-dimensional-integrals.html b/docs/2_39/stan-users-guide/one-dimensional-integrals.html
new file mode 100644
index 000000000..b48f60b63
--- /dev/null
+++ b/docs/2_39/stan-users-guide/one-dimensional-integrals.html
@@ -0,0 +1,1414 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Computing One Dimensional Integrals</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/complex-numbers.html" rel="next">
+<link href="../stan-users-guide/odes.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Computing One Dimensional Integrals">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/one-dimensional-integrals.html">Computing One Dimensional Integrals</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#integrate-1d" id="toc-integrate-1d" class="nav-link active" data-scroll-target="#integrate-1d">Computing One Dimensional Integrals</a>
+  <ul class="collapse">
+  <li><a href="#calling-the-integrator" id="toc-calling-the-integrator" class="nav-link" data-scroll-target="#calling-the-integrator">Calling the integrator</a>
+  <ul class="collapse">
+  <li><a href="#limits-of-integration" id="toc-limits-of-integration" class="nav-link" data-scroll-target="#limits-of-integration">Limits of integration</a></li>
+  <li><a href="#data-vs.-parameters" id="toc-data-vs.-parameters" class="nav-link" data-scroll-target="#data-vs.-parameters">Data vs.&nbsp;parameters</a></li>
+  </ul></li>
+  <li><a href="#integrator-convergence" id="toc-integrator-convergence" class="nav-link" data-scroll-target="#integrator-convergence">Integrator convergence</a>
+  <ul class="collapse">
+  <li><a href="#zero-crossing" id="toc-zero-crossing" class="nav-link" data-scroll-target="#zero-crossing">Zero-crossing integrals</a></li>
+  <li><a href="#integral-precision" id="toc-integral-precision" class="nav-link" data-scroll-target="#integral-precision">Avoiding precision loss near limits of integration in definite integrals</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/one-dimensional-integrals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/one-dimensional-integrals.html">Computing One Dimensional Integrals</a></li></ol></nav></header>
+
+
+
+
+<section id="integrate-1d" class="level1">
+<h1>Computing One Dimensional Integrals</h1>
+<p>Definite and indefinite one dimensional integrals can be performed in Stan using the <a href="https://mc-stan.org/docs/functions-reference/higher-order_functions.html#functions-1d-integrator"><code>integrate_1d</code> function</a></p>
+<p>As an example, the normalizing constant of a left-truncated normal distribution is</p>
+<p><span class="math display">\[
+  \int_a^\infty \frac{1}{\sqrt{2 \pi \sigma^2}} e^{-\frac{1}{2}\frac{(x - \mu)^2}{\sigma^2}}
+\]</span></p>
+<p>To compute this integral in Stan, the integrand must first be defined as a Stan function (see the Stan Reference Manual chapter on User-Defined Functions for more information on coding user-defined functions).</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> normal_density(<span class="dt">real</span> x,             <span class="co">// Function argument</span></span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">real</span> xc,            <span class="co">// Complement of function argument</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>                                        <span class="co">//  on the domain (defined later)</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span>[] <span class="dt">real</span> theta, <span class="co">// parameters</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span>[] <span class="dt">real</span> x_r,   <span class="co">// data (real)</span></span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>                    <span class="dt">array</span>[] <span class="dt">int</span> x_i) {  <span class="co">// data (integer)</span></span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu = theta[<span class="dv">1</span>];</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma = theta[<span class="dv">2</span>];</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> <span class="dv">1</span> / (sqrt(<span class="dv">2</span> * pi()) * sigma) * exp(-<span class="fl">0.5</span> * ((x - mu) / sigma)^<span class="dv">2</span>);</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This function is expected to return the value of the integrand evaluated at point <code>x</code>. The argument <code>xc</code> is used in definite integrals to avoid loss of precision near the limits of integration and is set to NaN when either limit is infinite (see the section on precision/loss in the chapter on Higher-Order Functions of the Stan Functions Reference for details on how to use this). The argument <code>theta</code> is used to pass in arguments of the integral that are a function of the parameters in our model. The arguments <code>x_r</code> and <code>x_i</code> are used to pass in real and integer arguments of the integral that are not a function of our parameters.</p>
+<p>The function defining the integrand must have exactly the argument types and return type of <code>normal_density</code> above, though argument naming is not important. Even if <code>x_r</code> and <code>x_i</code> are unused in the integrand, they must be included in the function signature. Even if the integral does not involve some of these, they must still be supplied some value. The most efficient will be a zero-length array or vector, which can be created with rep_array(0, 0) and rep_vector(0, 0), respectively. Other options include an uninitialized variable declared with size 0, which is equivalent to the above, or any easy value, such as size 1 array created with {0}.</p>
+<section id="calling-the-integrator" class="level2">
+<h2 class="anchored" data-anchor-id="calling-the-integrator">Calling the integrator</h2>
+<p>Suppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.</p>
+<p>Such a model might look like (include the function defined at the beginning of this chapter to make this code compile):</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">real</span> x_r;</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">int</span> x_i;</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="fl">0.0</span>&gt; sigma;</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> left_limit;</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>  left_limit ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> normal_lpdf(y | mu, sigma);</span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -log(integrate_1d(normal_density,</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>                              left_limit,</span>
+<span id="cb2-24"><a href="#cb2-24" aria-hidden="true" tabindex="-1"></a>                              positive_infinity(),</span>
+<span id="cb2-25"><a href="#cb2-25" aria-hidden="true" tabindex="-1"></a>                              { mu, sigma }, x_r, x_i));</span>
+<span id="cb2-26"><a href="#cb2-26" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="limits-of-integration" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="limits-of-integration">Limits of integration</h3>
+<p>The limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls <code>negative_infinity()</code> and <code>positive_infinity()</code>.</p>
+<p>If both limits are either <code>negative_infinity()</code> or <code>positive_infinity()</code>, the integral and its gradients are set to zero.</p>
+</section>
+<section id="data-vs.-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-vs.-parameters">Data vs.&nbsp;parameters</h3>
+<p>The arguments for the real data <code>x_r</code> and the integer data <code>x_i</code> must be expressions that only involve data or transformed data variables. <code>theta</code>, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.</p>
+<p>The endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).</p>
+</section>
+</section>
+<section id="integrator-convergence" class="level2">
+<h2 class="anchored" data-anchor-id="integrator-convergence">Integrator convergence</h2>
+<p>The integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library <span class="citation" data-cites="BoostQuadrature:2017">(<a href="#ref-BoostQuadrature:2017" role="doc-biblioref">Agrawal et al. 2017</a>)</span>. If the <span class="math inline">\(n\)</span>th estimate of the integral is denoted <span class="math inline">\(I_n\)</span> and the <span class="math inline">\(n\)</span>th estimate of the norm of the integral is denoted <span class="math inline">\(|I|_n\)</span>, the iteration is terminated when</p>
+<p><span class="math display">\[
+  \frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \text{relative tolerance}.
+\]</span></p>
+<p>The <code>relative_tolerance</code> parameter can be optionally specified as the last argument to <code>integrate_1d</code>. By default, <code>integrate_1d</code> follows the Boost library recommendation of setting <code>relative_tolerance</code> to the square root of the machine epsilon of double precision floating point numbers (about <code>1e-8</code>). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If <code>integrate_1d</code> causes an exception in transformed parameters block or model block, the result has the same effect as assigning a <span class="math inline">\(-\infty\)</span> log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If <code>integrate_1d</code> causes an exception in generated quantities block, the returned output from <code>integrate_1d</code> is NaN. In these cases, a bigger <code>relative_tolerance</code> value can be specified.</p>
+<section id="zero-crossing" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="zero-crossing">Zero-crossing integrals</h3>
+<p>Integrals on the (possibly infinite) interval <span class="math inline">\((a, b)\)</span> that cross zero are split into two integrals, one from <span class="math inline">\((a, 0)\)</span> and one from <span class="math inline">\((0, b)\)</span>. This is because the quadrature methods employed internally can have difficulty near zero.</p>
+<p>In this case, each integral is separately integrated to the given <code>relative_tolerance</code>.</p>
+</section>
+<section id="integral-precision" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="integral-precision">Avoiding precision loss near limits of integration in definite integrals</h3>
+<p>If care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.</p>
+<p>For instance, in integrating the pdf of a beta distribution when the values of <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> are small, most of the probability mass is lumped near zero and one.</p>
+<p>The pdf of a beta distribution is proportional to</p>
+<p><span class="math display">\[
+p(x) \propto x^{\alpha - 1}(1 - x)^{\beta - 1}
+\]</span></p>
+<p>Normalizing this distribution requires computing the integral of <span class="math inline">\(p(x)\)</span> from zero to one. In Stan code, the integrand might look like:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> beta(<span class="dt">real</span> x, <span class="dt">real</span> xc, <span class="dt">array</span>[] <span class="dt">real</span> theta, <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="dt">array</span>[] <span class="dt">int</span> x_i) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha = theta[<span class="dv">1</span>];</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta = theta[<span class="dv">2</span>];</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> x^(alpha - <span class="fl">1.0</span>) * (<span class="fl">1.0</span> - x)^(beta - <span class="fl">1.0</span>);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The issue is that there will be numerical breakdown in the precision of <code>1.0 - x</code> as <code>x</code> gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of <code>alpha</code> and <code>beta</code> much less than one.</p>
+<p>This is where <code>xc</code> is useful. It is defined, for definite integrals, as a high precision version of the distance from <code>x</code> to the nearest endpoint — <code>a - x</code> or <code>b - x</code> for a lower endpoint <code>a</code> and an upper endpoint <code>b</code>. To make use of this for the beta integral, the integrand can be re-coded:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> beta(<span class="dt">real</span> x, <span class="dt">real</span> xc, <span class="dt">array</span>[] <span class="dt">real</span> theta, <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="dt">array</span>[] <span class="dt">int</span> x_i) {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha = theta[<span class="dv">1</span>];</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta = theta[<span class="dv">2</span>];</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> v;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span>(x &gt; <span class="fl">0.5</span>) {</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>    v = x^(alpha - <span class="fl">1.0</span>) * xc^(beta - <span class="fl">1.0</span>);</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>    v = x^(alpha - <span class="fl">1.0</span>) * (<span class="fl">1.0</span> - x)^(beta - <span class="fl">1.0</span>);</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> v;</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case, as we approach the upper limit of integration <span class="math inline">\(a = 1\)</span>, <code>xc</code> will take on the value of <span class="math inline">\(a - x = 1 - x\)</span>. This version of the integrand will converge for much smaller values of <code>alpha</code> and <code>beta</code> than otherwise possible.</p>
+<p>Consider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount <span class="math inline">\(\delta\)</span>, and truncated at some value <span class="math inline">\(b\)</span>. If we were interested in calculating the expectation of a variable <span class="math inline">\(X\)</span> distributed in this way, we would need to calculate <span class="math display">\[
+\int_a^b xf(x)\,dx = \int_{\delta}^b xf(x)\,dx
+\]</span> in the numerator, where <span class="math inline">\(f(x)\)</span> is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> shift_lognormal_pdf(<span class="dt">real</span> x,</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>                         <span class="dt">real</span> mu,</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>                         <span class="dt">real</span> sigma,</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>                         <span class="dt">real</span> delta) {</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> p;</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  p = (<span class="fl">1.0</span> / ((x - delta) * sigma * sqrt(<span class="dv">2</span> * pi()))) *</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>    exp(-<span class="dv">1</span> * (log(x - delta) - mu)^<span class="dv">2</span> / (<span class="dv">2</span> * sigma^<span class="dv">2</span>));</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> p;</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Therefore, the function that we want to integrate is:</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> integrand(<span class="dt">real</span> x,</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>               <span class="dt">real</span> xc,</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">real</span> theta,</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">real</span> x_r,</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">int</span> x_i) {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> numerator;</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> p;</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu = theta[<span class="dv">1</span>];</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma = theta[<span class="dv">2</span>];</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = theta[<span class="dv">3</span>];</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b = theta[<span class="dv">4</span>];</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  p = shift_lognormal_pdf(x, mu, sigma, delta);</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a>  numerator = x * p;</span>
+<span id="cb6-17"><a href="#cb6-17" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-18"><a href="#cb6-18" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> numerator;</span>
+<span id="cb6-19"><a href="#cb6-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>What happens here is that, given that the log-normal distribution is shifted by <span class="math inline">\(\delta\)</span>, when we then try to integrate the numerator, our <code>x</code> starts at values just above <code>delta</code>. This, in turn, causes the <code>x - delta</code> term to be near zero, leading to a breakdown.</p>
+<p>We can use <code>xc</code>, and define the <code>integrand</code> as:</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> integrand(<span class="dt">real</span> x,</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>               <span class="dt">real</span> xc,</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">real</span> theta,</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">real</span> x_r,</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">int</span> x_i) {</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> numerator;</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> p;</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu = theta[<span class="dv">1</span>];</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma = theta[<span class="dv">2</span>];</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = theta[<span class="dv">3</span>];</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b = theta[<span class="dv">4</span>];</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (x &lt; delta + <span class="dv">1</span>) {</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>    p = shift_lognormal_pdf(xc, mu, sigma, delta);</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>    p = shift_lognormal_pdf(x, mu, sigma, delta);</span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>  numerator = x * p;</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> numerator;</span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Why does this work? When our values of <code>x</code> are less than <code>delta + 1</code> (so, when they’re near <code>delta</code>, given that our lower bound of integration is equal to <span class="math inline">\(\delta\)</span>), we pass <code>xc</code> as an argument to our <code>shift_lognormal_pdf</code> function. This way, instead of dealing with <code>x - delta</code> in <code>shift_lognormal_pdf</code>, we are working with <code>xc - delta</code> which is equal to <code>delta - x - delta</code>, as <code>delta</code> is the lower endpoint in that case. The <code>delta</code> terms cancel out, and we are left with a high-precision version of <code>x</code>. We don’t encounter the same problem at the upper limit <span class="math inline">\(b\)</span> so we don’t adjust the code for that case.</p>
+<p>Note, <code>xc</code> is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, <code>xc</code> will be NaN.</p>
+<p>For zero-crossing definite integrals (see section <a href="#zero-crossing">Zero Crossing</a>) the integrals are broken into two pieces (<span class="math inline">\((a, 0)\)</span> and <span class="math inline">\((0, b)\)</span> for endpoints <span class="math inline">\(a &lt; 0\)</span> and <span class="math inline">\(b &gt; 0\)</span>) and <code>xc</code> is a high precision version of the distance to the limits of each of the two integrals separately. This means <code>xc</code> will be a high precision version of <code>a - x</code>, <code>x</code>, or <code>b - x</code>, depending on the value of x and the endpoints.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-BoostQuadrature:2017" class="csl-entry" role="listitem">
+Agrawal, Nikhar, Anton Bikineev, Paul Bristow, Marco Guazzone, Christopher Kormanyos, Hubert Holin, Bruno Lalande, et al. 2017. <span>“Double-Exponential Quadrature.”</span> https://www.boost.org/doc/libs/1_66_0/libs/math/doc/html/math_toolkit/double_exponential.html.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/odes.html" class="pagination-link" aria-label="Ordinary Differential Equations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Ordinary Differential Equations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/complex-numbers.html" class="pagination-link" aria-label="Complex Numbers">
+        <span class="nav-page-text">Complex Numbers</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/one-dimensional-integrals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/parallelization.html b/docs/2_39/stan-users-guide/parallelization.html
new file mode 100644
index 000000000..dc87548c5
--- /dev/null
+++ b/docs/2_39/stan-users-guide/parallelization.html
@@ -0,0 +1,1636 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Parallelization</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/posterior-prediction.html" rel="next">
+<link href="../stan-users-guide/efficiency-tuning.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Parallelization">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/parallelization.html">Parallelization</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#parallelization.chapter" id="toc-parallelization.chapter" class="nav-link active" data-scroll-target="#parallelization.chapter">Parallelization</a>
+  <ul class="collapse">
+  <li><a href="#reduce-sum" id="toc-reduce-sum" class="nav-link" data-scroll-target="#reduce-sum">Reduce-sum</a>
+  <ul class="collapse">
+  <li><a href="#example-logistic-regression" id="toc-example-logistic-regression" class="nav-link" data-scroll-target="#example-logistic-regression">Example: logistic regression</a></li>
+  <li><a href="#reduce-sum-grainsize" id="toc-reduce-sum-grainsize" class="nav-link" data-scroll-target="#reduce-sum-grainsize">Picking the grainsize</a></li>
+  </ul></li>
+  <li><a href="#map-rect" id="toc-map-rect" class="nav-link" data-scroll-target="#map-rect">Map-rect</a>
+  <ul class="collapse">
+  <li><a href="#map-function" id="toc-map-function" class="nav-link" data-scroll-target="#map-function">Map function</a></li>
+  <li><a href="#map-function-signature" id="toc-map-function-signature" class="nav-link" data-scroll-target="#map-function-signature">Map function signature</a></li>
+  <li><a href="#example-logistic-regression-1" id="toc-example-logistic-regression-1" class="nav-link" data-scroll-target="#example-logistic-regression-1">Example: logistic regression</a></li>
+  <li><a href="#example-hierarchical-logistic-regression" id="toc-example-hierarchical-logistic-regression" class="nav-link" data-scroll-target="#example-hierarchical-logistic-regression">Example: hierarchical logistic regression</a></li>
+  <li><a href="#ragged-inputs-and-outputs" id="toc-ragged-inputs-and-outputs" class="nav-link" data-scroll-target="#ragged-inputs-and-outputs">Ragged inputs and outputs</a></li>
+  </ul></li>
+  <li><a href="#opencl" id="toc-opencl" class="nav-link" data-scroll-target="#opencl">OpenCL</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/parallelization.html">Parallelization</a></li></ol></nav></header>
+
+
+
+
+<section id="parallelization.chapter" class="level1">
+<h1>Parallelization</h1>
+<p>Stan has support for different types of parallelization: multi-threading with Intel Threading Building Blocks (TBB), multi-processing with Message Passing Interface (MPI) and manycore processing with OpenCL.</p>
+<p>Multi-threading in Stan can be used with two mechanisms: reduce with summation and rectangular map. The latter can also be used with multi-processing.</p>
+<p>The advantages of reduce with summation are:</p>
+<ol type="1">
+<li>More flexible argument interface, avoiding the packing and unpacking that is necessary with rectangular map.</li>
+<li>Partitions data for parallelization automatically (this is done manually in rectangular map).</li>
+<li>Is easier to use.</li>
+</ol>
+<p>The advantages of rectangular map are:</p>
+<ol type="1">
+<li>Returns a list of vectors, while the reduce summation returns only a scalar.</li>
+<li>Can be parallelized across multiple cores and multiple computers, while reduce summation can only parallelized across multiple cores on a single machine.</li>
+</ol>
+<p>The actual speedup gained from using these functions will depend on many details. It is strongly recommended to only parallelize the computationally most expensive operations in a Stan program. Oftentimes this is the evaluation of the log likelihood for the observed data. When it is not clear which parts of the model is the most computationally expensive, we recommend using profiling, which is available in Stan 2.26 and newer.</p>
+<p>Since only portions of a Stan program will run in parallel, the maximal speedup one can achieve is capped, a phenomen described by <a href="https://en.wikipedia.org/wiki/Amdahl%27s_law">Amdahl’s law</a>.</p>
+<section id="reduce-sum" class="level2">
+<h2 class="anchored" data-anchor-id="reduce-sum">Reduce-sum</h2>
+<p>It is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If <code>g: U -&gt; real</code> is the function and <code>{ x1, x2, ... }</code> is an array of inputs, then that sum looks like:</p>
+<p><code>g(x1) + g(x2) + ...</code></p>
+<p><code>reduce_sum</code> and <code>reduce_sum_static</code> are tools for parallelizing these calculations.</p>
+<p>For efficiency reasons the reduce function doesn’t work with the element-wise evaluated function <code>g</code>, but instead the partial sum function <code>f: U[] -&gt; real</code>, where <code>f</code> computes the partial sum corresponding to a slice of the sequence <code>x</code> passed in. Due to the associativity of the sum reduction it holds that:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>g(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>                      = f({ x1, x2 }) + f({ x3 })</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>                      = f({ x1 }) + f({ x2, x3 })</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>                      = f({ x1 }) + f({ x2 }) + f({ x3 })</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With the partial sum function <code>f: U[] -&gt; real</code> reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:</p>
+<ul>
+<li><code>reduce_sum</code>: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.</li>
+<li><code>reduce_sum_static</code>: Compute the same sum as <code>reduce_sum</code>, but partition the input in the same way for given data set (in <code>reduce_sum</code> this partitioning might change depending on computer load).</li>
+</ul>
+<p><code>grainsize</code> is the one tuning parameter. For <code>reduce_sum</code>, <code>grainsize</code> is a suggested partial sum size. A <code>grainsize</code> of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using <code>reduce_sum</code> unless time is spent carefully picking <code>grainsize</code>. For picking a <code>grainsize</code>, see details <a href="#reduce-sum-grainsize">below</a>.</p>
+<p>For <code>reduce_sum_static</code>, <code>grainsize</code> specifies the maximal partial sum size. With <code>reduce_sum_static</code> it is more important to choose <code>grainsize</code> carefully since it entirely determines the partitioning of work. See details <a href="#reduce-sum-grainsize">below</a>.</p>
+<p>For efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array <code>{ x1, x2, ... }</code> and the shared arguments <code>s1, s2, ...</code> the effective sum (with individual terms) looks like:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>g(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>which can be written equivalently with partial sums to look like:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>f({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the particular slicing of the <code>x</code> array can change.</p>
+<p>Given this, the signatures are:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> reduce_sum(F f, <span class="dt">array</span>[] T x, <span class="dt">int</span> grainsize, T1 s1, T2 s2, ...)</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> reduce_sum_static(F f, <span class="dt">array</span>[] T x, <span class="dt">int</span> grainsize, T1 s1, T2 s2, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<ol type="1">
+<li><code>f</code> - User defined function that computes partial sums</li>
+<li><code>x</code> - Array to slice, each element corresponds to a term in the summation</li>
+<li><code>grainsize</code> - Target for size of slices</li>
+<li><code>s1, s2, ...</code> - Arguments shared in every term</li>
+</ol>
+<p>The user-defined partial sum functions have the signature:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> f(<span class="dt">array</span>[] T x_slice, <span class="dt">int</span> start, <span class="dt">int</span> end, T1 s1, T2 s2, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and take the arguments:</p>
+<ol type="1">
+<li><code>x_slice</code> - The subset of <code>x</code> (from <code>reduce_sum</code> / <code>reduce_sum_static</code>) for which this partial sum is responsible (<code>x_slice = x[start:end]</code>)</li>
+<li><code>start</code> - An integer specifying the first term in the partial sum</li>
+<li><code>end</code> - An integer specifying the last term in the partial sum (inclusive)</li>
+<li><code>s1, s2, ...</code> - Arguments shared in every term (passed on without modification from the <code>reduce_sum</code> / <code>reduce_sum_static</code> call)</li>
+</ol>
+<p>The user-provided function <code>f</code> is expected to compute the partial sum with the terms <code>start</code> through <code>end</code> of the overall sum. The user function is passed the subset <code>x[start:end]</code> as <code>x_slice</code>. <code>start</code> and <code>end</code> are passed so that <code>f</code> can index any of the tailing <code>sM</code> arguments as necessary. The trailing <code>sM</code> arguments are passed without modification to every call of <code>f</code>.</p>
+<p>A <code>reduce_sum</code> (or <code>reduce_sum_static</code>) call:</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum = reduce_sum(f, x, grainsize, s1, s2, ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>can be replaced by either:</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum = f(x, <span class="dv">1</span>, size(x), s1, s2, ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>or the code:</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum = <span class="fl">0.0</span>;</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span>(i <span class="cf">in</span> <span class="dv">1</span>:size(x)) {</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  sum += f({ x[i] }, i, i, s1, s2, ...);</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="example-logistic-regression" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="example-logistic-regression">Example: logistic regression</h3>
+<p>Logistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> y;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] beta;</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x);</span>
+<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this model predictions are made about the <code>N</code> outputs <code>y</code> using the covariate <code>x</code>. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>y ~ bernoulli_logit(beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>can be rewritten (up to a proportionality constant) as:</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span>(n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> bernoulli_logit_lpmf(y[n] | beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x[n])</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in <a href="#reduce-sum">Reduce-Sum</a>, such a function can be written like:</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> partial_sum(<span class="dt">array</span>[] <span class="dt">int</span> y_slice,</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>                   <span class="dt">int</span> start, <span class="dt">int</span> end,</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>                   <span class="dt">vector</span> x,</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>                   <span class="dt">vector</span> beta) {</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> bernoulli_logit_lpmf(y_slice | beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x[start:end]);</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The likelihood statement in the model can now be written:</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> partial_sum(y, <span class="dv">1</span>, N, x, beta); <span class="co">// Sum terms 1 to N of the likelihood</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this example, <code>y</code> was chosen to be sliced over because there is one term in the summation per value of <code>y</code>. Technically <code>x</code> would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g.&nbsp;slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because <code>x</code> is a shared argument, it is subset accordingly with <code>start:end</code>. With this function, reduce summation can be used to automatically parallelize the likelihood:</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> grainsize = <span class="dv">1</span>;</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> reduce_sum(partial_sum, y,</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>                     grainsize,</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>                     x, beta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The reduce summation facility automatically breaks the sum into pieces and computes them in parallel. <code>grainsize = 1</code> specifies that the <code>grainsize</code> should be estimated automatically. The final model is:</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> partial_sum(<span class="dt">array</span>[] <span class="dt">int</span> y_slice,</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>                   <span class="dt">int</span> start, <span class="dt">int</span> end,</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>                   <span class="dt">vector</span> x,</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>                   <span class="dt">vector</span> beta) {</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> bernoulli_logit_lpmf(y_slice | beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x[start:end]);</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb15-11"><a href="#cb15-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> y;</span>
+<span id="cb15-12"><a href="#cb15-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb15-13"><a href="#cb15-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-14"><a href="#cb15-14" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb15-15"><a href="#cb15-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] beta;</span>
+<span id="cb15-16"><a href="#cb15-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-17"><a href="#cb15-17" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-18"><a href="#cb15-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> grainsize = <span class="dv">1</span>;</span>
+<span id="cb15-19"><a href="#cb15-19" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();</span>
+<span id="cb15-20"><a href="#cb15-20" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> reduce_sum(partial_sum, y,</span>
+<span id="cb15-21"><a href="#cb15-21" aria-hidden="true" tabindex="-1"></a>                       grainsize,</span>
+<span id="cb15-22"><a href="#cb15-22" aria-hidden="true" tabindex="-1"></a>                       x, beta);</span>
+<span id="cb15-23"><a href="#cb15-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="reduce-sum-grainsize" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reduce-sum-grainsize">Picking the grainsize</h3>
+<p>The rational for choosing a sensible <code>grainsize</code> is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.</p>
+<p>In <code>reduce_sum</code>, <code>grainsize</code> is a recommendation on how to partition the work in the partial sum into smaller pieces. A <code>grainsize</code> of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.</p>
+<p>In <code>reduce_sum_static</code>, <code>grainsize</code> is an upper limit on the worksize. Work will be split until all partial sums are just smaller than <code>grainsize</code> (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible <code>grainsize</code>.</p>
+<p>In order to figure out an optimal <code>grainsize</code>, if there are <code>N</code> terms and <code>M</code> cores, run a quick test model with <code>grainsize</code> set roughly to <code>N / M</code>. Record the time, cut the <code>grainsize</code> in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable <code>grainsize</code> for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.</p>
+<p>For instance, in a model with <code>N=10000</code> and <code>M = 4</code>, start with <code>grainsize = 2500</code>, and sequentially try <code>grainsize = 1250</code>, <code>grainsize = 625</code>, etc.</p>
+<p>It is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller <code>grainsize</code> that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a <code>grainsize</code> of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.</p>
+</section>
+</section>
+<section id="map-rect" class="level2">
+<h2 class="anchored" data-anchor-id="map-rect">Map-rect</h2>
+<p>Map-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).</p>
+<p>A <em>map function</em> is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, <span class="math inline">\(f(x) = x^2\)</span>, over the vector <span class="math inline">\([3, 5, 10]\)</span> produces the vector <span class="math inline">\([9, 25, 100]\)</span>. In other words, map applies the square function elementwise.</p>
+<p>The output of mapping a sequence is often fed into a reduction. A <em>reduction function</em> takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, <em>map-reduce</em>.</p>
+<section id="map-function" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="map-function">Map function</h3>
+<p>In order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.</p>
+</section>
+<section id="map-function-signature" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="map-function-signature">Map function signature</h3>
+<p>Stan’s map function has the following signature</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> map_rect((<span class="dt">vector</span>, <span class="dt">vector</span>, <span class="dt">array</span>[] <span class="dt">real</span>, <span class="dt">array</span>[] <span class="dt">int</span>):<span class="dt">vector</span> f,</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>                <span class="dt">vector</span> phi, <span class="dt">array</span>[] <span class="dt">vector</span> thetas,</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>                <span class="kw">data</span> <span class="dt">array</span>[,] <span class="dt">real</span> x_rs, <span class="kw">data</span> <span class="dt">array</span>[,] <span class="dt">int</span> x_is);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The arrays <code>thetas</code> of parameters, <code>x_rs</code> of real data, and <code>x_is</code> of integer data have the suffix “<code>s</code>” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function <code>f</code>. The value of <code>phi</code> is reused in each mapped operation.</p>
+<p>The <code>_rect</code> suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.</p>
+<p>The last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the <code>data</code> qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.</p>
+<p>The notation <code>(vector, vector, array[] real, array[] int):vector</code> indicates that the function argument <code>f</code> must have the following signature.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> f(<span class="dt">vector</span> phi, <span class="dt">vector</span> theta,</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>         <span class="kw">data</span> <span class="dt">array</span>[] <span class="dt">real</span> x_r, <span class="kw">data</span> <span class="dt">array</span>[] <span class="dt">int</span> x_i);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although <code>f</code> will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.</p>
+<section id="map-function-semantics" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="map-function-semantics">Map function semantics</h4>
+<p>Stan’s map function applies the function <code>f</code> to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments <code>theta</code>, <code>x_r</code>, and <code>x_i</code> must be arrays of the same size. If the arrays are all size <code>N</code>, the result is defined as follows.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a>map_rect(f, phi, thetas, xs, ns)</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>= f(phi, thetas[<span class="dv">1</span>], xs[<span class="dv">1</span>], ns[<span class="dv">1</span>]) . f(phi, thetas[<span class="dv">2</span>], xs[<span class="dv">2</span>], ns[<span class="dv">2</span>])</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>  . ... . f(phi, thetas[N], xs[N], ns[N])</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The dot operators in the notation above are meant to indicate concatenation (implemented as <code>append_row</code> in Stan). The output of each application of <code>f</code> is a vector, and the sequence of <code>N</code> vectors is concatenated together to return a single vector.</p>
+</section>
+</section>
+<section id="example-logistic-regression-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="example-logistic-regression-1">Example: logistic regression</h3>
+<p>An example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.</p>
+<section id="unmapped-logistic-regression" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="unmapped-logistic-regression">Unmapped logistic regression</h4>
+<p>Consider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">int</span> y;</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">real</span> x;</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] beta;</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();</span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * to_vector(x));</span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The <code>bernoulli_logit</code> distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.</p>
+</section>
+<section id="mapped-logistic-regression" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="mapped-logistic-regression">Mapped logistic regression</h4>
+<p>The unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> lr(<span class="dt">vector</span> beta, <span class="dt">vector</span> theta, <span class="dt">array</span>[] <span class="dt">real</span> x, <span class="dt">array</span>[] <span class="dt">int</span> y) {</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> lp = bernoulli_logit_lpmf(y | beta[<span class="dv">1</span>]</span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>                                       + to_vector(x) * beta[<span class="dv">2</span>]);</span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> [lp]';</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb20-9"><a href="#cb20-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">int</span> y;</span>
+<span id="cb20-10"><a href="#cb20-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">12</span>] <span class="dt">real</span> x;</span>
+<span id="cb20-11"><a href="#cb20-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-12"><a href="#cb20-12" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb20-13"><a href="#cb20-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// K = 3 shards</span></span>
+<span id="cb20-14"><a href="#cb20-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">int</span> ys = { y[<span class="dv">1</span>:<span class="dv">4</span>], y[<span class="dv">5</span>:<span class="dv">8</span>], y[<span class="dv">9</span>:<span class="dv">12</span>] };</span>
+<span id="cb20-15"><a href="#cb20-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">3</span>, <span class="dv">4</span>] <span class="dt">real</span> xs = { x[<span class="dv">1</span>:<span class="dv">4</span>], x[<span class="dv">5</span>:<span class="dv">8</span>], x[<span class="dv">9</span>:<span class="dv">12</span>] };</span>
+<span id="cb20-16"><a href="#cb20-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">vector</span>[<span class="dv">0</span>] theta;</span>
+<span id="cb20-17"><a href="#cb20-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-18"><a href="#cb20-18" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-19"><a href="#cb20-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] beta;</span>
+<span id="cb20-20"><a href="#cb20-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-21"><a href="#cb20-21" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb20-22"><a href="#cb20-22" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();</span>
+<span id="cb20-23"><a href="#cb20-23" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> sum(map_rect(lr, beta, theta, xs, ys));</span>
+<span id="cb20-24"><a href="#cb20-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The first piece of the code is the actual function to compute the logistic regression. The argument <code>beta</code> will contain the regression coefficients (intercept and slope), as before. The second argument <code>theta</code> of job-specific parameters is not used, but nevertheless must be present. The modeled data <code>y</code> is passed as an array of integers and the predictors <code>x</code> as an array of real values. The function body then computes the log probability mass of <code>y</code> and assigns it to the local variable <code>lp</code>. This variable is then used in <code>[lp]'</code> to construct a row vector and then transpose it to a vector to return.</p>
+<p>The data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>The value <code>3</code> is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array <code>ys</code> contains the modeled data variables; each element of the array <code>ys</code> is an array of size four. The second array <code>xs</code> is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors <code>x</code> stand in a one-to-one relationship with the modeled data <code>y</code>. The final array <code>theta</code> is also of size three; its elements are empty vectors, because there are no shard-specific parameters.</p>
+<p>The parameters and the prior are as before. The likelihood is now coded using map-reduce. The function <code>lr</code> to compute the log probability mass is mapped over the data <code>xs</code> and <code>ys</code>, which contain the original predictors and outcomes broken into shards. The parameters <code>beta</code> are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters <code>theta</code> contains only empty vectors.</p>
+</section>
+</section>
+<section id="example-hierarchical-logistic-regression" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="example-hierarchical-logistic-regression">Example: hierarchical logistic regression</h3>
+<p>Consider a hierarchical model of American presidential voting behavior based on state of residence.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>Each of the fifty states <span class="math inline">\(k \in \{1,\dotsc,50\}\)</span> will have its own slope <span class="math inline">\(\beta_k\)</span> and intercept <span class="math inline">\(\alpha_k\)</span> to model the log odds of voting for the Republican candidate as a function of income. Suppose there are <span class="math inline">\(N\)</span> voters and with voter <span class="math inline">\(n \in 1{:}N\)</span> being in state <span class="math inline">\(s[n]\)</span> with income <span class="math inline">\(x_n\)</span>. The data model for the vote <span class="math inline">\(y_n \in \{ 0, 1 \}\)</span> is <span class="math display">\[
+y_n \sim \textsf{Bernoulli}
+\Big(
+  \operatorname{logit}^{-1}\left( \alpha_{s[n]} + \beta_{s[n]} \, x_n \right)
+\Big).
+\]</span></p>
+<p>The slopes and intercepts get hierarchical priors, <span class="math display">\[\begin{align*}
+\alpha_k &amp;\sim \textsf{normal}(\mu_{\alpha}, \sigma_{\alpha}) \\
+\beta_k  &amp;\sim \textsf{normal}(\mu_{\beta}, \sigma_{\beta})
+\end{align*}\]</span></p>
+<section id="unmapped-implementation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="unmapped-implementation">Unmapped implementation</h4>
+<p>This model can be coded up in Stan directly as follows.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;</span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, <span class="dv">2</span>] beta;</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] mu;</span>
+<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[<span class="dv">2</span>] sigma;</span>
+<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb21-14"><a href="#cb21-14" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb21-15"><a href="#cb21-15" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb21-16"><a href="#cb21-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">2</span>) {</span>
+<span id="cb21-17"><a href="#cb21-17" aria-hidden="true" tabindex="-1"></a>    beta[ , i] ~ normal(mu[i], sigma[i]);</span>
+<span id="cb21-18"><a href="#cb21-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-19"><a href="#cb21-19" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(beta[kk, <span class="dv">1</span>] + beta[kk, <span class="dv">2</span>] .* x);</span>
+<span id="cb21-20"><a href="#cb21-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For this model the vector of predictors <code>x</code> is coded as a vector, corresponding to how it is used in the model. The priors for <code>mu</code> and <code>sigma</code> are vectorized. The priors on the two components of <code>beta</code> (intercept and slope, respectively) are stored in a <span class="math inline">\(K \times 2\)</span> matrix.</p>
+<p>The distribution statement is also vectorized using multi-indexing with index <code>kk</code> for the states and elementwise multiplication (<code>.*</code>) for the income <code>x</code>. The vectorized distribution statement works out to the same thing as the following less efficient looped form.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(beta[kk[n], <span class="dv">1</span>] + beta[kk[n], <span class="dv">2</span>] * x[n]);</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="mapped-implementation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="mapped-implementation">Mapped implementation</h4>
+<p>The mapped version of the model will map over the states <code>K</code>. This means the group-level parameters, real data, and integer-data must be arrays of the same size.</p>
+<p>The mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name <code>bernoulli_logit_glm</code> derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a> <span class="dt">vector</span> bl_glm(<span class="dt">vector</span> mu_sigma, <span class="dt">vector</span> beta,</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>               <span class="dt">array</span>[] <span class="dt">real</span> x, <span class="dt">array</span>[] <span class="dt">int</span> y) {</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>   <span class="dt">vector</span>[<span class="dv">2</span>] mu = mu_sigma[<span class="dv">1</span>:<span class="dv">2</span>];</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>   <span class="dt">vector</span>[<span class="dv">2</span>] sigma = mu_sigma[<span class="dv">3</span>:<span class="dv">4</span>];</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>   <span class="dt">real</span> lp = normal_lpdf(beta | mu, sigma);</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>   <span class="dt">real</span> ll = bernoulli_logit_lpmf(y | beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * to_vector(x));</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a>   <span class="cf">return</span> [lp + ll]';</span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a> }</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The shared parameter <code>mu_sigma</code> contains the locations (<code>mu_sigma[1:2]</code>) and scales (<code>mu_sigma[3:4]</code>) of the priors, which are extracted in the first two lines of the program. The variable <code>lp</code> is assigned the log density of the prior on <code>beta</code>. The vector <code>beta</code> is of size two, as are the vectors <code>mu</code> and <code>sigma</code>, so everything lines up for the vectorization. Next, the variable <code>ll</code> is assigned to the log likelihood contribution for the group. Here <code>beta[1]</code> is the intercept of the regression and <code>beta[2]</code> the slope. The predictor array <code>x</code> needs to be converted to a vector allow the multiplication.</p>
+<p>The data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;</span>
+<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;</span>
+<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb24-8"><a href="#cb24-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb24-9"><a href="#cb24-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J = N / K;</span>
+<span id="cb24-10"><a href="#cb24-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K, J] <span class="dt">real</span> x_r;</span>
+<span id="cb24-11"><a href="#cb24-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K, J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; x_i;</span>
+<span id="cb24-12"><a href="#cb24-12" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb24-13"><a href="#cb24-13" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> pos = <span class="dv">1</span>;</span>
+<span id="cb24-14"><a href="#cb24-14" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb24-15"><a href="#cb24-15" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> end = pos + J - <span class="dv">1</span>;</span>
+<span id="cb24-16"><a href="#cb24-16" aria-hidden="true" tabindex="-1"></a>      x_r[k] = to_array_1d(x[pos:end]);</span>
+<span id="cb24-17"><a href="#cb24-17" aria-hidden="true" tabindex="-1"></a>      x_i[k] = to_array_1d(y[pos:end]);</span>
+<span id="cb24-18"><a href="#cb24-18" aria-hidden="true" tabindex="-1"></a>      pos += J;</span>
+<span id="cb24-19"><a href="#cb24-19" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb24-20"><a href="#cb24-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb24-21"><a href="#cb24-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The integer <code>J</code> is set to the number of observations per group.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>The real data array <code>x_r</code> holds the predictors and the integer data array <code>x_i</code> holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector <code>x</code> (and converting it to an array) and slicing the outcome array <code>y</code>.</p>
+<p>Given the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters <code>mu</code> and <code>sigma</code>, but moves the prior for <code>beta</code> and the likelihood to the mapped function.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">vector</span>[<span class="dv">2</span>] beta;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] mu;</span>
+<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[<span class="dv">2</span>] sigma;</span>
+<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb25-9"><a href="#cb25-9" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));</span>
+<span id="cb25-10"><a href="#cb25-10" aria-hidden="true" tabindex="-1"></a>                         </span>
+<span id="cb25-11"><a href="#cb25-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.</p>
+</section>
+</section>
+<section id="ragged-inputs-and-outputs" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ragged-inputs-and-outputs">Ragged inputs and outputs</h3>
+<p>The previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by <code>map_rect</code>.</p>
+<section id="ragged-inputs" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="ragged-inputs">Ragged inputs</h4>
+<p>If each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.</p>
+</section>
+<section id="ragged-outputs" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="ragged-outputs">Ragged outputs</h4>
+<p>The output of each mapped function is concatenated in order of inputs to produce the output of <code>map_rect</code>. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the <code>to_matrix</code> function if they are rectangular.</p>
+<p>If the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.</p>
+</section>
+</section>
+</section>
+<section id="opencl" class="level2">
+<h2 class="anchored" data-anchor-id="opencl">OpenCL</h2>
+<p>OpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.</p>
+<p>Stan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the <code>STAN_OPENCL</code> makefile flag. Setting this flag means that the Stan-to-C++ translator (<code>stanc3</code>) will be supplied the <code>--use-opencl</code> flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.</p>
+<p>In Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:</p>
+<ul>
+<li>bernoulli_lpmf</li>
+<li>bernoulli_logit_lpmf</li>
+<li>bernoulli_logit_glm_lpmf*</li>
+<li>beta_lpdf</li>
+<li>beta_proportion_lpdf</li>
+<li>binomial_lpmf</li>
+<li>categorical_logit_glm_lpmf*</li>
+<li>cauchy_lpdf</li>
+<li>chi_square_lpdf</li>
+<li>double_exponential_lpdf</li>
+<li>exp_mod_normal_lpdf</li>
+<li>exponential_lpdf</li>
+<li>frechet_lpdf</li>
+<li>gamma_lpdf</li>
+<li>gumbel_lpdf</li>
+<li>inv_chi_square_lpdf</li>
+<li>inv_gamma_lpdf</li>
+<li>logistic_lpdf</li>
+<li>lognormal_lpdf</li>
+<li>neg_binomial_lpmf</li>
+<li>neg_binomial_2_lpmf</li>
+<li>neg_binomial_2_log_lpmf</li>
+<li>neg_binomial_2_log_glm_lpmf*</li>
+<li>normal_lpdf</li>
+<li>normal_id_glm_lpdf*</li>
+<li>ordered_logistic_glm_lpmf*</li>
+<li>pareto_lpdf</li>
+<li>pareto_type_2_lpdf</li>
+<li>poisson_lpmf</li>
+<li>poisson_log_lpmf</li>
+<li>poisson_log_glm_lpmf*</li>
+<li>rayleigh_lpdf</li>
+<li>scaled_inv_chi_square_lpdf</li>
+<li>skew_normal_lpdf</li>
+<li>std_normal_lpdf</li>
+<li>student_t_lpdf</li>
+<li>uniform_lpdf</li>
+<li>weibull_lpdf</li>
+</ul>
+<p>* OpenCL is not used when the covariate argument to the GLM functions is a <code>row_vector</code>.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanHill:2007" class="csl-entry" role="listitem">
+Gelman, Andrew, and Jennifer Hill. 2007. <em>Data Analysis Using Regression and Multilevel-Hierarchical Models</em>. Cambridge, United Kingdom: Cambridge University Press.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The term “shard” is borrowed from databases, where it refers to a slice of the rows of a database. That is exactly what it is here if we think of rows of a dataframe. Stan’s shards are more general in that they need not correspond to rows of a dataframe.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>This example is a simplified form of the model described in <span class="citation" data-cites="GelmanHill:2007">(<a href="#ref-GelmanHill:2007" role="doc-biblioref">Gelman and Hill 2007, sec. 14.2</a>)</span><a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>This makes the strong assumption that each group has the same number of observations!<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/efficiency-tuning.html" class="pagination-link" aria-label="Efficiency Tuning">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Efficiency Tuning</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/posterior-prediction.html" class="pagination-link" aria-label="Posterior Predictive Sampling">
+        <span class="nav-page-text">Posterior Predictive Sampling</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/posterior-prediction.html b/docs/2_39/stan-users-guide/posterior-prediction.html
new file mode 100644
index 000000000..b176170af
--- /dev/null
+++ b/docs/2_39/stan-users-guide/posterior-prediction.html
@@ -0,0 +1,1476 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Posterior Predictive Sampling</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/simulation-based-calibration.html" rel="next">
+<link href="../stan-users-guide/parallelization.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Posterior Predictive Sampling">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Predictive Sampling</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#posterior-prediction.chapter" id="toc-posterior-prediction.chapter" class="nav-link active" data-scroll-target="#posterior-prediction.chapter">Posterior Predictive Sampling</a>
+  <ul class="collapse">
+  <li><a href="#posterior-predictive-distribution" id="toc-posterior-predictive-distribution" class="nav-link" data-scroll-target="#posterior-predictive-distribution">Posterior predictive distribution</a></li>
+  <li><a href="#computing-the-posterior-predictive-distribution" id="toc-computing-the-posterior-predictive-distribution" class="nav-link" data-scroll-target="#computing-the-posterior-predictive-distribution">Computing the posterior predictive distribution</a></li>
+  <li><a href="#sampling-from-the-posterior-predictive-distribution" id="toc-sampling-from-the-posterior-predictive-distribution" class="nav-link" data-scroll-target="#sampling-from-the-posterior-predictive-distribution">Sampling from the posterior predictive distribution</a></li>
+  <li><a href="#posterior-predictive-simulation-in-stan" id="toc-posterior-predictive-simulation-in-stan" class="nav-link" data-scroll-target="#posterior-predictive-simulation-in-stan">Posterior predictive simulation in Stan</a>
+  <ul class="collapse">
+  <li><a href="#simple-poisson-model" id="toc-simple-poisson-model" class="nav-link" data-scroll-target="#simple-poisson-model">Simple Poisson model</a></li>
+  <li><a href="#stan-code" id="toc-stan-code" class="nav-link" data-scroll-target="#stan-code">Stan code</a></li>
+  <li><a href="#analytic-posterior-and-posterior-predictive" id="toc-analytic-posterior-and-posterior-predictive" class="nav-link" data-scroll-target="#analytic-posterior-and-posterior-predictive">Analytic posterior and posterior predictive</a></li>
+  </ul></li>
+  <li><a href="#posterior-prediction-for-regressions" id="toc-posterior-prediction-for-regressions" class="nav-link" data-scroll-target="#posterior-prediction-for-regressions">Posterior prediction for regressions</a>
+  <ul class="collapse">
+  <li><a href="#posterior-predictive-distributions-for-regressions" id="toc-posterior-predictive-distributions-for-regressions" class="nav-link" data-scroll-target="#posterior-predictive-distributions-for-regressions">Posterior predictive distributions for regressions</a></li>
+  <li><a href="#stan-program" id="toc-stan-program" class="nav-link" data-scroll-target="#stan-program">Stan program</a></li>
+  </ul></li>
+  <li><a href="#estimating-event-probabilities" id="toc-estimating-event-probabilities" class="nav-link" data-scroll-target="#estimating-event-probabilities">Estimating event probabilities</a></li>
+  <li><a href="#stand-alone-generated-quantities-and-ongoing-prediction" id="toc-stand-alone-generated-quantities-and-ongoing-prediction" class="nav-link" data-scroll-target="#stand-alone-generated-quantities-and-ongoing-prediction">Stand-alone generated quantities and ongoing prediction</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-prediction.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Predictive Sampling</a></li></ol></nav></header>
+
+
+
+
+<section id="posterior-prediction.chapter" class="level1">
+<h1>Posterior Predictive Sampling</h1>
+<p>The goal of inference is often posterior prediction, that is evaluating or sampling from the posterior predictive distribution <span class="math inline">\(p(\tilde{y} \mid y),\)</span> where <span class="math inline">\(y\)</span> is observed data and <span class="math inline">\(\tilde{y}\)</span> is yet to be observed data. Often there are unmodeled predictors <span class="math inline">\(x\)</span> and <span class="math inline">\(\tilde{x}\)</span> for the observed data <span class="math inline">\(y\)</span> and unobserved data <span class="math inline">\(\tilde{y}\)</span>. With predictors, the posterior predictive density is <span class="math inline">\(p(\tilde{y} \mid
+\tilde{x}, x, y).\)</span> All of these variables may represent multivariate quantities.</p>
+<p>This chapter explains how to sample from the posterior predictive distribution in Stan, including applications to posterior predictive simulation and calculating event probabilities. These techniques can be coded in Stan using random number generation in the generated quantities block. Further, a technique for fitting and performing inference in two stages is presented in a section on stand-alone generated quantities in Stan</p>
+<section id="posterior-predictive-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="posterior-predictive-distribution">Posterior predictive distribution</h2>
+<p>Given a full Bayesian model <span class="math inline">\(p(y, \theta)\)</span>, the posterior predictive density for new data <span class="math inline">\(\tilde{y}\)</span> given observed data <span class="math inline">\(y\)</span> is <span class="math display">\[
+p(\tilde{y} \mid y)
+=
+\int p(\tilde{y} \mid \theta) \cdot p(\theta \mid y)
+\, \textrm{d}\theta.
+\]</span> The product under the integral reduces to the joint posterior density <span class="math inline">\(p(\tilde{y}, \theta \mid y),\)</span> so that the integral is simply marginalizing out the parameters <span class="math inline">\(\theta,\)</span> leaving the predictive density <span class="math inline">\(p(\tilde{y} \mid y)\)</span> of future observations given past observations.</p>
+</section>
+<section id="computing-the-posterior-predictive-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="computing-the-posterior-predictive-distribution">Computing the posterior predictive distribution</h2>
+<p>The posterior predictive density (or mass) of a prediction <span class="math inline">\(\tilde{y}\)</span> given observed data <span class="math inline">\(y\)</span> can be computed using <span class="math inline">\(M\)</span> Monte Carlo draws</p>
+<p><span class="math display">\[
+\theta^{(m)} \sim p(\theta \mid y)
+\]</span> from the posterior as <span class="math display">\[
+p(\tilde{y} \mid y)
+\approx
+\frac{1}{M} \sum_{m = 1}^M p(\tilde{y} \mid \theta^{(m)}).
+\]</span></p>
+<p>Computing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, <span class="math inline">\(\log
+p(\tilde{y} \mid y)\)</span> may be computed using the stable log sum of exponents function as <span class="math display">\[\begin{eqnarray*}
+\log p(\tilde{y} \mid y)
+&amp; \approx &amp;
+\log \frac{1}{M} \sum_{m = 1}^M p(\tilde{y} \mid \theta^{(m)}).
+\\[4pt]
+&amp; = &amp;
+- \log M
++ \textrm{log-sum-exp}_{m = 1}^M \log p(\tilde{y} \mid \theta^{(m)}),
+\end{eqnarray*}\]</span> where <span class="math display">\[
+\textrm{log-sum-exp}_{m = 1}^M v_m
+= \log \sum_{m = 1}^M \exp v_m
+\]</span> is used to maintain arithmetic precision. See the <a href="../stan-users-guide/floating-point.html#log-sum-of-exponentials">section on log sum of exponentials</a> for more details.</p>
+</section>
+<section id="sampling-from-the-posterior-predictive-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="sampling-from-the-posterior-predictive-distribution">Sampling from the posterior predictive distribution</h2>
+<p>Given draws from the posterior <span class="math inline">\(\theta^{(m)} \sim p(\theta \mid y),\)</span> draws from the posterior predictive <span class="math inline">\(\tilde{y}^{(m)} \sim p(\tilde{y}
+\mid y)\)</span> can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, <span class="math display">\[
+\tilde{y}^{(m)} \sim p(y \mid \theta^{(m)}).
+\]</span></p>
+<p>Randomly drawing <span class="math inline">\(\tilde{y}\)</span> from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because <span class="math inline">\(\theta\)</span> is unknown and estimated based only on a finite sample of data <span class="math inline">\(y\)</span>. Aleatoric uncertainty arises because even a known value of <span class="math inline">\(\theta\)</span> leads to uncertainty about new <span class="math inline">\(\tilde{y}\)</span> as described by the data model <span class="math inline">\(p(\tilde{y} \mid \theta)\)</span>. Both forms of uncertainty show up in the factored form of the posterior predictive distribution, <span class="math display">\[
+p(\tilde{y} \mid y)
+=
+\int
+\underbrace{p(\tilde{y} \mid \theta)}_{\begin{array}{l}
+                                         \textrm{aleatoric}
+                                         \\[-2pt] \textrm{uncertainty}
+                                       \end{array}}
+\cdot \underbrace{p(\theta \mid y)}_{\begin{array}{l}
+                                         \textrm{epistemic}
+                                         \\[-2pt] \textrm{uncertainty}
+                                       \end{array}}
+\, \textrm{d}\theta.
+\]</span></p>
+</section>
+<section id="posterior-predictive-simulation-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="posterior-predictive-simulation-in-stan">Posterior predictive simulation in Stan</h2>
+<p>Posterior predictive quantities can be coded in Stan using the generated quantities block.</p>
+<section id="simple-poisson-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="simple-poisson-model">Simple Poisson model</h3>
+<p>For example, consider a simple Poisson model for count data with a rate parameter <span class="math inline">\(\lambda &gt; 0\)</span> having a gamma-distributed prior, <span class="math display">\[
+\lambda \sim \textrm{gamma}(1, 1).
+\]</span> The <span class="math inline">\(N\)</span> observations <span class="math inline">\(y_1, \ldots, y_N\)</span> are modeled as Poisson distributed, <span class="math display">\[
+y_n \sim \textrm{poisson}(\lambda).
+\]</span></p>
+</section>
+<section id="stan-code" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-code">Stan code</h3>
+<p>The following Stan program defines a variable for <span class="math inline">\(\tilde{y}\)</span> by random number generation in the generated quantities block.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  lambda ~ gamma(<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  y ~ poisson(lambda);</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y_tilde = poisson_rng(lambda);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The random draw from the data model for <span class="math inline">\(\tilde{y}\)</span> is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new <span class="math inline">\(\tilde{y}^{(m)} \sim p(y \mid
+\lambda^{(m)})\)</span> for each posterior draw <span class="math inline">\(\lambda^{(m)} \sim p(\theta
+\mid y).\)</span></p>
+<p>The posterior draws <span class="math inline">\(\tilde{y}^{(m)}\)</span> may be used to estimate the expected value of <span class="math inline">\(\tilde{y}\)</span> or any of its quantiles or posterior intervals, as well as event probabilities involving <span class="math inline">\(\tilde{y}\)</span>. In general, <span class="math inline">\(\mathbb{E}[f(\tilde{y}, \theta) \mid y]\)</span> may be evaluated as <span class="math display">\[
+\mathbb{E}[f(\tilde{y}, \theta) \mid y]
+\approx \frac{1}{M} \sum_{m=1}^M f(\tilde{y}^{(m)}, \theta^{(m)}),
+\]</span> which is just the posterior mean of <span class="math inline">\(f(\tilde{y}, \theta).\)</span> This quantity is computed by Stan if the value of <span class="math inline">\(f(\tilde{y}, \theta)\)</span> is assigned to a variable in the generated quantities block. That is, if we have</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> f_val = f(y_tilde, theta);</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where the value of <span class="math inline">\(f(\tilde{y}, \theta)\)</span> is assigned to variable <code>f_val</code>, then the posterior mean of <code>f_val</code> will be the expectation <span class="math inline">\(\mathbb{E}[f(\tilde{y}, \theta) \mid y]\)</span>.</p>
+</section>
+<section id="analytic-posterior-and-posterior-predictive" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="analytic-posterior-and-posterior-predictive">Analytic posterior and posterior predictive</h3>
+<p>The gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density <span class="math inline">\(p(\lambda \mid y)\)</span> will also follow a gamma distribution.</p>
+<p>Because the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive <span class="math inline">\(p(\tilde{y} \mid
+y)\)</span> will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, <span class="math inline">\(y \sim
+\textrm{negative-binomial}(\alpha, \beta)\)</span> if <span class="math inline">\(\lambda \sim
+\textrm{gamma}(\alpha, \beta)\)</span> and <span class="math inline">\(y \sim \textrm{poisson}(\lambda).\)</span> Rather than marginalizing out the rate parameter <span class="math inline">\(\lambda\)</span> analytically as can be done to define the negative binomial probability mass function, the rate <span class="math inline">\(\lambda^{(m)} \sim p(\lambda \mid y)\)</span> is sampled from the posterior and then used to generate a draw of <span class="math inline">\(\tilde{y}^{(m)} \sim p(y \mid \lambda^{(m)}).\)</span></p>
+</section>
+</section>
+<section id="posterior-prediction-for-regressions" class="level2">
+<h2 class="anchored" data-anchor-id="posterior-prediction-for-regressions">Posterior prediction for regressions</h2>
+<section id="posterior-predictive-distributions-for-regressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="posterior-predictive-distributions-for-regressions">Posterior predictive distributions for regressions</h3>
+<p>Consider a regression with a single predictor <span class="math inline">\(x_n\)</span> for the training outcome <span class="math inline">\(y_n\)</span> and <span class="math inline">\(\tilde{x}_n\)</span> for the test outcome <span class="math inline">\(\tilde{y}_n.\)</span> Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in <span class="math display">\[\begin{eqnarray}
+p(\tilde{y} \mid \tilde{x}, y, x)
+&amp; = &amp; \int p(\tilde{y} \mid \tilde{x}, \theta) \cdot p(\theta \mid y, x) \,
+\textrm{d}\theta
+\\[4pt]
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m=1}^M \, p(\tilde{y} \mid \tilde{x}, \theta^{(m)}),
+\end{eqnarray}\]</span> where <span class="math inline">\(\theta^{(m)} \sim p(\theta \mid x, y).\)</span></p>
+</section>
+<section id="stan-program" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-program">Stan program</h3>
+<p>The following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed <span class="math inline">\(y\)</span> values are coded as data. The predictive quantities <span class="math inline">\(\tilde{y}\)</span> appear in the generated quantities block, where they are generated by random number generation.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  y ~ poisson_log(alpha + beta * x);</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  { alpha, beta } ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb3-17"><a href="#cb3-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_tilde] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y_tilde</span>
+<span id="cb3-18"><a href="#cb3-18" aria-hidden="true" tabindex="-1"></a>    = poisson_log_rng(alpha + beta * x_tilde);</span>
+<span id="cb3-19"><a href="#cb3-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s <code>poisson_log</code> vs.&nbsp;<code>poisson</code>, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept <code>alpha</code> and slope <code>beta</code>, are given standard normal priors.</p>
+<p>In the model block, the log rate for the Poisson is a linear function of the training data <span class="math inline">\(x\)</span>, whereas in the generated quantities block it is a function of the test data <span class="math inline">\(\tilde{x}\)</span>. Because the generated quantities block does not affect the posterior draws, the model fits <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> using only the training data, reserving <span class="math inline">\(\tilde{x}\)</span> to generate <span class="math inline">\(\tilde{y}.\)</span></p>
+<p>The result from running Stan is a predictive sample <span class="math inline">\(\tilde{y}^{(1)},
+\ldots \tilde{y}^{(M)}\)</span> where each <span class="math inline">\(\tilde{y}^{(m)} \sim p(\tilde{y}
+\mid \tilde{x}, x, y).\)</span></p>
+<p>The mean of the posterior predictive distribution is the expected value <span class="math display">\[\begin{align}
+\mathbb{E}[\tilde{y} \mid \tilde{x}, x, y]
+&amp; =
+\int
+\tilde{y}
+\cdot p(\tilde{y} \mid \tilde{x}, \theta)
+\cdot p(\theta \mid x, y)
+\, \textrm{d}\theta
+\\[4pt]
+&amp; \approx \frac{1}{M} \sum_{m = 1}^M \tilde{y}^{(m)},
+\end{align}\]</span> where the <span class="math inline">\(\tilde{y}^{(m)} \sim p(\tilde{y} \mid \tilde{x}, x, y)\)</span> are drawn from the posterior predictive distribution. Thus the posterior mean of <code>y_tilde[n]</code> after running Stan is the expected value of <span class="math inline">\(\tilde{y}_n\)</span> conditioned on the training data <span class="math inline">\(x, y\)</span> and predictor <span class="math inline">\(\tilde{x}_n.\)</span> This is the Bayesian estimate for <span class="math inline">\(\tilde{y}\)</span> with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for <span class="math inline">\(\tilde{y}\)</span>, as well as covariance of the <span class="math inline">\(\tilde{y_n}.\)</span> The posterior draws <span class="math inline">\(\tilde{y}^{(m)}\)</span> may also be used to estimate predictive event probabilities, such as <span class="math inline">\(\Pr[\tilde{y}_1 &gt; 0]\)</span> or <span class="math inline">\(\Pr[\prod_{n =
+1}^{\tilde{N}}(\tilde{y_n}) &gt; 1],\)</span> as expectations of indicator functions.</p>
+<p>All of this can be carried out by running Stan only a single time to draw a single sample of <span class="math inline">\(M\)</span> draws, <span class="math display">\[
+\tilde{y}^{(1)}, \ldots, \tilde{y}^{(M)} \sim p(\tilde{y} \mid
+\tilde{x}, x, y).
+\]</span> It’s only when moving to cross-validation where multiple runs are required.</p>
+</section>
+</section>
+<section id="estimating-event-probabilities" class="level2">
+<h2 class="anchored" data-anchor-id="estimating-event-probabilities">Estimating event probabilities</h2>
+<p>Event probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate <span class="math inline">\(\Pr[\lambda &gt; 5 \mid y]\)</span> in the simple Poisson example with only a rate parameter <span class="math inline">\(\lambda\)</span>, it suffices to define a generated quantity</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; lambda_gt_5 = lambda &gt; <span class="dv">5</span>;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The value of the expression <code>lambda &gt; 5</code> is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability <span class="math display">\[\begin{eqnarray*}
+\Pr[\lambda &gt; 5 \mid y]
+&amp; = &amp;
+\int \textrm{I}(\lambda &gt; 5) \cdot p(\lambda \mid y)
+\, \textrm{d}\lambda
+\\[4pt]
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m = 1}^M \textrm{I}[\lambda^{(m)} &gt; 5],
+\end{eqnarray*}\]</span> where each <span class="math inline">\(\lambda^{(m)} \sim p(\lambda \mid y)\)</span> is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter <code>lambda_gt_5</code>.</p>
+<p>In general, event probabilities may be expressed as expectations of indicator functions. For example, <span class="math display">\[\begin{eqnarray*}
+\Pr[\lambda &gt; 5 \mid y]
+&amp; = &amp; \mathbb{E}[\textrm{I}[\lambda &gt; 5] \mid y]
+\\[4pt]
+&amp; = &amp;
+\int
+\textrm{I}(\lambda &gt; 5) \cdot p(\lambda \mid y)
+\, \textrm{d}\lambda
+\\[4pt]
+&amp; \approx &amp; \frac{1}{M} \sum_{m = 1}^M \textrm{I}(\lambda^{(m)} &gt; 5).
+\end{eqnarray*}\]</span> The last line above is the posterior mean of the indicator function as coded in Stan.</p>
+<p>Event probabilities involving posterior predictive quantities <span class="math inline">\(\tilde{y}\)</span> work exactly the same way as those for parameters. For example, if <span class="math inline">\(\tilde{y}_n\)</span> is the prediction for the <span class="math inline">\(n\)</span>-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then <span class="math display">\[\begin{eqnarray*}
+\Pr[\tilde{y}_3 &gt; \tilde{y}_7 \mid \tilde{x}, x, y]
+&amp; = &amp;
+\mathbb{E}\!\left[I[\tilde{y}_3 &gt; \tilde{y}_7] \mid \tilde{x}, x, y\right]
+\\[4pt]
+&amp; = &amp;
+\int
+\textrm{I}(\tilde{y}_3 &gt; \tilde{y}_7)
+\cdot p(\tilde{y} \mid \tilde{x}, x, y)
+\, \textrm{d}\tilde{y}
+\\[4pt]
+&amp; \approx &amp;
+\frac{1}{M} \sum_{m = 1}^M
+\textrm{I}(\tilde{y}^{(m)}_3 &gt; \tilde{y}^{(m)}_7),
+\end{eqnarray*}\]</span> where <span class="math inline">\(\tilde{y}^{(m)} \sim p(\tilde{y} \mid \tilde{x}, x, y).\)</span></p>
+</section>
+<section id="stand-alone-generated-quantities-and-ongoing-prediction" class="level2">
+<h2 class="anchored" data-anchor-id="stand-alone-generated-quantities-and-ongoing-prediction">Stand-alone generated quantities and ongoing prediction</h2>
+<p>Stan’s sampling algorithms take a Stan program representing a posterior <span class="math inline">\(p(\theta \mid y, x)\)</span> along with actual data <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span> to produce a set of draws <span class="math inline">\(\theta^{(1)}, \ldots, \theta^{(M)}\)</span> from the posterior. Posterior predictive draws <span class="math inline">\(\tilde{y}^{(m)} \sim p(\tilde{y} \mid
+\tilde{x}, x, y)\)</span> can be generated by drawing <span class="math display">\[
+\tilde{y}^{(m)} \sim p(y \mid \tilde{x}, \theta^{(m)})
+\]</span> from the data model. Note that drawing <span class="math inline">\(\tilde{y}^{(m)}\)</span> only depends on the new predictors <span class="math inline">\(\tilde{x}\)</span> and the posterior draws <span class="math inline">\(\theta^{(m)}\)</span>. Most importantly, neither the original data or the model density is required.</p>
+<p>By saving the posterior draws, predictions for new data items <span class="math inline">\(\tilde{x}\)</span> may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="fl">0.5</span>);</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] y_tilde</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>    = normal_rng(alpha + beta * x_tilde, sigma);</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for <span class="math inline">\(\tilde{y}\)</span> or derived quantities such as event probabilities.</p>
+<p>Any posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/parallelization.html" class="pagination-link" aria-label="Parallelization">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Parallelization</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/simulation-based-calibration.html" class="pagination-link" aria-label="Simulation-Based Calibration Checking">
+        <span class="nav-page-text">Simulation-Based Calibration Checking</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-prediction.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/posterior-predictive-checks.html b/docs/2_39/stan-users-guide/posterior-predictive-checks.html
new file mode 100644
index 000000000..e76588b67
--- /dev/null
+++ b/docs/2_39/stan-users-guide/posterior-predictive-checks.html
@@ -0,0 +1,1594 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Posterior and Prior Predictive Checks</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/cross-validation.html" rel="next">
+<link href="../stan-users-guide/simulation-based-calibration.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script src="../site_libs/quarto-contrib/glightbox/glightbox.min.js"></script>
+<link href="../site_libs/quarto-contrib/glightbox/glightbox.min.css" rel="stylesheet">
+<link href="../site_libs/quarto-contrib/glightbox/lightbox.css" rel="stylesheet">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Posterior and Prior Predictive Checks">
+<meta property="og:description" content="">
+<meta property="og:image" content="./img/ppc-pois-pois.jpg">
+<meta property="og:site_name" content="Stan Docs">
+<meta property="og:image:alt" content="Posterior predictive checks for Poisson data generating process and Poisson model.">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-predictive-checks.html">Posterior and Prior Predictive Checks</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#ppcs.chapter" id="toc-ppcs.chapter" class="nav-link active" data-scroll-target="#ppcs.chapter">Posterior and Prior Predictive Checks</a>
+  <ul class="collapse">
+  <li><a href="#simulating-from-the-posterior-predictive-distribution" id="toc-simulating-from-the-posterior-predictive-distribution" class="nav-link" data-scroll-target="#simulating-from-the-posterior-predictive-distribution">Simulating from the posterior predictive distribution</a></li>
+  <li><a href="#plotting-multiples" id="toc-plotting-multiples" class="nav-link" data-scroll-target="#plotting-multiples">Plotting multiples</a></li>
+  <li><a href="#posterior-p-values" id="toc-posterior-p-values" class="nav-link" data-scroll-target="#posterior-p-values">Posterior ‘’p-values’’</a>
+  <ul class="collapse">
+  <li><a href="#which-statistics-to-test" id="toc-which-statistics-to-test" class="nav-link" data-scroll-target="#which-statistics-to-test">Which statistics to test?</a></li>
+  </ul></li>
+  <li><a href="#prior-predictive-checks" id="toc-prior-predictive-checks" class="nav-link" data-scroll-target="#prior-predictive-checks">Prior predictive checks</a>
+  <ul class="collapse">
+  <li><a href="#coding-prior-predictive-checks-in-stan" id="toc-coding-prior-predictive-checks-in-stan" class="nav-link" data-scroll-target="#coding-prior-predictive-checks-in-stan">Coding prior predictive checks in Stan</a></li>
+  </ul></li>
+  <li><a href="#example-of-prior-predictive-checks" id="toc-example-of-prior-predictive-checks" class="nav-link" data-scroll-target="#example-of-prior-predictive-checks">Example of prior predictive checks</a></li>
+  <li><a href="#mixed-replication" id="toc-mixed-replication" class="nav-link" data-scroll-target="#mixed-replication">Mixed predictive replication for hierarchical models</a></li>
+  <li><a href="#joint-model-representation" id="toc-joint-model-representation" class="nav-link" data-scroll-target="#joint-model-representation">Joint model representation</a>
+  <ul class="collapse">
+  <li><a href="#posterior-predictive-model" id="toc-posterior-predictive-model" class="nav-link" data-scroll-target="#posterior-predictive-model">Posterior predictive model</a></li>
+  <li><a href="#prior-predictive-model" id="toc-prior-predictive-model" class="nav-link" data-scroll-target="#prior-predictive-model">Prior predictive model</a></li>
+  <li><a href="#mixed-replication-for-hierarchical-models" id="toc-mixed-replication-for-hierarchical-models" class="nav-link" data-scroll-target="#mixed-replication-for-hierarchical-models">Mixed replication for hierarchical models</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-predictive-checks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-predictive-checks.html">Posterior and Prior Predictive Checks</a></li></ol></nav></header>
+
+
+
+
+<section id="ppcs.chapter" class="level1">
+<h1>Posterior and Prior Predictive Checks</h1>
+<p>Posterior predictive checks are a way of measuring whether a model does a good job of capturing relevant aspects of the data, such as means, standard deviations, and quantiles <span class="citation" data-cites="Rubin:1984 GelmanEtAl:1996">(<a href="#ref-Rubin:1984" role="doc-biblioref">Rubin 1984</a>; <a href="#ref-GelmanEtAl:1996" role="doc-biblioref">Andrew Gelman, Meng, and Stern 1996</a>)</span>. Posterior predictive checking works by simulating new replicated data sets based on the fitted model parameters and then comparing statistics applied to the replicated data set with the same statistic applied to the original data set.</p>
+<p>Prior predictive checks evaluate the prior the same way. Specifically, they evaluate what data sets would be consistent with the prior. They will not be calibrated with actual data, but extreme values help diagnose priors that are either too strong, too weak, poorly shaped, or poorly located.</p>
+<p>Prior and posterior predictive checks are two cases of the general concept of predictive checks, just conditioning on different things (no data and the observed data, respectively). For hierarchical models, there are intermediate versions, as discussed in the section on <a href="#mixed-replication">hierarchical models and mixed replication</a>.</p>
+<section id="simulating-from-the-posterior-predictive-distribution" class="level2">
+<h2 class="anchored" data-anchor-id="simulating-from-the-posterior-predictive-distribution">Simulating from the posterior predictive distribution</h2>
+<p>The posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data <span class="math inline">\(y\)</span>.</p>
+<p>The posterior predictive distribution for replications <span class="math inline">\(y^{\textrm{rep}}\)</span> of the original data set <span class="math inline">\(y\)</span> given model parameters <span class="math inline">\(\theta\)</span> is defined by <span class="math display">\[
+p(y^{\textrm{rep}} \mid y)
+= \int p(y^{\textrm{rep}} \mid \theta)
+       \cdot p(\theta \mid y) \, \textrm{d}\theta.
+\]</span></p>
+<p>As with other posterior predictive quantities, generating a replicated data set <span class="math inline">\(y^{\textrm{rep}}\)</span> from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters <span class="math inline">\(\theta = (\alpha, \beta, \sigma).\)</span></p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To generate a replicated data set <code>y_rep</code> for this simple model, the following generated quantities block suffices.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y_rep = normal_rng(alpha + beta * x, sigma);</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The vectorized form of the normal random number generator is used with the original predictors <code>x</code> and the model parameters <code>alpha, beta</code>, and <code>sigma.</code> The replicated data variable <code>y_rep</code> is declared to be the same size as the original data <code>y</code>, but instead of a vector type, it is declared to be an array of reals to match the return type of the function <code>normal_rng</code>. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.</p>
+<p>The posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in <a href="#posterior-prediction.chapter">the chapter on posterior predictions</a> in that the original predictors <span class="math inline">\(x\)</span> are used. That is, the posterior predictions are for the original data.</p>
+</section>
+<section id="plotting-multiples" class="level2">
+<h2 class="anchored" data-anchor-id="plotting-multiples">Plotting multiples</h2>
+<p>A standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.</p>
+<p>Here’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; mean_y = mean(to_vector(y));</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sd_y = sd(to_vector(y));</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  y ~ poisson(lambda);</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  lambda ~ exponential(<span class="fl">0.2</span>);</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb3-17"><a href="#cb3-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y_rep = poisson_rng(rep_array(lambda, N));</span>
+<span id="cb3-18"><a href="#cb3-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; mean_y_rep = mean(to_vector(y_rep));</span>
+<span id="cb3-19"><a href="#cb3-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sd_y_rep = sd(to_vector(y_rep));</span>
+<span id="cb3-20"><a href="#cb3-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; mean_gte = (mean_y_rep &gt;= mean_y);</span>
+<span id="cb3-21"><a href="#cb3-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; sd_gte = (sd_y_rep &gt;= sd_y);</span>
+<span id="cb3-22"><a href="#cb3-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The generated quantities block creates a variable <code>y_rep</code> for the replicated data, variables <code>mean_y_rep</code> and <code>sd_y_rep</code> for the statistics of the replicated data, and indicator variables <code>mean_gte</code> and <code>sd_gte</code> for whether the replicated statistic is greater than or equal to the statistic applied to the original data.</p>
+<p>Now consider generating data <span class="math inline">\(y \sim \textrm{Poisson}(5)\)</span>. The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><a href="./img/ppc-pois-pois.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="Posterior predictive checks for Poisson data generating process and Poisson model."><img src="./img/ppc-pois-pois.jpg" class="img-fluid figure-img" style="width:50.0%" alt="Posterior predictive checks for Poisson data generating process and Poisson model."></a></p>
+<figcaption>Posterior predictive checks for Poisson data generating process and Poisson model.</figcaption>
+</figure>
+</div>
+<p>With a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.</p>
+<p>Now consider generating over-dispersed data <span class="math inline">\(y \sim \textrm{negative-binomial2}(5, 1).\)</span> This has the same mean as <span class="math inline">\(\textrm{Poisson}(5)\)</span>, namely <span class="math inline">\(5\)</span>, but a standard deviation of <span class="math inline">\(\sqrt{5 + 5^2 /1} \approx 5.5.\)</span> There is no way to fit this data with the Poisson model, because a variable distributed as <span class="math inline">\(\textrm{Poisson}(\lambda)\)</span> has mean <span class="math inline">\(\lambda\)</span> and standard deviation <span class="math inline">\(\sqrt{\lambda},\)</span> which is <span class="math inline">\(\sqrt{5}\)</span> for <span class="math inline">\(\textrm{Poisson}(5).\)</span> Here’s the resulting small multiples plot, again with original data in the upper left.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><a href="./img/ppc-nb-pois.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-2" title="Posterior predictive checks for negative binomial data generating process and Poisson model."><img src="./img/ppc-nb-pois.jpg" class="img-fluid figure-img" style="width:50.0%" alt="Posterior predictive checks for negative binomial data generating process and Poisson model."></a></p>
+<figcaption>Posterior predictive checks for negative binomial data generating process and Poisson model.</figcaption>
+</figure>
+</div>
+<p>This time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.</p>
+</section>
+<section id="posterior-p-values" class="level2">
+<h2 class="anchored" data-anchor-id="posterior-p-values">Posterior ‘’p-values’’</h2>
+<p>If a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic <span class="math inline">\(s(\cdot)\)</span> in a replicated data set exceeds that in the original data, <span class="math display">\[
+\Pr\!\left[ s(y^{\textrm{rep}}) \geq s(y) \mid y \right]
+=
+\int
+\textrm{I}\left( s(y^{\textrm{rep}}) \geq s(y) \mid y \right)
+\cdot p\left( y^{\textrm{rep}} \mid y \right)
+\, \textrm{d}{y^{\textrm{rep}}}.
+\]</span> It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified <span class="citation" data-cites="BayarriBerger:2000">(<a href="#ref-BayarriBerger:2000" role="doc-biblioref">Bayarri and Berger 2000</a>)</span>.</p>
+<p>Nevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.</p>
+<p>To calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; mean_gt;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; sd_gt;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[N] <span class="dt">real</span> y_rep = normal_rng(alpha + beta * x, sigma);</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>    mean_gt = mean(y_rep) &gt; mean(y);</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>    sd_gt = sd(y_rep) &gt; sd(y);</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The indicator variable <code>mean_gt</code> will have value 1 if the mean of the simulated data <code>y_rep</code> is greater than or equal to the mean of he original data <code>y</code>. Because the values of <code>y_rep</code> are not needed for the posterior predictive checks, the program saves output space by using a local variable for <code>y_rep</code>. The statistics <code>mean(u)</code> and <code>sd(y)</code> could also be computed in the transformed data block and saved.</p>
+<p>For the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><a href="./img/ppc-pvalue-nb-pois-mean.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="Histogram of means of replicated data sets; vertical red line at mean of original data."><img src="./img/ppc-pvalue-nb-pois-mean.jpg" class="img-fluid figure-img" style="width:50.0%" alt="Histogram of means of replicated data sets; vertical red line at mean of original data."></a></p>
+<figcaption>Histogram of means of replicated data sets; vertical red line at mean of original data.</figcaption>
+</figure>
+</div>
+<p>The p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior <span class="math inline">\(p\)</span>-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable <code>mean_gt</code>.</p>
+<p>The standard deviation statistic tells a different story.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><a href="./img/ppc-pvalue-nb-pois-sd.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-4" title="Scatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data."><img src="./img/ppc-pvalue-nb-pois-sd.jpg" class="img-fluid figure-img" style="width:50.0%" alt="Scatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data."></a></p>
+<figcaption>Scatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.</figcaption>
+</figure>
+</div>
+<p>Here, the original data has much higher standard deviation than any of the replicated data sets. The resulting <span class="math inline">\(p\)</span>-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small <span class="math inline">\(p\)</span>-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.</p>
+<section id="which-statistics-to-test" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="which-statistics-to-test">Which statistics to test?</h3>
+<p>Any statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.</p>
+<p>Despite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.</p>
+</section>
+</section>
+<section id="prior-predictive-checks" class="level2">
+<h2 class="anchored" data-anchor-id="prior-predictive-checks">Prior predictive checks</h2>
+<p>Prior predictive checks generate data according to the prior in order to asses whether a prior is appropriate <span class="citation" data-cites="GabryEtAl:2019">(<a href="#ref-GabryEtAl:2019" role="doc-biblioref">Gabry et al. 2019</a>)</span>. A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, <span class="math display">\[
+y^{\textrm{sim}} \sim p(y).
+\]</span> The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.</p>
+<p>This is easy to carry out mechanically by simulating parameters <span class="math display">\[
+\theta^{\textrm{sim}} \sim p(\theta)
+\]</span> according to the priors, then simulating data <span class="math display">\[
+y^{\textrm{sim}} \sim p(y \mid \theta^{\textrm{sim}})
+\]</span> according to the data model given the simulated parameters. The result is a simulation from the joint distribution, <span class="math display">\[
+(y^{\textrm{sim}}, \theta^{\textrm{sim}}) \sim p(y, \theta)
+\]</span> and thus <span class="math display">\[
+y^{\textrm{sim}} \sim p(y)
+\]</span> is a simulation from the prior predictive distribution.</p>
+<section id="coding-prior-predictive-checks-in-stan" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="coding-prior-predictive-checks-in-stan">Coding prior predictive checks in Stan</h3>
+<p>A prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.</p>
+<p>Prior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha = normal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta = normal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y_sim = poisson_log_rng(alpha + beta * x);</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Running this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.</p>
+</section>
+</section>
+<section id="example-of-prior-predictive-checks" class="level2">
+<h2 class="anchored" data-anchor-id="example-of-prior-predictive-checks">Example of prior predictive checks</h2>
+<p>Suppose we have a model for a football (aka soccer) league where there are <span class="math inline">\(J\)</span> teams. Each team has a scoring rate <span class="math inline">\(\lambda_j\)</span> and in each game will be assumed to score <span class="math inline">\(\textrm{poisson}(\lambda_j)\)</span> points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed <span class="math display">\[
+\lambda_j \sim \textrm{gamma}(\epsilon_1, \epsilon_2).
+\]</span> This is not just a manufactured example; <em>The BUGS Book</em> recommends setting <span class="math inline">\(\epsilon = (0.5, 0.00001)\)</span>, which corresponds to a Jeffreys prior for a Poisson rate parameter prior <span class="citation" data-cites="LunnEtAl:2012">(<a href="#ref-LunnEtAl:2012" role="doc-biblioref">Lunn et al. 2012, 85</a>)</span>.</p>
+<p>Suppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; epsilon;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J, J] <span class="dt">int</span> y;</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) lambda[j] = gamma_rng(epsilon[<span class="dv">1</span>], epsilon[<span class="dv">2</span>]);</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are <span class="math inline">\(\binom{J}{2} = \frac{J \cdot (J - 1)}{2}.\)</span></p>
+<p>Using the <span class="math inline">\(\textrm{gamma}(0.5, 0.00001)\)</span> reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, <span class="math inline">\(y^{(1:20)}_{1, 2}\)</span>.</p>
+<pre><code>2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987
+7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66</code></pre>
+<p>That’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this <span class="math inline">\(\textrm{gamma}(0.5, 0.00001)\)</span> prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.</p>
+<p>The posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values <span class="citation" data-cites="Gelman:2006">(<a href="#ref-Gelman:2006" role="doc-biblioref">A. Gelman 2006</a>)</span>.</p>
+<p>Just as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.</p>
+<p>In this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see <span class="citation" data-cites="GabryEtAl:2019">(<a href="#ref-GabryEtAl:2019" role="doc-biblioref">Gabry et al. 2019</a>)</span>.</p>
+<p>Prior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.</p>
+</section>
+<section id="mixed-replication" class="level2">
+<h2 class="anchored" data-anchor-id="mixed-replication">Mixed predictive replication for hierarchical models</h2>
+<p><span class="citation" data-cites="GelmanEtAl:1996">Andrew Gelman, Meng, and Stern (<a href="#ref-GelmanEtAl:1996" role="doc-biblioref">1996</a>)</span> discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.</p>
+<p>For example, consider a simple varying intercept logistic regression, with intercepts <span class="math inline">\(\alpha_k\)</span> for <span class="math inline">\(k \in 1:K\)</span>. Each data item <span class="math inline">\(y_n \in \{ 0, 1 \}\)</span> is assumed to correspond to group <span class="math inline">\(kk_n \in 1:K.\)</span> The data model is thus <span class="math display">\[
+y_n \sim \textrm{bernoulli}(\textrm{logit}^{-1}(\alpha_{kk[n]})).
+\]</span> The varying intercepts have a hierarchical normal prior, <span class="math display">\[
+\alpha_k \sim \textrm{normal}(\mu, \sigma).
+\]</span> The hyperparameters are themselves given weakly informative priors, <span class="math display">\[\begin{eqnarray*}
+\mu &amp; \sim &amp; \textrm{normal}(0, 2)
+\\[4pt]
+\sigma &amp; \sim &amp; \textrm{lognormal}(0, 1).
+\end{eqnarray*}\]</span></p>
+<p>Like in a posterior predictive check, the hyperparameters <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\sigma\)</span> are drawn from the posterior, <span class="math display">\[
+\mu^{(m)}, \sigma^{(m)} \sim p(\mu, \sigma \mid y)
+\]</span> Like in a prior predictive check, replicated values of <span class="math inline">\(\alpha\)</span> are drawn from the hyperparameters, <span class="math display">\[
+\alpha^{\textrm{rep}(m)}_k \sim \textrm{normal}(\alpha_k \mid
+\mu^{(m)}, \sigma^{(m)}).
+\]</span> The data items are then each replicated using the replicated intercepts, <span class="math display">\[
+y^{\textrm{rep}(m)}_n \sim
+\textrm{bernoulli}
+  (\textrm{logit}^{-1}(\alpha^{\textrm{rep}(m)}_{kk[n]})).
+\]</span> Thus the <span class="math inline">\(y^{\textrm{rep}(m)}\)</span> can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original <span class="math inline">\(K\)</span> groups.</p>
+<p>In Stan, mixed predictive replications <span class="math inline">\(y^{\textrm{rep}(m)}\)</span> can be programmed directly.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">offset</span>=mu, <span class="dt">multiplier</span>=sigma&gt;[K] alpha;</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);               <span class="co">// hyperprior</span></span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(mu, sigma);       <span class="co">// hierarchical prior</span></span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(alpha[kk]);  <span class="co">// data model</span></span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb8-19"><a href="#cb8-19" aria-hidden="true" tabindex="-1"></a>  <span class="co">// alpha replicated;  mu and sigma not replicated</span></span>
+<span id="cb8-20"><a href="#cb8-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">real</span> alpha_rep</span>
+<span id="cb8-21"><a href="#cb8-21" aria-hidden="true" tabindex="-1"></a>    = normal_rng(rep_vector(mu, K), sigma);</span>
+<span id="cb8-22"><a href="#cb8-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y_rep</span>
+<span id="cb8-23"><a href="#cb8-23" aria-hidden="true" tabindex="-1"></a>    = bernoulli_logit_rng(alpha_rep[kk]);</span>
+<span id="cb8-24"><a href="#cb8-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="joint-model-representation" class="level2">
+<h2 class="anchored" data-anchor-id="joint-model-representation">Joint model representation</h2>
+<p>Following <span class="citation" data-cites="GelmanEtAl:1996">Andrew Gelman, Meng, and Stern (<a href="#ref-GelmanEtAl:1996" role="doc-biblioref">1996</a>)</span>, prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.</p>
+<section id="posterior-predictive-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="posterior-predictive-model">Posterior predictive model</h3>
+<p>For example, posterior predictive replication may be formulated using distribution notation as follows. <span class="math display">\[\begin{eqnarray*}
+\theta &amp; \sim &amp; p(\theta)
+\\[2pt]
+y &amp; \sim &amp; p(y \mid \theta)
+\\[2pt]
+y^{\textrm{rep}} &amp; \sim &amp; p(y \mid \theta)
+\end{eqnarray*}\]</span> The heavily overloaded distribution notation is meant to indicate that both <span class="math inline">\(y\)</span> and <span class="math inline">\(y^{\textrm{rep}}\)</span> are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities <span class="math inline">\(p_{Y^{\textrm{rep}} \mid
+\Theta}\)</span> and <span class="math inline">\(p_{Y \mid \Theta}\)</span> are the same.</p>
+<p>The joint density is <span class="math display">\[
+p(\theta, y, y^{\textrm{rep}})
+= p(\theta) \cdot p(y \mid \theta) \cdot p(y^{\textrm{rep}} \mid \theta).
+\]</span> This again is assuming that the two distributions for <span class="math inline">\(y\)</span> and <span class="math inline">\(y^{\textrm{rep}}\)</span> are identical.</p>
+<p>The variable <span class="math inline">\(y\)</span> is observed, with the predictive simulation <span class="math inline">\(y^{\textrm{rep}}\)</span> and parameter vector <span class="math inline">\(\theta\)</span> not observed. The posterior is <span class="math inline">\(p(y^{\textrm{rep}}, \theta \mid y)\)</span>. Given draws from the posterior, the posterior predictive simulations <span class="math inline">\(y^{\textrm{rep}}\)</span> are retained.</p>
+</section>
+<section id="prior-predictive-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="prior-predictive-model">Prior predictive model</h3>
+<p>The prior predictive model simply drops the data component of the posterior predictive model. <span class="math display">\[\begin{eqnarray*}
+\theta &amp; \sim &amp; p(\theta)
+\\[2pt]
+y^{\textrm{rep}} &amp; \sim &amp; p(y \mid \theta)
+\end{eqnarray*}\]</span> This corresponds to the joint density <span class="math display">\[
+p(\theta, y^{\textrm{rep}}) = p(\theta) \cdot p(y^{\textrm{rep}} \mid
+\theta).
+\]</span></p>
+<p>It is typically straightforward to draw <span class="math inline">\(\theta\)</span> from the prior and <span class="math inline">\(y^{\textrm{rep}}\)</span> from the data model given <span class="math inline">\(\theta\)</span> efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.</p>
+</section>
+<section id="mixed-replication-for-hierarchical-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mixed-replication-for-hierarchical-models">Mixed replication for hierarchical models</h3>
+<p>The mixed replication corresponds to the model <span class="math display">\[\begin{eqnarray*}
+\phi &amp; \sim &amp; p(\phi)
+\\[2pt]
+\alpha &amp; \sim &amp; p(\alpha \mid \phi)
+\\[2pt]
+y &amp; \sim &amp; p(y \mid \alpha)
+\\[2pt]
+\alpha^{\textrm{rep}} &amp; \sim &amp; p(\alpha \mid \phi)
+\\[2pt]
+y^{\textrm{rep}} &amp; \sim &amp; p(y \mid \phi)
+\end{eqnarray*}\]</span> The notation here is meant to indicate that <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\alpha^{\textrm{rep}}\)</span> have identical distributions, as do <span class="math inline">\(y\)</span> and <span class="math inline">\(y^{\textrm{rep}}\)</span>.</p>
+<p>This corresponds to a joint model <span class="math display">\[
+p(\phi, \alpha, \alpha^{\textrm{rep}}, y, y^{\textrm{rep}})
+=
+p(\phi)
+\cdot p(\alpha \mid \phi)
+\cdot p(y \mid \alpha)
+\cdot p(\alpha^{\textrm{rep}} \mid \phi)
+\cdot p(y^{\textrm{rep}} \mid \alpha^{\textrm{rep}}),
+\]</span> where <span class="math inline">\(y\)</span> is the only observed variable, <span class="math inline">\(\alpha\)</span> contains the lower-level parameters and <span class="math inline">\(\phi\)</span> the hyperparameters. Note that <span class="math inline">\(\phi\)</span> is not replicated and instead appears in the distribution for both <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\alpha^{\textrm{rep}}\)</span>.</p>
+<p>The posterior is <span class="math inline">\(p(\phi, \alpha, \alpha^{\textrm{rep}},
+y^{\textrm{rep}} \mid y)\)</span>. From posterior draws, the posterior predictive simulations <span class="math inline">\(y^{\textrm{rep}}\)</span> are kept.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-BayarriBerger:2000" class="csl-entry" role="listitem">
+Bayarri, MJ, and James O Berger. 2000. <span>“P Values for Composite Null Models.”</span> <em>Journal of the American Statistical Association</em> 95 (452): 1127–42.
+</div>
+<div id="ref-GabryEtAl:2019" class="csl-entry" role="listitem">
+Gabry, Jonah, Daniel Simpson, Aki Vehtari, Michael Betancourt, and Andrew Gelman. 2019. <span>“Visualization in <span>B</span>ayesian Workflow.”</span> <em>Journal of the Royal Statistical Society: Series A (Statistics in Society)</em> 182 (2): 389–402.
+</div>
+<div id="ref-Gelman:2006" class="csl-entry" role="listitem">
+Gelman, A. 2006. <span>“Prior Distributions for Variance Parameters in Hierarchical Models.”</span> <em>Bayesian Analysis</em> 1 (3): 515–34.
+</div>
+<div id="ref-GelmanEtAl:1996" class="csl-entry" role="listitem">
+Gelman, Andrew, Xiao-Li Meng, and Hal Stern. 1996. <span>“Posterior Predictive Assessment of Model Fitness via Realized Discrepancies.”</span> <em>Statistica Sinica</em>, 733–60.
+</div>
+<div id="ref-LunnEtAl:2012" class="csl-entry" role="listitem">
+Lunn, David, Christopher Jackson, Nicky Best, Andrew Thomas, and David Spiegelhalter. 2012. <em>The <span>BUGS</span> Book: A Practical Introduction to <span>B</span>ayesian Analysis</em>. CRC Press/Chapman &amp; Hall.
+</div>
+<div id="ref-Rubin:1984" class="csl-entry" role="listitem">
+Rubin, Donald B. 1984. <span>“Bayesianly Justifiable and Relevant Frequency Calculations for the Applied Statistician.”</span> <em>The Annals of Statistics</em>, 1151–72.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/simulation-based-calibration.html" class="pagination-link" aria-label="Simulation-Based Calibration Checking">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Simulation-Based Calibration Checking</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/cross-validation.html" class="pagination-link" aria-label="Held-Out Evaluation and Cross-Validation">
+        <span class="nav-page-text">Held-Out Evaluation and Cross-Validation</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-predictive-checks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+<script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","descPosition":"bottom","loop":false,"openEffect":"zoom","selector":".lightbox"});
+(function() {
+  let previousOnload = window.onload;
+  window.onload = () => {
+    if (previousOnload) {
+      previousOnload();
+    }
+    lightboxQuarto.on('slide_before_load', (data) => {
+      const { slideIndex, slideNode, slideConfig, player, trigger } = data;
+      const href = trigger.getAttribute('href');
+      if (href !== null) {
+        const imgEl = window.document.querySelector(`a[href="${href}"] img`);
+        if (imgEl !== null) {
+          const srcAttr = imgEl.getAttribute("src");
+          if (srcAttr && srcAttr.startsWith("data:")) {
+            slideConfig.href = srcAttr;
+          }
+        }
+      } 
+    });
+  
+    lightboxQuarto.on('slide_after_load', (data) => {
+      const { slideIndex, slideNode, slideConfig, player, trigger } = data;
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(slideNode);
+      }
+    });
+  
+  };
+  
+})();
+          </script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/poststratification.html b/docs/2_39/stan-users-guide/poststratification.html
new file mode 100644
index 000000000..179eaca67
--- /dev/null
+++ b/docs/2_39/stan-users-guide/poststratification.html
@@ -0,0 +1,1511 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Poststratification</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/decision-analysis.html" rel="next">
+<link href="../stan-users-guide/cross-validation.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Poststratification">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/poststratification.html">Poststratification</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#poststratification" id="toc-poststratification" class="nav-link active" data-scroll-target="#poststratification">Poststratification</a>
+  <ul class="collapse">
+  <li><a href="#some-examples" id="toc-some-examples" class="nav-link" data-scroll-target="#some-examples">Some examples</a>
+  <ul class="collapse">
+  <li><a href="#earth-science" id="toc-earth-science" class="nav-link" data-scroll-target="#earth-science">Earth science</a></li>
+  <li><a href="#polling" id="toc-polling" class="nav-link" data-scroll-target="#polling">Polling</a></li>
+  </ul></li>
+  <li><a href="#bayesian-poststratification" id="toc-bayesian-poststratification" class="nav-link" data-scroll-target="#bayesian-poststratification">Bayesian poststratification</a></li>
+  <li><a href="#poststratification-in-stan" id="toc-poststratification-in-stan" class="nav-link" data-scroll-target="#poststratification-in-stan">Poststratification in Stan</a></li>
+  <li><a href="#regression-and-poststratification" id="toc-regression-and-poststratification" class="nav-link" data-scroll-target="#regression-and-poststratification">Regression and poststratification</a></li>
+  <li><a href="#multilevel-regression-and-poststratification" id="toc-multilevel-regression-and-poststratification" class="nav-link" data-scroll-target="#multilevel-regression-and-poststratification">Multilevel regression and poststratification</a>
+  <ul class="collapse">
+  <li><a href="#dealing-with-small-partitions-and-non-identifiability" id="toc-dealing-with-small-partitions-and-non-identifiability" class="nav-link" data-scroll-target="#dealing-with-small-partitions-and-non-identifiability">Dealing with small partitions and non-identifiability</a></li>
+  </ul></li>
+  <li><a href="#coding-mrp-in-stan" id="toc-coding-mrp-in-stan" class="nav-link" data-scroll-target="#coding-mrp-in-stan">Coding MRP in Stan</a>
+  <ul class="collapse">
+  <li><a href="#binomial-coding" id="toc-binomial-coding" class="nav-link" data-scroll-target="#binomial-coding">Binomial coding</a></li>
+  <li><a href="#coding-binary-groups" id="toc-coding-binary-groups" class="nav-link" data-scroll-target="#coding-binary-groups">Coding binary groups</a></li>
+  </ul></li>
+  <li><a href="#adding-group-level-predictors" id="toc-adding-group-level-predictors" class="nav-link" data-scroll-target="#adding-group-level-predictors">Adding group-level predictors</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/poststratification.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/poststratification.html">Poststratification</a></li></ol></nav></header>
+
+
+
+
+<section id="poststratification" class="level1">
+<h1>Poststratification</h1>
+<p>Stratification is a technique developed for survey sampling in which a population is partitioned into subgroups (i.e., stratified) and each group (i.e., stratum) is sampled independently. If the subgroups are more homogeneous (i.e., lower variance) than the population as a whole, this can reduce variance in the estimate of a quantity of interest at the population level.</p>
+<p>Poststratification is a technique for adjusting a non-representative sample (i.e., a convenience sample or other observational data) for which there are demographic predictors characterizing the strata. It is carried out after a model is fit to the observed data, hence the name <em>post</em>stratification <span class="citation" data-cites="Little:1993">(<a href="#ref-Little:1993" role="doc-biblioref">Little 1993</a>)</span>. Poststratification can be fruitfully combined with regression modeling (or more general parametric modeling), which provides estimates based on combinations of predictors (or general parameters) rather than raw counts in each stratum. Multilevel modeling is useful in determining how much partial pooling to apply in the regressions, leading to the popularity of the combination of multilevel regression and poststratification (MRP) <span class="citation" data-cites="Park:2004">(<a href="#ref-Park:2004" role="doc-biblioref">Park, Gelman, and Bafumi 2004</a>)</span>.</p>
+<section id="some-examples" class="level2">
+<h2 class="anchored" data-anchor-id="some-examples">Some examples</h2>
+<section id="earth-science" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="earth-science">Earth science</h3>
+<p>Stratification and poststratification can be applied to many applications beyond survey sampling <span class="citation" data-cites="Kennedy:2019">(<a href="#ref-Kennedy:2019" role="doc-biblioref">Kennedy and Gelman 2019</a>)</span>. For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid <span class="citation" data-cites="Paustian:1997">(<a href="#ref-Paustian:1997" role="doc-biblioref">Paustian et al. 1997</a>)</span>. Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both <span class="citation" data-cites="Field:1998">(<a href="#ref-Field:1998" role="doc-biblioref">Field et al. 1998</a>)</span>.</p>
+</section>
+<section id="polling" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="polling">Polling</h3>
+<p>Suppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.</p>
+<p>The proportion of support in the poll among students in each group provides a simple maximum likelihood estimate <span class="math inline">\(\theta^* = (0.25, 0.5,
+0.8)\)</span> of support in each group for a simple Bernoulli model where student <span class="math inline">\(n\)</span>’s vote is modeled as <span class="math display">\[
+y_n \sim \textrm{bernoulli}(\theta_{jj[n]}),
+\]</span> where <span class="math inline">\(jj[n] \in 1:3\)</span> is the subgroup to which the <span class="math inline">\(n\)</span>-th student belongs.</p>
+<p>An estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting <span class="math inline">\(N = (20\,000,\, 5\,000,\, 2\,000)\)</span> be the subgroup sizes, the poststratified estimate of support in the population <span class="math inline">\(\phi^*\)</span> is estimated by <span class="math display">\[
+\phi^*
+= \frac{\displaystyle \sum_{j = 1}^3 \theta_j^* \cdot N_j}
+       {\displaystyle \sum_{j = 1}^3 N_j}.
+\]</span> Plugging in our estimates and population counts yields <span class="math display">\[\begin{eqnarray*}
+\phi*
+&amp; = &amp; \frac{0.25 \cdot 20\,000 + 0.4 \cdot 5\,000 + 0.8 \cdot 2\,000}
+           {20\,000 + 5\,000 + 2\,000}
+\\[4pt] &amp; = &amp; \frac{8\,600}{27\,000}
+\\[4pt] &amp; \approx &amp; 0.32.
+\end{eqnarray*}\]</span></p>
+</section>
+</section>
+<section id="bayesian-poststratification" class="level2">
+<h2 class="anchored" data-anchor-id="bayesian-poststratification">Bayesian poststratification</h2>
+<p>Considering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, <span class="math display">\[
+y_n \sim \textrm{bernoulli}(\theta_{jj[n]}),
+\]</span> where <span class="math inline">\(jj[n] \in 1:J\)</span> is the group to which item <span class="math inline">\(n\)</span> belongs and <span class="math inline">\(\theta_j\)</span> is the proportion of support in group <span class="math inline">\(j\)</span>.</p>
+<p>This can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting <span class="math inline">\(A_j\)</span> be the number of respondents in group <span class="math inline">\(j\)</span> and <span class="math inline">\(a_j\)</span> be the number of positive responses in group <span class="math inline">\(j\)</span>, the data model may be reduced to the form <span class="math display">\[
+a_j \sim \textrm{binomial}(A_j, \theta_j).
+\]</span> A simple uniform prior on the proportion of support in each group completes the model, <span class="math display">\[
+\theta_j \sim \textrm{beta}(1, 1).
+\]</span> A more informative prior could be used if there is prior information available about support among the student body.</p>
+<p>Using sampling, draws <span class="math inline">\(\theta^{(m)} \sim p(\theta \mid y)\)</span> from the posterior may be combined with the population sizes <span class="math inline">\(N\)</span> to estimate <span class="math inline">\(\phi\)</span>, the proportion of support in the population, <span class="math display">\[
+\phi^{(m)}
+= \frac{\displaystyle \sum_{j = 1}^J \theta_j^{(m)} \cdot N_j}
+       {\displaystyle \sum_{j = 1}^J N_j}.
+\]</span> The posterior draws for <span class="math inline">\(\phi^{(m)}\)</span> characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the <span class="math inline">\(\phi^{(m)}\)</span>), posterior intervals (quantiles of the <span class="math inline">\(\phi^{(m)}\)</span>), or to plot a histogram.</p>
+</section>
+<section id="poststratification-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="poststratification-in-stan">Poststratification in Stan</h2>
+<p>The maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where <span class="math inline">\(A_j\)</span> is the number of voters from group <span class="math inline">\(j\)</span>, <span class="math inline">\(a_j\)</span> is the number of positive responses from group <span class="math inline">\(j\)</span>, and <span class="math inline">\(N_j\)</span> is the size of group <span class="math inline">\(j\)</span> in the population.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; A; </span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; a;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[J] N;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[J] theta;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  a ~ binomial(A, theta);</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {t</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; phi = dot(N, theta) / sum(N);</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the <span class="math inline">\(J\)</span> terms. The prior is implicitly uniform on <span class="math inline">\((0, 1),\)</span> the support of <span class="math inline">\(\theta.\)</span> The summation is computed using a dot product and the sum function, which is why <code>N</code> was declared as a vector rather than as an array of integers.</p>
+</section>
+<section id="regression-and-poststratification" class="level2">
+<h2 class="anchored" data-anchor-id="regression-and-poststratification">Regression and poststratification</h2>
+<p>In applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.</p>
+<p>The problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to <span class="math inline">\(5 \cdot 2 \cdot 5 \cdot
+50 = 2000\)</span> strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter <span class="math inline">\(\theta_j\)</span> for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.</p>
+<p>A practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only <span class="math inline">\(4 + 2 + 5 + 50 =
+61\)</span> regression coefficients to estimate. Now suppose that item <span class="math inline">\(n\)</span> has demographic features <span class="math inline">\(\textrm{age}_n \in 1:5\)</span>, <span class="math inline">\(\textrm{sex}_n \in 1:2\)</span>, <span class="math inline">\(\textrm{income}_n \in 1:5,\)</span> and <span class="math inline">\(\textrm{state}_n \in 1:50\)</span>. A logistic regression may be formulated as <span class="math display">\[
+y_n \sim
+\textrm{bernoulli}(\textrm{logit}^{-1}(
+\alpha + \beta_{\textrm{age}[n]}
++ \gamma_{\textrm{sex}[n]}
++ \delta_{\textrm{income}[n]}
++ \epsilon_{\textrm{state}[n]}
+)),
+\]</span> where <span class="math inline">\(\textrm{age}[n]\)</span> is the age of the <span class="math inline">\(n\)</span>-th respondent, <span class="math inline">\(\textrm{sex}[n]\)</span> is their sex, <span class="math inline">\(\textrm{income}[n]\)</span> their income and <span class="math inline">\(\textrm{state}[n]\)</span> their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.</p>
+<p>To poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as <span class="math display">\[
+\sum_{i = 1}^5 \sum_{j = 1}^2 \sum_{k = 1}^5 \sum_{m = 1}^{50}
+\textrm{logit}^{-1}(\alpha + \beta_i + \gamma_j + \delta_k + \eta_m)
+\cdot \textrm{pop}_{i, j, k, m},
+\]</span> where <span class="math inline">\(\textrm{pop}_{i, j, k, m}\)</span> is the size of the subpopulation with age <span class="math inline">\(i\)</span>, sex <span class="math inline">\(j\)</span>, income level <span class="math inline">\(k\)</span>, and state of residence <span class="math inline">\(m\)</span>.</p>
+<p>As formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes <span class="math inline">\(y\)</span> given predictors <span class="math inline">\(x\)</span>.</p>
+</section>
+<section id="multilevel-regression-and-poststratification" class="level2">
+<h2 class="anchored" data-anchor-id="multilevel-regression-and-poststratification">Multilevel regression and poststratification</h2>
+<p>With large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.</p>
+<p>The model introduced in the previous section had the data model <span class="math display">\[
+y_n \sim
+\textrm{bernoulli}(\textrm{logit}^{-1}(
+\alpha + \beta_{\textrm{age}[n]}
++ \gamma_{\textrm{sex}[n]}
++ \delta_{\textrm{income}[n]}
++ \epsilon_{\textrm{state}[n]}
+)).
+\]</span> The overall intercept can be given a broad fixed prior, <span class="math display">\[
+\alpha \sim \textrm{normal}(0, 5).
+\]</span> The other regression parameters can be given hierarchical priors, <span class="math display">\[\begin{eqnarray*}
+\beta_{1:4} &amp; \sim &amp; \textrm{normal}(0, \sigma^{\beta})
+\\[2pt]
+\gamma_{1:2} &amp; \sim &amp; \textrm{normal}(0, \sigma^{\gamma})
+\\[2pt]
+\delta_{1:5} &amp; \sim &amp; \textrm{normal}(0, \sigma^{\delta})
+\\[2pt]
+\epsilon_{1:50} &amp; \sim &amp; \textrm{normal}(0, \sigma^{\epsilon}).
+\end{eqnarray*}\]</span></p>
+<p>The hyperparameters for scale of variation within a group can be given simple standard hyperpriors, <span class="math display">\[
+\sigma^{\beta}, \sigma^{\gamma}, \sigma^{\delta}, \sigma^{\epsilon}
+\sim \textrm{normal}(0, 1).
+\]</span> The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).</p>
+<section id="dealing-with-small-partitions-and-non-identifiability" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="dealing-with-small-partitions-and-non-identifiability">Dealing with small partitions and non-identifiability</h3>
+<p>The multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.</p>
+<p>This is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, <span class="math inline">\(\gamma_2 =
+-\gamma_1.\)</span></p>
+<p>To implement sum-to-zero constraints,</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K - <span class="dv">1</span>] alpha_raw;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">multiplier</span>=sigma_alpha&gt;[K] alpha</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>    = append_row(alpha_raw, -sum(alpha_raw));</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a><span class="co">// ...    </span></span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, sigma_alpha);</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This prior is hard to interpret in that there are <code>K</code> normal distributions, but only <code>K - 1</code> free parameters. An alternative is to put the prior only on <code>alpha_raw</code>, but that is also difficult to interpret.</p>
+<p>Soft constraints can be more computationally tractable. They are also simpler to implement.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">multiplier</span>=alpha&gt;[K] alpha;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, sigma_alpha);</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  sum(alpha) ~ normal(<span class="dv">0</span>, <span class="fl">0.001</span>);</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.</p>
+<p>Note that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept <span class="math inline">\(\alpha.\)</span> Even with a wide prior on the intercept, the hyperprior on <span class="math inline">\(\sigma^{\gamma}\)</span> may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.</p>
+<p>None of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.</p>
+</section>
+</section>
+<section id="coding-mrp-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="coding-mrp-in-stan">Coding MRP in Stan</h2>
+<p>Multilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the <code>multiplier</code> transform is used on each of the parameters. The combination of</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="dt">multiplier</span>=s&gt;[K] a;</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>a ~ normal(<span class="dv">0</span>, s);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>implements a non-centered parameterization for <code>a</code>; a centered parameterization would drop the <code>multiplier</code> specification. The prior scale <code>s</code> is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the <code>multiplier</code>.</p>
+<p>Here is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable <code>P</code>.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">4</span>&gt; age;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">5</span>&gt; income;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">50</span>&gt; state;</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">4</span>, <span class="dv">5</span>, <span class="dv">50</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; P;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">multiplier</span>=sigma_beta&gt;[<span class="dv">4</span>] beta;</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_gamma;</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">multiplier</span>=sigma_gamma&gt;[<span class="dv">5</span>] gamma;</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_delta;</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="dt">multiplier</span>=sigma_delta&gt;[<span class="dv">50</span>] delta;</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>  alpha ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, sigma_beta);</span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>  gamma ~ normal(<span class="dv">0</span>, sigma_gamma);</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>  delta ~ normal(<span class="dv">0</span>, sigma_delta);</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb5-27"><a href="#cb5-27" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> expect_pos = <span class="dv">0</span>;</span>
+<span id="cb5-28"><a href="#cb5-28" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> total = <span class="dv">0</span>;</span>
+<span id="cb5-29"><a href="#cb5-29" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (b <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">4</span>) {</span>
+<span id="cb5-30"><a href="#cb5-30" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (c <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">5</span>) {</span>
+<span id="cb5-31"><a href="#cb5-31" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">50</span>) {</span>
+<span id="cb5-32"><a href="#cb5-32" aria-hidden="true" tabindex="-1"></a>        total += P[b, c, d];</span>
+<span id="cb5-33"><a href="#cb5-33" aria-hidden="true" tabindex="-1"></a>        expect_pos</span>
+<span id="cb5-34"><a href="#cb5-34" aria-hidden="true" tabindex="-1"></a>          += P[b, c, d]</span>
+<span id="cb5-35"><a href="#cb5-35" aria-hidden="true" tabindex="-1"></a>             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);</span>
+<span id="cb5-36"><a href="#cb5-36" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb5-37"><a href="#cb5-37" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb5-38"><a href="#cb5-38" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-39"><a href="#cb5-39" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; phi = expect_pos / total;</span>
+<span id="cb5-40"><a href="#cb5-40" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Unlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).</p>
+<section id="binomial-coding" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="binomial-coding">Binomial coding</h3>
+<p>In some cases, it can be more efficient to break the data down by group. Suppose there are <span class="math inline">\(4 \times 5 \times 2 \times 50 = 2000\)</span> groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; G;</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[G] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">4</span>&gt; age;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[G] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">5</span>&gt; income;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[G] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">50</span>&gt; state;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[G] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; pos_votes;</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[G] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; total_votes;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Finally, the data model is converted to binomial.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a>pos_votes ~ binomial_logit(total_votes,</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>                           alpha + beta[age] + ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The predictors look the same because of the way the <code>age</code> and other data items are coded.</p>
+</section>
+<section id="coding-binary-groups" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="coding-binary-groups">Coding binary groups</h3>
+<p>In this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the <code>N</code> data points.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">2</span>&gt; sex;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then add a single regression coefficient as a parameter,</p>
+<pre><code>  real epsilon;</code></pre>
+<p>In the log odds calculation, introduce a new term</p>
+<pre><code>[epsilon, -epsilon][sex]';</code></pre>
+<p>That is, the data model will now look like</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>                      + [epsilon, -epsilon][sex]');</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For data point <code>n</code>, the expression <code>[epsilon, -epsilon][sex]</code> takes on value <code>[epsilon, -epsilon][sex][n]</code>, which with Stan’s multi-indexing reduces to <code>[epsilon, -epsilon][sex[n]]</code>. This term evaluates to <code>epsilon</code> if <code>sex[n]</code> is 1 and to <code>-epsilon</code> if <code>sex[n]</code> is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The <code>'</code> at the end transposes <code>[epsilon, -epsilon][sex]</code> which is a <code>row_vector</code> into a <code>vector</code> that can be added to the other variables.</p>
+<p>Finally, a prior is needed for the coefficient in the model block,</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a>epsilon ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As with other priors in multilevel models, the posterior for <code>epsilon</code> should be investigated to make sure it is not unrealistically wide.</p>
+</section>
+</section>
+<section id="adding-group-level-predictors" class="level2">
+<h2 class="anchored" data-anchor-id="adding-group-level-predictors">Adding group-level predictors</h2>
+<p>If there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">50</span>] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; income;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>then a regression coefficient <code>psi</code> can be added for the effect of average state income,</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> psi;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with a fixed prior,</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a>psi ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This prior assumes the <code>income</code> predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a>y ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>                    + income[state] * psi);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>And finally, the formula in the generated quantities block is also updated,</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a>expect_pos</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  += P[b, c, d]</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]</span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>             + income[d] * psi);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here <code>d</code> is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Field:1998" class="csl-entry" role="listitem">
+Field, Christopher B, Michael J Behrenfeld, James T Randerson, and Paul Falkowski. 1998. <span>“Primary Production of the Biosphere: Integrating Terrestrial and Oceanic Components.”</span> <em>Science</em> 281 (5374): 237–40.
+</div>
+<div id="ref-Kennedy:2019" class="csl-entry" role="listitem">
+Kennedy, Lauren, and Andrew Gelman. 2019. <span>“Know Your Population and Know Your Model: Using Model-Based Regression and Poststratification to Generalize Findings Beyond the Observed Sample.”</span> <em>arXiv</em>, no. 1906.11323.
+</div>
+<div id="ref-Little:1993" class="csl-entry" role="listitem">
+Little, Roderick JA. 1993. <span>“Post-Stratification: A Modeler’s Perspective.”</span> <em>Journal of the American Statistical Association</em> 88 (423): 1001–12.
+</div>
+<div id="ref-Park:2004" class="csl-entry" role="listitem">
+Park, David K, Andrew Gelman, and Joseph Bafumi. 2004. <span>“Bayesian Multilevel Estimation with Poststratification: State-Level Estimates from National Polls.”</span> <em>Political Analysis</em> 12 (4): 375–85.
+</div>
+<div id="ref-Paustian:1997" class="csl-entry" role="listitem">
+Paustian, Keith, Elissa Levine, Wilfred M Post, and Irene M Ryzhova. 1997. <span>“The Use of Models to Integrate Information and Understanding of Soil <span>C</span> at the Regional Scale.”</span> <em>Geoderma</em> 79 (1-4): 227–60.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/cross-validation.html" class="pagination-link" aria-label="Held-Out Evaluation and Cross-Validation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Held-Out Evaluation and Cross-Validation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/decision-analysis.html" class="pagination-link" aria-label="Decision Analysis">
+        <span class="nav-page-text">Decision Analysis</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/poststratification.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/problematic-posteriors.html b/docs/2_39/stan-users-guide/problematic-posteriors.html
new file mode 100644
index 000000000..651f38493
--- /dev/null
+++ b/docs/2_39/stan-users-guide/problematic-posteriors.html
@@ -0,0 +1,1619 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Problematic Posteriors</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/reparameterization.html" rel="next">
+<link href="../stan-users-guide/proportionality-constants.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Problematic Posteriors">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/non-identified.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/problematic-posteriors.html">Problematic Posteriors</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#problematic-posteriors.chapter" id="toc-problematic-posteriors.chapter" class="nav-link active" data-scroll-target="#problematic-posteriors.chapter">Problematic Posteriors</a>
+  <ul class="collapse">
+  <li><a href="#collinearity.section" id="toc-collinearity.section" class="nav-link" data-scroll-target="#collinearity.section">Collinearity of predictors in regressions</a>
+  <ul class="collapse">
+  <li><a href="#examples-of-collinearity" id="toc-examples-of-collinearity" class="nav-link" data-scroll-target="#examples-of-collinearity">Examples of collinearity</a></li>
+  <li><a href="#mitigating-the-invariances" id="toc-mitigating-the-invariances" class="nav-link" data-scroll-target="#mitigating-the-invariances">Mitigating the invariances</a></li>
+  </ul></li>
+  <li><a href="#label-switching-problematic.section" id="toc-label-switching-problematic.section" class="nav-link" data-scroll-target="#label-switching-problematic.section">Label switching in mixture models</a>
+  <ul class="collapse">
+  <li><a href="#mixture-models" id="toc-mixture-models" class="nav-link" data-scroll-target="#mixture-models">Mixture models</a></li>
+  <li><a href="#convergence-monitoring-and-effective-sample-size" id="toc-convergence-monitoring-and-effective-sample-size" class="nav-link" data-scroll-target="#convergence-monitoring-and-effective-sample-size">Convergence monitoring and effective sample size</a></li>
+  <li><a href="#some-inferences-are-invariant" id="toc-some-inferences-are-invariant" class="nav-link" data-scroll-target="#some-inferences-are-invariant">Some inferences are invariant</a></li>
+  <li><a href="#highly-multimodal-posteriors" id="toc-highly-multimodal-posteriors" class="nav-link" data-scroll-target="#highly-multimodal-posteriors">Highly multimodal posteriors</a></li>
+  <li><a href="#hacks-as-fixes" id="toc-hacks-as-fixes" class="nav-link" data-scroll-target="#hacks-as-fixes">Hacks as fixes</a></li>
+  </ul></li>
+  <li><a href="#component-collapsing-in-mixture-models" id="toc-component-collapsing-in-mixture-models" class="nav-link" data-scroll-target="#component-collapsing-in-mixture-models">Component collapsing in mixture models</a></li>
+  <li><a href="#posteriors-with-unbounded-densities" id="toc-posteriors-with-unbounded-densities" class="nav-link" data-scroll-target="#posteriors-with-unbounded-densities">Posteriors with unbounded densities</a>
+  <ul class="collapse">
+  <li><a href="#mixture-models-with-varying-scales" id="toc-mixture-models-with-varying-scales" class="nav-link" data-scroll-target="#mixture-models-with-varying-scales">Mixture models with varying scales</a></li>
+  <li><a href="#beta-binomial-models-with-skewed-data-and-weak-priors" id="toc-beta-binomial-models-with-skewed-data-and-weak-priors" class="nav-link" data-scroll-target="#beta-binomial-models-with-skewed-data-and-weak-priors">Beta-binomial models with skewed data and weak priors</a></li>
+  </ul></li>
+  <li><a href="#posteriors-with-unbounded-parameters" id="toc-posteriors-with-unbounded-parameters" class="nav-link" data-scroll-target="#posteriors-with-unbounded-parameters">Posteriors with unbounded parameters</a>
+  <ul class="collapse">
+  <li><a href="#separability-in-logistic-regression" id="toc-separability-in-logistic-regression" class="nav-link" data-scroll-target="#separability-in-logistic-regression">Separability in logistic regression</a></li>
+  </ul></li>
+  <li><a href="#uniform-posteriors" id="toc-uniform-posteriors" class="nav-link" data-scroll-target="#uniform-posteriors">Uniform posteriors</a></li>
+  <li><a href="#sampling-difficulties-with-problematic-priors" id="toc-sampling-difficulties-with-problematic-priors" class="nav-link" data-scroll-target="#sampling-difficulties-with-problematic-priors">Sampling difficulties with problematic priors</a>
+  <ul class="collapse">
+  <li><a href="#gibbs-sampling" id="toc-gibbs-sampling" class="nav-link" data-scroll-target="#gibbs-sampling">Gibbs sampling</a></li>
+  <li><a href="#hamiltonian-monte-carlo-sampling" id="toc-hamiltonian-monte-carlo-sampling" class="nav-link" data-scroll-target="#hamiltonian-monte-carlo-sampling">Hamiltonian Monte Carlo sampling</a></li>
+  <li><a href="#no-u-turn-sampling" id="toc-no-u-turn-sampling" class="nav-link" data-scroll-target="#no-u-turn-sampling">No-U-turn sampling</a></li>
+  <li><a href="#examples-fits-in-stan" id="toc-examples-fits-in-stan" class="nav-link" data-scroll-target="#examples-fits-in-stan">Examples: fits in Stan</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/problematic-posteriors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/problematic-posteriors.html">Problematic Posteriors</a></li></ol></nav></header>
+
+
+
+
+<section id="problematic-posteriors.chapter" class="level1">
+<h1>Problematic Posteriors</h1>
+<p>Mathematically speaking, with a proper posterior, one can do Bayesian inference and that’s that. There is not even a need to require a finite variance or even a finite mean—all that’s needed is a finite integral. Nevertheless, modeling is a tricky business and even experienced modelers sometimes code models that lead to improper priors. Furthermore, some posteriors are mathematically sound, but ill-behaved in practice. This chapter discusses issues in models that create problematic posterior inferences, either in general for Bayesian inference or in practice for Stan.</p>
+<section id="collinearity.section" class="level2">
+<h2 class="anchored" data-anchor-id="collinearity.section">Collinearity of predictors in regressions</h2>
+<p>This section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.</p>
+<section id="examples-of-collinearity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="examples-of-collinearity">Examples of collinearity</h3>
+<section id="redundant-intercepts" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="redundant-intercepts">Redundant intercepts</h4>
+<p>The first example of collinearity is an artificial example involving redundant intercept parameters.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>Suppose there are observations <span class="math inline">\(y_n\)</span> for <span class="math inline">\(n \in \{1,\dotsc,N\}\)</span>, two intercept parameters <span class="math inline">\(\lambda_1\)</span> and <span class="math inline">\(\lambda_2\)</span>, a scale parameter <span class="math inline">\(\sigma &gt; 0\)</span>, and the data model <span class="math display">\[
+y_n \sim \textsf{normal}(\lambda_1 + \lambda_2, \sigma).
+\]</span></p>
+<p>For any constant <span class="math inline">\(q\)</span>, the sampling density for <span class="math inline">\(y\)</span> does not change if we add <span class="math inline">\(q\)</span> to <span class="math inline">\(\lambda_1\)</span> and subtract it from <span class="math inline">\(\lambda_2\)</span>, i.e., <span class="math display">\[
+p(y \mid \lambda_1, \lambda_2,\sigma)
+=
+p(y \mid \lambda_1 + q, \lambda_2 - q, \sigma).
+\]</span></p>
+<p>The consequence is that an improper uniform prior <span class="math inline">\(p(\mu,\sigma)
+\propto 1\)</span> leads to an improper posterior. This impropriety arises because the neighborhoods around <span class="math inline">\(\lambda_1 + q, \lambda_2 - q\)</span> have the same mass no matter what <span class="math inline">\(q\)</span> is. Therefore, a sampler would need to spend as much time in the neighborhood of <span class="math inline">\(\lambda_1=1\,000\,000\,000\)</span> and <span class="math inline">\(\lambda_2=-1\,000\,000\,000\)</span> as it does in the neighborhood of <span class="math inline">\(\lambda_1=0\)</span> and <span class="math inline">\(\lambda_2=0\)</span>, and so on for ever more far-ranging values.</p>
+<p>The marginal posterior <span class="math inline">\(p(\lambda_1,\lambda_2 \mid y)\)</span> for this model is thus improper.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>The impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where <span class="math inline">\(\lambda_2 = \lambda_1 + c\)</span> for some constant <span class="math inline">\(c\)</span>.</p>
+<p>Contrast this model with a simple regression with a single intercept parameter <span class="math inline">\(\mu\)</span> and data model <span class="math display">\[
+y_n \sim \textsf{normal}(\mu,\sigma).
+\]</span> Even with an improper prior, the posterior is proper as long as there are at least two data points <span class="math inline">\(y_n\)</span> with distinct values.</p>
+</section>
+<section id="ability-and-difficulty-in-irt-models" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="ability-and-difficulty-in-irt-models">Ability and difficulty in IRT models</h4>
+<p>Consider an item-response theory model for students <span class="math inline">\(j \in 1{:}J\)</span> with abilities <span class="math inline">\(\alpha_j\)</span> and test items <span class="math inline">\(i \in 1{:}I\)</span> with difficulties <span class="math inline">\(\beta_i\)</span>. The observed data are an <span class="math inline">\(I \times J\)</span> array with entries <span class="math inline">\(y_{i, j} \in \{ 0, 1 \}\)</span> coded such that <span class="math inline">\(y_{i, j} = 1\)</span> indicates that student <span class="math inline">\(j\)</span> answered question <span class="math inline">\(i\)</span> correctly. The sampling distribution for the data is <span class="math display">\[
+y_{i, j} \sim \textsf{Bernoulli}(\operatorname{logit}^{-1}(\alpha_j - \beta_i)).
+\]</span></p>
+<p>For any constant <span class="math inline">\(c\)</span>, the probability of <span class="math inline">\(y\)</span> is unchanged by adding a constant <span class="math inline">\(c\)</span> to all the abilities and subtracting it from all the difficulties, i.e., <span class="math display">\[
+p(y \mid \alpha, \beta)
+=
+p(y \mid \alpha + c, \beta - c).
+\]</span></p>
+<p>This leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.</p>
+</section>
+<section id="general-collinear-regression-predictors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="general-collinear-regression-predictors">General collinear regression predictors</h4>
+<p>The general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model <span class="math display">\[
+y_n \sim \textsf{normal}(x_n \beta, \sigma)
+\]</span> for an <span class="math inline">\(N\)</span>-dimensional observation vector <span class="math inline">\(y\)</span>, an <span class="math inline">\(N \times K\)</span> predictor matrix <span class="math inline">\(x\)</span>, and a <span class="math inline">\(K\)</span>-dimensional coefficient vector <span class="math inline">\(\beta\)</span>.</p>
+<p>Now suppose that column <span class="math inline">\(k\)</span> of the predictor matrix is a multiple of column <span class="math inline">\(k'\)</span>, i.e., there is some constant <span class="math inline">\(c\)</span> such that <span class="math inline">\(x_{n,k} = c
+\, x_{n,k'}\)</span> for all <span class="math inline">\(n\)</span>. In this case, the coefficients <span class="math inline">\(\beta_k\)</span> and <span class="math inline">\(\beta_{k'}\)</span> can covary without changing the predictions, so that for any <span class="math inline">\(d \neq 0\)</span>, <span class="math display">\[
+p(y \mid \ldots, \beta_k, \dotsc, \beta_{k'}, \dotsc, \sigma)
+=
+p(y \mid \ldots, d \beta_k, \dotsc, \frac{d}{c} \, \beta_{k'}, \dotsc,
+\sigma).
+\]</span></p>
+<p>Even if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.</p>
+</section>
+<section id="multiplicative-issues-with-discrimination-in-irt" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="multiplicative-issues-with-discrimination-in-irt">Multiplicative issues with discrimination in IRT</h4>
+<p>Consider adding a discrimination parameter <span class="math inline">\(\delta_i\)</span> for each question in an IRT model, with data model <span class="math display">\[
+y_{i, j} \sim \textsf{Bernoulli}(\operatorname{logit}^{-1}(\delta_i(\alpha_j - \beta_i))).
+\]</span> For any constant <span class="math inline">\(c \neq 0\)</span>, multiplying <span class="math inline">\(\delta\)</span> by <span class="math inline">\(c\)</span> and dividing <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> by <span class="math inline">\(c\)</span> produces the same likelihood, <span class="math display">\[
+p(y \mid \delta,\alpha,\beta)
+= p(y \mid c \delta, \frac{1}{c}\alpha, \frac{1}{c}\beta).
+\]</span> If <span class="math inline">\(c &lt; 0\)</span>, this switches the signs of every component in <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(\beta\)</span>, and <span class="math inline">\(\delta\)</span> without changing the density.</p>
+</section>
+<section id="softmax-with-k-vs.-k-1-parameters" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="softmax-with-k-vs.-k-1-parameters">Softmax with <span class="math inline">\(K\)</span> vs.&nbsp;<span class="math inline">\(K-1\)</span> parameters</h4>
+<p>In order to parameterize a <span class="math inline">\(K\)</span>-simplex (i.e., a <span class="math inline">\(K\)</span>-vector with non-negative values that sum to one), only <span class="math inline">\(K - 1\)</span> parameters are necessary because the <span class="math inline">\(K\)</span>th is just one minus the sum of the first <span class="math inline">\(K
+- 1\)</span> parameters, so that if <span class="math inline">\(\theta\)</span> is a <span class="math inline">\(K\)</span>-simplex, <span class="math display">\[
+\theta_K = 1 - \sum_{k=1}^{K-1} \theta_k.
+\]</span></p>
+<p>The <code>softmax</code> function maps a <span class="math inline">\(K\)</span>-vector <span class="math inline">\(\alpha\)</span> of linear predictors to a <span class="math inline">\(K\)</span>-simplex <span class="math inline">\(\theta = \texttt{softmax}(\alpha)\)</span> by defining <span class="math display">\[
+\theta_k = \frac{\exp(\alpha_k)}{\sum_{k'=1}^K \exp(\alpha_{k'})}.
+\]</span></p>
+<p>The <code>softmax</code> function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters <span class="math inline">\(\alpha\)</span>. In particular, adding or subtracting a constant from each <span class="math inline">\(\alpha_k\)</span> produces the same simplex <span class="math inline">\(\theta\)</span>.</p>
+</section>
+</section>
+<section id="mitigating-the-invariances" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mitigating-the-invariances">Mitigating the invariances</h3>
+<p>All of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.</p>
+<section id="removing-redundant-parameters-or-predictors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="removing-redundant-parameters-or-predictors">Removing redundant parameters or predictors</h4>
+<p>In the case of the multiple intercepts, <span class="math inline">\(\lambda_1\)</span> and <span class="math inline">\(\lambda_2\)</span>, the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter <span class="math inline">\(\mu\)</span> and sampling distribution <span class="math inline">\(y_n \sim \textsf{normal}(\mu, \sigma)\)</span>. The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix <span class="math inline">\(x\)</span>.</p>
+</section>
+<section id="pinning-parameters" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="pinning-parameters">Pinning parameters</h4>
+<p>The IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability <span class="math inline">\(\alpha_1\)</span> can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.</p>
+<p>This solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters <span class="math inline">\(\delta_i\)</span>. To solve this problem, one of the discrimination parameters, say <span class="math inline">\(\delta_1\)</span>, must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.</p>
+<p>The many-to-one nature of <span class="math inline">\(\texttt{softmax}(\alpha)\)</span> is typically mitigated by pinning a component of <span class="math inline">\(\alpha\)</span>, for instance fixing <span class="math inline">\(\alpha_K = 0\)</span>. The resulting mapping is one-to-one from <span class="math inline">\(K-1\)</span> unconstrained parameters to a <span class="math inline">\(K\)</span>-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a <span class="math inline">\(K-1\)</span>-vector can be written as</p>
+<pre><code>vector softmax_id(vector alpha) {
+  vector[num_elements(alpha) + 1] alphac1;
+  for (k in 1:num_elements(alpha)) {
+    alphac1[k] = alpha[k];
+  }
+  alphac1[num_elements(alphac1)] = 0;
+  return softmax(alphac1);
+}</code></pre>
+</section>
+<section id="adding-priors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="adding-priors">Adding priors</h4>
+<p>So far, the models have been discussed as if the priors on the parameters were improper uniform priors.</p>
+<p>A more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.</p>
+<p>For example, normal priors on the multiple intercepts, <span class="math display">\[
+\lambda_1, \lambda_2 \sim \textsf{normal}(0,\tau),
+\]</span> with a constant scale <span class="math inline">\(\tau\)</span>, ensure that the posterior mode is located at a point where <span class="math inline">\(\lambda_1 = \lambda_2\)</span>, because this minimizes <span class="math inline">\(\log \textsf{normal}(\lambda_1 \mid 0,\tau) + \log
+\textsf{normal}(\lambda_2 \mid 0,\tau)\)</span>.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>The following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.</p>
+<p>The two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><img src="img/non-identified.png" class="img-fluid figure-img"></p>
+<figcaption>Two intercepts with improper prior</figcaption>
+</figure>
+</div>
+<p>Adding a standard normal prior for the intercepts results in a proper posterior.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><img src="img/non-identified-plus-prior.png" class="img-fluid figure-img"></p>
+<figcaption>Two intercepts with proper prior</figcaption>
+</figure>
+</div>
+<p>The single intercept parameterization with no prior also has a proper posterior.</p>
+<div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><img src="img/one-param-identified.png" class="img-fluid figure-img"></p>
+<figcaption>Single intercepts with improper prior</figcaption>
+</figure>
+</div>
+<p>The addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.</p>
+<p>An alternative strategy for identifying a <span class="math inline">\(K\)</span>-simplex parameterization <span class="math inline">\(\theta = \texttt{softmax}(\alpha)\)</span> in terms of an unconstrained <span class="math inline">\(K\)</span>-vector <span class="math inline">\(\alpha\)</span> is to place a prior on the components of <span class="math inline">\(\alpha\)</span> with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning <span class="math inline">\(\alpha_K =
+0\)</span>, the prior-based approach models the <span class="math inline">\(K\)</span> outcomes symmetrically rather than modeling <span class="math inline">\(K-1\)</span> outcomes relative to the <span class="math inline">\(K\)</span>-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.</p>
+</section>
+<section id="vague-strongly-informative-and-weakly-informative-priors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="vague-strongly-informative-and-weakly-informative-priors">Vague, strongly informative, and weakly informative priors</h4>
+<p>Care must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.</p>
+<p>Ideally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.</p>
+<p>When there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.</p>
+<p>Priors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters <span class="math inline">\(\alpha\)</span> to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.</p>
+</section>
+</section>
+</section>
+<section id="label-switching-problematic.section" class="level2">
+<h2 class="anchored" data-anchor-id="label-switching-problematic.section">Label switching in mixture models</h2>
+<p>Where collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.</p>
+<section id="mixture-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mixture-models">Mixture models</h3>
+<p>Consider a normal mixture model with two location parameters <span class="math inline">\(\mu_1\)</span> and <span class="math inline">\(\mu_2\)</span>, a shared scale <span class="math inline">\(\sigma &gt; 0\)</span>, a mixture ratio <span class="math inline">\(\theta \in
+[0,1]\)</span>, and data model <span class="math display">\[
+p(y \mid \theta,\mu_1,\mu_2,\sigma)
+= \prod_{n=1}^N \big( \theta \, \textsf{normal}(y_n \mid \mu_1,\sigma)
+                       + (1 - \theta) \, \textsf{normal}(y_n \mid \mu_2,\sigma) \big).
+\]</span> The issue here is exchangeability of the mixture components, because <span class="math display">\[
+p(\theta,\mu_1,\mu_2,\sigma \mid y) = p\big((1-\theta),\mu_2,\mu_1,\sigma \mid y\big).
+\]</span> The problem is exacerbated as the number of mixture components <span class="math inline">\(K\)</span> grows, as in clustering models, leading to <span class="math inline">\(K!\)</span> identical posterior maxima.</p>
+</section>
+<section id="convergence-monitoring-and-effective-sample-size" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="convergence-monitoring-and-effective-sample-size">Convergence monitoring and effective sample size</h3>
+<p>The analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the <span class="math inline">\(\hat{R}\)</span> convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for <span class="math inline">\(\mu_1\)</span> that is equal to that of <span class="math inline">\(\mu_2\)</span>, and a posterior mean for <span class="math inline">\(\theta\)</span> that is always 1/2, no matter what the data are.</p>
+</section>
+<section id="some-inferences-are-invariant" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="some-inferences-are-invariant">Some inferences are invariant</h3>
+<p>In some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for <span class="math inline">\(\theta\)</span> in the two component mixture model will always be 1/2.</p>
+</section>
+<section id="highly-multimodal-posteriors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="highly-multimodal-posteriors">Highly multimodal posteriors</h3>
+<p>Theoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.</p>
+<p>Being able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for <span class="math inline">\(\mu_1\)</span> to move to a new mode when sampled conditioned on the current values of <span class="math inline">\(\mu_2\)</span> and <span class="math inline">\(\theta\)</span>. For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.</p>
+<p>Even with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.</p>
+</section>
+<section id="hacks-as-fixes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="hacks-as-fixes">Hacks as fixes</h3>
+<p>Several hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.</p>
+<section id="parameter-ordering-constraints" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="parameter-ordering-constraints">Parameter ordering constraints</h4>
+<p>One common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining <span class="math inline">\(\mu_1 &lt; \mu_2\)</span> in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering <span class="math inline">\(\mu_1 &gt; \mu_2\)</span>. In these cases, the posteriors are affected by the constraint and true posterior uncertainty in <span class="math inline">\(\mu_1\)</span> and <span class="math inline">\(\mu_2\)</span> is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use <span class="math inline">\(M\)</span> posterior draws to estimate <span class="math inline">\(\Pr[\mu_1 &gt; \mu_2]\)</span>, will fail, because the estimator <span class="math display">\[
+\Pr[\mu_1 &gt; \mu_2]
+\approx
+\sum_{m=1}^M \textrm{I}\left(\mu_1^{(m)} &gt; \mu_2^{(m)}\right)
+\]</span> will result in an estimate of 0 because the posterior respects the constraint in the model.</p>
+</section>
+<section id="initialization-around-a-single-mode" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="initialization-around-a-single-mode">Initialization around a single mode</h4>
+<p>Another common approach is to run a single chain or to initialize the parameters near realistic values.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
+<p>This can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.</p>
+</section>
+</section>
+</section>
+<section id="component-collapsing-in-mixture-models" class="level2">
+<h2 class="anchored" data-anchor-id="component-collapsing-in-mixture-models">Component collapsing in mixture models</h2>
+<p>It is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of <span class="math inline">\(K\)</span> normals might devolve to have <span class="math inline">\(\mu_i =
+\mu_j\)</span> and <span class="math inline">\(\sigma_i = \sigma_j\)</span> for <span class="math inline">\(i \neq j\)</span>.</p>
+<p>This will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw <span class="math inline">\((m)\)</span>, it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.</p>
+<p>It may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.</p>
+<p>In general, it is difficult to recover exactly the right <span class="math inline">\(K\)</span> mixture components in a mixture model as <span class="math inline">\(K\)</span> increases beyond one (yes, even a two-component mixture can have this problem).</p>
+</section>
+<section id="posteriors-with-unbounded-densities" class="level2">
+<h2 class="anchored" data-anchor-id="posteriors-with-unbounded-densities">Posteriors with unbounded densities</h2>
+<p>In some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.</p>
+<section id="mixture-models-with-varying-scales" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="mixture-models-with-varying-scales">Mixture models with varying scales</h3>
+<p>One such example is a binary mixture model with scales varying by component, <span class="math inline">\(\sigma_1\)</span> and <span class="math inline">\(\sigma_2\)</span> for locations <span class="math inline">\(\mu_1\)</span> and <span class="math inline">\(\mu_2\)</span>. In this situation, the density grows without bound as <span class="math inline">\(\sigma_1 \rightarrow 0\)</span> and <span class="math inline">\(\mu_1 \rightarrow y_n\)</span> for some <span class="math inline">\(n\)</span>; that is, one of the mixture components concentrates all of its mass around a single data item <span class="math inline">\(y_n\)</span>.</p>
+</section>
+<section id="beta-binomial-models-with-skewed-data-and-weak-priors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="beta-binomial-models-with-skewed-data-and-weak-priors">Beta-binomial models with skewed data and weak priors</h3>
+<p>Another example of unbounded densities arises with a posterior such as <span class="math inline">\(\textsf{beta}(\phi \mid 0.5,0.5)\)</span>, which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as <span class="math inline">\(\phi \rightarrow 0\)</span> and <span class="math inline">\(\phi \rightarrow 1\)</span>. Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior <span class="math display">\[\begin{align*}
+p(\phi \mid y)
+&amp;\propto
+   \textsf{beta}(\phi \mid 0.5,0.5) \times \prod_{n=1}^N \textsf{Bernoulli}(y_n \mid \phi) \\
+&amp;=
+   \textsf{beta}\left(\phi \,\middle|\, 0.5 + \sum_{n=1}^N y_n, 0.5 + N - \sum_{n=1}^N y_n\right).
+\end{align*}\]</span></p>
+<p>If <span class="math inline">\(N = 9\)</span> and each <span class="math inline">\(y_n = 1\)</span>, the posterior is <span class="math inline">\(\textsf{beta}(\phi \mid 9.5,0.5)\)</span>. This posterior is unbounded as <span class="math inline">\(\phi
+\rightarrow 1\)</span>. Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.</p>
+<section id="constrained-vs.-unconstrained-scales" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="constrained-vs.-unconstrained-scales">Constrained vs.&nbsp;unconstrained scales</h4>
+<p>Stan does not sample directly on the constrained <span class="math inline">\((0,1)\)</span> space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values <span class="math inline">\(\phi\)</span> are logit-transformed to <span class="math inline">\((-\infty,\infty)\)</span>. The boundaries at 0 and 1 are pushed out to <span class="math inline">\(-\infty\)</span> and <span class="math inline">\(\infty\)</span> respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of <span class="math inline">\((0,1)\)</span> is <span class="math inline">\(\log
+\operatorname{logit}^{-1}(\phi) + \log \operatorname{logit}(1 - \phi)\)</span>.</p>
+<p>There are two problems that still arise, though. The first is that if the posterior mass for <span class="math inline">\(\phi\)</span> is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.</p>
+</section>
+</section>
+</section>
+<section id="posteriors-with-unbounded-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="posteriors-with-unbounded-parameters">Posteriors with unbounded parameters</h2>
+<p>In some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.</p>
+<section id="separability-in-logistic-regression" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="separability-in-logistic-regression">Separability in logistic regression</h3>
+<p>Consider a logistic regression model with <span class="math inline">\(N\)</span> observed outcomes <span class="math inline">\(y_n
+\in \{ 0, 1 \}\)</span>, an <span class="math inline">\(N \times K\)</span> matrix <span class="math inline">\(x\)</span> of predictors, a <span class="math inline">\(K\)</span>-dimensional coefficient vector <span class="math inline">\(\beta\)</span>, and data model <span class="math display">\[
+y_n \sim \textsf{Bernoulli}(\operatorname{logit}^{-1}(x_n \beta)).
+\]</span> Now suppose that column <span class="math inline">\(k\)</span> of the predictor matrix is such that <span class="math inline">\(x_{n,k} &gt; 0\)</span> if and only if <span class="math inline">\(y_n = 1\)</span>, a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as <span class="math inline">\(\beta_k \rightarrow \infty\)</span>, because for cases with <span class="math inline">\(y_n = 1\)</span>, <span class="math inline">\(x_n \beta \rightarrow \infty\)</span> and hence <span class="math inline">\(\operatorname{logit}^{-1}(x_n \beta) \rightarrow 1\)</span>.</p>
+<p>With separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for <span class="math inline">\(\beta_k\)</span> is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for <span class="math inline">\(\beta\)</span>, which ensures a proper posterior.</p>
+</section>
+</section>
+<section id="uniform-posteriors" class="level2">
+<h2 class="anchored" data-anchor-id="uniform-posteriors">Uniform posteriors</h2>
+<p>Suppose your model includes a parameter <span class="math inline">\(\psi\)</span> that is defined on <span class="math inline">\([0,1]\)</span> and is given a flat prior <span class="math inline">\(\textsf{uniform}(\psi \mid 0,1)\)</span>. Now if the data don’t tell us anything about <span class="math inline">\(\psi\)</span>, the posterior is also <span class="math inline">\(\textsf{uniform}(\psi \mid 0,1)\)</span>.</p>
+<p>Although there is no maximum likelihood estimate for <span class="math inline">\(\psi\)</span>, the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on <span class="math inline">\([0,1]\)</span>, the posterior mean for <span class="math inline">\(\psi\)</span> is well-defined with value <span class="math inline">\(1/2\)</span>. Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for <span class="math inline">\(\psi\)</span> at all points in <span class="math inline">\([0,1]\)</span>.</p>
+</section>
+<section id="sampling-difficulties-with-problematic-priors" class="level2">
+<h2 class="anchored" data-anchor-id="sampling-difficulties-with-problematic-priors">Sampling difficulties with problematic priors</h2>
+<p>With an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the <a href="#collinearity.section">collinearity section</a> and illustrated in the non-identifiable density plot.</p>
+<section id="gibbs-sampling" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="gibbs-sampling">Gibbs sampling</h3>
+<p>Gibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.</p>
+<p>Consider what happens with initial values <span class="math inline">\(\lambda_1^{(0)}, \lambda_2^{(0)}\)</span>. Gibbs sampling proceeds in iteration <span class="math inline">\(m\)</span> by drawing <span class="math display">\[\begin{align*}
+\lambda_1^{(m)} &amp;\sim p(\lambda_1 \mid \lambda_2^{(m-1)}, \sigma^{(m-1)},  y) \\
+\lambda_2^{(m)} &amp;\sim p(\lambda_2 \mid \lambda_1^{(m)},   \sigma^{(m-1)},  y) \\
+\sigma^{(m)}    &amp;\sim p(\sigma    \mid \lambda_1^{(m)},   \lambda_2^{(m)}, y).
+\end{align*}\]</span></p>
+<p>Now consider the draw for <span class="math inline">\(\lambda_1\)</span> (the draw for <span class="math inline">\(\lambda_2\)</span> is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next <span class="math inline">\(\lambda_1\)</span> can be drawn is highly constrained by the current values of <span class="math inline">\(\lambda_2\)</span> and <span class="math inline">\(\sigma\)</span>. Gibbs will run quickly and provide seemingly reasonable inferences for <span class="math inline">\(\lambda_1 + \lambda_2\)</span>. But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.</p>
+</section>
+<section id="hamiltonian-monte-carlo-sampling" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="hamiltonian-monte-carlo-sampling">Hamiltonian Monte Carlo sampling</h3>
+<p>Hamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for <span class="math inline">\(\lambda_1\)</span> and <span class="math inline">\(\lambda_2\)</span>, the gradient of the log posterior is going to adjust for the correlation and the simulation will run <span class="math inline">\(\lambda_1\)</span> and <span class="math inline">\(\lambda_2\)</span> in opposite directions along the valley corresponding to the ridge in the posterior log density.</p>
+</section>
+<section id="no-u-turn-sampling" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="no-u-turn-sampling">No-U-turn sampling</h3>
+<p>Stan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior <span class="citation" data-cites="Hoffman-Gelman:2014">(see <a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">Hoffman and Gelman 2014</a>)</span>. NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.</p>
+<p>Here are results of Stan runs with default parameters fit to <span class="math inline">\(N=100\)</span> data points generated from <span class="math inline">\(y_n \sim \textsf{normal}(0,1)\)</span>:</p>
+<p><em>Two Scale Parameters, Improper Prior</em></p>
+<pre><code>Inference for Stan model: improper_stan
+Warmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total
+Sampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total
+
+                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat
+lp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0
+n_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0
+lambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2
+lambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2
+sigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1
+mu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0</code></pre>
+<p><em>Two Scale Parameters, Weak Prior</em></p>
+<pre><code>Warmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total
+Sampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total
+
+                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat
+lp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0
+n_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0
+lambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0
+lambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0
+sigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0
+mu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0</code></pre>
+<p><em>One Scale Parameter, Improper Prior</em></p>
+<pre><code>Warmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total
+Sampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total
+
+                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat
+lp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0
+n_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0
+mu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0
+sigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0</code></pre>
+<p>On the top is the non-identified model with improper uniform priors and data model <span class="math inline">\(y_n \sim \textsf{normal}(\lambda_1 + \lambda_2,
+\sigma)\)</span>.</p>
+<p>In the middle is the same data model as in top plus priors <span class="math inline">\(\lambda_k \sim \textsf{normal}(0,10)\)</span>.</p>
+<p>On the bottom is an identified model with an improper prior, with data model <span class="math inline">\(y_n \sim \textsf{normal}(\mu,\sigma)\)</span>. All models estimate <span class="math inline">\(\mu\)</span> at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value <span class="math inline">\(\mu=0\)</span> is within the 90% posterior intervals in all three models.</p>
+</section>
+<section id="examples-fits-in-stan" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="examples-fits-in-stan">Examples: fits in Stan</h3>
+<p>To illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the <a href="#collinearity.section">collinearity section</a>.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> lambda1;</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> lambda2;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  mu = lambda1 + lambda2;</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The second adds priors to the model block for <code>lambda1</code> and <code>lambda2</code> to the previous model.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a>lambda1 ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>lambda2 ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The third involves a single location parameter, but no priors.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>All three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.</p>
+<ul>
+<li><p>As indicated by <code>R_hat</code> column, all parameters have converged other than <span class="math inline">\(\lambda_1\)</span> and <span class="math inline">\(\lambda_2\)</span> in the non-identified model.</p></li>
+<li><p>The average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.</p></li>
+<li><p>The effective sample size per second for <span class="math inline">\(\mu\)</span> is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for <span class="math inline">\(\sigma\)</span>.</p></li>
+<li><p>In the non-identified model, the 95% interval for <span class="math inline">\(\lambda_1\)</span> is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.</p></li>
+<li><p>In all three models, the simulated value of <span class="math inline">\(\mu=0\)</span> and <span class="math inline">\(\sigma=1\)</span> are well within the posterior 90% intervals.</p></li>
+</ul>
+<p>The first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Hoffman-Gelman:2014" class="csl-entry" role="listitem">
+Hoffman, Matthew D., and Andrew Gelman. 2014. <span>“<span>T</span>he <span>N</span>o-<span>U</span>-<span>T</span>urn <span>S</span>ampler: <span>A</span>daptively <span>S</span>etting <span>P</span>ath <span>L</span>engths in <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>Journal of Machine Learning Research</em> 15: 1593–623. <a href="http://jmlr.org/papers/v15/hoffman14a.html">http://jmlr.org/papers/v15/hoffman14a.html</a>.
+</div>
+<div id="ref-Neal:1996b" class="csl-entry" role="listitem">
+Neal, Radford M. 1996. <span>“Sampling from Multimodal Distributions Using Tempered Transitions.”</span> <em>Statistics and Computing</em> 6 (4): 353–66.
+</div>
+<div id="ref-SwendsenWang:1986" class="csl-entry" role="listitem">
+Swendsen, Robert H., and Jian-Sheng Wang. 1986. <span>“Replica <span>M</span>onte <span>C</span>arlo Simulation of Spin Glasses.”</span> <em>Physical Review Letters</em> 57: 2607–9.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>This example was raised by Richard McElreath on the Stan users group in a query about the difference in behavior between Gibbs sampling as used in BUGS and JAGS and the Hamiltonian Monte Carlo (HMC) and no-U-turn samplers (NUTS) used by Stan.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>The marginal posterior <span class="math inline">\(p(\sigma \mid y)\)</span> for <span class="math inline">\(\sigma\)</span> is proper here as long as there are at least two distinct data points.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>A Laplace prior (or an L1 regularizer for penalized maximum likelihood estimation) is not sufficient to remove this additive invariance. It provides shrinkage, but does not in and of itself identify the parameters because adding a constant to <span class="math inline">\(\lambda_1\)</span> and subtracting it from <span class="math inline">\(\lambda_2\)</span> results in the same value for the prior density.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn4"><p>Tempering methods may be viewed as automated ways to carry out such a search for modes, though most MCMC tempering methods continue to search for modes on an ongoing basis; see <span class="citation" data-cites="SwendsenWang:1986 Neal:1996b">(<a href="#ref-SwendsenWang:1986" role="doc-biblioref">Swendsen and Wang 1986</a>; <a href="#ref-Neal:1996b" role="doc-biblioref">Neal 1996</a>)</span>.<a href="#fnref4" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/proportionality-constants.html" class="pagination-link" aria-label="Proportionality Constants">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Proportionality Constants</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/reparameterization.html" class="pagination-link" aria-label="Reparameterization and Change of Variables">
+        <span class="nav-page-text">Reparameterization and Change of Variables</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/problematic-posteriors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/proportionality-constants.html b/docs/2_39/stan-users-guide/proportionality-constants.html
new file mode 100644
index 000000000..ece81d6b6
--- /dev/null
+++ b/docs/2_39/stan-users-guide/proportionality-constants.html
@@ -0,0 +1,1299 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Proportionality Constants</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/problematic-posteriors.html" rel="next">
+<link href="../stan-users-guide/custom-probability.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Proportionality Constants">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/proportionality-constants.html">Proportionality Constants</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#proportionality-constants.chapter" id="toc-proportionality-constants.chapter" class="nav-link active" data-scroll-target="#proportionality-constants.chapter">Proportionality Constants</a>
+  <ul class="collapse">
+  <li><a href="#dropping-proportionality-constants" id="toc-dropping-proportionality-constants" class="nav-link" data-scroll-target="#dropping-proportionality-constants">Dropping Proportionality Constants</a></li>
+  <li><a href="#keeping-proportionality-constants" id="toc-keeping-proportionality-constants" class="nav-link" data-scroll-target="#keeping-proportionality-constants">Keeping Proportionality Constants</a></li>
+  <li><a href="#user-defined-distributions" id="toc-user-defined-distributions" class="nav-link" data-scroll-target="#user-defined-distributions">User-defined Distributions</a></li>
+  <li><a href="#limitations-on-using-_lupdf-and-_lupmf-functions" id="toc-limitations-on-using-_lupdf-and-_lupmf-functions" class="nav-link" data-scroll-target="#limitations-on-using-_lupdf-and-_lupmf-functions">Limitations on Using <code>_lupdf</code> and <code>_lupmf</code> Functions</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/proportionality-constants.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/proportionality-constants.html">Proportionality Constants</a></li></ol></nav></header>
+
+
+
+
+<section id="proportionality-constants.chapter" class="level1">
+<h1>Proportionality Constants</h1>
+<p>When evaluating a likelihood or prior as part of the log density computation in MCMC, variational inference, or optimization, it is usually only necessary to compute the functions up to a proportionality constant (or similarly compute log densities up to an additive constant). In MCMC this comes from the fact that the distribution being sampled does not need to be normalized (and so it is the normalization constant that is ignored). Similarly the distribution does not need normalized to perform variational inference or do optimizations. The advantage of working with unnormalized distributions is they can make computation quite a bit cheaper.</p>
+<p>There are three different syntaxes to build the model in Stan. The way to select between them is by determining if the proportionality constants are necessary. If performance is not a problem, it is always safe to use the normalized densities.</p>
+<p>The distribution statement (<code>~</code>) and log density increment statement (<code>target +=</code>) with <code>_lupdf()</code> use unnormalized densities for <span class="math inline">\(x\)</span> (dropping proportionality constants):</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>x ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lupdf(x | <span class="dv">0</span>, <span class="dv">1</span>); <span class="co">// the 'u' is for unnormalized</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The log density increment statement (<code>target +=</code>) with <code>_lpdf()</code> uses the full normalized density for <span class="math inline">\(x\)</span> (dropping no constants):</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(x | <span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For discrete distributions, the <code>target +=</code> syntax is using <code>_lupmf</code> and <code>_lpmf</code> instead:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>y ~ bernoulli(<span class="fl">0.5</span>);</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> bernoulli_lupmf(y | <span class="fl">0.5</span>);</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> bernoulli_lpmf(y | <span class="fl">0.5</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="dropping-proportionality-constants" class="level2">
+<h2 class="anchored" data-anchor-id="dropping-proportionality-constants">Dropping Proportionality Constants</h2>
+<p>If a density <span class="math inline">\(p(\theta)\)</span> can be factored into <span class="math inline">\(K g(\theta)\)</span> where <span class="math inline">\(K\)</span> are all the factors that are a not a function of <span class="math inline">\(\theta\)</span> and <span class="math inline">\(g(\theta)\)</span> are all the terms that are a function of <span class="math inline">\(\theta\)</span>, then it is said that <span class="math inline">\(g(\theta)\)</span> is proportional to <span class="math inline">\(p(\theta)\)</span> up to a constant.</p>
+<p>The advantage of all this is that sometimes <span class="math inline">\(K\)</span> is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.</p>
+<p>Stan takes advantage of the proportionality constant fact with the <code>~</code> syntax. Take for instance the normal data model:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="fl">0.0</span>&gt; sigma;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  x ~ normal(mu, sigma);</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Syntactically, this is just shorthand for the equivalent model that replaces the <code>~</code> syntax with a <code>target +=</code> statement and a <code>normal_lupdf</code> function call:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="fl">0.0</span>&gt; sigma;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> normal_lupdf(x | mu, sigma);</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function <code>normal_lupdf</code> is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:</p>
+<p><span class="math display">\[
+\textsf{normal\_lpdf}(x | \mu, \sigma) =
+-\log \left( \sigma \sqrt{2 \pi} \right)
+-\frac{1}{2} \left( \frac{x - \mu}{\sigma} \right)^2.
+\]</span></p>
+<p>Now because the density here is only a function of <span class="math inline">\(x\)</span>, the additive terms in the log density that are not a function of <span class="math inline">\(x\)</span> can be dropped. In this case it is enough to know only the quadratic term:</p>
+<p><span class="math display">\[
+\textsf{normal\_lupdf}(x | \mu, \sigma) =
+-\frac{1}{2} \left( \frac{x - \mu}{\sigma} \right)^2.
+\]</span></p>
+</section>
+<section id="keeping-proportionality-constants" class="level2">
+<h2 class="anchored" data-anchor-id="keeping-proportionality-constants">Keeping Proportionality Constants</h2>
+<p>In the case that the proportionality constants were needed for a normal log density the function <code>normal_lpdf</code> can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the <code>~</code> or <code>target += normal_lupdf</code> syntaxes if it is absolutely clear that the proportionality constants are not necessary.</p>
+</section>
+<section id="user-defined-distributions" class="level2">
+<h2 class="anchored" data-anchor-id="user-defined-distributions">User-defined Distributions</h2>
+<p>When a custom <code>_lpdf</code> or <code>_lpmf</code> function is defined, the compiler will automatically make available a <code>_lupdf</code> or <code>_lupmf</code> version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.</p>
+<p>The difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.</p>
+<p>The following code demonstrates the different behaviors:</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> custom1_lpdf(x) {</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> normal_lupdf(x | <span class="fl">0.0</span>, <span class="fl">1.0</span>);</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> custom2_lpdf(x) {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> normal_lpdf(x | <span class="fl">0.0</span>, <span class="fl">1.0</span>);</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>  mu ~ custom1(); <span class="co">// Normalization constants dropped</span></span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> custom1_lupdf(mu); <span class="co">// Normalization constants dropped</span></span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> custom1_lpdf(mu);  <span class="co">// Normalization constants kept</span></span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-17"><a href="#cb6-17" aria-hidden="true" tabindex="-1"></a>  mu ~ custom2();  <span class="co">// Normalization constants kept</span></span>
+<span id="cb6-18"><a href="#cb6-18" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> custom2_lupdf(mu);  <span class="co">// Normalization constants kept</span></span>
+<span id="cb6-19"><a href="#cb6-19" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> custom2_lpdf(mu);  <span class="co">// Normalization constants kept</span></span>
+<span id="cb6-20"><a href="#cb6-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="limitations-on-using-_lupdf-and-_lupmf-functions" class="level2">
+<h2 class="anchored" data-anchor-id="limitations-on-using-_lupdf-and-_lupmf-functions">Limitations on Using <code>_lupdf</code> and <code>_lupmf</code> Functions</h2>
+<p>To avoid ambiguities in how the normalization constants work, functions ending in <code>_lupdf</code> and <code>_lupmf</code> can only be used in the model block or user-defined probability functions (functions ending in <code>_lpdf</code> or <code>_lpmf</code>).</p>
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/custom-probability.html" class="pagination-link" aria-label="Custom Probability Functions">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Custom Probability Functions</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/problematic-posteriors.html" class="pagination-link" aria-label="Problematic Posteriors">
+        <span class="nav-page-text">Problematic Posteriors</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/proportionality-constants.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/references.html b/docs/2_39/stan-users-guide/references.html
new file mode 100644
index 000000000..a4117cb99
--- /dev/null
+++ b/docs/2_39/stan-users-guide/references.html
@@ -0,0 +1,821 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>references</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Docs">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#references" id="toc-references" class="nav-link active" data-scroll-target="#references"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></a></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"></header>
+
+
+
+
+<section id="references" class="level1 unnumbered">
+<h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">References</i></h1>
+
+
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/regression.html b/docs/2_39/stan-users-guide/regression.html
new file mode 100644
index 000000000..379f4b6f1
--- /dev/null
+++ b/docs/2_39/stan-users-guide/regression.html
@@ -0,0 +1,2305 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Regression Models</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/time-series.html" rel="next">
+<link href="../stan-users-guide/index.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Regression Models">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Regression Models</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#regression-models" id="toc-regression-models" class="nav-link active" data-scroll-target="#regression-models">Regression Models</a>
+  <ul class="collapse">
+  <li><a href="#linear-regression" id="toc-linear-regression" class="nav-link" data-scroll-target="#linear-regression">Linear regression</a>
+  <ul class="collapse">
+  <li><a href="#vectorization.section" id="toc-vectorization.section" class="nav-link" data-scroll-target="#vectorization.section">Matrix notation and vectorization</a></li>
+  </ul></li>
+  <li><a href="#QR-reparameterization.section" id="toc-QR-reparameterization.section" class="nav-link" data-scroll-target="#QR-reparameterization.section">The QR reparameterization</a></li>
+  <li><a href="#regression-priors.section" id="toc-regression-priors.section" class="nav-link" data-scroll-target="#regression-priors.section">Priors for coefficients and scales</a></li>
+  <li><a href="#robust-noise-models" id="toc-robust-noise-models" class="nav-link" data-scroll-target="#robust-noise-models">Robust noise models</a></li>
+  <li><a href="#logistic-probit-regression.section" id="toc-logistic-probit-regression.section" class="nav-link" data-scroll-target="#logistic-probit-regression.section">Logistic and probit regression</a></li>
+  <li><a href="#multi-logit.section" id="toc-multi-logit.section" class="nav-link" data-scroll-target="#multi-logit.section">Multi-logit regression</a>
+  <ul class="collapse">
+  <li><a href="#identifiability" id="toc-identifiability" class="nav-link" data-scroll-target="#identifiability">Identifiability</a></li>
+  </ul></li>
+  <li><a href="#parameterizing-centered-vectors" id="toc-parameterizing-centered-vectors" class="nav-link" data-scroll-target="#parameterizing-centered-vectors">Parameterizing centered vectors</a>
+  <ul class="collapse">
+  <li><a href="#built-in-sum-to-zero-vector" id="toc-built-in-sum-to-zero-vector" class="nav-link" data-scroll-target="#built-in-sum-to-zero-vector">Built-in sum-to-zero vector</a></li>
+  <li><a href="#soft-centering" id="toc-soft-centering" class="nav-link" data-scroll-target="#soft-centering">Soft centering</a></li>
+  </ul></li>
+  <li><a href="#ordered-logistic.section" id="toc-ordered-logistic.section" class="nav-link" data-scroll-target="#ordered-logistic.section">Ordered logistic and probit regression</a>
+  <ul class="collapse">
+  <li><a href="#ordered-logistic-regression" id="toc-ordered-logistic-regression" class="nav-link" data-scroll-target="#ordered-logistic-regression">Ordered logistic regression</a></li>
+  </ul></li>
+  <li><a href="#hierarchical-regression" id="toc-hierarchical-regression" class="nav-link" data-scroll-target="#hierarchical-regression">Hierarchical regression</a></li>
+  <li><a href="#hierarchical-priors.section" id="toc-hierarchical-priors.section" class="nav-link" data-scroll-target="#hierarchical-priors.section">Hierarchical priors</a>
+  <ul class="collapse">
+  <li><a href="#boundary-avoiding-priors-for-mle-in-hierarchical-models" id="toc-boundary-avoiding-priors-for-mle-in-hierarchical-models" class="nav-link" data-scroll-target="#boundary-avoiding-priors-for-mle-in-hierarchical-models">Boundary-avoiding priors for MLE in hierarchical models</a></li>
+  </ul></li>
+  <li><a href="#item-response-models.section" id="toc-item-response-models.section" class="nav-link" data-scroll-target="#item-response-models.section">Item-response theory models</a>
+  <ul class="collapse">
+  <li><a href="#data-declaration-with-missingness" id="toc-data-declaration-with-missingness" class="nav-link" data-scroll-target="#data-declaration-with-missingness">Data declaration with missingness</a></li>
+  <li><a href="#pl-rasch-model" id="toc-pl-rasch-model" class="nav-link" data-scroll-target="#pl-rasch-model">1PL (Rasch) model</a></li>
+  <li><a href="#multilevel-2pl-model" id="toc-multilevel-2pl-model" class="nav-link" data-scroll-target="#multilevel-2pl-model">Multilevel 2PL model</a></li>
+  </ul></li>
+  <li><a href="#priors-for-identification.section" id="toc-priors-for-identification.section" class="nav-link" data-scroll-target="#priors-for-identification.section">Priors for identifiability</a>
+  <ul class="collapse">
+  <li><a href="#location-and-scale-invariance" id="toc-location-and-scale-invariance" class="nav-link" data-scroll-target="#location-and-scale-invariance">Location and scale invariance</a></li>
+  <li><a href="#collinearity" id="toc-collinearity" class="nav-link" data-scroll-target="#collinearity">Collinearity</a></li>
+  <li><a href="#separability" id="toc-separability" class="nav-link" data-scroll-target="#separability">Separability</a></li>
+  </ul></li>
+  <li><a href="#multivariate-hierarchical-priors.section" id="toc-multivariate-hierarchical-priors.section" class="nav-link" data-scroll-target="#multivariate-hierarchical-priors.section">Multivariate priors for hierarchical models</a>
+  <ul class="collapse">
+  <li><a href="#multivariate-regression-example" id="toc-multivariate-regression-example" class="nav-link" data-scroll-target="#multivariate-regression-example">Multivariate regression example</a></li>
+  </ul></li>
+  <li><a href="#prediction-forecasting-and-backcasting" id="toc-prediction-forecasting-and-backcasting" class="nav-link" data-scroll-target="#prediction-forecasting-and-backcasting">Prediction, forecasting, and backcasting</a>
+  <ul class="collapse">
+  <li><a href="#programming-predictions" id="toc-programming-predictions" class="nav-link" data-scroll-target="#programming-predictions">Programming predictions</a></li>
+  <li><a href="#predictions-as-generated-quantities" id="toc-predictions-as-generated-quantities" class="nav-link" data-scroll-target="#predictions-as-generated-quantities">Predictions as generated quantities</a></li>
+  </ul></li>
+  <li><a href="#multivariate-outcomes" id="toc-multivariate-outcomes" class="nav-link" data-scroll-target="#multivariate-outcomes">Multivariate outcomes</a>
+  <ul class="collapse">
+  <li><a href="#seemingly-unrelated-regressions" id="toc-seemingly-unrelated-regressions" class="nav-link" data-scroll-target="#seemingly-unrelated-regressions">Seemingly unrelated regressions</a></li>
+  <li><a href="#multivariate-probit-regression" id="toc-multivariate-probit-regression" class="nav-link" data-scroll-target="#multivariate-probit-regression">Multivariate probit regression</a></li>
+  </ul></li>
+  <li><a href="#applications-of-pseudorandom-number-generation" id="toc-applications-of-pseudorandom-number-generation" class="nav-link" data-scroll-target="#applications-of-pseudorandom-number-generation">Applications of pseudorandom number generation</a>
+  <ul class="collapse">
+  <li><a href="#prediction" id="toc-prediction" class="nav-link" data-scroll-target="#prediction">Prediction</a></li>
+  <li><a href="#posterior-predictive-checks" id="toc-posterior-predictive-checks" class="nav-link" data-scroll-target="#posterior-predictive-checks">Posterior predictive checks</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/regression.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Regression Models</a></li></ol></nav></header>
+
+
+
+
+<section id="regression-models" class="level1">
+<h1>Regression Models</h1>
+<p>Stan supports regression models from simple linear regressions to multilevel generalized linear models.</p>
+<section id="linear-regression" class="level2">
+<h2 class="anchored" data-anchor-id="linear-regression">Linear regression</h2>
+<p>The simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as <span class="math display">\[
+y_n = \alpha + \beta x_n + \epsilon_n
+\quad\text{where}\quad
+\epsilon_n \sim \operatorname{normal}(0,\sigma).
+\]</span></p>
+<p>This is equivalent to the following sampling involving the residual, <span class="math display">\[
+y_n - (\alpha + \beta X_n) \sim \operatorname{normal}(0,\sigma),
+\]</span> and reducing still further, to <span class="math display">\[
+y_n \sim \operatorname{normal}(\alpha + \beta X_n, \, \sigma).
+\]</span></p>
+<p>This latter form of the model is coded in Stan as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There are <code>N</code> observations and for each observation, <span class="math inline">\(n \in N\)</span>, we have predictor <code>x[n]</code> and outcome <code>y[n]</code>. The intercept and slope parameters are <code>alpha</code> and <code>beta</code>. The model assumes a normally distributed noise term with scale <code>sigma</code>. This model has improper priors for the two regression coefficients.</p>
+<section id="vectorization.section" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectorization.section">Matrix notation and vectorization</h3>
+<p>The distribution statement in the previous model is vectorized, with</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(alpha + beta * x, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>providing the same model as the unvectorized version,</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(alpha + beta * x[n], sigma);</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In addition to being more concise, the vectorized form is much faster.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>In general, Stan allows the arguments to distributions such as <code>normal</code> to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.</p>
+<p>The other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because <code>x</code> is of type <code>vector</code> and <code>beta</code> of type <code>real</code>, the expression <code>beta * x</code> is of type <code>vector</code>. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;   <span class="co">// number of data items</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;   <span class="co">// number of predictors</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;   <span class="co">// predictor matrix</span></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;      <span class="co">// outcome vector</span></span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;           <span class="co">// intercept</span></span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;       <span class="co">// coefficients for predictors</span></span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;  <span class="co">// error scale</span></span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta + alpha, sigma);  <span class="co">// data model</span></span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The constraint <code>lower=0</code> in the declaration of <code>sigma</code> constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.</p>
+<p>In the model above, <code>x</code> is an <span class="math inline">\(N \times K\)</span> matrix of predictors and <code>beta</code> a <span class="math inline">\(K\)</span>-vector of coefficients, so <code>x * beta</code> is an <span class="math inline">\(N\)</span>-vector of predictions, one for each of the <span class="math inline">\(N\)</span> data items. These predictions line up with the outcomes in the <span class="math inline">\(N\)</span>-vector <code>y</code>, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix <code>x</code> to remove the <code>alpha</code> parameter.</p>
+<p>The distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(x[n] * beta, sigma);</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With Stan’s matrix indexing scheme, <code>x[n]</code> picks out row <code>n</code> of the matrix <code>x</code>; because <code>beta</code> is a column vector, the product <code>x[n] * beta</code> is a scalar of type <code>real</code>.</p>
+<section id="intercepts-as-inputs" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="intercepts-as-inputs">Intercepts as inputs</h4>
+<p>In the model formulation</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(x * beta, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>there is no longer an intercept coefficient <code>alpha</code>. Instead, we have assumed that the first column of the input matrix <code>x</code> is a column of 1 values. This way, <code>beta[1]</code> plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.</p>
+</section>
+</section>
+</section>
+<section id="QR-reparameterization.section" class="level2">
+<h2 class="anchored" data-anchor-id="QR-reparameterization.section">The QR reparameterization</h2>
+<p>In the previous example, the linear predictor can be written as <span class="math inline">\(\eta
+= x \beta\)</span>, where <span class="math inline">\(\eta\)</span> is a <span class="math inline">\(N\)</span>-vector of predictions, <span class="math inline">\(x\)</span> is a <span class="math inline">\(N
+\times K\)</span> matrix, and <span class="math inline">\(\beta\)</span> is a <span class="math inline">\(K\)</span>-vector of coefficients. Presuming <span class="math inline">\(N \geq K\)</span>, we can exploit the fact that any design matrix <span class="math inline">\(x\)</span> can be decomposed using the thin QR decomposition into an orthogonal matrix <span class="math inline">\(Q\)</span> and an upper-triangular matrix <span class="math inline">\(R\)</span>, i.e.&nbsp;<span class="math inline">\(x = Q
+R\)</span>.</p>
+<p>The functions <code>qr_thin_Q</code> and <code>qr_thin_R</code> implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using <code>qr_Q</code> and <code>qr_R</code>, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the <code>qr_thin_Q</code> and <code>qr_thin_R</code> functions). In practice, it is best to write <span class="math inline">\(x = Q^\ast
+R^\ast\)</span> where <span class="math inline">\(Q^\ast = Q * \sqrt{n - 1}\)</span> and <span class="math inline">\(R^\ast =
+\frac{1}{\sqrt{n - 1}} R\)</span>. Thus, we can equivalently write <span class="math inline">\(\eta = x
+\beta = Q R \beta = Q^\ast R^\ast \beta\)</span>. If we let <span class="math inline">\(\theta = R^\ast
+\beta\)</span>, then we have <span class="math inline">\(\eta = Q^\ast \theta\)</span> and <span class="math inline">\(\beta = R^{\ast^{-1}}
+\theta\)</span>. In that case, the previous Stan program becomes</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;   <span class="co">// number of data items</span></span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;   <span class="co">// number of predictors</span></span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;   <span class="co">// predictor matrix</span></span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;      <span class="co">// outcome vector</span></span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] Q_ast;</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] R_ast;</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] R_ast_inverse;</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// thin and scale the QR decomposition</span></span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>  Q_ast = qr_thin_Q(x) * sqrt(N - <span class="dv">1</span>);</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a>  R_ast = qr_thin_R(x) / sqrt(N - <span class="dv">1</span>);</span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  R_ast_inverse = inverse(R_ast);</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;           <span class="co">// intercept</span></span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] theta;      <span class="co">// coefficients on Q_ast</span></span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;  <span class="co">// error scale</span></span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a>  y ~ normal(Q_ast * theta + alpha, sigma);  <span class="co">// data model</span></span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-24"><a href="#cb7-24" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb7-25"><a href="#cb7-25" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb7-26"><a href="#cb7-26" aria-hidden="true" tabindex="-1"></a>  beta = R_ast_inverse * theta; <span class="co">// coefficients on x</span></span>
+<span id="cb7-27"><a href="#cb7-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Since this Stan program generates equivalent predictions for <span class="math inline">\(y\)</span> and the same posterior distribution for <span class="math inline">\(\alpha\)</span>, <span class="math inline">\(\beta\)</span>, and <span class="math inline">\(\sigma\)</span> as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:</p>
+<ol type="1">
+<li><p>The columns of <span class="math inline">\(Q^\ast\)</span> are orthogonal whereas the columns of <span class="math inline">\(x\)</span> generally are not. Thus, it is easier for a Markov Chain to move around in <span class="math inline">\(\theta\)</span>-space than in <span class="math inline">\(\beta\)</span>-space.</p></li>
+<li><p>The columns of <span class="math inline">\(Q^\ast\)</span> have the same scale whereas the columns of <span class="math inline">\(x\)</span> generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps</p></li>
+<li><p>Since the covariance matrix for the columns of <span class="math inline">\(Q^\ast\)</span> is an identity matrix, <span class="math inline">\(\theta\)</span> typically has a reasonable scale if the units of <span class="math inline">\(y\)</span> are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.</p></li>
+</ol>
+<p>Consequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever <span class="math inline">\(K &gt; 1\)</span> and you do not have an informative prior on the location of <span class="math inline">\(\beta\)</span>. It can also be worthwhile to subtract the mean from each column of <span class="math inline">\(x\)</span> before obtaining the QR decomposition, which does not affect the posterior distribution of <span class="math inline">\(\theta\)</span> or <span class="math inline">\(\beta\)</span> but does affect <span class="math inline">\(\alpha\)</span> and allows you to interpret <span class="math inline">\(\alpha\)</span> as the expectation of <span class="math inline">\(y\)</span> in a linear model.</p>
+</section>
+<section id="regression-priors.section" class="level2">
+<h2 class="anchored" data-anchor-id="regression-priors.section">Priors for coefficients and scales</h2>
+<p>See our <a href="https://github.com/stan-dev/stan/wiki/Prior-Choice-Recommendations">general discussion of priors</a> for tips on priors for parameters in regression models.</p>
+<p>Later sections discuss <a href="#hierarchical-priors.section">univariate hierarchical priors</a> and <a href="#multivariate-hierarchical-priors.section">multivariate hierarchical priors</a>, as well as <a href="#priors-for-identification.section">priors used to identify models</a>.</p>
+<p>However, as described in <a href="#QR-reparameterization.section">QR-reparameterization section</a>, if you do not have an informative prior on the <em>location</em> of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.</p>
+</section>
+<section id="robust-noise-models" class="level2">
+<h2 class="anchored" data-anchor-id="robust-noise-models">Robust noise models</h2>
+<p>The standard approach to linear regression is to model the noise term <span class="math inline">\(\epsilon\)</span> as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-<span class="math inline">\(t\)</span> distribution. To code this in Stan, the distribution distribution is changed to the following.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; nu;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  y ~ student_t(nu, alpha + beta * x, sigma);</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The degrees of freedom constant <code>nu</code> is specified as data.</p>
+</section>
+<section id="logistic-probit-regression.section" class="level2">
+<h2 class="anchored" data-anchor-id="logistic-probit-regression.section">Logistic and probit regression</h2>
+<p>For binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in <span class="math inline">\((-\infty,\infty)\)</span> to probability values in <span class="math inline">\((0,1)\)</span>. Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both <em>S</em>-shaped).</p>
+<p>A logistic regression model with one predictor and an intercept is coded as follows.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(alpha + beta * x);</span>
+<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The noise parameter is built into the Bernoulli formulation here rather than specified directly.</p>
+<p>Logistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by <span class="math display">\[
+\operatorname{logit}(v) = \log \left( \frac{v}{1-v} \right).
+\]</span></p>
+<p>The inverse of the link function appears in the model: <span class="math display">\[
+\operatorname{logit}^{-1}(u) = \texttt{inv}\mathtt{\_}\texttt{logit}(u) = \frac{1}{1 + \exp(-u)}.
+\]</span></p>
+<p>The model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by <span class="math display">\[
+\texttt{bernoulli}\mathtt{\_}\texttt{logit}\left(y \mid \alpha \right)
+=
+\texttt{bernoulli}\left(y \mid \operatorname{logit}^{-1}(\alpha)\right).
+\]</span></p>
+<p>The formulation is also vectorized in the sense that <code>alpha</code> and <code>beta</code> are scalars and <code>x</code> is a vector, so that <code>alpha   + beta * x</code> is a vector. The vectorized formulation is equivalent to the less efficient version</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(alpha + beta * x[n]);</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Expanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Other link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as</p>
+<p><span class="math display">\[
+\Phi(x) = \int_{-\infty}^x \textsf{normal}\left(y \mid 0,1 \right) \,\textrm{d}y.
+\]</span></p>
+<p>The cumulative standard normal distribution function <span class="math inline">\(\Phi\)</span> is implemented in Stan as the function <code>Phi</code>. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ bernoulli(Phi(alpha + beta * x[n]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A fast approximation to the cumulative standard normal distribution function <span class="math inline">\(\Phi\)</span> is implemented in Stan as the function <code>Phi_approx</code>.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> The approximate probit regression model may be coded with the following.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="multi-logit.section" class="level2">
+<h2 class="anchored" data-anchor-id="multi-logit.section">Multi-logit regression</h2>
+<p>Multiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are <span class="math inline">\(K\)</span> possible outcomes for each output variable <span class="math inline">\(y_n\)</span>. Also suppose that there is a <span class="math inline">\(D\)</span>-dimensional vector <span class="math inline">\(x_n\)</span> of predictors for <span class="math inline">\(y_n\)</span>. The multi-logit model with <span class="math inline">\(\textsf{normal}(0,5)\)</span> priors on the coefficients is coded as follows.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> K;</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> D;</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> y;</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] x;</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, K] beta;</span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb14-12"><a href="#cb14-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x_beta = x * beta;</span>
+<span id="cb14-13"><a href="#cb14-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-14"><a href="#cb14-14" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb14-15"><a href="#cb14-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-16"><a href="#cb14-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb14-17"><a href="#cb14-17" aria-hidden="true" tabindex="-1"></a>    y[n] ~ categorical_logit(x_beta[n]');</span>
+<span id="cb14-18"><a href="#cb14-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-19"><a href="#cb14-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb14-20"><a href="#cb14-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where <code>x_beta[n]'</code> is the transpose of <code>x_beta[n]</code>. The prior on <code>beta</code> is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, <span class="math display">\[
+\texttt{categorical}\mathtt{\_}\texttt{logit}\left(y \mid \alpha\right)
+=
+\texttt{categorical}\left(y \mid \texttt{softmax}(\alpha)\right),
+\]</span> where <span class="math display">\[
+\texttt{softmax}(u) = \exp(u) / \operatorname{sum}\left(\exp(u)\right).
+\]</span></p>
+<p>The categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ categorical(softmax(x[n] * beta));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="constraints-on-data-declarations" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="constraints-on-data-declarations">Constraints on data declarations</h4>
+<p>The data block in the above model is defined without constraints on sizes <code>K</code>, <code>N</code>, and <code>D</code> or on the outcome array <code>y</code>. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These constraints arise because the number of categories, <code>K</code>, must be at least two in order for a categorical model to be useful. The number of data items, <code>N</code>, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of <code>1:N</code> when <code>N</code> is equal to zero ensures the loop’s body will not be executed. The number of predictors, <code>D</code>, must be at least one in order for <code>beta * x[n]</code> to produce an appropriate argument for <code>softmax()</code>. The categorical outcomes <code>y[n]</code> must be between <code>1</code> and <code>K</code> in order for the discrete sampling to be well defined.</p>
+<p>Constraints on data declarations are optional. Constraints on parameters declared in the <code>parameters</code> block, on the other hand, are <em>not</em> optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.</p>
+</section>
+<section id="identifiability" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="identifiability">Identifiability</h3>
+<p>Because softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.</p>
+<p>An alternative is to use <span class="math inline">\((K-1)\)</span>-vectors by fixing one of them to be zero. The <a href="../stan-users-guide/missing-data.html#partially-known-parameters.section">partially known parameters section</a> discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use <span class="math inline">\((K - 1)\)</span>-vectors</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, K - <span class="dv">1</span>] beta_raw;</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[D] zeros = rep_vector(<span class="dv">0</span>, D);</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>which can then be appended to <code>beta_raw</code> to produce the coefficient matrix <code>beta</code>,</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, K] beta = append_col(beta_raw, zeros);</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>rep_vector(0, D)</code> call creates a column vector of size <code>D</code> with all entries set to zero. The derived matrix <code>beta</code> is then defined to be the result of appending the vector <code>zeros</code> as a new column at the end of <code>beta_raw</code>; the vector <code>zeros</code> is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.</p>
+<p>This is not the same model as using <span class="math inline">\(K\)</span>-vectors as parameters, because now the prior only applies to <span class="math inline">\((K-1)\)</span>-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.</p>
+</section>
+</section>
+<section id="parameterizing-centered-vectors" class="level2">
+<h2 class="anchored" data-anchor-id="parameterizing-centered-vectors">Parameterizing centered vectors</h2>
+<p>When there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept <span class="math inline">\(\alpha\)</span> and then a varying effect <span class="math inline">\(\beta_k\)</span> for age group <span class="math inline">\(k\)</span> to make a linear predictor <span class="math inline">\(\alpha +
+\beta_k\)</span>. With this predictor, we can add a constant to <span class="math inline">\(\alpha\)</span> and subtract from each <span class="math inline">\(\beta_k\)</span> and get exactly the same likelihood.</p>
+<p>The traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., <span class="math inline">\(\beta_1 = 0\)</span>. With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to <span class="math inline">\(\beta_1\)</span>.</p>
+<p>In a Bayesian setting, a proper prior on each of the <span class="math inline">\(\beta\)</span> is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the <a href="../stan-users-guide/problematic-posteriors.html#collinearity.section">collinearity section</a>.</p>
+<p>An alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., <span class="math inline">\(\sum_{k=1}^K \beta_k = 0.\)</span></p>
+<p>A parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the <a href="#multi-logit.section">multi-logit section</a> for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the <a href="#item-response-models.section">item-response model section</a> for details).</p>
+<section id="built-in-sum-to-zero-vector" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="built-in-sum-to-zero-vector">Built-in sum-to-zero vector</h3>
+<p>As of Stan 2.36, there is a built in <code>sum_to_zero_vector</code> type, which can be used as follows.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">sum_to_zero_vector</span>[K] beta;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This produces a vector of size <code>K</code> such that <code>sum(beta) = 0</code>. In the unconstrained representation requires only <code>K - 1</code> values because the last is determined by the first <code>K - 1</code>.</p>
+<p>Placing a prior on <code>beta</code> in this parameterization, for example,</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>leads to a subtly different posterior than what you would get with the same prior on an unconstrained size-<code>K</code> vector. As explained below, the variance is reduced.</p>
+<p>The sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., <span class="math inline">\(\beta_K
+= -\sum_{k=1}^{K-1} \beta_k.\)</span> But that leads to high correlation among the <span class="math inline">\(\beta_k\)</span>.</p>
+<p>The transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; <span class="citation" data-cites="seyboldt:2024">Seyboldt (<a href="#ref-seyboldt:2024" role="doc-biblioref">2024</a>)</span> provides an explanation. The <em>Stan Reference Manual</em> provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.</p>
+<section id="marginal-distribution-of-sum-to-zero-components" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="marginal-distribution-of-sum-to-zero-components">Marginal distribution of sum-to-zero components</h4>
+<p>On the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of <code>beta</code>,</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, inv(sqrt(<span class="dv">1</span> - inv(K))));</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The scale component can be multiplied by <code>sigma</code> to produce a <code>normal(0, sigma)</code> prior marginally.</p>
+<p>To generate distributions with marginals other than standard normal, the resulting <code>beta</code> may be scaled by some factor <code>sigma</code> and translated to some new location <code>mu</code>.</p>
+</section>
+</section>
+<section id="soft-centering" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="soft-centering">Soft centering</h3>
+<p>Adding a prior such as <span class="math inline">\(\beta \sim \textsf{normal}(0,\epsilon)\)</span> for a small <span class="math inline">\(\epsilon\)</span> will provide a kind of soft centering of a parameter vector <span class="math inline">\(\beta\)</span> by preferring, all else being equal, that <span class="math inline">\(\sum_{k=1}^K
+\beta_k = 0\)</span>. This approach is only guaranteed to roughly center if <span class="math inline">\(\beta\)</span> and the elementwise addition <span class="math inline">\(\beta + c\)</span> for a scalar constant <span class="math inline">\(c\)</span> produce the same likelihood (perhaps by another vector <span class="math inline">\(\alpha\)</span> being transformed to <span class="math inline">\(\alpha - c\)</span>, as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an <span class="math inline">\(\epsilon\)</span>. If <span class="math inline">\(\epsilon\)</span> is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.</p>
+</section>
+</section>
+<section id="ordered-logistic.section" class="level2">
+<h2 class="anchored" data-anchor-id="ordered-logistic.section">Ordered logistic and probit regression</h2>
+<p>Ordered regression for an outcome <span class="math inline">\(y_n \in \{ 1, \dotsc, k \}\)</span> with predictors <span class="math inline">\(x_n \in \mathbb{R}^D\)</span> is determined by a single coefficient vector <span class="math inline">\(\beta \in \mathbb{R}^D\)</span> along with a sequence of cutpoints <span class="math inline">\(c \in
+\mathbb{R}^{K-1}\)</span> sorted so that <span class="math inline">\(c_d &lt; c_{d+1}\)</span>. The discrete output is <span class="math inline">\(k\)</span> if the linear predictor <span class="math inline">\(x_n \beta\)</span> falls between <span class="math inline">\(c_{k-1}\)</span> and <span class="math inline">\(c_k\)</span>, assuming <span class="math inline">\(c_0 = -\infty\)</span> and <span class="math inline">\(c_K = \infty\)</span>. The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.</p>
+<section id="ordered-logistic-regression" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ordered-logistic-regression">Ordered logistic regression</h3>
+<p>The ordered logistic model can be coded in Stan using the <code>ordered</code> data type for the cutpoints and the built-in <code>ordered_logistic</code> distribution.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; y;</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] x;</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[D] beta;</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">ordered</span>[K - <span class="dv">1</span>] c;</span>
+<span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  y ~ ordered_logistic(x * beta, c);</span>
+<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The vector of cutpoints <code>c</code> is declared as <code>ordered[K - 1]</code>, which guarantees that <code>c[k]</code> is less than <code>c[k + 1]</code>.</p>
+<p>If the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.</p>
+<p>The equivalent model can be written using <code>ordered_logistic_glm</code> distribution, which can provide more efficient computation in case of higher dimensional <code>beta</code>.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a>  y ~ ordered_logistic_glm(x,  beta, c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="ordered-probit" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="ordered-probit">Ordered probit</h4>
+<p>An ordered probit model can be coded in exactly the same way by using the built-in <code>ordered_probit</code> distribution.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  ordered_probit(x * beta, c);</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There is not yet an <code>ordered_probit_glm</code> distribution in Stan.</p>
+</section>
+</section>
+</section>
+<section id="hierarchical-regression" class="level2">
+<h2 class="anchored" data-anchor-id="hierarchical-regression">Hierarchical regression</h2>
+<p>The simplest multilevel model is a hierarchical model in which the data are grouped into <span class="math inline">\(L\)</span> distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients <span class="math inline">\(\beta\)</span>. At the other extreme, an approach with no pooling assigns each level <span class="math inline">\(l\)</span> its own coefficient vector <span class="math inline">\(\beta_l\)</span> that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.</p>
+<p>Suppose each binary outcome <span class="math inline">\(y_n \in \{ 0, 1 \}\)</span> has an associated level, <span class="math inline">\(ll_n \in \{ 1, \dotsc, L \}\)</span>. Each outcome will also have an associated predictor vector <span class="math inline">\(x_n \in \mathbb{R}^D\)</span>. Each level <span class="math inline">\(l\)</span> gets its own coefficient vector <span class="math inline">\(\beta_l \in \mathbb{R}^D\)</span>. The hierarchical structure involves drawing the coefficients <span class="math inline">\(\beta_{l,d}
+\in \mathbb{R}\)</span> from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.</p>
+<p>The following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.</p>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb26-2"><a href="#cb26-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb26-3"><a href="#cb26-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb26-4"><a href="#cb26-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; L;</span>
+<span id="cb26-5"><a href="#cb26-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb26-6"><a href="#cb26-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=L&gt; ll;</span>
+<span id="cb26-7"><a href="#cb26-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">row_vector</span>[D] x;</span>
+<span id="cb26-8"><a href="#cb26-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb26-9"><a href="#cb26-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb26-10"><a href="#cb26-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span> mu;</span>
+<span id="cb26-11"><a href="#cb26-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb26-12"><a href="#cb26-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[L] <span class="dt">vector</span>[D] beta;</span>
+<span id="cb26-13"><a href="#cb26-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb26-14"><a href="#cb26-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb26-15"><a href="#cb26-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
+<span id="cb26-16"><a href="#cb26-16" aria-hidden="true" tabindex="-1"></a>    mu[d] ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
+<span id="cb26-17"><a href="#cb26-17" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (l <span class="cf">in</span> <span class="dv">1</span>:L) {</span>
+<span id="cb26-18"><a href="#cb26-18" aria-hidden="true" tabindex="-1"></a>      beta[l, d] ~ normal(mu[d], sigma[d]);</span>
+<span id="cb26-19"><a href="#cb26-19" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb26-20"><a href="#cb26-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb26-21"><a href="#cb26-21" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb26-22"><a href="#cb26-22" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));</span>
+<span id="cb26-23"><a href="#cb26-23" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb26-24"><a href="#cb26-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The standard deviation parameter <code>sigma</code> gets an implicit uniform prior on <span class="math inline">\((0,\infty)\)</span> because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the <a href="#regression-priors.section">regression priors section</a> for recommendations on priors for regression coefficients and scales.</p>
+<section id="optimizing-the-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="optimizing-the-model">Optimizing the model</h4>
+<p>Where possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.</p>
+<p>The first optimization vectorizes the for-loop over <code>D</code> as</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a>mu ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
+<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (l <span class="cf">in</span> <span class="dv">1</span>:L) {</span>
+<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a>  beta[l] ~ normal(mu, sigma);</span>
+<span id="cb27-4"><a href="#cb27-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The declaration of <code>beta</code> as an array of vectors means that the expression <code>beta[l]</code> denotes a vector. Although <code>beta</code> could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the <a href="../stan-users-guide/matrices-arrays.html#indexing-efficiency.section">indexing efficiency section</a> for more information on the efficiency tradeoffs among arrays, vectors, and matrices.</p>
+<p>This model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Unlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_beta_ll;</span>
+<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a>    x_beta_ll[n] = x[n] * beta[ll[n]];</span>
+<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(x_beta_ll);</span>
+<span id="cb29-7"><a href="#cb29-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The brackets introduce a new scope for the local variable <code>x_beta_ll</code>; alternatively, the variable may be declared at the top of the model block.</p>
+<p>In some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.</p>
+</section>
+</section>
+<section id="hierarchical-priors.section" class="level2">
+<h2 class="anchored" data-anchor-id="hierarchical-priors.section">Hierarchical priors</h2>
+<p>Priors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.</p>
+<section id="boundary-avoiding-priors-for-mle-in-hierarchical-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="boundary-avoiding-priors-for-mle-in-hierarchical-models">Boundary-avoiding priors for MLE in hierarchical models</h3>
+<p>The fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of <span class="math inline">\(y_n \in
+\mathbb{R}\)</span> on <span class="math inline">\(x_n \in \mathbb{R}^K\)</span>, formulated as <span class="math display">\[\begin{align*}
+y_n     &amp; \sim \textsf{normal}(x_n \beta, \sigma) \\
+\beta_k &amp; \sim \textsf{normal}(0,\tau) \\
+\tau    &amp; \sim \textsf{Cauchy}(0,2.5) \\
+\end{align*}\]</span></p>
+<p>In this case, as <span class="math inline">\(\tau \rightarrow 0\)</span> and <span class="math inline">\(\beta_k \rightarrow 0\)</span>, the posterior density <span class="math display">\[ p(\beta,\tau,\sigma|y,x) \propto p(y|x,\beta,\tau,\sigma) \]</span> grows without bound. See the <a href="../stan-users-guide/efficiency-tuning.html#funnel.figure">Neal’s funnel density</a>, which has similar behavior.</p>
+<p>There is obviously no MLE estimate for <span class="math inline">\(\beta,\tau,\sigma\)</span> in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by <span class="citation" data-cites="ChungEtAl:2013">Chung et al. (<a href="#ref-ChungEtAl:2013" role="doc-biblioref">2013</a>)</span> is to use a gamma distribution as a prior, such as <span class="math display">\[
+\sigma \sim \textsf{Gamma}(2, 1/A),
+\]</span> for a reasonably large value of <span class="math inline">\(A\)</span>, such as <span class="math inline">\(A = 10\)</span>.</p>
+</section>
+</section>
+<section id="item-response-models.section" class="level2">
+<h2 class="anchored" data-anchor-id="item-response-models.section">Item-response theory models</h2>
+<p>Item-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see <span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007</a>, pps. 314–320)</span> for a textbook introduction to hierarchical IRT models and <span class="citation" data-cites="Curtis:2010">Curtis (<a href="#ref-Curtis:2010" role="doc-biblioref">2010</a>)</span> for encodings of a range of IRT models in BUGS.</p>
+<section id="data-declaration-with-missingness" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-declaration-with-missingness">Data declaration with missingness</h3>
+<p>The data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.</p>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;                     <span class="co">// number of students</span></span>
+<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;                     <span class="co">// number of questions</span></span>
+<span id="cb30-4"><a href="#cb30-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;                     <span class="co">// number of observations</span></span>
+<span id="cb30-5"><a href="#cb30-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;  <span class="co">// student for observation n</span></span>
+<span id="cb30-6"><a href="#cb30-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;  <span class="co">// question for observation n</span></span>
+<span id="cb30-7"><a href="#cb30-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;   <span class="co">// correctness for observation n</span></span>
+<span id="cb30-8"><a href="#cb30-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This declares a total of <code>N</code> student-question pairs in the data set, where each <code>n</code> in <code>1:N</code> indexes a binary observation <code>y[n]</code> of the correctness of the answer of student <code>jj[n]</code> on question <code>kk[n]</code>.</p>
+<p>The prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.</p>
+</section>
+<section id="pl-rasch-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="pl-rasch-model">1PL (Rasch) model</h3>
+<p>The 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).</p>
+<p>The model parameters are declared as follows.</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta;            <span class="co">// mean student ability</span></span>
+<span id="cb31-3"><a href="#cb31-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> alpha;   <span class="co">// ability of student j - mean ability</span></span>
+<span id="cb31-4"><a href="#cb31-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">real</span> beta;    <span class="co">// difficulty of question k</span></span>
+<span id="cb31-5"><a href="#cb31-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameter <code>alpha[J]</code> is the ability coefficient for student <code>j</code> and <code>beta[k]</code> is the difficulty coefficient for question <code>k</code>. The non-standard parameterization used here also includes an intercept term <code>delta</code>, which represents the average student’s response to the average question.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
+<p>The model itself is as follows.</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();         <span class="co">// informative true prior</span></span>
+<span id="cb32-3"><a href="#cb32-3" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();          <span class="co">// informative true prior</span></span>
+<span id="cb32-4"><a href="#cb32-4" aria-hidden="true" tabindex="-1"></a>  delta ~ normal(<span class="fl">0.75</span>, <span class="dv">1</span>);      <span class="co">// informative true prior</span></span>
+<span id="cb32-5"><a href="#cb32-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb32-6"><a href="#cb32-6" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);</span>
+<span id="cb32-7"><a href="#cb32-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb32-8"><a href="#cb32-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This model uses the logit-parameterized Bernoulli distribution, where <span class="math display">\[
+\texttt{bernoulli}\mathtt{\_}\texttt{logit}\left(y \mid \alpha\right)
+=
+\texttt{bernoulli}\left(y \mid \operatorname{logit}^{-1}(\alpha)\right).
+\]</span></p>
+<p>The key to understanding it is the term inside the <code>bernoulli_logit</code> distribution, from which it follows that <span class="math display">\[
+\Pr[y_n = 1] = \operatorname{logit}^{-1}\left(\alpha_{jj[n]} - \beta_{kk[n]}
++ \delta\right).
+\]</span></p>
+<p>The model suffers from additive identifiability issues without the priors. For example, adding a term <span class="math inline">\(\xi\)</span> to each <span class="math inline">\(\alpha_j\)</span> and <span class="math inline">\(\beta_k\)</span> results in the same predictions. The use of priors for <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> located at 0 identifies the parameters; see <span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007</a>)</span> for a discussion of identifiability issues and alternative approaches to identification.</p>
+<p>For testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.</p>
+</section>
+<section id="multilevel-2pl-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multilevel-2pl-model">Multilevel 2PL model</h3>
+<p>The simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.</p>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb33-2"><a href="#cb33-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;                <span class="co">// mean question difficulty</span></span>
+<span id="cb33-3"><a href="#cb33-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[J] alpha;             <span class="co">// ability for j - mean</span></span>
+<span id="cb33-4"><a href="#cb33-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;              <span class="co">// difficulty for k</span></span>
+<span id="cb33-5"><a href="#cb33-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] gamma;    <span class="co">// discrimination of k</span></span>
+<span id="cb33-6"><a href="#cb33-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;    <span class="co">// scale of difficulties</span></span>
+<span id="cb33-7"><a href="#cb33-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_gamma;   <span class="co">// scale of log discrimination</span></span>
+<span id="cb33-8"><a href="#cb33-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameters should be clearer after the model definition.</p>
+<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb34-2"><a href="#cb34-2" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb34-3"><a href="#cb34-3" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, sigma_beta);</span>
+<span id="cb34-4"><a href="#cb34-4" aria-hidden="true" tabindex="-1"></a>  gamma ~ lognormal(<span class="dv">0</span>, sigma_gamma);</span>
+<span id="cb34-5"><a href="#cb34-5" aria-hidden="true" tabindex="-1"></a>  mu_beta ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb34-6"><a href="#cb34-6" aria-hidden="true" tabindex="-1"></a>  sigma_beta ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb34-7"><a href="#cb34-7" aria-hidden="true" tabindex="-1"></a>  sigma_gamma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb34-8"><a href="#cb34-8" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));</span>
+<span id="cb34-9"><a href="#cb34-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>std_normal</code> function is used here, defined by <span class="math display">\[
+\texttt{std}\mathtt{\_}\texttt{normal}(y)
+=
+\textsf{normal}\left(y \mid 0, 1\right).
+\]</span></p>
+<p>The distribution statement is also vectorized using elementwise multiplication; it is equivalent to</p>
+<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb35-2"><a href="#cb35-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(gamma[kk[n]]</span>
+<span id="cb35-3"><a href="#cb35-3" aria-hidden="true" tabindex="-1"></a>                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));</span>
+<span id="cb35-4"><a href="#cb35-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The 2PL model is similar to the 1PL model, with the additional parameter <code>gamma[k]</code> modeling how discriminative question <code>k</code> is. If <code>gamma[k]</code> is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter <code>gamma[k]</code> is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.</p>
+<p>The model is parameterized here with student abilities <code>alpha</code> being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the <a href="../stan-users-guide/problematic-posteriors.html">problematic posteriors chapter</a> for further discussion of identifiability. The difficulty and discrimination parameters <code>beta</code> and <code>gamma</code> then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as</p>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a>beta ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb36-2"><a href="#cb36-2" aria-hidden="true" tabindex="-1"></a>gamma ~ lognormal(<span class="dv">0</span>, <span class="dv">2</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The point is that the <code>alpha</code> determines the scale and location and <code>beta</code> and <code>gamma</code> are allowed to float.</p>
+<p>The <code>beta</code> parameter is here given a non-centered parameterization, with parameter <code>mu_beta</code> serving as the mean <code>beta</code> location. An alternative would’ve been to take:</p>
+<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a>beta ~ normal(mu_beta, sigma_beta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and</p>
+<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Non-centered parameterizations tend to be more efficient in hierarchical models; see the <a href="../stan-users-guide/efficiency-tuning.html#reparameterization.section">reparameterization section</a> for more information on non-centered reparameterizations.</p>
+<p>The intercept term <code>mu_beta</code> can’t itself be modeled hierarchically, so it is given a weakly informative <span class="math inline">\(\textsf{Cauchy}(0,5)\)</span> prior. Similarly, the scale terms, <code>sigma_beta</code>, and <code>sigma_gamma</code>, are given half-Cauchy priors. As mentioned earlier, the scale and location for <code>alpha</code> are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to <span class="math inline">\((0,\infty)\)</span> by their declarations.</p>
+</section>
+</section>
+<section id="priors-for-identification.section" class="level2">
+<h2 class="anchored" data-anchor-id="priors-for-identification.section">Priors for identifiability</h2>
+<section id="location-and-scale-invariance" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="location-and-scale-invariance">Location and scale invariance</h3>
+<p>One application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the <a href="../stan-users-guide/problematic-posteriors.html#collinearity.section">collinearity section</a> for a simple example of the problems this poses for estimation.</p>
+<p>The non-identifiability is resolved by providing a standard normal (i.e., <span class="math inline">\(\textsf{normal}(0,1)\)</span>) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.</p>
+</section>
+<section id="collinearity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="collinearity">Collinearity</h3>
+<p>Another case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.</p>
+</section>
+<section id="separability" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="separability">Separability</h3>
+<p>In a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.</p>
+<p>Similar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.</p>
+</section>
+</section>
+<section id="multivariate-hierarchical-priors.section" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-hierarchical-priors.section">Multivariate priors for hierarchical models</h2>
+<p>In hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.</p>
+<p>As an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.</p>
+<section id="multivariate-regression-example" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multivariate-regression-example">Multivariate regression example</h3>
+<p><span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007, chap. 13</a>, Chapter 17)</span> provide a discussion of a hierarchical model with <span class="math inline">\(N\)</span> individuals organized into <span class="math inline">\(J\)</span> groups. Each individual has a predictor row vector <span class="math inline">\(x_n\)</span> of size <span class="math inline">\(K\)</span>; to unify the notation, they assume that <span class="math inline">\(x_{n,1} = 1\)</span> is a fixed “intercept” predictor. To encode group membership, they assume individual <span class="math inline">\(n\)</span> belongs to group <span class="math inline">\(jj[n] \in \{ 1, \dotsc, J \}\)</span>. Each individual <span class="math inline">\(n\)</span> also has an observed outcome <span class="math inline">\(y_n\)</span> taking on real values.</p>
+<section id="data-model" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="data-model">Data model</h4>
+<p>The model is a linear regression with slope and intercept coefficients varying by group, so that <span class="math inline">\(\beta_j\)</span> is the coefficient <span class="math inline">\(K\)</span>-vector for group <span class="math inline">\(j\)</span>. The data model for individual <span class="math inline">\(n\)</span> is then just <span class="math display">\[
+y_n \sim \textsf{normal}(x_n \, \beta_{jj[n]}, \, \sigma)
+\quad\text{for}\quad n \in \{ 1, \dotsc, N \}.
+\]</span></p>
+</section>
+<section id="coefficient-prior" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="coefficient-prior">Coefficient prior</h4>
+<p>Gelman and Hill model the coefficient vectors <span class="math inline">\(\beta_j\)</span> as being drawn from a multivariate distribution with mean vector <span class="math inline">\(\mu\)</span> and covariance matrix <span class="math inline">\(\Sigma\)</span>, <span class="math display">\[
+\beta_j \sim \textsf{multivariate normal}(\mu_j, \, \Sigma)
+\quad\text{for}\quad j \in \{ 1, \dotsc, J \}.
+\]</span></p>
+<p>Below, we discuss the full model of Gelman and Hill, which uses group-level predictors to model <span class="math inline">\(\mu\)</span>; for now, we assume <span class="math inline">\(\mu\)</span> is a simple vector parameter.</p>
+</section>
+<section id="hyperpriors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="hyperpriors">Hyperpriors</h4>
+<p>For hierarchical modeling, the group-level mean vector <span class="math inline">\(\mu\)</span> and covariance matrix <span class="math inline">\(\Sigma\)</span> must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as <span class="math display">\[
+\mu_j \sim \textsf{normal}(0,5).
+\]</span> If more is known about the expected coefficient values <span class="math inline">\(\beta_{j, k}\)</span>, this information can be incorporated into the prior for <span class="math inline">\(\mu_j\)</span>.</p>
+<p>For the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling</p>
+<p>In Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define <span class="math display">\[
+\Sigma = \texttt{diag}\mathtt{\_}\texttt{matrix}(\tau) \times \Omega \times \texttt{diag}\mathtt{\_}\texttt{matrix}(\tau),
+\]</span> where <span class="math inline">\(\Omega\)</span> is a correlation matrix and <span class="math inline">\(\tau\)</span> is the vector of coefficient scales. This mapping from scale vector <span class="math inline">\(\tau\)</span> and correlation matrix <span class="math inline">\(\Omega\)</span> can be inverted, using <span class="math display">\[
+\tau_k = \sqrt{\Sigma_{k,k}}
+\quad\textsf{and}\quad
+\Omega_{i, j} = \frac{\Sigma_{i, j}}{\tau_i \, \tau_j}.
+\]</span></p>
+<p>The components of the scale vector <span class="math inline">\(\tau\)</span> can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as <span class="math display">\[
+\tau_k \sim \textsf{Cauchy}(0, 2.5)
+\quad\text{for}\quad k \in \{ 1, \dotsc, K \}
+\quad\text{constrained\ by}\quad \tau_k &gt; 0.
+\]</span> As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for <span class="math inline">\(\tau\)</span>. For large numbers of exchangeable coefficients, the components of <span class="math inline">\(\tau\)</span> itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.</p>
+<p>Our final recommendation is to give the correlation matrix <span class="math inline">\(\Omega\)</span> an LKJ prior with shape <span class="math inline">\(\eta \geq 1\)</span>,<a href="#fn5" class="footnote-ref" id="fnref5" role="doc-noteref"><sup>5</sup></a></p>
+<p><span class="math display">\[
+\Omega \sim \textsf{LKJCorr}(\eta).
+\]</span></p>
+<p>The LKJ correlation distribution is defined by <span class="math display">\[
+\textsf{LKJCorr}\left(\Sigma \mid \eta\right)
+\propto
+\operatorname{det}\left(\Sigma\right)^{\eta - 1}.
+\]</span></p>
+<p>The basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For <span class="math inline">\(\eta = 1\)</span>, the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.</p>
+<p>For <span class="math inline">\(\eta &gt; 1\)</span>, the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For <span class="math inline">\(\eta &lt; 1\)</span>, it increasingly concentrates mass in the other direction, i.e., favoring more correlation.</p>
+<p>The LKJ prior may thus be used to control the expected amount of correlation among the parameters <span class="math inline">\(\beta_j\)</span>. For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see <span class="citation" data-cites="barnard-mcculloch-meng:2000">Barnard, McCulloch, and Meng (<a href="#ref-barnard-mcculloch-meng:2000" role="doc-biblioref">2000</a>)</span>.</p>
+</section>
+<section id="group-level-predictors-for-prior-mean" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="group-level-predictors-for-prior-mean">Group-level predictors for prior mean</h4>
+<p>To complete Gelman and Hill’s model, suppose each group <span class="math inline">\(j \in \{ 1, \dotsc, J \}\)</span> is supplied with an <span class="math inline">\(L\)</span>-dimensional row-vector of group-level predictors <span class="math inline">\(u_j\)</span>. The prior mean for the <span class="math inline">\(\beta_j\)</span> can then itself be modeled as a regression, using an <span class="math inline">\(L\)</span>-dimensional coefficient vector <span class="math inline">\(\gamma\)</span>. The prior for the group-level coefficients then becomes <span class="math display">\[
+\beta_j \sim \textsf{multivariate normal}(u_j \, \gamma, \Sigma)
+\]</span></p>
+<p>The group-level coefficients <span class="math inline">\(\gamma\)</span> may themselves be given independent weakly informative priors, such as <span class="math display">\[
+\gamma_l \sim \textsf{normal}(0,5).
+\]</span> As usual, information about the group-level means should be incorporated into this prior.</p>
+</section>
+<section id="coding-the-model-in-stan" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="coding-the-model-in-stan">Coding the model in Stan</h4>
+<p>The Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.</p>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb39-2"><a href="#cb39-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;              <span class="co">// num individuals</span></span>
+<span id="cb39-3"><a href="#cb39-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;              <span class="co">// num ind predictors</span></span>
+<span id="cb39-4"><a href="#cb39-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;              <span class="co">// num groups</span></span>
+<span id="cb39-5"><a href="#cb39-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; L;              <span class="co">// num group predictors</span></span>
+<span id="cb39-6"><a href="#cb39-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;  <span class="co">// group for individual</span></span>
+<span id="cb39-7"><a href="#cb39-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;              <span class="co">// individual predictors</span></span>
+<span id="cb39-8"><a href="#cb39-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">row_vector</span>[L] u;    <span class="co">// group predictors</span></span>
+<span id="cb39-9"><a href="#cb39-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;                 <span class="co">// outcomes</span></span>
+<span id="cb39-10"><a href="#cb39-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb39-11"><a href="#cb39-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb39-12"><a href="#cb39-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[K] Omega;        <span class="co">// prior correlation</span></span>
+<span id="cb39-13"><a href="#cb39-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] tau;      <span class="co">// prior scale</span></span>
+<span id="cb39-14"><a href="#cb39-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[L, K] gamma;          <span class="co">// group coeffs</span></span>
+<span id="cb39-15"><a href="#cb39-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">vector</span>[K] beta;     <span class="co">// indiv coeffs by group</span></span>
+<span id="cb39-16"><a href="#cb39-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;         <span class="co">// prediction error scale</span></span>
+<span id="cb39-17"><a href="#cb39-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb39-18"><a href="#cb39-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb39-19"><a href="#cb39-19" aria-hidden="true" tabindex="-1"></a>  tau ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb39-20"><a href="#cb39-20" aria-hidden="true" tabindex="-1"></a>  Omega ~ lkj_corr(<span class="dv">2</span>);</span>
+<span id="cb39-21"><a href="#cb39-21" aria-hidden="true" tabindex="-1"></a>  to_vector(gamma) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb39-22"><a href="#cb39-22" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb39-23"><a href="#cb39-23" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[J] <span class="dt">row_vector</span>[K] u_gamma;</span>
+<span id="cb39-24"><a href="#cb39-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb39-25"><a href="#cb39-25" aria-hidden="true" tabindex="-1"></a>      u_gamma[j] = u[j] * gamma;</span>
+<span id="cb39-26"><a href="#cb39-26" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb39-27"><a href="#cb39-27" aria-hidden="true" tabindex="-1"></a>    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));</span>
+<span id="cb39-28"><a href="#cb39-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb39-29"><a href="#cb39-29" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb39-30"><a href="#cb39-30" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(x[n] * beta[jj[n]], sigma);</span>
+<span id="cb39-31"><a href="#cb39-31" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb39-32"><a href="#cb39-32" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix <code>Omega</code> and scale vector <code>tau</code> are more natural to inspect in the output; to output <code>Sigma</code>, define it as a transformed parameter. The function <code>quad_form_diag</code> is defined so that <code>quad_form_diag(Sigma, tau)</code> is equivalent to <code>diag_matrix(tau) * Sigma * diag_matrix(tau)</code>, where <code>diag_matrix(tau)</code> returns the matrix with <code>tau</code> on the diagonal and zeroes off diagonal; the version using <code>quad_form_diag</code> should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.</p>
+</section>
+<section id="optimization-through-vectorization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="optimization-through-vectorization">Optimization through vectorization</h4>
+<p>The code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:</p>
+<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb40-2"><a href="#cb40-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(x[n] * beta[jj[n]], sigma);</span>
+<span id="cb40-3"><a href="#cb40-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>with the vectorized distribution statement:</p>
+<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb41-2"><a href="#cb41-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_beta_jj;</span>
+<span id="cb41-3"><a href="#cb41-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb41-4"><a href="#cb41-4" aria-hidden="true" tabindex="-1"></a>    x_beta_jj[n] = x[n] * beta[jj[n]];</span>
+<span id="cb41-5"><a href="#cb41-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb41-6"><a href="#cb41-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x_beta_jj, sigma);</span>
+<span id="cb41-7"><a href="#cb41-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The outer brackets create a local scope in which to define the variable <code>x_beta_jj</code>, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.</p>
+<p>Although it is tempting to redeclare <code>beta</code> and include a revised model block distribution statement,</p>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb42-2"><a href="#cb42-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[J, K] beta;</span>
+<span id="cb42-3"><a href="#cb42-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb42-4"><a href="#cb42-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb42-5"><a href="#cb42-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb42-6"><a href="#cb42-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(rows_dot_product(x, beta[jj]), sigma);</span>
+<span id="cb42-7"><a href="#cb42-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb42-8"><a href="#cb42-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>this fails because it breaks the vectorization for <code>beta</code>,<a href="#fn6" class="footnote-ref" id="fnref6" role="doc-noteref"><sup>6</sup></a></p>
+<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a>beta ~ multi_normal(...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>which requires <code>beta</code> to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because <code>rows_dot_product</code> cannot do much optimization in and of itself because there are no shared computations.</p>
+<p>The code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.</p>
+<div class="sourceCode" id="cb44"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb44-1"><a href="#cb44-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb44-2"><a href="#cb44-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] Sigma_beta;</span>
+<span id="cb44-3"><a href="#cb44-3" aria-hidden="true" tabindex="-1"></a>  Sigma_beta = quad_form_diag(Omega, tau);</span>
+<span id="cb44-4"><a href="#cb44-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb44-5"><a href="#cb44-5" aria-hidden="true" tabindex="-1"></a>    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);</span>
+<span id="cb44-6"><a href="#cb44-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb44-7"><a href="#cb44-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this example, the covariance matrix <code>Sigma_beta</code> is defined as a local variable so as not to have to repeat the quadratic form computation <span class="math inline">\(J\)</span> times. This vectorization can be combined with the Cholesky-factor optimization in the next section.</p>
+</section>
+<section id="optimization-through-cholesky-factorization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="optimization-through-cholesky-factorization">Optimization through Cholesky factorization</h4>
+<p>The multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.</p>
+<div class="sourceCode" id="cb45"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb45-1"><a href="#cb45-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb45-2"><a href="#cb45-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[L, J] u;              <span class="co">// group predictors transposed</span></span>
+<span id="cb45-3"><a href="#cb45-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb45-4"><a href="#cb45-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-5"><a href="#cb45-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb45-6"><a href="#cb45-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] z;</span>
+<span id="cb45-7"><a href="#cb45-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
+<span id="cb45-8"><a href="#cb45-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, L] gamma;</span>
+<span id="cb45-9"><a href="#cb45-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb45-10"><a href="#cb45-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-11"><a href="#cb45-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb45-12"><a href="#cb45-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
+<span id="cb45-13"><a href="#cb45-13" aria-hidden="true" tabindex="-1"></a>  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;</span>
+<span id="cb45-14"><a href="#cb45-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-15"><a href="#cb45-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb45-16"><a href="#cb45-16" aria-hidden="true" tabindex="-1"></a>  to_vector(z) ~ std_normal();</span>
+<span id="cb45-17"><a href="#cb45-17" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
+<span id="cb45-18"><a href="#cb45-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb45-19"><a href="#cb45-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data variable <code>u</code> was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with <code>gamma</code> and <code>beta</code>, to minimize the number of transposition operations. The new parameter <code>L_Omega</code> is the Cholesky factor of the original correlation matrix <code>Omega</code>, so that</p>
+<pre><code>Omega = L_Omega * L_Omega'</code></pre>
+<p>The prior scale vector <code>tau</code> is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,</p>
+<pre><code>Sigma_beta
+  = quad_form_diag(Omega, tau)
+  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'</code></pre>
+<p>where the diagonal pre-multiply compound operation is defined by</p>
+<pre><code>diag_pre_multiply(a, b) = diag_matrix(a) * b</code></pre>
+<p>The new variable <code>z</code> is declared as a matrix, the entries of which are given independent standard normal priors; the <code>to_vector</code> operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of <code>z</code> being a <span class="math inline">\(K\)</span>-variate normal random vector with the identity as covariance matrix. Therefore, multiplying <code>z</code> by the Cholesky factor of the covariance matrix and adding the mean <code>(u * gamma)'</code> produces a <code>beta</code> distributed as in the original model, where the variance is, letting <span class="math inline">\(L = \mathrm{diag}(\tau)\,\Omega_L\)</span>,</p>
+<p><span class="math display">\[
+\begin{aligned}
+\mathbb{V}[\beta] &amp;= \mathbb{E}\big((L \, z)(L \, z)^\top) \\
+&amp;= \mathbb{E}\big((L \, z \, z^\top \, L^\top) \\
+&amp;= L \, \mathbb{E}(z \, z^\top) \, L^\top \\
+&amp;= L \, L^\top =(\mathrm{diag}(\tau)\,\Omega_L)\,(\mathrm{diag}(\tau)\,\Omega_L)^\top \\
+&amp;= \mathrm{diag}(\tau)\,\Omega\,\mathrm{diag}(\tau) \\
+&amp;= \Sigma.
+\end{aligned}
+\]</span> Where we have used the linearity of expectations (line 2 to 3), the definition of <span class="math inline">\(\Omega = \Omega_L \, \Omega_L^\top\)</span>, and the fact that <span class="math inline">\(\mathbb{E}(z \, z^\top) = I\)</span> since <span class="math inline">\(z \sim \mathcal{N}(0, I)\)</span>.</p>
+<p>Omitting the remaining data declarations, which are the same as before with the exception of <code>u</code>, the optimized model is as follows.</p>
+<div class="sourceCode" id="cb49"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb49-1"><a href="#cb49-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb49-2"><a href="#cb49-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] z;</span>
+<span id="cb49-3"><a href="#cb49-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
+<span id="cb49-4"><a href="#cb49-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=pi() / <span class="dv">2</span>&gt;[K] tau_unif;  <span class="co">// prior scale</span></span>
+<span id="cb49-5"><a href="#cb49-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, L] gamma;                        <span class="co">// group coeffs</span></span>
+<span id="cb49-6"><a href="#cb49-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;                       <span class="co">// prediction error scale</span></span>
+<span id="cb49-7"><a href="#cb49-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb49-8"><a href="#cb49-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb49-9"><a href="#cb49-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] tau = <span class="fl">2.5</span> * tan(tau_unif);</span>
+<span id="cb49-10"><a href="#cb49-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;</span>
+<span id="cb49-11"><a href="#cb49-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb49-12"><a href="#cb49-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb49-13"><a href="#cb49-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu;</span>
+<span id="cb49-14"><a href="#cb49-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span>(n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb49-15"><a href="#cb49-15" aria-hidden="true" tabindex="-1"></a>    mu[n] = x[n, ] * beta[, jj[n]];</span>
+<span id="cb49-16"><a href="#cb49-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb49-17"><a href="#cb49-17" aria-hidden="true" tabindex="-1"></a>  to_vector(z) ~ std_normal();</span>
+<span id="cb49-18"><a href="#cb49-18" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
+<span id="cb49-19"><a href="#cb49-19" aria-hidden="true" tabindex="-1"></a>  to_vector(gamma) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb49-20"><a href="#cb49-20" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
+<span id="cb49-21"><a href="#cb49-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This model also reparameterizes the prior scale <code>tau</code> to avoid potential problems with the heavy tails of the Cauchy distribution. The statement <code>tau_unif ~ uniform(0, pi() / 2)</code> can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.</p>
+</section>
+</section>
+</section>
+<section id="prediction-forecasting-and-backcasting" class="level2">
+<h2 class="anchored" data-anchor-id="prediction-forecasting-and-backcasting">Prediction, forecasting, and backcasting</h2>
+<p>Stan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).</p>
+<section id="programming-predictions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="programming-predictions">Programming predictions</h3>
+<p>As a simple example, the following linear regression provides the same setup for estimating the coefficients <code>beta</code> as in our very first example, using <code>y</code> for the <code>N</code> observations and <code>x</code> for the <code>N</code> predictor vectors. The model parameters and model for observations are exactly the same as before.</p>
+<p>To make predictions, we need to be given the number of predictions, <code>N_new</code>, and their predictor matrix, <code>x_new</code>. The predictions themselves are modeled as a parameter <code>y_new</code>. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector <code>y_new</code> and prediction matrix <code>x_new</code>.</p>
+<div class="sourceCode" id="cb50"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb50-1"><a href="#cb50-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb50-2"><a href="#cb50-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb50-3"><a href="#cb50-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb50-4"><a href="#cb50-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
+<span id="cb50-5"><a href="#cb50-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb50-6"><a href="#cb50-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb50-7"><a href="#cb50-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_new;</span>
+<span id="cb50-8"><a href="#cb50-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N_new, K] x_new;</span>
+<span id="cb50-9"><a href="#cb50-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb50-10"><a href="#cb50-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb50-11"><a href="#cb50-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb50-12"><a href="#cb50-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb50-13"><a href="#cb50-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb50-14"><a href="#cb50-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_new] y_new;                  <span class="co">// predictions</span></span>
+<span id="cb50-15"><a href="#cb50-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb50-16"><a href="#cb50-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb50-17"><a href="#cb50-17" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, sigma);          <span class="co">// observed model</span></span>
+<span id="cb50-18"><a href="#cb50-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb50-19"><a href="#cb50-19" aria-hidden="true" tabindex="-1"></a>  y_new ~ normal(x_new * beta, sigma);  <span class="co">// prediction model</span></span>
+<span id="cb50-20"><a href="#cb50-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="predictions-as-generated-quantities" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="predictions-as-generated-quantities">Predictions as generated quantities</h3>
+<p>Where possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.</p>
+<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...data as above...</span></span>
+<span id="cb51-2"><a href="#cb51-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb51-3"><a href="#cb51-3" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb51-4"><a href="#cb51-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb51-5"><a href="#cb51-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb51-6"><a href="#cb51-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb51-7"><a href="#cb51-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb51-8"><a href="#cb51-8" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, sigma);</span>
+<span id="cb51-9"><a href="#cb51-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb51-10"><a href="#cb51-10" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb51-11"><a href="#cb51-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_new] y_new;</span>
+<span id="cb51-12"><a href="#cb51-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_new) {</span>
+<span id="cb51-13"><a href="#cb51-13" aria-hidden="true" tabindex="-1"></a>    y_new[n] = normal_rng(x_new[n] * beta, sigma);</span>
+<span id="cb51-14"><a href="#cb51-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb51-15"><a href="#cb51-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now the data are just as before, but the parameter <code>y_new</code> is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.</p>
+<section id="overflow-in-generated-quantities" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="overflow-in-generated-quantities">Overflow in generated quantities</h4>
+<p>It is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of <span class="math inline">\(2^{31}\)</span>, must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by <span class="math inline">\(2^{31} - 1\)</span> in Stan.</p>
+</section>
+</section>
+</section>
+<section id="multivariate-outcomes" class="level2">
+<h2 class="anchored" data-anchor-id="multivariate-outcomes">Multivariate outcomes</h2>
+<p>Most regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.</p>
+<p>This section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.</p>
+<section id="seemingly-unrelated-regressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="seemingly-unrelated-regressions">Seemingly unrelated regressions</h3>
+<p>The first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors <span class="citation" data-cites="Zellner:1962 Greene:2011">(<a href="#ref-Zellner:1962" role="doc-biblioref">Zellner 1962</a>; <a href="#ref-Greene:2011" role="doc-biblioref">Greene 2011</a>)</span>.</p>
+<p>The model is easy to write down as a regression, <span class="math display">\[\begin{align*}
+y_n        &amp;=  x_n \, \beta + \epsilon_n \\
+\epsilon_n &amp;\sim \textsf{multivariate normal}(0, \Sigma)
+\end{align*}\]</span></p>
+<p>where <span class="math inline">\(x_n\)</span> is a <span class="math inline">\(J\)</span>-row-vector of predictors (<span class="math inline">\(x\)</span> is an <span class="math inline">\((N \times
+J)\)</span> matrix), <span class="math inline">\(y_n\)</span> is a <span class="math inline">\(K\)</span>-vector of observations, <span class="math inline">\(\beta\)</span> is a <span class="math inline">\((K
+\times J)\)</span> matrix of regression coefficients (vector <span class="math inline">\(\beta_k\)</span> holds coefficients for outcome <span class="math inline">\(k\)</span>), and <span class="math inline">\(\Sigma\)</span> is covariance matrix governing the error. As usual, the intercept can be rolled into <span class="math inline">\(x\)</span> as a column of ones.</p>
+<p>The basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).</p>
+<div class="sourceCode" id="cb52"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb52-1"><a href="#cb52-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb52-2"><a href="#cb52-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb52-3"><a href="#cb52-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;</span>
+<span id="cb52-4"><a href="#cb52-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb52-5"><a href="#cb52-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[J] x;</span>
+<span id="cb52-6"><a href="#cb52-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] y;</span>
+<span id="cb52-7"><a href="#cb52-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb52-8"><a href="#cb52-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb52-9"><a href="#cb52-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
+<span id="cb52-10"><a href="#cb52-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb52-11"><a href="#cb52-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb52-12"><a href="#cb52-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb52-13"><a href="#cb52-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] mu;</span>
+<span id="cb52-14"><a href="#cb52-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb52-15"><a href="#cb52-15" aria-hidden="true" tabindex="-1"></a>    mu[n] = beta * x[n];</span>
+<span id="cb52-16"><a href="#cb52-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb52-17"><a href="#cb52-17" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, Sigma);</span>
+<span id="cb52-18"><a href="#cb52-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.</p>
+<p>Following the advice in the <a href="#multivariate-hierarchical-priors.section">multivariate hierarchical priors section</a>, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.</p>
+<div class="sourceCode" id="cb53"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb53-1"><a href="#cb53-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb53-2"><a href="#cb53-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb53-3"><a href="#cb53-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
+<span id="cb53-4"><a href="#cb53-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
+<span id="cb53-5"><a href="#cb53-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] L_sigma;</span>
+<span id="cb53-6"><a href="#cb53-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb53-7"><a href="#cb53-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb53-8"><a href="#cb53-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] mu;</span>
+<span id="cb53-9"><a href="#cb53-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L_Sigma;</span>
+<span id="cb53-10"><a href="#cb53-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-11"><a href="#cb53-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb53-12"><a href="#cb53-12" aria-hidden="true" tabindex="-1"></a>    mu[n] = beta * x[n];</span>
+<span id="cb53-13"><a href="#cb53-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-14"><a href="#cb53-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb53-15"><a href="#cb53-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-16"><a href="#cb53-16" aria-hidden="true" tabindex="-1"></a>  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);</span>
+<span id="cb53-17"><a href="#cb53-17" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-18"><a href="#cb53-18" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb53-19"><a href="#cb53-19" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">4</span>);</span>
+<span id="cb53-20"><a href="#cb53-20" aria-hidden="true" tabindex="-1"></a>  L_sigma ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb53-21"><a href="#cb53-21" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-22"><a href="#cb53-22" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal_cholesky(mu, L_Sigma);</span>
+<span id="cb53-23"><a href="#cb53-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix <code>beta</code> to a vector.</p>
+<p>If required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.</p>
+</section>
+<section id="multivariate-probit-regression" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multivariate-probit-regression">Multivariate probit regression</h3>
+<p>The multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.</p>
+<p>The observations <span class="math inline">\(y_n\)</span> are <span class="math inline">\(D\)</span>-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations <span class="math inline">\(y_n\)</span> are based on latent values <span class="math inline">\(z_n\)</span> drawn from a seemingly unrelated regression model (see the previous section), <span class="math display">\[\begin{align*}
+z_n        &amp;=     x_n \, \beta + \epsilon_n \\
+\epsilon_n &amp;\sim  \textsf{multivariate normal}(0, \Sigma)
+\end{align*}\]</span></p>
+<p>These are then put through the step function to produce a <span class="math inline">\(K\)</span>-vector <span class="math inline">\(z_n\)</span> of boolean values with elements defined by <span class="math display">\[
+y_{n, k} = \operatorname{I}\left(z_{n, k} &gt; 0\right),
+\]</span> where <span class="math inline">\(\operatorname{I}()\)</span> is the indicator function taking the value 1 if its argument is true and 0 otherwise.</p>
+<p>Unlike in the seemingly unrelated regressions case, here the covariance matrix <span class="math inline">\(\Sigma\)</span> has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see <span class="citation" data-cites="Greene:2011">Greene (<a href="#ref-Greene:2011" role="doc-biblioref">2011</a>)</span>).</p>
+<p>Multivariate probit regression can be coded in Stan using the trick introduced by <span class="citation" data-cites="AlbertChib:1993">Albert and Chib (<a href="#ref-AlbertChib:1993" role="doc-biblioref">1993</a>)</span>, where the underlying continuous value vectors <span class="math inline">\(y_n\)</span> are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector <span class="math inline">\(z\)</span> in two parts, based on whether the corresponding value of <span class="math inline">\(y\)</span> is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.</p>
+<p>First, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in <span class="math inline">\(y\)</span>.</p>
+<div class="sourceCode" id="cb54"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb54-1"><a href="#cb54-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb54-2"><a href="#cb54-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> sum2d(<span class="dt">array</span>[,] <span class="dt">int</span> a) {</span>
+<span id="cb54-3"><a href="#cb54-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> s = <span class="dv">0</span>;</span>
+<span id="cb54-4"><a href="#cb54-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:size(a)) {</span>
+<span id="cb54-5"><a href="#cb54-5" aria-hidden="true" tabindex="-1"></a>      s += sum(a[i]);</span>
+<span id="cb54-6"><a href="#cb54-6" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb54-7"><a href="#cb54-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> s;</span>
+<span id="cb54-8"><a href="#cb54-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb54-9"><a href="#cb54-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.</p>
+<p>The data declaration block is much like for the seemingly unrelated regressions, but the observations <code>y</code> are now integers constrained to be 0 or 1.</p>
+<div class="sourceCode" id="cb55"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb55-1"><a href="#cb55-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb55-2"><a href="#cb55-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb55-3"><a href="#cb55-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb55-4"><a href="#cb55-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb55-5"><a href="#cb55-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N, D] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb55-6"><a href="#cb55-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] x;</span>
+<span id="cb55-7"><a href="#cb55-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array <code>y</code> into positive and negative components, keeping track of indexes so that <code>z</code> can be easily reassembled in the transformed parameters block.</p>
+<div class="sourceCode" id="cb56"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb56-1"><a href="#cb56-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb56-2"><a href="#cb56-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_pos;</span>
+<span id="cb56-3"><a href="#cb56-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[sum2d(y)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; n_pos;</span>
+<span id="cb56-4"><a href="#cb56-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[size(n_pos)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=D&gt; d_pos;</span>
+<span id="cb56-5"><a href="#cb56-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_neg;</span>
+<span id="cb56-6"><a href="#cb56-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[(N * D) - size(n_pos)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; n_neg;</span>
+<span id="cb56-7"><a href="#cb56-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[size(n_neg)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=D&gt; d_neg;</span>
+<span id="cb56-8"><a href="#cb56-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb56-9"><a href="#cb56-9" aria-hidden="true" tabindex="-1"></a>  N_pos = size(n_pos);</span>
+<span id="cb56-10"><a href="#cb56-10" aria-hidden="true" tabindex="-1"></a>  N_neg = size(n_neg);</span>
+<span id="cb56-11"><a href="#cb56-11" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb56-12"><a href="#cb56-12" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> i;</span>
+<span id="cb56-13"><a href="#cb56-13" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> j;</span>
+<span id="cb56-14"><a href="#cb56-14" aria-hidden="true" tabindex="-1"></a>    i = <span class="dv">1</span>;</span>
+<span id="cb56-15"><a href="#cb56-15" aria-hidden="true" tabindex="-1"></a>    j = <span class="dv">1</span>;</span>
+<span id="cb56-16"><a href="#cb56-16" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb56-17"><a href="#cb56-17" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
+<span id="cb56-18"><a href="#cb56-18" aria-hidden="true" tabindex="-1"></a>        <span class="cf">if</span> (y[n, d] == <span class="dv">1</span>) {</span>
+<span id="cb56-19"><a href="#cb56-19" aria-hidden="true" tabindex="-1"></a>          n_pos[i] = n;</span>
+<span id="cb56-20"><a href="#cb56-20" aria-hidden="true" tabindex="-1"></a>          d_pos[i] = d;</span>
+<span id="cb56-21"><a href="#cb56-21" aria-hidden="true" tabindex="-1"></a>          i += <span class="dv">1</span>;</span>
+<span id="cb56-22"><a href="#cb56-22" aria-hidden="true" tabindex="-1"></a>        } <span class="cf">else</span> {</span>
+<span id="cb56-23"><a href="#cb56-23" aria-hidden="true" tabindex="-1"></a>          n_neg[j] = n;</span>
+<span id="cb56-24"><a href="#cb56-24" aria-hidden="true" tabindex="-1"></a>          d_neg[j] = d;</span>
+<span id="cb56-25"><a href="#cb56-25" aria-hidden="true" tabindex="-1"></a>          j += <span class="dv">1</span>;</span>
+<span id="cb56-26"><a href="#cb56-26" aria-hidden="true" tabindex="-1"></a>        }</span>
+<span id="cb56-27"><a href="#cb56-27" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb56-28"><a href="#cb56-28" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb56-29"><a href="#cb56-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb56-30"><a href="#cb56-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The variables <code>N_pos</code> and <code>N_neg</code> are set to the number of true (1) and number of false (0) observations in <code>y</code>. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.</p>
+<p>The parameters are declared as follows.</p>
+<div class="sourceCode" id="cb57"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb57-1"><a href="#cb57-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb57-2"><a href="#cb57-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, K] beta;</span>
+<span id="cb57-3"><a href="#cb57-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
+<span id="cb57-4"><a href="#cb57-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N_pos] z_pos;</span>
+<span id="cb57-5"><a href="#cb57-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt;[N_neg] z_neg;</span>
+<span id="cb57-6"><a href="#cb57-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These include the regression coefficients <code>beta</code> and the Cholesky factor of the correlation matrix, <code>L_Omega</code>. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).</p>
+<p>The critical part of the parameter declaration is that the latent real value <span class="math inline">\(z\)</span> is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct <span class="math inline">\(z\)</span>.</p>
+<div class="sourceCode" id="cb58"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb58-1"><a href="#cb58-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb58-2"><a href="#cb58-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] z;</span>
+<span id="cb58-3"><a href="#cb58-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_pos) {</span>
+<span id="cb58-4"><a href="#cb58-4" aria-hidden="true" tabindex="-1"></a>    z[n_pos[n], d_pos[n]] = z_pos[n];</span>
+<span id="cb58-5"><a href="#cb58-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb58-6"><a href="#cb58-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_neg) {</span>
+<span id="cb58-7"><a href="#cb58-7" aria-hidden="true" tabindex="-1"></a>    z[n_neg[n], d_neg[n]] = z_neg[n];</span>
+<span id="cb58-8"><a href="#cb58-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb58-9"><a href="#cb58-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>At this point, the model is simple, pretty much recreating the seemingly unrelated regression.</p>
+<div class="sourceCode" id="cb59"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb59-1"><a href="#cb59-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb59-2"><a href="#cb59-2" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">4</span>);</span>
+<span id="cb59-3"><a href="#cb59-3" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb59-4"><a href="#cb59-4" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb59-5"><a href="#cb59-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[N] <span class="dt">vector</span>[D] beta_x;</span>
+<span id="cb59-6"><a href="#cb59-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb59-7"><a href="#cb59-7" aria-hidden="true" tabindex="-1"></a>      beta_x[n] = beta * x[n];</span>
+<span id="cb59-8"><a href="#cb59-8" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb59-9"><a href="#cb59-9" aria-hidden="true" tabindex="-1"></a>    z ~ multi_normal_cholesky(beta_x, L_Omega);</span>
+<span id="cb59-10"><a href="#cb59-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb59-11"><a href="#cb59-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This simple form of model is made possible by the Albert and Chib-style constraints on <code>z</code>.</p>
+<p>Finally, the correlation matrix itself can be put back together in the generated quantities block if desired.</p>
+<div class="sourceCode" id="cb60"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb60-1"><a href="#cb60-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb60-2"><a href="#cb60-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[D] Omega;</span>
+<span id="cb60-3"><a href="#cb60-3" aria-hidden="true" tabindex="-1"></a>  Omega = multiply_lower_tri_self_transpose(L_Omega);</span>
+<span id="cb60-4"><a href="#cb60-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The same could be done for the seemingly unrelated regressions in the previous section.</p>
+</section>
+</section>
+<section id="applications-of-pseudorandom-number-generation" class="level2">
+<h2 class="anchored" data-anchor-id="applications-of-pseudorandom-number-generation">Applications of pseudorandom number generation</h2>
+<p>The main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.</p>
+<section id="prediction" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="prediction">Prediction</h3>
+<p>Consider predicting unobserved outcomes using linear regression. Given predictors <span class="math inline">\(x_1, \dotsc, x_N\)</span> and observed outcomes <span class="math inline">\(y_1, \dotsc, y_N\)</span>, and assuming a standard linear regression with intercept <span class="math inline">\(\alpha\)</span>, slope <span class="math inline">\(\beta\)</span>, and error scale <span class="math inline">\(\sigma\)</span>, along with improper uniform priors, the posterior over the parameters given <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span> is <span class="math display">\[
+p\left(\alpha, \beta, \sigma \mid x, y \right)
+\propto
+\prod_{n=1}^N
+  \textsf{normal}\left(y_n \mid \alpha + \beta x_n, \sigma\right).
+\]</span></p>
+<p>For this model, the posterior predictive inference for a new outcome <span class="math inline">\(\tilde{y}_m\)</span> given a predictor <span class="math inline">\(\tilde{x}_m\)</span>, conditioned on the observed data <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span>, is <span class="math display">\[
+p\left(\tilde{y}_n \mid \tilde{x}_n, x, y\right)
+= \int_{(\alpha,\beta,\sigma)}
+  \textsf{normal}\left(\tilde{y}_n \mid \alpha + \beta \tilde{x}_n, \sigma\right)
+  \times
+  p\left(\alpha, \beta, \sigma \mid x, y\right)
+  \,\textrm{d}(\alpha,\beta,\sigma).
+\]</span></p>
+<p>To code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.</p>
+<div class="sourceCode" id="cb61"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb61-1"><a href="#cb61-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb61-2"><a href="#cb61-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb61-3"><a href="#cb61-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb61-4"><a href="#cb61-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb61-5"><a href="#cb61-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb61-6"><a href="#cb61-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
+<span id="cb61-7"><a href="#cb61-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb61-8"><a href="#cb61-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb61-9"><a href="#cb61-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb61-10"><a href="#cb61-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb61-11"><a href="#cb61-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb61-12"><a href="#cb61-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb61-13"><a href="#cb61-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb61-14"><a href="#cb61-14" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb61-15"><a href="#cb61-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb61-16"><a href="#cb61-16" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb61-17"><a href="#cb61-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] y_tilde;</span>
+<span id="cb61-18"><a href="#cb61-18" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_tilde) {</span>
+<span id="cb61-19"><a href="#cb61-19" aria-hidden="true" tabindex="-1"></a>    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);</span>
+<span id="cb61-20"><a href="#cb61-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb61-21"><a href="#cb61-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Given observed predictors <span class="math inline">\(x\)</span> and outcomes <span class="math inline">\(y\)</span>, <code>y_tilde</code> will be drawn according to <span class="math inline">\(p\left(\tilde{y} \mid \tilde{x}, y, x\right)\)</span>. This means that, for example, the posterior mean for <code>y_tilde</code> is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).</p>
+</section>
+<section id="posterior-predictive-checks" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="posterior-predictive-checks">Posterior predictive checks</h3>
+<p>A good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors <span class="math inline">\(x\)</span> are used in place of new predictors <span class="math inline">\(\tilde{x}\)</span> for unobserved outcomes. If the model fits the data well, the predictions for <span class="math inline">\(\tilde{y}\)</span> based on <span class="math inline">\(x\)</span> should match the observed data <span class="math inline">\(y\)</span>.</p>
+<p>To code posterior predictive checks in Stan requires only a slight modification of the prediction code to use <span class="math inline">\(x\)</span> and <span class="math inline">\(N\)</span> in place of <span class="math inline">\(\tilde{x}\)</span> and <span class="math inline">\(\tilde{N}\)</span>,</p>
+<div class="sourceCode" id="cb62"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb62-1"><a href="#cb62-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb62-2"><a href="#cb62-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y_tilde;</span>
+<span id="cb62-3"><a href="#cb62-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb62-4"><a href="#cb62-4" aria-hidden="true" tabindex="-1"></a>    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);</span>
+<span id="cb62-5"><a href="#cb62-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb62-6"><a href="#cb62-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> recommend choosing several posterior draws <span class="math inline">\(\tilde{y}^{(1)}, \dotsc, \tilde{y}^{(M)}\)</span> and plotting each of them alongside the data <span class="math inline">\(y\)</span> that was actually observed. If the model fits well, the simulated <span class="math inline">\(\tilde{y}\)</span> will look like the actual data <span class="math inline">\(y\)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-AlbertChib:1993" class="csl-entry" role="listitem">
+Albert, J. H., and S. Chib. 1993. <span>“Bayesian Analysis of Binary and Polychotomous Response Data.”</span> <em>Journal of the American Statistical Association</em> 88: 669–79.
+</div>
+<div id="ref-barnard-mcculloch-meng:2000" class="csl-entry" role="listitem">
+Barnard, John, Robert McCulloch, and Xiao-Li Meng. 2000. <span>“Modeling Covariance Matrices in Terms of Standard Deviations and Correlations, with Application to Shrinkage.”</span> <em>Statistica Sinica</em>, 1281–311.
+</div>
+<div id="ref-ChungEtAl:2013" class="csl-entry" role="listitem">
+Chung, Yeojin, Sophia Rabe-Hesketh, Vincent Dorie, Andrew Gelman, and Jingchen Liu. 2013. <span>“A Nondegenerate Penalized Likelihood Estimator for Variance Parameters in Multilevel Models.”</span> <em>Psychometrika</em> 78 (4): 685–709.
+</div>
+<div id="ref-Curtis:2010" class="csl-entry" role="listitem">
+Curtis, S. McKay. 2010. <span>“<span>BUGS</span> Code for Item Response Theory.”</span> <em>Journal of Statistical Software</em> 36 (1): 1–34.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-GelmanHill:2007" class="csl-entry" role="listitem">
+Gelman, Andrew, and Jennifer Hill. 2007. <em>Data Analysis Using Regression and Multilevel-Hierarchical Models</em>. Cambridge, United Kingdom: Cambridge University Press.
+</div>
+<div id="ref-Greene:2011" class="csl-entry" role="listitem">
+Greene, William H. 2011. <em>Econometric Analysis</em>. 7th ed. Prentice-Hall.
+</div>
+<div id="ref-LewandowskiKurowickaJoe:2009" class="csl-entry" role="listitem">
+Lewandowski, Daniel, Dorota Kurowicka, and Harry Joe. 2009. <span>“Generating Random Correlation Matrices Based on Vines and Extended Onion Method.”</span> <em>Journal of Multivariate Analysis</em> 100: 1989–2001.
+</div>
+<div id="ref-seyboldt:2024" class="csl-entry" role="listitem">
+Seyboldt, Adrian. 2024. <span>“Add ZeroSumNormal Distribution.”</span> <a href="https://github.com/pyro-ppl/numpyro/pull/1751#issuecomment-1980569811">https://github.com/pyro-ppl/numpyro/pull/1751#issuecomment-1980569811</a>.
+</div>
+<div id="ref-Zellner:1962" class="csl-entry" role="listitem">
+Zellner, Arnold. 1962. <span>“An Efficient Method of Estimating Seemingly Unrelated Regression Equations and Tests for Aggregation Bias.”</span> <em>Journal of the American Statistical Association</em> 57: 348–68.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Unlike in Python and R, which are interpreted, Stan is translated to C++ and compiled, so loops and assignment statements are fast. Vectorized code is faster in Stan because (a) the expression tree used to compute derivatives can be simplified, leading to fewer virtual function calls, and (b) computations that would be repeated in the looping version, such as <code>log(sigma)</code> in the above model, will be computed once and reused.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>The <code>Phi_approx</code> function is a rescaled version of the inverse logit function, so while the scale is roughly the same <span class="math inline">\(\Phi\)</span>, the tails do not match.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>The Bernoulli-logit distribution builds in the log link function, taking <span class="math display">\[\texttt{bernoulli}\mathtt{\_}\texttt{logit}\left(y \mid \alpha\right) = \texttt{bernoulli}\left(y \mid \operatorname{logit}^{-1}(\alpha)\right).\]</span><a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn4"><p><span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007</a>)</span> treat the <span class="math inline">\(\delta\)</span> term equivalently as the location parameter in the distribution of student abilities.<a href="#fnref4" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn5"><p>The prior is named for Lewandowski, Kurowicka, and Joe, as it was derived by inverting the random correlation matrix generation strategy of <span class="citation" data-cites="LewandowskiKurowickaJoe:2009">Lewandowski, Kurowicka, and Joe (<a href="#ref-LewandowskiKurowickaJoe:2009" role="doc-biblioref">2009</a>)</span>.<a href="#fnref5" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn6"><p>Thanks to Mike Lawrence for pointing this out in the GitHub issue for the manual.<a href="#fnref6" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/index.html" class="pagination-link" aria-label="Stan User's Guide">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Stan User’s Guide</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/time-series.html" class="pagination-link" aria-label="Time-Series Models">
+        <span class="nav-page-text">Time-Series Models</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/regression.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/reparameterization.html b/docs/2_39/stan-users-guide/reparameterization.html
new file mode 100644
index 000000000..f88283870
--- /dev/null
+++ b/docs/2_39/stan-users-guide/reparameterization.html
@@ -0,0 +1,1629 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Reparameterization and Change of Variables</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/efficiency-tuning.html" rel="next">
+<link href="../stan-users-guide/problematic-posteriors.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Reparameterization and Change of Variables">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/reparameterization.html">Reparameterization and Change of Variables</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#change-of-variables.chapter" id="toc-change-of-variables.chapter" class="nav-link active" data-scroll-target="#change-of-variables.chapter">Reparameterization and Change of Variables</a>
+  <ul class="collapse">
+  <li><a href="#theoretical-and-practical-background" id="toc-theoretical-and-practical-background" class="nav-link" data-scroll-target="#theoretical-and-practical-background">Theoretical and practical background</a></li>
+  <li><a href="#reparameterizations" id="toc-reparameterizations" class="nav-link" data-scroll-target="#reparameterizations">Reparameterizations</a>
+  <ul class="collapse">
+  <li><a href="#beta-and-dirichlet-priors" id="toc-beta-and-dirichlet-priors" class="nav-link" data-scroll-target="#beta-and-dirichlet-priors">Beta and Dirichlet priors</a></li>
+  <li><a href="#transforming-unconstrained-priors-probit-and-logit" id="toc-transforming-unconstrained-priors-probit-and-logit" class="nav-link" data-scroll-target="#transforming-unconstrained-priors-probit-and-logit">Transforming unconstrained priors: probit and logit</a></li>
+  </ul></li>
+  <li><a href="#changes-of-variables" id="toc-changes-of-variables" class="nav-link" data-scroll-target="#changes-of-variables">Changes of variables</a>
+  <ul class="collapse">
+  <li><a href="#change-of-variables-vs.-transformations" id="toc-change-of-variables-vs.-transformations" class="nav-link" data-scroll-target="#change-of-variables-vs.-transformations">Change of variables vs.&nbsp;transformations</a></li>
+  <li><a href="#multivariate-changes-of-variables" id="toc-multivariate-changes-of-variables" class="nav-link" data-scroll-target="#multivariate-changes-of-variables">Multivariate changes of variables</a></li>
+  </ul></li>
+  <li><a href="#vectors-with-varying-bounds" id="toc-vectors-with-varying-bounds" class="nav-link" data-scroll-target="#vectors-with-varying-bounds">Vectors with varying bounds</a>
+  <ul class="collapse">
+  <li><a href="#varying-lower-bounds" id="toc-varying-lower-bounds" class="nav-link" data-scroll-target="#varying-lower-bounds">Varying lower bounds</a></li>
+  <li><a href="#varying-upper-and-lower-bounds" id="toc-varying-upper-and-lower-bounds" class="nav-link" data-scroll-target="#varying-upper-and-lower-bounds">Varying upper and lower bounds</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/reparameterization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/reparameterization.html">Reparameterization and Change of Variables</a></li></ol></nav></header>
+
+
+
+
+<section id="change-of-variables.chapter" class="level1">
+<h1>Reparameterization and Change of Variables</h1>
+<p>Stan supports a direct encoding of reparameterizations. Stan also supports changes of variables by directly incrementing the log probability accumulator with the log Jacobian of the transform.</p>
+<section id="theoretical-and-practical-background" class="level2">
+<h2 class="anchored" data-anchor-id="theoretical-and-practical-background">Theoretical and practical background</h2>
+<p>A Bayesian posterior is technically a probability <em>measure</em>, which is a parameterization-invariant, abstract mathematical object.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>Stan’s modeling language, on the other hand, defines a probability <em>density</em>, which is a non-unique, parameterization-dependent function in <span class="math inline">\(\mathbb{R}^N \rightarrow \mathbb{R}^{+}\)</span>. In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.</p>
+<p>As pointed out by <span class="citation" data-cites="Gelman:2004">Gelman (<a href="#ref-Gelman:2004" role="doc-biblioref">2004</a>)</span> in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not <em>just</em> that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.</p>
+</section>
+<section id="reparameterizations" class="level2">
+<h2 class="anchored" data-anchor-id="reparameterizations">Reparameterizations</h2>
+<p>Reparameterizations may be implemented directly using the transformed parameters block or just in the model block.</p>
+<section id="beta-and-dirichlet-priors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="beta-and-dirichlet-priors">Beta and Dirichlet priors</h3>
+<p>The beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.</p>
+<section id="beta-distribution" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="beta-distribution">Beta distribution</h4>
+<p>For example, the Beta distribution is parameterized by two positive count parameters <span class="math inline">\(\alpha, \beta &gt; 0\)</span>. The following example illustrates a hierarchical Stan model with a vector of parameters <code>theta</code> are drawn i.i.d.&nbsp;for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; beta;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  alpha ~ ...</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  beta ~ ...</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>    theta[n] ~ beta(alpha, beta);</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter <span class="math display">\[
+\phi = \alpha / (\alpha + \beta)
+\]</span> and total count parameter <span class="math display">\[
+\lambda = \alpha + \beta.
+\]</span> Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with <span class="math inline">\(p(\lambda) \propto \lambda^{-2.5}\)</span>.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; phi;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="fl">0.1</span>&gt; lambda;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha = lambda * phi;</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; beta = lambda * (<span class="dv">1</span> - phi);</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  phi ~ beta(<span class="dv">1</span>, <span class="dv">1</span>); <span class="co">// uniform on phi, could drop</span></span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>  lambda ~ pareto(<span class="fl">0.1</span>, <span class="fl">1.5</span>);</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>    theta[n] ~ beta(alpha, beta);</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The new parameters, <code>phi</code> and <code>lambda</code>, are declared in the parameters block and the parameters for the Beta distribution, <code>alpha</code> and <code>beta</code>, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha = lambda * phi;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta = lambda * (<span class="dv">1</span> - phi);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>    theta[n] ~ beta(alpha, beta);</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With vectorization, this could be expressed more compactly and efficiently as follows.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  theta ~ beta(lambda * phi, lambda * (<span class="dv">1</span> - phi));</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the variables <code>alpha</code> and <code>beta</code> are of interest, they can be defined in the transformed parameter block and then used in the model.</p>
+</section>
+<section id="jacobians-not-necessary" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="jacobians-not-necessary">Jacobians not necessary</h4>
+<p>Because the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, <code>alpha</code> and <code>beta</code> have the correct posterior distribution.</p>
+</section>
+<section id="dirichlet-priors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="dirichlet-priors">Dirichlet priors</h4>
+<p>The same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are <span class="math inline">\(K &gt; 0\)</span> dimensions being considered (<span class="math inline">\(K=1\)</span> is trivial and <span class="math inline">\(K=2\)</span> reduces to the beta distribution case). The traditional prior is</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] alpha;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">simplex</span>[K] theta;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  alpha ~ <span class="co">// ...</span></span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>    theta[n] ~ dirichlet(alpha);</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This provides essentially <span class="math inline">\(K\)</span> degrees of freedom, one for each dimension of <code>alpha</code>, and it is not obvious how to specify a reasonable prior for <code>alpha</code>.</p>
+<p>An alternative coding is to use the mean, which is a simplex, and a total count.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] phi;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; kappa;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">simplex</span>[K] theta;</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] alpha = kappa * phi;</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>  phi ~ <span class="co">// ...</span></span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>  kappa ~ <span class="co">// ...</span></span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>    theta[n] ~ dirichlet(alpha);</span>
+<span id="cb6-16"><a href="#cb6-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-17"><a href="#cb6-17" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb6-18"><a href="#cb6-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now it is much easier to formulate priors, because <code>phi</code> is the expected value of <code>theta</code> and <code>kappa</code> (minus <code>K</code>) is the strength of the prior mean measured in number of prior observations.</p>
+</section>
+</section>
+<section id="transforming-unconstrained-priors-probit-and-logit" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="transforming-unconstrained-priors-probit-and-logit">Transforming unconstrained priors: probit and logit</h3>
+<p>If the variable <span class="math inline">\(u\)</span> has a <span class="math inline">\(\textsf{uniform}(0, 1)\)</span> distribution, then <span class="math inline">\(\operatorname{logit}(u)\)</span> is distributed as <span class="math inline">\(\textsf{logistic}(0, 1)\)</span>. This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in <span class="math inline">\((0, 1)\)</span> to a logistically-distributed quantity.</p>
+<p>Things work the same way for the probit case: if <span class="math inline">\(u\)</span> has a <span class="math inline">\(\textsf{uniform}(0, 1)\)</span> distribution, then <span class="math inline">\(\Phi^{-1}(u)\)</span> has a <span class="math inline">\(\textsf{normal}(0, 1)\)</span> distribution. The other way around, if <span class="math inline">\(v\)</span> has a <span class="math inline">\(\textsf{normal}(0, 1)\)</span> distribution, then <span class="math inline">\(\Phi(v)\)</span> has a <span class="math inline">\(\textsf{uniform}(0, 1)\)</span> distribution.</p>
+<p>In order to use the probit and logistic as priors on variables constrained to <span class="math inline">\((0, 1)\)</span>, create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a <span class="math inline">\((0, 1)\)</span>-constrained parameter <code>theta</code> and gives it a beta prior, then uses it as a parameter in a distribution (here using <code>foo</code> as a placeholder).</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>  theta ~ beta(a, b);</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  y ~ foo(theta);</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the variables <code>a</code> and <code>b</code> are one, then this imposes a uniform distribution <code>theta</code>. If <code>a</code> and <code>b</code> are both less than one, then the density on <code>theta</code> has a U shape, whereas if they are both greater than one, the density of <code>theta</code> has an inverted-U or more bell-like shape.</p>
+<p>Roughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta_raw;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta = inv_logit(theta_raw);</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>  theta_raw ~ logistic(mu, sigma);</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>  y ~ foo(theta);</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this model, an unconstrained parameter <code>theta_raw</code> gets a logistic prior, and then the transformed parameter <code>theta</code> is defined to be the inverse logit of <code>theta_raw</code>. In this parameterization, <code>inv_logit(mu)</code> is the mean of the implied prior on <code>theta</code>. The prior distribution on <code>theta</code> will be flat if <code>sigma</code> is one and <code>mu</code> is zero, and will be U-shaped if <code>sigma</code> is larger than one and bell shaped if <code>sigma</code> is less than one.</p>
+<p>When moving from a variable in <span class="math inline">\((0, 1)\)</span> to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] theta;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>  theta ~ dirichlet(a);</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  y ~ foo(theta);</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Now <code>a</code> is a vector with <code>K</code> rows, but it has the same shape properties as the pair <code>a</code> and <code>b</code> for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector <span class="math inline">\([a b]^{\top}\)</span>. To formulate an unconstrained prior, the exact same strategy works as for the beta.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] theta_raw;</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[K] theta = softmax(theta_raw);</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a>  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);</span>
+<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by <code>softmax(mu)</code>, but we have additional control of covariance through <code>L_Sigma</code> at the expense of having on the order of <span class="math inline">\(K^2\)</span> parameters in the prior rather than order <span class="math inline">\(K\)</span>. If no covariance is desired, the number of parameters can be reduced back to <span class="math inline">\(K\)</span> using a vectorized normal distribution as follows.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a>theta_raw ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>where either or both of <code>mu</code> and <code>sigma</code> can be vectors.</p>
+</section>
+</section>
+<section id="changes-of-variables" class="level2">
+<h2 class="anchored" data-anchor-id="changes-of-variables">Changes of variables</h2>
+<p>Changes of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable <span class="math inline">\(y &gt; 0\)</span> whose logarithm <span class="math inline">\(\log y\)</span> has a normal distribution. The distribution is being assigned to <span class="math inline">\(\log y\)</span>.</p>
+<p>The change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map <span class="math inline">\(\mathbb{R}^N \rightarrow \mathbb{R}^N\)</span>.</p>
+<p>The probability must be scaled by a <em>Jacobian adjustment</em> equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.</p>
+<p>In the case of log normals, if <span class="math inline">\(y\)</span>’s logarithm is normal with mean <span class="math inline">\(\mu\)</span> and deviation <span class="math inline">\(\sigma\)</span>, then the distribution of <span class="math inline">\(y\)</span> is given by <span class="math display">\[
+p(y)
+= \textsf{normal}(\log y \mid \mu, \sigma) \, \left| \frac{d}{dy} \log y \right|
+= \textsf{normal}(\log y \mid \mu, \sigma) \, \frac{1}{y}.
+\]</span> Stan works on the log scale to prevent underflow, where <span class="math display">\[
+\log p(y)
+=
+\log \textsf{normal}(\log y \mid \mu, \sigma) - \log y.
+\]</span></p>
+<p>In Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>  log(y) ~ normal(mu, sigma);</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -log(y);</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It is important, as always, to declare appropriate constraints on parameters; here <code>y</code> is constrained to be positive.</p>
+<p>It would be slightly more efficient to define a local variable for the logarithm, as follows.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_y;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  log_y = log(y);</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>  log_y ~ normal(mu, sigma);</span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -log_y;</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If <code>y</code> were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.</p>
+<section id="change-of-variables-vs.-transformations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="change-of-variables-vs.-transformations">Change of variables vs.&nbsp;transformations</h3>
+<p>This section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.</p>
+<p>It does not matter whether the probability function is expressed using a distribution statement, such as</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a>log(y) ~ normal(mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>or as an increment to the log probability function, as in</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> normal_lpdf(log(y) | mu, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="jacobian-adjustment.section" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="jacobian-adjustment.section">Gamma and inverse gamma distribution</h4>
+<p>Like the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.</p>
+<p>The transform based approach to defining <code>y_inv</code> to have an inverse gamma distribution can be coded as follows.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y_inv;</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>  y_inv = <span class="dv">1</span> / y;</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>  y ~ gamma(<span class="dv">2</span>,<span class="dv">4</span>);</span>
+<span id="cb16-10"><a href="#cb16-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The change-of-variables approach to defining <code>y_inv</code> to have an inverse gamma distribution can be coded as follows.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y_inv;</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; y;</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a>  y = <span class="dv">1</span> / y_inv;  <span class="co">// change variables</span></span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>  <span class="kw">jacobian +=</span> -<span class="dv">2</span> * log(y_inv); <span class="co">// Jacobian adjustment</span></span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb17-10"><a href="#cb17-10" aria-hidden="true" tabindex="-1"></a>  y ~ gamma(<span class="dv">2</span>,<span class="dv">4</span>);</span>
+<span id="cb17-11"><a href="#cb17-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is</p>
+<p><span class="math display">\[
+\log \left| \frac{d}{du} \left( \frac{1}{u} \right) \right|
+= \log \left| - u^{-2} \right|
+= \log u^{-2}
+=  -2 \log u.
+\]</span></p>
+</section>
+</section>
+<section id="multivariate-changes-of-variables" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="multivariate-changes-of-variables">Multivariate changes of variables</h3>
+<p>In the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] u;      <span class="co">// multivariate parameter</span></span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>   <span class="co">// ...</span></span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] v;     <span class="co">// transformed parameter</span></span>
+<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] J;   <span class="co">// Jacobian matrix of transform</span></span>
+<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... compute v as a function of u ...</span></span>
+<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... compute J[m, n] = d.v[m] / d.u[n] ...</span></span>
+<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a>  <span class="kw">jacobian +=</span> log(abs(determinant(J)));</span>
+<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb18-12"><a href="#cb18-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-13"><a href="#cb18-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb18-14"><a href="#cb18-14" aria-hidden="true" tabindex="-1"></a>  v ~ <span class="co">// ...</span></span>
+<span id="cb18-15"><a href="#cb18-15" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb18-16"><a href="#cb18-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.</p>
+<p>In many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element <code>v[k]</code> of the transformed parameter vector only depends on elements <code>u[1]</code>, …, <code>u[k]</code> of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] J_diag;  <span class="co">// diagonals of Jacobian matrix</span></span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... compute J[k, k] = d.v[k] / d.u[k] ...</span></span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>  <span class="kw">jacobian +=</span> sum(log(J_diag));</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="vectors-with-varying-bounds" class="level2">
+<h2 class="anchored" data-anchor-id="vectors-with-varying-bounds">Vectors with varying bounds</h2>
+<p>Stan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.</p>
+<section id="varying-lower-bounds" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="varying-lower-bounds">Varying lower bounds</h3>
+<p>For example, suppose there is a vector parameter <span class="math inline">\(\alpha\)</span> with a vector <span class="math inline">\(L\)</span> of lower bounds. The simplest way to deal with this if <span class="math inline">\(L\)</span> is a constant is to shift a lower-bounded parameter.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] L;  <span class="co">// lower bounds</span></span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=L&gt;[N] alpha_raw;</span>
+<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb20-9"><a href="#cb20-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The above is equivalent to manually calculating the vector bounds by the following.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] L;  <span class="co">// lower bounds</span></span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N] alpha_raw;</span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] alpha = L + alpha_raw;</span>
+<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Jacobian for adding a constant is one, so its log drops out of the log density.</p>
+<p>Even if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from <span class="math inline">\((L, \alpha_{\textrm{raw}})\)</span> to <span class="math inline">\((L + \alpha_{\textrm{raw}}, \alpha_{\textrm{raw}})\)</span> produces a Jacobian derivative matrix with a unit determinant.</p>
+<p>It’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] L;  <span class="co">// lower bounds</span></span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-6"><a href="#cb22-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb22-7"><a href="#cb22-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N] alpha_raw;</span>
+<span id="cb22-8"><a href="#cb22-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=L + alpha_raw&gt;[N] alpha;</span>
+<span id="cb22-9"><a href="#cb22-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb22-10"><a href="#cb22-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] L;  <span class="co">// lower bounds</span></span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] alpha_raw;</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] alpha = L + exp(alpha_raw);</span>
+<span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a>  <span class="kw">jacobian +=</span> sum(alpha_raw); <span class="co">// log Jacobian</span></span>
+<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-15"><a href="#cb23-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb23-16"><a href="#cb23-16" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb23-17"><a href="#cb23-17" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The adjustment in the log Jacobian determinant of the transform mapping <span class="math inline">\(\alpha_{\textrm{raw}}\)</span> to <span class="math inline">\(\alpha = L + \exp(\alpha_{\textrm{raw}})\)</span>. The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here <span class="math inline">\(L\)</span> can even be a vector containing parameters that don’t depend on <span class="math inline">\(\alpha_{\textrm{raw}}\)</span>; if the bounds do depend on <span class="math inline">\(\alpha_{\textrm{raw}}\)</span> then a revised Jacobian needs to be calculated taking into account the dependencies.</p>
+</section>
+<section id="varying-upper-and-lower-bounds" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="varying-upper-and-lower-bounds">Varying upper and lower bounds</h3>
+<p>Suppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to <span class="math inline">\((0, 1)\)</span>. This is easily accomplished as the following.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] L;  <span class="co">// lower bounds</span></span>
+<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] U;  <span class="co">// upper bounds</span></span>
+<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb24-8"><a href="#cb24-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=L, <span class="kw">upper</span>=U&gt;[N] alpha;</span>
+<span id="cb24-9"><a href="#cb24-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb24-10"><a href="#cb24-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The same may be accomplished by manually constructing the transform as follows.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] L;  <span class="co">// lower bounds</span></span>
+<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] U;  <span class="co">// upper bounds</span></span>
+<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;[N] alpha_raw;</span>
+<span id="cb25-9"><a href="#cb25-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb25-10"><a href="#cb25-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb25-11"><a href="#cb25-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb25-12"><a href="#cb25-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] alpha = L + (U - L) .* alpha_raw;</span>
+<span id="cb25-13"><a href="#cb25-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The expression <code>U - L</code> is multiplied by <code>alpha_raw</code> elementwise to produce a vector of variables in <span class="math inline">\((0, U-L)\)</span>, then adding <span class="math inline">\(L\)</span> results in a variable ranging between <span class="math inline">\((L, U)\)</span>.</p>
+<p>In this case, it is important that <span class="math inline">\(L\)</span> and <span class="math inline">\(U\)</span> are constants, otherwise a Jacobian would be required when multiplying by <span class="math inline">\(U - L\)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Gelman:2004" class="csl-entry" role="listitem">
+Gelman, Andrew. 2004. <span>“Parameterization and <span>B</span>ayesian Modeling.”</span> <em>Journal of the American Statistical Association</em> 99: 537–45.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>This is in contrast to (penalized) maximum likelihood estimates, which are not parameterization invariant.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>This example is for illustrative purposes only; the recommended way to implement the lognormal distribution in Stan is with the built-in <code>lognormal</code> probability function; see the functions reference manual for details.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/problematic-posteriors.html" class="pagination-link" aria-label="Problematic Posteriors">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Problematic Posteriors</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/efficiency-tuning.html" class="pagination-link" aria-label="Efficiency Tuning">
+        <span class="nav-page-text">Efficiency Tuning</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/reparameterization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/simulation-based-calibration.html b/docs/2_39/stan-users-guide/simulation-based-calibration.html
new file mode 100644
index 000000000..b3a86f5bf
--- /dev/null
+++ b/docs/2_39/stan-users-guide/simulation-based-calibration.html
@@ -0,0 +1,1619 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Simulation-Based Calibration Checking</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/posterior-predictive-checks.html" rel="next">
+<link href="../stan-users-guide/posterior-prediction.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Simulation-Based Calibration Checking">
+<meta property="og:description" content="">
+<meta property="og:image" content="./img/sbc-normal-normal.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration Checking</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#simulation-based-calibration-checking" id="toc-simulation-based-calibration-checking" class="nav-link active" data-scroll-target="#simulation-based-calibration-checking">Simulation-Based Calibration Checking</a>
+  <ul class="collapse">
+  <li><a href="#bayes-is-calibrated-by-construction" id="toc-bayes-is-calibrated-by-construction" class="nav-link" data-scroll-target="#bayes-is-calibrated-by-construction">Bayes is calibrated by construction</a></li>
+  <li><a href="#simulation-based-calibration-checking-1" id="toc-simulation-based-calibration-checking-1" class="nav-link" data-scroll-target="#simulation-based-calibration-checking-1">Simulation-based calibration checking</a></li>
+  <li><a href="#sbc-in-stan" id="toc-sbc-in-stan" class="nav-link" data-scroll-target="#sbc-in-stan">SBC in Stan</a>
+  <ul class="collapse">
+  <li><a href="#example-model" id="toc-example-model" class="nav-link" data-scroll-target="#example-model">Example model</a></li>
+  <li><a href="#testing-a-stan-program-with-simulation-based-calibration-checking" id="toc-testing-a-stan-program-with-simulation-based-calibration-checking" class="nav-link" data-scroll-target="#testing-a-stan-program-with-simulation-based-calibration-checking">Testing a Stan program with simulation-based calibration checking</a></li>
+  <li><a href="#pseudocode-for-simulation-based-calibration-checking" id="toc-pseudocode-for-simulation-based-calibration-checking" class="nav-link" data-scroll-target="#pseudocode-for-simulation-based-calibration-checking">Pseudocode for simulation-based calibration checking</a></li>
+  <li><a href="#the-importance-of-thinning" id="toc-the-importance-of-thinning" class="nav-link" data-scroll-target="#the-importance-of-thinning">The importance of thinning</a></li>
+  </ul></li>
+  <li><a href="#testing-uniformity" id="toc-testing-uniformity" class="nav-link" data-scroll-target="#testing-uniformity">Testing uniformity</a>
+  <ul class="collapse">
+  <li><a href="#indexing-to-simplify-arithmetic" id="toc-indexing-to-simplify-arithmetic" class="nav-link" data-scroll-target="#indexing-to-simplify-arithmetic">Indexing to simplify arithmetic</a></li>
+  </ul></li>
+  <li><a href="#examples-of-simulation-based-calibration-checking" id="toc-examples-of-simulation-based-calibration-checking" class="nav-link" data-scroll-target="#examples-of-simulation-based-calibration-checking">Examples of simulation-based calibration checking</a>
+  <ul class="collapse">
+  <li><a href="#when-things-go-right" id="toc-when-things-go-right" class="nav-link" data-scroll-target="#when-things-go-right">When things go right</a></li>
+  <li><a href="#when-things-go-wrong" id="toc-when-things-go-wrong" class="nav-link" data-scroll-target="#when-things-go-wrong">When things go wrong</a></li>
+  <li><a href="#when-stans-sampler-goes-wrong" id="toc-when-stans-sampler-goes-wrong" class="nav-link" data-scroll-target="#when-stans-sampler-goes-wrong">When Stan’s sampler goes wrong</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/simulation-based-calibration.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration Checking</a></li></ol></nav></header>
+
+
+
+
+<section id="simulation-based-calibration-checking" class="level1">
+<h1>Simulation-Based Calibration Checking</h1>
+<p>A Bayesian posterior is calibrated if the posterior intervals have appropriate coverage. For example, 80% intervals are expected to contain the true parameter 80% of the time. If data is generated according to a model, Bayesian posterior inference with respect to that model is calibrated by construction. Simulation-based calibration checking (SBC) exploits this property of Bayesian inference to assess the soundness of a posterior sampler. Roughly, the way it works is by simulating parameters according to the prior, then simulating data conditioned on the simulated parameters, then testing posterior calibration of the inference algorithm over independently simulated data sets. This chapter follows <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span>, which improves on the original approach developed by <span class="citation" data-cites="CookGelmanRubin:2006">Cook, Gelman, and Rubin (<a href="#ref-CookGelmanRubin:2006" role="doc-biblioref">2006</a>)</span>. See also <span class="citation" data-cites="ModrakEtAl:2023">Modrák et al. (<a href="#ref-ModrakEtAl:2023" role="doc-biblioref">2023</a>)</span> for further improvements.</p>
+<section id="bayes-is-calibrated-by-construction" class="level2">
+<h2 class="anchored" data-anchor-id="bayes-is-calibrated-by-construction">Bayes is calibrated by construction</h2>
+<p>Suppose a Bayesian model is given in the form of a prior density <span class="math inline">\(p(\theta)\)</span> and sampling density <span class="math inline">\(p(y \mid \theta).\)</span> Now consider a process that first simulates parameters from the prior, <span class="math display">\[
+\theta^{\textrm{sim}} \sim p(\theta),
+\]</span> and then simulates data given the parameters, <span class="math display">\[
+y^{\textrm{sim}} \sim p(y \mid \theta^{\textrm{sim}}).
+\]</span> By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, <span class="math display">\[
+(y^{\textrm{sim}}, \theta^{\textrm{sim}}) \sim p(y, \theta).
+\]</span> From Bayes’s rule, it follows that for any observed (fixed) data <span class="math inline">\(y\)</span>, <span class="math display">\[
+p(\theta \mid y) \propto p(y, \theta).
+\]</span> Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, <span class="math display">\[
+\theta^{\textrm{sim}} \sim p(\theta \mid y^{\textrm{sim}}).
+\]</span> Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, <span class="math display">\[
+\theta^{(1)}, \ldots, \theta^{(M)}
+\sim p(\theta \mid y^{\textrm{sim}}).
+\]</span> Because <span class="math inline">\(\theta^{\textrm{sim}}\)</span> is also distributed as a draw from the posterior, the rank statistics of <span class="math inline">\(\theta^{\textrm{sim}}\)</span> with respect to <span class="math inline">\(\theta^{(1)}, \ldots \theta^{(M)}\)</span> should be uniform.</p>
+<p>This is one way to define calibration, because it follows that posterior intervals will have appropriate coverage <span class="citation" data-cites="Dawid:1982 GneitingEtAl:2007">(<a href="#ref-Dawid:1982" role="doc-biblioref">Dawid 1982</a>; <a href="#ref-GneitingEtAl:2007" role="doc-biblioref">Gneiting, Balabdaoui, and Raftery 2007</a>)</span>. If the rank of <span class="math inline">\(\theta^{\textrm{sim}}\)</span> is uniform among the draws <span class="math inline">\(\theta^{(1)}, \ldots, \theta^{(M)},\)</span> then for any 90% interval selected, the probability the true value <span class="math inline">\(\theta^{\textrm{sim}}\)</span> falls in it will also be 90%. The same goes for any other posterior interval.</p>
+</section>
+<section id="simulation-based-calibration-checking-1" class="level2">
+<h2 class="anchored" data-anchor-id="simulation-based-calibration-checking-1">Simulation-based calibration checking</h2>
+<p>Suppose the Bayesian model to test has joint density <span class="math display">\[
+p(y, \theta) = p(y \mid \theta) \cdot p(\theta),
+\]</span> with data <span class="math inline">\(y\)</span> and parameters <span class="math inline">\(\theta\)</span> (both are typically multivariate). Simulation-based calibration checking works by generating <span class="math inline">\(N\)</span> simulated parameter and data pairs according to the joint density, <span class="math display">\[
+(y^{\textrm{sim}(1)}, \theta^{\textrm{sim}(1)}),
+\ldots, (y^{\textrm{sim}(N)}, \theta^{\textrm{sim}(N)}),
+\sim p(y, \theta).
+\]</span> For each simulated data set <span class="math inline">\(y^{\textrm{sim}(n)}\)</span>, use the algorithm to be tested to generate <span class="math inline">\(M\)</span> posterior draws, which if everything is working properly, will be distributed marginally as <span class="math display">\[
+\theta^{(n, 1)}, \ldots, \theta^{(n, M)}
+\sim p(\theta \mid y^{\textrm{sim}(n)}).
+\]</span> For a simulation <span class="math inline">\(n\)</span> and parameter <span class="math inline">\(k\)</span>, the rank of the simulated parameter among the posterior draws is <span class="math display">\[\begin{eqnarray*}
+r_{n, k}
+&amp; = &amp;
+\textrm{rank}(\theta_k^{\textrm{sim}(n)},
+              (\theta^{(n, 1)}, \ldots, \theta^{(n,M)}))
+\\[4pt]
+&amp; = &amp;
+\sum_{m = 1}^M
+  \textrm{I}[\theta_k^{(n,m)} &lt; \theta_k^{\textrm{sim}(n)}].
+\end{eqnarray*}\]</span> That is, the rank is the number of posterior draws <span class="math inline">\(\theta^{(n,m)}_k\)</span> that are less than the simulated draw <span class="math inline">\(\theta^{\textrm{sim}(n)}_k.\)</span></p>
+<p>If the algorithm generates posterior draws according to the posterior, the ranks should have uniform discrete distribution from <span class="math inline">\(0\)</span> to <span class="math inline">\(M\)</span>, so that the ranks plus one are uniformly distributed from <span class="math inline">\(1\)</span> to <span class="math inline">\(M + 1\)</span>, <span class="math display">\[
+r_{n, k} + 1
+\sim
+\textrm{categorical}\! \left(\frac{1}{M + 1}, \ldots, \frac{1}{M + 1}\right).
+\]</span> Simulation-based calibration checking uses this expected behavior to test the calibration of each parameter of a model on simulated data. <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span> suggest plotting binned counts of <span class="math inline">\(r_{1:N,
+k}\)</span> for different parameters <span class="math inline">\(k\)</span>; <span class="citation" data-cites="SailynojaEtAl:2022">Säilynoja, Bürkner, and Vehtari (<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">2022</a>)</span> provide a graphical test for discrete uniformity testing. Before uniformity testing the Markov chains should be thinned to remove autocorrelation as these uniformity tests assume independence <span class="citation" data-cites="SailynojaEtAl:2022">(<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">Säilynoja, Bürkner, and Vehtari 2022</a>)</span>.</p>
+</section>
+<section id="sbc-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="sbc-in-stan">SBC in Stan</h2>
+<p>Running simulation-based calibration checking in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.</p>
+<section id="example-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="example-model">Example model</h3>
+<p>For illustration, a very simple model will suffice. Suppose there are two parameters <span class="math inline">\((\mu, \sigma)\)</span> with independent priors, <span class="math display">\[
+\mu \sim \textrm{normal}(0, 1),
+\]</span> and <span class="math display">\[
+\sigma \sim \textrm{lognormal}(0, 1).
+\]</span> The data <span class="math inline">\(y = y_1, \ldots, y_N\)</span> is drawn conditionally independently given the parameters, <span class="math display">\[
+y_n \sim \textrm{normal}(\mu, \sigma).
+\]</span> The joint prior density is thus <span class="math display">\[
+p(\mu, \sigma)
+= \textrm{normal}(\mu \mid 0, 1)
+  \cdot \textrm{lognormal}(\sigma \mid 0, 1),
+\]</span> and the data model is <span class="math display">\[
+p(y \mid \mu, \sigma)
+= \prod_{n=1}^N \textrm{normal}(y_n \mid \mu, \sigma).
+\]</span></p>
+<p>For example, suppose the following two parameter values are drawn from the prior in the first simulation, <span class="math display">\[
+(\mu^{\textrm{sim(1)}}, \sigma^{\textrm{sim(1)}}) = (1.01, 0.23).
+\]</span> Then data <span class="math inline">\(y^{\textrm{sim}(1)} \sim p(y \mid \mu^{\textrm{sim(1)}},
+\sigma^{\textrm{sim(1)}})\)</span> is drawn according to the data model. Next, <span class="math inline">\(M = 4\)</span> draws are taken from the posterior <span class="math inline">\(\mu^{(1,m)}, \sigma^{(1,m)} \sim p(\mu, \sigma \mid y^{\textrm{sim}(1)})\)</span>, <span class="math display">\[
+\begin{array}{r|rr}
+m &amp; \mu^{(1,m)} &amp; \sigma^{(1,m)}
+\\ \hline
+1 &amp; 1.07 &amp; 0.33
+\\
+2 &amp; -0.32 &amp; 0.14
+\\
+3 &amp; -0.99 &amp; 0.26
+\\
+4 &amp; 1.51 &amp; 0.31
+\end{array}
+\]</span> Then the comparisons on which ranks are based look as follows, <span class="math display">\[
+\begin{array}{r|cc}
+m &amp; \textrm{I}(\mu^{(1,m)} &lt; \mu^{\textrm{sim}(1)})
+&amp; \textrm{I}(\sigma^{(1,m)} &lt; \sigma^{\textrm{sim}(1)})
+\\ \hline
+1 &amp; 0 &amp; 0
+\\
+2 &amp; 1 &amp; 1
+\\
+3 &amp; 1 &amp; 0
+\\
+4 &amp; 0 &amp; 0
+\end{array}
+\]</span> The ranks are the column sums, <span class="math inline">\(r_{1,1} = 2\)</span> and <span class="math inline">\(r_{1,2} = 1\)</span>. Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between <span class="math inline">\(0\)</span> and <span class="math inline">\(M\)</span>, the number of posterior draws.</p>
+</section>
+<section id="testing-a-stan-program-with-simulation-based-calibration-checking" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="testing-a-stan-program-with-simulation-based-calibration-checking">Testing a Stan program with simulation-based calibration checking</h3>
+<p>To code simulation-based calibration checking in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_sim = normal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_sim = lognormal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J = <span class="dv">10</span>;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[J] y_sim;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>    y_sim[j] = normal_rng(mu_sim, sigma_sim);</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>  y_sim ~ normal(mu, sigma);</span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; lt_sim</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>      = { mu &lt; mu_sim, sigma &lt; sigma_sim };</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To avoid confusion with the number of simulated data sets used for simulation-based calibration checking, <code>J</code> is used for the number of simulated data points.</p>
+<p>The model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.</p>
+</section>
+<section id="pseudocode-for-simulation-based-calibration-checking" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="pseudocode-for-simulation-based-calibration-checking">Pseudocode for simulation-based calibration checking</h3>
+<p>The entire simulation-based calibration checking process is as follows, where</p>
+<ul>
+<li><code>p(theta)</code> is the prior density</li>
+<li><code>p(y | theta)</code> is the sampling density</li>
+<li><code>K</code> is the number of parameters</li>
+<li><code>N</code> is the total number of simulated data sets and fits</li>
+<li><code>M</code> is the number of posterior draws per simulated data set</li>
+</ul>
+<pre><code>SBC(p(theta), p(y | theta), K, N, M)
+------------------------------------
+for (n in 1:N) {
+    // simulate parameters and data
+    theta(sim(n)) ~ p(theta)
+    y(sim(n)) ~ p(y | theta(sim(n)))
+
+    // posterior draws given simulated data
+    for (m in 1:M) {
+        theta(n, m) ~ p(theta | y(sim(n)))
+    }
+    // calculate rank of sim among posterior draws
+    for (k in 1:K) {
+        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))
+    }
+}
+// test uniformity of each parameter
+for (k in 1:K) {
+    test uniformity of rank(1:N, k)
+}</code></pre>
+</section>
+<section id="the-importance-of-thinning" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="the-importance-of-thinning">The importance of thinning</h3>
+<p>The draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws <span class="citation" data-cites="SailynojaEtAl:2022">(<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">Säilynoja, Bürkner, and Vehtari 2022</a>)</span>. Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of <code>M</code>, then thinning down to <code>M</code> draws.</p>
+</section>
+</section>
+<section id="testing-uniformity" class="level2">
+<h2 class="anchored" data-anchor-id="testing-uniformity">Testing uniformity</h2>
+<p>A simple, though not very highly powered, <span class="math inline">\(\chi^2\)</span>-squared test for uniformity can be formulated by binning the ranks <span class="math inline">\(0:M\)</span> into <span class="math inline">\(J\)</span> bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, <span class="citation" data-cites="SailynojaEtAl:2022">Säilynoja, Bürkner, and Vehtari (<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">2022</a>)</span> use binomial model pointiwise for the empirical cumlative distribution function and adjust to obtain simulatenous envelope to be used as graphical uniformity test.</p>
+<p>The bins don’t need to be exactly the same size. In general, if <span class="math inline">\(b_j\)</span> is the number of ranks that fall into bin <span class="math inline">\(j\)</span> and <span class="math inline">\(e_j\)</span> is the number of ranks expected to fall into bin <span class="math inline">\(j\)</span> (which will be proportional to its size under uniformity), the test statistic is <span class="math display">\[
+X^2 = \sum_{j = 1}^J \frac{(b_j - e_j)^2}{e_j}.
+\]</span> The terms are approximately square standard normal, so that under the null hypothesis of uniformity, <span class="math display">\[
+X^2 \sim \textrm{chiSquared}(J - 1),
+\]</span> with the corresponding <span class="math inline">\(p\)</span>-value given by the complementary cumulative distribution function (CCDF) of <span class="math inline">\(\textrm{chiSquared}(J - 1)\)</span> applied to <span class="math inline">\(X^2\)</span>. Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., <span class="math inline">\(e_j \geq 5.\)</span></p>
+<section id="indexing-to-simplify-arithmetic" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="indexing-to-simplify-arithmetic">Indexing to simplify arithmetic</h3>
+<p>Because there are <span class="math inline">\(M + 1\)</span> possible ranks, with <span class="math inline">\(J\)</span> bins, it is easiest to have <span class="math inline">\(M + 1\)</span> be divisible by <span class="math inline">\(J\)</span>. For instance, if <span class="math inline">\(J = 20\)</span> and <span class="math inline">\(M = 999\)</span>, then there are <span class="math inline">\(1000\)</span> possible ranks and an expected count in each bin of <span class="math inline">\(\frac{M + 1}{J} = 50.\)</span></p>
+<p>Distributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, <span class="math display">\[
+\textrm{bin}(r_{n, m}, M, J)
+= 1 + \left\lfloor \frac{r_{n, m}}{(M + 1) / J} \right\rfloor.
+\]</span> For example, with <span class="math inline">\(M = 999\)</span> and <span class="math inline">\(J = 20\)</span>, <span class="math inline">\((M + 1) / J = 50\)</span>. The lowest rank checks out, <span class="math display">\[
+\textrm{bin}(0, 999, 20) = 1 + \lfloor 0 / 50 \rfloor = 1,
+\]</span> as does the 50th rank, <span class="math display">\[
+\textrm{bin}(49, 999, 20) = 1 + \lfloor 49 / 50 \rfloor = 1,
+\]</span> and the 51st is appropriately put in the second bin, <span class="math display">\[
+\textrm{bin}(50, 999, 20) = 1 + \lfloor 50 / 50 \rfloor = 2.
+\]</span> The highest rank also checks out, with <span class="math inline">\(\textrm{bin}(1000, 999, 20) = 50.\)</span></p>
+<p>To summarize, the following pseudocode computes the <span class="math inline">\(b_j\)</span> values for the <span class="math inline">\(\chi^2\)</span> test or for visualization in a histogram.</p>
+<pre><code>Inputs: M draws, J bins, N parameters, ranks r[n, m]
+b[1:J] = 0
+for (m in 1:M) {
+  ++b[1 + floor(r[n, m] * J / (M + 1))]
+}</code></pre>
+<p>where the <code>++b[n]</code> notation is a common form of syntactic sugar for <code>b[n] = b[n] + 1.</code></p>
+<p>In general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.</p>
+</section>
+</section>
+<section id="examples-of-simulation-based-calibration-checking" class="level2">
+<h2 class="anchored" data-anchor-id="examples-of-simulation-based-calibration-checking">Examples of simulation-based calibration checking</h2>
+<p>This section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.</p>
+<section id="when-things-go-right" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="when-things-go-right">When things go right</h3>
+<p>Consider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. <span class="math display">\[\begin{eqnarray*}
+\mu &amp; \sim &amp; \textrm{normal}(0, 1)
+\\[4pt]
+\sigma &amp; \sim &amp; \textrm{lognormal}(0, 1)
+\\[4pt]
+y_{1:10} &amp; \sim &amp; \textrm{normal}(\mu, \sigma).
+\end{eqnarray*}\]</span> The Stan program for evaluating SBC for this model is</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_sim = normal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_sim = lognormal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J = <span class="dv">10</span>;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[J] y_sim;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>    y_sim[j] = student_t_rng(<span class="dv">4</span>, mu_sim, sigma_sim);</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>  y_sim ~ normal(mu, sigma);</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; I_lt_sim</span>
+<span id="cb4-23"><a href="#cb4-23" aria-hidden="true" tabindex="-1"></a>      = { mu &lt; mu_sim, sigma &lt; sigma_sim };</span>
+<span id="cb4-24"><a href="#cb4-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>After running this for enough iterations so that the effective sample size is larger than <span class="math inline">\(M\)</span>, then thinning to <span class="math inline">\(M\)</span> draws (here <span class="math inline">\(M = 999\)</span>), the ranks are computed and binned, and then plotted.</p>
+<div id="fig-sbc-good" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-float-fig figure">
+<div aria-describedby="fig-sbc-good-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="./img/sbc-normal-normal.png" class="img-fluid figure-img" style="width:90.0%">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-sbc-good-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+Figure&nbsp;1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.
+</figcaption>
+</figure>
+</div>
+</section>
+<section id="when-things-go-wrong" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="when-things-go-wrong">When things go wrong</h3>
+<p>Now consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.</p>
+<div id="fig-sbc-bad" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-float-fig figure">
+<div aria-describedby="fig-sbc-bad-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="./img/sbc-student-t-normal.png" class="img-fluid figure-img" style="width:90.0%">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-sbc-bad-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+Figure&nbsp;2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.
+</figcaption>
+</figure>
+</div>
+</section>
+<section id="when-stans-sampler-goes-wrong" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="when-stans-sampler-goes-wrong">When Stan’s sampler goes wrong</h3>
+<p>The example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.</p>
+<p>This can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means <span class="math inline">\(y_j\)</span> and standard deviations <span class="math inline">\(\sigma_j\)</span> of differences in test score after the same intervention in <span class="math inline">\(J = 8\)</span> different schools. <span class="citation" data-cites="Rubin:1981">Rubin (<a href="#ref-Rubin:1981" role="doc-biblioref">1981</a>)</span> applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span>; this section replicates their results.</p>
+<p>The meta-analysis model has parameters for a population mean <span class="math inline">\(\mu\)</span> and standard deviation <span class="math inline">\(\tau &gt; 0\)</span> as well as the effect <span class="math inline">\(\theta_j\)</span> of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, <span class="math display">\[\begin{eqnarray*}
+\mu &amp; \sim &amp; \textrm{normal}(0, 5)
+\\[4pt]
+\tau &amp; \sim &amp; \textrm{normal}_{+}(0, 5).
+\end{eqnarray*}\]</span> School level effects are modeled as normal given the population parameters, <span class="math display">\[
+\theta_j \sim \textrm{normal}(\mu, \tau).
+\]</span> The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, <span class="math display">\[
+y_j \sim \textrm{normal}(\theta_j, \sigma_j).
+\]</span></p>
+<p>This model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_sim = normal_rng(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> tau_sim = abs(normal_rng(<span class="dv">0</span>, <span class="dv">5</span>));</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J = <span class="dv">8</span>;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma = abs(normal_rng(rep_vector(<span class="dv">0</span>, J), <span class="dv">5</span>));</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> y = normal_rng(theta_sim, sigma);</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; tau;</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> theta;</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>  tau ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(mu, tau);</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>  y ~ normal(theta, sigma);</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; mu_lt_sim = mu &lt; mu_sim;</span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; tau_lt_sim = tau &lt; tau_sim;</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta1_lt_sim = theta[<span class="dv">1</span>] &lt; theta_sim[<span class="dv">1</span>];</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As usual for simulation-based calibration checking, the transformed data encodes the data-generating process using random number generators. Here, the population parameters <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\tau\)</span> are first simulated, then the school-level effects <span class="math inline">\(\theta\)</span>, and then finally the observed data <span class="math inline">\(\sigma_j\)</span> and <span class="math inline">\(y_j.\)</span> The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only <span class="math inline">\(\theta_1\)</span> because the schools are exchangeable in the simulation.</p>
+<p>When fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.</p>
+<p>SBC also diagnoses the problem. Here’s the rank plots for running <span class="math inline">\(N =
+200\)</span> simulations with 1000 warmup iterations and <span class="math inline">\(M = 999\)</span> draws per simulation used to compute the ranks.</p>
+<div id="fig-8schools" class="quarto-layout-panel">
+<figure class="quarto-float quarto-float-fig figure">
+<div aria-describedby="fig-8schools-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<div class="quarto-layout-row">
+<div class="quarto-layout-cell-subref quarto-layout-cell" data-ref-parent="fig-8schools" style="flex-basis: 33.3%;justify-content: flex-start;">
+<div id="fig-mu" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-subfloat-fig figure">
+<div aria-describedby="fig-mu-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="img/sbc-ctr-8-schools-mu.png" class="img-fluid figure-img" data-ref-parent="fig-8schools">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-subfloat-caption quarto-subfloat-fig" id="fig-mu-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+(a) <span class="math inline">\(\mu\)</span>
+</figcaption>
+</figure>
+</div>
+</div>
+<div class="quarto-layout-cell-subref quarto-layout-cell" data-ref-parent="fig-8schools" style="flex-basis: 33.3%;justify-content: flex-start;">
+<div id="fig-tau" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-subfloat-fig figure">
+<div aria-describedby="fig-tau-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="img/sbc-ctr-8-schools-tau.png" class="img-fluid figure-img" data-ref-parent="fig-8schools">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-subfloat-caption quarto-subfloat-fig" id="fig-tau-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+(b) <span class="math inline">\(\tau\)</span>
+</figcaption>
+</figure>
+</div>
+</div>
+<div class="quarto-layout-cell-subref quarto-layout-cell" data-ref-parent="fig-8schools" style="flex-basis: 33.3%;justify-content: flex-start;">
+<div id="fig-theta1" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-subfloat-fig figure">
+<div aria-describedby="fig-theta1-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="img/sbc-ctr-8-schools-theta1.png" class="img-fluid figure-img" data-ref-parent="fig-8schools">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-subfloat-caption quarto-subfloat-fig" id="fig-theta1-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+(c) <span class="math inline">\({\theta}_1\)</span>
+</figcaption>
+</figure>
+</div>
+</div>
+</div>
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-8schools-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+Figure&nbsp;3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for <span class="math inline">\(\theta_1\)</span> (<a href="#fig-theta1" class="quarto-xref">Figure&nbsp;3 (c)</a>).
+</figcaption>
+</figure>
+</div>
+<p>Although the population mean and standard deviation <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\tau\)</span> appear well calibrated, <span class="math inline">\(\theta_1\)</span> tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as <span class="math inline">\(\tau\)</span> approaches zero and the <span class="math inline">\(\theta_j\)</span> become highly constrained. The <a href="../stan-users-guide/reparameterization.html">chapter on reparameterization</a> explains how to remedy this problem and fit this kind of hierarchical model with Stan.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-CookGelmanRubin:2006" class="csl-entry" role="listitem">
+Cook, Samantha R., Andrew Gelman, and Donald B Rubin. 2006. <span>“Validation of Software for <span>B</span>ayesian Models Using Posterior Quantiles.”</span> <em>Journal of Computational and Graphical Statistics</em> 15 (3): 675–92. <a href="https://doi.org/10.1198/106186006X136976">https://doi.org/10.1198/106186006X136976</a>.
+</div>
+<div id="ref-Dawid:1982" class="csl-entry" role="listitem">
+Dawid, A Philip. 1982. <span>“The Well-Calibrated <span>B</span>ayesian.”</span> <em>Journal of the American Statistical Association</em> 77 (379): 605–10.
+</div>
+<div id="ref-GneitingEtAl:2007" class="csl-entry" role="listitem">
+Gneiting, Tilmann, Fadoua Balabdaoui, and Adrian E Raftery. 2007. <span>“Probabilistic Forecasts, Calibration and Sharpness.”</span> <em>Journal of the Royal Statistical Society: Series B (Statistical Methodology)</em> 69 (2): 243–68.
+</div>
+<div id="ref-ModrakEtAl:2023" class="csl-entry" role="listitem">
+Modrák, Martin, Angie H Moon, Shinyoung Kim, Paul Bürkner, Niko Huurre, Kateřina Faltejsková, Andrew Gelman, and Aki Vehtari. 2023. <span>“Simulation-Based Calibration Checking for <span>Bayesian</span> Computation: The Choice of Test Quantities Shapes Sensitivity.”</span> <em>Bayesian Analysis</em> 20 (2): 461.
+</div>
+<div id="ref-Rubin:1981" class="csl-entry" role="listitem">
+Rubin, Donald B. 1981. <span>“Estimation in Parallel Randomized Experiments.”</span> <em>Journal of Educational Statistics</em> 6: 377–401.
+</div>
+<div id="ref-SailynojaEtAl:2022" class="csl-entry" role="listitem">
+Säilynoja, Teemu, Paul-Christian Bürkner, and Aki Vehtari. 2022. <span>“Graphical Test for Discrete Uniformity and Its Applications in Goodness-of-Fit Evaluation and Multiple Sample Comparison.”</span> <em>Statistics and Computing</em> 32 (2): 32.
+</div>
+<div id="ref-TaltsEtAl:2018" class="csl-entry" role="listitem">
+Talts, Sean, Michael Betancourt, Daniel Simpson, Aki Vehtari, and Andrew Gelman. 2018. <span>“Validating <span>B</span>ayesian Inference Algorithms with Simulation-Based Calibration.”</span> <em>arXiv</em>, no. 1804.06788.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/posterior-prediction.html" class="pagination-link" aria-label="Posterior Predictive Sampling">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Posterior Predictive Sampling</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/posterior-predictive-checks.html" class="pagination-link" aria-label="Posterior and Prior Predictive Checks">
+        <span class="nav-page-text">Posterior and Prior Predictive Checks</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/simulation-based-calibration.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/sparse-ragged.html b/docs/2_39/stan-users-guide/sparse-ragged.html
new file mode 100644
index 000000000..f7b72e969
--- /dev/null
+++ b/docs/2_39/stan-users-guide/sparse-ragged.html
@@ -0,0 +1,1371 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Sparse and Ragged Data Structures</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/clustering.html" rel="next">
+<link href="../stan-users-guide/latent-discrete.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Sparse and Ragged Data Structures">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/sparse-ragged.html">Sparse and Ragged Data Structures</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#sparse-ragged.chapter" id="toc-sparse-ragged.chapter" class="nav-link active" data-scroll-target="#sparse-ragged.chapter">Sparse and Ragged Data Structures</a>
+  <ul class="collapse">
+  <li><a href="#sparse-data-structures" id="toc-sparse-data-structures" class="nav-link" data-scroll-target="#sparse-data-structures">Sparse data structures</a></li>
+  <li><a href="#ragged-data-structs.section" id="toc-ragged-data-structs.section" class="nav-link" data-scroll-target="#ragged-data-structs.section">Ragged data structures</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/sparse-ragged.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/sparse-ragged.html">Sparse and Ragged Data Structures</a></li></ol></nav></header>
+
+
+
+
+<section id="sparse-ragged.chapter" class="level1">
+<h1>Sparse and Ragged Data Structures</h1>
+<p>Stan does not directly support either sparse or ragged data structures, though both can be accommodated with some programming effort. The <a href="https://mc-stan.org/docs/functions-reference/sparse_matrix_operations.html">sparse matrices chapter</a> introduces a special-purpose sparse matrix times dense vector multiplication, which should be used where applicable; this chapter covers more general data structures.</p>
+<section id="sparse-data-structures" class="level2">
+<h2 class="anchored" data-anchor-id="sparse-data-structures">Sparse data structures</h2>
+<p>Coding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the <a href="../stan-users-guide/regression.html#item-response-models.section">item-response model section</a>. There are <span class="math inline">\(J\)</span> students and <span class="math inline">\(K\)</span> questions, and if every student answers every question, then it is practical to declare the data as a <span class="math inline">\(J \times K\)</span> array of answers.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J, K] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.</p>
+<p>The following missing data example shows an example with <span class="math inline">\(J=3\)</span> and <span class="math inline">\(K=4\)</span>, with missing responses shown as NA, as in R.</p>
+<p><span class="math display">\[\begin{equation*}
+y
+=
+\left[
+\begin{array}{cccc}
+0 &amp; 1 &amp; \mbox{NA} &amp; 1
+\\
+0 &amp; \mbox{NA} &amp; \mbox{NA} &amp; 1
+\\
+\mbox{NA} &amp; 0 &amp; \mbox{NA} &amp; \mbox{NA}
+\end{array}
+\right]
+\end{equation*}\]</span></p>
+<p>There is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns <span class="math inline">\(jj\)</span> and <span class="math inline">\(kk\)</span> used for the indexes (following <span class="citation" data-cites="GelmanHill:2007">Gelman and Hill (<a href="#ref-GelmanHill:2007" role="doc-biblioref">2007</a>)</span>), the 2-D array <span class="math inline">\(y\)</span> is recoded as a table. The number of rows in the table equals the number of defined array elements, here <span class="math inline">\(y_{1,1} = 0\)</span>, <span class="math inline">\(y_{1,2} = 1\)</span>, up to <span class="math inline">\(y_{3,2} = 1\)</span>. As the array becomes larger and sparser, the long form becomes the more economical encoding.</p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;"><em>jj</em></th>
+<th style="text-align: left;"><em>kk</em></th>
+<th style="text-align: left;"><em>y</em></th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;">1</td>
+<td style="text-align: left;">1</td>
+<td style="text-align: left;">0</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">1</td>
+<td style="text-align: left;">2</td>
+<td style="text-align: left;">1</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">1</td>
+<td style="text-align: left;">4</td>
+<td style="text-align: left;">1</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">2</td>
+<td style="text-align: left;">1</td>
+<td style="text-align: left;">0</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;">2</td>
+<td style="text-align: left;">4</td>
+<td style="text-align: left;">1</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;">3</td>
+<td style="text-align: left;">2</td>
+<td style="text-align: left;">0</td>
+</tr>
+</tbody>
+</table>
+<p>Letting <span class="math inline">\(N\)</span> be the number of <span class="math inline">\(y\)</span> that are defined, here <span class="math inline">\(N=6\)</span>, the data and model can be formulated as follows.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli_logit(delta[kk[n]]</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>                           * (alpha[jj[n]] - beta[kk[n]]));</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.</p>
+</section>
+<section id="ragged-data-structs.section" class="level2">
+<h2 class="anchored" data-anchor-id="ragged-data-structs.section">Ragged data structures</h2>
+<p>Ragged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.</p>
+<p>A general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.</p>
+<p>For example, consider a data structure for three groups, each of which has a different number of observations.</p>
+<div>
+
+</div>
+<div class="quarto-layout-panel" data-layout-ncol="2">
+<div class="quarto-layout-row">
+<div id="first-column" class="quarto-layout-cell" style="flex-basis: 50.0%;justify-content: flex-start;">
+<p><span class="math inline">\(y_1 =  \left[1.3 \ \ 2.4 \ \ 0.9\right]\\\)</span> <span class="math inline">\(y_2 = \left[-1.8 \ \ -0.1\right]\\\)</span> <span class="math inline">\(y_3 = \left[12.9 \ \ 18.7 \ \ 42.9 \ \ 4.7\right]\)</span></p>
+</div>
+<div id="second-column" class="quarto-layout-cell" style="flex-basis: 50.0%;justify-content: flex-start;">
+<p><span class="math inline">\(z = [1.3 \ \ 2.4 \ \ 0.9 \ \ -1.8 \ \ -0.1 \ \ 12.9 \ \ 18.7 \ \ 42.9 \ \ 4.7]\\\)</span> <span class="math inline">\(s  =  \{ 3 \ \ 2 \ \ 4 \}\)</span></p>
+</div>
+</div>
+</div>
+<p>On the left is the definition of a ragged data structure <span class="math inline">\(y\)</span> with three rows of different sizes (<span class="math inline">\(y_1\)</span> is size 3, <span class="math inline">\(y_2\)</span> size 2, and <span class="math inline">\(y_3\)</span> size 4). On the right is an example of how to code the data in Stan, using a single vector <span class="math inline">\(z\)</span> to hold all the values and a separate array of integers <span class="math inline">\(s\)</span> to hold the group row sizes. In this example, <span class="math inline">\(y_1 = z_{1:3}\)</span>, <span class="math inline">\(y_2 = z_{4:5}\)</span>, and <span class="math inline">\(y_3 = z_{6:9}\)</span>.</p>
+<p>Suppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood <span class="math display">\[\begin{equation*}
+\sum_{n=1}^3 \log \textsf{normal}(y_n \mid \mu_n, \sigma).
+\end{equation*}\]</span> There’s no direct way to encode this in Stan.</p>
+<p>A full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.</p>
+<p>The model can then be coded up using slicing operations as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;   <span class="co">// # observations</span></span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;   <span class="co">// # of groups</span></span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;      <span class="co">// observations</span></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">int</span> s;   <span class="co">// group sizes</span></span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> pos;</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  pos = <span class="dv">1</span>;</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>    segment(y, pos, s[k]) ~ normal(mu[k], sigma);</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>    pos = pos + s[k];</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This coding allows for efficient vectorization, which is worth the copy cost entailed by the <code>segment()</code> vector slicing operation.</p>
+
+
+
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-GelmanHill:2007" class="csl-entry" role="listitem">
+Gelman, Andrew, and Jennifer Hill. 2007. <em>Data Analysis Using Regression and Multilevel-Hierarchical Models</em>. Cambridge, United Kingdom: Cambridge University Press.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/latent-discrete.html" class="pagination-link" aria-label="Latent Discrete Parameters">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Latent Discrete Parameters</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/clustering.html" class="pagination-link" aria-label="Clustering Models">
+        <span class="nav-page-text">Clustering Models</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/sparse-ragged.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/style-guide.html b/docs/2_39/stan-users-guide/style-guide.html
new file mode 100644
index 000000000..d2fc86cd0
--- /dev/null
+++ b/docs/2_39/stan-users-guide/style-guide.html
@@ -0,0 +1,1467 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Stan Program Style Guide</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/for-bugs-users.html" rel="next">
+<link href="../stan-users-guide/using-stanc.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Stan Program Style Guide">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/style-guide.html">Stan Program Style Guide</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#stan-program-style-guide" id="toc-stan-program-style-guide" class="nav-link active" data-scroll-target="#stan-program-style-guide">Stan Program Style Guide</a>
+  <ul class="collapse">
+  <li><a href="#choose-a-consistent-style" id="toc-choose-a-consistent-style" class="nav-link" data-scroll-target="#choose-a-consistent-style">Choose a consistent style</a></li>
+  <li><a href="#line-length" id="toc-line-length" class="nav-link" data-scroll-target="#line-length">Line length</a></li>
+  <li><a href="#file-extensions" id="toc-file-extensions" class="nav-link" data-scroll-target="#file-extensions">File extensions</a></li>
+  <li><a href="#variable-naming" id="toc-variable-naming" class="nav-link" data-scroll-target="#variable-naming">Variable naming</a></li>
+  <li><a href="#local-variable-scope" id="toc-local-variable-scope" class="nav-link" data-scroll-target="#local-variable-scope">Local variable scope</a></li>
+  <li><a href="#parentheses-and-brackets" id="toc-parentheses-and-brackets" class="nav-link" data-scroll-target="#parentheses-and-brackets">Parentheses and brackets</a>
+  <ul class="collapse">
+  <li><a href="#braces-for-single-statement-blocks" id="toc-braces-for-single-statement-blocks" class="nav-link" data-scroll-target="#braces-for-single-statement-blocks">Braces for single-statement blocks</a></li>
+  <li><a href="#parentheses-in-nested-operator-expressions" id="toc-parentheses-in-nested-operator-expressions" class="nav-link" data-scroll-target="#parentheses-in-nested-operator-expressions">Parentheses in nested operator expressions</a></li>
+  <li><a href="#no-open-brackets-on-own-line" id="toc-no-open-brackets-on-own-line" class="nav-link" data-scroll-target="#no-open-brackets-on-own-line">No open brackets on own line</a></li>
+  </ul></li>
+  <li><a href="#conditionals" id="toc-conditionals" class="nav-link" data-scroll-target="#conditionals">Conditionals</a></li>
+  <li><a href="#functions" id="toc-functions" class="nav-link" data-scroll-target="#functions">Functions</a></li>
+  <li><a href="#white-space" id="toc-white-space" class="nav-link" data-scroll-target="#white-space">White space</a>
+  <ul class="collapse">
+  <li><a href="#line-breaks-between-statements-and-declarations" id="toc-line-breaks-between-statements-and-declarations" class="nav-link" data-scroll-target="#line-breaks-between-statements-and-declarations">Line breaks between statements and declarations</a></li>
+  <li><a href="#no-tabs" id="toc-no-tabs" class="nav-link" data-scroll-target="#no-tabs">No tabs</a></li>
+  <li><a href="#two-character-indents" id="toc-two-character-indents" class="nav-link" data-scroll-target="#two-character-indents">Two-character indents</a></li>
+  <li><a href="#space-between-if-and-condition" id="toc-space-between-if-and-condition" class="nav-link" data-scroll-target="#space-between-if-and-condition">Space between <code>if</code>, <code>{</code> and condition</a></li>
+  <li><a href="#no-space-for-function-calls" id="toc-no-space-for-function-calls" class="nav-link" data-scroll-target="#no-space-for-function-calls">No space for function calls</a></li>
+  <li><a href="#spaces-around-operators" id="toc-spaces-around-operators" class="nav-link" data-scroll-target="#spaces-around-operators">Spaces around operators</a></li>
+  <li><a href="#no-spaces-in-type-constraints" id="toc-no-spaces-in-type-constraints" class="nav-link" data-scroll-target="#no-spaces-in-type-constraints">No spaces in type constraints</a></li>
+  <li><a href="#breaking-expressions-across-lines" id="toc-breaking-expressions-across-lines" class="nav-link" data-scroll-target="#breaking-expressions-across-lines">Breaking expressions across lines</a></li>
+  <li><a href="#spaces-after-commas" id="toc-spaces-after-commas" class="nav-link" data-scroll-target="#spaces-after-commas">Spaces after commas</a></li>
+  <li><a href="#unix-newlines" id="toc-unix-newlines" class="nav-link" data-scroll-target="#unix-newlines">Unix newlines</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/style-guide.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/style-guide.html">Stan Program Style Guide</a></li></ol></nav></header>
+
+
+
+
+<section id="stan-program-style-guide" class="level1">
+<h1>Stan Program Style Guide</h1>
+<p>This chapter describes the preferred style for laying out Stan models. These are not rules of the language, but simply recommendations for laying out programs in a text editor. Although these recommendations may seem arbitrary, they are similar to those of many teams for many programming languages. Like rules for typesetting text, the goal is to achieve readability without wasting white space either vertically or horizontally. This is the style used in the Stan documentation, and should align with the auto-formatting ability of <code>stanc3</code>.</p>
+<section id="choose-a-consistent-style" class="level2">
+<h2 class="anchored" data-anchor-id="choose-a-consistent-style">Choose a consistent style</h2>
+<p>The most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.</p>
+</section>
+<section id="line-length" class="level2">
+<h2 class="anchored" data-anchor-id="line-length">Line length</h2>
+<p>Line lengths should not exceed 80 characters.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<p>This is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.</p>
+</section>
+<section id="file-extensions" class="level2">
+<h2 class="anchored" data-anchor-id="file-extensions">File extensions</h2>
+<p>The recommended file extension for Stan model files is <code>.stan</code>. Files which contain only function definitions (intended for use with <code>#include</code>) should be given the <code>.stanfunctions</code> extension. A <code>.stanfunctions</code> file only includes the function definition and does not require the <code>functions{}</code> block wrapped around the function. A simple example of usage where the function is defined and saved in the file <code>foo.stanfunctions</code>:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> x, <span class="dt">real</span> y) {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> sqrt(x * log(y));</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function <code>foo</code> can be accessed in the Stan program by including the path to the <code>foo.stanfunctions</code> file as:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">#include foo.stanfunctions;</span></span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="co">// ...body...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For Stan data dump files, the recommended extension is <code>.R</code>, or more informatively, <code>.data.R</code>. For JSON output, the recommended extension is <code>.json</code>.</p>
+</section>
+<section id="variable-naming" class="level2">
+<h2 class="anchored" data-anchor-id="variable-naming">Variable naming</h2>
+<p>The recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (<code>_</code>) used as a separator. Thus it is preferred to use <code>sigma_y</code>, rather than the run together <code>sigmay</code>, camel-case <code>sigmaY</code>, or capitalized camel-case <code>SigmaY</code>. An exception is often made for terms appearing in mathematical expressions with standard names, like <code>A</code> for a matrix.</p>
+<p>Another exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an <span class="math inline">\(M \times N\)</span> matrix <span class="math inline">\(a\)</span> would look as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>     a[m, n] = ...</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="local-variable-scope" class="level2">
+<h2 class="anchored" data-anchor-id="local-variable-scope">Local variable scope</h2>
+<p>Declaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.</p>
+<p>The following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of <code>mu</code> in each iteration.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> mu;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>    mu[n] = alpha * x[n] + beta;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(mu[n],sigma);</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> mu;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>    mu = alpha * x[n] + beta;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(mu,sigma);</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The local variable can be eliminated altogether, as follows.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(alpha * x[n] + beta, sigma);</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.</p>
+<section id="scope-of-compound-structures-with-componentwise-assignment" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="scope-of-compound-structures-with-componentwise-assignment">Scope of compound structures with componentwise assignment</h4>
+<p>In the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] mu;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>      mu[k] = <span class="co">// ...</span></span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>    y[n] ~ multi_normal(mu,Sigma);</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this case, the vector <code>mu</code> will be allocated outside of both loops, and used a total of <code>N</code> times.</p>
+</section>
+</section>
+<section id="parentheses-and-brackets" class="level2">
+<h2 class="anchored" data-anchor-id="parentheses-and-brackets">Parentheses and brackets</h2>
+<section id="braces-for-single-statement-blocks" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="braces-for-single-statement-blocks">Braces for single-statement blocks</h3>
+<p>Single-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu,<span class="dv">1</span>);</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The use of loops and conditionals without brackets can be dangerous. For instance, consider this program.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N)</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  z[n] ~ normal(nu,<span class="dv">1</span>);</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu,<span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  z[n] ~ normal(nu,<span class="dv">1</span>);</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>y[n] ~ normal(mu,<span class="dv">1</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Therefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the <code>else</code> branch contains exactly one conditional. Then, it is preferred to place the following <code>if</code> on the same line, as in the following.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (x) {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> <span class="cf">if</span> (y) {</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>} <span class="cf">else</span> {</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="parentheses-in-nested-operator-expressions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="parentheses-in-nested-operator-expressions">Parentheses in nested operator expressions</h3>
+<p>The preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression <code>a + b * c</code> is preferred to the equivalent <code>a + (b * c)</code> or <code>(a + (b * c))</code>. The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.</p>
+<p>Similarly, comparison operators can usually be written with minimal bracketing, with the form <code>y[n] &gt; 0 || x[n] != 0</code> preferred to the bracketed form <code>(y[n] &gt; 0) || (x[n] != 0)</code>.</p>
+</section>
+<section id="no-open-brackets-on-own-line" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="no-open-brackets-on-own-line">No open brackets on own line</h3>
+<p>Vertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(mu,<span class="dv">1</span>);</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This also goes for parameters blocks, transformed data blocks, which should look as follows.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.</p>
+</section>
+</section>
+<section id="conditionals" class="level2">
+<h2 class="anchored" data-anchor-id="conditionals">Conditionals</h2>
+<p>While Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued <code>x</code>, one should use</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (x != <span class="dv">0</span>) { ...</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>in place of</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (x) { ...</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Beyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.</p>
+</section>
+<section id="functions" class="level2">
+<h2 class="anchored" data-anchor-id="functions">Functions</h2>
+<p>Functions are laid out the same way as in languages such as Java and C++. For example,</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> x, <span class="dt">real</span> y) {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> sqrt(x * log(y));</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.</p>
+<p>If function names or argument lists are long, they can be written as</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span></span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>function_to_do_some_hairy_algebra(<span class="dt">matrix</span> thingamabob,</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>                                  <span class="dt">vector</span> doohickey2) {</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...body...</span></span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function starts a new line, under the type. The arguments are aligned under each other.</p>
+<p>Function documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the <a href="../stan-users-guide/user-functions.html#documenting-functions.section">documenting functions section</a>.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="co">/**</span></span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a><span class="co"> * Return a data matrix of specified size with rows</span></span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="co"> * corresponding to items and the first column filled</span></span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a><span class="co"> * with the value 1 to represent the intercept and the</span></span>
+<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a><span class="co"> * remaining columns randomly filled with unit-normal draws.</span></span>
+<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a><span class="co"> *</span></span>
+<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@param</span><span class="co"> N Number of rows correspond to data items</span></span>
+<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@param</span><span class="co"> K Number of predictors, counting the intercept, per</span></span>
+<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a><span class="co"> *          item.</span></span>
+<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@return</span><span class="co"> Simulated predictor matrix.</span></span>
+<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a><span class="co"> */</span></span>
+<span id="cb18-12"><a href="#cb18-12" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> predictors_rng(<span class="dt">int</span> N, <span class="dt">int</span> K) {</span>
+<span id="cb18-13"><a href="#cb18-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb18-14"><a href="#cb18-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The open comment is <code>/**</code>, asterisks are aligned below the first asterisk of the open comment, and the end comment <code>*/</code> is also aligned on the asterisk. The tags <code>@param</code> and <code>@return</code> are used to label function arguments (i.e., parameters) and return values.</p>
+</section>
+<section id="white-space" class="level2">
+<h2 class="anchored" data-anchor-id="white-space">White space</h2>
+<p>Stan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII <code>0x20</code>), line feed (ASCII <code>0x0A</code>), carriage return (<code>0x0D</code>), and tab (<code>0x09</code>). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.</p>
+<section id="line-breaks-between-statements-and-declarations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="line-breaks-between-statements-and-declarations">Line breaks between statements and declarations</h3>
+<p>Each statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> mu, sigma;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="no-tabs" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="no-tabs">No tabs</h3>
+<p>Stan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.</p>
+</section>
+<section id="two-character-indents" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="two-character-indents">Two-character indents</h3>
+<p>Stan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.</p>
+</section>
+<section id="space-between-if-and-condition" class="level3">
+<h3 class="anchored" data-anchor-id="space-between-if-and-condition">Space between <code>if</code>, <code>{</code> and condition</h3>
+<p>Use a space after <code>if</code>s. For instance, use <code>if (x &lt; y) {...</code>, not <code>if(x &lt; y){ ...</code>.</p>
+</section>
+<section id="no-space-for-function-calls" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="no-space-for-function-calls">No space for function calls</h3>
+<p>There should not be space between a function name and the arguments it applies to. For instance, use <code>normal(0, 1)</code>, not <code>normal (0,1)</code>.</p>
+</section>
+<section id="spaces-around-operators" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="spaces-around-operators">Spaces around operators</h3>
+<p>There should be spaces around binary operators. For instance, use <code>y[1] = x</code>, not <code>y[1]=x</code>, use <code>(x + y) * z</code> not <code>(x+y)*z</code>.</p>
+<p>Unary operators are written without a space, such as in <code>-x</code>, <code>!y</code>.</p>
+</section>
+<section id="no-spaces-in-type-constraints" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="no-spaces-in-type-constraints">No spaces in type constraints</h3>
+<p>Another exception to the above rule is when the assignment operator (<code>=</code>) is used inside a type constraint, such as</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Spaces should still be used in arithmetic and following commas, as in</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=a * x + b&gt; x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="breaking-expressions-across-lines" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="breaking-expressions-across-lines">Breaking expressions across lines</h3>
+<p>Sometimes expressions are too long to fit on a single line. In that case, the recommended form is to break <em>before</em> an operator,<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> aligning the operator to a term above to indicate scoping. For example, use the following form</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[J] p_distance = Phi((distance_tolerance - overshot)</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a>                           ./ ((x + overshot) * sigma_distance))</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>                       - Phi((-overshot)</span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a>                             ./ ((x + overshot) * sigma_distance));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here, the elementwise division operator (<code>./</code>) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (<code>Phi</code>).</p>
+<p>For functions with multiple arguments, break after a comma and line the next argument up underneath as follows.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ normal(alpha + beta * x + gamma * y,</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>              pow(tau,-<span class="fl">0.5</span>));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="spaces-after-commas" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="spaces-after-commas">Spaces after commas</h3>
+<p>Commas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.</p>
+<p>For example,</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a>normal(alpha * x[n] + beta, sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>is preferred over</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a>normal(alpha * x[n] + beta,sigma);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="unix-newlines" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unix-newlines">Unix newlines</h3>
+<p>Wherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.</p>
+<section id="platform-specificity-of-newlines" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="platform-specificity-of-newlines">Platform specificity of newlines</h4>
+<p>Newlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point <code>0x0A</code>). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point <code>0x0D</code>) followed by a line-feed (LF) character.</p>
+
+
+</section>
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Even 80 characters may be too many for rendering in print; for instance, in this manual, the number of code characters that fit on a line is about 65.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>This is the usual convention in both typesetting and other programming languages. Neither R nor BUGS allows breaks before an operator because they allow newlines to signal the end of an expression or statement.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/using-stanc.html" class="pagination-link" aria-label="Using the Stan Compiler">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Using the Stan Compiler</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/for-bugs-users.html" class="pagination-link" aria-label="Transitioning from BUGS">
+        <span class="nav-page-text">Transitioning from BUGS</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/style-guide.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/survival.html b/docs/2_39/stan-users-guide/survival.html
new file mode 100644
index 000000000..314136c57
--- /dev/null
+++ b/docs/2_39/stan-users-guide/survival.html
@@ -0,0 +1,1670 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Survival Models</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/wiener_diffusion_model.html" rel="next">
+<link href="../stan-users-guide/dae.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Survival Models">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/survival.html">Survival Models</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#survival-models" id="toc-survival-models" class="nav-link active" data-scroll-target="#survival-models">Survival Models</a>
+  <ul class="collapse">
+  <li><a href="#exponential-survival-model" id="toc-exponential-survival-model" class="nav-link" data-scroll-target="#exponential-survival-model">Exponential survival model</a>
+  <ul class="collapse">
+  <li><a href="#stan-program" id="toc-stan-program" class="nav-link" data-scroll-target="#stan-program">Stan program</a></li>
+  </ul></li>
+  <li><a href="#weibull-survival-model" id="toc-weibull-survival-model" class="nav-link" data-scroll-target="#weibull-survival-model">Weibull survival model</a>
+  <ul class="collapse">
+  <li><a href="#stan-program-1" id="toc-stan-program-1" class="nav-link" data-scroll-target="#stan-program-1">Stan program</a></li>
+  </ul></li>
+  <li><a href="#survival-with-covariates" id="toc-survival-with-covariates" class="nav-link" data-scroll-target="#survival-with-covariates">Survival with covariates</a></li>
+  <li><a href="#hazard-and-survival-functions" id="toc-hazard-and-survival-functions" class="nav-link" data-scroll-target="#hazard-and-survival-functions">Hazard and survival functions</a></li>
+  <li><a href="#proportional-hazards-model" id="toc-proportional-hazards-model" class="nav-link" data-scroll-target="#proportional-hazards-model">Proportional hazards model</a>
+  <ul class="collapse">
+  <li><a href="#partial-likelihood-function" id="toc-partial-likelihood-function" class="nav-link" data-scroll-target="#partial-likelihood-function">Partial likelihood function</a></li>
+  <li><a href="#stan-program-2" id="toc-stan-program-2" class="nav-link" data-scroll-target="#stan-program-2">Stan program</a></li>
+  <li><a href="#stan-model-for-tied-survival-times" id="toc-stan-model-for-tied-survival-times" class="nav-link" data-scroll-target="#stan-model-for-tied-survival-times">Stan model for tied survival times</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/survival.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/survival.html">Survival Models</a></li></ol></nav></header>
+
+
+
+
+<section id="survival-models" class="level1">
+<h1>Survival Models</h1>
+<p>Survival models apply to animals and plants as well as inanimate objects such as machine parts or electrical components. Survival models arise when there is an event of interest for a group of subjects, machine component, or other item that is</p>
+<ul>
+<li>certain to occur after some amount of time,</li>
+<li>but only measured for a fixed period of time, during which the event may not have occurred for all subjects.</li>
+</ul>
+<p>For example, one might wish to estimate the the distribution of time to failure for solid state drives in a data center, but only measure drives for a two year period, after which some number will have failed and some will still be in service.</p>
+<p>Survival models are often used comparatively, such as comparing time to death of patients diagnosed with stage one liver cancer under a new treatment and a standard treatment (pure controls are not allowed when there is an effective existing treatment for a serious condition). During a two year trial, some patients will die and others will survive.</p>
+<p>Survival models may involve covariates, such as the factory at which a component is manufactured, the day on which it is manufactured, and the amount of usage it gets. A clinical trial might be adjusted for the sex and age of a cancer patient or the hospital at which treatment is received.</p>
+<p>Survival models come in two main flavors, parametric and semi-parametric. In a parametric model, the survival time of a subject is modeled explicitly using a parametric probability distribution. There is a great deal of flexibility in how the parametric probability distribution is constructed. The sections below consider exponential and Weibull distributed survival times.</p>
+<p>Rather than explicitly modeling a parametric survival probability, semi-parametric survival models instead model the relative effect on survival of covariates. The final sections of this chapter consider the proportional hazards survival model.</p>
+<section id="exponential-survival-model" class="level2">
+<h2 class="anchored" data-anchor-id="exponential-survival-model">Exponential survival model</h2>
+<p>The exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if <span class="math inline">\(T \sim \textrm{exponential}(\lambda)\)</span> for some rate <span class="math inline">\(\lambda &gt; 0,\)</span> then <span class="math display">\[\begin{equation*}
+\Pr[T &gt; t] = \Pr[T &gt; t + t' \mid T &gt; t'].
+\end{equation*}\]</span> If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.</p>
+<p>The exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable <span class="math inline">\(T\)</span> is <span class="math display">\[\begin{equation*}
+\mathbb{E}[T \mid \lambda] = \frac{1}{\lambda}.
+\end{equation*}\]</span> The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) <span class="math inline">\(\beta = 1 / \lambda\)</span>.</p>
+<p>The data for a survival model consists of two components. First, there is a vector <span class="math inline">\(t \in (0, \infty)^N\)</span> of <span class="math inline">\(N\)</span> observed failure times. Second, there is a censoring time <span class="math inline">\(t^{\textrm{cens}}\)</span> such that failure times greater than <span class="math inline">\(t^{\textrm{cens}}\)</span> are not observed. The censoring time assumption imposes a constraint which requires <span class="math inline">\(t_n &lt; t^{\textrm{cens}}\)</span> for all <span class="math inline">\(n \in 1{:}N.\)</span> For the censored subjects, the only thing required in the model is their total count, <span class="math inline">\(N^\textrm{cens}\)</span> (their covariates are also required for models with covariates).</p>
+<p>The model for the observed failure times is exponential, so that for <span class="math inline">\(n \in 1{:}N,\)</span> <span class="math display">\[\begin{equation*}
+t_n \sim \textrm{exponential}(\lambda).
+\end{equation*}\]</span></p>
+<p>The model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of <span class="math display">\[\begin{equation*}
+\Pr[T &gt; t^{\textrm{cens}}] = 1 - F_T(t^{\textrm{cens}}),
+\end{equation*}\]</span> where <span class="math inline">\(F_T\)</span> is the cumulative distribution function (cdf) of survival time <span class="math inline">\(T\)</span> (<span class="math inline">\(F_X(x) = \Pr[X \leq x]\)</span> is standard notation for the cdf of a random variable <span class="math inline">\(X\)</span>). The function <span class="math inline">\(1 - F_T(t)\)</span> is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood <span class="math display">\[\begin{eqnarray*}
+p(t, t^{\textrm{cens}}, N^{\textrm{cens}} \mid \lambda)
+&amp; = &amp;
+\prod_{n=1}^N \textrm{exponential}(t_n \mid \lambda)
+\cdot
+\prod_{n=1}^{N^{\textrm{cens}}}
+\textrm{exponentialCCDF}(t^{\textrm{cens}} \mid \lambda)
+\\
+&amp; = &amp;
+\prod_{n=1}^N \textrm{exponential}(t_n \mid \lambda)
+\cdot
+\textrm{exponentialCCDF}(t^{\textrm{cens}} \mid \lambda)^{N^{\textrm{cens}}}.
+\end{eqnarray*}\]</span></p>
+<p>On the log scale, that’s <span class="math display">\[\begin{eqnarray*}
+\log p(t, t^{\textrm{cens}}, N^{\textrm{cens}} \mid \lambda)
+&amp; = &amp;
+\sum_{n=1}^N \log \textrm{exponential}(t_n \mid \lambda)
+\\
+&amp; &amp; { } + N^{\textrm{cens}} \cdot \log \textrm{exponentialCCDF}(t^{\textrm{cens}} \mid \lambda).
+\end{eqnarray*}\]</span></p>
+<p>The model can be completed with a standard lognormal prior on <span class="math inline">\(\lambda,\)</span> <span class="math display">\[\begin{equation*}
+\lambda \sim \textrm{lognormal}(0, 1),
+\end{equation*}\]</span> which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed <span class="math inline">\(\textrm{lognormal}(0, 1)\)</span>. In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.</p>
+<section id="stan-program" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-program">Stan program</h3>
+<p>The data for a simple survival analysis without covariates can be coded as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] t;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_cens;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; t_cens;</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In this program, <code>N</code> is the number of uncensored observations and <code>t</code> contains the times of the uncensored observations. There are a further <code>N_cens</code> items that are right censored at time <code>t_cens</code>. Right censoring means that if the time to failure is greater than</p>
+<p><code>t_cens</code>, it is only observed that the part survived until time <code>t_cens</code>. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.</p>
+<p>There is a single rate parameter, the inverse of which is the expected time to failure.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  t ~ exponential(lambda);</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens * exponential_lccdf(t_cens | lambda);</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  lambda ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The likelihood for rate <code>lambda</code> is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector <code>t</code> as a having an exponential distribution with rate <code>lambda</code>. This data model could have been written as</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  t[n] ~ exponential(lambda);</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate <code>lambda</code>. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N)</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> exponential_lccdf(t_cens | lambda);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>to define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.</p>
+</section>
+</section>
+<section id="weibull-survival-model" class="level2">
+<h2 class="anchored" data-anchor-id="weibull-survival-model">Weibull survival model</h2>
+<p>The Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.</p>
+<p>The Weibull distribution is parameterized by a shape <span class="math inline">\(\alpha &gt; 0\)</span> and scale <span class="math inline">\(\sigma &gt; 0.\)</span> For an outcome <span class="math inline">\(t \geq 0\)</span>, the Weibull distribution’s probability density function is <span class="math display">\[\begin{equation*}
+\textrm{Weibull}(t \mid \alpha, \sigma)
+= \frac{\alpha}{\sigma}
+  \cdot \left( \frac{t}{\sigma} \right)^{\alpha - 1}
+  \cdot \exp\left(-\left(\frac{t}{\sigma}\right)^{\alpha}\right).
+\end{equation*}\]</span> In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter <span class="math inline">\(\beta &gt; 0\)</span> with probability density function <span class="math display">\[\begin{equation*}
+\textrm{exponential}(t \mid \beta) =
+\beta
+\cdot
+\exp(-\beta \cdot t).
+\end{equation*}\]</span> When <span class="math inline">\(\alpha = 1,\)</span> the Weibull distribution reduces to an exponential distribution, <span class="math display">\[\begin{equation*}
+\textrm{Weibull}(t \mid 1, \sigma)
+=
+\textrm{exponential}\!\left(t \,\bigg|\, \frac{1}{\sigma}\right).
+\end{equation*}\]</span> In other words, the Weibull is a continuous expansion of the exponential distribution.</p>
+<p>If <span class="math inline">\(T \sim \textrm{Weibull}(\alpha, \sigma),\)</span> then the expected survival time is <span class="math display">\[\begin{equation*}
+\mathbb{E}[T] = \sigma \cdot \Gamma\!\left(1 + \frac{1}{\alpha}\right),
+\end{equation*}\]</span> where the <span class="math inline">\(\Gamma\)</span> function is the continuous completion of the factorial function (i.e., <span class="math inline">\(\Gamma(1 + n) = n!\ \)</span> for <span class="math inline">\(n \in
+\mathbb{N}\)</span>). As <span class="math inline">\(\alpha \rightarrow 0\)</span> for a fixed <span class="math inline">\(\sigma\)</span> or as <span class="math inline">\(\sigma \rightarrow \infty\)</span> for a fixed <span class="math inline">\(\alpha\)</span>, the expected survival time goes to infinity.</p>
+<p>There are three regimes of the Weibull distribution.</p>
+<ul>
+<li><p><span class="math inline">\(\alpha &lt; 1.\)</span> A subject is more likely to fail early. When <span class="math inline">\(\alpha
+&lt; 1,\)</span> the Weibull density approaches infinity as <span class="math inline">\(t \rightarrow 0.\)</span></p></li>
+<li><p><span class="math inline">\(\alpha = 1.\)</span> The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When <span class="math inline">\(\alpha = 1,\)</span> the Weibull distribution approaches <span class="math inline">\(\sigma\)</span> as <span class="math inline">\(t
+\rightarrow 0.\)</span></p></li>
+<li><p><span class="math inline">\(\alpha &gt; 1.\)</span> Subjects are less likely to fail early. When <span class="math inline">\(\alpha &lt; 1,\)</span> the Weibull density approaches zero as <span class="math inline">\(t \rightarrow 0.\)</span></p></li>
+</ul>
+<p>With <span class="math inline">\(\alpha \leq 1,\)</span> the mode is zero (<span class="math inline">\(t = 0\)</span>), whereas with <span class="math inline">\(\alpha &gt; 1,\)</span> the mode is nonzero (<span class="math inline">\(t &gt; 0\)</span>).</p>
+<section id="stan-program-1" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-program-1">Stan program</h3>
+<p>With Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; beta;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  t ~ exponential(beta);</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens * exponential_lccdf(t_cens | beta);</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  beta ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (<code>alpha</code>) and scale (<code>sigma</code>) parameters.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>  t ~ weibull(alpha, sigma);</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens * weibull_lccdf(t_cens | alpha, sigma);</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  alpha ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  sigma ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As usual, if more is known about expected survival times, <code>alpha</code> and <code>sigma</code> should be given more informative priors.</p>
+</section>
+</section>
+<section id="survival-with-covariates" class="level2">
+<h2 class="anchored" data-anchor-id="survival-with-covariates">Survival with covariates</h2>
+<p>Suppose that for each of <span class="math inline">\(n \in 1{:}N\)</span> items observed, both censored and uncensored, there is a covariate (row) vector <span class="math inline">\(x_n \in
+\mathbb{R}^K.\)</span> For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.</p>
+<p>Survival with covariates replaces what is essentially a simple regression with only an intercept <span class="math inline">\(\lambda\)</span> with a generalized linear model with a log link, where the rate for item <span class="math inline">\(n\)</span> is <span class="math display">\[\begin{equation*}
+\lambda_n = \exp(x_n \cdot \beta),
+\end{equation*}\]</span> where <span class="math inline">\(\beta \in \mathbb{R}^K\)</span> is a <span class="math inline">\(K\)</span>-vector of regression coefficients. Thus <span class="math display">\[\begin{equation*}
+t_n \sim \textrm{exponential}(\lambda_n).
+\end{equation*}\]</span> The censored items have probability <span class="math display">\[\begin{equation*}
+\Pr[n\textrm{-th censored}] =
+\textrm{exponentialCCDF}(t^{\textrm{cens}} \mid x^{\textrm{cens}}_n
+\cdot \beta).
+\end{equation*}\]</span></p>
+<p>The covariates form an <span class="math inline">\(N \times K\)</span> data matrix, <span class="math inline">\(x \in
+\mathbb{R}^{N \times K}\)</span>. An intercept can be introduced by adding a column of 1 values to <span class="math inline">\(x\)</span>.</p>
+<p>A Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (<code>t</code> and <code>t_cens</code>) corresponds to the ordering of items in the covariate matrices (<code>x</code> and <code>x_cens</code>).</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] t;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_cens;</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; t_cens;</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N_cens, K] x_cens;</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] gamma;</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  gamma ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a>  t ~ exponential(exp(x * gamma));</span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> exponential_lccdf(t_cens | exp(x_cens * gamma));</span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Both the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.</p>
+</section>
+<section id="hazard-and-survival-functions" class="level2">
+<h2 class="anchored" data-anchor-id="hazard-and-survival-functions">Hazard and survival functions</h2>
+<p>Suppose <span class="math inline">\(T\)</span> is a random variable representing a survival time, with a smooth cumulative distribution function <span class="math display">\[\begin{equation*}
+F_T(t) = \Pr[T \leq t],
+\end{equation*}\]</span> so that its probability density function is <span class="math display">\[\begin{equation*}
+p_T(t) = \frac{\textrm{d}}{\textrm{d}t} F_T(t).
+\end{equation*}\]</span></p>
+<p>The <em>survival function</em> <span class="math inline">\(S(t)\)</span> is the probability of surviving until at least time <span class="math inline">\(t\)</span>, which is just the complementary cumulative distribution function (ccdf) of the survival random variable <span class="math inline">\(T\)</span>, <span class="math display">\[\begin{equation*}
+S(t) = 1 - F_T(t).
+\end{equation*}\]</span> The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).</p>
+<p>The <em>hazard function</em> <span class="math inline">\(h(t)\)</span> is the instantaneous risk of not surviving past time <span class="math inline">\(t\)</span> assuming survival until time <span class="math inline">\(t\)</span>, which is given by <span class="math display">\[\begin{equation*}
+h(t) = \frac{p_T(t)}{S(t)} = \frac{p_T(t)}{1 - F_T(t)}.
+\end{equation*}\]</span> The <em>cumulative hazard function</em> <span class="math inline">\(H(t)\)</span> is defined to be the accumulated hazard over time, <span class="math display">\[\begin{equation*}
+H(t) = \int_0^t h(u) \, \textrm{d}u.
+\end{equation*}\]</span></p>
+<p>The hazard function and survival function are related through the differential equation <span class="math display">\[\begin{eqnarray*}
+h(t) &amp; = &amp; -\frac{\textrm{d}}{\textrm{d}t} \log S(t).
+\\[4pt]
+&amp; = &amp; -\frac{1}{S(t)} \frac{\textrm{d}}{\textrm{d}t} S(t)
+\\[4pt]
+&amp; = &amp; \frac{1}{S(t)} \frac{\textrm{d}}{\textrm{d}t} -(1 - F_T(t))
+\\[4pt]
+&amp; = &amp; \frac{1}{S(t)} \frac{\textrm{d}}{\textrm{d}t} (F_T(t) - 1)
+\\[4pt]
+&amp; = &amp; \frac{1}{S(t)} \frac{\textrm{d}}{\textrm{d}t} F_T(t)
+\\[4pt]
+&amp; = &amp; \frac{p_T(t)}{S(t)}.
+\end{eqnarray*}\]</span></p>
+<p>If <span class="math inline">\(T \sim \textrm{exponential}(\beta)\)</span> has an exponential distribution, then its hazard function is constant, <span class="math display">\[\begin{eqnarray*}
+h(t \mid \beta)
+&amp; = &amp; \frac{p_T(t \mid \beta)}{S(t \mid \beta)}
+\\[4pt]
+&amp; = &amp; \frac{\textrm{exponential}(t \mid \beta)}{1 - \textrm{exponentialCCDF}(t \mid \beta)}
+\\[4pt]
+&amp; = &amp; \frac{\beta \cdot \exp(-\beta \cdot t)}
+           {1 - (1 - \exp(-\beta \cdot t))}
+\\[4pt]
+&amp; = &amp; \frac{\beta \cdot \exp(-\beta \cdot t)}
+           {\exp(-\beta \cdot t)}
+\\[4pt]
+&amp; = &amp; \beta.
+\end{eqnarray*}\]</span> The exponential distribution is the only distribution of survival times with a constant hazard function.</p>
+<p>If <span class="math inline">\(T \sim \textrm{Weibull}(\alpha, \sigma),\)</span> then its hazard function is <span class="math display">\[\begin{eqnarray*}
+h(t \mid \alpha, \sigma)
+&amp; = &amp; \frac{p_T(t \mid \alpha, \sigma)}{S(t \mid \alpha, \sigma)}
+\\[4pt]
+&amp; = &amp; \frac{\textrm{Weibull}(t \mid \alpha, \sigma)}{1 - \textrm{WeibullCCDF}(t \mid \alpha, \sigma)}
+\\[4pt]
+&amp; = &amp;
+\frac{\frac{\alpha}{\sigma} \cdot \left( \frac{t}{\sigma} \right)^{\alpha - 1}
+      \cdot \exp\left(-\left(\frac{t}{\sigma} \right)^\alpha\right)}
+     {1 - \left(1 -
+       \exp\left(-\left(\frac{t}{\sigma}\right)^\alpha
+     \right)\right)}      
+\\[4pt]
+&amp; = &amp; \frac{\alpha}{\sigma}
+  \cdot
+  \left( \frac{t}{\sigma} \right)^{\alpha - 1}.
+\end{eqnarray*}\]</span></p>
+<p>If <span class="math inline">\(\alpha = 1\)</span> the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when <span class="math inline">\(\alpha = 1\)</span>). When <span class="math inline">\(\alpha &gt; 1,\)</span> the hazard grows as time passes, whereas when <span class="math inline">\(\alpha &lt; 1,\)</span> it decreases as time passes.</p>
+</section>
+<section id="proportional-hazards-model" class="level2">
+<h2 class="anchored" data-anchor-id="proportional-hazards-model">Proportional hazards model</h2>
+<p>The exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. <span class="citation" data-cites="cox:1972">Cox (<a href="#ref-cox:1972" role="doc-biblioref">1972</a>)</span> introduced a semi-parametric survival model specified directly in terms of a hazard function <span class="math inline">\(h(t)\)</span> rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.</p>
+<p>Let <span class="math inline">\(x_n \in \mathbb{R}^K\)</span> be a (row) vector of covariates for subject <span class="math inline">\(n\)</span> so that the full covariate data matrix is <span class="math inline">\(x \in \mathbb{R}^{N \times
+K}\)</span>. In Cox’s model, the hazard function for subject <span class="math inline">\(n\)</span> is defined conditionally in terms of their covariates <span class="math inline">\(x_n\)</span> and the parameter vector <span class="math inline">\(\gamma \in \mathbb{R}^K\)</span> as <span class="math display">\[\begin{equation*}
+h(t \mid x_n, \beta) = h_0(t) \cdot \exp(x_n \cdot \gamma),
+\end{equation*}\]</span> where <span class="math inline">\(h_0(t)\)</span> is a shared baseline hazard function and <span class="math inline">\(x_n \cdot
+\gamma = \sum_{k=1}^K x_{n, k} \cdot \beta_k\)</span> is a row vector-vector product.</p>
+<p>In the semi-parametric, proportional hazards model, the baseline hazard function <span class="math inline">\(h_0(t)\)</span> is not modeled. This is why it is called “semi-parametric.” Only the factor <span class="math inline">\(\exp(x_n \cdot \gamma),\)</span> which determines how individual <span class="math inline">\(n\)</span> varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”</p>
+<p>Cox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function <span class="math inline">\(h_0(t)\)</span> is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a>.</p>
+<section id="partial-likelihood-function" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="partial-likelihood-function">Partial likelihood function</h3>
+<p>Cox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function <span class="math inline">\(h_0(t)\)</span> is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.</p>
+<p>The hazard function <span class="math inline">\(h(t \mid x_n, \beta) = h_0(t) \cdot \exp(x_n
+\cdot \beta)\)</span> for subject <span class="math inline">\(n\)</span> represents the instantaneous probability that subject <span class="math inline">\(n\)</span> fails at time <span class="math inline">\(t\)</span> given that it has survived until time <span class="math inline">\(t.\)</span> The probability that subject <span class="math inline">\(n\)</span> is the first to fail among <span class="math inline">\(N\)</span> subjects is thus proportional to subject <span class="math inline">\(n\)</span>’s hazard function, <span class="math display">\[\begin{equation*}
+\Pr[n \textrm{ first to fail at time } t]
+\propto h(t \mid x_n, \beta).
+\end{equation*}\]</span> Normalizing yields <span class="math display">\[\begin{eqnarray*}
+\Pr[n \textrm{ first to fail at time } t]
+&amp; = &amp;  \frac{h(t \mid x_n, \beta)}
+            {\sum_{n' = 1}^N h(t \mid x_{n'}, \beta)}
+\\[4pt]
+&amp; = &amp;  \frac{h_0(t) \cdot \exp(x_n \cdot \beta)}
+            {\sum_{n' = 1}^N h_0(t) \cdot \exp(x_{n'} \cdot \beta)}
+\\[4pt]
+&amp; = &amp;  \frac{\exp(x_n \cdot \beta)}
+            {\sum_{n' = 1}^N \exp(x_{n'} \cdot \beta)}.
+\end{eqnarray*}\]</span></p>
+<p>Suppose there are <span class="math inline">\(N\)</span> subjects with strictly <em>ordered</em> survival times <span class="math inline">\(t_1 &lt;
+t_2 &lt; \cdots &lt; t_N\)</span> and covariate (row) vectors <span class="math inline">\(x_1, \ldots, x_N\)</span>. Let <span class="math inline">\(t^{\textrm{cens}}\)</span> be the (right) censoring time and let <span class="math inline">\(N^{\textrm{obs}}\)</span> be the largest value of <span class="math inline">\(n\)</span> such that <span class="math inline">\(t_n \leq
+t^{\textrm{cens}}\)</span>. This means <span class="math inline">\(N^{\textrm{obs}}\)</span> is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.</p>
+<p>With failure times sorted in decreasing order, the partial likelihood for each observed subject <span class="math inline">\(n \in 1{:}N^{\textrm{obs}}\)</span> can be expressed as <span class="math display">\[\begin{equation*}
+\Pr[n \textrm{ first to fail among } n, n + 1, \ldots N]
+= \frac{\exp(x_n \cdot \beta)}
+       {\sum_{n' = n}^N \exp(x_{n'} \cdot \beta)}.
+\end{equation*}\]</span> The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.</p>
+<p>The partial likelihood, defined in this form by <span class="citation" data-cites="breslow:1975">Breslow (<a href="#ref-breslow:1975" role="doc-biblioref">1975</a>)</span>, is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). <span class="math display">\[\begin{equation*}
+\Pr[\textrm{observed failures ordered } 1, \ldots, N^{\textrm{obs}} |
+x, \beta]
+= \prod_{n = 1}^{N^{\textrm{obs}}}
+  \frac{\exp(x_n \cdot \beta)}
+       {\sum_{n' = n}^N \exp(x_{n'} \cdot \beta)}.
+\end{equation*}\]</span> On the log scale, <span class="math display">\[\begin{eqnarray*}
+\log \Pr[\textrm{obs.\ fail ordered } 1, \ldots, N^{\textrm{obs}} |
+x, \beta]
+&amp; = &amp;
+\sum_{n = 1}^{N^{\textrm{obs}}}
+  \log \left(
+          \frac{\exp(x_n \cdot \beta)}
+               {\sum_{n' = n}^N \exp(x_{n'} \cdot \beta)}
+       \right)
+\\[4pt]
+&amp; = &amp; x_n \cdot \beta - \log \sum_{n' = n}^N \exp(x_{n'} \cdot \beta)
+\\
+&amp; = &amp; x_n \cdot \beta - \textrm{logSumExp}_{n' = n}^N \ x_{n'} \cdot \beta,
+\end{eqnarray*}\]</span> where <span class="math display">\[\begin{equation*}
+\textrm{logSumExp}_{n = a}^b \ x_n
+= \log \sum_{n = a}^b \exp(x_n)
+\end{equation*}\]</span> is implemented so as to preserve numerical precision.</p>
+<p>This likelihood follows the same approach to ranking as that developed by <span class="citation" data-cites="plackett:1975">Plackett (<a href="#ref-plackett:1975" role="doc-biblioref">1975</a>)</span> for estimating the probability of the order of the first few finishers in a horse race.</p>
+<p>A simple normal prior on the components of <span class="math inline">\(\beta\)</span> completes the model, <span class="math display">\[\begin{equation*}
+\beta \sim \textrm{normal}(0, 2).
+\end{equation*}\]</span> This should be scaled based on knowledge of the predictors.</p>
+</section>
+<section id="stan-program-2" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-program-2">Stan program</h3>
+<p>To simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;          <span class="co">// num covariates</span></span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;          <span class="co">// num uncensored obs</span></span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] t;             <span class="co">// event time (non-strict decreasing)</span></span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;          <span class="co">// covariates for uncensored obs</span></span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N_c;                 <span class="co">// num censored obs</span></span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=t[N]&gt; t_c;    <span class="co">// censoring time</span></span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N_c, K] x_c;      <span class="co">// covariates for censored obs</span></span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameters are just the coefficients.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;          <span class="co">// slopes (no intercept)</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  ...</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] log_theta = x * beta;</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_c] log_theta_c = x_c * beta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_denom = log_sum_exp(log_theta_c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Then, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>    log_denom = log_sum_exp(log_denom, log_theta[n]);</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> log_theta[n] - log_denom;   <span class="co">// log likelihood</span></span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">2</span>:N) {</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (!(t[n] &lt; t[n - <span class="dv">1</span>])) {</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>      <span class="kw">reject</span>(<span class="st">"times must be strictly decreasing, but found"</span></span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>             <span class="st">"!(t["</span>, n, <span class="st">"] &lt; t[, "</span>, (n - <span class="dv">1</span>), <span class="st">"])"</span>);</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>    }   </span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="stan-model-for-tied-survival-times" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="stan-model-for-tied-survival-times">Stan model for tied survival times</h3>
+<p>Technically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all <span class="math inline">\(N!\)</span> permutations for a set of <span class="math inline">\(N\)</span> subjects with tied survival times is not tractable. As an alternative, <span class="citation" data-cites="efron:1977">Efron (<a href="#ref-efron:1977" role="doc-biblioref">1977</a>)</span> introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.</p>
+<p>In the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> num_unique_starts(<span class="dt">vector</span> t) {</span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (size(t) == <span class="dv">0</span>) <span class="cf">return</span> <span class="dv">0</span>;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> us = <span class="dv">1</span>;</span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">2</span>:size(t)) {</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (t[n] != t[n - <span class="dv">1</span>]) us += <span class="dv">1</span>;</span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> us;</span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This is then used to compute the value <code>J</code> to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[] <span class="dt">int</span> unique_starts(<span class="dt">vector</span> t, <span class="dt">int</span> J) {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[J + <span class="dv">1</span>] <span class="dt">int</span> starts;</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (J == <span class="dv">0</span>) <span class="cf">return</span> starts;</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>    starts[<span class="dv">1</span>] = <span class="dv">1</span>;</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> pos = <span class="dv">2</span>;</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">2</span>:size(t)) {</span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (t[n] != t[n - <span class="dv">1</span>]) {</span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>    starts[pos] = n;</span>
+<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a>    pos += <span class="dv">1</span>;</span>
+<span id="cb17-10"><a href="#cb17-10" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb17-11"><a href="#cb17-11" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb17-12"><a href="#cb17-12" aria-hidden="true" tabindex="-1"></a>    starts[J + <span class="dv">1</span>] = size(t) + <span class="dv">1</span>;</span>
+<span id="cb17-13"><a href="#cb17-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> starts;</span>
+<span id="cb17-14"><a href="#cb17-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-15"><a href="#cb17-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; J = num_unique_starts(t);</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J + <span class="dv">1</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; starts = unique_starts(t, J);</span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For each unique survival time <code>j</code> in <code>1:J</code>, the subjects indexed from <code>starts[j]</code> to <code>starts[j + 1] - 1</code> (inclusive) share the same survival time. The number of elements with survival time <code>j</code> is thus <code>(starts[j + 1] - 1) - starts[j] + 1</code>, or just <code>starts[j + 1] - starts[j]</code>.</p>
+<p>The parameters and prior are also the same—just a vector <code>beta</code> of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables <code>J</code> and <code>starts</code>.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] log_theta = x * beta;</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_c] log_theta_c = x_c * beta;</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_denom_lhs = log_sum_exp(log_theta_c);</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> start = starts[j];</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> end = starts[j + <span class="dv">1</span>] - <span class="dv">1</span>;</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> len = end - start + <span class="dv">1</span>;</span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> log_len = log(len);</span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> numerator = sum(log_theta[start:end]);</span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a>    log_denom_lhs = log_sum_exp(log_denom_lhs,</span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>                                log_sum_exp(log_theta[start:end]));</span>
+<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[len] diff;</span>
+<span id="cb19-13"><a href="#cb19-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (ell <span class="cf">in</span> <span class="dv">1</span>:len) {</span>
+<span id="cb19-14"><a href="#cb19-14" aria-hidden="true" tabindex="-1"></a>      diff[ell] = log_diff_exp(log_denom_lhs,</span>
+<span id="cb19-15"><a href="#cb19-15" aria-hidden="true" tabindex="-1"></a>                               log(ell - <span class="dv">1</span>) - log_len</span>
+<span id="cb19-16"><a href="#cb19-16" aria-hidden="true" tabindex="-1"></a>                               + log_sum_exp(log_theta[start:end]));</span>
+<span id="cb19-17"><a href="#cb19-17" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb19-18"><a href="#cb19-18" aria-hidden="true" tabindex="-1"></a>    <span class="kw">target +=</span> numerator - sum(diff);</span>
+<span id="cb19-19"><a href="#cb19-19" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The special function <code>log_diff_exp</code> is defined as</p>
+<p><span class="math display">\[\begin{equation*}
+\textrm{logDiffExp}(u, v) = \log(\exp(u) - \exp(v)).
+\end{equation*}\]</span></p>
+<p>Because of how <code>J</code> and <code>starts</code> are constructed, the length <code>len</code> will always be strictly positive so that the log is well defined.</p>
+
+
+
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-breslow:1975" class="csl-entry" role="listitem">
+Breslow, N. E. 1975. <span>“Analysis of Survival Data Under the Proportional Hazards Model.”</span> <em>International Statisticas Review</em> 43 (1): 45–58.
+</div>
+<div id="ref-cox:1972" class="csl-entry" role="listitem">
+Cox, David R. 1972. <span>“Regression Models and Life-Tables.”</span> <em>Journal of the Royal Statistical Society: Series B (Methodological)</em> 34 (2): 187–202.
+</div>
+<div id="ref-efron:1977" class="csl-entry" role="listitem">
+Efron, Bradley. 1977. <span>“The Efficiency of <span>C</span>ox’s Likelihood Function for Censored Data.”</span> <em>Journal of the American Statistical Association</em> 72 (359): 557–65.
+</div>
+<div id="ref-plackett:1975" class="csl-entry" role="listitem">
+Plackett, Robin L. 1975. <span>“The Analysis of Permutations.”</span> <em>Journal of the Royal Statistical Society Series C: Applied Statistics</em> 24 (2): 193–202.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Cox mentioned in his seminal paper that modeling the baseline hazard function would improve statistical efficiency, but he did not do it for computational reasons.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/dae.html" class="pagination-link" aria-label="Differential-Algebraic Equations">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Differential-Algebraic Equations</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/wiener_diffusion_model.html" class="pagination-link" aria-label="Wiener diffusion model">
+        <span class="nav-page-text">Wiener diffusion model</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/survival.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/time-series.html b/docs/2_39/stan-users-guide/time-series.html
new file mode 100644
index 000000000..0b0120064
--- /dev/null
+++ b/docs/2_39/stan-users-guide/time-series.html
@@ -0,0 +1,1689 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Time-Series Models</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/missing-data.html" rel="next">
+<link href="../stan-users-guide/regression.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Time-Series Models">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/time-series.html">Time-Series Models</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#time-series.chapter" id="toc-time-series.chapter" class="nav-link active" data-scroll-target="#time-series.chapter">Time-Series Models</a>
+  <ul class="collapse">
+  <li><a href="#autoregressive.section" id="toc-autoregressive.section" class="nav-link" data-scroll-target="#autoregressive.section">Autoregressive models</a>
+  <ul class="collapse">
+  <li><a href="#ar1-models" id="toc-ar1-models" class="nav-link" data-scroll-target="#ar1-models">AR(1) models</a></li>
+  <li><a href="#extensions-to-the-ar1-model" id="toc-extensions-to-the-ar1-model" class="nav-link" data-scroll-target="#extensions-to-the-ar1-model">Extensions to the AR(1) model</a></li>
+  <li><a href="#ar2-models" id="toc-ar2-models" class="nav-link" data-scroll-target="#ar2-models">AR(2) models</a></li>
+  <li><a href="#ark-models" id="toc-ark-models" class="nav-link" data-scroll-target="#ark-models">AR(<span class="math inline">\(K\)</span>) models</a></li>
+  <li><a href="#arch1-models" id="toc-arch1-models" class="nav-link" data-scroll-target="#arch1-models">ARCH(1) models</a></li>
+  </ul></li>
+  <li><a href="#modeling-temporal-heteroscedasticity" id="toc-modeling-temporal-heteroscedasticity" class="nav-link" data-scroll-target="#modeling-temporal-heteroscedasticity">Modeling temporal heteroscedasticity</a>
+  <ul class="collapse">
+  <li><a href="#garch11-models" id="toc-garch11-models" class="nav-link" data-scroll-target="#garch11-models">GARCH(1,1) models</a></li>
+  </ul></li>
+  <li><a href="#moving-average-models" id="toc-moving-average-models" class="nav-link" data-scroll-target="#moving-average-models">Moving average models</a>
+  <ul class="collapse">
+  <li><a href="#ma2-example" id="toc-ma2-example" class="nav-link" data-scroll-target="#ma2-example">MA(2) example</a></li>
+  <li><a href="#vectorized-maq-model" id="toc-vectorized-maq-model" class="nav-link" data-scroll-target="#vectorized-maq-model">Vectorized MA(Q) model</a></li>
+  </ul></li>
+  <li><a href="#autoregressive-moving-average-models" id="toc-autoregressive-moving-average-models" class="nav-link" data-scroll-target="#autoregressive-moving-average-models">Autoregressive moving average models</a>
+  <ul class="collapse">
+  <li><a href="#identifiability-and-stationarity" id="toc-identifiability-and-stationarity" class="nav-link" data-scroll-target="#identifiability-and-stationarity">Identifiability and stationarity</a></li>
+  </ul></li>
+  <li><a href="#stochastic-volatility-models" id="toc-stochastic-volatility-models" class="nav-link" data-scroll-target="#stochastic-volatility-models">Stochastic volatility models</a></li>
+  <li><a href="#hmms.section" id="toc-hmms.section" class="nav-link" data-scroll-target="#hmms.section">Hidden Markov models</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/time-series.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/time-series.html">Time-Series Models</a></li></ol></nav></header>
+
+
+
+
+<section id="time-series.chapter" class="level1">
+<h1>Time-Series Models</h1>
+<p>Times series data come arranged in temporal order. This chapter presents two kinds of time series models, regression-like models such as autoregressive and moving average models, and hidden Markov models.</p>
+<p>The <a href="../stan-users-guide/gaussian-processes.html">Gaussian processes chapter</a> presents Gaussian processes, which may also be used for time-series (and spatial) data.</p>
+<section id="autoregressive.section" class="level2">
+<h2 class="anchored" data-anchor-id="autoregressive.section">Autoregressive models</h2>
+<p>A first-order autoregressive model (AR(1)) with normal noise takes each point <span class="math inline">\(y_n\)</span> in a sequence <span class="math inline">\(y\)</span> to be generated according to <span class="math display">\[
+y_n \sim \textsf{normal}(\alpha + \beta y_{n-1}, \sigma).
+\]</span></p>
+<p>That is, the expected value of <span class="math inline">\(y_n\)</span> is <span class="math inline">\(\alpha + \beta y_{n-1}\)</span>, with noise scaled as <span class="math inline">\(\sigma\)</span>.</p>
+<section id="ar1-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ar1-models">AR(1) models</h3>
+<p>With improper flat priors on the regression coefficients <span class="math inline">\(\alpha\)</span> and <span class="math inline">\(\beta\)</span> and on the positively-constrained noise scale (<span class="math inline">\(\sigma\)</span>), the Stan program for the AR(1) model is as follows.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">2</span>:N) {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(alpha + beta * y[n<span class="dv">-1</span>], sigma);</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The first observed data point, <code>y[1]</code>, is not modeled here because there is nothing to condition on; instead, it acts to condition <code>y[2]</code>. This model also uses an improper prior for <code>sigma</code>, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in <code>y</code> over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of <code>y</code> is available.</p>
+<section id="slicing-for-efficiency" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="slicing-for-efficiency">Slicing for efficiency</h4>
+<p>Although perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  y[<span class="dv">2</span>:N] ~ normal(alpha + beta * y[<span class="dv">1</span>:(N - <span class="dv">1</span>)], sigma);</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The left-hand side slicing operation pulls out the last <span class="math inline">\(N-1\)</span> elements and the right-hand side version pulls out the first <span class="math inline">\(N-1\)</span>.</p>
+</section>
+</section>
+<section id="extensions-to-the-ar1-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="extensions-to-the-ar1-model">Extensions to the AR(1) model</h3>
+<p>Proper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-<span class="math inline">\(t\)</span> distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.</p>
+<p>To enforce the estimation of a stationary AR(1) process, the slope coefficient <code>beta</code> may be constrained with bounds as follows.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; beta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of <code>beta</code> near zero.</p>
+</section>
+<section id="ar2-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ar2-models">AR(2) models</h3>
+<p>Extending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient <code>gamma</code> and the following model statement.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">3</span>:N) {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(alpha + beta*y[n<span class="dv">-1</span>] + gamma*y[n<span class="dv">-2</span>], sigma);</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="ark-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ark-models">AR(<span class="math inline">\(K\)</span>) models</h3>
+<p>A general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; K;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">real</span> beta;</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma;</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> (K+<span class="dv">1</span>):N) {</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> mu = alpha;</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:K) {</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>      mu += beta[k] * y[n-k];</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(mu, sigma);</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="arch1-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="arch1-models">ARCH(1) models</h3>
+<p>Econometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model <span class="citation" data-cites="Engle:1982">(<a href="#ref-Engle:1982" role="doc-biblioref">Engle 1982</a>)</span>. Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.</p>
+<p>The ARCH(1) model is typically presented as the following sequence of equations, where <span class="math inline">\(r_t\)</span> is the observed return at time point <span class="math inline">\(t\)</span> and <span class="math inline">\(\mu\)</span>, <span class="math inline">\(\alpha_0\)</span>, and <span class="math inline">\(\alpha_1\)</span> are unknown regression coefficient parameters.</p>
+<p><span class="math display">\[\begin{align*}
+r_t        &amp;=  \mu + a_t \\
+a_t        &amp;=  \sigma_t \epsilon_t \\
+\epsilon_t &amp;\sim \textsf{normal}(0,1) \\
+\sigma^2_t &amp;=  \alpha_0 + \alpha_1 a_{t-1}^2
+\end{align*}\]</span></p>
+<p>In order to ensure the noise terms <span class="math inline">\(\sigma^2_t\)</span> are positive, the scale coefficients are constrained to be positive, <span class="math inline">\(\alpha_0, \alpha_1
+&gt; 0\)</span>. To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., <span class="math inline">\(\alpha_1 &lt; 1\)</span>.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>The ARCH(1) model may be coded directly in Stan as follows.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; T;                <span class="co">// number of time points</span></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span> r;               <span class="co">// return at time t</span></span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;                       <span class="co">// average return</span></span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha0;          <span class="co">// noise intercept</span></span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; alpha1; <span class="co">// noise slope</span></span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:T) {</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>    r[t] ~ normal(mu, sqrt(alpha0 + alpha1</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>                                    * pow(r[t - <span class="dv">1</span>] - mu,<span class="dv">2</span>)));</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The loop in the model is defined so that the return at time <span class="math inline">\(t=1\)</span> is not modeled; the model in the next section shows how to model the return at <span class="math inline">\(t=1\)</span>. The model can be vectorized to be more efficient; the model in the next section provides an example.</p>
+</section>
+</section>
+<section id="modeling-temporal-heteroscedasticity" class="level2">
+<h2 class="anchored" data-anchor-id="modeling-temporal-heteroscedasticity">Modeling temporal heteroscedasticity</h2>
+<p>A set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.</p>
+<section id="garch11-models" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="garch11-models">GARCH(1,1) models</h3>
+<p>The basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time <span class="math inline">\(t-1\)</span> as a predictor of volatility at time <span class="math inline">\(t\)</span>, defining <span class="math display">\[
+\sigma^2_t = \alpha_0 + \alpha_1 a^2_{t-1} + \beta_1 \sigma^2_{t-1}.
+\]</span></p>
+<p>To ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy <span class="math inline">\(\alpha_0, \alpha_1,
+\beta_1 &gt; 0\)</span> and the slopes <span class="math inline">\(\alpha_1 + \beta_1 &lt; 1\)</span>.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; T;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span> r;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma1;</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha0;</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; alpha1;</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=(<span class="dv">1</span>-alpha1)&gt; beta1;</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  sigma[<span class="dv">1</span>] = sigma1;</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:T) {</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a>    sigma[t] = sqrt(alpha0</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>                     + alpha1 * pow(r[t - <span class="dv">1</span>] - mu, <span class="dv">2</span>)</span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a>                     + beta1 * pow(sigma[t - <span class="dv">1</span>], <span class="dv">2</span>));</span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a>  r ~ normal(mu, sigma);</span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value <code>sigma1</code> for the scale of the noise at <span class="math inline">\(t = 1\)</span>.</p>
+<p>The constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on <code>beta1</code> depends on the value of <code>alpha1</code>.</p>
+<p>A transformed parameter array of non-negative values <code>sigma</code> is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept <code>alpha0</code>, a slope <code>alpha1</code> for the squared difference in return from the mean at the previous time, and a slope <code>beta1</code> for the previous noise scale squared. Finally, the whole regression is inside the <code>sqrt</code> function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.</p>
+<p>With the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because <code>r</code> and <code>sigma</code> are of length <code>T</code>, all of the data are modeled directly.</p>
+</section>
+</section>
+<section id="moving-average-models" class="level2">
+<h2 class="anchored" data-anchor-id="moving-average-models">Moving average models</h2>
+<p>A moving average model uses previous errors as predictors for future outcomes. For a moving average model of order <span class="math inline">\(Q\)</span>, <span class="math inline">\(\mbox{MA}(Q)\)</span>, there is an overall mean parameter <span class="math inline">\(\mu\)</span> and regression coefficients <span class="math inline">\(\theta_q\)</span> for previous error terms. With <span class="math inline">\(\epsilon_t\)</span> being the noise at time <span class="math inline">\(t\)</span>, the model for outcome <span class="math inline">\(y_t\)</span> is defined by <span class="math display">\[
+y_t = \mu + \theta_1 \epsilon_{t-1} + \dotsb + \theta_Q \epsilon_{t-Q}
++ \epsilon_t,
+\]</span> with the noise term <span class="math inline">\(\epsilon_t\)</span> for outcome <span class="math inline">\(y_t\)</span> modeled as normal, <span class="math display">\[
+\epsilon_t \sim \textsf{normal}(0,\sigma).
+\]</span> In a proper Bayesian model, the parameters <span class="math inline">\(\mu\)</span>, <span class="math inline">\(\theta\)</span>, and <span class="math inline">\(\sigma\)</span> must all be given priors.</p>
+<section id="ma2-example" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="ma2-example">MA(2) example</h3>
+<p>An <span class="math inline">\(\mbox{MA}(2)\)</span> model can be coded in Stan as follows.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">3</span>&gt; T;          <span class="co">// number of observations</span></span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] y;             <span class="co">// observation at time T</span></span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;                 <span class="co">// mean</span></span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;     <span class="co">// error scale</span></span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">2</span>] theta;         <span class="co">// lag coefficients</span></span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] epsilon;       <span class="co">// error terms</span></span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>  epsilon[<span class="dv">1</span>] = y[<span class="dv">1</span>] - mu;</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  epsilon[<span class="dv">2</span>] = y[<span class="dv">2</span>] - mu - theta[<span class="dv">1</span>] * epsilon[<span class="dv">1</span>];</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">3</span>:T) {</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>    epsilon[t] = ( y[t] - mu</span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a>                    - theta[<span class="dv">1</span>] * epsilon[t - <span class="dv">1</span>]</span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a>                    - theta[<span class="dv">2</span>] * epsilon[t - <span class="dv">2</span>] );</span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-19"><a href="#cb8-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-20"><a href="#cb8-20" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-21"><a href="#cb8-21" aria-hidden="true" tabindex="-1"></a>  mu ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb8-22"><a href="#cb8-22" aria-hidden="true" tabindex="-1"></a>  theta ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb8-23"><a href="#cb8-23" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb8-24"><a href="#cb8-24" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">3</span>:T) {</span>
+<span id="cb8-25"><a href="#cb8-25" aria-hidden="true" tabindex="-1"></a>    y[t] ~ normal(mu</span>
+<span id="cb8-26"><a href="#cb8-26" aria-hidden="true" tabindex="-1"></a>                  + theta[<span class="dv">1</span>] * epsilon[t - <span class="dv">1</span>]</span>
+<span id="cb8-27"><a href="#cb8-27" aria-hidden="true" tabindex="-1"></a>                  + theta[<span class="dv">2</span>] * epsilon[t - <span class="dv">2</span>],</span>
+<span id="cb8-28"><a href="#cb8-28" aria-hidden="true" tabindex="-1"></a>                  sigma);</span>
+<span id="cb8-29"><a href="#cb8-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-30"><a href="#cb8-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The error terms <span class="math inline">\(\epsilon_t\)</span> are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to <span class="math inline">\(y_n\)</span> for <span class="math inline">\(n &gt; Q\)</span>. In this example, the parameters are all given Cauchy (half-Cauchy for <span class="math inline">\(\sigma\)</span>) priors, although other priors can be used just as easily.</p>
+<p>This model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the <span class="math inline">\(\epsilon_t\)</span> could also be sped up by using a dot product instead of a loop.</p>
+</section>
+<section id="vectorized-maq-model" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="vectorized-maq-model">Vectorized MA(Q) model</h3>
+<p>A general <span class="math inline">\(\mbox{MA}(Q)\)</span> model with a vectorized distribution statement may be defined as follows.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; Q;       <span class="co">// num previous noise terms</span></span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">3</span>&gt; T;       <span class="co">// num observations</span></span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] y;          <span class="co">// observation at time t</span></span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;              <span class="co">// mean</span></span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;  <span class="co">// error scale</span></span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[Q] theta;      <span class="co">// error coeff, lag -t</span></span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] epsilon;    <span class="co">// error term at time t</span></span>
+<span id="cb9-13"><a href="#cb9-13" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb9-14"><a href="#cb9-14" aria-hidden="true" tabindex="-1"></a>    epsilon[t] = y[t] - mu;</span>
+<span id="cb9-15"><a href="#cb9-15" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (q <span class="cf">in</span> <span class="dv">1</span>:min(t - <span class="dv">1</span>, Q)) {</span>
+<span id="cb9-16"><a href="#cb9-16" aria-hidden="true" tabindex="-1"></a>      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];</span>
+<span id="cb9-17"><a href="#cb9-17" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb9-18"><a href="#cb9-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-19"><a href="#cb9-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-20"><a href="#cb9-20" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-21"><a href="#cb9-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] eta;</span>
+<span id="cb9-22"><a href="#cb9-22" aria-hidden="true" tabindex="-1"></a>  mu ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb9-23"><a href="#cb9-23" aria-hidden="true" tabindex="-1"></a>  theta ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb9-24"><a href="#cb9-24" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb9-25"><a href="#cb9-25" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb9-26"><a href="#cb9-26" aria-hidden="true" tabindex="-1"></a>    eta[t] = mu;</span>
+<span id="cb9-27"><a href="#cb9-27" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (q <span class="cf">in</span> <span class="dv">1</span>:min(t - <span class="dv">1</span>, Q)) {</span>
+<span id="cb9-28"><a href="#cb9-28" aria-hidden="true" tabindex="-1"></a>      eta[t] = eta[t] + theta[q] * epsilon[t - q];</span>
+<span id="cb9-29"><a href="#cb9-29" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb9-30"><a href="#cb9-30" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb9-31"><a href="#cb9-31" aria-hidden="true" tabindex="-1"></a>  y ~ normal(eta, sigma);</span>
+<span id="cb9-32"><a href="#cb9-32" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).</p>
+</section>
+</section>
+<section id="autoregressive-moving-average-models" class="level2">
+<h2 class="anchored" data-anchor-id="autoregressive-moving-average-models">Autoregressive moving average models</h2>
+<p>Autoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; T;            <span class="co">// num observations</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[T] <span class="dt">real</span> y;                 <span class="co">// observed outputs</span></span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;                   <span class="co">// mean coeff</span></span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> phi;                  <span class="co">// autoregression coeff</span></span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> theta;                <span class="co">// moving avg coeff</span></span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;       <span class="co">// noise scale</span></span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-12"><a href="#cb10-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] nu;              <span class="co">// prediction for time t</span></span>
+<span id="cb10-13"><a href="#cb10-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] err;             <span class="co">// error for time t</span></span>
+<span id="cb10-14"><a href="#cb10-14" aria-hidden="true" tabindex="-1"></a>  nu[<span class="dv">1</span>] = mu + phi * mu;     <span class="co">// assume err[0] == 0</span></span>
+<span id="cb10-15"><a href="#cb10-15" aria-hidden="true" tabindex="-1"></a>  err[<span class="dv">1</span>] = y[<span class="dv">1</span>] - nu[<span class="dv">1</span>];</span>
+<span id="cb10-16"><a href="#cb10-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:T) {</span>
+<span id="cb10-17"><a href="#cb10-17" aria-hidden="true" tabindex="-1"></a>    nu[t] = mu + phi * y[t - <span class="dv">1</span>] + theta * err[t - <span class="dv">1</span>];</span>
+<span id="cb10-18"><a href="#cb10-18" aria-hidden="true" tabindex="-1"></a>    err[t] = y[t] - nu[t];</span>
+<span id="cb10-19"><a href="#cb10-19" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb10-20"><a href="#cb10-20" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);        <span class="co">// priors</span></span>
+<span id="cb10-21"><a href="#cb10-21" aria-hidden="true" tabindex="-1"></a>  phi ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb10-22"><a href="#cb10-22" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb10-23"><a href="#cb10-23" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb10-24"><a href="#cb10-24" aria-hidden="true" tabindex="-1"></a>  err ~ normal(<span class="dv">0</span>, sigma);    <span class="co">// error model</span></span>
+<span id="cb10-25"><a href="#cb10-25" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The data are declared in the same way as the other time-series regressions and the parameters are documented in the code.</p>
+<p>In the model block, the local vector <code>nu</code> stores the predictions and <code>err</code> the errors. These are computed similarly to the errors in the moving average models described in the previous section.</p>
+<p>The priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.</p>
+<p>Often in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring <code>err</code> as a transformed parameter, then defining it the same way as in the model above. The vector <code>nu</code> could still be a local variable, only now it will be in the transformed parameter block.</p>
+<p>Wayne Folta suggested encoding the model without local vector variables as follows.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> err;</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  phi ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>  theta ~ normal(<span class="dv">0</span>, <span class="dv">2</span>);</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>  err = y[<span class="dv">1</span>] - (mu + phi * mu);</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>  err ~ normal(<span class="dv">0</span>, sigma);</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:T) {</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>    err = y[t] - (mu + phi * y[t - <span class="dv">1</span>] + theta * err);</span>
+<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a>    err ~ normal(<span class="dv">0</span>, sigma);</span>
+<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This approach to ARMA models illustrates how local variables, such as <code>err</code> in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.</p>
+<p>Both encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just <code>err</code>.</p>
+<section id="identifiability-and-stationarity" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="identifiability-and-stationarity">Identifiability and stationarity</h3>
+<p>MA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (<code>theta</code>, <code>phi</code>) outside the <span class="math inline">\([-1,1]\)</span> interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).</p>
+<p>Further, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; phi;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="stochastic-volatility-models" class="level2">
+<h2 class="anchored" data-anchor-id="stochastic-volatility-models">Stochastic volatility models</h2>
+<p>Stochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time <span class="citation" data-cites="KimShephardChib:1998">(<a href="#ref-KimShephardChib:1998" role="doc-biblioref">Kim, Shephard, and Chib 1998</a>)</span>. The data consist of mean corrected (i.e., centered) returns <span class="math inline">\(y_t\)</span> on an underlying asset at <span class="math inline">\(T\)</span> equally spaced time points. Kim et al.&nbsp;formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter <span class="math inline">\(h_t\)</span> for the log volatility, along with parameters <span class="math inline">\(\mu\)</span> for the mean log volatility, and <span class="math inline">\(\phi\)</span> for the persistence of the volatility term. The variable <span class="math inline">\(\epsilon_t\)</span> represents the white-noise shock (i.e., multiplicative error) on the asset return at time <span class="math inline">\(t\)</span>, whereas <span class="math inline">\(\delta_t\)</span> represents the shock on volatility at time <span class="math inline">\(t\)</span>. <span class="math display">\[\begin{align*}
+y_t        &amp;=    \epsilon_t \exp(h_t / 2) \\
+h_{t+1}    &amp;=    \mu + \phi (h_t - \mu) + \delta_t \sigma \\
+h_1        &amp;\sim \textsf{normal}\left( \mu, \frac{\sigma}{\sqrt{1 - \phi^2}} \right) \\
+\epsilon_t &amp;\sim \textsf{normal}(0,1) \\
+\delta_t   &amp;\sim \textsf{normal}(0,1)
+\end{align*}\]</span></p>
+<p>Rearranging the first line, <span class="math inline">\(\epsilon_t = y_t \exp(-h_t / 2)\)</span>, allowing the distribution for <span class="math inline">\(y_t\)</span> to be written as <span class="math display">\[
+y_t \sim \textsf{normal}(0,\exp(h_t/2)).
+\]</span> The recurrence equation for <span class="math inline">\(h_{t+1}\)</span> may be combined with the scaling of <span class="math inline">\(\delta_t\)</span> to yield the distribution <span class="math display">\[
+h_t \sim \mathsf{normal}(\mu + \phi(h_{t-1} - \mu), \sigma).
+\]</span> This formulation can be directly encoded, as shown in the following Stan model.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; T;   <span class="co">// # time points (equally spaced)</span></span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] y;      <span class="co">// mean corrected return at time t</span></span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;                     <span class="co">// mean log volatility</span></span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; phi; <span class="co">// persistence of volatility</span></span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;         <span class="co">// white noise shock scale</span></span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] h;                 <span class="co">// log volatility at time t</span></span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb14-12"><a href="#cb14-12" aria-hidden="true" tabindex="-1"></a>  phi ~ uniform(-<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb14-13"><a href="#cb14-13" aria-hidden="true" tabindex="-1"></a>  sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb14-14"><a href="#cb14-14" aria-hidden="true" tabindex="-1"></a>  mu ~ cauchy(<span class="dv">0</span>, <span class="dv">10</span>);</span>
+<span id="cb14-15"><a href="#cb14-15" aria-hidden="true" tabindex="-1"></a>  h[<span class="dv">1</span>] ~ normal(mu, sigma / sqrt(<span class="dv">1</span> - phi * phi));</span>
+<span id="cb14-16"><a href="#cb14-16" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:T) {</span>
+<span id="cb14-17"><a href="#cb14-17" aria-hidden="true" tabindex="-1"></a>    h[t] ~ normal(mu + phi * (h[t - <span class="dv">1</span>] -  mu), sigma);</span>
+<span id="cb14-18"><a href="#cb14-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb14-19"><a href="#cb14-19" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">1</span>:T) {</span>
+<span id="cb14-20"><a href="#cb14-20" aria-hidden="true" tabindex="-1"></a>    y[t] ~ normal(<span class="dv">0</span>, exp(h[t] / <span class="dv">2</span>));</span>
+<span id="cb14-21"><a href="#cb14-21" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb14-22"><a href="#cb14-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compared to the Kim et al.&nbsp;formulation, the Stan model adds priors for the parameters <span class="math inline">\(\phi\)</span>, <span class="math inline">\(\sigma\)</span>, and <span class="math inline">\(\mu\)</span>. The shock terms <span class="math inline">\(\epsilon_t\)</span> and <span class="math inline">\(\delta_t\)</span> do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.</p>
+<p>The posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with <span class="math inline">\(\mu = -1.02\)</span>, <span class="math inline">\(\phi = 0.95\)</span>, and <span class="math inline">\(\sigma = 0.25\)</span> leads to 95% posterior intervals for <span class="math inline">\(\mu\)</span> of <span class="math inline">\((-1.23, -0.54)\)</span>, for <span class="math inline">\(\phi\)</span> of <span class="math inline">\((0.82, 0.98)\)</span>, and for <span class="math inline">\(\sigma\)</span> of <span class="math inline">\((0.16, 0.38)\)</span>.</p>
+<p>The NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in <span class="citation" data-cites="Hoffman-Gelman:2014">(<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">Hoffman and Gelman 2014</a>)</span>. Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of <span class="math inline">\(T\)</span>.</p>
+<p>It is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return <span class="math inline">\(y\)</span> is easily vectorized to</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(<span class="dv">0</span>, exp(h / <span class="dv">2</span>));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter <code>h_std</code> to be declared instead of <code>h</code>.</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] h_std;  <span class="co">// std log volatility time t</span></span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The original value of <code>h</code> is then defined in a transformed parameter block.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[T] h = h_std * sigma;  <span class="co">// now h ~ normal(0, sigma)</span></span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>  h[<span class="dv">1</span>] /= sqrt(<span class="dv">1</span> - phi * phi);  <span class="co">// rescale h[1]</span></span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>  h += mu;</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (t <span class="cf">in</span> <span class="dv">2</span>:T) {</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a>    h[t] += phi * (h[t - <span class="dv">1</span>] - mu);</span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The first assignment rescales <code>h_std</code> to have a <span class="math inline">\(\textsf{normal}(0,\sigma)\)</span> distribution and temporarily assigns it to <code>h</code>. The second assignment rescales <code>h[1]</code> so that its prior differs from that of <code>h[2]</code> through <code>h[T]</code>. The next assignment supplies a <code>mu</code> offset, so that <code>h[2]</code> through <code>h[T]</code> are now distributed <span class="math inline">\(\textsf{normal}(\mu,\sigma)\)</span>; note that this shift must be done after the rescaling of <code>h[1]</code>. The final loop adds in the moving average so that <code>h[2]</code> through <code>h[T]</code> are appropriately modeled relative to <code>phi</code> and <code>mu</code>.</p>
+<p>As a final improvement, the distribution statements for <code>h[1]</code> to <code>h[T]</code> are replaced with a single vectorized standard normal distribution statement.</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>  h_std ~ std_normal();</span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.</p>
+</section>
+<section id="hmms.section" class="level2">
+<h2 class="anchored" data-anchor-id="hmms.section">Hidden Markov models</h2>
+<p>A Hidden Markov model is a probabilistic model over <span class="math inline">\(N\)</span> observations <span class="math inline">\(y_{1:N}\)</span> and <span class="math inline">\(N\)</span> hidden states <span class="math inline">\(z_{1:N}\)</span>. This models is defined by the conditional distributions <span class="math inline">\(p(y_n \mid z_n, \phi)\)</span> and <span class="math inline">\(p(z_n \mid z_{n-1}, \phi)\)</span>. Here we make the dependency on additional model parameters <span class="math inline">\(\phi\)</span> explicit. (<span class="math inline">\(\phi\)</span> may be a vector of parameters.) The complete data likelihood is then <span class="math display">\[
+p(y, z \mid \phi) = \prod_n p(y_n \mid z_n, \phi) p(z_n \mid z_{n - 1}, \phi)
+\]</span> When <span class="math inline">\(z_{1:N}\)</span> is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate <span class="math inline">\(z\)</span> out.</p>
+<p>When each state <span class="math inline">\(z\)</span> takes a value over a discrete and finite set, say <span class="math inline">\(\{1, 2, ..., K\}\)</span>, we can use Stan’s suite of HMM functions to marginalize out <span class="math inline">\(z_{1:N}\)</span> and compute <span class="math display">\[
+p(y_{1:N} \mid \phi) = \int_{\mathcal Z} p(y, z \mid \phi) \text d z.
+\]</span> We start by defining the conditional observation distribution, stored in a <span class="math inline">\(K \times N\)</span> matrix <span class="math inline">\(\omega\)</span> with <span class="math display">\[
+\omega_{kn} = p(y_n \mid z_n = k, \phi).
+\]</span> Next, we introduce the <span class="math inline">\(K \times K\)</span> transition matrix, <span class="math inline">\(\Gamma\)</span>, with <span class="math display">\[
+\Gamma_{ij} = p(z_n = j \mid z_{n - 1} = i, \phi).
+\]</span> (This is a right-stochastic matrix.) Finally, we define the initial state <span class="math inline">\(K\)</span>-vector <span class="math inline">\(\rho\)</span>, with <span class="math display">\[
+\rho_k = p(z_0 = k \mid \phi).
+\]</span> It is common practice to set <span class="math inline">\(\rho\)</span> to be the stationary distribution of the HMM, that is <span class="math inline">\(\rho\)</span> is the first eigenvector of <span class="math inline">\(\Gamma\)</span> and solves <span class="math inline">\(\Gamma \rho = \rho\)</span>.</p>
+<p>As an example, consider a three-state model with <span class="math inline">\(K=3\)</span>. The observations are normally distributed conditional on the HMM states with <span class="math display">\[
+  y_n \sim \text{normal}(\mu_k, \sigma),
+\]</span> where <span class="math inline">\(\mu = (1, 5, 9)\)</span> and the standard deviation <span class="math inline">\(\sigma\)</span> is the same across all observations. The model is then</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;  <span class="co">// Number of observations</span></span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Rows of the transition matrix</span></span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">simplex</span>[<span class="dv">3</span>] gamma_arr;</span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Initial state</span></span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">simplex</span>[<span class="dv">3</span>] rho;</span>
+<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-13"><a href="#cb19-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Parameters of measurement model</span></span>
+<span id="cb19-14"><a href="#cb19-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] mu;</span>
+<span id="cb19-15"><a href="#cb19-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span> = <span class="fl">0.0</span>&gt; sigma;</span>
+<span id="cb19-16"><a href="#cb19-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-17"><a href="#cb19-17" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-18"><a href="#cb19-18" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb19-19"><a href="#cb19-19" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Build transition matrix</span></span>
+<span id="cb19-20"><a href="#cb19-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">3</span>] gamma;</span>
+<span id="cb19-21"><a href="#cb19-21" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">3</span>) gamma[k, ] = to_row_vector(gamma_arr[k]);</span>
+<span id="cb19-22"><a href="#cb19-22" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-23"><a href="#cb19-23" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Compute the log likelihoods in each possible state</span></span>
+<span id="cb19-24"><a href="#cb19-24" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">3</span>, N] log_omega;</span>
+<span id="cb19-25"><a href="#cb19-25" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb19-26"><a href="#cb19-26" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">3</span>) {</span>
+<span id="cb19-27"><a href="#cb19-27" aria-hidden="true" tabindex="-1"></a>      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);</span>
+<span id="cb19-28"><a href="#cb19-28" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb19-29"><a href="#cb19-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb19-30"><a href="#cb19-30" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-31"><a href="#cb19-31" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-32"><a href="#cb19-32" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb19-33"><a href="#cb19-33" aria-hidden="true" tabindex="-1"></a>  <span class="co">// prior</span></span>
+<span id="cb19-34"><a href="#cb19-34" aria-hidden="true" tabindex="-1"></a>  mu ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb19-35"><a href="#cb19-35" aria-hidden="true" tabindex="-1"></a>  sigma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb19-36"><a href="#cb19-36" aria-hidden="true" tabindex="-1"></a>  </span>
+<span id="cb19-37"><a href="#cb19-37" aria-hidden="true" tabindex="-1"></a>  <span class="co">// no explicit prior on gamma_arr, meaning we default to a</span></span>
+<span id="cb19-38"><a href="#cb19-38" aria-hidden="true" tabindex="-1"></a>  <span class="co">// uniform prior over the simplexes.</span></span>
+<span id="cb19-39"><a href="#cb19-39" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-40"><a href="#cb19-40" aria-hidden="true" tabindex="-1"></a>  <span class="co">// Increment target by log p(y | mu, sigma, Gamma, rho)</span></span>
+<span id="cb19-41"><a href="#cb19-41" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> hmm_marginal(log_omega, gamma, rho);</span>
+<span id="cb19-42"><a href="#cb19-42" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The last function <code>hmm_marginal</code> takes in all the ingredients of the HMM and computes the relevant log marginal distribution, <span class="math inline">\(\log p(y \mid \phi)\)</span>.</p>
+<p>If we desire draws from the posterior distribution of <span class="math inline">\(z\)</span>, we use the generated quantities block and draw, for each sample <span class="math inline">\(\phi\)</span>, a sample from <span class="math inline">\(p(z \mid y, \phi)\)</span>. In effect, MCMC produces draws from <span class="math inline">\(p(\phi \mid y)\)</span> and with the draws in generated quantities, we obtain draws from <span class="math inline">\(p(\phi \mid y) p(z \mid y, \phi) = p(z, \phi \mid y)\)</span>. It is also possible to compute the posterior probbability of each hidden state, that is <span class="math inline">\(\text{Pr}(z_n = k \mid \phi, y)\)</span>. Averagging these probabilities over all MCMC draws, we obtain <span class="math inline">\(\text{Pr}(z_n = k \mid y)\)</span>.</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> latent_states = hmm_latent_rng(log_omega, gamma, rho);</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[<span class="dv">3</span>, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);</span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><code>hmm_hidden_state_prob</code> returns the marginal probabilities of each state, <span class="math inline">\(\text{Pr}(z_n = k \mid \phi, y)\)</span>. This function cannot be used to compute the joint probability <span class="math inline">\(\text{Pr}(z \mid \phi, y)\)</span>, because such calculation requires accounting for the posterior correlation between the different components of <span class="math inline">\(z\)</span>. Therefore, <code>hidden_probs</code> should <em>not</em> be used to obtain posterior draws. Instead, users should rely on <code>hmm_latent_rng</code>.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>   <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; z = hmm_latent_rng(...fill-<span class="cf">in</span> params here to match example...);</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The example in this section is derived from the more detailed case study by Ben Bales: <a href="https://mc-stan.org/users/documentation/case-studies/hmm-example.html">https://mc-stan.org/users/documentation/case-studies/hmm-example.html</a>.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Engle:1982" class="csl-entry" role="listitem">
+Engle, Robert F. 1982. <span>“Autoregressive Conditional Heteroscedasticity with Estimates of Variance of <span>U</span>nited <span>K</span>ingdom Inflation.”</span> <em>Econometrica</em> 50: 987–1008.
+</div>
+<div id="ref-Hoffman-Gelman:2014" class="csl-entry" role="listitem">
+Hoffman, Matthew D., and Andrew Gelman. 2014. <span>“<span>T</span>he <span>N</span>o-<span>U</span>-<span>T</span>urn <span>S</span>ampler: <span>A</span>daptively <span>S</span>etting <span>P</span>ath <span>L</span>engths in <span>H</span>amiltonian <span>M</span>onte <span>C</span>arlo.”</span> <em>Journal of Machine Learning Research</em> 15: 1593–623. <a href="http://jmlr.org/papers/v15/hoffman14a.html">http://jmlr.org/papers/v15/hoffman14a.html</a>.
+</div>
+<div id="ref-KimShephardChib:1998" class="csl-entry" role="listitem">
+Kim, Sangjoon, Neil Shephard, and Siddhartha Chib. 1998. <span>“Stochastic Volatility: Likelihood Inference and Comparison with <span>ARCH</span> Models.”</span> <em>Review of Economic Studies</em> 65: 361–93.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The intercept in this model is <span class="math inline">\(\alpha / (1 - \beta)\)</span>. An alternative parameterization in terms of an intercept <span class="math inline">\(\gamma\)</span> suggested Mark Scheuerell on GitHub is <span class="math inline">\(y_n \sim \textsf{normal}\left(\gamma + \beta \cdot (y_{n-1} - \gamma), \sigma\right)\)</span>.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>In practice, it can be useful to remove the constraint to test whether a non-stationary set of coefficients provides a better fit to the data. It can also be useful to add a trend term to the model, because an unfitted trend will manifest as non-stationarity.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>This subsection is a lightly edited comment of Jonathan Gilligan’s on GitHub; see <a href="https://github.com/stan-dev/stan/issues/1617#issuecomment-160249142">https://github.com/stan-dev/stan/issues/1617#issuecomment-160249142</a>.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/regression.html" class="pagination-link" aria-label="Regression Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Regression Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/missing-data.html" class="pagination-link" aria-label="Missing Data and Partially Known Parameters">
+        <span class="nav-page-text">Missing Data and Partially Known Parameters</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/time-series.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/truncation-censoring.html b/docs/2_39/stan-users-guide/truncation-censoring.html
new file mode 100644
index 000000000..fe4b0bdc9
--- /dev/null
+++ b/docs/2_39/stan-users-guide/truncation-censoring.html
@@ -0,0 +1,1355 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Truncated or Censored Data</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/finite-mixtures.html" rel="next">
+<link href="../stan-users-guide/missing-data.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Truncated or Censored Data">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/truncation-censoring.html">Truncated or Censored Data</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#truncated-or-censored-data" id="toc-truncated-or-censored-data" class="nav-link active" data-scroll-target="#truncated-or-censored-data">Truncated or Censored Data</a>
+  <ul class="collapse">
+  <li><a href="#truncation.section" id="toc-truncation.section" class="nav-link" data-scroll-target="#truncation.section">Truncated distributions</a></li>
+  <li><a href="#truncated-data.section" id="toc-truncated-data.section" class="nav-link" data-scroll-target="#truncated-data.section">Truncated data</a>
+  <ul class="collapse">
+  <li><a href="#constraints-and-out-of-bounds-returns" id="toc-constraints-and-out-of-bounds-returns" class="nav-link" data-scroll-target="#constraints-and-out-of-bounds-returns">Constraints and out-of-bounds returns</a></li>
+  <li><a href="#unknown-truncation-points" id="toc-unknown-truncation-points" class="nav-link" data-scroll-target="#unknown-truncation-points">Unknown truncation points</a></li>
+  </ul></li>
+  <li><a href="#censored.section" id="toc-censored.section" class="nav-link" data-scroll-target="#censored.section">Censored data</a>
+  <ul class="collapse">
+  <li><a href="#estimating-censored-values" id="toc-estimating-censored-values" class="nav-link" data-scroll-target="#estimating-censored-values">Estimating censored values</a></li>
+  <li><a href="#integrating-out-censored-values" id="toc-integrating-out-censored-values" class="nav-link" data-scroll-target="#integrating-out-censored-values">Integrating out censored values</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/truncation-censoring.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/truncation-censoring.html">Truncated or Censored Data</a></li></ol></nav></header>
+
+
+
+
+<section id="truncated-or-censored-data" class="level1">
+<h1>Truncated or Censored Data</h1>
+<p>Data in which measurements have been truncated or censored can be coded in Stan following their respective probability models.</p>
+<section id="truncation.section" class="level2">
+<h2 class="anchored" data-anchor-id="truncation.section">Truncated distributions</h2>
+<p>Truncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.</p>
+</section>
+<section id="truncated-data.section" class="level2">
+<h2 class="anchored" data-anchor-id="truncated-data.section">Truncated data</h2>
+<p>Truncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.</p>
+<p>Truncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are <span class="math inline">\(y_n\)</span> with an upper truncation point of <span class="math inline">\(U = 300\)</span> so that <span class="math inline">\(y_n &lt; 300\)</span>. In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> U;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span>&lt;<span class="kw">upper</span>=U&gt; y;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma) <span class="kw">T</span>[ , U];</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The model declares an upper bound <code>U</code> as data and constrains the data for <code>y</code> to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.</p>
+<p>See the Stan Reference Manual’s <a href="https://mc-stan.org/docs/reference-manual/statements.html#truncated-distributions">Statements chapter</a> for how to use truncated distributions with the log probability increment statements (<code>target += ...</code>).</p>
+<p>This model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.</p>
+<section id="constraints-and-out-of-bounds-returns" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="constraints-and-out-of-bounds-returns">Constraints and out-of-bounds returns</h3>
+<p>If the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to <span class="math inline">\(-\infty\)</span>. For instance, if variate <code>y</code> is sampled with the statement.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a>y ~ normal(mu, sigma) <span class="kw">T</span>[L, U];</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>then if any value inside <code>y</code> is less than the value of <code>L</code> or greater than the value of <code>U</code>, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.</p>
+<p>To avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if <code>y</code> is a parameter in the above model, the declaration should constrain it to fall between the values of <code>L</code> and <code>U</code>.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span>&lt;<span class="kw">lower</span>=L, <span class="kw">upper</span>=U&gt; y;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If in the above model, <code>L</code> or <code>U</code> is a parameter and <code>y</code> is data, then <code>L</code> and <code>U</code> must be appropriately constrained so that all data are in range and the value of <code>L</code> is less than that of <code>U</code> (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">upper</span>=min(y)&gt; L;           <span class="co">// L &lt; y[n]</span></span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=fmax(L, max(y))&gt; U;  <span class="co">// L &lt; U; y[n] &lt; U</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For pairs of real numbers, the function <code>fmax</code> is used rather than <code>max</code>.</p>
+</section>
+<section id="unknown-truncation-points" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unknown-truncation-points">Unknown truncation points</h3>
+<p>If the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> y;</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">upper</span>=min(y)&gt; L;</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=max(y)&gt; U;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>  L ~ <span class="co">// ...</span></span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>  U ~ <span class="co">// ...</span></span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma) <span class="kw">T</span>[L, U];</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Here there is a lower truncation point <code>L</code> which is declared to be less than or equal to the minimum value of <code>y</code>. The upper truncation point <code>U</code> is declared to be larger than the maximum value of <code>y</code>. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With <code>N</code> declared as type <code>int&lt;lower=1&gt;</code>, there must be at least one data point. The constraint that <code>L</code> is less than <code>U</code> is enforced indirectly, based on the non-empty data.</p>
+<p>The ellipses where the priors for the bounds <code>L</code> and <code>U</code> should go should be filled in with a an informative prior in order for this model to not concentrate <code>L</code> strongly around <code>min(y)</code> and <code>U</code> strongly around <code>max(y)</code>.</p>
+</section>
+</section>
+<section id="censored.section" class="level2">
+<h2 class="anchored" data-anchor-id="censored.section">Censored data</h2>
+<p>Censoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.</p>
+<section id="estimating-censored-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="estimating-censored-values">Estimating censored values</h3>
+<p>One way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_cens;</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">real</span> y_obs;</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=max(y_obs)&gt; U;</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_cens] <span class="dt">real</span>&lt;<span class="kw">lower</span>=U&gt; y_cens;</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>  y_obs ~ normal(mu, sigma);</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>  y_cens ~ normal(mu, sigma);</span>
+<span id="cb6-15"><a href="#cb6-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Because the censored data array <code>y_cens</code> is declared to be a parameter, it will be sampled along with the location and scale parameters <code>mu</code> and <code>sigma</code>. Because the censored data array <code>y_cens</code> is declared to have values of type <code>real&lt;lower=U&gt;</code>, all imputed values for censored data will be greater than <code>U</code>. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.</p>
+</section>
+<section id="integrating-out-censored-values" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="integrating-out-censored-values">Integrating out censored values</h3>
+<p>Although it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of <span class="math display">\[\begin{align*}
+\Pr[y_{\mathrm{cens},m} &gt; U]
+  &amp;= \int_U^{\infty} \textsf{normal}\left(y_{\mathrm{cens},m} \mid \mu,\sigma \right) \,\textsf{d}y_{\mathrm{cens},m} \\
+  &amp;= 1 - \Phi\left(\frac{U - \mu}{\sigma}\right),
+\end{align*}\]</span></p>
+<p>where <span class="math inline">\(\Phi()\)</span> is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that <span class="math inline">\(y_{\mathrm{cens},m}&gt;U\)</span>. With <span class="math inline">\(M\)</span> censored observations, the likelihood on the log scale is <span class="math display">\[\begin{align*}
+\log \prod_{m=1}^M \Pr[y_{\mathrm{cens},m} &gt; U]
+  &amp;= \log \left( 1 - \Phi\left(\left(\frac{U - \mu}{\sigma}\right)\right)^{M}\right) \\
+  &amp;= M \times \texttt{normal}\mathtt{\_}\texttt{lccdf}\left(U \mid \mu, \sigma \right),
+\end{align*}\]</span></p>
+<p>where <code>normal_lccdf</code> is the log of complementary CDF (Stan provides <code>&lt;distr&gt;_lccdf</code> for each distribution implemented in Stan).</p>
+<p>The following right-censored model assumes that the censoring point is known, so it is declared as data.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_cens;</span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">real</span> y_obs;</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=max(y_obs)&gt; U;</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>  y_obs ~ normal(mu, sigma);</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens * normal_lccdf(U | mu, sigma);</span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For the observed values in <code>y_obs</code>, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.</p>
+<p>For the left-censored data the CDF (<code>normal_lcdf</code>) has to be used instead of complementary CDF. If the censoring point variable (<code>L</code>) is unknown, its declaration should be moved from the data to the parameters block.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_cens;</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">real</span> y_obs;</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">upper</span>=min(y_obs)&gt; L;</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>  L ~ normal(mu, sigma);</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  y_obs ~ normal(mu, sigma);</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens * normal_lcdf(L | mu, sigma);</span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/missing-data.html" class="pagination-link" aria-label="Missing Data and Partially Known Parameters">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Missing Data and Partially Known Parameters</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/finite-mixtures.html" class="pagination-link" aria-label="Finite Mixtures">
+        <span class="nav-page-text">Finite Mixtures</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/truncation-censoring.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/user-functions.html b/docs/2_39/stan-users-guide/user-functions.html
new file mode 100644
index 000000000..f5a2f7e71
--- /dev/null
+++ b/docs/2_39/stan-users-guide/user-functions.html
@@ -0,0 +1,1642 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>User-Defined Functions</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/custom-probability.html" rel="next">
+<link href="../stan-users-guide/multi-indexing.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="User-Defined Functions">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/user-functions.html">User-Defined Functions</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#functions-programming.chapter" id="toc-functions-programming.chapter" class="nav-link active" data-scroll-target="#functions-programming.chapter">User-Defined Functions</a>
+  <ul class="collapse">
+  <li><a href="#basic-functions.section" id="toc-basic-functions.section" class="nav-link" data-scroll-target="#basic-functions.section">Basic functions</a>
+  <ul class="collapse">
+  <li><a href="#user-defined-functions-block" id="toc-user-defined-functions-block" class="nav-link" data-scroll-target="#user-defined-functions-block">User-defined functions block</a></li>
+  <li><a href="#function-bodies" id="toc-function-bodies" class="nav-link" data-scroll-target="#function-bodies">Function bodies</a></li>
+  <li><a href="#return-statements" id="toc-return-statements" class="nav-link" data-scroll-target="#return-statements">Return statements</a></li>
+  <li><a href="#reject-and-error-statements" id="toc-reject-and-error-statements" class="nav-link" data-scroll-target="#reject-and-error-statements">Reject and error statements</a></li>
+  <li><a href="#type-declarations-for-functions" id="toc-type-declarations-for-functions" class="nav-link" data-scroll-target="#type-declarations-for-functions">Type declarations for functions</a></li>
+  <li><a href="#array-types-for-function-declarations" id="toc-array-types-for-function-declarations" class="nav-link" data-scroll-target="#array-types-for-function-declarations">Array types for function declarations</a></li>
+  <li><a href="#data-only-function-arguments" id="toc-data-only-function-arguments" class="nav-link" data-scroll-target="#data-only-function-arguments">Data-only function arguments</a></li>
+  </ul></li>
+  <li><a href="#functions-as-statements" id="toc-functions-as-statements" class="nav-link" data-scroll-target="#functions-as-statements">Functions as statements</a></li>
+  <li><a href="#functions-accessing-the-log-probability-accumulator" id="toc-functions-accessing-the-log-probability-accumulator" class="nav-link" data-scroll-target="#functions-accessing-the-log-probability-accumulator">Functions accessing the log probability accumulator</a></li>
+  <li><a href="#functions-implementing-change-of-variable-adjustments" id="toc-functions-implementing-change-of-variable-adjustments" class="nav-link" data-scroll-target="#functions-implementing-change-of-variable-adjustments">Functions implementing change-of-variable adjustments</a></li>
+  <li><a href="#functions-acting-as-random-number-generators" id="toc-functions-acting-as-random-number-generators" class="nav-link" data-scroll-target="#functions-acting-as-random-number-generators">Functions acting as random number generators</a></li>
+  <li><a href="#user-defined-probability-functions" id="toc-user-defined-probability-functions" class="nav-link" data-scroll-target="#user-defined-probability-functions">User-defined probability functions</a></li>
+  <li><a href="#overloading-functions" id="toc-overloading-functions" class="nav-link" data-scroll-target="#overloading-functions">Overloading functions</a>
+  <ul class="collapse">
+  <li><a href="#warning-on-usage" id="toc-warning-on-usage" class="nav-link" data-scroll-target="#warning-on-usage">Warning on usage</a></li>
+  <li><a href="#function-resolution" id="toc-function-resolution" class="nav-link" data-scroll-target="#function-resolution">Function resolution</a></li>
+  </ul></li>
+  <li><a href="#documenting-functions.section" id="toc-documenting-functions.section" class="nav-link" data-scroll-target="#documenting-functions.section">Documenting functions</a></li>
+  <li><a href="#summary-of-function-types" id="toc-summary-of-function-types" class="nav-link" data-scroll-target="#summary-of-function-types">Summary of function types</a>
+  <ul class="collapse">
+  <li><a href="#void-vs.-non-void-return" id="toc-void-vs.-non-void-return" class="nav-link" data-scroll-target="#void-vs.-non-void-return">Void vs.&nbsp;non-void return</a></li>
+  <li><a href="#suffixed-or-non-suffixed" id="toc-suffixed-or-non-suffixed" class="nav-link" data-scroll-target="#suffixed-or-non-suffixed">Suffixed or non-suffixed</a></li>
+  </ul></li>
+  <li><a href="#recursive-functions" id="toc-recursive-functions" class="nav-link" data-scroll-target="#recursive-functions">Recursive functions</a></li>
+  <li><a href="#truncated-random-number-generation" id="toc-truncated-random-number-generation" class="nav-link" data-scroll-target="#truncated-random-number-generation">Truncated random number generation</a>
+  <ul class="collapse">
+  <li><a href="#generation-with-inverse-cdfs" id="toc-generation-with-inverse-cdfs" class="nav-link" data-scroll-target="#generation-with-inverse-cdfs">Generation with inverse CDFs</a></li>
+  <li><a href="#truncated-variate-generation" id="toc-truncated-variate-generation" class="nav-link" data-scroll-target="#truncated-variate-generation">Truncated variate generation</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/floating-point.html">Programming Techniques</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/user-functions.html">User-Defined Functions</a></li></ol></nav></header>
+
+
+
+
+<section id="functions-programming.chapter" class="level1">
+<h1>User-Defined Functions</h1>
+<p>This chapter explains functions from a user perspective with examples; see the language reference for a full specification. User-defined functions allow computations to be encapsulated into a single named unit and invoked elsewhere by name. Similarly, functions allow complex procedures to be broken down into more understandable components. Writing modular code using descriptively named functions is easier to understand than a monolithic program, even if the latter is heavily commented.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></p>
+<section id="basic-functions.section" class="level2">
+<h2 class="anchored" data-anchor-id="basic-functions.section">Basic functions</h2>
+<p>Here’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> relative_diff(<span class="dt">real</span> x, <span class="dt">real</span> y) {</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> abs_diff;</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> avg_scale;</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>    abs_diff = abs(x - y);</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>    avg_scale = (abs(x) + abs(y)) / <span class="dv">2</span>;</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> abs_diff / avg_scale;</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> rdiff;</span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  rdiff = relative_diff(alpha, beta);</span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The function is named <code>relative_diff</code>, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.</p>
+<section id="user-defined-functions-block" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="user-defined-functions-block">User-defined functions block</h3>
+<p>All functions are defined in their own block, which is labeled <code>functions</code> and must appear before all other program blocks. The user-defined functions block is optional.</p>
+</section>
+<section id="function-bodies" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="function-bodies">Function bodies</h3>
+<p>The body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.</p>
+</section>
+<section id="return-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="return-statements">Return statements</h3>
+<p>Return statements, such as the one on the last line of the definition of <code>relative_diff</code> above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.</p>
+</section>
+<section id="reject-and-error-statements" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="reject-and-error-statements">Reject and error statements</h3>
+<p>The Stan <code>reject</code> statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.</p>
+<p>If an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a <code>fatal_error</code> statement.</p>
+<section id="catching-errors" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="catching-errors">Catching errors</h4>
+<p>Rejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).</p>
+<p>The most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> dbl_sqrt(<span class="dt">real</span> x) {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (!(x &gt;= <span class="dv">0</span>)) {</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>    <span class="kw">reject</span>(<span class="st">"dbl_sqrt(x): x must be positive; found x = "</span>, x);</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> <span class="dv">2</span> * sqrt(x);</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The negation of the positive test is important, because it also catches the case where <code>x</code> is a not-a-number value. If the condition had been coded as <code>(x &lt; 0)</code> it would not catch the not-a-number case, though it could be written as <code>(x &lt; 0 || is_nan(x))</code>. The positive infinite case is allowed through, but could also be checked with the <code>is_inf(x)</code> function. The square root function does not itself reject, but some downstream consumer of <code>dbl_sqrt(-2)</code> would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.</p>
+<p>The effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
+<p>In generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.</p>
+</section>
+</section>
+<section id="type-declarations-for-functions" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="type-declarations-for-functions">Type declarations for functions</h3>
+<p>Function argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the <a href="https://mc-stan.org/docs/reference-manual/types.html#variable-declaration">reference manual</a> for full details.</p>
+<p>For example, here’s a function to compute the entropy of a categorical distribution with simplex parameter <code>theta</code>.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> entropy(<span class="dt">vector</span> theta) {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> sum(theta .* log(theta));</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Although <code>theta</code> must be a simplex, only the type <code>vector</code> is used.<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
+<p>Upper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.</p>
+</section>
+<section id="array-types-for-function-declarations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="array-types-for-function-declarations">Array types for function declarations</h3>
+<p>Array arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[] <span class="dt">real</span> baz(<span class="dt">array</span>[,] <span class="dt">real</span> x);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The notation <code>[ ]</code> is used for one-dimensional arrays (as in the return above), <code>[ , ]</code> for two-dimensional arrays, <code>[ , , ]</code> for three-dimensional arrays, and so on.</p>
+<p>Functions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.</p>
+</section>
+<section id="data-only-function-arguments" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="data-only-function-arguments">Data-only function arguments</h3>
+<p>A function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier <code>data</code>. The following is an example of a data-only function argument.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> y, <span class="kw">data</span> <span class="dt">real</span> mu) {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> -<span class="fl">0.5</span> * (y - mu)^<span class="dv">2</span>;</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>This qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.</p>
+<p>Use of the <code>data</code> qualifier must be consistent between the forward declaration and the definition of a functions.</p>
+<p>This qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the <a href="../stan-users-guide/odes.html">ODE solver chapter</a> for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the <code>data</code> prefix qualifier. This allows for compile-time type checking and increases overall program understandability.</p>
+</section>
+</section>
+<section id="functions-as-statements" class="level2">
+<h2 class="anchored" data-anchor-id="functions-as-statements">Functions as statements</h2>
+<p>In some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">void</span> pretty_print_tri_lower(<span class="dt">matrix</span> x) {</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (rows(x) == <span class="dv">0</span>) {</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>      <span class="kw">print</span>(<span class="st">"empty matrix"</span>);</span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>      <span class="cf">return</span>;</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>    <span class="kw">print</span>(<span class="st">"rows="</span>, rows(x), <span class="st">" cols="</span>, cols(x));</span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:rows(x)) {</span>
+<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:m) {</span>
+<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a>        <span class="kw">print</span>(<span class="st">"["</span>, m, <span class="st">","</span>, n, <span class="st">"]="</span>, x[m, n]);</span>
+<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The special symbol <code>void</code> is used as the return type. This is not a type itself in that there are no values of type <code>void</code>; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.</p>
+<p>Void functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ... code to set Sigma ...</span></span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>  pretty_print_tri_lower(Sigma);</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="functions-accessing-the-log-probability-accumulator" class="level2">
+<h2 class="anchored" data-anchor-id="functions-accessing-the-log-probability-accumulator">Functions accessing the log probability accumulator</h2>
+<p>Functions whose names end in <code>_lp</code> are allowed to use sampling statements and <code>target +=</code> statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.</p>
+<p>Here is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the <a href="../stan-users-guide/efficiency-tuning.html#reparameterization.section">reparameterization section</a> for more information on efficient non-centered parameterizations</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> center_lp(<span class="dt">vector</span> beta_raw, <span class="dt">real</span> mu, <span class="dt">real</span> sigma) {</span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>    beta_raw ~ std_normal();</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>    sigma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>    mu ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> sigma * beta_raw + mu;</span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta_raw;</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;</span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;</span>
+<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-19"><a href="#cb8-19" aria-hidden="true" tabindex="-1"></a>  beta = center_lp(beta_raw, mu_beta, sigma_beta);</span>
+<span id="cb8-20"><a href="#cb8-20" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb8-21"><a href="#cb8-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="functions-implementing-change-of-variable-adjustments" class="level2">
+<h2 class="anchored" data-anchor-id="functions-implementing-change-of-variable-adjustments">Functions implementing change-of-variable adjustments</h2>
+<p>Functions whose names end in <code>_jacobian</code> can use the <code>jacobian +=</code> statement. This can be used to implement a custom change of variables for arbitrary parameters.</p>
+<p>For example, this function recreates the built-in <code>&lt;upper=x&gt;</code> transform on real numbers:</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> my_upper_bound_jacobian(<span class="dt">real</span> x, <span class="dt">real</span> ub) {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="kw">jacobian +=</span> x;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> ub - exp(x);</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It can be used as a replacement for <code>real&lt;lower=ub&gt;</code> as follows:</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// my_upper_bound_jacobian as above</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ub;</span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b_raw;</span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b = my_upper_bound_jacobian(b_raw, ub);</span>
+<span id="cb10-12"><a href="#cb10-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb10-13"><a href="#cb10-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-14"><a href="#cb10-14" aria-hidden="true" tabindex="-1"></a>  b ~ lognormal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb10-15"><a href="#cb10-15" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb10-16"><a href="#cb10-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="functions-acting-as-random-number-generators" class="level2">
+<h2 class="anchored" data-anchor-id="functions-acting-as-random-number-generators">Functions acting as random number generators</h2>
+<p>A user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in <code>_rng</code>. Giving a function a name that ends in <code>_rng</code> allows it to access built-in functions and user-defined functions that end in <code>_rng</code>, which includes all the built-in PRNG functions. Only functions ending in <code>_rng</code> are able access the built-in PRNG functions. The use of functions ending in <code>_rng</code> must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in <code>_rng</code>.</p>
+<p>For example, the following function generates an <span class="math inline">\(N \times K\)</span> data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> predictors_rng(<span class="dt">int</span> N, <span class="dt">int</span> K) {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>    x[n, <span class="dv">1</span>] = <span class="fl">1.0</span>;  <span class="co">// intercept</span></span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">2</span>:K) {</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>      x[n, k] = normal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> x;</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The following function defines a simulator for regression outcomes based on a data matrix <code>x</code>, coefficients <code>beta</code>, and noise scale <code>sigma</code>.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span> regression_rng(<span class="dt">vector</span> beta, <span class="dt">matrix</span> x, <span class="dt">real</span> sigma) {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[rows(x)] y;</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[rows(x)] mu;</span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>  mu = x * beta;</span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:rows(x)) {</span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>    y[n] = normal_rng(mu[n], sigma);</span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> y;</span>
+<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N_sim, K] x_sim;</span>
+<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_sim] y_sim;</span>
+<span id="cb13-9"><a href="#cb13-9" aria-hidden="true" tabindex="-1"></a>  x_sim = predictors_rng(N_sim, K);</span>
+<span id="cb13-10"><a href="#cb13-10" aria-hidden="true" tabindex="-1"></a>  y_sim = regression_rng(beta, x_sim, sigma);</span>
+<span id="cb13-11"><a href="#cb13-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A more sophisticated simulation might fit a multivariate&nbsp;normal to the predictors <code>x</code> and use the resulting parameters to generate multivariate normal draws for <code>x_sim</code>.</p>
+</section>
+<section id="user-defined-probability-functions" class="level2">
+<h2 class="anchored" data-anchor-id="user-defined-probability-functions">User-defined probability functions</h2>
+<p>Probability functions are distinguished in Stan by names ending in <code>_lpdf</code> for density functions and <code>_lpmf</code> for mass functions; in both cases, they must have <code>real</code> return types.</p>
+<p>Suppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> unit_normal_lpdf(<span class="dt">real</span> y) {</span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> normal_lpdf(y | <span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a>  alpha ~ unit_normal();</span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a>  beta ~ unit_normal();</span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The ability to use the <code>unit_normal</code> function as a density is keyed off its name ending in <code>_lpdf</code> (names ending in <code>_lpmf</code> for probability mass functions work the same way).</p>
+<p>In general, if <code>foo_lpdf</code> is defined to consume <span class="math inline">\(N + 1\)</span> arguments, then</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a>y ~ foo(theta1, ..., thetaN);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>can be used as shorthand for</p>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">target +=</span> foo_lpdf(y | theta1, ..., thetaN);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As with the built-in functions, the suffix <code>_lpdf</code> is dropped and the first argument moves to the left of the tilde symbol (<code>~</code>) in the distribution statement.</p>
+<p>Functions ending in <code>_lpmf</code> (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (<code>_lpdf</code>) must be continuous (not an integer or integer array), whereas the first argument of a mass function (<code>_lpmf</code>) must be discrete (integer or integer array).</p>
+</section>
+<section id="overloading-functions" class="level2">
+<h2 class="anchored" data-anchor-id="overloading-functions">Overloading functions</h2>
+<p>As described in the <a href="https://mc-stan.org/docs/reference-manual/user-functions.html#function-names">reference manual</a> function overloading is permitted in Stan, beginning in version 2.29.</p>
+<p>This means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.</p>
+<section id="warning-on-usage" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="warning-on-usage">Warning on usage</h3>
+<p>Overloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.</p>
+<p>For example, consider a function <code>triple</code> which has the following three signatures</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> triple(<span class="dt">real</span> x);</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> triple(<span class="dt">complex</span> x);</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[] <span class="dt">real</span> triple(<span class="dt">array</span>[] <span class="dt">real</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>One should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect</p>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="co">// The function does what it says</span></span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>triple(<span class="fl">3.0</span>) == <span class="fl">9.0</span></span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="co">// It is defined reasonably for different types</span></span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>triple(to_complex(<span class="fl">3.0</span>)) == to_complex(triple(<span class="fl">3.0</span>))</span>
+<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a><span class="co">// A container version of this function works by element</span></span>
+<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a>triple({<span class="fl">3.0</span>, <span class="fl">4.0</span>})[<span class="dv">0</span>] == triple({<span class="fl">3.0</span>, <span class="fl">4.0</span>}[<span class="dv">0</span>])</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Note that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.</p>
+</section>
+<section id="function-resolution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="function-resolution">Function resolution</h3>
+<p>Stan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.</p>
+<p>Consider the following function signatures</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">int</span> a, <span class="dt">real</span> b);</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> foo(<span class="dt">real</span> a, <span class="dt">real</span> b);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Given these, the function call <code>foo(1.5, 2.5)</code> is unambiguous - it must resolve to the second signature. But, the function call <code>foo(1, 1.5)</code> could be valid for <em>either</em> under Stan’s promotion rules, which allow integers to be promoted to real numbers.</p>
+<p>To resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call <code>foo(1, 1.5)</code> would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).</p>
+<p>Furthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures</p>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> bar(<span class="dt">int</span> x, <span class="dt">real</span> y);</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> bar(<span class="dt">real</span> x, <span class="dt">int</span> y);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>These signatures do not have a unique minimum number of promotions for the call <code>bar(1, 2)</code>. Both signatures require one <code>int</code> to <code>real</code> promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.</p>
+<p>Promotion from integers to complex numbers is considered to be two separate promotions, first from <code>int</code> to <code>real</code>, then from <code>real</code> to <code>complex</code>. This means that integer arguments will “prefer” a signature with real types over complex types.</p>
+<p>For example, consider the function signatures</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> pop(<span class="dt">real</span> x);</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> pop(<span class="dt">complex</span> x);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Stan will select the first signature when <code>pop</code> is called with an integer argument such as <code>pop(0)</code>.</p>
+</section>
+</section>
+<section id="documenting-functions.section" class="level2">
+<h2 class="anchored" data-anchor-id="documenting-functions.section">Documenting functions</h2>
+<p>Functions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.</p>
+<p>For example, here’s some documentation for the prediction matrix generator.</p>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="co">/**</span></span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a><span class="co"> * Return a data matrix of specified size with rows</span></span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a><span class="co"> * corresponding to items and the first column filled</span></span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a><span class="co"> * with the value 1 to represent the intercept and the</span></span>
+<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a><span class="co"> * remaining columns randomly filled with unit-normal draws.</span></span>
+<span id="cb22-6"><a href="#cb22-6" aria-hidden="true" tabindex="-1"></a><span class="co"> *</span></span>
+<span id="cb22-7"><a href="#cb22-7" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@param</span><span class="co"> N Number of rows corresponding to data items</span></span>
+<span id="cb22-8"><a href="#cb22-8" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@param</span><span class="co"> K Number of predictors, counting the intercept, per</span></span>
+<span id="cb22-9"><a href="#cb22-9" aria-hidden="true" tabindex="-1"></a><span class="co"> *          item.</span></span>
+<span id="cb22-10"><a href="#cb22-10" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@return</span><span class="co"> Simulated predictor matrix.</span></span>
+<span id="cb22-11"><a href="#cb22-11" aria-hidden="true" tabindex="-1"></a><span class="co"> */</span></span>
+<span id="cb22-12"><a href="#cb22-12" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> predictors_rng(<span class="dt">int</span> N, <span class="dt">int</span> K) {</span>
+<span id="cb22-13"><a href="#cb22-13" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The comment begins with <code>/**</code>, ends with <code>*/</code>, and has an asterisk (<code>*</code>) on each line. It uses <code>@param</code> followed by the argument’s identifier to document a function argument. The tag <code>@return</code> is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with <code>/*</code> and ending with <code>*/</code> to the Stan parser.</p>
+<p>For functions that raise exceptions, exceptions can be documented using <code>@throws</code>.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
+<p>For example,</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a> <span class="co">/** ...</span></span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a><span class="co"> * </span><span class="an">@param</span><span class="co"> theta</span></span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a><span class="co"> * @throws If any of the entries of theta is negative.</span></span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a><span class="co"> */</span></span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> entropy(<span class="dt">vector</span> theta) {</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Usually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.</p>
+</section>
+<section id="summary-of-function-types" class="level2">
+<h2 class="anchored" data-anchor-id="summary-of-function-types">Summary of function types</h2>
+<p>Functions may have a void or non-void return type and they may or may not have one of the special suffixes, <code>_lpdf</code>, <code>_lpmf</code>, <code>_lp</code>, or <code>_rng</code>.</p>
+<section id="void-vs.-non-void-return" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="void-vs.-non-void-return">Void vs.&nbsp;non-void return</h3>
+<p>Only functions declared to return <code>void</code> may be used as statements. These are also the only functions that use <code>return</code> statements with no arguments.</p>
+<p>Only functions declared to return non-<code>void</code> values may be used as expressions. These functions require <code>return</code> statements with arguments of a type that matches the declared return type.</p>
+</section>
+<section id="suffixed-or-non-suffixed" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="suffixed-or-non-suffixed">Suffixed or non-suffixed</h3>
+<p>Only functions ending in <code>_lpmf</code> or <code>_lpdf</code> and with return type <code>real</code> may be used as probability functions in distribution statements.</p>
+<p>Only functions ending in <code>_lp</code> may access the log probability accumulator through distribution statements or <code>target +=</code> statements. Such functions may only be used in the transformed parameters or model blocks.</p>
+<p>Only functions ending in <code>_rng</code> may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in <code>_rng</code>.</p>
+</section>
+</section>
+<section id="recursive-functions" class="level2">
+<h2 class="anchored" data-anchor-id="recursive-functions">Recursive functions</h2>
+<p>Stan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, <span class="math inline">\(A^n\)</span>, which is defined for a square matrix <span class="math inline">\(A\)</span> and positive integer <span class="math inline">\(n\)</span> by <span class="math display">\[
+A^n
+=
+\begin{cases}
+\textrm{I} &amp; \quad\text{if } n = 0, \text{ and} \\
+A \, A^{n-1} &amp; \quad\text{if } n &gt; 0.
+\end{cases}
+\]</span></p>
+<p>where <span class="math inline">\(\textrm{I}\)</span> is the identity matrix. This definition can be directly translated to a recursive function definition.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> matrix_pow(<span class="dt">matrix</span> a, <span class="dt">int</span> n) {</span>
+<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (n == <span class="dv">0</span>) {</span>
+<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> diag_matrix(rep_vector(<span class="dv">1</span>, rows(a)));</span>
+<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> a *  matrix_pow(a, n - <span class="dv">1</span>);</span>
+<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>It would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="cf">else</span> <span class="cf">if</span> (n == <span class="dv">1</span>) {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> a;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="truncated-random-number-generation" class="level2">
+<h2 class="anchored" data-anchor-id="truncated-random-number-generation">Truncated random number generation</h2>
+<section id="generation-with-inverse-cdfs" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="generation-with-inverse-cdfs">Generation with inverse CDFs</h3>
+<p>To generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate <span class="math inline">\(u \sim \textsf{uniform}(0, 1)\)</span>, then run through the inverse cumulative distribution function, <span class="math inline">\(y = \textrm{logit}(u)\)</span>. If this were not already built in as <code>logistic_rng(0, 1)</code>, it could be coded in Stan directly as</p>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> standard_logistic_rng() {</span>
+<span id="cb26-2"><a href="#cb26-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> u = uniform_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb26-3"><a href="#cb26-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y = logit(u);</span>
+<span id="cb26-4"><a href="#cb26-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> y;</span>
+<span id="cb26-5"><a href="#cb26-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Following the same pattern, a standard normal RNG could be coded as</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> standard_normal_rng() {</span>
+<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> u = uniform_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y = inv_Phi(u);</span>
+<span id="cb27-4"><a href="#cb27-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> y;</span>
+<span id="cb27-5"><a href="#cb27-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>that is, <span class="math inline">\(y = \Phi^{-1}(u)\)</span>, where <span class="math inline">\(\Phi^{-1}\)</span> is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function <code>inv_Phi</code>.</p>
+<p>In order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate <span class="math inline">\(\textsf{normal}(\mu, \sigma)\)</span> variates, it is enough to generate a uniform variate <span class="math inline">\(u \sim \textsf{uniform}(0, 1)\)</span>, then convert it to a standard normal variate, <span class="math inline">\(z = \Phi^{-1}(u)\)</span>, where <span class="math inline">\(\Phi^{-1}(\cdot)\)</span> is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, <span class="math inline">\(y = \mu +
+\sigma \times z\)</span>. In code,</p>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> my_normal_rng(<span class="dt">real</span> mu, <span class="dt">real</span> sigma) {</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> u = uniform_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> z = inv_Phi(u);</span>
+<span id="cb28-4"><a href="#cb28-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y = mu + sigma * z;</span>
+<span id="cb28-5"><a href="#cb28-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> y;</span>
+<span id="cb28-6"><a href="#cb28-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>A robust version of this function would test that the arguments are finite and that <code>sigma</code> is non-negative, e.g.,</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (is_nan(mu) || is_inf(mu)) {</span>
+<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a>    <span class="kw">reject</span>(<span class="st">"my_normal_rng: mu must be finite; "</span>,</span>
+<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>           <span class="st">"found mu = "</span>, mu);</span>
+<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (is_nan(sigma) || is_inf(sigma) || sigma &lt; <span class="dv">0</span>) {</span>
+<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a>    <span class="kw">reject</span>(<span class="st">"my_normal_rng: sigma must be finite and non-negative; "</span>,</span>
+<span id="cb29-7"><a href="#cb29-7" aria-hidden="true" tabindex="-1"></a>           <span class="st">"found sigma = "</span>, sigma);</span>
+<span id="cb29-8"><a href="#cb29-8" aria-hidden="true" tabindex="-1"></a>  }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="truncated-variate-generation" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="truncated-variate-generation">Truncated variate generation</h3>
+<p>Often truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.</p>
+<section id="truncating-below" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncating-below">Truncating below</h4>
+<p>For example, the following code generates a <span class="math inline">\(\textsf{Weibull}(\alpha, \sigma)\)</span> variate truncated below at a time <span class="math inline">\(t\)</span>,<a href="#fn5" class="footnote-ref" id="fnref5" role="doc-noteref"><sup>5</sup></a></p>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> weibull_lb_rng(<span class="dt">real</span> alpha, <span class="dt">real</span> sigma, <span class="dt">real</span> t) {</span>
+<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> p = weibull_cdf(t | alpha, sigma);   <span class="co">// cdf for lb</span></span>
+<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> u = uniform_rng(p, <span class="dv">1</span>);               <span class="co">// unif in bounds</span></span>
+<span id="cb30-4"><a href="#cb30-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y = sigma * (-log1m(u))^inv(alpha);  <span class="co">// inverse cdf</span></span>
+<span id="cb30-5"><a href="#cb30-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> y;</span>
+<span id="cb30-6"><a href="#cb30-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="truncating-above-and-below" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="truncating-above-and-below">Truncating above and below</h4>
+<p>If there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a <span class="math inline">\(\textsf{normal}(\mu, \sigma)\)</span> truncated to a region <span class="math inline">\((a, b)\)</span>, the following code suffices,</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> normal_lub_rng(<span class="dt">real</span> mu, <span class="dt">real</span> sigma, <span class="dt">real</span> lb, <span class="dt">real</span> ub) {</span>
+<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> p_lb = normal_cdf(lb | mu, sigma);</span>
+<span id="cb31-3"><a href="#cb31-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> p_ub = normal_cdf(ub | mu, sigma);</span>
+<span id="cb31-4"><a href="#cb31-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> u = uniform_rng(p_lb, p_ub);</span>
+<span id="cb31-5"><a href="#cb31-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y = mu + sigma * inv_Phi(u);</span>
+<span id="cb31-6"><a href="#cb31-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> y;</span>
+<span id="cb31-7"><a href="#cb31-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>To make this more robust, all variables should be tested for finiteness, <code>sigma</code> should be tested for positiveness, and <code>lb</code> and <code>ub</code> should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a>  <span class="cf">return</span> mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),</span>
+<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a>                                          normal_cdf(ub | mu, sigma)));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>for readability. The names like <code>p</code> indicate probabilities, and <code>p_lb</code> and <code>p_ub</code> indicate the probabilities of the bounds. The variable <code>u</code> is clearly named as a uniform variate, and <code>y</code> is used to denote the variate being generated itself.</p>
+
+
+</section>
+</section>
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>The main problem with comments is that they can be misleading, either due to misunderstandings on the programmer’s part or because the program’s behavior is modified after the comment is written. The program always behaves the way the code is written, which is why refactoring complex code into understandable units is preferable to simply adding comments.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Just because this makes it possible to code a rejection sampler does not make it a good idea. Rejections break differentiability and the smooth exploration of the posterior. In Hamiltonian Monte Carlo, it can cause the sampler to be reduced to a diffusive random walk.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn3"><p>A range of built-in validation routines is coming to Stan soon! Alternatively, the <code>reject</code> statement can be used to check constraints on the simplex.<a href="#fnref3" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn4"><p>As of Stan 2.9.0, the only way a user-defined producer will raise an exception is if a function it calls (including distribution statements) raises an exception via the reject statement.<a href="#fnref4" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn5"><p>The original code and impetus for including this in the manual came from the <a href="http://discourse.mc-stan.org/t/rng-for-truncated-distributions/3122/7">Stan forums post</a>; by user <code>lcomm</code>, who also explained truncation above and below.<a href="#fnref5" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/multi-indexing.html" class="pagination-link" aria-label="Multiple Indexing and Range Indexing">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Multiple Indexing and Range Indexing</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/custom-probability.html" class="pagination-link" aria-label="Custom Probability Functions">
+        <span class="nav-page-text">Custom Probability Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/using-stanc.html b/docs/2_39/stan-users-guide/using-stanc.html
new file mode 100644
index 000000000..27987f747
--- /dev/null
+++ b/docs/2_39/stan-users-guide/using-stanc.html
@@ -0,0 +1,2048 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Using the Stan Compiler</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/style-guide.html" rel="next">
+<link href="../stan-users-guide/multiple-imputation.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Using the Stan Compiler">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Using the Stan Compiler</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#using-the-stan-compiler" id="toc-using-the-stan-compiler" class="nav-link active" data-scroll-target="#using-the-stan-compiler">Using the Stan Compiler</a>
+  <ul class="collapse">
+  <li><a href="#stanc-args" id="toc-stanc-args" class="nav-link" data-scroll-target="#stanc-args">Command-line options for stanc3</a></li>
+  <li><a href="#understanding-stanc3-errors-and-warnings" id="toc-understanding-stanc3-errors-and-warnings" class="nav-link" data-scroll-target="#understanding-stanc3-errors-and-warnings">Understanding stanc3 errors and warnings</a>
+  <ul class="collapse">
+  <li><a href="#warnings" id="toc-warnings" class="nav-link" data-scroll-target="#warnings">Warnings</a></li>
+  <li><a href="#errors" id="toc-errors" class="nav-link" data-scroll-target="#errors">Errors</a></li>
+  </ul></li>
+  <li><a href="#pedantic-mode" id="toc-pedantic-mode" class="nav-link" data-scroll-target="#pedantic-mode">Pedantic mode</a>
+  <ul class="collapse">
+  <li><a href="#distribution-argument-and-variate-constraint-issues" id="toc-distribution-argument-and-variate-constraint-issues" class="nav-link" data-scroll-target="#distribution-argument-and-variate-constraint-issues">Distribution argument and variate constraint issues</a></li>
+  <li><a href="#special-case-distribution-issues" id="toc-special-case-distribution-issues" class="nav-link" data-scroll-target="#special-case-distribution-issues">Special-case distribution issues</a></li>
+  <li><a href="#unused-parameters" id="toc-unused-parameters" class="nav-link" data-scroll-target="#unused-parameters">Unused parameters</a></li>
+  <li><a href="#large-or-small-constants-in-a-distribution" id="toc-large-or-small-constants-in-a-distribution" class="nav-link" data-scroll-target="#large-or-small-constants-in-a-distribution">Large or small constants in a distribution</a></li>
+  <li><a href="#control-flow-depends-on-a-parameter" id="toc-control-flow-depends-on-a-parameter" class="nav-link" data-scroll-target="#control-flow-depends-on-a-parameter">Control flow depends on a parameter</a></li>
+  <li><a href="#parameters-with-multiple-tildes" id="toc-parameters-with-multiple-tildes" class="nav-link" data-scroll-target="#parameters-with-multiple-tildes">Parameters with multiple tildes</a></li>
+  <li><a href="#parameters-with-zero-or-multiple-priors" id="toc-parameters-with-zero-or-multiple-priors" class="nav-link" data-scroll-target="#parameters-with-zero-or-multiple-priors">Parameters with zero or multiple priors</a></li>
+  <li><a href="#variables-used-before-assignment" id="toc-variables-used-before-assignment" class="nav-link" data-scroll-target="#variables-used-before-assignment">Variables used before assignment</a></li>
+  <li><a href="#strict-or-nonsensical-parameter-bounds" id="toc-strict-or-nonsensical-parameter-bounds" class="nav-link" data-scroll-target="#strict-or-nonsensical-parameter-bounds">Strict or nonsensical parameter bounds</a></li>
+  <li><a href="#nonlinear-transformations" id="toc-nonlinear-transformations" class="nav-link" data-scroll-target="#nonlinear-transformations">Nonlinear transformations</a></li>
+  <li><a href="#pedantic-mode-limitations" id="toc-pedantic-mode-limitations" class="nav-link" data-scroll-target="#pedantic-mode-limitations">Pedantic mode limitations</a></li>
+  </ul></li>
+  <li><a href="#stanc-pretty-printing" id="toc-stanc-pretty-printing" class="nav-link" data-scroll-target="#stanc-pretty-printing">Automatic updating and formatting of Stan programs</a>
+  <ul class="collapse">
+  <li><a href="#automatic-formatting" id="toc-automatic-formatting" class="nav-link" data-scroll-target="#automatic-formatting">Automatic formatting</a></li>
+  <li><a href="#canonicalizing" id="toc-canonicalizing" class="nav-link" data-scroll-target="#canonicalizing">Canonicalizing</a></li>
+  <li><a href="#known-issues" id="toc-known-issues" class="nav-link" data-scroll-target="#known-issues">Known issues</a></li>
+  </ul></li>
+  <li><a href="#optimization" id="toc-optimization" class="nav-link" data-scroll-target="#optimization">Optimization</a>
+  <ul class="collapse">
+  <li><a href="#optimization-levels" id="toc-optimization-levels" class="nav-link" data-scroll-target="#optimization-levels">Optimization levels</a></li>
+  <li><a href="#o1-optimizations" id="toc-o1-optimizations" class="nav-link" data-scroll-target="#o1-optimizations"><strong>O1</strong> Optimizations</a></li>
+  <li><a href="#experimental-optimizations" id="toc-experimental-optimizations" class="nav-link" data-scroll-target="#experimental-optimizations"><strong>0experimental</strong> Optimizations</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/using-stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Appendices</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/using-stanc.html">Using the Stan Compiler</a></li></ol></nav></header>
+
+
+
+
+<section id="using-the-stan-compiler" class="level1">
+<h1>Using the Stan Compiler</h1>
+<p>Stan is used in most of our <a href="https://mc-stan.org/users/interfaces/">interfaces</a> through the Stan compiler <code>stanc</code>. Since version 2.22, the Stan compiler has been implemented in OCaml and is referred to as <a href="https://github.com/stan-dev/stanc3">stanc3</a>. The binary name is still simply <code>stanc</code>, so this document uses both stanc and stanc3 interchangeably.</p>
+<section id="stanc-args" class="level2">
+<h2 class="anchored" data-anchor-id="stanc-args">Command-line options for stanc3</h2>
+<p>The stanc3 compiler has the following command-line syntax:</p>
+<pre><code>&gt; stanc (options) &lt;model_file&gt;</code></pre>
+<p>where <code>&lt;model_file&gt;</code> is either a path to a file ending in <code>.stan</code> or <code>.stanfunctions</code> (which automatically sets <code>--standalone-functions</code>), or <code>'-'</code> to read from standard input.</p>
+<p>The stanc3 options are:</p>
+<ul>
+<li><p><code>--help</code> / <code>-?</code> - Displays the complete list of stanc3 options, then exits.</p></li>
+<li><p><code>--version</code> - Display stanc version number</p></li>
+<li><p><code>--info</code> - Print information about the model, such as the type information for variables and the list of used distributions.</p></li>
+<li><p><code>--name=&lt;model_name&gt;</code> - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.</p></li>
+<li><p><code>--o=&lt;file_name&gt;</code> / <code>-o=&lt;filename&gt;</code> / <code>--output=&lt;filename&gt;</code> - Specify a path to an output file for generated C++ code (default = <model_name>.hpp) or auto-formatting output (default: no file/print to stdout)</model_name></p></li>
+<li><p><code>--auto-format</code> - Pretty prints the program to the console. See <a href="#stanc-pretty-printing">more on auto formatting</a>.</p></li>
+<li><p><code>--allow-undefined</code> - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.</p></li>
+<li><p><code>--canonicalize</code> - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.</p></li>
+<li><p><code>--include_paths=&lt;dir1,...dirN&gt;</code> - Takes a comma-separated list of directories that may contain a file in an <code>#include</code> directive.</p></li>
+<li><p><code>--max-line-length=&lt;number&gt;</code> - Set the column number at which formatting with <code>--auto-format</code> attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.</p></li>
+<li><p><code>--print-canonical</code> - Synonymous with <code>--auto-format --canonicalize=deprecations,includes,parentheses,braces</code>.</p></li>
+<li><p><code>--print-cpp</code> - If set, output the generated C++ Stan model class to stdout.</p></li>
+<li><p><code>--filename-in-msg=&lt;name&gt;</code> - Sets the filename used in compiler and runtime errors. If absent, the <code>&lt;model_file&gt;</code> argument is used.</p></li>
+<li><p><code>--standalone-functions</code> - If set, only generate the code for the functions defined in the file. This is the default behavior for <code>.stanfunctions</code> files.</p></li>
+<li><p><code>--O0</code> (Default) Do not apply optimizations to the Stan code.</p></li>
+<li><p><code>--O1</code> Apply level 1 compiler <a href="#optimization">optimizations</a> (only basic optimizations).</p></li>
+<li><p><code>--Oexperimental</code> <strong>WARNING:</strong> <em>This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance!</em> Allow the compiler to apply all <a href="#optimization">optimizations</a> to the Stan code.</p></li>
+<li><p><code>--O</code> Synonym for <code>--O1</code> as of Stan 2.37. In earlier versions this was a synonym for <code>--Oexperimental</code>.</p></li>
+<li><p><code>--use-opencl</code> - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.</p></li>
+<li><p><code>--warn-pedantic</code> - Emit warnings in <a href="#pedantic-mode">Pedantic mode</a> which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.</p></li>
+<li><p><code>--warn-uninitialized</code> - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.</p></li>
+<li><p><code>--color</code> - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are <code>auto</code> (the default), <code>always</code>, <code>never</code>. Can also be controlled by the <code>STANC_COLOR</code> environment variable.</p></li>
+</ul>
+<p>The compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the <code>--help</code> option to see the full set.</p>
+</section>
+<section id="understanding-stanc3-errors-and-warnings" class="level2">
+<h2 class="anchored" data-anchor-id="understanding-stanc3-errors-and-warnings">Understanding stanc3 errors and warnings</h2>
+<p>During model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).</p>
+<section id="warnings" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="warnings">Warnings</h3>
+<p>Even without the optional <code>--warn-pedantic</code> and <code>--warn-uninitialized</code> <a href="#stanc-args">command line flags</a>, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations</p>
+<ol type="1">
+<li><p>A completely blank Stan program will produce the following warning message</p>
+<pre><code>Warning in 'empty.stan', line 1, column 0 to line 2, column 0:
+    Empty model detected; this is a valid Stan model but likely unintended!</code></pre></li>
+<li><p>The use of any <a href="https://mc-stan.org/docs/reference-manual/deprecations.html">deprecated features</a> will lead to warnings which will look as follows</p>
+<pre><code> Warning in 'deprecated.stan', line 2, column 10 to column 17:
+     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with
+     an independent lognormal distribution on the scales, see:
+     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution</code></pre>
+<p>A single Stan program can produce many warnings during compilation.</p></li>
+</ol>
+</section>
+<section id="errors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="errors">Errors</h3>
+<p>Errors differ from warnings in their severity and format. In particular, errors are <em>fatal</em> and stop compilation, so at most one error is displayed per run of stanc.</p>
+<p>There are five kinds of errors emitted by stanc3</p>
+<ol type="1">
+<li><p>File errors occur when the file passed to stanc is either missing or cannot be opened (i.e.&nbsp;has permissions issues). They look like</p>
+<pre><code>Error: file 'notfound.stan' not found or cannot be opened</code></pre></li>
+<li><p>Syntactic errors occur whenever a program violates the Stan language’s <a href="https://mc-stan.org/docs/reference-manual/syntax.html">syntax</a> requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the <code>#include</code> directive fails, and “parsing” errors which result when the structure of the program is incorrect.</p>
+<ul>
+<li><p>The lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of <code>$</code> in a variable name will look like the following.</p>
+<pre><code>Syntax error in 'char.stan', line 2, column 7 to column 8, lexing error:
+-------------------------------------------------
+  1:  data {
+  2:     int $ome_variable;
+             ^
+  3:  }
+-------------------------------------------------
+Invalid character found.</code></pre></li>
+<li><p>When an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).</p>
+<pre><code>Syntax error in './incl.stan', line 1, column 0, included from
+'./incl.stan', line 1, column 0, included from
+'incl.stan', line 1, column 0, include error:
+   -------------------------------------------------
+     1:  #include &lt;incl.stan&gt;
+         ^
+   -------------------------------------------------
+File incl.stan recursively included itself.</code></pre></li>
+<li><p>It is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.</p>
+<pre><code>Syntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:
+   -------------------------------------------------
+     1:  data {
+     2:     int&lt;lower=0&gt; N;
+     3:     vector x;
+                   ^
+     4:  }
+   -------------------------------------------------
+Ill-formed type. Expected "[" expression "]" for vector size.</code></pre></li>
+</ul></li>
+<li><p>Semantic errors (also known as type errors) occur when a program is structured correctly but features an error in the <a href="https://mc-stan.org/docs/reference-manual/syntax.html#extra-grammatical-constraints">type rules</a> imposed by the language. An example of this is assigning a real value to a variable defined as an integer.</p>
+<pre><code>Semantic error in 'type.stan', line 2, column 3 to column 15:
+   -------------------------------------------------
+     1:  transformed data {
+     2:     int x = 1.5;
+            ^
+     3:  }
+   -------------------------------------------------
+Ill-typed arguments supplied to assignment operator =:
+The left hand side has type
+  int
+and the right hand side has type
+  real</code></pre></li>
+<li><p>The compiler will raise an error for use of any <a href="https://mc-stan.org/docs/reference-manual/removals.html">removed features</a> for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.</p></li>
+<li><p>Finally, the compiler can raise an internal error. These are caused by bugs in the compiler, <strong>not</strong> your model, and we would appreciate it if you report them on the <a href="https://github.com/stan-dev/stanc3/issues">stanc3 repo</a> with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.</p></li>
+</ol>
+</section>
+</section>
+<section id="pedantic-mode" class="level2">
+<h2 class="anchored" data-anchor-id="pedantic-mode">Pedantic mode</h2>
+<p>Pedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> sigma;</span>
+<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu;</span>
+<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a>  x ~ normal(mu, sigma);</span>
+<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When pedantic mode is turned on, the compiler will produce the following warnings.</p>
+<pre><code>Warning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:
+    The parameter sigma has no priors. This means either no prior is
+    provided, or the prior(s) depend on data variables. In the later case,
+    this may be a false positive.
+Warning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:
+    The variable mu may not have been assigned a value before its use.
+Warning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:
+    A normal distribution is given parameter sigma as a scale parameter
+    (argument 2), but sigma was not constrained to be strictly positive.</code></pre>
+<p>Here are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):</p>
+<ul>
+<li><em>Distribution usages issues.</em> Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.</li>
+<li><em>Unused parameter.</em> A parameter is defined but doesn’t contribute to target.</li>
+<li><em>Large or small constant in a distribution.</em> Very large or very small constants are used as distribution arguments.</li>
+<li><em>Control flow depends on a parameter.</em> Branching control flow (like if/else) depends on a parameter value .</li>
+<li><em>Parameter has multiple tildes.</em> A parameter is on the left-hand side of multiple tildes.</li>
+<li><em>Parameter has zero or multiple priors.</em> A parameter has zero or more than one prior distribution.</li>
+<li><em>Variable is used before assignment.</em> A variable is used before being assigned a value.</li>
+<li><em>Strict or nonsensical parameter bounds.</em> A parameter is given questionable bounds.</li>
+<li><em>Nonlinear transformations.</em> When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.</li>
+</ul>
+<p>Some important limitations of pedantic mode are listed at the end of this chapter.</p>
+<section id="distribution-argument-and-variate-constraint-issues" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="distribution-argument-and-variate-constraint-issues">Distribution argument and variate constraint issues</h3>
+<p>When an argument to a built-in distribution certainly does not match that distribution’s specification in the <a href="https://mc-stan.org/docs/functions-reference/index.html">Stan Functions Reference</a>, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g.&nbsp;<code>simplex</code>) is incompatible with the domain of the distribution. x</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> unb_p;</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; pos_p;</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>  <span class="dv">1</span> ~ poisson(unb_p);</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>  <span class="dv">1</span> ~ poisson(pos_p);</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The parameter of <code>poisson</code> should be strictly positive, but <code>unb_p</code> is not constrained to be positive.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'ex-dist-args.stan', line 6, column 14 to column 19:
+  A poisson distribution is given parameter unb_p as a rate parameter
+  (argument 1), but unb_p was not constrained to be strictly positive.</code></pre>
+</section>
+<section id="special-case-distribution-issues" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="special-case-distribution-issues">Special-case distribution issues</h3>
+<p>Pedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:</p>
+<section id="uniform-distributions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="uniform-distributions">Uniform distributions</h4>
+<p>Any use of uniform distribution generates a warning, except when the variate parameter’s declared <code>upper</code> and <code>lower</code> bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; b;</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>  a ~ uniform(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a>  b ~ uniform(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><code>a</code> is assigned a uniform distribution that doesn’t match its constraints.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'uniform-warn.stan', line 6, column 2 to column 20:
+  Parameter a is given a uniform distribution. The uniform distribution is
+  not recommended, for two reasons: (a) Except when there are logical or
+  physical constraints, it is very unusual for you to be sure that a
+  parameter will fall inside a specified range, and (b) The infinite gradient
+  induced by a uniform density can cause difficulties for Stan's sampling
+  algorithm. As a consequence, we recommend soft constraints rather than hard
+  constraints; for example, instead of giving an elasticity parameter a
+  uniform(0, 1) distribution, try normal(0.5, 0.5).</code></pre>
+</section>
+<section id="inverse--gamma-distributions" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="inverse--gamma-distributions">(Inverse-) Gamma distributions</h4>
+<p>Gamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.</p>
+</section>
+<section id="lkj_corr-distribution" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="lkj_corr-distribution">lkj_corr distribution</h4>
+<p>Any use of the <code>lkj_corr</code> distribution generates a warning that suggests using the Cholesky variant instead. See <a href="https://mc-stan.org/docs/functions-reference/correlation_matrix_distributions.html#lkj-correlation">the LKJ correlation distribution section</a> of the Stan Functions Reference for details.</p>
+</section>
+</section>
+<section id="unused-parameters" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="unused-parameters">Unused parameters</h3>
+<p>A warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the <code>target</code> variable depends in any way on each of the parameters.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p><code>a</code> participates in the density function but <code>b</code> does not.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'unused.stan', line 3, column 2 to column 9:
+    The parameter b was declared but was not used in the density calculation.</code></pre>
+</section>
+<section id="large-or-small-constants-in-a-distribution" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="large-or-small-constants-in-a-distribution">Large or small constants in a distribution</h3>
+<p>When numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See <a href="https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors-and-outputs">the efficiency tuning section</a> of the Stan User’s guide for a discussion of scaling parameters.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a>  x ~ normal(-<span class="dv">100</span>, <span class="dv">100</span>);</span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>  y ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The constants <code>-100</code> and <code>100</code> suggest that <code>x</code> is not unit scaled.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'constants-warn.stan', line 6, column 13 to column 17:
+    Argument -100 suggests there may be parameters that are not unit scale;
+    consider rescaling with a multiplier, see:
+    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors
+Warning in 'constants-warn.stan', line 6, column 19 to column 22:
+    Argument 100 suggests there may be parameters that are not unit scale;
+    consider rescaling with a multiplier, see:
+    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors</code></pre>
+</section>
+<section id="control-flow-depends-on-a-parameter" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="control-flow-depends-on-a-parameter">Control flow depends on a parameter</h3>
+<p>Control flow statements, such as <code>if</code>, <code>for</code> and <code>while</code> should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>  <span class="co">// x depends on parameter a</span></span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x = a * a;</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> m;</span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a>  <span class="co">// the if-then-else depends on x which depends on a</span></span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span>(x &gt; <span class="dv">0</span>) {</span>
+<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a>    <span class="co">//now m depends on x which depends on a</span></span>
+<span id="cb19-13"><a href="#cb19-13" aria-hidden="true" tabindex="-1"></a>    m = <span class="dv">1</span>;</span>
+<span id="cb19-14"><a href="#cb19-14" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb19-15"><a href="#cb19-15" aria-hidden="true" tabindex="-1"></a>    m = <span class="dv">2</span>;</span>
+<span id="cb19-16"><a href="#cb19-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb19-17"><a href="#cb19-17" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb19-18"><a href="#cb19-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// for loop depends on m -&gt; x -&gt; a</span></span>
+<span id="cb19-19"><a href="#cb19-19" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">0</span>:m) {</span>
+<span id="cb19-20"><a href="#cb19-20" aria-hidden="true" tabindex="-1"></a>    a ~ normal(i, <span class="dv">1</span>);</span>
+<span id="cb19-21"><a href="#cb19-21" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb19-22"><a href="#cb19-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>if</code> and <code>for</code> statements are control flow that depend (indirectly) on the value of the parameter <code>m</code>.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:
+  A control flow statement depends on parameter(s): a.
+Warning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:
+  A control flow statement depends on parameter(s): a.</code></pre>
+</section>
+<section id="parameters-with-multiple-tildes" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="parameters-with-multiple-tildes">Parameters with multiple tildes</h3>
+<p>A warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent <code>target +=</code> conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.</p>
+<p>Pedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b;</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a>  a ~ normal(x, <span class="dv">1</span>);</span>
+<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a>  b ~ normal(<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'multi-tildes.stan', line 9, column 2 to column 19:
+  The parameter a is on the left-hand side of more than one tildes
+  statement.</code></pre>
+</section>
+<section id="parameters-with-zero-or-multiple-priors" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="parameters-with-zero-or-multiple-priors">Parameters with zero or multiple priors</h3>
+<p>A warning is generated when a parameter appears to have greater than or less than one prior distribution factor.</p>
+<p>This analysis depends on a <a href="https://en.wikipedia.org/wiki/Factor_graph"><em>factor graph</em></a> representation of a Stan program. A factor F that depends on a parameter P is called a <em>prior factor for P</em> if there is no path in the factor graph from F to any data variable except through P.</p>
+<p>One limitation of this approach is that the compiler cannot distinguish between <em>modeled</em> <code>data</code> variables and other convenient uses of <code>data</code> variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for <code>int</code> variables) are modeled data, which may cause extra warnings.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b;</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> c;</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> d;</span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span></span>
+<span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">0</span>, <span class="dv">1</span>); <span class="co">// this is a prior</span></span>
+<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  x ~ normal(a, <span class="dv">1</span>); <span class="co">// this is not a prior, since data is involved</span></span>
+<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-15"><a href="#cb23-15" aria-hidden="true" tabindex="-1"></a>  b ~ normal(x, <span class="dv">1</span>); <span class="co">// this is also not a prior, since data is involved</span></span>
+<span id="cb23-16"><a href="#cb23-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-17"><a href="#cb23-17" aria-hidden="true" tabindex="-1"></a>  <span class="co">// this is not a prior for c, since data is involved through b</span></span>
+<span id="cb23-18"><a href="#cb23-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// but it is a prior for b, since the data is only involved through b</span></span>
+<span id="cb23-19"><a href="#cb23-19" aria-hidden="true" tabindex="-1"></a>  c ~ normal(b, <span class="dv">1</span>);</span>
+<span id="cb23-20"><a href="#cb23-20" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-21"><a href="#cb23-21" aria-hidden="true" tabindex="-1"></a>  <span class="co">//these are multiple priors:</span></span>
+<span id="cb23-22"><a href="#cb23-22" aria-hidden="true" tabindex="-1"></a>  d ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb23-23"><a href="#cb23-23" aria-hidden="true" tabindex="-1"></a>  <span class="dv">1</span> ~ normal(d, <span class="dv">1</span>);</span>
+<span id="cb23-24"><a href="#cb23-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>One prior is found for <code>a</code> and for <code>b</code>, while <code>c</code> only has a factor that touches a <code>data</code> variable and <code>d</code> has multiple priors.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'priors.stan', line 7, column 2 to column 9:
+    The parameter c has no priors. This means either no prior is provided, or
+    the prior(s) depend on data variables. In the later case, this may be a
+    false positive.
+Warning in 'priors.stan', line 8, column 2 to column 9:
+    The parameter d has 2 priors.
+</code></pre>
+</section>
+<section id="variables-used-before-assignment" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="variables-used-before-assignment">Variables used before assignment</h3>
+<p>A warning is generated when any variable is used before it has been assigned a value.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (<span class="dv">1</span> &gt; <span class="dv">2</span>) {</span>
+<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a>    x = <span class="dv">1</span>;</span>
+<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a>    <span class="kw">print</span>(<span class="st">"oops"</span>);</span>
+<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a>  <span class="kw">print</span>(x);</span>
+<span id="cb25-9"><a href="#cb25-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Since <code>x</code> is only assigned in one of the branches of the <code>if</code> statement, it might get to <code>print(x)</code> without having been assigned to.</p>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'uninit-warn.stan', line 7, column 8 to column 9:
+  The variable x may not have been assigned a value before its use.</code></pre>
+</section>
+<section id="strict-or-nonsensical-parameter-bounds" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="strict-or-nonsensical-parameter-bounds">Strict or nonsensical parameter bounds</h3>
+<p>Except when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds <code>&lt;lower=.., upper=..&gt;</code> except for <code>&lt;lower=0, upper=1&gt;</code> or <code>&lt;lower=-1, upper=1&gt;</code>.</p>
+<p>In addition, a warning is generated when a parameter bound is found to have <code>lower &gt;= upper</code>.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; a;</span>
+<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">1</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; b;</span>
+<span id="cb27-4"><a href="#cb27-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=-<span class="dv">2</span>, <span class="kw">upper</span>=<span class="dv">1012</span>&gt; c;</span>
+<span id="cb27-5"><a href="#cb27-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb27-6"><a href="#cb27-6" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb27-7"><a href="#cb27-7" aria-hidden="true" tabindex="-1"></a>  c ~ normal(b, a);</span>
+<span id="cb27-8"><a href="#cb27-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'hard-constraint.stan', line 4, column 2 to column 31:
+    Your Stan program has a parameter c with a lower and upper bound in its
+    declaration. These hard constraints are not recommended, for two reasons:
+    (a) Except when there are logical or physical constraints, it is very
+    unusual for you to be sure that a parameter will fall inside a specified
+    range, and (b) The infinite gradient induced by a hard constraint can
+    cause difficulties for Stan's sampling algorithm. As a consequence, we
+    recommend soft constraints rather than hard constraints; for example,
+    instead of constraining an elasticity parameter to fall between 0, and 1,
+    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.</code></pre>
+</section>
+<section id="nonlinear-transformations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="nonlinear-transformations">Nonlinear transformations</h3>
+<p>When a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the <a href="../stan-users-guide/reparameterization.html">Changes of variables</a> section.</p>
+<p>This portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.</p>
+<p>For example, consider the following program.</p>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a>  log(y) ~ normal(<span class="dv">0</span>,<span class="dv">1</span>);</span>
+<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Pedantic mode produces the following warning.</p>
+<pre><code>Warning in 'jacobian.stan', line 5, column 2 to column 23:
+    Left-hand side of distribution statement (~) may contain a non-linear
+    transform of a parameter or local variable. If it does, you need to
+    include a target += statement with the log absolute determinant of the
+    Jacobian of the transform. You could also consider defining a transformed
+    parameter and using jacobian += in the transformed parameters block.</code></pre>
+</section>
+<section id="pedantic-mode-limitations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="pedantic-mode-limitations">Pedantic mode limitations</h3>
+<ul>
+<li><p>Constant values are sometimes uncomputable</p>
+<p>Pedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code <code>normal(x, 1 - 2)</code>, the expression <code>1 - 2</code> will be evaluated to <code>-1</code>, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.</p></li>
+<li><p>Container types</p>
+<p>Currently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).</p></li>
+<li><p>Data variables</p>
+<p>The declaration information for <code>data</code> variables is currently not considered, so using <code>data</code> as incompatible arguments to distributions may not generate the appropriate warnings.</p></li>
+<li><p>Control flow dependent on parameters in nested functions</p>
+<p>If a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.</p></li>
+</ul>
+</section>
+</section>
+<section id="stanc-pretty-printing" class="level2">
+<h2 class="anchored" data-anchor-id="stanc-pretty-printing">Automatic updating and formatting of Stan programs</h2>
+<p>In addition to compiling Stan programs, stanc3 features several <a href="#stanc-args">flags</a> which can be used to format Stan programs and update them to the most recent Stan syntax by removing any <a href="https://mc-stan.org/docs/reference-manual/deprecations.html">deprecation features</a> which can be automatically replaced.</p>
+<p>These flags work for both <code>.stan</code> model files and <code>.stanfunctions</code> function files. They can be combined with <code>--o</code> to redirect the formatted output to a new file.</p>
+<section id="automatic-formatting" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="automatic-formatting">Automatic formatting</h3>
+<p>Invoking <code>stanc --auto-format &lt;model_file&gt;</code> will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the <a href="../stan-users-guide/style-guide.html">Stan Program Style Guide</a>. This means spacing, indentation, and line length are all regularized. <em>Some</em> deprecated features, like the use of <code>#</code> for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.</p>
+<p>By default, this will try to split lines at or before column 78. This number can be changed using <code>--max-line-length</code>.</p>
+</section>
+<section id="canonicalizing" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="canonicalizing">Canonicalizing</h3>
+<p>In addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of <code>if</code> statements and <code>for</code> and <code>while</code> loops.</p>
+<p>This can be done by using <code>stanc --auto-format --canonicalize=...</code> where <code>...</code> is a comma-separated list of options. Currently these options are:</p>
+<ul>
+<li><p><code>deprecations</code></p>
+<p>Removes deprecated syntax such as replacing deprecated functions with their drop-in replacements.</p></li>
+<li><p><code>parentheses</code></p>
+<p>Removes unnecessary extra parentheses, such as converting <code>y = ((x-1))</code> to <code>y = x - 1</code></p></li>
+<li><p><code>braces</code></p>
+<p>Places braces around all blocks. For example, the following statement</p>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (cond)</span>
+<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">//result</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>will be formatted as</p>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> (cond) {</span>
+<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">//result</span></span>
+<span id="cb32-3"><a href="#cb32-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>and similarly for both kinds of loops containing a single statement.</p></li>
+<li><p><code>includes</code></p>
+<p>This will pretty-print code from other files included with <code>#include</code> as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same <code>#include</code> directives as the input program.</p></li>
+</ul>
+<p>Invoking <code>stanc --print-canonical &lt;model_file&gt;</code> is synonymous with running <code>stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes</code></p>
+</section>
+<section id="known-issues" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="known-issues">Known issues</h3>
+<p>The formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:</p>
+<ul>
+<li><p>Oddly placed comments</p>
+<p>If your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string <code>^^^:</code> to indicate they originally appeared higher in the program.</p>
+<p>We hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.</p></li>
+<li><p>Failure to recreate strange <code>#include</code> structure</p>
+<p>Printing <em>without</em> include inlining (<code>--canonicalize=includes</code>) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.</p></li>
+</ul>
+</section>
+</section>
+<section id="optimization" class="level2">
+<h2 class="anchored" data-anchor-id="optimization">Optimization</h2>
+<p>The stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.</p>
+<p>This section introduces the available optimization options and describes their effect.</p>
+<p>To print out a representation of the optimized Stan program, use the stanc3 command-line flag <code>--debug-optimized-mir-pretty</code>. To print an analogous representation of the Stan program prior to optimization, use the flag <code>--debug-transformed-mir-pretty</code>.</p>
+<section id="optimization-levels" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="optimization-levels">Optimization levels</h3>
+<p>To turn optimizations on, the user specifies the desired optimization <em>level</em>. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.</p>
+<p>Optimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:</p>
+<ul>
+<li><strong>O0</strong> No optimizations are applied.</li>
+<li><strong>O1</strong> Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.</li>
+<li><strong>Oexperimental</strong> All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.</li>
+</ul>
+<p>O0 is the default setting.</p>
+<p>The levels include these optimizations:</p>
+<ul>
+<li><strong>O0</strong> includes no optimizations.</li>
+<li><strong>O1</strong> includes:
+<ul>
+<li><a href="#dead-code-elimination">Dead code elimination</a></li>
+<li><a href="#copy-propagation">Copy propagation</a></li>
+<li><a href="#constant-propagation">Constant propagation</a></li>
+<li><a href="#partial-evaluation">Partial evaluation</a></li>
+<li><a href="#function-inlining">Function inlining</a></li>
+<li><a href="#memory-patterns">Matrix memory layout optimization</a></li>
+</ul></li>
+<li><strong>Oexperimental</strong> includes optimizations specified by <strong>O1</strong> and also:
+<ul>
+<li><a href="#automatic-differentiation-level-optimization">Automatic-differentiation level optimization</a></li>
+<li><a href="#one-step-loop-unrolling">One step loop unrolling</a></li>
+<li><a href="#expression-propagation">Expression propagation</a></li>
+<li><a href="#lazy-code-motion">Lazy code motion</a></li>
+<li><a href="#static-loop-unrolling">Static loop unrolling</a></li>
+</ul></li>
+</ul>
+<p>In addition, <strong>Oexperimental</strong> will apply more repetitions of the optimizations, which may increase compile times.</p>
+</section>
+<section id="o1-optimizations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="o1-optimizations"><strong>O1</strong> Optimizations</h3>
+<section id="dead-code-elimination" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="dead-code-elimination">Dead code elimination</h4>
+<p>Dead code is code that does not affect the behavior of the program. Code is not dead if it affects <code>target</code>, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb33-2"><a href="#cb33-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> i;</span>
+<span id="cb33-3"><a href="#cb33-3" aria-hidden="true" tabindex="-1"></a>  i = <span class="dv">5</span>;</span>
+<span id="cb33-4"><a href="#cb33-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">10</span>);</span>
+<span id="cb33-5"><a href="#cb33-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">if</span> (<span class="dv">0</span>) {</span>
+<span id="cb33-6"><a href="#cb33-6" aria-hidden="true" tabindex="-1"></a>    <span class="kw">print</span>(<span class="st">"Dead code"</span>);</span>
+<span id="cb33-7"><a href="#cb33-7" aria-hidden="true" tabindex="-1"></a>  } <span class="cf">else</span> {</span>
+<span id="cb33-8"><a href="#cb33-8" aria-hidden="true" tabindex="-1"></a>    <span class="kw">print</span>(<span class="st">"Hi!"</span>);</span>
+<span id="cb33-9"><a href="#cb33-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb33-10"><a href="#cb33-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before dead code elimination</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  int i = 5;
+  for(j in 1:10) {
+    ;
+  }
+  if(0) {
+    FnPrint__("Dead code");
+  } else {
+    FnPrint__("Hi!");
+  }
+}</code></pre>
+<p>Compiler representation of program <strong>after dead code elimination</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  int i;
+  FnPrint__("Hi!");
+}</code></pre>
+</section>
+<section id="constant-propagation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="constant-propagation">Constant propagation</h4>
+<p>Constant propagation replaces uses of a variable which is known to have a constant value <code>C</code> with that constant <code>C</code>. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb36-2"><a href="#cb36-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> n = <span class="dv">100</span>;</span>
+<span id="cb36-3"><a href="#cb36-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> a[n];</span>
+<span id="cb36-4"><a href="#cb36-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n) {</span>
+<span id="cb36-5"><a href="#cb36-5" aria-hidden="true" tabindex="-1"></a>    a[i] = i;</span>
+<span id="cb36-6"><a href="#cb36-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb36-7"><a href="#cb36-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before constant propagation</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int n = 100;
+  data array[int, n] a;
+  for(i in 1:n) {
+    a[i] = i;
+  }
+}</code></pre>
+<p>Compiler representation of program <strong>after constant propagation</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int n = 100;
+  data array[int, 100] a;
+  for(i in 1:100) {
+    a[i] = i;
+  }
+}</code></pre>
+</section>
+<section id="copy-propagation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="copy-propagation">Copy propagation</h4>
+<p>Copy propagation is similar to <a href="#expression-propagation">expression propagation</a>, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb39-2"><a href="#cb39-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> i = <span class="dv">1</span>;</span>
+<span id="cb39-3"><a href="#cb39-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> j = i;</span>
+<span id="cb39-4"><a href="#cb39-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> k = i + j;</span>
+<span id="cb39-5"><a href="#cb39-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before copy propagation</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>log_prob {
+    int i = 1;
+    int j = i;
+    int k = (i + j);
+}</code></pre>
+<p>Compiler representation of program <strong>after copy propagation</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  int i = 1;
+  int j = i;
+  int k = (i + i);
+}</code></pre>
+</section>
+<section id="partial-evaluation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="partial-evaluation">Partial evaluation</h4>
+<p>Partial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb42-2"><a href="#cb42-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a = <span class="dv">1</span> + <span class="dv">1</span>;</span>
+<span id="cb42-3"><a href="#cb42-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> b = log(<span class="dv">1</span> - a);</span>
+<span id="cb42-4"><a href="#cb42-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> c = a + b * <span class="dv">5</span>;</span>
+<span id="cb42-5"><a href="#cb42-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before partial evaluation</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  real a = (1 + 1);
+  real b = log((1 - a));
+  real c = (a + (b * 5));
+}</code></pre>
+<p>Compiler representation of program <strong>after partial evaluation</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  real a = 2;
+  real b = log1m(a);
+  real c = fma(b, 5, a);
+}</code></pre>
+</section>
+<section id="function-inlining" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="function-inlining">Function inlining</h4>
+<p>Function inlining replaces each function call to each user-defined function <code>f</code> with the body of <code>f</code>. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb45"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb45-1"><a href="#cb45-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb45-2"><a href="#cb45-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> incr(<span class="dt">int</span> x) {</span>
+<span id="cb45-3"><a href="#cb45-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> y = <span class="dv">1</span>;</span>
+<span id="cb45-4"><a href="#cb45-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> x + y;</span>
+<span id="cb45-5"><a href="#cb45-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb45-6"><a href="#cb45-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-7"><a href="#cb45-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb45-8"><a href="#cb45-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> a = <span class="dv">2</span>;</span>
+<span id="cb45-9"><a href="#cb45-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> b = incr(a);</span>
+<span id="cb45-10"><a href="#cb45-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before function inlining</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>functions {
+  int incr(int x) {
+    int y = 1;
+    return (x + y);
+  }
+}
+
+prepare_data {
+  data int a = 2;
+  data int b = incr(a);
+}</code></pre>
+<p>Compiler representation of program <strong>after function inlining</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int a;
+  a = 2;
+  data int b;
+  data int inline_sym1__;
+  data int inline_sym3__;
+  inline_sym3__ = 0;
+  for(inline_sym4__ in 1:1) {
+    int inline_sym2__;
+    inline_sym2__ = 1;
+    inline_sym3__ = 1;
+    inline_sym1__ = (a + inline_sym2__);
+    break;
+  }
+  b = inline_sym1__;
+}</code></pre>
+<p>In this code, the <code>for</code> loop and <code>break</code> is used to simulate the behavior of a <code>return</code> statement. The value to be returned is held in <code>inline_sym1__</code>. The flag variable <code>inline_sym3__</code> indicates whether a return has occurred and is necessary to handle <code>return</code> statements nested inside loops within the function body.</p>
+</section>
+<section id="memory-patterns" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="memory-patterns">Matrix memory layout optimization</h4>
+<p>Matrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.</p>
+<p>The first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.</p>
+<p>The second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.</p>
+<p><em>More general reading on AoS vs SoA can be found on <a href="https://en.wikipedia.org/wiki/AoS_and_SoA">Wikipedia</a></em></p>
+<p>This optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:</p>
+<ul>
+<li>Any Stan Math Library functions the matrix is passed to must be able to support it.</li>
+<li>The matrix should not be accessed/assigned elementwise in a loop.</li>
+</ul>
+<p>The debug flag <code>--debug-mem-patterns</code> will list each variable and whether it is using the AoS representation or the SoA representation.</p>
+</section>
+</section>
+<section id="experimental-optimizations" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="experimental-optimizations"><strong>0experimental</strong> Optimizations</h3>
+<section id="automatic-differentiation-level-optimization" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="automatic-differentiation-level-optimization">Automatic-differentiation level optimization</h4>
+<p>Stan variables can have two auto-differentiation (AD) <em>levels</em>: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb48"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb48-1"><a href="#cb48-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb48-2"><a href="#cb48-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb48-3"><a href="#cb48-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb48-4"><a href="#cb48-4" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb48-5"><a href="#cb48-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x = y + <span class="dv">1</span>;</span>
+<span id="cb48-6"><a href="#cb48-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before AD-level optimization</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>input_vars {
+  real y;
+}
+
+log_prob {
+  real x = (y + 1);
+}</code></pre>
+<p>Compiler representation of program <strong>after AD-level optimization</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>input_vars {
+  real y;
+}
+
+log_prob {
+  data real x = (y + 1);
+}</code></pre>
+</section>
+<section id="one-step-loop-unrolling" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="one-step-loop-unrolling">One step loop unrolling</h4>
+<p>One step loop unrolling is similar to <a href="#static-loop-unrolling">static loop unrolling</a>. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb51-2"><a href="#cb51-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> n;</span>
+<span id="cb51-3"><a href="#cb51-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb51-4"><a href="#cb51-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb51-5"><a href="#cb51-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> x = <span class="dv">0</span>;</span>
+<span id="cb51-6"><a href="#cb51-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n) {</span>
+<span id="cb51-7"><a href="#cb51-7" aria-hidden="true" tabindex="-1"></a>    x += i;</span>
+<span id="cb51-8"><a href="#cb51-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb51-9"><a href="#cb51-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before one step static loop unrolling</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int n = FnReadData__("n")[1];
+  data int x = 0;
+  for(i in 1:n) {
+    x = (x + i);
+  }
+}</code></pre>
+<p>Compiler representation of program <strong>after one step static loop unrolling</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int n = FnReadData__("n")[1];
+  int x = 0;
+  if((n &gt;= 1)) {
+    x = (x + 1);
+    for(i in (1 + 1):n) {
+      x = (x + i);
+    }
+  }
+}</code></pre>
+</section>
+<section id="expression-propagation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="expression-propagation">Expression propagation</h4>
+<p>Constant propagation replaces the uses of a variable which is known to have a constant value <code>E</code> with that constant <code>E</code>. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by <a href="#lazy-code-motion">lazy code motion</a> to avoid unnecessarily recomputing expressions.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb54"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb54-1"><a href="#cb54-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb54-2"><a href="#cb54-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> m;</span>
+<span id="cb54-3"><a href="#cb54-3" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb54-4"><a href="#cb54-4" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb54-5"><a href="#cb54-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> n = m+<span class="dv">1</span>;</span>
+<span id="cb54-6"><a href="#cb54-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> a[n];</span>
+<span id="cb54-7"><a href="#cb54-7" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n<span class="dv">-1</span>) {</span>
+<span id="cb54-8"><a href="#cb54-8" aria-hidden="true" tabindex="-1"></a>    a[i] = i;</span>
+<span id="cb54-9"><a href="#cb54-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb54-10"><a href="#cb54-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before expression propagation</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int m = FnReadData__("m")[1];
+  data int n = (m + 1);
+  data array[int, n] a;
+  for(i in 1:(n - 1)) {
+    a[i] = i;
+  }
+}</code></pre>
+<p>Compiler representation of program <strong>after expression propagation</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int m = FnReadData__("m")[1];
+  data int n = (m + 1);
+  data array[int, (m + 1)] a;
+  for(i in 1:((m + 1) - 1)) {
+    a[i] = i;
+  }
+}</code></pre>
+</section>
+<section id="lazy-code-motion" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="lazy-code-motion">Lazy code motion</h4>
+<p>Lazy code motion rearranges the statements and expressions in a program with the goals of:</p>
+<ul>
+<li>Avoiding computing expressions more than once, and</li>
+<li>Computing expressions as late as possible (to minimize the strain on the working memory set).</li>
+</ul>
+<p>To accomplish these goals, lazy code motion will perform optimizations such as:</p>
+<ul>
+<li>Moving a repeatedly calculated expression to its own variable (also referred to as <em>common-subexpression elimination</em>)</li>
+<li>Moving an expression outside of a loop if it does not need to be in the loop (also referred to as <em>loop-invariant code motion</em>)</li>
+</ul>
+<p>Lazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.</p>
+<p>As currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb57"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb57-1"><a href="#cb57-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb57-2"><a href="#cb57-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> x;</span>
+<span id="cb57-3"><a href="#cb57-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> y;</span>
+<span id="cb57-4"><a href="#cb57-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> z;</span>
+<span id="cb57-5"><a href="#cb57-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb57-6"><a href="#cb57-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">10</span>) {</span>
+<span id="cb57-7"><a href="#cb57-7" aria-hidden="true" tabindex="-1"></a>    x = sqrt(<span class="dv">10</span>);</span>
+<span id="cb57-8"><a href="#cb57-8" aria-hidden="true" tabindex="-1"></a>    y = sqrt(i);</span>
+<span id="cb57-9"><a href="#cb57-9" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb57-10"><a href="#cb57-10" aria-hidden="true" tabindex="-1"></a>  z = sqrt(<span class="dv">10</span>);</span>
+<span id="cb57-11"><a href="#cb57-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program before lazy code motion (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  real x;
+  real y;
+  real z;
+  for(i in 1:10) {
+    x = sqrt(10);
+    y = sqrt(i);
+  }
+  z = sqrt(10);
+}</code></pre>
+<p>Compiler representation of program after lazy code motion (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>log_prob {
+  data real lcm_sym4__;
+  data real lcm_sym3__;
+  real x;
+  real y;
+  lcm_sym4__ = sqrt(10);
+  real z;
+  for(i in 1:10) {
+    x = lcm_sym4__;
+    y = sqrt(i);
+  }
+  z = lcm_sym4__;
+}</code></pre>
+</section>
+<section id="static-loop-unrolling" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="static-loop-unrolling">Static loop unrolling</h4>
+<p>Static loop unrolling takes a loop with a predictable number of iterations <code>X</code> and replaces it by writing out the loop body <code>X</code> times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).</p>
+<p>Example Stan program:</p>
+<div class="sourceCode" id="cb60"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb60-1"><a href="#cb60-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb60-2"><a href="#cb60-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> x = <span class="dv">0</span>;</span>
+<span id="cb60-3"><a href="#cb60-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:<span class="dv">4</span>) {</span>
+<span id="cb60-4"><a href="#cb60-4" aria-hidden="true" tabindex="-1"></a>    x += i;</span>
+<span id="cb60-5"><a href="#cb60-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb60-6"><a href="#cb60-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Compiler representation of program <strong>before static loop unrolling</strong> (simplified from the output of <code>--debug-transformed-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int x = 0;
+  for(i in 1:4) {
+    x = (x + i);
+  }
+}</code></pre>
+<p>Compiler representation of program <strong>after static loop unrolling</strong> (simplified from the output of <code>--debug-optimized-mir-pretty</code>):</p>
+<pre><code>prepare_data {
+  data int x;
+  x = 0;
+  x = (x + 1);
+  x = (x + 2);
+  x = (x + 3);
+  x = (x + 4);
+}</code></pre>
+
+
+</section>
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/multiple-imputation.html" class="pagination-link" aria-label="Multiple imputation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Multiple imputation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/style-guide.html" class="pagination-link" aria-label="Stan Program Style Guide">
+        <span class="nav-page-text">Stan Program Style Guide</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/using-stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/wiener_diffusion_model.html b/docs/2_39/stan-users-guide/wiener_diffusion_model.html
new file mode 100644
index 000000000..b8cbc39e1
--- /dev/null
+++ b/docs/2_39/stan-users-guide/wiener_diffusion_model.html
@@ -0,0 +1,1598 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Wiener Diffusion Model</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/floating-point.html" rel="next">
+<link href="../stan-users-guide/survival.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Wiener Diffusion Model">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/Figure_DDM.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/wiener_diffusion_model.html">Wiener diffusion model</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#wiener-diffusion-model" id="toc-wiener-diffusion-model" class="nav-link active" data-scroll-target="#wiener-diffusion-model">Wiener diffusion model</a>
+  <ul class="collapse">
+  <li><a href="#function-call-example" id="toc-function-call-example" class="nav-link" data-scroll-target="#function-call-example">Function call example</a>
+  <ul class="collapse">
+  <li><a href="#the-data-block" id="toc-the-data-block" class="nav-link" data-scroll-target="#the-data-block">The data block</a></li>
+  <li><a href="#the-parameters-block" id="toc-the-parameters-block" class="nav-link" data-scroll-target="#the-parameters-block">The parameters block</a></li>
+  <li><a href="#the-model-block" id="toc-the-model-block" class="nav-link" data-scroll-target="#the-model-block">The model block</a></li>
+  </ul></li>
+  <li><a href="#wiener-truncation" id="toc-wiener-truncation" class="nav-link" data-scroll-target="#wiener-truncation">Truncated and censored data</a>
+  <ul class="collapse">
+  <li><a href="#modeling-truncated-data-with-the-diffusion-model" id="toc-modeling-truncated-data-with-the-diffusion-model" class="nav-link" data-scroll-target="#modeling-truncated-data-with-the-diffusion-model">Modeling truncated data with the diffusion model</a></li>
+  <li><a href="#modeling-censored-data-with-the-diffusion-model" id="toc-modeling-censored-data-with-the-diffusion-model" class="nav-link" data-scroll-target="#modeling-censored-data-with-the-diffusion-model">Modeling censored data with the diffusion model</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/wiener_diffusion_model.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div><div class="quarto-alternate-formats"><h2>Other Formats</h2><ul><li><a href="wiener_diffusion_model.pdf"><i class="bi bi-file-pdf"></i>PDF</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/wiener_diffusion_model.html">Wiener diffusion model</a></li></ol></nav></header>
+
+
+
+
+<section id="wiener-diffusion-model" class="level1">
+<h1>Wiener diffusion model</h1>
+<p>Diffusion models, sometimes also called Wiener diffusion models, are among the most frequently used model families in modeling two-alternative forced-choice tasks (see <span class="citation" data-cites="Wagenmakers2009">Wagenmakers (<a href="#ref-Wagenmakers2009" role="doc-biblioref">2009</a>)</span>, for a review). Diffusion models allow to model response times and responses jointly. The basic version of a diffusion model comprises four parameters: the boundary separation, <span class="math inline">\(a\)</span>, the relative starting point, <span class="math inline">\(w\)</span>, the drift rate, <span class="math inline">\(v\)</span>, and the non-decision time, <span class="math inline">\(t0\)</span> <span class="citation" data-cites="Ratcliff1978">(<a href="#ref-Ratcliff1978" role="doc-biblioref">Ratcliff 1978</a>)</span>. In the basic model, it is assumed that the four basic parameters are the same for the whole experiment. As this assumption is very strict and there are examples that suggest that the basic parameters can be different from trial to trial, so called inter-trial variabilities were introduced and the basic four- parameter model was extended to a seven-parameter model. In the seven-parameter extension of the diffusion model there are the following three parameters added: the inter-trial variability in relative starting point, <span class="math inline">\(s_w\)</span>, the inter-trial variability in drift rate, <span class="math inline">\(s_v\)</span>, and the inter-trial variability in non-decision time, <span class="math inline">\(s_{t0}\)</span> <span class="citation" data-cites="Ratcliff1998">Nicenboim, Schad, and Vasishth (<a href="#ref-nicenboim2025introduction" role="doc-biblioref">2025</a>)</span>.</p>
+<p>Data for the diffusion model is two-dimensional: There is one vector for the reaction times, <span class="math inline">\(y\)</span>, and one vector for the given responses, <span class="math inline">\(\text{resp}\)</span>. The reaction times shall be positive, continuous and in seconds, the responses shall be binary.</p>
+<p>As a diffusion model describes the decision process for a decision with exactly two choices, there exist reaction time distributions for each response alternative. This means that the probability density function (<span class="math inline">\(p\)</span>) splits into one part for one response alternative and one part for the other response alternative. In the following, we will refer to one alternative as the <em>upper response boundary</em> and to the other alternative as the <em>lower response boundary</em>. <span class="math inline">\(p\)</span> of the lower response boundary can be obtained when inserting <span class="math inline">\(-v\)</span> and <span class="math inline">\(1-w\)</span> to <span class="math inline">\(p\)</span> of the upper response boundary. Let’s call <span class="math inline">\(p\)</span> for the lower response boundary <span class="math inline">\(p_0\)</span> and <span class="math inline">\(p\)</span> for the upper response boundary <span class="math inline">\(p_1\)</span>. Then:</p>
+<p><span class="math display">\[
+p_0(a,t0,v,w,sv,sw,st0) = p_1(a,t0,-v,1-w,sv,sw,st0)
+\]</span></p>
+<p>Usually, a <span class="math inline">\(PDF\)</span> integrates to 1. In the case of the diffusion model, only the sum of both parts, <span class="math inline">\(p_0\)</span> and <span class="math inline">\(p_1\)</span>, integrates to 1. This is called <em>defective</em>.</p>
+<div id="fig-DDM" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-float-fig figure">
+<div aria-describedby="fig-DDM-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="img/Figure_DDM.png" class="img-fluid figure-img">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-DDM-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+Figure&nbsp;1: Figure 1: Realization of a Four-Parameter Diffusion Process Modeling the Binary Decision Process. Image from <span class="citation" data-cites="Henrich2024">Henrich et al. (<a href="#ref-Henrich2024" role="doc-biblioref">2024</a>)</span>, distributed under the Creative Commons Attribution 4.0 International License. <em>Note.</em> The parameters are the boundary separation a for two response alternatives, the relative starting point w, the drift rate v, and the non-decision time t0. The decision process is illustrated as a jagged line between the two boundaries. The predicted distributions of the reaction times are depicted as curved lines below and above the response boundaries (blue).
+</figcaption>
+</figure>
+</div>
+<p>In this model it is assumed that the decision process behaves like a random walk and we are interested in the first time that the random walk crosses one of the two decision boundaries. Hence, we are interested in the first-passage time of the decision process. The Stan function <code>wiener_lpdf()</code> returns the logarithm of the first-passage time density function for a diffusion model with up to seven parameters for upper boundary responses, <span class="math inline">\(\log(p_1)\)</span>. As can be seen above, it suffices to implement the density for only one response boundary, as the other can be obtained by mirroring the starting point and drift rate. Any combination of fixed and estimated parameters can be specified. In other words, with this implementation it is not only possible to estimate parameters of the fullseven- parameter model, but also to estimate restricted models such as the basic four- parameter model, or a five or six-parameter model, or even a one-parameter model when fixing the other six parameters.</p>
+<p>For example, it is possible to permit variability in just one or two parameters and to fix the other variabilities to 0, or even to estimate a three-parameter model when fixing more parameters (e.g., fixing the relative starting point at 0.5).</p>
+<p>It is assumed that the reaction time data that correspond to the upper response boundary <span class="math inline">\(y_\text{upper}\)</span> is distributed according to <code>wiener_lpdf()</code>:</p>
+<p><span class="math display">\[
+y_\text{lower} \sim \operatorname{wiener\_lpdf}(a, t0, w, v, s_v, s_w, s_{t0})
+\]</span> and the reaction time data that correspond to the lower response boundary <span class="math inline">\(y_\text{lower}\)</span> is distributed according to <code>wiener_lpdf()</code> with mirrored starting point and drift rate:</p>
+<p><span class="math display">\[
+y_\text{upper} \sim \operatorname{wiener\_lpdf}(a, t0, 1-w, -v, s_v, s_w, s_{t0})
+\]</span></p>
+<section id="function-call-example" class="level2">
+<h2 class="anchored" data-anchor-id="function-call-example">Function call example</h2>
+<p>The following example demonstrates a diffusion model call in Stan:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N; <span class="co">// Number of trials</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> rt; <span class="co">// response times (in seconds )</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; resp; <span class="co">// responses {0 ,1}</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span>{</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> min_rt = min(rt);</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; a;                <span class="co">// boundary separation</span></span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> v;                          <span class="co">// drift</span></span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; w;       <span class="co">// relative starting point</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=min_rt&gt; t0; <span class="co">// non-decision time</span></span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sv;               <span class="co">// variability in drift</span></span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  <span class="co">// variability in starting point</span></span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=fmin(<span class="dv">2</span> * w, <span class="dv">2</span> * (<span class="dv">1</span> - w))&gt; sw; </span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; st0;             <span class="co">// variability in non-decision time</span></span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span>{</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> one_minus_w = <span class="dv">1</span> - w;</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> neg_v = -v;</span>
+<span id="cb1-23"><a href="#cb1-23" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-24"><a href="#cb1-24" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-25"><a href="#cb1-25" aria-hidden="true" tabindex="-1"></a>  <span class="co">// prior</span></span>
+<span id="cb1-26"><a href="#cb1-26" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb1-27"><a href="#cb1-27" aria-hidden="true" tabindex="-1"></a>  w ~ normal(<span class="fl">0.5</span>, <span class="fl">0.1</span>);</span>
+<span id="cb1-28"><a href="#cb1-28" aria-hidden="true" tabindex="-1"></a>  v ~ normal(<span class="dv">2</span>, <span class="dv">3</span>);</span>
+<span id="cb1-29"><a href="#cb1-29" aria-hidden="true" tabindex="-1"></a>  t0 ~ normal(<span class="fl">0.435</span>, <span class="fl">0.12</span>);</span>
+<span id="cb1-30"><a href="#cb1-30" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-31"><a href="#cb1-31" aria-hidden="true" tabindex="-1"></a>  sv ~ normal(<span class="dv">1</span>, <span class="dv">3</span>);</span>
+<span id="cb1-32"><a href="#cb1-32" aria-hidden="true" tabindex="-1"></a>  st0 ~ normal(<span class="fl">0.183</span>, <span class="fl">0.09</span>);</span>
+<span id="cb1-33"><a href="#cb1-33" aria-hidden="true" tabindex="-1"></a>  sw ~ beta(<span class="dv">1</span>, <span class="dv">3</span>);</span>
+<span id="cb1-34"><a href="#cb1-34" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-35"><a href="#cb1-35" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood (diffusion model)</span></span>
+<span id="cb1-36"><a href="#cb1-36" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-37"><a href="#cb1-37" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb1-38"><a href="#cb1-38" aria-hidden="true" tabindex="-1"></a>      <span class="co">// upper boundary</span></span>
+<span id="cb1-39"><a href="#cb1-39" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_full_lpdf(rt[i] | a, t0, w, v,</span>
+<span id="cb1-40"><a href="#cb1-40" aria-hidden="true" tabindex="-1"></a>                                         sv, sw, st0);</span>
+<span id="cb1-41"><a href="#cb1-41" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> {</span>
+<span id="cb1-42"><a href="#cb1-42" aria-hidden="true" tabindex="-1"></a>      <span class="co">// lower boundary: mirror drift and starting point</span></span>
+<span id="cb1-43"><a href="#cb1-43" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_full_lpdf(rt[i] | a, t0, one_minus_w,</span>
+<span id="cb1-44"><a href="#cb1-44" aria-hidden="true" tabindex="-1"></a>                                         neg_v, sv, sw, st0);</span>
+<span id="cb1-45"><a href="#cb1-45" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-46"><a href="#cb1-46" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-47"><a href="#cb1-47" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="the-data-block" class="level3">
+<h3 class="anchored" data-anchor-id="the-data-block">The data block</h3>
+<p>The data should consist of at least three variables:</p>
+<ol type="1">
+<li>The number of trials N,</li>
+<li>the response, coded as 0 = “lower bound” and 1 = “upper bound”, and</li>
+<li>the reaction times in seconds (not milliseconds).</li>
+</ol>
+<p>Note that two different ways of coding responses are commonly used: First, in <em>response coding</em>, the boundaries correspond to the two response alternatives. Second, in <em>accuracy coding</em>, the boundaries correspond to correct (upper bound) and wrong (lower bound) responses. This means, depending on the coding you choose, the bounds mentioned in the second variable above differ and the <em>response</em> variable will have a different form.</p>
+<p>Most often, an experimenter wants to find out whether an experimental manipulation influences the model parameters. As there exists psychological interpretations for each diffusion model parameter, the experimenter can draw conclusions from differing parameters. Therefore, usually an own diffusion model is being computed for each experimental group to enable a comparison of the parameters between the groups. This can be manipulation between different subjects, like an experimental group and a control group (so called between-subject manipulations). However, this can also be manipulations within the same subject by presenting stimuli from different experimental groups (so called within-subject manipulations). Depending on the experimental design, one would typically also provide the number of conditions and the condition associated with each trial as a vector. Then, one model for each condition will be computed. This means that the parameters also have to be defined for each condition.</p>
+<p>In a hierarchical setting, the data block would also specify the number of participants and the participant associated with each trial as a vector. It is also possible to hand over a precision value in the data block.</p>
+</section>
+<section id="the-parameters-block" class="level3">
+<h3 class="anchored" data-anchor-id="the-parameters-block">The parameters block</h3>
+<p>The model arguments of the <code>wiener_lpdf()</code> function that are not fixed to a certain value are defned as parameters in the parameters block. In this block, it is also possible to insert restrictions on the parameters. Note that the MCMC algorithm iteratively searches for the next parameter set. If the suggested sample falls outside the internally defined parameter ranges, the program will throw an error, which causes the algorithm to restart the current iteration. Since this slows down the sampling process, it is advisable to include the parameter ranges in the defnition of the parameters in the parameters block to improve the sampling process (see table below for the parameter ranges). In addition, the parameter space is further constrained by the following conditions:</p>
+<ol type="1">
+<li>The non-decision time <span class="math inline">\(t_0\)</span> has to be smaller or equal to the observed reaction time: <span class="math inline">\(t0 \leq y\)</span>.</li>
+<li>The varying relative starting point <span class="math inline">\(w\)</span> has to be in the interval (0,1) and thus,</li>
+</ol>
+<p><span class="math display">\[
+\begin{aligned}
+&amp;w + \frac{s_w}{2} &lt; 1 \text{, and} \\
+&amp;0 &lt; w-\frac{s_w}{2}
+\end{aligned}
+\]</span></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;"><em>Parameter</em></th>
+<th style="text-align: left;"><em>Range</em></th>
+<th style="text-align: left;"></th>
+<th style="text-align: left;"><em>Parameter</em></th>
+<th style="text-align: left;"><em>Range</em></th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><span class="math inline">\(a\)</span></td>
+<td style="text-align: left;">(0, <span class="math inline">\(\infty\)</span>)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(y\)</span></td>
+<td style="text-align: left;">(0, <span class="math inline">\(\infty\)</span>)</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><span class="math inline">\(v\)</span></td>
+<td style="text-align: left;">(-<span class="math inline">\(\infty\)</span>, <span class="math inline">\(\infty\)</span>)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(s_v\)</span></td>
+<td style="text-align: left;">[0, <span class="math inline">\(\infty\)</span>)</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><span class="math inline">\(w\)</span></td>
+<td style="text-align: left;">(0,1)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(s_w\)</span></td>
+<td style="text-align: left;">[0,min(2<em>w, 2</em>(1-w)))</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><span class="math inline">\(t_0\)</span></td>
+<td style="text-align: left;">[0,<span class="math inline">\(\infty\)</span>)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(s_{t0}\)</span></td>
+<td style="text-align: left;">[0,<span class="math inline">\(\infty\)</span>)</td>
+</tr>
+</tbody>
+</table>
+</section>
+<section id="the-model-block" class="level3">
+<h3 class="anchored" data-anchor-id="the-model-block">The model block</h3>
+<p>In the model block, the priors and likelihood are defined for the upper and the lower response boundary. Different kinds of priors can be specifed here. Generally, the regularization induced by mildly informative priors can help both statistically and computationally.</p>
+<p>In the second part of the model block, the data generating distribution is applied to all responses. The drift rate <span class="math inline">\(v\)</span> and relative starting point <span class="math inline">\(w\)</span> have to be mirrored for responses at the lower boundary.</p>
+<p>For more details regarding the application of the diffusion model in Stan, see <span class="citation" data-cites="Henrich2024">Henrich et al. (<a href="#ref-Henrich2024" role="doc-biblioref">2024</a>)</span>.</p>
+</section>
+</section>
+<section id="wiener-truncation" class="level2">
+<h2 class="anchored" data-anchor-id="wiener-truncation">Truncated and censored data</h2>
+<p>Truncation and censoring frequently occur in psychological data collection. For reaction time data, truncated and censored data regularly arise in psychological studies as a consequence of using response windows or deadlines. These are sometimes introduced in the analysis of data to exclude reaction times that appear too short or too long, but they are also sometimes already built into the study procedures to push participants to respond within a specifc temporal window.</p>
+<p>Depending on the implementation of the response window, two different types of data arise: <em>truncated</em> data or <em>censored</em> data. Since the effects of truncation or censoring on summary statistics such as mean, median, standard deviation, and skewness is regularly too large to ignore <span class="citation" data-cites="Ulrich1994">(<a href="#ref-Ulrich1994" role="doc-biblioref">Ulrich and Miller 1994</a>)</span>, data analysts are well advised to account for these effects.</p>
+<p>As described in the <a href="../stan-users-guide/truncation-censoring.html">Truncated or Censored Data</a> chapter, the cumulative distribution function (<span class="math inline">\(F\)</span>) and its complement (<span class="math inline">\(\text{CCDF}\)</span>) are needed to model truncated and censored data.</p>
+<p>As explained above, <span class="math inline">\(p\)</span> is defined <em>defectively</em>, meaning that only the sum of <span class="math inline">\(p\)</span>s for both response alternatives integrates to 1. For the same reason, <span class="math inline">\(F\)</span> and <span class="math inline">\(\text{CCDF}\)</span> are also implemented defectively. Analogously, only the sum of the <span class="math inline">\(F\)</span>s and <span class="math inline">\(\text{CCDF}\)</span>s for both response alternatives asymptotes above at 1.</p>
+<p>In the case of the diffusion model, <span class="math inline">\(F\)</span> asymptotes above at the probability <span class="math inline">\(PROB\)</span> to hit the corresponding response boundary: (for simplicity, we omit the inter-trial variabilities in the following)</p>
+<p><span class="math display">\[
+\begin{aligned}
+F_1(\infty\mid a,w,v) &amp;= \text{PROB}(a,w,v) \text{ and} \\
+F_0(\infty\mid a,w,v) &amp;= F_1(\infty\mid a,1-w,-v) = \text{PROB}(a,1-w,-v)
+\end{aligned}
+\]</span></p>
+<section id="modeling-truncated-data-with-the-diffusion-model" class="level3">
+<h3 class="anchored" data-anchor-id="modeling-truncated-data-with-the-diffusion-model">Modeling truncated data with the diffusion model</h3>
+<p>Data are called <em>truncated</em> when there is no information available for analysis from trials with values larger (or smaller) than a right (or left) reaction-time bound. In reaction time experiments, reaction time data are truncated if trials with reaction times outside the response window are excluded from the analysis. Not even a count of those omitted trials is kept.</p>
+<p>Let <span class="math inline">\(L\)</span> denote the left reaction-time bound and <span class="math inline">\(U\)</span> denote the right reaction-time bound of a response window.</p>
+<p>Then, the density of truncated data for both response boundaries 0 and 1, here denoted as <span class="math inline">\(\text{resp}\in\{0,1\}\)</span>, can be formulated as follows:</p>
+<p><span class="math display">\[
+\begin{aligned}
+&amp;p_{\text{resp}}(y \mid L&lt;X\leq U, a, w, v) = \\ &amp;\frac{p_{\text{resp}}(y \mid a, w, v)\cdot \mathbb{I}_{\{L&lt;y\leq U\}}}
+{\bigl(F_0(U \mid a, w, v)+F_1(U \mid a, w, v)\bigr) -
+\bigl(F_0(L\mid a, w, v)+F_1(L\mid a, w, v)\bigr)}
+\end{aligned}
+\]</span></p>
+<p>The density of left truncated data can be formulated as follows. <span class="math display">\[
+\begin{aligned}
+p_{\text{resp}}(y \mid L&lt;X, a, w, v) = \frac{p_{\text{resp}}(y \mid a, w, v)\cdot \mathbb{I}_{\{L&lt;y\}}}
+{1-\bigl(F_0(L \mid a, w, v)+F_1(L \mid a, w, v)\bigr)},
+\end{aligned}
+\]</span></p>
+<p>The density of right truncated data can be formulated as follows.</p>
+<p><span class="math display">\[
+\begin{aligned}
+p_{\text{resp}}(y \mid X\leq U, a, w, v) = \frac{p_{\text{resp}}(y \mid a, w, v)\cdot \mathbb{I}_{\{y\leq U\}}}{F_0(U \mid a, w, v)+F_1(U \mid a, w, v)}
+\end{aligned}
+\]</span></p>
+<p>As the functions are implemented defectively, a truncated diffusion model cannot be calculated with the truncation functor <span class="math inline">\(T[,]\)</span> as it would usually be done in Stan. This means the function call: <code>y ~ wiener(...)T[L,U]</code> does not work the way it is supposed to. When the truncation functor is called in Stan, Stan searches for a CDF implementation internally. In the case of the diffusion model, Stan would find the CDF, but is not aware of its defective implementation and calculates the computations as if it were a non-defective CDF. This causes misleading and incorrect results.</p>
+<p>To implement the truncated model, write out the function shown above on the log-scale with <code>left_bound = L</code> and <code>right_bound = U</code>, where <code>wiener_lcdf_unnorm()</code> calls the logarithmized CDF of the diffusion model at the response-1-boundary:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_denom = log_diff_exp(</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>    log_sum_exp(</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>                                       sv, sw, st)),</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>    log_sum_exp(</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st),</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>                                      sv, sw, st)));</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood</span></span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>      <span class="co">// response -1 boundary</span></span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_lpdf (rt[i] | a, t0, w, v, sv, sw, st);</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> { </span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>      <span class="co">// response -0 boundary ( mirror v and w)</span></span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_lpdf (rt[i] | a, t0, one_minus_w, neg_v,</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>                                     sv, sw, st);</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>  } <span class="co">// end for</span></span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -N * log_denom;</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For details of how to call a truncated model within the parallelization routine of <code>reduce_sum</code> or with truncation to only on side, see <span class="citation" data-cites="Henrich2026">Henrich and Klauer (<a href="#ref-Henrich2026" role="doc-biblioref">2026</a>)</span>.</p>
+</section>
+<section id="modeling-censored-data-with-the-diffusion-model" class="level3">
+<h3 class="anchored" data-anchor-id="modeling-censored-data-with-the-diffusion-model">Modeling censored data with the diffusion model</h3>
+<p>Data are <em>censored</em> when observations that are above or below a right or left boundary value are reported as occurrences of the event <span class="math inline">\((y &gt;
+U)\)</span>, for <span class="math inline">\(U\)</span> the right bound, or as occurrences of the event <span class="math inline">\((y \leq
+L)\)</span>, for <span class="math inline">\(L\)</span> the left bound, respectively. Like for truncated data, the range of the possible values is restricted, but the number of observations that fall outside the boundaries is kept, whereas in truncation, no count would be kept.</p>
+<p>For the censored model, we distinguish two cases. In the first case, the responses of the censored trials are known, but the reaction times are not known. In the second case, neither the responses nor the reaction times of the censored trials are known. Note that the second case differs from a truncated model in the fact that the number of censored trials is still known. Consider first the case where the response is known even for censored data.</p>
+<p>To model such data in Stan, the left and right reaction time bounds, <code>left_bound</code> and <code>right_bound</code>, respectively, are handed over in the data block, as well as a vector <code>censored</code> that tracks whether a trial is censored (= 1) or not (= 0), and counts of trials censored at the left reaction time bound and counts of trials censored at the right reaction time bound for each response in {0,1}. There are four such count variables: <code>N_cens_left_0</code>, <code>N_cens_left_1</code>, <code>N_cens_right_0</code>, <code>N_cens_right_1</code>:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (censored[i] == <span class="dv">0</span>) {</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>        y[i] ~ wiener(a, t0, w, v, sv, sw, st0);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (resp[i] == <span class="dv">0</span>) {</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>        y[i] ~ wiener(a, t0, one_minus_w, neg_v, sv, sw, st0);</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood (response = 0)</span></span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_left_0 </span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>         * wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>                                           sv, sw, st0);</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-17"><a href="#cb3-17" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_right_0 </span>
+<span id="cb3-18"><a href="#cb3-18" aria-hidden="true" tabindex="-1"></a>         * wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb3-19"><a href="#cb3-19" aria-hidden="true" tabindex="-1"></a>                                             sv, sw, st0);</span>
+<span id="cb3-20"><a href="#cb3-20" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-21"><a href="#cb3-21" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood (response = 1)</span></span>
+<span id="cb3-22"><a href="#cb3-22" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_left_1 </span>
+<span id="cb3-23"><a href="#cb3-23" aria-hidden="true" tabindex="-1"></a>         * wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st0);</span>
+<span id="cb3-24"><a href="#cb3-24" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-25"><a href="#cb3-25" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_right_1 </span>
+<span id="cb3-26"><a href="#cb3-26" aria-hidden="true" tabindex="-1"></a>         * wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st0);</span>
+<span id="cb3-27"><a href="#cb3-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When data are censored at only one side, meaning that the reaction time constraint only exists for one of the two boundaries, omit the lines for the other side in the code. A both sided reaction time window would be, for example, when only reaction times are accepted that occur between 0.2 and 0.8 seconds. A one sided reaction time constraint would be, for example, when all reaction times below 0.8 seconds are accepted.</p>
+<p>When data consist of many conditions (as explained in the beginning), it is sometimes more convenient to loop over all trials instead of using count variables as described above, using the following notation and code. A vector containing the information whether a trial is censored or not, here <code>censored</code>, needs to be handed over in the data block. This vector splits the data into three bins: all trials <span class="math inline">\(i\)</span> with<code>censored[i]=0</code> are censored below the left reaction time bound, all trials <span class="math inline">\(i\)</span> with <code>censored[i]=1</code> fall between the reaction time bounds, and all trials <span class="math inline">\(i\)</span> with <code>censored[i]=2</code> are censored above the right reaction time bound. For non-censored trials, the log-PDF is computed, for left censored trials, the log-CDF is computed, and for right censored trials, the log-CCDF is computed:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> { </span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) { </span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>    <span class="co">// right censored at right_bound</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (resp [i] == <span class="dv">1</span>) { </span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>      <span class="co">// upper response boundary</span></span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (censored[i] == <span class="dv">0</span>) {</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lcdf_unnorm(left_bound | a, t0, w, v, </span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>                                                  sv, sw, st0);</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">1</span>) {</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lpdf(y[i] | a, t0, w, v, sv, sw, st0);</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">2</span>) {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>         <span class="kw">target +=</span> wiener_lccdf_unnorm(right_bound | a, t0, w, v,</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>                                                     sv, sw, st0);</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> { </span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>      <span class="co">// lower response boundary (mirror drift and // starting point!)</span></span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (censored[i] == <span class="dv">0</span>) {</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w,</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>                                                  neg_v, sv, sw, st0);</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">1</span>) {</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lpdf(y[i] | a, t0, one_minus_w, neg_v,</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>                                     sv, sw, st0);</span>
+<span id="cb4-23"><a href="#cb4-23" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">2</span>) {</span>
+<span id="cb4-24"><a href="#cb4-24" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,</span>
+<span id="cb4-25"><a href="#cb4-25" aria-hidden="true" tabindex="-1"></a>                                                    neg_v, sv, sw, st0);</span>
+<span id="cb4-26"><a href="#cb4-26" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb4-27"><a href="#cb4-27" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb4-28"><a href="#cb4-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-29"><a href="#cb4-29" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When the data are censored on only one side, omit the case that is not needed.</p>
+<p>Note that this block can be inserted in the defnition of the parallelization function, <code>partial_sum_wiener()</code>, as defined below.</p>
+<p>Sometimes also the response is missing (i.e., it is known that the reaction time in a trial fell outside the response window, but which response was given is unknown). One method that has been used to model such data has involved inferring the numbers of missing responses of either kind from the observed relative frequencies of the two responses. This approach has the problem that quite specifc assumptions on the missing data have to be made (namely, that the proportions of the two kinds of responses are the same for responses within and outside the response window).</p>
+<p>The following is a more principled approach that uses the cumulative distribution functions and their complements to provide the data-generating distribution of censored data. As before, let <span class="math inline">\(L\)</span> be the left reaction time bound, and <span class="math inline">\(U\)</span> the right reaction time bound, and consider decision times without inter-trial variabilities for the sake of simplicity. It follows that the likelihood contribution <span class="math inline">\(\textit{lik}_l\)</span> for a left-censored data point is given by</p>
+<p><span class="math display">\[
+\begin{aligned}
+\textit{lik}_l(a,w,v) = F_0(L\mid a,w,v) + F_1(L\mid a,w,v),
+\end{aligned}
+\]</span></p>
+<p>whereas the likelihood contribution <span class="math inline">\(lik_r\)</span> due to a right-censored data point is given by</p>
+<p><span class="math display">\[
+\begin{aligned}
+\textit{lik}_r(a,w,v) = \text{CCDF}_0(U\mid a,w,v) + \text{CCDF}_1(U\mid a,w,v).
+\end{aligned}
+\]</span></p>
+<p>See the following code for an example of Stan code implementing this second case of censoring. This model call deals with the problem of unknown responses by computing the probability of choosing the response-1 or response-0 boundary outside the response window. Here, the CDF and/or the CCDF are required, depending upon whether there is only left-censoring, right-censoring, or censoring both to the left and to the right. The following code shows the functions block for a model that is right-censored using the function <code>partial_sum_wiener()</code> to parallelize the execution of a single Stan chain across multiple cores:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> partial_sum_wiener(<span class="dt">array</span>[] <span class="dt">real</span> rt_slice, <span class="dt">int</span> start,</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">int</span> end, <span class="dt">real</span> a, <span class="dt">real</span> t0, <span class="dt">real</span> w,</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">real</span> v, <span class="dt">real</span> sv, <span class="dt">real</span> sw, <span class="dt">real</span> st,</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">array</span>[] <span class="dt">int</span> resp, <span class="dt">real</span> right_bound,</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">array</span>[] <span class="dt">int</span> censored) {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> ans = <span class="dv">0</span>;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> start:end) {</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (censored[i] == <span class="dv">1</span>) {</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>        <span class="co">// not censored</span></span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>        <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>          <span class="co">// upper boundary</span></span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>          ans += wiener_lpdf(rt_slice[i+<span class="dv">1</span>- start ] | a, t0, w, v,</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>                                                     sv, sw, st);</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>        } <span class="cf">else</span> {</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>          <span class="co">// lower boundary(mirror v and w)</span></span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>          ans += wiener_lpdf(rt_slice[i+<span class="dv">1</span>- start ] | a, t0, one_minus_w,</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>                                                     neg_v, sv, sw, st);</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>        }</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> { </span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>        <span class="co">// censored</span></span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>        ans += log_sum_exp (</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>          wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>          wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a>                                            neg_v, sv, sw, st);</span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb5-27"><a href="#cb5-27" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb5-28"><a href="#cb5-28" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> ans;</span>
+<span id="cb5-29"><a href="#cb5-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-30"><a href="#cb5-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Combine this block with the model block in the example above by using the function <code>reduce_sum()</code>.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> reduce_sum(partial_sum_wiener, rt, <span class="dv">1</span>,</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>    a, t0, w, v, sv, sw, st, resp, right_bound, censored);</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For more details, see <span class="citation" data-cites="Henrich2026">Henrich and Klauer (<a href="#ref-Henrich2026" role="doc-biblioref">2026</a>)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Henrich2024" class="csl-entry" role="listitem">
+Henrich, Franziska, Raphael Hartmann, Valentin Pratz, Andreas Voss, and Karl Christoph Klauer. 2024. <span>“The Seven-Parameter Diffusion Model: An Implementation in Stan for Bayesian Analyses.”</span> <em>Behavior Research Methods</em> 56 (4): 3102–16. <a href="https://doi.org/10.3758/s13428-023-02179-1">https://doi.org/10.3758/s13428-023-02179-1</a>.
+</div>
+<div id="ref-Henrich2026" class="csl-entry" role="listitem">
+Henrich, Franziska, and Karl Christoph Klauer. 2026. <span>“Modeling Truncated and Censored Data with the Diffusion Model in <span>S</span>tan.”</span> <em>Behavior Research Methods</em> 58 (42). <a href="https://doi.org/10.3758/s13428-025-02822-z">https://doi.org/10.3758/s13428-025-02822-z</a>.
+</div>
+<div id="ref-nicenboim2025introduction" class="csl-entry" role="listitem">
+Nicenboim, Bruno, Daniel J Schad, and Shravan Vasishth. 2025. <em>Introduction to <span>B</span>ayesian Data Analysis for Cognitive Science</em>. CRC Press.
+</div>
+<div id="ref-Ratcliff1978" class="csl-entry" role="listitem">
+Ratcliff, Roger. 1978. <span>“A Theory of Memory Retrieval.”</span> <em>Psychological Review</em> 85 (2): 59–108.
+</div>
+<div id="ref-Ratcliff1998" class="csl-entry" role="listitem">
+Ratcliff, Roger, and Jeffrey N. Rouder. 1998. <span>“Modelling Response Times for Two-Choice Decisions.”</span> <em>Psychological Science</em> 9 (5): 347–56.
+</div>
+<div id="ref-Ulrich1994" class="csl-entry" role="listitem">
+Ulrich, Rolf, and Jeff Miller. 1994. <span>“Effects of Truncation on Reaction Time Analysis.”</span> <em>Journal of Experimental Psychology: General</em> 123 (1): 34–80.
+</div>
+<div id="ref-Wagenmakers2009" class="csl-entry" role="listitem">
+Wagenmakers, Eric-Jan. 2009. <span>“Methodological and Empirical Developments for the <span>R</span>atcliff Diffusion Model of Response Times and Accuracy.”</span> <em>European Journal of Cognitive Psychology</em> 21 (5): 641–71. <a href="https://doi.org/10.1080/09541440802205067">https://doi.org/10.1080/09541440802205067</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/survival.html" class="pagination-link" aria-label="Survival Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Survival Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/floating-point.html" class="pagination-link" aria-label="Floating Point Arithmetic">
+        <span class="nav-page-text">Floating Point Arithmetic</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/wiener_diffusion_model.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/2_39/stan-users-guide/wiener_diffusion_model.pdf b/docs/2_39/stan-users-guide/wiener_diffusion_model.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..fdec7b5f6f658697b7ba4465430774cfafd01b6b
GIT binary patch
literal 168306
zcmagEW2`Vxvu1g0+qP}nwr$(CZQHhOzsI(1pZ+?1XXYkzlm4?)yM9+CtMV*T1rae?
zMmkm~(&PJ^cPJ)C0tNzmBP%E#9(plL8y8b2dNCVA7gG^aV|x=*dKptYa~BH&CJuI1
zK0YXC7bjCgTPTms8|}$Bl6J)2FX~J1EMc%<Y=!e+z*vAinCIF1LiAKQzV|<RGE5k|
z$(gCkMF}RB3`-Z(WN8|DH8cHRZ$)&>zsl&~cYVFxKji)&9Dl#wudh4$bb9?2(ZU6b
zgGZ-s3ets)>EtA$3<uAT_cc|it9;;*y-QuG1*M{NrQxUKHnun{)c(%r*-`s@G2(Z`
zwNRFN-?sbjyL5jYKd%i^l02&5OGrTkrVgP9X>8#wmY=pH$Utmf*Lp`5T*&9#2}2-8
zWQ_jGCKCHo^QktJ!o;C{nIz6r$v5r*WWA5+WX7Q)T5OB2>tqMy(a3U=0g{?PCs@%I
zNl?k}#C?vcdXc{2N0NsyRl(2(T0*E|mtmR6NN*#EljOuIza;-i`Mt<J)P`pW8q%e4
zMeNs?Un$^=qmxOn!l24ktW<w$v@F6Tg}nVlgVNnAi^gLUn0@BHJVNaVUV*q(p{k3K
zek2J$@qHgKp<q214tPTNA;Mds2^4r^(<hdVFR<64jb4A`mKxN-s>l@l_0QW3%;VZ0
z&lwPEf@|Y2Vp_ri(M@|GNb$OZYy8A3csK4~wmE?aWkS;6`wHzM$<73vCqg(qNCAf<
zABwL+c)UeY;VMY;$^0cf+Z`gDTjN>Gzz)Jk@P<C*2r`<Cr#Oi~G%f+n$a?91VM<pz
z>|G3~PO47mwjZ3SR3zg0{+Dh$=7!;=;|0T8B9-a$F4{Su_kTo}fHW{Lydj{t{~%Q2
zjSrOPj^+b{&%Hr8xS9X__l@xcMh3E0-?Sj3^TM8lB%i=vke;i$0T)jMT|Muo9||po
z`M4~&d`J!sTLnAjmzmE$yg^90JZ*VFwFZlg+Y2)Rd3TcaG`+uIF%O!A<3Zeg%S?u!
zOWHZG4m{bS1H@ac+hUonq{$*6gSVfV>_xa8-*X~om|{S<uq)kuH#>|k0-(h+>3QYK
znc>$zh}EENW$`ZO(u#?TD-&mP#JLHonlntw&+5#isW);>V<okUnKA1Srt1=PO;+n|
zFujpFDI*FvRrw09^DA0r^yZdjlXL2d<rKj(R?p`MiHjGRd|OXtro5FF|2j2;FG4!z
zKtBg9^u~}{6e<JTErtj6qNV5aG4SNrPDt4E-6ZH{l7#z>n_AWYm$n9Hmk(9c41BDj
zFq(@wH|<KkKO`H`2bfx4O`|8KpF@q$Ee=7v5vB&)e!wDN(I>;UErj*j2?%<`m-d?C
zNo6ZBs56+Bmy%r+f@-l17%K_u>vOF%Rb{_+onId{E<YS?ebVu2Xge@VlFZeXeNm1P
z9m~0B9&#kuP6L=r&_#l8Rg|1JwYNAp>A2!1v*K~J=QlKxy!w1tg;iD=Q^2ru%~n-&
zbj)HYGacj}aCR>s8L2i7Olp9ckEczWlQ%>=srLjzsJ~vv#6?-o+&%jd>ye4pY^Z99
z@CutC-A2u5OcA}=jsguQm6Sk_h{6jNn7L8#q`h*YC1dT6Gg_C6owXB~HJ-?{mL$C9
zvgYCX25@$<GQg)RzMk<+<TDu87z-8iYcy*)A_gKL>S;*$?YdgHk&iY^s=WevdPY-l
z)%bG>%W=hUGa;QcG8?cBJ9yoX;C?*Q+Rn~pFI&HB?Pju1^W)|0*z*QQ5T?IK>~tcV
zyF<FgH#=hGCeLaqCCpoLJ$4Z^z~eTjP-LtR7wBqEd1CI`n?7Fo`owkYr#Lx;=WzBo
zFkV(PRz&f1FNnJDKA`C4@z&-P$-UQnl&A0_FG{3)pHbKyU29OdBiTR05`UokQQJ`k
zS+qK4Dp$0lPj+Hvxo}$Bmlly6R{4QAuD-I4isKwa{@E&_*Jk&4!8mx1Y4U}d>g`>H
zHq~K?b4+h$j_)q(YB#bV@}WvUyTGxb4m!nSiJvo-2?NYe_&(8nn3t&iXv_V2+pv8&
zn&4dF{Tnq*(qwGMY3BRf=C?C>hB0oo^uXv1kNGxn(ap?x2az)jds~;EjJY)t0J(p)
z;s4o!WSxVpG5JAL^JG1f?Y(hs;SuM+0Y$KiGl-U>cE%p5dA1i!-AtY#J3M=`BCCB)
z%In5#<x<Pvbd9n)LUETc@Ue}qqU4;ZdGc~2Yk|*uGoa=*weEd6FM8hLVt%7y@EMXP
z;5#L+aFJ7lYP@4uxG;aT<$BQTll}4~+6yZXvB=B^tZ2?@*d#h>m)e*{hL}d*Ase8I
zZ5fjGzBpAp>6aAeqxSi;Jm_}AJ1JL!wtvP^RbL6>*QdoG6*xK?(m6e542<jiqLIO6
zLwEF3);vxOH#X3F!F*6^9bTs^=hs^EaP6uCM5d;D<FNRlO*%|}fse3&3~qSny{ltY
zJpG<{5qn*=9;<4UROeO7D-_#7L}7(0@$<%av|FyZ5cZ7YY+uD#c`<6U?<w8egg?m;
zh^%~}WieR$z_U{|_?HKq;g~9ZD~1&?EFm1v=}Hw|^HuJH(svU&PNz7di^bm3*p<=B
z(aP3*C4oQlQ9D2u*9>ITo~PRH)sC>-1lh6o=|~h{=kRt#7Q$+-&40wpH`g>l9Nx6x
zl@`jDebwF)?UyaA<$vSW)Tvn$9*2>Zg0ARffT15g$(mA<c3sHCa_efss(;yl+rB0+
zTc$un`O2e#Q5`B+@N)e6YPVK%LodMBRnptC8=7KOL)b=dwohSU5mn<mrcJ-~a&z%<
z>&$)H=4I}Nq~?na*#(VQb@pTm(w<m!v^47Ykc!)m0ia5WeEyHOU8MQ-E)t;&ak}8l
zPKdalzZYGj_>f{|k#W2BR&V5z3)r=JcfiZ-U|G=0QCjcy<`Z*;>^%u~4UUsHajh7-
zB8-f|P*<i!V>vi|a<GmSh%M*7q`@NWQLE#H`?ZSZJ!A9MMV`#O@vIjMcII2=;Gh;>
zme(6|f{)0X4T{3=x>&H++;oTvn2m-@s^NlH_+q}@4z?Xu%`>l<$NrI5c$G)|%uX{!
z(iRrC#}U3W(t3sp$cxrSx_aft@0&Tz{a9d;-M*>fyGl_6?V>C~!(zA(;Ajz$c9&i~
zPgpYV6GJLvefRGU(J}2CkleL?F$)r!Z=}X^PTO2<=MroAh#Yq4czVMtexKct`p%~g
zc~G(wGt0flI(OP}#&}$htFX=Kd!g?!+oy``F`h!r{lcHu*9OY}<1)<s9}d*nl__c|
z-10d4<z_FV&`-L?`Q%kW4UhGbYUga^YUywp_>S7Oj&I$R7Bvr>FfJ2%XfMGES*eOy
z?AYV6A!4=Q$de~DBJ~x>1ep<#dZ(tLRtrD<E=M;?s^_b5kqXu7;C|M>r)g@LgB9F8
z%#i0YTV?Io4wZrDA(1yX30>!*J;a6u2w-g;*7ip4ayY}jzJf%;#Tj9o-HyPp9c&d$
zH%~PExD_bS&sE=9*Y??V)nFkiM4?^}-`HKFvr7%mKs@!;hD|ldlLA>eI4;3=R!iul
zWB1$XRurYJQ;dnBH1;tdEF`x@piSFFAU|c;@I`3eScSlUkjo{Tz&bE1z#$WHOU7-H
zb#BQkwLSQu0bBjA+Ny_o9`5rQhvOJ2)jpp9(q}Na3LWN(rn8ugi)Q8{ElRzidiAlL
zjg?`Et=lTpu_d`-S_tPAo{?t-xbWe31RJ`~spI{@8j2wH_^8!>?cBVE%fT9$R}mC3
zPtI(o-sNf8Ce_beN=TA1jTA%L{%U__WGLW5rV~lYOrL70yYIC1#rF}#Yu0&>)<HW8
zk1uP6m^I`>Je_fmob}&frJtoew4J9)fKZWM;)Hv1<m17SpibwNhs5D|d^e3ObZzDz
z{D0y9xO(?6MGXkWD0=?d*Z#h@{s7cpZ6u*g?M(iQqW@F<Z|-K|{Qm&_|Ijug0V4}D
z!+)VS69F3=Gy8vim<j%$HUT>uBlG_N?Ef9oV_iU1mUmY9a;1U;0)(ZQT_oJyB_W0x
z1`wDTm{S5;h;XIKfsmxqU4l!Jq{ayfBr08eiu>OB-g^K3wD#J~YWH$_-<sbxzH-hu
zb85oKfTZFaz$3VD!jKTqK&Ajxo}XJv00jyv2oNZQh9)hcL_39losu20gb^etaLAtU
z0Z>6eLWbT|1Sw$T;NgHRyt{!zL;;78n1+&=1PBB`AX0wO5#ka6D}sCq;sAIB1K4oD
z1d|-926%rE70l65%(DGy0k;kC0vcLcNdE3z0+PT%06~Kd0C*9mKo0{q5Mi9aGX@3~
zAg2ECNjpq+6)5B}2?_P}^#v2al@Q__(#{UR--Q(E0E{7!g6;!3fc#)#7Qj6Per3oc
zh9DVS1Bw3LW%uPM^bvr71~|404Ip45C-1{Jg%AK{P63c<bpry)L0-XYzQ70I->+-{
zhJuFv=-$*{>xC57?@SOdVBYRQ2EPj#%ma{@ae){V)=))03f}<<GLGm8CU7CgkAr;#
z6x<Vl!S1`A5d?H6ApnM8-q$=N7=SJ#g`kKEZ~MhiK2^b-r>q#nQ9HuHLBfQRzAxob
zqXGgL$8YJM&&+lV9P~8?^uv%*PS2mMA;j(F37IjLr?HDlKhZ&sh(5N?U<3h20+o`A
zl#Bo(&;bs?-4H)jcbE5&AGs;Mr(-h-C<c&kAsvQb03n081R4BE03{OOK>~z4goA*7
zneXMJMKmz>AwY`(;1J9yVvg@FG1x~g;$u5~K$hVRU`HI}F#x__zTT$Lgz3dOW_JGU
z-|xOqnUvLf>#K&}%n$v?Q&JE70ssIJEdv4~I1mV6Qc%DpjL-!CXpA9(epkTm?{sK0
zka3b9%5`Trf2vLM`!NQ7(z!eU{%FeJ#~09m_J6Za*aRF2FrE;9`JI2<r~cv(`>Vd~
zY5wlQmV0-0{KGr_$N%yfCV($*|Ih)3wd_&E0KFVy;0*p?TLZooYw<=9FK-_96lqZ)
zia|&-n8Rj)hKzy)8u|SwlyEP>9a%z(2I%}bp2qE8rLAujFh&sKLVmq;7&Ha;{YM#q
zfV&Ox5J<u@{DlZe#B3|}*50SX44$9o`IZd~I*#ZMA|s>$1Ry{r@<E(mFr;x12x1f~
zh)3?&TLFcF7&xGkfd*!9KpcXICH!=oTSx>To~K{7kkHZqAPRqnQ-B~ozQZYD5gvbR
ze~ZBz*nt8qAK6_QV1K87e~<(S{16x>%P~3L^+K#=MXn@Mmu&R|V;MESd!Lb|5d#~C
zz(p$fS|PhvOp<f1Idhp_<a>uM;dQv5Q+wr{(#`t83#{*!KcaCVw-xuC`i>GfoP9%h
zQ;{5IM%aXRLsShCt7eFvpccbTCvb!pC2RK2x6Own@jTkg&Wxkz(PbmMO59AU{k+Xp
zCEMt5lUqLKvKTLGhYK#k-r+yHJ+1K|53U;7<o`W0Z@8Fj$x9~rQd6pdg@K_{cxQGS
z3AOqTvAuHQ6;rI-!ZKAxA*yB3_YibXc%MA6CQ}Uv25UW38|!~C$zMvw{uV-i-dEF7
z*Q?vx5&P0F$|cr`Jr>Udy%(B{E|yf?NwP85=MSV=e20v4p=?75eN$nI$rK;S<D<ih
zQbk`vv=;@Qsk-3ktMjW~6OA_O9dtUYM&f+FyoC4PHn^l|I7#xKyR$VODE?tgu-@-O
zodrhX!NdFe$*hOqG)_9&brilEjXp26(K*tg0V>NB6)vGC8HBINfhd=@FE_xz!7p0&
zlv2Ys8=k!z`NU0B*xzNmobQiEd?c(72OP{-Z}A5bO$Pe1rWd8Eg1{yO2sq=KhTeYp
zD*Vud8vMvP-GN>!pf%DlU^cB5g%9||F+Kn~V2TgQ-nuO3STU_z0B5Zt%t7^I1SZ)q
zIx1taaFL9D#*G=clP#9w&05N%L=~1!a{z^(6}+21F)co2xi9NaK;jU2Jwb}ZVDav}
zOTCsot$gN{D65brkkbk8B5od2fs>2r3X`!Vm~6K%vwGJ{cxK~KlA2EDjlGW+DjU&W
zo&87`yHGi56y?`b^!D}EcC0juzl`O7-&zL$%7Ti@@m9BUVsf^$r;FKyYT3`Jh8V!&
zIf7d6I4RYmbn%;Ja8k}5$u*?6R+cx6B0zNV`?yQqD_*3p>#*JPc|>GWg&?6QJM%Ut
z%N_ww8OPm+%i*NiB7=@tX~lP(hdVd`oiiVo(L#4}lA5R3&Ag^R{pCQhZ?8~N0}~(Z
z!Am(6V}t6Loa;yL*7Nq_nw~v*vfDO=d?K`*2+wl3M4vvnn(Z;1ua@2+E@Xaf92G9U
zBMOf*r@#5wSg$KC*L=#W6BtoO%d*1mp_m)VidDfkBZ>SwzApQ?8n0UOy=~D|26CVh
zcCoSeFn!E`W}o_E8E~Y|cZZf$37vP=x~vhg=D){upu%aZ*5Y5+dc;uf6s=?;>1bs3
ztmBK*vV%{%M?LbLJxI7wW-=;p<yRf!6y-~@s}El|S5`A_{xex41+5{~681X3xSHKc
zyAI{gTPL9?*~P5zNUn?>U6ysZpNLuWY9;00JuIAEx2hDN)!Odfx|u~GxV0UarEcqu
z3AAp7$9vioG`U%w*<h%}$+`{d9aIkL8#PHW^2|mmWonztXoLE7NEQJT^xs402?_v3
zRBPiL$KHoQ<wN8&eNKKP6LlVIcOUmQx^`y+mB(?k%iJYlymg}Xi|Y11>GV5Htmx}l
z5t~~KMF-c6@}ZnY&9)ibln&>~h^qLG*^0TLx6-P%oG_BS$xKC}TUg5;19l;IS{y=0
z+K^{!uCB~K^pf$yEFs(I12^6tksH#JfHzNM47$UCX)shI$u9TdeR{P!k|IwAC0&G;
zD;LxMTmiWuVR6}@9lgsx%aHr?X}^|dn+&L5ovB?__v<}=n7OXWwWns{ZY+*g-sWHW
ziHhj$+sksTLoSsyMY80u-25aB@h&yvPj6L_a1^%h7+34p7eF`&gEn<ZA@}Y%msW?S
zBrR`Zpee*jZjYJ36`oB_R!jE-5!0wOZ7`C~+FOdm_h%;nU2tF$gGnccC<+zdqbE&V
z*y453Wt?<0#NJ+Yn-b2;;*j=y>u&0}<)WvlitV7CJVW!K_M#pkH{2TO@q^GGM?7D?
z=&y?ZgUqf&p;EqEVUomU?hS0`#{L^~qS1eJRS4NJqajK8zFhDQZB{2_?8TX3{r73m
z5TAc#O2p<e?thb!^t#_i)Sj_@X1s;{`2HIt*@R^e*wsMWxOugo76@BqT<I*x*_~}J
z=qwn@rKMi~$n&J(*~KCR365LarIE5byN}-J3+@R{e^tfaxeMQ_wHs!R&5KyCO~7jI
zYKZm96iv)7LqF_km1q2wWI$iyRBZOo6`86>;l|mDKee{?Q@Ppcee%p{<3?6`7*YR?
zU9x(^aa%OZRhj?ble>`5hZ2Y2+OCw!yZ*fIv9wiX_B^0x(Y6;@@j@ZBg39z6M|32{
zXw<{qT4}meYNWRbJ0ui{4$zalP!?FOqh*+M#5^7+JCw+ImD4+bTK`yhs9&oj&7k@A
zYU%nzbCdJy%r7|&ifR){Cws!%2PepS1Xq&M!&^Kn`gnjkzC$rE>K$3lW4+B8uKNXR
zB~0e35BUr#!s(QWhtgBcbO?t(s%KOjKm8pISpS_XFWo@mGpWN0(o2q0Y5b#C08{S6
z#R{&gdVFw4R<dBfjTe0)>k30M1Zc2G$(YP}yf4}p0pv+v#<1Uj?9XY-UvU;m#N4NG
zxx;yAKuEJ{{rIU@K|gAP@9+Bj@?oYp9B!}o0CPZj@dyfYFz?;Flu2uJK;e@<unO_g
z84l<*P|)5BIlU&eGmF3H$X3c0rT(9RUaYcvr89a0Y$M^VJy3LS3zQ}NZ1i`>agUy|
zG@s_{5(U;(Ax22UTvtH9<>XTR;zoOM)Wu?Isl-mpL?41USZ)&lXJunRBJ<9}txx<s
zo#Q(49mOy2x8l5?PSFv`K&#(&%loG7`9CINS>OiirO)`}t<GlKRDMWl38Hw^Z2L%w
zuY-_qVgcaR3noo>iVK-x4v4<XTz_wbnbaLL3y+jt1}CQ=^7)W)2FC{m7CgD@xNY}8
zBkpt0qi`ZCBjrtvu=%+9a7M;>1v6HFQet&C)>b@aL7o{?dkb#zYN7J3AJG!7uGu-r
z>{>s6)M$#G`YSOpK>QXF3t2)9Qa?pYRg0auZU9sJ250RlfU~H2xvSWht?+i{GEb2R
z;WY449~m+rQ|$>2lk_o2cUjD%Xv|HAaseC?*P%y#J{ps!9uzUY4WS879D9D~;!elh
zk@ihjP|Fy5l?S<JlJri?muGBr`S~sR79^kMy*~E8mbJ+(LTJaNS$Ap_eXt4AH6AZ=
zb}L#6C>iP>g2z%v#tYn>z8a<-xT4R^%o<`9R0Fa-P&!yXll+!q)Pi0SJjF*;NJJP5
z^wC^(=q1}SkbP9MeBjZ0%?$j@2?V>P_SC&u_fbat0FFY=L+wE{V0+{o+Ch81i))Q#
z`P7+RMEgW3wn8g7&A)yAx9!xz+?qYkc7<k!u;+1HxDti%zTf8(Ax3$%rWm6Q2w>Y2
zq?5{Yxou3?48wWK=$c!GWWy7XoT1UbHou_DY@?#Bt1A)4Z{DBQo>}isH@6Xm>h$t)
zy7)4bHPg8sc)t-#anD%N35bIy%d+p1?=rq2%oC+$6zaz9DofS{@p)@$o=Y<S#ND}r
z6338*1A-Yl-+z+G`h$3LcxaT=PmSj??jn^RhpRp6D74C{HdaT8V<=tI=hc4Y2b5?s
zRyrTXO46D$q79j<>!$iG$^)7E+Ss+3_vo!b?un2dy2YR@#U~m;BPWP8O(oqTiI99&
zEztFNUk2^A>$?ux7#BH$A(UEU;``lMvHXi!6cVvc`+PPwGjc4PHndl&TP&gVg_g}h
zpjhd$%_i90kMhvVnYY-=I`)10G}JK1DSA^~R8@7Pbz|7<v~kX|TXk`FW~NfQzBFkc
zi#qFcuNfQ|_&{&pN#`9kX(_EY5+uttTBisY2{O|SEd3UQqUv|Gc(ExO6-c<mOQfEY
zr>_sZ0n)pU^{u6pR9mKQZ4>rk9^|hyCM)SYEt`J9G?!eqy^xd$HT%>?;60v$o}Uk}
zB#nKt!xzWep>-9PA5I~;m6Y{6P~njaB|R)#h}@y8HUwUo@Y57#T{A}<IAvycSAtwW
zs$UEA)(i~-O)kvxVa|iGhQvwNF>aIt5~Yi*JmqX8OBVYU_-*HllBS}ib3(_SOjN|#
zY^l9#C{fAptv^)vS$ad>OLZjFm#tQbHl;=8eJ54etm%dQA;<Sz{-y23m1!IAHV?u#
z<eBc|UpPg%Ityp`4(e5wDk)>ROc6WqEW$4m$s4!2>uNnV#j2Fu#B>4Ndt1Is%n4L5
z>px`ZX;6BH65kc3<C`QZlu^Sh3Hn7jY8XMQ9s$*NwCcvo&h*WzBeRD@s)9|SO6s%4
zn4$v2Pr#lGv2amK{t`H*b&ylqWstNkN11&J-4@?CkIJntn!<?{Nh^|;j-5vVpLM&#
z3YOwcw_EfC^}{B&EG9WFfvs1}_p=qXaSNIAdK_RW7&{gpQRyWMGR??tA0bG%)Vx-*
z`n~izno=*Ba{Ghs9Pl;7!YN?#PEKGgn(T8r#B1^!{_-KJVseK`jr4CsAsIKp&1s8g
z_5TQPBmu3lSfl1)qe|f|ZbDa67q~XHLgT?uY4HBFF{@Ht5Lx&k<bv`9-;y$n^MdDg
zE?B9c3t|rWh{sW@@d&WIA@lge9F7~}Vvpl-T7}BnZ1vLUrqeiltBa>S6fv8V=i;a|
zj5$8j-c6Iu2+ewG!pj8<bSEv{iNAe?NfD2NENPI(pHuc0#ms+}9moUXvF{~!y-2P?
zhq%B2KCsU2@>2IA-Q?Ti%WCkIA4CMF@e050XG1|Ua1w3J-NPtzV?;9_OzK9*hO&S2
zFu&^M`j*qxZe=DSYR;>A%K9OjLkzCI1a%HP4{Od>x_$`Q@KKP38A=>6wIo!rmVP;z
z32~9vHmT`2n#;2-iCfdkriGx>{Q}>)b$Y$aPDBSoNu_~(x^!8v42yRzBOr}pC8E4)
z-6mngw#F5)G-VgkjVDZwvfYrPXiisiuu%(uPE}6r;ZXzH(-67%g}yHhxA;3EL;HAc
z8UNAhDSF(9_CD!FhCAg)Osjz%iKR+K@xpt)$J@2b*0vPT>-|I<I_v9EGjzzJd2llV
z)?6sr_^F^@M63;6PE<rJ)~vjZlT_Y2YnHVNDpyP_nnUr+;t87S7s-1y?zJ^xT0N0Q
zQ|57yWaT#Qr+~oqLSoNQ!OEZfvit)<aO7`T-YQ~3kN`A8(EFFnJkpW5+w^gW>*OO!
zhEJ4?r@JIuJzVJ~1){sW=_|!p@0sv&n*6#R4Owl>{QSyr`z*<Khj1)tcn#k(=N9S*
z@3sX;c2PK<l03Ue6m0AyRcl5z-+g}qX}|CC$r)&A54u9FYT&psI*>4Nv2u29$0=K1
zO84fOw|<72txpN+)<e%uLzH!tQiWRoY%=SDDXjb;N!c%t1l8Cl)$L+<X<X@~j_ujv
z^QKIJxM!HD&xQGC*|;u)<+z^2eFD)zpsV#Q0!@JA^M1HvJTAbq4$JuSI_2DjK5vHj
zjU~v0sTy|HsX&%BwJo&%Lm*jb;#N(0(e6&VueZmH7)@D7Z41+|nY%M<mf3Y)uLe1K
zSED9DZh6cPb|oWb_QcL0zI!aX_gwSlE;<rP0S15SnP4|^o?$0`06`W_Xz{G#Nn$|m
z=h21#HU4lmzlRTkQN#>rwVZ~cj0Muxi>db7oj-OcX*pMV52q7-51!}qjx@ucH=(Qb
zU?*#vQ!Tn{c~N(Mve28;l1HSb<QgniE6Vgi&u%As?!~8<9jeRp<gpc=rFk8^Uu1Mf
zQruC4ez?>vuhS)OUssErxNq(-HklvyA!`vWl5x38kEf<KNQ`dgZkl)4L+i^);$Ji5
zxtqFXzjL+tt9cm97W`SxXf^w`dO4-{uf%~GDLPvDc~Qnkb&ch=F2zBCTY4@w>cx5^
zOeyh-y+dn9HOqotg{9B-y4i?z^H9`FF8i5bTqo`zuEj=9aE{?l37G;^_#=0V%xayO
z-`R{fpvgcsXJ~}_5w>H=6o8570>cogL?)@*wHoFmMQ5V6kJtJplLzzPROan3rcOXz
ze`E=seCpERuy3X|9m83IX|)My>taXoteNv)54AIYUBxD2$7d!3L9`}00{o+oQ46n+
z;cipOMyzoOusj<v$EpNMEudNaOKK_=^uI5&ft{h!0E|o56+&0_JRkwPT_IU{F`+&E
zTa!b*8Ro<RU{^`^s(gM{4a$uww%>#rawH)~StF>zjR}A)-s*n8*Vbv5V@f+RPw<Of
z-{oq^M*P}LHf5JrS#x<qN$NXIV4~MNJG$b^!=Dt}K&ozV#v1!jr*I%2EHOM$B~z8g
z#niGY&dOB0EjH9ORV(``--!nyB$Fv`>5J<a^myGnwoOOB1J0#XCS{#mx--R{qHmD~
zKKB1yF77S~^QvWqn}_=2>MR`&gL$X^!ISDloqH6FI4wh|FE41EK29G8hrGqbqTdeV
zG`Csez;;+|kxM@|RCL;qkadwtuJExwlizJBY3lH_0eZr0ygFGM5>{T>uQy(Sg3`jO
z*TLj_dvr6dPCqI&0j@KSA^sH1`IXfjL<fD054q;8yLZ=hUuv@JMD^d@7t2_nz4asX
zdA+upFv+&x6gQ7x9wQ`qM7{-Ih`Op$T}to7?Q<HWQ4wpo0&tF)`XRHUa+5$^`EwG`
zEZE(2VKt<@Fh(x!2_l_six#@mOd7M)a&8kV2blX(n?uU@h#ornsJb79u;t8}ZcrBx
z^Qx`Y#7TEX<nhB8COHm!Cus~kgt`m5ab0#dneyHLHa92z+<Cvjk19EZ(W0{r>?0Qh
z$~q0eD-x?19bR{d*12=CFYr|TBLo}g#sO8~Bf|q_4$tnt;997v@R#LQpwdmp<304p
z#icP$xkbvWE?@sTxcl}Ux42Xt#|@Gnw%z>#$C$f}I*V^kcCE_@StOlweaT<2@DEbq
zyTC7rNkIv)B$DaN{t+%7`FTxRv^-lyJpM$FfL#0_Vq-gxAJjb6G*MrEJ}3cdZsRgG
zLg0sv<H94?@Cfxm7FAcRmLA}{apS^sV{`E5BW_y~_lx;&S^ZG%Ci^SD{i}|c7e?#Y
zs_FxD<=pw8?IQfgBHO1yy|EOOZ=Eu`xB2X^Fc^7)q!~LgXjWF^UzBf8S*)|z4N!Z|
z!=3z;=;W{a)_4~O=<PhHASjomwr{1;bckXoM;2O1RB9<K?d{&f=}q1GaF7VWeLNCK
ze>Bt|LN=dTZ}`2j?A3|RDhkj^H=WEr*8G(n%PoQfh`6;g|8y%Qu1dlG$Dn?xOc?a{
zP@V0Hxx}+yJ4hziA$Bud_n5446MBsIg1`+WTSFzSo8uia<2JZ|%?jUG_15h(DmF53
zF2A9tF9b{ULI%vXjj~%go|RzX275Du=E992$JpT(-)XjCUGss9K+(E;&0Nm_SH4Kf
z-*FFU#fS&g_*S%hc8t8*cj0ej0!_5H^KK%6pF)2BYHoXHP@frSZ?s_?747~NoC);A
zOTPzybL&cPo*)t&WL2WyR&<V(yhm9Dn3iJ;iJB7Qzb{o1;=3!}5jIhyiTfFTGHSVu
z(78&gG%c7z90IwAFp=f^LhpHe&f_W^i^{g`%js1$kBR!k0nf2JR=V!kQ(HYfdc7Cp
zJ7ZKg**?2i?bYgEaq!L55Uh6|GgMP>6enpOd}g|_Z&6{W^Hh~Qi`Miqm6maAIpPL!
zKNahmRnQI(-OrX?xFY9bzY<%#oew7)AN-`-PL?9j%*JJV&ypt7Kq0pl$Npp!u-y`<
z+*>)5u&zCQ*<nmu8*qQ|TfKyOT5(>Ifrq8ORUh|R0Vu=_=L=9P5C80YBGeUGgLc`Y
zw-1gZ>!AC)Tk1^Y%kN6ZH6jD3PZ8V4n{c;tRaC3$xz!d<5(nw%`yG8dHtT@x+9W^L
zJ~^uO(fyCCRCz=xr4(Q}WZ(^j*I=HW7VSwFx213o&e*Pcty7hcs8xdnv#Iy2&mE+-
zK>>rVvXHN;e5StMuQUy~eC^IdM6%Znc+bQ3D;jmO5aDGfX>1(qSEH6F<+dSzn3LZM
zL%gjKA`e=!S+YZU+Bb^|y^<+iIAXdf+a)_4oEs8xY5!iQQ<C28d}$kVkB;<#iKEpe
zea-=#H5^8U^u^6vQF?+WzmGD0Yd>8s(>!9ksUs%sNO8WtvESKccJsRg+g%|=^v@IY
zrceJflSHyfr$<0A%u*!0C*TJZS7Y%fYB2$#@0*UW=SoN1)KSe7P|EXQwEE_oiHkSh
z{n>=NRj|61Q{d@I?V307NxJcl=wA&8TZxO^8hv#hIMaSP>nj_<?j9}uwBc49%(J;$
z*=v>+JF_@Z*+sS%WvgT-)N|D-1v_bP=ZlJQp8`HBZl62(J^P64_jZz|&n~%_-H?=Y
zVu{CG2L7xwQnA69A3ug>r8tlsOJPmIbb6ZK$cUu>IK9wi>60^M&u9U3=D62hUO&bv
zAw9*w+i;Qu!7yQ^sCvo0Qtj^bJO$n1Iq}dDy;NDAyr;epsC17m+-TceA8CdB>_kh<
zugW^tirAu(=H~Xnl3h=-JMdW)LrZ+OUg43Q&i8&s6CJnx^A8Z(k&^*7F2WgiO0Xb+
z#>{<E4?rjcO#P}Kx3<gRiwe$_N-Qbw_Y{Y|_?*JfoE~HGy#woo)5PGrA)6Fgh)2wu
zfcmI+DO;cd+I(9j^Nn}O3^>PB@A_}|p~X0utXlJ2GeGbK#S09ip6*=!im7`l5uj%k
zG-)aIJZF;}H*s)QfzY)q@dl1pY-JZbYak)N{(08e1+oh~TE|0LN^hf;Hw!JZ>Pyjc
zB98!7b){AzZ-EAZkQ}=`KswJy&uxyIkVD;+6YI>q9n}HAdw`efkIwG{gOWKS46j{x
z{YRWaP+I|*MG~KIMs$JKD50q0jEzWR@_W+WisqbLM<BD3ugy>&yfnj!KY)Hw1>*ld
zCT9L`p5}i^nQWY_|Ca>VRz=qSkO86hOx*)MbrX0LKSUrj;W|VDr>$11b(5N;3_H??
zB1HJ_=Q9>ktr{<aY~J2%_dT@p>!mCEyoq-`((CwoHg8K>e_is^dr#EeT~k}fY71t5
zHLVRteqR2%)jm2Gcip#ZwoR6mhvTu0QZ9Wdt|nh{%uhwOpa8zxzuli)R9Ag^@<1s!
ztSZyNVXyk&n1dgeIlPe`qd0}XjbGjNX=H@>Y#T7UOlAv<Ij8tS6+Am`Do$1TRh{t~
zz%ntRrY7p(Q*6G-Ej{pf=ghO=rH!*~(Q35J{`Q{dIlttdFstvm^rreWc*kGyRL(8E
zj$5<BfJdud(=xuUAp6HtUz_d6asd0}r@2{Xsr$Yvs;%YBPJi8R&u86Y^dqbG(Fb><
z?Awj3Q=St?-0rZ$Cop{JVG|JSAq;>Rhd@Dg1TO<vAnKly47dTL08)YzU>PA393^0T
zRLO(DF+ikC2gynaKq5U1psH{|EpV_Ch(Xv(M)w_N$f&{CCGm;VWoA%<#~3b5480d-
z*mA~H1Y?c@AetdZu#3@<qkIlf6JxGGY--q{5~RRxxKkUy-n|W@zWcrXu71v;RRk*>
zhLW?(>EZ#@NZ-;DVcw@l@S1jEVLyqYjBD+KBHd0=qAC+P&)81T$ByE4*DRv+PP*gS
zmsuzJKhJVvpS@hi-4&Ue<Dv0kL0Jr}rnKHT<@;1NVWB=I{;+MTmp<hjPag5EN6x#7
zxY7^in|o*<<BJkf^LS@`;Cv)8$bE<xBq%K8OA^#((u}i%I^=(N>3(!YZa>SLx5Jz^
zNM8%AzXi(gLi1jFe2|_O-tU6sz0~zy+I*0nI9T5d^IK|kmj1KpTzZHfTl>6?IZAr3
z+f}GKY83=cWMzGe?P?QF^SPLngHE@n_ju9&W-I9Xf0%ui|3Q&pX65*Q6$w^Omj6MK
zVB_Fm{J$(ntU9R5Ki~dxL<Iy$Kvc1qxJB5dC0L~)3}CW`SVn{5Qthr1&_p@PL?uZ}
zKueeeI8ew)?ibI!Grqgtvu*F+v+KVvUv9qqKThB#EDUH0PJ-*gO9&1v0SyEaV1-52
zB!B>cq9B3<2u=hGAmMJIKZ(geGldjz7{K%odoNTJl$fzel|hS~OgbRAg+~`qh-jb?
zQj-u;Qvrei2ndoNdSMVLumY&pAWi@mP(UpRj^$vSiXZ_aMFw(n7P{&FH~>2ccmWM9
zt)#zpZh=YIs35@shX6W^8yMGN(~tmNz%d92n5eN|VvzP6W~|s0M5M>pS2U1rPY5Cf
zIVhfQ0DFv0OaqXYfTErPxB&edpyvU*g}qqFBw#?YIfi(CBjXU_P3)K;AO|3I3=pWd
z0Zxx0+yV*!I`#n#W$=RdkfA?}t8d1A@NZVEfQEwm{^6gKA8G`c-z`j-K!WWA0}eDW
zcOZ@-ok9R#S6o3BuvvJ3VO&4VAYEPr4*v#pj8mw#am;Tz92f<4CXfKfpP!Xna1c=8
zB3=+)gSvh$luuMJb`#c)V5p&4SSWGtgug0z)UY5S2jy|7AH8a5oCDamA7{6aL0laF
z*8TlR6sFMOZQjDHDu2y=kz@bqoC1pg4hWE>lq4{L9Y6(`5zTXT2V@$gPi@L?>6q%h
zJ@iNcARI?pfgi&-h3)9=HQ;Md&;rFg1^>Ojy7#kKo&V?&0Ko--uL|TCIp25Z9E{5k
z^SJGvC@{bcKnDTjFaSSZzrIa_hSPw6?xVl9zuuifSW{GDZ?W%w?BDjOO-w+*_oxVf
zfFL3xf&hkwmWlx+xBGwd6krDH`l<oHuPY(kL;y?vQKPaK`k>xFt-iPY5re?*<}|z@
zaE%H8Z+igkV3END9ry=->@)t|-}dLfs3(5_@BXXBw-=Y6+S5PMKm08Q>>1Rr^t^EO
zc@{XYu81691AqUPaX!QAx)i`O%D4QjSQia&z*F|+>i+PEW`uFW0=X~-77)((6+VdB
ze?%G7F>tH`?gM^43jjbu1PlHa4m3a=2Yp$196Y{jgAOI0=2NU3AchXmyG2Gw0Sr)}
zVo1kpG2D^C-=U5y&H+Zh#TW!65aI-mbO1VDzX9VQP(JE?qa*=Cd<rN==qW$=iG+C+
zDRAoj?D5iH_22eL!h{$VAU<YLfQYa^8|YeYI>av0+6Sf@zCj#Dd4k0U+`(R^o>G!n
zQp>rxJs@9%DO>x$KEKUbZ-={^Y0mnxX<<tcYbV3#{XcUz$gpB-dr1Bmr1qGVS}Ilk
zUg0)u0h7C{{myRv%;I$*wC}lF0zPi@lv2P#S;lwmSr}gIr%!i7<!dqO)TAwxb5diW
zM>5B>_^CYMX&UA;O$>+yp|?ASyp7?7Uw-e$M`C*trO|3d_?-S~ITnrdr%e;zlO^**
zp^y>d<;38`q{FyF!r6CY9Y3k+OD%VDUVZg?LcG%XB>$|=i?yQ~uCUtYqw*FRt$wzE
z<@O_B*olJ?2y4w`u9dEON!pQCNst<IdOd4NV@}y8(Y)dPen{3!`>IWF6-P&yS5>_?
zx456-+0xa<vBE&vmQg4~r5Bp8N*bN0o9e!p>L=r(Ut*Q((fwFXSJZuIchdNqSe_g?
ze+o{;b{%P^*KGr%ArA9PNyc{pYLJ6b^5Vily{a^6Lx+C!V3AZx7%&7kx`q12?)rKd
z(RY%BPVy9x<)qaP#B3|Ah&dNFi|&XD47&|}COs0$z^W5{E0Vn(@{&;V3dTuXFP}}M
zM-oQ#&3;@LUsU^*Q46QPzk9;mW%&2r$h-7#rTvi0o7xGC)-X}(k0`lX1X{n7^jtIE
zq3Bmir0gp>o~j#XP~+c>xd_W*=wR&Lrs}J@=QBCrG@NR1uT|r|Oy}y^qI-rr=F~34
zJ>-lxlXF;BL%#1NP1-nK_5;0fw0^9@s=L>l>!S&c()aeLQ-~UM^fN@6h3N2y&_K@S
z$TGb0cgV3);39r`cue4n?saAx(^ct{57`vj3GTDYR9V5u3{U8Xa`$V(;%WX7i7`7_
zPdyvX6A$~>sPV~njN|{J(pb7@ptW0BQ?|<=r8L`bIQDU}xD*%C082Hcq@!bSWTIc^
zVNs$8JSI#_$$A7{tM#`d`{%ovDog>A4%tcxs#giNfvG}8sCCLqdz2?~5)Q>|hQ_U4
z1P3Tsj1Gn^?l`7l@~`=N^Q4Lk*%I>&vTlOTw27Fif-v~iQ63-;|Hw>&PZrf$(T*dV
zc54SRx|eW5aMSiL%5WLj>MYJYR9t;5KYGGwY63$wvb1ap4!9rfUk%mQxOXh7<X?cU
zWccD`JBv4gUtRV}+oWW`?&qo{=U#T;s9qS-)PV6L@C`+!)3`Wrwj1D1TJu)a&qGP^
zVpSs6Uxo}+uyTxm=L&RhlGUMyOf+9~Gefb(q{ro(LClQzhs{Z1@vSm!$!OLEpzFe8
z%p|uH>b1)>M3>TSOTwM$l0qbuXc$kIPW3-}u%R`59!z|}RdfavkbG`0uMvUMaTr7|
zUNcP8(=>XRSqz)J5jCb%h^xoDrjTU*He4O~-nnM>u6Pk|IJlUuf`?t3@L`vaWWi!h
zzqR}rqOZ*%;_aGFT8EXzdo*R27*%}^Eas=+1$g(OFz`CsgWZ9)Zw!H(0o~)7j=eMA
zHtEcPk6{!ygRwXjky6*iS+_Gvh()z<pQ=<H`vnBDoNtHlq6Q1w$wi4emTlPh<<3^P
zVRWg$XNh;&E+nvJ?^cvK+0ErwOkWut-_y0#tvV5;9$vA`sVCE4^x(Tf+GDXugUHQL
z#;MDnSi)#ThK}^9;T~je;rYuXo%bUWp}z&0Lr3(;D1deq>x?1n3Eeq6Cy}(hu?a;K
zT_7Ff_1Le#C#tTI&`?=0DNTM?`BDk&d%QQ?IFS49nrA$lDi5wxJn;~CJ73LwIM;yh
zG1)T}R|>eXP9vqhWINw(={{De!ST4yqXmkb7YwuDy`C{kDt0knNIjAtdg07thQR*Q
zrK-%CR^Qvc@7BeAj~Z`+Z{n6`-$5h3%TM4UTKovu&vi#cUw-N(e(1flmQah?7bK_5
z2EUq2d2l`s=c4EBq4vGds9X-fAE?=mXi)Uh+Olnbt5WKnO{{py`t>@)uy>Z<s+&(n
z+EIIsm3)jnelY80Yp0#c7DWlPJx;B)XZI)yZHZOV0{SU<<aQPwF%1D#e0^^#Ophfv
zYF!9qd00JXP&|Ogwfo6X4Q;7T#)SRkEMInE&7FDEtmOlr_n9D%YUjRqBXasaJ^Ncc
zX9<*vp$*F;2?);`Nw(X%x=g|fxAZjYcAbHC-k4q2DRZ<{I8lvE$59?7t!9Kw1*40y
z4{xAvOiwg2tYvW;;YUMz154gMQDxx(@Miy)b;mX1vVz18>T{=##Fhf3hIFqT)r8*m
zR9WL20d657k?rkfX6f6j(16~m?ZwK!FI_J#PPPU86`Rc_wmvf>Bd3y2b>7u1m-aw1
zFBTdfa$EIX?ve$~R)^;M)7lb^xBWfoBl@nyStpP4zBVCC-!JpYW?=VLe0%(3-Q;(`
z<yqGoV)I)GdbYH-C#2(rH*ZtLQDw@H#IM>Oi5ELqHkkWk*YDinl$>Is646Z@@N{Th
zE-2gB^bDOUEu&Dw@-cY`wqZ;#JWlL)!K37JiTFdksg1nF1%vq{qStPD_b|Rh6*C@=
zJF<J4ZXrQvaHM==`5Wi7n%GRJdYG)Sa=Ge7_~U9?M{F2Qr*M2;h*O<VE%EX1`Ia6V
z_fUqnvNcyO-Rxbm*Mwc&W6}b>xs=7ZRh(cId_((_#+;M%02%W4m&_7d+y$E7vc%c0
zI?lNsUI>D$#2uQ6r{G_VA?NOv7@|@VbdBpUYlKLA2^yHU%hU?MiBK;QaN1E_ulHN-
zXPCrB>Q&&d%S@ya`GII<gS;mtEu}WNx!kDp8jF$=5~&gieI0c9G-jjhs3YePHK<6F
zYyppKZ}oq|dGBw$$SGXS`?E5(yo$|yZ*@`7IFQJYK)PAC)AGR9G*wNMNB0aWI+m{?
zFK8V()wwqJQmddlDvVF&t?){84Mf`oANrS%xzL$27QfJO>&<~doj=0Z4@x~BhV?lM
z8D&4FHN!3H(X^9b)~6g#xr@X{rz^u++F9y{v7~l~>X{$XIr}o~v)DJoRPTJG7d<s$
zo^WmjZ#{OFf3`C$`Shxx+Sqs34VGi-i8Wq|VE;*M5=lC>37RFvU~K3{f9N286h503
z-v!?hCc2m+ZoPBbGWT45%{tZ>i6lV35k4+9%EbFql;$_5bDrx{^26EmbkH;qA|P#K
zZ7x}Ae0sI`ResFk0mK@+8;b4JuAr|65q6b`+j_M!XRX<*ivKoUuDCtzquj(x#{~Xx
z;rj=e*lWvo1RTdgFG~`4%;BW;Se#<u)2lawuOdKaBAwbz(<<IiE>dXOsOYw#rqY<(
z3T!AGSxiY)RSgyXUN5$ixQe#4p}wbgnK5Lw2MAksiK`+%kGe5svJkQA8w%GEuvWyb
zX}lp38jPk^9eR?P@S^5z)d#$r=$|`fo9EWN<zdna;my80KgjIXvdHW^LC>ijX4D<5
z)ahF9AL_2H!KZ@;e9QREg0!RC89eu9J2BGUa#HS*(>Wn}xpu?@L0BdqpWUY^0l+l0
zwC|D@Uy@!4_RgcwOwvVOm)2K7OW_0y?KV5LgXq3EI}ttCyGD}-!rHDM8RHheYo`i!
z?tzQ6dHv=|w3vQ0)R>Sz@vM2uR%Gm(ypPTg{h+V?(RXBQCNqaq*J=MwqAVQGy}gNM
zk;=EqIuF%_3ItPqdCy9v>Z>05kH7Zz`Kdp~o>KL|#ym>Fo6c+~k2V|ic=3Ex<#?qZ
zc@q)xm)|bou7YOim<K<v$04lKzeGzEzT3XnR7HH)^Il;XB|Xro*@slR5<cZAtn3vk
z;ay6v3{`By9*;BZ-JTLPHG=pI!kF)#Z4pE@G{Iy*%Ts(Ht>;Ck@ci5DlM>`SN_<Lw
zbMgg=^^kg;o2O$&jI8)B&d5GrTkVD>@Vf%_5I_;VoL|q1Pz&m_DRkqc7&cB=Wxd$z
z$&2^oUf4G`@}83ag-ke{0jsiUqq|GH8p^u6hu6S6*H}ryO`O3W6)jXAO}ji=JC=2Q
zIZ&Hwckv~%Xb*8rbvw+%h~_%_aj7~hZ@Qy5-Z9R4#60HxR$kl1RwB8RF-Wc58sH&d
zYyBPrFD3Er#vH!%aw73=gaBde5H}yT4wSVAsTuU95;X8We9!_<F8N)nc%v*uZ<LOh
zyzaC_BI4iIlI+fHpr&<su2yVnf{h=6xl#orj~z1U#R;0nL=-V5jMXpVX*_)#^7J}L
z$0P3$7Y`Ja@f#GDwpNS!Xr`E8=IgZiRr<&Wpwa_uvCT!~_@MLZ1^#jJvp9KHZ5n=I
zwZ<=qEnXy9EItZ%jS%h1-EwQagURnYIjEwmQu#7++SzomYJ?eD`*<7V6QU8AIp;ey
zxw2ONoXdr|q_#wESHF1z%eSyY<cGbSNdD8d{4~SJp$cILjZnt&6yoS~pE~!K2{@hZ
zajLj_eR?u^^H~#}XB7FRw+#%s5qb4gc>2b@Y!X7OU^J6P<=2S%u>7DMQ9Iof-|I}z
zy^^)w_*IwS>N)s61@r!BQRS0|STbrhEq{+8S~N5hetYAehF8>x_vW)!l_T}-{@F><
z9<2^v(0<ru*${p<1<13~Lj(rv_2oib^1`Y$OE6gpFC3!IM~JG1rbV!XI<3Sy(zdgh
zDw!(A!LGO&59vjgw&0>Fl}T9WIa3K%$Ucms_y(wN*6)28+iZQzl}lm{F?gEMi)fPB
z+A|%dxiV5rong*b3d*&&%N4{Dl+tPZ#|u(j-amK~g6hs1w-Imgd$F}EEoJy&l`+b2
zsbzyqaTYpE$TYSsDL_8h#{uv8oQ!DAdnP<|W1>RjXo}gj#l5Xri2HUZbacw>X{nqT
znzRy6`20dk#d<cM(N8Y3Tl~HpDlk^uA!io39LxUoQJ2|wiJx=EVRwydS87;d;h2_7
zh`lmG4?*C|lB0$~GaRG9d&}&rI$5CRSF{@=IuT?>A9Q<7NjDce--#JR#n1@b^j7wo
zv#L`Z%Z|8abYjW{r_<?a<nv?~d0R!&jdHe}Hqe}u>~l^0et~^K9t}=Dd2}etidEQ3
z7M*<eB|Hlx32&ExpIUDJlzFs*rOkL@-x0Y)iq*j>Qv0Nw{#1$q9?RsCn|_^bgBsF}
zV+>!9J?W^h0PCv>LG(r81lc#l`CFXHcee=sVU+sz!S80xcL}SXT$XNQLSOxb!mv<`
z?d*}kYp=9bxn6KB{ARmpVN*fczmU*2SKi~Yc?uRQE|WQVx;q#$s2jW>Bh|zJ8@MYP
zkl11;z7QpUGWAw~il<l=QiHqGhF>!JM&&HY*xp4y+aZx01=>l@V<l&r4*rv7Bl#@*
zNaiegXCEiM#b}KXuE%@Cz>rTFq;(y3Js&&CT?M#@e!<K*MvSWxjlK=%^BEyszPs*Y
zQwUw}&8RAK{SUJ+cSt;Bv4q)~@QtiCm6Kz^mUPw~I~xs~hq{@;+MJ%_V|g=YrN~>Y
z!~E$wqrKhXv+~*eQiA9iGMIyV%JvmGeUl{xX^=5TRU@g3VeX6=6g^2jgdeBcd;<5V
zjwIKQd16*R|3&66F@k-xD@P=d3nz(wrmJZA(0STs`Fxdf2YYrV9UK4hT}Fosk-fB<
z>C~j+e-ZW$!J<UV+U@e%%eHOXwr$(C_Ofl;wr$(CZU6hch}(JhMBGNTs#g(NF=vh(
zU#5IaSIZH0BZAZ-=?5kaohzz(b!7MKb8L|FChfIf^>b4U&eUJ<pwdM-9OT9!jE9d$
zY$?QA6@vLbXDK&smT!z?%Igk-c;e~o>z3WX%^q0idB;8bkh7Dev5MZV|HJf}*V?x8
zVT!pAENHy40XLoB+C)Ye`?%W#`_=m`4b?^IC^Aj+H=$cC#i!F8jGHzAOpM%I{ZCV>
zM#j<+!Y^oNtpT21v*KTzyI50g&u?ySkRqO{bd6G!_30_SCl=5w^*Jp=);pacf^S5^
zCpV|j-54NH+3wf9o<)=wh8dpv%N)GZqeHzKx@^xG>HhwHad#2C!!79c6aj&0C%CKY
zo0=$ymibsPXweMe!fHmdke#I8Z&2MwTCU%#gN+)>gp&=8^{$AhNnl;mmEf@ui3*X~
zibedIdO_&KBJRy9yY71J@yi0?PZ2|=#nBvd<wy;N3BNR647K%3{%48;Y>Gvd$NoEm
zeK-@eRNdp9<?mf7biP{}$tG^DIjbpO6SRa{)eUp%=wFShM;WaX(WA@v;nqmhJgPA2
z%Z}B&zApryeK+KjbK*?85&$8d6hfbo*0$1L=8&#)ROBj}f{9{@(NTdc_tQmlkBMlC
z*P&s=wt<3G^P+0NE@9>ON?ekZ94Sbes(K@5*wJQ%I&-W&dvPkZ$$Or1#TmA4gd+;8
zEJ$IcDwx9#7lRb<qm~-PpNSq(grk&(ef_sE{wOTYgx;{^gV`$%AA*_$>or4PLF+?E
ze#yIjg12BFI*x*ZyRKIw5Wh=TL~Yf#yAgCh4xx{l^p)6>tE}y&yDBEuC*x`r_$9uE
z9Tq&4FJ|2k_s@n&6p*@K-W`#j16X=o@nhthv&;8d$$30!0&~OimKV0x_FBWf!h(%T
z4xPpJM@H#+b$?aR-4d*<pUgbaUMcNehI8~w>6cTl;3Bmd8@QDsSq-VZzA_Qnco^bp
zIhiDz_u|sQtBS!nliXbtB&^?ap4zG9c;jEqz{k*H%c3GeG@F4zf{SPFPw^~svNF#6
z=(|L1(H$j56EXv`Pia*P53j$<=M?&zk?WBOaXq5f-!=8iG~_=5aDF%P(Il6SWV^<L
zZWoTJ6ga<M+8?*EH%Hps+o$*EAeL!(8xo|xM9exwO3FADOf&rTG23IQ==-C2ZoQ-#
zS&#;?ZPbL7+woYk`4Z`Bw<vP;S45SM+U$>>w#m=;lA{-}a3sO%GAotr6XqC`aqDA$
ztMm#@@sT47pXi<hG$>zougrdDVAuJkdh7Wff{Nkl1fg{`WFkaxL5JE*aNdY}))r1`
zjN}7x|8<kiqSU%@WV^l=nm3$qb!dV`i5IiZiIh76U1#3}Le4<YamA{BmFY9_G8eJF
z$ud-q1e*~Y(_JjVOuDESrLrn{XDa2&Ga8gPn+Va)K#+W)gZR<YmAz<Q`Qk%cv^586
z;Ov%gX-IlqW<JvGNJgy}fU6&YgJwq3@1Prb2vs-SpZ5#vIBG{4@G=IhH#62rZ!Iig
zeZA@rCb73SC-=rKLy4zGbEs~gcm^wq3en9Td#h@5ll?l?6EQdEAoe1>*k6rm>`x0q
z&%y-}7T&n;IxxH(Q%j=;sKh!EBw4kNu|c?lm$hnnvJF9~iU@z4K}>4dBVW}YPVcr$
zBfJev4;T`sh%#Gwl~qvVs76C1nqa@yMAf@?+~B+)^Ax#<5h2V4?`NsVw!QBpLga6&
zQ4ws+ghcO2&N_=0#Ge!)C><Jx^kk@ecg4ybZQ>Wc7%sNa^0t?{F^Wts5fpoN+29b~
zH0u^$skqeE4Z2HdpEbKAt{+-g%k9Fe0<wFoK6+f|6c(f|Ac!AZ$+}>+A<-1ox=NO;
zjc(JKj+b|z%euyV2srvBm2HKk)z*u`U?FTZMbEn%DnNp1vd_Wi!!%}55UNx{sfau?
zc70|PYxCT{vH<8#y=~<PEaEO{%7sv{7fVue`d~YAZVfA&-ijO=c#BHr0(Dw!RyuD-
z<q3Iv#P7*j-BQj<N+RkMOF!R*%fdo@NOL_uZXQlXmEDDxkq&1}yEe!0s$phEjS@?P
zlbm{<vbpDheQ2xUaM3Y8d7Lpsz+)R)nWFb8oY*``@HpNAw%R&o3ma;3`OcOUXSWA^
z8z7FF;`1U}K1`8*f;?U(ZakhP2@f8nigrp_-ZBh!uEo<oF|!o@vRrmhHKEal5AKc?
z$(Vkr9AEfJ-%KBfvG4Fe9%QAndg110$Y*8AmaXD72B1<GpvkdwvsSJKQ<rafhG|>E
zy7)j$tNpq7H^BO;rDy|JXAA?^!v)W_A^u5AdYx`u&m}^}gW!iq9Y@foT9;}qM&g^u
zl?<zqN#g-m+X8wC6v{Ph@`W~D#XtXFP#Ww1AEn{5F>(AKl*~ruKNuOg>whq^t2mHK
zknyR|LEUDBa%H!mVMjc^00H*L?dqYJ{rU?4CSPo>*;W76x4VX%^0v)1vG>u{a#n|v
zUb@7KlN0pK%|i_*o;24NV+rdsyO*aSosU}ebl***N{?mcet&|RO*7xBsj-jT!b!IU
z59nvzC)=Z$BF=|<0}tD0z6BZfQmg5j)&yJQRLa^Zq7#&F=FgJzH4;*6mXj9w2InO#
zy**?J8iF<F#0HDawq8;ljWeiVLq~bY<d}~dxEG>&*AF|)Nu8yePF#ughb5Vu(C_{e
z56I$a53A2&Pvht8z}3))H}?&UFg-4HY85Nj=IqB$GY2~B_sDkg{fh&$bVohDjEY+3
zqJvVp?_y6T{;~J?LW4F~N%e20l=-kLn<A3bv6FW&r6yQ8G))Qv<d&gUpK)0^eQR)C
z6MI3xumC1hB+@Y?D!4_!II$2B?VM14sHV1CeTOLMxVN(M_x2OoHMh1thL{nB2Q+m#
zEGYNC{2{X-sJ$@Py*>tm)zgMXEvRfo@zq4>AqHM_GvBU0_SZw`!viG0qoh`ryP!vW
z!K~oF;9dz3b;Pt!5U2i^7<G^MOrs9C#<0}^H3Iu_B7!cU+@;Mihh6tu4rj2C?g%rP
zVS9RUwF~#-N+;}~4#f2D_MeghtuWt;3oIvn5C(6s<)uJmML7iQ1ljQ+IcJD~SrWoo
zf3nK(d*|SjkDz#rWWCMB9V-HJ;rYaJgPc1$m9v}=xRdYN6&6lKgL&C3Pn-$~bFol9
zx5|c2MJaNzL^KrU3Tbz6_+C8}hi!6{Ju=Cr?H-GUSGg_TS>!mMJQkR8$#CAe&(}^R
z$av*EZyt-#c;z(Txh<|Ai~haJ{^UL%IF(4{mBV=Tm<XCxBK65>Sbv_`qTBsmxM$r*
z>E`uZZoj?uI?K^~o!bw$(sc5Pj@?G6+#2OMJmdotoi(|G3k@mt_%C(x->iHl_W#lE
z8Sz;-=sEtkmCwq~!u)>}Q?xRuva(KR2{O8{WCs|a2%q7ZEyy|TxPj?ke*@Dr3{DV)
zqE3YlE@ELxL69O9<)jo*LC&X```_K~AFtI8x9sK_j!&Q5))(KL=}8lP>O5!>j=m)Y
z@vOgf_~Kw?6;{?1Fd+WIpaS>^0Ccqw*r9*F8GzlV_0g%ZgOlFv092r%qXsvZ`q;T;
z(SkuNJ@Eey6!05BU>HbX!2cl!0+RTJ3L6*$SnkWQCzn@)FAEkNr~}YdLEZodc5@ur
zpZvN+8v}0#MNUacIr>F`Rdx+D*n_7>EQ1>17~Hx;A<zezLkIOAwCDX5?<YHr8q7Bd
z3hL_W0_nF^7eF47*hl4?gNkti!|bo4k0OJH{NkXOhr0s%<VnZ>i)MZRJ^IsvRfz4N
zs}Btw#)iKKkB%4_d&e(mfR5k8KKG9~m&Jw<_yxN70>AnDy?~KZlYiYh`xW?+1`zpy
z1`@z0VPQr+`4eUTjj;!TE~~8}3VR&62ksX?lfy^%7ui4T&(UWGy7~+Lb%717ti}ow
za1Z>00SDFJH_u*$j~esFQ7|{l{HD_r7*tWPwWTkPje6I}$HEQ<)^FDtJ9nGtB2v&5
z=j(?WV^6$ve+n-T*Az#Acyj}|sQd}sp9$c%nWfL3$1biZD2NRb)DCR018<g}doJTW
zSEnbtpKJOB0_bawjCmG@2ZRdX7PQX?z(>7?1U6*Atq<t?yL~^4!mp10Cs@dyAN-#-
z5AnvuwFB++)jfj83w8t31hn5y1`YD<{qt=WSc--mEd2Hb^Xt+hm_ZqWnKI$`o%6QK
zi-MBGHYdj~j7g49Lko(Gikb)@DJcmA{F6Qc0R6iNc6O^SQ$z$w{@tnd#rj=oG|LCF
z_em6^)#rDzAhfRm1^n<W<7gGf{wKf>;<vZ_hw8}B=xvtz7j68vk-!BG<RCTkHucx9
zt>4x@XciA3rwNz63&2^y9RvRt$rARHMpFkJ`DXtmS62oboespL2q|akKSDl!{jMv!
zD$Wf|AZDKf{ONAeyA9hn1q%o^1QWl(KHn{>H}an+Jl?Jz{F=8<;K<WsIy7jyT)&m>
zz@WT5J{M9#T0jsZdVM^AKXPOU4i0&_^ty5KX|Hsoz`-E#<k-9*VYv(dXJG>XzPOe2
z@_(ZFAd!*(jgy|e!3=i-{*nXU_vzow=t=jj{;qzc))>&+p@ZD*<3JLA&K3U{Y&ujk
zR;Ko)?oT#8;vn3o`0O9FVJWry&Ttl`V<O#9zRXdU<TD4pniw43ereeG@!;OK_};U-
zlRn$R#_Y-N7EHwO@AQqgqRG<3l1B_%C^OO+#+!2W$7N6KL=0V+CcJk(%8%R5Ctn;7
zM8aN8@SMZsVb#6GWl$xfpcQ6q*9#z<j1M!#Vl_3D#1s#!7kj0dXT<2ji}V}CDsL2|
z%Xy<#8#00M`T0J-ndoJl>GrtFMGLE-<2<tH(bQb!5BZvY1dVD=s|R3F9)wnyP~5Mk
zM{W(+PQh<V&1A`(^m^ya(Ot}+6_sL2=i|L~j28kTq>~V!|5cnYIXJW?y~(j|E05RC
zxxm=y6x+4#K&Yjae>CEEv7O@-2Xm5bPBPD#nd;;kO3Y70a*FK57v($iam&*m>dzt?
zA2OH`9IS$WYnzZNnLgM5G0u^evo7VW=V@(oyzs4^(s?-%m&iUa8wxMKQk<r}mFn@~
zK{{)Hp~#ZpaTPKt=kT|4<H;UgmD}Jg)(!@+|8CeJ@z*qv3~5uqT3qjsA1BR~(83(Z
zD7)ds{8)#Z<}8$wEIAemeV*lb)}vCo^+7>$k&?Ury?lK11*gz8NG0RNM*MkQ56Kkz
zV`cS9$znQbZcAaPbb!%F4EfqC`}(KW>n~{3tu6;eZ5GbO*2QY?Z&U5IgX^x*p~l2z
z7re2JGi5Mr2k(*>>_0C*e@SEUAk}%+Z8$4$Pc)snZ*rAf>y`Y3W}mH2I*-mtSu>VL
z_5#?!rA$eB=zSawCbwsXhZQJfCT6dAE1*QFO&4}!=JjmP6VgpKkU?!l{lm4~hoV0^
ziX2b`wN}8hRN!{|UBjluZs!NLKd{q!okAMRsIp2yskpqcb8D3`Y_vJO;ElE~BG3d)
z<+^G1x|Y(hto?ztX16TdE&Rus#+AR{rFXC|*PznX@*sZsbEL4kMgdh5E@;cPdIwNs
ztNB4rWjV!wnAFAmY!MVt;tVddsAX?HjPTn8h$^OY1+WDZ2}=^tw@dKFxG9yY5#E|r
zN`u=#8^{w6z0*NYdj2OW=4ZiW<x5>XJ=xTsvU~vz+;S{_GZk6)eDE(@`;e`RARSET
zz=Q5&b6E{%8qLcN7g~m@ZMJwv;k|j>!Xd0*>WaP@Q5b=U063UZTgxLdrX$Na>EGN9
zpHU&ljT_+sA?WNR_8v)ndfARuBQ8a2CO;!cY^paBIVLM1KG|oCH{irHRfPlr9o|wK
zlS|eCC*=bfk;DR2p)f~?Q?}xUGSfpSS)Kj%Dn{|V2Q=}8qk02bTx9jhN|f$IQ7N(<
zWj>1W5D1-p-0|&nhnysfeQN_sl;Eh@q2-KZ?c%12lJo-)i$(|T=Wy=5l(PCl;0>|Y
z(NK{`HHA~;aU^e0OhQvBn&;MdW3H0psr7mZO(C9nX&k(BN-^3$jJ5sC>=UFbHyFB4
z+;uvqIjqbaC`$Flv}F)6o`xIHbj@TIsI{jBB$fwqR}0TR=PWspZ}R8jx|mCkKfYd-
zHGU+tj5pt@!9X(ac2ZwI*eO~20z;w@=UVyFG%Hcm-Y=9iT~0I8U|*b>TSFwsgj-5T
z&db^lyzAd`#@K#Sfp72U3tR=mHkz&!Ku+=(E@1~=OE<~iK;ojicZ%njbRW`tD#Vko
zf*W4<Ae7&{!w<WaX0WuE!%edsZ?0BC&uu+s=NS{cjPKTiu8cW}2NP3v6a$-MAf&n4
zA^3RB1@0}Yp&WFlp?>do!#?e_L~Ipj`0JjhXBcWy1aTzdJF|7|+4&MG5DMh$7PurR
zy2hw}T$+ARmnl4at;*MvX->PR<Q#eC0n#gCz$+xj6}V-4;aid8UdQhmgn-wyo9?rC
zpJl0|F+9z*A`PL31LLxmf#;ib)7F2{vY4Gtqvtg*rBEmPSfE{^NRNRobrs@DkEcB9
z-fCvsT+Vl{vVKk3C}oy5imk#2g}?R>eFO-a0|W{e6sEpP<itKJ(+CRI*Dq_u0?cFq
zqf0oAnp3EKXgVwrv^_*rJuU)mb=9uFNKA=^1S?Qj^=^6QG&Ous)Z0xA5DX>r3kKfr
zG#o<xIo|7C6IIJSM9bcEZhT}cN;`foHO$KwUgvvy^a9S9#iU+)%UjveMY|&%e)KDt
z;<%9<b9}W+Xki}p41Ub-f^=ZRLsGXe&ZK~6q4eGbQdhz;D;jzNKg$w?K;;!Y=lh$z
zA14x5xSDcnmzKnDT1}C=8lL6@jNOuh_}ZWH_^l38xChv~2lk89*V!Gyc)v$*gNFEb
z9u8=?swVE+oRVy9Wq$mjJk>S|xHwD>Tw@8-#t>X%S3=Asn|`BBqnZ?nn)?|pBZ&(g
zX-a9e)C6s=bb0nZLL~4H26atOehTGEWvSV)j=BOH&E20M9)&nFn|W)DrM3fhvJKHt
zCJM;WRug2dOL~Q-W4dK_E387B^Duo&x(Z12sb=uu0m`cAb72tr*CV?FT}th|IPy;9
zRG7*r#e$Z|Lg@)W{TspWeQpQnAa3;(j(b|1d1x2x_E>DqlorSRk&)6JSz4`|7Ot--
zs(RDsNWM&a*<4sfkiQ|~1<%8tt0{9d%sjrfgFtw6)a8f&eO+2YC4zDummN^sN{nzJ
zaOj;3;T(?%>l7@R<%%N~KtX%>7JYrm(teq!3avz=a&0iv<>i$AYFQ%7E)jtT7v-w=
zB~&F!j7`*|8G66%J)SwS`%iJTN@q~SC1Ci}3cJjVO)Oq}Dsf*GGhs8!kADG<i091s
zi<Z`7u3S+MH93_@#Gtk4E+<rI=_H`?`4Gzz8}~c7)8hQ5R!6OZ6Nh<@cs=ykBJs=a
zn{~1L9QBQA#Ob4FB(<Z9TRnE~gAPO_E0AJc;3$PU&Zg}aB~J4-m&F$vVQ$&|or~VX
zt~57}+oA=4J(kgujIOxj%~TMP@mFNrxYeJDF8-ZMS^gS8eC)Q%y&b(6tc=!|H2p^#
ztx#>DqHAR}37{oKemIGDh?g%6W-{D5*3bq2+E)a*W@g0wFg>fs?F7<}eR=f*f`Vy_
zhwiG`B<M~>ji($dt4nNb!*}!Yg=}jitl~dwplX&aZ%tHeMy3kVk<d=zZgFKDJUE<)
zEfaa=K=Re)HZ{a~!LGLH-X=U`_oyG(!o{BsX8bm>M2Ixxu3MiO<fTs&X~E^2P}Q=d
z(v8D*IQ_5w6*_7wlVfxYErleo1G^bzwn?G7>x;1XO8n4Mn<jOd?LXp4R{9#wdYZmP
zI4x}x=i)N66GR>dqvvl$=$@}1QNfXdf8ABDl&{$?GAH6fmaZ4Wiq54$2721-<34@q
zAqVL^>qsph`1|69f6fOUtKZWi(v6r7#gGIvi~2bx9Fn@uXlyAN`y%1UTv0jTZt9{O
zycn2l!p05P)5S=a!w#7TaG`~%KZtuUy~oT05lOiE5qvEr;dHv`diq|9WUoE_wyp&u
zUdc|jPOcv$GK8pWH;}L}=AGKl>LpGOu7bGqq%2d;0gTm<Dd_9aa<{_;J3Hd4%6;Oq
zUN+U7BAmJ)8z{NKg<8{yH9TEm)6(Dhe5e~Im?`0T#?m55Ax;81-1t~IwA3vb?8V)_
z4?Hfv40u5aY~?MCs|n4=SFY-O<^p8wA*Ram>vbtoBMG)j?`G9*{?s?>UNHGvD9g#g
z)DFkiBv~nY=p1NMIy;`{)r!{6Z#iJ@jW1U5iK5skf5aHND#+)IJgxG8#y(U|M?yT_
zFbJOC&N~%nvRr`cwrE@B6=t4Lb8^_#mNZotNTx2^m^j7s_}=mp>92oIxMNavh)44(
zQ3**Vh7DB(jXocK5EFpH<dt6NccofLo+uI<WpcW_UnF~2Q6u*!+4XyHGS$BI7sDF4
zs5b%o*|vwv+;mm9ZO<Rr)f5}|jHP@M^cIlkY7(isgjU4mke9C}CCUU^@%JYR+D4De
z?jJ8+z5sJlslmKI$hm8m!Gb#nb-hMe-i9J?T=M+8#vYuYJ~MMWdh^@o^QrPA+kftL
zIWoHpm1a&owFAWj>NTOT6z-z)p+^U!pt$itcK~TPmg&VoT)EKAuph62N>qc14Kb|1
z#(`Z^%35|$t^48*{NOmoxn6gb4kw$!HpM=xqUSzh!aKxW*O^Iw(5igTOa@L-3gr#O
z9_)}(wwFXWhpdJ2GPb6@M}cGf#+WPT5N%=2|F8s{zb_BW8jEfsjkt8nN-uiduL&nv
zbZjw4*NrfD#$*j@k47Qio5GK_l4AwLgKIC#r($Oz8nZO$+K(FU8~R9|FMFj5t5}Mz
zHji^of&*%MJfavG8HFu}#5a}*a=?)~gllbxLvQQXUL7YTvXu^I*lF?3N7r{|=rt;Q
z{`i|--J~uCw%rP98mFd`Ekeo@?5k@mCFA<*N^<nlvE&>I!?8BZn;-lP+>f!sL5y#R
znh^{JCG}^|X^0%)k_h{_jF2IUZZCvS%f7lV-w|$9%6l0}RiryYxwJ#s==C$gJ6p5T
zf!kQbdk#%Z(s<f4PO1uh4XU_A1L*t17eb$IC$^tmoT0rRFt>&&EidGb)qfE1Swhw0
zK*DD5k+}FV(UjaO&eC&+9`xLL{jqIert%bxeAJ}r@r-IR>iSXuM^Bd9ss)`noR<=>
zb5kV&Ht9s!Ix3^>2#}(I*isU}dELLkg=0u5mC!QMBBp);T>GlH0upSfb@aB<kO)Wc
zw}PFeB<w0g`=d1T4P{|r+=f>EhnJA~i8U65iK!uDv}pA!&#X$RlJKz>k}v;pMz`xE
zQQrH*CsBB#fYdrE+{x)i5}nntUVg$`RLT}x&AaYj3xh1hgbfZec9J?3XY(b@4M=W?
zc;gtN$1NI%fYE_cTJ;8o6uDT5Gbd(UGCM>zd()(Sh8DJhS#d;+8UVJ0E}Oib11S^>
zbMmLS@BV_6#o~F=V9rDbF8ou;*Lsh_5wQjEVSo^>BOYo13k5v0#g~zm@%{vqLoU11
z#F$N$Y&I2bm)UrVGbwJN(EL)?P@^K60BgJrVf?n>b2B1x2ED@J;1&d)XeJvk^$dOG
zy!9J1hUhdM-S%IsVpGXf(6Y;=I-O5@rl;=fDz6QIgn1@m{E^adlr{(I(}0&~im}8^
zC|u^b-`?C~C|f0F3h~KANrz)-@;bm-dC9?`D1^+6T`a{1i{UxhbM0sN<{M>B`pGEQ
z+HV_v@8s*vjl<PmvwiII-^aPCwyYz@?~Pi8duBRgX5rKy#*C4&v`qL7_qLo^G4=@S
zMF#eJbfpgs4!<Ig&#C2BsVDy@4M6S>BZJ}cfqcgWQgJKrM+rg`yo=Lk?aSba%QV5@
z7yTL?m`Br!6^--<lUOe+wpWUVtwG%*!x~*~yh0X}+@tp^rj$|z@_gdz7IeRMg@J6(
zJ(6pZ+f`iM>zh~}f=t_6$-(!e-UQhioLp{|xgy(+SL*t`mLrJ;p*nvR#Pf4EM|Cf8
zw=|2W`qtGKX!^h$hmoi1hqk|$nsB892m87V#U=Ut;}?hFNPk?#SR_0#6!5Wso&DL)
zv*?Ox)N7+vXIHE%WS-N<Yt|;Uy2ps#W>TEkx78YaL5rN@Voe7^Aa4+|nd5`f+b7;j
zLPgl65=UcZQ{fpH5KNdNWI>TAGA54_XdE|g9~wk>&0xe3%&RnKL(`e&hhWQ(VtUc$
zQ*5?`R1mgfFiWR8JBqdm!GG*6m6+-nF08ty<S7u}3(jaI`^&+Hk*XALT_pDC9EvCH
z4h=?<4_-(hLs8BFN?!r#PxG01m}1s53Ag7H1Q#w0KA?i+20TD}WZtp)UJmJre&y|3
zAzQNgf-YS?m_1G5d-5}MR#Il$`~v%t*^oEHIM=DS__KCISNjQt5)PRe*bB(r`=4C4
zMoex}!V(WQJgH2*EKB$VDdFMW4kesg>382Y-4jlcyEpcxrL2%^d5~C+Z{J#oceoN#
zdwN>;J+8^;R=ReKXbg*19huS$nJ{O50$3C#p8P?!(NEVrS0}sH0x~8?2f8kvbA9x#
zphz#7$`x`L^}Je`ex77KUZ>!XQyStVrY)ONfs6ddvSF1X;*xn!^#~Y)u~C>tB#$F>
z&9~gT!YKR*FC$|r^-p9<b-isxWXD%PPQ4oMIywM9sDnbA-;b5l7Lqs29|Avz^XEb-
zTtmG%tD@r0rEYz>sb95uY{nSHP<V*+i5*B!!vL7h;kEL-s!Mir<ZrNWmlsXoag2{K
z-_@2*!C`%=PB=x&oH0~^_M)B?8?`XZ>|B;iwGcH82Ko+rTU4AlN4{ZR%I3O!@QF$q
z{7WLk;AOkVu?@`dkM=HUT8!&km`qO;+>WC-7qAqN<)Bs9TsBZLU1QgqYX|Lq>Kgor
zusbkva4`(6lG_%)zKs}-2C}k{7V2DQ;j9gU^Q+c6-m3@904`JMGm_R7YGv)Z0tiRx
z#XhFI11Pp3^dsjmd&oW6aDKZ+_8%w~UB<d*brj4`B*>NvG_O06D_mSHnR1(S1p@e8
z)mVK_8lQOgL(5w-{AWYyDA;$XqvTl>pGGW47xJCqcVL6Kxv3m@m<3Q$<qe+_QW!(2
zywwUNCa!K){0Zfay9W0oyz%{8R~Lp<B;LAE$VA*KRI3?ofRCX+4m7#qOD@?F!um3?
zW<krbUFg2!<i;L34<jk&Zjnq>F1HaiC`zam{^al2gBCBXR4iau>v4FIETepR2tEiL
zm}51ah=%`dwjS1oaV|996l6W1GnqWND~DyIliIptHGIa^j+oQDYRH!l?-D*uxE#eI
z3B@~4!%FG1Pj^rn1efbkls1)WIcgMDLzspOWUEjYWD_HkH|d)K1NJOz$ti<SE;2KA
zP7Zf*Q!omx^tcil2z-x`ah?mYNS3%!*2g%$vz1pr8cCs+`Ri&tfyre)JOe5fO75Hx
z*10Gpru=ZP-pdq7!PGeE3a{*Z8>s7*VtU$-<8k6|cmEJB3eXn}bCdQI1~uQA5TXk6
z<)AR_*&AV|k3Wy+Y5YF4R31jKcSVZSPUS*rM404b*z2u<KWk)Vwp%L~0zD%#(s<m<
zWiRXF!s8HxoC2?R>D#nCqA@gx{>&h}`YhnoFDAN}Iq6Xh+B(}Fo*K|-@tAzZw|`X7
zD~v}fR8gR^rR(cdJ*w!p)y>Is_p=_jbsPe^?i<nwl8#W*GaBSwqcn6uxg?qNQd4^3
zH!U049v_Ay<JGSh6GqVbFg#R6xw^yk%Ci2n5IcTtm*~D~2SSN3%}k8H>K3rz*9Os`
z0z2-UBDnx@sVCZ?s<P1#6@?mql^p-Z2k6TKglx@Y5nF1GB%P{D8c=9{vNCIiq{~#N
zh;K{6vgeC_H{4v~t=?>mnnOyI0H`X8(Kw}0M}u<XT@VczA?CP9aH9Jd@1v2YeO)p0
zM_=0N67$wAT&6uKUqxa??~6u9p~022ESiLPlLV^juop9};NJ2Dl(Cw<_2n7rJ@MbM
zZx(zlBkgfe%%BT(p=48egZaBod=iifhGX(WG50nfN?MJ{KX!;ItrSv7HvEX(9Bw*j
z;6g2dMDU8fl@e~i`Ohd}B)twBek?FuQL(hiN~XB3t~TJ=5iB~9u~6!kR4eA?#cP@H
zB|GV50{8yW*yz=^B5k=(j>b7%=nBk=Ywfj?rhu#@=F-rT3v-N_3rWsSiA3XcgHwYC
zG5;*jJW|B7VpVh<&0hGIEx$XB3cU2Wm~2!h@AC_SS$oO6BXKxjk(UFuBl&>f#+k@6
ztEH%hO2LTBT*u?L(4uOHVa1H<3eI8joo>K1?8S9Nv>e-E&fmZ$B8eD4jzqxY7hbb+
z=ZD=~w8ZPC+`D!(ZL=l&1z#LhPjthpV57-rZb(2glL=Rl<0<G+F?+c>>nSC~Y!?cJ
zM9Xh9dR^I#|IN#o|F@pZK~MjGps@dN<xNibu1mGLP=ND(7CvG{T4>-RV_=RaV41tS
znwX&@h%ghO7P$A<*hZ^Nn|Pt9W&3t6F6<DuTwmT9?_R$crbm$oz56ykIh`z)-ImaU
zo0@I6Qypx%U85@_KOG)l6uwfQS9H099uIT8k^LtQ627SI&Gy*Us@Oz9Z=8{=0N#Ti
ziXW~^*U{JEt9S0M><|ai_)q4n9mN^3NkW4IIO(|*zXNCNmqWGu9&rk!I^0jw-a+V5
zL*IRZ6E#!1!8JKF#siEc$YK#b=h8{EZ`V#@Fjl@EJUNF~`kt%EW5p@I;^9LJPqL8Q
z$oJUt*hTuDh)=$>;^fiC{X5it!7;;TL?3Sn(8a6W)=u&hJQ#N=4^KYf{BtdaO55Gh
zLxas>3$C6KLyBWbYPL)jgwdZ`mTULP$4Qn}+4cqrK25p`m3n2);R$$oEdY|g5dd01
zbpY4^w`>61fBUqU0hW|#?Ct?p09ZlT0WvW28dRZx|6;0xVFOf%4Jk)L2V#a;x!;ki
zL{L4YFb)ql2#OA4V6spL)kU=aO&hjP@x+3_MMbtA=7$0BTm$_lBkuK$bD;jw!A8Z|
zuq-zL;s&^F48l$HIuU3DVgAEvg(*+>uVelfv)@bx*fPjvg*i``sIdg*#PUgI(RpoE
zse1W2sq?pGaTs2cYAHSf1Zt0jWte8z`zYcd;SOfl`vFE>a?*<+&{zUFMsz`8vPDEX
ztCD3T2}?DcyB}Nfh@2NOk~2bu9_9e_l;4eJ9ZFuEDn>!fDyCkIDKr6Exlx7Oub<)A
zD~6)?xO=Z;6`KzJQG>#sX-N(9ie-waoo%>^Ien;3l@Xs5NzI>H(?QxroXUti*DPtz
z)Lm5K-C8$x9cC8~0vMyEn=p%m>W)qWDKIPIk)=xh{(CvC_`pgnu<`iOanAy0bkDTC
zR>;SzJTN2WGa;Pj)rLY>$yemPN+n!lYglb>Pn^n5U)OCO^2ebn84O(QA^UGC_Gbib
zZ>0wK1l$!vRy+~!ilXc!!_06sN6BT|KDzhCF83kH@;+8Y!*BP@&nxNYZtD7P&&1Cw
z>*sFgm7kv3FL?Slp7M*|-1=|7@oy;OCtuLj@0{^38|myqQY*%J%3}2hpBuMut63qJ
zoRX))S#v=uklhgd$Y!AH8eU!G&lyQqPv_f5J;e`rO7gkte<|eu=32AR|BqhHi2rXB
z6wChx-yCeL|5q46bQ7p@(lsh8_4YOi@V_W}Hz_yh|1XOk8iuf#c|kHTC@4wXO`J_!
zT&fjl;V%EyYsPEl{kOTtZd!A8%S+Ge)@!ucPPF6HGpoT>9xDzJ8R+!r_y7TT%<3Wn
z$ld*|{og~6f$`jLk2a^LDJ?1lKLrdO9QF-P3kM9iAIk}f|1~d-1PNH?(E;H84%j7V
z*fn6t%?-edi#zX!A!uL-0Gfxk2AekyDjz7wkDC_tAkU_UzSfWM<NhDh8dXabpth`x
zeiYB`Zxz)auoh53fDHaNqOs?DQLQ{EemF}guzsIkDv-z^BJ3N|k?HH}>uFHu=e>Xd
z>>_mzfF1odB7U?ZzyJ^bD}dffP;)>X0YAvY0b`Iet^l4N#Y=(f{T+VvKX+yJfSUpM
zxUsM~@LcfFckXQbaw@=i*I)r(CRO*TfZ%(#D**28?!P!UcDK9!eQ|s<{90ONm00ix
zP$BF9SN*j3e_Im3(6+#5009KAe6fIZxbtFv`04Q1K+PY`@5mhR`6X1r0lcejZ!fs5
z4fNUCz*E83j@^Q{dIz~#H4qHR;9Oil1@>^$dLNYl2KifJbzYBum{iUFEMXn|uCC)k
z*Ejy^1Z0<DwE}`%-9RX+yloyR1i#150D^%3dAPg#6Zrthe*+K~TTS$OyZ6y!o5zQL
zYxivQ<NsqT&-VA*2qKTaj_2$iZgCg*3>2u(ASaLi_ZQ?xYvAY@bj{yp9{?*37(L)!
z?n#Vg?1%2%;eoymt}p5|-n|PTXSdfcmm0U-1U$Iy?Z@cHtZnO|p<+W*^2!hLhmDvF
z{2B1g=^+w`{T;*upo{xU5MX~)XwOf~0~X{r8+4Yhrvl!Kod2oT7_alGHmLh2hj(%}
zvLEv8R$X&WY$O1H`WJpHODt~m<$mz<SNf5U_Sf&~2j}=#@a}h7Y<zL}7;fSL>*qI*
zU#%|kcK3})k7qx2Yc+`17U0NFCYs-`x{`DZ_Uinjw~P||tqr&V9z4~L4dE&u<OOh@
z2Yk>5=eJ-&&ogir>I5j{Z;lX8UoRdq02v+J?5}>TmdWMUo4Gsp!#x~`r*_6~k0OF0
zRP$%Ij(30{06sL>)37)8eIfCmyHg)+A+EStJTTz4c6|A_La4p^F8;qf$RY1uIfUK-
zw%fZEy~Kb3tv|gaJ7O&kfj>gXAohX3Ldc-@oj*hm4*+chzXZG=g1<D7+#9?-zQ2Hc
zJC`qb2S(<xepMm>TfX>fdN;o+HPvL$FKt(TXK!nfe&c?+sPJRp#*rMqD3(;XC(1}G
zjv6FV$J6Z`lpMot7EdY3i+;E(7aq}(DFd467!5$5B$<qJ!M$%f@yFVx{x36rz4fTq
zE=mcqk1~DLDgie`rc5VW?B}WV0Y2+;-OT7qSs#pAfNnQ{A^kJr(SNQjquOtB2-{{K
z!)d#{l6-idQ+%;}pIwy9Diy_XE?b-vm{#ydxZB(_7-mVWLb7mCW+Vb<MBNVp*?fMi
zBIwPo{8efEahUG#Q+nrios;QqEJy6hwR2|+0}n8QDL7yjIQH*A5_`8~D=^oqenSU_
z>700@5rdmJ*s(1lv);71yfKpXx6Q?U+?rSKys>UYZ6Wca(l4OqMHO8nc<NE2bTf+@
z)+sao=uY247pz+}rfONfo2*&fCW&_AbH1Mg(i2IiDn}>cobxmoQ?K|>=cX#<o%nDo
z+Na&As$?%&T#w`ooTXgJCO`b0T$l1SUmLTOsN-x|+s}+Y_sE#1yWT{LM0LJ+;{D$C
zaf(NxpZyp1JAiIc91oiRP1Q`z_&P?WC=SYZM7=AwcE~7o*G#fTY)!0q=_CA1G%c8N
zVC$^A4%AWu*TnNHe*08N#qJP~-GgZ^+SbRPV0zM$TxSt%798yQMwBaVx(ae_pYw*c
z&3_m@%HB_2eHuBeVSMukK?c5>8?V!EKn>5a%gnkN959$IvNlVj@uCvB2JQo_?){jo
zFrg4yQ7gPm0i`RnM%Fd4ZP&R}T9_oI7ftj*dgaiI{n1hv6Nb0X7oq+b>X)U+R*bLB
zLe4CM1(kNXS#5Py&7Il1=Jta_?`FnC>X@3zO?<wOdP<q1sANd<bSib((<8w8exV;$
zIKL=E^$*kDSHa7HKjp^h*1r5+a>xC=SeE<m%zNI09Gb*A1-pUPK<Dg@6`Nnza>{%c
zg%6!g4nxp{V6gypc$Y3>YRt`OS|GPjc}=lvrt!9tDm}*AvUnLmFIlQKD$ffCN>AL1
ztu&ZVQk{#_q+!Bo6zYjrH<3|#uJv9yZj2Y+vR%6M#@16gj{0+&z3#mkEy4F7Toilv
z-?pw=Hd|;9$E$b$Mku;WJ6egfUgcHf^9zj=)C5@`ypJ!~H_dbJR><h$YY|l#k83(@
zu&t4pgt|JCg9p27gJLH~YNU!KX>&#$5Du+fF;;V6B97Iah<Ce>30{wtJxbSbY{YSH
zbaiGG=_6x{p4YiW;^*x{Tdbsw+quq@P{GmqH;-_x^~a)|3?zZ387BIA_@q0HWi*p1
zn~Ir344JHLR+-<b6xi3dB=dF<7ucJkX9u#FWS2EAwR=#&SvGzxM`OG-v5prKXBZCa
z_m~v6oTs*0h_PjqmxG)`Atf?R>)J2ktY@7KqhboiRq8X_l$~kc#0op@AK%?qX4e|p
zDFW*EZ{@bcB4idj7x{*Y$}^Vi=V%J0906Rw@EnPf8F*yjx^=xkhRvS9=|XK{(6-8%
zI8?5^kNRiax=NE*l3Bv!qVENKg}R0+<%cqUYebKb#vn_Jja!OpcwIqMj)p-(g2b6u
zN+p%XSwvBal?9jCj>mkuFzs{JoL9oA8$3$!kery2gA$Kzx|ex7o2=SbXWBScRXD#t
zsQB|6_slHSi8BL>b1sV<U+WJ$_YH(;mU9x`PP;EtKtm009>iIOjX;NB2)m}?gxG6Q
zGd1+u6&`GM6f;c#_s;a%-*rzvh!P7#q16riV_l@=esBa$jG36vOFyQ^I+jwlrI&q_
z={4=G8E7MEHJh~+c$w66x;VgD!*vgYhA{uOs{uT#uv{8W5?O_NSg_hwMm%1H4snCv
zAsE|SN8KVcOfWKsT~#m5E?vULp4=QPwZUr}8Ue?4c%P$bQ*oYEp`%9lSd-FifizFu
ztULSgEF|P!N3PrB`6d9SbtU2`_llA}wEoifJ{(lr4<lcLj-Y&GD3e;5K}=!~(6e&X
z%hHJ&hZe)j6gF(B<5Y6D`<$Sqj=_e4jo-g^${F79K7%Bqu^POzGr&`vgxroUvkaFC
zV)XBQCaL-$SU%Cx!HSal<#Twde4OSeg3+!eio`|r9bRo`J@?!e-@eY7^1RNuWywFF
z`ZV4vQ_~1FFztx<7R92(SKs&s!=o*>RQbxCnX71-Fn6Ezuq-r>87en`%Zi!jA&&{%
zhqY?jPEr=qo$g@O?Z5HXi)%)F_>9u=c=<V1PSY1n?srd~N=4Wasz#Gf;ALfvcpoJ{
zO?vHM&9~lpDwr`8lVBu_gG_`IW1*m(vcdE@DF|QB<6zq=(53VidQ2HomoK(03>krb
zq-DR(JETu=QIDT!`%snQ@QQIJ!arkrRrE0$oRV{n+%scV@DkAfD0Xb48%uEGAj^F;
ziQTseQ8Ko}E~8X2iSQC9SKr?eFdd#>6PO<Lor50%r05he^Ombn{`+(9=~J3ckdI`S
zaaZeGGj7!mEQKw;+bV*v^F`|mUw4^T1IyJleX*PjuT>Ux+(OE&$9IS+D$QaIK_7(p
z=GWhF{a`1M>juzGb@Nj@9}KUI*X$j^d8a<)xGLZ!s?R1a24JUO2hDBd9y-BC8f?2i
zy3$@tB8&#8*?763)8KqQsVPMkFr^528jV@mqZyf)JJw6=mZua2*b((MgQb&MBJ&+v
zAI@D1K6)&<7GGzj>5u*8#dr5P2@5ZVvPv5cip;#)YTBm9=giPicO@j2&qJ=hAvu|z
zuqAr2OtzCEns1f(p(0&rzYTDD+5odX2L!IkeOR)@UEFP5mPq+EB_pup2r;-%KeD}t
z*<f667?a*Rn%@C-To)0j^b!axl~tZGp5l($lrXvylY%a_UVad>NUxiBew}};kQ!Xk
zP0`tVM+#v?*i{jd{Flh23Y=lPd?_5ydAgDRmV>aJ`3P{|sKBS}njE#CC&)zEXbph#
z79eM<&W@i7JbtmQRgsmkC!A9WAxk@rKGvlc00!=fx~~@Uvs7=uGtG~65Q(usSWPv(
z@K|o3?{PU@_3^@qYj}t9AUR(iGli+R9r%Qjzv_hRl)QJ0`sX~dizw$Y3ylS6N-I$c
zX95Q2B~1=;9Q3BIz~p!&X0Bnyd4j?JREfoRv83zpaYSraOb%e5o6ZWXtMa3x*|2!>
ztsOqJ#D-Ia_oacz(cDI3|0eVi4Bk~F*lS5S)cbIM0Z;u#3kfWd+m7cG126uf3#QXA
z7Xo)<B2BWBEE7eHGP5<Ml^7fFbu*jGZO`7(@^eJ}!A$YUJXl>2<&x{(R?_*72F*KY
z_FYJ`nC5N3cvOpVXwDi8YKv;<wDDG3_xsBB^`ej^${nW}!B6ee049{!!s|$VF5cY|
zlH)ZU;%DF|f5)=!f!dI~Pw4gpv=~Ui0Mv~<Q1ja`%CuIF)QfHr)1^H7N2;a^WVDLT
zVFuJUX-#T}C;F;At&k$Et-yRIDd2W1W3?Dau;~di9a;+ah8Qz<f(Y9l#b?YfqepEg
z0T<g@-VEKZQCWHaf=8nYdqQY|Lb$>zA-ubCfcOF)BP4~+g=Llt+Lu#bIen#?@|%RF
z>Q%`P0dKjoCJ712VYeI}ocw&*+d=7BhN(b_n8CfXzu>NLK3Y~JwN|AQKG>7?7L5u6
z5yj6cchBa_J$I2(nkp0%{84w=)BOm+M$|>yIVjHx&ImT_gYfTZT_9#USC}5&)v3>4
z><)*p16=kr%`%rB7ZvZ`k8@mtOidNkB5!}9acnc{*f2(aF3JZ2Q#?!~0v;zUWlZ3W
zCt0u2g9%cT6{m9Q4kay@C5uTDqN*@PycqPi`fH;49XZtDfOLdrv_nXNwhq>PUi1~#
zxV%|wG6WThT>6V-h!=~6JjW^)jv<7$0b=KH&wQ^%k-}H16?tn5=K0EqF+;Imn__h{
z1r8u-n4O5o@7NxBR2;|SAT#b;(@N?$KE^8oNApUQaF8x_u7X$&TrSJ3^UIYQf6lrm
z=Fh9D$y#=7Qno@!5MDpXfdQj{j;6RrVpv^zx4k7Muyk>qSWH$c0l0wAlY+q+?H=m=
z0xlD&h)?X%tygdEN%X6O(M@BK>{2D(;8o_?^-2x-pvJshxQBPTh5&=Edcv)8*L2eE
zZ_=&?SVS<j#__?JPOw*hvt0ui7x9#=jF~HiLMb`x9zRE~{Rf(UtvIOGZcXH{+T93T
zy2s?^ysA$M16m!do@AUx@)7B1dJgzeZt*vN0Yzzs2lXSC?a>SHa^6?bWO=#Rh;@8Z
z-}{=LK{kkRq4f?eO?%WaEDCAv4@K=O0clB_u251d!SyC{a{t)rzIwu$bz{#E)=14Q
z#X$`iZvYt%lB!t*bE`oJ-fgkFnq?%M%0_Xsc+)aFk##e43chx4Lhgral=y)tB5-ga
zQW8PSLCyEREC2eV*4!gTh878iCY1d`bqSz_8%3;nXYkAcjH!nXqcs-8+O3?7#;k!P
zm`as+a6Ybw*dSE3gZmQo;F-iz9mAd+i(E*q%hm9wx_ir!ASDrL=hi3g#0Q>)ty68d
zwHQh@^HLq<c&bi2OkYjDq3P}B`X-Ek1N!}sXGjgr+ED3p$xKMSdL%;b<xX$?Iz%*+
zxF$2c-Xy-wg0M)o8>iR#=DPOI>*ax_<xVy7VMx%9KA(&ANt8b#r{tSaM<j#?p2e$@
z6Og+)2Lih@7asG}-m#{}SV1i)(K`)>*zmXK$k^cU*gli0x4B6#z&A>}v1URK%}xuT
z4F8tjSrPoLQ7biPI|<dBrYJ|ox#YrYoO+#<3XtCAy0MtTFt+y}hP8{c%|(EW;dK&*
z!a@3#y!=?`D-B}V)446NjnCu(%M4$TkAOr_2nDAlF(W6X(He8rswEbCU;;$${g%-_
zrN=r&vZOoz!{Q^tmxE$N1<vzf@?GpY+;U!o@L62i+H5DYP*?UW9$O{Cbiatoi{g%7
zAadBoqrD;d1nZlYsG9aChqtk;>m#`*ynbCa;|G7Wlv+lUNKj7-?+K8XqQbpgY-U4_
zeVwXxp9l=J<~n%uHcw5IioRGwL1L?M?@unDv3ap2JsX3p#{EvyB~i7?pK!wW4hj*L
zud8^?VLk>ugOpbR&;0!5Db95->$YIT-YoB*^S5%=S;TJ{72OYwBjjM}KJMbyBM`c>
z-chb^U*px+ExE`XJ_=@nSR8Z5DoMFm15^#exf7baEtXHtWD{o_n$JaTlJ|Cig6tp#
zD#o^zX~g5?J({ukZ?M(SsK$A$PVY8H1b0_Q(qKp}_h)|Pjyr$UwEILFpj7qLv~Anc
zz(FpOR=e%GHI`P^ACbxOtD$e^%YBo-7a?}02J$|f$<P?4KbhbfBWupMVMu?zU9yF-
zdx4Tcbt!07nuw!6#+}P8@GG(16Z>BWP&J}0c0HyaFLT7^x(BlmCJ>7YFsf-`(%BI6
z-<Yic*PDA1#wG8}&B<cPz){mhK{9+d9^|(Kwm@arvLoZL7SJb;v2<8S{l~@+rsr3E
zA*s;_@e3c>Iw>MxHvTFv4T_>-q0r0^b^{46lNFlPd*1+O+^;_&SQgo$DO2VC(drw#
z*THP$E%@xBR+lyS;Z2z}nVU6>EhURi#eP~VgHwr8qO<PcmsTJF84E6rDN)_PV}*it
zL2r0n8uFT;kOYT+yWb8zS&9yHxsn14Wv~rpzULEH*6Z!w9`@SNXWKb!3$A^M=wKnn
z1=txPQo7WOWALv~=>GW&nzxoEHl1E~tNSPU8RwB1gefZfMlq8654{Ry*z%?Y6M-|i
z#Mzx@d$;S&_mha=%`;kw%6l+~j8(xW<BWXH^z?)~B)G%nr26Wtdfq-8;BN8Jn#Tc)
zLwTWZ+UQ8$pAGW1G#V_HJvKAWYsqpO&maBpRh_Ro&YYFc4z0%%awYC1{&LGN`(vkZ
z?Z!BouF&Hqqce+9GG(0AZGKfXu_)7m4V{7#ReKydYZnN>gfQ{RjPgVuY-9;~wauM-
zizb`hOs;jm+V>{>eb?p;V!>oRZ)yA)gX^ilLQiiirUdTFN1fAFp?}%lSJkMee=-8?
zsd!%Gcg=%_DhlrK6n~tQ$(kU8hc3RvyR(*jg{L0&?O=)65v|Y4r0i9$WSEV~bhp|=
zZT(0NIle|ra&bq&Ia-4qh1J-pAyfyz&~;h9AlG1=s*ez~MX@%rf-17xpUZ~6;%ciy
zJwJ?a7T{60_)cMnzas!i%@#%-Zl+4P99Th{q2UDk6u8ys<U|JiYi<PEoM_LZo#)-X
zE1St)A(py_Nil-Xg$br{&HXJ@BI=7UA(JPJJguu3H3{Pbsvh)u9E{_<a`k+3q|xH6
zvr!9Hp8`isw7e5ljGE`fCxnARhwI7(6{(z#Xj)y)l(irt$HiTsk(erFcSLMP(pI(*
zH$?Agrn(sV{8Z?Pcq3@h5?5j>jUR<9N*GB5j;O<+F$}XFIb@Pa5j0eR^)Z}s`c_a9
z5WtQ%WLSDvxHLaS+Bsse?FueD_M~jD?iDirT%Ob0vC4>jp1XIc`!h-g>m0Xm4TLgP
zn_0pZldr<V#KQP-xFrfekwqlUzPyq;Le+h@l?~QC=I*gR<SX@KvUQ^`HORipv9#Y9
zrV)r26!79}DMM{j`0>447Ol!G^lZp@>b1A5sXb#<*)6YG3a#-fn-7Yb4%fMJSZOBt
zzdcEgu+o@&+j%u7uaN-kTI7s&BD|hAW^fw-hM@D78S(lks&89ou*b7udl4Za)Gw9L
zq^L9kR<=ZKA}&%RK9djkz|E%Hr$QJhiFdJjSOw!TliUk0i%R~0k6n@1D4716x?l(@
zDf=Zs3GA?Ewpn@Fd6w<1Q4%a+%C!zzB1Lj3Gnrn14X)V@D&i41vZcH%4otlI7acom
zh1GuQikQ3$MMHMTUUGKV@!0kMsQb#MxPm6!2ogL%a18`^cL@Xu9^9S97I$~o;4Z;~
z>*DSnWMOf4cUTrK?^|!x{c``n{W8<jJ$>fXsWYeQbU*VnfLm|BX6^v%FXS8j-<Dge
z&(OBdR9s;aA7d=XI}fj_^~SDhtwWYhXRE!0eqr@02`X|5?DdWjWwX!EApgqy^SIV1
ze=MzlbuH|la`D4QN5-B#LpEN*=QP7qCX(b-jkZm+{ycgdF~KaUqS>J(ost84DKurr
zDcX@x;aDdyEopcsn2Zw1Ib<N~J=K0|Xa?(7)HZWCg<3qbH5~HMtm3%ZFve(Iux^*{
zR<}R`YvqXF4WfkU`SahHunSxM4;SRV*z2OLTIDPR<Qi2({bNgN;rG|o!yRqyjY5!y
zCuWE2+=oR;vkos-Z;@3x?W&nwX5|4MC05DqS4zK|{V7OZdUXA>Qk7wcv+A8MI{|e&
zxD@U>2PrC+k*FY9u+_VAR}AEVWq6ON_{YaPO7<D6&;wcJ*H_#&Fq(}NDfL~!QBtU(
z(~~G7nbs9-Ffo{9>vof|U`kc`?L@dDw(c(4-qFCZ%f`Lxdi^;4A2|&fB&4j)!Ucph
z^21NlcXlJP@o9mBd$GfLM41c7ESL6<*V9;hN2aZTF5CyqOxG+M`<RLY3V$Q7>AH>9
zTEe1T%AD;A<XH)EuO@cQ2A`OHk&Tz9T?${|fDvU=6!rQ5hKM)_4e=A*v2V>|@Wx_X
z__v7Vx>=<!w8qBOw?V#Xx^&`S@Tu}oQVYzPDx@Uyn5T~oWX>!E@o6v1xYJ$adwjeU
zn^6_b90ENv-4xR>)v;N0{zZuzea#Clx`c55rg*<Sa#ds$_YEbNoi2=)rVpVuF7#W*
zO^_#wK)!z&?M8ZU=Z<R_3%{23&oHlU39rE-U1)W^GC$KwWGz~V`#I7E%adWl`#Wmk
z8dR4|ox0EagDB3kGeKmF*AzqZUz>z~vi6lORS~X<Mx@RUXr2{$<`hccSk|^9ENMT^
z<{HR_OXRzV4#=SxTrH{r_Ueed{BYHCw#+it3_Gu_TX{md&L6>QPCchG4l$K3)w>S4
z$iiGq*C3QT_1G71dntCM6sFU}eCyrd5fU;@#PadghpiDqvpQUGUT)#vp^)CSBffW$
zSv(AY=gm*-V*PCvV?m#)Wk$X;jG^ON%+$4>@mraAXq{;3>ISFz+sU|45^cJ7JU{cQ
zJn0znntb2$sgt!OsM`+=8%%t33+_O!Y)@@L1UUZt1?>noD5m46#6<sD?g}~W5~ISI
zIiHl5wD}OBqlo&`PUVjcf&`1k_{BH^#l=+r`chX1liy|26awL+RR@_}W{{?ko>a)-
zcYSz2<ynM9!1sfm_Ailj(S`A#gR|EtUcV+p4g0FAm))I68;9=X!^CWxDZux%QVtt&
z5BQg-U#wwD`<^zuCazT)XMN?Y`|U>=cjpQk>qy0=)-*Y{zChV-jM?AyLmd8r>!LZo
z!9*Je2`Nb>)WvhxV?K|<EC0S0r+O>c+!hHU)9frNGWIimj(-{FVp-2nwb%KAPs{^I
z-kO?VOLo|-@OI}lemZHJ@%{eM^;z&dGpDDMLxWd+Ybwg~!kCrj_H4xUGZ(_33mS{j
zgIr8ZP|g!N<#bV>HRc|}R!?+OAuBt%aYxS1&#W@XW5cN-NMNZF&PCm?En<Axq)7HN
zVmIS;`+NT~4`VUs<L}gB|K<Qx;UsZjn_H@gR!M3`5+c9Q{fv_u9wlngZ{)5IK7f9Q
ztb$WxJAe-(Y)UTU>Kux{OQp>y@uK_FVGj`P4Z@^JF|nh6cN?^K!$DNaVxtI?J0@v;
zZ-BqMK$5WSY70?=k0fEi9>K5h{Jp4>b@wpRO`9w)*VmUyfhDMhK**B~aWpA6P^ohc
zzi<AIP!gm^>6;==Q$O#@T%fHRYz*ToV-)LFotV3~Ydi0n`HdnGQ$VD}SS9;gMmJab
z0P7zM<L5Hco21~ri!E@0sap<`QIfmIc>~>`LJJ=?u&hKNNzkr+m*H=-c&;x!n+nJ5
z364s{CPjAWTBNjFSt?D`)W&Aa?KAx*KOCj+Q72J>tF=Io-j%iTnlgu>1sC`5A{T)c
z;yHVQ4{h3D7`#yO(M_u*?`&F>SgjE5JC2<8a(<GZqr0(-BF;`Q-Uo-R8!T)aH}kf?
z3Ga%o5dgmSEj?g01T`yZ)A49b2bu-H?u9=A6eLu%5eGTalJ@a+N|)tg-{cQuR=G8q
z0w`BL&<r!5@lNvg@RAI_`*uAzqqV54#h+e$VQCjEUs|>dk5=%#2E!6tNlg@P2)2+H
zG<eXbAV#jOFsy7#jW~}Z%(hQMBMVER|BUojUvqARm_EJ~zWJ;!R){AZGcQa}Se58V
z47=FTNte5=#2Z}hI26)@axvB~x2I@G{qOjNZILz4c}3w>OoD1uB{*|bX~%c9o1;+H
zp{N$t<gO^OySi+$J1i?8EBdKCZ9@YE;=77$2PT8^&HA3Sb>v<+oMM9zC(7EnF)7`H
zFHTIJy5+~qn);q+lpHLR_e60P-(B2xw!C3)rw-;Lmd6wVS39!Wu@dzE6`|p2jG1Hp
z>7~2tLxp{F6wpgBqo-NyANW9e9MpT7JT-qxbTQeI;kh5}sgaxXS<asBk7INSl77Ig
zA3-LGIc=$c`cA_dQ_g~Ar%^2=N109%3mDTIHnrYu9#7$Wg!>mIkF;7?ifPh7Iq@f!
zcm;v^VI>FqVE77d-Nx8!h!Ltp#PN3x(jP#Q7KR~6sH19(Of%{R-xQGn8}4~f4HN(N
zK#l7T@K@5@9i-bO!Lr(vk?w3QVjcR4xu|tr0+W44!I<P%LbvX4=m*0=$CjeDq)DR1
z+%Xi5`Wu&>=B8T=ap`c>2@IZurY+(}+;f`bs}m*e&Q+CK0Z++JJA!|tCULzBfVy~u
z>^O%4cLAaU;uKl?ODU_@LJI2Yjf9lQ(<v7VDW|%;;ptrLn{1}DkcI(8=#=D04&Bjc
z7j3ew-uj?O*RL!J&WI_5ue+Atcq@Dn8M@e+_8QD?DK{1?_&ctqUK_hZrf*v~!1Yw2
zdo}gH9TOF#X3y|;OS+3k<5}?F1D80i21*#O7R;CJXTKj*Eij}S^enIv%ig4ZD3cx#
zD<#8d@4Hogm=^C#%5pACaLBU%`j$Ox6gi8x#Q00^?g&)!TcPV$X!qt{M=W_YTUyC6
z%I`{z5$1Z#7<s7llA5XVX42OBfFblj@@DpB@#TSNPEj$Kpt^oNhV`&RHE4T{G3)}=
zQ!CF3ISi4`Doo_2OrmaA-}E6Hie%dVV@14tg**5j^OkQGsE^!!|CvdI7<TG}#fE%x
z@fCn>>m+5TjMC2{U^Dk!$rgF;rkshJVPVz0X38K23BS^!mc`r*%19j0ppkPuZb7z6
zRtJeJWhQ-g%tWjqptq&eR(wh+jxk@88F9E^6N?<-Tr>udNF(*YS6#Xw;9O6_v2=dm
zc9uf5E%W5q5obAK>k}RG=r>_Rvdcj}8Ip$)XA=KV{yy|&ZJ(Lo(t(F;H}uQWgN5}#
zT@)tce2Wjub?UE}FyTsRad54FSP?q=Q);ffXKoqlXX!FWEA;z)<MJ}L%s+<H3NEtu
z=|#-#GObimzQsS4U0)2H^19aaiI%2zs^bqbPBqMnEj)(H<Fom2|6Tl+ST02=>R0t#
zJagaujFkM|yiTPT-wLp!l6gUla!3s=kqe*vGe{G`llpvkg-iXj>zm3qmH9%K<Y)cT
zmI{H5Zt`Js{Uk&j52q~XU#=1Rxn~*XuKQLw*Q(SWyjb~>6Cq^6U+QqaDqH93Ll<2Q
z?e&rbGrp1{#rT1K@`n+OufMA#ima&#E!f6?&_N3MPk%_zlQy3j@cd5VX_qAUr|ntA
z<*I-m(bPsf#2F5TE3^5GIvAeP;V2iHfHk66fw?E&l~}6hr0I%t!f3eK(KT}_x!L74
z(ISVdX<g(&u=FMQtYgH2a67hXn=K8_5k`A-LgY`WA(Yrf(An8^xx-Xhgz5Q%8-kdS
zutB&PGH6a(-Jy1bi?Xx&z|*JDzRJrfeu<gU`fZ%N<njc(y<2b?)Kqns`0X7Ux5&>g
z**>;NI-Vexjo)^L`N~)$GFDZg{8XoC-#>uaO&#L$t^n60Wp~z`R`ATHeZtfFT3!j4
zi2c<|D4qw+{jLMU@yFFk%ij~%7_=3{i;5b{@CO~05Sx+u@K&4x3mZra8=4&MOLb9Y
za6sqQgGoJM^?BT2o=@Lz8d`@s17?9lS7`cI5;91qEPp~j7Ao`4Q3xG(z)$`ApgY6s
zwbcbr9aYnHb)+~LW7}Y^{6o@1qNQs#-BMnSt^$uH?j@b#;4R2Z%SzW%_B}>L6Qw+z
zbVguOC6MrWhn9PxvZlKGB6cz<r3yH0Hqlthi9F-VO?k#OcOi>98F$>;7HsOgLMGZ-
z;}@q*^2)Th2$})9AFZBNCWqALLQk8~vAZwnWdC@TG=n1duo@o+xKT=aRWI$WQM}9W
zHQZ_ruzD_3;MhEjIYeU5zgUM-MVpoSgTEQJf0v`GHxp)N5Edj4JWolaaI8T$Zler3
z{}IJyqC_LjU`)O!qW_p49`S<Zh9!94^KmDa;GIq=((5y0Be&kfGod=Ve<)5eujO5u
zx4^%J;xfl?_k=t*<ckXVtZSd@T+?4Zr}{{dE$LmKok-WR24ZY?Z_2lc4#jwYThS7r
zy7E|pN=lg+H--9M>Tz1b$JeRU3+gIf(J;-;>;8d?$m2^*=RNgXSrp-57kvu3CZeCs
z@f(|%9l-(~_q@jG4{_KLaIrMv$*^(3`_L}_Q|eRMM1X~mqZWj>8Ob$GeA$HCR_>1y
z!3AaUbEvZFdCcM}Ng7w-SZq!!(DIe+;tlfr|A3?Yf6P}jVCrRiCqApdTWU8^14^<3
zq%+k{QSITr@otnIF40hv^6fV%HMoFD3^%^Wu}~~JCpB6xwRX`O=^P*jo4Tl=rT&Ii
zq2GP0afdk&jvLLNwUWz2ZsrwoLSOn0y*?FB`Pe)a`<UP7hCm{2AKh1o!VLvAnYC<O
zn{sbq)=sRzr<gW|?bB_GG-s3V8Goy|OSX&IVZ}Z?-$x%4ip*M2<#b+oQl}$NZ40SW
zM)x0()f!Nh3A7nazBCUs`Hv{67}=s2SvvfHj|$?%#ibd=VML#SOZ*usshbxjhSAu3
zX6zD!obXs$`qXyFvgFa+%^o|be1)&6OoHq#t@&~46J9swX?K80Z`G8UIS7}(D87m&
zJ>0~bb@JIgz~O9&U~quuWth&+dIM?ydoVA`dyKnp*oHrK53nacE^rv`e+R^F3l0-&
z!|Mh0evJw~LUvcMCK>YBZrPqD`Eo&##e><?MQ%{EomevNfV?Gbc(r+#9B51QR8(j^
z9)R6@L87b(uOKZXZZF19i6-=q>H}|LI4>seukXJ3U+G8iJO{HLrsH>15k<cG#tI`H
zI=hwe9`<|CZ#oqePQ@Vm*v^gbOG*gy(BHNEikON~7Uqep`;#Z7*Dm1uMnn|3ELe8M
zt(dXC_cN;6W9G^_$NAv8&|FxN|H@;gb}CW9JLhI)|0j!gPQ#mGY5(VcO!po$AybLW
zJ~`}X*GWjZC2~)KX7$I}Ee4&?qHUXA22Y=xzfSAR_tQd+XZd~D3k`cugk)`$>Me0z
zgF}9((K%xqWUCz|;s1|)!t;N5y;MEy%~&)QjjhZ~oLN*{jGg~8mb7!U`ET<dD>)mB
zn3$cr4(po}2L~%N7auv>TgSu6p~oU`<m_nauJa!}H!JyDi_EIWqHN}5=i+E$<|HV{
zB5nt?b5yZ6GBG1({|{DFnMKLS+1bp|mYkjIze5RkXDJnDBj-2!e>Si3+p4zCZ;q00
zC9K&0bDM7(4pw9i_Wy|-<Q)Igvh$I1{#Pq3%%bFIXQE=}tiz)8Ly|?+%-vb<O(|&!
zd|Qb{68LuYA7&<Yre^;`qLOp7@$zvB3nM!@JDM5UAiFPBbjZ(`@x>bQRI5E8jK;i@
ztH*DrAZUwX$OJK`CsQwV+Kc+1M@gL@kOpfNOFbY7C#0t0!AXCn?RwePc)YFlV|b17
zQ-{_+|LyX86gqyIUp{Yqs(yXReZ4d2su=I`_2}_q=xRHAJbs13H@(6@WPafFw?mMx
z`(3a1uj68`r;if`zGOlV_qSeg2Cw?JZk93_FL*fn=g;OZL+f{~fea5_SAM&0YBu`a
zIXUlZMP6?UOl`WxR$AL3ug{nFmu(YPYOimu;DvK~*?Q@hTd4>&UmlXX^ItnJFsXDQ
z-?$}H^K|GvnVA)&%baD#){Xf=D^i2^u64qvdrH{}meMG!nBw=U1xQ`@isQ4*NAXRv
zGJ9bzLK{&gyQo*4U2ja*nTJ8A*y~?yeeM<C>?v4}t0fcv7?b>7A=loDt-jWDCeORO
zWpP7ExpwE~J!8H}ZRaQEJs8iwt~y(CsYEWMMy8ITvOFR;<`5{>tMZ+=U&@5rW9lEg
z7e)_*1A|E4gSfifg{S%xjHh~F=W5rNjtlb<XymK%&3|U{!2UCn_R;U9_3AV-_ci-m
zh$?!1?5$e4H`(j<b1GoGcya1AgH|RQ<~2U^qP$Z_O9@s??|?BnX(c^A`fKg6ONxOy
z^7+qsY-ca-t+c;Ep>&r83%SZ7PcXX2`up5JBG6Yl=rx_NnV;}b$%%pf0u%JwBj^5h
zTjJSF@J08WgW`2tB?xx>dW!bEV1SE;hwyZ~-d5@P>J>GihkX+FWbz(nC!+qAkWzng
zx>^mvJ*0Sb9{Y8?-gL5M^)kyORq3~Uh3WS{op<4($V-l5rQdqp*q+}TLuRAjdJ_uN
z?^*4?QZ?!8bG%fe-*Qu><g0U=>f^mlUBx^s4n#En;<MbT@Nzvx{9ozfW&5139Jbt4
zFma{W_*Br^WqHP!@e0g*eJQ)K)4t<+AhUe{Kk}UB+&;faDu?S0<4Z3XpiDJk!odQ8
zXRUz;^&fJT4_&^n<yR+8S-XQ*Se(V>>%+njH*0nt7~nz@ls~1_R!&xsL|8mf>eOJ+
z_yVR%xEqY8!)bOoW8{O<BQ_cBd5Y?-=8sbo+RLll#Z>CmOO^rV{jEln!C6QICQeB?
z_HJGxgP}<oxF_nEEEcP_JrxGU!vx12!Re;wrAt!*Heb6wj0$&@Ey?sQ)eibsf>89t
zc{GdBM7z1?5o)xf0LOFw9mXpkCytJh+{!*Sn?oM2*+vgU*%B!>870swqV+|IG7xmd
zr*ycRGniVtTb9AWZB<7#Rq0Bh@|-5ctW!9^)IE_iuM;4kdQS)>bpVOvCi-Y%6YB{%
z8!%e+<h&T@KZJF28_(H9CU+92;8YJS9$9(b+0Cn|*MKFRz^Cg8>6PXxtDcP-;DxaC
z<QiiR=0)0iWd+IO$K~2Rf2P`z&%h?SkLpFbO&V-Or}tTNUW<g~eWBcF-$GVt#5>mZ
zxysch@LyI_Wz<2Li+()nK-xr`bXE>YD!=lNnaiPyqaJm(;G<$yo8yE=o1=0V$GR!#
zwLQeRh>>9A4h*wav6Kf_T(eqxgA6-LJMc`@hvb@oM}&z*8C@i+oj96}EML=ciKc7}
z>{uYFd{Nyel(bs8K01J(1tBtH>&#`@+@S|H8lHMdM0c(A!oM<I2U|PZo=<ZoiVp#z
zRJr2vz)PKa5tceIt@=FwHt?CbV*&a2$!Po-)^>tf?eJX1{rd}`0IL+j8aWRck}-CY
zHF7ZkKTa9f*Go;Irkmnu{!1C(tBEO2LnY1-X?*0Tr77}Gi%>h_pn7_7b?N+Ei+KhC
znX$VBN``xx7*pT?V&c3Ru|JNgpdOnE!*DHYI&~%-)j0sy=eAL@+~wT)ZEUk1VnD^F
z((z)<qz|h-a<S-|$(W~v*wm%u0@#gJ@T$gH(WS~=TIWAoJ9uDb<!MRj^O@62liHR-
z=@CraYDK1}+@G;$WX48QB9z#u6SmgVekz&Y9WbQ9yuu)G9aG=9?NPv<az~F|I&Wbb
zVX<yG9a~IQl>EADP-|M+7<FsK?55e&qBqgv-W1`8YxPabIib^Jg=u+f-w7F$CW9Ok
zv_$C4%=+QeVxCpgzsv}Ile@>cAKZ$Nz^ac0CpByceOj_em*yQTiKB99R@|(o#|3FI
z`2aSrD%e<VwVr9<td?Fx{}Q|Lx!+!*ZKB=>>*GE-+g@NH6_HwECJI<3m|%Ha&dZ4~
zW+|Ahc0>`PlVk|ER?pC*S}-%Q#=su~yntJ6kN?)vms2lmTNp1!O*}nKEAV08Ue=vb
z)=f1K9D7XxP15=+YS^}kaPOpux=xD>3Q;eA9OI9GZ}P3FD?P~*8>uxOy-m%}fh`9~
z9EToW#vEbqX*G5Tl>z5Bsii%0KPcyumrMlZyVDm<$yQBw<xGteE4_7Gtn;jt6P4Ai
z$F1KurXy+}jQ~**^`3rK@{fZpKhx8hoL^0&wHhB)eqb$cQRm5|a5PG0Po&RJ@DYjV
z6|?LprqXp~(aRAPKbO&+a+?me4ldZrYJfsM0kkzzu!Thvhp0ucPg{7=@VPs#Q@L}K
zRj|C~`Rt30xFyS`2(L!L240@#cjFA{5%MC24y(5{Rz6PLX$WCl20x<H|1ymLjr)#9
zZ<mt7&!Aa%GQd&l`sy4b7DD=f!*ATL4|Ig;xEy`GVf78QUUd^Y4UIE5cZ}w}{WjWM
zHgsc<Kc;b$?Bx&vsIj;BSg{{L4-k!Dku}s-W1JV$JXzw1m4rGy2I55GB7-b&-9T>=
zmosdNDv)#sBeRiF`bIRYy2NP4-+nHX_Kuf?4bW@HTVC5uSdVdW!hfc8Y;;%CiOA3C
z$A1DEJFRJ~w*!~J6V=WTWl-}%T=1{G`9jr=90{#z^?QI<SF)M>0%MSpjZAT$QOx6-
zj=vmU#z=q|G)bgoMlyB1p|>3);sp%V+yqtLyR+6yXIQGRuDP-oCzgfI#I^#45!5Yv
z^_PFS#6S*s>yq1Pv@^|_%n1ouiKOujytXJq4KvFiJ}zUez<4=5u2nF=!qp^W<I5r;
z$a+?|2JP5s&n^{d5=netzPMtoSY&y8WIa+xaWHhx#OC<<U89L&AP_6y$FEcL`l5gR
zBHCWw$E>rAS?OZVIZcl$Z^2h{K1@&|-fo2*8x0|)^;VbB<ellAGGUkA9j+I9r-t~Q
z4WL$7U`oV<gq{5%iIC%(v6%c)%5c1%XjXeaOg!!ggS~bHKYdT6ULrGO0+J?>JNj~c
zlY7-205?#SUvH%jk_XP{$PxYooKHgV6BG?U^uKf%(U5k02hyjhNyKGV3NfUIi})mB
z-9mJa>+lN!9KPRFimBB%e%+u#+_p_Dt#ziTTZ!hPVZ7yVa&iaVK){J%0kpjAPER-3
zBZ!Y*=8vTT*+8V_F-O|{cyTu1?Rm&cmZe=NQh$tC@cuqQvq;Uki>LB8i4o*eunElB
z2~TSQsthlt;n^rivo=8g5f1EdJHi7}G||-MEv{xe#{L-^joo}-TwzE@`!n(~YWvf>
zdJy=zEN&uzWc-V~PQpp%j5p_7dfEq>6^ZFfQ)p$8w^!~<`i04UZ|d#I`U;?>8@OW5
zT8wVsh@M3E;VyuVvEg&Pk2q(I<5rq_kFmezrvlW5)VSi2$&-*)M;*2}F>ghlXacI)
z`8arV9sFmU7>dq>!^A#6Yw(^7Kx~#kTa&eZDH*c~N%a~;?MY;_Na5`4xhTfP*`$q2
z6ch9wZo3qcY~-1&v7kP}Tu;ts&&{QiQvW*29Pg7YCGi1?$SG9!ikP}-`cehENJ~yG
zRP00oAO(y7aC&L_1cLQp=DV{TQ2YZC?JGJnYX1?r!3p7Y_v^t153@VT?f2t<Sm_3u
zA-PfoJb+>n0c&m|YiVB-u#-7e@y9&{yXOX5G(Pk3AH^Jp41QKeBY+ipZMhn)usAa=
z*eN@9reJ8tTE=bq9ix7g`h;uhpOHm=ODvvcR;Qp-8SGDL*PynX7E6@_u(@jlyC-*|
zU0c%(r#bM3YF0wy0XU^N;h%mI3i&nx-wVW1rPf*G;%*SVdkk<D0yKUl(!Sqi)521j
zOrt1SJUht{IX!{zb@b{^tGve!d%r70%~5;U#f44jWjwEXmw{;nIpb^@v8!w_^|AhY
zZ#Op0XznSw30e-_tkdH^V{Ou`2D)XoM(4&Om82buc^%z0>PAE}okRwDI5d$5lVtE`
zC+o^Dj+6;j3CXt)PB@ox_TV~Vgj)40h77PltfKvE`iO^9>~4=vuE&pn-tM2E^u~c1
zwZ;jFsYV#9iSl08FSA%vh8DFrwx0QCpV<HvoO>m;BnYVEa5O@-W`B0;myk7{LDr`g
z=O-Rd=uhoHl2pn&#(H5h9YTN;`>~1_TXbzD9^1sYJ^{6ZkxR%8tJeSy>&^09I8J6E
zySs|hcQ&sO61K&Z>33iL^_a-rEQ?y~!N=I`V8{=<NKt_S;?~!ifXA4VLEy5K(;8j?
z><@8hJ_5WE$BJw^s|>ZKQ4OKa4_3Of0Cw35YCxH@J_+p!d)iM0e)fjdSX#mp77%i*
zNR}3TX%M+qi5=@g7joR6k#XR1pBAv7_hx+JHxYANGf~VdSYImTm{E~6F@!-Y9$ZIL
z0G<2sY?hxdEOC<9Wuwq(ZGIabJHc=dG8fx&T>zqyjaNZW@NK9&Dt#xlM{t9`2AtAq
zO{R37b#W23A#iAI6<Dt3Kq^L^Bj#%%*<acCoc%+R+w|z$`j6xi4b`*T=<l>7*;O9(
zjMHsBJNeV{`0y(BIm&sWHh-4_tEiL{74mS8TiI07yj*a#w<PfvUjowq>eRPJWMdEt
z{oC>}c02mVFM4EP#BwdyAA@YhzN$6eIuW4t9y00<H6qm4Fh2~5=(Ujj3>>SqY@*Af
z<^|Q0RuXdb9Ln{kEYkM6JlRn(epsv$&hj;MP$B|)<1u!pofL1TZUX^U#e0gCY?ryl
z4v0jZdZVUmT$x<5O+4n?2eU-jL*w@5waE>&F~F3*!8I-0biR`=lIs`s6PB>s<xpU5
z!(Fc=#c_L7d8p-QxhStF?##%?2JQiGx$$^mj8HD?v22dmjX;}V3nS>or^^`liB#mM
ziTRkzjyt0_;9Q^HUyu>JHLHr-pplt5k&g-a7j8I{*z}$)0b<sc(~M`w{Q0A|dNptF
znf6H}(5<mDjN}*6lILK3Q%aom7zl+)1;v`ZKEyk;N!!XbeZuv%^3WGz#x;M*8v8CV
z{rM)JJ|DrB!_<4gHeU(9rm1}b%UqlM(rjkB)*6+|tFPoQL4oPG>Ao;yRQh|44>7m9
zRj6Z<VTU7EYPJ~{x_*_DC*h5q+Im0+$@3XX?P=cauQMPp+<3Y%6*^vwjm}G8C8N0=
zHG<(iZ~<<fS8mGo6tM3)>S2Ur9GUb$I@xL8KAYA%cq16VBrK$qxnDWAxXo?q2_ujR
zwqoZaYpl#99QGyOFF6zkj_>TH?>iva0VjPz>*gyafQ^(@+Fc@g)Z7@ELYbcvGxtl3
zT)S0Q_umU}L%2r{#plFnxQo}|UxUq?tUHJ<i*jSiIxZ~Y`xc{w#*6PU+p6LaoMx|v
zQ%T23EklSC#+r`5a^L=80MteA9sP}H%$SIEr2bjqi!wVch3-NdPoU75M&S;4`yNiJ
zvp{wT)5L0A$iB+~NpXv@fibS%d<1Y$_Dq4nHyF!gO<~8nxoXC>R1TeetJ@qrUPW>m
z8q4S$NKM7o6!PLU8#Q($MmV0N7S+DflcPW?2Z-GV9##B$jnT*uY#YHvtW>p*lnR?H
z97GabY#To7GNx~kI6hY)lm|!$t7*bjDYZP<jGDve{kg-t7B<5s(k69oqw9GyKQcOL
zHp98MzI;c#6iJ@%??~r(@SYLd>x4R6x<^CqBDeqHV^v$*05L#4wz^0!@*4-isZ+fy
zs{hwAb6&TYL#siVYEMmIRZo&0S5ouh&g!yT9Y$h`{P-vVzl>3uE-RECL)Ox4Pv#!a
zy2(wNL_A9|Xv=^7GeAvf^iU3PR}r@_>Lf#LLP`yS=W1!uqdXHroa)3lAO1AvkqHZ&
z4P}`9+H)zFWgC<>NivnUl5Vdu>U<qP4)}ZHa$`yW;-Bq~ed@YqqDL;3^~(gZ6^!ea
zZ51b#EeRZRe4K1E-|HWCEjc9XOP@O-d&<oe;kxy!H{NqS6wx7$CrUiRwrcWT#b?uA
zV)FyI+PS%e%&WJi#~IBoR+e3uLl#+u>raJ<vGphPEzMmOn*Ju*JjKFX*UG<o4Gg7W
z@ex&xQmFa7uhF+o;kX^7>7H}tG~Nv=KV}uu&VEw8ovfsw%p{@a&#nz30;cRSbk~`E
z^|RJn9r}y5s`Yrj$_1Xkw@H<MstMOg*>+!mpzd>;;Tbi*uTh8Aa6L2VVBw)8>0xX5
z+g$QPWAlVo{y5H~nBtE8l9&IIYt|d@(?81tCl(zPyXH<}odm1h{Toq(m;+=<)#7ce
zFr>EghOepKNvUsSXkyr_v`f1fw}~N)wrudeFBmbS7S49qzy+p1I5&;iEHn@pyqCiz
z)k4x)##415I;yc53$VLO$`W2=^W^ro;`O9+jcyfcv?&wZtuaxZDxMB^jrn;gVHn5j
z#HJ4YMRPA`4KE#4{bAGAtdQi?Pd0|GhfU%ZXa0n~<)m~^NA|^jqj_I?m;K(?M3&LJ
zCtCh_*!vhr)4O+NZ27Ib&<ll-2sp~8a^F(u;aTP@xhtEg0C`ke_R;pPSG)op4N_A#
zQ~<?Rw*w#XGXt&ekY7+bb7|T51i}U;c-U0fc+`DZh(XejM;#SJpZl(cMbnUU@Moe-
zCarVd1Es-9+*X%)oloZYyEUrb>#EiXBk2TkTY03^TLLin`+3814z5Rk<6QjWWw+M}
z@3-G{`>hSHj$1~^=*ab5Us4ABJI1Dd8@O|mGP{bJ^69C!Isj-oI_d>_CwN~t?*8m+
zMMme_LVfGnWD%BcaCrWGzB@91>3p^O`=XT>bL;53#&*kmy^7y7B^!;yj5Lp-LbxYa
zH*u(Ui&j7g%-BSW)DyC0!x713<#0035BYxJ1!ef1w&Psrc#IWn;iA#zT@iz{A%$49
z7hl@OR8yurwFRfsr}$3pX7D)uf%M@mM45@wug+=_!e&<SRK`C<=pX|aM`aa}I}}i-
zc{bm>yG~ZNL&+xjq-lfqwHM}1#VKL(G5u4$gb2i$Z2%%ioY<Z6U=(TTI&{a?l=jLR
zk*Ja;Pirn9U~Y0Ma+7>Vu0g^3Zc5Q-YluF&h3N+5i_Nfa@fWm=vFf|UDTNP=L@L-f
zwHmZcHtt>IK}VS<{<gn6r5ES;mp9UA<LN7;GFH-v;dvW%b7jli#U<6Wz%ll$<eD-Q
zpPDB-#`SqBT5BK1YBs^zVpNQLK>?0LGd3Z<U{fd#Uwtaf*%mzoX<>W%rcmUrL6!ty
zsC9c{cH`g5>t8%HZ&k{UP+kaV>NO_C7qRd?Kjwr3Q)=P|IMO<1Ajp-?W&Lt^?LU`=
zX$rq90bs<9(x3b|G>&cx)>nJYKTw~T*&{#?5d{n;2cKAXN8j&*95!sn=Qu^6t4+`5
z=W^MnzLVDQ*6hEX?#f$2kzSm9@2ss|CZ2%Ft_1nTSNl#tXd1rpsS%_c+Vte^NZg(X
zk#lPTRBWWN2@73fb_}P=DRdR6MdM?a5GSNwsB(Y_c@Fq^<Ip4*nXid!!Zadlg(KUj
zlk8>KO>=!*g4Ytp^^TNqdYS&EF*%>58yXn2N1rN)PI9hT#6WINSrZ2{Cax~!?uo)X
z=+YR5(d2nK#hy#NoVfemU04&GIBqX9--7^UKsSv=E-=M&6Q6^GT{_$P<+-Z{My7CW
z&KA7z3Qs_G-)~_3@95N#$ZEt~N4Ic``W4(~Cp+L`Q<R+R-!`L@3}C}{NPeYh0e}tl
z{?ktb9~0^}iQZZ9p!8nt(yIHD44)n044)IItcrMQJr1*G%|yxE^Yw7aI<-KyC4Dxa
z_6z##<p~)CZoeUVW>^-dnB;SO{?&6|1=MwO@JIjat_6Gz|6GS3a5Sr$Xpvq+(mp4m
zaW((&IQnXFvl``Ol_ox>ZVgb;b!NtHT|wT>20P6V6*MN?3rx;FzkR&?(peiCbd=uO
z9fC*g`@NQKf4sjGZzrBJAChsV)+j?uFY48JhHUyl*Y4-Ix|>Lyn+w@L+z+`F*5PaV
z^nczp_eNCUb{%i|fcTU=#c<)*f-PMMSl+?HIb6GFe`$zuAjJ8pp>!<gPx@w6^;*2&
zLf+HY;IO@XS7~#P%+3a~&F<@zmK+)16vo=>-+a;EzCJEHV>COiS*fb_0919<I<%dk
z5Oyj;UvSm>{RA`DgwL4IKuxLMIe<)i7n(*A;cC%rsC!k}PQPRRz8KoS3cSdtcTSul
z*cr83q#FE3n0M&O9m(6CE`GL61&VUXp3Ak3&DVmhPSb>-I9&apqPb`H2jNKm3WjxF
zmG)KxlaX!Pu;R*1YE4OBtcT{!^nR*j{-uy6_uj=&IqGfI(%Sux9CQKqhD;*bds*=J
za_Hq}ae5Km=8UnD*)~mHwT4HLCB@dm#z73Y8irg($ESt49pZ^iXM|-8U5S%rF;HG^
z>kfl(YlfOPymw99A&*j~&|3-iy(LG*iAPKCMT#v(kZ}O5=*uzFQ-D_swk^h2^m@AU
zob-+}MGocuReQdcG(2V%SAUaps2~xU=?3Ag@Z%k3otEGkWd_v|b5nMkI3l_&Ma#z%
z#0?bx>f@8t^oD!}?g;g+@o&}5xxkg)_f7d{k=AI1rF}N7&99|Qe7U)9bVEcS)efqO
zJk22qji*W{i_2OrK8E}KS<~uuG}1PGiz{Exk$ER|v@@^fFpm9ZP~lIR`_V=aSmZ5V
z&U~iH@O2qO5Nef6Z$sk9vd<;sOX<Gk<~Gii-O>M1zKV*8B|>p8q<PX&#%;HPKr&Is
ztNqq>jH^>7tFTY}UwT%Hmn+_f*3J`B@RQLf|4=l+wrkjgO4uLI&59Mv4N360RMtaz
z)9&PIEei%`o70sOv-=7ulb&#=SN{utyiNgnNrjH1lglUCj#We+6#ZhaOuH7ny4?|y
z+=-_Xkc-Q{lAGF8WAYedTFWXdDlz@4bZDvrJoZCRyY{pkNykmmF7Zg7S*2;O)gWpK
z32OYEbfpIWeqs9uI=#Fd!^0=XU)`JpEqeVj|F-LLohB*@0?X`_0G;&%jZ|IchFeZ+
zO-1<(+nDq=x)yb*neVyu>H_W%23y2x&Nb6=sBY@^0HxwU!7jhN7F0(O6h!RK<8%($
z=62^m!sLwH^qR&$q-YUP6eJtgj4=NrrUk6;EZOrcKTf=HbcKn0Lkendoi-)YxA;SR
zeT6AT6FY`~N6S|FniZrdu1_qfbmax?i;1=88Wzk$1%X>R%OrJn9D$S=tl1j=%RE;t
zgnv#y%0GW_C~ApWc02>|A(RZ_*>bq}3rVa2P%6E8p2xZG)7uj@BcQzl5ij<sktjOb
zf~1zEyB{mbDNVO2c`~q@Dh)C(u}l5?RB?Q5&$*?71TvZcA2e2vt9$aE!4uD-w~kB8
z64ss})F(IWWx{}BwJIXUXeb+M3(}ZxzR$?~rta>?+wt{X8Ew(q<(-d~d>`o7jq<<F
z-D4!3^~rh(8Qsof2Z-z;THU`-L8XHOHs`_|5qF<2=BLO#Wa^8Qx9AqKn#7Y`HxJAO
zU6<0(aSXTwv+i4AI~IeovH$Fj@6?gS+t2M%AI2Bi7i{`btLG<uDt%Eh^rzlgwv0tk
zxY3nVRzev8RdG(Q*85RPu4zJ@nR`fLyH8CK2Gx$14QQ%Wb3`kfiHMwhjnDjqZBIz2
z^G8=ui<UJF5pP@H^va<L@(qXO4Bgw=Wlx^(Q$C#uZXkc)EMHP{5PBBcIV<j5v(&ad
zUf0u5ZzEO)K(#q-2y&}H3$KzbKjN92P)Ypi;{L9i8w&E(V{z>vt`m>Nd(3g;{Gy1-
z%_;Rsmg19rQFrz0NSMPyZh77HpQToVS8EPNk+cYu%jMuQKI~~j`^%{fbQU4<7rqzk
z7$>_hYVUxKJI~_=D&tux@6(}>Oo7`KTVjy7Qw{2a-fDc(E%56VZfhYQqVCwa8YvN9
z9^F6TCJX8N{eFXu{el4-jty6pHx7MaMax9Uxq3$F+_GEK6h063!q?ph!+*`<Vbwku
z$q2)H;2SUU)cMW&(H~}|<@0Vv_=ewho5wr&9p-XDzt`r57U0ah-MS}6Ld#{^sENfD
zlL>7xy0u#R19JUjJo0K-_+gsYFDtl!OY0h+DX3F@N7SpJURMD3msPu7W|;GcI%HR5
z-9ZSPzp*LYa-_lX?dC&YsQwbm+}?sl(K^o|PY-u>AOkLgkxOd9VeS~yNK?_Pu6ieO
zSKHpbA6Helz${1DsjU3_o}!jb19%=p^WI-RSjJUjW`TY0^u;!j;Ie$phCr+Lutoyv
zzRg$*+-|e21-~jwLN(`;UGa)T&HZW5p{d5QwS+tQr_bA>PCDz-GF=>cz5eB;xwOEQ
z#<W^ObxvuhvwtUDu2O$>Os0<kGbdBklZy?rhHmLHOKj$85=Y9k_e!r6UKJ?M=gQY9
zivi-9zi5qkMtCGEA%1y$2DtWKB0YJ{mj>wonX550cE{AOmmiWdlT#tGx9bY&puA`&
zvh{BVt@e1r#k+V8cOg*>dn=)Kj$}%D`t}n)Y<mhC3i?3an-!G%3c?64G8PHzQlLs*
zcj{k1pOt;FTPhlz=yL6R3@oBZMVij9KFTZLBD90ZY_QjMR}NEk>?@qux~8Z>8{c#h
zFOIysPu;GewxIPWLo8O<Gx#I5y*f5OorMQ>fdndg+Wd?Ml1YQ^(gSjH&eUlJZv*1d
zhR$tbj`x>1glTsUkIS`mBm-ZYyQd1I%T7~#o2?Mbk2}V!(>qH{5wBTNO88}Z^EhE~
zKfcl)bFblb!t0(zTvL(nE50}}-Eup-+SC>ecEYk&RzIj`gT8(^;yo$w$;)#8beu7P
z5OS>FJuhzZ(UG|q<8FU}C^NWRLW4$Ax8Z7pL1OPT-cuFP@?Kv0K=gQfZBr?;O9u`r
zqe*j@jo3a0=!XdjwGJ?EzqD_5V$7vSX4SZTd(M)L15L)dwY(MCBRUk-Ic<{yA3#+c
zx1gw|0qsimT{Lxlu<nVfqhRx%`I;lxqcS+<7HV6i@u9%pFE^haD*PD_Cv)f}?eyb-
zK@}6uF?YcmGIUZWcgqt|yKD|Tk*%;a-(~@1)_*)u_SK~&K5?+ngoX?`eDQjSJdH8g
zCa;n`Y`Hg_P9ekCQr>F@s-9hJ^0W-W8ZAMjgCUv08Sw)vew8Bl`bhmMW%sje;?@oU
zq+MeW+n<-#E?dh(_g`(qfMNC3P$wPe-&aAF;t!`Bgy?P)Qw~mOQM_Ei^AYqBfiZ2h
z50@i|w&zOLZdz)%0TL^z$8IP$l%ki5LNN9(;!OzZ^wN<i`xk+24LT;~7^DYQF5PU+
zm0x`s>Zx}UhOcV4Ess9pks^72JjUhhQx$rdsI;@Gn%xM^C=p6tRGviy$woTHeQ)f@
z1G_A<MjV+4jL!cVFSM$dSm_@VN7)!;3UJyab<D#V&)E_oD`O?7SgaL&-ecAko|U^z
zkI=sUM?fs83-J4@tMn7ry4hy+2U*ixu3?WkC;E?kX*?GzcKcCJC<Q)xJXZ9U9K)B;
z){luR(^jPAavh_qw=dfCUaz{0Nsj@_f}j5VZ7G?%KwuDVFOONCl+5V`s<l(*`#MB>
zMK}bNazAIBTq9K3<2um-ryZO-MW`Hb983AAGKYe6Rw+4GOPq&@%vR*QVtpngn|a!C
zcs=c|U$T5C7sIzb)54_0S3)$}I7syKe>lHe%o5g+)1w2e{H4)XwaO;{itr(!ZZ%iC
zJcj%E$(eokNaVLk!x{~nO7Hqz4qx;GYE~RfmribYlSs)CvB*JP(GZzpDp)02T#N^k
zqP!4yC1ZK~x_{yb^iApp-7C1?f%YK+pV)ze)Sa0Pdr}gNShu=y+0tu(8;-syvaV-4
za(EtrW3);tZ&dEH?SxOQ-8V?_rTfkvcO*PX-E+mksaub8y{>H-_3SX&A>>L7y9M2B
z9OYEVORll*{QhXj;iE+$r_>O}TyRruiMm+%pU+~IO&6?=&6z`L<(r}JtUr<=+<)y7
zIfK8Bkngvrb!@gbzJO-|Cl4ZFKp8TFvU~(g0XKU+IwO>mgI`gI2p@{}pH_IO*iNw)
zdj$r+zMt@?Y=88h0XXnhpHPa8-_S24K?O4*{?*VIXT8<2c%0^7f7{L|zpeh;nsXxP
z#R&{6%|fpMlDAiEC%HogCwbX(O6gYd$yfUmX;0)@U&ROJNoH4GRk#z~Kkr5-F_!xf
zA1^#V<r<m*$yR@Pwvc3;-PTgh0eS}{$p38b{&Kp$*y-D(&CO+*O!I3I_v?yRaSmHS
zR%m7zin1WJUdLS8ABm8OmtLLO7juFM#$J7OkRVSPk!&tI>Umwli3iqfx}N`gz1q&e
z3<9(f3S(u&?B#CiR(j{v@r+aktuR@iaBnX7<4wu5h*|LVBIJ>TX#5pFBwhbJ|MKFU
zfx3yQVV#?x66Qv_D+1|VN9XS4!3=l0xu^+qaO=*Oi*smJCr#(Nqbv@?)xEDEDE)!K
zdrxT_Cr8Q<G?IZwg-29Cy07%txkKq$hkC8g41fguoAm4EWeI$@g-K-o)PsLxk-Dr1
z1=O7(eOnCZ!f2VcoZCd#Fx)s;{wVE8P_ADJ{Sm(-L1RkCXGC?K*Va<0Mw0O!79hs@
z{$@&W(@InD=-0Q(Jw@@h&kFv>r`on6Nb`Jx#b%GF#5{>4dpKrcZSLQGauD^ATf)Va
zzx=csjQ9W!1ig3L(})%Dbhx!h$Z+)%ueX^u0j{ZiS^XlV+pv=$D{{G=QBBqK=VW=D
z_B8ll1$ONO<t5s-%^hf_Dltgw`vn|5eTo*%b2)@Hc7SD2|IN5|bbN^=2oMtCGWzi8
ztBON^TNgK4zI%_F$VA$=BYowilH(tLzb5=8Fo*D#(@dfrIUImP9?)S|_#ethZQP2T
zG-svYn<Lo$Gj+5PV&DVa(Xys2gg;-ch&O??j%-S>Y2P<H?usd6xrys7-k{uDn(=hN
z(rJ8w@*wX;`ep^g%xD*#rs_8S_T~Huk%68(P)Zw7cO~gWqsyb<ea;W9o6>4eFH34$
z^SR}fG8Wcf7xvxAtr<Kl@y#FmRcYzm@ok)iZ0Yp`{PQpUr`=QbQHy8YdF?e_2Fjn*
z><0U8=7HJdPA79^h{i2c6Bn3a_uEwCy7$s8z|pn9de1^#K`gtN#0@XebeI?svrC$9
zI(mecxHW^C^X?k5VxP@@1`w)Myx#uAx~T~#B1v#xJMS~f>(Q1F&D!@cYecgbHiP0e
zB#2OZ7_;3g!&_uP8c#7NOuiZ6|A(!fU*l)0BBc)RG805F7+uQS=*+q5)j_fh^App?
z#XA<#B0E3wO=W-3@56kli=J^kPx9Y_f`$&QA<k!9Hm}4;snb}<J=FeKm<=RYr-!MC
zAEPT9VvCAUr{`Awo{_?ZIx<w5@opv1>?dHh_T7ABPI%l{V$WgHh%4Q#*%w=B@%bgx
zdrjz&{B9UcpM{U|pt)}V3whkBSidiRXr43KgWOF=F=;;yjwJPE@5*-7<SCX+7?)Ay
zEG5(T_wVqp9%u6rWvfEd9He4o!$JavvQvLzqxdvfc>U$|O1AMPe+n<dUp?~KHlmxK
zmAKsx0=$3v+!$rT<g1=47e@}x;3f>ELzn&4s*2umnWN+k0Ri<4zp)ALH+TES2F9S<
zPN}KyV#|`81>G>GR%lOdi;uU~8v1^pRfuB1PkA{0Q2CGn(Rw2{{|g$8P~?=Gp*l*$
zi4@@TEN|e+LkBf2H`CE~W~9-vpBn<(rI4`o9ruyLv8ikJa()geCgj0)-eNIG4b9g!
z_kULyM_G+k1qmkRKRG(E>v4%5^<ryrB{Uv7B@}g19X}AhcY`Z*8jb;^4H{2JDDNL8
z+8fU-k18e6PJ(4mKDtrNJMewbTqsenEhk&rEAbGlO;w5W#8Kr{cjI@&COTTSn1-Ac
zEE5Z|osJ1XPTVye)REspk$9?tmhYVRr?K6=a%%3(WGA)aH0eq0b2gZTEN{iXtsuz2
z)dp~5_+V7FxWM6LAqsYTGaF{Ju8f&1J3TKl&&`5kU}(fADgrQ579z}(2@*VI@qnve
z9~Z`d5&KTeNws_OUK9d%vJD?>)=9%loP(^Eevwc{-}YL<#kkR*3}|2=|FgATqKxy8
zksSTxR?Zs%ryT{YQ^3nXn>YCytpmpO<1-t|&U>Jln^FAnnt&mf)h10n4M5iWu!;%G
z^IUn~5(Bxa3X@qm^+#WbFCW4fM%+Us<<k&KnP}HzRnfSaz6m(U3dv1fJD=NU%ryLz
zlR%CkCJ@j@S?=#6cAJ7@Ftid@xGKXl7SQ*bfuN@i=KAM6I2H;QY46BCYhK~BI4Y$t
z<O>vjzh88X>Yn=S%zK;VctztK%(|aak*PqrQaLN(4R>#&1p`m($X)&#EYYtwyq#Wh
zJl*-GsDKdm?`&(tXyq_a=(ba0;)lqr$Bkgm!H?TP`06JisKc1!)Rrm{o^t^ET_h@&
zj@>WrR&-c}ahl~nErMLc*+B@Ab4egNR~aT}oFIP~&2BVrDz}w#mfLf7`epW2f|5tS
z^o28?&01aMHmSo(fblE*U!Vc+#6Xab-yhm{UQeN}+(%j~M@XVzW*p;A@oUk_sP)?e
zL>Zv{cO0#EdX_e>uBAvN?~+%l4uLfH4$7|Kqm|#oLwxJJhD%JKav6_W6^o%H`?GpQ
zzNB)iIs;{U<20WhFayqJljIg_D}A$tcPZK;ucLX;X`#xATP1-x%){9>?Jk2}F)MB-
zou^=o?8C$NIPFwf#Rldy)V#$uxBhO|U{*nOKGm7+j&ppA17^qEUR1~`Mn~)Wo*p+L
zu7d}`kQ<T+4H?*Xms~aR(U-XzK11P0&El03rMO0n2_kI;p_la8Pr?9{_}tI@Cx;Jn
zYzb4XG>10m+lxz^QtahYk^C3*)r%oq2QKNSM*%Y4U+nfimeQ>-9Kdo|9gno~<P>(E
z$uCI%;pb`ZfLUR#Q7Vsuy@aQXN{d(^ePY%QgyYQ;bMS=2)w=3oDk;t!?9diN?m27M
z@}6wU@ESJmpc5cIKur5{yF&4fzpp)zxtAkLSD=<Zdds{BvAB)LK6ihK;XXWU=m~;&
zWY&>|XN^sXfYJ;eaqk11pC<i`#&I6{i?Vvmr~v2Rdb&*Gs!o$>at&@<(dxEDb~s2F
zXGjXG=rj3|wFy=i_W68ONZKm=(WLlHi?{)_4DWKe!K&Dp7d4G`)}59ri~!Eb*^QtC
z`Ox{Qgh>*Au*m*xIP8mMFuSql#rNdzxs_h1Jo)tFjm$b|2={JBrgDYZS?Fl$vKT#l
z?ox4_U)i<`_>-5fOa{<ye*}`qE|z2yoVgI>46@`?+rkh?R_ro;LcJfZ$~9-5wa+5d
z_S_exa#T;wDX|K&2En=e$A))qszS6J2#sy*H(jz&)pk1(Z7a9zH(4jY`uF~dh9>NB
zgEyLJHF)z9z_1JQRj4XC!#-BPh!8kPsvC{U-hSILh9g=4kGC{ThT2}dt8tQ)b_&^u
zbSTXby=8>&(=TS49`%WW=zq`3V*zCa??$j39UbPEC(qoy_@f!673Rf79s+FJ8jo%^
zFs3;r5km9A{4UZy=QD^6d=wSEXY!Q$v@xy{@x56j?)`iDpBQ?i@V(;F2SptojPIC#
z?olJn^>On}U?=tX{Pkq7$}d{4V?_M!w2s+o?3)=;YYUmz#>g0?n4?s$-<&x*P19o2
z!I01YKLC?JY`;7R0qCJ2_*U~68H$$bF#Y2O(agI~c|wk_3->tL*X%pXcjotBBK=zL
zq1g^@urO7f#%*$^t}`9a=5foK!P8C+BOR=vd!8ZZcDp3)Bj$IT`&K)Ohl@DTGWw;<
zgOq#XgAe6~t-qZnJqLjh=N>fzn$Ea`YYZMYg7_MF_&B(TfG4ZdeS2#kjrz-J+HYHK
zHad%xcn<fOGrjb32a)trdT4qx4ot#Rr|FjY-D45zc{jc02I;^e)RP#m<^3k$)})ex
z*4@rqB+*N)a^N}c(Il$JfVtIni5cEeoRf}T&$r9lxUJdRJp3*6nFqpy5P}|7!fyiC
z+nR}^(hpU+^)6{=E?$q<frGL~!UyL4=ixra>SNpG&JK4IZfC?KFCh&aeR008zN?rM
z0ax9A-cH|=Uk=W;M<n9<^MNM6&)bP8xImK{EAtbEVmokKJ&6o6{;o5bElTRRzwb66
zUa1M6)FyBMWnMe|z%kd!nP{f?yuHG{IYrac3@9s7DP}@Ds4dhbXx{sSde~>DpU#Re
z^|glgRZyG1+7I3RRzL!eLudO80r$;M4RfVn#u?9J=6ev9urCG{p@*S>&XmDegnDG<
zh%g=w<;-RKCdY+GR1>6aChFzPtg+L13%JVB0@xtBzaB(pI5Y<UC#!m-<j}lZH*l1y
zBWOAY4+*=B8x&U`#WAUsfOiom3}HKBa+W+BfO5Uk>1PrUim)9q$rjdq1}*)J_H1?C
zF0jcEbf!GpuD}G%YZ1e<;~hD9LZk-e)_*%RP7DV=A>-r%wFeKgez*rm2|D(z06Y0q
zF!J;Zx{o}W^QzF>Q<vzW>$WwC)afk!2!MVOruuGvuA&_FuOlbt)ep8`u02C5k!KVt
zL(WG#B`?g4?bi@aw`AwH$Z2#;u6b|T433WX$qYJkR}r{lcF)}09Ryt8=c}EX`ZtE4
z9z9@cnupNERRq6ZlP53(Q&5i{uu5KZ7=9dB#-UPfy~JMVkyUji9M$a@cPRT59sJDU
z{XjT}!qvZen$cW=I3qfei#Ya@aLprkR=bm)6BM^wQLXFeX+>gwn?1NLHDM7MVEfhT
z+4+yvYF&HB_vDY1T$zu?3=MAo&C{>3<GG+FyH5o(0`5WCZ8;feXX?pN9cNiG-BfSA
z&6B<Qji$W#FxiWoM|iB2O6~N}oJUTslxOp7xSdgyPrry0b$&17Lz(O6-AMZF&l8;g
zRIu|#5>(EjvUbjcS$>}lBjCF2`;L;c^yr$Hz(-O3c&bmVN<tDiO~*s1>=tFylWBS|
zxapKQiTzUDM^4%*Qg2yPkZaI9f1Mux>}Xw5>ksM~LY?Nm?nHekM^Z}7C3yOJdWKsv
zoR(9+H{di~R;NeYI%<Gq_DZKaX1KJ-9syk*OS5`hDY=wWkDzzb5I?FNg9$6|j>B$e
zy^%~^>@dcW#5%XPN~+aE^U~`iWYYO?xzsc+T>UIfGIQ&05~`>lK`ApY>OuE7ikI9Y
zjY)X-F`|a_Sf2#@8E0+ccK_`{U-QUHSs5E=HV*)C26yf9p*u1<i)5IW+5E2D83jov
zt4YL~brvUEutIHsk~!_r4AwG<8nce(WlCBsL(nZT<kzcs67yY-LMA&Sh0Nxsgr~I~
zfwi5I)5)o(@5?0id%R!G`yjxO*F;IK39Ea{In~(UCNF)=uwBx01};77fPK54qP^+)
z{s;{)F4dUj_%7YX26;uC)|yg<U>N0C-yH1TvIMsYy4I~1@jDJpUn(f+8BV_LI&N7u
zg6E<A)3fbCQJk)J2W(3_0@KWwc}Gk=5)-tib0$>#5t#cLf0NjX{aC6ygCs-EDqc}T
zt|rh}rT0ijph;kXqmdBhU|bl2WseMLGP&G;ram`IZ~Ma|&N|Jm4i5%(TAC4Q9BT9a
z2wXg*Zdr2Ho<W>5rK?|HKcD&?7A~N#BQSC5{!jv`9vZ3jpxbkxI*n8UPJQRq=~}m6
z%9Z>>(}`hn%X%M$z%-uh7x$!h4;n##p|&s{jR1$5cB#f3-*VS2OhP>ZlK_afhEA9~
zr~8U>Ov1C8?0w}Okr*RM{PjDZ(_<zdC_msj@@H>tw^yby>v=c1QwbzR(1Gb&H@TWx
zkGMSX_Ji_qo)>nnHbPgcQA3)JlPjL>47irO`U>6ch)XSb*gFZ&*h`eXla2t5y$6ke
za!&mknbV!h199g2ksn%fj|j@lJ~XefOoGi5Lc{4uM*fTV{rvO`bLqgmRXaUn@ra^2
zZc&uiQ#Lo+CimdZv!Zt<vw=Gb7@-+%`}naOn2QLv$<|q%(2#GP_L^(_K3za2!RIB~
z>KWB(iU+JcBiJ+U#q^6*XL6YM&9}&1sNJK6mCF#fnRaxnXCS4CJ-CkorXzQTCnljD
zMc&NsT!yd-^=R_M1-h*bPnYd<<eVX#W2sT^{YbeeQ)0l!zFi<4bwIs6Sn&$3pY`Nk
zN$o-JXUk2Hqr-B_cN@Yb=fFIqm=f{EoXWf{nH@2L-fxNf4(@%zz522~RC@_}wLNYa
z^rOj}X_?Q;eSm)SFJ1Ekr5`z|>#okksvj+AgpqdYnGv0)Njh<FA7^>C?^bWwS?7_I
z>Q+Tk$@xt*%%qD9tcT9~WT?6UV-sNFM7@G#jIOzTNj*?MEqN*);5~<&?lgSVc#bV@
z^BRh+Y2mq4;gKs66{PJs@s!;qXgat;(RU7&)bwyU-Dh1=r01o&k0PKMRT<i+wxH=L
zGqDIY##>|4e7djW4%qkY*zrSjdS&YI5DdEV8^ukp9K%Cz{E3MJT`fT)X@k>^7in!j
z&S@r2XT%Y3nt{V6j$c)yY|F*bowC}2@p2YkQGCp>+{_2!)b239kcB&tA2c<;<lK^a
z7(IuPO3UGLZzB&`J`-}4K^)e+v<%%N+R9aN;c4M#i=I1F6Hf1;!`r91-0K9EZWxOL
zdhUifT_3ppGzBQ1N{>b1bbkhgIDkPTRfp5X<r$BhQEvYg<@LY}8n`^vcst|YT71gO
z53FWXkm}Lm$cH!_PBU;I6jj<`9`h_eDi}f2lVthM`5BicAIP;(XZyp3hi7-HvCu8w
z7MzFn48yd1t81TE&@+G!luWCK<~6>>UV!_YK+~;iRZk5Zbc|cCi2Iy)s@Q&?c6%0i
zr|IOpTz^^THJGlaNjVq}nwuEQXG%=u{$b)iC+)Zv1hkKOA2{;p9;)tg0<HNH2pH?C
zbMx>BHL3Mgg647ymGb~&wEO$P=BmT_>FMBh$ANn5oHO3n$gQ|Vj8b2znopeDkS*Ke
zyaOV)BxU11XlUs`#e6*QGXiFa_VZ$9x*jY7Zn|#L{KnE97ZXbwG;&s$V#fQ3!Jd%=
z$f@5|Jya7`wgwNpvzq{ty9WJH7GJD0UNetd7hxQ+b`K3p2kPORoyPvw!VC<lH*;To
zVFhY@wlgg|D3^H-lnd90I#)(OS_Bp>@orXREpnP8yCPi-aFciH&<uJA-E6Bl&3HnK
zn3C>p;xvT(BB*2yA-OW_7L*Jhho)D=@(^}&1r56!y4~Fb8fn*{Cs~uzbdy@cuK`Bc
zNTJ|#<y>BQ+PDbPhX2c<dg~=M{gFuN1+oa0#)Z1tet5wy;+-)w50zWNS|Jjp_6u!x
z%^LJmt?XD7B$7n|_lV`sQ}6^dOQLJgbvtk(7&;wsn!Sj>?;a;k^J^dkzr?a6v){6H
zxW|dpeap5YA%JC#Th_GtSehiSBQ1{}6<@z(&?e}1j}xc)oni)RjCh9JTbO|wlJZ`(
zfG#3M96B#0RP3R<e9A5@p$6=dla#oVC0@j?v9GxZJQ=fm*V_a9>}o*0R6XOE)Aktz
z)1g*WkZZ0-+lg;)r>U?`$eg61|1Xbw6N}SM-~4q#*F?1$E)wdNt<yXlUCVE#CY`qr
zcz^}XZCt2~Se!C36m*gTJm|iCn55M{H20wwvHM33c_g5OO2v>r;kC+#%8cbHBQGt&
zh|HaE-)gt6Se~|i8Z(u2B+XTxm>g~&HBT1uq4Bty7g`n}Fr;V6Fae+{G30vX9g!%w
ztbAzB8iM-o&<8Z{b3!E~5Ss%DN8hnP=~aougEB_Pa+7^w3cBA(bGMmWc6rvmL)U$X
z<yRyQ{nQ|yR?ca92roiYy5ot{dAn;fAQZ5?9<zu|S!0FrtT0J;jTP#Kd|ljiuaYzu
z1Fow_rFq!{w#c_~Xil&li7yY8Tej`^9yFs4EiPZ@?>_}c$r+x}cY%1G2Fy3ZHRKT=
zG*5Er9zE%WoY_RJJ7p-hafc?=TUI<Ly;Hqi#Thc;xc%CD=HyVhWz}yJADRN6%bn}!
z@ioJp1J^o(UEQJzS_w!WR4&}o6Y8y?4Ov0+dhXKWO@CnObX?dYR?49{<G4$Ao}iMT
z;lh0+QvDbQn1UKVPgkpjC8#}%jt}(<TZgXO0kc#z>Arm>#aYmv;)I(<<t$EYja^E(
zpcQo78r0*B9vb^dQaqY@(*2=YC7yT0qY2kQT8Vd)i>Y{y%^dL&`$*zx7IVbofyHkh
zXhOM-b?s+yGOD*Rf{6!BGfzA_+>YVR;3ZHQHEDcef**Fyq<MR=bgX#isY43>!||}G
zoeSQ$GY34KA)$n13fgY3w9g%{VfEW9o7FYi3Uf|yOI#z1hHfl4_AQCGkhe*oA|*PO
z2@(W<FzG>ar&c^$(Mo!Ynth-#1QHsJ9LJb=3+b{9m7U|A@wfz!hP|`w2YSqd1NUtV
zr}d<IpG-oxwJG`OtZc4j99nO+$Gh2K%w_Q{<{5*yZ>5+x3F86W7|;0~$()?-I^u=p
zHjp$AVw&CBw`{ZTIH}&Ujf;<mMm8BJj+9C3hjI%q=!N)aC+rC)SHTkMrK|6xp(+Xq
z1xAgL(R?*z6c{<)y6z~52h%=OZfmCXm8AQ2jNtlN@9lzyfpKyuCEW26NIZ6+=RBxM
za|ll6f}|g~LeSO7WCAq%CWae1bt!(qK`-pbr?$mam!M|MYtnh!8pomY(4BO)9Upnk
zJD1SoOX&1gi;w(;x3^L>#L11N4P6>56LTi%?j?r%^U8^PS^A~7V|tu-n&)qejuH8?
z#4?gu-JD8oNkcgoVqa!=)58#`HAXz69}2NAThXMq)?QoBvMD79f7v9VBpkivh-YpK
zC`X$~Z?`5VgGp~=i&M3r^ES3*)Jtz`!f{T}bs^^=32$SYlZ2$VF?)|aX&gR5&o=g<
zF*bx+o|^9=MIFJC@5|ZA>J0l8*ps&0y@RexJ%-Ssny`22`5ID85D{2%5OWnIq;I|%
zy?r9BPSE?~nTB$$c_TnPFm6uo)KH+Fiz?{tDc1{D(%V%t_ay19cB4}#yj@TaSspYR
zBRXbTBdJ`vie9>tQd_*3Tw4i{S?7f{>AAKUv?uMX0ZG@T$j_v=3!~+n{Rww3w^~g)
znf5_%Tlkq>czesWmwXQ?(}-s|>L2JH<I+1|+nu+9-i~RR<qveterXKjiN%<YHZI&#
z=QlLfujOk96H&x-w_n;XH4#1^Y9D^hw=-_Nrb&+WP<t?`4<!xYrR&nQ`%70#*^z<o
zfP3o$WoUgVkv$XYg-c~WRC_7%QZ4!AhJ`5v&@5|xLuC95rMCGETd(tlTJLD*(niFs
zR>wn!zmSJYlb70pU)~x>fY=_oJTPX^o>b~*1U^tkYY*Lh%opmN^dD%Jb@`?Q^||x#
z2lO0Z-?ZWe|56X1@=ZBinGfxC-O2TXiqu<nCMEj%AE=6qht9dnOP5Fed#LR7QX#9x
zOI6!T`O51<wd#K1=erBI34-urf?tY({g$n(d{?`<d|%risHcSZrhU}?Qt}DDH6W_y
zLuYI92fBNC>2fcHZ%SsR@zT|`KXkXz{y_JbZw)%e{(3)~n{~6+m#(J*Zg4##=C_<4
zW&*P5L;DY*^nq*b{ZPVky%mft<L$s4B5%EBkCnIH<za@uDG{{(K=;;n3~)btsj?II
z(xv<R<@f<-Z7=nBKHs#?B5w^1^X8?Wb?*aD-rt96ec_nx?ZEu8Jba)uk%z9{%O9v`
zvi+tq`!g3vf6(<0)RWSEQ|)8EWl0$EU6vFG^IK$Yav;pnUn*1H8qD4<?WMZ<TZ7pc
zt3K4M34LoAA}^d>hQ9Qp6MWzab@)v^W7-2{FL~(hTmL}azvP?h)Mv=C3-ythgzhYL
zJhXa_&~HlEhd<C&ABjo)&iV{F0X*w!4;`M8@tbnf`K@94QRz2b?O8a8;@OKnR7QWF
zVVC|6jV-A9)@o9vy|wBdkH{rW!gyrn^j*pI(lwt&id+d(-g-?8ek%r>YzeEsbd>e2
z7(mz3mlEjmtzm9?hMcSkvg<>+j_?P%_IPX9ZNHDU_f~zV4t-uQ2I3=yun%?XLu)Vf
z4^+!|Yaoc@OS|v$JXGqlP7&s_uD1s4ivG|!TYGC*wavHelJ*D;Vtw{!ogzO$A8!qN
z_D3BfNx&#?4JuCs1m=H6OwLKzH#F1a(Zz^wKgS=aywltx%9poZLi@a3utjFQH4wpI
zzLdz|4_#g6AE?THM?vPUFWnyA?xC{frP{T;HK=O;emz);&pgLU)MuR{255O<8;^h{
z255QKDWZVZ^3u`jquOu3<S7p&zvf%R@*}N>u2)ed9BA8be~AM<9=gXnA3Z|tqm2<z
zc05v(NUx)}m+tXMO>!>`eQ7JNLPp@xN6{om=-e-KkJN8E#(F!;cEMW#u~+1wX1_iQ
z74aM84_&~Q<)z#GI3DU5H@+z`J=RONeh;}~DeAY-)Z%9m?WIfgt)cbO-m*>0x6oAL
zj+O6)N(uT^>s!N+`Zh*1^|ykl@6XqK^bjS)QTtZyq4F$$T0BzZTf^38mMei_`u7*C
z1XX#ycBqVRp{YdgcqSQChrTt~zqXgM0o7hAbHCJj@GUf<y2eXMgYnj|s=Q2|X|8^C
z@LRWss=P(_-1QwsHCg^PhI7PsS@#z$f1s-0N~K~b>UwFb`Yqz2yR?@Q<z_yVD;NEx
zt3DH&xT5u29fR&FZw+g-ZxdFDE!z591D73Ny4&-9nqOdgppHi@kj|m+7q0QG87TQ2
z#xvq1bQs^tn3B?A=v!}h*MNtT%I=|~&2ME)J@<Bf9|C9Am#(tpt-<t|@=~^B>r1zM
zFH=f-2YG}O2_AO)16|*1t&-wlKOz%TQD0ccBc3r7b?Hl&elJr>q|x%oOhS#8Z)Hr0
zHCpyt!}Jr)hmvLBp{rMU`)d||d21l@Q-3I#Vcr_f?Z0oz-uOFYTyXt<5oCVo`=ylt
z&o7lR>O)(99+>P8?>r~{!?RG5{$W1r6!{<Kek)*qf4x-KAG)wS6v;Q0D&?(#V0tel
z{mgu+N%NQL+}}2=k_lozV@@iF{RqwFq4MiXRmWRHo6nRc-sp>V>2IB5JVQXN(eiD;
zLRYOXU9FBk&|RNZifE&)ynVGL<`KF?8*S??s~(W*p~RW^qU43>&lD5)6dNz~P!!*^
zwK3ir%m9{`_D8yUuCV7tkRD>boi)wo5pCp$Snu!GSZ`A{lOM%`6cJKhx|=);8c8B#
zz5UW5&-5Zq#C`^xJP~`ow6^`och5vr-?DA!xB05ri7J1fQohYsCEjT1f1vo+KXlG#
za0oa0NIRm9*7p$vU*eZ)d!c>#{cVb@5AWD2E|g!CSfl+}qevXlAGLtY5&ao)Qb&wu
z#K|2o#!Gu2^48FGmY0q>zxVu9Jq6t#sLXHcRY@nX9<_jo{%gMU=lSV{CTt#YNho@G
zd!Sw30n@MbsTD$}?=PhhTI-%~S(A19rLAlI16BQ6pr9N8uZI!<dc1U(FL?=_qpX*%
zQuGhhgWh~oY4uB9I^A`@bdOfw8nm?W2fEt&9!Za}`K8r6{XSR*)^EcR*h=4Tt)>{&
zm-aTse(4_a(lKU{huXydLm%dv>!qvBXUsi#PkX71@556l@k8FSdwrkULA8DFkviSq
zr$w#!{k!r~(zE`7uDu@E+B4?Zi0?edMtlYV6Y<;pbt<L26_A$Z?Ez6g-?D2Km4}*8
zXm1VpM7%xQz3x9yZS|!(+xUKIgdT132P(bvht}TvAE@@#9@=Kt2X>DS^U!a{4&dm|
z+$SBy_u2=Y^G|Ip&nEg%nflf%eKmQhuK7|^7WPZkWchljypKjg3X%R&-H+5HA;s=5
zZFR3dQ2F**f%dNL4^(x&bWGj#q0=K;J#>HXFF@CL#1OGZi|R{PZ7<!e&bQj_S%AN(
zdh+nU(&<<6|3FuJmMQ{}emku=-MzoGjs6TeIVrmSfv#_-73gSVzV(j%?X&`&p6}qB
z641B4bWM4wv{n8<_j*R1v=nQ;6rr_Wx*i!#S_*l-*A*F8zG)r1tcSL;`&*VMxNlig
zDjzY8src=<0v$u=OQ&DKdZ?6Vts)L-8GoQ!-tNk*<sBSCE7zC45`!?Lk0L-6()LiI
zkhbyCTHB)@5Qemumo^V$^H9%W{!RP*_FRDy9{LYdzCBl<lRVEa)mg_w*PPFgi>EyJ
z1NBU}-*orY|3G!U9aHj~e$nmWEB-)5<^yYghMd$C`<Wo5rr6I~MQVz@-Ws}RG5V&X
zYkTV@$=iD9oVs61obEr+UB>%pGRmVK5RbH$KhRa5AtxSbU4NjvK5G>bN!#-U6Oy#e
z`cm@6JoH2x|Ay}x>d6G<fp}u>m#)!XYM<YHYaoH&Tb4v5Z&@jGte39&%rSCPtn${d
z_v&w1!aaRGu&40)_FQqgq`lOH$ovD{^1V`dHm6-*+RJ{s&`ou!5B+uvY9|BFALtx9
zUy4iYeyNoDQt9PeIk?$HzUiLdjw{X6UDR&{-!nzjZ#&krEL8mi)o;g@=8^l`_d0cg
z-hbP%Ckp&C<itSl;}2B3ymZV_-g;*pe}M8mNl(07eWA|!QknC8T3nOmr5WwxFI4;L
zZ&}rEzm;}kmwYc%{ug;hoD3B*U)u6k{a>l>?`5izp<+J+Muv(lf1sk@szJqE)cyym
zy6QtsMbo!lX@96zc6pzHL<ss&eUy9PdY+b;r1cr`?$Ie;YI?!9Vo<y1Y~L^aamD0A
zd+XmuEOhtr_Kai>Zw<42yRA6ICH<kRjc+3sx`)109zSB}F7tbpf_li9Z>sIvKg;Q3
z)KwBytZy^64!@kdU%I5d6mNZb>6ZP{KK8e2P{~xW#~-MtrTwN8Uy_$@?eDXZSL3Bp
z-f2#7($ZeaorCsLEzh76oU}eOo#3Q({DDe+geJjB>$e>X)z)6x<}?3^P1>H7irA#R
zzI60H-WtZU6AYd6`#cP~-W^D&jAx}HU&Z%vv{Rk!rF*^-eR`zN{!&@bN=3ekHU2<#
zJ=y^oE2O-%)knmVvO?BdZSV5E{Rh?k7Cq^gzv-;+M8{Ut@>a_gmi1D7?2fuT@b!G?
zZl%4gkzVR8OM0pA_d}S{`fbHRSN+!SD`86Cfh0`nyQmPR^jo?9KY#sO`Sl;a%CCR@
z+kg8N|NDRZ&uXyzs=xl7%a!rxfB(;)|HV<ojZovi{MWDl>7S-F{VMVQpa1^f|1%5y
zFMj{$?))#R{YzK#H~+;m|LZ*XzY$2ow1xjKivNp0{=@%r{^yxhn-?(cm7%gf|MjEf
z38H2y3(fxLkH2$qyq;h|r1B)lQ54C){wP5xNX5F2BKPk<%5Yu4_XeWO{ritH6`QdW
z%p$vU!Ov1RadMWr;Ag1}!<&Wj??21jn_OAwf*++mn|e{|1Adkw`8=^wvQYm0XBmQ^
zqe9$57yK;6ZCb9YXQBN2&z|Rd;>_Q}NB?|z>UV|CW&M0zKMP*=quhOGKMQ`~kCOMz
z{Ve%`KT3XH-(R}<p+C!h(qaEB`=LL}=DYk$(F=c;yf5@GML+aM+0SeJS@sA0SvKG9
zUy5G%v*dlne<}K*Kg;Hu{!7sdf0TSA-~Z<K?vf@0Vp!!KA0pTJ2$DgW=?266StL=E
zz!z{3H}FT1%Xk(J@OS+gfP;s@X{+6>%V_aa_D8#)qCZ;xl>FKLr|8cfJY$sN^*`eY
z)+h4%k9z+|DgQq{HQ2xY-`>D~{Y6lpIFrBrozNryTk`#iVf(}W_Orit$%sm`&Y!P4
z%lP*!|8Toe>EE?`#KS+_ZdCer?H*zC54RhY{$0CAa{bfq_f-0~?Yb2Fr(bZ^`nN5+
zP@VsD%Tep!w(K(cpMJ?%>)*AEOX)w`DmDIXi_PPl|I@8Tt$*9HuS5U%^G2<I+p;ff
z|MbU{wf<epr$t|X{;77o#(%$6{?A)IOW8l$?5EEEdAq(Y%0Jugr_N`2eHP9?udm;$
zu&t^8ms@ad<j~Up`ybKd|M9PnGD_6ShOhn6P|1<mfjg7t5n8^I_JBB&vQ97=AQ_;5
zP4$E;BDQ?zoSG<(5ibd*t2>iXPC#%X%j330P{paKF$;;7iIOPmgeqR1R@^>1k8#+t
z;Z53O@l@8?!L1@*%xqoB`gC$=N65e)`BVre?;-2O^S`bOg@b8g_+@Brg2{KZfARcm
z{X*9W#gpPhhqx3(I0f;@$tV6HvE(bpB;oHo%~?RdXHH1F9YMi}B|kWl_JDLfdo&9V
zCw@glvYoj&kS1{4?j?Us-VBG$ll}(Wgx(atv427d5=9=rG{Un<pV`aJ4KhNjE!p{F
z_Yxym+KSj~%hR;%Unc-sSgD?GqBtBQ&7KVJPDHZaP86?o0%kfS*7t>r)CPnI%gBZV
zk8eaQCPtX{`}4MOwIITm$2vc-d;UivcddvhN|<=!umfd8*iE3{DJQ^In<o?;vEv(o
z69|u$vEPB4(`TY%?MUfL==j0mGqJ5!#FryTyx%7m&uiMVq!~<B6(OiH0uF)W8%NK?
znj%_Y!Y%d+2q5J_dk<860TH<*!ga>IpY7#lq^1HbPMyUwx|{%kM2v47#u1Du;`zch
zN6y4oDr;oH=6A?Flv2cyA}+j#MI=5_k1&@&&!SFnqsW0rtoY739HEUe@j6gBY!(k*
zdt!bODt?l%jmSiq?vzmR9-)u8LL=V<(~&bcLIsuHvmxnZ6d)$hDkDpZUmf@0KRJ}R
za5bi|x>H^LTr1~q#OU#aL;(dcK#U$w8VkgKiv%46*tvw=ZDQ--0vW7Ma&r(Er$-7U
z!oycDw<KatgM{->4|H>kF(SjW^-DY(iD<MivMv8298CnkBPzV#6ek?bY9}x?QQ=MU
zdcjdv$%ZExa({U&nSFb+I#E`)r(;|flA9H{Ju}6DVxrffu86!zNO-b>{@N$f=}ag@
zLfD?)Z)V@MbJ9jwmK`BYfqp^UBd|mkMB>4d1rbQj)`a%N!=a}P3r6NW7@6Z*z}D54
zlI>wKLlkh4$7X=XJ&B0?1cE158nBnh5<?t#(u@H|CRGIDd%82b`C$`5L2Qy8xlD-m
zPI4&Vc5je?q*Vf%WNO(-`?l>@L=ague=VNShxj2oqWls7eh`$O;2$NzT@&fO;39#o
z)FSNwkW4JVIkHY%vWR&=IFHQ!4&0oo5zS+#RK_8ThzA7j$XL-tc_;bAFJf^-bV!&d
zZXo7HkBk-ses3J85nRJBc?a}-0mRmr`}Sy%K#mw15}`v0@6Nw^tc)6~4p(;ODQL2B
zWG5F#hs3!kt!BYRNO2(1m$Mb&YVP)%OZQ<m(X4?U;3=42{w8ciulX!W2xXW_wUR&9
z?X#EuN9IT3ybsRL2tz?~>41dR^|sfDD8E30#sZ0tw*xjpuKx145)nunxH&r`fP_H}
z*ax}Ui4`%lWy#@JzKInPS!ZFyn1r>=f<lD%&aoN6A4X)C1gZ}@vEn1cF7e$Fuf%xB
zj59=cC!!t@@seG)&!~Gzsgdl2*zTkR@JJ4AWWntr&<M^z;P!wzIW6;u4fEtmAh3HS
z04@O=b_#i1tvD<rG=oI8N&>r^?Ddi+@nZhsYQ;4sf-rb$vjgoZ#NqI$st0!B^8mS_
zm?gVMw((dCkt30~?gWK#sHr&NWyK69rh-ILH=??eD;l_GWCuV$_UqlmP8c<-Tm--P
z<&hJ5&4SX2VL*t4(obMiAm^g6-Deiu1eo;j2$2q+aPD042X>OQC-i*11VS)ja)yYD
zI7BZ1YA(3yA?y9;rj^HNa1F=4MBo|`EE|2DyTCvbUNwK-xJ2tvXUInG;IIvNq8*eh
z**!HWVGTxP4JMMi4;N7lGEgcuKt)=sr^Y$~>Ve#~1KOHe^XH3eYQPp@N3j+P=V2bu
zp!O^%emBbF8R&)N8Lhx^XExqc#47-5zk{TMvoPWoOg)cItoG>83WW&71qx?70$F31
zFxiP&Fx$~+U^wjq%^W*Ii_q=v?+y&58v6xDahQ%sFah)n+yo}z<T~KMGI6--ywI<2
z6PBPw;9;V+;|}X^m!1tRTuFhg?-Q~YxJWAi)X26%(02S6fqkA&5{BO&Y(zwBY${yl
zoUOUCL+ZD-)$<?Ag^Pp&z(|CL@#R;x2~p4@#XXS82HPS33D)lSwe_SyUI81SqV?q7
z&4P`aDe6gFB`fT>Ksb!N#<1<1*q)#9L<q;i8$=7&b3J7SoSatw@)!ndWR-JqCrdqF
zN_J;;GGwSH#Hz`Z6ap~GMr4+A%+p!~*%!~UTm6KhNqpZxAw@3n3~HoQ5zkX8enMd?
z-N>Jt;aMDFEuNpB1XI2rLCo70kSiyrWF)(RiG4*dTM>?qgk{>v8vvBPFXZdybR=R<
z6O?`Sa}+ga&6bCMas<p9Cu}!N&a+65)Z+wXcl8S>ZC$7&#9+&#h7B)Xd&@J2d=&1<
z_^<bl90owYi%q12R^!cZd4`Asz!|Atfu8mrsO>(DPErLBKB38x+*s6=Ef9~XaExe}
zpb5it>?~;No(;5zS%}xor5T^OL}`yiYMA2_EfJ^@oR~1|btC5p-QJJx3>4xF05jVc
zhC4e6u${jhgn&`McN61-m8Igv$QGCI^soDh8)Oq*fVd1-&UD|6QHPnLoPB_T?ggAD
z1%U4ju<_|p9X`)|#-5-K<)Y0I2Pjo;HW3EvfqGYW!b&bbhkKkg5xB6KJRn45>Sn*8
zk?;c8+tCJqn|uJknZ3l@iFg7)Vk89|g>VAu`N(Ge+y=S&ph=6+*92Sld)B}aDL;Uo
zDG9hDK^+?#!`qg9ltLn^7nWf{PLz@L$)anpOk3#NA7{i#y_;=p8(tq>)ZZKn(j@s%
z;@EXcvQH8dcWb210Gi14pObhqq0P<iKi3wC+JKuB0KgFz2;tR-oLPmF;0W0LYBf;S
z$~yZZ{y$J-JB;Y+jnDx^#fTIkFi(iC9%^yqugUP#97b*;!l-+K1=xsGB0wRt1#rd&
z8@oAfbps_59RO_!a<(J1JAu`ijp{ahO{T6fFzLGHZ<Wi`uo^E6HzvXbz}kp`#EYq&
z3hj|&mBWdN2;v^~qwR#YEHCd)oo6!?Ca68mG~Ob{2(kXv9sE|d)BO@Pkd$Hp$0SUE
zdUm>!KPPi|z<rqz<{zeEwE4yqVg5DaRR&a|{5xDB`*2;F-G{tIJbd55^63kk_ZiJ}
zQXk6JBAF`NT7GldVb}b*l2F5$gg_L}MCh#NqQ(UmIj!`3UtDjAe)oCH*?=!SFf!Bv
zmAC*7Be|ZRR}kADXtFM$DbI5S3>}1vEsUVUocb6{dS-^f5Ez(PMD{<hW$SBd^mCM%
z4M`)gzjsOFgAn|8dyjtJF6}|q3Ww`HCh5TA*+<WQlEo|rDe@YyrPtW(@)6c}kr-U{
zJbFuAVlzi;yq+)x365TfR_|{&%xXDC*kL17txb{^Jyq7o_6ywcI4M#$JHS(8vfZZ5
zB-}fu<*{9Wix2>3{;sSM&3nd7ANe%P6BnOvLKB)Le!hA$mTES%{bsbodP|zZxcd3{
zMy&nZU*TS~f4X5Q_zopO{QV=}@g}*w!qtC=B{Txex^=`|ZLxh1^yH8(Oj;IfYy=*m
z6N$Bf`#cZ%*;X(&6p&%PWqqTBNR`Np?z^D-$hwu<8z(IL3~!mKs$pGW*SJhAE#Vp(
z&7?|zb)`q#<yXQ6D!Ugh%u|dgPPskAe)tIjmDt$6Kxk-xoa7LQ)R&TVe$G~_-+u=B
z4r)NDM+?&YMhS-A*~a%ZD|YdKo-y86x7HFiKbi7%P>#f5x0W@};SO^T7${S65(S-G
z&#)OAMcXaJv#xeIl9G+y84B1YDp|lq#4umyvpO?^#y;vmWD6qpdFKcbC}Ac5{kk)7
z#%?^@#M4!4_Pg~yG@GC+^po2V`32=@GDq(q>Um$G2eJK?P>Jp05ws$U6Wh$wZ6-9P
zawZ!4%+*)H?%|-4iR<ue)yG^R=(&#z>RNV^n<yk!hso^^dPP0DUQ8hNs4F4)T`P>t
z-az&$U^Uc7_O1I-qM5AO?}bw>%=gs~6TjE&Dr!H1%J;X+VZD9WMjHp#*b!mR+vKi~
zqpcb7fSw#&{WYB2bpaWIqg$Un8aBrp`=_urj@Sk6dIp!nkrvRDi0WfzD)Uht*)jLz
zb*|}0CPUyN0za_#kqr&M%M7gLY*YX@`&=H(Eq1KfaQB<cZhS;uJ2u=)KWcOJ3*!OT
z2#ciVo8ZRDIBIjzayu{kOll6fayU%gJz6u#p7YmAYDk!;&kVMXIjpfTv4d}7*BvBH
zG{k4jAtslXJ>Xr5r{Su}ItHJ81ZTk^aJgSyb}LY$<6Jb6dd$U@$Ykj8EVR%4ENj3`
zy!~Tq#=GT?$v&QiW-H7rX-4}oeK{b>w(Y|?{*oq*XLf!0mF0*4n60m^$C_SpRy(4Y
z4>HSo071DAzKc&_Ozg-}1{A{VV~cCOKiAF{sCrtnp50UvzIsggq3I3G!8)jfGQsHs
z$HD9ZNNkh9uy}pDV4s8P?P|sL`ZzMlf=x*w!4`yz`U7pK#(^s~p?*55cHgxKDE;^k
z+Keyh=Q=y-BYXoSc71@_kC!PQOt59L_iUfU{v<~nb;Qog4#f;8;dUA)f1G-BLh6HB
z$==w@Izv%*4mdYRxRtE#<r`ymjz@gJ&#qO#vsYz0Z9ED`Cae3t6dMdPp`I-|_8XHm
z;#R;S!5fDt6_*CKs&bj+U|Vuat0fx-{kP<v8wr&N`fP~hb_&NWbDteU61ZkF3nAe!
za<IWzrF`E_Vh@3^<nCGl9Fa^5sJJw!C#Klh&Yrq+drxw7%tESOa)yZv+w2U(NV4m2
zdFptFL*`MQIKnC9h5&kI?||-b;hT_hj1LZW8bR|ttZL-h1hTyi>tSGV+A~OV1uS!B
z<DS`;{ZEPaz;S-tXP-2($8-ME`q_}RSsRjOkqqT*8CnlV!ZmPd@aVBE8xg;7%+9SD
z&RHc}2XeM^+wIQw>&AL^M>XMQn~5!A4SSE%qd}N_k8`_XfVdQbM+2~h&j)aE(9W*&
zjw#8F5JKd`Mye-V2q8kg?^)+aa?XUz^N73>qCq%NRo^a`kt>@)ldT7q*be6K$=5J8
zn!}#~wuy}oydQ*cWL+Ij-<vydk-j>%<#+xZ{l)j&XP&TdOw|aC&#wMnw;}QS#lR6s
zb2+>k$ZL;r+%D8Y_6Lp+?i@5IXraK!q0QNq8>z6&f(uj^1*U<n9z(nlJ!PJwixqEe
z|NpGLOR^=oj%C@;QTPPn3ICrFM7yLKHGphnH8M&hS~>mV+5mX29p3QFhKs!CZnm6<
zs!FAjAPH<Ra8*0xz#S{#%+|zbYV)VW<AS47VD{q>;4XF~K_d0|`I)EfjD)kL?67C~
zG@u{mn=L>^6a&L^iKDMtbquvjSbI7Sv*Vo=_S#ut^_FyowR{(z!>}y325Di(Mb;U6
z#mL|ox3en&^T3+#(@33$GpOyku{bsB2;c*|o0Pa$5=IQO9Td=%Qv-)uzAww6S>@FO
z>x}vedFt&*%-7f)q;+)fmrkeLG~yslJ`Ma~+tQ3b&uRCFPlGdbrJJ3{d8$U>M)DpQ
zDU0<CxF!8O&Z7m$%(P6zATBc%Htc7C@z{9m*r8#oJE~kLA?qNql_<Eooe!Ql-d6JT
zg+kvo0e5Q*9k(Tp+*#Q++dig$F__vRroU|ZY(nT>TJi2ypx!N?2A0rfiSQXHTo~pX
z>98f221M!Gh#xn_3Ikhf0nehN>-cv1kuGeowJrI3KYY*;9E}C^J#!tw$)RB>DOD6a
z1MxK2%7`i#$+rZPXJ7;>M40bq<lAhUAwiP<>9)mbERNjS-|!r(jL*$Ze$NbGUpymz
zvWLN?CI`c8U6KSz!E8X#XfSQ930$mhbr5b{g^5a=5wPlCgXTV>f;H-RR%o5Tar~`h
z2=&5VrXRRRYD3@|e20$aXO8O(1q3VYLb^9J6C2qKn9H;C0teX|G>Xm@&%FxgNkBQj
zIb811Fh{=^-)#?rN}Ud5kwvNo2KH5K!Uaw&7GTRZ5$bRaa~2ptF~H0o-4FFK4q1;V
zyT!EeLZ@id!rUW~Mu*ef8Cvx<8@6tBcLt1~xS1v|M_~7CwwTm-3JRtjm$Ah$v*ws_
z(_^~VL}3(gIL-NMN`5=omM;VD9>r!_7ANM0gx87Z3Dp`1UkRpmC8cv{E-{mh8#C-!
ze^vQ1kgepRGwNoc>igNO{E%0Zr^<HkUM(@vsW<nP5haUaeSvA&@__|zHE$eqbabXb
zUmzJ{9Ae#IB)NN<R{99k%W%qZ;WONKF|-ik8l$S?k`iXN3pZPvK%T(il;MsgH)}lg
zQh2W?pJ@SgRInr}6?eE5+beT%&bY~~eQRKHr$ueAu?c~(iLa-_%ogfjW|n%BvwO*M
z7z1|tv5RdFFEx3VQk=oczL@-)_UA%X+e<*Dr!j~sn!)S_GH|iD!O$N2!j7|<4DuLR
z8VN^+{HzwD*3MYH3O2W5Q2Aza**vmhlx$iY!^mtKzg{d$1ha4apyiEiFCVYjKX4$*
z1FUvtgS__{cg4;s*GG%a`TEvfc4fpcm@#(*)AUZ%*>UcI&HX=NXXG{JOL2s~3z8SC
zt5&vsBOo(@VBI#SqU+6wDt|XLZyk0<CBc3#(+9T!%k|z5GIVl`J44WY#xrzogC%-3
z{>aPx@0KqiQ0`84<7bX((H4te3nt&(Ld?66S%Ed{$hM^iB;6k@yZMEZccVKjId1yE
zG&~Qvmd5?Esg9v!&_3wnI>#a2lH|Kj4})wX%V2gN6Lh?LHpjy;+$#wg@60iZ(vS_a
z9D#HbIWw+iYYkMqD{-CjaP%!J;O6})<5{lpK%+CeFNqoUsB`j8we#TaHL%I)Cwmll
z2suRgPw{P)<MDv!r3jWC(-1y-cq}Z%1p6Sy0Bg@vl(^j48GThShr>L|-4%6tU>PM!
z<}>~ZrZA+@?UvdCu=z7il7}pFAx+s|(9~lLIhzhYYD2ZV5`6*m1lh37R-iVJW{zv0
zq0^jNabu6`j67KZ;6~lz&c|e-3hOZ!O|piVlkGfjIj%D|KGSS-&2*pfOvYUfwG1f%
z&&@6Q?RKeoJDeQXEXx659gaoT*2R13w$B5T*D70dK76yn!Zi_8{Ja&J?t<2U_V+w9
zIdGA%9@uQ-9F6WX{)&|yCft%Agu(L!@#|wCTJxSE_rk96T^`5ofdpJ^%mZtTtBV{V
zRq%8dw>>a;%g~!hVc9JbgxGVRsov)!?CNT+)_Z#LZw2`<2q=AIzCS;2&s?oJg|R!@
z9NCzw^+G%@Ae+pv6ayY1)y-V3kBR5g1j`Y8{<IssQozD0gKcA_F*Rl*gkmCWUk7#u
z;4PTkc?v~teYb2}`lcCd@!ywXFGj%AADB<r-1fL7bhi_l@xbQ68Nk6bzvKXQ@BEm<
zL$I<T6L1mW42)fRq`1!rEEs)glMBQ=*^e!>;t}`2gAAD5_d6rSjT55W<HHCLU3^M9
z7x1MhG;6Pg6NHG}fpUVXCOMzJ-GS&NBZ7_~swb#Nio2RC|5i$XfM*SxMT@)8LZvLK
z!;W?@Wur66Up?_QY~tRqw3}FNorCi!&Qx?*8gO&Ss6?~AqX9i5#F!Re;2#2XsS88f
zO!o%X#6d@CZRfbRdKL=Z!WPCVAG%SBhn_LS@8PVwdb4L}Yex-4OEh{kCr!!ItwSfC
z4$zzS9vSY6OEM=_(-(uHyH}&N>=pEU#Y3%lX6${hcJ5#s#e3z_^ZBaHN*=*_wSWqu
z5-!_Ajd(uXatn^Gj?E6?ZkO5|$qr%q{Qk_~RzWJhg$_MEim8UxYOVMF*CxcwawgA9
zcjq{J$yYPlyWmN`RKptX04X~0dr;uYt|jPMOf^yDaoVLD>vma}$U#qENs4os+5Hot
zXQr<nB{TA^m(ZFjTIZ+U)>w>l_PY1F^o&eJHT03^ymW91p#0VM8!N=ZEd}#5(pg=~
znI5o(FiQEClJvrIO84h0Su57*hs}b6TnNzWzG_f2R`5+8>lWSG)zJy{j(1-5y&Elg
z2F&uJ=Y*TUkV{9iSfMi!&09e!yi{co_svz&g{P2C({KM^;=`90D)%3!O*^%m(eLT0
z*`edu^`awv>c!&f7CKcw578C8HYHM7?$ypTuoQcRwA!tACN|ljxzt;BYV%xf2EIms
zo_<T*nsT&{4?PVgAB9iT>_#pGoEb*<I<c6RBFETD+82(cf_;0C5AY0Ax~1;RN3lg3
zgegDu=mf0SI7-0hVwEw2Rx-GYwoz{lmE7l|*{w8aQ|54-J4if#sXifSz2Dt2c9pSO
z!EgC<wL!9HXbF#NQ6);2cgY^NUR(^+u6fsc3scK`V4wR|yU1)ST1vT1b0)SVn(iAo
zkaVhou~(guZmIY(LbC@Rh~iO=AI;qHfp)xZveDL}6Lk3&38od_eJ5ycDyVSHdM_o@
z9adMgcPINtRh%Mugl^GBnmp*l&owB4XbN9E4YY6BPIfj0Fz&f;y`v6`e;lJmpCMpr
z(B5oDuFust`V#HC*82>LWK^PkjaviDFg-81S8C8r8YJkM+dfUR$si<3hIpfF1yJ48
zPGz+CAck(MhBI}>@S|OaqOzyy;wvBWL=((2K4(^SBtgLGYt__(R#J6{wzclX8jCex
zZZW0~BI$;Ok*+jv(XOZgd7X`Qs~w~z5Y2ba9fZ?$i%y*$J`SCz;fuE3?$stj@}fD<
z`()-ti-+7?F}Dn2>Im#5ZBXm?Y8%~e(VV^7N$3udy(Fid@j*?jwqdkgJ8NolUKOk#
z6%;g%fw9~cL?2kNwRwMEQPvgB`E#u*w|2$yuCanPPPKupe=6vxp^D~mf8PhumbfUc
z^tmmaLIB&Ww^GbDE1Fx_s@$VSOAhNwzSlbM^mKTrHgbcaHYj>lT=y0>LVXHbUblZ7
zipHI2FFl8RRg#(zw744*^!(rMx2$!G=7Ea^u#a@<8TpD8E!SM86K`v<yzf-V!VgtI
z=x`4{=ozWAYCwzk)_aWCrni&iz4C|J7Id`u%=qrDWUb}0(XG<6gpXNw=}jx}d)`l%
zU^9{*$Zysic%kRh&qmV>BV%ckb}F8Hj?cYrh2TS5k8aP1%$h--yDi!ay#`z^roNyU
z1UJsiXQY?s@BMw2<vaBS-k%dZt)Gvb%KdrGtPG)je=4Yx&`Id$HJ*u4PG9&a?{$kJ
z;;-0yA~9~YXJ(#O^sc28>75ik-_wXsY-3DXME-Pl7vPyZhNACV8~68<nG%*qx^Us-
zf!Fow``+o!Y&SzWWVlq$Gtd$vEjrtka5y}?btV^dOIm`B;^E6SbhbNRHuMZDx9oY{
zTP<6XdrpZ?qC)5REF{p`*3lENd4Q*=Y~-oAz!zyKZ4c>75E>lg2};{V0u#(8yWX;>
zV;GG&525WQiHRYXxEhp93gs{JoA=iU9XNtPTbcJ__)BI{-?9)S76$)(-!6o09D3xl
z?-M3!+BmJkit7*v()s1^f1imsIO$Au_gW2#=a=a}#1xn%^})2no^^FY4=rsStrBvu
z+Na(+m^G&IHP&19%rr|kiz#sL_ApbMV1ud$J=SeAVTzg8r6Wmif=%ukO^CF}NLE9H
z4?@pGZOz@*jP9inxS<+&$@k9Ay>Ba9{X_?PhLNII(^41{&>__cxf^AFV!1jk)ljA2
z8<dY2Slrx4j|bPPscf371}p>&0Y@w?fJ4#~#Fw}mGV~Z?*5b!v&A05ak-JHLtdV+9
z64M&(9>oUjnMa4ZW3AvsA_AEDhn@+4s%h~d`p}I43AE>NZ|pQn)sv;-u2#_VriX-3
z_|!gh;)M^LdEeXJY9~eR^nS0l<?%4JNgpEuTY<>H6Z*4DbdW0l52Is@Tb4ar)v%iD
zU_ejbpMtsFl5AaU;PePcVD0QPD;&uqj-ZWwPGI%$9>K{RH?`5fF2!kAml`V`3pZd6
zEP#D~Y9&3eE5F1z#}+aU96-7F1f9bkpx<c;dF?u^y}O44339|GSYt7Xy0(J+plQbU
zC|CBbCTuH5wfNEF#{unbcxAczXSM5TpBUE(RKSr#abv^H3)Ho{v{lY>Z<Wu*YUa2c
z+V5D!iV%|?THO@{+NwRn>5}_itZjQ=B(x|2`>;w&cLz=2z|~|bDx_Z~yL&1!Bf3^&
z*h^|yas=M8$2P3Geja(;vS(poqCC0ITa@H9Rzd!8S7+`nqPF2F%DI(9R#<Vzyu>$H
zthI3yZkG;}DyCjhjB_0$j^^gvHTF57i%=`*Ofg$9IAPX6+t~LOh=sKpO8wke8_m>@
z?yt$NV>(B3j9X5p7gMnS^}?YK7+T$D6FNCqP<-Twp4B-9SIN0zosXb)VVZ}I^nnZK
zJz$MxCwkfq7Y}IeF^4hpvQQ&xbH_FmbRhf=9exr9#rL+0w&>dfpR;U5?~E{5(lxc}
znLOq#TF>CE-}kxeEn5deuEI<2)3OdE;(l6qXTOvfThZ*>%E)Oq)GHLTn#wIY<bVww
zE8nt1`XSI3JA}w-6N4y9Al9WEA^}tRRJ#aDCX9BacHt>F)U>s)Ps?z$H08z+(<*j6
zVf~8D?J+{J4k-4*Im2y1b4Sw(I^91B%50|gK9k4(P@+IY%MJqvZ6~%Xp~>ZMM2Sj(
z<~3ttoNNSCy`1bua@OK)7z-~h(a8oiwA9ZXjM`1yD}OA!NeE!CCL$IqzC({?xzx18
zKi{jNB4&bPEscA%HSLWjHi82x&lJ(sd)M2x8`nzF8H=5a?~+NI?Y?4&VQ&CU3us+2
z-{T}jpa}G15iiBji7uRBF2&nTbh<_ev~mh5+GoCHTSlBkVdHoHU=@h}0W{C(psWJ%
z>n84eyK}ZEz9OPCcC9k|M`%ev`}EC+LubJ+1X|~(8RkOYLP759Q|}^JMPhI}w&ODU
zW!WX-=KkE-MjuF#^E+&r?Z*Ys7VXH^p54}&NuuZDZ^-sNG{vsUq1tC6Mt?<iTA&P4
z#UZp?@=VChindm7t(7ARQDu#E+8T~v^GAdN^cY)pm>k}WZ)6i9Xnm=n$rSVqW?L_(
z^>m9KtE5;3^_oKH40dbpmR*Am%Gt2Wp06z|36xESwKtk7_jz#&Qw=r3n?x7RBGBq_
zD7hrWgOb%ylGjj@WVDPXcQ6wj))|WjkHbo__ByM^b}KFDssdXWuLqKIO)rcP+WUK*
zZv%7jO>p4A;}zkD?V>`^UReK04q{%mv_vkFqANj1L;M2T!cHz6?XYbFEiK>MIx8@f
zKRl-OxD;niJ?pU<^1a&H+7#iyDc@?z8fW^6Jx(l3)?S~{2#OG$+Kyz|l6<vL<b+?1
zCii0y-P#C4>!W)=D4ga|1bVvHrB&hc(C$$Xq9{N}kt5ax2b#iR#?j$(UuOWeN>lzk
zLf1Zr4rf=ek=t|u$$YO)$j}z;A9$`^bfM*Xje;9R49SVa_)z#6yA#v&9Pg!e)&RlM
zvF}@UO8~_{I={mc8a8f`@X6GzPS=7F+_;edL*x)j-aextMFYi6!YIp|OD7SUQ=)Xy
zxr+Sk^<Kt|;RL$T#+W)J9Rbf`d|3s}mGYqH6=kDw012ErhoFUBmk5f;q+U!95f7+Z
z<zw?er2NR)bo~@t1&NF)p}o(It}}2Ot?rNvJ?}s36`TcAq!mEFk=R^G%*1e7G|5ig
z!Ws9XVVS(~DuK@8wiFZr3DYL^ZLxxiW3kq5t;Bak{2_4`&@#)t7@<mfwJqF(p{?Nj
zhf2<#{a&nb+#<@sT+ohIn&^nVrh*}fU-FcEP>fNSr|d36>st4I@eMUvTDkW-{e@Q#
z%7;l1WZFUfg=ez1?p>BkUW#hJ4?UpRLMwzg<Fr1ri|7YM-&C}^7%Mc(X8~Gb{ZVlD
zn)ZF;vC%IbsAnKk#A#{Jdd6<ma3vZ<&y%ag&(N7KZEabL=iY7w{S4j)+vi%Ou(aQo
zG4Jp38Zw%1w__cMv&l4%g8K}!C+p7X9uO6_3y{&68bs@Do8^r_8{eY4CL1I<huN3<
zXLqWJ_R>y@o&d*!8Zc*88PL?-5Q54TV=BVQO}~QBW2r{xF1KtsgSRAo&ujM6hJpei
z{wB~KCx8X9N5bc)#I-(U+eB0YTIQ$T^U*Rd5oX<@+p>elGDEsnd0aZMY$PkZR!r^e
zB6742o#WFCb23Fh<AC_U#kV0yB;2K&El=n@+&uIQ?2MJ{pvGQc!R?px_Nid4=r-eD
za;pKR9P<(m#$%~kjO|<Wtml*35m&!k*3@s4qSH#isyrrBjveV%Kp3LYx?^s|zZ#tl
zsfc-uo+xUsqVg$s%H(7;-NBXRAa()bWZTsuMJJ9i&=p7Oh2`LO&F68yUtH7qK&Ql{
z`@I?(zIwLN+O1$0eC2?(`l<)I*yM(m9D^X*j4Wj(rRJK5(9D@PwC_8cbGF2WICRYz
znK+#)dJ$(4Fg|c0x&a+6MrWiP%QV`{zVs|`w_3NiI4y({NGDojXXMacyqlj{vEHjG
zH7zi$a^YzaQ`2rc7DI>5Gmxu-w#E-SFdV0JzVp4<l9NC58mmN(puYj^^+Wd=$@O+x
zyR$S}#Bw@G+IZ`ij$uuHrnC5@7Q<gRdV2Y|6gdZz1P?c%!jUyAXXx6|Llzrus(^LZ
zz1T5zrruc%4R7d~my%w@Y0VhNIHLD!HFtsp_GTl<S<B~J(gJ0nWqfXH(78op5nosj
z5{4{Mz|n5Y5SuGn_W0ZulA=aGF9)s2RP`E4xI2qZIi4gPKp8|V5?KylAJOwsi$s@0
zT7@Hw*e&fMy|SV#olZ-Z--#j|`z=clx0;5?uv@n36atD=)1@164&)N|^j(iaR9;O+
zS1j%e&N$WJJ=6!HtSLH{?|vQ1Mzhhr)4hd3Amo%A<}E3gBe4|b67ETyjX-N!_xh)!
z*c^3i<@m@p3xKNEa9s?6yW13n%8OmUCPEu+iaKM9AB$CfH>|;N2d|gncqux&k6=};
zZyoHsPG9FiC0Yf&YtpLT+VpZ#3lGrB2Q#hXr@DP$gU)3Qo~nBeL%Z~RpylQS(<nF8
zj5sM@imR{Hn>^YZI+1CSnE#H$59lOlX=vHJ?h`LUq=9qY=iSM22JF~vgb}zb=8EyT
z;tGk^iLwTSH4Q~jL9}_a88jnIAX@vlWeHZdvezbn4uUnIdF@-a5^`?zHkT~p)B4a<
ze1D_L4xcJ}+Q&`4iqVgXhr~$ZoVaR<QT1%6+pl}I!8a#5Ja|Y{dhVj1YR?-l274X%
zA(mnlp{~^}I0!reo@xBjUF^-E3Vs{+Ox?|Ccu~|MZ8pLD$t;!$i4sgP3qS{Y9%!`g
z*mA8!W5zNXTTjr1+az>_6IDghpGdtS7r}Y!MK-|bSouQ-&m)@pt@kWFlU~DF#zt3Q
zD+@DA2`FMdzM0U5To3cD;TTKX>8FB;cD3L=f~l+y5%+-Rumh_!O030aitUBUUMGhM
zqB%Z)c2JY$)i`68+h)*-+X-~VMrI|{AR9|Qio@830#27T29|EU)`M_B6^Ajg?i#G3
z#PJJ;qhytUPP3nfM8D|fgdZ9bX3&m<jM03biI8)HCG4-NoD-Js<8<S3>BP6es7Kk0
z$}<3(+J&cccENtX4&q=qkrs@tH|XrvEIstskLmm`;;D?z@JWYql4#V$@1R-7a6?;>
z(YZKj^E@WKuTR;c6D85uDXUFK41%r4D@66*uT|r4uW~AhT#TseLqOR?Se8d+lv`^@
zP{D<j?`;isT#be+G?dLs(G*@Im31L*C!*0~Sr%>^DUp-DMTwzM(RTOmD(gbBnuxkb
zQYBW6+J?gDqaUJ7nXD(Bulhr?<bt82-wHA-)4xncI^Uv&Wu4J*wW?i?vPQ$f893Oq
z6U@<wL<AO_ms*W2CFt_eFNQ9BGT4I!v9S_#BzxVc2lfkfx-IJI9H6YH^}1S$y-H|I
zcI_@ID`A(-n!9R_bGl=otvhUcD1(&fYVqNc48T9}dzc7Zy*X@AEj~lI5JE=`M)Enb
z$<?-of{tO+5(e-sTZ?yZ9Qn$n(yoEF`Kh5bA5m(H^Y=Q`0R@lUT`wHom06DEpypiz
zEu-FQ3COW(u|$K;w(Ju_Ie7+UYtCuf9~yU?`o7PwCFmp?Yjdb%!Iae~c1|h**vVSF
zC<w<X*-Z4#a6i9gvjlfUZ@)*uGTY3A#r~mZ0PjX!q6}I)N9WKn+O4-`mdT>ACWLCQ
z5ZdEhyoT<-XzUL{cd@Cd+|^7)$g*aqvFBHDQ>|tVaM*UAc+ToMpKATou&m#d&=D5X
zA&Y3cw{>jMuD64ifp-v`dYi+DOae}_p@^n=X_h7^G>sTnNg2@X^DQYhiR+T%s9OA7
zE9Md0pmFOG4bNxj9A*okN_p#Tl^t@?2x)=tS-fmYx8B~_MupBkZw*5ss-m}>lRIE#
zt6^Lt6hHN46BI&jw}xX(*`$$sbW$ZJFpHV&Fg+;Qah#@cZ){JFpP>;nohz$C`%I2i
z7pn%uMr_v0apVnC{vAP=B!wOe(P0W3*!sKCIPt0)3T$@DB|MO-4TwfRV=Gfaw%IA?
z_Nk#OCxB?0x2zpc4ry}UI~c7jhqnnaxvG7<2vB_J8DW~y*x{_$zJ&LQ=Fi2>0A%yi
zG3mlo+8)^;>Rr5PVKf}ZkvVjrEn*4Od{}C%FuP|}3;z?~vgg7B>B7=%sN#UF1U!p4
zt22b91d&q_?Jnm5wZsUb-rP;wJ=@_ec>0q~PUo%r!m%mST6gMtjiO^(YkypQ8vB9E
zTbq^ArDO;;>XI~2V#1x~3(p8^vr{VT2OS*Nh=%iFV?nrw)A09%u5BYFdW=2j`6_I7
z3In=ni{ESk4jb2{Guh;HCWdHP4m7eu7djE#X89x+UW1|7wMn6OjI?ww)iYMMX!q?-
zs6x#|i|;`+q%iS5fv(#L5W~dc5Owid=+s8IXdSm+c8k4+bf193CMEpat{ei~m6Hd-
z@SBEiMm*85YMahyhh1y9ZZPzGt=r0uev6*3r&NRP{N6aXW|u?oBD_F^`i%F&u{-@-
zG|uW9$N6lRiN@9pl!F;yTc11S6^Zp*uG5ra^>ZPeH43JBe>~X-0oLi00dTVN5pDa@
zg<)p2ln*7jglPFZ>3QwRu2n{K06EcAr}s%&0uuEq>c&EomghE@-3Dp+5fP4W>0{yx
zKpYc~6VZs1Xw@AQL}NAI6hLnoA9t8R4QDJ$2#$NN#|WAZUFutOM8{6ygVFFkD^>Kg
z#lF>Q!&Pv^hlNNcf+V8h>(#14yqrc%1(d@T;EGVlCg;CpaJU%;TxgPr#$?j)N^I3>
z@dZpnr5O)!=!}IGbg+Xc>N9$4p3y@v0+gVuTYH~REFOWB1m3sTOt;pS*en2!Fg2js
z=@az~n@~coEg9#CbqQ2Q4Wh-D=#8`l2$U$dP#n{=Ch<p4=Xt+*yn&G@5seLzMs|Yb
z71a9(I`<OGpQj~xoxPgyL2E1my~?{H0FLdirggVNn}kuU37bo!36ihOXhc{w6j+is
zcikUKf<${W*L@2*y?f&?aqB(T$X-L77(zMyGa4JZ4Ixds8nnboWJ5`l(=gF+^@7R;
zP&DFM8)BN+KBBeX8s;YaR5VQYM&JWZ?9dzs8BJjm-L96ySr|I2znmPq_>IVo<`uLs
zAdnQo8%s1+Y>lJ^(ZwU>3KS1O*bT>o&2Cc?QD8f0mC}W1Z?|mSIM#@|WD%6yAznk?
zw+0q9UN(Y>ph~`_G+8SeOj$LP)3I(1L#GF#z1(LY8KTqLZ@si$HJbM=yOjA+G+l@Q
zfmLvWW}us+AJG|KnUuJ4Hrm^zS(z?G@4z^v0%b6qMH@Psh9YXRAuMLy3tyRAu8w3T
zz??JQ7c!eTdx?&F9TYXULAS6Qv46Qaj8KU-<>@z!=1;9-l;rfj8MNDq4V805Ko@is
zo0-mgtG5Bp0^NqnIl|MZp|)AQwG|Q>ba)3RsLdaOga*}{d*u0{tJz?Qdrl3C3WPOx
zUopRu)Bt4@(fNU-o-{aOSR$=KuU<imu~Nab_6)s-)j(#0Zj;mbD$1=^3oOC(xq+_L
zI<$H2IOvc@oM=QqK-aX{>AbQ}Bb>ifFB{H?P2ZteU9+`Y?U_5gcb!lF7G;5NZ>PC_
zt>}sl-eMk-fp4Ggx#YMsr)|c-tZ@UZ0rC15j83B<bTbAxn)}dAo(R#lZv{wkcm*}O
zl7i-ky_;IbE!&tY1#^sE?2R%?i~6D+xNY>BpTU~VY#Vy!joEH&&}+R4jSWiZj1FF6
z`=)VFQp|dvrQK#hVAhM)bsu(XUsAMqm+Ksu;aMdb-xyS?#iGfl{m#i_Gvob?wB7_z
zod$?TFnQ<vw)sez<Kyp~D|a6~;()oEx~*5xFKo4Yr_JqC>v>B)PUiZ#1Iz+@HM#hZ
z*pt%kdJXZ7b}4&UBi!YA$5+I^#T0R$z-_bZ3`4Ug*(Jv2zPGwH^%+gK&aFh+qA>?{
ze35WD5slqz=w3chrU>HhECILoj#T|q>MR1c*<NS#^C<`_4L{NQut;?#A1ae4DTN;B
z?H%b)L`Slra)cB<w3lSIz}T*iD7w6ZbNF{_7>Z&P^#l*lsa-D7*rJ4vlD&oqCh2H)
z+JwWQ_`QYY?Kt+QX*Q{y4f;B!OO(hj6muwWRGXQGe0iN_lJFSyFBG_%*<RzS_;e)`
zwsCTV%0S0Wy4lP$R-7<FM^9eCY`5B0x|v?rg6`6f8eT!1wRT+7yRA*f>h}o-X;cix
z`VdNzM5Fb>S%%qN<Gw7C225}2RX(*I!`G~_vBI;ZlObVFzgBEsL#0$9nxY2MB=~z}
z)9*qbriauW9A@eFi+ai$=<Qi^?prB~&F%EUTDE<51YoaAw|DMC)2+3U$H!=okA`rc
zG`xn`(1DUx#%Z}_ndj{#%=)?7AkT@@bSc}u_7=j0wWI1;vbRh)XG4|2+-SWQI}7pY
ztq?XAbl2oHBsU#9(w@~8sbfS0MOWyX4U}U<um#ko=9^XpSZec<E8X80A5O*y&)QCu
zp!fL^eqtR3!=9YR=g}mO?^n7VnefLxylB5gNqjvV20IID|GGyw(|3Y8M-O@#D!<0U
z!MryushZPt%Mxkh4^=+b-zc~`^$N(Lu##7hW1dnJZVejW0b#sr_zOt$mL+v}(C9^7
zqDt1-S4*I?6x=;%z4ca~KREC10Tn@Lu;07wZm%4e?$7)1QK!cSQ12c)WAt-Z=d<?a
zR*D^`grEBjga=QTwyJ;2Jx>0hqxc}%L+q!c`B}Wddz^|uSMgD@->(kY&zih!Sco0n
z&su}Vqmf&o4u4?Z#q7gVw;A+<4k!XEMnkuyqnC><Z!}&n`4yUvlRZN7pd*U<Lb8)$
zq5C%@XS1M$XF0u<`g(C(38GGpg;Bg^Q@lk95=YT|?{pD`F532|g0-gE03=esm)dRm
z#aaw7sCh*9ZZT1z1+u(zv^(^gnuz~Cww$5FeHx9E+RhSinaPERWeZvfIdob-bZj%-
z(bu~*oYj`M_SFf4?&PCnyPXa#(A1)Av<)IM$w$f7KhHM&9Ypd^*s;TF(0ZF@?(7eC
zjSb{Lv$uD<7usj&4j3~Xian(T4RZrMP<fe%h8cw-X!Sv|MWAv=6tq@<4fVnz8`fa?
zP*yljgPm?(oRv?d$lV2vz9B&-pO;(t-0bPSK2El9mWIyg-7OlM)C=8hfR_8MU?`Bl
zXuNQ(Qv>E(>&nrM;`A9fhc{umb?!QgzpU^gfbQtVp}NyK^qDh$??(h(=W(|D8gldq
z%D#rvap|%6d`gl`U{KC1oR&+^in;_%m$tmUz3cj1t>HlCblm$TEk$?QE*&YgpnlE(
z9m!T2BbJ+8+0K2auQR@J9MH1cxRLw~dhAx8m+I{IVw=r+r}bV;$yg2gPU66q(m&=J
zpY2c@bZ;*=<_8|BHT#vqJ!0TJ>yu@TBbA}7BmTx~xD5~7jcc@bxD7*ByVtr!n-sAG
z;|OZtA>P~!sD0l1>ZAJ?USn1o_{n!Sxm-G2aQPw1h(zMhGoJdSR6~_x5U}7I_~&2V
zZJ}eh#Z_bY9m&s0jLzE%oQL~Ma9!o|stCgDQG<WEgG>!`1$E<KSX7R~aw|YuIB32V
zlod_W0#zTls%Nc$ozJoaowv$%#_piw{+_wiUqV=2(4BJ7TlCCk%`fjc4WX>FgXVh+
zb7r^QYiQgIA_sGHq=+tna#rMY-LeEe2K6CvAQyUja-8ZzXYqw;<lBSh8Vi_O+N}Yv
z1b+#o-m;~K>0CVtGxXf<44QA*8Xt2>x7JphUn*RUfZHeWntcH^ijomJoEMH9b5)J=
zP3R0Oy}A@PlrW-8FFd1n!1x#mG9NxGRnNQwJ<UDhzRG>7J;x5}>*!Gx?CR~EeX|a_
zXTLNDJ(J1umI-;*fqM1ILDz@Q6%Cf+V?qgG%}X!b+ZU%bTo?w+z!>{deEO^ccYL@Z
zPctMYu>+QXOg;|ly{k#8QK#kpe%_{yQ}v-doG~+n?;?~OR!-BcU@MuJQ?Eh#>ONAI
z9eqYARm-IT>GZ>4w4t)o`r5SYXhQ@;tD9XV4(kU}FSSKkpTwXe`tlWp802_pDQLep
zM#yOn8W00NF4k_*vzl=Gz3h9p!#C>=tM~V(bM>h7-s@aGEtFmE&sQ5!@V4Mi!|i|)
znvKvg#-#>PQW`rdjqyn756Y#tEW&2~3Oqpvw4}K%Ete{6+Nc|@1C^@PF721D<%Um*
zv6@xjj>cZ|UaY5GFKw5OJuVGMKM6SQlcA7Tg4Up9^A6hY-I|}2!CS#umbf8MR=<zC
zHM$sW_Ziqvj2dE~Mcpm?cUMAKV+;kX1a+eY%AR-75wz5@qTU$35V@;4Uc`k2HU@N~
zav-x;=cTvoSzD!eJNLc_6^oaXZw1FLxh)=FIYNW&Ho2eo4)Q?sa_NO+Gk2J7wFom`
zdaGUAy-c9@#<ueFiiY~yn4aO(>9{nWHe{>^*&t@UCG({u@C_QUC=af|=wKv-SZ)iH
zh!26z7<(e#a*z>X&0F-sy;I!Q=UTnHqY3nu9ZFj1^j>Ssy)Ugnr;;c-EkRduSCbLm
zqrj0?wBfk7uoc|r^p@S6UI+CjLn<`SOH0t4Zms#UQR<}u@fnPl5Ee8-q46n%%H}{P
z^!_|1HWEbK?j5prP?ios?@e_jxpI2pYW0E0dq=IlJhLN;ed)ZE{gt5iX7PHw^fuO#
z+SfPTWIPM%&w!`eXz>!laFJu-ZSEE@lbzs!Yu12aQ_G`xTsniIclJW*u{-TS+nriN
zO-(@tF&zDXI<j?G0*<tAbKjD-tNR6K*)cKAFsc=uR9=e2dc2u(A7CR``%=lhI<@zu
zNW0vAp>i&D+Ja_uIv-k4x_l`bkwI%vmI=R5eA9yFpa{jkv?Gl7(%JH*0nLB*iy>*|
z(AOpkC+`m<i|=|VI#fYt(2a=GOZUk8i=tvM+I(o1@Q%}d>2xoXKWLHY%R~1Zm$sm+
znA<~Jb}y4Z>0GzLKeX4iZrO6Floq}za*HpW*Q)&qS4#Pz=EiI=*d`+Tfg`Oih~mzt
zhkxM8ZGFQ`{Ha4<#2`?VO8SKo`1sJ%W5+L)kh_OAZT5ZBt^`^SW%nrP3OWcidg*Gn
zxjz(D@0XIJ`4`H5`a}0!Tl|FUzJ-9uu6?@rhZ2qY3mtc?b9-CbbMw>WA1c+EZ(75V
z_7B?9+`rIp<otv7vhJ7ugN6^`9~6i9mrkYHdg$19a{ADvwcSH;I(q01hn*i3G0JqO
z61NJWGnbd<`nlZ?+}BY1z`_atLu)SeEn8=q4{WvUUnmEg5AB=_K6G%d_R`t%OD6|9
zFXcewp^L+Whn^Q%f1!6L`$Nef(RX?29xu%OE?|Gqa(D0kpmXMz)*5HFf6zVW16yAY
z>v@%5=x4S3gW3l6L-)?_wMeHp9=a3~^-#>~hvKdNO(oeq+Dk{;FU^=0FJ0kp@{<-N
z|Nf?x>ENXdr4Mc8>AEO~*7R13zyE&dpx?YynS;OS<bGeeT6^h%PopEk-_tL2^!iQ7
zA<@TtsRY-)=`jUMduU0c|3b%DZw-|=!I!St%0thi&HT`wR{DjGRbR?h#tYZl4?L@x
z{L+<vX$0gA)<btq>21bMpurbKKQxW;(45o$h0a|1Lrcy57rOI$DW@0hp-2SPm(DU@
zicH*F1L<SRL+NArp)>!|40E>fR<LS&pJDB%?-NAR(>GOcRD0p<<1LzNf95ITDo1(g
z`AVJ^SLp83!?x7v-?FXF_6yx}zV~|;5$UBPuUcO^*LW$Jo61Y~E-zim&iqYx-fxrb
z=i?()S9^N7l6|H5mdzu5S8JKY$}e=~J|9|JUN3E_<hKUKtNzel%KU|{nqErs_3=>B
zX0?}&y6Zz%Y2_EXT7PRe#^d))wIo6JJyR*!-ctdBdrym@?Kd8VY|?&ceFi4r7~iwh
zl0;*ED`?8K`Ata(_11fqXz8u@T<qI4P$tJUexYmaw++bU_R^hpe{0w{v3}8_gd*jo
zJGGaheE!yOOpE73DgSyWKXQEPsP3W9d_>~VUfvt%&jbOS<=aRk!D+4O7rJYG>Fo9C
z3kXZs{R^G#X>o+GTYt;$-PU(mQcCpiA(QO1Qks0|?DMVREW*F7G3Cr1^{wG7Ild<;
zvh(J#-x`!%{Tuc=+kWX|7?4@Gzr7s6{9hVj&h>4IDFyP{ztFwQc<Ao?X>yi{`x!Bj
z6iwfD0U4lMYA-D*&0pxq&xnbnXnvX;NzwehL?Q!r%ijxFCD1PAt)Y+7-x@}FCJr=3
z>-P$l1mLZ{&p_PQ3rF2=tw?^pwQi|Z{YgnJ(Vj6=`CYyju+Y(;Ns8POeZKXcVes29
zAR&Ak^`$M1Z^J;D=f{3&*WT<y*L<3s1QYXn1-nQvvGOl;toGKh%6cDdt+n?<ciYbk
zW~KOyLM#;bw`IQ&6;1gUx>I{=KpXO<72SrH&iu?%sEX$C*3fo-+8kBUQr;RyF8wWw
z&PIDEIWEgfcWrMCtJlXIBh5rhZw-4s6_97*X))4FJmVl)Ci-KGk!7N{U+8?hIaww~
zdh0z?PknFUOfuuG0SSZt(mBgp13q%U&^_1N3bU{L(v{aUYLaJS&6n=F(^~^+C%#_~
zE5)auvryd6Jf%FAFD&gDpwSdf&pd^wXihIB3U9x3<)_I}6@69=3t7=p-e<rI|E&Qn
ziT8fbYj`Fq#a_3!-qW;?HHKP4dqzzOaeOae*LHc!Z)({}zG+FXL8hFy{edBf@h4p4
z88elk_q|!R4BiS(*MDq-?MHwQUGtY)Aa(6akI9nXXINvkmm=yhUb^;olW$gv`+Lyb
zN(=pM8$ie2Zw;lU@|G<-eGi%&Y0>mOQEj9}bAD@RGrt$>HP5$#X9IXAMGdTn&Ro7t
zd+2<oznv7_|JJbf)62KC!@s}ajehy_wL@q79yB-kCZ0|P-EF@$l-|or*?}rAoqfHu
zDZ=)ld$gC53;C^Kd~ZX;=F`o$Qayaz_Ru}wqU2bAn{nsW-iw{d;AwIeiaY&6_x?62
zx6+f0mzKQ0hj{2p<)sMS^oO>TzUL|E$WMnxUNnE3W6+i7TLWPM`JqIgy)}@O`lYKp
z?+1C&Uw*&aGc1sL;`_h}NC?40+02|TUG3W#+oYao`>nU8UfvoiIRPKqO8+*-pljs!
zA$l9_r6Y~`)<9_8OWBgmFJ1F{n%X3vn9txu_K8(~q0j#OCi}#ChD>G6dtq+R@XSJS
zkNwiIzo)5<#AtfPOf*K*w=uR68BOc0p`W4reFpNIzh&{Be((3J_|sbh(oglFE0^)s
zun<FkD0|~?lbvzl88gX0QP)c|`p++&?fZCYlYruRU?n+y`Z@V0o{5t56aAT|NI%in
zTLJs~<E4B2atou6`O=wZdTX#WWZ!hv`MvK4-E+Tm_w~Jo-Q=HGPn(l}Vm*T<YsLL-
zk3sitZw=)+JJ^VgzG$WT*4f(A1PG0$?*%M$<owc6a{Gm@{LE4)jh6J@>zQJnK^LXb
zGTySgl>Oc>@&DhH>=X6rVq~AF&zMQ}iI!hlN^5ToX8}B4`kWSRl27#KMUZ@=zc+g<
zn`dYv_r!RAKiYUNHPd(u3sO(a^wK4N!-uXp-uv2ZKHZDl6YFVka!;)O(p;VcedyZx
zEnC|DUcPSDiF^J+XZl{gZj?q-{e|Lh|Ipr_#zAQG8SNAo|31RRz4)cOyzqSa^}Q6C
zBi^>%_*H&UbRFw6N1?4*AF}|}()zSG@~@B63RF@!-xN*Sw}!fVdTHyA)5=CdyN_S!
z?BA=`y%ld)Ub-Kzm5m(zc$<MNlli63f#xQM!F(HRf6Y%NksD%t8I4d>R^PH`tXnTV
zuRQ-k_x_ropd0{?ho0f`%MKG=-##nQ-qLvKNZ+0Vq9cDbICPe8pB2%S*GpF``K@86
z(te?%jPI62aI}6|^`hSg%fS4-a0HfA*IVl`jPgrs`F2?mU2VR!wVvifPsjh|9{T9x
zrK9wx&B;G8%1dYaK0Jl4^lcK0j_*@D=q}%Tq@wHlw5TLfP<-#rL38?E35k9?6aY(k
z+MG4w+s|1eK8?T<@q76ziUiO@$!YW60n(pu+0oOU4<!-7TLV53@13oh*DrLJ{L<Y^
z`+jNYkCyohoi){m=3477bk|WHT6*0NtY`V0U;6FX0c`c@`=p`xp8KG^|8}jjnjF8-
z+4o!TF`dhN=^p*1$5>b|-FfEm()m6bnJDJ0FWu`IHOWM=>Pt&r;}<%=eO92g?(z%W
zyT7!Z#>o#QJ;*O~eeW+oM|*}LL`c)#FFnKgOG!%oR(n>}>!oL1k-x8o5NT<@&{3YL
z3MJBSrxnpv>q|?kPotBLqV8Yl_;y->w$l1rZ(H9^E70D0ekroP`K6=Jm(DWMFLaHk
z)yYXQ`b!zK)=S4TMw60aKHn=jDdwCH&26ReP!giQWy@IOEqjd0XP6$!{MH`Y+U_qs
z<H-7j&h*SxLt)In(4F4fojs?waS&L}FMUT0v`8OA04dV)P^3spd+C`x9`gVx(wbgc
zYOTLerE>qKwSRlAKv9YQh0br!6=;|8`$Kon?V+Rhr^(4l(dRF8j`Y%1NBxEF@!r@b
z-|817E5$6o&^h}9bA6heq!jDvAS9(&&s;@PiZ$LEPObZ;t?u&H+xC8%oTcJkFGa8W
z7rIh=A8oeum<P+#jy`ncr^%&YK7OGqKXVnDq~-a7(IhQBzf@7b-!xRm|Ag-xN@9v<
zxFIh^eVUxS6s^4Ue183{p|7^yvgqHwWoPPbyma)ZkCB*Sq_>6@dgy=9jhgtMG?SU)
z+jB*9%<|H#sMm+C`8`u*a#O7QQfW^-@R&~dq2F#nVr&0Gd)xh`xWulP&Xiw@e(Zed
zPV=R!e><)~H}Zflo%wqmo6#oC`xm;u9ao?kVYG+h*jryZ>(k^UsHp81y6b#t>n*+Y
z_S}Af^ShJgETNtk?w(&ddw-u6*W~%qtNihk?mFsQcJFV$mD5&)e|}F>nG6;4X>n3i
z%>L5yBhdZ_-Rpar%A~1SPlJ)BV$EOZ+~1}_X1TcQFLdvbABr60TkkBtR6AFCpMiu3
z`=R?W?t$ZZTI5Od)8gop=Ka!R7knEA8G+JoHxAJ!&IDS^cf_6Os_nfqk~+LK^z`kv
zB8p4;Lq}=fi&*Gt`=#^oBZjWjzh^1v=<B7seEVmKK8w0c!iw>|jLl6+kJd}aEHA}d
zf4+3h_0rnbw`q_`S+UwLbgcQ-(D5aC>Du*uHu7t{bf&kPH%ZgKl$e9^(w&}0N31kI
zJsq{u+<u`mKZ7P}rTKdu3*Du>wDhO{BUW0TnF_JenqPW)X?tsE&rUG3_wVyC=y-P^
zp|d?RRhL|Qd+F}wrK`UqI(aMlGgFbbVmvbyc`HWyh3@fK2PCeT>7^w<LoAsq=6I{E
zb$)OELHBx#lCNbwwCA^@vsT>ct(Fj$@zVXQJ2H#K{a6R6mZtJvZLQSbvb()j7P6)J
zdld^k-Jw4;|I(1?mcEk;x~1R7_5c0Xe@lP;kH6Ai|Nh_pufP0%|DXPu4@`gMzy6IF
zOaJG8|Bs*lMd%0r&%gf5e;$+RujK##`RD)sA5rLk^ZVa#=6_S|uQ&RiJn_HVWB)g#
zHIA+DzyA6Y|7ZLUC4V$9{}+GkG}QU$zn_Z#o!b8gNyIs@IR*=t{f{4iw<vPO_LIaD
z4IM8MfBjJcX~@mm+Kb%3|0qMSfOZDF%>Db1GMN=)MHe=1=YpT5ZtCPHb-~Y4A<7$t
z;_pAp+&Hc%bit2OpDjHv^#MOik-Z-IlqeK`|5*k#bPR}F=z^c62&g4=JqpF&f8Ke#
zCxm`F*}{IlJT&39>6pda`dRL>PqF*PewO>dPr>`zeir<|Pr=Wt`{Ncr^heQ8F6^I0
zKlDe@A5zPok{A9c_%lTBQ}RPUMgQ2cy+wb}A4TJ({!#M69|iA={iEcE{wNx+_m7en
zehNNg-T&b?9;3>sRz0ZMH}OC1zhvo!ovHW7f4<DiD9Lo+KZ^wN5+s|*oSy$Ea+#O;
z`Fv64G8%#d|HUCwbbS1~3cH+szn>C6@A^~h=Y4<5{<!;3(I5B!Df#m$K1F{%$EWNc
zZ|j!*`8?12`tc(EXpi6PeOlyyerh;>{C{<U|NLX3&;YRi`Zt77{txDtAM=Rc=a2in
zQ<hg6b^h#nQAPvG{D-gWmHypzZGifRuj`fm-F0oe`-iXVmHypzHF}nR`~`che|u#O
z#{cvyj#~fr${O7JKYeAd^>44N$>N`W%~9*$T^X0wfA*@>__tRqQeOY3uj;k_?Ugkr
z{^QT<wf^muHOJO}`pRDG-(A_RnEt72ero%#FQ%z+{i9d>w@;n_^>uBY{b#THsq??S
zu1(ti>~%kN{@2&F>HD9(?x)Vjj(9AD-wlaxbK@8f|Bn{R|M|K6wf2A5OX&EYAD!OZ
zbzBu|g_cP&Q3<Iw;ak|m2cQ!8>{&1o@vpc5RD1!tDR=_+s4^M|*{BZX5>~rX1xA&V
zT&2~#jC-(&S|c#6`{PAk4&}8kD+;cS%4}64iyfEVoikAqwRg{mhID`C;GewNq1_Ce
zocoi{e0gGTg!`-3#>$Z%NX9oA*7x24@njPm$x|Sq+n-shllDAS)4v5Oh`*Bi(Gw4#
z+uu0vuSgW%o9qdhii%Q_XG;eTPWvlf)UJ(>Xn#5Ls+w}QTN2S~y@=(K!YKCR^H<sM
ziMv_E=G7&w$;<G+tElAOC)vcxRpgc0iu>D(1OAHEvL|dtrN42$&n3<iWrsPDgR3Id
z<*6xvgCr7FA<-pZDdc9Ucm$Yc9&F*lQc>c~Ca;V>f8$2#R7xLbE%E0{3~nD6Dpa4#
z^kGWS=<};@%HE9rNK6%n`jvQz8J8K2BmIg>pr;c>hhI_pz_x2V_*ssgw^^%_YR9cF
zDVkQ|ggl)DaFK40Z!YBDQRe5rQ)P(`Ks43))G(O#2PsipMW!*C3OAbkg(J>Nf`aZ(
zLnIX^`4v|+&kTwpze4jXe;6jwiS~Zled^kt_yaaL#jj1-?&CV7>(j4W8!4`Xw7Eu*
zp;X8QdnYer#jPp%Ab@uqC-@8(W86EkJZhC)W>2At82@Ply&lp@#cf(rUEW#V7p`84
zT79*0?{jC4HM#leg5JQ15I7~XCul*4zmf^1>c9H8h*w<IuTN|G>%k|EJSg-HQi3@|
zq>B-8dA1VJ<YbvUPdP54P4%iI$EMN_BcC<hH<264VN(!3Ul}SY4o{cHUS7(Xz5*)W
zEJK4b1{)k_<Tcoaf22f9aavy`ZoT4Z&W#Lx;nl^_(cUNHDz{1A=qQM9Y*WEVh>rGt
z<qVVYVY^H=P@(r#!3HMci}XGzCMkKkhWA#Yq|ldOgjmINx`9(&SmE=<S*hstGd9iU
z-Hi~bZI_8GYiRD1lTHO!ZBdZOc1<PKrR`zM#?EsslaWSmp9tKl)WdPxgCtzmy4&=8
zfuyg}m1`aW&(~cAH$H9wl{$&}AvB<o+Hd3xspN%i_X*MOZ<NF(Fs{Byc*EROv*&$9
zXP;a#nn~s!7cPz~EBRwR!5=F7%2L8DEzi@1oIWW&^yI7ih46E0)*PE;N)Bt`sIp2h
zU`Cx)i5aSrC)N0QW1iE!yO=Uk`b<=mTfN0yc9irfxPSb3a45^;C^MTg6x@MbDGm)z
zf}XjGlzxmYl*2VAp9k0h6<5#fo#Oq_ml2QPI@uF=08xClS=oj8QPpppO>%LG6&lPa
z4eLU^nw~EZjb6RU6_YygPnaf^W%Y+GjwR`qo{A1d{WD`!>w=!a81Z}s)7LH0gZOnr
z$W!mJx{5b~`jr#ON+zw{qt3q7>Qjz<@pBT`*-T$@JkCU~P@i~hD=t3uEikOEbxSU?
zFI4rL71v%;m&2aeE-#Oee&-;P;k$Yo5p?u3YQv1Go)8Q+Ifkr=4twrsWc2x#D0prm
zMahf^l_sw3sl!mw@3Qw$7v4VD;=;yZyUk|nLTLt7@~L>DRYdfW+tJk1dbw~T5WzrS
zJS(6KW{w}Lj-_W@eJF6gO3UX1(G}77-)dAzUTR|{`N*n#@~ZKSI{tbHa~ifInbwSE
z+sG$qhFbl$#zuNTaK=PTh!eUHRatK=RTyw{=2#VY>$VdCPAdzhy?ja`zKOGSCj8}D
z!GMG(>%k)*7P|RWJEz1%7yqCMZbA{FN_*LdQDU30GHTHtcHAe)L1ZWua?S=^>ccqv
z_ZDyv0Lag0Z7J@PV29IkGE#Hg(p1<PFxMJz5wZm=*@d@(1Uvmzk@M^KmUiQqk)M0o
z8l>@cL#UdUAVj9{Q#olFQ%;jg+OKVCY+wXKP;M?yF^5jR;t05z?%7We#K!|)?L$8D
zfQpQ-S*dvfHrbCAZzJ!kK*hyZvS7^?B|Cn`YP6M8+&XK8CabX&>jpsbeCkCkx5|;o
z=l*mdCcJ=nRI23s{Cwe!s1LfwWwyfM>6g9sxS>5r-oOowWL7!q_zH$sjcyYG)!C3*
z^Hdo~<0}$g#<J%DL<2v$)s{S46pWakN)U&qLPixI=Mw#RmU`jx?z{6Xh$@V}GU>S)
z>$m~vzH`28Y_p$_T{+^m$qv<RIWXf6$4+KGpXqqSv&@b$OgEdJum>spO!d6PUd$E)
z9)rL;dGB#?vfrp7Qyy{SXVsaibwHosrfOp4r<8mb63nBLX9nLFPBJCcEVte5@HAB^
z)&+(oZvgt|eMVBVX;8uM)hz$*Ik(Wj_it`b|72n00;%NNj_dX_^y;T)HM-)beUpBq
zVu^Y?7A2jvWk~j{x+vdgk~37YoOzmAprYLQmLgdgy7v_=he~#Xz2mme+Pw)TmdFmz
zL`COInpwpNNy5FLk~||`A_Ejz8TdSNF`EF*`-;)4(IWf36OW{>o0R~$IXs^ija%df
z5f>tON^h2As8RCR5w_Ax#nZi^cVD+Y8+^iRF!XikYQ<A&O|9>2^_Tka@dDp6wa>VJ
zUkEpF+3YDF(YS9q=0Sj^?zrqyl6kd`5BWePP-aN;>^|t*7t)N-%SImY>%DY>BM(j=
zlJF<fW;A<g@g3q$Mh*t9qhyi}2b&sv413r3dC|jin`xP)30!O%u;%UAOcbQ;8_Pwm
z-aHW#BK8Tb2JSt1LR=uZPSr7Te4qG{5yZ}DuD;}Ok$Zuwud$<~_bR^axXeu5=7+u0
zj=P*1JGl1h_b!*uEz*T>w1%~BFXKKUX{j_Sb#|r%IiMaka<(()UC7#7jX?`L%u0Y`
zOL%{gtjDJr9^6E8cs|x`7D_2mWmVjIRa$HB`-B)RJ6$DFXZ)HiG|d>RtPfnIF_$e<
zeGX))g(`z=Ku@5EUVS3b0{y(T8Mk?!C&`W?HhMWEK2USPSqYaYIoT&-e5j>HL77i5
zb_bkoSq+6*yv0rKX(Ps2o7#>5()ATN1{%Fazrst5^a{)cyF|f!MqUk9Bl&6t+4@{;
znm*%2i(ez{C3rrK5+!S`Hrf%^&;lw*UDIMR10Bu!jtqm$tvmzvj0!`wgrD*A^UXCM
z0zbDzBGcymGJbu2lWZbt<^Gx^W*SGb<Sz`atPaeM*}9K@)=#_GD4*|jwnEc2Uly}T
zf!M6o5;59!h#?wkU{X*GXIYjS25Edd#1Bn%$}{}g_8<!Nb=o$JrFMesX*0=rzbh_r
zED|%C$F3uO+cKI9l-!=;3^>WQiXq-EI}=D&FYipM0r$3D9F`Kr;poUMup}zJ?}{#c
zA$>yCxaBQAd9Z40jkb2&=4xaJ!90<fZE&Y#A7k0;14s6K#9iE8TKGnHZSXeVJb*Dm
z(mrfNIm?PbXHhbJ{~Y1@MA8Jlc`auh5|YX$R~sX)ats7p_L0RVE>o93jaIJ48vA9o
zoT?>YPoztzT|w+|ne$C&jV);=x0-SAQ|IkUvOWWz!PoAn_ZDzAQYC0F&J+G2H?Oes
zDxc6>=Njy|RXY6VmbQY<IiTpb+4dq!0;~<cIFUiaR()ql;gb=8L1Sqp$#As;j%mxj
z8e6r<&+8Rv73`@Xfh;Pm;Ed~NwLntP>EAptK{lkeW&6t@_VnlhB=B=~8o&%8KIo{f
zvryM~;hS6t&NtNf8Dx8y@GP&knas81ap6p6GN0v&e-Z32B{vCoiz<(4J@h%vz2ee@
zO4}#4Jb5ru^NKCk+xDjcPAmbMG4By2bHya52YLEZ*!IqA>SzY>B+{=*%a9#bCOq9x
z^Tz=iIxi>%YSsEKg5gnN+tqJzq_{E>-uk=}Z`GkJIo@eZ^03*ganIa5qQ<k1o!j>0
z{lo;geP(ha8pnVx%UUR|e7dmBB4>F&qe5~y${IMfy>$Qt<*SZrB<*e;9JrCqsnSpP
z`ytdvi>4xXsqFAfD7o8dLh<L8iDwHEQXo02N>t8N**)1b@C;BgWpRBEc&1WY?KO>f
z1~Jj9S@CY#1GQa&o$LNvD9KlwSdP83;&H~_Zs-l$E<i<ILE|z7>Wm*RjC2{yex3+h
z$yR%(3y*l3-S~602n?}8BAaa{RQ0pWCBwZeTbudzA?opk6bBft_PK%<r-L<plF6bR
z-*i5}apJP=2Q=c7T<J)Z;KnFW`5LP<Q%1n1FmWbx=|?+U)s}m`C4Y;B^ff~lDlG*&
zM#+{bd%lzvY&2W!9h=F&210q7napd(l_w7AysYLH5OVRYo4tG*zQ+@8vT#<_QKBiN
zUqCO-W-cHUUpPn6V%D=a0u>X^>R~nSxtM^QK-H`W_u;T8BAh{PZJVipgG_tO(Y63Y
zAzqm*F(P_fQDa@xcC~W5zdc{8pF_n7HA3)-m&Dh;eQ5<8ybo>hdD3#!;S~rj!(_y-
z&#e{PAWXh%zH~a+HT#rh=SXP4Cv{x1!#10yfHT=K314oM+)89^$0YT_GwAA2gG{x<
zy>prUDK|a0CX(=Y<1_CuEe0d@n?sH2b_H_qHQOP{+I5H1Hb$<F(bMiLeqGg(*ksN#
z{plC119#8w3Y^ss-1+z%9*eZ<_yaAP3pe=+fP2NmjLuTDN}9=Mi^iEr(N6lj!yM5$
zc=d8e(iAdynspf@A9uhlhla(B7`MI^?af_AvQ#AC&zEhd;W<Wf%?UbZb~_vO_d2TH
z*LhUGaYAy2o$~pP4a)D`jMdS-KN5{I%|K%<8ue8Iyy?r$Cr!4~=kzs<ufl<wyaWu_
zB!G9=JIR!pEXX9_@GK&2ggdXL#T_2|!c67HPJzQ&NlL&-8qO8KjWYvdbIRr^fY^Iy
zz+<={wMNOaFtQm0p`;mbv81t`9NUdW>RnOYx{x0Oq3ND-^7HZHgZp4V!Oz_H=mCpt
z6iO~njtkW0wgR8L`wjdeiW|1g4)0KN*S4KxMOqX*X1rNH-IEBT?_R1c^y(S3TR<&g
zvCyjTccbp!sx4#Mi0NyPW5y<tGlaBw#yRBDaHd8-4MEvmaok!~`QaIGH8b)0auXI^
z_tx!*gmk-hn?&1mJ9FQ@WepqMY@JxNpz2<RjdX-z_ke5I0m<-la3tLoy-kJ57h7nz
zCwVk9rQeFn)WH`h8?D^?X~jD5j=bEkHEoLC;c9@Gds}wie$31r*BNQq=Hjtiy<1Om
z%8YGuv}aF5$z3C<yd-?@1gNlW?M-^_RCL_W*t=|lv5lw`n@y{U`r|_T48A(-C1;)8
z=OUX-v*Y;~&KV73B<S%?owZdbpGH-@|A>+$s{?QL5d7oi5L-SCOZ(2tVXZhS=?J#f
z_&MS^IwRcu1J5ko?O1m<e{GCs@@YU+d|swFZGPQ)4LBG}8L*T&&KZiwFlD+mc3LZG
zGC4+&e~)JsmEEU-4!Q$QPS>i&%pQ-!x#F_LQR(q>ss8Xi_Ar<@zU;Q>=_w{C{6@Pk
zj#BbbXA`!PG=ZP1_(EZmIHnFeQrkUBmOXqy;yXd8wRchq_$I!yGb-2tNuKT)1Dw;$
zpyXj@D7DvV$4{qmTNK3b)MjK)zMYLuyJE7jtV;J6!9l<6qp}wnk2n>OjirZ=hQn1E
z5V0PU9}57{n660{-*mQRM{$_#`*g^&7j<7&I?YBB;9{t$azLvAb0dcU$6=$}H;xK>
zB87kkr(~uFaLB8nDC=vKtct|e5nHh)+VYg!T58ldW=5A?ao9lRrZU(QJwodbUtt4B
zu|>)1-l@@LFV#*-*7Rd3pgcR6e73&}RM}g#Qxft?RELYyX4?ENd0(giY~C98;c&Yt
zz6b^nw#f8dwhsQRAO|kpjPBt}Z*<p*jilYG=TC>Z^%@RF2(e2-L-zhQixzlBV7-ah
z`9_6ZTP`|H={bJg6ml&eu5qt?CY)j>cJKE(UD@Te%V)C@jI$>#h}??>0d@sjYmwv|
z0Xw^Qj&^R+y`edl?Z4&aQ}foLY0@Nf+Phr1wymhyZjEQ$4L=q6DK;<3B8#ILEX&~@
zO28=>L~@X9n>|oouWr?nvSkHSjI~DP)}mz5!lX+wqB^TN+l*b(iXT@k81QWKy?qqd
z<uE~O9DA<>G+##~v}=i9qo=Q*uboQV`A}%J41;{HGiJ35g>xuc>*fMRW-8mRfbHs5
zU!C_d%e;}Woc#C%rS;tWyuFVU*WPlBkvTQI&Pu))J8z%vL<l?Y*<2FNT3|WM=+|i`
zjFVAR)-!f0TbyHg6LDxed~Gt*t+Q5h&l_o7q(EvdIhL8|=&N_lFlPw&5ra*sX_HA;
zbg{#U_pzKt<2_VTyewJNQyYs8_c9kbLe&FK&S=!FO1~wSBB<p(R$E+X*>M4x9ovix
zZN*Q^eXNXVp>12&OO9<;l9iVBBIE{Z_7tM(Y^B<T-0K*)v^K_1ZJKq7rn+tGG1Cq2
zMB0YRn#0d%WgCh>+h>w|@Q6B5p0oF}d_xMjbSg|*pf%z`o%Ohugzl6>v>N@EA9wex
zg|U{^qNV!!ZEN~8s~WGn8qinAXWm_LlS}U5S{<hn1vhrSZ|n36N3*k%vfMcY4fQkp
z+RlB7FSnD*W!v8F@SQyr)G>bS8{fA3gjRY&5kTI$JzaQ4U^)Ai@Et-wy`U_DQPbQD
z5y+HaE~e#{Y>cfkA>3;onRLb{^Nt<y<1-r9VqP^p#jPXumu-rw_cCV?wNu=ceLae9
zI{Opso9tbKtz<Ld85qqph4E|K`KT@K@ff?w_k5GTGge&Nrc5POfU)4(wg#*zUJnZ{
zLQCaG<wxhhnQRSsj9y!TB9>m0xs0GJQvz<jc+JfvMakN;otLCB{gWOX^_nLD>4;ju
znQaYNiuX$#-itYL;mQzf7h3^h8D)pZY8k~hb4%RNM9)ztmu<#_&slAW!cJaX2Cp-l
zC(v6CxU0Ev)#C2iNThc;<#5#Mehu4Z9L;;hPikOi2;MfA`f&;J=oJZn?Vi@l&Jb*K
z`|wpI+UQcW(2-8cQDx28;#vxgTaxT3X6+{Tvqu{J48i6%J^Z)?c;2|L3;T|%_m0oK
z0y0{Ok|WP*9@aUmq)B!b8`sY(x8)VW=x0>20cPeECFfqnLlg47PgbbDjpdt(sG{d(
zb9Rd<k|h_K{c_1K-q~4l*-GfH<zc6@7IRo!K47xs+S?p^Q;fHZ?p(I5;_|JsXvWjE
zvwx<Gj4VrVE<b-n^*m+U@`{~I*pPn%XKG~x<P3J$wBWu7z=f+64!_Iyk`O{~#df?-
zXSVHVB+nCG>g;$$crkD&`gBXHGwM9;YqQFE6Em*PMzUbbzWzl8y;Yl4&M@ljvFW>6
z6A1e%Bj9Mwf@`YJAB3^DnQeb744X}?i!B4&^z$tA&zqSHP1!Mx-7#Vmb8&^8G1#`;
z-A4fBbN_TdmpF5ntpJY+?4#x8+|iEv(s1W9)XmfM@3BpPi|pD>TLGR`$%<z=cCoDh
z&q6Rool_g_OkFEF+gV+m5!kof_PzQU+swNmCqFp@>XVh>=SRF5$@8+aEk{cIz=fZM
zth{1F9yB-iZ%Eb8z@a6x1y76QWGhC|$;rD$(q)zvlx38js<|7y;7)crwfeMq2`T~C
zs$3cMLK~iOhEDVT{;X5yvX6Dw8FY{Wg+tVC^PkveJ+0uPQtrc^Pj|>g826+5`<08u
z^z817MA0c<9R}gPZoOLvEjqFFQ!de6Z3b_MaTz@;uueFbY!crMf@W>%YhY~gWH`Mw
zJCrYTu3NU$ZY!7b?|jRa-EHa8d+Rf3?cKU-pLLa>Eq8NI?BP`>dcNLl`(drztMzK<
z24`qaw`j-5vnr*`4t7?QY1_+n?;@y8lXwu+`hj!W?@yQSJ?J{avu&RB{ki_y%YEt6
zrxCYy5N(?8UX()_KSMqXt+X$l@_~>SX<0$fs;YJKQ_DSHdPeE)2V&#aa2CEv%j=PI
zzw}J@)%@&oTesf2Fh&a2a&7@nZ{LE){Ogx?o!XWqQ>0va)^r8h-MJQcUd$O93eMI(
zQHbKlLanK!`+FrrE@uiGy9ZlXkOkkA`nk|j$X3z*@25_gD}~K;U-=8t+y5#$ekfim
zZI44lVdd5H2zpH5V*?bdT90569r!#JOy!bi&>~x<5KPzhh1xPMQl5(URmrq+hJn&`
zewBl={og2iOs{dvqQ)$m!*YRQb9GvQ4eNWmeT_306l{VYw|lzqX{aLcsu=&|DSO7G
zLJr!LZ@qJ%#m}9kT>_pyI|<U}SigSg($mh}`hDEpNI5~tT;gkPS}f^aSBn8fL+Mu6
z+BwD^)@)ndWoLJC5M1P&plf7zH;0zS_*P6c_|lr$o8ktLZ@rWx^XfBWq<dRU+hEX@
z;*~RA;OA2~MB4Gj7>C$@FVbHgIp|P47>!VJ{ZM%@7j4v!Tkj;9v%Rs|ZYxZp5>>;f
zpQl&=T|lD0<Pg@gh2;S~Q(H3nQE9@Ytmo22fi}^)UGB1K$@7Jk4`qbo#&SGEXUFrH
z2)yyJVjAg$&~?6%q9uI5sUEH)zId?yFB$NqNsjR{(kcSayNd}A3x!?o!ixDvt(|Os
z&}09l4<!%YUqvgLd9O9JNm=l5=(pN4K-1p%QsjQ0fEZ{|T*Gg*i>zONW%9VxeXB*9
zyhB&6^VYk`H}zN1rf|PyH>sL*wb5drB7S<ECe_vAB@htFnd*UT?$t|`U#GC@DADa$
z)}0Yfu*JI>(i_^M#l0k;m6RwnO)4MQnAc1BSGY8bk%a?jX=<IajIv5&fesxce}Iw;
z<K6&!a*HN7b8Fr~Cpib8%MYgoQR~-T`DrO|H5|@qmOR_NSM~txh3)jIb)&6Ltu2R<
zGjUI=YQpn<rF<!h-V}{MCScacU$R0KYtVv@wC<qQd@Cr3FzdPcwVjB%WVgFWaX4B=
zaf%F-*n2tOt2GX}i#F|JYMO-EoryLpM^DV6ti@RLWzN`py=4cBmrUxQ9W65nwd*Qc
zk{|v+&)0T_$h;JVf1m7pET{WqmLzH1i!C%Ap<T+tPB(Ih+%5;9_sKR&g3#EWLw_bU
z&aHXeJ=WC^JO3Z{9?FdVxd^$lUMtmQho#W``W$P$aBXWZR7G1**5rko9cg|lASSRP
zlR>J(51mBFWZUEW(ltZUu<ZCod3KfT5hwv0_!?xa77x?`&TLIc^6y&`S%amb`&t+a
zN9Hq4W=SAK!!Ba(&fAmxV!M|=cE#zzXg%ipF>7g=OwfxoqnA!hNiH_-x3CGoJBBK6
zapV|PwG+7|xHWT~w_YMOhq@OBbwO)pS;&x<YU-Cx0u1S(`wo?Cx_aB3ue9H*kt?C;
zHKdv@?VaOn8h6YX(@rPm?(Hd7Si6T(BFib}^T20)cwR~-57w?9ink#zH}8Dm>h^BQ
z0C1azfM>mjv$Xg$!m$NTSK&P@pYG5zSM;LPHQfzUbfYnmIRT0989nok?0&X=@*ecO
z@-s^y?3z4|OmuP}BDvS);{<}{nv;-w&AwqQx=Ahp^^G}T8@R0@s@k@XpeULbv?j8q
zF1J#ki`5}2rNe#Lpra94hQ4j<J`15;p3s{>K6uA)3mwcp()kU08%6Uopl4_*%R6b7
z9+Ik^W3O9u)8Q$ZNBlnjOJ~87^L6tLhMGvWNc|iz7i@pDozM4piyl+-OkQY4=lrSQ
z7(-`nf)4k;fZ}nrkW@QEynp)8fo)`><onVlhjCg6sf0@x3KaaI{<yO<tut0~{uS#R
zl*<#|XWR@PF_&uVf<H{(4~5Qx$5FNf%=c=9pETXW@B?$`YPfBv<<h5yozUFZTk_yK
z(SZ~N^=bkZWxnUggGVp%=|I`YLzdi+xU(8+=@HnNx{#PwJ?(I}5y62RMOI9b5BL0~
zY-@2$ja#-ZqFRv@pCx(+>%ndH&XuCuo|Rw5rj{o8{M^Fn*iC(5?i5d3h~As(V5oH8
zvSVk)OBU<(^JJ&8s!u1Ad_3sNggWt+ja}}Y5!kDG!>0>~$GN&V`vUvYUAlcw)ulsn
z*4FpZm$PeKmy*H7zNggohfZQU5Eb9mH$CX{bXerAd-wSPxYf%E5OCKPXzk6uo{0-j
z(8gvbjW9>QMVZ2uDu@GD=rJy`=AyNI*IaaWo1u=;Uv62FF{_4J`@Pv?PuL7~tgC$Q
z_E-)!UUv8C1fH*V_*CR>1EQPF5~y+kJ3XD|yb}x++>DJfm~ak-QPbTINfgI8O<ODh
zgK}_hb4D1Pz>RYU=6Q=xTI-e5SMrv}e+ybO?j2bDpAn2zA>jj<doGFqY%B|8r;L~!
zSQDDXOo+)0*axdI4pZy6*B?AL>z*C9IdCDi&k85+&+lhLxPr1dxmig6OX#i?zL0jW
ze+KBG&52_?x%oz?EU4`lI$18aS<VDlM32E7ovV955qFz&43^Ale8+`OQa*50Gvr(D
zZj@>IVn?Ci&ba2Sw#d!9@LBHon9!%<Gv*muZkM8@RQ0_5j&vN8U?Q=XQuDs`lFq1T
zQJH(bHRv3JMd@c~XZxbt`tw{fCSyWFY>#IsCOIG{@@nuB127>%qiX$o{+({or3gYc
ziE^y$Q7~^@SnYL92P9<>f+%)4lrJ4f?Q>tP#@iW~&r3T$dHErhC?WAGJ9{+3AoS@N
zU^GJV;|nPMpk41bWnUP}IKTmfdl;kMdK><ZYO&@{528fhLidQB2sF$ax)$`Yo<8-W
zYGtn{xetjbL6v%nJLe!5#l|0RJ^M+XD62`edAkp`m~Cm>Sz|oloLqgJZPZwzK-mXT
z5{&4IQzygv9Bt+astByYzGnTDJytGDRh({0?m^kD(Xteqdt6Fvh-#QAMwCfF;8YE#
zBeYw#605+7()z5oUIr5FBa{`Jh>V-K&)Yu2H664#*&69~PPc+liHu}nTUNdgF*aTy
zMXpo#_fuX*YK3bQVLuOBtajx956bzW2Gu%l1xoG3NGIDpAa0ks0VN_3Mf_)KMYnUL
zoW<17NCXsS)Y!R)PQpj%1=E#@paq9CUT#nB^CY_QxGV}=jmb?4LeoB{HTS)RmNCGK
zq6DIMu_$DFNmpB8Ub85~fsjtSVuxA_Vm!Ud(E9S0#YEGnw0lnX&St?)gio*BlIIEx
zmdeI3FI^3nex5<0_)d4i!BoSo`<RQ4tdV(rCWp-)a?-a{s_Vgta*oTkV4X3fY`ix|
zwr$K>FVzWa;Xt^A`L+KgyOiy}x#APwcpx|Jx#pGc^@-Hc_{U=P7Htj1n}XYX``o~>
z&#D^sOq{-=gAN=<v<q6xt#=eeFp9Ff$M;O^0l6@=b?)IM(5aII>fL?bd*GB<FYGf1
zY<4-bP9IjAF{Evhs4!G=FJs+mQCcz)h7|`ZY=P*wl4Fq_{g!W&g}AwedX$Q&D-x9H
zzM{aw>K=E(6pJk2vS9XIqYZ4cqs#TAZg-7<+Cg_#_`wH`Z6Rpe-C2OW$QM{?kJ5)?
zW|=yPN0mcIb6t#9Tzm;SbFu@cW7oPp1f9v_#Z&s+EwX{boXuknfif_gdq?b&Gd%R%
zdWb?eC1^jJ8qm51By`G?geWqPTglqc<1iM_-c&<LKg({aRyNMJ*puv`k#V=Ww5RAI
z8RrY~P7ZZ#2ppbDKodX2IXF&-fV-JK11G=%JDVO-Sx6svJ`FjGrZheeJ6OEa6Oq}t
zMbFjGebYZnDfd#Rt9H@wA_AVli6azk(TrQcF(McxI3Qp1mfdr>Q*5j8tc5Vc*-YHW
zy>CVLv<S6*Pr(n(ny4sCLYtttO_-F%bX79ZIX#d(!cw894NZ@yHFHxC+|tcp()>~M
ztjw3byffPS7|tx7w@C!0doElQOj)<R#@lIF6G}ZJD`exAwfMf&Zwn~jqkM(KMB>Px
zNI!$6aUS=c))~K@_H-*aeZUkR&cx(r8ik0_+~f(2v&p`6Afi#&vu)9Ys665mQ8c{A
ztHC=9H<KdKt#oOT8e+k{=YHl`a>a(q9#PJsquSwTiW6w&lTy2io;BXMMccyFgj2>I
z$M6`b4UJrmKf6auK+!?fY2X>X+AWI?rs%o9a&LDQHkUY6P2HD?&}D~=>|ozpSR9e)
zRCezB`}3yg;fSX7`?N^fz;;dn<$z*S)9IZt{Bk0Mdv%d_f4m$*2B$SRJK#GgP82Fj
z^y@8K=-2w1PFvioogo@ifKcld;|T{?>>U<!O}RB-Z*=HqD@Jw<`m{Qsiw~L=A!95@
zeA{CpvX&r(M{?@c=I4Rg%u_2KYj)ilW@ihf$RjW27S<pfyQ*yZG|o~hVB+4%jg1Q2
zPq>)7w+iyv>8Lg)F(S0{|6kj=>}rxE$9dmRah5ZSLuK4OwggB5+>O*biVJ}p3Ia5g
z1|+E0SA6E-o~Np-38)DHGu>TQ@n=TfBK%@*?)l8PSCo8{c+nB_3qt{tz@>8Vx0t2K
zSgD>$Tr<31&`si+w<sJe66yn5Mpe_~cmzZ)*eAIW&t+u_UCo`QoeRVecR&fk6%|WD
zH`J4=vdEk1#fTZ*4X<M+TtAl&fS|uS)iZ~##WGMOfOKPb2eo6yb<aD7tqi)6+h7%4
z&wCALQQw{ltSwdvecGOJB6C?1UG4*zMerZPrcF;c?a@je;7t7Sq??K+FRGeY*-B5W
zbi{PlO*loRh0uj1sk}~bfOw@E;v4Ly8QK6<-7p(psUuN#CZWB?`++aG-G1$v#{yjl
zaIlh!pLx;a*)0VTjxq9%J=gvyUv!N1_qfRWDsi`hiwH`}QxBq+EAArHL_UG|Kf)p+
z*)Q@5H3B6}I#eP>K4GJcjG}DkV)7}Ey&NkQI+3+_Fej%=-U2O->P-ItTj!~2=nYV@
zzCG0g16HocmC%X%P`#orQO=-u{5-mM0+OVwp^<yRW7HNoNaw9yi5?%U>i}o1hy|Dr
zv9IaL_23Nf(N;1aRV35AgtK(D>3&xpba$kgtH{mQ(P#5|F8L&riB&}+_3~7DY<NaZ
zqjMqPl3AjLEJv00PtU``ZEWsJJw1=B)VV&k5@w=N*SOq0D*L)e2kjLxP?oV>-idVt
zkF?%hmj&Wc_qAj9$nar>JtQsk^VRcIb0W|ou8!427gFq6e4qW!m6BbL$d&N3MQD25
zYl1eEW6ws{mfxAD=bhP0>=QP93oDHjF-jdG&{%)bUEFIr>%@&r(9dBMw*}6I9Lo<P
zl9zqLD&7)Zmwj@4hgcGjkrexcz{SXu(No%Yruoc7`)H3F6R!<SJk<$S#1UT*nJ&$%
zM>0<~6Xe#aChR@uHMbih?I|^<CH+aRAiP{`<kFq)6Dmdm7t}ajvymGv$@k%9TxG}W
zNy1a|Mvp3p{n2314~cga#Z4?wF9Q*sE_PASCHW3wGgn=#s8o!KsjAw?cWhP3t1hLy
zXKM&Sfu0ZdE>o%+M8gAg#5W2mkCXpsR|FMwxWLb&9WCdJb>xL`nM=bU9~2+xxNYam
zmR3_f7WMKC^ce3#pHQd*6}^LR2*rc|Hws{~b}#!{Aoa|EJHB!(YB_7XsQs!53YC-;
zWBcV2gO)D(35q|l<474!^b=wn>)HdLlL}5b`};1dx6;uc&y5L!sYFg?oe?*e`vRj#
zuElP7)bU(h4{PC7gL;K7sl01s$nVB9karNT9jIMtkMnfzXHL06hq+!w$sTz!Q|Y2O
zl#$T@orE4(%g?;?!bDSd8#>R0Xskf!vm-jUJ#b|)g3e;U11uCr>2t2xvyZAeXG@l_
zpmFx7((!|aD+g-RZFnQ$9%!Rv#N8Xqp<MwH9{7CLi%vcwUZW?>H`gt5^L1D&c?Cm~
zESxOj;MsbPZKFiYNHS&Oo=EJg<drSu|9q2di5h%j`5g0V90-MttK+UqbqC!rS^Wi{
z@eo`MuFCb>S0_X0dYq?s9!GqnC{e0Xwd@&P5>@Ki{FM!xkZK!(#QHw&+DB!LK4I;K
z<C^uZsH(FgJDNj#qfnNw!y*B$YWwQt#G$)ZRH4|s-1XV-?2&WO<JxuSMB(qxtC(py
z)P*Q@(1_GZ2SV~6&owPC>i2?0Wjrlsu$7Wl)U3UCfTG%Vg&|`#p=@(_*>)R|An=;5
z<w1!VX;&8_^vZ40KO<1YK8lA{o>-M>0)Uass$6eJg`#St^9?g}+KX)u9G1QEm|gS*
zt_uOOKnDSDf3jDORwz+hK)dqu#dRIp4I>2+0(6_#Z$G`1oWN4|$fjoUHBnRkP%1wC
zy;!d%l#Wj%G6~|TAE}~1SFNC^cIxpW?la^XPAcnoTr!j!Gy^U5xzVLo^8f{B)Jxb`
zEVQ_GYINeI<vUOm-#rS{+Q^d~<6!%4)k@2;T^6_sw{K05P^%t5Jx1}&IgG9dEbn?)
zu+)(}E^zfO-!unkE;EH!m$O~<YJ{AN>3OcD?M{)yNF-9m(Ypsw>&gWbcJ$Q=*)u}v
zptE}w-V?D_N<Q?JbFzM`?uFX$%45=tIo3Byt@%&fT*Tg|$4^mEe^Dj+q#S!|fk0cm
zEl}ntv1ijKV0!)Wd7o<@;dl|Z9&g#sIe3Q<_}Ou5XccfZL+ZZrQJ}L*G=y?(+c^X3
za{M#32yv-aSh69G21r0|heB>n%jb#%zTbOrmMB+NhtBz;wy${oinU54b-&@Gg$GY<
zRgrQC0p48f(XxCzz>f_64z#x7B@%VS*Fa2%0FrFGh=Wt>Cc#x-s`9Ah*t?5z1r<c1
zMAh{z&Wj!^4XRJob%*JroMXqXc6E!NSn%2D+Lc(3Y;F@5(8JhgA$sz#OXN5iFzcOy
zdBh!ghn|?02Y(aa8*`j_522owGCl@2`|jUei2xDc8s2?|yVU|cH>+P!qg+>ER1W`g
zSJnWj4}-KsnvA7HTalQEO1|TAX(KRQx^%baI`!1uq1)&h`!!E?HbpNcYp<nABLO&$
z>KQP?NTHL!a`{MawI1j)FK?2#$OTukD{>fqE}&Qu%Ba+qLyr~XJ-0epr9i#k25ZZV
z3LNRjY;OyNF>)^#e{g%vH$6oH^cppsrYJmXvbnnVi|AKf5VTfrH&LU2QlR*lDo2gf
zTtn_s>LR$Q^F2UU*ABk58H}x_Sm<`mo#cp6QCX|kvX@o!u)`!U0@zj#J*JfnZ-TY*
zs1!wcL9O=RKqncGCbCn@0~|MDsr<EtyqZw!-YSX|NM6J4-P>?k=!F-GLLau2CO+<6
zXw(~~-}=2Bdlr30V27!~KbWlEdCY6t$9hL9_@YwX`z@&R$$hdSXazW@_xIpi&!Ol2
zpLPXLbEd4FKJ=aa`MT=uh4T(PX7{*{@YcTzoy+g_c%tm0(`CS1?3MNuJ5e)Ndp*wG
zJ6G%Go5K2X*!}FsRYB`J0}O0e_FO^2??uh3Tmlr~pG0kkqHgSr9oG00yV8|^0j=fn
z-duY`#2$1#zAr6_X<RgVM@1J5VOREgkDcd7!ONwrETR`(anPr@EmW_`0lKP}ibEaB
zGa<Hei^`*F4)i=w7Zw}iNpz`$oMhSGsh(xUTT;LlN^mu|0%m4?J1OAq{%KwB4ZybR
zH<x?^gq`j5Hl;Y0(A}y!f9U&eRV9ZOv03Flm+PE$+7Ap@*4d#;CG^~T9pK!TYq)jw
z6)A2j*EaIFokoFv+;9Rwm$!3(oL2%Ip5jKAKIb}Ao9lewJpUZNSC8~edUZtz#Y<Mp
zL!QmW#g|Hmc!L^^{j`IZTW2kFo4cng+5R0baNnd}yF#63@qxeBBGq@Dc~EtL!RO3+
z#*P=b-9Jrz7tl&0a}Ts^%Olh&x(SilD}w|I$HJXq3#+Y5WZel6Wk(#X!-|1lr5I@B
zd^_TqB>t*s`SFFJp2-`IH#yfkcIw4exM=nqt6;D$c)u6CI>r)y)UlItx~P-;qlkt@
zLJ=0Js@JZ0097vIxYqa1Y(t==sun-qL|PP3?7=-K_M#}0n5W4in*ut!6MPn01%X!m
zfmla-4^Tx6s=oFIc+8)OoX(kOr%Qz5FCo4lbj};2q^nGz)Lwh~E|eUF+VZ-+^ZH!Z
zq*GYT%_n|Qc5D@NE1+gV0nwbn`Di?HN<DARz_1XN@rvLtpgFs6w)LK$mUfDEw^8H4
z;0Ze2udXlHgjY@qsz&s=;|F^Fu2^t|0S&<t1>0ASBwf6`{xrh59aTTn2$TZ|iWI2Y
z5~|>~UcO4l{3>2>OAf0iy2-4Q^b6r_P<p6#XCX%AV;|}X;#q(h34^YXm{;7M_vp#7
z=Lt+5VeY8Z1l_7a#mjp^8<gNzd1Kca?Ucwj&xO-nH7|IpJYC7vpsRF}x80wj?O0Rk
zU_g^9-k8~5)K%z|vzxJ2L&ee+ta8=<ge}oc#$RO8yGtZ+=z)(BZ!A3w#HLcSy)ai$
z9h&7*@Z#B0C(H^^)%dXIq|WN=#M9YU&guNO)ACvm;~$jzW93_ODWM1l7NQdXF14o9
z$N(9Ws;b%dUfB}=O@b4aW04^62R=gw0aUFPRZJcAqa80$#8VyW8rdB$Q27Q}zd>&V
zrBwkBvF2((nXacGg5%Yx=Y2efvS!>YxblxjO^<I~675nr(;Az1n<WI=Fe`zDxI;&L
z6Sqp-iLw%%qW2PwEm0%@Dgh3yJCP*o>XGQH<oWDk!&em9_oY-H0zDtLXP7D-S_0iq
zt6u5fgACQxfN=O;)AF4|%}Z<(G)$hj;i$3D`N9){Em4#pw9383lbt39h@0?mEcbVl
zt10x!EN-XI!DzR7J*eEE=1WRuAUll?P_d{g%Z}AgQD>3H*&IrOL^t)?j$%21WAH5%
z)@}=j5!BF%5Z5SIT1a2Z6Sf&b>LR4f>7Pi%=RoJWJbi_hFN%jhoH`%T6Hqhf>6a2|
zs*ZvTvLlcZM8>;lc@$Ju);r!u{J`x*eY(q(K+h7pM9zkQZi#LL;Gn+rT>lW~L|?2*
z|N0$!eEzAt_kN;A8Wfp~s_tI#l1g5+Iz7BZo(pvQ`ZX+g_8o;zUw*J1a{}ire$12M
z`<=jAddZ~^x)QZU6%XiZszP+uo+VX@*f1bTz{H0kS0b)FP`nVOiU4fbEud!&UtVM#
z0hB=}1k}P*<{ix9;d#dH$Bk`s5IpKmfhhZS2n1~f7fO#Zd{NLfe8HS+y~pis_`5r%
zwVXrWRMChRbN!=19rJXazE|s9lm<|~lhv^KL6;Ykpmmd=;zeT9L^(3xcG}F@xDzwP
zV70V|$1#HD2Sx6->lL`nD^bXd`NJKwMiAiy0mLAP4=*704{}EGe&7pke?5+Qd{^dC
zv>*;A?!*cPy0&jmJD4zof*r(8pefFX`VDESI>(2&Qy!pCj9IR%&3Ejmk5{7L(C$P{
zZ#Wtv>pdI2(Y2TL9hviiif3_J0>{m<$kVv}_Qc0K(DAJL9^tCz$xOEqX}q8V4Rk!)
z<FX4iGR#9%R<8mH;bDPOf+s^JcB02!GiN*Os|hUBBzQ9Kox5o5gs()Y3AV$4NGiq&
z`*K*DpmVqa$60;ur#DcW8=lxJ7JCt>4JYci2w4r<CWvz)g4W2Pa1op`@;vXQO&Tc5
z*ju*^66mTDJP{G3=B3fe*jaS_%M_3#al0-WKY(uKo4K?R1nHq*zZ>%Ocf4bH<xN(>
z=%Eq;s0G{9xFknnxD|H^qq=qA5^nbn`h2mu-6Ux3X^FNf$rHUZu_mF#r+1txvBWz4
zeS}3!cOdckcuf=4n>|t5%{XtUC5W%&me$VuiXt9`J-l=KYD<GU#er?;2_JE=EsAi$
z?Fnxk6oGJ|Yrc2z_}IQVk;2QddpBPxkMkQ#f{H<#m%ZFP<e;`lVkazOk;nkm1`V|!
zvZI52raNOVO*lZUC)5DUk|P^R*kap<Q8s9-Kt!jCnJ{xsXPv*ND*48u9}D9|?7-V}
zQ|Ol3iI~?pfvNmQRV?Jmp<Y@kdT->DKv{T&t}jQhw!Efl25o+5adH(Gn$70j+Niqf
zqkC?fl{0I$u18NCnnF)ht$|cPb{qA8mS$PA7tdC&io2;6;naagNCS>4?;%!j>AVjo
zI$WD@sGps=xJ@X24TL|PY*O>vB#QaSG^hPq9&t=xbWnUu@Ytu@4eP8~1hwK$YKbPj
zpsR~&g70E0ugRX`(nb6I;)&ZM`^48dgDp7NV<DRmet5!q1U1MeVzuSPvIJoRv4_ta
zff;Ck!jF~x`-~Tz%~he!3;UwS;J>Rz$(M^Y?CNEqnZE)(HnKBFDqe7){kde5v$Dmm
zfE(&V=Z+lJz%LqZ#P{854n?)hX<n<fJnVF3tQA?5!TlhRs~AocXM@g5e~ZSRfC&9=
ze_f)C22{(XR@0zt4iq;aFcC0{_S$98f<bS$Bt_yWK%vJq6u++SL`yYqb=ix>O=O26
zR@S7v5Z^f7k~zDlWYT8f5pM7ERQG#;rj{g-QF_VWAFs>BWzQBJ>ohm?alOKGD`(DY
zjK@=`=*;F=`rLaXza9w#&~^KCw3!n)H>6!ObT2@+Cs+8=^L8P@rsUrfrJw_5K4L%K
z`>N!%Zn27(75x$&C5~CwZ5tRzMIW?XE!=2I1BxDuLcOi;nFpvvntZCbsd<Ddc_x)*
z1QUE9(?WhwjX-lnL_w@OF^_IfdA3fh$W`CpDU33_J>%1Bf6>MJExuR;kq5mQ#tlMe
zxg6-8Q9V|2UTZiX%t{FD4carzE5RR2I^7w(a_iWrIgW0(xA#VWN4j=jQ%muCrG#d0
z_nA?0`lwc$H%H0TBOr7JPp*Vj0!Pt2yHrPw&l@lB_Mh33GrMS0O}G-cp?Nv*0=J7p
zu4srVRCz?D=|^nUsD`Qs%{NvfvY}9ZNzc}(vH+dpnwYo*dE9ej9x0<*jRJL0QyUK{
zAmss$-8!Y0`em}HODRXVvZw@9{V>V9#d<ms4s|8WYN2~FB^zrloqkvOiP%7Mjoi9l
z(8Uwg)~V(pD+(lj`H_g4>IZ$5I7?Iufm-)dp-I>R?On-4R=72D>`+HKr8RG(JfXCH
z{ZRfapjoM9Q}V3b*9{b5QbciwMOo#fWdn99Nx6PoKY}9vD754a1f5jybXjxvxlXZS
z&)Q6qM}j1QG6uAUVKa+#E_Hw&xs@5GApO<a{fa>;kRYd)B}|h-F*;t7{tBKWAQcPb
z*@C4wwb1CALI)bG(@0F{6VaII?OHeUJUesA-;YZCK?%&6CnBeB41I&-lL^8vA)ge9
z+lgK%>@y}B9~K~4lMm%u<57!rdR`$?tu-_<KsT}`U2U@8JGDja;izWUG@ZmAc!fb%
z#34b|ik&oF38!{K{$dAFimBuQU?rRfAbEwmZ%v3-Sj<W-g#6^>Xtd}i4KPBc1N3~t
zUCP%xpic3gqR;!CYNuj^5bL}R(*c#n=px_Top7>$@QxBAt$DL(<$!HYWUWWd%A}ry
zf(*hu7NmYqu`r2sjj+up{Tzsr5O}fro{b8?DfC*vV2kQI;%Yruhf}Nax!Ht%BGUx8
z;Q<i&Y2gV@S~{R;LO4We;fu1T9F%)8W66MH2DJpA?#At>I$kYd4;|?f-6)!XK0(Wv
zxP9l{!C#*dl$^MI?6wL|<m-alcT8h~17V{~?*{y#D1^N;J^OmsNSMYsRMxysKW3i(
zxQge=4@pzB7d2Fy8C&X3ls0hd!=oFQQ!uG$p=yip2EF$Zop8R0@jMZ8nrMY0R!=CU
z-kzUA4Pwbq5Bh6@PTUs)kpvg)9okfR#uQ3$EJ2mt;ADNzJiuA09z8|d@4UJZVI5++
zKY4(&=uVyM{OvXFC%hmOO^C2W7et}N=Lp}GP|=#W=O~&G>Fe#i++{=Wv7u@opcWg#
z>@z<vFW>f$CNe636O%)EjR@Avp>@spKKlz9Lbc>oY{aI6+7iqX5)e||xoCahfRfml
zGgdvQBQQ(4`a0A+Zc~W}xt$u@HmJcYA?0Q9A>sAGEa~!&R!(7{Dh4`Jp%-)#I_m0W
z@uH{!3teNp(?v>MYuWkuW}a;gysa*kpPs0QKFVK50W;u2bBct0`*Ih~y2*A_zv%0N
zq6Q<hw|MQ*2`%Tl>!N92y2G3jq|xqkUfx){#iAM-i?nhNv^GpPjvFEpQ^rW~-R}u=
zZ_Z14W#|lL?1}sZ?LoTK)GQBsoGErCRES&6DVzgf0918kc+rhgM3*>Km{9nq-S7EX
zsxQ4_(g=cjVv?dgG)O^dY7W|D;a$|E56~43TzNL5$+ljR5b&JD<sj#iteugKm^j0@
zL!?Dp$UuN{xkIFb9oV@C=hG$Sy~2i42G4JJ3x4*b?*W1)upVs2>w$tU`YezsdWhYC
zLbPJ9ToFT=#6gXH30nJ%dx+;p!wMJ{VFgbdKlmkGvPbU>rdn9?+P2_JfG*@rL^=pL
zlDWj98TxeAS~&%yF`@6ct>)QmQcX+>vqVB=A{9!b1|@O(d$z4K7kO|Cn(8!_xe4vT
z1)mb|mSrGq(7-0F6yA*y9_>Ua6y_LTX}evkgd!o$nR;5SN=xdf=ks~;Oy^bQ{pR;!
z^@C3C2M=sT-eua&?Q&~vy+@!woKRfq9w_qO`a;eUIPwVVA&%G;9JU;bEZY=X({e}r
zf#MD4OwBx-(TRhvw5=d4Oj94mYkG%QI4KcB(F8NSMy3gr{tD`7QA1Z|(knuWCMskJ
zV_>U4jL;IzW<gp0y<cH;W&stKR;<c!_yHx!4pFW}o@LT9!mnYD+#cu}!I~u3py%%g
zOiJ938IkFU`v=9ut>=}$pJ+JibIPR9#Qj6kL>6iAj6mtzu2rH%4ctU4ly)oZ4))%H
zjcl!6kQ(ZFpi+J9Y9w~8>V;S+5*oZ>Jwk|}yWIWhY4>uMtUlkM&edFVU_mRx|L&3<
zs(FNpMx<W69S?nBVbaZ5UvJR)KoKX@*l{g~g*shxZ|T@?_c^XdP2~0Mj2+y0P=%ae
z7uEGnR|MKVvb&Ekw9K?`O3CG-7_eQsaPOoH4J|F8(UL|e`ctwTvO&TkhBMqCfE$}*
z&q7uFlTJTs74gh`VNPMeMF<@o62{gLwg*MAM$f!Qba)}qiKGeFkP$%8O{u*$fhd<)
zkRIU^b?<MMV>>b(x(a(IcHXL?brDuY4|LEj-0hd!2uG;26|k22?VtsUWPimj%iU$+
z3~t>Ti_j<-4b0u^+3!e^m^o9sE%5-i8bLxC>am1^PDD+(SP35rRT~Jmz-VLJ5;b%V
zI2ldV11;2_GFLii_CV`H<9QPFLE{Ca@R6a%WU@Tjd1bHH^gQp$8Z~z^Vx#*xC`4@c
zn|fb4g+amBhtjG&E*PT)^8{a@hDh$Bc?6;+q{vo8JV7bb5jw*6Tf|=xH6is^`g;}U
z`MaA_BmDt55vv|5m&}{6uY=K3mtbj6kULcBZIyVoK$h+zv}Pxj)zqd_C~j;?lwbte
zw3Jm|G>>3c;sSak+(4lMbAX+UmW6W&jdn;2HEuX|8!QtNdiLy`SQmkl<3?TYqIm?&
z^`Z8PmQxra4-Qpfw44HT_=B<}S<NduX69K^NG92Qr?B9CeULWl;gl|vrzlq-@zNt)
zv(&G#2|dP<<=M8_IwXF|?L&OF19BW^Cm|6~%FRN7LUz*FXiE!Ps3s-DGJ{Y82#vrq
zA+aK0iXdzPy27SdSYB{bmCgyFb5IDl$`XxtXeS^0LB^oxP3X#z2t{Kc(LpxwUc%*0
zuP&)Vp(R&Kh;TSELp;QpSG!pG?i5O1*dhtSb663`S#*y;*kpWFbXx>>u}Oyt_}7an
zbUoDJ@tWL=ydLGZWEW~U4xox7poq|8p%C|SXlTm)t#O+_4av{NBCfta=POWt!D0z!
z7bj0kDk8sf9Dt>JNX=!QA9IvBwcT^zz~9&DlkE>GXPNt2LojDSiq#7)P0DzcaSBm2
zA)i#|dj$(0M@LYg<)X(pT6#2Xm3Oj8y66+J5K>%ncbdK?%Jl9~G$ar;8FD7BYrG`1
zu>ETqxhgXE^D1&O6kAuK23>@VXAfoOg;XeI*h9T2TY!yiz!$|ZwLn=xJrrV=2T)ZV
zE5e9HjXe=6pkh=}Ubm2|K-7daDOU;fp_;FX=basS1ys0bsiWpM56kOgiqO>l-WhZX
zO;M_>hF-jtP)yhvMrhkF`bSWcCL+&mXeV@1tp{~$gj~U*F8~JkP{ZaMK?XItHguWU
zBgdZ41z#0Ye_V%qv0GAP3Yv8Xpc6JCti^0NwCEtge(Uy6$+I1oqsvG*u58>LNF>Xh
zRhsuEdUn>QxVLMGO><|pgc#Z4({$6eW?&c4+*c(CmXkt;<Hj8FYs4qqvYpPOvmAP?
zpd~wewvi)|!jr4I(ZWh}(_lwtX$0^VN}+J3alTpzB`u^v*^)eN65At-1KKUhBLq`u
zAr*?=1wji7iY_Q?#)s-XqSSnu2<^%_tniT*N}<ST;7eG@{(vr2O_Xz(2@ZmUe!u$6
zYZz!a%HQrG@4x6DfvO4DJEKq7qJhM{2{m$WfD4@hW{`A_WlqJx)hsnf8cP`up`=Yo
z`-tJLDq5NM=Uc_ya?59wFaa(&CeWQD@w8|jG2GI06_P;ZCx|SZw}q%U@4x7%!OdSZ
zkHEIU`dcBP0=hVyc$8!3SYnhy$xG$gXwK9@sZCGxxD1cbT|>5K(L4fnV^=ee_tNIV
zYZSpcx*@DBhL1j_=dm2R=zyoMt~{{~;t`^0;!Nr=skSJkaxY4Hd6JCtZ*G@60efx-
zDL1X}1PqSQ<6kYI&z0zW{IH{j>t_gs5?AgLhI>IP?x-G1?@6?Ec*rX22&>T0G@$3m
z4R9n#>qs5oHW&pQ(m%u_ZINXF71dU5(|t6CP-ycY(;g)+YstM;*k-*Kux4{Fm9S*a
z%2Z3rWwrJmfx<zB|IT7n6Wvl%<+^sgISni7x8QE-j^!6r>nypA$SE1*siDB8t!Tyo
zXiA{jj@Yb8)y(sHWxt11hc7}a*CGZILx{l2h~`0t&Kc~SraGlO4HZ9&p=Y1D@0B4{
zK`$LOhJF$}ZWxRQC8ul7X)1F}sc5h7C7m1W%|<)*A&HvZZleh11!^6?kR!Btw|m1E
z<QlnF3r#A}Nl96kI7^%jjZo5@l+R(ACc2@Ru*yE6<ZI+6=ruyPHwniEQ;pV3U38Rd
zPl9TvlQeA!?7R=2MV%p@3V3&)(!-1~1VV2lzC~j|6RSaQD>p08KF-3u<jb<91Ueh6
zj;<cQ<wZ9PlkP0jMAtyl1T>UpjE)fyOyp$OnipVdIH+$v6l@HEYw~$rEz&G6VQ=WT
zUPE*37^7tb!ito;Op8Z@qcv{$Gz$ko7n**I&CrmsM^YX`pdghILHG1U5*3sgD5cs;
z06lFcNZeEhIL_nqqL6k-)%1`DHf>OzbuA@2l!3|!AObwD^=}AD@P>11tmW_3=}x6s
zp($(9;+|UJ3i|Vb9`k%_h1<`_Ng3_*JPaSI9i}2(zu#j7&BfF_!So`uv?f6fK@$=I
z>SuV-(jO0bu|MV!C|6M!d}OW&bRrz1a8Iy*Otk8z?T3<`k*EY^wY4@&w0BzRkcOii
zD^Ti~eOsILo$vXB2ba#bbiF9MtW9c7R(Knxh4k2kUE5}?b6p}JN;0cQ-~|np2$kE!
zha)=h)<LU`go~c#r0Sr3th_PXh^kTg(W8df@Zu?ICe1!(ymG__MKmD@6g35=imX07
z4i*~Erhub=yD%DG)X+0gBrCi#V~ax1L~mU#w)pn2`Bf1Xbt|CQV6qJ?1apnOclpPA
ztf~2;w)}>l5AwaX+2Is($D`c9B(lJ^MCd%!7C<L74}68x2r<HwraQJq5VQe}QhF8E
zgkGZI3hw)KgZXDt6P)*cUA_t+>QdK`iIo;l;PLP&RoOYVNNP=J=qi}6Thl*>5ZX6n
zLeaKYWnct*K^IwiUl&)Nt&4ww*;Jo*EG~qe!&w^Mv(Y_IN0_1JVY>T6567j?0#(Z(
zO(Ar#4Vlp9;nO;UhXE?dXO@Z*Y)zwU1SKk6iT|0GwhN>MndRYefi8Ce)T~<=Z6msT
zkT@bOBf55`U4z_g&|!yX_G%3cGNTAjloK|S6;GCq5yMldE!*fAoP7u|T|Aff2xq%#
zktmuug|l~fs_ESAnqv_rkhg=$ZHcbssiw8VIW)o{%TrA$js_@?y-OiH!8Pn|+1aq#
zR9NjSv9_WB0dx^Ez{au=Z8JPU6og$;;vKs;dL~lT$n6nNaLY>CBR2sYLF7@@rab}`
zrUX@GDE>+fuuATL{VtTB%x)t~sfvWyEa)t5eC_0kn<*3`lkU8wH{1lO)IV}3Pg)+R
zgyI&+d|M+gVXEhev^uVf3@8E0E_pM{DIgEwNxZV(>7IvBWYf(R1TZs6xKBUXlS8Sh
zQ(kVBXblZIp`CyRv(@MrAvqJ1P4J@DJ+XJ*`Rj3+vzEk_q~&l8^mq<c(kHdOLo24Z
zde}$n5u;_q(0#E8AA=eEFhqiq*<f^xn9hORq2g}%gjddF=Hph62Iren2b@FCCV4g{
zh?Ohq2SAM`Bd9v1GMS-)%w_<Iaz`xC(0ViHo30sA29vB?m$f~j2Euqz%t??>I{u(0
z-U)G~JFkr!xZQ>yJ+v7%ZXl1DXV`*&4=P39E;gnGcpSP~Jr5d!f?Fz5!#~Cm&1N<V
zp(IwcpoB8dhMtZ#hvYuSjWvO;8~cbaVM|i&u93%(&7-o8xjA+}`;9tib>Fl+QRKN%
z2NmU{tI<^co_23C-4Kuo@HhzWn`XCFz^z3VeS|1--%_{ssM1;2uF<jiQux9}thhT0
z;Og$2@_qL}>Aa)O<J^b1SvY#c`7ppKxDT}QxqDhM`dV+)>QFS1gB}{Zpo=WJ3m*hV
z$$j%Yuw6%Tdw)NtrdVQhGp@c<#)rI#UpgbIMKKiE;@!sZTwmq{t5*SvULz6<O3Pn1
z+R@&L+7R7jec2{9tWxxRls#<k1bl}E9U+0f*$DRIk)jdhx1gIkbQ$GA<=wb}+YR_$
z;(1Gkdtak`_7^3k>sxhvi5my>@kH{@xWR@P{5ZV_rPVxHpW>TPB^u&FyGmAH+xqd5
zW?lplJ!cc!r6TUC_+z=ejGjZ!j^5<p?)|)DUDJn;CbN<o)|IobR#?wq=086h==<}1
z*-+=d@1C}yKsuoT(S@1d>}~_J{EnKmLc(0Qp#xCs9;84#I>%O;Tj^3;6hLf<l~{mX
z&8XrjQZoy+)fH}Pj_vgBzF&CBiFFg(1mVgsLpCfXg2>$!6?0}&nIgiY=`(`T9?Ju?
zHHR+wPF&iK5}mCtdp2jVMP=P?>fGZd+gh<qQdVvrG+in54XP-6Q&rd<s*a^roZ%Dt
zdZ3h_b@iBxr4vWcN$E9fus41nZ2<WFlaCBPa67-}d?_#5RF%SHv!oVOPAn9X^`5H_
zHm)ua_#0W;+-E#nmn+GkL+u>^TO{*00-?qI)90Bh4=~kB5wz#6h1gi#KkXI`-ZuiF
z2%0c<hTH<E%BHZrg9pFW(y|thGDkxJP!&aptrEHAO-*3)C?4Uu8E8SY*@HM?c7ZP1
zF-p`Z5)*sqE!NlAqY4gi>m7&MJZqwDIAa~f`uu$_!>M`a<%?^8c82;~mumh^uy5|^
zEN<y%ktXOC-GtK9_8x(!>JNQIdT1lDLf}LzLX_lys->D@W?!z%n&>7??o=L^(4;RF
zZT6hQ9&{LyJXVmCHw`1iH-WY>VhvS%**W?ptp;4d?F7`3Oz5o~r3dqlDG+y8eYczv
zI)Wy((Fb)0U|Ei%%e{D7D{{RXR?nj~DVqW8sr5+_Rzh+)rJ*;iOZ4<T=&|>__TsG6
zHOC^B>d$LNnsB2HDy|9NWzZxxRixBb>T4<`=VfSg_$m>7_qMtdI2W?K5W)}&F^?$d
z*nC3xX*{eFP*MJ?OI<bQq$b}*D20$U(BrvwkFe%eMVE-~o6=coVlj}tt&d<wQ@2f*
z2yAd=i-PB}rR(i)YpNJ+>K*7|PLO{_U=9>DVOdxG9Xk8GS+q!6?xiHZ47p=}wLI0D
z$aP56Ds*Bt#Z}99u+>8ddP{W7GkGP7MmHA!@--5|`#l+5B6v%r$7iuw(<Z{)F`Xsk
z#5W-_yDNIUSonXGCbBw+KyuOA$vyE}%*Y6DJivYTq!r9=(1kRQk`KWf+0>FfPRLs$
z_@$um<2~w{mI1P$&-N(MekV3%cM~s+VmCl#!MqQAj$GR36@d#@5hz;wfKPZCn`m0S
z^!%(!-iU(g#U>v8yocRTAQdg_J>DaVSeplwAnMdack>_~<TDGj3@lb+S}k{WnnVx)
zmDWp*H;E<@bc9kOZoS8aMW+<uu|2~^q)Wu`U1?i%limasA(^1h^8`m^Z<!iG>y8cw
zpfrG2BvPG#FM7PZT0Ql+@*+qR2|U*lU+(@3&wc7en>qBXFr|3#JMwd&%Qx2c%*1wc
z-djy(C!mK+&;IP*@_EnC?k7gFX`kzX7LiyaP_#X~D|DINM{{-2K+gj&LRT~Z7OHv!
zNfcSjBapn+qoK`Y3-m0JUVqYQTArx!ar4O6HIG1#qdG;hYqzThAdS3-CuDbyJg>Fi
z>}~T!R~5wszwh#7gc`N}h_cLEhXi_zk0LT1GUz4HO?O98?1K6$bdq{kA<&iXJ%&3)
ztH%;Yuc2Nptx)n$fTNcdV8J&;8KTB5pd<)ZKa?CUJJGXz3q?Sn2I_<pC{Xj-eL{`>
z5n7Plz8;`mOfT4B(d`qImMWqJJ>{m>g-%Rq7ZK<=GqyM}w#=@Ik<hb*i!AhvE2#y#
zI=N35(-?WZ#u6&-UAab&&`IVEDT(=l=e6nr;S9W}k@SItG~xwQQMZ?l(ZhQON|CD2
zb8lj{Rl1P^Z!`j|5(Ve<37F`01bJ3qt^2i~ySNiPF459cc`l7lpbE!b!lLIf={7U{
z0p7;0LO1t8{Hv?F2s(8Rsa4#vH_)>T)ahhIYdHld4B~*5D02aYY2wqR&MDE2;+~W(
zZp#ADvzIyb5oP2%2bJhR#G49Llth)X<i)VtqXLwir#N@Vpo*0PX?8-jLrnnp2(+Rs
zGStaAo)yLm!5wKs?3stv?SmM$o*dbUA)v}o^4Z|yvDd*{4DET)F^2Q-_~a5j&f;{E
zWUKFC=YjwNNpj7&<I+v4`eatUy{(1Hqa3a?oL6hg%ME1?^9uI)sGM}SZfUYmb?4w7
z*o_3&hhiEOdL*TBgA_tH1QRqEF4Y~d3~D<#K#K(i+LTR=9PJij30-VvA}+Q(XhWe=
zPsU<7#6QBQ5pnX0EP{4O)FLjfTL=M)z=rczZCaiSR*4>t?O}O@i3%#;(zKZ~IG@PY
zH%H6bJnWh*&msucSd1W$AY?K`tSoti<1m}hy5oK$3<_)1*`s%)8h!V5-}*TdRSr|a
zBUW>4sYD_4e)m{3q!ZeF;ft6v=-K&a_a$M(JA(!d>^Ud=j+N-_PJNVBc0u}@`A2HG
z;qnIvCR|t>PoC^-l0|$-+m})Vips9Gnth|$0w*#i+=RVGqN`5dmD*k{=e6XgGWNNZ
z0idSNJVSA<ys$0Vdv2q8sRl~KP-N;7PmB%`t7`?!9;`x*wmBTN-ncWUQ_{Gw`oz4@
zBxQ*k@)qJ_nA&{(nRSba;Q>;YHC*L0LFhTYB<!6hL*}aMYp72=2Uaa45>S13Z02-{
zY_<>AsAzfI*{BwSGdxQkV8WY0cB9Unzdv6*yAmee+Bp=qbOE(@A0ZcfmI<T`p38a$
zqy_}gqU(&nT^&76QoAqcQlqW?A%%)q?YVL&I!KQU3zGq9P?SuVK$-Ex6>n{hwpf)#
z2{`7%6vNB%T(q#!WUYlELxTuh2bJ>YTF;sR&4l{+u}2<Yh`Xh#`7dMPw)GE+&O_k~
zbn4ThJbdPx1Tz~U6KdIm>myXHbEj+r%PTt<)P%0RVzE@yxGB)C(mfz^FPcI;FB(Lk
zWkPxPzPe{Zk>ug$rN=ufW7_=mxg4NrjZTly5=)Yv$l02v{N|D8YV;QTP<i-@E!Wh`
zMCghefFFp}t$O7iA3lB{EE9ge^Grj%%BOk^jhq#F*+QSESe}2<#EXY$_LMw7?KwN>
zr7(dR-J^I15H-q?R8BO#FLjqZ>ZWxOui-J|;$a~RO7d!|8@Jsn*y@F%bJ5Q>z(T_r
zfmHG)O0$*^X`;ltxz%v_!(~?D9illz(_CA=@1lk1&8GWio(OuR`H5!9pLaSt=yi?G
zW}xSTJF7d=AL~ZGg=*KBYMM~<2<Ofh|Bh~fA|Q-xeMZL^uP-|3m%~+=wVbV0#K0s;
z(ncQEq(sgFvqs4yY=peI-O=Z{7Pa}_TFZDuJyEG=Nou7g2;)aXk3W6yHYobV41MOU
zzyLkVaH)Q4FQo{yJ-(VAzwgPRZBorOq>s9CD9jyxKH7V#j?}3}KDSpPC=7UwB@0pl
zg`}!w>)AIRB2+mkR>Np62D(sp)XHtK_JbmfsI~jKSwB!4&`VT=vEB<KfIPLJUY4dn
zMD4Vt*o*_93Ca!Ouy(6<go;En#ByGV7Szau+SNq{g=y>CPn+{a7x58p?z|;JpiMmv
zhpJuIi=xBiYkJP>n0AjI`7uo6;m{$X3v_X~01+eeudW0((VsyVWrn{4h<uB`oi<0H
zi$de6J{KiV1XFnVLGQ8$DCrFrliT(gaHCqGZ|0Us2;9>*q}lQ<RIjBSU(($Y3(X;*
zwM=z9dva`{PXF$-*N7=(q^L5UN{YraB*S8NN7^f%A%Rl!9f_uCd0fcOO$+jvCC7G=
z-%vB?5oalQg3wk`9jG@;E0mg2(3O^c!4sstohr+Bf~#^=w%7N<&PzSOj$8bmi3S84
zbUnAp9q49ppDv-Br>E6B;Jl3)YC(MqX311NYpf;g&m>`yK|yvjP`EnUdyX4A62;Kd
zp19vPi#!UI9*CF>i|-!g9>}anfL!%ffkvgTBq`d?10W-!1I%%|E`yE-CykBrxjKqM
zVTbEv{RT%tC**3Oc4@IfAt&o1_T~M6E>U}@ywsIvLvvoc5(%ILCyozdG<MrEg6;}^
z3PqrxISIubLb>PGAfGMk7R+<dpaFswSacs!-nojsLd9!mx2|7~0Bc@f*lq1D4DOTf
zKab_e260H}o^fNOjYgT6f{8t6(5Lx#yKeRbaKQKwxO189x^lm+)=Q#OEu0Yav8Uz4
zmYO$Wutbf3P%_=1y^WD)gZQs!GoxRw&`pMUN=ahpaW^SSy`mNmmk<r2COq=lVv$sY
z<qTm-vlg!sdZ3g|OesG1JVH@l2NsXjY965y_eXSYULpfvBP-zRbdS}SAW!p+Y4xjg
z3hE8IP)7?^mQY1=St_5{FV+6w(W1P@esTs=VMGaD#UtZGCX2#_?Ms#gJ)X?g)5#8Z
zLJ0xsRcc{n7d5vSvU2H%le66fA7p#4XXnH=uDgFv7V8fUBFOYfl_wq<%vsXT$g4ax
zu8ew7taBtwkZ223lCmhKtXm=^aH01jeU`CuXv0p{3id(>RH6hE(ujbxL@}ZJro2EA
z&0j)qo?76djBTny(DI_FAeo@&xwM@B-0~=fdYjBZnf738vv6*U^n$S7)nvWqaZOz;
zKov2$=;nh2oit%`lyo||Y$=}fSeca~$9|`uLEHqjRaibWVlsfHEyN^yHUgyzrhuYO
zsssF&kDs28|NilO{LN2)`1tA1@1Kqx_QC)Ebbd(&$Ou9L{LilddL+@Qp1=J5&-Vwv
zdKN8CIE(jG`RA_(`qf|m-3PVpYNc`i{r5W1Kl;hlKJ{23UymEuFAnSf`)B>L9Ps%l
zAD@KE?mz$GPhbB=&=>sAAK(1*&+ka|jUQ?E!n8P}%e^i9|LcGM!=LW8{f*D-oyAbp
z>iio5JBRqopMUyKe|;XLH~jth`JX=S5FhdLKYsk(@Ba1ofB*B}o>9*S-|qS5&nUKF
z`}mZ)g^zj|_0wPdoG1I{K;K^T=e?BA_gen&gA3-bt9M+SzwLTS9?Wyw`2hEiuh)m%
z5xSE9bFa638Kq*yb4(5RujiAxW4=8Sw&@-7-@d;c-<;>$+o8++e6RBlKYaMm!rF&Q
zqxWyS8x^gFwm&}0qwH+nkaNKIGGbH=(&hZ;KmN=A{{3gQ-~F$D{+B=g{`0uRzboHg
z;rBoM#WVLS_<S$n-@DLHeBt#UyilpC&wu|q@0ee^%;(oG^F8*Dt}<uu*T{T7)nMlA
zuPlTRZWmdNM139MPJDO0KN#zmj*~+@jjwp0Ui?=+&}DHy>|HnWFF)VE{ilyVeEi@6
zc#NGxkSJl7rpvZ%-m-1mwr$(CZQinN-m-1m_EbmAqC2J|=HGprjL67E=9A~eIoT_T
zLy6t6^8`}#@^W8Yh2--EU>FHI{r>sn`+gj~xp`?F?CIQMSaYl4^DdL7JZtm&kxP4=
z8tfkRfIl!x;u|ZSQ-*W#9&<u5r)Rb-4E{&EH-vvnJlEBjt=&F6^BVUWrT6*xbMZjN
zu$n#I>3hIg>3cn@{9c^i`ZX?z9dwPsbL;mpyzuQh`x5st{q&G?3C{*Mz-UJEt(^6-
z<1P`r#SYT#_Fmi7&wC8_cE|d9m2OviBFkj<X0;-iL6sPO!tlLpML6})PI_=lB{w>{
zNyxW7*n|5JWo3|=cf;Mw6f%$Vnd0}O=fr=`2xVex?BwidVqo)MCp$w+C>BNndV>Et
zaC6g%T39=qIMRt)8#tQ?n;6*{o6t#{*qS+;6EL!}v-9#o{ohO7*Q+(hVh`Ejx?j|W
z;De>0n=v2`4d|7CC#wUxVBg~t6xPZ{@_&996U8Nysl#wX1XYL@T7(s}@zBriUqtkj
z-^=Lwi+|4V-_$(c-QM4v9zO7I((HDwL@G^GDto&8beeSg+9z0g9WRc)?sl>iznQ+$
zA0Ol;%X$g-<!<ffN^g$uTv@ETj%w}vx_-zJT@op(q!!~<sSmRq?bKk;OLwh&It9Od
z$}Di{Om(DlN(dML;@FbX?kW{oc{aR$B@3m`>8v4XZHwfE&%=2(^aW2<6Uz>HWa2Be
z+!?#coN}zU<q{3n!>Nk5D!M}^@BF&^%g@V>rQK=ox#UULRI?;v*>c5go%m*;7mlcr
zdB!&u*;?QL*YaB_vJ{N;@*AvWQdl`r19Wg>6j{HedV{-VA-r?6@I6PCC_UU*U{te?
z7-$eb#Dm#7JPiUSylCC%*ri}uT3fy^YX^0rRogpi536eWM*D%uDwcssdtVkN|4*92
zk^JdhmUZMTEH9N(-4F!M6KBi0IkhzXjrxMs;00kCC*?``louw*a?r`&eUXcEz1wUM
zcP}M-5ph!TnAo2k?0F7H8!x?!M|nS;pSj;x29wa~$VoYW`MEXXvM2`p@`0>TJD9w1
z?9RBl)v?9ikDZF?GcXrG*e5g?T^Rj^KhBq}0b-SvQL~qjvM42-^WP5pUXuY>zM4~2
zb_xyTN>!4esmu_O*U}~=-53>FRa5>NC$aM75~JMdC#VWhl8L8A4elC&|C^0y2zD<I
zrrnHVu)%r2o7GW)X9+pmM|FG7$vrFHGAbRT*&08Uvo*W1R`ZZ`B_#>Wx3o#%VE~|h
z8Gs2C7FD*I&=c90j$qkVb)of<s|-Q)+knBz?E_B^$Io?ETWKh+ZXN)xR_yzkE6!t)
z(F-1OKtB$$X>RS!HzyaB>=ir!mkJ_AA0xD4G_85+?mxm?$%zReTIl+~*;XwTVrUQA
z7QA41Ku@*L!BzkLwFdifbpR^^`z)-sz#J(HPjXZZgIp$Mwy=OC?yBC84{*7T&bj3U
z^&cV}KRHUGA@_dteT=|a3+e-I0os#N>vT+%e`ca*Js)<267w_BqgX`kl~~qUhB=OD
zh<q1e(Q_h5j4cgq0pJ{ALS;*;iWoy>jyBz*fpu7^TKH#gWlOq%2dN@nYt}0T?k;ZL
zJ|C{46N`p{u|^8jRrNOEiR}dCn^nYW8dxD0<*wQ)7|<B@4?_hw=+`dXbAEl3sOpQY
z8c1^h4n4{+`pD7mjU^oZHib+sh?9q*Wc<D~6rkbhm1zme@BgLaUp_|I46U)I8J*J3
zwC8E|woE^MxaY|{DX4!=W75w{lgCbnu_sHKr5r?N51GnvT?VL+PGQeRE*J%}l9-1@
zjD)nkXqrV-{O>ZU%3ww!^K8smbAYg9R1dbhbc3pB6-rZbiX$-?38z=ZgH5fX8^m5`
zi6Y%N&w2v7WFBO=g-9v{ElI)sFIrDd|8#D<(coXu2qDr&F;tCrGmamJQd`wt_R6((
zc3<{0@aBJ%c5as2?-@#E&&p+vnA?^*1(UWCAmMk_)yc8;6P!+9t>NiS$l5;2T<~i1
zyItXuvJQPBu_;Y73Bjqgn4*E<dAr!4260>J8&V1}61u?uB-}(s9;XiiZJFj5Mj9rv
zZg-W#k`+hC@)u@mpf(HIsR*<k$nxc*49Cyck}Es|5wA@<*+8G4-7^Zj*iB6L%nbXa
ztr}<zsA!gg^~8v8v7Kq`Os`^fx<6!Ze39WkRd1Aw1aC%kwvNhN70T6c-I;VQ<mb2X
zxR^!N!vJtcNxN)0jEFVv+r*GRNsfI!dL&PARM4dt>|q5ptobO;0c_|9f;%n9se(yg
zS0fRZBxBZzclB#n-xci-5W?Q$OVu1cr->_BT`9V&JV2SI(pp)B!|_C(*k`zL$_1h;
zT%-$h30i9&it!bp=#d=NP*>d1Lfu1cQuT`w9<=0!J>hO2g?)rPiL~O+pjeGJsaQ-3
zNZF{>(9T3Ipx&6&lSO2cW1)Ez8Zx#AOmlRgP46n!>NMTfJ1>wF#}(E)>08{ky1JpI
zAB?uaSS_+o19W{HY<<!Os;<^XMKzbB*NLl98zt=9{T+gKR-_rEnM4L|%#9UMs?49l
zIzj5*Q?h0YSnF?+nUth3E%k(z8<SQ|XWaENR^C+5Mc!6{{yx<&<zAzKc+E3MHDzvE
z(jZ(qyIoNWKTZZ-x$s;GHEc)j2Qbtp%i|D%WS^u*6wMF);R29)1-xv_&L}T%f+GHL
zF1Mc@A(+l>W&jxB48_QK{9`^>05Efn6iDu7u-+yhR9O57!!QqbwOvmM!MJ;n$&n9&
z5yF9kV&h1Z&|ZfhodZl=CKvSQ6cL=?&OX2E97fZWY;P{|k%$vt<S01F%r!!fJ;{iQ
z($X;cX~7@3Q4?k-S162!g)%@oCO7|kH%EG@F#y6&)aZ#Q>0rG1C>Zr@eCT976>}F;
z2Hz}4XSv^}!s2xLO>SfR?hWSXU}ie>Z&x`kcE&OjF-4}X{Ij}z(#VfGSW{43J&yAX
zh8W2+QJ@PsIUwiIGxm^clgTvE*=5McGKR9;(CY+c2A>(OkfalLuKfIHgSY_2ni<?|
z%&`qRgg+sv?|3?&zV%G&EkL-A(Ii>BpGuZcM@Z#Dm&Qj^ff6tHf<4Un9w)HpigT;$
zI%bxrg%wx8a=qNz2pqN=S#>h2Wj)w=0qwTQUk|6bCem&O1;~sw&g|3(to&sRxG@+x
z0_@-#f>ZKkro@jn#rYbi)*kD@PetZ=N|z{{oVmmH4*pmH)(t9PA{#3^c~;*tz`#X0
zaMekCV>CuY5Ui;^18A5RZDE+<sXR2*9gz?$cxWkctCH1=XcQ@xmC|r*^D1&8^mNZ}
zyF(A!!Fwv&C=^$nR92QY>h7F2YFmH!pgd<(xlx_2g+2D>A-sU7d&6m}RfltN1Z-Y0
zvA|%unOKmG{$b6Gl0h5nPSXzv&it`GXK)-&oThGvp;eZvops5xU4PNZBp(cKEvo7i
z>$|aWQH-GlsM;MthX|f!w-2<f>D=Xnq0I<R$9X9X>UGQ><a`&mhsnST+gz9IH0-dU
zx652&YCLH-e494H=lSM%Pdt|!l>z0J=_F?QGxc-tAI6p}xZZLjNa!}vMQOTd-Db@+
zma4R&L!=-dl~-?TF`e7$&yUU!QwgM```)n8=ojzLqboTG<CY33F8WeCuM)jK9L3MP
zH>~8{nLELsWGhmw*D^hlm3CXqOr2PmXt*}yB(3K;#~D#MG<fl{k@p7q)~3l{NGX0P
zj+d09bPffJPw0|QyFw48J4@#zQN`Dl)5kpRKAtRDNwA(y{zW#rZq)-Dkg?BJ^s)1+
zYW9on@oG74>iE<3Tof$Z*&&MiHz)avoUQ!jYVLY2$NNJ{W-Wsvdk~&br83Cm{U{r4
z40DdY%OgA>>JKRT-Dml?=zCV$$}rf4o9LFq8&w~x;0bz}zg4%SramL79WGZLGY{xK
zF-uVrbNlBq>Hx1jHN7_puV1Y{!18}QV*VG3{q@BE|0$M%gPr-mQ|z#ol-(gKOwWtj
zE4*+@IUB;Ee7;V8D~%(Sr#NE<&r{vU4?Jphm`IEwGhz-=Mi>vHL=wJUVwca(BH2RT
zI(M;Od-rEfY)|n@Yu1b&UTdCTV^$*3czMg{MJYd&m3}8MaTUzQ!O@JD1tIt6)2kO&
zC6dZs6BGL~l#MnPrFN4?zB4*yb^ppB?vS&f5GUotZ_|V!!PK(O=Ovv<HW?D7LALC^
zDco6Xae(xy_Mf0fPnl6dr!Xtd8#8H}M7(^&Bn8bB-ov9kS&lHbiW23D+&hctKxZm|
zmwU>D$sOvpFfN~+7|Gs}MpoRv-%X(Aia4IoqA;F~3RA@_M60uv<>=0$9FeC3{FtW6
zIS}*RX$@t`gDO~<ri(%Hkb=SQ3%u7WVo*FG%Q4G>EKSWDIZ_bP>=WpgFZn8fXqni#
z3oq$}{-sP3)-i@Rz2#{zK*+aMbgvJkpZD2&0mRJGpjZjyI*uW`A`k^cICMSNAlFS`
z#(Pl|1qAXu{x|U*MUU+<zFPUrKmyH0=;O72^llK27&K0Ki*Rd2SuRiQ)GE?)!S0{s
zG+f}oQm%>^!$8hw8)0MqmCt!2K9XdQ2VNe`^3uuF5@AK%s2+4GZQ$#Xk^6ajjsClV
z)Ept*spo|Xj;S{<YQwY(0qXs7->TVIdAYSkRDkMx;EcP1^HZ+d4ki_@R1_WBh2oTt
zSYkal310QnM~e873fK|FyAG`p*NtQF>z%mwkwgUOC5uD|URvU!l0+U*C2b)pw-@>$
zcK+Lulv7>L7AzFn1fFPtW=kMm02J-r^wu^Wl=F9!u9EW)%rsjf*^G1qgi8?pVymOo
zPWRP(5|fgyI;#H)G1JLy2#k!0`<Mdgf;odYBKTA=Br|@2pFWPz;S$1uKi~>jFih#c
zR<A;CqEfd?NjrKHZ-T?lu-7}T+tMLqK}^yW8L`TFs=reF?FJ3l!LT&fp`Zd#g@Q}_
z2$I9ACY+NY;Ih(VAuKDX+-t<BDfOff{Ap|qcy259!m?ILJ7*7wqOWyC(elSpUvtpZ
zXkzad^*{bJC6jocDXIfye+>1TkiJr$nW}5U{7jt3lu;<sTwJu27@-Tin_R8W;W~^_
zd0oP=(VOe2&6t&bYfr@H)kjdp`~PXsvj?6<bHEn+yaKlZ*WwlMFlfJ<d@U-j4&x?*
zM)=}Jq}T)OxsYmGg{?bYvT1tOb(d1K1J$tX+^Qalu@f2ACm+p0LYyZb?jS+$1&mu#
z?lvz2p%RawTCuFf3@&O}fnAi^{}2T7iy|`f`2#pRHmDN^av-_c47$lhY&rk-ftR19
z-XHTrnv)X+D=7$i*fZPnj=l%b{}z3kNotN!HqRBJ@Mv{2i+lc6kYZr9_kXZPzKoX+
zl-a-&rW3o<{u=76pTz4nPfyVjYfl+)_oE$$gr}R4XYYosxV@MIH3D-=5U`mJN8y}N
z!QSh~#G%-G2n606M8bg(8QJAObyR*~Jf!RZr1)I&x$=HZ-txqgOe`2+Gk}U#-6IBh
z69R7TYpM{#KP{?EGxh|DU3<|?c#?uMSub)0wQGI9foIx=zSwFv2)*5fVM@OB_&pdv
zG6fSEr4u@948nkpM;MsOKWlZ%=o#)M#F@3Rc3bVQdqJyYK9ZPdxd#zlx2dvaLF%~a
z)5O?sXP@^agBJnA2g^C>kB1EgP>S$^<B{+nz>zOC1N%Y@baN{z>g@F&4<rYhO0IKe
zLUAVj=$-GwL+aJn28!FfQ8xVeQED7qHm{s`GVk@D&HdgGp3G~VJ9ELm)45<;Gf6yu
zM_oz0l$`DiSyEbOU|pG;lMEXjL;8R!0aOD>DkU76a7L+~85Y^T1rohEH?5fbp_NX$
zI(uY*@EV6iOg)b3_RQt&)3I$_wir~b-OV-$cl!|;$=9wniPyRz8NmbwTSW;8(jna&
zsWfsDFqnibI4aZ-u1dYvU7(m2$}sogUcn+Bh=)7Wo*9jK4kRr)n$g5qQ9h@;O$=do
zW#OxJkKg2L$H612kVGp+TODlSWfev+K#L7RGW`@jwFZ%>!O|ES5qE{lqf>owHuc0f
z+eRn6XPbhiIQlEkra!Xmo9z}R;p!Eq>AI)n+hn#=i3{^z2t`}o`bN92crkk2nVy)+
zr2(o9)*O}Cb57YmOSkH5Dm5tHEH6se7ra_j0FrJc`L<e&Z6Y>23;gLOD?*JmQp4F+
zTtRzD)l%&Txp3Y%0d|d)CUwT9CXV^p6O;PanWYrKrcU4)^h>1xN90*q_Ih?4Sxm?s
z+|$w`KKo9lg6V4=bH%*EZYh`bN7vfWY*Gl2nL<V~Uqkz=PX++v=3|FbH>C{+tRkTn
zIUThrjC(A-px)Z`SGz$oPPdl6e*lGAXnj~pI*)gU;lWL+x!kAVgW%k0s;c8FT5roZ
zJ(|H13f)kuHIyS}?B&U<;B+}p+BaV<bkj^+9ZPXcAF>FlF21}JYnw;&#sy~;_i5#d
z5l1d;buIt}U5AHYGlT`?tW8L|{YKd%)MzfM?Ol6V8m1R8-K;0p)W~bshMlN3__CDq
z`i4C!ePAE%m!-FCsJQ(U&%WZCWbvNKt2v6BI?Wux$9Q6X(2I^{YWco^@i9{B{$C1U
zrT<STfRTlX{=ZYel;%Y2B`e(4jM^*sdV+8_46`Q#KN3MF%(X3jlvZnpDNEGh4?1x|
zq41bvb7kd8K=J!}gS4VIo`UYT`eWg|pC=^QS8Xc(kEB~~=?>q^PfeD~TW!$i%Eggg
zuPoS)xA8H)ly>#YMAY*c(d#g0_n%&x^+8Lv*3Y+|&D0(IJ)IKO%CQ4}r(v7Srxm*i
zxn6U1XHPjB&o@`C=&>{jq*l%>(HcKL6{tW(S*oL@4ZfbQme0>C#P=)`yDcmioPS(p
z$t-yUTT#~r`BpOTI&RKt_;c&!osK-HtQ9I^Oi)(2O9IZws9E^Ged$i$SHqZTVxJ3G
zfSrM@6H#l+Lv@s(1C3VV-)D&{F3F9U-I9yQ;Ca5&r6R)n=B&DCtDhfT3QZD)pcv|W
zHYhfq{7A3{C!y+u;d~TMY>q20)fznX*f$SxDF?O1EuTUx73Z+>g14DIoLf7|^L*la
zdmeHIYuR$NdL|oK@NFm&6@*)?EQdv&YIVmGxlZ)M3b56w&I!IZ$<gxA3BDfGKT+4N
zRAUuwhPwURpfLxPU7SwU>vD~QWMU0C`?7mmNmQ)MNx&aAum~C}rSo*QVZ^w8cA3$%
zAF$k1(B$C|l@S3-3a^ZA=ck1X#5BOUaI5-i7n>CC=wau_D-ga|6{n7w(vRZh$I5yZ
z+W%PQR8*?#rB6hQlIu-QO)P~yoMDfXYH7e{N<ynh@jgiuKL1EvFoVr~(J_k6y#J88
z<mYDmcr~(|e`!)b>k55VCPuQ>+4ha=12(|Mukj9+GYo#VI>v_9t()gbaBkqVlU>5x
znm9L?3Kw<j=mZ9ZIr9NdpXotWV#*a=RuCBp=WQa|rgQ7{eW)d9(r@3JQ$U8#D`Cxf
z5PWRYvJh!utda$hx5U4VVrtPdwEzsO+;?Su5SdVTW}57#-;#>@NI)|a^U3^c+C=o}
znP_7CUE|xq)`(+ob(%!-yP(Q0wkXAkQk0fnrZC0vK5U)A6_x>exFkwG!R&oFPA{|#
zD4OnuC98#s428g)&W@~*V@KCMeqvHuM+4L`K-s+V-FGr^GKaL(hvGC6n67LELwdEl
zdrYLTYKx4JhPLOq2apt0Vuo^oUIzP2{Sa<xcZpIlq=3rD>dxO2CKP4k(L#NSI--2*
z-ISE7rgIp=bh-xcO|YqT3s=|iC~(JPL8Mz$FXiV;VByIQKNHUAblx6JLynVpnUnTw
z3Q>kF&W6SjxG<L!UkiApuq+j}Ej5*My4_ed3QBEGFtS*gf?<>t|HwL)%T5Q)y1}MN
zowtk#88-G=SGeYHl%1)<zzY}`L}+rpC{{%m_H3Up@t~mIo1-yuZ_bC5wi*Zzvb^w4
zxG_ZOi2@20N_t9`Yg(5hW-c30gKCkR7oah;;Xhkz)`b@X-*-}HTl9Gs!J9~2+7hd;
zb()d0EzLx1ip2LR$_D=huOo!s>8T(I(yn?c%DZ7rbsc``22QT0Z+klH&U-~%-hU0F
z2tTvZ=}JIi@Tu1TN|dH>5dGN?z+{S6L(ASQGy4s(zE#_rTLqkHEob34Pu#2YE+TDq
zPS{8k^GguPlK&FZ6>Fy8*{6*0r8DlFSZ5;UB(u;FLrlk@kpf{x642=bYldH<#}=N7
z$ja1K4^{2JK`7Sly*G<-QqqZr6Kan~(C!kBqIIE4ZG<_7XSJn=+$_x#K53OS&)c;{
zw$YMN=_BU3v7}(@$5TSn4xt0<@IcH@m4^$*oGl9^2Hh9|XoBMy8M~8lRNZB~>OmH(
z5z3W7P6-pdYVLAveR2jg%|XRfPr$Opv1*2$Tg}*O%Uwe7i7x&Ym`mz-I*<)dlel5{
zQ1XoE3*o?@nQo}&)RkA!pC3RmjS>G9RbQE3j(AFIwigSWl^l*@`}6bg)T*<DDKBu-
zq?D;R1ie4Us?r^OHIKJ)>$~A=mm5aHs+S;E2~~c2NwluK^sxEKefc04_gPG?%4J&<
z#LpZh1S@rNY>X@Hy7%X@XDoVi2tJfYz!TKzX@)MKf($y{td48lakn2^PeY7~eOq&t
zbmYdQB@;|irFUqS!X*qM{Lnf~*ztY-*@#WVgHBKcsQD=R0x;*D(QMV8n86gZWva^?
z{>eacp;5}G+6M}BJS%Fk#_r6*K2K1k8TFEdCPy4c2Py;lm;#AG4AWA#i8Asl_FQ)A
zQ1M};Jt<Qx8)s2p{os;`L9|w_MEBxfOvqmo?a6A5Z^gbnSx6IXfdNoJG}^?$^x6*G
z35(%6sd==bCrRNvF=K9XpT#-x<}FW~wlp_Z)rls4H`s&Q@{d)n@y@UUbv!1;p-#>D
zei(v4yeJ*a6Esx>G{bCCo~~$V?(F5;vtTvd-f*W#sszGR*gnsM&Jj;@m$gSIDwTX5
zYSm>Gx<%`*BrRh;Q(-KQ6Fuqb#=*AkK+p8Pg_q0cNW^QaN4&w*mkt12kv)mH)C|G*
zdX^l2%!S!M1}cO(&5phE{_F&SyQVQuU5%4o9^}xCo;CE9#C)U!xSrI+-b;-7MOqIV
zkn3%fOrVYB(_Qs0S=c8@XyXujLLz3Fd%?9h{xLpZRWgkrFAQtFGcUfd9W9n%lGj<Q
z2Lk=OcmAE7%hje!{TSfPkLoz|N>m8H-xQEl7j={uQmtkEKRJ;5iy$J7^NA<CDq-rA
z-oLANdLd^s%lE|zmg!1d-5ttDN`kXjc|OA=fzdnYgGDQC8fWv%oPuPC3`|2+!|@+d
zkMDhjNFID!tnvphbdpy+)Hib_?G=EaM+m|TJ!(tyxqMAiyY|*`XZ98{{kZ>Dj2^<$
zA^zx062U${v{chMpU)rIeiX{$dm=*fK1Oa_^n96537~}=bw**p=L8W8JBstLDOGbP
zFe82HXe#j+vO?IN>~V8$Wf4Suz3I+mQ|faN!xef>ET#7|x;TodPvL>g#o?iBZrPaD
z9jIOe9c3?6XSv7?KqQ$Wy5V!C4VW4LO|C;K?a1MD%6z>&eivU-1EK2BV9Jww3ODls
zeFQ=_T#Z#HeQ!1~p~pZtd9b3VCk3dk%UD`7*R!ZxS?!_zmJIcGB*p~l81Y-or47ZO
zj-SBzh94B(Rf1xn;!7pBkpfy*?}TM2ol*KS)CzfJ3B~swTF&az9km;aw*@8{h+rO^
z=uQ6GeP@@Qsmk*);E;`CY8tDEUdVGi*rhnrO(ej@EQT|Rs43H#K$&}km^2cj2S#V$
zhRAX)V4vMkz8kdYBwbeUU3G9e(H;f}gl$(S>0)=KTX7&+G@6rGmnSUu-8VC^*L7=7
z)_u|qM}WV1x2E_Y&jYP40<E5P26Ae#`P1S|Wz?&y^O9}FNPK@LdI42+Qa%0`PGtIz
zIFXr|@xOCoHCAUVaXZ6{+9Nzn9CZ(%0T3`R@HAZRq&>Z6HSHZ3`qoWn<nVq`R`uK^
ztSfk8$CzPv-sbn}Wa(Yr8quy_x6emF?{|b>Z=Ub>m(9nKS|et4!?dYar`-s}$=9=B
zocH#&?N7ToZsPKDtl#asp-><BwuMbkx2B98i_glsjs@uV^b>MkSPL+94L|$#F0v53
zeLHmRkI%=-b6hJXapxdJzcn2zrPUV@G5Fa)8S>>l@O5;c)O{ubNia1nbZcx?XSZM%
z{d@sHk>2+1289>2fH#K0$3(5izzl~;x5y=Cv$(uKK77@yF6QnZU&LOoPiWM0VVB8q
ze=B|q2Z4`~XyW8k1VI#uRYJ}1De!r&THCJ1+K5*YPEoXLV($IFeF=8<xc|mg-sugI
zo+@9Stcf@`u|9MGs^-TEVl|FL@`)2!`w4`IbY8{F?=~8Gs+whR^W=vc)Z~x4gPbou
zW(2TNI~~JR#`iempiZ4Q7XKB{<4Be}GvWmTt!LW|=LwHyaYA5V!YwWS@%z(9<g&}R
zPC@RvFLI2_&cN44aYrOJ1>`&>aoSPr80niv40o=1Hx(#vAT{l!ar;qb+#d}W`x*H`
zHcn%IKfpXa?Q|1<*XcdZxmVGxx~f?$KkS~8VjG5?H###%5y`y9w0Fbb8Ts<}5>-;<
z`k9s+^)qsX0)yIIJwik}tg7?3E>5mKMYR;Lxe)!|VY=he@jk)Uw1DcX2&C;Ky;lxu
z(OWv09z8*oF!mcw0@b8hb1GX1B%8yg%F^}0HfH4_Y%hm5BGR2XcL>O!b2xHj>l&mA
z4)5`PMoIEfYX*n$-t-&KIRS6PbILzw+_+*?*KjWBU~W|Lt~7}_&<e!vp0RaPxehCj
z(TLrsiSu{L*Mt;w)#&L$1HJ{LyryzlKyX5JcT5AmFvd=9Xv<0T_2ahQmP~L{YT4+j
z{xu@n1s>8-QD#|ORbSmG7iB&@P$D~_rS>}+hdAUW!0+F_6J)QHhmQNEHb(n;%Ul-v
zk<<nU(JVNk!@_?Rc)Xwqq2`c8u#t75iGb$DL%;@i3#CW+nrM+@hlJM;wGwjUNj@Wl
zlJumSCMFZ`as)hC0++WyU!b#iB#2GM;}`FDD-u%bFZ9}|_%H`OwwmuoUs@T<#nE^9
z(ZSAt>Al@ZTDVb1+tfI8;ad`|;73I0Hfde&OA`oqcQg9b6*%tpdrE!a?&F{D@ahTL
zYMY|25E_1<(<7APm-b^ZLbKGVMM8@2Xg1WtXxd^SwdL+TpZQ46Qk4$=1QN)R3H@&V
zh#cA;dx*)M9DoaBB0~#%zB(8{%8rbX+$u)}LU=kfz8~Cb#@V{UyHE`GEN_k!n%f#f
zF=IF0-`6L|lq!n|(S{c0OUdD^DA!s$OIT_6Q=uP3C0=2Ttg8LTc=RG$cEV9<7^lTR
zM=lkcbe{!duz!0!F!DBCl4XiIEAuHu?{Ru%b<%5DOr}MC8yUwD&`5*Pbfq{DHSs#*
zslr~jzVq))>=8m@AO=_}@n#@tu2;y-BhB0cX%rz<6^NF|PjoB`tc-csW4O!Zy98{G
zTU>Xxp^{OxE(WG*>Izbsv+80~O4y<`{Ed8KhmmND+DkmD`-(^JWv2W~1+@9R0dF%Y
z+48kZI)3xEf3uXEi>e?3@?gmYLt|i|`sCqCr4b~M7!f$vgs*Cdx0T0QAUa(+`f!bn
z{Pxgv0y+RvvFR14JZmCb-QVhfkfhu3^gh|bV}<gvLR~-W@iqNcJ~n#yOfYS8h{I=K
zc=7Q7E4lC1FRzanREHMqB$Xy-kRe0h=A%Kp3IFjCLSEX48-`X(<@G+rNr9tP8Aw~`
znwr<+Zb$J0XmwP><Uwh!%VlbH_qSLLJORg{31xT8p1?E`4w-cJqNcbZX|c)ZuQ^IK
z8V9BWK|7RSZl|O;2}HuLb`y#m3CED}d;nkTa)du^tN5^GLwCt7s5xN_s@bWijlNIs
zj=5n3$dR4B@DrMu$|bUllUulO@l!_sZZR00`jpVUcKOFoVGQ>9eVj+k)+|{l4umsB
zYf%{e9sAu3GuLW$GKgs6hdnT9Uai6~RO79cck@4noFlV7EH@{?%e&({!;PVh;LiH_
z#(2xr<cTd^UvW;BoxxB;f&Fr&AN@_EC1>728aZc+yc{gB(JaN`OG5I$(#(+}CmO{v
zhf#S;u4w*MvtQGzeu0$jE@T^y5@L$|C9XyNcep<r$H_BeBS~^sve-@Q#2zCo9gN(c
zye#(jbATfXmlMaYF1)deM=qLuY{N&9qIjk}x*t}rfB^&1<vuUftRpY+-s0-VLNQff
zzR+FT*(f|-42OspFLV766>I)%QQ0*4w?WV}nRTf5H>7vMHCf{6yF?~yx}(}K!kw9Z
ze-Rg^Fha8@86-Ad#Y)R$n)k|xH_rt1;6AkU8`?yH{TDmCzB-!@j_Y+iz|6~F8o@`p
zBP>)YPx8V}tlz)h{?imx=;m|Tj$#S>=z=`LY6^qA_o3e1eUHF2MvvXIB6-!FM%9VN
zUgDwBP_5yWrka3_l{-69O$GU&i3w`IF@fs|dNe2evI<WK*^HFC;rfDkBb(vF7rW->
zgo8Ji+LFoBw6sYO?#e*mqKj5jqVl5WD!1A3q-#pUk2ZgB=WZ3=ET3X2By}9<iqq@b
ze1)6(0Q9*CP`wpV*0Osow6cgBdB7@tRdTMrcd|X2cw^5qy)Zq#y*!r$_jm=Nb!kYt
zu?{I9Yh>TD>pDGydkl@?8(M$7DWBYGRNE|_`shq0Tt+(iKt;jHdmxV+OuA{h-L5-8
zv3u$AE9kjOh3%;>cx-6!PH{M6y${z+`K>a*!C{H^H1`L}rYUG&b+_dv^_R1UxZ8*N
zjK_^V@V2#l5DBWBS)aI;80=0;Frh}FB6~45Rw*{z3-xnaI5N(-{&PPP!)@?{J#xZj
zw5oHB!gu*cZ^LA|Lg<yk#np|OI_ZoK0=Lz`>`9kfW$5C2hG*7O%LKX&G9%LWHJmE>
z3JubMi~6#j^pZ0bX0STYKgI)BM#W}y*W4Yg#%SmSB4nC4+6l4{^Oq!zYo8Xaxsu=|
z$M$frk2w#MTzjQH$}OCWD~N(zX`29cj5zs$nsKx~Z#*nL;FI&laYXl60sl8{(&=~M
z7V+K<oD~$#G;xJ-$F8DDCn~U635g!8tv*{eWn<bZ)r8M{Od~K-NXV}|EkQMx8~A!<
z_XDmgdqMZV@Z5i-{xPz#v;23S8~*jM5w|0De^5OF5AXE=K;pu<gWXPvZo%M7>X7Dn
z*ZutHGX@HPr6nudz(?B_6id_Ak46#D{dAN^P5vyBo_^}-`hGI~ywmu7JG{Td-K6RH
zs-Tw$JId;9-yBuCT;G=(`zF)g<oR(GKi*|JzfnD$C`o4Vy_1!^C-{+5#!UC#tgsz?
z1;`nIa6^K}(in{^<^6EFur$=LO)tIlRJ=QUeSGB6`FY}xv7`xRNmnFhsdaT_P((ZA
zY?|Z6TLlE)bK6;6N;7h6NQ^vL`<U7?&j(M3ckacqKU6?J)EOl+zA_fgrmXob)Y}$b
zpwL+eYzF#rM1g?ljqE2L>tcq#u58+)F<21Wu#6=!h&@b<F18oh-|uwAHrEyPiu%N3
zn7bHuUe-<16>ZMtX(CXbDyJJa#9;1;n-Lehxn<NC=6o;SE*iRz^UUtE+aV%NmN=CD
zDTja(l_;HQp<5GOBW2RqVOAN4<g1n}7E{ykCK^^H_R(E{$QN>}XhRV%(Y>KgArWy3
zNCc=zEGOSE6wnFg3T(-QgaNUsjEMV~pY21NB>+qoE&tr5U}XJN8DTN%R68IgGT$()
z8yAG(U&G?Lky#%gKR!)0sBDe~o1kn+(o1vc%Mv1;^>Zzv0t>~C$5iGZC5lN>UgZxd
z3W$&PA#VH=5)p)LV?qijz%`7T7$|Z-PG&47I)mnSK5gk@X6rD7kvLr4pzg~Axel~1
zuGIiKd|y&i*6GI&Vs6z!|7gzHj-p8!lB{fOR)I_21CjyF*^(t45|X~!fZ;S{2F5&<
zR>W|^<HkuWXiy&xqJ_?usvh3Yw%JuBmGw^nV~7nQ;V?O2)|HaV1@7F?rm+Q}{UwKw
zbS;6}8P@vI)}DAO;-(TEwemnfdCo>El46Lb9jAu|Zr+4JooG&F-W_f{T1tt9{ZdoC
z!l9kwvf8Ynv`NQ4>Zv`7`p1GYx~2HM7AxHbmOPY(AwSj}{y6D2>H;TkRn%MV?zx31
ziH0nadvmweB`j@#2hIoOx}OR;6yDSGDi?=xIpuAy$d=MflrzHxmQP~dik9PW;@Z)#
zX(KvUj0orJXr&MEG%LLETdmL0<$gE+=p4mb2OS5YT4n$%kGLNW9rXc*=iBDA{;iV1
z$8*aIR6u+XSu=Z<Hg`5~jrEVbzadnQ!lgzLqD2sjghnYiz`ShU7gQvEvEoTj;O<AR
z4{CU7-YL!VvQ0tgX_Xa~AQZ883Bp9ggJ-mFABmu*jCcS)K{p|U8KGRlAgjJscvE6G
z;f%VNedNEJa8!t_bV$mRv7Tvd(A`W+pw%4Yi`uEF0g!*3GpbJ0x_~P#huNEz=Bpq=
zXrttp^M9+k1Z|4cZP@_WbbJTyna-=NAjD!yi1It*a`Pz{pRvYL*z{f*LYVc8wk3Q_
z@pc6qQ+O+!^eqpq{}L<WiYp79=LKmV1Rid#l%6+ntAKM5Mzz&Bj%6vSq^q9B*ge<m
zH0)>9&C6OSsEp1rFgiUo0sI}ijn=hylg4oyr9^4&5wuAAd}@6^+Sz~$au%$ZM~o0i
zN6@`wa26!0HTI|ex@-GW-R0|3i<$`SFRhmec@#n>;6Q|6D3Bm`@1zqLoana90w!0j
zK=jWl#XAF4y~P%7BU!lI3ue0wy-6XRpiDs@e;uQsB(rP1J+pH&zB9NbRWYr5$MuZ{
z3fr^7#rK)-_#o<~x$vFx-~L@mIUX-5)~m<NToz@lVz%F5?emDhXO18M{%2E|<G}j@
z-Z)vjOPZ>rFt%*=I;%C3bsAB8-#tX2x1jvg5_Y&X7{o{y11M?CzXS0(S(|>@?{n|Q
z^xRtglT>m(@D5ghNy-GF5&sNn8ak-e#8LBYfoFRi#1WW*Li>&d&^7kU*?tJ?!%~7)
z|0jytR&K6nk^E9H0%0R{EpxPbwE@pB;UQV~GA!~*ARur%0-`~y8sTA5J1niCUOSHC
z6j)P)X{o;@f9exf_NlQHmyhObC<ybhInj4u$NtS#zegWe-%co0LIHivYuPYq_M1Rj
z0Zz5<QC^U|A~xGR76&1ZpOQN$SQPkJMq(KPA+Uuq(9WU_h$31|e^)kBIyEF}GIFL1
z{haDFnmx72%ivzDQ@ewYVxF@OP@ne(5|vMBn4Jensx&$j=)_c!qch5kZB%%cQRgh^
zIPORq_FVCp!=B|O!pE{Og%8$qfmca)tG;ICIn`&Lra%}hvu3yl3>x#DswYjj{PW7C
zex;wBEA=v^!3!-m`OAg0S#`m$<62n9XY&r{%NNzm>TcEjWT?L_;mIfC&I?(#=|txy
z0IU0*=p5uF(b4aCmRq-j9Ul(wprI;fvAwW%j}`iPT0F5N`PV-uSl8n%F=i2!Avgy^
z8inyme}#x;R;*h1u)mr%3s?}vK!PtA4$yh(8z~g-cyUW=&jcNKk(cxQMDudcLW!0}
zCmuBzZacSpfmh6AyUFAaKra|V;iA0bye3Fw)@tfLP~K%C#Sl10D(n>sH3u`pNh#dc
zws|QsWBS(gfug!fJ5GA=U2^h^L+3capxq_oA<^WK<+3<pNfz#$PT(!s;1na{N?OS!
zD({lRaoA28Gea7)b#`(_5K5GUi7)S`+8Y{tzUKI3*`LuH*)$lJS!BIj6mrV*Bz%`h
zWj(GF2o5+eRU#A0PA{98)P0AnR>xb`o)-e9glK|Z1QfDBR}6OOxe;7~HS3?!Y~;D*
z#4TvE{TE|9lTW{Yc)d9Y8N||`xUj+%ik%nPuzb7T_1;=8$vZxS_KHXiD?2~LgBlPl
zT#$6Z*P7&(9U?$!y3&7Cq%Y8X7o`cxin)fY@O7(eeO6gMz13xFUn^)#-*2Hb{S|`I
zqOr07Yd1>)In+@bNj+Q{S>mnJd_!2kdmH*}Q>?8Hcfk+u^zhP!3q0T9I@s_9h+His
z;Dk3+OV@(120w96IqDXI8oh*7C}1JlT{kPew169w4?p1BS2NLRfge3UvhzFYo>z}#
zc+DCZFK5itaF?dU(zCx;?IH=jp*Oz0jsSmW{Bvzk&J~7iU_jijtdT6#2wz|wV@MF_
z3dnp;v)I-EbfY>4b8dhZB?e_-Zn&IK2Iy($lS+M{A*2H?!()D))rIxEwrFA77*_Vg
zk^@1;xL3#VL23(EI5H2%b4-Pn(B?$roXtE*$x6SQBl1K<vZKInAG+2e@2+sDMr&r>
zUiWCqRUgW(ZQgmEm{|Vl@}KSWlF*}+I&$!?ZSU^otdcUOm#{HxdTfqjc6@>>+<~hE
z0FzePj!yW#?}n+rkLVa%gevK9%Yz?;5f{~GuDVpo8RcT`bA20%rBwIubSD}B>zu5)
znib10zjP?sSXq;Lt7rCKeKIIClF(iiH3idozAF_#$0a`(M4|b@RI1QAyYy=AK8x#|
z!F2oO9xMRhAKZ+2+>9@62(E@SWMrJ4Oy3td`Z8~ez36V*2bkA!`MTBP_ElmcnwpXW
zvfW%Qd4@J+^JPCMYTH)AC+JqR>(I??JSq09kFQoYSeV=f;z>Y8kiokl??x<WJ2QQ1
z4Ub~$v_+n?F^*_Wn2prg>_e4l5;_ke!AO(QRV}KA9+hoaj=LwQC#kjCZ_vB-D$K?W
zc!-|9y7(n`<^sK&@Y)4GZggCkqEOM`Wh<Uw$%Rpi)&|wJ;FfhZl78QZ{Qz+F`}{9a
z54QhI9%uTW^7zuPN$wYudrs6I@ezd3cKr8*{HK80W%jk=gGe>J#_&vO-#-3=kCAC5
z(`e*AVKHI3G)kz}<LUVrTczGaX)u2?NP&D-rVZY>zfk%5a(#SnEZnd3ehizZxUfzv
zSVV{hJWaqwJKB11et!sfIeC80Eb!Hp^>e7y^()BLYDcarglKb~=Qp?b@4VjCz+jcK
z$sWb0`NmJL3wds8T;V@DxqEC1NK>d}IhIV-H}9o{eCEIOeGxVD@Lj{iEIlrQSTc3=
z2Kg7QKk(Dw)xxmwuo2Q#*)jgHXw~O4$@x+S?rI5lZ-D^7xs-0_<i*c=R5{RBFyX7X
zgkKLM#?<(@#B(#$z9Jr%BAJac@b%E_q5B<fr2QDE?=3YrshS+5?x1IN-`BHzy1lL3
z+n{MPPI93tBE7BUHgR2%k7E@DN)%S5P0N;L$mB?3RO;}tQfds7wXGMQH~@a$%1d*u
zFnWS+!)vY!y$&`luALR0HO7c6Ot8wxayaj%ur{BpXC}Jvt(3~1CD$~ceQAzF$cUt0
zg%I9@gjfj<&z=EVPLof*As_KfTR13S$p!*lQ;6?EAl?(_<H%T$c8ilbW;7Q^NHLCx
zUTbLzp-C5jl&tB;X(FlP9%9D$D_zMrULZS}S9@8wj4m6_{R#FcWMoNDR}2qs?=Bw~
zT#&|4Mx7o}@_TnzO4aJO$1)BFM_|CIR*he1*1y;RZu!nx8p$F|w;Md{sl$lsH#$gE
zzCV-GQagtHKy)G1p&-pCUWVz5j6%nQQAsMkv(V0>T}pDPlE4sH8U?fD=>kGxc+Q^I
z144Sh2anhTB6+Xk2CsLmm-WjIGFcRdgwdgZ{ub&_REX?zlxsw78gIVgFjs{^zWv<K
z&j_4P#$lzzo;(RW4vj5?#O+09>8C_yDl+|EZ2s0>e8jaxYx2$tIoO<}bR*LM<wS4}
zttx!BS`az4He#wUkQC_v;FTo)>&GX!E$(pQl~oz;-Udu~v*;KZuwHQr(URb{p**pm
z%MTz};*N?~5<;mTrn49{8(Xly8fU^j_oaS*@{Dp#RZ{FFuo{_6MNJrUMW0UckR)rl
z$QWrR9zBK=Z%>UHqTr-6f#cy7f$HM%Y<8xTptroL+Gr+^vS|}S)GUD|J`q;W-de_^
zFJ%Aaue{6-ZskZC%Dqf$l;Q^JhLKvr>98blGce+Jdc#&VTdUwX(?ug<0;RtEXVc>p
zC?r!d^FX8!SrA@)y=T%8DcrFYayf9win6)Y98G|GW*tSUzvFtJ<=8!}Iox4t`2<x-
zt#yT7R}1$M46+pMU|;32wXud6zUr8>BiQW<A55&OCV)*T>Z+LuPfB8X|1%Q7=dcbJ
zZn({Zo)(pACt9lXmOs;IY=CM9RH1}hJ8#xzdqJ{OF8&D!GCJATJr0UL*JB9)NQPYF
zQX2AIUW!1)^4TGc0&Q}6M%d(EP<|^I&b#4sqm|LchGEzT3;LZ-46-08^$-#&H5_Ls
ziH_83;|i_}j*Lp7u*3yhg{#SWL>#Uy&L>%~-k5L|OoyHdV^X>~w1tm`&UGiOrJ!8a
z%wTtHP3T_;_z+!Hp>Q-$f&vw^Ms;7X@S1-_O|VaE%})lAYF{0pk^>4(vA}}bvkAP2
z1nma7{Q*HdCRfeP^nQ+6+w9g9IdeKns)fDUs**BN4*RH>?vRKg?p5_p6<=bYU%LsK
z0_{s^E>24IuWg%8!!cSEKZbJ@HW>g`N^a6y-tBo09hrJ&oy4}LxmFhZrIIVDrmDGK
z@S!#b8u|)~lN?Ir(+8dR<WBUoqu;Tx({4*oESqX$(tgH64k>$&uw-b0OCeySG3IU6
z9!1TT(Q(U!*li6rf2Rw#S-9G+FTJo)^d_}cz?FfShR%HjY<9M6S}Q(Cj@MOP%Mc#p
z;jBywa%@U~P|_20!zzPfJnfRCNA!e*qKa4I0c_;i4myl{Wu!#_6-cF3Jc8^O9S9EI
zfdM=;aTnW^FADLoTN7=wD8W_sg?J3&g^(ks0gRRnW-@-*s@z6AyT%z<uNiWU4^Bh?
zY648hy!M2am#EOqCdu~TwqTu`qn4_P@~gBYX5A&7vl#d2RYn$$#v5XXq=gt1sir#C
z9E<W6#^09Y)5;0X)D7Z|7$voqT2+8`fYQs2(&q(}QkE*;TB1V;gwSTgqM9D}28@jP
z8X6fW?bsgO#p4$u`5A2F`oJU&uioZ#<)ANFw1)4$59d`C@Kzprr36{UM5u&Za0e8Z
zU+=d_jqxR{NMj4Gkbd9Ov*ail;+!;J*pv4&dgL*DzMTapFQk-WwMF*5`OL4w9!1Q!
zW*Oyzp+}<=j+#ANM4eyU{8ag_R6d;>_(x*ErpV?PYq^2I%Hu6W0+g6JTooKO=**_^
zY7yIzagGGpQ<Z}I_yV_*6Hm^fLokvSBv}F+;8{sR5~*IzqOI;Ht&Hy=M(XEq10iG5
z^p1=pDxnJcfyd!B)-9ji9a?3l{;+IR8b{W&r`wX>-)5)s-PdjO8F>JCt!Yr4?@3g&
zGoz~P_QsPw&}VB~X9aB1TKXFUS_*57-pMqCe{y=$DOZY$3O9U@%C5N2O|1Gj3&E@y
zUUNc0CSmGHQlIoUavw$Ge52K@QaNvu|MFG8RZ+t#h+z8(ns1Fz_j-qt6%qK5a((}u
zM{X#b$eBnlW0hW5=-!j{(9>REUUkY0aOn9SG&VB#NO5$}`7H|(ObVVw_hi*ke<`lH
zLZYhn3`>a4sQ?l+x#$R*y+$1ZAG&Af8ztDR=KhMe$!^P*>B)j2ierVGUJUV?l7EiL
z@(B2MB(Psqtl73f(TwCSMK@$)mhKq-4#@4A-HG<)@}az~3D8?@j+5$f+4(C(PuLB!
z4%|G)1lY4$c-b*+XmpVuY2mR8lUkKg_BQXeeOUMPdCj8HKR3gkZuKuN!sUAhUaxlN
zu%#Kb&*EG?EY7uhif9GgC_HMZ#2ej^oIr}?AmtQ4dMNwCiKU|?y4~~OpNWOIh=JT4
zaPvxQQ^`XPbEU+|!a=i+%PC$xJ4KE8*<nEEvH%eSzVOw2h0;g^srK}K9sRgqWxf5s
zV@S6DOjc%L{U5UO6qdyA81n8~?GQYfEJzCp;ap#z+uzRUaq6}}MI|!{p6Q1V2D2Z6
zb}YF<@;r2|Ix$h|5<8>*v;I`1<kx!`|6O_X<ac(q^7!WS(=B>&uV;3zSDW^ml*asg
zV50g_Z&~u}&2Cb(qK|U%RN<watL`5;nt3_u_<Y8k6s&}t!q?u8DQ^CGZTbAlbC|Pe
zjQlko%B-%)*hOIcm-BG@+evl&<I{tK!~RWz(uBaLEUxh^38K~R(b`|F1QL~qFQCBX
zJ7#%6h4M?XE|A|<)>UsF+??0XzIK0&pqCLQklZQb&(Hmi-H)d4+cJfGb$p1ixcwAy
zlt5Mn4nfVI1`I>eb?skcy-1pA%ootU?6+nH%g-f(SuQogz+|<-SXbFeOjA6mrC^G&
zMnZUCjH}R6sMQg;UEx}~%W`a;E#iUK67+fJDR-P1)CKK!LP&5Ulo>t>tiMJ~>%`c=
zc?<<8i0^={M{yDO-{>{k`Bsl)IHL%d6SZyzOaweLx*u&>-cpJ>+@P;SEv6~l+w)}&
zYisSL6lsJ}uf?47_Nl76ki1j5EJ?{gK!;eF=M+rv2(nYDj&+)MWkhJtfg{Z4q9@B7
zE!nx+tUET8ueI>F@=2H!R!fz>z0KFOscb8G7SSB4#wt|j2O2Ip-GYgC97FT&_TvO}
z2#kDplr3rKNvVJoq1h4pN2^Y(L6owP2HrR`%?6tJT<Uii^p?;?s6=8C+b+jw%eiza
zld^%x>QXgG<MJjx>y-3Q>3Vtnip)88(BW<YZH?+8He0~PC_^#%N!tD(shBp5&IWf{
zP{p!c<RHmEvXEIt3QN?N!AW9MxN9$$nX<K|63`mT96qf@4MqAUCIW0m{vs)2lU^=O
z(#Y4H@ivIV^dw&EX1H_y&QgXbo{mUe@{$uf+X_(j8>dz%^`*JU`xQY4eRtjB0xqhn
z=0AyKKJ03^jwZ=15nJIFUVgTg1?t*^VHxxGO^wC_N>x1GwV3Tn_%((v0)y6@W@KR&
zAb>||C+vZJAN-E`UkF~q%0g-iN6s2Hynk2D9w>;F^)SnBNXjC%Hj-xupz5fpiTi&u
zrMr%MW6z<(ra&_xa&~8=^PpSd8LCGL3lA&yFr~tzkfG<Gc^8ZhV%~|c*4#QMH1==k
zUckqrRMo_h&;A_#M_oF9ZBMysdLj9#aA3ad3MZA6D;EsHnR0}wnUAlmqf}rCw<6nj
zMWOI;Ht8Pk?3zE*)`NkIc9XJ&SG8XF>%hbb1*Bt=lMaPUk2tO3T(=65I~5f;{z-t*
z+mX%_l9kK)_(_a5Q&&7%8Rn8z@vZ<^U#yW-Z>1>%zbo%{V7O;x4KCM}ekwYNjo!;7
z)xW25ASBc@SCmyk(m;B1-fHrtD~^UW<f-aF(al|>rlmLsya+q5Hi$nJj4Qg&_o!v4
zzNq}hAFxu+{*KT0Z%?s=W?$o=i`iL8UTi6JK~j}xAZ>GEDh`z1(~XTPV!@IIdFWDd
zLd0<F{tJ_2E$p4J!d>Hn6vurY5eq2nJl_rMz7*Dy{G;W=X_-&skSZzxf`-Ff5^kVz
z5?%&JtXr>*KSK&Wp+se$8f<eorGj&hW!V0fkCou?%z@+3g-Tb#8TUDXLRMVbKScz$
z`C0-a7JbQxv*D>$3yntW^ig3ujug8&#p#{5Cy`_c0Z!yv^;uz{cy>*^l0<Lpry4Iz
z!)T9Twf2!>r&T`>7@xO0->?L-DmW#@(xXTmv{)x|kG_k1EE!SZxR>$=*UR3q(SRDt
zX0>$KOz-z`1Tm?!bIi3Egz~pOk$Le~dqa1cOlwnlC~n*54ArUM&IZ>TyZ#@_{wYeg
zV2joTGgsQSZRbkcwr$(CZQHhO+qP|^)_tn8@7b-|u9v^fr~hHbm~HeJ@kQ@gjQWxw
zlGM*qgj_RR#~XPKI|Ef+R|HanroP$YDnfbduP(sT&~G`AQI^v40!V=k#%!D^s%!?`
z=tN@_{?g0A3_ES{w~PyH;+YgjQ}|8!E}K?9LaOZq<!Z6y(;v?0>({WGd?ZU+q_pnG
z>*+bsk<-_Po_KNrXHxRsR}Q%hP#UydhH)>f{J0WYFW1Dl9n|Psusy$F&Eq}g%>+?D
zZihGL@Pl36#@($;SBaBZ7eFw0?TX5-GSMDvUlYY`QwHqCZx=D5kL!>}Nv^d?`N^TJ
zvc?@nXyAs<iV)=)c&qhEF)qGk9C^i7q{1AoQ3a_3X2{9SA4?$L(ZT?&(#91tx80)6
zXu{LO%kcY_A2qG+CI;Bd9duAJ1YUi|n;He>)Foa_K#4@qvrWOx^zzNgGpVQHSc?<+
zb+n8;)ub8udUsf8(Bp<UTa3mTCzm5HFVpPY?Io0ZkgCS5HaD@78V?p#^4OqX_j7{q
z9cUgmEA}a69cXG$QM7drXZq0>{5z`AtgSmWI{sP59;6eRf!m!k>*;fuC<ffNEMBs6
zIn+Uaji(u$oeSQ%K(@_SDPPaR8v@Q9#FtJn&(%f@7UX4bXCf?xmQ!x+T|J_h7O4wG
zYF6JS0^*EvNffL-dl8&f?^sIli3@a9=bHxf2%KL*x&bkIOTyN*0(YY37pZ>H5rkY%
zwu`m4dvpnI%eIK{+6tf&yz7-<FUQJrOd)gP)5?}bBM`O{ZBL_<$vmuQWjl)VsAxmG
z#{}J*CDEXLycF@7uV<}jmK6-nHPz!yct`Y?gL`~+&<FE{MQ(ywD+h_Y=Fj-u4$oCY
zg051c%!0H0fUl)QK)$<~%H;U3Ze$t={nVd+_YW{?zD1mst4Q?rn^oWJDDqoaYW79$
zw!Hao0Q_&y;+FrJN0rf!XarF;_1;fwvx!}UJkAG)Kn~)lSKfGJsxY2hQt$P*R0Bwl
zc+s$1(8JvrV>FtFHR7XRS!7#3t^zmfRyXU4T<uMU$*z!ULby2HV<BS~Kd^tUE$7`f
z+;;!Me~ONs5|Y!P1biv)5dMtau%S5|R9)M1kSHJx0)5-(UhD6x=C715OB{yK^f-Pg
zJU4z5Y<a940{%9UVrN>+w%H%w$)(O_-NE|IETiI*@+emV?ICizS~K|!`p7K#zI!+R
zHN!{4wPAN)Vi>e+r-`>2B{NIf?649W1v13yPYeB=PsnAz6j@WxY-Rea+Mio9r&|qG
zE>`3s(RLn6+IthDOWBTS@Sx2tC$^7FBV0AfsUUuefTeZyX~F5><s%TO1ss5VADk(a
zCYNskqqdIFRS>gX14J4hFLz2|jUsrHmucyGsP!Ih#NDF4_0(?4*E(e;vbd7Xj*{|J
z8(jQQTc^$5nQO;2GrES+0p0>+S&$_+%@d8YOS+Ed8m%mX=mh&o)S@~|*U5pubO@V7
z@ZP(qS7UXp#>S2Mfv8EQbA5Ubm%jMLqfPhedudv6I9GGs?C5{gT+#Mh4tg6>Il1@T
zO#>evnp@#2iqday-js6U`)5LPO=olk!uMZvB+LJVCo(eqhmPFVkc=hz3*Yld^#M1u
zcHD|Q4G5uf0Gfeyz-Pje#dh`c3q+ebS$k7Nme^Nk#+n|pS#x??@nHDgP$pdBYjGd<
zGkJVm`*f-J@zwrARZ#zC{nv-H`Rl{H@cxp9yO)|D{WJxL4}PX%SdFA863P#ChEA84
z?({fSQ%zOB`L@U5HJ{}9d*A*2r?jn%SxuywD;G)COKO2!Us)?h*Oyb`9NQ-c7S-y*
zSm1;2qeunjx|K)L9L0NYs=Hd+kmT9v+z{rOzV=BI=yfS1*GtNKYw>oLmyVamRVVI`
zoV@geE3Z0AajVlj3s%!vf>~uM;bQq1N@Wwnq&??pe;7X`ik^X0-1(_d#y}E`pS78Q
z)nX3t5DVTd%Ha929%7jiIia5rH;e|Gx;9SaZUp^0Vgp1_z(6~A^2Qd`^B|lD{AUJ|
zyPG87T(K_3B0XSA?i|wHM6Vt3M^7`o0yEZyQr)YB)4HGeflHwq@CR#e^80o63Gihw
zST8m<c(RE}Ka$c%70A?s5e3m92lr*DKBo}ij1NEy*vM*Kw!T7^Kfyu$n+l?k+{rgU
zFW^xs^ZPE~w(vu}f{d9dOT5ia<WuKD0wag9<Xc{s(SE+Wc(EU0@`v9zr$f(QCUv*5
zsb>YX*J+aygjXz#y+en92p;7GJo&q*tldNu+&~&R2|wsvFO@f0KdU+$#;ST$RVop;
zNn!Xm=d@nu?gV!jH<?(1kRdAJ$~<-$cVQROUv8YJMbQTp5qA;JmJGvC&$4e)>IT+&
z81AL$$gjx7KLvxFB}<GfuYw++kPZ=eQcj`+kS+MzQsMQk6^2CS%WRi5&4&b7BugNk
z=z8WK_UWVLc#+?q$f@y^kxE9=ccVGKMeHAHYe&z43j>%rr{uA$LWK*$HIVHG*o0Ju
z|1P7Kt_^HFe)3xiuMmtcd<DmSK4Q(uhAu40HlKLpT*TlMRDu@|O<Yhm8gMNkW0TjE
zHKTI&#=_KS?*dY8?a~*~%@Kj224fhraN%f$g>4|g;Z!G_8=a1iJ5miDK+`|auodbs
z%{WwsnlO(It3K;6`cnHmbuS9+XYZ<<UlKd`w6w|N+%fveDh=BA*dB680TMyAXDr>S
zbmkHEV*HUwiKnpN<Hu+e9ox#(NjLC^{aEAdk(-?S6J9%Q)#Tn|W_EE2&^4T(<Vr@`
zR}b;chut2#vNejr;Z4jB*BKy)W!+Ekd~51aNKX{gcAef6l?yLwXd7-N+zJIwd|3rw
zb{N!0(|g@7Z@K1IxROyF+fQmYS;O7b10d|dbmmv|X|5)8jw5wM%h86Ggs`JQOT+~P
zRPCHM1z6tVC)uvW*&077d3ki0VqF({+W`J78VO$#xnUED_`pc7*-D<}X&9Ya40tl=
zWAe+A<eWx|QK91lc_y5&Bif<)5G(u##TG+g2Ul~2umRM=zP+s6Z|z@P3Fq_iZ1K@f
zlyLsgB&vei#6izij03-sv*X-EFXSIhjRsKNWGflm3o%2zKrPN^rb2|1r5Rg!ngSEO
zC?IFpY<I51ppQ;iS*=8QUX|>L2vmNt8sh5a(fYYDFQ?3B;$&VHV3di|8glmLd&4dE
z#xS9l4IArfXqa+33nY;z=l)37JGQ3y;&90{2lj;p6`W3IIV&dTFlm)(43btYYCWl=
zKOr0i4qVG@Z}vB<Dl%<P+l;Gy9YZAugHMeWffA9IG*(s08G)}_>-7s?O(Ev9th0ks
zj;7)b$>tguN2In#4Wp-jp0`$7hEFS#8|ZOmdpD$Z;|(?PFsK3`qxp`tUrF+d;S!Pq
z0~mawe7F3e@L*@Eshw2?gc*dY@3ypfGpmON1f=F6WQKrQ5fQR@R)J~Z*%hI9dV!P-
z+m=qH^gov8sf}b3Bt!rSX87G^!t!DrjRhgi&d?aIdZ69utE^6q-1aLx%lIghquC_+
z>INmY(GiODiPcbA_6~$2givU()jN*IyUr_tpYA#$W}hfPL%NEU;m}P8y29!guq!a=
zZP)Pim~^OW^mIBG^X#b_G~8=v0;4-v4MC~~yfYvu$<ue$;1w~bOl3da5=#RSMTWIS
zL&00wHXyCHI&rM^6~pkY%oOtGMTZBaD@l7R_Qo7McOLaEml+vzwU70VX=Mt1{wh-i
zwzA*l6`<yof;;xd^OGCfI2uvn?_Q2Jp>Dqz#<_TojNGMQU>*9v3=e_AiN4rv?q09N
zFZs^?GT6NHj#&Q8YH~CU?&ml>lsCgAx7BX{1bZo-@Q{NH*cqI|MXlCRq;HKalPXmO
zd#TqH)w38Qzhwy@<7N@)lP^s~u&I`9Kh)i0wk{or3>B?RUM$@B(4r}M#rfQcmg=sQ
zU8okjXZm~#kxFo4K{7qLc?(y*YX3{d3y>r=<EA`voGgLhU`o<CbQ#RS(78b9=XH;q
z0Hq*a^N5rH4AUFHSuL~S4n9OeJnXv5kQH;QpC+k>c+D9~I`BLIisj5<p;#5XTl`bc
z_lyt#_w=Yvvqi(T?S27~O6l&*b1cyFn)<DqQ=4NgbFkwrgK?FTz8^dzZQZ?6w?FTy
zoj*sqdn_h-tyF2*+99AHqiLG19FeWJIK#A=Mq}Zw-{sBWQG2P*PZc=V?{c&HqhnfZ
zW}q0AW>$L}r{uW2P_j+9&SmDy1dY;$^4ZBy)>J@cjkF&c7<<|!ZP}p~>oEGs2@ay!
zqU^4{>(}HfYYJ1YLi6!CXrRo$sU&tDIsjF;B8Ha#{POXxkKZW5c5R}6(_2xRMV;ul
z)F6@rc0_Xt;>KJQE@08^`8~{a^&zQ2oP0-Lck;Zz>Kc8%^BFjwT7H0E+y8r<*{~Tu
z${9bfeGEl)8?wol_8KT*){zQ<20@fK3XDV2+&3Ej1}2=Nlq^Sz8Qjh}OhTc^ZKD!Y
zThg|P264FitW7GG&N4Xck2bP&@bsv``|&yH>I+1K)V}y%)DQdrq<)y#S^rO%Z&_p7
zhKLoh=ak|DWGZghm+wyBU)rzRn4uLv^t`^gl$-e0SFK@;P~B=GK@^iS5@XYXrSs>q
z(xAD8m$iBPN!A*7!MEM})3)Q|@Auc~o2roEqiyA*ZE1_|Z-?5~B{%2qRsj8swd`It
zxDS&G#Q6cJPB{F*%j%`ej*lnYC213K+^X|Br&fvf?_0z7SL8yb#O!!k3*G6o2cV9v
z>;$Cc!o=h)SvqmNg14k^agv8DZ`^85E_IsIYDwn?*YEwY?j~<<SEwU429+_QD5t$l
zPaG(u=%WHV8O4@m2W)t2JmQS`cIhn3wJa+^k|_X5Q#hpBM^dvNTE~RPJ6nyB&tS4q
zr+)jNM@ac*an#b==E!=&vN%E_NQ_G5K3}lq=3LPRvm?2rIXl)!y+_fRKi=oJF(~E2
zKVxF1uY)^+B$oQhFdF3g4$v^xDg9Jhe0+Uh6}hVgLPT6{v=N)P%k<;qQGXLdYCxJg
zH#uq9cm3=*bfg_V+lax`2V(4dQjXRu-xQI!YdcdLSbyVEr*#Sdcm<aaw3sJn{c1cf
z9qoMJ`L@hasu8c$$&}1`^yCaNm$-P#7f1V|BL^<+;CAd#qUUIX9q0ZiczdEp#kWP>
zY_p9By^<OiWDE0>=FR!M1^4x<>@d4Z8S8RB_`DcRI#WAV2<ePRoqR`Fg*la65gti;
zC|cna^qjWaYB}s2eK90?COlNkv%{qyi&~8>x$)S;kV}MH{#cQ(zLEl(^Sxywj1pdO
zQ%WdlEcWv>p&DZ!JMosot0&?wPY=Agv6^tIJGd?^ATV?1p*PncHHB^XNNkR)?J6bM
zP>Z_v+K4~4DF~g%G<3f_4X0Jwr><I#ceE0P_mIZF3;+v?D}jvt@SFG+rl=n)CauKU
zX0Y$$EwMO~!r5f8-^<!$lj-m2Pd^I1#a}x@fu!CM^_AkI8h+1{9lI95{bjOjYg^ri
z=#&h9;3u!Hahk^q;*l@DQ{@HlM&HK^=7}0FoDW6#wR+MY;JQ2GF@GnT11p;5<OTFX
zKm9EQ`WkhS))={<KnZ%iw^%je@trH<H*@=%$72M##S=lxi$SwS2E`RsEqB-f%SCr{
zO~2_3y&scw6hbB-&9poZB2>{yE~8t2`YLL!iPuswFUC1yxHWGk;M;vAZCR*EK>MmN
zvc*NbTb9{eo(iR;vx8)o3$rz=S55byul|hiX^Lpt$O*x=*P5TJ1pRIX%qFXv4%|5M
zFdI#^6)LoI-JxrHR8A<UGLfPr$~YwhFPmdJ!WucIv5{lQOvSCp<dg+Eyeol|7?Yd{
zZLv%#QzWQnnke|qi`5((G&b##Ao(UN-l9{W&cDmgHIl9#;duf-!j@y@#ee3xwK(G_
z&YdEgQZn3_+r$PFC5Br;DARh0*{sws`Xo1(8PHfHF6J1am!qxsnq2yloc)76+JcR_
zwGr^gkN`?K@Xnk(Ul5;TSy2Q|#&T61S`<PCMi;gNxW>_7pN9vhDqxyu;K6RP(m_}q
zio4fbNmcAn2}qfe`a!oTBXlwDzUIiPUO}~wZDQL7TL`yx>Wk_^zk4^9VJ-L41pE0K
z8-a7v=7KvS{ahyI81;QPK9sLI8Uj(jE-fr55YhOU6c2>+NsV~`r#iGHwx&QHAwX?!
zt*ro+KXC>D>&X*@9Imy9?@tG4WNI>VItzWD2xRrKCMe&zp+XARdKdt4r~wi=Jq7EN
zd(gk2+Wtb*e_nH(i+^ooHcR~OOLq`jG|a(i7@qGSJ6C8SK_KPH{0v6uyiWhveDU|_
zU{+KoWM{?-^xpjWmn<c>wjh*S+>q|y%cIGTB1f|{FL0zoyb@H%v0!bFzzI2ciO`DA
z%eD{$ri=cU+?EPo;}6E!Xi!50i8$iI!i8C2L&B{gCCso})SZC+lpa&^Ku@L%qAl=L
z_gZQM@>RDPJE<_s^}l}MxB8WzmR8(haFk#pLc{~MvyF_ZPU6L2*s%|J@S<090izEU
zCLGu&*@Z2pfE4^PmXA1CfxJh#Uh|H-@guM4w78?I3YZXGx?>%a4Ft*bOjd(n^EWtL
z8GY1J9&eCx`M$jXTlPvIIbRCeadK_FfN7=>rH-w4=}*h+nUuAc-^LEQu?>)9No8!}
zrkYta3?`F)9Ff%Ca{Lu#$|O$Cxb0jQhMHH_thi|Y1Z0#+k#bhvifW<qTc$b@ATNQc
zrRd$w8%y)0dwyfv1=E$GZRNw-K+m$t(npkb)tvk+jTJh8L}<T)lZzgR?-;_Ij~VJj
zBfI28se*22>jdR+emXvLEm9J`l&S^mUYbt!_r#-(gOPT0<4E1>z<R`40ka$-T8<s=
zo#7e3C+n!^rp<QijH>X=m>}eA|5+y#gP<-TK5}R`o|MQYi&?%4L5y=?9aMp78o%1;
z=Ap9xBiNGEEOg@Zy%noQ>#&m3uHpKb`o1&kl<Z2w%S44sLliG_rVnANJYR+)tK;x-
zu5Z09c0#q%WMDP24nOZ|10><{KPfVxtJ3Pd3ac57$!WE3VaVwVQA~ard(Tc$*@ZBV
zzA!jZd#Tt3o`3G46g8VfVe+W`o)5Vo$SJyK9~Kf^e8ur765^s#Kfy)dp?@EVC+OIP
zjkkC1Ma$FOc7)@>VKWB|SxrWl71vGq9oa0XU5vA~0mDR;5G6TmNr|li%&lU$s(Crk
z`S*`cmUv8<u?UeuWjrQfmdO!O>ENDfA<j;PPR%_JbGZ&(F&Rb2lXo?a$#;0Jh3av)
zs0;-WYoCTV+H6k?u)uo?ZJ=F}ACGV5*wU23d*%7SFs2U6n=)ialD|IIi+$?_i`D#x
z)0c)6Z0)AV@?}4(t(6+IBHqyHD4Hl*8>{hO;qd{l4M@YsJ)Ow8gqzOlu{{|mF=SDy
zi7*cF^dLprWMU|HV&4LkTve-&q(g>$!w8})cb&&lr^T$-U_O@KI7lf%qCp8lqJVj@
z@S=*N-j3A?hqwj2A}25kK-&`@xV?lVh)?4^g$vu8grt*sg@XseB_t9Wg|iYkii>1C
zouC(WAv%hh)`L(q{9O>}>tk_@8bsW;*S&rvqvWuf<+Y_oHI#+fRb_HhqEL#IR1-*w
zly=!W;8(`pi~hx(Ah^o_Zz`ESvVCCBd$;puJU;ST0Bkq?DBE2x4PV{HukY*9U%+2>
za*F?3qOtvNp274?|KS<DtvMaD$pYW=O7Zl^WITO84j9#wd=B)CW_Tw8_u$z^=Ks-X
z<i8<=#_P`-4)dtt2H}LKzC5tv#vd%{;-kDK4)DJh=g+fmPZT=3vp&8a={{e>(jt|4
zf$2Isht={Mi%tLv$R(+Zx3U_Ev1mh)xxa4je_`tr6{DowgXBXDeTVx~epYfddSjR=
zK+`&WG{CH-zvfF%1NCEw{Bz;nh-qk>*W64M7E@1w*zj(rG)ZE)-7S+0qbyK1?uvD6
zH4Zbj%#hKUz-2aSp>7~%fn^`^Ldm{^cPost8E>R8coab&CRhI&uTkMQieq>k`|@Fo
zO$=RnOGg6%ms0C>=^sbeVzVcRh^ZbOCrYvpmA>|XWX}*Dtv#OqU=(0)@JA0TUO6W1
z*iCUCD0--o3~6XgSz_ko3@9QD$y-YsmC<KW6wRL%`6BX!b1JqI@?pjm<AA50etA-A
zkOUN})HF(X3^rEhMbS`CqPwBll>ZGMu~P>z+mh@x$k7Q{Q&HH@qZfL`P-^K1{FJAD
z`chKHripu!uGJ<}y)T7M@OdbSyZqL&!vFqcqxeoxPP`Ha>)RP%6mg>aephlLrtKBf
z%2uHJ{5MHcx|CCvh@yc_3kInYQUhl`RT?F0kt93ZJDQ{<IW$_<3o(F=Hgs5wNENc4
zV22!C;I*I7acG93dnk9h4X&VRRSQa3vJNxC!T16PM9Y|4b4D61jvEK8iR|XoSsp1s
zjDm!w+$N8QzL5vp87&J>cL<}1kNQ-BP#J(f42Ksz8Q1pMuY}!%6Y9`c(jVDcPnc3j
z-D*f9u&7)!zacAFH*7YeyYWOx|3tP)>6iH>Fo4`&Oh-6YW<*r6Xgs?7pgu;4qL4kv
zN&=}dr=ZA<<2g<1fjQV~x_~}U=@2>sOwZ-vd-1#K1m<I5&9}K^AL<6ck1SUBnURIz
zM4z!D_&zb~2#!%_2*6i{bkUSLNHw&{Zis&_#%N>=1yza$7>uaP{w&HEkAPCbHl53U
z<XFfLpp~AY6dD$jAeS+b=S<>1QNE#_bzK^imgzqNAE3>$%!Lx^K{Y3l7v|&1CVCtD
z2jyI*lpkxJO|lB%JPx!lE-gS)Fm_CD*$O92?W^Ss-xq`ywKm@$EC<M|(JCkupy(5f
zXeopNK}xDpv1tw1TE|z&LYDjZpgvJ@Yre<j{<|D2jlP&85)(;T-}KUG=R+h6CNRA~
z9hp;VVuM^Ba~$C-t#RSL+Ww=LGZH7>lt=y7$0xm@KT%V;F73yd>HEdJ`2KSh?&*ug
zT^L(6V8x;8eOky~7_~S{j5C<$#XWPUbQW)aTcuKi-(qy3)$>bdwaz50ppBR$(_jqr
z*k%-jVdmOu&|vzcFIvlgwv6({-8-kb=C$$(V8OcRRL=jJw6tAshtlCmPP`>&*hk%`
z_a>xp&fc6;Vx_1e{MM}!FiIB_y&wgi#Z9`A5f}G@g>Rzr+!43^tC|OqJSn{>h4a-M
zg|B!szOWB3v>G#CU4&o|I<H_I73L)Dp-Kz9vr^A1X=guyU-z%=10QhnA26WzY@kS^
z7o4=CaP!Ek6=gH8qO16qh(N`yw&?mmj;eMS-FJlTj6j?#<k$EZfd{yF?!mLMzVm1X
zBDN2)cC0wsv_9Ae$thkXpYV5-cm2{d-=Z&?liA~4m|KDxVlTT`8Uhj0&SiL__3sl(
zKe>{}`2IG67KR@TLj;*rwJYRbg_JC&^wniY<s_A_$?73*1L!Qm1F+GxKBHcAtT7(K
z5s~)V8_Oea|3Hw9HkXJt9c(114EQ&K{?WD~x=kzdc^d*8ApGrGK|4#1w(H=Ek&Co>
zHaD<+`|a&28;aEU$B9mx0&DG>4h|BSr$rmYev^HiWa+?ox9$9OayR<(A{TK_=kP}k
zS_}z}wP&ZdvIrx~=0P420B6dAO%kksd^nwN>NURQ620kEOBKWX(C|tjeP*kxNgW=9
z?l6_pHm~|i{;@~WC;HzYMCvTsePtQi$r%Tj*vrphO~FzKlqyC*7ntl6cc5!z9#fEU
zc-kh>Hux2DJEI<Mt+CYw7h*4(+*?Srbr&HVTkx5I;g-R6hw*XE<Uh4JH8)Kp5de@1
zi^1j#Ay>m6PUeCP$#(C3As7CUnT1CeBdXplniC}J7#PayxMX}<jF!Yw8iZe7DP}#H
zw^duq)tL96UGMYjEuIU<MF29LR(k{^5Fs;Y^Vy`Zn4<CZ7pCtsm_{gE@c{z6GppMs
z09N^2@F_&4AI*I$jQJ{FQ2Vc-kf11YwSR=!;i^Sn@Bkt-f^kmaAU6n7ZUU`g71(2P
z1{ZF?SYuG!e|!Bh`IEW#P4K_S+u|3StrlAoJ`hRP$(l9d9gDjC$jwl`NhXv5<zTCx
z5PelN_U_Sz?Fme{ZTrjWRCW^WC-S!8J~R4&_Py(8Hx+1Uf_B?WNTu9GEB`*<pgeT|
z3)K@<8z5!tC+^h;-8@_=^BC4y4^5O`W|K>pgKZkR!rH@-BtAbV$1_S&?P}z7)Mm#p
zMeMYQwsw!LC){izl&>#h1cU$bd&ZFem;u{zjvk4-+x`@j@655dEvuYw@@u`!S#jJx
zf9ksxLaC^$;p(55u`iQzTsP<8xlgOLhQ1$PqDS(ZpSm{odT|=jkc0l)GUDViRDYdf
zVLLw*UmSKB-)Jni>bxiK&hmbNT(tw#|KHky{eOed8Cm|rskK^tGKQ!VX8ZJCer1(d
zQ4Zf=B)~XGLnr@j0JYZkb?3yv%r6~bLO~H_>gJ0S{h>?p&cWeboX(Hdmcb=&D{<NH
z^~3${$BX9om;K{+>+5mtZ{M)GqLuyRr?dUc8+;1V-{6KW-=EKEjW#u`tY#l>!4lW1
zSvlU^6~Zi8g5(WMN$*Yx*`gIIXRDI;(?qG7$fA$yLK&$!()&Lg@~K!@Z_YX)oo3f)
zE(wBoH+?isdJiidu`QXYB%>H*V0G4Vt_Ls)DD6@D&S6ha)rVx_#tEAr(+mkwrS^3b
z;2EPFZ(P5p-?_@x#wwZBn1V=M7StN;(Bc#=%4gg@;}mYr9eQQCYbsml0#l^?6;P7_
z1dn;Wy>!lDa8IoG-_bVjtF~a#ygzRGN4b(LJR`?R{Xq$6cv<DGI$cZzeGiA0?X<nU
z4UVmjmVO)ZBQmwT8PA>5)qrRA?91PU(Uen!9;W#f_>#(~zzNXhv##e{y^kXP>zJwf
zz4ufaPls_9bXFtpE?nLJAemRgbf{bp25kr)TUt{ClPLZ7*3&^t{%+NwwoX+)T@<t<
zGy0G14lpurd+nzNE9enr^d#f;Q+4VdbwSF;Jh7z3O_eaawLhOJYApr4xmtrn7bM3v
zD1#N6AZ%u|<w*c>_yvNa&N;;+hgk6UW8-}S-r;YRf0o|jDZSKcJ;Ym4pDL()kt!|(
zXCl1NXEAo4JWi1apXTz>N>s88!Z_Fr9Z~P&_KcLQbQn+eSBWP|pkY{$`;26D;+OLt
zz2ctNWjklh>D=wLhTG0KLFFfayJtgfp+Tz@)3?~^MypJScLWo0nUNMBORE^N1<vij
zZpVcNl{LnM>bWhoUiLO_u#7ZDH>h)aJi8`tA~YRL*h>vCgKclg3qY|?jY_9FP{kLk
z2O*ZG+==JfubQwnL7=NA)lII)-4=3o47Xz15OhMdmeb9Z;~^|(5p?!xCGa+ZEWeW0
zLQFjS7@ymywt3)A_WadDEQv>0EUh$Nn<yiQpJ62&mVlNpyW>jf+GRVe+9eP1t$BUf
zi;SM%54|e-c2VZZ+PE7rL-CbM*>_UDQ``^iaIHJ-Tz7k*5XwLSh`&Z+kZ1j+^O2fF
zwntE$X`sy+iT_R@gAvomoHPzfVctiJ7DtP2V6IEv2tguCXVemN(vGXer7LO8_REY`
zV8Yg1%H&|oL1;GEW}IboHK@3OwD__axCJ3BCw`2_#nH_*I|>rDw2@Cd1qb4tV|7S#
zym*<Z{oO|{#zUk{K_IuoGEanPVZ=r2&%@2v5zX@UYOMTxfqjqxJ9uU5Y44sUOGNJ3
z-Fa-u2A<Z?-SpHXYkF2(pNULmsmTU9t{`(NUcl>hv?N}&QW+R%4<la>rw}I~GpxuH
z^EHpQw>U~aBGFYmrE%m413Z%=q&-)lZ_m9b9T1}ht-SAn7dg64rXTQW^Vigj*;`pZ
z-E+zCh0`(|Xbse><c6+s&9|P}GpbA*b^bSA|96=LHG1Z=GDdQ{*yQyP%qyB3`<<Mq
zpC|hzirg`Uf(8zu=Ws%QfhN+4IYpYq><b<%e*>ey7ZVGp4XvI*?C1x=ikfse$Z14G
zC^_7KsMFjF<-MXGt4jSE!fKkuJEM+n4b;O6!lw@ArwC+25QGmj&*Gj`V9oSbo=56Z
zJ5fL9$|BiE*x`Z*ny@WCvj<#{`5d&*Em@hTkck(RzHYeAt785W(HF4SLd1x2jePq<
zZfZk8vgWj?m4J$n7+N}kq;I9sJn3oRC~&Gl&jD`K9qa<BePj#)^NYf)ptBz&yHIZx
zLbCbpjRgU>)pX`!^a^XWKWF^`;-KiF$!2+VsPj<{*j5757{A0em*%+XK=?@d##*zu
zE#GF~Hc~H~-pNcU5_}o(I{`pb8nDdgj^+l!(7rP3G_1;8T7i692yRoW+>S*{J(T4c
zSNu^VhpuC5MOIH)iDM}72>Pu(LnV06`LMDKl{SOJ!lKZ?oa@o~6K<TX*1WC6Ld#N3
zl)exE|Gb9Oi8xE=yum|9Mu8t4LJ&lT8=I*Ta)XlSFGii<u$<5)HcVoSvRh7~6Gr8!
z{M}75oVdL}T7q!Q-(#Pua!hjAP{_Jug94FAlWKQ0=9=uICvd@@X#LswMXx$Ts5&!%
zkso5NDQ$dxAtaMcYO9dt+Wv-d-OM>%)?g(f;eImd{zy~2w1Km2lQAu`QBwa&dR+yh
z%Wtz2LN!mE;R>D5`7N{qBa0__#lsx1Co9?c50}D>mCM&aM{(HfjPnl)n&@uR5f~h*
zHce+9m8QTF3!(V9NA5&H^s<Iy16h&`g0v^!##_Ur#DXKBqQ?F;F5#%!_MYMRlFc@g
zq$!%NwRWR?Uh_kebR|CZIF;VpB1(9>20>a8;_G|PPtYf<&ytk6yh+)RRzRuK7Zt4%
z+!h0;1DXm?Q1-cRWYXhn`4omxO+K&8-LIoRQ(P;B>^~zP&CfarL%gcM<>9i%PJ?XU
zO64ZcPsN&RbY_&iYwY5&nAi>KPC4D~ML;f>9TLhdA>U(vlYoi;E(zHMrGCefoB{^E
z&!IH@<r%OOpL6ncfw16qg0Kfuiumc^gm=GmzCS9Wor#n=IA30Chc_v;AfFXr{fyVc
zrm&ZIl35Q4S6v9T((y1^+A%f&>JNoAF~i(=1bo#G#X&jEAhd$($gZUCq=-F{=f1nN
zo=@M;HMQN#)npNOzU%C}xMub1h3L?j_DV}|?QzOVfYP_HA(Z4nd*d+egz#!DhoHOE
zpcH)7qkqR<#?{+3=N;rSFdj}~)B@%tH?tvtrR5EgmX8QR*!LtaC$dWc>*XGQ!U%a=
zmF!gAe!Q~>hV+l9Da58u?g1lEkgUf~<J;owt272fvry6A&kN8q9%x5<egYdJr0w(h
zy?6W@lwRmyg8>F>r_E2>zpLhAxayC?oU-{RRSORlI<Bo6>a(&tARXYZz<Y;KkdYVN
z1h~<z=E-aC7u+^p%KiVXju=?~C-#Sl<v$#3wHKr<1Nq=V_d0tGVkt=8u1oL*3Yx*C
zF)ajUM{pDgY8o;Gf4rS{NkSnZV>>wy*$hjCRmNNlUgV#h#gQyED3GJSwU>uzove$G
zJ*1kRjf!%eRbf)HZz}iYoQJM*7e{~WyuP<mld-i;yDx1Gym^johJgRBxqCnlaUBv;
z;T;NT6J+KL*AJ(CbA#cMJ?m9gpV2)vbbTrWC{fi~Im<|Xmvq4mMsp+}IaU$R?`wH!
zY(rs&R*))>Q>oW8=|V(4zDrOi9PD<C+Fva!^(Jeybe?x@+Y_>?AHBzcUE||T<nv9w
z=D|la1%Mx@F)_-s85LMH{Xsg05lAuvv(7`_AQS{mPZPm-szK?HgCz)z{TSej<kW}<
z%HIIzWUq*h;}&Xu1iSc4baYsjO{5$4(H3rC^pLO7^Y<6z*Tg3Fd~l)6xFcs!c{e#@
zvszRRZeih3Wx-J*)JIL`{1Xm#qMi4@t=9i!Jo1<IKYT~DlXN3D_>hNpA1O}kCDo-E
zf02pvM*QkJt)0;}M$@8L>PYpY;Cwzw9bsve<ld&U(>-UhFSD&(-fOHu89vYd!MdRK
z78@I#madP<NTRMb8|31`+e$zbUn;2$u_l?T4Zc)#GE+Xk?wh|J?>`S6x3tEYL$jW)
zJ>Drl?}y!DQDxf@ox8W)NEGg=csHy4+Ai1SNR!6ILsncXT426tvXvd29kz=baJrnk
zO{Vux_3S~(gVndZ+r!O)VTIICt0Zx%&dr@K#>ZakIoeH8qq3wa!*An+t1G|5*Vv;m
zr5Cp6rPt#gTezc?<A1FGV8~?<rUVch5F?G%@S`RLp7=8xQf(3PkC}>%k*Vf!&9$MM
zMLI$%?O}_NhMB`VVIn$Bu#5TUNeMJF;+L@ym!)(W0~+vn)Zu&J%ac9IlZg%61zkl{
zpy5}L@;3{RAS+=dh(JFp-RX(l<;gLDz=a4cPsuD3E9qGCmvs^J#|JUOz*PYB2NLuL
z3iX<~p~C?OfJp{~B0JCD#W4aYm5$ZqFFwdDDHdKrH>n{unLQ==qe>oh9!5hqNOm>p
zySUL`sdP2<cWNS1lDm^7iJORm{QVft3rGoqv@pp7GirU}D*!uddg3zxk!PCp#kMfX
z-zX5I?nG&Bp?Y~PY`re(Qj=Zu|2BE7<<?sFsSf%PA7B<cByN+1gb=8b%fmL}i==mA
z(=NGZnw}nAtFqaN6{Ln<rN=E5Uun8w@cpdGl?nbA5P2%j`rnF>?SDBi|I7IQDnjYl
zOnUf`+b>AG)(Vw-uoJ-uLO@Uo<wcduZAn|;Eb4_3{7~$NTT2u27(R))&nVNcnEGyt
z7Yo~tc&nea9`=j)EY3v@@5|ek`J*DbYzTBR>(!{nfZr1hxc(BhrNP>dit&@T!)x1r
z*>mc#+3Zi%EhloZxwW<*4!;%Gm!qD;gbNo;CtbXt(r;C-7jJFRA04cxIaaNJ1X!sn
z+ajO%bn2VAPp#qKEj1RVv`5E<Y=78yLiZ(gn=v3@TsX-`=Zp@iw?aDRTUgL=tjaxI
zj5OO#(aaWW4x$V;&bPc)P(w;>*owK<{U_v#%OI85DT6Ecl=CV1A^vruB64{OIYdK-
z5#kYpsN!(NtVYDCnVL7>SV`F^eng}@L}Uky4}G&C_CCZk;sxcxf-pPm%ml#<u4Kr{
zHHu{La!7*sqMRWy2U$ILV&fbov4P+q?C>Wc{Cme>#PD3IsCc6sVzDH$y0`>+Sqb8C
z0i&Ry)p)hgcUoc>z)M!7H}ZtXU?Jr2yC8ylZ{anxST2ZoH(rE~@FkDo1G8;&#9ACu
z9!Ucgh3fAjt821jOJ4J~hs1~r^T$#G?~SkYFnh}Z*C_$e3O~9OVG?p!rekcBnS@ye
z$iI(j{)l#FN-a(WTSl^)J{;wX!i`%@)7u)ZJNqmFnMNE>+4GaZf1d`EKze=ud9X_S
zx03pw7z}2H|Lj7ElNN*Y@WQV@kTi-CcakZH;33R#^vu@bi^8cfT;edzR>-FDy|vYx
zPV)XRqJ`~`dZb&(n*%I?;JjL<5|~59_05HvuC5_N={3<X2SQ8H7-?U-P0DydnS#r`
z$~571YdAWSwWq6t^_EmvqK^h;KA23uT-Syq79D%{=ktkTX_40NZ7@ryFAL$zl<<UH
zF&nM@+(TB{uBILtH#^$WHT9|tpIb>kC0(!IaLq<w2q5k^#|B2G9T_&>#$(Da68)5z
zva~9n_eX&1AGz8hcAUCXuE!;(_mgj25k^^_8*U1D#QrpZ#dQqT`jY%|@=|i*t9jL>
z_X~@#K8Q`Bc;wOglYC({Y0P5&Y#pHXPNXYZZV6TG!1bg^lnG;slxsSCQ1A^Y*tSpy
zZc=3`H+*F({)|GZp70082x1-!Qemlecot)GmMHrqNX$%68u^6NO~CoNknSZ{p;OEs
z<?2Y^OH~1|7BkQ|OJ|WfAN~j9=^~aI;l0uHuBRh_&z^^(z8MQ63%eH2+Pv=Xw`jkq
z+eH6mecAq}Aq?|>rsZfSZCh0GAqU@jN3>h9BRl(Of(uw&@NfJB1iS$XIm_pw7#Yyq
zKQF4J=_rZs&Zl%NukABao=aZ7i5GhH&0Lwo1#dg~PWKTt5Is3wc*cv*K3ohUey__h
z5}USI4lNxC<5zxtRA*;<by_z2;oMzbb7N9xNw>OtmD^EpaO&Q!ITGTgo=sf2Wt}R%
z_2Da(357UpFJ~3oA~Ky>ug+cCU&)-1J$#W_Z;xqr*0o3O6J?Ib;z@)NCXek_VR5oY
zON!^-h)61}dLZj7PAbL5TzPY_BCS6KKj(vt-F0Rr8>dc$izpL3BZwQt9AFgGDEL9C
zB;!E11F+42I%199)`7xuxeH4E^hX+W!z`%-&Q!$|5s&%pNo@x*gF`~{kUUVZHKGEW
zMpX8t8-Tc+@ZA;6p$+Tgm9|UHEnZy#2KA^Jt;(&673JHct7V5Tx08Mp<k##EGji+W
zJxRrwqJZllH()&t8{^+VF`~aEX@SfTBSUA28`X(j-l3|Ivt{KTJMJZGA?Z-}f}GU|
zOOG~<@YDocMWDoO??q8JaFQCa%Gq5(n_jMA%*@)m6ghvj5LO?~nA+>PF{aF`i~M*V
zY)XPZ0*u@WU|$<UddlwR#@M5$YTv(t=*UBG{I^PB_+Od}BP;v=31ex;Z(0QXKh5P|
zrEDpn!g<a?A)r>{U%O0p@bF2AV5ud-KM%CliW|{!#PzY3PJP`!R+Itf{tIhNSYhKI
z1zBX9P*)5>^`z~h|1)yl6yJT$mvnlnTF^7@x97VvaZ>2@+p%-ZC)Chu$jckP!y3!T
z1vEHA!iyRVZ>@#$%qYXARFD`~g6r3ta&Nx?bww-H@@C@2IxXioKUgzvd0B2+oWquv
zhJV*E)dIEcq{N7gl3x~nHZK{t%qTXw%m@eai6*P1yg9F1ApJt1(HOYrH35x-B-&oA
zcvj}*)PxL{GUH!37JrF|hm8`WHR~2R?2>bB%#<$wL_(h3tn=E~pm{u%ptW~tnxs@R
zk;5Av7(D&oUhe;Nj{5)d<IrB<w++Nc2KXoT0}NA(5vr2^GLTlz3{`1ouHV?~ucs^F
zo56j*BITUx6K8ncx%CCo*G5P<6G6nPL}H38xJYQAKW#(eqqVtE>LJ8uqnp0>o?#&c
zz4tKaO2ZLv{b0MC&aQ6z)F$slN71%^kfxq0jw2H053el_0)(5W3V)-EoKY=XB$NOW
z?7t>?pkD`4?-cj+MP72ML9{5UHn5#G-<IEwY~_LdMZ`2pmmRfp6p%!hBI;wgfaI4b
zO^I*H(m?e~_#IgXE!Ebz-&8}Jd7#Kz2r`zLM{O2JE?cui$zS>RtDMIcLsG^L^%`;N
zj<tkEDLNgWMnWrxr(+pHDc^=$S{A+9q+J?xpcXXn%$7sYFBW4eW(N{uy5~szTAX2d
z5OH$z;l6xebFJj@i~m*J@!<-YV>@DYX!3pwCu&&e^xxKl`Ttw&ip~a3?smp>g5p9V
zwl+>m|DNdu|9>UqveW;^HxR7~rG#XPnuE>=lyQ&(kH*Nxq;hFk>8G~qd|9X_TuoOY
zCx3*1{|ADY01z0$b^uQRoOGW;2zd??P?&OWbEBiWqvibmY~`?Kd1J+fyQAe3?#F6Y
z2HfKOJm)$qaeRvRwb!-Bv-cAn_hjGy-i_<F`I2RHOt~@8lt#5^s8eOf1ZXXsPplq!
zWPvLMV^HRxu?N?zBF<OyjG+lB8?x^f*li31i6N@hvg^|8%$QB*4bt@(hdAp%qYg@U
z7#uei{^P^<bJ2IJBXyX{pt$m|*r8b}8wC*^`|ATzSR+LtC9TLrARzE`$|)O-aS|xK
zF0IWdSS6X-(BSA^N)5k}HqKo%|5(L%$|*Q2JehR3h?qVCm{be1->9re0wu9b$*8E@
z1ekGOoDro^nLH8WMwdc5ex$7EUMhr3r^0AST;b%}l{J)!3dP}XqY5fna%Is}l{$1`
zf<!uTGl#XjTpR}xgWNxyN8;bCm!HChKsnBGw<`H^7b3UW<vC)`l_eCkf@xVQKwx}G
za+#q4S$U6KmZ!77jR+t!syhTkpJ0v4il+tL6G=0sdPqfIe+|<Lrv+g%DA<s*UX2<e
zQNLV`qhx4FjRo(p{S`E7zlN)hdyUs!(>-;x3RiLR^_ceo{Phf3F2Fg!)*i+u8ysJ0
z$Ti22oA(*f9mQ=MQ<Fu8zD5ikG}f)rB^!ySW8l&sY0m(?&-56`U`?tIo_P@-14X?;
zP1qb#V!|n*pdva%pkhsrk|B|l$aRj)Lw5aFAg6M-*<Xe*K`t9t8mhbQ%4XDgve`#N
zk~q`#RZiQjwYVfv1dulz$TXlKjal>+E{>%YgWUFo(DXeA$;8Y(RS#h6h;awlH4+S9
zD~N?UYAZ*VuV3rnyWm5O5>R$(kz`H~E=<T$2hADUmJK)t^K7hVF)t^G5nxc(o@HWq
zvQO#=p^!_ivQQ64zmmPFH`WU{AZ204s^*5Y%R6!Pn6dFYX-#9Jv^GD=L(-=l_vloS
zeO<mDYJ7ZezO+@!*Vn7v*|j2YWu20b#Jqe>(>#x=f(A`0p<6h4eL2Znvw|c^i0O~u
zJY>_^jKZ77<!DGurITiMzh3;;pETK2EcA!ap8~w~1we++Jq&#FpB!RL^Y9&<_!DGA
zVIbt69T&fy;M@5EtWi&E?`|gnYag&s=b=E>q4hkznqn8K<qs|g%D0mb=R2GyGu!Fa
zHFmG%ta)`)eZR$1dgN5WAGkQWjF}3(X!MBHU4AC=c!=(w5i4dS8n-(1nVX+Erz1$~
z){JPD@b;?it_%(>4I6TN?VY)K1gsS_lEi{$u{hV;MMg~AzE?c@tUQEWYE)~!`!@39
zg!=Oj)uU8rtW<RN!#Z`KSue0wuKn%M#kZ^C{`&(9*17yt5rNlePN&jsJ3W#q0$A3x
zSm*0{JDV~8_-?kPBnXa|K)?;G9m|zEq%6Av#gnIYpzl-r8^g6!F!uwVg5oWFCdGm<
z%68DBILUZ=tnY9gDYFDV0|T(S#WNlPN=%mJ;7FDBXnV<hptfm1>=>aqQBgizA9(D}
zFCRsI6C@G?$Z3)8@G8I*4_FD_2xkGhF`yYIGOgRE^WYvk$|q%RR2+S{My8=;5*iPv
zWd%sehOwkEuI3~j?B-?%GK|0zfF}AG%=PX_8A*V9l7y)W?P+r*nfsNhzX?el)wyHk
zPjIy<uZej5(*Qkx0_TbS#5}Na(x9x`oy?70w)w-?<@C!6gjyFrdH+lw*|UN51*0)>
zKvg*jlFdEOeMmI!h^b^&OfgGjQ^L|k!okp7^7!t!JvGioE!bU*!xQKP+r?pz`ptg8
z1%4wx4GDK?jyfW-TcNenZ#xC(Tlh@LKf<8ndqmP4GXB65aJ<)^<*1fg>z0odPv}$q
zY`UMmm~z}N59crep*jNxZ)}D0Jd9!891FAxLIua&v4fQogg|YRv;{NWKb9$p@FEfM
zWA2bavNWC!e{K;X+#HP&AC*k$#4wCjV#3+m?ygi}KrcSj@YC^VTA=~evG#@tOzM+D
z2PJ~ARL-l>n6CtASrFwULoi!lO7kNEe(h8%F=0v+Zr&gj^t9B}^q|V{`Myk=?J@de
z_gP+J$A-kWU1<mWc%k{Bm2*nWg9F@_LJS`8yzakQ+i_8a9ttDKj`pmv;HgqR*RL}_
z2(mIGK=22P{~Wvd6TRf0P(K8(et9$`@}sr`p#6<ipkr)4D0tCFva0s4yJ7Qxqru7}
zyLj1LU;D!`7K1vsDdtGrZDGz11~dis4I|k6W+$Z~s$(4|DeJPJoI`aaxy8E+68Jr+
z?+Wr=eWzWCRDSC^%QO;~&Y-sn%OPAXO*?T<&3jzpZw1^iT5EpyTzr!Vq%w_)2IF&D
zh(_%t)2}>B?l$nuQhVjbWDWQ1EBz}A+VHJa+<SV2bJ9fCS>cT{1wRN9d7JlccAqOa
zr^Wm`<G_YB^D_Z~CP&^!>H`GK1P-L@>khQ_$vV`RMWW?;0VadW#_HuHeF=1IS8xWJ
zc>&Lf=obTN+&ld0|6*oc&O;YFu|Xr9R`WU~L6<=05`d|(<Wsh&v@HX-Oil387Lj0y
z;BvS^yJzPwI>~Goy|^49;@q7GU?Wt?Y9&98SYD$QdBr|G69d{46UJqayOcn0UUYsw
zY!C2G=+9B<VT)UA0Z<|WK}|@#LiFi5J_^drlF@iEj=*=<CvBWkJp3($L~lxwa!N!U
zLD_rd;{0Hy8VipVf&lm9@{}l1F)s%Q>#%K&aJo~=d+N*`<cRm64l0AFD|S6qQF?$a
zuFSBUv!Bv-QJ9cJv@BZ7k%R$O?loDPY4)9>p-o~OEy+@p$QcKmH-5XWsHyB^`Y>Mr
z=bUp*QVLZ-B7ZM@Kc8tyD<of%z-Y*jUQ0(#Tl(DYp7}tczN+ju!Ltj5Ikq@giYRkR
zo-;$x^&ABD+4jTKW+C_1^_%5fNJ-q%?8gFW!wMOKU8sR@iLa-Wo>x7*?7nv1k*Pv<
z<Z+JI+|_CKC-?aR4eR;jz@(yFNLQIJB`FmRZ85l`DGFU7dC@L7q{DOQaR$+0^Lhfg
zbsoe*Ahl1v575d9sbZj%!$oKF6R?a|@NXI|%MBJT3T-`8b63379vs<#Sbh{12Es%5
zvUujshL)CRN`Ut)Zo8!!TECW*_reGJn<#HGkHy}D&&ENGj_f3uY53-1TSZxaGbDVr
zs4QlI0}-SPijd=TX%UK2*$rV-_P(o9>EH43<u(&FGuV#eteQwlsg>K4A;{UX8xvnf
zG@t#|Ofs<aBIsFZAO`>mN=NfHy|Z}yk&Th&N-8%;th2^>-Ug#>cE?QPzFogf78MGt
zL0q`*qtgU)!fe{~LlqmWPcb=lC@H=J?lh(?CO4Wu%eO%LD+hZ`byP8e!iR~LlZ8mR
zTP`;hJv!Qr-s~xPAZJ?bIHc$l5`x;{x&77n+QyLt4K76>WaB4U!JdHYzhM3U62)$7
zYdB4cUw2*Xkl79C-;}LC0D_l<+(|X0(@V#C54uq`LxUsT*2C;wpjf{)hL?+zg*_>l
z;cWbC2a^9p!YE}ByU>i~yBw&y<CCS*Qb2feYjUF*_)M>J1ZG5yyc+pM5gL}a$A#49
zE0Y6TqadRtryxr(Ugf1+496;=$F7}o&b=~Z5YDY^u<3gHt+I4>Ozj2D7s6_Ba9zHA
zIjpqr=Qy7b^k7QAFIO>9D&OXM@efzu5S&ifLbKl;1f6#Dq8EFb&61;j>jdxAzzy=&
z71>d~I)_d>dY@O{jn_4DV&+v%oAs5~<L*P|qrt~G#smW%&fHX1Frr5#hfo&slY5yp
z+PiLb&l@54Jn8CQmzMrF<^De%n=t+dQ(~#7w+9cegsQZW&|kMF7V-xG6A>8vzy10D
zF<|}wcU!>9#=!J{ZjFmIz}%2VTXYB{Ei(Y=2<s$Gq#@uvZ2jjv8T35tZ@b4q3^i#R
zCd3?Wm0Psbur|>%cRa(9@QL$*!NVaal7fP!BJR6)HP`vv=(e_8cN@RPe_TX4PrpjP
zPoso_l2Hf{!>v?nG|-Uu5ZUy<(wuR19DGydCYOi5_<p?CfbRVE)Ddh~gYk$@z%=*u
zxraVI)S72UKfxrJ>FSTX1K^9F5O+GGE(Qm21A~7<lwZ*;+Ubi%!GfI81caQ(JdnP_
zU&BX-c!d<?V|iIKPH~Se>Zi7e=(otYJPy|G56S4jA!A^l?;l`WGueE;isn;ZF<c%;
zH)ve#AE6?kA0O)-A)Hz@c~)$e)jF5?;|G-SIuZE&U21T5Xn%5C_;Xw?xmh;H2e||W
zuet;RkNWlTVwyE?uaQj9Fpw`y$-ZEo?_TX79_?Ba!8$yHHiA7p*J#%F>UM1G<@DON
zCGc2d`{?0184Rzk2DrGcx&{XWO=GaoXkRh0kT3o}l$}#_X3@5-V_OyTk8RsV#Yu&K
zY}-yMw(X>1+fK!{ZS&N9JG-_0a8GOVea*Sv#^}9&oqhVJCOXG-kCXBW*TnHKU9*0d
z@8q}%JXW=<CL0GI&Bin<)xz{9*~ZX_p!J1*o{y8Ei;}s2lpx6m)1Z(M6LT=O1OlZ5
z3R#5mpt%tOa|*Vk95Dr2V))pgsu3=8cAD@cK!8+$;t0hl!eh|e2(&IvQ;d!x1w}UE
za1hRjye?8xzKUcC?HmT$h`%miMar6@8C4;ebyuVn?J5|0SK1w2AeeDi@&z4Kj2Kl6
z0-Z#R3ROW2MOlokE-{?MD1vKNXx6S^Tih=kavxUkDY$Fsx4KwSI7YEc-K7e74)qpY
zw-Nh>WtdvdpbS`X@WRix;4G0Po#&>n_Vc1YX6=`k-39NfS1q0h*ZC*6vWj6jo$@hL
z*Ig3NJge(9&&WRb?;nO8gU!QFIENDaXS^+Tw%b_g*%z)oA)k}i|3HCPxl27Ry;-0Z
z4+cEBrvt$((vx=JzqZV_1fdf&b(W&K>{3~&449ZTd0U-fSOH$6HmYXE3c>hm{kOcN
z;-;y-x@;CinaZUGbv;_7W3scrK_+HP6KlAr%{9Nst&3)>AXk6E*GkZMOIS%;SavmB
z=T#nlFPJV$VI@zr*!UAw&*DfVd;h-PTsAh{ATOMaA0}h9x{*~g^bv?P>ez0MWjDL1
zg*KP+2d<By5s~|3>p(LRg3+dYnUI<@16-l?F|3t?MVO3@77SZsQ@X8gx;$qy6Bh>y
z6I+>0Bmy0M7q-$AZqFqj(sOBBbw-sMgxZ{Q^!Blc5)4+NFg;uT#%p`nyrYg{^)`Id
zafTzmslC{k398K#NP!rgU+1yewrydiMc{b%rl5i9WR`o@s9wm~UVPs;E;Ew_nDusM
z-VsA>I}T%uwE8pKF&$Cw0Z$BHm~)w)AwSmjggq@Vh=Q2>eDjjvB*6)rKS%erPP(m!
zmMk<8ZlI-TtLM$|p~_m4+Fx6J&m@|`wiD?$*o>)qeTvrEaa`ys#H^&o&nzZZimMR)
zR17;;0;lof&7v-&mQ0@Gz6f2?7$v?~#C%>bWX*W(Vq-v0a3jKK0AB}9lLyt-@5mJO
zy9t2KJd=qZ2xnjG=m6u=b6^hcyO87#(e7?XKt3Nq7Y**f$_k4)RY^Rp7MKfn$d26n
z=_K5dfoQlSq|T|O=CKTXN#880w;Uu99<|6<ZsaRBmvl;XFf0!f?bN-2QBV^P^x3%u
zTfzontW~|R?s=rlGrrY=4RWWRd|~my@r<S+*wD}+=CVrlymNW^sj~vwZz2JQdoaXa
z+EHdE3INRDFcfr+sum@s6*}6SVROo6%@>6cS9D8)RsLUCMU>j_L_o~gR514iZ8{6-
z%j7Z($pCG77UIu|oLO^P52UNr1sKaL4d5#?EUeP(>&!ZJbc~kX5)|Zu`@A~<2ZZw+
zU1+PGmP8)TYGVqT0iobl#St_qEF|#c<fc}f<{YD1rnm@mmFZn;{r2x8NHwuva7gkZ
zv0zWdnPY{EL?}aMlQ2HVg%`JHE{w}Mr9}D)@jB<-;mOGR&HML;_*xqOVU`C1CC_~p
z@)kWutZ-eubsB1^|2PDFuV})BsR3#OuH9BpOd-1Yd`^_q0tX>7t%>oYQubdA5mB5d
zv{BNue5G;&Z?qK_#krlf*A*(L^?ho0G}mEUbDA7~^+F3pl&C7mxFssNvMc36>BeKi
zwFjw5MUhRDRam9$GG%IZ1M8hTcSY0vqEB<U4A+)J;F3wT2LE}5I@I7S$x!@aa<O=a
zVCMV-^A#%+dMoGEyE8<+gfsnFJR*e2E<DDA$iq$ucJ;9hklhe7Ai&TtAOO>logEXC
z9b~)yZGs(M%!zC!^ap9<B$b2$&uVh5Dd+XcPF!5UPH5oA!v%g`E@tDJMMslL`C`S%
z0h^lk)#Z9P1{e#XqbVw-$WGs^8P!$#N+)d7*bL8P9<2LxhiGjS#*_$1w2;-x?q@C6
z_?8tdOpI=eD>iEt>)cBhP?AX?QLmjw-#u%yhr(*yx}EF@V*>{%8>=AZ-mVVwhSs8(
z?<LX~zm1A|pK9V;LM6lw690)34=kIOlc{xm2e}U6r=kS$X>yR4rThg;LPiu&eO9S(
z1-7ij1~kUeY<RZP(GDALu1a8Cj`NIh0Uttlb*ufSoX-S<WX#kmEc<L@v`2GX5Q9B3
z7n}TymjT|IrAVP1RN{(tm?JP>g^z>HTOem*2ejr+;#Bm`A!b{3DL7KJ_$S5}Ssz5w
z2hHYak-6NW=Ok7=g#}82>>6VyXw?OxfxZ-PHbox76lY|`b^hRzhF)TkqVxh)BAY?k
zGelaiwZr-*_B0fJKs6#;MKUg8?Zs*6NGozPA*ZT)sUJ~<6wnDJLUzT(+#i8bYBucj
zS>7w{cs0$TWw54mZDRsp$*t~+P#zocN?@K+CG0R$2q~641gu#x`}EkKMX!D+j4&HP
z&gHhVZrrLf;T?^zLtwvb)WhQ%GKAniSuXA&8&iD#0)d~csdu?rzvu$J@{z1JZn0EE
z^|?(<Ip_Ritqn51W>W9r#5t8#zcbh6)TGgkns>skqe9@qWzghr4~tui9KA`&V@ycK
z;gx4rKSzZPzvh=REYUYQil*6nCny6iV?Z6nsSW$=8t_C^+`)5w*zoap&U=C%@1D3%
zC5+oystgN!ePhkGB_v}81$}Y!yU#AsrPt@A#$J{>JO2yI+9;zZ><lX9p1j<p&`^&*
zsu6iKw?QAzQ9JOOiV3n<wj7yPepNJ5A_Jd^3d3>QWAqjcNxnPR0)wP-Npd?fesx23
z=|ro=VRj)p2!z7yb$ggoS`t#1CP@##F7-qZdgep-LwI`rfUWjK7I0{<0hIT3I~WBO
z1(jm}xve;9RpRMns;?-?YXo!Fgm>{^IU`&%61(2ic>t90^qYq_@;|W^*qr?YSQWlK
z9dYInuq^ltv&vaJ#oRuDnWIZ<rBchDCs^SuoUc85oSroTQ_f_B*yWUi21l#oWpOIt
zcbmxyt$8znO2Se<y(9s>Mc7zG<b{SRzM)%jIdE`JxQaegSxlGV(?EFs$15FkJ9DBV
zu#EveQCFt><!rv5#NwmrMO=h6{E|PTs>*fq2YBb%ok7LpZ;EX^D#X}JJzu}6q6h=i
z&0|vG#rF?fn6JZ3M_YG4IO;J<MgSG`TA2Fg&+#vX31e7fQ{bpyDx&wWEB%dG!Ds(o
z<0H#(0Mw)xx9qMQ9M+{iR-zm6n~C2nZH|~Sq!J&FUy-19$b>`GtMa3NoyofaYQ(L;
z{frQvxLtaS;p-{B>L=P(>>ZHp+b(+5?m;Y-U~*^AiKI08aS=$dFKxaV&6|AuTzl8)
zW6LdvHA!bM>s?RHeu*BHvC@QJ&Ffa%S&QFy_u{C-b}CEN`4)+ZQNC07idtXrqWvsI
zSzHO9S6eVWG(>DR%%JyCmpD6s(Kmb<`Se)wdC%7>WzaMKcQsn?#+kofc+;2Hw^HEg
z<;YU0%Yj~ZDws#~7t47nZr|cud2MGI1_V2MrED6}AHA+s?0IbM=_?C!DO%IAXACBA
z+(w~%8Rfxw-upfH23eMIP`z+it=c6!4XSWOIAlj-iUG4jwe~eIblG3Z2c{_+*T&r;
z0|QzM#a$iN1O4<wNIKph2l-vm&<H!751{pdKUWm$JtIWDc$^MkL4A*Xb+KRThLpS^
zSM&b1!P0P*uXzzZ81Nm51aY403%d@I#tcv@ZEQ2VHe^X5KI+4Cx?HC!{rcOz1%jD}
z-nx@58_YrqaqO=_4&l-k$@fQv^F=#9)EX|NqlwOum`G^42NYxIE#;3%3HPekP_6bu
zRFdZhM=Gr+dM6rqeZA5}F`RpxIPKW>pWHS_ZnFU%f8%X!vN~bpxKD0nBpR$jTp%NF
zSk!U?#|cpl+(|D7v*%1KJc(JkJ_&HN4UvZa>@V7aTkc}MFS*D1jf_u|5f)6{iL1^`
z^{H!^t;lB!fbbg&;A!|iMZ~H@jUE_iI-zPIPsd16-mo%V&`+r)lh1>@9lkt7Twb$m
z{#>j2{rJIHUvV=ILX@A6z<|8kRob`);IfO^fQes(0+Msyp=Q~z)@uh;`FaT1?D9{#
zrj#X=lQSgd^#wiJ5$E=D*){v{a@$4?NC+XSGXxxWYFeH|Y-}bY3Kt90H7Q-8pAV|o
zZ@IiT>)YwRvp}D)<ohB|)sSS~@0tW8WQ{Nk_g}x%BFS$M5_}tL(LSQzZIJF`0tjtp
zcTH1<5L}Cl_mVKFcbOO=Fp5%+G-i`$i-m<}<4H27A+kU3=8~^L7nk`~&T44VWxvF>
z@EQCYNk|`sK%$Iikj<$xQqw)|UY6TI@yn5JgZh3i^O)ZlevbL9J=}=#P>_&N<Lsz=
ziv<4p?!k8frLJvZFvnp?ETS|W8kz^aVw#(aUesx4#@kxZk`|$xek*vKSSnFJBh00i
zI;Re!<kytTCr7ImAm0=!)$57NhGihKn@v5ff~*7IH$d5MlBtQp9Aua^rQJIjU2%_E
z395|_My4!(b9-+X^>6RIJw852o=?aO(#9`1Ez7Q;n(*l-E-#<8v<xOrg%Kvi8B_FD
zGf)nScnj{?6b=$l8iY$SCSetoiM|H9bd2#ceN!rpNSC8p*4)k{JN)YdqB$G>21E?a
zA(>$z03%c!XcxojwB1GrzY~l8bnCd*p_nlWdYfo=SgI;Tqjh8TVDykZ?ySzCltn9;
zIw-wE;`inzV}}@_Pw%YSaRp=`gsz8a=d3UKw`()uz#(hYc?ZRajr%#r4;B}bo3e6;
zPG+eF5Ttp@IA1Oo+f8o1qzjQ|2!xzTfR;1zT$C|&k8}0gZ3iqU-<S7D^Ahr|+8G}o
za<SaE@YcW?`B9eF+~x_k*{t@+2`$XJmF9i_%BTlkO}CaEnW@5Cbg*`JR5)k^)xJ8l
z@~7G4ithdJ&IfQRS4J7#jC`@r1*%jx`_pTjH!3@OjrD;1373<dwfLF*aq=|sx!>Tt
zhCo4NnpFFtTn|U+i^P+7ufwIx{>{VM6L5I;jh2$(S6m+=&Cb`Z@4E^?ln-0dMX{a|
zW`eAI9k$Sr2nO9vD=H!9%i-|3WG-p8bB2Azk~-|u{f?sb%i5M(1L;EwpA$H;mi-SY
z5ZD(mnL)}>B}9lGchdckbj~S7k#rI;8cXmj#j$C203c4<Oe46d7sT6yi6tmj>>DtN
z-qBp#>hwHP%8oTgmDNIT^DU=!(+sceoa5UB?lLO;rI&k&N8sL977=a3XY4b+0bbDD
z#MHU-Y5z_sR5nL*gb=+qX~pglwArAN^9#FEX25+~?XsJV3*nmR7sjCQ^_E1gd(R26
zcpX>T-et5`+%N-R!QRrm+hCkUdVbxB5u47ln&{aFZNPGkXM?pcO)Hp$1+LD3-xz^!
zLMW=?gM?KerCu5%U8M4A)F)5dS3__19ll6hMG74T_MqI)zd86GK)5K7P2`f%<_X|e
zX|lt<F@FJ1Jv2Q6ZGKob^7w+?Gyq5Jb_GngTL@gL3P;v-A{d;SAOG%x4>%p%WW5bo
z57C+^N_dBDL%$5^<4w3xu86T^it(e|KC-%+_gpHSq(h>=82OL$>eO;jXFx#PG~Ix8
z#)}q4e}FJ70pIs%XgVbyec5MmUNLm0Tpbc51J&=?5nFTeosHJJwz(3LL8ZT(TeN^!
zIEhR3yOo?lIlisO1s3-kd@`9q!qzwo$#Yb{i{1!pxV4?K+8i9vQeapRdS_7?-NYkv
za?lIkxwK6E-?GI)o`suz&FOJP3tzovy%*pgK|R`(?DAMup-lRN$<wsKIrbTDHTj}*
zvV**GhSB`Je!183WlAfu13MoLFv$4YAaQ^#7-h$v0Ms{zcj{aKlT5Nqd1yw0`O{c<
zgFD(&-Yvdj0PNFnGcg09<w7HiT5`_k4Xao-^oVk*E}&b{Z4!?s43YAi#|kBKnO02F
zb&?0JCh`@oQ~8M}DmOWiaG!k9;ZLfdQSIYSd@Iq_y*qY&_flui@aP`79SN=dH!n-i
z39$!uc=SE|*7T(QCPicYX3ggNp2EaIbl^ZR{70}TW%P+y=|u0bk4+hx(~Voba~gW7
zf5ez}t<ot0!ML?N&zC42o*U)c_)_~LIXI!lM!S2|7mn=p{w|2$TZ)cyfE|A2ZaaCh
z{`Cp8KCe0t=~zG3$<M!6kAYvZr-w7utKdo_T^WkBZmcblSLzL#(~1f7NSP6mAaoLK
zbB<|NY#=_ah;$}|q_mh{Lba2<%Q)BK=TXo}1d2C4z@{(u>HDq}o)NHxp@i)fx5Jhm
zj%Z_P%fb1S<{Fj<rAlZ^8=yA~fok)FruBGr(}8+EiS}SV_l7vqR25cs+ubQ<<6UH$
zM*d>G*~FCsEo3VrrIy-`?-{<&<-(B82^m?=RT@$fPuZm5oq;xNl&dqF@R+uZH_MG4
zWjU<9P0lEcEvLoIL*l(1WCM4_{_V1>I{CH2pQlC!__3$Q;vcqJO!W2&YqQY?m(-7#
zHbE}U9IzGXTxs}IdzS|aX<3Q+^T$9CbE#@xSnpO^JsX5G&tup#(q3>32_IO&vccvn
z7)O3I-{6dxHD&$PG5^imw%(Q{{(18GH4-z6E$;U2`d84ifA!D$$j%YmBHL>uw~Pc}
zrRE0S_)3*%Xlj<{mdF~dQGFbQvIL#*K@|tqHd0M@_nKB`p~^0trl4J`>o!kpH4WQh
z0enpy?7PWDpS%G1`BNDge?8>KA@U^#Xe7%CY%#*O2d9Y}F}yy(RxR4Wm3~Phfwc|&
zP&F0n!ZBLqx_lM=3mRr=#1O62d&lZ|R@**4s5a@4VW$Ehn5j4OJ7_it0Wv`#bVHxR
zUE{9v5p|q)&i76)`Cy*QXG5?(V<9gMreN#G*7G|ZJKU5P_CuWI51FKfxFU+L-$Sx|
zoOV$Vm_<9;U3f7XjZ{9r4N~!(A8J-i=r*z!o%>Eloi~24YKDs1Q`hUgfXik}?B)83
zeXhdtta7xPl>VqiUin9PszPX0$^#Z65AY5>RIiyK!z1^p;Zr>YXSQ@_2<Y66;e7Ek
z+u0!{9Rf_LjnrUaE&csIj=iO2@Z|y;NokoW1=)GmL1tnZk7ef|qJ;9RJm465c))B9
z#1JWxRzq3jeV{XrZ!=AlQ_>L<Bn;d9F*6eMmBJ*Lhmw^Nr&7xENVBl|=b(p?plnlM
zduwh`oK8GOAbO+m5$>NgSAJ1XGAcm6WzlH%=P5&WyUTQwe^Sex6iOux<|oHD_xwU+
zCs`Bg65Gz(s&B|YAM*eu%#q!F76)-KhY1VQT<JehDtDFe-wdO%;+A&U-od55dKC2@
zcsx7T;4x!dhCC8z#CLf=Opp#976|OGWTaj(in+KN*AX4>-K=n0ww-sQ%*J82j`_sb
zH&#W|u&m;4Cb$BVY+LGtDtT@67U?v#2}j4PosuM|Xj-BOu}DY<*wk|br~`XR?S(vB
z9LB=jclLxRkK4qLyH?*Yrdv~OVHgk*?xS3AzXw<VA{^U(-&P|She!LF=C{{9_RS-{
zes{bdIi6#7Lidm~IcRYgNwpI^)E+oVtnTFxLr$=`vArQb|C$4`?V;BV#yTC)31Wmr
zm3?fTOq`3|5}+&>cM|V@MjxkLKmX0@Q_1yiG+BC61F0G5MhgM6>^ILzQn`n0raFF+
z-PoPx@RL|~x<WL*C>h?|W|H{$Js2eE%jU^@Wh!zGvVDBrQIZ2746drXQ4$mQ0F{pN
z&&M#mESyEGM<To+j!ZGUK}EXstBls)62{_{=!1!0@a6mX+u4?u)oC;KVNo$a39<<8
zBdRi$?Pe$ZD25daFdMive@F*kOttKVNXVe23KcFMy<@2SkTim!)}4_^D|IYDyXUCm
zt`!xJ?#bi`?o)|Mp=>lE*>z6PE>{NYEZ_&@z444-Au17RKcZs;%_qCprlX3ao;XvK
z$&(a_a&78K5;G|(zfU9e(e}lS>VL7j!>hCum`6mZFmr_vR9K0P#k{xpP@3@F4Op``
ztv>{fUdy_n;!!SCq3(s|Lf6UHj$rv)uVaEvhvhmz+5}jUFGWUa@0lo<`R2ny9FjuI
z{G&K#e5cY-AChv&?@U|SY3^59yFS53;>Eu!(gro%{9=OL^jcj1a|p!G24vo1RI#bJ
zTe(ps8>s96?4Zq=9FcA;&v;<-eCF0%&EG?disVRHTzF_OnW0T(XXv+n=i>{}kZOdo
z3Ej=MQuIyF*h`-%xV$u#F6fJk8#y><y@NnCyRk4g^?LK`ph^OP{^~xkzWutOkQk)^
zQy~50kB*sn95IEIEj0$-<Y-!TxQeK4>`Yu8oRP%EL}(U`;;(XO`Do#+hE9ih@DFYP
z0pYs0pjj3%VLG404T_`u(_ZF#woEmF6tK0hYz3*ro8v*JaY-p<74;oASb~!PA@q0}
z`06W+*s~`QLE5yRkPp89=j43PSXa*{Oq^ud!4-siT~|-C{FSfoS2;Ug!6W-FUCC_c
zrq?rJbs2syJKGQZX13dmIFP<RFpe`9hd}qFU+EOA452a7;j+qsY7J)Jq|%e2Z_e#`
z#FeVhWqiUyBJ>d}f99d#Cu&X`pTbs4HJdjHEt_lbYF*cH<%5N~9+R{FB)pFGtChDo
ziaP3n+EkTyyl}r*wgP^}|0pa%M|Xp?0xiC>eyYr9%-F07nQFt1x)vq79$Zb?C`I~O
zq~^BSomS!zu9S{|9;j8{;gg7<(MimZ(ctJ$_})#Dl*wyB0V7)%&$j}c9|}ilFfB|a
z*=yntIhLmni?p*;F!8An{@7XBbGdN5Uv0nyp+7j5V$dL5#{5?zi)+634p)t^tv2cB
ze?d>lxQRxVRm&dKA&oln90xqqluJ#8C<N^$QQM&wQGR9bDnfNJj8l!YTKAGc3sFrG
z88L;K%lA@|)#l82#}k)-Rw|hx;x5Fi=cRV*>;VzHPE$VVU-^rSb3Xn?uaN$U%FpQ{
za*xmC^c_7~I9XmZ?#t7=q$l6ul9`i3n+cyvrS3o5u>{G*a%CkaDA^DRe>3ax24{^K
z$#bT6XOa?8ZWT?a+;7U;%S{{N=vnGCnP`eZB1(&)CnDU)@tnZ!e(j*Se~>Q&8XbUS
z(Kl&s-35zIm-6*|4_*fv@qNK>OZwN(saeRe6|bIRJqBOkWx4?pU!N>WS#lV1R_2$v
z8mZDq-=I5g;tI6%=!lz+0Jk;)WZ!u?Q~}{4jyR)seClCKP4w$drgU_gvj(z4f;l@b
zHth<ebde&_MMq-?2}vhLt7VcfbEBP!uC#~8?$!Od&XK(KaWgIMLRd$Vt>N&_pUk?G
zZOgMq&S#2*FqS=CE#0ptcNiizRu(sTBTWGP-)L$DP?7arTJy(?^^$wcM`~d;yHJ#H
zORL&i$A<s|<TK|H=6$Q`TAKoZMCGi=8~tM0Q_oLZx+?}Mh_7ssYmB#ODH??>k^~Lf
z#~1p5FKJI|a++#|8l(G9-TXkD-{MH3U!pgF?hoB`NO<bS_pllyA5g02Dw;ISDT)ds
zG?9~&$zS3!@PCEUB~TC4Qo2cgnp$g&N#kwhLNm^UO3fZiuq~T&d%4D`T~*yh{+3sw
z2>^-55|Q_%`IUM5dMHcW(%>R5Ew3z>8`QC?!F5%`_@@HG+_>laLo}tU$Ad(Kme<Yo
zkZ6=4n84ZB$NN`CZgvimvu78Qh)-4IldN8?BWC#Zi8=l83iorj$R-`{_c0OLzp>8o
zxb1cP`+k}ug|YDQD4lR%ixS2t^6=b5()v)sR~Qv;Qkqjnn6vUxn7_TBaI?);qXW;y
za{`?8!E<1UkY8>p)Y+%K&p=96@pAUQ2V~H~yz##%-2Z=2hX3S^|C7(-n6~skU%$Kk
z(6>MR9-~3g?U{BUR6y|1;Mo4#NB)=F`Tr`TbFi}ghjyQ=1?&D(t=mq@&NGkPz!(K3
z5iat?_=gq*Fi~EpIq<?J|Gm3NKC{_k?c&wUCec}cE~TVV$O=A5rPiMcY%jdN-Jg|s
z?@0{wFJkH6=V^L|d(*7z%rSY}E5|E8ecxlkoZOxsqSz6<ZWa-&)J__578-sJwO~5e
z{`Ru8$-!{vt1U-YjCTwOY0U^C+QT{$@Ricn3wytM`tWU>gKydSvUlOa&<{4t8-gQh
zLBXw85@*4k#Ssa_Eh4rILYWjO{()cV=nqx;y&!@)+s3{F9{}I)A+J!*%$7T%SNd?d
zRh%Y178)%q%86C%<Q&x01bh~gy{c4GdKu{{>DZ`<<Pf=RK9jRxSZ6_c01r8Tk^Wtr
zWI>pHqlkJK1MF%#dcD2Ts1<4ln#utVti!(Ok{p5-qpXG)`IX86AXp3_>8P<ZC=Phu
zM#hnRz4#p&E%h5oIsyqcdJQ8XU3fKYu&5MRV#+7T3}EA6AQQ_;$UzbLo-afzBxNZe
z6=g}JUPsZgRvn!~JQo#BDFRH0sQ=ra%1npDr8AKL$Y^K)>J+j*9&-RK)5*c4hcxL~
zXQ|<$K}lm43Wr#E&HA^;8Ccp>M(vNcw;^<s)q3A9(n27rLj(@`cZk`Sw2j~!&I#uz
z<EUNjHj3q;fnfah&xJXa9&X>77n}R_U5#PCWujMOzg$5>_uqm8OrVoOrYNjwIO7Or
z5-f@=DyD3#p&ltX6TgiU(&!}qYGUK<8${w9_j-UIBWEz1_vIb*Mg)unyaqkrmb0Fk
zm2AsRSAQi9feQ9#j;v32@CfnbIrma``wK=8{jIrwevUOEG+Sck2lMb}+QBOl)kkDQ
zoP)MSfWyV-d!aXciF*`{L2eRwS}>m8842>k`iH<)=hNHZu^y|PSFFMn@ESBGOOVVW
zxcwaMBP;z)hwX7b#+lO?FDruP^$bDeCmcCwz^)yg!n~ZaW8+_%j|b}fO|s>BoiSuo
z7l?Vm(}DEcfDK0;*gH83#EbJVmjKz^rd+@APsc*v>_u{Y6Hj7LSLuhgFG?$9vt!$m
zfYRm6IECVwnPtSLgNnM{+Cb)(nSY~}*AZ!sPPbyiMKk~zpGLx}O`|vE+j;X*sPK>q
z$IRwSIX!qUs8hROaI63n;km)LlBp@9x=FNyw~{3{jM5K1$n_GmWh=Bm%*TUn#H&jE
z=>;d$yV~7nYIyPXlmvG2@sl@k*whqZj9W95E*dd|e`?CbA;TMKjEq#Os(OIWtvtFu
zmulP!jgjH{`SgzDj|@%WAdZdU;EatS;!I5;B2J<0uE@pOwT3M_5TxoJH|)L?U<)kz
z;CflSw=~d@8L-pIt#dOQM%Xe;hQ2maLmv@%z7~8?<)Pab|2fZGLcUzyL1$}FMYuOC
zw-n!cagpXFMq6S#@~-4T)bH!CN19H}D(eeK&5rhe>VM73aQO4cgse`wt7&eyuH(-2
zh?bkpTt!Y|5ZSlMV6qKs4iY6-Z#fhd2SUZ1m!fjI8U02BW`qFw695|#__@?0Ou6|5
z^^S<2UitF%*+Jhoa<yN(ei~iyegV~~l*KSZZqW~S&vteg2OCZo1TDx+ilsQ5Fi*Jy
z9Sr$XfLx@XE{)CQlet#?l99ib*UYU-Y-;td>i1W~liLmlz9)`p-^G-Zhz1Fo%wD?h
zy(uw=5b*;&1KKmea53E{Gor1yPCMM=FETxe{LtSt>_%<-5^8EXp_#)2;R<|piIM}m
zVUoKyf?$2uSI@8zpH~!kpWlc4`PXVyC>qH^K?T9}eR>p|)!`OOBu`A#mxY29qH37)
zQd;q}gKMw$-eyc#JF`IN(i8X#K%}IuIMK8x*Oybk349ENo48DH&w*k7PK*8fqcSpK
zSCp4MMO`G(5$Y~cKF!<Bz5eju0u0|CQheDeL~FqF;cn6W$4HOC@f;uh(klDCXzY_r
zEj!58>RMOTRZx45tvmGW{Z^tR<B#>rp^dcJ#BEI~k|qD9{QP@%o@@!@WlVhmoc5HC
z=9?$H=Qf7~!F^%$bJW<xYkNKhg)4&3k$au3ot&^H{&m4o?UW`L-YW5jkLVftp8Hn6
zj%U*64PHt`J>S6~g2J=awQ-~iX>&y1_5F@r)M(FE7|KXyKV~H20JJ*<Ir9Ca^a%mh
zbXKc4D&~BPsXlJD#u>hTvLKjmFxcJzl@ElC{{3-2qE2fLkNWts3iij%a77S9XUO{-
zVvD2NW30dFB&$fg=WhRsS63Z4Otv%q+8l{k{Ry+MS;f~{=zr45{MP2NW6Mv?*|NW$
zAY$-#6kS7v28Ok`qa)Wm43ijwHP%Jj@acZw<jUhNVPW3jCtW7j+d;D7Oi?Fb(o$x+
zf<@-1I}<8+D(gE4@loX_iVeZspy83p#d-ORPMmi|KW`s3n7kM>?|_agHtS1W=q++*
zS|&jQp9{o=EFB1^!FAv3oE_7#v3kJ{WnK@8(>n}!JGTJeeBZxb@EQ8VTU8wDuM(YH
zpj-QA@btTy9fLej<q8c2F|}_zkgv>PO^G%ly6M~ZS!<z(Qd4rtgwY&j<9@+7CZk&x
zfw(A{RB`ag=RtIh8BhrUf!hKlOmk%=6zY7q8uUW~0VfPOk<h!s+p@d+P1l_}63>=b
z`hj+FC)2>pm!Z^Q^!(u8A*?HtT+cdHVP(5RftfB={Y3>G=VK<ft>uhy!e}fqaf+;3
zRt#)-lS(4*p+kDn0)y4=^i11JbTFxKb@<xeome@Hdv@O2hOYV8g8T&Ui9g;B)rDOH
z&@GEilx%4JS&9gtls?lpVuu%{ZV^Fa1~VNH6yg=Eehgr7r*mf`NX6u^#v<fHnSIIO
ze_*VCW}*qgclUJ1Bc#G1t3(l6^lhp#C0|?AHAzFC`oATCz1B2aKa0d*JQyq-60>yL
zJ%1+_V^pc|ZA9jBGEJ+LQP2Kk;pKHHNLT^Lfzr*m)=Cn*G7;5a*?lS3{%yJmnRJkd
z80HDTc@k5$e)2vb*@J49|E6?6Gj^GWGKXYphtEfm<MFXw($$)F7E0X3FR?f3SzQv}
zWYYByTNvpfEuMS`RV_@QZOm)1%-Vr!7*Gic&K0QVdaq5t_5ur9F{NTiz)+6N*l3kD
z6cfS%$+#&^@l3EW%d$~_UiN=V5|S^6b5>!Q7Lw4!k?-@vHojmG22bl#4iV^!_&qgM
zE+OjSd1SjiDX}%nTeXk1ojS#UzMq;i?0evmAf*zKb&!}^_pSMX5Hp0s1zAb{lXtL|
zDp`(DGSu3~XDI`R2?@&&%ALzOdB3ia-Ov5{eM`|C%%_cT^j^x{pXX_6^M%gk;@<5=
zfMA`jyZr;rLqz1ym)Y&owQ>AoR>ZCUh$eJdD}v>wcpO2tNE3M2kgcoIw)QP;@6N2I
zC-S>2B}X`k%9jdo@`#{v(U6TeH?(JGcsuY~r|)sUT-IDhZt6jaI@ZgcQQmysZ9-t(
ztF%m+ZILs9pCYt%gFHUo!U7MyWajvD*GG4n3%BYIZsq(2T$8^*zc)4|ri-zjerszZ
zh=2Q18MEcwk6>CL3NkLTvryDNy$&s!iPkthyAqN{YEx1Bm8bm<WNXwH&mq293@7AW
zf#2U^?u}MQ5Nf1mqxb-7>Q6J#tdT`7l(hB5bK~N;QiX-_xH6qY6%0g2u@8HcB<b!I
zzZP8@Ri#!DZFj_}3&pV4Zl!H8W<<A;q8g1jP>MtfylRRX7K%cKRMYSj4(tpP(~iPB
zh@8d_+*t4hlAAz1ddTYhUf|B4wP9odVg-tJY14w0t-VxcuayPZL&$4;gheR71S9be
z%6i<nqEpoYwv1d$T2&P!q>Nm5h;x(3E)L1^MT=Egd50fJ=Y|Xir(};76kT=xN2w_j
zjAH1GgDL?THzLE<tR20P!MSG5FFEGBKK9YqtnVcepc^vMe1o`q#7G6zxBiW~TdjFw
zp5q781&4hGlZB3oBmxvMzX720IYGs>`S4#dA*;t%BHMLj=1|(YKR_=T@>wc-iOFN}
z!dJ%2)clSZs%Zk(c<zUmVVN>8Mho#!xAeS;{tXxTX{%9Tg*w=*f4~>L1e!>_%G-{v
zBkweN#<C!8L&d^FsjLvpskP#QhHa@hxaH|3qyoH?RelAJI@jAnZB@`3n`<xQH?~bO
zu>K2-ndN7r*oNn^J14V$x)o0pmG&v0Op3XqSdcS|O`E$=B&xTV>^E=7Llu|95NDoM
z(**I?F0X-I&KR15ALLCtVx|de0~H$mf6&m<$WF|}u3S%CU!yhg=YqO4ld9CPV5SQN
zlug*6;ik)xb(-a!w5DMl%!w$;rlv=$i&K&sd56AEguHXrF0OCfm=0K2rLV#Yoga`U
zhh^{Vy7&gu3hXuT@JU?IfCU9F>2;Cl-)c{1WXHjUZnB;7rJ$YR{Y|yu3YLJQpXVr}
zv8N^mzhgm6=e``1=QXcK0*H<|0WX3RCu*dvcGE5&B*{_V_q_}R7nU`e)ti#Di!tN)
zLY5<itrjwz%~@9mPaJ#GN@3tP(=)*`NxlK{TgKIm{cbItUaei|TVZ!vdBye%tBiDn
zuC%SII(~06r&BFX8Dhu|9kHm(M;hZQKj&2%CNL2b-uNHAiVD6Ro=iSV5V4&-xM}zG
zw;|wWTJs2T2@W5x^JCuSELb&l<;k0#94%u8U@tQ%#s0Yz;bbl3^R>+CG&p=V+sue;
zkky1JQMEGYeDZm0poQCZLoCMKM@813hMuIlZAbp{6!)*vAPe|ZP6ibsL+v-S<`Z=~
zQChd0m@Q2sp<c&dZjwj`T%L(H>=%%_i67>BT};UN?Ti`Z^Gua-l*`2hnLGSBsWXQo
zhvN2Vz9ucawo@UlT;#Ukt+wZpxttzwPcC%G-M!b<R9G+HQVJInFb17c!K^ynSd<kl
zFQ1#=S~(OD7mZ>C<#VJ;U`(=sD+Dcd7R6ItJ+n!cdT6MQe4B&3uCdO!vMvADY-4xA
zz2$R4#bth8pC%LC$Zd(5{Ud{JxH_~zib|;M>VjhI%2W{=^3AT2PvBNqZ@CS+<8|fV
zkNJ{Ju1A#*iX?Q3&-h8*`nMNbRRZy#zx~e&S?v#8tID;CZ94u><N_MK3B3mR1Q2bD
zTU(~}OE&FXN<5vLm#=Uei(zEJqXwBy{g21?`+Gp9p_QCBa5v(n-5id(fh@|=aWAQo
z+p!qmkSUXZQ1wyS26+*0d=UkbB}k?w@3PpAzA%Q>(!7!;k6+PdId$f2)%sCo_6bKD
zbhRx<1Lid|e}y_^F34TN!C0%F^NAiQe1GId^CQbyITd?pFwsA&taV6dzpz{<X}SCp
z#Cl1^8N<#fGBcxd5%XOqBdYSmxq$?0X~v!K&mDeG#K^C+Cy3`Q<(z)TtTlJF2mkB{
zt=ruj@cYr56SDyT)3b_R16Bm#6V_7&F~%H_RSG4$um`Mu-|h_e+5gAlaf)d%xyhxb
zo8_|p?E@0uOD)7&Js!Nw&#tj2CQ3B^aOzbxgs}X!-LA|Tm=QM2cxnm{3Sfb^S&rnM
zMcE!-dHyMTL5SNE9g)77nIk{7<Xu4?Vr=uW%CEP?O%Upm0!<8Qy(|!fkcd^tMj=@{
zIIMWru0W{kP|iYy4|lj&dWUl1zq|B73YY?>#uE5IVMClI^Pu7t^$>>LoSJ}gmsztA
z2!0yjsjV%iKl3k5qTU=S;(O>8$=s(n7U^h8YfZLl*`0KVMoNR{fG~s&!1^|AKuab=
zvo7PGx73Jr?`I{Sd%K8Cy|jukSTd<Lf_@9u?)3HO<E(y8PjV4_q1V4jy!?%258c;%
zpZnZZ0)Vqe`eKnRxoV}N(r5VH8J<O3HDg@skk*6ad5&V6-_e_3>%=a|SzvY!QPuba
zgHG5TBa28#fOchhdRl6K{3Z$88Edk0b_<M$BOxSt<w7mW^|T~x*oFWFj|!B=HQtyV
zk}Lig$1?GPt%2K1ongsuQOx%F*?OWZC5ycz<zjOTL!eThEADWjV0afo8C;CeamXgE
zz`|9)#EgsUvqj^zHZ376Kil*Rpx^1{sBe35VIzY7SRh&>3!R^9nr78u&({#1L_;UZ
zY@r<eky}c@3tDUV<i7t0W0-v2S%FV^$Ju0eW->HYWU+{^Ul!}Dp!Dl=6H6D9d}7!6
zX_m`^%+J3RU9t}F7nOIO$#xjlt4+-G3^X3Ri2DTU;*#X_)`!JX5AzPG^J)^Cs|l90
zNkhlBs~ea;5_jky&g)S)5A=AGE8`aX)FmjMkl9Up965HryxTuQ=#468r_$HU=8q(z
zkqdEaPIBSMGbY*0{D!#5R<V#)JG#LZ_P~(S#+`~;q}WZ<W~}GYSQ8;})VXaoJmS(7
z*BO2OU{yrBNJPai0YIV!kmFH(^h&6v%X~vff#kr$WJf9@i0H=dq=nWDhW%)zz~DO$
zeFMDf%1rNV@@<SdQ4JL;X@4pozK{~sW`GdSxQLdiqfBgC#T}KwOu93L&z2SZ7_7;j
z=N(t{jaDs!-X`=nBH|L1B!7sQ(Qw;e6~|=Hb_70-AX#ep(2zIa+Fv7K!<0-`y47zP
zXB)#Jj>P^#7eE;G2CQYI)#wb-4<(^Ua(_vq@2(eEXY_4jU4rE5T<y57e%+DeAdwoo
z9n`J#iL~Y65rS@RqmC^t;`jkMmv&|2an9MJRDGl?!kIGiJZ8&i3<AVUS|fxI3QUuZ
zJyj!Qw%(kC&9L*eiS1(k5;<vj7#lNfT!ZncRoZ^)<{7p*W?QT1#pv23{^>OOB}dzY
zhYH@P?8`%f)86Ht!v=*V{S7?r7U?^vi&41stoyC=KM0Wvb0%%n^`jq6&i&<drHPe}
zOgd4SN<&`kk~{{S7-lL^tKzd6s!81B0esUZ$!pFuBBm~e<f=l|4Mj_P_HNv`)LD<l
zlQD0M^schf_jOB9Zqh$5N!HkL<(tNX3j#5;*H|j=&d#5g@RMN1HbfYCTch2MS#k-+
zy8#RuDmMazYqwm()-5aL1&gnTj%Q1JkW+6D^*K!SDEzde>QB6-+HZ(eBs3D4dKlm`
zjYpc3)RTfyqT^iWGbkK5i;+pmnM+}F3F-P0>x~oYW)AtW4#%;*3rfebA<{`b3wYkZ
z?^D{hDEE8jS)XRe@0yo2js_BXa&~zzxYTw{on-w?sE(PKLqGsTHv*A{bY92X<rA;e
z!!POV!a%H?ty-VVoOR)x@x$Mg`NIu5n~1qYXBS1Vx&;NJ3MJDtMTu$eyTLVZrH5`C
zO~JHie8*r{Z&_d`ax}vjZ6W9flO{AU?FH?D8U*M*MR$wJ4h&)IY<`pqL7jAvXmKoo
zAgKIkrDN(^zHlCOf@JM^G`2LCpu7u5DKu0<W7IP=2xt8;@}6Hja7fE!v#7}|Nnnw{
zOx^74_@c>V+C5@4j^qJ$_X)4^r++)myo$4W?s7eT-gVC3o5B**UkWwD=V3fwGnG_7
zOT`KlRnJE?GY}($pgyzQz$|yD@W1kWN5&?fcd(p)RrC8JZJPJT3Ui{FCfHwZ@n-i}
zS%2Fen+%9|9wwahc|T$BN=(8s<b#Nmj&qnV!t!vBXS79i4)(@aIHmT#FQRcN`?>yA
zv;5pIu6~;P7Mg^<Z$UFOWDA}JZZE~pB)BRqiwTSA3~lj`E2)VM9xV+1y2S&vFwdV_
zlLg)UKu4;8&07LW8hUnmznA2k!~)zvk=vF6iA1++`z2i#gmnf~es3M^>r0hlS5Z{<
ziM((g?E`vi3cNs_{ctD^zYr^jCac4^9yn~iGqDomAry+)IbcUE*(m(4nECqxex6zf
z6~c3@*9=xCy>z%orI9sf+&G}YcAC^<zoAy^DY`*5^jQvm;G}XJ)jXklxd=#;e&9eA
zM(AM7u06zjxrvy<BXx$nX4zK~RY!i*k=~Wow2fhQzAhia@m%FD)1G|xDhT+-Qw<`E
zg8S+#zLK-;JU^n7z!q7j=zX<pPj`&<OQJpg`W7X9daFUpM!p_=-AH|21-%-&M_jFe
z=fQ&jLZECZhPY!JEyHDi@caaQtj27;VVZo0jm=YX3fhU~ymbC>zJu5-Njy#a+CQDs
zzn*W*ij{oa%kS!J!PC?S-~JKccjebny?DlkHV`~oOU{y<Q#pfYIMG)dgD9!t%y?Wf
z8S4EhOD#R|06}t9+&o~&n4;Npb<(bMUEg*p_RWPKeG<u6cewF%-{`8YNo%d|{_Jfq
z5YE$%#j{Gp<FU+ZrxLWbHb{rGVpU5arH2AId4!!4z>hPSZ;|OnZ`7{#D(<k}`>{%7
z3m3e>a;x+-mRG0v`ua-0qwdoJ^=?UM*7L*6b46T@3ui+QbkzbW4&e@29u`*<pA9Au
zj2GG?%Qdrl;R*)s4B^5#Y5f8zb845W84@v+c=sDL1_?3#zlkLO3qc}bVeM?<$S7fL
z=xicpVr&OAVU#nmHFGv6{!ciz)hNAz*}pJjpb?6>K_0k)5%Dm`FvxaNXtXKQzl|xv
z?|<2Ja1iWUXgIPU@W@ho_mqf$5Sl-RpddF-P-Xw^68|er`F~Yv*tog=Bk)-L|CAb&
zb^($sr95@5f+u`_N;*=iq$xv8$hvyQkbfaLueB13f3zNpH8%OVqNzJdi)M<`tr>os
zQyGiZ9|85|!r}x<%?`N^R|EvdiboIA8;Mw$KF$noSMoVo07~xH@9)p8>wgCs9ETUn
zkO@1H!)}#iQVxYqm&q;&KoXRS!PmC<o%C&iBJ%<}sL1k>#r$O<_x(^exEVb;=ocE3
zZyLhlB{G6EqT-{Nf9i*vr$`0^&WFqA%LA)`kgd=I*Ah03kn_UzfCB&`^b1iC$C4@%
zE0Xw$Q4MPj2NR9xB+|(p*h(G20r;e^3c?}b^)zWQ4kJCvEzqV5AmOq8A`+%L=|${V
z%Sr9QJJrANDFhOyH!ue_kOwxXrHSSqI+(&kW{KHqi%7spru9c0T^#|<9EP;ARO^^@
zlGrH&(X)pq_6}f%ypcZXj!f;*>s!aNtc6)LCgY17xX)cLe;;*HU`TOgdPwUEW2Lg%
z5r!oW9Xied&}JiF@5xuoQAHfiJ}};RweDFZdXm@$-8Vazn{A?VT092UMyq+r$N(&`
zl;TuJ04H+T&fitx`d}aX@6{p>rXo-x+xEDrr=ti5+y@>&1{UN9p(8T*c=1scAY?t7
z2aZ-yXJD1Mp6C)8UcCIMDG(kl<U*_<ep3n`Efzo=0fW3tmL3j+N`@vjUbI_c=M={+
zyfZ>-^4kpCZ1@7-DTF(9%&hshkLiV^Igu4^rC$~uP-JxW_HV2pn_=Hz9lA+A9--Fj
z=(6FVp`W+v*Gm6L0mW4IK_*@1$5c{MIctfYQP0Oq+DJ$43O_!fE5XZ-Uap@$(!Tc&
zs;Pkw=6WIM9d=h?pU@z*X4$fN<ETTMLM07pIqyL3U=MzcENgJ(?_pj4Q~3~vg6`G1
z9jnEb9d+a~R}24PF|2fzDS=4~7hc+k=N%x%xF&-oWJY4~Zw=}Ixl<^tf`*Yfms-*7
znDvdk$RfRvA?wJAK%HHAHo3#czxOQNI7V+>_D~(F;Eghc%pKt?<#9GgyYIrr=>)1P
znWH3{u_;NQ4VaZv4Hd82dUf$9iMO;bTth_^N>1fkvQVUgCtSK-Rbd<FXGw&EyjJHv
zhd{N|cd*_v7Zwiv1`~0%Zbhf=!&c;&8XhsEC1m{TA>`!|07&7Il2|A~5vb=ne233(
zkSAzU18$6ZyJ{lMbAW;7{$mG`;k8YV95c)ri|n5QjU$l7Zi%>3XsGPKqMKd9q*Cg+
zssTv>=Ri2LEu+3udwl$7f^^X9y421Ggdx&0O06IHkeBvr!gxrOwYI5$CWs5m&M%JD
zr}KtU7L2P94C>bDiYf3s5St6_+0SLyub&bPG29?%(j3D<N4g-zubDm_YBN7uY(|<L
zJX8c?TcOsK5{Go*n9SnE3pMbfJ&l^7)Ns`#a7t((O(dtZoAhcx1}_ZJ*v}RtbgfVG
zU9Jb}?D@Qr1Z0(BRnAw9OuP?lb+h!|`wWl!Oh-vCU3zu((C6ki7Pms4Q?GU?!^<K9
z<$){l*7^oHv7osfBM4bARYsXIa?NN6*s_ZdGl3$QbsmVG>;e@=$O_~t-ZYsiDidEE
z=x9t(ECz40D(qAYb70XqMe3I&6gJwSjK93G6nZp~e{2OCk1~dKW52+BckjRc82udb
z<}YD@R8JjJ)e`wl4BBD);Duv9$)08(cB=RuW3$-G##5+VO>$JNRWvc+7V-W!+w#L|
zK>1E)<TmKDwQ@U7Q{jpl>=q=k;{Y1>yz|${9xtbItg^Kw54+XVE25wOPrFY^?$7sc
zkZsZ3=R^`X9Ffs9N9ah#x~@BI5<z}eA3rrc(T<wxkQeN)93LS)GC+y-vB?c90Fzt(
zvRj>qpv+IHve<t48R5;70$nDib{zxp1K^P1?<e$2=jSd>9oj47<!U1%wV1Tiudq{k
zZ7)o33;{4oG()!l#@;;G)-NnprQPk(P*(Mif7bei<S_D0NUYz1yd=n`i4*NB>yQi0
z<bTn4pSm+@(Q;owT8MWAV+fts<ZWJZHHo&vi*twvd`YoIj*ZXlj*Q!?V}e;fM3JRi
zs474n52J9=2^rLuWmj=Bs%ibm)d-+}_tVs?>e0j*dKOMAi3M|?CUxy}oCnGJ{EiAp
zO%G^7sr@CBSE?qfBBdmzx+Y=CtPsLq`%qa6+ckn)KOwGd+$|vl79<f%EgLXp!mN>u
z@^An4)xRN=M+@R<fqGjZm(m5M--jD4(Kv+)fv8|4TqJO>7N0MaD~=l#9r9q;DkPm=
zUNkO;?;rnfhKG^5-Bnf3!9BoQ5uW1CqlOSFoS>w(ZoNzmxHqe_lq+%MeWk`IplApo
z^h-J=Wz?1aneBp!<@QOl$XT&<c|~*C!kMsHmEuOXC^?2#y%R}*Cw02TmYu;I6{+j)
zJhohT5EF*<#AeAYMOj4GT6L`>OTFbJG>B(t6WH=rqJI>0@2h`q5^gsT)sp_T89;r}
zJqRA*G`O(OM^92boQD@EH<?eStDpz$V&T(k6VKKJARb%58L2olO{)AM)u}RZVNc_1
zQDwk7E*g50=IuETp(b$EE~I)uS2v+9E09$rY>QlkOU=thS}sT0P_~Nm8l5x}S4T?u
zRs5%{riNQ8_>+Ad8dKn#_1R?@mft#TT3ZTu^YHPaDa-eVI`4R|;CRo#gC<%6S*y|w
z8#65isUmJv52$1UCz`Ndn5^r4b=P9}(#F&JVm$is<BkQ!<wR6!^}0I~s?_iHOUR5h
z{qUir-M%CJ2-ySUN(^~rvN+5kHtLUyGlYfaE=7%>r&9GvRYq+(BF<N|ximLZ>hA@X
zVllEA)G0XSpL`0@#ly-C<rcCrRqs1u`ZeqFGO0Ar7}6n~H$KwR#6*mvAPlHl`}WJ~
zMmhawpKUmi2Sw8x>wcd%m&lc{+`scDHkQ<8Ul<0${uI*Ndtaj7Ie+ec*?t5%m4`0S
z&Oj%8Os`h-<<512VBfg}OOa)uhsZ`BnmiUG#_-f1sWRewA-x0EI^2A;wN0UJ1upzX
z8%s@*w?(6&3gHZB%zRy{OXo~T-ga;+m({HvLwR}Sjw+VomNL3!mv07+G*45wEDJ~@
z=yp!wi#2tBp3O3@@8Z+aj}R0lyrlT?qWkJixmpQLV;Q)C8Zc5*5?@`5=@4)nt93SG
zsCFz0Nm7uXprh~8N3^PQXLwvrH3D(remhrtShoI<U&gRp+Mg4NtE|Vn0RAPBwX@6Z
zhHjbvLy-nchT^06!@t}tHY@!OweyoT90GUtUb$7m(9-&*(Sv7f;>M)0arL5Xc-sE_
z=h1B$_mD7$hul#;TEW!n%^~;3Ax#Rf96}o>y^YR<qvC%kd#51Lf^OZiY}YQ^wr$(q
zW!tuG+qP}nwr#8T?U(*f-_xh#zOR`1l94Op%Q?r`yHMs(chET)uie;lBpLc6TgVr!
zlX*LR^vLwKj<RsFqN+>P{gA%CL$&*c!aJh2%Lr@;6uB-i75=2mL+6r90!OF3vkJ#n
zxc~~IQj~h);}iGE2`Uf_Q<_(4%tcsuKh*T`7uK9P(;w&dp=Fir$nBqhWv|$|G?CW#
zc^Ev`Q-Xab*PnI$I>x7Kiv>tf-aUl(C5rKQ?D)|sIr+N^|B5!QI&x4X1egW<N`{om
zyc(1gp7qaq+qwicd1bZrQ*ajFI*%Q8tqVC|dLE7Jq8_NMMqq1sMI}Ki&&%^$5HW7O
zVW4(S$}8IM7TGV2V)I&NmdOD~ln_@Wom3$`&D=Rk5G}mf^4bQ`!OY-Jiv`;LH6>P)
zp8g=T$_~IM!eJ$}^qPw=&9F$Gx}4dtnNbe4z}xLIo5eyMF3UxP<3s%_mtu^|GV_XO
z*hHXl>N!KcRZ^Gj@3)A_9_e*kGd>v<dlmq@==i2c;_7}|&?@JxB_T_4a&>9CgrY8-
z?$$Y9s0S=qT(Hu9uL(d2Tqg=w`-Hkso2c;=P1E_ckyPjKs)zh77Q?1ZwS>eCJr?E!
z19V-oG6Mb9Zcv7L;NWh7yGb6+D35CAr0B;eIQo;h#XsvE_+<jL`q+$(l$u&?4{|H`
zzmE4Tri(RTrGN{>wrBbad#(x--J_hgHtB;3(h^S8-4?cQ8C-YiNz3q2B}d|P2gMZS
zIuyM^bHfMCa7&ftQ7<=Ogt8-#bXnP&650F^Lc@AoC5{Knt>hOAu6bbHHae0V!@jE;
zSdN`7*BeS+0GHFqCLke#_V|Y{TiOqr?{;3y{zyPPraNQ3`R|X!n$t)36qv3yfH7s+
z!bEO!yCd;>!N8R~f-;7|%rfgA$)<uavxI}CB0NJ=K3HaevuBG04~I4d#B9a8lfluL
zYgeJ4J2kVOWh3%I9g-3YDoBo>{)kCan&}d|$W3L55PQ+XL%i_V-<83N&j+@aaj3B0
z(X=dpl8KYJj)ot2-mlaLUe4{jL|0wI59_(nACJ#vG@rU-COYeOP`p5^j(Crhuq%^M
z)7c|iq5{xP!`Fjxik{aPHhz;!bBDS|GbuKoDheAe4g<(e%sr9~VLn+`dmvw`&zFw)
zk?SQPw=?|XP7j)Gp6-tu#EIXydJ?B@S|ad#$lyl}JI0}3pwYxgtaXlm(`EL;4jZ`Z
ztXUaTm;uqRnVp?gmSiuB-bl|K@H05k<E$O#jk~rRFC(Q7AVVC+Ra#$sKz~7wGzlc?
zLkw9kFufZXxXZpYE$g$btXi0xTfi|6F(`Z@w)_71^4OA;<8bUyGfZpcxWgB}SA_qO
zxd`yi?h2M%u~9X5v5yeZgIl<0ci0~rZ)n#*K)KMt9@RQ?>!8>FVxB7fbz5Lt&xOXg
z&Xp;LDe~!W!LkAc9Wf}*SxZkHpY3CMj;@AH;gkY4=g{NPj?9+fYU+&eTjUD4>5koY
zd#7Xx&!hCEqE%Y(!AQrIQhTixJAta=Cm7&#QtVO9r&GQnR{3;-!eA|SoyPxNxK6pn
z8PfKJ>J&$-w?rct+al5koi&(WUDLug>Ezh5vISj!JGSlP_BhW|hT`i~kOjAbv-U^%
z5nCr?>sZj0+D@b)lPyoET>PzuF|TLini@h;kI|I=j`K!qWd#(0O)pt^j1;3ZOs=iU
zggBhDehwf7^D=}OnfJNJr#51<^n7haGRR$EY?UyXwnNPnayxdf?p`Cq^Y%EcEOEAc
z!D{ZX8YtuUZKx;HUjSxpSu;~cL8|APcTLb`Io@N2M}IQZj(UA|BwQ$YMNPhpp~&G*
zznxP%32FONbERz6Utx+?{dfC*8_=?FSR~D=tP<f51*`x@cz#>izsJ1Gdcw*X6Sj0I
zoS@0=In4{m^<{UCT+%J0u^{v;X~S~`_uURiqL7$H4qSUwS)a&rUrgG6LJF1SGvc5Y
zWidQeH7=*ddZp|u!Z*%n@#u0Zcjsi-n;!q68`!0{p;N=Z9R+5Tgm9Ig%boN+X2(-k
z<d={f_5)CGz*QZ6M9C132CX?#-e|n5po3s$yH6N2l>5Bsi9*P^v@bt+QPA`!7GZ1-
ziK@4KB6&RqPNwIa+%Y?lm}M5O1aXN}8rgO)z*;f}%}IY97@N^`1|~l9H9E&NrJm<l
zbjj58JU0?#e@?>@X(oG2UIkmxToxmBc7TMs?M6DKHJeg@_A-w+2^HeIe0XT3U6RHm
z#4H+^fYneiqre{CwqHHHRS4)0c23a<umTTNv;T@r|CJrKnW`Ut-DL+6>a2i*NBx7c
zm;!nx4K`3hO(Aa(kYuivY+<>W48P}szTzp|6y<ReT-?}-Twf4W6_ge?R$QSb#&N-#
zX%4(|RCZz3ua)WWa526Grt1O^HwxRQrtMUg(?vJ@2tGN>_<j0w?6T!qhsLrgtqOgz
zkaTKf0g4HH54EhD1q~Anx-XcH&jc4s!@b42_M-8Pa9-A&qGC9@=zWi9zI0WPdsRO>
z|K>c0d--ZwE)rqn;CuAL!mtK?3Q{W9+}gB+*je!rQQ_Odb2rF~4or(q%>_N`JMmQc
zoqrW5nafX@A`W$>nVC|yX}d4Qu+Li&gxM<P+;y#e**)?KVHsTgm2VPXPi(r$Yt?Vu
z@BVfG@Kp;&SoQXk1`W<*dJ8kqoXZHT<zu)+{PWZ$Oh}rU`$pO$w&OWwUg3yC3g3^W
zklET)Y6L>(uPS`l9GH-=nXgdMr;IWAHx^`-rxdH^lb7IXc_lRS6!(InPB-WtyDNlR
zaBZWuc1R1}L$)c7;+Uu4@hw=W`?e*HMrfP<;7_zxz=YPlC;O3cRtIeCVn@oalHc+0
z(5<j%0`eVp{HIZN<?Os{%V6nzqH(lo%wCE|UKxJkk{`5BQaQ4Mjeyg~-}H>Oxb9Vg
zk=$(&9s-PXa(Wf@@y=+nPS3j&3kk&(2U~q83JpJ-Bq=qU!Do#;>R2m#_RGn7u<~kt
z>XZ^PyaX6*>r0DrVwWXsC6>M296xnfnNgYk8Fy!d*KuE7Xfs*a_e9_NobTvdei_zB
zx-?f16tlz@>)&2`n?AMEdxRy2n~XM|H>?O+D*E(m(gjmzHdVc52Q^xT;R<Qm4UEM}
zuE*`u-JkXx80eq-hcrC+>}Tcxd^)gSLG!Gv>*reMerm_yR1s-(EE*UbctJ7`vaIa0
z=D7&z-A*bgtNeVO<R{C{_9>(4m(>v716OHSYNKDaX&HU9|1Q(}4~W_Scd*F#U$VRZ
zyT!Xsj+uG`dHxOG2nB#Z6V&&AugysRA2|8{<Ghfanc+VHY0>JC+DfAxd<;h*hzjgE
zE18C;M;1CBDtVeT7S{fhP*B|}Ngad@#8vSK;`sWA10Dmo{}8E0DD|3_D5()rXPDI<
zm(?#V$t#*#*1L_aF*u)|wgZGuFE2f43CJchQ|alqpTE7gzaso=>xe-9eeQ+5y@ay&
zU7rUs`wc^B2*uU^0KSpFA&wy2_-?9wQe^kOs1J?TFfl+)6AEU|O`h06Rk441(S>Ib
zg82ESUA~twI5}oUetEoS?=`eWXJZ^8KP7_v?C?tv<&Yc0JVB5kKofJTj!;V0ux_wW
zaZ4kVN$BjIU48dehI@>Ri==B3Xzg8fJH}je2%cS%KZ1iqsg@w;;}_!E=*x2}mv&V!
z^W+bA#Lt05vYc$&xQR&xE5hQyLV5{gq#@Ku#f>FJ4tj!4wTt~9#z4w7^P^G@Y>nNd
z-NXgm(wG@&skJFo#s`ZpTdd%cKFAN07oy|y_w6C~7ltE?7lQ8W%C(C(S{J5elF1!a
z$c=3f-J}DwrpEF>*fzU{d22yIXm`#&FE;$9Ep=2dlbHqclR|<k1qlt2w11bI46gx#
z8Ci)#`f4>mq{PEHhO`ZE8FJP6xy7i7Q4u2{*$g1-lid8(MZ3h^5U9dbhAa+H1`_W`
zy~3P_JO)zN_^pV)!T}DP4UiGVtq59>Y!NoYc!vHGCWYZ22@@lV2qQBHQzMF95hVph
zh!Z9RuVKA42dL{mM&iUvZ6OIm^8VS%Ey{S3KckI8eSyZ?plPy1mO(NX`6iYSO=KIs
zyjEK)UFXaCdO8vY=L(&MLxJ<nO<*|AH*7jgRVr3GEcLlPI!W47K5;V4cx~!Ge|l<O
z#47zg$h!REqxICTY5S74F2$XppR;KK^WZSn0;0K#7(Pxyv8YZ8VRN3(mXt+n=q)0L
z?nA#%rupTcy-A;RA7wSyvfPaJ29vj;XGR4Ql3_75`f(vOfsI@_$v59bl`MnFef{#e
zz@on71SFqLJ4pl$O@A?|i4>{)ja0Rl&>#1!kd41SX+Zz&2H(B<ZGfMlG0Ar0dqc5a
zNN>Ve=UDg13yWSYVf<{2QZcu3&gh^>STCI`DN}EzrDf14s9w&Xl#-I1<o7hqEDtFs
zpOvmPS**ZHgkuEXka4E5GPD+R>>h$7^(8-Y9U$+Z-A=OwnCNikxo3%MV{fde>ZlS>
zoN^8VV*XZ9(9hf-gbR<q439V{U_6BB3BCtFybX^kf6vNMC?&aM{zq{3K!y5=2<wel
zcs5_Rkf23#1?fA_4SP6*_Btd^*H)H;Up2fUf1~e(RVO6EqzhfMg#Y7kEN$rQO%quS
zQ>&3TM)<Y^B~>l>7u?W{-Fh6}cc-3<gA|FM&tA<%g!qoMS(28tCAv~SmXDTXSzGZ|
zpk^LMFa~$>$|P^^^A>;jgS*+TdUs>_VUw(1<$f62bhp!`{}Xf%WH~kXE-A|ADWWan
zGI^52o%8J(#h#Pw_WIbP(l0R-G^e2(RNgTA9+VEcfGa6P+Rp^VKD2|_E`2S_mAMYN
z%edy6A?<o!x;~vtQ5g<yb(i+aSQ1N0s}rg1=pXgDg*ljJ?4*T)cs9~FOSS}*bG-!!
zC9!%&Hq4r)Om7n^55Lwfl1Gyw=GY}S9Jr!&=3I{Pxf&g^cN4UW-<1^oB@hHSLphC4
z4_+;Nv;9y2C(n9x#DF6);2Lo^PHRvwZg#wY*espBiS;85$|`1NH=Rm=_kdkC<m3c+
z^Mb*o=BDx%({`mHMDa$s^T_g^pynkGatW8y+zf{|!1`Ov@$NwK3glT86Hfwqu0YIj
zkFa57%6Vgh`hleIDSiH`7Hf#1_!A!JjLYkGd`r0!AqP4i&l=qx6Sqdmp`M{SD?B*T
zCp!>T#_ThJ$9UvMAHd53QM-8{h=2eO5Y)*NJGFW{w{|74$!c1b&g8u^wU=WZZaB`|
zBXQE`I3#3AwWPVc=I{dBhV8)%c3Lw6`*tRnHjd{_2<LK0!1N3{8`O@>33ULg3uhZ`
z6BVZ>Dm?rw>Xz5)<^t%fVEgvN>e0&a1ajY-$r{p=iCq~53uh<)ih_pePYPi6udKC4
z&G%2&5q7h%TBhRa_hP3V5;`O@-X{JP&}Lp`>@s-Aj9tNk8&M3}y*B0AbZSEHAibub
zi|{qewSJu5KA~{EGGfNhv#t-W+a<pkS00d&_64F{GIP2$!2n8+xKekft7li&Ni&tu
zE3L2H`}fbZe@k+^ly4`9sk(NVx7S+PaYEw*v&Wt2L5ei!;ko;T6b-aoq71Ay9x>aO
zUyh&>>WtW;dIdbkH#_`--AQ=uv>t4hOd;5V#q_XaYC2QkglTk@QV1xj<{oY2<mL)L
zl~q5Q1!8T$&XaR*<X@S+M&QxUSt%QuUjGvOQ8Zo6_Va9O-&pV^l-VL!-{+*S<IUnV
zNLp7LUwH04M2|$@I<`EIuJ2g*EDZMdSFBT4kb0IgnT=7`M({jV_cu6o8nm6ylbQEu
zz$#>z4zJ4wKe>IKvmFR4UZ-5?=)I!{+?t5s-Z&K=ZMsO*^=DE0bAWHb60v<x@3o>k
zUX8`jTMm2pH!zlwdl`Z+KfI#Z8=9G^t;|l8r2HmAW|cp-W``B&LL$dvQ0oR~B#c=U
zysoMbZ4)&-^&0u&(#qr+$eWv)nVSrR)OK?~*$j1Tn!Uf0W1kPV3#qF18!B|S9)MQa
zbhs|1>6OO5!hw(CI@}>HST_Bt4K1)tq&bWAk9+t|8T|OA$Vbt!ZL)EeFqhU=4ErK=
z`JL2u=^}QUP|)<R279C*qp809a1VOk2Y&(B1iIb4d>$bH6$07Qm7G`6Qk9beZ-B)v
z>F8TyJU^>7A2%W+V2@A#SV>5@YZMj<8A~n<n}oQ38PYTb&b;=94D2f*qW)l56<7iX
z0t@m&-=OP%TD-+w&8Lb1fu_eP<;fbxgMAZDNFP2dCy*ERK=&aDza?%2%_|K)AA&!I
z4n@bS`*n4i6%pqn{SZ3{j;H3WzB{JwQKcHtwL59gsk3S9dsu!2YhPZXm7Rm=PRtf*
zG3#hCwx9{$03`t{lqa)Y8KBnQYxk7WR;Ugt0uc&jrlV8EuaZZf`z^@5ZcJBC#!-8s
z1DN(<z=QV!B&9ady)R*b^J{2|KR!-rGEzlTO&K_^Ys+v-P0x&#QJ9?&4+pahqu}?R
z)m7xSJ3c28q-Y+jr?1B^z-3fuMpnYUd`fCPp-AgUsenhW5D|1!u(O*<jfFS*edHZ|
zZQAC#zgA~!0TLq=1oFI2<=(nqU3(b1x)RZ+M{lW^Kt!||vl~wCi*#^SI{0Gn>kimu
zQ9Z(>N2w;(EbRTwqKd`W4dN?+@zzWcEwB)0!-#|#!341FmyfS+>gW<Mc$^zG<X}J$
z8d#HXrKk@K$dCEDk2a*M-o+`O2SC5GRW!7?bAzTOv9w|zQd-w*OYyS1zB_MZiC6_t
z42{-%^|M|&*jGq%t4?!ixqh9nUg{7mIh3ZjZJxh@(pct_rokep;j#0NhwrLR3f}c+
z3sHbG8Pp;ofSb<Y2S&scRvK*y+$0M%?^nDrW8IbhMcgjY>!=RmVuiN~;WC*zkP=M|
zBbaxzqsjssF4(d6_%~V2{t|D-p-CJ7FE(|+mSg#WS89_R=dL}f-}z1YzRto9+8%Cn
z!Jy0)Wrb`TDmUzjsIK24lZsvZZ*OQT6=f-r0pcZ1m%@cv!bM#hz6W{#OR~{z?_&zx
zZJz8gzo<1%%{;s<s9E-$=x<s|aZ3q31V2~I5{{2|=U~l`G0-)|>_P_d6pYhh>1i@3
ztxf(4kiw<3%9oKdZGzEo*9pS!rIKqWX(%b{8A-#c&=k_bG__^n_$gW4Vl<Zd^kCYO
zhbIE21;#>*lhk`I4*c26lU2YQ79K4jC9PYUp7CNyof3|HYYpcbDiElrHvc2sDyAKf
z+dtrJP~m%a7G6lhLg52Y9Km8J-uaU$@T)(TR^Nl;C&A8t{w^nW^?YjIn1M$HCEgT2
zi~L-48S%n!Ir44klU0Fj`^|l0MN&&2XLoqW9pAN~JU8N`ih^VEuGr`{)Mc20L5k!2
z1+Olq(b>K+w|nx?|I5!=arm2u6Z1<2rjPd$hSIX5?ilPbCN>CfmKGhfA|lQ{3oIjY
zbWLKnkj~^R7Hg>A<NF*CEL}3CCutBVB`JAHTs<CrFX=34<@Ak-jbv&N!#;Se_gw0<
zmm-5p)#~GPT}cI;i*2b^F5}m53&&)!A}*@3>Q2SHd1tn`HB<8%kJI3&<Clf_yiLds
zDf@AQ>jcPa+*H|Hnwt9iY}@<37f6-cb$e=4NO^1IFUuZ!wD%5J=;m^IDUNI!vuspa
zdg^y~EP#>W<!LRT*dtg_HCt}YuIce+edku(>Bb%WC)NOqf@W?S^(6%7t_4N1dTGj+
z2CVE8=bdH07uN0C*X*<V2oDWBX)@@c4;ruE6<Vi~PPS&?<$IL&jm(V9-44A${sER=
zB?V<wa>5bw@Nn**#2RBh$u;M88?0;g@9x$_Q!;bP;hZ}J^lm7-(*Q+hPf9M3TB`>f
zG5v&<b4rb(Zo;S_JaL)G<QMTH5ke%oYu;vVa$%xTa>X=`9lT5WefU?C#7h4z)yn>v
z6_wFj5*eZgYnBi%h1nc7vlcgQy=fuG(Le6gjDO=WG9h~lgSJ($Ak{1A@r>m=g_Gfj
z7Ot(8_zpS*TtvB?nBYQ@J<_&tAQc(NTm~edIz+a_#d;^&ZGU;2V!2Xg`ibbGKZC|^
zIOvL2kQcpE$uHKt%sh#nPBM9oo6d4(Ld#9fEpKgjG#wB{$UqWpNvfrVQMCb<50Zn@
z^Lrizd)gwsS+Yg&)V2OZwsH&KWn)uyLRc92Yj!4;ExCZVtc@`@(P_fcmF|$H?7BHK
z#z%!;)vuKg7jUYRug}A;3QSU7=Tm6?_I`t#NKt|84*Wt(kfoEx%es5H$93I;ZO<!h
z6R@n;P~p2%ca3az*=n&d@Y8!OFJX6!-(&*uY6&xI>auDo(Azr^o<jO~U$T^F;F{P9
z*PCq!c<!L-qMDe)qG3qX#6*Na_b?{yectua9`)Da;S&CRTYUEjuo^-xl&K9#Zd5bI
zt6`Q@9ek?Zm5o2G6VEDfETy<4;16i#OJ!PWJCg~jiC`?`!u%oXdLN6}#ucb)7d`D?
z+d^)K{n3WXRU2F^%#_05+rew~SNLvLWgZZms(e*S8})bHb@%yJWEgA<zTcXoj8`%7
zDD2gnyKu%*R@UaF=2auLBNYFd8mg@L_qQY~*DBc_y7=fAwUfz*YqsNyHumsqy*-?S
ziZ)-5?@Pc;wvNyqZFfRk%jJdzQHjbDg>uOy5xVyuXYBr*pM#m!L4DHM0|yfbv|@Wz
z%Gg*A$<}Z&xO@eGBnibux||-@?=?ivi_KTC9rs@*NT)1U91eoeR8{I&v)Qsn>nh4p
z&PJFnCC`FOY{$z*9jR3t-)*)yk$ZCfr1a$MM9ThQhL|XlJ2;8;*rN-W&5nRe)inQU
zt<l{&6bDYf#0O;i#qgU1#ofHTY9G@d$h~b@1`}bmQ+gsFrVHNi67S=ey?!;<J!XJg
z*N9N5j`#w5JA+~N$c1RNC_{$}HVx8*qK{nNblh*G)kRpn$=hK(#suhUNVQnpYO`D5
zr5Q6JWVP8S)OcC5WSQ;!^k0GPWs-Y1Q~|0j7ib&g2lubWLQ3NY_UBCrV*b^5fWK(X
za%C2zv1BK*ND=d`1!I&|$!Q^|HKhr|#h*F~r)GJx-&ow%h-rH<opzydbo-|Z*^t_d
z{0591fOHE3EOp;UtWKKB{3k8N-HAR+Kix-Nk{peAg?xe2i=l*Zl`VAmfqUi(d>XqE
zt$7E`S8HFST_YsS#EK|r=P~qR5m6}~D_)!ON=sI@_)-BMb4Lf%pGg%SH4*1jj4Wu4
zse5r{jpHWy`*UXw05)R&cApww&4@OoyvNUlK+OM)J$sDe1E2*&M{RdlEV4|le$N@N
z6R`y?CQ-o0XW4J_M!rEoaAtrRF)87K5))rVc(S+*geeMFsdgI(8KjEDF~uX}gOz%-
z=G1_5B<Pjaky2P&AuGE9ymnY5??mw35rm6`4T;YTzLfZmn2ecEur>00i(j|f%OVHL
z_u|T3xs!*)Sr5a^B((JhK*sMKK%O;2>NNS>)8o%=k83?iM{jX~`QPv{gXzh`)J})4
z-1d|wQtcvk6HvQDUg3dgX6W>2ywRito$ulSU-Er&XXe!8SGfF;Pr$s%%$j?l*s8TJ
zf&XUlvP`>uuP=^I=e+ieK1F;vXW3^wL-50rTXfHUM@b43e4CMKYN%_7{R4IK2J9J6
z*@qAGvK`ixE$X@@F(ajLa6B`iA}3*BLn2%~DX~i*rkQ>ue(5A6ElvNs6zxCw2>(Mu
z_g^%w|I1;J<+J*7ap36=&ibu&3;@7238C@dRsWC1!vE8Zj)94h?LRX*og_<nWVJt|
zu8{hGh9H7f(+4-f`8`nn`3n`g8w2Dj@i76C)qeie2oM_50b}z0cUyLk2g!;8{N#D)
zh`N5{EiE{o4b9ck^K{uUgJ_rR*WmXx-A*M~&76ylrZ<OMzuAs6wl~-b5HQ5z2{3ZI
z^^tUo>8wMy2N9G}oQ=EfB2h?cf}li<L1PJo5D10F2GaO&3S%>rKxnUr<>c?%y}4s*
zNFW-mQ+%$WM3RU^*GKLV>b-?J_n3&-{BgqM^-*rYVT8B{X-V+}bO&9Pr;qK`9y3NP
z>Xivi;o+A`?+=uX`DJTR2}sDhVUM#|0uiWIizrDC$tjA!bLmfdnW$HD0Wdv^514(J
zJ0K732HEupcft6PlVB*QGgg2iCOrjP4XV%$5wNK8<C#+aO&J_SrHytT^7;#oCVDg{
z{2Mn@)OEpZ8Uh{peoPwd2zc;*!H4|y%bGnD1pS%{MaR=3kiiKf33^_n4a9G|!?x~x
zL{z-qmKP^cQ`4H4`c&jc)BLj<&R~TipT2d!*-UPZ=`-TuDZ*W1SHulrz{0!=qG`rR
zG~<F{2z;&l1LAMuB<~XZ@_lL2v{*vC#b!;>1px$2`a#?*u^TqyLpb93F3{O5V%M<I
zE+#q(4e}zVCr`SxhLf~|w{qrlPIKX^?DW01fMSPr&h_jtm5BA2BKXTwhk&fG=1XlX
z5dJZ{;G^b$4o2ox!7GK;^5nio)fh4S1x-~B%WT_YH~vDGs@umaeh01MCa21`ZeYFC
zN9ris&}_{l$=5vU4&Zj&a{ifVl(ZjWr4;3paj^JH3N@A|x?dmGM|BsWF1=@m{o_Ux
zA$$=WN}eTgA%D8o=!T;tpQDv5cDn)_tx|UqIk1c|&GGhp2ExSZRBoH`V>O`7-3NDx
z-3iSsd5^nuC{!EUr|^_o>%a^&elmB~MXBBu5~WzB;(Jj1k=@e0KGm9ZanKm#d(V8a
zCUPS(fShQT#_J9Vg?#mRF()dS9M9uj{lgrsOuk1DVU(C+Na;LSa*$=nR4i=`@APC5
z=ifn3&-8=ieR`6amJ}hApob`@KW-9l#cQkzIwY=8j!T6zzM0ZW_lU6gKrYS_Y4}%f
zaGS2)8xZ^)OO|}g$U9j%V%sxC#Ij6R4Dr5iyet4K)B0I8+Wt;_SI_rVLX@J%7z<I&
zS29LN40hGH0IGahO`7kJgbdx(+r*T~+G=yxSHSUNsF+Va&y?ua=mN)D5NjKD(_{Tp
zneD2|97)faWznFv64{KrPn7Mjn$_LW945-v_pQHeoTLtI0xZL~3O1*v(~(%FkS{H<
zZo=vH5woPSeOO6KZSZ4|8m_P+zq)dH|BPqGjeCbSmW<ukVuP31KQrrFAi!ak)!r3}
zMJC~F2RljBD9KIeL&W^qT213AY`~sQzes-Eb8%O~CMQjQ&UV)JBrk60z*(7cso?o_
zcIW1fK1N1hg6w)n$RI1$%Rx9w;gdXn2o8R(_;^lbB1JyY<+h(vrh3lIR_~yW2^Drc
zvm<Y^8iVs1-H5o^2fk?@<$X?tw7l1z_4?cxnO-}PH>I^Lv)g9+j>B!MmMab#eo<w8
z&N#t|DSNsrRRr@({70d(vL^G!Uh233`fLh_m#Fr>E%s$@`$o1lQ$%Y1I>tX|{2#wB
z`G*#hbRj}Tk&Zaw+j}0#zD!q|C**DB<p-aXnLbi2h<_MK_iQTxF?5$3H?;AaD%fDi
z*=lWh<HW&(yE>xy7l@wlFTj6y7XL?o_5X0v{};Qmr5*qvC|C#>d<ZBt6aYLP5Y2yA
z_&*+b{!b5iMkZ#)|8mHydOIPjpytk*5F%7MWw^+J??gpFg!%v;<2(HE2Rc?tBFq~W
zN+^-oh=Y{aAZ`lYNHHK=gqOsE`7X$%_?BE$g$^d_5L-fF7q8cMZ7#p+X!8l6eH-!q
zska5f=J-6x8sBoYx#@g$^?ltTBO)Y%Oqc$L*{0EAXKG#VAt$kCF`Y<wSr}eFz-~nf
z^FzQ$ACpDS4MQ*_1TBW4Q4*D=B=Yrz^?t~F?!A8C03o1qllFL>3V=FJe6tkIa;RyV
zFY(HdmnJeAaw3cfBQZPRI$+Zn3N%zTw#VEKq0U~nY3f8JD=sRk*{7zhRIsgUj8sxg
zAr`+qzPccGDb&oI<gjWNRkfZmU?Z*9N?5DT0=~suf6q!^K7+APVx2T@#OTnm%8@Z`
zuJ@oyHLa%+H(-Jh#k$<9YN!5bzL4f@mPs8xNpEygU9p*~zjU>l)mybU;p$v3TTZv;
zWMzmqw&ujKtlOlSB_Z2<e3Hmg$<2y(wi++-lx#6N_YJ@NyO9y!=02Fwtkqg8)4+9Y
z#cFEY=;YEoZzT1>r~Zoo*=2$Z9d2a65HCqhq>4}(?qq;mAMBPshVO#x2??uD{DQaw
z5ry-YK{zVNfLKA;i*}RRGS!=hhQ~lwSbxYfem(v-VK&)y@;1)mA?AQ&_+3o3UqvA_
z@G`)T8r|C^p&?8LFnO1{ZAwH%r5K<BWz}Vi|8d9dqB?oRMNQ36_q!9w?p2jV8@JJ>
z63A>Bk@;~s$gP6?`rv~L7;pJw2H-+OYV<rpprf{XVKy(iRav%`sT4t;1yr=TGihe*
z1{eMNlLB39a4&HGv;VA;pp8E1lgU|&Sx|A=7GWYQl+uc1PrhBKJKTr&dtvE(rqd2p
zi01*9ZWXFJSf{DByFE7FM%TB161jRK;)Gf>)`tyE@vi(w7qh?HO*C*9v`}AZEIK!h
z{37qH-~SpJ4xn3I{9`i4QULRu`EmV2$&2VvtllV-;<GJAbP<&pA4NzdUt^d0?O8jo
zco4j-TKjnZfWRZm>IM}IqX~#sK>@gvnXYQP?(pGg6^*aQN#E(!=3JZI8)zbcQNfol
zt}B7aokTD8uvw8RJLl&nDANDiNUYNH4<+kY&_SF-Klc;6$JJ@6K6Mt)i#1u}pBbh6
z!J|^x)7h_J-baDI6`|n)wbD9S3E{y9hZXuV#(?~6<3KXOs>plYFkLbpzylD0Ct@*Y
zL4<xbi6WeT3vtTL^!8*L^otzQ!X3c(<?n3d!B%BN!!m~(D<UT;btZFty%E<s#rK)j
z&YriukR6d>i{om%yLdnUr05Z?J4xQb-ymyCLtzKR8yT{p(gY+5!~;GoKC*1*m3>&(
zq}p6sTo@L_Sp$*AfRpNd2s^E^92z|zMVGHwD#H-qA{(mU#Q0ker+1md-?celzAQki
z)k-gCjq&X3?NF0XA0Eg&zjP4td+T=D#Ci(ZdtGG9pvSForP_dGow|PTUkf;25_;gA
zPte&?8cA|{HFEVxvTJHUOLljIgP+?5BC~lH>jY;{g=^X-mp;UQJ6{{)b`(!;&gQ`g
zm$?|5GXXkhx#_&ltiOhYjLd*7rBCTRMM@@?$bl()*gItS7VW3ic81@#thDnLlKDE<
zcEN9j%HzVa#7V;{<c8V8R1k~v6LxvU3Pb2kQNrj?`G^XcZ2&9kLxc_@Fe*a<8FW{S
zZp3F=HfK0+Rdsjc;B*6IqAjyieF`BOz_m<5%KDk`>XAu)1m@VC+*&Xn61WFg4Xn4_
z?rrEj<&N6wbhF?xWpa|S!dHA`v3LMvM<k-yc_5Wi%`p6np;M4aH>}(jjZMqwv?j9x
zB3XlZX3c3wjX8_vxF#K+c-G$3jFeLMZmG)Vne2iCIA2A6&GOo7^tkx52y<OiPDd`i
zy@=kvayI-uEp%a`;99UdOOM?0mhdUEJ=SJ@Lm7oW=2i(T#H)Jo>H#yxf4ew4ygFBI
zn)a2Q?u<tp&9pPIPmU>i{PTGq3kiTz2Kx=)Y0{(>b)u2Drqg1;+OoRbjrXR~Mn5R|
zN4z8Gr4@<o0M`+CdACfB@PhP`L0Fh8@ssb6|LynzFX8+&Jk$XQ^!q{hIH1{;ij<xP
zm+ud?nw_+dw?nB^J`Zz@Fu@Z1qQE;R_JDDo#1V-$y@ZcMH~9u;s$(&i+LEyh;IejW
zuW&KCcN@6RoiTZypI!Gsp>Qcutihx<{Pq|_U7-^YAmeT3te3@C`VbA=PNT(@XupC5
zY26sc!wC2tywAId&r3DGGOc&J$v;~MaRQUdLFy2EiqU4p9<jmpMX7izW^Obo2l7<~
z;**xlgNn9cBKAh{c{mt|A>9BY{p!K4=&w)|Sjhs}e_*rl0?PRg<Rga8D^`R|Q@VZ}
z_GhyDD)ZLpAvF>_=8bFU>6KkuVr{_=&B&vj((PmWkbo_J&k$$%6N(IMvVO^kD*%#B
zEFGO)T3DWgG7skQ(V~8+A!0$wtqZ_LDsBqeTCy<dYbAPq%Z3+G<?5|@u66gv@7q^R
zY7MXjlMUl=5jQr|sMZospmGUnn~I(w%#<y+X)`{kG}Bv^bKOSotr3|tQMsQ)yXN&Y
zdR*1M+i;q2Y58mlSh}W1h$*uI)^&H!ZalHlMrKr;=IzXH`DTL8?J!rjnFz-ALfVi=
z%4It?mo>Vk0bTAor5?_mUegm{rUD_M%BP4k7|HiB*C$-f{)u1I9|H)b<?w1Adh>yg
zzA8H|SIOjj9etl*^!4cUOH4!TL{KB1_xqIqo)l-vd2&E;3B&M=6PDUfRZ%imA9?Zo
zU?ac_&jG)`%d)cgdVW9j@@ks#<kMR=P<u2mMR>U*3NRnV0@<^v0_T+@)TwxjG507R
z0V9YXsoN-=AFQSQl!<mQl638HYi3K}k2_!SjGV-q(V!jx)+40WdV3dQJUP>6RnVho
z#60Z!b@-7Tq@17q7oR#=MrqmTadtuG79B!cgI^bJmD!wCnb7uvx=M>Xa>Jo?Aw@Jf
z;C1)IPKz$L2F?s(->SfQFUw!z^hAxoO0#ZMXJBR^rm-z6QHxX-wY`<gyEG{vn$<1J
zkuQ>BBdQsCD}KNwAS!eJn=2T@fAqrq4|lu&cc0Ck9>UTe`Lu)mx;_y=0OSQ=<o~@k
z+y7%H|6e+CIynOiB`0e-S$qaowtr@F2S+D-24)tH|3OmDLeKhN7M|ALbifiz@XoD0
zYb9|F*mmxfX8%JFkuqIxG@iuL0u~4!W~H!307whqe*ONzQ(6JY>}L~}H<YacaVDi=
zY?95LoV~1)M<q;fue(nKZTv?T`r3fNS-*g$=%3O(xG9*eCawlz=MPkKkbGH;;SkG(
z#4&U*rsAQ`-aHznU~>d$M3H@4NnRMy#b+M0ax(9PV4+V^J$-?^VhE5-au9uC{=@(R
zw2CZhNP!Hwos`HebYz8MpJ150*(r=NxdywM8E_@eL^X1-SbHOsj-S~jH1l$QBw?7K
z7;^xHXr6<g<hcqWwQ`{n!Jm=0FkZBgd?rBB4?U<fE5az@)#|y1FnS3nC%80q47UTl
zBL3F<dS)FUBg=ULNa1QzV|1%P{#C$I*f<jX2IJl}S2FAi{_?~*e3oziE&t?2$hEK^
z3w*fz8O50UGmb8D8vHw_nC8B|{N&)iupIH1ziUT)S#$k@KzQU)m|}kpb5hUTor2Zu
zpuUciQADEH!Nfx;Yx|Mf{*YZEeH}K-83Bw0s0x<ld$K6V@%Xz9$_~8mBrw!%Ye`My
zHNb*Ei^!5b><#s4Kg@tSASd%sijF-RQ6N0!aD(=GA%K8{fDr;?zVEpK^%FdRR2LMW
zxeeS%DaZI{`+Wu%aqXGvOJhO_xS@xE`R640G0JxllNA+Ca?27n2eL#+(p->pYTgkq
zKw}OSO!lqi(bxw-$v^;_(gz`wfCxbd9P2^kAL6_T0x`G2NlM7e>BQj@jzYcRK$bZj
z<>^sD`}&_CZO0Ip$38KTc*NK(#B?Q<6(K3<2Nq8@8F{X(;6@!m(hyn(kpz2l3UP_G
zsCxuv1PN^~!99^HlKAN$up_Q%$+5)qgXid}2QkCJ7&98o-J{E}0}x0kG5lH6XXFP2
zP9A`$5eI3Ys7ogT@jrraqNrDdno4Ad#>TJdGj{_^qZBAW3{V!LLP&IF7TD)sM%}`v
z!U$C-S}-Io^dAQmO9tzfkThI#Hn7B=*T->*&0mB98r?TS38^>|XWFCWuu4Qb=hq9>
za>5s=b3*$7036>&_q>D5ij;zbK4szJ=O=74z+fKtx9utS`8lkq$&n>p6nQ>b`~<E2
z8%6e8C0p89>A{KsabckDt(&bj3&yC+Ez-!g?7Em+y1YK$)uSC%n)ploe4j|R3|a#w
zkQ%Z-khx&rH6rifHZP(;X&WM!^L1`{_Wk5jyiu(@stzV<OC|+S8bm=On)O)YMCv6n
zO>?YYj2+k+6;Qt<g!6tR6^#yxAc=T*Se9y_*2_8v&R8Z-_(b@TyuzE{(p;$%o*+jS
zfX85x<x-6BfVf3F9kUxuMI_c4Q!F0KL9f80xVkO7Y7O#;_{xHvGwlyNWO%+ylJo2b
z7%#Lbk68jepW>#;{E;h128|nM&;KFvjIF?&OOJiTB#M?J@Nl{Q^dQrXu3S5tqsF(|
zOIPu@Iq&>UjVJrK-xcre`n_-C@{OA;7AeA$sciaedpP-V(7C9s|DCL#M`QZ@<-ppb
zC2e~CFgkCo)6K)<>H0F5o7?T>j$3)kR(R>IuXuib{`A@pi91K;wIR-{+)X0*IO);A
zZ^f|OIKTb+{Psaq6Hr5*O+tnwavC{5{q}Hv^ieVbtJJNr)aomQr0~7bM6anj%Nt{I
zcBjBnfx2wF{EJOCJUePuh$;WWTn9z~TAoD?+XKjzz%<OYlPZ3XaPdYX$&FP&8UL%P
zL3pfm9@Fx+vPYiNJWzl#0%hTurD0AmxfaGM?yxqZl=8wjp0I?balw$YWd}TtlItd>
zT3YCyZDU)|4#<Igd31ko)uT5XpQS=4I?chDc9D9&d~lc*&UX3D?ahUuD_d((G!BcJ
z`*`eHI!o)*?e_w`dMr8LmLRtc-l51GBb+TO_EVW7qxE;m<XohvrML?(1gm>32*G@j
zG>Ei{VCWr5pfftJlPuu0l+XM1hQg`WeU%Nfz&K>^Y5fx=$nWuPUeANStJ3uF97Suj
z@uLFv)IuJgkMjC^Z|iCSHD@y^iAI!mo_W)Y`Stj)vGlao$4geu#lKuFpG1bVCoI<6
zAJjz6nWe_2Qi7$rB(bBT@??{mg(^j}977lPDO<bYKDsNH0pzsfea5v5jpWq0=O{|*
zp4Y)Jbpm@LuAR_ey#)6_K|fw?Y)O-#T`zJ>Lm?$`(|G)Svw<b?N7SPxwTA)Ru6U#E
zIN{F9p_8g}+oA)yA52lu@dU$(gW+%T$;LS@4E?mTdU|tB`Xt9CGsNtPPt;D{GHL#F
z4}ib>Sj6Hc;y@GO&_jLqskshMY!-A8lXOs(8SZ{>kH@d>82s7?Dm><I9NDMhx8|j!
z&}4~<tp9m!b0#Z~vSDah&Czi_g_JL6)5^<n_&k9`?%tJC`cro|`{`*}6>6hg;v`~X
z4{}iUyAx}UcIvOUXS3{YoF@b%BUKa|cnQ({-cizdI85&KO@_$i4KJLSrmVKdOMV2W
zuyg$V<+D}q0A2FV_)~3KHr?~p!M*eL<FL_hoBG!DucC_DqCH2pj+$c9#-{vqF)L*j
zMDcw*Z_e0tdaj5lF0ifxzAK#9@IF0?<FDvo)KF8Vtv!ti%tmRcNyC!Jl*09Ix-n{Q
z-qLvQN?b5ftS5}T^6{a3ifS=rSnU3;8EPpTr$)fK<KB>D(m?s2m8Sbxv-Cd#%7mIk
z4E4~06#BuBRD%f8QqX}-OeCu?Fbg1|Zk52CBwKZ-9NQR7xST<{QMRLz?|Wxf9_|k&
z3>h>ru`}_ODGzSy44vEL5fPLrC<YUPEk{L+%Mc8f1k>MUt4L4wPM1`U*xtF^o?cEI
zf0vG{e(WB+dD5fl_$c$YQvS7#A%o~RPcx_?dq8b7sG++!*H-+ELQQ2mgvT>_$~T_a
z#Da=rX!%&*4v_-HnpiE4FE(jxB1waW>r83xTDG2sKdgMmC#~21*!QR6dp~=5enyKk
zbm00h<;kQXFpvUt81s1W;p_o5a^cpPDpAW_;3|ySv6b47m=Z<%47Lw8S?z&8yp(I~
z=)&HvM_O}m0!Q(ZYFF%03Z}a)XI|w#nV9%c#CC)<Ra2l%j`nIs?WNuenAMEpJn5Ld
z$VlyxtFQ^$pvnJGKk3h;({Dq%l=7TX$r0jiQA|+pGwwceHP0{{1TX%8agK)RH6{m(
zV6>e4Vo=NUFw)Db(xtE&2Ru&Jodo20&?<C+F7Gg_tHIxV&<bm@9Q%?jcZQd@Y<7kR
zu<TY#wLMqVsqho_!CLGH{d<hYc-^91a0%`$B;k_du&&WzUtp7!wrl>RQKi-3725Ad
z(0i-C3j}2XP4hHS=55i}Jbj#?Bg0S7>$c`INE^M~4(pR%v)9zgJb%C7Rqa{Y!mM9x
z=@{>SnXN-oM?A~KYY3^{gWla+zWHajdXQxaj)r<{Hn<pucK*ioQd8?h%m*b7b!BmW
z;$#gst-ZpmqO3&1XEp_OC$l0jj*Pvx=aXw;JVAoQfs@|z;rzs8O(m*4Pa9>kjZ#+1
z`CO&Z@q1Wy5d3>sTUWE$#iDYJQ=@`sc!UV#Ni)W)$~t4+Vmg_JyD@d(TD4xx?`9Z;
z33Sskg%!XlS1*{jrxl&UCY;P%>B{vZiUJU5s>WERG>GQ8!ns7CliiK>@|mzCeylda
zTfJ3v+yP7TpN7V_eqE~y-fy!{N%W=F$@s~;c*P*0)Ke-x<<k4let|Zf=b}_1?KIA}
z%4ejjxNyqanz4r1Ws7kx;*N^dE7~pE%`5clOPvo^WViw!au%hKZ&Uq7^2AlY@)NwK
zrB;2Bq4LE_-Zgr1KJz4fsF|1QEZSq!w0U<A{ok#nU9U1eFN6V10_{HVN3$cxfqTyD
z#PLxlc-<Q$yd_u6+*di?-qgKkvyo-co7p^nJ_D1qmb<M<C^}3kW!{cT?Ubs9_nuYw
zMp|2zHOAZLsvjC_7K?R@oA5Q2!p{2W7Va_HWm`3Ps}Vr3imJv45IHO5m1v9(|GNfb
zOad57GvqDmRyp2A9ugX$D!&8DKBTgX^(SPt#HSAEg6NX#b3t&Aa_FlB+ciDR*w&7Y
zgR*Js&$dhNlvAu)n&WG`98R4ddHD5a;NzHTxtsb{1JF?G!F@b9u?rty5ynR$Cix#e
zAw<qRv$5-s(e>^w+DwN>dAA%d8^@W1=bCAcU6bw&6YoW~^~Kf^6e+G$hU?1?Jwv(8
z2Ka_;@Xc*-EG=kP8LoL7GR<qFjflCfEAx{tnTuVcpXsdp@!v;+S)j7Lf*d)%Y1&Ko
zcwNLJ*TPx`2^~XBEbm0nn$y4&SP>t8PB|=2wU?UV<MkPyq4k%h#%12ER&URg?h0Pp
zJ9E$>YWnF|M)K(+Geg`K*1Y@W8o<9cn0vymo{qaOm1s!7tH<bQH&kP`9HQ3O7H{12
z>7F7!1Ak^*UZ!3N3)x^Fw#VH5ec933-D~UV%Fe>r9TBCo--!rOFbmXZEO2F0d%MnC
z_Uy$IyB!DaIfncdw|g7B)}Kj!TTUE7v$GcHeu%~8aP<$Cd!xugM}b*}g-f^H+GMNV
zT~#|mWK)&zO?#cJdlTrzS}7KJZuP#^GX?)=6U4C7tEYq1ZgP6a<-KoWTwhb9{liIP
zHpZv-!`jRBdH?1008dPk+y?UXmz{U!0%3YqH1gx@icVV$K~>cJ4(5?e@C*D0&vf$d
z_Htq=#S?rA7aQu}(r@ZwWrqh>t**D`Ruq{ZQMZt!=e35dyXQN=SX#B7EIlt4jVRmV
zWtT2g{32X4jyhSWD7<u-(BWb-3NV#D{Z^5x*>wmjTosB?sZzTmu<`a>VMdBj?L#t#
z@GM^TTmGy;fKg!pY@mc|0hU>92m?%%q7Vg42)S>SfgP?AEX6ZF)U-8!8^t!j4Z378
z*i0EgX)i95K)>@~zNjT)+XUy5|Fc`GJTFZ2z?7uDEiv8VC{b2JNjNZgN_|)}k40mq
z+lJZCSTW+jwwUt1@`?$LJU_~i6IA2Zv;Uq?YafY95Lf0C63{sV=c3KPQj}r<dQnES
z-Gv-ddlvYd5PF?Gh$a*`w6r|BO3k0}Xmw}0eu1qyJ}7mKt+PXh>_;&vAl)2(WU;Xn
z#-l={KoX!7KBGCz*FF$Tn7BXNpdzFr{sP2<;92tH(r@^1DJ4?L9wVO%B?jkg(kcup
zX1qdViX^EbBFRldRnnTEvyYfh4R8a&s_(+?kbFhR(QzYiZqTB07I83s4<LDa3gS6?
z18lHBX2Es`K$j!Nm*U*=RsKb+DP&#1x1xV6A=MzJk(uaYB5M-Ra1DccNgdko0}w@l
zK|$!=!~_2ggE8|Hvd}{rH0?o(rmAuSj)bWSaQyAV;Dt+e$r;Swi<5_|rV#a4f+P+O
z7vhDKJuWE2&UwNm18Z}O7N9=jw`|157e&MFHTO#cdYob=jBDO++PiIgyI(2M#75*T
z(nL*x=4-l=uo(r6<%9!uv#Y}h1fw?C_t*;^!$SxrG<Ttl_F{(S43{k>2q*1SF;dfj
zC(5-6u`3|*vPdHIf5ESm)FjT@(}n{h#X;l=kn*-RKo^UlN!gDA@9kRrp`WnUF43ge
zuvTeuDI_e!gOfcOl!OA-4;*a)v30cxZcfM6y-uj34A(Hw-t8yHJt{oESdxUepAcKb
z)^VI7mq?yI{iN;J&;CMd)PXo`qHuVEjbf${o=nR|Y((ZJZw2EqY!Ba`GeKtbqFTd@
zbI<@I<sno}V|IfPvZ?bII1u9<iT-)Mit=*t8L>nSdj=cV_4&Cuj;;OuIylapH=-RL
zKYu!jP}%)r&|PXhO@v=e`p}L&ja<hQldCPcy*#`?adnI$TNOK16IUOJ<crBazPTys
zG5#TBr`g=Hq{!^Lx=IxURrpTu*>83=v}^2TgJ^9pWV_u@-e_+5Zdh2{ty%P$9{6|}
z(T}-A9>$A$e>#*73aFOpcW-1Uh-lg(_yQ+w+{G74lRkZBGkB*9i}sbvz%9w1%!Z(4
zgRy*Sl7R9zkH)5Px~AJ*>-oUA&7v0Id++&hap33<AgWmDhFM*P0Fz#1rakq#{H|QM
zWy#MZ`}dQ+KlrB@J2LR#%IN!Jcd$5Z_JXc`zo{LrJAi`bS#DsENy(`|@}0F*y366J
zH2mB!bfYgp=VM(G1Y06(51{mMgg24WHi_!XuC?>^d~iD}_xlCZ{vEXbZuQ{M^!65(
zWnN-rIV1c{lEqLE%jCA#!X79VODP*i1X&YY@kI3`q+W<*=|pGA8)z8G<*v4#!F2*~
zbG7O~9{4bmCIb~bNa6HzJ&9(!(+rWV;0<%LW_|NUw;yxu%bY$J!<LqF;x-6_bUy;u
z>ieYxbu_5}z8N<b2|k>|OTPJJpZNp{K36LCmEWackji!s*!6zidCK<%VEI>;Y*EZx
z_BcAha)+GYb{I=qWu;3&(wV1J-E<m@fYpr%(c#S%>Ft#egj6RRaP&4`@~Tu4-ovtj
zoM1y)K@dkU+2%e}arlb2IDkEilCR6-=VkL$>L6saFapUVlTtV7=i>(fc?nDfTL|Wt
zS>y?a;AB!UgZFJin;}!1gZT9t(bw*S%C(&k{&eDe_iwjbr_oL1aN$3I(cPwUyY9>7
zX0NyMLand&W&~y1MKondILz;n)Q2H~{p>%WR65ynXbLXKrq%kGZ!$OW?{$g~y~Y-;
zM#}xQFW|O7sm=c$6f!a}{J)|?bwy(nItgJdZgys4Hhm5Q1~z&Ib~aWPLt`U-Lw0=*
zCS!UNBV!{rdfxy4FLdY>ZEc;P{?`(UbP_fuw)p?6aY>O*$=t&jpOJx)PECsep9!B)
z=YI+!8UM@BAIp-2m5e@m=>6+AjCJ!8x02!Wf}-Q}9~xSf#cFp|XBifyrVT}lGS6jY
zCYpKzbLutqx`sNve$jyiRb)gE=b6M;8z1)vK3yA!#77*&_!AQ{Br?-_O!x!iuAqXN
zBJrS}5H<OEeQ2xj>N=K@eb}y;*87qO;!WF;LjvhO*0iu|dI=H+gVW6lkQ^=7KBMF(
z)0QT+@o&>@46X_gCF}DdMf8T75FO=zTuLXEaPH@y+9jW+W89s7kFHwH=Y<@X(ZM$e
zEak%FmnQI(vK5v8Yqu{6_piEUe(Mr<e{%j+^{3muA2hlu-k<xjTRkNH`0YtePgmuC
z&VAWFb(Q?5oR|ITq4wrCjTlxJ2=E;Vk?C;`*`Idfr%6oHWy1zz1L^cf0drb|3#IrD
zFWtc^o@ABwAZJeVVMG4#B)h%?t~(^c6V=uo+`5C${gFmYzjT2~-Vxs&JjWlz%xQO&
z>Hb{E*j6CZcbqAvPuqYyJX!DA1C~2ot~+?ulO(tcjr|Utv|!!-xb9B-?n3r=M`Cw~
zcNg-{JE&a2;Qdh|>_Mi@(qFnS?y~0odis0smG8%^@6Yeux8MHfzmuE3FA4klUix+M
zD<Q2l>#P1A{k3P;mhfw(ub4JQS31U)UHCQa`ujE6S6ZdkykDg)@VfieA<Jw0d-~>f
zuFyWWrS3)Pn|rQ%^&=+zmtEc2^@siNrW`CwC5lTDi%KerQq#DAW89`@mRzc;uKsRZ
E0CFexxc~qF

literal 0
HcmV?d00001

diff --git a/docs/404.html b/docs/404.html
index 71cb3a24b..7f4b1223b 100644
--- a/docs/404.html
+++ b/docs/404.html
@@ -837,8 +837,35 @@ <h1 class="title">Page Not Found</h1>
 });
 </script>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/404.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/404.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/bib.html b/docs/cmdstan-guide/bib.html
index cda8b7cf5..5bed838fb 100644
--- a/docs/cmdstan-guide/bib.html
+++ b/docs/cmdstan-guide/bib.html
@@ -787,8 +787,35 @@ <h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">Re
 });
 </script>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/bib.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/bib.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/command_line_options.html b/docs/cmdstan-guide/command_line_options.html
index 81addd6f9..e821bbec5 100644
--- a/docs/cmdstan-guide/command_line_options.html
+++ b/docs/cmdstan-guide/command_line_options.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -464,7 +464,9 @@ <h1>Command-Line Interface Overview</h1>
 <ul>
 <li><code>sample</code> - obtain a sample (set of draws) from the posterior using HMC</li>
 <li><code>optimize</code> - penalized maximum likelihood estimation</li>
-<li><code>variational</code> - automatic variational inference</li>
+<li><code>pathfinder</code> - quasi-Newton variational inference</li>
+<li><code>variational</code> - automatic differentation variational inference (ADVI)</li>
+<li><code>laplace</code> - sample from a normal approximation centered at the mode</li>
 <li><code>generate_quantities</code> - run model’s <code>generated quantities</code> block on existing sample to obtain new quantities of interest.</li>
 <li><code>log_prob</code> - compute the log probability and gradient of the model for one set of parameters.</li>
 <li><code>diagnose</code> - compute and compare sampler gradient calculations to finite differences.</li>
@@ -1115,8 +1117,35 @@ <h2 class="anchored" data-anchor-id="error-messages-and-return-codes">Error mess
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/command_line_options.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/command_line_options.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/compiling_stan_programs.html b/docs/cmdstan-guide/compiling_stan_programs.html
index d4b8592ce..2b151827e 100644
--- a/docs/cmdstan-guide/compiling_stan_programs.html
+++ b/docs/cmdstan-guide/compiling_stan_programs.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1157,8 +1157,35 @@ <h3 class="anchored" data-anchor-id="optimizing-by-ignoring-range-checks">Optimi
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/compiling_stan_programs.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/compiling_stan_programs.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/diagnose_config.html b/docs/cmdstan-guide/diagnose_config.html
index 7fb35fc53..e590ceb56 100644
--- a/docs/cmdstan-guide/diagnose_config.html
+++ b/docs/cmdstan-guide/diagnose_config.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1068,8 +1068,35 @@ <h1>Diagnosing HMC by Comparison of Gradients</h1>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/diagnose_utility.html b/docs/cmdstan-guide/diagnose_utility.html
index b23b61ea8..804b2d293 100644
--- a/docs/cmdstan-guide/diagnose_utility.html
+++ b/docs/cmdstan-guide/diagnose_utility.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1192,8 +1192,35 @@ <h3 class="anchored" data-anchor-id="high-hatr">High <span class="math inline">\
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_utility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/diagnose_utility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/example_model_data.html b/docs/cmdstan-guide/example_model_data.html
index 6e9039a47..7bd1e760e 100644
--- a/docs/cmdstan-guide/example_model_data.html
+++ b/docs/cmdstan-guide/example_model_data.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1057,8 +1057,35 @@ <h1>Example Model and Data</h1>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/example_model_data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/example_model_data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/external_code.html b/docs/cmdstan-guide/external_code.html
index 0fd42c59f..591f51b58 100644
--- a/docs/cmdstan-guide/external_code.html
+++ b/docs/cmdstan-guide/external_code.html
@@ -259,7 +259,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1163,8 +1163,35 @@ <h2 class="anchored" data-anchor-id="special-functions-rngs-distributions-editin
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/external_code.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/external_code.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/generate_quantities_config.html b/docs/cmdstan-guide/generate_quantities_config.html
index 9f631fb8f..d2c731c4a 100644
--- a/docs/cmdstan-guide/generate_quantities_config.html
+++ b/docs/cmdstan-guide/generate_quantities_config.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1125,8 +1125,35 @@ <h2 class="anchored" data-anchor-id="errors">Errors</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/generate_quantities_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/generate_quantities_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/index.html b/docs/cmdstan-guide/index.html
index fe5f177f6..db95069e3 100644
--- a/docs/cmdstan-guide/index.html
+++ b/docs/cmdstan-guide/index.html
@@ -74,7 +74,7 @@
 
 <link rel="stylesheet" href="../quarto-config/quarto_styles.css">
 <meta property="og:title" content="CmdStan User’s Guide – Stan Docs">
-<meta property="og:description" content="Version 2.38">
+<meta property="og:description" content="Version 2.39">
 <meta property="og:image" content="img/logo_tm.png">
 <meta property="og:site_name" content="Stan Docs">
 </head>
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -410,7 +410,7 @@
 <header id="title-block-header" class="quarto-title-block default">
 <div class="quarto-title">
 <h1 class="title">CmdStan User’s Guide</h1>
-<p class="subtitle lead">Version 2.38</p>
+<p class="subtitle lead">Version 2.39</p>
 </div>
 
 
@@ -436,7 +436,7 @@ <h1 class="title">CmdStan User’s Guide</h1>
 <li><p>generate data from the model according to a given set of parameters.</p></li>
 </ul>
 <p>The packages CmdStanR and CmdStanPy provide interfaces to CmdStan from R and Python, respectively, similarly, JuliaStan also interfaces with CmdStan.</p>
-<p><a href="https://mc-stan.org/docs/2_38/cmdstan-guide-2_38.pdf">Download the pdf version of this manual</a>.</p>
+<p><a href="https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf">Download the pdf version of this manual</a>.</p>
 <section id="benefits-of-cmdstan" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="benefits-of-cmdstan">Benefits of CmdStan</h4>
 <ul>
@@ -1043,8 +1043,35 @@ <h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/installation.html b/docs/cmdstan-guide/installation.html
index 74fa8aa25..abdc6b9c4 100644
--- a/docs/cmdstan-guide/installation.html
+++ b/docs/cmdstan-guide/installation.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1323,8 +1323,35 @@ <h2 class="anchored" data-anchor-id="gnu-make">Using GNU Make</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/installation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/installation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/json_apdx.html b/docs/cmdstan-guide/json_apdx.html
index 5b2b06ab7..f958afc7a 100644
--- a/docs/cmdstan-guide/json_apdx.html
+++ b/docs/cmdstan-guide/json_apdx.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1239,8 +1239,35 @@ <h3 class="anchored" data-anchor-id="empty-arrays-in-json">Empty arrays in JSON<
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/json_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/json_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/laplace_sample_config.html b/docs/cmdstan-guide/laplace_sample_config.html
index 6d639198d..65f2b4133 100644
--- a/docs/cmdstan-guide/laplace_sample_config.html
+++ b/docs/cmdstan-guide/laplace_sample_config.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1100,8 +1100,35 @@ <h2 class="anchored" data-anchor-id="example">Example</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/laplace_sample_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/laplace_sample_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/log_prob_config.html b/docs/cmdstan-guide/log_prob_config.html
index 01ea7fdcc..b60d63e64 100644
--- a/docs/cmdstan-guide/log_prob_config.html
+++ b/docs/cmdstan-guide/log_prob_config.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1087,8 +1087,35 @@ <h2 class="anchored" data-anchor-id="csv-output">CSV output</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/log_prob_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/log_prob_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/mcmc_config.html b/docs/cmdstan-guide/mcmc_config.html
index dd4f21874..d418b1d8a 100644
--- a/docs/cmdstan-guide/mcmc_config.html
+++ b/docs/cmdstan-guide/mcmc_config.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1563,8 +1563,35 @@ <h3 class="anchored" data-anchor-id="everything-example">Everything example</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/mcmc_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/mcmc_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/optimize_config.html b/docs/cmdstan-guide/optimize_config.html
index a16b09626..56a348a61 100644
--- a/docs/cmdstan-guide/optimize_config.html
+++ b/docs/cmdstan-guide/optimize_config.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -439,6 +439,7 @@ <h2 id="toc-title">On this page</h2>
   <ul>
   <li><a href="#optimization" id="toc-optimization" class="nav-link active" data-scroll-target="#optimization">Optimization</a>
   <ul class="collapse">
+  <li><a href="#meaning-of-the-converged__-column" id="toc-meaning-of-the-converged__-column" class="nav-link" data-scroll-target="#meaning-of-the-converged__-column">Meaning of the <code>converged__</code> column</a></li>
   <li><a href="#jacobian-adjustments" id="toc-jacobian-adjustments" class="nav-link" data-scroll-target="#jacobian-adjustments">Jacobian adjustments</a></li>
   <li><a href="#optimization-algorithms" id="toc-optimization-algorithms" class="nav-link" data-scroll-target="#optimization-algorithms">Optimization algorithms</a></li>
   <li><a href="#the-quasi-newton-optimizers" id="toc-the-quasi-newton-optimizers" class="nav-link" data-scroll-target="#the-quasi-newton-optimizers">The quasi-Newton optimizers</a></li>
@@ -516,20 +517,72 @@ <h1>Optimization</h1>
 #     iter = 2000 (Default)
 #     save_iterations = false (Default)</code></pre>
 <p>Following the config information are two lines of output, the CSV headers and the recorded values:</p>
-<pre><code>lp__,theta
--5.00402,0.200003</code></pre>
-<p>Note that everything is a comment other than a line for the header, and a line for the values. Here, the header indicates the unnormalized log probability with <code>lp__</code> and the model parameter <code>theta</code>. The maximum log probability is -5.0 and the posterior mode for <code>theta</code> is 0.20. The mode exactly matches what we would expect from the data. Because the prior was uniform, the result 0.20 represents the maximum likelihood estimate (MLE) for the very simple Bernoulli model. Note that no uncertainty is reported.</p>
+<pre><code>lp__,converged__,theta
+-5.00402,31,0.200003</code></pre>
+<p>Note that everything is a comment other than a line for the header, and a line for the values. Here, the header indicates the unnormalized log probability with <code>lp__</code>, algorithm status in <code>converged__</code>, and the model parameter <code>theta</code>. The maximum log probability is -5.0 and the posterior mode for <code>theta</code> is 0.20. The mode exactly matches what we would expect from the data. Because the prior was uniform, the result 0.20 represents the maximum likelihood estimate (MLE) for the very simple Bernoulli model. Note that no uncertainty is reported.</p>
 <p>All of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument <code>save_iterations</code> specifies whether or not to save the intermediate iterations to the output file. Allowed values are <code>true</code> or <code>false</code>. The default value is <code>false</code>, i.e., intermediate iterations are not saved to the output file. Running the optimizer with <code>save_iterations=true</code> writes both the initial log joint probability and values for all iterations to the output CSV file.</p>
 <p>Running the example model with option <code>save_iterations=true</code>, i.e., the command</p>
 <pre><code>&gt; ./bernoulli optimize save_iterations=1 data file=bernoulli.data.json</code></pre>
 <p>produces CSV file output rows:</p>
-<pre><code>lp__,theta
--6.85653,0.493689
--6.10128,0.420936
--5.02953,0.22956
--5.00517,0.206107
--5.00403,0.200299
--5.00402,0.200003</code></pre>
+<pre><code>lp__,converged__,theta
+-6.85653,0,0.493689
+-6.10128,0,0.420936
+-5.02953,0,0.22956
+-5.00517,0,0.206107
+-5.00403,0,0.200299
+-5.00402,31,0.200003</code></pre>
+<section id="meaning-of-the-converged__-column" class="level2">
+<h2 class="anchored" data-anchor-id="meaning-of-the-converged__-column">Meaning of the <code>converged__</code> column</h2>
+<p>The <code>converged__</code> column is used to indicate the status of the algorithm. It can take on the following values.</p>
+<table class="caption-top table">
+<colgroup>
+<col style="width: 15%">
+<col style="width: 84%">
+</colgroup>
+<thead>
+<tr class="header">
+<th>Numeric value</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td>-1</td>
+<td>Line search failed to achieve a sufficient decrease, no more progress can be made</td>
+</tr>
+<tr class="even">
+<td>0</td>
+<td>Successful step completed</td>
+</tr>
+<tr class="odd">
+<td>10</td>
+<td>Convergence detected: absolute parameter change was below tolerance</td>
+</tr>
+<tr class="even">
+<td>20</td>
+<td>Convergence detected: absolute change in objective function was below tolerance</td>
+</tr>
+<tr class="odd">
+<td>21</td>
+<td>Convergence detected: relative change in objective function was below tolerance</td>
+</tr>
+<tr class="even">
+<td>30</td>
+<td>Convergence detected: gradient norm is below tolerance</td>
+</tr>
+<tr class="odd">
+<td>31</td>
+<td>Convergence detected: relative gradient magnitude is below tolerance</td>
+</tr>
+<tr class="even">
+<td>40</td>
+<td>Maximum number of iterations hit, may not be at an optima</td>
+</tr>
+</tbody>
+</table>
+<p>Note that not all algorithms can produce all codes, and some, such as <code>0</code>, will only be observed if <code>save_iterations</code> is used.</p>
+<p>A comment explaining the final value (with the same contents as the above table) is added at the end of the CSV file.</p>
+</section>
 <section id="jacobian-adjustments" class="level2">
 <h2 class="anchored" data-anchor-id="jacobian-adjustments">Jacobian adjustments</h2>
 <p>If there are constrained parameters, Stan makes <a href="https://mc-stan.org/docs/reference-manual/transforms.html">a transformation to an unconstrained space</a> and runs the optimization algorithm in the unconstrained space.</p>
@@ -1147,8 +1200,35 @@ <h2 class="anchored" data-anchor-id="the-newton-optimizer">The Newton optimizer<
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/optimize_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/optimize_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/parallelization.html b/docs/cmdstan-guide/parallelization.html
index ba2d09864..a6c4315fc 100644
--- a/docs/cmdstan-guide/parallelization.html
+++ b/docs/cmdstan-guide/parallelization.html
@@ -226,7 +226,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1128,8 +1128,35 @@ <h3 class="anchored" data-anchor-id="running-2">Running</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/pathfinder_config.html b/docs/cmdstan-guide/pathfinder_config.html
index 247e91e8b..3ad17a27a 100644
--- a/docs/cmdstan-guide/pathfinder_config.html
+++ b/docs/cmdstan-guide/pathfinder_config.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1235,8 +1235,35 @@ <h2 class="anchored" data-anchor-id="single-path-pathfinder-outputs">Single-path
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/pathfinder_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/pathfinder_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/print.html b/docs/cmdstan-guide/print.html
index 1a33ecfea..38a6da455 100644
--- a/docs/cmdstan-guide/print.html
+++ b/docs/cmdstan-guide/print.html
@@ -226,7 +226,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1000,8 +1000,35 @@ <h1><code>print</code> (deprecated): MCMC Output Analysis</h1>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/print.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/print.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/rdump_apdx.html b/docs/cmdstan-guide/rdump_apdx.html
index fd5461561..aea8c2afe 100644
--- a/docs/cmdstan-guide/rdump_apdx.html
+++ b/docs/cmdstan-guide/rdump_apdx.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1256,8 +1256,35 @@ <h2 class="anchored" data-anchor-id="bnf-grammar-for-dump-data">BNF grammar for
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/rdump_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/rdump_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/stan_csv_apdx.html b/docs/cmdstan-guide/stan_csv_apdx.html
index f764c5ff8..0c202f475 100644
--- a/docs/cmdstan-guide/stan_csv_apdx.html
+++ b/docs/cmdstan-guide/stan_csv_apdx.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1245,8 +1245,35 @@ <h2 class="anchored" data-anchor-id="diagnose-method-outputs">Diagnose method ou
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stan_csv_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stan_csv_apdx.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/stanc.html b/docs/cmdstan-guide/stanc.html
index 7ed425e4a..0a8d4f5ff 100644
--- a/docs/cmdstan-guide/stanc.html
+++ b/docs/cmdstan-guide/stanc.html
@@ -226,7 +226,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1030,8 +1030,35 @@ <h2 class="anchored" data-anchor-id="the-stan-compiler-program">The Stan compile
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/stansummary.html b/docs/cmdstan-guide/stansummary.html
index 4611817d4..e9070745a 100644
--- a/docs/cmdstan-guide/stansummary.html
+++ b/docs/cmdstan-guide/stansummary.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1178,8 +1178,35 @@ <h2 class="anchored" data-anchor-id="command-line-options">Command-line options<
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stansummary.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/stansummary.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/cmdstan-guide/variational_config.html b/docs/cmdstan-guide/variational_config.html
index e0ca7d9f0..96e9ed10f 100644
--- a/docs/cmdstan-guide/variational_config.html
+++ b/docs/cmdstan-guide/variational_config.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -1206,8 +1206,35 @@ <h2 class="anchored" data-anchor-id="csv-output">CSV output</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/variational_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/cmdstan-guide/variational_config.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/array_operations.html b/docs/functions-reference/array_operations.html
index a61e53b3a..dc0b99ca2 100644
--- a/docs/functions-reference/array_operations.html
+++ b/docs/functions-reference/array_operations.html
@@ -288,7 +288,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -511,6 +511,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1467,8 +1473,35 @@ <h2 class="anchored" data-anchor-id="reversing-functions">Reversing functions</h
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/array_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/array_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/binary_distributions.html b/docs/functions-reference/binary_distributions.html
index 6db376df1..18a7b19d6 100644
--- a/docs/functions-reference/binary_distributions.html
+++ b/docs/functions-reference/binary_distributions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1291,8 +1297,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-2">Stan Functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/binary_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/binary_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/bounded_continuous_distributions.html b/docs/functions-reference/bounded_continuous_distributions.html
index f69007d73..eda015ead 100644
--- a/docs/functions-reference/bounded_continuous_distributions.html
+++ b/docs/functions-reference/bounded_continuous_distributions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1157,8 +1163,35 @@ <h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/bounded_discrete_distributions.html b/docs/functions-reference/bounded_discrete_distributions.html
index 5b142af7e..36da4d4c7 100644
--- a/docs/functions-reference/bounded_discrete_distributions.html
+++ b/docs/functions-reference/bounded_discrete_distributions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1728,8 +1734,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-10">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/bounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/circular_distributions.html b/docs/functions-reference/circular_distributions.html
index 7e81ff5cb..a2e8bc372 100644
--- a/docs/functions-reference/circular_distributions.html
+++ b/docs/functions-reference/circular_distributions.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -512,6 +512,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1208,8 +1214,35 @@ <h3 class="anchored" data-anchor-id="numerical-stability">Numerical stability</h
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/circular_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/circular_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/complex-valued_basic_functions.html b/docs/functions-reference/complex-valued_basic_functions.html
index 80143c22c..70ee5f82b 100644
--- a/docs/functions-reference/complex-valued_basic_functions.html
+++ b/docs/functions-reference/complex-valued_basic_functions.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -512,6 +512,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1504,8 +1510,35 @@ <h2 class="anchored" data-anchor-id="complex-h-trig">Complex hyperbolic trigonom
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/complex_matrix_operations.html b/docs/functions-reference/complex_matrix_operations.html
index 03bb5f7e1..35eaadeb8 100644
--- a/docs/functions-reference/complex_matrix_operations.html
+++ b/docs/functions-reference/complex_matrix_operations.html
@@ -288,7 +288,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -511,6 +511,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -2066,8 +2072,35 @@ <h2 class="anchored" data-anchor-id="reverse-functions-for-complex-matrices">Rev
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/complex_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/compound_arithmetic_and_assignment.html b/docs/functions-reference/compound_arithmetic_and_assignment.html
index 8d34e2cbc..1b6ce2cfe 100644
--- a/docs/functions-reference/compound_arithmetic_and_assignment.html
+++ b/docs/functions-reference/compound_arithmetic_and_assignment.html
@@ -226,7 +226,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -449,6 +449,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1137,8 +1143,35 @@ <h2 class="anchored" data-anchor-id="compound-elementwise-division-and-assignmen
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/compound_arithmetic_and_assignment.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/compound_arithmetic_and_assignment.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/continuous_distributions_on_0_1.html b/docs/functions-reference/continuous_distributions_on_0_1.html
index 252cce468..df8c1baeb 100644
--- a/docs/functions-reference/continuous_distributions_on_0_1.html
+++ b/docs/functions-reference/continuous_distributions_on_0_1.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1208,8 +1214,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/continuous_distributions_on_0_1.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/continuous_distributions_on_0_1.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/conventions_for_probability_functions.html b/docs/functions-reference/conventions_for_probability_functions.html
index 838febad9..892c0238d 100644
--- a/docs/functions-reference/conventions_for_probability_functions.html
+++ b/docs/functions-reference/conventions_for_probability_functions.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -512,6 +512,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1339,8 +1345,35 @@ <h4 class="anchored" data-anchor-id="return-type">Return type</h4>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/conventions_for_probability_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/conventions_for_probability_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/correlation_matrix_distributions.html b/docs/functions-reference/correlation_matrix_distributions.html
index b1b16ce4c..2d9c7541a 100644
--- a/docs/functions-reference/correlation_matrix_distributions.html
+++ b/docs/functions-reference/correlation_matrix_distributions.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -532,6 +532,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1268,8 +1274,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-1">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/correlation_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/correlation_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/covariance_matrix_distributions.html b/docs/functions-reference/covariance_matrix_distributions.html
index 77e397def..26470d87e 100644
--- a/docs/functions-reference/covariance_matrix_distributions.html
+++ b/docs/functions-reference/covariance_matrix_distributions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1272,8 +1278,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-3">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/covariance_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/covariance_matrix_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/deprecated_functions.html b/docs/functions-reference/deprecated_functions.html
index 2b5f328d0..1a9def5ee 100644
--- a/docs/functions-reference/deprecated_functions.html
+++ b/docs/functions-reference/deprecated_functions.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -512,6 +512,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1317,8 +1323,35 @@ <h4 class="anchored" data-anchor-id="sizes-and-parallel-arrays-1">Sizes and para
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/deprecated_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/deprecated_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/distributions_over_unbounded_vectors.html b/docs/functions-reference/distributions_over_unbounded_vectors.html
index dc0eee2d7..aa66e4846 100644
--- a/docs/functions-reference/distributions_over_unbounded_vectors.html
+++ b/docs/functions-reference/distributions_over_unbounded_vectors.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -498,6 +498,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1584,8 +1590,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/distributions_over_unbounded_vectors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/distributions_over_unbounded_vectors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/embedded_laplace.html b/docs/functions-reference/embedded_laplace.html
new file mode 100644
index 000000000..e3e23970e
--- /dev/null
+++ b/docs/functions-reference/embedded_laplace.html
@@ -0,0 +1,1517 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Embedded Laplace Approximation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../functions-reference/mathematical_functions.html" rel="next">
+<link href="../functions-reference/hidden_markov_models.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Embedded Laplace Approximation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../functions-reference/index.html" aria-current="page"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/embedded_laplace.html">Embedded Laplace Approximation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Functions Reference</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Built-in Functions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/void_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Void Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/integer-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Integer-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/real-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Real-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex-valued_basic_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex-Valued Basic Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/array_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Array Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/complex_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/sparse_matrix_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse Matrix Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mixed_operations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mixed Operations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/compound_arithmetic_and_assignment.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Compound Arithmetic and Assignment</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/higher-order_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Higher-Order Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/transform_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variable Transformation Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/deprecated_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/removed_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/conventions_for_probability_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Conventions for Probability Functions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Discrete Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/binary_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Binary Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Discrete Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/multivariate_discrete_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multivariate Discrete Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Continuous Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/unbounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Unbounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/positive_lower-bounded_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Positive Lower-Bounded Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/continuous_distributions_on_0_1.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Continuous Distributions on [0, 1]</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/circular_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Circular Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/bounded_continuous_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Bounded Continuous Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/distributions_over_unbounded_vectors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Distributions over Unbounded Vectors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/simplex_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simplex Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/correlation_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Correlation Matrix Distributions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/covariance_matrix_distributions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Covariance Matrix Distributions</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Additional Distributions</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Hidden Markov Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendix</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-6" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-6" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/mathematical_functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Mathematical Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/functions_index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Alphabetical Index</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#embedded-laplace-approximation" id="toc-embedded-laplace-approximation" class="nav-link active" data-scroll-target="#embedded-laplace-approximation">Embedded Laplace Approximation</a>
+  <ul class="collapse">
+  <li><a href="#approximating-the-log-marginal-likelihood-log-py-mid-phi" id="toc-approximating-the-log-marginal-likelihood-log-py-mid-phi" class="nav-link" data-scroll-target="#approximating-the-log-marginal-likelihood-log-py-mid-phi">Approximating the log marginal likelihood <span class="math inline">\(\log p(y \mid \phi)\)</span></a></li>
+  <li><a href="#laplace-likelihood_spec" id="toc-laplace-likelihood_spec" class="nav-link" data-scroll-target="#laplace-likelihood_spec">Specifying the log likelihood function</a></li>
+  <li><a href="#specifying-the-covariance-function" id="toc-specifying-the-covariance-function" class="nav-link" data-scroll-target="#specifying-the-covariance-function">Specifying the covariance function</a></li>
+  <li><a href="#control-parameters" id="toc-control-parameters" class="nav-link" data-scroll-target="#control-parameters">Control parameters</a></li>
+  <li><a href="#sample-from-the-approximate-conditional-hatptheta-mid-y-phi" id="toc-sample-from-the-approximate-conditional-hatptheta-mid-y-phi" class="nav-link" data-scroll-target="#sample-from-the-approximate-conditional-hatptheta-mid-y-phi">Sample from the approximate conditional <span class="math inline">\(\hat{p}(\theta \mid y, \phi)\)</span></a></li>
+  <li><a href="#built-in-laplace-marginal-likelihood-functions" id="toc-built-in-laplace-marginal-likelihood-functions" class="nav-link" data-scroll-target="#built-in-laplace-marginal-likelihood-functions">Built-in Laplace marginal likelihood functions</a>
+  <ul class="collapse">
+  <li><a href="#poisson-with-log-link" id="toc-poisson-with-log-link" class="nav-link" data-scroll-target="#poisson-with-log-link">Poisson with log link</a></li>
+  <li><a href="#negative-binomial-with-log-link" id="toc-negative-binomial-with-log-link" class="nav-link" data-scroll-target="#negative-binomial-with-log-link">Negative Binomial with log link</a></li>
+  <li><a href="#bernoulli-with-logit-link" id="toc-bernoulli-with-logit-link" class="nav-link" data-scroll-target="#bernoulli-with-logit-link">Bernoulli with logit link</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/embedded_laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../functions-reference/hidden_markov_models.html">Additional Distributions</a></li><li class="breadcrumb-item"><a href="../functions-reference/embedded_laplace.html">Embedded Laplace Approximation</a></li></ol></nav></header>
+
+
+
+
+<section id="embedded-laplace-approximation" class="level1">
+<h1>Embedded Laplace Approximation</h1>
+<p>The embedded Laplace approximation can be used to approximate certain marginal and conditional distributions that arise in latent Gaussian models. Embedded Laplace replaces explicit sampling of (high-dimensional) Gaussian latent variables with a local Gaussian approximation. In doing so, it marginalizes out the latent Gaussian variables. Inference can then be performed on the remaining, often low-dimensional, parameters. The embedded Laplace approximation in Stan is best suited for latent Gaussian models when jointly sampling over all model parameters is expensive and the conditional posterior of the Gaussian latent variables is reasonably close to Gaussian.</p>
+<p>For observed data <span class="math inline">\(y\)</span>, latent Gaussian variables <span class="math inline">\(\theta\)</span>, and hyperparameters <span class="math inline">\(\phi\)</span>, a latent Gaussian model observes the following hierarchical structure: <span class="math display">\[\begin{eqnarray}
+  \phi &amp;\sim&amp; p(\phi), \\
+  \theta &amp;\sim&amp; \text{MultiNormal}(0, K(\phi)), \\
+  y &amp;\sim&amp; p(y \mid \theta, \phi).
+\end{eqnarray}\]</span> In this formulation, <span class="math inline">\(p(y \mid \theta, \phi)\)</span> is the data model that specifies how observations are generated conditional on <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span>. <span class="math inline">\(K(\phi)\)</span> denotes the prior covariance matrix for the latent Gaussian variables <span class="math inline">\(\theta\)</span> and is parameterized by <span class="math inline">\(\phi\)</span>. The prior on <span class="math inline">\(\theta\)</span> is centered at 0, however an offset can always be added when specifying the data model <span class="math inline">\(p(y \mid \theta, \phi)\)</span>.</p>
+<p>Conditioning on observations <span class="math inline">\(y\)</span> we obtain the joint posterior <span class="math inline">\(p(\phi, \theta \mid y) \propto p(y \mid \theta, \phi) p(\theta |
+\phi) p(\phi)\)</span>, where <span class="math inline">\(p(y \mid \theta, \phi)\)</span> as function of <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span> is the likelihood function. To sample from the joint posterior, we can either use a standard method, such as Markov chain Monte Carlo, or we can follow a two-step procedure:</p>
+<ol type="1">
+<li>sample from the <em>marginal posterior</em> <span class="math inline">\(p(\phi \mid y)\)</span>,</li>
+<li>sample from the <em>conditional posterior</em> <span class="math inline">\(p(\theta \mid y, \phi)\)</span>.</li>
+</ol>
+<p>In the above procedure, neither the marginal posterior nor the conditional posterior are typically available in closed form and so they must be approximated. The marginal posterior can be written as <span class="math inline">\(p(\phi \mid y) \propto p(y \mid \phi) p(\phi)\)</span>, where <span class="math inline">\(p(y \mid \phi) = \int p(y \mid \phi, \theta) p(\theta) \text{d}\theta\)</span> is called the marginal likelihood. The Laplace method approximates <span class="math inline">\(p(y \mid \phi, \theta) p(\theta)\)</span> with a normal distribution centered at the mode, <span class="math display">\[
+  \theta^* = \underset{\theta}{\text{argmax}} \ \log p(\theta \mid y, \phi),
+\]</span> and <span class="math inline">\(\theta^*\)</span> is obtained using a numerical optimizer. The resulting Gaussian integral can be evaluated analytically to obtain an approximation to the log marginal likelihood <span class="math inline">\(\log \hat p(y \mid \phi) \approx \log p(y \mid \phi)\)</span>. Specifically: <span class="math display">\[
+  \hat p(y \mid \phi) = \frac{p(\theta^* \mid \phi) p(y \mid \theta^*, \phi)}{\hat p (\theta^* \mid \phi, y)}.
+\]</span></p>
+<p>Combining this marginal likelihood with the prior in the <code>model</code> block, we can then sample from the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span> using one of Stan’s algorithms. The marginal posterior is lower dimensional and likely to have a simpler geometry leading to more efficient inference. On the other hand each marginal likelihood computation is more costly, and the combined change in efficiency depends on the application.</p>
+<p>To obtain posterior draws for <span class="math inline">\(\theta\)</span>, we sample from the normal approximation to <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in <code>generated quantities</code>. The process of iteratively sampling from <span class="math inline">\(p(\phi \mid y)\)</span> (say, with MCMC) and then <span class="math inline">\(p(\theta \mid y, \phi)\)</span> produces posterior draws from the joint posterior <span class="math inline">\(p(\theta, \phi \mid y)\)</span>.</p>
+<p>The Laplace approximation is especially useful if <span class="math inline">\(p(y \mid \phi, \theta)\)</span> as function of <span class="math inline">\(\theta\)</span> is log-concave, e.g., in case of Poisson, binomial, negative-binomial, and Bernoulli. (The likelihood of normal model is also log concave, however when the likelihood is normal, marginalization can be performed exactly and does not required an approximation.) Stan’s embedded Laplace approximation is restricted to the case where the prior <span class="math inline">\(p(\theta \mid \phi)\)</span> is multivariate normal. Furthermore, the likelihood <span class="math inline">\(p(y \mid \phi, \theta)\)</span> must be computed using only operations which support higher-order derivatives (see section <a href="#laplace_likelihood_spec">specifying the likelihood function</a>).</p>
+<p>The Laplace approximation can also be useful in generated quantities to marginalize out latent variables even if the sampling had been done using the full joint posterior.</p>
+<section id="approximating-the-log-marginal-likelihood-log-py-mid-phi" class="level2">
+<h2 class="anchored" data-anchor-id="approximating-the-log-marginal-likelihood-log-py-mid-phi">Approximating the log marginal likelihood <span class="math inline">\(\log p(y \mid \phi)\)</span></h2>
+<p>In the <code>model</code> block, we increment <code>target</code> with <code>laplace_marginal</code>, a function that approximates the log marginal likelihood <span class="math inline">\(\log p(y \mid \phi)\)</span>. The signature of the function is:</p>
+<a id="index-entry-7ffad9885f7c8f209fffbfae2182cbc3ac432b59" class="index-entry"></a> <!-- real; laplace_marginal; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal" class="unlink" title="Jump to index entry"><code>laplace_marginal</code></a></strong><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code></p>
+<p>which returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span>.</p>
+<p>The embedded Laplace functions accept two functors whose user defined arguments are passed in as tuples to <code>laplace_marginal</code>.</p>
+<ol type="1">
+<li><code>likelihood_function</code> - user-specified log likelihood whose first argument is the vector of latent Gaussian variables <span class="math inline">\(\theta\)</span>. The subsequent arguments are user defined.</li>
+</ol>
+<ul>
+<li><code>real likelihood_function(vector theta, likelihood_arguments_1, likelihood_arguments_2, ...)</code>.</li>
+</ul>
+<ol start="2" type="1">
+<li><code>likelihood_arguments</code> - A tuple of arguments whose internal members are be passed to the log likelihood function. This tuple does NOT include the latent variable <span class="math inline">\(\theta\)</span>.</li>
+<li><code>hessian_block_size</code> - the block size of the Hessian of the log likelihood, <span class="math inline">\(\partial^2 \log p(y \mid \theta, \phi) / \partial \theta^2\)</span>.</li>
+<li><code>covariance_function</code> - A function that returns the covariance matrix of the multivariate normal prior on <span class="math inline">\(\theta\)</span>.</li>
+</ol>
+<ul>
+<li><code>matrix covariance_function(covariance_argument_1, covariance_argument_2, ...)</code>.</li>
+</ul>
+<ol start="5" type="1">
+<li><code>covariance_arguments</code> A tuple of the arguments whose internal members will be passed to the the covariance function.</li>
+</ol>
+<small><i>Available since 2.39</i></small>
+<p>Below we go over each argument in more detail.</p>
+</section>
+<section id="laplace-likelihood_spec" class="level2">
+<h2 class="anchored" data-anchor-id="laplace-likelihood_spec">Specifying the log likelihood function</h2>
+<p>The first step to use the embedded Laplace approximation is to write down a function in the <code>functions</code> block which returns the log likelihood <span class="math inline">\(\log p(y \mid \theta, \phi)\)</span>.</p>
+<p>There are a few constraints on this function:</p>
+<ol type="1">
+<li><p>The function return type must be <code>real</code>.</p></li>
+<li><p>The first argument must be the latent Gaussian variable <span class="math inline">\(\theta\)</span> and must have type <code>vector</code>.</p></li>
+<li><p>The operations in the function must support higher-order automatic differentiation (AD). Most functions in Stan support higher-order AD. The exceptions are functions with specialized calls for reverse-mode AD, and these are higher-order functions (algebraic solvers, differential equation solvers, and integrators), the marginalization function for hidden Markov models (HMM) function, and the embedded Laplace approximation itself.</p></li>
+</ol>
+<p>The base signature of the function is</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> likelihood_function(<span class="dt">vector</span> theta, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>...</code> represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments <code>...</code> and each argument can be passed as data or parameter.</p>
+<p>The tuple after <code>likelihood_function</code> contains the arguments that get passed to <code>likelihood_function</code> <em>excluding <span class="math inline">\(\theta\)</span></em>. For instance, if a user defined likelihood uses a real and a matrix, the likelihood function’s signature would first have a vector and then a real and matrix argument.</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> likelihood_fun(<span class="dt">vector</span> theta, <span class="dt">real</span> a, <span class="dt">matrix</span> X)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The call to the laplace marginal would start with this likelihood and tuple holding the other likelihood arguments. We do not need to pass <code>theta</code>, since it is marginalized out and therefore does not appear explicitly as a passed parameter.</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> val = laplace_marginal(likelihood_fun, (a, X), hessian_block_size, ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the <code>likelihood_function</code> has only one argument, the tuple syntax is <code>(a, )</code>.</p>
+<p>As always, users should use parameter arguments only when necessary in order to speed up differentiation. In general, we recommend marking data only arguments with the keyword <code>data</code>, for example,</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> likelihood_function(<span class="dt">vector</span> theta, <span class="kw">data</span> <span class="dt">vector</span> x, ...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>In addition to the likelihood function, users must specify the block size of the Hessian, <span class="math inline">\(\partial^2 \log p(y \mid \theta, \phi) / \partial \theta^2\)</span>. The Hessian is often block diagonal and this structure can be taken advantage of for fast computation. For example, if <span class="math inline">\(y_i\)</span> only depends on <span class="math inline">\(\theta_i\)</span>, then the Hessian is diagonal and <code>hessian_block_size=1</code>,</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> val = laplace_marginal(likelihood_fun, (a, X), <span class="dv">1</span>, ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>On the other hand, if the Hessian is not block diagonal, we can always set <code>hessian_block_size=n</code> where <span class="math inline">\(n\)</span> is the size of <span class="math inline">\(\theta\)</span>.</p>
+</section>
+<section id="specifying-the-covariance-function" class="level2">
+<h2 class="anchored" data-anchor-id="specifying-the-covariance-function">Specifying the covariance function</h2>
+<p>The argument <code>covariance_function</code> returns the prior covariance matrix <span class="math inline">\(K\)</span>. The signature for this function is the same as a standard stan function. It’s return type must be a matrix of size <span class="math inline">\(n \times n\)</span> where <span class="math inline">\(n\)</span> is the size of <span class="math inline">\(\theta\)</span>.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> covariance_function(...)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>...</code> represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments <code>...</code> and each argument can be passed as data or parameter. The variables <span class="math inline">\(\phi\)</span> is implicitly defined as the collection of all non-data arguments passed to <code>likelihood_function</code> (excluding <span class="math inline">\(\theta\)</span>) and <code>covariance_function</code>.</p>
+<p>The tuple after <code>covariance_function</code> contains the arguments that get passed to <code>covariance_function</code>. For instance, if a user defined covariance function uses two vectors</p>
+<div class="sourceCode" id="cb7"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span> cov_fun(<span class="dt">real</span> b, <span class="dt">matrix</span> Z)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>the call to the Laplace marginal would include the covariance function and a tuple holding the covariance function arguments.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> val = laplace_marginal(likelihood_fun, (a, X), cov_fun, (b, Z), ...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If the <code>covariance_function</code> has only one argument, the tuple syntax is <code>(b, )</code>.</p>
+</section>
+<section id="control-parameters" class="level2">
+<h2 class="anchored" data-anchor-id="control-parameters">Control parameters</h2>
+<p>It also possible to specify control parameters, which can help improve the optimization that underlies the Laplace approximation, using <code>laplace_marginal_tol</code> with the following signature:</p>
+<a id="index-entry-9d40f0ee623d8ad92f054b2d83550d7a8ddf4b56" class="index-entry"></a> <!-- real; laplace_marginal_tol; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol</code></a></strong><code>(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...),  tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>The final argument, tolerances, is a tuple with the following elements</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">vector</span> theta_init, <span class="dt">real</span> tol, <span class="dt">int</span> max_steps, <span class="dt">int</span> solver,</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>      <span class="dt">int</span> max_steps_linesearch, <span class="dt">int</span> allow_fallback)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> and allows the user to tune the control parameters of the approximation.</p>
+<ul>
+<li><p><code>theta_init</code>: the initial guess for a Newton solver when finding the mode of <span class="math inline">\(p(\theta \mid y, \phi)\)</span>. By default, it is a zero-vector.</p></li>
+<li><p><code>tol</code>: the tolerance <span class="math inline">\(\epsilon\)</span> of the optimizer. Specifically, the optimizer stops when <span class="math inline">\(||\nabla \log p(\theta \mid y, \phi)|| \le \epsilon\)</span>. By default, the value is <span class="math inline">\(\epsilon \approx 1.49 \times 10^{-8}\)</span>, which is the square-root of machine precision.</p></li>
+<li><p><code>max_num_steps</code>: the maximum number of steps taken by the optimizer before it gives up (in which case the Metropolis proposal gets rejected). The default is 500 steps.</p></li>
+<li><p><code>solver</code>: choice of Newton solver. The optimizer underlying the Laplace approximation does one of three matrix decompositions to compute a Newton step. The problem determines which decomposition is numerically stable. By default (<code>solver=1</code>), the solver attempts a Cholesky decomposition of the negative Hessian of the log likelihood, <span class="math inline">\(- \partial^2 \log p(y \mid \theta, \phi) / \partial^2 \theta\)</span>. This operation is legal if the negative Hessian is positive-definite, which will always be true when the likelihood as function of <span class="math inline">\(\theta\)</span> is log concave. If <code>solver=2</code>, the solver makes a Cholesky decomposition of the covariance matrix <span class="math inline">\(K(\phi)\)</span>. Since a covariance matrix is always positive-definite, computing its Cholesky decomposition is always a legal operation, at least in theory. In practice, we may not be able to compute the Cholesky decomposition of the negative Hessian nor of the covariance matrix, either because it does not exist or because of numerical issues. In that case, we can use <code>solver=3</code> which uses a more expensive but less specialized approach to compute a Newton step.</p></li>
+<li><p><code>max_steps_linesearch</code>: maximum number of steps in linesearch. The linesearch adjusts to step size to ensure that a Newton step leads to an increase in the objective function (i.e., <span class="math inline">\(f(\theta) = p(\theta \mid \phi, y)\)</span>). If a standard Newton step does not improve the objective function, the step is adjusted iteratively until the objective function increases or the maximum number of steps in the linesearch is reached. By default, <code>max_steps_linesearch=1000</code>. Setting <code>max_steps_linesearch=0</code> results in no linesearch.</p></li>
+<li><p><code>allow_fallback</code>: If user set solver fails, this flag determines whether to fallback to the next solver. For example, if the user specifies <code>solver=1</code> but the Cholesky decomposition of the negative Hessian <span class="math inline">\(- \partial^2 \log p(y \mid \theta, \phi) / \partial^2 \theta\)</span> fails, the optimizer will try <code>solver=2</code> instead. By default, <code>allow_fallback = 1</code> (TRUE).</p></li>
+</ul>
+<small><i>Available since 2.39</i></small>
+<p>The embedded Laplace approximation’s options have a helper callable <code>generate_laplace_options(int theta_size)</code> that will generate the tuple for the user. This can be useful for quickly setting up the control parameters in the <code>transformed data</code> block to reuse within the model.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">vector</span>[theta_size], <span class="dt">real</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>) laplace_ops = generate_laplace_options(theta_size);</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a><span class="co">// Modify solver type</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>laplace_ops<span class="fl">.5</span> = <span class="dv">2</span>;</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a><span class="co">// Turn off fallthrough</span></span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>laplace_ops<span class="fl">.7</span> = <span class="dv">0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<a id="index-entry-60827bf41179724b5e6e8a0fc73da3515b44fc46" class="index-entry"></a> <!-- tuple(vector, real, int, int, int, int); generate_laplace_options; (int dimension); -->
+<p></p>
+<p><code>tuple(vector, real, int, int, int, int)</code> <strong><a href="../functions-reference/functions_index.html#generate_laplace_options" class="unlink" title="Jump to index entry"><code>generate_laplace_options</code></a></strong><code>(int dimension)</code><br></p>
+<p>Create a default laplace options tuple for a <code>theta_init</code> of size <code>dimension</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-d1c421fc2e179180c0a9a45ea4733480792940bc" class="index-entry"></a> <!-- tuple(vector, real, int, int, int, int); generate_laplace_options; (vector theta_init); -->
+<p></p>
+<p><code>tuple(vector, real, int, int, int, int)</code> <strong><a href="../functions-reference/functions_index.html#generate_laplace_options" class="unlink" title="Jump to index entry"><code>generate_laplace_options</code></a></strong><code>(vector theta_init)</code><br></p>
+<p>Create a default Laplace options tuple containing <code>theta_init</code>.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="sample-from-the-approximate-conditional-hatptheta-mid-y-phi" class="level2">
+<h2 class="anchored" data-anchor-id="sample-from-the-approximate-conditional-hatptheta-mid-y-phi">Sample from the approximate conditional <span class="math inline">\(\hat{p}(\theta \mid y, \phi)\)</span></h2>
+<p>In <code>generated quantities</code>, it is possible to sample from the Laplace approximation of <span class="math inline">\(p(\theta \mid \phi, y)\)</span> using <code>laplace_latent_rng</code>. The signature for <code>laplace_latent_rng</code> follows closely the signature for <code>laplace_marginal</code>:</p>
+<a id="index-entry-60d4efd35a460871aff97a758f5d366faee26465" class="index-entry"></a> <!-- vector; laplace_latent_rng; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_rng</code></a></strong><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Samples from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span>.</p>
+<small><i>Available since 2.39</i></small>
+<p>Once again, it is possible to specify control parameters:</p>
+<a id="index-entry-eb06251dd7ccd5f9b7882b7e6950fa99cc0c2d57" class="index-entry"></a> <!-- vector; laplace_latent_rng_tol; (function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><code>laplace_latent_tol_rng</code></strong><code>(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...), tuple(vector, real, int, int, int, int) tolerances)</code><br> Samples from the approximate conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="built-in-laplace-marginal-likelihood-functions" class="level2">
+<h2 class="anchored" data-anchor-id="built-in-laplace-marginal-likelihood-functions">Built-in Laplace marginal likelihood functions</h2>
+<p>Stan provides convenient wrappers for the embedded Laplace approximation when applied to latent Gaussian models with certain likelihoods arising from some common data models. With this wrapper, the likelihood is pre-specified and does not need to be specified by the user. The selection of supported likelihoods is currently narrow and expected to grow. The wrappers exist for the user’s convenience but are not more computationally efficient than specifying log likelihoods in the <code>functions</code> block.</p>
+<section id="poisson-with-log-link" class="level3">
+<h3 class="anchored" data-anchor-id="poisson-with-log-link">Poisson with log link</h3>
+<p>Given count data, with each observed count <span class="math inline">\(y_i\)</span> associated with a group <span class="math inline">\(g(i)\)</span> and a corresponding latent variable <span class="math inline">\(\theta_{g(i)}\)</span>, and a Poisson model, the likelihood is <span class="math display">\[
+p(y \mid \theta, \phi) = \prod_i\text{Poisson} (y_i \mid \exp(\theta_{g(i)} + m_{g(i)})),
+\]</span> where <span class="math inline">\(m_{g(i)}\)</span> acts as an offset for <span class="math inline">\(\theta_{g(i)}\)</span>. This can also be interpreted as a prior mean on <span class="math inline">\(\theta_{g(i)}\)</span>. The arguments required to compute this likelihood are:</p>
+<ul>
+<li><code>y</code>: an array of counts.</li>
+<li><code>y_index</code>: an array whose <span class="math inline">\(i^\text{th}\)</span> element indicates to which group the <span class="math inline">\(i^\text{th}\)</span> observation belongs to.</li>
+<li><code>m</code>: a vector of offsets or prior means for <span class="math inline">\(\theta\)</span>.</li>
+</ul>
+<a id="index-entry-b2c6946b65561039eb8bf3999fb59d38a6336e10" class="index-entry"></a> <!-- real; laplace_marginal_poisson_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_poisson_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_poisson_log</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-4a7ecca812a308ea648bb31996559cba79738d83" class="index-entry"></a> <!-- real; laplace_marginal_tol_poisson_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_poisson_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_poisson_log</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_tol_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covaraince_arguments, tolerances)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<p>The signatures for the embedded Laplace approximation function with a Poisson likelihood are</p>
+<a id="index-entry-ca1a086a83293f252043ab8b127ec6c9ef16deec" class="index-entry"></a> <!-- real; laplace_marginal_poisson_log_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments; -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_poisson_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_poisson_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-51cad53596dd0d74c0fe99b84a3c975cbec3866a" class="index-entry"></a> <!-- real; laplace_marginal_tol_poisson_log_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_poisson_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_poisson_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-d0ecc660f5107f7409bd76ba059c5ea82ec81ba0" class="index-entry"></a> <!-- real; laplace_marginal_poisson_log_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments; -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_poisson_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_poisson_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-23e9ed91fd3f98c314ddbce86a77005fdf8ed5f3" class="index-entry"></a> <!-- real; laplace_marginal_tol_poisson_log_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p> <code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_poisson_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_poisson_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-a6bf063f06956735a484fd5e960352833373ec15" class="index-entry"></a> <!-- vector; laplace_latent_poisson_log_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p> <code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_poisson_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_poisson_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br> Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-8fec77181949b3ca47b1916246cd0384394f87bc" class="index-entry"></a> <!-- vector; laplace_latent_tol_poisson_log_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_tol_poisson_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_tol_poisson_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Poisson distribution with a log link and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="negative-binomial-with-log-link" class="level3">
+<h3 class="anchored" data-anchor-id="negative-binomial-with-log-link">Negative Binomial with log link</h3>
+<p>The negative Binomial distribution generalizes the Poisson distribution by introducing the dispersion parameter <span class="math inline">\(\eta\)</span>. The corresponding likelihood is then <span class="math display">\[
+p(y \mid \theta, \phi) = \prod_i\text{NegBinomial2} (y_i \mid \exp(\theta_{g(i)} + m_{g(i)}), \eta).
+\]</span> Here we use the alternative parameterization implemented in Stan, meaning that <span class="math display">\[
+\mathbb E(y_i) = \exp (\theta_{g(i)} + m_{g(i)}), \\
+\text{Var}(y_i) = \mathbb E(y_i) + \frac{(\mathbb E(y_i))^2}{\eta}.
+\]</span> The arguments for the likelihood function are:</p>
+<ul>
+<li><code>y</code>: the observed counts</li>
+<li><code>y_index</code>: an array whose <span class="math inline">\(i^\text{th}\)</span> element indicates to which group the <span class="math inline">\(i^\text{th}\)</span> observation belongs to.</li>
+<li><code>eta</code>: the overdispersion parameter.</li>
+<li><code>m</code>: a vector of offsets or prior means for <span class="math inline">\(\theta\)</span>.</li>
+</ul>
+<a id="index-entry-7f807083cfb7347ece100da5dfe719186a678d02" class="index-entry"></a> <!-- real; laplace_marginal_neg_binomial_2_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_neg_binomial_2_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_neg_binomial_2_log</code></a></strong><code>(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-40016f7d5d96b9d999ca42f446720cf3f3cd5769" class="index-entry"></a> <!-- real; laplace_marginal_tol_neg_binomial_2_log ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_neg_binomial_2_log" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_neg_binomial_2_log</code></a></strong><code>(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_tol_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<p>The function signatures for the embedded Laplace approximation with a negative Binomial likelihood are</p>
+<a id="index-entry-b74e5e572a36211903f16f5deed7158384c21355" class="index-entry"></a> <!-- real; laplace_marginal_neg_binomial_2_log_lpmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_neg_binomial_2_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_neg_binomial_2_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-cefee8b5275ca255f121386fb14c9354cf95594c" class="index-entry"></a> <!-- real; laplace_marginal_tol_neg_binomial_2_log_lpmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_neg_binomial_2_log_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_neg_binomial_2_log_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-5d9163f94a110e178c54b2bd192e9aded134cd3a" class="index-entry"></a> <!-- real; laplace_marginal_neg_binomial_2_log_lupmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_neg_binomial_2_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_neg_binomial_2_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, function covariance_function, data int hessian_block_size, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-efb046fcbadbcaaa59a274ab3c5b1ed15f23add9" class="index-entry"></a> <!-- real; laplace_marginal_tol_neg_binomial_2_log_lupmf; (array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_neg_binomial_2_log_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_neg_binomial_2_log_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-79fe43b1435d4b3821038d2cd934d8c136e8808c" class="index-entry"></a> <!-- vector; laplace_latent_neg_binomial_2_log_rng; (array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_neg_binomial_2_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_neg_binomial_2_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative binomial distribution with a log link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-259ff7a39109d52d1194cf72a125b72127ddd2b1" class="index-entry"></a> <!-- vector; laplace_latent_tol_neg_binomial_2_log_rng; (array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_tol_neg_binomial_2_log_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_tol_neg_binomial_2_log_rng</code></a></strong><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi, \eta)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta, \eta)\)</span> is a Negative binomial distribution with a log link and allows the user to tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+</section>
+<section id="bernoulli-with-logit-link" class="level3">
+<h3 class="anchored" data-anchor-id="bernoulli-with-logit-link">Bernoulli with logit link</h3>
+<p>Given binary outcome <span class="math inline">\(y_i \in \{0, 1\}\)</span> and Bernoulli model, the likelihood is <span class="math display">\[
+p(y \mid \theta, \phi) = \prod_i\text{Bernoulli} (y_i \mid \text{logit}^{-1}(\theta_{g(i)} + m_{g(i)})).
+\]</span> The arguments of the likelihood function are:</p>
+<ul>
+<li><code>y</code>: the observed counts</li>
+<li><code>y_index</code>: an array whose <span class="math inline">\(i^\text{th}\)</span> element indicates to which group the <span class="math inline">\(i^\text{th}\)</span> observation belongs to.</li>
+<li><code>m</code>: a vector of offsets or prior means for <span class="math inline">\(\theta\)</span>.</li>
+</ul>
+<a id="index-entry-1d93ab799518d0aac88e63c01f9655f36c7cbeb6" class="index-entry"></a> <!-- real; laplace_marginal_bernoulli_logit ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_bernoulli_logit" class="unlink" title="Jump to index entry"><code>laplace_marginal_bernoulli_logit</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-92d43f7c3643c7d85966bbfc0b2a71546facb37c" class="index-entry"></a> <!-- real; laplace_marginal_tol_bernoulli_logit ~; -->
+<p></p>
+<p><code>y ~</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_bernoulli_logit" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_bernoulli_logit</code></a></strong><code>(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code><br></p>
+<p>Increment target log probability density with <code>laplace_marginal_tol_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<p>The function signatures for the embedded Laplace approximation with a Bernoulli likelihood are</p>
+<a id="index-entry-fd129c664f37d1cdd280abaf7ee82bfaf9c8e54e" class="index-entry"></a> <!-- real; laplace_marginal_bernoulli_logit_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_bernoulli_logit_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_bernoulli_logit_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-5126f7d7829072e478ea2f7257d226219b43e1d4" class="index-entry"></a> <!-- real; laplace_marginal_tol_bernoulli_logit_lpmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_bernoulli_logit_lpmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_bernoulli_logit_lpmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link and allows the user to tune the control parameters.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-53e5e8065cf24c3e72a435d5fccf4d14a587b245" class="index-entry"></a> <!-- real; laplace_marginal_bernoulli_logit_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_bernoulli_logit_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_bernoulli_logit_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-7d93f40162970812a81fc61df49dcb53f993e293" class="index-entry"></a> <!-- real; laplace_marginal_tol_bernoulli_logit_lupmf; (array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#laplace_marginal_tol_bernoulli_logit_lupmf" class="unlink" title="Jump to index entry"><code>laplace_marginal_tol_bernoulli_logit_lupmf</code></a></strong><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns an approximation to the log marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a bernoulli distribution with a logit link and allows the user to tune the control parameters, dropping constant terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-59f1ae97dbd720c3dd7cd15f92e7e3c6f93ae22e" class="index-entry"></a> <!-- vector; laplace_latent_bernoulli_logit_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_bernoulli_logit_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_bernoulli_logit_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Bernoulli distribution with a logit link.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-4566ad7c6cd5f0ee057539a6e2a0bea28554f082" class="index-entry"></a> <!-- vector; laplace_latent_tol_bernoulli_logit_rng; (array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances); -->
+<p></p>
+<p><code>vector</code> <strong><a href="../functions-reference/functions_index.html#laplace_latent_tol_bernoulli_logit_rng" class="unlink" title="Jump to index entry"><code>laplace_latent_tol_bernoulli_logit_rng</code></a></strong><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)</code><br></p>
+<p>Returns a draw from the Laplace approximation to the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> in the special case where the likelihood <span class="math inline">\(p(y \mid \theta)\)</span> is a Bernoulli distribution with a logit link, and lets the user tune the control parameters of the approximation.</p>
+<small><i>Available since 2.39</i></small>
+<!-- ## Draw approximate samples for out-of-sample latent variables. -->
+<!-- In many applications, it is of interest to draw latent variables for -->
+<!-- in-sample and out-of-sample predictions. We respectively denote these latent -->
+<!-- variables $\theta$ and $\theta^*$. In a latent Gaussian model,  -->
+<!-- $(\theta, \theta^*)$ jointly follow a prior multivariate normal distribution: -->
+<!-- $$ -->
+<!--   \theta, \theta^* \sim \text{MultiNormal}(0, {\bf K}(\phi)), -->
+<!-- $$ -->
+<!-- where $\bf K$ designates the joint covariance matrix over $\theta, \theta^*$. -->
+<!-- We can break $\bf K$ into three components, -->
+<!-- $$ -->
+<!-- {\bf K} = \begin{bmatrix} -->
+<!--   K & \\ -->
+<!--   K^* & K^{**} -->
+<!-- \end{bmatrix}, -->
+<!-- $$ -->
+<!-- where $K$ is the prior covariance matrix for $\theta$, $K^{**}$ the prior -->
+<!-- covariance matrix for $\theta^*$, and $K^*$ the covariance matrix between -->
+<!-- $\theta$ and $\theta^*$. -->
+<!-- Stan supports the case where $\theta$ is associated with an in-sample -->
+<!-- covariate $X$ and $\theta^*$ with an out-of-sample covariate $X^*$. -->
+<!-- Furthermore, the covariance function is written in such a way that -->
+<!-- $$ -->
+<!-- K = f(..., X, X), \ \ K^{**} = f(..., X^*, X^*), \ \ K^* = f(..., X, X^*), -->
+<!-- $$ -->
+<!-- as is typically the case in Gaussian process models. -->
+<!-- The -->
+<!-- function `laplace_latent_rng` samples from the Laplace approximation -->
+<!-- and admits nearly the same arguments as `laplace_marginal`. A key difference -->
+<!-- is that  -->
+<!-- ``` -->
+<!-- vector laplace_latent_rng(function likelihood_function, tuple(...), vector theta_0,  -->
+<a id="index-entry-770ccdd326df6b55b9b82f28548a93c005c11324" class="index-entry"></a> <!--                           function covariance_function, tuple(...)); -->
+<!-- ``` -->
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../functions-reference/hidden_markov_models.html" class="pagination-link" aria-label="Hidden Markov Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Hidden Markov Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../functions-reference/mathematical_functions.html" class="pagination-link" aria-label="Mathematical Functions">
+        <span class="nav-page-text">Mathematical Functions</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/embedded_laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/functions_index.html b/docs/functions-reference/functions_index.html
index b89c87236..cf99a8514 100644
--- a/docs/functions-reference/functions_index.html
+++ b/docs/functions-reference/functions_index.html
@@ -224,7 +224,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -447,6 +447,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -506,6 +512,7 @@ <h2 id="toc-title">On this page</h2>
   <li><a href="#u" id="toc-u" class="nav-link" data-scroll-target="#u">U</a></li>
   <li><a href="#v" id="toc-v" class="nav-link" data-scroll-target="#v">V</a></li>
   <li><a href="#w" id="toc-w" class="nav-link" data-scroll-target="#w">W</a></li>
+  <li><a href="#y" id="toc-y" class="nav-link" data-scroll-target="#y">Y</a></li>
   <li><a href="#z" id="toc-z" class="nav-link" data-scroll-target="#z">Z</a></li>
   </ul></li>
   </ul>
@@ -2322,6 +2329,15 @@ <h2 class="anchored" data-anchor-id="g">G</h2>
 <a href="../functions-reference/matrix_operations.html#index-entry-49a205d1f503afa341f4ec2f12b8d9484c5bc3f0"><code>(matrix A) : matrix</code></a> <span class="detail">(matrix_operations.html)</span>
 </div></li>
 </ul>
+<p><a id="generate_laplace_options" href="#generate_laplace_options" class="anchored unlink"><strong>generate_laplace_options</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-60827bf41179724b5e6e8a0fc73da3515b44fc46"><code>(int dimension) : tuple(vector, real, int, int, int, int)</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-d1c421fc2e179180c0a9a45ea4733480792940bc"><code>(vector theta_init) : tuple(vector, real, int, int, int, int)</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
 <p><a id="get_imag" href="#get_imag" class="anchored unlink"><strong>get_imag</strong>:</a></p>
 <ul>
 <li><div class="index-container">
@@ -2880,6 +2896,174 @@ <h2 class="anchored" data-anchor-id="l">L</h2>
 <a href="../functions-reference/real-valued_basic_functions.html#index-entry-ab5ca07ba7ba53020939ca3ba7c6e64ccf05cf19"><code>(T x) : R</code></a> <span class="detail">(real-valued_basic_functions.html)</span>
 </div></li>
 </ul>
+<p><a id="laplace_latent_bernoulli_logit_rng" href="#laplace_latent_bernoulli_logit_rng" class="anchored unlink"><strong>laplace_latent_bernoulli_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-59f1ae97dbd720c3dd7cd15f92e7e3c6f93ae22e"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_neg_binomial_2_log_rng" href="#laplace_latent_neg_binomial_2_log_rng" class="anchored unlink"><strong>laplace_latent_neg_binomial_2_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-79fe43b1435d4b3821038d2cd934d8c136e8808c"><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_poisson_log_rng" href="#laplace_latent_poisson_log_rng" class="anchored unlink"><strong>laplace_latent_poisson_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-a6bf063f06956735a484fd5e960352833373ec15"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_rng" href="#laplace_latent_rng" class="anchored unlink"><strong>laplace_latent_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-60d4efd35a460871aff97a758f5d366faee26465"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_rng_tol" href="#laplace_latent_rng_tol" class="anchored unlink"><strong>laplace_latent_rng_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-eb06251dd7ccd5f9b7882b7e6950fa99cc0c2d57"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_tol_bernoulli_logit_rng" href="#laplace_latent_tol_bernoulli_logit_rng" class="anchored unlink"><strong>laplace_latent_tol_bernoulli_logit_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-4566ad7c6cd5f0ee057539a6e2a0bea28554f082"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_tol_neg_binomial_2_log_rng" href="#laplace_latent_tol_neg_binomial_2_log_rng" class="anchored unlink"><strong>laplace_latent_tol_neg_binomial_2_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-259ff7a39109d52d1194cf72a125b72127ddd2b1"><code>(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_latent_tol_poisson_log_rng" href="#laplace_latent_tol_poisson_log_rng" class="anchored unlink"><strong>laplace_latent_tol_poisson_log_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-8fec77181949b3ca47b1916246cd0384394f87bc"><code>(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal" href="#laplace_marginal" class="anchored unlink"><strong>laplace_marginal</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-7ffad9885f7c8f209fffbfae2182cbc3ac432b59"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_bernoulli_logit" href="#laplace_marginal_bernoulli_logit" class="anchored unlink"><strong>laplace_marginal_bernoulli_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-1d93ab799518d0aac88e63c01f9655f36c7cbeb6">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_bernoulli_logit_lpmf" href="#laplace_marginal_bernoulli_logit_lpmf" class="anchored unlink"><strong>laplace_marginal_bernoulli_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-fd129c664f37d1cdd280abaf7ee82bfaf9c8e54e"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_bernoulli_logit_lupmf" href="#laplace_marginal_bernoulli_logit_lupmf" class="anchored unlink"><strong>laplace_marginal_bernoulli_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-53e5e8065cf24c3e72a435d5fccf4d14a587b245"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_neg_binomial_2_log" href="#laplace_marginal_neg_binomial_2_log" class="anchored unlink"><strong>laplace_marginal_neg_binomial_2_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-7f807083cfb7347ece100da5dfe719186a678d02">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_neg_binomial_2_log_lpmf" href="#laplace_marginal_neg_binomial_2_log_lpmf" class="anchored unlink"><strong>laplace_marginal_neg_binomial_2_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-b74e5e572a36211903f16f5deed7158384c21355"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_neg_binomial_2_log_lupmf" href="#laplace_marginal_neg_binomial_2_log_lupmf" class="anchored unlink"><strong>laplace_marginal_neg_binomial_2_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-5d9163f94a110e178c54b2bd192e9aded134cd3a"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_poisson_log" href="#laplace_marginal_poisson_log" class="anchored unlink"><strong>laplace_marginal_poisson_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-b2c6946b65561039eb8bf3999fb59d38a6336e10">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_poisson_log_lpmf" href="#laplace_marginal_poisson_log_lpmf" class="anchored unlink"><strong>laplace_marginal_poisson_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-ca1a086a83293f252043ab8b127ec6c9ef16deec"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_poisson_log_lupmf" href="#laplace_marginal_poisson_log_lupmf" class="anchored unlink"><strong>laplace_marginal_poisson_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-d0ecc660f5107f7409bd76ba059c5ea82ec81ba0"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol" href="#laplace_marginal_tol" class="anchored unlink"><strong>laplace_marginal_tol</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-9d40f0ee623d8ad92f054b2d83550d7a8ddf4b56"><code>(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_bernoulli_logit" href="#laplace_marginal_tol_bernoulli_logit" class="anchored unlink"><strong>laplace_marginal_tol_bernoulli_logit</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-92d43f7c3643c7d85966bbfc0b2a71546facb37c">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_bernoulli_logit_lpmf" href="#laplace_marginal_tol_bernoulli_logit_lpmf" class="anchored unlink"><strong>laplace_marginal_tol_bernoulli_logit_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-5126f7d7829072e478ea2f7257d226219b43e1d4"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_bernoulli_logit_lupmf" href="#laplace_marginal_tol_bernoulli_logit_lupmf" class="anchored unlink"><strong>laplace_marginal_tol_bernoulli_logit_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-7d93f40162970812a81fc61df49dcb53f993e293"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_neg_binomial_2_log" href="#laplace_marginal_tol_neg_binomial_2_log" class="anchored unlink"><strong>laplace_marginal_tol_neg_binomial_2_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-40016f7d5d96b9d999ca42f446720cf3f3cd5769">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_neg_binomial_2_log_lpmf" href="#laplace_marginal_tol_neg_binomial_2_log_lpmf" class="anchored unlink"><strong>laplace_marginal_tol_neg_binomial_2_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-cefee8b5275ca255f121386fb14c9354cf95594c"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_neg_binomial_2_log_lupmf" href="#laplace_marginal_tol_neg_binomial_2_log_lupmf" class="anchored unlink"><strong>laplace_marginal_tol_neg_binomial_2_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-efb046fcbadbcaaa59a274ab3c5b1ed15f23add9"><code>(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_poisson_log" href="#laplace_marginal_tol_poisson_log" class="anchored unlink"><strong>laplace_marginal_tol_poisson_log</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-4a7ecca812a308ea648bb31996559cba79738d83">distribution statement</a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_poisson_log_lpmf" href="#laplace_marginal_tol_poisson_log_lpmf" class="anchored unlink"><strong>laplace_marginal_tol_poisson_log_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-51cad53596dd0d74c0fe99b84a3c975cbec3866a"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
+<p><a id="laplace_marginal_tol_poisson_log_lupmf" href="#laplace_marginal_tol_poisson_log_lupmf" class="anchored unlink"><strong>laplace_marginal_tol_poisson_log_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/embedded_laplace.html#index-entry-23e9ed91fd3f98c314ddbce86a77005fdf8ed5f3"><code>(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real</code></a> <span class="detail">(embedded_laplace.html)</span>
+</div></li>
+</ul>
 <p><a id="lbeta" href="#lbeta" class="anchored unlink"><strong>lbeta</strong>:</a></p>
 <ul>
 <li><div class="index-container">
@@ -6474,6 +6658,12 @@ <h2 class="anchored" data-anchor-id="t">T</h2>
 <a href="../functions-reference/matrix_operations.html#index-entry-932fc0bd0a556886a853004f8ad6bc3432760b62"><code>(matrix A) : real</code></a> <span class="detail">(matrix_operations.html)</span>
 </div></li>
 </ul>
+<p><a id="trace_dot" href="#trace_dot" class="anchored unlink"><strong>trace_dot</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/matrix_operations.html#index-entry-b067f1dc33e4e60522d490bc635d4305d0e9e7fb"><code>(matrix A, matrix B) : real</code></a> <span class="detail">(matrix_operations.html)</span>
+</div></li>
+</ul>
 <p><a id="trace_gen_quad_form" href="#trace_gen_quad_form" class="anchored unlink"><strong>trace_gen_quad_form</strong>:</a></p>
 <ul>
 <li><div class="index-container">
@@ -6784,6 +6974,51 @@ <h2 class="anchored" data-anchor-id="w">W</h2>
 </div></li>
 </ul>
 </section>
+<section id="y" class="level2">
+<h2 class="anchored" data-anchor-id="y">Y</h2>
+<p><a id="yule_simon" href="#yule_simon" class="anchored unlink"><strong>yule_simon</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-0a6380458dc845b883a913fb2227d90dee08322f">distribution statement</a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_cdf" href="#yule_simon_cdf" class="anchored unlink"><strong>yule_simon_cdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-be440f4170e8f674bbbf23f2e94e8803e94cec24"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lccdf" href="#yule_simon_lccdf" class="anchored unlink"><strong>yule_simon_lccdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-3def60eb9f4fdb6c157d57d9c12a79c1be38a69e"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lcdf" href="#yule_simon_lcdf" class="anchored unlink"><strong>yule_simon_lcdf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-17da2af212f09a989d004314b9af2fe44bbe25a8"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lpmf" href="#yule_simon_lpmf" class="anchored unlink"><strong>yule_simon_lpmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-0fa5c053a53c4eb74a8936030170ca24edf5997d"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_lupmf" href="#yule_simon_lupmf" class="anchored unlink"><strong>yule_simon_lupmf</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-ab47efb6b4bbb3f4bdaa849c799ee6e10bbe6036"><code>(ints n | reals alpha) : real</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+<p><a id="yule_simon_rng" href="#yule_simon_rng" class="anchored unlink"><strong>yule_simon_rng</strong>:</a></p>
+<ul>
+<li><div class="index-container">
+<a href="../functions-reference/unbounded_discrete_distributions.html#index-entry-fb10a835dac9bc498feff76a66b8bf9737013d89"><code>(reals alpha) : R</code></a> <span class="detail">(unbounded_discrete_distributions.html)</span>
+</div></li>
+</ul>
+</section>
 <section id="z" class="level2">
 <h2 class="anchored" data-anchor-id="z">Z</h2>
 <p><a id="zeros_array" href="#zeros_array" class="anchored unlink"><strong>zeros_array</strong>:</a></p>
@@ -7387,8 +7622,35 @@ <h2 class="anchored" data-anchor-id="z">Z</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/functions_index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/functions_index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/hidden_markov_models.html b/docs/functions-reference/hidden_markov_models.html
index 639e59fb4..4b055c0fc 100644
--- a/docs/functions-reference/hidden_markov_models.html
+++ b/docs/functions-reference/hidden_markov_models.html
@@ -30,7 +30,7 @@
 <script src="../site_libs/quarto-search/fuse.min.js"></script>
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
-<link href="../functions-reference/mathematical_functions.html" rel="next">
+<link href="../functions-reference/embedded_laplace.html" rel="next">
 <link href="../functions-reference/covariance_matrix_distributions.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link active">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1132,14 +1138,41 @@ <h2 class="anchored" data-anchor-id="hmm-stan-functions">Stan functions</h2>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
-      <a href="../functions-reference/mathematical_functions.html" class="pagination-link" aria-label="Mathematical Functions">
-        <span class="nav-page-text">Mathematical Functions</span> <i class="bi bi-arrow-right-short"></i>
+      <a href="../functions-reference/embedded_laplace.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <span class="nav-page-text">Embedded Laplace Approximation</span> <i class="bi bi-arrow-right-short"></i>
       </a>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/hidden_markov_models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/hidden_markov_models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/higher-order_functions.html b/docs/functions-reference/higher-order_functions.html
index d37d0bdb1..1b73b613a 100644
--- a/docs/functions-reference/higher-order_functions.html
+++ b/docs/functions-reference/higher-order_functions.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -532,6 +532,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1630,8 +1636,35 @@ <h3 class="anchored" data-anchor-id="rectangular-map">Rectangular map</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/higher-order_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/higher-order_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/index.html b/docs/functions-reference/index.html
index 273f5b5b9..820f6d50b 100644
--- a/docs/functions-reference/index.html
+++ b/docs/functions-reference/index.html
@@ -74,7 +74,7 @@
 
 <link rel="stylesheet" href="../quarto-config/quarto_styles.css">
 <meta property="og:title" content="Stan Functions Reference – Stan Docs">
-<meta property="og:description" content="Version 2.38">
+<meta property="og:description" content="Version 2.39">
 <meta property="og:image" content="img/logo_tm.png">
 <meta property="og:site_name" content="Stan Docs">
 </head>
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -448,6 +448,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -487,7 +493,7 @@
 <header id="title-block-header" class="quarto-title-block default">
 <div class="quarto-title">
 <h1 class="title">Stan Functions Reference</h1>
-<p class="subtitle lead">Version 2.38</p>
+<p class="subtitle lead">Version 2.39</p>
 </div>
 
 
@@ -511,7 +517,7 @@ <h1 class="title">Stan Functions Reference</h1>
 <li><p>the <a href="https://mc-stan.org/docs/stan-users-guide/index.html"><strong>Stan User’s Guide</strong></a>. The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:</p></li>
 <li><p>the <a href="https://mc-stan.org/docs/reference-manual/index.html"><strong>Stan Reference Manual</strong></a>. Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.</p></li>
 </ul>
-<p><a href="https://mc-stan.org/docs/2_38/functions-reference-2_38.pdf">Download the pdf version of this manual</a>.</p>
+<p><a href="https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf">Download the pdf version of this manual</a>.</p>
 <section id="copyright-and-trademark" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
 <ul>
@@ -1102,8 +1108,35 @@ <h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/integer-valued_basic_functions.html b/docs/functions-reference/integer-valued_basic_functions.html
index 61f8975e8..4788a0d0f 100644
--- a/docs/functions-reference/integer-valued_basic_functions.html
+++ b/docs/functions-reference/integer-valued_basic_functions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1234,8 +1240,35 @@ <h2 class="anchored" data-anchor-id="casting-functions">Casting functions</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/integer-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/integer-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/mathematical_functions.html b/docs/functions-reference/mathematical_functions.html
index 47259bf12..d81ac39a9 100644
--- a/docs/functions-reference/mathematical_functions.html
+++ b/docs/functions-reference/mathematical_functions.html
@@ -31,7 +31,7 @@
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
 <link href="../functions-reference/functions_index.html" rel="next">
-<link href="../functions-reference/hidden_markov_models.html" rel="prev">
+<link href="../functions-reference/embedded_laplace.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
 <script src="../site_libs/quarto-html/popper.min.js"></script>
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1127,8 +1133,8 @@ <h2 class="anchored" data-anchor-id="digamma-appendix">Digamma</h2>
 </script>
 <nav class="page-navigation">
   <div class="nav-page nav-page-previous">
-      <a href="../functions-reference/hidden_markov_models.html" class="pagination-link" aria-label="Hidden Markov Models">
-        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Hidden Markov Models</span>
+      <a href="../functions-reference/embedded_laplace.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Embedded Laplace Approximation</span>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
@@ -1138,8 +1144,35 @@ <h2 class="anchored" data-anchor-id="digamma-appendix">Digamma</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mathematical_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mathematical_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/matrix_operations.html b/docs/functions-reference/matrix_operations.html
index c4aba8c04..40457a087 100644
--- a/docs/functions-reference/matrix_operations.html
+++ b/docs/functions-reference/matrix_operations.html
@@ -308,7 +308,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -531,6 +531,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1059,6 +1065,10 @@ <h3 class="anchored" data-anchor-id="specialized-products">Specialized products<
 <p></p>
 <p><code>real</code> <strong><a href="../functions-reference/functions_index.html#quad_form_sym" class="unlink" title="Jump to index entry"><code>quad_form_sym</code></a></strong><code>(matrix A, vector B)</code><br> Similarly to quad_form, gives <code>B' * A * B</code>, but additionally checks if A is symmetric and ensures that the result is also symmetric.</p>
 <small><i>Available since 2.3</i></small>
+<a id="index-entry-b067f1dc33e4e60522d490bc635d4305d0e9e7fb" class="index-entry"></a> <!-- real; trace_dot; (matrix A, matrix B); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace_dot" class="unlink" title="Jump to index entry"><code>trace_dot</code></a></strong><code>(matrix A, matrix B)</code><br> The trace of the matrix product, i.e., <code>trace(A * B)</code>.</p>
+<small><i>Available since 2.39</i></small>
 <a id="index-entry-9e673dba39e3d7e6297d26eda1aaa01ff737cd89" class="index-entry"></a> <!-- real; trace_quad_form; (matrix A, matrix B); -->
 <p></p>
 <p><code>real</code> <strong><a href="../functions-reference/functions_index.html#trace_quad_form" class="unlink" title="Jump to index entry"><code>trace_quad_form</code></a></strong><code>(matrix A, matrix B)</code><br> The trace of the quadratic form, i.e., <code>trace(B' * A * B)</code>.</p>
@@ -1957,7 +1967,7 @@ <h4 class="anchored" data-anchor-id="QR-decomposition">QR decomposition</h4>
 <p><code>matrix</code> <strong><a href="../functions-reference/functions_index.html#qr_R" class="unlink" title="Jump to index entry"><code>qr_R</code></a></strong><code>(matrix A)</code><br> The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A</p>
 <small><i>Available since 2.3</i></small>
 <a id="index-entry-0f3b3496deb99ba4839dce3aea55e6de916877e0" class="index-entry"></a> <!-- tuple(matrix, matrix); qr; (matrix A); -->
-<p> <code>tuple(matrix, matrix)</code> <strong><a href="../functions-reference/functions_index.html#qr" class="unlink" title="Jump to index entry"><code>qr</code></a></strong><code>(matrix A)</code><br> Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to <code>(qr_Q(A), qr_R(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
+<p> <code>tuple(matrix, matrix)</code> <strong><a href="../functions-reference/functions_index.html#qr" class="unlink" title="Jump to index entry"><code>qr</code></a></strong><code>(matrix A)</code><br> Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthogonal matrix in the fat QR decomposition and the second element (“R”) is upper trapezoidal. This function is equivalent to <code>(qr_Q(A), qr_R(A))</code> but with a lower computational cost due to the shared work between the two results.</p>
 <small><i>Available since 2.33</i></small>
 <p>The thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as <span class="math display">\[\begin{equation*} A = Q \, R. \end{equation*}\]</span> Multiplying a column of an orthogonal matrix by <span class="math inline">\(-1\)</span> still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by <span class="math inline">\(-1\)</span> without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.</p>
 </section>
@@ -2637,8 +2647,35 @@ <h2 class="anchored" data-anchor-id="reverse-functions">Reverse functions</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/mixed_operations.html b/docs/functions-reference/mixed_operations.html
index 8101005d9..af4d23e2a 100644
--- a/docs/functions-reference/mixed_operations.html
+++ b/docs/functions-reference/mixed_operations.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1338,8 +1344,35 @@ <h1>Mixed Operations</h1>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mixed_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/mixed_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/multivariate_discrete_distributions.html b/docs/functions-reference/multivariate_discrete_distributions.html
index 348a6c101..007d82fcc 100644
--- a/docs/functions-reference/multivariate_discrete_distributions.html
+++ b/docs/functions-reference/multivariate_discrete_distributions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1233,8 +1239,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-2">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/multivariate_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/multivariate_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/positive_continuous_distributions.html b/docs/functions-reference/positive_continuous_distributions.html
index c29d8e03e..6c27f5814 100644
--- a/docs/functions-reference/positive_continuous_distributions.html
+++ b/docs/functions-reference/positive_continuous_distributions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1659,8 +1665,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-10">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/positive_lower-bounded_distributions.html b/docs/functions-reference/positive_lower-bounded_distributions.html
index c658d57df..aa3c8aca2 100644
--- a/docs/functions-reference/positive_lower-bounded_distributions.html
+++ b/docs/functions-reference/positive_lower-bounded_distributions.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -498,6 +498,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1412,8 +1418,35 @@ <h3 class="anchored" data-anchor-id="tolerance-tuning">Tolerance tuning</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_lower-bounded_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/positive_lower-bounded_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/real-valued_basic_functions.html b/docs/functions-reference/real-valued_basic_functions.html
index b46001086..96f0b6ebe 100644
--- a/docs/functions-reference/real-valued_basic_functions.html
+++ b/docs/functions-reference/real-valued_basic_functions.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -532,6 +532,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -2220,8 +2226,35 @@ <h2 class="anchored" data-anchor-id="hypergeometric-functions">Hypergeometric Fu
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/real-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/real-valued_basic_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/references.html b/docs/functions-reference/references.html
index 4458bfda0..8c2cd1b2a 100644
--- a/docs/functions-reference/references.html
+++ b/docs/functions-reference/references.html
@@ -787,8 +787,35 @@ <h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">Re
 });
 </script>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/removed_functions.html b/docs/functions-reference/removed_functions.html
index 041785972..869196b5c 100644
--- a/docs/functions-reference/removed_functions.html
+++ b/docs/functions-reference/removed_functions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1169,8 +1175,35 @@ <h2 class="anchored" data-anchor-id="real-arguments-to-logical-operators-operato
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/removed_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/removed_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/simplex_distributions.html b/docs/functions-reference/simplex_distributions.html
index bd3e01b16..8d7aafd24 100644
--- a/docs/functions-reference/simplex_distributions.html
+++ b/docs/functions-reference/simplex_distributions.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -512,6 +512,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1216,8 +1222,35 @@ <h3 class="anchored" data-anchor-id="stan-functions">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/simplex_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/simplex_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/sparse_matrix_operations.html b/docs/functions-reference/sparse_matrix_operations.html
index 25534a49a..e7596edd0 100644
--- a/docs/functions-reference/sparse_matrix_operations.html
+++ b/docs/functions-reference/sparse_matrix_operations.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1177,8 +1183,35 @@ <h3 class="anchored" data-anchor-id="sparse-matrix-multiplication">Sparse matrix
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/sparse_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/sparse_matrix_operations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/transform_functions.html b/docs/functions-reference/transform_functions.html
index 43ab4cbc9..7311ee935 100644
--- a/docs/functions-reference/transform_functions.html
+++ b/docs/functions-reference/transform_functions.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -478,6 +478,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1429,8 +1435,35 @@ <h3 class="anchored" data-anchor-id="sum-to-zero-matrices">Sum-to-zero matrices<
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/transform_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/transform_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/unbounded_continuous_distributions.html b/docs/functions-reference/unbounded_continuous_distributions.html
index b08557e47..807dfed4c 100644
--- a/docs/functions-reference/unbounded_continuous_distributions.html
+++ b/docs/functions-reference/unbounded_continuous_distributions.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -498,6 +498,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1765,8 +1771,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-10">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_continuous_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/unbounded_discrete_distributions.html b/docs/functions-reference/unbounded_discrete_distributions.html
index a8a99e58b..d06a2783e 100644
--- a/docs/functions-reference/unbounded_discrete_distributions.html
+++ b/docs/functions-reference/unbounded_discrete_distributions.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -498,6 +498,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -583,6 +589,12 @@ <h2 id="toc-title">On this page</h2>
   <li><a href="#distribution-statement-7" id="toc-distribution-statement-7" class="nav-link" data-scroll-target="#distribution-statement-7">Distribution statement</a></li>
   <li><a href="#stan-functions-7" id="toc-stan-functions-7" class="nav-link" data-scroll-target="#stan-functions-7">Stan functions</a></li>
   </ul></li>
+  <li><a href="#yule-simon" id="toc-yule-simon" class="nav-link" data-scroll-target="#yule-simon">Yule Simon distribution</a>
+  <ul class="collapse">
+  <li><a href="#probability-mass-function-7" id="toc-probability-mass-function-7" class="nav-link" data-scroll-target="#probability-mass-function-7">Probability mass function</a></li>
+  <li><a href="#distribution-statement-8" id="toc-distribution-statement-8" class="nav-link" data-scroll-target="#distribution-statement-8">Distribution statement</a></li>
+  <li><a href="#stan-functions-8" id="toc-stan-functions-8" class="nav-link" data-scroll-target="#stan-functions-8">Stan functions</a></li>
+  </ul></li>
   </ul></li>
   </ul>
 <div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
@@ -986,6 +998,50 @@ <h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
 <p></p>
 <p><code>R</code> <strong><a href="../functions-reference/functions_index.html#beta_neg_binomial_rng" class="unlink" title="Jump to index entry"><code>beta_neg_binomial_rng</code></a></strong><code>(reals r, reals alpha, reals beta)</code><br> Generate a beta negative binomial variate with parameters <code>r</code>, <code>alpha</code> and <code>beta</code>; may only be used in transformed data and generated quantities blocks. <code>r</code> <span class="math inline">\(\cdot\)</span> <code>beta</code> <span class="math inline">\(/\)</span> (<code>alpha</code><span class="math inline">\(-1\)</span>) must be less than <span class="math inline">\(2 ^ {29}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
 <small><i>Available since 2.36</i></small>
+</section>
+</section>
+<section id="yule-simon" class="level2">
+<h2 class="anchored" data-anchor-id="yule-simon">Yule Simon distribution</h2>
+<section id="probability-mass-function-7" class="level3">
+<h3 class="anchored" data-anchor-id="probability-mass-function-7">Probability mass function</h3>
+<p>If <span class="math inline">\(\alpha \in \mathbb{R}^+\)</span>, then for <span class="math inline">\(n \in \mathbb{N}^+=\{1,2,...\}\)</span>, <span class="math display">\[\begin{equation*}
+\text{YuleSimon}(n|\alpha) = \alpha \, \mathrm{B}(\alpha + 1, n) = \alpha \, \frac{\Gamma(n) \, \Gamma(\alpha + 1)}{\Gamma(n + \alpha + 1)}.
+\end{equation*}\]</span></p>
+</section>
+<section id="distribution-statement-8" class="level3">
+<h3 class="anchored" data-anchor-id="distribution-statement-8">Distribution statement</h3>
+<p><code>n ~</code> <strong><a href="../functions-reference/functions_index.html#yule_simon" class="unlink" title="Jump to index entry"><code>yule_simon</code></a></strong><code>(alpha)</code></p>
+<p>Increment target log probability density with <code>yule_simon_lupmf(n | alpha)</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-0a6380458dc845b883a913fb2227d90dee08322f" class="index-entry"></a> <!-- real; yule_simon ~; -->
+<p></p>
+</section>
+<section id="stan-functions-8" class="level3">
+<h3 class="anchored" data-anchor-id="stan-functions-8">Stan functions</h3>
+<a id="index-entry-0fa5c053a53c4eb74a8936030170ca24edf5997d" class="index-entry"></a> <!-- real; yule_simon_lpmf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lpmf" class="unlink" title="Jump to index entry"><code>yule_simon_lpmf</code></a></strong><code>(ints n | reals alpha)</code><br> The log Yule Simon probability mass of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-ab47efb6b4bbb3f4bdaa849c799ee6e10bbe6036" class="index-entry"></a> <!-- real; yule_simon_lupmf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lupmf" class="unlink" title="Jump to index entry"><code>yule_simon_lupmf</code></a></strong><code>(ints n | reals alpha)</code><br> The log Yule Simon probability mass of <code>n</code> given parameter <code>alpha</code> dropping constant additive terms.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-be440f4170e8f674bbbf23f2e94e8803e94cec24" class="index-entry"></a> <!-- real; yule_simon_cdf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_cdf" class="unlink" title="Jump to index entry"><code>yule_simon_cdf</code></a></strong><code>(ints n | reals alpha)</code><br> The Yule Simon cumulative distribution function of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-17da2af212f09a989d004314b9af2fe44bbe25a8" class="index-entry"></a> <!-- real; yule_simon_lcdf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lcdf" class="unlink" title="Jump to index entry"><code>yule_simon_lcdf</code></a></strong><code>(ints n | reals alpha)</code><br> The log of the Yule Simon cumulative distribution function of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-3def60eb9f4fdb6c157d57d9c12a79c1be38a69e" class="index-entry"></a> <!-- real; yule_simon_lccdf; (ints n | reals alpha); -->
+<p></p>
+<p><code>real</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_lccdf" class="unlink" title="Jump to index entry"><code>yule_simon_lccdf</code></a></strong><code>(ints n | reals alpha)</code><br> The log of the Yule Simon complementary cumulative distribution function of <code>n</code> given parameter <code>alpha</code>.</p>
+<small><i>Available since 2.39</i></small>
+<a id="index-entry-fb10a835dac9bc498feff76a66b8bf9737013d89" class="index-entry"></a> <!-- R; yule_simon_rng; (reals alpha); -->
+<p></p>
+<p><code>R</code> <strong><a href="../functions-reference/functions_index.html#yule_simon_rng" class="unlink" title="Jump to index entry"><code>yule_simon_rng</code></a></strong><code>(reals alpha)</code><br> Generate a Yule Simon variate with parameter <code>alpha</code>; may only be used in transformed data and generated quantities blocks. <code>alpha</code> <span class="math inline">\(/\)</span> (<code>alpha</code><span class="math inline">\(-1\)</span>) must be less than <span class="math inline">\(2 ^ {29}\)</span>. For a description of argument and return types, see section <a href="../functions-reference/conventions_for_probability_functions.html#prob-vectorization">vectorized function signatures</a>.</p>
+<small><i>Available since 2.39</i></small>
 
 
 
@@ -1572,8 +1628,35 @@ <h3 class="anchored" data-anchor-id="stan-functions-7">Stan functions</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/unbounded_discrete_distributions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/functions-reference/void_functions.html b/docs/functions-reference/void_functions.html
index 5752c5b39..3cf06665a 100644
--- a/docs/functions-reference/void_functions.html
+++ b/docs/functions-reference/void_functions.html
@@ -226,7 +226,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -449,6 +449,12 @@
   <a href="../functions-reference/hidden_markov_models.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Hidden Markov Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../functions-reference/embedded_laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1108,8 +1114,35 @@ <h2 class="anchored" data-anchor-id="fatal-error-statement">Fatal error statemen
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/void_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/functions-reference/void_functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/index.html b/docs/index.html
index 8c2fee151..5ba34cc12 100644
--- a/docs/index.html
+++ b/docs/index.html
@@ -74,7 +74,7 @@
 
 <link rel="stylesheet" href="quarto-config/quarto_styles.css">
 <meta property="og:title" content="Stan Documentation – Stan Docs">
-<meta property="og:description" content="Version 2.38">
+<meta property="og:description" content="Version 2.39">
 <meta property="og:image" content="img/logo_tm.png">
 <meta property="og:site_name" content="Stan Docs">
 </head>
@@ -213,7 +213,7 @@
 <header id="title-block-header" class="quarto-title-block default">
 <div class="quarto-title">
 <h1 class="title">Stan Documentation</h1>
-<p class="subtitle lead">Version 2.38</p>
+<p class="subtitle lead">Version 2.39</p>
 </div>
 
 
@@ -239,14 +239,14 @@ <h1 class="title">Stan Documentation</h1>
 <div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
 <p>This is the official documentation for <a href="https://mc-stan.org/">Stan</a>.</p>
 <ul>
-<li><p>The <a href="stan-users-guide/index.html"><strong>Stan User’s Guide</strong></a> (<a href="https://mc-stan.org/docs/2_38/stan-users-guide-2_38.pdf">pdf</a>) provides example models and programming techniques for coding statistical models in Stan.</p></li>
-<li><p>The <a href="reference-manual/index.html"><strong>Stan Reference Manual</strong></a> (<a href="https://mc-stan.org/docs/2_38/reference-manual-2_38.pdf">pdf</a>) specifies the Stan programming language and inference algorithms.</p></li>
-<li><p>The <a href="functions-reference/index.html"><strong>Stan Functions Reference</strong></a> (<a href="https://mc-stan.org/docs/2_38/functions-reference-2_38.pdf">pdf</a>) specifies the functions built into the Stan programming language.</p></li>
+<li><p>The <a href="stan-users-guide/index.html"><strong>Stan User’s Guide</strong></a> (<a href="https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf">pdf</a>) provides example models and programming techniques for coding statistical models in Stan.</p></li>
+<li><p>The <a href="reference-manual/index.html"><strong>Stan Reference Manual</strong></a> (<a href="https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf">pdf</a>) specifies the Stan programming language and inference algorithms.</p></li>
+<li><p>The <a href="functions-reference/index.html"><strong>Stan Functions Reference</strong></a> (<a href="https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf">pdf</a>) specifies the functions built into the Stan programming language.</p></li>
 </ul>
-<p>There are also separate installation and getting started guides for <a href="cmdstan-guide/index.html"><em>CmdStan</em></a> (<a href="https://mc-stan.org/docs/2_38/cmdstan-guide-2_38.pdf">pdf</a>), the command-line interface to the Stan inference engine, and the R, Python, and Julia interfaces.</p>
+<p>There are also separate installation and getting started guides for <a href="cmdstan-guide/index.html"><em>CmdStan</em></a> (<a href="https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf">pdf</a>), the command-line interface to the Stan inference engine, and the R, Python, and Julia interfaces.</p>
 <section id="older-versions" class="level3">
 <h3 class="anchored" data-anchor-id="older-versions">Older Versions</h3>
-<p>This documentation is for Stan 2.38. Older versions of each of the documents linked above can be found in the table below:</p>
+<p>This documentation is for Stan 2.39. Older versions of each of the documents linked above can be found in the table below:</p>
 <table class="caption-top table">
 <thead>
 <tr class="header">
@@ -259,146 +259,153 @@ <h3 class="anchored" data-anchor-id="older-versions">Older Versions</h3>
 </thead>
 <tbody>
 <tr class="odd">
+<td>2.39</td>
+<td><a href="https://mc-stan.org/docs/2_39/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_39/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_39/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf">pdf</a></td>
+<td><a href="https://mc-stan.org/docs/2_39/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf">pdf</a></td>
+</tr>
+<tr class="even">
 <td>2.38</td>
 <td><a href="https://mc-stan.org/docs/2_38/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_38/reference-manual-2_38.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_38/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_38/stan-users-guide-2_38.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_38/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_38/cmdstan-guide-2_38.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_38/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_38/functions-reference-2_38.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.37</td>
 <td><a href="https://mc-stan.org/docs/2_37/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_37/reference-manual-2_37.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_37/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_37/stan-users-guide-2_37.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_37/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_37/cmdstan-guide-2_37.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_37/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_37/functions-reference-2_37.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.36</td>
 <td><a href="https://mc-stan.org/docs/2_36/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_36/reference-manual-2_36.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_36/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_36/stan-users-guide-2_36.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_36/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_36/cmdstan-guide-2_36.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_36/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_36/functions-reference-2_36.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.35</td>
 <td><a href="https://mc-stan.org/docs/2_35/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_35/reference-manual-2_35.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_35/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_35/stan-users-guide-2_35.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_35/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_35/cmdstan-guide-2_35.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_35/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_35/functions-reference-2_35.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.34</td>
 <td><a href="https://mc-stan.org/docs/2_34/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_34/reference-manual-2_34.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_34/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_34/stan-users-guide-2_34.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_34/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_34/cmdstan-guide-2_34.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_34/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_34/functions-reference-2_34.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.33</td>
 <td><a href="https://mc-stan.org/docs/2_33/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_33/reference-manual-2_33.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_33/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_33/stan-users-guide-2_33.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_33/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_33/cmdstan-guide-2_33.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_33/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_33/functions-reference-2_33.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.32</td>
 <td><a href="https://mc-stan.org/docs/2_32/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_32/reference-manual-2_32.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_32/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_32/stan-users-guide-2_32.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_32/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_32/cmdstan-guide-2_32.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_32/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_32/functions-reference-2_32.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.31</td>
 <td><a href="https://mc-stan.org/docs/2_31/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_31/reference-manual-2_31.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_31/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_31/stan-users-guide-2_31.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_31/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_31/cmdstan-guide-2_31.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_31/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_31/functions-reference-2_31.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.30</td>
 <td><a href="https://mc-stan.org/docs/2_30/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_30/reference-manual-2_30.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_30/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_30/stan-users-guide-2_30.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_30/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_30/cmdstan-guide-2_30.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_30/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_30/functions-reference-2_30.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.29</td>
 <td><a href="https://mc-stan.org/docs/2_29/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_29/reference-manual-2_29.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_29/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_29/stan-users-guide-2_29.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_29/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_29/cmdstan-guide-2_29.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_29/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_29/functions-reference-2_29.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.28</td>
 <td><a href="https://mc-stan.org/docs/2_28/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_28/reference-manual-2_28.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_28/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_28/stan-users-guide-2_28.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_28/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_28/cmdstan-guide-2_28.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_28/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_28/functions-reference-2_28.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.27</td>
 <td><a href="https://mc-stan.org/docs/2_27/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_27/reference-manual-2_27.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_27/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_27/stan-users-guide-2_27.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_27/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_27/cmdstan-guide-2_27.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_27/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_27/functions-reference-2_27.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.26</td>
 <td><a href="https://mc-stan.org/docs/2_26/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_26/reference-manual-2_26.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_26/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_26/stan-users-guide-2_26.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_26/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_26/cmdstan-guide-2_26.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_26/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_26/functions-reference-2_26.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.25</td>
 <td><a href="https://mc-stan.org/docs/2_25/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_25/reference-manual-2_25.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_25/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_25/stan-users-guide-2_25.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_25/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_25/cmdstan-guide-2_25.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_25/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_25/functions-reference-2_25.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.24</td>
 <td><a href="https://mc-stan.org/docs/2_24/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_24/reference-manual-2_24.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_24/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_24/stan-users-guide-2_24.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_24/cmdstan-guide/">html</a> <a href="https://mc-stan.org/docs/2_24/cmdstan-guide-2_24.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_24/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_24/functions-reference-2_24.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.23</td>
 <td><a href="https://mc-stan.org/docs/2_23/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_23/reference-manual-2_23.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_23/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_23/stan-users-guide-2_23.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_23/cmdstan-guide/">html</a></td>
 <td><a href="https://mc-stan.org/docs/2_23/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_23/functions-reference-2_23.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.22</td>
 <td><a href="https://mc-stan.org/docs/2_22/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_22/reference-manual-2_22.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_22/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_22/stan-users-guide-2_22.pdf">pdf</a></td>
 <td></td>
 <td><a href="https://mc-stan.org/docs/2_22/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_22/functions-reference-2_22.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.21</td>
 <td><a href="https://mc-stan.org/docs/2_21/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_21/reference-manual-2_21.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_21/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_21/stan-users-guide-2_21.pdf">pdf</a></td>
 <td></td>
 <td><a href="https://mc-stan.org/docs/2_21/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_21/functions-reference-2_21.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.20</td>
 <td><a href="https://mc-stan.org/docs/2_20/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_20/reference-manual-2_20.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_20/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_20/stan-users-guide-2_20.pdf">pdf</a></td>
 <td></td>
 <td><a href="https://mc-stan.org/docs/2_20/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_20/functions-reference-2_20.pdf">pdf</a></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td>2.19</td>
 <td><a href="https://mc-stan.org/docs/2_19/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_19/reference-manual-2_19.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_19/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_19/stan-users-guide-2_19.pdf">pdf</a></td>
 <td></td>
 <td><a href="https://mc-stan.org/docs/2_19/functions-reference/">html</a> <a href="https://mc-stan.org/docs/2_19/functions-reference-2_19.pdf">pdf</a></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td>2.18</td>
 <td><a href="https://mc-stan.org/docs/2_18/reference-manual/">html</a> <a href="https://mc-stan.org/docs/2_18/reference-manual-2_18.pdf">pdf</a></td>
 <td><a href="https://mc-stan.org/docs/2_18/stan-users-guide/">html</a> <a href="https://mc-stan.org/docs/2_18/stan-users-guide-2_18.pdf">pdf</a></td>
@@ -1088,8 +1095,35 @@ <h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
 });
 </script>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/analysis.html b/docs/reference-manual/analysis.html
index 2831fdcf7..70de8e8bb 100644
--- a/docs/reference-manual/analysis.html
+++ b/docs/reference-manual/analysis.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -416,6 +416,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1319,8 +1325,35 @@ <h3 class="unnumbered anchored" data-anchor-id="thinning">Thinning</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/blocks.html b/docs/reference-manual/blocks.html
index cab4af7f5..15e8de1b5 100644
--- a/docs/reference-manual/blocks.html
+++ b/docs/reference-manual/blocks.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -450,6 +450,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1516,8 +1522,35 @@ <h2 class="anchored" data-anchor-id="program-block-generated-quantities">Program
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/blocks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/blocks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/comments.html b/docs/reference-manual/comments.html
index 9cb0493df..ac1c5115d 100644
--- a/docs/reference-manual/comments.html
+++ b/docs/reference-manual/comments.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -401,6 +401,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1051,8 +1057,35 @@ <h2 class="anchored" data-anchor-id="bracketed-comments">Bracketed comments</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/comments.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/comments.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/deprecations.html b/docs/reference-manual/deprecations.html
index 9ba0faea4..eca6bfe04 100644
--- a/docs/reference-manual/deprecations.html
+++ b/docs/reference-manual/deprecations.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -401,6 +401,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1070,8 +1076,35 @@ <h2 class="anchored" data-anchor-id="deprecated-functions">Deprecated Functions<
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/deprecations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/deprecations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/diagnostics.html b/docs/reference-manual/diagnostics.html
index 3feadead2..d2888edbd 100644
--- a/docs/reference-manual/diagnostics.html
+++ b/docs/reference-manual/diagnostics.html
@@ -31,7 +31,7 @@
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
 <link href="../reference-manual/reproducibility.html" rel="next">
-<link href="../reference-manual/laplace.html" rel="prev">
+<link href="../reference-manual/laplace_embedded.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
 <script src="../site_libs/quarto-html/popper.min.js"></script>
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -396,6 +396,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1081,8 +1087,8 @@ <h2 class="anchored" data-anchor-id="speed-warning-and-data-trimming">Speed warn
 </script>
 <nav class="page-navigation">
   <div class="nav-page nav-page-previous">
-      <a href="../reference-manual/laplace.html" class="pagination-link" aria-label="Laplace Approximation">
-        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Laplace Approximation</span>
+      <a href="../reference-manual/laplace_embedded.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Embedded Laplace Approximation</span>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
@@ -1092,8 +1098,35 @@ <h2 class="anchored" data-anchor-id="speed-warning-and-data-trimming">Speed warn
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/diagnostics.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/diagnostics.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/encoding.html b/docs/reference-manual/encoding.html
index 4c2d756d6..9d12a9011 100644
--- a/docs/reference-manual/encoding.html
+++ b/docs/reference-manual/encoding.html
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -366,6 +366,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1025,8 +1031,35 @@ <h2 class="anchored" data-anchor-id="string-literals">String literals</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/encoding.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/encoding.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/execution.html b/docs/reference-manual/execution.html
index 4f0598b44..04dd08047 100644
--- a/docs/reference-manual/execution.html
+++ b/docs/reference-manual/execution.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -450,6 +450,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1223,8 +1229,35 @@ <h3 class="unnumbered anchored" data-anchor-id="write">Write</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/execution.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/execution.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/expressions.html b/docs/reference-manual/expressions.html
index 273c729f1..08aab3a8f 100644
--- a/docs/reference-manual/expressions.html
+++ b/docs/reference-manual/expressions.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -450,6 +450,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -773,7 +779,7 @@ <h3 class="unnumbered anchored" data-anchor-id="tuple-expressions-and-types">Tup
 <p>Stan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.</p>
 <div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">vector</span>[<span class="dv">3</span>]) xy = (<span class="dv">42</span>, [<span class="dv">1</span>, <span class="fl">2.9</span>, -<span class="fl">1.3</span>]');</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The expression <code>42</code> is of type <code>int</code> and the expression <code>[1, 2.9, -1.3]</code> is of type <code>row_vector</code> so that <code>[1, 2.9, -1.3]'</code> is of type <code>vector</code> and of size 3. The whole tuple expression <code>(42, [1, 2.9, -1.3]')</code> thus has a sized type of <code>tuple(int, vector[3])</code> and an unsized type (e.g., for a function argument) of <code>tuple(int, vector)</code>.</p>
-<p>Stan does <em>not</em> support the Python notation with trailing commas, such as <code>(1, 2, 3, )</code> for a 3-tuple.</p>
+<p>A tuple of one element can be created using the same style as languages like Python, with a trailing comma, e.g., <code>(3.14,)</code>. For longer tuples, Stan does <em>not</em> support trailing commas.</p>
 </section>
 <section id="restrictions-on-values" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="restrictions-on-values">Restrictions on values</h3>
@@ -790,9 +796,10 @@ <h4 class="unnumbered anchored" data-anchor-id="no-empty-array-expressions">No e
 <p>Because there is no way to infer the type of the result, the empty array expression (<code>{ }</code>) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.</p>
 <div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">0</span>] <span class="dt">int</span> a;   <span class="co">// a is fully defined as zero element array</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
-<section id="no-zero-tuples-or-one-tuples" class="level4 unnumbered">
-<h4 class="unnumbered anchored" data-anchor-id="no-zero-tuples-or-one-tuples">No zero-tuples or one-tuples</h4>
-<p>There is no way to declare or construct a zero-tuple or one-tuple in Stan. Tuples must be at least two elements long. The expression <code>()</code> does not pick out a zero-tuple—it is ill formed. Similarly, the expression <code>(1)</code> is of type <code>int</code> rather than a tuple.</p>
+<section id="no-zero-tuples" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="no-zero-tuples">No zero-tuples</h4>
+<p>There is no way to declare or construct a zero-tuple in Stan. Tuples must be at least one element long. The expression <code>()</code> does not pick out a zero-tuple—it is ill formed.</p>
+<p>One-tuples need a trailing comma, like <code>(1,)</code>. The expression <code>(1)</code>, without the comma, is of type <code>int</code> rather than a tuple.</p>
 </section>
 </section>
 </section>
@@ -2252,8 +2259,35 @@ <h3 class="unnumbered anchored" data-anchor-id="diagnosing-problems-with-derivat
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/expressions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/expressions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/includes.html b/docs/reference-manual/includes.html
index 2b2fd3e95..d34a68824 100644
--- a/docs/reference-manual/includes.html
+++ b/docs/reference-manual/includes.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -401,6 +401,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1112,8 +1118,35 @@ <h3 class="unnumbered anchored" data-anchor-id="slashes-in-include-paths">Slashe
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/includes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/includes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/index.html b/docs/reference-manual/index.html
index 3697e8e63..c4a192f88 100644
--- a/docs/reference-manual/index.html
+++ b/docs/reference-manual/index.html
@@ -74,7 +74,7 @@
 
 <link rel="stylesheet" href="../quarto-config/quarto_styles.css">
 <meta property="og:title" content="Stan Reference Manual – Stan Docs">
-<meta property="og:description" content="Version 2.38">
+<meta property="og:description" content="Version 2.39">
 <meta property="og:image" content="img/logo_tm.png">
 <meta property="og:site_name" content="Stan Docs">
 </head>
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -366,6 +366,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -411,7 +417,7 @@
 <header id="title-block-header" class="quarto-title-block default">
 <div class="quarto-title">
 <h1 class="title">Stan Reference Manual</h1>
-<p class="subtitle lead">Version 2.38</p>
+<p class="subtitle lead">Version 2.39</p>
 </div>
 
 
@@ -431,7 +437,7 @@ <h1 class="title">Stan Reference Manual</h1>
 <div class="page-columns page-full"><p></p><div class="no-row-height column-margin column-container"><img src="img/logo_tm.png" title="Stan Logo" class="d-none d-md-block img-fluid" width="225"></div></div>
 <p>This is the official reference manual for <a href="https://mc-stan.org/">Stan</a>’s <em>programming language</em> for coding probability models, <em>inference algorithms</em> for fitting models and making predictions, and <em>posterior analysis</em> tools for evaluating the results. This manual applies to all Stan interfaces.</p>
 <p>The first part of the reference manual provides a full specification of the Stan programming language. The language is responsible for defining a log density function conditioned on data. Typically, this is a Bayesian posterior, but it may also be a penalized likelihood function. The second part of the manual specifies the inference algorithms and posterior inference tools. The third part provides auxiliary information about the use of Stan.</p>
-<p><a href="https://mc-stan.org/docs/2_38/reference-manual-2_38.pdf">Download the pdf version of this manual</a>.</p>
+<p><a href="https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf">Download the pdf version of this manual</a>.</p>
 <section id="copyright-and-trademark" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
 <ul>
@@ -1022,8 +1028,35 @@ <h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/laplace.html b/docs/reference-manual/laplace.html
index 533cc32b2..14b05ecbd 100644
--- a/docs/reference-manual/laplace.html
+++ b/docs/reference-manual/laplace.html
@@ -30,7 +30,7 @@
 <script src="../site_libs/quarto-search/fuse.min.js"></script>
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
-<link href="../reference-manual/diagnostics.html" rel="next">
+<link href="../reference-manual/laplace_embedded.html" rel="next">
 <link href="../reference-manual/variational.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -396,6 +396,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link active">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1032,14 +1038,41 @@ <h1>Laplace Approximation</h1>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
-      <a href="../reference-manual/diagnostics.html" class="pagination-link" aria-label="Diagnostic Mode">
-        <span class="nav-page-text">Diagnostic Mode</span> <i class="bi bi-arrow-right-short"></i>
+      <a href="../reference-manual/laplace_embedded.html" class="pagination-link" aria-label="Embedded Laplace Approximation">
+        <span class="nav-page-text">Embedded Laplace Approximation</span> <i class="bi bi-arrow-right-short"></i>
       </a>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/laplace_embedded.html b/docs/reference-manual/laplace_embedded.html
new file mode 100644
index 000000000..9e3a54307
--- /dev/null
+++ b/docs/reference-manual/laplace_embedded.html
@@ -0,0 +1,1226 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Embedded Laplace Approximation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../reference-manual/diagnostics.html" rel="next">
+<link href="../reference-manual/laplace.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Embedded Laplace Approximation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../stan-users-guide/index.html"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../reference-manual/index.html" aria-current="page"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace_embedded.html">Embedded Laplace Approximation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Reference Manual</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Language</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/encoding.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Character Encoding</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/includes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Includes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/comments.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Comments</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/whitespace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Whitespace</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/types.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Data Types and Declarations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/expressions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Expressions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/statements.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Statements</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/blocks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Blocks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/transforms.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Constraint Transforms</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/syntax.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Language Syntax</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/execution.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Program Execution</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/deprecations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Deprecated Features</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/removals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Removed Features</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Algorithms</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/mcmc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">MCMC Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/optimization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Optimization</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/pathfinder.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Pathfinder</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/variational.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Variational Inference</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/diagnostics.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Diagnostic Mode</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Usage</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/reproducibility.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reproducibility</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/licenses.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Licenses and Dependencies</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#embedded-laplace-approximation" id="toc-embedded-laplace-approximation" class="nav-link active" data-scroll-target="#embedded-laplace-approximation">Embedded Laplace Approximation</a>
+  <ul class="collapse">
+  <li><a href="#latent-gaussian-models" id="toc-latent-gaussian-models" class="nav-link" data-scroll-target="#latent-gaussian-models">Latent Gaussian models</a></li>
+  <li><a href="#approximation-of-the-conditional-posterior-and-marginal-likelihood" id="toc-approximation-of-the-conditional-posterior-and-marginal-likelihood" class="nav-link" data-scroll-target="#approximation-of-the-conditional-posterior-and-marginal-likelihood">Approximation of the conditional posterior and marginal likelihood</a></li>
+  <li><a href="#trade-offs-of-the-approximation" id="toc-trade-offs-of-the-approximation" class="nav-link" data-scroll-target="#trade-offs-of-the-approximation">Trade-offs of the approximation</a>
+  <ul class="collapse">
+  <li><a href="#when-the-approximation-is-appropriate" id="toc-when-the-approximation-is-appropriate" class="nav-link" data-scroll-target="#when-the-approximation-is-appropriate">When the approximation is appropriate</a></li>
+  </ul></li>
+  <li><a href="#details-of-the-approximation" id="toc-details-of-the-approximation" class="nav-link" data-scroll-target="#details-of-the-approximation">Details of the approximation</a>
+  <ul class="collapse">
+  <li><a href="#tuning-the-newton-solver" id="toc-tuning-the-newton-solver" class="nav-link" data-scroll-target="#tuning-the-newton-solver">Tuning the Newton solver</a></li>
+  <li><a href="#solver-strategies" id="toc-solver-strategies" class="nav-link" data-scroll-target="#solver-strategies">Solver Strategies</a></li>
+  <li><a href="#sparse-hessian-of-the-log-likelihood" id="toc-sparse-hessian-of-the-log-likelihood" class="nav-link" data-scroll-target="#sparse-hessian-of-the-log-likelihood">Sparse Hessian of the log likelihood</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace_embedded.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../reference-manual/mcmc.html">Algorithms</a></li><li class="breadcrumb-item"><a href="../reference-manual/laplace_embedded.html">Embedded Laplace Approximation</a></li></ol></nav></header>
+
+
+
+
+<section id="embedded-laplace-approximation" class="level1">
+<h1>Embedded Laplace Approximation</h1>
+<p>The embedded Laplace approximation replaces explicit sampling of potentially high-dimensional Gaussian latent variables with a local Gaussian approximation, marginalizing them out so that inference proceeds over the remaining hyperparameters alone. This approach is often referred to as the integrated Laplace approximation, although the exact details of the method can vary. The details of Stan’s implementation can be found in references <span class="citation" data-cites="Margossian:2020 Margossian:2023">(<a href="#ref-Margossian:2020" role="doc-biblioref">Margossian et al. 2020</a>; <a href="#ref-Margossian:2023" role="doc-biblioref">Margossian 2023</a>)</span>.</p>
+<p>A standard approach to fit a latent Gaussian model would be to perform inference jointly over the latent Gaussian variables and the hyperparameters. Instead, the embedded Laplace approximation can be used to do <em>approximate</em> marginalization of the latent Gaussian variables; we can then use any inference over the remaining hyperparameters. By marginalizing out the latent variables, the sampler explores a lower-dimensional, better-behaved marginal posterior. Individual iterations are more expensive (each requires an inner optimization), but the sampler typically needs far fewer iterations to achieve the same effective sample size. How this trade-off resolves depends on the specific problem at hand.</p>
+<p>For complete function signatures and the built-in likelihood wrappers (Poisson, Negative Binomial, Bernoulli), see the <a href="../functions-reference/embedded_laplace.html">Embedded Laplace functions reference</a>. For worked examples with full data blocks, see the <a href="../stan-users-guide/gaussian-processes.html#poisson-gp-using-an-embedded-laplace-approximation">Gaussian Processes chapter</a>.</p>
+<section id="latent-gaussian-models" class="level2">
+<h2 class="anchored" data-anchor-id="latent-gaussian-models">Latent Gaussian models</h2>
+<p>The embedded Laplace approximation is used for latent Gaussian models. A latent Gaussian model is defined by three components:</p>
+<ul>
+<li><span class="math inline">\(\phi\)</span>: hyperparameters (e.g., GP kernel length-scale and magnitude, or variance components in a hierarchical model),</li>
+<li><span class="math inline">\(\theta\)</span>: latent Gaussian variables (the high-dimensional quantity to be marginalized out),</li>
+<li><span class="math inline">\(y\)</span>: observed data.</li>
+</ul>
+<p>These components are related through a hierarchical structure. The hyperparameters <span class="math inline">\(\phi\)</span> are given a prior <span class="math inline">\(p(\phi)\)</span>. The latent variables <span class="math inline">\(\theta\)</span> have a multivariate normal prior centered at 0 with covariance matrix <span class="math inline">\(K(\phi)\)</span>. An non-zero mean offset can be incorporated into the likelihood function. The observations <span class="math inline">\(y\)</span> have a data model <span class="math inline">\(p(y \mid \theta, \phi)\)</span>. The prior on <span class="math inline">\(\theta\)</span> is centered at zero; an offset can be incorporated into the data model if a non-zero mean is needed.</p>
+<p><span class="math display">\[\begin{eqnarray*}
+  \phi &amp; \sim &amp; p(\phi) \\
+  \theta &amp; \sim &amp; \text{Multi-Normal}(0, K(\phi)) \\
+  y &amp; \sim &amp; p(y \mid \theta, \phi).
+\end{eqnarray*}\]</span></p>
+<p>The generative model above defines a joint distribution over all three quantities, <span class="math inline">\(p(\phi, \theta, y) = p(\phi) \, p(\theta \mid \phi) \, p(y \mid \theta, \phi)\)</span>. After observing data <span class="math inline">\(y\)</span>, Bayes’ theorem gives the joint posterior <span class="math inline">\(p(\phi, \theta \mid y) \propto p(\phi) \, p(\theta \mid \phi) \, p(y \mid \theta, \phi)\)</span>, where <span class="math inline">\(p(y \mid \theta, \phi)\)</span> as function of <span class="math inline">\(\theta\)</span> and <span class="math inline">\(\phi\)</span> is the joint likelihood function.</p>
+<p>Sampling directly from the joint posterior <span class="math inline">\(p(\phi, \theta \mid y)\)</span> of this model is often difficult. Challenging geometries (e.g., funnels) frustrate inference algorithms, including Hamiltonian Monte Carlo and variational inference. However, the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span> is often well-behaved and low-dimensional, making it much easier to sample. With an embedded Laplace approximation, we can obtain an approximation of the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span>. This is done via an intermediate approximation of the conditional posterior <span class="math inline">\(p(\theta \mid \phi, y)\)</span> by a normal distribution and this normal approximation is well-justified when the likelihood <span class="math inline">\(p(y \mid \theta, \phi)\)</span> as function of <span class="math inline">\(\theta\)</span> is log concave (given that we already have a normal prior <span class="math inline">\(p(\theta \mid \phi)\)</span>). Once we obtain (approximate) samples <span class="math inline">\(\phi^{(i)} \sim p(\phi \mid y)\)</span>, we can in turn generate posterior draws for <span class="math inline">\(\theta\)</span> using the normal approximation to <span class="math inline">\(p(\theta \mid y, \phi)\)</span>, evaluated at the posterior draws <span class="math inline">\(\phi^{(i)}\)</span>.</p>
+</section>
+<section id="approximation-of-the-conditional-posterior-and-marginal-likelihood" class="level2">
+<h2 class="anchored" data-anchor-id="approximation-of-the-conditional-posterior-and-marginal-likelihood">Approximation of the conditional posterior and marginal likelihood</h2>
+<p>The two-step inference strategy for using embedded Laplace in a latent Gaussian model requires approximations to both the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> and the marginal likelihood <span class="math inline">\(p(y \mid \phi)\)</span>. The Laplace approximation is the normal distribution that matches the mode and curvature of the conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span>. The mode, defined as the value of <span class="math inline">\(\theta\)</span> that maximizes the conditional posterior, is estimated by a Newton solver, <span class="math display">\[
+  \theta^* = \underset{\theta}{\text{argmax}} \ p(\theta \mid y, \phi),
+\]</span></p>
+<p>Since the approximation is normal, the curvature is matched by setting the covariance to the negative Hessian of the log conditional posterior, evaluated at the mode,</p>
+<p><span class="math display">\[
+  \Sigma^* = -  \left . \frac{\partial^2}{\partial \theta^2}
+    \log p (\theta \mid \phi, y) \right |_{\theta =\theta^*}.
+\]</span></p>
+<p>The resulting Laplace approximation is a multivariate normal centered at the mode with covariance given by the inverse curvature,</p>
+<p><span class="math display">\[
+\hat p_\mathcal{L} (\theta \mid y, \phi) = \text{Multi-Normal}(\theta^*, \Sigma^*)
+\approx p(\theta \mid y, \phi).
+\]</span></p>
+<p>This approximation also yields an approximation to the marginal likelihood, obtained by evaluating the prior, likelihood, and approximate posterior at the mode <span class="math inline">\(\theta^*\)</span>,</p>
+<p><span class="math display">\[
+   \hat p_\mathcal{L}(y \mid \phi) := \frac{p(\theta^* \mid \phi) \
+   p(y \mid \theta^*, \phi) }{ \hat p_\mathcal{L} (\theta^* \mid \phi, y) }
+   \approx p(y \mid \phi).
+\]</span></p>
+<p>Hence, a strategy to approximate the posterior of the latent Gaussian model is to first estimate the marginal posterior <span class="math inline">\(\hat p_\mathcal{L}(\phi \mid y) \propto p(\phi) p_\mathcal{L} (y \mid \phi)\)</span> using any algorithm supported by Stan. Approximate posterior draws for the latent Gaussian variables are then obtained by first sampling <span class="math inline">\(\phi \sim \hat p_\mathcal{L}(\phi \mid y)\)</span> and then <span class="math inline">\(\theta \sim  \hat p_\mathcal{L}(\theta \mid \phi, y)\)</span>.</p>
+</section>
+<section id="trade-offs-of-the-approximation" class="level2">
+<h2 class="anchored" data-anchor-id="trade-offs-of-the-approximation">Trade-offs of the approximation</h2>
+<p>The embedded Laplace approximation presents several trade-offs with standard inference over the joint posterior <span class="math inline">\(p(\theta, \phi \mid y)\)</span>. The main advantage of the embedded Laplace approximation is that it side-steps the intricate geometry of hierarchical models. The marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span> can then be handled by Hamiltonian Monte Carlo sampling without extensive tuning or reparameterization, and the mixing time is faster, meaning we can run shorter chains to achieve a desired precision. One additional benefit is that approximate methods, e.g.&nbsp;variational inference, which work poorly on the joint <span class="math inline">\(p(\theta, \phi \mid y)\)</span> can work well on the marginal posterior <span class="math inline">\(p(\phi \mid y)\)</span>.</p>
+<p>On the other hand, the embedded Laplace approximation presents certain disadvantages. First, we need to perform a Laplace approximation each time the log marginal likelihood is evaluated, meaning each iteration can be expensive. Secondly, the approximation can introduce non-negligible error, especially with non-log-concave likelihoods (note the prior is always multivariate normal). How these trade-offs are resolved depends on the application; see <span class="citation" data-cites="Margossian:2020">Margossian et al. (<a href="#ref-Margossian:2020" role="doc-biblioref">2020</a>)</span> for some examples.</p>
+<section id="when-the-approximation-is-appropriate" class="level3">
+<h3 class="anchored" data-anchor-id="when-the-approximation-is-appropriate">When the approximation is appropriate</h3>
+<p>The quality of the Laplace approximation depends on how close the true conditional posterior <span class="math inline">\(p(\theta \mid y, \phi)\)</span> is to Gaussian.</p>
+<p><strong>Works well.</strong> Log-concave likelihoods, for example from a Poisson model with log link or negative binomial with log link. These produce unimodal conditional posteriors when combined with a Gaussian prior. The approximation error is typically negligible with these likelihoods, especially with moderate-to-large counts <span class="citation" data-cites="Kuss:2005 Vanhatalo:2010 Cseke:2011 Vehtari:2016">(<a href="#ref-Kuss:2005" role="doc-biblioref">Kuss and Rasmussen 2005</a>; <a href="#ref-Vanhatalo:2010" role="doc-biblioref">Vanhatalo, Pietiläinen, and Vehtari 2010</a>; <a href="#ref-Cseke:2011" role="doc-biblioref">Cseke and Heskes 2011</a>; <a href="#ref-Vehtari:2016" role="doc-biblioref">Vehtari et al. 2016</a>)</span>. If the likelihood is normal, there is no error in the approximation but in this case the marginalization can be worked analytically and the resulting implementation is much faster than using the embedded Laplace approximation.</p>
+<p><strong>Works adequately.</strong> Bernoulli model with logit link has technically log-concave likelihood, but the likelihood can be very skewed making the Gaussian approximation less accurate than for count data. The embedded Laplace is still useful when <span class="math inline">\(\theta\)</span> is high-dimensional and joint sampling is infeasible; see <span class="citation" data-cites="Vehtari:2016">Vehtari et al. (<a href="#ref-Vehtari:2016" role="doc-biblioref">2016</a>)</span> and <span class="citation" data-cites="Margossian:2020">Margossian et al. (<a href="#ref-Margossian:2020" role="doc-biblioref">2020</a>)</span> for discussion.</p>
+<p><strong>Not appropriate.</strong> For likelihoods that are not log-concave in <span class="math inline">\(\theta\)</span>, the conditional posterior may be multimodal and the Newton solver finds only a single mode or can fail completely. When <span class="math inline">\(\theta\)</span> is low-dimensional (a few dozen or fewer), the overhead of the inner optimization may not pay for itself and standard joint HMC sampling is often adequate.</p>
+</section>
+</section>
+<section id="details-of-the-approximation" class="level2">
+<h2 class="anchored" data-anchor-id="details-of-the-approximation">Details of the approximation</h2>
+<p>When the embedded Laplace approximation does not converge or produces unexpected results, the solver configuration may need adjustment. This section describes the internals of the Newton solver and the options available for tuning it.</p>
+<section id="tuning-the-newton-solver" class="level3">
+<h3 class="anchored" data-anchor-id="tuning-the-newton-solver">Tuning the Newton solver</h3>
+<p>A critical component of the embedded Laplace approximation is the Newton solver used to estimate the mode <span class="math inline">\(\theta^*\)</span> of <span class="math inline">\(p(\theta \mid \phi, y)\)</span>. The objective function being maximized is the log joint density of the prior and likelihood with respect to <span class="math inline">\(\theta\)</span>.</p>
+<p><span class="math display">\[
+\Psi(\theta) = \log p(\theta \mid \phi) + \log p(y \mid \theta, \phi),
+\]</span></p>
+<p>Convergence is declared when the change in the objective between successive iterations falls below a <em>tolerance</em> <span class="math inline">\(\Delta\)</span>.</p>
+<p><span class="math display">\[
+| \Psi (\theta^{(i + 1)}) - \Psi (\theta^{(i)}) | \le \Delta.
+\]</span></p>
+<p>The solver also stops after reaching a pre-specified <em>maximum number of steps</em>. In that case, Stan throws a warning, but still returns the last iteration’s parameters. If you see this warning you should check the diagnostics to understand why the solver failed to converge.</p>
+<p>To help with cases where the Newton step does not lead to a decrease in the objective function, the Newton iteration is augmented with a wolfe line-search to ensure that at each iteration the objective function <span class="math inline">\(\Psi\)</span> decreases. Specifically, suppose the objective increases after a Newton step, indicating the step overshot a region of improvement,</p>
+<p><span class="math display">\[
+\Psi (\theta^{(i + 1)}) &lt; \Psi (\theta^{(i)}).
+\]</span></p>
+<p>This can indicate that the Newton step <span class="math inline">\(\alpha\)</span> at iteration <span class="math inline">\(i\)</span> is too large and that we skipped a region where the objective function decreases. In that case, we can fallback to a Wolfe line search to find a step size which satisfies the Wolfe conditions. The wolfe line search attempts to find a search direction <span class="math inline">\(p_i\)</span> and step size <span class="math inline">\(\alpha_k\)</span> such that an accepted step both increases our objective while ensuring the slope of the accepted step is flatter than our previous position. Together these help push the algorithm towards a minimum.</p>
+<p><span class="math display">\[
+f(x_i + \alpha_k p_i) \le f(x_i) + c_1 \alpha_k \nabla f(x_i)^T p_i
+-p^T_i \Delta f(x_i + \alpha_k p_i) \le -c_2 p^T_i \Delta f(x_i)
+\]</span></p>
+<p><span class="math display">\[
+  \theta^{(i + 1)} \leftarrow \frac{\theta^{(i + 1)} + \theta^{(i)}}{2}.
+\]</span></p>
+<p>We repeat this halving of steps until <span class="math inline">\(\Psi (\theta^{(i + 1)}) \ge \Psi (\theta^{(i)})\)</span>, or until a maximum number of linesearch steps is reached. For certain problems, adding a linesearch can make the optimization more stable.</p>
+</section>
+<section id="solver-strategies" class="level3">
+<h3 class="anchored" data-anchor-id="solver-strategies">Solver Strategies</h3>
+<p>The embedded Laplace approximation uses a custom Newton solver, specialized to find the mode of <span class="math inline">\(p(\theta \mid \phi, y)\)</span>. A key step for efficient optimization is to ensure all matrix inversions are numerically stable. This can be done using the Woodbury-Sherman-Morrison formula and requires one of three matrix decompositions:</p>
+<ol type="1">
+<li><p>Cholesky decomposition of the Hessian of the negative log likelihood <span class="math inline">\(W = - \partial^2_\theta \log p(y \mid \theta, \phi)\)</span>.</p></li>
+<li><p>Cholesky decomposition of the prior covariance matrix <span class="math inline">\(K(\phi)\)</span>.</p></li>
+<li><p>LU-decomposition of <span class="math inline">\(I + KW\)</span>, where <span class="math inline">\(I\)</span> is the identity matrix.</p></li>
+</ol>
+<p>The first solver (1) should be used if the negative log likelihood is positive-definite. Otherwise the user should rely on (2). In rarer cases where it is not numerically safe to invert the covariance matrix <span class="math inline">\(K\)</span>, users can use the third solver as a last-resort option.</p>
+</section>
+<section id="sparse-hessian-of-the-log-likelihood" class="level3">
+<h3 class="anchored" data-anchor-id="sparse-hessian-of-the-log-likelihood">Sparse Hessian of the log likelihood</h3>
+<p>A key step to speed up computation is to take advantage of the sparsity of <span class="math inline">\(H\)</span>, the Hessian of the log likelihood with respect to the latent variables, <span class="math display">\[
+  H = \frac{\partial^2}{\partial \theta^2} \log p(y \mid \theta, \phi).
+\]</span> For example, if the observations <span class="math inline">\((y_1, \cdots, y_N)\)</span> are conditionally independent and each depends on only one component of <span class="math inline">\(\theta\)</span>, the log likelihood decomposes into a sum of per-observation terms, <span class="math display">\[
+  \log p(y \mid \theta, \phi) = \sum_{i = 1}^N \log p(y_i \mid \theta_i, \phi),
+\]</span> and the Hessian is diagonal. This leads to faster calculations of the Hessian and subsequently sparse matrix operations. This case is common in Gaussian process models, and certain hierarchical models.</p>
+<p>Stan’s suite of functions for the embedded Laplace approximation exploits block-diagonal structure in the Hessian, where the user specifies the block size B. The user can specify the size <span class="math inline">\(B\)</span> of these blocks. The user is responsible for working out what <span class="math inline">\(B\)</span> is. If the Hessian is dense, then we simply set <span class="math inline">\(B = N\)</span>. The diagonal case above corresponds to B = 1. Arbitrary sparsity patterns beyond block-diagonal structure are not currently supported.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Cseke:2011" class="csl-entry" role="listitem">
+Cseke, Botond, and Tom Heskes. 2011. <span>“Approximate Marginals in Latent <span>Gaussian</span> Models.”</span> <em>Journal of Machine Learning Research</em> 12.
+</div>
+<div id="ref-Kuss:2005" class="csl-entry" role="listitem">
+Kuss, Malte, and Carl E Rasmussen. 2005. <span>“Assessing Approximate Inference for Binary <span>Gaussian</span> Process Classification.”</span> <em>Journal of Machine Learning Research</em> 6: 1679–1704.
+</div>
+<div id="ref-Margossian:2023" class="csl-entry" role="listitem">
+Margossian, Charles C. 2023. <span>“General Adjoint-Differentiated Laplace Approximation.”</span> <em>arXiv:2306.14976</em>.
+</div>
+<div id="ref-Margossian:2020" class="csl-entry" role="listitem">
+Margossian, Charles C, Aki Vehtari, Daniel Simpson, and Raj Agrawal. 2020. <span>“Hamiltonian Monte Carlo Using an Adjoint-Differentiated Laplace Approximation: Bayesian Inference for Latent Gaussian Models and Beyond.”</span> <em>Advances in Neural Information Processing Systems</em> 34.
+</div>
+<div id="ref-Vanhatalo:2010" class="csl-entry" role="listitem">
+Vanhatalo, Jarno, Ville Pietiläinen, and Aki Vehtari. 2010. <span>“Approximate Inference for Disease Mapping with Sparse <span>Gaussian</span> Processes.”</span> <em>Statistics in Medicine</em> 29 (15): 1580–1607.
+</div>
+<div id="ref-Vehtari:2016" class="csl-entry" role="listitem">
+Vehtari, Aki, Tommi Mononen, Ville Tolvanen, Tuomas Sivula, and Ole Winther. 2016. <span>“Bayesian Leave-One-Out Cross-Validation Approximations for <span>Gaussian</span> Latent Variable Models.”</span> <em>Journal of Machine Learning Research</em> 17 (103): 1–38. <a href="http://jmlr.org/papers/v17/14-540.html">http://jmlr.org/papers/v17/14-540.html</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../reference-manual/laplace.html" class="pagination-link" aria-label="Laplace Approximation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Laplace Approximation</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../reference-manual/diagnostics.html" class="pagination-link" aria-label="Diagnostic Mode">
+        <span class="nav-page-text">Diagnostic Mode</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/laplace_embedded.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/licenses.html b/docs/reference-manual/licenses.html
index e6ff6813a..0e58ea5e0 100644
--- a/docs/reference-manual/licenses.html
+++ b/docs/reference-manual/licenses.html
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -366,6 +366,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1061,8 +1067,35 @@ <h2 class="anchored" data-anchor-id="google-test-license">Google test license</h
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/licenses.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/licenses.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/mcmc.html b/docs/reference-manual/mcmc.html
index 768f46a75..0c9a17252 100644
--- a/docs/reference-manual/mcmc.html
+++ b/docs/reference-manual/mcmc.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -450,6 +450,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -727,8 +733,8 @@ <h3 class="unnumbered anchored" data-anchor-id="automatic-parameter-tuning">Auto
 <section id="discretization-interval-adaptation-parameters" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="discretization-interval-adaptation-parameters">Discretization-interval adaptation parameters</h3>
 <p>Stan’s HMC algorithms utilize dual averaging <span class="citation" data-cites="Nesterov:2009">Nesterov (<a href="#ref-Nesterov:2009" role="doc-biblioref">2009</a>)</span> to optimize the step size.<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a></p>
-<p>This warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>. In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman-Gelman:2014.</p>
-<p>The full set of dual-averaging parameters are</p>
+<p>This warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>. In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in <span class="citation" data-cites="Hoffman-Gelman:2014">Hoffman and Gelman (<a href="#ref-Hoffman-Gelman:2014" role="doc-biblioref">2014</a>)</span>.</p>
+<p>The full set of dual-averaging parameters are:</p>
 <p><strong>Step Size Adaptation Parameters Table</strong> <em>The parameters controlling step size adaptation, with constraints and default values.</em></p>
 <table class="caption-top table">
 <thead>
@@ -1511,8 +1517,35 @@ <h3 class="unnumbered anchored" data-anchor-id="diagnosing-and-eliminating-diver
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/mcmc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/mcmc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/optimization.html b/docs/reference-manual/optimization.html
index d0542e3ed..074423a20 100644
--- a/docs/reference-manual/optimization.html
+++ b/docs/reference-manual/optimization.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -450,6 +450,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1173,8 +1179,35 @@ <h3 class="unnumbered anchored" data-anchor-id="constrained-vs.-unconstrained-pa
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/optimization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/optimization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/pathfinder.html b/docs/reference-manual/pathfinder.html
index 077f145f0..e7ce67d5b 100644
--- a/docs/reference-manual/pathfinder.html
+++ b/docs/reference-manual/pathfinder.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -416,6 +416,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1077,8 +1083,35 @@ <h2 class="anchored" data-anchor-id="using-pathfinder-for-initializing-mcmc">Usi
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/pathfinder.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/pathfinder.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/references.html b/docs/reference-manual/references.html
index 1c0ed4dae..d08beeba7 100644
--- a/docs/reference-manual/references.html
+++ b/docs/reference-manual/references.html
@@ -787,8 +787,35 @@ <h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">Re
 });
 </script>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/removals.html b/docs/reference-manual/removals.html
index 82d0e1fc1..c4c380cba 100644
--- a/docs/reference-manual/removals.html
+++ b/docs/reference-manual/removals.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -401,6 +401,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1163,8 +1169,35 @@ <h2 class="anchored" data-anchor-id="real-values-in-conditionals">Real values in
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/removals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/removals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/reproducibility.html b/docs/reference-manual/reproducibility.html
index 980cc8215..127106597 100644
--- a/docs/reference-manual/reproducibility.html
+++ b/docs/reference-manual/reproducibility.html
@@ -246,7 +246,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -387,6 +387,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1054,8 +1060,35 @@ <h2 class="anchored" data-anchor-id="notable-changes-across-versions">Notable ch
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/reproducibility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/reproducibility.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/statements.html b/docs/reference-manual/statements.html
index b7678fe74..6fd7ff97d 100644
--- a/docs/reference-manual/statements.html
+++ b/docs/reference-manual/statements.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -430,6 +430,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1911,8 +1917,35 @@ <h2 class="anchored" data-anchor-id="fatal-error-statements">Fatal error stateme
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/statements.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/statements.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/syntax.html b/docs/reference-manual/syntax.html
index 45bf930b2..4d89e58a0 100644
--- a/docs/reference-manual/syntax.html
+++ b/docs/reference-manual/syntax.html
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -396,6 +396,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -532,8 +538,8 @@ <h3 class="unnumbered anchored" data-anchor-id="function-declarations-and-defini
                  | &lt;basic_type&gt;
                  | &lt;unsized_tuple_type&gt;
 
-&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA &lt;unsized_type&gt;
-                         (COMMA &lt;unsized_type&gt;)* RPAREN
+&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA [&lt;unsized_type&gt;
+                         (COMMA &lt;unsized_type&gt;)*] RPAREN
 
 &lt;basic_type&gt; ::= INT
                | REAL
@@ -552,7 +558,6 @@ <h3 class="unnumbered anchored" data-anchor-id="function-declarations-and-defini
 <h3 class="unnumbered anchored" data-anchor-id="variable-declarations-and-compound-definitions">Variable declarations and compound definitions</h3>
 <pre><code>&lt;identifier&gt; ::= IDENTIFIER
                | TRUNCATE
-               | JACOBIAN
 
 &lt;decl_identifier&gt; ::= &lt;identifier&gt;
 
@@ -577,8 +582,8 @@ <h3 class="unnumbered anchored" data-anchor-id="variable-declarations-and-compou
                           | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;
 
 &lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA
-                            &lt;higher_type(type_rule)&gt; (COMMA
-                            &lt;higher_type(type_rule)&gt;)* RPAREN
+                            [&lt;higher_type(type_rule)&gt; (COMMA
+                            &lt;higher_type(type_rule)&gt;)*] RPAREN
 
 &lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;
 
@@ -599,8 +604,7 @@ <h3 class="unnumbered anchored" data-anchor-id="variable-declarations-and-compou
                        RBRACK
 
 &lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]
-                 | REAL &lt;type_constraint&gt;                  | TUPLE
-
+                 | REAL &lt;type_constraint&gt;
                  | COMPLEX &lt;type_constraint&gt;
                  | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK
                  | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK
@@ -615,13 +619,13 @@ <h3 class="unnumbered anchored" data-anchor-id="variable-declarations-and-compou
                  | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK
                  | SIMPLEX LBRACK &lt;expression&gt; RBRACK
                  | UNITVECTOR LBRACK &lt;expression&gt; RBRACK
-                 | SUMTOZERO LBRACK &lt;expression&gt; RBRACK
-                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK
+                 | SUMTOZEROVEC LBRACK &lt;expression&gt; RBRACK
                  | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK
                  | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]
                    RBRACK
                  | CORRMATRIX LBRACK &lt;expression&gt; RBRACK
                  | COVMATRIX LBRACK &lt;expression&gt; RBRACK
+                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK
                  | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA
                    &lt;expression&gt; RBRACK
                  | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;
@@ -673,8 +677,8 @@ <h3 class="unnumbered anchored" data-anchor-id="expressions">Expressions</h3>
                       | TARGET LPAREN RPAREN
                       | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;
                         (COMMA &lt;expression&gt;)*] RPAREN
-                      | LPAREN &lt;expression&gt; COMMA &lt;expression&gt; (COMMA
-                        &lt;expression&gt;)* RPAREN
+                      | LPAREN &lt;expression&gt; COMMA [&lt;expression&gt; (COMMA
+                        &lt;expression&gt;)*] RPAREN
                       | &lt;common_expression&gt; DOTNUMERAL
                       | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK
                       | LPAREN &lt;expression&gt; RPAREN
@@ -1442,8 +1446,35 @@ <h3 class="unnumbered anchored" data-anchor-id="indexes">Indexes</h3>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/syntax.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/syntax.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/transforms.html b/docs/reference-manual/transforms.html
index ec54cce49..bdf7d1c0e 100644
--- a/docs/reference-manual/transforms.html
+++ b/docs/reference-manual/transforms.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -450,6 +450,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -2192,8 +2198,35 @@ <h3 class="unnumbered anchored" data-anchor-id="absolute-jacobian-determinant-of
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/transforms.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/transforms.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/types.html b/docs/reference-manual/types.html
index 9a11f4252..b9fc1d933 100644
--- a/docs/reference-manual/types.html
+++ b/docs/reference-manual/types.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -430,6 +430,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1206,21 +1212,23 @@ <h2 class="anchored" data-anchor-id="tuple-data-type">Tuple data type</h2>
 <h3 class="unnumbered anchored" data-anchor-id="declaring-tuple-variables">Declaring tuple variables</h3>
 <p>Tuples are declared with the keyword <code>tuple</code> followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as</p>
 <div class="sourceCode" id="cb75"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb75-1"><a href="#cb75-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">vector</span>[<span class="dv">3</span>], <span class="dt">complex</span>) abc;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
-<p>Tuples must have at least two entries, so the following declarations are illegal.</p>
-<div class="sourceCode" id="cb76"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb76-1"><a href="#cb76-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>() nil;  <span class="co">// ILLEGAL</span></span>
-<span id="cb76-2"><a href="#cb76-2" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>) n;  <span class="co">// ILLEGAL</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Tuples must have at least one entry, so the following declaration is illegal.</p>
+<div class="sourceCode" id="cb76"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb76-1"><a href="#cb76-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>() nil;  <span class="co">// ILLEGAL</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Tuples of length one must use a trailing comma, to align with the expression syntax.</p>
+<div class="sourceCode" id="cb77"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb77-1"><a href="#cb77-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>,) m; <span class="co">// CORRECT</span></span>
+<span id="cb77-2"><a href="#cb77-2" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>) n;  <span class="co">// ILLEGAL</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Tuples can be assigned as a whole if their elements can be assigned individually. For example, <code>a</code> can be assigned to <code>b</code> in the following example because <code>int</code> can be promoted to <code>complex</code>.</p>
-<div class="sourceCode" id="cb77"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb77-1"><a href="#cb77-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">real</span>) a;</span>
-<span id="cb77-2"><a href="#cb77-2" aria-hidden="true" tabindex="-1"></a>...</span>
-<span id="cb77-3"><a href="#cb77-3" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">complex</span>, <span class="dt">real</span>) b = a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb78"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb78-1"><a href="#cb78-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">real</span>) a;</span>
+<span id="cb78-2"><a href="#cb78-2" aria-hidden="true" tabindex="-1"></a>...</span>
+<span id="cb78-3"><a href="#cb78-3" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">complex</span>, <span class="dt">real</span>) b = a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Tuple types may have elements which are declared as tuples, such as the following example.</p>
-<div class="sourceCode" id="cb78"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb78-1"><a href="#cb78-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">tuple</span>(<span class="dt">real</span>, <span class="dt">complex</span>)) x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb79"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb79-1"><a href="#cb79-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">tuple</span>(<span class="dt">real</span>, <span class="dt">complex</span>)) x;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>In this case, it would probably be simpler to use a 3-tuple type, <code>tuple(int, real, complex)</code>.</p>
 <p>Tuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration</p>
-<div class="sourceCode" id="cb79"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb79-1"><a href="#cb79-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb79-2"><a href="#cb79-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb80"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb80-1"><a href="#cb80-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb80-2"><a href="#cb80-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>it is legal to have a tuple with constraints such as</p>
-<div class="sourceCode" id="cb80"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb80-1"><a href="#cb80-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;, <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;) sigma_theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb81"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb81-1"><a href="#cb81-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;, <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt;) sigma_theta;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="accessing-tuple-elements" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="accessing-tuple-elements">Accessing tuple elements</h3>
@@ -1229,24 +1237,24 @@ <h3 class="unnumbered anchored" data-anchor-id="accessing-tuple-elements">Access
 <section id="assigning-tuple-elements" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="assigning-tuple-elements">Assigning tuple elements</h3>
 <p>Tuple elements can be assigned individually, allowing, e.g.,</p>
-<div class="sourceCode" id="cb81"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb81-1"><a href="#cb81-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">real</span>) ab;</span>
-<span id="cb81-2"><a href="#cb81-2" aria-hidden="true" tabindex="-1"></a>ab<span class="fl">.1</span> = <span class="dv">123</span>;</span>
-<span id="cb81-3"><a href="#cb81-3" aria-hidden="true" tabindex="-1"></a>ab<span class="fl">.2</span> = <span class="fl">12.9</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb82"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb82-1"><a href="#cb82-1" aria-hidden="true" tabindex="-1"></a><span class="dt">tuple</span>(<span class="dt">int</span>, <span class="dt">real</span>) ab;</span>
+<span id="cb82-2"><a href="#cb82-2" aria-hidden="true" tabindex="-1"></a>ab<span class="fl">.1</span> = <span class="dv">123</span>;</span>
+<span id="cb82-3"><a href="#cb82-3" aria-hidden="true" tabindex="-1"></a>ab<span class="fl">.2</span> = <span class="fl">12.9</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>As with other assignments, promotions will happen if necessary (of <code>int</code> to <code>real</code> and of <code>real</code> to <code>complex</code>, along with the corresponding container type promotions).</p>
 </section>
 <section id="unpacking-assignment-of-tuples" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="unpacking-assignment-of-tuples">Unpacking assignment of tuples</h3>
 <p>For convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.</p>
 <p>Given a tuple <code>t</code> of type <code>tuple(T1, ..., Tn)</code> and a sequence of <a href="#rhs-expressions">assignable expressions</a> of types <code>v1</code>, …, <code>vn</code>, where each <code>vi</code> has a <a href="../reference-manual/statements.html#promotion">type which is assignable</a> from type <code>Ti</code>, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement</p>
-<div class="sourceCode" id="cb82"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb82-1"><a href="#cb82-1" aria-hidden="true" tabindex="-1"></a>(v1, <span class="co">/*...*/</span>, vn) = t;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb83"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb83-1"><a href="#cb83-1" aria-hidden="true" tabindex="-1"></a>(v1, <span class="co">/*...*/</span>, vn) = t;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Note that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).</p>
 <p>These unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.</p>
 <p>For example, if <code>T</code> is a tuple of type <code>tuple(int, (real, real), complex)</code>, then the program</p>
-<div class="sourceCode" id="cb83"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb83-1"><a href="#cb83-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> i;</span>
-<span id="cb83-2"><a href="#cb83-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x, y;</span>
-<span id="cb83-3"><a href="#cb83-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z;</span>
-<span id="cb83-4"><a href="#cb83-4" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb83-5"><a href="#cb83-5" aria-hidden="true" tabindex="-1"></a>(i, (x, y), z) = T;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb84"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb84-1"><a href="#cb84-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> i;</span>
+<span id="cb84-2"><a href="#cb84-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x, y;</span>
+<span id="cb84-3"><a href="#cb84-3" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z;</span>
+<span id="cb84-4"><a href="#cb84-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb84-5"><a href="#cb84-5" aria-hidden="true" tabindex="-1"></a>(i, (x, y), z) = T;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Assigns the result of <code>T.1</code> to <code>i</code>, the result of <code>T.2.1</code> to <code>x</code>, the result of <code>T.2.2</code> to <code>y</code>, and the result of <code>T.3</code> to <code>z</code>.</p>
 <p>The left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.</p>
 </section>
@@ -1257,7 +1265,7 @@ <h2 class="anchored" data-anchor-id="variable-types-vs.-constraints-and-sizes">V
 <section id="type-information-excludes-sizes" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="type-information-excludes-sizes">Type information excludes sizes</h3>
 <p>The size associated with a given variable is not part of its data type. For example, declaring a variable using</p>
-<div class="sourceCode" id="cb84"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb84-1"><a href="#cb84-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span> a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb85"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb85-1"><a href="#cb85-1" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">real</span> a;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>declares the variable <code>a</code> to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.</p>
 <section id="when-are-sizes-checked" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="when-are-sizes-checked">When are sizes checked?</h4>
@@ -1289,24 +1297,24 @@ <h3 class="unnumbered anchored" data-anchor-id="type-naming-notation">Type namin
 <section id="variable-declaration.section" class="level2">
 <h2 class="anchored" data-anchor-id="variable-declaration.section">Variable declaration</h2>
 <p>Variables in Stan are declared by giving a type and a name. For example</p>
-<div class="sourceCode" id="cb85"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb85-1"><a href="#cb85-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N;</span>
-<span id="cb85-2"><a href="#cb85-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] y;</span>
-<span id="cb85-3"><a href="#cb85-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] A;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb86"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb86-1"><a href="#cb86-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N;</span>
+<span id="cb86-2"><a href="#cb86-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>[N] y;</span>
+<span id="cb86-3"><a href="#cb86-3" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">5</span>] <span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">4</span>] A;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>declares a variable <code>N</code> that is an integer, a variable <code>y</code> that is a vector of length <code>N</code> (the previously declared variable), and a variable <code>A</code>, which is a length-5 array where each element is a 3 by 4 matrix.</p>
 <p>The size of top-level variables in the <code>parameters</code>, <code>transformed parameters</code>, and <code>generated quantities</code> must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.</p>
-<div class="sourceCode" id="cb86"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb86-1"><a href="#cb86-1" aria-hidden="true" tabindex="-1"></a><span class="co">// illegal and will be flagged by the compiler:</span></span>
-<span id="cb86-2"><a href="#cb86-2" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb86-3"><a href="#cb86-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N = <span class="dv">10</span>;</span>
-<span id="cb86-4"><a href="#cb86-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> foo;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb87"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb87-1"><a href="#cb87-1" aria-hidden="true" tabindex="-1"></a><span class="co">// illegal and will be flagged by the compiler:</span></span>
+<span id="cb87-2"><a href="#cb87-2" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb87-3"><a href="#cb87-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> N = <span class="dv">10</span>;</span>
+<span id="cb87-4"><a href="#cb87-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> foo;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Depending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.</p>
-<div class="sourceCode" id="cb87"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb87-1"><a href="#cb87-1" aria-hidden="true" tabindex="-1"></a><span class="co">// valid block variables, but not locals or function parameters</span></span>
-<span id="cb87-2"><a href="#cb87-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N] u;</span>
-<span id="cb87-3"><a href="#cb87-3" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb87-4"><a href="#cb87-4" aria-hidden="true" tabindex="-1"></a><span class="co">// valid as a block or local variable, but not a function parameter</span></span>
-<span id="cb87-5"><a href="#cb87-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> is;</span>
-<span id="cb87-6"><a href="#cb87-6" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb87-7"><a href="#cb87-7" aria-hidden="true" tabindex="-1"></a><span class="co">// function parameters exclude sizes and cannot be constrained</span></span>
-<span id="cb87-8"><a href="#cb87-8" aria-hidden="true" tabindex="-1"></a><span class="dt">void</span> pretty_print_tri_lower(<span class="dt">matrix</span> x) { ... }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb88"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb88-1"><a href="#cb88-1" aria-hidden="true" tabindex="-1"></a><span class="co">// valid block variables, but not locals or function parameters</span></span>
+<span id="cb88-2"><a href="#cb88-2" aria-hidden="true" tabindex="-1"></a><span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N] u;</span>
+<span id="cb88-3"><a href="#cb88-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb88-4"><a href="#cb88-4" aria-hidden="true" tabindex="-1"></a><span class="co">// valid as a block or local variable, but not a function parameter</span></span>
+<span id="cb88-5"><a href="#cb88-5" aria-hidden="true" tabindex="-1"></a><span class="dt">array</span>[<span class="dv">3</span>] <span class="dt">int</span> is;</span>
+<span id="cb88-6"><a href="#cb88-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb88-7"><a href="#cb88-7" aria-hidden="true" tabindex="-1"></a><span class="co">// function parameters exclude sizes and cannot be constrained</span></span>
+<span id="cb88-8"><a href="#cb88-8" aria-hidden="true" tabindex="-1"></a><span class="dt">void</span> pretty_print_tri_lower(<span class="dt">matrix</span> x) { ... }</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Top-level variables can have <a href="#constrained-data-types">constraints</a> and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as <a href="../reference-manual/user-functions.html#argument-types-and-qualifiers">function parameters</a> are not constrained types and <em>exclude</em> sizes.</p>
 <p>In the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables <code>M</code> and <code>N</code> indicate number of columns and rows, respectively. The variable <code>K</code> is used for square matrices, i.e., <code>K</code> denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types <code>cov_matrix[K]</code>, <code>corr_matrix[K]</code>, <code>cholesky_factor_cov[K]</code>, and <code>cholesky_factor_corr[K]</code> are only assignable to matrices of dimensions <code>matrix[K, K]</code> types.</p>
 <table class="caption-top table">
@@ -1654,24 +1662,24 @@ <h2 class="anchored" data-anchor-id="compound-variable-declaration-and-definitio
 <li>variables declared in the transformed data, transformed parameters, or generated quantities blocks.</li>
 </ul>
 <p>For example, the statement</p>
-<div class="sourceCode" id="cb90"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb90-1"><a href="#cb90-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N = <span class="dv">5</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb91"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb91-1"><a href="#cb91-1" aria-hidden="true" tabindex="-1"></a><span class="dt">int</span> N = <span class="dv">5</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>declares the variable <code>N</code> to be an integer scalar type and at the same time defines it to be the value of the expression <code>5</code>.</p>
 <section id="assignment-typing" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="assignment-typing">Assignment typing</h3>
 <p>The type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have</p>
-<div class="sourceCode" id="cb91"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb91-1"><a href="#cb91-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum = <span class="dv">0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb92"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb92-1"><a href="#cb92-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> sum = <span class="dv">0</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>even though <code>0</code> is of type <code>int</code> and <code>sum</code> is of type <code>real</code>, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.</p>
 <p>Variables of any type may have values assigned to them. For example,</p>
-<div class="sourceCode" id="cb92"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb92-1"><a href="#cb92-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] a = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb93"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb93-1"><a href="#cb93-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] a = b;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>declares a <span class="math inline">\(3 \times 2\)</span> matrix variable <code>a</code> and assigns a copy of the value of <code>b</code> to the variable <code>a</code>. The variable <code>b</code> must be of type <code>matrix</code> for the statement to be well formed. For the code to execute successfully, <code>b</code> must be the same shape as <code>a</code>, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to <code>a</code> do not affect <code>b</code> and subsequent changes to <code>b</code> do not affect <code>a</code>.</p>
 </section>
 <section id="rhs-expressions" class="level3">
 <h3 class="anchored" data-anchor-id="rhs-expressions">Right-hand side expressions</h3>
 <p>The right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,</p>
-<div class="sourceCode" id="cb93"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb93-1"><a href="#cb93-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] a = <span class="fl">0.5</span> * (b + c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb94"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb94-1"><a href="#cb94-1" aria-hidden="true" tabindex="-1"></a><span class="dt">matrix</span>[<span class="dv">3</span>, <span class="dv">2</span>] a = <span class="fl">0.5</span> * (b + c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>assigns the matrix variable <code>a</code> to half of the sum of <code>b</code> and <code>c</code>. The only requirement on <code>b</code> and <code>c</code> is that the expression <code>b + c</code> be of type <code>matrix</code>. For example, <code>b</code> could be of type <code>matrix</code> and <code>c</code> of type <code>real</code>, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.</p>
 <p>Similarly,</p>
-<div class="sourceCode" id="cb94"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb94-1"><a href="#cb94-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = <span class="dv">2</span> + <span class="fl">3i</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb95"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb95-1"><a href="#cb95-1" aria-hidden="true" tabindex="-1"></a><span class="dt">complex</span> z = <span class="dv">2</span> + <span class="fl">3i</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>assigns the the complex number <span class="math inline">\(2 + 3i\)</span> to the complex scalar <code>z</code>. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).</p>
 </section>
 <section id="scope-within-expressions" class="level3 unnumbered">
@@ -1685,18 +1693,18 @@ <h2 class="anchored" data-anchor-id="declaring-multiple-variables-at-once">Decla
 <section id="types-for-multiple-declarations" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="types-for-multiple-declarations">Types for multiple declarations</h3>
 <p>The code:</p>
-<div class="sourceCode" id="cb95"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb95-1"><a href="#cb95-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x, y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb96"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb96-1"><a href="#cb96-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x, y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>is equivalent to</p>
-<div class="sourceCode" id="cb96"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb96-1"><a href="#cb96-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x;</span>
-<span id="cb96-2"><a href="#cb96-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb97"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb97-1"><a href="#cb97-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x;</span>
+<span id="cb97-2"><a href="#cb97-2" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>As a result, all declarations on the same line must be of the same type.</p>
 </section>
 <section id="combining-with-other-features" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="combining-with-other-features">Combining with other features</h3>
 <p>The ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in <a href="#compound-variable-declaration-and-definition">the section introducing compound declarations and definitions</a> :</p>
-<div class="sourceCode" id="cb97"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb97-1"><a href="#cb97-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="dv">3</span>, y = <span class="fl">5.6</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb98"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb98-1"><a href="#cb98-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span> x = <span class="dv">3</span>, y = <span class="fl">5.6</span>;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p><a href="#constrained-data-types">Constrained data types</a> can also be declared together, so long as the constraint for each variable is the same:</p>
-<div class="sourceCode" id="cb98"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb98-1"><a href="#cb98-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; x, y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb99"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb99-1"><a href="#cb99-1" aria-hidden="true" tabindex="-1"></a><span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; x, y;</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 
 
 </section>
@@ -2284,8 +2292,35 @@ <h3 class="unnumbered anchored" data-anchor-id="combining-with-other-features">C
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/types.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/types.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/user-functions.html b/docs/reference-manual/user-functions.html
index a9b5cb0f0..2c1e2cbc5 100644
--- a/docs/reference-manual/user-functions.html
+++ b/docs/reference-manual/user-functions.html
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -401,6 +401,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1311,8 +1317,35 @@ <h2 class="anchored" data-anchor-id="forward-declarations.section">Declarations<
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/variational.html b/docs/reference-manual/variational.html
index e979ed016..98e2fbaf2 100644
--- a/docs/reference-manual/variational.html
+++ b/docs/reference-manual/variational.html
@@ -275,7 +275,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -416,6 +416,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1087,8 +1093,35 @@ <h3 class="unnumbered anchored" data-anchor-id="assessing-convergence">Assessing
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/variational.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/variational.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/reference-manual/whitespace.html b/docs/reference-manual/whitespace.html
index f7df00591..4ba49ebf7 100644
--- a/docs/reference-manual/whitespace.html
+++ b/docs/reference-manual/whitespace.html
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -366,6 +366,12 @@
   <a href="../reference-manual/laplace.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Laplace Approximation</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../reference-manual/laplace_embedded.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Embedded Laplace Approximation</span></a>
+  </div>
 </li>
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
@@ -1018,8 +1024,35 @@ <h2 class="anchored" data-anchor-id="whitespace-location">Whitespace location</h
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/whitespace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/reference-manual/whitespace.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/search.json b/docs/search.json
index c6f90957a..4a90c7f3e 100644
--- a/docs/search.json
+++ b/docs/search.json
@@ -1,9195 +1,9494 @@
 [
   {
-    "objectID": "stan-users-guide/using-stanc.html",
-    "href": "stan-users-guide/using-stanc.html",
-    "title": "Using the Stan Compiler",
+    "objectID": "stan-users-guide/wiener_diffusion_model.html",
+    "href": "stan-users-guide/wiener_diffusion_model.html",
+    "title": "Wiener diffusion model",
     "section": "",
-    "text": "Stan is used in most of our interfaces through the Stan compiler stanc. Since version 2.22, the Stan compiler has been implemented in OCaml and is referred to as stanc3. The binary name is still simply stanc, so this document uses both stanc and stanc3 interchangeably.\n\n\nThe stanc3 compiler has the following command-line syntax:\n&gt; stanc (options) &lt;model_file&gt;\nwhere &lt;model_file&gt; is either a path to a file ending in .stan or .stanfunctions (which automatically sets --standalone-functions), or '-' to read from standard input.\nThe stanc3 options are:\n\n--help / -? - Displays the complete list of stanc3 options, then exits.\n--version - Display stanc version number\n--info - Print information about the model, such as the type information for variables and the list of used distributions.\n--name=&lt;model_name&gt; - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.\n--o=&lt;file_name&gt; / -o=&lt;filename&gt; / --output=&lt;filename&gt; - Specify a path to an output file for generated C++ code (default = .hpp) or auto-formatting output (default: no file/print to stdout)\n--auto-format - Pretty prints the program to the console. See more on auto formatting.\n--allow-undefined - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.\n--canonicalize - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.\n--include_paths=&lt;dir1,...dirN&gt; - Takes a comma-separated list of directories that may contain a file in an #include directive.\n--max-line-length=&lt;number&gt; - Set the column number at which formatting with --auto-format attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.\n--print-canonical - Synonymous with --auto-format --canonicalize=deprecations,includes,parentheses,braces.\n--print-cpp - If set, output the generated C++ Stan model class to stdout.\n--filename-in-msg=&lt;name&gt; - Sets the filename used in compiler and runtime errors. If absent, the &lt;model_file&gt; argument is used.\n--standalone-functions - If set, only generate the code for the functions defined in the file. This is the default behavior for .stanfunctions files.\n--O0 (Default) Do not apply optimizations to the Stan code.\n--O1 Apply level 1 compiler optimizations (only basic optimizations).\n--Oexperimental WARNING: This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance! Allow the compiler to apply all optimizations to the Stan code.\n--O Synonym for --O1 as of Stan 2.37. In earlier versions this was a synonym for --Oexperimental.\n--use-opencl - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.\n--warn-pedantic - Emit warnings in Pedantic mode which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.\n--warn-uninitialized - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.\n--color - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are auto (the default), always, never. Can also be controlled by the STANC_COLOR environment variable.\n\nThe compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the --help option to see the full set.\n\n\n\nDuring model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).\n\n\nEven without the optional --warn-pedantic and --warn-uninitialized command line flags, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations\n\nA completely blank Stan program will produce the following warning message\nWarning in 'empty.stan', line 1, column 0 to line 2, column 0:\n    Empty model detected; this is a valid Stan model but likely unintended!\nThe use of any deprecated features will lead to warnings which will look as follows\n Warning in 'deprecated.stan', line 2, column 10 to column 17:\n     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with\n     an independent lognormal distribution on the scales, see:\n     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution\nA single Stan program can produce many warnings during compilation.\n\n\n\n\nErrors differ from warnings in their severity and format. In particular, errors are fatal and stop compilation, so at most one error is displayed per run of stanc.\nThere are five kinds of errors emitted by stanc3\n\nFile errors occur when the file passed to stanc is either missing or cannot be opened (i.e. has permissions issues). They look like\nError: file 'notfound.stan' not found or cannot be opened\nSyntactic errors occur whenever a program violates the Stan language’s syntax requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the #include directive fails, and “parsing” errors which result when the structure of the program is incorrect.\n\nThe lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of $ in a variable name will look like the following.\nSyntax error in 'char.stan', line 2, column 7 to column 8, lexing error:\n-------------------------------------------------\n  1:  data {\n  2:     int $ome_variable;\n             ^\n  3:  }\n-------------------------------------------------\nInvalid character found.\nWhen an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).\nSyntax error in './incl.stan', line 1, column 0, included from\n'./incl.stan', line 1, column 0, included from\n'incl.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include &lt;incl.stan&gt;\n         ^\n   -------------------------------------------------\nFile incl.stan recursively included itself.\nIt is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.\nSyntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:\n   -------------------------------------------------\n     1:  data {\n     2:     int&lt;lower=0&gt; N;\n     3:     vector x;\n                   ^\n     4:  }\n   -------------------------------------------------\nIll-formed type. Expected \"[\" expression \"]\" for vector size.\n\nSemantic errors (also known as type errors) occur when a program is structured correctly but features an error in the type rules imposed by the language. An example of this is assigning a real value to a variable defined as an integer.\nSemantic error in 'type.stan', line 2, column 3 to column 15:\n   -------------------------------------------------\n     1:  transformed data {\n     2:     int x = 1.5;\n            ^\n     3:  }\n   -------------------------------------------------\nIll-typed arguments supplied to assignment operator =:\nThe left hand side has type\n  int\nand the right hand side has type\n  real\nThe compiler will raise an error for use of any removed features for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.\nFinally, the compiler can raise an internal error. These are caused by bugs in the compiler, not your model, and we would appreciate it if you report them on the stanc3 repo with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.\n\n\n\n\n\nPedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.\nFor example, consider the following program.\ndata {\n  int N;\n  array[N] real x;\n}\nparameters {\n  real sigma;\n}\nmodel {\n  real mu;\n  x ~ normal(mu, sigma);\n}\nWhen pedantic mode is turned on, the compiler will produce the following warnings.\nWarning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:\n    The parameter sigma has no priors. This means either no prior is\n    provided, or the prior(s) depend on data variables. In the later case,\n    this may be a false positive.\nWarning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:\n    The variable mu may not have been assigned a value before its use.\nWarning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:\n    A normal distribution is given parameter sigma as a scale parameter\n    (argument 2), but sigma was not constrained to be strictly positive.\nHere are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):\n\nDistribution usages issues. Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.\nUnused parameter. A parameter is defined but doesn’t contribute to target.\nLarge or small constant in a distribution. Very large or very small constants are used as distribution arguments.\nControl flow depends on a parameter. Branching control flow (like if/else) depends on a parameter value .\nParameter has multiple tildes. A parameter is on the left-hand side of multiple tildes.\nParameter has zero or multiple priors. A parameter has zero or more than one prior distribution.\nVariable is used before assignment. A variable is used before being assigned a value.\nStrict or nonsensical parameter bounds. A parameter is given questionable bounds.\nNonlinear transformations. When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.\n\nSome important limitations of pedantic mode are listed at the end of this chapter.\n\n\nWhen an argument to a built-in distribution certainly does not match that distribution’s specification in the Stan Functions Reference, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g. simplex) is incompatible with the domain of the distribution. x\nFor example, consider the following program.\nparameters {\n  real unb_p;\n  real&lt;lower=0&gt; pos_p;\n}\nmodel {\n  1 ~ poisson(unb_p);\n  1 ~ poisson(pos_p);\n}\nThe parameter of poisson should be strictly positive, but unb_p is not constrained to be positive.\nPedantic mode produces the following warning.\nWarning in 'ex-dist-args.stan', line 6, column 14 to column 19:\n  A poisson distribution is given parameter unb_p as a rate parameter\n  (argument 1), but unb_p was not constrained to be strictly positive.\n\n\n\nPedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:\n\n\nAny use of uniform distribution generates a warning, except when the variate parameter’s declared upper and lower bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.\nFor example, consider the following program.\nparameters {\n  real a;\n  real&lt;lower=0, upper=1&gt; b;\n}\nmodel {\n  a ~ uniform(0, 1);\n  b ~ uniform(0, 1);\n}\na is assigned a uniform distribution that doesn’t match its constraints.\nPedantic mode produces the following warning.\nWarning in 'uniform-warn.stan', line 6, column 2 to column 20:\n  Parameter a is given a uniform distribution. The uniform distribution is\n  not recommended, for two reasons: (a) Except when there are logical or\n  physical constraints, it is very unusual for you to be sure that a\n  parameter will fall inside a specified range, and (b) The infinite gradient\n  induced by a uniform density can cause difficulties for Stan's sampling\n  algorithm. As a consequence, we recommend soft constraints rather than hard\n  constraints; for example, instead of giving an elasticity parameter a\n  uniform(0, 1) distribution, try normal(0.5, 0.5).\n\n\n\nGamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.\n\n\n\nAny use of the lkj_corr distribution generates a warning that suggests using the Cholesky variant instead. See the LKJ correlation distribution section of the Stan Functions Reference for details.\n\n\n\n\nA warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the target variable depends in any way on each of the parameters.\nFor example, consider the following program.\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(1, 1);\n}\na participates in the density function but b does not.\nPedantic mode produces the following warning.\nWarning in 'unused.stan', line 3, column 2 to column 9:\n    The parameter b was declared but was not used in the density calculation.\n\n\n\nWhen numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See the efficiency tuning section of the Stan User’s guide for a discussion of scaling parameters.\nFor example, consider the following program.\nparameters {\n  real x;\n  real y;\n}\nmodel {\n  x ~ normal(-100, 100);\n  y ~ normal(0, 1);\n}\nThe constants -100 and 100 suggest that x is not unit scaled.\nPedantic mode produces the following warning.\nWarning in 'constants-warn.stan', line 6, column 13 to column 17:\n    Argument -100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\nWarning in 'constants-warn.stan', line 6, column 19 to column 22:\n    Argument 100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\n\n\n\nControl flow statements, such as if, for and while should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.\nFor example, consider the following program.\nparameters {\n  real a;\n}\nmodel {\n  // x depends on parameter a\n  real x = a * a;\n\n  int m;\n\n  // the if-then-else depends on x which depends on a\n  if(x &gt; 0) {\n    //now m depends on x which depends on a\n    m = 1;\n  } else {\n    m = 2;\n  }\n\n  // for loop depends on m -&gt; x -&gt; a\n  for (i in 0:m) {\n    a ~ normal(i, 1);\n  }\n}\nThe if and for statements are control flow that depend (indirectly) on the value of the parameter m.\nPedantic mode produces the following warning.\nWarning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:\n  A control flow statement depends on parameter(s): a.\nWarning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:\n  A control flow statement depends on parameter(s): a.\n\n\n\nA warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent target += conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.\nPedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(0, 1);\n  a ~ normal(x, 1);\n\n  b ~ normal(1, 1);\n}\nPedantic mode produces the following warning.\nWarning in 'multi-tildes.stan', line 9, column 2 to column 19:\n  The parameter a is on the left-hand side of more than one tildes\n  statement.\n\n\n\nA warning is generated when a parameter appears to have greater than or less than one prior distribution factor.\nThis analysis depends on a factor graph representation of a Stan program. A factor F that depends on a parameter P is called a prior factor for P if there is no path in the factor graph from F to any data variable except through P.\nOne limitation of this approach is that the compiler cannot distinguish between modeled data variables and other convenient uses of data variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for int variables) are modeled data, which may cause extra warnings.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n  real c;\n  real d;\n}\nmodel\n{\n  a ~ normal(0, 1); // this is a prior\n  x ~ normal(a, 1); // this is not a prior, since data is involved\n\n  b ~ normal(x, 1); // this is also not a prior, since data is involved\n\n  // this is not a prior for c, since data is involved through b\n  // but it is a prior for b, since the data is only involved through b\n  c ~ normal(b, 1);\n\n  //these are multiple priors:\n  d ~ normal(0, 1);\n  1 ~ normal(d, 1);\n}\nOne prior is found for a and for b, while c only has a factor that touches a data variable and d has multiple priors.\nPedantic mode produces the following warning.\nWarning in 'priors.stan', line 7, column 2 to column 9:\n    The parameter c has no priors. This means either no prior is provided, or\n    the prior(s) depend on data variables. In the later case, this may be a\n    false positive.\nWarning in 'priors.stan', line 8, column 2 to column 9:\n    The parameter d has 2 priors.\n\n\n\n\nA warning is generated when any variable is used before it has been assigned a value.\nFor example, consider the following program.\ntransformed data {\n  real x;\n  if (1 &gt; 2) {\n    x = 1;\n  } else {\n    print(\"oops\");\n  }\n  print(x);\n}\nSince x is only assigned in one of the branches of the if statement, it might get to print(x) without having been assigned to.\nPedantic mode produces the following warning.\nWarning in 'uninit-warn.stan', line 7, column 8 to column 9:\n  The variable x may not have been assigned a value before its use.\n\n\n\nExcept when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds &lt;lower=.., upper=..&gt; except for &lt;lower=0, upper=1&gt; or &lt;lower=-1, upper=1&gt;.\nIn addition, a warning is generated when a parameter bound is found to have lower &gt;= upper.\nFor example, consider the following program.\nparameters {\n  real&lt;lower=0, upper=1&gt; a;\n  real&lt;lower=-1, upper=1&gt; b;\n  real&lt;lower=-2, upper=1012&gt; c;\n}\nmodel {\n  c ~ normal(b, a);\n}\nPedantic mode produces the following warning.\nWarning in 'hard-constraint.stan', line 4, column 2 to column 31:\n    Your Stan program has a parameter c with a lower and upper bound in its\n    declaration. These hard constraints are not recommended, for two reasons:\n    (a) Except when there are logical or physical constraints, it is very\n    unusual for you to be sure that a parameter will fall inside a specified\n    range, and (b) The infinite gradient induced by a hard constraint can\n    cause difficulties for Stan's sampling algorithm. As a consequence, we\n    recommend soft constraints rather than hard constraints; for example,\n    instead of constraining an elasticity parameter to fall between 0, and 1,\n    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.\n\n\n\nWhen a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the Changes of variables section.\nThis portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.\nFor example, consider the following program.\nparameters {\n  real y;\n}\nmodel {\n  log(y) ~ normal(0,1);\n}\nPedantic mode produces the following warning.\nWarning in 'jacobian.stan', line 5, column 2 to column 23:\n    Left-hand side of distribution statement (~) may contain a non-linear\n    transform of a parameter or local variable. If it does, you need to\n    include a target += statement with the log absolute determinant of the\n    Jacobian of the transform. You could also consider defining a transformed\n    parameter and using jacobian += in the transformed parameters block.\n\n\n\n\nConstant values are sometimes uncomputable\nPedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code normal(x, 1 - 2), the expression 1 - 2 will be evaluated to -1, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.\nContainer types\nCurrently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).\nData variables\nThe declaration information for data variables is currently not considered, so using data as incompatible arguments to distributions may not generate the appropriate warnings.\nControl flow dependent on parameters in nested functions\nIf a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.\n\n\n\n\n\nIn addition to compiling Stan programs, stanc3 features several flags which can be used to format Stan programs and update them to the most recent Stan syntax by removing any deprecation features which can be automatically replaced.\nThese flags work for both .stan model files and .stanfunctions function files. They can be combined with --o to redirect the formatted output to a new file.\n\n\nInvoking stanc --auto-format &lt;model_file&gt; will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the Stan Program Style Guide. This means spacing, indentation, and line length are all regularized. Some deprecated features, like the use of # for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.\nBy default, this will try to split lines at or before column 78. This number can be changed using --max-line-length.\n\n\n\nIn addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of if statements and for and while loops.\nThis can be done by using stanc --auto-format --canonicalize=... where ... is a comma-separated list of options. Currently these options are:\n\ndeprecations\nRemoves deprecated syntax such as replacing deprecated functions with their drop-in replacements.\nparentheses\nRemoves unnecessary extra parentheses, such as converting y = ((x-1)) to y = x - 1\nbraces\nPlaces braces around all blocks. For example, the following statement\nif (cond)\n  //result\nwill be formatted as\nif (cond) {\n  //result\n}\nand similarly for both kinds of loops containing a single statement.\nincludes\nThis will pretty-print code from other files included with #include as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same #include directives as the input program.\n\nInvoking stanc --print-canonical &lt;model_file&gt; is synonymous with running stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes\n\n\n\nThe formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:\n\nOddly placed comments\nIf your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string ^^^: to indicate they originally appeared higher in the program.\nWe hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.\nFailure to recreate strange #include structure\nPrinting without include inlining (--canonicalize=includes) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.\n\n\n\n\n\nThe stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.\nThis section introduces the available optimization options and describes their effect.\nTo print out a representation of the optimized Stan program, use the stanc3 command-line flag --debug-optimized-mir-pretty. To print an analogous representation of the Stan program prior to optimization, use the flag --debug-transformed-mir-pretty.\n\n\nTo turn optimizations on, the user specifies the desired optimization level. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.\nOptimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:\n\nO0 No optimizations are applied.\nO1 Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.\nOexperimental All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.\n\nO0 is the default setting.\nThe levels include these optimizations:\n\nO0 includes no optimizations.\nO1 includes:\n\nDead code elimination\nCopy propagation\nConstant propagation\nPartial evaluation\nFunction inlining\nMatrix memory layout optimization\n\nOexperimental includes optimizations specified by O1 and also:\n\nAutomatic-differentiation level optimization\nOne step loop unrolling\nExpression propagation\nLazy code motion\nStatic loop unrolling\n\n\nIn addition, Oexperimental will apply more repetitions of the optimizations, which may increase compile times.\n\n\n\n\n\nDead code is code that does not affect the behavior of the program. Code is not dead if it affects target, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.\nExample Stan program:\nmodel {\n  int i;\n  i = 5;\n  for (j in 1:10);\n  if (0) {\n    print(\"Dead code\");\n  } else {\n    print(\"Hi!\");\n  }\n}\nCompiler representation of program before dead code elimination (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  int i = 5;\n  for(j in 1:10) {\n    ;\n  }\n  if(0) {\n    FnPrint__(\"Dead code\");\n  } else {\n    FnPrint__(\"Hi!\");\n  }\n}\nCompiler representation of program after dead code elimination (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i;\n  FnPrint__(\"Hi!\");\n}\n\n\n\nConstant propagation replaces uses of a variable which is known to have a constant value C with that constant C. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).\nExample Stan program:\ntransformed data {\n  int n = 100;\n  int a[n];\n  for (i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before constant propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, n] a;\n  for(i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after constant propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, 100] a;\n  for(i in 1:100) {\n    a[i] = i;\n  }\n}\n\n\n\nCopy propagation is similar to expression propagation, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.\nExample Stan program:\nmodel {\n  int i = 1;\n  int j = i;\n  int k = i + j;\n}\nCompiler representation of program before copy propagation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n    int i = 1;\n    int j = i;\n    int k = (i + j);\n}\nCompiler representation of program after copy propagation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i = 1;\n  int j = i;\n  int k = (i + i);\n}\n\n\n\nPartial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.\nExample Stan program:\nmodel {\n  real a = 1 + 1;\n  real b = log(1 - a);\n  real c = a + b * 5;\n}\nCompiler representation of program before partial evaluation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real a = (1 + 1);\n  real b = log((1 - a));\n  real c = (a + (b * 5));\n}\nCompiler representation of program after partial evaluation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  real a = 2;\n  real b = log1m(a);\n  real c = fma(b, 5, a);\n}\n\n\n\nFunction inlining replaces each function call to each user-defined function f with the body of f. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return x + y;\n  }\n}\ntransformed data {\n  int a = 2;\n  int b = incr(a);\n}\nCompiler representation of program before function inlining (simplified from the output of --debug-transformed-mir-pretty):\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return (x + y);\n  }\n}\n\nprepare_data {\n  data int a = 2;\n  data int b = incr(a);\n}\nCompiler representation of program after function inlining (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int a;\n  a = 2;\n  data int b;\n  data int inline_sym1__;\n  data int inline_sym3__;\n  inline_sym3__ = 0;\n  for(inline_sym4__ in 1:1) {\n    int inline_sym2__;\n    inline_sym2__ = 1;\n    inline_sym3__ = 1;\n    inline_sym1__ = (a + inline_sym2__);\n    break;\n  }\n  b = inline_sym1__;\n}\nIn this code, the for loop and break is used to simulate the behavior of a return statement. The value to be returned is held in inline_sym1__. The flag variable inline_sym3__ indicates whether a return has occurred and is necessary to handle return statements nested inside loops within the function body.\n\n\n\nMatrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.\nThe first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.\nThe second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.\nMore general reading on AoS vs SoA can be found on Wikipedia\nThis optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:\n\nAny Stan Math Library functions the matrix is passed to must be able to support it.\nThe matrix should not be accessed/assigned elementwise in a loop.\n\nThe debug flag --debug-mem-patterns will list each variable and whether it is using the AoS representation or the SoA representation.\n\n\n\n\n\n\nStan variables can have two auto-differentiation (AD) levels: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.\nExample Stan program:\ndata {\n  real y;\n}\nmodel {\n  real x = y + 1;\n}\nCompiler representation of program before AD-level optimization (simplified from the output of --debug-transformed-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  real x = (y + 1);\n}\nCompiler representation of program after AD-level optimization (simplified from the output of --debug-optimized-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  data real x = (y + 1);\n}\n\n\n\nOne step loop unrolling is similar to static loop unrolling. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.\nExample Stan program:\ndata {\n  int n;\n}\ntransformed data {\n  int x = 0;\n  for (i in 1:n) {\n    x += i;\n  }\n}\nCompiler representation of program before one step static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  data int x = 0;\n  for(i in 1:n) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after one step static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  int x = 0;\n  if((n &gt;= 1)) {\n    x = (x + 1);\n    for(i in (1 + 1):n) {\n      x = (x + i);\n    }\n  }\n}\n\n\n\nConstant propagation replaces the uses of a variable which is known to have a constant value E with that constant E. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by lazy code motion to avoid unnecessarily recomputing expressions.\nExample Stan program:\ndata {\n  int m;\n}\ntransformed data {\n  int n = m+1;\n  int a[n];\n  for (i in 1:n-1) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before expression propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, n] a;\n  for(i in 1:(n - 1)) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after expression propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, (m + 1)] a;\n  for(i in 1:((m + 1) - 1)) {\n    a[i] = i;\n  }\n}\n\n\n\nLazy code motion rearranges the statements and expressions in a program with the goals of:\n\nAvoiding computing expressions more than once, and\nComputing expressions as late as possible (to minimize the strain on the working memory set).\n\nTo accomplish these goals, lazy code motion will perform optimizations such as:\n\nMoving a repeatedly calculated expression to its own variable (also referred to as common-subexpression elimination)\nMoving an expression outside of a loop if it does not need to be in the loop (also referred to as loop-invariant code motion)\n\nLazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.\nAs currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.\nExample Stan program:\nmodel {\n  real x;\n  real y;\n  real z;\n\n  for (i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program before lazy code motion (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real x;\n  real y;\n  real z;\n  for(i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program after lazy code motion (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  data real lcm_sym4__;\n  data real lcm_sym3__;\n  real x;\n  real y;\n  lcm_sym4__ = sqrt(10);\n  real z;\n  for(i in 1:10) {\n    x = lcm_sym4__;\n    y = sqrt(i);\n  }\n  z = lcm_sym4__;\n}\n\n\n\nStatic loop unrolling takes a loop with a predictable number of iterations X and replaces it by writing out the loop body X times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\ntransformed data {\n  int x = 0;\n  for (i in 1:4) {\n    x += i;\n  }\n}\nCompiler representation of program before static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int x = 0;\n  for(i in 1:4) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int x;\n  x = 0;\n  x = (x + 1);\n  x = (x + 2);\n  x = (x + 3);\n  x = (x + 4);\n}",
+    "text": "Diffusion models, sometimes also called Wiener diffusion models, are among the most frequently used model families in modeling two-alternative forced-choice tasks (see Wagenmakers (2009), for a review). Diffusion models allow to model response times and responses jointly. The basic version of a diffusion model comprises four parameters: the boundary separation, \\(a\\), the relative starting point, \\(w\\), the drift rate, \\(v\\), and the non-decision time, \\(t0\\) (Ratcliff 1978). In the basic model, it is assumed that the four basic parameters are the same for the whole experiment. As this assumption is very strict and there are examples that suggest that the basic parameters can be different from trial to trial, so called inter-trial variabilities were introduced and the basic four- parameter model was extended to a seven-parameter model. In the seven-parameter extension of the diffusion model there are the following three parameters added: the inter-trial variability in relative starting point, \\(s_w\\), the inter-trial variability in drift rate, \\(s_v\\), and the inter-trial variability in non-decision time, \\(s_{t0}\\) Nicenboim, Schad, and Vasishth (2025).\nData for the diffusion model is two-dimensional: There is one vector for the reaction times, \\(y\\), and one vector for the given responses, \\(\\text{resp}\\). The reaction times shall be positive, continuous and in seconds, the responses shall be binary.\nAs a diffusion model describes the decision process for a decision with exactly two choices, there exist reaction time distributions for each response alternative. This means that the probability density function (\\(p\\)) splits into one part for one response alternative and one part for the other response alternative. In the following, we will refer to one alternative as the upper response boundary and to the other alternative as the lower response boundary. \\(p\\) of the lower response boundary can be obtained when inserting \\(-v\\) and \\(1-w\\) to \\(p\\) of the upper response boundary. Let’s call \\(p\\) for the lower response boundary \\(p_0\\) and \\(p\\) for the upper response boundary \\(p_1\\). Then:\n\\[\np_0(a,t0,v,w,sv,sw,st0) = p_1(a,t0,-v,1-w,sv,sw,st0)\n\\]\nUsually, a \\(PDF\\) integrates to 1. In the case of the diffusion model, only the sum of both parts, \\(p_0\\) and \\(p_1\\), integrates to 1. This is called defective.\n\n\n\n\n\n\nFigure 1: Figure 1: Realization of a Four-Parameter Diffusion Process Modeling the Binary Decision Process. Image from Henrich et al. (2024), distributed under the Creative Commons Attribution 4.0 International License. Note. The parameters are the boundary separation a for two response alternatives, the relative starting point w, the drift rate v, and the non-decision time t0. The decision process is illustrated as a jagged line between the two boundaries. The predicted distributions of the reaction times are depicted as curved lines below and above the response boundaries (blue).\n\n\n\nIn this model it is assumed that the decision process behaves like a random walk and we are interested in the first time that the random walk crosses one of the two decision boundaries. Hence, we are interested in the first-passage time of the decision process. The Stan function wiener_lpdf() returns the logarithm of the first-passage time density function for a diffusion model with up to seven parameters for upper boundary responses, \\(\\log(p_1)\\). As can be seen above, it suffices to implement the density for only one response boundary, as the other can be obtained by mirroring the starting point and drift rate. Any combination of fixed and estimated parameters can be specified. In other words, with this implementation it is not only possible to estimate parameters of the fullseven- parameter model, but also to estimate restricted models such as the basic four- parameter model, or a five or six-parameter model, or even a one-parameter model when fixing the other six parameters.\nFor example, it is possible to permit variability in just one or two parameters and to fix the other variabilities to 0, or even to estimate a three-parameter model when fixing more parameters (e.g., fixing the relative starting point at 0.5).\nIt is assumed that the reaction time data that correspond to the upper response boundary \\(y_\\text{upper}\\) is distributed according to wiener_lpdf():\n\\[\ny_\\text{lower} \\sim \\operatorname{wiener\\_lpdf}(a, t0, w, v, s_v, s_w, s_{t0})\n\\] and the reaction time data that correspond to the lower response boundary \\(y_\\text{lower}\\) is distributed according to wiener_lpdf() with mirrored starting point and drift rate:\n\\[\ny_\\text{upper} \\sim \\operatorname{wiener\\_lpdf}(a, t0, 1-w, -v, s_v, s_w, s_{t0})\n\\]\n\n\nThe following example demonstrates a diffusion model call in Stan:\ndata {\n  int &lt;lower=0&gt; N; // Number of trials\n  array[N] real rt; // response times (in seconds )\n  array[N] int &lt;lower=0, upper=1&gt; resp; // responses {0 ,1}\n}\ntransformed data{\n  real min_rt = min(rt);\n}\nparameters {\n  real &lt;lower=0&gt; a;                // boundary separation\n  real v;                          // drift\n  real &lt;lower=0, upper=1&gt; w;       // relative starting point\n  real &lt;lower=0, upper=min_rt&gt; t0; // non-decision time\n\n  real &lt;lower=0&gt; sv;               // variability in drift\n  // variability in starting point\n  real &lt;lower=0, upper=fmin(2 * w, 2 * (1 - w))&gt; sw; \n  real &lt;lower=0&gt; st0;             // variability in non-decision time\n}\ntransformed parameters{\n  real one_minus_w = 1 - w;\n  real neg_v = -v;\n}\nmodel {\n  // prior\n  a ~ normal(1, 1);\n  w ~ normal(0.5, 0.1);\n  v ~ normal(2, 3);\n  t0 ~ normal(0.435, 0.12);\n\n  sv ~ normal(1, 3);\n  st0 ~ normal(0.183, 0.09);\n  sw ~ beta(1, 3);\n\n  // likelihood (diffusion model)\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // upper boundary\n      target += wiener_full_lpdf(rt[i] | a, t0, w, v,\n                                         sv, sw, st0);\n    } else {\n      // lower boundary: mirror drift and starting point\n      target += wiener_full_lpdf(rt[i] | a, t0, one_minus_w,\n                                         neg_v, sv, sw, st0);\n    }\n  }\n}\n\n\nThe data should consist of at least three variables:\n\nThe number of trials N,\nthe response, coded as 0 = “lower bound” and 1 = “upper bound”, and\nthe reaction times in seconds (not milliseconds).\n\nNote that two different ways of coding responses are commonly used: First, in response coding, the boundaries correspond to the two response alternatives. Second, in accuracy coding, the boundaries correspond to correct (upper bound) and wrong (lower bound) responses. This means, depending on the coding you choose, the bounds mentioned in the second variable above differ and the response variable will have a different form.\nMost often, an experimenter wants to find out whether an experimental manipulation influences the model parameters. As there exists psychological interpretations for each diffusion model parameter, the experimenter can draw conclusions from differing parameters. Therefore, usually an own diffusion model is being computed for each experimental group to enable a comparison of the parameters between the groups. This can be manipulation between different subjects, like an experimental group and a control group (so called between-subject manipulations). However, this can also be manipulations within the same subject by presenting stimuli from different experimental groups (so called within-subject manipulations). Depending on the experimental design, one would typically also provide the number of conditions and the condition associated with each trial as a vector. Then, one model for each condition will be computed. This means that the parameters also have to be defined for each condition.\nIn a hierarchical setting, the data block would also specify the number of participants and the participant associated with each trial as a vector. It is also possible to hand over a precision value in the data block.\n\n\n\nThe model arguments of the wiener_lpdf() function that are not fixed to a certain value are defned as parameters in the parameters block. In this block, it is also possible to insert restrictions on the parameters. Note that the MCMC algorithm iteratively searches for the next parameter set. If the suggested sample falls outside the internally defined parameter ranges, the program will throw an error, which causes the algorithm to restart the current iteration. Since this slows down the sampling process, it is advisable to include the parameter ranges in the defnition of the parameters in the parameters block to improve the sampling process (see table below for the parameter ranges). In addition, the parameter space is further constrained by the following conditions:\n\nThe non-decision time \\(t_0\\) has to be smaller or equal to the observed reaction time: \\(t0 \\leq y\\).\nThe varying relative starting point \\(w\\) has to be in the interval (0,1) and thus,\n\n\\[\n\\begin{aligned}\n&w + \\frac{s_w}{2} &lt; 1 \\text{, and} \\\\\n&0 &lt; w-\\frac{s_w}{2}\n\\end{aligned}\n\\]\n\n\n\nParameter\nRange\n\nParameter\nRange\n\n\n\n\n\\(a\\)\n(0, \\(\\infty\\))\n\n\\(y\\)\n(0, \\(\\infty\\))\n\n\n\\(v\\)\n(-\\(\\infty\\), \\(\\infty\\))\n\n\\(s_v\\)\n[0, \\(\\infty\\))\n\n\n\\(w\\)\n(0,1)\n\n\\(s_w\\)\n[0,min(2w, 2(1-w)))\n\n\n\\(t_0\\)\n[0,\\(\\infty\\))\n\n\\(s_{t0}\\)\n[0,\\(\\infty\\))\n\n\n\n\n\n\nIn the model block, the priors and likelihood are defined for the upper and the lower response boundary. Different kinds of priors can be specifed here. Generally, the regularization induced by mildly informative priors can help both statistically and computationally.\nIn the second part of the model block, the data generating distribution is applied to all responses. The drift rate \\(v\\) and relative starting point \\(w\\) have to be mirrored for responses at the lower boundary.\nFor more details regarding the application of the diffusion model in Stan, see Henrich et al. (2024).\n\n\n\n\nTruncation and censoring frequently occur in psychological data collection. For reaction time data, truncated and censored data regularly arise in psychological studies as a consequence of using response windows or deadlines. These are sometimes introduced in the analysis of data to exclude reaction times that appear too short or too long, but they are also sometimes already built into the study procedures to push participants to respond within a specifc temporal window.\nDepending on the implementation of the response window, two different types of data arise: truncated data or censored data. Since the effects of truncation or censoring on summary statistics such as mean, median, standard deviation, and skewness is regularly too large to ignore (Ulrich and Miller 1994), data analysts are well advised to account for these effects.\nAs described in the Truncated or Censored Data chapter, the cumulative distribution function (\\(F\\)) and its complement (\\(\\text{CCDF}\\)) are needed to model truncated and censored data.\nAs explained above, \\(p\\) is defined defectively, meaning that only the sum of \\(p\\)s for both response alternatives integrates to 1. For the same reason, \\(F\\) and \\(\\text{CCDF}\\) are also implemented defectively. Analogously, only the sum of the \\(F\\)s and \\(\\text{CCDF}\\)s for both response alternatives asymptotes above at 1.\nIn the case of the diffusion model, \\(F\\) asymptotes above at the probability \\(PROB\\) to hit the corresponding response boundary: (for simplicity, we omit the inter-trial variabilities in the following)\n\\[\n\\begin{aligned}\nF_1(\\infty\\mid a,w,v) &= \\text{PROB}(a,w,v) \\text{ and} \\\\\nF_0(\\infty\\mid a,w,v) &= F_1(\\infty\\mid a,1-w,-v) = \\text{PROB}(a,1-w,-v)\n\\end{aligned}\n\\]\n\n\nData are called truncated when there is no information available for analysis from trials with values larger (or smaller) than a right (or left) reaction-time bound. In reaction time experiments, reaction time data are truncated if trials with reaction times outside the response window are excluded from the analysis. Not even a count of those omitted trials is kept.\nLet \\(L\\) denote the left reaction-time bound and \\(U\\) denote the right reaction-time bound of a response window.\nThen, the density of truncated data for both response boundaries 0 and 1, here denoted as \\(\\text{resp}\\in\\{0,1\\}\\), can be formulated as follows:\n\\[\n\\begin{aligned}\n&p_{\\text{resp}}(y \\mid L&lt;X\\leq U, a, w, v) = \\\\ &\\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\leq U\\}}}\n{\\bigl(F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)\\bigr) -\n\\bigl(F_0(L\\mid a, w, v)+F_1(L\\mid a, w, v)\\bigr)}\n\\end{aligned}\n\\]\nThe density of left truncated data can be formulated as follows. \\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid L&lt;X, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\}}}\n{1-\\bigl(F_0(L \\mid a, w, v)+F_1(L \\mid a, w, v)\\bigr)},\n\\end{aligned}\n\\]\nThe density of right truncated data can be formulated as follows.\n\\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid X\\leq U, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{y\\leq U\\}}}{F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)}\n\\end{aligned}\n\\]\nAs the functions are implemented defectively, a truncated diffusion model cannot be calculated with the truncation functor \\(T[,]\\) as it would usually be done in Stan. This means the function call: y ~ wiener(...)T[L,U] does not work the way it is supposed to. When the truncation functor is called in Stan, Stan searches for a CDF implementation internally. In the case of the diffusion model, Stan would find the CDF, but is not aware of its defective implementation and calculates the computations as if it were a non-defective CDF. This causes misleading and incorrect results.\nTo implement the truncated model, write out the function shown above on the log-scale with left_bound = L and right_bound = U, where wiener_lcdf_unnorm() calls the logarithmized CDF of the diffusion model at the response-1-boundary:\nmodel {\n  real log_denom = log_diff_exp(\n    log_sum_exp(\n      wiener_lcdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                       sv, sw, st)),\n    log_sum_exp(\n      wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                      sv, sw, st)));\n  // likelihood\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // response -1 boundary\n      target += wiener_lpdf (rt[i] | a, t0, w, v, sv, sw, st);\n    } else { \n      // response -0 boundary ( mirror v and w)\n      target += wiener_lpdf (rt[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st);\n    }\n  } // end for\n  target += -N * log_denom;\n}\nFor details of how to call a truncated model within the parallelization routine of reduce_sum or with truncation to only on side, see Henrich and Klauer (2026).\n\n\n\nData are censored when observations that are above or below a right or left boundary value are reported as occurrences of the event \\((y &gt;\nU)\\), for \\(U\\) the right bound, or as occurrences of the event \\((y \\leq\nL)\\), for \\(L\\) the left bound, respectively. Like for truncated data, the range of the possible values is restricted, but the number of observations that fall outside the boundaries is kept, whereas in truncation, no count would be kept.\nFor the censored model, we distinguish two cases. In the first case, the responses of the censored trials are known, but the reaction times are not known. In the second case, neither the responses nor the reaction times of the censored trials are known. Note that the second case differs from a truncated model in the fact that the number of censored trials is still known. Consider first the case where the response is known even for censored data.\nTo model such data in Stan, the left and right reaction time bounds, left_bound and right_bound, respectively, are handed over in the data block, as well as a vector censored that tracks whether a trial is censored (= 1) or not (= 0), and counts of trials censored at the left reaction time bound and counts of trials censored at the right reaction time bound for each response in {0,1}. There are four such count variables: N_cens_left_0, N_cens_left_1, N_cens_right_0, N_cens_right_1:\nmodel {\n  for (i in 1:N) {\n    if (censored[i] == 0) {\n      if (resp[i] == 1) {\n        y[i] ~ wiener(a, t0, w, v, sv, sw, st0);\n      } else if (resp[i] == 0) {\n        y[i] ~ wiener(a, t0, one_minus_w, neg_v, sv, sw, st0);\n      }\n    }\n  }\n\n  // likelihood (response = 0)\n  target += N_cens_left_0 \n         * wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                           sv, sw, st0);\n\n  target += N_cens_right_0 \n         * wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                             sv, sw, st0);\n\n  // likelihood (response = 1)\n  target += N_cens_left_1 \n         * wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st0);\n\n  target += N_cens_right_1 \n         * wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st0);\n}\nWhen data are censored at only one side, meaning that the reaction time constraint only exists for one of the two boundaries, omit the lines for the other side in the code. A both sided reaction time window would be, for example, when only reaction times are accepted that occur between 0.2 and 0.8 seconds. A one sided reaction time constraint would be, for example, when all reaction times below 0.8 seconds are accepted.\nWhen data consist of many conditions (as explained in the beginning), it is sometimes more convenient to loop over all trials instead of using count variables as described above, using the following notation and code. A vector containing the information whether a trial is censored or not, here censored, needs to be handed over in the data block. This vector splits the data into three bins: all trials \\(i\\) withcensored[i]=0 are censored below the left reaction time bound, all trials \\(i\\) with censored[i]=1 fall between the reaction time bounds, and all trials \\(i\\) with censored[i]=2 are censored above the right reaction time bound. For non-censored trials, the log-PDF is computed, for left censored trials, the log-CDF is computed, and for right censored trials, the log-CCDF is computed:\nmodel { \n  for (i in 1:N) { \n    // right censored at right_bound\n    if (resp [i] == 1) { \n      // upper response boundary\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, w, v, \n                                                  sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, w, v, sv, sw, st0);\n      } else if (censored[i] == 2) {\n         target += wiener_lccdf_unnorm(right_bound | a, t0, w, v,\n                                                     sv, sw, st0);\n      }\n    } else { \n      // lower response boundary (mirror drift and // starting point!)\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w,\n                                                  neg_v, sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st0);\n      } else if (censored[i] == 2) {\n        target += wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                                    neg_v, sv, sw, st0);\n      }\n    }\n  }\n}\nWhen the data are censored on only one side, omit the case that is not needed.\nNote that this block can be inserted in the defnition of the parallelization function, partial_sum_wiener(), as defined below.\nSometimes also the response is missing (i.e., it is known that the reaction time in a trial fell outside the response window, but which response was given is unknown). One method that has been used to model such data has involved inferring the numbers of missing responses of either kind from the observed relative frequencies of the two responses. This approach has the problem that quite specifc assumptions on the missing data have to be made (namely, that the proportions of the two kinds of responses are the same for responses within and outside the response window).\nThe following is a more principled approach that uses the cumulative distribution functions and their complements to provide the data-generating distribution of censored data. As before, let \\(L\\) be the left reaction time bound, and \\(U\\) the right reaction time bound, and consider decision times without inter-trial variabilities for the sake of simplicity. It follows that the likelihood contribution \\(\\textit{lik}_l\\) for a left-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_l(a,w,v) = F_0(L\\mid a,w,v) + F_1(L\\mid a,w,v),\n\\end{aligned}\n\\]\nwhereas the likelihood contribution \\(lik_r\\) due to a right-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_r(a,w,v) = \\text{CCDF}_0(U\\mid a,w,v) + \\text{CCDF}_1(U\\mid a,w,v).\n\\end{aligned}\n\\]\nSee the following code for an example of Stan code implementing this second case of censoring. This model call deals with the problem of unknown responses by computing the probability of choosing the response-1 or response-0 boundary outside the response window. Here, the CDF and/or the CCDF are required, depending upon whether there is only left-censoring, right-censoring, or censoring both to the left and to the right. The following code shows the functions block for a model that is right-censored using the function partial_sum_wiener() to parallelize the execution of a single Stan chain across multiple cores:\nfunctions {\n  real partial_sum_wiener(array[] real rt_slice, int start,\n                          int end, real a, real t0, real w,\n                          real v, real sv, real sw, real st,\n                          array[] int resp, real right_bound,\n                          array[] int censored) {\n    real ans = 0;\n    for (i in start:end) {\n      if (censored[i] == 1) {\n        // not censored\n        if (resp[i] == 1) {\n          // upper boundary\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, w, v,\n                                                     sv, sw, st);\n        } else {\n          // lower boundary(mirror v and w)\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, one_minus_w,\n                                                     neg_v, sv, sw, st);\n        }\n      } else { \n        // censored\n        ans += log_sum_exp (\n          wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n          wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                            neg_v, sv, sw, st);\n      }\n    }\n    return ans;\n  }\n}\nCombine this block with the model block in the example above by using the function reduce_sum().\n  target += reduce_sum(partial_sum_wiener, rt, 1,\n    a, t0, w, v, sv, sw, st, resp, right_bound, censored);\n}\nFor more details, see Henrich and Klauer (2026).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Using the Stan Compiler"
+      "Example Models",
+      "Wiener diffusion model"
     ]
   },
   {
-    "objectID": "stan-users-guide/using-stanc.html#stanc-args",
-    "href": "stan-users-guide/using-stanc.html#stanc-args",
-    "title": "Using the Stan Compiler",
+    "objectID": "stan-users-guide/wiener_diffusion_model.html#function-call-example",
+    "href": "stan-users-guide/wiener_diffusion_model.html#function-call-example",
+    "title": "Wiener diffusion model",
     "section": "",
-    "text": "The stanc3 compiler has the following command-line syntax:\n&gt; stanc (options) &lt;model_file&gt;\nwhere &lt;model_file&gt; is either a path to a file ending in .stan or .stanfunctions (which automatically sets --standalone-functions), or '-' to read from standard input.\nThe stanc3 options are:\n\n--help / -? - Displays the complete list of stanc3 options, then exits.\n--version - Display stanc version number\n--info - Print information about the model, such as the type information for variables and the list of used distributions.\n--name=&lt;model_name&gt; - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.\n--o=&lt;file_name&gt; / -o=&lt;filename&gt; / --output=&lt;filename&gt; - Specify a path to an output file for generated C++ code (default = .hpp) or auto-formatting output (default: no file/print to stdout)\n--auto-format - Pretty prints the program to the console. See more on auto formatting.\n--allow-undefined - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.\n--canonicalize - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.\n--include_paths=&lt;dir1,...dirN&gt; - Takes a comma-separated list of directories that may contain a file in an #include directive.\n--max-line-length=&lt;number&gt; - Set the column number at which formatting with --auto-format attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.\n--print-canonical - Synonymous with --auto-format --canonicalize=deprecations,includes,parentheses,braces.\n--print-cpp - If set, output the generated C++ Stan model class to stdout.\n--filename-in-msg=&lt;name&gt; - Sets the filename used in compiler and runtime errors. If absent, the &lt;model_file&gt; argument is used.\n--standalone-functions - If set, only generate the code for the functions defined in the file. This is the default behavior for .stanfunctions files.\n--O0 (Default) Do not apply optimizations to the Stan code.\n--O1 Apply level 1 compiler optimizations (only basic optimizations).\n--Oexperimental WARNING: This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance! Allow the compiler to apply all optimizations to the Stan code.\n--O Synonym for --O1 as of Stan 2.37. In earlier versions this was a synonym for --Oexperimental.\n--use-opencl - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.\n--warn-pedantic - Emit warnings in Pedantic mode which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.\n--warn-uninitialized - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.\n--color - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are auto (the default), always, never. Can also be controlled by the STANC_COLOR environment variable.\n\nThe compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the --help option to see the full set.",
+    "text": "The following example demonstrates a diffusion model call in Stan:\ndata {\n  int &lt;lower=0&gt; N; // Number of trials\n  array[N] real rt; // response times (in seconds )\n  array[N] int &lt;lower=0, upper=1&gt; resp; // responses {0 ,1}\n}\ntransformed data{\n  real min_rt = min(rt);\n}\nparameters {\n  real &lt;lower=0&gt; a;                // boundary separation\n  real v;                          // drift\n  real &lt;lower=0, upper=1&gt; w;       // relative starting point\n  real &lt;lower=0, upper=min_rt&gt; t0; // non-decision time\n\n  real &lt;lower=0&gt; sv;               // variability in drift\n  // variability in starting point\n  real &lt;lower=0, upper=fmin(2 * w, 2 * (1 - w))&gt; sw; \n  real &lt;lower=0&gt; st0;             // variability in non-decision time\n}\ntransformed parameters{\n  real one_minus_w = 1 - w;\n  real neg_v = -v;\n}\nmodel {\n  // prior\n  a ~ normal(1, 1);\n  w ~ normal(0.5, 0.1);\n  v ~ normal(2, 3);\n  t0 ~ normal(0.435, 0.12);\n\n  sv ~ normal(1, 3);\n  st0 ~ normal(0.183, 0.09);\n  sw ~ beta(1, 3);\n\n  // likelihood (diffusion model)\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // upper boundary\n      target += wiener_full_lpdf(rt[i] | a, t0, w, v,\n                                         sv, sw, st0);\n    } else {\n      // lower boundary: mirror drift and starting point\n      target += wiener_full_lpdf(rt[i] | a, t0, one_minus_w,\n                                         neg_v, sv, sw, st0);\n    }\n  }\n}\n\n\nThe data should consist of at least three variables:\n\nThe number of trials N,\nthe response, coded as 0 = “lower bound” and 1 = “upper bound”, and\nthe reaction times in seconds (not milliseconds).\n\nNote that two different ways of coding responses are commonly used: First, in response coding, the boundaries correspond to the two response alternatives. Second, in accuracy coding, the boundaries correspond to correct (upper bound) and wrong (lower bound) responses. This means, depending on the coding you choose, the bounds mentioned in the second variable above differ and the response variable will have a different form.\nMost often, an experimenter wants to find out whether an experimental manipulation influences the model parameters. As there exists psychological interpretations for each diffusion model parameter, the experimenter can draw conclusions from differing parameters. Therefore, usually an own diffusion model is being computed for each experimental group to enable a comparison of the parameters between the groups. This can be manipulation between different subjects, like an experimental group and a control group (so called between-subject manipulations). However, this can also be manipulations within the same subject by presenting stimuli from different experimental groups (so called within-subject manipulations). Depending on the experimental design, one would typically also provide the number of conditions and the condition associated with each trial as a vector. Then, one model for each condition will be computed. This means that the parameters also have to be defined for each condition.\nIn a hierarchical setting, the data block would also specify the number of participants and the participant associated with each trial as a vector. It is also possible to hand over a precision value in the data block.\n\n\n\nThe model arguments of the wiener_lpdf() function that are not fixed to a certain value are defned as parameters in the parameters block. In this block, it is also possible to insert restrictions on the parameters. Note that the MCMC algorithm iteratively searches for the next parameter set. If the suggested sample falls outside the internally defined parameter ranges, the program will throw an error, which causes the algorithm to restart the current iteration. Since this slows down the sampling process, it is advisable to include the parameter ranges in the defnition of the parameters in the parameters block to improve the sampling process (see table below for the parameter ranges). In addition, the parameter space is further constrained by the following conditions:\n\nThe non-decision time \\(t_0\\) has to be smaller or equal to the observed reaction time: \\(t0 \\leq y\\).\nThe varying relative starting point \\(w\\) has to be in the interval (0,1) and thus,\n\n\\[\n\\begin{aligned}\n&w + \\frac{s_w}{2} &lt; 1 \\text{, and} \\\\\n&0 &lt; w-\\frac{s_w}{2}\n\\end{aligned}\n\\]\n\n\n\nParameter\nRange\n\nParameter\nRange\n\n\n\n\n\\(a\\)\n(0, \\(\\infty\\))\n\n\\(y\\)\n(0, \\(\\infty\\))\n\n\n\\(v\\)\n(-\\(\\infty\\), \\(\\infty\\))\n\n\\(s_v\\)\n[0, \\(\\infty\\))\n\n\n\\(w\\)\n(0,1)\n\n\\(s_w\\)\n[0,min(2w, 2(1-w)))\n\n\n\\(t_0\\)\n[0,\\(\\infty\\))\n\n\\(s_{t0}\\)\n[0,\\(\\infty\\))\n\n\n\n\n\n\nIn the model block, the priors and likelihood are defined for the upper and the lower response boundary. Different kinds of priors can be specifed here. Generally, the regularization induced by mildly informative priors can help both statistically and computationally.\nIn the second part of the model block, the data generating distribution is applied to all responses. The drift rate \\(v\\) and relative starting point \\(w\\) have to be mirrored for responses at the lower boundary.\nFor more details regarding the application of the diffusion model in Stan, see Henrich et al. (2024).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Using the Stan Compiler"
+      "Example Models",
+      "Wiener diffusion model"
     ]
   },
   {
-    "objectID": "stan-users-guide/using-stanc.html#understanding-stanc3-errors-and-warnings",
-    "href": "stan-users-guide/using-stanc.html#understanding-stanc3-errors-and-warnings",
-    "title": "Using the Stan Compiler",
+    "objectID": "stan-users-guide/wiener_diffusion_model.html#wiener-truncation",
+    "href": "stan-users-guide/wiener_diffusion_model.html#wiener-truncation",
+    "title": "Wiener diffusion model",
     "section": "",
-    "text": "During model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).\n\n\nEven without the optional --warn-pedantic and --warn-uninitialized command line flags, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations\n\nA completely blank Stan program will produce the following warning message\nWarning in 'empty.stan', line 1, column 0 to line 2, column 0:\n    Empty model detected; this is a valid Stan model but likely unintended!\nThe use of any deprecated features will lead to warnings which will look as follows\n Warning in 'deprecated.stan', line 2, column 10 to column 17:\n     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with\n     an independent lognormal distribution on the scales, see:\n     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution\nA single Stan program can produce many warnings during compilation.\n\n\n\n\nErrors differ from warnings in their severity and format. In particular, errors are fatal and stop compilation, so at most one error is displayed per run of stanc.\nThere are five kinds of errors emitted by stanc3\n\nFile errors occur when the file passed to stanc is either missing or cannot be opened (i.e. has permissions issues). They look like\nError: file 'notfound.stan' not found or cannot be opened\nSyntactic errors occur whenever a program violates the Stan language’s syntax requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the #include directive fails, and “parsing” errors which result when the structure of the program is incorrect.\n\nThe lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of $ in a variable name will look like the following.\nSyntax error in 'char.stan', line 2, column 7 to column 8, lexing error:\n-------------------------------------------------\n  1:  data {\n  2:     int $ome_variable;\n             ^\n  3:  }\n-------------------------------------------------\nInvalid character found.\nWhen an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).\nSyntax error in './incl.stan', line 1, column 0, included from\n'./incl.stan', line 1, column 0, included from\n'incl.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include &lt;incl.stan&gt;\n         ^\n   -------------------------------------------------\nFile incl.stan recursively included itself.\nIt is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.\nSyntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:\n   -------------------------------------------------\n     1:  data {\n     2:     int&lt;lower=0&gt; N;\n     3:     vector x;\n                   ^\n     4:  }\n   -------------------------------------------------\nIll-formed type. Expected \"[\" expression \"]\" for vector size.\n\nSemantic errors (also known as type errors) occur when a program is structured correctly but features an error in the type rules imposed by the language. An example of this is assigning a real value to a variable defined as an integer.\nSemantic error in 'type.stan', line 2, column 3 to column 15:\n   -------------------------------------------------\n     1:  transformed data {\n     2:     int x = 1.5;\n            ^\n     3:  }\n   -------------------------------------------------\nIll-typed arguments supplied to assignment operator =:\nThe left hand side has type\n  int\nand the right hand side has type\n  real\nThe compiler will raise an error for use of any removed features for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.\nFinally, the compiler can raise an internal error. These are caused by bugs in the compiler, not your model, and we would appreciate it if you report them on the stanc3 repo with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.",
+    "text": "Truncation and censoring frequently occur in psychological data collection. For reaction time data, truncated and censored data regularly arise in psychological studies as a consequence of using response windows or deadlines. These are sometimes introduced in the analysis of data to exclude reaction times that appear too short or too long, but they are also sometimes already built into the study procedures to push participants to respond within a specifc temporal window.\nDepending on the implementation of the response window, two different types of data arise: truncated data or censored data. Since the effects of truncation or censoring on summary statistics such as mean, median, standard deviation, and skewness is regularly too large to ignore (Ulrich and Miller 1994), data analysts are well advised to account for these effects.\nAs described in the Truncated or Censored Data chapter, the cumulative distribution function (\\(F\\)) and its complement (\\(\\text{CCDF}\\)) are needed to model truncated and censored data.\nAs explained above, \\(p\\) is defined defectively, meaning that only the sum of \\(p\\)s for both response alternatives integrates to 1. For the same reason, \\(F\\) and \\(\\text{CCDF}\\) are also implemented defectively. Analogously, only the sum of the \\(F\\)s and \\(\\text{CCDF}\\)s for both response alternatives asymptotes above at 1.\nIn the case of the diffusion model, \\(F\\) asymptotes above at the probability \\(PROB\\) to hit the corresponding response boundary: (for simplicity, we omit the inter-trial variabilities in the following)\n\\[\n\\begin{aligned}\nF_1(\\infty\\mid a,w,v) &= \\text{PROB}(a,w,v) \\text{ and} \\\\\nF_0(\\infty\\mid a,w,v) &= F_1(\\infty\\mid a,1-w,-v) = \\text{PROB}(a,1-w,-v)\n\\end{aligned}\n\\]\n\n\nData are called truncated when there is no information available for analysis from trials with values larger (or smaller) than a right (or left) reaction-time bound. In reaction time experiments, reaction time data are truncated if trials with reaction times outside the response window are excluded from the analysis. Not even a count of those omitted trials is kept.\nLet \\(L\\) denote the left reaction-time bound and \\(U\\) denote the right reaction-time bound of a response window.\nThen, the density of truncated data for both response boundaries 0 and 1, here denoted as \\(\\text{resp}\\in\\{0,1\\}\\), can be formulated as follows:\n\\[\n\\begin{aligned}\n&p_{\\text{resp}}(y \\mid L&lt;X\\leq U, a, w, v) = \\\\ &\\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\leq U\\}}}\n{\\bigl(F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)\\bigr) -\n\\bigl(F_0(L\\mid a, w, v)+F_1(L\\mid a, w, v)\\bigr)}\n\\end{aligned}\n\\]\nThe density of left truncated data can be formulated as follows. \\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid L&lt;X, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{L&lt;y\\}}}\n{1-\\bigl(F_0(L \\mid a, w, v)+F_1(L \\mid a, w, v)\\bigr)},\n\\end{aligned}\n\\]\nThe density of right truncated data can be formulated as follows.\n\\[\n\\begin{aligned}\np_{\\text{resp}}(y \\mid X\\leq U, a, w, v) = \\frac{p_{\\text{resp}}(y \\mid a, w, v)\\cdot \\mathbb{I}_{\\{y\\leq U\\}}}{F_0(U \\mid a, w, v)+F_1(U \\mid a, w, v)}\n\\end{aligned}\n\\]\nAs the functions are implemented defectively, a truncated diffusion model cannot be calculated with the truncation functor \\(T[,]\\) as it would usually be done in Stan. This means the function call: y ~ wiener(...)T[L,U] does not work the way it is supposed to. When the truncation functor is called in Stan, Stan searches for a CDF implementation internally. In the case of the diffusion model, Stan would find the CDF, but is not aware of its defective implementation and calculates the computations as if it were a non-defective CDF. This causes misleading and incorrect results.\nTo implement the truncated model, write out the function shown above on the log-scale with left_bound = L and right_bound = U, where wiener_lcdf_unnorm() calls the logarithmized CDF of the diffusion model at the response-1-boundary:\nmodel {\n  real log_denom = log_diff_exp(\n    log_sum_exp(\n      wiener_lcdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                       sv, sw, st)),\n    log_sum_exp(\n      wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st),\n      wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                      sv, sw, st)));\n  // likelihood\n  for (i in 1:N) {\n    if (resp[i] == 1) {\n      // response -1 boundary\n      target += wiener_lpdf (rt[i] | a, t0, w, v, sv, sw, st);\n    } else { \n      // response -0 boundary ( mirror v and w)\n      target += wiener_lpdf (rt[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st);\n    }\n  } // end for\n  target += -N * log_denom;\n}\nFor details of how to call a truncated model within the parallelization routine of reduce_sum or with truncation to only on side, see Henrich and Klauer (2026).\n\n\n\nData are censored when observations that are above or below a right or left boundary value are reported as occurrences of the event \\((y &gt;\nU)\\), for \\(U\\) the right bound, or as occurrences of the event \\((y \\leq\nL)\\), for \\(L\\) the left bound, respectively. Like for truncated data, the range of the possible values is restricted, but the number of observations that fall outside the boundaries is kept, whereas in truncation, no count would be kept.\nFor the censored model, we distinguish two cases. In the first case, the responses of the censored trials are known, but the reaction times are not known. In the second case, neither the responses nor the reaction times of the censored trials are known. Note that the second case differs from a truncated model in the fact that the number of censored trials is still known. Consider first the case where the response is known even for censored data.\nTo model such data in Stan, the left and right reaction time bounds, left_bound and right_bound, respectively, are handed over in the data block, as well as a vector censored that tracks whether a trial is censored (= 1) or not (= 0), and counts of trials censored at the left reaction time bound and counts of trials censored at the right reaction time bound for each response in {0,1}. There are four such count variables: N_cens_left_0, N_cens_left_1, N_cens_right_0, N_cens_right_1:\nmodel {\n  for (i in 1:N) {\n    if (censored[i] == 0) {\n      if (resp[i] == 1) {\n        y[i] ~ wiener(a, t0, w, v, sv, sw, st0);\n      } else if (resp[i] == 0) {\n        y[i] ~ wiener(a, t0, one_minus_w, neg_v, sv, sw, st0);\n      }\n    }\n  }\n\n  // likelihood (response = 0)\n  target += N_cens_left_0 \n         * wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,\n                                           sv, sw, st0);\n\n  target += N_cens_right_0 \n         * wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,\n                                             sv, sw, st0);\n\n  // likelihood (response = 1)\n  target += N_cens_left_1 \n         * wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st0);\n\n  target += N_cens_right_1 \n         * wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st0);\n}\nWhen data are censored at only one side, meaning that the reaction time constraint only exists for one of the two boundaries, omit the lines for the other side in the code. A both sided reaction time window would be, for example, when only reaction times are accepted that occur between 0.2 and 0.8 seconds. A one sided reaction time constraint would be, for example, when all reaction times below 0.8 seconds are accepted.\nWhen data consist of many conditions (as explained in the beginning), it is sometimes more convenient to loop over all trials instead of using count variables as described above, using the following notation and code. A vector containing the information whether a trial is censored or not, here censored, needs to be handed over in the data block. This vector splits the data into three bins: all trials \\(i\\) withcensored[i]=0 are censored below the left reaction time bound, all trials \\(i\\) with censored[i]=1 fall between the reaction time bounds, and all trials \\(i\\) with censored[i]=2 are censored above the right reaction time bound. For non-censored trials, the log-PDF is computed, for left censored trials, the log-CDF is computed, and for right censored trials, the log-CCDF is computed:\nmodel { \n  for (i in 1:N) { \n    // right censored at right_bound\n    if (resp [i] == 1) { \n      // upper response boundary\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, w, v, \n                                                  sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, w, v, sv, sw, st0);\n      } else if (censored[i] == 2) {\n         target += wiener_lccdf_unnorm(right_bound | a, t0, w, v,\n                                                     sv, sw, st0);\n      }\n    } else { \n      // lower response boundary (mirror drift and // starting point!)\n      if (censored[i] == 0) {\n        target += wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w,\n                                                  neg_v, sv, sw, st0);\n      } else if (censored[i] == 1) {\n        target += wiener_lpdf(y[i] | a, t0, one_minus_w, neg_v,\n                                     sv, sw, st0);\n      } else if (censored[i] == 2) {\n        target += wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                                    neg_v, sv, sw, st0);\n      }\n    }\n  }\n}\nWhen the data are censored on only one side, omit the case that is not needed.\nNote that this block can be inserted in the defnition of the parallelization function, partial_sum_wiener(), as defined below.\nSometimes also the response is missing (i.e., it is known that the reaction time in a trial fell outside the response window, but which response was given is unknown). One method that has been used to model such data has involved inferring the numbers of missing responses of either kind from the observed relative frequencies of the two responses. This approach has the problem that quite specifc assumptions on the missing data have to be made (namely, that the proportions of the two kinds of responses are the same for responses within and outside the response window).\nThe following is a more principled approach that uses the cumulative distribution functions and their complements to provide the data-generating distribution of censored data. As before, let \\(L\\) be the left reaction time bound, and \\(U\\) the right reaction time bound, and consider decision times without inter-trial variabilities for the sake of simplicity. It follows that the likelihood contribution \\(\\textit{lik}_l\\) for a left-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_l(a,w,v) = F_0(L\\mid a,w,v) + F_1(L\\mid a,w,v),\n\\end{aligned}\n\\]\nwhereas the likelihood contribution \\(lik_r\\) due to a right-censored data point is given by\n\\[\n\\begin{aligned}\n\\textit{lik}_r(a,w,v) = \\text{CCDF}_0(U\\mid a,w,v) + \\text{CCDF}_1(U\\mid a,w,v).\n\\end{aligned}\n\\]\nSee the following code for an example of Stan code implementing this second case of censoring. This model call deals with the problem of unknown responses by computing the probability of choosing the response-1 or response-0 boundary outside the response window. Here, the CDF and/or the CCDF are required, depending upon whether there is only left-censoring, right-censoring, or censoring both to the left and to the right. The following code shows the functions block for a model that is right-censored using the function partial_sum_wiener() to parallelize the execution of a single Stan chain across multiple cores:\nfunctions {\n  real partial_sum_wiener(array[] real rt_slice, int start,\n                          int end, real a, real t0, real w,\n                          real v, real sv, real sw, real st,\n                          array[] int resp, real right_bound,\n                          array[] int censored) {\n    real ans = 0;\n    for (i in start:end) {\n      if (censored[i] == 1) {\n        // not censored\n        if (resp[i] == 1) {\n          // upper boundary\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, w, v,\n                                                     sv, sw, st);\n        } else {\n          // lower boundary(mirror v and w)\n          ans += wiener_lpdf(rt_slice[i+1- start ] | a, t0, one_minus_w,\n                                                     neg_v, sv, sw, st);\n        }\n      } else { \n        // censored\n        ans += log_sum_exp (\n          wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),\n          wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,\n                                            neg_v, sv, sw, st);\n      }\n    }\n    return ans;\n  }\n}\nCombine this block with the model block in the example above by using the function reduce_sum().\n  target += reduce_sum(partial_sum_wiener, rt, 1,\n    a, t0, w, v, sv, sw, st, resp, right_bound, censored);\n}\nFor more details, see Henrich and Klauer (2026).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Using the Stan Compiler"
+      "Example Models",
+      "Wiener diffusion model"
     ]
   },
   {
-    "objectID": "stan-users-guide/using-stanc.html#pedantic-mode",
-    "href": "stan-users-guide/using-stanc.html#pedantic-mode",
-    "title": "Using the Stan Compiler",
+    "objectID": "stan-users-guide/user-functions.html",
+    "href": "stan-users-guide/user-functions.html",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Pedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.\nFor example, consider the following program.\ndata {\n  int N;\n  array[N] real x;\n}\nparameters {\n  real sigma;\n}\nmodel {\n  real mu;\n  x ~ normal(mu, sigma);\n}\nWhen pedantic mode is turned on, the compiler will produce the following warnings.\nWarning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:\n    The parameter sigma has no priors. This means either no prior is\n    provided, or the prior(s) depend on data variables. In the later case,\n    this may be a false positive.\nWarning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:\n    The variable mu may not have been assigned a value before its use.\nWarning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:\n    A normal distribution is given parameter sigma as a scale parameter\n    (argument 2), but sigma was not constrained to be strictly positive.\nHere are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):\n\nDistribution usages issues. Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.\nUnused parameter. A parameter is defined but doesn’t contribute to target.\nLarge or small constant in a distribution. Very large or very small constants are used as distribution arguments.\nControl flow depends on a parameter. Branching control flow (like if/else) depends on a parameter value .\nParameter has multiple tildes. A parameter is on the left-hand side of multiple tildes.\nParameter has zero or multiple priors. A parameter has zero or more than one prior distribution.\nVariable is used before assignment. A variable is used before being assigned a value.\nStrict or nonsensical parameter bounds. A parameter is given questionable bounds.\nNonlinear transformations. When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.\n\nSome important limitations of pedantic mode are listed at the end of this chapter.\n\n\nWhen an argument to a built-in distribution certainly does not match that distribution’s specification in the Stan Functions Reference, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g. simplex) is incompatible with the domain of the distribution. x\nFor example, consider the following program.\nparameters {\n  real unb_p;\n  real&lt;lower=0&gt; pos_p;\n}\nmodel {\n  1 ~ poisson(unb_p);\n  1 ~ poisson(pos_p);\n}\nThe parameter of poisson should be strictly positive, but unb_p is not constrained to be positive.\nPedantic mode produces the following warning.\nWarning in 'ex-dist-args.stan', line 6, column 14 to column 19:\n  A poisson distribution is given parameter unb_p as a rate parameter\n  (argument 1), but unb_p was not constrained to be strictly positive.\n\n\n\nPedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:\n\n\nAny use of uniform distribution generates a warning, except when the variate parameter’s declared upper and lower bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.\nFor example, consider the following program.\nparameters {\n  real a;\n  real&lt;lower=0, upper=1&gt; b;\n}\nmodel {\n  a ~ uniform(0, 1);\n  b ~ uniform(0, 1);\n}\na is assigned a uniform distribution that doesn’t match its constraints.\nPedantic mode produces the following warning.\nWarning in 'uniform-warn.stan', line 6, column 2 to column 20:\n  Parameter a is given a uniform distribution. The uniform distribution is\n  not recommended, for two reasons: (a) Except when there are logical or\n  physical constraints, it is very unusual for you to be sure that a\n  parameter will fall inside a specified range, and (b) The infinite gradient\n  induced by a uniform density can cause difficulties for Stan's sampling\n  algorithm. As a consequence, we recommend soft constraints rather than hard\n  constraints; for example, instead of giving an elasticity parameter a\n  uniform(0, 1) distribution, try normal(0.5, 0.5).\n\n\n\nGamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.\n\n\n\nAny use of the lkj_corr distribution generates a warning that suggests using the Cholesky variant instead. See the LKJ correlation distribution section of the Stan Functions Reference for details.\n\n\n\n\nA warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the target variable depends in any way on each of the parameters.\nFor example, consider the following program.\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(1, 1);\n}\na participates in the density function but b does not.\nPedantic mode produces the following warning.\nWarning in 'unused.stan', line 3, column 2 to column 9:\n    The parameter b was declared but was not used in the density calculation.\n\n\n\nWhen numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See the efficiency tuning section of the Stan User’s guide for a discussion of scaling parameters.\nFor example, consider the following program.\nparameters {\n  real x;\n  real y;\n}\nmodel {\n  x ~ normal(-100, 100);\n  y ~ normal(0, 1);\n}\nThe constants -100 and 100 suggest that x is not unit scaled.\nPedantic mode produces the following warning.\nWarning in 'constants-warn.stan', line 6, column 13 to column 17:\n    Argument -100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\nWarning in 'constants-warn.stan', line 6, column 19 to column 22:\n    Argument 100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\n\n\n\nControl flow statements, such as if, for and while should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.\nFor example, consider the following program.\nparameters {\n  real a;\n}\nmodel {\n  // x depends on parameter a\n  real x = a * a;\n\n  int m;\n\n  // the if-then-else depends on x which depends on a\n  if(x &gt; 0) {\n    //now m depends on x which depends on a\n    m = 1;\n  } else {\n    m = 2;\n  }\n\n  // for loop depends on m -&gt; x -&gt; a\n  for (i in 0:m) {\n    a ~ normal(i, 1);\n  }\n}\nThe if and for statements are control flow that depend (indirectly) on the value of the parameter m.\nPedantic mode produces the following warning.\nWarning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:\n  A control flow statement depends on parameter(s): a.\nWarning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:\n  A control flow statement depends on parameter(s): a.\n\n\n\nA warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent target += conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.\nPedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(0, 1);\n  a ~ normal(x, 1);\n\n  b ~ normal(1, 1);\n}\nPedantic mode produces the following warning.\nWarning in 'multi-tildes.stan', line 9, column 2 to column 19:\n  The parameter a is on the left-hand side of more than one tildes\n  statement.\n\n\n\nA warning is generated when a parameter appears to have greater than or less than one prior distribution factor.\nThis analysis depends on a factor graph representation of a Stan program. A factor F that depends on a parameter P is called a prior factor for P if there is no path in the factor graph from F to any data variable except through P.\nOne limitation of this approach is that the compiler cannot distinguish between modeled data variables and other convenient uses of data variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for int variables) are modeled data, which may cause extra warnings.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n  real c;\n  real d;\n}\nmodel\n{\n  a ~ normal(0, 1); // this is a prior\n  x ~ normal(a, 1); // this is not a prior, since data is involved\n\n  b ~ normal(x, 1); // this is also not a prior, since data is involved\n\n  // this is not a prior for c, since data is involved through b\n  // but it is a prior for b, since the data is only involved through b\n  c ~ normal(b, 1);\n\n  //these are multiple priors:\n  d ~ normal(0, 1);\n  1 ~ normal(d, 1);\n}\nOne prior is found for a and for b, while c only has a factor that touches a data variable and d has multiple priors.\nPedantic mode produces the following warning.\nWarning in 'priors.stan', line 7, column 2 to column 9:\n    The parameter c has no priors. This means either no prior is provided, or\n    the prior(s) depend on data variables. In the later case, this may be a\n    false positive.\nWarning in 'priors.stan', line 8, column 2 to column 9:\n    The parameter d has 2 priors.\n\n\n\n\nA warning is generated when any variable is used before it has been assigned a value.\nFor example, consider the following program.\ntransformed data {\n  real x;\n  if (1 &gt; 2) {\n    x = 1;\n  } else {\n    print(\"oops\");\n  }\n  print(x);\n}\nSince x is only assigned in one of the branches of the if statement, it might get to print(x) without having been assigned to.\nPedantic mode produces the following warning.\nWarning in 'uninit-warn.stan', line 7, column 8 to column 9:\n  The variable x may not have been assigned a value before its use.\n\n\n\nExcept when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds &lt;lower=.., upper=..&gt; except for &lt;lower=0, upper=1&gt; or &lt;lower=-1, upper=1&gt;.\nIn addition, a warning is generated when a parameter bound is found to have lower &gt;= upper.\nFor example, consider the following program.\nparameters {\n  real&lt;lower=0, upper=1&gt; a;\n  real&lt;lower=-1, upper=1&gt; b;\n  real&lt;lower=-2, upper=1012&gt; c;\n}\nmodel {\n  c ~ normal(b, a);\n}\nPedantic mode produces the following warning.\nWarning in 'hard-constraint.stan', line 4, column 2 to column 31:\n    Your Stan program has a parameter c with a lower and upper bound in its\n    declaration. These hard constraints are not recommended, for two reasons:\n    (a) Except when there are logical or physical constraints, it is very\n    unusual for you to be sure that a parameter will fall inside a specified\n    range, and (b) The infinite gradient induced by a hard constraint can\n    cause difficulties for Stan's sampling algorithm. As a consequence, we\n    recommend soft constraints rather than hard constraints; for example,\n    instead of constraining an elasticity parameter to fall between 0, and 1,\n    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.\n\n\n\nWhen a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the Changes of variables section.\nThis portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.\nFor example, consider the following program.\nparameters {\n  real y;\n}\nmodel {\n  log(y) ~ normal(0,1);\n}\nPedantic mode produces the following warning.\nWarning in 'jacobian.stan', line 5, column 2 to column 23:\n    Left-hand side of distribution statement (~) may contain a non-linear\n    transform of a parameter or local variable. If it does, you need to\n    include a target += statement with the log absolute determinant of the\n    Jacobian of the transform. You could also consider defining a transformed\n    parameter and using jacobian += in the transformed parameters block.\n\n\n\n\nConstant values are sometimes uncomputable\nPedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code normal(x, 1 - 2), the expression 1 - 2 will be evaluated to -1, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.\nContainer types\nCurrently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).\nData variables\nThe declaration information for data variables is currently not considered, so using data as incompatible arguments to distributions may not generate the appropriate warnings.\nControl flow dependent on parameters in nested functions\nIf a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.",
+    "text": "This chapter explains functions from a user perspective with examples; see the language reference for a full specification. User-defined functions allow computations to be encapsulated into a single named unit and invoked elsewhere by name. Similarly, functions allow complex procedures to be broken down into more understandable components. Writing modular code using descriptively named functions is easier to understand than a monolithic program, even if the latter is heavily commented.1\n\n\nHere’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.\nfunctions {\n  real relative_diff(real x, real y) {\n    real abs_diff;\n    real avg_scale;\n    abs_diff = abs(x - y);\n    avg_scale = (abs(x) + abs(y)) / 2;\n    return abs_diff / avg_scale;\n  }\n}\n// ...\ngenerated quantities {\n  real rdiff;\n  rdiff = relative_diff(alpha, beta);\n}\nThe function is named relative_diff, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.\n\n\nAll functions are defined in their own block, which is labeled functions and must appear before all other program blocks. The user-defined functions block is optional.\n\n\n\nThe body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.\n\n\n\nReturn statements, such as the one on the last line of the definition of relative_diff above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.\nIf an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a fatal_error statement.\n\n\nRejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).\nThe most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.\nreal dbl_sqrt(real x) {\n  if (!(x &gt;= 0)) {\n    reject(\"dbl_sqrt(x): x must be positive; found x = \", x);\n  }\n  return 2 * sqrt(x);\n}\nThe negation of the positive test is important, because it also catches the case where x is a not-a-number value. If the condition had been coded as (x &lt; 0) it would not catch the not-a-number case, though it could be written as (x &lt; 0 || is_nan(x)). The positive infinite case is allowed through, but could also be checked with the is_inf(x) function. The square root function does not itself reject, but some downstream consumer of dbl_sqrt(-2) would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.\nThe effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.2\nIn generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.\n\n\n\n\nFunction argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the reference manual for full details.\nFor example, here’s a function to compute the entropy of a categorical distribution with simplex parameter theta.\nreal entropy(vector theta) {\n  return sum(theta .* log(theta));\n}\nAlthough theta must be a simplex, only the type vector is used.3\nUpper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.\n\n\n\nArray arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.\narray[] real baz(array[,] real x);\nThe notation [ ] is used for one-dimensional arrays (as in the return above), [ , ] for two-dimensional arrays, [ , , ] for three-dimensional arrays, and so on.\nFunctions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.\n\n\n\nA function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier data. The following is an example of a data-only function argument.\nreal foo(real y, data real mu) {\n  return -0.5 * (y - mu)^2;\n}\nThis qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.\nUse of the data qualifier must be consistent between the forward declaration and the definition of a functions.\nThis qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the ODE solver chapter for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the data prefix qualifier. This allows for compile-time type checking and increases overall program understandability.\n\n\n\n\nIn some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.\nfunctions {\n  void pretty_print_tri_lower(matrix x) {\n    if (rows(x) == 0) {\n      print(\"empty matrix\");\n      return;\n    }\n    print(\"rows=\", rows(x), \" cols=\", cols(x));\n    for (m in 1:rows(x)) {\n      for (n in 1:m) {\n        print(\"[\", m, \",\", n, \"]=\", x[m, n]);\n      }\n    }\n  }\n}\nThe special symbol void is used as the return type. This is not a type itself in that there are no values of type void; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.\nVoid functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ... code to set Sigma ...\n  pretty_print_tri_lower(Sigma);\n  // ...\n}\n\n\n\nFunctions whose names end in _lp are allowed to use sampling statements and target += statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.\nHere is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the reparameterization section for more information on efficient non-centered parameterizations\nfunctions {\n  vector center_lp(vector beta_raw, real mu, real sigma) {\n    beta_raw ~ std_normal();\n    sigma ~ cauchy(0, 5);\n    mu ~ cauchy(0, 2.5);\n    return sigma * beta_raw + mu;\n  }\n  // ...\n}\nparameters {\n  vector[K] beta_raw;\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // ...\n  beta = center_lp(beta_raw, mu_beta, sigma_beta);\n  // ...\n}\n\n\n\nFunctions whose names end in _jacobian can use the jacobian += statement. This can be used to implement a custom change of variables for arbitrary parameters.\nFor example, this function recreates the built-in &lt;upper=x&gt; transform on real numbers:\nreal my_upper_bound_jacobian(real x, real ub) {\n  jacobian += x;\n  return ub - exp(x);\n}\nIt can be used as a replacement for real&lt;lower=ub&gt; as follows:\nfunctions {\n  // my_upper_bound_jacobian as above\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  b ~ lognormal(0, 1);\n  // ...\n}\n\n\n\nA user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in _rng. Giving a function a name that ends in _rng allows it to access built-in functions and user-defined functions that end in _rng, which includes all the built-in PRNG functions. Only functions ending in _rng are able access the built-in PRNG functions. The use of functions ending in _rng must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in _rng.\nFor example, the following function generates an \\(N \\times K\\) data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.\nmatrix predictors_rng(int N, int K) {\n  matrix[N, K] x;\n  for (n in 1:N) {\n    x[n, 1] = 1.0;  // intercept\n    for (k in 2:K) {\n      x[n, k] = normal_rng(0, 1);\n    }\n  }\n  return x;\n}\nThe following function defines a simulator for regression outcomes based on a data matrix x, coefficients beta, and noise scale sigma.\nvector regression_rng(vector beta, matrix x, real sigma) {\n  vector[rows(x)] y;\n  vector[rows(x)] mu;\n  mu = x * beta;\n  for (n in 1:rows(x)) {\n    y[n] = normal_rng(mu[n], sigma);\n  }\n  return y;\n}\nThese might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n  // ...\n}\ngenerated quantities {\n  matrix[N_sim, K] x_sim;\n  vector[N_sim] y_sim;\n  x_sim = predictors_rng(N_sim, K);\n  y_sim = regression_rng(beta, x_sim, sigma);\n}\nA more sophisticated simulation might fit a multivariate normal to the predictors x and use the resulting parameters to generate multivariate normal draws for x_sim.\n\n\n\nProbability functions are distinguished in Stan by names ending in _lpdf for density functions and _lpmf for mass functions; in both cases, they must have real return types.\nSuppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.\nfunctions {\n  real unit_normal_lpdf(real y) {\n    return normal_lpdf(y | 0, 1);\n  }\n}\n// ...\nmodel {\n  alpha ~ unit_normal();\n  beta ~ unit_normal();\n  // ...\n}\nThe ability to use the unit_normal function as a density is keyed off its name ending in _lpdf (names ending in _lpmf for probability mass functions work the same way).\nIn general, if foo_lpdf is defined to consume \\(N + 1\\) arguments, then\ny ~ foo(theta1, ..., thetaN);\ncan be used as shorthand for\ntarget += foo_lpdf(y | theta1, ..., thetaN);\nAs with the built-in functions, the suffix _lpdf is dropped and the first argument moves to the left of the tilde symbol (~) in the distribution statement.\nFunctions ending in _lpmf (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (_lpdf) must be continuous (not an integer or integer array), whereas the first argument of a mass function (_lpmf) must be discrete (integer or integer array).\n\n\n\nAs described in the reference manual function overloading is permitted in Stan, beginning in version 2.29.\nThis means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.\n\n\nOverloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.\nFor example, consider a function triple which has the following three signatures\nreal triple(real x);\ncomplex triple(complex x);\narray[] real triple(array[] real);\nOne should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect\n// The function does what it says\ntriple(3.0) == 9.0\n// It is defined reasonably for different types\ntriple(to_complex(3.0)) == to_complex(triple(3.0))\n// A container version of this function works by element\ntriple({3.0, 4.0})[0] == triple({3.0, 4.0}[0])\nNote that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.\n\n\n\nStan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.\nConsider the following function signatures\nreal foo(int a, real b);\nreal foo(real a, real b);\nGiven these, the function call foo(1.5, 2.5) is unambiguous - it must resolve to the second signature. But, the function call foo(1, 1.5) could be valid for either under Stan’s promotion rules, which allow integers to be promoted to real numbers.\nTo resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call foo(1, 1.5) would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).\nFurthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures\nreal bar(int x, real y);\nreal bar(real x, int y);\nThese signatures do not have a unique minimum number of promotions for the call bar(1, 2). Both signatures require one int to real promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.\nPromotion from integers to complex numbers is considered to be two separate promotions, first from int to real, then from real to complex. This means that integer arguments will “prefer” a signature with real types over complex types.\nFor example, consider the function signatures\nreal pop(real x);\nreal pop(complex x);\nStan will select the first signature when pop is called with an integer argument such as pop(0).\n\n\n\n\nFunctions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.\nFor example, here’s some documentation for the prediction matrix generator.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows corresponding to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\nThe comment begins with /**, ends with */, and has an asterisk (*) on each line. It uses @param followed by the argument’s identifier to document a function argument. The tag @return is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with /* and ending with */ to the Stan parser.\nFor functions that raise exceptions, exceptions can be documented using @throws.4\nFor example,\n /** ...\n * @param theta\n * @throws If any of the entries of theta is negative.\n */\nreal entropy(vector theta) {\n  // ...\n}\nUsually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.\n\n\n\nFunctions may have a void or non-void return type and they may or may not have one of the special suffixes, _lpdf, _lpmf, _lp, or _rng.\n\n\nOnly functions declared to return void may be used as statements. These are also the only functions that use return statements with no arguments.\nOnly functions declared to return non-void values may be used as expressions. These functions require return statements with arguments of a type that matches the declared return type.\n\n\n\nOnly functions ending in _lpmf or _lpdf and with return type real may be used as probability functions in distribution statements.\nOnly functions ending in _lp may access the log probability accumulator through distribution statements or target += statements. Such functions may only be used in the transformed parameters or model blocks.\nOnly functions ending in _rng may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in _rng.\n\n\n\n\nStan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, \\(A^n\\), which is defined for a square matrix \\(A\\) and positive integer \\(n\\) by \\[\nA^n\n=\n\\begin{cases}\n\\textrm{I} & \\quad\\text{if } n = 0, \\text{ and} \\\\\nA \\, A^{n-1} & \\quad\\text{if } n &gt; 0.\n\\end{cases}\n\\]\nwhere \\(\\textrm{I}\\) is the identity matrix. This definition can be directly translated to a recursive function definition.\nmatrix matrix_pow(matrix a, int n) {\n  if (n == 0) {\n    return diag_matrix(rep_vector(1, rows(a)));\n  } else {\n    return a *  matrix_pow(a, n - 1);\n  }\n}\nIt would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.\nelse if (n == 1) {\n  return a;\n}\n\n\n\n\n\nTo generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then run through the inverse cumulative distribution function, \\(y = \\textrm{logit}(u)\\). If this were not already built in as logistic_rng(0, 1), it could be coded in Stan directly as\nreal standard_logistic_rng() {\n  real u = uniform_rng(0, 1);\n  real y = logit(u);\n  return y;\n}\nFollowing the same pattern, a standard normal RNG could be coded as\nreal standard_normal_rng() {\n  real u = uniform_rng(0, 1);\n  real y = inv_Phi(u);\n  return y;\n}\nthat is, \\(y = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}\\) is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function inv_Phi.\nIn order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate \\(\\textsf{normal}(\\mu, \\sigma)\\) variates, it is enough to generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then convert it to a standard normal variate, \\(z = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}(\\cdot)\\) is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, \\(y = \\mu +\n\\sigma \\times z\\). In code,\nreal my_normal_rng(real mu, real sigma) {\n  real u = uniform_rng(0, 1);\n  real z = inv_Phi(u);\n  real y = mu + sigma * z;\n  return y;\n}\nA robust version of this function would test that the arguments are finite and that sigma is non-negative, e.g.,\n  if (is_nan(mu) || is_inf(mu)) {\n    reject(\"my_normal_rng: mu must be finite; \",\n           \"found mu = \", mu);\n  }\n  if (is_nan(sigma) || is_inf(sigma) || sigma &lt; 0) {\n    reject(\"my_normal_rng: sigma must be finite and non-negative; \",\n           \"found sigma = \", sigma);\n  }\n\n\n\nOften truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.\n\n\nFor example, the following code generates a \\(\\textsf{Weibull}(\\alpha, \\sigma)\\) variate truncated below at a time \\(t\\),5\nreal weibull_lb_rng(real alpha, real sigma, real t) {\n  real p = weibull_cdf(t | alpha, sigma);   // cdf for lb\n  real u = uniform_rng(p, 1);               // unif in bounds\n  real y = sigma * (-log1m(u))^inv(alpha);  // inverse cdf\n  return y;\n}\n\n\n\nIf there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a \\(\\textsf{normal}(\\mu, \\sigma)\\) truncated to a region \\((a, b)\\), the following code suffices,\nreal normal_lub_rng(real mu, real sigma, real lb, real ub) {\n  real p_lb = normal_cdf(lb | mu, sigma);\n  real p_ub = normal_cdf(ub | mu, sigma);\n  real u = uniform_rng(p_lb, p_ub);\n  real y = mu + sigma * inv_Phi(u);\n  return y;\n}\nTo make this more robust, all variables should be tested for finiteness, sigma should be tested for positiveness, and lb and ub should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement\n  return mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),\n                                          normal_cdf(ub | mu, sigma)));\nfor readability. The names like p indicate probabilities, and p_lb and p_ub indicate the probabilities of the bounds. The variable u is clearly named as a uniform variate, and y is used to denote the variate being generated itself.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Using the Stan Compiler"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/using-stanc.html#stanc-pretty-printing",
-    "href": "stan-users-guide/using-stanc.html#stanc-pretty-printing",
-    "title": "Using the Stan Compiler",
+    "objectID": "stan-users-guide/user-functions.html#basic-functions.section",
+    "href": "stan-users-guide/user-functions.html#basic-functions.section",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "In addition to compiling Stan programs, stanc3 features several flags which can be used to format Stan programs and update them to the most recent Stan syntax by removing any deprecation features which can be automatically replaced.\nThese flags work for both .stan model files and .stanfunctions function files. They can be combined with --o to redirect the formatted output to a new file.\n\n\nInvoking stanc --auto-format &lt;model_file&gt; will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the Stan Program Style Guide. This means spacing, indentation, and line length are all regularized. Some deprecated features, like the use of # for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.\nBy default, this will try to split lines at or before column 78. This number can be changed using --max-line-length.\n\n\n\nIn addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of if statements and for and while loops.\nThis can be done by using stanc --auto-format --canonicalize=... where ... is a comma-separated list of options. Currently these options are:\n\ndeprecations\nRemoves deprecated syntax such as replacing deprecated functions with their drop-in replacements.\nparentheses\nRemoves unnecessary extra parentheses, such as converting y = ((x-1)) to y = x - 1\nbraces\nPlaces braces around all blocks. For example, the following statement\nif (cond)\n  //result\nwill be formatted as\nif (cond) {\n  //result\n}\nand similarly for both kinds of loops containing a single statement.\nincludes\nThis will pretty-print code from other files included with #include as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same #include directives as the input program.\n\nInvoking stanc --print-canonical &lt;model_file&gt; is synonymous with running stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes\n\n\n\nThe formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:\n\nOddly placed comments\nIf your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string ^^^: to indicate they originally appeared higher in the program.\nWe hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.\nFailure to recreate strange #include structure\nPrinting without include inlining (--canonicalize=includes) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.",
+    "text": "Here’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.\nfunctions {\n  real relative_diff(real x, real y) {\n    real abs_diff;\n    real avg_scale;\n    abs_diff = abs(x - y);\n    avg_scale = (abs(x) + abs(y)) / 2;\n    return abs_diff / avg_scale;\n  }\n}\n// ...\ngenerated quantities {\n  real rdiff;\n  rdiff = relative_diff(alpha, beta);\n}\nThe function is named relative_diff, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.\n\n\nAll functions are defined in their own block, which is labeled functions and must appear before all other program blocks. The user-defined functions block is optional.\n\n\n\nThe body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.\n\n\n\nReturn statements, such as the one on the last line of the definition of relative_diff above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.\nIf an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a fatal_error statement.\n\n\nRejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).\nThe most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.\nreal dbl_sqrt(real x) {\n  if (!(x &gt;= 0)) {\n    reject(\"dbl_sqrt(x): x must be positive; found x = \", x);\n  }\n  return 2 * sqrt(x);\n}\nThe negation of the positive test is important, because it also catches the case where x is a not-a-number value. If the condition had been coded as (x &lt; 0) it would not catch the not-a-number case, though it could be written as (x &lt; 0 || is_nan(x)). The positive infinite case is allowed through, but could also be checked with the is_inf(x) function. The square root function does not itself reject, but some downstream consumer of dbl_sqrt(-2) would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.\nThe effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.2\nIn generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.\n\n\n\n\nFunction argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the reference manual for full details.\nFor example, here’s a function to compute the entropy of a categorical distribution with simplex parameter theta.\nreal entropy(vector theta) {\n  return sum(theta .* log(theta));\n}\nAlthough theta must be a simplex, only the type vector is used.3\nUpper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.\n\n\n\nArray arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.\narray[] real baz(array[,] real x);\nThe notation [ ] is used for one-dimensional arrays (as in the return above), [ , ] for two-dimensional arrays, [ , , ] for three-dimensional arrays, and so on.\nFunctions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.\n\n\n\nA function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier data. The following is an example of a data-only function argument.\nreal foo(real y, data real mu) {\n  return -0.5 * (y - mu)^2;\n}\nThis qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.\nUse of the data qualifier must be consistent between the forward declaration and the definition of a functions.\nThis qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the ODE solver chapter for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the data prefix qualifier. This allows for compile-time type checking and increases overall program understandability.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Using the Stan Compiler"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/using-stanc.html#optimization",
-    "href": "stan-users-guide/using-stanc.html#optimization",
-    "title": "Using the Stan Compiler",
+    "objectID": "stan-users-guide/user-functions.html#functions-as-statements",
+    "href": "stan-users-guide/user-functions.html#functions-as-statements",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "The stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.\nThis section introduces the available optimization options and describes their effect.\nTo print out a representation of the optimized Stan program, use the stanc3 command-line flag --debug-optimized-mir-pretty. To print an analogous representation of the Stan program prior to optimization, use the flag --debug-transformed-mir-pretty.\n\n\nTo turn optimizations on, the user specifies the desired optimization level. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.\nOptimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:\n\nO0 No optimizations are applied.\nO1 Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.\nOexperimental All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.\n\nO0 is the default setting.\nThe levels include these optimizations:\n\nO0 includes no optimizations.\nO1 includes:\n\nDead code elimination\nCopy propagation\nConstant propagation\nPartial evaluation\nFunction inlining\nMatrix memory layout optimization\n\nOexperimental includes optimizations specified by O1 and also:\n\nAutomatic-differentiation level optimization\nOne step loop unrolling\nExpression propagation\nLazy code motion\nStatic loop unrolling\n\n\nIn addition, Oexperimental will apply more repetitions of the optimizations, which may increase compile times.\n\n\n\n\n\nDead code is code that does not affect the behavior of the program. Code is not dead if it affects target, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.\nExample Stan program:\nmodel {\n  int i;\n  i = 5;\n  for (j in 1:10);\n  if (0) {\n    print(\"Dead code\");\n  } else {\n    print(\"Hi!\");\n  }\n}\nCompiler representation of program before dead code elimination (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  int i = 5;\n  for(j in 1:10) {\n    ;\n  }\n  if(0) {\n    FnPrint__(\"Dead code\");\n  } else {\n    FnPrint__(\"Hi!\");\n  }\n}\nCompiler representation of program after dead code elimination (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i;\n  FnPrint__(\"Hi!\");\n}\n\n\n\nConstant propagation replaces uses of a variable which is known to have a constant value C with that constant C. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).\nExample Stan program:\ntransformed data {\n  int n = 100;\n  int a[n];\n  for (i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before constant propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, n] a;\n  for(i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after constant propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, 100] a;\n  for(i in 1:100) {\n    a[i] = i;\n  }\n}\n\n\n\nCopy propagation is similar to expression propagation, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.\nExample Stan program:\nmodel {\n  int i = 1;\n  int j = i;\n  int k = i + j;\n}\nCompiler representation of program before copy propagation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n    int i = 1;\n    int j = i;\n    int k = (i + j);\n}\nCompiler representation of program after copy propagation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i = 1;\n  int j = i;\n  int k = (i + i);\n}\n\n\n\nPartial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.\nExample Stan program:\nmodel {\n  real a = 1 + 1;\n  real b = log(1 - a);\n  real c = a + b * 5;\n}\nCompiler representation of program before partial evaluation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real a = (1 + 1);\n  real b = log((1 - a));\n  real c = (a + (b * 5));\n}\nCompiler representation of program after partial evaluation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  real a = 2;\n  real b = log1m(a);\n  real c = fma(b, 5, a);\n}\n\n\n\nFunction inlining replaces each function call to each user-defined function f with the body of f. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return x + y;\n  }\n}\ntransformed data {\n  int a = 2;\n  int b = incr(a);\n}\nCompiler representation of program before function inlining (simplified from the output of --debug-transformed-mir-pretty):\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return (x + y);\n  }\n}\n\nprepare_data {\n  data int a = 2;\n  data int b = incr(a);\n}\nCompiler representation of program after function inlining (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int a;\n  a = 2;\n  data int b;\n  data int inline_sym1__;\n  data int inline_sym3__;\n  inline_sym3__ = 0;\n  for(inline_sym4__ in 1:1) {\n    int inline_sym2__;\n    inline_sym2__ = 1;\n    inline_sym3__ = 1;\n    inline_sym1__ = (a + inline_sym2__);\n    break;\n  }\n  b = inline_sym1__;\n}\nIn this code, the for loop and break is used to simulate the behavior of a return statement. The value to be returned is held in inline_sym1__. The flag variable inline_sym3__ indicates whether a return has occurred and is necessary to handle return statements nested inside loops within the function body.\n\n\n\nMatrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.\nThe first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.\nThe second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.\nMore general reading on AoS vs SoA can be found on Wikipedia\nThis optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:\n\nAny Stan Math Library functions the matrix is passed to must be able to support it.\nThe matrix should not be accessed/assigned elementwise in a loop.\n\nThe debug flag --debug-mem-patterns will list each variable and whether it is using the AoS representation or the SoA representation.\n\n\n\n\n\n\nStan variables can have two auto-differentiation (AD) levels: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.\nExample Stan program:\ndata {\n  real y;\n}\nmodel {\n  real x = y + 1;\n}\nCompiler representation of program before AD-level optimization (simplified from the output of --debug-transformed-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  real x = (y + 1);\n}\nCompiler representation of program after AD-level optimization (simplified from the output of --debug-optimized-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  data real x = (y + 1);\n}\n\n\n\nOne step loop unrolling is similar to static loop unrolling. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.\nExample Stan program:\ndata {\n  int n;\n}\ntransformed data {\n  int x = 0;\n  for (i in 1:n) {\n    x += i;\n  }\n}\nCompiler representation of program before one step static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  data int x = 0;\n  for(i in 1:n) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after one step static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  int x = 0;\n  if((n &gt;= 1)) {\n    x = (x + 1);\n    for(i in (1 + 1):n) {\n      x = (x + i);\n    }\n  }\n}\n\n\n\nConstant propagation replaces the uses of a variable which is known to have a constant value E with that constant E. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by lazy code motion to avoid unnecessarily recomputing expressions.\nExample Stan program:\ndata {\n  int m;\n}\ntransformed data {\n  int n = m+1;\n  int a[n];\n  for (i in 1:n-1) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before expression propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, n] a;\n  for(i in 1:(n - 1)) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after expression propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, (m + 1)] a;\n  for(i in 1:((m + 1) - 1)) {\n    a[i] = i;\n  }\n}\n\n\n\nLazy code motion rearranges the statements and expressions in a program with the goals of:\n\nAvoiding computing expressions more than once, and\nComputing expressions as late as possible (to minimize the strain on the working memory set).\n\nTo accomplish these goals, lazy code motion will perform optimizations such as:\n\nMoving a repeatedly calculated expression to its own variable (also referred to as common-subexpression elimination)\nMoving an expression outside of a loop if it does not need to be in the loop (also referred to as loop-invariant code motion)\n\nLazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.\nAs currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.\nExample Stan program:\nmodel {\n  real x;\n  real y;\n  real z;\n\n  for (i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program before lazy code motion (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real x;\n  real y;\n  real z;\n  for(i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program after lazy code motion (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  data real lcm_sym4__;\n  data real lcm_sym3__;\n  real x;\n  real y;\n  lcm_sym4__ = sqrt(10);\n  real z;\n  for(i in 1:10) {\n    x = lcm_sym4__;\n    y = sqrt(i);\n  }\n  z = lcm_sym4__;\n}\n\n\n\nStatic loop unrolling takes a loop with a predictable number of iterations X and replaces it by writing out the loop body X times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\ntransformed data {\n  int x = 0;\n  for (i in 1:4) {\n    x += i;\n  }\n}\nCompiler representation of program before static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int x = 0;\n  for(i in 1:4) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int x;\n  x = 0;\n  x = (x + 1);\n  x = (x + 2);\n  x = (x + 3);\n  x = (x + 4);\n}",
+    "text": "In some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.\nfunctions {\n  void pretty_print_tri_lower(matrix x) {\n    if (rows(x) == 0) {\n      print(\"empty matrix\");\n      return;\n    }\n    print(\"rows=\", rows(x), \" cols=\", cols(x));\n    for (m in 1:rows(x)) {\n      for (n in 1:m) {\n        print(\"[\", m, \",\", n, \"]=\", x[m, n]);\n      }\n    }\n  }\n}\nThe special symbol void is used as the return type. This is not a type itself in that there are no values of type void; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.\nVoid functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ... code to set Sigma ...\n  pretty_print_tri_lower(Sigma);\n  // ...\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Using the Stan Compiler"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/truncation-censoring.html",
-    "href": "stan-users-guide/truncation-censoring.html",
-    "title": "Truncated or Censored Data",
+    "objectID": "stan-users-guide/user-functions.html#functions-accessing-the-log-probability-accumulator",
+    "href": "stan-users-guide/user-functions.html#functions-accessing-the-log-probability-accumulator",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Data in which measurements have been truncated or censored can be coded in Stan following their respective probability models.\n\n\nTruncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.\n\n\n\nTruncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.\nTruncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are \\(y_n\\) with an upper truncation point of \\(U = 300\\) so that \\(y_n &lt; 300\\). In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  real U;\n  array[N] real&lt;upper=U&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma) T[ , U];\n}\nThe model declares an upper bound U as data and constrains the data for y to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.\nSee the Stan Reference Manual’s Statements chapter for how to use truncated distributions with the log probability increment statements (target += ...).\nThis model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.\n\n\nIf the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to \\(-\\infty\\). For instance, if variate y is sampled with the statement.\ny ~ normal(mu, sigma) T[L, U];\nthen if any value inside y is less than the value of L or greater than the value of U, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.\nTo avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if y is a parameter in the above model, the declaration should constrain it to fall between the values of L and U.\nparameters {\n  array[N] real&lt;lower=L, upper=U&gt; y;\n  // ...\n}\nIf in the above model, L or U is a parameter and y is data, then L and U must be appropriately constrained so that all data are in range and the value of L is less than that of U (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.\nparameters {\n  real&lt;upper=min(y)&gt; L;           // L &lt; y[n]\n  real&lt;lower=fmax(L, max(y))&gt; U;  // L &lt; U; y[n] &lt; U\nFor pairs of real numbers, the function fmax is used rather than max.\n\n\n\nIf the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real y;\n}\nparameters {\n  real&lt;upper=min(y)&gt; L;\n  real&lt;lower=max(y)&gt; U;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ // ...\n  U ~ // ...\n  y ~ normal(mu, sigma) T[L, U];\n}\nHere there is a lower truncation point L which is declared to be less than or equal to the minimum value of y. The upper truncation point U is declared to be larger than the maximum value of y. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With N declared as type int&lt;lower=1&gt;, there must be at least one data point. The constraint that L is less than U is enforced indirectly, based on the non-empty data.\nThe ellipses where the priors for the bounds L and U should go should be filled in with a an informative prior in order for this model to not concentrate L strongly around min(y) and U strongly around max(y).\n\n\n\n\nCensoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.\n\n\nOne way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  array[N_cens] real&lt;lower=U&gt; y_cens;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_cens ~ normal(mu, sigma);\n}\nBecause the censored data array y_cens is declared to be a parameter, it will be sampled along with the location and scale parameters mu and sigma. Because the censored data array y_cens is declared to have values of type real&lt;lower=U&gt;, all imputed values for censored data will be greater than U. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.\n\n\n\nAlthough it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of \\[\\begin{align*}\n\\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\int_U^{\\infty} \\textsf{normal}\\left(y_{\\mathrm{cens},m} \\mid \\mu,\\sigma \\right) \\,\\textsf{d}y_{\\mathrm{cens},m} \\\\\n  &= 1 - \\Phi\\left(\\frac{U - \\mu}{\\sigma}\\right),\n\\end{align*}\\]\nwhere \\(\\Phi()\\) is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that \\(y_{\\mathrm{cens},m}&gt;U\\). With \\(M\\) censored observations, the likelihood on the log scale is \\[\\begin{align*}\n\\log \\prod_{m=1}^M \\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\log \\left( 1 - \\Phi\\left(\\left(\\frac{U - \\mu}{\\sigma}\\right)\\right)^{M}\\right) \\\\\n  &= M \\times \\texttt{normal}\\mathtt{\\_}\\texttt{lccdf}\\left(U \\mid \\mu, \\sigma \\right),\n\\end{align*}\\]\nwhere normal_lccdf is the log of complementary CDF (Stan provides &lt;distr&gt;_lccdf for each distribution implemented in Stan).\nThe following right-censored model assumes that the censoring point is known, so it is declared as data.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lccdf(U | mu, sigma);\n}\nFor the observed values in y_obs, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.\nFor the left-censored data the CDF (normal_lcdf) has to be used instead of complementary CDF. If the censoring point variable (L) is unknown, its declaration should be moved from the data to the parameters block.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real&lt;upper=min(y_obs)&gt; L;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ normal(mu, sigma);\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lcdf(L | mu, sigma);\n}",
+    "text": "Functions whose names end in _lp are allowed to use sampling statements and target += statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.\nHere is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the reparameterization section for more information on efficient non-centered parameterizations\nfunctions {\n  vector center_lp(vector beta_raw, real mu, real sigma) {\n    beta_raw ~ std_normal();\n    sigma ~ cauchy(0, 5);\n    mu ~ cauchy(0, 2.5);\n    return sigma * beta_raw + mu;\n  }\n  // ...\n}\nparameters {\n  vector[K] beta_raw;\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // ...\n  beta = center_lp(beta_raw, mu_beta, sigma_beta);\n  // ...\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Truncated or Censored Data"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/truncation-censoring.html#truncation.section",
-    "href": "stan-users-guide/truncation-censoring.html#truncation.section",
-    "title": "Truncated or Censored Data",
+    "objectID": "stan-users-guide/user-functions.html#functions-implementing-change-of-variable-adjustments",
+    "href": "stan-users-guide/user-functions.html#functions-implementing-change-of-variable-adjustments",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Truncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.",
+    "text": "Functions whose names end in _jacobian can use the jacobian += statement. This can be used to implement a custom change of variables for arbitrary parameters.\nFor example, this function recreates the built-in &lt;upper=x&gt; transform on real numbers:\nreal my_upper_bound_jacobian(real x, real ub) {\n  jacobian += x;\n  return ub - exp(x);\n}\nIt can be used as a replacement for real&lt;lower=ub&gt; as follows:\nfunctions {\n  // my_upper_bound_jacobian as above\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  b ~ lognormal(0, 1);\n  // ...\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Truncated or Censored Data"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/truncation-censoring.html#truncated-data.section",
-    "href": "stan-users-guide/truncation-censoring.html#truncated-data.section",
-    "title": "Truncated or Censored Data",
+    "objectID": "stan-users-guide/user-functions.html#functions-acting-as-random-number-generators",
+    "href": "stan-users-guide/user-functions.html#functions-acting-as-random-number-generators",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Truncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.\nTruncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are \\(y_n\\) with an upper truncation point of \\(U = 300\\) so that \\(y_n &lt; 300\\). In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  real U;\n  array[N] real&lt;upper=U&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma) T[ , U];\n}\nThe model declares an upper bound U as data and constrains the data for y to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.\nSee the Stan Reference Manual’s Statements chapter for how to use truncated distributions with the log probability increment statements (target += ...).\nThis model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.\n\n\nIf the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to \\(-\\infty\\). For instance, if variate y is sampled with the statement.\ny ~ normal(mu, sigma) T[L, U];\nthen if any value inside y is less than the value of L or greater than the value of U, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.\nTo avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if y is a parameter in the above model, the declaration should constrain it to fall between the values of L and U.\nparameters {\n  array[N] real&lt;lower=L, upper=U&gt; y;\n  // ...\n}\nIf in the above model, L or U is a parameter and y is data, then L and U must be appropriately constrained so that all data are in range and the value of L is less than that of U (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.\nparameters {\n  real&lt;upper=min(y)&gt; L;           // L &lt; y[n]\n  real&lt;lower=fmax(L, max(y))&gt; U;  // L &lt; U; y[n] &lt; U\nFor pairs of real numbers, the function fmax is used rather than max.\n\n\n\nIf the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real y;\n}\nparameters {\n  real&lt;upper=min(y)&gt; L;\n  real&lt;lower=max(y)&gt; U;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ // ...\n  U ~ // ...\n  y ~ normal(mu, sigma) T[L, U];\n}\nHere there is a lower truncation point L which is declared to be less than or equal to the minimum value of y. The upper truncation point U is declared to be larger than the maximum value of y. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With N declared as type int&lt;lower=1&gt;, there must be at least one data point. The constraint that L is less than U is enforced indirectly, based on the non-empty data.\nThe ellipses where the priors for the bounds L and U should go should be filled in with a an informative prior in order for this model to not concentrate L strongly around min(y) and U strongly around max(y).",
+    "text": "A user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in _rng. Giving a function a name that ends in _rng allows it to access built-in functions and user-defined functions that end in _rng, which includes all the built-in PRNG functions. Only functions ending in _rng are able access the built-in PRNG functions. The use of functions ending in _rng must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in _rng.\nFor example, the following function generates an \\(N \\times K\\) data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.\nmatrix predictors_rng(int N, int K) {\n  matrix[N, K] x;\n  for (n in 1:N) {\n    x[n, 1] = 1.0;  // intercept\n    for (k in 2:K) {\n      x[n, k] = normal_rng(0, 1);\n    }\n  }\n  return x;\n}\nThe following function defines a simulator for regression outcomes based on a data matrix x, coefficients beta, and noise scale sigma.\nvector regression_rng(vector beta, matrix x, real sigma) {\n  vector[rows(x)] y;\n  vector[rows(x)] mu;\n  mu = x * beta;\n  for (n in 1:rows(x)) {\n    y[n] = normal_rng(mu[n], sigma);\n  }\n  return y;\n}\nThese might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n  // ...\n}\ngenerated quantities {\n  matrix[N_sim, K] x_sim;\n  vector[N_sim] y_sim;\n  x_sim = predictors_rng(N_sim, K);\n  y_sim = regression_rng(beta, x_sim, sigma);\n}\nA more sophisticated simulation might fit a multivariate normal to the predictors x and use the resulting parameters to generate multivariate normal draws for x_sim.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Truncated or Censored Data"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/truncation-censoring.html#censored.section",
-    "href": "stan-users-guide/truncation-censoring.html#censored.section",
-    "title": "Truncated or Censored Data",
+    "objectID": "stan-users-guide/user-functions.html#user-defined-probability-functions",
+    "href": "stan-users-guide/user-functions.html#user-defined-probability-functions",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Censoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.\n\n\nOne way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  array[N_cens] real&lt;lower=U&gt; y_cens;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_cens ~ normal(mu, sigma);\n}\nBecause the censored data array y_cens is declared to be a parameter, it will be sampled along with the location and scale parameters mu and sigma. Because the censored data array y_cens is declared to have values of type real&lt;lower=U&gt;, all imputed values for censored data will be greater than U. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.\n\n\n\nAlthough it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of \\[\\begin{align*}\n\\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\int_U^{\\infty} \\textsf{normal}\\left(y_{\\mathrm{cens},m} \\mid \\mu,\\sigma \\right) \\,\\textsf{d}y_{\\mathrm{cens},m} \\\\\n  &= 1 - \\Phi\\left(\\frac{U - \\mu}{\\sigma}\\right),\n\\end{align*}\\]\nwhere \\(\\Phi()\\) is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that \\(y_{\\mathrm{cens},m}&gt;U\\). With \\(M\\) censored observations, the likelihood on the log scale is \\[\\begin{align*}\n\\log \\prod_{m=1}^M \\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\log \\left( 1 - \\Phi\\left(\\left(\\frac{U - \\mu}{\\sigma}\\right)\\right)^{M}\\right) \\\\\n  &= M \\times \\texttt{normal}\\mathtt{\\_}\\texttt{lccdf}\\left(U \\mid \\mu, \\sigma \\right),\n\\end{align*}\\]\nwhere normal_lccdf is the log of complementary CDF (Stan provides &lt;distr&gt;_lccdf for each distribution implemented in Stan).\nThe following right-censored model assumes that the censoring point is known, so it is declared as data.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lccdf(U | mu, sigma);\n}\nFor the observed values in y_obs, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.\nFor the left-censored data the CDF (normal_lcdf) has to be used instead of complementary CDF. If the censoring point variable (L) is unknown, its declaration should be moved from the data to the parameters block.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real&lt;upper=min(y_obs)&gt; L;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ normal(mu, sigma);\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lcdf(L | mu, sigma);\n}",
+    "text": "Probability functions are distinguished in Stan by names ending in _lpdf for density functions and _lpmf for mass functions; in both cases, they must have real return types.\nSuppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.\nfunctions {\n  real unit_normal_lpdf(real y) {\n    return normal_lpdf(y | 0, 1);\n  }\n}\n// ...\nmodel {\n  alpha ~ unit_normal();\n  beta ~ unit_normal();\n  // ...\n}\nThe ability to use the unit_normal function as a density is keyed off its name ending in _lpdf (names ending in _lpmf for probability mass functions work the same way).\nIn general, if foo_lpdf is defined to consume \\(N + 1\\) arguments, then\ny ~ foo(theta1, ..., thetaN);\ncan be used as shorthand for\ntarget += foo_lpdf(y | theta1, ..., thetaN);\nAs with the built-in functions, the suffix _lpdf is dropped and the first argument moves to the left of the tilde symbol (~) in the distribution statement.\nFunctions ending in _lpmf (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (_lpdf) must be continuous (not an integer or integer array), whereas the first argument of a mass function (_lpmf) must be discrete (integer or integer array).",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Truncated or Censored Data"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html",
-    "href": "stan-users-guide/survival.html",
-    "title": "Survival Models",
+    "objectID": "stan-users-guide/user-functions.html#overloading-functions",
+    "href": "stan-users-guide/user-functions.html#overloading-functions",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Survival models apply to animals and plants as well as inanimate objects such as machine parts or electrical components. Survival models arise when there is an event of interest for a group of subjects, machine component, or other item that is\n\ncertain to occur after some amount of time,\nbut only measured for a fixed period of time, during which the event may not have occurred for all subjects.\n\nFor example, one might wish to estimate the the distribution of time to failure for solid state drives in a data center, but only measure drives for a two year period, after which some number will have failed and some will still be in service.\nSurvival models are often used comparatively, such as comparing time to death of patients diagnosed with stage one liver cancer under a new treatment and a standard treatment (pure controls are not allowed when there is an effective existing treatment for a serious condition). During a two year trial, some patients will die and others will survive.\nSurvival models may involve covariates, such as the factory at which a component is manufactured, the day on which it is manufactured, and the amount of usage it gets. A clinical trial might be adjusted for the sex and age of a cancer patient or the hospital at which treatment is received.\nSurvival models come in two main flavors, parametric and semi-parametric. In a parametric model, the survival time of a subject is modeled explicitly using a parametric probability distribution. There is a great deal of flexibility in how the parametric probability distribution is constructed. The sections below consider exponential and Weibull distributed survival times.\nRather than explicitly modeling a parametric survival probability, semi-parametric survival models instead model the relative effect on survival of covariates. The final sections of this chapter consider the proportional hazards survival model.\n\n\nThe exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if \\(T \\sim \\textrm{exponential}(\\lambda)\\) for some rate \\(\\lambda &gt; 0,\\) then \\[\\begin{equation*}\n\\Pr[T &gt; t] = \\Pr[T &gt; t + t' \\mid T &gt; t'].\n\\end{equation*}\\] If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.\nThe exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable \\(T\\) is \\[\\begin{equation*}\n\\mathbb{E}[T \\mid \\lambda] = \\frac{1}{\\lambda}.\n\\end{equation*}\\] The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) \\(\\beta = 1 / \\lambda\\).\nThe data for a survival model consists of two components. First, there is a vector \\(t \\in (0, \\infty)^N\\) of \\(N\\) observed failure times. Second, there is a censoring time \\(t^{\\textrm{cens}}\\) such that failure times greater than \\(t^{\\textrm{cens}}\\) are not observed. The censoring time assumption imposes a constraint which requires \\(t_n &lt; t^{\\textrm{cens}}\\) for all \\(n \\in 1{:}N.\\) For the censored subjects, the only thing required in the model is their total count, \\(N^\\textrm{cens}\\) (their covariates are also required for models with covariates).\nThe model for the observed failure times is exponential, so that for \\(n \\in 1{:}N,\\) \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda).\n\\end{equation*}\\]\nThe model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of \\[\\begin{equation*}\n\\Pr[T &gt; t^{\\textrm{cens}}] = 1 - F_T(t^{\\textrm{cens}}),\n\\end{equation*}\\] where \\(F_T\\) is the cumulative distribution function (cdf) of survival time \\(T\\) (\\(F_X(x) = \\Pr[X \\leq x]\\) is standard notation for the cdf of a random variable \\(X\\)). The function \\(1 - F_T(t)\\) is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood \\[\\begin{eqnarray*}\np(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\prod_{n=1}^{N^{\\textrm{cens}}}\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)\n\\\\\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)^{N^{\\textrm{cens}}}.\n\\end{eqnarray*}\\]\nOn the log scale, that’s \\[\\begin{eqnarray*}\n\\log p(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\sum_{n=1}^N \\log \\textrm{exponential}(t_n \\mid \\lambda)\n\\\\\n& & { } + N^{\\textrm{cens}} \\cdot \\log \\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda).\n\\end{eqnarray*}\\]\nThe model can be completed with a standard lognormal prior on \\(\\lambda,\\) \\[\\begin{equation*}\n\\lambda \\sim \\textrm{lognormal}(0, 1),\n\\end{equation*}\\] which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed \\(\\textrm{lognormal}(0, 1)\\). In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.\n\n\nThe data for a simple survival analysis without covariates can be coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n}\nIn this program, N is the number of uncensored observations and t contains the times of the uncensored observations. There are a further N_cens items that are right censored at time t_cens. Right censoring means that if the time to failure is greater than\nt_cens, it is only observed that the part survived until time t_cens. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.\nThere is a single rate parameter, the inverse of which is the expected time to failure.\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nThe exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.\nmodel {\n  t ~ exponential(lambda);\n  target += N_cens * exponential_lccdf(t_cens | lambda);\n\n  lambda ~ lognormal(0, 1);\n}\nThe likelihood for rate lambda is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector t as a having an exponential distribution with rate lambda. This data model could have been written as\nfor (n in 1:N) {\n  t[n] ~ exponential(lambda);\n}\nThe log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate lambda. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,\nfor (n in 1:N)\n  target += exponential_lccdf(t_cens | lambda);\nto define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.\n\n\n\n\nThe Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.\nThe Weibull distribution is parameterized by a shape \\(\\alpha &gt; 0\\) and scale \\(\\sigma &gt; 0.\\) For an outcome \\(t \\geq 0\\), the Weibull distribution’s probability density function is \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid \\alpha, \\sigma)\n= \\frac{\\alpha}{\\sigma}\n  \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n  \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^{\\alpha}\\right).\n\\end{equation*}\\] In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter \\(\\beta &gt; 0\\) with probability density function \\[\\begin{equation*}\n\\textrm{exponential}(t \\mid \\beta) =\n\\beta\n\\cdot\n\\exp(-\\beta \\cdot t).\n\\end{equation*}\\] When \\(\\alpha = 1,\\) the Weibull distribution reduces to an exponential distribution, \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid 1, \\sigma)\n=\n\\textrm{exponential}\\!\\left(t \\,\\bigg|\\, \\frac{1}{\\sigma}\\right).\n\\end{equation*}\\] In other words, the Weibull is a continuous expansion of the exponential distribution.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then the expected survival time is \\[\\begin{equation*}\n\\mathbb{E}[T] = \\sigma \\cdot \\Gamma\\!\\left(1 + \\frac{1}{\\alpha}\\right),\n\\end{equation*}\\] where the \\(\\Gamma\\) function is the continuous completion of the factorial function (i.e., \\(\\Gamma(1 + n) = n!\\ \\) for \\(n \\in\n\\mathbb{N}\\)). As \\(\\alpha \\rightarrow 0\\) for a fixed \\(\\sigma\\) or as \\(\\sigma \\rightarrow \\infty\\) for a fixed \\(\\alpha\\), the expected survival time goes to infinity.\nThere are three regimes of the Weibull distribution.\n\n\\(\\alpha &lt; 1.\\) A subject is more likely to fail early. When \\(\\alpha\n&lt; 1,\\) the Weibull density approaches infinity as \\(t \\rightarrow 0.\\)\n\\(\\alpha = 1.\\) The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When \\(\\alpha = 1,\\) the Weibull distribution approaches \\(\\sigma\\) as \\(t\n\\rightarrow 0.\\)\n\\(\\alpha &gt; 1.\\) Subjects are less likely to fail early. When \\(\\alpha &lt; 1,\\) the Weibull density approaches zero as \\(t \\rightarrow 0.\\)\n\nWith \\(\\alpha \\leq 1,\\) the mode is zero (\\(t = 0\\)), whereas with \\(\\alpha &gt; 1,\\) the mode is nonzero (\\(t &gt; 0\\)).\n\n\nWith Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.\nparameters {\n  real&lt;lower=0&gt; beta;\n}\nmodel {\n  t ~ exponential(beta);\n  target += N_cens * exponential_lccdf(t_cens | beta);\n\n  beta ~ lognormal(0, 1);\n}\nThe Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (alpha) and scale (sigma) parameters.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  t ~ weibull(alpha, sigma);\n  target += N_cens * weibull_lccdf(t_cens | alpha, sigma);\n\n  alpha ~ lognormal(0, 1);\n  sigma ~ lognormal(0, 1);\n}\nAs usual, if more is known about expected survival times, alpha and sigma should be given more informative priors.\n\n\n\n\nSuppose that for each of \\(n \\in 1{:}N\\) items observed, both censored and uncensored, there is a covariate (row) vector \\(x_n \\in\n\\mathbb{R}^K.\\) For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.\nSurvival with covariates replaces what is essentially a simple regression with only an intercept \\(\\lambda\\) with a generalized linear model with a log link, where the rate for item \\(n\\) is \\[\\begin{equation*}\n\\lambda_n = \\exp(x_n \\cdot \\beta),\n\\end{equation*}\\] where \\(\\beta \\in \\mathbb{R}^K\\) is a \\(K\\)-vector of regression coefficients. Thus \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda_n).\n\\end{equation*}\\] The censored items have probability \\[\\begin{equation*}\n\\Pr[n\\textrm{-th censored}] =\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid x^{\\textrm{cens}}_n\n\\cdot \\beta).\n\\end{equation*}\\]\nThe covariates form an \\(N \\times K\\) data matrix, \\(x \\in\n\\mathbb{R}^{N \\times K}\\). An intercept can be introduced by adding a column of 1 values to \\(x\\).\nA Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (t and t_cens) corresponds to the ordering of items in the covariate matrices (x and x_cens).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n  int&lt;lower=0&gt; K;\n  matrix[N, K] x;\n  matrix[N_cens, K] x_cens;\n}\nparameters {\n  vector[K] gamma;\n}\nmodel {\n  gamma ~ normal(0, 2);\n\n  t ~ exponential(exp(x * gamma));\n  target += exponential_lccdf(t_cens | exp(x_cens * gamma));\n}\nBoth the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.\n\n\n\nSuppose \\(T\\) is a random variable representing a survival time, with a smooth cumulative distribution function \\[\\begin{equation*}\nF_T(t) = \\Pr[T \\leq t],\n\\end{equation*}\\] so that its probability density function is \\[\\begin{equation*}\np_T(t) = \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t).\n\\end{equation*}\\]\nThe survival function \\(S(t)\\) is the probability of surviving until at least time \\(t\\), which is just the complementary cumulative distribution function (ccdf) of the survival random variable \\(T\\), \\[\\begin{equation*}\nS(t) = 1 - F_T(t).\n\\end{equation*}\\] The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).\nThe hazard function \\(h(t)\\) is the instantaneous risk of not surviving past time \\(t\\) assuming survival until time \\(t\\), which is given by \\[\\begin{equation*}\nh(t) = \\frac{p_T(t)}{S(t)} = \\frac{p_T(t)}{1 - F_T(t)}.\n\\end{equation*}\\] The cumulative hazard function \\(H(t)\\) is defined to be the accumulated hazard over time, \\[\\begin{equation*}\nH(t) = \\int_0^t h(u) \\, \\textrm{d}u.\n\\end{equation*}\\]\nThe hazard function and survival function are related through the differential equation \\[\\begin{eqnarray*}\nh(t) & = & -\\frac{\\textrm{d}}{\\textrm{d}t} \\log S(t).\n\\\\[4pt]\n& = & -\\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} S(t)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} -(1 - F_T(t))\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} (F_T(t) - 1)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t)\n\\\\[4pt]\n& = & \\frac{p_T(t)}{S(t)}.\n\\end{eqnarray*}\\]\nIf \\(T \\sim \\textrm{exponential}(\\beta)\\) has an exponential distribution, then its hazard function is constant, \\[\\begin{eqnarray*}\nh(t \\mid \\beta)\n& = & \\frac{p_T(t \\mid \\beta)}{S(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\textrm{exponential}(t \\mid \\beta)}{1 - \\textrm{exponentialCCDF}(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {1 - (1 - \\exp(-\\beta \\cdot t))}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {\\exp(-\\beta \\cdot t)}\n\\\\[4pt]\n& = & \\beta.\n\\end{eqnarray*}\\] The exponential distribution is the only distribution of survival times with a constant hazard function.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then its hazard function is \\[\\begin{eqnarray*}\nh(t \\mid \\alpha, \\sigma)\n& = & \\frac{p_T(t \\mid \\alpha, \\sigma)}{S(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = & \\frac{\\textrm{Weibull}(t \\mid \\alpha, \\sigma)}{1 - \\textrm{WeibullCCDF}(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = &\n\\frac{\\frac{\\alpha}{\\sigma} \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n      \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma} \\right)^\\alpha\\right)}\n     {1 - \\left(1 -\n       \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^\\alpha\n     \\right)\\right)}      \n\\\\[4pt]\n& = & \\frac{\\alpha}{\\sigma}\n  \\cdot\n  \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}.\n\\end{eqnarray*}\\]\nIf \\(\\alpha = 1\\) the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when \\(\\alpha = 1\\)). When \\(\\alpha &gt; 1,\\) the hazard grows as time passes, whereas when \\(\\alpha &lt; 1,\\) it decreases as time passes.\n\n\n\nThe exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. Cox (1972) introduced a semi-parametric survival model specified directly in terms of a hazard function \\(h(t)\\) rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.\nLet \\(x_n \\in \\mathbb{R}^K\\) be a (row) vector of covariates for subject \\(n\\) so that the full covariate data matrix is \\(x \\in \\mathbb{R}^{N \\times\nK}\\). In Cox’s model, the hazard function for subject \\(n\\) is defined conditionally in terms of their covariates \\(x_n\\) and the parameter vector \\(\\gamma \\in \\mathbb{R}^K\\) as \\[\\begin{equation*}\nh(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n \\cdot \\gamma),\n\\end{equation*}\\] where \\(h_0(t)\\) is a shared baseline hazard function and \\(x_n \\cdot\n\\gamma = \\sum_{k=1}^K x_{n, k} \\cdot \\beta_k\\) is a row vector-vector product.\nIn the semi-parametric, proportional hazards model, the baseline hazard function \\(h_0(t)\\) is not modeled. This is why it is called “semi-parametric.” Only the factor \\(\\exp(x_n \\cdot \\gamma),\\) which determines how individual \\(n\\) varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”\nCox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function \\(h_0(t)\\) is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function1.\n\n\nCox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function \\(h_0(t)\\) is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.\nThe hazard function \\(h(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n\n\\cdot \\beta)\\) for subject \\(n\\) represents the instantaneous probability that subject \\(n\\) fails at time \\(t\\) given that it has survived until time \\(t.\\) The probability that subject \\(n\\) is the first to fail among \\(N\\) subjects is thus proportional to subject \\(n\\)’s hazard function, \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail at time } t]\n\\propto h(t \\mid x_n, \\beta).\n\\end{equation*}\\] Normalizing yields \\[\\begin{eqnarray*}\n\\Pr[n \\textrm{ first to fail at time } t]\n& = &  \\frac{h(t \\mid x_n, \\beta)}\n            {\\sum_{n' = 1}^N h(t \\mid x_{n'}, \\beta)}\n\\\\[4pt]\n& = &  \\frac{h_0(t) \\cdot \\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N h_0(t) \\cdot \\exp(x_{n'} \\cdot \\beta)}\n\\\\[4pt]\n& = &  \\frac{\\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{eqnarray*}\\]\nSuppose there are \\(N\\) subjects with strictly ordered survival times \\(t_1 &lt;\nt_2 &lt; \\cdots &lt; t_N\\) and covariate (row) vectors \\(x_1, \\ldots, x_N\\). Let \\(t^{\\textrm{cens}}\\) be the (right) censoring time and let \\(N^{\\textrm{obs}}\\) be the largest value of \\(n\\) such that \\(t_n \\leq\nt^{\\textrm{cens}}\\). This means \\(N^{\\textrm{obs}}\\) is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.\nWith failure times sorted in decreasing order, the partial likelihood for each observed subject \\(n \\in 1{:}N^{\\textrm{obs}}\\) can be expressed as \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail among } n, n + 1, \\ldots N]\n= \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.\nThe partial likelihood, defined in this form by Breslow (1975), is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). \\[\\begin{equation*}\n\\Pr[\\textrm{observed failures ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n= \\prod_{n = 1}^{N^{\\textrm{obs}}}\n  \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] On the log scale, \\[\\begin{eqnarray*}\n\\log \\Pr[\\textrm{obs.\\ fail ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n& = &\n\\sum_{n = 1}^{N^{\\textrm{obs}}}\n  \\log \\left(\n          \\frac{\\exp(x_n \\cdot \\beta)}\n               {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}\n       \\right)\n\\\\[4pt]\n& = & x_n \\cdot \\beta - \\log \\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)\n\\\\\n& = & x_n \\cdot \\beta - \\textrm{logSumExp}_{n' = n}^N \\ x_{n'} \\cdot \\beta,\n\\end{eqnarray*}\\] where \\[\\begin{equation*}\n\\textrm{logSumExp}_{n = a}^b \\ x_n\n= \\log \\sum_{n = a}^b \\exp(x_n)\n\\end{equation*}\\] is implemented so as to preserve numerical precision.\nThis likelihood follows the same approach to ranking as that developed by Plackett (1975) for estimating the probability of the order of the first few finishers in a horse race.\nA simple normal prior on the components of \\(\\beta\\) completes the model, \\[\\begin{equation*}\n\\beta \\sim \\textrm{normal}(0, 2).\n\\end{equation*}\\] This should be scaled based on knowledge of the predictors.\n\n\n\nTo simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.\ndata {\n  int&lt;lower=0&gt; K;          // num covariates\n\n  int&lt;lower=0&gt; N;          // num uncensored obs\n  vector[N] t;             // event time (non-strict decreasing)\n  matrix[N, K] x;          // covariates for uncensored obs\n\n  int N_c;                 // num censored obs\n  real&lt;lower=t[N]&gt; t_c;    // censoring time\n  matrix[N_c, K] x_c;      // covariates for censored obs\n}\nThe parameters are just the coefficients.\nparameters {\n  vector[K] beta;          // slopes (no intercept)\n}\nThe prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.\nmodel {\n  beta ~ normal(0, 2);\n  ...\nThe log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\nThese vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.\n  real log_denom = log_sum_exp(log_theta_c);\nThen, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.\n  for (n in 1:N) {\n    log_denom = log_sum_exp(log_denom, log_theta[n]);\n    target += log_theta[n] - log_denom;   // log likelihood\n  }\nThe running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.\ntransformed data {\n  for (n in 2:N) {\n    if (!(t[n] &lt; t[n - 1])) {\n      reject(\"times must be strictly decreasing, but found\"\n             \"!(t[\", n, \"] &lt; t[, \", (n - 1), \"])\");\n    }   \n  }\n}\n\n\n\nTechnically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all \\(N!\\) permutations for a set of \\(N\\) subjects with tied survival times is not tractable. As an alternative, Efron (1977) introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.\nIn the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.\nfunctions {\n  int num_unique_starts(vector t) {\n    if (size(t) == 0) return 0;\n    int us = 1;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) us += 1;\n    }\n    return us;\n  }\nThis is then used to compute the value J to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.\n  array[] int unique_starts(vector t, int J) {\n    array[J + 1] int starts;\n    if (J == 0) return starts;\n    starts[1] = 1;\n    int pos = 2;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) {\n    starts[pos] = n;\n    pos += 1;\n      }\n    }\n    starts[J + 1] = size(t) + 1;\n    return starts;\n  }\n}\nThe data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.\ntransformed data {\n  int&lt;lower=0&gt; J = num_unique_starts(t);\n  array[J + 1] int&lt;lower=0&gt; starts = unique_starts(t, J);\n}\nFor each unique survival time j in 1:J, the subjects indexed from starts[j] to starts[j + 1] - 1 (inclusive) share the same survival time. The number of elements with survival time j is thus (starts[j + 1] - 1) - starts[j] + 1, or just starts[j + 1] - starts[j].\nThe parameters and prior are also the same—just a vector beta of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables J and starts.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\n  real log_denom_lhs = log_sum_exp(log_theta_c);\n  for (j in 1:J) {\n    int start = starts[j];\n    int end = starts[j + 1] - 1;\n    int len = end - start + 1;\n    real log_len = log(len);\n    real numerator = sum(log_theta[start:end]);\n    log_denom_lhs = log_sum_exp(log_denom_lhs,\n                                log_sum_exp(log_theta[start:end]));\n    vector[len] diff;\n    for (ell in 1:len) {\n      diff[ell] = log_diff_exp(log_denom_lhs,\n                               log(ell - 1) - log_len\n                               + log_sum_exp(log_theta[start:end]));\n    }\n    target += numerator - sum(diff);\n  }\nThe special function log_diff_exp is defined as\n\\[\\begin{equation*}\n\\textrm{logDiffExp}(u, v) = \\log(\\exp(u) - \\exp(v)).\n\\end{equation*}\\]\nBecause of how J and starts are constructed, the length len will always be strictly positive so that the log is well defined.",
+    "text": "As described in the reference manual function overloading is permitted in Stan, beginning in version 2.29.\nThis means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.\n\n\nOverloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.\nFor example, consider a function triple which has the following three signatures\nreal triple(real x);\ncomplex triple(complex x);\narray[] real triple(array[] real);\nOne should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect\n// The function does what it says\ntriple(3.0) == 9.0\n// It is defined reasonably for different types\ntriple(to_complex(3.0)) == to_complex(triple(3.0))\n// A container version of this function works by element\ntriple({3.0, 4.0})[0] == triple({3.0, 4.0}[0])\nNote that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.\n\n\n\nStan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.\nConsider the following function signatures\nreal foo(int a, real b);\nreal foo(real a, real b);\nGiven these, the function call foo(1.5, 2.5) is unambiguous - it must resolve to the second signature. But, the function call foo(1, 1.5) could be valid for either under Stan’s promotion rules, which allow integers to be promoted to real numbers.\nTo resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call foo(1, 1.5) would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).\nFurthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures\nreal bar(int x, real y);\nreal bar(real x, int y);\nThese signatures do not have a unique minimum number of promotions for the call bar(1, 2). Both signatures require one int to real promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.\nPromotion from integers to complex numbers is considered to be two separate promotions, first from int to real, then from real to complex. This means that integer arguments will “prefer” a signature with real types over complex types.\nFor example, consider the function signatures\nreal pop(real x);\nreal pop(complex x);\nStan will select the first signature when pop is called with an integer argument such as pop(0).",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Survival Models"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html#exponential-survival-model",
-    "href": "stan-users-guide/survival.html#exponential-survival-model",
-    "title": "Survival Models",
+    "objectID": "stan-users-guide/user-functions.html#documenting-functions.section",
+    "href": "stan-users-guide/user-functions.html#documenting-functions.section",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "The exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if \\(T \\sim \\textrm{exponential}(\\lambda)\\) for some rate \\(\\lambda &gt; 0,\\) then \\[\\begin{equation*}\n\\Pr[T &gt; t] = \\Pr[T &gt; t + t' \\mid T &gt; t'].\n\\end{equation*}\\] If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.\nThe exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable \\(T\\) is \\[\\begin{equation*}\n\\mathbb{E}[T \\mid \\lambda] = \\frac{1}{\\lambda}.\n\\end{equation*}\\] The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) \\(\\beta = 1 / \\lambda\\).\nThe data for a survival model consists of two components. First, there is a vector \\(t \\in (0, \\infty)^N\\) of \\(N\\) observed failure times. Second, there is a censoring time \\(t^{\\textrm{cens}}\\) such that failure times greater than \\(t^{\\textrm{cens}}\\) are not observed. The censoring time assumption imposes a constraint which requires \\(t_n &lt; t^{\\textrm{cens}}\\) for all \\(n \\in 1{:}N.\\) For the censored subjects, the only thing required in the model is their total count, \\(N^\\textrm{cens}\\) (their covariates are also required for models with covariates).\nThe model for the observed failure times is exponential, so that for \\(n \\in 1{:}N,\\) \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda).\n\\end{equation*}\\]\nThe model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of \\[\\begin{equation*}\n\\Pr[T &gt; t^{\\textrm{cens}}] = 1 - F_T(t^{\\textrm{cens}}),\n\\end{equation*}\\] where \\(F_T\\) is the cumulative distribution function (cdf) of survival time \\(T\\) (\\(F_X(x) = \\Pr[X \\leq x]\\) is standard notation for the cdf of a random variable \\(X\\)). The function \\(1 - F_T(t)\\) is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood \\[\\begin{eqnarray*}\np(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\prod_{n=1}^{N^{\\textrm{cens}}}\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)\n\\\\\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)^{N^{\\textrm{cens}}}.\n\\end{eqnarray*}\\]\nOn the log scale, that’s \\[\\begin{eqnarray*}\n\\log p(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\sum_{n=1}^N \\log \\textrm{exponential}(t_n \\mid \\lambda)\n\\\\\n& & { } + N^{\\textrm{cens}} \\cdot \\log \\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda).\n\\end{eqnarray*}\\]\nThe model can be completed with a standard lognormal prior on \\(\\lambda,\\) \\[\\begin{equation*}\n\\lambda \\sim \\textrm{lognormal}(0, 1),\n\\end{equation*}\\] which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed \\(\\textrm{lognormal}(0, 1)\\). In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.\n\n\nThe data for a simple survival analysis without covariates can be coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n}\nIn this program, N is the number of uncensored observations and t contains the times of the uncensored observations. There are a further N_cens items that are right censored at time t_cens. Right censoring means that if the time to failure is greater than\nt_cens, it is only observed that the part survived until time t_cens. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.\nThere is a single rate parameter, the inverse of which is the expected time to failure.\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nThe exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.\nmodel {\n  t ~ exponential(lambda);\n  target += N_cens * exponential_lccdf(t_cens | lambda);\n\n  lambda ~ lognormal(0, 1);\n}\nThe likelihood for rate lambda is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector t as a having an exponential distribution with rate lambda. This data model could have been written as\nfor (n in 1:N) {\n  t[n] ~ exponential(lambda);\n}\nThe log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate lambda. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,\nfor (n in 1:N)\n  target += exponential_lccdf(t_cens | lambda);\nto define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.",
+    "text": "Functions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.\nFor example, here’s some documentation for the prediction matrix generator.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows corresponding to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\nThe comment begins with /**, ends with */, and has an asterisk (*) on each line. It uses @param followed by the argument’s identifier to document a function argument. The tag @return is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with /* and ending with */ to the Stan parser.\nFor functions that raise exceptions, exceptions can be documented using @throws.4\nFor example,\n /** ...\n * @param theta\n * @throws If any of the entries of theta is negative.\n */\nreal entropy(vector theta) {\n  // ...\n}\nUsually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Survival Models"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html#weibull-survival-model",
-    "href": "stan-users-guide/survival.html#weibull-survival-model",
-    "title": "Survival Models",
+    "objectID": "stan-users-guide/user-functions.html#summary-of-function-types",
+    "href": "stan-users-guide/user-functions.html#summary-of-function-types",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "The Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.\nThe Weibull distribution is parameterized by a shape \\(\\alpha &gt; 0\\) and scale \\(\\sigma &gt; 0.\\) For an outcome \\(t \\geq 0\\), the Weibull distribution’s probability density function is \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid \\alpha, \\sigma)\n= \\frac{\\alpha}{\\sigma}\n  \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n  \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^{\\alpha}\\right).\n\\end{equation*}\\] In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter \\(\\beta &gt; 0\\) with probability density function \\[\\begin{equation*}\n\\textrm{exponential}(t \\mid \\beta) =\n\\beta\n\\cdot\n\\exp(-\\beta \\cdot t).\n\\end{equation*}\\] When \\(\\alpha = 1,\\) the Weibull distribution reduces to an exponential distribution, \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid 1, \\sigma)\n=\n\\textrm{exponential}\\!\\left(t \\,\\bigg|\\, \\frac{1}{\\sigma}\\right).\n\\end{equation*}\\] In other words, the Weibull is a continuous expansion of the exponential distribution.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then the expected survival time is \\[\\begin{equation*}\n\\mathbb{E}[T] = \\sigma \\cdot \\Gamma\\!\\left(1 + \\frac{1}{\\alpha}\\right),\n\\end{equation*}\\] where the \\(\\Gamma\\) function is the continuous completion of the factorial function (i.e., \\(\\Gamma(1 + n) = n!\\ \\) for \\(n \\in\n\\mathbb{N}\\)). As \\(\\alpha \\rightarrow 0\\) for a fixed \\(\\sigma\\) or as \\(\\sigma \\rightarrow \\infty\\) for a fixed \\(\\alpha\\), the expected survival time goes to infinity.\nThere are three regimes of the Weibull distribution.\n\n\\(\\alpha &lt; 1.\\) A subject is more likely to fail early. When \\(\\alpha\n&lt; 1,\\) the Weibull density approaches infinity as \\(t \\rightarrow 0.\\)\n\\(\\alpha = 1.\\) The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When \\(\\alpha = 1,\\) the Weibull distribution approaches \\(\\sigma\\) as \\(t\n\\rightarrow 0.\\)\n\\(\\alpha &gt; 1.\\) Subjects are less likely to fail early. When \\(\\alpha &lt; 1,\\) the Weibull density approaches zero as \\(t \\rightarrow 0.\\)\n\nWith \\(\\alpha \\leq 1,\\) the mode is zero (\\(t = 0\\)), whereas with \\(\\alpha &gt; 1,\\) the mode is nonzero (\\(t &gt; 0\\)).\n\n\nWith Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.\nparameters {\n  real&lt;lower=0&gt; beta;\n}\nmodel {\n  t ~ exponential(beta);\n  target += N_cens * exponential_lccdf(t_cens | beta);\n\n  beta ~ lognormal(0, 1);\n}\nThe Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (alpha) and scale (sigma) parameters.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  t ~ weibull(alpha, sigma);\n  target += N_cens * weibull_lccdf(t_cens | alpha, sigma);\n\n  alpha ~ lognormal(0, 1);\n  sigma ~ lognormal(0, 1);\n}\nAs usual, if more is known about expected survival times, alpha and sigma should be given more informative priors.",
+    "text": "Functions may have a void or non-void return type and they may or may not have one of the special suffixes, _lpdf, _lpmf, _lp, or _rng.\n\n\nOnly functions declared to return void may be used as statements. These are also the only functions that use return statements with no arguments.\nOnly functions declared to return non-void values may be used as expressions. These functions require return statements with arguments of a type that matches the declared return type.\n\n\n\nOnly functions ending in _lpmf or _lpdf and with return type real may be used as probability functions in distribution statements.\nOnly functions ending in _lp may access the log probability accumulator through distribution statements or target += statements. Such functions may only be used in the transformed parameters or model blocks.\nOnly functions ending in _rng may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in _rng.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Survival Models"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html#survival-with-covariates",
-    "href": "stan-users-guide/survival.html#survival-with-covariates",
-    "title": "Survival Models",
+    "objectID": "stan-users-guide/user-functions.html#recursive-functions",
+    "href": "stan-users-guide/user-functions.html#recursive-functions",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Suppose that for each of \\(n \\in 1{:}N\\) items observed, both censored and uncensored, there is a covariate (row) vector \\(x_n \\in\n\\mathbb{R}^K.\\) For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.\nSurvival with covariates replaces what is essentially a simple regression with only an intercept \\(\\lambda\\) with a generalized linear model with a log link, where the rate for item \\(n\\) is \\[\\begin{equation*}\n\\lambda_n = \\exp(x_n \\cdot \\beta),\n\\end{equation*}\\] where \\(\\beta \\in \\mathbb{R}^K\\) is a \\(K\\)-vector of regression coefficients. Thus \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda_n).\n\\end{equation*}\\] The censored items have probability \\[\\begin{equation*}\n\\Pr[n\\textrm{-th censored}] =\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid x^{\\textrm{cens}}_n\n\\cdot \\beta).\n\\end{equation*}\\]\nThe covariates form an \\(N \\times K\\) data matrix, \\(x \\in\n\\mathbb{R}^{N \\times K}\\). An intercept can be introduced by adding a column of 1 values to \\(x\\).\nA Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (t and t_cens) corresponds to the ordering of items in the covariate matrices (x and x_cens).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n  int&lt;lower=0&gt; K;\n  matrix[N, K] x;\n  matrix[N_cens, K] x_cens;\n}\nparameters {\n  vector[K] gamma;\n}\nmodel {\n  gamma ~ normal(0, 2);\n\n  t ~ exponential(exp(x * gamma));\n  target += exponential_lccdf(t_cens | exp(x_cens * gamma));\n}\nBoth the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.",
+    "text": "Stan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, \\(A^n\\), which is defined for a square matrix \\(A\\) and positive integer \\(n\\) by \\[\nA^n\n=\n\\begin{cases}\n\\textrm{I} & \\quad\\text{if } n = 0, \\text{ and} \\\\\nA \\, A^{n-1} & \\quad\\text{if } n &gt; 0.\n\\end{cases}\n\\]\nwhere \\(\\textrm{I}\\) is the identity matrix. This definition can be directly translated to a recursive function definition.\nmatrix matrix_pow(matrix a, int n) {\n  if (n == 0) {\n    return diag_matrix(rep_vector(1, rows(a)));\n  } else {\n    return a *  matrix_pow(a, n - 1);\n  }\n}\nIt would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.\nelse if (n == 1) {\n  return a;\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Survival Models"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html#hazard-and-survival-functions",
-    "href": "stan-users-guide/survival.html#hazard-and-survival-functions",
-    "title": "Survival Models",
+    "objectID": "stan-users-guide/user-functions.html#truncated-random-number-generation",
+    "href": "stan-users-guide/user-functions.html#truncated-random-number-generation",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Suppose \\(T\\) is a random variable representing a survival time, with a smooth cumulative distribution function \\[\\begin{equation*}\nF_T(t) = \\Pr[T \\leq t],\n\\end{equation*}\\] so that its probability density function is \\[\\begin{equation*}\np_T(t) = \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t).\n\\end{equation*}\\]\nThe survival function \\(S(t)\\) is the probability of surviving until at least time \\(t\\), which is just the complementary cumulative distribution function (ccdf) of the survival random variable \\(T\\), \\[\\begin{equation*}\nS(t) = 1 - F_T(t).\n\\end{equation*}\\] The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).\nThe hazard function \\(h(t)\\) is the instantaneous risk of not surviving past time \\(t\\) assuming survival until time \\(t\\), which is given by \\[\\begin{equation*}\nh(t) = \\frac{p_T(t)}{S(t)} = \\frac{p_T(t)}{1 - F_T(t)}.\n\\end{equation*}\\] The cumulative hazard function \\(H(t)\\) is defined to be the accumulated hazard over time, \\[\\begin{equation*}\nH(t) = \\int_0^t h(u) \\, \\textrm{d}u.\n\\end{equation*}\\]\nThe hazard function and survival function are related through the differential equation \\[\\begin{eqnarray*}\nh(t) & = & -\\frac{\\textrm{d}}{\\textrm{d}t} \\log S(t).\n\\\\[4pt]\n& = & -\\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} S(t)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} -(1 - F_T(t))\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} (F_T(t) - 1)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t)\n\\\\[4pt]\n& = & \\frac{p_T(t)}{S(t)}.\n\\end{eqnarray*}\\]\nIf \\(T \\sim \\textrm{exponential}(\\beta)\\) has an exponential distribution, then its hazard function is constant, \\[\\begin{eqnarray*}\nh(t \\mid \\beta)\n& = & \\frac{p_T(t \\mid \\beta)}{S(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\textrm{exponential}(t \\mid \\beta)}{1 - \\textrm{exponentialCCDF}(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {1 - (1 - \\exp(-\\beta \\cdot t))}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {\\exp(-\\beta \\cdot t)}\n\\\\[4pt]\n& = & \\beta.\n\\end{eqnarray*}\\] The exponential distribution is the only distribution of survival times with a constant hazard function.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then its hazard function is \\[\\begin{eqnarray*}\nh(t \\mid \\alpha, \\sigma)\n& = & \\frac{p_T(t \\mid \\alpha, \\sigma)}{S(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = & \\frac{\\textrm{Weibull}(t \\mid \\alpha, \\sigma)}{1 - \\textrm{WeibullCCDF}(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = &\n\\frac{\\frac{\\alpha}{\\sigma} \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n      \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma} \\right)^\\alpha\\right)}\n     {1 - \\left(1 -\n       \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^\\alpha\n     \\right)\\right)}      \n\\\\[4pt]\n& = & \\frac{\\alpha}{\\sigma}\n  \\cdot\n  \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}.\n\\end{eqnarray*}\\]\nIf \\(\\alpha = 1\\) the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when \\(\\alpha = 1\\)). When \\(\\alpha &gt; 1,\\) the hazard grows as time passes, whereas when \\(\\alpha &lt; 1,\\) it decreases as time passes.",
+    "text": "To generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then run through the inverse cumulative distribution function, \\(y = \\textrm{logit}(u)\\). If this were not already built in as logistic_rng(0, 1), it could be coded in Stan directly as\nreal standard_logistic_rng() {\n  real u = uniform_rng(0, 1);\n  real y = logit(u);\n  return y;\n}\nFollowing the same pattern, a standard normal RNG could be coded as\nreal standard_normal_rng() {\n  real u = uniform_rng(0, 1);\n  real y = inv_Phi(u);\n  return y;\n}\nthat is, \\(y = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}\\) is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function inv_Phi.\nIn order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate \\(\\textsf{normal}(\\mu, \\sigma)\\) variates, it is enough to generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then convert it to a standard normal variate, \\(z = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}(\\cdot)\\) is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, \\(y = \\mu +\n\\sigma \\times z\\). In code,\nreal my_normal_rng(real mu, real sigma) {\n  real u = uniform_rng(0, 1);\n  real z = inv_Phi(u);\n  real y = mu + sigma * z;\n  return y;\n}\nA robust version of this function would test that the arguments are finite and that sigma is non-negative, e.g.,\n  if (is_nan(mu) || is_inf(mu)) {\n    reject(\"my_normal_rng: mu must be finite; \",\n           \"found mu = \", mu);\n  }\n  if (is_nan(sigma) || is_inf(sigma) || sigma &lt; 0) {\n    reject(\"my_normal_rng: sigma must be finite and non-negative; \",\n           \"found sigma = \", sigma);\n  }\n\n\n\nOften truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.\n\n\nFor example, the following code generates a \\(\\textsf{Weibull}(\\alpha, \\sigma)\\) variate truncated below at a time \\(t\\),5\nreal weibull_lb_rng(real alpha, real sigma, real t) {\n  real p = weibull_cdf(t | alpha, sigma);   // cdf for lb\n  real u = uniform_rng(p, 1);               // unif in bounds\n  real y = sigma * (-log1m(u))^inv(alpha);  // inverse cdf\n  return y;\n}\n\n\n\nIf there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a \\(\\textsf{normal}(\\mu, \\sigma)\\) truncated to a region \\((a, b)\\), the following code suffices,\nreal normal_lub_rng(real mu, real sigma, real lb, real ub) {\n  real p_lb = normal_cdf(lb | mu, sigma);\n  real p_ub = normal_cdf(ub | mu, sigma);\n  real u = uniform_rng(p_lb, p_ub);\n  real y = mu + sigma * inv_Phi(u);\n  return y;\n}\nTo make this more robust, all variables should be tested for finiteness, sigma should be tested for positiveness, and lb and ub should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement\n  return mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),\n                                          normal_cdf(ub | mu, sigma)));\nfor readability. The names like p indicate probabilities, and p_lb and p_ub indicate the probabilities of the bounds. The variable u is clearly named as a uniform variate, and y is used to denote the variate being generated itself.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Survival Models"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html#proportional-hazards-model",
-    "href": "stan-users-guide/survival.html#proportional-hazards-model",
-    "title": "Survival Models",
-    "section": "",
-    "text": "The exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. Cox (1972) introduced a semi-parametric survival model specified directly in terms of a hazard function \\(h(t)\\) rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.\nLet \\(x_n \\in \\mathbb{R}^K\\) be a (row) vector of covariates for subject \\(n\\) so that the full covariate data matrix is \\(x \\in \\mathbb{R}^{N \\times\nK}\\). In Cox’s model, the hazard function for subject \\(n\\) is defined conditionally in terms of their covariates \\(x_n\\) and the parameter vector \\(\\gamma \\in \\mathbb{R}^K\\) as \\[\\begin{equation*}\nh(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n \\cdot \\gamma),\n\\end{equation*}\\] where \\(h_0(t)\\) is a shared baseline hazard function and \\(x_n \\cdot\n\\gamma = \\sum_{k=1}^K x_{n, k} \\cdot \\beta_k\\) is a row vector-vector product.\nIn the semi-parametric, proportional hazards model, the baseline hazard function \\(h_0(t)\\) is not modeled. This is why it is called “semi-parametric.” Only the factor \\(\\exp(x_n \\cdot \\gamma),\\) which determines how individual \\(n\\) varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”\nCox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function \\(h_0(t)\\) is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function1.\n\n\nCox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function \\(h_0(t)\\) is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.\nThe hazard function \\(h(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n\n\\cdot \\beta)\\) for subject \\(n\\) represents the instantaneous probability that subject \\(n\\) fails at time \\(t\\) given that it has survived until time \\(t.\\) The probability that subject \\(n\\) is the first to fail among \\(N\\) subjects is thus proportional to subject \\(n\\)’s hazard function, \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail at time } t]\n\\propto h(t \\mid x_n, \\beta).\n\\end{equation*}\\] Normalizing yields \\[\\begin{eqnarray*}\n\\Pr[n \\textrm{ first to fail at time } t]\n& = &  \\frac{h(t \\mid x_n, \\beta)}\n            {\\sum_{n' = 1}^N h(t \\mid x_{n'}, \\beta)}\n\\\\[4pt]\n& = &  \\frac{h_0(t) \\cdot \\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N h_0(t) \\cdot \\exp(x_{n'} \\cdot \\beta)}\n\\\\[4pt]\n& = &  \\frac{\\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{eqnarray*}\\]\nSuppose there are \\(N\\) subjects with strictly ordered survival times \\(t_1 &lt;\nt_2 &lt; \\cdots &lt; t_N\\) and covariate (row) vectors \\(x_1, \\ldots, x_N\\). Let \\(t^{\\textrm{cens}}\\) be the (right) censoring time and let \\(N^{\\textrm{obs}}\\) be the largest value of \\(n\\) such that \\(t_n \\leq\nt^{\\textrm{cens}}\\). This means \\(N^{\\textrm{obs}}\\) is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.\nWith failure times sorted in decreasing order, the partial likelihood for each observed subject \\(n \\in 1{:}N^{\\textrm{obs}}\\) can be expressed as \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail among } n, n + 1, \\ldots N]\n= \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.\nThe partial likelihood, defined in this form by Breslow (1975), is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). \\[\\begin{equation*}\n\\Pr[\\textrm{observed failures ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n= \\prod_{n = 1}^{N^{\\textrm{obs}}}\n  \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] On the log scale, \\[\\begin{eqnarray*}\n\\log \\Pr[\\textrm{obs.\\ fail ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n& = &\n\\sum_{n = 1}^{N^{\\textrm{obs}}}\n  \\log \\left(\n          \\frac{\\exp(x_n \\cdot \\beta)}\n               {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}\n       \\right)\n\\\\[4pt]\n& = & x_n \\cdot \\beta - \\log \\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)\n\\\\\n& = & x_n \\cdot \\beta - \\textrm{logSumExp}_{n' = n}^N \\ x_{n'} \\cdot \\beta,\n\\end{eqnarray*}\\] where \\[\\begin{equation*}\n\\textrm{logSumExp}_{n = a}^b \\ x_n\n= \\log \\sum_{n = a}^b \\exp(x_n)\n\\end{equation*}\\] is implemented so as to preserve numerical precision.\nThis likelihood follows the same approach to ranking as that developed by Plackett (1975) for estimating the probability of the order of the first few finishers in a horse race.\nA simple normal prior on the components of \\(\\beta\\) completes the model, \\[\\begin{equation*}\n\\beta \\sim \\textrm{normal}(0, 2).\n\\end{equation*}\\] This should be scaled based on knowledge of the predictors.\n\n\n\nTo simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.\ndata {\n  int&lt;lower=0&gt; K;          // num covariates\n\n  int&lt;lower=0&gt; N;          // num uncensored obs\n  vector[N] t;             // event time (non-strict decreasing)\n  matrix[N, K] x;          // covariates for uncensored obs\n\n  int N_c;                 // num censored obs\n  real&lt;lower=t[N]&gt; t_c;    // censoring time\n  matrix[N_c, K] x_c;      // covariates for censored obs\n}\nThe parameters are just the coefficients.\nparameters {\n  vector[K] beta;          // slopes (no intercept)\n}\nThe prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.\nmodel {\n  beta ~ normal(0, 2);\n  ...\nThe log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\nThese vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.\n  real log_denom = log_sum_exp(log_theta_c);\nThen, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.\n  for (n in 1:N) {\n    log_denom = log_sum_exp(log_denom, log_theta[n]);\n    target += log_theta[n] - log_denom;   // log likelihood\n  }\nThe running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.\ntransformed data {\n  for (n in 2:N) {\n    if (!(t[n] &lt; t[n - 1])) {\n      reject(\"times must be strictly decreasing, but found\"\n             \"!(t[\", n, \"] &lt; t[, \", (n - 1), \"])\");\n    }   \n  }\n}\n\n\n\nTechnically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all \\(N!\\) permutations for a set of \\(N\\) subjects with tied survival times is not tractable. As an alternative, Efron (1977) introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.\nIn the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.\nfunctions {\n  int num_unique_starts(vector t) {\n    if (size(t) == 0) return 0;\n    int us = 1;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) us += 1;\n    }\n    return us;\n  }\nThis is then used to compute the value J to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.\n  array[] int unique_starts(vector t, int J) {\n    array[J + 1] int starts;\n    if (J == 0) return starts;\n    starts[1] = 1;\n    int pos = 2;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) {\n    starts[pos] = n;\n    pos += 1;\n      }\n    }\n    starts[J + 1] = size(t) + 1;\n    return starts;\n  }\n}\nThe data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.\ntransformed data {\n  int&lt;lower=0&gt; J = num_unique_starts(t);\n  array[J + 1] int&lt;lower=0&gt; starts = unique_starts(t, J);\n}\nFor each unique survival time j in 1:J, the subjects indexed from starts[j] to starts[j + 1] - 1 (inclusive) share the same survival time. The number of elements with survival time j is thus (starts[j + 1] - 1) - starts[j] + 1, or just starts[j + 1] - starts[j].\nThe parameters and prior are also the same—just a vector beta of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables J and starts.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\n  real log_denom_lhs = log_sum_exp(log_theta_c);\n  for (j in 1:J) {\n    int start = starts[j];\n    int end = starts[j + 1] - 1;\n    int len = end - start + 1;\n    real log_len = log(len);\n    real numerator = sum(log_theta[start:end]);\n    log_denom_lhs = log_sum_exp(log_denom_lhs,\n                                log_sum_exp(log_theta[start:end]));\n    vector[len] diff;\n    for (ell in 1:len) {\n      diff[ell] = log_diff_exp(log_denom_lhs,\n                               log(ell - 1) - log_len\n                               + log_sum_exp(log_theta[start:end]));\n    }\n    target += numerator - sum(diff);\n  }\nThe special function log_diff_exp is defined as\n\\[\\begin{equation*}\n\\textrm{logDiffExp}(u, v) = \\log(\\exp(u) - \\exp(v)).\n\\end{equation*}\\]\nBecause of how J and starts are constructed, the length len will always be strictly positive so that the log is well defined.",
+    "objectID": "stan-users-guide/user-functions.html#footnotes",
+    "href": "stan-users-guide/user-functions.html#footnotes",
+    "title": "User-Defined Functions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe main problem with comments is that they can be misleading, either due to misunderstandings on the programmer’s part or because the program’s behavior is modified after the comment is written. The program always behaves the way the code is written, which is why refactoring complex code into understandable units is preferable to simply adding comments.↩︎\nJust because this makes it possible to code a rejection sampler does not make it a good idea. Rejections break differentiability and the smooth exploration of the posterior. In Hamiltonian Monte Carlo, it can cause the sampler to be reduced to a diffusive random walk.↩︎\nA range of built-in validation routines is coming to Stan soon! Alternatively, the reject statement can be used to check constraints on the simplex.↩︎\nAs of Stan 2.9.0, the only way a user-defined producer will raise an exception is if a function it calls (including distribution statements) raises an exception via the reject statement.↩︎\nThe original code and impetus for including this in the manual came from the Stan forums post; by user lcomm, who also explained truncation above and below.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Survival Models"
+      "Programming Techniques",
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/survival.html#footnotes",
-    "href": "stan-users-guide/survival.html#footnotes",
-    "title": "Survival Models",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nCox mentioned in his seminal paper that modeling the baseline hazard function would improve statistical efficiency, but he did not do it for computational reasons.↩︎",
+    "objectID": "stan-users-guide/time-series.html",
+    "href": "stan-users-guide/time-series.html",
+    "title": "Time-Series Models",
+    "section": "",
+    "text": "Times series data come arranged in temporal order. This chapter presents two kinds of time series models, regression-like models such as autoregressive and moving average models, and hidden Markov models.\nThe Gaussian processes chapter presents Gaussian processes, which may also be used for time-series (and spatial) data.\n\n\nA first-order autoregressive model (AR(1)) with normal noise takes each point \\(y_n\\) in a sequence \\(y\\) to be generated according to \\[\ny_n \\sim \\textsf{normal}(\\alpha + \\beta y_{n-1}, \\sigma).\n\\]\nThat is, the expected value of \\(y_n\\) is \\(\\alpha + \\beta y_{n-1}\\), with noise scaled as \\(\\sigma\\).\n\n\nWith improper flat priors on the regression coefficients \\(\\alpha\\) and \\(\\beta\\) and on the positively-constrained noise scale (\\(\\sigma\\)), the Stan program for the AR(1) model is as follows.1\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  for (n in 2:N) {\n    y[n] ~ normal(alpha + beta * y[n-1], sigma);\n  }\n}\nThe first observed data point, y[1], is not modeled here because there is nothing to condition on; instead, it acts to condition y[2]. This model also uses an improper prior for sigma, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in y over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of y is available.\n\n\nAlthough perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner\nmodel {\n  y[2:N] ~ normal(alpha + beta * y[1:(N - 1)], sigma);\n}\nThe left-hand side slicing operation pulls out the last \\(N-1\\) elements and the right-hand side version pulls out the first \\(N-1\\).\n\n\n\n\nProper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-\\(t\\) distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.\nTo enforce the estimation of a stationary AR(1) process, the slope coefficient beta may be constrained with bounds as follows.\nreal&lt;lower=-1, upper=1&gt; beta;\nIn practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of beta near zero.\n\n\n\nExtending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient gamma and the following model statement.\nfor (n in 3:N) {\n  y[n] ~ normal(alpha + beta*y[n-1] + gamma*y[n-2], sigma);\n}\n\n\n\nA general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] real y;\n}\nparameters {\n  real alpha;\n  array[K] real beta;\n  real sigma;\n}\nmodel {\n  for (n in (K+1):N) {\n    real mu = alpha;\n    for (k in 1:K) {\n      mu += beta[k] * y[n-k];\n    }\n    y[n] ~ normal(mu, sigma);\n  }\n}\n\n\n\nEconometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model (Engle 1982). Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.\nThe ARCH(1) model is typically presented as the following sequence of equations, where \\(r_t\\) is the observed return at time point \\(t\\) and \\(\\mu\\), \\(\\alpha_0\\), and \\(\\alpha_1\\) are unknown regression coefficient parameters.\n\\[\\begin{align*}\nr_t        &=  \\mu + a_t \\\\\na_t        &=  \\sigma_t \\epsilon_t \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\sigma^2_t &=  \\alpha_0 + \\alpha_1 a_{t-1}^2\n\\end{align*}\\]\nIn order to ensure the noise terms \\(\\sigma^2_t\\) are positive, the scale coefficients are constrained to be positive, \\(\\alpha_0, \\alpha_1\n&gt; 0\\). To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., \\(\\alpha_1 &lt; 1\\).2\nThe ARCH(1) model may be coded directly in Stan as follows.\ndata {\n  int&lt;lower=0&gt; T;                // number of time points\n  array[T] real r;               // return at time t\n}\nparameters {\n  real mu;                       // average return\n  real&lt;lower=0&gt; alpha0;          // noise intercept\n  real&lt;lower=0, upper=1&gt; alpha1; // noise slope\n}\nmodel {\n  for (t in 2:T) {\n    r[t] ~ normal(mu, sqrt(alpha0 + alpha1\n                                    * pow(r[t - 1] - mu,2)));\n  }\n}\nThe loop in the model is defined so that the return at time \\(t=1\\) is not modeled; the model in the next section shows how to model the return at \\(t=1\\). The model can be vectorized to be more efficient; the model in the next section provides an example.\n\n\n\n\nA set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.\n\n\nThe basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time \\(t-1\\) as a predictor of volatility at time \\(t\\), defining \\[\n\\sigma^2_t = \\alpha_0 + \\alpha_1 a^2_{t-1} + \\beta_1 \\sigma^2_{t-1}.\n\\]\nTo ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy \\(\\alpha_0, \\alpha_1,\n\\beta_1 &gt; 0\\) and the slopes \\(\\alpha_1 + \\beta_1 &lt; 1\\).\ndata {\n  int&lt;lower=0&gt; T;\n  array[T] real r;\n  real&lt;lower=0&gt; sigma1;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; alpha0;\n  real&lt;lower=0, upper=1&gt; alpha1;\n  real&lt;lower=0, upper=(1-alpha1)&gt; beta1;\n}\ntransformed parameters {\n  array[T] real&lt;lower=0&gt; sigma;\n  sigma[1] = sigma1;\n  for (t in 2:T) {\n    sigma[t] = sqrt(alpha0\n                     + alpha1 * pow(r[t - 1] - mu, 2)\n                     + beta1 * pow(sigma[t - 1], 2));\n  }\n}\nmodel {\n  r ~ normal(mu, sigma);\n}\nTo get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value sigma1 for the scale of the noise at \\(t = 1\\).\nThe constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on beta1 depends on the value of alpha1.\nA transformed parameter array of non-negative values sigma is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept alpha0, a slope alpha1 for the squared difference in return from the mean at the previous time, and a slope beta1 for the previous noise scale squared. Finally, the whole regression is inside the sqrt function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.\nWith the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because r and sigma are of length T, all of the data are modeled directly.\n\n\n\n\nA moving average model uses previous errors as predictors for future outcomes. For a moving average model of order \\(Q\\), \\(\\mbox{MA}(Q)\\), there is an overall mean parameter \\(\\mu\\) and regression coefficients \\(\\theta_q\\) for previous error terms. With \\(\\epsilon_t\\) being the noise at time \\(t\\), the model for outcome \\(y_t\\) is defined by \\[\ny_t = \\mu + \\theta_1 \\epsilon_{t-1} + \\dotsb + \\theta_Q \\epsilon_{t-Q}\n+ \\epsilon_t,\n\\] with the noise term \\(\\epsilon_t\\) for outcome \\(y_t\\) modeled as normal, \\[\n\\epsilon_t \\sim \\textsf{normal}(0,\\sigma).\n\\] In a proper Bayesian model, the parameters \\(\\mu\\), \\(\\theta\\), and \\(\\sigma\\) must all be given priors.\n\n\nAn \\(\\mbox{MA}(2)\\) model can be coded in Stan as follows.\ndata {\n  int&lt;lower=3&gt; T;          // number of observations\n  vector[T] y;             // observation at time T\n}\nparameters {\n  real mu;                 // mean\n  real&lt;lower=0&gt; sigma;     // error scale\n  vector[2] theta;         // lag coefficients\n}\ntransformed parameters {\n  vector[T] epsilon;       // error terms\n  epsilon[1] = y[1] - mu;\n  epsilon[2] = y[2] - mu - theta[1] * epsilon[1];\n  for (t in 3:T) {\n    epsilon[t] = ( y[t] - mu\n                    - theta[1] * epsilon[t - 1]\n                    - theta[2] * epsilon[t - 2] );\n  }\n}\nmodel {\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 3:T) {\n    y[t] ~ normal(mu\n                  + theta[1] * epsilon[t - 1]\n                  + theta[2] * epsilon[t - 2],\n                  sigma);\n  }\n}\nThe error terms \\(\\epsilon_t\\) are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to \\(y_n\\) for \\(n &gt; Q\\). In this example, the parameters are all given Cauchy (half-Cauchy for \\(\\sigma\\)) priors, although other priors can be used just as easily.\nThis model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the \\(\\epsilon_t\\) could also be sped up by using a dot product instead of a loop.\n\n\n\nA general \\(\\mbox{MA}(Q)\\) model with a vectorized distribution statement may be defined as follows.\ndata {\n  int&lt;lower=0&gt; Q;       // num previous noise terms\n  int&lt;lower=3&gt; T;       // num observations\n  vector[T] y;          // observation at time t\n}\nparameters {\n  real mu;              // mean\n  real&lt;lower=0&gt; sigma;  // error scale\n  vector[Q] theta;      // error coeff, lag -t\n}\ntransformed parameters {\n  vector[T] epsilon;    // error term at time t\n  for (t in 1:T) {\n    epsilon[t] = y[t] - mu;\n    for (q in 1:min(t - 1, Q)) {\n      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];\n    }\n  }\n}\nmodel {\n  vector[T] eta;\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 1:T) {\n    eta[t] = mu;\n    for (q in 1:min(t - 1, Q)) {\n      eta[t] = eta[t] + theta[q] * epsilon[t - q];\n    }\n  }\n  y ~ normal(eta, sigma);\n}\nHere all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).\n\n\n\n\nAutoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.\ndata {\n  int&lt;lower=1&gt; T;            // num observations\n  array[T] real y;                 // observed outputs\n}\nparameters {\n  real mu;                   // mean coeff\n  real phi;                  // autoregression coeff\n  real theta;                // moving avg coeff\n  real&lt;lower=0&gt; sigma;       // noise scale\n}\nmodel {\n  vector[T] nu;              // prediction for time t\n  vector[T] err;             // error for time t\n  nu[1] = mu + phi * mu;     // assume err[0] == 0\n  err[1] = y[1] - nu[1];\n  for (t in 2:T) {\n    nu[t] = mu + phi * y[t - 1] + theta * err[t - 1];\n    err[t] = y[t] - nu[t];\n  }\n  mu ~ normal(0, 10);        // priors\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err ~ normal(0, sigma);    // error model\n}\nThe data are declared in the same way as the other time-series regressions and the parameters are documented in the code.\nIn the model block, the local vector nu stores the predictions and err the errors. These are computed similarly to the errors in the moving average models described in the previous section.\nThe priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.\nOften in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring err as a transformed parameter, then defining it the same way as in the model above. The vector nu could still be a local variable, only now it will be in the transformed parameter block.\nWayne Folta suggested encoding the model without local vector variables as follows.\nmodel {\n  real err;\n  mu ~ normal(0, 10);\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err = y[1] - (mu + phi * mu);\n  err ~ normal(0, sigma);\n  for (t in 2:T) {\n    err = y[t] - (mu + phi * y[t - 1] + theta * err);\n    err ~ normal(0, sigma);\n  }\n}\nThis approach to ARMA models illustrates how local variables, such as err in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.\nBoth encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just err.\n\n\nMA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint3\nreal&lt;lower=-1, upper=1&gt; theta;\nWhen the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (theta, phi) outside the \\([-1,1]\\) interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).\nFurther, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.\nreal&lt;lower=-1, upper=1&gt; phi;\n\n\n\n\nStochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time (Kim, Shephard, and Chib 1998). The data consist of mean corrected (i.e., centered) returns \\(y_t\\) on an underlying asset at \\(T\\) equally spaced time points. Kim et al. formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter \\(h_t\\) for the log volatility, along with parameters \\(\\mu\\) for the mean log volatility, and \\(\\phi\\) for the persistence of the volatility term. The variable \\(\\epsilon_t\\) represents the white-noise shock (i.e., multiplicative error) on the asset return at time \\(t\\), whereas \\(\\delta_t\\) represents the shock on volatility at time \\(t\\). \\[\\begin{align*}\ny_t        &=    \\epsilon_t \\exp(h_t / 2) \\\\\nh_{t+1}    &=    \\mu + \\phi (h_t - \\mu) + \\delta_t \\sigma \\\\\nh_1        &\\sim \\textsf{normal}\\left( \\mu, \\frac{\\sigma}{\\sqrt{1 - \\phi^2}} \\right) \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\delta_t   &\\sim \\textsf{normal}(0,1)\n\\end{align*}\\]\nRearranging the first line, \\(\\epsilon_t = y_t \\exp(-h_t / 2)\\), allowing the distribution for \\(y_t\\) to be written as \\[\ny_t \\sim \\textsf{normal}(0,\\exp(h_t/2)).\n\\] The recurrence equation for \\(h_{t+1}\\) may be combined with the scaling of \\(\\delta_t\\) to yield the distribution \\[\nh_t \\sim \\mathsf{normal}(\\mu + \\phi(h_{t-1} - \\mu), \\sigma).\n\\] This formulation can be directly encoded, as shown in the following Stan model.\ndata {\n  int&lt;lower=0&gt; T;   // # time points (equally spaced)\n  vector[T] y;      // mean corrected return at time t\n}\nparameters {\n  real mu;                     // mean log volatility\n  real&lt;lower=-1, upper=1&gt; phi; // persistence of volatility\n  real&lt;lower=0&gt; sigma;         // white noise shock scale\n  vector[T] h;                 // log volatility at time t\n}\nmodel {\n  phi ~ uniform(-1, 1);\n  sigma ~ cauchy(0, 5);\n  mu ~ cauchy(0, 10);\n  h[1] ~ normal(mu, sigma / sqrt(1 - phi * phi));\n  for (t in 2:T) {\n    h[t] ~ normal(mu + phi * (h[t - 1] -  mu), sigma);\n  }\n  for (t in 1:T) {\n    y[t] ~ normal(0, exp(h[t] / 2));\n  }\n}\nCompared to the Kim et al. formulation, the Stan model adds priors for the parameters \\(\\phi\\), \\(\\sigma\\), and \\(\\mu\\). The shock terms \\(\\epsilon_t\\) and \\(\\delta_t\\) do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.\nThe posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with \\(\\mu = -1.02\\), \\(\\phi = 0.95\\), and \\(\\sigma = 0.25\\) leads to 95% posterior intervals for \\(\\mu\\) of \\((-1.23, -0.54)\\), for \\(\\phi\\) of \\((0.82, 0.98)\\), and for \\(\\sigma\\) of \\((0.16, 0.38)\\).\nThe NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in (Hoffman and Gelman 2014). Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of \\(T\\).\nIt is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return \\(y\\) is easily vectorized to\ny ~ normal(0, exp(h / 2));\nThis speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter h_std to be declared instead of h.\nparameters {\n  // ...\n  vector[T] h_std;  // std log volatility time t\n}\nThe original value of h is then defined in a transformed parameter block.\ntransformed parameters {\n  vector[T] h = h_std * sigma;  // now h ~ normal(0, sigma)\n  h[1] /= sqrt(1 - phi * phi);  // rescale h[1]\n  h += mu;\n  for (t in 2:T) {\n    h[t] += phi * (h[t - 1] - mu);\n  }\n}\nThe first assignment rescales h_std to have a \\(\\textsf{normal}(0,\\sigma)\\) distribution and temporarily assigns it to h. The second assignment rescales h[1] so that its prior differs from that of h[2] through h[T]. The next assignment supplies a mu offset, so that h[2] through h[T] are now distributed \\(\\textsf{normal}(\\mu,\\sigma)\\); note that this shift must be done after the rescaling of h[1]. The final loop adds in the moving average so that h[2] through h[T] are appropriately modeled relative to phi and mu.\nAs a final improvement, the distribution statements for h[1] to h[T] are replaced with a single vectorized standard normal distribution statement.\nmodel {\n  // ...\n  h_std ~ std_normal();\n}\nAlthough the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.\n\n\n\nA Hidden Markov model is a probabilistic model over \\(N\\) observations \\(y_{1:N}\\) and \\(N\\) hidden states \\(z_{1:N}\\). This models is defined by the conditional distributions \\(p(y_n \\mid z_n, \\phi)\\) and \\(p(z_n \\mid z_{n-1}, \\phi)\\). Here we make the dependency on additional model parameters \\(\\phi\\) explicit. (\\(\\phi\\) may be a vector of parameters.) The complete data likelihood is then \\[\np(y, z \\mid \\phi) = \\prod_n p(y_n \\mid z_n, \\phi) p(z_n \\mid z_{n - 1}, \\phi)\n\\] When \\(z_{1:N}\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(z\\) out.\nWhen each state \\(z\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can use Stan’s suite of HMM functions to marginalize out \\(z_{1:N}\\) and compute \\[\np(y_{1:N} \\mid \\phi) = \\int_{\\mathcal Z} p(y, z \\mid \\phi) \\text d z.\n\\] We start by defining the conditional observation distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid z_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n\\Gamma_{ij} = p(z_n = j \\mid z_{n - 1} = i, \\phi).\n\\] (This is a right-stochastic matrix.) Finally, we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n\\rho_k = p(z_0 = k \\mid \\phi).\n\\] It is common practice to set \\(\\rho\\) to be the stationary distribution of the HMM, that is \\(\\rho\\) is the first eigenvector of \\(\\Gamma\\) and solves \\(\\Gamma \\rho = \\rho\\).\nAs an example, consider a three-state model with \\(K=3\\). The observations are normally distributed conditional on the HMM states with \\[\n  y_n \\sim \\text{normal}(\\mu_k, \\sigma),\n\\] where \\(\\mu = (1, 5, 9)\\) and the standard deviation \\(\\sigma\\) is the same across all observations. The model is then\ndata {\n  int N;  // Number of observations\n  array[N] real y;\n}\n\nparameters {\n  // Rows of the transition matrix\n  array[3] simplex[3] gamma_arr;\n\n  // Initial state\n  simplex[3] rho;\n\n  // Parameters of measurement model\n  vector[3] mu;\n  real&lt;lower = 0.0&gt; sigma;\n}\n\ntransformed parameters {\n  // Build transition matrix\n  matrix[3, 3] gamma;\n  for (k in 1:3) gamma[k, ] = to_row_vector(gamma_arr[k]);\n\n  // Compute the log likelihoods in each possible state\n  matrix[3, N] log_omega;\n  for (n in 1:N) {\n    for (i in 1:3) {\n      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);\n    }\n  }\n}\n\nmodel {\n  // prior\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  \n  // no explicit prior on gamma_arr, meaning we default to a\n  // uniform prior over the simplexes.\n\n  // Increment target by log p(y | mu, sigma, Gamma, rho)\n  target += hmm_marginal(log_omega, gamma, rho);\n}\nThe last function hmm_marginal takes in all the ingredients of the HMM and computes the relevant log marginal distribution, \\(\\log p(y \\mid \\phi)\\).\nIf we desire draws from the posterior distribution of \\(z\\), we use the generated quantities block and draw, for each sample \\(\\phi\\), a sample from \\(p(z \\mid y, \\phi)\\). In effect, MCMC produces draws from \\(p(\\phi \\mid y)\\) and with the draws in generated quantities, we obtain draws from \\(p(\\phi \\mid y) p(z \\mid y, \\phi) = p(z, \\phi \\mid y)\\). It is also possible to compute the posterior probbability of each hidden state, that is \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). Averagging these probabilities over all MCMC draws, we obtain \\(\\text{Pr}(z_n = k \\mid y)\\).\ngenerated quantities {\n  array[N] int latent_states = hmm_latent_rng(log_omega, gamma, rho);\n  matrix[3, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);\n}\nhmm_hidden_state_prob returns the marginal probabilities of each state, \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). This function cannot be used to compute the joint probability \\(\\text{Pr}(z \\mid \\phi, y)\\), because such calculation requires accounting for the posterior correlation between the different components of \\(z\\). Therefore, hidden_probs should not be used to obtain posterior draws. Instead, users should rely on hmm_latent_rng.\ngenerated quantities {\n   array[N] int&lt;lower=1, upper=K&gt; z = hmm_latent_rng(...fill-in params here to match example...);\n}\nThe example in this section is derived from the more detailed case study by Ben Bales: https://mc-stan.org/users/documentation/case-studies/hmm-example.html.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Survival Models"
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/sparse-ragged.html",
-    "href": "stan-users-guide/sparse-ragged.html",
-    "title": "Sparse and Ragged Data Structures",
+    "objectID": "stan-users-guide/time-series.html#autoregressive.section",
+    "href": "stan-users-guide/time-series.html#autoregressive.section",
+    "title": "Time-Series Models",
     "section": "",
-    "text": "Stan does not directly support either sparse or ragged data structures, though both can be accommodated with some programming effort. The sparse matrices chapter introduces a special-purpose sparse matrix times dense vector multiplication, which should be used where applicable; this chapter covers more general data structures.\n\n\nCoding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the item-response model section. There are \\(J\\) students and \\(K\\) questions, and if every student answers every question, then it is practical to declare the data as a \\(J \\times K\\) array of answers.\ndata {\n  int&lt;lower=1&gt; J;\n  int&lt;lower=1&gt; K;\n  array[J, K] int&lt;lower=0, upper=1&gt; y;\n  // ...\nmodel {\n  for (j in 1:J) {\n    for (k in 1:K) {\n      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));\n    }\n  }\n  // ...\n}\nWhen not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.\nThe following missing data example shows an example with \\(J=3\\) and \\(K=4\\), with missing responses shown as NA, as in R.\n\\[\\begin{equation*}\ny\n=\n\\left[\n\\begin{array}{cccc}\n0 & 1 & \\mbox{NA} & 1\n\\\\\n0 & \\mbox{NA} & \\mbox{NA} & 1\n\\\\\n\\mbox{NA} & 0 & \\mbox{NA} & \\mbox{NA}\n\\end{array}\n\\right]\n\\end{equation*}\\]\nThere is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns \\(jj\\) and \\(kk\\) used for the indexes (following Gelman and Hill (2007)), the 2-D array \\(y\\) is recoded as a table. The number of rows in the table equals the number of defined array elements, here \\(y_{1,1} = 0\\), \\(y_{1,2} = 1\\), up to \\(y_{3,2} = 1\\). As the array becomes larger and sparser, the long form becomes the more economical encoding.\n\n\n\njj\nkk\ny\n\n\n\n\n1\n1\n0\n\n\n1\n2\n1\n\n\n1\n4\n1\n\n\n2\n1\n0\n\n\n2\n4\n1\n\n\n3\n2\n0\n\n\n\nLetting \\(N\\) be the number of \\(y\\) that are defined, here \\(N=6\\), the data and model can be formulated as follows.\ndata {\n  // ...\n  int&lt;lower=1&gt; N;\n  array[N] int&lt;lower=1, upper=J&gt; jj;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  // ...\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(delta[kk[n]]\n                           * (alpha[jj[n]] - beta[kk[n]]));\n  }\n  // ...\n}\nIn the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.\n\n\n\nRagged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.\nA general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.\nFor example, consider a data structure for three groups, each of which has a different number of observations.\n\n\n\n\n\n\n\\(y_1 =  \\left[1.3 \\ \\ 2.4 \\ \\ 0.9\\right]\\\\\\) \\(y_2 = \\left[-1.8 \\ \\ -0.1\\right]\\\\\\) \\(y_3 = \\left[12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7\\right]\\)\n\n\n\\(z = [1.3 \\ \\ 2.4 \\ \\ 0.9 \\ \\ -1.8 \\ \\ -0.1 \\ \\ 12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7]\\\\\\) \\(s  =  \\{ 3 \\ \\ 2 \\ \\ 4 \\}\\)\n\n\n\nOn the left is the definition of a ragged data structure \\(y\\) with three rows of different sizes (\\(y_1\\) is size 3, \\(y_2\\) size 2, and \\(y_3\\) size 4). On the right is an example of how to code the data in Stan, using a single vector \\(z\\) to hold all the values and a separate array of integers \\(s\\) to hold the group row sizes. In this example, \\(y_1 = z_{1:3}\\), \\(y_2 = z_{4:5}\\), and \\(y_3 = z_{6:9}\\).\nSuppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood \\[\\begin{equation*}\n\\sum_{n=1}^3 \\log \\textsf{normal}(y_n \\mid \\mu_n, \\sigma).\n\\end{equation*}\\] There’s no direct way to encode this in Stan.\nA full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.\nThe model can then be coded up using slicing operations as follows.\ndata {\n  int&lt;lower=0&gt; N;   // # observations\n  int&lt;lower=0&gt; K;   // # of groups\n  vector[N] y;      // observations\n  array[K] int s;   // group sizes\n  // ...\n}\nmodel {\n  int pos;\n  pos = 1;\n  for (k in 1:K) {\n    segment(y, pos, s[k]) ~ normal(mu[k], sigma);\n    pos = pos + s[k];\n  }\nThis coding allows for efficient vectorization, which is worth the copy cost entailed by the segment() vector slicing operation.",
+    "text": "A first-order autoregressive model (AR(1)) with normal noise takes each point \\(y_n\\) in a sequence \\(y\\) to be generated according to \\[\ny_n \\sim \\textsf{normal}(\\alpha + \\beta y_{n-1}, \\sigma).\n\\]\nThat is, the expected value of \\(y_n\\) is \\(\\alpha + \\beta y_{n-1}\\), with noise scaled as \\(\\sigma\\).\n\n\nWith improper flat priors on the regression coefficients \\(\\alpha\\) and \\(\\beta\\) and on the positively-constrained noise scale (\\(\\sigma\\)), the Stan program for the AR(1) model is as follows.1\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  for (n in 2:N) {\n    y[n] ~ normal(alpha + beta * y[n-1], sigma);\n  }\n}\nThe first observed data point, y[1], is not modeled here because there is nothing to condition on; instead, it acts to condition y[2]. This model also uses an improper prior for sigma, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in y over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of y is available.\n\n\nAlthough perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner\nmodel {\n  y[2:N] ~ normal(alpha + beta * y[1:(N - 1)], sigma);\n}\nThe left-hand side slicing operation pulls out the last \\(N-1\\) elements and the right-hand side version pulls out the first \\(N-1\\).\n\n\n\n\nProper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-\\(t\\) distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.\nTo enforce the estimation of a stationary AR(1) process, the slope coefficient beta may be constrained with bounds as follows.\nreal&lt;lower=-1, upper=1&gt; beta;\nIn practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of beta near zero.\n\n\n\nExtending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient gamma and the following model statement.\nfor (n in 3:N) {\n  y[n] ~ normal(alpha + beta*y[n-1] + gamma*y[n-2], sigma);\n}\n\n\n\nA general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] real y;\n}\nparameters {\n  real alpha;\n  array[K] real beta;\n  real sigma;\n}\nmodel {\n  for (n in (K+1):N) {\n    real mu = alpha;\n    for (k in 1:K) {\n      mu += beta[k] * y[n-k];\n    }\n    y[n] ~ normal(mu, sigma);\n  }\n}\n\n\n\nEconometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model (Engle 1982). Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.\nThe ARCH(1) model is typically presented as the following sequence of equations, where \\(r_t\\) is the observed return at time point \\(t\\) and \\(\\mu\\), \\(\\alpha_0\\), and \\(\\alpha_1\\) are unknown regression coefficient parameters.\n\\[\\begin{align*}\nr_t        &=  \\mu + a_t \\\\\na_t        &=  \\sigma_t \\epsilon_t \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\sigma^2_t &=  \\alpha_0 + \\alpha_1 a_{t-1}^2\n\\end{align*}\\]\nIn order to ensure the noise terms \\(\\sigma^2_t\\) are positive, the scale coefficients are constrained to be positive, \\(\\alpha_0, \\alpha_1\n&gt; 0\\). To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., \\(\\alpha_1 &lt; 1\\).2\nThe ARCH(1) model may be coded directly in Stan as follows.\ndata {\n  int&lt;lower=0&gt; T;                // number of time points\n  array[T] real r;               // return at time t\n}\nparameters {\n  real mu;                       // average return\n  real&lt;lower=0&gt; alpha0;          // noise intercept\n  real&lt;lower=0, upper=1&gt; alpha1; // noise slope\n}\nmodel {\n  for (t in 2:T) {\n    r[t] ~ normal(mu, sqrt(alpha0 + alpha1\n                                    * pow(r[t - 1] - mu,2)));\n  }\n}\nThe loop in the model is defined so that the return at time \\(t=1\\) is not modeled; the model in the next section shows how to model the return at \\(t=1\\). The model can be vectorized to be more efficient; the model in the next section provides an example.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Sparse and Ragged Data Structures"
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/sparse-ragged.html#sparse-data-structures",
-    "href": "stan-users-guide/sparse-ragged.html#sparse-data-structures",
-    "title": "Sparse and Ragged Data Structures",
+    "objectID": "stan-users-guide/time-series.html#modeling-temporal-heteroscedasticity",
+    "href": "stan-users-guide/time-series.html#modeling-temporal-heteroscedasticity",
+    "title": "Time-Series Models",
     "section": "",
-    "text": "Coding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the item-response model section. There are \\(J\\) students and \\(K\\) questions, and if every student answers every question, then it is practical to declare the data as a \\(J \\times K\\) array of answers.\ndata {\n  int&lt;lower=1&gt; J;\n  int&lt;lower=1&gt; K;\n  array[J, K] int&lt;lower=0, upper=1&gt; y;\n  // ...\nmodel {\n  for (j in 1:J) {\n    for (k in 1:K) {\n      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));\n    }\n  }\n  // ...\n}\nWhen not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.\nThe following missing data example shows an example with \\(J=3\\) and \\(K=4\\), with missing responses shown as NA, as in R.\n\\[\\begin{equation*}\ny\n=\n\\left[\n\\begin{array}{cccc}\n0 & 1 & \\mbox{NA} & 1\n\\\\\n0 & \\mbox{NA} & \\mbox{NA} & 1\n\\\\\n\\mbox{NA} & 0 & \\mbox{NA} & \\mbox{NA}\n\\end{array}\n\\right]\n\\end{equation*}\\]\nThere is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns \\(jj\\) and \\(kk\\) used for the indexes (following Gelman and Hill (2007)), the 2-D array \\(y\\) is recoded as a table. The number of rows in the table equals the number of defined array elements, here \\(y_{1,1} = 0\\), \\(y_{1,2} = 1\\), up to \\(y_{3,2} = 1\\). As the array becomes larger and sparser, the long form becomes the more economical encoding.\n\n\n\njj\nkk\ny\n\n\n\n\n1\n1\n0\n\n\n1\n2\n1\n\n\n1\n4\n1\n\n\n2\n1\n0\n\n\n2\n4\n1\n\n\n3\n2\n0\n\n\n\nLetting \\(N\\) be the number of \\(y\\) that are defined, here \\(N=6\\), the data and model can be formulated as follows.\ndata {\n  // ...\n  int&lt;lower=1&gt; N;\n  array[N] int&lt;lower=1, upper=J&gt; jj;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  // ...\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(delta[kk[n]]\n                           * (alpha[jj[n]] - beta[kk[n]]));\n  }\n  // ...\n}\nIn the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.",
+    "text": "A set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.\n\n\nThe basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time \\(t-1\\) as a predictor of volatility at time \\(t\\), defining \\[\n\\sigma^2_t = \\alpha_0 + \\alpha_1 a^2_{t-1} + \\beta_1 \\sigma^2_{t-1}.\n\\]\nTo ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy \\(\\alpha_0, \\alpha_1,\n\\beta_1 &gt; 0\\) and the slopes \\(\\alpha_1 + \\beta_1 &lt; 1\\).\ndata {\n  int&lt;lower=0&gt; T;\n  array[T] real r;\n  real&lt;lower=0&gt; sigma1;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; alpha0;\n  real&lt;lower=0, upper=1&gt; alpha1;\n  real&lt;lower=0, upper=(1-alpha1)&gt; beta1;\n}\ntransformed parameters {\n  array[T] real&lt;lower=0&gt; sigma;\n  sigma[1] = sigma1;\n  for (t in 2:T) {\n    sigma[t] = sqrt(alpha0\n                     + alpha1 * pow(r[t - 1] - mu, 2)\n                     + beta1 * pow(sigma[t - 1], 2));\n  }\n}\nmodel {\n  r ~ normal(mu, sigma);\n}\nTo get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value sigma1 for the scale of the noise at \\(t = 1\\).\nThe constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on beta1 depends on the value of alpha1.\nA transformed parameter array of non-negative values sigma is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept alpha0, a slope alpha1 for the squared difference in return from the mean at the previous time, and a slope beta1 for the previous noise scale squared. Finally, the whole regression is inside the sqrt function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.\nWith the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because r and sigma are of length T, all of the data are modeled directly.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Sparse and Ragged Data Structures"
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/sparse-ragged.html#ragged-data-structs.section",
-    "href": "stan-users-guide/sparse-ragged.html#ragged-data-structs.section",
-    "title": "Sparse and Ragged Data Structures",
+    "objectID": "stan-users-guide/time-series.html#moving-average-models",
+    "href": "stan-users-guide/time-series.html#moving-average-models",
+    "title": "Time-Series Models",
     "section": "",
-    "text": "Ragged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.\nA general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.\nFor example, consider a data structure for three groups, each of which has a different number of observations.\n\n\n\n\n\n\n\\(y_1 =  \\left[1.3 \\ \\ 2.4 \\ \\ 0.9\\right]\\\\\\) \\(y_2 = \\left[-1.8 \\ \\ -0.1\\right]\\\\\\) \\(y_3 = \\left[12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7\\right]\\)\n\n\n\\(z = [1.3 \\ \\ 2.4 \\ \\ 0.9 \\ \\ -1.8 \\ \\ -0.1 \\ \\ 12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7]\\\\\\) \\(s  =  \\{ 3 \\ \\ 2 \\ \\ 4 \\}\\)\n\n\n\nOn the left is the definition of a ragged data structure \\(y\\) with three rows of different sizes (\\(y_1\\) is size 3, \\(y_2\\) size 2, and \\(y_3\\) size 4). On the right is an example of how to code the data in Stan, using a single vector \\(z\\) to hold all the values and a separate array of integers \\(s\\) to hold the group row sizes. In this example, \\(y_1 = z_{1:3}\\), \\(y_2 = z_{4:5}\\), and \\(y_3 = z_{6:9}\\).\nSuppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood \\[\\begin{equation*}\n\\sum_{n=1}^3 \\log \\textsf{normal}(y_n \\mid \\mu_n, \\sigma).\n\\end{equation*}\\] There’s no direct way to encode this in Stan.\nA full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.\nThe model can then be coded up using slicing operations as follows.\ndata {\n  int&lt;lower=0&gt; N;   // # observations\n  int&lt;lower=0&gt; K;   // # of groups\n  vector[N] y;      // observations\n  array[K] int s;   // group sizes\n  // ...\n}\nmodel {\n  int pos;\n  pos = 1;\n  for (k in 1:K) {\n    segment(y, pos, s[k]) ~ normal(mu[k], sigma);\n    pos = pos + s[k];\n  }\nThis coding allows for efficient vectorization, which is worth the copy cost entailed by the segment() vector slicing operation.",
+    "text": "A moving average model uses previous errors as predictors for future outcomes. For a moving average model of order \\(Q\\), \\(\\mbox{MA}(Q)\\), there is an overall mean parameter \\(\\mu\\) and regression coefficients \\(\\theta_q\\) for previous error terms. With \\(\\epsilon_t\\) being the noise at time \\(t\\), the model for outcome \\(y_t\\) is defined by \\[\ny_t = \\mu + \\theta_1 \\epsilon_{t-1} + \\dotsb + \\theta_Q \\epsilon_{t-Q}\n+ \\epsilon_t,\n\\] with the noise term \\(\\epsilon_t\\) for outcome \\(y_t\\) modeled as normal, \\[\n\\epsilon_t \\sim \\textsf{normal}(0,\\sigma).\n\\] In a proper Bayesian model, the parameters \\(\\mu\\), \\(\\theta\\), and \\(\\sigma\\) must all be given priors.\n\n\nAn \\(\\mbox{MA}(2)\\) model can be coded in Stan as follows.\ndata {\n  int&lt;lower=3&gt; T;          // number of observations\n  vector[T] y;             // observation at time T\n}\nparameters {\n  real mu;                 // mean\n  real&lt;lower=0&gt; sigma;     // error scale\n  vector[2] theta;         // lag coefficients\n}\ntransformed parameters {\n  vector[T] epsilon;       // error terms\n  epsilon[1] = y[1] - mu;\n  epsilon[2] = y[2] - mu - theta[1] * epsilon[1];\n  for (t in 3:T) {\n    epsilon[t] = ( y[t] - mu\n                    - theta[1] * epsilon[t - 1]\n                    - theta[2] * epsilon[t - 2] );\n  }\n}\nmodel {\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 3:T) {\n    y[t] ~ normal(mu\n                  + theta[1] * epsilon[t - 1]\n                  + theta[2] * epsilon[t - 2],\n                  sigma);\n  }\n}\nThe error terms \\(\\epsilon_t\\) are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to \\(y_n\\) for \\(n &gt; Q\\). In this example, the parameters are all given Cauchy (half-Cauchy for \\(\\sigma\\)) priors, although other priors can be used just as easily.\nThis model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the \\(\\epsilon_t\\) could also be sped up by using a dot product instead of a loop.\n\n\n\nA general \\(\\mbox{MA}(Q)\\) model with a vectorized distribution statement may be defined as follows.\ndata {\n  int&lt;lower=0&gt; Q;       // num previous noise terms\n  int&lt;lower=3&gt; T;       // num observations\n  vector[T] y;          // observation at time t\n}\nparameters {\n  real mu;              // mean\n  real&lt;lower=0&gt; sigma;  // error scale\n  vector[Q] theta;      // error coeff, lag -t\n}\ntransformed parameters {\n  vector[T] epsilon;    // error term at time t\n  for (t in 1:T) {\n    epsilon[t] = y[t] - mu;\n    for (q in 1:min(t - 1, Q)) {\n      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];\n    }\n  }\n}\nmodel {\n  vector[T] eta;\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 1:T) {\n    eta[t] = mu;\n    for (q in 1:min(t - 1, Q)) {\n      eta[t] = eta[t] + theta[q] * epsilon[t - q];\n    }\n  }\n  y ~ normal(eta, sigma);\n}\nHere all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Sparse and Ragged Data Structures"
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/reparameterization.html",
-    "href": "stan-users-guide/reparameterization.html",
-    "title": "Reparameterization and Change of Variables",
+    "objectID": "stan-users-guide/time-series.html#autoregressive-moving-average-models",
+    "href": "stan-users-guide/time-series.html#autoregressive-moving-average-models",
+    "title": "Time-Series Models",
     "section": "",
-    "text": "Stan supports a direct encoding of reparameterizations. Stan also supports changes of variables by directly incrementing the log probability accumulator with the log Jacobian of the transform.\n\n\nA Bayesian posterior is technically a probability measure, which is a parameterization-invariant, abstract mathematical object.1\nStan’s modeling language, on the other hand, defines a probability density, which is a non-unique, parameterization-dependent function in \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^{+}\\). In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.\nAs pointed out by Gelman (2004) in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not just that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.\n\n\n\nReparameterizations may be implemented directly using the transformed parameters block or just in the model block.\n\n\nThe beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.\n\n\nFor example, the Beta distribution is parameterized by two positive count parameters \\(\\alpha, \\beta &gt; 0\\). The following example illustrates a hierarchical Stan model with a vector of parameters theta are drawn i.i.d. for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n  // ...\n}\nmodel {\n  alpha ~ ...\n  beta ~ ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nIt is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter \\[\n\\phi = \\alpha / (\\alpha + \\beta)\n\\] and total count parameter \\[\n\\lambda = \\alpha + \\beta.\n\\] Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with \\(p(\\lambda) \\propto \\lambda^{-2.5}\\).\nparameters {\n  real&lt;lower=0, upper=1&gt; phi;\n  real&lt;lower=0.1&gt; lambda;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0&gt; alpha = lambda * phi;\n  real&lt;lower=0&gt; beta = lambda * (1 - phi);\n  // ...\n}\nmodel {\n  phi ~ beta(1, 1); // uniform on phi, could drop\n  lambda ~ pareto(0.1, 1.5);\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nThe new parameters, phi and lambda, are declared in the parameters block and the parameters for the Beta distribution, alpha and beta, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.\nmodel {\n  real alpha = lambda * phi;\n  real beta = lambda * (1 - phi);\n  // ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nWith vectorization, this could be expressed more compactly and efficiently as follows.\nmodel {\n  theta ~ beta(lambda * phi, lambda * (1 - phi));\n  // ...\n}\nIf the variables alpha and beta are of interest, they can be defined in the transformed parameter block and then used in the model.\n\n\n\nBecause the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, alpha and beta have the correct posterior distribution.\n\n\n\nThe same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are \\(K &gt; 0\\) dimensions being considered (\\(K=1\\) is trivial and \\(K=2\\) reduces to the beta distribution case). The traditional prior is\nparameters {\n  vector[K] alpha;\n  array[N] simplex[K] theta;\n  // ...\n}\nmodel {\n  alpha ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n}\nThis provides essentially \\(K\\) degrees of freedom, one for each dimension of alpha, and it is not obvious how to specify a reasonable prior for alpha.\nAn alternative coding is to use the mean, which is a simplex, and a total count.\nparameters {\n  simplex[K] phi;\n  real&lt;lower=0&gt; kappa;\n  array[N] simplex[K] theta;\n  // ...\n}\ntransformed parameters {\n  vector[K] alpha = kappa * phi;\n  // ...\n}\nmodel {\n  phi ~ // ...\n  kappa ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n  // ...\n}\nNow it is much easier to formulate priors, because phi is the expected value of theta and kappa (minus K) is the strength of the prior mean measured in number of prior observations.\n\n\n\n\nIf the variable \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\operatorname{logit}(u)\\) is distributed as \\(\\textsf{logistic}(0, 1)\\). This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in \\((0, 1)\\) to a logistically-distributed quantity.\nThings work the same way for the probit case: if \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\Phi^{-1}(u)\\) has a \\(\\textsf{normal}(0, 1)\\) distribution. The other way around, if \\(v\\) has a \\(\\textsf{normal}(0, 1)\\) distribution, then \\(\\Phi(v)\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution.\nIn order to use the probit and logistic as priors on variables constrained to \\((0, 1)\\), create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a \\((0, 1)\\)-constrained parameter theta and gives it a beta prior, then uses it as a parameter in a distribution (here using foo as a placeholder).\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nmodel {\n  theta ~ beta(a, b);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIf the variables a and b are one, then this imposes a uniform distribution theta. If a and b are both less than one, then the density on theta has a U shape, whereas if they are both greater than one, the density of theta has an inverted-U or more bell-like shape.\nRoughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,\nparameters {\n  real theta_raw;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0, upper=1&gt; theta = inv_logit(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ logistic(mu, sigma);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIn this model, an unconstrained parameter theta_raw gets a logistic prior, and then the transformed parameter theta is defined to be the inverse logit of theta_raw. In this parameterization, inv_logit(mu) is the mean of the implied prior on theta. The prior distribution on theta will be flat if sigma is one and mu is zero, and will be U-shaped if sigma is larger than one and bell shaped if sigma is less than one.\nWhen moving from a variable in \\((0, 1)\\) to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.\nparameters {\n  simplex[K] theta;\n  // ...\n}\nmodel {\n  theta ~ dirichlet(a);\n  // ...\n  y ~ foo(theta);\n}\nNow a is a vector with K rows, but it has the same shape properties as the pair a and b for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector \\([a b]^{\\top}\\). To formulate an unconstrained prior, the exact same strategy works as for the beta.\nparameters {\n  vector[K] theta_raw;\n  // ...\n}\ntransformed parameters {\n  simplex[K] theta = softmax(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by softmax(mu), but we have additional control of covariance through L_Sigma at the expense of having on the order of \\(K^2\\) parameters in the prior rather than order \\(K\\). If no covariance is desired, the number of parameters can be reduced back to \\(K\\) using a vectorized normal distribution as follows.\ntheta_raw ~ normal(mu, sigma);\nwhere either or both of mu and sigma can be vectors.\n\n\n\n\nChanges of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable \\(y &gt; 0\\) whose logarithm \\(\\log y\\) has a normal distribution. The distribution is being assigned to \\(\\log y\\).\nThe change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^N\\).\nThe probability must be scaled by a Jacobian adjustment equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.\nIn the case of log normals, if \\(y\\)’s logarithm is normal with mean \\(\\mu\\) and deviation \\(\\sigma\\), then the distribution of \\(y\\) is given by \\[\np(y)\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\left| \\frac{d}{dy} \\log y \\right|\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\frac{1}{y}.\n\\] Stan works on the log scale to prevent underflow, where \\[\n\\log p(y)\n=\n\\log \\textsf{normal}(\\log y \\mid \\mu, \\sigma) - \\log y.\n\\]\nIn Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.2\nparameters {\n  real&lt;lower=0&gt; y;\n  // ...\n}\nmodel {\n  log(y) ~ normal(mu, sigma);\n  target += -log(y);\n  // ...\n}\nIt is important, as always, to declare appropriate constraints on parameters; here y is constrained to be positive.\nIt would be slightly more efficient to define a local variable for the logarithm, as follows.\nmodel {\n  real log_y;\n  log_y = log(y);\n  log_y ~ normal(mu, sigma);\n  target += -log_y;\n  // ...\n}\nIf y were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.\n\n\nThis section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.\nIt does not matter whether the probability function is expressed using a distribution statement, such as\nlog(y) ~ normal(mu, sigma);\nor as an increment to the log probability function, as in\ntarget += normal_lpdf(log(y) | mu, sigma);\n\n\nLike the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.\nThe transform based approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y_inv;\n  y_inv = 1 / y;\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe change-of-variables approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y_inv;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y;\n  y = 1 / y_inv;  // change variables\n  jacobian += -2 * log(y_inv); // Jacobian adjustment\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is\n\\[\n\\log \\left| \\frac{d}{du} \\left( \\frac{1}{u} \\right) \\right|\n= \\log \\left| - u^{-2} \\right|\n= \\log u^{-2}\n=  -2 \\log u.\n\\]\n\n\n\n\nIn the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.\nparameters {\n  vector[K] u;      // multivariate parameter\n   // ...\n}\ntransformed parameters {\n  vector[K] v;     // transformed parameter\n  matrix[K, K] J;   // Jacobian matrix of transform\n  // ... compute v as a function of u ...\n  // ... compute J[m, n] = d.v[m] / d.u[n] ...\n  jacobian += log(abs(determinant(J)));\n  // ...\n}\nmodel {\n  v ~ // ...\n  // ...\n}\nIf the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.\nIn many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element v[k] of the transformed parameter vector only depends on elements u[1], …, u[k] of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.\ntransformed parameters {\n  // ...\n  vector[K] J_diag;  // diagonals of Jacobian matrix\n  // ...\n  // ... compute J[k, k] = d.v[k] / d.u[k] ...\n  jacobian += sum(log(J_diag));\n  // ...\n}\n\n\n\n\nStan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.\n\n\nFor example, suppose there is a vector parameter \\(\\alpha\\) with a vector \\(L\\) of lower bounds. The simplest way to deal with this if \\(L\\) is a constant is to shift a lower-bounded parameter.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L&gt;[N] alpha_raw;\n  // ...\n}\nThe above is equivalent to manually calculating the vector bounds by the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + alpha_raw;\n  // ...\n}\nThe Jacobian for adding a constant is one, so its log drops out of the log density.\nEven if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from \\((L, \\alpha_{\\textrm{raw}})\\) to \\((L + \\alpha_{\\textrm{raw}}, \\alpha_{\\textrm{raw}})\\) produces a Jacobian derivative matrix with a unit determinant.\nIt’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  vector&lt;lower=L + alpha_raw&gt;[N] alpha;\n  // ...\n}\nThis is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + exp(alpha_raw);\n  jacobian += sum(alpha_raw); // log Jacobian\n  // ...\n}\nmodel {\n  // ...\n}\nThe adjustment in the log Jacobian determinant of the transform mapping \\(\\alpha_{\\textrm{raw}}\\) to \\(\\alpha = L + \\exp(\\alpha_{\\textrm{raw}})\\). The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here \\(L\\) can even be a vector containing parameters that don’t depend on \\(\\alpha_{\\textrm{raw}}\\); if the bounds do depend on \\(\\alpha_{\\textrm{raw}}\\) then a revised Jacobian needs to be calculated taking into account the dependencies.\n\n\n\nSuppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to \\((0, 1)\\). This is easily accomplished as the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L, upper=U&gt;[N] alpha;\n  // ...\n}\nThe same may be accomplished by manually constructing the transform as follows.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + (U - L) .* alpha_raw;\n}\nThe expression U - L is multiplied by alpha_raw elementwise to produce a vector of variables in \\((0, U-L)\\), then adding \\(L\\) results in a variable ranging between \\((L, U)\\).\nIn this case, it is important that \\(L\\) and \\(U\\) are constants, otherwise a Jacobian would be required when multiplying by \\(U - L\\).",
+    "text": "Autoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.\ndata {\n  int&lt;lower=1&gt; T;            // num observations\n  array[T] real y;                 // observed outputs\n}\nparameters {\n  real mu;                   // mean coeff\n  real phi;                  // autoregression coeff\n  real theta;                // moving avg coeff\n  real&lt;lower=0&gt; sigma;       // noise scale\n}\nmodel {\n  vector[T] nu;              // prediction for time t\n  vector[T] err;             // error for time t\n  nu[1] = mu + phi * mu;     // assume err[0] == 0\n  err[1] = y[1] - nu[1];\n  for (t in 2:T) {\n    nu[t] = mu + phi * y[t - 1] + theta * err[t - 1];\n    err[t] = y[t] - nu[t];\n  }\n  mu ~ normal(0, 10);        // priors\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err ~ normal(0, sigma);    // error model\n}\nThe data are declared in the same way as the other time-series regressions and the parameters are documented in the code.\nIn the model block, the local vector nu stores the predictions and err the errors. These are computed similarly to the errors in the moving average models described in the previous section.\nThe priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.\nOften in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring err as a transformed parameter, then defining it the same way as in the model above. The vector nu could still be a local variable, only now it will be in the transformed parameter block.\nWayne Folta suggested encoding the model without local vector variables as follows.\nmodel {\n  real err;\n  mu ~ normal(0, 10);\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err = y[1] - (mu + phi * mu);\n  err ~ normal(0, sigma);\n  for (t in 2:T) {\n    err = y[t] - (mu + phi * y[t - 1] + theta * err);\n    err ~ normal(0, sigma);\n  }\n}\nThis approach to ARMA models illustrates how local variables, such as err in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.\nBoth encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just err.\n\n\nMA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint3\nreal&lt;lower=-1, upper=1&gt; theta;\nWhen the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (theta, phi) outside the \\([-1,1]\\) interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).\nFurther, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.\nreal&lt;lower=-1, upper=1&gt; phi;",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Reparameterization and Change of Variables"
+      "Example Models",
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/reparameterization.html#theoretical-and-practical-background",
-    "href": "stan-users-guide/reparameterization.html#theoretical-and-practical-background",
-    "title": "Reparameterization and Change of Variables",
+    "objectID": "stan-users-guide/time-series.html#stochastic-volatility-models",
+    "href": "stan-users-guide/time-series.html#stochastic-volatility-models",
+    "title": "Time-Series Models",
     "section": "",
-    "text": "A Bayesian posterior is technically a probability measure, which is a parameterization-invariant, abstract mathematical object.1\nStan’s modeling language, on the other hand, defines a probability density, which is a non-unique, parameterization-dependent function in \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^{+}\\). In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.\nAs pointed out by Gelman (2004) in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not just that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.",
+    "text": "Stochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time (Kim, Shephard, and Chib 1998). The data consist of mean corrected (i.e., centered) returns \\(y_t\\) on an underlying asset at \\(T\\) equally spaced time points. Kim et al. formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter \\(h_t\\) for the log volatility, along with parameters \\(\\mu\\) for the mean log volatility, and \\(\\phi\\) for the persistence of the volatility term. The variable \\(\\epsilon_t\\) represents the white-noise shock (i.e., multiplicative error) on the asset return at time \\(t\\), whereas \\(\\delta_t\\) represents the shock on volatility at time \\(t\\). \\[\\begin{align*}\ny_t        &=    \\epsilon_t \\exp(h_t / 2) \\\\\nh_{t+1}    &=    \\mu + \\phi (h_t - \\mu) + \\delta_t \\sigma \\\\\nh_1        &\\sim \\textsf{normal}\\left( \\mu, \\frac{\\sigma}{\\sqrt{1 - \\phi^2}} \\right) \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\delta_t   &\\sim \\textsf{normal}(0,1)\n\\end{align*}\\]\nRearranging the first line, \\(\\epsilon_t = y_t \\exp(-h_t / 2)\\), allowing the distribution for \\(y_t\\) to be written as \\[\ny_t \\sim \\textsf{normal}(0,\\exp(h_t/2)).\n\\] The recurrence equation for \\(h_{t+1}\\) may be combined with the scaling of \\(\\delta_t\\) to yield the distribution \\[\nh_t \\sim \\mathsf{normal}(\\mu + \\phi(h_{t-1} - \\mu), \\sigma).\n\\] This formulation can be directly encoded, as shown in the following Stan model.\ndata {\n  int&lt;lower=0&gt; T;   // # time points (equally spaced)\n  vector[T] y;      // mean corrected return at time t\n}\nparameters {\n  real mu;                     // mean log volatility\n  real&lt;lower=-1, upper=1&gt; phi; // persistence of volatility\n  real&lt;lower=0&gt; sigma;         // white noise shock scale\n  vector[T] h;                 // log volatility at time t\n}\nmodel {\n  phi ~ uniform(-1, 1);\n  sigma ~ cauchy(0, 5);\n  mu ~ cauchy(0, 10);\n  h[1] ~ normal(mu, sigma / sqrt(1 - phi * phi));\n  for (t in 2:T) {\n    h[t] ~ normal(mu + phi * (h[t - 1] -  mu), sigma);\n  }\n  for (t in 1:T) {\n    y[t] ~ normal(0, exp(h[t] / 2));\n  }\n}\nCompared to the Kim et al. formulation, the Stan model adds priors for the parameters \\(\\phi\\), \\(\\sigma\\), and \\(\\mu\\). The shock terms \\(\\epsilon_t\\) and \\(\\delta_t\\) do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.\nThe posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with \\(\\mu = -1.02\\), \\(\\phi = 0.95\\), and \\(\\sigma = 0.25\\) leads to 95% posterior intervals for \\(\\mu\\) of \\((-1.23, -0.54)\\), for \\(\\phi\\) of \\((0.82, 0.98)\\), and for \\(\\sigma\\) of \\((0.16, 0.38)\\).\nThe NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in (Hoffman and Gelman 2014). Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of \\(T\\).\nIt is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return \\(y\\) is easily vectorized to\ny ~ normal(0, exp(h / 2));\nThis speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter h_std to be declared instead of h.\nparameters {\n  // ...\n  vector[T] h_std;  // std log volatility time t\n}\nThe original value of h is then defined in a transformed parameter block.\ntransformed parameters {\n  vector[T] h = h_std * sigma;  // now h ~ normal(0, sigma)\n  h[1] /= sqrt(1 - phi * phi);  // rescale h[1]\n  h += mu;\n  for (t in 2:T) {\n    h[t] += phi * (h[t - 1] - mu);\n  }\n}\nThe first assignment rescales h_std to have a \\(\\textsf{normal}(0,\\sigma)\\) distribution and temporarily assigns it to h. The second assignment rescales h[1] so that its prior differs from that of h[2] through h[T]. The next assignment supplies a mu offset, so that h[2] through h[T] are now distributed \\(\\textsf{normal}(\\mu,\\sigma)\\); note that this shift must be done after the rescaling of h[1]. The final loop adds in the moving average so that h[2] through h[T] are appropriately modeled relative to phi and mu.\nAs a final improvement, the distribution statements for h[1] to h[T] are replaced with a single vectorized standard normal distribution statement.\nmodel {\n  // ...\n  h_std ~ std_normal();\n}\nAlthough the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Reparameterization and Change of Variables"
+      "Example Models",
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/reparameterization.html#reparameterizations",
-    "href": "stan-users-guide/reparameterization.html#reparameterizations",
-    "title": "Reparameterization and Change of Variables",
+    "objectID": "stan-users-guide/time-series.html#hmms.section",
+    "href": "stan-users-guide/time-series.html#hmms.section",
+    "title": "Time-Series Models",
     "section": "",
-    "text": "Reparameterizations may be implemented directly using the transformed parameters block or just in the model block.\n\n\nThe beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.\n\n\nFor example, the Beta distribution is parameterized by two positive count parameters \\(\\alpha, \\beta &gt; 0\\). The following example illustrates a hierarchical Stan model with a vector of parameters theta are drawn i.i.d. for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n  // ...\n}\nmodel {\n  alpha ~ ...\n  beta ~ ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nIt is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter \\[\n\\phi = \\alpha / (\\alpha + \\beta)\n\\] and total count parameter \\[\n\\lambda = \\alpha + \\beta.\n\\] Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with \\(p(\\lambda) \\propto \\lambda^{-2.5}\\).\nparameters {\n  real&lt;lower=0, upper=1&gt; phi;\n  real&lt;lower=0.1&gt; lambda;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0&gt; alpha = lambda * phi;\n  real&lt;lower=0&gt; beta = lambda * (1 - phi);\n  // ...\n}\nmodel {\n  phi ~ beta(1, 1); // uniform on phi, could drop\n  lambda ~ pareto(0.1, 1.5);\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nThe new parameters, phi and lambda, are declared in the parameters block and the parameters for the Beta distribution, alpha and beta, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.\nmodel {\n  real alpha = lambda * phi;\n  real beta = lambda * (1 - phi);\n  // ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nWith vectorization, this could be expressed more compactly and efficiently as follows.\nmodel {\n  theta ~ beta(lambda * phi, lambda * (1 - phi));\n  // ...\n}\nIf the variables alpha and beta are of interest, they can be defined in the transformed parameter block and then used in the model.\n\n\n\nBecause the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, alpha and beta have the correct posterior distribution.\n\n\n\nThe same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are \\(K &gt; 0\\) dimensions being considered (\\(K=1\\) is trivial and \\(K=2\\) reduces to the beta distribution case). The traditional prior is\nparameters {\n  vector[K] alpha;\n  array[N] simplex[K] theta;\n  // ...\n}\nmodel {\n  alpha ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n}\nThis provides essentially \\(K\\) degrees of freedom, one for each dimension of alpha, and it is not obvious how to specify a reasonable prior for alpha.\nAn alternative coding is to use the mean, which is a simplex, and a total count.\nparameters {\n  simplex[K] phi;\n  real&lt;lower=0&gt; kappa;\n  array[N] simplex[K] theta;\n  // ...\n}\ntransformed parameters {\n  vector[K] alpha = kappa * phi;\n  // ...\n}\nmodel {\n  phi ~ // ...\n  kappa ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n  // ...\n}\nNow it is much easier to formulate priors, because phi is the expected value of theta and kappa (minus K) is the strength of the prior mean measured in number of prior observations.\n\n\n\n\nIf the variable \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\operatorname{logit}(u)\\) is distributed as \\(\\textsf{logistic}(0, 1)\\). This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in \\((0, 1)\\) to a logistically-distributed quantity.\nThings work the same way for the probit case: if \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\Phi^{-1}(u)\\) has a \\(\\textsf{normal}(0, 1)\\) distribution. The other way around, if \\(v\\) has a \\(\\textsf{normal}(0, 1)\\) distribution, then \\(\\Phi(v)\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution.\nIn order to use the probit and logistic as priors on variables constrained to \\((0, 1)\\), create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a \\((0, 1)\\)-constrained parameter theta and gives it a beta prior, then uses it as a parameter in a distribution (here using foo as a placeholder).\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nmodel {\n  theta ~ beta(a, b);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIf the variables a and b are one, then this imposes a uniform distribution theta. If a and b are both less than one, then the density on theta has a U shape, whereas if they are both greater than one, the density of theta has an inverted-U or more bell-like shape.\nRoughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,\nparameters {\n  real theta_raw;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0, upper=1&gt; theta = inv_logit(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ logistic(mu, sigma);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIn this model, an unconstrained parameter theta_raw gets a logistic prior, and then the transformed parameter theta is defined to be the inverse logit of theta_raw. In this parameterization, inv_logit(mu) is the mean of the implied prior on theta. The prior distribution on theta will be flat if sigma is one and mu is zero, and will be U-shaped if sigma is larger than one and bell shaped if sigma is less than one.\nWhen moving from a variable in \\((0, 1)\\) to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.\nparameters {\n  simplex[K] theta;\n  // ...\n}\nmodel {\n  theta ~ dirichlet(a);\n  // ...\n  y ~ foo(theta);\n}\nNow a is a vector with K rows, but it has the same shape properties as the pair a and b for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector \\([a b]^{\\top}\\). To formulate an unconstrained prior, the exact same strategy works as for the beta.\nparameters {\n  vector[K] theta_raw;\n  // ...\n}\ntransformed parameters {\n  simplex[K] theta = softmax(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by softmax(mu), but we have additional control of covariance through L_Sigma at the expense of having on the order of \\(K^2\\) parameters in the prior rather than order \\(K\\). If no covariance is desired, the number of parameters can be reduced back to \\(K\\) using a vectorized normal distribution as follows.\ntheta_raw ~ normal(mu, sigma);\nwhere either or both of mu and sigma can be vectors.",
+    "text": "A Hidden Markov model is a probabilistic model over \\(N\\) observations \\(y_{1:N}\\) and \\(N\\) hidden states \\(z_{1:N}\\). This models is defined by the conditional distributions \\(p(y_n \\mid z_n, \\phi)\\) and \\(p(z_n \\mid z_{n-1}, \\phi)\\). Here we make the dependency on additional model parameters \\(\\phi\\) explicit. (\\(\\phi\\) may be a vector of parameters.) The complete data likelihood is then \\[\np(y, z \\mid \\phi) = \\prod_n p(y_n \\mid z_n, \\phi) p(z_n \\mid z_{n - 1}, \\phi)\n\\] When \\(z_{1:N}\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(z\\) out.\nWhen each state \\(z\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can use Stan’s suite of HMM functions to marginalize out \\(z_{1:N}\\) and compute \\[\np(y_{1:N} \\mid \\phi) = \\int_{\\mathcal Z} p(y, z \\mid \\phi) \\text d z.\n\\] We start by defining the conditional observation distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid z_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n\\Gamma_{ij} = p(z_n = j \\mid z_{n - 1} = i, \\phi).\n\\] (This is a right-stochastic matrix.) Finally, we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n\\rho_k = p(z_0 = k \\mid \\phi).\n\\] It is common practice to set \\(\\rho\\) to be the stationary distribution of the HMM, that is \\(\\rho\\) is the first eigenvector of \\(\\Gamma\\) and solves \\(\\Gamma \\rho = \\rho\\).\nAs an example, consider a three-state model with \\(K=3\\). The observations are normally distributed conditional on the HMM states with \\[\n  y_n \\sim \\text{normal}(\\mu_k, \\sigma),\n\\] where \\(\\mu = (1, 5, 9)\\) and the standard deviation \\(\\sigma\\) is the same across all observations. The model is then\ndata {\n  int N;  // Number of observations\n  array[N] real y;\n}\n\nparameters {\n  // Rows of the transition matrix\n  array[3] simplex[3] gamma_arr;\n\n  // Initial state\n  simplex[3] rho;\n\n  // Parameters of measurement model\n  vector[3] mu;\n  real&lt;lower = 0.0&gt; sigma;\n}\n\ntransformed parameters {\n  // Build transition matrix\n  matrix[3, 3] gamma;\n  for (k in 1:3) gamma[k, ] = to_row_vector(gamma_arr[k]);\n\n  // Compute the log likelihoods in each possible state\n  matrix[3, N] log_omega;\n  for (n in 1:N) {\n    for (i in 1:3) {\n      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);\n    }\n  }\n}\n\nmodel {\n  // prior\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  \n  // no explicit prior on gamma_arr, meaning we default to a\n  // uniform prior over the simplexes.\n\n  // Increment target by log p(y | mu, sigma, Gamma, rho)\n  target += hmm_marginal(log_omega, gamma, rho);\n}\nThe last function hmm_marginal takes in all the ingredients of the HMM and computes the relevant log marginal distribution, \\(\\log p(y \\mid \\phi)\\).\nIf we desire draws from the posterior distribution of \\(z\\), we use the generated quantities block and draw, for each sample \\(\\phi\\), a sample from \\(p(z \\mid y, \\phi)\\). In effect, MCMC produces draws from \\(p(\\phi \\mid y)\\) and with the draws in generated quantities, we obtain draws from \\(p(\\phi \\mid y) p(z \\mid y, \\phi) = p(z, \\phi \\mid y)\\). It is also possible to compute the posterior probbability of each hidden state, that is \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). Averagging these probabilities over all MCMC draws, we obtain \\(\\text{Pr}(z_n = k \\mid y)\\).\ngenerated quantities {\n  array[N] int latent_states = hmm_latent_rng(log_omega, gamma, rho);\n  matrix[3, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);\n}\nhmm_hidden_state_prob returns the marginal probabilities of each state, \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). This function cannot be used to compute the joint probability \\(\\text{Pr}(z \\mid \\phi, y)\\), because such calculation requires accounting for the posterior correlation between the different components of \\(z\\). Therefore, hidden_probs should not be used to obtain posterior draws. Instead, users should rely on hmm_latent_rng.\ngenerated quantities {\n   array[N] int&lt;lower=1, upper=K&gt; z = hmm_latent_rng(...fill-in params here to match example...);\n}\nThe example in this section is derived from the more detailed case study by Ben Bales: https://mc-stan.org/users/documentation/case-studies/hmm-example.html.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Reparameterization and Change of Variables"
-    ]
+      "Example Models",
+      "Time-Series Models"
+    ]
   },
   {
-    "objectID": "stan-users-guide/reparameterization.html#changes-of-variables",
-    "href": "stan-users-guide/reparameterization.html#changes-of-variables",
-    "title": "Reparameterization and Change of Variables",
-    "section": "",
-    "text": "Changes of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable \\(y &gt; 0\\) whose logarithm \\(\\log y\\) has a normal distribution. The distribution is being assigned to \\(\\log y\\).\nThe change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^N\\).\nThe probability must be scaled by a Jacobian adjustment equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.\nIn the case of log normals, if \\(y\\)’s logarithm is normal with mean \\(\\mu\\) and deviation \\(\\sigma\\), then the distribution of \\(y\\) is given by \\[\np(y)\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\left| \\frac{d}{dy} \\log y \\right|\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\frac{1}{y}.\n\\] Stan works on the log scale to prevent underflow, where \\[\n\\log p(y)\n=\n\\log \\textsf{normal}(\\log y \\mid \\mu, \\sigma) - \\log y.\n\\]\nIn Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.2\nparameters {\n  real&lt;lower=0&gt; y;\n  // ...\n}\nmodel {\n  log(y) ~ normal(mu, sigma);\n  target += -log(y);\n  // ...\n}\nIt is important, as always, to declare appropriate constraints on parameters; here y is constrained to be positive.\nIt would be slightly more efficient to define a local variable for the logarithm, as follows.\nmodel {\n  real log_y;\n  log_y = log(y);\n  log_y ~ normal(mu, sigma);\n  target += -log_y;\n  // ...\n}\nIf y were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.\n\n\nThis section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.\nIt does not matter whether the probability function is expressed using a distribution statement, such as\nlog(y) ~ normal(mu, sigma);\nor as an increment to the log probability function, as in\ntarget += normal_lpdf(log(y) | mu, sigma);\n\n\nLike the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.\nThe transform based approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y_inv;\n  y_inv = 1 / y;\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe change-of-variables approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y_inv;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y;\n  y = 1 / y_inv;  // change variables\n  jacobian += -2 * log(y_inv); // Jacobian adjustment\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is\n\\[\n\\log \\left| \\frac{d}{du} \\left( \\frac{1}{u} \\right) \\right|\n= \\log \\left| - u^{-2} \\right|\n= \\log u^{-2}\n=  -2 \\log u.\n\\]\n\n\n\n\nIn the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.\nparameters {\n  vector[K] u;      // multivariate parameter\n   // ...\n}\ntransformed parameters {\n  vector[K] v;     // transformed parameter\n  matrix[K, K] J;   // Jacobian matrix of transform\n  // ... compute v as a function of u ...\n  // ... compute J[m, n] = d.v[m] / d.u[n] ...\n  jacobian += log(abs(determinant(J)));\n  // ...\n}\nmodel {\n  v ~ // ...\n  // ...\n}\nIf the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.\nIn many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element v[k] of the transformed parameter vector only depends on elements u[1], …, u[k] of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.\ntransformed parameters {\n  // ...\n  vector[K] J_diag;  // diagonals of Jacobian matrix\n  // ...\n  // ... compute J[k, k] = d.v[k] / d.u[k] ...\n  jacobian += sum(log(J_diag));\n  // ...\n}",
+    "objectID": "stan-users-guide/time-series.html#footnotes",
+    "href": "stan-users-guide/time-series.html#footnotes",
+    "title": "Time-Series Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe intercept in this model is \\(\\alpha / (1 - \\beta)\\). An alternative parameterization in terms of an intercept \\(\\gamma\\) suggested Mark Scheuerell on GitHub is \\(y_n \\sim \\textsf{normal}\\left(\\gamma + \\beta \\cdot (y_{n-1} - \\gamma), \\sigma\\right)\\).↩︎\nIn practice, it can be useful to remove the constraint to test whether a non-stationary set of coefficients provides a better fit to the data. It can also be useful to add a trend term to the model, because an unfitted trend will manifest as non-stationarity.↩︎\nThis subsection is a lightly edited comment of Jonathan Gilligan’s on GitHub; see https://github.com/stan-dev/stan/issues/1617#issuecomment-160249142.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Reparameterization and Change of Variables"
+      "Example Models",
+      "Time-Series Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/reparameterization.html#vectors-with-varying-bounds",
-    "href": "stan-users-guide/reparameterization.html#vectors-with-varying-bounds",
-    "title": "Reparameterization and Change of Variables",
+    "objectID": "stan-users-guide/style-guide.html",
+    "href": "stan-users-guide/style-guide.html",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "Stan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.\n\n\nFor example, suppose there is a vector parameter \\(\\alpha\\) with a vector \\(L\\) of lower bounds. The simplest way to deal with this if \\(L\\) is a constant is to shift a lower-bounded parameter.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L&gt;[N] alpha_raw;\n  // ...\n}\nThe above is equivalent to manually calculating the vector bounds by the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + alpha_raw;\n  // ...\n}\nThe Jacobian for adding a constant is one, so its log drops out of the log density.\nEven if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from \\((L, \\alpha_{\\textrm{raw}})\\) to \\((L + \\alpha_{\\textrm{raw}}, \\alpha_{\\textrm{raw}})\\) produces a Jacobian derivative matrix with a unit determinant.\nIt’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  vector&lt;lower=L + alpha_raw&gt;[N] alpha;\n  // ...\n}\nThis is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + exp(alpha_raw);\n  jacobian += sum(alpha_raw); // log Jacobian\n  // ...\n}\nmodel {\n  // ...\n}\nThe adjustment in the log Jacobian determinant of the transform mapping \\(\\alpha_{\\textrm{raw}}\\) to \\(\\alpha = L + \\exp(\\alpha_{\\textrm{raw}})\\). The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here \\(L\\) can even be a vector containing parameters that don’t depend on \\(\\alpha_{\\textrm{raw}}\\); if the bounds do depend on \\(\\alpha_{\\textrm{raw}}\\) then a revised Jacobian needs to be calculated taking into account the dependencies.\n\n\n\nSuppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to \\((0, 1)\\). This is easily accomplished as the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L, upper=U&gt;[N] alpha;\n  // ...\n}\nThe same may be accomplished by manually constructing the transform as follows.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + (U - L) .* alpha_raw;\n}\nThe expression U - L is multiplied by alpha_raw elementwise to produce a vector of variables in \\((0, U-L)\\), then adding \\(L\\) results in a variable ranging between \\((L, U)\\).\nIn this case, it is important that \\(L\\) and \\(U\\) are constants, otherwise a Jacobian would be required when multiplying by \\(U - L\\).",
+    "text": "This chapter describes the preferred style for laying out Stan models. These are not rules of the language, but simply recommendations for laying out programs in a text editor. Although these recommendations may seem arbitrary, they are similar to those of many teams for many programming languages. Like rules for typesetting text, the goal is to achieve readability without wasting white space either vertically or horizontally. This is the style used in the Stan documentation, and should align with the auto-formatting ability of stanc3.\n\n\nThe most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.\n\n\n\nLine lengths should not exceed 80 characters.1\nThis is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.\n\n\n\nThe recommended file extension for Stan model files is .stan. Files which contain only function definitions (intended for use with #include) should be given the .stanfunctions extension. A .stanfunctions file only includes the function definition and does not require the functions{} block wrapped around the function. A simple example of usage where the function is defined and saved in the file foo.stanfunctions:\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe function foo can be accessed in the Stan program by including the path to the foo.stanfunctions file as:\nfunctions {\n  #include foo.stanfunctions;\n}\n// ...body...\nFor Stan data dump files, the recommended extension is .R, or more informatively, .data.R. For JSON output, the recommended extension is .json.\n\n\n\nThe recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (_) used as a separator. Thus it is preferred to use sigma_y, rather than the run together sigmay, camel-case sigmaY, or capitalized camel-case SigmaY. An exception is often made for terms appearing in mathematical expressions with standard names, like A for a matrix.\nAnother exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an \\(M \\times N\\) matrix \\(a\\) would look as follows.\nfor (m in 1:M) {\n  for (n in 1:N) {\n     a[m, n] = ...\n  }\n}\n\n\n\nDeclaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.\nThe following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of mu in each iteration.\nmodel {\n  array[N] real mu;\n  for (n in 1:N) {\n    mu[n] = alpha * x[n] + beta;\n    y[n] ~ normal(mu[n],sigma);\n  }\n}\nBecause variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.\nmodel {\n  for (n in 1:N) {\n    real mu;\n    mu = alpha * x[n] + beta;\n    y[n] ~ normal(mu,sigma);\n  }\n}\nThe local variable can be eliminated altogether, as follows.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(alpha * x[n] + beta, sigma);\n  }\n}\nThere is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.\n\n\nIn the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.\nmodel {\n  vector[K] mu;\n  for (n in 1:N) {\n    for (k in 1:K) {\n      mu[k] = // ...\n    }\n    y[n] ~ multi_normal(mu,Sigma);\n}\nIn this case, the vector mu will be allocated outside of both loops, and used a total of N times.\n\n\n\n\n\n\nSingle-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThe use of loops and conditionals without brackets can be dangerous. For instance, consider this program.\nfor (n in 1:N)\n  z[n] ~ normal(nu,1);\n  y[n] ~ normal(mu,1);\nBecause Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.\nfor (n in 1:N) {\n  z[n] ~ normal(nu,1);\n}\ny[n] ~ normal(mu,1);\nTherefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the else branch contains exactly one conditional. Then, it is preferred to place the following if on the same line, as in the following.\nif (x) {\n  // ...\n} else if (y) {\n  // ...\n} else {\n  // ...\n}\n\n\n\nThe preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression a + b * c is preferred to the equivalent a + (b * c) or (a + (b * c)). The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.\nSimilarly, comparison operators can usually be written with minimal bracketing, with the form y[n] &gt; 0 || x[n] != 0 preferred to the bracketed form (y[n] &gt; 0) || (x[n] != 0).\n\n\n\nVertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThis also goes for parameters blocks, transformed data blocks, which should look as follows.\ntransformed parameters {\n  real sigma;\n  // ...\n}\nThe exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.\n\n\n\n\nWhile Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued x, one should use\nif (x != 0) { ...\nin place of\nif (x) { ...\nBeyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.\n\n\n\nFunctions are laid out the same way as in languages such as Java and C++. For example,\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.\nIf function names or argument lists are long, they can be written as\nmatrix\nfunction_to_do_some_hairy_algebra(matrix thingamabob,\n                                  vector doohickey2) {\n  // ...body...\n}\nThe function starts a new line, under the type. The arguments are aligned under each other.\nFunction documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the documenting functions section.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows correspond to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\n}\nThe open comment is /**, asterisks are aligned below the first asterisk of the open comment, and the end comment */ is also aligned on the asterisk. The tags @param and @return are used to label function arguments (i.e., parameters) and return values.\n\n\n\nStan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII 0x20), line feed (ASCII 0x0A), carriage return (0x0D), and tab (0x09). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.\n\n\nEach statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax\nreal mu, sigma;\n\n\n\nStan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.\n\n\n\nStan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.\n\n\n\nUse a space after ifs. For instance, use if (x &lt; y) {..., not if(x &lt; y){ ....\n\n\n\nThere should not be space between a function name and the arguments it applies to. For instance, use normal(0, 1), not normal (0,1).\n\n\n\nThere should be spaces around binary operators. For instance, use y[1] = x, not y[1]=x, use (x + y) * z not (x+y)*z.\nUnary operators are written without a space, such as in -x, !y.\n\n\n\nAnother exception to the above rule is when the assignment operator (=) is used inside a type constraint, such as\nreal&lt;lower=0&gt; x;\nSpaces should still be used in arithmetic and following commas, as in\nreal&lt;lower=0, upper=a * x + b&gt; x;\n\n\n\nSometimes expressions are too long to fit on a single line. In that case, the recommended form is to break before an operator,2 aligning the operator to a term above to indicate scoping. For example, use the following form\nvector[J] p_distance = Phi((distance_tolerance - overshot)\n                           ./ ((x + overshot) * sigma_distance))\n                       - Phi((-overshot)\n                             ./ ((x + overshot) * sigma_distance));\nHere, the elementwise division operator (./) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (Phi).\nFor functions with multiple arguments, break after a comma and line the next argument up underneath as follows.\ny[n] ~ normal(alpha + beta * x + gamma * y,\n              pow(tau,-0.5));\n\n\n\nCommas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.\nFor example,\nnormal(alpha * x[n] + beta, sigma);\nis preferred over\nnormal(alpha * x[n] + beta,sigma);\n\n\n\nWherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.\n\n\nNewlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point 0x0A). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point 0x0D) followed by a line-feed (LF) character.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Reparameterization and Change of Variables"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/reparameterization.html#footnotes",
-    "href": "stan-users-guide/reparameterization.html#footnotes",
-    "title": "Reparameterization and Change of Variables",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThis is in contrast to (penalized) maximum likelihood estimates, which are not parameterization invariant.↩︎\nThis example is for illustrative purposes only; the recommended way to implement the lognormal distribution in Stan is with the built-in lognormal probability function; see the functions reference manual for details.↩︎",
+    "objectID": "stan-users-guide/style-guide.html#choose-a-consistent-style",
+    "href": "stan-users-guide/style-guide.html#choose-a-consistent-style",
+    "title": "Stan Program Style Guide",
+    "section": "",
+    "text": "The most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Reparameterization and Change of Variables"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/references.html",
-    "href": "stan-users-guide/references.html",
-    "title": "References",
-    "section": "",
-    "text": "References\n\n\n\n\n Back to top"
-  },
-  {
-    "objectID": "stan-users-guide/problematic-posteriors.html",
-    "href": "stan-users-guide/problematic-posteriors.html",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#line-length",
+    "href": "stan-users-guide/style-guide.html#line-length",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "Mathematically speaking, with a proper posterior, one can do Bayesian inference and that’s that. There is not even a need to require a finite variance or even a finite mean—all that’s needed is a finite integral. Nevertheless, modeling is a tricky business and even experienced modelers sometimes code models that lead to improper priors. Furthermore, some posteriors are mathematically sound, but ill-behaved in practice. This chapter discusses issues in models that create problematic posterior inferences, either in general for Bayesian inference or in practice for Stan.\n\n\nThis section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.\n\n\n\n\nThe first example of collinearity is an artificial example involving redundant intercept parameters.1\nSuppose there are observations \\(y_n\\) for \\(n \\in \\{1,\\dotsc,N\\}\\), two intercept parameters \\(\\lambda_1\\) and \\(\\lambda_2\\), a scale parameter \\(\\sigma &gt; 0\\), and the data model \\[\ny_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2, \\sigma).\n\\]\nFor any constant \\(q\\), the sampling density for \\(y\\) does not change if we add \\(q\\) to \\(\\lambda_1\\) and subtract it from \\(\\lambda_2\\), i.e., \\[\np(y \\mid \\lambda_1, \\lambda_2,\\sigma)\n=\np(y \\mid \\lambda_1 + q, \\lambda_2 - q, \\sigma).\n\\]\nThe consequence is that an improper uniform prior \\(p(\\mu,\\sigma)\n\\propto 1\\) leads to an improper posterior. This impropriety arises because the neighborhoods around \\(\\lambda_1 + q, \\lambda_2 - q\\) have the same mass no matter what \\(q\\) is. Therefore, a sampler would need to spend as much time in the neighborhood of \\(\\lambda_1=1\\,000\\,000\\,000\\) and \\(\\lambda_2=-1\\,000\\,000\\,000\\) as it does in the neighborhood of \\(\\lambda_1=0\\) and \\(\\lambda_2=0\\), and so on for ever more far-ranging values.\nThe marginal posterior \\(p(\\lambda_1,\\lambda_2 \\mid y)\\) for this model is thus improper.2\nThe impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where \\(\\lambda_2 = \\lambda_1 + c\\) for some constant \\(c\\).\nContrast this model with a simple regression with a single intercept parameter \\(\\mu\\) and data model \\[\ny_n \\sim \\textsf{normal}(\\mu,\\sigma).\n\\] Even with an improper prior, the posterior is proper as long as there are at least two data points \\(y_n\\) with distinct values.\n\n\n\nConsider an item-response theory model for students \\(j \\in 1{:}J\\) with abilities \\(\\alpha_j\\) and test items \\(i \\in 1{:}I\\) with difficulties \\(\\beta_i\\). The observed data are an \\(I \\times J\\) array with entries \\(y_{i, j} \\in \\{ 0, 1 \\}\\) coded such that \\(y_{i, j} = 1\\) indicates that student \\(j\\) answered question \\(i\\) correctly. The sampling distribution for the data is \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\alpha_j - \\beta_i)).\n\\]\nFor any constant \\(c\\), the probability of \\(y\\) is unchanged by adding a constant \\(c\\) to all the abilities and subtracting it from all the difficulties, i.e., \\[\np(y \\mid \\alpha, \\beta)\n=\np(y \\mid \\alpha + c, \\beta - c).\n\\]\nThis leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.\n\n\n\nThe general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model \\[\ny_n \\sim \\textsf{normal}(x_n \\beta, \\sigma)\n\\] for an \\(N\\)-dimensional observation vector \\(y\\), an \\(N \\times K\\) predictor matrix \\(x\\), and a \\(K\\)-dimensional coefficient vector \\(\\beta\\).\nNow suppose that column \\(k\\) of the predictor matrix is a multiple of column \\(k'\\), i.e., there is some constant \\(c\\) such that \\(x_{n,k} = c\n\\, x_{n,k'}\\) for all \\(n\\). In this case, the coefficients \\(\\beta_k\\) and \\(\\beta_{k'}\\) can covary without changing the predictions, so that for any \\(d \\neq 0\\), \\[\np(y \\mid \\ldots, \\beta_k, \\dotsc, \\beta_{k'}, \\dotsc, \\sigma)\n=\np(y \\mid \\ldots, d \\beta_k, \\dotsc, \\frac{d}{c} \\, \\beta_{k'}, \\dotsc,\n\\sigma).\n\\]\nEven if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.\n\n\n\nConsider adding a discrimination parameter \\(\\delta_i\\) for each question in an IRT model, with data model \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\delta_i(\\alpha_j - \\beta_i))).\n\\] For any constant \\(c \\neq 0\\), multiplying \\(\\delta\\) by \\(c\\) and dividing \\(\\alpha\\) and \\(\\beta\\) by \\(c\\) produces the same likelihood, \\[\np(y \\mid \\delta,\\alpha,\\beta)\n= p(y \\mid c \\delta, \\frac{1}{c}\\alpha, \\frac{1}{c}\\beta).\n\\] If \\(c &lt; 0\\), this switches the signs of every component in \\(\\alpha\\), \\(\\beta\\), and \\(\\delta\\) without changing the density.\n\n\n\nIn order to parameterize a \\(K\\)-simplex (i.e., a \\(K\\)-vector with non-negative values that sum to one), only \\(K - 1\\) parameters are necessary because the \\(K\\)th is just one minus the sum of the first \\(K\n- 1\\) parameters, so that if \\(\\theta\\) is a \\(K\\)-simplex, \\[\n\\theta_K = 1 - \\sum_{k=1}^{K-1} \\theta_k.\n\\]\nThe softmax function maps a \\(K\\)-vector \\(\\alpha\\) of linear predictors to a \\(K\\)-simplex \\(\\theta = \\texttt{softmax}(\\alpha)\\) by defining \\[\n\\theta_k = \\frac{\\exp(\\alpha_k)}{\\sum_{k'=1}^K \\exp(\\alpha_{k'})}.\n\\]\nThe softmax function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters \\(\\alpha\\). In particular, adding or subtracting a constant from each \\(\\alpha_k\\) produces the same simplex \\(\\theta\\).\n\n\n\n\nAll of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.\n\n\nIn the case of the multiple intercepts, \\(\\lambda_1\\) and \\(\\lambda_2\\), the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter \\(\\mu\\) and sampling distribution \\(y_n \\sim \\textsf{normal}(\\mu, \\sigma)\\). The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix \\(x\\).\n\n\n\nThe IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability \\(\\alpha_1\\) can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.\nThis solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters \\(\\delta_i\\). To solve this problem, one of the discrimination parameters, say \\(\\delta_1\\), must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.\nThe many-to-one nature of \\(\\texttt{softmax}(\\alpha)\\) is typically mitigated by pinning a component of \\(\\alpha\\), for instance fixing \\(\\alpha_K = 0\\). The resulting mapping is one-to-one from \\(K-1\\) unconstrained parameters to a \\(K\\)-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a \\(K-1\\)-vector can be written as\nvector softmax_id(vector alpha) {\n  vector[num_elements(alpha) + 1] alphac1;\n  for (k in 1:num_elements(alpha)) {\n    alphac1[k] = alpha[k];\n  }\n  alphac1[num_elements(alphac1)] = 0;\n  return softmax(alphac1);\n}\n\n\n\nSo far, the models have been discussed as if the priors on the parameters were improper uniform priors.\nA more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.\nFor example, normal priors on the multiple intercepts, \\[\n\\lambda_1, \\lambda_2 \\sim \\textsf{normal}(0,\\tau),\n\\] with a constant scale \\(\\tau\\), ensure that the posterior mode is located at a point where \\(\\lambda_1 = \\lambda_2\\), because this minimizes \\(\\log \\textsf{normal}(\\lambda_1 \\mid 0,\\tau) + \\log\n\\textsf{normal}(\\lambda_2 \\mid 0,\\tau)\\).3\nThe following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.\nThe two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.\n\n\n\nTwo intercepts with improper prior\n\n\nAdding a standard normal prior for the intercepts results in a proper posterior.\n\n\n\nTwo intercepts with proper prior\n\n\nThe single intercept parameterization with no prior also has a proper posterior.\n\n\n\nSingle intercepts with improper prior\n\n\nThe addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.\nAn alternative strategy for identifying a \\(K\\)-simplex parameterization \\(\\theta = \\texttt{softmax}(\\alpha)\\) in terms of an unconstrained \\(K\\)-vector \\(\\alpha\\) is to place a prior on the components of \\(\\alpha\\) with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning \\(\\alpha_K =\n0\\), the prior-based approach models the \\(K\\) outcomes symmetrically rather than modeling \\(K-1\\) outcomes relative to the \\(K\\)-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.\n\n\n\nCare must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.\nIdeally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.\nWhen there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.\nPriors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters \\(\\alpha\\) to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.\n\n\n\n\n\nWhere collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.\n\n\nConsider a normal mixture model with two location parameters \\(\\mu_1\\) and \\(\\mu_2\\), a shared scale \\(\\sigma &gt; 0\\), a mixture ratio \\(\\theta \\in\n[0,1]\\), and data model \\[\np(y \\mid \\theta,\\mu_1,\\mu_2,\\sigma)\n= \\prod_{n=1}^N \\big( \\theta \\, \\textsf{normal}(y_n \\mid \\mu_1,\\sigma)\n                       + (1 - \\theta) \\, \\textsf{normal}(y_n \\mid \\mu_2,\\sigma) \\big).\n\\] The issue here is exchangeability of the mixture components, because \\[\np(\\theta,\\mu_1,\\mu_2,\\sigma \\mid y) = p\\big((1-\\theta),\\mu_2,\\mu_1,\\sigma \\mid y\\big).\n\\] The problem is exacerbated as the number of mixture components \\(K\\) grows, as in clustering models, leading to \\(K!\\) identical posterior maxima.\n\n\n\nThe analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the \\(\\hat{R}\\) convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for \\(\\mu_1\\) that is equal to that of \\(\\mu_2\\), and a posterior mean for \\(\\theta\\) that is always 1/2, no matter what the data are.\n\n\n\nIn some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for \\(\\theta\\) in the two component mixture model will always be 1/2.\n\n\n\nTheoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.\nBeing able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for \\(\\mu_1\\) to move to a new mode when sampled conditioned on the current values of \\(\\mu_2\\) and \\(\\theta\\). For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.\nEven with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.\n\n\n\nSeveral hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.\n\n\nOne common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining \\(\\mu_1 &lt; \\mu_2\\) in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering \\(\\mu_1 &gt; \\mu_2\\). In these cases, the posteriors are affected by the constraint and true posterior uncertainty in \\(\\mu_1\\) and \\(\\mu_2\\) is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use \\(M\\) posterior draws to estimate \\(\\Pr[\\mu_1 &gt; \\mu_2]\\), will fail, because the estimator \\[\n\\Pr[\\mu_1 &gt; \\mu_2]\n\\approx\n\\sum_{m=1}^M \\textrm{I}\\left(\\mu_1^{(m)} &gt; \\mu_2^{(m)}\\right)\n\\] will result in an estimate of 0 because the posterior respects the constraint in the model.\n\n\n\nAnother common approach is to run a single chain or to initialize the parameters near realistic values.4\nThis can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.\n\n\n\n\n\nIt is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of \\(K\\) normals might devolve to have \\(\\mu_i =\n\\mu_j\\) and \\(\\sigma_i = \\sigma_j\\) for \\(i \\neq j\\).\nThis will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw \\((m)\\), it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.\nIt may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.\nIn general, it is difficult to recover exactly the right \\(K\\) mixture components in a mixture model as \\(K\\) increases beyond one (yes, even a two-component mixture can have this problem).\n\n\n\nIn some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.\n\n\nOne such example is a binary mixture model with scales varying by component, \\(\\sigma_1\\) and \\(\\sigma_2\\) for locations \\(\\mu_1\\) and \\(\\mu_2\\). In this situation, the density grows without bound as \\(\\sigma_1 \\rightarrow 0\\) and \\(\\mu_1 \\rightarrow y_n\\) for some \\(n\\); that is, one of the mixture components concentrates all of its mass around a single data item \\(y_n\\).\n\n\n\nAnother example of unbounded densities arises with a posterior such as \\(\\textsf{beta}(\\phi \\mid 0.5,0.5)\\), which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as \\(\\phi \\rightarrow 0\\) and \\(\\phi \\rightarrow 1\\). Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior \\[\\begin{align*}\np(\\phi \\mid y)\n&\\propto\n   \\textsf{beta}(\\phi \\mid 0.5,0.5) \\times \\prod_{n=1}^N \\textsf{Bernoulli}(y_n \\mid \\phi) \\\\\n&=\n   \\textsf{beta}\\left(\\phi \\,\\middle|\\, 0.5 + \\sum_{n=1}^N y_n, 0.5 + N - \\sum_{n=1}^N y_n\\right).\n\\end{align*}\\]\nIf \\(N = 9\\) and each \\(y_n = 1\\), the posterior is \\(\\textsf{beta}(\\phi \\mid 9.5,0.5)\\). This posterior is unbounded as \\(\\phi\n\\rightarrow 1\\). Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.\n\n\nStan does not sample directly on the constrained \\((0,1)\\) space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values \\(\\phi\\) are logit-transformed to \\((-\\infty,\\infty)\\). The boundaries at 0 and 1 are pushed out to \\(-\\infty\\) and \\(\\infty\\) respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of \\((0,1)\\) is \\(\\log\n\\operatorname{logit}^{-1}(\\phi) + \\log \\operatorname{logit}(1 - \\phi)\\).\nThere are two problems that still arise, though. The first is that if the posterior mass for \\(\\phi\\) is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.\n\n\n\n\n\nIn some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.\n\n\nConsider a logistic regression model with \\(N\\) observed outcomes \\(y_n\n\\in \\{ 0, 1 \\}\\), an \\(N \\times K\\) matrix \\(x\\) of predictors, a \\(K\\)-dimensional coefficient vector \\(\\beta\\), and data model \\[\ny_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(x_n \\beta)).\n\\] Now suppose that column \\(k\\) of the predictor matrix is such that \\(x_{n,k} &gt; 0\\) if and only if \\(y_n = 1\\), a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as \\(\\beta_k \\rightarrow \\infty\\), because for cases with \\(y_n = 1\\), \\(x_n \\beta \\rightarrow \\infty\\) and hence \\(\\operatorname{logit}^{-1}(x_n \\beta) \\rightarrow 1\\).\nWith separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for \\(\\beta_k\\) is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for \\(\\beta\\), which ensures a proper posterior.\n\n\n\n\nSuppose your model includes a parameter \\(\\psi\\) that is defined on \\([0,1]\\) and is given a flat prior \\(\\textsf{uniform}(\\psi \\mid 0,1)\\). Now if the data don’t tell us anything about \\(\\psi\\), the posterior is also \\(\\textsf{uniform}(\\psi \\mid 0,1)\\).\nAlthough there is no maximum likelihood estimate for \\(\\psi\\), the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on \\([0,1]\\), the posterior mean for \\(\\psi\\) is well-defined with value \\(1/2\\). Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for \\(\\psi\\) at all points in \\([0,1]\\).\n\n\n\nWith an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the collinearity section and illustrated in the non-identifiable density plot.\n\n\nGibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.\nConsider what happens with initial values \\(\\lambda_1^{(0)}, \\lambda_2^{(0)}\\). Gibbs sampling proceeds in iteration \\(m\\) by drawing \\[\\begin{align*}\n\\lambda_1^{(m)} &\\sim p(\\lambda_1 \\mid \\lambda_2^{(m-1)}, \\sigma^{(m-1)},  y) \\\\\n\\lambda_2^{(m)} &\\sim p(\\lambda_2 \\mid \\lambda_1^{(m)},   \\sigma^{(m-1)},  y) \\\\\n\\sigma^{(m)}    &\\sim p(\\sigma    \\mid \\lambda_1^{(m)},   \\lambda_2^{(m)}, y).\n\\end{align*}\\]\nNow consider the draw for \\(\\lambda_1\\) (the draw for \\(\\lambda_2\\) is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next \\(\\lambda_1\\) can be drawn is highly constrained by the current values of \\(\\lambda_2\\) and \\(\\sigma\\). Gibbs will run quickly and provide seemingly reasonable inferences for \\(\\lambda_1 + \\lambda_2\\). But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.\n\n\n\nHamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for \\(\\lambda_1\\) and \\(\\lambda_2\\), the gradient of the log posterior is going to adjust for the correlation and the simulation will run \\(\\lambda_1\\) and \\(\\lambda_2\\) in opposite directions along the valley corresponding to the ridge in the posterior log density.\n\n\n\nStan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior (see Hoffman and Gelman 2014). NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.\nHere are results of Stan runs with default parameters fit to \\(N=100\\) data points generated from \\(y_n \\sim \\textsf{normal}(0,1)\\):\nTwo Scale Parameters, Improper Prior\nInference for Stan model: improper_stan\nWarmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total\nSampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total\n\n                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat\nlp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0\nn_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0\nlambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2\nlambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2\nsigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1\nmu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0\nTwo Scale Parameters, Weak Prior\nWarmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total\nSampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total\n\n                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat\nlp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0\nn_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0\nlambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nlambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nsigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0\nmu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0\nOne Scale Parameter, Improper Prior\nWarmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total\nSampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total\n\n                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat\nlp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0\nn_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0\nmu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0\nsigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0\nOn the top is the non-identified model with improper uniform priors and data model \\(y_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2,\n\\sigma)\\).\nIn the middle is the same data model as in top plus priors \\(\\lambda_k \\sim \\textsf{normal}(0,10)\\).\nOn the bottom is an identified model with an improper prior, with data model \\(y_n \\sim \\textsf{normal}(\\mu,\\sigma)\\). All models estimate \\(\\mu\\) at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value \\(\\mu=0\\) is within the 90% posterior intervals in all three models.\n\n\n\nTo illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the collinearity section.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real lambda1;\n  real lambda2;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  real mu;\n  mu = lambda1 + lambda2;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nThe second adds priors to the model block for lambda1 and lambda2 to the previous model.\nlambda1 ~ normal(0, 10);\nlambda2 ~ normal(0, 10);\nThe third involves a single location parameter, but no priors.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nAll three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.\n\nAs indicated by R_hat column, all parameters have converged other than \\(\\lambda_1\\) and \\(\\lambda_2\\) in the non-identified model.\nThe average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.\nThe effective sample size per second for \\(\\mu\\) is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for \\(\\sigma\\).\nIn the non-identified model, the 95% interval for \\(\\lambda_1\\) is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.\nIn all three models, the simulated value of \\(\\mu=0\\) and \\(\\sigma=1\\) are well within the posterior 90% intervals.\n\nThe first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.",
+    "text": "Line lengths should not exceed 80 characters.1\nThis is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#collinearity.section",
-    "href": "stan-users-guide/problematic-posteriors.html#collinearity.section",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#file-extensions",
+    "href": "stan-users-guide/style-guide.html#file-extensions",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "This section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.\n\n\n\n\nThe first example of collinearity is an artificial example involving redundant intercept parameters.1\nSuppose there are observations \\(y_n\\) for \\(n \\in \\{1,\\dotsc,N\\}\\), two intercept parameters \\(\\lambda_1\\) and \\(\\lambda_2\\), a scale parameter \\(\\sigma &gt; 0\\), and the data model \\[\ny_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2, \\sigma).\n\\]\nFor any constant \\(q\\), the sampling density for \\(y\\) does not change if we add \\(q\\) to \\(\\lambda_1\\) and subtract it from \\(\\lambda_2\\), i.e., \\[\np(y \\mid \\lambda_1, \\lambda_2,\\sigma)\n=\np(y \\mid \\lambda_1 + q, \\lambda_2 - q, \\sigma).\n\\]\nThe consequence is that an improper uniform prior \\(p(\\mu,\\sigma)\n\\propto 1\\) leads to an improper posterior. This impropriety arises because the neighborhoods around \\(\\lambda_1 + q, \\lambda_2 - q\\) have the same mass no matter what \\(q\\) is. Therefore, a sampler would need to spend as much time in the neighborhood of \\(\\lambda_1=1\\,000\\,000\\,000\\) and \\(\\lambda_2=-1\\,000\\,000\\,000\\) as it does in the neighborhood of \\(\\lambda_1=0\\) and \\(\\lambda_2=0\\), and so on for ever more far-ranging values.\nThe marginal posterior \\(p(\\lambda_1,\\lambda_2 \\mid y)\\) for this model is thus improper.2\nThe impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where \\(\\lambda_2 = \\lambda_1 + c\\) for some constant \\(c\\).\nContrast this model with a simple regression with a single intercept parameter \\(\\mu\\) and data model \\[\ny_n \\sim \\textsf{normal}(\\mu,\\sigma).\n\\] Even with an improper prior, the posterior is proper as long as there are at least two data points \\(y_n\\) with distinct values.\n\n\n\nConsider an item-response theory model for students \\(j \\in 1{:}J\\) with abilities \\(\\alpha_j\\) and test items \\(i \\in 1{:}I\\) with difficulties \\(\\beta_i\\). The observed data are an \\(I \\times J\\) array with entries \\(y_{i, j} \\in \\{ 0, 1 \\}\\) coded such that \\(y_{i, j} = 1\\) indicates that student \\(j\\) answered question \\(i\\) correctly. The sampling distribution for the data is \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\alpha_j - \\beta_i)).\n\\]\nFor any constant \\(c\\), the probability of \\(y\\) is unchanged by adding a constant \\(c\\) to all the abilities and subtracting it from all the difficulties, i.e., \\[\np(y \\mid \\alpha, \\beta)\n=\np(y \\mid \\alpha + c, \\beta - c).\n\\]\nThis leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.\n\n\n\nThe general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model \\[\ny_n \\sim \\textsf{normal}(x_n \\beta, \\sigma)\n\\] for an \\(N\\)-dimensional observation vector \\(y\\), an \\(N \\times K\\) predictor matrix \\(x\\), and a \\(K\\)-dimensional coefficient vector \\(\\beta\\).\nNow suppose that column \\(k\\) of the predictor matrix is a multiple of column \\(k'\\), i.e., there is some constant \\(c\\) such that \\(x_{n,k} = c\n\\, x_{n,k'}\\) for all \\(n\\). In this case, the coefficients \\(\\beta_k\\) and \\(\\beta_{k'}\\) can covary without changing the predictions, so that for any \\(d \\neq 0\\), \\[\np(y \\mid \\ldots, \\beta_k, \\dotsc, \\beta_{k'}, \\dotsc, \\sigma)\n=\np(y \\mid \\ldots, d \\beta_k, \\dotsc, \\frac{d}{c} \\, \\beta_{k'}, \\dotsc,\n\\sigma).\n\\]\nEven if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.\n\n\n\nConsider adding a discrimination parameter \\(\\delta_i\\) for each question in an IRT model, with data model \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\delta_i(\\alpha_j - \\beta_i))).\n\\] For any constant \\(c \\neq 0\\), multiplying \\(\\delta\\) by \\(c\\) and dividing \\(\\alpha\\) and \\(\\beta\\) by \\(c\\) produces the same likelihood, \\[\np(y \\mid \\delta,\\alpha,\\beta)\n= p(y \\mid c \\delta, \\frac{1}{c}\\alpha, \\frac{1}{c}\\beta).\n\\] If \\(c &lt; 0\\), this switches the signs of every component in \\(\\alpha\\), \\(\\beta\\), and \\(\\delta\\) without changing the density.\n\n\n\nIn order to parameterize a \\(K\\)-simplex (i.e., a \\(K\\)-vector with non-negative values that sum to one), only \\(K - 1\\) parameters are necessary because the \\(K\\)th is just one minus the sum of the first \\(K\n- 1\\) parameters, so that if \\(\\theta\\) is a \\(K\\)-simplex, \\[\n\\theta_K = 1 - \\sum_{k=1}^{K-1} \\theta_k.\n\\]\nThe softmax function maps a \\(K\\)-vector \\(\\alpha\\) of linear predictors to a \\(K\\)-simplex \\(\\theta = \\texttt{softmax}(\\alpha)\\) by defining \\[\n\\theta_k = \\frac{\\exp(\\alpha_k)}{\\sum_{k'=1}^K \\exp(\\alpha_{k'})}.\n\\]\nThe softmax function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters \\(\\alpha\\). In particular, adding or subtracting a constant from each \\(\\alpha_k\\) produces the same simplex \\(\\theta\\).\n\n\n\n\nAll of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.\n\n\nIn the case of the multiple intercepts, \\(\\lambda_1\\) and \\(\\lambda_2\\), the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter \\(\\mu\\) and sampling distribution \\(y_n \\sim \\textsf{normal}(\\mu, \\sigma)\\). The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix \\(x\\).\n\n\n\nThe IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability \\(\\alpha_1\\) can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.\nThis solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters \\(\\delta_i\\). To solve this problem, one of the discrimination parameters, say \\(\\delta_1\\), must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.\nThe many-to-one nature of \\(\\texttt{softmax}(\\alpha)\\) is typically mitigated by pinning a component of \\(\\alpha\\), for instance fixing \\(\\alpha_K = 0\\). The resulting mapping is one-to-one from \\(K-1\\) unconstrained parameters to a \\(K\\)-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a \\(K-1\\)-vector can be written as\nvector softmax_id(vector alpha) {\n  vector[num_elements(alpha) + 1] alphac1;\n  for (k in 1:num_elements(alpha)) {\n    alphac1[k] = alpha[k];\n  }\n  alphac1[num_elements(alphac1)] = 0;\n  return softmax(alphac1);\n}\n\n\n\nSo far, the models have been discussed as if the priors on the parameters were improper uniform priors.\nA more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.\nFor example, normal priors on the multiple intercepts, \\[\n\\lambda_1, \\lambda_2 \\sim \\textsf{normal}(0,\\tau),\n\\] with a constant scale \\(\\tau\\), ensure that the posterior mode is located at a point where \\(\\lambda_1 = \\lambda_2\\), because this minimizes \\(\\log \\textsf{normal}(\\lambda_1 \\mid 0,\\tau) + \\log\n\\textsf{normal}(\\lambda_2 \\mid 0,\\tau)\\).3\nThe following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.\nThe two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.\n\n\n\nTwo intercepts with improper prior\n\n\nAdding a standard normal prior for the intercepts results in a proper posterior.\n\n\n\nTwo intercepts with proper prior\n\n\nThe single intercept parameterization with no prior also has a proper posterior.\n\n\n\nSingle intercepts with improper prior\n\n\nThe addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.\nAn alternative strategy for identifying a \\(K\\)-simplex parameterization \\(\\theta = \\texttt{softmax}(\\alpha)\\) in terms of an unconstrained \\(K\\)-vector \\(\\alpha\\) is to place a prior on the components of \\(\\alpha\\) with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning \\(\\alpha_K =\n0\\), the prior-based approach models the \\(K\\) outcomes symmetrically rather than modeling \\(K-1\\) outcomes relative to the \\(K\\)-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.\n\n\n\nCare must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.\nIdeally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.\nWhen there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.\nPriors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters \\(\\alpha\\) to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.",
+    "text": "The recommended file extension for Stan model files is .stan. Files which contain only function definitions (intended for use with #include) should be given the .stanfunctions extension. A .stanfunctions file only includes the function definition and does not require the functions{} block wrapped around the function. A simple example of usage where the function is defined and saved in the file foo.stanfunctions:\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe function foo can be accessed in the Stan program by including the path to the foo.stanfunctions file as:\nfunctions {\n  #include foo.stanfunctions;\n}\n// ...body...\nFor Stan data dump files, the recommended extension is .R, or more informatively, .data.R. For JSON output, the recommended extension is .json.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#label-switching-problematic.section",
-    "href": "stan-users-guide/problematic-posteriors.html#label-switching-problematic.section",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#variable-naming",
+    "href": "stan-users-guide/style-guide.html#variable-naming",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "Where collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.\n\n\nConsider a normal mixture model with two location parameters \\(\\mu_1\\) and \\(\\mu_2\\), a shared scale \\(\\sigma &gt; 0\\), a mixture ratio \\(\\theta \\in\n[0,1]\\), and data model \\[\np(y \\mid \\theta,\\mu_1,\\mu_2,\\sigma)\n= \\prod_{n=1}^N \\big( \\theta \\, \\textsf{normal}(y_n \\mid \\mu_1,\\sigma)\n                       + (1 - \\theta) \\, \\textsf{normal}(y_n \\mid \\mu_2,\\sigma) \\big).\n\\] The issue here is exchangeability of the mixture components, because \\[\np(\\theta,\\mu_1,\\mu_2,\\sigma \\mid y) = p\\big((1-\\theta),\\mu_2,\\mu_1,\\sigma \\mid y\\big).\n\\] The problem is exacerbated as the number of mixture components \\(K\\) grows, as in clustering models, leading to \\(K!\\) identical posterior maxima.\n\n\n\nThe analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the \\(\\hat{R}\\) convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for \\(\\mu_1\\) that is equal to that of \\(\\mu_2\\), and a posterior mean for \\(\\theta\\) that is always 1/2, no matter what the data are.\n\n\n\nIn some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for \\(\\theta\\) in the two component mixture model will always be 1/2.\n\n\n\nTheoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.\nBeing able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for \\(\\mu_1\\) to move to a new mode when sampled conditioned on the current values of \\(\\mu_2\\) and \\(\\theta\\). For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.\nEven with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.\n\n\n\nSeveral hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.\n\n\nOne common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining \\(\\mu_1 &lt; \\mu_2\\) in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering \\(\\mu_1 &gt; \\mu_2\\). In these cases, the posteriors are affected by the constraint and true posterior uncertainty in \\(\\mu_1\\) and \\(\\mu_2\\) is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use \\(M\\) posterior draws to estimate \\(\\Pr[\\mu_1 &gt; \\mu_2]\\), will fail, because the estimator \\[\n\\Pr[\\mu_1 &gt; \\mu_2]\n\\approx\n\\sum_{m=1}^M \\textrm{I}\\left(\\mu_1^{(m)} &gt; \\mu_2^{(m)}\\right)\n\\] will result in an estimate of 0 because the posterior respects the constraint in the model.\n\n\n\nAnother common approach is to run a single chain or to initialize the parameters near realistic values.4\nThis can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.",
+    "text": "The recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (_) used as a separator. Thus it is preferred to use sigma_y, rather than the run together sigmay, camel-case sigmaY, or capitalized camel-case SigmaY. An exception is often made for terms appearing in mathematical expressions with standard names, like A for a matrix.\nAnother exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an \\(M \\times N\\) matrix \\(a\\) would look as follows.\nfor (m in 1:M) {\n  for (n in 1:N) {\n     a[m, n] = ...\n  }\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#component-collapsing-in-mixture-models",
-    "href": "stan-users-guide/problematic-posteriors.html#component-collapsing-in-mixture-models",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#local-variable-scope",
+    "href": "stan-users-guide/style-guide.html#local-variable-scope",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "It is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of \\(K\\) normals might devolve to have \\(\\mu_i =\n\\mu_j\\) and \\(\\sigma_i = \\sigma_j\\) for \\(i \\neq j\\).\nThis will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw \\((m)\\), it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.\nIt may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.\nIn general, it is difficult to recover exactly the right \\(K\\) mixture components in a mixture model as \\(K\\) increases beyond one (yes, even a two-component mixture can have this problem).",
+    "text": "Declaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.\nThe following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of mu in each iteration.\nmodel {\n  array[N] real mu;\n  for (n in 1:N) {\n    mu[n] = alpha * x[n] + beta;\n    y[n] ~ normal(mu[n],sigma);\n  }\n}\nBecause variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.\nmodel {\n  for (n in 1:N) {\n    real mu;\n    mu = alpha * x[n] + beta;\n    y[n] ~ normal(mu,sigma);\n  }\n}\nThe local variable can be eliminated altogether, as follows.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(alpha * x[n] + beta, sigma);\n  }\n}\nThere is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.\n\n\nIn the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.\nmodel {\n  vector[K] mu;\n  for (n in 1:N) {\n    for (k in 1:K) {\n      mu[k] = // ...\n    }\n    y[n] ~ multi_normal(mu,Sigma);\n}\nIn this case, the vector mu will be allocated outside of both loops, and used a total of N times.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-densities",
-    "href": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-densities",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#parentheses-and-brackets",
+    "href": "stan-users-guide/style-guide.html#parentheses-and-brackets",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "In some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.\n\n\nOne such example is a binary mixture model with scales varying by component, \\(\\sigma_1\\) and \\(\\sigma_2\\) for locations \\(\\mu_1\\) and \\(\\mu_2\\). In this situation, the density grows without bound as \\(\\sigma_1 \\rightarrow 0\\) and \\(\\mu_1 \\rightarrow y_n\\) for some \\(n\\); that is, one of the mixture components concentrates all of its mass around a single data item \\(y_n\\).\n\n\n\nAnother example of unbounded densities arises with a posterior such as \\(\\textsf{beta}(\\phi \\mid 0.5,0.5)\\), which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as \\(\\phi \\rightarrow 0\\) and \\(\\phi \\rightarrow 1\\). Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior \\[\\begin{align*}\np(\\phi \\mid y)\n&\\propto\n   \\textsf{beta}(\\phi \\mid 0.5,0.5) \\times \\prod_{n=1}^N \\textsf{Bernoulli}(y_n \\mid \\phi) \\\\\n&=\n   \\textsf{beta}\\left(\\phi \\,\\middle|\\, 0.5 + \\sum_{n=1}^N y_n, 0.5 + N - \\sum_{n=1}^N y_n\\right).\n\\end{align*}\\]\nIf \\(N = 9\\) and each \\(y_n = 1\\), the posterior is \\(\\textsf{beta}(\\phi \\mid 9.5,0.5)\\). This posterior is unbounded as \\(\\phi\n\\rightarrow 1\\). Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.\n\n\nStan does not sample directly on the constrained \\((0,1)\\) space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values \\(\\phi\\) are logit-transformed to \\((-\\infty,\\infty)\\). The boundaries at 0 and 1 are pushed out to \\(-\\infty\\) and \\(\\infty\\) respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of \\((0,1)\\) is \\(\\log\n\\operatorname{logit}^{-1}(\\phi) + \\log \\operatorname{logit}(1 - \\phi)\\).\nThere are two problems that still arise, though. The first is that if the posterior mass for \\(\\phi\\) is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.",
+    "text": "Single-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThe use of loops and conditionals without brackets can be dangerous. For instance, consider this program.\nfor (n in 1:N)\n  z[n] ~ normal(nu,1);\n  y[n] ~ normal(mu,1);\nBecause Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.\nfor (n in 1:N) {\n  z[n] ~ normal(nu,1);\n}\ny[n] ~ normal(mu,1);\nTherefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the else branch contains exactly one conditional. Then, it is preferred to place the following if on the same line, as in the following.\nif (x) {\n  // ...\n} else if (y) {\n  // ...\n} else {\n  // ...\n}\n\n\n\nThe preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression a + b * c is preferred to the equivalent a + (b * c) or (a + (b * c)). The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.\nSimilarly, comparison operators can usually be written with minimal bracketing, with the form y[n] &gt; 0 || x[n] != 0 preferred to the bracketed form (y[n] &gt; 0) || (x[n] != 0).\n\n\n\nVertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThis also goes for parameters blocks, transformed data blocks, which should look as follows.\ntransformed parameters {\n  real sigma;\n  // ...\n}\nThe exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-parameters",
-    "href": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-parameters",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#conditionals",
+    "href": "stan-users-guide/style-guide.html#conditionals",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "In some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.\n\n\nConsider a logistic regression model with \\(N\\) observed outcomes \\(y_n\n\\in \\{ 0, 1 \\}\\), an \\(N \\times K\\) matrix \\(x\\) of predictors, a \\(K\\)-dimensional coefficient vector \\(\\beta\\), and data model \\[\ny_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(x_n \\beta)).\n\\] Now suppose that column \\(k\\) of the predictor matrix is such that \\(x_{n,k} &gt; 0\\) if and only if \\(y_n = 1\\), a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as \\(\\beta_k \\rightarrow \\infty\\), because for cases with \\(y_n = 1\\), \\(x_n \\beta \\rightarrow \\infty\\) and hence \\(\\operatorname{logit}^{-1}(x_n \\beta) \\rightarrow 1\\).\nWith separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for \\(\\beta_k\\) is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for \\(\\beta\\), which ensures a proper posterior.",
+    "text": "While Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued x, one should use\nif (x != 0) { ...\nin place of\nif (x) { ...\nBeyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#uniform-posteriors",
-    "href": "stan-users-guide/problematic-posteriors.html#uniform-posteriors",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#functions",
+    "href": "stan-users-guide/style-guide.html#functions",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "Suppose your model includes a parameter \\(\\psi\\) that is defined on \\([0,1]\\) and is given a flat prior \\(\\textsf{uniform}(\\psi \\mid 0,1)\\). Now if the data don’t tell us anything about \\(\\psi\\), the posterior is also \\(\\textsf{uniform}(\\psi \\mid 0,1)\\).\nAlthough there is no maximum likelihood estimate for \\(\\psi\\), the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on \\([0,1]\\), the posterior mean for \\(\\psi\\) is well-defined with value \\(1/2\\). Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for \\(\\psi\\) at all points in \\([0,1]\\).",
+    "text": "Functions are laid out the same way as in languages such as Java and C++. For example,\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.\nIf function names or argument lists are long, they can be written as\nmatrix\nfunction_to_do_some_hairy_algebra(matrix thingamabob,\n                                  vector doohickey2) {\n  // ...body...\n}\nThe function starts a new line, under the type. The arguments are aligned under each other.\nFunction documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the documenting functions section.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows correspond to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\n}\nThe open comment is /**, asterisks are aligned below the first asterisk of the open comment, and the end comment */ is also aligned on the asterisk. The tags @param and @return are used to label function arguments (i.e., parameters) and return values.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#sampling-difficulties-with-problematic-priors",
-    "href": "stan-users-guide/problematic-posteriors.html#sampling-difficulties-with-problematic-priors",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#white-space",
+    "href": "stan-users-guide/style-guide.html#white-space",
+    "title": "Stan Program Style Guide",
     "section": "",
-    "text": "With an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the collinearity section and illustrated in the non-identifiable density plot.\n\n\nGibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.\nConsider what happens with initial values \\(\\lambda_1^{(0)}, \\lambda_2^{(0)}\\). Gibbs sampling proceeds in iteration \\(m\\) by drawing \\[\\begin{align*}\n\\lambda_1^{(m)} &\\sim p(\\lambda_1 \\mid \\lambda_2^{(m-1)}, \\sigma^{(m-1)},  y) \\\\\n\\lambda_2^{(m)} &\\sim p(\\lambda_2 \\mid \\lambda_1^{(m)},   \\sigma^{(m-1)},  y) \\\\\n\\sigma^{(m)}    &\\sim p(\\sigma    \\mid \\lambda_1^{(m)},   \\lambda_2^{(m)}, y).\n\\end{align*}\\]\nNow consider the draw for \\(\\lambda_1\\) (the draw for \\(\\lambda_2\\) is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next \\(\\lambda_1\\) can be drawn is highly constrained by the current values of \\(\\lambda_2\\) and \\(\\sigma\\). Gibbs will run quickly and provide seemingly reasonable inferences for \\(\\lambda_1 + \\lambda_2\\). But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.\n\n\n\nHamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for \\(\\lambda_1\\) and \\(\\lambda_2\\), the gradient of the log posterior is going to adjust for the correlation and the simulation will run \\(\\lambda_1\\) and \\(\\lambda_2\\) in opposite directions along the valley corresponding to the ridge in the posterior log density.\n\n\n\nStan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior (see Hoffman and Gelman 2014). NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.\nHere are results of Stan runs with default parameters fit to \\(N=100\\) data points generated from \\(y_n \\sim \\textsf{normal}(0,1)\\):\nTwo Scale Parameters, Improper Prior\nInference for Stan model: improper_stan\nWarmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total\nSampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total\n\n                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat\nlp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0\nn_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0\nlambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2\nlambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2\nsigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1\nmu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0\nTwo Scale Parameters, Weak Prior\nWarmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total\nSampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total\n\n                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat\nlp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0\nn_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0\nlambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nlambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nsigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0\nmu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0\nOne Scale Parameter, Improper Prior\nWarmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total\nSampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total\n\n                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat\nlp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0\nn_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0\nmu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0\nsigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0\nOn the top is the non-identified model with improper uniform priors and data model \\(y_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2,\n\\sigma)\\).\nIn the middle is the same data model as in top plus priors \\(\\lambda_k \\sim \\textsf{normal}(0,10)\\).\nOn the bottom is an identified model with an improper prior, with data model \\(y_n \\sim \\textsf{normal}(\\mu,\\sigma)\\). All models estimate \\(\\mu\\) at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value \\(\\mu=0\\) is within the 90% posterior intervals in all three models.\n\n\n\nTo illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the collinearity section.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real lambda1;\n  real lambda2;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  real mu;\n  mu = lambda1 + lambda2;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nThe second adds priors to the model block for lambda1 and lambda2 to the previous model.\nlambda1 ~ normal(0, 10);\nlambda2 ~ normal(0, 10);\nThe third involves a single location parameter, but no priors.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nAll three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.\n\nAs indicated by R_hat column, all parameters have converged other than \\(\\lambda_1\\) and \\(\\lambda_2\\) in the non-identified model.\nThe average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.\nThe effective sample size per second for \\(\\mu\\) is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for \\(\\sigma\\).\nIn the non-identified model, the 95% interval for \\(\\lambda_1\\) is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.\nIn all three models, the simulated value of \\(\\mu=0\\) and \\(\\sigma=1\\) are well within the posterior 90% intervals.\n\nThe first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.",
+    "text": "Stan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII 0x20), line feed (ASCII 0x0A), carriage return (0x0D), and tab (0x09). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.\n\n\nEach statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax\nreal mu, sigma;\n\n\n\nStan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.\n\n\n\nStan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.\n\n\n\nUse a space after ifs. For instance, use if (x &lt; y) {..., not if(x &lt; y){ ....\n\n\n\nThere should not be space between a function name and the arguments it applies to. For instance, use normal(0, 1), not normal (0,1).\n\n\n\nThere should be spaces around binary operators. For instance, use y[1] = x, not y[1]=x, use (x + y) * z not (x+y)*z.\nUnary operators are written without a space, such as in -x, !y.\n\n\n\nAnother exception to the above rule is when the assignment operator (=) is used inside a type constraint, such as\nreal&lt;lower=0&gt; x;\nSpaces should still be used in arithmetic and following commas, as in\nreal&lt;lower=0, upper=a * x + b&gt; x;\n\n\n\nSometimes expressions are too long to fit on a single line. In that case, the recommended form is to break before an operator,2 aligning the operator to a term above to indicate scoping. For example, use the following form\nvector[J] p_distance = Phi((distance_tolerance - overshot)\n                           ./ ((x + overshot) * sigma_distance))\n                       - Phi((-overshot)\n                             ./ ((x + overshot) * sigma_distance));\nHere, the elementwise division operator (./) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (Phi).\nFor functions with multiple arguments, break after a comma and line the next argument up underneath as follows.\ny[n] ~ normal(alpha + beta * x + gamma * y,\n              pow(tau,-0.5));\n\n\n\nCommas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.\nFor example,\nnormal(alpha * x[n] + beta, sigma);\nis preferred over\nnormal(alpha * x[n] + beta,sigma);\n\n\n\nWherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.\n\n\nNewlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point 0x0A). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point 0x0D) followed by a line-feed (LF) character.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/problematic-posteriors.html#footnotes",
-    "href": "stan-users-guide/problematic-posteriors.html#footnotes",
-    "title": "Problematic Posteriors",
+    "objectID": "stan-users-guide/style-guide.html#footnotes",
+    "href": "stan-users-guide/style-guide.html#footnotes",
+    "title": "Stan Program Style Guide",
     "section": "Footnotes",
-    "text": "Footnotes\n\n\nThis example was raised by Richard McElreath on the Stan users group in a query about the difference in behavior between Gibbs sampling as used in BUGS and JAGS and the Hamiltonian Monte Carlo (HMC) and no-U-turn samplers (NUTS) used by Stan.↩︎\nThe marginal posterior \\(p(\\sigma \\mid y)\\) for \\(\\sigma\\) is proper here as long as there are at least two distinct data points.↩︎\nA Laplace prior (or an L1 regularizer for penalized maximum likelihood estimation) is not sufficient to remove this additive invariance. It provides shrinkage, but does not in and of itself identify the parameters because adding a constant to \\(\\lambda_1\\) and subtracting it from \\(\\lambda_2\\) results in the same value for the prior density.↩︎\nTempering methods may be viewed as automated ways to carry out such a search for modes, though most MCMC tempering methods continue to search for modes on an ongoing basis; see (Swendsen and Wang 1986; Neal 1996).↩︎",
+    "text": "Footnotes\n\n\nEven 80 characters may be too many for rendering in print; for instance, in this manual, the number of code characters that fit on a line is about 65.↩︎\nThis is the usual convention in both typesetting and other programming languages. Neither R nor BUGS allows breaks before an operator because they allow newlines to signal the end of an expression or statement.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Problematic Posteriors"
+      "Appendices",
+      "Stan Program Style Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html",
-    "href": "stan-users-guide/posterior-predictive-checks.html",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/simulation-based-calibration.html",
+    "href": "stan-users-guide/simulation-based-calibration.html",
+    "title": "Simulation-Based Calibration Checking",
     "section": "",
-    "text": "Posterior predictive checks are a way of measuring whether a model does a good job of capturing relevant aspects of the data, such as means, standard deviations, and quantiles (Rubin 1984; Andrew Gelman, Meng, and Stern 1996). Posterior predictive checking works by simulating new replicated data sets based on the fitted model parameters and then comparing statistics applied to the replicated data set with the same statistic applied to the original data set.\nPrior predictive checks evaluate the prior the same way. Specifically, they evaluate what data sets would be consistent with the prior. They will not be calibrated with actual data, but extreme values help diagnose priors that are either too strong, too weak, poorly shaped, or poorly located.\nPrior and posterior predictive checks are two cases of the general concept of predictive checks, just conditioning on different things (no data and the observed data, respectively). For hierarchical models, there are intermediate versions, as discussed in the section on hierarchical models and mixed replication.\n\n\nThe posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data \\(y\\).\nThe posterior predictive distribution for replications \\(y^{\\textrm{rep}}\\) of the original data set \\(y\\) given model parameters \\(\\theta\\) is defined by \\[\np(y^{\\textrm{rep}} \\mid y)\n= \\int p(y^{\\textrm{rep}} \\mid \\theta)\n       \\cdot p(\\theta \\mid y) \\, \\textrm{d}\\theta.\n\\]\nAs with other posterior predictive quantities, generating a replicated data set \\(y^{\\textrm{rep}}\\) from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters \\(\\theta = (\\alpha, \\beta, \\sigma).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  y ~ normal(alpha + beta * x, sigma);\n}\nTo generate a replicated data set y_rep for this simple model, the following generated quantities block suffices.\ngenerated quantities {\n  array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n}\nThe vectorized form of the normal random number generator is used with the original predictors x and the model parameters alpha, beta, and sigma. The replicated data variable y_rep is declared to be the same size as the original data y, but instead of a vector type, it is declared to be an array of reals to match the return type of the function normal_rng. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.\nThe posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in the chapter on posterior predictions in that the original predictors \\(x\\) are used. That is, the posterior predictions are for the original data.\n\n\n\nA standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.\nHere’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\ntransformed data {\n  real&lt;lower=0&gt; mean_y = mean(to_vector(y));\n  real&lt;lower=0&gt; sd_y = sd(to_vector(y));\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  y ~ poisson(lambda);\n  lambda ~ exponential(0.2);\n}\ngenerated quantities {\n  array[N] int&lt;lower=0&gt; y_rep = poisson_rng(rep_array(lambda, N));\n  real&lt;lower=0&gt; mean_y_rep = mean(to_vector(y_rep));\n  real&lt;lower=0&gt; sd_y_rep = sd(to_vector(y_rep));\n  int&lt;lower=0, upper=1&gt; mean_gte = (mean_y_rep &gt;= mean_y);\n  int&lt;lower=0, upper=1&gt; sd_gte = (sd_y_rep &gt;= sd_y);\n}\nThe generated quantities block creates a variable y_rep for the replicated data, variables mean_y_rep and sd_y_rep for the statistics of the replicated data, and indicator variables mean_gte and sd_gte for whether the replicated statistic is greater than or equal to the statistic applied to the original data.\nNow consider generating data \\(y \\sim \\textrm{Poisson}(5)\\). The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.\n\n\n\nPosterior predictive checks for Poisson data generating process and Poisson model.\n\n\nWith a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.\nNow consider generating over-dispersed data \\(y \\sim \\textrm{negative-binomial2}(5, 1).\\) This has the same mean as \\(\\textrm{Poisson}(5)\\), namely \\(5\\), but a standard deviation of \\(\\sqrt{5 + 5^2 /1} \\approx 5.5.\\) There is no way to fit this data with the Poisson model, because a variable distributed as \\(\\textrm{Poisson}(\\lambda)\\) has mean \\(\\lambda\\) and standard deviation \\(\\sqrt{\\lambda},\\) which is \\(\\sqrt{5}\\) for \\(\\textrm{Poisson}(5).\\) Here’s the resulting small multiples plot, again with original data in the upper left.\n\n\n\nPosterior predictive checks for negative binomial data generating process and Poisson model.\n\n\nThis time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.\n\n\n\nIf a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic \\(s(\\cdot)\\) in a replicated data set exceeds that in the original data, \\[\n\\Pr\\!\\left[ s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right]\n=\n\\int\n\\textrm{I}\\left( s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right)\n\\cdot p\\left( y^{\\textrm{rep}} \\mid y \\right)\n\\, \\textrm{d}{y^{\\textrm{rep}}}.\n\\] It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified (Bayarri and Berger 2000).\nNevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.\nTo calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mean_gt;\n  int&lt;lower=0, upper=1&gt; sd_gt;\n  {\n    array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n    mean_gt = mean(y_rep) &gt; mean(y);\n    sd_gt = sd(y_rep) &gt; sd(y);\n  }\n}\nThe indicator variable mean_gt will have value 1 if the mean of the simulated data y_rep is greater than or equal to the mean of he original data y. Because the values of y_rep are not needed for the posterior predictive checks, the program saves output space by using a local variable for y_rep. The statistics mean(u) and sd(y) could also be computed in the transformed data block and saved.\nFor the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.\n\n\n\nHistogram of means of replicated data sets; vertical red line at mean of original data.\n\n\nThe p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior \\(p\\)-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable mean_gt.\nThe standard deviation statistic tells a different story.\n\n\n\nScatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.\n\n\nHere, the original data has much higher standard deviation than any of the replicated data sets. The resulting \\(p\\)-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small \\(p\\)-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.\n\n\nAny statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.\nDespite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.\n\n\n\n\nPrior predictive checks generate data according to the prior in order to asses whether a prior is appropriate (Gabry et al. 2019). A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, \\[\ny^{\\textrm{sim}} \\sim p(y).\n\\] The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.\nThis is easy to carry out mechanically by simulating parameters \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta)\n\\] according to the priors, then simulating data \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}})\n\\] according to the data model given the simulated parameters. The result is a simulation from the joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta)\n\\] and thus \\[\ny^{\\textrm{sim}} \\sim p(y)\n\\] is a simulation from the prior predictive distribution.\n\n\nA prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.\nPrior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n}\ngenerated quantities {\n  real alpha = normal_rng(0, 1);\n  real beta = normal_rng(0, 1);\n  array[N] real y_sim = poisson_log_rng(alpha + beta * x);\n}\nRunning this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.\n\n\n\n\nSuppose we have a model for a football (aka soccer) league where there are \\(J\\) teams. Each team has a scoring rate \\(\\lambda_j\\) and in each game will be assumed to score \\(\\textrm{poisson}(\\lambda_j)\\) points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed \\[\n\\lambda_j \\sim \\textrm{gamma}(\\epsilon_1, \\epsilon_2).\n\\] This is not just a manufactured example; The BUGS Book recommends setting \\(\\epsilon = (0.5, 0.00001)\\), which corresponds to a Jeffreys prior for a Poisson rate parameter prior (Lunn et al. 2012, 85).\nSuppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.\ndata {\n  int&lt;lower=0&gt; J;\n  array[2] real&lt;lower=0&gt; epsilon;\n}\ngenerated quantities {\n  array[J] real&lt;lower=0&gt; lambda;\n  array[J, J] int y;\n  for (j in 1:J) lambda[j] = gamma_rng(epsilon[1], epsilon[2]);\n  for (i in 1:J) {\n    for (j in 1:J) {\n      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);\n    }\n  }\n}\nIn this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are \\(\\binom{J}{2} = \\frac{J \\cdot (J - 1)}{2}.\\)\nUsing the \\(\\textrm{gamma}(0.5, 0.00001)\\) reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, \\(y^{(1:20)}_{1, 2}\\).\n2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987\n7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66\nThat’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this \\(\\textrm{gamma}(0.5, 0.00001)\\) prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.\nThe posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values (A. Gelman 2006).\nJust as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.\nIn this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see (Gabry et al. 2019).\nPrior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.\n\n\n\nAndrew Gelman, Meng, and Stern (1996) discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.\nFor example, consider a simple varying intercept logistic regression, with intercepts \\(\\alpha_k\\) for \\(k \\in 1:K\\). Each data item \\(y_n \\in \\{ 0, 1 \\}\\) is assumed to correspond to group \\(kk_n \\in 1:K.\\) The data model is thus \\[\ny_n \\sim \\textrm{bernoulli}(\\textrm{logit}^{-1}(\\alpha_{kk[n]})).\n\\] The varying intercepts have a hierarchical normal prior, \\[\n\\alpha_k \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The hyperparameters are themselves given weakly informative priors, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 2)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\]\nLike in a posterior predictive check, the hyperparameters \\(\\mu\\) and \\(\\sigma\\) are drawn from the posterior, \\[\n\\mu^{(m)}, \\sigma^{(m)} \\sim p(\\mu, \\sigma \\mid y)\n\\] Like in a prior predictive check, replicated values of \\(\\alpha\\) are drawn from the hyperparameters, \\[\n\\alpha^{\\textrm{rep}(m)}_k \\sim \\textrm{normal}(\\alpha_k \\mid\n\\mu^{(m)}, \\sigma^{(m)}).\n\\] The data items are then each replicated using the replicated intercepts, \\[\ny^{\\textrm{rep}(m)}_n \\sim\n\\textrm{bernoulli}\n  (\\textrm{logit}^{-1}(\\alpha^{\\textrm{rep}(m)}_{kk[n]})).\n\\] Thus the \\(y^{\\textrm{rep}(m)}\\) can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original \\(K\\) groups.\nIn Stan, mixed predictive replications \\(y^{\\textrm{rep}(m)}\\) can be programmed directly.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;offset=mu, multiplier=sigma&gt;[K] alpha;\n}\nmodel {\n  mu ~ normal(0, 2);               // hyperprior\n  sigma ~ lognormal(0, 1);\n  alpha ~ normal(mu, sigma);       // hierarchical prior\n  y ~ bernoulli_logit(alpha[kk]);  // data model\n}\ngenerated quantities {\n  // alpha replicated;  mu and sigma not replicated\n  array[K] real alpha_rep\n    = normal_rng(rep_vector(mu, K), sigma);\n  array[N] int&lt;lower=0, upper=1&gt; y_rep\n    = bernoulli_logit_rng(alpha_rep[kk]);\n}\n\n\n\nFollowing Andrew Gelman, Meng, and Stern (1996), prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.\n\n\nFor example, posterior predictive replication may be formulated using distribution notation as follows. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] The heavily overloaded distribution notation is meant to indicate that both \\(y\\) and \\(y^{\\textrm{rep}}\\) are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities \\(p_{Y^{\\textrm{rep}} \\mid\n\\Theta}\\) and \\(p_{Y \\mid \\Theta}\\) are the same.\nThe joint density is \\[\np(\\theta, y, y^{\\textrm{rep}})\n= p(\\theta) \\cdot p(y \\mid \\theta) \\cdot p(y^{\\textrm{rep}} \\mid \\theta).\n\\] This again is assuming that the two distributions for \\(y\\) and \\(y^{\\textrm{rep}}\\) are identical.\nThe variable \\(y\\) is observed, with the predictive simulation \\(y^{\\textrm{rep}}\\) and parameter vector \\(\\theta\\) not observed. The posterior is \\(p(y^{\\textrm{rep}}, \\theta \\mid y)\\). Given draws from the posterior, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are retained.\n\n\n\nThe prior predictive model simply drops the data component of the posterior predictive model. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] This corresponds to the joint density \\[\np(\\theta, y^{\\textrm{rep}}) = p(\\theta) \\cdot p(y^{\\textrm{rep}} \\mid\n\\theta).\n\\]\nIt is typically straightforward to draw \\(\\theta\\) from the prior and \\(y^{\\textrm{rep}}\\) from the data model given \\(\\theta\\) efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.\n\n\n\nThe mixed replication corresponds to the model \\[\\begin{eqnarray*}\n\\phi & \\sim & p(\\phi)\n\\\\[2pt]\n\\alpha & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\alpha)\n\\\\[2pt]\n\\alpha^{\\textrm{rep}} & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\phi)\n\\end{eqnarray*}\\] The notation here is meant to indicate that \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\) have identical distributions, as do \\(y\\) and \\(y^{\\textrm{rep}}\\).\nThis corresponds to a joint model \\[\np(\\phi, \\alpha, \\alpha^{\\textrm{rep}}, y, y^{\\textrm{rep}})\n=\np(\\phi)\n\\cdot p(\\alpha \\mid \\phi)\n\\cdot p(y \\mid \\alpha)\n\\cdot p(\\alpha^{\\textrm{rep}} \\mid \\phi)\n\\cdot p(y^{\\textrm{rep}} \\mid \\alpha^{\\textrm{rep}}),\n\\] where \\(y\\) is the only observed variable, \\(\\alpha\\) contains the lower-level parameters and \\(\\phi\\) the hyperparameters. Note that \\(\\phi\\) is not replicated and instead appears in the distribution for both \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\).\nThe posterior is \\(p(\\phi, \\alpha, \\alpha^{\\textrm{rep}},\ny^{\\textrm{rep}} \\mid y)\\). From posterior draws, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are kept.",
+    "text": "A Bayesian posterior is calibrated if the posterior intervals have appropriate coverage. For example, 80% intervals are expected to contain the true parameter 80% of the time. If data is generated according to a model, Bayesian posterior inference with respect to that model is calibrated by construction. Simulation-based calibration checking (SBC) exploits this property of Bayesian inference to assess the soundness of a posterior sampler. Roughly, the way it works is by simulating parameters according to the prior, then simulating data conditioned on the simulated parameters, then testing posterior calibration of the inference algorithm over independently simulated data sets. This chapter follows Talts et al. (2018), which improves on the original approach developed by Cook, Gelman, and Rubin (2006). See also Modrák et al. (2023) for further improvements.\n\n\nSuppose a Bayesian model is given in the form of a prior density \\(p(\\theta)\\) and sampling density \\(p(y \\mid \\theta).\\) Now consider a process that first simulates parameters from the prior, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta),\n\\] and then simulates data given the parameters, \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}}).\n\\] By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta).\n\\] From Bayes’s rule, it follows that for any observed (fixed) data \\(y\\), \\[\np(\\theta \\mid y) \\propto p(y, \\theta).\n\\] Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, \\[\n\\theta^{(1)}, \\ldots, \\theta^{(M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Because \\(\\theta^{\\textrm{sim}}\\) is also distributed as a draw from the posterior, the rank statistics of \\(\\theta^{\\textrm{sim}}\\) with respect to \\(\\theta^{(1)}, \\ldots \\theta^{(M)}\\) should be uniform.\nThis is one way to define calibration, because it follows that posterior intervals will have appropriate coverage (Dawid 1982; Gneiting, Balabdaoui, and Raftery 2007). If the rank of \\(\\theta^{\\textrm{sim}}\\) is uniform among the draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)},\\) then for any 90% interval selected, the probability the true value \\(\\theta^{\\textrm{sim}}\\) falls in it will also be 90%. The same goes for any other posterior interval.\n\n\n\nSuppose the Bayesian model to test has joint density \\[\np(y, \\theta) = p(y \\mid \\theta) \\cdot p(\\theta),\n\\] with data \\(y\\) and parameters \\(\\theta\\) (both are typically multivariate). Simulation-based calibration checking works by generating \\(N\\) simulated parameter and data pairs according to the joint density, \\[\n(y^{\\textrm{sim}(1)}, \\theta^{\\textrm{sim}(1)}),\n\\ldots, (y^{\\textrm{sim}(N)}, \\theta^{\\textrm{sim}(N)}),\n\\sim p(y, \\theta).\n\\] For each simulated data set \\(y^{\\textrm{sim}(n)}\\), use the algorithm to be tested to generate \\(M\\) posterior draws, which if everything is working properly, will be distributed marginally as \\[\n\\theta^{(n, 1)}, \\ldots, \\theta^{(n, M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}(n)}).\n\\] For a simulation \\(n\\) and parameter \\(k\\), the rank of the simulated parameter among the posterior draws is \\[\\begin{eqnarray*}\nr_{n, k}\n& = &\n\\textrm{rank}(\\theta_k^{\\textrm{sim}(n)},\n              (\\theta^{(n, 1)}, \\ldots, \\theta^{(n,M)}))\n\\\\[4pt]\n& = &\n\\sum_{m = 1}^M\n  \\textrm{I}[\\theta_k^{(n,m)} &lt; \\theta_k^{\\textrm{sim}(n)}].\n\\end{eqnarray*}\\] That is, the rank is the number of posterior draws \\(\\theta^{(n,m)}_k\\) that are less than the simulated draw \\(\\theta^{\\textrm{sim}(n)}_k.\\)\nIf the algorithm generates posterior draws according to the posterior, the ranks should have uniform discrete distribution from \\(0\\) to \\(M\\), so that the ranks plus one are uniformly distributed from \\(1\\) to \\(M + 1\\), \\[\nr_{n, k} + 1\n\\sim\n\\textrm{categorical}\\! \\left(\\frac{1}{M + 1}, \\ldots, \\frac{1}{M + 1}\\right).\n\\] Simulation-based calibration checking uses this expected behavior to test the calibration of each parameter of a model on simulated data. Talts et al. (2018) suggest plotting binned counts of \\(r_{1:N,\nk}\\) for different parameters \\(k\\); Säilynoja, Bürkner, and Vehtari (2022) provide a graphical test for discrete uniformity testing. Before uniformity testing the Markov chains should be thinned to remove autocorrelation as these uniformity tests assume independence (Säilynoja, Bürkner, and Vehtari 2022).\n\n\n\nRunning simulation-based calibration checking in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.\n\n\nFor illustration, a very simple model will suffice. Suppose there are two parameters \\((\\mu, \\sigma)\\) with independent priors, \\[\n\\mu \\sim \\textrm{normal}(0, 1),\n\\] and \\[\n\\sigma \\sim \\textrm{lognormal}(0, 1).\n\\] The data \\(y = y_1, \\ldots, y_N\\) is drawn conditionally independently given the parameters, \\[\ny_n \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The joint prior density is thus \\[\np(\\mu, \\sigma)\n= \\textrm{normal}(\\mu \\mid 0, 1)\n  \\cdot \\textrm{lognormal}(\\sigma \\mid 0, 1),\n\\] and the data model is \\[\np(y \\mid \\mu, \\sigma)\n= \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, \\sigma).\n\\]\nFor example, suppose the following two parameter values are drawn from the prior in the first simulation, \\[\n(\\mu^{\\textrm{sim(1)}}, \\sigma^{\\textrm{sim(1)}}) = (1.01, 0.23).\n\\] Then data \\(y^{\\textrm{sim}(1)} \\sim p(y \\mid \\mu^{\\textrm{sim(1)}},\n\\sigma^{\\textrm{sim(1)}})\\) is drawn according to the data model. Next, \\(M = 4\\) draws are taken from the posterior \\(\\mu^{(1,m)}, \\sigma^{(1,m)} \\sim p(\\mu, \\sigma \\mid y^{\\textrm{sim}(1)})\\), \\[\n\\begin{array}{r|rr}\nm & \\mu^{(1,m)} & \\sigma^{(1,m)}\n\\\\ \\hline\n1 & 1.07 & 0.33\n\\\\\n2 & -0.32 & 0.14\n\\\\\n3 & -0.99 & 0.26\n\\\\\n4 & 1.51 & 0.31\n\\end{array}\n\\] Then the comparisons on which ranks are based look as follows, \\[\n\\begin{array}{r|cc}\nm & \\textrm{I}(\\mu^{(1,m)} &lt; \\mu^{\\textrm{sim}(1)})\n& \\textrm{I}(\\sigma^{(1,m)} &lt; \\sigma^{\\textrm{sim}(1)})\n\\\\ \\hline\n1 & 0 & 0\n\\\\\n2 & 1 & 1\n\\\\\n3 & 1 & 0\n\\\\\n4 & 0 & 0\n\\end{array}\n\\] The ranks are the column sums, \\(r_{1,1} = 2\\) and \\(r_{1,2} = 1\\). Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between \\(0\\) and \\(M\\), the number of posterior draws.\n\n\n\nTo code simulation-based calibration checking in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = normal_rng(mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nTo avoid confusion with the number of simulated data sets used for simulation-based calibration checking, J is used for the number of simulated data points.\nThe model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.\n\n\n\nThe entire simulation-based calibration checking process is as follows, where\n\np(theta) is the prior density\np(y | theta) is the sampling density\nK is the number of parameters\nN is the total number of simulated data sets and fits\nM is the number of posterior draws per simulated data set\n\nSBC(p(theta), p(y | theta), K, N, M)\n------------------------------------\nfor (n in 1:N) {\n    // simulate parameters and data\n    theta(sim(n)) ~ p(theta)\n    y(sim(n)) ~ p(y | theta(sim(n)))\n\n    // posterior draws given simulated data\n    for (m in 1:M) {\n        theta(n, m) ~ p(theta | y(sim(n)))\n    }\n    // calculate rank of sim among posterior draws\n    for (k in 1:K) {\n        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))\n    }\n}\n// test uniformity of each parameter\nfor (k in 1:K) {\n    test uniformity of rank(1:N, k)\n}\n\n\n\nThe draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws (Säilynoja, Bürkner, and Vehtari 2022). Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of M, then thinning down to M draws.\n\n\n\n\nA simple, though not very highly powered, \\(\\chi^2\\)-squared test for uniformity can be formulated by binning the ranks \\(0:M\\) into \\(J\\) bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, Säilynoja, Bürkner, and Vehtari (2022) use binomial model pointiwise for the empirical cumlative distribution function and adjust to obtain simulatenous envelope to be used as graphical uniformity test.\nThe bins don’t need to be exactly the same size. In general, if \\(b_j\\) is the number of ranks that fall into bin \\(j\\) and \\(e_j\\) is the number of ranks expected to fall into bin \\(j\\) (which will be proportional to its size under uniformity), the test statistic is \\[\nX^2 = \\sum_{j = 1}^J \\frac{(b_j - e_j)^2}{e_j}.\n\\] The terms are approximately square standard normal, so that under the null hypothesis of uniformity, \\[\nX^2 \\sim \\textrm{chiSquared}(J - 1),\n\\] with the corresponding \\(p\\)-value given by the complementary cumulative distribution function (CCDF) of \\(\\textrm{chiSquared}(J - 1)\\) applied to \\(X^2\\). Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., \\(e_j \\geq 5.\\)\n\n\nBecause there are \\(M + 1\\) possible ranks, with \\(J\\) bins, it is easiest to have \\(M + 1\\) be divisible by \\(J\\). For instance, if \\(J = 20\\) and \\(M = 999\\), then there are \\(1000\\) possible ranks and an expected count in each bin of \\(\\frac{M + 1}{J} = 50.\\)\nDistributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, \\[\n\\textrm{bin}(r_{n, m}, M, J)\n= 1 + \\left\\lfloor \\frac{r_{n, m}}{(M + 1) / J} \\right\\rfloor.\n\\] For example, with \\(M = 999\\) and \\(J = 20\\), \\((M + 1) / J = 50\\). The lowest rank checks out, \\[\n\\textrm{bin}(0, 999, 20) = 1 + \\lfloor 0 / 50 \\rfloor = 1,\n\\] as does the 50th rank, \\[\n\\textrm{bin}(49, 999, 20) = 1 + \\lfloor 49 / 50 \\rfloor = 1,\n\\] and the 51st is appropriately put in the second bin, \\[\n\\textrm{bin}(50, 999, 20) = 1 + \\lfloor 50 / 50 \\rfloor = 2.\n\\] The highest rank also checks out, with \\(\\textrm{bin}(1000, 999, 20) = 50.\\)\nTo summarize, the following pseudocode computes the \\(b_j\\) values for the \\(\\chi^2\\) test or for visualization in a histogram.\nInputs: M draws, J bins, N parameters, ranks r[n, m]\nb[1:J] = 0\nfor (m in 1:M) {\n  ++b[1 + floor(r[n, m] * J / (M + 1))]\n}\nwhere the ++b[n] notation is a common form of syntactic sugar for b[n] = b[n] + 1.\nIn general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.\n\n\n\n\nThis section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.\n\n\nConsider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 1)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1)\n\\\\[4pt]\ny_{1:10} & \\sim & \\textrm{normal}(\\mu, \\sigma).\n\\end{eqnarray*}\\] The Stan program for evaluating SBC for this model is\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = student_t_rng(4, mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; I_lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nAfter running this for enough iterations so that the effective sample size is larger than \\(M\\), then thinning to \\(M\\) draws (here \\(M = 999\\)), the ranks are computed and binned, and then plotted.\n\n\n\n\n\n\nFigure 1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.\n\n\n\n\n\n\nNow consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.\n\n\n\n\n\n\nFigure 2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.\n\n\n\n\n\n\nThe example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.\nThis can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means \\(y_j\\) and standard deviations \\(\\sigma_j\\) of differences in test score after the same intervention in \\(J = 8\\) different schools. Rubin (1981) applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by Talts et al. (2018); this section replicates their results.\nThe meta-analysis model has parameters for a population mean \\(\\mu\\) and standard deviation \\(\\tau &gt; 0\\) as well as the effect \\(\\theta_j\\) of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 5)\n\\\\[4pt]\n\\tau & \\sim & \\textrm{normal}_{+}(0, 5).\n\\end{eqnarray*}\\] School level effects are modeled as normal given the population parameters, \\[\n\\theta_j \\sim \\textrm{normal}(\\mu, \\tau).\n\\] The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, \\[\ny_j \\sim \\textrm{normal}(\\theta_j, \\sigma_j).\n\\]\nThis model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.\ntransformed data {\n  real mu_sim = normal_rng(0, 5);\n  real tau_sim = abs(normal_rng(0, 5));\n  int&lt;lower=0&gt; J = 8;\n  array[J] real theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);\n  array[J] real&lt;lower=0&gt; sigma = abs(normal_rng(rep_vector(0, J), 5));\n  array[J] real y = normal_rng(theta_sim, sigma);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; tau;\n  array[J] real theta;\n}\nmodel {\n  tau ~ normal(0, 5);\n  mu ~ normal(0, 5);\n  theta ~ normal(mu, tau);\n  y ~ normal(theta, sigma);\n}\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mu_lt_sim = mu &lt; mu_sim;\n  int&lt;lower=0, upper=1&gt; tau_lt_sim = tau &lt; tau_sim;\n  int&lt;lower=0, upper=1&gt; theta1_lt_sim = theta[1] &lt; theta_sim[1];\n}\nAs usual for simulation-based calibration checking, the transformed data encodes the data-generating process using random number generators. Here, the population parameters \\(\\mu\\) and \\(\\tau\\) are first simulated, then the school-level effects \\(\\theta\\), and then finally the observed data \\(\\sigma_j\\) and \\(y_j.\\) The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only \\(\\theta_1\\) because the schools are exchangeable in the simulation.\nWhen fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.\nSBC also diagnoses the problem. Here’s the rank plots for running \\(N =\n200\\) simulations with 1000 warmup iterations and \\(M = 999\\) draws per simulation used to compute the ranks.\n\n\n\n\n\n\n\n\n\n\n\n(a) \\(\\mu\\)\n\n\n\n\n\n\n\n\n\n\n\n(b) \\(\\tau\\)\n\n\n\n\n\n\n\n\n\n\n\n(c) \\({\\theta}_1\\)\n\n\n\n\n\n\n\nFigure 3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for \\(\\theta_1\\) (Figure 3 (c)).\n\n\n\nAlthough the population mean and standard deviation \\(\\mu\\) and \\(\\tau\\) appear well calibrated, \\(\\theta_1\\) tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as \\(\\tau\\) approaches zero and the \\(\\theta_j\\) become highly constrained. The chapter on reparameterization explains how to remedy this problem and fit this kind of hierarchical model with Stan.",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Simulation-Based Calibration Checking"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#simulating-from-the-posterior-predictive-distribution",
-    "href": "stan-users-guide/posterior-predictive-checks.html#simulating-from-the-posterior-predictive-distribution",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/simulation-based-calibration.html#bayes-is-calibrated-by-construction",
+    "href": "stan-users-guide/simulation-based-calibration.html#bayes-is-calibrated-by-construction",
+    "title": "Simulation-Based Calibration Checking",
     "section": "",
-    "text": "The posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data \\(y\\).\nThe posterior predictive distribution for replications \\(y^{\\textrm{rep}}\\) of the original data set \\(y\\) given model parameters \\(\\theta\\) is defined by \\[\np(y^{\\textrm{rep}} \\mid y)\n= \\int p(y^{\\textrm{rep}} \\mid \\theta)\n       \\cdot p(\\theta \\mid y) \\, \\textrm{d}\\theta.\n\\]\nAs with other posterior predictive quantities, generating a replicated data set \\(y^{\\textrm{rep}}\\) from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters \\(\\theta = (\\alpha, \\beta, \\sigma).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  y ~ normal(alpha + beta * x, sigma);\n}\nTo generate a replicated data set y_rep for this simple model, the following generated quantities block suffices.\ngenerated quantities {\n  array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n}\nThe vectorized form of the normal random number generator is used with the original predictors x and the model parameters alpha, beta, and sigma. The replicated data variable y_rep is declared to be the same size as the original data y, but instead of a vector type, it is declared to be an array of reals to match the return type of the function normal_rng. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.\nThe posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in the chapter on posterior predictions in that the original predictors \\(x\\) are used. That is, the posterior predictions are for the original data.",
+    "text": "Suppose a Bayesian model is given in the form of a prior density \\(p(\\theta)\\) and sampling density \\(p(y \\mid \\theta).\\) Now consider a process that first simulates parameters from the prior, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta),\n\\] and then simulates data given the parameters, \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}}).\n\\] By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta).\n\\] From Bayes’s rule, it follows that for any observed (fixed) data \\(y\\), \\[\np(\\theta \\mid y) \\propto p(y, \\theta).\n\\] Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, \\[\n\\theta^{(1)}, \\ldots, \\theta^{(M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Because \\(\\theta^{\\textrm{sim}}\\) is also distributed as a draw from the posterior, the rank statistics of \\(\\theta^{\\textrm{sim}}\\) with respect to \\(\\theta^{(1)}, \\ldots \\theta^{(M)}\\) should be uniform.\nThis is one way to define calibration, because it follows that posterior intervals will have appropriate coverage (Dawid 1982; Gneiting, Balabdaoui, and Raftery 2007). If the rank of \\(\\theta^{\\textrm{sim}}\\) is uniform among the draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)},\\) then for any 90% interval selected, the probability the true value \\(\\theta^{\\textrm{sim}}\\) falls in it will also be 90%. The same goes for any other posterior interval.",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Simulation-Based Calibration Checking"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#plotting-multiples",
-    "href": "stan-users-guide/posterior-predictive-checks.html#plotting-multiples",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/simulation-based-calibration.html#simulation-based-calibration-checking-1",
+    "href": "stan-users-guide/simulation-based-calibration.html#simulation-based-calibration-checking-1",
+    "title": "Simulation-Based Calibration Checking",
     "section": "",
-    "text": "A standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.\nHere’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\ntransformed data {\n  real&lt;lower=0&gt; mean_y = mean(to_vector(y));\n  real&lt;lower=0&gt; sd_y = sd(to_vector(y));\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  y ~ poisson(lambda);\n  lambda ~ exponential(0.2);\n}\ngenerated quantities {\n  array[N] int&lt;lower=0&gt; y_rep = poisson_rng(rep_array(lambda, N));\n  real&lt;lower=0&gt; mean_y_rep = mean(to_vector(y_rep));\n  real&lt;lower=0&gt; sd_y_rep = sd(to_vector(y_rep));\n  int&lt;lower=0, upper=1&gt; mean_gte = (mean_y_rep &gt;= mean_y);\n  int&lt;lower=0, upper=1&gt; sd_gte = (sd_y_rep &gt;= sd_y);\n}\nThe generated quantities block creates a variable y_rep for the replicated data, variables mean_y_rep and sd_y_rep for the statistics of the replicated data, and indicator variables mean_gte and sd_gte for whether the replicated statistic is greater than or equal to the statistic applied to the original data.\nNow consider generating data \\(y \\sim \\textrm{Poisson}(5)\\). The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.\n\n\n\nPosterior predictive checks for Poisson data generating process and Poisson model.\n\n\nWith a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.\nNow consider generating over-dispersed data \\(y \\sim \\textrm{negative-binomial2}(5, 1).\\) This has the same mean as \\(\\textrm{Poisson}(5)\\), namely \\(5\\), but a standard deviation of \\(\\sqrt{5 + 5^2 /1} \\approx 5.5.\\) There is no way to fit this data with the Poisson model, because a variable distributed as \\(\\textrm{Poisson}(\\lambda)\\) has mean \\(\\lambda\\) and standard deviation \\(\\sqrt{\\lambda},\\) which is \\(\\sqrt{5}\\) for \\(\\textrm{Poisson}(5).\\) Here’s the resulting small multiples plot, again with original data in the upper left.\n\n\n\nPosterior predictive checks for negative binomial data generating process and Poisson model.\n\n\nThis time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.",
+    "text": "Suppose the Bayesian model to test has joint density \\[\np(y, \\theta) = p(y \\mid \\theta) \\cdot p(\\theta),\n\\] with data \\(y\\) and parameters \\(\\theta\\) (both are typically multivariate). Simulation-based calibration checking works by generating \\(N\\) simulated parameter and data pairs according to the joint density, \\[\n(y^{\\textrm{sim}(1)}, \\theta^{\\textrm{sim}(1)}),\n\\ldots, (y^{\\textrm{sim}(N)}, \\theta^{\\textrm{sim}(N)}),\n\\sim p(y, \\theta).\n\\] For each simulated data set \\(y^{\\textrm{sim}(n)}\\), use the algorithm to be tested to generate \\(M\\) posterior draws, which if everything is working properly, will be distributed marginally as \\[\n\\theta^{(n, 1)}, \\ldots, \\theta^{(n, M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}(n)}).\n\\] For a simulation \\(n\\) and parameter \\(k\\), the rank of the simulated parameter among the posterior draws is \\[\\begin{eqnarray*}\nr_{n, k}\n& = &\n\\textrm{rank}(\\theta_k^{\\textrm{sim}(n)},\n              (\\theta^{(n, 1)}, \\ldots, \\theta^{(n,M)}))\n\\\\[4pt]\n& = &\n\\sum_{m = 1}^M\n  \\textrm{I}[\\theta_k^{(n,m)} &lt; \\theta_k^{\\textrm{sim}(n)}].\n\\end{eqnarray*}\\] That is, the rank is the number of posterior draws \\(\\theta^{(n,m)}_k\\) that are less than the simulated draw \\(\\theta^{\\textrm{sim}(n)}_k.\\)\nIf the algorithm generates posterior draws according to the posterior, the ranks should have uniform discrete distribution from \\(0\\) to \\(M\\), so that the ranks plus one are uniformly distributed from \\(1\\) to \\(M + 1\\), \\[\nr_{n, k} + 1\n\\sim\n\\textrm{categorical}\\! \\left(\\frac{1}{M + 1}, \\ldots, \\frac{1}{M + 1}\\right).\n\\] Simulation-based calibration checking uses this expected behavior to test the calibration of each parameter of a model on simulated data. Talts et al. (2018) suggest plotting binned counts of \\(r_{1:N,\nk}\\) for different parameters \\(k\\); Säilynoja, Bürkner, and Vehtari (2022) provide a graphical test for discrete uniformity testing. Before uniformity testing the Markov chains should be thinned to remove autocorrelation as these uniformity tests assume independence (Säilynoja, Bürkner, and Vehtari 2022).",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Simulation-Based Calibration Checking"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#posterior-p-values",
-    "href": "stan-users-guide/posterior-predictive-checks.html#posterior-p-values",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/simulation-based-calibration.html#sbc-in-stan",
+    "href": "stan-users-guide/simulation-based-calibration.html#sbc-in-stan",
+    "title": "Simulation-Based Calibration Checking",
     "section": "",
-    "text": "If a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic \\(s(\\cdot)\\) in a replicated data set exceeds that in the original data, \\[\n\\Pr\\!\\left[ s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right]\n=\n\\int\n\\textrm{I}\\left( s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right)\n\\cdot p\\left( y^{\\textrm{rep}} \\mid y \\right)\n\\, \\textrm{d}{y^{\\textrm{rep}}}.\n\\] It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified (Bayarri and Berger 2000).\nNevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.\nTo calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mean_gt;\n  int&lt;lower=0, upper=1&gt; sd_gt;\n  {\n    array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n    mean_gt = mean(y_rep) &gt; mean(y);\n    sd_gt = sd(y_rep) &gt; sd(y);\n  }\n}\nThe indicator variable mean_gt will have value 1 if the mean of the simulated data y_rep is greater than or equal to the mean of he original data y. Because the values of y_rep are not needed for the posterior predictive checks, the program saves output space by using a local variable for y_rep. The statistics mean(u) and sd(y) could also be computed in the transformed data block and saved.\nFor the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.\n\n\n\nHistogram of means of replicated data sets; vertical red line at mean of original data.\n\n\nThe p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior \\(p\\)-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable mean_gt.\nThe standard deviation statistic tells a different story.\n\n\n\nScatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.\n\n\nHere, the original data has much higher standard deviation than any of the replicated data sets. The resulting \\(p\\)-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small \\(p\\)-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.\n\n\nAny statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.\nDespite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.",
+    "text": "Running simulation-based calibration checking in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.\n\n\nFor illustration, a very simple model will suffice. Suppose there are two parameters \\((\\mu, \\sigma)\\) with independent priors, \\[\n\\mu \\sim \\textrm{normal}(0, 1),\n\\] and \\[\n\\sigma \\sim \\textrm{lognormal}(0, 1).\n\\] The data \\(y = y_1, \\ldots, y_N\\) is drawn conditionally independently given the parameters, \\[\ny_n \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The joint prior density is thus \\[\np(\\mu, \\sigma)\n= \\textrm{normal}(\\mu \\mid 0, 1)\n  \\cdot \\textrm{lognormal}(\\sigma \\mid 0, 1),\n\\] and the data model is \\[\np(y \\mid \\mu, \\sigma)\n= \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, \\sigma).\n\\]\nFor example, suppose the following two parameter values are drawn from the prior in the first simulation, \\[\n(\\mu^{\\textrm{sim(1)}}, \\sigma^{\\textrm{sim(1)}}) = (1.01, 0.23).\n\\] Then data \\(y^{\\textrm{sim}(1)} \\sim p(y \\mid \\mu^{\\textrm{sim(1)}},\n\\sigma^{\\textrm{sim(1)}})\\) is drawn according to the data model. Next, \\(M = 4\\) draws are taken from the posterior \\(\\mu^{(1,m)}, \\sigma^{(1,m)} \\sim p(\\mu, \\sigma \\mid y^{\\textrm{sim}(1)})\\), \\[\n\\begin{array}{r|rr}\nm & \\mu^{(1,m)} & \\sigma^{(1,m)}\n\\\\ \\hline\n1 & 1.07 & 0.33\n\\\\\n2 & -0.32 & 0.14\n\\\\\n3 & -0.99 & 0.26\n\\\\\n4 & 1.51 & 0.31\n\\end{array}\n\\] Then the comparisons on which ranks are based look as follows, \\[\n\\begin{array}{r|cc}\nm & \\textrm{I}(\\mu^{(1,m)} &lt; \\mu^{\\textrm{sim}(1)})\n& \\textrm{I}(\\sigma^{(1,m)} &lt; \\sigma^{\\textrm{sim}(1)})\n\\\\ \\hline\n1 & 0 & 0\n\\\\\n2 & 1 & 1\n\\\\\n3 & 1 & 0\n\\\\\n4 & 0 & 0\n\\end{array}\n\\] The ranks are the column sums, \\(r_{1,1} = 2\\) and \\(r_{1,2} = 1\\). Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between \\(0\\) and \\(M\\), the number of posterior draws.\n\n\n\nTo code simulation-based calibration checking in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = normal_rng(mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nTo avoid confusion with the number of simulated data sets used for simulation-based calibration checking, J is used for the number of simulated data points.\nThe model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.\n\n\n\nThe entire simulation-based calibration checking process is as follows, where\n\np(theta) is the prior density\np(y | theta) is the sampling density\nK is the number of parameters\nN is the total number of simulated data sets and fits\nM is the number of posterior draws per simulated data set\n\nSBC(p(theta), p(y | theta), K, N, M)\n------------------------------------\nfor (n in 1:N) {\n    // simulate parameters and data\n    theta(sim(n)) ~ p(theta)\n    y(sim(n)) ~ p(y | theta(sim(n)))\n\n    // posterior draws given simulated data\n    for (m in 1:M) {\n        theta(n, m) ~ p(theta | y(sim(n)))\n    }\n    // calculate rank of sim among posterior draws\n    for (k in 1:K) {\n        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))\n    }\n}\n// test uniformity of each parameter\nfor (k in 1:K) {\n    test uniformity of rank(1:N, k)\n}\n\n\n\nThe draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws (Säilynoja, Bürkner, and Vehtari 2022). Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of M, then thinning down to M draws.",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Simulation-Based Calibration Checking"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#prior-predictive-checks",
-    "href": "stan-users-guide/posterior-predictive-checks.html#prior-predictive-checks",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/simulation-based-calibration.html#testing-uniformity",
+    "href": "stan-users-guide/simulation-based-calibration.html#testing-uniformity",
+    "title": "Simulation-Based Calibration Checking",
     "section": "",
-    "text": "Prior predictive checks generate data according to the prior in order to asses whether a prior is appropriate (Gabry et al. 2019). A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, \\[\ny^{\\textrm{sim}} \\sim p(y).\n\\] The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.\nThis is easy to carry out mechanically by simulating parameters \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta)\n\\] according to the priors, then simulating data \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}})\n\\] according to the data model given the simulated parameters. The result is a simulation from the joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta)\n\\] and thus \\[\ny^{\\textrm{sim}} \\sim p(y)\n\\] is a simulation from the prior predictive distribution.\n\n\nA prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.\nPrior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n}\ngenerated quantities {\n  real alpha = normal_rng(0, 1);\n  real beta = normal_rng(0, 1);\n  array[N] real y_sim = poisson_log_rng(alpha + beta * x);\n}\nRunning this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.",
+    "text": "A simple, though not very highly powered, \\(\\chi^2\\)-squared test for uniformity can be formulated by binning the ranks \\(0:M\\) into \\(J\\) bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, Säilynoja, Bürkner, and Vehtari (2022) use binomial model pointiwise for the empirical cumlative distribution function and adjust to obtain simulatenous envelope to be used as graphical uniformity test.\nThe bins don’t need to be exactly the same size. In general, if \\(b_j\\) is the number of ranks that fall into bin \\(j\\) and \\(e_j\\) is the number of ranks expected to fall into bin \\(j\\) (which will be proportional to its size under uniformity), the test statistic is \\[\nX^2 = \\sum_{j = 1}^J \\frac{(b_j - e_j)^2}{e_j}.\n\\] The terms are approximately square standard normal, so that under the null hypothesis of uniformity, \\[\nX^2 \\sim \\textrm{chiSquared}(J - 1),\n\\] with the corresponding \\(p\\)-value given by the complementary cumulative distribution function (CCDF) of \\(\\textrm{chiSquared}(J - 1)\\) applied to \\(X^2\\). Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., \\(e_j \\geq 5.\\)\n\n\nBecause there are \\(M + 1\\) possible ranks, with \\(J\\) bins, it is easiest to have \\(M + 1\\) be divisible by \\(J\\). For instance, if \\(J = 20\\) and \\(M = 999\\), then there are \\(1000\\) possible ranks and an expected count in each bin of \\(\\frac{M + 1}{J} = 50.\\)\nDistributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, \\[\n\\textrm{bin}(r_{n, m}, M, J)\n= 1 + \\left\\lfloor \\frac{r_{n, m}}{(M + 1) / J} \\right\\rfloor.\n\\] For example, with \\(M = 999\\) and \\(J = 20\\), \\((M + 1) / J = 50\\). The lowest rank checks out, \\[\n\\textrm{bin}(0, 999, 20) = 1 + \\lfloor 0 / 50 \\rfloor = 1,\n\\] as does the 50th rank, \\[\n\\textrm{bin}(49, 999, 20) = 1 + \\lfloor 49 / 50 \\rfloor = 1,\n\\] and the 51st is appropriately put in the second bin, \\[\n\\textrm{bin}(50, 999, 20) = 1 + \\lfloor 50 / 50 \\rfloor = 2.\n\\] The highest rank also checks out, with \\(\\textrm{bin}(1000, 999, 20) = 50.\\)\nTo summarize, the following pseudocode computes the \\(b_j\\) values for the \\(\\chi^2\\) test or for visualization in a histogram.\nInputs: M draws, J bins, N parameters, ranks r[n, m]\nb[1:J] = 0\nfor (m in 1:M) {\n  ++b[1 + floor(r[n, m] * J / (M + 1))]\n}\nwhere the ++b[n] notation is a common form of syntactic sugar for b[n] = b[n] + 1.\nIn general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Simulation-Based Calibration Checking"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#example-of-prior-predictive-checks",
-    "href": "stan-users-guide/posterior-predictive-checks.html#example-of-prior-predictive-checks",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/simulation-based-calibration.html#examples-of-simulation-based-calibration-checking",
+    "href": "stan-users-guide/simulation-based-calibration.html#examples-of-simulation-based-calibration-checking",
+    "title": "Simulation-Based Calibration Checking",
     "section": "",
-    "text": "Suppose we have a model for a football (aka soccer) league where there are \\(J\\) teams. Each team has a scoring rate \\(\\lambda_j\\) and in each game will be assumed to score \\(\\textrm{poisson}(\\lambda_j)\\) points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed \\[\n\\lambda_j \\sim \\textrm{gamma}(\\epsilon_1, \\epsilon_2).\n\\] This is not just a manufactured example; The BUGS Book recommends setting \\(\\epsilon = (0.5, 0.00001)\\), which corresponds to a Jeffreys prior for a Poisson rate parameter prior (Lunn et al. 2012, 85).\nSuppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.\ndata {\n  int&lt;lower=0&gt; J;\n  array[2] real&lt;lower=0&gt; epsilon;\n}\ngenerated quantities {\n  array[J] real&lt;lower=0&gt; lambda;\n  array[J, J] int y;\n  for (j in 1:J) lambda[j] = gamma_rng(epsilon[1], epsilon[2]);\n  for (i in 1:J) {\n    for (j in 1:J) {\n      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);\n    }\n  }\n}\nIn this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are \\(\\binom{J}{2} = \\frac{J \\cdot (J - 1)}{2}.\\)\nUsing the \\(\\textrm{gamma}(0.5, 0.00001)\\) reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, \\(y^{(1:20)}_{1, 2}\\).\n2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987\n7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66\nThat’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this \\(\\textrm{gamma}(0.5, 0.00001)\\) prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.\nThe posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values (A. Gelman 2006).\nJust as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.\nIn this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see (Gabry et al. 2019).\nPrior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.",
+    "text": "This section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.\n\n\nConsider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 1)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1)\n\\\\[4pt]\ny_{1:10} & \\sim & \\textrm{normal}(\\mu, \\sigma).\n\\end{eqnarray*}\\] The Stan program for evaluating SBC for this model is\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = student_t_rng(4, mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; I_lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nAfter running this for enough iterations so that the effective sample size is larger than \\(M\\), then thinning to \\(M\\) draws (here \\(M = 999\\)), the ranks are computed and binned, and then plotted.\n\n\n\n\n\n\nFigure 1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.\n\n\n\n\n\n\nNow consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.\n\n\n\n\n\n\nFigure 2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.\n\n\n\n\n\n\nThe example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.\nThis can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means \\(y_j\\) and standard deviations \\(\\sigma_j\\) of differences in test score after the same intervention in \\(J = 8\\) different schools. Rubin (1981) applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by Talts et al. (2018); this section replicates their results.\nThe meta-analysis model has parameters for a population mean \\(\\mu\\) and standard deviation \\(\\tau &gt; 0\\) as well as the effect \\(\\theta_j\\) of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 5)\n\\\\[4pt]\n\\tau & \\sim & \\textrm{normal}_{+}(0, 5).\n\\end{eqnarray*}\\] School level effects are modeled as normal given the population parameters, \\[\n\\theta_j \\sim \\textrm{normal}(\\mu, \\tau).\n\\] The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, \\[\ny_j \\sim \\textrm{normal}(\\theta_j, \\sigma_j).\n\\]\nThis model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.\ntransformed data {\n  real mu_sim = normal_rng(0, 5);\n  real tau_sim = abs(normal_rng(0, 5));\n  int&lt;lower=0&gt; J = 8;\n  array[J] real theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);\n  array[J] real&lt;lower=0&gt; sigma = abs(normal_rng(rep_vector(0, J), 5));\n  array[J] real y = normal_rng(theta_sim, sigma);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; tau;\n  array[J] real theta;\n}\nmodel {\n  tau ~ normal(0, 5);\n  mu ~ normal(0, 5);\n  theta ~ normal(mu, tau);\n  y ~ normal(theta, sigma);\n}\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mu_lt_sim = mu &lt; mu_sim;\n  int&lt;lower=0, upper=1&gt; tau_lt_sim = tau &lt; tau_sim;\n  int&lt;lower=0, upper=1&gt; theta1_lt_sim = theta[1] &lt; theta_sim[1];\n}\nAs usual for simulation-based calibration checking, the transformed data encodes the data-generating process using random number generators. Here, the population parameters \\(\\mu\\) and \\(\\tau\\) are first simulated, then the school-level effects \\(\\theta\\), and then finally the observed data \\(\\sigma_j\\) and \\(y_j.\\) The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only \\(\\theta_1\\) because the schools are exchangeable in the simulation.\nWhen fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.\nSBC also diagnoses the problem. Here’s the rank plots for running \\(N =\n200\\) simulations with 1000 warmup iterations and \\(M = 999\\) draws per simulation used to compute the ranks.\n\n\n\n\n\n\n\n\n\n\n\n(a) \\(\\mu\\)\n\n\n\n\n\n\n\n\n\n\n\n(b) \\(\\tau\\)\n\n\n\n\n\n\n\n\n\n\n\n(c) \\({\\theta}_1\\)\n\n\n\n\n\n\n\nFigure 3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for \\(\\theta_1\\) (Figure 3 (c)).\n\n\n\nAlthough the population mean and standard deviation \\(\\mu\\) and \\(\\tau\\) appear well calibrated, \\(\\theta_1\\) tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as \\(\\tau\\) approaches zero and the \\(\\theta_j\\) become highly constrained. The chapter on reparameterization explains how to remedy this problem and fit this kind of hierarchical model with Stan.",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Simulation-Based Calibration Checking"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#mixed-replication",
-    "href": "stan-users-guide/posterior-predictive-checks.html#mixed-replication",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/regression.html",
+    "href": "stan-users-guide/regression.html",
+    "title": "Regression Models",
     "section": "",
-    "text": "Andrew Gelman, Meng, and Stern (1996) discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.\nFor example, consider a simple varying intercept logistic regression, with intercepts \\(\\alpha_k\\) for \\(k \\in 1:K\\). Each data item \\(y_n \\in \\{ 0, 1 \\}\\) is assumed to correspond to group \\(kk_n \\in 1:K.\\) The data model is thus \\[\ny_n \\sim \\textrm{bernoulli}(\\textrm{logit}^{-1}(\\alpha_{kk[n]})).\n\\] The varying intercepts have a hierarchical normal prior, \\[\n\\alpha_k \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The hyperparameters are themselves given weakly informative priors, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 2)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\]\nLike in a posterior predictive check, the hyperparameters \\(\\mu\\) and \\(\\sigma\\) are drawn from the posterior, \\[\n\\mu^{(m)}, \\sigma^{(m)} \\sim p(\\mu, \\sigma \\mid y)\n\\] Like in a prior predictive check, replicated values of \\(\\alpha\\) are drawn from the hyperparameters, \\[\n\\alpha^{\\textrm{rep}(m)}_k \\sim \\textrm{normal}(\\alpha_k \\mid\n\\mu^{(m)}, \\sigma^{(m)}).\n\\] The data items are then each replicated using the replicated intercepts, \\[\ny^{\\textrm{rep}(m)}_n \\sim\n\\textrm{bernoulli}\n  (\\textrm{logit}^{-1}(\\alpha^{\\textrm{rep}(m)}_{kk[n]})).\n\\] Thus the \\(y^{\\textrm{rep}(m)}\\) can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original \\(K\\) groups.\nIn Stan, mixed predictive replications \\(y^{\\textrm{rep}(m)}\\) can be programmed directly.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;offset=mu, multiplier=sigma&gt;[K] alpha;\n}\nmodel {\n  mu ~ normal(0, 2);               // hyperprior\n  sigma ~ lognormal(0, 1);\n  alpha ~ normal(mu, sigma);       // hierarchical prior\n  y ~ bernoulli_logit(alpha[kk]);  // data model\n}\ngenerated quantities {\n  // alpha replicated;  mu and sigma not replicated\n  array[K] real alpha_rep\n    = normal_rng(rep_vector(mu, K), sigma);\n  array[N] int&lt;lower=0, upper=1&gt; y_rep\n    = bernoulli_logit_rng(alpha_rep[kk]);\n}",
+    "text": "Stan supports regression models from simple linear regressions to multilevel generalized linear models.\n\n\nThe simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\quad\\text{where}\\quad\n\\epsilon_n \\sim \\operatorname{normal}(0,\\sigma).\n\\]\nThis is equivalent to the following sampling involving the residual, \\[\ny_n - (\\alpha + \\beta X_n) \\sim \\operatorname{normal}(0,\\sigma),\n\\] and reducing still further, to \\[\ny_n \\sim \\operatorname{normal}(\\alpha + \\beta X_n, \\, \\sigma).\n\\]\nThis latter form of the model is coded in Stan as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\nThere are N observations and for each observation, \\(n \\in N\\), we have predictor x[n] and outcome y[n]. The intercept and slope parameters are alpha and beta. The model assumes a normally distributed noise term with scale sigma. This model has improper priors for the two regression coefficients.\n\n\nThe distribution statement in the previous model is vectorized, with\ny ~ normal(alpha + beta * x, sigma);\nproviding the same model as the unvectorized version,\nfor (n in 1:N) {\n  y[n] ~ normal(alpha + beta * x[n], sigma);\n}\nIn addition to being more concise, the vectorized form is much faster.1\nIn general, Stan allows the arguments to distributions such as normal to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.\nThe other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because x is of type vector and beta of type real, the expression beta * x is of type vector. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] beta;       // coefficients for predictors\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(x * beta + alpha, sigma);  // data model\n}\nThe constraint lower=0 in the declaration of sigma constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.\nIn the model above, x is an \\(N \\times K\\) matrix of predictors and beta a \\(K\\)-vector of coefficients, so x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. These predictions line up with the outcomes in the \\(N\\)-vector y, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix x to remove the alpha parameter.\nThe distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, sigma);\n  }\n}\nWith Stan’s matrix indexing scheme, x[n] picks out row n of the matrix x; because beta is a column vector, the product x[n] * beta is a scalar of type real.\n\n\nIn the model formulation\ny ~ normal(x * beta, sigma);\nthere is no longer an intercept coefficient alpha. Instead, we have assumed that the first column of the input matrix x is a column of 1 values. This way, beta[1] plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.\n\n\n\n\n\nIn the previous example, the linear predictor can be written as \\(\\eta\n= x \\beta\\), where \\(\\eta\\) is a \\(N\\)-vector of predictions, \\(x\\) is a \\(N\n\\times K\\) matrix, and \\(\\beta\\) is a \\(K\\)-vector of coefficients. Presuming \\(N \\geq K\\), we can exploit the fact that any design matrix \\(x\\) can be decomposed using the thin QR decomposition into an orthogonal matrix \\(Q\\) and an upper-triangular matrix \\(R\\), i.e. \\(x = Q\nR\\).\nThe functions qr_thin_Q and qr_thin_R implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using qr_Q and qr_R, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the qr_thin_Q and qr_thin_R functions). In practice, it is best to write \\(x = Q^\\ast\nR^\\ast\\) where \\(Q^\\ast = Q * \\sqrt{n - 1}\\) and \\(R^\\ast =\n\\frac{1}{\\sqrt{n - 1}} R\\). Thus, we can equivalently write \\(\\eta = x\n\\beta = Q R \\beta = Q^\\ast R^\\ast \\beta\\). If we let \\(\\theta = R^\\ast\n\\beta\\), then we have \\(\\eta = Q^\\ast \\theta\\) and \\(\\beta = R^{\\ast^{-1}}\n\\theta\\). In that case, the previous Stan program becomes\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\ntransformed data {\n  matrix[N, K] Q_ast;\n  matrix[K, K] R_ast;\n  matrix[K, K] R_ast_inverse;\n  // thin and scale the QR decomposition\n  Q_ast = qr_thin_Q(x) * sqrt(N - 1);\n  R_ast = qr_thin_R(x) / sqrt(N - 1);\n  R_ast_inverse = inverse(R_ast);\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] theta;      // coefficients on Q_ast\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(Q_ast * theta + alpha, sigma);  // data model\n}\ngenerated quantities {\n  vector[K] beta;\n  beta = R_ast_inverse * theta; // coefficients on x\n}\nSince this Stan program generates equivalent predictions for \\(y\\) and the same posterior distribution for \\(\\alpha\\), \\(\\beta\\), and \\(\\sigma\\) as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:\n\nThe columns of \\(Q^\\ast\\) are orthogonal whereas the columns of \\(x\\) generally are not. Thus, it is easier for a Markov Chain to move around in \\(\\theta\\)-space than in \\(\\beta\\)-space.\nThe columns of \\(Q^\\ast\\) have the same scale whereas the columns of \\(x\\) generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps\nSince the covariance matrix for the columns of \\(Q^\\ast\\) is an identity matrix, \\(\\theta\\) typically has a reasonable scale if the units of \\(y\\) are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.\n\nConsequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever \\(K &gt; 1\\) and you do not have an informative prior on the location of \\(\\beta\\). It can also be worthwhile to subtract the mean from each column of \\(x\\) before obtaining the QR decomposition, which does not affect the posterior distribution of \\(\\theta\\) or \\(\\beta\\) but does affect \\(\\alpha\\) and allows you to interpret \\(\\alpha\\) as the expectation of \\(y\\) in a linear model.\n\n\n\nSee our general discussion of priors for tips on priors for parameters in regression models.\nLater sections discuss univariate hierarchical priors and multivariate hierarchical priors, as well as priors used to identify models.\nHowever, as described in QR-reparameterization section, if you do not have an informative prior on the location of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.\n\n\n\nThe standard approach to linear regression is to model the noise term \\(\\epsilon\\) as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-\\(t\\) distribution. To code this in Stan, the distribution distribution is changed to the following.\ndata {\n  // ...\n  real&lt;lower=0&gt; nu;\n}\n// ...\nmodel {\n  y ~ student_t(nu, alpha + beta * x, sigma);\n}\nThe degrees of freedom constant nu is specified as data.\n\n\n\nFor binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in \\((-\\infty,\\infty)\\) to probability values in \\((0,1)\\). Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both S-shaped).\nA logistic regression model with one predictor and an intercept is coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta * x);\n}\nThe noise parameter is built into the Bernoulli formulation here rather than specified directly.\nLogistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by \\[\n\\operatorname{logit}(v) = \\log \\left( \\frac{v}{1-v} \\right).\n\\]\nThe inverse of the link function appears in the model: \\[\n\\operatorname{logit}^{-1}(u) = \\texttt{inv}\\mathtt{\\_}\\texttt{logit}(u) = \\frac{1}{1 + \\exp(-u)}.\n\\]\nThe model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha \\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe formulation is also vectorized in the sense that alpha and beta are scalars and x is a vector, so that alpha   + beta * x is a vector. The vectorized formulation is equivalent to the less efficient version\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(alpha + beta * x[n]);\n}\nExpanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable\nfor (n in 1:N) {\n  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));\n}\nOther link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as\n\\[\n\\Phi(x) = \\int_{-\\infty}^x \\textsf{normal}\\left(y \\mid 0,1 \\right) \\,\\textrm{d}y.\n\\]\nThe cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.\ny[n] ~ bernoulli(Phi(alpha + beta * x[n]));\nA fast approximation to the cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi_approx.2 The approximate probit regression model may be coded with the following.\ny[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));\n\n\n\nMultiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are \\(K\\) possible outcomes for each output variable \\(y_n\\). Also suppose that there is a \\(D\\)-dimensional vector \\(x_n\\) of predictors for \\(y_n\\). The multi-logit model with \\(\\textsf{normal}(0,5)\\) priors on the coefficients is coded as follows.\ndata {\n  int K;\n  int N;\n  int D;\n  array[N] int y;\n  matrix[N, D] x;\n}\nparameters {\n  matrix[D, K] beta;\n}\nmodel {\n  matrix[N, K] x_beta = x * beta;\n\n  to_vector(beta) ~ normal(0, 5);\n\n  for (n in 1:N) {\n    y[n] ~ categorical_logit(x_beta[n]');\n\n  }\n}\nwhere x_beta[n]' is the transpose of x_beta[n]. The prior on beta is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, \\[\n\\texttt{categorical}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{categorical}\\left(y \\mid \\texttt{softmax}(\\alpha)\\right),\n\\] where \\[\n\\texttt{softmax}(u) = \\exp(u) / \\operatorname{sum}\\left(\\exp(u)\\right).\n\\]\nThe categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing\ny[n] ~ categorical(softmax(x[n] * beta));\n\n\nThe data block in the above model is defined without constraints on sizes K, N, and D or on the outcome array y. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to\nint&lt;lower=2&gt; K;\nint&lt;lower=0&gt; N;\nint&lt;lower=1&gt; D;\narray[N] int&lt;lower=1, upper=K&gt; y;\nThese constraints arise because the number of categories, K, must be at least two in order for a categorical model to be useful. The number of data items, N, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of 1:N when N is equal to zero ensures the loop’s body will not be executed. The number of predictors, D, must be at least one in order for beta * x[n] to produce an appropriate argument for softmax(). The categorical outcomes y[n] must be between 1 and K in order for the discrete sampling to be well defined.\nConstraints on data declarations are optional. Constraints on parameters declared in the parameters block, on the other hand, are not optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.\n\n\n\nBecause softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.\nAn alternative is to use \\((K-1)\\)-vectors by fixing one of them to be zero. The partially known parameters section discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use \\((K - 1)\\)-vectors\nparameters {\n  matrix[D, K - 1] beta_raw;\n}\nand then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\nwhich can then be appended to beta_raw to produce the coefficient matrix beta,\ntransformed parameters {\n  matrix[D, K] beta = append_col(beta_raw, zeros);\n}\nThe rep_vector(0, D) call creates a column vector of size D with all entries set to zero. The derived matrix beta is then defined to be the result of appending the vector zeros as a new column at the end of beta_raw; the vector zeros is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.\nThis is not the same model as using \\(K\\)-vectors as parameters, because now the prior only applies to \\((K-1)\\)-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.\n\n\n\n\nWhen there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept \\(\\alpha\\) and then a varying effect \\(\\beta_k\\) for age group \\(k\\) to make a linear predictor \\(\\alpha +\n\\beta_k\\). With this predictor, we can add a constant to \\(\\alpha\\) and subtract from each \\(\\beta_k\\) and get exactly the same likelihood.\nThe traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., \\(\\beta_1 = 0\\). With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to \\(\\beta_1\\).\nIn a Bayesian setting, a proper prior on each of the \\(\\beta\\) is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the collinearity section.\nAn alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., \\(\\sum_{k=1}^K \\beta_k = 0.\\)\nA parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the multi-logit section for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the item-response model section for details).\n\n\nAs of Stan 2.36, there is a built in sum_to_zero_vector type, which can be used as follows.\nparameters {\n  sum_to_zero_vector[K] beta;\n  // ...\n}\nThis produces a vector of size K such that sum(beta) = 0. In the unconstrained representation requires only K - 1 values because the last is determined by the first K - 1.\nPlacing a prior on beta in this parameterization, for example,\n  beta ~ normal(0, 1);\nleads to a subtly different posterior than what you would get with the same prior on an unconstrained size-K vector. As explained below, the variance is reduced.\nThe sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., \\(\\beta_K\n= -\\sum_{k=1}^{K-1} \\beta_k.\\) But that leads to high correlation among the \\(\\beta_k\\).\nThe transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; Seyboldt (2024) provides an explanation. The Stan Reference Manual provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.\n\n\nOn the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of beta,\nmodel {\n  beta ~ normal(0, inv(sqrt(1 - inv(K))));\n  // ...\n}\nThe scale component can be multiplied by sigma to produce a normal(0, sigma) prior marginally.\nTo generate distributions with marginals other than standard normal, the resulting beta may be scaled by some factor sigma and translated to some new location mu.\n\n\n\n\nAdding a prior such as \\(\\beta \\sim \\textsf{normal}(0,\\epsilon)\\) for a small \\(\\epsilon\\) will provide a kind of soft centering of a parameter vector \\(\\beta\\) by preferring, all else being equal, that \\(\\sum_{k=1}^K\n\\beta_k = 0\\). This approach is only guaranteed to roughly center if \\(\\beta\\) and the elementwise addition \\(\\beta + c\\) for a scalar constant \\(c\\) produce the same likelihood (perhaps by another vector \\(\\alpha\\) being transformed to \\(\\alpha - c\\), as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an \\(\\epsilon\\). If \\(\\epsilon\\) is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.\n\n\n\n\nOrdered regression for an outcome \\(y_n \\in \\{ 1, \\dotsc, k \\}\\) with predictors \\(x_n \\in \\mathbb{R}^D\\) is determined by a single coefficient vector \\(\\beta \\in \\mathbb{R}^D\\) along with a sequence of cutpoints \\(c \\in\n\\mathbb{R}^{K-1}\\) sorted so that \\(c_d &lt; c_{d+1}\\). The discrete output is \\(k\\) if the linear predictor \\(x_n \\beta\\) falls between \\(c_{k-1}\\) and \\(c_k\\), assuming \\(c_0 = -\\infty\\) and \\(c_K = \\infty\\). The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.\n\n\nThe ordered logistic model can be coded in Stan using the ordered data type for the cutpoints and the built-in ordered_logistic distribution.\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  matrix[N, D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  y ~ ordered_logistic(x * beta, c);\n}\nThe vector of cutpoints c is declared as ordered[K - 1], which guarantees that c[k] is less than c[k + 1].\nIf the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.\nThe equivalent model can be written using ordered_logistic_glm distribution, which can provide more efficient computation in case of higher dimensional beta.\n  y ~ ordered_logistic_glm(x,  beta, c);\n\n\nAn ordered probit model can be coded in exactly the same way by using the built-in ordered_probit distribution.\nmodel {\n  ordered_probit(x * beta, c);\n}\nThere is not yet an ordered_probit_glm distribution in Stan.\n\n\n\n\n\nThe simplest multilevel model is a hierarchical model in which the data are grouped into \\(L\\) distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients \\(\\beta\\). At the other extreme, an approach with no pooling assigns each level \\(l\\) its own coefficient vector \\(\\beta_l\\) that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.\nSuppose each binary outcome \\(y_n \\in \\{ 0, 1 \\}\\) has an associated level, \\(ll_n \\in \\{ 1, \\dotsc, L \\}\\). Each outcome will also have an associated predictor vector \\(x_n \\in \\mathbb{R}^D\\). Each level \\(l\\) gets its own coefficient vector \\(\\beta_l \\in \\mathbb{R}^D\\). The hierarchical structure involves drawing the coefficients \\(\\beta_{l,d}\n\\in \\mathbb{R}\\) from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.\nThe following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.\ndata {\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; L;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  array[N] int&lt;lower=1, upper=L&gt; ll;\n  array[N] row_vector[D] x;\n}\nparameters {\n  array[D] real mu;\n  array[D] real&lt;lower=0&gt; sigma;\n  array[L] vector[D] beta;\n}\nmodel {\n  for (d in 1:D) {\n    mu[d] ~ normal(0, 100);\n    for (l in 1:L) {\n      beta[l, d] ~ normal(mu[d], sigma[d]);\n    }\n  }\n  for (n in 1:N) {\n    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));\n  }\n}\nThe standard deviation parameter sigma gets an implicit uniform prior on \\((0,\\infty)\\) because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the regression priors section for recommendations on priors for regression coefficients and scales.\n\n\nWhere possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.\nThe first optimization vectorizes the for-loop over D as\nmu ~ normal(0, 100);\nfor (l in 1:L) {\n  beta[l] ~ normal(mu, sigma);\n}\nThe declaration of beta as an array of vectors means that the expression beta[l] denotes a vector. Although beta could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the indexing efficiency section for more information on the efficiency tradeoffs among arrays, vectors, and matrices.\nThis model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,3\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);\n}\nUnlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.\n{\n  vector[N] x_beta_ll;\n  for (n in 1:N) {\n    x_beta_ll[n] = x[n] * beta[ll[n]];\n  }\n  y ~ bernoulli_logit(x_beta_ll);\n}\nThe brackets introduce a new scope for the local variable x_beta_ll; alternatively, the variable may be declared at the top of the model block.\nIn some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.\n\n\n\n\nPriors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.\n\n\nThe fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of \\(y_n \\in\n\\mathbb{R}\\) on \\(x_n \\in \\mathbb{R}^K\\), formulated as \\[\\begin{align*}\ny_n     & \\sim \\textsf{normal}(x_n \\beta, \\sigma) \\\\\n\\beta_k & \\sim \\textsf{normal}(0,\\tau) \\\\\n\\tau    & \\sim \\textsf{Cauchy}(0,2.5) \\\\\n\\end{align*}\\]\nIn this case, as \\(\\tau \\rightarrow 0\\) and \\(\\beta_k \\rightarrow 0\\), the posterior density \\[ p(\\beta,\\tau,\\sigma|y,x) \\propto p(y|x,\\beta,\\tau,\\sigma) \\] grows without bound. See the Neal’s funnel density, which has similar behavior.\nThere is obviously no MLE estimate for \\(\\beta,\\tau,\\sigma\\) in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by Chung et al. (2013) is to use a gamma distribution as a prior, such as \\[\n\\sigma \\sim \\textsf{Gamma}(2, 1/A),\n\\] for a reasonably large value of \\(A\\), such as \\(A = 10\\).\n\n\n\n\nItem-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see Gelman and Hill (2007, pps. 314–320) for a textbook introduction to hierarchical IRT models and Curtis (2010) for encodings of a range of IRT models in BUGS.\n\n\nThe data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.\ndata {\n  int&lt;lower=1&gt; J;                     // number of students\n  int&lt;lower=1&gt; K;                     // number of questions\n  int&lt;lower=1&gt; N;                     // number of observations\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // student for observation n\n  array[N] int&lt;lower=1, upper=K&gt; kk;  // question for observation n\n  array[N] int&lt;lower=0, upper=1&gt; y;   // correctness for observation n\n}\nThis declares a total of N student-question pairs in the data set, where each n in 1:N indexes a binary observation y[n] of the correctness of the answer of student jj[n] on question kk[n].\nThe prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.\n\n\n\nThe 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).\nThe model parameters are declared as follows.\nparameters {\n  real delta;            // mean student ability\n  array[J] real alpha;   // ability of student j - mean ability\n  array[K] real beta;    // difficulty of question k\n}\nThe parameter alpha[J] is the ability coefficient for student j and beta[k] is the difficulty coefficient for question k. The non-standard parameterization used here also includes an intercept term delta, which represents the average student’s response to the average question.4\nThe model itself is as follows.\nmodel {\n  alpha ~ std_normal();         // informative true prior\n  beta ~ std_normal();          // informative true prior\n  delta ~ normal(0.75, 1);      // informative true prior\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);\n  }\n}\nThis model uses the logit-parameterized Bernoulli distribution, where \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe key to understanding it is the term inside the bernoulli_logit distribution, from which it follows that \\[\n\\Pr[y_n = 1] = \\operatorname{logit}^{-1}\\left(\\alpha_{jj[n]} - \\beta_{kk[n]}\n+ \\delta\\right).\n\\]\nThe model suffers from additive identifiability issues without the priors. For example, adding a term \\(\\xi\\) to each \\(\\alpha_j\\) and \\(\\beta_k\\) results in the same predictions. The use of priors for \\(\\alpha\\) and \\(\\beta\\) located at 0 identifies the parameters; see Gelman and Hill (2007) for a discussion of identifiability issues and alternative approaches to identification.\nFor testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.\n\n\n\nThe simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.\nparameters {\n  real mu_beta;                // mean question difficulty\n  vector[J] alpha;             // ability for j - mean\n  vector[K] beta;              // difficulty for k\n  vector&lt;lower=0&gt;[K] gamma;    // discrimination of k\n  real&lt;lower=0&gt; sigma_beta;    // scale of difficulties\n  real&lt;lower=0&gt; sigma_gamma;   // scale of log discrimination\n}\nThe parameters should be clearer after the model definition.\nmodel {\n  alpha ~ std_normal();\n  beta ~ normal(0, sigma_beta);\n  gamma ~ lognormal(0, sigma_gamma);\n  mu_beta ~ cauchy(0, 5);\n  sigma_beta ~ cauchy(0, 5);\n  sigma_gamma ~ cauchy(0, 5);\n  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));\n}\nThe std_normal function is used here, defined by \\[\n\\texttt{std}\\mathtt{\\_}\\texttt{normal}(y)\n=\n\\textsf{normal}\\left(y \\mid 0, 1\\right).\n\\]\nThe distribution statement is also vectorized using elementwise multiplication; it is equivalent to\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(gamma[kk[n]]\n                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));\n}\nThe 2PL model is similar to the 1PL model, with the additional parameter gamma[k] modeling how discriminative question k is. If gamma[k] is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter gamma[k] is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.\nThe model is parameterized here with student abilities alpha being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the problematic posteriors chapter for further discussion of identifiability. The difficulty and discrimination parameters beta and gamma then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as\nbeta ~ normal(0, 5);\ngamma ~ lognormal(0, 2);\nThe point is that the alpha determines the scale and location and beta and gamma are allowed to float.\nThe beta parameter is here given a non-centered parameterization, with parameter mu_beta serving as the mean beta location. An alternative would’ve been to take:\nbeta ~ normal(mu_beta, sigma_beta);\nand\ny[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));\nNon-centered parameterizations tend to be more efficient in hierarchical models; see the reparameterization section for more information on non-centered reparameterizations.\nThe intercept term mu_beta can’t itself be modeled hierarchically, so it is given a weakly informative \\(\\textsf{Cauchy}(0,5)\\) prior. Similarly, the scale terms, sigma_beta, and sigma_gamma, are given half-Cauchy priors. As mentioned earlier, the scale and location for alpha are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to \\((0,\\infty)\\) by their declarations.\n\n\n\n\n\n\nOne application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the collinearity section for a simple example of the problems this poses for estimation.\nThe non-identifiability is resolved by providing a standard normal (i.e., \\(\\textsf{normal}(0,1)\\)) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.\n\n\n\nAnother case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.\n\n\n\nIn a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.\nSimilar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.\n\n\n\n\nIn hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.\nAs an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.\n\n\nGelman and Hill (2007, chap. 13, Chapter 17) provide a discussion of a hierarchical model with \\(N\\) individuals organized into \\(J\\) groups. Each individual has a predictor row vector \\(x_n\\) of size \\(K\\); to unify the notation, they assume that \\(x_{n,1} = 1\\) is a fixed “intercept” predictor. To encode group membership, they assume individual \\(n\\) belongs to group \\(jj[n] \\in \\{ 1, \\dotsc, J \\}\\). Each individual \\(n\\) also has an observed outcome \\(y_n\\) taking on real values.\n\n\nThe model is a linear regression with slope and intercept coefficients varying by group, so that \\(\\beta_j\\) is the coefficient \\(K\\)-vector for group \\(j\\). The data model for individual \\(n\\) is then just \\[\ny_n \\sim \\textsf{normal}(x_n \\, \\beta_{jj[n]}, \\, \\sigma)\n\\quad\\text{for}\\quad n \\in \\{ 1, \\dotsc, N \\}.\n\\]\n\n\n\nGelman and Hill model the coefficient vectors \\(\\beta_j\\) as being drawn from a multivariate distribution with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\), \\[\n\\beta_j \\sim \\textsf{multivariate normal}(\\mu_j, \\, \\Sigma)\n\\quad\\text{for}\\quad j \\in \\{ 1, \\dotsc, J \\}.\n\\]\nBelow, we discuss the full model of Gelman and Hill, which uses group-level predictors to model \\(\\mu\\); for now, we assume \\(\\mu\\) is a simple vector parameter.\n\n\n\nFor hierarchical modeling, the group-level mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\) must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as \\[\n\\mu_j \\sim \\textsf{normal}(0,5).\n\\] If more is known about the expected coefficient values \\(\\beta_{j, k}\\), this information can be incorporated into the prior for \\(\\mu_j\\).\nFor the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling\nIn Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define \\[\n\\Sigma = \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau) \\times \\Omega \\times \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau),\n\\] where \\(\\Omega\\) is a correlation matrix and \\(\\tau\\) is the vector of coefficient scales. This mapping from scale vector \\(\\tau\\) and correlation matrix \\(\\Omega\\) can be inverted, using \\[\n\\tau_k = \\sqrt{\\Sigma_{k,k}}\n\\quad\\textsf{and}\\quad\n\\Omega_{i, j} = \\frac{\\Sigma_{i, j}}{\\tau_i \\, \\tau_j}.\n\\]\nThe components of the scale vector \\(\\tau\\) can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as \\[\n\\tau_k \\sim \\textsf{Cauchy}(0, 2.5)\n\\quad\\text{for}\\quad k \\in \\{ 1, \\dotsc, K \\}\n\\quad\\text{constrained\\ by}\\quad \\tau_k &gt; 0.\n\\] As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for \\(\\tau\\). For large numbers of exchangeable coefficients, the components of \\(\\tau\\) itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.\nOur final recommendation is to give the correlation matrix \\(\\Omega\\) an LKJ prior with shape \\(\\eta \\geq 1\\),5\n\\[\n\\Omega \\sim \\textsf{LKJCorr}(\\eta).\n\\]\nThe LKJ correlation distribution is defined by \\[\n\\textsf{LKJCorr}\\left(\\Sigma \\mid \\eta\\right)\n\\propto\n\\operatorname{det}\\left(\\Sigma\\right)^{\\eta - 1}.\n\\]\nThe basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For \\(\\eta = 1\\), the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.\nFor \\(\\eta &gt; 1\\), the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For \\(\\eta &lt; 1\\), it increasingly concentrates mass in the other direction, i.e., favoring more correlation.\nThe LKJ prior may thus be used to control the expected amount of correlation among the parameters \\(\\beta_j\\). For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see Barnard, McCulloch, and Meng (2000).\n\n\n\nTo complete Gelman and Hill’s model, suppose each group \\(j \\in \\{ 1, \\dotsc, J \\}\\) is supplied with an \\(L\\)-dimensional row-vector of group-level predictors \\(u_j\\). The prior mean for the \\(\\beta_j\\) can then itself be modeled as a regression, using an \\(L\\)-dimensional coefficient vector \\(\\gamma\\). The prior for the group-level coefficients then becomes \\[\n\\beta_j \\sim \\textsf{multivariate normal}(u_j \\, \\gamma, \\Sigma)\n\\]\nThe group-level coefficients \\(\\gamma\\) may themselves be given independent weakly informative priors, such as \\[\n\\gamma_l \\sim \\textsf{normal}(0,5).\n\\] As usual, information about the group-level means should be incorporated into this prior.\n\n\n\nThe Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.\ndata {\n  int&lt;lower=0&gt; N;              // num individuals\n  int&lt;lower=1&gt; K;              // num ind predictors\n  int&lt;lower=1&gt; J;              // num groups\n  int&lt;lower=1&gt; L;              // num group predictors\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // group for individual\n  matrix[N, K] x;              // individual predictors\n  array[J] row_vector[L] u;    // group predictors\n  vector[N] y;                 // outcomes\n}\nparameters {\n  corr_matrix[K] Omega;        // prior correlation\n  vector&lt;lower=0&gt;[K] tau;      // prior scale\n  matrix[L, K] gamma;          // group coeffs\n  array[J] vector[K] beta;     // indiv coeffs by group\n  real&lt;lower=0&gt; sigma;         // prediction error scale\n}\nmodel {\n  tau ~ cauchy(0, 2.5);\n  Omega ~ lkj_corr(2);\n  to_vector(gamma) ~ normal(0, 5);\n  {\n    array[J] row_vector[K] u_gamma;\n    for (j in 1:J) {\n      u_gamma[j] = u[j] * gamma;\n    }\n    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));\n  }\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n  }\n}\nThe hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix Omega and scale vector tau are more natural to inspect in the output; to output Sigma, define it as a transformed parameter. The function quad_form_diag is defined so that quad_form_diag(Sigma, tau) is equivalent to diag_matrix(tau) * Sigma * diag_matrix(tau), where diag_matrix(tau) returns the matrix with tau on the diagonal and zeroes off diagonal; the version using quad_form_diag should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.\n\n\n\nThe code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:\nfor (n in 1:N) {\n  y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n}\nwith the vectorized distribution statement:\n{\n  vector[N] x_beta_jj;\n  for (n in 1:N) {\n    x_beta_jj[n] = x[n] * beta[jj[n]];\n  }\n  y ~ normal(x_beta_jj, sigma);\n}\nThe outer brackets create a local scope in which to define the variable x_beta_jj, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.\nAlthough it is tempting to redeclare beta and include a revised model block distribution statement,\nparameters {\n  matrix[J, K] beta;\n// ...\n}\nmodel {\n  y ~ normal(rows_dot_product(x, beta[jj]), sigma);\n  // ...\n}\nthis fails because it breaks the vectorization for beta,6\nbeta ~ multi_normal(...);\nwhich requires beta to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because rows_dot_product cannot do much optimization in and of itself because there are no shared computations.\nThe code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.\n{\n  matrix[K, K] Sigma_beta;\n  Sigma_beta = quad_form_diag(Omega, tau);\n  for (j in 1:J) {\n    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);\n  }\n}\nIn this example, the covariance matrix Sigma_beta is defined as a local variable so as not to have to repeat the quadratic form computation \\(J\\) times. This vectorization can be combined with the Cholesky-factor optimization in the next section.\n\n\n\nThe multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.\ndata {\n  matrix[L, J] u;              // group predictors transposed\n  // ...\n}\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  matrix[K, L] gamma;\n  // ...\n}\ntransformed parameters {\n  matrix[K, J] beta;\n  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  // ...\n}\nThe data variable u was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with gamma and beta, to minimize the number of transposition operations. The new parameter L_Omega is the Cholesky factor of the original correlation matrix Omega, so that\nOmega = L_Omega * L_Omega'\nThe prior scale vector tau is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,\nSigma_beta\n  = quad_form_diag(Omega, tau)\n  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'\nwhere the diagonal pre-multiply compound operation is defined by\ndiag_pre_multiply(a, b) = diag_matrix(a) * b\nThe new variable z is declared as a matrix, the entries of which are given independent standard normal priors; the to_vector operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of z being a \\(K\\)-variate normal random vector with the identity as covariance matrix. Therefore, multiplying z by the Cholesky factor of the covariance matrix and adding the mean (u * gamma)' produces a beta distributed as in the original model, where the variance is, letting \\(L = \\mathrm{diag}(\\tau)\\,\\Omega_L\\),\n\\[\n\\begin{aligned}\n\\mathbb{V}[\\beta] &= \\mathbb{E}\\big((L \\, z)(L \\, z)^\\top) \\\\\n&= \\mathbb{E}\\big((L \\, z \\, z^\\top \\, L^\\top) \\\\\n&= L \\, \\mathbb{E}(z \\, z^\\top) \\, L^\\top \\\\\n&= L \\, L^\\top =(\\mathrm{diag}(\\tau)\\,\\Omega_L)\\,(\\mathrm{diag}(\\tau)\\,\\Omega_L)^\\top \\\\\n&= \\mathrm{diag}(\\tau)\\,\\Omega\\,\\mathrm{diag}(\\tau) \\\\\n&= \\Sigma.\n\\end{aligned}\n\\] Where we have used the linearity of expectations (line 2 to 3), the definition of \\(\\Omega = \\Omega_L \\, \\Omega_L^\\top\\), and the fact that \\(\\mathbb{E}(z \\, z^\\top) = I\\) since \\(z \\sim \\mathcal{N}(0, I)\\).\nOmitting the remaining data declarations, which are the same as before with the exception of u, the optimized model is as follows.\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0, upper=pi() / 2&gt;[K] tau_unif;  // prior scale\n  matrix[K, L] gamma;                        // group coeffs\n  real&lt;lower=0&gt; sigma;                       // prediction error scale\n}\ntransformed parameters {\n  vector&lt;lower=0&gt;[K] tau = 2.5 * tan(tau_unif);\n  matrix[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  vector[N] mu;\n  for(n in 1:N) {\n    mu[n] = x[n, ] * beta[, jj[n]];\n  }\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  to_vector(gamma) ~ normal(0, 5);\n  y ~ normal(mu, sigma);\n}\nThis model also reparameterizes the prior scale tau to avoid potential problems with the heavy tails of the Cauchy distribution. The statement tau_unif ~ uniform(0, pi() / 2) can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.\n\n\n\n\n\nStan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).\n\n\nAs a simple example, the following linear regression provides the same setup for estimating the coefficients beta as in our very first example, using y for the N observations and x for the N predictor vectors. The model parameters and model for observations are exactly the same as before.\nTo make predictions, we need to be given the number of predictions, N_new, and their predictor matrix, x_new. The predictions themselves are modeled as a parameter y_new. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector y_new and prediction matrix x_new.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=0&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n\n  int&lt;lower=0&gt; N_new;\n  matrix[N_new, K] x_new;\n}\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n\n  vector[N_new] y_new;                  // predictions\n}\nmodel {\n  y ~ normal(x * beta, sigma);          // observed model\n\n  y_new ~ normal(x_new * beta, sigma);  // prediction model\n}\n\n\n\nWhere possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.\n// ...data as above...\n\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(x * beta, sigma);\n}\ngenerated quantities {\n  vector[N_new] y_new;\n  for (n in 1:N_new) {\n    y_new[n] = normal_rng(x_new[n] * beta, sigma);\n  }\n}\nNow the data are just as before, but the parameter y_new is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.\n\n\nIt is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of \\(2^{31}\\), must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by \\(2^{31} - 1\\) in Stan.\n\n\n\n\n\nMost regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.\nThis section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.\n\n\nThe first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors (Zellner 1962; Greene 2011).\nThe model is easy to write down as a regression, \\[\\begin{align*}\ny_n        &=  x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nwhere \\(x_n\\) is a \\(J\\)-row-vector of predictors (\\(x\\) is an \\((N \\times\nJ)\\) matrix), \\(y_n\\) is a \\(K\\)-vector of observations, \\(\\beta\\) is a \\((K\n\\times J)\\) matrix of regression coefficients (vector \\(\\beta_k\\) holds coefficients for outcome \\(k\\)), and \\(\\Sigma\\) is covariance matrix governing the error. As usual, the intercept can be rolled into \\(x\\) as a column of ones.\nThe basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; J;\n  int&lt;lower=0&gt; N;\n  array[N] vector[J] x;\n  array[N] vector[K] y;\n}\nparameters {\n  matrix[K, J] beta;\n  cov_matrix[K] Sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n  }\n  y ~ multi_normal(mu, Sigma);\n}\nFor efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.\nFollowing the advice in the multivariate hierarchical priors section, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.\n// ...\nparameters {\n  matrix[K, J] beta;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0&gt;[K] L_sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  matrix[K, K] L_Sigma;\n\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n\n  }\n\n  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);\n\n  to_vector(beta) ~ normal(0, 5);\n  L_Omega ~ lkj_corr_cholesky(4);\n  L_sigma ~ cauchy(0, 2.5);\n\n  y ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix beta to a vector.\nIf required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.\n\n\n\nThe multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.\nThe observations \\(y_n\\) are \\(D\\)-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations \\(y_n\\) are based on latent values \\(z_n\\) drawn from a seemingly unrelated regression model (see the previous section), \\[\\begin{align*}\nz_n        &=     x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim  \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nThese are then put through the step function to produce a \\(K\\)-vector \\(z_n\\) of boolean values with elements defined by \\[\ny_{n, k} = \\operatorname{I}\\left(z_{n, k} &gt; 0\\right),\n\\] where \\(\\operatorname{I}()\\) is the indicator function taking the value 1 if its argument is true and 0 otherwise.\nUnlike in the seemingly unrelated regressions case, here the covariance matrix \\(\\Sigma\\) has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see Greene (2011)).\nMultivariate probit regression can be coded in Stan using the trick introduced by Albert and Chib (1993), where the underlying continuous value vectors \\(y_n\\) are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector \\(z\\) in two parts, based on whether the corresponding value of \\(y\\) is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.\nFirst, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in \\(y\\).\nfunctions {\n  int sum2d(array[,] int a) {\n    int s = 0;\n    for (i in 1:size(a)) {\n      s += sum(a[i]);\n    }\n    return s;\n  }\n}\nThe function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.\nThe data declaration block is much like for the seemingly unrelated regressions, but the observations y are now integers constrained to be 0 or 1.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  array[N, D] int&lt;lower=0, upper=1&gt; y;\n  array[N] vector[K] x;\n}\nAfter declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array y into positive and negative components, keeping track of indexes so that z can be easily reassembled in the transformed parameters block.\ntransformed data {\n  int&lt;lower=0&gt; N_pos;\n  array[sum2d(y)] int&lt;lower=1, upper=N&gt; n_pos;\n  array[size(n_pos)] int&lt;lower=1, upper=D&gt; d_pos;\n  int&lt;lower=0&gt; N_neg;\n  array[(N * D) - size(n_pos)] int&lt;lower=1, upper=N&gt; n_neg;\n  array[size(n_neg)] int&lt;lower=1, upper=D&gt; d_neg;\n\n  N_pos = size(n_pos);\n  N_neg = size(n_neg);\n  {\n    int i;\n    int j;\n    i = 1;\n    j = 1;\n    for (n in 1:N) {\n      for (d in 1:D) {\n        if (y[n, d] == 1) {\n          n_pos[i] = n;\n          d_pos[i] = d;\n          i += 1;\n        } else {\n          n_neg[j] = n;\n          d_neg[j] = d;\n          j += 1;\n        }\n      }\n    }\n  }\n}\nThe variables N_pos and N_neg are set to the number of true (1) and number of false (0) observations in y. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.\nThe parameters are declared as follows.\nparameters {\n  matrix[D, K] beta;\n  cholesky_factor_corr[D] L_Omega;\n  vector&lt;lower=0&gt;[N_pos] z_pos;\n  vector&lt;upper=0&gt;[N_neg] z_neg;\n}\nThese include the regression coefficients beta and the Cholesky factor of the correlation matrix, L_Omega. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).\nThe critical part of the parameter declaration is that the latent real value \\(z\\) is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct \\(z\\).\ntransformed parameters {\n  array[N] vector[D] z;\n  for (n in 1:N_pos) {\n    z[n_pos[n], d_pos[n]] = z_pos[n];\n  }\n  for (n in 1:N_neg) {\n    z[n_neg[n], d_neg[n]] = z_neg[n];\n  }\n}\nAt this point, the model is simple, pretty much recreating the seemingly unrelated regression.\nmodel {\n  L_Omega ~ lkj_corr_cholesky(4);\n  to_vector(beta) ~ normal(0, 5);\n  {\n    array[N] vector[D] beta_x;\n    for (n in 1:N) {\n      beta_x[n] = beta * x[n];\n    }\n    z ~ multi_normal_cholesky(beta_x, L_Omega);\n  }\n}\nThis simple form of model is made possible by the Albert and Chib-style constraints on z.\nFinally, the correlation matrix itself can be put back together in the generated quantities block if desired.\ngenerated quantities {\n  corr_matrix[D] Omega;\n  Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\nThe same could be done for the seemingly unrelated regressions in the previous section.\n\n\n\n\nThe main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.\n\n\nConsider predicting unobserved outcomes using linear regression. Given predictors \\(x_1, \\dotsc, x_N\\) and observed outcomes \\(y_1, \\dotsc, y_N\\), and assuming a standard linear regression with intercept \\(\\alpha\\), slope \\(\\beta\\), and error scale \\(\\sigma\\), along with improper uniform priors, the posterior over the parameters given \\(x\\) and \\(y\\) is \\[\np\\left(\\alpha, \\beta, \\sigma \\mid x, y \\right)\n\\propto\n\\prod_{n=1}^N\n  \\textsf{normal}\\left(y_n \\mid \\alpha + \\beta x_n, \\sigma\\right).\n\\]\nFor this model, the posterior predictive inference for a new outcome \\(\\tilde{y}_m\\) given a predictor \\(\\tilde{x}_m\\), conditioned on the observed data \\(x\\) and \\(y\\), is \\[\np\\left(\\tilde{y}_n \\mid \\tilde{x}_n, x, y\\right)\n= \\int_{(\\alpha,\\beta,\\sigma)}\n  \\textsf{normal}\\left(\\tilde{y}_n \\mid \\alpha + \\beta \\tilde{x}_n, \\sigma\\right)\n  \\times\n  p\\left(\\alpha, \\beta, \\sigma \\mid x, y\\right)\n  \\,\\textrm{d}(\\alpha,\\beta,\\sigma).\n\\]\nTo code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde;\n  for (n in 1:N_tilde) {\n    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);\n  }\n}\nGiven observed predictors \\(x\\) and outcomes \\(y\\), y_tilde will be drawn according to \\(p\\left(\\tilde{y} \\mid \\tilde{x}, y, x\\right)\\). This means that, for example, the posterior mean for y_tilde is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).\n\n\n\nA good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors \\(x\\) are used in place of new predictors \\(\\tilde{x}\\) for unobserved outcomes. If the model fits the data well, the predictions for \\(\\tilde{y}\\) based on \\(x\\) should match the observed data \\(y\\).\nTo code posterior predictive checks in Stan requires only a slight modification of the prediction code to use \\(x\\) and \\(N\\) in place of \\(\\tilde{x}\\) and \\(\\tilde{N}\\),\ngenerated quantities {\n  vector[N] y_tilde;\n  for (n in 1:N) {\n    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);\n  }\n}\nGelman et al. (2013) recommend choosing several posterior draws \\(\\tilde{y}^{(1)}, \\dotsc, \\tilde{y}^{(M)}\\) and plotting each of them alongside the data \\(y\\) that was actually observed. If the model fits well, the simulated \\(\\tilde{y}\\) will look like the actual data \\(y\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-predictive-checks.html#joint-model-representation",
-    "href": "stan-users-guide/posterior-predictive-checks.html#joint-model-representation",
-    "title": "Posterior and Prior Predictive Checks",
+    "objectID": "stan-users-guide/regression.html#linear-regression",
+    "href": "stan-users-guide/regression.html#linear-regression",
+    "title": "Regression Models",
     "section": "",
-    "text": "Following Andrew Gelman, Meng, and Stern (1996), prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.\n\n\nFor example, posterior predictive replication may be formulated using distribution notation as follows. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] The heavily overloaded distribution notation is meant to indicate that both \\(y\\) and \\(y^{\\textrm{rep}}\\) are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities \\(p_{Y^{\\textrm{rep}} \\mid\n\\Theta}\\) and \\(p_{Y \\mid \\Theta}\\) are the same.\nThe joint density is \\[\np(\\theta, y, y^{\\textrm{rep}})\n= p(\\theta) \\cdot p(y \\mid \\theta) \\cdot p(y^{\\textrm{rep}} \\mid \\theta).\n\\] This again is assuming that the two distributions for \\(y\\) and \\(y^{\\textrm{rep}}\\) are identical.\nThe variable \\(y\\) is observed, with the predictive simulation \\(y^{\\textrm{rep}}\\) and parameter vector \\(\\theta\\) not observed. The posterior is \\(p(y^{\\textrm{rep}}, \\theta \\mid y)\\). Given draws from the posterior, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are retained.\n\n\n\nThe prior predictive model simply drops the data component of the posterior predictive model. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] This corresponds to the joint density \\[\np(\\theta, y^{\\textrm{rep}}) = p(\\theta) \\cdot p(y^{\\textrm{rep}} \\mid\n\\theta).\n\\]\nIt is typically straightforward to draw \\(\\theta\\) from the prior and \\(y^{\\textrm{rep}}\\) from the data model given \\(\\theta\\) efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.\n\n\n\nThe mixed replication corresponds to the model \\[\\begin{eqnarray*}\n\\phi & \\sim & p(\\phi)\n\\\\[2pt]\n\\alpha & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\alpha)\n\\\\[2pt]\n\\alpha^{\\textrm{rep}} & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\phi)\n\\end{eqnarray*}\\] The notation here is meant to indicate that \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\) have identical distributions, as do \\(y\\) and \\(y^{\\textrm{rep}}\\).\nThis corresponds to a joint model \\[\np(\\phi, \\alpha, \\alpha^{\\textrm{rep}}, y, y^{\\textrm{rep}})\n=\np(\\phi)\n\\cdot p(\\alpha \\mid \\phi)\n\\cdot p(y \\mid \\alpha)\n\\cdot p(\\alpha^{\\textrm{rep}} \\mid \\phi)\n\\cdot p(y^{\\textrm{rep}} \\mid \\alpha^{\\textrm{rep}}),\n\\] where \\(y\\) is the only observed variable, \\(\\alpha\\) contains the lower-level parameters and \\(\\phi\\) the hyperparameters. Note that \\(\\phi\\) is not replicated and instead appears in the distribution for both \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\).\nThe posterior is \\(p(\\phi, \\alpha, \\alpha^{\\textrm{rep}},\ny^{\\textrm{rep}} \\mid y)\\). From posterior draws, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are kept.",
+    "text": "The simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\quad\\text{where}\\quad\n\\epsilon_n \\sim \\operatorname{normal}(0,\\sigma).\n\\]\nThis is equivalent to the following sampling involving the residual, \\[\ny_n - (\\alpha + \\beta X_n) \\sim \\operatorname{normal}(0,\\sigma),\n\\] and reducing still further, to \\[\ny_n \\sim \\operatorname{normal}(\\alpha + \\beta X_n, \\, \\sigma).\n\\]\nThis latter form of the model is coded in Stan as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\nThere are N observations and for each observation, \\(n \\in N\\), we have predictor x[n] and outcome y[n]. The intercept and slope parameters are alpha and beta. The model assumes a normally distributed noise term with scale sigma. This model has improper priors for the two regression coefficients.\n\n\nThe distribution statement in the previous model is vectorized, with\ny ~ normal(alpha + beta * x, sigma);\nproviding the same model as the unvectorized version,\nfor (n in 1:N) {\n  y[n] ~ normal(alpha + beta * x[n], sigma);\n}\nIn addition to being more concise, the vectorized form is much faster.1\nIn general, Stan allows the arguments to distributions such as normal to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.\nThe other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because x is of type vector and beta of type real, the expression beta * x is of type vector. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] beta;       // coefficients for predictors\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(x * beta + alpha, sigma);  // data model\n}\nThe constraint lower=0 in the declaration of sigma constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.\nIn the model above, x is an \\(N \\times K\\) matrix of predictors and beta a \\(K\\)-vector of coefficients, so x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. These predictions line up with the outcomes in the \\(N\\)-vector y, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix x to remove the alpha parameter.\nThe distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, sigma);\n  }\n}\nWith Stan’s matrix indexing scheme, x[n] picks out row n of the matrix x; because beta is a column vector, the product x[n] * beta is a scalar of type real.\n\n\nIn the model formulation\ny ~ normal(x * beta, sigma);\nthere is no longer an intercept coefficient alpha. Instead, we have assumed that the first column of the input matrix x is a column of 1 values. This way, beta[1] plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior and Prior Predictive Checks"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/parallelization.html",
-    "href": "stan-users-guide/parallelization.html",
-    "title": "Parallelization",
+    "objectID": "stan-users-guide/regression.html#QR-reparameterization.section",
+    "href": "stan-users-guide/regression.html#QR-reparameterization.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "Stan has support for different types of parallelization: multi-threading with Intel Threading Building Blocks (TBB), multi-processing with Message Passing Interface (MPI) and manycore processing with OpenCL.\nMulti-threading in Stan can be used with two mechanisms: reduce with summation and rectangular map. The latter can also be used with multi-processing.\nThe advantages of reduce with summation are:\n\nMore flexible argument interface, avoiding the packing and unpacking that is necessary with rectangular map.\nPartitions data for parallelization automatically (this is done manually in rectangular map).\nIs easier to use.\n\nThe advantages of rectangular map are:\n\nReturns a list of vectors, while the reduce summation returns only a scalar.\nCan be parallelized across multiple cores and multiple computers, while reduce summation can only parallelized across multiple cores on a single machine.\n\nThe actual speedup gained from using these functions will depend on many details. It is strongly recommended to only parallelize the computationally most expensive operations in a Stan program. Oftentimes this is the evaluation of the log likelihood for the observed data. When it is not clear which parts of the model is the most computationally expensive, we recommend using profiling, which is available in Stan 2.26 and newer.\nSince only portions of a Stan program will run in parallel, the maximal speedup one can achieve is capped, a phenomen described by Amdahl’s law.\n\n\nIt is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If g: U -&gt; real is the function and { x1, x2, ... } is an array of inputs, then that sum looks like:\ng(x1) + g(x2) + ...\nreduce_sum and reduce_sum_static are tools for parallelizing these calculations.\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g, but instead the partial sum function f: U[] -&gt; real, where f computes the partial sum corresponding to a slice of the sequence x passed in. Due to the associativity of the sum reduction it holds that:\ng(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })\n                      = f({ x1, x2 }) + f({ x3 })\n                      = f({ x1 }) + f({ x2, x3 })\n                      = f({ x1 }) + f({ x2 }) + f({ x3 })\nWith the partial sum function f: U[] -&gt; real reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load).\n\ngrainsize is the one tuning parameter. For reduce_sum, grainsize is a suggested partial sum size. A grainsize of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using reduce_sum unless time is spent carefully picking grainsize. For picking a grainsize, see details below.\nFor reduce_sum_static, grainsize specifies the maximal partial sum size. With reduce_sum_static it is more important to choose grainsize carefully since it entirely determines the partitioning of work. See details below.\nFor efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array { x1, x2, ... } and the shared arguments s1, s2, ... the effective sum (with individual terms) looks like:\ng(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...\nwhich can be written equivalently with partial sums to look like:\nf({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)\nwhere the particular slicing of the x array can change.\nGiven this, the signatures are:\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nreal reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n\nf - User defined function that computes partial sums\nx - Array to slice, each element corresponds to a term in the summation\ngrainsize - Target for size of slices\ns1, s2, ... - Arguments shared in every term\n\nThe user-defined partial sum functions have the signature:\nreal f(array[] T x_slice, int start, int end, T1 s1, T2 s2, ...)\nand take the arguments:\n\nx_slice - The subset of x (from reduce_sum / reduce_sum_static) for which this partial sum is responsible (x_slice = x[start:end])\nstart - An integer specifying the first term in the partial sum\nend - An integer specifying the last term in the partial sum (inclusive)\ns1, s2, ... - Arguments shared in every term (passed on without modification from the reduce_sum / reduce_sum_static call)\n\nThe user-provided function f is expected to compute the partial sum with the terms start through end of the overall sum. The user function is passed the subset x[start:end] as x_slice. start and end are passed so that f can index any of the tailing sM arguments as necessary. The trailing sM arguments are passed without modification to every call of f.\nA reduce_sum (or reduce_sum_static) call:\nreal sum = reduce_sum(f, x, grainsize, s1, s2, ...);\ncan be replaced by either:\nreal sum = f(x, 1, size(x), s1, s2, ...);\nor the code:\nreal sum = 0.0;\nfor(i in 1:size(x)) {\n  sum += f({ x[i] }, i, i, s1, s2, ...);\n}\n\n\nLogistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * x);\n}\nIn this model predictions are made about the N outputs y using the covariate x. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:\ny ~ bernoulli_logit(beta[1] + beta[2] * x);\ncan be rewritten (up to a proportionality constant) as:\nfor(n in 1:N) {\n  target += bernoulli_logit_lpmf(y[n] | beta[1] + beta[2] * x[n])\n}\nNow it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in Reduce-Sum, such a function can be written like:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\nThe likelihood statement in the model can now be written:\ntarget += partial_sum(y, 1, N, x, beta); // Sum terms 1 to N of the likelihood\nIn this example, y was chosen to be sliced over because there is one term in the summation per value of y. Technically x would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g. slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because x is a shared argument, it is subset accordingly with start:end. With this function, reduce summation can be used to automatically parallelize the likelihood:\nint grainsize = 1;\ntarget += reduce_sum(partial_sum, y,\n                     grainsize,\n                     x, beta);\nThe reduce summation facility automatically breaks the sum into pieces and computes them in parallel. grainsize = 1 specifies that the grainsize should be estimated automatically. The final model is:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  int grainsize = 1;\n  beta ~ std_normal();\n  target += reduce_sum(partial_sum, y,\n                       grainsize,\n                       x, beta);\n}\n\n\n\nThe rational for choosing a sensible grainsize is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.\nIn reduce_sum, grainsize is a recommendation on how to partition the work in the partial sum into smaller pieces. A grainsize of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.\nIn reduce_sum_static, grainsize is an upper limit on the worksize. Work will be split until all partial sums are just smaller than grainsize (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible grainsize.\nIn order to figure out an optimal grainsize, if there are N terms and M cores, run a quick test model with grainsize set roughly to N / M. Record the time, cut the grainsize in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable grainsize for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.\nFor instance, in a model with N=10000 and M = 4, start with grainsize = 2500, and sequentially try grainsize = 1250, grainsize = 625, etc.\nIt is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller grainsize that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a grainsize of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.\n\n\n\n\nMap-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).\nA map function is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, \\(f(x) = x^2\\), over the vector \\([3, 5, 10]\\) produces the vector \\([9, 25, 100]\\). In other words, map applies the square function elementwise.\nThe output of mapping a sequence is often fed into a reduction. A reduction function takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, map-reduce.\n\n\nIn order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.\n\n\n\nStan’s map function has the following signature\nvector map_rect((vector, vector, array[] real, array[] int):vector f,\n                vector phi, array[] vector thetas,\n                data array[,] real x_rs, data array[,] int x_is);\nThe arrays thetas of parameters, x_rs of real data, and x_is of integer data have the suffix “s” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function f. The value of phi is reused in each mapped operation.\nThe _rect suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.\nThe last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the data qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.\nThe notation (vector, vector, array[] real, array[] int):vector indicates that the function argument f must have the following signature.\nvector f(vector phi, vector theta,\n         data array[] real x_r, data array[] int x_i);\nAlthough f will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.\n\n\nStan’s map function applies the function f to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments theta, x_r, and x_i must be arrays of the same size. If the arrays are all size N, the result is defined as follows.\nmap_rect(f, phi, thetas, xs, ns)\n= f(phi, thetas[1], xs[1], ns[1]) . f(phi, thetas[2], xs[2], ns[2])\n  . ... . f(phi, thetas[N], xs[N], ns[N])\nThe dot operators in the notation above are meant to indicate concatenation (implemented as append_row in Stan). The output of each application of f is a vector, and the sequence of N vectors is concatenated together to return a single vector.\n\n\n\n\nAn example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.\n\n\nConsider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.\ndata {\n  array[12] int y;\n  array[12] real x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * to_vector(x));\n}\nThe program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The bernoulli_logit distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.\n\n\n\nThe unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.\nfunctions {\n  vector lr(vector beta, vector theta, array[] real x, array[] int y) {\n    real lp = bernoulli_logit_lpmf(y | beta[1]\n                                       + to_vector(x) * beta[2]);\n    return [lp]';\n  }\n}\ndata {\n  array[12] int y;\n  array[12] real x;\n}\ntransformed data {\n  // K = 3 shards\n  array[3, 4] int ys = { y[1:4], y[5:8], y[9:12] };\n  array[3, 4] real xs = { x[1:4], x[5:8], x[9:12] };\n  array[3] vector[0] theta;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  target += sum(map_rect(lr, beta, theta, xs, ys));\n}\nThe first piece of the code is the actual function to compute the logistic regression. The argument beta will contain the regression coefficients (intercept and slope), as before. The second argument theta of job-specific parameters is not used, but nevertheless must be present. The modeled data y is passed as an array of integers and the predictors x as an array of real values. The function body then computes the log probability mass of y and assigns it to the local variable lp. This variable is then used in [lp]' to construct a row vector and then transpose it to a vector to return.\nThe data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.1\nThe value 3 is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array ys contains the modeled data variables; each element of the array ys is an array of size four. The second array xs is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors x stand in a one-to-one relationship with the modeled data y. The final array theta is also of size three; its elements are empty vectors, because there are no shard-specific parameters.\nThe parameters and the prior are as before. The likelihood is now coded using map-reduce. The function lr to compute the log probability mass is mapped over the data xs and ys, which contain the original predictors and outcomes broken into shards. The parameters beta are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters theta contains only empty vectors.\n\n\n\n\nConsider a hierarchical model of American presidential voting behavior based on state of residence.2\nEach of the fifty states \\(k \\in \\{1,\\dotsc,50\\}\\) will have its own slope \\(\\beta_k\\) and intercept \\(\\alpha_k\\) to model the log odds of voting for the Republican candidate as a function of income. Suppose there are \\(N\\) voters and with voter \\(n \\in 1{:}N\\) being in state \\(s[n]\\) with income \\(x_n\\). The data model for the vote \\(y_n \\in \\{ 0, 1 \\}\\) is \\[\ny_n \\sim \\textsf{Bernoulli}\n\\Big(\n  \\operatorname{logit}^{-1}\\left( \\alpha_{s[n]} + \\beta_{s[n]} \\, x_n \\right)\n\\Big).\n\\]\nThe slopes and intercepts get hierarchical priors, \\[\\begin{align*}\n\\alpha_k &\\sim \\textsf{normal}(\\mu_{\\alpha}, \\sigma_{\\alpha}) \\\\\n\\beta_k  &\\sim \\textsf{normal}(\\mu_{\\beta}, \\sigma_{\\beta})\n\\end{align*}\\]\n\n\nThis model can be coded up in Stan directly as follows.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  matrix[K, 2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  for (i in 1:2) {\n    beta[ , i] ~ normal(mu[i], sigma[i]);\n  }\n  y ~ bernoulli_logit(beta[kk, 1] + beta[kk, 2] .* x);\n}\nFor this model the vector of predictors x is coded as a vector, corresponding to how it is used in the model. The priors for mu and sigma are vectorized. The priors on the two components of beta (intercept and slope, respectively) are stored in a \\(K \\times 2\\) matrix.\nThe distribution statement is also vectorized using multi-indexing with index kk for the states and elementwise multiplication (.*) for the income x. The vectorized distribution statement works out to the same thing as the following less efficient looped form.\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(beta[kk[n], 1] + beta[kk[n], 2] * x[n]);\n}\n\n\n\nThe mapped version of the model will map over the states K. This means the group-level parameters, real data, and integer-data must be arrays of the same size.\nThe mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name bernoulli_logit_glm derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).\nfunctions {\n vector bl_glm(vector mu_sigma, vector beta,\n               array[] real x, array[] int y) {\n   vector[2] mu = mu_sigma[1:2];\n   vector[2] sigma = mu_sigma[3:4];\n   real lp = normal_lpdf(beta | mu, sigma);\n   real ll = bernoulli_logit_lpmf(y | beta[1] + beta[2] * to_vector(x));\n   return [lp + ll]';\n }\n}\nThe shared parameter mu_sigma contains the locations (mu_sigma[1:2]) and scales (mu_sigma[3:4]) of the priors, which are extracted in the first two lines of the program. The variable lp is assigned the log density of the prior on beta. The vector beta is of size two, as are the vectors mu and sigma, so everything lines up for the vectorization. Next, the variable ll is assigned to the log likelihood contribution for the group. Here beta[1] is the intercept of the regression and beta[2] the slope. The predictor array x needs to be converted to a vector allow the multiplication.\nThe data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\ntransformed data {\n  int&lt;lower=0&gt; J = N / K;\n  array[K, J] real x_r;\n  array[K, J] int&lt;lower=0, upper=1&gt; x_i;\n  {\n    int pos = 1;\n    for (k in 1:K) {\n      int end = pos + J - 1;\n      x_r[k] = to_array_1d(x[pos:end]);\n      x_i[k] = to_array_1d(y[pos:end]);\n      pos += J;\n    }\n  }\n}\nThe integer J is set to the number of observations per group.3\nThe real data array x_r holds the predictors and the integer data array x_i holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector x (and converting it to an array) and slicing the outcome array y.\nGiven the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters mu and sigma, but moves the prior for beta and the likelihood to the mapped function.\nparameters {\n  array[K] vector[2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  target += sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));\n                         \n}\nThe model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.\n\n\n\n\nThe previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by map_rect.\n\n\nIf each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.\n\n\n\nThe output of each mapped function is concatenated in order of inputs to produce the output of map_rect. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the to_matrix function if they are rectangular.\nIf the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.\n\n\n\n\n\nOpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.\nStan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the STAN_OPENCL makefile flag. Setting this flag means that the Stan-to-C++ translator (stanc3) will be supplied the --use-opencl flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.\nIn Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:\n\nbernoulli_lpmf\nbernoulli_logit_lpmf\nbernoulli_logit_glm_lpmf*\nbeta_lpdf\nbeta_proportion_lpdf\nbinomial_lpmf\ncategorical_logit_glm_lpmf*\ncauchy_lpdf\nchi_square_lpdf\ndouble_exponential_lpdf\nexp_mod_normal_lpdf\nexponential_lpdf\nfrechet_lpdf\ngamma_lpdf\ngumbel_lpdf\ninv_chi_square_lpdf\ninv_gamma_lpdf\nlogistic_lpdf\nlognormal_lpdf\nneg_binomial_lpmf\nneg_binomial_2_lpmf\nneg_binomial_2_log_lpmf\nneg_binomial_2_log_glm_lpmf*\nnormal_lpdf\nnormal_id_glm_lpdf*\nordered_logistic_glm_lpmf*\npareto_lpdf\npareto_type_2_lpdf\npoisson_lpmf\npoisson_log_lpmf\npoisson_log_glm_lpmf*\nrayleigh_lpdf\nscaled_inv_chi_square_lpdf\nskew_normal_lpdf\nstd_normal_lpdf\nstudent_t_lpdf\nuniform_lpdf\nweibull_lpdf\n\n* OpenCL is not used when the covariate argument to the GLM functions is a row_vector.",
+    "text": "In the previous example, the linear predictor can be written as \\(\\eta\n= x \\beta\\), where \\(\\eta\\) is a \\(N\\)-vector of predictions, \\(x\\) is a \\(N\n\\times K\\) matrix, and \\(\\beta\\) is a \\(K\\)-vector of coefficients. Presuming \\(N \\geq K\\), we can exploit the fact that any design matrix \\(x\\) can be decomposed using the thin QR decomposition into an orthogonal matrix \\(Q\\) and an upper-triangular matrix \\(R\\), i.e. \\(x = Q\nR\\).\nThe functions qr_thin_Q and qr_thin_R implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using qr_Q and qr_R, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the qr_thin_Q and qr_thin_R functions). In practice, it is best to write \\(x = Q^\\ast\nR^\\ast\\) where \\(Q^\\ast = Q * \\sqrt{n - 1}\\) and \\(R^\\ast =\n\\frac{1}{\\sqrt{n - 1}} R\\). Thus, we can equivalently write \\(\\eta = x\n\\beta = Q R \\beta = Q^\\ast R^\\ast \\beta\\). If we let \\(\\theta = R^\\ast\n\\beta\\), then we have \\(\\eta = Q^\\ast \\theta\\) and \\(\\beta = R^{\\ast^{-1}}\n\\theta\\). In that case, the previous Stan program becomes\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\ntransformed data {\n  matrix[N, K] Q_ast;\n  matrix[K, K] R_ast;\n  matrix[K, K] R_ast_inverse;\n  // thin and scale the QR decomposition\n  Q_ast = qr_thin_Q(x) * sqrt(N - 1);\n  R_ast = qr_thin_R(x) / sqrt(N - 1);\n  R_ast_inverse = inverse(R_ast);\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] theta;      // coefficients on Q_ast\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(Q_ast * theta + alpha, sigma);  // data model\n}\ngenerated quantities {\n  vector[K] beta;\n  beta = R_ast_inverse * theta; // coefficients on x\n}\nSince this Stan program generates equivalent predictions for \\(y\\) and the same posterior distribution for \\(\\alpha\\), \\(\\beta\\), and \\(\\sigma\\) as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:\n\nThe columns of \\(Q^\\ast\\) are orthogonal whereas the columns of \\(x\\) generally are not. Thus, it is easier for a Markov Chain to move around in \\(\\theta\\)-space than in \\(\\beta\\)-space.\nThe columns of \\(Q^\\ast\\) have the same scale whereas the columns of \\(x\\) generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps\nSince the covariance matrix for the columns of \\(Q^\\ast\\) is an identity matrix, \\(\\theta\\) typically has a reasonable scale if the units of \\(y\\) are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.\n\nConsequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever \\(K &gt; 1\\) and you do not have an informative prior on the location of \\(\\beta\\). It can also be worthwhile to subtract the mean from each column of \\(x\\) before obtaining the QR decomposition, which does not affect the posterior distribution of \\(\\theta\\) or \\(\\beta\\) but does affect \\(\\alpha\\) and allows you to interpret \\(\\alpha\\) as the expectation of \\(y\\) in a linear model.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Parallelization"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/parallelization.html#reduce-sum",
-    "href": "stan-users-guide/parallelization.html#reduce-sum",
-    "title": "Parallelization",
+    "objectID": "stan-users-guide/regression.html#regression-priors.section",
+    "href": "stan-users-guide/regression.html#regression-priors.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "It is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If g: U -&gt; real is the function and { x1, x2, ... } is an array of inputs, then that sum looks like:\ng(x1) + g(x2) + ...\nreduce_sum and reduce_sum_static are tools for parallelizing these calculations.\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g, but instead the partial sum function f: U[] -&gt; real, where f computes the partial sum corresponding to a slice of the sequence x passed in. Due to the associativity of the sum reduction it holds that:\ng(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })\n                      = f({ x1, x2 }) + f({ x3 })\n                      = f({ x1 }) + f({ x2, x3 })\n                      = f({ x1 }) + f({ x2 }) + f({ x3 })\nWith the partial sum function f: U[] -&gt; real reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load).\n\ngrainsize is the one tuning parameter. For reduce_sum, grainsize is a suggested partial sum size. A grainsize of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using reduce_sum unless time is spent carefully picking grainsize. For picking a grainsize, see details below.\nFor reduce_sum_static, grainsize specifies the maximal partial sum size. With reduce_sum_static it is more important to choose grainsize carefully since it entirely determines the partitioning of work. See details below.\nFor efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array { x1, x2, ... } and the shared arguments s1, s2, ... the effective sum (with individual terms) looks like:\ng(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...\nwhich can be written equivalently with partial sums to look like:\nf({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)\nwhere the particular slicing of the x array can change.\nGiven this, the signatures are:\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nreal reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n\nf - User defined function that computes partial sums\nx - Array to slice, each element corresponds to a term in the summation\ngrainsize - Target for size of slices\ns1, s2, ... - Arguments shared in every term\n\nThe user-defined partial sum functions have the signature:\nreal f(array[] T x_slice, int start, int end, T1 s1, T2 s2, ...)\nand take the arguments:\n\nx_slice - The subset of x (from reduce_sum / reduce_sum_static) for which this partial sum is responsible (x_slice = x[start:end])\nstart - An integer specifying the first term in the partial sum\nend - An integer specifying the last term in the partial sum (inclusive)\ns1, s2, ... - Arguments shared in every term (passed on without modification from the reduce_sum / reduce_sum_static call)\n\nThe user-provided function f is expected to compute the partial sum with the terms start through end of the overall sum. The user function is passed the subset x[start:end] as x_slice. start and end are passed so that f can index any of the tailing sM arguments as necessary. The trailing sM arguments are passed without modification to every call of f.\nA reduce_sum (or reduce_sum_static) call:\nreal sum = reduce_sum(f, x, grainsize, s1, s2, ...);\ncan be replaced by either:\nreal sum = f(x, 1, size(x), s1, s2, ...);\nor the code:\nreal sum = 0.0;\nfor(i in 1:size(x)) {\n  sum += f({ x[i] }, i, i, s1, s2, ...);\n}\n\n\nLogistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * x);\n}\nIn this model predictions are made about the N outputs y using the covariate x. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:\ny ~ bernoulli_logit(beta[1] + beta[2] * x);\ncan be rewritten (up to a proportionality constant) as:\nfor(n in 1:N) {\n  target += bernoulli_logit_lpmf(y[n] | beta[1] + beta[2] * x[n])\n}\nNow it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in Reduce-Sum, such a function can be written like:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\nThe likelihood statement in the model can now be written:\ntarget += partial_sum(y, 1, N, x, beta); // Sum terms 1 to N of the likelihood\nIn this example, y was chosen to be sliced over because there is one term in the summation per value of y. Technically x would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g. slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because x is a shared argument, it is subset accordingly with start:end. With this function, reduce summation can be used to automatically parallelize the likelihood:\nint grainsize = 1;\ntarget += reduce_sum(partial_sum, y,\n                     grainsize,\n                     x, beta);\nThe reduce summation facility automatically breaks the sum into pieces and computes them in parallel. grainsize = 1 specifies that the grainsize should be estimated automatically. The final model is:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  int grainsize = 1;\n  beta ~ std_normal();\n  target += reduce_sum(partial_sum, y,\n                       grainsize,\n                       x, beta);\n}\n\n\n\nThe rational for choosing a sensible grainsize is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.\nIn reduce_sum, grainsize is a recommendation on how to partition the work in the partial sum into smaller pieces. A grainsize of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.\nIn reduce_sum_static, grainsize is an upper limit on the worksize. Work will be split until all partial sums are just smaller than grainsize (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible grainsize.\nIn order to figure out an optimal grainsize, if there are N terms and M cores, run a quick test model with grainsize set roughly to N / M. Record the time, cut the grainsize in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable grainsize for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.\nFor instance, in a model with N=10000 and M = 4, start with grainsize = 2500, and sequentially try grainsize = 1250, grainsize = 625, etc.\nIt is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller grainsize that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a grainsize of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.",
+    "text": "See our general discussion of priors for tips on priors for parameters in regression models.\nLater sections discuss univariate hierarchical priors and multivariate hierarchical priors, as well as priors used to identify models.\nHowever, as described in QR-reparameterization section, if you do not have an informative prior on the location of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Parallelization"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/parallelization.html#map-rect",
-    "href": "stan-users-guide/parallelization.html#map-rect",
-    "title": "Parallelization",
+    "objectID": "stan-users-guide/regression.html#robust-noise-models",
+    "href": "stan-users-guide/regression.html#robust-noise-models",
+    "title": "Regression Models",
     "section": "",
-    "text": "Map-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).\nA map function is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, \\(f(x) = x^2\\), over the vector \\([3, 5, 10]\\) produces the vector \\([9, 25, 100]\\). In other words, map applies the square function elementwise.\nThe output of mapping a sequence is often fed into a reduction. A reduction function takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, map-reduce.\n\n\nIn order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.\n\n\n\nStan’s map function has the following signature\nvector map_rect((vector, vector, array[] real, array[] int):vector f,\n                vector phi, array[] vector thetas,\n                data array[,] real x_rs, data array[,] int x_is);\nThe arrays thetas of parameters, x_rs of real data, and x_is of integer data have the suffix “s” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function f. The value of phi is reused in each mapped operation.\nThe _rect suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.\nThe last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the data qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.\nThe notation (vector, vector, array[] real, array[] int):vector indicates that the function argument f must have the following signature.\nvector f(vector phi, vector theta,\n         data array[] real x_r, data array[] int x_i);\nAlthough f will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.\n\n\nStan’s map function applies the function f to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments theta, x_r, and x_i must be arrays of the same size. If the arrays are all size N, the result is defined as follows.\nmap_rect(f, phi, thetas, xs, ns)\n= f(phi, thetas[1], xs[1], ns[1]) . f(phi, thetas[2], xs[2], ns[2])\n  . ... . f(phi, thetas[N], xs[N], ns[N])\nThe dot operators in the notation above are meant to indicate concatenation (implemented as append_row in Stan). The output of each application of f is a vector, and the sequence of N vectors is concatenated together to return a single vector.\n\n\n\n\nAn example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.\n\n\nConsider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.\ndata {\n  array[12] int y;\n  array[12] real x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * to_vector(x));\n}\nThe program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The bernoulli_logit distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.\n\n\n\nThe unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.\nfunctions {\n  vector lr(vector beta, vector theta, array[] real x, array[] int y) {\n    real lp = bernoulli_logit_lpmf(y | beta[1]\n                                       + to_vector(x) * beta[2]);\n    return [lp]';\n  }\n}\ndata {\n  array[12] int y;\n  array[12] real x;\n}\ntransformed data {\n  // K = 3 shards\n  array[3, 4] int ys = { y[1:4], y[5:8], y[9:12] };\n  array[3, 4] real xs = { x[1:4], x[5:8], x[9:12] };\n  array[3] vector[0] theta;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  target += sum(map_rect(lr, beta, theta, xs, ys));\n}\nThe first piece of the code is the actual function to compute the logistic regression. The argument beta will contain the regression coefficients (intercept and slope), as before. The second argument theta of job-specific parameters is not used, but nevertheless must be present. The modeled data y is passed as an array of integers and the predictors x as an array of real values. The function body then computes the log probability mass of y and assigns it to the local variable lp. This variable is then used in [lp]' to construct a row vector and then transpose it to a vector to return.\nThe data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.1\nThe value 3 is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array ys contains the modeled data variables; each element of the array ys is an array of size four. The second array xs is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors x stand in a one-to-one relationship with the modeled data y. The final array theta is also of size three; its elements are empty vectors, because there are no shard-specific parameters.\nThe parameters and the prior are as before. The likelihood is now coded using map-reduce. The function lr to compute the log probability mass is mapped over the data xs and ys, which contain the original predictors and outcomes broken into shards. The parameters beta are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters theta contains only empty vectors.\n\n\n\n\nConsider a hierarchical model of American presidential voting behavior based on state of residence.2\nEach of the fifty states \\(k \\in \\{1,\\dotsc,50\\}\\) will have its own slope \\(\\beta_k\\) and intercept \\(\\alpha_k\\) to model the log odds of voting for the Republican candidate as a function of income. Suppose there are \\(N\\) voters and with voter \\(n \\in 1{:}N\\) being in state \\(s[n]\\) with income \\(x_n\\). The data model for the vote \\(y_n \\in \\{ 0, 1 \\}\\) is \\[\ny_n \\sim \\textsf{Bernoulli}\n\\Big(\n  \\operatorname{logit}^{-1}\\left( \\alpha_{s[n]} + \\beta_{s[n]} \\, x_n \\right)\n\\Big).\n\\]\nThe slopes and intercepts get hierarchical priors, \\[\\begin{align*}\n\\alpha_k &\\sim \\textsf{normal}(\\mu_{\\alpha}, \\sigma_{\\alpha}) \\\\\n\\beta_k  &\\sim \\textsf{normal}(\\mu_{\\beta}, \\sigma_{\\beta})\n\\end{align*}\\]\n\n\nThis model can be coded up in Stan directly as follows.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  matrix[K, 2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  for (i in 1:2) {\n    beta[ , i] ~ normal(mu[i], sigma[i]);\n  }\n  y ~ bernoulli_logit(beta[kk, 1] + beta[kk, 2] .* x);\n}\nFor this model the vector of predictors x is coded as a vector, corresponding to how it is used in the model. The priors for mu and sigma are vectorized. The priors on the two components of beta (intercept and slope, respectively) are stored in a \\(K \\times 2\\) matrix.\nThe distribution statement is also vectorized using multi-indexing with index kk for the states and elementwise multiplication (.*) for the income x. The vectorized distribution statement works out to the same thing as the following less efficient looped form.\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(beta[kk[n], 1] + beta[kk[n], 2] * x[n]);\n}\n\n\n\nThe mapped version of the model will map over the states K. This means the group-level parameters, real data, and integer-data must be arrays of the same size.\nThe mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name bernoulli_logit_glm derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).\nfunctions {\n vector bl_glm(vector mu_sigma, vector beta,\n               array[] real x, array[] int y) {\n   vector[2] mu = mu_sigma[1:2];\n   vector[2] sigma = mu_sigma[3:4];\n   real lp = normal_lpdf(beta | mu, sigma);\n   real ll = bernoulli_logit_lpmf(y | beta[1] + beta[2] * to_vector(x));\n   return [lp + ll]';\n }\n}\nThe shared parameter mu_sigma contains the locations (mu_sigma[1:2]) and scales (mu_sigma[3:4]) of the priors, which are extracted in the first two lines of the program. The variable lp is assigned the log density of the prior on beta. The vector beta is of size two, as are the vectors mu and sigma, so everything lines up for the vectorization. Next, the variable ll is assigned to the log likelihood contribution for the group. Here beta[1] is the intercept of the regression and beta[2] the slope. The predictor array x needs to be converted to a vector allow the multiplication.\nThe data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\ntransformed data {\n  int&lt;lower=0&gt; J = N / K;\n  array[K, J] real x_r;\n  array[K, J] int&lt;lower=0, upper=1&gt; x_i;\n  {\n    int pos = 1;\n    for (k in 1:K) {\n      int end = pos + J - 1;\n      x_r[k] = to_array_1d(x[pos:end]);\n      x_i[k] = to_array_1d(y[pos:end]);\n      pos += J;\n    }\n  }\n}\nThe integer J is set to the number of observations per group.3\nThe real data array x_r holds the predictors and the integer data array x_i holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector x (and converting it to an array) and slicing the outcome array y.\nGiven the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters mu and sigma, but moves the prior for beta and the likelihood to the mapped function.\nparameters {\n  array[K] vector[2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  target += sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));\n                         \n}\nThe model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.\n\n\n\n\nThe previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by map_rect.\n\n\nIf each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.\n\n\n\nThe output of each mapped function is concatenated in order of inputs to produce the output of map_rect. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the to_matrix function if they are rectangular.\nIf the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.",
+    "text": "The standard approach to linear regression is to model the noise term \\(\\epsilon\\) as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-\\(t\\) distribution. To code this in Stan, the distribution distribution is changed to the following.\ndata {\n  // ...\n  real&lt;lower=0&gt; nu;\n}\n// ...\nmodel {\n  y ~ student_t(nu, alpha + beta * x, sigma);\n}\nThe degrees of freedom constant nu is specified as data.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Parallelization"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/parallelization.html#opencl",
-    "href": "stan-users-guide/parallelization.html#opencl",
-    "title": "Parallelization",
+    "objectID": "stan-users-guide/regression.html#logistic-probit-regression.section",
+    "href": "stan-users-guide/regression.html#logistic-probit-regression.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "OpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.\nStan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the STAN_OPENCL makefile flag. Setting this flag means that the Stan-to-C++ translator (stanc3) will be supplied the --use-opencl flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.\nIn Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:\n\nbernoulli_lpmf\nbernoulli_logit_lpmf\nbernoulli_logit_glm_lpmf*\nbeta_lpdf\nbeta_proportion_lpdf\nbinomial_lpmf\ncategorical_logit_glm_lpmf*\ncauchy_lpdf\nchi_square_lpdf\ndouble_exponential_lpdf\nexp_mod_normal_lpdf\nexponential_lpdf\nfrechet_lpdf\ngamma_lpdf\ngumbel_lpdf\ninv_chi_square_lpdf\ninv_gamma_lpdf\nlogistic_lpdf\nlognormal_lpdf\nneg_binomial_lpmf\nneg_binomial_2_lpmf\nneg_binomial_2_log_lpmf\nneg_binomial_2_log_glm_lpmf*\nnormal_lpdf\nnormal_id_glm_lpdf*\nordered_logistic_glm_lpmf*\npareto_lpdf\npareto_type_2_lpdf\npoisson_lpmf\npoisson_log_lpmf\npoisson_log_glm_lpmf*\nrayleigh_lpdf\nscaled_inv_chi_square_lpdf\nskew_normal_lpdf\nstd_normal_lpdf\nstudent_t_lpdf\nuniform_lpdf\nweibull_lpdf\n\n* OpenCL is not used when the covariate argument to the GLM functions is a row_vector.",
+    "text": "For binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in \\((-\\infty,\\infty)\\) to probability values in \\((0,1)\\). Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both S-shaped).\nA logistic regression model with one predictor and an intercept is coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta * x);\n}\nThe noise parameter is built into the Bernoulli formulation here rather than specified directly.\nLogistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by \\[\n\\operatorname{logit}(v) = \\log \\left( \\frac{v}{1-v} \\right).\n\\]\nThe inverse of the link function appears in the model: \\[\n\\operatorname{logit}^{-1}(u) = \\texttt{inv}\\mathtt{\\_}\\texttt{logit}(u) = \\frac{1}{1 + \\exp(-u)}.\n\\]\nThe model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha \\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe formulation is also vectorized in the sense that alpha and beta are scalars and x is a vector, so that alpha   + beta * x is a vector. The vectorized formulation is equivalent to the less efficient version\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(alpha + beta * x[n]);\n}\nExpanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable\nfor (n in 1:N) {\n  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));\n}\nOther link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as\n\\[\n\\Phi(x) = \\int_{-\\infty}^x \\textsf{normal}\\left(y \\mid 0,1 \\right) \\,\\textrm{d}y.\n\\]\nThe cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.\ny[n] ~ bernoulli(Phi(alpha + beta * x[n]));\nA fast approximation to the cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi_approx.2 The approximate probit regression model may be coded with the following.\ny[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Parallelization"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/parallelization.html#footnotes",
-    "href": "stan-users-guide/parallelization.html#footnotes",
-    "title": "Parallelization",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe term “shard” is borrowed from databases, where it refers to a slice of the rows of a database. That is exactly what it is here if we think of rows of a dataframe. Stan’s shards are more general in that they need not correspond to rows of a dataframe.↩︎\nThis example is a simplified form of the model described in (Gelman and Hill 2007, sec. 14.2)↩︎\nThis makes the strong assumption that each group has the same number of observations!↩︎",
+    "objectID": "stan-users-guide/regression.html#multi-logit.section",
+    "href": "stan-users-guide/regression.html#multi-logit.section",
+    "title": "Regression Models",
+    "section": "",
+    "text": "Multiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are \\(K\\) possible outcomes for each output variable \\(y_n\\). Also suppose that there is a \\(D\\)-dimensional vector \\(x_n\\) of predictors for \\(y_n\\). The multi-logit model with \\(\\textsf{normal}(0,5)\\) priors on the coefficients is coded as follows.\ndata {\n  int K;\n  int N;\n  int D;\n  array[N] int y;\n  matrix[N, D] x;\n}\nparameters {\n  matrix[D, K] beta;\n}\nmodel {\n  matrix[N, K] x_beta = x * beta;\n\n  to_vector(beta) ~ normal(0, 5);\n\n  for (n in 1:N) {\n    y[n] ~ categorical_logit(x_beta[n]');\n\n  }\n}\nwhere x_beta[n]' is the transpose of x_beta[n]. The prior on beta is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, \\[\n\\texttt{categorical}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{categorical}\\left(y \\mid \\texttt{softmax}(\\alpha)\\right),\n\\] where \\[\n\\texttt{softmax}(u) = \\exp(u) / \\operatorname{sum}\\left(\\exp(u)\\right).\n\\]\nThe categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing\ny[n] ~ categorical(softmax(x[n] * beta));\n\n\nThe data block in the above model is defined without constraints on sizes K, N, and D or on the outcome array y. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to\nint&lt;lower=2&gt; K;\nint&lt;lower=0&gt; N;\nint&lt;lower=1&gt; D;\narray[N] int&lt;lower=1, upper=K&gt; y;\nThese constraints arise because the number of categories, K, must be at least two in order for a categorical model to be useful. The number of data items, N, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of 1:N when N is equal to zero ensures the loop’s body will not be executed. The number of predictors, D, must be at least one in order for beta * x[n] to produce an appropriate argument for softmax(). The categorical outcomes y[n] must be between 1 and K in order for the discrete sampling to be well defined.\nConstraints on data declarations are optional. Constraints on parameters declared in the parameters block, on the other hand, are not optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.\n\n\n\nBecause softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.\nAn alternative is to use \\((K-1)\\)-vectors by fixing one of them to be zero. The partially known parameters section discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use \\((K - 1)\\)-vectors\nparameters {\n  matrix[D, K - 1] beta_raw;\n}\nand then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\nwhich can then be appended to beta_raw to produce the coefficient matrix beta,\ntransformed parameters {\n  matrix[D, K] beta = append_col(beta_raw, zeros);\n}\nThe rep_vector(0, D) call creates a column vector of size D with all entries set to zero. The derived matrix beta is then defined to be the result of appending the vector zeros as a new column at the end of beta_raw; the vector zeros is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.\nThis is not the same model as using \\(K\\)-vectors as parameters, because now the prior only applies to \\((K-1)\\)-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Parallelization"
+      "Example Models",
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html",
-    "href": "stan-users-guide/odes.html",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#parameterizing-centered-vectors",
+    "href": "stan-users-guide/regression.html#parameterizing-centered-vectors",
+    "title": "Regression Models",
     "section": "",
-    "text": "Stan provides a number of different methods for solving systems of ordinary differential equations (ODEs). All of these methods adaptively refine their solutions in order to satisfy given tolerances, but internally they handle calculations quite a bit differently.\nBecause Stan’s algorithms requires gradients of the log density, the ODE solvers must not only provide the solution to the ODE itself, but also the gradient of the ODE solution with respect to parameters (the sensitivities). Two fundamentally different approaches are available in Stan to solve this problem, each having very different computational cost depending on the number of ODE states \\(N\\) and the number of parameters \\(M\\) being used:\n\nA forward sensitivity solver expands the base ODE system with additional ODE equations for the gradients of the solution. For each parameter, an additional full set of \\(N\\) sensitivity states are added meaning that the full ODE solved has \\(N \\, + N \\cdot M\\) states.\nAn adjoint sensitivity solver starts by solving the base ODE system forward in time to get the ODE solution and then solves another ODE system (the adjoint) backward in time to get the gradients. The forward and reverse solves both have \\(N\\) states each. There is additionally one quadrature problem solved for every parameter.\n\nThe adjoint sensitivity approach scales much better than the forward sensitivity approach. Whereas the computational cost of the forward approach scales multiplicatively in the number of ODE states \\(N\\) and parameters \\(M\\), the adjoint sensitivity approach scales linear in states \\(N\\) and parameters \\(M\\). However, the adjoint problem is harder to configure and the overhead for small problems actually makes it slower than solving the full forward sensitivity system. With that in mind, the rest of this introduction focuses on the forward sensitivity interfaces. For information on the adjoint sensitivity interface see the Adjoint ODE solver\nTwo interfaces are provided for each forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. For instance, if a solution component slowly varies between 3.0 and 5.0 and measurements of the ODE state are noisy, then perhaps the tolerances do not need to be as tight as for a situation where the solutions vary between 3.0 and 3.1 and very high precision measurements of the ODE state are available. It is also often useful to reduce the absolute tolerance when a component of the solution is expected to approach zero. For information on choosing tolerances, see the control parameters section.\nThe advantage of adaptive solvers is that as long as reasonable tolerances are provided and an ODE solver well-suited to the problem is chosen the technical details of solving the ODE can be abstracted away. The catch is that it is not always clear from the outset what reasonable tolerances are or which ODE solver is best suited to a problem. In addition, as changes are made to an ODE model, the optimal solver and tolerances may change.\nWith this in mind, the four forward solvers are rk45, bdf, adams, and ckrk. If no other information about the ODE is available, start with the rk45 solver. The list below has information on when each solver is useful.\nIf there is any uncertainty about which solver is the best, it can be useful to measure the performance of all the interesting solvers using profile statements. It is difficult to always know exactly what solver is the best in all situations, but a profile can provide a quick check.\n\nrk45: a fourth and fifth order Runge-Kutta method for non-stiff systems (Dormand and Prince 1980; Ahnert and Mulansky 2011). rk45 is the most generic solver and should be tried first.\nbdf: a variable-step, variable-order, backward-differentiation formula implementation for stiff systems (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). bdf is often useful for ODEs modeling chemical reactions.\nadams: a variable-step, variable-order, Adams-Moulton formula implementation for non-stiff systems (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). The method has order up to 12, hence is commonly used when high-accuracy is desired for a very smooth solution, such as in modeling celestial mechanics and orbital dynamics (Montenbruck and Gill 2000).\nckrk: a fourth and fifth order explicit Runge-Kutta method for non-stiff and semi-stiff systems (Cash and Karp 1990; Mazzia, Cash, and Soetaert 2012). The difference between ckrk and rk45 is that ckrk should perform better for systems that exhibit rapidly varying solutions. Often in those situations the derivatives become large or even nearly discontinuous, and ckrk is designed to address such problems.\n\nFor a discussion of stiff ODE systems, see the stiff ODE section. For information on the adjoint sensitivity interface see the Adjoint ODE solver section. The function signatures for Stan’s ODE solvers can be found in the function reference manual section on ODE solvers.\n\n\nAn ODE is defined by a set of differential equations, \\(y(t, \\theta)' = f(t, y, \\theta)\\), and initial conditions, \\(y(t_0, \\theta) = y_0\\). The function \\(f(t, y, \\theta)\\) is called the system function. The \\(\\theta\\) dependence is included in the notation for \\(y(t, \\theta)\\) and \\(f(t, y, \\theta)\\) as a reminder that the solution is a function of any parameters used in the computation.\n\n\n\nAs an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair \\(y = (y_1, y_2)\\) representing position and speed. The change in the system with respect to time is given by the following differential equations.1\n\\[\\begin{align*}\n&\\frac{d}{dt} y_1 = y_2  \\\\\n&\\frac{d}{dt} y_2 = -y_1 - \\theta y_2\n\\end{align*}\\]\nThe state equations implicitly defines the state at future times as a function of an initial state and the system parameters.\n\n\n\nThe first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the system function is time, passed as a real; the second argument to the system function is the system state, passed as a vector, and the return value from the system function are the current time derivatives of the state defined as a vector. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.\nThe simple harmonic oscillator can be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector sho(real t,        // time\n           vector y,      // state\n           real theta) {  // friction parameter\n  vector[2] dydt;\n  dydt[1] = y[2];\n  dydt[2] = -y[1] - theta * y[2];\n  return dydt;\n}\nThe function takes in a time t (a real), the system state y (a vector), and the parameter theta (a real). The function returns a vector of time derivatives of the system state at time t, state y, and parameter theta. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, t does not show up in the definition of dydt, however it is still required.\n\n\nThe types in the ODE system function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible ODE signatures:\nvector myode1(real t, vector y, real a0);\nvector myode2(real t, vector y, array[] int a0, vector a1);\nvector myode3(real t, vector y, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector myode1(real t, array[] real y, real a0);\n// Second argument is not a vector\narray[] real myode2(real t, vector y, real a0);\n// Return type is not a vector\nvector myode3(vector y, real a0);\n// First argument is not a real and second is not a vector\n\n\n\n\nNoisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.\n\n\nAs an example, suppose the simple harmonic oscillator has a parameter value of \\(\\theta = 0.15\\) and an initial state \\(y(t = 0, \\theta = 0.15) = (1, 0)\\). Assume the system is measured at 10 time points, \\(t = 1, 2, \\cdots, 10\\), where each measurement of \\(y(t, \\theta)\\) has independent \\(\\textsf{normal}(0, 0.1)\\) error in both dimensions (\\(y_1(t, \\theta)\\) and \\(y_2(t, \\theta)\\)).\nThe following model can be used to generate data like this:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  real t0;\n  array[T] real ts;\n  real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThe system parameters theta and initial state y0 are read in as data along with the initial time t0 and observation times ts. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations y_sim. Because the system is not stiff, the ode_rk45 solver is used.\nThis program illustrates the way in which the ODE solver is called in a Stan program,\narray[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\nthis returns the solution of the ODE initial value problem defined by system function sho, initial state y0, initial time t0, and parameter theta at the times ts. The call explicitly specifies the non-stiff RK45 solver.\nThe parameter theta is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, \\(\\theta\\) and \\(\\beta\\), the system function definition would look like:\nvector sho(real t, vector y, real theta, real beta) { ... }\nand the appropriate ODE solver call would be:\node_rk45(sho, y0, t0, ts, theta, beta);\nAny number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).\nBecause all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the T outputs with normal_rng.\n\n\n\nTypical realization of harmonic oscillator trajectory.\n\n\n\n\n\nThese ten noisy observations of the state can be used to estimate the friction parameter, \\(\\theta\\), the initial conditions, \\(y(t_0, \\theta)\\), and the scale of the noise in the problem. The full Stan model is:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  array[T] vector[2] y;\n  real t0;\n  array[T] real ts;\n}\nparameters {\n  vector[2] y0;\n  vector&lt;lower=0&gt;[2] sigma;\n  real theta;\n}\nmodel {\n  array[T] vector[2] mu = ode_rk45(sho, y0, t0, ts, theta);\n  sigma ~ normal(0, 2.5);\n  theta ~ std_normal();\n  y0 ~ std_normal();\n  for (t in 1:T) {\n    y[t] ~ normal(mu[t], sigma);\n  }\n}\nBecause the solves are now a function of model parameters, the ode_rk45 call is now made in the model block. There are half-normal priors on the measurement error scales sigma, and standard normal priors on theta and the initial state vector y0. The solutions to the ODE are assigned to mu, which is used as the location for the normal observation model.\nAs with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).\n\n\n\n\nStiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.\nStan provides a specialized solver for stiff ODEs (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the rk45 suffix is replaced with bdf, as in\node_bdf(sho, y0, t0, ts, theta);\nUsing the stiff (bdf) solver on a system that is not stiff may be much slower than using the non-stiff (rk45) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.\nIf it is not known for sure that an ODE system is stiff, run the model with both the rk45 and bdf solvers and see which is faster. If the rk45 solver is faster, then the problem is probably non-stiff, and then it makes sense to try the adams solver as well. The adams solver uses higher order methods which can take larger timesteps than the rk45 solver, though similar to the bdf solver each of these steps is more expensive to compute.\n\n\n\nFor additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters. These are the same as the regular function names but with _tol appended to the end. All three control arguments must be supplied with this signature (there are no defaults).\narray[T] vector[2] y_sim = ode_bdf_tol(sho, y0, t0, ts,\n                                 relative_tolerance,\n                                 absolute_tolerance,\n                                 max_num_steps,\n                                 theta);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nFor the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both \\(10^{-6}\\) and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million.\n\n\nIf there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.\nNonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times \\(0 &lt; t &lt; t'\\) and then positive for \\(t \\geq t'\\). In this example example, we would use code in the system such as\nif (t &lt; t_lag) {\n  return [0, 0]';\n} else {\n  // ... return non-zero vector...\n}\nIn general it is better to integrate up to t_lag in one solve and then integrate from t_lag onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.\nIf the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.\n\n\n\nThe relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns \\(y=(y_1,\\dots,y_n)^T\\), at every step the solver controls estimated local error \\(e=(e_1,\\dots,e_n)^T\\) through its weighted root-mean-square norm (Serban and Hindmarsh (2005), Hairer, Nørsett, and Wanner (1993))\n\\[\\begin{equation*}\n\\sqrt{\\sum_{i=1}^n{\\frac{1}{n}\\frac{e_i^2}{(\\text{RTOL}\\times y_i + \\text{ATOL})^2}}} &lt; 1\n\\end{equation*}\\] by reducing the stepsize when the inequality is not satisfied.\nTo understand the roles of the two tolerances it helps to assume \\(y\\) at opposite scales in the above expression: on one hand the absolute tolerance has little effect when \\(y_i \\gg 1\\), on the other the relative tolerance can not affect the norm when \\(y_i = 0\\). Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow Brenan, Campbell, and Petzold (1995) for a rule of thumb: let \\(m\\) be the number of significant digits required for \\(y\\), set \\(\\text{RTOL}=10^{-(m+1)}\\), and set ATOL at which \\(y\\) becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in bdf and adams solvers, and the same tolerances are used to control forward sensitivity calculation. See Serban and Hindmarsh (2005) for details.\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.\n\n\n\n\nThe adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with \\(N\\) additional states for each parameter for which gradients are needed. If there are \\(M\\) parameters for which sensitivities are required, then the augmented ODE system has a total of \\(N \\cdot (M +\n1)\\) states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.\nIn contrast, the adjoint ODE solver integrates forward in time a system of \\(N\\) equations to compute the ODE solution and then integrates backwards in time another system of \\(N\\) equations to get the sensitivities. Additionally, for \\(M\\) parameters there are \\(M\\) additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the bdf and adams forward sensitivity interfaces).\nThe solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.\nThe solver type (either bdf or adams) can be individually set for both the forward and backward solves.\nThe tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:\narray[T] vector[2] y_sim\n    = ode_adjoint_tol_ctl(sho, y0, t0, ts,\n                          relative_tolerance/9.0,                // forward tolerance\n                          rep_vector(absolute_tolerance/9.0, 2), // forward tolerance\n                          relative_tolerance/3.0,                // backward tolerance\n                          rep_vector(absolute_tolerance/3.0, 2), // backward tolerance\n                          relative_tolerance,                    // quadrature tolerance\n                          absolute_tolerance,                    // quadrature tolerance\n                          max_num_steps,\n                          150,                                   // number of steps between checkpoints\n                          1,                                     // interpolation polynomial: 1=Hermite, 2=polynomial\n                          2,                                     // solver for forward phase: 1=Adams, 2=BDF\n                          2,                                     // solver for backward phase: 1=Adams, 2=BDF\n                          theta);\nFor a detailed information on each argument please see the Stan function reference manual.\n\n\n\nLinear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.\nThe solution to \\(\\frac{d}{dt} y = ay\\) is \\(y = y_0e^{at}\\), where the constant \\(y_0\\) is determined by boundary conditions. We can extend this solution to the vector case: \\[\n\\frac{d}{dt}y = A \\, y\n\\]  where \\(y\\) is now a vector of length \\(n\\) and \\(A\\) is an \\(n\\) by \\(n\\) matrix. The solution is then given by: \\[\ny = e^{tA} \\, y_0\n\\]  where the matrix exponential is formally defined by the convergent power series: \\[\ne^{tA} = \\sum_{n=0}^{\\infty} \\dfrac{tA^n}{n!} = I + tA + \\frac{t^2A^2}{2!} + \\dotsb\n\\] \nWe can apply this technique to the simple harmonic oscillator example, by setting \\[\ny = \\begin{bmatrix} y_1 \\\\ y_2 \\end{bmatrix} \\qquad\nA = \\begin{bmatrix} 0 & 1 \\\\ -1 & -\\theta \\end{bmatrix}\n\\] \nThe Stan model to simulate noisy observations using a matrix exponential function is given below.\nIn general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  array[T] real ts;\n  array[1] real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim;\n  matrix[2, 2] A = [[ 0,  1],\n                    [-1, -theta[1]]]\n  for (t in 1:T) {\n    y_sim[t] = matrix_exp((t - 1) * A) * y0;\n  }\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThis Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters theta and initial state y0 are read in as data along observation times ts. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations y_sim. Because the ODEs are linear, we can use the matrix_exp function to solve the system.",
+    "text": "When there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept \\(\\alpha\\) and then a varying effect \\(\\beta_k\\) for age group \\(k\\) to make a linear predictor \\(\\alpha +\n\\beta_k\\). With this predictor, we can add a constant to \\(\\alpha\\) and subtract from each \\(\\beta_k\\) and get exactly the same likelihood.\nThe traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., \\(\\beta_1 = 0\\). With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to \\(\\beta_1\\).\nIn a Bayesian setting, a proper prior on each of the \\(\\beta\\) is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the collinearity section.\nAn alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., \\(\\sum_{k=1}^K \\beta_k = 0.\\)\nA parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the multi-logit section for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the item-response model section for details).\n\n\nAs of Stan 2.36, there is a built in sum_to_zero_vector type, which can be used as follows.\nparameters {\n  sum_to_zero_vector[K] beta;\n  // ...\n}\nThis produces a vector of size K such that sum(beta) = 0. In the unconstrained representation requires only K - 1 values because the last is determined by the first K - 1.\nPlacing a prior on beta in this parameterization, for example,\n  beta ~ normal(0, 1);\nleads to a subtly different posterior than what you would get with the same prior on an unconstrained size-K vector. As explained below, the variance is reduced.\nThe sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., \\(\\beta_K\n= -\\sum_{k=1}^{K-1} \\beta_k.\\) But that leads to high correlation among the \\(\\beta_k\\).\nThe transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; Seyboldt (2024) provides an explanation. The Stan Reference Manual provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.\n\n\nOn the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of beta,\nmodel {\n  beta ~ normal(0, inv(sqrt(1 - inv(K))));\n  // ...\n}\nThe scale component can be multiplied by sigma to produce a normal(0, sigma) prior marginally.\nTo generate distributions with marginals other than standard normal, the resulting beta may be scaled by some factor sigma and translated to some new location mu.\n\n\n\n\nAdding a prior such as \\(\\beta \\sim \\textsf{normal}(0,\\epsilon)\\) for a small \\(\\epsilon\\) will provide a kind of soft centering of a parameter vector \\(\\beta\\) by preferring, all else being equal, that \\(\\sum_{k=1}^K\n\\beta_k = 0\\). This approach is only guaranteed to roughly center if \\(\\beta\\) and the elementwise addition \\(\\beta + c\\) for a scalar constant \\(c\\) produce the same likelihood (perhaps by another vector \\(\\alpha\\) being transformed to \\(\\alpha - c\\), as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an \\(\\epsilon\\). If \\(\\epsilon\\) is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#notation",
-    "href": "stan-users-guide/odes.html#notation",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#ordered-logistic.section",
+    "href": "stan-users-guide/regression.html#ordered-logistic.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "An ODE is defined by a set of differential equations, \\(y(t, \\theta)' = f(t, y, \\theta)\\), and initial conditions, \\(y(t_0, \\theta) = y_0\\). The function \\(f(t, y, \\theta)\\) is called the system function. The \\(\\theta\\) dependence is included in the notation for \\(y(t, \\theta)\\) and \\(f(t, y, \\theta)\\) as a reminder that the solution is a function of any parameters used in the computation.",
+    "text": "Ordered regression for an outcome \\(y_n \\in \\{ 1, \\dotsc, k \\}\\) with predictors \\(x_n \\in \\mathbb{R}^D\\) is determined by a single coefficient vector \\(\\beta \\in \\mathbb{R}^D\\) along with a sequence of cutpoints \\(c \\in\n\\mathbb{R}^{K-1}\\) sorted so that \\(c_d &lt; c_{d+1}\\). The discrete output is \\(k\\) if the linear predictor \\(x_n \\beta\\) falls between \\(c_{k-1}\\) and \\(c_k\\), assuming \\(c_0 = -\\infty\\) and \\(c_K = \\infty\\). The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.\n\n\nThe ordered logistic model can be coded in Stan using the ordered data type for the cutpoints and the built-in ordered_logistic distribution.\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  matrix[N, D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  y ~ ordered_logistic(x * beta, c);\n}\nThe vector of cutpoints c is declared as ordered[K - 1], which guarantees that c[k] is less than c[k + 1].\nIf the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.\nThe equivalent model can be written using ordered_logistic_glm distribution, which can provide more efficient computation in case of higher dimensional beta.\n  y ~ ordered_logistic_glm(x,  beta, c);\n\n\nAn ordered probit model can be coded in exactly the same way by using the built-in ordered_probit distribution.\nmodel {\n  ordered_probit(x * beta, c);\n}\nThere is not yet an ordered_probit_glm distribution in Stan.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#example-simple-harmonic-oscillator",
-    "href": "stan-users-guide/odes.html#example-simple-harmonic-oscillator",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#hierarchical-regression",
+    "href": "stan-users-guide/regression.html#hierarchical-regression",
+    "title": "Regression Models",
     "section": "",
-    "text": "As an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair \\(y = (y_1, y_2)\\) representing position and speed. The change in the system with respect to time is given by the following differential equations.1\n\\[\\begin{align*}\n&\\frac{d}{dt} y_1 = y_2  \\\\\n&\\frac{d}{dt} y_2 = -y_1 - \\theta y_2\n\\end{align*}\\]\nThe state equations implicitly defines the state at future times as a function of an initial state and the system parameters.",
+    "text": "The simplest multilevel model is a hierarchical model in which the data are grouped into \\(L\\) distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients \\(\\beta\\). At the other extreme, an approach with no pooling assigns each level \\(l\\) its own coefficient vector \\(\\beta_l\\) that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.\nSuppose each binary outcome \\(y_n \\in \\{ 0, 1 \\}\\) has an associated level, \\(ll_n \\in \\{ 1, \\dotsc, L \\}\\). Each outcome will also have an associated predictor vector \\(x_n \\in \\mathbb{R}^D\\). Each level \\(l\\) gets its own coefficient vector \\(\\beta_l \\in \\mathbb{R}^D\\). The hierarchical structure involves drawing the coefficients \\(\\beta_{l,d}\n\\in \\mathbb{R}\\) from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.\nThe following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.\ndata {\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; L;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  array[N] int&lt;lower=1, upper=L&gt; ll;\n  array[N] row_vector[D] x;\n}\nparameters {\n  array[D] real mu;\n  array[D] real&lt;lower=0&gt; sigma;\n  array[L] vector[D] beta;\n}\nmodel {\n  for (d in 1:D) {\n    mu[d] ~ normal(0, 100);\n    for (l in 1:L) {\n      beta[l, d] ~ normal(mu[d], sigma[d]);\n    }\n  }\n  for (n in 1:N) {\n    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));\n  }\n}\nThe standard deviation parameter sigma gets an implicit uniform prior on \\((0,\\infty)\\) because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the regression priors section for recommendations on priors for regression coefficients and scales.\n\n\nWhere possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.\nThe first optimization vectorizes the for-loop over D as\nmu ~ normal(0, 100);\nfor (l in 1:L) {\n  beta[l] ~ normal(mu, sigma);\n}\nThe declaration of beta as an array of vectors means that the expression beta[l] denotes a vector. Although beta could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the indexing efficiency section for more information on the efficiency tradeoffs among arrays, vectors, and matrices.\nThis model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,3\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);\n}\nUnlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.\n{\n  vector[N] x_beta_ll;\n  for (n in 1:N) {\n    x_beta_ll[n] = x[n] * beta[ll[n]];\n  }\n  y ~ bernoulli_logit(x_beta_ll);\n}\nThe brackets introduce a new scope for the local variable x_beta_ll; alternatively, the variable may be declared at the top of the model block.\nIn some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#coding-the-ode-system-function",
-    "href": "stan-users-guide/odes.html#coding-the-ode-system-function",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#hierarchical-priors.section",
+    "href": "stan-users-guide/regression.html#hierarchical-priors.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "The first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the system function is time, passed as a real; the second argument to the system function is the system state, passed as a vector, and the return value from the system function are the current time derivatives of the state defined as a vector. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.\nThe simple harmonic oscillator can be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector sho(real t,        // time\n           vector y,      // state\n           real theta) {  // friction parameter\n  vector[2] dydt;\n  dydt[1] = y[2];\n  dydt[2] = -y[1] - theta * y[2];\n  return dydt;\n}\nThe function takes in a time t (a real), the system state y (a vector), and the parameter theta (a real). The function returns a vector of time derivatives of the system state at time t, state y, and parameter theta. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, t does not show up in the definition of dydt, however it is still required.\n\n\nThe types in the ODE system function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible ODE signatures:\nvector myode1(real t, vector y, real a0);\nvector myode2(real t, vector y, array[] int a0, vector a1);\nvector myode3(real t, vector y, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector myode1(real t, array[] real y, real a0);\n// Second argument is not a vector\narray[] real myode2(real t, vector y, real a0);\n// Return type is not a vector\nvector myode3(vector y, real a0);\n// First argument is not a real and second is not a vector",
+    "text": "Priors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.\n\n\nThe fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of \\(y_n \\in\n\\mathbb{R}\\) on \\(x_n \\in \\mathbb{R}^K\\), formulated as \\[\\begin{align*}\ny_n     & \\sim \\textsf{normal}(x_n \\beta, \\sigma) \\\\\n\\beta_k & \\sim \\textsf{normal}(0,\\tau) \\\\\n\\tau    & \\sim \\textsf{Cauchy}(0,2.5) \\\\\n\\end{align*}\\]\nIn this case, as \\(\\tau \\rightarrow 0\\) and \\(\\beta_k \\rightarrow 0\\), the posterior density \\[ p(\\beta,\\tau,\\sigma|y,x) \\propto p(y|x,\\beta,\\tau,\\sigma) \\] grows without bound. See the Neal’s funnel density, which has similar behavior.\nThere is obviously no MLE estimate for \\(\\beta,\\tau,\\sigma\\) in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by Chung et al. (2013) is to use a gamma distribution as a prior, such as \\[\n\\sigma \\sim \\textsf{Gamma}(2, 1/A),\n\\] for a reasonably large value of \\(A\\), such as \\(A = 10\\).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#measurement-error-models",
-    "href": "stan-users-guide/odes.html#measurement-error-models",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#item-response-models.section",
+    "href": "stan-users-guide/regression.html#item-response-models.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "Noisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.\n\n\nAs an example, suppose the simple harmonic oscillator has a parameter value of \\(\\theta = 0.15\\) and an initial state \\(y(t = 0, \\theta = 0.15) = (1, 0)\\). Assume the system is measured at 10 time points, \\(t = 1, 2, \\cdots, 10\\), where each measurement of \\(y(t, \\theta)\\) has independent \\(\\textsf{normal}(0, 0.1)\\) error in both dimensions (\\(y_1(t, \\theta)\\) and \\(y_2(t, \\theta)\\)).\nThe following model can be used to generate data like this:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  real t0;\n  array[T] real ts;\n  real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThe system parameters theta and initial state y0 are read in as data along with the initial time t0 and observation times ts. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations y_sim. Because the system is not stiff, the ode_rk45 solver is used.\nThis program illustrates the way in which the ODE solver is called in a Stan program,\narray[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\nthis returns the solution of the ODE initial value problem defined by system function sho, initial state y0, initial time t0, and parameter theta at the times ts. The call explicitly specifies the non-stiff RK45 solver.\nThe parameter theta is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, \\(\\theta\\) and \\(\\beta\\), the system function definition would look like:\nvector sho(real t, vector y, real theta, real beta) { ... }\nand the appropriate ODE solver call would be:\node_rk45(sho, y0, t0, ts, theta, beta);\nAny number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).\nBecause all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the T outputs with normal_rng.\n\n\n\nTypical realization of harmonic oscillator trajectory.\n\n\n\n\n\nThese ten noisy observations of the state can be used to estimate the friction parameter, \\(\\theta\\), the initial conditions, \\(y(t_0, \\theta)\\), and the scale of the noise in the problem. The full Stan model is:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  array[T] vector[2] y;\n  real t0;\n  array[T] real ts;\n}\nparameters {\n  vector[2] y0;\n  vector&lt;lower=0&gt;[2] sigma;\n  real theta;\n}\nmodel {\n  array[T] vector[2] mu = ode_rk45(sho, y0, t0, ts, theta);\n  sigma ~ normal(0, 2.5);\n  theta ~ std_normal();\n  y0 ~ std_normal();\n  for (t in 1:T) {\n    y[t] ~ normal(mu[t], sigma);\n  }\n}\nBecause the solves are now a function of model parameters, the ode_rk45 call is now made in the model block. There are half-normal priors on the measurement error scales sigma, and standard normal priors on theta and the initial state vector y0. The solutions to the ODE are assigned to mu, which is used as the location for the normal observation model.\nAs with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).",
+    "text": "Item-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see Gelman and Hill (2007, pps. 314–320) for a textbook introduction to hierarchical IRT models and Curtis (2010) for encodings of a range of IRT models in BUGS.\n\n\nThe data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.\ndata {\n  int&lt;lower=1&gt; J;                     // number of students\n  int&lt;lower=1&gt; K;                     // number of questions\n  int&lt;lower=1&gt; N;                     // number of observations\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // student for observation n\n  array[N] int&lt;lower=1, upper=K&gt; kk;  // question for observation n\n  array[N] int&lt;lower=0, upper=1&gt; y;   // correctness for observation n\n}\nThis declares a total of N student-question pairs in the data set, where each n in 1:N indexes a binary observation y[n] of the correctness of the answer of student jj[n] on question kk[n].\nThe prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.\n\n\n\nThe 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).\nThe model parameters are declared as follows.\nparameters {\n  real delta;            // mean student ability\n  array[J] real alpha;   // ability of student j - mean ability\n  array[K] real beta;    // difficulty of question k\n}\nThe parameter alpha[J] is the ability coefficient for student j and beta[k] is the difficulty coefficient for question k. The non-standard parameterization used here also includes an intercept term delta, which represents the average student’s response to the average question.4\nThe model itself is as follows.\nmodel {\n  alpha ~ std_normal();         // informative true prior\n  beta ~ std_normal();          // informative true prior\n  delta ~ normal(0.75, 1);      // informative true prior\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);\n  }\n}\nThis model uses the logit-parameterized Bernoulli distribution, where \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe key to understanding it is the term inside the bernoulli_logit distribution, from which it follows that \\[\n\\Pr[y_n = 1] = \\operatorname{logit}^{-1}\\left(\\alpha_{jj[n]} - \\beta_{kk[n]}\n+ \\delta\\right).\n\\]\nThe model suffers from additive identifiability issues without the priors. For example, adding a term \\(\\xi\\) to each \\(\\alpha_j\\) and \\(\\beta_k\\) results in the same predictions. The use of priors for \\(\\alpha\\) and \\(\\beta\\) located at 0 identifies the parameters; see Gelman and Hill (2007) for a discussion of identifiability issues and alternative approaches to identification.\nFor testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.\n\n\n\nThe simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.\nparameters {\n  real mu_beta;                // mean question difficulty\n  vector[J] alpha;             // ability for j - mean\n  vector[K] beta;              // difficulty for k\n  vector&lt;lower=0&gt;[K] gamma;    // discrimination of k\n  real&lt;lower=0&gt; sigma_beta;    // scale of difficulties\n  real&lt;lower=0&gt; sigma_gamma;   // scale of log discrimination\n}\nThe parameters should be clearer after the model definition.\nmodel {\n  alpha ~ std_normal();\n  beta ~ normal(0, sigma_beta);\n  gamma ~ lognormal(0, sigma_gamma);\n  mu_beta ~ cauchy(0, 5);\n  sigma_beta ~ cauchy(0, 5);\n  sigma_gamma ~ cauchy(0, 5);\n  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));\n}\nThe std_normal function is used here, defined by \\[\n\\texttt{std}\\mathtt{\\_}\\texttt{normal}(y)\n=\n\\textsf{normal}\\left(y \\mid 0, 1\\right).\n\\]\nThe distribution statement is also vectorized using elementwise multiplication; it is equivalent to\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(gamma[kk[n]]\n                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));\n}\nThe 2PL model is similar to the 1PL model, with the additional parameter gamma[k] modeling how discriminative question k is. If gamma[k] is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter gamma[k] is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.\nThe model is parameterized here with student abilities alpha being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the problematic posteriors chapter for further discussion of identifiability. The difficulty and discrimination parameters beta and gamma then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as\nbeta ~ normal(0, 5);\ngamma ~ lognormal(0, 2);\nThe point is that the alpha determines the scale and location and beta and gamma are allowed to float.\nThe beta parameter is here given a non-centered parameterization, with parameter mu_beta serving as the mean beta location. An alternative would’ve been to take:\nbeta ~ normal(mu_beta, sigma_beta);\nand\ny[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));\nNon-centered parameterizations tend to be more efficient in hierarchical models; see the reparameterization section for more information on non-centered reparameterizations.\nThe intercept term mu_beta can’t itself be modeled hierarchically, so it is given a weakly informative \\(\\textsf{Cauchy}(0,5)\\) prior. Similarly, the scale terms, sigma_beta, and sigma_gamma, are given half-Cauchy priors. As mentioned earlier, the scale and location for alpha are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to \\((0,\\infty)\\) by their declarations.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#stiff-ode.section",
-    "href": "stan-users-guide/odes.html#stiff-ode.section",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#priors-for-identification.section",
+    "href": "stan-users-guide/regression.html#priors-for-identification.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "Stiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.\nStan provides a specialized solver for stiff ODEs (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the rk45 suffix is replaced with bdf, as in\node_bdf(sho, y0, t0, ts, theta);\nUsing the stiff (bdf) solver on a system that is not stiff may be much slower than using the non-stiff (rk45) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.\nIf it is not known for sure that an ODE system is stiff, run the model with both the rk45 and bdf solvers and see which is faster. If the rk45 solver is faster, then the problem is probably non-stiff, and then it makes sense to try the adams solver as well. The adams solver uses higher order methods which can take larger timesteps than the rk45 solver, though similar to the bdf solver each of these steps is more expensive to compute.",
+    "text": "One application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the collinearity section for a simple example of the problems this poses for estimation.\nThe non-identifiability is resolved by providing a standard normal (i.e., \\(\\textsf{normal}(0,1)\\)) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.\n\n\n\nAnother case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.\n\n\n\nIn a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.\nSimilar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#control-ode.section",
-    "href": "stan-users-guide/odes.html#control-ode.section",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#multivariate-hierarchical-priors.section",
+    "href": "stan-users-guide/regression.html#multivariate-hierarchical-priors.section",
+    "title": "Regression Models",
     "section": "",
-    "text": "For additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters. These are the same as the regular function names but with _tol appended to the end. All three control arguments must be supplied with this signature (there are no defaults).\narray[T] vector[2] y_sim = ode_bdf_tol(sho, y0, t0, ts,\n                                 relative_tolerance,\n                                 absolute_tolerance,\n                                 max_num_steps,\n                                 theta);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nFor the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both \\(10^{-6}\\) and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million.\n\n\nIf there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.\nNonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times \\(0 &lt; t &lt; t'\\) and then positive for \\(t \\geq t'\\). In this example example, we would use code in the system such as\nif (t &lt; t_lag) {\n  return [0, 0]';\n} else {\n  // ... return non-zero vector...\n}\nIn general it is better to integrate up to t_lag in one solve and then integrate from t_lag onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.\nIf the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.\n\n\n\nThe relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns \\(y=(y_1,\\dots,y_n)^T\\), at every step the solver controls estimated local error \\(e=(e_1,\\dots,e_n)^T\\) through its weighted root-mean-square norm (Serban and Hindmarsh (2005), Hairer, Nørsett, and Wanner (1993))\n\\[\\begin{equation*}\n\\sqrt{\\sum_{i=1}^n{\\frac{1}{n}\\frac{e_i^2}{(\\text{RTOL}\\times y_i + \\text{ATOL})^2}}} &lt; 1\n\\end{equation*}\\] by reducing the stepsize when the inequality is not satisfied.\nTo understand the roles of the two tolerances it helps to assume \\(y\\) at opposite scales in the above expression: on one hand the absolute tolerance has little effect when \\(y_i \\gg 1\\), on the other the relative tolerance can not affect the norm when \\(y_i = 0\\). Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow Brenan, Campbell, and Petzold (1995) for a rule of thumb: let \\(m\\) be the number of significant digits required for \\(y\\), set \\(\\text{RTOL}=10^{-(m+1)}\\), and set ATOL at which \\(y\\) becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in bdf and adams solvers, and the same tolerances are used to control forward sensitivity calculation. See Serban and Hindmarsh (2005) for details.\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
+    "text": "In hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.\nAs an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.\n\n\nGelman and Hill (2007, chap. 13, Chapter 17) provide a discussion of a hierarchical model with \\(N\\) individuals organized into \\(J\\) groups. Each individual has a predictor row vector \\(x_n\\) of size \\(K\\); to unify the notation, they assume that \\(x_{n,1} = 1\\) is a fixed “intercept” predictor. To encode group membership, they assume individual \\(n\\) belongs to group \\(jj[n] \\in \\{ 1, \\dotsc, J \\}\\). Each individual \\(n\\) also has an observed outcome \\(y_n\\) taking on real values.\n\n\nThe model is a linear regression with slope and intercept coefficients varying by group, so that \\(\\beta_j\\) is the coefficient \\(K\\)-vector for group \\(j\\). The data model for individual \\(n\\) is then just \\[\ny_n \\sim \\textsf{normal}(x_n \\, \\beta_{jj[n]}, \\, \\sigma)\n\\quad\\text{for}\\quad n \\in \\{ 1, \\dotsc, N \\}.\n\\]\n\n\n\nGelman and Hill model the coefficient vectors \\(\\beta_j\\) as being drawn from a multivariate distribution with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\), \\[\n\\beta_j \\sim \\textsf{multivariate normal}(\\mu_j, \\, \\Sigma)\n\\quad\\text{for}\\quad j \\in \\{ 1, \\dotsc, J \\}.\n\\]\nBelow, we discuss the full model of Gelman and Hill, which uses group-level predictors to model \\(\\mu\\); for now, we assume \\(\\mu\\) is a simple vector parameter.\n\n\n\nFor hierarchical modeling, the group-level mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\) must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as \\[\n\\mu_j \\sim \\textsf{normal}(0,5).\n\\] If more is known about the expected coefficient values \\(\\beta_{j, k}\\), this information can be incorporated into the prior for \\(\\mu_j\\).\nFor the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling\nIn Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define \\[\n\\Sigma = \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau) \\times \\Omega \\times \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau),\n\\] where \\(\\Omega\\) is a correlation matrix and \\(\\tau\\) is the vector of coefficient scales. This mapping from scale vector \\(\\tau\\) and correlation matrix \\(\\Omega\\) can be inverted, using \\[\n\\tau_k = \\sqrt{\\Sigma_{k,k}}\n\\quad\\textsf{and}\\quad\n\\Omega_{i, j} = \\frac{\\Sigma_{i, j}}{\\tau_i \\, \\tau_j}.\n\\]\nThe components of the scale vector \\(\\tau\\) can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as \\[\n\\tau_k \\sim \\textsf{Cauchy}(0, 2.5)\n\\quad\\text{for}\\quad k \\in \\{ 1, \\dotsc, K \\}\n\\quad\\text{constrained\\ by}\\quad \\tau_k &gt; 0.\n\\] As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for \\(\\tau\\). For large numbers of exchangeable coefficients, the components of \\(\\tau\\) itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.\nOur final recommendation is to give the correlation matrix \\(\\Omega\\) an LKJ prior with shape \\(\\eta \\geq 1\\),5\n\\[\n\\Omega \\sim \\textsf{LKJCorr}(\\eta).\n\\]\nThe LKJ correlation distribution is defined by \\[\n\\textsf{LKJCorr}\\left(\\Sigma \\mid \\eta\\right)\n\\propto\n\\operatorname{det}\\left(\\Sigma\\right)^{\\eta - 1}.\n\\]\nThe basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For \\(\\eta = 1\\), the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.\nFor \\(\\eta &gt; 1\\), the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For \\(\\eta &lt; 1\\), it increasingly concentrates mass in the other direction, i.e., favoring more correlation.\nThe LKJ prior may thus be used to control the expected amount of correlation among the parameters \\(\\beta_j\\). For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see Barnard, McCulloch, and Meng (2000).\n\n\n\nTo complete Gelman and Hill’s model, suppose each group \\(j \\in \\{ 1, \\dotsc, J \\}\\) is supplied with an \\(L\\)-dimensional row-vector of group-level predictors \\(u_j\\). The prior mean for the \\(\\beta_j\\) can then itself be modeled as a regression, using an \\(L\\)-dimensional coefficient vector \\(\\gamma\\). The prior for the group-level coefficients then becomes \\[\n\\beta_j \\sim \\textsf{multivariate normal}(u_j \\, \\gamma, \\Sigma)\n\\]\nThe group-level coefficients \\(\\gamma\\) may themselves be given independent weakly informative priors, such as \\[\n\\gamma_l \\sim \\textsf{normal}(0,5).\n\\] As usual, information about the group-level means should be incorporated into this prior.\n\n\n\nThe Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.\ndata {\n  int&lt;lower=0&gt; N;              // num individuals\n  int&lt;lower=1&gt; K;              // num ind predictors\n  int&lt;lower=1&gt; J;              // num groups\n  int&lt;lower=1&gt; L;              // num group predictors\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // group for individual\n  matrix[N, K] x;              // individual predictors\n  array[J] row_vector[L] u;    // group predictors\n  vector[N] y;                 // outcomes\n}\nparameters {\n  corr_matrix[K] Omega;        // prior correlation\n  vector&lt;lower=0&gt;[K] tau;      // prior scale\n  matrix[L, K] gamma;          // group coeffs\n  array[J] vector[K] beta;     // indiv coeffs by group\n  real&lt;lower=0&gt; sigma;         // prediction error scale\n}\nmodel {\n  tau ~ cauchy(0, 2.5);\n  Omega ~ lkj_corr(2);\n  to_vector(gamma) ~ normal(0, 5);\n  {\n    array[J] row_vector[K] u_gamma;\n    for (j in 1:J) {\n      u_gamma[j] = u[j] * gamma;\n    }\n    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));\n  }\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n  }\n}\nThe hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix Omega and scale vector tau are more natural to inspect in the output; to output Sigma, define it as a transformed parameter. The function quad_form_diag is defined so that quad_form_diag(Sigma, tau) is equivalent to diag_matrix(tau) * Sigma * diag_matrix(tau), where diag_matrix(tau) returns the matrix with tau on the diagonal and zeroes off diagonal; the version using quad_form_diag should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.\n\n\n\nThe code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:\nfor (n in 1:N) {\n  y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n}\nwith the vectorized distribution statement:\n{\n  vector[N] x_beta_jj;\n  for (n in 1:N) {\n    x_beta_jj[n] = x[n] * beta[jj[n]];\n  }\n  y ~ normal(x_beta_jj, sigma);\n}\nThe outer brackets create a local scope in which to define the variable x_beta_jj, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.\nAlthough it is tempting to redeclare beta and include a revised model block distribution statement,\nparameters {\n  matrix[J, K] beta;\n// ...\n}\nmodel {\n  y ~ normal(rows_dot_product(x, beta[jj]), sigma);\n  // ...\n}\nthis fails because it breaks the vectorization for beta,6\nbeta ~ multi_normal(...);\nwhich requires beta to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because rows_dot_product cannot do much optimization in and of itself because there are no shared computations.\nThe code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.\n{\n  matrix[K, K] Sigma_beta;\n  Sigma_beta = quad_form_diag(Omega, tau);\n  for (j in 1:J) {\n    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);\n  }\n}\nIn this example, the covariance matrix Sigma_beta is defined as a local variable so as not to have to repeat the quadratic form computation \\(J\\) times. This vectorization can be combined with the Cholesky-factor optimization in the next section.\n\n\n\nThe multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.\ndata {\n  matrix[L, J] u;              // group predictors transposed\n  // ...\n}\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  matrix[K, L] gamma;\n  // ...\n}\ntransformed parameters {\n  matrix[K, J] beta;\n  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  // ...\n}\nThe data variable u was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with gamma and beta, to minimize the number of transposition operations. The new parameter L_Omega is the Cholesky factor of the original correlation matrix Omega, so that\nOmega = L_Omega * L_Omega'\nThe prior scale vector tau is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,\nSigma_beta\n  = quad_form_diag(Omega, tau)\n  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'\nwhere the diagonal pre-multiply compound operation is defined by\ndiag_pre_multiply(a, b) = diag_matrix(a) * b\nThe new variable z is declared as a matrix, the entries of which are given independent standard normal priors; the to_vector operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of z being a \\(K\\)-variate normal random vector with the identity as covariance matrix. Therefore, multiplying z by the Cholesky factor of the covariance matrix and adding the mean (u * gamma)' produces a beta distributed as in the original model, where the variance is, letting \\(L = \\mathrm{diag}(\\tau)\\,\\Omega_L\\),\n\\[\n\\begin{aligned}\n\\mathbb{V}[\\beta] &= \\mathbb{E}\\big((L \\, z)(L \\, z)^\\top) \\\\\n&= \\mathbb{E}\\big((L \\, z \\, z^\\top \\, L^\\top) \\\\\n&= L \\, \\mathbb{E}(z \\, z^\\top) \\, L^\\top \\\\\n&= L \\, L^\\top =(\\mathrm{diag}(\\tau)\\,\\Omega_L)\\,(\\mathrm{diag}(\\tau)\\,\\Omega_L)^\\top \\\\\n&= \\mathrm{diag}(\\tau)\\,\\Omega\\,\\mathrm{diag}(\\tau) \\\\\n&= \\Sigma.\n\\end{aligned}\n\\] Where we have used the linearity of expectations (line 2 to 3), the definition of \\(\\Omega = \\Omega_L \\, \\Omega_L^\\top\\), and the fact that \\(\\mathbb{E}(z \\, z^\\top) = I\\) since \\(z \\sim \\mathcal{N}(0, I)\\).\nOmitting the remaining data declarations, which are the same as before with the exception of u, the optimized model is as follows.\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0, upper=pi() / 2&gt;[K] tau_unif;  // prior scale\n  matrix[K, L] gamma;                        // group coeffs\n  real&lt;lower=0&gt; sigma;                       // prediction error scale\n}\ntransformed parameters {\n  vector&lt;lower=0&gt;[K] tau = 2.5 * tan(tau_unif);\n  matrix[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  vector[N] mu;\n  for(n in 1:N) {\n    mu[n] = x[n, ] * beta[, jj[n]];\n  }\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  to_vector(gamma) ~ normal(0, 5);\n  y ~ normal(mu, sigma);\n}\nThis model also reparameterizes the prior scale tau to avoid potential problems with the heavy tails of the Cauchy distribution. The statement tau_unif ~ uniform(0, pi() / 2) can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#adjoint-ode.section",
-    "href": "stan-users-guide/odes.html#adjoint-ode.section",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#prediction-forecasting-and-backcasting",
+    "href": "stan-users-guide/regression.html#prediction-forecasting-and-backcasting",
+    "title": "Regression Models",
     "section": "",
-    "text": "The adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with \\(N\\) additional states for each parameter for which gradients are needed. If there are \\(M\\) parameters for which sensitivities are required, then the augmented ODE system has a total of \\(N \\cdot (M +\n1)\\) states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.\nIn contrast, the adjoint ODE solver integrates forward in time a system of \\(N\\) equations to compute the ODE solution and then integrates backwards in time another system of \\(N\\) equations to get the sensitivities. Additionally, for \\(M\\) parameters there are \\(M\\) additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the bdf and adams forward sensitivity interfaces).\nThe solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.\nThe solver type (either bdf or adams) can be individually set for both the forward and backward solves.\nThe tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:\narray[T] vector[2] y_sim\n    = ode_adjoint_tol_ctl(sho, y0, t0, ts,\n                          relative_tolerance/9.0,                // forward tolerance\n                          rep_vector(absolute_tolerance/9.0, 2), // forward tolerance\n                          relative_tolerance/3.0,                // backward tolerance\n                          rep_vector(absolute_tolerance/3.0, 2), // backward tolerance\n                          relative_tolerance,                    // quadrature tolerance\n                          absolute_tolerance,                    // quadrature tolerance\n                          max_num_steps,\n                          150,                                   // number of steps between checkpoints\n                          1,                                     // interpolation polynomial: 1=Hermite, 2=polynomial\n                          2,                                     // solver for forward phase: 1=Adams, 2=BDF\n                          2,                                     // solver for backward phase: 1=Adams, 2=BDF\n                          theta);\nFor a detailed information on each argument please see the Stan function reference manual.",
+    "text": "Stan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).\n\n\nAs a simple example, the following linear regression provides the same setup for estimating the coefficients beta as in our very first example, using y for the N observations and x for the N predictor vectors. The model parameters and model for observations are exactly the same as before.\nTo make predictions, we need to be given the number of predictions, N_new, and their predictor matrix, x_new. The predictions themselves are modeled as a parameter y_new. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector y_new and prediction matrix x_new.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=0&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n\n  int&lt;lower=0&gt; N_new;\n  matrix[N_new, K] x_new;\n}\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n\n  vector[N_new] y_new;                  // predictions\n}\nmodel {\n  y ~ normal(x * beta, sigma);          // observed model\n\n  y_new ~ normal(x_new * beta, sigma);  // prediction model\n}\n\n\n\nWhere possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.\n// ...data as above...\n\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(x * beta, sigma);\n}\ngenerated quantities {\n  vector[N_new] y_new;\n  for (n in 1:N_new) {\n    y_new[n] = normal_rng(x_new[n] * beta, sigma);\n  }\n}\nNow the data are just as before, but the parameter y_new is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.\n\n\nIt is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of \\(2^{31}\\), must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by \\(2^{31} - 1\\) in Stan.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#solving-a-system-of-linear-odes-using-a-matrix-exponential",
-    "href": "stan-users-guide/odes.html#solving-a-system-of-linear-odes-using-a-matrix-exponential",
-    "title": "Ordinary Differential Equations",
+    "objectID": "stan-users-guide/regression.html#multivariate-outcomes",
+    "href": "stan-users-guide/regression.html#multivariate-outcomes",
+    "title": "Regression Models",
     "section": "",
-    "text": "Linear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.\nThe solution to \\(\\frac{d}{dt} y = ay\\) is \\(y = y_0e^{at}\\), where the constant \\(y_0\\) is determined by boundary conditions. We can extend this solution to the vector case: \\[\n\\frac{d}{dt}y = A \\, y\n\\]  where \\(y\\) is now a vector of length \\(n\\) and \\(A\\) is an \\(n\\) by \\(n\\) matrix. The solution is then given by: \\[\ny = e^{tA} \\, y_0\n\\]  where the matrix exponential is formally defined by the convergent power series: \\[\ne^{tA} = \\sum_{n=0}^{\\infty} \\dfrac{tA^n}{n!} = I + tA + \\frac{t^2A^2}{2!} + \\dotsb\n\\] \nWe can apply this technique to the simple harmonic oscillator example, by setting \\[\ny = \\begin{bmatrix} y_1 \\\\ y_2 \\end{bmatrix} \\qquad\nA = \\begin{bmatrix} 0 & 1 \\\\ -1 & -\\theta \\end{bmatrix}\n\\] \nThe Stan model to simulate noisy observations using a matrix exponential function is given below.\nIn general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  array[T] real ts;\n  array[1] real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim;\n  matrix[2, 2] A = [[ 0,  1],\n                    [-1, -theta[1]]]\n  for (t in 1:T) {\n    y_sim[t] = matrix_exp((t - 1) * A) * y0;\n  }\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThis Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters theta and initial state y0 are read in as data along observation times ts. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations y_sim. Because the ODEs are linear, we can use the matrix_exp function to solve the system.",
+    "text": "Most regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.\nThis section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.\n\n\nThe first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors (Zellner 1962; Greene 2011).\nThe model is easy to write down as a regression, \\[\\begin{align*}\ny_n        &=  x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nwhere \\(x_n\\) is a \\(J\\)-row-vector of predictors (\\(x\\) is an \\((N \\times\nJ)\\) matrix), \\(y_n\\) is a \\(K\\)-vector of observations, \\(\\beta\\) is a \\((K\n\\times J)\\) matrix of regression coefficients (vector \\(\\beta_k\\) holds coefficients for outcome \\(k\\)), and \\(\\Sigma\\) is covariance matrix governing the error. As usual, the intercept can be rolled into \\(x\\) as a column of ones.\nThe basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; J;\n  int&lt;lower=0&gt; N;\n  array[N] vector[J] x;\n  array[N] vector[K] y;\n}\nparameters {\n  matrix[K, J] beta;\n  cov_matrix[K] Sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n  }\n  y ~ multi_normal(mu, Sigma);\n}\nFor efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.\nFollowing the advice in the multivariate hierarchical priors section, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.\n// ...\nparameters {\n  matrix[K, J] beta;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0&gt;[K] L_sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  matrix[K, K] L_Sigma;\n\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n\n  }\n\n  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);\n\n  to_vector(beta) ~ normal(0, 5);\n  L_Omega ~ lkj_corr_cholesky(4);\n  L_sigma ~ cauchy(0, 2.5);\n\n  y ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix beta to a vector.\nIf required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.\n\n\n\nThe multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.\nThe observations \\(y_n\\) are \\(D\\)-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations \\(y_n\\) are based on latent values \\(z_n\\) drawn from a seemingly unrelated regression model (see the previous section), \\[\\begin{align*}\nz_n        &=     x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim  \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nThese are then put through the step function to produce a \\(K\\)-vector \\(z_n\\) of boolean values with elements defined by \\[\ny_{n, k} = \\operatorname{I}\\left(z_{n, k} &gt; 0\\right),\n\\] where \\(\\operatorname{I}()\\) is the indicator function taking the value 1 if its argument is true and 0 otherwise.\nUnlike in the seemingly unrelated regressions case, here the covariance matrix \\(\\Sigma\\) has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see Greene (2011)).\nMultivariate probit regression can be coded in Stan using the trick introduced by Albert and Chib (1993), where the underlying continuous value vectors \\(y_n\\) are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector \\(z\\) in two parts, based on whether the corresponding value of \\(y\\) is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.\nFirst, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in \\(y\\).\nfunctions {\n  int sum2d(array[,] int a) {\n    int s = 0;\n    for (i in 1:size(a)) {\n      s += sum(a[i]);\n    }\n    return s;\n  }\n}\nThe function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.\nThe data declaration block is much like for the seemingly unrelated regressions, but the observations y are now integers constrained to be 0 or 1.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  array[N, D] int&lt;lower=0, upper=1&gt; y;\n  array[N] vector[K] x;\n}\nAfter declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array y into positive and negative components, keeping track of indexes so that z can be easily reassembled in the transformed parameters block.\ntransformed data {\n  int&lt;lower=0&gt; N_pos;\n  array[sum2d(y)] int&lt;lower=1, upper=N&gt; n_pos;\n  array[size(n_pos)] int&lt;lower=1, upper=D&gt; d_pos;\n  int&lt;lower=0&gt; N_neg;\n  array[(N * D) - size(n_pos)] int&lt;lower=1, upper=N&gt; n_neg;\n  array[size(n_neg)] int&lt;lower=1, upper=D&gt; d_neg;\n\n  N_pos = size(n_pos);\n  N_neg = size(n_neg);\n  {\n    int i;\n    int j;\n    i = 1;\n    j = 1;\n    for (n in 1:N) {\n      for (d in 1:D) {\n        if (y[n, d] == 1) {\n          n_pos[i] = n;\n          d_pos[i] = d;\n          i += 1;\n        } else {\n          n_neg[j] = n;\n          d_neg[j] = d;\n          j += 1;\n        }\n      }\n    }\n  }\n}\nThe variables N_pos and N_neg are set to the number of true (1) and number of false (0) observations in y. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.\nThe parameters are declared as follows.\nparameters {\n  matrix[D, K] beta;\n  cholesky_factor_corr[D] L_Omega;\n  vector&lt;lower=0&gt;[N_pos] z_pos;\n  vector&lt;upper=0&gt;[N_neg] z_neg;\n}\nThese include the regression coefficients beta and the Cholesky factor of the correlation matrix, L_Omega. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).\nThe critical part of the parameter declaration is that the latent real value \\(z\\) is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct \\(z\\).\ntransformed parameters {\n  array[N] vector[D] z;\n  for (n in 1:N_pos) {\n    z[n_pos[n], d_pos[n]] = z_pos[n];\n  }\n  for (n in 1:N_neg) {\n    z[n_neg[n], d_neg[n]] = z_neg[n];\n  }\n}\nAt this point, the model is simple, pretty much recreating the seemingly unrelated regression.\nmodel {\n  L_Omega ~ lkj_corr_cholesky(4);\n  to_vector(beta) ~ normal(0, 5);\n  {\n    array[N] vector[D] beta_x;\n    for (n in 1:N) {\n      beta_x[n] = beta * x[n];\n    }\n    z ~ multi_normal_cholesky(beta_x, L_Omega);\n  }\n}\nThis simple form of model is made possible by the Albert and Chib-style constraints on z.\nFinally, the correlation matrix itself can be put back together in the generated quantities block if desired.\ngenerated quantities {\n  corr_matrix[D] Omega;\n  Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\nThe same could be done for the seemingly unrelated regressions in the previous section.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/odes.html#footnotes",
-    "href": "stan-users-guide/odes.html#footnotes",
-    "title": "Ordinary Differential Equations",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThis example is drawn from the documentation for the Boost Numeric Odeint library (Ahnert and Mulansky 2011), which Stan uses to implement the rk45 and ckrk solver.↩︎",
+    "objectID": "stan-users-guide/regression.html#applications-of-pseudorandom-number-generation",
+    "href": "stan-users-guide/regression.html#applications-of-pseudorandom-number-generation",
+    "title": "Regression Models",
+    "section": "",
+    "text": "The main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.\n\n\nConsider predicting unobserved outcomes using linear regression. Given predictors \\(x_1, \\dotsc, x_N\\) and observed outcomes \\(y_1, \\dotsc, y_N\\), and assuming a standard linear regression with intercept \\(\\alpha\\), slope \\(\\beta\\), and error scale \\(\\sigma\\), along with improper uniform priors, the posterior over the parameters given \\(x\\) and \\(y\\) is \\[\np\\left(\\alpha, \\beta, \\sigma \\mid x, y \\right)\n\\propto\n\\prod_{n=1}^N\n  \\textsf{normal}\\left(y_n \\mid \\alpha + \\beta x_n, \\sigma\\right).\n\\]\nFor this model, the posterior predictive inference for a new outcome \\(\\tilde{y}_m\\) given a predictor \\(\\tilde{x}_m\\), conditioned on the observed data \\(x\\) and \\(y\\), is \\[\np\\left(\\tilde{y}_n \\mid \\tilde{x}_n, x, y\\right)\n= \\int_{(\\alpha,\\beta,\\sigma)}\n  \\textsf{normal}\\left(\\tilde{y}_n \\mid \\alpha + \\beta \\tilde{x}_n, \\sigma\\right)\n  \\times\n  p\\left(\\alpha, \\beta, \\sigma \\mid x, y\\right)\n  \\,\\textrm{d}(\\alpha,\\beta,\\sigma).\n\\]\nTo code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde;\n  for (n in 1:N_tilde) {\n    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);\n  }\n}\nGiven observed predictors \\(x\\) and outcomes \\(y\\), y_tilde will be drawn according to \\(p\\left(\\tilde{y} \\mid \\tilde{x}, y, x\\right)\\). This means that, for example, the posterior mean for y_tilde is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).\n\n\n\nA good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors \\(x\\) are used in place of new predictors \\(\\tilde{x}\\) for unobserved outcomes. If the model fits the data well, the predictions for \\(\\tilde{y}\\) based on \\(x\\) should match the observed data \\(y\\).\nTo code posterior predictive checks in Stan requires only a slight modification of the prediction code to use \\(x\\) and \\(N\\) in place of \\(\\tilde{x}\\) and \\(\\tilde{N}\\),\ngenerated quantities {\n  vector[N] y_tilde;\n  for (n in 1:N) {\n    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);\n  }\n}\nGelman et al. (2013) recommend choosing several posterior draws \\(\\tilde{y}^{(1)}, \\dotsc, \\tilde{y}^{(M)}\\) and plotting each of them alongside the data \\(y\\) that was actually observed. If the model fits well, the simulated \\(\\tilde{y}\\) will look like the actual data \\(y\\).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Ordinary Differential Equations"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html",
-    "href": "stan-users-guide/missing-data.html",
-    "title": "Missing Data and Partially Known Parameters",
-    "section": "",
-    "text": "Bayesian inference supports a general approach to missing data in which any missing data item is represented as a parameter that is estimated in the posterior (Gelman et al. 2013). If the missing data are not explicitly modeled, as in the predictors for most regression models, then the result is an improper prior on the parameter representing the missing predictor.\nMixing arrays of observed and missing data can be difficult to include in Stan, partly because it can be tricky to model discrete unknowns in Stan and partly because unlike some other statistical languages (for example, R and Bugs), Stan requires observed and unknown quantities to be defined in separate places in the model. Thus it can be necessary to include code in a Stan program to splice together observed and missing parts of a data structure. Examples are provided later in the chapter.\n\n\nStan treats variables declared in the data and transformed data blocks as known and the variables in the parameters block as unknown.\nAn example involving missing normal observations could be coded as follows.1\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  array[N_mis] real y_mis;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_mis ~ normal(mu, sigma);\n}\nThe number of observed and missing data points are coded as data with non-negative integer variables N_obs and N_mis. The observed data are provided as an array data variable y_obs. The missing data are coded as an array parameter, y_mis. The ordinary parameters being estimated, the location mu and scale sigma, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.\n\n\n\nIn some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the transformed parameters block and assigning to it.\nThe following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] vector[2] y;\n  real&lt;lower=0&gt; var1;\n  real&lt;lower=0&gt; var2;\n}\ntransformed data {\n  real&lt;lower=0&gt; max_cov = sqrt(var1 * var2);\n  real&lt;upper=0&gt; min_cov = -max_cov;\n}\nparameters {\n  vector[2] mu;\n  real&lt;lower=min_cov, upper=max_cov&gt; cov;\n}\ntransformed parameters {\n  matrix[2, 2] Sigma;\n  Sigma[1, 1] = var1;\n  Sigma[1, 2] = cov;\n  Sigma[2, 1] = cov;\n  Sigma[2, 2] = var2;\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);\n}\nThe variances are defined as data in variables var1 and var2, whereas the covariance is defined as a parameter in variable cov. The \\(2 \\times 2\\) covariance matrix Sigma is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.\nThe constraint on the covariance declaration ensures that the resulting covariance matrix sigma is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.\nThe vectorization of the multivariate normal is critical for efficiency here. The transformed parameter Sigma could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.\n\n\n\nIf the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_obs;\n  array[N_mis] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_mis;\n  array[N_obs] real y_obs;\n}\ntransformed data {\n  int&lt;lower=0&gt; N = N_obs + N_mis;\n}\nparameters {\n  array[N_mis] real y_mis;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  array[N] real y;\n  y[ii_obs] = y_obs;\n  y[ii_mis] = y_mis;\n}\nmodel {\n  sigma ~ gamma(1, 1);\n  y[1] ~ normal(0, 100);\n  y[2:N] ~ normal(y[1:(N - 1)], sigma);\n}\nThe index arrays ii_obs and ii_mis contain the indexes into the final array y of the observed data (coded as a data vector y_obs) and the missing data (coded as a parameter vector y_mis). See the time series chapter for further discussion of time-series model and specifically the autoregression section for an explanation of the vectorization for y as well as an explanation of how to convert this example to a full AR(1) model. To ensure y[1] has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.\nAnother potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in\nx[N_obs_2, 2] = x_obs_2;\nx[N_mis_2, 2] = x_mis_2;\nwhere the relevant variables are all hard coded with index 2 because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the ragged data structures section for a general discussion of coding ragged data structures in Stan).\n\n\n\nRick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis (Aguilar and West 2000). This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.\ndata {\n  int&lt;lower=2&gt; K;\n}\ntransformed data {\n  int&lt;lower=1&gt; K_choose_2;\n  K_choose_2 = (K * (K - 1)) / 2;\n}\nparameters {\n  vector[K_choose_2] L_lower;\n}\ntransformed parameters {\n  cholesky_factor_cov[K] L;\n  for (k in 1:K) {\n    L[k, k] = 1;\n  }\n  {\n    int i;\n    for (m in 2:K) {\n      for (n in 1:(m - 1)) {\n        L[m, n] = L_lower[i];\n        L[n, m] = 0;\n        i += 1;\n      }\n    }\n  }\n}\nIt is most convenient to place a prior directly on L_lower. An alternative would be a prior for the full Cholesky factor L, because the transform from L_lower to L is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix L   * L', because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.\n\n\n\nIt’s often the case that one or more components of a multivariate outcome are missing.2\nAs an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations y and a boolean array y_observed to indicate which values were observed (others can have dummy values in the input).\narray[N] vector[2] y;\narray[N, 2] int&lt;lower=0, upper=1&gt; y_observed;\nIf both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    y[n] ~ multi_normal(mu, Sigma);\n  } else if (y_observed[n, 1]) {\n    y[n, 1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\n  } else if (y_observed[n, 2]) {\n    y[n, 2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\n  }\n}\nIt’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:\ntransformed data {\n  array[observed_12(y_observed)] int ns12;\n  array[observed_1(y_observed)] int ns1;\n  array[observed_2(y_observed)] int ns2;\n}\nYou will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.\nint n12 = 1;\nint n1 = 1;\nint n2 = 1;\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    ns12[n12] = n;\n    n12 += 1;\n  } else if (y_observed[n, 1]) {\n    ns1[n1] = n;\n    n1 += 1;\n  } else if (y_observed[n, 2]) {\n    ns2[n2] = n;\n    n2 += 1;\n  }\n}\nThen, in the model block, everything is vectorizable using those indexes constructed once in transformed data:\ny[ns12] ~ multi_normal(mu, Sigma);\ny[ns1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\ny[ns2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\nThe result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.\nThis approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).",
+    "objectID": "stan-users-guide/regression.html#footnotes",
+    "href": "stan-users-guide/regression.html#footnotes",
+    "title": "Regression Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nUnlike in Python and R, which are interpreted, Stan is translated to C++ and compiled, so loops and assignment statements are fast. Vectorized code is faster in Stan because (a) the expression tree used to compute derivatives can be simplified, leading to fewer virtual function calls, and (b) computations that would be repeated in the looping version, such as log(sigma) in the above model, will be computed once and reused.↩︎\nThe Phi_approx function is a rescaled version of the inverse logit function, so while the scale is roughly the same \\(\\Phi\\), the tails do not match.↩︎\nThe Bernoulli-logit distribution builds in the log link function, taking \\[\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right) = \\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\\]↩︎\nGelman and Hill (2007) treat the \\(\\delta\\) term equivalently as the location parameter in the distribution of student abilities.↩︎\nThe prior is named for Lewandowski, Kurowicka, and Joe, as it was derived by inverting the random correlation matrix generation strategy of Lewandowski, Kurowicka, and Joe (2009).↩︎\nThanks to Mike Lawrence for pointing this out in the GitHub issue for the manual.↩︎",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Regression Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html#missing-data",
-    "href": "stan-users-guide/missing-data.html#missing-data",
-    "title": "Missing Data and Partially Known Parameters",
+    "objectID": "stan-users-guide/proportionality-constants.html",
+    "href": "stan-users-guide/proportionality-constants.html",
+    "title": "Proportionality Constants",
     "section": "",
-    "text": "Stan treats variables declared in the data and transformed data blocks as known and the variables in the parameters block as unknown.\nAn example involving missing normal observations could be coded as follows.1\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  array[N_mis] real y_mis;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_mis ~ normal(mu, sigma);\n}\nThe number of observed and missing data points are coded as data with non-negative integer variables N_obs and N_mis. The observed data are provided as an array data variable y_obs. The missing data are coded as an array parameter, y_mis. The ordinary parameters being estimated, the location mu and scale sigma, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.",
+    "text": "When evaluating a likelihood or prior as part of the log density computation in MCMC, variational inference, or optimization, it is usually only necessary to compute the functions up to a proportionality constant (or similarly compute log densities up to an additive constant). In MCMC this comes from the fact that the distribution being sampled does not need to be normalized (and so it is the normalization constant that is ignored). Similarly the distribution does not need normalized to perform variational inference or do optimizations. The advantage of working with unnormalized distributions is they can make computation quite a bit cheaper.\nThere are three different syntaxes to build the model in Stan. The way to select between them is by determining if the proportionality constants are necessary. If performance is not a problem, it is always safe to use the normalized densities.\nThe distribution statement (~) and log density increment statement (target +=) with _lupdf() use unnormalized densities for \\(x\\) (dropping proportionality constants):\nx ~ normal(0, 1);\ntarget += normal_lupdf(x | 0, 1); // the 'u' is for unnormalized\nThe log density increment statement (target +=) with _lpdf() uses the full normalized density for \\(x\\) (dropping no constants):\ntarget += normal_lpdf(x | 0, 1);\nFor discrete distributions, the target += syntax is using _lupmf and _lpmf instead:\ny ~ bernoulli(0.5);\ntarget += bernoulli_lupmf(y | 0.5);\ntarget += bernoulli_lpmf(y | 0.5);\n\n\nIf a density \\(p(\\theta)\\) can be factored into \\(K g(\\theta)\\) where \\(K\\) are all the factors that are a not a function of \\(\\theta\\) and \\(g(\\theta)\\) are all the terms that are a function of \\(\\theta\\), then it is said that \\(g(\\theta)\\) is proportional to \\(p(\\theta)\\) up to a constant.\nThe advantage of all this is that sometimes \\(K\\) is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.\nStan takes advantage of the proportionality constant fact with the ~ syntax. Take for instance the normal data model:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nSyntactically, this is just shorthand for the equivalent model that replaces the ~ syntax with a target += statement and a normal_lupdf function call:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  target += normal_lupdf(x | mu, sigma);\n}\nThe function normal_lupdf is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\nNow because the density here is only a function of \\(x\\), the additive terms in the log density that are not a function of \\(x\\) can be dropped. In this case it is enough to know only the quadratic term:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\n\n\n\nIn the case that the proportionality constants were needed for a normal log density the function normal_lpdf can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the ~ or target += normal_lupdf syntaxes if it is absolutely clear that the proportionality constants are not necessary.\n\n\n\nWhen a custom _lpdf or _lpmf function is defined, the compiler will automatically make available a _lupdf or _lupmf version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.\nThe difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.\nThe following code demonstrates the different behaviors:\nfunctions {\n  real custom1_lpdf(x) {\n    return normal_lupdf(x | 0.0, 1.0);\n  }\n  real custom2_lpdf(x) {\n    return normal_lpdf(x | 0.0, 1.0);\n  }\n}\nparameters {\n  real mu;\n}\nmodel {\n  mu ~ custom1(); // Normalization constants dropped\n  target += custom1_lupdf(mu); // Normalization constants dropped\n  target += custom1_lpdf(mu);  // Normalization constants kept\n\n  mu ~ custom2();  // Normalization constants kept\n  target += custom2_lupdf(mu);  // Normalization constants kept\n  target += custom2_lpdf(mu);  // Normalization constants kept\n}\n\n\n\nTo avoid ambiguities in how the normalization constants work, functions ending in _lupdf and _lupmf can only be used in the model block or user-defined probability functions (functions ending in _lpdf or _lpmf).",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Programming Techniques",
+      "Proportionality Constants"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html#partially-known-parameters.section",
-    "href": "stan-users-guide/missing-data.html#partially-known-parameters.section",
-    "title": "Missing Data and Partially Known Parameters",
+    "objectID": "stan-users-guide/proportionality-constants.html#dropping-proportionality-constants",
+    "href": "stan-users-guide/proportionality-constants.html#dropping-proportionality-constants",
+    "title": "Proportionality Constants",
     "section": "",
-    "text": "In some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the transformed parameters block and assigning to it.\nThe following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] vector[2] y;\n  real&lt;lower=0&gt; var1;\n  real&lt;lower=0&gt; var2;\n}\ntransformed data {\n  real&lt;lower=0&gt; max_cov = sqrt(var1 * var2);\n  real&lt;upper=0&gt; min_cov = -max_cov;\n}\nparameters {\n  vector[2] mu;\n  real&lt;lower=min_cov, upper=max_cov&gt; cov;\n}\ntransformed parameters {\n  matrix[2, 2] Sigma;\n  Sigma[1, 1] = var1;\n  Sigma[1, 2] = cov;\n  Sigma[2, 1] = cov;\n  Sigma[2, 2] = var2;\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);\n}\nThe variances are defined as data in variables var1 and var2, whereas the covariance is defined as a parameter in variable cov. The \\(2 \\times 2\\) covariance matrix Sigma is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.\nThe constraint on the covariance declaration ensures that the resulting covariance matrix sigma is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.\nThe vectorization of the multivariate normal is critical for efficiency here. The transformed parameter Sigma could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.",
+    "text": "If a density \\(p(\\theta)\\) can be factored into \\(K g(\\theta)\\) where \\(K\\) are all the factors that are a not a function of \\(\\theta\\) and \\(g(\\theta)\\) are all the terms that are a function of \\(\\theta\\), then it is said that \\(g(\\theta)\\) is proportional to \\(p(\\theta)\\) up to a constant.\nThe advantage of all this is that sometimes \\(K\\) is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.\nStan takes advantage of the proportionality constant fact with the ~ syntax. Take for instance the normal data model:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nSyntactically, this is just shorthand for the equivalent model that replaces the ~ syntax with a target += statement and a normal_lupdf function call:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  target += normal_lupdf(x | mu, sigma);\n}\nThe function normal_lupdf is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\nNow because the density here is only a function of \\(x\\), the additive terms in the log density that are not a function of \\(x\\) can be dropped. In this case it is enough to know only the quadratic term:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Programming Techniques",
+      "Proportionality Constants"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html#sliced-missing-data",
-    "href": "stan-users-guide/missing-data.html#sliced-missing-data",
-    "title": "Missing Data and Partially Known Parameters",
+    "objectID": "stan-users-guide/proportionality-constants.html#keeping-proportionality-constants",
+    "href": "stan-users-guide/proportionality-constants.html#keeping-proportionality-constants",
+    "title": "Proportionality Constants",
     "section": "",
-    "text": "If the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_obs;\n  array[N_mis] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_mis;\n  array[N_obs] real y_obs;\n}\ntransformed data {\n  int&lt;lower=0&gt; N = N_obs + N_mis;\n}\nparameters {\n  array[N_mis] real y_mis;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  array[N] real y;\n  y[ii_obs] = y_obs;\n  y[ii_mis] = y_mis;\n}\nmodel {\n  sigma ~ gamma(1, 1);\n  y[1] ~ normal(0, 100);\n  y[2:N] ~ normal(y[1:(N - 1)], sigma);\n}\nThe index arrays ii_obs and ii_mis contain the indexes into the final array y of the observed data (coded as a data vector y_obs) and the missing data (coded as a parameter vector y_mis). See the time series chapter for further discussion of time-series model and specifically the autoregression section for an explanation of the vectorization for y as well as an explanation of how to convert this example to a full AR(1) model. To ensure y[1] has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.\nAnother potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in\nx[N_obs_2, 2] = x_obs_2;\nx[N_mis_2, 2] = x_mis_2;\nwhere the relevant variables are all hard coded with index 2 because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the ragged data structures section for a general discussion of coding ragged data structures in Stan).",
+    "text": "In the case that the proportionality constants were needed for a normal log density the function normal_lpdf can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the ~ or target += normal_lupdf syntaxes if it is absolutely clear that the proportionality constants are not necessary.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Programming Techniques",
+      "Proportionality Constants"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html#loading-matrix-for-factor-analysis",
-    "href": "stan-users-guide/missing-data.html#loading-matrix-for-factor-analysis",
-    "title": "Missing Data and Partially Known Parameters",
+    "objectID": "stan-users-guide/proportionality-constants.html#user-defined-distributions",
+    "href": "stan-users-guide/proportionality-constants.html#user-defined-distributions",
+    "title": "Proportionality Constants",
     "section": "",
-    "text": "Rick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis (Aguilar and West 2000). This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.\ndata {\n  int&lt;lower=2&gt; K;\n}\ntransformed data {\n  int&lt;lower=1&gt; K_choose_2;\n  K_choose_2 = (K * (K - 1)) / 2;\n}\nparameters {\n  vector[K_choose_2] L_lower;\n}\ntransformed parameters {\n  cholesky_factor_cov[K] L;\n  for (k in 1:K) {\n    L[k, k] = 1;\n  }\n  {\n    int i;\n    for (m in 2:K) {\n      for (n in 1:(m - 1)) {\n        L[m, n] = L_lower[i];\n        L[n, m] = 0;\n        i += 1;\n      }\n    }\n  }\n}\nIt is most convenient to place a prior directly on L_lower. An alternative would be a prior for the full Cholesky factor L, because the transform from L_lower to L is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix L   * L', because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.",
+    "text": "When a custom _lpdf or _lpmf function is defined, the compiler will automatically make available a _lupdf or _lupmf version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.\nThe difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.\nThe following code demonstrates the different behaviors:\nfunctions {\n  real custom1_lpdf(x) {\n    return normal_lupdf(x | 0.0, 1.0);\n  }\n  real custom2_lpdf(x) {\n    return normal_lpdf(x | 0.0, 1.0);\n  }\n}\nparameters {\n  real mu;\n}\nmodel {\n  mu ~ custom1(); // Normalization constants dropped\n  target += custom1_lupdf(mu); // Normalization constants dropped\n  target += custom1_lpdf(mu);  // Normalization constants kept\n\n  mu ~ custom2();  // Normalization constants kept\n  target += custom2_lupdf(mu);  // Normalization constants kept\n  target += custom2_lpdf(mu);  // Normalization constants kept\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Programming Techniques",
+      "Proportionality Constants"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html#missing-multivariate-data",
-    "href": "stan-users-guide/missing-data.html#missing-multivariate-data",
-    "title": "Missing Data and Partially Known Parameters",
+    "objectID": "stan-users-guide/proportionality-constants.html#limitations-on-using-_lupdf-and-_lupmf-functions",
+    "href": "stan-users-guide/proportionality-constants.html#limitations-on-using-_lupdf-and-_lupmf-functions",
+    "title": "Proportionality Constants",
     "section": "",
-    "text": "It’s often the case that one or more components of a multivariate outcome are missing.2\nAs an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations y and a boolean array y_observed to indicate which values were observed (others can have dummy values in the input).\narray[N] vector[2] y;\narray[N, 2] int&lt;lower=0, upper=1&gt; y_observed;\nIf both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    y[n] ~ multi_normal(mu, Sigma);\n  } else if (y_observed[n, 1]) {\n    y[n, 1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\n  } else if (y_observed[n, 2]) {\n    y[n, 2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\n  }\n}\nIt’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:\ntransformed data {\n  array[observed_12(y_observed)] int ns12;\n  array[observed_1(y_observed)] int ns1;\n  array[observed_2(y_observed)] int ns2;\n}\nYou will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.\nint n12 = 1;\nint n1 = 1;\nint n2 = 1;\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    ns12[n12] = n;\n    n12 += 1;\n  } else if (y_observed[n, 1]) {\n    ns1[n1] = n;\n    n1 += 1;\n  } else if (y_observed[n, 2]) {\n    ns2[n2] = n;\n    n2 += 1;\n  }\n}\nThen, in the model block, everything is vectorizable using those indexes constructed once in transformed data:\ny[ns12] ~ multi_normal(mu, Sigma);\ny[ns1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\ny[ns2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\nThe result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.\nThis approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).",
+    "text": "To avoid ambiguities in how the normalization constants work, functions ending in _lupdf and _lupmf can only be used in the model block or user-defined probability functions (functions ending in _lpdf or _lpmf).",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Programming Techniques",
+      "Proportionality Constants"
     ]
   },
   {
-    "objectID": "stan-users-guide/missing-data.html#footnotes",
-    "href": "stan-users-guide/missing-data.html#footnotes",
-    "title": "Missing Data and Partially Known Parameters",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nA more meaningful estimation example would involve a regression of the observed and missing observations using predictors that were known for each and specified in the data block.↩︎\nThis is not the same as missing components of a multivariate predictor in a regression problem; in that case, you will need to represent the missing data as a parameter and impute missing values in order to feed them into the regression.↩︎",
+    "objectID": "stan-users-guide/poststratification.html",
+    "href": "stan-users-guide/poststratification.html",
+    "title": "Poststratification",
+    "section": "",
+    "text": "Stratification is a technique developed for survey sampling in which a population is partitioned into subgroups (i.e., stratified) and each group (i.e., stratum) is sampled independently. If the subgroups are more homogeneous (i.e., lower variance) than the population as a whole, this can reduce variance in the estimate of a quantity of interest at the population level.\nPoststratification is a technique for adjusting a non-representative sample (i.e., a convenience sample or other observational data) for which there are demographic predictors characterizing the strata. It is carried out after a model is fit to the observed data, hence the name poststratification (Little 1993). Poststratification can be fruitfully combined with regression modeling (or more general parametric modeling), which provides estimates based on combinations of predictors (or general parameters) rather than raw counts in each stratum. Multilevel modeling is useful in determining how much partial pooling to apply in the regressions, leading to the popularity of the combination of multilevel regression and poststratification (MRP) (Park, Gelman, and Bafumi 2004).\n\n\n\n\nStratification and poststratification can be applied to many applications beyond survey sampling (Kennedy and Gelman 2019). For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid (Paustian et al. 1997). Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both (Field et al. 1998).\n\n\n\nSuppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.\nThe proportion of support in the poll among students in each group provides a simple maximum likelihood estimate \\(\\theta^* = (0.25, 0.5,\n0.8)\\) of support in each group for a simple Bernoulli model where student \\(n\\)’s vote is modeled as \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:3\\) is the subgroup to which the \\(n\\)-th student belongs.\nAn estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting \\(N = (20\\,000,\\, 5\\,000,\\, 2\\,000)\\) be the subgroup sizes, the poststratified estimate of support in the population \\(\\phi^*\\) is estimated by \\[\n\\phi^*\n= \\frac{\\displaystyle \\sum_{j = 1}^3 \\theta_j^* \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^3 N_j}.\n\\] Plugging in our estimates and population counts yields \\[\\begin{eqnarray*}\n\\phi*\n& = & \\frac{0.25 \\cdot 20\\,000 + 0.4 \\cdot 5\\,000 + 0.8 \\cdot 2\\,000}\n           {20\\,000 + 5\\,000 + 2\\,000}\n\\\\[4pt] & = & \\frac{8\\,600}{27\\,000}\n\\\\[4pt] & \\approx & 0.32.\n\\end{eqnarray*}\\]\n\n\n\n\nConsidering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:J\\) is the group to which item \\(n\\) belongs and \\(\\theta_j\\) is the proportion of support in group \\(j\\).\nThis can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting \\(A_j\\) be the number of respondents in group \\(j\\) and \\(a_j\\) be the number of positive responses in group \\(j\\), the data model may be reduced to the form \\[\na_j \\sim \\textrm{binomial}(A_j, \\theta_j).\n\\] A simple uniform prior on the proportion of support in each group completes the model, \\[\n\\theta_j \\sim \\textrm{beta}(1, 1).\n\\] A more informative prior could be used if there is prior information available about support among the student body.\nUsing sampling, draws \\(\\theta^{(m)} \\sim p(\\theta \\mid y)\\) from the posterior may be combined with the population sizes \\(N\\) to estimate \\(\\phi\\), the proportion of support in the population, \\[\n\\phi^{(m)}\n= \\frac{\\displaystyle \\sum_{j = 1}^J \\theta_j^{(m)} \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^J N_j}.\n\\] The posterior draws for \\(\\phi^{(m)}\\) characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the \\(\\phi^{(m)}\\)), posterior intervals (quantiles of the \\(\\phi^{(m)}\\)), or to plot a histogram.\n\n\n\nThe maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where \\(A_j\\) is the number of voters from group \\(j\\), \\(a_j\\) is the number of positive responses from group \\(j\\), and \\(N_j\\) is the size of group \\(j\\) in the population.\ndata {\n  int&lt;lower=1&gt; J;\n  array[J] int&lt;lower=0&gt; A; \n  array[J] int&lt;lower=0&gt; a;\n  vector&lt;lower=0&gt;[J] N;\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[J] theta;\n}\nmodel {\n  a ~ binomial(A, theta);\n}\ngenerated quantities {t\n  real&lt;lower=0, upper=1&gt; phi = dot(N, theta) / sum(N);\n}\nThe binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the \\(J\\) terms. The prior is implicitly uniform on \\((0, 1),\\) the support of \\(\\theta.\\) The summation is computed using a dot product and the sum function, which is why N was declared as a vector rather than as an array of integers.\n\n\n\nIn applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.\nThe problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to \\(5 \\cdot 2 \\cdot 5 \\cdot\n50 = 2000\\) strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter \\(\\theta_j\\) for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.\nA practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only \\(4 + 2 + 5 + 50 =\n61\\) regression coefficients to estimate. Now suppose that item \\(n\\) has demographic features \\(\\textrm{age}_n \\in 1:5\\), \\(\\textrm{sex}_n \\in 1:2\\), \\(\\textrm{income}_n \\in 1:5,\\) and \\(\\textrm{state}_n \\in 1:50\\). A logistic regression may be formulated as \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)),\n\\] where \\(\\textrm{age}[n]\\) is the age of the \\(n\\)-th respondent, \\(\\textrm{sex}[n]\\) is their sex, \\(\\textrm{income}[n]\\) their income and \\(\\textrm{state}[n]\\) their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.\nTo poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as \\[\n\\sum_{i = 1}^5 \\sum_{j = 1}^2 \\sum_{k = 1}^5 \\sum_{m = 1}^{50}\n\\textrm{logit}^{-1}(\\alpha + \\beta_i + \\gamma_j + \\delta_k + \\eta_m)\n\\cdot \\textrm{pop}_{i, j, k, m},\n\\] where \\(\\textrm{pop}_{i, j, k, m}\\) is the size of the subpopulation with age \\(i\\), sex \\(j\\), income level \\(k\\), and state of residence \\(m\\).\nAs formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes \\(y\\) given predictors \\(x\\).\n\n\n\nWith large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.\nThe model introduced in the previous section had the data model \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)).\n\\] The overall intercept can be given a broad fixed prior, \\[\n\\alpha \\sim \\textrm{normal}(0, 5).\n\\] The other regression parameters can be given hierarchical priors, \\[\\begin{eqnarray*}\n\\beta_{1:4} & \\sim & \\textrm{normal}(0, \\sigma^{\\beta})\n\\\\[2pt]\n\\gamma_{1:2} & \\sim & \\textrm{normal}(0, \\sigma^{\\gamma})\n\\\\[2pt]\n\\delta_{1:5} & \\sim & \\textrm{normal}(0, \\sigma^{\\delta})\n\\\\[2pt]\n\\epsilon_{1:50} & \\sim & \\textrm{normal}(0, \\sigma^{\\epsilon}).\n\\end{eqnarray*}\\]\nThe hyperparameters for scale of variation within a group can be given simple standard hyperpriors, \\[\n\\sigma^{\\beta}, \\sigma^{\\gamma}, \\sigma^{\\delta}, \\sigma^{\\epsilon}\n\\sim \\textrm{normal}(0, 1).\n\\] The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).\n\n\nThe multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.\nThis is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, \\(\\gamma_2 =\n-\\gamma_1.\\)\nTo implement sum-to-zero constraints,\nparameters {\n  vector[K - 1] alpha_raw;\n// ...\n}\ntransformed parameters {\n  vector&lt;multiplier=sigma_alpha&gt;[K] alpha\n    = append_row(alpha_raw, -sum(alpha_raw));\n// ...    \n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n}\nThis prior is hard to interpret in that there are K normal distributions, but only K - 1 free parameters. An alternative is to put the prior only on alpha_raw, but that is also difficult to interpret.\nSoft constraints can be more computationally tractable. They are also simpler to implement.\nparameters {\n  vector&lt;multiplier=alpha&gt;[K] alpha;\n// ...\n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n  sum(alpha) ~ normal(0, 0.001);\n}\nThis leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.\nNote that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept \\(\\alpha.\\) Even with a wide prior on the intercept, the hyperprior on \\(\\sigma^{\\gamma}\\) may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.\nNone of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.\n\n\n\n\nMultilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the multiplier transform is used on each of the parameters. The combination of\nvector&lt;multiplier=s&gt;[K] a;\n// ...\na ~ normal(0, s);\nimplements a non-centered parameterization for a; a centered parameterization would drop the multiplier specification. The prior scale s is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the multiplier.\nHere is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable P.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; age;\n  array[N] int&lt;lower=1, upper=5&gt; income;\n  array[N] int&lt;lower=1, upper=50&gt; state;\n  array[N] int&lt;lower=0&gt; y;\n  array[4, 5, 50] int&lt;lower=0&gt; P;\n}\nparameters {\n  real alpha;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;multiplier=sigma_beta&gt;[4] beta;\n  real&lt;lower=0&gt; sigma_gamma;\n  vector&lt;multiplier=sigma_gamma&gt;[5] gamma;\n  real&lt;lower=0&gt; sigma_delta;\n  vector&lt;multiplier=sigma_delta&gt;[50] delta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, sigma_beta);\n  gamma ~ normal(0, sigma_gamma);\n  delta ~ normal(0, sigma_delta);\n  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(0, 1);\n}\ngenerated quantities {\n  real expect_pos = 0;\n  int total = 0;\n  for (b in 1:4) {\n    for (c in 1:5) {\n      for (d in 1:50) {\n        total += P[b, c, d];\n        expect_pos\n          += P[b, c, d]\n             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);\n      }\n    }\n  }\n  real&lt;lower=0, upper=1&gt; phi = expect_pos / total;\n}\nUnlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).\n\n\nIn some cases, it can be more efficient to break the data down by group. Suppose there are \\(4 \\times 5 \\times 2 \\times 50 = 2000\\) groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group\nint&lt;lower=0&gt; G;\narray[G] int&lt;lower=1, upper=4&gt; age;\narray[G] int&lt;lower=1, upper=5&gt; income;\narray[G] int&lt;lower=1, upper=50&gt; state;\nThen the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.\narray[G] int&lt;lower=0&gt; pos_votes;\narray[G] int&lt;lower=0&gt; total_votes;\nFinally, the data model is converted to binomial.\npos_votes ~ binomial_logit(total_votes,\n                           alpha + beta[age] + ...);\nThe predictors look the same because of the way the age and other data items are coded.\n\n\n\nIn this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the N data points.\n  array[N] int&lt;lower=1, upper=2&gt; sex;\nThen add a single regression coefficient as a parameter,\n  real epsilon;\nIn the log odds calculation, introduce a new term\n[epsilon, -epsilon][sex]';\nThat is, the data model will now look like\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]\n                      + [epsilon, -epsilon][sex]');\nFor data point n, the expression [epsilon, -epsilon][sex] takes on value [epsilon, -epsilon][sex][n], which with Stan’s multi-indexing reduces to [epsilon, -epsilon][sex[n]]. This term evaluates to epsilon if sex[n] is 1 and to -epsilon if sex[n] is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The ' at the end transposes [epsilon, -epsilon][sex] which is a row_vector into a vector that can be added to the other variables.\nFinally, a prior is needed for the coefficient in the model block,\nepsilon ~ normal(0, 2);\nAs with other priors in multilevel models, the posterior for epsilon should be investigated to make sure it is not unrealistically wide.\n\n\n\n\nIf there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable\narray[50] real&lt;lower=0&gt; income;\nthen a regression coefficient psi can be added for the effect of average state income,\nreal psi;\nwith a fixed prior,\npsi ~ normal(0, 2);\nThis prior assumes the income predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,\ny ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]\n                    + income[state] * psi);\nAnd finally, the formula in the generated quantities block is also updated,\nexpect_pos\n  += P[b, c, d]\n     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]\n             + income[d] * psi);\nHere d is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Missing Data and Partially Known Parameters"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html",
-    "href": "stan-users-guide/matrices-arrays.html",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#some-examples",
+    "href": "stan-users-guide/poststratification.html#some-examples",
+    "title": "Poststratification",
     "section": "",
-    "text": "This chapter provides pointers as to how to choose among the various container types (matrix, vector, array, and tuple) provided by Stan.\n\n\nStan provides three basic scalar types, int, real, and complex, as well as three basic linear algebra types, vector, row_vector, and matrix. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).\nThis leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.\narray[N] real a;\nvector[N] a;\nrow_vector[N] a;\nThese distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.\nThe following code fragment shows all four ways to declare a two-dimensional container of size \\(M \\times N\\).\narray[M, N] real b;          // b[m] : array[] real     (efficient)\narray[M] vector[N] b;        // b[m] : vector     (efficient)\narray[M] row_vector[N] b;    // b[m] : row_vector (efficient)\nmatrix[M, N] b;              // b[m] : row_vector (inefficient)\nThe main differences among these choices involve efficiency for various purposes and the type of b[m], which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.\nThe inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).\n\n\n\nArrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single complex typed entry, all the entries are complex).\nWith arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a complex_vector[3] holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.\nTuples provide a way to represent a sequence of values of heterogeneous types. For example, tuple(int, real) is the type of a pair consisting of an integer and a real number and tuple(array[5] int, vector[6]) is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.\n\n\nTuples are declared using the keyword tuple followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.\ntuple(int, vector[3]) ny = (5, [3, 2.9, 1.8]');\nThe elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using\nint n = ny.1;\nvector[3] y = ny.2;\nWe can also assign into the elements of a tuple.\ntuple(int, vector[3], complex) abc;\nabc.1 = 5;\nabc.2[1] = 3;\nabc.2[2] = 2.9;\nabc.2[3] = 1.4798;\nabc.3 = 2 + 1.9j;\nAs the cascaded indexing example shows, the result of abc.1 is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., abc.2[1] pulls out the first element of the vector value of the second element of the tuple.)\nThere are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.\nint a = 5;\nmatrix[10, 100] b = ...;\ntuple(int, matrix[10, 100]) ab = (a, b);  // COPIES b\nb[1,1] = 10.3;  // does NOT change ab\n\n\n\nTuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, array[N] tuple(int, real, vector[5]) is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as tuple(array[N] int, array[N] real, array[N] vector[5]).\nThe second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix \\(A = Q \\cdot R\\), where \\(Q\\) is orthonormal and \\(R\\) is upper triangular. In the past, this required two function calls.\nmatrix[M, N] A = ...;\nmatrix[M, M] Q = qr_Q(A);\nmatrix[M, N] R = qr_R(A);\nWith tuples, this can be simplified to the following,\ntuple(matrix[M, M], matrix[M, N]) QR = qr(A);\nwith QR.1 being Q and QR.2 giving R.\n\n\n\n\nStan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.\nWhen an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.\n\n\n\nThe underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.\nArrays, on the other hand, are implemented as instances of the C++\nstd::vector class (not to be confused with Eigen’s Eigen::Vector class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.\n\n\nIn Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.\nFirst, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.\nSecond, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).\nThird, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.\nmatrix[M, N] a;\n//...\nfor (n in 1:N) {\n  for (m in 1:M) {\n    // ... do something with a[m, n] ...\n  }\n}\nArrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  for (n in 1:N) {\n    // ... do something with a[m, n] ...\n  }\n}\nThe first use of a[m ,n] should bring a[m] into memory. Overall, traversing matrices is more efficient than traversing arrays.\nThis is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is\narray[I, J] matrix[M, N] b;\n// ...\nfor (i in 1:I) {\n  for (j in 1:J) {\n    for (n in 1:N) {\n      for (m in 1:M) {\n        // ... do something with b[i, j, m, n] ...\n      }\n    }\n  }\n}\nIf a is a matrix, the notation a[m] picks out row m of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this\narray[M] row_vector[N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nis much more efficient than the pure matrix version\nmatrix[M, N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nSimilarly, indexing an array of column vectors is more efficient than using the col function to pick out a column of a matrix.\nIn contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this\nmatrix[N, k] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\ny_hat = x * beta;\nthan it is to do this\narray[N] row_vector[K] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\nfor (n in 1:N) {\n  y_hat[n] = x[n] * beta;\n}\n\n\n\nFor use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The Eigen::Vector template specialization and the std::vector template class are implemented similarly as containers of double values (the type real in Stan). Only arrays in Stan are allowed to store integer values.\n\n\n\n\nThe key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.\n\n\nCPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is much slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.\n\n\n\nMatrices are stored internally in column-major order. That is, an \\(M\n\\times N\\) matrix stores its elements in the order \\[\n(1,1), (2, 1), \\dotsc, (M, 1), (1, 2), \\dotsc, (M, 2), \\dotsc, (1, N),\n\\dotsc, (M, N).\n\\]\nThis means that it’s much more efficient to write loops over matrices column by column, as in the following example.\nmatrix[M, N] a;\n// ...\nfor (n in 1:N) {\n  for (m in 1:M) {\n     // ... do something with a[m, n] ...\n  }\n}\nIt also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.\narray[M] row_vector[N] a;\n// ...\nfor (m in 1:M) {\n  // ... do something with row vector a[m] ...\n}\nEven if what is done involves a function call, the row vector a[m] will not have to be copied.\n\n\n\nArrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  // ... do something with a[m] ...\n}\nA difference with matrices is that the entries a[m] in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”\n\n\n\n\nThere is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the multi-indexing chapter for some reshaping operations involving multiple indexing and range indexing.\n\n\n\nStan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to u and the compound slicing assignment to x, where u and x start with the same values.\ntransformed data {\n  vector[4] x = [ 1, 2, 3, 4 ]';\n  vector[4] u = [ 1, 2, 3, 4 ]';\n\n  for (t in 2:4) {\n    u[t] = u[t - 1] * 3;\n  }\n\n  x[2:4] = x[1:3] * 3;\n\n  print(\"u = \", u);\n  print(\"x = \", x);\n}\nThe output it produces is,\nu = [1, 3, 9, 27]\nx = [1, 3, 6, 9]\nIn the loop version assigning to u, the values are updated before being used to define subsequent values; in the sliced expression assigning to x, the entire right-hand side is evaluated before assigning to the left-hand side.",
+    "text": "Stratification and poststratification can be applied to many applications beyond survey sampling (Kennedy and Gelman 2019). For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid (Paustian et al. 1997). Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both (Field et al. 1998).\n\n\n\nSuppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.\nThe proportion of support in the poll among students in each group provides a simple maximum likelihood estimate \\(\\theta^* = (0.25, 0.5,\n0.8)\\) of support in each group for a simple Bernoulli model where student \\(n\\)’s vote is modeled as \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:3\\) is the subgroup to which the \\(n\\)-th student belongs.\nAn estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting \\(N = (20\\,000,\\, 5\\,000,\\, 2\\,000)\\) be the subgroup sizes, the poststratified estimate of support in the population \\(\\phi^*\\) is estimated by \\[\n\\phi^*\n= \\frac{\\displaystyle \\sum_{j = 1}^3 \\theta_j^* \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^3 N_j}.\n\\] Plugging in our estimates and population counts yields \\[\\begin{eqnarray*}\n\\phi*\n& = & \\frac{0.25 \\cdot 20\\,000 + 0.4 \\cdot 5\\,000 + 0.8 \\cdot 2\\,000}\n           {20\\,000 + 5\\,000 + 2\\,000}\n\\\\[4pt] & = & \\frac{8\\,600}{27\\,000}\n\\\\[4pt] & \\approx & 0.32.\n\\end{eqnarray*}\\]",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#basic-motivation",
-    "href": "stan-users-guide/matrices-arrays.html#basic-motivation",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#bayesian-poststratification",
+    "href": "stan-users-guide/poststratification.html#bayesian-poststratification",
+    "title": "Poststratification",
     "section": "",
-    "text": "Stan provides three basic scalar types, int, real, and complex, as well as three basic linear algebra types, vector, row_vector, and matrix. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).\nThis leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.\narray[N] real a;\nvector[N] a;\nrow_vector[N] a;\nThese distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.\nThe following code fragment shows all four ways to declare a two-dimensional container of size \\(M \\times N\\).\narray[M, N] real b;          // b[m] : array[] real     (efficient)\narray[M] vector[N] b;        // b[m] : vector     (efficient)\narray[M] row_vector[N] b;    // b[m] : row_vector (efficient)\nmatrix[M, N] b;              // b[m] : row_vector (inefficient)\nThe main differences among these choices involve efficiency for various purposes and the type of b[m], which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.\nThe inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).",
+    "text": "Considering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:J\\) is the group to which item \\(n\\) belongs and \\(\\theta_j\\) is the proportion of support in group \\(j\\).\nThis can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting \\(A_j\\) be the number of respondents in group \\(j\\) and \\(a_j\\) be the number of positive responses in group \\(j\\), the data model may be reduced to the form \\[\na_j \\sim \\textrm{binomial}(A_j, \\theta_j).\n\\] A simple uniform prior on the proportion of support in each group completes the model, \\[\n\\theta_j \\sim \\textrm{beta}(1, 1).\n\\] A more informative prior could be used if there is prior information available about support among the student body.\nUsing sampling, draws \\(\\theta^{(m)} \\sim p(\\theta \\mid y)\\) from the posterior may be combined with the population sizes \\(N\\) to estimate \\(\\phi\\), the proportion of support in the population, \\[\n\\phi^{(m)}\n= \\frac{\\displaystyle \\sum_{j = 1}^J \\theta_j^{(m)} \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^J N_j}.\n\\] The posterior draws for \\(\\phi^{(m)}\\) characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the \\(\\phi^{(m)}\\)), posterior intervals (quantiles of the \\(\\phi^{(m)}\\)), or to plot a histogram.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#tuple-types",
-    "href": "stan-users-guide/matrices-arrays.html#tuple-types",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#poststratification-in-stan",
+    "href": "stan-users-guide/poststratification.html#poststratification-in-stan",
+    "title": "Poststratification",
     "section": "",
-    "text": "Arrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single complex typed entry, all the entries are complex).\nWith arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a complex_vector[3] holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.\nTuples provide a way to represent a sequence of values of heterogeneous types. For example, tuple(int, real) is the type of a pair consisting of an integer and a real number and tuple(array[5] int, vector[6]) is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.\n\n\nTuples are declared using the keyword tuple followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.\ntuple(int, vector[3]) ny = (5, [3, 2.9, 1.8]');\nThe elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using\nint n = ny.1;\nvector[3] y = ny.2;\nWe can also assign into the elements of a tuple.\ntuple(int, vector[3], complex) abc;\nabc.1 = 5;\nabc.2[1] = 3;\nabc.2[2] = 2.9;\nabc.2[3] = 1.4798;\nabc.3 = 2 + 1.9j;\nAs the cascaded indexing example shows, the result of abc.1 is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., abc.2[1] pulls out the first element of the vector value of the second element of the tuple.)\nThere are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.\nint a = 5;\nmatrix[10, 100] b = ...;\ntuple(int, matrix[10, 100]) ab = (a, b);  // COPIES b\nb[1,1] = 10.3;  // does NOT change ab\n\n\n\nTuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, array[N] tuple(int, real, vector[5]) is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as tuple(array[N] int, array[N] real, array[N] vector[5]).\nThe second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix \\(A = Q \\cdot R\\), where \\(Q\\) is orthonormal and \\(R\\) is upper triangular. In the past, this required two function calls.\nmatrix[M, N] A = ...;\nmatrix[M, M] Q = qr_Q(A);\nmatrix[M, N] R = qr_R(A);\nWith tuples, this can be simplified to the following,\ntuple(matrix[M, M], matrix[M, N]) QR = qr(A);\nwith QR.1 being Q and QR.2 giving R.",
+    "text": "The maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where \\(A_j\\) is the number of voters from group \\(j\\), \\(a_j\\) is the number of positive responses from group \\(j\\), and \\(N_j\\) is the size of group \\(j\\) in the population.\ndata {\n  int&lt;lower=1&gt; J;\n  array[J] int&lt;lower=0&gt; A; \n  array[J] int&lt;lower=0&gt; a;\n  vector&lt;lower=0&gt;[J] N;\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[J] theta;\n}\nmodel {\n  a ~ binomial(A, theta);\n}\ngenerated quantities {t\n  real&lt;lower=0, upper=1&gt; phi = dot(N, theta) / sum(N);\n}\nThe binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the \\(J\\) terms. The prior is implicitly uniform on \\((0, 1),\\) the support of \\(\\theta.\\) The summation is computed using a dot product and the sum function, which is why N was declared as a vector rather than as an array of integers.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#fixed-sizes-and-indexing-out-of-bounds",
-    "href": "stan-users-guide/matrices-arrays.html#fixed-sizes-and-indexing-out-of-bounds",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#regression-and-poststratification",
+    "href": "stan-users-guide/poststratification.html#regression-and-poststratification",
+    "title": "Poststratification",
     "section": "",
-    "text": "Stan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.\nWhen an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.",
+    "text": "In applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.\nThe problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to \\(5 \\cdot 2 \\cdot 5 \\cdot\n50 = 2000\\) strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter \\(\\theta_j\\) for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.\nA practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only \\(4 + 2 + 5 + 50 =\n61\\) regression coefficients to estimate. Now suppose that item \\(n\\) has demographic features \\(\\textrm{age}_n \\in 1:5\\), \\(\\textrm{sex}_n \\in 1:2\\), \\(\\textrm{income}_n \\in 1:5,\\) and \\(\\textrm{state}_n \\in 1:50\\). A logistic regression may be formulated as \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)),\n\\] where \\(\\textrm{age}[n]\\) is the age of the \\(n\\)-th respondent, \\(\\textrm{sex}[n]\\) is their sex, \\(\\textrm{income}[n]\\) their income and \\(\\textrm{state}[n]\\) their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.\nTo poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as \\[\n\\sum_{i = 1}^5 \\sum_{j = 1}^2 \\sum_{k = 1}^5 \\sum_{m = 1}^{50}\n\\textrm{logit}^{-1}(\\alpha + \\beta_i + \\gamma_j + \\delta_k + \\eta_m)\n\\cdot \\textrm{pop}_{i, j, k, m},\n\\] where \\(\\textrm{pop}_{i, j, k, m}\\) is the size of the subpopulation with age \\(i\\), sex \\(j\\), income level \\(k\\), and state of residence \\(m\\).\nAs formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes \\(y\\) given predictors \\(x\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#indexing-efficiency.section",
-    "href": "stan-users-guide/matrices-arrays.html#indexing-efficiency.section",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#multilevel-regression-and-poststratification",
+    "href": "stan-users-guide/poststratification.html#multilevel-regression-and-poststratification",
+    "title": "Poststratification",
     "section": "",
-    "text": "The underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.\nArrays, on the other hand, are implemented as instances of the C++\nstd::vector class (not to be confused with Eigen’s Eigen::Vector class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.\n\n\nIn Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.\nFirst, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.\nSecond, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).\nThird, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.\nmatrix[M, N] a;\n//...\nfor (n in 1:N) {\n  for (m in 1:M) {\n    // ... do something with a[m, n] ...\n  }\n}\nArrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  for (n in 1:N) {\n    // ... do something with a[m, n] ...\n  }\n}\nThe first use of a[m ,n] should bring a[m] into memory. Overall, traversing matrices is more efficient than traversing arrays.\nThis is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is\narray[I, J] matrix[M, N] b;\n// ...\nfor (i in 1:I) {\n  for (j in 1:J) {\n    for (n in 1:N) {\n      for (m in 1:M) {\n        // ... do something with b[i, j, m, n] ...\n      }\n    }\n  }\n}\nIf a is a matrix, the notation a[m] picks out row m of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this\narray[M] row_vector[N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nis much more efficient than the pure matrix version\nmatrix[M, N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nSimilarly, indexing an array of column vectors is more efficient than using the col function to pick out a column of a matrix.\nIn contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this\nmatrix[N, k] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\ny_hat = x * beta;\nthan it is to do this\narray[N] row_vector[K] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\nfor (n in 1:N) {\n  y_hat[n] = x[n] * beta;\n}\n\n\n\nFor use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The Eigen::Vector template specialization and the std::vector template class are implemented similarly as containers of double values (the type real in Stan). Only arrays in Stan are allowed to store integer values.",
+    "text": "With large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.\nThe model introduced in the previous section had the data model \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)).\n\\] The overall intercept can be given a broad fixed prior, \\[\n\\alpha \\sim \\textrm{normal}(0, 5).\n\\] The other regression parameters can be given hierarchical priors, \\[\\begin{eqnarray*}\n\\beta_{1:4} & \\sim & \\textrm{normal}(0, \\sigma^{\\beta})\n\\\\[2pt]\n\\gamma_{1:2} & \\sim & \\textrm{normal}(0, \\sigma^{\\gamma})\n\\\\[2pt]\n\\delta_{1:5} & \\sim & \\textrm{normal}(0, \\sigma^{\\delta})\n\\\\[2pt]\n\\epsilon_{1:50} & \\sim & \\textrm{normal}(0, \\sigma^{\\epsilon}).\n\\end{eqnarray*}\\]\nThe hyperparameters for scale of variation within a group can be given simple standard hyperpriors, \\[\n\\sigma^{\\beta}, \\sigma^{\\gamma}, \\sigma^{\\delta}, \\sigma^{\\epsilon}\n\\sim \\textrm{normal}(0, 1).\n\\] The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).\n\n\nThe multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.\nThis is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, \\(\\gamma_2 =\n-\\gamma_1.\\)\nTo implement sum-to-zero constraints,\nparameters {\n  vector[K - 1] alpha_raw;\n// ...\n}\ntransformed parameters {\n  vector&lt;multiplier=sigma_alpha&gt;[K] alpha\n    = append_row(alpha_raw, -sum(alpha_raw));\n// ...    \n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n}\nThis prior is hard to interpret in that there are K normal distributions, but only K - 1 free parameters. An alternative is to put the prior only on alpha_raw, but that is also difficult to interpret.\nSoft constraints can be more computationally tractable. They are also simpler to implement.\nparameters {\n  vector&lt;multiplier=alpha&gt;[K] alpha;\n// ...\n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n  sum(alpha) ~ normal(0, 0.001);\n}\nThis leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.\nNote that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept \\(\\alpha.\\) Even with a wide prior on the intercept, the hyperprior on \\(\\sigma^{\\gamma}\\) may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.\nNone of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#memory-locality",
-    "href": "stan-users-guide/matrices-arrays.html#memory-locality",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#coding-mrp-in-stan",
+    "href": "stan-users-guide/poststratification.html#coding-mrp-in-stan",
+    "title": "Poststratification",
     "section": "",
-    "text": "The key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.\n\n\nCPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is much slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.\n\n\n\nMatrices are stored internally in column-major order. That is, an \\(M\n\\times N\\) matrix stores its elements in the order \\[\n(1,1), (2, 1), \\dotsc, (M, 1), (1, 2), \\dotsc, (M, 2), \\dotsc, (1, N),\n\\dotsc, (M, N).\n\\]\nThis means that it’s much more efficient to write loops over matrices column by column, as in the following example.\nmatrix[M, N] a;\n// ...\nfor (n in 1:N) {\n  for (m in 1:M) {\n     // ... do something with a[m, n] ...\n  }\n}\nIt also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.\narray[M] row_vector[N] a;\n// ...\nfor (m in 1:M) {\n  // ... do something with row vector a[m] ...\n}\nEven if what is done involves a function call, the row vector a[m] will not have to be copied.\n\n\n\nArrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  // ... do something with a[m] ...\n}\nA difference with matrices is that the entries a[m] in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”",
+    "text": "Multilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the multiplier transform is used on each of the parameters. The combination of\nvector&lt;multiplier=s&gt;[K] a;\n// ...\na ~ normal(0, s);\nimplements a non-centered parameterization for a; a centered parameterization would drop the multiplier specification. The prior scale s is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the multiplier.\nHere is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable P.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; age;\n  array[N] int&lt;lower=1, upper=5&gt; income;\n  array[N] int&lt;lower=1, upper=50&gt; state;\n  array[N] int&lt;lower=0&gt; y;\n  array[4, 5, 50] int&lt;lower=0&gt; P;\n}\nparameters {\n  real alpha;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;multiplier=sigma_beta&gt;[4] beta;\n  real&lt;lower=0&gt; sigma_gamma;\n  vector&lt;multiplier=sigma_gamma&gt;[5] gamma;\n  real&lt;lower=0&gt; sigma_delta;\n  vector&lt;multiplier=sigma_delta&gt;[50] delta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, sigma_beta);\n  gamma ~ normal(0, sigma_gamma);\n  delta ~ normal(0, sigma_delta);\n  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(0, 1);\n}\ngenerated quantities {\n  real expect_pos = 0;\n  int total = 0;\n  for (b in 1:4) {\n    for (c in 1:5) {\n      for (d in 1:50) {\n        total += P[b, c, d];\n        expect_pos\n          += P[b, c, d]\n             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);\n      }\n    }\n  }\n  real&lt;lower=0, upper=1&gt; phi = expect_pos / total;\n}\nUnlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).\n\n\nIn some cases, it can be more efficient to break the data down by group. Suppose there are \\(4 \\times 5 \\times 2 \\times 50 = 2000\\) groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group\nint&lt;lower=0&gt; G;\narray[G] int&lt;lower=1, upper=4&gt; age;\narray[G] int&lt;lower=1, upper=5&gt; income;\narray[G] int&lt;lower=1, upper=50&gt; state;\nThen the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.\narray[G] int&lt;lower=0&gt; pos_votes;\narray[G] int&lt;lower=0&gt; total_votes;\nFinally, the data model is converted to binomial.\npos_votes ~ binomial_logit(total_votes,\n                           alpha + beta[age] + ...);\nThe predictors look the same because of the way the age and other data items are coded.\n\n\n\nIn this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the N data points.\n  array[N] int&lt;lower=1, upper=2&gt; sex;\nThen add a single regression coefficient as a parameter,\n  real epsilon;\nIn the log odds calculation, introduce a new term\n[epsilon, -epsilon][sex]';\nThat is, the data model will now look like\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]\n                      + [epsilon, -epsilon][sex]');\nFor data point n, the expression [epsilon, -epsilon][sex] takes on value [epsilon, -epsilon][sex][n], which with Stan’s multi-indexing reduces to [epsilon, -epsilon][sex[n]]. This term evaluates to epsilon if sex[n] is 1 and to -epsilon if sex[n] is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The ' at the end transposes [epsilon, -epsilon][sex] which is a row_vector into a vector that can be added to the other variables.\nFinally, a prior is needed for the coefficient in the model block,\nepsilon ~ normal(0, 2);\nAs with other priors in multilevel models, the posterior for epsilon should be investigated to make sure it is not unrealistically wide.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#converting-among-matrix-vector-and-array-types",
-    "href": "stan-users-guide/matrices-arrays.html#converting-among-matrix-vector-and-array-types",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/poststratification.html#adding-group-level-predictors",
+    "href": "stan-users-guide/poststratification.html#adding-group-level-predictors",
+    "title": "Poststratification",
     "section": "",
-    "text": "There is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the multi-indexing chapter for some reshaping operations involving multiple indexing and range indexing.",
+    "text": "If there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable\narray[50] real&lt;lower=0&gt; income;\nthen a regression coefficient psi can be added for the effect of average state income,\nreal psi;\nwith a fixed prior,\npsi ~ normal(0, 2);\nThis prior assumes the income predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,\ny ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]\n                    + income[state] * psi);\nAnd finally, the formula in the generated quantities block is also updated,\nexpect_pos\n  += P[b, c, d]\n     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]\n             + income[d] * psi);\nHere d is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Poststratification"
     ]
   },
   {
-    "objectID": "stan-users-guide/matrices-arrays.html#aliasing-in-stan-containers",
-    "href": "stan-users-guide/matrices-arrays.html#aliasing-in-stan-containers",
-    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "objectID": "stan-users-guide/posterior-prediction.html",
+    "href": "stan-users-guide/posterior-prediction.html",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "Stan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to u and the compound slicing assignment to x, where u and x start with the same values.\ntransformed data {\n  vector[4] x = [ 1, 2, 3, 4 ]';\n  vector[4] u = [ 1, 2, 3, 4 ]';\n\n  for (t in 2:4) {\n    u[t] = u[t - 1] * 3;\n  }\n\n  x[2:4] = x[1:3] * 3;\n\n  print(\"u = \", u);\n  print(\"x = \", x);\n}\nThe output it produces is,\nu = [1, 3, 9, 27]\nx = [1, 3, 6, 9]\nIn the loop version assigning to u, the values are updated before being used to define subsequent values; in the sliced expression assigning to x, the entire right-hand side is evaluated before assigning to the left-hand side.",
+    "text": "The goal of inference is often posterior prediction, that is evaluating or sampling from the posterior predictive distribution \\(p(\\tilde{y} \\mid y),\\) where \\(y\\) is observed data and \\(\\tilde{y}\\) is yet to be observed data. Often there are unmodeled predictors \\(x\\) and \\(\\tilde{x}\\) for the observed data \\(y\\) and unobserved data \\(\\tilde{y}\\). With predictors, the posterior predictive density is \\(p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\\) All of these variables may represent multivariate quantities.\nThis chapter explains how to sample from the posterior predictive distribution in Stan, including applications to posterior predictive simulation and calculating event probabilities. These techniques can be coded in Stan using random number generation in the generated quantities block. Further, a technique for fitting and performing inference in two stages is presented in a section on stand-alone generated quantities in Stan\n\n\nGiven a full Bayesian model \\(p(y, \\theta)\\), the posterior predictive density for new data \\(\\tilde{y}\\) given observed data \\(y\\) is \\[\np(\\tilde{y} \\mid y)\n=\n\\int p(\\tilde{y} \\mid \\theta) \\cdot p(\\theta \\mid y)\n\\, \\textrm{d}\\theta.\n\\] The product under the integral reduces to the joint posterior density \\(p(\\tilde{y}, \\theta \\mid y),\\) so that the integral is simply marginalizing out the parameters \\(\\theta,\\) leaving the predictive density \\(p(\\tilde{y} \\mid y)\\) of future observations given past observations.\n\n\n\nThe posterior predictive density (or mass) of a prediction \\(\\tilde{y}\\) given observed data \\(y\\) can be computed using \\(M\\) Monte Carlo draws\n\\[\n\\theta^{(m)} \\sim p(\\theta \\mid y)\n\\] from the posterior as \\[\np(\\tilde{y} \\mid y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\]\nComputing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, \\(\\log\np(\\tilde{y} \\mid y)\\) may be computed using the stable log sum of exponents function as \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\\\[4pt]\n& = &\n- \\log M\n+ \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\theta^{(m)}),\n\\end{eqnarray*}\\] where \\[\n\\textrm{log-sum-exp}_{m = 1}^M v_m\n= \\log \\sum_{m = 1}^M \\exp v_m\n\\] is used to maintain arithmetic precision. See the section on log sum of exponentials for more details.\n\n\n\nGiven draws from the posterior \\(\\theta^{(m)} \\sim p(\\theta \\mid y),\\) draws from the posterior predictive \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid y)\\) can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\theta^{(m)}).\n\\]\nRandomly drawing \\(\\tilde{y}\\) from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because \\(\\theta\\) is unknown and estimated based only on a finite sample of data \\(y\\). Aleatoric uncertainty arises because even a known value of \\(\\theta\\) leads to uncertainty about new \\(\\tilde{y}\\) as described by the data model \\(p(\\tilde{y} \\mid \\theta)\\). Both forms of uncertainty show up in the factored form of the posterior predictive distribution, \\[\np(\\tilde{y} \\mid y)\n=\n\\int\n\\underbrace{p(\\tilde{y} \\mid \\theta)}_{\\begin{array}{l}\n                                         \\textrm{aleatoric}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\cdot \\underbrace{p(\\theta \\mid y)}_{\\begin{array}{l}\n                                         \\textrm{epistemic}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\, \\textrm{d}\\theta.\n\\]\n\n\n\nPosterior predictive quantities can be coded in Stan using the generated quantities block.\n\n\nFor example, consider a simple Poisson model for count data with a rate parameter \\(\\lambda &gt; 0\\) having a gamma-distributed prior, \\[\n\\lambda \\sim \\textrm{gamma}(1, 1).\n\\] The \\(N\\) observations \\(y_1, \\ldots, y_N\\) are modeled as Poisson distributed, \\[\ny_n \\sim \\textrm{poisson}(\\lambda).\n\\]\n\n\n\nThe following Stan program defines a variable for \\(\\tilde{y}\\) by random number generation in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  lambda ~ gamma(1, 1);\n  y ~ poisson(lambda);\n}\ngenerated quantities {\n  int&lt;lower=0&gt; y_tilde = poisson_rng(lambda);\n}\nThe random draw from the data model for \\(\\tilde{y}\\) is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new \\(\\tilde{y}^{(m)} \\sim p(y \\mid\n\\lambda^{(m)})\\) for each posterior draw \\(\\lambda^{(m)} \\sim p(\\theta\n\\mid y).\\)\nThe posterior draws \\(\\tilde{y}^{(m)}\\) may be used to estimate the expected value of \\(\\tilde{y}\\) or any of its quantiles or posterior intervals, as well as event probabilities involving \\(\\tilde{y}\\). In general, \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\) may be evaluated as \\[\n\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\n\\approx \\frac{1}{M} \\sum_{m=1}^M f(\\tilde{y}^{(m)}, \\theta^{(m)}),\n\\] which is just the posterior mean of \\(f(\\tilde{y}, \\theta).\\) This quantity is computed by Stan if the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to a variable in the generated quantities block. That is, if we have\ngenerated quantities {\n  real f_val = f(y_tilde, theta);\n  // ...\n}\nwhere the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to variable f_val, then the posterior mean of f_val will be the expectation \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\).\n\n\n\nThe gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density \\(p(\\lambda \\mid y)\\) will also follow a gamma distribution.\nBecause the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive \\(p(\\tilde{y} \\mid\ny)\\) will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, \\(y \\sim\n\\textrm{negative-binomial}(\\alpha, \\beta)\\) if \\(\\lambda \\sim\n\\textrm{gamma}(\\alpha, \\beta)\\) and \\(y \\sim \\textrm{poisson}(\\lambda).\\) Rather than marginalizing out the rate parameter \\(\\lambda\\) analytically as can be done to define the negative binomial probability mass function, the rate \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is sampled from the posterior and then used to generate a draw of \\(\\tilde{y}^{(m)} \\sim p(y \\mid \\lambda^{(m)}).\\)\n\n\n\n\n\n\nConsider a regression with a single predictor \\(x_n\\) for the training outcome \\(y_n\\) and \\(\\tilde{x}_n\\) for the test outcome \\(\\tilde{y}_n.\\) Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in \\[\\begin{eqnarray}\np(\\tilde{y} \\mid \\tilde{x}, y, x)\n& = & \\int p(\\tilde{y} \\mid \\tilde{x}, \\theta) \\cdot p(\\theta \\mid y, x) \\,\n\\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m=1}^M \\, p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\end{eqnarray}\\] where \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y).\\)\n\n\n\nThe following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed \\(y\\) values are coded as data. The predictive quantities \\(\\tilde{y}\\) appear in the generated quantities block, where they are generated by random number generation.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0&gt; y;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ poisson_log(alpha + beta * x);\n  { alpha, beta } ~ normal(0, 1);\n}\ngenerated quantities {\n  array[N_tilde] int&lt;lower=0&gt; y_tilde\n    = poisson_log_rng(alpha + beta * x_tilde);\n}\nThe Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s poisson_log vs. poisson, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept alpha and slope beta, are given standard normal priors.\nIn the model block, the log rate for the Poisson is a linear function of the training data \\(x\\), whereas in the generated quantities block it is a function of the test data \\(\\tilde{x}\\). Because the generated quantities block does not affect the posterior draws, the model fits \\(\\alpha\\) and \\(\\beta\\) using only the training data, reserving \\(\\tilde{x}\\) to generate \\(\\tilde{y}.\\)\nThe result from running Stan is a predictive sample \\(\\tilde{y}^{(1)},\n\\ldots \\tilde{y}^{(M)}\\) where each \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid \\tilde{x}, x, y).\\)\nThe mean of the posterior predictive distribution is the expected value \\[\\begin{align}\n\\mathbb{E}[\\tilde{y} \\mid \\tilde{x}, x, y]\n& =\n\\int\n\\tilde{y}\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n\\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}^{(m)},\n\\end{align}\\] where the \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y)\\) are drawn from the posterior predictive distribution. Thus the posterior mean of y_tilde[n] after running Stan is the expected value of \\(\\tilde{y}_n\\) conditioned on the training data \\(x, y\\) and predictor \\(\\tilde{x}_n.\\) This is the Bayesian estimate for \\(\\tilde{y}\\) with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for \\(\\tilde{y}\\), as well as covariance of the \\(\\tilde{y_n}.\\) The posterior draws \\(\\tilde{y}^{(m)}\\) may also be used to estimate predictive event probabilities, such as \\(\\Pr[\\tilde{y}_1 &gt; 0]\\) or \\(\\Pr[\\prod_{n =\n1}^{\\tilde{N}}(\\tilde{y_n}) &gt; 1],\\) as expectations of indicator functions.\nAll of this can be carried out by running Stan only a single time to draw a single sample of \\(M\\) draws, \\[\n\\tilde{y}^{(1)}, \\ldots, \\tilde{y}^{(M)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\n\\] It’s only when moving to cross-validation where multiple runs are required.\n\n\n\n\nEvent probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate \\(\\Pr[\\lambda &gt; 5 \\mid y]\\) in the simple Poisson example with only a rate parameter \\(\\lambda\\), it suffices to define a generated quantity\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; lambda_gt_5 = lambda &gt; 5;\n  // ...\n}\nThe value of the expression lambda &gt; 5 is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = &\n\\int \\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}[\\lambda^{(m)} &gt; 5],\n\\end{eqnarray*}\\] where each \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter lambda_gt_5.\nIn general, event probabilities may be expressed as expectations of indicator functions. For example, \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = & \\mathbb{E}[\\textrm{I}[\\lambda &gt; 5] \\mid y]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}(\\lambda^{(m)} &gt; 5).\n\\end{eqnarray*}\\] The last line above is the posterior mean of the indicator function as coded in Stan.\nEvent probabilities involving posterior predictive quantities \\(\\tilde{y}\\) work exactly the same way as those for parameters. For example, if \\(\\tilde{y}_n\\) is the prediction for the \\(n\\)-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then \\[\\begin{eqnarray*}\n\\Pr[\\tilde{y}_3 &gt; \\tilde{y}_7 \\mid \\tilde{x}, x, y]\n& = &\n\\mathbb{E}\\!\\left[I[\\tilde{y}_3 &gt; \\tilde{y}_7] \\mid \\tilde{x}, x, y\\right]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\tilde{y}_3 &gt; \\tilde{y}_7)\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\, \\textrm{d}\\tilde{y}\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n\\textrm{I}(\\tilde{y}^{(m)}_3 &gt; \\tilde{y}^{(m)}_7),\n\\end{eqnarray*}\\] where \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y).\\)\n\n\n\nStan’s sampling algorithms take a Stan program representing a posterior \\(p(\\theta \\mid y, x)\\) along with actual data \\(x\\) and \\(y\\) to produce a set of draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)}\\) from the posterior. Posterior predictive draws \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y)\\) can be generated by drawing \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\tilde{x}, \\theta^{(m)})\n\\] from the data model. Note that drawing \\(\\tilde{y}^{(m)}\\) only depends on the new predictors \\(\\tilde{x}\\) and the posterior draws \\(\\theta^{(m)}\\). Most importantly, neither the original data or the model density is required.\nBy saving the posterior draws, predictions for new data items \\(\\tilde{x}\\) may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 5);\n  beta ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.5);\n}\nA second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde\n    = normal_rng(alpha + beta * x_tilde, sigma);\n}\nWhen running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for \\(\\tilde{y}\\) or derived quantities such as event probabilities.\nAny posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Matrices, Vectors, Arrays, and Tuples"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/index.html",
-    "href": "stan-users-guide/index.html",
-    "title": "Stan User’s Guide",
+    "objectID": "stan-users-guide/posterior-prediction.html#posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-prediction.html#posterior-predictive-distribution",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "This is the official user’s guide for Stan. It provides example models and programming techniques for coding statistical models in Stan.\n\nPart 1 gives Stan code and discussions for several important classes of models.\nPart 2 discusses various general Stan programming techniques that are not tied to any particular model.\nPart 3 introduces algorithms for calibration and model checking that require multiple runs of Stan.\nThe appendices provide an introduction to the stanc3 compiler used in the various interfaces to Stan, a style guide, and advice for users of BUGS and JAGS.\n\nWe recommend working through this guide using the textbooks Bayesian Data Analysis and Statistical Rethinking: A Bayesian Course with Examples in R and Stan as references on the concepts, and using the Stan Reference Manual when necessary to clarify programming issues.\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "text": "Given a full Bayesian model \\(p(y, \\theta)\\), the posterior predictive density for new data \\(\\tilde{y}\\) given observed data \\(y\\) is \\[\np(\\tilde{y} \\mid y)\n=\n\\int p(\\tilde{y} \\mid \\theta) \\cdot p(\\theta \\mid y)\n\\, \\textrm{d}\\theta.\n\\] The product under the integral reduces to the joint posterior density \\(p(\\tilde{y}, \\theta \\mid y),\\) so that the integral is simply marginalizing out the parameters \\(\\theta,\\) leaving the predictive density \\(p(\\tilde{y} \\mid y)\\) of future observations given past observations.",
     "crumbs": [
       "Stan Users Guide",
-      "Stan User's Guide"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/gaussian-processes.html",
-    "href": "stan-users-guide/gaussian-processes.html",
-    "title": "Gaussian Processes",
+    "objectID": "stan-users-guide/posterior-prediction.html#computing-the-posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-prediction.html#computing-the-posterior-predictive-distribution",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "Gaussian processes are continuous stochastic processes and thus may be interpreted as providing a probability distribution over functions. A probability distribution over continuous functions may be viewed, roughly, as an uncountably infinite collection of random variables, one for each valid input. The generality of the supported functions makes Gaussian priors popular choices for priors in general multivariate (non-linear) regression problems.\nThe defining feature of a Gaussian process is that the joint distribution of the function’s value at a finite number of input points is a multivariate normal distribution. This makes it tractable to both fit models from finite amounts of observed data and make predictions for finitely many new data points.\nUnlike a simple multivariate normal distribution, which is parameterized by a mean vector and covariance matrix, a Gaussian process is parameterized by a mean function and covariance function. The mean and covariance functions apply to vectors of inputs and return a mean vector and covariance matrix which provide the mean and covariance of the outputs corresponding to those input points in the functions drawn from the process.\nGaussian processes can be encoded in Stan by implementing their mean and covariance functions or by using the specialized covariance functions outlined below, and plugging the result into the Gaussian model.\nThis form of model is straightforward and may be used for simulation, model fitting, or posterior predictive inference. A more efficient Stan implementation for the GP with a normally distributed outcome marginalizes over the latent Gaussian process, and applies a Cholesky-factor reparameterization of the Gaussian to compute the likelihood and the posterior predictive distribution analytically.\nAfter defining Gaussian processes, this chapter covers the basic implementations for simulation, hyperparameter estimation, and posterior predictive inference for univariate regressions, multivariate regressions, and multivariate logistic regressions. Gaussian processes are general, and by necessity this chapter only touches on some basic models. For more information, see Rasmussen and Williams (2006).\nNote that fitting Gaussian processes as described below using exact inference by computing Cholesky of the covariance matrix scales cubicly with the size of data. Due to how Stan autodiff is implemented, Stan is also slower than Gaussian process specialized software. It is likely that Gaussian processes using exact inference by computing Cholesky of the covariance matrix with \\(N&gt;1000\\) are too slow for practical purposes in Stan. There are many approximations to speed-up Gaussian process computation, from which the basis function approaches for 1-3 dimensional \\(x\\) are easiest to implement in Stan (see, e.g., Riutort-Mayol et al. (2023)).\n\n\nThe data for a multivariate Gaussian process regression consists of a series of \\(N\\) inputs \\(x_1,\\dotsc,x_N \\in \\mathbb{R}^D\\) paired with outputs \\(y_1,\\dotsc,y_N \\in \\mathbb{R}\\). The defining feature of Gaussian processes is that the probability of a finite number of outputs \\(y\\) conditioned on their inputs \\(x\\) is Gaussian: \\[\ny \\sim \\textsf{multivariate normal}(m(x), K(x \\mid \\theta)),\n\\] where \\(m(x)\\) is an \\(N\\)-vector and \\(K(x \\mid \\theta)\\) is an \\(N \\times N\\) covariance matrix. The mean function \\(m : \\mathbb{R}^{N \\times D}\n\\rightarrow \\mathbb{R}^{N}\\) can be anything, but the covariance function \\(K : \\mathbb{R}^{N \\times D} \\rightarrow \\mathbb{R}^{N \\times N}\\) must produce a positive-definite matrix for any input \\(x\\).1\nA popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, \\[\n  K(x \\mid \\alpha, \\rho, \\sigma)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j} \\sigma^2,\n\\] where \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\) are hyperparameters defining the covariance function and where \\(\\delta_{i, j}\\) is the Kronecker delta function with value 1 if \\(i = j\\) and value 0 otherwise; this test is between the indexes \\(i\\) and \\(j\\), not between values \\(x_i\\) and \\(x_j\\). This kernel is obtained through a convolution of two independent Gaussian processes, \\(f_1\\) and \\(f_2\\), with kernels \\[\n  K_1(x \\mid \\alpha, \\rho)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n\\] and \\[\n  K_2(x \\mid \\sigma)_{i, j}\n=\n\\delta_{i, j} \\sigma^2.\n\\]\nThe addition of \\(\\sigma^2\\) on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs \\(x_i = x_j\\). In statistical terms, \\(\\sigma\\) is the scale of the noise term in the regression.\nThe hyperparameter \\(\\rho\\) is the length-scale, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of \\(\\rho\\) closer to zero lead the GP to represent high-frequency functions, whereas larger values of \\(\\rho\\) lead to low-frequency functions. The hyperparameter \\(\\alpha\\) is the marginal standard deviation. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP \\(f_1\\) prior at a single input \\(x\\) conditional on one value of \\(\\alpha\\) one would recover \\(\\alpha\\).\nThe only term in the squared exponential covariance function involving the inputs \\(x_i\\) and \\(x_j\\) is their vector difference, \\(x_i - x_j\\). This produces a process with stationary covariance in the sense that if an input vector \\(x\\) is translated by a vector \\(\\epsilon\\) to \\(x +\n\\epsilon\\), the covariance at any pair of outputs is unchanged, because \\(K(x \\mid \\theta) = K(x + \\epsilon \\mid \\theta)\\).\nThe summation involved is just the squared Euclidean distance between \\(x_i\\) and \\(x_j\\) (i.e., the squared \\(L_2\\) norm of their difference, \\(x_i -\nx_j\\)). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\).\nChanging the notion of distance from Euclidean to taxicab distance (i.e., an \\(L_1\\) norm) changes the support to functions which are continuous but not smooth.\n\n\n\nIt is simplest to start with a Stan model that does nothing more than simulate draws of functions \\(f\\) from a Gaussian process. In practical terms, the model will draw values \\(y_n = f(x_n)\\) for finitely many input points \\(x_n\\).\nThe Stan model defines the mean and covariance functions in a transformed data block and then samples outputs \\(y\\) in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to \\(\\alpha^2 = 1\\), \\(\\rho^2 = 1\\), and \\(\\sigma^2 = 0.1\\), and the mean function \\(m\\) is defined to always return the zero vector, \\(m(x) = \\textbf{0}\\). Consider the following implementation of a Gaussian process simulator.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K;\n  vector[N] mu = rep_vector(0, N);\n  for (i in 1:(N - 1)) {\n    K[i, i] = 1 + 0.1;\n    for (j in (i + 1):N) {\n      K[i, j] = exp(-0.5 * square(x[i] - x[j]));\n      K[j, i] = K[i, j];\n    }\n  }\n  K[N, N] = 1 + 0.1;\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K = gp_exp_quad_cov(x, 1.0, 1.0);\n  vector[N] mu = rep_vector(0, N);\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + 0.1;\n  }\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe input data are just the vector of inputs x and its size N. Such a model can be used with values of x evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.\n\n\nOnly the input data needs to change in moving from a univariate model to a multivariate model.\nThe only lines that change from the univariate model above are as follows.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n}\ntransformed data {\n  // ...\n}\nThe data are now declared as an array of vectors instead of an array of scalars; the dimensionality D is also declared.\nIn the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.\n\n\n\nA more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose \\(\\eta\\) is an an isotropic standard normal variate \\[\n\\eta \\sim \\textsf{normal}(\\textbf{0}, \\textrm{I}),\n\\] where \\(\\textbf{0}\\) is an \\(N\\)-vector of 0 values and \\(\\textrm{I}\\) is the \\(N\n\\times N\\) identity matrix. Let \\(L\\) be the Cholesky decomposition of \\(K(x \\mid \\theta)\\), i.e., the lower-triangular matrix \\(L\\) such that \\(LL^{\\top} =\nK(x \\mid \\theta)\\). Then the transformed variable \\(\\mu + L\\eta\\) has the intended target distribution, \\[\n  \\mu + L\\eta \\sim \\textsf{multivariate normal}(\\mu(x), K(x \\mid \\theta)).\n\\]\nThis transform can be applied directly to Gaussian process simulation.\nThis model has the same data declarations for N and x, and the same transformed data definitions of mu and K as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.\n// ...\ntransformed data {\n  matrix[N, N] L;\n  // ...\n  L = cholesky_decompose(K);\n}\nparameters {\n  vector[N] eta;\n}\nmodel {\n  eta ~ std_normal();\n}\ngenerated quantities {\n  vector[N] y;\n  y = mu + L * eta;\n}\nThe Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix K computed. The isotropic normal distribution for eta is specified as a vectorized univariate distribution for efficiency; this specifies that each eta[n] has an independent standard normal distribution. The sampled vector y is then defined as a generated quantity using a direct encoding of the transform described above.\n\n\n\n\n\n\nThe full generative model for a GP with a normal outcome, \\(y \\in \\mathbb{R}^N\\), with inputs \\(x \\in \\mathbb{R}^N\\), for a finite \\(N\\): \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\nf      &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i    &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\}\n\\end{align*}\\] With a normal outcome, it is possible to integrate out the Gaussian process \\(f\\), yielding the more parsimonious model: \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\ny      &\\sim \\textsf{multivariate normal}\n         \\left(0, K(x \\mid \\alpha, \\rho) + \\textbf{I}_N \\sigma^2\\right) \\\\\n\\end{align*}\\]\nIt can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.\nThe hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (\\(\\alpha\\)), the scale of the output noise (\\(\\sigma\\)), and the scale at which distances are measured among inputs (\\(\\rho\\)). See the Gaussian process priors section for more information about how to specify appropriate priors for the hyperparameters.\nThe Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix K in the model block, rather than the transformed data block.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  vector[N] mu = rep_vector(0, N);\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N, N] L_K;\n  matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n  real sq_sigma = square(sigma);\n\n  // diagonal elements\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + sq_sigma;\n  }\n\n  L_K = cholesky_decompose(K);\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y ~ multi_normal_cholesky(mu, L_K);\n}\nThe data block declares a vector y of observed values y[n] for inputs x[n]. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix K is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value y is known and the covariance matrix K is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.\nWe have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the cholesky_decompose function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (\\(N \\gtrsim 100\\)) the Cholesky decomposition version will be faster.\nHamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model (Neal 1997). If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.\n\n\nWe can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, \\(\\delta\\) to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\nTwo differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length \\(N\\) called eta. This is used in the model block to generate a multivariate normal vector called \\(f\\), corresponding to the latent GP. We put a \\(\\textsf{normal}(0,1)\\) prior on eta like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for \\(y\\) with one \\(N\\)-dimensional multivariate normal with an identity covariance matrix multiplied by \\(\\sigma^2\\), we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.\n\n\n\n\nGaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.\n\n\nIf we want to model count data, we can remove the \\(\\sigma\\) parameter, and use poisson_log, which implements a Poisson distribution with log link function, rather than normal. We can also add an overall mean parameter, \\(a\\), which will account for the marginal expected value for \\(y\\). We do this because we cannot center count data like we would for normally distributed data.\ndata {\n  // ...\n  array[N] int&lt;lower=0&gt; y;\n  // ...\n}\n// ...\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\nmodel {\n  // ...\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ poisson_log(a + f);\n}\n\n\n\nFor binary classification problems, the observed outputs \\(z_n \\in\n\\{ 0,1 \\}\\) are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs \\(y_n\\) through the logistic link, \\[\nz_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(y_n)),\n\\] or in other words, \\[\n\\Pr[z_n = 1] = \\operatorname{logit}^{-1}(y_n).\n\\]\nWe can extend our latent variable GP Stan program to deal with classification problems. Below a is the bias term, which can help account for imbalanced classes in the training data:\ndata {\n  // ...\n  array[N] int&lt;lower=0, upper=1&gt; z;\n  // ...\n}\n// ...\nmodel {\n  // ...\n  y ~ bernoulli_logit(a + f);\n}\n\n\n\n\nIf we have multivariate inputs \\(x \\in \\mathbb{R}^D\\), the squared exponential covariance function can be further generalized by fitting a scale parameter \\(\\rho_d\\) for each dimension \\(d\\), \\[\n  k(x \\mid \\alpha, \\vec{\\rho}, \\sigma)_{i, j} = \\alpha^2 \\exp\n\\left(-\\dfrac{1}{2}\n\\sum_{d=1}^D \\dfrac{1}{\\rho_d^2} (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j}\\sigma^2.\n\\] The estimation of \\(\\rho\\) was termed “automatic relevance determination” by Neal (1996), but this is misleading, because the magnitude of the scale of the posterior for each \\(\\rho_d\\) is dependent on the scaling of the input data along dimension \\(d\\). Moreover, the scale of the parameters \\(\\rho_d\\) measures non-linearity along the \\(d\\)-th dimension, rather than “relevance” (Piironen and Vehtari 2016).\nA priori, the closer \\(\\rho_d\\) is to zero, the more nonlinear the conditional mean in dimension \\(d\\) is. A posteriori, the actual dependencies between \\(x\\) and \\(y\\) play a role. With one covariate \\(x_1\\) having a linear effect and another covariate \\(x_2\\) having a nonlinear effect, it is possible that \\(\\rho_1 &gt; \\rho_2\\) even if the predictive relevance of \\(x_1\\) is higher (Rasmussen and Williams 2006, 80). The collection of \\(\\rho_d\\) (or \\(1/\\rho_d\\)) parameters can also be modeled hierarchically.\nThe implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying rho to be an array.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  array[D] real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\n\n\n\nFormulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.\nPerhaps most importantly, the parameters \\(\\rho\\) and \\(\\alpha\\) are weakly identified (Zhang 2004). The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.\n\n\nGPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.\nWe may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.\nIf our model consists of only the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\} \\\\\n    & x \\in \\mathbb{R}^{N \\times D}, \\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, inv_gamma_lpdf in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.2.\nIf we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\big(0, K(x \\mid \\alpha, \\rho)\\big) \\\\\ny_i &\\sim \\textsf{normal}\\left(\\beta_0 + x_i \\beta_{[1:D]} + f_i, \\sigma\\right) \\, \\forall i\n  \\in \\{1, \\dots, N\\} \\\\\n    & x_i^T, \\beta_{[1:D]} \\in \\mathbb{R}^D,\\quad\n      x \\in \\mathbb{R}^{N \\times D},\\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: \\[\\begin{align*}\nf(x \\mid a, b, p) &= \\dfrac{\\left(a/b\\right)^{p/2}}{2K_p\\left(\\sqrt{ab}\\right)} x^{p - 1}\\exp\\big(-(ax + b\n  / x)/2\\big) \\\\\n  & x, a, b \\in \\mathbb{R}^{+},\\quad\n    p \\in \\mathbb{Z}\n\\end{align*}\\]\nwhich has an inverse gamma left tail if \\(p \\leq 0\\) and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:\nfunctions {\n  real generalized_inverse_gaussian_lpdf(real x, int p,\n                                        real a, real b) {\n    return p * 0.5 * log(a / b)\n      - log(2 * modified_bessel_second_kind(p, sqrt(a * b)))\n      + (p - 1) * log(x)\n      - (a * x + b / x) * 0.5;\n }\n}\ndata {\n  // ...\n}\nIf we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of \\(\\rho\\) in a fixed-domain \\([0,T]\\), we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is \\(T / \\pi\n\\rho\\). You can see that as \\(\\rho\\) decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.\n\n\n\nThe parameter \\(\\alpha\\) corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-\\(t\\) prior on \\(\\alpha\\).\nA half-\\(t\\) or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.\n\n\n\n\nSuppose for a given sequence of inputs \\(x\\) that the corresponding outputs \\(y\\) are observed. Given a new sequence of inputs \\(\\tilde{x}\\), the posterior predictive distribution of their labels is computed by sampling outputs \\(\\tilde{y}\\) according to \\[\np\\left(\\tilde{y} \\mid \\tilde{x},x,y\\right)\n\\ = \\\n\\frac{p\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right)}\n     {p(y \\mid x)}\n\\ \\propto \\\np\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right).\n\\]\nA direct implementation in Stan defines a model in terms of the joint distribution of the observed \\(y\\) and unobserved \\(\\tilde{y}\\).\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y1 ~ normal(f[1:N1], sigma);\n}\ngenerated quantities {\n  vector[N2] y2;\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f[N1 + n2], sigma);\n  }\n}\nThe input vectors x1 and x2 are declared as data, as is the observed output vector y1. The unknown output vector y2, which corresponds to input vector x2, is declared in the generated quantities block and will be sampled when the model is executed.\nA transformed data block is used to combine the input vectors x1 and x2 into a single vector x.\nThe model block declares and defines a local variable for the combined output vector f, which consists of the concatenation of the conditional mean for known outputs y1 and unknown outputs y2. Thus the combined output vector f is aligned with the combined input vector x. All that is left is to define the univariate normal distribution statement for y.\nThe generated quantities block defines the quantity y2. We generate y2 by randomly generating N2 values from univariate normals with each mean corresponding to the appropriate element in f.\n\n\nWe can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.\nConsider the following full model for prediction using logistic Gaussian process regression.\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  array[N1] int&lt;lower=0, upper=1&gt; z1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  z1 ~ bernoulli_logit(a + f[1:N1]);\n}\ngenerated quantities {\n  array[N2] int z2;\n  for (n2 in 1:N2) {\n    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);\n  }\n}\n\n\n\nBayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.\nJumping straight to the result, \\[\np\\left(\\tilde{y} \\mid \\tilde{x},y,x\\right)\n=\n\\textsf{normal}\\left(K^{\\top}\\Sigma^{-1}y,\\\n                \\Omega - K^{\\top}\\Sigma^{-1}K\\right),\n\\] where \\(\\Sigma = K(x \\mid \\alpha, \\rho, \\sigma)\\) is the result of applying the covariance function to the inputs \\(x\\) with observed outputs \\(y\\), \\(\\Omega =\nK(\\tilde{x} \\mid \\alpha, \\rho)\\) is the result of applying the covariance function to the inputs \\(\\tilde{x}\\) for which predictions are to be inferred, and \\(K\\) is the matrix of covariances between inputs \\(x\\) and \\(\\tilde{x}\\), which in the case of the exponentiated quadratic covariance function would be \\[\nK(x \\mid \\alpha, \\rho)_{i, j} = \\alpha^2 \\exp\\left(-\\dfrac{1}{2 \\rho^2}\n\\sum_{d=1}^D \\left(x_{i,d} - \\tilde{x}_{j,d}\\right)^2\\right).\n\\]\nThere is no noise term including \\(\\sigma^2\\) because the indexes of elements in \\(x\\) and \\(\\tilde{x}\\) are never the same.\nThis Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called gp_pred_rng which will generate a draw from the posterior predictive mean conditioned on observed data y1. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of \\(p(\\tilde{y})\\).\nfunctions {\n  vector gp_pred_rng(array[] real x2,\n                     vector y1,\n                     array[] real x1,\n                     real alpha,\n                     real rho,\n                     real sigma,\n                     real delta) {\n    int N1 = rows(y1);\n    int N2 = size(x2);\n    vector[N2] f2;\n    {\n      matrix[N1, N1] L_K;\n      vector[N1] K_div_y1;\n      matrix[N1, N2] k_x1_x2;\n      matrix[N1, N2] v_pred;\n      vector[N2] f2_mu;\n      matrix[N2, N2] cov_f2;\n      matrix[N2, N2] diag_delta;\n      matrix[N1, N1] K;\n      K = gp_exp_quad_cov(x1, alpha, rho);\n      for (n in 1:N1) {\n        K[n, n] = K[n, n] + square(sigma);\n      }\n      L_K = cholesky_decompose(K);\n      K_div_y1 = mdivide_left_tri_low(L_K, y1);\n      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';\n      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);\n      f2_mu = (k_x1_x2' * K_div_y1);\n      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);\n      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;\n      diag_delta = diag_matrix(rep_vector(delta, N2));\n\n      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);\n    }\n    return f2;\n  }\n}\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  vector[N1] mu = rep_vector(0, N1);\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N1, N1] L_K;\n  {\n    matrix[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);\n    real sq_sigma = square(sigma);\n\n    // diagonal elements\n    for (n1 in 1:N1) {\n      K[n1, n1] = K[n1, n1] + sq_sigma;\n    }\n\n    L_K = cholesky_decompose(K);\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y1 ~ multi_normal_cholesky(mu, L_K);\n}\ngenerated quantities {\n  vector[N2] f2;\n  vector[N2] y2;\n\n  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f2[n2], sigma);\n  }\n}\n\n\n\n\nSuppose we have observations \\(y_i \\in \\mathbb{R}^M\\) observed at \\(x_i \\in \\mathbb{R}^K\\). One can model the data like so: \\[\\begin{align*}\ny_i  &\\sim \\textsf{multivariate normal}\\left(f(x_i), \\textbf{I}_M \\sigma^2\\right) \\\\\nf(x) &\\sim \\textsf{GP}\\big(m(x), K(x \\mid \\theta, \\phi)\\big) \\\\\n     & K(x \\mid \\theta) \\in \\mathbb{R}^{M \\times M}, \\quad\n       f(x), m(x) \\in \\mathbb{R}^M\n\\end{align*}\\] where the \\(K(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]}\\) entry defines the covariance between \\(f_m(x)\\) and \\(f_{m^\\prime}(x^\\prime)(x)\\). This construction of Gaussian processes allows us to learn the covariance between the output dimensions of \\(f(x)\\). If we parameterize our kernel \\(K\\): \\[\nK(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]} = k\\left(x, x^\\prime \\mid\n\\theta\\right) k\\left(m, m^\\prime \\mid \\phi\\right)\n\\] then our finite dimensional generative model for the above is: \\[\\begin{align*}\nf        &\\sim \\textsf{matrixnormal}\\big(m(x), K(x \\mid \\alpha, \\rho), C(\\phi)\\big) \\\\\ny_{i, m} &\\sim \\textsf{normal}(f_{i,m}, \\sigma) \\\\\nf        &\\in  \\mathbb{R}^{N \\times M}\n\\end{align*}\\] where \\(K(x \\mid \\alpha, \\rho)\\) is the exponentiated quadratic kernel we’ve used throughout this chapter, and \\(C(\\phi)\\) is a positive-definite matrix, parameterized by some vector \\(\\phi\\).\nThe matrix normal distribution has two covariance matrices: \\(K(x \\mid\n\\alpha, \\rho)\\) to encode column covariance, and \\(C(\\phi)\\) to define row covariance. The salient features of the matrix normal are that the rows of the matrix \\(f\\) are distributed: \\[\nf_{[n,]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[n,]}, K(x \\mid \\alpha,\n\\rho)_{[n,n]} C(\\phi)\\big)\n\\] and that the columns of the matrix \\(f\\) are distributed: \\[\nf_{[,m]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[,m]}, K(x\n  \\mid \\alpha, \\rho) C(\\phi)_{[m,m]}\\big)\n\\] This also means means that \\(\\mathbb{E}\\left[f^T f\\right]\\) is equal to \\(\\operatorname{trace}\\!\\big(K(x \\mid \\alpha, \\rho)\\big) \\times C\\), whereas \\(\\mathbb{E}\\left[ff^T\\right]\\) is \\(\\operatorname{trace}(C) \\times K(x \\mid \\alpha, \\rho)\\). We can derive this using properties of expectation and the matrix normal density.\nWe should set \\(\\alpha\\) to \\(1.0\\) because the parameter is not identified unless we constrain \\(\\operatorname{trace}(C) = 1\\). Otherwise, we can multiply \\(\\alpha\\) by a scalar \\(d\\) and \\(C\\) by \\(1/d\\) and our likelihood will not change.\nWe can generate a random variable \\(f\\) from a matrix normal density in \\(\\mathbb{R}^{N \\times M}\\) using the following algorithm: \\[\\begin{align*}\n\\eta_{i,j} &\\sim \\textsf{normal}(0, 1) \\, \\forall i,j \\\\\nf          &= L_{K(x \\mid 1.0, \\rho)} \\, \\eta \\, L_C(\\phi)^T \\\\\nf          &\\sim \\textsf{matrixnormal}\\big(0, K(x \\mid 1.0, \\rho), C(\\phi)\\big) \\\\\n\\eta       &\\in \\mathbb{R}^{N \\times M} \\\\\nL_C(\\phi)  &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(C(\\phi)\\big) \\\\\nL_{K(x \\mid 1.0, \\rho)} &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(K(x \\mid 1.0, \\rho)\\big)\n\\end{align*}\\]\nThis can be implemented in Stan using a latent-variable GP formulation. We’ve used \\(\\textsf{LKJCorr}\\) for \\(C(\\phi)\\), but any positive-definite matrix will do.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] real x;\n  matrix[N, D] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  vector&lt;lower=0&gt;[D] alpha;\n  real&lt;lower=0&gt; sigma;\n  cholesky_factor_corr[D] L_Omega;\n  matrix[N, D] eta;\n}\nmodel {\n  matrix[N, D] f;\n  {\n    matrix[N, N] K = gp_exp_quad_cov(x, 1.0, rho);\n    matrix[N, N] L_K;\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta\n        * diag_pre_multiply(alpha, L_Omega)';\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(3);\n  to_vector(eta) ~ std_normal();\n\n  to_vector(y) ~ normal(to_vector(f), sigma);\n}\ngenerated quantities {\n  matrix[D, D] Omega;\n  Omega = L_Omega * L_Omega';\n}",
+    "text": "The posterior predictive density (or mass) of a prediction \\(\\tilde{y}\\) given observed data \\(y\\) can be computed using \\(M\\) Monte Carlo draws\n\\[\n\\theta^{(m)} \\sim p(\\theta \\mid y)\n\\] from the posterior as \\[\np(\\tilde{y} \\mid y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\]\nComputing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, \\(\\log\np(\\tilde{y} \\mid y)\\) may be computed using the stable log sum of exponents function as \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\\\[4pt]\n& = &\n- \\log M\n+ \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\theta^{(m)}),\n\\end{eqnarray*}\\] where \\[\n\\textrm{log-sum-exp}_{m = 1}^M v_m\n= \\log \\sum_{m = 1}^M \\exp v_m\n\\] is used to maintain arithmetic precision. See the section on log sum of exponentials for more details.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Gaussian Processes"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/gaussian-processes.html#gaussian-process-regression",
-    "href": "stan-users-guide/gaussian-processes.html#gaussian-process-regression",
-    "title": "Gaussian Processes",
+    "objectID": "stan-users-guide/posterior-prediction.html#sampling-from-the-posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-prediction.html#sampling-from-the-posterior-predictive-distribution",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "The data for a multivariate Gaussian process regression consists of a series of \\(N\\) inputs \\(x_1,\\dotsc,x_N \\in \\mathbb{R}^D\\) paired with outputs \\(y_1,\\dotsc,y_N \\in \\mathbb{R}\\). The defining feature of Gaussian processes is that the probability of a finite number of outputs \\(y\\) conditioned on their inputs \\(x\\) is Gaussian: \\[\ny \\sim \\textsf{multivariate normal}(m(x), K(x \\mid \\theta)),\n\\] where \\(m(x)\\) is an \\(N\\)-vector and \\(K(x \\mid \\theta)\\) is an \\(N \\times N\\) covariance matrix. The mean function \\(m : \\mathbb{R}^{N \\times D}\n\\rightarrow \\mathbb{R}^{N}\\) can be anything, but the covariance function \\(K : \\mathbb{R}^{N \\times D} \\rightarrow \\mathbb{R}^{N \\times N}\\) must produce a positive-definite matrix for any input \\(x\\).1\nA popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, \\[\n  K(x \\mid \\alpha, \\rho, \\sigma)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j} \\sigma^2,\n\\] where \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\) are hyperparameters defining the covariance function and where \\(\\delta_{i, j}\\) is the Kronecker delta function with value 1 if \\(i = j\\) and value 0 otherwise; this test is between the indexes \\(i\\) and \\(j\\), not between values \\(x_i\\) and \\(x_j\\). This kernel is obtained through a convolution of two independent Gaussian processes, \\(f_1\\) and \\(f_2\\), with kernels \\[\n  K_1(x \\mid \\alpha, \\rho)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n\\] and \\[\n  K_2(x \\mid \\sigma)_{i, j}\n=\n\\delta_{i, j} \\sigma^2.\n\\]\nThe addition of \\(\\sigma^2\\) on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs \\(x_i = x_j\\). In statistical terms, \\(\\sigma\\) is the scale of the noise term in the regression.\nThe hyperparameter \\(\\rho\\) is the length-scale, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of \\(\\rho\\) closer to zero lead the GP to represent high-frequency functions, whereas larger values of \\(\\rho\\) lead to low-frequency functions. The hyperparameter \\(\\alpha\\) is the marginal standard deviation. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP \\(f_1\\) prior at a single input \\(x\\) conditional on one value of \\(\\alpha\\) one would recover \\(\\alpha\\).\nThe only term in the squared exponential covariance function involving the inputs \\(x_i\\) and \\(x_j\\) is their vector difference, \\(x_i - x_j\\). This produces a process with stationary covariance in the sense that if an input vector \\(x\\) is translated by a vector \\(\\epsilon\\) to \\(x +\n\\epsilon\\), the covariance at any pair of outputs is unchanged, because \\(K(x \\mid \\theta) = K(x + \\epsilon \\mid \\theta)\\).\nThe summation involved is just the squared Euclidean distance between \\(x_i\\) and \\(x_j\\) (i.e., the squared \\(L_2\\) norm of their difference, \\(x_i -\nx_j\\)). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\).\nChanging the notion of distance from Euclidean to taxicab distance (i.e., an \\(L_1\\) norm) changes the support to functions which are continuous but not smooth.",
+    "text": "Given draws from the posterior \\(\\theta^{(m)} \\sim p(\\theta \\mid y),\\) draws from the posterior predictive \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid y)\\) can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\theta^{(m)}).\n\\]\nRandomly drawing \\(\\tilde{y}\\) from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because \\(\\theta\\) is unknown and estimated based only on a finite sample of data \\(y\\). Aleatoric uncertainty arises because even a known value of \\(\\theta\\) leads to uncertainty about new \\(\\tilde{y}\\) as described by the data model \\(p(\\tilde{y} \\mid \\theta)\\). Both forms of uncertainty show up in the factored form of the posterior predictive distribution, \\[\np(\\tilde{y} \\mid y)\n=\n\\int\n\\underbrace{p(\\tilde{y} \\mid \\theta)}_{\\begin{array}{l}\n                                         \\textrm{aleatoric}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\cdot \\underbrace{p(\\theta \\mid y)}_{\\begin{array}{l}\n                                         \\textrm{epistemic}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\, \\textrm{d}\\theta.\n\\]",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Gaussian Processes"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/gaussian-processes.html#simulating-from-a-gaussian-process",
-    "href": "stan-users-guide/gaussian-processes.html#simulating-from-a-gaussian-process",
-    "title": "Gaussian Processes",
+    "objectID": "stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan",
+    "href": "stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "It is simplest to start with a Stan model that does nothing more than simulate draws of functions \\(f\\) from a Gaussian process. In practical terms, the model will draw values \\(y_n = f(x_n)\\) for finitely many input points \\(x_n\\).\nThe Stan model defines the mean and covariance functions in a transformed data block and then samples outputs \\(y\\) in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to \\(\\alpha^2 = 1\\), \\(\\rho^2 = 1\\), and \\(\\sigma^2 = 0.1\\), and the mean function \\(m\\) is defined to always return the zero vector, \\(m(x) = \\textbf{0}\\). Consider the following implementation of a Gaussian process simulator.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K;\n  vector[N] mu = rep_vector(0, N);\n  for (i in 1:(N - 1)) {\n    K[i, i] = 1 + 0.1;\n    for (j in (i + 1):N) {\n      K[i, j] = exp(-0.5 * square(x[i] - x[j]));\n      K[j, i] = K[i, j];\n    }\n  }\n  K[N, N] = 1 + 0.1;\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K = gp_exp_quad_cov(x, 1.0, 1.0);\n  vector[N] mu = rep_vector(0, N);\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + 0.1;\n  }\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe input data are just the vector of inputs x and its size N. Such a model can be used with values of x evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.\n\n\nOnly the input data needs to change in moving from a univariate model to a multivariate model.\nThe only lines that change from the univariate model above are as follows.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n}\ntransformed data {\n  // ...\n}\nThe data are now declared as an array of vectors instead of an array of scalars; the dimensionality D is also declared.\nIn the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.\n\n\n\nA more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose \\(\\eta\\) is an an isotropic standard normal variate \\[\n\\eta \\sim \\textsf{normal}(\\textbf{0}, \\textrm{I}),\n\\] where \\(\\textbf{0}\\) is an \\(N\\)-vector of 0 values and \\(\\textrm{I}\\) is the \\(N\n\\times N\\) identity matrix. Let \\(L\\) be the Cholesky decomposition of \\(K(x \\mid \\theta)\\), i.e., the lower-triangular matrix \\(L\\) such that \\(LL^{\\top} =\nK(x \\mid \\theta)\\). Then the transformed variable \\(\\mu + L\\eta\\) has the intended target distribution, \\[\n  \\mu + L\\eta \\sim \\textsf{multivariate normal}(\\mu(x), K(x \\mid \\theta)).\n\\]\nThis transform can be applied directly to Gaussian process simulation.\nThis model has the same data declarations for N and x, and the same transformed data definitions of mu and K as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.\n// ...\ntransformed data {\n  matrix[N, N] L;\n  // ...\n  L = cholesky_decompose(K);\n}\nparameters {\n  vector[N] eta;\n}\nmodel {\n  eta ~ std_normal();\n}\ngenerated quantities {\n  vector[N] y;\n  y = mu + L * eta;\n}\nThe Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix K computed. The isotropic normal distribution for eta is specified as a vectorized univariate distribution for efficiency; this specifies that each eta[n] has an independent standard normal distribution. The sampled vector y is then defined as a generated quantity using a direct encoding of the transform described above.",
+    "text": "Posterior predictive quantities can be coded in Stan using the generated quantities block.\n\n\nFor example, consider a simple Poisson model for count data with a rate parameter \\(\\lambda &gt; 0\\) having a gamma-distributed prior, \\[\n\\lambda \\sim \\textrm{gamma}(1, 1).\n\\] The \\(N\\) observations \\(y_1, \\ldots, y_N\\) are modeled as Poisson distributed, \\[\ny_n \\sim \\textrm{poisson}(\\lambda).\n\\]\n\n\n\nThe following Stan program defines a variable for \\(\\tilde{y}\\) by random number generation in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  lambda ~ gamma(1, 1);\n  y ~ poisson(lambda);\n}\ngenerated quantities {\n  int&lt;lower=0&gt; y_tilde = poisson_rng(lambda);\n}\nThe random draw from the data model for \\(\\tilde{y}\\) is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new \\(\\tilde{y}^{(m)} \\sim p(y \\mid\n\\lambda^{(m)})\\) for each posterior draw \\(\\lambda^{(m)} \\sim p(\\theta\n\\mid y).\\)\nThe posterior draws \\(\\tilde{y}^{(m)}\\) may be used to estimate the expected value of \\(\\tilde{y}\\) or any of its quantiles or posterior intervals, as well as event probabilities involving \\(\\tilde{y}\\). In general, \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\) may be evaluated as \\[\n\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\n\\approx \\frac{1}{M} \\sum_{m=1}^M f(\\tilde{y}^{(m)}, \\theta^{(m)}),\n\\] which is just the posterior mean of \\(f(\\tilde{y}, \\theta).\\) This quantity is computed by Stan if the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to a variable in the generated quantities block. That is, if we have\ngenerated quantities {\n  real f_val = f(y_tilde, theta);\n  // ...\n}\nwhere the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to variable f_val, then the posterior mean of f_val will be the expectation \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\).\n\n\n\nThe gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density \\(p(\\lambda \\mid y)\\) will also follow a gamma distribution.\nBecause the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive \\(p(\\tilde{y} \\mid\ny)\\) will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, \\(y \\sim\n\\textrm{negative-binomial}(\\alpha, \\beta)\\) if \\(\\lambda \\sim\n\\textrm{gamma}(\\alpha, \\beta)\\) and \\(y \\sim \\textrm{poisson}(\\lambda).\\) Rather than marginalizing out the rate parameter \\(\\lambda\\) analytically as can be done to define the negative binomial probability mass function, the rate \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is sampled from the posterior and then used to generate a draw of \\(\\tilde{y}^{(m)} \\sim p(y \\mid \\lambda^{(m)}).\\)",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Gaussian Processes"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/gaussian-processes.html#fit-gp.section",
-    "href": "stan-users-guide/gaussian-processes.html#fit-gp.section",
-    "title": "Gaussian Processes",
+    "objectID": "stan-users-guide/posterior-prediction.html#posterior-prediction-for-regressions",
+    "href": "stan-users-guide/posterior-prediction.html#posterior-prediction-for-regressions",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "The full generative model for a GP with a normal outcome, \\(y \\in \\mathbb{R}^N\\), with inputs \\(x \\in \\mathbb{R}^N\\), for a finite \\(N\\): \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\nf      &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i    &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\}\n\\end{align*}\\] With a normal outcome, it is possible to integrate out the Gaussian process \\(f\\), yielding the more parsimonious model: \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\ny      &\\sim \\textsf{multivariate normal}\n         \\left(0, K(x \\mid \\alpha, \\rho) + \\textbf{I}_N \\sigma^2\\right) \\\\\n\\end{align*}\\]\nIt can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.\nThe hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (\\(\\alpha\\)), the scale of the output noise (\\(\\sigma\\)), and the scale at which distances are measured among inputs (\\(\\rho\\)). See the Gaussian process priors section for more information about how to specify appropriate priors for the hyperparameters.\nThe Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix K in the model block, rather than the transformed data block.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  vector[N] mu = rep_vector(0, N);\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N, N] L_K;\n  matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n  real sq_sigma = square(sigma);\n\n  // diagonal elements\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + sq_sigma;\n  }\n\n  L_K = cholesky_decompose(K);\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y ~ multi_normal_cholesky(mu, L_K);\n}\nThe data block declares a vector y of observed values y[n] for inputs x[n]. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix K is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value y is known and the covariance matrix K is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.\nWe have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the cholesky_decompose function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (\\(N \\gtrsim 100\\)) the Cholesky decomposition version will be faster.\nHamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model (Neal 1997). If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.\n\n\nWe can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, \\(\\delta\\) to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\nTwo differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length \\(N\\) called eta. This is used in the model block to generate a multivariate normal vector called \\(f\\), corresponding to the latent GP. We put a \\(\\textsf{normal}(0,1)\\) prior on eta like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for \\(y\\) with one \\(N\\)-dimensional multivariate normal with an identity covariance matrix multiplied by \\(\\sigma^2\\), we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.\n\n\n\n\nGaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.\n\n\nIf we want to model count data, we can remove the \\(\\sigma\\) parameter, and use poisson_log, which implements a Poisson distribution with log link function, rather than normal. We can also add an overall mean parameter, \\(a\\), which will account for the marginal expected value for \\(y\\). We do this because we cannot center count data like we would for normally distributed data.\ndata {\n  // ...\n  array[N] int&lt;lower=0&gt; y;\n  // ...\n}\n// ...\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\nmodel {\n  // ...\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ poisson_log(a + f);\n}\n\n\n\nFor binary classification problems, the observed outputs \\(z_n \\in\n\\{ 0,1 \\}\\) are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs \\(y_n\\) through the logistic link, \\[\nz_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(y_n)),\n\\] or in other words, \\[\n\\Pr[z_n = 1] = \\operatorname{logit}^{-1}(y_n).\n\\]\nWe can extend our latent variable GP Stan program to deal with classification problems. Below a is the bias term, which can help account for imbalanced classes in the training data:\ndata {\n  // ...\n  array[N] int&lt;lower=0, upper=1&gt; z;\n  // ...\n}\n// ...\nmodel {\n  // ...\n  y ~ bernoulli_logit(a + f);\n}\n\n\n\n\nIf we have multivariate inputs \\(x \\in \\mathbb{R}^D\\), the squared exponential covariance function can be further generalized by fitting a scale parameter \\(\\rho_d\\) for each dimension \\(d\\), \\[\n  k(x \\mid \\alpha, \\vec{\\rho}, \\sigma)_{i, j} = \\alpha^2 \\exp\n\\left(-\\dfrac{1}{2}\n\\sum_{d=1}^D \\dfrac{1}{\\rho_d^2} (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j}\\sigma^2.\n\\] The estimation of \\(\\rho\\) was termed “automatic relevance determination” by Neal (1996), but this is misleading, because the magnitude of the scale of the posterior for each \\(\\rho_d\\) is dependent on the scaling of the input data along dimension \\(d\\). Moreover, the scale of the parameters \\(\\rho_d\\) measures non-linearity along the \\(d\\)-th dimension, rather than “relevance” (Piironen and Vehtari 2016).\nA priori, the closer \\(\\rho_d\\) is to zero, the more nonlinear the conditional mean in dimension \\(d\\) is. A posteriori, the actual dependencies between \\(x\\) and \\(y\\) play a role. With one covariate \\(x_1\\) having a linear effect and another covariate \\(x_2\\) having a nonlinear effect, it is possible that \\(\\rho_1 &gt; \\rho_2\\) even if the predictive relevance of \\(x_1\\) is higher (Rasmussen and Williams 2006, 80). The collection of \\(\\rho_d\\) (or \\(1/\\rho_d\\)) parameters can also be modeled hierarchically.\nThe implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying rho to be an array.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  array[D] real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\n\n\n\nFormulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.\nPerhaps most importantly, the parameters \\(\\rho\\) and \\(\\alpha\\) are weakly identified (Zhang 2004). The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.\n\n\nGPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.\nWe may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.\nIf our model consists of only the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\} \\\\\n    & x \\in \\mathbb{R}^{N \\times D}, \\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, inv_gamma_lpdf in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.2.\nIf we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\big(0, K(x \\mid \\alpha, \\rho)\\big) \\\\\ny_i &\\sim \\textsf{normal}\\left(\\beta_0 + x_i \\beta_{[1:D]} + f_i, \\sigma\\right) \\, \\forall i\n  \\in \\{1, \\dots, N\\} \\\\\n    & x_i^T, \\beta_{[1:D]} \\in \\mathbb{R}^D,\\quad\n      x \\in \\mathbb{R}^{N \\times D},\\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: \\[\\begin{align*}\nf(x \\mid a, b, p) &= \\dfrac{\\left(a/b\\right)^{p/2}}{2K_p\\left(\\sqrt{ab}\\right)} x^{p - 1}\\exp\\big(-(ax + b\n  / x)/2\\big) \\\\\n  & x, a, b \\in \\mathbb{R}^{+},\\quad\n    p \\in \\mathbb{Z}\n\\end{align*}\\]\nwhich has an inverse gamma left tail if \\(p \\leq 0\\) and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:\nfunctions {\n  real generalized_inverse_gaussian_lpdf(real x, int p,\n                                        real a, real b) {\n    return p * 0.5 * log(a / b)\n      - log(2 * modified_bessel_second_kind(p, sqrt(a * b)))\n      + (p - 1) * log(x)\n      - (a * x + b / x) * 0.5;\n }\n}\ndata {\n  // ...\n}\nIf we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of \\(\\rho\\) in a fixed-domain \\([0,T]\\), we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is \\(T / \\pi\n\\rho\\). You can see that as \\(\\rho\\) decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.\n\n\n\nThe parameter \\(\\alpha\\) corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-\\(t\\) prior on \\(\\alpha\\).\nA half-\\(t\\) or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.\n\n\n\n\nSuppose for a given sequence of inputs \\(x\\) that the corresponding outputs \\(y\\) are observed. Given a new sequence of inputs \\(\\tilde{x}\\), the posterior predictive distribution of their labels is computed by sampling outputs \\(\\tilde{y}\\) according to \\[\np\\left(\\tilde{y} \\mid \\tilde{x},x,y\\right)\n\\ = \\\n\\frac{p\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right)}\n     {p(y \\mid x)}\n\\ \\propto \\\np\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right).\n\\]\nA direct implementation in Stan defines a model in terms of the joint distribution of the observed \\(y\\) and unobserved \\(\\tilde{y}\\).\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y1 ~ normal(f[1:N1], sigma);\n}\ngenerated quantities {\n  vector[N2] y2;\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f[N1 + n2], sigma);\n  }\n}\nThe input vectors x1 and x2 are declared as data, as is the observed output vector y1. The unknown output vector y2, which corresponds to input vector x2, is declared in the generated quantities block and will be sampled when the model is executed.\nA transformed data block is used to combine the input vectors x1 and x2 into a single vector x.\nThe model block declares and defines a local variable for the combined output vector f, which consists of the concatenation of the conditional mean for known outputs y1 and unknown outputs y2. Thus the combined output vector f is aligned with the combined input vector x. All that is left is to define the univariate normal distribution statement for y.\nThe generated quantities block defines the quantity y2. We generate y2 by randomly generating N2 values from univariate normals with each mean corresponding to the appropriate element in f.\n\n\nWe can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.\nConsider the following full model for prediction using logistic Gaussian process regression.\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  array[N1] int&lt;lower=0, upper=1&gt; z1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  z1 ~ bernoulli_logit(a + f[1:N1]);\n}\ngenerated quantities {\n  array[N2] int z2;\n  for (n2 in 1:N2) {\n    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);\n  }\n}\n\n\n\nBayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.\nJumping straight to the result, \\[\np\\left(\\tilde{y} \\mid \\tilde{x},y,x\\right)\n=\n\\textsf{normal}\\left(K^{\\top}\\Sigma^{-1}y,\\\n                \\Omega - K^{\\top}\\Sigma^{-1}K\\right),\n\\] where \\(\\Sigma = K(x \\mid \\alpha, \\rho, \\sigma)\\) is the result of applying the covariance function to the inputs \\(x\\) with observed outputs \\(y\\), \\(\\Omega =\nK(\\tilde{x} \\mid \\alpha, \\rho)\\) is the result of applying the covariance function to the inputs \\(\\tilde{x}\\) for which predictions are to be inferred, and \\(K\\) is the matrix of covariances between inputs \\(x\\) and \\(\\tilde{x}\\), which in the case of the exponentiated quadratic covariance function would be \\[\nK(x \\mid \\alpha, \\rho)_{i, j} = \\alpha^2 \\exp\\left(-\\dfrac{1}{2 \\rho^2}\n\\sum_{d=1}^D \\left(x_{i,d} - \\tilde{x}_{j,d}\\right)^2\\right).\n\\]\nThere is no noise term including \\(\\sigma^2\\) because the indexes of elements in \\(x\\) and \\(\\tilde{x}\\) are never the same.\nThis Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called gp_pred_rng which will generate a draw from the posterior predictive mean conditioned on observed data y1. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of \\(p(\\tilde{y})\\).\nfunctions {\n  vector gp_pred_rng(array[] real x2,\n                     vector y1,\n                     array[] real x1,\n                     real alpha,\n                     real rho,\n                     real sigma,\n                     real delta) {\n    int N1 = rows(y1);\n    int N2 = size(x2);\n    vector[N2] f2;\n    {\n      matrix[N1, N1] L_K;\n      vector[N1] K_div_y1;\n      matrix[N1, N2] k_x1_x2;\n      matrix[N1, N2] v_pred;\n      vector[N2] f2_mu;\n      matrix[N2, N2] cov_f2;\n      matrix[N2, N2] diag_delta;\n      matrix[N1, N1] K;\n      K = gp_exp_quad_cov(x1, alpha, rho);\n      for (n in 1:N1) {\n        K[n, n] = K[n, n] + square(sigma);\n      }\n      L_K = cholesky_decompose(K);\n      K_div_y1 = mdivide_left_tri_low(L_K, y1);\n      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';\n      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);\n      f2_mu = (k_x1_x2' * K_div_y1);\n      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);\n      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;\n      diag_delta = diag_matrix(rep_vector(delta, N2));\n\n      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);\n    }\n    return f2;\n  }\n}\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  vector[N1] mu = rep_vector(0, N1);\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N1, N1] L_K;\n  {\n    matrix[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);\n    real sq_sigma = square(sigma);\n\n    // diagonal elements\n    for (n1 in 1:N1) {\n      K[n1, n1] = K[n1, n1] + sq_sigma;\n    }\n\n    L_K = cholesky_decompose(K);\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y1 ~ multi_normal_cholesky(mu, L_K);\n}\ngenerated quantities {\n  vector[N2] f2;\n  vector[N2] y2;\n\n  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f2[n2], sigma);\n  }\n}\n\n\n\n\nSuppose we have observations \\(y_i \\in \\mathbb{R}^M\\) observed at \\(x_i \\in \\mathbb{R}^K\\). One can model the data like so: \\[\\begin{align*}\ny_i  &\\sim \\textsf{multivariate normal}\\left(f(x_i), \\textbf{I}_M \\sigma^2\\right) \\\\\nf(x) &\\sim \\textsf{GP}\\big(m(x), K(x \\mid \\theta, \\phi)\\big) \\\\\n     & K(x \\mid \\theta) \\in \\mathbb{R}^{M \\times M}, \\quad\n       f(x), m(x) \\in \\mathbb{R}^M\n\\end{align*}\\] where the \\(K(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]}\\) entry defines the covariance between \\(f_m(x)\\) and \\(f_{m^\\prime}(x^\\prime)(x)\\). This construction of Gaussian processes allows us to learn the covariance between the output dimensions of \\(f(x)\\). If we parameterize our kernel \\(K\\): \\[\nK(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]} = k\\left(x, x^\\prime \\mid\n\\theta\\right) k\\left(m, m^\\prime \\mid \\phi\\right)\n\\] then our finite dimensional generative model for the above is: \\[\\begin{align*}\nf        &\\sim \\textsf{matrixnormal}\\big(m(x), K(x \\mid \\alpha, \\rho), C(\\phi)\\big) \\\\\ny_{i, m} &\\sim \\textsf{normal}(f_{i,m}, \\sigma) \\\\\nf        &\\in  \\mathbb{R}^{N \\times M}\n\\end{align*}\\] where \\(K(x \\mid \\alpha, \\rho)\\) is the exponentiated quadratic kernel we’ve used throughout this chapter, and \\(C(\\phi)\\) is a positive-definite matrix, parameterized by some vector \\(\\phi\\).\nThe matrix normal distribution has two covariance matrices: \\(K(x \\mid\n\\alpha, \\rho)\\) to encode column covariance, and \\(C(\\phi)\\) to define row covariance. The salient features of the matrix normal are that the rows of the matrix \\(f\\) are distributed: \\[\nf_{[n,]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[n,]}, K(x \\mid \\alpha,\n\\rho)_{[n,n]} C(\\phi)\\big)\n\\] and that the columns of the matrix \\(f\\) are distributed: \\[\nf_{[,m]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[,m]}, K(x\n  \\mid \\alpha, \\rho) C(\\phi)_{[m,m]}\\big)\n\\] This also means means that \\(\\mathbb{E}\\left[f^T f\\right]\\) is equal to \\(\\operatorname{trace}\\!\\big(K(x \\mid \\alpha, \\rho)\\big) \\times C\\), whereas \\(\\mathbb{E}\\left[ff^T\\right]\\) is \\(\\operatorname{trace}(C) \\times K(x \\mid \\alpha, \\rho)\\). We can derive this using properties of expectation and the matrix normal density.\nWe should set \\(\\alpha\\) to \\(1.0\\) because the parameter is not identified unless we constrain \\(\\operatorname{trace}(C) = 1\\). Otherwise, we can multiply \\(\\alpha\\) by a scalar \\(d\\) and \\(C\\) by \\(1/d\\) and our likelihood will not change.\nWe can generate a random variable \\(f\\) from a matrix normal density in \\(\\mathbb{R}^{N \\times M}\\) using the following algorithm: \\[\\begin{align*}\n\\eta_{i,j} &\\sim \\textsf{normal}(0, 1) \\, \\forall i,j \\\\\nf          &= L_{K(x \\mid 1.0, \\rho)} \\, \\eta \\, L_C(\\phi)^T \\\\\nf          &\\sim \\textsf{matrixnormal}\\big(0, K(x \\mid 1.0, \\rho), C(\\phi)\\big) \\\\\n\\eta       &\\in \\mathbb{R}^{N \\times M} \\\\\nL_C(\\phi)  &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(C(\\phi)\\big) \\\\\nL_{K(x \\mid 1.0, \\rho)} &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(K(x \\mid 1.0, \\rho)\\big)\n\\end{align*}\\]\nThis can be implemented in Stan using a latent-variable GP formulation. We’ve used \\(\\textsf{LKJCorr}\\) for \\(C(\\phi)\\), but any positive-definite matrix will do.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] real x;\n  matrix[N, D] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  vector&lt;lower=0&gt;[D] alpha;\n  real&lt;lower=0&gt; sigma;\n  cholesky_factor_corr[D] L_Omega;\n  matrix[N, D] eta;\n}\nmodel {\n  matrix[N, D] f;\n  {\n    matrix[N, N] K = gp_exp_quad_cov(x, 1.0, rho);\n    matrix[N, N] L_K;\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta\n        * diag_pre_multiply(alpha, L_Omega)';\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(3);\n  to_vector(eta) ~ std_normal();\n\n  to_vector(y) ~ normal(to_vector(f), sigma);\n}\ngenerated quantities {\n  matrix[D, D] Omega;\n  Omega = L_Omega * L_Omega';\n}",
+    "text": "Consider a regression with a single predictor \\(x_n\\) for the training outcome \\(y_n\\) and \\(\\tilde{x}_n\\) for the test outcome \\(\\tilde{y}_n.\\) Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in \\[\\begin{eqnarray}\np(\\tilde{y} \\mid \\tilde{x}, y, x)\n& = & \\int p(\\tilde{y} \\mid \\tilde{x}, \\theta) \\cdot p(\\theta \\mid y, x) \\,\n\\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m=1}^M \\, p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\end{eqnarray}\\] where \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y).\\)\n\n\n\nThe following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed \\(y\\) values are coded as data. The predictive quantities \\(\\tilde{y}\\) appear in the generated quantities block, where they are generated by random number generation.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0&gt; y;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ poisson_log(alpha + beta * x);\n  { alpha, beta } ~ normal(0, 1);\n}\ngenerated quantities {\n  array[N_tilde] int&lt;lower=0&gt; y_tilde\n    = poisson_log_rng(alpha + beta * x_tilde);\n}\nThe Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s poisson_log vs. poisson, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept alpha and slope beta, are given standard normal priors.\nIn the model block, the log rate for the Poisson is a linear function of the training data \\(x\\), whereas in the generated quantities block it is a function of the test data \\(\\tilde{x}\\). Because the generated quantities block does not affect the posterior draws, the model fits \\(\\alpha\\) and \\(\\beta\\) using only the training data, reserving \\(\\tilde{x}\\) to generate \\(\\tilde{y}.\\)\nThe result from running Stan is a predictive sample \\(\\tilde{y}^{(1)},\n\\ldots \\tilde{y}^{(M)}\\) where each \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid \\tilde{x}, x, y).\\)\nThe mean of the posterior predictive distribution is the expected value \\[\\begin{align}\n\\mathbb{E}[\\tilde{y} \\mid \\tilde{x}, x, y]\n& =\n\\int\n\\tilde{y}\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n\\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}^{(m)},\n\\end{align}\\] where the \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y)\\) are drawn from the posterior predictive distribution. Thus the posterior mean of y_tilde[n] after running Stan is the expected value of \\(\\tilde{y}_n\\) conditioned on the training data \\(x, y\\) and predictor \\(\\tilde{x}_n.\\) This is the Bayesian estimate for \\(\\tilde{y}\\) with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for \\(\\tilde{y}\\), as well as covariance of the \\(\\tilde{y_n}.\\) The posterior draws \\(\\tilde{y}^{(m)}\\) may also be used to estimate predictive event probabilities, such as \\(\\Pr[\\tilde{y}_1 &gt; 0]\\) or \\(\\Pr[\\prod_{n =\n1}^{\\tilde{N}}(\\tilde{y_n}) &gt; 1],\\) as expectations of indicator functions.\nAll of this can be carried out by running Stan only a single time to draw a single sample of \\(M\\) draws, \\[\n\\tilde{y}^{(1)}, \\ldots, \\tilde{y}^{(M)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\n\\] It’s only when moving to cross-validation where multiple runs are required.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Gaussian Processes"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/gaussian-processes.html#footnotes",
-    "href": "stan-users-guide/gaussian-processes.html#footnotes",
-    "title": "Gaussian Processes",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nGaussian processes can be extended to covariance functions producing positive semi-definite matrices, but Stan does not support inference in the resulting models because the resulting distribution does not have unconstrained support.↩︎\nA boundary-avoiding prior is just one where the limit of the density is zero at the boundary, the result of which is estimates that are pushed away from the boundary.↩︎",
+    "objectID": "stan-users-guide/posterior-prediction.html#estimating-event-probabilities",
+    "href": "stan-users-guide/posterior-prediction.html#estimating-event-probabilities",
+    "title": "Posterior Predictive Sampling",
+    "section": "",
+    "text": "Event probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate \\(\\Pr[\\lambda &gt; 5 \\mid y]\\) in the simple Poisson example with only a rate parameter \\(\\lambda\\), it suffices to define a generated quantity\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; lambda_gt_5 = lambda &gt; 5;\n  // ...\n}\nThe value of the expression lambda &gt; 5 is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = &\n\\int \\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}[\\lambda^{(m)} &gt; 5],\n\\end{eqnarray*}\\] where each \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter lambda_gt_5.\nIn general, event probabilities may be expressed as expectations of indicator functions. For example, \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = & \\mathbb{E}[\\textrm{I}[\\lambda &gt; 5] \\mid y]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}(\\lambda^{(m)} &gt; 5).\n\\end{eqnarray*}\\] The last line above is the posterior mean of the indicator function as coded in Stan.\nEvent probabilities involving posterior predictive quantities \\(\\tilde{y}\\) work exactly the same way as those for parameters. For example, if \\(\\tilde{y}_n\\) is the prediction for the \\(n\\)-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then \\[\\begin{eqnarray*}\n\\Pr[\\tilde{y}_3 &gt; \\tilde{y}_7 \\mid \\tilde{x}, x, y]\n& = &\n\\mathbb{E}\\!\\left[I[\\tilde{y}_3 &gt; \\tilde{y}_7] \\mid \\tilde{x}, x, y\\right]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\tilde{y}_3 &gt; \\tilde{y}_7)\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\, \\textrm{d}\\tilde{y}\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n\\textrm{I}(\\tilde{y}^{(m)}_3 &gt; \\tilde{y}^{(m)}_7),\n\\end{eqnarray*}\\] where \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y).\\)",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Gaussian Processes"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html",
-    "href": "stan-users-guide/floating-point.html",
-    "title": "Floating Point Arithmetic",
+    "objectID": "stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction",
+    "href": "stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction",
+    "title": "Posterior Predictive Sampling",
     "section": "",
-    "text": "Computers approximate real values in \\(\\mathbb{R}\\) using a fixed number of bits. This chapter explains how this is done and why it is important for writing robust Stan (and other numerical) programs. The subfield of computer science devoted to studying how real arithmetic works on computers is called numerical analysis.\n\n\nStan’s arithmetic is implemented using double-precision arithmetic. The behavior of most1 modern computers follows the floating-point arithmetic, IEEE Standard for Floating-Point Arithmetic (IEEE 754).\n\n\nThe double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:\n\nsignificand (53 bits): base value representing significant digits\nexponent (11 bits): power of two multiplied by the base\n\nThe value of a finite floating point number is\n\\[\nv = (-1)^s \\times c \\, 2^q\n\\]\n\n\n\nA normal floating-point value does not use any leading zeros in its significand; subnormal numbers may use leading zeros. Not all I/O systems support subnormal numbers.\n\n\n\nThere are some reserved exponent values so that legal exponent values range between\\(-(2^{10}) + 2 = -1022\\) and \\(2^{10} - 1 = 1023\\). Legal significand values are between \\(-2^{52}\\) and \\(2^{52} - 1\\). Floating point allows the representation of both really big and really small values. Some extreme values are\n\nlargest normal finite number: \\(\\approx 1.8 \\times 10^{308}\\)\nlargest subnormal finite number: \\(\\approx 2.2 \\times 10^{308}\\)\nsmallest positive normal number: \\(\\approx 2.2 \\times 10^{-308}\\)\nsmallest positive subnormal number: \\(\\approx 4.9 \\times 10^{-324}\\)\n\n\n\n\nBecause of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., 0 == -0 evaluates to true).\n\n\n\nA specially chosen bit pattern is used for the not-a-number value (often written as NaN in programming language output, including Stan’s).\nStan provides a value function not_a_number() that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.\nStan also provides a test function is_nan(x) that returns 1 if x is not-a-number and 0 otherwise.\nNot-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as log and exp have the same behavior, propagating not-a-number values.\nMost of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.\nComparisons with not-a-number always return false, up to and including comparison with itself. That is, not_a_number() == not_a_number() somewhat confusingly returns false. That is why there is a built-in is_nan() function in Stan (and in C++). The only exception is negation, which remains coherent. This means not_a_number() != not_a_number() returns true.\nUndefined operations often return not-a-number values. For example, sqrt(-1) will evaluate to not-a-number.\n\n\n\nThere are also two special values representing positive infinity (\\(\\infty)\\) and negative infinity (\\(-\\infty\\)). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, log(0) evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that 0 == exp(log(0)). Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.\nThere are value functions positive_infinity() and negative_infinity() as well as a test function is_inf().\nPositive and negative infinity have the expected comparison behavior, so that negative_infinity() &lt; 0 evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.\nPositive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.\n\n\n\n\nIn programming languages such as Stan, numbers may be represented in standard decimal (base 10) notation. For example, 2.39 or -1567846.276452. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using scientific notation, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, 36.29e-3 represents the number \\(36.29 \\times\n10^{-3}\\), which is the same number as is represented by 0.03629.\n\n\n\nThe choice of significand provides \\(\\log_{10} 2^{53} \\approx 15.95\\) decimal (base 10) digits of arithmetic precision. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.\n\n\nIn practice, the finite amount of arithmetic precision leads to rounding, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,\n1 + 1e-20 == 1\nThe closest floating point number to \\(1 + 10^{-20}\\) turns out to be \\(1\\) itself. By contrast,\n0 + 1e-20 == 1e-20\nThis highlights the fact that precision depends on scale. Even though 1 + 1e-20 == 1, we have 1e-20 + 1e-20 == 2e-20, as expected.\nRounding also manifests itself in a lack of transitivity. In particular, it does not usually hold for three floating point numbers \\(a, b, c\\) that \\((a + b) + c = a + (b + c)\\).\nIn statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix \\(L\\) with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix \\(L \\, L^{\\top}\\). In practice, rounding and loss of precision may render the result \\(L \\, L^{\\top}\\) neither symmetric nor positive definite.\nIn practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding \\(L \\, L^{\\top}\\) with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.\n\n\n\nThe smallest number greater than zero is roughly \\(0 + 10^{-323}\\). The largest number less than one is roughly \\(1 - 10^{-15.95}\\). The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of \\(0.9999999999999999\\).\nFor this reason, the machine precision is said to be roughly \\(10^{-15.95}\\). This constant is available as machine_precision() in Stan.\n\n\n\nSpecial operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation log(1 + x) for positive x. When x is small (less than \\(10^{-16}\\) for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing \\(f(x) = \\log (1 + x)\\). In Stan (as in C++), this operation is written as log1p(x). Because x itself may be close to zero, the function log1p(x) can take the logarithm of values very close to one, the results of which are close to zero.\nSimilarly, the complementary cumulative distribution functions (CCDF), defined by \\(F^{\\complement}_Y(y) = 1 - F_Y(y)\\), where \\(F_Y\\) is the cumulative distribution function (CDF) for the random variable \\(Y\\). This allows values very close to one to be represented in complementary form.\n\n\n\nAnother downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider \\[\\begin{align*}\n  1&.23456789012345 \\\\\n- 1&.23456789012344 \\\\\n= 0&.00000000000001\n\\end{align*}\\] We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.\nCatastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses Welford’s algorithm for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.\n\n\n\nEven though 1e200 may be represented as a double precision floating point value, there is no finite value large enough to represent 1e200 * 1e200. The result of 1e200 * 1e200 is said to overflow. The IEEE 754 standard requires the result to be positive infinity.\nOverflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.\n\n\n\nWhen there is no number small enough to represent a result, it is said to underflow. For instance, 1e-200 may be represented, but 1e-200 * 1e-200 underflows so that the result is zero.\nUnderflow is a ubiquitous problem in likelihood calculations, For example, if \\(p(y_n \\mid \\theta) &lt; 0.1\\), then \\[\np(y \\mid \\theta) = \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\] will underflow as soon as \\(N &gt; 350\\) or so.\nTo deal with underflow, work on the log scale. Even though \\(p(y \\mid\n\\theta)\\) can’t be represented, there is no problem representing \\[\n\\begin{array}{rcl}\n\\log p(y \\mid \\theta)\n& = & \\log \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\\\[4pt]\n& = & \\sum_{n = 1}^N \\log p(y_n \\mid \\theta)\n\\end{array}\n\\]\nThis is why all of Stan’s probability functions operate on the log scale.\n\n\n\n\nWorking on the log scale, multiplication is converted to addition, \\[\n\\log (a \\cdot b) = \\log a + \\log b.\n\\] Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given \\(\\log a\\) and \\(\\log b\\), how do we get \\(\\log (a + b)\\)? Working out the algebra, \\[\n\\log (a + b)\n=\n\\log (\\exp(\\log a) + \\exp(\\log b)).\n\\]\n\n\nThe nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\log (\\exp(u) + \\exp(v)).\n\\] so that \\[\n\\log (a + b)\n=\n\\textrm{log-sum-exp}(\\log a, \\log b).\n\\]\nAlthough it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\max(u, v) + \\log\\big( \\exp(u - \\max(u, v)) + \\exp(v - \\max(u, v)) \\big).\n\\]\nBecause the terms inside the exponentiations are \\(u - \\max(u, v)\\) and \\(v - \\max(u, v)\\), one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that \\(u \\geq v\\), so that \\[\n\\textrm{log-sum-exp}(u, v) = u + \\log\\big(1 + \\exp(v - u)\\big).\n\\]\nAlthough the inner term may itself be evaluated using the built-in function log1p, there is only limited gain because \\(\\exp(v - u)\\) is only near zero when \\(u\\) is much larger than \\(v\\), meaning the final result is likely to round to \\(u\\) anyway.\nTo conclude, when evaluating \\(\\log (a + b)\\) given \\(\\log a\\) and \\(\\log\nb\\), and assuming \\(\\log a &gt; \\log b\\), return\n\\[\n\\log (a + b) =\n\\log a + \\textrm{log1p}\\big(\\exp(\\log b - \\log a)\\big).\n\\]\n\n\n\nThe log sum of exponentials function may be generalized to sequences in the obvious way, so that if \\(v = v_1, \\ldots, v_N\\), then \\[\\begin{eqnarray*}\n\\textrm{log-sum-exp}(v)\n& = & \\log \\sum_{n = 1}^N \\exp(v_n)\n\\\\[4pt]\n& = & \\max(v) + \\log \\sum_{n = 1}^N \\exp(v_n - \\max(v)).\n\\end{eqnarray*}\\] The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate \\(\\log (u_1 + \\cdots + u_N)\\) given only \\(\\log u_n\\).\n\n\n\nAn immediate application is to computing the mean of a vector \\(u\\) entirely on the log scale. That is, given \\(\\log u\\) and returning \\(\\log \\textrm{mean}(u)\\). \\[\\begin{eqnarray*}\n\\log \\left( \\frac{1}{N} \\sum_{n = 1}^N u_n \\right)\n& = & \\log \\frac{1}{N} + \\log \\sum_{n = 1}^N \\exp(\\log u_n)\n\\\\[4pt]\n& = & -\\log N + \\textrm{log-sum-exp}(\\log u).\n\\end{eqnarray*}\\] where \\(\\log u = (\\log u_1, \\ldots, \\log u_N)\\) is understood elementwise.\n\n\n\n\nBecause floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing x == y, an approximate test may be used given an absolute or relative tolerance.\nGiven a positive absolute tolerance of epsilon, x can be compared to y using the conditional\nabs(x - y) &lt;= epsilon.\nAbsolute tolerances work when the scale of x and y and the relevant comparison is known.\nGiven a positive relative tolerance of epsilon, a typical comparison is\n2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.",
+    "text": "Stan’s sampling algorithms take a Stan program representing a posterior \\(p(\\theta \\mid y, x)\\) along with actual data \\(x\\) and \\(y\\) to produce a set of draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)}\\) from the posterior. Posterior predictive draws \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y)\\) can be generated by drawing \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\tilde{x}, \\theta^{(m)})\n\\] from the data model. Note that drawing \\(\\tilde{y}^{(m)}\\) only depends on the new predictors \\(\\tilde{x}\\) and the posterior draws \\(\\theta^{(m)}\\). Most importantly, neither the original data or the model density is required.\nBy saving the posterior draws, predictions for new data items \\(\\tilde{x}\\) may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 5);\n  beta ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.5);\n}\nA second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde\n    = normal_rng(alpha + beta * x_tilde, sigma);\n}\nWhen running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for \\(\\tilde{y}\\) or derived quantities such as event probabilities.\nAny posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Posterior Inference & Model Checking",
+      "Posterior Predictive Sampling"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html#floating-point-representations",
-    "href": "stan-users-guide/floating-point.html#floating-point-representations",
-    "title": "Floating Point Arithmetic",
+    "objectID": "stan-users-guide/one-dimensional-integrals.html",
+    "href": "stan-users-guide/one-dimensional-integrals.html",
+    "title": "Computing One Dimensional Integrals",
     "section": "",
-    "text": "Stan’s arithmetic is implemented using double-precision arithmetic. The behavior of most1 modern computers follows the floating-point arithmetic, IEEE Standard for Floating-Point Arithmetic (IEEE 754).\n\n\nThe double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:\n\nsignificand (53 bits): base value representing significant digits\nexponent (11 bits): power of two multiplied by the base\n\nThe value of a finite floating point number is\n\\[\nv = (-1)^s \\times c \\, 2^q\n\\]\n\n\n\nA normal floating-point value does not use any leading zeros in its significand; subnormal numbers may use leading zeros. Not all I/O systems support subnormal numbers.\n\n\n\nThere are some reserved exponent values so that legal exponent values range between\\(-(2^{10}) + 2 = -1022\\) and \\(2^{10} - 1 = 1023\\). Legal significand values are between \\(-2^{52}\\) and \\(2^{52} - 1\\). Floating point allows the representation of both really big and really small values. Some extreme values are\n\nlargest normal finite number: \\(\\approx 1.8 \\times 10^{308}\\)\nlargest subnormal finite number: \\(\\approx 2.2 \\times 10^{308}\\)\nsmallest positive normal number: \\(\\approx 2.2 \\times 10^{-308}\\)\nsmallest positive subnormal number: \\(\\approx 4.9 \\times 10^{-324}\\)\n\n\n\n\nBecause of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., 0 == -0 evaluates to true).\n\n\n\nA specially chosen bit pattern is used for the not-a-number value (often written as NaN in programming language output, including Stan’s).\nStan provides a value function not_a_number() that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.\nStan also provides a test function is_nan(x) that returns 1 if x is not-a-number and 0 otherwise.\nNot-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as log and exp have the same behavior, propagating not-a-number values.\nMost of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.\nComparisons with not-a-number always return false, up to and including comparison with itself. That is, not_a_number() == not_a_number() somewhat confusingly returns false. That is why there is a built-in is_nan() function in Stan (and in C++). The only exception is negation, which remains coherent. This means not_a_number() != not_a_number() returns true.\nUndefined operations often return not-a-number values. For example, sqrt(-1) will evaluate to not-a-number.\n\n\n\nThere are also two special values representing positive infinity (\\(\\infty)\\) and negative infinity (\\(-\\infty\\)). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, log(0) evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that 0 == exp(log(0)). Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.\nThere are value functions positive_infinity() and negative_infinity() as well as a test function is_inf().\nPositive and negative infinity have the expected comparison behavior, so that negative_infinity() &lt; 0 evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.\nPositive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.",
+    "text": "Definite and indefinite one dimensional integrals can be performed in Stan using the integrate_1d function\nAs an example, the normalizing constant of a left-truncated normal distribution is\n\\[\n  \\int_a^\\infty \\frac{1}{\\sqrt{2 \\pi \\sigma^2}} e^{-\\frac{1}{2}\\frac{(x - \\mu)^2}{\\sigma^2}}\n\\]\nTo compute this integral in Stan, the integrand must first be defined as a Stan function (see the Stan Reference Manual chapter on User-Defined Functions for more information on coding user-defined functions).\nreal normal_density(real x,             // Function argument\n                    real xc,            // Complement of function argument\n                                        //  on the domain (defined later)\n                    array[] real theta, // parameters\n                    array[] real x_r,   // data (real)\n                    array[] int x_i) {  // data (integer)\n  real mu = theta[1];\n  real sigma = theta[2];\n\n  return 1 / (sqrt(2 * pi()) * sigma) * exp(-0.5 * ((x - mu) / sigma)^2);\n}\nThis function is expected to return the value of the integrand evaluated at point x. The argument xc is used in definite integrals to avoid loss of precision near the limits of integration and is set to NaN when either limit is infinite (see the section on precision/loss in the chapter on Higher-Order Functions of the Stan Functions Reference for details on how to use this). The argument theta is used to pass in arguments of the integral that are a function of the parameters in our model. The arguments x_r and x_i are used to pass in real and integer arguments of the integral that are not a function of our parameters.\nThe function defining the integrand must have exactly the argument types and return type of normal_density above, though argument naming is not important. Even if x_r and x_i are unused in the integrand, they must be included in the function signature. Even if the integral does not involve some of these, they must still be supplied some value. The most efficient will be a zero-length array or vector, which can be created with rep_array(0, 0) and rep_vector(0, 0), respectively. Other options include an uninitialized variable declared with size 0, which is equivalent to the above, or any easy value, such as size 1 array created with {0}.\n\n\nSuppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.\nSuch a model might look like (include the function defined at the beginning of this chapter to make this code compile):\ndata {\n  int N;\n  array[N] real y;\n}\n\ntransformed data {\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\nparameters {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n  real left_limit;\n}\n\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  left_limit ~ normal(0, 1);\n  target += normal_lpdf(y | mu, sigma);\n  target += -log(integrate_1d(normal_density,\n                              left_limit,\n                              positive_infinity(),\n                              { mu, sigma }, x_r, x_i));\n}\n\n\nThe limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls negative_infinity() and positive_infinity().\nIf both limits are either negative_infinity() or positive_infinity(), the integral and its gradients are set to zero.\n\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.\nThe endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).\n\n\n\n\nThe integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library (Agrawal et al. 2017). If the \\(n\\)th estimate of the integral is denoted \\(I_n\\) and the \\(n\\)th estimate of the norm of the integral is denoted \\(|I|_n\\), the iteration is terminated when\n\\[\n  \\frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \\text{relative tolerance}.\n\\]\nThe relative_tolerance parameter can be optionally specified as the last argument to integrate_1d. By default, integrate_1d follows the Boost library recommendation of setting relative_tolerance to the square root of the machine epsilon of double precision floating point numbers (about 1e-8). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If integrate_1d causes an exception in transformed parameters block or model block, the result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If integrate_1d causes an exception in generated quantities block, the returned output from integrate_1d is NaN. In these cases, a bigger relative_tolerance value can be specified.\n\n\nIntegrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). This is because the quadrature methods employed internally can have difficulty near zero.\nIn this case, each integral is separately integrated to the given relative_tolerance.\n\n\n\nIf care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.\nFor instance, in integrating the pdf of a beta distribution when the values of \\(\\alpha\\) and \\(\\beta\\) are small, most of the probability mass is lumped near zero and one.\nThe pdf of a beta distribution is proportional to\n\\[\np(x) \\propto x^{\\alpha - 1}(1 - x)^{\\beta - 1}\n\\]\nNormalizing this distribution requires computing the integral of \\(p(x)\\) from zero to one. In Stan code, the integrand might look like:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n\n  return x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n}\nThe issue is that there will be numerical breakdown in the precision of 1.0 - x as x gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of alpha and beta much less than one.\nThis is where xc is useful. It is defined, for definite integrals, as a high precision version of the distance from x to the nearest endpoint — a - x or b - x for a lower endpoint a and an upper endpoint b. To make use of this for the beta integral, the integrand can be re-coded:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n  real v;\n\n  if(x &gt; 0.5) {\n    v = x^(alpha - 1.0) * xc^(beta - 1.0);\n  } else {\n    v = x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n  }\n\n  return v;\n}\nIn this case, as we approach the upper limit of integration \\(a = 1\\), xc will take on the value of \\(a - x = 1 - x\\). This version of the integrand will converge for much smaller values of alpha and beta than otherwise possible.\nConsider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount \\(\\delta\\), and truncated at some value \\(b\\). If we were interested in calculating the expectation of a variable \\(X\\) distributed in this way, we would need to calculate \\[\n\\int_a^b xf(x)\\,dx = \\int_{\\delta}^b xf(x)\\,dx\n\\] in the numerator, where \\(f(x)\\) is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:\nreal shift_lognormal_pdf(real x,\n                         real mu,\n                         real sigma,\n                         real delta) {\n  real p;\n\n  p = (1.0 / ((x - delta) * sigma * sqrt(2 * pi()))) *\n    exp(-1 * (log(x - delta) - mu)^2 / (2 * sigma^2));\n\n  return p;\n}\nTherefore, the function that we want to integrate is:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  p = shift_lognormal_pdf(x, mu, sigma, delta);\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhat happens here is that, given that the log-normal distribution is shifted by \\(\\delta\\), when we then try to integrate the numerator, our x starts at values just above delta. This, in turn, causes the x - delta term to be near zero, leading to a breakdown.\nWe can use xc, and define the integrand as:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  if (x &lt; delta + 1) {\n    p = shift_lognormal_pdf(xc, mu, sigma, delta);\n  } else {\n    p = shift_lognormal_pdf(x, mu, sigma, delta);\n  }\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhy does this work? When our values of x are less than delta + 1 (so, when they’re near delta, given that our lower bound of integration is equal to \\(\\delta\\)), we pass xc as an argument to our shift_lognormal_pdf function. This way, instead of dealing with x - delta in shift_lognormal_pdf, we are working with xc - delta which is equal to delta - x - delta, as delta is the lower endpoint in that case. The delta terms cancel out, and we are left with a high-precision version of x. We don’t encounter the same problem at the upper limit \\(b\\) so we don’t adjust the code for that case.\nNote, xc is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, xc will be NaN.\nFor zero-crossing definite integrals (see section Zero Crossing) the integrals are broken into two pieces (\\((a, 0)\\) and \\((0, b)\\) for endpoints \\(a &lt; 0\\) and \\(b &gt; 0\\)) and xc is a high precision version of the distance to the limits of each of the two integrals separately. This means xc will be a high precision version of a - x, x, or b - x, depending on the value of x and the endpoints.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Example Models",
+      "Computing One Dimensional Integrals"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html#literals-decimal-and-scientific-notation",
-    "href": "stan-users-guide/floating-point.html#literals-decimal-and-scientific-notation",
-    "title": "Floating Point Arithmetic",
+    "objectID": "stan-users-guide/one-dimensional-integrals.html#calling-the-integrator",
+    "href": "stan-users-guide/one-dimensional-integrals.html#calling-the-integrator",
+    "title": "Computing One Dimensional Integrals",
     "section": "",
-    "text": "In programming languages such as Stan, numbers may be represented in standard decimal (base 10) notation. For example, 2.39 or -1567846.276452. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using scientific notation, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, 36.29e-3 represents the number \\(36.29 \\times\n10^{-3}\\), which is the same number as is represented by 0.03629.",
+    "text": "Suppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.\nSuch a model might look like (include the function defined at the beginning of this chapter to make this code compile):\ndata {\n  int N;\n  array[N] real y;\n}\n\ntransformed data {\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\nparameters {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n  real left_limit;\n}\n\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  left_limit ~ normal(0, 1);\n  target += normal_lpdf(y | mu, sigma);\n  target += -log(integrate_1d(normal_density,\n                              left_limit,\n                              positive_infinity(),\n                              { mu, sigma }, x_r, x_i));\n}\n\n\nThe limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls negative_infinity() and positive_infinity().\nIf both limits are either negative_infinity() or positive_infinity(), the integral and its gradients are set to zero.\n\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.\nThe endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Example Models",
+      "Computing One Dimensional Integrals"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html#arithmetic-precision",
-    "href": "stan-users-guide/floating-point.html#arithmetic-precision",
-    "title": "Floating Point Arithmetic",
+    "objectID": "stan-users-guide/one-dimensional-integrals.html#integrator-convergence",
+    "href": "stan-users-guide/one-dimensional-integrals.html#integrator-convergence",
+    "title": "Computing One Dimensional Integrals",
     "section": "",
-    "text": "The choice of significand provides \\(\\log_{10} 2^{53} \\approx 15.95\\) decimal (base 10) digits of arithmetic precision. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.\n\n\nIn practice, the finite amount of arithmetic precision leads to rounding, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,\n1 + 1e-20 == 1\nThe closest floating point number to \\(1 + 10^{-20}\\) turns out to be \\(1\\) itself. By contrast,\n0 + 1e-20 == 1e-20\nThis highlights the fact that precision depends on scale. Even though 1 + 1e-20 == 1, we have 1e-20 + 1e-20 == 2e-20, as expected.\nRounding also manifests itself in a lack of transitivity. In particular, it does not usually hold for three floating point numbers \\(a, b, c\\) that \\((a + b) + c = a + (b + c)\\).\nIn statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix \\(L\\) with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix \\(L \\, L^{\\top}\\). In practice, rounding and loss of precision may render the result \\(L \\, L^{\\top}\\) neither symmetric nor positive definite.\nIn practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding \\(L \\, L^{\\top}\\) with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.\n\n\n\nThe smallest number greater than zero is roughly \\(0 + 10^{-323}\\). The largest number less than one is roughly \\(1 - 10^{-15.95}\\). The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of \\(0.9999999999999999\\).\nFor this reason, the machine precision is said to be roughly \\(10^{-15.95}\\). This constant is available as machine_precision() in Stan.\n\n\n\nSpecial operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation log(1 + x) for positive x. When x is small (less than \\(10^{-16}\\) for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing \\(f(x) = \\log (1 + x)\\). In Stan (as in C++), this operation is written as log1p(x). Because x itself may be close to zero, the function log1p(x) can take the logarithm of values very close to one, the results of which are close to zero.\nSimilarly, the complementary cumulative distribution functions (CCDF), defined by \\(F^{\\complement}_Y(y) = 1 - F_Y(y)\\), where \\(F_Y\\) is the cumulative distribution function (CDF) for the random variable \\(Y\\). This allows values very close to one to be represented in complementary form.\n\n\n\nAnother downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider \\[\\begin{align*}\n  1&.23456789012345 \\\\\n- 1&.23456789012344 \\\\\n= 0&.00000000000001\n\\end{align*}\\] We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.\nCatastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses Welford’s algorithm for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.\n\n\n\nEven though 1e200 may be represented as a double precision floating point value, there is no finite value large enough to represent 1e200 * 1e200. The result of 1e200 * 1e200 is said to overflow. The IEEE 754 standard requires the result to be positive infinity.\nOverflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.\n\n\n\nWhen there is no number small enough to represent a result, it is said to underflow. For instance, 1e-200 may be represented, but 1e-200 * 1e-200 underflows so that the result is zero.\nUnderflow is a ubiquitous problem in likelihood calculations, For example, if \\(p(y_n \\mid \\theta) &lt; 0.1\\), then \\[\np(y \\mid \\theta) = \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\] will underflow as soon as \\(N &gt; 350\\) or so.\nTo deal with underflow, work on the log scale. Even though \\(p(y \\mid\n\\theta)\\) can’t be represented, there is no problem representing \\[\n\\begin{array}{rcl}\n\\log p(y \\mid \\theta)\n& = & \\log \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\\\[4pt]\n& = & \\sum_{n = 1}^N \\log p(y_n \\mid \\theta)\n\\end{array}\n\\]\nThis is why all of Stan’s probability functions operate on the log scale.",
+    "text": "The integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library (Agrawal et al. 2017). If the \\(n\\)th estimate of the integral is denoted \\(I_n\\) and the \\(n\\)th estimate of the norm of the integral is denoted \\(|I|_n\\), the iteration is terminated when\n\\[\n  \\frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \\text{relative tolerance}.\n\\]\nThe relative_tolerance parameter can be optionally specified as the last argument to integrate_1d. By default, integrate_1d follows the Boost library recommendation of setting relative_tolerance to the square root of the machine epsilon of double precision floating point numbers (about 1e-8). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If integrate_1d causes an exception in transformed parameters block or model block, the result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If integrate_1d causes an exception in generated quantities block, the returned output from integrate_1d is NaN. In these cases, a bigger relative_tolerance value can be specified.\n\n\nIntegrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). This is because the quadrature methods employed internally can have difficulty near zero.\nIn this case, each integral is separately integrated to the given relative_tolerance.\n\n\n\nIf care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.\nFor instance, in integrating the pdf of a beta distribution when the values of \\(\\alpha\\) and \\(\\beta\\) are small, most of the probability mass is lumped near zero and one.\nThe pdf of a beta distribution is proportional to\n\\[\np(x) \\propto x^{\\alpha - 1}(1 - x)^{\\beta - 1}\n\\]\nNormalizing this distribution requires computing the integral of \\(p(x)\\) from zero to one. In Stan code, the integrand might look like:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n\n  return x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n}\nThe issue is that there will be numerical breakdown in the precision of 1.0 - x as x gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of alpha and beta much less than one.\nThis is where xc is useful. It is defined, for definite integrals, as a high precision version of the distance from x to the nearest endpoint — a - x or b - x for a lower endpoint a and an upper endpoint b. To make use of this for the beta integral, the integrand can be re-coded:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n  real v;\n\n  if(x &gt; 0.5) {\n    v = x^(alpha - 1.0) * xc^(beta - 1.0);\n  } else {\n    v = x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n  }\n\n  return v;\n}\nIn this case, as we approach the upper limit of integration \\(a = 1\\), xc will take on the value of \\(a - x = 1 - x\\). This version of the integrand will converge for much smaller values of alpha and beta than otherwise possible.\nConsider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount \\(\\delta\\), and truncated at some value \\(b\\). If we were interested in calculating the expectation of a variable \\(X\\) distributed in this way, we would need to calculate \\[\n\\int_a^b xf(x)\\,dx = \\int_{\\delta}^b xf(x)\\,dx\n\\] in the numerator, where \\(f(x)\\) is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:\nreal shift_lognormal_pdf(real x,\n                         real mu,\n                         real sigma,\n                         real delta) {\n  real p;\n\n  p = (1.0 / ((x - delta) * sigma * sqrt(2 * pi()))) *\n    exp(-1 * (log(x - delta) - mu)^2 / (2 * sigma^2));\n\n  return p;\n}\nTherefore, the function that we want to integrate is:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  p = shift_lognormal_pdf(x, mu, sigma, delta);\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhat happens here is that, given that the log-normal distribution is shifted by \\(\\delta\\), when we then try to integrate the numerator, our x starts at values just above delta. This, in turn, causes the x - delta term to be near zero, leading to a breakdown.\nWe can use xc, and define the integrand as:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  if (x &lt; delta + 1) {\n    p = shift_lognormal_pdf(xc, mu, sigma, delta);\n  } else {\n    p = shift_lognormal_pdf(x, mu, sigma, delta);\n  }\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhy does this work? When our values of x are less than delta + 1 (so, when they’re near delta, given that our lower bound of integration is equal to \\(\\delta\\)), we pass xc as an argument to our shift_lognormal_pdf function. This way, instead of dealing with x - delta in shift_lognormal_pdf, we are working with xc - delta which is equal to delta - x - delta, as delta is the lower endpoint in that case. The delta terms cancel out, and we are left with a high-precision version of x. We don’t encounter the same problem at the upper limit \\(b\\) so we don’t adjust the code for that case.\nNote, xc is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, xc will be NaN.\nFor zero-crossing definite integrals (see section Zero Crossing) the integrals are broken into two pieces (\\((a, 0)\\) and \\((0, b)\\) for endpoints \\(a &lt; 0\\) and \\(b &gt; 0\\)) and xc is a high precision version of the distance to the limits of each of the two integrals separately. This means xc will be a high precision version of a - x, x, or b - x, depending on the value of x and the endpoints.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Example Models",
+      "Computing One Dimensional Integrals"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html#log-sum-of-exponentials",
-    "href": "stan-users-guide/floating-point.html#log-sum-of-exponentials",
-    "title": "Floating Point Arithmetic",
+    "objectID": "stan-users-guide/multiple-imputation.html",
+    "href": "stan-users-guide/multiple-imputation.html",
+    "title": "Multiple imputation",
     "section": "",
-    "text": "Working on the log scale, multiplication is converted to addition, \\[\n\\log (a \\cdot b) = \\log a + \\log b.\n\\] Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given \\(\\log a\\) and \\(\\log b\\), how do we get \\(\\log (a + b)\\)? Working out the algebra, \\[\n\\log (a + b)\n=\n\\log (\\exp(\\log a) + \\exp(\\log b)).\n\\]\n\n\nThe nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\log (\\exp(u) + \\exp(v)).\n\\] so that \\[\n\\log (a + b)\n=\n\\textrm{log-sum-exp}(\\log a, \\log b).\n\\]\nAlthough it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\max(u, v) + \\log\\big( \\exp(u - \\max(u, v)) + \\exp(v - \\max(u, v)) \\big).\n\\]\nBecause the terms inside the exponentiations are \\(u - \\max(u, v)\\) and \\(v - \\max(u, v)\\), one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that \\(u \\geq v\\), so that \\[\n\\textrm{log-sum-exp}(u, v) = u + \\log\\big(1 + \\exp(v - u)\\big).\n\\]\nAlthough the inner term may itself be evaluated using the built-in function log1p, there is only limited gain because \\(\\exp(v - u)\\) is only near zero when \\(u\\) is much larger than \\(v\\), meaning the final result is likely to round to \\(u\\) anyway.\nTo conclude, when evaluating \\(\\log (a + b)\\) given \\(\\log a\\) and \\(\\log\nb\\), and assuming \\(\\log a &gt; \\log b\\), return\n\\[\n\\log (a + b) =\n\\log a + \\textrm{log1p}\\big(\\exp(\\log b - \\log a)\\big).\n\\]\n\n\n\nThe log sum of exponentials function may be generalized to sequences in the obvious way, so that if \\(v = v_1, \\ldots, v_N\\), then \\[\\begin{eqnarray*}\n\\textrm{log-sum-exp}(v)\n& = & \\log \\sum_{n = 1}^N \\exp(v_n)\n\\\\[4pt]\n& = & \\max(v) + \\log \\sum_{n = 1}^N \\exp(v_n - \\max(v)).\n\\end{eqnarray*}\\] The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate \\(\\log (u_1 + \\cdots + u_N)\\) given only \\(\\log u_n\\).\n\n\n\nAn immediate application is to computing the mean of a vector \\(u\\) entirely on the log scale. That is, given \\(\\log u\\) and returning \\(\\log \\textrm{mean}(u)\\). \\[\\begin{eqnarray*}\n\\log \\left( \\frac{1}{N} \\sum_{n = 1}^N u_n \\right)\n& = & \\log \\frac{1}{N} + \\log \\sum_{n = 1}^N \\exp(\\log u_n)\n\\\\[4pt]\n& = & -\\log N + \\textrm{log-sum-exp}(\\log u).\n\\end{eqnarray*}\\] where \\(\\log u = (\\log u_1, \\ldots, \\log u_N)\\) is understood elementwise.",
+    "text": "Missing data is common in applied data analysis. Ignoring this missingness can distort posterior inferences and reduce their precision, and modeling it can improve inference quality. There are several ways to model missing data. Chapter 18 in Gelman et al. (2013) offers an approximately Bayesian perspective, and the chapter Missing Data and Partially Known Parameters in the Reference Manual shows an approach that handles missing data as parameters.\nAnother way to model missing data is through multiple imputation, which replaces missing values with sampled values to obtain different versions of a complete data set. Then, the model of interest can be fitted to each of these complete versions separately. Combining the resulting draws from these different fits produces a sample that accounts for the uncertainty in the missing values.\n\n\nSuppose that we have a data set \\(x\\) with columns \\(x_{\\cdot, 1}, \\ldots, x_{\\cdot, K}\\) that make up covariates in a regression with quantities of interest \\(\\theta\\). With the completely observed data, \\(x^\\text{comp}\\), we could estimate the posterior distribution of \\(\\theta\\) as \\(p(\\theta \\mid x^\\text{comp})\\). But with missing data, our matrix is split into \\(x^{\\text{obs}}\\) (the observed values of \\(x\\)) and \\(x^{\\text{mis}}\\) (the missing values of \\(x\\)).\nFortunately, we can treat \\(x^{\\text{mis}}\\) as additional, nuisance parameters that we can estimate along with \\(\\theta\\) as \\(p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\\). So, we can express the marginal distribution of \\(\\theta\\) given only the observed values \\(x^{\\text{obs}}\\) as \\[\\begin{align*}\np(\\theta \\mid x^{\\text{obs}}) =&\n\\int\n    p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=&\n\\int\n    p(\\theta \\mid x^{\\text{obs}}, x^{\\text{mis}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=& \\int\n    p(\\theta \\mid x^{\\text{imp}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}},\n\\end{align*}\\] where \\(x^{\\text{imp}}\\) is a data set that includes imputed values of \\(x^{\\text{mis}}\\).\nThe equations above show that we do not need to describe \\(p(\\theta \\mid x^{\\text{obs}})\\) directly. Instead, we can first find a way to sample from \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), and then use these samples to fit the model \\(p(\\theta \\mid x^\\text{imp})\\), treating \\(x^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nNote that the model for \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\) needs to be explicit. In a typical regression setting with covariates \\(x\\) and outcome \\(y\\), this model for \\(x\\) is not needed because inferences for the regression parameters are independent of the model of \\(x\\) when \\(x\\) is fully observed. But with missing data and multiple imputation, we need to generate values for \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), which in turn needs an explicit model for \\(x^{\\text{mis}}\\).\nThus, the general outline for multiple imputation with a number \\(M\\) of imputations is:\n\nDraw \\(M\\) random samples \\(x^{\\text{mis}}_1, \\ldots,\nx^{\\text{mis}}_M\\) from the posterior predictive distribution \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\).\nUse these samples to get \\(M\\) different complete data sets \\(x_1^\\text{imp}, \\ldots, x_M^\\text{imp}\\).\nFor each of the \\(M\\) imputed data sets, sample from the model of interest \\(p(\\theta \\mid x_m^\\text{imp})\\), each time treating \\(x_m^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nCombine the draws for \\(\\theta\\) from all \\(M\\) fits.\n\n\n\nFollowing the notation above, we can express the outline with the pseudocode below:\ntheta_estimates = []\nfor (m in 1:M) {\n  x_mis &lt;- get_imputations(x_obs)\n  x_imp &lt;- add_imputations(x_obs, x_mis)\n  imputation_fits &lt;- my_model.fit(x_imp).draws()\n  theta_estimates &lt;- theta_estimates.append(imputation_fits)\n}\nEach of the functions in this pseudocode can encapsulate simple or complex procedures. get_imputations(), for example, may impute missing data by taking values directly from x_obs; or it may model a regression where an \\(x_{\\cdot, k}\\) with missing values is the outcome and the \\(x_{\\cdot, j}\\) without missing observations are predictors. Below we illustrate how to implement this second approach in Stan.\n\n\n\n\nImagine our data set is composed of two numerical, continuous variables \\(x\\) and \\(y\\), and one binary variable \\(z\\). We want to estimate the parameters that govern the conditional association \\(p(y \\mid x, z)\\), but several values of \\(x\\) are missing. So, before we fit this model for \\(y\\), we impute the missing values in \\(x\\).\nWe first find all the fully observed data points and identify their values as \\(x^{\\text{obs}}\\), \\(y^{\\text{obs}}\\), and \\(z^{\\text{obs}}\\). Then we use these data to fit the model \\[\nx^{\\text{obs}} \\sim \\text{normal}(\\gamma_0 + \\gamma_1 y^{\\text{obs}} +\n    \\gamma_2 z^{\\text{obs}}, \\lambda).\n\\]\nThis model can then give us samples for the missing values \\(x^{\\text{mis}}\\) conditional on the corresponding values of \\(y\\) and \\(z\\), which we call \\(y^{\\text{aux}}\\) and \\(z^{\\text{aux}}\\). The Stan code for this imputation is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  vector[N_obs] x_obs;\n  vector[N_obs] y_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;  \n  int&lt;lower=0&gt; N_mis;   \n  vector[N_mis] y_aux;  \n  array[N_mis] int&lt;lower=0, upper=1&gt; z_aux; \n}\nparameters {\n  vector[3] gamma;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  lambda ~ exponential(1);\n  x_obs ~ normal(gamma[1] + gamma[2] * y_obs + gamma[3] * z_obs,\n          lambda);\n}\ngenerated quantities {\n  array[N_mis] x_imp = normal_rng(gamma[1] + gamma[2] * y_aux[n]\n                      + gamma[3] * z_aux[n], lambda);\n}\nThe generated quantities block automatically samples \\(\\gamma_0\\), \\(\\gamma_1\\), \\(\\gamma_2\\), and \\(\\lambda\\) from their posterior distributions. So, the random draws of x_imp incorporate uncertainty from the estimated parameters and from the sampling variation in \\(x^{\\text{mis}}\\).\nMultiple posterior draws of \\(x^{\\text{mis}}\\) give us multiple imputed data sets. With these data sets we can model \\(p(y \\mid \\beta, \\sigma, x, z)\\) as \\[\ny \\sim \\text{normal}(\\beta_0 + \\beta_1 x^\\text{comp} + \\beta_2 z,\n        \\sigma),\n\\] where \\(x^\\text{comp}\\) contains observed and imputed values. The Stan code for this model is:\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  vector[N] z;\n}\nparameters {\n  vector[3] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  beta ~ normal(0, 1);\n  sigma ~ exponential(1);\n  y ~ normal(beta[1] + beta[2] * x + beta[3] * z, sigma);\n}\nWith multiple imputation, the model for \\(p(y \\mid \\beta, \\sigma, x, z)\\) does not need to distinguish between observed and imputed values of \\(x\\). Instead, this model can treat all imputed data sets as complete because we can combine all the posterior draws from multiple fits.\n\n\n\nA more general scenario involves an outcome variable \\(y\\) and several explanatory variables \\(x_1, \\ldots, x_K\\), each of which can have missing values that we want to impute.\nOne solution is to do multiple imputation through chained equations, a procedure often called “MICE”1. The MICE procedure in this scenario is:\n\nInitialize missing values in all \\(x_i\\). For each variable \\(x_i\\) with missing entries, fill its missing values with random samples from its observed values (or use another simple initialization rule).\nUpdate each \\(x_i\\) given \\(y\\) and the other \\(x\\)’s. For \\(i=1, \\ldots, K\\), fit a model for the observed \\(x_i\\) conditional on the current versions (with observed and imputed values) of \\(y\\) and \\(x_{-k} = x_1, \\ldots x_{k-1}, x_{k+1}, \\ldots,\nx_{K}\\). Use this model to draw impuations from the predictive distribution of the missing \\(x_i\\). Completing this step for all \\(i = 1, \\ldots, K\\) constitutes a single imputation cycle.\nWarmup period. Repeat the imputation cycle in step 1 several times as warmup to let the imputations stabilize.\nCreate M imputed datasets. After the warmup, record the current complete dataset as one imputed dataset. Then either restart from step 0 and repeat steps 1–3 until you have \\(M\\) imputed datasets; or do a single long run, i.e., continue iterating steps 1–2 and save the imputed dataset at \\(M\\) well-spaced iterations (e.g., every \\(S\\) cycles) to obtain \\(M\\) imputed datasets, without restarting from step 0 each time.\nFit the target model \\(p(y \\mid \\beta, \\sigma, x_1, \\ldots,\nx_K)\\) separately to each of the \\(M\\) imputed datasets and save the posterior draws for the parameters of interest.\nCombine the draws (or other summaries) from all \\(M\\) fits.\n\nNote that, as described here, the MICE procedure does not guarantee that the conditional distributions of all variables will be compatible. Compatibility means that there is a joint distribution for all the variables used in the imputation that can be decomposed as the conditional distributions we used. An incompatible imputation model will technically not sample from any well-defined probability distribution. But the consequences of this are not always serious. See section 6.4 in Carpenter et al. (2023) for a more detailed explanation of compatibility and of the related concept of congeniality.\n\n\nImagine that we again want to use numerical variables \\(x\\) and \\(y\\), and dichotomic variable \\(z\\) to estimate the parameters in \\(p(y \\mid \\beta, \\sigma, x, z)\\). But now both \\(x\\) and \\(z\\) have missing values that we want to impute.\nTo apply the MICE procedure in this example, we can reuse the models for \\(x\\) and \\(y\\) that we defined above. We also need a new model to impute \\(z^{\\text{mis}}\\), so we use the logistic regression \\[\nz^{\\text{obs}} \\sim \\text{Bernoulli}(\n  \\text{logit}^{-1}(\n    \\alpha_0 + \\alpha_1 y^{\\text{obs}} + \\alpha_2 x^{\\text{obs}})\n  ).\n\\] Here, \\(z^{\\text{obs}}\\) contains only the completely observed values from our original data set, while \\(x^{\\text{obs}}\\) and \\(y^{\\text{obs}}\\) contain all the values that correspond to \\(z^{\\text{obs}}\\). Thus, \\(x^{\\text{obs}}\\) can include observed and imputed values, and \\(y^{\\text{obs}}\\) need not be the same as in the model for \\(x^{\\text{obs}}\\).\nWith this model we can sample values for \\(z^{\\text{mis}}\\) conditional on its corresponding values \\(y^{\\text{aux}}\\) and \\(x^{\\text{aux}}\\).\nThe Stan code to impute \\(z^{\\text{mis}}\\) is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;\n  vector[N_obs] y_obs;\n  vector[N_obs] x_obs;\n  int&lt;lower=0&gt; N_mis;\n  vector[N_mis] y_aux;\n  vector[N_mis] x_aux;\n}\nparameters {\n  vector[3] gamma;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  z_obs ~ bernoulli_logit(gamma[1] + gamma[2] * y_obs \n                          + gamma[3] * x_obs);\n}\ngenerated quantities {\n  array[N_mis] int z_imp = bernoulli_logit_rng(gamma[1]\n                      + gamma[2] * y_aux\n                      + gamma[3] * x_aux);\n}\nThe pseudocode for this example of MICE, restarting the imputations after each imputation cycle, is shown below. Note that we do not need to initialize \\(x^{\\text{mis}}\\) in step 0 because we can impute it directly with the model for \\(x^{\\text{obs}}\\).\ncompleted_datasets &lt;- []\nfor (m in 1:M) {\n  data_m &lt;- copy(data_orig)\n  data_m.z[missing_idz] &lt;- random_sample(data_m.z[observed_idz])\n  for (cycle in 1:n_warm) {\n    stanmod_x &lt;- build_model(\"model_for_x.stan\")\n    fit_x &lt;- stanmod_x.fit(\n      obs_data=data_m[observed_idx],\n      mis_data=data_m[missing_idx]\n    )\n    data_m.x[missing_idx] &lt;- get_imputations(fit_x, \"x_imp\")\n    stanmod_z &lt;- build_model(\"model_for_z.stan\")\n    fit_z &lt;- stanmod_z.fit(\n      obs_data=data_m[observed_idz],\n      mis_data=data_m[missing_idz]\n    )\n    data_m.z[missing_idz] &lt;- get_imputations(fit_z, \"z_imp\")\n  }\n  completed_datasets.append(data_m)\n}\nall_draws &lt;- []\nfor (dataset in completed_datasets) {\n  stanmod_y &lt;- build_model(\"model_for_y.stan\")\n  fit_y &lt;- stanmod_y.fit(dataset)\n  all_draws.append(extract_draws(model_results))\n}\nall_results &lt;- combine_results(all_draws)\n\n\n\n\nWith Stan’s MCMC sampler, we can treat posterior draw chains from imputed data sets as if they were chains based on complete data. There is one important difference. Multiple imputation expresses uncertainty in the missing values as consistent differences in the estimates obtained from different imputated data sets. This means that chains obtained from the same imputed data set should converge, but chains obtained from different data sets do not have to. So, we need not worry if diagnostics2 signal that the chains from different imputed data sets are not converging properly. See an example in Bürkner (2025).\n\n\n\nA full Bayesian probability model includes a feedback flow of information between all parameters and all data. Cut models separate some parts of this feedback flow so that different subsets of data influence only some parameters in the model (see Plummer (2015)).\nMultiple imputation interrupts the flow of information from data to parameters. In our regression above, for example, the imputations influence the distribution of the parameters in \\(p(y \\mid x_1, \\ldots, x_K)\\), but these parameters do not influence the imputations. So, we could use multiple imputation to implement a cut model.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html#comparing-floating-point-numbers",
-    "href": "stan-users-guide/floating-point.html#comparing-floating-point-numbers",
-    "title": "Floating Point Arithmetic",
+    "objectID": "stan-users-guide/multiple-imputation.html#motivation-and-outline",
+    "href": "stan-users-guide/multiple-imputation.html#motivation-and-outline",
+    "title": "Multiple imputation",
     "section": "",
-    "text": "Because floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing x == y, an approximate test may be used given an absolute or relative tolerance.\nGiven a positive absolute tolerance of epsilon, x can be compared to y using the conditional\nabs(x - y) &lt;= epsilon.\nAbsolute tolerances work when the scale of x and y and the relevant comparison is known.\nGiven a positive relative tolerance of epsilon, a typical comparison is\n2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.",
+    "text": "Suppose that we have a data set \\(x\\) with columns \\(x_{\\cdot, 1}, \\ldots, x_{\\cdot, K}\\) that make up covariates in a regression with quantities of interest \\(\\theta\\). With the completely observed data, \\(x^\\text{comp}\\), we could estimate the posterior distribution of \\(\\theta\\) as \\(p(\\theta \\mid x^\\text{comp})\\). But with missing data, our matrix is split into \\(x^{\\text{obs}}\\) (the observed values of \\(x\\)) and \\(x^{\\text{mis}}\\) (the missing values of \\(x\\)).\nFortunately, we can treat \\(x^{\\text{mis}}\\) as additional, nuisance parameters that we can estimate along with \\(\\theta\\) as \\(p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\\). So, we can express the marginal distribution of \\(\\theta\\) given only the observed values \\(x^{\\text{obs}}\\) as \\[\\begin{align*}\np(\\theta \\mid x^{\\text{obs}}) =&\n\\int\n    p(\\theta, x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=&\n\\int\n    p(\\theta \\mid x^{\\text{obs}}, x^{\\text{mis}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}} \\\\\n=& \\int\n    p(\\theta \\mid x^{\\text{imp}})\n    p(x^{\\text{mis}} \\mid x^{\\text{obs}})\n    \\mathrm{d}x^{\\text{mis}},\n\\end{align*}\\] where \\(x^{\\text{imp}}\\) is a data set that includes imputed values of \\(x^{\\text{mis}}\\).\nThe equations above show that we do not need to describe \\(p(\\theta \\mid x^{\\text{obs}})\\) directly. Instead, we can first find a way to sample from \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), and then use these samples to fit the model \\(p(\\theta \\mid x^\\text{imp})\\), treating \\(x^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nNote that the model for \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\) needs to be explicit. In a typical regression setting with covariates \\(x\\) and outcome \\(y\\), this model for \\(x\\) is not needed because inferences for the regression parameters are independent of the model of \\(x\\) when \\(x\\) is fully observed. But with missing data and multiple imputation, we need to generate values for \\(x^{\\text{mis}}\\) based on \\(x^{\\text{obs}}\\), which in turn needs an explicit model for \\(x^{\\text{mis}}\\).\nThus, the general outline for multiple imputation with a number \\(M\\) of imputations is:\n\nDraw \\(M\\) random samples \\(x^{\\text{mis}}_1, \\ldots,\nx^{\\text{mis}}_M\\) from the posterior predictive distribution \\(p(x^{\\text{mis}} \\mid x^{\\text{obs}})\\).\nUse these samples to get \\(M\\) different complete data sets \\(x_1^\\text{imp}, \\ldots, x_M^\\text{imp}\\).\nFor each of the \\(M\\) imputed data sets, sample from the model of interest \\(p(\\theta \\mid x_m^\\text{imp})\\), each time treating \\(x_m^\\text{imp}\\) as if it was \\(x^\\text{comp}\\).\nCombine the draws for \\(\\theta\\) from all \\(M\\) fits.\n\n\n\nFollowing the notation above, we can express the outline with the pseudocode below:\ntheta_estimates = []\nfor (m in 1:M) {\n  x_mis &lt;- get_imputations(x_obs)\n  x_imp &lt;- add_imputations(x_obs, x_mis)\n  imputation_fits &lt;- my_model.fit(x_imp).draws()\n  theta_estimates &lt;- theta_estimates.append(imputation_fits)\n}\nEach of the functions in this pseudocode can encapsulate simple or complex procedures. get_imputations(), for example, may impute missing data by taking values directly from x_obs; or it may model a regression where an \\(x_{\\cdot, k}\\) with missing values is the outcome and the \\(x_{\\cdot, j}\\) without missing observations are predictors. Below we illustrate how to implement this second approach in Stan.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/floating-point.html#footnotes",
-    "href": "stan-users-guide/floating-point.html#footnotes",
-    "title": "Floating Point Arithmetic",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe notable exception is Intel’s optimizing compilers under certain optimization settings.↩︎",
+    "objectID": "stan-users-guide/multiple-imputation.html#imputing-one-variable-in-stan",
+    "href": "stan-users-guide/multiple-imputation.html#imputing-one-variable-in-stan",
+    "title": "Multiple imputation",
+    "section": "",
+    "text": "Imagine our data set is composed of two numerical, continuous variables \\(x\\) and \\(y\\), and one binary variable \\(z\\). We want to estimate the parameters that govern the conditional association \\(p(y \\mid x, z)\\), but several values of \\(x\\) are missing. So, before we fit this model for \\(y\\), we impute the missing values in \\(x\\).\nWe first find all the fully observed data points and identify their values as \\(x^{\\text{obs}}\\), \\(y^{\\text{obs}}\\), and \\(z^{\\text{obs}}\\). Then we use these data to fit the model \\[\nx^{\\text{obs}} \\sim \\text{normal}(\\gamma_0 + \\gamma_1 y^{\\text{obs}} +\n    \\gamma_2 z^{\\text{obs}}, \\lambda).\n\\]\nThis model can then give us samples for the missing values \\(x^{\\text{mis}}\\) conditional on the corresponding values of \\(y\\) and \\(z\\), which we call \\(y^{\\text{aux}}\\) and \\(z^{\\text{aux}}\\). The Stan code for this imputation is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  vector[N_obs] x_obs;\n  vector[N_obs] y_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;  \n  int&lt;lower=0&gt; N_mis;   \n  vector[N_mis] y_aux;  \n  array[N_mis] int&lt;lower=0, upper=1&gt; z_aux; \n}\nparameters {\n  vector[3] gamma;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  lambda ~ exponential(1);\n  x_obs ~ normal(gamma[1] + gamma[2] * y_obs + gamma[3] * z_obs,\n          lambda);\n}\ngenerated quantities {\n  array[N_mis] x_imp = normal_rng(gamma[1] + gamma[2] * y_aux[n]\n                      + gamma[3] * z_aux[n], lambda);\n}\nThe generated quantities block automatically samples \\(\\gamma_0\\), \\(\\gamma_1\\), \\(\\gamma_2\\), and \\(\\lambda\\) from their posterior distributions. So, the random draws of x_imp incorporate uncertainty from the estimated parameters and from the sampling variation in \\(x^{\\text{mis}}\\).\nMultiple posterior draws of \\(x^{\\text{mis}}\\) give us multiple imputed data sets. With these data sets we can model \\(p(y \\mid \\beta, \\sigma, x, z)\\) as \\[\ny \\sim \\text{normal}(\\beta_0 + \\beta_1 x^\\text{comp} + \\beta_2 z,\n        \\sigma),\n\\] where \\(x^\\text{comp}\\) contains observed and imputed values. The Stan code for this model is:\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  vector[N] z;\n}\nparameters {\n  vector[3] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  beta ~ normal(0, 1);\n  sigma ~ exponential(1);\n  y ~ normal(beta[1] + beta[2] * x + beta[3] * z, sigma);\n}\nWith multiple imputation, the model for \\(p(y \\mid \\beta, \\sigma, x, z)\\) does not need to distinguish between observed and imputed values of \\(x\\). Instead, this model can treat all imputed data sets as complete because we can combine all the posterior draws from multiple fits.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Floating Point Arithmetic"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html",
-    "href": "stan-users-guide/efficiency-tuning.html",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/multiple-imputation.html#imputing-two-or-more-variables",
+    "href": "stan-users-guide/multiple-imputation.html#imputing-two-or-more-variables",
+    "title": "Multiple imputation",
     "section": "",
-    "text": "This chapter provides a grab bag of techniques for optimizing Stan code, including vectorization, sufficient statistics, and conjugacy. At a coarse level, efficiency involves both the amount of time required for a computation and the amount of memory required. For practical applied statistical modeling, we are mainly concerned with reducing wall time (how long a program takes as measured by a clock on the wall) and keeping memory requirements within available bounds.\n\n\nThe standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).\n\n\n\nStan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.\nWith probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.\nA well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.\n\n\n\nThere is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.\nStatistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.\n\n\n\nBecause Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.\nA second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density \\(\\log\np(\\theta)\\) with respect to the parameter vector \\(\\theta\\), defined as \\[\nH(\\theta) = \\nabla \\, \\nabla \\, \\log p(\\theta \\mid y),\n\\] so that \\[\nH_{i, j}(\\theta) = \\frac{\\partial^2 \\log p(\\theta \\mid y)}\n                {\\partial \\theta_i \\ \\partial \\theta_j}.\n\\] For pure penalized maximum likelihood problems, the posterior log density \\(\\log p(\\theta \\mid y)\\) is replaced by the penalized likelihood function \\(\\mathcal{L}(\\theta) = \\log p(y \\mid \\theta) - \\lambda(\\theta)\\).\n\n\nA good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix \\(H\\), which is the ratio of the largest to the smallest eigenvalue of \\(H\\) (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.\n\n\n\nIdeally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.\n\n\n\nIn all but very simple models (such as multivariate normals), the Hessian will vary as \\(\\theta\\) varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.\nFor all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.\nWith sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.\n\n\n\nImproving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the change of variables chapter.\nOne has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.\n\n\n\n\nModel misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.\nCounterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as normal(0, 1000)). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as normal(mu,   sigma), where mu and sigma are new parameters with their own hyperpriors, can be beneficial.\n\n\n\nStan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ...\n}                               // first validation\nmodel {\n  y ~ multi_normal(mu, Sigma);  // second validation\n  // ...\n}\nBecause Sigma is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that Sigma is positive definite. This test is expensive, having cubic run time (i.e., \\(\\mathcal{O}(N^3)\\) for \\(N \\times N\\) matrices), so it should not be done twice.\nThe test may be avoided by simply declaring Sigma to be a simple unconstrained matrix.\ntransformed parameters {\n  matrix[K, K] Sigma;\n  // ...\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);  // only validation\n}\nNow the only validation is carried out by the multivariate normal.\n\n\n\nStan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.\n\n\nIn this section, we discuss a general transform from a centered to a non-centered parameterization (Papaspiliopoulos, Roberts, and Sköld 2007).1\nThis reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.\nNeal (2003) defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for \\(y \\in\n\\mathbb{R}\\) and \\(x \\in \\mathbb{R}^9\\) with density\n\\[\np(y,x) = \\textsf{normal}(y \\mid 0,3) \\times \\prod_{n=1}^9\n\\textsf{normal}(x_n \\mid 0,\\exp(y/2)).\n\\]\nThe probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to \\(y\\). A plot of the log marginal density of \\(y\\) and the first dimension \\(x_1\\) is shown in the following plot.\nThe funnel can be implemented directly in Stan as follows.\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y/2));\n}\nWhen the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where \\(y\\) is small and thus \\(x\\) is constrained to be near 0. This is due to the fact that the density’s scale changes with \\(y\\), so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.\n\n\n\n\n\n\n\n\n\n\n\n\n\nFigure 1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable \\(y\\) and one lower-level variable \\(x_1\\) (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of \\(x_1\\) and \\(y\\) values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density \\(p(y) = \\textsf{normal}(y \\mid 0,3)\\), which has mean 0 and standard deviation 3.\n\n\n\nIn this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y/2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y/2))\n}\nIn this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.\n\n\n\nSampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.\nThe practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.\nSuppose a random variable of interest \\(X\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), so that \\(X \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\). The variable \\(X\\) has a cumulative distribution function \\(F_X:\\mathbb{R} \\rightarrow (0,1)\\) defined by \\[\nF_X(x) = \\frac{1}{\\pi} \\arctan \\left( \\frac{x - \\mu}{\\tau} \\right) +\n\\frac{1}{2}.\n\\] The inverse of the cumulative distribution function, \\(F_X^{-1}:(0,1) \\rightarrow \\mathbb{R}\\), is thus\n\\[\nF^{-1}_X(y) = \\mu + \\tau \\tan \\left( \\pi \\left( y - \\frac{1}{2} \\right) \\right).\n\\] Thus if the random variable \\(Y\\) has a unit uniform distribution, \\(Y\n\\sim \\textsf{uniform}(0,1)\\), then \\(F^{-1}_X(Y)\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), i.e., \\(F^{-1}_X(Y) \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\).\nConsider a Stan program involving a Cauchy-distributed parameter beta.\nparameters {\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ cauchy(mu, tau);\n  // ...\n}\nThis declaration of beta as a parameter may be replaced with a transformed parameter beta defined in terms of a uniform-distributed parameter beta_unif.\nparameters {\n  real&lt;lower=-pi() / 2, upper=pi() / 2&gt; beta_unif;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = mu + tau * tan(beta_unif);  // beta ~ cauchy(mu, tau)\n}\nmodel {\n  beta_unif ~ uniform(-pi() / 2, pi() / 2);  // not necessary\n  // ...\n}\nIt is more convenient in Stan to transform a uniform variable on \\((-\\pi/2, \\pi/2)\\) than one on \\((0,1)\\). The Cauchy location and scale parameters, mu and tau, may be defined as data or may themselves be parameters. The variable beta could also be defined as a local variable if it does not need to be included in the sampler’s output.\nThe uniform distribution on beta_unif is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because \\(\\log\n\\textsf{uniform}(\\beta_{\\textsf{unif}} \\mid -\\pi/2,\\pi/2) =\n-\\log \\pi\\) is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that beta_unif is between -pi() / 2 and pi() / 2, but this condition is guaranteed by the constraints in the declaration of beta_unif.\n\n\n\nOne thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable \\(\\beta\\), \\[\n\\beta \\sim \\textsf{Student-t}(\\nu, 0, 1),\n\\] by first generating a gamma-distributed precision (inverse variance) \\(\\tau\\) according to \\[\n\\tau \\sim \\textsf{Gamma}(\\nu/2, \\nu/2),\n\\] and then generating \\(\\beta\\) from the normal distribution, \\[\n\\beta \\sim \\textsf{normal}\\left(0,\\tau^{-\\frac{1}{2}}\\right).\n\\]\nBecause \\(\\tau\\) is precision, \\(\\tau^{-\\frac{1}{2}}\\) is the scale (standard deviation), which is the parameterization used by Stan.\nThe marginal distribution of \\(\\beta\\) when you integrate out \\(\\tau\\) is \\(\\textsf{Student-t}(\\nu, 0, 1)\\), i.e., \\[\n\\textsf{Student-t}(\\beta \\mid \\nu, 0, 1)\n=\n\\int_0^{\\infty}\n\\,\n\\textsf{normal}\\left(\\beta \\middle| 0, \\tau^{-0.5}\\right)\n\\times\n\\textsf{Gamma}\\left(\\tau \\middle| \\nu/2, \\nu/2\\right)\n\\\n\\text{d} \\tau.\n\\]\nTo go one step further, instead of defining a \\(\\beta\\) drawn from a normal with precision \\(\\tau\\), define \\(\\alpha\\) to be drawn from a unit normal, \\[\n\\alpha \\sim \\textsf{normal}(0,1)\n\\] and rescale by defining \\[\n\\beta = \\alpha \\, \\tau^{-\\frac{1}{2}}.\n\\]\nNow suppose \\(\\mu = \\beta x\\) is the product of \\(\\beta\\) with a regression predictor \\(x\\). Then the reparameterization \\(\\mu = \\alpha\n\\tau^{-\\frac{1}{2}} x\\) has the same distribution, but in the original, direct parameterization, \\(\\beta\\) has (potentially) heavy tails, whereas in the second, neither \\(\\tau\\) nor \\(\\alpha\\) have heavy tails.\nTo translate into Stan notation, this reparameterization replaces\nparameters {\n  real&lt;lower=0&gt; nu;\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ student_t(nu, 0, 1);\n  // ...\n}\nwith\nparameters {\n  real&lt;lower=0&gt; nu;\n  real&lt;lower=0&gt; tau;\n  real alpha;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = alpha / sqrt(tau);\n  // ...\n}\nmodel {\n  real half_nu;\n  half_nu = 0.5 * nu;\n  tau ~ gamma(half_nu, half_nu);\n  alpha ~ std_normal();\n  // ...\n}\nAlthough set to 0 here, in most cases, the lower bound for the degrees of freedom parameter nu can be set to 1 or higher; when nu is 1, the result is a Cauchy distribution with fat tails and as nu approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter nu characterizes the heaviness of the tails of the model.\n\n\n\nUnfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.\n\n\nFor example, a vectorized hierarchical model might draw a vector of coefficients \\(\\beta\\) with definitions as follows. The so-called centered parameterization is as follows.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nAlthough not shown, a full model will have priors on both mu_beta and sigma_beta along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix x and binary outcome vector y would include a likelihood statement such as form y ~ bernoulli_logit(x * beta), leading to an analytically intractable posterior.\nA hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of beta, mu_beta and sigma_beta are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See Betancourt and Girolami (2013) for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.\n\n\n\n\nSometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see Betancourt and Girolami (2013)), and in more extreme cases will be necessary to achieve convergence.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta_raw;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // implies: beta ~ normal(mu_beta, sigma_beta)\n  beta = mu_beta + sigma_beta * beta_raw;\n}\nmodel {\n  beta_raw ~ std_normal();\n  // ...\n}\nAny priors defined for mu_beta and sigma_beta remain as defined in the original model.\nAlternatively, Stan’s affine transform can be used to decouple sigma and beta:\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;offset=mu_beta, multiplier=sigma_beta&gt;[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nReparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let \\(\\beta = l + s\\alpha\\) where \\(l\\) is a location parameter and \\(s\\) is a scale parameter. The parameter \\(l\\) need not be the mean, \\(s\\) need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If \\(\\alpha\\) and \\(\\beta\\) are from the same distributional family but \\(\\alpha\\) has location zero and unit scale, while \\(\\beta\\) has location \\(l\\) and scale \\(s\\), then that distribution is a location-scale distribution. Thus, if \\(\\alpha\\) were a parameter and \\(\\beta\\) were a transformed parameter, then a prior distribution from the location-scale family on \\(\\alpha\\) with location zero and unit scale implies a prior distribution on \\(\\beta\\) with location \\(l\\) and scale \\(s\\). Doing so would reduce the dependence between \\(\\alpha\\), \\(l\\), and \\(s\\).\nThere are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if \\(\\alpha\\) is distributed standard normal, then \\(\\beta\\) is distributed normal with mean \\(\\mu = l\\) and standard deviation \\(\\sigma = s\\). The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.\nAlso, if \\(z\\) is distributed standard normal, then \\(z^2\\) is distributed chi-squared with one degree of freedom. By summing the squares of \\(K\\) independent standard normal variates, one can obtain a single variate that is distributed chi-squared with \\(K\\) degrees of freedom. However, for large \\(K\\), the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying \\(K\\) primitive parameters just to obtain one transformed parameter to use in a model.\n\n\n\nThe benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for \\(\\beta\\) to be multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). Such a belief is reflected by the following code.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\nparameters {\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ multi_normal(mu, Sigma);\n  // ...\n}\nIn this case mu and Sigma are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of beta.\nIf \\(\\alpha\\) has the same dimensions as \\(\\beta\\) but the elements of \\(\\alpha\\) are independently and identically distributed standard normal such that \\(\\beta = \\mu + L\\alpha\\), where \\(LL^\\top = \\Sigma\\), then \\(\\beta\\) is distributed multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). One choice for \\(L\\) is the Cholesky factor of \\(\\Sigma\\). Thus, the model above could be reparameterized as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Sigma);\n}\nparameters {\n  vector[K] alpha;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  beta = mu + L * alpha;\n}\nmodel {\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu, Sigma)\n  // ...\n}\nThis reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of alpha and second, it avoids the need to invert Sigma every time multi_normal is evaluated.\nThe Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix Tau were fixed data but the vector of standard deviations sigma were unknown parameters, then a reparameterization of beta in terms of alpha could be implemented as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  corr_matrix[K] Tau;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Tau);\n}\nparameters {\n  vector[K] alpha;\n  vector&lt;lower=0&gt;[K] sigma;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // This equals mu + diag_matrix(sigma) * L * alpha;\n  beta = mu + sigma .* (L * alpha);\n}\nmodel {\n  sigma ~ cauchy(0, 5);\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu,\n  //  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))\n  // ...\n}\nThis reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.\nA Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let \\(L\\) be the Cholesky factor of a \\(K \\times K\\) positive definite scale matrix \\(S\\) and let \\(\\nu\\) be the degrees of freedom. If \\[\nA = \\begin{pmatrix}\n\\sqrt{c_{1}} & 0            & \\cdots                & 0 \\\\\nz_{21}       & \\sqrt{c_{2}} & \\ddots                & \\vdots \\\\\n\\vdots       & \\ddots       & \\ddots                & 0 \\\\\nz_{K1}       & \\cdots       & z_{K\\left(K-1\\right)} & \\sqrt{c_{K}}\n\\end{pmatrix},\n\\] where each \\(c_i\\) is distributed chi-squared with \\(\\nu - i + 1\\) degrees of freedom and each \\(z_{ij}\\) is distributed standard normal, then \\(W = LAA^{\\top}L^{\\top}\\) is distributed Wishart with scale matrix \\(S = LL^{\\top}\\) and degrees of freedom \\(\\nu\\). Such a reparameterization can be implemented by the following Stan code:\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  vector[K] mu;\n  matrix[N, K] y;\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  int count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal();\n  // implies: L * A * A' * L' ~ wishart(nu, L * L')\n  y ~ multi_normal_cholesky(mu, L * A);\n  // ...\n}\nThis reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of z and second, it avoids the need to invert the covariance matrix, \\(W\\) every time wishart is evaluated. Third, if \\(W\\) is to be used with a multivariate normal distribution, you can pass \\(L A\\) to the more efficient multi_normal_cholesky function, rather than passing \\(W\\) to multi_normal.\nIf \\(W\\) is distributed Wishart with scale matrix \\(S\\) and degrees of freedom \\(\\nu\\), then \\(W^{-1}\\) is distributed inverse Wishart with inverse scale matrix \\(S^{-1}\\) and degrees of freedom \\(\\nu\\). Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since \\(W = L A A^{\\top} L^{\\top}\\), \\(W^{-1} = L^{{\\top}^{-1}} A^{{\\top}^{-1}} A^{-1} L^{-1}\\), where all four inverses exist, but \\(L^{{-1}^{\\top}} = L^{{\\top}^{-1}}\\) and \\(A^{{-1}^{\\top}} = A^{{\\top}^{-1}}\\). We can slightly modify the above Stan code for this case:\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  // ...\n}\ntransformed data {\n  matrix[K, K] eye;\n  matrix[K, K] L_inv;\n  for (j in 1:K) {\n    for (i in 1:K) {\n      eye[i, j] = 0.0;\n    }\n    eye[j, j] = 1.0;\n  }\n  L_inv = mdivide_left_tri_low(L, eye);\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  matrix[K, K] A_inv_L_inv;\n  int count;\n  count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal(); // implies: crossprod(A_inv_L_inv) ~\n  // inv_wishart(nu, L_inv' * L_inv)\n  // ...\n}\nAnother candidate for reparameterization is the Dirichlet distribution with all \\(K\\) shape parameters equal. Zyczkowski and Sommers (2001) shows that if \\(\\theta_i\\) is equal to the sum of \\(\\beta\\) independent squared standard normal variates and \\(\\rho_i = \\frac{\\theta_i}{\\sum \\theta_i}\\), then the \\(K\\)-vector \\(\\rho\\) is distributed Dirichlet with all shape parameters equal to \\(\\frac{\\beta}{2}\\). In particular, if \\(\\beta = 2\\), then \\(\\rho\\) is uniformly distributed on the unit simplex. Thus, we can make \\(\\rho\\) be a transformed parameter to reduce dependence, as in:\ndata {\n  int&lt;lower=1&gt; beta;\n  // ...\n}\nparameters {\n  array[K] vector[beta] z;\n  // ...\n}\ntransformed parameters {\n  simplex[K] rho;\n  for (k in 1:K) {\n    rho[k] = dot_self(z[k]); // sum-of-squares\n  }\n  rho = rho / sum(rho);\n}\nmodel {\n  for (k in 1:K) {\n    z[k] ~ std_normal();\n  }\n  // implies: rho ~ dirichlet(0.5 * beta * ones)\n  // ...\n}\n\n\n\n\n\n\nStan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.2 This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.\n\n\n\nBecause of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the sum() operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where foo() is some operation that depends on n.\nfor (n in 1:N) {\n  total += foo(n,...);\n}\nThis code has to create intermediate representations for each of the N summands.\nA faster alternative is to copy the values into a vector, then apply the sum() operator, as in the following refactoring.\n{\n  vector[N] summands;\n  for (n in 1:N) {\n    summands[n] = foo(n,...);\n  }\n  total = sum(summands);\n}\nSyntactically, the replacement is a statement block delineated by curly brackets ({, }), starting with the definition of the local variable summands.\nEven though it involves extra work to allocate the summands vector and copy N values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing total within the loop.\n\n\n\nThe following program directly encodes a linear regression with fixed unit noise using a two-dimensional array x of predictors, an array y of outcomes, and an array beta of regression coefficients.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[K, N] real x;\n  array[N] real y;\n}\nparameters {\n  array[K] real beta;\n}\nmodel {\n  for (n in 1:N) {\n    real gamma = 0;\n    for (k in 1:K) {\n      gamma += x[n, k] * beta[k];\n    }\n    y[n] ~ normal(gamma, 1);\n  }\n}\nThe following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[N] vector[K] x;\n  array[N] real y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(dot_product(x[n], beta), 1);\n  }\n}\nAlthough it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in\ny[n] ~ normal(x[n]' * beta, 1);\nThe relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.3\nThe inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.\n// ...\ntransformed parameters {\n  row_vector[K] beta_t;\n  beta_t = beta';\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(beta_t * x[n], 1);\n  }\n}\nThe problem with transposition could be completely solved by directly encoding the x as a row vector, as in the following example.\ndata {\n  // ...\n  array[N] row_vector[K] x;\n  // ...\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, 1);\n  }\n}\nDeclaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.\nHaving said all this, the most efficient way to code this model is with direct matrix multiplication, as in\ndata {\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nIn general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.\n\n\n\nThe final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nThe variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication x * beta in the model block is a vector of the same length as y.\nThe probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, y is a vector of size N, x * beta is a vector of size N, and 1 is a scalar.\n\n\n\nSometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).\nJohn Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu[ii[n]], sigma);\n}\nThe brute force vectorization would build up a mean vector and then vectorize all at once.\n{\n  vector[N] mu_ii;\n  for (n in 1:N) {\n    mu_ii[n] = mu[ii[n]];\n  }\n  y ~ normal(mu_ii, sigma);\n}\nIf there aren’t many levels (values ii[n] can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector y, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have\ny_1 ~ normal(mu[1], sigma);\n// ...\ny_5 ~ normal(mu[5], sigma);\nThis way, both the mu and sigma parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.\n\n\n\n\nIn some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.\n\n\nConsider the following Bernoulli sampling model.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(alpha, beta);\n  for (n in 1:N) {\n    y[n] ~ bernoulli(theta);\n  }\n}\nIn this model, the sum of positive outcomes in y is a sufficient statistic for the chance of success theta. The model may be recoded using the binomial distribution as follows.\ntheta ~ beta(alpha, beta);\nsum(y) ~ binomial(N, theta);\nBecause truth is represented as one and falsehood as zero, the sum sum(y) of a binary vector y is equal to the number of positive outcomes out of a total of N trials.\nThis can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only \\(K\\) possible discrete outcomes, \\(z_1, \\dotsc, z_K\\), but there are \\(N\\) observations, where \\(N\\) is much larger than \\(K\\). If \\(f_k\\) is the frequency of outcome \\(z_k\\), then the entire likelihood with distribution foo can be coded as follows.\nfor (k in 1:K) {\n  target += f[k] * foo_lpmf(z[k] | ...);\n}\nwhere the ellipses are the parameters of the log probability mass function for distribution foo (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).\nThe resulting program looks like a “weighted” regression, but here the weights f[k] are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies f[k] should be counted once in the transformed data block and stored.\nThe same trick works for combining multiple binomial observations.\n\n\n\nConsider the following Stan model for fitting a normal distribution to data.\ndata {\n  int N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nWith the vectorized form used for y, Stan is clever enough to only evaluate log(sigma) once, but it still has to evaluate the normal for all of y[1] to y[N], which involves adding up all the squared differences from the mean and then dividing by sigma squared.\nAn equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.\ndata {\n  int N;\n  vector[N] y;\n}\ntransformed data {\n  real mean_y = mean(y);\n  real&lt;lower=0&gt; var_y = variance(y);\n  real nm1_over2 = 0.5 * (N - 1);\n  real sqrt_N = sqrt(N);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mean_y ~ normal(mu, sigma / sqrt_N);\n  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^2);\n}\nThe data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.\nThe original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on mu and sigma are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.\nThis transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.\n\n\n\nThe Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace\ny ~ poisson(lambda);\nwith\nsum(y) ~ poisson(size(y) * lambda);\nThis will work even if y is a parameter vector because no Jacobian adjustment is required for summation.\n\n\n\n\nIf an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using exp(theta) in multiple places, declare a local variable to store its value and reuse the local variable.\nAnother case that may not be so obvious is with two multilevel parameters, say a[ii[n]] + b[jj[n]]. If a and b are small (i.e., do not have many levels), then a table a_b of their sums can be created, with\nmatrix[size(a), size(b)] a_b;\nfor (i in 1:size(a)) {\n  for (j in 1:size(b)) {\n    a_b[i, j] = a[i] + b[j];\n  }\n}\nThen the sum can be replaced with a_b[ii[n], jj[n]].\n\n\n\nContinuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.\ntheta ~ beta(alpha + sum(y), beta + N - sum(y));\nTo make the model even more efficient, a transformed data variable defined to be sum(y) could be used in the place of sum(y).\n\n\n\nStandardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:\n\nIt helps in faster convergence of MCMC chains.\nIt makes the model less sensitive to the specifics of the parameterization.\nIt aids in the interpretation and comparison of the importance of coefficients across different predictors.\n\nWhen there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.\nSuppose that \\(y = (y_1,\\dotsc,y_N)\\) is a vector of \\(N\\) outcomes and \\(x = (x_1,\\dotsc,x_N)\\) the corresponding vector of \\(N\\) predictors. A simple linear regression involving an intercept coefficient \\(\\alpha\\) and slope coefficient \\(\\beta\\) can be expressed as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n,\n\\] where \\[\n\\epsilon_n \\sim \\textsf{normal}(0,\\sigma).\n\\]\nIf \\(x\\) has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values \\(x_n\\). First the elements of \\(x\\) are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.\nThe mean of \\(x\\) is given by:\n\\[\nmean_x = \\frac{1}{N} \\sum_{n=1}^{N} x_n\n\\]\nThe standard deviation of \\(x\\) is calculated as: \\[\nsd_x = {\\left({\\frac{1}{N} \\sum_{n=1}^{N} (x_n - mean_x)^2}\\right)}^{1/2}\n\\]\nWith these, we compute the \\(z\\), the standardized predictors\n\\[\nz_n = \\frac{x_n - mean_x}{sd_x}\n\\]\nwhere \\(z_n\\) is the standardized value corresponding to \\(x_n\\).\nThe inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.\n\\[\nx_n = z_n sd_x + mean_x\n\\]\nStandardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.\nConsider the following initial model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  // priors\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ normal(0, 5);\n  // likelihood\n  y ~ normal(x * beta + alpha, sigma);\n}\nThe data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\ntransformed data {\n  real mean_x = mean(x);\n  real sd_x = sd(x);\n  vector[N] x_std = (x - mean_x) / sd_x;\n}\nparameters {\n  real alpha_std;\n  real beta_std;\n  real&lt;lower=0&gt; sigma_std;\n}\nmodel {\n  alpha_std ~ normal(0, 10);\n  beta_std ~ normal(0, 10);\n  sigma_std ~ normal(0, 5);\n  y ~ normal(x_std * beta_std + alpha_std, sigma_std);\n}\nThe parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data x_std is defined in terms of variables mean_x and sd_x; by declaring these variables in the transformed data block, they will be available in all following blocks, and therefore can be used in the generated quantities block to record the “natural” parameters alpha and beta.\nThe fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.\nThe original regression \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\] has been transformed to a regression on the standardized data variable \\(z\\),\n\\[\ny_n = \\alpha' + \\beta' z_n + \\epsilon_n.\n\\]\nThe likelihood is specified in terms of the standardized parameters. The original slope \\(\\beta\\) is the standardized slope \\(\\beta'\\) scaled by the inverse of the standard deviation of \\(x\\). The original intercept \\(\\alpha\\) is the intercept from the standardized model \\(\\alpha'\\), corrected for the effect of scaling and centering \\(x\\). Thus, the formulas to retrieve \\(\\alpha\\) and \\(\\beta\\) from \\(\\alpha'\\) and \\(\\beta'\\) are:\n\\[\\begin{align*}\n\\beta = \\frac{\\beta'}{\\sigma_x} \\\\\n\\alpha = \\alpha' - \\beta' \\frac{\\mu_x}{\\sigma_x}\n\\end{align*}\\]\nThese recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,\ngenerated quantities {\n  real beta = beta_std / sd_x;\n  real alpha = alpha_std - beta_std * mean_x / sd_x;\n\n}\nWhen there are multiple real-valued predictors, i.e., when K is the number of predictors, x is an \\(N \\times K\\) matrix, and beta ia \\(K\\)-vector of coefficients, then x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. When \\(K \\ll N\\) the QR reparameterization is recommended for linear and generalized linear models unless there is an informative prior on the location of \\(\\beta\\).\n\n\nFor many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use\ny ~ std_normal();\nthan to use\ny ~ normal(0, 1);\nbecause the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.\n\n\n\n\nThe map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the map-reduce chapter for more information on implementing map-reduce operations.\nMap-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.",
+    "text": "A more general scenario involves an outcome variable \\(y\\) and several explanatory variables \\(x_1, \\ldots, x_K\\), each of which can have missing values that we want to impute.\nOne solution is to do multiple imputation through chained equations, a procedure often called “MICE”1. The MICE procedure in this scenario is:\n\nInitialize missing values in all \\(x_i\\). For each variable \\(x_i\\) with missing entries, fill its missing values with random samples from its observed values (or use another simple initialization rule).\nUpdate each \\(x_i\\) given \\(y\\) and the other \\(x\\)’s. For \\(i=1, \\ldots, K\\), fit a model for the observed \\(x_i\\) conditional on the current versions (with observed and imputed values) of \\(y\\) and \\(x_{-k} = x_1, \\ldots x_{k-1}, x_{k+1}, \\ldots,\nx_{K}\\). Use this model to draw impuations from the predictive distribution of the missing \\(x_i\\). Completing this step for all \\(i = 1, \\ldots, K\\) constitutes a single imputation cycle.\nWarmup period. Repeat the imputation cycle in step 1 several times as warmup to let the imputations stabilize.\nCreate M imputed datasets. After the warmup, record the current complete dataset as one imputed dataset. Then either restart from step 0 and repeat steps 1–3 until you have \\(M\\) imputed datasets; or do a single long run, i.e., continue iterating steps 1–2 and save the imputed dataset at \\(M\\) well-spaced iterations (e.g., every \\(S\\) cycles) to obtain \\(M\\) imputed datasets, without restarting from step 0 each time.\nFit the target model \\(p(y \\mid \\beta, \\sigma, x_1, \\ldots,\nx_K)\\) separately to each of the \\(M\\) imputed datasets and save the posterior draws for the parameters of interest.\nCombine the draws (or other summaries) from all \\(M\\) fits.\n\nNote that, as described here, the MICE procedure does not guarantee that the conditional distributions of all variables will be compatible. Compatibility means that there is a joint distribution for all the variables used in the imputation that can be decomposed as the conditional distributions we used. An incompatible imputation model will technically not sample from any well-defined probability distribution. But the consequences of this are not always serious. See section 6.4 in Carpenter et al. (2023) for a more detailed explanation of compatibility and of the related concept of congeniality.\n\n\nImagine that we again want to use numerical variables \\(x\\) and \\(y\\), and dichotomic variable \\(z\\) to estimate the parameters in \\(p(y \\mid \\beta, \\sigma, x, z)\\). But now both \\(x\\) and \\(z\\) have missing values that we want to impute.\nTo apply the MICE procedure in this example, we can reuse the models for \\(x\\) and \\(y\\) that we defined above. We also need a new model to impute \\(z^{\\text{mis}}\\), so we use the logistic regression \\[\nz^{\\text{obs}} \\sim \\text{Bernoulli}(\n  \\text{logit}^{-1}(\n    \\alpha_0 + \\alpha_1 y^{\\text{obs}} + \\alpha_2 x^{\\text{obs}})\n  ).\n\\] Here, \\(z^{\\text{obs}}\\) contains only the completely observed values from our original data set, while \\(x^{\\text{obs}}\\) and \\(y^{\\text{obs}}\\) contain all the values that correspond to \\(z^{\\text{obs}}\\). Thus, \\(x^{\\text{obs}}\\) can include observed and imputed values, and \\(y^{\\text{obs}}\\) need not be the same as in the model for \\(x^{\\text{obs}}\\).\nWith this model we can sample values for \\(z^{\\text{mis}}\\) conditional on its corresponding values \\(y^{\\text{aux}}\\) and \\(x^{\\text{aux}}\\).\nThe Stan code to impute \\(z^{\\text{mis}}\\) is:\ndata {\n  int&lt;lower=0&gt; N_obs;\n  array[N_obs] int&lt;lower=0, upper=1&gt; z_obs;\n  vector[N_obs] y_obs;\n  vector[N_obs] x_obs;\n  int&lt;lower=0&gt; N_mis;\n  vector[N_mis] y_aux;\n  vector[N_mis] x_aux;\n}\nparameters {\n  vector[3] gamma;\n}\nmodel {\n  gamma ~ normal(0, 1);\n  z_obs ~ bernoulli_logit(gamma[1] + gamma[2] * y_obs \n                          + gamma[3] * x_obs);\n}\ngenerated quantities {\n  array[N_mis] int z_imp = bernoulli_logit_rng(gamma[1]\n                      + gamma[2] * y_aux\n                      + gamma[3] * x_aux);\n}\nThe pseudocode for this example of MICE, restarting the imputations after each imputation cycle, is shown below. Note that we do not need to initialize \\(x^{\\text{mis}}\\) in step 0 because we can impute it directly with the model for \\(x^{\\text{obs}}\\).\ncompleted_datasets &lt;- []\nfor (m in 1:M) {\n  data_m &lt;- copy(data_orig)\n  data_m.z[missing_idz] &lt;- random_sample(data_m.z[observed_idz])\n  for (cycle in 1:n_warm) {\n    stanmod_x &lt;- build_model(\"model_for_x.stan\")\n    fit_x &lt;- stanmod_x.fit(\n      obs_data=data_m[observed_idx],\n      mis_data=data_m[missing_idx]\n    )\n    data_m.x[missing_idx] &lt;- get_imputations(fit_x, \"x_imp\")\n    stanmod_z &lt;- build_model(\"model_for_z.stan\")\n    fit_z &lt;- stanmod_z.fit(\n      obs_data=data_m[observed_idz],\n      mis_data=data_m[missing_idz]\n    )\n    data_m.z[missing_idz] &lt;- get_imputations(fit_z, \"z_imp\")\n  }\n  completed_datasets.append(data_m)\n}\nall_draws &lt;- []\nfor (dataset in completed_datasets) {\n  stanmod_y &lt;- build_model(\"model_for_y.stan\")\n  fit_y &lt;- stanmod_y.fit(dataset)\n  all_draws.append(extract_draws(model_results))\n}\nall_results &lt;- combine_results(all_draws)",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#what-is-efficiency",
-    "href": "stan-users-guide/efficiency-tuning.html#what-is-efficiency",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/multiple-imputation.html#combining-posterior-draws",
+    "href": "stan-users-guide/multiple-imputation.html#combining-posterior-draws",
+    "title": "Multiple imputation",
     "section": "",
-    "text": "The standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).",
+    "text": "With Stan’s MCMC sampler, we can treat posterior draw chains from imputed data sets as if they were chains based on complete data. There is one important difference. Multiple imputation expresses uncertainty in the missing values as consistent differences in the estimates obtained from different imputated data sets. This means that chains obtained from the same imputed data set should converge, but chains obtained from different data sets do not have to. So, we need not worry if diagnostics2 signal that the chains from different imputed data sets are not converging properly. See an example in Bürkner (2025).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#efficiency-for-probabilistic-models-and-algorithms",
-    "href": "stan-users-guide/efficiency-tuning.html#efficiency-for-probabilistic-models-and-algorithms",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/multiple-imputation.html#cut-models",
+    "href": "stan-users-guide/multiple-imputation.html#cut-models",
+    "title": "Multiple imputation",
     "section": "",
-    "text": "Stan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.\nWith probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.\nA well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.",
+    "text": "A full Bayesian probability model includes a feedback flow of information between all parameters and all data. Cut models separate some parts of this feedback flow so that different subsets of data influence only some parameters in the model (see Plummer (2015)).\nMultiple imputation interrupts the flow of information from data to parameters. In our regression above, for example, the imputations influence the distribution of the parameters in \\(p(y \\mid x_1, \\ldots, x_K)\\), but these parameters do not influence the imputations. So, we could use multiple imputation to implement a cut model.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#statistical-vs.-computational-efficiency",
-    "href": "stan-users-guide/efficiency-tuning.html#statistical-vs.-computational-efficiency",
-    "title": "Efficiency Tuning",
-    "section": "",
-    "text": "There is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.\nStatistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.",
+    "objectID": "stan-users-guide/multiple-imputation.html#footnotes",
+    "href": "stan-users-guide/multiple-imputation.html#footnotes",
+    "title": "Multiple imputation",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nSection 5 of chapter 4 in van Buuren (2018) details the MICE procedure in a frequentist context.↩︎\nSuch as \\(\\hat{R}\\). See Split R-hat for detecting non-stationarity in the Reference Manual.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Posterior Inference & Model Checking",
+      "Multiple imputation"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature",
-    "href": "stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html",
+    "href": "stan-users-guide/missing-data.html",
+    "title": "Missing Data and Partially Known Parameters",
     "section": "",
-    "text": "Because Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.\nA second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density \\(\\log\np(\\theta)\\) with respect to the parameter vector \\(\\theta\\), defined as \\[\nH(\\theta) = \\nabla \\, \\nabla \\, \\log p(\\theta \\mid y),\n\\] so that \\[\nH_{i, j}(\\theta) = \\frac{\\partial^2 \\log p(\\theta \\mid y)}\n                {\\partial \\theta_i \\ \\partial \\theta_j}.\n\\] For pure penalized maximum likelihood problems, the posterior log density \\(\\log p(\\theta \\mid y)\\) is replaced by the penalized likelihood function \\(\\mathcal{L}(\\theta) = \\log p(y \\mid \\theta) - \\lambda(\\theta)\\).\n\n\nA good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix \\(H\\), which is the ratio of the largest to the smallest eigenvalue of \\(H\\) (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.\n\n\n\nIdeally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.\n\n\n\nIn all but very simple models (such as multivariate normals), the Hessian will vary as \\(\\theta\\) varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.\nFor all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.\nWith sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.\n\n\n\nImproving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the change of variables chapter.\nOne has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.",
+    "text": "Bayesian inference supports a general approach to missing data in which any missing data item is represented as a parameter that is estimated in the posterior (Gelman et al. 2013). If the missing data are not explicitly modeled, as in the predictors for most regression models, then the result is an improper prior on the parameter representing the missing predictor.\nMixing arrays of observed and missing data can be difficult to include in Stan, partly because it can be tricky to model discrete unknowns in Stan and partly because unlike some other statistical languages (for example, R and Bugs), Stan requires observed and unknown quantities to be defined in separate places in the model. Thus it can be necessary to include code in a Stan program to splice together observed and missing parts of a data structure. Examples are provided later in the chapter.\n\n\nStan treats variables declared in the data and transformed data blocks as known and the variables in the parameters block as unknown.\nAn example involving missing normal observations could be coded as follows.1\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  array[N_mis] real y_mis;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_mis ~ normal(mu, sigma);\n}\nThe number of observed and missing data points are coded as data with non-negative integer variables N_obs and N_mis. The observed data are provided as an array data variable y_obs. The missing data are coded as an array parameter, y_mis. The ordinary parameters being estimated, the location mu and scale sigma, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.\n\n\n\nIn some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the transformed parameters block and assigning to it.\nThe following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] vector[2] y;\n  real&lt;lower=0&gt; var1;\n  real&lt;lower=0&gt; var2;\n}\ntransformed data {\n  real&lt;lower=0&gt; max_cov = sqrt(var1 * var2);\n  real&lt;upper=0&gt; min_cov = -max_cov;\n}\nparameters {\n  vector[2] mu;\n  real&lt;lower=min_cov, upper=max_cov&gt; cov;\n}\ntransformed parameters {\n  matrix[2, 2] Sigma;\n  Sigma[1, 1] = var1;\n  Sigma[1, 2] = cov;\n  Sigma[2, 1] = cov;\n  Sigma[2, 2] = var2;\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);\n}\nThe variances are defined as data in variables var1 and var2, whereas the covariance is defined as a parameter in variable cov. The \\(2 \\times 2\\) covariance matrix Sigma is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.\nThe constraint on the covariance declaration ensures that the resulting covariance matrix sigma is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.\nThe vectorization of the multivariate normal is critical for efficiency here. The transformed parameter Sigma could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.\n\n\n\nIf the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_obs;\n  array[N_mis] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_mis;\n  array[N_obs] real y_obs;\n}\ntransformed data {\n  int&lt;lower=0&gt; N = N_obs + N_mis;\n}\nparameters {\n  array[N_mis] real y_mis;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  array[N] real y;\n  y[ii_obs] = y_obs;\n  y[ii_mis] = y_mis;\n}\nmodel {\n  sigma ~ gamma(1, 1);\n  y[1] ~ normal(0, 100);\n  y[2:N] ~ normal(y[1:(N - 1)], sigma);\n}\nThe index arrays ii_obs and ii_mis contain the indexes into the final array y of the observed data (coded as a data vector y_obs) and the missing data (coded as a parameter vector y_mis). See the time series chapter for further discussion of time-series model and specifically the autoregression section for an explanation of the vectorization for y as well as an explanation of how to convert this example to a full AR(1) model. To ensure y[1] has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.\nAnother potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in\nx[N_obs_2, 2] = x_obs_2;\nx[N_mis_2, 2] = x_mis_2;\nwhere the relevant variables are all hard coded with index 2 because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the ragged data structures section for a general discussion of coding ragged data structures in Stan).\n\n\n\nRick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis (Aguilar and West 2000). This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.\ndata {\n  int&lt;lower=2&gt; K;\n}\ntransformed data {\n  int&lt;lower=1&gt; K_choose_2;\n  K_choose_2 = (K * (K - 1)) / 2;\n}\nparameters {\n  vector[K_choose_2] L_lower;\n}\ntransformed parameters {\n  cholesky_factor_cov[K] L;\n  for (k in 1:K) {\n    L[k, k] = 1;\n  }\n  {\n    int i;\n    for (m in 2:K) {\n      for (n in 1:(m - 1)) {\n        L[m, n] = L_lower[i];\n        L[n, m] = 0;\n        i += 1;\n      }\n    }\n  }\n}\nIt is most convenient to place a prior directly on L_lower. An alternative would be a prior for the full Cholesky factor L, because the transform from L_lower to L is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix L   * L', because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.\n\n\n\nIt’s often the case that one or more components of a multivariate outcome are missing.2\nAs an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations y and a boolean array y_observed to indicate which values were observed (others can have dummy values in the input).\narray[N] vector[2] y;\narray[N, 2] int&lt;lower=0, upper=1&gt; y_observed;\nIf both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    y[n] ~ multi_normal(mu, Sigma);\n  } else if (y_observed[n, 1]) {\n    y[n, 1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\n  } else if (y_observed[n, 2]) {\n    y[n, 2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\n  }\n}\nIt’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:\ntransformed data {\n  array[observed_12(y_observed)] int ns12;\n  array[observed_1(y_observed)] int ns1;\n  array[observed_2(y_observed)] int ns2;\n}\nYou will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.\nint n12 = 1;\nint n1 = 1;\nint n2 = 1;\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    ns12[n12] = n;\n    n12 += 1;\n  } else if (y_observed[n, 1]) {\n    ns1[n1] = n;\n    n1 += 1;\n  } else if (y_observed[n, 2]) {\n    ns2[n2] = n;\n    n2 += 1;\n  }\n}\nThen, in the model block, everything is vectorizable using those indexes constructed once in transformed data:\ny[ns12] ~ multi_normal(mu, Sigma);\ny[ns1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\ny[ns2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\nThe result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.\nThis approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#well-specified-models",
-    "href": "stan-users-guide/efficiency-tuning.html#well-specified-models",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html#missing-data",
+    "href": "stan-users-guide/missing-data.html#missing-data",
+    "title": "Missing Data and Partially Known Parameters",
     "section": "",
-    "text": "Model misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.\nCounterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as normal(0, 1000)). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as normal(mu,   sigma), where mu and sigma are new parameters with their own hyperpriors, can be beneficial.",
+    "text": "Stan treats variables declared in the data and transformed data blocks as known and the variables in the parameters block as unknown.\nAn example involving missing normal observations could be coded as follows.1\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  array[N_mis] real y_mis;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_mis ~ normal(mu, sigma);\n}\nThe number of observed and missing data points are coded as data with non-negative integer variables N_obs and N_mis. The observed data are provided as an array data variable y_obs. The missing data are coded as an array parameter, y_mis. The ordinary parameters being estimated, the location mu and scale sigma, are also coded as parameters. The model is vectorized on the observed and missing data; combining them in this case would be less efficient because the data observations would be promoted and have needless derivatives calculated.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#avoiding-validation",
-    "href": "stan-users-guide/efficiency-tuning.html#avoiding-validation",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html#partially-known-parameters.section",
+    "href": "stan-users-guide/missing-data.html#partially-known-parameters.section",
+    "title": "Missing Data and Partially Known Parameters",
     "section": "",
-    "text": "Stan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ...\n}                               // first validation\nmodel {\n  y ~ multi_normal(mu, Sigma);  // second validation\n  // ...\n}\nBecause Sigma is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that Sigma is positive definite. This test is expensive, having cubic run time (i.e., \\(\\mathcal{O}(N^3)\\) for \\(N \\times N\\) matrices), so it should not be done twice.\nThe test may be avoided by simply declaring Sigma to be a simple unconstrained matrix.\ntransformed parameters {\n  matrix[K, K] Sigma;\n  // ...\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);  // only validation\n}\nNow the only validation is carried out by the multivariate normal.",
+    "text": "In some situations, such as when a multivariate probability function has partially observed outcomes or parameters, it will be necessary to create a vector mixing known (data) and unknown (parameter) values. This can be done in Stan by creating a vector or array in the transformed parameters block and assigning to it.\nThe following example involves a bivariate covariance matrix in which the variances are known, but the covariance is not.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] vector[2] y;\n  real&lt;lower=0&gt; var1;\n  real&lt;lower=0&gt; var2;\n}\ntransformed data {\n  real&lt;lower=0&gt; max_cov = sqrt(var1 * var2);\n  real&lt;upper=0&gt; min_cov = -max_cov;\n}\nparameters {\n  vector[2] mu;\n  real&lt;lower=min_cov, upper=max_cov&gt; cov;\n}\ntransformed parameters {\n  matrix[2, 2] Sigma;\n  Sigma[1, 1] = var1;\n  Sigma[1, 2] = cov;\n  Sigma[2, 1] = cov;\n  Sigma[2, 2] = var2;\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);\n}\nThe variances are defined as data in variables var1 and var2, whereas the covariance is defined as a parameter in variable cov. The \\(2 \\times 2\\) covariance matrix Sigma is defined as a transformed parameter, with the variances assigned to the two diagonal elements and the covariance to the two off-diagonal elements.\nThe constraint on the covariance declaration ensures that the resulting covariance matrix sigma is positive definite. The bound, plus or minus the square root of the product of the variances, is defined as transformed data so that it is only calculated once.\nThe vectorization of the multivariate normal is critical for efficiency here. The transformed parameter Sigma could be defined as a local variable within the model block if it does not need to be included in the sampler’s output.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#reparameterization.section",
-    "href": "stan-users-guide/efficiency-tuning.html#reparameterization.section",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html#sliced-missing-data",
+    "href": "stan-users-guide/missing-data.html#sliced-missing-data",
+    "title": "Missing Data and Partially Known Parameters",
     "section": "",
-    "text": "Stan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.\n\n\nIn this section, we discuss a general transform from a centered to a non-centered parameterization (Papaspiliopoulos, Roberts, and Sköld 2007).1\nThis reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.\nNeal (2003) defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for \\(y \\in\n\\mathbb{R}\\) and \\(x \\in \\mathbb{R}^9\\) with density\n\\[\np(y,x) = \\textsf{normal}(y \\mid 0,3) \\times \\prod_{n=1}^9\n\\textsf{normal}(x_n \\mid 0,\\exp(y/2)).\n\\]\nThe probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to \\(y\\). A plot of the log marginal density of \\(y\\) and the first dimension \\(x_1\\) is shown in the following plot.\nThe funnel can be implemented directly in Stan as follows.\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y/2));\n}\nWhen the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where \\(y\\) is small and thus \\(x\\) is constrained to be near 0. This is due to the fact that the density’s scale changes with \\(y\\), so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.\n\n\n\n\n\n\n\n\n\n\n\n\n\nFigure 1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable \\(y\\) and one lower-level variable \\(x_1\\) (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of \\(x_1\\) and \\(y\\) values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density \\(p(y) = \\textsf{normal}(y \\mid 0,3)\\), which has mean 0 and standard deviation 3.\n\n\n\nIn this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y/2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y/2))\n}\nIn this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.\n\n\n\nSampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.\nThe practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.\nSuppose a random variable of interest \\(X\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), so that \\(X \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\). The variable \\(X\\) has a cumulative distribution function \\(F_X:\\mathbb{R} \\rightarrow (0,1)\\) defined by \\[\nF_X(x) = \\frac{1}{\\pi} \\arctan \\left( \\frac{x - \\mu}{\\tau} \\right) +\n\\frac{1}{2}.\n\\] The inverse of the cumulative distribution function, \\(F_X^{-1}:(0,1) \\rightarrow \\mathbb{R}\\), is thus\n\\[\nF^{-1}_X(y) = \\mu + \\tau \\tan \\left( \\pi \\left( y - \\frac{1}{2} \\right) \\right).\n\\] Thus if the random variable \\(Y\\) has a unit uniform distribution, \\(Y\n\\sim \\textsf{uniform}(0,1)\\), then \\(F^{-1}_X(Y)\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), i.e., \\(F^{-1}_X(Y) \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\).\nConsider a Stan program involving a Cauchy-distributed parameter beta.\nparameters {\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ cauchy(mu, tau);\n  // ...\n}\nThis declaration of beta as a parameter may be replaced with a transformed parameter beta defined in terms of a uniform-distributed parameter beta_unif.\nparameters {\n  real&lt;lower=-pi() / 2, upper=pi() / 2&gt; beta_unif;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = mu + tau * tan(beta_unif);  // beta ~ cauchy(mu, tau)\n}\nmodel {\n  beta_unif ~ uniform(-pi() / 2, pi() / 2);  // not necessary\n  // ...\n}\nIt is more convenient in Stan to transform a uniform variable on \\((-\\pi/2, \\pi/2)\\) than one on \\((0,1)\\). The Cauchy location and scale parameters, mu and tau, may be defined as data or may themselves be parameters. The variable beta could also be defined as a local variable if it does not need to be included in the sampler’s output.\nThe uniform distribution on beta_unif is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because \\(\\log\n\\textsf{uniform}(\\beta_{\\textsf{unif}} \\mid -\\pi/2,\\pi/2) =\n-\\log \\pi\\) is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that beta_unif is between -pi() / 2 and pi() / 2, but this condition is guaranteed by the constraints in the declaration of beta_unif.\n\n\n\nOne thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable \\(\\beta\\), \\[\n\\beta \\sim \\textsf{Student-t}(\\nu, 0, 1),\n\\] by first generating a gamma-distributed precision (inverse variance) \\(\\tau\\) according to \\[\n\\tau \\sim \\textsf{Gamma}(\\nu/2, \\nu/2),\n\\] and then generating \\(\\beta\\) from the normal distribution, \\[\n\\beta \\sim \\textsf{normal}\\left(0,\\tau^{-\\frac{1}{2}}\\right).\n\\]\nBecause \\(\\tau\\) is precision, \\(\\tau^{-\\frac{1}{2}}\\) is the scale (standard deviation), which is the parameterization used by Stan.\nThe marginal distribution of \\(\\beta\\) when you integrate out \\(\\tau\\) is \\(\\textsf{Student-t}(\\nu, 0, 1)\\), i.e., \\[\n\\textsf{Student-t}(\\beta \\mid \\nu, 0, 1)\n=\n\\int_0^{\\infty}\n\\,\n\\textsf{normal}\\left(\\beta \\middle| 0, \\tau^{-0.5}\\right)\n\\times\n\\textsf{Gamma}\\left(\\tau \\middle| \\nu/2, \\nu/2\\right)\n\\\n\\text{d} \\tau.\n\\]\nTo go one step further, instead of defining a \\(\\beta\\) drawn from a normal with precision \\(\\tau\\), define \\(\\alpha\\) to be drawn from a unit normal, \\[\n\\alpha \\sim \\textsf{normal}(0,1)\n\\] and rescale by defining \\[\n\\beta = \\alpha \\, \\tau^{-\\frac{1}{2}}.\n\\]\nNow suppose \\(\\mu = \\beta x\\) is the product of \\(\\beta\\) with a regression predictor \\(x\\). Then the reparameterization \\(\\mu = \\alpha\n\\tau^{-\\frac{1}{2}} x\\) has the same distribution, but in the original, direct parameterization, \\(\\beta\\) has (potentially) heavy tails, whereas in the second, neither \\(\\tau\\) nor \\(\\alpha\\) have heavy tails.\nTo translate into Stan notation, this reparameterization replaces\nparameters {\n  real&lt;lower=0&gt; nu;\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ student_t(nu, 0, 1);\n  // ...\n}\nwith\nparameters {\n  real&lt;lower=0&gt; nu;\n  real&lt;lower=0&gt; tau;\n  real alpha;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = alpha / sqrt(tau);\n  // ...\n}\nmodel {\n  real half_nu;\n  half_nu = 0.5 * nu;\n  tau ~ gamma(half_nu, half_nu);\n  alpha ~ std_normal();\n  // ...\n}\nAlthough set to 0 here, in most cases, the lower bound for the degrees of freedom parameter nu can be set to 1 or higher; when nu is 1, the result is a Cauchy distribution with fat tails and as nu approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter nu characterizes the heaviness of the tails of the model.\n\n\n\nUnfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.\n\n\nFor example, a vectorized hierarchical model might draw a vector of coefficients \\(\\beta\\) with definitions as follows. The so-called centered parameterization is as follows.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nAlthough not shown, a full model will have priors on both mu_beta and sigma_beta along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix x and binary outcome vector y would include a likelihood statement such as form y ~ bernoulli_logit(x * beta), leading to an analytically intractable posterior.\nA hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of beta, mu_beta and sigma_beta are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See Betancourt and Girolami (2013) for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.\n\n\n\n\nSometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see Betancourt and Girolami (2013)), and in more extreme cases will be necessary to achieve convergence.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta_raw;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // implies: beta ~ normal(mu_beta, sigma_beta)\n  beta = mu_beta + sigma_beta * beta_raw;\n}\nmodel {\n  beta_raw ~ std_normal();\n  // ...\n}\nAny priors defined for mu_beta and sigma_beta remain as defined in the original model.\nAlternatively, Stan’s affine transform can be used to decouple sigma and beta:\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;offset=mu_beta, multiplier=sigma_beta&gt;[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nReparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let \\(\\beta = l + s\\alpha\\) where \\(l\\) is a location parameter and \\(s\\) is a scale parameter. The parameter \\(l\\) need not be the mean, \\(s\\) need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If \\(\\alpha\\) and \\(\\beta\\) are from the same distributional family but \\(\\alpha\\) has location zero and unit scale, while \\(\\beta\\) has location \\(l\\) and scale \\(s\\), then that distribution is a location-scale distribution. Thus, if \\(\\alpha\\) were a parameter and \\(\\beta\\) were a transformed parameter, then a prior distribution from the location-scale family on \\(\\alpha\\) with location zero and unit scale implies a prior distribution on \\(\\beta\\) with location \\(l\\) and scale \\(s\\). Doing so would reduce the dependence between \\(\\alpha\\), \\(l\\), and \\(s\\).\nThere are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if \\(\\alpha\\) is distributed standard normal, then \\(\\beta\\) is distributed normal with mean \\(\\mu = l\\) and standard deviation \\(\\sigma = s\\). The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.\nAlso, if \\(z\\) is distributed standard normal, then \\(z^2\\) is distributed chi-squared with one degree of freedom. By summing the squares of \\(K\\) independent standard normal variates, one can obtain a single variate that is distributed chi-squared with \\(K\\) degrees of freedom. However, for large \\(K\\), the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying \\(K\\) primitive parameters just to obtain one transformed parameter to use in a model.\n\n\n\nThe benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for \\(\\beta\\) to be multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). Such a belief is reflected by the following code.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\nparameters {\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ multi_normal(mu, Sigma);\n  // ...\n}\nIn this case mu and Sigma are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of beta.\nIf \\(\\alpha\\) has the same dimensions as \\(\\beta\\) but the elements of \\(\\alpha\\) are independently and identically distributed standard normal such that \\(\\beta = \\mu + L\\alpha\\), where \\(LL^\\top = \\Sigma\\), then \\(\\beta\\) is distributed multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). One choice for \\(L\\) is the Cholesky factor of \\(\\Sigma\\). Thus, the model above could be reparameterized as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Sigma);\n}\nparameters {\n  vector[K] alpha;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  beta = mu + L * alpha;\n}\nmodel {\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu, Sigma)\n  // ...\n}\nThis reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of alpha and second, it avoids the need to invert Sigma every time multi_normal is evaluated.\nThe Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix Tau were fixed data but the vector of standard deviations sigma were unknown parameters, then a reparameterization of beta in terms of alpha could be implemented as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  corr_matrix[K] Tau;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Tau);\n}\nparameters {\n  vector[K] alpha;\n  vector&lt;lower=0&gt;[K] sigma;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // This equals mu + diag_matrix(sigma) * L * alpha;\n  beta = mu + sigma .* (L * alpha);\n}\nmodel {\n  sigma ~ cauchy(0, 5);\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu,\n  //  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))\n  // ...\n}\nThis reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.\nA Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let \\(L\\) be the Cholesky factor of a \\(K \\times K\\) positive definite scale matrix \\(S\\) and let \\(\\nu\\) be the degrees of freedom. If \\[\nA = \\begin{pmatrix}\n\\sqrt{c_{1}} & 0            & \\cdots                & 0 \\\\\nz_{21}       & \\sqrt{c_{2}} & \\ddots                & \\vdots \\\\\n\\vdots       & \\ddots       & \\ddots                & 0 \\\\\nz_{K1}       & \\cdots       & z_{K\\left(K-1\\right)} & \\sqrt{c_{K}}\n\\end{pmatrix},\n\\] where each \\(c_i\\) is distributed chi-squared with \\(\\nu - i + 1\\) degrees of freedom and each \\(z_{ij}\\) is distributed standard normal, then \\(W = LAA^{\\top}L^{\\top}\\) is distributed Wishart with scale matrix \\(S = LL^{\\top}\\) and degrees of freedom \\(\\nu\\). Such a reparameterization can be implemented by the following Stan code:\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  vector[K] mu;\n  matrix[N, K] y;\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  int count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal();\n  // implies: L * A * A' * L' ~ wishart(nu, L * L')\n  y ~ multi_normal_cholesky(mu, L * A);\n  // ...\n}\nThis reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of z and second, it avoids the need to invert the covariance matrix, \\(W\\) every time wishart is evaluated. Third, if \\(W\\) is to be used with a multivariate normal distribution, you can pass \\(L A\\) to the more efficient multi_normal_cholesky function, rather than passing \\(W\\) to multi_normal.\nIf \\(W\\) is distributed Wishart with scale matrix \\(S\\) and degrees of freedom \\(\\nu\\), then \\(W^{-1}\\) is distributed inverse Wishart with inverse scale matrix \\(S^{-1}\\) and degrees of freedom \\(\\nu\\). Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since \\(W = L A A^{\\top} L^{\\top}\\), \\(W^{-1} = L^{{\\top}^{-1}} A^{{\\top}^{-1}} A^{-1} L^{-1}\\), where all four inverses exist, but \\(L^{{-1}^{\\top}} = L^{{\\top}^{-1}}\\) and \\(A^{{-1}^{\\top}} = A^{{\\top}^{-1}}\\). We can slightly modify the above Stan code for this case:\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  // ...\n}\ntransformed data {\n  matrix[K, K] eye;\n  matrix[K, K] L_inv;\n  for (j in 1:K) {\n    for (i in 1:K) {\n      eye[i, j] = 0.0;\n    }\n    eye[j, j] = 1.0;\n  }\n  L_inv = mdivide_left_tri_low(L, eye);\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  matrix[K, K] A_inv_L_inv;\n  int count;\n  count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal(); // implies: crossprod(A_inv_L_inv) ~\n  // inv_wishart(nu, L_inv' * L_inv)\n  // ...\n}\nAnother candidate for reparameterization is the Dirichlet distribution with all \\(K\\) shape parameters equal. Zyczkowski and Sommers (2001) shows that if \\(\\theta_i\\) is equal to the sum of \\(\\beta\\) independent squared standard normal variates and \\(\\rho_i = \\frac{\\theta_i}{\\sum \\theta_i}\\), then the \\(K\\)-vector \\(\\rho\\) is distributed Dirichlet with all shape parameters equal to \\(\\frac{\\beta}{2}\\). In particular, if \\(\\beta = 2\\), then \\(\\rho\\) is uniformly distributed on the unit simplex. Thus, we can make \\(\\rho\\) be a transformed parameter to reduce dependence, as in:\ndata {\n  int&lt;lower=1&gt; beta;\n  // ...\n}\nparameters {\n  array[K] vector[beta] z;\n  // ...\n}\ntransformed parameters {\n  simplex[K] rho;\n  for (k in 1:K) {\n    rho[k] = dot_self(z[k]); // sum-of-squares\n  }\n  rho = rho / sum(rho);\n}\nmodel {\n  for (k in 1:K) {\n    z[k] ~ std_normal();\n  }\n  // implies: rho ~ dirichlet(0.5 * beta * ones)\n  // ...\n}",
+    "text": "If the missing data are part of some larger data structure, then it can often be effectively reassembled using index arrays and slicing. Here’s an example for time-series data, where only some entries in the series are observed.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_mis;\n  array[N_obs] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_obs;\n  array[N_mis] int&lt;lower=1, upper=N_obs + N_mis&gt; ii_mis;\n  array[N_obs] real y_obs;\n}\ntransformed data {\n  int&lt;lower=0&gt; N = N_obs + N_mis;\n}\nparameters {\n  array[N_mis] real y_mis;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  array[N] real y;\n  y[ii_obs] = y_obs;\n  y[ii_mis] = y_mis;\n}\nmodel {\n  sigma ~ gamma(1, 1);\n  y[1] ~ normal(0, 100);\n  y[2:N] ~ normal(y[1:(N - 1)], sigma);\n}\nThe index arrays ii_obs and ii_mis contain the indexes into the final array y of the observed data (coded as a data vector y_obs) and the missing data (coded as a parameter vector y_mis). See the time series chapter for further discussion of time-series model and specifically the autoregression section for an explanation of the vectorization for y as well as an explanation of how to convert this example to a full AR(1) model. To ensure y[1] has a proper posterior in case it is missing, we have given it an explicit, albeit broad, prior.\nAnother potential application would be filling the columns of a data matrix of predictors for which some predictors are missing; matrix columns can be accessed as vectors and assigned the same way, as in\nx[N_obs_2, 2] = x_obs_2;\nx[N_mis_2, 2] = x_mis_2;\nwhere the relevant variables are all hard coded with index 2 because Stan doesn’t support ragged arrays. These could all be packed into a single array with more fiddly indexing that slices out vectors from longer vectors (see the ragged data structures section for a general discussion of coding ragged data structures in Stan).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#vectorization",
-    "href": "stan-users-guide/efficiency-tuning.html#vectorization",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html#loading-matrix-for-factor-analysis",
+    "href": "stan-users-guide/missing-data.html#loading-matrix-for-factor-analysis",
+    "title": "Missing Data and Partially Known Parameters",
     "section": "",
-    "text": "Stan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.2 This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.\n\n\n\nBecause of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the sum() operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where foo() is some operation that depends on n.\nfor (n in 1:N) {\n  total += foo(n,...);\n}\nThis code has to create intermediate representations for each of the N summands.\nA faster alternative is to copy the values into a vector, then apply the sum() operator, as in the following refactoring.\n{\n  vector[N] summands;\n  for (n in 1:N) {\n    summands[n] = foo(n,...);\n  }\n  total = sum(summands);\n}\nSyntactically, the replacement is a statement block delineated by curly brackets ({, }), starting with the definition of the local variable summands.\nEven though it involves extra work to allocate the summands vector and copy N values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing total within the loop.\n\n\n\nThe following program directly encodes a linear regression with fixed unit noise using a two-dimensional array x of predictors, an array y of outcomes, and an array beta of regression coefficients.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[K, N] real x;\n  array[N] real y;\n}\nparameters {\n  array[K] real beta;\n}\nmodel {\n  for (n in 1:N) {\n    real gamma = 0;\n    for (k in 1:K) {\n      gamma += x[n, k] * beta[k];\n    }\n    y[n] ~ normal(gamma, 1);\n  }\n}\nThe following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[N] vector[K] x;\n  array[N] real y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(dot_product(x[n], beta), 1);\n  }\n}\nAlthough it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in\ny[n] ~ normal(x[n]' * beta, 1);\nThe relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.3\nThe inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.\n// ...\ntransformed parameters {\n  row_vector[K] beta_t;\n  beta_t = beta';\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(beta_t * x[n], 1);\n  }\n}\nThe problem with transposition could be completely solved by directly encoding the x as a row vector, as in the following example.\ndata {\n  // ...\n  array[N] row_vector[K] x;\n  // ...\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, 1);\n  }\n}\nDeclaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.\nHaving said all this, the most efficient way to code this model is with direct matrix multiplication, as in\ndata {\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nIn general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.\n\n\n\nThe final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nThe variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication x * beta in the model block is a vector of the same length as y.\nThe probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, y is a vector of size N, x * beta is a vector of size N, and 1 is a scalar.\n\n\n\nSometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).\nJohn Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu[ii[n]], sigma);\n}\nThe brute force vectorization would build up a mean vector and then vectorize all at once.\n{\n  vector[N] mu_ii;\n  for (n in 1:N) {\n    mu_ii[n] = mu[ii[n]];\n  }\n  y ~ normal(mu_ii, sigma);\n}\nIf there aren’t many levels (values ii[n] can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector y, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have\ny_1 ~ normal(mu[1], sigma);\n// ...\ny_5 ~ normal(mu[5], sigma);\nThis way, both the mu and sigma parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.",
+    "text": "Rick Farouni, on the Stan users group, inquired as to how to build a Cholesky factor for a covariance matrix with a unit diagonal, as used in Bayesian factor analysis (Aguilar and West 2000). This can be accomplished by declaring the below-diagonal elements as parameters, then filling the full matrix as a transformed parameter.\ndata {\n  int&lt;lower=2&gt; K;\n}\ntransformed data {\n  int&lt;lower=1&gt; K_choose_2;\n  K_choose_2 = (K * (K - 1)) / 2;\n}\nparameters {\n  vector[K_choose_2] L_lower;\n}\ntransformed parameters {\n  cholesky_factor_cov[K] L;\n  for (k in 1:K) {\n    L[k, k] = 1;\n  }\n  {\n    int i;\n    for (m in 2:K) {\n      for (n in 1:(m - 1)) {\n        L[m, n] = L_lower[i];\n        L[n, m] = 0;\n        i += 1;\n      }\n    }\n  }\n}\nIt is most convenient to place a prior directly on L_lower. An alternative would be a prior for the full Cholesky factor L, because the transform from L_lower to L is just the identity and thus does not require a Jacobian adjustment (despite the warning from the parser, which is not smart enough to do the code analysis to infer that the transform is linear). It would not be at all convenient to place a prior on the full covariance matrix L   * L', because that would require a Jacobian adjustment; the exact adjustment is detailed in the reference manual.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#exploiting-sufficient-statistics",
-    "href": "stan-users-guide/efficiency-tuning.html#exploiting-sufficient-statistics",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html#missing-multivariate-data",
+    "href": "stan-users-guide/missing-data.html#missing-multivariate-data",
+    "title": "Missing Data and Partially Known Parameters",
     "section": "",
-    "text": "In some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.\n\n\nConsider the following Bernoulli sampling model.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(alpha, beta);\n  for (n in 1:N) {\n    y[n] ~ bernoulli(theta);\n  }\n}\nIn this model, the sum of positive outcomes in y is a sufficient statistic for the chance of success theta. The model may be recoded using the binomial distribution as follows.\ntheta ~ beta(alpha, beta);\nsum(y) ~ binomial(N, theta);\nBecause truth is represented as one and falsehood as zero, the sum sum(y) of a binary vector y is equal to the number of positive outcomes out of a total of N trials.\nThis can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only \\(K\\) possible discrete outcomes, \\(z_1, \\dotsc, z_K\\), but there are \\(N\\) observations, where \\(N\\) is much larger than \\(K\\). If \\(f_k\\) is the frequency of outcome \\(z_k\\), then the entire likelihood with distribution foo can be coded as follows.\nfor (k in 1:K) {\n  target += f[k] * foo_lpmf(z[k] | ...);\n}\nwhere the ellipses are the parameters of the log probability mass function for distribution foo (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).\nThe resulting program looks like a “weighted” regression, but here the weights f[k] are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies f[k] should be counted once in the transformed data block and stored.\nThe same trick works for combining multiple binomial observations.\n\n\n\nConsider the following Stan model for fitting a normal distribution to data.\ndata {\n  int N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nWith the vectorized form used for y, Stan is clever enough to only evaluate log(sigma) once, but it still has to evaluate the normal for all of y[1] to y[N], which involves adding up all the squared differences from the mean and then dividing by sigma squared.\nAn equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.\ndata {\n  int N;\n  vector[N] y;\n}\ntransformed data {\n  real mean_y = mean(y);\n  real&lt;lower=0&gt; var_y = variance(y);\n  real nm1_over2 = 0.5 * (N - 1);\n  real sqrt_N = sqrt(N);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mean_y ~ normal(mu, sigma / sqrt_N);\n  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^2);\n}\nThe data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.\nThe original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on mu and sigma are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.\nThis transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.\n\n\n\nThe Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace\ny ~ poisson(lambda);\nwith\nsum(y) ~ poisson(size(y) * lambda);\nThis will work even if y is a parameter vector because no Jacobian adjustment is required for summation.",
+    "text": "It’s often the case that one or more components of a multivariate outcome are missing.2\nAs an example, we’ll consider the bivariate distribution, which is easily marginalized. The coding here is brute force, representing both an array of vector observations y and a boolean array y_observed to indicate which values were observed (others can have dummy values in the input).\narray[N] vector[2] y;\narray[N, 2] int&lt;lower=0, upper=1&gt; y_observed;\nIf both components are observed, we model them using the full multi-normal, otherwise we model the marginal distribution of the component that is observed.\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    y[n] ~ multi_normal(mu, Sigma);\n  } else if (y_observed[n, 1]) {\n    y[n, 1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\n  } else if (y_observed[n, 2]) {\n    y[n, 2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\n  }\n}\nIt’s a bit more work, but much more efficient to vectorize these distribution statements. In transformed data, build up three vectors of indices, for the three cases above:\ntransformed data {\n  array[observed_12(y_observed)] int ns12;\n  array[observed_1(y_observed)] int ns1;\n  array[observed_2(y_observed)] int ns2;\n}\nYou will need to write functions that pull out the count of observations in each of the three situations. This must be done with functions because the result needs to go in top-level block variable size declaration. Then the rest of transformed data just fills in the values using three counters.\nint n12 = 1;\nint n1 = 1;\nint n2 = 1;\nfor (n in 1:N) {\n  if (y_observed[n, 1] && y_observed[n, 2]) {\n    ns12[n12] = n;\n    n12 += 1;\n  } else if (y_observed[n, 1]) {\n    ns1[n1] = n;\n    n1 += 1;\n  } else if (y_observed[n, 2]) {\n    ns2[n2] = n;\n    n2 += 1;\n  }\n}\nThen, in the model block, everything is vectorizable using those indexes constructed once in transformed data:\ny[ns12] ~ multi_normal(mu, Sigma);\ny[ns1] ~ normal(mu[1], sqrt(Sigma[1, 1]));\ny[ns2] ~ normal(mu[2], sqrt(Sigma[2, 2]));\nThe result will be much more efficient than using latent variables for the missing data, but it requires the multivariate distribution to be marginalized analytically. It’d be more efficient still to precompute the three arrays in the transformed data block, though the efficiency improvement will be relatively minor compared to vectorizing the probability functions.\nThis approach can easily be generalized with some index fiddling to the general multivariate case. The trick is to pull out entries in the covariance matrix for the missing components. It can also be used in situations such as multivariate differential equation solutions where only one component is observed, as in a phase-space experiment recording only time and position of a pendulum (and not recording momentum).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Efficiency Tuning"
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#aggregating-common-subexpressions",
-    "href": "stan-users-guide/efficiency-tuning.html#aggregating-common-subexpressions",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/missing-data.html#footnotes",
+    "href": "stan-users-guide/missing-data.html#footnotes",
+    "title": "Missing Data and Partially Known Parameters",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nA more meaningful estimation example would involve a regression of the observed and missing observations using predictors that were known for each and specified in the data block.↩︎\nThis is not the same as missing components of a multivariate predictor in a regression problem; in that case, you will need to represent the missing data as a parameter and impute missing values in order to feed them into the regression.↩︎",
+    "crumbs": [
+      "Stan Users Guide",
+      "Example Models",
+      "Missing Data and Partially Known Parameters"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/matrices-arrays.html",
+    "href": "stan-users-guide/matrices-arrays.html",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "If an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using exp(theta) in multiple places, declare a local variable to store its value and reuse the local variable.\nAnother case that may not be so obvious is with two multilevel parameters, say a[ii[n]] + b[jj[n]]. If a and b are small (i.e., do not have many levels), then a table a_b of their sums can be created, with\nmatrix[size(a), size(b)] a_b;\nfor (i in 1:size(a)) {\n  for (j in 1:size(b)) {\n    a_b[i, j] = a[i] + b[j];\n  }\n}\nThen the sum can be replaced with a_b[ii[n], jj[n]].",
+    "text": "This chapter provides pointers as to how to choose among the various container types (matrix, vector, array, and tuple) provided by Stan.\n\n\nStan provides three basic scalar types, int, real, and complex, as well as three basic linear algebra types, vector, row_vector, and matrix. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).\nThis leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.\narray[N] real a;\nvector[N] a;\nrow_vector[N] a;\nThese distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.\nThe following code fragment shows all four ways to declare a two-dimensional container of size \\(M \\times N\\).\narray[M, N] real b;          // b[m] : array[] real     (efficient)\narray[M] vector[N] b;        // b[m] : vector     (efficient)\narray[M] row_vector[N] b;    // b[m] : row_vector (efficient)\nmatrix[M, N] b;              // b[m] : row_vector (inefficient)\nThe main differences among these choices involve efficiency for various purposes and the type of b[m], which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.\nThe inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).\n\n\n\nArrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single complex typed entry, all the entries are complex).\nWith arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a complex_vector[3] holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.\nTuples provide a way to represent a sequence of values of heterogeneous types. For example, tuple(int, real) is the type of a pair consisting of an integer and a real number and tuple(array[5] int, vector[6]) is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.\n\n\nTuples are declared using the keyword tuple followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.\ntuple(int, vector[3]) ny = (5, [3, 2.9, 1.8]');\nThe elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using\nint n = ny.1;\nvector[3] y = ny.2;\nWe can also assign into the elements of a tuple.\ntuple(int, vector[3], complex) abc;\nabc.1 = 5;\nabc.2[1] = 3;\nabc.2[2] = 2.9;\nabc.2[3] = 1.4798;\nabc.3 = 2 + 1.9j;\nAs the cascaded indexing example shows, the result of abc.1 is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., abc.2[1] pulls out the first element of the vector value of the second element of the tuple.)\nThere are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.\nint a = 5;\nmatrix[10, 100] b = ...;\ntuple(int, matrix[10, 100]) ab = (a, b);  // COPIES b\nb[1,1] = 10.3;  // does NOT change ab\n\n\n\nTuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, array[N] tuple(int, real, vector[5]) is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as tuple(array[N] int, array[N] real, array[N] vector[5]).\nThe second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix \\(A = Q \\cdot R\\), where \\(Q\\) is orthonormal and \\(R\\) is upper triangular. In the past, this required two function calls.\nmatrix[M, N] A = ...;\nmatrix[M, M] Q = qr_Q(A);\nmatrix[M, N] R = qr_R(A);\nWith tuples, this can be simplified to the following,\ntuple(matrix[M, M], matrix[M, N]) QR = qr(A);\nwith QR.1 being Q and QR.2 giving R.\n\n\n\n\nStan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.\nWhen an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.\n\n\n\nThe underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.\nArrays, on the other hand, are implemented as instances of the C++\nstd::vector class (not to be confused with Eigen’s Eigen::Vector class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.\n\n\nIn Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.\nFirst, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.\nSecond, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).\nThird, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.\nmatrix[M, N] a;\n//...\nfor (n in 1:N) {\n  for (m in 1:M) {\n    // ... do something with a[m, n] ...\n  }\n}\nArrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  for (n in 1:N) {\n    // ... do something with a[m, n] ...\n  }\n}\nThe first use of a[m ,n] should bring a[m] into memory. Overall, traversing matrices is more efficient than traversing arrays.\nThis is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is\narray[I, J] matrix[M, N] b;\n// ...\nfor (i in 1:I) {\n  for (j in 1:J) {\n    for (n in 1:N) {\n      for (m in 1:M) {\n        // ... do something with b[i, j, m, n] ...\n      }\n    }\n  }\n}\nIf a is a matrix, the notation a[m] picks out row m of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this\narray[M] row_vector[N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nis much more efficient than the pure matrix version\nmatrix[M, N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nSimilarly, indexing an array of column vectors is more efficient than using the col function to pick out a column of a matrix.\nIn contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this\nmatrix[N, k] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\ny_hat = x * beta;\nthan it is to do this\narray[N] row_vector[K] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\nfor (n in 1:N) {\n  y_hat[n] = x[n] * beta;\n}\n\n\n\nFor use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The Eigen::Vector template specialization and the std::vector template class are implemented similarly as containers of double values (the type real in Stan). Only arrays in Stan are allowed to store integer values.\n\n\n\n\nThe key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.\n\n\nCPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is much slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.\n\n\n\nMatrices are stored internally in column-major order. That is, an \\(M\n\\times N\\) matrix stores its elements in the order \\[\n(1,1), (2, 1), \\dotsc, (M, 1), (1, 2), \\dotsc, (M, 2), \\dotsc, (1, N),\n\\dotsc, (M, N).\n\\]\nThis means that it’s much more efficient to write loops over matrices column by column, as in the following example.\nmatrix[M, N] a;\n// ...\nfor (n in 1:N) {\n  for (m in 1:M) {\n     // ... do something with a[m, n] ...\n  }\n}\nIt also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.\narray[M] row_vector[N] a;\n// ...\nfor (m in 1:M) {\n  // ... do something with row vector a[m] ...\n}\nEven if what is done involves a function call, the row vector a[m] will not have to be copied.\n\n\n\nArrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  // ... do something with a[m] ...\n}\nA difference with matrices is that the entries a[m] in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”\n\n\n\n\nThere is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the multi-indexing chapter for some reshaping operations involving multiple indexing and range indexing.\n\n\n\nStan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to u and the compound slicing assignment to x, where u and x start with the same values.\ntransformed data {\n  vector[4] x = [ 1, 2, 3, 4 ]';\n  vector[4] u = [ 1, 2, 3, 4 ]';\n\n  for (t in 2:4) {\n    u[t] = u[t - 1] * 3;\n  }\n\n  x[2:4] = x[1:3] * 3;\n\n  print(\"u = \", u);\n  print(\"x = \", x);\n}\nThe output it produces is,\nu = [1, 3, 9, 27]\nx = [1, 3, 6, 9]\nIn the loop version assigning to u, the values are updated before being used to define subsequent values; in the sliced expression assigning to x, the entire right-hand side is evaluated before assigning to the left-hand side.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Efficiency Tuning"
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#exploiting-conjugacy",
-    "href": "stan-users-guide/efficiency-tuning.html#exploiting-conjugacy",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/matrices-arrays.html#basic-motivation",
+    "href": "stan-users-guide/matrices-arrays.html#basic-motivation",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "Continuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.\ntheta ~ beta(alpha + sum(y), beta + N - sum(y));\nTo make the model even more efficient, a transformed data variable defined to be sum(y) could be used in the place of sum(y).",
+    "text": "Stan provides three basic scalar types, int, real, and complex, as well as three basic linear algebra types, vector, row_vector, and matrix. Stan allows arrays of any dimensionality, containing any type of element (though that type must be declared and must be the same for all elements).\nThis leaves us in the awkward situation of having three one-dimensional containers, as exemplified by the following declarations.\narray[N] real a;\nvector[N] a;\nrow_vector[N] a;\nThese distinctions matter. Matrix types, like vector and row vector, are required for linear algebra operations. There is no automatic promotion of arrays to vectors because the target, row vector or column vector, is ambiguous. Similarly, row vectors are separated from column vectors because multiplying a row vector by a column vector produces a scalar, whereas multiplying in the opposite order produces a matrix.\nThe following code fragment shows all four ways to declare a two-dimensional container of size \\(M \\times N\\).\narray[M, N] real b;          // b[m] : array[] real     (efficient)\narray[M] vector[N] b;        // b[m] : vector     (efficient)\narray[M] row_vector[N] b;    // b[m] : row_vector (efficient)\nmatrix[M, N] b;              // b[m] : row_vector (inefficient)\nThe main differences among these choices involve efficiency for various purposes and the type of b[m], which is shown in comments to the right of the declarations. Thus the only way to efficiently iterate over row vectors is to use the third declaration, but if you need linear algebra on matrices, but the only way to use matrix operations is to use the fourth declaration.\nThe inefficiencies due to any manual reshaping of containers is usually slight compared to what else is going on in a Stan program (typically a lot of gradient calculations).",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Efficiency Tuning"
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#standardizing-predictors",
-    "href": "stan-users-guide/efficiency-tuning.html#standardizing-predictors",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/matrices-arrays.html#tuple-types",
+    "href": "stan-users-guide/matrices-arrays.html#tuple-types",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "Standardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:\n\nIt helps in faster convergence of MCMC chains.\nIt makes the model less sensitive to the specifics of the parameterization.\nIt aids in the interpretation and comparison of the importance of coefficients across different predictors.\n\nWhen there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.\nSuppose that \\(y = (y_1,\\dotsc,y_N)\\) is a vector of \\(N\\) outcomes and \\(x = (x_1,\\dotsc,x_N)\\) the corresponding vector of \\(N\\) predictors. A simple linear regression involving an intercept coefficient \\(\\alpha\\) and slope coefficient \\(\\beta\\) can be expressed as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n,\n\\] where \\[\n\\epsilon_n \\sim \\textsf{normal}(0,\\sigma).\n\\]\nIf \\(x\\) has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values \\(x_n\\). First the elements of \\(x\\) are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.\nThe mean of \\(x\\) is given by:\n\\[\nmean_x = \\frac{1}{N} \\sum_{n=1}^{N} x_n\n\\]\nThe standard deviation of \\(x\\) is calculated as: \\[\nsd_x = {\\left({\\frac{1}{N} \\sum_{n=1}^{N} (x_n - mean_x)^2}\\right)}^{1/2}\n\\]\nWith these, we compute the \\(z\\), the standardized predictors\n\\[\nz_n = \\frac{x_n - mean_x}{sd_x}\n\\]\nwhere \\(z_n\\) is the standardized value corresponding to \\(x_n\\).\nThe inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.\n\\[\nx_n = z_n sd_x + mean_x\n\\]\nStandardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.\nConsider the following initial model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  // priors\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ normal(0, 5);\n  // likelihood\n  y ~ normal(x * beta + alpha, sigma);\n}\nThe data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\ntransformed data {\n  real mean_x = mean(x);\n  real sd_x = sd(x);\n  vector[N] x_std = (x - mean_x) / sd_x;\n}\nparameters {\n  real alpha_std;\n  real beta_std;\n  real&lt;lower=0&gt; sigma_std;\n}\nmodel {\n  alpha_std ~ normal(0, 10);\n  beta_std ~ normal(0, 10);\n  sigma_std ~ normal(0, 5);\n  y ~ normal(x_std * beta_std + alpha_std, sigma_std);\n}\nThe parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data x_std is defined in terms of variables mean_x and sd_x; by declaring these variables in the transformed data block, they will be available in all following blocks, and therefore can be used in the generated quantities block to record the “natural” parameters alpha and beta.\nThe fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.\nThe original regression \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\] has been transformed to a regression on the standardized data variable \\(z\\),\n\\[\ny_n = \\alpha' + \\beta' z_n + \\epsilon_n.\n\\]\nThe likelihood is specified in terms of the standardized parameters. The original slope \\(\\beta\\) is the standardized slope \\(\\beta'\\) scaled by the inverse of the standard deviation of \\(x\\). The original intercept \\(\\alpha\\) is the intercept from the standardized model \\(\\alpha'\\), corrected for the effect of scaling and centering \\(x\\). Thus, the formulas to retrieve \\(\\alpha\\) and \\(\\beta\\) from \\(\\alpha'\\) and \\(\\beta'\\) are:\n\\[\\begin{align*}\n\\beta = \\frac{\\beta'}{\\sigma_x} \\\\\n\\alpha = \\alpha' - \\beta' \\frac{\\mu_x}{\\sigma_x}\n\\end{align*}\\]\nThese recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,\ngenerated quantities {\n  real beta = beta_std / sd_x;\n  real alpha = alpha_std - beta_std * mean_x / sd_x;\n\n}\nWhen there are multiple real-valued predictors, i.e., when K is the number of predictors, x is an \\(N \\times K\\) matrix, and beta ia \\(K\\)-vector of coefficients, then x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. When \\(K \\ll N\\) the QR reparameterization is recommended for linear and generalized linear models unless there is an informative prior on the location of \\(\\beta\\).\n\n\nFor many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use\ny ~ std_normal();\nthan to use\ny ~ normal(0, 1);\nbecause the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.",
+    "text": "Arrays may contain entries of any type, but the types must be the same for all entries. Matrices and vectors contain either real numbers or complex numbers, but all the contained types are the same (e.g., if a vector has a single complex typed entry, all the entries are complex).\nWith arrays or vectors, we can represent pairs of real numbers or pairs of complex numbers. For example, a complex_vector[3] holds exactly three complex numbers. With arrays and vectors, there is no way to represent a pair consisting of an integer and a real number.\nTuples provide a way to represent a sequence of values of heterogeneous types. For example, tuple(int, real) is the type of a pair consisting of an integer and a real number and tuple(array[5] int, vector[6]) is the type of pairs where the first element is a five-element array of integers, and the second is a six-element vector.\n\n\nTuples are declared using the keyword tuple followed by a sequence of type declarations in parentheses. Tuples are constructed using only parentheses. The following example illustrations both declaration and construction.\ntuple(int, vector[3]) ny = (5, [3, 2.9, 1.8]');\nThe elements of a tuple are accessed by position, starting from 1. For example, we can extract the elements of the tuple above using\nint n = ny.1;\nvector[3] y = ny.2;\nWe can also assign into the elements of a tuple.\ntuple(int, vector[3], complex) abc;\nabc.1 = 5;\nabc.2[1] = 3;\nabc.2[2] = 2.9;\nabc.2[3] = 1.4798;\nabc.3 = 2 + 1.9j;\nAs the cascaded indexing example shows, the result of abc.1 is an lvalue (i.e., something to which values may be assigned), and we can further index into it to create new lvalues (e.g., abc.2[1] pulls out the first element of the vector value of the second element of the tuple.)\nThere are two efficiency considerations for tuples. First, like the other container types, tuples are passed to functions by constant reference, which means only a pointer gets passed rather than copying the data. Second, like the array types, creating a tuple requires copying the data for all of its elements. For example, in the following code, the matrix is copied, entailing 1000 copies of scalar values.\nint a = 5;\nmatrix[10, 100] b = ...;\ntuple(int, matrix[10, 100]) ab = (a, b);  // COPIES b\nb[1,1] = 10.3;  // does NOT change ab\n\n\n\nTuples are primarily useful for two things. First, they provide a way to encapsulate a group of heterogeneous items so that they may be passed as a group. This lets us define arrays of structures as well as structures of arrays. For example, array[N] tuple(int, real, vector[5]) is an array of tuples, each of which has an integer, real, and vector component. Alternatively, we can represent the same information using a tuple of parallel arrays as tuple(array[N] int, array[N] real, array[N] vector[5]).\nThe second use is for function return values. Here, if a function computes two different things with different types, and the computation shares work, it’s best to write one function that returns both things. For example, an eigendecomposition returns a pair consisting of a vector of eigenvalues and a matrix of eigenvectors, whereas a singular value decomposition returns three matrices of different shapes. Before introducing tuples in version 2.33, the QR decomposition of matrix \\(A = Q \\cdot R\\), where \\(Q\\) is orthonormal and \\(R\\) is upper triangular. In the past, this required two function calls.\nmatrix[M, N] A = ...;\nmatrix[M, M] Q = qr_Q(A);\nmatrix[M, N] R = qr_R(A);\nWith tuples, this can be simplified to the following,\ntuple(matrix[M, M], matrix[M, N]) QR = qr(A);\nwith QR.1 being Q and QR.2 giving R.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Efficiency Tuning"
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#using-map-reduce",
-    "href": "stan-users-guide/efficiency-tuning.html#using-map-reduce",
-    "title": "Efficiency Tuning",
+    "objectID": "stan-users-guide/matrices-arrays.html#fixed-sizes-and-indexing-out-of-bounds",
+    "href": "stan-users-guide/matrices-arrays.html#fixed-sizes-and-indexing-out-of-bounds",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "The map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the map-reduce chapter for more information on implementing map-reduce operations.\nMap-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.",
+    "text": "Stan’s matrices, vectors, and array variables are sized when they are declared and may not be dynamically resized. Function arguments do not have sizes, but these sizes are fixed when the function is called and the container is instantiated. Also, declarations may be inside loops and thus may change over the course of running a program, but each time a declaration is visited, it declares a fixed size object.\nWhen an index is provided that is out of bounds, Stan throws a rejection error and computation on the current log density and gradient evaluation is halted and the algorithm is left to clean up the error. All of Stan’s containers check the sizes of all indexes.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Efficiency Tuning"
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/efficiency-tuning.html#footnotes",
-    "href": "stan-users-guide/efficiency-tuning.html#footnotes",
-    "title": "Efficiency Tuning",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThis parameterization came to be known on our mailing lists as the “Matt trick” after Matt Hoffman, who independently came up with it while fitting hierarchical models in Stan.↩︎\nStan uses its own arena-based allocation, so allocation and deallocation are faster than with a raw call to new.↩︎\nFuture versions of Stan may remove this inefficiency by more fully exploiting expression templates inside the Eigen C++ matrix library. This will require enhancing Eigen to deal with mixed-type arguments, such as the type double used for constants and the algorithmic differentiation type stan::math::var used for variables.↩︎",
+    "objectID": "stan-users-guide/matrices-arrays.html#indexing-efficiency.section",
+    "href": "stan-users-guide/matrices-arrays.html#indexing-efficiency.section",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
+    "section": "",
+    "text": "The underlying matrix and linear algebra operations are implemented in terms of data types from the Eigen C++ library. By having vectors and matrices as basic types, no conversion is necessary when invoking matrix operations or calling linear algebra functions.\nArrays, on the other hand, are implemented as instances of the C++\nstd::vector class (not to be confused with Eigen’s Eigen::Vector class or Stan vectors). By implementing arrays this way, indexing is efficient because values can be returned by reference rather than copied by value.\n\n\nIn Stan models, there are a few minor efficiency considerations in deciding between a two-dimensional array and a matrix, which may seem interchangeable at first glance.\nFirst, matrices use a bit less memory than two-dimensional arrays. This is because they don’t store a sequence of arrays, but just the data and the two dimensions.\nSecond, matrices store their data in column-major order. Furthermore, all of the data in a matrix is guaranteed to be contiguous in memory. This is an important consideration for optimized code because bringing in data from memory to cache is much more expensive than performing arithmetic operations with contemporary CPUs. Arrays, on the other hand, only guarantee that the values of primitive types are contiguous in memory; otherwise, they hold copies of their values (which are returned by reference wherever possible).\nThird, both data structures are best traversed in the order in which they are stored. This also helps with memory locality. This is column-major for matrices, so the following order is appropriate.\nmatrix[M, N] a;\n//...\nfor (n in 1:N) {\n  for (m in 1:M) {\n    // ... do something with a[m, n] ...\n  }\n}\nArrays, on the other hand, should be traversed in row-major order (i.e., last index fastest), as in the following example.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  for (n in 1:N) {\n    // ... do something with a[m, n] ...\n  }\n}\nThe first use of a[m ,n] should bring a[m] into memory. Overall, traversing matrices is more efficient than traversing arrays.\nThis is true even for arrays of matrices. For example, the ideal order in which to traverse a two-dimensional array of matrices is\narray[I, J] matrix[M, N] b;\n// ...\nfor (i in 1:I) {\n  for (j in 1:J) {\n    for (n in 1:N) {\n      for (m in 1:M) {\n        // ... do something with b[i, j, m, n] ...\n      }\n    }\n  }\n}\nIf a is a matrix, the notation a[m] picks out row m of that matrix. This is a rather inefficient operation for matrices. If indexing of vectors is needed, it is much better to declare an array of vectors. That is, this\narray[M] row_vector[N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nis much more efficient than the pure matrix version\nmatrix[M, N] b;\n// ...\nfor (m in 1:M) {\n   // ... do something with row vector b[m] ...\n}\nSimilarly, indexing an array of column vectors is more efficient than using the col function to pick out a column of a matrix.\nIn contrast, whatever can be done as pure matrix algebra will be the fastest. So if I want to create a row of predictor-coefficient dot-products, it’s more efficient to do this\nmatrix[N, k] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\ny_hat = x * beta;\nthan it is to do this\narray[N] row_vector[K] x;    // predictors (aka covariates)\n// ...\nvector[K] beta;   // coeffs\n// ...\nvector[N] y_hat;  // linear prediction\n// ...\nfor (n in 1:N) {\n  y_hat[n] = x[n] * beta;\n}\n\n\n\nFor use purely as a container, there is really nothing to decide among vectors, row vectors and one-dimensional arrays. The Eigen::Vector template specialization and the std::vector template class are implemented similarly as containers of double values (the type real in Stan). Only arrays in Stan are allowed to store integer values.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Efficiency Tuning"
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html",
-    "href": "stan-users-guide/dae.html",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/matrices-arrays.html#memory-locality",
+    "href": "stan-users-guide/matrices-arrays.html#memory-locality",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "Stan support solving systems of differential-algebraic equations (DAEs) of index 1 (Serban et al. 2021). The solver adaptively refines the solutions in order to satisfy given tolerances.\nOne can think a differential-algebraic system of equations as ODEs with additional algebraic constraints applied to some of the variables. In such a system, the variable derivatives may not be expressed explicitly with a right-hand-side as in ODEs, but implicitly constrained.\nSimilar to ODE solvers, the DAE solvers must not only provide the solution to the DAE itself, but also the gradient of the DAE solution with respect to parameters (the sensitivities). Stan’s DAE solver uses the forward sensitivity calculation to expand the base DAE system with additional DAE equations for the gradients of the solution. For each parameter, an additional full set of \\(N\\) sensitivity states are added meaning that the full DAE solved has \\(N \\, + N \\cdot M\\) states.\nTwo interfaces are provided for the forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. The same principles in the control parameters section apply here.\nInternally Stan’s DAE solver uses a variable-step, variable-order, backward-differentiation formula implementation (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005).\n\n\nA DAE is defined by a set of expressions for the residuals of differential equations and algebraic equations \\(r(y', y, t, \\theta)\\), and consistent initial conditions \\(y(t_0, \\theta) = y_0, y'(t_0, \\theta)=y'_0\\). The DAE is define by residual function as \\(r(y', y, t, \\theta)=0\\). The \\(\\theta\\) dependence is included in the notation to highlight that the solution \\(y(t)\\) is a function of any parameters used in the computation.\n\n\n\nAs an example of a system of DAEs, consider following chemical kinetics problem(Robertson 1966). The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants \\(y_1\\) and \\(y_2\\) to the product \\(y_3\\), and the algebraic equation describes the mass conservation. (Serban and Hindmarsh 2021).\n\\[\\begin{align*}\n\\frac{dy_1}{dt} + \\alpha y_1 - \\beta y_2 y_3 = 0 \\\\\n\\frac{dy_2}{dt} - \\alpha y_1 + \\beta y_2 y_3 + \\gamma y_2^2 = 0 \\\\\ny_1 + y_2 + y_3 - 1.0 = 0\n\\end{align*}\\]\nThe state equations implicitly defines the state \\((y_1(t), y_2(t), y_3(t))\\) at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients \\((\\alpha, \\beta, \\gamma)\\).\nUnlike solving ODEs, solving DAEs requires a consistent initial condition. That is, one must specify both \\(y(t_0)\\) and \\(y'(t_0)\\) so that residual function becomes zero at initial time \\(t_0\\) \\[\\begin{equation*}\nr(y'(t_0), y(t_0), t_0) = 0\n\\end{equation*}\\]\n\n\n\nThe index along a DAE solution \\(y(t)\\) is the minimum number of differentiations of some of the components of the system required to solve for \\(y'\\) uniquely in terms of \\(y\\) and \\(t\\), so that the DAE is converted into an ODE for \\(y\\). Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for \\(y_3\\).\nMost DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically (Ascher and Petzold 1998).\n\n\n\nThe first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the residual function is time, passed as a real; the second argument to the residual function is the system state \\(y\\), passed as a vector, the third argument to the residual function is the state derivative \\(y'\\), also passed as a vector. The residual function’s return value is a vector of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient \\(\\alpha\\), \\(\\beta\\), and \\(\\gamma\\)), data, or any quantities that are needed to define the DAE.\nThe above reaction be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector chem(real t, vector yy, vector yp,\n            real alpha, real beta, real gamma) {\n  vector[3] res;\n  res[1] = yp[1] + alpha * yy[1] - beta * yy[2] * yy[3];\n  res[2] = yp[2] - alpha * yy[1] + beta * yy[2] * yy[3] + gamma * yy[2] * yy[2];\n  res[3] = yy[1] + yy[2] + yy[3] - 1.0;\n  return res;\n}\nThe function takes in a time t (a real), the system state yy (a vector), state derivative yp (a vector), as well as parameter alpha (a real), beta (a real), and gamma (a real). The function returns a vector of the residuals at time t. The DAE coded here does not explicitly depend on t, however one still needs to specify t as an argument.\n\n\nThe types in the DAE residual function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, the third argument is the state derivative passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible DAE signatures:\nvector my_dae1(real t, vector y, vector yp, real a0);\nvector my_dae2(real t, vector y, vector yp, array[] int a0, vector a1);\nvector my_dae3(real t, vector y, vector yp, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector my_dae1(real t, array[] real y, vector yp);\n// Second argument is not a vector\narray[] real my_dae2(real t, vector y, vector yp);\n// Return type is not a vector\nvector my_dae3(real t, vector y);\n// First argument is not a real and missing the third argument\n\n\n\n\nStan provides a dae function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.\ndata {\n  int N;\n  vector[3] yy0;\n  vector[3] yp0;\n  real t0;\n  real alpha;\n  real beta;\n  array[N] real ts;\n  array[N] vector[3] y;\n}\nparameters {\n  real gamma;\n}\ntransformed parameters {\n  vector[3] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);\n}\nSince gamma is a parameter, the DAE solver is called in the transformed parameters block.\n\n\n\nUsing dae_tol one can specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters in order to control the DAE solution.\nvector[3] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,\n                             relative_tolerance,\n                             absolute_tolerance,\n                             max_num_steps,\n                             alpha, beta, gamma);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nThe default value of relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
+    "text": "The key to understanding efficiency of matrix and vector representations is memory locality and reference passing versus copying.\n\n\nCPUs on computers bring in memory in blocks through layers of caches. Fetching from memory is much slower than performing arithmetic operations. The only way to make container operations fast is to respect memory locality and access elements that are close together in memory sequentially in the program.\n\n\n\nMatrices are stored internally in column-major order. That is, an \\(M\n\\times N\\) matrix stores its elements in the order \\[\n(1,1), (2, 1), \\dotsc, (M, 1), (1, 2), \\dotsc, (M, 2), \\dotsc, (1, N),\n\\dotsc, (M, N).\n\\]\nThis means that it’s much more efficient to write loops over matrices column by column, as in the following example.\nmatrix[M, N] a;\n// ...\nfor (n in 1:N) {\n  for (m in 1:M) {\n     // ... do something with a[m, n] ...\n  }\n}\nIt also follows that pulling a row out of a matrix is not memory local, as it has to stride over the whole sequence of values. It also requires a copy operation into a new data structure as it is not stored internally as a unit in a matrix. For sequential access to row vectors in a matrix, it is much better to use an array of row vectors, as in the following example.\narray[M] row_vector[N] a;\n// ...\nfor (m in 1:M) {\n  // ... do something with row vector a[m] ...\n}\nEven if what is done involves a function call, the row vector a[m] will not have to be copied.\n\n\n\nArrays are stored internally following their data structure. That means a two dimensional array is stored in row-major order. Thus it is efficient to pull out a “row” of a two-dimensional array.\narray[M, N] real a;\n// ...\nfor (m in 1:M) {\n  // ... do something with a[m] ...\n}\nA difference with matrices is that the entries a[m] in the two dimensional array are not necessarily adjacent in memory, so there are no guarantees on iterating over all the elements in a two-dimensional array will provide memory locality across the “rows.”",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Differential-Algebraic Equations"
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html#notation",
-    "href": "stan-users-guide/dae.html#notation",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/matrices-arrays.html#converting-among-matrix-vector-and-array-types",
+    "href": "stan-users-guide/matrices-arrays.html#converting-among-matrix-vector-and-array-types",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "A DAE is defined by a set of expressions for the residuals of differential equations and algebraic equations \\(r(y', y, t, \\theta)\\), and consistent initial conditions \\(y(t_0, \\theta) = y_0, y'(t_0, \\theta)=y'_0\\). The DAE is define by residual function as \\(r(y', y, t, \\theta)=0\\). The \\(\\theta\\) dependence is included in the notation to highlight that the solution \\(y(t)\\) is a function of any parameters used in the computation.",
+    "text": "There is no automatic conversion among matrices, vectors, and arrays in Stan. But there are a wide range of conversion functions to convert a matrix into a vector, or a multi-dimensional array into a one-dimensional array, or convert a vector to an array. See the section on mixed matrix and array operations in the functions reference manual for a complete list of conversion operators and the multi-indexing chapter for some reshaping operations involving multiple indexing and range indexing.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Differential-Algebraic Equations"
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html#example-chemical-kinetics",
-    "href": "stan-users-guide/dae.html#example-chemical-kinetics",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/matrices-arrays.html#aliasing-in-stan-containers",
+    "href": "stan-users-guide/matrices-arrays.html#aliasing-in-stan-containers",
+    "title": "Matrices, Vectors, Arrays, and Tuples",
     "section": "",
-    "text": "As an example of a system of DAEs, consider following chemical kinetics problem(Robertson 1966). The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants \\(y_1\\) and \\(y_2\\) to the product \\(y_3\\), and the algebraic equation describes the mass conservation. (Serban and Hindmarsh 2021).\n\\[\\begin{align*}\n\\frac{dy_1}{dt} + \\alpha y_1 - \\beta y_2 y_3 = 0 \\\\\n\\frac{dy_2}{dt} - \\alpha y_1 + \\beta y_2 y_3 + \\gamma y_2^2 = 0 \\\\\ny_1 + y_2 + y_3 - 1.0 = 0\n\\end{align*}\\]\nThe state equations implicitly defines the state \\((y_1(t), y_2(t), y_3(t))\\) at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients \\((\\alpha, \\beta, \\gamma)\\).\nUnlike solving ODEs, solving DAEs requires a consistent initial condition. That is, one must specify both \\(y(t_0)\\) and \\(y'(t_0)\\) so that residual function becomes zero at initial time \\(t_0\\) \\[\\begin{equation*}\nr(y'(t_0), y(t_0), t_0) = 0\n\\end{equation*}\\]",
+    "text": "Stan expressions are all evaluated before assignment happens, so there is no danger of so-called aliasing in array, vector, or matrix operations. In the following Stan program, contrast the behavior of the loop assignment to u and the compound slicing assignment to x, where u and x start with the same values.\ntransformed data {\n  vector[4] x = [ 1, 2, 3, 4 ]';\n  vector[4] u = [ 1, 2, 3, 4 ]';\n\n  for (t in 2:4) {\n    u[t] = u[t - 1] * 3;\n  }\n\n  x[2:4] = x[1:3] * 3;\n\n  print(\"u = \", u);\n  print(\"x = \", x);\n}\nThe output it produces is,\nu = [1, 3, 9, 27]\nx = [1, 3, 6, 9]\nIn the loop version assigning to u, the values are updated before being used to define subsequent values; in the sliced expression assigning to x, the entire right-hand side is evaluated before assigning to the left-hand side.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Differential-Algebraic Equations"
+      "Programming Techniques",
+      "Matrices, Vectors, Arrays, and Tuples"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html#index-of-daes",
-    "href": "stan-users-guide/dae.html#index-of-daes",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/index.html",
+    "href": "stan-users-guide/index.html",
+    "title": "Stan User’s Guide",
     "section": "",
-    "text": "The index along a DAE solution \\(y(t)\\) is the minimum number of differentiations of some of the components of the system required to solve for \\(y'\\) uniquely in terms of \\(y\\) and \\(t\\), so that the DAE is converted into an ODE for \\(y\\). Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for \\(y_3\\).\nMost DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically (Ascher and Petzold 1998).",
+    "text": "This is the official user’s guide for Stan. It provides example models and programming techniques for coding statistical models in Stan.\n\nPart 1 gives Stan code and discussions for several important classes of models.\nPart 2 discusses various general Stan programming techniques that are not tied to any particular model.\nPart 3 introduces algorithms for calibration and model checking that require multiple runs of Stan.\nThe appendices provide an introduction to the stanc3 compiler used in the various interfaces to Stan, a style guide, and advice for users of BUGS and JAGS.\n\nWe recommend working through this guide using the textbooks Bayesian Data Analysis and Statistical Rethinking: A Bayesian Course with Examples in R and Stan as references on the concepts, and using the Stan Reference Manual when necessary to clarify programming issues.\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Differential-Algebraic Equations"
+      "Stan User's Guide"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html#coding-the-dae-system-function",
-    "href": "stan-users-guide/dae.html#coding-the-dae-system-function",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/gaussian-processes.html",
+    "href": "stan-users-guide/gaussian-processes.html",
+    "title": "Gaussian Processes",
     "section": "",
-    "text": "The first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the residual function is time, passed as a real; the second argument to the residual function is the system state \\(y\\), passed as a vector, the third argument to the residual function is the state derivative \\(y'\\), also passed as a vector. The residual function’s return value is a vector of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient \\(\\alpha\\), \\(\\beta\\), and \\(\\gamma\\)), data, or any quantities that are needed to define the DAE.\nThe above reaction be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector chem(real t, vector yy, vector yp,\n            real alpha, real beta, real gamma) {\n  vector[3] res;\n  res[1] = yp[1] + alpha * yy[1] - beta * yy[2] * yy[3];\n  res[2] = yp[2] - alpha * yy[1] + beta * yy[2] * yy[3] + gamma * yy[2] * yy[2];\n  res[3] = yy[1] + yy[2] + yy[3] - 1.0;\n  return res;\n}\nThe function takes in a time t (a real), the system state yy (a vector), state derivative yp (a vector), as well as parameter alpha (a real), beta (a real), and gamma (a real). The function returns a vector of the residuals at time t. The DAE coded here does not explicitly depend on t, however one still needs to specify t as an argument.\n\n\nThe types in the DAE residual function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, the third argument is the state derivative passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible DAE signatures:\nvector my_dae1(real t, vector y, vector yp, real a0);\nvector my_dae2(real t, vector y, vector yp, array[] int a0, vector a1);\nvector my_dae3(real t, vector y, vector yp, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector my_dae1(real t, array[] real y, vector yp);\n// Second argument is not a vector\narray[] real my_dae2(real t, vector y, vector yp);\n// Return type is not a vector\nvector my_dae3(real t, vector y);\n// First argument is not a real and missing the third argument",
+    "text": "Gaussian processes are continuous stochastic processes and thus may be interpreted as providing a probability distribution over functions. A probability distribution over continuous functions may be viewed, roughly, as an uncountably infinite collection of random variables, one for each valid input. The generality of the supported functions makes Gaussian priors popular choices for priors in general multivariate (non-linear) regression problems.\nThe defining feature of a Gaussian process is that the joint distribution of the function’s value at a finite number of input points is a multivariate normal distribution. This makes it tractable to both fit models from finite amounts of observed data and make predictions for finitely many new data points.\nUnlike a simple multivariate normal distribution, which is parameterized by a mean vector and covariance matrix, a Gaussian process is parameterized by a mean function and covariance function. The mean and covariance functions apply to vectors of inputs and return a mean vector and covariance matrix which provide the mean and covariance of the outputs corresponding to those input points in the functions drawn from the process.\nGaussian processes can be encoded in Stan by implementing their mean and covariance functions or by using the specialized covariance functions outlined below, and plugging the result into the Gaussian model. This form of model is straightforward and may be used for simulation, model fitting, or posterior predictive inference. A more efficient Stan implementation for the GP with a normally distributed outcome marginalizes over the latent Gaussian process, and applies a Cholesky-factor reparameterization of the Gaussian to compute the likelihood and the posterior predictive distribution analytically.\nAfter defining Gaussian processes, this chapter covers the basic implementations for simulation, hyperparameter estimation, and posterior predictive inference for univariate regressions, multivariate regressions, and multivariate logistic regressions. Gaussian processes are general, and by necessity this chapter only touches on some basic models. For more information, see Rasmussen and Williams (2006).\nNote that fitting Gaussian processes as described below using exact inference by computing Cholesky of the covariance matrix scales cubicly with the size of data. Due to how Stan autodiff is implemented, Stan is also slower than Gaussian process specialized software. It is likely that Gaussian processes using exact inference by computing Cholesky of the covariance matrix with \\(N&gt;1000\\) are too slow for practical purposes in Stan. There are many approximations to speed-up Gaussian process computation, from which the basis function approaches for 1-3 dimensional \\(x\\) are easiest to implement in Stan (see, e.g., Riutort-Mayol et al. (2023)).\n\n\nThe data for a multivariate Gaussian process regression consists of a series of \\(N\\) inputs \\(x_1,\\dotsc,x_N \\in \\mathbb{R}^D\\) paired with outputs \\(y_1,\\dotsc,y_N \\in \\mathbb{R}\\). The defining feature of Gaussian processes is that the probability of a finite number of outputs \\(y\\) conditioned on their inputs \\(x\\) is Gaussian: \\[\ny \\sim \\textsf{multivariate normal}(m(x), K(x \\mid \\theta)),\n\\] where \\(m(x)\\) is an \\(N\\)-vector and \\(K(x \\mid \\theta)\\) is an \\(N \\times N\\) covariance matrix. The mean function \\(m : \\mathbb{R}^{N \\times D}\n\\rightarrow \\mathbb{R}^{N}\\) can be anything, but the covariance function \\(K : \\mathbb{R}^{N \\times D} \\rightarrow \\mathbb{R}^{N \\times N}\\) must produce a positive-definite matrix for any input \\(x\\).1\nA popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, \\[\n  K(x \\mid \\alpha, \\rho, \\sigma)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j} \\sigma^2,\n\\] where \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\) are hyperparameters defining the covariance function and where \\(\\delta_{i, j}\\) is the Kronecker delta function with value 1 if \\(i = j\\) and value 0 otherwise; this test is between the indexes \\(i\\) and \\(j\\), not between values \\(x_i\\) and \\(x_j\\). This kernel is obtained through a convolution of two independent Gaussian processes, \\(f_1\\) and \\(f_2\\), with kernels \\[\n  K_1(x \\mid \\alpha, \\rho)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n\\] and \\[\n  K_2(x \\mid \\sigma)_{i, j}\n=\n\\delta_{i, j} \\sigma^2.\n\\]\nThe addition of \\(\\sigma^2\\) on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs \\(x_i = x_j\\). In statistical terms, \\(\\sigma\\) is the scale of the noise term in the regression.\nThe hyperparameter \\(\\rho\\) is the length-scale, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of \\(\\rho\\) closer to zero lead the GP to represent high-frequency functions, whereas larger values of \\(\\rho\\) lead to low-frequency functions. The hyperparameter \\(\\alpha\\) is the marginal standard deviation. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP \\(f_1\\) prior at a single input \\(x\\) conditional on one value of \\(\\alpha\\) one would recover \\(\\alpha\\).\nThe only term in the squared exponential covariance function involving the inputs \\(x_i\\) and \\(x_j\\) is their vector difference, \\(x_i - x_j\\). This produces a process with stationary covariance in the sense that if an input vector \\(x\\) is translated by a vector \\(\\epsilon\\) to \\(x +\n\\epsilon\\), the covariance at any pair of outputs is unchanged, because \\(K(x \\mid \\theta) = K(x + \\epsilon \\mid \\theta)\\).\nThe summation involved is just the squared Euclidean distance between \\(x_i\\) and \\(x_j\\) (i.e., the squared \\(L_2\\) norm of their difference, \\(x_i -\nx_j\\)). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\).\nChanging the notion of distance from Euclidean to taxicab distance (i.e., an \\(L_1\\) norm) changes the support to functions which are continuous but not smooth.\n\n\n\nIt is simplest to start with a Stan model that does nothing more than simulate draws of functions \\(f\\) from a Gaussian process. In practical terms, the model will draw values \\(y_n = f(x_n)\\) for finitely many input points \\(x_n\\).\nThe Stan model defines the mean and covariance functions in a transformed data block and then samples outputs \\(y\\) in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to \\(\\alpha^2 = 1\\), \\(\\rho^2 = 1\\), and \\(\\sigma^2 = 0.1\\), and the mean function \\(m\\) is defined to always return the zero vector, \\(m(x) = \\textbf{0}\\). Consider the following implementation of a Gaussian process simulator.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K;\n  vector[N] mu = rep_vector(0, N);\n  for (i in 1:(N - 1)) {\n    K[i, i] = 1 + 0.1;\n    for (j in (i + 1):N) {\n      K[i, j] = exp(-0.5 * square(x[i] - x[j]));\n      K[j, i] = K[i, j];\n    }\n  }\n  K[N, N] = 1 + 0.1;\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K = gp_exp_quad_cov(x, 1.0, 1.0);\n  vector[N] mu = rep_vector(0, N);\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + 0.1;\n  }\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe input data are just the vector of inputs x and its size N. Such a model can be used with values of x evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.\n\n\nOnly the input data needs to change in moving from a univariate model to a multivariate model.\nThe only lines that change from the univariate model above are as follows.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n}\ntransformed data {\n  // ...\n}\nThe data are now declared as an array of vectors instead of an array of scalars; the dimensionality D is also declared.\nIn the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.\n\n\n\nA more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose \\(\\eta\\) is an an isotropic standard normal variate \\[\n\\eta \\sim \\textsf{normal}(\\textbf{0}, \\textrm{I}),\n\\] where \\(\\textbf{0}\\) is an \\(N\\)-vector of 0 values and \\(\\textrm{I}\\) is the \\(N\n\\times N\\) identity matrix. Let \\(L\\) be the Cholesky decomposition of \\(K(x \\mid \\theta)\\), i.e., the lower-triangular matrix \\(L\\) such that \\(LL^{\\top} =\nK(x \\mid \\theta)\\). Then the transformed variable \\(\\mu + L\\eta\\) has the intended target distribution, \\[\n  \\mu + L\\eta \\sim \\textsf{multivariate normal}(\\mu(x), K(x \\mid \\theta)).\n\\]\nThis transform can be applied directly to Gaussian process simulation.\nThis model has the same data declarations for N and x, and the same transformed data definitions of mu and K as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.\n// ...\ntransformed data {\n  matrix[N, N] L;\n  // ...\n  L = cholesky_decompose(K);\n}\nparameters {\n  vector[N] eta;\n}\nmodel {\n  eta ~ std_normal();\n}\ngenerated quantities {\n  vector[N] y;\n  y = mu + L * eta;\n}\nThe Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix K computed. The isotropic normal distribution for eta is specified as a vectorized univariate distribution for efficiency; this specifies that each eta[n] has an independent standard normal distribution. The sampled vector y is then defined as a generated quantity using a direct encoding of the transform described above.\n\n\n\n\n\n\nThe full generative model for a GP with a normal outcome, \\(y \\in \\mathbb{R}^N\\), with inputs \\(x \\in \\mathbb{R}^N\\), for a finite \\(N\\): \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\nf      &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i    &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\}\n\\end{align*}\\] With a normal outcome, it is possible to integrate out the Gaussian process \\(f\\), yielding the more parsimonious model: \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\ny      &\\sim \\textsf{multivariate normal}\n         \\left(0, K(x \\mid \\alpha, \\rho) + \\textbf{I}_N \\sigma^2\\right) \\\\\n\\end{align*}\\]\nIt can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.\nThe hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (\\(\\alpha\\)), the scale of the output noise (\\(\\sigma\\)), and the scale at which distances are measured among inputs (\\(\\rho\\)). See the Gaussian process priors section for more information about how to specify appropriate priors for the hyperparameters.\nThe Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix K in the model block, rather than the transformed data block.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  vector[N] mu = rep_vector(0, N);\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N, N] L_K;\n  matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n  real sq_sigma = square(sigma);\n\n  // diagonal elements\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + sq_sigma;\n  }\n\n  L_K = cholesky_decompose(K);\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y ~ multi_normal_cholesky(mu, L_K);\n}\nThe data block declares a vector y of observed values y[n] for inputs x[n]. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix K is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value y is known and the covariance matrix K is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.\nWe have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the cholesky_decompose function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (\\(N \\gtrsim 100\\)) the Cholesky decomposition version will be faster.\nHamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model (Neal 1997). If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.\n\n\nWe can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, \\(\\delta\\) to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\nTwo differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length \\(N\\) called eta. This is used in the model block to generate a multivariate normal vector called \\(f\\), corresponding to the latent GP. We put a \\(\\textsf{normal}(0,1)\\) prior on eta like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for \\(y\\) with one \\(N\\)-dimensional multivariate normal with an identity covariance matrix multiplied by \\(\\sigma^2\\), we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.\n\n\n\n\nGaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.\n\n\nIf we want to model count data, we can remove the \\(\\sigma\\) parameter, and use poisson_log, which implements a Poisson distribution with log link function, rather than normal. We can also add an overall mean parameter, \\(a\\), which will account for the marginal expected value for \\(y\\). We do this because we cannot center count data like we would for normally distributed data.\ndata {\n  // ...\n  array[N] int&lt;lower=0&gt; y;\n  // ...\n}\n// ...\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ poisson_log(a + f);\n}\n\n\n\nFor computational reasons, we may want to integrate out the latent variable \\(f\\), as was done in the normal output model. Unfortunately, exact marginalization over \\(f\\) is not possible when the outcome model is not normal. Instead, we may perform approximate marginalization with an embedded Laplace approximation (Rasmussen and Williams 2006; Rue, Martino, and Chopin 2009; Margossian et al. 2020). To do so, we first use the function laplace_marginal to approximate the marginal likelihood \\(p(y \\mid \\rho, \\alpha, a)\\) and sample the hyperparameters with Hamiltonian Monte Carlo sampling. Then, we recover the integrated out \\(f\\) in the generated quantities block using laplace_latent_rng.\nThe embedded Laplace approximation computes a Gaussian approximation of the conditional posterior, \\[\n  \\hat p_\\mathcal{L}(f \\mid \\rho, \\alpha, a, y) \\approx p(f \\mid \\rho, \\alpha, a, y),\n\\] where \\(\\hat p_\\mathcal{L}\\) is a Gaussian that matches the mode and curvature of \\(p(f \\mid \\rho, \\alpha, a, y)\\). We then obtain an approximation of the marginal likelihood as follows: \\[\n  \\hat p_\\mathcal{L}(y \\mid \\rho, \\alpha, a)\n    = \\frac{p(f^* \\mid \\alpha, \\rho) p(y \\mid f^*, a)}{\n    \\hat p_\\mathcal{L}(f^* \\mid \\rho, \\alpha, a, y)},\n\\] where \\(f^*\\) is the mode of \\(p(f \\mid \\rho, \\alpha, a, y)\\), obtained via numerical optimization.\nTo use Stan’s embedded Laplace approximation, we must define the prior covariance function and the log likelihood function in the functions block.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int y) {\n      return poisson_log_lpmf(y | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n\n}\nThe embedded Laplace relies on calculations of the log likelihood’s Hessian, \\(\\partial^2 \\log p(y \\mid f, a, \\rho, \\alpha) / \\partial f^2\\), and these calculations can be much faster when the Hessian is sparse. In particular, it is expected that the Hessian is block diagonal. In the transformed data block we can specify the block size of the Hessian.\ntransformed data {\n  int hessian_block_size = 1;\n}\nFor example, if \\(y_i\\) depends only on \\(f_i\\), then the Hessian of the log likelihood is diagonal and the block size is 1. On the other hand, if the Hessian is not sparse, then we set the hessian block size to \\(N\\), where \\(N\\) is the dimension of \\(f\\). Currently, Stan does not check the block size of the Hessian and so the user is responsible for correctly specifying the block size.\nFinally, we increment target in the model block with the approximation to \\(\\log p(y \\mid \\rho, \\alpha, a)\\).\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  target += laplace_marginal(ll_function, (a, y), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\nNotice that we do not need to construct \\(f\\) explicitly, since it is marginalized out. Instead, we can recover the latent variables in generated quantities:\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nUsers can set the control parameters of the embedded Laplace approximation, via laplace_marginal_tol and laplace_latent_tol_rng. When using these functions, the user must set all the control options and store them in a tuple. These control parameters mostly concern the numerical optimizer used to find the mode \\(f^*\\) of \\(p(f \\mid \\rho, \\alpha, a)\\).\ntransformed data {\n  tuple(vector[N], real, int, int, int, int) laplace_ops;\n  laplace_ops.1 = rep_vector(0, N);  // starting point for Laplace optimizer\n  laplace_ops.2 = 1.49e-8;           // tolerance for optimizer\n  laplace_ops.3 = 500;               // maximum number of steps for optimizer.\n  laplace_ops.4 = 1;              // solver type being used.\n  laplace_ops.5 = 1000;           // max number of steps for linesearch.\n  laplace_ops.6 = 1;              // allow_fallback (1: TRUE, 0: FALSE)\nIf users want to depart from the defaults for only some of the control parameters, a tuple with the default values (as above) can be created with the helper callable generate_laplace_options(), and the specific control parameter can then be modified,\ntransformed data {\n  tuple(vector[N], real, int, int, int, int, int) laplace_ops =\n    generate_laplace_options(N);\n\n  laplace_ops.2 = 1e-6; // make tolerance of the optimizer less strict.\n}\nThe tuple laplace_ops is then passed to laplace_marginal_tol and laplace_rng_tol.\nmodel {\n// ...\n\n  target += laplace_marginal_tol(ll_function, (a, y), hessian_block_size,\n                                 cov_function, (rho, alpha, x, N, delta),\n                                 laplace_ops);\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta),\n                                   laplace_ops);\n}\nStan also provides support for a limited menu of built-in functions, including the Poisson distribution with a log link and and prior mean \\(m\\). When using such a built-in function, the user does not need to specify a likelihood in the functions block. However, the user must strictly follow the signature of the likelihood: in this case, \\(m\\) must be a vector of length \\(N\\) (to allow for different offsets for each observation \\(y_i\\)) and we must indicate which element of \\(f\\) each component of \\(y\\) matches using the variable \\(y_\\text{index}\\). In our example, there is a simple pairing \\((y_i, f_i)\\), however we could imagine a scenario where multiple observations \\((y_{j1}, y_{j2}, ...)\\) are observed for a single \\(f_j\\).\ntransformed data {\n  // ...\n  array[n_obs] int y_index;\n  for (i in 1:n_obs) y_index[i] = i - 1;\n}\n\n// ...\n\ntransformed parameter {\n  vector[N] m = rep_vector(a, N);\n}\n\nmodel {\n  // ...\n  target += laplace_marginal_poisson_log_lpmf(y | y_index, m,\n                                       cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_poisson_log_rng(y, y_index, m,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nAs before, we could specify the control parameters for the embedded Laplace approximation using laplace_marginal_tol_poisson_log_lpmf and laplace_latent_tol_poisson_log_nrg.\nMarginalization with a Laplace approximation can lead to faster inference, however it also introduces an approximation error. In practice, this error is negligible when using a Poisson likelihood and the approximation works well for log concave likelihoods (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). Still, users should exercise caution, especially when trying unconventional likelihoods.\n\n\n\nFor binary classification problems, the observed outputs \\(z_n \\in\n\\{ 0,1 \\}\\) are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs \\(y_n\\) through the logistic link, \\[\nz_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(y_n)),\n\\] or in other words, \\[\n\\Pr[z_n = 1] = \\operatorname{logit}^{-1}(y_n).\n\\]\nWe can extend our latent variable GP Stan program to deal with classification problems. Below a is the bias term, which can help account for imbalanced classes in the training data:\ndata {\n  // ...\n  array[N] int&lt;lower=0, upper=1&gt; z;\n  // ...\n}\n// ...\nmodel {\n  // ...\n  z ~ bernoulli_logit(a + f);\n}\n\n\n\nAs with the Poisson GP, we cannot marginalize the latent variables exactly, however we can resort to an embedded Laplace approximation.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int z) {\n      return bernoulli_logit_lpmf(z | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n}\n\n// ...\n\ntransformed data {\n  int hessian_block_size = 1;\n}\n\n// ...\n\nmodel {\n  target += laplace_marginal(ll_function, (a, z), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, z), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nWhile marginalization with a Laplace approximation can lead to faster inference, it also introduces an approximation error. In practice, this error may not be negligible with a Bernoulli likelihood; for more discussion see, e.g. (Vehtari et al. 2016; Margossian et al. 2020).\n\n\n\n\nIf we have multivariate inputs \\(x \\in \\mathbb{R}^D\\), the squared exponential covariance function can be further generalized by fitting a scale parameter \\(\\rho_d\\) for each dimension \\(d\\), \\[\n  k(x \\mid \\alpha, \\vec{\\rho}, \\sigma)_{i, j} = \\alpha^2 \\exp\n\\left(-\\dfrac{1}{2}\n\\sum_{d=1}^D \\dfrac{1}{\\rho_d^2} (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j}\\sigma^2.\n\\] The estimation of \\(\\rho\\) was termed “automatic relevance determination” by Neal (1996), but this is misleading, because the magnitude of the scale of the posterior for each \\(\\rho_d\\) is dependent on the scaling of the input data along dimension \\(d\\). Moreover, the scale of the parameters \\(\\rho_d\\) measures non-linearity along the \\(d\\)-th dimension, rather than “relevance” (Piironen and Vehtari 2016).\nA priori, the closer \\(\\rho_d\\) is to zero, the more nonlinear the conditional mean in dimension \\(d\\) is. A posteriori, the actual dependencies between \\(x\\) and \\(y\\) play a role. With one covariate \\(x_1\\) having a linear effect and another covariate \\(x_2\\) having a nonlinear effect, it is possible that \\(\\rho_1 &gt; \\rho_2\\) even if the predictive relevance of \\(x_1\\) is higher (Rasmussen and Williams 2006, 80). The collection of \\(\\rho_d\\) (or \\(1/\\rho_d\\)) parameters can also be modeled hierarchically.\nThe implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying rho to be an array.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  array[D] real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\n\n\n\nFormulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.\nPerhaps most importantly, the parameters \\(\\rho\\) and \\(\\alpha\\) are weakly identified (Zhang 2004). The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.\n\n\nGPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.\nWe may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.\nIf our model consists of only the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\} \\\\\n    & x \\in \\mathbb{R}^{N \\times D}, \\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, inv_gamma_lpdf in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.2.\nIf we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\big(0, K(x \\mid \\alpha, \\rho)\\big) \\\\\ny_i &\\sim \\textsf{normal}\\left(\\beta_0 + x_i \\beta_{[1:D]} + f_i, \\sigma\\right) \\, \\forall i\n  \\in \\{1, \\dots, N\\} \\\\\n    & x_i^T, \\beta_{[1:D]} \\in \\mathbb{R}^D,\\quad\n      x \\in \\mathbb{R}^{N \\times D},\\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: \\[\\begin{align*}\nf(x \\mid a, b, p) &= \\dfrac{\\left(a/b\\right)^{p/2}}{2K_p\\left(\\sqrt{ab}\\right)} x^{p - 1}\\exp\\big(-(ax + b\n  / x)/2\\big) \\\\\n  & x, a, b \\in \\mathbb{R}^{+},\\quad\n    p \\in \\mathbb{Z}\n\\end{align*}\\]\nwhich has an inverse gamma left tail if \\(p \\leq 0\\) and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:\nfunctions {\n  real generalized_inverse_gaussian_lpdf(real x, int p,\n                                        real a, real b) {\n    return p * 0.5 * log(a / b)\n      - log(2 * modified_bessel_second_kind(p, sqrt(a * b)))\n      + (p - 1) * log(x)\n      - (a * x + b / x) * 0.5;\n }\n}\ndata {\n  // ...\n}\nIf we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of \\(\\rho\\) in a fixed-domain \\([0,T]\\), we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is \\(T / \\pi\n\\rho\\). You can see that as \\(\\rho\\) decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.\n\n\n\nThe parameter \\(\\alpha\\) corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-\\(t\\) prior on \\(\\alpha\\).\nA half-\\(t\\) or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.\n\n\n\n\nSuppose for a given sequence of inputs \\(x\\) that the corresponding outputs \\(y\\) are observed. Given a new sequence of inputs \\(\\tilde{x}\\), the posterior predictive distribution of their labels is computed by sampling outputs \\(\\tilde{y}\\) according to \\[\np\\left(\\tilde{y} \\mid \\tilde{x},x,y\\right)\n\\ = \\\n\\frac{p\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right)}\n     {p(y \\mid x)}\n\\ \\propto \\\np\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right).\n\\]\nA direct implementation in Stan defines a model in terms of the joint distribution of the observed \\(y\\) and unobserved \\(\\tilde{y}\\).\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y1 ~ normal(f[1:N1], sigma);\n}\ngenerated quantities {\n  vector[N2] y2;\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f[N1 + n2], sigma);\n  }\n}\nThe input vectors x1 and x2 are declared as data, as is the observed output vector y1. The unknown output vector y2, which corresponds to input vector x2, is declared in the generated quantities block and will be sampled when the model is executed.\nA transformed data block is used to combine the input vectors x1 and x2 into a single vector x.\nThe model block declares and defines a local variable for the combined output vector f, which consists of the concatenation of the conditional mean for known outputs y1 and unknown outputs y2. Thus the combined output vector f is aligned with the combined input vector x. All that is left is to define the univariate normal distribution statement for y.\nThe generated quantities block defines the quantity y2. We generate y2 by randomly generating N2 values from univariate normals with each mean corresponding to the appropriate element in f.\n\n\nWe can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.\nConsider the following full model for prediction using logistic Gaussian process regression.\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  array[N1] int&lt;lower=0, upper=1&gt; z1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  z1 ~ bernoulli_logit(a + f[1:N1]);\n}\ngenerated quantities {\n  array[N2] int z2;\n  for (n2 in 1:N2) {\n    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);\n  }\n}\n\n\n\nBayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.\nJumping straight to the result, \\[\np\\left(\\tilde{y} \\mid \\tilde{x},y,x\\right)\n=\n\\textsf{normal}\\left(K^{\\top}\\Sigma^{-1}y,\\\n                \\Omega - K^{\\top}\\Sigma^{-1}K\\right),\n\\] where \\(\\Sigma = K(x \\mid \\alpha, \\rho, \\sigma)\\) is the result of applying the covariance function to the inputs \\(x\\) with observed outputs \\(y\\), \\(\\Omega =\nK(\\tilde{x} \\mid \\alpha, \\rho)\\) is the result of applying the covariance function to the inputs \\(\\tilde{x}\\) for which predictions are to be inferred, and \\(K\\) is the matrix of covariances between inputs \\(x\\) and \\(\\tilde{x}\\), which in the case of the exponentiated quadratic covariance function would be \\[\nK(x \\mid \\alpha, \\rho)_{i, j} = \\alpha^2 \\exp\\left(-\\dfrac{1}{2 \\rho^2}\n\\sum_{d=1}^D \\left(x_{i,d} - \\tilde{x}_{j,d}\\right)^2\\right).\n\\]\nThere is no noise term including \\(\\sigma^2\\) because the indexes of elements in \\(x\\) and \\(\\tilde{x}\\) are never the same.\nThis Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called gp_pred_rng which will generate a draw from the posterior predictive mean conditioned on observed data y1. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of \\(p(\\tilde{y})\\).\nfunctions {\n  vector gp_pred_rng(array[] real x2,\n                     vector y1,\n                     array[] real x1,\n                     real alpha,\n                     real rho,\n                     real sigma,\n                     real delta) {\n    int N1 = rows(y1);\n    int N2 = size(x2);\n    vector[N2] f2;\n    {\n      matrix[N1, N1] L_K;\n      vector[N1] K_div_y1;\n      matrix[N1, N2] k_x1_x2;\n      matrix[N1, N2] v_pred;\n      vector[N2] f2_mu;\n      matrix[N2, N2] cov_f2;\n      matrix[N2, N2] diag_delta;\n      matrix[N1, N1] K;\n      K = gp_exp_quad_cov(x1, alpha, rho);\n      for (n in 1:N1) {\n        K[n, n] = K[n, n] + square(sigma);\n      }\n      L_K = cholesky_decompose(K);\n      K_div_y1 = mdivide_left_tri_low(L_K, y1);\n      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';\n      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);\n      f2_mu = (k_x1_x2' * K_div_y1);\n      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);\n      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;\n      diag_delta = diag_matrix(rep_vector(delta, N2));\n\n      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);\n    }\n    return f2;\n  }\n}\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  vector[N1] mu = rep_vector(0, N1);\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N1, N1] L_K;\n  {\n    matrix[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);\n    real sq_sigma = square(sigma);\n\n    // diagonal elements\n    for (n1 in 1:N1) {\n      K[n1, n1] = K[n1, n1] + sq_sigma;\n    }\n\n    L_K = cholesky_decompose(K);\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y1 ~ multi_normal_cholesky(mu, L_K);\n}\ngenerated quantities {\n  vector[N2] f2;\n  vector[N2] y2;\n\n  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f2[n2], sigma);\n  }\n}\n\n\n\n\nSuppose we have observations \\(y_i \\in \\mathbb{R}^M\\) observed at \\(x_i \\in \\mathbb{R}^K\\). One can model the data like so: \\[\\begin{align*}\ny_i  &\\sim \\textsf{multivariate normal}\\left(f(x_i), \\textbf{I}_M \\sigma^2\\right) \\\\\nf(x) &\\sim \\textsf{GP}\\big(m(x), K(x \\mid \\theta, \\phi)\\big) \\\\\n     & K(x \\mid \\theta) \\in \\mathbb{R}^{M \\times M}, \\quad\n       f(x), m(x) \\in \\mathbb{R}^M\n\\end{align*}\\] where the \\(K(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]}\\) entry defines the covariance between \\(f_m(x)\\) and \\(f_{m^\\prime}(x^\\prime)(x)\\). This construction of Gaussian processes allows us to learn the covariance between the output dimensions of \\(f(x)\\). If we parameterize our kernel \\(K\\): \\[\nK(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]} = k\\left(x, x^\\prime \\mid\n\\theta\\right) k\\left(m, m^\\prime \\mid \\phi\\right)\n\\] then our finite dimensional generative model for the above is: \\[\\begin{align*}\nf        &\\sim \\textsf{matrixnormal}\\big(m(x), K(x \\mid \\alpha, \\rho), C(\\phi)\\big) \\\\\ny_{i, m} &\\sim \\textsf{normal}(f_{i,m}, \\sigma) \\\\\nf        &\\in  \\mathbb{R}^{N \\times M}\n\\end{align*}\\] where \\(K(x \\mid \\alpha, \\rho)\\) is the exponentiated quadratic kernel we’ve used throughout this chapter, and \\(C(\\phi)\\) is a positive-definite matrix, parameterized by some vector \\(\\phi\\).\nThe matrix normal distribution has two covariance matrices: \\(K(x \\mid\n\\alpha, \\rho)\\) to encode column covariance, and \\(C(\\phi)\\) to define row covariance. The salient features of the matrix normal are that the rows of the matrix \\(f\\) are distributed: \\[\nf_{[n,]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[n,]}, K(x \\mid \\alpha,\n\\rho)_{[n,n]} C(\\phi)\\big)\n\\] and that the columns of the matrix \\(f\\) are distributed: \\[\nf_{[,m]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[,m]}, K(x\n  \\mid \\alpha, \\rho) C(\\phi)_{[m,m]}\\big)\n\\] This also means means that \\(\\mathbb{E}\\left[f^T f\\right]\\) is equal to \\(\\operatorname{trace}\\!\\big(K(x \\mid \\alpha, \\rho)\\big) \\times C\\), whereas \\(\\mathbb{E}\\left[ff^T\\right]\\) is \\(\\operatorname{trace}(C) \\times K(x \\mid \\alpha, \\rho)\\). We can derive this using properties of expectation and the matrix normal density.\nWe should set \\(\\alpha\\) to \\(1.0\\) because the parameter is not identified unless we constrain \\(\\operatorname{trace}(C) = 1\\). Otherwise, we can multiply \\(\\alpha\\) by a scalar \\(d\\) and \\(C\\) by \\(1/d\\) and our likelihood will not change.\nWe can generate a random variable \\(f\\) from a matrix normal density in \\(\\mathbb{R}^{N \\times M}\\) using the following algorithm: \\[\\begin{align*}\n\\eta_{i,j} &\\sim \\textsf{normal}(0, 1) \\, \\forall i,j \\\\\nf          &= L_{K(x \\mid 1.0, \\rho)} \\, \\eta \\, L_C(\\phi)^T \\\\\nf          &\\sim \\textsf{matrixnormal}\\big(0, K(x \\mid 1.0, \\rho), C(\\phi)\\big) \\\\\n\\eta       &\\in \\mathbb{R}^{N \\times M} \\\\\nL_C(\\phi)  &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(C(\\phi)\\big) \\\\\nL_{K(x \\mid 1.0, \\rho)} &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(K(x \\mid 1.0, \\rho)\\big)\n\\end{align*}\\]\nThis can be implemented in Stan using a latent-variable GP formulation. We’ve used \\(\\textsf{LKJCorr}\\) for \\(C(\\phi)\\), but any positive-definite matrix will do.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] real x;\n  matrix[N, D] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  vector&lt;lower=0&gt;[D] alpha;\n  real&lt;lower=0&gt; sigma;\n  cholesky_factor_corr[D] L_Omega;\n  matrix[N, D] eta;\n}\nmodel {\n  matrix[N, D] f;\n  {\n    matrix[N, N] K = gp_exp_quad_cov(x, 1.0, rho);\n    matrix[N, N] L_K;\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta\n        * diag_pre_multiply(alpha, L_Omega)';\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(3);\n  to_vector(eta) ~ std_normal();\n\n  to_vector(y) ~ normal(to_vector(f), sigma);\n}\ngenerated quantities {\n  matrix[D, D] Omega;\n  Omega = L_Omega * L_Omega';\n}",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Differential-Algebraic Equations"
+      "Gaussian Processes"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html#solving-daes",
-    "href": "stan-users-guide/dae.html#solving-daes",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/gaussian-processes.html#gaussian-process-regression",
+    "href": "stan-users-guide/gaussian-processes.html#gaussian-process-regression",
+    "title": "Gaussian Processes",
     "section": "",
-    "text": "Stan provides a dae function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.\ndata {\n  int N;\n  vector[3] yy0;\n  vector[3] yp0;\n  real t0;\n  real alpha;\n  real beta;\n  array[N] real ts;\n  array[N] vector[3] y;\n}\nparameters {\n  real gamma;\n}\ntransformed parameters {\n  vector[3] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);\n}\nSince gamma is a parameter, the DAE solver is called in the transformed parameters block.",
+    "text": "The data for a multivariate Gaussian process regression consists of a series of \\(N\\) inputs \\(x_1,\\dotsc,x_N \\in \\mathbb{R}^D\\) paired with outputs \\(y_1,\\dotsc,y_N \\in \\mathbb{R}\\). The defining feature of Gaussian processes is that the probability of a finite number of outputs \\(y\\) conditioned on their inputs \\(x\\) is Gaussian: \\[\ny \\sim \\textsf{multivariate normal}(m(x), K(x \\mid \\theta)),\n\\] where \\(m(x)\\) is an \\(N\\)-vector and \\(K(x \\mid \\theta)\\) is an \\(N \\times N\\) covariance matrix. The mean function \\(m : \\mathbb{R}^{N \\times D}\n\\rightarrow \\mathbb{R}^{N}\\) can be anything, but the covariance function \\(K : \\mathbb{R}^{N \\times D} \\rightarrow \\mathbb{R}^{N \\times N}\\) must produce a positive-definite matrix for any input \\(x\\).1\nA popular covariance function, which will be used in the implementations later in this chapter, is an exponentiated quadratic function, \\[\n  K(x \\mid \\alpha, \\rho, \\sigma)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j} \\sigma^2,\n\\] where \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\) are hyperparameters defining the covariance function and where \\(\\delta_{i, j}\\) is the Kronecker delta function with value 1 if \\(i = j\\) and value 0 otherwise; this test is between the indexes \\(i\\) and \\(j\\), not between values \\(x_i\\) and \\(x_j\\). This kernel is obtained through a convolution of two independent Gaussian processes, \\(f_1\\) and \\(f_2\\), with kernels \\[\n  K_1(x \\mid \\alpha, \\rho)_{i, j}\n= \\alpha^2\n\\exp \\left(\n- \\dfrac{1}{2 \\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2\n\\right)\n\\] and \\[\n  K_2(x \\mid \\sigma)_{i, j}\n=\n\\delta_{i, j} \\sigma^2.\n\\]\nThe addition of \\(\\sigma^2\\) on the diagonal is important to ensure the positive definiteness of the resulting matrix in the case of two identical inputs \\(x_i = x_j\\). In statistical terms, \\(\\sigma\\) is the scale of the noise term in the regression.\nThe hyperparameter \\(\\rho\\) is the length-scale, and corresponds to the frequency of the functions represented by the Gaussian process prior with respect to the domain. Values of \\(\\rho\\) closer to zero lead the GP to represent high-frequency functions, whereas larger values of \\(\\rho\\) lead to low-frequency functions. The hyperparameter \\(\\alpha\\) is the marginal standard deviation. It controls the magnitude of the range of the function represented by the GP. If you were to take the standard deviation of many draws from the GP \\(f_1\\) prior at a single input \\(x\\) conditional on one value of \\(\\alpha\\) one would recover \\(\\alpha\\).\nThe only term in the squared exponential covariance function involving the inputs \\(x_i\\) and \\(x_j\\) is their vector difference, \\(x_i - x_j\\). This produces a process with stationary covariance in the sense that if an input vector \\(x\\) is translated by a vector \\(\\epsilon\\) to \\(x +\n\\epsilon\\), the covariance at any pair of outputs is unchanged, because \\(K(x \\mid \\theta) = K(x + \\epsilon \\mid \\theta)\\).\nThe summation involved is just the squared Euclidean distance between \\(x_i\\) and \\(x_j\\) (i.e., the squared \\(L_2\\) norm of their difference, \\(x_i -\nx_j\\)). This results in support for smooth functions in the process. The amount of variation in the function is controlled by the free hyperparameters \\(\\alpha\\), \\(\\rho\\), and \\(\\sigma\\).\nChanging the notion of distance from Euclidean to taxicab distance (i.e., an \\(L_1\\) norm) changes the support to functions which are continuous but not smooth.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Differential-Algebraic Equations"
+      "Gaussian Processes"
     ]
   },
   {
-    "objectID": "stan-users-guide/dae.html#control-dae.section",
-    "href": "stan-users-guide/dae.html#control-dae.section",
-    "title": "Differential-Algebraic Equations",
+    "objectID": "stan-users-guide/gaussian-processes.html#simulating-from-a-gaussian-process",
+    "href": "stan-users-guide/gaussian-processes.html#simulating-from-a-gaussian-process",
+    "title": "Gaussian Processes",
     "section": "",
-    "text": "Using dae_tol one can specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters in order to control the DAE solution.\nvector[3] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,\n                             relative_tolerance,\n                             absolute_tolerance,\n                             max_num_steps,\n                             alpha, beta, gamma);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nThe default value of relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
+    "text": "It is simplest to start with a Stan model that does nothing more than simulate draws of functions \\(f\\) from a Gaussian process. In practical terms, the model will draw values \\(y_n = f(x_n)\\) for finitely many input points \\(x_n\\).\nThe Stan model defines the mean and covariance functions in a transformed data block and then samples outputs \\(y\\) in the model using a multivariate normal distribution. To make the model concrete, the squared exponential covariance function described in the previous section will be used with hyperparameters set to \\(\\alpha^2 = 1\\), \\(\\rho^2 = 1\\), and \\(\\sigma^2 = 0.1\\), and the mean function \\(m\\) is defined to always return the zero vector, \\(m(x) = \\textbf{0}\\). Consider the following implementation of a Gaussian process simulator.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K;\n  vector[N] mu = rep_vector(0, N);\n  for (i in 1:(N - 1)) {\n    K[i, i] = 1 + 0.1;\n    for (j in (i + 1):N) {\n      K[i, j] = exp(-0.5 * square(x[i] - x[j]));\n      K[j, i] = K[i, j];\n    }\n  }\n  K[N, N] = 1 + 0.1;\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe above model can also be written more compactly using the specialized covariance function that implements the exponentiated quadratic kernel.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n}\ntransformed data {\n  matrix[N, N] K = gp_exp_quad_cov(x, 1.0, 1.0);\n  vector[N] mu = rep_vector(0, N);\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + 0.1;\n  }\n}\nparameters {\n  vector[N] y;\n}\nmodel {\n  y ~ multi_normal(mu, K);\n}\nThe input data are just the vector of inputs x and its size N. Such a model can be used with values of x evenly spaced over some interval in order to plot sample draws of functions from a Gaussian process.\n\n\nOnly the input data needs to change in moving from a univariate model to a multivariate model.\nThe only lines that change from the univariate model above are as follows.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n}\ntransformed data {\n  // ...\n}\nThe data are now declared as an array of vectors instead of an array of scalars; the dimensionality D is also declared.\nIn the remainder of the chapter, univariate models will be used for simplicity, but any of the models could be changed to multivariate in the same way as the simple sampling model. The only extra computational overhead from a multivariate model is in the distance calculation.\n\n\n\nA more efficient implementation of the simulation model can be coded in Stan by relocating, rescaling and rotating an isotropic standard normal variate. Suppose \\(\\eta\\) is an an isotropic standard normal variate \\[\n\\eta \\sim \\textsf{normal}(\\textbf{0}, \\textrm{I}),\n\\] where \\(\\textbf{0}\\) is an \\(N\\)-vector of 0 values and \\(\\textrm{I}\\) is the \\(N\n\\times N\\) identity matrix. Let \\(L\\) be the Cholesky decomposition of \\(K(x \\mid \\theta)\\), i.e., the lower-triangular matrix \\(L\\) such that \\(LL^{\\top} =\nK(x \\mid \\theta)\\). Then the transformed variable \\(\\mu + L\\eta\\) has the intended target distribution, \\[\n  \\mu + L\\eta \\sim \\textsf{multivariate normal}(\\mu(x), K(x \\mid \\theta)).\n\\]\nThis transform can be applied directly to Gaussian process simulation.\nThis model has the same data declarations for N and x, and the same transformed data definitions of mu and K as the previous model, with the addition of a transformed data variable for the Cholesky decomposition. The parameters change to the raw parameters sampled from an isotropic standard normal, and the actual parameters are defined in generated quantities.\n// ...\ntransformed data {\n  matrix[N, N] L;\n  // ...\n  L = cholesky_decompose(K);\n}\nparameters {\n  vector[N] eta;\n}\nmodel {\n  eta ~ std_normal();\n}\ngenerated quantities {\n  vector[N] y;\n  y = mu + L * eta;\n}\nThe Cholesky decomposition is only computed once, after the data are loaded and the covariance matrix K computed. The isotropic normal distribution for eta is specified as a vectorized univariate distribution for efficiency; this specifies that each eta[n] has an independent standard normal distribution. The sampled vector y is then defined as a generated quantity using a direct encoding of the transform described above.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Differential-Algebraic Equations"
+      "Gaussian Processes"
     ]
   },
   {
-    "objectID": "stan-users-guide/cross-validation.html",
-    "href": "stan-users-guide/cross-validation.html",
-    "title": "Held-Out Evaluation and Cross-Validation",
+    "objectID": "stan-users-guide/gaussian-processes.html#fit-gp.section",
+    "href": "stan-users-guide/gaussian-processes.html#fit-gp.section",
+    "title": "Gaussian Processes",
     "section": "",
-    "text": "Held-out evaluation involves splitting a data set into two parts, a training data set and a test data set. The training data is used to estimate the model and the test data is used for evaluation. Held-out evaluation is commonly used to declare winners in predictive modeling competitions such as those run by Kaggle.\nCross-validation involves repeated held-out evaluations performed by partitioning a single data set in different ways. The training/test split can be done either by randomly selecting the test set, or by partitioning the data set into several equally-sized subsets and then using each subset in turn as the test data with the other folds as training data.\nHeld-out evaluation and cross-validation may involve any kind of predictive statistics, with common choices being the predictive log density on test data, squared error of parameter estimates, or accuracy in a classification task.\n\n\nGiven training data \\((x, y)\\) consisting of parallel sequences of predictors and observations and test data \\((\\tilde{x}, \\tilde{y})\\) of the same structure, the posterior predictive density is \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n=\n\\int\n  p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n  \\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta,\n\\]\nwhere \\(\\theta\\) is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors \\(\\tilde{x}\\) and the training data \\((x, y).\\)\nThis integral may be calculated with Monte Carlo methods as usual, \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\] where the \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y)\\) are draws from the posterior given only the training data \\((x, y).\\)\nTo avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid \\tilde{x}, x, y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M \\exp(\\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}))\n\\\\[4pt]\n& = & -\\log M + \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)})\n\\end{eqnarray*}\\] where the log sum of exponentials function is defined so as to make the above equation hold, \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\, \\mu_m\n= \\log \\sum_{m=1}^M \\exp(\\mu_m).\n\\] The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\mu_m\n= \\textrm{max}(\\mu)\n+ \\log \\sum_{m = 1}^M \\exp(\\mu_m - \\textrm{max}(\\mu)).\n\\] Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms \\(\\mu_m - \\textrm{max}(\\mu) \\leq 0\\), and thus will not overflow.\n\n\nTo evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of \\(\\log p(\\tilde{y} \\mid \\tilde{x},\n\\theta^{(m)}).\\)\nHere is the code for evaluating the log posterior predictive density in a simple linear regression of the test data \\(\\tilde{y}\\) given predictors \\(\\tilde{x}\\) and training data \\((x, y).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  vector[N_tilde] y_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  real log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);\n}\nOnly the training data x and y are used in the model block. The test data y_tilde and test predictors x_tilde appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use x_tilde in the model block—only the test observations y_tilde are unknown before they are predicted.\nGiven \\(M\\) posterior draws from Stan, the sequence log_p[1:M] will be available, so that the log posterior predictive density of the test data given training data and predictors is just log_sum_exp(log_p) - log(M).\n\n\n\n\n\n\nEstimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If \\(\\theta\\) is the true value and \\(\\hat{\\theta}\\) the estimate, then error is just the difference between the prediction and the true value, \\[\n\\textrm{err} = \\hat{\\theta} - \\theta.\n\\]\nIf the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. Absolute error is defined as \\[\n\\textrm{abs-err} = \\left| \\hat{\\theta} - \\theta \\right|\n\\] and squared error as \\[\n\\textrm{sq-err} = \\left( \\hat{\\theta} - \\theta \\right)^2.\n\\] Gneiting and Raftery (2007) provide a thorough overview of such scoring rules and their properties.\nBayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.\n\n\n\nIn addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, \\(z_1, \\ldots, z_N,\\) then it is common to report mean square error (MSE), \\[\n\\textrm{mse}\n= \\frac{1}{N} \\sum_{n = 1}^N \\left( \\hat{z}_n - z_n\\right)^2.\n\\] To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as root mean square error (RMSE), \\[\n\\textrm{rmse} = \\sqrt{\\textrm{mean-sq-err}}.\n\\]\n\n\n\nConsider a simple linear regression model, parameters for the intercept \\(\\alpha\\) and slope \\(\\beta\\), along with predictors \\(\\tilde{x}_n\\). The standard Bayesian estimate is the expected value of \\(\\tilde{y}\\) given the predictors and training data, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}[\\tilde{y}_n \\mid \\tilde{x}_n, x, y]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}_n^{(m)}\n\\end{eqnarray*}\\] where \\(\\tilde{y}_n^{(m)}\\) is drawn from the data model \\[\n\\tilde{y}_n^{(m)}\n\\sim p(\\tilde{y}_n \\mid \\tilde{x}_n, \\alpha^{(m)}, \\beta^{(m)}),\n\\] for parameters \\(\\alpha^{(m)}\\) and \\(\\beta^{(m)}\\) drawn from the posterior, \\[\n(\\alpha^{(m)}, \\beta^{(m)}) \\sim p(\\alpha, \\beta \\mid x, y).\n\\]\nIn the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws \\(\\tilde{y}_n^{(m)}\\), the same result is obtained by averaging linear predictions, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}\\left[\n          \\alpha + \\beta \\cdot \\tilde{x}_n\n          \\mid \\tilde{x}_n, x, y\n       \\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\] This is possible because \\[\n\\tilde{y}_n^{(m)} \\sim \\textrm{normal}(\\tilde{y}_n \\mid \\alpha^{(m)} +\n\\beta^{(m)} \\cdot \\tilde{x}_n, \\sigma^{(m)}),\n\\] and the normal distribution has symmetric error so that the expectation of \\(\\tilde{y}_n^{(m)}\\) is the same as \\(\\alpha^{(m)} + \\beta^{(m)} \\cdot\n\\tilde{x}_n\\). Replacing the sampled quantity \\(\\tilde{y}_n^{(m)}\\) with its expectation is a general variance reduction technique for Monte Carlo estimates known as Rao-Blackwellization (Rao 1945; Blackwell 1947).\nIn the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, \\[\\begin{eqnarray*}\n\\tilde{y}_n^{(m)}\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\left( \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n \\right)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\frac{1}{M} \\sum_{m = 1}^M (\\beta^{(m)} \\cdot \\tilde{x}_n)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\left( \\frac{1}{M} \\sum_{m = 1}^M \\beta^{(m)}\\right) \\cdot \\tilde{x}_n\n\\\\[4pt]\n& = & \\hat{\\alpha} + \\hat{\\beta} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\]\nIn Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  // ...\n}\n// ...\ngenerated quantities {\n  vector[N_tilde] tilde_y = alpha + beta * x_tilde;\n}\nThe posterior mean of tilde_y calculated by Stan is the Bayesian estimate \\(\\hat{\\tilde{y}}.\\) The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.\n\n\n\n\nCross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation (Vehtari, Gelman, and Gabry 2017).\nPartitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.\n\n\nFor the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.\nfunctions {\n  array[] int permutation_rng(int N) {\n    array[N] int y;\n    for (n in 1 : N) {\n      y[n] = n;\n    }\n    vector[N] theta = rep_vector(1.0 / N, N);\n    for (n in 1 : size(y)) {\n      int i = categorical_rng(theta);\n      int temp = y[n];\n      y[n] = y[i];\n      y[i] = temp;\n    }\n    return y;\n  }\n}\nThe name of the function must end in _rng because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers 1:N.1\nThe transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=N&gt; N_test;\n}\ntransformed data {\n  int N_train = N - N_test;\n  array[N] int permutation = permutation_rng(N);\n  vector[N_train] x_train = x[permutation[1 : N_train]];\n  vector[N_train] y_train = y[permutation[1 : N_train]];\n  vector[N_test] x_test = x[permutation[N_train + 1 : N]];\n  vector[N_test] y_test = y[permutation[N_train + 1 : N]];\n}\nRecall that in Stan, permutation[1:N_train] is an array of integers, so that x[permutation[1 : N_train]] is a vector defined for i in 1:N_train by\nx[permutation[1 : N_train]][i] = x[permutation[1:N_train][i]]\n                               = x[permutation[i]]\nGiven the test/train split, the rest of the model is straightforward.\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_train ~ normal(alpha + beta * x_train, sigma);\n  { alpha, beta, sigma } ~ normal(0, 1);\n}\ngenerated quantities {\n  vector[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);\n  vector[N] err = y_test_sim - y_hat;\n}\nThe prediction y_test_hat is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, \\[\\begin{eqnarray*}\n\\hat{y}^{\\textrm{test}}\n& = & \\mathbb{E}\\left[ y^{\\textrm{test}} \\mid x^{\\textrm{test}}, x^{\\textrm{train}} y^{\\textrm{train}} \\right]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)}.\n\\end{eqnarray*}\\]\nBecause the test set is constant and the expectation operator is linear, the posterior mean of err as defined in the Stan program will be the error of the posterior mean estimate, \\[\\begin{eqnarray*}\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n  - y^{\\textrm{test}}\n\\\\[4pt]\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)} - y^{\\textrm{test}},\n\\end{eqnarray*}\\] where \\[\n\\hat{y}^{\\textrm{test}(m)}\n\\sim p(y \\mid x^{\\textrm{test}}, x^{\\textrm{train}},\ny^{\\textrm{train}}).\n\\] This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should not be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.\nBecause the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.\n\n\n\nIt is straightforward to declare the variable permutation in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.\n\n\n\nCross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to mixed replication in posterior predictive checking, where there is a choice to simulate new elements of existing groups or generate entirely new groups.\nEducation testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.\n\n\n\nOften data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.\n\n\n\nVehtari, Gelman, and Gabry (2017) introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo (Gabry et al. 2019).",
+    "text": "The full generative model for a GP with a normal outcome, \\(y \\in \\mathbb{R}^N\\), with inputs \\(x \\in \\mathbb{R}^N\\), for a finite \\(N\\): \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\nf      &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i    &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\}\n\\end{align*}\\] With a normal outcome, it is possible to integrate out the Gaussian process \\(f\\), yielding the more parsimonious model: \\[\\begin{align*}\n\\rho   &\\sim \\textsf{InvGamma}(5, 5) \\\\\n\\alpha &\\sim \\textsf{normal}(0, 1) \\\\\n\\sigma &\\sim \\textsf{normal}(0, 1) \\\\\ny      &\\sim \\textsf{multivariate normal}\n         \\left(0, K(x \\mid \\alpha, \\rho) + \\textbf{I}_N \\sigma^2\\right) \\\\\n\\end{align*}\\]\nIt can be more computationally efficient when dealing with a normal outcome to integrate out the Gaussian process, because this yields a lower-dimensional parameter space over which to do inference. We’ll fit both models in Stan. The former model will be referred to as the latent variable GP, while the latter will be called the marginal likelihood GP.\nThe hyperparameters controlling the covariance function of a Gaussian process can be fit by assigning them priors, like we have in the generative models above, and then computing the posterior distribution of the hyperparameters given observed data. The priors on the parameters should be defined based on prior knowledge of the scale of the output values (\\(\\alpha\\)), the scale of the output noise (\\(\\sigma\\)), and the scale at which distances are measured among inputs (\\(\\rho\\)). See the Gaussian process priors section for more information about how to specify appropriate priors for the hyperparameters.\nThe Stan program implementing the marginal likelihood GP is shown below. The program is similar to the Stan programs that implement the simulation GPs above, but because we are doing inference on the hyperparameters, we need to calculate the covariance matrix K in the model block, rather than the transformed data block.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  vector[N] mu = rep_vector(0, N);\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N, N] L_K;\n  matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n  real sq_sigma = square(sigma);\n\n  // diagonal elements\n  for (n in 1:N) {\n    K[n, n] = K[n, n] + sq_sigma;\n  }\n\n  L_K = cholesky_decompose(K);\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y ~ multi_normal_cholesky(mu, L_K);\n}\nThe data block declares a vector y of observed values y[n] for inputs x[n]. The transformed data block now only defines the mean vector to be zero. The three hyperparameters are defined as parameters constrained to be non-negative. The computation of the covariance matrix K is now in the model block because it involves unknown parameters and thus can’t simply be precomputed as transformed data. The rest of the model consists of the priors for the hyperparameters and the multivariate Cholesky-parameterized normal distribution, only now the value y is known and the covariance matrix K is an unknown dependent on the hyperparameters, allowing us to learn the hyperparameters.\nWe have used the Cholesky parameterized multivariate normal rather than the standard parameterization because it allows us to the cholesky_decompose function which has been optimized for both small and large matrices. When working with small matrices the differences in computational speed between the two approaches will not be noticeable, but for larger matrices (\\(N \\gtrsim 100\\)) the Cholesky decomposition version will be faster.\nHamiltonian Monte Carlo sampling is fast and effective for hyperparameter inference in this model (Neal 1997). If the posterior is well-concentrated for the hyperparameters the Stan implementation will fit hyperparameters in models with a few hundred data points in seconds.\n\n\nWe can also explicitly code the latent variable formulation of a GP in Stan. This will be useful for when the outcome is not normal. We’ll need to add a small positive term, \\(\\delta\\) to the diagonal of the covariance matrix in order to ensure that our covariance matrix remains positive definite.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\nTwo differences between the latent variable GP and the marginal likelihood GP are worth noting. The first is that we have augmented our parameter block with a new parameter vector of length \\(N\\) called eta. This is used in the model block to generate a multivariate normal vector called \\(f\\), corresponding to the latent GP. We put a \\(\\textsf{normal}(0,1)\\) prior on eta like we did in the Cholesky-parameterized GP in the simulation section. The second difference is that although we could code the distribution statement for \\(y\\) with one \\(N\\)-dimensional multivariate normal with an identity covariance matrix multiplied by \\(\\sigma^2\\), we instead use vectorized univariate normal distribution, which is equivalent but more efficient to use.\n\n\n\n\nGaussian processes can be generalized the same way as standard linear models by introducing a link function. This allows them to be used as discrete data models.\n\n\nIf we want to model count data, we can remove the \\(\\sigma\\) parameter, and use poisson_log, which implements a Poisson distribution with log link function, rather than normal. We can also add an overall mean parameter, \\(a\\), which will account for the marginal expected value for \\(y\\). We do this because we cannot center count data like we would for normally distributed data.\ndata {\n  // ...\n  array[N] int&lt;lower=0&gt; y;\n  // ...\n}\n// ...\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ poisson_log(a + f);\n}\n\n\n\nFor computational reasons, we may want to integrate out the latent variable \\(f\\), as was done in the normal output model. Unfortunately, exact marginalization over \\(f\\) is not possible when the outcome model is not normal. Instead, we may perform approximate marginalization with an embedded Laplace approximation (Rasmussen and Williams 2006; Rue, Martino, and Chopin 2009; Margossian et al. 2020). To do so, we first use the function laplace_marginal to approximate the marginal likelihood \\(p(y \\mid \\rho, \\alpha, a)\\) and sample the hyperparameters with Hamiltonian Monte Carlo sampling. Then, we recover the integrated out \\(f\\) in the generated quantities block using laplace_latent_rng.\nThe embedded Laplace approximation computes a Gaussian approximation of the conditional posterior, \\[\n  \\hat p_\\mathcal{L}(f \\mid \\rho, \\alpha, a, y) \\approx p(f \\mid \\rho, \\alpha, a, y),\n\\] where \\(\\hat p_\\mathcal{L}\\) is a Gaussian that matches the mode and curvature of \\(p(f \\mid \\rho, \\alpha, a, y)\\). We then obtain an approximation of the marginal likelihood as follows: \\[\n  \\hat p_\\mathcal{L}(y \\mid \\rho, \\alpha, a)\n    = \\frac{p(f^* \\mid \\alpha, \\rho) p(y \\mid f^*, a)}{\n    \\hat p_\\mathcal{L}(f^* \\mid \\rho, \\alpha, a, y)},\n\\] where \\(f^*\\) is the mode of \\(p(f \\mid \\rho, \\alpha, a, y)\\), obtained via numerical optimization.\nTo use Stan’s embedded Laplace approximation, we must define the prior covariance function and the log likelihood function in the functions block.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int y) {\n      return poisson_log_lpmf(y | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n\n}\nThe embedded Laplace relies on calculations of the log likelihood’s Hessian, \\(\\partial^2 \\log p(y \\mid f, a, \\rho, \\alpha) / \\partial f^2\\), and these calculations can be much faster when the Hessian is sparse. In particular, it is expected that the Hessian is block diagonal. In the transformed data block we can specify the block size of the Hessian.\ntransformed data {\n  int hessian_block_size = 1;\n}\nFor example, if \\(y_i\\) depends only on \\(f_i\\), then the Hessian of the log likelihood is diagonal and the block size is 1. On the other hand, if the Hessian is not sparse, then we set the hessian block size to \\(N\\), where \\(N\\) is the dimension of \\(f\\). Currently, Stan does not check the block size of the Hessian and so the user is responsible for correctly specifying the block size.\nFinally, we increment target in the model block with the approximation to \\(\\log p(y \\mid \\rho, \\alpha, a)\\).\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  target += laplace_marginal(ll_function, (a, y), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\nNotice that we do not need to construct \\(f\\) explicitly, since it is marginalized out. Instead, we can recover the latent variables in generated quantities:\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nUsers can set the control parameters of the embedded Laplace approximation, via laplace_marginal_tol and laplace_latent_tol_rng. When using these functions, the user must set all the control options and store them in a tuple. These control parameters mostly concern the numerical optimizer used to find the mode \\(f^*\\) of \\(p(f \\mid \\rho, \\alpha, a)\\).\ntransformed data {\n  tuple(vector[N], real, int, int, int, int) laplace_ops;\n  laplace_ops.1 = rep_vector(0, N);  // starting point for Laplace optimizer\n  laplace_ops.2 = 1.49e-8;           // tolerance for optimizer\n  laplace_ops.3 = 500;               // maximum number of steps for optimizer.\n  laplace_ops.4 = 1;              // solver type being used.\n  laplace_ops.5 = 1000;           // max number of steps for linesearch.\n  laplace_ops.6 = 1;              // allow_fallback (1: TRUE, 0: FALSE)\nIf users want to depart from the defaults for only some of the control parameters, a tuple with the default values (as above) can be created with the helper callable generate_laplace_options(), and the specific control parameter can then be modified,\ntransformed data {\n  tuple(vector[N], real, int, int, int, int, int) laplace_ops =\n    generate_laplace_options(N);\n\n  laplace_ops.2 = 1e-6; // make tolerance of the optimizer less strict.\n}\nThe tuple laplace_ops is then passed to laplace_marginal_tol and laplace_rng_tol.\nmodel {\n// ...\n\n  target += laplace_marginal_tol(ll_function, (a, y), hessian_block_size,\n                                 cov_function, (rho, alpha, x, N, delta),\n                                 laplace_ops);\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta),\n                                   laplace_ops);\n}\nStan also provides support for a limited menu of built-in functions, including the Poisson distribution with a log link and and prior mean \\(m\\). When using such a built-in function, the user does not need to specify a likelihood in the functions block. However, the user must strictly follow the signature of the likelihood: in this case, \\(m\\) must be a vector of length \\(N\\) (to allow for different offsets for each observation \\(y_i\\)) and we must indicate which element of \\(f\\) each component of \\(y\\) matches using the variable \\(y_\\text{index}\\). In our example, there is a simple pairing \\((y_i, f_i)\\), however we could imagine a scenario where multiple observations \\((y_{j1}, y_{j2}, ...)\\) are observed for a single \\(f_j\\).\ntransformed data {\n  // ...\n  array[n_obs] int y_index;\n  for (i in 1:n_obs) y_index[i] = i - 1;\n}\n\n// ...\n\ntransformed parameter {\n  vector[N] m = rep_vector(a, N);\n}\n\nmodel {\n  // ...\n  target += laplace_marginal_poisson_log_lpmf(y | y_index, m,\n                                       cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_poisson_log_rng(y, y_index, m,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nAs before, we could specify the control parameters for the embedded Laplace approximation using laplace_marginal_tol_poisson_log_lpmf and laplace_latent_tol_poisson_log_nrg.\nMarginalization with a Laplace approximation can lead to faster inference, however it also introduces an approximation error. In practice, this error is negligible when using a Poisson likelihood and the approximation works well for log concave likelihoods (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). Still, users should exercise caution, especially when trying unconventional likelihoods.\n\n\n\nFor binary classification problems, the observed outputs \\(z_n \\in\n\\{ 0,1 \\}\\) are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs \\(y_n\\) through the logistic link, \\[\nz_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(y_n)),\n\\] or in other words, \\[\n\\Pr[z_n = 1] = \\operatorname{logit}^{-1}(y_n).\n\\]\nWe can extend our latent variable GP Stan program to deal with classification problems. Below a is the bias term, which can help account for imbalanced classes in the training data:\ndata {\n  // ...\n  array[N] int&lt;lower=0, upper=1&gt; z;\n  // ...\n}\n// ...\nmodel {\n  // ...\n  z ~ bernoulli_logit(a + f);\n}\n\n\n\nAs with the Poisson GP, we cannot marginalize the latent variables exactly, however we can resort to an embedded Laplace approximation.\nfunctions {\n  // log likelihood function\n  real ll_function(vector f, real a, array[] int z) {\n      return bernoulli_logit_lpmf(z | a + f);\n  }\n\n  // covariance function\n  matrix cov_function(real rho, real alpha, array[] real x, int N, real delta) {\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    return add_diag(K, delta)\n  }\n}\n\n// ...\n\ntransformed data {\n  int hessian_block_size = 1;\n}\n\n// ...\n\nmodel {\n  target += laplace_marginal(ll_function, (a, z), hessian_block_size,\n                             cov_function, (rho, alpha, x, N, delta));\n}\n\ngenerated quantities {\n  vector[N] f = laplace_latent_rng(ll_function, (a, z), hessian_block_size,\n                                   cov_function, (rho, alpha, x, N, delta));\n}\nWhile marginalization with a Laplace approximation can lead to faster inference, it also introduces an approximation error. In practice, this error may not be negligible with a Bernoulli likelihood; for more discussion see, e.g. (Vehtari et al. 2016; Margossian et al. 2020).\n\n\n\n\nIf we have multivariate inputs \\(x \\in \\mathbb{R}^D\\), the squared exponential covariance function can be further generalized by fitting a scale parameter \\(\\rho_d\\) for each dimension \\(d\\), \\[\n  k(x \\mid \\alpha, \\vec{\\rho}, \\sigma)_{i, j} = \\alpha^2 \\exp\n\\left(-\\dfrac{1}{2}\n\\sum_{d=1}^D \\dfrac{1}{\\rho_d^2} (x_{i,d} - x_{j,d})^2\n\\right)\n+ \\delta_{i, j}\\sigma^2.\n\\] The estimation of \\(\\rho\\) was termed “automatic relevance determination” by Neal (1996), but this is misleading, because the magnitude of the scale of the posterior for each \\(\\rho_d\\) is dependent on the scaling of the input data along dimension \\(d\\). Moreover, the scale of the parameters \\(\\rho_d\\) measures non-linearity along the \\(d\\)-th dimension, rather than “relevance” (Piironen and Vehtari 2016).\nA priori, the closer \\(\\rho_d\\) is to zero, the more nonlinear the conditional mean in dimension \\(d\\) is. A posteriori, the actual dependencies between \\(x\\) and \\(y\\) play a role. With one covariate \\(x_1\\) having a linear effect and another covariate \\(x_2\\) having a nonlinear effect, it is possible that \\(\\rho_1 &gt; \\rho_2\\) even if the predictive relevance of \\(x_1\\) is higher (Rasmussen and Williams 2006, 80). The collection of \\(\\rho_d\\) (or \\(1/\\rho_d\\)) parameters can also be modeled hierarchically.\nThe implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying rho to be an array.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] vector[D] x;\n  vector[N] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  array[D] real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\nmodel {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y ~ normal(f, sigma);\n}\n\n\n\nFormulating priors for GP hyperparameters requires the analyst to consider the inherent statistical properties of a GP, the GP’s purpose in the model, and the numerical issues that may arise in Stan when estimating a GP.\nPerhaps most importantly, the parameters \\(\\rho\\) and \\(\\alpha\\) are weakly identified (Zhang 2004). The ratio of the two parameters is well-identified, but in practice we put independent priors on the two hyperparameters because these two quantities are more interpretable than their ratio.\n\n\nGPs are a flexible class of priors and, as such, can represent a wide spectrum of functions. For length scales below the minimum spacing of the covariates the GP likelihood plateaus. Unless regularized by a prior, this flat likelihood induces considerable posterior mass at small length scales where the observation variance drops to zero and the functions supported by the GP begin to exactly interpolate between the input data. The resulting posterior not only significantly overfits to the input data, it also becomes hard to accurately sample using Euclidean HMC.\nWe may wish to put further soft constraints on the length-scale, but these are dependent on how the GP is used in our statistical model.\nIf our model consists of only the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\left(0, K(x \\mid \\alpha, \\rho)\\right) \\\\\ny_i &\\sim \\textsf{normal}(f_i, \\sigma) \\, \\forall i \\in \\{1, \\dots, N\\} \\\\\n    & x \\in \\mathbb{R}^{N \\times D}, \\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe likely don’t need constraints beyond penalizing small length-scales. We’d like to allow the GP prior to represent both high-frequency and low-frequency functions, so our prior should put non-negligible mass on both sets of functions. In this case, an inverse gamma, inv_gamma_lpdf in Stan’s language, will work well as it has a sharp left tail that puts negligible mass on infinitesimal length-scales, but a generous right tail, allowing for large length-scales. Inverse gamma priors will avoid infinitesimal length-scales because the density is zero at zero, so the posterior for length-scale will be pushed away from zero. An inverse gamma distribution is one of many zero-avoiding or boundary-avoiding distributions.2.\nIf we’re using the GP as a component in a larger model that includes an overall mean and fixed effects for the same variables we’re using as the domain for the GP, i.e.: \\[\\begin{align*}\nf   &\\sim \\textsf{multivariate normal}\\big(0, K(x \\mid \\alpha, \\rho)\\big) \\\\\ny_i &\\sim \\textsf{normal}\\left(\\beta_0 + x_i \\beta_{[1:D]} + f_i, \\sigma\\right) \\, \\forall i\n  \\in \\{1, \\dots, N\\} \\\\\n    & x_i^T, \\beta_{[1:D]} \\in \\mathbb{R}^D,\\quad\n      x \\in \\mathbb{R}^{N \\times D},\\quad\n      f \\in \\mathbb{R}^N\n\\end{align*}\\]\nwe’ll likely want to constrain large length-scales as well. A length scale that is larger than the scale of the data yields a GP posterior that is practically linear (with respect to the particular covariate) and increasing the length scale has little impact on the likelihood. This will introduce nonidentifiability in our model, as both the fixed effects and the GP will explain similar variation. In order to limit the amount of overlap between the GP and the linear regression, we should use a prior with a sharper right tail to limit the GP to higher-frequency functions. We can use a generalized inverse Gaussian distribution: \\[\\begin{align*}\nf(x \\mid a, b, p) &= \\dfrac{\\left(a/b\\right)^{p/2}}{2K_p\\left(\\sqrt{ab}\\right)} x^{p - 1}\\exp\\big(-(ax + b\n  / x)/2\\big) \\\\\n  & x, a, b \\in \\mathbb{R}^{+},\\quad\n    p \\in \\mathbb{Z}\n\\end{align*}\\]\nwhich has an inverse gamma left tail if \\(p \\leq 0\\) and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:\nfunctions {\n  real generalized_inverse_gaussian_lpdf(real x, int p,\n                                        real a, real b) {\n    return p * 0.5 * log(a / b)\n      - log(2 * modified_bessel_second_kind(p, sqrt(a * b)))\n      + (p - 1) * log(x)\n      - (a * x + b / x) * 0.5;\n }\n}\ndata {\n  // ...\n}\nIf we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of \\(\\rho\\) in a fixed-domain \\([0,T]\\), we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is \\(T / \\pi\n\\rho\\). You can see that as \\(\\rho\\) decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.\n\n\n\nThe parameter \\(\\alpha\\) corresponds to how much of the variation is explained by the regression function and has a similar role to the prior variance for linear model weights. This means the prior can be the same as used in linear models, such as a half-\\(t\\) prior on \\(\\alpha\\).\nA half-\\(t\\) or half-Gaussian prior on alpha also has the benefit of putting nontrivial prior mass around zero. This allows the GP support the zero functions and allows the possibility that the GP won’t contribute to the conditional mean of the total output.\n\n\n\n\nSuppose for a given sequence of inputs \\(x\\) that the corresponding outputs \\(y\\) are observed. Given a new sequence of inputs \\(\\tilde{x}\\), the posterior predictive distribution of their labels is computed by sampling outputs \\(\\tilde{y}\\) according to \\[\np\\left(\\tilde{y} \\mid \\tilde{x},x,y\\right)\n\\ = \\\n\\frac{p\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right)}\n     {p(y \\mid x)}\n\\ \\propto \\\np\\left(\\tilde{y}, y \\mid \\tilde{x},x\\right).\n\\]\nA direct implementation in Stan defines a model in terms of the joint distribution of the observed \\(y\\) and unobserved \\(\\tilde{y}\\).\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  eta ~ std_normal();\n\n  y1 ~ normal(f[1:N1], sigma);\n}\ngenerated quantities {\n  vector[N2] y2;\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f[N1 + n2], sigma);\n  }\n}\nThe input vectors x1 and x2 are declared as data, as is the observed output vector y1. The unknown output vector y2, which corresponds to input vector x2, is declared in the generated quantities block and will be sampled when the model is executed.\nA transformed data block is used to combine the input vectors x1 and x2 into a single vector x.\nThe model block declares and defines a local variable for the combined output vector f, which consists of the concatenation of the conditional mean for known outputs y1 and unknown outputs y2. Thus the combined output vector f is aligned with the combined input vector x. All that is left is to define the univariate normal distribution statement for y.\nThe generated quantities block defines the quantity y2. We generate y2 by randomly generating N2 values from univariate normals with each mean corresponding to the appropriate element in f.\n\n\nWe can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.\nConsider the following full model for prediction using logistic Gaussian process regression.\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  array[N1] int&lt;lower=0, upper=1&gt; z1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  real delta = 1e-9;\n  int&lt;lower=1&gt; N = N1 + N2;\n  array[N] real x;\n  for (n1 in 1:N1) {\n    x[n1] = x1[n1];\n  }\n  for (n2 in 1:N2) {\n    x[N1 + n2] = x2[n2];\n  }\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real a;\n  vector[N] eta;\n}\ntransformed parameters {\n  vector[N] f;\n  {\n    matrix[N, N] L_K;\n    matrix[N, N] K = gp_exp_quad_cov(x, alpha, rho);\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta;\n  }\n}\nmodel {\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  a ~ std_normal();\n  eta ~ std_normal();\n\n  z1 ~ bernoulli_logit(a + f[1:N1]);\n}\ngenerated quantities {\n  array[N2] int z2;\n  for (n2 in 1:N2) {\n    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);\n  }\n}\n\n\n\nBayesian predictive inference for Gaussian processes with Gaussian observations can be sped up by deriving the posterior analytically, then directly sampling from it.\nJumping straight to the result, \\[\np\\left(\\tilde{y} \\mid \\tilde{x},y,x\\right)\n=\n\\textsf{normal}\\left(K^{\\top}\\Sigma^{-1}y,\\\n                \\Omega - K^{\\top}\\Sigma^{-1}K\\right),\n\\] where \\(\\Sigma = K(x \\mid \\alpha, \\rho, \\sigma)\\) is the result of applying the covariance function to the inputs \\(x\\) with observed outputs \\(y\\), \\(\\Omega =\nK(\\tilde{x} \\mid \\alpha, \\rho)\\) is the result of applying the covariance function to the inputs \\(\\tilde{x}\\) for which predictions are to be inferred, and \\(K\\) is the matrix of covariances between inputs \\(x\\) and \\(\\tilde{x}\\), which in the case of the exponentiated quadratic covariance function would be \\[\nK(x \\mid \\alpha, \\rho)_{i, j} = \\alpha^2 \\exp\\left(-\\dfrac{1}{2 \\rho^2}\n\\sum_{d=1}^D \\left(x_{i,d} - \\tilde{x}_{j,d}\\right)^2\\right).\n\\]\nThere is no noise term including \\(\\sigma^2\\) because the indexes of elements in \\(x\\) and \\(\\tilde{x}\\) are never the same.\nThis Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called gp_pred_rng which will generate a draw from the posterior predictive mean conditioned on observed data y1. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of \\(p(\\tilde{y})\\).\nfunctions {\n  vector gp_pred_rng(array[] real x2,\n                     vector y1,\n                     array[] real x1,\n                     real alpha,\n                     real rho,\n                     real sigma,\n                     real delta) {\n    int N1 = rows(y1);\n    int N2 = size(x2);\n    vector[N2] f2;\n    {\n      matrix[N1, N1] L_K;\n      vector[N1] K_div_y1;\n      matrix[N1, N2] k_x1_x2;\n      matrix[N1, N2] v_pred;\n      vector[N2] f2_mu;\n      matrix[N2, N2] cov_f2;\n      matrix[N2, N2] diag_delta;\n      matrix[N1, N1] K;\n      K = gp_exp_quad_cov(x1, alpha, rho);\n      for (n in 1:N1) {\n        K[n, n] = K[n, n] + square(sigma);\n      }\n      L_K = cholesky_decompose(K);\n      K_div_y1 = mdivide_left_tri_low(L_K, y1);\n      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';\n      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);\n      f2_mu = (k_x1_x2' * K_div_y1);\n      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);\n      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;\n      diag_delta = diag_matrix(rep_vector(delta, N2));\n\n      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);\n    }\n    return f2;\n  }\n}\ndata {\n  int&lt;lower=1&gt; N1;\n  array[N1] real x1;\n  vector[N1] y1;\n  int&lt;lower=1&gt; N2;\n  array[N2] real x2;\n}\ntransformed data {\n  vector[N1] mu = rep_vector(0, N1);\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  matrix[N1, N1] L_K;\n  {\n    matrix[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);\n    real sq_sigma = square(sigma);\n\n    // diagonal elements\n    for (n1 in 1:N1) {\n      K[n1, n1] = K[n1, n1] + sq_sigma;\n    }\n\n    L_K = cholesky_decompose(K);\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n\n  y1 ~ multi_normal_cholesky(mu, L_K);\n}\ngenerated quantities {\n  vector[N2] f2;\n  vector[N2] y2;\n\n  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);\n  for (n2 in 1:N2) {\n    y2[n2] = normal_rng(f2[n2], sigma);\n  }\n}\n\n\n\n\nSuppose we have observations \\(y_i \\in \\mathbb{R}^M\\) observed at \\(x_i \\in \\mathbb{R}^K\\). One can model the data like so: \\[\\begin{align*}\ny_i  &\\sim \\textsf{multivariate normal}\\left(f(x_i), \\textbf{I}_M \\sigma^2\\right) \\\\\nf(x) &\\sim \\textsf{GP}\\big(m(x), K(x \\mid \\theta, \\phi)\\big) \\\\\n     & K(x \\mid \\theta) \\in \\mathbb{R}^{M \\times M}, \\quad\n       f(x), m(x) \\in \\mathbb{R}^M\n\\end{align*}\\] where the \\(K(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]}\\) entry defines the covariance between \\(f_m(x)\\) and \\(f_{m^\\prime}(x^\\prime)(x)\\). This construction of Gaussian processes allows us to learn the covariance between the output dimensions of \\(f(x)\\). If we parameterize our kernel \\(K\\): \\[\nK(x, x^\\prime \\mid \\theta, \\phi)_{[m, m^\\prime]} = k\\left(x, x^\\prime \\mid\n\\theta\\right) k\\left(m, m^\\prime \\mid \\phi\\right)\n\\] then our finite dimensional generative model for the above is: \\[\\begin{align*}\nf        &\\sim \\textsf{matrixnormal}\\big(m(x), K(x \\mid \\alpha, \\rho), C(\\phi)\\big) \\\\\ny_{i, m} &\\sim \\textsf{normal}(f_{i,m}, \\sigma) \\\\\nf        &\\in  \\mathbb{R}^{N \\times M}\n\\end{align*}\\] where \\(K(x \\mid \\alpha, \\rho)\\) is the exponentiated quadratic kernel we’ve used throughout this chapter, and \\(C(\\phi)\\) is a positive-definite matrix, parameterized by some vector \\(\\phi\\).\nThe matrix normal distribution has two covariance matrices: \\(K(x \\mid\n\\alpha, \\rho)\\) to encode column covariance, and \\(C(\\phi)\\) to define row covariance. The salient features of the matrix normal are that the rows of the matrix \\(f\\) are distributed: \\[\nf_{[n,]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[n,]}, K(x \\mid \\alpha,\n\\rho)_{[n,n]} C(\\phi)\\big)\n\\] and that the columns of the matrix \\(f\\) are distributed: \\[\nf_{[,m]} \\sim \\textsf{multivariate normal}\\big(m(x)_{[,m]}, K(x\n  \\mid \\alpha, \\rho) C(\\phi)_{[m,m]}\\big)\n\\] This also means means that \\(\\mathbb{E}\\left[f^T f\\right]\\) is equal to \\(\\operatorname{trace}\\!\\big(K(x \\mid \\alpha, \\rho)\\big) \\times C\\), whereas \\(\\mathbb{E}\\left[ff^T\\right]\\) is \\(\\operatorname{trace}(C) \\times K(x \\mid \\alpha, \\rho)\\). We can derive this using properties of expectation and the matrix normal density.\nWe should set \\(\\alpha\\) to \\(1.0\\) because the parameter is not identified unless we constrain \\(\\operatorname{trace}(C) = 1\\). Otherwise, we can multiply \\(\\alpha\\) by a scalar \\(d\\) and \\(C\\) by \\(1/d\\) and our likelihood will not change.\nWe can generate a random variable \\(f\\) from a matrix normal density in \\(\\mathbb{R}^{N \\times M}\\) using the following algorithm: \\[\\begin{align*}\n\\eta_{i,j} &\\sim \\textsf{normal}(0, 1) \\, \\forall i,j \\\\\nf          &= L_{K(x \\mid 1.0, \\rho)} \\, \\eta \\, L_C(\\phi)^T \\\\\nf          &\\sim \\textsf{matrixnormal}\\big(0, K(x \\mid 1.0, \\rho), C(\\phi)\\big) \\\\\n\\eta       &\\in \\mathbb{R}^{N \\times M} \\\\\nL_C(\\phi)  &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(C(\\phi)\\big) \\\\\nL_{K(x \\mid 1.0, \\rho)} &= \\texttt{cholesky}\\mathtt{\\_}\\texttt{decompose}\\big(K(x \\mid 1.0, \\rho)\\big)\n\\end{align*}\\]\nThis can be implemented in Stan using a latent-variable GP formulation. We’ve used \\(\\textsf{LKJCorr}\\) for \\(C(\\phi)\\), but any positive-definite matrix will do.\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] real x;\n  matrix[N, D] y;\n}\ntransformed data {\n  real delta = 1e-9;\n}\nparameters {\n  real&lt;lower=0&gt; rho;\n  vector&lt;lower=0&gt;[D] alpha;\n  real&lt;lower=0&gt; sigma;\n  cholesky_factor_corr[D] L_Omega;\n  matrix[N, D] eta;\n}\nmodel {\n  matrix[N, D] f;\n  {\n    matrix[N, N] K = gp_exp_quad_cov(x, 1.0, rho);\n    matrix[N, N] L_K;\n\n    // diagonal elements\n    for (n in 1:N) {\n      K[n, n] = K[n, n] + delta;\n    }\n\n    L_K = cholesky_decompose(K);\n    f = L_K * eta\n        * diag_pre_multiply(alpha, L_Omega)';\n  }\n\n  rho ~ inv_gamma(5, 5);\n  alpha ~ std_normal();\n  sigma ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(3);\n  to_vector(eta) ~ std_normal();\n\n  to_vector(y) ~ normal(to_vector(f), sigma);\n}\ngenerated quantities {\n  matrix[D, D] Omega;\n  Omega = L_Omega * L_Omega';\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Held-Out Evaluation and Cross-Validation"
+      "Example Models",
+      "Gaussian Processes"
     ]
   },
   {
-    "objectID": "stan-users-guide/cross-validation.html#evaluating-posterior-predictive.section",
-    "href": "stan-users-guide/cross-validation.html#evaluating-posterior-predictive.section",
-    "title": "Held-Out Evaluation and Cross-Validation",
-    "section": "",
-    "text": "Given training data \\((x, y)\\) consisting of parallel sequences of predictors and observations and test data \\((\\tilde{x}, \\tilde{y})\\) of the same structure, the posterior predictive density is \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n=\n\\int\n  p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n  \\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta,\n\\]\nwhere \\(\\theta\\) is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors \\(\\tilde{x}\\) and the training data \\((x, y).\\)\nThis integral may be calculated with Monte Carlo methods as usual, \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\] where the \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y)\\) are draws from the posterior given only the training data \\((x, y).\\)\nTo avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid \\tilde{x}, x, y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M \\exp(\\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}))\n\\\\[4pt]\n& = & -\\log M + \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)})\n\\end{eqnarray*}\\] where the log sum of exponentials function is defined so as to make the above equation hold, \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\, \\mu_m\n= \\log \\sum_{m=1}^M \\exp(\\mu_m).\n\\] The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\mu_m\n= \\textrm{max}(\\mu)\n+ \\log \\sum_{m = 1}^M \\exp(\\mu_m - \\textrm{max}(\\mu)).\n\\] Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms \\(\\mu_m - \\textrm{max}(\\mu) \\leq 0\\), and thus will not overflow.\n\n\nTo evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of \\(\\log p(\\tilde{y} \\mid \\tilde{x},\n\\theta^{(m)}).\\)\nHere is the code for evaluating the log posterior predictive density in a simple linear regression of the test data \\(\\tilde{y}\\) given predictors \\(\\tilde{x}\\) and training data \\((x, y).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  vector[N_tilde] y_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  real log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);\n}\nOnly the training data x and y are used in the model block. The test data y_tilde and test predictors x_tilde appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use x_tilde in the model block—only the test observations y_tilde are unknown before they are predicted.\nGiven \\(M\\) posterior draws from Stan, the sequence log_p[1:M] will be available, so that the log posterior predictive density of the test data given training data and predictors is just log_sum_exp(log_p) - log(M).",
+    "objectID": "stan-users-guide/gaussian-processes.html#footnotes",
+    "href": "stan-users-guide/gaussian-processes.html#footnotes",
+    "title": "Gaussian Processes",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nGaussian processes can be extended to covariance functions producing positive semi-definite matrices, but Stan does not support inference in the resulting models because the resulting distribution does not have unconstrained support.↩︎\nA boundary-avoiding prior is just one where the limit of the density is zero at the boundary, the result of which is estimates that are pushed away from the boundary.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Held-Out Evaluation and Cross-Validation"
+      "Example Models",
+      "Gaussian Processes"
     ]
   },
   {
-    "objectID": "stan-users-guide/cross-validation.html#estimation-error",
-    "href": "stan-users-guide/cross-validation.html#estimation-error",
-    "title": "Held-Out Evaluation and Cross-Validation",
+    "objectID": "stan-users-guide/floating-point.html",
+    "href": "stan-users-guide/floating-point.html",
+    "title": "Floating Point Arithmetic",
     "section": "",
-    "text": "Estimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If \\(\\theta\\) is the true value and \\(\\hat{\\theta}\\) the estimate, then error is just the difference between the prediction and the true value, \\[\n\\textrm{err} = \\hat{\\theta} - \\theta.\n\\]\nIf the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. Absolute error is defined as \\[\n\\textrm{abs-err} = \\left| \\hat{\\theta} - \\theta \\right|\n\\] and squared error as \\[\n\\textrm{sq-err} = \\left( \\hat{\\theta} - \\theta \\right)^2.\n\\] Gneiting and Raftery (2007) provide a thorough overview of such scoring rules and their properties.\nBayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.\n\n\n\nIn addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, \\(z_1, \\ldots, z_N,\\) then it is common to report mean square error (MSE), \\[\n\\textrm{mse}\n= \\frac{1}{N} \\sum_{n = 1}^N \\left( \\hat{z}_n - z_n\\right)^2.\n\\] To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as root mean square error (RMSE), \\[\n\\textrm{rmse} = \\sqrt{\\textrm{mean-sq-err}}.\n\\]\n\n\n\nConsider a simple linear regression model, parameters for the intercept \\(\\alpha\\) and slope \\(\\beta\\), along with predictors \\(\\tilde{x}_n\\). The standard Bayesian estimate is the expected value of \\(\\tilde{y}\\) given the predictors and training data, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}[\\tilde{y}_n \\mid \\tilde{x}_n, x, y]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}_n^{(m)}\n\\end{eqnarray*}\\] where \\(\\tilde{y}_n^{(m)}\\) is drawn from the data model \\[\n\\tilde{y}_n^{(m)}\n\\sim p(\\tilde{y}_n \\mid \\tilde{x}_n, \\alpha^{(m)}, \\beta^{(m)}),\n\\] for parameters \\(\\alpha^{(m)}\\) and \\(\\beta^{(m)}\\) drawn from the posterior, \\[\n(\\alpha^{(m)}, \\beta^{(m)}) \\sim p(\\alpha, \\beta \\mid x, y).\n\\]\nIn the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws \\(\\tilde{y}_n^{(m)}\\), the same result is obtained by averaging linear predictions, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}\\left[\n          \\alpha + \\beta \\cdot \\tilde{x}_n\n          \\mid \\tilde{x}_n, x, y\n       \\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\] This is possible because \\[\n\\tilde{y}_n^{(m)} \\sim \\textrm{normal}(\\tilde{y}_n \\mid \\alpha^{(m)} +\n\\beta^{(m)} \\cdot \\tilde{x}_n, \\sigma^{(m)}),\n\\] and the normal distribution has symmetric error so that the expectation of \\(\\tilde{y}_n^{(m)}\\) is the same as \\(\\alpha^{(m)} + \\beta^{(m)} \\cdot\n\\tilde{x}_n\\). Replacing the sampled quantity \\(\\tilde{y}_n^{(m)}\\) with its expectation is a general variance reduction technique for Monte Carlo estimates known as Rao-Blackwellization (Rao 1945; Blackwell 1947).\nIn the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, \\[\\begin{eqnarray*}\n\\tilde{y}_n^{(m)}\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\left( \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n \\right)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\frac{1}{M} \\sum_{m = 1}^M (\\beta^{(m)} \\cdot \\tilde{x}_n)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\left( \\frac{1}{M} \\sum_{m = 1}^M \\beta^{(m)}\\right) \\cdot \\tilde{x}_n\n\\\\[4pt]\n& = & \\hat{\\alpha} + \\hat{\\beta} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\]\nIn Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  // ...\n}\n// ...\ngenerated quantities {\n  vector[N_tilde] tilde_y = alpha + beta * x_tilde;\n}\nThe posterior mean of tilde_y calculated by Stan is the Bayesian estimate \\(\\hat{\\tilde{y}}.\\) The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.",
+    "text": "Computers approximate real values in \\(\\mathbb{R}\\) using a fixed number of bits. This chapter explains how this is done and why it is important for writing robust Stan (and other numerical) programs. The subfield of computer science devoted to studying how real arithmetic works on computers is called numerical analysis.\n\n\nStan’s arithmetic is implemented using double-precision arithmetic. The behavior of most1 modern computers follows the floating-point arithmetic, IEEE Standard for Floating-Point Arithmetic (IEEE 754).\n\n\nThe double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:\n\nsignificand (53 bits): base value representing significant digits\nexponent (11 bits): power of two multiplied by the base\n\nThe value of a finite floating point number is\n\\[\nv = (-1)^s \\times c \\, 2^q\n\\]\n\n\n\nA normal floating-point value does not use any leading zeros in its significand; subnormal numbers may use leading zeros. Not all I/O systems support subnormal numbers.\n\n\n\nThere are some reserved exponent values so that legal exponent values range between\\(-(2^{10}) + 2 = -1022\\) and \\(2^{10} - 1 = 1023\\). Legal significand values are between \\(-2^{52}\\) and \\(2^{52} - 1\\). Floating point allows the representation of both really big and really small values. Some extreme values are\n\nlargest normal finite number: \\(\\approx 1.8 \\times 10^{308}\\)\nlargest subnormal finite number: \\(\\approx 2.2 \\times 10^{308}\\)\nsmallest positive normal number: \\(\\approx 2.2 \\times 10^{-308}\\)\nsmallest positive subnormal number: \\(\\approx 4.9 \\times 10^{-324}\\)\n\n\n\n\nBecause of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., 0 == -0 evaluates to true).\n\n\n\nA specially chosen bit pattern is used for the not-a-number value (often written as NaN in programming language output, including Stan’s).\nStan provides a value function not_a_number() that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.\nStan also provides a test function is_nan(x) that returns 1 if x is not-a-number and 0 otherwise.\nNot-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as log and exp have the same behavior, propagating not-a-number values.\nMost of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.\nComparisons with not-a-number always return false, up to and including comparison with itself. That is, not_a_number() == not_a_number() somewhat confusingly returns false. That is why there is a built-in is_nan() function in Stan (and in C++). The only exception is negation, which remains coherent. This means not_a_number() != not_a_number() returns true.\nUndefined operations often return not-a-number values. For example, sqrt(-1) will evaluate to not-a-number.\n\n\n\nThere are also two special values representing positive infinity (\\(\\infty)\\) and negative infinity (\\(-\\infty\\)). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, log(0) evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that 0 == exp(log(0)). Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.\nThere are value functions positive_infinity() and negative_infinity() as well as a test function is_inf().\nPositive and negative infinity have the expected comparison behavior, so that negative_infinity() &lt; 0 evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.\nPositive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.\n\n\n\n\nIn programming languages such as Stan, numbers may be represented in standard decimal (base 10) notation. For example, 2.39 or -1567846.276452. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using scientific notation, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, 36.29e-3 represents the number \\(36.29 \\times\n10^{-3}\\), which is the same number as is represented by 0.03629.\n\n\n\nThe choice of significand provides \\(\\log_{10} 2^{53} \\approx 15.95\\) decimal (base 10) digits of arithmetic precision. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.\n\n\nIn practice, the finite amount of arithmetic precision leads to rounding, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,\n1 + 1e-20 == 1\nThe closest floating point number to \\(1 + 10^{-20}\\) turns out to be \\(1\\) itself. By contrast,\n0 + 1e-20 == 1e-20\nThis highlights the fact that precision depends on scale. Even though 1 + 1e-20 == 1, we have 1e-20 + 1e-20 == 2e-20, as expected.\nRounding also manifests itself in a lack of transitivity. In particular, it does not usually hold for three floating point numbers \\(a, b, c\\) that \\((a + b) + c = a + (b + c)\\).\nIn statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix \\(L\\) with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix \\(L \\, L^{\\top}\\). In practice, rounding and loss of precision may render the result \\(L \\, L^{\\top}\\) neither symmetric nor positive definite.\nIn practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding \\(L \\, L^{\\top}\\) with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.\n\n\n\nThe smallest number greater than zero is roughly \\(0 + 10^{-323}\\). The largest number less than one is roughly \\(1 - 10^{-15.95}\\). The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of \\(0.9999999999999999\\).\nFor this reason, the machine precision is said to be roughly \\(10^{-15.95}\\). This constant is available as machine_precision() in Stan.\n\n\n\nSpecial operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation log(1 + x) for positive x. When x is small (less than \\(10^{-16}\\) for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing \\(f(x) = \\log (1 + x)\\). In Stan (as in C++), this operation is written as log1p(x). Because x itself may be close to zero, the function log1p(x) can take the logarithm of values very close to one, the results of which are close to zero.\nSimilarly, the complementary cumulative distribution functions (CCDF), defined by \\(F^{\\complement}_Y(y) = 1 - F_Y(y)\\), where \\(F_Y\\) is the cumulative distribution function (CDF) for the random variable \\(Y\\). This allows values very close to one to be represented in complementary form.\n\n\n\nAnother downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider \\[\\begin{align*}\n  1&.23456789012345 \\\\\n- 1&.23456789012344 \\\\\n= 0&.00000000000001\n\\end{align*}\\] We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.\nCatastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses Welford’s algorithm for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.\n\n\n\nEven though 1e200 may be represented as a double precision floating point value, there is no finite value large enough to represent 1e200 * 1e200. The result of 1e200 * 1e200 is said to overflow. The IEEE 754 standard requires the result to be positive infinity.\nOverflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.\n\n\n\nWhen there is no number small enough to represent a result, it is said to underflow. For instance, 1e-200 may be represented, but 1e-200 * 1e-200 underflows so that the result is zero.\nUnderflow is a ubiquitous problem in likelihood calculations, For example, if \\(p(y_n \\mid \\theta) &lt; 0.1\\), then \\[\np(y \\mid \\theta) = \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\] will underflow as soon as \\(N &gt; 350\\) or so.\nTo deal with underflow, work on the log scale. Even though \\(p(y \\mid\n\\theta)\\) can’t be represented, there is no problem representing \\[\n\\begin{array}{rcl}\n\\log p(y \\mid \\theta)\n& = & \\log \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\\\[4pt]\n& = & \\sum_{n = 1}^N \\log p(y_n \\mid \\theta)\n\\end{array}\n\\]\nThis is why all of Stan’s probability functions operate on the log scale.\n\n\n\n\nWorking on the log scale, multiplication is converted to addition, \\[\n\\log (a \\cdot b) = \\log a + \\log b.\n\\] Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given \\(\\log a\\) and \\(\\log b\\), how do we get \\(\\log (a + b)\\)? Working out the algebra, \\[\n\\log (a + b)\n=\n\\log (\\exp(\\log a) + \\exp(\\log b)).\n\\]\n\n\nThe nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\log (\\exp(u) + \\exp(v)).\n\\] so that \\[\n\\log (a + b)\n=\n\\textrm{log-sum-exp}(\\log a, \\log b).\n\\]\nAlthough it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\max(u, v) + \\log\\big( \\exp(u - \\max(u, v)) + \\exp(v - \\max(u, v)) \\big).\n\\]\nBecause the terms inside the exponentiations are \\(u - \\max(u, v)\\) and \\(v - \\max(u, v)\\), one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that \\(u \\geq v\\), so that \\[\n\\textrm{log-sum-exp}(u, v) = u + \\log\\big(1 + \\exp(v - u)\\big).\n\\]\nAlthough the inner term may itself be evaluated using the built-in function log1p, there is only limited gain because \\(\\exp(v - u)\\) is only near zero when \\(u\\) is much larger than \\(v\\), meaning the final result is likely to round to \\(u\\) anyway.\nTo conclude, when evaluating \\(\\log (a + b)\\) given \\(\\log a\\) and \\(\\log\nb\\), and assuming \\(\\log a &gt; \\log b\\), return\n\\[\n\\log (a + b) =\n\\log a + \\textrm{log1p}\\big(\\exp(\\log b - \\log a)\\big).\n\\]\n\n\n\nThe log sum of exponentials function may be generalized to sequences in the obvious way, so that if \\(v = v_1, \\ldots, v_N\\), then \\[\\begin{eqnarray*}\n\\textrm{log-sum-exp}(v)\n& = & \\log \\sum_{n = 1}^N \\exp(v_n)\n\\\\[4pt]\n& = & \\max(v) + \\log \\sum_{n = 1}^N \\exp(v_n - \\max(v)).\n\\end{eqnarray*}\\] The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate \\(\\log (u_1 + \\cdots + u_N)\\) given only \\(\\log u_n\\).\n\n\n\nAn immediate application is to computing the mean of a vector \\(u\\) entirely on the log scale. That is, given \\(\\log u\\) and returning \\(\\log \\textrm{mean}(u)\\). \\[\\begin{eqnarray*}\n\\log \\left( \\frac{1}{N} \\sum_{n = 1}^N u_n \\right)\n& = & \\log \\frac{1}{N} + \\log \\sum_{n = 1}^N \\exp(\\log u_n)\n\\\\[4pt]\n& = & -\\log N + \\textrm{log-sum-exp}(\\log u).\n\\end{eqnarray*}\\] where \\(\\log u = (\\log u_1, \\ldots, \\log u_N)\\) is understood elementwise.\n\n\n\n\nBecause floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing x == y, an approximate test may be used given an absolute or relative tolerance.\nGiven a positive absolute tolerance of epsilon, x can be compared to y using the conditional\nabs(x - y) &lt;= epsilon.\nAbsolute tolerances work when the scale of x and y and the relevant comparison is known.\nGiven a positive relative tolerance of epsilon, a typical comparison is\n2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Held-Out Evaluation and Cross-Validation"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/cross-validation.html#cross-validation",
-    "href": "stan-users-guide/cross-validation.html#cross-validation",
-    "title": "Held-Out Evaluation and Cross-Validation",
+    "objectID": "stan-users-guide/floating-point.html#floating-point-representations",
+    "href": "stan-users-guide/floating-point.html#floating-point-representations",
+    "title": "Floating Point Arithmetic",
     "section": "",
-    "text": "Cross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation (Vehtari, Gelman, and Gabry 2017).\nPartitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.\n\n\nFor the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.\nfunctions {\n  array[] int permutation_rng(int N) {\n    array[N] int y;\n    for (n in 1 : N) {\n      y[n] = n;\n    }\n    vector[N] theta = rep_vector(1.0 / N, N);\n    for (n in 1 : size(y)) {\n      int i = categorical_rng(theta);\n      int temp = y[n];\n      y[n] = y[i];\n      y[i] = temp;\n    }\n    return y;\n  }\n}\nThe name of the function must end in _rng because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers 1:N.1\nThe transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=N&gt; N_test;\n}\ntransformed data {\n  int N_train = N - N_test;\n  array[N] int permutation = permutation_rng(N);\n  vector[N_train] x_train = x[permutation[1 : N_train]];\n  vector[N_train] y_train = y[permutation[1 : N_train]];\n  vector[N_test] x_test = x[permutation[N_train + 1 : N]];\n  vector[N_test] y_test = y[permutation[N_train + 1 : N]];\n}\nRecall that in Stan, permutation[1:N_train] is an array of integers, so that x[permutation[1 : N_train]] is a vector defined for i in 1:N_train by\nx[permutation[1 : N_train]][i] = x[permutation[1:N_train][i]]\n                               = x[permutation[i]]\nGiven the test/train split, the rest of the model is straightforward.\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_train ~ normal(alpha + beta * x_train, sigma);\n  { alpha, beta, sigma } ~ normal(0, 1);\n}\ngenerated quantities {\n  vector[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);\n  vector[N] err = y_test_sim - y_hat;\n}\nThe prediction y_test_hat is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, \\[\\begin{eqnarray*}\n\\hat{y}^{\\textrm{test}}\n& = & \\mathbb{E}\\left[ y^{\\textrm{test}} \\mid x^{\\textrm{test}}, x^{\\textrm{train}} y^{\\textrm{train}} \\right]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)}.\n\\end{eqnarray*}\\]\nBecause the test set is constant and the expectation operator is linear, the posterior mean of err as defined in the Stan program will be the error of the posterior mean estimate, \\[\\begin{eqnarray*}\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n  - y^{\\textrm{test}}\n\\\\[4pt]\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)} - y^{\\textrm{test}},\n\\end{eqnarray*}\\] where \\[\n\\hat{y}^{\\textrm{test}(m)}\n\\sim p(y \\mid x^{\\textrm{test}}, x^{\\textrm{train}},\ny^{\\textrm{train}}).\n\\] This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should not be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.\nBecause the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.\n\n\n\nIt is straightforward to declare the variable permutation in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.\n\n\n\nCross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to mixed replication in posterior predictive checking, where there is a choice to simulate new elements of existing groups or generate entirely new groups.\nEducation testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.\n\n\n\nOften data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.\n\n\n\nVehtari, Gelman, and Gabry (2017) introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo (Gabry et al. 2019).",
+    "text": "Stan’s arithmetic is implemented using double-precision arithmetic. The behavior of most1 modern computers follows the floating-point arithmetic, IEEE Standard for Floating-Point Arithmetic (IEEE 754).\n\n\nThe double-precision component of the IEEE 754 standard specifies the representation of real values using a fixed pattern of 64 bits (8 bytes). All values are represented in base two (i.e., binary). The representation is divided into two signed components:\n\nsignificand (53 bits): base value representing significant digits\nexponent (11 bits): power of two multiplied by the base\n\nThe value of a finite floating point number is\n\\[\nv = (-1)^s \\times c \\, 2^q\n\\]\n\n\n\nA normal floating-point value does not use any leading zeros in its significand; subnormal numbers may use leading zeros. Not all I/O systems support subnormal numbers.\n\n\n\nThere are some reserved exponent values so that legal exponent values range between\\(-(2^{10}) + 2 = -1022\\) and \\(2^{10} - 1 = 1023\\). Legal significand values are between \\(-2^{52}\\) and \\(2^{52} - 1\\). Floating point allows the representation of both really big and really small values. Some extreme values are\n\nlargest normal finite number: \\(\\approx 1.8 \\times 10^{308}\\)\nlargest subnormal finite number: \\(\\approx 2.2 \\times 10^{308}\\)\nsmallest positive normal number: \\(\\approx 2.2 \\times 10^{-308}\\)\nsmallest positive subnormal number: \\(\\approx 4.9 \\times 10^{-324}\\)\n\n\n\n\nBecause of the sign bit, there are two ways to represent zero, often called “positive zero” and “negative zero”. This distinction is irrelevant in Stan (as it is in R), because the two values are equal (i.e., 0 == -0 evaluates to true).\n\n\n\nA specially chosen bit pattern is used for the not-a-number value (often written as NaN in programming language output, including Stan’s).\nStan provides a value function not_a_number() that returns this special not-a-number value. It is meant to represent error conditions, not missing values. Usually when not-a-number is an argument to a function, the result will not-a-number if an exception (a rejection in Stan) is not raised.\nStan also provides a test function is_nan(x) that returns 1 if x is not-a-number and 0 otherwise.\nNot-a-number values propagate under almost all mathematical operations. For example, all of the built-in binary arithmetic operations (addition, subtraction, multiplication, division, negation) return not-a-number if any of their arguments are not-a-number. The built-in functions such as log and exp have the same behavior, propagating not-a-number values.\nMost of Stan’s built-in functions will throw exceptions (i.e., reject) when any of their arguments is not-a-number.\nComparisons with not-a-number always return false, up to and including comparison with itself. That is, not_a_number() == not_a_number() somewhat confusingly returns false. That is why there is a built-in is_nan() function in Stan (and in C++). The only exception is negation, which remains coherent. This means not_a_number() != not_a_number() returns true.\nUndefined operations often return not-a-number values. For example, sqrt(-1) will evaluate to not-a-number.\n\n\n\nThere are also two special values representing positive infinity (\\(\\infty)\\) and negative infinity (\\(-\\infty\\)). These are not as pathological as not-a-number, but are often used to represent error conditions such as overflow and underflow. For example, rather than raising an error or returning not-a-number, log(0) evaluates to negative infinity. Exponentiating negative infinity leads back to zero, so that 0 == exp(log(0)). Nevertheless, this should not be done in Stan because the chain rule used to calculate the derivatives will attempt illegal operations and return not-a-number.\nThere are value functions positive_infinity() and negative_infinity() as well as a test function is_inf().\nPositive and negative infinity have the expected comparison behavior, so that negative_infinity() &lt; 0 evaluates to true (represented with 1 in Stan). Also, negating positive infinity leads to negative infinity and vice-versa.\nPositive infinity added to either itself or a finite value produces positive infinity. Negative infinity behaves the same way. However, attempts to subtract positive infinity from itself produce not-a-number, not zero. Similarly, attempts to divide infinite values results in a not-a-number value.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Held-Out Evaluation and Cross-Validation"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/cross-validation.html#footnotes",
-    "href": "stan-users-guide/cross-validation.html#footnotes",
-    "title": "Held-Out Evaluation and Cross-Validation",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe traditional approach is to walk through a vector and replace each item with a random element from the remaining elements, which is guaranteed to only move each item once. This was not done here as it’d require new categorical theta because Stan does not have a uniform discrete RNG built in.↩︎",
+    "objectID": "stan-users-guide/floating-point.html#literals-decimal-and-scientific-notation",
+    "href": "stan-users-guide/floating-point.html#literals-decimal-and-scientific-notation",
+    "title": "Floating Point Arithmetic",
+    "section": "",
+    "text": "In programming languages such as Stan, numbers may be represented in standard decimal (base 10) notation. For example, 2.39 or -1567846.276452. Remember there is no point in writing more than 16 significant digits as they cannot be represented. A number may be coded in Stan using scientific notation, which consists of a signed decimal representation of a base and a signed integer decimal exponent. For example, 36.29e-3 represents the number \\(36.29 \\times\n10^{-3}\\), which is the same number as is represented by 0.03629.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Held-Out Evaluation and Cross-Validation"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/complex-numbers.html",
-    "href": "stan-users-guide/complex-numbers.html",
-    "title": "Complex Numbers",
+    "objectID": "stan-users-guide/floating-point.html#arithmetic-precision",
+    "href": "stan-users-guide/floating-point.html#arithmetic-precision",
+    "title": "Floating Point Arithmetic",
     "section": "",
-    "text": "Stan supports complex scalars, matrices, and vectors as well as real-based ones.\n\n\nThis section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.\n\n\nComplex numbers can be constructed using imaginary literals. For example,\ncomplex z = -1.1 + 2.3i;\nproduces the complex number \\(-1.1 + 2.3i\\). This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the to_complex() function may be used. For example, the following code will work if x and y are parameters, transformed data, or local variables in a function or model block.\nreal x = // ...\nreal y = // ...\ncomplex z = to_complex(x, y);\nThe real and imaginary parts of the complex number can be accessed with getters as follows.\nreal x = get_real(z);  // x = -1.1\nreal y = get_imag(z);  // y = 2.3\nComplex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.\ncomplex a = 3.2 + 2i;\ncomplex b = to_complex(3.2, 2);\nif (a == b) print(\"hello\");\n\n\n\nInteger- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.\ncomplex z1 = 3;  // int promoted to 3 + 0i\ncomplex z2 = 3.2;  // real promoted to 3.2 + 0.0i\n\n\n\nArrays of complex numbers work as usual and allow the usual curly bracket constructors.\ncomplex z1;  complex z2;  complex z3;\n// ...\narray[3] complex zs = { z1, z2, z3 };\nfor (z in zs) {\n  print(z);\n}\nComplex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.\n\n\n\nAll of the standard complex functions are available, including natural logarithm log(z), natural exponentiation exp(z), and powers pow(z1, z2), as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as sin(z), acos(z), tanh(z) and asinh(z).\nPromotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.\ncomplex times_i(complex z) {\n  complex i = to_complex(0, 1);\n  return i * z;\n}\nFor instance, times_i(1) evaluates to the imaginary base \\(i\\), as does times_i(1.0).\n\n\n\n\nThe simplest way to model a distribution over a complex random number \\(z = x + yi\\) is to consider its real part \\(x\\) and imaginary part \\(y\\) to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.\ncomplex z;\nvector[2] mu;\ncholesky_factor_cov[2] L_Sigma;\n// ...\n[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);\nFor example, if z is data, this can be used to estimate mu and the covariance Cholesky factor L_Sigma. Alternatively, if z is a parameter, mu and L_Sigma may constants defining a prior or further parameters defining a hierarchical model.\n\n\n\nStan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.\ncomplex_vector[3] v;\ncomplex_row_vector[2] rv;\ncomplex_matrix[3, 2] m;\nWe can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of rv above, we could assign it to a constructed row vector.\nrv =  [2 + 3i, 1.9 - 2.3i];\nComplex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.\nFurthermore, it is possible to convert back and forth between arrays and matrices using the to_array functions.\n\n\n\nComplex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take x to be predictors, y to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.\ndata {\n  int&lt;lower=0&gt; N;\n  complex_vector[N] x;\n  complex_vector[N] y;\n}\nparameters {\n  complex alpha;\n  complex beta;\n}\nmodel {\n  complex_vector[N] eps = y - (alpha + beta * x);\n  eps ~  // ...error distribution...\n}\nThe question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.\n\n\nThe simplest approach to error in complex regression is to give the real and imaginary parts of eps_n independent independent normal distributions, as follows.\nparameters {\n  // ...\n  vector[2] sigma;\n}\n// ...\nmodel {\n  // ...\n  get_real(eps) ~ normal(0, sigma[1]);\n  get_imag(eps) ~ normal(0, sigma[2]);\n  sigma ~ //...hyperprior...\n}\nA new error scale vector sigma is introduced, and it should itself get a prior based on the expected scale of error for the problem.\n\n\n\nThe next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.\nparameters {\n  cholesky_factor_corr[2] L_Omega;  // correlation matrix\n  vector[2] sigma;                  // real, imag error scales\n  // ...\n}\n// ...\nmodel {\n  array[N] vector[2] eps_arr;\n  for (n in 1:N) {\n    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };\n  }\n  eps_arr ~ multi_normal_cholesky([0, 0]',\n                                  diag_pre_multiply(sigma, L_Omega));\n  L_Omega ~ lkj_cholesky(4);  // shrink toward diagonal correlation\n  sigma ~ // ... hyperprior ...\n}\nHere, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector sigma and a Cholesky factor or a correlation matrix, L_Omega. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to multi_normal_cholesky, the vector of complex numbers needs to be converted to an array of size 2 vectors.",
+    "text": "The choice of significand provides \\(\\log_{10} 2^{53} \\approx 15.95\\) decimal (base 10) digits of arithmetic precision. This is just the precision of the floating-point representation. After several operations are chained together, the realized arithmetic precision is often much lower.\n\n\nIn practice, the finite amount of arithmetic precision leads to rounding, whereby a number is represented by the closest floating-point number. For example, with only 16 decimal digits of accuracy,\n1 + 1e-20 == 1\nThe closest floating point number to \\(1 + 10^{-20}\\) turns out to be \\(1\\) itself. By contrast,\n0 + 1e-20 == 1e-20\nThis highlights the fact that precision depends on scale. Even though 1 + 1e-20 == 1, we have 1e-20 + 1e-20 == 2e-20, as expected.\nRounding also manifests itself in a lack of transitivity. In particular, it does not usually hold for three floating point numbers \\(a, b, c\\) that \\((a + b) + c = a + (b + c)\\).\nIn statistical applications, problems often manifest in situations where users expect the usual rules of real-valued arithmetic to hold. Suppose we have a lower triangular matrix \\(L\\) with strictly positive diagonal, so that it is the Cholesky factor of a positive-definite matrix \\(L \\, L^{\\top}\\). In practice, rounding and loss of precision may render the result \\(L \\, L^{\\top}\\) neither symmetric nor positive definite.\nIn practice, care must be taken to defend against rounding. For example, symmetry may be produced by adding \\(L \\, L^{\\top}\\) with its transpose and dividing by two, or by copying the lower triangular portion into the upper portion. Positive definiteness may be maintained by adding a small quantity to the diagonal.\n\n\n\nThe smallest number greater than zero is roughly \\(0 + 10^{-323}\\). The largest number less than one is roughly \\(1 - 10^{-15.95}\\). The asymmetry is apparent when considering the representation of that largest number smaller than one—the exponent is of no help, and the number is represented as the binary equivalent of \\(0.9999999999999999\\).\nFor this reason, the machine precision is said to be roughly \\(10^{-15.95}\\). This constant is available as machine_precision() in Stan.\n\n\n\nSpecial operations are available to mitigate this problem with numbers rounding when they get close to one. For example, consider the operation log(1 + x) for positive x. When x is small (less than \\(10^{-16}\\) for double-precision floating point), the sum in the argument will round to 1 and the result will round to zero. To allow more granularity, programming languages provide a library function directly implementing \\(f(x) = \\log (1 + x)\\). In Stan (as in C++), this operation is written as log1p(x). Because x itself may be close to zero, the function log1p(x) can take the logarithm of values very close to one, the results of which are close to zero.\nSimilarly, the complementary cumulative distribution functions (CCDF), defined by \\(F^{\\complement}_Y(y) = 1 - F_Y(y)\\), where \\(F_Y\\) is the cumulative distribution function (CDF) for the random variable \\(Y\\). This allows values very close to one to be represented in complementary form.\n\n\n\nAnother downside to floating point representations is that subtraction of two numbers close to each other results in a loss of precision that depends on how close they are. This is easy to see in practice. Consider \\[\\begin{align*}\n  1&.23456789012345 \\\\\n- 1&.23456789012344 \\\\\n= 0&.00000000000001\n\\end{align*}\\] We start with fifteen decimal places of accuracy in the arguments and are left with a single decimal place of accuracy in the result.\nCatastrophic cancellation arises in statistical computations whenever we calculate variance for a distribution with small standard deviations relative to its location. When calculating summary statistics, Stan uses Welford’s algorithm for computing variances. This avoids catastrophic cancellation and may also be carried out in a single pass.\n\n\n\nEven though 1e200 may be represented as a double precision floating point value, there is no finite value large enough to represent 1e200 * 1e200. The result of 1e200 * 1e200 is said to overflow. The IEEE 754 standard requires the result to be positive infinity.\nOverflow is rarely a problem in statistical computations. If it is, it’s possible to work on the log scale, just as for underflow as described below.\n\n\n\nWhen there is no number small enough to represent a result, it is said to underflow. For instance, 1e-200 may be represented, but 1e-200 * 1e-200 underflows so that the result is zero.\nUnderflow is a ubiquitous problem in likelihood calculations, For example, if \\(p(y_n \\mid \\theta) &lt; 0.1\\), then \\[\np(y \\mid \\theta) = \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\] will underflow as soon as \\(N &gt; 350\\) or so.\nTo deal with underflow, work on the log scale. Even though \\(p(y \\mid\n\\theta)\\) can’t be represented, there is no problem representing \\[\n\\begin{array}{rcl}\n\\log p(y \\mid \\theta)\n& = & \\log \\prod_{n=1}^N p(y_n \\mid \\theta)\n\\\\[4pt]\n& = & \\sum_{n = 1}^N \\log p(y_n \\mid \\theta)\n\\end{array}\n\\]\nThis is why all of Stan’s probability functions operate on the log scale.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Complex Numbers"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/complex-numbers.html#working-with-complex-numbers",
-    "href": "stan-users-guide/complex-numbers.html#working-with-complex-numbers",
-    "title": "Complex Numbers",
+    "objectID": "stan-users-guide/floating-point.html#log-sum-of-exponentials",
+    "href": "stan-users-guide/floating-point.html#log-sum-of-exponentials",
+    "title": "Floating Point Arithmetic",
     "section": "",
-    "text": "This section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.\n\n\nComplex numbers can be constructed using imaginary literals. For example,\ncomplex z = -1.1 + 2.3i;\nproduces the complex number \\(-1.1 + 2.3i\\). This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the to_complex() function may be used. For example, the following code will work if x and y are parameters, transformed data, or local variables in a function or model block.\nreal x = // ...\nreal y = // ...\ncomplex z = to_complex(x, y);\nThe real and imaginary parts of the complex number can be accessed with getters as follows.\nreal x = get_real(z);  // x = -1.1\nreal y = get_imag(z);  // y = 2.3\nComplex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.\ncomplex a = 3.2 + 2i;\ncomplex b = to_complex(3.2, 2);\nif (a == b) print(\"hello\");\n\n\n\nInteger- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.\ncomplex z1 = 3;  // int promoted to 3 + 0i\ncomplex z2 = 3.2;  // real promoted to 3.2 + 0.0i\n\n\n\nArrays of complex numbers work as usual and allow the usual curly bracket constructors.\ncomplex z1;  complex z2;  complex z3;\n// ...\narray[3] complex zs = { z1, z2, z3 };\nfor (z in zs) {\n  print(z);\n}\nComplex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.\n\n\n\nAll of the standard complex functions are available, including natural logarithm log(z), natural exponentiation exp(z), and powers pow(z1, z2), as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as sin(z), acos(z), tanh(z) and asinh(z).\nPromotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.\ncomplex times_i(complex z) {\n  complex i = to_complex(0, 1);\n  return i * z;\n}\nFor instance, times_i(1) evaluates to the imaginary base \\(i\\), as does times_i(1.0).",
+    "text": "Working on the log scale, multiplication is converted to addition, \\[\n\\log (a \\cdot b) = \\log a + \\log b.\n\\] Thus sequences of multiplication operations can remain on the log scale. But what about addition? Given \\(\\log a\\) and \\(\\log b\\), how do we get \\(\\log (a + b)\\)? Working out the algebra, \\[\n\\log (a + b)\n=\n\\log (\\exp(\\log a) + \\exp(\\log b)).\n\\]\n\n\nThe nested log of sum of exponentials is so common, it has its own name, “log-sum-exp”, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\log (\\exp(u) + \\exp(v)).\n\\] so that \\[\n\\log (a + b)\n=\n\\textrm{log-sum-exp}(\\log a, \\log b).\n\\]\nAlthough it appears this might overflow as soon as exponentiation is introduced, evaluation does not proceed by evaluating the terms as written. Instead, with a little algebra, the terms are rearranged into a stable form, \\[\n\\textrm{log-sum-exp}(u, v)\n=\n\\max(u, v) + \\log\\big( \\exp(u - \\max(u, v)) + \\exp(v - \\max(u, v)) \\big).\n\\]\nBecause the terms inside the exponentiations are \\(u - \\max(u, v)\\) and \\(v - \\max(u, v)\\), one will be zero and the other will be negative. Because the operation is symmetric, it may be assumed without loss of generality that \\(u \\geq v\\), so that \\[\n\\textrm{log-sum-exp}(u, v) = u + \\log\\big(1 + \\exp(v - u)\\big).\n\\]\nAlthough the inner term may itself be evaluated using the built-in function log1p, there is only limited gain because \\(\\exp(v - u)\\) is only near zero when \\(u\\) is much larger than \\(v\\), meaning the final result is likely to round to \\(u\\) anyway.\nTo conclude, when evaluating \\(\\log (a + b)\\) given \\(\\log a\\) and \\(\\log\nb\\), and assuming \\(\\log a &gt; \\log b\\), return\n\\[\n\\log (a + b) =\n\\log a + \\textrm{log1p}\\big(\\exp(\\log b - \\log a)\\big).\n\\]\n\n\n\nThe log sum of exponentials function may be generalized to sequences in the obvious way, so that if \\(v = v_1, \\ldots, v_N\\), then \\[\\begin{eqnarray*}\n\\textrm{log-sum-exp}(v)\n& = & \\log \\sum_{n = 1}^N \\exp(v_n)\n\\\\[4pt]\n& = & \\max(v) + \\log \\sum_{n = 1}^N \\exp(v_n - \\max(v)).\n\\end{eqnarray*}\\] The exponent cannot overflow because its argument is either zero or negative. This form makes it easy to calculate \\(\\log (u_1 + \\cdots + u_N)\\) given only \\(\\log u_n\\).\n\n\n\nAn immediate application is to computing the mean of a vector \\(u\\) entirely on the log scale. That is, given \\(\\log u\\) and returning \\(\\log \\textrm{mean}(u)\\). \\[\\begin{eqnarray*}\n\\log \\left( \\frac{1}{N} \\sum_{n = 1}^N u_n \\right)\n& = & \\log \\frac{1}{N} + \\log \\sum_{n = 1}^N \\exp(\\log u_n)\n\\\\[4pt]\n& = & -\\log N + \\textrm{log-sum-exp}(\\log u).\n\\end{eqnarray*}\\] where \\(\\log u = (\\log u_1, \\ldots, \\log u_N)\\) is understood elementwise.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Complex Numbers"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/complex-numbers.html#complex-random-variables",
-    "href": "stan-users-guide/complex-numbers.html#complex-random-variables",
-    "title": "Complex Numbers",
+    "objectID": "stan-users-guide/floating-point.html#comparing-floating-point-numbers",
+    "href": "stan-users-guide/floating-point.html#comparing-floating-point-numbers",
+    "title": "Floating Point Arithmetic",
     "section": "",
-    "text": "The simplest way to model a distribution over a complex random number \\(z = x + yi\\) is to consider its real part \\(x\\) and imaginary part \\(y\\) to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.\ncomplex z;\nvector[2] mu;\ncholesky_factor_cov[2] L_Sigma;\n// ...\n[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);\nFor example, if z is data, this can be used to estimate mu and the covariance Cholesky factor L_Sigma. Alternatively, if z is a parameter, mu and L_Sigma may constants defining a prior or further parameters defining a hierarchical model.",
+    "text": "Because floating-point representations are inexact, it is rarely a good idea to test exact inequality. The general recommendation is that rather than testing x == y, an approximate test may be used given an absolute or relative tolerance.\nGiven a positive absolute tolerance of epsilon, x can be compared to y using the conditional\nabs(x - y) &lt;= epsilon.\nAbsolute tolerances work when the scale of x and y and the relevant comparison is known.\nGiven a positive relative tolerance of epsilon, a typical comparison is\n2 * abs(x - y) / (abs(x) + abs(y)) &lt;= epsilon.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Complex Numbers"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/complex-numbers.html#complex-matrices-and-vectors",
-    "href": "stan-users-guide/complex-numbers.html#complex-matrices-and-vectors",
-    "title": "Complex Numbers",
-    "section": "",
-    "text": "Stan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.\ncomplex_vector[3] v;\ncomplex_row_vector[2] rv;\ncomplex_matrix[3, 2] m;\nWe can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of rv above, we could assign it to a constructed row vector.\nrv =  [2 + 3i, 1.9 - 2.3i];\nComplex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.\nFurthermore, it is possible to convert back and forth between arrays and matrices using the to_array functions.",
+    "objectID": "stan-users-guide/floating-point.html#footnotes",
+    "href": "stan-users-guide/floating-point.html#footnotes",
+    "title": "Floating Point Arithmetic",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe notable exception is Intel’s optimizing compilers under certain optimization settings.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Complex Numbers"
+      "Programming Techniques",
+      "Floating Point Arithmetic"
     ]
   },
   {
-    "objectID": "stan-users-guide/complex-numbers.html#complex-linear-regression",
-    "href": "stan-users-guide/complex-numbers.html#complex-linear-regression",
-    "title": "Complex Numbers",
+    "objectID": "stan-users-guide/efficiency-tuning.html",
+    "href": "stan-users-guide/efficiency-tuning.html",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "Complex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take x to be predictors, y to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.\ndata {\n  int&lt;lower=0&gt; N;\n  complex_vector[N] x;\n  complex_vector[N] y;\n}\nparameters {\n  complex alpha;\n  complex beta;\n}\nmodel {\n  complex_vector[N] eps = y - (alpha + beta * x);\n  eps ~  // ...error distribution...\n}\nThe question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.\n\n\nThe simplest approach to error in complex regression is to give the real and imaginary parts of eps_n independent independent normal distributions, as follows.\nparameters {\n  // ...\n  vector[2] sigma;\n}\n// ...\nmodel {\n  // ...\n  get_real(eps) ~ normal(0, sigma[1]);\n  get_imag(eps) ~ normal(0, sigma[2]);\n  sigma ~ //...hyperprior...\n}\nA new error scale vector sigma is introduced, and it should itself get a prior based on the expected scale of error for the problem.\n\n\n\nThe next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.\nparameters {\n  cholesky_factor_corr[2] L_Omega;  // correlation matrix\n  vector[2] sigma;                  // real, imag error scales\n  // ...\n}\n// ...\nmodel {\n  array[N] vector[2] eps_arr;\n  for (n in 1:N) {\n    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };\n  }\n  eps_arr ~ multi_normal_cholesky([0, 0]',\n                                  diag_pre_multiply(sigma, L_Omega));\n  L_Omega ~ lkj_cholesky(4);  // shrink toward diagonal correlation\n  sigma ~ // ... hyperprior ...\n}\nHere, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector sigma and a Cholesky factor or a correlation matrix, L_Omega. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to multi_normal_cholesky, the vector of complex numbers needs to be converted to an array of size 2 vectors.",
+    "text": "This chapter provides a grab bag of techniques for optimizing Stan code, including vectorization, sufficient statistics, and conjugacy. At a coarse level, efficiency involves both the amount of time required for a computation and the amount of memory required. For practical applied statistical modeling, we are mainly concerned with reducing wall time (how long a program takes as measured by a clock on the wall) and keeping memory requirements within available bounds.\n\n\nThe standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).\n\n\n\nStan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.\nWith probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.\nA well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.\n\n\n\nThere is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.\nStatistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.\n\n\n\nBecause Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.\nA second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density \\(\\log\np(\\theta)\\) with respect to the parameter vector \\(\\theta\\), defined as \\[\nH(\\theta) = \\nabla \\, \\nabla \\, \\log p(\\theta \\mid y),\n\\] so that \\[\nH_{i, j}(\\theta) = \\frac{\\partial^2 \\log p(\\theta \\mid y)}\n                {\\partial \\theta_i \\ \\partial \\theta_j}.\n\\] For pure penalized maximum likelihood problems, the posterior log density \\(\\log p(\\theta \\mid y)\\) is replaced by the penalized likelihood function \\(\\mathcal{L}(\\theta) = \\log p(y \\mid \\theta) - \\lambda(\\theta)\\).\n\n\nA good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix \\(H\\), which is the ratio of the largest to the smallest eigenvalue of \\(H\\) (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.\n\n\n\nIdeally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.\n\n\n\nIn all but very simple models (such as multivariate normals), the Hessian will vary as \\(\\theta\\) varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.\nFor all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.\nWith sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.\n\n\n\nImproving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the change of variables chapter.\nOne has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.\n\n\n\n\nModel misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.\nCounterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as normal(0, 1000)). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as normal(mu,   sigma), where mu and sigma are new parameters with their own hyperpriors, can be beneficial.\n\n\n\nStan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ...\n}                               // first validation\nmodel {\n  y ~ multi_normal(mu, Sigma);  // second validation\n  // ...\n}\nBecause Sigma is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that Sigma is positive definite. This test is expensive, having cubic run time (i.e., \\(\\mathcal{O}(N^3)\\) for \\(N \\times N\\) matrices), so it should not be done twice.\nThe test may be avoided by simply declaring Sigma to be a simple unconstrained matrix.\ntransformed parameters {\n  matrix[K, K] Sigma;\n  // ...\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);  // only validation\n}\nNow the only validation is carried out by the multivariate normal.\n\n\n\nStan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.\n\n\nIn this section, we discuss a general transform from a centered to a non-centered parameterization (Papaspiliopoulos, Roberts, and Sköld 2007).1\nThis reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.\nNeal (2003) defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for \\(y \\in\n\\mathbb{R}\\) and \\(x \\in \\mathbb{R}^9\\) with density\n\\[\np(y,x) = \\textsf{normal}(y \\mid 0,3) \\times \\prod_{n=1}^9\n\\textsf{normal}(x_n \\mid 0,\\exp(y/2)).\n\\]\nThe probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to \\(y\\). A plot of the log marginal density of \\(y\\) and the first dimension \\(x_1\\) is shown in the following plot.\nThe funnel can be implemented directly in Stan as follows.\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y/2));\n}\nWhen the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where \\(y\\) is small and thus \\(x\\) is constrained to be near 0. This is due to the fact that the density’s scale changes with \\(y\\), so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.\n\n\n\n\n\n\n\n\n\n\n\n\n\nFigure 1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable \\(y\\) and one lower-level variable \\(x_1\\) (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of \\(x_1\\) and \\(y\\) values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density \\(p(y) = \\textsf{normal}(y \\mid 0,3)\\), which has mean 0 and standard deviation 3.\n\n\n\nIn this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y/2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y/2))\n}\nIn this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.\n\n\n\nSampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.\nThe practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.\nSuppose a random variable of interest \\(X\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), so that \\(X \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\). The variable \\(X\\) has a cumulative distribution function \\(F_X:\\mathbb{R} \\rightarrow (0,1)\\) defined by \\[\nF_X(x) = \\frac{1}{\\pi} \\arctan \\left( \\frac{x - \\mu}{\\tau} \\right) +\n\\frac{1}{2}.\n\\] The inverse of the cumulative distribution function, \\(F_X^{-1}:(0,1) \\rightarrow \\mathbb{R}\\), is thus\n\\[\nF^{-1}_X(y) = \\mu + \\tau \\tan \\left( \\pi \\left( y - \\frac{1}{2} \\right) \\right).\n\\] Thus if the random variable \\(Y\\) has a unit uniform distribution, \\(Y\n\\sim \\textsf{uniform}(0,1)\\), then \\(F^{-1}_X(Y)\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), i.e., \\(F^{-1}_X(Y) \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\).\nConsider a Stan program involving a Cauchy-distributed parameter beta.\nparameters {\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ cauchy(mu, tau);\n  // ...\n}\nThis declaration of beta as a parameter may be replaced with a transformed parameter beta defined in terms of a uniform-distributed parameter beta_unif.\nparameters {\n  real&lt;lower=-pi() / 2, upper=pi() / 2&gt; beta_unif;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = mu + tau * tan(beta_unif);  // beta ~ cauchy(mu, tau)\n}\nmodel {\n  beta_unif ~ uniform(-pi() / 2, pi() / 2);  // not necessary\n  // ...\n}\nIt is more convenient in Stan to transform a uniform variable on \\((-\\pi/2, \\pi/2)\\) than one on \\((0,1)\\). The Cauchy location and scale parameters, mu and tau, may be defined as data or may themselves be parameters. The variable beta could also be defined as a local variable if it does not need to be included in the sampler’s output.\nThe uniform distribution on beta_unif is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because \\(\\log\n\\textsf{uniform}(\\beta_{\\textsf{unif}} \\mid -\\pi/2,\\pi/2) =\n-\\log \\pi\\) is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that beta_unif is between -pi() / 2 and pi() / 2, but this condition is guaranteed by the constraints in the declaration of beta_unif.\n\n\n\nOne thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable \\(\\beta\\), \\[\n\\beta \\sim \\textsf{Student-t}(\\nu, 0, 1),\n\\] by first generating a gamma-distributed precision (inverse variance) \\(\\tau\\) according to \\[\n\\tau \\sim \\textsf{Gamma}(\\nu/2, \\nu/2),\n\\] and then generating \\(\\beta\\) from the normal distribution, \\[\n\\beta \\sim \\textsf{normal}\\left(0,\\tau^{-\\frac{1}{2}}\\right).\n\\]\nBecause \\(\\tau\\) is precision, \\(\\tau^{-\\frac{1}{2}}\\) is the scale (standard deviation), which is the parameterization used by Stan.\nThe marginal distribution of \\(\\beta\\) when you integrate out \\(\\tau\\) is \\(\\textsf{Student-t}(\\nu, 0, 1)\\), i.e., \\[\n\\textsf{Student-t}(\\beta \\mid \\nu, 0, 1)\n=\n\\int_0^{\\infty}\n\\,\n\\textsf{normal}\\left(\\beta \\middle| 0, \\tau^{-0.5}\\right)\n\\times\n\\textsf{Gamma}\\left(\\tau \\middle| \\nu/2, \\nu/2\\right)\n\\\n\\text{d} \\tau.\n\\]\nTo go one step further, instead of defining a \\(\\beta\\) drawn from a normal with precision \\(\\tau\\), define \\(\\alpha\\) to be drawn from a unit normal, \\[\n\\alpha \\sim \\textsf{normal}(0,1)\n\\] and rescale by defining \\[\n\\beta = \\alpha \\, \\tau^{-\\frac{1}{2}}.\n\\]\nNow suppose \\(\\mu = \\beta x\\) is the product of \\(\\beta\\) with a regression predictor \\(x\\). Then the reparameterization \\(\\mu = \\alpha\n\\tau^{-\\frac{1}{2}} x\\) has the same distribution, but in the original, direct parameterization, \\(\\beta\\) has (potentially) heavy tails, whereas in the second, neither \\(\\tau\\) nor \\(\\alpha\\) have heavy tails.\nTo translate into Stan notation, this reparameterization replaces\nparameters {\n  real&lt;lower=0&gt; nu;\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ student_t(nu, 0, 1);\n  // ...\n}\nwith\nparameters {\n  real&lt;lower=0&gt; nu;\n  real&lt;lower=0&gt; tau;\n  real alpha;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = alpha / sqrt(tau);\n  // ...\n}\nmodel {\n  real half_nu;\n  half_nu = 0.5 * nu;\n  tau ~ gamma(half_nu, half_nu);\n  alpha ~ std_normal();\n  // ...\n}\nAlthough set to 0 here, in most cases, the lower bound for the degrees of freedom parameter nu can be set to 1 or higher; when nu is 1, the result is a Cauchy distribution with fat tails and as nu approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter nu characterizes the heaviness of the tails of the model.\n\n\n\nUnfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.\n\n\nFor example, a vectorized hierarchical model might draw a vector of coefficients \\(\\beta\\) with definitions as follows. The so-called centered parameterization is as follows.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nAlthough not shown, a full model will have priors on both mu_beta and sigma_beta along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix x and binary outcome vector y would include a likelihood statement such as form y ~ bernoulli_logit(x * beta), leading to an analytically intractable posterior.\nA hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of beta, mu_beta and sigma_beta are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See Betancourt and Girolami (2013) for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.\n\n\n\n\nSometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see Betancourt and Girolami (2013)), and in more extreme cases will be necessary to achieve convergence.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta_raw;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // implies: beta ~ normal(mu_beta, sigma_beta)\n  beta = mu_beta + sigma_beta * beta_raw;\n}\nmodel {\n  beta_raw ~ std_normal();\n  // ...\n}\nAny priors defined for mu_beta and sigma_beta remain as defined in the original model.\nAlternatively, Stan’s affine transform can be used to decouple sigma and beta:\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;offset=mu_beta, multiplier=sigma_beta&gt;[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nReparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let \\(\\beta = l + s\\alpha\\) where \\(l\\) is a location parameter and \\(s\\) is a scale parameter. The parameter \\(l\\) need not be the mean, \\(s\\) need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If \\(\\alpha\\) and \\(\\beta\\) are from the same distributional family but \\(\\alpha\\) has location zero and unit scale, while \\(\\beta\\) has location \\(l\\) and scale \\(s\\), then that distribution is a location-scale distribution. Thus, if \\(\\alpha\\) were a parameter and \\(\\beta\\) were a transformed parameter, then a prior distribution from the location-scale family on \\(\\alpha\\) with location zero and unit scale implies a prior distribution on \\(\\beta\\) with location \\(l\\) and scale \\(s\\). Doing so would reduce the dependence between \\(\\alpha\\), \\(l\\), and \\(s\\).\nThere are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if \\(\\alpha\\) is distributed standard normal, then \\(\\beta\\) is distributed normal with mean \\(\\mu = l\\) and standard deviation \\(\\sigma = s\\). The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.\nAlso, if \\(z\\) is distributed standard normal, then \\(z^2\\) is distributed chi-squared with one degree of freedom. By summing the squares of \\(K\\) independent standard normal variates, one can obtain a single variate that is distributed chi-squared with \\(K\\) degrees of freedom. However, for large \\(K\\), the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying \\(K\\) primitive parameters just to obtain one transformed parameter to use in a model.\n\n\n\nThe benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for \\(\\beta\\) to be multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). Such a belief is reflected by the following code.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\nparameters {\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ multi_normal(mu, Sigma);\n  // ...\n}\nIn this case mu and Sigma are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of beta.\nIf \\(\\alpha\\) has the same dimensions as \\(\\beta\\) but the elements of \\(\\alpha\\) are independently and identically distributed standard normal such that \\(\\beta = \\mu + L\\alpha\\), where \\(LL^\\top = \\Sigma\\), then \\(\\beta\\) is distributed multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). One choice for \\(L\\) is the Cholesky factor of \\(\\Sigma\\). Thus, the model above could be reparameterized as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Sigma);\n}\nparameters {\n  vector[K] alpha;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  beta = mu + L * alpha;\n}\nmodel {\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu, Sigma)\n  // ...\n}\nThis reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of alpha and second, it avoids the need to invert Sigma every time multi_normal is evaluated.\nThe Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix Tau were fixed data but the vector of standard deviations sigma were unknown parameters, then a reparameterization of beta in terms of alpha could be implemented as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  corr_matrix[K] Tau;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Tau);\n}\nparameters {\n  vector[K] alpha;\n  vector&lt;lower=0&gt;[K] sigma;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // This equals mu + diag_matrix(sigma) * L * alpha;\n  beta = mu + sigma .* (L * alpha);\n}\nmodel {\n  sigma ~ cauchy(0, 5);\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu,\n  //  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))\n  // ...\n}\nThis reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.\nA Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let \\(L\\) be the Cholesky factor of a \\(K \\times K\\) positive definite scale matrix \\(S\\) and let \\(\\nu\\) be the degrees of freedom. If \\[\nA = \\begin{pmatrix}\n\\sqrt{c_{1}} & 0            & \\cdots                & 0 \\\\\nz_{21}       & \\sqrt{c_{2}} & \\ddots                & \\vdots \\\\\n\\vdots       & \\ddots       & \\ddots                & 0 \\\\\nz_{K1}       & \\cdots       & z_{K\\left(K-1\\right)} & \\sqrt{c_{K}}\n\\end{pmatrix},\n\\] where each \\(c_i\\) is distributed chi-squared with \\(\\nu - i + 1\\) degrees of freedom and each \\(z_{ij}\\) is distributed standard normal, then \\(W = LAA^{\\top}L^{\\top}\\) is distributed Wishart with scale matrix \\(S = LL^{\\top}\\) and degrees of freedom \\(\\nu\\). Such a reparameterization can be implemented by the following Stan code:\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  vector[K] mu;\n  matrix[N, K] y;\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  int count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal();\n  // implies: L * A * A' * L' ~ wishart(nu, L * L')\n  y ~ multi_normal_cholesky(mu, L * A);\n  // ...\n}\nThis reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of z and second, it avoids the need to invert the covariance matrix, \\(W\\) every time wishart is evaluated. Third, if \\(W\\) is to be used with a multivariate normal distribution, you can pass \\(L A\\) to the more efficient multi_normal_cholesky function, rather than passing \\(W\\) to multi_normal.\nIf \\(W\\) is distributed Wishart with scale matrix \\(S\\) and degrees of freedom \\(\\nu\\), then \\(W^{-1}\\) is distributed inverse Wishart with inverse scale matrix \\(S^{-1}\\) and degrees of freedom \\(\\nu\\). Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since \\(W = L A A^{\\top} L^{\\top}\\), \\(W^{-1} = L^{{\\top}^{-1}} A^{{\\top}^{-1}} A^{-1} L^{-1}\\), where all four inverses exist, but \\(L^{{-1}^{\\top}} = L^{{\\top}^{-1}}\\) and \\(A^{{-1}^{\\top}} = A^{{\\top}^{-1}}\\). We can slightly modify the above Stan code for this case:\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  // ...\n}\ntransformed data {\n  matrix[K, K] eye;\n  matrix[K, K] L_inv;\n  for (j in 1:K) {\n    for (i in 1:K) {\n      eye[i, j] = 0.0;\n    }\n    eye[j, j] = 1.0;\n  }\n  L_inv = mdivide_left_tri_low(L, eye);\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  matrix[K, K] A_inv_L_inv;\n  int count;\n  count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal(); // implies: crossprod(A_inv_L_inv) ~\n  // inv_wishart(nu, L_inv' * L_inv)\n  // ...\n}\nAnother candidate for reparameterization is the Dirichlet distribution with all \\(K\\) shape parameters equal. Zyczkowski and Sommers (2001) shows that if \\(\\theta_i\\) is equal to the sum of \\(\\beta\\) independent squared standard normal variates and \\(\\rho_i = \\frac{\\theta_i}{\\sum \\theta_i}\\), then the \\(K\\)-vector \\(\\rho\\) is distributed Dirichlet with all shape parameters equal to \\(\\frac{\\beta}{2}\\). In particular, if \\(\\beta = 2\\), then \\(\\rho\\) is uniformly distributed on the unit simplex. Thus, we can make \\(\\rho\\) be a transformed parameter to reduce dependence, as in:\ndata {\n  int&lt;lower=1&gt; beta;\n  // ...\n}\nparameters {\n  array[K] vector[beta] z;\n  // ...\n}\ntransformed parameters {\n  simplex[K] rho;\n  for (k in 1:K) {\n    rho[k] = dot_self(z[k]); // sum-of-squares\n  }\n  rho = rho / sum(rho);\n}\nmodel {\n  for (k in 1:K) {\n    z[k] ~ std_normal();\n  }\n  // implies: rho ~ dirichlet(0.5 * beta * ones)\n  // ...\n}\n\n\n\n\n\n\nStan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.2 This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.\n\n\n\nBecause of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the sum() operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where foo() is some operation that depends on n.\nfor (n in 1:N) {\n  total += foo(n,...);\n}\nThis code has to create intermediate representations for each of the N summands.\nA faster alternative is to copy the values into a vector, then apply the sum() operator, as in the following refactoring.\n{\n  vector[N] summands;\n  for (n in 1:N) {\n    summands[n] = foo(n,...);\n  }\n  total = sum(summands);\n}\nSyntactically, the replacement is a statement block delineated by curly brackets ({, }), starting with the definition of the local variable summands.\nEven though it involves extra work to allocate the summands vector and copy N values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing total within the loop.\n\n\n\nThe following program directly encodes a linear regression with fixed unit noise using a two-dimensional array x of predictors, an array y of outcomes, and an array beta of regression coefficients.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[K, N] real x;\n  array[N] real y;\n}\nparameters {\n  array[K] real beta;\n}\nmodel {\n  for (n in 1:N) {\n    real gamma = 0;\n    for (k in 1:K) {\n      gamma += x[n, k] * beta[k];\n    }\n    y[n] ~ normal(gamma, 1);\n  }\n}\nThe following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[N] vector[K] x;\n  array[N] real y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(dot_product(x[n], beta), 1);\n  }\n}\nAlthough it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in\ny[n] ~ normal(x[n]' * beta, 1);\nThe relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.3\nThe inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.\n// ...\ntransformed parameters {\n  row_vector[K] beta_t;\n  beta_t = beta';\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(beta_t * x[n], 1);\n  }\n}\nThe problem with transposition could be completely solved by directly encoding the x as a row vector, as in the following example.\ndata {\n  // ...\n  array[N] row_vector[K] x;\n  // ...\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, 1);\n  }\n}\nDeclaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.\nHaving said all this, the most efficient way to code this model is with direct matrix multiplication, as in\ndata {\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nIn general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.\n\n\n\nThe final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nThe variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication x * beta in the model block is a vector of the same length as y.\nThe probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, y is a vector of size N, x * beta is a vector of size N, and 1 is a scalar.\n\n\n\nSometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).\nJohn Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu[ii[n]], sigma);\n}\nThe brute force vectorization would build up a mean vector and then vectorize all at once.\n{\n  vector[N] mu_ii;\n  for (n in 1:N) {\n    mu_ii[n] = mu[ii[n]];\n  }\n  y ~ normal(mu_ii, sigma);\n}\nIf there aren’t many levels (values ii[n] can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector y, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have\ny_1 ~ normal(mu[1], sigma);\n// ...\ny_5 ~ normal(mu[5], sigma);\nThis way, both the mu and sigma parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.\n\n\n\n\nIn some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.\n\n\nConsider the following Bernoulli sampling model.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(alpha, beta);\n  for (n in 1:N) {\n    y[n] ~ bernoulli(theta);\n  }\n}\nIn this model, the sum of positive outcomes in y is a sufficient statistic for the chance of success theta. The model may be recoded using the binomial distribution as follows.\ntheta ~ beta(alpha, beta);\nsum(y) ~ binomial(N, theta);\nBecause truth is represented as one and falsehood as zero, the sum sum(y) of a binary vector y is equal to the number of positive outcomes out of a total of N trials.\nThis can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only \\(K\\) possible discrete outcomes, \\(z_1, \\dotsc, z_K\\), but there are \\(N\\) observations, where \\(N\\) is much larger than \\(K\\). If \\(f_k\\) is the frequency of outcome \\(z_k\\), then the entire likelihood with distribution foo can be coded as follows.\nfor (k in 1:K) {\n  target += f[k] * foo_lpmf(z[k] | ...);\n}\nwhere the ellipses are the parameters of the log probability mass function for distribution foo (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).\nThe resulting program looks like a “weighted” regression, but here the weights f[k] are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies f[k] should be counted once in the transformed data block and stored.\nThe same trick works for combining multiple binomial observations.\n\n\n\nConsider the following Stan model for fitting a normal distribution to data.\ndata {\n  int N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nWith the vectorized form used for y, Stan is clever enough to only evaluate log(sigma) once, but it still has to evaluate the normal for all of y[1] to y[N], which involves adding up all the squared differences from the mean and then dividing by sigma squared.\nAn equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.\ndata {\n  int N;\n  vector[N] y;\n}\ntransformed data {\n  real mean_y = mean(y);\n  real&lt;lower=0&gt; var_y = variance(y);\n  real nm1_over2 = 0.5 * (N - 1);\n  real sqrt_N = sqrt(N);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mean_y ~ normal(mu, sigma / sqrt_N);\n  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^2);\n}\nThe data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.\nThe original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on mu and sigma are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.\nThis transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.\n\n\n\nThe Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace\ny ~ poisson(lambda);\nwith\nsum(y) ~ poisson(size(y) * lambda);\nThis will work even if y is a parameter vector because no Jacobian adjustment is required for summation.\n\n\n\n\nIf an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using exp(theta) in multiple places, declare a local variable to store its value and reuse the local variable.\nAnother case that may not be so obvious is with two multilevel parameters, say a[ii[n]] + b[jj[n]]. If a and b are small (i.e., do not have many levels), then a table a_b of their sums can be created, with\nmatrix[size(a), size(b)] a_b;\nfor (i in 1:size(a)) {\n  for (j in 1:size(b)) {\n    a_b[i, j] = a[i] + b[j];\n  }\n}\nThen the sum can be replaced with a_b[ii[n], jj[n]].\n\n\n\nContinuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.\ntheta ~ beta(alpha + sum(y), beta + N - sum(y));\nTo make the model even more efficient, a transformed data variable defined to be sum(y) could be used in the place of sum(y).\n\n\n\nStandardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:\n\nIt helps in faster convergence of MCMC chains.\nIt makes the model less sensitive to the specifics of the parameterization.\nIt aids in the interpretation and comparison of the importance of coefficients across different predictors.\n\nWhen there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.\nSuppose that \\(y = (y_1,\\dotsc,y_N)\\) is a vector of \\(N\\) outcomes and \\(x = (x_1,\\dotsc,x_N)\\) the corresponding vector of \\(N\\) predictors. A simple linear regression involving an intercept coefficient \\(\\alpha\\) and slope coefficient \\(\\beta\\) can be expressed as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n,\n\\] where \\[\n\\epsilon_n \\sim \\textsf{normal}(0,\\sigma).\n\\]\nIf \\(x\\) has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values \\(x_n\\). First the elements of \\(x\\) are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.\nThe mean of \\(x\\) is given by:\n\\[\nmean_x = \\frac{1}{N} \\sum_{n=1}^{N} x_n\n\\]\nThe standard deviation of \\(x\\) is calculated as: \\[\nsd_x = {\\left({\\frac{1}{N} \\sum_{n=1}^{N} (x_n - mean_x)^2}\\right)}^{1/2}\n\\]\nWith these, we compute the \\(z\\), the standardized predictors\n\\[\nz_n = \\frac{x_n - mean_x}{sd_x}\n\\]\nwhere \\(z_n\\) is the standardized value corresponding to \\(x_n\\).\nThe inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.\n\\[\nx_n = z_n sd_x + mean_x\n\\]\nStandardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.\nConsider the following initial model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  // priors\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ normal(0, 5);\n  // likelihood\n  y ~ normal(x * beta + alpha, sigma);\n}\nThe data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\ntransformed data {\n  real mean_x = mean(x);\n  real sd_x = sd(x);\n  vector[N] x_std = (x - mean_x) / sd_x;\n}\nparameters {\n  real alpha_std;\n  real beta_std;\n  real&lt;lower=0&gt; sigma_std;\n}\nmodel {\n  alpha_std ~ normal(0, 10);\n  beta_std ~ normal(0, 10);\n  sigma_std ~ normal(0, 5);\n  y ~ normal(x_std * beta_std + alpha_std, sigma_std);\n}\nThe parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data x_std is defined in terms of variables mean_x and sd_x; by declaring these variables in the transformed data block, they will be available in all following blocks, and therefore can be used in the generated quantities block to record the “natural” parameters alpha and beta.\nThe fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.\nThe original regression \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\] has been transformed to a regression on the standardized data variable \\(z\\),\n\\[\ny_n = \\alpha' + \\beta' z_n + \\epsilon_n.\n\\]\nThe likelihood is specified in terms of the standardized parameters. The original slope \\(\\beta\\) is the standardized slope \\(\\beta'\\) scaled by the inverse of the standard deviation of \\(x\\). The original intercept \\(\\alpha\\) is the intercept from the standardized model \\(\\alpha'\\), corrected for the effect of scaling and centering \\(x\\). Thus, the formulas to retrieve \\(\\alpha\\) and \\(\\beta\\) from \\(\\alpha'\\) and \\(\\beta'\\) are:\n\\[\\begin{align*}\n\\beta = \\frac{\\beta'}{\\sigma_x} \\\\\n\\alpha = \\alpha' - \\beta' \\frac{\\mu_x}{\\sigma_x}\n\\end{align*}\\]\nThese recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,\ngenerated quantities {\n  real beta = beta_std / sd_x;\n  real alpha = alpha_std - beta_std * mean_x / sd_x;\n\n}\nWhen there are multiple real-valued predictors, i.e., when K is the number of predictors, x is an \\(N \\times K\\) matrix, and beta ia \\(K\\)-vector of coefficients, then x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. When \\(K \\ll N\\) the QR reparameterization is recommended for linear and generalized linear models unless there is an informative prior on the location of \\(\\beta\\).\n\n\nFor many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use\ny ~ std_normal();\nthan to use\ny ~ normal(0, 1);\nbecause the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.\n\n\n\n\nThe map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the map-reduce chapter for more information on implementing map-reduce operations.\nMap-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Complex Numbers"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "stan-users-guide/bootstrap.html",
-    "href": "stan-users-guide/bootstrap.html",
-    "title": "The Bootstrap and Bagging",
+    "objectID": "stan-users-guide/efficiency-tuning.html#what-is-efficiency",
+    "href": "stan-users-guide/efficiency-tuning.html#what-is-efficiency",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "The bootstrap is a technique for approximately sampling from the error distribution for an estimator. Thus it can be used as a Monte Carlo method to estimate standard errors and confidence intervals for point estimates (Efron and Tibshirani 1986; 1994). It works by subsampling the original data and computing sample estimates from the subsample. Like other Monte Carlo methods, the bootstrap is plug-and-play, allowing great flexibility in both model choice and estimator.\nBagging is a technique for combining bootstrapped estimators for model criticism and more robust inference (Breiman 1996; Huggins and Miller 2019).\n\n\n\n\nAn estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set \\(y_{1,\\ldots, N}\\) to a number by \\[\n\\textrm{mean}(y) = \\frac{1}{N} \\sum_{n=1}^N y_n,\n\\] and hence meets the definition of an estimator. Given the likelihood function \\[\np(y \\mid \\mu) = \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, 1),\n\\] the mean is the maximum likelihood estimator,\n\\[\n\\textrm{mean}(y) = \\textrm{arg max}_{\\mu} \\ p(y \\mid \\mu, 1)\n\\] A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.\nIn analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.\n\n\n\nThe bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.\nThe following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate \\(\\hat{\\theta}\\) that is a function of data \\(y\\).\nfor (m in 1:M) {\n  y_rep[m] &lt;- sample_uniform(y)\n  theta_hat[m] &lt;- estimate_theta(y_rep[m])\n}\nstd_error = sd(theta_hat)\nconf_95pct = [ quantile(theta_hat, 0.025),\n               quantile(theta_hat, 0.975) ]\nThe sample_uniform function works by independently assigning each element of y_rep an element of y drawn uniformly at random. This produces a sample with replacement. That is, some elements of y may show up more than once in y_rep and some may not appear at all.\n\n\n\n\nThe bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector x of predictors in addition to outcomes y. To allow a single program to fit both the original data and random subsamples, the variable resample is set to 1 to resample and 0 to use the original data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=1&gt; resample;\n}\ntransformed data {\n  simplex[N] uniform = rep_vector(1.0 / N, N);\n  array[N] int&lt;lower=1, upper=N&gt; boot_idxs;\n  for (n in 1:N) {\n    boot_idxs[n] = resample ? categorical_rng(uniform) : n;\n  }\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);\n}\nThe model accepts data in the usual form for a linear regression as a number of observations \\(N\\) with a size \\(N\\) vector \\(x\\) of predictors and a size \\(N\\) vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of boot_idxs from 1:N, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array boot_idxs is defined to be the sequence 1:N, because x == x[1:N] and y = y[1:N].\nFor example, when resample == 1, if \\(N = 4,\\) the value of boot_idxs might be {2, 1, 1, 3}, resulting in a bootstrap sample {y[2], y[1], y[1], y[3]} with the first element repeated twice and the fourth element not sampled at all.\nThe parameters are the usual regression coefficients for the intercept alpha, slope beta, and error scale sigma. The model uses the bootstrap index variable boot_idx to index the predictors as x[boot_idx] and outcomes as y[boot_idx]. This generates a new size-\\(N\\) vector whose entries are defined by x[boot_idx][n] = x[boot_idx[n]] and similarly for y. For example, if \\(N = 4\\) and boot_idxs = {2, 1, 1, 3}, then x[boot_idxs] = [x[2], x[1], x[1], x[3]]' and y[boot_idxs] = [y[2], y[1], y[1], y[3]]'. The predictor and outcome vectors remain aligned, with both elements of the pair x[1] and y[1] repeated twice.\nWith the model defined this way, if resample is 1, the model is fit to a bootstrap subsample of the data. If resample is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.\n\n\n\nRunning the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.\nTo estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data \\(y\\), it is usually not worth calculating bootstrap estimates to high precision.\n\n\nHere’s the result of calculating standard errors for the linear regression model above with \\(N = 50\\) data points, \\(\\alpha = 1.2, \\beta\n= -0.5,\\) and \\(\\sigma = 1.5.\\) With a total of \\(M = 100\\) bootstrap samples, there are 100 estimates of \\(\\alpha\\), 100 of \\(\\beta\\), and 100 of \\(\\sigma\\). These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for \\(\\alpha\\) provide the bootstrap estimate of the standard error in the estimate for \\(\\alpha\\). Here’s what it looks like for the above model with \\(M =\n100\\)\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.218\n      beta     -0.610      0.204\n     sigma      1.537      0.142\nWith the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.206\n     alpha      1.359      0.240\n     alpha      1.359      0.234\n     alpha      1.359      0.249\n     alpha      1.359      0.227\nIncreasing \\(M\\) will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample \\(y\\).\n\n\n\nAs usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are \\(M = 1000\\) estimates of \\(\\hat{\\alpha}\\) in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set \\(y\\). To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).\n\n\n\n\nWhen bootstrapping is carried through inference it is known as bootstrap aggregation, or bagging, in the machine-learning literature (Breiman 1996). In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate \\(\\hat{\\sigma}\\) from the original data set, bootstrapped data sets \\(y^{\\textrm{boot}(1)}, \\ldots,\ny^{\\textrm{boot}(N)}\\) are generated. Each is used to generate an estimate \\(\\hat{\\sigma}^{\\textrm{boot}(n)}.\\) The final estimate is \\[\n\\hat{\\sigma} = \\frac{1}{N} \\sum_{n = 1}^N \\hat{\\sigma}^{\\textrm{boot}(n)}.\n\\] The same would be done to estimate a predictive quantity \\(\\tilde{y}\\) for as yet unseen data. \\[\n\\hat{\\tilde{y}} = \\frac{1}{N} \\sum_{n = 1}^N\n\\hat{\\tilde{y}}^{\\textrm{boot}(n)}.\n\\] For discrete parameters, voting is used to select the outcome.\nOne way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.\n\n\n\nA Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.\n(Huggins and Miller 2019) use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.",
+    "text": "The standard algorithm analyses in computer science measure efficiency asymptotically as a function of problem size (such as data, number of parameters, etc.) and typically do not consider constant additive factors like startup times or multiplicative factors like speed of operations. In practice, the constant factors are important; if run time can be cut in half or more, that’s a huge gain. This chapter focuses on both the constant factors involved in efficiency (such as using built-in matrix operations as opposed to naive loops) and on asymptotic efficiency factors (such as using linear algorithms instead of quadratic algorithms in loops).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "The Bootstrap and Bagging"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "stan-users-guide/bootstrap.html#the-bootstrap",
-    "href": "stan-users-guide/bootstrap.html#the-bootstrap",
-    "title": "The Bootstrap and Bagging",
+    "objectID": "stan-users-guide/efficiency-tuning.html#efficiency-for-probabilistic-models-and-algorithms",
+    "href": "stan-users-guide/efficiency-tuning.html#efficiency-for-probabilistic-models-and-algorithms",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "An estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set \\(y_{1,\\ldots, N}\\) to a number by \\[\n\\textrm{mean}(y) = \\frac{1}{N} \\sum_{n=1}^N y_n,\n\\] and hence meets the definition of an estimator. Given the likelihood function \\[\np(y \\mid \\mu) = \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, 1),\n\\] the mean is the maximum likelihood estimator,\n\\[\n\\textrm{mean}(y) = \\textrm{arg max}_{\\mu} \\ p(y \\mid \\mu, 1)\n\\] A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.\nIn analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.\n\n\n\nThe bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.\nThe following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate \\(\\hat{\\theta}\\) that is a function of data \\(y\\).\nfor (m in 1:M) {\n  y_rep[m] &lt;- sample_uniform(y)\n  theta_hat[m] &lt;- estimate_theta(y_rep[m])\n}\nstd_error = sd(theta_hat)\nconf_95pct = [ quantile(theta_hat, 0.025),\n               quantile(theta_hat, 0.975) ]\nThe sample_uniform function works by independently assigning each element of y_rep an element of y drawn uniformly at random. This produces a sample with replacement. That is, some elements of y may show up more than once in y_rep and some may not appear at all.",
+    "text": "Stan programs express models which are intrinsically statistical in nature. The algorithms applied to these models may or may not themselves be probabilistic. For example, given an initial value for parameters (which may itself be given deterministically or generated randomly), Stan’s optimization algorithm (L-BFGS) for penalized maximum likelihood estimation is purely deterministic. Stan’s sampling algorithms are based on Markov chain Monte Carlo algorithms, which are probabilistic by nature at every step. Stan’s variational inference algorithm (ADVI) is probabilistic despite being an optimization algorithm; the randomization lies in a nested Monte Carlo calculation for an expected gradient.\nWith probabilistic algorithms, there will be variation in run times (and maybe memory usage) based on the randomization involved. For example, by starting too far out in the tail, iterative algorithms underneath the hood, such as the solvers for ordinary differential equations, may take different numbers of steps. Ideally this variation will be limited; when there is a lot of variation it can be a sign that there is a problem with the model’s parameterization in a Stan program or with initialization.\nA well-behaved Stan program will have low variance between runs with different random initializations and differently seeded random number generators. But sometimes an algorithm can get stuck in one part of the posterior, typically due to high curvature. Such sticking almost always indicates the need to reparameterize the model. Just throwing away Markov chains with apparently poor behavior (slow, or stuck) can lead to bias in posterior estimates. This problem with getting stuck can often be overcome by lowering the initial step size to avoid getting stuck during adaptation and increasing the target acceptance rate in order to target a lower step size. This is because smaller step sizes allow Stan’s gradient-based algorithms to better follow the curvature in the density or penalized maximum likelihood being fit.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "The Bootstrap and Bagging"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "stan-users-guide/bootstrap.html#coding-the-bootstrap-in-stan",
-    "href": "stan-users-guide/bootstrap.html#coding-the-bootstrap-in-stan",
-    "title": "The Bootstrap and Bagging",
+    "objectID": "stan-users-guide/efficiency-tuning.html#statistical-vs.-computational-efficiency",
+    "href": "stan-users-guide/efficiency-tuning.html#statistical-vs.-computational-efficiency",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "The bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector x of predictors in addition to outcomes y. To allow a single program to fit both the original data and random subsamples, the variable resample is set to 1 to resample and 0 to use the original data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=1&gt; resample;\n}\ntransformed data {\n  simplex[N] uniform = rep_vector(1.0 / N, N);\n  array[N] int&lt;lower=1, upper=N&gt; boot_idxs;\n  for (n in 1:N) {\n    boot_idxs[n] = resample ? categorical_rng(uniform) : n;\n  }\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);\n}\nThe model accepts data in the usual form for a linear regression as a number of observations \\(N\\) with a size \\(N\\) vector \\(x\\) of predictors and a size \\(N\\) vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of boot_idxs from 1:N, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array boot_idxs is defined to be the sequence 1:N, because x == x[1:N] and y = y[1:N].\nFor example, when resample == 1, if \\(N = 4,\\) the value of boot_idxs might be {2, 1, 1, 3}, resulting in a bootstrap sample {y[2], y[1], y[1], y[3]} with the first element repeated twice and the fourth element not sampled at all.\nThe parameters are the usual regression coefficients for the intercept alpha, slope beta, and error scale sigma. The model uses the bootstrap index variable boot_idx to index the predictors as x[boot_idx] and outcomes as y[boot_idx]. This generates a new size-\\(N\\) vector whose entries are defined by x[boot_idx][n] = x[boot_idx[n]] and similarly for y. For example, if \\(N = 4\\) and boot_idxs = {2, 1, 1, 3}, then x[boot_idxs] = [x[2], x[1], x[1], x[3]]' and y[boot_idxs] = [y[2], y[1], y[1], y[3]]'. The predictor and outcome vectors remain aligned, with both elements of the pair x[1] and y[1] repeated twice.\nWith the model defined this way, if resample is 1, the model is fit to a bootstrap subsample of the data. If resample is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.",
+    "text": "There is a difference between pure computational efficiency and statistical efficiency for Stan programs fit with sampling-based algorithms. Computational efficiency measures the amount of time or memory required for a given step in a calculation, such as an evaluation of a log posterior or penalized likelihood.\nStatistical efficiency typically involves requiring fewer steps in algorithms by making the statistical formulation of a model better behaved. The typical way to do this is by applying a change of variables (i.e., reparameterization) so that sampling algorithms mix better or optimization algorithms require less adaptation.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "The Bootstrap and Bagging"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "stan-users-guide/bootstrap.html#error-statistics-from-the-bootstrap",
-    "href": "stan-users-guide/bootstrap.html#error-statistics-from-the-bootstrap",
-    "title": "The Bootstrap and Bagging",
+    "objectID": "stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature",
+    "href": "stan-users-guide/efficiency-tuning.html#model-conditioning-and-curvature",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "Running the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.\nTo estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data \\(y\\), it is usually not worth calculating bootstrap estimates to high precision.\n\n\nHere’s the result of calculating standard errors for the linear regression model above with \\(N = 50\\) data points, \\(\\alpha = 1.2, \\beta\n= -0.5,\\) and \\(\\sigma = 1.5.\\) With a total of \\(M = 100\\) bootstrap samples, there are 100 estimates of \\(\\alpha\\), 100 of \\(\\beta\\), and 100 of \\(\\sigma\\). These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for \\(\\alpha\\) provide the bootstrap estimate of the standard error in the estimate for \\(\\alpha\\). Here’s what it looks like for the above model with \\(M =\n100\\)\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.218\n      beta     -0.610      0.204\n     sigma      1.537      0.142\nWith the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.206\n     alpha      1.359      0.240\n     alpha      1.359      0.234\n     alpha      1.359      0.249\n     alpha      1.359      0.227\nIncreasing \\(M\\) will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample \\(y\\).\n\n\n\nAs usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are \\(M = 1000\\) estimates of \\(\\hat{\\alpha}\\) in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set \\(y\\). To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).",
+    "text": "Because Stan’s algorithms rely on step-based gradient-based approximations of the density (or penalized maximum likelihood) being fitted, posterior curvature not captured by this first-order approximation plays a central role in determining the statistical efficiency of Stan’s algorithms.\nA second-order approximation to curvature is provided by the Hessian, the matrix of second derivatives of the log density \\(\\log\np(\\theta)\\) with respect to the parameter vector \\(\\theta\\), defined as \\[\nH(\\theta) = \\nabla \\, \\nabla \\, \\log p(\\theta \\mid y),\n\\] so that \\[\nH_{i, j}(\\theta) = \\frac{\\partial^2 \\log p(\\theta \\mid y)}\n                {\\partial \\theta_i \\ \\partial \\theta_j}.\n\\] For pure penalized maximum likelihood problems, the posterior log density \\(\\log p(\\theta \\mid y)\\) is replaced by the penalized likelihood function \\(\\mathcal{L}(\\theta) = \\log p(y \\mid \\theta) - \\lambda(\\theta)\\).\n\n\nA good gauge of how difficult a problem the curvature presents is given by the condition number of the Hessian matrix \\(H\\), which is the ratio of the largest to the smallest eigenvalue of \\(H\\) (assuming the Hessian is positive definite). This essentially measures the difference between the flattest direction of movement and the most curved. Typically, the step size of a gradient-based algorithm is bounded by the most sharply curved direction. With better conditioned log densities or penalized likelihood functions, it is easier for Stan’s adaptation, especially the diagonal adaptations that are used as defaults.\n\n\n\nIdeally, all parameters should be programmed so that they have unit scale and so that posterior correlation is reduced; together, these properties mean that there is no rotation or scaling required for optimal performance of Stan’s algorithms. For Hamiltonian Monte Carlo, this implies a unit mass matrix, which requires no adaptation as it is where the algorithm initializes.\n\n\n\nIn all but very simple models (such as multivariate normals), the Hessian will vary as \\(\\theta\\) varies (an extreme example is Neal’s funnel, as naturally arises in hierarchical models with little or no data). The more the curvature varies, the harder it is for all of the algorithms with fixed adaptation parameters to find adaptations that cover the entire density well. Many of the variable transforms proposed are aimed at improving the conditioning of the Hessian and/or making it more consistent across the relevant portions of the density (or penalized maximum likelihood function) being fit.\nFor all of Stan’s algorithms, the curvature along the path from the initial values of the parameters to the solution is relevant. For penalized maximum likelihood and variational inference, the solution of the iterative algorithm will be a single point, so this is all that matters. For sampling, the relevant “solution” is the typical set, which is the posterior volume where almost all draws from the posterior lies; thus, the typical set contains almost all of the posterior probability mass.\nWith sampling, the curvature may vary dramatically between the points on the path from the initialization point to the typical set and within the typical set. This is why adaptation needs to run long enough to visit enough points in the typical set to get a good first-order estimate of the curvature within the typical set. If adaptation is not run long enough, sampling within the typical set after adaptation will not be efficient. We generally recommend at least one hundred iterations after the typical set is reached (and the first effective draw is ready to be realized). Whether adaptation has run long enough can be measured by comparing the adaptation parameters derived from a set of diffuse initial parameter values.\n\n\n\nImproving statistical efficiency is achieved by reparameterizing the model so that the same result may be calculated using a density or penalized maximum likelihood that is better conditioned. Again, see the example of reparameterizing Neal’s funnel for an example, and also the examples in the change of variables chapter.\nOne has to be careful in using change-of-variables reparameterizations when using maximum likelihood estimation, because they can change the result if the Jacobian term is inadvertently included in the revised likelihood model.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "The Bootstrap and Bagging"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "stan-users-guide/bootstrap.html#bagging",
-    "href": "stan-users-guide/bootstrap.html#bagging",
-    "title": "The Bootstrap and Bagging",
+    "objectID": "stan-users-guide/efficiency-tuning.html#well-specified-models",
+    "href": "stan-users-guide/efficiency-tuning.html#well-specified-models",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "When bootstrapping is carried through inference it is known as bootstrap aggregation, or bagging, in the machine-learning literature (Breiman 1996). In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate \\(\\hat{\\sigma}\\) from the original data set, bootstrapped data sets \\(y^{\\textrm{boot}(1)}, \\ldots,\ny^{\\textrm{boot}(N)}\\) are generated. Each is used to generate an estimate \\(\\hat{\\sigma}^{\\textrm{boot}(n)}.\\) The final estimate is \\[\n\\hat{\\sigma} = \\frac{1}{N} \\sum_{n = 1}^N \\hat{\\sigma}^{\\textrm{boot}(n)}.\n\\] The same would be done to estimate a predictive quantity \\(\\tilde{y}\\) for as yet unseen data. \\[\n\\hat{\\tilde{y}} = \\frac{1}{N} \\sum_{n = 1}^N\n\\hat{\\tilde{y}}^{\\textrm{boot}(n)}.\n\\] For discrete parameters, voting is used to select the outcome.\nOne way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.",
+    "text": "Model misspecification, which roughly speaking means using a model that doesn’t match the data, can be a major source of slow code. This can be seen in cases where simulated data according to the model runs robustly and efficiently, whereas the real data for which it was intended runs slowly or may even have convergence and mixing issues. While some of the techniques recommended in the remaining sections of this chapter may mitigate the problem, the best remedy is a better model specification.\nCounterintuitively, more complicated models often run faster than simpler models. One common pattern is with a group of parameters with a wide fixed prior such as normal(0, 1000)). This can fit slowly due to the mismatch between prior and posterior (the prior has support for values in the hundreds or even thousands, whereas the posterior may be concentrated near zero). In such cases, replacing the fixed prior with a hierarchical prior such as normal(mu,   sigma), where mu and sigma are new parameters with their own hyperpriors, can be beneficial.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "The Bootstrap and Bagging"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "stan-users-guide/bootstrap.html#bayesian-bootstrap-and-bagging",
-    "href": "stan-users-guide/bootstrap.html#bayesian-bootstrap-and-bagging",
-    "title": "The Bootstrap and Bagging",
+    "objectID": "stan-users-guide/efficiency-tuning.html#avoiding-validation",
+    "href": "stan-users-guide/efficiency-tuning.html#avoiding-validation",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "A Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.\n(Huggins and Miller 2019) use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.",
+    "text": "Stan validates all of its data structure constraints. For example, consider a transformed parameter defined to be a covariance matrix and then used as a covariance parameter in the model block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ...\n}                               // first validation\nmodel {\n  y ~ multi_normal(mu, Sigma);  // second validation\n  // ...\n}\nBecause Sigma is declared to be a covariance matrix, it will be factored at the end of the transformed parameter block to ensure that it is positive definite. The multivariate normal log density function also validates that Sigma is positive definite. This test is expensive, having cubic run time (i.e., \\(\\mathcal{O}(N^3)\\) for \\(N \\times N\\) matrices), so it should not be done twice.\nThe test may be avoided by simply declaring Sigma to be a simple unconstrained matrix.\ntransformed parameters {\n  matrix[K, K] Sigma;\n  // ...\n}\nmodel {\n  y ~ multi_normal(mu, Sigma);  // only validation\n}\nNow the only validation is carried out by the multivariate normal.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "The Bootstrap and Bagging"
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/whitespace.html",
-    "href": "reference-manual/whitespace.html",
-    "title": "Whitespace",
+    "objectID": "stan-users-guide/efficiency-tuning.html#reparameterization.section",
+    "href": "stan-users-guide/efficiency-tuning.html#reparameterization.section",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).\n\n\n\nStan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.\nOther than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.\n\n\n\nZero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as a * b, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.\nIdentifiers and literals may not be separated by whitespace. Thus it is not legal to write the number 10000 as 10 000 or to write the identifier normal_lpdf as normal _ lpdf.",
+    "text": "Stan’s sampler can be slow in sampling from distributions with difficult posterior geometries. One way to speed up such models is through reparameterization. In some cases, reparameterization can dramatically increase effective sample size for the same number of iterations or even make programs that would not converge well behaved.\n\n\nIn this section, we discuss a general transform from a centered to a non-centered parameterization (Papaspiliopoulos, Roberts, and Sköld 2007).1\nThis reparameterization is helpful when there is not much data, because it separates the hierarchical parameters and lower-level parameters in the prior.\nNeal (2003) defines a distribution that exemplifies the difficulties of sampling from some hierarchical models. Neal’s example is fairly extreme, but can be trivially reparameterized in such a way as to make sampling straightforward. Neal’s example has support for \\(y \\in\n\\mathbb{R}\\) and \\(x \\in \\mathbb{R}^9\\) with density\n\\[\np(y,x) = \\textsf{normal}(y \\mid 0,3) \\times \\prod_{n=1}^9\n\\textsf{normal}(x_n \\mid 0,\\exp(y/2)).\n\\]\nThe probability contours are shaped like ten-dimensional funnels. The funnel’s neck is particularly sharp because of the exponential function applied to \\(y\\). A plot of the log marginal density of \\(y\\) and the first dimension \\(x_1\\) is shown in the following plot.\nThe funnel can be implemented directly in Stan as follows.\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y/2));\n}\nWhen the model is expressed this way, Stan has trouble sampling from the neck of the funnel, where \\(y\\) is small and thus \\(x\\) is constrained to be near 0. This is due to the fact that the density’s scale changes with \\(y\\), so that a step size that works well in the body will be too large for the neck, and a step size that works in the neck will be inefficient in the body. This can be seen in the following plots.\n\n\n\n\n\n\n\n\n\n\n\n\n\nFigure 1: Neal’s funnel. (Left) The marginal density of Neal’s funnel for the upper-level variable \\(y\\) and one lower-level variable \\(x_1\\) (see the text for the formula). The blue region has log density greater than -8, the yellow region density greater than -16, and the gray background a density less than -16. (Right) 4000 draws are taken from a run of Stan’s sampler with default settings. Both plots are restricted to the shown window of \\(x_1\\) and \\(y\\) values; some draws fell outside of the displayed area as would be expected given the density. The draws are consistent with the marginal density \\(p(y) = \\textsf{normal}(y \\mid 0,3)\\), which has mean 0 and standard deviation 3.\n\n\n\nIn this particular instance, because the analytic form of the density is known, the model can be converted to the following more efficient form.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y/2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y/2))\n}\nIn this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan. These are then transformed into draws from the funnel. In this case, the same transform may be used to define Monte Carlo directly based on independent standard normal draws; Markov chain Monte Carlo methods are not necessary. If such a reparameterization were used in Stan code, it is useful to provide a comment indicating what the distribution for the parameter implies for the distribution of the transformed parameter.\n\n\n\nSampling from heavy tailed distributions such as the Cauchy is difficult for Hamiltonian Monte Carlo, which operates within a Euclidean geometry.\nThe practical problem is that tail of the Cauchy requires a relatively large step size compared to the trunk. With a small step size, the No-U-Turn sampler requires many steps when starting in the tail of the distribution; with a large step size, there will be too much rejection in the central portion of the distribution. This problem may be mitigated by defining the Cauchy-distributed variable as the transform of a uniformly distributed variable using the Cauchy inverse cumulative distribution function.\nSuppose a random variable of interest \\(X\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), so that \\(X \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\). The variable \\(X\\) has a cumulative distribution function \\(F_X:\\mathbb{R} \\rightarrow (0,1)\\) defined by \\[\nF_X(x) = \\frac{1}{\\pi} \\arctan \\left( \\frac{x - \\mu}{\\tau} \\right) +\n\\frac{1}{2}.\n\\] The inverse of the cumulative distribution function, \\(F_X^{-1}:(0,1) \\rightarrow \\mathbb{R}\\), is thus\n\\[\nF^{-1}_X(y) = \\mu + \\tau \\tan \\left( \\pi \\left( y - \\frac{1}{2} \\right) \\right).\n\\] Thus if the random variable \\(Y\\) has a unit uniform distribution, \\(Y\n\\sim \\textsf{uniform}(0,1)\\), then \\(F^{-1}_X(Y)\\) has a Cauchy distribution with location \\(\\mu\\) and scale \\(\\tau\\), i.e., \\(F^{-1}_X(Y) \\sim\n\\textsf{Cauchy}(\\mu,\\tau)\\).\nConsider a Stan program involving a Cauchy-distributed parameter beta.\nparameters {\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ cauchy(mu, tau);\n  // ...\n}\nThis declaration of beta as a parameter may be replaced with a transformed parameter beta defined in terms of a uniform-distributed parameter beta_unif.\nparameters {\n  real&lt;lower=-pi() / 2, upper=pi() / 2&gt; beta_unif;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = mu + tau * tan(beta_unif);  // beta ~ cauchy(mu, tau)\n}\nmodel {\n  beta_unif ~ uniform(-pi() / 2, pi() / 2);  // not necessary\n  // ...\n}\nIt is more convenient in Stan to transform a uniform variable on \\((-\\pi/2, \\pi/2)\\) than one on \\((0,1)\\). The Cauchy location and scale parameters, mu and tau, may be defined as data or may themselves be parameters. The variable beta could also be defined as a local variable if it does not need to be included in the sampler’s output.\nThe uniform distribution on beta_unif is defined explicitly in the model block, but it could be safely removed from the program without changing sampling behavior. This is because \\(\\log\n\\textsf{uniform}(\\beta_{\\textsf{unif}} \\mid -\\pi/2,\\pi/2) =\n-\\log \\pi\\) is a constant and Stan only needs the total log probability up to an additive constant. Stan will spend some time checking that that beta_unif is between -pi() / 2 and pi() / 2, but this condition is guaranteed by the constraints in the declaration of beta_unif.\n\n\n\nOne thing that sometimes works when you’re having trouble with the heavy-tailedness of Student-t distributions is to use the gamma-mixture representation, which says that you can generate a Student-t distributed variable \\(\\beta\\), \\[\n\\beta \\sim \\textsf{Student-t}(\\nu, 0, 1),\n\\] by first generating a gamma-distributed precision (inverse variance) \\(\\tau\\) according to \\[\n\\tau \\sim \\textsf{Gamma}(\\nu/2, \\nu/2),\n\\] and then generating \\(\\beta\\) from the normal distribution, \\[\n\\beta \\sim \\textsf{normal}\\left(0,\\tau^{-\\frac{1}{2}}\\right).\n\\]\nBecause \\(\\tau\\) is precision, \\(\\tau^{-\\frac{1}{2}}\\) is the scale (standard deviation), which is the parameterization used by Stan.\nThe marginal distribution of \\(\\beta\\) when you integrate out \\(\\tau\\) is \\(\\textsf{Student-t}(\\nu, 0, 1)\\), i.e., \\[\n\\textsf{Student-t}(\\beta \\mid \\nu, 0, 1)\n=\n\\int_0^{\\infty}\n\\,\n\\textsf{normal}\\left(\\beta \\middle| 0, \\tau^{-0.5}\\right)\n\\times\n\\textsf{Gamma}\\left(\\tau \\middle| \\nu/2, \\nu/2\\right)\n\\\n\\text{d} \\tau.\n\\]\nTo go one step further, instead of defining a \\(\\beta\\) drawn from a normal with precision \\(\\tau\\), define \\(\\alpha\\) to be drawn from a unit normal, \\[\n\\alpha \\sim \\textsf{normal}(0,1)\n\\] and rescale by defining \\[\n\\beta = \\alpha \\, \\tau^{-\\frac{1}{2}}.\n\\]\nNow suppose \\(\\mu = \\beta x\\) is the product of \\(\\beta\\) with a regression predictor \\(x\\). Then the reparameterization \\(\\mu = \\alpha\n\\tau^{-\\frac{1}{2}} x\\) has the same distribution, but in the original, direct parameterization, \\(\\beta\\) has (potentially) heavy tails, whereas in the second, neither \\(\\tau\\) nor \\(\\alpha\\) have heavy tails.\nTo translate into Stan notation, this reparameterization replaces\nparameters {\n  real&lt;lower=0&gt; nu;\n  real beta;\n  // ...\n}\nmodel {\n  beta ~ student_t(nu, 0, 1);\n  // ...\n}\nwith\nparameters {\n  real&lt;lower=0&gt; nu;\n  real&lt;lower=0&gt; tau;\n  real alpha;\n  // ...\n}\ntransformed parameters {\n  real beta;\n  beta = alpha / sqrt(tau);\n  // ...\n}\nmodel {\n  real half_nu;\n  half_nu = 0.5 * nu;\n  tau ~ gamma(half_nu, half_nu);\n  alpha ~ std_normal();\n  // ...\n}\nAlthough set to 0 here, in most cases, the lower bound for the degrees of freedom parameter nu can be set to 1 or higher; when nu is 1, the result is a Cauchy distribution with fat tails and as nu approaches infinity, the Student-t distribution approaches a normal distribution. Thus the parameter nu characterizes the heaviness of the tails of the model.\n\n\n\nUnfortunately, the usual situation in applied Bayesian modeling involves complex geometries and interactions that are not known analytically. Nevertheless, the non-centered parameterization can still be effective for separating parameters.\n\n\nFor example, a vectorized hierarchical model might draw a vector of coefficients \\(\\beta\\) with definitions as follows. The so-called centered parameterization is as follows.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nAlthough not shown, a full model will have priors on both mu_beta and sigma_beta along with data modeled based on these coefficients. For instance, a standard binary logistic regression with data matrix x and binary outcome vector y would include a likelihood statement such as form y ~ bernoulli_logit(x * beta), leading to an analytically intractable posterior.\nA hierarchical model such as the above will suffer from the same kind of inefficiencies as Neal’s funnel, because the values of beta, mu_beta and sigma_beta are highly correlated in the posterior. The extremity of the correlation depends on the amount of data, with Neal’s funnel being the extreme with no data. In these cases, the non-centered parameterization, discussed in the next section, is preferable; when there is a lot of data, the centered parameterization is more efficient. See Betancourt and Girolami (2013) for more information on the effects of centering in hierarchical models fit with Hamiltonian Monte Carlo.\n\n\n\n\nSometimes the group-level effects do not constrain the hierarchical distribution tightly. Examples arise when there are not many groups, or when the inter-group variation is high. In such cases, hierarchical models can be made much more efficient by shifting the data’s correlation with the parameters to the hyperparameters. Similar to the funnel example, this will be much more efficient in terms of effective sample size when there is not much data (see Betancourt and Girolami (2013)), and in more extreme cases will be necessary to achieve convergence.\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector[K] beta_raw;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // implies: beta ~ normal(mu_beta, sigma_beta)\n  beta = mu_beta + sigma_beta * beta_raw;\n}\nmodel {\n  beta_raw ~ std_normal();\n  // ...\n}\nAny priors defined for mu_beta and sigma_beta remain as defined in the original model.\nAlternatively, Stan’s affine transform can be used to decouple sigma and beta:\nparameters {\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;offset=mu_beta, multiplier=sigma_beta&gt;[K] beta;\n  // ...\n}\nmodel {\n  beta ~ normal(mu_beta, sigma_beta);\n  // ...\n}\nReparameterization of hierarchical models is not limited to the normal distribution, although the normal distribution is the best candidate for doing so. In general, any distribution of parameters in the location-scale family is a good candidate for reparameterization. Let \\(\\beta = l + s\\alpha\\) where \\(l\\) is a location parameter and \\(s\\) is a scale parameter. The parameter \\(l\\) need not be the mean, \\(s\\) need not be the standard deviation, and neither the mean nor the standard deviation need to exist. If \\(\\alpha\\) and \\(\\beta\\) are from the same distributional family but \\(\\alpha\\) has location zero and unit scale, while \\(\\beta\\) has location \\(l\\) and scale \\(s\\), then that distribution is a location-scale distribution. Thus, if \\(\\alpha\\) were a parameter and \\(\\beta\\) were a transformed parameter, then a prior distribution from the location-scale family on \\(\\alpha\\) with location zero and unit scale implies a prior distribution on \\(\\beta\\) with location \\(l\\) and scale \\(s\\). Doing so would reduce the dependence between \\(\\alpha\\), \\(l\\), and \\(s\\).\nThere are several univariate distributions in the location-scale family, such as the Student t distribution, including its special cases of the Cauchy distribution (with one degree of freedom) and the normal distribution (with infinite degrees of freedom). As shown above, if \\(\\alpha\\) is distributed standard normal, then \\(\\beta\\) is distributed normal with mean \\(\\mu = l\\) and standard deviation \\(\\sigma = s\\). The logistic, the double exponential, the generalized extreme value distributions, and the stable distribution are also in the location-scale family.\nAlso, if \\(z\\) is distributed standard normal, then \\(z^2\\) is distributed chi-squared with one degree of freedom. By summing the squares of \\(K\\) independent standard normal variates, one can obtain a single variate that is distributed chi-squared with \\(K\\) degrees of freedom. However, for large \\(K\\), the computational gains of this reparameterization may be overwhelmed by the computational cost of specifying \\(K\\) primitive parameters just to obtain one transformed parameter to use in a model.\n\n\n\nThe benefits of reparameterization are not limited to univariate distributions. A parameter with a multivariate normal prior distribution is also an excellent candidate for reparameterization. Suppose you intend the prior for \\(\\beta\\) to be multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). Such a belief is reflected by the following code.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\nparameters {\n  vector[K] beta;\n  // ...\n}\nmodel {\n  beta ~ multi_normal(mu, Sigma);\n  // ...\n}\nIn this case mu and Sigma are fixed data, but they could be unknown parameters, in which case their priors would be unaffected by a reparameterization of beta.\nIf \\(\\alpha\\) has the same dimensions as \\(\\beta\\) but the elements of \\(\\alpha\\) are independently and identically distributed standard normal such that \\(\\beta = \\mu + L\\alpha\\), where \\(LL^\\top = \\Sigma\\), then \\(\\beta\\) is distributed multivariate normal with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\). One choice for \\(L\\) is the Cholesky factor of \\(\\Sigma\\). Thus, the model above could be reparameterized as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  cov_matrix[K] Sigma;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Sigma);\n}\nparameters {\n  vector[K] alpha;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  beta = mu + L * alpha;\n}\nmodel {\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu, Sigma)\n  // ...\n}\nThis reparameterization is more efficient for two reasons. First, it reduces dependence among the elements of alpha and second, it avoids the need to invert Sigma every time multi_normal is evaluated.\nThe Cholesky factor is also useful when a covariance matrix is decomposed into a correlation matrix that is multiplied from both sides by a diagonal matrix of standard deviations, where either the standard deviations or the correlations are unknown parameters. The Cholesky factor of the covariance matrix is equal to the product of a diagonal matrix of standard deviations and the Cholesky factor of the correlation matrix. Furthermore, the product of a diagonal matrix of standard deviations and a vector is equal to the elementwise product between the standard deviations and that vector. Thus, if for example the correlation matrix Tau were fixed data but the vector of standard deviations sigma were unknown parameters, then a reparameterization of beta in terms of alpha could be implemented as follows.\ndata {\n  int&lt;lower=2&gt; K;\n  vector[K] mu;\n  corr_matrix[K] Tau;\n  // ...\n}\ntransformed data {\n  matrix[K, K] L;\n  L = cholesky_decompose(Tau);\n}\nparameters {\n  vector[K] alpha;\n  vector&lt;lower=0&gt;[K] sigma;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // This equals mu + diag_matrix(sigma) * L * alpha;\n  beta = mu + sigma .* (L * alpha);\n}\nmodel {\n  sigma ~ cauchy(0, 5);\n  alpha ~ std_normal();\n  // implies: beta ~ multi_normal(mu,\n  //  diag_matrix(sigma) * L * L' * diag_matrix(sigma)))\n  // ...\n}\nThis reparameterization of a multivariate normal distribution in terms of standard normal variates can be extended to other multivariate distributions that can be conceptualized as contaminations of the multivariate normal, such as the multivariate Student t and the skew multivariate normal distribution.\nA Wishart distribution can also be reparameterized in terms of standard normal variates and chi-squared variates. Let \\(L\\) be the Cholesky factor of a \\(K \\times K\\) positive definite scale matrix \\(S\\) and let \\(\\nu\\) be the degrees of freedom. If \\[\nA = \\begin{pmatrix}\n\\sqrt{c_{1}} & 0            & \\cdots                & 0 \\\\\nz_{21}       & \\sqrt{c_{2}} & \\ddots                & \\vdots \\\\\n\\vdots       & \\ddots       & \\ddots                & 0 \\\\\nz_{K1}       & \\cdots       & z_{K\\left(K-1\\right)} & \\sqrt{c_{K}}\n\\end{pmatrix},\n\\] where each \\(c_i\\) is distributed chi-squared with \\(\\nu - i + 1\\) degrees of freedom and each \\(z_{ij}\\) is distributed standard normal, then \\(W = LAA^{\\top}L^{\\top}\\) is distributed Wishart with scale matrix \\(S = LL^{\\top}\\) and degrees of freedom \\(\\nu\\). Such a reparameterization can be implemented by the following Stan code:\ndata {\n  int&lt;lower=1&gt; N;\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  vector[K] mu;\n  matrix[N, K] y;\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  int count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal();\n  // implies: L * A * A' * L' ~ wishart(nu, L * L')\n  y ~ multi_normal_cholesky(mu, L * A);\n  // ...\n}\nThis reparameterization is more efficient for three reasons. First, it reduces dependence among the elements of z and second, it avoids the need to invert the covariance matrix, \\(W\\) every time wishart is evaluated. Third, if \\(W\\) is to be used with a multivariate normal distribution, you can pass \\(L A\\) to the more efficient multi_normal_cholesky function, rather than passing \\(W\\) to multi_normal.\nIf \\(W\\) is distributed Wishart with scale matrix \\(S\\) and degrees of freedom \\(\\nu\\), then \\(W^{-1}\\) is distributed inverse Wishart with inverse scale matrix \\(S^{-1}\\) and degrees of freedom \\(\\nu\\). Thus, the previous result can be used to reparameterize the inverse Wishart distribution. Since \\(W = L A A^{\\top} L^{\\top}\\), \\(W^{-1} = L^{{\\top}^{-1}} A^{{\\top}^{-1}} A^{-1} L^{-1}\\), where all four inverses exist, but \\(L^{{-1}^{\\top}} = L^{{\\top}^{-1}}\\) and \\(A^{{-1}^{\\top}} = A^{{\\top}^{-1}}\\). We can slightly modify the above Stan code for this case:\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=K + 2&gt; nu\n  matrix[K, K] L; // Cholesky factor of scale matrix\n  // ...\n}\ntransformed data {\n  matrix[K, K] eye;\n  matrix[K, K] L_inv;\n  for (j in 1:K) {\n    for (i in 1:K) {\n      eye[i, j] = 0.0;\n    }\n    eye[j, j] = 1.0;\n  }\n  L_inv = mdivide_left_tri_low(L, eye);\n}\nparameters {\n  vector&lt;lower=0&gt;[K] c;\n  vector[0.5 * K * (K - 1)] z;\n  // ...\n}\nmodel {\n  matrix[K, K] A;\n  matrix[K, K] A_inv_L_inv;\n  int count;\n  count = 1;\n  for (j in 1:(K - 1)) {\n    for (i in (j + 1):K) {\n      A[i, j] = z[count];\n      count += 1;\n    }\n    for (i in 1:(j - 1)) {\n      A[i, j] = 0.0;\n    }\n    A[j, j] = sqrt(c[j]);\n  }\n  for (i in 1:(K - 1)) {\n    A[i, K] = 0;\n  }\n  A[K, K] = sqrt(c[K]);\n\n  A_inv_L_inv = mdivide_left_tri_low(A, L_inv);\n  for (i in 1:K) {\n    c[i] ~ chi_square(nu - i + 1);\n  }\n\n  z ~ std_normal(); // implies: crossprod(A_inv_L_inv) ~\n  // inv_wishart(nu, L_inv' * L_inv)\n  // ...\n}\nAnother candidate for reparameterization is the Dirichlet distribution with all \\(K\\) shape parameters equal. Zyczkowski and Sommers (2001) shows that if \\(\\theta_i\\) is equal to the sum of \\(\\beta\\) independent squared standard normal variates and \\(\\rho_i = \\frac{\\theta_i}{\\sum \\theta_i}\\), then the \\(K\\)-vector \\(\\rho\\) is distributed Dirichlet with all shape parameters equal to \\(\\frac{\\beta}{2}\\). In particular, if \\(\\beta = 2\\), then \\(\\rho\\) is uniformly distributed on the unit simplex. Thus, we can make \\(\\rho\\) be a transformed parameter to reduce dependence, as in:\ndata {\n  int&lt;lower=1&gt; beta;\n  // ...\n}\nparameters {\n  array[K] vector[beta] z;\n  // ...\n}\ntransformed parameters {\n  simplex[K] rho;\n  for (k in 1:K) {\n    rho[k] = dot_self(z[k]); // sum-of-squares\n  }\n  rho = rho / sum(rho);\n}\nmodel {\n  for (k in 1:K) {\n    z[k] ~ std_normal();\n  }\n  // implies: rho ~ dirichlet(0.5 * beta * ones)\n  // ...\n}",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Whitespace"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/whitespace.html#whitespace-characters",
-    "href": "reference-manual/whitespace.html#whitespace-characters",
-    "title": "Whitespace",
+    "objectID": "stan-users-guide/efficiency-tuning.html#vectorization",
+    "href": "stan-users-guide/efficiency-tuning.html#vectorization",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).",
+    "text": "Stan spends the vast majority of its time computing the gradient of the log probability function, making gradients the obvious target for optimization. Stan’s gradient calculations with algorithmic differentiation require a template expression to be allocated and constructed for each subexpression of a Stan program involving parameters or transformed parameters.2 This section defines optimization strategies based on vectorizing these subexpressions to reduce the work done during algorithmic differentiation.\n\n\n\nBecause of the gradient bottleneck described in the previous section, it is more efficient to collect a sequence of summands into a vector or array and then apply the sum() operation than it is to continually increment a variable by assignment and addition. For example, consider the following code snippet, where foo() is some operation that depends on n.\nfor (n in 1:N) {\n  total += foo(n,...);\n}\nThis code has to create intermediate representations for each of the N summands.\nA faster alternative is to copy the values into a vector, then apply the sum() operator, as in the following refactoring.\n{\n  vector[N] summands;\n  for (n in 1:N) {\n    summands[n] = foo(n,...);\n  }\n  total = sum(summands);\n}\nSyntactically, the replacement is a statement block delineated by curly brackets ({, }), starting with the definition of the local variable summands.\nEven though it involves extra work to allocate the summands vector and copy N values into it, the savings in differentiation more than make up for it. Perhaps surprisingly, it will also use substantially less memory overall than incrementing total within the loop.\n\n\n\nThe following program directly encodes a linear regression with fixed unit noise using a two-dimensional array x of predictors, an array y of outcomes, and an array beta of regression coefficients.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[K, N] real x;\n  array[N] real y;\n}\nparameters {\n  array[K] real beta;\n}\nmodel {\n  for (n in 1:N) {\n    real gamma = 0;\n    for (k in 1:K) {\n      gamma += x[n, k] * beta[k];\n    }\n    y[n] ~ normal(gamma, 1);\n  }\n}\nThe following model computes the same log probability function as the previous model, even supporting the same input files for data and initialization.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  array[N] vector[K] x;\n  array[N] real y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(dot_product(x[n], beta), 1);\n  }\n}\nAlthough it produces equivalent results, the dot product should not be replaced with a transpose and multiply, as in\ny[n] ~ normal(x[n]' * beta, 1);\nThe relative inefficiency of the transpose and multiply approach is that the transposition operator allocates a new vector into which the result of the transposition is copied. This consumes both time and memory.3\nThe inefficiency of transposition could itself be mitigated by reordering the product and pulling the transposition out of the loop, as follows.\n// ...\ntransformed parameters {\n  row_vector[K] beta_t;\n  beta_t = beta';\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(beta_t * x[n], 1);\n  }\n}\nThe problem with transposition could be completely solved by directly encoding the x as a row vector, as in the following example.\ndata {\n  // ...\n  array[N] row_vector[K] x;\n  // ...\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, 1);\n  }\n}\nDeclaring the data as a matrix and then computing all the predictors at once using matrix multiplication is more efficient still, as in the example discussed in the next section.\nHaving said all this, the most efficient way to code this model is with direct matrix multiplication, as in\ndata {\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nIn general, encapsulated single operations that do the work of loops will be more efficient in their encapsulated forms. Rather than performing a sequence of row-vector/vector multiplications, it is better to encapsulate it as a single matrix/vector multiplication.\n\n\n\nThe final and most efficient version replaces the loops and transformed parameters by using the vectorized form of the normal probability function, as in the following example.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n}\nparameters {\n  vector[K] beta;\n}\nmodel {\n  y ~ normal(x * beta, 1);\n}\nThe variables are all declared as either matrix or vector types. The result of the matrix-vector multiplication x * beta in the model block is a vector of the same length as y.\nThe probability function documentation in the function reference manual indicates which of Stan’s probability functions support vectorization; see the function reference manual for full details. Vectorized probability functions accept either vector or scalar inputs for all arguments, with the only restriction being that all vector arguments are the same dimensionality. In the example above, y is a vector of size N, x * beta is a vector of size N, and 1 is a scalar.\n\n\n\nSometimes data does not arrive in a shape that is ideal for vectorization, but can be put into such shape with some munging (either inside Stan’s transformed data block or outside).\nJohn Hall provided a simple example on the Stan users group. Simplifying notation a bit, the original model had a sampling statement in a loop, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu[ii[n]], sigma);\n}\nThe brute force vectorization would build up a mean vector and then vectorize all at once.\n{\n  vector[N] mu_ii;\n  for (n in 1:N) {\n    mu_ii[n] = mu[ii[n]];\n  }\n  y ~ normal(mu_ii, sigma);\n}\nIf there aren’t many levels (values ii[n] can take), then it behooves us to reorganize the data by group in a case like this. Rather than having a single observation vector y, there are K of them. And because Stan doesn’t support ragged arrays, it means K declarations. For instance, with 5 levels, we have\ny_1 ~ normal(mu[1], sigma);\n// ...\ny_5 ~ normal(mu[5], sigma);\nThis way, both the mu and sigma parameters are shared. Which way works out to be more efficient will depend on the shape of the data; if the sizes are small, the simple vectorization may be faster, but for moderate to large sized groups, the full expansion should be faster.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Whitespace"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/whitespace.html#whitespace-neutrality",
-    "href": "reference-manual/whitespace.html#whitespace-neutrality",
-    "title": "Whitespace",
+    "objectID": "stan-users-guide/efficiency-tuning.html#exploiting-sufficient-statistics",
+    "href": "stan-users-guide/efficiency-tuning.html#exploiting-sufficient-statistics",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "Stan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.\nOther than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.",
+    "text": "In some cases, models can be recoded to exploit sufficient statistics in estimation. This can lead to large efficiency gains compared to an expanded model. This section provides examples for Bernoulli and normal distributions, but the same approach can be applied to other members of the exponential family.\n\n\nConsider the following Bernoulli sampling model.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(alpha, beta);\n  for (n in 1:N) {\n    y[n] ~ bernoulli(theta);\n  }\n}\nIn this model, the sum of positive outcomes in y is a sufficient statistic for the chance of success theta. The model may be recoded using the binomial distribution as follows.\ntheta ~ beta(alpha, beta);\nsum(y) ~ binomial(N, theta);\nBecause truth is represented as one and falsehood as zero, the sum sum(y) of a binary vector y is equal to the number of positive outcomes out of a total of N trials.\nThis can be generalized to other discrete cases (one wouldn’t expect continuous observations to be duplicated if they are random). Suppose there are only \\(K\\) possible discrete outcomes, \\(z_1, \\dotsc, z_K\\), but there are \\(N\\) observations, where \\(N\\) is much larger than \\(K\\). If \\(f_k\\) is the frequency of outcome \\(z_k\\), then the entire likelihood with distribution foo can be coded as follows.\nfor (k in 1:K) {\n  target += f[k] * foo_lpmf(z[k] | ...);\n}\nwhere the ellipses are the parameters of the log probability mass function for distribution foo (there’s no distribution called “foo”; this is just a placeholder for any discrete distribution name).\nThe resulting program looks like a “weighted” regression, but here the weights f[k] are counts and thus sufficient statistics for the PMF and simply amount to an alternative, more efficient coding of the same likelihood. For efficiency, the frequencies f[k] should be counted once in the transformed data block and stored.\nThe same trick works for combining multiple binomial observations.\n\n\n\nConsider the following Stan model for fitting a normal distribution to data.\ndata {\n  int N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nWith the vectorized form used for y, Stan is clever enough to only evaluate log(sigma) once, but it still has to evaluate the normal for all of y[1] to y[N], which involves adding up all the squared differences from the mean and then dividing by sigma squared.\nAn equivalent density to the one above (up to normalizing constants that do not depend on parameters), is given in the following Stan program.\ndata {\n  int N;\n  vector[N] y;\n}\ntransformed data {\n  real mean_y = mean(y);\n  real&lt;lower=0&gt; var_y = variance(y);\n  real nm1_over2 = 0.5 * (N - 1);\n  real sqrt_N = sqrt(N);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mean_y ~ normal(mu, sigma / sqrt_N);\n  var_y ~ gamma(nm1_over2, nm1_over2 / sigma^2);\n}\nThe data and parameters are the same in this program as in the first. The second version adds a transformed data block to compute the mean and variance of the data, which are the sufficient statistics here. These are stored along with two other useful constants. Then the program can define distributions over the mean and variance, both of which are scalars here.\nThe original Stan program and this one define the same model in the sense that they define the same log density up to a constant additive term that does not depend on the parameters. The priors on mu and sigma are both improper, but proper priors could be added as additional statements in the model block without affecting the sufficiency.\nThis transform explicitly relies on aggregating the data. Using this trick on parameters leads to more computation than just computing the normal log density, even before accounting for the non-linear change of variables in the variance.\n\n\n\nThe Poisson distribution is the easiest case, because the sum of observations is sufficient. Specifically, we can replace\ny ~ poisson(lambda);\nwith\nsum(y) ~ poisson(size(y) * lambda);\nThis will work even if y is a parameter vector because no Jacobian adjustment is required for summation.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Whitespace"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/whitespace.html#whitespace-location",
-    "href": "reference-manual/whitespace.html#whitespace-location",
-    "title": "Whitespace",
+    "objectID": "stan-users-guide/efficiency-tuning.html#aggregating-common-subexpressions",
+    "href": "stan-users-guide/efficiency-tuning.html#aggregating-common-subexpressions",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "Zero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as a * b, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.\nIdentifiers and literals may not be separated by whitespace. Thus it is not legal to write the number 10000 as 10 000 or to write the identifier normal_lpdf as normal _ lpdf.",
+    "text": "If an expression is calculated once, the value should be saved and reused wherever possible. That is, rather than using exp(theta) in multiple places, declare a local variable to store its value and reuse the local variable.\nAnother case that may not be so obvious is with two multilevel parameters, say a[ii[n]] + b[jj[n]]. If a and b are small (i.e., do not have many levels), then a table a_b of their sums can be created, with\nmatrix[size(a), size(b)] a_b;\nfor (i in 1:size(a)) {\n  for (j in 1:size(b)) {\n    a_b[i, j] = a[i] + b[j];\n  }\n}\nThen the sum can be replaced with a_b[ii[n], jj[n]].",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Whitespace"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html",
-    "href": "reference-manual/user-functions.html",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/efficiency-tuning.html#exploiting-conjugacy",
+    "href": "stan-users-guide/efficiency-tuning.html#exploiting-conjugacy",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "Stan allows users to define their own functions. The basic syntax is a simplified version of that used in C and C++. This chapter specifies how functions are declared, defined, and used in Stan.\n\n\nUser-defined functions appear in a special function-definition block before all of the other program blocks.\nfunctions {\n   // ... function declarations and definitions ...\n}\ndata {\n  // ...\nFunction definitions and declarations may appear in any order. Forward declarations are allowed but not required.\n\n\n\nThe rules for function naming and function-argument naming are the same as for other variables; see the section on variables for more information on valid identifiers. For example,\nreal foo(real mu, real sigma);\ndeclares a function named foo with two argument variables of types real and real. The arguments are named mu and sigma, but that is not part of the declaration.\n\n\nMultiple user-defined functions may have the same name if they have different sequences of argument types. This is known as function overloading.\nFor example, the following two functions are both defined with the name add_up\nreal add_up(real a, real b){\n  return a + b;\n}\n\nreal add_up(real a, real b, real c){\n  return a + b + c;\n}\nThe return types of overloaded functions do not need to be the same. One could define an additional add_up function as follows\nint add_up(int a, int b){\n  return a + b;\n}\nThat being said, functions may not use the same name if their signature only differs by the return type.\nFor example, the following is not permitted\n// illegal\nreal baz(int x);\nint baz(int x);\nFunction names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the reduce_sum family of functions and ODE integrators, which cannot be overloaded.\n\n\n\n\nAll function arguments are mandatory—there are no default values.\n\n\nFunctions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.\n\n\n\nFunctions with void return types may be applied to arguments and used as statements. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.\n\n\n\nOverloaded functions alongside type promotion can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.\nConsider the following two overloaded functions\nreal foo(int a, real b);\nreal foo(real a, int b);\nThese functions do not have a unique minimum when called with two integer arguments foo(1,2), and therefore cannot be called as such.\nPromotion of integers to complex numbers is considered as two separate promotions, one from int to real and a second from real to complex. Consider the following functions with real and complex signatures\nreal bar(real x);\nreal bar(complex z);\nA call bar(5) with an integer argument will be resolved to bar(real) because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.\n\n\n\nThe rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.\nreal foo(real x) { return ... };\n...\nint a = 5;\nreal b = foo(a); // a promoted to type real\nIn addition to promoting int to real, Stan also promotes real to complex, and by transitivity, int to complex. This also works for containers, so an array of int may be assigned to an array of real of the same shape. And we can also promote vector to complex_vector and similarly for row vectors and matrices.\n\n\n\nFunctions whose name ends in _lpdf or _lpmf (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of statements.qmd#distribution-statements.section.\n\n\n\nFunctions of certain types are restricted on scope of usage. Functions whose names end in _lp assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.\nFunctions whose name end in _jacobian assume access to the log probability accumulator may only be used within the transformed parameters block.\nFunctions whose names end in _rng assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in _rng.\nFunctions whose names end in _lpdf and _lpmf can be used anywhere. However, _lupdf and _lupmf functions can only be used in the model block or user-defined probability functions.\nSee the section on function bodies for more information on these special types of function.\n\n\n\n\nStan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.\nThe type void may not be used as an argument type, only a return type for a function with side effects.\n\n\nThe base variable types are integer, real, complex, vector, row_vector, and matrix. No lower-bound or upper-bound constraints are allowed (e.g., real&lt;lower=0&gt; is illegal). Specialized constrained types are also not allowed (e.g., simplex is illegal).\nTuple types of the form tuple(T1, ..., TN) are also allowed, with all of the types T1 to TN being function argument types (i.e., no constraints and no sizes).\n\n\n\nArguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, int denotes a single integer argument or return, whereas array[] real indicates a one-dimensional array of reals, array[,] real a two-dimensional array and array[,,] real a three-dimensional array; whitespace is optional, as usual.\nThe dimensions for vectors and matrices are not included, so that matrix is the type of a single matrix argument or return type. Thus if a variable is declared as matrix a, then a has two indexing dimensions, so that a[1] is a row vector and a[1, 1] a real value. Matrices implicitly have two indexing dimensions. The type declaration matrix[ , ] b specifies that b is a two-dimensional array of matrices, for a total of four indexing dimensions, with b[1, 1, 1, 1] picking out a real value.\n\n\n\nFunction argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.\nDimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.\nAny errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.\n\n\n\nSome of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.\nIn user-defined functions, the qualifier data may be placed before an argument type declaration to indicate that the argument must be data only. For example,\nreal foo(data real x) {\n  return x^2;\n}\nrequires the argument x to be data only.\nDeclaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.\n\n\n\n\nThe body of a function is between an open curly brace ({) and close curly brace (}). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.\nAny user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.\nThe only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.\n\n\nFunctions that call random number generating functions in their bodies must have a name that ends in _rng; attempts to use random-number generators in other functions lead to a compile-time error.\nLike other random number generating functions, user-defined functions with names that end in _rng may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in _rng. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions that include distribution statements or log probability increment statements must have a name that ends in _lp. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.\nLike the target log density increment statement and distribution statements, user-defined functions with names that end in _lp may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions whose names end in _lpdf and _lpmf (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (~) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of y given parameter theta allows the use of the distribution statement is defined as follows.\nreal foo_lpdf(real y, vector theta) { ... }\nNote that for function definitions, the comma is used rather than the vertical bar.\nFor every custom _lpdf and _lpmf defined there is a corresponding _lupdf and _lupmf defined automatically. The _lupdf and _lupmf versions of the functions cannot be defined directly (to do so will produce an error). The difference in the _lpdf and _lpmf and the corresponding _lupdf and _lupmf functions is that if any other unnormalized density functions are used inside the user-defined function, the _lpdf and _lpmf forms of the user-defined function will change these densities to be normalized. The _lupdf and _lupmf forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.\nThe distribution statement shorthand\nz ~ foo(phi);\nwill have the same effect as incrementing the target with the log of the unnormalized density:\ntarget += foo_lupdf(z | phi);\nOther _lupdf and _lupmf functions used in the definition of foo_lpdf will drop additive constants when foo_lupdf is called and will not drop additive constants when foo_lpdf is called.\nIf there are _lupdf and _lupmf functions used inside the following call to foo_lpdf, they will be forced to normalize (return the equivalent of their _lpdf and _lpmf forms):\ntarget += foo_lpdf(z | phi);\nIf there are no _lupdf or _lupmf functions used in the definition of foo_lpdf, then there will be no difference between a foo_lpdf or foo_lupdf call.\nThe unnormalized _lupdf and _lupmf functions can only be used in the model block or in user-defined probability functions (those ending in _lpdf or _lpmf).\nThe same syntax and shorthand that works for _lpdf also works for log probability mass functions with suffixes _lpmf.\nA function that is going to be accessed as distributions must return the log of the density or mass function it defines.\n\n\n\n\nWithin function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (=) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.1\nLocal variables may be declared at the top of the function block and scope as usual.\n\n\n\nNon-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.\nVoid functions may use return only without an argument, but return statements are not mandatory.\n\n\nUnlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that\n\na return statement qualifies,\n\nand the recursive cases are that\n\na sequence of statements qualifies if its last statement qualifies,\na for loop or while loop qualifies if its body qualifies, and\na conditional statement qualifies if it has a default else clause and all of its body statements qualify.\n\nAn exception is made for “obviously infinite” loops like while (1), which contain a return statement and no break statements. The only way to exit such a loop is to return, so they are considered as returning statements.\nThese rules disqualify\nreal foo(real x) {\n  if (x &gt; 2) {\n    return 1.0;\n  } else if (x &lt;= 2) {\n    return -1.0;\n  }\n}\nbecause there is no default else clause, and disqualify\nreal foo(real x) {\n  real y;\n  y = x;\n  while (x &lt; 10) {\n    if (x &gt; 0) {\n      return x;\n    }\n    y = x / 2;\n  }\n}\nbecause the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow\nreal log_fancy(real x) {\n  if (x &lt; 1e-30) {\n    return x;\n  } else if (x &lt; 1e-14) {\n    return x * x;\n  } else {\n    return log(x);\n  }\n}\nbecause there’s a default else clause and each condition body has return as its final statement.\n\n\n\n\n\n\nA function can be declared without a return value by using void in place of a return type. Note that the type void may only be used as a return type—arguments may not be declared to be of type void.\n\n\n\nA void function may be used as a statement.\nBecause there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.\n\n\n\nIn a return statement within a void function’s definition, the return keyword is followed immediately by a semicolon (;) rather than by the expression whose value is returned.\n\n\n\n\nStan supports forward declarations, which look like function definitions without bodies. For example,\nreal unit_normal_lpdf(real y);\ndeclares a function named unit_normal_lpdf that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.\nA function definition with a body simultaneously declares and defines the named function, as in\nreal unit_normal_lpdf(real y) {\n  return -0.5 * square(y);\n}\nA function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the section on function overloading.",
+    "text": "Continuing the model from the previous section, the conjugacy of the beta prior and binomial distribution allow the model to be further optimized to the following equivalent form.\ntheta ~ beta(alpha + sum(y), beta + N - sum(y));\nTo make the model even more efficient, a transformed data variable defined to be sum(y) could be used in the place of sum(y).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#function-definition-block",
-    "href": "reference-manual/user-functions.html#function-definition-block",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/efficiency-tuning.html#standardizing-predictors",
+    "href": "stan-users-guide/efficiency-tuning.html#standardizing-predictors",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "User-defined functions appear in a special function-definition block before all of the other program blocks.\nfunctions {\n   // ... function declarations and definitions ...\n}\ndata {\n  // ...\nFunction definitions and declarations may appear in any order. Forward declarations are allowed but not required.",
+    "text": "Standardizing the data so that all predictors have a zero sample mean and unit sample variance has the following potential benefits:\n\nIt helps in faster convergence of MCMC chains.\nIt makes the model less sensitive to the specifics of the parameterization.\nIt aids in the interpretation and comparison of the importance of coefficients across different predictors.\n\nWhen there are large differences between the units and scales of the predictors, standardizing the predictors is especially useful. This section illustrates the principle for a simple linear regression.\nSuppose that \\(y = (y_1,\\dotsc,y_N)\\) is a vector of \\(N\\) outcomes and \\(x = (x_1,\\dotsc,x_N)\\) the corresponding vector of \\(N\\) predictors. A simple linear regression involving an intercept coefficient \\(\\alpha\\) and slope coefficient \\(\\beta\\) can be expressed as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n,\n\\] where \\[\n\\epsilon_n \\sim \\textsf{normal}(0,\\sigma).\n\\]\nIf \\(x\\) has very large or very small values or if the mean of the values is far away from 0 (on the scale of the values), then it can be more efficient to standardize the predictor values \\(x_n\\). First the elements of \\(x\\) are zero-centered by subtracting the mean, then scaled by dividing by the standard deviation.\nThe mean of \\(x\\) is given by:\n\\[\nmean_x = \\frac{1}{N} \\sum_{n=1}^{N} x_n\n\\]\nThe standard deviation of \\(x\\) is calculated as: \\[\nsd_x = {\\left({\\frac{1}{N} \\sum_{n=1}^{N} (x_n - mean_x)^2}\\right)}^{1/2}\n\\]\nWith these, we compute the \\(z\\), the standardized predictors\n\\[\nz_n = \\frac{x_n - mean_x}{sd_x}\n\\]\nwhere \\(z_n\\) is the standardized value corresponding to \\(x_n\\).\nThe inverse transform is defined by reversing the two normalization steps, first rescaling by the same deviation and relocating by the sample mean.\n\\[\nx_n = z_n sd_x + mean_x\n\\]\nStandardizing the predictors standardizes the scale of the variables, and hence the scale of the priors.\nConsider the following initial model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  // priors\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ normal(0, 5);\n  // likelihood\n  y ~ normal(x * beta + alpha, sigma);\n}\nThe data block for the standardized model is identical. The mean and standard deviation of the data are defined in the transformed data block, along with the standardized predictors.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n}\ntransformed data {\n  real mean_x = mean(x);\n  real sd_x = sd(x);\n  vector[N] x_std = (x - mean_x) / sd_x;\n}\nparameters {\n  real alpha_std;\n  real beta_std;\n  real&lt;lower=0&gt; sigma_std;\n}\nmodel {\n  alpha_std ~ normal(0, 10);\n  beta_std ~ normal(0, 10);\n  sigma_std ~ normal(0, 5);\n  y ~ normal(x_std * beta_std + alpha_std, sigma_std);\n}\nThe parameters are renamed to indicate that they aren’t the “natural” parameters. The transformed data x_std is defined in terms of variables mean_x and sd_x; by declaring these variables in the transformed data block, they will be available in all following blocks, and therefore can be used in the generated quantities block to record the “natural” parameters alpha and beta.\nThe fairly diffuse priors on the coefficients are the same. These could have been transformed as well, but here they are left as is, because the scales make sense as diffuse priors for standardized data.\nThe original regression \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\] has been transformed to a regression on the standardized data variable \\(z\\),\n\\[\ny_n = \\alpha' + \\beta' z_n + \\epsilon_n.\n\\]\nThe likelihood is specified in terms of the standardized parameters. The original slope \\(\\beta\\) is the standardized slope \\(\\beta'\\) scaled by the inverse of the standard deviation of \\(x\\). The original intercept \\(\\alpha\\) is the intercept from the standardized model \\(\\alpha'\\), corrected for the effect of scaling and centering \\(x\\). Thus, the formulas to retrieve \\(\\alpha\\) and \\(\\beta\\) from \\(\\alpha'\\) and \\(\\beta'\\) are:\n\\[\\begin{align*}\n\\beta = \\frac{\\beta'}{\\sigma_x} \\\\\n\\alpha = \\alpha' - \\beta' \\frac{\\mu_x}{\\sigma_x}\n\\end{align*}\\]\nThese recovered parameter values on the original scales can be calculated within Stan using a generated quantities block following the model block,\ngenerated quantities {\n  real beta = beta_std / sd_x;\n  real alpha = alpha_std - beta_std * mean_x / sd_x;\n\n}\nWhen there are multiple real-valued predictors, i.e., when K is the number of predictors, x is an \\(N \\times K\\) matrix, and beta ia \\(K\\)-vector of coefficients, then x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. When \\(K \\ll N\\) the QR reparameterization is recommended for linear and generalized linear models unless there is an informative prior on the location of \\(\\beta\\).\n\n\nFor many applications on the standard scale, normal distributions with location zero and scale one will be used. In these cases, it is more efficient to use\ny ~ std_normal();\nthan to use\ny ~ normal(0, 1);\nbecause the subtraction of the location and division by the scale cancel, as does subtracting the log of the scale.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#function-names",
-    "href": "reference-manual/user-functions.html#function-names",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/efficiency-tuning.html#using-map-reduce",
+    "href": "stan-users-guide/efficiency-tuning.html#using-map-reduce",
+    "title": "Efficiency Tuning",
     "section": "",
-    "text": "The rules for function naming and function-argument naming are the same as for other variables; see the section on variables for more information on valid identifiers. For example,\nreal foo(real mu, real sigma);\ndeclares a function named foo with two argument variables of types real and real. The arguments are named mu and sigma, but that is not part of the declaration.\n\n\nMultiple user-defined functions may have the same name if they have different sequences of argument types. This is known as function overloading.\nFor example, the following two functions are both defined with the name add_up\nreal add_up(real a, real b){\n  return a + b;\n}\n\nreal add_up(real a, real b, real c){\n  return a + b + c;\n}\nThe return types of overloaded functions do not need to be the same. One could define an additional add_up function as follows\nint add_up(int a, int b){\n  return a + b;\n}\nThat being said, functions may not use the same name if their signature only differs by the return type.\nFor example, the following is not permitted\n// illegal\nreal baz(int x);\nint baz(int x);\nFunction names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the reduce_sum family of functions and ODE integrators, which cannot be overloaded.",
+    "text": "The map-reduce operation, even without multi-core MPI support, can be used to make programs more scalable and also more efficient. See the map-reduce chapter for more information on implementing map-reduce operations.\nMap-reduce allows greater scalability because only the Jacobian of the mapped function for each shard is stored. The Jacobian consists of all of the derivatives of the outputs with respect to the parameters. During execution, the derivatives of the shard are evaluated using nested automatic differentiation. As often happens with modern CPUs, reduced memory overhead leads to increased memory locality and faster execution. The Jacobians are all computed with local memory and their outputs stored contiguously in memory.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#calling-functions",
-    "href": "reference-manual/user-functions.html#calling-functions",
-    "title": "User-Defined Functions",
-    "section": "",
-    "text": "All function arguments are mandatory—there are no default values.\n\n\nFunctions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.\n\n\n\nFunctions with void return types may be applied to arguments and used as statements. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.\n\n\n\nOverloaded functions alongside type promotion can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.\nConsider the following two overloaded functions\nreal foo(int a, real b);\nreal foo(real a, int b);\nThese functions do not have a unique minimum when called with two integer arguments foo(1,2), and therefore cannot be called as such.\nPromotion of integers to complex numbers is considered as two separate promotions, one from int to real and a second from real to complex. Consider the following functions with real and complex signatures\nreal bar(real x);\nreal bar(complex z);\nA call bar(5) with an integer argument will be resolved to bar(real) because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.\n\n\n\nThe rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.\nreal foo(real x) { return ... };\n...\nint a = 5;\nreal b = foo(a); // a promoted to type real\nIn addition to promoting int to real, Stan also promotes real to complex, and by transitivity, int to complex. This also works for containers, so an array of int may be assigned to an array of real of the same shape. And we can also promote vector to complex_vector and similarly for row vectors and matrices.\n\n\n\nFunctions whose name ends in _lpdf or _lpmf (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of statements.qmd#distribution-statements.section.\n\n\n\nFunctions of certain types are restricted on scope of usage. Functions whose names end in _lp assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.\nFunctions whose name end in _jacobian assume access to the log probability accumulator may only be used within the transformed parameters block.\nFunctions whose names end in _rng assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in _rng.\nFunctions whose names end in _lpdf and _lpmf can be used anywhere. However, _lupdf and _lupmf functions can only be used in the model block or user-defined probability functions.\nSee the section on function bodies for more information on these special types of function.",
+    "objectID": "stan-users-guide/efficiency-tuning.html#footnotes",
+    "href": "stan-users-guide/efficiency-tuning.html#footnotes",
+    "title": "Efficiency Tuning",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis parameterization came to be known on our mailing lists as the “Matt trick” after Matt Hoffman, who independently came up with it while fitting hierarchical models in Stan.↩︎\nStan uses its own arena-based allocation, so allocation and deallocation are faster than with a raw call to new.↩︎\nFuture versions of Stan may remove this inefficiency by more fully exploiting expression templates inside the Eigen C++ matrix library. This will require enhancing Eigen to deal with mixed-type arguments, such as the type double used for constants and the algorithmic differentiation type stan::math::var used for variables.↩︎",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Efficiency Tuning"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#argument-types-and-qualifiers",
-    "href": "reference-manual/user-functions.html#argument-types-and-qualifiers",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/dae.html",
+    "href": "stan-users-guide/dae.html",
+    "title": "Differential-Algebraic Equations",
     "section": "",
-    "text": "Stan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.\nThe type void may not be used as an argument type, only a return type for a function with side effects.\n\n\nThe base variable types are integer, real, complex, vector, row_vector, and matrix. No lower-bound or upper-bound constraints are allowed (e.g., real&lt;lower=0&gt; is illegal). Specialized constrained types are also not allowed (e.g., simplex is illegal).\nTuple types of the form tuple(T1, ..., TN) are also allowed, with all of the types T1 to TN being function argument types (i.e., no constraints and no sizes).\n\n\n\nArguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, int denotes a single integer argument or return, whereas array[] real indicates a one-dimensional array of reals, array[,] real a two-dimensional array and array[,,] real a three-dimensional array; whitespace is optional, as usual.\nThe dimensions for vectors and matrices are not included, so that matrix is the type of a single matrix argument or return type. Thus if a variable is declared as matrix a, then a has two indexing dimensions, so that a[1] is a row vector and a[1, 1] a real value. Matrices implicitly have two indexing dimensions. The type declaration matrix[ , ] b specifies that b is a two-dimensional array of matrices, for a total of four indexing dimensions, with b[1, 1, 1, 1] picking out a real value.\n\n\n\nFunction argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.\nDimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.\nAny errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.\n\n\n\nSome of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.\nIn user-defined functions, the qualifier data may be placed before an argument type declaration to indicate that the argument must be data only. For example,\nreal foo(data real x) {\n  return x^2;\n}\nrequires the argument x to be data only.\nDeclaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.",
+    "text": "Stan support solving systems of differential-algebraic equations (DAEs) of index 1 (Serban et al. 2021). The solver adaptively refines the solutions in order to satisfy given tolerances.\nOne can think a differential-algebraic system of equations as ODEs with additional algebraic constraints applied to some of the variables. In such a system, the variable derivatives may not be expressed explicitly with a right-hand-side as in ODEs, but implicitly constrained.\nSimilar to ODE solvers, the DAE solvers must not only provide the solution to the DAE itself, but also the gradient of the DAE solution with respect to parameters (the sensitivities). Stan’s DAE solver uses the forward sensitivity calculation to expand the base DAE system with additional DAE equations for the gradients of the solution. For each parameter, an additional full set of \\(N\\) sensitivity states are added meaning that the full DAE solved has \\(N \\, + N \\cdot M\\) states.\nTwo interfaces are provided for the forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. The same principles in the control parameters section apply here.\nInternally Stan’s DAE solver uses a variable-step, variable-order, backward-differentiation formula implementation (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005).\n\n\nA DAE is defined by a set of expressions for the residuals of differential equations and algebraic equations \\(r(y', y, t, \\theta)\\), and consistent initial conditions \\(y(t_0, \\theta) = y_0, y'(t_0, \\theta)=y'_0\\). The DAE is define by residual function as \\(r(y', y, t, \\theta)=0\\). The \\(\\theta\\) dependence is included in the notation to highlight that the solution \\(y(t)\\) is a function of any parameters used in the computation.\n\n\n\nAs an example of a system of DAEs, consider following chemical kinetics problem(Robertson 1966). The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants \\(y_1\\) and \\(y_2\\) to the product \\(y_3\\), and the algebraic equation describes the mass conservation. (Serban and Hindmarsh 2021).\n\\[\\begin{align*}\n\\frac{dy_1}{dt} + \\alpha y_1 - \\beta y_2 y_3 = 0 \\\\\n\\frac{dy_2}{dt} - \\alpha y_1 + \\beta y_2 y_3 + \\gamma y_2^2 = 0 \\\\\ny_1 + y_2 + y_3 - 1.0 = 0\n\\end{align*}\\]\nThe state equations implicitly defines the state \\((y_1(t), y_2(t), y_3(t))\\) at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients \\((\\alpha, \\beta, \\gamma)\\).\nUnlike solving ODEs, solving DAEs requires a consistent initial condition. That is, one must specify both \\(y(t_0)\\) and \\(y'(t_0)\\) so that residual function becomes zero at initial time \\(t_0\\) \\[\\begin{equation*}\nr(y'(t_0), y(t_0), t_0) = 0\n\\end{equation*}\\]\n\n\n\nThe index along a DAE solution \\(y(t)\\) is the minimum number of differentiations of some of the components of the system required to solve for \\(y'\\) uniquely in terms of \\(y\\) and \\(t\\), so that the DAE is converted into an ODE for \\(y\\). Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for \\(y_3\\).\nMost DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically (Ascher and Petzold 1998).\n\n\n\nThe first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the residual function is time, passed as a real; the second argument to the residual function is the system state \\(y\\), passed as a vector, the third argument to the residual function is the state derivative \\(y'\\), also passed as a vector. The residual function’s return value is a vector of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient \\(\\alpha\\), \\(\\beta\\), and \\(\\gamma\\)), data, or any quantities that are needed to define the DAE.\nThe above reaction be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector chem(real t, vector yy, vector yp,\n            real alpha, real beta, real gamma) {\n  vector[3] res;\n  res[1] = yp[1] + alpha * yy[1] - beta * yy[2] * yy[3];\n  res[2] = yp[2] - alpha * yy[1] + beta * yy[2] * yy[3] + gamma * yy[2] * yy[2];\n  res[3] = yy[1] + yy[2] + yy[3] - 1.0;\n  return res;\n}\nThe function takes in a time t (a real), the system state yy (a vector), state derivative yp (a vector), as well as parameter alpha (a real), beta (a real), and gamma (a real). The function returns a vector of the residuals at time t. The DAE coded here does not explicitly depend on t, however one still needs to specify t as an argument.\n\n\nThe types in the DAE residual function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, the third argument is the state derivative passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible DAE signatures:\nvector my_dae1(real t, vector y, vector yp, real a0);\nvector my_dae2(real t, vector y, vector yp, array[] int a0, vector a1);\nvector my_dae3(real t, vector y, vector yp, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector my_dae1(real t, array[] real y, vector yp);\n// Second argument is not a vector\narray[] real my_dae2(real t, vector y, vector yp);\n// Return type is not a vector\nvector my_dae3(real t, vector y);\n// First argument is not a real and missing the third argument\n\n\n\n\nStan provides a dae function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.\ndata {\n  int N;\n  vector[3] yy0;\n  vector[3] yp0;\n  real t0;\n  real alpha;\n  real beta;\n  array[N] real ts;\n  array[N] vector[3] y;\n}\nparameters {\n  real gamma;\n}\ntransformed parameters {\n  vector[3] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);\n}\nSince gamma is a parameter, the DAE solver is called in the transformed parameters block.\n\n\n\nUsing dae_tol one can specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters in order to control the DAE solution.\nvector[3] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,\n                             relative_tolerance,\n                             absolute_tolerance,\n                             max_num_steps,\n                             alpha, beta, gamma);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nThe default value of relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#function-bodies.section",
-    "href": "reference-manual/user-functions.html#function-bodies.section",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/dae.html#notation",
+    "href": "stan-users-guide/dae.html#notation",
+    "title": "Differential-Algebraic Equations",
     "section": "",
-    "text": "The body of a function is between an open curly brace ({) and close curly brace (}). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.\nAny user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.\nThe only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.\n\n\nFunctions that call random number generating functions in their bodies must have a name that ends in _rng; attempts to use random-number generators in other functions lead to a compile-time error.\nLike other random number generating functions, user-defined functions with names that end in _rng may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in _rng. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions that include distribution statements or log probability increment statements must have a name that ends in _lp. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.\nLike the target log density increment statement and distribution statements, user-defined functions with names that end in _lp may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions whose names end in _lpdf and _lpmf (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (~) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of y given parameter theta allows the use of the distribution statement is defined as follows.\nreal foo_lpdf(real y, vector theta) { ... }\nNote that for function definitions, the comma is used rather than the vertical bar.\nFor every custom _lpdf and _lpmf defined there is a corresponding _lupdf and _lupmf defined automatically. The _lupdf and _lupmf versions of the functions cannot be defined directly (to do so will produce an error). The difference in the _lpdf and _lpmf and the corresponding _lupdf and _lupmf functions is that if any other unnormalized density functions are used inside the user-defined function, the _lpdf and _lpmf forms of the user-defined function will change these densities to be normalized. The _lupdf and _lupmf forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.\nThe distribution statement shorthand\nz ~ foo(phi);\nwill have the same effect as incrementing the target with the log of the unnormalized density:\ntarget += foo_lupdf(z | phi);\nOther _lupdf and _lupmf functions used in the definition of foo_lpdf will drop additive constants when foo_lupdf is called and will not drop additive constants when foo_lpdf is called.\nIf there are _lupdf and _lupmf functions used inside the following call to foo_lpdf, they will be forced to normalize (return the equivalent of their _lpdf and _lpmf forms):\ntarget += foo_lpdf(z | phi);\nIf there are no _lupdf or _lupmf functions used in the definition of foo_lpdf, then there will be no difference between a foo_lpdf or foo_lupdf call.\nThe unnormalized _lupdf and _lupmf functions can only be used in the model block or in user-defined probability functions (those ending in _lpdf or _lpmf).\nThe same syntax and shorthand that works for _lpdf also works for log probability mass functions with suffixes _lpmf.\nA function that is going to be accessed as distributions must return the log of the density or mass function it defines.",
+    "text": "A DAE is defined by a set of expressions for the residuals of differential equations and algebraic equations \\(r(y', y, t, \\theta)\\), and consistent initial conditions \\(y(t_0, \\theta) = y_0, y'(t_0, \\theta)=y'_0\\). The DAE is define by residual function as \\(r(y', y, t, \\theta)=0\\). The \\(\\theta\\) dependence is included in the notation to highlight that the solution \\(y(t)\\) is a function of any parameters used in the computation.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#parameters-are-constant",
-    "href": "reference-manual/user-functions.html#parameters-are-constant",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/dae.html#example-chemical-kinetics",
+    "href": "stan-users-guide/dae.html#example-chemical-kinetics",
+    "title": "Differential-Algebraic Equations",
     "section": "",
-    "text": "Within function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (=) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.1\nLocal variables may be declared at the top of the function block and scope as usual.",
+    "text": "As an example of a system of DAEs, consider following chemical kinetics problem(Robertson 1966). The nondimensionalized DAE consists of two differential equations and one algebraic constraint. The differential equations describe the reactions from reactants \\(y_1\\) and \\(y_2\\) to the product \\(y_3\\), and the algebraic equation describes the mass conservation. (Serban and Hindmarsh 2021).\n\\[\\begin{align*}\n\\frac{dy_1}{dt} + \\alpha y_1 - \\beta y_2 y_3 = 0 \\\\\n\\frac{dy_2}{dt} - \\alpha y_1 + \\beta y_2 y_3 + \\gamma y_2^2 = 0 \\\\\ny_1 + y_2 + y_3 - 1.0 = 0\n\\end{align*}\\]\nThe state equations implicitly defines the state \\((y_1(t), y_2(t), y_3(t))\\) at future times as a function of an initial state and the system parameters, in this example the reaction rate coefficients \\((\\alpha, \\beta, \\gamma)\\).\nUnlike solving ODEs, solving DAEs requires a consistent initial condition. That is, one must specify both \\(y(t_0)\\) and \\(y'(t_0)\\) so that residual function becomes zero at initial time \\(t_0\\) \\[\\begin{equation*}\nr(y'(t_0), y(t_0), t_0) = 0\n\\end{equation*}\\]",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#return-value",
-    "href": "reference-manual/user-functions.html#return-value",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/dae.html#index-of-daes",
+    "href": "stan-users-guide/dae.html#index-of-daes",
+    "title": "Differential-Algebraic Equations",
     "section": "",
-    "text": "Non-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.\nVoid functions may use return only without an argument, but return statements are not mandatory.\n\n\nUnlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that\n\na return statement qualifies,\n\nand the recursive cases are that\n\na sequence of statements qualifies if its last statement qualifies,\na for loop or while loop qualifies if its body qualifies, and\na conditional statement qualifies if it has a default else clause and all of its body statements qualify.\n\nAn exception is made for “obviously infinite” loops like while (1), which contain a return statement and no break statements. The only way to exit such a loop is to return, so they are considered as returning statements.\nThese rules disqualify\nreal foo(real x) {\n  if (x &gt; 2) {\n    return 1.0;\n  } else if (x &lt;= 2) {\n    return -1.0;\n  }\n}\nbecause there is no default else clause, and disqualify\nreal foo(real x) {\n  real y;\n  y = x;\n  while (x &lt; 10) {\n    if (x &gt; 0) {\n      return x;\n    }\n    y = x / 2;\n  }\n}\nbecause the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow\nreal log_fancy(real x) {\n  if (x &lt; 1e-30) {\n    return x;\n  } else if (x &lt; 1e-14) {\n    return x * x;\n  } else {\n    return log(x);\n  }\n}\nbecause there’s a default else clause and each condition body has return as its final statement.",
+    "text": "The index along a DAE solution \\(y(t)\\) is the minimum number of differentiations of some of the components of the system required to solve for \\(y'\\) uniquely in terms of \\(y\\) and \\(t\\), so that the DAE is converted into an ODE for \\(y\\). Thus an ODE system is of index 0. The above chemical kinetics DAE is of index 1, as we can perform differentiation of the third equation followed by introducing the first two equations in order to obtain the ODE for \\(y_3\\).\nMost DAE solvers, including the one in Stan, support only index-1 DAEs. For a high index DAE problem the user must first convert it to a lower index system. This often can be done by carrying out differentiations analytically (Ascher and Petzold 1998).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#void-functions-as-statements",
-    "href": "reference-manual/user-functions.html#void-functions-as-statements",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/dae.html#coding-the-dae-system-function",
+    "href": "stan-users-guide/dae.html#coding-the-dae-system-function",
+    "title": "Differential-Algebraic Equations",
     "section": "",
-    "text": "A function can be declared without a return value by using void in place of a return type. Note that the type void may only be used as a return type—arguments may not be declared to be of type void.\n\n\n\nA void function may be used as a statement.\nBecause there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.\n\n\n\nIn a return statement within a void function’s definition, the return keyword is followed immediately by a semicolon (;) rather than by the expression whose value is returned.",
+    "text": "The first step in coding an DAE system in Stan is defining the DAE residual function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the residual function is time, passed as a real; the second argument to the residual function is the system state \\(y\\), passed as a vector, the third argument to the residual function is the state derivative \\(y'\\), also passed as a vector. The residual function’s return value is a vector of the same size as state and state derivatives. Additional arguments can be included in the residual function to pass other information into the solve (these will be passed through the function that starts the DAE solution). These argument can be parameters (in our example, the reaction rate coefficient \\(\\alpha\\), \\(\\beta\\), and \\(\\gamma\\)), data, or any quantities that are needed to define the DAE.\nThe above reaction be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector chem(real t, vector yy, vector yp,\n            real alpha, real beta, real gamma) {\n  vector[3] res;\n  res[1] = yp[1] + alpha * yy[1] - beta * yy[2] * yy[3];\n  res[2] = yp[2] - alpha * yy[1] + beta * yy[2] * yy[3] + gamma * yy[2] * yy[2];\n  res[3] = yy[1] + yy[2] + yy[3] - 1.0;\n  return res;\n}\nThe function takes in a time t (a real), the system state yy (a vector), state derivative yp (a vector), as well as parameter alpha (a real), beta (a real), and gamma (a real). The function returns a vector of the residuals at time t. The DAE coded here does not explicitly depend on t, however one still needs to specify t as an argument.\n\n\nThe types in the DAE residual function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, the third argument is the state derivative passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The fourth argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible DAE signatures:\nvector my_dae1(real t, vector y, vector yp, real a0);\nvector my_dae2(real t, vector y, vector yp, array[] int a0, vector a1);\nvector my_dae3(real t, vector y, vector yp, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector my_dae1(real t, array[] real y, vector yp);\n// Second argument is not a vector\narray[] real my_dae2(real t, vector y, vector yp);\n// Return type is not a vector\nvector my_dae3(real t, vector y);\n// First argument is not a real and missing the third argument",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#forward-declarations.section",
-    "href": "reference-manual/user-functions.html#forward-declarations.section",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/dae.html#solving-daes",
+    "href": "stan-users-guide/dae.html#solving-daes",
+    "title": "Differential-Algebraic Equations",
     "section": "",
-    "text": "Stan supports forward declarations, which look like function definitions without bodies. For example,\nreal unit_normal_lpdf(real y);\ndeclares a function named unit_normal_lpdf that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.\nA function definition with a body simultaneously declares and defines the named function, as in\nreal unit_normal_lpdf(real y) {\n  return -0.5 * square(y);\n}\nA function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the section on function overloading.",
+    "text": "Stan provides a dae function for solving DAEs, so that the above chemical reaction equation can be solved in the following code.\ndata {\n  int N;\n  vector[3] yy0;\n  vector[3] yp0;\n  real t0;\n  real alpha;\n  real beta;\n  array[N] real ts;\n  array[N] vector[3] y;\n}\nparameters {\n  real gamma;\n}\ntransformed parameters {\n  vector[3] y_hat[N] = dae(chem, yy0, yp0, t0, ts, alpha, beta, gamma);\n}\nSince gamma is a parameter, the DAE solver is called in the transformed parameters block.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/user-functions.html#footnotes",
-    "href": "reference-manual/user-functions.html#footnotes",
-    "title": "User-Defined Functions",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nDespite being declared constant and appearing to have a pass-by-value syntax in Stan, the implementation of the language passes function arguments by constant reference in C++.↩︎",
+    "objectID": "stan-users-guide/dae.html#control-dae.section",
+    "href": "stan-users-guide/dae.html#control-dae.section",
+    "title": "Differential-Algebraic Equations",
+    "section": "",
+    "text": "Using dae_tol one can specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters in order to control the DAE solution.\nvector[3] y_hat[N] = dae_tol(chem, yy0, yp0, t0, ts,\n                             relative_tolerance,\n                             absolute_tolerance,\n                             max_num_steps,\n                             alpha, beta, gamma);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nThe default value of relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million. We suggest the user choose the control parameters according to the problem in hand, and resort to the defaults only when no knowledge of the DAE system or the physics it models is available.\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "User-Defined Functions"
+      "Stan Users Guide",
+      "Example Models",
+      "Differential-Algebraic Equations"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html",
-    "href": "reference-manual/transforms.html",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/cross-validation.html",
+    "href": "stan-users-guide/cross-validation.html",
+    "title": "Held-Out Evaluation and Cross-Validation",
     "section": "",
-    "text": "To avoid having to deal with constraints while simulating the Hamiltonian dynamics during sampling, every (multivariate) parameter in a Stan model is transformed to an unconstrained variable behind the scenes by the model compiler. The transform is based on the constraints, if any, in the parameter’s definition. Scalars or the scalar values in vectors, row vectors or matrices may be constrained with lower and/or upper bounds. Vectors may alternatively be constrained to be ordered, positive ordered, or simplexes. Matrices may be constrained to be correlation matrices or covariance matrices. This chapter provides a definition of the transforms used for each type of variable. For examples of how to declare and define these variables in a Stan program, see section Variable declaration. To directly access the functional form of these transformations from inside the Stan language, see Variable Transformation Functions in the Functions Reference.\nStan converts models to C++ classes which define probability functions with support on all of \\(\\mathbb{R}^K\\), where \\(K\\) is the number of unconstrained parameters needed to define the constrained parameters defined in the program. The C++ classes also include code to transform the parameters from unconstrained to constrained and apply the appropriate Jacobians.\n\n\nIn this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.\nStan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with lower=0. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about floating point arithmetic in the Stan User’s Guide.\nThese issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the CmdStan Command-Line Interface Overview.\n\n\n\nThe support of a random variable \\(X\\) with density \\(p_X(x)\\) is that subset of values for which it has non-zero density,\n\\[\n\\mathrm{supp}(X) = \\{ x | p_X(x) &gt; 0 \\}.\n\\]\nIf \\(f\\) is a total function defined on the support of \\(X\\), then \\(Y =\nf(X)\\) is a new random variable. This section shows how to compute the probability density function of \\(Y\\) for well-behaved transforms \\(f\\). The rest of the chapter details the transforms used by Stan.\n\n\nSuppose \\(X\\) is one dimensional and \\(f: \\mathrm{supp}(X) \\rightarrow\n\\mathbb{R}\\) is a one-to-one, monotonic function with a differentiable inverse \\(f^{-1}\\). Then the density of \\(Y\\) is given by\n\\[\np_Y(y) = p_X(f^{-1}(y))\n         \\,\n         \\left| \\, \\frac{d}{dy} f^{-1}(y)\\, \\right|.\n\\]\nThe absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.\n\n\n\nThe multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.\nSuppose \\(X\\) is a \\(K\\)-dimensional random variable with probability density function \\(p_X(x)\\). A new random variable \\(Y = f(X)\\) may be defined by transforming \\(X\\) with a suitably well-behaved function \\(f\\). It suffices for what follows to note that if \\(f\\) is one-to-one and its inverse \\(f^{-1}\\) has a well-defined Jacobian, then the density of \\(Y\\) is\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\, \\left| \\, \\det \\, J_{f^{-1}}(y) \\, \\right|,\n\\]\nwhere \\(\\det{}\\) is the matrix determinant operation and \\(J_{f^{-1}}(y)\\) is the Jacobian matrix of \\(f^{-1}\\) evaluated at \\(y\\). Taking \\(x =\nf^{-1}(y)\\), the Jacobian matrix is defined by\n\\[\nJ_{f^{-1}}(y) =\n\\left[\n\\begin{array}{ccc}\\displaystyle\n\\frac{\\partial x_1}{\\partial y_1}\n& \\cdots\n& \\displaystyle \\frac{\\partial x_1}{\\partial y_{K}}\n\\\\\n\\vdots & \\vdots & \\vdots\n\\\\\n\\displaystyle\\frac{\\partial x_{K}}{\\partial y_1}\n& \\cdots\n& \\displaystyle\\frac{\\partial x_{K}}{\\partial y_{K}}\n\\end{array}\n\\right].\n\\]\nIf the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,\n\\[\n\\det \\, J_{f^{-1}}(y)\n= \\prod_{k=1}^K \\frac{\\partial x_k}{\\partial y_k}.\n\\]\nTriangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.\n\n\n\n\nStan uses a logarithmic transform for lower and upper bounds.\n\n\nIf a variable \\(X\\) is declared to have lower bound \\(a\\), it is transformed to an unbounded variable \\(Y\\), where\n\\[\nY = \\log(X - a).\n\\]\n\n\n\nThe inverse of the lower-bound transform maps an unbounded variable \\(Y\\) to a variable \\(X\\) that is bounded below by \\(a\\) by\n\\[\nX = \\exp(Y) + a.\n\\]\n\n\n\nThe absolute derivative of the inverse transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( \\exp(y) + a \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, given the density \\(p_X\\) of \\(X\\), the density of \\(Y\\) is\n\\[\np_Y(y)\n= p_X\\!\\left( \\exp(y) + a \\right) \\cdot \\exp(y).\n\\]\n\n\n\n\nStan uses a negated logarithmic transform for upper bounds.\n\n\nIf a variable \\(X\\) is declared to have an upper bound \\(b\\), it is transformed to the unbounded variable \\(Y\\) by\n\\[\nY = \\log(b - X).\n\\]\n\n\n\nThe inverse of the upper bound transform converts the unbounded variable \\(Y\\) to the variable \\(X\\) bounded above by \\(b\\) through\n\\[\nX = b - \\exp(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse of the upper bound transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( b - \\exp(y) \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, the density of the unconstrained variable \\(Y\\) is defined in terms of the density of the variable \\(X\\) with an upper bound of \\(b\\) by\n\\[\np_Y(y)\n=   p_X \\!\\left( b - \\exp(y) \\right) \\cdot \\exp(y).\n\\]\n\n\n\n\nFor lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.\n\n\nThe log-odds function is defined for \\(u \\in (0,1)\\) by\n\\[\n\\mathrm{logit}(u) = \\log \\frac{u}{1 - u}.\n\\]\nThe inverse of the log odds function is the logistic sigmoid, defined for \\(v \\in (-\\infty,\\infty)\\) by\n\\[\n\\mathrm{logit}^{-1}(v) = \\frac{1}{1 + \\exp(-v)}.\n\\]\nThe derivative of the logistic sigmoid is\n\\[\n\\frac{d}{dy} \\mathrm{logit}^{-1}(y)\n= \\mathrm{logit}^{-1}(y) \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\n\n\n\nFor variables constrained to be in the open interval \\((a, b)\\), Stan uses a scaled and translated log-odds transform. If variable \\(X\\) is declared to have lower bound \\(a\\) and upper bound \\(b\\), then it is transformed to a new variable \\(Y\\), where\n\\[\nY = \\mathrm{logit} \\left( \\frac{X - a}{b - a} \\right).\n\\]\n\n\n\nThe inverse of this transform is\n\\[\nX = a + (b - a) \\cdot \\mathrm{logit}^{-1}(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse transform is given by\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      a + (b - a) \\cdot \\mathrm{logit}^{-1}(y)\n    \\right)\n  \\right|\n= (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( a + (b - a) \\cdot \\mathrm{logit}^{-1}(y) \\right)\n    \\cdot (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nDespite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, \\(\\mathrm{logit}^{-1}(y)\\) only needs to be evaluated once, so there is only one call to \\(\\exp(-y)\\).\n\n\n\n\nStan uses an affine transform to be able to specify parameters with a given offset and multiplier.\n\n\nFor variables with expected offset \\(\\mu\\) and/or (positive) multiplier \\(\\sigma\\), Stan uses an affine transform. Such a variable \\(X\\) is transformed to a new variable \\(Y\\), where\n\\[\nY = \\frac{X - \\mu}{\\sigma}.\n\\]\nThe default value for the offset \\(\\mu\\) is \\(0\\) and for the multiplier \\(\\sigma\\) is \\(1\\) in case not both are specified.\n\n\n\nThe inverse of this transform is\n\\[\nX =  \\mu + \\sigma \\cdot Y.\n\\]\n\n\n\nThe absolute derivative of the affine inverse transform is\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      \\mu + \\sigma \\cdot y\n    \\right)\n  \\right|\n= \\sigma.\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( \\mu + \\sigma \\cdot y \\right)\n    \\cdot \\sigma.\n\\]\nFor an example of how to code this in Stan, see section Affinely Transformed Real.\n\n\n\n\nFor some modeling tasks, a vector-valued random variable \\(X\\) is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.\nIn constraint terms, an ordered \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies\n\\[\nx_k &lt; x_{k+1}\n\\]\nfor \\(k \\in \\{ 1, \\ldots, K-1 \\}\\).\n\n\nStan’s transform follows the constraint directly. It maps an increasing vector \\(x \\in \\mathbb{R}^{K}\\) to an unconstrained vector \\(y \\in\n\\mathbb{R}^K\\) by setting\n\\[\ny_k\n=\n\\left\\{\n\\begin{array}{ll}\nx_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\log \\left( x_{k} - x_{k-1} \\right) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform for an unconstrained \\(y \\in \\mathbb{R}^K\\) to an ordered sequence \\(x \\in \\mathbb{R}^K\\) is defined by the recursion\n\\[\nx_k\n=\n\\left\\{\n\\begin{array}{ll}\ny_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\nx_{k-1} + \\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\\(x_k\\) can also be expressed iteratively as\n\\[\nx_k = y_1 + \\sum_{k'=2}^k \\exp(y_{k'}).\n\\]\n\n\n\nThe Jacobian of the inverse transform \\(f^{-1}\\) is lower triangular, with diagonal elements for \\(1 \\leq k \\leq K\\) of\n\\[\nJ_{k,k} =\n\\left\\{\n\\begin{array}{ll}\n1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\nBecause \\(J\\) is triangular, the absolute Jacobian determinant is\n\\[\n\\left| \\, \\det \\, J \\, \\right|\n\\ = \\\n\\left| \\, \\prod_{k=1}^K J_{k,k} \\, \\right|\n\\ = \\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\nPutting this all together, if \\(p_X\\) is the density of \\(X\\), then the transformed variable \\(Y\\) has density \\(p_Y\\) given by\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n\\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\n\n\n\n\nPlaceholder.\nThe positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.\n\n\n\nStan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length \\(N\\) with real values and the sum of the vector equals zero. The sum-to-zero matrix is an \\(N \\times M\\) matrix where both the rows and columns sum-to-zero.\nStan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix \\(H\\) such that \\(H\\in\\mathbb R^{N\\times (N-1)}\\) and \\(H^{\\mathsf T}H=I_{N-1},\\;H^{\\mathsf T}\\mathbf 1=0\\). The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,(Seyboldt 2024). Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.\nIn many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:\n\\[\n  \\sigma^2 \\begin{pmatrix}\n  1-\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots&-\\tfrac{1}{N} \\\\\n  -\\tfrac{1}{N} & 1-\\tfrac{1}{N} & \\cdots & -\\tfrac{1}{N} \\\\\n  \\vdots & \\vdots & \\ddots & \\vdots \\\\\n  -\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots &1-\\tfrac{1}{N}\n  \\end{pmatrix}.\n\\]\nThe marginal standard deviation no longer corresponds to \\(\\sigma\\) but is \\(\\sigma \\sqrt{1-\\tfrac{1}{N}}\\). The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of \\(\\sqrt{1-\\tfrac{1}{N}}\\) to adjust the variance to the intended \\(\\sigma\\):\nsum_to_zero_vector[N] x;\nx ~ normal(0, sqrt(N %/% (N - 1)));\nWhen \\(\\sigma\\) is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let \\(y\\) be the unconstrained \\(N - 1\\) vector and \\(y\\) is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of \\(\\sigma\\) is given by \\[\nx = \\underbrace{\\sigma\\sqrt{\\frac{N}{N-1}}\\;H}_{\\text{ size of } N \\times (N - 1)}\\,y.\n\\] This is the classic non-centered model. The crucial detail from above is that the operation of multiplying \\(\\sigma\\) to \\(y\\) is on \\(N - 1\\) dimensions, not \\(N\\) dimensions. When writing the centered model using \\[\nx \\sim \\mathcal N \\bigg(0,\\, \\sigma\\sqrt{\\frac{N}{N-1}} \\bigg),\n\\] we are incrementing the log density by an additional \\(-\\log(\\sigma)\\) and so must increment the log density by\ntarget += log(sigma * sqrt(N * inv(N - 1)));\n\n// or\n// because `sqrt(N * inv(N - 1))` is constant\n\ntarget += log(sigma);\nwhich correctly adjusts for the \\(N - 1\\) free variables.\nMore details of the Helmert matrix are in (Lancaster 1965), for the basic definitions of the isometric log ratio transform see (Egozcue et al. 2003) and Chapter 3 of (Filzmoser, Hron, and Templ 2018) for the pivot coordinate version used here.\n\n\nVectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).\nA sum-to-zero \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies the constraint \\[\n\\sum_{k=1}^K x_k = 0.\n\\]\n\n\n\nThe transform is defined iteratively. Given an \\(x \\in \\mathbb{R}^{N + 1}\\) that sums to zero (i.e., \\(\\sum_{n=1}^{N+1} x_n = 0\\)), the transform proceeds as follows to produce an unconstrained \\(y \\in \\mathbb{R}^N\\). This is mathematically equivalent to pre-multiplying the unconstrained \\(y\\) by an orthogonal standard basis, e.g. constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.\nThe transform is initialized by setting \\[\nS_N = 0\n\\] and \\[\ny_N = -x_{N + 1} \\sqrt{1 + \\frac{1}{N}}.\n\\] The for each \\(n\\) from \\(N - 1\\) down to \\(1\\), let \\[\nw_{n + 1} = \\frac{y_{n + 1}}{\\sqrt{(n + 1)(n + 2)}},\n\\] \\[\nS_n = S_{n + 1} + w_{n + 1},\n\\] and \\[\ny_n = (S_n - x_{n + 1}) \\frac{\\sqrt{n (n + 1)}}{n}.\n\\]\nThis transform is expressed in Stan code as:\n  vector manual_sum_to_zero_jacobian(vector x) {\n    int N = size(x) - 1;\n    vector[N] y;\n    y[N] = -x[N+1] * sqrt(1 + 1. / N);\n    real sum_w = 0;\n    for (n in 1:(N-1)) {\n      int i = N - n;\n      int i_p_1 = i + 1;\n      real w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + 1));\n      sum_w += w;\n      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;\n    }\n    return y;\n  }\n\n\n\nThe inverse transform follows the isometric logratio tranform. It maps an unconstrained vector \\(y \\in \\mathbb{R}^N\\) to a sum-to-zero vector \\(x \\in \\mathbb{R}^{N + 1}\\) such that \\[\n\\sum_{n=1}^{N + 1} x_n = 0.\n\\] The values are defined inductively, starting with \\[\nx_1 = \\sum_{n=1}^N \\frac{y_n}{\\sqrt{n (n + 1)}}\n\\] and then setting \\[\nx_{n + 1} = \\sum_{i = n + 1}^N \\frac{y_i}{\\sqrt{i (i + 1)}}\n- n \\cdot \\frac{y_n}{\\sqrt{n (n + 1)}}.\n\\] for \\(n \\in 1{:}N\\).\nThe definition is such that \\[\n\\sum_{n = 1}^{N + 1} x_n = 0\n\\] by construction, because each of the terms added to \\(x_{n}\\) is then subtracted from \\(x_{n + 1}\\) the number of times it shows up in earlier terms.\n\n\n\nThe inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.\nThe sum-to-zero inverse transform is expressed within Stan as:\n  vector manual_sum_to_zero_inverse_jacobian(vector y) {\n    int N = num_elements(y);\n    vector[N + 1] x = zeros_vector(N + 1);\n    real sum_w = 0;\n    for (n in 1:N) {\n      int i = N - n + 1;\n      real w = y[i] * inv_sqrt(i * (i + 1));\n      sum_w += w;\n      x[i] += sum_w;\n      x[i + 1] -= w * i;\n    }\n    return x;\n  }\nNote that there is no target += increment because the Jacobian is zero.\n\n\n\nThe matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any \\(N\\)-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.\nLet the unconstrained matrix be \\[\n\\mathcal Y \\in \\mathbb R^{n_1 \\times n_2}\n\\] and the zero sum vector transform as \\[\n\\mathbf z = \\mathcal C_n(\\mathbf y)\\;=\\;\n\\begin{bmatrix}H_n\\\\[2pt]-\\mathbf 1_{1\\times d}\\end{bmatrix}\\mathbf y\n\\;\\in\\mathbb R^{n+1},\n\\] where \\(H_n\\in\\mathbb R^{n \\times n}\\) is the orthogonal Helmert matrix and satisfies \\(\\mathbf 1_{1\\times d}A_d^{\\!\\top}=0\\).\nApplying \\(C_n\\) to each slice results in\n\\[\n\\mathcal Z \\;=\\;\n\\mathcal X\n\\times_1 \\bigl[\\mathcal C_{n_1}\\bigr]\n\\times_2 \\bigl[\\mathcal C_{n_2}\\bigr]\n\\] where \\[\n\\mathcal Z \\in\n\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}.\n\\]\nBecause each \\(\\mathbb R^{d_1\\times\\cdots\\times d_N}\\) is invertible on the \\(\\mathbf 1^\\perp\\) subspace the composite map applied to \\(\\mathcal Z\\) is a linear bijection between \\(\\mathbb R^{n_1\\times n_2}\\) and the codomain \\(\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}\\).\n\n\n\n\nVariables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).\nThe unit \\(K\\)-simplex is the set of points \\(x \\in \\mathbb{R}^K\\) such that for \\(1 \\leq k \\leq K\\),\n\\[\nx_k &gt; 0,\n\\]\nand\n\\[\n\\sum_{k=1}^K x_k = 1.\n\\]\nAn alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values \\((0,1)\\), and \\((1,0)\\) and the unit 2-simplex is the line connecting these two points; values such as \\((0.3,0.7)\\) and \\((0.99,0.01)\\) lie on the line. In 3-dimensions, the basis is \\((0,0,1)\\), \\((0,1,0)\\) and \\((1,0,0)\\) and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include \\((0.5,0.5,0)\\), \\((0.2,0.7,0.1)\\) and all other triplets of non-negative values summing to 1.\nAs these examples illustrate, the simplex always picks out a subspace of \\(K-1\\) dimensions from \\(\\mathbb{R}^K\\). Therefore a point \\(x\\) in the \\(K\\)-simplex is fully determined by its first \\(K-1\\) elements \\(x_1, x_2,\n\\ldots, x_{K-1}\\), with\n\\[\nx_K = 1 - \\sum_{k=1}^{K-1} x_k.\n\\]\n\n\nThe length-\\(K\\) unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length \\(K\\).\nLet \\(y\\) represent the unconstrained \\(K - 1\\) values in \\((-\\infty, \\infty)\\). The intermediate sum-to-zero vector \\(z = \\text{sum\\_to\\_zero\\_transform}(y)\\) is length \\(K\\). The unit simplex is then given by \\[\nx_i = \\text{softmax}(z) = \\frac{\\exp(z_i)}{\\sum_{i = 1}^K \\exp(z_i)}\n\\]\nThe sum-to-zero vector transform is described in further detail at the sum-to-zero vector section of the Reference Manual.\n\n\n\n\n\n\nNote\n\n\n\nAll versions of Stan pre-2.37 used the stick-breaking transform. This is documented at Stan 2.36 Reference Manual: Simplex Transform.\n\n\n\n\nThe Jacobian \\(J\\) of the inverse unit-simplex transform is found by restricting \\(J\\) to the subspace spanned by the sum-to-zero vector \\(z\\). The Jacobian is given as the \\((K - 1) \\times (K - 1)\\) matrix \\(J\\) where\n\\[\nJ_{ij} = \\frac{\\partial x_i}{\\partial z_j} =\n\\frac{\\partial}{\\partial z_i} \\left( \\frac{\\exp(z_i)}{{\\sum_{i = 1}^K \\exp(z_i)}} \\right)\n\\] and \\(i,j \\in 1, \\ldots, K - 1\\).\nThe diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification\n\\[\nJ_{ij} =\n\\begin{cases}\nx_i (1 - x_i), & \\text{if } i = j, \\\\\n-x_i x_j, & \\text{if } i \\neq j.\n\\end{cases}\n\\]\nIn matrix form this can be expressed as\n\\[\nJ = \\text{diag}(x) - x x^\\top\n\\]\nThe determinant of this matrix can be found using the Matrix Determinant Lemma:\n\\[\n\\det\\bigl(A + u v^{\\top}\\bigr)\n=\n\\det(A)\\,\\bigl(1 + v^{\\top}A^{-1}u\\bigr).\n\\]\nHere,\n\\[\nA \\;=\\; \\operatorname{diag}(x_{1},\\ldots, x_{K-1}),\n\\quad\nu \\;=\\; -\\bigl(x_1,\\ldots, x_{K-1}\\bigr)^{\\!\\top},\n\\quad\nv \\;=\\; \\bigl(x_{1}, \\ldots, x_{K-1}\\bigr)^{\\!\\top}.\n\\] Therefore,\n\\[\n\\begin{aligned}\n\\det(J)\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_i \\bigg)\n\\bigg(1 + (x_{1},\\ldots, x_{K-1})\\,\\mathrm{diag}\\bigl(x_{1}^{-1},\\ldots,x_{K-1}^{-1}\\bigr)\\,\n\\big(-x_{1},\\ldots,-x_{K-1}\\big)^{\\top}\n\\bigg) \\\\\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg)\n\\bigg(1 - \\sum_{i=1}^{K-1} x_{i}\\bigg)\n=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg) x_{K} \\\\\n&=\n\\prod_{i=1}^{K} x_{i}.\n\\end{aligned}\n\\]\n\n\n\n\nThe transform \\(Y = f(X)\\) can be derived by reversing the stages of the inverse transform,\n\\[\ny_k\n= H^\\top \\bigg(\\log(x_k)\n- \\frac{1}{K}\\sum_{i=1}^K\\log(x_i) \\bigg)\n.\n\\]\nThe matrix \\(H\\) is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.\n\n\n\n\nThe column_stochastic_matrix[N, M] and row_stochastic_matrix[M, N] type in Stan represents an \\(N \\times M\\) matrix where each column (row) is a unit simplex of dimension \\(N\\). In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.\n\n\nA column stochastic matrix \\(X \\in \\mathbb{R}^{N \\times M}\\) is defined such that each column is a simplex. For column \\(m\\) (where \\(1 \\leq m \\leq M\\)):\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{n=1}^N X_{n, m} = 1.\n\\]\nA row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e. the rows of the matrix are simplexes)\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{m=1}^N X_{n, m} = 1.\n\\]\nThis definition ensures that each column (row) of the matrix \\(X\\) lies on the \\(N-1\\) dimensional unit simplex, similar to the simplex[N] type, but extended across multiple columns(rows).\n\n\n\nFor the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).\n\n\n\nThe Jacobian determinant of the inverse transform for each column \\(m\\) in the matrix is given by the product of the diagonal entries \\(J_{n, m}\\) of the lower-triangular Jacobian matrix. This determinant is calculated as:\n\\[\n\\left| \\det J_m \\right| = \\prod_{n=1}^{N-1} \\left( z_{n, m} (1 - z_{n, m}) \\left( 1 - \\sum_{n'=1}^{n-1} X_{n', m} \\right) \\right).\n\\]\nThus, the overall Jacobian determinant for the entire column_stochastic_matrix and row_stochastic_matrix is the product of the determinants for each column (row):\n\\[\n\\left| \\det J \\right| = \\prod_{m=1}^{M} \\left| \\det J_m \\right|.\n\\]\n\n\n\nFor the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).\n\n\n\n\nAn \\(n\\)-dimensional vector \\(x \\in \\mathbb{R}^n\\) is said to be a unit vector if it has unit Euclidean length, so that\n\\[\n\\Vert x \\Vert\n\\ = \\ \\sqrt{x^{\\top}\\,x}\n\\ = \\ \\sqrt{x_1^2 + x_2^2 + \\cdots + x_n^2}\n\\ = \\ 1\\ .\n\\]\n\n\nStan divides an unconstrained vector \\(y \\in \\mathbb{R}^{n}\\) by its norm, \\(\\Vert y \\Vert = \\sqrt{y^\\top y}\\), to obtain a unit vector \\(x\\),\n\\[\nx = \\frac{y}{\\Vert y \\Vert}.\n\\]\nTo generate a unit vector, Stan generates points at random in \\(\\mathbb{R}^n\\) with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. Muller (1959) showed this generates points uniformly at random on \\(S^{n-1}\\). That is, if we draw \\(y_n \\sim \\mathsf{Normal}(0, 1)\\) for \\(n \\in 1{:}n\\), then \\(x = \\frac{y}{\\Vert y \\Vert}\\) has a uniform distribution over \\(S^{n-1}\\). This allows us to use an \\(n\\)-dimensional basis for \\(S^{n-1}\\) that preserves local neighborhoods in that points that are close to each other in \\(\\mathbb{R}^n\\) map to points near each other in \\(S^{n-1}\\). The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in \\(\\mathbb{R}^n\\) that map to identical points in \\(S^{n-1}\\).\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nThe Jacobian matrix relating the input vector \\(y\\) to the output vector \\(x\\) is singular because \\(x^\\top x = 1\\) for any non-zero input vector \\(y\\). Thus, there technically is no unique transformation from \\(x\\) to \\(y\\). To circumvent this issue, let \\(r = \\sqrt{y^\\top y}\\) so that \\(y = r\nx\\). The transformation from \\(\\left(r, x_{-n}\\right)\\) to \\(y\\) is well-defined but \\(r\\) is arbitrary, so we set \\(r = 1\\). In this case, the determinant of the Jacobian is proportional to \\(e^{-\\frac{1}{2} y^\\top y}\\), which is the kernel of a standard multivariate normal distribution with \\(n\\) independent dimensions.\n\n\n\n\nA \\(K \\times K\\) correlation matrix \\(x\\) must be symmetric, so that\n\\[\nx_{k,k'} = x_{k',k}\n\\]\nfor all \\(k,k' \\in \\{ 1, \\ldots, K \\}\\), it must have a unit diagonal, so that\n\\[\nx_{k,k} = 1\n\\]\nfor all \\(k \\in \\{ 1, \\ldots, K \\}\\), and it must be positive definite, so that for every non-zero \\(K\\)-vector \\(a\\),\n\\[\na^{\\top} x a &gt; 0.\n\\]\nThe number of free parameters required to specify a \\(K \\times K\\) correlation matrix is \\(\\binom{K}{2}\\).\nThere is more than one way to map from \\(\\binom{K}{2}\\) unconstrained parameters to a \\(K \\times K\\) correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform Lewandowski, Kurowicka, and Joe (2009).\n\n\nIt is easiest to specify the inverse, going from its \\(\\binom{K}{2}\\) parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose \\(y\\) is a vector containing \\(\\binom{K}{2}\\) unconstrained values. These are first transformed via the bijective function \\(\\tanh : \\mathbb{R} \\rightarrow\n(-1, 1)\\)\n\\[\n\\tanh y = \\frac{\\exp(2y) - 1}{\\exp(2y) + 1}.\n\\]\nThen, define a \\(K \\times K\\) matrix \\(z\\), the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the \\(4 \\times 4\\) case, there are \\(\\binom{4}{2}\\) values arranged as\n\\[\nz\n=\n\\left[\n\\begin{array}{cccc}\n1 & \\tanh y_1 & \\tanh y_2 & \\tanh y_4\n\\\\\n0 & 1 & \\tanh y_3 & \\tanh y_5\n\\\\\n0 & 0 & 1 & \\tanh y_6\n\\\\\n0 & 0 & 0 & 1\n\\end{array}\n\\right]\n.\n\\]\nLewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array \\(z\\) to a correlation matrix \\(x\\). The entry \\(z_{i,j}\\) for \\(i &lt;\nj\\) is interpreted as the canonical partial correlation (CPC) between \\(i\\) and \\(j\\), which is the correlation between \\(i\\)’s residuals and \\(j\\)’s residuals when both \\(i\\) and \\(j\\) are regressed on all variables \\(i'\\) such that \\(i'&lt; i\\). In the case of \\(i=1\\), there are no earlier variables, so \\(z_{1,j}\\) is just the Pearson correlation between \\(i\\) and \\(j\\).\nIn Stan, the LKJ transform is reformulated in terms of a Cholesky factor \\(w\\) of the final correlation matrix, defined for \\(1 \\leq i,j \\leq K\\) by\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\n\\prod_{i'=1}^{i - 1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{if } 1 &lt; i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\\\nz_{i,j} \\, \\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThis does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n\\frac{z_{i,j}}{z_{i-1,j}} \\ w_{i-1,j} \\left( 1 - z_{i-1,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i \\leq j.\n\\end{array}\n\\right.\n\\]\nGiven the upper-triangular Cholesky factor \\(w\\), the final correlation matrix is\n\\[\nx = w^{\\top} w.\n\\]\nLewandowski, Kurowicka, and Joe (2009) show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as\n\\[\n\\mbox{det} \\, x = \\prod_{i=1}^{K-1} \\ \\prod_{j=i+1}^K \\ (1 - z_{i,j}^2)\n= \\prod_{1 \\leq i &lt; j \\leq K} (1 - z_{i,j}^2),\n\\]\n\n\n\nFrom the inverse of equation 11 in (Lewandowski, Kurowicka, and Joe 2009), the absolute Jacobian determinant is\n\\[\n\\sqrt{\\prod_{i=1}^{K-1}\\prod_{j=i+1}^K \\left(1-z_{i,j}^2\\right)^{K-i-1}} \\\n\\times \\prod_{i=1}^{K-1}\\prod_{j=i+1}^K\n\\frac{\\partial z_{i,j}}{\\partial y_{i,j}}\n\\]\n\n\n\nThe correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.\nStarting with a correlation matrix \\(x\\), the first step is to find the unique upper triangular \\(w\\) such that \\(x = w w^{\\top}\\). Because \\(x\\) is positive definite, this can be done by applying the Cholesky decomposition,\n\\[\nw = \\mbox{chol}(x).\n\\]\nThe next step from the Cholesky factor \\(w\\) back to the array \\(z\\) of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of \\(w\\), which when inverted yields\n\\[\nz_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i \\leq j,\n\\\\\nw_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n{w_{i,j}}\n\\\n\\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,j}^2 \\right)^{-1/2}\n& \\mbox{if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThe final stage of the transform reverses the hyperbolic tangent transform, which is defined by\n\\[\ny = \\tanh^{-1} z = \\frac{1}{2} \\log \\left( \\frac{1 + z}{1 - z} \\right).\n\\]\nThe inverse hyperbolic tangent function, \\(\\tanh^{-1}\\), is also called the Fisher transformation.\n\n\n\n\nA \\(K \\times K\\) matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires \\(K + \\binom{K}{2}\\) free parameters to specify a \\(K \\times K\\) covariance matrix.\n\n\nStan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.1\nIf \\(x\\) is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix \\(z =\n\\mathrm{chol}(x)\\) with positive diagonal entries, called a Cholesky factor, such that\n\\[\nx = z \\, z^{\\top}.\n\\]\nThe off-diagonal entries of the Cholesky factor \\(z\\) are unconstrained, but the diagonal entries \\(z_{k,k}\\) must be positive for \\(1 \\leq k\n\\leq K\\).\nTo complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix \\(y\\) defined by\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log z_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nz_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular \\(K \\times K\\) matrix \\(y\\), the first step is to recover the intermediate matrix \\(z\\) by reversing the log transform,\n\\[\nz_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe covariance matrix \\(x\\) is recovered from its Cholesky factor \\(z\\) by taking\n\\[\nx = z \\, z^{\\top}.\n\\]\n\n\n\nThe Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix \\(y\\) to matrix \\(z\\) with positive diagonals and the product transform from the Cholesky factor \\(z\\) to \\(x\\).\nThe transform from unconstrained \\(y\\) to Cholesky factor \\(z\\) has a diagonal Jacobian matrix, the absolute determinant of which is thus\n\\[\n\\prod_{k=1}^K  \\frac{\\partial}{\\partial_{y_{k,k}}} \\, \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K z_{k,k}.\n\\]\nThe Jacobian matrix of the second transform from the Cholesky factor \\(z\\) to the covariance matrix \\(x\\) is also triangular, with diagonal entries corresponding to pairs \\((m,n)\\) with \\(m \\geq n\\), defined by\n\\[\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( z \\, z^{\\top} \\right)_{m,n}\n\\ = \\\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( \\sum_{k=1}^K z_{m,k} \\, z_{n,k} \\right)\n\\ = \\\n\\left\\{\n\\begin{array}{cl}\n2 \\, z_{n,n} & \\mbox{if } m = n \\mbox{ and }\n\\\\\nz_{n,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe absolute Jacobian determinant of the second transform is thus\n\\[\n2^{K} \\ \\prod_{m = 1}^{K} \\ \\prod_{n=1}^{m} z_{n,n}\n\\ = \\\n\\prod_{n=1}^K \\ \\prod_{m=n}^K z_{n,n}\n\\ = \\\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}.\n\\]\nFinally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular \\(y\\) to a symmetric, positive definite matrix \\(x\\) is the product of the Jacobian determinants of the exponentiation and product transforms,\n\\[\n\\left( \\prod_{k=1}^K z_{k,k} \\right)\n\\left(\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}\n\\right)\n\\ = \\\n2^K\n\\, \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\nLet \\(f^{-1}\\) be the inverse transform from a \\(K + \\binom{K}{2}\\)-vector \\(y\\) to the \\(K \\times K\\) covariance matrix \\(x\\). A density function \\(p_X(x)\\) defined on \\(K \\times K\\) covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(K + \\binom{K}{2}\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\ 2^K \\ \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\n\n\n\n\nAn \\(M \\times M\\) covariance matrix \\(\\Sigma\\) can be Cholesky factored to a lower triangular matrix \\(L\\) such that \\(L\\,L^{\\top} = \\Sigma\\). If \\(\\Sigma\\) is positive definite, then \\(L\\) will be \\(M \\times M\\). If \\(\\Sigma\\) is only positive semi-definite, then \\(L\\) will be \\(M \\times N\\), with \\(N &lt; M\\).\nA matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and \\(M \\geq\nN\\). A matrix satisfying these conditions ensures that \\(L \\,\nL^{\\top}\\) is positive semi-definite if \\(M &gt; N\\) and positive definite if \\(M = N\\).\nA Cholesky factor of a covariance matrix requires \\(N + \\binom{N}{2} +\n(M - N)N\\) unconstrained parameters.\n\n\nStan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose \\(x\\) is an \\(M \\times N\\) Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log x_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nx_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform need only invert the logarithm with an exponentiation. If \\(y\\) is the unconstrained matrix representation, then the elements of the constrained matrix \\(x\\) is defined by\n\\[\nx_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe transform has a diagonal Jacobian matrix, the absolute determinant of which is\n\\[\n\\prod_{n=1}^N  \\frac{\\partial}{\\partial_{y_{n,n}}} \\, \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N x_{n,n}.\n\\]\nLet \\(x = f^{-1}(y)\\) be the inverse transform from a \\(N + \\binom{N}{2}\n+ (M - N)N\\) vector to an \\(M \\times N\\) Cholesky factor for a covariance matrix \\(x\\) defined in the previous section. A density function \\(p_X(x)\\) defined on \\(M \\times N\\) Cholesky factors of covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(N + \\binom{N}{2}\n+ (M - N)N\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\prod_{N=1}^N x_{n,n}.\n\\]\n\n\n\n\nA \\(K \\times K\\) correlation matrix \\(\\Omega\\) is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a \\(K \\times K\\) lower-triangular matrix \\(L\\) with positive diagonal elements such that \\(\\Omega = L\\,L^{\\top}\\). Because the correlation matrix has a unit diagonal,\n\\[\n\\Omega_{k,k} = L_k\\,L_k^{\\top} = 1,\n\\]\neach row vector \\(L_k\\) of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of \\(L\\) to be calculated from the off-diagonal elements, so that a Cholesky factor for a \\(K \\times K\\) correlation matrix requires only \\(\\binom{K}{2}\\) unconstrained parameters.\n\n\nIt is easiest to start with the inverse transform from the \\(\\binom{K}{2}\\) unconstrained parameters \\(y\\) to the \\(K \\times K\\) lower-triangular Cholesky factor \\(x\\). The inverse transform is based on the hyperbolic tangent function, \\(\\tanh\\), which satisfies \\(\\tanh(x) \\in (-1,1)\\). Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the section on correlation matrix transforms for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.\nSuppose \\(y\\) is a vector of \\(\\binom{K}{2}\\) unconstrained values. Let \\(z\\) be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the \\(3 \\times 3\\) case,\n\\[\nz =\n\\left[\n\\begin{array}{ccc}\n0 & 0 & 0\n\\\\\n\\tanh y_1 & 0 & 0\n\\\\\n\\tanh y_2 & \\tanh y_3 & 0\n\\end{array}\n\\right]\n\\]\nThe matrix \\(z\\), with entries in the range \\((-1, 1)\\), is then transformed to the Cholesky factor \\(x\\), by taking2\n\\[\nx_{i,j}\n=\n\\left\\{\n\\begin{array}{lll}\n0 & \\mbox{ if } i &lt; j & \\mbox{ [above diagonal]}\n\\\\\n\\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i = j & \\mbox{ [on diagonal]}\n\\\\\nz_{i,j} \\ \\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i &gt; j & \\mbox{ [below diagonal]}\n\\end{array}\n\\right.\n\\]\nIn the \\(3 \\times 3\\) case, this yields\n\\[\nx =\n\\left[\n\\begin{array}{ccc}\n1 & 0 & 0\n\\\\\nz_{2,1} & \\sqrt{1 - x_{2,1}^2} & 0\n\\\\\nz_{3,1} & z_{3,2} \\sqrt{1 - x_{3,1}^2}\n        & \\sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}\n\\end{array}\n\\right],\n\\]\nwhere the \\(z_{i,j} \\in (-1,1)\\) are the \\(\\tanh\\)-transformed \\(y\\).\nThe approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at \\(j=1\\), each below-diagonal entry \\(x_{i,j}\\) is determined by the (signed) fraction \\(z_{i,j}\\) of the remaining length for the row that it consumes. The diagonal entries \\(x_{i,i}\\) get any leftover length from earlier entries in their row. The above-diagonal entries are zero.\n\n\n\nSuppose \\(x\\) is a \\(K \\times K\\) Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values \\(z\\) from \\(x\\),\n\\[\nz_{i,j} = \\frac{x_{i,j}}{\\sqrt{1 - \\sum_{j' &lt; j}x_{i,j'}^2}}.\n\\]\nThe mapping from the resulting \\(z\\) to \\(y\\) inverts \\(\\tanh\\),\n\\[\ny\n\\ = \\\n\\tanh^{-1} z\n\\ = \\\n\\frac{1}{2} \\left( \\log (1 + z) - \\log (1 - z) \\right).\n\\]\n\n\n\nThe Jacobian of the full transform is the product of the Jacobians of its component transforms.\nFirst, for the inverse transform \\(z = \\tanh y\\), the derivative is\n\\[\n\\frac{d}{dy} \\tanh y = \\frac{1}{(\\cosh y)^2}.\n\\]\nSecond, for the inverse transform of \\(z\\) to \\(x\\), the resulting Jacobian matrix \\(J\\) is of dimension \\(\\binom{K}{2} \\times\n\\binom{K}{2}\\), with indexes \\((i,j)\\) for \\((i &gt; j)\\). The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each \\((i,j)\\) pair,\n\\[\n\\left| \\, \\mbox{det} \\, J \\, \\right|\n  \\ = \\ \\prod_{i &gt; j} \\left| \\frac{d}{dz_{i,j}} x_{i,j} \\right|,\n\\]\nwhere\n\\[\n\\frac{d}{dz_{i,j}} x_{i,j}\n= \\sqrt{1 - \\sum_{j' &lt; j} x^2_{i,j'}}.\n\\]\nSo the combined density for unconstrained \\(y\\) is\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n  \\ \\\n  \\prod_{n &lt; \\binom{K}{2}} \\frac{1}{(\\cosh y)^2}\n  \\ \\\n  \\prod_{i &gt; j} \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2\n  \\right)^{1/2},\n\\]\nwhere \\(x = f^{-1}(y)\\) is used for notational convenience. The log Jacobian determinant of the complete inverse transform \\(x = f^{-1}(y)\\) is given by\n\\[\n\\log \\left| \\, \\det J \\, \\right|\n=\n-2 \\sum_{n \\leq \\binom{K}{2}}\n\\log \\cosh y\n\\\n+\n\\\n\\frac{1}{2} \\\n\\sum_{i &gt; j}\n\\log \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2 \\right)\n.\n\\]",
+    "text": "Held-out evaluation involves splitting a data set into two parts, a training data set and a test data set. The training data is used to estimate the model and the test data is used for evaluation. Held-out evaluation is commonly used to declare winners in predictive modeling competitions such as those run by Kaggle.\nCross-validation involves repeated held-out evaluations performed by partitioning a single data set in different ways. The training/test split can be done either by randomly selecting the test set, or by partitioning the data set into several equally-sized subsets and then using each subset in turn as the test data with the other folds as training data.\nHeld-out evaluation and cross-validation may involve any kind of predictive statistics, with common choices being the predictive log density on test data, squared error of parameter estimates, or accuracy in a classification task.\n\n\nGiven training data \\((x, y)\\) consisting of parallel sequences of predictors and observations and test data \\((\\tilde{x}, \\tilde{y})\\) of the same structure, the posterior predictive density is \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n=\n\\int\n  p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n  \\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta,\n\\]\nwhere \\(\\theta\\) is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors \\(\\tilde{x}\\) and the training data \\((x, y).\\)\nThis integral may be calculated with Monte Carlo methods as usual, \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\] where the \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y)\\) are draws from the posterior given only the training data \\((x, y).\\)\nTo avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid \\tilde{x}, x, y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M \\exp(\\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}))\n\\\\[4pt]\n& = & -\\log M + \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)})\n\\end{eqnarray*}\\] where the log sum of exponentials function is defined so as to make the above equation hold, \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\, \\mu_m\n= \\log \\sum_{m=1}^M \\exp(\\mu_m).\n\\] The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\mu_m\n= \\textrm{max}(\\mu)\n+ \\log \\sum_{m = 1}^M \\exp(\\mu_m - \\textrm{max}(\\mu)).\n\\] Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms \\(\\mu_m - \\textrm{max}(\\mu) \\leq 0\\), and thus will not overflow.\n\n\nTo evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of \\(\\log p(\\tilde{y} \\mid \\tilde{x},\n\\theta^{(m)}).\\)\nHere is the code for evaluating the log posterior predictive density in a simple linear regression of the test data \\(\\tilde{y}\\) given predictors \\(\\tilde{x}\\) and training data \\((x, y).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  vector[N_tilde] y_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  real log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);\n}\nOnly the training data x and y are used in the model block. The test data y_tilde and test predictors x_tilde appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use x_tilde in the model block—only the test observations y_tilde are unknown before they are predicted.\nGiven \\(M\\) posterior draws from Stan, the sequence log_p[1:M] will be available, so that the log posterior predictive density of the test data given training data and predictors is just log_sum_exp(log_p) - log(M).\n\n\n\n\n\n\nEstimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If \\(\\theta\\) is the true value and \\(\\hat{\\theta}\\) the estimate, then error is just the difference between the prediction and the true value, \\[\n\\textrm{err} = \\hat{\\theta} - \\theta.\n\\]\nIf the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. Absolute error is defined as \\[\n\\textrm{abs-err} = \\left| \\hat{\\theta} - \\theta \\right|\n\\] and squared error as \\[\n\\textrm{sq-err} = \\left( \\hat{\\theta} - \\theta \\right)^2.\n\\] Gneiting and Raftery (2007) provide a thorough overview of such scoring rules and their properties.\nBayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.\n\n\n\nIn addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, \\(z_1, \\ldots, z_N,\\) then it is common to report mean square error (MSE), \\[\n\\textrm{mse}\n= \\frac{1}{N} \\sum_{n = 1}^N \\left( \\hat{z}_n - z_n\\right)^2.\n\\] To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as root mean square error (RMSE), \\[\n\\textrm{rmse} = \\sqrt{\\textrm{mean-sq-err}}.\n\\]\n\n\n\nConsider a simple linear regression model, parameters for the intercept \\(\\alpha\\) and slope \\(\\beta\\), along with predictors \\(\\tilde{x}_n\\). The standard Bayesian estimate is the expected value of \\(\\tilde{y}\\) given the predictors and training data, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}[\\tilde{y}_n \\mid \\tilde{x}_n, x, y]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}_n^{(m)}\n\\end{eqnarray*}\\] where \\(\\tilde{y}_n^{(m)}\\) is drawn from the data model \\[\n\\tilde{y}_n^{(m)}\n\\sim p(\\tilde{y}_n \\mid \\tilde{x}_n, \\alpha^{(m)}, \\beta^{(m)}),\n\\] for parameters \\(\\alpha^{(m)}\\) and \\(\\beta^{(m)}\\) drawn from the posterior, \\[\n(\\alpha^{(m)}, \\beta^{(m)}) \\sim p(\\alpha, \\beta \\mid x, y).\n\\]\nIn the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws \\(\\tilde{y}_n^{(m)}\\), the same result is obtained by averaging linear predictions, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}\\left[\n          \\alpha + \\beta \\cdot \\tilde{x}_n\n          \\mid \\tilde{x}_n, x, y\n       \\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\] This is possible because \\[\n\\tilde{y}_n^{(m)} \\sim \\textrm{normal}(\\tilde{y}_n \\mid \\alpha^{(m)} +\n\\beta^{(m)} \\cdot \\tilde{x}_n, \\sigma^{(m)}),\n\\] and the normal distribution has symmetric error so that the expectation of \\(\\tilde{y}_n^{(m)}\\) is the same as \\(\\alpha^{(m)} + \\beta^{(m)} \\cdot\n\\tilde{x}_n\\). Replacing the sampled quantity \\(\\tilde{y}_n^{(m)}\\) with its expectation is a general variance reduction technique for Monte Carlo estimates known as Rao-Blackwellization (Rao 1945; Blackwell 1947).\nIn the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, \\[\\begin{eqnarray*}\n\\tilde{y}_n^{(m)}\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\left( \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n \\right)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\frac{1}{M} \\sum_{m = 1}^M (\\beta^{(m)} \\cdot \\tilde{x}_n)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\left( \\frac{1}{M} \\sum_{m = 1}^M \\beta^{(m)}\\right) \\cdot \\tilde{x}_n\n\\\\[4pt]\n& = & \\hat{\\alpha} + \\hat{\\beta} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\]\nIn Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  // ...\n}\n// ...\ngenerated quantities {\n  vector[N_tilde] tilde_y = alpha + beta * x_tilde;\n}\nThe posterior mean of tilde_y calculated by Stan is the Bayesian estimate \\(\\hat{\\tilde{y}}.\\) The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.\n\n\n\n\nCross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation (Vehtari, Gelman, and Gabry 2017).\nPartitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.\n\n\nFor the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.\nfunctions {\n  array[] int permutation_rng(int N) {\n    array[N] int y;\n    for (n in 1 : N) {\n      y[n] = n;\n    }\n    vector[N] theta = rep_vector(1.0 / N, N);\n    for (n in 1 : size(y)) {\n      int i = categorical_rng(theta);\n      int temp = y[n];\n      y[n] = y[i];\n      y[i] = temp;\n    }\n    return y;\n  }\n}\nThe name of the function must end in _rng because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers 1:N.1\nThe transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=N&gt; N_test;\n}\ntransformed data {\n  int N_train = N - N_test;\n  array[N] int permutation = permutation_rng(N);\n  vector[N_train] x_train = x[permutation[1 : N_train]];\n  vector[N_train] y_train = y[permutation[1 : N_train]];\n  vector[N_test] x_test = x[permutation[N_train + 1 : N]];\n  vector[N_test] y_test = y[permutation[N_train + 1 : N]];\n}\nRecall that in Stan, permutation[1:N_train] is an array of integers, so that x[permutation[1 : N_train]] is a vector defined for i in 1:N_train by\nx[permutation[1 : N_train]][i] = x[permutation[1:N_train][i]]\n                               = x[permutation[i]]\nGiven the test/train split, the rest of the model is straightforward.\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_train ~ normal(alpha + beta * x_train, sigma);\n  { alpha, beta, sigma } ~ normal(0, 1);\n}\ngenerated quantities {\n  vector[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);\n  vector[N] err = y_test_sim - y_hat;\n}\nThe prediction y_test_hat is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, \\[\\begin{eqnarray*}\n\\hat{y}^{\\textrm{test}}\n& = & \\mathbb{E}\\left[ y^{\\textrm{test}} \\mid x^{\\textrm{test}}, x^{\\textrm{train}} y^{\\textrm{train}} \\right]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)}.\n\\end{eqnarray*}\\]\nBecause the test set is constant and the expectation operator is linear, the posterior mean of err as defined in the Stan program will be the error of the posterior mean estimate, \\[\\begin{eqnarray*}\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n  - y^{\\textrm{test}}\n\\\\[4pt]\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)} - y^{\\textrm{test}},\n\\end{eqnarray*}\\] where \\[\n\\hat{y}^{\\textrm{test}(m)}\n\\sim p(y \\mid x^{\\textrm{test}}, x^{\\textrm{train}},\ny^{\\textrm{train}}).\n\\] This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should not be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.\nBecause the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.\n\n\n\nIt is straightforward to declare the variable permutation in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.\n\n\n\nCross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to mixed replication in posterior predictive checking, where there is a choice to simulate new elements of existing groups or generate entirely new groups.\nEducation testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.\n\n\n\nOften data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.\n\n\n\nVehtari, Gelman, and Gabry (2017) introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo (Gabry et al. 2019).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#limitations-due-to-finite-accuracy-presentation",
-    "href": "reference-manual/transforms.html#limitations-due-to-finite-accuracy-presentation",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/cross-validation.html#evaluating-posterior-predictive.section",
+    "href": "stan-users-guide/cross-validation.html#evaluating-posterior-predictive.section",
+    "title": "Held-Out Evaluation and Cross-Validation",
     "section": "",
-    "text": "In this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.\nStan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with lower=0. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about floating point arithmetic in the Stan User’s Guide.\nThese issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the CmdStan Command-Line Interface Overview.",
+    "text": "Given training data \\((x, y)\\) consisting of parallel sequences of predictors and observations and test data \\((\\tilde{x}, \\tilde{y})\\) of the same structure, the posterior predictive density is \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n=\n\\int\n  p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n  \\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta,\n\\]\nwhere \\(\\theta\\) is the vector of model parameters. This predictive density is the density of the test observations, conditioned on both the test predictors \\(\\tilde{x}\\) and the training data \\((x, y).\\)\nThis integral may be calculated with Monte Carlo methods as usual, \\[\np(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\] where the \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y)\\) are draws from the posterior given only the training data \\((x, y).\\)\nTo avoid underflow in calculations, it will be more stable to compute densities on the log scale. Taking the logarithm and pushing it through results in a stable computation, \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid \\tilde{x}, x, y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\\\[4pt]\n& = & -\\log M + \\log \\sum_{m = 1}^M \\exp(\\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}))\n\\\\[4pt]\n& = & -\\log M + \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)})\n\\end{eqnarray*}\\] where the log sum of exponentials function is defined so as to make the above equation hold, \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\, \\mu_m\n= \\log \\sum_{m=1}^M \\exp(\\mu_m).\n\\] The log sum of exponentials function can be implemented so as to avoid underflow and maintain high arithmetic precision as \\[\n\\textrm{log-sum-exp}_{m = 1}^M \\mu_m\n= \\textrm{max}(\\mu)\n+ \\log \\sum_{m = 1}^M \\exp(\\mu_m - \\textrm{max}(\\mu)).\n\\] Pulling the maximum out preserves all of its precision. By subtracting the maximum, the terms \\(\\mu_m - \\textrm{max}(\\mu) \\leq 0\\), and thus will not overflow.\n\n\nTo evaluate the log predictive density of a model, it suffices to implement the log predictive density of the test data in the generated quantities block. The log sum of exponentials calculation must be done on the outside of Stan using the posterior draws of \\(\\log p(\\tilde{y} \\mid \\tilde{x},\n\\theta^{(m)}).\\)\nHere is the code for evaluating the log posterior predictive density in a simple linear regression of the test data \\(\\tilde{y}\\) given predictors \\(\\tilde{x}\\) and training data \\((x, y).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  vector[N_tilde] y_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  real log_p = normal_lpdf(y_tilde | alpha + beta * x_tilde, sigma);\n}\nOnly the training data x and y are used in the model block. The test data y_tilde and test predictors x_tilde appear in only the generated quantities block. Thus the program is not cheating by using the test data during training. Although this model does not do so, it would be fair to use x_tilde in the model block—only the test observations y_tilde are unknown before they are predicted.\nGiven \\(M\\) posterior draws from Stan, the sequence log_p[1:M] will be available, so that the log posterior predictive density of the test data given training data and predictors is just log_sum_exp(log_p) - log(M).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#change-of-variables.section",
-    "href": "reference-manual/transforms.html#change-of-variables.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/cross-validation.html#estimation-error",
+    "href": "stan-users-guide/cross-validation.html#estimation-error",
+    "title": "Held-Out Evaluation and Cross-Validation",
     "section": "",
-    "text": "The support of a random variable \\(X\\) with density \\(p_X(x)\\) is that subset of values for which it has non-zero density,\n\\[\n\\mathrm{supp}(X) = \\{ x | p_X(x) &gt; 0 \\}.\n\\]\nIf \\(f\\) is a total function defined on the support of \\(X\\), then \\(Y =\nf(X)\\) is a new random variable. This section shows how to compute the probability density function of \\(Y\\) for well-behaved transforms \\(f\\). The rest of the chapter details the transforms used by Stan.\n\n\nSuppose \\(X\\) is one dimensional and \\(f: \\mathrm{supp}(X) \\rightarrow\n\\mathbb{R}\\) is a one-to-one, monotonic function with a differentiable inverse \\(f^{-1}\\). Then the density of \\(Y\\) is given by\n\\[\np_Y(y) = p_X(f^{-1}(y))\n         \\,\n         \\left| \\, \\frac{d}{dy} f^{-1}(y)\\, \\right|.\n\\]\nThe absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.\n\n\n\nThe multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.\nSuppose \\(X\\) is a \\(K\\)-dimensional random variable with probability density function \\(p_X(x)\\). A new random variable \\(Y = f(X)\\) may be defined by transforming \\(X\\) with a suitably well-behaved function \\(f\\). It suffices for what follows to note that if \\(f\\) is one-to-one and its inverse \\(f^{-1}\\) has a well-defined Jacobian, then the density of \\(Y\\) is\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\, \\left| \\, \\det \\, J_{f^{-1}}(y) \\, \\right|,\n\\]\nwhere \\(\\det{}\\) is the matrix determinant operation and \\(J_{f^{-1}}(y)\\) is the Jacobian matrix of \\(f^{-1}\\) evaluated at \\(y\\). Taking \\(x =\nf^{-1}(y)\\), the Jacobian matrix is defined by\n\\[\nJ_{f^{-1}}(y) =\n\\left[\n\\begin{array}{ccc}\\displaystyle\n\\frac{\\partial x_1}{\\partial y_1}\n& \\cdots\n& \\displaystyle \\frac{\\partial x_1}{\\partial y_{K}}\n\\\\\n\\vdots & \\vdots & \\vdots\n\\\\\n\\displaystyle\\frac{\\partial x_{K}}{\\partial y_1}\n& \\cdots\n& \\displaystyle\\frac{\\partial x_{K}}{\\partial y_{K}}\n\\end{array}\n\\right].\n\\]\nIf the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,\n\\[\n\\det \\, J_{f^{-1}}(y)\n= \\prod_{k=1}^K \\frac{\\partial x_k}{\\partial y_k}.\n\\]\nTriangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.",
+    "text": "Estimation is usually considered for unknown parameters. If the data from which the parameters were estimated came from simulated data, the true value of the parameters may be known. If \\(\\theta\\) is the true value and \\(\\hat{\\theta}\\) the estimate, then error is just the difference between the prediction and the true value, \\[\n\\textrm{err} = \\hat{\\theta} - \\theta.\n\\]\nIf the estimate is larger than the true value, the error is positive, and if it’s smaller, then error is negative. If an estimator’s unbiased, then expected error is zero. So typically, absolute error or squared error are used, which will always have positive expectations for an imperfect estimator. Absolute error is defined as \\[\n\\textrm{abs-err} = \\left| \\hat{\\theta} - \\theta \\right|\n\\] and squared error as \\[\n\\textrm{sq-err} = \\left( \\hat{\\theta} - \\theta \\right)^2.\n\\] Gneiting and Raftery (2007) provide a thorough overview of such scoring rules and their properties.\nBayesian posterior means minimize expected square error, whereas posterior medians minimize expected absolute error. Estimates based on modes rather than probability, such as (penalized) maximum likelihood estimates or maximum a posterior estimates, do not have these properties.\n\n\n\nIn addition to parameters, other unknown quantities may be estimated, such as the score of a football match or the effect of a medical treatment given to a subject. In these cases, square error is defined in the same way. If there are multiple exchangeable outcomes being estimated, \\(z_1, \\ldots, z_N,\\) then it is common to report mean square error (MSE), \\[\n\\textrm{mse}\n= \\frac{1}{N} \\sum_{n = 1}^N \\left( \\hat{z}_n - z_n\\right)^2.\n\\] To put the error back on the scale of the original value, the square root may be applied, resulting in what is known prosaically as root mean square error (RMSE), \\[\n\\textrm{rmse} = \\sqrt{\\textrm{mean-sq-err}}.\n\\]\n\n\n\nConsider a simple linear regression model, parameters for the intercept \\(\\alpha\\) and slope \\(\\beta\\), along with predictors \\(\\tilde{x}_n\\). The standard Bayesian estimate is the expected value of \\(\\tilde{y}\\) given the predictors and training data, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}[\\tilde{y}_n \\mid \\tilde{x}_n, x, y]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}_n^{(m)}\n\\end{eqnarray*}\\] where \\(\\tilde{y}_n^{(m)}\\) is drawn from the data model \\[\n\\tilde{y}_n^{(m)}\n\\sim p(\\tilde{y}_n \\mid \\tilde{x}_n, \\alpha^{(m)}, \\beta^{(m)}),\n\\] for parameters \\(\\alpha^{(m)}\\) and \\(\\beta^{(m)}\\) drawn from the posterior, \\[\n(\\alpha^{(m)}, \\beta^{(m)}) \\sim p(\\alpha, \\beta \\mid x, y).\n\\]\nIn the linear regression case, two stages of simplification can be carried out, the first of which helpfully reduces the variance of the estimator. First, rather than averaging draws \\(\\tilde{y}_n^{(m)}\\), the same result is obtained by averaging linear predictions, \\[\\begin{eqnarray*}\n\\hat{\\tilde{y}}_n\n& = & \\mathbb{E}\\left[\n          \\alpha + \\beta \\cdot \\tilde{x}_n\n          \\mid \\tilde{x}_n, x, y\n       \\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\] This is possible because \\[\n\\tilde{y}_n^{(m)} \\sim \\textrm{normal}(\\tilde{y}_n \\mid \\alpha^{(m)} +\n\\beta^{(m)} \\cdot \\tilde{x}_n, \\sigma^{(m)}),\n\\] and the normal distribution has symmetric error so that the expectation of \\(\\tilde{y}_n^{(m)}\\) is the same as \\(\\alpha^{(m)} + \\beta^{(m)} \\cdot\n\\tilde{x}_n\\). Replacing the sampled quantity \\(\\tilde{y}_n^{(m)}\\) with its expectation is a general variance reduction technique for Monte Carlo estimates known as Rao-Blackwellization (Rao 1945; Blackwell 1947).\nIn the linear case, because the predictor is linear in the coefficients, the estimate can be further simplified to use the estimated coefficients, \\[\\begin{eqnarray*}\n\\tilde{y}_n^{(m)}\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n  \\left( \\alpha^{(m)} + \\beta^{(m)} \\cdot \\tilde{x}_n \\right)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\frac{1}{M} \\sum_{m = 1}^M (\\beta^{(m)} \\cdot \\tilde{x}_n)\n\\\\[4pt]\n& = & \\frac{1}{M} \\sum_{m = 1}^M \\alpha^{(m)}\n      + \\left( \\frac{1}{M} \\sum_{m = 1}^M \\beta^{(m)}\\right) \\cdot \\tilde{x}_n\n\\\\[4pt]\n& = & \\hat{\\alpha} + \\hat{\\beta} \\cdot \\tilde{x}_n.\n\\end{eqnarray*}\\]\nIn Stan, only the first of the two steps (the important variance reduction step) can be coded in the object model. The linear predictor is defined in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n  // ...\n}\n// ...\ngenerated quantities {\n  vector[N_tilde] tilde_y = alpha + beta * x_tilde;\n}\nThe posterior mean of tilde_y calculated by Stan is the Bayesian estimate \\(\\hat{\\tilde{y}}.\\) The posterior median may also be calculated and used as an estimate, though square error and the posterior mean are more commonly reported.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#lower-bound-transform.section",
-    "href": "reference-manual/transforms.html#lower-bound-transform.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/cross-validation.html#cross-validation",
+    "href": "stan-users-guide/cross-validation.html#cross-validation",
+    "title": "Held-Out Evaluation and Cross-Validation",
     "section": "",
-    "text": "Stan uses a logarithmic transform for lower and upper bounds.\n\n\nIf a variable \\(X\\) is declared to have lower bound \\(a\\), it is transformed to an unbounded variable \\(Y\\), where\n\\[\nY = \\log(X - a).\n\\]\n\n\n\nThe inverse of the lower-bound transform maps an unbounded variable \\(Y\\) to a variable \\(X\\) that is bounded below by \\(a\\) by\n\\[\nX = \\exp(Y) + a.\n\\]\n\n\n\nThe absolute derivative of the inverse transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( \\exp(y) + a \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, given the density \\(p_X\\) of \\(X\\), the density of \\(Y\\) is\n\\[\np_Y(y)\n= p_X\\!\\left( \\exp(y) + a \\right) \\cdot \\exp(y).\n\\]",
+    "text": "Cross-validation involves choosing multiple subsets of a data set as the test set and using the other data as training. This can be done by partitioning the data and using each subset in turn as the test set with the remaining subsets as training data. A partition into ten subsets is common to reduce computational overhead. In the limit, when the test set is just a single item, the result is known as leave-one-out (LOO) cross-validation (Vehtari, Gelman, and Gabry 2017).\nPartitioning the data and reusing the partitions is very fiddly in the indexes and may not lead to even divisions of the data. It’s far easier to use random partitions, which support arbitrarily sized test/training splits and can be easily implemented in Stan. The drawback is that the variance of the resulting estimate is higher than with a balanced block partition.\n\n\nFor the simple linear regression model, randomized cross-validation can be implemented in a single model. To randomly permute a vector in Stan, the simplest approach is the following.\nfunctions {\n  array[] int permutation_rng(int N) {\n    array[N] int y;\n    for (n in 1 : N) {\n      y[n] = n;\n    }\n    vector[N] theta = rep_vector(1.0 / N, N);\n    for (n in 1 : size(y)) {\n      int i = categorical_rng(theta);\n      int temp = y[n];\n      y[n] = y[i];\n      y[i] = temp;\n    }\n    return y;\n  }\n}\nThe name of the function must end in _rng because it uses other random functions internally. This will restrict its usage to the transformed data and generated quantities block. The code walks through an array of integers exchanging each item with another randomly chosen item, resulting in a uniformly drawn permutation of the integers 1:N.1\nThe transformed data block uses the permutation RNG to generate training data and test data by taking prefixes and suffixes of the permuted data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=N&gt; N_test;\n}\ntransformed data {\n  int N_train = N - N_test;\n  array[N] int permutation = permutation_rng(N);\n  vector[N_train] x_train = x[permutation[1 : N_train]];\n  vector[N_train] y_train = y[permutation[1 : N_train]];\n  vector[N_test] x_test = x[permutation[N_train + 1 : N]];\n  vector[N_test] y_test = y[permutation[N_train + 1 : N]];\n}\nRecall that in Stan, permutation[1:N_train] is an array of integers, so that x[permutation[1 : N_train]] is a vector defined for i in 1:N_train by\nx[permutation[1 : N_train]][i] = x[permutation[1:N_train][i]]\n                               = x[permutation[i]]\nGiven the test/train split, the rest of the model is straightforward.\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_train ~ normal(alpha + beta * x_train, sigma);\n  { alpha, beta, sigma } ~ normal(0, 1);\n}\ngenerated quantities {\n  vector[N] y_test_hat = normal_rng(alpha + beta * x_test, sigma);\n  vector[N] err = y_test_sim - y_hat;\n}\nThe prediction y_test_hat is defined in the generated quantities block using the general form involving all uncertainty. The posterior of this quantity corresponds to using a posterior mean estimator, \\[\\begin{eqnarray*}\n\\hat{y}^{\\textrm{test}}\n& = & \\mathbb{E}\\left[ y^{\\textrm{test}} \\mid x^{\\textrm{test}}, x^{\\textrm{train}} y^{\\textrm{train}} \\right]\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)}.\n\\end{eqnarray*}\\]\nBecause the test set is constant and the expectation operator is linear, the posterior mean of err as defined in the Stan program will be the error of the posterior mean estimate, \\[\\begin{eqnarray*}\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n  - y^{\\textrm{test}}\n\\\\[4pt]\n& = &\n\\mathbb{E}\\left[\n  \\hat{y}^{\\textrm{test}} - y^{\\textrm{test}}\n  \\mid x^{\\textrm{test}}, x^{\\textrm{train}}, y^{\\textrm{train}}\n\\right]\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\hat{y}^{\\textrm{test}(m)} - y^{\\textrm{test}},\n\\end{eqnarray*}\\] where \\[\n\\hat{y}^{\\textrm{test}(m)}\n\\sim p(y \\mid x^{\\textrm{test}}, x^{\\textrm{train}},\ny^{\\textrm{train}}).\n\\] This just calculates error; taking absolute value or squaring will compute absolute error and mean square error. Note that the absolute value and square operation should not be done within the Stan program because neither is a linear function and the result of averaging squares is not the same as squaring an average in general.\nBecause the test set size is chosen for convenience in cross-validation, results should be presented on a per-item scale, such as average absolute error or root mean square error, not on the scale of error in the fold being evaluated.\n\n\n\nIt is straightforward to declare the variable permutation in the data block instead of the transformed data block and read it in as data. This allows an external program to control the blocking, allowing non-random partitions to be evaluated.\n\n\n\nCross-validation must be done with care if the data is inherently structured. For example, in a simple natural language application, data might be structured by document. For cross-validation, one needs to cross-validate at the document level, not at the individual word level. This is related to mixed replication in posterior predictive checking, where there is a choice to simulate new elements of existing groups or generate entirely new groups.\nEducation testing applications are typically grouped by school district, by school, by classroom, and by demographic features of the individual students or the school as a whole. Depending on the variables of interest, different structured subsets should be evaluated. For example, the focus of interest may be on the performance of entire classrooms, so it would make sense to cross-validate at the class or school level on classroom performance.\n\n\n\nOften data measurements have spatial or temporal properties. For example, home energy consumption varies by time of day, day of week, on holidays, by season, and by ambient temperature (e.g., a hot spell or a cold snap). Cross-validation must be tailored to the predictive goal. For example, in predicting energy consumption, the quantity of interest may be the prediction for next week’s energy consumption given historical data and current weather covariates. This suggests an alternative to cross-validation, wherein individual weeks are each tested given previous data. This often allows comparing how well prediction performs with more or less historical data.\n\n\n\nVehtari, Gelman, and Gabry (2017) introduce a method that approximates the evaluation of leave-one-out cross validation inexpensively using only the data point log likelihoods from a single model fit. This method is documented and implemented in the R package loo (Gabry et al. 2019).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#upper-bounded-scalar",
-    "href": "reference-manual/transforms.html#upper-bounded-scalar",
-    "title": "Constraint Transforms",
-    "section": "",
-    "text": "Stan uses a negated logarithmic transform for upper bounds.\n\n\nIf a variable \\(X\\) is declared to have an upper bound \\(b\\), it is transformed to the unbounded variable \\(Y\\) by\n\\[\nY = \\log(b - X).\n\\]\n\n\n\nThe inverse of the upper bound transform converts the unbounded variable \\(Y\\) to the variable \\(X\\) bounded above by \\(b\\) through\n\\[\nX = b - \\exp(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse of the upper bound transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( b - \\exp(y) \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, the density of the unconstrained variable \\(Y\\) is defined in terms of the density of the variable \\(X\\) with an upper bound of \\(b\\) by\n\\[\np_Y(y)\n=   p_X \\!\\left( b - \\exp(y) \\right) \\cdot \\exp(y).\n\\]",
+    "objectID": "stan-users-guide/cross-validation.html#footnotes",
+    "href": "stan-users-guide/cross-validation.html#footnotes",
+    "title": "Held-Out Evaluation and Cross-Validation",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe traditional approach is to walk through a vector and replace each item with a random element from the remaining elements, which is guaranteed to only move each item once. This was not done here as it’d require new categorical theta because Stan does not have a uniform discrete RNG built in.↩︎",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "Held-Out Evaluation and Cross-Validation"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#logit-transform-jacobian.section",
-    "href": "reference-manual/transforms.html#logit-transform-jacobian.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/complex-numbers.html",
+    "href": "stan-users-guide/complex-numbers.html",
+    "title": "Complex Numbers",
     "section": "",
-    "text": "For lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.\n\n\nThe log-odds function is defined for \\(u \\in (0,1)\\) by\n\\[\n\\mathrm{logit}(u) = \\log \\frac{u}{1 - u}.\n\\]\nThe inverse of the log odds function is the logistic sigmoid, defined for \\(v \\in (-\\infty,\\infty)\\) by\n\\[\n\\mathrm{logit}^{-1}(v) = \\frac{1}{1 + \\exp(-v)}.\n\\]\nThe derivative of the logistic sigmoid is\n\\[\n\\frac{d}{dy} \\mathrm{logit}^{-1}(y)\n= \\mathrm{logit}^{-1}(y) \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\n\n\n\nFor variables constrained to be in the open interval \\((a, b)\\), Stan uses a scaled and translated log-odds transform. If variable \\(X\\) is declared to have lower bound \\(a\\) and upper bound \\(b\\), then it is transformed to a new variable \\(Y\\), where\n\\[\nY = \\mathrm{logit} \\left( \\frac{X - a}{b - a} \\right).\n\\]\n\n\n\nThe inverse of this transform is\n\\[\nX = a + (b - a) \\cdot \\mathrm{logit}^{-1}(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse transform is given by\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      a + (b - a) \\cdot \\mathrm{logit}^{-1}(y)\n    \\right)\n  \\right|\n= (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( a + (b - a) \\cdot \\mathrm{logit}^{-1}(y) \\right)\n    \\cdot (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nDespite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, \\(\\mathrm{logit}^{-1}(y)\\) only needs to be evaluated once, so there is only one call to \\(\\exp(-y)\\).",
+    "text": "Stan supports complex scalars, matrices, and vectors as well as real-based ones.\n\n\nThis section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.\n\n\nComplex numbers can be constructed using imaginary literals. For example,\ncomplex z = -1.1 + 2.3i;\nproduces the complex number \\(-1.1 + 2.3i\\). This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the to_complex() function may be used. For example, the following code will work if x and y are parameters, transformed data, or local variables in a function or model block.\nreal x = // ...\nreal y = // ...\ncomplex z = to_complex(x, y);\nThe real and imaginary parts of the complex number can be accessed with getters as follows.\nreal x = get_real(z);  // x = -1.1\nreal y = get_imag(z);  // y = 2.3\nComplex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.\ncomplex a = 3.2 + 2i;\ncomplex b = to_complex(3.2, 2);\nif (a == b) print(\"hello\");\n\n\n\nInteger- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.\ncomplex z1 = 3;  // int promoted to 3 + 0i\ncomplex z2 = 3.2;  // real promoted to 3.2 + 0.0i\n\n\n\nArrays of complex numbers work as usual and allow the usual curly bracket constructors.\ncomplex z1;  complex z2;  complex z3;\n// ...\narray[3] complex zs = { z1, z2, z3 };\nfor (z in zs) {\n  print(z);\n}\nComplex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.\n\n\n\nAll of the standard complex functions are available, including natural logarithm log(z), natural exponentiation exp(z), and powers pow(z1, z2), as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as sin(z), acos(z), tanh(z) and asinh(z).\nPromotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.\ncomplex times_i(complex z) {\n  complex i = to_complex(0, 1);\n  return i * z;\n}\nFor instance, times_i(1) evaluates to the imaginary base \\(i\\), as does times_i(1.0).\n\n\n\n\nThe simplest way to model a distribution over a complex random number \\(z = x + yi\\) is to consider its real part \\(x\\) and imaginary part \\(y\\) to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.\ncomplex z;\nvector[2] mu;\ncholesky_factor_cov[2] L_Sigma;\n// ...\n[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);\nFor example, if z is data, this can be used to estimate mu and the covariance Cholesky factor L_Sigma. Alternatively, if z is a parameter, mu and L_Sigma may constants defining a prior or further parameters defining a hierarchical model.\n\n\n\nStan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.\ncomplex_vector[3] v;\ncomplex_row_vector[2] rv;\ncomplex_matrix[3, 2] m;\nWe can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of rv above, we could assign it to a constructed row vector.\nrv =  [2 + 3i, 1.9 - 2.3i];\nComplex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.\nFurthermore, it is possible to convert back and forth between arrays and matrices using the to_array functions.\n\n\n\nComplex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take x to be predictors, y to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.\ndata {\n  int&lt;lower=0&gt; N;\n  complex_vector[N] x;\n  complex_vector[N] y;\n}\nparameters {\n  complex alpha;\n  complex beta;\n}\nmodel {\n  complex_vector[N] eps = y - (alpha + beta * x);\n  eps ~  // ...error distribution...\n}\nThe question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.\n\n\nThe simplest approach to error in complex regression is to give the real and imaginary parts of eps_n independent independent normal distributions, as follows.\nparameters {\n  // ...\n  vector[2] sigma;\n}\n// ...\nmodel {\n  // ...\n  get_real(eps) ~ normal(0, sigma[1]);\n  get_imag(eps) ~ normal(0, sigma[2]);\n  sigma ~ //...hyperprior...\n}\nA new error scale vector sigma is introduced, and it should itself get a prior based on the expected scale of error for the problem.\n\n\n\nThe next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.\nparameters {\n  cholesky_factor_corr[2] L_Omega;  // correlation matrix\n  vector[2] sigma;                  // real, imag error scales\n  // ...\n}\n// ...\nmodel {\n  array[N] vector[2] eps_arr;\n  for (n in 1:N) {\n    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };\n  }\n  eps_arr ~ multi_normal_cholesky([0, 0]',\n                                  diag_pre_multiply(sigma, L_Omega));\n  L_Omega ~ lkj_cholesky(4);  // shrink toward diagonal correlation\n  sigma ~ // ... hyperprior ...\n}\nHere, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector sigma and a Cholesky factor or a correlation matrix, L_Omega. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to multi_normal_cholesky, the vector of complex numbers needs to be converted to an array of size 2 vectors.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#affinely-transformed-scalar",
-    "href": "reference-manual/transforms.html#affinely-transformed-scalar",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/complex-numbers.html#working-with-complex-numbers",
+    "href": "stan-users-guide/complex-numbers.html#working-with-complex-numbers",
+    "title": "Complex Numbers",
     "section": "",
-    "text": "Stan uses an affine transform to be able to specify parameters with a given offset and multiplier.\n\n\nFor variables with expected offset \\(\\mu\\) and/or (positive) multiplier \\(\\sigma\\), Stan uses an affine transform. Such a variable \\(X\\) is transformed to a new variable \\(Y\\), where\n\\[\nY = \\frac{X - \\mu}{\\sigma}.\n\\]\nThe default value for the offset \\(\\mu\\) is \\(0\\) and for the multiplier \\(\\sigma\\) is \\(1\\) in case not both are specified.\n\n\n\nThe inverse of this transform is\n\\[\nX =  \\mu + \\sigma \\cdot Y.\n\\]\n\n\n\nThe absolute derivative of the affine inverse transform is\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      \\mu + \\sigma \\cdot y\n    \\right)\n  \\right|\n= \\sigma.\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( \\mu + \\sigma \\cdot y \\right)\n    \\cdot \\sigma.\n\\]\nFor an example of how to code this in Stan, see section Affinely Transformed Real.",
+    "text": "This section describes the complex scalar type, including how to build complex numbers, assign them, and use them in arrays and functions.\n\n\nComplex numbers can be constructed using imaginary literals. For example,\ncomplex z = -1.1 + 2.3i;\nproduces the complex number \\(-1.1 + 2.3i\\). This only works if the real and imaginary components are literal numerals. To construct a complex number out of arbitrary real variables, the to_complex() function may be used. For example, the following code will work if x and y are parameters, transformed data, or local variables in a function or model block.\nreal x = // ...\nreal y = // ...\ncomplex z = to_complex(x, y);\nThe real and imaginary parts of the complex number can be accessed with getters as follows.\nreal x = get_real(z);  // x = -1.1\nreal y = get_imag(z);  // y = 2.3\nComplex numbers can be compared using equality (or inequality), but not with greater than or less than operators. For example, after running the code above, the following code snippet will print “hello”.\ncomplex a = 3.2 + 2i;\ncomplex b = to_complex(3.2, 2);\nif (a == b) print(\"hello\");\n\n\n\nInteger- or real-valued expressions may be assigned to complex numbers, with the corresponding imaginary component set to zero.\ncomplex z1 = 3;  // int promoted to 3 + 0i\ncomplex z2 = 3.2;  // real promoted to 3.2 + 0.0i\n\n\n\nArrays of complex numbers work as usual and allow the usual curly bracket constructors.\ncomplex z1;  complex z2;  complex z3;\n// ...\narray[3] complex zs = { z1, z2, z3 };\nfor (z in zs) {\n  print(z);\n}\nComplex arrays allow assignment into their elements, with integer or real assigned values being promoted to complex.\n\n\n\nAll of the standard complex functions are available, including natural logarithm log(z), natural exponentiation exp(z), and powers pow(z1, z2), as well as all of the trig and hyperbolic trigonometric functions and their inverse, such as sin(z), acos(z), tanh(z) and asinh(z).\nPromotion also works for complex-valued function arguments, which may be passed integer or real values, which will be promoted before the function is evaluated. For example, the following user-defined complex function will accept integer, real, or complex arguments.\ncomplex times_i(complex z) {\n  complex i = to_complex(0, 1);\n  return i * z;\n}\nFor instance, times_i(1) evaluates to the imaginary base \\(i\\), as does times_i(1.0).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#ordered-vector",
-    "href": "reference-manual/transforms.html#ordered-vector",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/complex-numbers.html#complex-random-variables",
+    "href": "stan-users-guide/complex-numbers.html#complex-random-variables",
+    "title": "Complex Numbers",
     "section": "",
-    "text": "For some modeling tasks, a vector-valued random variable \\(X\\) is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.\nIn constraint terms, an ordered \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies\n\\[\nx_k &lt; x_{k+1}\n\\]\nfor \\(k \\in \\{ 1, \\ldots, K-1 \\}\\).\n\n\nStan’s transform follows the constraint directly. It maps an increasing vector \\(x \\in \\mathbb{R}^{K}\\) to an unconstrained vector \\(y \\in\n\\mathbb{R}^K\\) by setting\n\\[\ny_k\n=\n\\left\\{\n\\begin{array}{ll}\nx_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\log \\left( x_{k} - x_{k-1} \\right) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform for an unconstrained \\(y \\in \\mathbb{R}^K\\) to an ordered sequence \\(x \\in \\mathbb{R}^K\\) is defined by the recursion\n\\[\nx_k\n=\n\\left\\{\n\\begin{array}{ll}\ny_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\nx_{k-1} + \\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\\(x_k\\) can also be expressed iteratively as\n\\[\nx_k = y_1 + \\sum_{k'=2}^k \\exp(y_{k'}).\n\\]\n\n\n\nThe Jacobian of the inverse transform \\(f^{-1}\\) is lower triangular, with diagonal elements for \\(1 \\leq k \\leq K\\) of\n\\[\nJ_{k,k} =\n\\left\\{\n\\begin{array}{ll}\n1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\nBecause \\(J\\) is triangular, the absolute Jacobian determinant is\n\\[\n\\left| \\, \\det \\, J \\, \\right|\n\\ = \\\n\\left| \\, \\prod_{k=1}^K J_{k,k} \\, \\right|\n\\ = \\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\nPutting this all together, if \\(p_X\\) is the density of \\(X\\), then the transformed variable \\(Y\\) has density \\(p_Y\\) given by\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n\\\n\\prod_{k=2}^K \\exp(y_k).\n\\]",
+    "text": "The simplest way to model a distribution over a complex random number \\(z = x + yi\\) is to consider its real part \\(x\\) and imaginary part \\(y\\) to have a bivariate normal distribution. For example, a complex prior can be expressed as follows.\ncomplex z;\nvector[2] mu;\ncholesky_factor_cov[2] L_Sigma;\n// ...\n[get_real(z), get_imag(z)]' ~ multi_normal_cholesky(mu, L_Sigma);\nFor example, if z is data, this can be used to estimate mu and the covariance Cholesky factor L_Sigma. Alternatively, if z is a parameter, mu and L_Sigma may constants defining a prior or further parameters defining a hierarchical model.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#positive-ordered-vector",
-    "href": "reference-manual/transforms.html#positive-ordered-vector",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/complex-numbers.html#complex-matrices-and-vectors",
+    "href": "stan-users-guide/complex-numbers.html#complex-matrices-and-vectors",
+    "title": "Complex Numbers",
     "section": "",
-    "text": "Placeholder.\nThe positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.",
+    "text": "Stan supports complex matrices, vectors, and row vectors. Variables of these types are declared with sizes in the same way as their real-based counterparts.\ncomplex_vector[3] v;\ncomplex_row_vector[2] rv;\ncomplex_matrix[3, 2] m;\nWe can construct vectors and matrices using brackets in the same way as for real-valued vectors and matrices. For example, given the declaration of rv above, we could assign it to a constructed row vector.\nrv =  [2 + 3i, 1.9 - 2.3i];\nComplex matrices and vectors support all of the standard arithmetic operations including negation, addition, subtraction, and multiplication (division involves a solve, and isn’t a simple arithmetic operation for matrices). They also support transposition.\nFurthermore, it is possible to convert back and forth between arrays and matrices using the to_array functions.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#sum-to-zero-transforms",
-    "href": "reference-manual/transforms.html#sum-to-zero-transforms",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/complex-numbers.html#complex-linear-regression",
+    "href": "stan-users-guide/complex-numbers.html#complex-linear-regression",
+    "title": "Complex Numbers",
     "section": "",
-    "text": "Stan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length \\(N\\) with real values and the sum of the vector equals zero. The sum-to-zero matrix is an \\(N \\times M\\) matrix where both the rows and columns sum-to-zero.\nStan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix \\(H\\) such that \\(H\\in\\mathbb R^{N\\times (N-1)}\\) and \\(H^{\\mathsf T}H=I_{N-1},\\;H^{\\mathsf T}\\mathbf 1=0\\). The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,(Seyboldt 2024). Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.\nIn many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:\n\\[\n  \\sigma^2 \\begin{pmatrix}\n  1-\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots&-\\tfrac{1}{N} \\\\\n  -\\tfrac{1}{N} & 1-\\tfrac{1}{N} & \\cdots & -\\tfrac{1}{N} \\\\\n  \\vdots & \\vdots & \\ddots & \\vdots \\\\\n  -\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots &1-\\tfrac{1}{N}\n  \\end{pmatrix}.\n\\]\nThe marginal standard deviation no longer corresponds to \\(\\sigma\\) but is \\(\\sigma \\sqrt{1-\\tfrac{1}{N}}\\). The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of \\(\\sqrt{1-\\tfrac{1}{N}}\\) to adjust the variance to the intended \\(\\sigma\\):\nsum_to_zero_vector[N] x;\nx ~ normal(0, sqrt(N %/% (N - 1)));\nWhen \\(\\sigma\\) is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let \\(y\\) be the unconstrained \\(N - 1\\) vector and \\(y\\) is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of \\(\\sigma\\) is given by \\[\nx = \\underbrace{\\sigma\\sqrt{\\frac{N}{N-1}}\\;H}_{\\text{ size of } N \\times (N - 1)}\\,y.\n\\] This is the classic non-centered model. The crucial detail from above is that the operation of multiplying \\(\\sigma\\) to \\(y\\) is on \\(N - 1\\) dimensions, not \\(N\\) dimensions. When writing the centered model using \\[\nx \\sim \\mathcal N \\bigg(0,\\, \\sigma\\sqrt{\\frac{N}{N-1}} \\bigg),\n\\] we are incrementing the log density by an additional \\(-\\log(\\sigma)\\) and so must increment the log density by\ntarget += log(sigma * sqrt(N * inv(N - 1)));\n\n// or\n// because `sqrt(N * inv(N - 1))` is constant\n\ntarget += log(sigma);\nwhich correctly adjusts for the \\(N - 1\\) free variables.\nMore details of the Helmert matrix are in (Lancaster 1965), for the basic definitions of the isometric log ratio transform see (Egozcue et al. 2003) and Chapter 3 of (Filzmoser, Hron, and Templ 2018) for the pivot coordinate version used here.\n\n\nVectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).\nA sum-to-zero \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies the constraint \\[\n\\sum_{k=1}^K x_k = 0.\n\\]\n\n\n\nThe transform is defined iteratively. Given an \\(x \\in \\mathbb{R}^{N + 1}\\) that sums to zero (i.e., \\(\\sum_{n=1}^{N+1} x_n = 0\\)), the transform proceeds as follows to produce an unconstrained \\(y \\in \\mathbb{R}^N\\). This is mathematically equivalent to pre-multiplying the unconstrained \\(y\\) by an orthogonal standard basis, e.g. constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.\nThe transform is initialized by setting \\[\nS_N = 0\n\\] and \\[\ny_N = -x_{N + 1} \\sqrt{1 + \\frac{1}{N}}.\n\\] The for each \\(n\\) from \\(N - 1\\) down to \\(1\\), let \\[\nw_{n + 1} = \\frac{y_{n + 1}}{\\sqrt{(n + 1)(n + 2)}},\n\\] \\[\nS_n = S_{n + 1} + w_{n + 1},\n\\] and \\[\ny_n = (S_n - x_{n + 1}) \\frac{\\sqrt{n (n + 1)}}{n}.\n\\]\nThis transform is expressed in Stan code as:\n  vector manual_sum_to_zero_jacobian(vector x) {\n    int N = size(x) - 1;\n    vector[N] y;\n    y[N] = -x[N+1] * sqrt(1 + 1. / N);\n    real sum_w = 0;\n    for (n in 1:(N-1)) {\n      int i = N - n;\n      int i_p_1 = i + 1;\n      real w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + 1));\n      sum_w += w;\n      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;\n    }\n    return y;\n  }\n\n\n\nThe inverse transform follows the isometric logratio tranform. It maps an unconstrained vector \\(y \\in \\mathbb{R}^N\\) to a sum-to-zero vector \\(x \\in \\mathbb{R}^{N + 1}\\) such that \\[\n\\sum_{n=1}^{N + 1} x_n = 0.\n\\] The values are defined inductively, starting with \\[\nx_1 = \\sum_{n=1}^N \\frac{y_n}{\\sqrt{n (n + 1)}}\n\\] and then setting \\[\nx_{n + 1} = \\sum_{i = n + 1}^N \\frac{y_i}{\\sqrt{i (i + 1)}}\n- n \\cdot \\frac{y_n}{\\sqrt{n (n + 1)}}.\n\\] for \\(n \\in 1{:}N\\).\nThe definition is such that \\[\n\\sum_{n = 1}^{N + 1} x_n = 0\n\\] by construction, because each of the terms added to \\(x_{n}\\) is then subtracted from \\(x_{n + 1}\\) the number of times it shows up in earlier terms.\n\n\n\nThe inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.\nThe sum-to-zero inverse transform is expressed within Stan as:\n  vector manual_sum_to_zero_inverse_jacobian(vector y) {\n    int N = num_elements(y);\n    vector[N + 1] x = zeros_vector(N + 1);\n    real sum_w = 0;\n    for (n in 1:N) {\n      int i = N - n + 1;\n      real w = y[i] * inv_sqrt(i * (i + 1));\n      sum_w += w;\n      x[i] += sum_w;\n      x[i + 1] -= w * i;\n    }\n    return x;\n  }\nNote that there is no target += increment because the Jacobian is zero.\n\n\n\nThe matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any \\(N\\)-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.\nLet the unconstrained matrix be \\[\n\\mathcal Y \\in \\mathbb R^{n_1 \\times n_2}\n\\] and the zero sum vector transform as \\[\n\\mathbf z = \\mathcal C_n(\\mathbf y)\\;=\\;\n\\begin{bmatrix}H_n\\\\[2pt]-\\mathbf 1_{1\\times d}\\end{bmatrix}\\mathbf y\n\\;\\in\\mathbb R^{n+1},\n\\] where \\(H_n\\in\\mathbb R^{n \\times n}\\) is the orthogonal Helmert matrix and satisfies \\(\\mathbf 1_{1\\times d}A_d^{\\!\\top}=0\\).\nApplying \\(C_n\\) to each slice results in\n\\[\n\\mathcal Z \\;=\\;\n\\mathcal X\n\\times_1 \\bigl[\\mathcal C_{n_1}\\bigr]\n\\times_2 \\bigl[\\mathcal C_{n_2}\\bigr]\n\\] where \\[\n\\mathcal Z \\in\n\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}.\n\\]\nBecause each \\(\\mathbb R^{d_1\\times\\cdots\\times d_N}\\) is invertible on the \\(\\mathbf 1^\\perp\\) subspace the composite map applied to \\(\\mathcal Z\\) is a linear bijection between \\(\\mathbb R^{n_1\\times n_2}\\) and the codomain \\(\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}\\).",
+    "text": "Complex valued linear regression with complex predictors and regression coefficients looks just like standard regression. For example, if we take x to be predictors, y to be an array of outcomes. For example, consider the following complete Stan program for an intercept and slope.\ndata {\n  int&lt;lower=0&gt; N;\n  complex_vector[N] x;\n  complex_vector[N] y;\n}\nparameters {\n  complex alpha;\n  complex beta;\n}\nmodel {\n  complex_vector[N] eps = y - (alpha + beta * x);\n  eps ~  // ...error distribution...\n}\nThe question remains of how to fill in the error distribution and there are several alternatives. We consider only two simple alternatives, and do not consider penalizing the absolute value of the error.\n\n\nThe simplest approach to error in complex regression is to give the real and imaginary parts of eps_n independent independent normal distributions, as follows.\nparameters {\n  // ...\n  vector[2] sigma;\n}\n// ...\nmodel {\n  // ...\n  get_real(eps) ~ normal(0, sigma[1]);\n  get_imag(eps) ~ normal(0, sigma[2]);\n  sigma ~ //...hyperprior...\n}\nA new error scale vector sigma is introduced, and it should itself get a prior based on the expected scale of error for the problem.\n\n\n\nThe next simplest approach is to treat the real and imaginary parts of the complex number as having a multivariate normal prior. This can be done by adding a parameter for correlation to the above, or just working with a multivariate covariance matrix, as we do here.\nparameters {\n  cholesky_factor_corr[2] L_Omega;  // correlation matrix\n  vector[2] sigma;                  // real, imag error scales\n  // ...\n}\n// ...\nmodel {\n  array[N] vector[2] eps_arr;\n  for (n in 1:N) {\n    eps_arr[n] = { to_real(eps[n]), to_imag(eps[n]) };\n  }\n  eps_arr ~ multi_normal_cholesky([0, 0]',\n                                  diag_pre_multiply(sigma, L_Omega));\n  L_Omega ~ lkj_cholesky(4);  // shrink toward diagonal correlation\n  sigma ~ // ... hyperprior ...\n}\nHere, the real and imaginary components of the error get a joint distribution with correlation and independent scales. The error gets a multivariate normal distribution with zero mean and a Cholesky factor representation of covariance, consisting of a scale vector sigma and a Cholesky factor or a correlation matrix, L_Omega. The prior on the correlations is concentrated loosely around diagonal covariance, and the prior on the scales is left open. In order to vectorize the call to multi_normal_cholesky, the vector of complex numbers needs to be converted to an array of size 2 vectors.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Example Models",
+      "Complex Numbers"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#simplex-transform.section",
-    "href": "reference-manual/transforms.html#simplex-transform.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/bootstrap.html",
+    "href": "stan-users-guide/bootstrap.html",
+    "title": "The Bootstrap and Bagging",
     "section": "",
-    "text": "Variables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).\nThe unit \\(K\\)-simplex is the set of points \\(x \\in \\mathbb{R}^K\\) such that for \\(1 \\leq k \\leq K\\),\n\\[\nx_k &gt; 0,\n\\]\nand\n\\[\n\\sum_{k=1}^K x_k = 1.\n\\]\nAn alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values \\((0,1)\\), and \\((1,0)\\) and the unit 2-simplex is the line connecting these two points; values such as \\((0.3,0.7)\\) and \\((0.99,0.01)\\) lie on the line. In 3-dimensions, the basis is \\((0,0,1)\\), \\((0,1,0)\\) and \\((1,0,0)\\) and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include \\((0.5,0.5,0)\\), \\((0.2,0.7,0.1)\\) and all other triplets of non-negative values summing to 1.\nAs these examples illustrate, the simplex always picks out a subspace of \\(K-1\\) dimensions from \\(\\mathbb{R}^K\\). Therefore a point \\(x\\) in the \\(K\\)-simplex is fully determined by its first \\(K-1\\) elements \\(x_1, x_2,\n\\ldots, x_{K-1}\\), with\n\\[\nx_K = 1 - \\sum_{k=1}^{K-1} x_k.\n\\]\n\n\nThe length-\\(K\\) unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length \\(K\\).\nLet \\(y\\) represent the unconstrained \\(K - 1\\) values in \\((-\\infty, \\infty)\\). The intermediate sum-to-zero vector \\(z = \\text{sum\\_to\\_zero\\_transform}(y)\\) is length \\(K\\). The unit simplex is then given by \\[\nx_i = \\text{softmax}(z) = \\frac{\\exp(z_i)}{\\sum_{i = 1}^K \\exp(z_i)}\n\\]\nThe sum-to-zero vector transform is described in further detail at the sum-to-zero vector section of the Reference Manual.\n\n\n\n\n\n\nNote\n\n\n\nAll versions of Stan pre-2.37 used the stick-breaking transform. This is documented at Stan 2.36 Reference Manual: Simplex Transform.\n\n\n\n\nThe Jacobian \\(J\\) of the inverse unit-simplex transform is found by restricting \\(J\\) to the subspace spanned by the sum-to-zero vector \\(z\\). The Jacobian is given as the \\((K - 1) \\times (K - 1)\\) matrix \\(J\\) where\n\\[\nJ_{ij} = \\frac{\\partial x_i}{\\partial z_j} =\n\\frac{\\partial}{\\partial z_i} \\left( \\frac{\\exp(z_i)}{{\\sum_{i = 1}^K \\exp(z_i)}} \\right)\n\\] and \\(i,j \\in 1, \\ldots, K - 1\\).\nThe diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification\n\\[\nJ_{ij} =\n\\begin{cases}\nx_i (1 - x_i), & \\text{if } i = j, \\\\\n-x_i x_j, & \\text{if } i \\neq j.\n\\end{cases}\n\\]\nIn matrix form this can be expressed as\n\\[\nJ = \\text{diag}(x) - x x^\\top\n\\]\nThe determinant of this matrix can be found using the Matrix Determinant Lemma:\n\\[\n\\det\\bigl(A + u v^{\\top}\\bigr)\n=\n\\det(A)\\,\\bigl(1 + v^{\\top}A^{-1}u\\bigr).\n\\]\nHere,\n\\[\nA \\;=\\; \\operatorname{diag}(x_{1},\\ldots, x_{K-1}),\n\\quad\nu \\;=\\; -\\bigl(x_1,\\ldots, x_{K-1}\\bigr)^{\\!\\top},\n\\quad\nv \\;=\\; \\bigl(x_{1}, \\ldots, x_{K-1}\\bigr)^{\\!\\top}.\n\\] Therefore,\n\\[\n\\begin{aligned}\n\\det(J)\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_i \\bigg)\n\\bigg(1 + (x_{1},\\ldots, x_{K-1})\\,\\mathrm{diag}\\bigl(x_{1}^{-1},\\ldots,x_{K-1}^{-1}\\bigr)\\,\n\\big(-x_{1},\\ldots,-x_{K-1}\\big)^{\\top}\n\\bigg) \\\\\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg)\n\\bigg(1 - \\sum_{i=1}^{K-1} x_{i}\\bigg)\n=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg) x_{K} \\\\\n&=\n\\prod_{i=1}^{K} x_{i}.\n\\end{aligned}\n\\]\n\n\n\n\nThe transform \\(Y = f(X)\\) can be derived by reversing the stages of the inverse transform,\n\\[\ny_k\n= H^\\top \\bigg(\\log(x_k)\n- \\frac{1}{K}\\sum_{i=1}^K\\log(x_i) \\bigg)\n.\n\\]\nThe matrix \\(H\\) is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.",
+    "text": "The bootstrap is a technique for approximately sampling from the error distribution for an estimator. Thus it can be used as a Monte Carlo method to estimate standard errors and confidence intervals for point estimates (Efron and Tibshirani 1986; 1994). It works by subsampling the original data and computing sample estimates from the subsample. Like other Monte Carlo methods, the bootstrap is plug-and-play, allowing great flexibility in both model choice and estimator.\nBagging is a technique for combining bootstrapped estimators for model criticism and more robust inference (Breiman 1996; Huggins and Miller 2019).\n\n\n\n\nAn estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set \\(y_{1,\\ldots, N}\\) to a number by \\[\n\\textrm{mean}(y) = \\frac{1}{N} \\sum_{n=1}^N y_n,\n\\] and hence meets the definition of an estimator. Given the likelihood function \\[\np(y \\mid \\mu) = \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, 1),\n\\] the mean is the maximum likelihood estimator,\n\\[\n\\textrm{mean}(y) = \\textrm{arg max}_{\\mu} \\ p(y \\mid \\mu, 1)\n\\] A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.\nIn analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.\n\n\n\nThe bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.\nThe following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate \\(\\hat{\\theta}\\) that is a function of data \\(y\\).\nfor (m in 1:M) {\n  y_rep[m] &lt;- sample_uniform(y)\n  theta_hat[m] &lt;- estimate_theta(y_rep[m])\n}\nstd_error = sd(theta_hat)\nconf_95pct = [ quantile(theta_hat, 0.025),\n               quantile(theta_hat, 0.975) ]\nThe sample_uniform function works by independently assigning each element of y_rep an element of y drawn uniformly at random. This produces a sample with replacement. That is, some elements of y may show up more than once in y_rep and some may not appear at all.\n\n\n\n\nThe bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector x of predictors in addition to outcomes y. To allow a single program to fit both the original data and random subsamples, the variable resample is set to 1 to resample and 0 to use the original data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=1&gt; resample;\n}\ntransformed data {\n  simplex[N] uniform = rep_vector(1.0 / N, N);\n  array[N] int&lt;lower=1, upper=N&gt; boot_idxs;\n  for (n in 1:N) {\n    boot_idxs[n] = resample ? categorical_rng(uniform) : n;\n  }\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);\n}\nThe model accepts data in the usual form for a linear regression as a number of observations \\(N\\) with a size \\(N\\) vector \\(x\\) of predictors and a size \\(N\\) vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of boot_idxs from 1:N, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array boot_idxs is defined to be the sequence 1:N, because x == x[1:N] and y = y[1:N].\nFor example, when resample == 1, if \\(N = 4,\\) the value of boot_idxs might be {2, 1, 1, 3}, resulting in a bootstrap sample {y[2], y[1], y[1], y[3]} with the first element repeated twice and the fourth element not sampled at all.\nThe parameters are the usual regression coefficients for the intercept alpha, slope beta, and error scale sigma. The model uses the bootstrap index variable boot_idx to index the predictors as x[boot_idx] and outcomes as y[boot_idx]. This generates a new size-\\(N\\) vector whose entries are defined by x[boot_idx][n] = x[boot_idx[n]] and similarly for y. For example, if \\(N = 4\\) and boot_idxs = {2, 1, 1, 3}, then x[boot_idxs] = [x[2], x[1], x[1], x[3]]' and y[boot_idxs] = [y[2], y[1], y[1], y[3]]'. The predictor and outcome vectors remain aligned, with both elements of the pair x[1] and y[1] repeated twice.\nWith the model defined this way, if resample is 1, the model is fit to a bootstrap subsample of the data. If resample is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.\n\n\n\nRunning the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.\nTo estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data \\(y\\), it is usually not worth calculating bootstrap estimates to high precision.\n\n\nHere’s the result of calculating standard errors for the linear regression model above with \\(N = 50\\) data points, \\(\\alpha = 1.2, \\beta\n= -0.5,\\) and \\(\\sigma = 1.5.\\) With a total of \\(M = 100\\) bootstrap samples, there are 100 estimates of \\(\\alpha\\), 100 of \\(\\beta\\), and 100 of \\(\\sigma\\). These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for \\(\\alpha\\) provide the bootstrap estimate of the standard error in the estimate for \\(\\alpha\\). Here’s what it looks like for the above model with \\(M =\n100\\)\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.218\n      beta     -0.610      0.204\n     sigma      1.537      0.142\nWith the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.206\n     alpha      1.359      0.240\n     alpha      1.359      0.234\n     alpha      1.359      0.249\n     alpha      1.359      0.227\nIncreasing \\(M\\) will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample \\(y\\).\n\n\n\nAs usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are \\(M = 1000\\) estimates of \\(\\hat{\\alpha}\\) in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set \\(y\\). To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).\n\n\n\n\nWhen bootstrapping is carried through inference it is known as bootstrap aggregation, or bagging, in the machine-learning literature (Breiman 1996). In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate \\(\\hat{\\sigma}\\) from the original data set, bootstrapped data sets \\(y^{\\textrm{boot}(1)}, \\ldots,\ny^{\\textrm{boot}(N)}\\) are generated. Each is used to generate an estimate \\(\\hat{\\sigma}^{\\textrm{boot}(n)}.\\) The final estimate is \\[\n\\hat{\\sigma} = \\frac{1}{N} \\sum_{n = 1}^N \\hat{\\sigma}^{\\textrm{boot}(n)}.\n\\] The same would be done to estimate a predictive quantity \\(\\tilde{y}\\) for as yet unseen data. \\[\n\\hat{\\tilde{y}} = \\frac{1}{N} \\sum_{n = 1}^N\n\\hat{\\tilde{y}}^{\\textrm{boot}(n)}.\n\\] For discrete parameters, voting is used to select the outcome.\nOne way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.\n\n\n\nA Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.\n(Huggins and Miller 2019) use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#stochastic-matrix-transform.section",
-    "href": "reference-manual/transforms.html#stochastic-matrix-transform.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/bootstrap.html#the-bootstrap",
+    "href": "stan-users-guide/bootstrap.html#the-bootstrap",
+    "title": "The Bootstrap and Bagging",
     "section": "",
-    "text": "The column_stochastic_matrix[N, M] and row_stochastic_matrix[M, N] type in Stan represents an \\(N \\times M\\) matrix where each column (row) is a unit simplex of dimension \\(N\\). In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.\n\n\nA column stochastic matrix \\(X \\in \\mathbb{R}^{N \\times M}\\) is defined such that each column is a simplex. For column \\(m\\) (where \\(1 \\leq m \\leq M\\)):\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{n=1}^N X_{n, m} = 1.\n\\]\nA row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e. the rows of the matrix are simplexes)\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{m=1}^N X_{n, m} = 1.\n\\]\nThis definition ensures that each column (row) of the matrix \\(X\\) lies on the \\(N-1\\) dimensional unit simplex, similar to the simplex[N] type, but extended across multiple columns(rows).\n\n\n\nFor the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).\n\n\n\nThe Jacobian determinant of the inverse transform for each column \\(m\\) in the matrix is given by the product of the diagonal entries \\(J_{n, m}\\) of the lower-triangular Jacobian matrix. This determinant is calculated as:\n\\[\n\\left| \\det J_m \\right| = \\prod_{n=1}^{N-1} \\left( z_{n, m} (1 - z_{n, m}) \\left( 1 - \\sum_{n'=1}^{n-1} X_{n', m} \\right) \\right).\n\\]\nThus, the overall Jacobian determinant for the entire column_stochastic_matrix and row_stochastic_matrix is the product of the determinants for each column (row):\n\\[\n\\left| \\det J \\right| = \\prod_{m=1}^{M} \\left| \\det J_m \\right|.\n\\]\n\n\n\nFor the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).",
+    "text": "An estimator is nothing more than a function mapping a data set to one or more numbers, which are called “estimates”. For example, the mean function maps a data set \\(y_{1,\\ldots, N}\\) to a number by \\[\n\\textrm{mean}(y) = \\frac{1}{N} \\sum_{n=1}^N y_n,\n\\] and hence meets the definition of an estimator. Given the likelihood function \\[\np(y \\mid \\mu) = \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, 1),\n\\] the mean is the maximum likelihood estimator,\n\\[\n\\textrm{mean}(y) = \\textrm{arg max}_{\\mu} \\ p(y \\mid \\mu, 1)\n\\] A Bayesian approach to point estimation would be to add a prior and use the posterior mean or median as an estimator. Alternatively, a penalty function could be added to the likelihood so that optimization produces a penalized maximum likelihood estimate. With any of these approaches, the estimator is just a function from data to a number.\nIn analyzing estimators, the data set is being modeled as a random variable. It is assumed that the observed data is just one of many possible random samples of data that may have been produced. If the data is modeled a random variable, then the estimator applied to the data is also a random variable. The simulations being done for the bootstrap are attempts to randomly sample replicated data sets and compute the random properties of the estimators using standard Monte Carlo methods.\n\n\n\nThe bootstrap works by applying an estimator to replicated data sets. These replicates are created by subsampling the original data with replacement. The sample quantiles may then be used to estimate standard errors and confidence intervals.\nThe following pseudocode estimates 95% confidence intervals and standard errors for a generic estimate \\(\\hat{\\theta}\\) that is a function of data \\(y\\).\nfor (m in 1:M) {\n  y_rep[m] &lt;- sample_uniform(y)\n  theta_hat[m] &lt;- estimate_theta(y_rep[m])\n}\nstd_error = sd(theta_hat)\nconf_95pct = [ quantile(theta_hat, 0.025),\n               quantile(theta_hat, 0.975) ]\nThe sample_uniform function works by independently assigning each element of y_rep an element of y drawn uniformly at random. This produces a sample with replacement. That is, some elements of y may show up more than once in y_rep and some may not appear at all.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#unit-vector.section",
-    "href": "reference-manual/transforms.html#unit-vector.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/bootstrap.html#coding-the-bootstrap-in-stan",
+    "href": "stan-users-guide/bootstrap.html#coding-the-bootstrap-in-stan",
+    "title": "The Bootstrap and Bagging",
     "section": "",
-    "text": "An \\(n\\)-dimensional vector \\(x \\in \\mathbb{R}^n\\) is said to be a unit vector if it has unit Euclidean length, so that\n\\[\n\\Vert x \\Vert\n\\ = \\ \\sqrt{x^{\\top}\\,x}\n\\ = \\ \\sqrt{x_1^2 + x_2^2 + \\cdots + x_n^2}\n\\ = \\ 1\\ .\n\\]\n\n\nStan divides an unconstrained vector \\(y \\in \\mathbb{R}^{n}\\) by its norm, \\(\\Vert y \\Vert = \\sqrt{y^\\top y}\\), to obtain a unit vector \\(x\\),\n\\[\nx = \\frac{y}{\\Vert y \\Vert}.\n\\]\nTo generate a unit vector, Stan generates points at random in \\(\\mathbb{R}^n\\) with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. Muller (1959) showed this generates points uniformly at random on \\(S^{n-1}\\). That is, if we draw \\(y_n \\sim \\mathsf{Normal}(0, 1)\\) for \\(n \\in 1{:}n\\), then \\(x = \\frac{y}{\\Vert y \\Vert}\\) has a uniform distribution over \\(S^{n-1}\\). This allows us to use an \\(n\\)-dimensional basis for \\(S^{n-1}\\) that preserves local neighborhoods in that points that are close to each other in \\(\\mathbb{R}^n\\) map to points near each other in \\(S^{n-1}\\). The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in \\(\\mathbb{R}^n\\) that map to identical points in \\(S^{n-1}\\).\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nThe Jacobian matrix relating the input vector \\(y\\) to the output vector \\(x\\) is singular because \\(x^\\top x = 1\\) for any non-zero input vector \\(y\\). Thus, there technically is no unique transformation from \\(x\\) to \\(y\\). To circumvent this issue, let \\(r = \\sqrt{y^\\top y}\\) so that \\(y = r\nx\\). The transformation from \\(\\left(r, x_{-n}\\right)\\) to \\(y\\) is well-defined but \\(r\\) is arbitrary, so we set \\(r = 1\\). In this case, the determinant of the Jacobian is proportional to \\(e^{-\\frac{1}{2} y^\\top y}\\), which is the kernel of a standard multivariate normal distribution with \\(n\\) independent dimensions.",
+    "text": "The bootstrap procedure can be coded quite generally in Stan models. The following code illustrates a Stan model coding the likelihood for a simple linear regression. There is a parallel vector x of predictors in addition to outcomes y. To allow a single program to fit both the original data and random subsamples, the variable resample is set to 1 to resample and 0 to use the original data.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n  int&lt;lower=0, upper=1&gt; resample;\n}\ntransformed data {\n  simplex[N] uniform = rep_vector(1.0 / N, N);\n  array[N] int&lt;lower=1, upper=N&gt; boot_idxs;\n  for (n in 1:N) {\n    boot_idxs[n] = resample ? categorical_rng(uniform) : n;\n  }\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y[boot_idxs] ~ normal(alpha + beta * x[boot_idxs], sigma);\n}\nThe model accepts data in the usual form for a linear regression as a number of observations \\(N\\) with a size \\(N\\) vector \\(x\\) of predictors and a size \\(N\\) vector of outcomes. The transformed data block generates a set of indexes into the data that is the same size as the data. This is done by independently sampling each entry of boot_idxs from 1:N, using a discrete uniform distribution coded as a categorical random number generator with an equal chance for each outcome. If resampling is not done, the array boot_idxs is defined to be the sequence 1:N, because x == x[1:N] and y = y[1:N].\nFor example, when resample == 1, if \\(N = 4,\\) the value of boot_idxs might be {2, 1, 1, 3}, resulting in a bootstrap sample {y[2], y[1], y[1], y[3]} with the first element repeated twice and the fourth element not sampled at all.\nThe parameters are the usual regression coefficients for the intercept alpha, slope beta, and error scale sigma. The model uses the bootstrap index variable boot_idx to index the predictors as x[boot_idx] and outcomes as y[boot_idx]. This generates a new size-\\(N\\) vector whose entries are defined by x[boot_idx][n] = x[boot_idx[n]] and similarly for y. For example, if \\(N = 4\\) and boot_idxs = {2, 1, 1, 3}, then x[boot_idxs] = [x[2], x[1], x[1], x[3]]' and y[boot_idxs] = [y[2], y[1], y[1], y[3]]'. The predictor and outcome vectors remain aligned, with both elements of the pair x[1] and y[1] repeated twice.\nWith the model defined this way, if resample is 1, the model is fit to a bootstrap subsample of the data. If resample is 0, the model is fit to the original data as given. By running the bootstrap fit multiple times, confidence intervals can be generated from quantiles of the results.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#correlation-matrix-transform.section",
-    "href": "reference-manual/transforms.html#correlation-matrix-transform.section",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/bootstrap.html#error-statistics-from-the-bootstrap",
+    "href": "stan-users-guide/bootstrap.html#error-statistics-from-the-bootstrap",
+    "title": "The Bootstrap and Bagging",
     "section": "",
-    "text": "A \\(K \\times K\\) correlation matrix \\(x\\) must be symmetric, so that\n\\[\nx_{k,k'} = x_{k',k}\n\\]\nfor all \\(k,k' \\in \\{ 1, \\ldots, K \\}\\), it must have a unit diagonal, so that\n\\[\nx_{k,k} = 1\n\\]\nfor all \\(k \\in \\{ 1, \\ldots, K \\}\\), and it must be positive definite, so that for every non-zero \\(K\\)-vector \\(a\\),\n\\[\na^{\\top} x a &gt; 0.\n\\]\nThe number of free parameters required to specify a \\(K \\times K\\) correlation matrix is \\(\\binom{K}{2}\\).\nThere is more than one way to map from \\(\\binom{K}{2}\\) unconstrained parameters to a \\(K \\times K\\) correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform Lewandowski, Kurowicka, and Joe (2009).\n\n\nIt is easiest to specify the inverse, going from its \\(\\binom{K}{2}\\) parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose \\(y\\) is a vector containing \\(\\binom{K}{2}\\) unconstrained values. These are first transformed via the bijective function \\(\\tanh : \\mathbb{R} \\rightarrow\n(-1, 1)\\)\n\\[\n\\tanh y = \\frac{\\exp(2y) - 1}{\\exp(2y) + 1}.\n\\]\nThen, define a \\(K \\times K\\) matrix \\(z\\), the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the \\(4 \\times 4\\) case, there are \\(\\binom{4}{2}\\) values arranged as\n\\[\nz\n=\n\\left[\n\\begin{array}{cccc}\n1 & \\tanh y_1 & \\tanh y_2 & \\tanh y_4\n\\\\\n0 & 1 & \\tanh y_3 & \\tanh y_5\n\\\\\n0 & 0 & 1 & \\tanh y_6\n\\\\\n0 & 0 & 0 & 1\n\\end{array}\n\\right]\n.\n\\]\nLewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array \\(z\\) to a correlation matrix \\(x\\). The entry \\(z_{i,j}\\) for \\(i &lt;\nj\\) is interpreted as the canonical partial correlation (CPC) between \\(i\\) and \\(j\\), which is the correlation between \\(i\\)’s residuals and \\(j\\)’s residuals when both \\(i\\) and \\(j\\) are regressed on all variables \\(i'\\) such that \\(i'&lt; i\\). In the case of \\(i=1\\), there are no earlier variables, so \\(z_{1,j}\\) is just the Pearson correlation between \\(i\\) and \\(j\\).\nIn Stan, the LKJ transform is reformulated in terms of a Cholesky factor \\(w\\) of the final correlation matrix, defined for \\(1 \\leq i,j \\leq K\\) by\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\n\\prod_{i'=1}^{i - 1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{if } 1 &lt; i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\\\nz_{i,j} \\, \\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThis does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n\\frac{z_{i,j}}{z_{i-1,j}} \\ w_{i-1,j} \\left( 1 - z_{i-1,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i \\leq j.\n\\end{array}\n\\right.\n\\]\nGiven the upper-triangular Cholesky factor \\(w\\), the final correlation matrix is\n\\[\nx = w^{\\top} w.\n\\]\nLewandowski, Kurowicka, and Joe (2009) show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as\n\\[\n\\mbox{det} \\, x = \\prod_{i=1}^{K-1} \\ \\prod_{j=i+1}^K \\ (1 - z_{i,j}^2)\n= \\prod_{1 \\leq i &lt; j \\leq K} (1 - z_{i,j}^2),\n\\]\n\n\n\nFrom the inverse of equation 11 in (Lewandowski, Kurowicka, and Joe 2009), the absolute Jacobian determinant is\n\\[\n\\sqrt{\\prod_{i=1}^{K-1}\\prod_{j=i+1}^K \\left(1-z_{i,j}^2\\right)^{K-i-1}} \\\n\\times \\prod_{i=1}^{K-1}\\prod_{j=i+1}^K\n\\frac{\\partial z_{i,j}}{\\partial y_{i,j}}\n\\]\n\n\n\nThe correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.\nStarting with a correlation matrix \\(x\\), the first step is to find the unique upper triangular \\(w\\) such that \\(x = w w^{\\top}\\). Because \\(x\\) is positive definite, this can be done by applying the Cholesky decomposition,\n\\[\nw = \\mbox{chol}(x).\n\\]\nThe next step from the Cholesky factor \\(w\\) back to the array \\(z\\) of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of \\(w\\), which when inverted yields\n\\[\nz_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i \\leq j,\n\\\\\nw_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n{w_{i,j}}\n\\\n\\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,j}^2 \\right)^{-1/2}\n& \\mbox{if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThe final stage of the transform reverses the hyperbolic tangent transform, which is defined by\n\\[\ny = \\tanh^{-1} z = \\frac{1}{2} \\log \\left( \\frac{1 + z}{1 - z} \\right).\n\\]\nThe inverse hyperbolic tangent function, \\(\\tanh^{-1}\\), is also called the Fisher transformation.",
+    "text": "Running the model multiple times produces a Monte Carlo sample of estimates from multiple alternative data sets subsampled from the original data set. The error distribution is just the distribution of the bootstrap estimates minus the estimate for the original data set.\nTo estimate standard errors and confidence intervals for maximum likelihood estimates the Stan program is executed multiple times using optimization (which turns off Jacobian adjustments for constraints and finds maximum likelihood estimates). On the order of one hundred replicates is typically enough to get a good sense of standard error; more will be needed to accurate estimate the boundaries of a 95% confidence interval. On the other hand, given that there is inherent variance due to sampling the original data \\(y\\), it is usually not worth calculating bootstrap estimates to high precision.\n\n\nHere’s the result of calculating standard errors for the linear regression model above with \\(N = 50\\) data points, \\(\\alpha = 1.2, \\beta\n= -0.5,\\) and \\(\\sigma = 1.5.\\) With a total of \\(M = 100\\) bootstrap samples, there are 100 estimates of \\(\\alpha\\), 100 of \\(\\beta\\), and 100 of \\(\\sigma\\). These are then treated like Monte Carlo draws. For example, the sample standard deviation of the draws for \\(\\alpha\\) provide the bootstrap estimate of the standard error in the estimate for \\(\\alpha\\). Here’s what it looks like for the above model with \\(M =\n100\\)\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.218\n      beta     -0.610      0.204\n     sigma      1.537      0.142\nWith the data set fixed, these estimates of standard error will display some Monte Carlo error. For example, here are the standard error estimates from five more runs holding the data the same, but allowing the subsampling to vary within Stan:\n parameter   estimate    std err\n ---------   --------    -------\n     alpha      1.359      0.206\n     alpha      1.359      0.240\n     alpha      1.359      0.234\n     alpha      1.359      0.249\n     alpha      1.359      0.227\nIncreasing \\(M\\) will reduce Monte Carlo error, but this is not usually worth the extra computation time as there is so much other uncertainty due to the original data sample \\(y\\).\n\n\n\nAs usual with Monte Carlo methods, confidence intervals are estimated using quantiles of the draws. That is, if there are \\(M = 1000\\) estimates of \\(\\hat{\\alpha}\\) in different subsamples, the 2.5% quantile and 97.5% quantile pick out the boundaries of the 95% confidence interval around the estimate for the actual data set \\(y\\). To get accurate 97.5% quantile estimates requires a much larger number of Monte Carlo simulations (roughly twenty times as large as needed for the median).",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#covariance-matrices",
-    "href": "reference-manual/transforms.html#covariance-matrices",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/bootstrap.html#bagging",
+    "href": "stan-users-guide/bootstrap.html#bagging",
+    "title": "The Bootstrap and Bagging",
     "section": "",
-    "text": "A \\(K \\times K\\) matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires \\(K + \\binom{K}{2}\\) free parameters to specify a \\(K \\times K\\) covariance matrix.\n\n\nStan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.1\nIf \\(x\\) is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix \\(z =\n\\mathrm{chol}(x)\\) with positive diagonal entries, called a Cholesky factor, such that\n\\[\nx = z \\, z^{\\top}.\n\\]\nThe off-diagonal entries of the Cholesky factor \\(z\\) are unconstrained, but the diagonal entries \\(z_{k,k}\\) must be positive for \\(1 \\leq k\n\\leq K\\).\nTo complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix \\(y\\) defined by\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log z_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nz_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular \\(K \\times K\\) matrix \\(y\\), the first step is to recover the intermediate matrix \\(z\\) by reversing the log transform,\n\\[\nz_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe covariance matrix \\(x\\) is recovered from its Cholesky factor \\(z\\) by taking\n\\[\nx = z \\, z^{\\top}.\n\\]\n\n\n\nThe Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix \\(y\\) to matrix \\(z\\) with positive diagonals and the product transform from the Cholesky factor \\(z\\) to \\(x\\).\nThe transform from unconstrained \\(y\\) to Cholesky factor \\(z\\) has a diagonal Jacobian matrix, the absolute determinant of which is thus\n\\[\n\\prod_{k=1}^K  \\frac{\\partial}{\\partial_{y_{k,k}}} \\, \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K z_{k,k}.\n\\]\nThe Jacobian matrix of the second transform from the Cholesky factor \\(z\\) to the covariance matrix \\(x\\) is also triangular, with diagonal entries corresponding to pairs \\((m,n)\\) with \\(m \\geq n\\), defined by\n\\[\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( z \\, z^{\\top} \\right)_{m,n}\n\\ = \\\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( \\sum_{k=1}^K z_{m,k} \\, z_{n,k} \\right)\n\\ = \\\n\\left\\{\n\\begin{array}{cl}\n2 \\, z_{n,n} & \\mbox{if } m = n \\mbox{ and }\n\\\\\nz_{n,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe absolute Jacobian determinant of the second transform is thus\n\\[\n2^{K} \\ \\prod_{m = 1}^{K} \\ \\prod_{n=1}^{m} z_{n,n}\n\\ = \\\n\\prod_{n=1}^K \\ \\prod_{m=n}^K z_{n,n}\n\\ = \\\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}.\n\\]\nFinally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular \\(y\\) to a symmetric, positive definite matrix \\(x\\) is the product of the Jacobian determinants of the exponentiation and product transforms,\n\\[\n\\left( \\prod_{k=1}^K z_{k,k} \\right)\n\\left(\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}\n\\right)\n\\ = \\\n2^K\n\\, \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\nLet \\(f^{-1}\\) be the inverse transform from a \\(K + \\binom{K}{2}\\)-vector \\(y\\) to the \\(K \\times K\\) covariance matrix \\(x\\). A density function \\(p_X(x)\\) defined on \\(K \\times K\\) covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(K + \\binom{K}{2}\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\ 2^K \\ \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]",
+    "text": "When bootstrapping is carried through inference it is known as bootstrap aggregation, or bagging, in the machine-learning literature (Breiman 1996). In the simplest case, this involves bootstrapping the original data, fitting a model to each bootstrapped data set, then averaging the predictions. For instance, rather than using an estimate \\(\\hat{\\sigma}\\) from the original data set, bootstrapped data sets \\(y^{\\textrm{boot}(1)}, \\ldots,\ny^{\\textrm{boot}(N)}\\) are generated. Each is used to generate an estimate \\(\\hat{\\sigma}^{\\textrm{boot}(n)}.\\) The final estimate is \\[\n\\hat{\\sigma} = \\frac{1}{N} \\sum_{n = 1}^N \\hat{\\sigma}^{\\textrm{boot}(n)}.\n\\] The same would be done to estimate a predictive quantity \\(\\tilde{y}\\) for as yet unseen data. \\[\n\\hat{\\tilde{y}} = \\frac{1}{N} \\sum_{n = 1}^N\n\\hat{\\tilde{y}}^{\\textrm{boot}(n)}.\n\\] For discrete parameters, voting is used to select the outcome.\nOne way of viewing bagging is as a classical attempt to get something like averaging over parameter estimation uncertainty.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#cholesky-factors-of-covariance-matrices",
-    "href": "reference-manual/transforms.html#cholesky-factors-of-covariance-matrices",
-    "title": "Constraint Transforms",
+    "objectID": "stan-users-guide/bootstrap.html#bayesian-bootstrap-and-bagging",
+    "href": "stan-users-guide/bootstrap.html#bayesian-bootstrap-and-bagging",
+    "title": "The Bootstrap and Bagging",
     "section": "",
-    "text": "An \\(M \\times M\\) covariance matrix \\(\\Sigma\\) can be Cholesky factored to a lower triangular matrix \\(L\\) such that \\(L\\,L^{\\top} = \\Sigma\\). If \\(\\Sigma\\) is positive definite, then \\(L\\) will be \\(M \\times M\\). If \\(\\Sigma\\) is only positive semi-definite, then \\(L\\) will be \\(M \\times N\\), with \\(N &lt; M\\).\nA matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and \\(M \\geq\nN\\). A matrix satisfying these conditions ensures that \\(L \\,\nL^{\\top}\\) is positive semi-definite if \\(M &gt; N\\) and positive definite if \\(M = N\\).\nA Cholesky factor of a covariance matrix requires \\(N + \\binom{N}{2} +\n(M - N)N\\) unconstrained parameters.\n\n\nStan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose \\(x\\) is an \\(M \\times N\\) Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log x_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nx_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform need only invert the logarithm with an exponentiation. If \\(y\\) is the unconstrained matrix representation, then the elements of the constrained matrix \\(x\\) is defined by\n\\[\nx_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe transform has a diagonal Jacobian matrix, the absolute determinant of which is\n\\[\n\\prod_{n=1}^N  \\frac{\\partial}{\\partial_{y_{n,n}}} \\, \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N x_{n,n}.\n\\]\nLet \\(x = f^{-1}(y)\\) be the inverse transform from a \\(N + \\binom{N}{2}\n+ (M - N)N\\) vector to an \\(M \\times N\\) Cholesky factor for a covariance matrix \\(x\\) defined in the previous section. A density function \\(p_X(x)\\) defined on \\(M \\times N\\) Cholesky factors of covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(N + \\binom{N}{2}\n+ (M - N)N\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\prod_{N=1}^N x_{n,n}.\n\\]",
+    "text": "A Bayesian estimator may be analyzed with the bootstrap in exactly the same way as a (penalized) maximum likelihood estimate. For example, the posterior mean and posterior median are two different Bayesian estimators. The bootstrap may be used estimate standard errors and confidence intervals, just as for any other estimator.\n(Huggins and Miller 2019) use the bootstrap to assess model calibration and fitting in a Bayesian framework and further suggest using bagged estimators as a guard against model misspecification. Bagged posteriors will typically have wider posterior intervals than those fit with just the original data, showing that the method is not a pure Bayesian approach to updating, and indicating it would not be calibrated if the model were well specified. The hope is that it can guard against over-certainty in a poorly specified model.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Constraint Transforms"
+      "Stan Users Guide",
+      "Posterior Inference & Model Checking",
+      "The Bootstrap and Bagging"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#cholesky-factors-of-correlation-matrices",
-    "href": "reference-manual/transforms.html#cholesky-factors-of-correlation-matrices",
-    "title": "Constraint Transforms",
+    "objectID": "reference-manual/whitespace.html",
+    "href": "reference-manual/whitespace.html",
+    "title": "Whitespace",
     "section": "",
-    "text": "A \\(K \\times K\\) correlation matrix \\(\\Omega\\) is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a \\(K \\times K\\) lower-triangular matrix \\(L\\) with positive diagonal elements such that \\(\\Omega = L\\,L^{\\top}\\). Because the correlation matrix has a unit diagonal,\n\\[\n\\Omega_{k,k} = L_k\\,L_k^{\\top} = 1,\n\\]\neach row vector \\(L_k\\) of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of \\(L\\) to be calculated from the off-diagonal elements, so that a Cholesky factor for a \\(K \\times K\\) correlation matrix requires only \\(\\binom{K}{2}\\) unconstrained parameters.\n\n\nIt is easiest to start with the inverse transform from the \\(\\binom{K}{2}\\) unconstrained parameters \\(y\\) to the \\(K \\times K\\) lower-triangular Cholesky factor \\(x\\). The inverse transform is based on the hyperbolic tangent function, \\(\\tanh\\), which satisfies \\(\\tanh(x) \\in (-1,1)\\). Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the section on correlation matrix transforms for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.\nSuppose \\(y\\) is a vector of \\(\\binom{K}{2}\\) unconstrained values. Let \\(z\\) be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the \\(3 \\times 3\\) case,\n\\[\nz =\n\\left[\n\\begin{array}{ccc}\n0 & 0 & 0\n\\\\\n\\tanh y_1 & 0 & 0\n\\\\\n\\tanh y_2 & \\tanh y_3 & 0\n\\end{array}\n\\right]\n\\]\nThe matrix \\(z\\), with entries in the range \\((-1, 1)\\), is then transformed to the Cholesky factor \\(x\\), by taking2\n\\[\nx_{i,j}\n=\n\\left\\{\n\\begin{array}{lll}\n0 & \\mbox{ if } i &lt; j & \\mbox{ [above diagonal]}\n\\\\\n\\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i = j & \\mbox{ [on diagonal]}\n\\\\\nz_{i,j} \\ \\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i &gt; j & \\mbox{ [below diagonal]}\n\\end{array}\n\\right.\n\\]\nIn the \\(3 \\times 3\\) case, this yields\n\\[\nx =\n\\left[\n\\begin{array}{ccc}\n1 & 0 & 0\n\\\\\nz_{2,1} & \\sqrt{1 - x_{2,1}^2} & 0\n\\\\\nz_{3,1} & z_{3,2} \\sqrt{1 - x_{3,1}^2}\n        & \\sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}\n\\end{array}\n\\right],\n\\]\nwhere the \\(z_{i,j} \\in (-1,1)\\) are the \\(\\tanh\\)-transformed \\(y\\).\nThe approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at \\(j=1\\), each below-diagonal entry \\(x_{i,j}\\) is determined by the (signed) fraction \\(z_{i,j}\\) of the remaining length for the row that it consumes. The diagonal entries \\(x_{i,i}\\) get any leftover length from earlier entries in their row. The above-diagonal entries are zero.\n\n\n\nSuppose \\(x\\) is a \\(K \\times K\\) Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values \\(z\\) from \\(x\\),\n\\[\nz_{i,j} = \\frac{x_{i,j}}{\\sqrt{1 - \\sum_{j' &lt; j}x_{i,j'}^2}}.\n\\]\nThe mapping from the resulting \\(z\\) to \\(y\\) inverts \\(\\tanh\\),\n\\[\ny\n\\ = \\\n\\tanh^{-1} z\n\\ = \\\n\\frac{1}{2} \\left( \\log (1 + z) - \\log (1 - z) \\right).\n\\]\n\n\n\nThe Jacobian of the full transform is the product of the Jacobians of its component transforms.\nFirst, for the inverse transform \\(z = \\tanh y\\), the derivative is\n\\[\n\\frac{d}{dy} \\tanh y = \\frac{1}{(\\cosh y)^2}.\n\\]\nSecond, for the inverse transform of \\(z\\) to \\(x\\), the resulting Jacobian matrix \\(J\\) is of dimension \\(\\binom{K}{2} \\times\n\\binom{K}{2}\\), with indexes \\((i,j)\\) for \\((i &gt; j)\\). The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each \\((i,j)\\) pair,\n\\[\n\\left| \\, \\mbox{det} \\, J \\, \\right|\n  \\ = \\ \\prod_{i &gt; j} \\left| \\frac{d}{dz_{i,j}} x_{i,j} \\right|,\n\\]\nwhere\n\\[\n\\frac{d}{dz_{i,j}} x_{i,j}\n= \\sqrt{1 - \\sum_{j' &lt; j} x^2_{i,j'}}.\n\\]\nSo the combined density for unconstrained \\(y\\) is\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n  \\ \\\n  \\prod_{n &lt; \\binom{K}{2}} \\frac{1}{(\\cosh y)^2}\n  \\ \\\n  \\prod_{i &gt; j} \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2\n  \\right)^{1/2},\n\\]\nwhere \\(x = f^{-1}(y)\\) is used for notational convenience. The log Jacobian determinant of the complete inverse transform \\(x = f^{-1}(y)\\) is given by\n\\[\n\\log \\left| \\, \\det J \\, \\right|\n=\n-2 \\sum_{n \\leq \\binom{K}{2}}\n\\log \\cosh y\n\\\n+\n\\\n\\frac{1}{2} \\\n\\sum_{i &gt; j}\n\\log \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2 \\right)\n.\n\\]",
+    "text": "The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).\n\n\n\nStan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.\nOther than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.\n\n\n\nZero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as a * b, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.\nIdentifiers and literals may not be separated by whitespace. Thus it is not legal to write the number 10000 as 10 000 or to write the identifier normal_lpdf as normal _ lpdf.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Constraint Transforms"
+      "Whitespace"
     ]
   },
   {
-    "objectID": "reference-manual/transforms.html#footnotes",
-    "href": "reference-manual/transforms.html#footnotes",
-    "title": "Constraint Transforms",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nAn alternative to the transform in this section, which can be coded directly in Stan, is to parameterize a covariance matrix as a scaled correlation matrix. An arbitrary \\(K \\times K\\) covariance matrix \\(\\Sigma\\) can be expressed in terms of a \\(K\\)-vector \\(\\sigma\\) and correlation matrix \\(\\Omega\\) as \\[\\Sigma = \\mbox{diag}(\\sigma) \\times \\Omega \\times \\mbox{diag}(\\sigma),\\] so that each entry is just a deviation-scaled correlation, \\[\\Sigma_{m,n} = \\sigma_m \\times \\sigma_n \\times \\Omega_{m,n}.\\]↩︎\nFor convenience, a summation with no terms, such as \\(\\sum_{j' &lt; 1} x_{i,j'}\\), is defined to be 0. This implies \\(x_{1,1} = 1\\) and that \\(x_{i,1} = z_{i,1}\\) for \\(i &gt; 1\\).↩︎",
+    "objectID": "reference-manual/whitespace.html#whitespace-characters",
+    "href": "reference-manual/whitespace.html#whitespace-characters",
+    "title": "Whitespace",
+    "section": "",
+    "text": "The whitespace characters (and their ASCII code points) are the space (0x20), tab (0x09), carriage return (0x0D), and line feed (0x0A).",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Constraint Transforms"
+      "Whitespace"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html",
-    "href": "reference-manual/statements.html",
-    "title": "Statements",
+    "objectID": "reference-manual/whitespace.html#whitespace-neutrality",
+    "href": "reference-manual/whitespace.html#whitespace-neutrality",
+    "title": "Whitespace",
     "section": "",
-    "text": "The blocks of a Stan program are made up of variable declarations and statements; see the blocks chapter for details. Unlike programs in BUGS, the declarations and statements making up a Stan program are executed in the order in which they are written. Variables must be defined to have some value (as well as declared to have some type) before they are used — if they do not, the behavior is undefined.\nThe basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters. Log probability functions can be constructed by using distribution statements and log probability increment statements. Statements may be grouped into sequences and into for-each loops. In addition, Stan allows local variables to be declared in blocks and also allows an empty statement consisting only of a semicolon.\n\n\nThe data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the blocks chapter for more information about the block structure of Stan programs.\n\n\n\nAn assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.\nn = 0;\nExecuting this statement assigns the value of the expression 0, which is the integer zero, to the variable n. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because 0 is an expression of type int, the variable n must be declared as being of type int or of type real. If the variable is of type real, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable n will have the value zero (either as an integer or a floating-point value, depending on its type).\nSyntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).\nBecause the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing\nn = n + 1;\nSuch self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.\nThe left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if Sigma is of type matrix, then\nSigma[1, 1] = 1.0;\nsets the value in the first column of the first row of Sigma to one.\nAssignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, a is an array of type array[,] real and b is an array of type array[] real, then the following two statements are both well-formed.\na[3] = b;\nb = a[4];\nSimilarly, if x is a variable declared to have type row_vector and Y is a variable declared as type matrix, then the following sequence of statements to swap the first two rows of Y is well formed.\nx = Y[1];\nY[1] = Y[2];\nY[2] = x;\n\n\nStan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type int to an lvalue of type real, and we can assign an expression of type real to an lvalue of type complex. Furthermore, promotion is transitive, so that we can assign an expression of type int to an lvalue of type complex.\nPromotion extends to containers, so that arrays of int can be promoted to arrays of real during assignment, and arrays of real can be assigned to an lvalue of type array of complex. Similarly, an expression of type vector may be assigned to an lvalue of type complex_vector, and similarly for row vectors and matrices.\n\n\n\nThe expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,\n\na variable, or\na variable with one or more indices, or\na comma separated list of lvalues surrounded by ( and )\n\nTo be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.\n\n\n\nMultiple indexes, as described in the multi-indexing section, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.\n\n\nAll assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.\n\n\n\n\nStan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.\nreal x = 5;\nx += 7;  // value of x is now 12\nThe compound arithmetic and assignment statement above is equivalent to the following long form.\nx = x + 7;\nIn general, the compound form\nx op= y\nwill be equivalent to\nx = x op y;\nThe compound statement will be legal whenever the long form is legal. This requires that the operation x op y must itself be well formed and that the result of the operation be assignable to x. For the expression x to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.\nvector[N] x;\nx[3] += 2;\nAs a further example, consider\nmatrix[M, M] x;\nvector[M] y;\nreal z;\nx *= x;  // OK, (x * x) is a matrix\nx *= z;  // OK, (x * z) is a matrix\nx *= y;  // BAD, (x * y) is a vector\nThe supported compound arithmetic and assignment operations are listed in the compound arithmetic/assignment table; they are also listed in the index prefaced by operator, e.g., operator+=.\nCompound Arithmetic/Assignment Table.  Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.\n\n\n\noperation\ncompound\nunfolded\n\n\n\n\naddition\nx += y\nx = x + y\n\n\nsubtraction\nx -= y\nx = x - y\n\n\nmultiplication\nx *= y\nx = x * y\n\n\ndivision\nx /= y\nx = x / y\n\n\nelementwise multiplication\nx .*= y\nx = x .* y\n\n\nelementwise division\nx ./= y\nx = x ./ y\n\n\n\n\n\n\n\nThe basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.1\ntarget += -0.5 * y * y;\nThe keyword target here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).\nIn this example, the unnormalized log probability of a unit normal variable \\(y\\) is added to the total log probability. In the general case, the argument can be any expression.2\nAn entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.\nparameters {\n  real y;\n}\nmodel {\n  target += -0.5 * y * y;\n}\nThis model defines a log probability function\n\\[\n\\log p(y) = - \\, \\frac{y^2}{2} - \\log Z\n\\]\nwhere \\(Z\\) is a normalizing constant that does not depend on \\(y\\). The constant \\(Z\\) is conventionally written this way because on the linear scale, \\[\np(y) = \\frac{1}{Z} \\exp\\left(-\\frac{y^2}{2}\\right).\n\\] which is typically written without reference to \\(Z\\) as \\[\np(y) \\propto \\exp\\left(-\\frac{y^2}{2}\\right).\n\\]\nStan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant \\(Z\\) is either time-consuming to compute or intractable to evaluate.\n\n\nThe built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.\nFor instance, the normal_lpdf function returns the log density of a normal distribution:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nThe normal_lupdf function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant sigma term, normal_lupdf would be equivalent to:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nAll functions ending in _lpdf have a corresponding _lupdf version which evaluates and returns the unnormalized density. The same is true for _lpmf and _lupmf.\n\n\n\nThe increment log density statement looks syntactically like compound addition and assignment (see the compound arithmetic/assignment section, it is treated as a primitive statement because target is not itself a variable. So, even though\ntarget += lp;\nis a legal statement, the corresponding long form is not legal.\ntarget = target + lp;  // BAD, target is not a variable\n\n\n\nThe target += ... statement accepts an argument in place of ... for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.\n\n\n\n\nA variant of the target += statement described above is the jacobian += statement. This can be used in the transformed parameters block or in functions ending with _jacobian to mimic the log Jacobian adjustments accrued by built-in variable transforms.\nSimilarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.\nFor example, here is a program which recreates the existing &lt;upper=x&gt; transform on real numbers:\nfunctions {\n  real my_upper_bound_jacobian(real x, real ub) {\n    jacobian += x;\n    return ub - exp(x);\n  }\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  // use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters\n  // e.g.\n  // b ~ lognormal(0, 1);\n}\n\n\nTo access the accumulated log density up to the current execution point, the function target() may be used.\n\n\n\n\nThe term “sampling statement” has been replaced with distribution statement.\n\n\n\nStan supports writing probability statements also using distribution statements, for example\ny ~ normal(mu, sigma);\nmu ~ normal(0, 10);\nsigma ~ normal(0, 1);\nThe symbol \\(\\sim\\) is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.\nIn general, we can read \\(\\sim\\) as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as \\[\n\\begin{aligned}\n   p(y| \\mu, \\sigma) & = \\mathrm{normal}(y |  \\mu, \\sigma)\\\\\n   p(\\mu) & = \\mathrm{normal}(\\mu |  0, 10)\\\\\n   p(\\sigma) & = \\mathrm{normal}^+(\\sigma |  0, 1).\n\\end{aligned}\n\\] A collection of distribution statements define a joint distribution as the product of component distributions \\[\np(y,\\mu,\\sigma) = p(y| \\mu, \\sigma )p(\\mu) p(\\sigma).\n\\]\nThis works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:\n  a ~ normal(0, 1);\n  a ~ normal(0, 1);\nThis is translated to \\[\n    p(a) = \\mathrm{normal}(a | 0, 1)\\mathrm{normal}(a |  0, 1),\n\\] which in this case is \\(\\mathrm{normal}(a|0,1/\\sqrt{2})\\). One might expect that the above two lines of code would represent a redundant expression of a \\(\\mathrm{normal}(a|0,1)\\) prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.\nWhen the joint distribution is considered as a function of parameters (e.g. \\(\\mu\\), \\(\\sigma\\)) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.\nStan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each \\(\\sim\\) statement corresponds to a multiplicative factor in the unnormalized posterior density.\nDistribution statements (~) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).\nIn Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as\ntarget += normal_lpdf(y | mu, sigma);\ntarget += normal_lpdf(mu | 0, 10);\ntarget += normal_lpdf(sigma | 0, 1);\nStan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g. censored data model) or coding language limitations (e.g. mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.\nIn general, a distribution statement of the form\ny ~ dist(theta1, ..., thetaN);\ninvolving subexpressions y and theta1 through thetaN (including the case where N is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real y values, the log probability density function is used,\ntarget += dist_lpdf(y | theta1, ..., thetaN);\nFor those restricted to integer y values, the log probability mass function is used,\ntarget += dist_lpmf(y | theta1, ..., thetaN);\nThis will be well formed if and only if dist_lpdf(y | theta1,   ..., thetaN) or dist_lpmf(y | theta1, ..., thetaN) is a well-formed expression of type real. User defined distributions can be defined in functions block by using function names ending with _lpdf.\n\n\nAlthough both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in\ny ~ normal(mu, sigma);\nand explicitly incrementing the log probability function, as in\ntarget += normal_lpdf(y | mu, sigma);\nThe distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to normal_lpdf adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, y, mu, or sigma, involve only constants, data variables, and transformed data variables. See the section Built in distributions above discussing _lupdf and _lupmf functions that also drops all the constant terms.\n\n\n\nThe left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write\nparameters {\n  real&lt;lower=0&gt; beta;\n}\n// ...\nmodel {\n  log(beta) ~ normal(mu, sigma);\n}\nUnfortunately, this is not enough to properly model beta as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to beta here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure beta gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the change of variables section for full definitions. For the case above, the following adjustment will account for the log transform.3\ntarget += - log(abs(y));\n\n\n\nStan supports truncating distributions with lower bounds, upper bounds, or both.\n\n\nA probability density function \\(p(x)\\) for a continuous distribution may be truncated to an interval \\([a, b]\\) to define a new density \\(p_{[a, b]}(x)\\) with support \\([a, b]\\) by setting\n\\[\np_{[a, b]}(x)\n= \\frac{p(x)}\n       {\\int_a^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) for a discrete distribution may be truncated to the closed interval \\([a, b]\\) by\n\\[\np_{[a, b]}(x) = \\frac{p(x)}\n                  {\\sum_{u = a}^b p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x)\n= \\frac{p(x)}\n       {\\int_a^{\\infty} p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x) = \\frac{p(x)}\n                  {\\sum_{a &lt;= u} p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty, b]}(x)\n= \\frac{p(x)}\n       {\\int_{-\\infty}^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty,b]}(x) = \\frac{p(x)}\n                  {\\sum_{u &lt;= b} p(u)}.\n\\]\n\n\n\nGiven a probability function \\(p_X(x)\\) for a random variable \\(X\\), its cumulative distribution function (cdf) \\(F_X(x)\\) is defined to be the probability that \\(X \\leq x\\),\n\\[\nF_X(x) = \\Pr[X \\leq x].\n\\]\nThe upper-case variable \\(X\\) is the random variable whereas the lower-case variable \\(x\\) is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to\n\\[\nF_X(x) \\ = \\ \\int_{-\\infty}^{x} p_X(u) \\, du,\n\\]\nFor discrete variables, the cdf is defined to include the upper bound given by the argument,\n\\[\nF_X(x) = \\sum_{u \\leq x} p_X(u).\n\\]\n\n\n\nThe complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by\n\\[\nF^C_X(x)\n\\ = \\ \\Pr[X &gt; x]\n\\ = \\ 1 - F_X(x).\n\\]\nUnlike the cdf, the ccdf is exclusive of the bound, hence the event \\(X &gt; x\\) rather than the cdf’s event \\(X \\leq x\\).\nFor continuous distributions, the ccdf works out to\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\int_{-\\infty}^x p_X(u) \\, du\n\\ = \\ \\int_x^{\\infty} p_X(u) \\, du.\n\\]\nThe lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over \\(u &gt; x\\),\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\sum_{u \\leq x} p_X(u)\n\\ = \\ \\sum_{u &gt; x} p_X(u).\n\\]\nCumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by \\[\n\\int_a^b p(u) \\, du = F_X(b) - F_X(a).\n\\] This allows truncated distributions to be defined as \\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a)}.\n\\]\nFor discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from \\(F_X(b) - F_X(a)\\),\n\\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a) + p_X(a)}.\n\\]\n\n\n\nStan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to \\([-0.5, 2.1]\\) can be coded with the following distribution statement.\ny ~ normal(0, 1) T[-0.5, 2.1];\nTruncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.\nIn general, the truncation bounds and parameters may be parameters or local variables.\nBecause the example above involves a continuous distribution, it behaves the same way as the following more verbose form.\ny ~ normal(0, 1);\nif (y &lt; -0.5 || y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -log_diff_exp(normal_lcdf(2.1 | 0, 1),\n                          normal_lcdf(-0.5 | 0, 1));\n}\nBecause a Stan program defines a log density function, all calculations are on the log scale. The function normal_lcdf is the log of the cumulative normal distribution function and the function log_diff_exp(a, b) is a more arithmetically stable form of log(exp(a) - exp(b)).\nFor a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution\ny ~ poisson(3.7) T[2, 10];\nbehaves in the same way as the following code.\ny ~ poisson(3.7);\nif (y &lt; 2 || y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         log_diff_exp(poisson_lcdf(10 | 3.7),\n                                      poisson_lcdf(2 | 3.7)));\n}\nRecall that log_sum_exp(a, b) is just the arithmetically stable form of log(exp(a) + exp(b)).\n\n\n\nFor truncating with only a lower bound, the upper limit is left blank.\ny ~ normal(0, 1) T[-0.5, ];\nThis truncated distribution statement has the same behavior as the following code.\ny ~ normal(0, 1);\nif (y &lt; -0.5) {\n  target += negative_infinity();\n} else {\n  target += -normal_lccdf(-0.5 | 0, 1);\n}\nThe normal_lccdf function is the normal complementary cumulative distribution function.\nAs with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus\ny ~ poisson(3.7) T[2, ];\nbehaves the same way as\ny ~ poisson(3.7);\nif (y &lt; 2) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         poisson_lccdf(2 | 3.7));\n}\n\n\n\nTo truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement\ny ~ normal(0, 1) T[ , 2.1];\nproduces the same result as the following code.\ntarget += normal_lpdf(y | 0, 1);\nif (y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -normal_lcdf(2.1 | 0, 1);\n}\nWith only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement\ny ~ poisson(3.7) T[ , 10];\nbehaves the same way as the following code.\ny ~ poisson(3.7);\nif (y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -poisson_lcdf(10 | 3.7);\n}\n\n\n\nIn all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.\n\n\n\nFor continuous distributions, truncation points must be expressions of type int or real. For discrete distributions, truncation points must be expressions of type int.\n\n\n\nFor a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds \\(-\\infty\\) to the total log probability, which in turn results in the sample being rejected.\n\n\n\nVectorization of distribution functions with truncation is available if the underlying distribution, lcdf, and lccdf functions meet the required signatures.\nThe equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):\n\nIf the variate y is the only non-scalar, the result is the same as described in the above sections, but the lcdf/lccdf calculation is multiplied by size(y).\nIf the other arguments to the distribution are non-scalars, then the vectorized version of the lcdf/lccdf is used. These functions return the sum of their terms, so no multiplication by the size is needed.\nThe exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the log_diff_exp of two sums is not the same as the sum of the pairwise log_diff_exp operations.\n\nNote that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.\n\n\n\n\n\nSuppose N is a variable of type int, y is a one-dimensional array of type array[] real, and mu and sigma are variables of type real. Furthermore, suppose that n has not been defined as a variable. Then the following is a well-formed for-loop statement.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nThe loop variable is n, the loop bounds are the values in the range 1:N, and the body is the statement following the loop bounds.\n\n\nThe type of the loop variable is int. Unlike in C++ and similarly to R, this variable must not be declared explicitly.\nThe bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range L:H will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to L and less than or equal to H. For example, the loop for (n in 2:5) will cause the body of the for loop to be executed with n equal to 2, 3, 4, and 5, in order. The variable and bound for (n in   5:2) will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.\nThe scope of the loop variable is limited to the body of the loop.\n\n\n\nUnlike in BUGS, Stan allows variables to be reassigned. For example, the variable theta in the following program is reassigned in each iteration of the loop.\nfor (n in 1:N) {\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nSuch reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.4\nIn Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.\nfor (n in 1:N) {\n  y[n] ~ bernoulli(theta);\n  theta = inv_logit(alpha + x[n] * beta);\n}\nIn this program, theta is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.\nStan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.\ntotal = 0.0;\nfor (n in 1:N) {\n  total = total + x[n];\n}\nAfter the for loop is executed, the variable total will hold the sum of the elements in the array x. This example was purely pedagogical; it is easier and more efficient to write\ntotal = sum(x);\nA variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.\nfor (n in 1:100) {\n  y += y * epsilon;\n  epsilon = 0.5 * epsilon;\n  y += y * epsilon;\n}\n\n\n\n\nA second form of for loops allows iteration over elements of containers. If ys is an expression denoting a container (vector, row vector, matrix, or array) with elements of type T, then the following is a well-formed foreach statement.\nfor (y in ys) {\n  // ... do something with y ...\n}\nThe order in which elements of ys are visited is defined for container types as follows.\n\nvector, row_vector: elements visited in order, y is of type double\nmatrix: elements visited in column-major order, y is of type double\narray[] T: elements visited in order, y is of type T.\n\nConsequently, if ys is a two dimensional array array[,] real, y will be a one-dimensional array of real values (type array[] real). If ’ysis a matrix, thenywill be a real value (typereal`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,\narray[2, 3] real yss;\nfor (ys in yss) {\n  for (y in ys) {\n    // ... do something with y ...\n  }\n}\nwhereas a matrix can be looped over in one foreach statement\nmatrix[2, 3] yss;\nfor (y in yss) {\n   // ... do something with y...\n}\nIn both cases, the loop variable y is of type real.  The elements of the matrix are visited in column-major order (e.g.,y[1, 1],y[2, 1],y[1, 2], ...,y[2, 3]), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,y[1, 1],y[1, 2],y[1, 3],y[2, 1], ...,y[2, 3]`).\n\n\n\nStan supports full conditional statements using the same if-then-else syntax as C++. The general format is\nif (condition1)\n  statement1\nelse if (condition2)\n  statement2\n// ...\nelse if (conditionN-1)\n  statementN-1\nelse\n  statementN\nThere must be a single leading if clause, which may be followed by any number of else if clauses, all of which may be optionally followed by an else clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.\nThe entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.\n\n\n\nStan supports standard while loops using the same syntax as C++. The general format is as follows.\nwhile (condition)\n  body\nThe condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).\nEvaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.\nThe rest of the body of a while loop may be skipped using a continue. The loop will be exited with a break statement. See the section on continue and break statements for more details.\n\n\n\nJust as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.\n\n\nBlocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nTo put multiple statements inside the body of a for loop, a block is used, as in the following example.\nfor (n in 1:N) {\n  lambda[n] ~ gamma(alpha, beta);\n  y[n] ~ poisson(lambda[n]);\n}\nThe open curly bracket ({) is the first character of the block and the close curly bracket (}) is the last character.\nBecause whitespace is ignored in Stan, the following program will not compile.\nfor (n in 1:N)\n  y[n] ~ normal(mu, sigma);\n  z[n] ~ normal(mu, sigma); // ERROR!\nThe problem is that the body of the for loop is taken to be the statement directly following it, which is y[n] ~ normal(mu, sigma). This leaves the probability statement for z[n] hanging, as is clear from the following equivalent program.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nz[n] ~ normal(mu, sigma); // ERROR!\nNeither of these programs will compile. If the loop variable n was defined before the for loop, the for-loop declaration will raise an error. If the loop variable n was not defined before the for loop, then the use of the expression z[n] will raise an error.\n\n\n\nA for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.\nfor (n in 1:N) {\n  real theta;\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nThe local variable theta is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus theta is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable theta if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.\nfor (m in 1:M) {\n  real theta;\n  for (n in 1:N) {\n    real theta; // ERROR!\n    theta = inv_logit(alpha + x[m, n] * beta);\n    y[m, n] ~ bernoulli(theta);\n// ...\nThe compiler will flag the second declaration of theta with a message that it is already defined.\n\n\n\nLocal variables may not have constraints on their declaration. The only types that may be used are listed in the types table under “local”.\n\n\n\nA block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following\nfor (m in 1:M) {\n  {\n     int n = 2 * m;\n     sum += n;\n  }\n  for (n in 1:N) {\n    sum += x[m, n];\n  }\n}\nThe variable declaration int n; is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of n are unrelated.\n\n\n\n\nThe one-token statements continue and break may be used within loops to alter control flow; continue causes the next iteration of the loop to run immediately, whereas break terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both break and continue scope to the most deeply nested loop, but pass through non-loop statements.\nAlthough these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.\n\n\nWhen a break statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:\nwhile (1) {\n  if (n &lt; 0) {\n    break;\n  }\n  foo(n);\n  n = n - 1;\n}\nThe while~(1) loop is a “forever” loop, because 1 is the true value, so the test always succeeds. Within the loop, if the value of n is less than 0, the loop terminates, otherwise it executes foo(n) and then decrements n. The statement above does exactly the same thing as\nwhile (n &gt;= 0) {\n  foo(n);\n  n = n - 1;\n}\nThis case is simply illustrative of the behavior; it is not a case where a break simplifies the loop.\n\n\n\nThe continue statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array x,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &lt;= 0) {\n    continue;\n  }\n  sum += x[n];\n}\nWhen the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an x[n] less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &gt; 0) {\n    sum += x[n];\n  }\n}\nWhile the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.\n\n\n\nIf there is a loop nested within a loop, a break or continue statement only breaks out of the inner loop. So\nwhile (cond1) {\n  // ...\n  while (cond2) {\n    // ...\n    if (cond3) {\n      break;\n    }\n    // ...\n  }\n  // execution continues here after break\n  // ...\n}\nIf the break is triggered by cond3 being true, execution will continue after the nested loop.\nAs with break statements, continue statements go back to the top of the most deeply nested loop in which the continue appears.\nAlthough break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.\n\n\n\n\nStan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.\nfor (n in 1:N) { print(\"loop iteration: \", n); ... }\nThe print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression n, followed by a new line.\n\n\nThe text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.\nPrinting a simple real or int typed variable always prints the variable’s value.5\nFor array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as\n[1, 2, 3]\nand a \\(2 \\times 3\\)-matrix as\n[[1, 2, 3], [4, 5, 6]]\nComplex numbers print as pairs. For example, the pair of statements\ncomplex z = to_complex(1.2, -3.5);\nprint(z)\nwill print as (1.2,-3.5), with no space after the comma or within the parentheses.\nPrinting a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.\ntransformed data {\n  matrix[2, 2] u;\n  u[1, 1] = 1.0;   u[1, 2] = 4.0;\n  u[2, 1] = 9.0;   u[2, 2] = 16.0;\n  for (n in 1:2) {\n    print(\"u[\", n, \"] = \", u[n]);\n  }\n}\nThis print statement executes twice, printing the following two lines of output.\nu[1] = [1, 4]\nu[2] = [9, 16]\n\n\n\nThe input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.\n\n\n\nPrinting for a print statement happens every time it is executed. The transformed data block is executed once per chain, the transformed parameter and model blocks once per leapfrog step, and the generated quantities block once per iteration.\n\n\n\nString literals begin and end with a double quote character (\"). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.\nThe Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.\n\n\n\nBecause Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.\nIt is particularly useful to print the value of the target log density accumulator (through the target() function), as in the following example.\nvector[2] y;\ny[1] = 1;\nprint(\"log density before =\", target());\ny ~ normal(0,1);  // bug!  y[2] not defined\nprint(\"log density after =\", target());\nThe example has a bug in that y[2] is not defined before the vector y is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).\nNote that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.\n\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.\nLike the print statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.\nReject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable x’s value is negative.\nif (x &lt; 0) {\n  reject(\"x must not be negative; found x=\", x);\n}\n\n\nReject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the normal_lpdf function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.\nIn all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.\n\n\nRejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.\n\n\n\nRejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.\nReject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.\nFatal errors in other blocks may also be signaled by use of the fatal_error statement.\n\n\n\nRejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.\nIf the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.\n\n\n\n\nRejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;\n  real&lt;lower=a, upper=b&gt; theta;\n  // ...\n}\nmodel {\n  // **wrong** needs explicit truncation\n  theta ~ normal(0, 1);\n  // ...\n}\nThis program is wrong because its truncation bounds on theta depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.\n  theta ~ normal(0, 1) T[a, b];\nThe conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over \\([a ,b]\\); in these simple univariate cases, we can overcome that with the T[ , ] notation, which essentially divides by whatever the prior integrates to over \\([a, b]\\).\nThis problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.\n  if (theta &lt; a || theta &gt; b) {\n    reject(\"theta not in (a, b)\");\n  }\n  // still **wrong**, needs T[a,b]\n  theta ~ normal(0, 1);\nIn this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.\nEven if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.\n\n\n\n\nThe Stan fatal_error statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.\nLike the print or reject statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.\nThe fatal error may be used to signal an unrecoverable error in blocks where reject leads to the algorithm attempting to try again, such as the model block.",
+    "text": "Stan treats all whitespace characters identically. Specifically, there is no significance to indentation, to tabs, to carriage returns or line feeds, or to any vertical alignment of text. Any whitespace character is exchangeable with any other.\nOther than for readability, the number of whitespaces is also irrelevant. One or more whitespace characters of any type are treated identically by the parser.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "Whitespace"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#statement-block-contexts",
-    "href": "reference-manual/statements.html#statement-block-contexts",
-    "title": "Statements",
+    "objectID": "reference-manual/whitespace.html#whitespace-location",
+    "href": "reference-manual/whitespace.html#whitespace-location",
+    "title": "Whitespace",
     "section": "",
-    "text": "The data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the blocks chapter for more information about the block structure of Stan programs.",
+    "text": "Zero or more whitespace characters may be placed between symbols in a Stan program. For example, zero or more whitespace characters of any variety may be included before and after a binary operation such as a * b, before a statement-ending semicolon, around parentheses or brackets, before or after commas separating function arguments, etc.\nIdentifiers and literals may not be separated by whitespace. Thus it is not legal to write the number 10000 as 10 000 or to write the identifier normal_lpdf as normal _ lpdf.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "Whitespace"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#assignment-statement.section",
-    "href": "reference-manual/statements.html#assignment-statement.section",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html",
+    "href": "reference-manual/user-functions.html",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "An assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.\nn = 0;\nExecuting this statement assigns the value of the expression 0, which is the integer zero, to the variable n. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because 0 is an expression of type int, the variable n must be declared as being of type int or of type real. If the variable is of type real, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable n will have the value zero (either as an integer or a floating-point value, depending on its type).\nSyntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).\nBecause the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing\nn = n + 1;\nSuch self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.\nThe left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if Sigma is of type matrix, then\nSigma[1, 1] = 1.0;\nsets the value in the first column of the first row of Sigma to one.\nAssignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, a is an array of type array[,] real and b is an array of type array[] real, then the following two statements are both well-formed.\na[3] = b;\nb = a[4];\nSimilarly, if x is a variable declared to have type row_vector and Y is a variable declared as type matrix, then the following sequence of statements to swap the first two rows of Y is well formed.\nx = Y[1];\nY[1] = Y[2];\nY[2] = x;\n\n\nStan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type int to an lvalue of type real, and we can assign an expression of type real to an lvalue of type complex. Furthermore, promotion is transitive, so that we can assign an expression of type int to an lvalue of type complex.\nPromotion extends to containers, so that arrays of int can be promoted to arrays of real during assignment, and arrays of real can be assigned to an lvalue of type array of complex. Similarly, an expression of type vector may be assigned to an lvalue of type complex_vector, and similarly for row vectors and matrices.\n\n\n\nThe expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,\n\na variable, or\na variable with one or more indices, or\na comma separated list of lvalues surrounded by ( and )\n\nTo be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.\n\n\n\nMultiple indexes, as described in the multi-indexing section, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.\n\n\nAll assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.\n\n\n\n\nStan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.\nreal x = 5;\nx += 7;  // value of x is now 12\nThe compound arithmetic and assignment statement above is equivalent to the following long form.\nx = x + 7;\nIn general, the compound form\nx op= y\nwill be equivalent to\nx = x op y;\nThe compound statement will be legal whenever the long form is legal. This requires that the operation x op y must itself be well formed and that the result of the operation be assignable to x. For the expression x to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.\nvector[N] x;\nx[3] += 2;\nAs a further example, consider\nmatrix[M, M] x;\nvector[M] y;\nreal z;\nx *= x;  // OK, (x * x) is a matrix\nx *= z;  // OK, (x * z) is a matrix\nx *= y;  // BAD, (x * y) is a vector\nThe supported compound arithmetic and assignment operations are listed in the compound arithmetic/assignment table; they are also listed in the index prefaced by operator, e.g., operator+=.\nCompound Arithmetic/Assignment Table.  Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.\n\n\n\noperation\ncompound\nunfolded\n\n\n\n\naddition\nx += y\nx = x + y\n\n\nsubtraction\nx -= y\nx = x - y\n\n\nmultiplication\nx *= y\nx = x * y\n\n\ndivision\nx /= y\nx = x / y\n\n\nelementwise multiplication\nx .*= y\nx = x .* y\n\n\nelementwise division\nx ./= y\nx = x ./ y",
+    "text": "Stan allows users to define their own functions. The basic syntax is a simplified version of that used in C and C++. This chapter specifies how functions are declared, defined, and used in Stan.\n\n\nUser-defined functions appear in a special function-definition block before all of the other program blocks.\nfunctions {\n   // ... function declarations and definitions ...\n}\ndata {\n  // ...\nFunction definitions and declarations may appear in any order. Forward declarations are allowed but not required.\n\n\n\nThe rules for function naming and function-argument naming are the same as for other variables; see the section on variables for more information on valid identifiers. For example,\nreal foo(real mu, real sigma);\ndeclares a function named foo with two argument variables of types real and real. The arguments are named mu and sigma, but that is not part of the declaration.\n\n\nMultiple user-defined functions may have the same name if they have different sequences of argument types. This is known as function overloading.\nFor example, the following two functions are both defined with the name add_up\nreal add_up(real a, real b){\n  return a + b;\n}\n\nreal add_up(real a, real b, real c){\n  return a + b + c;\n}\nThe return types of overloaded functions do not need to be the same. One could define an additional add_up function as follows\nint add_up(int a, int b){\n  return a + b;\n}\nThat being said, functions may not use the same name if their signature only differs by the return type.\nFor example, the following is not permitted\n// illegal\nreal baz(int x);\nint baz(int x);\nFunction names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the reduce_sum family of functions and ODE integrators, which cannot be overloaded.\n\n\n\n\nAll function arguments are mandatory—there are no default values.\n\n\nFunctions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.\n\n\n\nFunctions with void return types may be applied to arguments and used as statements. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.\n\n\n\nOverloaded functions alongside type promotion can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.\nConsider the following two overloaded functions\nreal foo(int a, real b);\nreal foo(real a, int b);\nThese functions do not have a unique minimum when called with two integer arguments foo(1,2), and therefore cannot be called as such.\nPromotion of integers to complex numbers is considered as two separate promotions, one from int to real and a second from real to complex. Consider the following functions with real and complex signatures\nreal bar(real x);\nreal bar(complex z);\nA call bar(5) with an integer argument will be resolved to bar(real) because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.\n\n\n\nThe rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.\nreal foo(real x) { return ... };\n...\nint a = 5;\nreal b = foo(a); // a promoted to type real\nIn addition to promoting int to real, Stan also promotes real to complex, and by transitivity, int to complex. This also works for containers, so an array of int may be assigned to an array of real of the same shape. And we can also promote vector to complex_vector and similarly for row vectors and matrices.\n\n\n\nFunctions whose name ends in _lpdf or _lpmf (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of statements.qmd#distribution-statements.section.\n\n\n\nFunctions of certain types are restricted on scope of usage. Functions whose names end in _lp assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.\nFunctions whose name end in _jacobian assume access to the log probability accumulator may only be used within the transformed parameters block.\nFunctions whose names end in _rng assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in _rng.\nFunctions whose names end in _lpdf and _lpmf can be used anywhere. However, _lupdf and _lupmf functions can only be used in the model block or user-defined probability functions.\nSee the section on function bodies for more information on these special types of function.\n\n\n\n\nStan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.\nThe type void may not be used as an argument type, only a return type for a function with side effects.\n\n\nThe base variable types are integer, real, complex, vector, row_vector, and matrix. No lower-bound or upper-bound constraints are allowed (e.g., real&lt;lower=0&gt; is illegal). Specialized constrained types are also not allowed (e.g., simplex is illegal).\nTuple types of the form tuple(T1, ..., TN) are also allowed, with all of the types T1 to TN being function argument types (i.e., no constraints and no sizes).\n\n\n\nArguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, int denotes a single integer argument or return, whereas array[] real indicates a one-dimensional array of reals, array[,] real a two-dimensional array and array[,,] real a three-dimensional array; whitespace is optional, as usual.\nThe dimensions for vectors and matrices are not included, so that matrix is the type of a single matrix argument or return type. Thus if a variable is declared as matrix a, then a has two indexing dimensions, so that a[1] is a row vector and a[1, 1] a real value. Matrices implicitly have two indexing dimensions. The type declaration matrix[ , ] b specifies that b is a two-dimensional array of matrices, for a total of four indexing dimensions, with b[1, 1, 1, 1] picking out a real value.\n\n\n\nFunction argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.\nDimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.\nAny errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.\n\n\n\nSome of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.\nIn user-defined functions, the qualifier data may be placed before an argument type declaration to indicate that the argument must be data only. For example,\nreal foo(data real x) {\n  return x^2;\n}\nrequires the argument x to be data only.\nDeclaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.\n\n\n\n\nThe body of a function is between an open curly brace ({) and close curly brace (}). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.\nAny user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.\nThe only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.\n\n\nFunctions that call random number generating functions in their bodies must have a name that ends in _rng; attempts to use random-number generators in other functions lead to a compile-time error.\nLike other random number generating functions, user-defined functions with names that end in _rng may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in _rng. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions that include distribution statements or log probability increment statements must have a name that ends in _lp. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.\nLike the target log density increment statement and distribution statements, user-defined functions with names that end in _lp may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions whose names end in _lpdf and _lpmf (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (~) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of y given parameter theta allows the use of the distribution statement is defined as follows.\nreal foo_lpdf(real y, vector theta) { ... }\nNote that for function definitions, the comma is used rather than the vertical bar.\nFor every custom _lpdf and _lpmf defined there is a corresponding _lupdf and _lupmf defined automatically. The _lupdf and _lupmf versions of the functions cannot be defined directly (to do so will produce an error). The difference in the _lpdf and _lpmf and the corresponding _lupdf and _lupmf functions is that if any other unnormalized density functions are used inside the user-defined function, the _lpdf and _lpmf forms of the user-defined function will change these densities to be normalized. The _lupdf and _lupmf forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.\nThe distribution statement shorthand\nz ~ foo(phi);\nwill have the same effect as incrementing the target with the log of the unnormalized density:\ntarget += foo_lupdf(z | phi);\nOther _lupdf and _lupmf functions used in the definition of foo_lpdf will drop additive constants when foo_lupdf is called and will not drop additive constants when foo_lpdf is called.\nIf there are _lupdf and _lupmf functions used inside the following call to foo_lpdf, they will be forced to normalize (return the equivalent of their _lpdf and _lpmf forms):\ntarget += foo_lpdf(z | phi);\nIf there are no _lupdf or _lupmf functions used in the definition of foo_lpdf, then there will be no difference between a foo_lpdf or foo_lupdf call.\nThe unnormalized _lupdf and _lupmf functions can only be used in the model block or in user-defined probability functions (those ending in _lpdf or _lpmf).\nThe same syntax and shorthand that works for _lpdf also works for log probability mass functions with suffixes _lpmf.\nA function that is going to be accessed as distributions must return the log of the density or mass function it defines.\n\n\n\n\nWithin function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (=) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.1\nLocal variables may be declared at the top of the function block and scope as usual.\n\n\n\nNon-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.\nVoid functions may use return only without an argument, but return statements are not mandatory.\n\n\nUnlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that\n\na return statement qualifies,\n\nand the recursive cases are that\n\na sequence of statements qualifies if its last statement qualifies,\na for loop or while loop qualifies if its body qualifies, and\na conditional statement qualifies if it has a default else clause and all of its body statements qualify.\n\nAn exception is made for “obviously infinite” loops like while (1), which contain a return statement and no break statements. The only way to exit such a loop is to return, so they are considered as returning statements.\nThese rules disqualify\nreal foo(real x) {\n  if (x &gt; 2) {\n    return 1.0;\n  } else if (x &lt;= 2) {\n    return -1.0;\n  }\n}\nbecause there is no default else clause, and disqualify\nreal foo(real x) {\n  real y;\n  y = x;\n  while (x &lt; 10) {\n    if (x &gt; 0) {\n      return x;\n    }\n    y = x / 2;\n  }\n}\nbecause the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow\nreal log_fancy(real x) {\n  if (x &lt; 1e-30) {\n    return x;\n  } else if (x &lt; 1e-14) {\n    return x * x;\n  } else {\n    return log(x);\n  }\n}\nbecause there’s a default else clause and each condition body has return as its final statement.\n\n\n\n\n\n\nA function can be declared without a return value by using void in place of a return type. Note that the type void may only be used as a return type—arguments may not be declared to be of type void.\n\n\n\nA void function may be used as a statement.\nBecause there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.\n\n\n\nIn a return statement within a void function’s definition, the return keyword is followed immediately by a semicolon (;) rather than by the expression whose value is returned.\n\n\n\n\nStan supports forward declarations, which look like function definitions without bodies. For example,\nreal unit_normal_lpdf(real y);\ndeclares a function named unit_normal_lpdf that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.\nA function definition with a body simultaneously declares and defines the named function, as in\nreal unit_normal_lpdf(real y) {\n  return -0.5 * square(y);\n}\nA function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the section on function overloading.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#increment-log-prob.section",
-    "href": "reference-manual/statements.html#increment-log-prob.section",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#function-definition-block",
+    "href": "reference-manual/user-functions.html#function-definition-block",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "The basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.1\ntarget += -0.5 * y * y;\nThe keyword target here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).\nIn this example, the unnormalized log probability of a unit normal variable \\(y\\) is added to the total log probability. In the general case, the argument can be any expression.2\nAn entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.\nparameters {\n  real y;\n}\nmodel {\n  target += -0.5 * y * y;\n}\nThis model defines a log probability function\n\\[\n\\log p(y) = - \\, \\frac{y^2}{2} - \\log Z\n\\]\nwhere \\(Z\\) is a normalizing constant that does not depend on \\(y\\). The constant \\(Z\\) is conventionally written this way because on the linear scale, \\[\np(y) = \\frac{1}{Z} \\exp\\left(-\\frac{y^2}{2}\\right).\n\\] which is typically written without reference to \\(Z\\) as \\[\np(y) \\propto \\exp\\left(-\\frac{y^2}{2}\\right).\n\\]\nStan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant \\(Z\\) is either time-consuming to compute or intractable to evaluate.\n\n\nThe built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.\nFor instance, the normal_lpdf function returns the log density of a normal distribution:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nThe normal_lupdf function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant sigma term, normal_lupdf would be equivalent to:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nAll functions ending in _lpdf have a corresponding _lupdf version which evaluates and returns the unnormalized density. The same is true for _lpmf and _lupmf.\n\n\n\nThe increment log density statement looks syntactically like compound addition and assignment (see the compound arithmetic/assignment section, it is treated as a primitive statement because target is not itself a variable. So, even though\ntarget += lp;\nis a legal statement, the corresponding long form is not legal.\ntarget = target + lp;  // BAD, target is not a variable\n\n\n\nThe target += ... statement accepts an argument in place of ... for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.",
+    "text": "User-defined functions appear in a special function-definition block before all of the other program blocks.\nfunctions {\n   // ... function declarations and definitions ...\n}\ndata {\n  // ...\nFunction definitions and declarations may appear in any order. Forward declarations are allowed but not required.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment",
-    "href": "reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#function-names",
+    "href": "reference-manual/user-functions.html#function-names",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "A variant of the target += statement described above is the jacobian += statement. This can be used in the transformed parameters block or in functions ending with _jacobian to mimic the log Jacobian adjustments accrued by built-in variable transforms.\nSimilarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.\nFor example, here is a program which recreates the existing &lt;upper=x&gt; transform on real numbers:\nfunctions {\n  real my_upper_bound_jacobian(real x, real ub) {\n    jacobian += x;\n    return ub - exp(x);\n  }\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  // use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters\n  // e.g.\n  // b ~ lognormal(0, 1);\n}\n\n\nTo access the accumulated log density up to the current execution point, the function target() may be used.",
+    "text": "The rules for function naming and function-argument naming are the same as for other variables; see the section on variables for more information on valid identifiers. For example,\nreal foo(real mu, real sigma);\ndeclares a function named foo with two argument variables of types real and real. The arguments are named mu and sigma, but that is not part of the declaration.\n\n\nMultiple user-defined functions may have the same name if they have different sequences of argument types. This is known as function overloading.\nFor example, the following two functions are both defined with the name add_up\nreal add_up(real a, real b){\n  return a + b;\n}\n\nreal add_up(real a, real b, real c){\n  return a + b + c;\n}\nThe return types of overloaded functions do not need to be the same. One could define an additional add_up function as follows\nint add_up(int a, int b){\n  return a + b;\n}\nThat being said, functions may not use the same name if their signature only differs by the return type.\nFor example, the following is not permitted\n// illegal\nreal baz(int x);\nint baz(int x);\nFunction names used in the Stan standard library may be overloaded by user-defined functions. Exceptions to this are the reduce_sum family of functions and ODE integrators, which cannot be overloaded.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#sampling-statements.section",
-    "href": "reference-manual/statements.html#sampling-statements.section",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#calling-functions",
+    "href": "reference-manual/user-functions.html#calling-functions",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "The term “sampling statement” has been replaced with distribution statement.",
+    "text": "All function arguments are mandatory—there are no default values.\n\n\nFunctions with non-void return types are called just like any other built-in function in Stan—they are applied to appropriately typed arguments to produce an expression, which has a value when executed.\n\n\n\nFunctions with void return types may be applied to arguments and used as statements. These act like distribution statements or print statements. Such uses are only appropriate for functions that act through side effects, such as incrementing the log probability accumulator, printing, or raising exceptions.\n\n\n\nOverloaded functions alongside type promotion can result in situations where there are multiple valid interpretations of a function call. Stan requires that there be a unique signature which minimizes the number of promotions required.\nConsider the following two overloaded functions\nreal foo(int a, real b);\nreal foo(real a, int b);\nThese functions do not have a unique minimum when called with two integer arguments foo(1,2), and therefore cannot be called as such.\nPromotion of integers to complex numbers is considered as two separate promotions, one from int to real and a second from real to complex. Consider the following functions with real and complex signatures\nreal bar(real x);\nreal bar(complex z);\nA call bar(5) with an integer argument will be resolved to bar(real) because it only requires a single promotion, whereas the promotion to a complex number requires two promotions.\n\n\n\nThe rules for calling functions work the same way as assignment as far as promotion goes. This means that we can promote arguments to the type expected by function arguments. For example, the following will work.\nreal foo(real x) { return ... };\n...\nint a = 5;\nreal b = foo(a); // a promoted to type real\nIn addition to promoting int to real, Stan also promotes real to complex, and by transitivity, int to complex. This also works for containers, so an array of int may be assigned to an array of real of the same shape. And we can also promote vector to complex_vector and similarly for row vectors and matrices.\n\n\n\nFunctions whose name ends in _lpdf or _lpmf (log density and mass functions) may be used as probability functions and may be used in place of parameterized distributions on the right side of statements.qmd#distribution-statements.section.\n\n\n\nFunctions of certain types are restricted on scope of usage. Functions whose names end in _lp assume access to the log probability accumulator and are only available in the transformed parameters and model blocks.\nFunctions whose name end in _jacobian assume access to the log probability accumulator may only be used within the transformed parameters block.\nFunctions whose names end in _rng assume access to the random number generator and may only be used within the generated quantities block, transformed data block, and within user-defined functions ending in _rng.\nFunctions whose names end in _lpdf and _lpmf can be used anywhere. However, _lupdf and _lupmf functions can only be used in the model block or user-defined probability functions.\nSee the section on function bodies for more information on these special types of function.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#distribution-statements.section",
-    "href": "reference-manual/statements.html#distribution-statements.section",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#argument-types-and-qualifiers",
+    "href": "reference-manual/user-functions.html#argument-types-and-qualifiers",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Stan supports writing probability statements also using distribution statements, for example\ny ~ normal(mu, sigma);\nmu ~ normal(0, 10);\nsigma ~ normal(0, 1);\nThe symbol \\(\\sim\\) is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.\nIn general, we can read \\(\\sim\\) as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as \\[\n\\begin{aligned}\n   p(y| \\mu, \\sigma) & = \\mathrm{normal}(y |  \\mu, \\sigma)\\\\\n   p(\\mu) & = \\mathrm{normal}(\\mu |  0, 10)\\\\\n   p(\\sigma) & = \\mathrm{normal}^+(\\sigma |  0, 1).\n\\end{aligned}\n\\] A collection of distribution statements define a joint distribution as the product of component distributions \\[\np(y,\\mu,\\sigma) = p(y| \\mu, \\sigma )p(\\mu) p(\\sigma).\n\\]\nThis works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:\n  a ~ normal(0, 1);\n  a ~ normal(0, 1);\nThis is translated to \\[\n    p(a) = \\mathrm{normal}(a | 0, 1)\\mathrm{normal}(a |  0, 1),\n\\] which in this case is \\(\\mathrm{normal}(a|0,1/\\sqrt{2})\\). One might expect that the above two lines of code would represent a redundant expression of a \\(\\mathrm{normal}(a|0,1)\\) prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.\nWhen the joint distribution is considered as a function of parameters (e.g. \\(\\mu\\), \\(\\sigma\\)) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.\nStan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each \\(\\sim\\) statement corresponds to a multiplicative factor in the unnormalized posterior density.\nDistribution statements (~) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).\nIn Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as\ntarget += normal_lpdf(y | mu, sigma);\ntarget += normal_lpdf(mu | 0, 10);\ntarget += normal_lpdf(sigma | 0, 1);\nStan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g. censored data model) or coding language limitations (e.g. mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.\nIn general, a distribution statement of the form\ny ~ dist(theta1, ..., thetaN);\ninvolving subexpressions y and theta1 through thetaN (including the case where N is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real y values, the log probability density function is used,\ntarget += dist_lpdf(y | theta1, ..., thetaN);\nFor those restricted to integer y values, the log probability mass function is used,\ntarget += dist_lpmf(y | theta1, ..., thetaN);\nThis will be well formed if and only if dist_lpdf(y | theta1,   ..., thetaN) or dist_lpmf(y | theta1, ..., thetaN) is a well-formed expression of type real. User defined distributions can be defined in functions block by using function names ending with _lpdf.\n\n\nAlthough both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in\ny ~ normal(mu, sigma);\nand explicitly incrementing the log probability function, as in\ntarget += normal_lpdf(y | mu, sigma);\nThe distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to normal_lpdf adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, y, mu, or sigma, involve only constants, data variables, and transformed data variables. See the section Built in distributions above discussing _lupdf and _lupmf functions that also drops all the constant terms.\n\n\n\nThe left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write\nparameters {\n  real&lt;lower=0&gt; beta;\n}\n// ...\nmodel {\n  log(beta) ~ normal(mu, sigma);\n}\nUnfortunately, this is not enough to properly model beta as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to beta here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure beta gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the change of variables section for full definitions. For the case above, the following adjustment will account for the log transform.3\ntarget += - log(abs(y));\n\n\n\nStan supports truncating distributions with lower bounds, upper bounds, or both.\n\n\nA probability density function \\(p(x)\\) for a continuous distribution may be truncated to an interval \\([a, b]\\) to define a new density \\(p_{[a, b]}(x)\\) with support \\([a, b]\\) by setting\n\\[\np_{[a, b]}(x)\n= \\frac{p(x)}\n       {\\int_a^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) for a discrete distribution may be truncated to the closed interval \\([a, b]\\) by\n\\[\np_{[a, b]}(x) = \\frac{p(x)}\n                  {\\sum_{u = a}^b p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x)\n= \\frac{p(x)}\n       {\\int_a^{\\infty} p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x) = \\frac{p(x)}\n                  {\\sum_{a &lt;= u} p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty, b]}(x)\n= \\frac{p(x)}\n       {\\int_{-\\infty}^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty,b]}(x) = \\frac{p(x)}\n                  {\\sum_{u &lt;= b} p(u)}.\n\\]\n\n\n\nGiven a probability function \\(p_X(x)\\) for a random variable \\(X\\), its cumulative distribution function (cdf) \\(F_X(x)\\) is defined to be the probability that \\(X \\leq x\\),\n\\[\nF_X(x) = \\Pr[X \\leq x].\n\\]\nThe upper-case variable \\(X\\) is the random variable whereas the lower-case variable \\(x\\) is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to\n\\[\nF_X(x) \\ = \\ \\int_{-\\infty}^{x} p_X(u) \\, du,\n\\]\nFor discrete variables, the cdf is defined to include the upper bound given by the argument,\n\\[\nF_X(x) = \\sum_{u \\leq x} p_X(u).\n\\]\n\n\n\nThe complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by\n\\[\nF^C_X(x)\n\\ = \\ \\Pr[X &gt; x]\n\\ = \\ 1 - F_X(x).\n\\]\nUnlike the cdf, the ccdf is exclusive of the bound, hence the event \\(X &gt; x\\) rather than the cdf’s event \\(X \\leq x\\).\nFor continuous distributions, the ccdf works out to\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\int_{-\\infty}^x p_X(u) \\, du\n\\ = \\ \\int_x^{\\infty} p_X(u) \\, du.\n\\]\nThe lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over \\(u &gt; x\\),\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\sum_{u \\leq x} p_X(u)\n\\ = \\ \\sum_{u &gt; x} p_X(u).\n\\]\nCumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by \\[\n\\int_a^b p(u) \\, du = F_X(b) - F_X(a).\n\\] This allows truncated distributions to be defined as \\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a)}.\n\\]\nFor discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from \\(F_X(b) - F_X(a)\\),\n\\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a) + p_X(a)}.\n\\]\n\n\n\nStan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to \\([-0.5, 2.1]\\) can be coded with the following distribution statement.\ny ~ normal(0, 1) T[-0.5, 2.1];\nTruncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.\nIn general, the truncation bounds and parameters may be parameters or local variables.\nBecause the example above involves a continuous distribution, it behaves the same way as the following more verbose form.\ny ~ normal(0, 1);\nif (y &lt; -0.5 || y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -log_diff_exp(normal_lcdf(2.1 | 0, 1),\n                          normal_lcdf(-0.5 | 0, 1));\n}\nBecause a Stan program defines a log density function, all calculations are on the log scale. The function normal_lcdf is the log of the cumulative normal distribution function and the function log_diff_exp(a, b) is a more arithmetically stable form of log(exp(a) - exp(b)).\nFor a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution\ny ~ poisson(3.7) T[2, 10];\nbehaves in the same way as the following code.\ny ~ poisson(3.7);\nif (y &lt; 2 || y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         log_diff_exp(poisson_lcdf(10 | 3.7),\n                                      poisson_lcdf(2 | 3.7)));\n}\nRecall that log_sum_exp(a, b) is just the arithmetically stable form of log(exp(a) + exp(b)).\n\n\n\nFor truncating with only a lower bound, the upper limit is left blank.\ny ~ normal(0, 1) T[-0.5, ];\nThis truncated distribution statement has the same behavior as the following code.\ny ~ normal(0, 1);\nif (y &lt; -0.5) {\n  target += negative_infinity();\n} else {\n  target += -normal_lccdf(-0.5 | 0, 1);\n}\nThe normal_lccdf function is the normal complementary cumulative distribution function.\nAs with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus\ny ~ poisson(3.7) T[2, ];\nbehaves the same way as\ny ~ poisson(3.7);\nif (y &lt; 2) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         poisson_lccdf(2 | 3.7));\n}\n\n\n\nTo truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement\ny ~ normal(0, 1) T[ , 2.1];\nproduces the same result as the following code.\ntarget += normal_lpdf(y | 0, 1);\nif (y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -normal_lcdf(2.1 | 0, 1);\n}\nWith only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement\ny ~ poisson(3.7) T[ , 10];\nbehaves the same way as the following code.\ny ~ poisson(3.7);\nif (y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -poisson_lcdf(10 | 3.7);\n}\n\n\n\nIn all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.\n\n\n\nFor continuous distributions, truncation points must be expressions of type int or real. For discrete distributions, truncation points must be expressions of type int.\n\n\n\nFor a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds \\(-\\infty\\) to the total log probability, which in turn results in the sample being rejected.\n\n\n\nVectorization of distribution functions with truncation is available if the underlying distribution, lcdf, and lccdf functions meet the required signatures.\nThe equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):\n\nIf the variate y is the only non-scalar, the result is the same as described in the above sections, but the lcdf/lccdf calculation is multiplied by size(y).\nIf the other arguments to the distribution are non-scalars, then the vectorized version of the lcdf/lccdf is used. These functions return the sum of their terms, so no multiplication by the size is needed.\nThe exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the log_diff_exp of two sums is not the same as the sum of the pairwise log_diff_exp operations.\n\nNote that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.",
+    "text": "Stan’s functions all have declared types for both arguments and returned value. As with built-in functions, user-defined functions are only declared for base argument type and dimensionality. This requires a different syntax than for declaring other variables. The choice of language was made so that return types and argument types could use the same declaration syntax.\nThe type void may not be used as an argument type, only a return type for a function with side effects.\n\n\nThe base variable types are integer, real, complex, vector, row_vector, and matrix. No lower-bound or upper-bound constraints are allowed (e.g., real&lt;lower=0&gt; is illegal). Specialized constrained types are also not allowed (e.g., simplex is illegal).\nTuple types of the form tuple(T1, ..., TN) are also allowed, with all of the types T1 to TN being function argument types (i.e., no constraints and no sizes).\n\n\n\nArguments and return types may be arrays, and these are indicated with optional brackets and commas as would be used for indexing. For example, int denotes a single integer argument or return, whereas array[] real indicates a one-dimensional array of reals, array[,] real a two-dimensional array and array[,,] real a three-dimensional array; whitespace is optional, as usual.\nThe dimensions for vectors and matrices are not included, so that matrix is the type of a single matrix argument or return type. Thus if a variable is declared as matrix a, then a has two indexing dimensions, so that a[1] is a row vector and a[1, 1] a real value. Matrices implicitly have two indexing dimensions. The type declaration matrix[ , ] b specifies that b is a two-dimensional array of matrices, for a total of four indexing dimensions, with b[1, 1, 1, 1] picking out a real value.\n\n\n\nFunction argument and return types are not themselves checked for dimensionality. A matrix of any size may be passed in as a matrix argument. Nevertheless, a user-defined function might call a function (such as a multivariate normal density) that itself does dimensionality checks.\nDimensions of function return values will be checked if they’re assigned to a previously declared variable. They may also be checked if they are used as the argument to a function.\nAny errors raised by calls to functions inside user functions or return type mismatches are simply passed on; this typically results in a warning message and rejection of a proposal during sampling or optimization.\n\n\n\nSome of Stan’s built-in functions, like the differential equation solvers, have arguments that must be data. Such data-only arguments must be expressions involving only data, transformed data, and generated quantity variables.\nIn user-defined functions, the qualifier data may be placed before an argument type declaration to indicate that the argument must be data only. For example,\nreal foo(data real x) {\n  return x^2;\n}\nrequires the argument x to be data only.\nDeclaring an argument data only allows type inference to proceed in the body of the function so that, for example, the variable may be used as a data-only argument to a built-in function.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#for-loops",
-    "href": "reference-manual/statements.html#for-loops",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#function-bodies.section",
+    "href": "reference-manual/user-functions.html#function-bodies.section",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Suppose N is a variable of type int, y is a one-dimensional array of type array[] real, and mu and sigma are variables of type real. Furthermore, suppose that n has not been defined as a variable. Then the following is a well-formed for-loop statement.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nThe loop variable is n, the loop bounds are the values in the range 1:N, and the body is the statement following the loop bounds.\n\n\nThe type of the loop variable is int. Unlike in C++ and similarly to R, this variable must not be declared explicitly.\nThe bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range L:H will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to L and less than or equal to H. For example, the loop for (n in 2:5) will cause the body of the for loop to be executed with n equal to 2, 3, 4, and 5, in order. The variable and bound for (n in   5:2) will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.\nThe scope of the loop variable is limited to the body of the loop.\n\n\n\nUnlike in BUGS, Stan allows variables to be reassigned. For example, the variable theta in the following program is reassigned in each iteration of the loop.\nfor (n in 1:N) {\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nSuch reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.4\nIn Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.\nfor (n in 1:N) {\n  y[n] ~ bernoulli(theta);\n  theta = inv_logit(alpha + x[n] * beta);\n}\nIn this program, theta is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.\nStan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.\ntotal = 0.0;\nfor (n in 1:N) {\n  total = total + x[n];\n}\nAfter the for loop is executed, the variable total will hold the sum of the elements in the array x. This example was purely pedagogical; it is easier and more efficient to write\ntotal = sum(x);\nA variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.\nfor (n in 1:100) {\n  y += y * epsilon;\n  epsilon = 0.5 * epsilon;\n  y += y * epsilon;\n}",
+    "text": "The body of a function is between an open curly brace ({) and close curly brace (}). The body may contain local variable declarations at the top of the function body’s block and these scope the same way as local variables used in any other statement block.\nAny user-defined function may be used in the function body regardless of the order in which the function definitions appear in the file. Self-recursive and mutually recursive functions are possible without any additional declarations.\nThe only restrictions on statements in function bodies are external, and determine whether the log probability accumulator or random number generators are available; see the rest of this section for details.\n\n\nFunctions that call random number generating functions in their bodies must have a name that ends in _rng; attempts to use random-number generators in other functions lead to a compile-time error.\nLike other random number generating functions, user-defined functions with names that end in _rng may be used only in the generated quantities block and transformed data block, or within the bodies of user-defined functions ending in _rng. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions that include distribution statements or log probability increment statements must have a name that ends in _lp. Attempts to use distribution statements or increment log probability statements in other functions lead to a compile-time error.\nLike the target log density increment statement and distribution statements, user-defined functions with names that end in _lp may only be used in blocks where the log probability accumulator is accessible, namely the transformed parameters and model blocks. An attempt to use such a function elsewhere results in a compile-time error.\n\n\n\nFunctions whose names end in _lpdf and _lpmf (density and mass functions) can be used as probability functions in distribution statements. As with the built-in functions, the first argument will appear on the left of the distribution statement operator (~) in the distribution statement and the other arguments follow. For example, suppose a function returning the log of the density of y given parameter theta allows the use of the distribution statement is defined as follows.\nreal foo_lpdf(real y, vector theta) { ... }\nNote that for function definitions, the comma is used rather than the vertical bar.\nFor every custom _lpdf and _lpmf defined there is a corresponding _lupdf and _lupmf defined automatically. The _lupdf and _lupmf versions of the functions cannot be defined directly (to do so will produce an error). The difference in the _lpdf and _lpmf and the corresponding _lupdf and _lupmf functions is that if any other unnormalized density functions are used inside the user-defined function, the _lpdf and _lpmf forms of the user-defined function will change these densities to be normalized. The _lupdf and _lupmf forms of the user-defined functions will instead allow other unnormalized density functions to drop additive constants.\nThe distribution statement shorthand\nz ~ foo(phi);\nwill have the same effect as incrementing the target with the log of the unnormalized density:\ntarget += foo_lupdf(z | phi);\nOther _lupdf and _lupmf functions used in the definition of foo_lpdf will drop additive constants when foo_lupdf is called and will not drop additive constants when foo_lpdf is called.\nIf there are _lupdf and _lupmf functions used inside the following call to foo_lpdf, they will be forced to normalize (return the equivalent of their _lpdf and _lpmf forms):\ntarget += foo_lpdf(z | phi);\nIf there are no _lupdf or _lupmf functions used in the definition of foo_lpdf, then there will be no difference between a foo_lpdf or foo_lupdf call.\nThe unnormalized _lupdf and _lupmf functions can only be used in the model block or in user-defined probability functions (those ending in _lpdf or _lpmf).\nThe same syntax and shorthand that works for _lpdf also works for log probability mass functions with suffixes _lpmf.\nA function that is going to be accessed as distributions must return the log of the density or mass function it defines.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#foreach-loops",
-    "href": "reference-manual/statements.html#foreach-loops",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#parameters-are-constant",
+    "href": "reference-manual/user-functions.html#parameters-are-constant",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "A second form of for loops allows iteration over elements of containers. If ys is an expression denoting a container (vector, row vector, matrix, or array) with elements of type T, then the following is a well-formed foreach statement.\nfor (y in ys) {\n  // ... do something with y ...\n}\nThe order in which elements of ys are visited is defined for container types as follows.\n\nvector, row_vector: elements visited in order, y is of type double\nmatrix: elements visited in column-major order, y is of type double\narray[] T: elements visited in order, y is of type T.\n\nConsequently, if ys is a two dimensional array array[,] real, y will be a one-dimensional array of real values (type array[] real). If ’ysis a matrix, thenywill be a real value (typereal`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,\narray[2, 3] real yss;\nfor (ys in yss) {\n  for (y in ys) {\n    // ... do something with y ...\n  }\n}\nwhereas a matrix can be looped over in one foreach statement\nmatrix[2, 3] yss;\nfor (y in yss) {\n   // ... do something with y...\n}\nIn both cases, the loop variable y is of type real.  The elements of the matrix are visited in column-major order (e.g.,y[1, 1],y[2, 1],y[1, 2], ...,y[2, 3]), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,y[1, 1],y[1, 2],y[1, 3],y[2, 1], ...,y[2, 3]`).",
+    "text": "Within function definition bodies, the parameters may be used like any other variable. But the parameters are constant in the sense that they can’t be assigned to (i.e., can’t appear on the left side of an assignment (=) statement). In other words, their value remains constant throughout the function body. Attempting to assign a value to a function parameter value will raise a compile-time error.1\nLocal variables may be declared at the top of the function block and scope as usual.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#conditional-statements",
-    "href": "reference-manual/statements.html#conditional-statements",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#return-value",
+    "href": "reference-manual/user-functions.html#return-value",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Stan supports full conditional statements using the same if-then-else syntax as C++. The general format is\nif (condition1)\n  statement1\nelse if (condition2)\n  statement2\n// ...\nelse if (conditionN-1)\n  statementN-1\nelse\n  statementN\nThere must be a single leading if clause, which may be followed by any number of else if clauses, all of which may be optionally followed by an else clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.\nThe entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.",
+    "text": "Non-void functions must have a return statement that returns an appropriately typed expression. If the expression in a return statement does not have the same type as the return type declared for the function, a compile-time error is raised.\nVoid functions may use return only without an argument, but return statements are not mandatory.\n\n\nUnlike C++, Stan enforces a syntactic guarantee for non-void functions that ensures control will leave a non-void function through an appropriately typed return statement or because an exception is raised in the execution of the function. To enforce this condition, functions must have a return statement as the last statement in their body. This notion of last is defined recursively in terms of statements that qualify as bodies for functions. The base case is that\n\na return statement qualifies,\n\nand the recursive cases are that\n\na sequence of statements qualifies if its last statement qualifies,\na for loop or while loop qualifies if its body qualifies, and\na conditional statement qualifies if it has a default else clause and all of its body statements qualify.\n\nAn exception is made for “obviously infinite” loops like while (1), which contain a return statement and no break statements. The only way to exit such a loop is to return, so they are considered as returning statements.\nThese rules disqualify\nreal foo(real x) {\n  if (x &gt; 2) {\n    return 1.0;\n  } else if (x &lt;= 2) {\n    return -1.0;\n  }\n}\nbecause there is no default else clause, and disqualify\nreal foo(real x) {\n  real y;\n  y = x;\n  while (x &lt; 10) {\n    if (x &gt; 0) {\n      return x;\n    }\n    y = x / 2;\n  }\n}\nbecause the return statement is not the last statement in the while loop. A bogus dummy return could be placed after the while loop in this case. The rules for returns allow\nreal log_fancy(real x) {\n  if (x &lt; 1e-30) {\n    return x;\n  } else if (x &lt; 1e-14) {\n    return x * x;\n  } else {\n    return log(x);\n  }\n}\nbecause there’s a default else clause and each condition body has return as its final statement.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#while-statements",
-    "href": "reference-manual/statements.html#while-statements",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#void-functions-as-statements",
+    "href": "reference-manual/user-functions.html#void-functions-as-statements",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Stan supports standard while loops using the same syntax as C++. The general format is as follows.\nwhile (condition)\n  body\nThe condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).\nEvaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.\nThe rest of the body of a while loop may be skipped using a continue. The loop will be exited with a break statement. See the section on continue and break statements for more details.",
+    "text": "A function can be declared without a return value by using void in place of a return type. Note that the type void may only be used as a return type—arguments may not be declared to be of type void.\n\n\n\nA void function may be used as a statement.\nBecause there is no return, such a usage is only for side effects, such as incrementing the log probability function, printing, or raising an error.\n\n\n\nIn a return statement within a void function’s definition, the return keyword is followed immediately by a semicolon (;) rather than by the expression whose value is returned.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#statement-blocks-and-local-variable-declarations",
-    "href": "reference-manual/statements.html#statement-blocks-and-local-variable-declarations",
-    "title": "Statements",
+    "objectID": "reference-manual/user-functions.html#forward-declarations.section",
+    "href": "reference-manual/user-functions.html#forward-declarations.section",
+    "title": "User-Defined Functions",
     "section": "",
-    "text": "Just as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.\n\n\nBlocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nTo put multiple statements inside the body of a for loop, a block is used, as in the following example.\nfor (n in 1:N) {\n  lambda[n] ~ gamma(alpha, beta);\n  y[n] ~ poisson(lambda[n]);\n}\nThe open curly bracket ({) is the first character of the block and the close curly bracket (}) is the last character.\nBecause whitespace is ignored in Stan, the following program will not compile.\nfor (n in 1:N)\n  y[n] ~ normal(mu, sigma);\n  z[n] ~ normal(mu, sigma); // ERROR!\nThe problem is that the body of the for loop is taken to be the statement directly following it, which is y[n] ~ normal(mu, sigma). This leaves the probability statement for z[n] hanging, as is clear from the following equivalent program.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nz[n] ~ normal(mu, sigma); // ERROR!\nNeither of these programs will compile. If the loop variable n was defined before the for loop, the for-loop declaration will raise an error. If the loop variable n was not defined before the for loop, then the use of the expression z[n] will raise an error.\n\n\n\nA for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.\nfor (n in 1:N) {\n  real theta;\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nThe local variable theta is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus theta is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable theta if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.\nfor (m in 1:M) {\n  real theta;\n  for (n in 1:N) {\n    real theta; // ERROR!\n    theta = inv_logit(alpha + x[m, n] * beta);\n    y[m, n] ~ bernoulli(theta);\n// ...\nThe compiler will flag the second declaration of theta with a message that it is already defined.\n\n\n\nLocal variables may not have constraints on their declaration. The only types that may be used are listed in the types table under “local”.\n\n\n\nA block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following\nfor (m in 1:M) {\n  {\n     int n = 2 * m;\n     sum += n;\n  }\n  for (n in 1:N) {\n    sum += x[m, n];\n  }\n}\nThe variable declaration int n; is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of n are unrelated.",
+    "text": "Stan supports forward declarations, which look like function definitions without bodies. For example,\nreal unit_normal_lpdf(real y);\ndeclares a function named unit_normal_lpdf that consumes a single real-valued input and produces a real-valued output. Declaring a function without a definition is only really useful when using an extension which supplies the definition in C++ rather than in the Stan code itself. How exactly this can be accomplished will differ depending on your Stan interface.\nA function definition with a body simultaneously declares and defines the named function, as in\nreal unit_normal_lpdf(real y) {\n  return -0.5 * square(y);\n}\nA function can be declared and (perhaps separately) defined at most once. However, functions with different argument types are considered distinct even if they have the same name; see the section on function overloading.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#break-continue-statements",
-    "href": "reference-manual/statements.html#break-continue-statements",
-    "title": "Statements",
-    "section": "",
-    "text": "The one-token statements continue and break may be used within loops to alter control flow; continue causes the next iteration of the loop to run immediately, whereas break terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both break and continue scope to the most deeply nested loop, but pass through non-loop statements.\nAlthough these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.\n\n\nWhen a break statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:\nwhile (1) {\n  if (n &lt; 0) {\n    break;\n  }\n  foo(n);\n  n = n - 1;\n}\nThe while~(1) loop is a “forever” loop, because 1 is the true value, so the test always succeeds. Within the loop, if the value of n is less than 0, the loop terminates, otherwise it executes foo(n) and then decrements n. The statement above does exactly the same thing as\nwhile (n &gt;= 0) {\n  foo(n);\n  n = n - 1;\n}\nThis case is simply illustrative of the behavior; it is not a case where a break simplifies the loop.\n\n\n\nThe continue statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array x,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &lt;= 0) {\n    continue;\n  }\n  sum += x[n];\n}\nWhen the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an x[n] less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &gt; 0) {\n    sum += x[n];\n  }\n}\nWhile the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.\n\n\n\nIf there is a loop nested within a loop, a break or continue statement only breaks out of the inner loop. So\nwhile (cond1) {\n  // ...\n  while (cond2) {\n    // ...\n    if (cond3) {\n      break;\n    }\n    // ...\n  }\n  // execution continues here after break\n  // ...\n}\nIf the break is triggered by cond3 being true, execution will continue after the nested loop.\nAs with break statements, continue statements go back to the top of the most deeply nested loop in which the continue appears.\nAlthough break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.",
+    "objectID": "reference-manual/user-functions.html#footnotes",
+    "href": "reference-manual/user-functions.html#footnotes",
+    "title": "User-Defined Functions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nDespite being declared constant and appearing to have a pass-by-value syntax in Stan, the implementation of the language passes function arguments by constant reference in C++.↩︎",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "User-Defined Functions"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#print-statements.section",
-    "href": "reference-manual/statements.html#print-statements.section",
-    "title": "Statements",
+    "objectID": "reference-manual/transforms.html",
+    "href": "reference-manual/transforms.html",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Stan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.\nfor (n in 1:N) { print(\"loop iteration: \", n); ... }\nThe print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression n, followed by a new line.\n\n\nThe text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.\nPrinting a simple real or int typed variable always prints the variable’s value.5\nFor array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as\n[1, 2, 3]\nand a \\(2 \\times 3\\)-matrix as\n[[1, 2, 3], [4, 5, 6]]\nComplex numbers print as pairs. For example, the pair of statements\ncomplex z = to_complex(1.2, -3.5);\nprint(z)\nwill print as (1.2,-3.5), with no space after the comma or within the parentheses.\nPrinting a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.\ntransformed data {\n  matrix[2, 2] u;\n  u[1, 1] = 1.0;   u[1, 2] = 4.0;\n  u[2, 1] = 9.0;   u[2, 2] = 16.0;\n  for (n in 1:2) {\n    print(\"u[\", n, \"] = \", u[n]);\n  }\n}\nThis print statement executes twice, printing the following two lines of output.\nu[1] = [1, 4]\nu[2] = [9, 16]\n\n\n\nThe input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.\n\n\n\nPrinting for a print statement happens every time it is executed. The transformed data block is executed once per chain, the transformed parameter and model blocks once per leapfrog step, and the generated quantities block once per iteration.\n\n\n\nString literals begin and end with a double quote character (\"). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.\nThe Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.\n\n\n\nBecause Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.\nIt is particularly useful to print the value of the target log density accumulator (through the target() function), as in the following example.\nvector[2] y;\ny[1] = 1;\nprint(\"log density before =\", target());\ny ~ normal(0,1);  // bug!  y[2] not defined\nprint(\"log density after =\", target());\nThe example has a bug in that y[2] is not defined before the vector y is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).\nNote that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.",
+    "text": "To avoid having to deal with constraints while simulating the Hamiltonian dynamics during sampling, every (multivariate) parameter in a Stan model is transformed to an unconstrained variable behind the scenes by the model compiler. The transform is based on the constraints, if any, in the parameter’s definition. Scalars or the scalar values in vectors, row vectors or matrices may be constrained with lower and/or upper bounds. Vectors may alternatively be constrained to be ordered, positive ordered, or simplexes. Matrices may be constrained to be correlation matrices or covariance matrices. This chapter provides a definition of the transforms used for each type of variable. For examples of how to declare and define these variables in a Stan program, see section Variable declaration. To directly access the functional form of these transformations from inside the Stan language, see Variable Transformation Functions in the Functions Reference.\nStan converts models to C++ classes which define probability functions with support on all of \\(\\mathbb{R}^K\\), where \\(K\\) is the number of unconstrained parameters needed to define the constrained parameters defined in the program. The C++ classes also include code to transform the parameters from unconstrained to constrained and apply the appropriate Jacobians.\n\n\nIn this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.\nStan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with lower=0. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about floating point arithmetic in the Stan User’s Guide.\nThese issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the CmdStan Command-Line Interface Overview.\n\n\n\nThe support of a random variable \\(X\\) with density \\(p_X(x)\\) is that subset of values for which it has non-zero density,\n\\[\n\\mathrm{supp}(X) = \\{ x | p_X(x) &gt; 0 \\}.\n\\]\nIf \\(f\\) is a total function defined on the support of \\(X\\), then \\(Y =\nf(X)\\) is a new random variable. This section shows how to compute the probability density function of \\(Y\\) for well-behaved transforms \\(f\\). The rest of the chapter details the transforms used by Stan.\n\n\nSuppose \\(X\\) is one dimensional and \\(f: \\mathrm{supp}(X) \\rightarrow\n\\mathbb{R}\\) is a one-to-one, monotonic function with a differentiable inverse \\(f^{-1}\\). Then the density of \\(Y\\) is given by\n\\[\np_Y(y) = p_X(f^{-1}(y))\n         \\,\n         \\left| \\, \\frac{d}{dy} f^{-1}(y)\\, \\right|.\n\\]\nThe absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.\n\n\n\nThe multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.\nSuppose \\(X\\) is a \\(K\\)-dimensional random variable with probability density function \\(p_X(x)\\). A new random variable \\(Y = f(X)\\) may be defined by transforming \\(X\\) with a suitably well-behaved function \\(f\\). It suffices for what follows to note that if \\(f\\) is one-to-one and its inverse \\(f^{-1}\\) has a well-defined Jacobian, then the density of \\(Y\\) is\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\, \\left| \\, \\det \\, J_{f^{-1}}(y) \\, \\right|,\n\\]\nwhere \\(\\det{}\\) is the matrix determinant operation and \\(J_{f^{-1}}(y)\\) is the Jacobian matrix of \\(f^{-1}\\) evaluated at \\(y\\). Taking \\(x =\nf^{-1}(y)\\), the Jacobian matrix is defined by\n\\[\nJ_{f^{-1}}(y) =\n\\left[\n\\begin{array}{ccc}\\displaystyle\n\\frac{\\partial x_1}{\\partial y_1}\n& \\cdots\n& \\displaystyle \\frac{\\partial x_1}{\\partial y_{K}}\n\\\\\n\\vdots & \\vdots & \\vdots\n\\\\\n\\displaystyle\\frac{\\partial x_{K}}{\\partial y_1}\n& \\cdots\n& \\displaystyle\\frac{\\partial x_{K}}{\\partial y_{K}}\n\\end{array}\n\\right].\n\\]\nIf the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,\n\\[\n\\det \\, J_{f^{-1}}(y)\n= \\prod_{k=1}^K \\frac{\\partial x_k}{\\partial y_k}.\n\\]\nTriangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.\n\n\n\n\nStan uses a logarithmic transform for lower and upper bounds.\n\n\nIf a variable \\(X\\) is declared to have lower bound \\(a\\), it is transformed to an unbounded variable \\(Y\\), where\n\\[\nY = \\log(X - a).\n\\]\n\n\n\nThe inverse of the lower-bound transform maps an unbounded variable \\(Y\\) to a variable \\(X\\) that is bounded below by \\(a\\) by\n\\[\nX = \\exp(Y) + a.\n\\]\n\n\n\nThe absolute derivative of the inverse transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( \\exp(y) + a \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, given the density \\(p_X\\) of \\(X\\), the density of \\(Y\\) is\n\\[\np_Y(y)\n= p_X\\!\\left( \\exp(y) + a \\right) \\cdot \\exp(y).\n\\]\n\n\n\n\nStan uses a negated logarithmic transform for upper bounds.\n\n\nIf a variable \\(X\\) is declared to have an upper bound \\(b\\), it is transformed to the unbounded variable \\(Y\\) by\n\\[\nY = \\log(b - X).\n\\]\n\n\n\nThe inverse of the upper bound transform converts the unbounded variable \\(Y\\) to the variable \\(X\\) bounded above by \\(b\\) through\n\\[\nX = b - \\exp(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse of the upper bound transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( b - \\exp(y) \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, the density of the unconstrained variable \\(Y\\) is defined in terms of the density of the variable \\(X\\) with an upper bound of \\(b\\) by\n\\[\np_Y(y)\n=   p_X \\!\\left( b - \\exp(y) \\right) \\cdot \\exp(y).\n\\]\n\n\n\n\nFor lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.\n\n\nThe log-odds function is defined for \\(u \\in (0,1)\\) by\n\\[\n\\mathrm{logit}(u) = \\log \\frac{u}{1 - u}.\n\\]\nThe inverse of the log odds function is the logistic sigmoid, defined for \\(v \\in (-\\infty,\\infty)\\) by\n\\[\n\\mathrm{logit}^{-1}(v) = \\frac{1}{1 + \\exp(-v)}.\n\\]\nThe derivative of the logistic sigmoid is\n\\[\n\\frac{d}{dy} \\mathrm{logit}^{-1}(y)\n= \\mathrm{logit}^{-1}(y) \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\n\n\n\nFor variables constrained to be in the open interval \\((a, b)\\), Stan uses a scaled and translated log-odds transform. If variable \\(X\\) is declared to have lower bound \\(a\\) and upper bound \\(b\\), then it is transformed to a new variable \\(Y\\), where\n\\[\nY = \\mathrm{logit} \\left( \\frac{X - a}{b - a} \\right).\n\\]\n\n\n\nThe inverse of this transform is\n\\[\nX = a + (b - a) \\cdot \\mathrm{logit}^{-1}(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse transform is given by\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      a + (b - a) \\cdot \\mathrm{logit}^{-1}(y)\n    \\right)\n  \\right|\n= (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( a + (b - a) \\cdot \\mathrm{logit}^{-1}(y) \\right)\n    \\cdot (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nDespite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, \\(\\mathrm{logit}^{-1}(y)\\) only needs to be evaluated once, so there is only one call to \\(\\exp(-y)\\).\n\n\n\n\nStan uses an affine transform to be able to specify parameters with a given offset and multiplier.\n\n\nFor variables with expected offset \\(\\mu\\) and/or (positive) multiplier \\(\\sigma\\), Stan uses an affine transform. Such a variable \\(X\\) is transformed to a new variable \\(Y\\), where\n\\[\nY = \\frac{X - \\mu}{\\sigma}.\n\\]\nThe default value for the offset \\(\\mu\\) is \\(0\\) and for the multiplier \\(\\sigma\\) is \\(1\\) in case not both are specified.\n\n\n\nThe inverse of this transform is\n\\[\nX =  \\mu + \\sigma \\cdot Y.\n\\]\n\n\n\nThe absolute derivative of the affine inverse transform is\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      \\mu + \\sigma \\cdot y\n    \\right)\n  \\right|\n= \\sigma.\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( \\mu + \\sigma \\cdot y \\right)\n    \\cdot \\sigma.\n\\]\nFor an example of how to code this in Stan, see section Affinely Transformed Real.\n\n\n\n\nFor some modeling tasks, a vector-valued random variable \\(X\\) is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.\nIn constraint terms, an ordered \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies\n\\[\nx_k &lt; x_{k+1}\n\\]\nfor \\(k \\in \\{ 1, \\ldots, K-1 \\}\\).\n\n\nStan’s transform follows the constraint directly. It maps an increasing vector \\(x \\in \\mathbb{R}^{K}\\) to an unconstrained vector \\(y \\in\n\\mathbb{R}^K\\) by setting\n\\[\ny_k\n=\n\\left\\{\n\\begin{array}{ll}\nx_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\log \\left( x_{k} - x_{k-1} \\right) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform for an unconstrained \\(y \\in \\mathbb{R}^K\\) to an ordered sequence \\(x \\in \\mathbb{R}^K\\) is defined by the recursion\n\\[\nx_k\n=\n\\left\\{\n\\begin{array}{ll}\ny_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\nx_{k-1} + \\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\\(x_k\\) can also be expressed iteratively as\n\\[\nx_k = y_1 + \\sum_{k'=2}^k \\exp(y_{k'}).\n\\]\n\n\n\nThe Jacobian of the inverse transform \\(f^{-1}\\) is lower triangular, with diagonal elements for \\(1 \\leq k \\leq K\\) of\n\\[\nJ_{k,k} =\n\\left\\{\n\\begin{array}{ll}\n1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\nBecause \\(J\\) is triangular, the absolute Jacobian determinant is\n\\[\n\\left| \\, \\det \\, J \\, \\right|\n\\ = \\\n\\left| \\, \\prod_{k=1}^K J_{k,k} \\, \\right|\n\\ = \\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\nPutting this all together, if \\(p_X\\) is the density of \\(X\\), then the transformed variable \\(Y\\) has density \\(p_Y\\) given by\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n\\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\n\n\n\n\nPlaceholder.\nThe positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.\n\n\n\nStan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length \\(N\\) with real values and the sum of the vector equals zero. The sum-to-zero matrix is an \\(N \\times M\\) matrix where both the rows and columns sum-to-zero.\nStan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix \\(H\\) such that \\(H\\in\\mathbb R^{N\\times (N-1)}\\) and \\(H^{\\mathsf T}H=I_{N-1},\\;H^{\\mathsf T}\\mathbf 1=0\\). The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,(Seyboldt 2024). Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.\nIn many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:\n\\[\n  \\sigma^2 \\begin{pmatrix}\n  1-\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots&-\\tfrac{1}{N} \\\\\n  -\\tfrac{1}{N} & 1-\\tfrac{1}{N} & \\cdots & -\\tfrac{1}{N} \\\\\n  \\vdots & \\vdots & \\ddots & \\vdots \\\\\n  -\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots &1-\\tfrac{1}{N}\n  \\end{pmatrix}.\n\\]\nThe marginal standard deviation no longer corresponds to \\(\\sigma\\) but is \\(\\sigma \\sqrt{1-\\tfrac{1}{N}}\\). The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of \\(\\sqrt{1-\\tfrac{1}{N}}\\) to adjust the variance to the intended \\(\\sigma\\):\nsum_to_zero_vector[N] x;\nx ~ normal(0, sqrt(N %/% (N - 1)));\nWhen \\(\\sigma\\) is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let \\(y\\) be the unconstrained \\(N - 1\\) vector and \\(y\\) is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of \\(\\sigma\\) is given by \\[\nx = \\underbrace{\\sigma\\sqrt{\\frac{N}{N-1}}\\;H}_{\\text{ size of } N \\times (N - 1)}\\,y.\n\\] This is the classic non-centered model. The crucial detail from above is that the operation of multiplying \\(\\sigma\\) to \\(y\\) is on \\(N - 1\\) dimensions, not \\(N\\) dimensions. When writing the centered model using \\[\nx \\sim \\mathcal N \\bigg(0,\\, \\sigma\\sqrt{\\frac{N}{N-1}} \\bigg),\n\\] we are incrementing the log density by an additional \\(-\\log(\\sigma)\\) and so must increment the log density by\ntarget += log(sigma * sqrt(N * inv(N - 1)));\n\n// or\n// because `sqrt(N * inv(N - 1))` is constant\n\ntarget += log(sigma);\nwhich correctly adjusts for the \\(N - 1\\) free variables.\nMore details of the Helmert matrix are in (Lancaster 1965), for the basic definitions of the isometric log ratio transform see (Egozcue et al. 2003) and Chapter 3 of (Filzmoser, Hron, and Templ 2018) for the pivot coordinate version used here.\n\n\nVectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).\nA sum-to-zero \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies the constraint \\[\n\\sum_{k=1}^K x_k = 0.\n\\]\n\n\n\nThe transform is defined iteratively. Given an \\(x \\in \\mathbb{R}^{N + 1}\\) that sums to zero (i.e., \\(\\sum_{n=1}^{N+1} x_n = 0\\)), the transform proceeds as follows to produce an unconstrained \\(y \\in \\mathbb{R}^N\\). This is mathematically equivalent to pre-multiplying the unconstrained \\(y\\) by an orthogonal standard basis, e.g. constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.\nThe transform is initialized by setting \\[\nS_N = 0\n\\] and \\[\ny_N = -x_{N + 1} \\sqrt{1 + \\frac{1}{N}}.\n\\] The for each \\(n\\) from \\(N - 1\\) down to \\(1\\), let \\[\nw_{n + 1} = \\frac{y_{n + 1}}{\\sqrt{(n + 1)(n + 2)}},\n\\] \\[\nS_n = S_{n + 1} + w_{n + 1},\n\\] and \\[\ny_n = (S_n - x_{n + 1}) \\frac{\\sqrt{n (n + 1)}}{n}.\n\\]\nThis transform is expressed in Stan code as:\n  vector manual_sum_to_zero_jacobian(vector x) {\n    int N = size(x) - 1;\n    vector[N] y;\n    y[N] = -x[N+1] * sqrt(1 + 1. / N);\n    real sum_w = 0;\n    for (n in 1:(N-1)) {\n      int i = N - n;\n      int i_p_1 = i + 1;\n      real w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + 1));\n      sum_w += w;\n      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;\n    }\n    return y;\n  }\n\n\n\nThe inverse transform follows the isometric logratio tranform. It maps an unconstrained vector \\(y \\in \\mathbb{R}^N\\) to a sum-to-zero vector \\(x \\in \\mathbb{R}^{N + 1}\\) such that \\[\n\\sum_{n=1}^{N + 1} x_n = 0.\n\\] The values are defined inductively, starting with \\[\nx_1 = \\sum_{n=1}^N \\frac{y_n}{\\sqrt{n (n + 1)}}\n\\] and then setting \\[\nx_{n + 1} = \\sum_{i = n + 1}^N \\frac{y_i}{\\sqrt{i (i + 1)}}\n- n \\cdot \\frac{y_n}{\\sqrt{n (n + 1)}}.\n\\] for \\(n \\in 1{:}N\\).\nThe definition is such that \\[\n\\sum_{n = 1}^{N + 1} x_n = 0\n\\] by construction, because each of the terms added to \\(x_{n}\\) is then subtracted from \\(x_{n + 1}\\) the number of times it shows up in earlier terms.\n\n\n\nThe inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.\nThe sum-to-zero inverse transform is expressed within Stan as:\n  vector manual_sum_to_zero_inverse_jacobian(vector y) {\n    int N = num_elements(y);\n    vector[N + 1] x = zeros_vector(N + 1);\n    real sum_w = 0;\n    for (n in 1:N) {\n      int i = N - n + 1;\n      real w = y[i] * inv_sqrt(i * (i + 1));\n      sum_w += w;\n      x[i] += sum_w;\n      x[i + 1] -= w * i;\n    }\n    return x;\n  }\nNote that there is no target += increment because the Jacobian is zero.\n\n\n\nThe matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any \\(N\\)-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.\nLet the unconstrained matrix be \\[\n\\mathcal Y \\in \\mathbb R^{n_1 \\times n_2}\n\\] and the zero sum vector transform as \\[\n\\mathbf z = \\mathcal C_n(\\mathbf y)\\;=\\;\n\\begin{bmatrix}H_n\\\\[2pt]-\\mathbf 1_{1\\times d}\\end{bmatrix}\\mathbf y\n\\;\\in\\mathbb R^{n+1},\n\\] where \\(H_n\\in\\mathbb R^{n \\times n}\\) is the orthogonal Helmert matrix and satisfies \\(\\mathbf 1_{1\\times d}A_d^{\\!\\top}=0\\).\nApplying \\(C_n\\) to each slice results in\n\\[\n\\mathcal Z \\;=\\;\n\\mathcal X\n\\times_1 \\bigl[\\mathcal C_{n_1}\\bigr]\n\\times_2 \\bigl[\\mathcal C_{n_2}\\bigr]\n\\] where \\[\n\\mathcal Z \\in\n\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}.\n\\]\nBecause each \\(\\mathbb R^{d_1\\times\\cdots\\times d_N}\\) is invertible on the \\(\\mathbf 1^\\perp\\) subspace the composite map applied to \\(\\mathcal Z\\) is a linear bijection between \\(\\mathbb R^{n_1\\times n_2}\\) and the codomain \\(\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}\\).\n\n\n\n\nVariables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).\nThe unit \\(K\\)-simplex is the set of points \\(x \\in \\mathbb{R}^K\\) such that for \\(1 \\leq k \\leq K\\),\n\\[\nx_k &gt; 0,\n\\]\nand\n\\[\n\\sum_{k=1}^K x_k = 1.\n\\]\nAn alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values \\((0,1)\\), and \\((1,0)\\) and the unit 2-simplex is the line connecting these two points; values such as \\((0.3,0.7)\\) and \\((0.99,0.01)\\) lie on the line. In 3-dimensions, the basis is \\((0,0,1)\\), \\((0,1,0)\\) and \\((1,0,0)\\) and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include \\((0.5,0.5,0)\\), \\((0.2,0.7,0.1)\\) and all other triplets of non-negative values summing to 1.\nAs these examples illustrate, the simplex always picks out a subspace of \\(K-1\\) dimensions from \\(\\mathbb{R}^K\\). Therefore a point \\(x\\) in the \\(K\\)-simplex is fully determined by its first \\(K-1\\) elements \\(x_1, x_2,\n\\ldots, x_{K-1}\\), with\n\\[\nx_K = 1 - \\sum_{k=1}^{K-1} x_k.\n\\]\n\n\nThe length-\\(K\\) unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length \\(K\\).\nLet \\(y\\) represent the unconstrained \\(K - 1\\) values in \\((-\\infty, \\infty)\\). The intermediate sum-to-zero vector \\(z = \\text{sum\\_to\\_zero\\_transform}(y)\\) is length \\(K\\). The unit simplex is then given by \\[\nx_i = \\text{softmax}(z) = \\frac{\\exp(z_i)}{\\sum_{i = 1}^K \\exp(z_i)}\n\\]\nThe sum-to-zero vector transform is described in further detail at the sum-to-zero vector section of the Reference Manual.\n\n\n\n\n\n\nNote\n\n\n\nAll versions of Stan pre-2.37 used the stick-breaking transform. This is documented at Stan 2.36 Reference Manual: Simplex Transform.\n\n\n\n\nThe Jacobian \\(J\\) of the inverse unit-simplex transform is found by restricting \\(J\\) to the subspace spanned by the sum-to-zero vector \\(z\\). The Jacobian is given as the \\((K - 1) \\times (K - 1)\\) matrix \\(J\\) where\n\\[\nJ_{ij} = \\frac{\\partial x_i}{\\partial z_j} =\n\\frac{\\partial}{\\partial z_i} \\left( \\frac{\\exp(z_i)}{{\\sum_{i = 1}^K \\exp(z_i)}} \\right)\n\\] and \\(i,j \\in 1, \\ldots, K - 1\\).\nThe diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification\n\\[\nJ_{ij} =\n\\begin{cases}\nx_i (1 - x_i), & \\text{if } i = j, \\\\\n-x_i x_j, & \\text{if } i \\neq j.\n\\end{cases}\n\\]\nIn matrix form this can be expressed as\n\\[\nJ = \\text{diag}(x) - x x^\\top\n\\]\nThe determinant of this matrix can be found using the Matrix Determinant Lemma:\n\\[\n\\det\\bigl(A + u v^{\\top}\\bigr)\n=\n\\det(A)\\,\\bigl(1 + v^{\\top}A^{-1}u\\bigr).\n\\]\nHere,\n\\[\nA \\;=\\; \\operatorname{diag}(x_{1},\\ldots, x_{K-1}),\n\\quad\nu \\;=\\; -\\bigl(x_1,\\ldots, x_{K-1}\\bigr)^{\\!\\top},\n\\quad\nv \\;=\\; \\bigl(x_{1}, \\ldots, x_{K-1}\\bigr)^{\\!\\top}.\n\\] Therefore,\n\\[\n\\begin{aligned}\n\\det(J)\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_i \\bigg)\n\\bigg(1 + (x_{1},\\ldots, x_{K-1})\\,\\mathrm{diag}\\bigl(x_{1}^{-1},\\ldots,x_{K-1}^{-1}\\bigr)\\,\n\\big(-x_{1},\\ldots,-x_{K-1}\\big)^{\\top}\n\\bigg) \\\\\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg)\n\\bigg(1 - \\sum_{i=1}^{K-1} x_{i}\\bigg)\n=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg) x_{K} \\\\\n&=\n\\prod_{i=1}^{K} x_{i}.\n\\end{aligned}\n\\]\n\n\n\n\nThe transform \\(Y = f(X)\\) can be derived by reversing the stages of the inverse transform,\n\\[\ny_k\n= H^\\top \\bigg(\\log(x_k)\n- \\frac{1}{K}\\sum_{i=1}^K\\log(x_i) \\bigg)\n.\n\\]\nThe matrix \\(H\\) is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.\n\n\n\n\nThe column_stochastic_matrix[N, M] and row_stochastic_matrix[M, N] type in Stan represents an \\(N \\times M\\) matrix where each column (row) is a unit simplex of dimension \\(N\\). In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.\n\n\nA column stochastic matrix \\(X \\in \\mathbb{R}^{N \\times M}\\) is defined such that each column is a simplex. For column \\(m\\) (where \\(1 \\leq m \\leq M\\)):\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{n=1}^N X_{n, m} = 1.\n\\]\nA row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e. the rows of the matrix are simplexes)\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{m=1}^N X_{n, m} = 1.\n\\]\nThis definition ensures that each column (row) of the matrix \\(X\\) lies on the \\(N-1\\) dimensional unit simplex, similar to the simplex[N] type, but extended across multiple columns(rows).\n\n\n\nFor the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).\n\n\n\nThe Jacobian determinant of the inverse transform for each column \\(m\\) in the matrix is given by the product of the diagonal entries \\(J_{n, m}\\) of the lower-triangular Jacobian matrix. This determinant is calculated as:\n\\[\n\\left| \\det J_m \\right| = \\prod_{n=1}^{N-1} \\left( z_{n, m} (1 - z_{n, m}) \\left( 1 - \\sum_{n'=1}^{n-1} X_{n', m} \\right) \\right).\n\\]\nThus, the overall Jacobian determinant for the entire column_stochastic_matrix and row_stochastic_matrix is the product of the determinants for each column (row):\n\\[\n\\left| \\det J \\right| = \\prod_{m=1}^{M} \\left| \\det J_m \\right|.\n\\]\n\n\n\nFor the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).\n\n\n\n\nAn \\(n\\)-dimensional vector \\(x \\in \\mathbb{R}^n\\) is said to be a unit vector if it has unit Euclidean length, so that\n\\[\n\\Vert x \\Vert\n\\ = \\ \\sqrt{x^{\\top}\\,x}\n\\ = \\ \\sqrt{x_1^2 + x_2^2 + \\cdots + x_n^2}\n\\ = \\ 1\\ .\n\\]\n\n\nStan divides an unconstrained vector \\(y \\in \\mathbb{R}^{n}\\) by its norm, \\(\\Vert y \\Vert = \\sqrt{y^\\top y}\\), to obtain a unit vector \\(x\\),\n\\[\nx = \\frac{y}{\\Vert y \\Vert}.\n\\]\nTo generate a unit vector, Stan generates points at random in \\(\\mathbb{R}^n\\) with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. Muller (1959) showed this generates points uniformly at random on \\(S^{n-1}\\). That is, if we draw \\(y_n \\sim \\mathsf{Normal}(0, 1)\\) for \\(n \\in 1{:}n\\), then \\(x = \\frac{y}{\\Vert y \\Vert}\\) has a uniform distribution over \\(S^{n-1}\\). This allows us to use an \\(n\\)-dimensional basis for \\(S^{n-1}\\) that preserves local neighborhoods in that points that are close to each other in \\(\\mathbb{R}^n\\) map to points near each other in \\(S^{n-1}\\). The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in \\(\\mathbb{R}^n\\) that map to identical points in \\(S^{n-1}\\).\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nThe Jacobian matrix relating the input vector \\(y\\) to the output vector \\(x\\) is singular because \\(x^\\top x = 1\\) for any non-zero input vector \\(y\\). Thus, there technically is no unique transformation from \\(x\\) to \\(y\\). To circumvent this issue, let \\(r = \\sqrt{y^\\top y}\\) so that \\(y = r\nx\\). The transformation from \\(\\left(r, x_{-n}\\right)\\) to \\(y\\) is well-defined but \\(r\\) is arbitrary, so we set \\(r = 1\\). In this case, the determinant of the Jacobian is proportional to \\(e^{-\\frac{1}{2} y^\\top y}\\), which is the kernel of a standard multivariate normal distribution with \\(n\\) independent dimensions.\n\n\n\n\nA \\(K \\times K\\) correlation matrix \\(x\\) must be symmetric, so that\n\\[\nx_{k,k'} = x_{k',k}\n\\]\nfor all \\(k,k' \\in \\{ 1, \\ldots, K \\}\\), it must have a unit diagonal, so that\n\\[\nx_{k,k} = 1\n\\]\nfor all \\(k \\in \\{ 1, \\ldots, K \\}\\), and it must be positive definite, so that for every non-zero \\(K\\)-vector \\(a\\),\n\\[\na^{\\top} x a &gt; 0.\n\\]\nThe number of free parameters required to specify a \\(K \\times K\\) correlation matrix is \\(\\binom{K}{2}\\).\nThere is more than one way to map from \\(\\binom{K}{2}\\) unconstrained parameters to a \\(K \\times K\\) correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform Lewandowski, Kurowicka, and Joe (2009).\n\n\nIt is easiest to specify the inverse, going from its \\(\\binom{K}{2}\\) parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose \\(y\\) is a vector containing \\(\\binom{K}{2}\\) unconstrained values. These are first transformed via the bijective function \\(\\tanh : \\mathbb{R} \\rightarrow\n(-1, 1)\\)\n\\[\n\\tanh y = \\frac{\\exp(2y) - 1}{\\exp(2y) + 1}.\n\\]\nThen, define a \\(K \\times K\\) matrix \\(z\\), the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the \\(4 \\times 4\\) case, there are \\(\\binom{4}{2}\\) values arranged as\n\\[\nz\n=\n\\left[\n\\begin{array}{cccc}\n1 & \\tanh y_1 & \\tanh y_2 & \\tanh y_4\n\\\\\n0 & 1 & \\tanh y_3 & \\tanh y_5\n\\\\\n0 & 0 & 1 & \\tanh y_6\n\\\\\n0 & 0 & 0 & 1\n\\end{array}\n\\right]\n.\n\\]\nLewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array \\(z\\) to a correlation matrix \\(x\\). The entry \\(z_{i,j}\\) for \\(i &lt;\nj\\) is interpreted as the canonical partial correlation (CPC) between \\(i\\) and \\(j\\), which is the correlation between \\(i\\)’s residuals and \\(j\\)’s residuals when both \\(i\\) and \\(j\\) are regressed on all variables \\(i'\\) such that \\(i'&lt; i\\). In the case of \\(i=1\\), there are no earlier variables, so \\(z_{1,j}\\) is just the Pearson correlation between \\(i\\) and \\(j\\).\nIn Stan, the LKJ transform is reformulated in terms of a Cholesky factor \\(w\\) of the final correlation matrix, defined for \\(1 \\leq i,j \\leq K\\) by\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\n\\prod_{i'=1}^{i - 1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{if } 1 &lt; i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\\\nz_{i,j} \\, \\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThis does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n\\frac{z_{i,j}}{z_{i-1,j}} \\ w_{i-1,j} \\left( 1 - z_{i-1,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i \\leq j.\n\\end{array}\n\\right.\n\\]\nGiven the upper-triangular Cholesky factor \\(w\\), the final correlation matrix is\n\\[\nx = w^{\\top} w.\n\\]\nLewandowski, Kurowicka, and Joe (2009) show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as\n\\[\n\\mbox{det} \\, x = \\prod_{i=1}^{K-1} \\ \\prod_{j=i+1}^K \\ (1 - z_{i,j}^2)\n= \\prod_{1 \\leq i &lt; j \\leq K} (1 - z_{i,j}^2),\n\\]\n\n\n\nFrom the inverse of equation 11 in (Lewandowski, Kurowicka, and Joe 2009), the absolute Jacobian determinant is\n\\[\n\\sqrt{\\prod_{i=1}^{K-1}\\prod_{j=i+1}^K \\left(1-z_{i,j}^2\\right)^{K-i-1}} \\\n\\times \\prod_{i=1}^{K-1}\\prod_{j=i+1}^K\n\\frac{\\partial z_{i,j}}{\\partial y_{i,j}}\n\\]\n\n\n\nThe correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.\nStarting with a correlation matrix \\(x\\), the first step is to find the unique upper triangular \\(w\\) such that \\(x = w w^{\\top}\\). Because \\(x\\) is positive definite, this can be done by applying the Cholesky decomposition,\n\\[\nw = \\mbox{chol}(x).\n\\]\nThe next step from the Cholesky factor \\(w\\) back to the array \\(z\\) of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of \\(w\\), which when inverted yields\n\\[\nz_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i \\leq j,\n\\\\\nw_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n{w_{i,j}}\n\\\n\\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,j}^2 \\right)^{-1/2}\n& \\mbox{if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThe final stage of the transform reverses the hyperbolic tangent transform, which is defined by\n\\[\ny = \\tanh^{-1} z = \\frac{1}{2} \\log \\left( \\frac{1 + z}{1 - z} \\right).\n\\]\nThe inverse hyperbolic tangent function, \\(\\tanh^{-1}\\), is also called the Fisher transformation.\n\n\n\n\nA \\(K \\times K\\) matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires \\(K + \\binom{K}{2}\\) free parameters to specify a \\(K \\times K\\) covariance matrix.\n\n\nStan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.1\nIf \\(x\\) is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix \\(z =\n\\mathrm{chol}(x)\\) with positive diagonal entries, called a Cholesky factor, such that\n\\[\nx = z \\, z^{\\top}.\n\\]\nThe off-diagonal entries of the Cholesky factor \\(z\\) are unconstrained, but the diagonal entries \\(z_{k,k}\\) must be positive for \\(1 \\leq k\n\\leq K\\).\nTo complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix \\(y\\) defined by\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log z_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nz_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular \\(K \\times K\\) matrix \\(y\\), the first step is to recover the intermediate matrix \\(z\\) by reversing the log transform,\n\\[\nz_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe covariance matrix \\(x\\) is recovered from its Cholesky factor \\(z\\) by taking\n\\[\nx = z \\, z^{\\top}.\n\\]\n\n\n\nThe Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix \\(y\\) to matrix \\(z\\) with positive diagonals and the product transform from the Cholesky factor \\(z\\) to \\(x\\).\nThe transform from unconstrained \\(y\\) to Cholesky factor \\(z\\) has a diagonal Jacobian matrix, the absolute determinant of which is thus\n\\[\n\\prod_{k=1}^K  \\frac{\\partial}{\\partial_{y_{k,k}}} \\, \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K z_{k,k}.\n\\]\nThe Jacobian matrix of the second transform from the Cholesky factor \\(z\\) to the covariance matrix \\(x\\) is also triangular, with diagonal entries corresponding to pairs \\((m,n)\\) with \\(m \\geq n\\), defined by\n\\[\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( z \\, z^{\\top} \\right)_{m,n}\n\\ = \\\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( \\sum_{k=1}^K z_{m,k} \\, z_{n,k} \\right)\n\\ = \\\n\\left\\{\n\\begin{array}{cl}\n2 \\, z_{n,n} & \\mbox{if } m = n \\mbox{ and }\n\\\\\nz_{n,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe absolute Jacobian determinant of the second transform is thus\n\\[\n2^{K} \\ \\prod_{m = 1}^{K} \\ \\prod_{n=1}^{m} z_{n,n}\n\\ = \\\n\\prod_{n=1}^K \\ \\prod_{m=n}^K z_{n,n}\n\\ = \\\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}.\n\\]\nFinally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular \\(y\\) to a symmetric, positive definite matrix \\(x\\) is the product of the Jacobian determinants of the exponentiation and product transforms,\n\\[\n\\left( \\prod_{k=1}^K z_{k,k} \\right)\n\\left(\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}\n\\right)\n\\ = \\\n2^K\n\\, \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\nLet \\(f^{-1}\\) be the inverse transform from a \\(K + \\binom{K}{2}\\)-vector \\(y\\) to the \\(K \\times K\\) covariance matrix \\(x\\). A density function \\(p_X(x)\\) defined on \\(K \\times K\\) covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(K + \\binom{K}{2}\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\ 2^K \\ \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\n\n\n\n\nAn \\(M \\times M\\) covariance matrix \\(\\Sigma\\) can be Cholesky factored to a lower triangular matrix \\(L\\) such that \\(L\\,L^{\\top} = \\Sigma\\). If \\(\\Sigma\\) is positive definite, then \\(L\\) will be \\(M \\times M\\). If \\(\\Sigma\\) is only positive semi-definite, then \\(L\\) will be \\(M \\times N\\), with \\(N &lt; M\\).\nA matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and \\(M \\geq\nN\\). A matrix satisfying these conditions ensures that \\(L \\,\nL^{\\top}\\) is positive semi-definite if \\(M &gt; N\\) and positive definite if \\(M = N\\).\nA Cholesky factor of a covariance matrix requires \\(N + \\binom{N}{2} +\n(M - N)N\\) unconstrained parameters.\n\n\nStan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose \\(x\\) is an \\(M \\times N\\) Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log x_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nx_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform need only invert the logarithm with an exponentiation. If \\(y\\) is the unconstrained matrix representation, then the elements of the constrained matrix \\(x\\) is defined by\n\\[\nx_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe transform has a diagonal Jacobian matrix, the absolute determinant of which is\n\\[\n\\prod_{n=1}^N  \\frac{\\partial}{\\partial_{y_{n,n}}} \\, \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N x_{n,n}.\n\\]\nLet \\(x = f^{-1}(y)\\) be the inverse transform from a \\(N + \\binom{N}{2}\n+ (M - N)N\\) vector to an \\(M \\times N\\) Cholesky factor for a covariance matrix \\(x\\) defined in the previous section. A density function \\(p_X(x)\\) defined on \\(M \\times N\\) Cholesky factors of covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(N + \\binom{N}{2}\n+ (M - N)N\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\prod_{N=1}^N x_{n,n}.\n\\]\n\n\n\n\nA \\(K \\times K\\) correlation matrix \\(\\Omega\\) is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a \\(K \\times K\\) lower-triangular matrix \\(L\\) with positive diagonal elements such that \\(\\Omega = L\\,L^{\\top}\\). Because the correlation matrix has a unit diagonal,\n\\[\n\\Omega_{k,k} = L_k\\,L_k^{\\top} = 1,\n\\]\neach row vector \\(L_k\\) of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of \\(L\\) to be calculated from the off-diagonal elements, so that a Cholesky factor for a \\(K \\times K\\) correlation matrix requires only \\(\\binom{K}{2}\\) unconstrained parameters.\n\n\nIt is easiest to start with the inverse transform from the \\(\\binom{K}{2}\\) unconstrained parameters \\(y\\) to the \\(K \\times K\\) lower-triangular Cholesky factor \\(x\\). The inverse transform is based on the hyperbolic tangent function, \\(\\tanh\\), which satisfies \\(\\tanh(x) \\in (-1,1)\\). Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the section on correlation matrix transforms for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.\nSuppose \\(y\\) is a vector of \\(\\binom{K}{2}\\) unconstrained values. Let \\(z\\) be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the \\(3 \\times 3\\) case,\n\\[\nz =\n\\left[\n\\begin{array}{ccc}\n0 & 0 & 0\n\\\\\n\\tanh y_1 & 0 & 0\n\\\\\n\\tanh y_2 & \\tanh y_3 & 0\n\\end{array}\n\\right]\n\\]\nThe matrix \\(z\\), with entries in the range \\((-1, 1)\\), is then transformed to the Cholesky factor \\(x\\), by taking2\n\\[\nx_{i,j}\n=\n\\left\\{\n\\begin{array}{lll}\n0 & \\mbox{ if } i &lt; j & \\mbox{ [above diagonal]}\n\\\\\n\\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i = j & \\mbox{ [on diagonal]}\n\\\\\nz_{i,j} \\ \\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i &gt; j & \\mbox{ [below diagonal]}\n\\end{array}\n\\right.\n\\]\nIn the \\(3 \\times 3\\) case, this yields\n\\[\nx =\n\\left[\n\\begin{array}{ccc}\n1 & 0 & 0\n\\\\\nz_{2,1} & \\sqrt{1 - x_{2,1}^2} & 0\n\\\\\nz_{3,1} & z_{3,2} \\sqrt{1 - x_{3,1}^2}\n        & \\sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}\n\\end{array}\n\\right],\n\\]\nwhere the \\(z_{i,j} \\in (-1,1)\\) are the \\(\\tanh\\)-transformed \\(y\\).\nThe approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at \\(j=1\\), each below-diagonal entry \\(x_{i,j}\\) is determined by the (signed) fraction \\(z_{i,j}\\) of the remaining length for the row that it consumes. The diagonal entries \\(x_{i,i}\\) get any leftover length from earlier entries in their row. The above-diagonal entries are zero.\n\n\n\nSuppose \\(x\\) is a \\(K \\times K\\) Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values \\(z\\) from \\(x\\),\n\\[\nz_{i,j} = \\frac{x_{i,j}}{\\sqrt{1 - \\sum_{j' &lt; j}x_{i,j'}^2}}.\n\\]\nThe mapping from the resulting \\(z\\) to \\(y\\) inverts \\(\\tanh\\),\n\\[\ny\n\\ = \\\n\\tanh^{-1} z\n\\ = \\\n\\frac{1}{2} \\left( \\log (1 + z) - \\log (1 - z) \\right).\n\\]\n\n\n\nThe Jacobian of the full transform is the product of the Jacobians of its component transforms.\nFirst, for the inverse transform \\(z = \\tanh y\\), the derivative is\n\\[\n\\frac{d}{dy} \\tanh y = \\frac{1}{(\\cosh y)^2}.\n\\]\nSecond, for the inverse transform of \\(z\\) to \\(x\\), the resulting Jacobian matrix \\(J\\) is of dimension \\(\\binom{K}{2} \\times\n\\binom{K}{2}\\), with indexes \\((i,j)\\) for \\((i &gt; j)\\). The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each \\((i,j)\\) pair,\n\\[\n\\left| \\, \\mbox{det} \\, J \\, \\right|\n  \\ = \\ \\prod_{i &gt; j} \\left| \\frac{d}{dz_{i,j}} x_{i,j} \\right|,\n\\]\nwhere\n\\[\n\\frac{d}{dz_{i,j}} x_{i,j}\n= \\sqrt{1 - \\sum_{j' &lt; j} x^2_{i,j'}}.\n\\]\nSo the combined density for unconstrained \\(y\\) is\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n  \\ \\\n  \\prod_{n &lt; \\binom{K}{2}} \\frac{1}{(\\cosh y)^2}\n  \\ \\\n  \\prod_{i &gt; j} \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2\n  \\right)^{1/2},\n\\]\nwhere \\(x = f^{-1}(y)\\) is used for notational convenience. The log Jacobian determinant of the complete inverse transform \\(x = f^{-1}(y)\\) is given by\n\\[\n\\log \\left| \\, \\det J \\, \\right|\n=\n-2 \\sum_{n \\leq \\binom{K}{2}}\n\\log \\cosh y\n\\\n+\n\\\n\\frac{1}{2} \\\n\\sum_{i &gt; j}\n\\log \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2 \\right)\n.\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#reject-statements.section",
-    "href": "reference-manual/statements.html#reject-statements.section",
-    "title": "Statements",
+    "objectID": "reference-manual/transforms.html#limitations-due-to-finite-accuracy-presentation",
+    "href": "reference-manual/transforms.html#limitations-due-to-finite-accuracy-presentation",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "The Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.\nLike the print statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.\nReject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable x’s value is negative.\nif (x &lt; 0) {\n  reject(\"x must not be negative; found x=\", x);\n}\n\n\nReject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the normal_lpdf function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.\nIn all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.\n\n\nRejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.\n\n\n\nRejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.\nReject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.\nFatal errors in other blocks may also be signaled by use of the fatal_error statement.\n\n\n\nRejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.\nIf the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.\n\n\n\n\nRejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;\n  real&lt;lower=a, upper=b&gt; theta;\n  // ...\n}\nmodel {\n  // **wrong** needs explicit truncation\n  theta ~ normal(0, 1);\n  // ...\n}\nThis program is wrong because its truncation bounds on theta depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.\n  theta ~ normal(0, 1) T[a, b];\nThe conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over \\([a ,b]\\); in these simple univariate cases, we can overcome that with the T[ , ] notation, which essentially divides by whatever the prior integrates to over \\([a, b]\\).\nThis problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.\n  if (theta &lt; a || theta &gt; b) {\n    reject(\"theta not in (a, b)\");\n  }\n  // still **wrong**, needs T[a,b]\n  theta ~ normal(0, 1);\nIn this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.\nEven if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.",
+    "text": "In this section the transformations are described mathematically. However, observed behavior can be different from the exact arithmetic.\nStan’s arithmetic is implemented using double-precision floating-point numbers, which may cause computation to behave differently than mathematics. For example, the lower bound constraint is defined above by an exponential inverse transform which mathematically excludes the lower bound, but if the closest floating-point number for the inverse transformed value is the boundary, then the value is rounded to the boundary. This may cause unexpected warnings or errors, if in other parts of the code the boundary value is invalid. For example, we may observe floating-point value 0 for a variance parameter that has been declared with lower=0. In general, double-precision floating-point numbers cannot reliably store more than 16 digits of a number in decimal. See more about floating point arithmetic in the Stan User’s Guide.\nThese issues are exacerbated by the fact that CmdStan stores the output to CSV files with 8 digits precision by default. More digits can be requested by the user at the cost of additional disk usage, as discussed in the CmdStan Command-Line Interface Overview.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#fatal-error-statements",
-    "href": "reference-manual/statements.html#fatal-error-statements",
-    "title": "Statements",
+    "objectID": "reference-manual/transforms.html#change-of-variables.section",
+    "href": "reference-manual/transforms.html#change-of-variables.section",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "The Stan fatal_error statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.\nLike the print or reject statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.\nThe fatal error may be used to signal an unrecoverable error in blocks where reject leads to the algorithm attempting to try again, such as the model block.",
+    "text": "The support of a random variable \\(X\\) with density \\(p_X(x)\\) is that subset of values for which it has non-zero density,\n\\[\n\\mathrm{supp}(X) = \\{ x | p_X(x) &gt; 0 \\}.\n\\]\nIf \\(f\\) is a total function defined on the support of \\(X\\), then \\(Y =\nf(X)\\) is a new random variable. This section shows how to compute the probability density function of \\(Y\\) for well-behaved transforms \\(f\\). The rest of the chapter details the transforms used by Stan.\n\n\nSuppose \\(X\\) is one dimensional and \\(f: \\mathrm{supp}(X) \\rightarrow\n\\mathbb{R}\\) is a one-to-one, monotonic function with a differentiable inverse \\(f^{-1}\\). Then the density of \\(Y\\) is given by\n\\[\np_Y(y) = p_X(f^{-1}(y))\n         \\,\n         \\left| \\, \\frac{d}{dy} f^{-1}(y)\\, \\right|.\n\\]\nThe absolute derivative of the inverse transform measures how the scale of the transformed variable changes with respect to the underlying variable.\n\n\n\nThe multivariate generalization of an absolute derivative is a Jacobian, or more fully the absolute value of the determinant of the Jacobian matrix of the transform. The Jacobian matrix measures the change of each output variable relative to every input variable and the absolute determinant uses that to determine the differential change in volume at a given point in the parameter space.\nSuppose \\(X\\) is a \\(K\\)-dimensional random variable with probability density function \\(p_X(x)\\). A new random variable \\(Y = f(X)\\) may be defined by transforming \\(X\\) with a suitably well-behaved function \\(f\\). It suffices for what follows to note that if \\(f\\) is one-to-one and its inverse \\(f^{-1}\\) has a well-defined Jacobian, then the density of \\(Y\\) is\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\, \\left| \\, \\det \\, J_{f^{-1}}(y) \\, \\right|,\n\\]\nwhere \\(\\det{}\\) is the matrix determinant operation and \\(J_{f^{-1}}(y)\\) is the Jacobian matrix of \\(f^{-1}\\) evaluated at \\(y\\). Taking \\(x =\nf^{-1}(y)\\), the Jacobian matrix is defined by\n\\[\nJ_{f^{-1}}(y) =\n\\left[\n\\begin{array}{ccc}\\displaystyle\n\\frac{\\partial x_1}{\\partial y_1}\n& \\cdots\n& \\displaystyle \\frac{\\partial x_1}{\\partial y_{K}}\n\\\\\n\\vdots & \\vdots & \\vdots\n\\\\\n\\displaystyle\\frac{\\partial x_{K}}{\\partial y_1}\n& \\cdots\n& \\displaystyle\\frac{\\partial x_{K}}{\\partial y_{K}}\n\\end{array}\n\\right].\n\\]\nIf the Jacobian matrix is triangular, the determinant reduces to the product of the diagonal entries,\n\\[\n\\det \\, J_{f^{-1}}(y)\n= \\prod_{k=1}^K \\frac{\\partial x_k}{\\partial y_k}.\n\\]\nTriangular matrices naturally arise in situations where the variables are ordered, for instance by dimension, and each variable’s transformed value depends on the previous variable’s transformed values. Diagonal matrices, a simple form of triangular matrix, arise if each transformed variable only depends on a single untransformed variable.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/statements.html#footnotes",
-    "href": "reference-manual/statements.html#footnotes",
-    "title": "Statements",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe current notation replaces two previous versions. Originally, a variable lp__ was directly exposed and manipulated; this is no longer allowed. The original statement syntax for target += u was increment_log_prob(u), but this form was removed in Stan 2.33↩︎\nWriting this model with the expression -0.5 * y * y is more efficient than with the equivalent expression y * y / -2 because multiplication is more efficient than division; in both cases, the negation is rolled into the numeric literal (-0.5 and -2). Writing square(y) instead of y * y would be even more efficient because the derivatives can be precomputed, reducing the memory and number of operations required for automatic differentiation.↩︎\nBecause \\(\\log | \\frac{d}{dy} \\log y | = \\log | 1/y | = - \\log |y|\\).↩︎\nA programming idiom in BUGS code simulates a local variable by replacing theta in the above example with theta[n], effectively creating N different variables, theta[1], …, theta[N]. Of course, this is not a hack if the value of theta[n] is required for all n.↩︎\nThe adjoint component is always zero during execution for the algorithmic differentiation variables used to implement parameters, transformed parameters, and local variables in the model.↩︎",
+    "objectID": "reference-manual/transforms.html#lower-bound-transform.section",
+    "href": "reference-manual/transforms.html#lower-bound-transform.section",
+    "title": "Constraint Transforms",
+    "section": "",
+    "text": "Stan uses a logarithmic transform for lower and upper bounds.\n\n\nIf a variable \\(X\\) is declared to have lower bound \\(a\\), it is transformed to an unbounded variable \\(Y\\), where\n\\[\nY = \\log(X - a).\n\\]\n\n\n\nThe inverse of the lower-bound transform maps an unbounded variable \\(Y\\) to a variable \\(X\\) that is bounded below by \\(a\\) by\n\\[\nX = \\exp(Y) + a.\n\\]\n\n\n\nThe absolute derivative of the inverse transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( \\exp(y) + a \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, given the density \\(p_X\\) of \\(X\\), the density of \\(Y\\) is\n\\[\np_Y(y)\n= p_X\\!\\left( \\exp(y) + a \\right) \\cdot \\exp(y).\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Statements"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html",
-    "href": "reference-manual/removals.html",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#upper-bounded-scalar",
+    "href": "reference-manual/transforms.html#upper-bounded-scalar",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "This chapter lists functionalities that were once present in the language but have since been removed, along with how to replace them.\n\n\nRemoved: The variable lp__ is no longer available for direct access or manipulation.\nReplacement: General manipulation of the value of the lp__ variable is not allowed, but\nlp__ &lt;- lp__ + e;\ncan be replaced with\ntarget += e;\nThe value of lp__ is available through the no-argument function target().\n\n\n\nRemoved: The operator &lt;- for assignment, e.g.,\na &lt;- b;\nis no longer available.\nReplacement: The new syntax uses the operator = for assignment, e.g.,\na = b;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The increment_log_prob(u) statement for incrementing the log density accumulator by u is no longer available.\nReplacement: Replace the above statement with\ntarget += u;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The built-in no-argument function get_lp() is no longer available.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33\n\n\n\nRemoved: Formerly, the probability function for the distribution foo would be applied to an outcome variable y and sequence of zero or more parameters ... to produce the expression foo_log(y, ...). This suffix is no longer a special value.\nReplacement: If y can be a real value (including vectors or matrices), replace\nfoo_log(y, ...)\nwith the log probability density function notation\nfoo_lpdf(y | ...).\nIf y must be an integer (including arrays), instead replace\nfoo_log(y, ...\nwith the log probability mass function\nfoo_lpmf(y | ...).\nRemoved In: Stan 2.33\n\n\n\nRemoved: The log cumulative distribution and complementary cumulative distribution functions for a distribution foo were formerly written as foo_cdf_log and foo_ccdf_log.\nReplacement:\nReplace foo_cdf_log(y, ...) with foo_lcdf(y | ...).\nReplace foo_ccdf_log(y, ...) with foo_lccdf(y | ...).\n\n\n\nRemoved: A user-defined function ending in _log can be no longer be used in statements.qmd#distribution-statements.section.\nReplacement: Replace the _log suffix with _lpdf for density functions or _lpmf for mass functions in the user-defined function.\nRemoved In: Stan 2.33\nNote: Following Stan 2.33, users can stil define a function ending in _log, it simply no longer has a special meaning or is supported in the ~ syntax.\n\n\n\nRemoved: The function if_else is no longer available.\nReplacement: Use the conditional operator which allows more flexibility in the types of b and c and is much more efficient in that it only evaluates whichever of b or c is returned.\nx = if_else(a, b, c);\nwith\nx = a ? b : c;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The use of # for line-based comments is no longer permitted. # may only be used for #include statements.\nReplacement: Use a pair of forward slashes, //, for line comments.\nRemoved In: Stan 2.33\n\n\n\nBefore Stan 2.26, arrays were declared by writing syntax after the variable.\nRemoved: The use of array declarations like\nint n[5];\nreal a[3, 4];\nreal&lt;lower=0&gt; z[5, 4, 2];\nvector[7] mu[3];\nmatrix[7, 2] mu[15, 12];\ncholesky_factor_cov[5, 6] mu[2, 3, 4];\nReplacement: The use of the array keyword, which replaces the above examples with\narray[5] int n;\narray[3, 4] real a;\narray[5, 4, 2] real&lt;lower=0&gt; z;\narray[3] vector[7] mu;\narray[15, 12] matrix[7, 2] mu;\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nRemoved In: Stan 2.33\n\n\n\nStan interprets nested indexing in assingments as flat indexing so that a statement like\na[:][1] = b;\nis the same as\na[:,1] = b;\nHowever, this is inconsistent with multiple indexing rules.\nTo avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.\nRemoved In: Stan 2.33\n\n\n\nRemoved: Using a real value in a conditional is no longer permitted.\nreal x = 1.0;\nif (x) {\nThe value was interpreted as true if it is nonzero.\nReplacement: For the exact equivalent, use a comparison operator to make the intent clear.\nreal x = 1.0;\nif (x != 0) {\nHowever, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative abs(x) &lt; machine_precision() is appropriate for the use case.\nRemoved In: Stan 2.34",
+    "text": "Stan uses a negated logarithmic transform for upper bounds.\n\n\nIf a variable \\(X\\) is declared to have an upper bound \\(b\\), it is transformed to the unbounded variable \\(Y\\) by\n\\[\nY = \\log(b - X).\n\\]\n\n\n\nThe inverse of the upper bound transform converts the unbounded variable \\(Y\\) to the variable \\(X\\) bounded above by \\(b\\) through\n\\[\nX = b - \\exp(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse of the upper bound transform is\n\\[\n\\left| \\,\n\\frac{d}{dy} \\left( b - \\exp(y) \\right)\n\\, \\right|\n= \\exp(y).\n\\]\nTherefore, the density of the unconstrained variable \\(Y\\) is defined in terms of the density of the variable \\(X\\) with an upper bound of \\(b\\) by\n\\[\np_Y(y)\n=   p_X \\!\\left( b - \\exp(y) \\right) \\cdot \\exp(y).\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#lp__-variable",
-    "href": "reference-manual/removals.html#lp__-variable",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#logit-transform-jacobian.section",
+    "href": "reference-manual/transforms.html#logit-transform-jacobian.section",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The variable lp__ is no longer available for direct access or manipulation.\nReplacement: General manipulation of the value of the lp__ variable is not allowed, but\nlp__ &lt;- lp__ + e;\ncan be replaced with\ntarget += e;\nThe value of lp__ is available through the no-argument function target().",
+    "text": "For lower and upper-bounded variables, Stan uses a scaled and translated log-odds transform.\n\n\nThe log-odds function is defined for \\(u \\in (0,1)\\) by\n\\[\n\\mathrm{logit}(u) = \\log \\frac{u}{1 - u}.\n\\]\nThe inverse of the log odds function is the logistic sigmoid, defined for \\(v \\in (-\\infty,\\infty)\\) by\n\\[\n\\mathrm{logit}^{-1}(v) = \\frac{1}{1 + \\exp(-v)}.\n\\]\nThe derivative of the logistic sigmoid is\n\\[\n\\frac{d}{dy} \\mathrm{logit}^{-1}(y)\n= \\mathrm{logit}^{-1}(y) \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\n\n\n\nFor variables constrained to be in the open interval \\((a, b)\\), Stan uses a scaled and translated log-odds transform. If variable \\(X\\) is declared to have lower bound \\(a\\) and upper bound \\(b\\), then it is transformed to a new variable \\(Y\\), where\n\\[\nY = \\mathrm{logit} \\left( \\frac{X - a}{b - a} \\right).\n\\]\n\n\n\nThe inverse of this transform is\n\\[\nX = a + (b - a) \\cdot \\mathrm{logit}^{-1}(Y).\n\\]\n\n\n\nThe absolute derivative of the inverse transform is given by\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      a + (b - a) \\cdot \\mathrm{logit}^{-1}(y)\n    \\right)\n  \\right|\n= (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( a + (b - a) \\cdot \\mathrm{logit}^{-1}(y) \\right)\n    \\cdot (b - a)\n    \\cdot \\mathrm{logit}^{-1}(y)\n    \\cdot \\left( 1 - \\mathrm{logit}^{-1}(y) \\right).\n\\]\nDespite the apparent complexity of this expression, most of the terms are repeated and thus only need to be evaluated once. Most importantly, \\(\\mathrm{logit}^{-1}(y)\\) only needs to be evaluated once, so there is only one call to \\(\\exp(-y)\\).",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#assignment-with--",
-    "href": "reference-manual/removals.html#assignment-with--",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#affinely-transformed-scalar",
+    "href": "reference-manual/transforms.html#affinely-transformed-scalar",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The operator &lt;- for assignment, e.g.,\na &lt;- b;\nis no longer available.\nReplacement: The new syntax uses the operator = for assignment, e.g.,\na = b;\nRemoved In: Stan 2.33",
+    "text": "Stan uses an affine transform to be able to specify parameters with a given offset and multiplier.\n\n\nFor variables with expected offset \\(\\mu\\) and/or (positive) multiplier \\(\\sigma\\), Stan uses an affine transform. Such a variable \\(X\\) is transformed to a new variable \\(Y\\), where\n\\[\nY = \\frac{X - \\mu}{\\sigma}.\n\\]\nThe default value for the offset \\(\\mu\\) is \\(0\\) and for the multiplier \\(\\sigma\\) is \\(1\\) in case not both are specified.\n\n\n\nThe inverse of this transform is\n\\[\nX =  \\mu + \\sigma \\cdot Y.\n\\]\n\n\n\nThe absolute derivative of the affine inverse transform is\n\\[\n\\left|\n  \\frac{d}{dy}\n    \\left(\n      \\mu + \\sigma \\cdot y\n    \\right)\n  \\right|\n= \\sigma.\n\\]\nTherefore, the density of the transformed variable \\(Y\\) is\n\\[\np_Y(y)\n=\np_X \\! \\left( \\mu + \\sigma \\cdot y \\right)\n    \\cdot \\sigma.\n\\]\nFor an example of how to code this in Stan, see section Affinely Transformed Real.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#increment_log_prob-statement",
-    "href": "reference-manual/removals.html#increment_log_prob-statement",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#ordered-vector",
+    "href": "reference-manual/transforms.html#ordered-vector",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The increment_log_prob(u) statement for incrementing the log density accumulator by u is no longer available.\nReplacement: Replace the above statement with\ntarget += u;\nRemoved In: Stan 2.33",
+    "text": "For some modeling tasks, a vector-valued random variable \\(X\\) is required with support on ordered sequences. One example is the set of cut points in ordered logistic regression.\nIn constraint terms, an ordered \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies\n\\[\nx_k &lt; x_{k+1}\n\\]\nfor \\(k \\in \\{ 1, \\ldots, K-1 \\}\\).\n\n\nStan’s transform follows the constraint directly. It maps an increasing vector \\(x \\in \\mathbb{R}^{K}\\) to an unconstrained vector \\(y \\in\n\\mathbb{R}^K\\) by setting\n\\[\ny_k\n=\n\\left\\{\n\\begin{array}{ll}\nx_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\log \\left( x_{k} - x_{k-1} \\right) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform for an unconstrained \\(y \\in \\mathbb{R}^K\\) to an ordered sequence \\(x \\in \\mathbb{R}^K\\) is defined by the recursion\n\\[\nx_k\n=\n\\left\\{\n\\begin{array}{ll}\ny_1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\nx_{k-1} + \\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\n\\(x_k\\) can also be expressed iteratively as\n\\[\nx_k = y_1 + \\sum_{k'=2}^k \\exp(y_{k'}).\n\\]\n\n\n\nThe Jacobian of the inverse transform \\(f^{-1}\\) is lower triangular, with diagonal elements for \\(1 \\leq k \\leq K\\) of\n\\[\nJ_{k,k} =\n\\left\\{\n\\begin{array}{ll}\n1 & \\mbox{if } k = 1, \\mbox{ and}\n\\\\\n\\exp(y_k) & \\mbox{if } 1 &lt; k \\leq K.\n\\end{array}\n\\right.\n\\]\nBecause \\(J\\) is triangular, the absolute Jacobian determinant is\n\\[\n\\left| \\, \\det \\, J \\, \\right|\n\\ = \\\n\\left| \\, \\prod_{k=1}^K J_{k,k} \\, \\right|\n\\ = \\\n\\prod_{k=2}^K \\exp(y_k).\n\\]\nPutting this all together, if \\(p_X\\) is the density of \\(X\\), then the transformed variable \\(Y\\) has density \\(p_Y\\) given by\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n\\\n\\prod_{k=2}^K \\exp(y_k).\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#get_lp-function",
-    "href": "reference-manual/removals.html#get_lp-function",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#positive-ordered-vector",
+    "href": "reference-manual/transforms.html#positive-ordered-vector",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The built-in no-argument function get_lp() is no longer available.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33",
+    "text": "Placeholder.\nThe positive ordered transformation is defined in the same style as the ordered transformation above, but with the first element being exponentiated to ensure positivity.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#log-density-and-mass-functions",
-    "href": "reference-manual/removals.html#log-density-and-mass-functions",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#sum-to-zero-transforms",
+    "href": "reference-manual/transforms.html#sum-to-zero-transforms",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: Formerly, the probability function for the distribution foo would be applied to an outcome variable y and sequence of zero or more parameters ... to produce the expression foo_log(y, ...). This suffix is no longer a special value.\nReplacement: If y can be a real value (including vectors or matrices), replace\nfoo_log(y, ...)\nwith the log probability density function notation\nfoo_lpdf(y | ...).\nIf y must be an integer (including arrays), instead replace\nfoo_log(y, ...\nwith the log probability mass function\nfoo_lpmf(y | ...).\nRemoved In: Stan 2.33",
+    "text": "Stan provides built-in constraint transforms for sum-to-zero vectors and sum-to-zero matrices. The sum-to-zero vector is a vector of length \\(N\\) with real values and the sum of the vector equals zero. The sum-to-zero matrix is an \\(N \\times M\\) matrix where both the rows and columns sum-to-zero.\nStan uses an orthogonal basis as the initial point of construction. The orthogonal basis balances the constraint across each unconstrained value. The basis is a matrix \\(H\\) such that \\(H\\in\\mathbb R^{N\\times (N-1)}\\) and \\(H^{\\mathsf T}H=I_{N-1},\\;H^{\\mathsf T}\\mathbf 1=0\\). The sum-to-zero vector lies in the subspace where the vector sums to zero. Although this seems tautological, the orthogonal basis construction allows all the marginal variances of the contrained vector to be the same, see, e.g.,(Seyboldt 2024). Simpler alternatives, such as setting the final element to the negative sum of the first elements, do not result in equal variances. It is worth noting that even with marginal variances being equal each value in the sum-to-zero constrained space is negatively correlated.\nIn many cases one wishes to model the sum-to-zero vector as normally distributed and the induced covariance matrix is fully known:\n\\[\n  \\sigma^2 \\begin{pmatrix}\n  1-\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots&-\\tfrac{1}{N} \\\\\n  -\\tfrac{1}{N} & 1-\\tfrac{1}{N} & \\cdots & -\\tfrac{1}{N} \\\\\n  \\vdots & \\vdots & \\ddots & \\vdots \\\\\n  -\\tfrac{1}{N} & -\\tfrac{1}{N} & \\cdots &1-\\tfrac{1}{N}\n  \\end{pmatrix}.\n\\]\nThe marginal standard deviation no longer corresponds to \\(\\sigma\\) but is \\(\\sigma \\sqrt{1-\\tfrac{1}{N}}\\). The properties of the normal distribution allow multiplying the sum-to-zero vector by the reciprocal of \\(\\sqrt{1-\\tfrac{1}{N}}\\) to adjust the variance to the intended \\(\\sigma\\):\nsum_to_zero_vector[N] x;\nx ~ normal(0, sqrt(N %/% (N - 1)));\nWhen \\(\\sigma\\) is a parameter there is an additional adjustment when using the centered version of the sum-to-zero constraint. If we let \\(y\\) be the unconstrained \\(N - 1\\) vector and \\(y\\) is implicitly given a standard normal prior, then the sum-to-zero vector distributed as normal with a mean of zero and a standard deviation of \\(\\sigma\\) is given by \\[\nx = \\underbrace{\\sigma\\sqrt{\\frac{N}{N-1}}\\;H}_{\\text{ size of } N \\times (N - 1)}\\,y.\n\\] This is the classic non-centered model. The crucial detail from above is that the operation of multiplying \\(\\sigma\\) to \\(y\\) is on \\(N - 1\\) dimensions, not \\(N\\) dimensions. When writing the centered model using \\[\nx \\sim \\mathcal N \\bigg(0,\\, \\sigma\\sqrt{\\frac{N}{N-1}} \\bigg),\n\\] we are incrementing the log density by an additional \\(-\\log(\\sigma)\\) and so must increment the log density by\ntarget += log(sigma * sqrt(N * inv(N - 1)));\n\n// or\n// because `sqrt(N * inv(N - 1))` is constant\n\ntarget += log(sigma);\nwhich correctly adjusts for the \\(N - 1\\) free variables.\nMore details of the Helmert matrix are in (Lancaster 1965), for the basic definitions of the isometric log ratio transform see (Egozcue et al. 2003) and Chapter 3 of (Filzmoser, Hron, and Templ 2018) for the pivot coordinate version used here.\n\n\nVectors that are constrained to sum-to-zero are useful for, among other things, additive varying effects, such as varying slopes or intercepts in a regression model (e.g., for income deciles).\nA sum-to-zero \\(K\\)-vector \\(x \\in \\mathbb{R}^K\\) satisfies the constraint \\[\n\\sum_{k=1}^K x_k = 0.\n\\]\n\n\n\nThe transform is defined iteratively. Given an \\(x \\in \\mathbb{R}^{N + 1}\\) that sums to zero (i.e., \\(\\sum_{n=1}^{N+1} x_n = 0\\)), the transform proceeds as follows to produce an unconstrained \\(y \\in \\mathbb{R}^N\\). This is mathematically equivalent to pre-multiplying the unconstrained \\(y\\) by an orthogonal standard basis, e.g. constructing orthogonal vectors from the standard basis using the Gram-Schmidt process. The single loop version below achieves low computational and memory costs as no matrices are created or multplied.\nThe transform is initialized by setting \\[\nS_N = 0\n\\] and \\[\ny_N = -x_{N + 1} \\sqrt{1 + \\frac{1}{N}}.\n\\] The for each \\(n\\) from \\(N - 1\\) down to \\(1\\), let \\[\nw_{n + 1} = \\frac{y_{n + 1}}{\\sqrt{(n + 1)(n + 2)}},\n\\] \\[\nS_n = S_{n + 1} + w_{n + 1},\n\\] and \\[\ny_n = (S_n - x_{n + 1}) \\frac{\\sqrt{n (n + 1)}}{n}.\n\\]\nThis transform is expressed in Stan code as:\n  vector manual_sum_to_zero_jacobian(vector x) {\n    int N = size(x) - 1;\n    vector[N] y;\n    y[N] = -x[N+1] * sqrt(1 + 1. / N);\n    real sum_w = 0;\n    for (n in 1:(N-1)) {\n      int i = N - n;\n      int i_p_1 = i + 1;\n      real w = y[i_p_1] * inv_sqrt(i_p_1 * (i_p_1 + 1));\n      sum_w += w;\n      y[i] = (sum_w - x[i_p_1]) * sqrt(i_p_1 * i) / i;\n    }\n    return y;\n  }\n\n\n\nThe inverse transform follows the isometric logratio tranform. It maps an unconstrained vector \\(y \\in \\mathbb{R}^N\\) to a sum-to-zero vector \\(x \\in \\mathbb{R}^{N + 1}\\) such that \\[\n\\sum_{n=1}^{N + 1} x_n = 0.\n\\] The values are defined inductively, starting with \\[\nx_1 = \\sum_{n=1}^N \\frac{y_n}{\\sqrt{n (n + 1)}}\n\\] and then setting \\[\nx_{n + 1} = \\sum_{i = n + 1}^N \\frac{y_i}{\\sqrt{i (i + 1)}}\n- n \\cdot \\frac{y_n}{\\sqrt{n (n + 1)}}.\n\\] for \\(n \\in 1{:}N\\).\nThe definition is such that \\[\n\\sum_{n = 1}^{N + 1} x_n = 0\n\\] by construction, because each of the terms added to \\(x_{n}\\) is then subtracted from \\(x_{n + 1}\\) the number of times it shows up in earlier terms.\n\n\n\nThe inverse transform is a linear operation, leading to a constant Jacobian determinant which is therefore not included.\nThe sum-to-zero inverse transform is expressed within Stan as:\n  vector manual_sum_to_zero_inverse_jacobian(vector y) {\n    int N = num_elements(y);\n    vector[N + 1] x = zeros_vector(N + 1);\n    real sum_w = 0;\n    for (n in 1:N) {\n      int i = N - n + 1;\n      real w = y[i] * inv_sqrt(i * (i + 1));\n      sum_w += w;\n      x[i] += sum_w;\n      x[i + 1] -= w * i;\n    }\n    return x;\n  }\nNote that there is no target += increment because the Jacobian is zero.\n\n\n\nThe matrix case of the sum-to-zero transform generalizes the vector case to ensure that every row of the matrix sums to zero and every column the matrix sums to zero. In fact, any \\(N\\)-dimensional array can be constructed into a sum-to-zero N-dimensional array using the sum-to-zero vector. This is because the vector transform is a linear bijection and produces an orthogonally constructed sum-to-zero object by applying the one dimensional transform across each array slice. For the matrix case, there are two slices present, the rows and columns, to perform the transform over. The sum-to-zero vector is applied over the vectorized slice of either the row or column slice and subsequently to the other slice.\nLet the unconstrained matrix be \\[\n\\mathcal Y \\in \\mathbb R^{n_1 \\times n_2}\n\\] and the zero sum vector transform as \\[\n\\mathbf z = \\mathcal C_n(\\mathbf y)\\;=\\;\n\\begin{bmatrix}H_n\\\\[2pt]-\\mathbf 1_{1\\times d}\\end{bmatrix}\\mathbf y\n\\;\\in\\mathbb R^{n+1},\n\\] where \\(H_n\\in\\mathbb R^{n \\times n}\\) is the orthogonal Helmert matrix and satisfies \\(\\mathbf 1_{1\\times d}A_d^{\\!\\top}=0\\).\nApplying \\(C_n\\) to each slice results in\n\\[\n\\mathcal Z \\;=\\;\n\\mathcal X\n\\times_1 \\bigl[\\mathcal C_{n_1}\\bigr]\n\\times_2 \\bigl[\\mathcal C_{n_2}\\bigr]\n\\] where \\[\n\\mathcal Z \\in\n\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}.\n\\]\nBecause each \\(\\mathbb R^{d_1\\times\\cdots\\times d_N}\\) is invertible on the \\(\\mathbf 1^\\perp\\) subspace the composite map applied to \\(\\mathcal Z\\) is a linear bijection between \\(\\mathbb R^{n_1\\times n_2}\\) and the codomain \\(\\mathbb R^{(n_1+1)\\times\\cdots\\times(n_2+1)}\\).",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#cdf_log-and-ccdf_log-cumulative-distribution-functions",
-    "href": "reference-manual/removals.html#cdf_log-and-ccdf_log-cumulative-distribution-functions",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#simplex-transform.section",
+    "href": "reference-manual/transforms.html#simplex-transform.section",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The log cumulative distribution and complementary cumulative distribution functions for a distribution foo were formerly written as foo_cdf_log and foo_ccdf_log.\nReplacement:\nReplace foo_cdf_log(y, ...) with foo_lcdf(y | ...).\nReplace foo_ccdf_log(y, ...) with foo_lccdf(y | ...).",
+    "text": "Variables constrained to the unit simplex show up in multivariate discrete models as both parameters (categorical and multinomial) and as variates generated by their priors (Dirichlet and multivariate logistic).\nThe unit \\(K\\)-simplex is the set of points \\(x \\in \\mathbb{R}^K\\) such that for \\(1 \\leq k \\leq K\\),\n\\[\nx_k &gt; 0,\n\\]\nand\n\\[\n\\sum_{k=1}^K x_k = 1.\n\\]\nAn alternative definition is to take the convex closure of the vertices. For instance, in 2-dimensions, the simplex vertices are the extreme values \\((0,1)\\), and \\((1,0)\\) and the unit 2-simplex is the line connecting these two points; values such as \\((0.3,0.7)\\) and \\((0.99,0.01)\\) lie on the line. In 3-dimensions, the basis is \\((0,0,1)\\), \\((0,1,0)\\) and \\((1,0,0)\\) and the unit 3-simplex is the boundary and interior of the triangle with these vertices. Points in the 3-simplex include \\((0.5,0.5,0)\\), \\((0.2,0.7,0.1)\\) and all other triplets of non-negative values summing to 1.\nAs these examples illustrate, the simplex always picks out a subspace of \\(K-1\\) dimensions from \\(\\mathbb{R}^K\\). Therefore a point \\(x\\) in the \\(K\\)-simplex is fully determined by its first \\(K-1\\) elements \\(x_1, x_2,\n\\ldots, x_{K-1}\\), with\n\\[\nx_K = 1 - \\sum_{k=1}^{K-1} x_k.\n\\]\n\n\nThe length-\\(K\\) unit simplex inverse transform is given by the softmax of a sum-to-zero vector of length \\(K\\).\nLet \\(y\\) represent the unconstrained \\(K - 1\\) values in \\((-\\infty, \\infty)\\). The intermediate sum-to-zero vector \\(z = \\text{sum\\_to\\_zero\\_transform}(y)\\) is length \\(K\\). The unit simplex is then given by \\[\nx_i = \\text{softmax}(z) = \\frac{\\exp(z_i)}{\\sum_{i = 1}^K \\exp(z_i)}\n\\]\nThe sum-to-zero vector transform is described in further detail at the sum-to-zero vector section of the Reference Manual.\n\n\n\n\n\n\nNote\n\n\n\nAll versions of Stan pre-2.37 used the stick-breaking transform. This is documented at Stan 2.36 Reference Manual: Simplex Transform.\n\n\n\n\nThe Jacobian \\(J\\) of the inverse unit-simplex transform is found by restricting \\(J\\) to the subspace spanned by the sum-to-zero vector \\(z\\). The Jacobian is given as the \\((K - 1) \\times (K - 1)\\) matrix \\(J\\) where\n\\[\nJ_{ij} = \\frac{\\partial x_i}{\\partial z_j} =\n\\frac{\\partial}{\\partial z_i} \\left( \\frac{\\exp(z_i)}{{\\sum_{i = 1}^K \\exp(z_i)}} \\right)\n\\] and \\(i,j \\in 1, \\ldots, K - 1\\).\nThe diagonal and off-diagonal derivatives are found using the derivative quotient rule and algebraic simplification\n\\[\nJ_{ij} =\n\\begin{cases}\nx_i (1 - x_i), & \\text{if } i = j, \\\\\n-x_i x_j, & \\text{if } i \\neq j.\n\\end{cases}\n\\]\nIn matrix form this can be expressed as\n\\[\nJ = \\text{diag}(x) - x x^\\top\n\\]\nThe determinant of this matrix can be found using the Matrix Determinant Lemma:\n\\[\n\\det\\bigl(A + u v^{\\top}\\bigr)\n=\n\\det(A)\\,\\bigl(1 + v^{\\top}A^{-1}u\\bigr).\n\\]\nHere,\n\\[\nA \\;=\\; \\operatorname{diag}(x_{1},\\ldots, x_{K-1}),\n\\quad\nu \\;=\\; -\\bigl(x_1,\\ldots, x_{K-1}\\bigr)^{\\!\\top},\n\\quad\nv \\;=\\; \\bigl(x_{1}, \\ldots, x_{K-1}\\bigr)^{\\!\\top}.\n\\] Therefore,\n\\[\n\\begin{aligned}\n\\det(J)\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_i \\bigg)\n\\bigg(1 + (x_{1},\\ldots, x_{K-1})\\,\\mathrm{diag}\\bigl(x_{1}^{-1},\\ldots,x_{K-1}^{-1}\\bigr)\\,\n\\big(-x_{1},\\ldots,-x_{K-1}\\big)^{\\top}\n\\bigg) \\\\\n&=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg)\n\\bigg(1 - \\sum_{i=1}^{K-1} x_{i}\\bigg)\n=\n\\bigg(\\prod_{i=1}^{K-1} x_{i}\\bigg) x_{K} \\\\\n&=\n\\prod_{i=1}^{K} x_{i}.\n\\end{aligned}\n\\]\n\n\n\n\nThe transform \\(Y = f(X)\\) can be derived by reversing the stages of the inverse transform,\n\\[\ny_k\n= H^\\top \\bigg(\\log(x_k)\n- \\frac{1}{K}\\sum_{i=1}^K\\log(x_i) \\bigg)\n.\n\\]\nThe matrix \\(H\\) is the orthogonal basis matrix the sum-to-zero vector uses. Since the matrix is orthonormal, the transpose is the same as the inverse.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#user-defined-function-with-_log-suffix",
-    "href": "reference-manual/removals.html#user-defined-function-with-_log-suffix",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#stochastic-matrix-transform.section",
+    "href": "reference-manual/transforms.html#stochastic-matrix-transform.section",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: A user-defined function ending in _log can be no longer be used in statements.qmd#distribution-statements.section.\nReplacement: Replace the _log suffix with _lpdf for density functions or _lpmf for mass functions in the user-defined function.\nRemoved In: Stan 2.33\nNote: Following Stan 2.33, users can stil define a function ending in _log, it simply no longer has a special meaning or is supported in the ~ syntax.",
+    "text": "The column_stochastic_matrix[N, M] and row_stochastic_matrix[M, N] type in Stan represents an \\(N \\times M\\) matrix where each column (row) is a unit simplex of dimension \\(N\\). In other words, each column (row) of the matrix is a vector constrained to have non-negative entries that sum to one.\n\n\nA column stochastic matrix \\(X \\in \\mathbb{R}^{N \\times M}\\) is defined such that each column is a simplex. For column \\(m\\) (where \\(1 \\leq m \\leq M\\)):\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{n=1}^N X_{n, m} = 1.\n\\]\nA row stochastic matrix is any matrix whose transpose is a column stochastic matrix (i.e. the rows of the matrix are simplexes)\n\\[\nX_{n, m} \\geq 0 \\quad \\text{for } 1 \\leq n \\leq N,\n\\]\nand\n\\[\n\\sum_{m=1}^N X_{n, m} = 1.\n\\]\nThis definition ensures that each column (row) of the matrix \\(X\\) lies on the \\(N-1\\) dimensional unit simplex, similar to the simplex[N] type, but extended across multiple columns(rows).\n\n\n\nFor the column and row stochastic matrices the inverse transform is the same as simplex, but applied to each column (row).\n\n\n\nThe Jacobian determinant of the inverse transform for each column \\(m\\) in the matrix is given by the product of the diagonal entries \\(J_{n, m}\\) of the lower-triangular Jacobian matrix. This determinant is calculated as:\n\\[\n\\left| \\det J_m \\right| = \\prod_{n=1}^{N-1} \\left( z_{n, m} (1 - z_{n, m}) \\left( 1 - \\sum_{n'=1}^{n-1} X_{n', m} \\right) \\right).\n\\]\nThus, the overall Jacobian determinant for the entire column_stochastic_matrix and row_stochastic_matrix is the product of the determinants for each column (row):\n\\[\n\\left| \\det J \\right| = \\prod_{m=1}^{M} \\left| \\det J_m \\right|.\n\\]\n\n\n\nFor the column and row stochastic matrices the transform is the same as simplex, but applied to each column (row).",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#if_else-function",
-    "href": "reference-manual/removals.html#if_else-function",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#unit-vector.section",
+    "href": "reference-manual/transforms.html#unit-vector.section",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The function if_else is no longer available.\nReplacement: Use the conditional operator which allows more flexibility in the types of b and c and is much more efficient in that it only evaluates whichever of b or c is returned.\nx = if_else(a, b, c);\nwith\nx = a ? b : c;\nRemoved In: Stan 2.33",
+    "text": "An \\(n\\)-dimensional vector \\(x \\in \\mathbb{R}^n\\) is said to be a unit vector if it has unit Euclidean length, so that\n\\[\n\\Vert x \\Vert\n\\ = \\ \\sqrt{x^{\\top}\\,x}\n\\ = \\ \\sqrt{x_1^2 + x_2^2 + \\cdots + x_n^2}\n\\ = \\ 1\\ .\n\\]\n\n\nStan divides an unconstrained vector \\(y \\in \\mathbb{R}^{n}\\) by its norm, \\(\\Vert y \\Vert = \\sqrt{y^\\top y}\\), to obtain a unit vector \\(x\\),\n\\[\nx = \\frac{y}{\\Vert y \\Vert}.\n\\]\nTo generate a unit vector, Stan generates points at random in \\(\\mathbb{R}^n\\) with independent unit normal distributions, which are then standardized by dividing by their Euclidean length. Muller (1959) showed this generates points uniformly at random on \\(S^{n-1}\\). That is, if we draw \\(y_n \\sim \\mathsf{Normal}(0, 1)\\) for \\(n \\in 1{:}n\\), then \\(x = \\frac{y}{\\Vert y \\Vert}\\) has a uniform distribution over \\(S^{n-1}\\). This allows us to use an \\(n\\)-dimensional basis for \\(S^{n-1}\\) that preserves local neighborhoods in that points that are close to each other in \\(\\mathbb{R}^n\\) map to points near each other in \\(S^{n-1}\\). The mapping is not perfectly distance preserving, because there are points arbitrarily far away from each other in \\(\\mathbb{R}^n\\) that map to identical points in \\(S^{n-1}\\).\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a very small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nThe Jacobian matrix relating the input vector \\(y\\) to the output vector \\(x\\) is singular because \\(x^\\top x = 1\\) for any non-zero input vector \\(y\\). Thus, there technically is no unique transformation from \\(x\\) to \\(y\\). To circumvent this issue, let \\(r = \\sqrt{y^\\top y}\\) so that \\(y = r\nx\\). The transformation from \\(\\left(r, x_{-n}\\right)\\) to \\(y\\) is well-defined but \\(r\\) is arbitrary, so we set \\(r = 1\\). In this case, the determinant of the Jacobian is proportional to \\(e^{-\\frac{1}{2} y^\\top y}\\), which is the kernel of a standard multivariate normal distribution with \\(n\\) independent dimensions.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#character-as-comment-prefix",
-    "href": "reference-manual/removals.html#character-as-comment-prefix",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#correlation-matrix-transform.section",
+    "href": "reference-manual/transforms.html#correlation-matrix-transform.section",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: The use of # for line-based comments is no longer permitted. # may only be used for #include statements.\nReplacement: Use a pair of forward slashes, //, for line comments.\nRemoved In: Stan 2.33",
+    "text": "A \\(K \\times K\\) correlation matrix \\(x\\) must be symmetric, so that\n\\[\nx_{k,k'} = x_{k',k}\n\\]\nfor all \\(k,k' \\in \\{ 1, \\ldots, K \\}\\), it must have a unit diagonal, so that\n\\[\nx_{k,k} = 1\n\\]\nfor all \\(k \\in \\{ 1, \\ldots, K \\}\\), and it must be positive definite, so that for every non-zero \\(K\\)-vector \\(a\\),\n\\[\na^{\\top} x a &gt; 0.\n\\]\nThe number of free parameters required to specify a \\(K \\times K\\) correlation matrix is \\(\\binom{K}{2}\\).\nThere is more than one way to map from \\(\\binom{K}{2}\\) unconstrained parameters to a \\(K \\times K\\) correlation matrix. Stan implements the Lewandowski-Kurowicka-Joe (LKJ) transform Lewandowski, Kurowicka, and Joe (2009).\n\n\nIt is easiest to specify the inverse, going from its \\(\\binom{K}{2}\\) parameter basis to a correlation matrix. The basis will actually be broken down into two steps. To start, suppose \\(y\\) is a vector containing \\(\\binom{K}{2}\\) unconstrained values. These are first transformed via the bijective function \\(\\tanh : \\mathbb{R} \\rightarrow\n(-1, 1)\\)\n\\[\n\\tanh y = \\frac{\\exp(2y) - 1}{\\exp(2y) + 1}.\n\\]\nThen, define a \\(K \\times K\\) matrix \\(z\\), the upper triangular values of which are filled by row with the transformed values, and the diagonal entries are set to one. For example, in the \\(4 \\times 4\\) case, there are \\(\\binom{4}{2}\\) values arranged as\n\\[\nz\n=\n\\left[\n\\begin{array}{cccc}\n1 & \\tanh y_1 & \\tanh y_2 & \\tanh y_4\n\\\\\n0 & 1 & \\tanh y_3 & \\tanh y_5\n\\\\\n0 & 0 & 1 & \\tanh y_6\n\\\\\n0 & 0 & 0 & 1\n\\end{array}\n\\right]\n.\n\\]\nLewandowski, Kurowicka and Joe (LKJ) show how to bijectively map the array \\(z\\) to a correlation matrix \\(x\\). The entry \\(z_{i,j}\\) for \\(i &lt;\nj\\) is interpreted as the canonical partial correlation (CPC) between \\(i\\) and \\(j\\), which is the correlation between \\(i\\)’s residuals and \\(j\\)’s residuals when both \\(i\\) and \\(j\\) are regressed on all variables \\(i'\\) such that \\(i'&lt; i\\). In the case of \\(i=1\\), there are no earlier variables, so \\(z_{1,j}\\) is just the Pearson correlation between \\(i\\) and \\(j\\).\nIn Stan, the LKJ transform is reformulated in terms of a Cholesky factor \\(w\\) of the final correlation matrix, defined for \\(1 \\leq i,j \\leq K\\) by\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\n\\prod_{i'=1}^{i - 1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{if } 1 &lt; i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\\\nz_{i,j} \\, \\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,\\,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThis does not require as much computation per matrix entry as it may appear; calculating the rows in terms of earlier rows yields the more manageable expression\n\\[\nw_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i &gt; j,\n\\\\\n1 & \\mbox{if } 1 = i = j,\n\\\\\nz_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n\\frac{z_{i,j}}{z_{i-1,j}} \\ w_{i-1,j} \\left( 1 - z_{i-1,j}^2 \\right)^{1/2}\n& \\mbox{ if } 1 &lt; i \\leq j.\n\\end{array}\n\\right.\n\\]\nGiven the upper-triangular Cholesky factor \\(w\\), the final correlation matrix is\n\\[\nx = w^{\\top} w.\n\\]\nLewandowski, Kurowicka, and Joe (2009) show that the determinant of the correlation matrix can be defined in terms of the canonical partial correlations as\n\\[\n\\mbox{det} \\, x = \\prod_{i=1}^{K-1} \\ \\prod_{j=i+1}^K \\ (1 - z_{i,j}^2)\n= \\prod_{1 \\leq i &lt; j \\leq K} (1 - z_{i,j}^2),\n\\]\n\n\n\nFrom the inverse of equation 11 in (Lewandowski, Kurowicka, and Joe 2009), the absolute Jacobian determinant is\n\\[\n\\sqrt{\\prod_{i=1}^{K-1}\\prod_{j=i+1}^K \\left(1-z_{i,j}^2\\right)^{K-i-1}} \\\n\\times \\prod_{i=1}^{K-1}\\prod_{j=i+1}^K\n\\frac{\\partial z_{i,j}}{\\partial y_{i,j}}\n\\]\n\n\n\nThe correlation transform is defined by reversing the steps of the inverse transform defined in the previous section.\nStarting with a correlation matrix \\(x\\), the first step is to find the unique upper triangular \\(w\\) such that \\(x = w w^{\\top}\\). Because \\(x\\) is positive definite, this can be done by applying the Cholesky decomposition,\n\\[\nw = \\mbox{chol}(x).\n\\]\nThe next step from the Cholesky factor \\(w\\) back to the array \\(z\\) of canonical partial correlations (CPCs) is simplified by the ordering of the elements in the definition of \\(w\\), which when inverted yields\n\\[\nz_{i,j} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } i \\leq j,\n\\\\\nw_{i,j} & \\mbox{if } 1 = i &lt; j, \\mbox{ and}\n\\\\\n{w_{i,j}}\n\\\n\\prod_{i'=1}^{i-1} \\left( 1 - z_{i'\\!,j}^2 \\right)^{-1/2}\n& \\mbox{if } 1 &lt; i &lt; j.\n\\end{array}\n\\right.\n\\]\nThe final stage of the transform reverses the hyperbolic tangent transform, which is defined by\n\\[\ny = \\tanh^{-1} z = \\frac{1}{2} \\log \\left( \\frac{1 + z}{1 - z} \\right).\n\\]\nThe inverse hyperbolic tangent function, \\(\\tanh^{-1}\\), is also called the Fisher transformation.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#postfix-brackets-array-syntax",
-    "href": "reference-manual/removals.html#postfix-brackets-array-syntax",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#covariance-matrices",
+    "href": "reference-manual/transforms.html#covariance-matrices",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Before Stan 2.26, arrays were declared by writing syntax after the variable.\nRemoved: The use of array declarations like\nint n[5];\nreal a[3, 4];\nreal&lt;lower=0&gt; z[5, 4, 2];\nvector[7] mu[3];\nmatrix[7, 2] mu[15, 12];\ncholesky_factor_cov[5, 6] mu[2, 3, 4];\nReplacement: The use of the array keyword, which replaces the above examples with\narray[5] int n;\narray[3, 4] real a;\narray[5, 4, 2] real&lt;lower=0&gt; z;\narray[3] vector[7] mu;\narray[15, 12] matrix[7, 2] mu;\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nRemoved In: Stan 2.33",
+    "text": "A \\(K \\times K\\) matrix is a covariance matrix if it is symmetric and positive definite (see the previous section for definitions). It requires \\(K + \\binom{K}{2}\\) free parameters to specify a \\(K \\times K\\) covariance matrix.\n\n\nStan’s covariance transform is based on a Cholesky decomposition composed with a log transform of the positive-constrained diagonal elements.1\nIf \\(x\\) is a covariance matrix (i.e., a symmetric, positive definite matrix), then there is a unique lower-triangular matrix \\(z =\n\\mathrm{chol}(x)\\) with positive diagonal entries, called a Cholesky factor, such that\n\\[\nx = z \\, z^{\\top}.\n\\]\nThe off-diagonal entries of the Cholesky factor \\(z\\) are unconstrained, but the diagonal entries \\(z_{k,k}\\) must be positive for \\(1 \\leq k\n\\leq K\\).\nTo complete the transform, the diagonal is log-transformed to produce a fully unconstrained lower-triangular matrix \\(y\\) defined by\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log z_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nz_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform reverses the two steps of the transform. Given an unconstrained lower-triangular \\(K \\times K\\) matrix \\(y\\), the first step is to recover the intermediate matrix \\(z\\) by reversing the log transform,\n\\[\nz_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe covariance matrix \\(x\\) is recovered from its Cholesky factor \\(z\\) by taking\n\\[\nx = z \\, z^{\\top}.\n\\]\n\n\n\nThe Jacobian is the product of the Jacobians of the exponential transform from the unconstrained lower-triangular matrix \\(y\\) to matrix \\(z\\) with positive diagonals and the product transform from the Cholesky factor \\(z\\) to \\(x\\).\nThe transform from unconstrained \\(y\\) to Cholesky factor \\(z\\) has a diagonal Jacobian matrix, the absolute determinant of which is thus\n\\[\n\\prod_{k=1}^K  \\frac{\\partial}{\\partial_{y_{k,k}}} \\, \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K \\exp(y_{k,k})\n\\ = \\\n\\prod_{k=1}^K z_{k,k}.\n\\]\nThe Jacobian matrix of the second transform from the Cholesky factor \\(z\\) to the covariance matrix \\(x\\) is also triangular, with diagonal entries corresponding to pairs \\((m,n)\\) with \\(m \\geq n\\), defined by\n\\[\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( z \\, z^{\\top} \\right)_{m,n}\n\\ = \\\n\\frac{\\partial}{\\partial z_{m,n}}\n\\left( \\sum_{k=1}^K z_{m,k} \\, z_{n,k} \\right)\n\\ = \\\n\\left\\{\n\\begin{array}{cl}\n2 \\, z_{n,n} & \\mbox{if } m = n \\mbox{ and }\n\\\\\nz_{n,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\nThe absolute Jacobian determinant of the second transform is thus\n\\[\n2^{K} \\ \\prod_{m = 1}^{K} \\ \\prod_{n=1}^{m} z_{n,n}\n\\ = \\\n\\prod_{n=1}^K \\ \\prod_{m=n}^K z_{n,n}\n\\ = \\\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}.\n\\]\nFinally, the full absolute Jacobian determinant of the inverse of the covariance matrix transform from the unconstrained lower-triangular \\(y\\) to a symmetric, positive definite matrix \\(x\\) is the product of the Jacobian determinants of the exponentiation and product transforms,\n\\[\n\\left( \\prod_{k=1}^K z_{k,k} \\right)\n\\left(\n2^{K} \\ \\prod_{k=1}^K z_{k,k}^{K - k + 1}\n\\right)\n\\ = \\\n2^K\n\\, \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]\nLet \\(f^{-1}\\) be the inverse transform from a \\(K + \\binom{K}{2}\\)-vector \\(y\\) to the \\(K \\times K\\) covariance matrix \\(x\\). A density function \\(p_X(x)\\) defined on \\(K \\times K\\) covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(K + \\binom{K}{2}\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\ 2^K \\ \\prod_{k=1}^K z_{k,k}^{K-k+2}.\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#nested-multiple-indexing-in-assignments",
-    "href": "reference-manual/removals.html#nested-multiple-indexing-in-assignments",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#cholesky-factors-of-covariance-matrices",
+    "href": "reference-manual/transforms.html#cholesky-factors-of-covariance-matrices",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Stan interprets nested indexing in assingments as flat indexing so that a statement like\na[:][1] = b;\nis the same as\na[:,1] = b;\nHowever, this is inconsistent with multiple indexing rules.\nTo avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.\nRemoved In: Stan 2.33",
+    "text": "An \\(M \\times M\\) covariance matrix \\(\\Sigma\\) can be Cholesky factored to a lower triangular matrix \\(L\\) such that \\(L\\,L^{\\top} = \\Sigma\\). If \\(\\Sigma\\) is positive definite, then \\(L\\) will be \\(M \\times M\\). If \\(\\Sigma\\) is only positive semi-definite, then \\(L\\) will be \\(M \\times N\\), with \\(N &lt; M\\).\nA matrix is a Cholesky factor for a covariance matrix if and only if it is lower triangular, the diagonal entries are positive, and \\(M \\geq\nN\\). A matrix satisfying these conditions ensures that \\(L \\,\nL^{\\top}\\) is positive semi-definite if \\(M &gt; N\\) and positive definite if \\(M = N\\).\nA Cholesky factor of a covariance matrix requires \\(N + \\binom{N}{2} +\n(M - N)N\\) unconstrained parameters.\n\n\nStan’s Cholesky factor transform only requires the first step of the covariance matrix transform, namely log transforming the positive diagonal elements. Suppose \\(x\\) is an \\(M \\times N\\) Cholesky factor. The above-diagonal entries are zero, the diagonal entries are positive, and the below-diagonal entries are unconstrained. The transform required is thus\n\\[\ny_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\log x_{m,m} & \\mbox{if } m = n, \\mbox{ and}\n\\\\\nx_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe inverse transform need only invert the logarithm with an exponentiation. If \\(y\\) is the unconstrained matrix representation, then the elements of the constrained matrix \\(x\\) is defined by\n\\[\nx_{m,n} =\n\\left\\{\n\\begin{array}{cl}\n0 & \\mbox{if } m &lt; n,\n\\\\\n\\exp(y_{m,m}) & \\mbox{if } m = n, \\mbox{ and}\n\\\\\ny_{m,n} & \\mbox{if } m &gt; n.\n\\end{array}\n\\right.\n\\]\n\n\n\nThe transform has a diagonal Jacobian matrix, the absolute determinant of which is\n\\[\n\\prod_{n=1}^N  \\frac{\\partial}{\\partial_{y_{n,n}}} \\, \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N \\exp(y_{n,n})\n\\ = \\\n\\prod_{n=1}^N x_{n,n}.\n\\]\nLet \\(x = f^{-1}(y)\\) be the inverse transform from a \\(N + \\binom{N}{2}\n+ (M - N)N\\) vector to an \\(M \\times N\\) Cholesky factor for a covariance matrix \\(x\\) defined in the previous section. A density function \\(p_X(x)\\) defined on \\(M \\times N\\) Cholesky factors of covariance matrices is transformed to the density \\(p_Y(y)\\) over \\(N + \\binom{N}{2}\n+ (M - N)N\\) vectors \\(y\\) by\n\\[\np_Y(y) = p_X(f^{-1}(y)) \\prod_{N=1}^N x_{n,n}.\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/removals.html#real-values-in-conditionals",
-    "href": "reference-manual/removals.html#real-values-in-conditionals",
-    "title": "Removed Features",
+    "objectID": "reference-manual/transforms.html#cholesky-factors-of-correlation-matrices",
+    "href": "reference-manual/transforms.html#cholesky-factors-of-correlation-matrices",
+    "title": "Constraint Transforms",
     "section": "",
-    "text": "Removed: Using a real value in a conditional is no longer permitted.\nreal x = 1.0;\nif (x) {\nThe value was interpreted as true if it is nonzero.\nReplacement: For the exact equivalent, use a comparison operator to make the intent clear.\nreal x = 1.0;\nif (x != 0) {\nHowever, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative abs(x) &lt; machine_precision() is appropriate for the use case.\nRemoved In: Stan 2.34",
+    "text": "A \\(K \\times K\\) correlation matrix \\(\\Omega\\) is positive definite and has a unit diagonal. Because it is positive definite, it can be Cholesky factored to a \\(K \\times K\\) lower-triangular matrix \\(L\\) with positive diagonal elements such that \\(\\Omega = L\\,L^{\\top}\\). Because the correlation matrix has a unit diagonal,\n\\[\n\\Omega_{k,k} = L_k\\,L_k^{\\top} = 1,\n\\]\neach row vector \\(L_k\\) of the Cholesky factor is of unit length. The length and positivity constraint allow the diagonal elements of \\(L\\) to be calculated from the off-diagonal elements, so that a Cholesky factor for a \\(K \\times K\\) correlation matrix requires only \\(\\binom{K}{2}\\) unconstrained parameters.\n\n\nIt is easiest to start with the inverse transform from the \\(\\binom{K}{2}\\) unconstrained parameters \\(y\\) to the \\(K \\times K\\) lower-triangular Cholesky factor \\(x\\). The inverse transform is based on the hyperbolic tangent function, \\(\\tanh\\), which satisfies \\(\\tanh(x) \\in (-1,1)\\). Here it will function like an inverse logit with a sign to pick out the direction of an underlying canonical partial correlation; see the section on correlation matrix transforms for more information on the relation between canonical partial correlations and the Cholesky factors of correlation matrices.\nSuppose \\(y\\) is a vector of \\(\\binom{K}{2}\\) unconstrained values. Let \\(z\\) be a lower-triangular matrix with zero diagonal and below diagonal entries filled by row. For example, in the \\(3 \\times 3\\) case,\n\\[\nz =\n\\left[\n\\begin{array}{ccc}\n0 & 0 & 0\n\\\\\n\\tanh y_1 & 0 & 0\n\\\\\n\\tanh y_2 & \\tanh y_3 & 0\n\\end{array}\n\\right]\n\\]\nThe matrix \\(z\\), with entries in the range \\((-1, 1)\\), is then transformed to the Cholesky factor \\(x\\), by taking2\n\\[\nx_{i,j}\n=\n\\left\\{\n\\begin{array}{lll}\n0 & \\mbox{ if } i &lt; j & \\mbox{ [above diagonal]}\n\\\\\n\\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i = j & \\mbox{ [on diagonal]}\n\\\\\nz_{i,j} \\ \\sqrt{1 - \\sum_{j' &lt; j} x_{i,j'}^2}\n  & \\mbox{ if } i &gt; j & \\mbox{ [below diagonal]}\n\\end{array}\n\\right.\n\\]\nIn the \\(3 \\times 3\\) case, this yields\n\\[\nx =\n\\left[\n\\begin{array}{ccc}\n1 & 0 & 0\n\\\\\nz_{2,1} & \\sqrt{1 - x_{2,1}^2} & 0\n\\\\\nz_{3,1} & z_{3,2} \\sqrt{1 - x_{3,1}^2}\n        & \\sqrt{1 - (x_{3,1}^2 + x_{3,2}^2)}\n\\end{array}\n\\right],\n\\]\nwhere the \\(z_{i,j} \\in (-1,1)\\) are the \\(\\tanh\\)-transformed \\(y\\).\nThe approach is a signed stick-breaking process on the quadratic (Euclidean length) scale. Starting from length 1 at \\(j=1\\), each below-diagonal entry \\(x_{i,j}\\) is determined by the (signed) fraction \\(z_{i,j}\\) of the remaining length for the row that it consumes. The diagonal entries \\(x_{i,i}\\) get any leftover length from earlier entries in their row. The above-diagonal entries are zero.\n\n\n\nSuppose \\(x\\) is a \\(K \\times K\\) Cholesky factor for some correlation matrix. The first step of the transform reconstructs the intermediate values \\(z\\) from \\(x\\),\n\\[\nz_{i,j} = \\frac{x_{i,j}}{\\sqrt{1 - \\sum_{j' &lt; j}x_{i,j'}^2}}.\n\\]\nThe mapping from the resulting \\(z\\) to \\(y\\) inverts \\(\\tanh\\),\n\\[\ny\n\\ = \\\n\\tanh^{-1} z\n\\ = \\\n\\frac{1}{2} \\left( \\log (1 + z) - \\log (1 - z) \\right).\n\\]\n\n\n\nThe Jacobian of the full transform is the product of the Jacobians of its component transforms.\nFirst, for the inverse transform \\(z = \\tanh y\\), the derivative is\n\\[\n\\frac{d}{dy} \\tanh y = \\frac{1}{(\\cosh y)^2}.\n\\]\nSecond, for the inverse transform of \\(z\\) to \\(x\\), the resulting Jacobian matrix \\(J\\) is of dimension \\(\\binom{K}{2} \\times\n\\binom{K}{2}\\), with indexes \\((i,j)\\) for \\((i &gt; j)\\). The Jacobian matrix is lower triangular, so that its determinant is the product of its diagonal entries, of which there is one for each \\((i,j)\\) pair,\n\\[\n\\left| \\, \\mbox{det} \\, J \\, \\right|\n  \\ = \\ \\prod_{i &gt; j} \\left| \\frac{d}{dz_{i,j}} x_{i,j} \\right|,\n\\]\nwhere\n\\[\n\\frac{d}{dz_{i,j}} x_{i,j}\n= \\sqrt{1 - \\sum_{j' &lt; j} x^2_{i,j'}}.\n\\]\nSo the combined density for unconstrained \\(y\\) is\n\\[\np_Y(y)\n= p_X(f^{-1}(y))\n  \\ \\\n  \\prod_{n &lt; \\binom{K}{2}} \\frac{1}{(\\cosh y)^2}\n  \\ \\\n  \\prod_{i &gt; j} \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2\n  \\right)^{1/2},\n\\]\nwhere \\(x = f^{-1}(y)\\) is used for notational convenience. The log Jacobian determinant of the complete inverse transform \\(x = f^{-1}(y)\\) is given by\n\\[\n\\log \\left| \\, \\det J \\, \\right|\n=\n-2 \\sum_{n \\leq \\binom{K}{2}}\n\\log \\cosh y\n\\\n+\n\\\n\\frac{1}{2} \\\n\\sum_{i &gt; j}\n\\log \\left( 1 - \\sum_{j' &lt; j} x_{i,j'}^2 \\right)\n.\n\\]",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Removed Features"
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/pathfinder.html",
-    "href": "reference-manual/pathfinder.html",
-    "title": "Pathfinder",
-    "section": "",
-    "text": "Stan supports the Pathfinder algorithm (Zhang et al. 2022). Pathfinder is a variational method for approximately sampling from differentiable log densities. Starting from a random initialization, Pathfinder locates normal approximations to the target density along a quasi-Newton optimization path, with local covariance estimated using the negative inverse Hessian estimates produced by the LBFGS optimizer. Pathfinder returns draws from the Gaussian approximation with the lowest estimated Kullback-Leibler (KL) divergence to the true posterior.\nStan provides two versions of the Pathfinder algorithm: single-path Pathfinder and multi-path Pathfinder. Single-path Pathfinder generates a set of approximate draws from one run of the basic Pathfinder algorithm. Multi-path Pathfinder uses importance resampling over the draws from multiple runs of Pathfinder. This better matches non-normal target densities and also mitigates the problem of L-BFGS getting stuck at local optima or in saddle points on plateaus. Compared to ADVI and short dynamic HMC runs, Pathfinder requires one to two orders of magnitude fewer log density and gradient evaluations, with greater reductions for more challenging posteriors. While the evaluations by Zhang et al. (2022) found that single-path and multi-path Pathfinder outperform ADVI for most of the models in the PosteriorDB (Magnusson et al. 2024) evaluation set, we recognize the need for further experiments on a wider range of models.\n\n\nPathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-\\(\\hat{k}\\) diagnostic (Vehtari et al. 2024) to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably (Vehtari et al. 2024, sec. 3), which is the first requirement for reliable resampling. If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest (Vehtari et al. 2024, sec. 2.2). If estimated Pareto-\\(\\hat{k}\\) is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow (Gelman et al. 2020).\n\n\n\nIf estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-\\(\\hat{k}\\) for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g. uniform from -2 to 2 used by Stan by default). If Pareto-\\(\\hat{k}\\) is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.",
+    "objectID": "reference-manual/transforms.html#footnotes",
+    "href": "reference-manual/transforms.html#footnotes",
+    "title": "Constraint Transforms",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nAn alternative to the transform in this section, which can be coded directly in Stan, is to parameterize a covariance matrix as a scaled correlation matrix. An arbitrary \\(K \\times K\\) covariance matrix \\(\\Sigma\\) can be expressed in terms of a \\(K\\)-vector \\(\\sigma\\) and correlation matrix \\(\\Omega\\) as \\[\\Sigma = \\mbox{diag}(\\sigma) \\times \\Omega \\times \\mbox{diag}(\\sigma),\\] so that each entry is just a deviation-scaled correlation, \\[\\Sigma_{m,n} = \\sigma_m \\times \\sigma_n \\times \\Omega_{m,n}.\\]↩︎\nFor convenience, a summation with no terms, such as \\(\\sum_{j' &lt; 1} x_{i,j'}\\), is defined to be 0. This implies \\(x_{1,1} = 1\\) and that \\(x_{i,1} = z_{i,1}\\) for \\(i &gt; 1\\).↩︎",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Pathfinder"
+      "Language",
+      "Constraint Transforms"
     ]
   },
   {
-    "objectID": "reference-manual/pathfinder.html#diagnosing-pathfinder",
-    "href": "reference-manual/pathfinder.html#diagnosing-pathfinder",
-    "title": "Pathfinder",
+    "objectID": "reference-manual/statements.html",
+    "href": "reference-manual/statements.html",
+    "title": "Statements",
     "section": "",
-    "text": "Pathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-\\(\\hat{k}\\) diagnostic (Vehtari et al. 2024) to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably (Vehtari et al. 2024, sec. 3), which is the first requirement for reliable resampling. If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest (Vehtari et al. 2024, sec. 2.2). If estimated Pareto-\\(\\hat{k}\\) is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow (Gelman et al. 2020).",
+    "text": "The blocks of a Stan program are made up of variable declarations and statements; see the blocks chapter for details. Unlike programs in BUGS, the declarations and statements making up a Stan program are executed in the order in which they are written. Variables must be defined to have some value (as well as declared to have some type) before they are used — if they do not, the behavior is undefined.\nThe basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters. Log probability functions can be constructed by using distribution statements and log probability increment statements. Statements may be grouped into sequences and into for-each loops. In addition, Stan allows local variables to be declared in blocks and also allows an empty statement consisting only of a semicolon.\n\n\nThe data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the blocks chapter for more information about the block structure of Stan programs.\n\n\n\nAn assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.\nn = 0;\nExecuting this statement assigns the value of the expression 0, which is the integer zero, to the variable n. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because 0 is an expression of type int, the variable n must be declared as being of type int or of type real. If the variable is of type real, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable n will have the value zero (either as an integer or a floating-point value, depending on its type).\nSyntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).\nBecause the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing\nn = n + 1;\nSuch self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.\nThe left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if Sigma is of type matrix, then\nSigma[1, 1] = 1.0;\nsets the value in the first column of the first row of Sigma to one.\nAssignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, a is an array of type array[,] real and b is an array of type array[] real, then the following two statements are both well-formed.\na[3] = b;\nb = a[4];\nSimilarly, if x is a variable declared to have type row_vector and Y is a variable declared as type matrix, then the following sequence of statements to swap the first two rows of Y is well formed.\nx = Y[1];\nY[1] = Y[2];\nY[2] = x;\n\n\nStan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type int to an lvalue of type real, and we can assign an expression of type real to an lvalue of type complex. Furthermore, promotion is transitive, so that we can assign an expression of type int to an lvalue of type complex.\nPromotion extends to containers, so that arrays of int can be promoted to arrays of real during assignment, and arrays of real can be assigned to an lvalue of type array of complex. Similarly, an expression of type vector may be assigned to an lvalue of type complex_vector, and similarly for row vectors and matrices.\n\n\n\nThe expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,\n\na variable, or\na variable with one or more indices, or\na comma separated list of lvalues surrounded by ( and )\n\nTo be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.\n\n\n\nMultiple indexes, as described in the multi-indexing section, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.\n\n\nAll assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.\n\n\n\n\nStan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.\nreal x = 5;\nx += 7;  // value of x is now 12\nThe compound arithmetic and assignment statement above is equivalent to the following long form.\nx = x + 7;\nIn general, the compound form\nx op= y\nwill be equivalent to\nx = x op y;\nThe compound statement will be legal whenever the long form is legal. This requires that the operation x op y must itself be well formed and that the result of the operation be assignable to x. For the expression x to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.\nvector[N] x;\nx[3] += 2;\nAs a further example, consider\nmatrix[M, M] x;\nvector[M] y;\nreal z;\nx *= x;  // OK, (x * x) is a matrix\nx *= z;  // OK, (x * z) is a matrix\nx *= y;  // BAD, (x * y) is a vector\nThe supported compound arithmetic and assignment operations are listed in the compound arithmetic/assignment table; they are also listed in the index prefaced by operator, e.g., operator+=.\nCompound Arithmetic/Assignment Table.  Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.\n\n\n\noperation\ncompound\nunfolded\n\n\n\n\naddition\nx += y\nx = x + y\n\n\nsubtraction\nx -= y\nx = x - y\n\n\nmultiplication\nx *= y\nx = x * y\n\n\ndivision\nx /= y\nx = x / y\n\n\nelementwise multiplication\nx .*= y\nx = x .* y\n\n\nelementwise division\nx ./= y\nx = x ./ y\n\n\n\n\n\n\n\nThe basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.1\ntarget += -0.5 * y * y;\nThe keyword target here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).\nIn this example, the unnormalized log probability of a unit normal variable \\(y\\) is added to the total log probability. In the general case, the argument can be any expression.2\nAn entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.\nparameters {\n  real y;\n}\nmodel {\n  target += -0.5 * y * y;\n}\nThis model defines a log probability function\n\\[\n\\log p(y) = - \\, \\frac{y^2}{2} - \\log Z\n\\]\nwhere \\(Z\\) is a normalizing constant that does not depend on \\(y\\). The constant \\(Z\\) is conventionally written this way because on the linear scale, \\[\np(y) = \\frac{1}{Z} \\exp\\left(-\\frac{y^2}{2}\\right).\n\\] which is typically written without reference to \\(Z\\) as \\[\np(y) \\propto \\exp\\left(-\\frac{y^2}{2}\\right).\n\\]\nStan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant \\(Z\\) is either time-consuming to compute or intractable to evaluate.\n\n\nThe built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.\nFor instance, the normal_lpdf function returns the log density of a normal distribution:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nThe normal_lupdf function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant sigma term, normal_lupdf would be equivalent to:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nAll functions ending in _lpdf have a corresponding _lupdf version which evaluates and returns the unnormalized density. The same is true for _lpmf and _lupmf.\n\n\n\nThe increment log density statement looks syntactically like compound addition and assignment (see the compound arithmetic/assignment section, it is treated as a primitive statement because target is not itself a variable. So, even though\ntarget += lp;\nis a legal statement, the corresponding long form is not legal.\ntarget = target + lp;  // BAD, target is not a variable\n\n\n\nThe target += ... statement accepts an argument in place of ... for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.\n\n\n\n\nA variant of the target += statement described above is the jacobian += statement. This can be used in the transformed parameters block or in functions ending with _jacobian to mimic the log Jacobian adjustments accrued by built-in variable transforms.\nSimilarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.\nFor example, here is a program which recreates the existing &lt;upper=x&gt; transform on real numbers:\nfunctions {\n  real my_upper_bound_jacobian(real x, real ub) {\n    jacobian += x;\n    return ub - exp(x);\n  }\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  // use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters\n  // e.g.\n  // b ~ lognormal(0, 1);\n}\n\n\nTo access the accumulated log density up to the current execution point, the function target() may be used.\n\n\n\n\nThe term “sampling statement” has been replaced with distribution statement.\n\n\n\nStan supports writing probability statements also using distribution statements, for example\ny ~ normal(mu, sigma);\nmu ~ normal(0, 10);\nsigma ~ normal(0, 1);\nThe symbol \\(\\sim\\) is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.\nIn general, we can read \\(\\sim\\) as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as \\[\n\\begin{aligned}\n   p(y| \\mu, \\sigma) & = \\mathrm{normal}(y |  \\mu, \\sigma)\\\\\n   p(\\mu) & = \\mathrm{normal}(\\mu |  0, 10)\\\\\n   p(\\sigma) & = \\mathrm{normal}^+(\\sigma |  0, 1).\n\\end{aligned}\n\\] A collection of distribution statements define a joint distribution as the product of component distributions \\[\np(y,\\mu,\\sigma) = p(y| \\mu, \\sigma )p(\\mu) p(\\sigma).\n\\]\nThis works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:\n  a ~ normal(0, 1);\n  a ~ normal(0, 1);\nThis is translated to \\[\n    p(a) = \\mathrm{normal}(a | 0, 1)\\mathrm{normal}(a |  0, 1),\n\\] which in this case is \\(\\mathrm{normal}(a|0,1/\\sqrt{2})\\). One might expect that the above two lines of code would represent a redundant expression of a \\(\\mathrm{normal}(a|0,1)\\) prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.\nWhen the joint distribution is considered as a function of parameters (e.g. \\(\\mu\\), \\(\\sigma\\)) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.\nStan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each \\(\\sim\\) statement corresponds to a multiplicative factor in the unnormalized posterior density.\nDistribution statements (~) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).\nIn Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as\ntarget += normal_lpdf(y | mu, sigma);\ntarget += normal_lpdf(mu | 0, 10);\ntarget += normal_lpdf(sigma | 0, 1);\nStan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g. censored data model) or coding language limitations (e.g. mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.\nIn general, a distribution statement of the form\ny ~ dist(theta1, ..., thetaN);\ninvolving subexpressions y and theta1 through thetaN (including the case where N is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real y values, the log probability density function is used,\ntarget += dist_lpdf(y | theta1, ..., thetaN);\nFor those restricted to integer y values, the log probability mass function is used,\ntarget += dist_lpmf(y | theta1, ..., thetaN);\nThis will be well formed if and only if dist_lpdf(y | theta1,   ..., thetaN) or dist_lpmf(y | theta1, ..., thetaN) is a well-formed expression of type real. User defined distributions can be defined in functions block by using function names ending with _lpdf.\n\n\nAlthough both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in\ny ~ normal(mu, sigma);\nand explicitly incrementing the log probability function, as in\ntarget += normal_lpdf(y | mu, sigma);\nThe distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to normal_lpdf adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, y, mu, or sigma, involve only constants, data variables, and transformed data variables. See the section Built in distributions above discussing _lupdf and _lupmf functions that also drops all the constant terms.\n\n\n\nThe left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write\nparameters {\n  real&lt;lower=0&gt; beta;\n}\n// ...\nmodel {\n  log(beta) ~ normal(mu, sigma);\n}\nUnfortunately, this is not enough to properly model beta as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to beta here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure beta gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the change of variables section for full definitions. For the case above, the following adjustment will account for the log transform.3\ntarget += - log(abs(y));\n\n\n\nStan supports truncating distributions with lower bounds, upper bounds, or both.\n\n\nA probability density function \\(p(x)\\) for a continuous distribution may be truncated to an interval \\([a, b]\\) to define a new density \\(p_{[a, b]}(x)\\) with support \\([a, b]\\) by setting\n\\[\np_{[a, b]}(x)\n= \\frac{p(x)}\n       {\\int_a^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) for a discrete distribution may be truncated to the closed interval \\([a, b]\\) by\n\\[\np_{[a, b]}(x) = \\frac{p(x)}\n                  {\\sum_{u = a}^b p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x)\n= \\frac{p(x)}\n       {\\int_a^{\\infty} p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x) = \\frac{p(x)}\n                  {\\sum_{a &lt;= u} p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty, b]}(x)\n= \\frac{p(x)}\n       {\\int_{-\\infty}^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty,b]}(x) = \\frac{p(x)}\n                  {\\sum_{u &lt;= b} p(u)}.\n\\]\n\n\n\nGiven a probability function \\(p_X(x)\\) for a random variable \\(X\\), its cumulative distribution function (cdf) \\(F_X(x)\\) is defined to be the probability that \\(X \\leq x\\),\n\\[\nF_X(x) = \\Pr[X \\leq x].\n\\]\nThe upper-case variable \\(X\\) is the random variable whereas the lower-case variable \\(x\\) is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to\n\\[\nF_X(x) \\ = \\ \\int_{-\\infty}^{x} p_X(u) \\, du,\n\\]\nFor discrete variables, the cdf is defined to include the upper bound given by the argument,\n\\[\nF_X(x) = \\sum_{u \\leq x} p_X(u).\n\\]\n\n\n\nThe complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by\n\\[\nF^C_X(x)\n\\ = \\ \\Pr[X &gt; x]\n\\ = \\ 1 - F_X(x).\n\\]\nUnlike the cdf, the ccdf is exclusive of the bound, hence the event \\(X &gt; x\\) rather than the cdf’s event \\(X \\leq x\\).\nFor continuous distributions, the ccdf works out to\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\int_{-\\infty}^x p_X(u) \\, du\n\\ = \\ \\int_x^{\\infty} p_X(u) \\, du.\n\\]\nThe lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over \\(u &gt; x\\),\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\sum_{u \\leq x} p_X(u)\n\\ = \\ \\sum_{u &gt; x} p_X(u).\n\\]\nCumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by \\[\n\\int_a^b p(u) \\, du = F_X(b) - F_X(a).\n\\] This allows truncated distributions to be defined as \\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a)}.\n\\]\nFor discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from \\(F_X(b) - F_X(a)\\),\n\\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a) + p_X(a)}.\n\\]\n\n\n\nStan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to \\([-0.5, 2.1]\\) can be coded with the following distribution statement.\ny ~ normal(0, 1) T[-0.5, 2.1];\nTruncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.\nIn general, the truncation bounds and parameters may be parameters or local variables.\nBecause the example above involves a continuous distribution, it behaves the same way as the following more verbose form.\ny ~ normal(0, 1);\nif (y &lt; -0.5 || y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -log_diff_exp(normal_lcdf(2.1 | 0, 1),\n                          normal_lcdf(-0.5 | 0, 1));\n}\nBecause a Stan program defines a log density function, all calculations are on the log scale. The function normal_lcdf is the log of the cumulative normal distribution function and the function log_diff_exp(a, b) is a more arithmetically stable form of log(exp(a) - exp(b)).\nFor a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution\ny ~ poisson(3.7) T[2, 10];\nbehaves in the same way as the following code.\ny ~ poisson(3.7);\nif (y &lt; 2 || y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         log_diff_exp(poisson_lcdf(10 | 3.7),\n                                      poisson_lcdf(2 | 3.7)));\n}\nRecall that log_sum_exp(a, b) is just the arithmetically stable form of log(exp(a) + exp(b)).\n\n\n\nFor truncating with only a lower bound, the upper limit is left blank.\ny ~ normal(0, 1) T[-0.5, ];\nThis truncated distribution statement has the same behavior as the following code.\ny ~ normal(0, 1);\nif (y &lt; -0.5) {\n  target += negative_infinity();\n} else {\n  target += -normal_lccdf(-0.5 | 0, 1);\n}\nThe normal_lccdf function is the normal complementary cumulative distribution function.\nAs with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus\ny ~ poisson(3.7) T[2, ];\nbehaves the same way as\ny ~ poisson(3.7);\nif (y &lt; 2) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         poisson_lccdf(2 | 3.7));\n}\n\n\n\nTo truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement\ny ~ normal(0, 1) T[ , 2.1];\nproduces the same result as the following code.\ntarget += normal_lpdf(y | 0, 1);\nif (y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -normal_lcdf(2.1 | 0, 1);\n}\nWith only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement\ny ~ poisson(3.7) T[ , 10];\nbehaves the same way as the following code.\ny ~ poisson(3.7);\nif (y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -poisson_lcdf(10 | 3.7);\n}\n\n\n\nIn all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.\n\n\n\nFor continuous distributions, truncation points must be expressions of type int or real. For discrete distributions, truncation points must be expressions of type int.\n\n\n\nFor a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds \\(-\\infty\\) to the total log probability, which in turn results in the sample being rejected.\n\n\n\nVectorization of distribution functions with truncation is available if the underlying distribution, lcdf, and lccdf functions meet the required signatures.\nThe equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):\n\nIf the variate y is the only non-scalar, the result is the same as described in the above sections, but the lcdf/lccdf calculation is multiplied by size(y).\nIf the other arguments to the distribution are non-scalars, then the vectorized version of the lcdf/lccdf is used. These functions return the sum of their terms, so no multiplication by the size is needed.\nThe exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the log_diff_exp of two sums is not the same as the sum of the pairwise log_diff_exp operations.\n\nNote that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.\n\n\n\n\n\nSuppose N is a variable of type int, y is a one-dimensional array of type array[] real, and mu and sigma are variables of type real. Furthermore, suppose that n has not been defined as a variable. Then the following is a well-formed for-loop statement.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nThe loop variable is n, the loop bounds are the values in the range 1:N, and the body is the statement following the loop bounds.\n\n\nThe type of the loop variable is int. Unlike in C++ and similarly to R, this variable must not be declared explicitly.\nThe bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range L:H will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to L and less than or equal to H. For example, the loop for (n in 2:5) will cause the body of the for loop to be executed with n equal to 2, 3, 4, and 5, in order. The variable and bound for (n in   5:2) will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.\nThe scope of the loop variable is limited to the body of the loop.\n\n\n\nUnlike in BUGS, Stan allows variables to be reassigned. For example, the variable theta in the following program is reassigned in each iteration of the loop.\nfor (n in 1:N) {\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nSuch reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.4\nIn Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.\nfor (n in 1:N) {\n  y[n] ~ bernoulli(theta);\n  theta = inv_logit(alpha + x[n] * beta);\n}\nIn this program, theta is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.\nStan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.\ntotal = 0.0;\nfor (n in 1:N) {\n  total = total + x[n];\n}\nAfter the for loop is executed, the variable total will hold the sum of the elements in the array x. This example was purely pedagogical; it is easier and more efficient to write\ntotal = sum(x);\nA variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.\nfor (n in 1:100) {\n  y += y * epsilon;\n  epsilon = 0.5 * epsilon;\n  y += y * epsilon;\n}\n\n\n\n\nA second form of for loops allows iteration over elements of containers. If ys is an expression denoting a container (vector, row vector, matrix, or array) with elements of type T, then the following is a well-formed foreach statement.\nfor (y in ys) {\n  // ... do something with y ...\n}\nThe order in which elements of ys are visited is defined for container types as follows.\n\nvector, row_vector: elements visited in order, y is of type double\nmatrix: elements visited in column-major order, y is of type double\narray[] T: elements visited in order, y is of type T.\n\nConsequently, if ys is a two dimensional array array[,] real, y will be a one-dimensional array of real values (type array[] real). If ’ysis a matrix, thenywill be a real value (typereal`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,\narray[2, 3] real yss;\nfor (ys in yss) {\n  for (y in ys) {\n    // ... do something with y ...\n  }\n}\nwhereas a matrix can be looped over in one foreach statement\nmatrix[2, 3] yss;\nfor (y in yss) {\n   // ... do something with y...\n}\nIn both cases, the loop variable y is of type real.  The elements of the matrix are visited in column-major order (e.g.,y[1, 1],y[2, 1],y[1, 2], ...,y[2, 3]), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,y[1, 1],y[1, 2],y[1, 3],y[2, 1], ...,y[2, 3]`).\n\n\n\nStan supports full conditional statements using the same if-then-else syntax as C++. The general format is\nif (condition1)\n  statement1\nelse if (condition2)\n  statement2\n// ...\nelse if (conditionN-1)\n  statementN-1\nelse\n  statementN\nThere must be a single leading if clause, which may be followed by any number of else if clauses, all of which may be optionally followed by an else clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.\nThe entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.\n\n\n\nStan supports standard while loops using the same syntax as C++. The general format is as follows.\nwhile (condition)\n  body\nThe condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).\nEvaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.\nThe rest of the body of a while loop may be skipped using a continue. The loop will be exited with a break statement. See the section on continue and break statements for more details.\n\n\n\nJust as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.\n\n\nBlocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nTo put multiple statements inside the body of a for loop, a block is used, as in the following example.\nfor (n in 1:N) {\n  lambda[n] ~ gamma(alpha, beta);\n  y[n] ~ poisson(lambda[n]);\n}\nThe open curly bracket ({) is the first character of the block and the close curly bracket (}) is the last character.\nBecause whitespace is ignored in Stan, the following program will not compile.\nfor (n in 1:N)\n  y[n] ~ normal(mu, sigma);\n  z[n] ~ normal(mu, sigma); // ERROR!\nThe problem is that the body of the for loop is taken to be the statement directly following it, which is y[n] ~ normal(mu, sigma). This leaves the probability statement for z[n] hanging, as is clear from the following equivalent program.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nz[n] ~ normal(mu, sigma); // ERROR!\nNeither of these programs will compile. If the loop variable n was defined before the for loop, the for-loop declaration will raise an error. If the loop variable n was not defined before the for loop, then the use of the expression z[n] will raise an error.\n\n\n\nA for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.\nfor (n in 1:N) {\n  real theta;\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nThe local variable theta is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus theta is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable theta if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.\nfor (m in 1:M) {\n  real theta;\n  for (n in 1:N) {\n    real theta; // ERROR!\n    theta = inv_logit(alpha + x[m, n] * beta);\n    y[m, n] ~ bernoulli(theta);\n// ...\nThe compiler will flag the second declaration of theta with a message that it is already defined.\n\n\n\nLocal variables may not have constraints on their declaration. The only types that may be used are listed in the types table under “local”.\n\n\n\nA block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following\nfor (m in 1:M) {\n  {\n     int n = 2 * m;\n     sum += n;\n  }\n  for (n in 1:N) {\n    sum += x[m, n];\n  }\n}\nThe variable declaration int n; is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of n are unrelated.\n\n\n\n\nThe one-token statements continue and break may be used within loops to alter control flow; continue causes the next iteration of the loop to run immediately, whereas break terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both break and continue scope to the most deeply nested loop, but pass through non-loop statements.\nAlthough these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.\n\n\nWhen a break statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:\nwhile (1) {\n  if (n &lt; 0) {\n    break;\n  }\n  foo(n);\n  n = n - 1;\n}\nThe while~(1) loop is a “forever” loop, because 1 is the true value, so the test always succeeds. Within the loop, if the value of n is less than 0, the loop terminates, otherwise it executes foo(n) and then decrements n. The statement above does exactly the same thing as\nwhile (n &gt;= 0) {\n  foo(n);\n  n = n - 1;\n}\nThis case is simply illustrative of the behavior; it is not a case where a break simplifies the loop.\n\n\n\nThe continue statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array x,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &lt;= 0) {\n    continue;\n  }\n  sum += x[n];\n}\nWhen the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an x[n] less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &gt; 0) {\n    sum += x[n];\n  }\n}\nWhile the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.\n\n\n\nIf there is a loop nested within a loop, a break or continue statement only breaks out of the inner loop. So\nwhile (cond1) {\n  // ...\n  while (cond2) {\n    // ...\n    if (cond3) {\n      break;\n    }\n    // ...\n  }\n  // execution continues here after break\n  // ...\n}\nIf the break is triggered by cond3 being true, execution will continue after the nested loop.\nAs with break statements, continue statements go back to the top of the most deeply nested loop in which the continue appears.\nAlthough break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.\n\n\n\n\nStan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.\nfor (n in 1:N) { print(\"loop iteration: \", n); ... }\nThe print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression n, followed by a new line.\n\n\nThe text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.\nPrinting a simple real or int typed variable always prints the variable’s value.5\nFor array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as\n[1, 2, 3]\nand a \\(2 \\times 3\\)-matrix as\n[[1, 2, 3], [4, 5, 6]]\nComplex numbers print as pairs. For example, the pair of statements\ncomplex z = to_complex(1.2, -3.5);\nprint(z)\nwill print as (1.2,-3.5), with no space after the comma or within the parentheses.\nPrinting a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.\ntransformed data {\n  matrix[2, 2] u;\n  u[1, 1] = 1.0;   u[1, 2] = 4.0;\n  u[2, 1] = 9.0;   u[2, 2] = 16.0;\n  for (n in 1:2) {\n    print(\"u[\", n, \"] = \", u[n]);\n  }\n}\nThis print statement executes twice, printing the following two lines of output.\nu[1] = [1, 4]\nu[2] = [9, 16]\n\n\n\nThe input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.\n\n\n\nPrinting for a print statement happens every time it is executed. The transformed data block is executed once per chain, the transformed parameter and model blocks once per leapfrog step, and the generated quantities block once per iteration.\n\n\n\nString literals begin and end with a double quote character (\"). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.\nThe Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.\n\n\n\nBecause Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.\nIt is particularly useful to print the value of the target log density accumulator (through the target() function), as in the following example.\nvector[2] y;\ny[1] = 1;\nprint(\"log density before =\", target());\ny ~ normal(0,1);  // bug!  y[2] not defined\nprint(\"log density after =\", target());\nThe example has a bug in that y[2] is not defined before the vector y is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).\nNote that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.\n\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.\nLike the print statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.\nReject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable x’s value is negative.\nif (x &lt; 0) {\n  reject(\"x must not be negative; found x=\", x);\n}\n\n\nReject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the normal_lpdf function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.\nIn all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.\n\n\nRejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.\n\n\n\nRejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.\nReject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.\nFatal errors in other blocks may also be signaled by use of the fatal_error statement.\n\n\n\nRejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.\nIf the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.\n\n\n\n\nRejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;\n  real&lt;lower=a, upper=b&gt; theta;\n  // ...\n}\nmodel {\n  // **wrong** needs explicit truncation\n  theta ~ normal(0, 1);\n  // ...\n}\nThis program is wrong because its truncation bounds on theta depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.\n  theta ~ normal(0, 1) T[a, b];\nThe conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over \\([a ,b]\\); in these simple univariate cases, we can overcome that with the T[ , ] notation, which essentially divides by whatever the prior integrates to over \\([a, b]\\).\nThis problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.\n  if (theta &lt; a || theta &gt; b) {\n    reject(\"theta not in (a, b)\");\n  }\n  // still **wrong**, needs T[a,b]\n  theta ~ normal(0, 1);\nIn this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.\nEven if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.\n\n\n\n\nThe Stan fatal_error statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.\nLike the print or reject statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.\nThe fatal error may be used to signal an unrecoverable error in blocks where reject leads to the algorithm attempting to try again, such as the model block.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Pathfinder"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/pathfinder.html#using-pathfinder-for-initializing-mcmc",
-    "href": "reference-manual/pathfinder.html#using-pathfinder-for-initializing-mcmc",
-    "title": "Pathfinder",
+    "objectID": "reference-manual/statements.html#statement-block-contexts",
+    "href": "reference-manual/statements.html#statement-block-contexts",
+    "title": "Statements",
     "section": "",
-    "text": "If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-\\(\\hat{k}\\) for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g. uniform from -2 to 2 used by Stan by default). If Pareto-\\(\\hat{k}\\) is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.",
+    "text": "The data and parameters blocks do not allow statements of any kind because these blocks are solely used to declare the data variables for input and the parameter variables for sampling. All other blocks allow statements. In these blocks, both variable declarations and statements are allowed. All top-level variables in a block are considered block variables. See the blocks chapter for more information about the block structure of Stan programs.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Pathfinder"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html",
-    "href": "reference-manual/mcmc.html",
-    "title": "MCMC Sampling",
+    "objectID": "reference-manual/statements.html#assignment-statement.section",
+    "href": "reference-manual/statements.html#assignment-statement.section",
+    "title": "Statements",
     "section": "",
-    "text": "This chapter presents the two Markov chain Monte Carlo (MCMC) algorithms used in Stan, the Hamiltonian Monte Carlo (HMC) algorithm and its adaptive variant the no-U-turn sampler (NUTS), along with details of their implementation and configuration.\n\n\nHamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., Betancourt and Girolami (2013), Neal (2011) for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.\nThis section translates the presentation of HMC by Betancourt and Girolami (2013) into the notation of Gelman et al. (2013).\n\n\nThe goal of sampling is to draw from a density \\(p(\\theta)\\) for parameters \\(\\theta\\). This is typically a Bayesian posterior \\(p(\\theta|y)\\) given data \\(y\\), and in particular, a Bayesian posterior coded as a Stan program.\n\n\n\nHMC introduces auxiliary momentum variables \\(\\rho\\) and draws from a joint density\n\\[\np(\\rho, \\theta) = p(\\rho | \\theta) p(\\theta).\n\\]\nIn most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters \\(\\theta\\),\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\n\\(M\\) is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see Betancourt (2017) for details.\nBy default Stan sets \\(M^{-1}\\) equal to a diagonal estimate of the covariance computed during warmup.\n\n\n\nThe joint density \\(p(\\rho, \\theta)\\) defines a Hamiltonian\n\\[\n\\begin{array}{rcl}\nH(\\rho, \\theta) & = & - \\log p(\\rho, \\theta)\n\\\\[3pt]\n& = & - \\log p(\\rho | \\theta) - \\log p(\\theta).\n\\\\[3pt]\n& = & T(\\rho | \\theta) + V(\\theta),\n\\end{array}\n\\]\nwhere the term\n\\[\nT(\\rho | \\theta) = - \\log p(\\rho | \\theta)\n\\]\nis called the “kinetic energy” and the term\n\\[\nV(\\theta) = - \\log p(\\theta)\n\\]\nis called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.\n\n\n\nStarting from the current value of the parameters \\(\\theta\\), a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.\nFirst, a value for the momentum is drawn independently of the current parameter values,\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\nThus momentum does not persist across iterations.\nNext, the joint system \\((\\theta,\\rho)\\) made up of the current parameter values \\(\\theta\\) and new momentum \\(\\rho\\) is evolved via Hamilton’s equations,\n\\[\n\\begin{array}{rcccl}\n\\displaystyle\n\\frac{d\\theta}{dt}\n& = &\n\\displaystyle\n+ \\frac{\\partial H}{\\partial \\rho}\n& = &\n\\displaystyle\n+ \\frac{\\partial T}{\\partial \\rho}\n\\\\[12pt]\n\\displaystyle\n\\frac{d\\rho}{dt}\n& = &\n\\displaystyle\n- \\frac{\\partial H}{\\partial \\theta }\n& = &\n\\displaystyle\n- \\frac{\\partial T}{\\partial \\theta}\n- \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nWith the momentum density being independent of the target density, i.e., \\(p(\\rho | \\theta) = p(\\rho)\\), the first term in the momentum time derivative, \\({\\partial T} / {\\partial \\theta}\\) is zero, yielding the pair time derivatives\n\\[\n\\begin{array}{rcl}\n\\frac{d \\theta}{d t} & = & +\\frac{\\partial T}{\\partial \\rho}\n\\\\[2pt]\n\\frac{d \\rho}{d t} & = & -\\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\n\n\n\nThe last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.\nLike most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval \\(\\epsilon\\). The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values \\(\\theta\\) or previous momentum value.\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\] It then alternates half-step updates of the momentum and full-step updates of the position.\n\\[\n\\begin{array}{rcl}\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}\n\\\\[6pt]\n\\theta & \\leftarrow\n       & \\theta \\, + \\, \\epsilon \\, M^{-1} \\, \\rho\n\\\\[6pt]\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nBy applying \\(L\\) leapfrog steps, a total of \\(L \\, \\epsilon\\) time is simulated. The resulting state at the end of the simulation (\\(L\\) repetitions of the above three steps) will be denoted \\((\\rho^{*}, \\theta^{*})\\).\nThe leapfrog integrator’s error is on the order of \\(\\epsilon^3\\) per step and \\(\\epsilon^2\\) globally, where \\(\\epsilon\\) is the time interval (also known as the step size); Leimkuhler and Reich (2004) provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.\n\n\n\nIf the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal \\((\\rho^{*}, \\theta^{*})\\) generated by transitioning from \\((\\rho, \\theta)\\) is\n\\[\n\\min \\!\n\\left(\n1,\n\\ \\exp \\! \\left( H(\\rho, \\theta) - H(\\rho^{*}, \\theta^{*}) \\right)\n\\right).\n\\]\nIf the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.\n\n\n\nThe Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters \\(\\theta\\); in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter \\(\\theta\\) is updated using the leapfrog integrator with discretization time \\(\\epsilon\\) and number of steps \\(L\\) according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state \\((\\theta^{*}, \\rho^{*})\\) or keep the existing state.\n\n\n\n\nThe Hamiltonian Monte Carlo algorithm has three parameters which must be set,\n\ndiscretization time \\(\\epsilon\\),\nmetric \\(M\\), and\nnumber of steps taken \\(L\\).\n\nIn practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters Neal (2011), Hoffman and Gelman (2014).\nIf \\(\\epsilon\\) is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If \\(\\epsilon\\) is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.\nIf \\(L\\) is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If \\(L\\) is too large, the algorithm will do too much work on each iteration.\nIf the inverse metric \\(M^{-1}\\) is a poor estimate of the posterior covariance, the step size \\(\\epsilon\\) must be kept small to maintain arithmetic precision. This would lead to a large \\(L\\) to compensate.\n\n\nThe actual integration time is \\(L \\, \\epsilon\\), a function of number of steps. Some interfaces to Stan set an approximate integration time \\(t\\) and the discretization interval (step size) \\(\\epsilon\\). In these cases, the number of steps will be rounded down as\n\\[\nL = \\left\\lfloor \\frac{t}{\\epsilon} \\right\\rfloor.\n\\]\nand the actual integration time will still be \\(L \\, \\epsilon\\).\n\n\n\nStan is able to automatically optimize \\(\\epsilon\\) to match an acceptance-rate target, able to estimate \\(M\\) based on warmup sample iterations, and able to dynamically adapt \\(L\\) on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm Hoffman and Gelman (2014).\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\nWhen adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the warmup adaptation figure, with two fast intervals surrounding a series of growing slow intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.\nThe motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.\n\nIn the initial fast interval the chain is allowed to converge towards the typical set,1 with only parameters that can learn from local information adapted.\nAfter this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.\nLastly, the fast parameters are allowed to adapt to the final update of the slow parameters.\n\nThese intervals may be controlled through the following configuration parameters, all of which must be positive integers:\nAdaptation Parameters Table. The parameters controlling adaptation and their default values.\n\n\n\n\n\n\n\n\nparameter\ndescription\ndefault\n\n\n\n\ninitial buffer\nwidth of initial fast adaptation interval\n75\n\n\nterm buffer\nwidth of final fast adaptation interval\n50\n\n\nwindow\ninitial width of slow adaptation interval\n25\n\n\n\n\n\n\nStan’s HMC algorithms utilize dual averaging Nesterov (2009) to optimize the step size.2\nThis warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of Hoffman and Gelman (2014). In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman-Gelman:2014.\nThe full set of dual-averaging parameters are\nStep Size Adaptation Parameters Table The parameters controlling step size adaptation, with constraints and default values.\n\n\n\nparameter\ndescription\nconstraint\ndefault\n\n\n\n\ndelta\ntarget Metropolis acceptance rate\n[0, 1]\n0.8\n\n\ngamma\nadaptation regularization scale\n(0, infty)\n0.05\n\n\nkappa\nadaptation relaxation exponent\n(0, infty)\n0.75\n\n\nt_0\nadaptation iteration offset\n(0, infty)\n10\n\n\n\nBy setting the target acceptance parameter \\(\\delta\\) to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of \\(\\delta\\) will also allow some models that would otherwise get stuck to overcome their blockages.\n\n\n\nAll implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.\nSmall step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See Neal (2011) for further discussion of step-size jittering.\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nIf the metric is constant then the resulting implementation is known as Euclidean HMC. Stan allows a choice among three Euclidean HMC implementations,\n\na unit metric (diagonal matrix of ones),\na diagonal metric (diagonal matrix with positive diagonal entries), and\na dense metric (a dense, symmetric positive definite matrix)\n\nto be configured by the user.\nIf the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the warmup adaptation stages figure). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.\nIf the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.\nVariances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.\n\n\nThe metric can compensate for linear (i.e. global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.\nIn complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e. local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.\n\n\n\nThe metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.3\nOne of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.\n\n\n\nStatistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.\n\n\n\nMCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.\nThis only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.\nPoor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.\n\n\n\n\nThe no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., Roberts, Gelman, and Gilks (1997)) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see Hoffman and Gelman (2014).\nNUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either\n\nthe NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or\nthe depth of the completed tree hits the maximum depth allowed.\n\nRather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory Betancourt (2016b).4\nConfiguring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.\nBoth the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as treedepth__ and n_leapfrog__, respectively. Because the final subtree may only be partially constructed, these two will always satisfy\n\\[\n2^{\\mathrm{treedepth} - 1} - 1\n\\ &lt; \\\nN_{\\mathrm{leapfrog}}\n\\ \\le \\\n2^{\\mathrm{treedepth} } - 1.\n\\]\nTree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\n\n\n\n\nIn some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.\nFor example, to generate a sequence of \\(N\\) draws from a binomial with trials \\(K\\) and chance of success \\(\\theta\\), the following program suffices.\ndata {\n  real&lt;lower=0, upper=1&gt; theta;\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n}\ngenerated quantities {\n  array[N] int&lt;lower=0, upper=K&gt; y;\n  for (n in 1:N) {\n    y[n] = binomial_rng(K, theta);\n  }\n}\nFor this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.\nMost models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, theta in the example above could be declared as a parameter and initialized as a parameter.\n\n\n\nStan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the optimization algorithms chapter, and the diagnostics chapter.\n\n\nThe random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nStan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.\n\n\nTogether, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.\n\n\n\n\nThe initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.\n\n\nIf the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).\n\n\n\nIt is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.\n\n\n\nRandom initialization by default initializes the parameter values with values drawn at random from a \\(\\mathsf{Uniform}(-2, 2)\\) distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.\nBecause zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\n\n\n\n\n\nThe Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.\nIn Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. Leimkuhler and Reich (2004) provide details and error analysis for the leapfrog algorithm.\nA divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,5 the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. Betancourt (2016a) provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.\nThe Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.\nIf the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.\nThe primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.\n\n\nIn some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.\nBefore reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.",
+    "text": "An assignment statement consists of a variable (possibly multivariate with indexing information) and an expression. Executing an assignment statement evaluates the expression on the right-hand side and assigns it to the (indexed) variable on the left-hand side. An example of a simple assignment is as follows.\nn = 0;\nExecuting this statement assigns the value of the expression 0, which is the integer zero, to the variable n. For an assignment to be well formed, the type of the expression on the right-hand side should be compatible with the type of the (indexed) variable on the left-hand side. For the above example, because 0 is an expression of type int, the variable n must be declared as being of type int or of type real. If the variable is of type real, the integer zero is promoted to a floating-point zero and assigned to the variable. After the assignment statement executes, the variable n will have the value zero (either as an integer or a floating-point value, depending on its type).\nSyntactically, every assignment statement must be followed by a semicolon. Otherwise, whitespace between the tokens does not matter (the tokens here being the left-hand-side (indexed) variable, the assignment operator, the right-hand-side expression and the semicolon).\nBecause the right-hand side is evaluated first, it is possible to increment a variable in Stan just as in C++ and other programming languages by writing\nn = n + 1;\nSuch self assignments are not allowed in BUGS, because they induce a cycle into the directed graphical model.\nThe left-hand side of an assignment may contain indices for array, matrix, or vector data structures. For instance, if Sigma is of type matrix, then\nSigma[1, 1] = 1.0;\nsets the value in the first column of the first row of Sigma to one.\nAssignments to subcomponents of larger multi-variate data structures are supported by Stan. For example, a is an array of type array[,] real and b is an array of type array[] real, then the following two statements are both well-formed.\na[3] = b;\nb = a[4];\nSimilarly, if x is a variable declared to have type row_vector and Y is a variable declared as type matrix, then the following sequence of statements to swap the first two rows of Y is well formed.\nx = Y[1];\nY[1] = Y[2];\nY[2] = x;\n\n\nStan allows assignment of lower types to higher types, but not vice-versa. That is, we can assign an expression of type int to an lvalue of type real, and we can assign an expression of type real to an lvalue of type complex. Furthermore, promotion is transitive, so that we can assign an expression of type int to an lvalue of type complex.\nPromotion extends to containers, so that arrays of int can be promoted to arrays of real during assignment, and arrays of real can be assigned to an lvalue of type array of complex. Similarly, an expression of type vector may be assigned to an lvalue of type complex_vector, and similarly for row vectors and matrices.\n\n\n\nThe expressions that are legal left-hand sides of assignment statements are known as “lvalues.” In Stan, there are three kinds of legal lvalues,\n\na variable, or\na variable with one or more indices, or\na comma separated list of lvalues surrounded by ( and )\n\nTo be used as an lvalue, an indexed variable must have at least as many dimensions as the number of indices provided. An array of real or integer types has as many dimensions as it is declared for. A matrix has two dimensions and a vector or row vector one dimension; this also holds for the constrained types, covariance and correlation matrices and their Cholesky factors and ordered, positive ordered, and simplex vectors. An array of matrices has two more dimensions than the array and an array of vectors or row vectors has one more dimension than the array. Note that the number of indices can be less than the number of dimensions of the variable, meaning that the right hand side must itself be multidimensional to match the remaining dimensions.\n\n\n\nMultiple indexes, as described in the multi-indexing section, are also permitted on the left-hand side of assignments. Indexing on the left side works exactly as it does for expressions, with multiple indexes preserving index positions and single indexes reducing them. The type on the left side must still match the type on the right side.\n\n\nAll assignment is carried out as if the right-hand side is copied before the assignment. This resolves any potential aliasing issues arising from he right-hand side changing in the middle of an assignment statement’s execution.\n\n\n\n\nStan’s arithmetic operators may be used in compound arithmetic and assignment operations. For example, consider the following example of compound addition and assignment.\nreal x = 5;\nx += 7;  // value of x is now 12\nThe compound arithmetic and assignment statement above is equivalent to the following long form.\nx = x + 7;\nIn general, the compound form\nx op= y\nwill be equivalent to\nx = x op y;\nThe compound statement will be legal whenever the long form is legal. This requires that the operation x op y must itself be well formed and that the result of the operation be assignable to x. For the expression x to be assignable, it must be an indexed variable where the variable is defined in the current block. For example, the following compound addition and assignment statement will increment a single element of a vector by two.\nvector[N] x;\nx[3] += 2;\nAs a further example, consider\nmatrix[M, M] x;\nvector[M] y;\nreal z;\nx *= x;  // OK, (x * x) is a matrix\nx *= z;  // OK, (x * z) is a matrix\nx *= y;  // BAD, (x * y) is a vector\nThe supported compound arithmetic and assignment operations are listed in the compound arithmetic/assignment table; they are also listed in the index prefaced by operator, e.g., operator+=.\nCompound Arithmetic/Assignment Table.  Stan allows compound arithmetic and assignment statements of the forms listed in the table. The compound form is legal whenever the corresponding long form would be legal and it has the same effect.\n\n\n\noperation\ncompound\nunfolded\n\n\n\n\naddition\nx += y\nx = x + y\n\n\nsubtraction\nx -= y\nx = x - y\n\n\nmultiplication\nx *= y\nx = x * y\n\n\ndivision\nx /= y\nx = x / y\n\n\nelementwise multiplication\nx .*= y\nx = x .* y\n\n\nelementwise division\nx ./= y\nx = x ./ y",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html#hamiltonian-monte-carlo",
-    "href": "reference-manual/mcmc.html#hamiltonian-monte-carlo",
-    "title": "MCMC Sampling",
+    "objectID": "reference-manual/statements.html#increment-log-prob.section",
+    "href": "reference-manual/statements.html#increment-log-prob.section",
+    "title": "Statements",
     "section": "",
-    "text": "Hamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., Betancourt and Girolami (2013), Neal (2011) for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.\nThis section translates the presentation of HMC by Betancourt and Girolami (2013) into the notation of Gelman et al. (2013).\n\n\nThe goal of sampling is to draw from a density \\(p(\\theta)\\) for parameters \\(\\theta\\). This is typically a Bayesian posterior \\(p(\\theta|y)\\) given data \\(y\\), and in particular, a Bayesian posterior coded as a Stan program.\n\n\n\nHMC introduces auxiliary momentum variables \\(\\rho\\) and draws from a joint density\n\\[\np(\\rho, \\theta) = p(\\rho | \\theta) p(\\theta).\n\\]\nIn most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters \\(\\theta\\),\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\n\\(M\\) is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see Betancourt (2017) for details.\nBy default Stan sets \\(M^{-1}\\) equal to a diagonal estimate of the covariance computed during warmup.\n\n\n\nThe joint density \\(p(\\rho, \\theta)\\) defines a Hamiltonian\n\\[\n\\begin{array}{rcl}\nH(\\rho, \\theta) & = & - \\log p(\\rho, \\theta)\n\\\\[3pt]\n& = & - \\log p(\\rho | \\theta) - \\log p(\\theta).\n\\\\[3pt]\n& = & T(\\rho | \\theta) + V(\\theta),\n\\end{array}\n\\]\nwhere the term\n\\[\nT(\\rho | \\theta) = - \\log p(\\rho | \\theta)\n\\]\nis called the “kinetic energy” and the term\n\\[\nV(\\theta) = - \\log p(\\theta)\n\\]\nis called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.\n\n\n\nStarting from the current value of the parameters \\(\\theta\\), a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.\nFirst, a value for the momentum is drawn independently of the current parameter values,\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\nThus momentum does not persist across iterations.\nNext, the joint system \\((\\theta,\\rho)\\) made up of the current parameter values \\(\\theta\\) and new momentum \\(\\rho\\) is evolved via Hamilton’s equations,\n\\[\n\\begin{array}{rcccl}\n\\displaystyle\n\\frac{d\\theta}{dt}\n& = &\n\\displaystyle\n+ \\frac{\\partial H}{\\partial \\rho}\n& = &\n\\displaystyle\n+ \\frac{\\partial T}{\\partial \\rho}\n\\\\[12pt]\n\\displaystyle\n\\frac{d\\rho}{dt}\n& = &\n\\displaystyle\n- \\frac{\\partial H}{\\partial \\theta }\n& = &\n\\displaystyle\n- \\frac{\\partial T}{\\partial \\theta}\n- \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nWith the momentum density being independent of the target density, i.e., \\(p(\\rho | \\theta) = p(\\rho)\\), the first term in the momentum time derivative, \\({\\partial T} / {\\partial \\theta}\\) is zero, yielding the pair time derivatives\n\\[\n\\begin{array}{rcl}\n\\frac{d \\theta}{d t} & = & +\\frac{\\partial T}{\\partial \\rho}\n\\\\[2pt]\n\\frac{d \\rho}{d t} & = & -\\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\n\n\n\nThe last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.\nLike most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval \\(\\epsilon\\). The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values \\(\\theta\\) or previous momentum value.\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\] It then alternates half-step updates of the momentum and full-step updates of the position.\n\\[\n\\begin{array}{rcl}\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}\n\\\\[6pt]\n\\theta & \\leftarrow\n       & \\theta \\, + \\, \\epsilon \\, M^{-1} \\, \\rho\n\\\\[6pt]\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nBy applying \\(L\\) leapfrog steps, a total of \\(L \\, \\epsilon\\) time is simulated. The resulting state at the end of the simulation (\\(L\\) repetitions of the above three steps) will be denoted \\((\\rho^{*}, \\theta^{*})\\).\nThe leapfrog integrator’s error is on the order of \\(\\epsilon^3\\) per step and \\(\\epsilon^2\\) globally, where \\(\\epsilon\\) is the time interval (also known as the step size); Leimkuhler and Reich (2004) provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.\n\n\n\nIf the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal \\((\\rho^{*}, \\theta^{*})\\) generated by transitioning from \\((\\rho, \\theta)\\) is\n\\[\n\\min \\!\n\\left(\n1,\n\\ \\exp \\! \\left( H(\\rho, \\theta) - H(\\rho^{*}, \\theta^{*}) \\right)\n\\right).\n\\]\nIf the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.\n\n\n\nThe Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters \\(\\theta\\); in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter \\(\\theta\\) is updated using the leapfrog integrator with discretization time \\(\\epsilon\\) and number of steps \\(L\\) according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state \\((\\theta^{*}, \\rho^{*})\\) or keep the existing state.",
+    "text": "The basis of Stan’s execution is the evaluation of a log probability function (specifically, a log probability density function) for a given set of (real-valued) parameters; this function returns the log density of the posterior up to an additive constant. Data and transformed data are fixed before the log density is evaluated. The total log probability is initialized to zero. Next, any log Jacobian adjustments accrued by the variable constraints are added to the log density (the Jacobian adjustment may be skipped for maximum likelihood estimation via optimization). Distribution statements and log probability increment statements may add to the log density in the model block. A log probability increment statement directly increments the log density with the value of an expression as follows.1\ntarget += -0.5 * y * y;\nThe keyword target here is actually not a variable, and may not be accessed as such (though see below on how to access the value of target through a special function).\nIn this example, the unnormalized log probability of a unit normal variable \\(y\\) is added to the total log probability. In the general case, the argument can be any expression.2\nAn entire Stan model can be implemented this way. For instance, the following model has a single variable according to a unit normal probability.\nparameters {\n  real y;\n}\nmodel {\n  target += -0.5 * y * y;\n}\nThis model defines a log probability function\n\\[\n\\log p(y) = - \\, \\frac{y^2}{2} - \\log Z\n\\]\nwhere \\(Z\\) is a normalizing constant that does not depend on \\(y\\). The constant \\(Z\\) is conventionally written this way because on the linear scale, \\[\np(y) = \\frac{1}{Z} \\exp\\left(-\\frac{y^2}{2}\\right).\n\\] which is typically written without reference to \\(Z\\) as \\[\np(y) \\propto \\exp\\left(-\\frac{y^2}{2}\\right).\n\\]\nStan only requires models to be defined up to a constant that does not depend on the parameters. This is convenient because often the normalizing constant \\(Z\\) is either time-consuming to compute or intractable to evaluate.\n\n\nThe built in distribution functions in Stan are all available in normalized and unnormalized form. The normalized forms include all of the terms in the log density, and the unnormalized forms drop terms which are not directly or indirectly a function of the model parameters.\nFor instance, the normal_lpdf function returns the log density of a normal distribution:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nThe normal_lupdf function returns the log density of an unnormalized distribution. With the unnormalized version of the function, Stan does not define what the normalization constant will be, though usually as many terms as possible are dropped to make the calculation fast. Dropping a constant sigma term, normal_lupdf would be equivalent to:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2\n\\]\nAll functions ending in _lpdf have a corresponding _lupdf version which evaluates and returns the unnormalized density. The same is true for _lpmf and _lupmf.\n\n\n\nThe increment log density statement looks syntactically like compound addition and assignment (see the compound arithmetic/assignment section, it is treated as a primitive statement because target is not itself a variable. So, even though\ntarget += lp;\nis a legal statement, the corresponding long form is not legal.\ntarget = target + lp;  // BAD, target is not a variable\n\n\n\nThe target += ... statement accepts an argument in place of ... for any expression type, including integers, reals, vectors, row vectors, matrices, and arrays of any dimensionality, including arrays of vectors and matrices. For container arguments, their sum will be added to the total log density.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html#hmc-algorithm-parameters",
-    "href": "reference-manual/mcmc.html#hmc-algorithm-parameters",
-    "title": "MCMC Sampling",
+    "objectID": "reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment",
+    "href": "reference-manual/statements.html#increment-log-density-with-a-change-of-variables-adjustment",
+    "title": "Statements",
     "section": "",
-    "text": "The Hamiltonian Monte Carlo algorithm has three parameters which must be set,\n\ndiscretization time \\(\\epsilon\\),\nmetric \\(M\\), and\nnumber of steps taken \\(L\\).\n\nIn practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters Neal (2011), Hoffman and Gelman (2014).\nIf \\(\\epsilon\\) is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If \\(\\epsilon\\) is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.\nIf \\(L\\) is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If \\(L\\) is too large, the algorithm will do too much work on each iteration.\nIf the inverse metric \\(M^{-1}\\) is a poor estimate of the posterior covariance, the step size \\(\\epsilon\\) must be kept small to maintain arithmetic precision. This would lead to a large \\(L\\) to compensate.\n\n\nThe actual integration time is \\(L \\, \\epsilon\\), a function of number of steps. Some interfaces to Stan set an approximate integration time \\(t\\) and the discretization interval (step size) \\(\\epsilon\\). In these cases, the number of steps will be rounded down as\n\\[\nL = \\left\\lfloor \\frac{t}{\\epsilon} \\right\\rfloor.\n\\]\nand the actual integration time will still be \\(L \\, \\epsilon\\).\n\n\n\nStan is able to automatically optimize \\(\\epsilon\\) to match an acceptance-rate target, able to estimate \\(M\\) based on warmup sample iterations, and able to dynamically adapt \\(L\\) on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm Hoffman and Gelman (2014).\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\nWhen adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the warmup adaptation figure, with two fast intervals surrounding a series of growing slow intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.\nThe motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.\n\nIn the initial fast interval the chain is allowed to converge towards the typical set,1 with only parameters that can learn from local information adapted.\nAfter this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.\nLastly, the fast parameters are allowed to adapt to the final update of the slow parameters.\n\nThese intervals may be controlled through the following configuration parameters, all of which must be positive integers:\nAdaptation Parameters Table. The parameters controlling adaptation and their default values.\n\n\n\n\n\n\n\n\nparameter\ndescription\ndefault\n\n\n\n\ninitial buffer\nwidth of initial fast adaptation interval\n75\n\n\nterm buffer\nwidth of final fast adaptation interval\n50\n\n\nwindow\ninitial width of slow adaptation interval\n25\n\n\n\n\n\n\nStan’s HMC algorithms utilize dual averaging Nesterov (2009) to optimize the step size.2\nThis warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of Hoffman and Gelman (2014). In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman-Gelman:2014.\nThe full set of dual-averaging parameters are\nStep Size Adaptation Parameters Table The parameters controlling step size adaptation, with constraints and default values.\n\n\n\nparameter\ndescription\nconstraint\ndefault\n\n\n\n\ndelta\ntarget Metropolis acceptance rate\n[0, 1]\n0.8\n\n\ngamma\nadaptation regularization scale\n(0, infty)\n0.05\n\n\nkappa\nadaptation relaxation exponent\n(0, infty)\n0.75\n\n\nt_0\nadaptation iteration offset\n(0, infty)\n10\n\n\n\nBy setting the target acceptance parameter \\(\\delta\\) to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of \\(\\delta\\) will also allow some models that would otherwise get stuck to overcome their blockages.\n\n\n\nAll implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.\nSmall step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See Neal (2011) for further discussion of step-size jittering.\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nIf the metric is constant then the resulting implementation is known as Euclidean HMC. Stan allows a choice among three Euclidean HMC implementations,\n\na unit metric (diagonal matrix of ones),\na diagonal metric (diagonal matrix with positive diagonal entries), and\na dense metric (a dense, symmetric positive definite matrix)\n\nto be configured by the user.\nIf the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the warmup adaptation stages figure). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.\nIf the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.\nVariances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.\n\n\nThe metric can compensate for linear (i.e. global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.\nIn complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e. local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.\n\n\n\nThe metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.3\nOne of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.\n\n\n\nStatistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.\n\n\n\nMCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.\nThis only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.\nPoor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.\n\n\n\n\nThe no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., Roberts, Gelman, and Gilks (1997)) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see Hoffman and Gelman (2014).\nNUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either\n\nthe NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or\nthe depth of the completed tree hits the maximum depth allowed.\n\nRather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory Betancourt (2016b).4\nConfiguring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.\nBoth the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as treedepth__ and n_leapfrog__, respectively. Because the final subtree may only be partially constructed, these two will always satisfy\n\\[\n2^{\\mathrm{treedepth} - 1} - 1\n\\ &lt; \\\nN_{\\mathrm{leapfrog}}\n\\ \\le \\\n2^{\\mathrm{treedepth} } - 1.\n\\]\nTree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.",
+    "text": "A variant of the target += statement described above is the jacobian += statement. This can be used in the transformed parameters block or in functions ending with _jacobian to mimic the log Jacobian adjustments accrued by built-in variable transforms.\nSimilarly to those implemented for the built-in transforms, these Jacobian adjustment may be skipped for maximum likelihood estimation via optimization.\nFor example, here is a program which recreates the existing &lt;upper=x&gt; transform on real numbers:\nfunctions {\n  real my_upper_bound_jacobian(real x, real ub) {\n    jacobian += x;\n    return ub - exp(x);\n  }\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  // use b as if it was declared `real&lt;upper=ub&gt; b;` in parameters\n  // e.g.\n  // b ~ lognormal(0, 1);\n}\n\n\nTo access the accumulated log density up to the current execution point, the function target() may be used.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html#sampling-without-parameters",
-    "href": "reference-manual/mcmc.html#sampling-without-parameters",
-    "title": "MCMC Sampling",
+    "objectID": "reference-manual/statements.html#sampling-statements.section",
+    "href": "reference-manual/statements.html#sampling-statements.section",
+    "title": "Statements",
     "section": "",
-    "text": "In some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.\nFor example, to generate a sequence of \\(N\\) draws from a binomial with trials \\(K\\) and chance of success \\(\\theta\\), the following program suffices.\ndata {\n  real&lt;lower=0, upper=1&gt; theta;\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n}\ngenerated quantities {\n  array[N] int&lt;lower=0, upper=K&gt; y;\n  for (n in 1:N) {\n    y[n] = binomial_rng(K, theta);\n  }\n}\nFor this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.\nMost models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, theta in the example above could be declared as a parameter and initialized as a parameter.",
+    "text": "The term “sampling statement” has been replaced with distribution statement.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html#general-config.section",
-    "href": "reference-manual/mcmc.html#general-config.section",
-    "title": "MCMC Sampling",
+    "objectID": "reference-manual/statements.html#distribution-statements.section",
+    "href": "reference-manual/statements.html#distribution-statements.section",
+    "title": "Statements",
     "section": "",
-    "text": "Stan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the optimization algorithms chapter, and the diagnostics chapter.\n\n\nThe random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nStan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.\n\n\nTogether, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.\n\n\n\n\nThe initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.\n\n\nIf the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).\n\n\n\nIt is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.\n\n\n\nRandom initialization by default initializes the parameter values with values drawn at random from a \\(\\mathsf{Uniform}(-2, 2)\\) distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.\nBecause zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).",
+    "text": "Stan supports writing probability statements also using distribution statements, for example\ny ~ normal(mu, sigma);\nmu ~ normal(0, 10);\nsigma ~ normal(0, 1);\nThe symbol \\(\\sim\\) is called tilde. Due to historical reasons, the distribution statements used to be called “sampling statements” in Stan, but that term is not recommended anymore as it is a less accurate description.\nIn general, we can read \\(\\sim\\) as “is distributed as,” and overall this notation is used as a shorthand for defining distributions, so that the above example can be written also as \\[\n\\begin{aligned}\n   p(y| \\mu, \\sigma) & = \\mathrm{normal}(y |  \\mu, \\sigma)\\\\\n   p(\\mu) & = \\mathrm{normal}(\\mu |  0, 10)\\\\\n   p(\\sigma) & = \\mathrm{normal}^+(\\sigma |  0, 1).\n\\end{aligned}\n\\] A collection of distribution statements define a joint distribution as the product of component distributions \\[\np(y,\\mu,\\sigma) = p(y| \\mu, \\sigma )p(\\mu) p(\\sigma).\n\\]\nThis works even if the model is not constructed generatively. For example, suppose you include the following code in a Stan model:\n  a ~ normal(0, 1);\n  a ~ normal(0, 1);\nThis is translated to \\[\n    p(a) = \\mathrm{normal}(a | 0, 1)\\mathrm{normal}(a |  0, 1),\n\\] which in this case is \\(\\mathrm{normal}(a|0,1/\\sqrt{2})\\). One might expect that the above two lines of code would represent a redundant expression of a \\(\\mathrm{normal}(a|0,1)\\) prior, but, no, each line of code corresponds to an additional term in the target, or log posterior density. You can think of each line as representing an additional piece of information.\nWhen the joint distribution is considered as a function of parameters (e.g. \\(\\mu\\), \\(\\sigma\\)) given fixed data, it is proportional to the posterior distribution. In general, the posterior distribution is not a normalized probability density function—that is, it will be positive but will not in general integrate to 1—but the proportionality is sufficient for the Stan algorithms.\nStan always constructs the target function—in Bayesian terms, the log posterior density function of the parameter vector—by adding terms in the model block. Equivalently, each \\(\\sim\\) statement corresponds to a multiplicative factor in the unnormalized posterior density.\nDistribution statements (~) accept only built-in or user-defined distributions on the right side. The left side of a distribution statement may be data, parameter, or a complex expression, but the evaluated type needs to match one of the allowed types of the distribution on the right (see more below).\nIn Stan, a distribution statement is merely a notational convenience following the typical notation used to present models in the literature. The above model defined with distribution statements could be expressed as a direct increment on the total log probability density as\ntarget += normal_lpdf(y | mu, sigma);\ntarget += normal_lpdf(mu | 0, 10);\ntarget += normal_lpdf(sigma | 0, 1);\nStan models can mix distribution statements and log probability increment statements. Although statistical models are usually defined with distributions in the literature, there are several scenarios in which we may want to code the log likelihood or parts of it directly, for example, due to computational efficiency (e.g. censored data model) or coding language limitations (e.g. mixture models in Stan). This is possible with log probability increment statements. See also the discussion below about Jacobians.\nIn general, a distribution statement of the form\ny ~ dist(theta1, ..., thetaN);\ninvolving subexpressions y and theta1 through thetaN (including the case where N is zero) will be well formed if and only if the corresponding log probability increment statement is well-formed. For densities allowing real y values, the log probability density function is used,\ntarget += dist_lpdf(y | theta1, ..., thetaN);\nFor those restricted to integer y values, the log probability mass function is used,\ntarget += dist_lpmf(y | theta1, ..., thetaN);\nThis will be well formed if and only if dist_lpdf(y | theta1,   ..., thetaN) or dist_lpmf(y | theta1, ..., thetaN) is a well-formed expression of type real. User defined distributions can be defined in functions block by using function names ending with _lpdf.\n\n\nAlthough both lead to the same inference algorithm behavior in Stan, there is one critical difference between using the distribution statement, as in\ny ~ normal(mu, sigma);\nand explicitly incrementing the log probability function, as in\ntarget += normal_lpdf(y | mu, sigma);\nThe distribution statement drops all the terms in the log probability function that are constant, whereas the explicit call to normal_lpdf adds all of the terms in the definition of the log normal probability function, including all of the constant normalizing terms. Therefore, the explicit increment form can be used to recreate the exact log probability values for the model. Otherwise, the distribution statement form will be faster if any of the input expressions, y, mu, or sigma, involve only constants, data variables, and transformed data variables. See the section Built in distributions above discussing _lupdf and _lupmf functions that also drops all the constant terms.\n\n\n\nThe left-hand side of a distribution statement may be an arbitrary expression (of compatible type)“. For instance, it is legal syntactically to write\nparameters {\n  real&lt;lower=0&gt; beta;\n}\n// ...\nmodel {\n  log(beta) ~ normal(mu, sigma);\n}\nUnfortunately, this is not enough to properly model beta as having a lognormal distribution. Whenever a nonlinear transform is applied to a parameter, such as the logarithm function being applied to beta here, and then used on the left-hand side of a distribution statement or on the left of a vertical bar in a log pdf function, an adjustment must be made to account for the differential change in scale and ensure beta gets the correct distribution. The correction required is to add the log Jacobian of the transform to the target log density; see the change of variables section for full definitions. For the case above, the following adjustment will account for the log transform.3\ntarget += - log(abs(y));\n\n\n\nStan supports truncating distributions with lower bounds, upper bounds, or both.\n\n\nA probability density function \\(p(x)\\) for a continuous distribution may be truncated to an interval \\([a, b]\\) to define a new density \\(p_{[a, b]}(x)\\) with support \\([a, b]\\) by setting\n\\[\np_{[a, b]}(x)\n= \\frac{p(x)}\n       {\\int_a^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) for a discrete distribution may be truncated to the closed interval \\([a, b]\\) by\n\\[\np_{[a, b]}(x) = \\frac{p(x)}\n                  {\\sum_{u = a}^b p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x)\n= \\frac{p(x)}\n       {\\int_a^{\\infty} p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([a, \\infty]\\) by defining\n\\[\np_{[a, \\infty]}(x) = \\frac{p(x)}\n                  {\\sum_{a &lt;= u} p(u)}.\n\\]\n\n\n\nA probability density function \\(p(x)\\) can be truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty, b]}(x)\n= \\frac{p(x)}\n       {\\int_{-\\infty}^b p(u) \\, du}.\n\\]\nA probability mass function \\(p(x)\\) is truncated to \\([-\\infty, b]\\) by defining\n\\[\np_{[-\\infty,b]}(x) = \\frac{p(x)}\n                  {\\sum_{u &lt;= b} p(u)}.\n\\]\n\n\n\nGiven a probability function \\(p_X(x)\\) for a random variable \\(X\\), its cumulative distribution function (cdf) \\(F_X(x)\\) is defined to be the probability that \\(X \\leq x\\),\n\\[\nF_X(x) = \\Pr[X \\leq x].\n\\]\nThe upper-case variable \\(X\\) is the random variable whereas the lower-case variable \\(x\\) is just an ordinary bound variable. For continuous random variables, the definition of the cdf works out to\n\\[\nF_X(x) \\ = \\ \\int_{-\\infty}^{x} p_X(u) \\, du,\n\\]\nFor discrete variables, the cdf is defined to include the upper bound given by the argument,\n\\[\nF_X(x) = \\sum_{u \\leq x} p_X(u).\n\\]\n\n\n\nThe complementary cumulative distribution function (ccdf) in both the continuous and discrete cases is given by\n\\[\nF^C_X(x)\n\\ = \\ \\Pr[X &gt; x]\n\\ = \\ 1 - F_X(x).\n\\]\nUnlike the cdf, the ccdf is exclusive of the bound, hence the event \\(X &gt; x\\) rather than the cdf’s event \\(X \\leq x\\).\nFor continuous distributions, the ccdf works out to\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\int_{-\\infty}^x p_X(u) \\, du\n\\ = \\ \\int_x^{\\infty} p_X(u) \\, du.\n\\]\nThe lower boundary can be included in the integration bounds because it is a single point on a line and hence has no probability mass. For the discrete case, the lower bound must be excluded in the summation explicitly by summing over \\(u &gt; x\\),\n\\[\nF^C_X(x)\n\\ = \\ 1 - \\sum_{u \\leq x} p_X(u)\n\\ = \\ \\sum_{u &gt; x} p_X(u).\n\\]\nCumulative distribution functions provide the necessary integral calculations to define truncated distributions. For truncation with lower and upper bounds, the denominator is defined by \\[\n\\int_a^b p(u) \\, du = F_X(b) - F_X(a).\n\\] This allows truncated distributions to be defined as \\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a)}.\n\\]\nFor discrete distributions, a slightly more complicated form is required to explicitly insert the lower truncation point, which is otherwise excluded from \\(F_X(b) - F_X(a)\\),\n\\[\np_{[a,b]}(x) = \\frac{p_X(x)}\n                  {F_X(b) - F_X(a) + p_X(a)}.\n\\]\n\n\n\nStan allows probability functions to be truncated. For example, a truncated unit normal distributions restricted to \\([-0.5, 2.1]\\) can be coded with the following distribution statement.\ny ~ normal(0, 1) T[-0.5, 2.1];\nTruncated distributions are translated as an additional term in the accumulated log density function plus error checking to make sure the variate in the distribution statement is within the bounds of the truncation.\nIn general, the truncation bounds and parameters may be parameters or local variables.\nBecause the example above involves a continuous distribution, it behaves the same way as the following more verbose form.\ny ~ normal(0, 1);\nif (y &lt; -0.5 || y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -log_diff_exp(normal_lcdf(2.1 | 0, 1),\n                          normal_lcdf(-0.5 | 0, 1));\n}\nBecause a Stan program defines a log density function, all calculations are on the log scale. The function normal_lcdf is the log of the cumulative normal distribution function and the function log_diff_exp(a, b) is a more arithmetically stable form of log(exp(a) - exp(b)).\nFor a discrete distribution, another term is necessary in the denominator to account for the excluded boundary. The truncated discrete distribution\ny ~ poisson(3.7) T[2, 10];\nbehaves in the same way as the following code.\ny ~ poisson(3.7);\nif (y &lt; 2 || y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         log_diff_exp(poisson_lcdf(10 | 3.7),\n                                      poisson_lcdf(2 | 3.7)));\n}\nRecall that log_sum_exp(a, b) is just the arithmetically stable form of log(exp(a) + exp(b)).\n\n\n\nFor truncating with only a lower bound, the upper limit is left blank.\ny ~ normal(0, 1) T[-0.5, ];\nThis truncated distribution statement has the same behavior as the following code.\ny ~ normal(0, 1);\nif (y &lt; -0.5) {\n  target += negative_infinity();\n} else {\n  target += -normal_lccdf(-0.5 | 0, 1);\n}\nThe normal_lccdf function is the normal complementary cumulative distribution function.\nAs with lower and upper truncation, the discrete case requires a more complicated denominator to add back in the probability mass for the lower bound. Thus\ny ~ poisson(3.7) T[2, ];\nbehaves the same way as\ny ~ poisson(3.7);\nif (y &lt; 2) {\n  target += negative_infinity();\n} else {\n  target += -log_sum_exp(poisson_lpmf(2 | 3.7),\n                         poisson_lccdf(2 | 3.7));\n}\n\n\n\nTo truncate with only an upper bound, the lower bound is left blank. The upper truncated distribution statement\ny ~ normal(0, 1) T[ , 2.1];\nproduces the same result as the following code.\ntarget += normal_lpdf(y | 0, 1);\nif (y &gt; 2.1) {\n  target += negative_infinity();\n} else {\n  target += -normal_lcdf(2.1 | 0, 1);\n}\nWith only an upper bound, the discrete case does not need a boundary adjustment. The upper-truncated distribution statement\ny ~ poisson(3.7) T[ , 10];\nbehaves the same way as the following code.\ny ~ poisson(3.7);\nif (y &gt; 10) {\n  target += negative_infinity();\n} else {\n  target += -poisson_lcdf(10 | 3.7);\n}\n\n\n\nIn all cases, the truncation is only well formed if the appropriate log density or mass function and necessary log cumulative distribution functions are defined. Not every distribution built into Stan has log cdf and log ccdfs defined, nor will every user-defined distribution. The discrete probability function documentations describes the available discrete and continuous cumulative distribution functions; most univariate distributions have log cdf and log ccdf functions.\n\n\n\nFor continuous distributions, truncation points must be expressions of type int or real. For discrete distributions, truncation points must be expressions of type int.\n\n\n\nFor a truncated distribution statement, if the value sampled is not within the bounds specified by the truncation expression, the result is zero probability and the entire statement adds \\(-\\infty\\) to the total log probability, which in turn results in the sample being rejected.\n\n\n\nVectorization of distribution functions with truncation is available if the underlying distribution, lcdf, and lccdf functions meet the required signatures.\nThe equivalent code for a vectorized truncation depends on which of the variables are non-scalars (arrays, vectors, etc.):\n\nIf the variate y is the only non-scalar, the result is the same as described in the above sections, but the lcdf/lccdf calculation is multiplied by size(y).\nIf the other arguments to the distribution are non-scalars, then the vectorized version of the lcdf/lccdf is used. These functions return the sum of their terms, so no multiplication by the size is needed.\nThe exception to the above is when a non-variate is a vector and both a lower and upper bound are specified in the truncation. In this case, a for loop is generated over the elements of the non-scalar arguments. This is required since the log_diff_exp of two sums is not the same as the sum of the pairwise log_diff_exp operations.\n\nNote that while a lower-and-upper truncated distribution may generate a for-loop internally as part of translating the truncation statement, this is still preferable to manually constructing a loop, since the distribution function itself can still be evaluated in a vectorized manner.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html#divergent-transitions",
-    "href": "reference-manual/mcmc.html#divergent-transitions",
-    "title": "MCMC Sampling",
+    "objectID": "reference-manual/statements.html#for-loops",
+    "href": "reference-manual/statements.html#for-loops",
+    "title": "Statements",
     "section": "",
-    "text": "The Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.\nIn Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. Leimkuhler and Reich (2004) provide details and error analysis for the leapfrog algorithm.\nA divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,5 the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. Betancourt (2016a) provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.\nThe Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.\nIf the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.\nThe primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.\n\n\nIn some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.\nBefore reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.",
+    "text": "Suppose N is a variable of type int, y is a one-dimensional array of type array[] real, and mu and sigma are variables of type real. Furthermore, suppose that n has not been defined as a variable. Then the following is a well-formed for-loop statement.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nThe loop variable is n, the loop bounds are the values in the range 1:N, and the body is the statement following the loop bounds.\n\n\nThe type of the loop variable is int. Unlike in C++ and similarly to R, this variable must not be declared explicitly.\nThe bounds in a for loop must be integers. Unlike in R, the loop is always interpreted as an upward counting loop. The range L:H will cause the loop to execute the loop with the loop variable taking on all integer values greater than or equal to L and less than or equal to H. For example, the loop for (n in 2:5) will cause the body of the for loop to be executed with n equal to 2, 3, 4, and 5, in order. The variable and bound for (n in   5:2) will not execute anything because there are no integers greater than or equal to 5 and less than or equal to 2.\nThe scope of the loop variable is limited to the body of the loop.\n\n\n\nUnlike in BUGS, Stan allows variables to be reassigned. For example, the variable theta in the following program is reassigned in each iteration of the loop.\nfor (n in 1:N) {\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nSuch reassignment is not permitted in BUGS. In BUGS, for loops are declarative, defining plates in directed graphical model notation, which can be thought of as repeated substructures in the graphical model. Therefore, it is illegal in BUGS or JAGS to have a for loop that repeatedly reassigns a value to a variable.4\nIn Stan, assignments are executed in the order they are encountered. As a consequence, the following Stan program has a very different interpretation than the previous one.\nfor (n in 1:N) {\n  y[n] ~ bernoulli(theta);\n  theta = inv_logit(alpha + x[n] * beta);\n}\nIn this program, theta is assigned after it is used in the probability statement. This presupposes it was defined before the first loop iteration (otherwise behavior is undefined), and then each loop uses the assignment from the previous iteration.\nStan loops may be used to accumulate values. Thus it is possible to sum the values of an array directly using code such as the following.\ntotal = 0.0;\nfor (n in 1:N) {\n  total = total + x[n];\n}\nAfter the for loop is executed, the variable total will hold the sum of the elements in the array x. This example was purely pedagogical; it is easier and more efficient to write\ntotal = sum(x);\nA variable inside (or outside) a loop may even be reassigned multiple times, as in the following legal code.\nfor (n in 1:100) {\n  y += y * epsilon;\n  epsilon = 0.5 * epsilon;\n  y += y * epsilon;\n}",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/mcmc.html#footnotes",
-    "href": "reference-manual/mcmc.html#footnotes",
-    "title": "MCMC Sampling",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe typical set is a concept borrowed from information theory and refers to the neighborhood (or neighborhoods in multimodal models) of substantial posterior probability mass through which the Markov chain will travel in equilibrium.↩︎\nThis optimization of step size during adaptation of the sampler should not be confused with running Stan’s optimization method.↩︎\nIn Riemannian HMC the metric compensates for nonlinear correlations.↩︎\nStan previously used slice sampling along the trajectory, following the original NUTS paper of Hoffman and Gelman (2014).↩︎\nThe current default threshold is a factor of \\(10^3\\), whereas when the leapfrog integrator is working properly, the divergences will be around \\(10^{-7}\\) and do not compound due to the symplectic nature of the leapfrog integrator.↩︎",
+    "objectID": "reference-manual/statements.html#foreach-loops",
+    "href": "reference-manual/statements.html#foreach-loops",
+    "title": "Statements",
+    "section": "",
+    "text": "A second form of for loops allows iteration over elements of containers. If ys is an expression denoting a container (vector, row vector, matrix, or array) with elements of type T, then the following is a well-formed foreach statement.\nfor (y in ys) {\n  // ... do something with y ...\n}\nThe order in which elements of ys are visited is defined for container types as follows.\n\nvector, row_vector: elements visited in order, y is of type double\nmatrix: elements visited in column-major order, y is of type double\narray[] T: elements visited in order, y is of type T.\n\nConsequently, if ys is a two dimensional array array[,] real, y will be a one-dimensional array of real values (type array[] real). If ’ysis a matrix, thenywill be a real value (typereal`). To loop over all values of a two-dimensional array using foreach statements would require a doubly-nested loop,\narray[2, 3] real yss;\nfor (ys in yss) {\n  for (y in ys) {\n    // ... do something with y ...\n  }\n}\nwhereas a matrix can be looped over in one foreach statement\nmatrix[2, 3] yss;\nfor (y in yss) {\n   // ... do something with y...\n}\nIn both cases, the loop variable y is of type real.  The elements of the matrix are visited in column-major order (e.g.,y[1, 1],y[2, 1],y[1, 2], ...,y[2, 3]), whereas the elements of the two-dimensional array are visited in row-major order (e.g.,y[1, 1],y[1, 2],y[1, 3],y[2, 1], ...,y[2, 3]`).",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "MCMC Sampling"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/laplace.html",
-    "href": "reference-manual/laplace.html",
-    "title": "Laplace Approximation",
+    "objectID": "reference-manual/statements.html#conditional-statements",
+    "href": "reference-manual/statements.html#conditional-statements",
+    "title": "Statements",
     "section": "",
-    "text": "Laplace Approximation\nStan provides a Laplace approximation algorithm which can be used to obtain draws from an approximated posterior. The Laplace approximation works in the unconstrained space, so that if there are constrained parameters, the normal approximation is centered at the mode in the unconstrained space and then the implemented method transforms the normal approximation sample to the constrained space before outputting them.\nGiven the estimate of the mode \\(\\widehat{\\theta}\\), the Hessian \\(H(\\widehat{\\theta})\\) is computed using central finite differences of the model functor. Next the algorithm computes the Cholesky factor of the negative inverse Hessian:\n\\(R^{-1} = \\textrm{chol}(-H(\\widehat{\\theta})) \\backslash \\mathbf{1}\\).\nEach draw is generated on the unconstrained scale by sampling\n\\(\\theta^{\\textrm{std}(m)} \\sim \\textrm{normal}(0, \\textrm{I})\\)\nand defining draw \\(m\\) to be\n\\(\\theta^{(m)} = \\widehat{\\theta} + R^{-1} \\cdot \\theta^{\\textrm{std}(m)}\\)\nFinally, each \\(\\theta^{(m)}\\) is transformed back to the constrained scale.\nThe one-time computation of the Cholesky factor incurs a high constant overhead of \\(\\mathcal{O}(N^3)\\) in \\(N\\) dimensions. It also requires \\(2N\\) gradient calculations to use as the basis, which scales at best as \\(\\mathcal{O}(N^2)\\) and is worse for models whose gradient calculation is super-linear in dimension. The algorithm also has a high per-draw overhead, requiring \\(N\\) standard normal pseudorandom numbers and \\(\\mathcal{O}(N^2)\\) per draw (to multiply by the Cholesky factor). For \\(M\\) draws, the total cost is proportional to \\(\\mathcal{O}(N^3 + M \\cdot N^2)\\).\n\n\n\n\n Back to top",
+    "text": "Stan supports full conditional statements using the same if-then-else syntax as C++. The general format is\nif (condition1)\n  statement1\nelse if (condition2)\n  statement2\n// ...\nelse if (conditionN-1)\n  statementN-1\nelse\n  statementN\nThere must be a single leading if clause, which may be followed by any number of else if clauses, all of which may be optionally followed by an else clause. Each condition must be an integer value, with non-zero values interpreted as true and the zero value as false.\nThe entire sequence of if-then-else clauses forms a single conditional statement for evaluation. The conditions are evaluated in order until one of the conditions evaluates to a non-zero value, at which point its corresponding statement is executed and the conditional statement finishes execution. If none of the conditions evaluate to a non-zero value and there is a final else clause, its statement is executed.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Laplace Approximation"
+      "Language",
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/includes.html",
-    "href": "reference-manual/includes.html",
-    "title": "Includes",
+    "objectID": "reference-manual/statements.html#while-statements",
+    "href": "reference-manual/statements.html#while-statements",
+    "title": "Statements",
     "section": "",
-    "text": "Stan allows one file to be included within another file using a syntax similar to that from C++. For example, suppose the file my-std-normal.stan defines the standard normal log probability density function (up to an additive constant).\nfunctions {\n  real my_std_normal_lpdf(vector y) {\n    return -0.5 * y' * y;\n  }\n}\nSuppose we also have a file containing a Stan program with an include statement.\n#include my-std-normal.stan\nparameters {\n  real y;\n}\nmodel {\n  y ~ my_std_normal();\n}\nThis Stan program behaves as if the contents of the file my-std-normal.stan replace the line with the #include statement, behaving as if a single Stan program were provided.\nfunctions {\n  real my_std_normal_lpdf(vector y) {\n    return -0.5 * y' * y;\n  }\n}\nparameters {\n  real y;\n}\nmodel {\n  y ~ my_std_normal();\n}\nThere are no restrictions on where include statements may be placed within a file or what the contents are of the replaced file.\n\n\nIt is possible to use includes on a line non-initially. For example, the previous example could’ve included space before the # in the include line:\n    #include my-std-normal.stan\nparameters {\n// ...\nIf there is initial space before an include, it will be discarded.\n\n\n\nIt is also possible to include line-based comments after the include. For example, the previous example can be coded as:\n#include my-std-normal.stan  // definition of standard normal\nparameters {\n// ...\nLine comments are discarded when the entire line is replaced with the contents of the included file.\n\n\n\nRecursive includes will lead to a compiler error. For example, suppose a.stan contains\n#include b.stan\nand b.stan contains\n#include a.stan\nThis will result in an error explaining the circular dependency:\nSyntax error in './b.stan', line 1, column 0, included from\n'./a.stan', line 1, column 0, included from\n'./b.stan', line 1, column 0, included from\n'a.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include a.stan\n         ^\n   -------------------------------------------------\n\nFile a.stan recursively included itself.\n\n\n\nThe Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.\n\n\nIf there is not a final / or \\ in the path, a / will be appended between the path and the included file name.",
+    "text": "Stan supports standard while loops using the same syntax as C++. The general format is as follows.\nwhile (condition)\n  body\nThe condition must be an integer expression and the body can be any statement (or sequence of statements in curly braces).\nEvaluation of a while loop starts by evaluating the condition. If the condition evaluates to a false (zero) value, the execution of the loop terminates and control moves to the position after the loop. If the loop’s condition evaluates to a true (non-zero) value, the body statement is executed, then the whole loop is executed again. Thus the loop is continually executed as long as the condition evaluates to a true value.\nThe rest of the body of a while loop may be skipped using a continue. The loop will be exited with a break statement. See the section on continue and break statements for more details.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Includes"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/includes.html#recursive-includes",
-    "href": "reference-manual/includes.html#recursive-includes",
-    "title": "Includes",
+    "objectID": "reference-manual/statements.html#statement-blocks-and-local-variable-declarations",
+    "href": "reference-manual/statements.html#statement-blocks-and-local-variable-declarations",
+    "title": "Statements",
     "section": "",
-    "text": "Recursive includes will lead to a compiler error. For example, suppose a.stan contains\n#include b.stan\nand b.stan contains\n#include a.stan\nThis will result in an error explaining the circular dependency:\nSyntax error in './b.stan', line 1, column 0, included from\n'./a.stan', line 1, column 0, included from\n'./b.stan', line 1, column 0, included from\n'a.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include a.stan\n         ^\n   -------------------------------------------------\n\nFile a.stan recursively included itself.",
+    "text": "Just as parentheses may be used to group expressions, curly brackets may be used to group a sequence of zero or more statements into a statement block. At the beginning of each block, local variables may be declared that are scoped over the rest of the statements in the block.\n\n\nBlocks are often used to group a sequence of statements together to be used in the body of a for loop. Because the body of a for loop can be any statement, for loops with bodies consisting of a single statement can be written as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nTo put multiple statements inside the body of a for loop, a block is used, as in the following example.\nfor (n in 1:N) {\n  lambda[n] ~ gamma(alpha, beta);\n  y[n] ~ poisson(lambda[n]);\n}\nThe open curly bracket ({) is the first character of the block and the close curly bracket (}) is the last character.\nBecause whitespace is ignored in Stan, the following program will not compile.\nfor (n in 1:N)\n  y[n] ~ normal(mu, sigma);\n  z[n] ~ normal(mu, sigma); // ERROR!\nThe problem is that the body of the for loop is taken to be the statement directly following it, which is y[n] ~ normal(mu, sigma). This leaves the probability statement for z[n] hanging, as is clear from the following equivalent program.\nfor (n in 1:N) {\n  y[n] ~ normal(mu, sigma);\n}\nz[n] ~ normal(mu, sigma); // ERROR!\nNeither of these programs will compile. If the loop variable n was defined before the for loop, the for-loop declaration will raise an error. If the loop variable n was not defined before the for loop, then the use of the expression z[n] will raise an error.\n\n\n\nA for loop has a statement as a body. It is often convenient in writing programs to be able to define a local variable that will be used temporarily and then forgotten. For instance, the for loop example of repeated assignment should use a local variable for maximum clarity and efficiency, as in the following example.\nfor (n in 1:N) {\n  real theta;\n  theta = inv_logit(alpha + x[n] * beta);\n  y[n] ~ bernoulli(theta);\n}\nThe local variable theta is declared here inside the for loop. The scope of a local variable is just the block in which it is defined. Thus theta is available for use inside the for loop, but not outside of it. As in other situations, Stan does not allow variable hiding. So it is illegal to declare a local variable theta if the variable theta is already defined in the scope of the for loop. For instance, the following is not legal.\nfor (m in 1:M) {\n  real theta;\n  for (n in 1:N) {\n    real theta; // ERROR!\n    theta = inv_logit(alpha + x[m, n] * beta);\n    y[m, n] ~ bernoulli(theta);\n// ...\nThe compiler will flag the second declaration of theta with a message that it is already defined.\n\n\n\nLocal variables may not have constraints on their declaration. The only types that may be used are listed in the types table under “local”.\n\n\n\nA block is itself a statement, so anywhere a sequence of statements is allowed, one or more of the statements may be a block. For instance, in a for loop, it is legal to have the following\nfor (m in 1:M) {\n  {\n     int n = 2 * m;\n     sum += n;\n  }\n  for (n in 1:N) {\n    sum += x[m, n];\n  }\n}\nThe variable declaration int n; is the first element of an embedded block and so has scope within that block. The for loop defines its own local block implicitly over the statement following it in which the loop variable is defined. As far as Stan is concerned, these two uses of n are unrelated.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Includes"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/includes.html#include-paths",
-    "href": "reference-manual/includes.html#include-paths",
-    "title": "Includes",
+    "objectID": "reference-manual/statements.html#break-continue-statements",
+    "href": "reference-manual/statements.html#break-continue-statements",
+    "title": "Statements",
     "section": "",
-    "text": "The Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.\n\n\nIf there is not a final / or \\ in the path, a / will be appended between the path and the included file name.",
+    "text": "The one-token statements continue and break may be used within loops to alter control flow; continue causes the next iteration of the loop to run immediately, whereas break terminates the loop and causes execution to resume after the loop. Both control structures must appear in loops. Both break and continue scope to the most deeply nested loop, but pass through non-loop statements.\nAlthough these control statements may seem undesirable because of their goto-like behavior, their judicious use can greatly improve readability by reducing the level of nesting or eliminating bookkeeping inside loops.\n\n\nWhen a break statement is executed, the most deeply nested loop currently being executed is ended and execution picks up with the next statement after the loop. For example, consider the following program:\nwhile (1) {\n  if (n &lt; 0) {\n    break;\n  }\n  foo(n);\n  n = n - 1;\n}\nThe while~(1) loop is a “forever” loop, because 1 is the true value, so the test always succeeds. Within the loop, if the value of n is less than 0, the loop terminates, otherwise it executes foo(n) and then decrements n. The statement above does exactly the same thing as\nwhile (n &gt;= 0) {\n  foo(n);\n  n = n - 1;\n}\nThis case is simply illustrative of the behavior; it is not a case where a break simplifies the loop.\n\n\n\nThe continue statement ends the current operation of the loop and returns to the condition at the top of the loop. Such loops are typically used to exclude some values from calculations. For example, we could use the following loop to sum the positive values in the array x,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &lt;= 0) {\n    continue;\n  }\n  sum += x[n];\n}\nWhen the continue statement is executed, control jumps back to the conditional part of the loop. With while and for loops, this causes control to return to the conditional of the loop. With for loops, this advances the loop variable, so the the above program will not go into an infinite loop when faced with an x[n] less than zero. Thus the above program could be rewritten with deeper nesting by reversing the conditional,\nreal sum;\nsum = 0;\nfor (n in 1:size(x)) {\n  if (x[n] &gt; 0) {\n    sum += x[n];\n  }\n}\nWhile the latter form may seem more readable in this simple case, the former has the main line of execution nested one level less deep. Instead, the conditional at the top finds cases to exclude and doesn’t require the same level of nesting for code that’s not excluded. When there are several such exclusion conditions, the break or continue versions tend to be much easier to read.\n\n\n\nIf there is a loop nested within a loop, a break or continue statement only breaks out of the inner loop. So\nwhile (cond1) {\n  // ...\n  while (cond2) {\n    // ...\n    if (cond3) {\n      break;\n    }\n    // ...\n  }\n  // execution continues here after break\n  // ...\n}\nIf the break is triggered by cond3 being true, execution will continue after the nested loop.\nAs with break statements, continue statements go back to the top of the most deeply nested loop in which the continue appears.\nAlthough break and continue must appear within loops, they may appear in nested statements within loops, such as within the conditionals shown above or within nested statements. The break and continue statements jump past any control structure other than while-loops and for-loops.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Includes"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html",
-    "href": "reference-manual/execution.html",
-    "title": "Program Execution",
+    "objectID": "reference-manual/statements.html#print-statements.section",
+    "href": "reference-manual/statements.html#print-statements.section",
+    "title": "Statements",
     "section": "",
-    "text": "This chapter provides a sketch of how a compiled Stan model is executed using sampling. Optimization shares the same data reading and initialization steps, but then does optimization rather than sampling.\nThis sketch is elaborated in the following chapters of this part, which cover variable declarations, expressions, statements, and blocks in more detail.\n\n\nThe reading and transforming data steps are the same for sampling, optimization and diagnostics.\n\n\nThe first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.1\nAll of the variables declared in the data block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.\nAfter each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable N is declared as int&lt;lower=0&gt;, after N is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.\n\n\n\nAfter data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.\nTransformed data variables are initialized with real values set to NaN and integer values set to the smallest integer (large absolute value negative number).\nAfter the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.\n\n\n\n\nInitialization is the same for all of Stan’s algorithms.\n\n\nIf there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.\nAfter being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.\n\n\nBecause of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nan initial value of 0 for theta leads to an unconstrained value of \\(-\\infty\\), whereas a value of 1 leads to an unconstrained value of \\(+\\infty\\). While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.\n\n\n\n\nIf there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval \\((-2,2)\\). The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.\nAn unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.\nFor parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to \\(\\exp(0) = 1\\) on the constrained scale. A value of -2 corresponds to \\(\\exp(-2) = .13\\) and a value of 2 corresponds to \\(\\exp(2) = 7.4\\).\nFor parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.\nSimplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is \\(1/K\\) in a \\(K\\)-simplex).\nCholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.\nThe initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the chapter on variable transforms.\n\n\n\nThe initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.\n\n\n\n\nSampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.\n\n\nFor each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.\nFor simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.\nIf the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively (Hoffman and Gelman 2014).\n\n\n\nDuring each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.\nThe first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.\nBecause the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.\nNext, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.\nThe final step in the log probability function calculation is to execute the statements defined in the model block.\nAs the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.\n\n\n\nA standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.\nIf step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).\nIf the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.\n\n\n\n\nOptimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.\n\n\n\nVariational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.\n\n\n\nModel diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.\n\n\n\nFor each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.\n\n\nBefore generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.\nAfter the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.\n\n\n\nThe final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.\nIn the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).2",
+    "text": "Stan provides print statements that can print literal strings and the values of expressions. Print statements accept any number of arguments. Consider the following for-each statement with a print statement in its body.\nfor (n in 1:N) { print(\"loop iteration: \", n); ... }\nThe print statement will execute every time the body of the loop does. Each time the loop body is executed, it will print the string “loop iteration:” (with the trailing space), followed by the value of the expression n, followed by a new line.\n\n\nThe text printed by a print statement varies based on its content. A literal (i.e., quoted) string in a print statement always prints exactly that string (without the quotes). Expressions in print statements result in the value of the expression being printed. But how the value of the expression is formatted will depend on its type.\nPrinting a simple real or int typed variable always prints the variable’s value.5\nFor array, vector, and matrix variables, the print format uses brackets. For example, a 3-vector will print as\n[1, 2, 3]\nand a \\(2 \\times 3\\)-matrix as\n[[1, 2, 3], [4, 5, 6]]\nComplex numbers print as pairs. For example, the pair of statements\ncomplex z = to_complex(1.2, -3.5);\nprint(z)\nwill print as (1.2,-3.5), with no space after the comma or within the parentheses.\nPrinting a more readable version of arrays or matrices can be done with loops. An example is the print statement in the following transformed data block.\ntransformed data {\n  matrix[2, 2] u;\n  u[1, 1] = 1.0;   u[1, 2] = 4.0;\n  u[2, 1] = 9.0;   u[2, 2] = 16.0;\n  for (n in 1:2) {\n    print(\"u[\", n, \"] = \", u[n]);\n  }\n}\nThis print statement executes twice, printing the following two lines of output.\nu[1] = [1, 4]\nu[2] = [9, 16]\n\n\n\nThe input type to a print function cannot be void. In particular, it can’t be the result of a user-defined void function. All other types are allowed as arguments to the print function.\n\n\n\nPrinting for a print statement happens every time it is executed. The transformed data block is executed once per chain, the transformed parameter and model blocks once per leapfrog step, and the generated quantities block once per iteration.\n\n\n\nString literals begin and end with a double quote character (\"). The characters between the double quote characters may be any byte sequence, with the exception of the double quote character.\nThe Stan interfaces preserve the byte sequences which they receive. The encoding of these byte sequences as characters and their rendering as glyphs will be handled by whatever display mechanism is being used to monitor Stan’s output (e.g., a terminal, a Jupyter notebook, RStudio, etc.). Stan does not enforce a character encoding for strings, and no attempt is made to validate the bytes as legal ASCII, UTF-8, etc.\n\n\n\nBecause Stan is an imperative language, print statements can be very useful for debugging. They can be used to display the values of variables or expressions at various points in the execution of a program. They are particularly useful for spotting problematic not-a-number of infinite values, both of which will be printed.\nIt is particularly useful to print the value of the target log density accumulator (through the target() function), as in the following example.\nvector[2] y;\ny[1] = 1;\nprint(\"log density before =\", target());\ny ~ normal(0,1);  // bug!  y[2] not defined\nprint(\"log density after =\", target());\nThe example has a bug in that y[2] is not defined before the vector y is used in the distribution statement. By printing the value of the log probability accumulator before and after each distribution statement, it’s possible to isolate where the log probability becomes ill-defined (i.e., becomes not-a-number).\nNote that print statements may not always be displayed immediately, but rather at the end of an operation (e.g., leapfrog step). As such, some issues such as infinite loops are difficult to debug effectively with this technique.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#reading-and-transforming-data",
-    "href": "reference-manual/execution.html#reading-and-transforming-data",
-    "title": "Program Execution",
+    "objectID": "reference-manual/statements.html#reject-statements.section",
+    "href": "reference-manual/statements.html#reject-statements.section",
+    "title": "Statements",
     "section": "",
-    "text": "The reading and transforming data steps are the same for sampling, optimization and diagnostics.\n\n\nThe first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.1\nAll of the variables declared in the data block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.\nAfter each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable N is declared as int&lt;lower=0&gt;, after N is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.\n\n\n\nAfter data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.\nTransformed data variables are initialized with real values set to NaN and integer values set to the smallest integer (large absolute value negative number).\nAfter the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.",
+    "text": "The Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution and either halt processing or reject iterations.\nLike the print statement, the reject statement accepts any number of quoted string literals or Stan expressions as arguments.\nReject statements are typically embedded in a conditional statement in order to detect variables in illegal states. For example, the following code handles the case where a variable x’s value is negative.\nif (x &lt; 0) {\n  reject(\"x must not be negative; found x=\", x);\n}\n\n\nReject statements have the same behavior as exceptions thrown by built-in Stan functions. For example, the normal_lpdf function raises an exception if the input scale is not positive and finite. The effect of a reject statement depends on the program block in which the rejection occurs.\nIn all cases of rejection, the interface accessing the Stan program should print the arguments to the reject statement.\n\n\nRejections in user-defined functions are just passed to the calling function or program block. Reject statements can be used in functions to validate the function arguments, allowing user-defined functions to fully emulate built-in function behavior. It is better to find out earlier rather than later when there is a problem.\n\n\n\nRejections are fatal in the transformed data block. This is because if initialization fails there is no way to recover values, so the algorithm will not begin execution.\nReject statements placed in the transformed data block can be used to validate both the data and transformed data (if any). This allows more complicated constraints to be enforced that can be specified with Stan’s constrained variable declarations.\nFatal errors in other blocks may also be signaled by use of the fatal_error statement.\n\n\n\nRejections in the transformed parameters and model blocks are not in and of themselves instantly fatal. The result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization.\nIf the log probability function results in a rejection every time it is called, the containing application (MCMC sampler or optimization) should diagnose this problem and terminate with an appropriate error message. To aid in diagnosing problems, the message for each reject statement will be printed as a result of executing it.\n\n\n\n\nRejection should be used for error handling, not defining arbitrary constraints. Consider the following errorful Stan program.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;\n  real&lt;lower=a, upper=b&gt; theta;\n  // ...\n}\nmodel {\n  // **wrong** needs explicit truncation\n  theta ~ normal(0, 1);\n  // ...\n}\nThis program is wrong because its truncation bounds on theta depend on parameters, and thus need to be accounted for using an explicit truncation on the distribution. This is the right way to do it.\n  theta ~ normal(0, 1) T[a, b];\nThe conceptual issue is that the prior does not integrate to one over the admissible parameter space; it integrates to one over all real numbers and integrates to something less than one over \\([a ,b]\\); in these simple univariate cases, we can overcome that with the T[ , ] notation, which essentially divides by whatever the prior integrates to over \\([a, b]\\).\nThis problem is exactly the same problem as you would get using reject statements to enforce complicated inequalities on multivariate functions. In this case, it is wrong to try to deal with truncation through constraints.\n  if (theta &lt; a || theta &gt; b) {\n    reject(\"theta not in (a, b)\");\n  }\n  // still **wrong**, needs T[a,b]\n  theta ~ normal(0, 1);\nIn this case, the prior integrates to something less than one over the region of the parameter space where the complicated inequalities are satisfied. But we don’t generally know what value the prior integrates to, so we can’t increment the log probability function to compensate.\nEven if this adjustment to a proper probability model may seem minor in particular models where the amount of truncated posterior density is negligible or constant, we can’t sample from that truncated posterior efficiently. Programs need to use one-to-one mappings that guarantee the constraints are satisfied and only use reject statements to raise errors or help with debugging.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#initialization",
-    "href": "reference-manual/execution.html#initialization",
-    "title": "Program Execution",
+    "objectID": "reference-manual/statements.html#fatal-error-statements",
+    "href": "reference-manual/statements.html#fatal-error-statements",
+    "title": "Statements",
     "section": "",
-    "text": "Initialization is the same for all of Stan’s algorithms.\n\n\nIf there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.\nAfter being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.\n\n\nBecause of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nan initial value of 0 for theta leads to an unconstrained value of \\(-\\infty\\), whereas a value of 1 leads to an unconstrained value of \\(+\\infty\\). While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.\n\n\n\n\nIf there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval \\((-2,2)\\). The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.\nAn unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.\nFor parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to \\(\\exp(0) = 1\\) on the constrained scale. A value of -2 corresponds to \\(\\exp(-2) = .13\\) and a value of 2 corresponds to \\(\\exp(2) = 7.4\\).\nFor parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.\nSimplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is \\(1/K\\) in a \\(K\\)-simplex).\nCholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.\nThe initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the chapter on variable transforms.\n\n\n\nThe initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.",
+    "text": "The Stan fatal_error statement provides a mechanism to report errors or problematic values encountered during program execution and uniformly halt processing.\nLike the print or reject statements, the fatal error statement accepts any number of quoted string literals or Stan expressions as arguments.\nThe fatal error may be used to signal an unrecoverable error in blocks where reject leads to the algorithm attempting to try again, such as the model block.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#sampling",
-    "href": "reference-manual/execution.html#sampling",
-    "title": "Program Execution",
-    "section": "",
-    "text": "Sampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.\n\n\nFor each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.\nFor simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.\nIf the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively (Hoffman and Gelman 2014).\n\n\n\nDuring each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.\nThe first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.\nBecause the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.\nNext, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.\nThe final step in the log probability function calculation is to execute the statements defined in the model block.\nAs the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.\n\n\n\nA standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.\nIf step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).\nIf the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.",
+    "objectID": "reference-manual/statements.html#footnotes",
+    "href": "reference-manual/statements.html#footnotes",
+    "title": "Statements",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe current notation replaces two previous versions. Originally, a variable lp__ was directly exposed and manipulated; this is no longer allowed. The original statement syntax for target += u was increment_log_prob(u), but this form was removed in Stan 2.33↩︎\nWriting this model with the expression -0.5 * y * y is more efficient than with the equivalent expression y * y / -2 because multiplication is more efficient than division; in both cases, the negation is rolled into the numeric literal (-0.5 and -2). Writing square(y) instead of y * y would be even more efficient because the derivatives can be precomputed, reducing the memory and number of operations required for automatic differentiation.↩︎\nBecause \\(\\log | \\frac{d}{dy} \\log y | = \\log | 1/y | = - \\log |y|\\).↩︎\nA programming idiom in BUGS code simulates a local variable by replacing theta in the above example with theta[n], effectively creating N different variables, theta[1], …, theta[N]. Of course, this is not a hack if the value of theta[n] is required for all n.↩︎\nThe adjoint component is always zero during execution for the algorithmic differentiation variables used to implement parameters, transformed parameters, and local variables in the model.↩︎",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Statements"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#optimization",
-    "href": "reference-manual/execution.html#optimization",
-    "title": "Program Execution",
+    "objectID": "reference-manual/removals.html",
+    "href": "reference-manual/removals.html",
+    "title": "Removed Features",
     "section": "",
-    "text": "Optimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.",
+    "text": "This chapter lists functionalities that were once present in the language but have since been removed, along with how to replace them.\n\n\nRemoved: The variable lp__ is no longer available for direct access or manipulation.\nReplacement: General manipulation of the value of the lp__ variable is not allowed, but\nlp__ &lt;- lp__ + e;\ncan be replaced with\ntarget += e;\nThe value of lp__ is available through the no-argument function target().\n\n\n\nRemoved: The operator &lt;- for assignment, e.g.,\na &lt;- b;\nis no longer available.\nReplacement: The new syntax uses the operator = for assignment, e.g.,\na = b;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The increment_log_prob(u) statement for incrementing the log density accumulator by u is no longer available.\nReplacement: Replace the above statement with\ntarget += u;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The built-in no-argument function get_lp() is no longer available.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33\n\n\n\nRemoved: Formerly, the probability function for the distribution foo would be applied to an outcome variable y and sequence of zero or more parameters ... to produce the expression foo_log(y, ...). This suffix is no longer a special value.\nReplacement: If y can be a real value (including vectors or matrices), replace\nfoo_log(y, ...)\nwith the log probability density function notation\nfoo_lpdf(y | ...).\nIf y must be an integer (including arrays), instead replace\nfoo_log(y, ...\nwith the log probability mass function\nfoo_lpmf(y | ...).\nRemoved In: Stan 2.33\n\n\n\nRemoved: The log cumulative distribution and complementary cumulative distribution functions for a distribution foo were formerly written as foo_cdf_log and foo_ccdf_log.\nReplacement:\nReplace foo_cdf_log(y, ...) with foo_lcdf(y | ...).\nReplace foo_ccdf_log(y, ...) with foo_lccdf(y | ...).\n\n\n\nRemoved: A user-defined function ending in _log can be no longer be used in statements.qmd#distribution-statements.section.\nReplacement: Replace the _log suffix with _lpdf for density functions or _lpmf for mass functions in the user-defined function.\nRemoved In: Stan 2.33\nNote: Following Stan 2.33, users can stil define a function ending in _log, it simply no longer has a special meaning or is supported in the ~ syntax.\n\n\n\nRemoved: The function if_else is no longer available.\nReplacement: Use the conditional operator which allows more flexibility in the types of b and c and is much more efficient in that it only evaluates whichever of b or c is returned.\nx = if_else(a, b, c);\nwith\nx = a ? b : c;\nRemoved In: Stan 2.33\n\n\n\nRemoved: The use of # for line-based comments is no longer permitted. # may only be used for #include statements.\nReplacement: Use a pair of forward slashes, //, for line comments.\nRemoved In: Stan 2.33\n\n\n\nBefore Stan 2.26, arrays were declared by writing syntax after the variable.\nRemoved: The use of array declarations like\nint n[5];\nreal a[3, 4];\nreal&lt;lower=0&gt; z[5, 4, 2];\nvector[7] mu[3];\nmatrix[7, 2] mu[15, 12];\ncholesky_factor_cov[5, 6] mu[2, 3, 4];\nReplacement: The use of the array keyword, which replaces the above examples with\narray[5] int n;\narray[3, 4] real a;\narray[5, 4, 2] real&lt;lower=0&gt; z;\narray[3] vector[7] mu;\narray[15, 12] matrix[7, 2] mu;\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nRemoved In: Stan 2.33\n\n\n\nStan interprets nested indexing in assingments as flat indexing so that a statement like\na[:][1] = b;\nis the same as\na[:,1] = b;\nHowever, this is inconsistent with multiple indexing rules.\nTo avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.\nRemoved In: Stan 2.33\n\n\n\nRemoved: Using a real value in a conditional is no longer permitted.\nreal x = 1.0;\nif (x) {\nThe value was interpreted as true if it is nonzero.\nReplacement: For the exact equivalent, use a comparison operator to make the intent clear.\nreal x = 1.0;\nif (x != 0) {\nHowever, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative abs(x) &lt; machine_precision() is appropriate for the use case.\nRemoved In: Stan 2.34",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#variational-inference",
-    "href": "reference-manual/execution.html#variational-inference",
-    "title": "Program Execution",
+    "objectID": "reference-manual/removals.html#lp__-variable",
+    "href": "reference-manual/removals.html#lp__-variable",
+    "title": "Removed Features",
     "section": "",
-    "text": "Variational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.",
+    "text": "Removed: The variable lp__ is no longer available for direct access or manipulation.\nReplacement: General manipulation of the value of the lp__ variable is not allowed, but\nlp__ &lt;- lp__ + e;\ncan be replaced with\ntarget += e;\nThe value of lp__ is available through the no-argument function target().",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#model-diagnostics",
-    "href": "reference-manual/execution.html#model-diagnostics",
-    "title": "Program Execution",
+    "objectID": "reference-manual/removals.html#assignment-with--",
+    "href": "reference-manual/removals.html#assignment-with--",
+    "title": "Removed Features",
     "section": "",
-    "text": "Model diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.",
+    "text": "Removed: The operator &lt;- for assignment, e.g.,\na &lt;- b;\nis no longer available.\nReplacement: The new syntax uses the operator = for assignment, e.g.,\na = b;\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#output",
-    "href": "reference-manual/execution.html#output",
-    "title": "Program Execution",
+    "objectID": "reference-manual/removals.html#increment_log_prob-statement",
+    "href": "reference-manual/removals.html#increment_log_prob-statement",
+    "title": "Removed Features",
     "section": "",
-    "text": "For each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.\n\n\nBefore generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.\nAfter the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.\n\n\n\nThe final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.\nIn the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).2",
+    "text": "Removed: The increment_log_prob(u) statement for incrementing the log density accumulator by u is no longer available.\nReplacement: Replace the above statement with\ntarget += u;\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/execution.html#footnotes",
-    "href": "reference-manual/execution.html#footnotes",
-    "title": "Program Execution",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe C++ code underlying Stan is flexible enough to allow data to be read from memory or file. Calls from R, for instance, can be configured to read data from file or directly from R’s memory.↩︎\nIn the R version of Stan, the values may either be written to a CSV file or directly back to R’s memory.↩︎",
+    "objectID": "reference-manual/removals.html#get_lp-function",
+    "href": "reference-manual/removals.html#get_lp-function",
+    "title": "Removed Features",
+    "section": "",
+    "text": "Removed: The built-in no-argument function get_lp() is no longer available.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Program Execution"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/diagnostics.html",
-    "href": "reference-manual/diagnostics.html",
-    "title": "Diagnostic Mode",
+    "objectID": "reference-manual/removals.html#log-density-and-mass-functions",
+    "href": "reference-manual/removals.html#log-density-and-mass-functions",
+    "title": "Removed Features",
     "section": "",
-    "text": "Stan’s diagnostic mode runs a Stan program with data, initializing parameters either randomly or with user-specified initial values, and then evaluates the log probability and its gradients. The gradients computed by the Stan program are compared to values calculated by finite differences.\nDiagnostic mode may be configured with two parameters.\nDiagnostic Mode Configuration Table. The diagnostic model configuration parameters, constraints, and default values.\n\n\n\n\n\n\n\n\n\nparameter\ndescription\nconstraints\ndefault\n\n\n\n\nepsilon\nfinite difference size\n(0, infty)\n1e–6\n\n\nerror\nerror threshold for matching\n(0, infty)\n1e–6\n\n\n\nIf the difference between the Stan program’s gradient value and that calculated by finite difference is higher than the specified threshold, the argument will be flagged.\n\n\nDiagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.\n\n\nThe output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an \\(N \\times N\\) correlation matrix, requires \\(\\binom{N}{2}\\) unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.\n\n\n\nThe log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.\n\n\n\n\nThe general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.\n\n\n\nDue to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.",
+    "text": "Removed: Formerly, the probability function for the distribution foo would be applied to an outcome variable y and sequence of zero or more parameters ... to produce the expression foo_log(y, ...). This suffix is no longer a special value.\nReplacement: If y can be a real value (including vectors or matrices), replace\nfoo_log(y, ...)\nwith the log probability density function notation\nfoo_lpdf(y | ...).\nIf y must be an integer (including arrays), instead replace\nfoo_log(y, ...\nwith the log probability mass function\nfoo_lpmf(y | ...).\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Diagnostic Mode"
+      "Language",
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/diagnostics.html#diagnostic-mode-output",
-    "href": "reference-manual/diagnostics.html#diagnostic-mode-output",
-    "title": "Diagnostic Mode",
+    "objectID": "reference-manual/removals.html#cdf_log-and-ccdf_log-cumulative-distribution-functions",
+    "href": "reference-manual/removals.html#cdf_log-and-ccdf_log-cumulative-distribution-functions",
+    "title": "Removed Features",
     "section": "",
-    "text": "Diagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.\n\n\nThe output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an \\(N \\times N\\) correlation matrix, requires \\(\\binom{N}{2}\\) unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.\n\n\n\nThe log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.",
+    "text": "Removed: The log cumulative distribution and complementary cumulative distribution functions for a distribution foo were formerly written as foo_cdf_log and foo_ccdf_log.\nReplacement:\nReplace foo_cdf_log(y, ...) with foo_lcdf(y | ...).\nReplace foo_ccdf_log(y, ...) with foo_lccdf(y | ...).",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Diagnostic Mode"
+      "Language",
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/diagnostics.html#configuration-options",
-    "href": "reference-manual/diagnostics.html#configuration-options",
-    "title": "Diagnostic Mode",
+    "objectID": "reference-manual/removals.html#user-defined-function-with-_log-suffix",
+    "href": "reference-manual/removals.html#user-defined-function-with-_log-suffix",
+    "title": "Removed Features",
     "section": "",
-    "text": "The general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.",
+    "text": "Removed: A user-defined function ending in _log can be no longer be used in statements.qmd#distribution-statements.section.\nReplacement: Replace the _log suffix with _lpdf for density functions or _lpmf for mass functions in the user-defined function.\nRemoved In: Stan 2.33\nNote: Following Stan 2.33, users can stil define a function ending in _log, it simply no longer has a special meaning or is supported in the ~ syntax.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Diagnostic Mode"
+      "Language",
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/diagnostics.html#speed-warning-and-data-trimming",
-    "href": "reference-manual/diagnostics.html#speed-warning-and-data-trimming",
-    "title": "Diagnostic Mode",
+    "objectID": "reference-manual/removals.html#if_else-function",
+    "href": "reference-manual/removals.html#if_else-function",
+    "title": "Removed Features",
     "section": "",
-    "text": "Due to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.",
+    "text": "Removed: The function if_else is no longer available.\nReplacement: Use the conditional operator which allows more flexibility in the types of b and c and is much more efficient in that it only evaluates whichever of b or c is returned.\nx = if_else(a, b, c);\nwith\nx = a ? b : c;\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Diagnostic Mode"
+      "Language",
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/comments.html",
-    "href": "reference-manual/comments.html",
-    "title": "Comments",
+    "objectID": "reference-manual/removals.html#character-as-comment-prefix",
+    "href": "reference-manual/removals.html#character-as-comment-prefix",
+    "title": "Removed Features",
     "section": "",
-    "text": "Stan supports C++-style line-based and bracketed comments. Comments may be used anywhere whitespace is allowed in a Stan program.\n\n\nAny characters on a line following two forward slashes (//) is ignored along with the slashes. These may be used, for example, to document variables,\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  array[N] real y;  // observations\n}\n\n\n\nFor bracketed comments, any text between a forward-slash and asterisk pair (/*) and an asterisk and forward-slash pair (*/) is ignored.",
+    "text": "Removed: The use of # for line-based comments is no longer permitted. # may only be used for #include statements.\nReplacement: Use a pair of forward slashes, //, for line comments.\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Comments"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/comments.html#line-based-comments",
-    "href": "reference-manual/comments.html#line-based-comments",
-    "title": "Comments",
+    "objectID": "reference-manual/removals.html#postfix-brackets-array-syntax",
+    "href": "reference-manual/removals.html#postfix-brackets-array-syntax",
+    "title": "Removed Features",
     "section": "",
-    "text": "Any characters on a line following two forward slashes (//) is ignored along with the slashes. These may be used, for example, to document variables,\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  array[N] real y;  // observations\n}",
+    "text": "Before Stan 2.26, arrays were declared by writing syntax after the variable.\nRemoved: The use of array declarations like\nint n[5];\nreal a[3, 4];\nreal&lt;lower=0&gt; z[5, 4, 2];\nvector[7] mu[3];\nmatrix[7, 2] mu[15, 12];\ncholesky_factor_cov[5, 6] mu[2, 3, 4];\nReplacement: The use of the array keyword, which replaces the above examples with\narray[5] int n;\narray[3, 4] real a;\narray[5, 4, 2] real&lt;lower=0&gt; z;\narray[3] vector[7] mu;\narray[15, 12] matrix[7, 2] mu;\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Comments"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/comments.html#bracketed-comments",
-    "href": "reference-manual/comments.html#bracketed-comments",
-    "title": "Comments",
+    "objectID": "reference-manual/removals.html#nested-multiple-indexing-in-assignments",
+    "href": "reference-manual/removals.html#nested-multiple-indexing-in-assignments",
+    "title": "Removed Features",
     "section": "",
-    "text": "For bracketed comments, any text between a forward-slash and asterisk pair (/*) and an asterisk and forward-slash pair (*/) is ignored.",
+    "text": "Stan interprets nested indexing in assingments as flat indexing so that a statement like\na[:][1] = b;\nis the same as\na[:,1] = b;\nHowever, this is inconsistent with multiple indexing rules.\nTo avoid confusion nested multiple indexing in assignment became an error in Stan 2.33. Nesting single indexing is still allowed as it cannot lead to ambiguity.\nRemoved In: Stan 2.33",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Comments"
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/analysis.html",
-    "href": "reference-manual/analysis.html",
-    "title": "Posterior Analysis",
+    "objectID": "reference-manual/removals.html#real-values-in-conditionals",
+    "href": "reference-manual/removals.html#real-values-in-conditionals",
+    "title": "Removed Features",
     "section": "",
-    "text": "Stan uses Markov chain Monte Carlo (MCMC) techniques to generate draws from the posterior distribution for full Bayesian inference. Markov chain Monte Carlo (MCMC) methods were developed for situations in which it is not straightforward to make independent draws Metropolis et al. (1953).\nStan’s variational inference algorithm provides draws from the variational approximation to the posterior which may be analyzed just as any other MCMC output, despite the fact that it is not actually a Markov chain.\nStan’s Laplace algorithm produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood.\n\n\nA Markov chain is a sequence of random variables \\(\\theta^{(1)},\n\\theta^{(2)},\\ldots\\) where each variable is conditionally independent of all other variables given the value of the previous value. Thus if \\(\\theta = \\theta^{(1)}, \\theta^{(2)},\\ldots, \\theta^{(N)}\\), then\n\\[\np(\\theta) = p(\\theta^{(1)}) \\prod_{n=2}^N p(\\theta^{(n)}|\\theta^{(n-1)}).\n\\]\nStan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the Hamiltonian Monte Carlo chapter.\nThe Markov chains Stan and other MCMC samplers generate are ergodic in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of \\(\\theta\\) from another. The Markov chains are also stationary, meaning that the transition probabilities do not change at different positions in the chain, so that for \\(n, n' \\geq 0\\), the probability function \\(p(\\theta^{(n+1)}|\\theta^{(n)})\\) is the same as \\(p(\\theta^{(n'+1)}|\\theta^{(n')})\\) (following the convention of overloading random and bound variables and picking out a probability function by its arguments).\nStationary Markov chains have an equilibrium distribution on states in which each has the same marginal probability function, so that \\(p(\\theta^{(n)})\\) is the same probability function as \\(p(\\theta^{(n+1)})\\). In Stan, this equilibrium distribution \\(p(\\theta^{(n)})\\) is the target density \\(p(\\theta)\\) defined by a Stan program, which is typically a proper Bayesian posterior density \\(p(\\theta | y)\\) defined on the log scale up to a constant.\nUsing MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.\nStan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.\n\n\n\nBy definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when \\(p(\\theta^{(n)})\\) is the target density). The following point cannot be expressed strongly enough:\n\nIn theory, convergence is only guaranteed asymptotically as the number of draws grows without bound.\nIn practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available.\n\n\n\n\nTo establish basic notation, suppose a target Bayesian posterior density \\(p(\\theta | y)\\) given real-valued vectors of parameters \\(\\theta\\) and real- and discrete-valued data \\(y\\).1\nAn MCMC sample consists of a set of a sequence of \\(M\\) Markov chains, each consisting of an ordered sequence of \\(N\\) draws from the posterior.2 The sample thus consists of \\(M \\times N\\) draws from the posterior.\n\n\nOne way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the Gelman and Rubin (1992) potential scale reduction statistic, \\(\\hat{R}\\). The \\(\\hat{R}\\) statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and \\(\\hat{R}\\) will be one. If the chains have not converged to a common distribution, the \\(\\hat{R}\\) statistic will be greater than one.\nGelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for \\(\\hat{R}\\) are below some threshold. Vehtari et al. (2021) suggest in general to use a threshold \\(1.01\\), but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.\nThe \\(\\hat{R}\\) statistic is defined for a set of \\(M\\) Markov chains, \\(\\theta_m\\), each of which has \\(N\\) draws \\(\\theta^{(n)}_m\\). The between-chain variance estimate is\n\\[\nB\n=\n\\frac{N}{M-1}\n\\,\n\\sum_{m=1}^M (\\bar{\\theta}^{(\\bullet)}_{m}\n                - \\bar{\\theta}^{(\\bullet)}_{\\bullet})^2,\n\\]\nwhere\n\\[\n\\bar{\\theta}_m^{(\\bullet)}\n= \\frac{1}{N} \\sum_{n = 1}^N \\theta_m^{(n)}\n\\]\nand\n\\[\n\\bar{\\theta}^{(\\bullet)}_{\\bullet}\n= \\frac{1}{M} \\, \\sum_{m=1}^M \\bar{\\theta}_m^{(\\bullet)}.\n\\]\nThe within-chain variance is averaged over the chains,\n\\[\nW = \\frac{1}{M} \\, \\sum_{m=1}^M s_m^2,\n\\]\nwhere\n\\[\ns_m^2\n=\n\\frac{1}{N-1}\n\\, \\sum_{n=1}^N (\\theta^{(n)}_m - \\bar{\\theta}^{(\\bullet)}_m)^2.\n\\]\nThe variance estimator is a mixture of the within-chain and cross-chain sample variances,\n\\[\n\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)\n= \\frac{N-1}{N}\\, W \\, + \\, \\frac{1}{N} \\, B.\n\\]\nFinally, the potential scale reduction statistic is defined by\n\\[\n\\hat{R}\n\\, = \\,\n\\sqrt{\\frac{\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)}{W}}.\n\\]\n\n\n\nBefore Stan calculating the potential-scale-reduction statistic \\(\\hat{R}\\), each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have \\(\\hat{R}\\) values near unity. In this case, splitting each chain into two parts leads to \\(\\hat{R}\\) values substantially greater than 1 because the first half of each chain has not mixed with the second half.\n\n\n\nSplit R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following Vehtari et al. (2021), we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.\nRank normalization proceeds as follows:\n\nFirst, replace each value \\(\\theta^{(nm)}\\) by its rank \\(r^{(nm)}\\) within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.\nSecond, transform ranks to normal scores using the inverse normal transformation and a fractional offset:\n\n\\[\nz_{(nm)} = \\Phi^{-1} \\left( \\frac{r_{(nm)} - 3/8}{S - 1/4} \\right)\n\\]\nTo further improve sensitivity to chains having different scales,\nrank normalized R-hat is computed also for the for the corresponding folded draws \\(\\zeta^{(mn)}\\), absolute deviations from the median, \\[\n\\label{zeta}\n\\zeta^{(mn)} = \\left|\\theta^{(nm)}-{\\rm median}(\\theta)\\right|.\n\\] The rank normalized split-\\(\\widehat{R}\\) measure computed on the \\(\\zeta^{(mn)}\\) values is called -\\(\\widehat{R}\\). This measures convergence in the tails rather than in the bulk of the distribution.\nTo obtain a single conservative \\(\\widehat{R}\\) estimate, we propose to report the maximum of rank normalized split-\\(\\widehat{R}\\) and rank normalized folded-split-\\(\\widehat{R}\\) for each parameter.\nBulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See Effective Sample Size section for details on how ESS is estimated.\n\n\n\nA question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.\nFor example, consider the value lp__, which is the log posterior density (up to a constant).3\nIt is thus a mistake to declare convergence in any practical sense if lp__ has not converged, because different chains are really in different parts of the space. Yet measuring convergence for lp__ is particularly tricky, as noted below.\n\n\nMarkov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.\n\n\n\nThe \\(\\hat{R}\\) statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.\nThe transformation from unconstrained space to constrained space is just another function, so does not effect convergence.\nDifferent functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, lp__ and other measured quantities should at least be consistent.\nThe obvious difference in lp__ is that it tends to vary quickly with position and is consequently susceptible to outliers.\n\n\n\nThe question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.\nSo what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.\n\n\n\nFirstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better \\(\\hat{R}\\). Functions vulnerable to such outliers may show fragility.\nSecondly, use of the \\(\\hat{R}\\) statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (lp__) almost never looks Gaussian, instead it features long tails that can lead to large \\(\\hat{R}\\) even in the large \\(N\\) limit. Tweaks to \\(\\hat{R}\\), such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the \\(\\hat{R}\\) statistic more accurate.\n\n\n\n“Convergence” is a global property and holds for all integrable functions at once, but employing the \\(\\hat{R}\\) statistic requires additional assumptions and thus may not work for all functions equally well.\nNote that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.\n\n\n\n\n\nThe second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see Charles J. Geyer (2011).\nStan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).\nUnlike most packages, the particular calculations used by Stan follow those for split-\\(\\hat{R}\\), which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see Gelman et al. (2013) and Vehtari et al. (2021).\n\n\nThe amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size \\(N\\). Given dependent sample, the sample size is replaced with the effective sample size \\(N_{\\mathrm{eff}}\\).\nFor example, Monte Carlo standard error (MCSE) is proportional to \\(1 / \\sqrt{N_{\\mathrm{eff}}}\\) rather than \\(1/\\sqrt{N}\\).\nThe effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation \\(\\rho_t\\) at lag \\(t \\geq 0\\) for a chain with joint probability function \\(p(\\theta)\\) with mean \\(\\mu\\) and variance \\(\\sigma^2\\) is defined to be\n\\[\n\\rho_t\n=\n\\frac{1}{\\sigma^2} \\, \\int_{\\Theta} (\\theta^{(n)} - \\mu)\n(\\theta^{(n+t)} - \\mu) \\, p(\\theta) \\, d\\theta.\n\\]\nThis is the correlation between the two chains offset by \\(t\\) positions (i.e., a lag in time-series terminology). Because we know \\(\\theta^{(n)}\\) and \\(\\theta^{(n+t)}\\) have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields\n\\[\n\\rho_t\n= \\frac{1}{\\sigma^2}\n\\, \\int_{\\Theta}\n      \\theta^{(n)} \\, \\theta^{(n+t)} \\, p(\\theta)\n   \\, d\\theta - \\frac{\\mu^2}{\\sigma^2}.\n\\]\nThe effective sample size of \\(N\\) draws generated by a process with autocorrelations \\(\\rho_t\\) is defined by \\[\nN_{\\mathrm{eff}}\n\\ = \\\n\\frac{N}{\\sum_{t = -\\infty}^{\\infty} \\rho_t}\n\\ = \\\n\\frac{N}{1 + 2 \\sum_{t = 1}^{\\infty} \\rho_t}.\n\\]\nFor independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.\n\n\n\nIn practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-\\(\\hat{R}\\) based effective sample size estimator, based on multiple chains. As before, each chain \\(\\theta_m\\) will be assumed to be of length \\(N\\).\nStan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see Charles J. Geyer (2011) for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates \\(\\hat{\\rho}_{t,m}\\) at lag \\(t\\) from multiple chains \\(m \\in\n(1,\\ldots,M)\\) are combined with within-sample variance estimate \\(W\\) and multi-chain variance estimate \\(\\widehat{\\mbox{var}}^{+}\\) introduced in the previous section to compute the combined autocorrelation at lag \\(t\\) as\n\\[\n\\hat{\\rho}_t\n= 1 - \\frac{\\displaystyle W\n              - \\textstyle \\frac{1}{M}\\sum_{m=1}^M s_m^2 \\hat{\\rho}_{t,m}}\n        {\\widehat{\\mbox{var}}^{+}}.\n\\]\nIf the chains have not converged, the variance estimator \\(\\widehat{\\mbox{var}}^{+}\\) will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.\nBecause of the noise in the correlation estimates \\(\\hat{\\rho}_t\\) as \\(t\\) increases, a typical truncated sum of \\(\\hat{\\rho}_t\\) is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise Charles J. Geyer (1992), Charles J. Geyer (2011). Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as\n\\[\n\\hat{N}_{\\mathrm{eff}} = \\frac{M \\cdot N}{\\hat{\\tau}},\n\\]\nwhere\n\\[\n\\hat{\\tau} = 1 + 2 \\sum_{t=1}^{2m+1} \\hat{\\rho}_t = -1 + 2 \\sum_{t'=0}^{m} \\hat{P}_{t'},\n\\]\nwhere \\(\\hat{P}_{t'}=\\hat{\\rho}_{2t'}+\\hat{\\rho}_{2t'+1}\\). Initial positive sequence estimators is obtained by choosing the largest \\(m\\) such that \\(\\hat{P}_{t'}&gt;0, \\quad t' = 1,\\ldots,m\\). The initial monotone sequence is obtained by further reducing \\(\\hat{P}_{t'}\\) to the minimum of the preceding ones so that the estimated sequence is monotone.\n\n\n\nThe posterior standard deviation of a parameter \\(\\theta_n\\) conditioned on observed data \\(y\\) is just the standard deviation of the posterior density \\(p(\\theta_n | y)\\). This is estimated by the standard deviation of the combined posterior draws across chains,\n\\[\n\\hat{\\sigma}_n = \\mathrm{sd}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n).\n\\]\nThe previous section showed how to estimate \\(N_{\\mathrm{eff}}\\) for a parameter \\(\\theta_n\\) based on multiple chains of posterior draws.\nThe mean of the posterior draws of \\(\\theta_n\\) \\[\n\\hat{\\theta}_n\n= \\mathrm{mean}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n)\n\\]\nis treated as an estimator of the true posterior mean,\n\\[\n\\mathbb{E}[\\theta_n \\mid y]\n\\ = \\\n\\int_{-\\infty}^{\\infty}\n    \\, \\theta \\, p(\\theta | y)\n\\, \\mathrm{d}\\theta_n,\n\\]\nbased the observed data \\(y\\).\nThe standard error for the estimator \\(\\hat{\\theta}_n\\) is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as \\(\\hat{\\sigma}_n / \\sqrt{N_{\\mathrm{eff}}}\\). The smaller the standard error, the closer the estimate \\(\\hat{\\theta}_n\\) is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see Charles J. Geyer (2011) for more details of the MCMC central limit theorem.\n\n\n\nIn complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag \\(t\\), \\(\\rho_t\\), decreases as the lag, \\(t\\), increases. In this situation, thinning the sample by keeping only every \\(N\\)-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.\nFor instance, consider generating one thousand posterior draws in one of the following two ways.\n\nGenerate 1000 draws after convergence and save all of them.\nGenerate 10,000 draws after convergence and save every tenth draw.\n\nEven though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation \\(\\rho_t\\) for the thinned sequence is equivalent to \\(\\rho_{10t}\\) in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.\nNow contrast the second approach above with the unthinned alternative,\n\nGenerate 10,000 draws after convergence and save every draw.\n\nThis will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, the only reason to thin a sample is to reduce memory requirements.\nIf draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.",
+    "text": "Removed: Using a real value in a conditional is no longer permitted.\nreal x = 1.0;\nif (x) {\nThe value was interpreted as true if it is nonzero.\nReplacement: For the exact equivalent, use a comparison operator to make the intent clear.\nreal x = 1.0;\nif (x != 0) {\nHowever, one should keep in mind that floating point calculations are subject to rounding errors and precise equality is fragile. It is worth considering whether the more robust alternative abs(x) &lt; machine_precision() is appropriate for the use case.\nRemoved In: Stan 2.34",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Posterior Analysis"
+      "Language",
+      "Removed Features"
     ]
   },
   {
-    "objectID": "reference-manual/analysis.html#markov-chains",
-    "href": "reference-manual/analysis.html#markov-chains",
-    "title": "Posterior Analysis",
+    "objectID": "reference-manual/pathfinder.html",
+    "href": "reference-manual/pathfinder.html",
+    "title": "Pathfinder",
     "section": "",
-    "text": "A Markov chain is a sequence of random variables \\(\\theta^{(1)},\n\\theta^{(2)},\\ldots\\) where each variable is conditionally independent of all other variables given the value of the previous value. Thus if \\(\\theta = \\theta^{(1)}, \\theta^{(2)},\\ldots, \\theta^{(N)}\\), then\n\\[\np(\\theta) = p(\\theta^{(1)}) \\prod_{n=2}^N p(\\theta^{(n)}|\\theta^{(n-1)}).\n\\]\nStan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the Hamiltonian Monte Carlo chapter.\nThe Markov chains Stan and other MCMC samplers generate are ergodic in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of \\(\\theta\\) from another. The Markov chains are also stationary, meaning that the transition probabilities do not change at different positions in the chain, so that for \\(n, n' \\geq 0\\), the probability function \\(p(\\theta^{(n+1)}|\\theta^{(n)})\\) is the same as \\(p(\\theta^{(n'+1)}|\\theta^{(n')})\\) (following the convention of overloading random and bound variables and picking out a probability function by its arguments).\nStationary Markov chains have an equilibrium distribution on states in which each has the same marginal probability function, so that \\(p(\\theta^{(n)})\\) is the same probability function as \\(p(\\theta^{(n+1)})\\). In Stan, this equilibrium distribution \\(p(\\theta^{(n)})\\) is the target density \\(p(\\theta)\\) defined by a Stan program, which is typically a proper Bayesian posterior density \\(p(\\theta | y)\\) defined on the log scale up to a constant.\nUsing MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.\nStan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.",
+    "text": "Stan supports the Pathfinder algorithm (Zhang et al. 2022). Pathfinder is a variational method for approximately sampling from differentiable log densities. Starting from a random initialization, Pathfinder locates normal approximations to the target density along a quasi-Newton optimization path, with local covariance estimated using the negative inverse Hessian estimates produced by the LBFGS optimizer. Pathfinder returns draws from the Gaussian approximation with the lowest estimated Kullback-Leibler (KL) divergence to the true posterior.\nStan provides two versions of the Pathfinder algorithm: single-path Pathfinder and multi-path Pathfinder. Single-path Pathfinder generates a set of approximate draws from one run of the basic Pathfinder algorithm. Multi-path Pathfinder uses importance resampling over the draws from multiple runs of Pathfinder. This better matches non-normal target densities and also mitigates the problem of L-BFGS getting stuck at local optima or in saddle points on plateaus. Compared to ADVI and short dynamic HMC runs, Pathfinder requires one to two orders of magnitude fewer log density and gradient evaluations, with greater reductions for more challenging posteriors. While the evaluations by Zhang et al. (2022) found that single-path and multi-path Pathfinder outperform ADVI for most of the models in the PosteriorDB (Magnusson et al. 2024) evaluation set, we recognize the need for further experiments on a wider range of models.\n\n\nPathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-\\(\\hat{k}\\) diagnostic (Vehtari et al. 2024) to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably (Vehtari et al. 2024, sec. 3), which is the first requirement for reliable resampling. If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest (Vehtari et al. 2024, sec. 2.2). If estimated Pareto-\\(\\hat{k}\\) is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow (Gelman et al. 2020).\n\n\n\nIf estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-\\(\\hat{k}\\) for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g. uniform from -2 to 2 used by Stan by default). If Pareto-\\(\\hat{k}\\) is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Posterior Analysis"
+      "Pathfinder"
     ]
   },
   {
-    "objectID": "reference-manual/analysis.html#convergence",
-    "href": "reference-manual/analysis.html#convergence",
-    "title": "Posterior Analysis",
+    "objectID": "reference-manual/pathfinder.html#diagnosing-pathfinder",
+    "href": "reference-manual/pathfinder.html#diagnosing-pathfinder",
+    "title": "Pathfinder",
     "section": "",
-    "text": "By definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when \\(p(\\theta^{(n)})\\) is the target density). The following point cannot be expressed strongly enough:\n\nIn theory, convergence is only guaranteed asymptotically as the number of draws grows without bound.\nIn practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available.",
+    "text": "Pathfinder diagnoses the accuracy of the approximation by computing the density ratio of the true posterior and the approximation and using Pareto-\\(\\hat{k}\\) diagnostic (Vehtari et al. 2024) to assess whether these ratios can be used to improve the approximation via resampling. The normalization for the posterior can be estimated reliably (Vehtari et al. 2024, sec. 3), which is the first requirement for reliable resampling. If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, there is still need to further diagnose reliability of importance sampling estimate for all quantities of interest (Vehtari et al. 2024, sec. 2.2). If estimated Pareto-\\(\\hat{k}\\) is larger than 0.7, then the estimate for the normalization is unreliable and any Monte Carlo estimate may have a big error. The resampled draws can still contain some useful information about the location and shape of the posterior which can be used in early parts of Bayesian workflow (Gelman et al. 2020).",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Posterior Analysis"
+      "Pathfinder"
     ]
   },
   {
-    "objectID": "reference-manual/analysis.html#notation-for-samples-chains-and-draws",
-    "href": "reference-manual/analysis.html#notation-for-samples-chains-and-draws",
-    "title": "Posterior Analysis",
+    "objectID": "reference-manual/pathfinder.html#using-pathfinder-for-initializing-mcmc",
+    "href": "reference-manual/pathfinder.html#using-pathfinder-for-initializing-mcmc",
+    "title": "Pathfinder",
     "section": "",
-    "text": "To establish basic notation, suppose a target Bayesian posterior density \\(p(\\theta | y)\\) given real-valued vectors of parameters \\(\\theta\\) and real- and discrete-valued data \\(y\\).1\nAn MCMC sample consists of a set of a sequence of \\(M\\) Markov chains, each consisting of an ordered sequence of \\(N\\) draws from the posterior.2 The sample thus consists of \\(M \\times N\\) draws from the posterior.\n\n\nOne way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the Gelman and Rubin (1992) potential scale reduction statistic, \\(\\hat{R}\\). The \\(\\hat{R}\\) statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and \\(\\hat{R}\\) will be one. If the chains have not converged to a common distribution, the \\(\\hat{R}\\) statistic will be greater than one.\nGelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for \\(\\hat{R}\\) are below some threshold. Vehtari et al. (2021) suggest in general to use a threshold \\(1.01\\), but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.\nThe \\(\\hat{R}\\) statistic is defined for a set of \\(M\\) Markov chains, \\(\\theta_m\\), each of which has \\(N\\) draws \\(\\theta^{(n)}_m\\). The between-chain variance estimate is\n\\[\nB\n=\n\\frac{N}{M-1}\n\\,\n\\sum_{m=1}^M (\\bar{\\theta}^{(\\bullet)}_{m}\n                - \\bar{\\theta}^{(\\bullet)}_{\\bullet})^2,\n\\]\nwhere\n\\[\n\\bar{\\theta}_m^{(\\bullet)}\n= \\frac{1}{N} \\sum_{n = 1}^N \\theta_m^{(n)}\n\\]\nand\n\\[\n\\bar{\\theta}^{(\\bullet)}_{\\bullet}\n= \\frac{1}{M} \\, \\sum_{m=1}^M \\bar{\\theta}_m^{(\\bullet)}.\n\\]\nThe within-chain variance is averaged over the chains,\n\\[\nW = \\frac{1}{M} \\, \\sum_{m=1}^M s_m^2,\n\\]\nwhere\n\\[\ns_m^2\n=\n\\frac{1}{N-1}\n\\, \\sum_{n=1}^N (\\theta^{(n)}_m - \\bar{\\theta}^{(\\bullet)}_m)^2.\n\\]\nThe variance estimator is a mixture of the within-chain and cross-chain sample variances,\n\\[\n\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)\n= \\frac{N-1}{N}\\, W \\, + \\, \\frac{1}{N} \\, B.\n\\]\nFinally, the potential scale reduction statistic is defined by\n\\[\n\\hat{R}\n\\, = \\,\n\\sqrt{\\frac{\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)}{W}}.\n\\]\n\n\n\nBefore Stan calculating the potential-scale-reduction statistic \\(\\hat{R}\\), each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have \\(\\hat{R}\\) values near unity. In this case, splitting each chain into two parts leads to \\(\\hat{R}\\) values substantially greater than 1 because the first half of each chain has not mixed with the second half.\n\n\n\nSplit R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following Vehtari et al. (2021), we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.\nRank normalization proceeds as follows:\n\nFirst, replace each value \\(\\theta^{(nm)}\\) by its rank \\(r^{(nm)}\\) within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.\nSecond, transform ranks to normal scores using the inverse normal transformation and a fractional offset:\n\n\\[\nz_{(nm)} = \\Phi^{-1} \\left( \\frac{r_{(nm)} - 3/8}{S - 1/4} \\right)\n\\]\nTo further improve sensitivity to chains having different scales,\nrank normalized R-hat is computed also for the for the corresponding folded draws \\(\\zeta^{(mn)}\\), absolute deviations from the median, \\[\n\\label{zeta}\n\\zeta^{(mn)} = \\left|\\theta^{(nm)}-{\\rm median}(\\theta)\\right|.\n\\] The rank normalized split-\\(\\widehat{R}\\) measure computed on the \\(\\zeta^{(mn)}\\) values is called -\\(\\widehat{R}\\). This measures convergence in the tails rather than in the bulk of the distribution.\nTo obtain a single conservative \\(\\widehat{R}\\) estimate, we propose to report the maximum of rank normalized split-\\(\\widehat{R}\\) and rank normalized folded-split-\\(\\widehat{R}\\) for each parameter.\nBulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See Effective Sample Size section for details on how ESS is estimated.\n\n\n\nA question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.\nFor example, consider the value lp__, which is the log posterior density (up to a constant).3\nIt is thus a mistake to declare convergence in any practical sense if lp__ has not converged, because different chains are really in different parts of the space. Yet measuring convergence for lp__ is particularly tricky, as noted below.\n\n\nMarkov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.\n\n\n\nThe \\(\\hat{R}\\) statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.\nThe transformation from unconstrained space to constrained space is just another function, so does not effect convergence.\nDifferent functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, lp__ and other measured quantities should at least be consistent.\nThe obvious difference in lp__ is that it tends to vary quickly with position and is consequently susceptible to outliers.\n\n\n\nThe question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.\nSo what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.\n\n\n\nFirstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better \\(\\hat{R}\\). Functions vulnerable to such outliers may show fragility.\nSecondly, use of the \\(\\hat{R}\\) statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (lp__) almost never looks Gaussian, instead it features long tails that can lead to large \\(\\hat{R}\\) even in the large \\(N\\) limit. Tweaks to \\(\\hat{R}\\), such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the \\(\\hat{R}\\) statistic more accurate.\n\n\n\n“Convergence” is a global property and holds for all integrable functions at once, but employing the \\(\\hat{R}\\) statistic requires additional assumptions and thus may not work for all functions equally well.\nNote that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.",
+    "text": "If estimated Pareto-\\(\\hat{k}\\) for the ratios is smaller than 0.7, the resampled posterior draws are almost as good for initializing MCMC as would independent draws from the posterior be. If estimated Pareto-\\(\\hat{k}\\) for the ratios is larger than 0.7, the Pathfinder draws are not reliable for posterior inference directly, but they are still very likely better for initializing MCMC than random draws from an arbitrary pre-defined distribution (e.g. uniform from -2 to 2 used by Stan by default). If Pareto-\\(\\hat{k}\\) is larger than 0.7, it is likely that one of the ratios is much bigger than others and the default resampling with replacement would produce copies of one unique draw. For initializing several Markov chains, it is better to use resampling without replacement to guarantee unique initialization for each chain. At the moment Stan allows turning off the resampling completely, and then the resampling without replacement can be done outside of Stan.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Posterior Analysis"
+      "Pathfinder"
     ]
   },
   {
-    "objectID": "reference-manual/analysis.html#effective-sample-size.section",
-    "href": "reference-manual/analysis.html#effective-sample-size.section",
-    "title": "Posterior Analysis",
+    "objectID": "reference-manual/mcmc.html",
+    "href": "reference-manual/mcmc.html",
+    "title": "MCMC Sampling",
     "section": "",
-    "text": "The second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see Charles J. Geyer (2011).\nStan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).\nUnlike most packages, the particular calculations used by Stan follow those for split-\\(\\hat{R}\\), which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see Gelman et al. (2013) and Vehtari et al. (2021).\n\n\nThe amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size \\(N\\). Given dependent sample, the sample size is replaced with the effective sample size \\(N_{\\mathrm{eff}}\\).\nFor example, Monte Carlo standard error (MCSE) is proportional to \\(1 / \\sqrt{N_{\\mathrm{eff}}}\\) rather than \\(1/\\sqrt{N}\\).\nThe effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation \\(\\rho_t\\) at lag \\(t \\geq 0\\) for a chain with joint probability function \\(p(\\theta)\\) with mean \\(\\mu\\) and variance \\(\\sigma^2\\) is defined to be\n\\[\n\\rho_t\n=\n\\frac{1}{\\sigma^2} \\, \\int_{\\Theta} (\\theta^{(n)} - \\mu)\n(\\theta^{(n+t)} - \\mu) \\, p(\\theta) \\, d\\theta.\n\\]\nThis is the correlation between the two chains offset by \\(t\\) positions (i.e., a lag in time-series terminology). Because we know \\(\\theta^{(n)}\\) and \\(\\theta^{(n+t)}\\) have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields\n\\[\n\\rho_t\n= \\frac{1}{\\sigma^2}\n\\, \\int_{\\Theta}\n      \\theta^{(n)} \\, \\theta^{(n+t)} \\, p(\\theta)\n   \\, d\\theta - \\frac{\\mu^2}{\\sigma^2}.\n\\]\nThe effective sample size of \\(N\\) draws generated by a process with autocorrelations \\(\\rho_t\\) is defined by \\[\nN_{\\mathrm{eff}}\n\\ = \\\n\\frac{N}{\\sum_{t = -\\infty}^{\\infty} \\rho_t}\n\\ = \\\n\\frac{N}{1 + 2 \\sum_{t = 1}^{\\infty} \\rho_t}.\n\\]\nFor independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.\n\n\n\nIn practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-\\(\\hat{R}\\) based effective sample size estimator, based on multiple chains. As before, each chain \\(\\theta_m\\) will be assumed to be of length \\(N\\).\nStan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see Charles J. Geyer (2011) for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates \\(\\hat{\\rho}_{t,m}\\) at lag \\(t\\) from multiple chains \\(m \\in\n(1,\\ldots,M)\\) are combined with within-sample variance estimate \\(W\\) and multi-chain variance estimate \\(\\widehat{\\mbox{var}}^{+}\\) introduced in the previous section to compute the combined autocorrelation at lag \\(t\\) as\n\\[\n\\hat{\\rho}_t\n= 1 - \\frac{\\displaystyle W\n              - \\textstyle \\frac{1}{M}\\sum_{m=1}^M s_m^2 \\hat{\\rho}_{t,m}}\n        {\\widehat{\\mbox{var}}^{+}}.\n\\]\nIf the chains have not converged, the variance estimator \\(\\widehat{\\mbox{var}}^{+}\\) will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.\nBecause of the noise in the correlation estimates \\(\\hat{\\rho}_t\\) as \\(t\\) increases, a typical truncated sum of \\(\\hat{\\rho}_t\\) is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise Charles J. Geyer (1992), Charles J. Geyer (2011). Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as\n\\[\n\\hat{N}_{\\mathrm{eff}} = \\frac{M \\cdot N}{\\hat{\\tau}},\n\\]\nwhere\n\\[\n\\hat{\\tau} = 1 + 2 \\sum_{t=1}^{2m+1} \\hat{\\rho}_t = -1 + 2 \\sum_{t'=0}^{m} \\hat{P}_{t'},\n\\]\nwhere \\(\\hat{P}_{t'}=\\hat{\\rho}_{2t'}+\\hat{\\rho}_{2t'+1}\\). Initial positive sequence estimators is obtained by choosing the largest \\(m\\) such that \\(\\hat{P}_{t'}&gt;0, \\quad t' = 1,\\ldots,m\\). The initial monotone sequence is obtained by further reducing \\(\\hat{P}_{t'}\\) to the minimum of the preceding ones so that the estimated sequence is monotone.\n\n\n\nThe posterior standard deviation of a parameter \\(\\theta_n\\) conditioned on observed data \\(y\\) is just the standard deviation of the posterior density \\(p(\\theta_n | y)\\). This is estimated by the standard deviation of the combined posterior draws across chains,\n\\[\n\\hat{\\sigma}_n = \\mathrm{sd}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n).\n\\]\nThe previous section showed how to estimate \\(N_{\\mathrm{eff}}\\) for a parameter \\(\\theta_n\\) based on multiple chains of posterior draws.\nThe mean of the posterior draws of \\(\\theta_n\\) \\[\n\\hat{\\theta}_n\n= \\mathrm{mean}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n)\n\\]\nis treated as an estimator of the true posterior mean,\n\\[\n\\mathbb{E}[\\theta_n \\mid y]\n\\ = \\\n\\int_{-\\infty}^{\\infty}\n    \\, \\theta \\, p(\\theta | y)\n\\, \\mathrm{d}\\theta_n,\n\\]\nbased the observed data \\(y\\).\nThe standard error for the estimator \\(\\hat{\\theta}_n\\) is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as \\(\\hat{\\sigma}_n / \\sqrt{N_{\\mathrm{eff}}}\\). The smaller the standard error, the closer the estimate \\(\\hat{\\theta}_n\\) is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see Charles J. Geyer (2011) for more details of the MCMC central limit theorem.\n\n\n\nIn complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag \\(t\\), \\(\\rho_t\\), decreases as the lag, \\(t\\), increases. In this situation, thinning the sample by keeping only every \\(N\\)-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.\nFor instance, consider generating one thousand posterior draws in one of the following two ways.\n\nGenerate 1000 draws after convergence and save all of them.\nGenerate 10,000 draws after convergence and save every tenth draw.\n\nEven though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation \\(\\rho_t\\) for the thinned sequence is equivalent to \\(\\rho_{10t}\\) in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.\nNow contrast the second approach above with the unthinned alternative,\n\nGenerate 10,000 draws after convergence and save every draw.\n\nThis will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, the only reason to thin a sample is to reduce memory requirements.\nIf draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.",
+    "text": "This chapter presents the two Markov chain Monte Carlo (MCMC) algorithms used in Stan, the Hamiltonian Monte Carlo (HMC) algorithm and its adaptive variant the no-U-turn sampler (NUTS), along with details of their implementation and configuration.\n\n\nHamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., Betancourt and Girolami (2013), Neal (2011) for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.\nThis section translates the presentation of HMC by Betancourt and Girolami (2013) into the notation of Gelman et al. (2013).\n\n\nThe goal of sampling is to draw from a density \\(p(\\theta)\\) for parameters \\(\\theta\\). This is typically a Bayesian posterior \\(p(\\theta|y)\\) given data \\(y\\), and in particular, a Bayesian posterior coded as a Stan program.\n\n\n\nHMC introduces auxiliary momentum variables \\(\\rho\\) and draws from a joint density\n\\[\np(\\rho, \\theta) = p(\\rho | \\theta) p(\\theta).\n\\]\nIn most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters \\(\\theta\\),\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\n\\(M\\) is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see Betancourt (2017) for details.\nBy default Stan sets \\(M^{-1}\\) equal to a diagonal estimate of the covariance computed during warmup.\n\n\n\nThe joint density \\(p(\\rho, \\theta)\\) defines a Hamiltonian\n\\[\n\\begin{array}{rcl}\nH(\\rho, \\theta) & = & - \\log p(\\rho, \\theta)\n\\\\[3pt]\n& = & - \\log p(\\rho | \\theta) - \\log p(\\theta).\n\\\\[3pt]\n& = & T(\\rho | \\theta) + V(\\theta),\n\\end{array}\n\\]\nwhere the term\n\\[\nT(\\rho | \\theta) = - \\log p(\\rho | \\theta)\n\\]\nis called the “kinetic energy” and the term\n\\[\nV(\\theta) = - \\log p(\\theta)\n\\]\nis called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.\n\n\n\nStarting from the current value of the parameters \\(\\theta\\), a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.\nFirst, a value for the momentum is drawn independently of the current parameter values,\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\nThus momentum does not persist across iterations.\nNext, the joint system \\((\\theta,\\rho)\\) made up of the current parameter values \\(\\theta\\) and new momentum \\(\\rho\\) is evolved via Hamilton’s equations,\n\\[\n\\begin{array}{rcccl}\n\\displaystyle\n\\frac{d\\theta}{dt}\n& = &\n\\displaystyle\n+ \\frac{\\partial H}{\\partial \\rho}\n& = &\n\\displaystyle\n+ \\frac{\\partial T}{\\partial \\rho}\n\\\\[12pt]\n\\displaystyle\n\\frac{d\\rho}{dt}\n& = &\n\\displaystyle\n- \\frac{\\partial H}{\\partial \\theta }\n& = &\n\\displaystyle\n- \\frac{\\partial T}{\\partial \\theta}\n- \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nWith the momentum density being independent of the target density, i.e., \\(p(\\rho | \\theta) = p(\\rho)\\), the first term in the momentum time derivative, \\({\\partial T} / {\\partial \\theta}\\) is zero, yielding the pair time derivatives\n\\[\n\\begin{array}{rcl}\n\\frac{d \\theta}{d t} & = & +\\frac{\\partial T}{\\partial \\rho}\n\\\\[2pt]\n\\frac{d \\rho}{d t} & = & -\\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\n\n\n\nThe last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.\nLike most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval \\(\\epsilon\\). The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values \\(\\theta\\) or previous momentum value.\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\] It then alternates half-step updates of the momentum and full-step updates of the position.\n\\[\n\\begin{array}{rcl}\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}\n\\\\[6pt]\n\\theta & \\leftarrow\n       & \\theta \\, + \\, \\epsilon \\, M^{-1} \\, \\rho\n\\\\[6pt]\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nBy applying \\(L\\) leapfrog steps, a total of \\(L \\, \\epsilon\\) time is simulated. The resulting state at the end of the simulation (\\(L\\) repetitions of the above three steps) will be denoted \\((\\rho^{*}, \\theta^{*})\\).\nThe leapfrog integrator’s error is on the order of \\(\\epsilon^3\\) per step and \\(\\epsilon^2\\) globally, where \\(\\epsilon\\) is the time interval (also known as the step size); Leimkuhler and Reich (2004) provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.\n\n\n\nIf the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal \\((\\rho^{*}, \\theta^{*})\\) generated by transitioning from \\((\\rho, \\theta)\\) is\n\\[\n\\min \\!\n\\left(\n1,\n\\ \\exp \\! \\left( H(\\rho, \\theta) - H(\\rho^{*}, \\theta^{*}) \\right)\n\\right).\n\\]\nIf the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.\n\n\n\nThe Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters \\(\\theta\\); in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter \\(\\theta\\) is updated using the leapfrog integrator with discretization time \\(\\epsilon\\) and number of steps \\(L\\) according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state \\((\\theta^{*}, \\rho^{*})\\) or keep the existing state.\n\n\n\n\nThe Hamiltonian Monte Carlo algorithm has three parameters which must be set,\n\ndiscretization time \\(\\epsilon\\),\nmetric \\(M\\), and\nnumber of steps taken \\(L\\).\n\nIn practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters Neal (2011), Hoffman and Gelman (2014).\nIf \\(\\epsilon\\) is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If \\(\\epsilon\\) is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.\nIf \\(L\\) is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If \\(L\\) is too large, the algorithm will do too much work on each iteration.\nIf the inverse metric \\(M^{-1}\\) is a poor estimate of the posterior covariance, the step size \\(\\epsilon\\) must be kept small to maintain arithmetic precision. This would lead to a large \\(L\\) to compensate.\n\n\nThe actual integration time is \\(L \\, \\epsilon\\), a function of number of steps. Some interfaces to Stan set an approximate integration time \\(t\\) and the discretization interval (step size) \\(\\epsilon\\). In these cases, the number of steps will be rounded down as\n\\[\nL = \\left\\lfloor \\frac{t}{\\epsilon} \\right\\rfloor.\n\\]\nand the actual integration time will still be \\(L \\, \\epsilon\\).\n\n\n\nStan is able to automatically optimize \\(\\epsilon\\) to match an acceptance-rate target, able to estimate \\(M\\) based on warmup sample iterations, and able to dynamically adapt \\(L\\) on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm Hoffman and Gelman (2014).\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\nWhen adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the warmup adaptation figure, with two fast intervals surrounding a series of growing slow intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.\nThe motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.\n\nIn the initial fast interval the chain is allowed to converge towards the typical set,1 with only parameters that can learn from local information adapted.\nAfter this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.\nLastly, the fast parameters are allowed to adapt to the final update of the slow parameters.\n\nThese intervals may be controlled through the following configuration parameters, all of which must be positive integers:\nAdaptation Parameters Table. The parameters controlling adaptation and their default values.\n\n\n\n\n\n\n\n\nparameter\ndescription\ndefault\n\n\n\n\ninitial buffer\nwidth of initial fast adaptation interval\n75\n\n\nterm buffer\nwidth of final fast adaptation interval\n50\n\n\nwindow\ninitial width of slow adaptation interval\n25\n\n\n\n\n\n\nStan’s HMC algorithms utilize dual averaging Nesterov (2009) to optimize the step size.2\nThis warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of Hoffman and Gelman (2014). In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman and Gelman (2014).\nThe full set of dual-averaging parameters are:\nStep Size Adaptation Parameters Table The parameters controlling step size adaptation, with constraints and default values.\n\n\n\nparameter\ndescription\nconstraint\ndefault\n\n\n\n\ndelta\ntarget Metropolis acceptance rate\n[0, 1]\n0.8\n\n\ngamma\nadaptation regularization scale\n(0, infty)\n0.05\n\n\nkappa\nadaptation relaxation exponent\n(0, infty)\n0.75\n\n\nt_0\nadaptation iteration offset\n(0, infty)\n10\n\n\n\nBy setting the target acceptance parameter \\(\\delta\\) to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of \\(\\delta\\) will also allow some models that would otherwise get stuck to overcome their blockages.\n\n\n\nAll implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.\nSmall step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See Neal (2011) for further discussion of step-size jittering.\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nIf the metric is constant then the resulting implementation is known as Euclidean HMC. Stan allows a choice among three Euclidean HMC implementations,\n\na unit metric (diagonal matrix of ones),\na diagonal metric (diagonal matrix with positive diagonal entries), and\na dense metric (a dense, symmetric positive definite matrix)\n\nto be configured by the user.\nIf the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the warmup adaptation stages figure). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.\nIf the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.\nVariances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.\n\n\nThe metric can compensate for linear (i.e. global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.\nIn complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e. local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.\n\n\n\nThe metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.3\nOne of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.\n\n\n\nStatistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.\n\n\n\nMCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.\nThis only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.\nPoor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.\n\n\n\n\nThe no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., Roberts, Gelman, and Gilks (1997)) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see Hoffman and Gelman (2014).\nNUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either\n\nthe NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or\nthe depth of the completed tree hits the maximum depth allowed.\n\nRather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory Betancourt (2016b).4\nConfiguring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.\nBoth the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as treedepth__ and n_leapfrog__, respectively. Because the final subtree may only be partially constructed, these two will always satisfy\n\\[\n2^{\\mathrm{treedepth} - 1} - 1\n\\ &lt; \\\nN_{\\mathrm{leapfrog}}\n\\ \\le \\\n2^{\\mathrm{treedepth} } - 1.\n\\]\nTree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\n\n\n\n\nIn some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.\nFor example, to generate a sequence of \\(N\\) draws from a binomial with trials \\(K\\) and chance of success \\(\\theta\\), the following program suffices.\ndata {\n  real&lt;lower=0, upper=1&gt; theta;\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n}\ngenerated quantities {\n  array[N] int&lt;lower=0, upper=K&gt; y;\n  for (n in 1:N) {\n    y[n] = binomial_rng(K, theta);\n  }\n}\nFor this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.\nMost models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, theta in the example above could be declared as a parameter and initialized as a parameter.\n\n\n\nStan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the optimization algorithms chapter, and the diagnostics chapter.\n\n\nThe random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nStan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.\n\n\nTogether, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.\n\n\n\n\nThe initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.\n\n\nIf the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).\n\n\n\nIt is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.\n\n\n\nRandom initialization by default initializes the parameter values with values drawn at random from a \\(\\mathsf{Uniform}(-2, 2)\\) distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.\nBecause zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\n\n\n\n\n\nThe Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.\nIn Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. Leimkuhler and Reich (2004) provide details and error analysis for the leapfrog algorithm.\nA divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,5 the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. Betancourt (2016a) provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.\nThe Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.\nIf the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.\nThe primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.\n\n\nIn some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.\nBefore reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Posterior Analysis"
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "reference-manual/analysis.html#footnotes",
-    "href": "reference-manual/analysis.html#footnotes",
-    "title": "Posterior Analysis",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nUsing vectors simplifies high level exposition at the expense of collapsing structure.↩︎\nThe structure is assumed to be rectangular; in the future, this needs to be generalized to ragged samples.↩︎\nThe lp__ value also represents the potential energy in the Hamiltonian system and is rate bounded by the randomly supplied kinetic energy each iteration, which follows a Chi-square distribution in the number of parameters.↩︎",
+    "objectID": "reference-manual/mcmc.html#hamiltonian-monte-carlo",
+    "href": "reference-manual/mcmc.html#hamiltonian-monte-carlo",
+    "title": "MCMC Sampling",
+    "section": "",
+    "text": "Hamiltonian Monte Carlo (HMC) is a Markov chain Monte Carlo (MCMC) method that uses the derivatives of the density function being sampled to generate efficient transitions spanning the posterior (see, e.g., Betancourt and Girolami (2013), Neal (2011) for more details). It uses an approximate Hamiltonian dynamics simulation based on numerical integration which is then corrected by performing a Metropolis acceptance step.\nThis section translates the presentation of HMC by Betancourt and Girolami (2013) into the notation of Gelman et al. (2013).\n\n\nThe goal of sampling is to draw from a density \\(p(\\theta)\\) for parameters \\(\\theta\\). This is typically a Bayesian posterior \\(p(\\theta|y)\\) given data \\(y\\), and in particular, a Bayesian posterior coded as a Stan program.\n\n\n\nHMC introduces auxiliary momentum variables \\(\\rho\\) and draws from a joint density\n\\[\np(\\rho, \\theta) = p(\\rho | \\theta) p(\\theta).\n\\]\nIn most applications of HMC, including Stan, the auxiliary density is a multivariate normal that does not depend on the parameters \\(\\theta\\),\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\n\\(M\\) is the Euclidean metric. It can be seen as a transform of parameter space that makes sampling more efficient; see Betancourt (2017) for details.\nBy default Stan sets \\(M^{-1}\\) equal to a diagonal estimate of the covariance computed during warmup.\n\n\n\nThe joint density \\(p(\\rho, \\theta)\\) defines a Hamiltonian\n\\[\n\\begin{array}{rcl}\nH(\\rho, \\theta) & = & - \\log p(\\rho, \\theta)\n\\\\[3pt]\n& = & - \\log p(\\rho | \\theta) - \\log p(\\theta).\n\\\\[3pt]\n& = & T(\\rho | \\theta) + V(\\theta),\n\\end{array}\n\\]\nwhere the term\n\\[\nT(\\rho | \\theta) = - \\log p(\\rho | \\theta)\n\\]\nis called the “kinetic energy” and the term\n\\[\nV(\\theta) = - \\log p(\\theta)\n\\]\nis called the “potential energy.” The potential energy is specified by the Stan program through its definition of a log density.\n\n\n\nStarting from the current value of the parameters \\(\\theta\\), a transition to a new state is generated in two stages before being subjected to a Metropolis accept step.\nFirst, a value for the momentum is drawn independently of the current parameter values,\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\]\nThus momentum does not persist across iterations.\nNext, the joint system \\((\\theta,\\rho)\\) made up of the current parameter values \\(\\theta\\) and new momentum \\(\\rho\\) is evolved via Hamilton’s equations,\n\\[\n\\begin{array}{rcccl}\n\\displaystyle\n\\frac{d\\theta}{dt}\n& = &\n\\displaystyle\n+ \\frac{\\partial H}{\\partial \\rho}\n& = &\n\\displaystyle\n+ \\frac{\\partial T}{\\partial \\rho}\n\\\\[12pt]\n\\displaystyle\n\\frac{d\\rho}{dt}\n& = &\n\\displaystyle\n- \\frac{\\partial H}{\\partial \\theta }\n& = &\n\\displaystyle\n- \\frac{\\partial T}{\\partial \\theta}\n- \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nWith the momentum density being independent of the target density, i.e., \\(p(\\rho | \\theta) = p(\\rho)\\), the first term in the momentum time derivative, \\({\\partial T} / {\\partial \\theta}\\) is zero, yielding the pair time derivatives\n\\[\n\\begin{array}{rcl}\n\\frac{d \\theta}{d t} & = & +\\frac{\\partial T}{\\partial \\rho}\n\\\\[2pt]\n\\frac{d \\rho}{d t} & = & -\\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\n\n\n\nThe last section leaves a two-state differential equation to solve. Stan, like most other HMC implementations, uses the leapfrog integrator, which is a numerical integration algorithm that’s specifically adapted to provide stable results for Hamiltonian systems of equations.\nLike most numerical integrators, the leapfrog algorithm takes discrete steps of some small time interval \\(\\epsilon\\). The leapfrog algorithm begins by drawing a fresh momentum term independently of the parameter values \\(\\theta\\) or previous momentum value.\n\\[\n\\rho \\sim \\mathsf{MultiNormal}(0, M).\n\\] It then alternates half-step updates of the momentum and full-step updates of the position.\n\\[\n\\begin{array}{rcl}\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}\n\\\\[6pt]\n\\theta & \\leftarrow\n       & \\theta \\, + \\, \\epsilon \\, M^{-1} \\, \\rho\n\\\\[6pt]\n\\rho & \\leftarrow\n     & \\rho \\, - \\, \\frac{\\epsilon}{2} \\frac{\\partial V}{\\partial \\theta}.\n\\end{array}\n\\]\nBy applying \\(L\\) leapfrog steps, a total of \\(L \\, \\epsilon\\) time is simulated. The resulting state at the end of the simulation (\\(L\\) repetitions of the above three steps) will be denoted \\((\\rho^{*}, \\theta^{*})\\).\nThe leapfrog integrator’s error is on the order of \\(\\epsilon^3\\) per step and \\(\\epsilon^2\\) globally, where \\(\\epsilon\\) is the time interval (also known as the step size); Leimkuhler and Reich (2004) provide a detailed analysis of numerical integration for Hamiltonian systems, including a derivation of the error bound for the leapfrog integrator.\n\n\n\nIf the leapfrog integrator were perfect numerically, there would no need to do any more randomization per transition than generating a random momentum vector. Instead, what is done in practice to account for numerical errors during integration is to apply a Metropolis acceptance step, where the probability of keeping the proposal \\((\\rho^{*}, \\theta^{*})\\) generated by transitioning from \\((\\rho, \\theta)\\) is\n\\[\n\\min \\!\n\\left(\n1,\n\\ \\exp \\! \\left( H(\\rho, \\theta) - H(\\rho^{*}, \\theta^{*}) \\right)\n\\right).\n\\]\nIf the proposal is not accepted, the previous parameter value is returned for the next draw and used to initialize the next iteration.\n\n\n\nThe Hamiltonian Monte Carlo algorithm starts at a specified initial set of parameters \\(\\theta\\); in Stan, this value is either user-specified or generated randomly. Then, for a given number of iterations, a new momentum vector is sampled and the current value of the parameter \\(\\theta\\) is updated using the leapfrog integrator with discretization time \\(\\epsilon\\) and number of steps \\(L\\) according to the Hamiltonian dynamics. Then a Metropolis acceptance step is applied, and a decision is made whether to update to the new state \\((\\theta^{*}, \\rho^{*})\\) or keep the existing state.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Posterior Analysis"
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html",
-    "href": "functions-reference/unbounded_discrete_distributions.html",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/mcmc.html#hmc-algorithm-parameters",
+    "href": "reference-manual/mcmc.html#hmc-algorithm-parameters",
+    "title": "MCMC Sampling",
     "section": "",
-    "text": "The unbounded discrete distributions have support over the natural numbers (i.e., the non-negative integers).\n\n\nFor the negative binomial distribution Stan uses the parameterization described in Gelman et al. (2013). For alternative parameterizations, see section negative binomial glm.\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n\n\\in \\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial}(n~|~\\alpha,\\beta) = \\binom{n +\n\\alpha - 1}{\\alpha - 1} \\, \\left( \\frac{\\beta}{\\beta+1}\n\\right)^{\\!\\alpha} \\, \\left( \\frac{1}{\\beta + 1} \\right)^{\\!n} \\!. \\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial}(\\alpha,\\beta)\\) are given by \\[\\begin{equation*} \\mathbb{E}[n] =\n\\frac{\\alpha}{\\beta} \\ \\ \\text{ and } \\ \\ \\text{Var}[n] =\n\\frac{\\alpha}{\\beta^2} (\\beta + 1). \\end{equation*}\\]\n\n\n\nn ~ neg_binomial(alpha, beta)\nIncrement target log probability density with neg_binomial_lupmf(n | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal neg_binomial_lpmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lupmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal neg_binomial_cdf(ints n | reals alpha, reals beta) The negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal neg_binomial_lcdf(ints n | reals alpha, reals beta) The log of the negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lccdf(ints n | reals alpha, reals beta) The log of the negative binomial complementary cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR neg_binomial_rng(reals alpha, reals beta) Generate a negative binomial variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) beta must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section combinatorial functions, below, provides a second alternative parameterization directly in terms of the log mean.\n\n\nThe first parameterization is for \\(\\mu \\in \\mathbb{R}^+\\) and \\(\\phi \\in\n\\mathbb{R}^+\\), which for \\(n \\in \\mathbb{N}\\) is defined as \\[\\begin{equation*}\n\\text{NegBinomial2}(n \\, | \\, \\mu, \\phi)\n= \\binom{n + \\phi - 1}{n} \\, \\left( \\frac{\\mu}{\\mu+\\phi} \\right)^{\\!n} \\, \\left( \\frac{\\phi}{\\mu+\\phi} \\right)^{\\!\\phi} \\!.\n\\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial2}(n~|~\\mu,\\phi)\\) are \\[\\begin{equation*}\n\\mathbb{E}[n] = \\mu \\ \\ \\ \\text{ and } \\ \\ \\ \\text{Var}[n] = \\mu + \\frac{\\mu^2}{\\phi}.\n\\end{equation*}\\] Recall that \\(\\text{Poisson}(\\mu)\\) has variance \\(\\mu\\), so \\(\\mu^2 / \\phi &gt; 0\\) is the additional variance of the negative binomial above that of the Poisson with mean \\(\\mu\\). So the inverse of parameter \\(\\phi\\) controls the overdispersion, scaled by the square of the mean, \\(\\mu^2\\).\n\n\n\nn ~ neg_binomial_2(mu, phi)\nIncrement target log probability density with neg_binomial_2_lupmf(n | mu, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_lpmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_lupmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_cdf(ints n | reals mu, reals phi) The negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.6\n \n\nreal neg_binomial_2_lcdf(ints n | reals mu, reals phi) The log of the negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nreal neg_binomial_2_lccdf(ints n | reals mu, reals phi) The log of the negative binomial complementary cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nR neg_binomial_2_rng(reals mu, reals phi) Generate a negative binomial variate with location mu and precision phi; may only be used in transformed data and generated quantities blocks. mu must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nRelated to the parameterization in section negative binomial, alternative parameterization, the following parameterization uses a log mean parameter \\(\\eta = \\log(\\mu)\\), defined for \\(\\eta \\in \\mathbb{R}\\), \\(\\phi \\in \\mathbb{R}^+\\), so that for \\(n \\in\n\\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial2Log}(n \\, | \\, \\eta, \\phi) =\n\\text{NegBinomial2}(n | \\exp(\\eta), \\phi). \\end{equation*}\\] This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.\n\n\nn ~ neg_binomial_2_log(eta, phi)\nIncrement target log probability density with neg_binomial_2_log_lupmf(n | eta, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_log_lpmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_log_lupmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nR neg_binomial_2_log_rng(reals eta, reals phi) Generate a negative binomial variate with log-location eta and inverse overdispersion control phi; may only be used in transformed data and generated quantities blocks. eta must be less than \\(29 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e. a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\phi\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{NegBinomial2LogGLM}(y~|~x, \\alpha, \\beta, \\phi) = \\prod_{1\\leq i\n\\leq n}\\text{NegBinomial2}(y_i~|~\\exp(\\alpha_i + x_i\\cdot \\beta),\n\\phi). \\end{equation*}\\]\n\n\n\ny ~ neg_binomial_2_log_glm(x, alpha, beta, phi)\nIncrement target log probability density with neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(\\lambda \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{Poisson}(n|\\lambda) = \\frac{1}{n!} \\, \\lambda^n \\,\n\\exp(-\\lambda). \\end{equation*}\\]\n\n\n\nn ~ poisson(lambda)\nIncrement target log probability density with poisson_lupmf(n | lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_lpmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lupmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal poisson_cdf(ints n | reals lambda) The Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.0\n \n\nreal poisson_lcdf(ints n | reals lambda) The log of the Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lccdf(ints n | reals lambda) The log of the Poisson complementary cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nR poisson_rng(reals lambda) Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than \\(2^{30}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also provides a parameterization of the Poisson using the log rate \\(\\alpha = \\log \\lambda\\) as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{PoissonLog}(n|\\alpha) = \\frac{1}{n!} \\, \\exp \\left(n\\alpha -\n\\exp(\\alpha) \\right). \\end{equation*}\\]\n\n\n\nn ~ poisson_log(alpha)\nIncrement target log probability density with poisson_log_lupmf(n | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_log_lpmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha\nAvailable since 2.12\n \n\nreal poisson_log_lupmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha dropping constant additive terms\nAvailable since 2.25\n \n\nR poisson_log_rng(reals alpha) Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than \\(30 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e. a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{PoissonLogGLM}(y|x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Poisson}(y_i|\\exp(\\alpha_i + x_i\\cdot \\beta)). \\end{equation*}\\]\n\n\n\ny ~ poisson_log_glm(x, alpha, beta)\nIncrement target log probability density with poisson_log_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(r \\in \\mathbb{R}^+\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{BetaNegBinomial}(n|r,\\alpha,\\beta) = \\frac {\\Gamma (n+r )}{n!\\;\\Gamma (r )}\n\\frac {\\mathrm {B} (\\beta+n,\\alpha +r )}{\\mathrm {B} (\\beta,\\alpha )}. \\end{equation*}\\]\n\n\n\nn ~ beta_neg_binomial(r,alpha,beta)\nIncrement target log probability density with beta_neg_binomial_lupmf(n | r, alpha, beta).\nAvailable since 2.36\n \n\n\n\n\n \n\nreal beta_neg_binomial_lpmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lupmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta dropping constant additive terms.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_cdf(ints n | reals r, reals alpha, reals beta) The beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lcdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lccdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial complementary cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nR beta_neg_binomial_rng(reals r, reals alpha, reals beta) Generate a beta negative binomial variate with parameters r, alpha and beta; may only be used in transformed data and generated quantities blocks. r \\(\\cdot\\) beta \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.36",
+    "text": "The Hamiltonian Monte Carlo algorithm has three parameters which must be set,\n\ndiscretization time \\(\\epsilon\\),\nmetric \\(M\\), and\nnumber of steps taken \\(L\\).\n\nIn practice, sampling efficiency, both in terms of iteration speed and iterations per effective sample, is highly sensitive to these three tuning parameters Neal (2011), Hoffman and Gelman (2014).\nIf \\(\\epsilon\\) is too large, the leapfrog integrator will be inaccurate and too many proposals will be rejected. If \\(\\epsilon\\) is too small, too many small steps will be taken by the leapfrog integrator leading to long simulation times per interval. Thus the goal is to balance the acceptance rate between these extremes.\nIf \\(L\\) is too small, the trajectory traced out in each iteration will be too short and sampling will devolve to a random walk. If \\(L\\) is too large, the algorithm will do too much work on each iteration.\nIf the inverse metric \\(M^{-1}\\) is a poor estimate of the posterior covariance, the step size \\(\\epsilon\\) must be kept small to maintain arithmetic precision. This would lead to a large \\(L\\) to compensate.\n\n\nThe actual integration time is \\(L \\, \\epsilon\\), a function of number of steps. Some interfaces to Stan set an approximate integration time \\(t\\) and the discretization interval (step size) \\(\\epsilon\\). In these cases, the number of steps will be rounded down as\n\\[\nL = \\left\\lfloor \\frac{t}{\\epsilon} \\right\\rfloor.\n\\]\nand the actual integration time will still be \\(L \\, \\epsilon\\).\n\n\n\nStan is able to automatically optimize \\(\\epsilon\\) to match an acceptance-rate target, able to estimate \\(M\\) based on warmup sample iterations, and able to dynamically adapt \\(L\\) on the fly during sampling (and during warmup) using the no-U-turn sampling (NUTS) algorithm Hoffman and Gelman (2014).\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\nWhen adaptation is engaged (it may be turned off by fixing a step size and metric), the warmup period is split into three stages, as illustrated in the warmup adaptation figure, with two fast intervals surrounding a series of growing slow intervals. Here fast and slow refer to parameters that adapt using local and global information, respectively; the Hamiltonian Monte Carlo samplers, for example, define the step size as a fast parameter and the (co)variance as a slow parameter. The size of the the initial and final fast intervals and the initial size of the slow interval are all customizable, although user-specified values may be modified slightly in order to ensure alignment with the warmup period.\nThe motivation behind this partitioning of the warmup period is to allow for more robust adaptation. The stages are as follows.\n\nIn the initial fast interval the chain is allowed to converge towards the typical set,1 with only parameters that can learn from local information adapted.\nAfter this initial stage parameters that require global information, for example (co)variances, are estimated in a series of expanding, memoryless windows; often fast parameters will be adapted here as well.\nLastly, the fast parameters are allowed to adapt to the final update of the slow parameters.\n\nThese intervals may be controlled through the following configuration parameters, all of which must be positive integers:\nAdaptation Parameters Table. The parameters controlling adaptation and their default values.\n\n\n\n\n\n\n\n\nparameter\ndescription\ndefault\n\n\n\n\ninitial buffer\nwidth of initial fast adaptation interval\n75\n\n\nterm buffer\nwidth of final fast adaptation interval\n50\n\n\nwindow\ninitial width of slow adaptation interval\n25\n\n\n\n\n\n\nStan’s HMC algorithms utilize dual averaging Nesterov (2009) to optimize the step size.2\nThis warmup optimization procedure is extremely flexible and for completeness, Stan exposes each tuning option for dual averaging, using the notation of Hoffman and Gelman (2014). In practice, the efficacy of the optimization is sensitive to the value of these parameters, but we do not recommend changing the defaults without experience with the dual-averaging algorithm. For more information, see the discussion of dual averaging in Hoffman and Gelman (2014).\nThe full set of dual-averaging parameters are:\nStep Size Adaptation Parameters Table The parameters controlling step size adaptation, with constraints and default values.\n\n\n\nparameter\ndescription\nconstraint\ndefault\n\n\n\n\ndelta\ntarget Metropolis acceptance rate\n[0, 1]\n0.8\n\n\ngamma\nadaptation regularization scale\n(0, infty)\n0.05\n\n\nkappa\nadaptation relaxation exponent\n(0, infty)\n0.75\n\n\nt_0\nadaptation iteration offset\n(0, infty)\n10\n\n\n\nBy setting the target acceptance parameter \\(\\delta\\) to a value closer to 1 (its value must be strictly less than 1 and its default value is 0.8), adaptation will be forced to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of \\(\\delta\\) will also allow some models that would otherwise get stuck to overcome their blockages.\n\n\n\nAll implementations of HMC use numerical integrators requiring a step size (equivalently, discretization time interval). Stan allows the step size to be adapted or set explicitly. Stan also allows the step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. The jitter is a proportion that may be added or subtracted, so the maximum amount of jitter is 1, which will cause step sizes to be selected in the range of 0 to twice the adapted step size. The default value is 0, producing no jitter.\nSmall step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes. The downside is that jittering below the adapted value will increase the number of leapfrog steps required and thus slow down iterations, whereas jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. See Neal (2011) for further discussion of step-size jittering.\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nIf the metric is constant then the resulting implementation is known as Euclidean HMC. Stan allows a choice among three Euclidean HMC implementations,\n\na unit metric (diagonal matrix of ones),\na diagonal metric (diagonal matrix with positive diagonal entries), and\na dense metric (a dense, symmetric positive definite matrix)\n\nto be configured by the user.\nIf the metric is specified to be diagonal, then regularized variances are estimated based on the iterations in each slow-stage block (labeled II in the warmup adaptation stages figure). Each of these estimates is based only on the iterations in that block. This allows early estimates to be used to help guide warmup and then be forgotten later so that they do not influence the final covariance estimate.\nIf the metric is specified to be dense, then regularized covariance estimates will be carried out, regularizing the estimate to a diagonal matrix, which is itself regularized toward a unit matrix.\nVariances or covariances are estimated using Welford accumulators to avoid a loss of precision over many floating point operations.\n\n\nThe metric can compensate for linear (i.e. global) correlations in the posterior which can dramatically improve the performance of HMC in some problems. This requires knowing the global correlations.\nIn complex models, the global correlations are usually difficult, if not impossible, to derive analytically; for example, nonlinear model components convolve the scales of the data, so standardizing the data does not always help. Therefore, Stan estimates these correlations online with an adaptive warmup. In models with strong nonlinear (i.e. local) correlations this learning can be slow, even with regularization. This is ultimately why warmup in Stan often needs to be so long, and why a sufficiently long warmup can yield such substantial performance improvements.\n\n\n\nThe metric compensates for only linear (equivalently global or position-independent) correlations in the posterior. The hierarchical parameterizations, on the other hand, affect some of the nasty nonlinear (equivalently local or position-dependent) correlations common in hierarchical models.3\nOne of the biggest difficulties with dense metrics is the estimation of the metric itself which introduces a bit of a chicken-and-egg scenario; in order to estimate an appropriate metric for sampling, convergence is required, and in order to converge, an appropriate metric is required.\n\n\n\nStatistical models for which sampling is problematic are not typically dominated by linear correlations for which a dense metric can adjust. Rather, they are governed by more complex nonlinear correlations that are best tackled with better parameterizations or more advanced algorithms, such as Riemannian HMC.\n\n\n\nMCMC convergence time is roughly equivalent to the autocorrelation time. Because HMC (and NUTS) chains tend to be lowly autocorrelated they also tend to converge quite rapidly.\nThis only applies when there is uniformity of curvature across the posterior, an assumption which is violated in many complex models. Quite often, the tails have large curvature while the bulk of the posterior mass is relatively well-behaved; in other words, warmup is slow not because the actual convergence time is slow but rather because the cost of an HMC iteration is more expensive out in the tails.\nPoor behavior in the tails is the kind of pathology that can be uncovered by running only a few warmup iterations. By looking at the acceptance probabilities and step sizes of the first few iterations provides an idea of how bad the problem is and whether it must be addressed with modeling efforts such as tighter priors or reparameterizations.\n\n\n\n\nThe no-U-turn sampler (NUTS) automatically selects an appropriate number of leapfrog steps in each iteration in order to allow the proposals to traverse the posterior without doing unnecessary work. The motivation is to maximize the expected squared jump distance (see, e.g., Roberts, Gelman, and Gilks (1997)) at each step and avoid the random-walk behavior that arises in random-walk Metropolis or Gibbs samplers when there is correlation in the posterior. For a precise definition of the NUTS algorithm and a proof of detailed balance, see Hoffman and Gelman (2014).\nNUTS generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then generates an independent standard normal random momentum vector. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. At each iteration of the NUTS algorithm the tree depth is increased by one, doubling the number of leapfrog steps and effectively doubles the computation time. The algorithm terminates in one of two ways, either\n\nthe NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree, or\nthe depth of the completed tree hits the maximum depth allowed.\n\nRather than using a standard Metropolis step, the final parameter value is selected via multinomial sampling with a bias toward the second half of the steps in the trajectory Betancourt (2016b).4\nConfiguring the no-U-turn sample involves putting a cap on the depth of the trees that it evaluates during each iteration. This is controlled through a maximum depth parameter. The number of leapfrog steps taken is then bounded by 2 to the power of the maximum depth minus 1.\nBoth the tree depth and the actual number of leapfrog steps computed are reported along with the parameters in the output as treedepth__ and n_leapfrog__, respectively. Because the final subtree may only be partially constructed, these two will always satisfy\n\\[\n2^{\\mathrm{treedepth} - 1} - 1\n\\ &lt; \\\nN_{\\mathrm{leapfrog}}\n\\ \\le \\\n2^{\\mathrm{treedepth} } - 1.\n\\]\nTree depth is an important diagnostic tool for NUTS. For example, a tree depth of zero occurs when the first leapfrog step is immediately rejected and the initial state returned, indicating extreme curvature and poorly-chosen step size (at least relative to the current position). On the other hand, a tree depth equal to the maximum depth indicates that NUTS is taking many leapfrog steps and being terminated prematurely to avoid excessively long execution time. Taking very many steps may be a sign of poor adaptation, may be due to targeting a very high acceptance rate, or may simply indicate a difficult posterior from which to sample. In the latter case, reparameterization may help with efficiency. But in the rare cases where the model is correctly specified and a large number of steps is necessary, the maximum depth should be increased to ensure that that the NUTS tree can grow as large as necessary.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#negative-binomial-distribution",
-    "href": "functions-reference/unbounded_discrete_distributions.html#negative-binomial-distribution",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/mcmc.html#sampling-without-parameters",
+    "href": "reference-manual/mcmc.html#sampling-without-parameters",
+    "title": "MCMC Sampling",
     "section": "",
-    "text": "For the negative binomial distribution Stan uses the parameterization described in Gelman et al. (2013). For alternative parameterizations, see section negative binomial glm.\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n\n\\in \\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial}(n~|~\\alpha,\\beta) = \\binom{n +\n\\alpha - 1}{\\alpha - 1} \\, \\left( \\frac{\\beta}{\\beta+1}\n\\right)^{\\!\\alpha} \\, \\left( \\frac{1}{\\beta + 1} \\right)^{\\!n} \\!. \\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial}(\\alpha,\\beta)\\) are given by \\[\\begin{equation*} \\mathbb{E}[n] =\n\\frac{\\alpha}{\\beta} \\ \\ \\text{ and } \\ \\ \\text{Var}[n] =\n\\frac{\\alpha}{\\beta^2} (\\beta + 1). \\end{equation*}\\]\n\n\n\nn ~ neg_binomial(alpha, beta)\nIncrement target log probability density with neg_binomial_lupmf(n | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal neg_binomial_lpmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lupmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal neg_binomial_cdf(ints n | reals alpha, reals beta) The negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal neg_binomial_lcdf(ints n | reals alpha, reals beta) The log of the negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lccdf(ints n | reals alpha, reals beta) The log of the negative binomial complementary cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR neg_binomial_rng(reals alpha, reals beta) Generate a negative binomial variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) beta must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "text": "In some situations, such as pure forward data simulation in a directed graphical model (e.g., where you can work down generatively from known hyperpriors to simulate parameters and data), there is no need to declare any parameters in Stan, the model block will be empty (and thus can be omitted), and all output quantities will be produced in the generated quantities block.\nFor example, to generate a sequence of \\(N\\) draws from a binomial with trials \\(K\\) and chance of success \\(\\theta\\), the following program suffices.\ndata {\n  real&lt;lower=0, upper=1&gt; theta;\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n}\ngenerated quantities {\n  array[N] int&lt;lower=0, upper=K&gt; y;\n  for (n in 1:N) {\n    y[n] = binomial_rng(K, theta);\n  }\n}\nFor this model, the sampler must be configured to use the fixed-parameters setting because there are no parameters. Without parameter sampling there is no need for adaptation and the number of warmup iterations should be set to zero.\nMost models that are written to be sampled without parameters will not declare any parameters, instead putting anything parameter-like in the data block. Nevertheless, it is possible to include parameters for fixed-parameters sampling and initialize them in any of the usual ways (randomly, fixed to zero on the unconstrained scale, or with user-specified values). For example, theta in the example above could be declared as a parameter and initialized as a parameter.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#nbalt",
-    "href": "functions-reference/unbounded_discrete_distributions.html#nbalt",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/mcmc.html#general-config.section",
+    "href": "reference-manual/mcmc.html#general-config.section",
+    "title": "MCMC Sampling",
     "section": "",
-    "text": "Stan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section combinatorial functions, below, provides a second alternative parameterization directly in terms of the log mean.\n\n\nThe first parameterization is for \\(\\mu \\in \\mathbb{R}^+\\) and \\(\\phi \\in\n\\mathbb{R}^+\\), which for \\(n \\in \\mathbb{N}\\) is defined as \\[\\begin{equation*}\n\\text{NegBinomial2}(n \\, | \\, \\mu, \\phi)\n= \\binom{n + \\phi - 1}{n} \\, \\left( \\frac{\\mu}{\\mu+\\phi} \\right)^{\\!n} \\, \\left( \\frac{\\phi}{\\mu+\\phi} \\right)^{\\!\\phi} \\!.\n\\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial2}(n~|~\\mu,\\phi)\\) are \\[\\begin{equation*}\n\\mathbb{E}[n] = \\mu \\ \\ \\ \\text{ and } \\ \\ \\ \\text{Var}[n] = \\mu + \\frac{\\mu^2}{\\phi}.\n\\end{equation*}\\] Recall that \\(\\text{Poisson}(\\mu)\\) has variance \\(\\mu\\), so \\(\\mu^2 / \\phi &gt; 0\\) is the additional variance of the negative binomial above that of the Poisson with mean \\(\\mu\\). So the inverse of parameter \\(\\phi\\) controls the overdispersion, scaled by the square of the mean, \\(\\mu^2\\).\n\n\n\nn ~ neg_binomial_2(mu, phi)\nIncrement target log probability density with neg_binomial_2_lupmf(n | mu, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_lpmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_lupmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_cdf(ints n | reals mu, reals phi) The negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.6\n \n\nreal neg_binomial_2_lcdf(ints n | reals mu, reals phi) The log of the negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nreal neg_binomial_2_lccdf(ints n | reals mu, reals phi) The log of the negative binomial complementary cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nR neg_binomial_2_rng(reals mu, reals phi) Generate a negative binomial variate with location mu and precision phi; may only be used in transformed data and generated quantities blocks. mu must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "text": "Stan’s interfaces provide a number of configuration options that are shared among the MCMC algorithms (this chapter), the optimization algorithms chapter, and the diagnostics chapter.\n\n\nThe random-number generator’s behavior is fully determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nStan also allows a chain identifier to be specified, which is useful when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. When running multiple chains from a single command, Stan’s interfaces will manage the chain identifiers.\n\n\nTogether, the seed and chain identifier determine the behavior of the underlying random number generator. For complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries.\n\n\n\n\nThe initial parameter values for Stan’s algorithms (MCMC, optimization, or diagnostic) may be either specified by the user or generated randomly. If user-specified values are provided, all parameters must be given initial values or Stan will abort with an error message.\n\n\nIf the user specifies initial values, they must satisfy the constraints declared in the model (i.e., they are on the constrained scale).\n\n\n\nIt is also possible to provide an initialization of 0, which causes all variables to be initialized with zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations for most parameters, such as 0 for unconstrained parameters, 1 for parameters constrained to be positive, 0.5 for variables to constrained to lie between 0 and 1, a symmetric (uniform) vector for simplexes, unit matrices for both correlation and covariance matrices, and so on.\n\n\n\nRandom initialization by default initializes the parameter values with values drawn at random from a \\(\\mathsf{Uniform}(-2, 2)\\) distribution. Alternatively, a value other than 2 may be specified for the absolute bounds. These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters.\nBecause zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log",
-    "href": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/mcmc.html#divergent-transitions",
+    "href": "reference-manual/mcmc.html#divergent-transitions",
+    "title": "MCMC Sampling",
     "section": "",
-    "text": "Related to the parameterization in section negative binomial, alternative parameterization, the following parameterization uses a log mean parameter \\(\\eta = \\log(\\mu)\\), defined for \\(\\eta \\in \\mathbb{R}\\), \\(\\phi \\in \\mathbb{R}^+\\), so that for \\(n \\in\n\\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial2Log}(n \\, | \\, \\eta, \\phi) =\n\\text{NegBinomial2}(n | \\exp(\\eta), \\phi). \\end{equation*}\\] This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.\n\n\nn ~ neg_binomial_2_log(eta, phi)\nIncrement target log probability density with neg_binomial_2_log_lupmf(n | eta, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_log_lpmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_log_lupmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nR neg_binomial_2_log_rng(reals eta, reals phi) Generate a negative binomial variate with log-location eta and inverse overdispersion control phi; may only be used in transformed data and generated quantities blocks. eta must be less than \\(29 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "text": "The Hamiltonian Monte Carlo algorithms (HMC and NUTS) simulate the trajectory of a fictitious particle representing parameter values when subject to a potential energy field, the value of which at a point is the negative log posterior density (up to a constant that does not depend on location). Random momentum is imparted independently in each direction, by drawing from a standard normal distribution. The Hamiltonian is defined to be the sum of the potential energy and kinetic energy of the system. The key feature of the Hamiltonian is that it is conserved along the trajectory the particle moves.\nIn Stan, we use the leapfrog algorithm to simulate the path of a particle along the trajectory defined by the initial random momentum and the potential energy field. This is done by alternating updates of the position based on the momentum and the momentum based on the position. The momentum updates involve the potential energy and are applied along the gradient. This is essentially a stepwise (discretized) first-order approximation of the trajectory. Leimkuhler and Reich (2004) provide details and error analysis for the leapfrog algorithm.\nA divergence arises when the simulated Hamiltonian trajectory departs from the true trajectory as measured by departure of the Hamiltonian value from its initial value. When this divergence is too high,5 the simulation has gone off the rails and cannot be trusted. The positions along the simulated trajectory after the Hamiltonian diverges will never be selected as the next draw of the MCMC algorithm, potentially reducing Hamiltonian Monte Carlo to a simple random walk and biasing estimates by not being able to thoroughly explore the posterior distribution. Betancourt (2016a) provides details of the theory, computation, and practical implications of divergent transitions in Hamiltonian Monte Carlo.\nThe Stan interfaces report divergences as warnings and provide ways to access which iterations encountered divergences. ShinyStan provides visualizations that highlight the starting point of divergent transitions to diagnose where the divergences arise in parameter space. A common location is in the neck of the funnel in a centered parameterization, an example of which is provided in the user’s guide.\nIf the posterior is highly curved, very small step sizes are required for this gradient-based simulation of the Hamiltonian to be accurate. When the step size is too large (relative to the curvature), the simulation diverges from the true Hamiltonian. This definition is imprecise in the same way that stiffness for a differential equation is imprecise; both are defined by the way they cause traditional stepwise algorithms to diverge from where they should be.\nThe primary cause of divergent transitions in Euclidean HMC (other than bugs in the code) is highly varying posterior curvature, for which small step sizes are too inefficient in some regions and diverge in other regions. If the step size is too small, the sampler becomes inefficient and halts before making a U-turn (hits the maximum tree depth in NUTS); if the step size is too large, the Hamiltonian simulation diverges.\n\n\nIn some cases, simply lowering the initial step size and increasing the target acceptance rate will keep the step size small enough that sampling can proceed. In other cases, a reparameterization is required so that the posterior curvature is more manageable; see the funnel example in the user’s guide for an example.\nBefore reparameterization, it may be helpful to plot the posterior draws, highlighting the divergent transitions to see where they arise. This is marked as a divergent transition in the interfaces; for example, ShinyStan and RStan have special plotting facilities to highlight where divergent transitions arise.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log-glm",
-    "href": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log-glm",
-    "title": "Unbounded Discrete Distributions",
-    "section": "",
-    "text": "Stan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e. a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\phi\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{NegBinomial2LogGLM}(y~|~x, \\alpha, \\beta, \\phi) = \\prod_{1\\leq i\n\\leq n}\\text{NegBinomial2}(y_i~|~\\exp(\\alpha_i + x_i\\cdot \\beta),\n\\phi). \\end{equation*}\\]\n\n\n\ny ~ neg_binomial_2_log_glm(x, alpha, beta, phi)\nIncrement target log probability density with neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25",
+    "objectID": "reference-manual/mcmc.html#footnotes",
+    "href": "reference-manual/mcmc.html#footnotes",
+    "title": "MCMC Sampling",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe typical set is a concept borrowed from information theory and refers to the neighborhood (or neighborhoods in multimodal models) of substantial posterior probability mass through which the Markov chain will travel in equilibrium.↩︎\nThis optimization of step size during adaptation of the sampler should not be confused with running Stan’s optimization method.↩︎\nIn Riemannian HMC the metric compensates for nonlinear correlations.↩︎\nStan previously used slice sampling along the trajectory, following the original NUTS paper of Hoffman and Gelman (2014).↩︎\nThe current default threshold is a factor of \\(10^3\\), whereas when the leapfrog integrator is working properly, the divergences will be around \\(10^{-7}\\) and do not compound due to the symplectic nature of the leapfrog integrator.↩︎",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "MCMC Sampling"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson",
-    "href": "functions-reference/unbounded_discrete_distributions.html#poisson",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/laplace_embedded.html",
+    "href": "reference-manual/laplace_embedded.html",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "If \\(\\lambda \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{Poisson}(n|\\lambda) = \\frac{1}{n!} \\, \\lambda^n \\,\n\\exp(-\\lambda). \\end{equation*}\\]\n\n\n\nn ~ poisson(lambda)\nIncrement target log probability density with poisson_lupmf(n | lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_lpmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lupmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal poisson_cdf(ints n | reals lambda) The Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.0\n \n\nreal poisson_lcdf(ints n | reals lambda) The log of the Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lccdf(ints n | reals lambda) The log of the Poisson complementary cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nR poisson_rng(reals lambda) Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than \\(2^{30}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "text": "The embedded Laplace approximation replaces explicit sampling of potentially high-dimensional Gaussian latent variables with a local Gaussian approximation, marginalizing them out so that inference proceeds over the remaining hyperparameters alone. This approach is often referred to as the integrated Laplace approximation, although the exact details of the method can vary. The details of Stan’s implementation can be found in references (Margossian et al. 2020; Margossian 2023).\nA standard approach to fit a latent Gaussian model would be to perform inference jointly over the latent Gaussian variables and the hyperparameters. Instead, the embedded Laplace approximation can be used to do approximate marginalization of the latent Gaussian variables; we can then use any inference over the remaining hyperparameters. By marginalizing out the latent variables, the sampler explores a lower-dimensional, better-behaved marginal posterior. Individual iterations are more expensive (each requires an inner optimization), but the sampler typically needs far fewer iterations to achieve the same effective sample size. How this trade-off resolves depends on the specific problem at hand.\nFor complete function signatures and the built-in likelihood wrappers (Poisson, Negative Binomial, Bernoulli), see the Embedded Laplace functions reference. For worked examples with full data blocks, see the Gaussian Processes chapter.\n\n\nThe embedded Laplace approximation is used for latent Gaussian models. A latent Gaussian model is defined by three components:\n\n\\(\\phi\\): hyperparameters (e.g., GP kernel length-scale and magnitude, or variance components in a hierarchical model),\n\\(\\theta\\): latent Gaussian variables (the high-dimensional quantity to be marginalized out),\n\\(y\\): observed data.\n\nThese components are related through a hierarchical structure. The hyperparameters \\(\\phi\\) are given a prior \\(p(\\phi)\\). The latent variables \\(\\theta\\) have a multivariate normal prior centered at 0 with covariance matrix \\(K(\\phi)\\). An non-zero mean offset can be incorporated into the likelihood function. The observations \\(y\\) have a data model \\(p(y \\mid \\theta, \\phi)\\). The prior on \\(\\theta\\) is centered at zero; an offset can be incorporated into the data model if a non-zero mean is needed.\n\\[\\begin{eqnarray*}\n  \\phi & \\sim & p(\\phi) \\\\\n  \\theta & \\sim & \\text{Multi-Normal}(0, K(\\phi)) \\\\\n  y & \\sim & p(y \\mid \\theta, \\phi).\n\\end{eqnarray*}\\]\nThe generative model above defines a joint distribution over all three quantities, \\(p(\\phi, \\theta, y) = p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\). After observing data \\(y\\), Bayes’ theorem gives the joint posterior \\(p(\\phi, \\theta \\mid y) \\propto p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\), where \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) and \\(\\phi\\) is the joint likelihood function.\nSampling directly from the joint posterior \\(p(\\phi, \\theta \\mid y)\\) of this model is often difficult. Challenging geometries (e.g., funnels) frustrate inference algorithms, including Hamiltonian Monte Carlo and variational inference. However, the marginal posterior \\(p(\\phi \\mid y)\\) is often well-behaved and low-dimensional, making it much easier to sample. With an embedded Laplace approximation, we can obtain an approximation of the marginal posterior \\(p(\\phi \\mid y)\\). This is done via an intermediate approximation of the conditional posterior \\(p(\\theta \\mid \\phi, y)\\) by a normal distribution and this normal approximation is well-justified when the likelihood \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) is log concave (given that we already have a normal prior \\(p(\\theta \\mid \\phi)\\)). Once we obtain (approximate) samples \\(\\phi^{(i)} \\sim p(\\phi \\mid y)\\), we can in turn generate posterior draws for \\(\\theta\\) using the normal approximation to \\(p(\\theta \\mid y, \\phi)\\), evaluated at the posterior draws \\(\\phi^{(i)}\\).\n\n\n\nThe two-step inference strategy for using embedded Laplace in a latent Gaussian model requires approximations to both the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and the marginal likelihood \\(p(y \\mid \\phi)\\). The Laplace approximation is the normal distribution that matches the mode and curvature of the conditional posterior \\(p(\\theta \\mid y, \\phi)\\). The mode, defined as the value of \\(\\theta\\) that maximizes the conditional posterior, is estimated by a Newton solver, \\[\n  \\theta^* = \\underset{\\theta}{\\text{argmax}} \\ p(\\theta \\mid y, \\phi),\n\\]\nSince the approximation is normal, the curvature is matched by setting the covariance to the negative Hessian of the log conditional posterior, evaluated at the mode,\n\\[\n  \\Sigma^* = -  \\left . \\frac{\\partial^2}{\\partial \\theta^2}\n    \\log p (\\theta \\mid \\phi, y) \\right |_{\\theta =\\theta^*}.\n\\]\nThe resulting Laplace approximation is a multivariate normal centered at the mode with covariance given by the inverse curvature,\n\\[\n\\hat p_\\mathcal{L} (\\theta \\mid y, \\phi) = \\text{Multi-Normal}(\\theta^*, \\Sigma^*)\n\\approx p(\\theta \\mid y, \\phi).\n\\]\nThis approximation also yields an approximation to the marginal likelihood, obtained by evaluating the prior, likelihood, and approximate posterior at the mode \\(\\theta^*\\),\n\\[\n   \\hat p_\\mathcal{L}(y \\mid \\phi) := \\frac{p(\\theta^* \\mid \\phi) \\\n   p(y \\mid \\theta^*, \\phi) }{ \\hat p_\\mathcal{L} (\\theta^* \\mid \\phi, y) }\n   \\approx p(y \\mid \\phi).\n\\]\nHence, a strategy to approximate the posterior of the latent Gaussian model is to first estimate the marginal posterior \\(\\hat p_\\mathcal{L}(\\phi \\mid y) \\propto p(\\phi) p_\\mathcal{L} (y \\mid \\phi)\\) using any algorithm supported by Stan. Approximate posterior draws for the latent Gaussian variables are then obtained by first sampling \\(\\phi \\sim \\hat p_\\mathcal{L}(\\phi \\mid y)\\) and then \\(\\theta \\sim  \\hat p_\\mathcal{L}(\\theta \\mid \\phi, y)\\).\n\n\n\nThe embedded Laplace approximation presents several trade-offs with standard inference over the joint posterior \\(p(\\theta, \\phi \\mid y)\\). The main advantage of the embedded Laplace approximation is that it side-steps the intricate geometry of hierarchical models. The marginal posterior \\(p(\\phi \\mid y)\\) can then be handled by Hamiltonian Monte Carlo sampling without extensive tuning or reparameterization, and the mixing time is faster, meaning we can run shorter chains to achieve a desired precision. One additional benefit is that approximate methods, e.g. variational inference, which work poorly on the joint \\(p(\\theta, \\phi \\mid y)\\) can work well on the marginal posterior \\(p(\\phi \\mid y)\\).\nOn the other hand, the embedded Laplace approximation presents certain disadvantages. First, we need to perform a Laplace approximation each time the log marginal likelihood is evaluated, meaning each iteration can be expensive. Secondly, the approximation can introduce non-negligible error, especially with non-log-concave likelihoods (note the prior is always multivariate normal). How these trade-offs are resolved depends on the application; see Margossian et al. (2020) for some examples.\n\n\nThe quality of the Laplace approximation depends on how close the true conditional posterior \\(p(\\theta \\mid y, \\phi)\\) is to Gaussian.\nWorks well. Log-concave likelihoods, for example from a Poisson model with log link or negative binomial with log link. These produce unimodal conditional posteriors when combined with a Gaussian prior. The approximation error is typically negligible with these likelihoods, especially with moderate-to-large counts (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). If the likelihood is normal, there is no error in the approximation but in this case the marginalization can be worked analytically and the resulting implementation is much faster than using the embedded Laplace approximation.\nWorks adequately. Bernoulli model with logit link has technically log-concave likelihood, but the likelihood can be very skewed making the Gaussian approximation less accurate than for count data. The embedded Laplace is still useful when \\(\\theta\\) is high-dimensional and joint sampling is infeasible; see Vehtari et al. (2016) and Margossian et al. (2020) for discussion.\nNot appropriate. For likelihoods that are not log-concave in \\(\\theta\\), the conditional posterior may be multimodal and the Newton solver finds only a single mode or can fail completely. When \\(\\theta\\) is low-dimensional (a few dozen or fewer), the overhead of the inner optimization may not pay for itself and standard joint HMC sampling is often adequate.\n\n\n\n\nWhen the embedded Laplace approximation does not converge or produces unexpected results, the solver configuration may need adjustment. This section describes the internals of the Newton solver and the options available for tuning it.\n\n\nA critical component of the embedded Laplace approximation is the Newton solver used to estimate the mode \\(\\theta^*\\) of \\(p(\\theta \\mid \\phi, y)\\). The objective function being maximized is the log joint density of the prior and likelihood with respect to \\(\\theta\\).\n\\[\n\\Psi(\\theta) = \\log p(\\theta \\mid \\phi) + \\log p(y \\mid \\theta, \\phi),\n\\]\nConvergence is declared when the change in the objective between successive iterations falls below a tolerance \\(\\Delta\\).\n\\[\n| \\Psi (\\theta^{(i + 1)}) - \\Psi (\\theta^{(i)}) | \\le \\Delta.\n\\]\nThe solver also stops after reaching a pre-specified maximum number of steps. In that case, Stan throws a warning, but still returns the last iteration’s parameters. If you see this warning you should check the diagnostics to understand why the solver failed to converge.\nTo help with cases where the Newton step does not lead to a decrease in the objective function, the Newton iteration is augmented with a wolfe line-search to ensure that at each iteration the objective function \\(\\Psi\\) decreases. Specifically, suppose the objective increases after a Newton step, indicating the step overshot a region of improvement,\n\\[\n\\Psi (\\theta^{(i + 1)}) &lt; \\Psi (\\theta^{(i)}).\n\\]\nThis can indicate that the Newton step \\(\\alpha\\) at iteration \\(i\\) is too large and that we skipped a region where the objective function decreases. In that case, we can fallback to a Wolfe line search to find a step size which satisfies the Wolfe conditions. The wolfe line search attempts to find a search direction \\(p_i\\) and step size \\(\\alpha_k\\) such that an accepted step both increases our objective while ensuring the slope of the accepted step is flatter than our previous position. Together these help push the algorithm towards a minimum.\n\\[\nf(x_i + \\alpha_k p_i) \\le f(x_i) + c_1 \\alpha_k \\nabla f(x_i)^T p_i\n-p^T_i \\Delta f(x_i + \\alpha_k p_i) \\le -c_2 p^T_i \\Delta f(x_i)\n\\]\n\\[\n  \\theta^{(i + 1)} \\leftarrow \\frac{\\theta^{(i + 1)} + \\theta^{(i)}}{2}.\n\\]\nWe repeat this halving of steps until \\(\\Psi (\\theta^{(i + 1)}) \\ge \\Psi (\\theta^{(i)})\\), or until a maximum number of linesearch steps is reached. For certain problems, adding a linesearch can make the optimization more stable.\n\n\n\nThe embedded Laplace approximation uses a custom Newton solver, specialized to find the mode of \\(p(\\theta \\mid \\phi, y)\\). A key step for efficient optimization is to ensure all matrix inversions are numerically stable. This can be done using the Woodbury-Sherman-Morrison formula and requires one of three matrix decompositions:\n\nCholesky decomposition of the Hessian of the negative log likelihood \\(W = - \\partial^2_\\theta \\log p(y \\mid \\theta, \\phi)\\).\nCholesky decomposition of the prior covariance matrix \\(K(\\phi)\\).\nLU-decomposition of \\(I + KW\\), where \\(I\\) is the identity matrix.\n\nThe first solver (1) should be used if the negative log likelihood is positive-definite. Otherwise the user should rely on (2). In rarer cases where it is not numerically safe to invert the covariance matrix \\(K\\), users can use the third solver as a last-resort option.\n\n\n\nA key step to speed up computation is to take advantage of the sparsity of \\(H\\), the Hessian of the log likelihood with respect to the latent variables, \\[\n  H = \\frac{\\partial^2}{\\partial \\theta^2} \\log p(y \\mid \\theta, \\phi).\n\\] For example, if the observations \\((y_1, \\cdots, y_N)\\) are conditionally independent and each depends on only one component of \\(\\theta\\), the log likelihood decomposes into a sum of per-observation terms, \\[\n  \\log p(y \\mid \\theta, \\phi) = \\sum_{i = 1}^N \\log p(y_i \\mid \\theta_i, \\phi),\n\\] and the Hessian is diagonal. This leads to faster calculations of the Hessian and subsequently sparse matrix operations. This case is common in Gaussian process models, and certain hierarchical models.\nStan’s suite of functions for the embedded Laplace approximation exploits block-diagonal structure in the Hessian, where the user specifies the block size B. The user can specify the size \\(B\\) of these blocks. The user is responsible for working out what \\(B\\) is. If the Hessian is dense, then we simply set \\(B = N\\). The diagonal case above corresponds to B = 1. Arbitrary sparsity patterns beyond block-diagonal structure are not currently supported.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson-distribution-log-parameterization",
-    "href": "functions-reference/unbounded_discrete_distributions.html#poisson-distribution-log-parameterization",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/laplace_embedded.html#latent-gaussian-models",
+    "href": "reference-manual/laplace_embedded.html#latent-gaussian-models",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "Stan also provides a parameterization of the Poisson using the log rate \\(\\alpha = \\log \\lambda\\) as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{PoissonLog}(n|\\alpha) = \\frac{1}{n!} \\, \\exp \\left(n\\alpha -\n\\exp(\\alpha) \\right). \\end{equation*}\\]\n\n\n\nn ~ poisson_log(alpha)\nIncrement target log probability density with poisson_log_lupmf(n | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_log_lpmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha\nAvailable since 2.12\n \n\nreal poisson_log_lupmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha dropping constant additive terms\nAvailable since 2.25\n \n\nR poisson_log_rng(reals alpha) Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than \\(30 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
+    "text": "The embedded Laplace approximation is used for latent Gaussian models. A latent Gaussian model is defined by three components:\n\n\\(\\phi\\): hyperparameters (e.g., GP kernel length-scale and magnitude, or variance components in a hierarchical model),\n\\(\\theta\\): latent Gaussian variables (the high-dimensional quantity to be marginalized out),\n\\(y\\): observed data.\n\nThese components are related through a hierarchical structure. The hyperparameters \\(\\phi\\) are given a prior \\(p(\\phi)\\). The latent variables \\(\\theta\\) have a multivariate normal prior centered at 0 with covariance matrix \\(K(\\phi)\\). An non-zero mean offset can be incorporated into the likelihood function. The observations \\(y\\) have a data model \\(p(y \\mid \\theta, \\phi)\\). The prior on \\(\\theta\\) is centered at zero; an offset can be incorporated into the data model if a non-zero mean is needed.\n\\[\\begin{eqnarray*}\n  \\phi & \\sim & p(\\phi) \\\\\n  \\theta & \\sim & \\text{Multi-Normal}(0, K(\\phi)) \\\\\n  y & \\sim & p(y \\mid \\theta, \\phi).\n\\end{eqnarray*}\\]\nThe generative model above defines a joint distribution over all three quantities, \\(p(\\phi, \\theta, y) = p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\). After observing data \\(y\\), Bayes’ theorem gives the joint posterior \\(p(\\phi, \\theta \\mid y) \\propto p(\\phi) \\, p(\\theta \\mid \\phi) \\, p(y \\mid \\theta, \\phi)\\), where \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) and \\(\\phi\\) is the joint likelihood function.\nSampling directly from the joint posterior \\(p(\\phi, \\theta \\mid y)\\) of this model is often difficult. Challenging geometries (e.g., funnels) frustrate inference algorithms, including Hamiltonian Monte Carlo and variational inference. However, the marginal posterior \\(p(\\phi \\mid y)\\) is often well-behaved and low-dimensional, making it much easier to sample. With an embedded Laplace approximation, we can obtain an approximation of the marginal posterior \\(p(\\phi \\mid y)\\). This is done via an intermediate approximation of the conditional posterior \\(p(\\theta \\mid \\phi, y)\\) by a normal distribution and this normal approximation is well-justified when the likelihood \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) is log concave (given that we already have a normal prior \\(p(\\theta \\mid \\phi)\\)). Once we obtain (approximate) samples \\(\\phi^{(i)} \\sim p(\\phi \\mid y)\\), we can in turn generate posterior draws for \\(\\theta\\) using the normal approximation to \\(p(\\theta \\mid y, \\phi)\\), evaluated at the posterior draws \\(\\phi^{(i)}\\).",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson-log-glm",
-    "href": "functions-reference/unbounded_discrete_distributions.html#poisson-log-glm",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/laplace_embedded.html#approximation-of-the-conditional-posterior-and-marginal-likelihood",
+    "href": "reference-manual/laplace_embedded.html#approximation-of-the-conditional-posterior-and-marginal-likelihood",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "Stan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e. a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{PoissonLogGLM}(y|x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Poisson}(y_i|\\exp(\\alpha_i + x_i\\cdot \\beta)). \\end{equation*}\\]\n\n\n\ny ~ poisson_log_glm(x, alpha, beta)\nIncrement target log probability density with poisson_log_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25",
+    "text": "The two-step inference strategy for using embedded Laplace in a latent Gaussian model requires approximations to both the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and the marginal likelihood \\(p(y \\mid \\phi)\\). The Laplace approximation is the normal distribution that matches the mode and curvature of the conditional posterior \\(p(\\theta \\mid y, \\phi)\\). The mode, defined as the value of \\(\\theta\\) that maximizes the conditional posterior, is estimated by a Newton solver, \\[\n  \\theta^* = \\underset{\\theta}{\\text{argmax}} \\ p(\\theta \\mid y, \\phi),\n\\]\nSince the approximation is normal, the curvature is matched by setting the covariance to the negative Hessian of the log conditional posterior, evaluated at the mode,\n\\[\n  \\Sigma^* = -  \\left . \\frac{\\partial^2}{\\partial \\theta^2}\n    \\log p (\\theta \\mid \\phi, y) \\right |_{\\theta =\\theta^*}.\n\\]\nThe resulting Laplace approximation is a multivariate normal centered at the mode with covariance given by the inverse curvature,\n\\[\n\\hat p_\\mathcal{L} (\\theta \\mid y, \\phi) = \\text{Multi-Normal}(\\theta^*, \\Sigma^*)\n\\approx p(\\theta \\mid y, \\phi).\n\\]\nThis approximation also yields an approximation to the marginal likelihood, obtained by evaluating the prior, likelihood, and approximate posterior at the mode \\(\\theta^*\\),\n\\[\n   \\hat p_\\mathcal{L}(y \\mid \\phi) := \\frac{p(\\theta^* \\mid \\phi) \\\n   p(y \\mid \\theta^*, \\phi) }{ \\hat p_\\mathcal{L} (\\theta^* \\mid \\phi, y) }\n   \\approx p(y \\mid \\phi).\n\\]\nHence, a strategy to approximate the posterior of the latent Gaussian model is to first estimate the marginal posterior \\(\\hat p_\\mathcal{L}(\\phi \\mid y) \\propto p(\\phi) p_\\mathcal{L} (y \\mid \\phi)\\) using any algorithm supported by Stan. Approximate posterior draws for the latent Gaussian variables are then obtained by first sampling \\(\\phi \\sim \\hat p_\\mathcal{L}(\\phi \\mid y)\\) and then \\(\\theta \\sim  \\hat p_\\mathcal{L}(\\theta \\mid \\phi, y)\\).",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_discrete_distributions.html#beta-neg-binomial",
-    "href": "functions-reference/unbounded_discrete_distributions.html#beta-neg-binomial",
-    "title": "Unbounded Discrete Distributions",
+    "objectID": "reference-manual/laplace_embedded.html#trade-offs-of-the-approximation",
+    "href": "reference-manual/laplace_embedded.html#trade-offs-of-the-approximation",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "If \\(r \\in \\mathbb{R}^+\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{BetaNegBinomial}(n|r,\\alpha,\\beta) = \\frac {\\Gamma (n+r )}{n!\\;\\Gamma (r )}\n\\frac {\\mathrm {B} (\\beta+n,\\alpha +r )}{\\mathrm {B} (\\beta,\\alpha )}. \\end{equation*}\\]\n\n\n\nn ~ beta_neg_binomial(r,alpha,beta)\nIncrement target log probability density with beta_neg_binomial_lupmf(n | r, alpha, beta).\nAvailable since 2.36\n \n\n\n\n\n \n\nreal beta_neg_binomial_lpmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lupmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta dropping constant additive terms.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_cdf(ints n | reals r, reals alpha, reals beta) The beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lcdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lccdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial complementary cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nR beta_neg_binomial_rng(reals r, reals alpha, reals beta) Generate a beta negative binomial variate with parameters r, alpha and beta; may only be used in transformed data and generated quantities blocks. r \\(\\cdot\\) beta \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.36",
+    "text": "The embedded Laplace approximation presents several trade-offs with standard inference over the joint posterior \\(p(\\theta, \\phi \\mid y)\\). The main advantage of the embedded Laplace approximation is that it side-steps the intricate geometry of hierarchical models. The marginal posterior \\(p(\\phi \\mid y)\\) can then be handled by Hamiltonian Monte Carlo sampling without extensive tuning or reparameterization, and the mixing time is faster, meaning we can run shorter chains to achieve a desired precision. One additional benefit is that approximate methods, e.g. variational inference, which work poorly on the joint \\(p(\\theta, \\phi \\mid y)\\) can work well on the marginal posterior \\(p(\\phi \\mid y)\\).\nOn the other hand, the embedded Laplace approximation presents certain disadvantages. First, we need to perform a Laplace approximation each time the log marginal likelihood is evaluated, meaning each iteration can be expensive. Secondly, the approximation can introduce non-negligible error, especially with non-log-concave likelihoods (note the prior is always multivariate normal). How these trade-offs are resolved depends on the application; see Margossian et al. (2020) for some examples.\n\n\nThe quality of the Laplace approximation depends on how close the true conditional posterior \\(p(\\theta \\mid y, \\phi)\\) is to Gaussian.\nWorks well. Log-concave likelihoods, for example from a Poisson model with log link or negative binomial with log link. These produce unimodal conditional posteriors when combined with a Gaussian prior. The approximation error is typically negligible with these likelihoods, especially with moderate-to-large counts (Kuss and Rasmussen 2005; Vanhatalo, Pietiläinen, and Vehtari 2010; Cseke and Heskes 2011; Vehtari et al. 2016). If the likelihood is normal, there is no error in the approximation but in this case the marginalization can be worked analytically and the resulting implementation is much faster than using the embedded Laplace approximation.\nWorks adequately. Bernoulli model with logit link has technically log-concave likelihood, but the likelihood can be very skewed making the Gaussian approximation less accurate than for count data. The embedded Laplace is still useful when \\(\\theta\\) is high-dimensional and joint sampling is infeasible; see Vehtari et al. (2016) and Margossian et al. (2020) for discussion.\nNot appropriate. For likelihoods that are not log-concave in \\(\\theta\\), the conditional posterior may be multimodal and the Newton solver finds only a single mode or can fail completely. When \\(\\theta\\) is low-dimensional (a few dozen or fewer), the overhead of the inner optimization may not pay for itself and standard joint HMC sampling is often adequate.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Unbounded Discrete Distributions"
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/transform_functions.html",
-    "href": "functions-reference/transform_functions.html",
-    "title": "Variable Transformation Functions",
+    "objectID": "reference-manual/laplace_embedded.html#details-of-the-approximation",
+    "href": "reference-manual/laplace_embedded.html#details-of-the-approximation",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "Variable transformation functions provide implementations of the built-in constraining and unconstraining transforms defined in Stan Reference Manual.\nFor each of the built-in variable transforms there are three functions named after the transform with differing suffixes. A _unconstrain function that maps from the constrained space back to free variables (the “transform”), A _constrain function that maps from free variables to constrained variables (the “inverse transform”), and a _jacobian function, which computes the same value as the _constrain function while also incrementing the Jacobian accumulator with the log Jacobian determinant.\nFor this page, variables named y are unconstrained, while variables named x are in the constrained space. The unconstraining functions will reject if their input does not satisfy the declared constraint.\n\n\nThese transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.\nThe functions are all overloaded to apply to containers elementwise. If the y argument is a container, the others must be either scalars or containers of exactly the same size.\n\n\nThese functions perform the transform and inverse transform described in the Lower Bounded Scalar section.\n \n\nreals lower_bound_constrain(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_bound_jacobian(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_bound_unconstrain(reals x, reals lb) Takes a value x which is greater than lb and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Upper Bounded Scalar section.\n \n\nreals upper_bound_constrain(reals y, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals upper_bound_jacobian(reals x, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals upper_bound_unconstrain(reals x, reals ub) Takes a value x which is less than ub and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Lower and Upper Bounded Scalar section.\n \n\nreals lower_upper_bound_constrain(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_upper_bound_jacobian(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_upper_bound_unconstrain(reals x, reals lb, reals ub) Takes a value x which is bounded between lb and ub and returns returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Affinely Transformed Scalar section.\n \n\nreals offset_multiplier_constrain(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nAvailable since 2.37\n \n\nreals offset_multiplier_jacobian(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals offset_multiplier_unconstrain(reals x, reals offset, reals mult) Takes a value x, shift offset, and scale mult and a value which has been un-scaled and un-shifted.\nAvailable since 2.37\n\n\n\n\nThese functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.\nWhere vectors is used, this indicates that either a vector or a (possibly multidimensional) array of vectors may be provided. The array will be processed element by element.\n\n\nThese functions perform the transform and inverse transform described in the Ordered Vector section.\n \n\nvectors ordered_constrain(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nAvailable since 2.37\n \n\nvectors ordered_jacobian(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors ordered_unconstrain(vectors x) Takes an ordered vector x and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Positive Ordered Vector section.\n \n\nvectors positive_ordered_constrain(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nAvailable since 2.37\n \n\nvectors positive_ordered_jacobian(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors positive_ordered_unconstrain(vectors x) Takes an ordered vector x with positive entries and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Simplex section.\n \n\nvectors simplex_constrain(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors simplex_jacobian(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors simplex_unconstrain(vectors x) Takes a simplex x and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Zero Sum Vector section.\n \n\nvectors sum_to_zero_constrain(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors sum_to_zero_jacobian(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThe returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors sum_to_zero_unconstrain(vectors x) Takes a vector x with elements that sum to 0 and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Vector section.\n \n\nvectors unit_vectors_constrain(vectors y) Takes a free vector y and returns a vector with unit length, i.e., norm2(unit_vectors_constrain(y)) == 1 for any y that has a positive and finite norm itself (if y does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.\nAvailable since 2.37\n \n\nvectors unit_vectors_jacobian(vectors y) Takes a free vector y and returns a vector with unit length. This function rejects if given a vector of all zeros.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors unit_vectors_unconstrain(vectors x) Takes a vector x of unit length and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\n\nSimilarly to the above, vectors means a vector or array thereof, and matrices means a matrix or array thereof.\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Correlation Matrices section.\n \n\nmatrices cholesky_factor_corr_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.\nAvailable since 2.37\n \n Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment. matrices cholesky_factor_corr_jacobian(vectors y, int K)\nAvailable since 2.37\n \n\nvectors cholesky_factor_corr_unconstrain(matrices x) Takes x, a (\\(K \\times K\\)) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Covariance Matrices section.\n \n\nmatrices cholesky_factor_cov_constrain(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nAvailable since 2.37\n \n\nmatrices cholesky_factor_cov_jacobian(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cholesky_factor_cov_unconstrain(matrices x) Takes a \\(M \\times N\\) matrix x which is a Cholesky factor of a covariance matrix (a matrix \\(L\\) such that \\(L^T L\\) is positive definite) and returns the corresponding free vector of length \\(N + \\binom{N}{2} + (M - N)N\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Correlation Matrices section.\n \n\nmatrices corr_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nAvailable since 2.37\n \n\nmatrices corr_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors corr_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) matrix x which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size \\(\\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Covariance Matrices section.\n \n\nmatrices cov_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nAvailable since 2.37\n \n\nmatrices cov_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cov_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) positive definite matrix x and returns the corresponding free vector of size \\(K + \\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for column (left) stochastic matrices.\n \n\nmatrices stochastic_column_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nAvailable since 2.37\n \n\nmatrices stochastic_column_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_column_unconstrain(matrices x) Takes a left stochastic matrix x of size \\(N+1 \\times M\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for row (right) stochastic matrices.\n \n\nmatrices stochastic_row_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nAvailable since 2.37\n \n\nmatrices stochastic_row_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_row_unconstrain(matrices x) Takes a right stochastic matrix x of size \\(N \\times M+1\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThe sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the sum-to-zero matrix section of the Reference Manual.\n \n\nmatrices sum_to_zero_constrain(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_jacobian(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to sum_to_zero_constrain.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_unconstrain(matrices x) This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.\nAvailable since 2.37",
+    "text": "When the embedded Laplace approximation does not converge or produces unexpected results, the solver configuration may need adjustment. This section describes the internals of the Newton solver and the options available for tuning it.\n\n\nA critical component of the embedded Laplace approximation is the Newton solver used to estimate the mode \\(\\theta^*\\) of \\(p(\\theta \\mid \\phi, y)\\). The objective function being maximized is the log joint density of the prior and likelihood with respect to \\(\\theta\\).\n\\[\n\\Psi(\\theta) = \\log p(\\theta \\mid \\phi) + \\log p(y \\mid \\theta, \\phi),\n\\]\nConvergence is declared when the change in the objective between successive iterations falls below a tolerance \\(\\Delta\\).\n\\[\n| \\Psi (\\theta^{(i + 1)}) - \\Psi (\\theta^{(i)}) | \\le \\Delta.\n\\]\nThe solver also stops after reaching a pre-specified maximum number of steps. In that case, Stan throws a warning, but still returns the last iteration’s parameters. If you see this warning you should check the diagnostics to understand why the solver failed to converge.\nTo help with cases where the Newton step does not lead to a decrease in the objective function, the Newton iteration is augmented with a wolfe line-search to ensure that at each iteration the objective function \\(\\Psi\\) decreases. Specifically, suppose the objective increases after a Newton step, indicating the step overshot a region of improvement,\n\\[\n\\Psi (\\theta^{(i + 1)}) &lt; \\Psi (\\theta^{(i)}).\n\\]\nThis can indicate that the Newton step \\(\\alpha\\) at iteration \\(i\\) is too large and that we skipped a region where the objective function decreases. In that case, we can fallback to a Wolfe line search to find a step size which satisfies the Wolfe conditions. The wolfe line search attempts to find a search direction \\(p_i\\) and step size \\(\\alpha_k\\) such that an accepted step both increases our objective while ensuring the slope of the accepted step is flatter than our previous position. Together these help push the algorithm towards a minimum.\n\\[\nf(x_i + \\alpha_k p_i) \\le f(x_i) + c_1 \\alpha_k \\nabla f(x_i)^T p_i\n-p^T_i \\Delta f(x_i + \\alpha_k p_i) \\le -c_2 p^T_i \\Delta f(x_i)\n\\]\n\\[\n  \\theta^{(i + 1)} \\leftarrow \\frac{\\theta^{(i + 1)} + \\theta^{(i)}}{2}.\n\\]\nWe repeat this halving of steps until \\(\\Psi (\\theta^{(i + 1)}) \\ge \\Psi (\\theta^{(i)})\\), or until a maximum number of linesearch steps is reached. For certain problems, adding a linesearch can make the optimization more stable.\n\n\n\nThe embedded Laplace approximation uses a custom Newton solver, specialized to find the mode of \\(p(\\theta \\mid \\phi, y)\\). A key step for efficient optimization is to ensure all matrix inversions are numerically stable. This can be done using the Woodbury-Sherman-Morrison formula and requires one of three matrix decompositions:\n\nCholesky decomposition of the Hessian of the negative log likelihood \\(W = - \\partial^2_\\theta \\log p(y \\mid \\theta, \\phi)\\).\nCholesky decomposition of the prior covariance matrix \\(K(\\phi)\\).\nLU-decomposition of \\(I + KW\\), where \\(I\\) is the identity matrix.\n\nThe first solver (1) should be used if the negative log likelihood is positive-definite. Otherwise the user should rely on (2). In rarer cases where it is not numerically safe to invert the covariance matrix \\(K\\), users can use the third solver as a last-resort option.\n\n\n\nA key step to speed up computation is to take advantage of the sparsity of \\(H\\), the Hessian of the log likelihood with respect to the latent variables, \\[\n  H = \\frac{\\partial^2}{\\partial \\theta^2} \\log p(y \\mid \\theta, \\phi).\n\\] For example, if the observations \\((y_1, \\cdots, y_N)\\) are conditionally independent and each depends on only one component of \\(\\theta\\), the log likelihood decomposes into a sum of per-observation terms, \\[\n  \\log p(y \\mid \\theta, \\phi) = \\sum_{i = 1}^N \\log p(y_i \\mid \\theta_i, \\phi),\n\\] and the Hessian is diagonal. This leads to faster calculations of the Hessian and subsequently sparse matrix operations. This case is common in Gaussian process models, and certain hierarchical models.\nStan’s suite of functions for the embedded Laplace approximation exploits block-diagonal structure in the Hessian, where the user specifies the block size B. The user can specify the size \\(B\\) of these blocks. The user is responsible for working out what \\(B\\) is. If the Hessian is dense, then we simply set \\(B = N\\). The diagonal case above corresponds to B = 1. Arbitrary sparsity patterns beyond block-diagonal structure are not currently supported.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Variable Transformation Functions"
+      "Reference Manual",
+      "Algorithms",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/transform_functions.html#transforms-for-scalars",
-    "href": "functions-reference/transform_functions.html#transforms-for-scalars",
-    "title": "Variable Transformation Functions",
+    "objectID": "reference-manual/index.html",
+    "href": "reference-manual/index.html",
+    "title": "Stan Reference Manual",
     "section": "",
-    "text": "These transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.\nThe functions are all overloaded to apply to containers elementwise. If the y argument is a container, the others must be either scalars or containers of exactly the same size.\n\n\nThese functions perform the transform and inverse transform described in the Lower Bounded Scalar section.\n \n\nreals lower_bound_constrain(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_bound_jacobian(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_bound_unconstrain(reals x, reals lb) Takes a value x which is greater than lb and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Upper Bounded Scalar section.\n \n\nreals upper_bound_constrain(reals y, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals upper_bound_jacobian(reals x, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals upper_bound_unconstrain(reals x, reals ub) Takes a value x which is less than ub and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Lower and Upper Bounded Scalar section.\n \n\nreals lower_upper_bound_constrain(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_upper_bound_jacobian(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_upper_bound_unconstrain(reals x, reals lb, reals ub) Takes a value x which is bounded between lb and ub and returns returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Affinely Transformed Scalar section.\n \n\nreals offset_multiplier_constrain(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nAvailable since 2.37\n \n\nreals offset_multiplier_jacobian(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals offset_multiplier_unconstrain(reals x, reals offset, reals mult) Takes a value x, shift offset, and scale mult and a value which has been un-scaled and un-shifted.\nAvailable since 2.37",
+    "text": "This is the official reference manual for Stan’s programming language for coding probability models, inference algorithms for fitting models and making predictions, and posterior analysis tools for evaluating the results. This manual applies to all Stan interfaces.\nThe first part of the reference manual provides a full specification of the Stan programming language. The language is responsible for defining a log density function conditioned on data. Typically, this is a Bayesian posterior, but it may also be a penalized likelihood function. The second part of the manual specifies the inference algorithms and posterior inference tools. The third part provides auxiliary information about the use of Stan.\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Variable Transformation Functions"
+      "Reference Manual",
+      "Stan Reference Manual"
     ]
   },
   {
-    "objectID": "functions-reference/transform_functions.html#transforms-for-constrained-vectors",
-    "href": "functions-reference/transform_functions.html#transforms-for-constrained-vectors",
-    "title": "Variable Transformation Functions",
+    "objectID": "reference-manual/expressions.html",
+    "href": "reference-manual/expressions.html",
+    "title": "Expressions",
     "section": "",
-    "text": "These functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.\nWhere vectors is used, this indicates that either a vector or a (possibly multidimensional) array of vectors may be provided. The array will be processed element by element.\n\n\nThese functions perform the transform and inverse transform described in the Ordered Vector section.\n \n\nvectors ordered_constrain(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nAvailable since 2.37\n \n\nvectors ordered_jacobian(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors ordered_unconstrain(vectors x) Takes an ordered vector x and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Positive Ordered Vector section.\n \n\nvectors positive_ordered_constrain(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nAvailable since 2.37\n \n\nvectors positive_ordered_jacobian(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors positive_ordered_unconstrain(vectors x) Takes an ordered vector x with positive entries and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Simplex section.\n \n\nvectors simplex_constrain(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors simplex_jacobian(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors simplex_unconstrain(vectors x) Takes a simplex x and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Zero Sum Vector section.\n \n\nvectors sum_to_zero_constrain(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors sum_to_zero_jacobian(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThe returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors sum_to_zero_unconstrain(vectors x) Takes a vector x with elements that sum to 0 and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Vector section.\n \n\nvectors unit_vectors_constrain(vectors y) Takes a free vector y and returns a vector with unit length, i.e., norm2(unit_vectors_constrain(y)) == 1 for any y that has a positive and finite norm itself (if y does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.\nAvailable since 2.37\n \n\nvectors unit_vectors_jacobian(vectors y) Takes a free vector y and returns a vector with unit length. This function rejects if given a vector of all zeros.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors unit_vectors_unconstrain(vectors x) Takes a vector x of unit length and returns the corresponding free vector.\nAvailable since 2.37",
+    "text": "An expression is the syntactic unit in a Stan program that denotes a value. Every expression in a well-formed Stan program has a type that is determined statically (at compile time), based only on the type of its variables and the types of the functions used in it. If an expressions type cannot be determined statically, the Stan compiler will report the location of the problem.\nThis chapter covers the syntax, typing, and usage of the various forms of expressions in Stan.\n\n\nThe simplest form of expression is a literal that denotes a primitive numerical value.\n\n\nInteger literals represent integers of type int. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression --1 is interpreted as the negation of the literal -1.)\nThe following list contains well-formed integer literals.\n0, 1, -1, 256, -127098, 24567898765\nInteger literals must have values that fall within the bounds for integer values (see the section on numerical data types).\nInteger literals may not contain decimal points (.). Thus the expressions 1. and 1.0 are of type real and may not be used where a value of type int is required.\n\n\n\nA number written with a period or with scientific notation is assigned to a the continuous numeric type real. Real literals are written in base 10 with a period (.) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.\n0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.\nThe notation e or E followed by a positive or negative integer denotes a power of 10 to multiply. For instance, 2.7e3 and 2.7e+3 denote \\(2.7 \\times 10^3\\), whereas -2E-5 denotes \\(-2 \\times\n10^{-5}\\).\n\n\n\nA number followed by the character i denotes an imaginary number and is assigned to the numeric type complex. The number preceding i may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.\n1i, 2i, -325.786i, 1e10i, 2.87e-10i.\nNote that the character i by itself is not a well-formed imaginary literal. The unit imaginary number must be written as 1i.\n\n\n\nStan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.\n1 + 2i, -3.2e9 + 1e10i\nThese will be assigned the type complex, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.\n\n\n\n\nA variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (_) character. Variables must start with a letter (a--z and A--Z) and may not end with two underscores (__).\nExamples of legal variable identifiers are as follows.\na, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable\nUnlike in R and BUGS, variable identifiers in Stan may not contain a period character.\n\n\nStan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the stanc translator halting with an error message indicating which reserved name was used and its location in the model code.\n\n\nThe name of the model cannot be used as a variable within the model. This is usually not a problem because the default in bin/stanc is to append _model to the name of the file containing the model specification. For example, if the model is in file foo.stan, it would not be legal to have a variable named foo_model when using the default model name through bin/stanc. With user-specified model names, variables cannot match the model.\n\n\n\nThe following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.\nfor, in, while, repeat, until, if, then, else,\ntrue, false, target, struct, typedef, export,\nauto, extern, var, static, lower, upper, offset,\nmultiplier\nVariables should not be named after types, either, and thus may not be any of the following.\nint, real, complex, vector, simplex, unit_vector,\nsum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,\nrow_vector, matrix, cholesky_factor_corr,\ncolumn_stochastic_matrix,row_stochastic_matrix,\ncholesky_factor_cov, corr_matrix, cov_matrix, array\nThe following keywords or built-in functions are also reserved and cannot be used as variable names:\nprint, reject, profile, fatal_error, target, jacobian\nThe following block identifiers are reserved and cannot be used as variable names:\nfunctions, model, data, parameters, quantities,\ntransformed, generated\n\n\n\nVariable names will also conflict with the names of distributions suffixed with _lpdf, _lpmf, _lcdf, and _lccdf, _cdf, and _ccdf, such as normal_lcdf_log. No user-defined variable can take a name ending in _lupdf or _lupmf even if a corresponding _lpdf or _lpmf is not defined.\nUsing any of these variable names causes the stanc translator to halt and report the name and location of the variable causing the conflict.\n\n\n\nStan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed _stan_NAME (e.g. _stan_public) behind the scenes (in the generated C++ code).\n\nalignas, alignof, and, and_eq, asm, bitand, bitor, bool,\ncase, catch, char, char16_t, char32_t, class, compl, const,\nconstexpr, const_cast, decltype, default, delete, do,\ndouble, dynamic_cast, enum, explicit, float, friend, goto,\ninline, long, mutable, namespace, new, noexcept, not, not_eq,\nnullptr, operator, or, or_eq, private, protected, public,\nregister, reinterpret_cast, short, signed, sizeof,\nstatic_assert, static_cast, switch, template, this, thread_local,\nthrow, try, typeid, typename, union, unsigned, using, virtual,\nvolatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,\nSTAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH\n\n\n\n\nThe legal characters for variable identifiers are given in the following table.\n\n\nThe alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.\n\n\n\ncharacters\nASCII code points\n\n\n\n\na -- z\n97 – 122\n\n\nA -- Z\n65 – 90\n\n\n0 -- 9\n48 – 57\n\n\n_\n95\n\n\n\nAlthough not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.\n\n\n\nWithin comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.\n\n\n\n\n\nExpressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.\n\n\nSquare brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression [ 1, 10, 100 ] denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.\nrow_vector[2] rv2 =  [ 1, 2 ];\nvector[3] v3 = [ 3, 4, 5 ]';\nThe vector expression values may be compound expressions or variable names, so it is legal to write [ 2 * 3, 1 + 4] or [ x, y ], providing that x and y are primitive variables.\n\n\n\nA matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.\nmatrix[3, 2] m1 = [ [ 1, 2 ], [ 3, 4 ], [5, 6 ] ];\nAny expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:\nvector[2] vX = [ 1, 10 ]';\nrow_vector[2] vY = [ 100, 1000 ];\nmatrix[3, 2] m2 = [ vX', vY, [ 1, 2 ]  ];\n\n\n\nComplex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.\ncomplex_vector[3] = [1 + 2i, 3 - 1.7i, 0]';\ncomplex_row_vector[2] = [12, -2i];\ncomplex_matrix[2, 3] = [[1 + 2i, 3 - 1.7i, 0],\n                        [3.9 - 1.234i, 176i, 1 + 1i]];\n\n\nThe empty expression [ ] is ambiguous and therefore is not allowed and similarly expressions such as [ [ ] ] or [ [ ], [ ] ] are not allowed.\n\n\n\n\nIf needed, it is possible to create an empty vector with\nrep_vector(e, 0)\nwhere the first expression e needs to scalar of type real.\nIf needed, it is possible to create an empty matrix with\nrep_matrix(e, 0, 0)\nwhere the first expression e needs to scalar of type real.\n\n\n\nCurly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression { 1, 10, 100 } denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.\narray[3] int a = { 1, 10, 100 };\nThe values may be compound expressions, so it is legal to write { 2 * 3, 1 + 4 }. It is also possible to write two dimensional arrays directly, as in the following example.\narray[2, 3] int b = { { 1, 2, 3 }, { 4, 5, 6 } };\nThis way, b[1] is { 1, 2, 3 } and b[2] is { 4, 5, 6 }.\nWhitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.\narray[2, 3] int b = { { 1, 2, 3 },\n                { 4, 5, 6 } };\n\n\n\nThe empty array expression ({ }) is not allowed. See more about restrictions on array expressions in subsection Restrictions on values.\nIf needed, it is possible to create an empty array with\nrep_array(e, 0)\nwhere the first expression e determines the type of the array. For example, rep_array(0.0, 0) returns an empty real array of type real[], whereas rep_array({123}, 0) returns an empty two dimensional integer array of type int[ , ]. Only the type of the first argument is used, so the integer arrays {123} and {0} produce equivalent values.\n\n\n\nAny type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.\nvector[3] b;\nvector[3] c;\n// ...\narray[2] vector[3] d = { b, c };\nThe elements may also be a mixture of int and real typed expressions, in which case the result is an array of real values.\narray[2] real b = { 1, 1.9 };\n\n\n\nStan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.\ntuple(int, vector[3]) xy = (42, [1, 2.9, -1.3]');\nThe expression 42 is of type int and the expression [1, 2.9, -1.3] is of type row_vector so that [1, 2.9, -1.3]' is of type vector and of size 3. The whole tuple expression (42, [1, 2.9, -1.3]') thus has a sized type of tuple(int, vector[3]) and an unsized type (e.g., for a function argument) of tuple(int, vector).\nA tuple of one element can be created using the same style as languages like Python, with a trailing comma, e.g., (3.14,). For longer tuples, Stan does not support trailing commas.\n\n\n\nThere are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.\n\n\nAlthough it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.\n{ { 1, 2, 3 }, { 4, 5 } }  // compile time error: size mismatch\nThis may appear to be OK, because it is creating a two-dimensional integer array (array[,] int) out of two one-dimensional array integer arrays (array[] int). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.\n{ { 1, 2, 3 }, m }  // runtime error if m not size 3\n\n\n\nBecause there is no way to infer the type of the result, the empty array expression ({ }) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.\narray[0] int a;   // a is fully defined as zero element array\n\n\n\nThere is no way to declare or construct a zero-tuple in Stan. Tuples must be at least one element long. The expression () does not pick out a zero-tuple—it is ill formed.\nOne-tuples need a trailing comma, like (1,). The expression (1), without the comma, is of type int rather than a tuple.\n\n\n\n\n\nAny expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, ( and ), are allowed. The square brackets [ and ] are reserved for array indexing and the curly braces { and } for grouping statements.\nWith parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression 1 + 2 * 3 has a subexpression 2 * 3 and evaluates to 7. With parentheses, this grouping may be made explicit with the expression 1 + (2 * 3). More importantly, the expression (1 + 2) *   3 has 1 + 2 as a subexpression and evaluates to 9.\n\n\n\nFor integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (+), subtraction (-), multiplication (*) and division (/) in the usual ways.\nFor integer expressions, Stan supports the modulus (%) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3.0 + 0.14\n-15\n2 * 3 + 1\n(x - y) / 2.0\n(n * (n + 1)) / 2\nx / n\nm % n\nThe negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (') is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.\nFor example, if y and mu are variables of type vector and Sigma is a variable of type matrix, then (y - mu)' * Sigma * (y - mu) is a well-formed expression of type real. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) y - mu are of type vector because the variables y and mu are of type vector. The transpose of this expression, the subexpression (y - mu)' is of type row_vector. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form (((y - mu)') * Sigma) * (y - mu).\nThe type of subexpression (y - mu)' * Sigma is inferred to be row_vector, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a real value.\nStan provides elementwise matrix multiplication (e.g., a .* b) and division (e.g., a ./ b) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,\nvector[N] a;\nvector[N] b;\nvector[N] c;\nthe assignment,\nc = a .* b;\nproduces the same result with roughly the same efficiency as the loop\nfor (n in 1:N) {\n  c[n] = a[n] * b[n];\n}\nStan supports exponentiation (^) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3 ^ 2\n3.0 ^ -2\n3.0 ^ 0.14\nx ^ n\nn ^ x\nn ^ m\nx ^ y\nExponentiation is right associative, so the expression 2 ^ 3 ^ 4 is equivalent to the fully specified form 2 ^ (3 ^ 4).\n\n\nThe precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.\n\n\nStan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e., operator*(int, int):int indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.\n\n\n\n\n\n\n\n\n\n\nOp.\nPrec.\nAssoc.\nPlacement\nDescription\n\n\n\n\n? ~ :\n10\nright\nternary infix\nconditional\n\n\n||\n9\nleft\nbinary infix\nlogical or\n\n\n&&\n8\nleft\nbinary infix\nlogical and\n\n\n==\n7\nleft\nbinary infix\nequality\n\n\n!=\n7\nleft\nbinary infix\ninequality\n\n\n&lt;\n6\nleft\nbinary infix\nless than\n\n\n&lt;=\n6\nleft\nbinary infix\nless than or equal\n\n\n&gt;\n6\nleft\nbinary infix\ngreater than\n\n\n&gt;=\n6\nleft\nbinary infix\ngreater than or equal\n\n\n+\n5\nleft\nbinary infix\naddition\n\n\n-\n5\nleft\nbinary infix\nsubtraction\n\n\n*\n4\nleft\nbinary infix\nmultiplication\n\n\n.*\n4\nleft\nbinary infix\nelementwise multiplication\n\n\n/\n4\nleft\nbinary infix\n(right) division\n\n\n./\n4\nleft\nbinary infix\nelementwise division\n\n\n%\n4\nleft\nbinary infix\nmodulus\n\n\n\\\n3\nleft\nbinary infix\nleft division\n\n\n%/%\n3\nleft\nbinary infix\ninteger division\n\n\n!\n2\nn/a\nunary prefix\nlogical negation\n\n\n-\n2\nn/a\nunary prefix\nnegation\n\n\n+\n2\nn/a\nunary prefix\npromotion (no-op in Stan)\n\n\n^\n1\nright\nbinary infix\nexponentiation\n\n\n.^\n1\nright\nbinary infix\nelementwise exponentiation\n\n\n'\n0\nn/a\nunary postfix\ntransposition\n\n\n()\n0\nn/a\nprefix, wrap\nfunction application\n\n\n[]\n0\nleft\nprefix, wrap\narray, matrix indexing\n\n\n\nOther expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.\nThe precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression a + b + c is interpreted as (a + b) + c. Similarly, a / b * c is interpreted as (a / b) * c.\nBecause multiplication has higher precedence than addition, the expression a * b + c is interpreted as (a * b) + c and the expression a + b * c is interpreted as a + (b * c). Similarly, 2 * x + 3 * - y is interpreted as (2 * x) + (3 * (-y)).\nTransposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, -u' is interpreted as -(u'), u * v' as u* (v'), and u' * v as (u') * v. For integer and reals, -n ^ 3 is interpreted as -(n ^ 3).\n\n\n\n\n\n\n\nThe ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If a is an expression of type int and b and c are expressions that can be converted to one another (e.g., compared with ==), then\na ? b : c\nis an expression of the promoted type of b and c. The result of this expression is b if condition a is true, c otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g. if one argument is of type int and the other of type real, the conditional expression as a whole is of type real. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.\n\n\nThe conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,\na &gt; 0 || b &lt; 0 ? c + d : e - f\nis equivalent to the explicitly grouped version\n(a &gt; 0 || b &lt; 0) ? (c + d) : (e - f)\nThe latter is easier to read even if the parentheses are not strictly necessary.\n\n\n\nThe conditional operator is right associative, so that\na ? b : c ? d : e\nparses as if explicitly grouped as\na ? b : (c ? d : e)\nAgain, the explicitly grouped version is easier to read.\n\n\n\n\nStan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable a in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value b is returned if the condition evaluates to a non-zero value and c is returned if the condition evaluates to zero.\n\n\nThe key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.\n\n\n\nIf one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full if-then conditional statement. For example,\ndata {\n  array[10] real x;\n  // ...\n}\nparameters {\n  array[10] real z;\n  // ...\n}\nmodel {\n  y ~ normal(cond ? x : z, sigma);\n  // ...\n}\nwould be more efficiently (if not more transparently) coded as\nif (cond) {\n  y ~ normal(x, sigma);\n} else {\n  y ~ normal(z, sigma);\n}\nThe conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable x will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.\n\n\n\n\n\nStan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if x is a variable of type array [] real (a one-dimensional array of reals) then x[1] is the value of the first element of the array.\nSubscripting has higher precedence than any of the arithmetic operations. For example, alpha * x[1] is equivalent to alpha * (x[1]).\nMultiple subscripts may be provided within a single pair of square brackets. If x is of type array[,] real, a two-dimensional array, then x[2, 501] is of type real.\n\n\nThe subscripting operator also returns subarrays of arrays. For example, if x is of type array[,,] real, then x[2] is of type array[,] real, and x[2, 3] is of type array[] real. As a result, the expressions x[2, 3] and x[2][3] have the same meaning.\n\n\n\nIf Sigma is a variable of type matrix, then Sigma[1] denotes the first row of Sigma and has the type row_vector.\n\n\n\nStan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if m is of type matrix[ , ], a two-dimensional array of matrices, then m[1] refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that m[1, 2] is of type matrix and denotes the matrix in the first row and second column of the array. Continuing to add indices, m[1, 2, 3] is of type row_vector and denotes the third row of the matrix denoted by m[1, 2]. Finally, m[1, 2, 3, 4] is of type real and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array m.\n\n\n\n\nIn addition to single integer indexes, as described in the language indexing section, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.\n\n\nTypes of indexes and examples with one-dimensional containers of size N and an integer array ii of type array [] real size K.\n\n\n\nindex type\nexample\nvalue\n\n\n\n\ninteger\na[11]\nvalue of a at index 11\n\n\ninteger array\na[ii]\na[ii[1]], …, a[ii[K]]\n\n\nlower bound\na[3:]\na[3], …, a[N]\n\n\nupper bound\na[:5]\na[1], …, a[5]\n\n\nrange\na[2:7]\na[2], …, a[7]\n\n\nrange\na[7:2]\n[]\n\n\nrange\na[5-3:5+2]\na[2], …, a[7]\n\n\nall\na[:]\na[1], …, a[N]\n\n\nall\na[]\na[1], …, a[N]\n\n\n\nThe range indexing with : allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:\n  array[6] int ii = reverse(linspaced_int_array(6, 2, 7));\nThen a[ii] evaluates to a[7], …, a[2].\n\n\n\nThe fundamental semantic rule for dealing with multiple indexes is the following. If idxs is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.\na[idxs, ...][i, ...] = a[idxs[i], ...][...]\nOn the other hand, if idx is a single index, it reduces the dimensionality of the output, so that\na[idx, ...] = a[idx][...]\nThe only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.\n\n\nSpecial rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix a, with integer single indexes i and j and integer array multiple indexes is and js. The same typing rules apply for all multiple indexes.\n\n\n\nexample\nrow index\ncolumn index\nresult type\n\n\n\n\na[i]\nsingle\nn/a\nrow vector\n\n\na[is]\nmultiple\nn/a\nmatrix\n\n\na[i, j]\nsingle\nsingle\nreal\n\n\na[i, js]\nsingle\nmultiple\nrow vector\n\n\na[is, j]\nmultiple\nsingle\nvector\n\n\na[is, js]\nmultiple\nmultiple\nmatrix\n\n\n\nEvaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.\nm[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]\nm[idxs, idx][j] = m[idxs[j], idx]\nm[idx, idxs][j] = m[idx, idxs[j]]\nEvaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.\n\n\n\n\n\nStan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.\nExpressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, log(2.0) is the expression of type real denoting the result of applying the natural logarithm to the value of the real literal 2.0.\nSyntactically, function application has higher precedence than any of the other operators, so that y + log(x) is interpreted as y + (log(x)).\n\n\nEach function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as\nreal log(real);\nand the signature for the lmultiply function is\nreal lmultiply(real, real);\nA function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.\nreal mean(array [] real);\nreal mean(vector);\nThe first applies to a one-dimensional array of real values and the second to a vector.\nThe identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.\n\n\n\nConstants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants \\(\\pi\\) and \\(e\\) are represented as nullary functions named pi() and e(). See the Stan Functions Reference built-in constants section for a list of built-in constants.\n\n\n\nBecause of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.\nFor example, consider a situation in which the following two function signatures have been registered for foo.\nreal foo(real, real);\nint foo(int, int);\nThe use of foo in the expression foo(1.0, 1.0) resolves to foo(real, real), and thus the expression foo(1.0, 1.0) itself is assigned a type of real.\nBecause integers may be promoted to real values, the expression foo(1, 1) could potentially match either foo(real, real) or foo(int, int). The former requires two type promotions and the latter requires none, so foo(1, 1) is resolved to function foo(int, int) and is thus assigned the type int.\nThe expression foo(1, 1.0) has argument types (int, real) and thus does not explicitly match either function signature. By promoting the integer expression 1 to type real, it is able to match foo(real, real), and hence the type of the function expression foo(1, 1.0) is real.\nIn some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named bar with the following signatures.\nreal bar(real, int);\nreal bar(int, real);\nWith these signatures, the expression bar(1.0, 1) and bar(1, 1.0) resolve to the first and second of the above functions, respectively. The expression bar(1.0, 1.0) is illegal because real values may not be demoted to integers. The expression bar(1, 1) is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name bar is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with bar(1, 1) given the two declarations above, the Stan compiler will flag the expression as illegal.\n\n\n\nFor most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix _rng. For example, a univariate normal random number can be generated by normal_rng(0, 1); only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.\n\n\nThe use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in _rng.\nThis allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.\n\n\n\nPosterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of (Gelman et al. 2013) for more information on posterior predictive checks.\n\n\n\n\n\nStan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.\n\n\nThe primitive implementation types for Stan are\nint, real, complex, vector, row_vector,  matrix, complex_vector,\ncomplex_row_vector, complex_matrix\nEvery basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.\n\n\nThe table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.\n\n\n\ntype\nprimitive type\n\n\n\n\nint\nint\n\n\nreal\nreal\n\n\nvector\nvector\n\n\nsimplex\nvector\n\n\nunit_vector\nvector\n\n\nsum_to_zero_vector\nvector\n\n\nordered\nvector\n\n\npositive_ordered\nvector\n\n\nrow_vector\nrow_vector\n\n\nmatrix\nmatrix\n\n\ncov_matrix\nmatrix\n\n\ncorr_matrix\nmatrix\n\n\ncholesky_factor_cov\nmatrix\n\n\ncholesky_factor_corr\nmatrix\n\n\ncolumn_stochastic_matrix\nmatrix\n\n\nrow_stochastic_matrix\nmatrix\n\n\nsum_to_zero_matrix\nmatrix\n\n\ncomplex_vector\ncomplex_vector\n\n\ncomplex_row_vector\ncomplex_row_vector\n\n\ncomplex_matrix\ncomplex_matrix\n\n\n\nA full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, array [] real has an array dimensionality of 1, int an array dimensionality of 0, and array [,,] int an array dimensionality of 3. The implementation type matrix[ , , ] has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.\nRecall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.\narray[I, J, K] matrix[M, N] a;\nThe matrix a is indexed as a[i, j, k, m, n] with the array indices first, followed by the matrix indices, with a[i, j, k] being a matrix and a[i, j, k, m] being a row vector.\n\n\n\n\nStan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.\n\n\n\nThere are two basic promotion rules,\n\nint types may be promoted to real, and\nreal types may be promoted to complex.\n\nPlus, promotion is transitive, so that\n\nif type U can be promoted to type V and type V can be promoted to type T, then U can be promoted to T.\n\nThe first rule means that expressions of type int may be used anywhere an expression of type real is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.\nThe remaining rules have to do with covariant typing rules, which say that a container of type U may be promoted to a container of the same shape of type T if U can be promoted to T. For vector and matrix types, this induces three rules,\n\nvector may be promoted to complex_vector,\nrow_vector may be promoted to complex_row_vector\nmatrix may be promoted to complex_matrix.\n\nFor array types, there’s a single rule\n\narray[...] U may be promoted to array[...] T if U can be promoted to T.\n\nFor example, this means array[,] int may be used where array [,] real or array [,] complex is required; as another example, array[] real may be used anywhere array[] complex is required.\nTuples have the natural extension of the above rules, applied to all sub-types at once\n\nA tuple(U1, ..., UN) may be promoted to a tuple(T1, ..., TN) if every Un can be promoted to Tn for n in 1:N\n\n\n\nAn integer literal expression such as 42 is of type int. Real literals such as 42.0 are of type real. Imaginary literals such as -17i are of type complex. the expression 7 - 2i acts like a complex literal, but technically it combines a real literal 7 and an imaginary literal 2i through subtraction.\n\n\n\nThe type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is int.\nThere is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.1\n\n\n\nIf x is an expression of total dimensionality greater than or equal to \\(N\\), then the type of expression e[i1, i2, ..., iN] is the same as that of e[i1][i2]...[iN], so it suffices to define the type of a singly-indexed function. Suppose e is an expression and i is an expression of primitive type int. Then\n\nif e is an expression of type array[i1, i2, ..., iN] T and k, i1, …, iN are expressions of type int, then e[k] is an expression of type array[i2, ..., iN] T,\nif e is an expression of type array[i] T with i and k expressions of type int, then e[k] is of type T,\nif e has implementation type vector or row_vector, dimensionality 0, then e[i] has implementation type real,\nif e has implementation type matrix, then e[i] has type row_vector,\nif e has implementation type complex_vector or complex_row_vector and i is an expression of type int, then e[i] is an expression of type complex, and\nif e has implementation type complex_matrix, and i is an expression of type int, then e[i] is an expression of type complex_row_vector.\n\n\n\n\nIf f is the name of a function and e1,...,eN are expressions for \\(N \\geq 0\\), then f(e1,...,eN) is an expression whose type is determined by the return type in the function signature for f given e1 through eN. Recall that a function signature is a declaration of the argument types and the result type.\nIn looking up functions, binary operators like real * real are defined as operator*(real, real) in the documentation and index.\nIn matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type int may be promoted to type real or complex if necessary (see the subsection on type promotion in the function application section, a real argument will be promoted to complex if necessary, a vector will be promoted to complex_vector if necessary, and so on.\nIn general, matrix operations return the lowest inferable type. For example, row_vector * vector returns a value of type real, which is declared in the function documentation and index as real operator*(row_vector, vector).\n\n\n\n\n\nThere are several expression constructions in Stan that act as higher-order functions.2\nThe higher-order functions and the signature of their argument functions are listed in the following pair of tables.\n\n\nHigher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.\n\n\n\n\n\n\n\n\n\nfunction\nparameter or data args\ndata args\nreturn type\n\n\n\n\nalgebra_solver\nvector, vector\narray [] real, array [] real\nvector\n\n\nalgebra_solver_newton\nvector, vector\narray [] real, array [] real\nvector\n\n\nintegrate_1d,\nreal, real, array [] real\narray [] real, array [] real\nreal\n\n\nintegrate_ode_X,\nreal, array [] real, array [] real\narray [] real, array [] real\narray [] real\n\n\nmap_rect\nvector, vector\narray [] real, array [] real\nvector\n\n\n\nFor example, the integrate_ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  array [] real foo(real t,\n                    array [] real y,\n                    array [] real theta,\n                    array [] real x_r,\n                    array [] real x_i) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\narray[2] real y0;\nreal t0;\narray[T] real ts;\narray[1] real theta;\narray[0] real x_r;\narray[0] int x_i;\n// ...\narray[T, 2] real y_hat = integrate_ode_rk45(foo, y0, t0,\n                                              ts, theta, x_r, x_i);\nThe function argument is foo, the name of the user-defined function; as shown in the higher-order functions table, integrate_ode_rk45 takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.\n\n\n\nVariadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.\n\n\n\nfunction\nrestricted args\nreturn type\n\n\n\n\nsolve_X\nvector\nvector\n\n\node_X,\nvector, real, array [] real\nvector[]\n\n\nreduce_sum\narray[] T, T1, T2\nreal\n\n\n\nT, T1, and T2 can be any Stan type.\nFor example, the ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  vector foo(real t, vector y, real theta, vector beta,\n            array [] real x_i, int index) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\nvector[2] y0;\nreal t0;\narray[T] real ts;\nreal theta;\nvector[7] beta;\narray[10] int x_i;\nint index;\n// ...\nvector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,\n                              beta, x_i, index);\nThe function argument is foo, the name of the user-defined function. As shown in the variadic higher-order functions table, ode_rk45 takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of foo and returns an array of vectors.\n\n\n\nThe function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.\n\n\n\nSome of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.\nFor user-defined functions the qualifier data may be prepended to the type to restrict the argument to data-only variables.\n\n\n\n\nDerivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.\n\n\nUnlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(sqrt(x - x), 1);\n}\nAlgebraically, the distribution statement in the model could be reduced to\n  x ~ normal(0, 1);\nand it would seem the model should produce unit normal draws for x. But rather than canceling, the expression sqrt(x -   x) causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,\n\\[\n\\begin{array}{rcl}\n\\frac{d}{dx} \\sqrt{x - x}\n& = &\n\\frac{1}{2 \\sqrt{x - x}} \\times \\frac{d}{dx} (x - x)\n\\\\[4pt]\n& = &\n\\frac{1}{0} \\times (1 - 1)\n\\\\[4pt]\n& = &\n\\infty \\times 0\n\\\\[4pt]\n& = & \\mathrm{NaN}.\n\\end{array}\n\\]\nRather than the \\(x - x\\) canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.\nThe only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like sqrt(x - x) for which the chain rule produces not-a-number values.\n\n\n\nThe best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).\nFor example, compiling the above model to an executable sqrt-x-minus-x in CmdStan, the test can be run as\n&gt; ./sqrt-x-minus-x diagnose test=gradient\nwhich produces\n...\nTEST GRADIENT MODE\n\n Log probability=-0.393734\n\n param idx           value           model     finite diff           error\n         0       -0.887393             nan               0             nan\nEven though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Variable Transformation Functions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/transform_functions.html#transforms-for-constrained-matrices",
-    "href": "functions-reference/transform_functions.html#transforms-for-constrained-matrices",
-    "title": "Variable Transformation Functions",
+    "objectID": "reference-manual/expressions.html#numeric-literals",
+    "href": "reference-manual/expressions.html#numeric-literals",
+    "title": "Expressions",
     "section": "",
-    "text": "Similarly to the above, vectors means a vector or array thereof, and matrices means a matrix or array thereof.\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Correlation Matrices section.\n \n\nmatrices cholesky_factor_corr_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.\nAvailable since 2.37\n \n Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment. matrices cholesky_factor_corr_jacobian(vectors y, int K)\nAvailable since 2.37\n \n\nvectors cholesky_factor_corr_unconstrain(matrices x) Takes x, a (\\(K \\times K\\)) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Covariance Matrices section.\n \n\nmatrices cholesky_factor_cov_constrain(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nAvailable since 2.37\n \n\nmatrices cholesky_factor_cov_jacobian(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cholesky_factor_cov_unconstrain(matrices x) Takes a \\(M \\times N\\) matrix x which is a Cholesky factor of a covariance matrix (a matrix \\(L\\) such that \\(L^T L\\) is positive definite) and returns the corresponding free vector of length \\(N + \\binom{N}{2} + (M - N)N\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Correlation Matrices section.\n \n\nmatrices corr_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nAvailable since 2.37\n \n\nmatrices corr_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors corr_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) matrix x which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size \\(\\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Covariance Matrices section.\n \n\nmatrices cov_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nAvailable since 2.37\n \n\nmatrices cov_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cov_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) positive definite matrix x and returns the corresponding free vector of size \\(K + \\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for column (left) stochastic matrices.\n \n\nmatrices stochastic_column_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nAvailable since 2.37\n \n\nmatrices stochastic_column_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_column_unconstrain(matrices x) Takes a left stochastic matrix x of size \\(N+1 \\times M\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for row (right) stochastic matrices.\n \n\nmatrices stochastic_row_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nAvailable since 2.37\n \n\nmatrices stochastic_row_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_row_unconstrain(matrices x) Takes a right stochastic matrix x of size \\(N \\times M+1\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThe sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the sum-to-zero matrix section of the Reference Manual.\n \n\nmatrices sum_to_zero_constrain(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_jacobian(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to sum_to_zero_constrain.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_unconstrain(matrices x) This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.\nAvailable since 2.37",
+    "text": "The simplest form of expression is a literal that denotes a primitive numerical value.\n\n\nInteger literals represent integers of type int. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression --1 is interpreted as the negation of the literal -1.)\nThe following list contains well-formed integer literals.\n0, 1, -1, 256, -127098, 24567898765\nInteger literals must have values that fall within the bounds for integer values (see the section on numerical data types).\nInteger literals may not contain decimal points (.). Thus the expressions 1. and 1.0 are of type real and may not be used where a value of type int is required.\n\n\n\nA number written with a period or with scientific notation is assigned to a the continuous numeric type real. Real literals are written in base 10 with a period (.) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.\n0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.\nThe notation e or E followed by a positive or negative integer denotes a power of 10 to multiply. For instance, 2.7e3 and 2.7e+3 denote \\(2.7 \\times 10^3\\), whereas -2E-5 denotes \\(-2 \\times\n10^{-5}\\).\n\n\n\nA number followed by the character i denotes an imaginary number and is assigned to the numeric type complex. The number preceding i may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.\n1i, 2i, -325.786i, 1e10i, 2.87e-10i.\nNote that the character i by itself is not a well-formed imaginary literal. The unit imaginary number must be written as 1i.\n\n\n\nStan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.\n1 + 2i, -3.2e9 + 1e10i\nThese will be assigned the type complex, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Variable Transformation Functions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/simplex_distributions.html",
-    "href": "functions-reference/simplex_distributions.html",
-    "title": "Simplex Distributions",
+    "objectID": "reference-manual/expressions.html#variables.section",
+    "href": "reference-manual/expressions.html#variables.section",
+    "title": "Expressions",
     "section": "",
-    "text": "The simplex probabilities have support on the unit \\(K\\)-simplex for a specified \\(K\\). A \\(K\\)-dimensional vector \\(\\theta\\) is a unit \\(K\\)-simplex if \\(\\theta_k \\geq 0\\) for \\(k \\in \\{1,\\ldots,K\\}\\) and \\(\\sum_{k = 1}^K \\theta_k = 1\\).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) and \\(\\alpha \\in (\\mathbb{R}^+)^{K}\\), then for \\(\\theta \\in \\text{$K$-simplex}\\),\n\\[\\begin{equation*}\n\\text{Dirichlet}(\\theta|\\alpha) =\n\\frac{\\Gamma \\! \\left( \\sum_{k=1}^K \\alpha_k \\right)}{\\prod_{k=1}^K \\Gamma(\\alpha_k)}  \\prod_{k=1}^K \\theta_k^{\\alpha_k -1}\n\\end{equation*}\\]\nWarning: If any of the components of \\(\\theta\\) satisfies \\(\\theta_i = 0\\) or \\(\\theta_i = 1\\), then the probability is \\(0\\) and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, with \\(\\alpha_i &gt; 0\\) for each \\(i\\).\n\n\n\nA symmetric Dirichlet prior is \\([\\alpha, \\ldots, \\alpha]^{\\top}\\). To code this in Stan,\n data {\n   int&lt;lower=1&gt; K;\n   real&lt;lower=0&gt; alpha;\n }\n generated quantities {\n   vector[K] theta = dirichlet_rng(rep_vector(alpha, K));\n }\nTaking \\(K = 10\\), here are the first five draws for \\(\\alpha = 1\\). For \\(\\alpha = 1\\), the distribution is uniform over simplexes.\n 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05\n 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06\n 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03\n 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18\n 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01\nThat does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) \\(1/K\\). When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on \\(10\\)-simplexes, the probability that a component is greater than the mean of \\(1/10\\) is only 39%. Most of the posterior marginal probability mass for each component is in the interval \\((0, 0.1)\\).\nWhen the \\(\\alpha\\) value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for \\(\\alpha = 0.001\\).\n 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279\n 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000\n 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112\n 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000\n 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000\nEach row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.\nAs \\(\\alpha\\) increases, the draws become increasingly uniform. For \\(\\alpha = 1000\\),\n 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10\n 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n\n\n\ntheta ~ dirichlet(alpha)\nIncrement target log probability density with dirichlet_lupdf(theta | alpha).\nAvailable since 2.0\n \n\n\n\n\nThe Dirichlet probability functions are overloaded to allow the simplex \\(\\theta\\) and prior counts (plus one) \\(\\alpha\\) to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal dirichlet_lpdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha\nAvailable since 2.12, vectorized in 2.21\n \n\nreal dirichlet_lupdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha dropping constant additive terms\nAvailable since 2.25\n \n\nvector dirichlet_rng(vector alpha) Generate a Dirichlet variate with prior counts (plus one) alpha; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "text": "A variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (_) character. Variables must start with a letter (a--z and A--Z) and may not end with two underscores (__).\nExamples of legal variable identifiers are as follows.\na, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable\nUnlike in R and BUGS, variable identifiers in Stan may not contain a period character.\n\n\nStan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the stanc translator halting with an error message indicating which reserved name was used and its location in the model code.\n\n\nThe name of the model cannot be used as a variable within the model. This is usually not a problem because the default in bin/stanc is to append _model to the name of the file containing the model specification. For example, if the model is in file foo.stan, it would not be legal to have a variable named foo_model when using the default model name through bin/stanc. With user-specified model names, variables cannot match the model.\n\n\n\nThe following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.\nfor, in, while, repeat, until, if, then, else,\ntrue, false, target, struct, typedef, export,\nauto, extern, var, static, lower, upper, offset,\nmultiplier\nVariables should not be named after types, either, and thus may not be any of the following.\nint, real, complex, vector, simplex, unit_vector,\nsum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,\nrow_vector, matrix, cholesky_factor_corr,\ncolumn_stochastic_matrix,row_stochastic_matrix,\ncholesky_factor_cov, corr_matrix, cov_matrix, array\nThe following keywords or built-in functions are also reserved and cannot be used as variable names:\nprint, reject, profile, fatal_error, target, jacobian\nThe following block identifiers are reserved and cannot be used as variable names:\nfunctions, model, data, parameters, quantities,\ntransformed, generated\n\n\n\nVariable names will also conflict with the names of distributions suffixed with _lpdf, _lpmf, _lcdf, and _lccdf, _cdf, and _ccdf, such as normal_lcdf_log. No user-defined variable can take a name ending in _lupdf or _lupmf even if a corresponding _lpdf or _lpmf is not defined.\nUsing any of these variable names causes the stanc translator to halt and report the name and location of the variable causing the conflict.\n\n\n\nStan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed _stan_NAME (e.g. _stan_public) behind the scenes (in the generated C++ code).\n\nalignas, alignof, and, and_eq, asm, bitand, bitor, bool,\ncase, catch, char, char16_t, char32_t, class, compl, const,\nconstexpr, const_cast, decltype, default, delete, do,\ndouble, dynamic_cast, enum, explicit, float, friend, goto,\ninline, long, mutable, namespace, new, noexcept, not, not_eq,\nnullptr, operator, or, or_eq, private, protected, public,\nregister, reinterpret_cast, short, signed, sizeof,\nstatic_assert, static_cast, switch, template, this, thread_local,\nthrow, try, typeid, typename, union, unsigned, using, virtual,\nvolatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,\nSTAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH\n\n\n\n\nThe legal characters for variable identifiers are given in the following table.\n\n\nThe alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.\n\n\n\ncharacters\nASCII code points\n\n\n\n\na -- z\n97 – 122\n\n\nA -- Z\n65 – 90\n\n\n0 -- 9\n48 – 57\n\n\n_\n95\n\n\n\nAlthough not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.\n\n\n\nWithin comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.",
     "crumbs": [
-      "Functions Reference",
-      "Continuous Distributions",
-      "Simplex Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/simplex_distributions.html#dirichlet-distribution",
-    "href": "functions-reference/simplex_distributions.html#dirichlet-distribution",
-    "title": "Simplex Distributions",
+    "objectID": "reference-manual/expressions.html#vector-matrix-array-expressions.section",
+    "href": "reference-manual/expressions.html#vector-matrix-array-expressions.section",
+    "title": "Expressions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\) and \\(\\alpha \\in (\\mathbb{R}^+)^{K}\\), then for \\(\\theta \\in \\text{$K$-simplex}\\),\n\\[\\begin{equation*}\n\\text{Dirichlet}(\\theta|\\alpha) =\n\\frac{\\Gamma \\! \\left( \\sum_{k=1}^K \\alpha_k \\right)}{\\prod_{k=1}^K \\Gamma(\\alpha_k)}  \\prod_{k=1}^K \\theta_k^{\\alpha_k -1}\n\\end{equation*}\\]\nWarning: If any of the components of \\(\\theta\\) satisfies \\(\\theta_i = 0\\) or \\(\\theta_i = 1\\), then the probability is \\(0\\) and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, with \\(\\alpha_i &gt; 0\\) for each \\(i\\).\n\n\n\nA symmetric Dirichlet prior is \\([\\alpha, \\ldots, \\alpha]^{\\top}\\). To code this in Stan,\n data {\n   int&lt;lower=1&gt; K;\n   real&lt;lower=0&gt; alpha;\n }\n generated quantities {\n   vector[K] theta = dirichlet_rng(rep_vector(alpha, K));\n }\nTaking \\(K = 10\\), here are the first five draws for \\(\\alpha = 1\\). For \\(\\alpha = 1\\), the distribution is uniform over simplexes.\n 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05\n 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06\n 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03\n 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18\n 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01\nThat does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) \\(1/K\\). When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on \\(10\\)-simplexes, the probability that a component is greater than the mean of \\(1/10\\) is only 39%. Most of the posterior marginal probability mass for each component is in the interval \\((0, 0.1)\\).\nWhen the \\(\\alpha\\) value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for \\(\\alpha = 0.001\\).\n 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279\n 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000\n 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112\n 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000\n 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000\nEach row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.\nAs \\(\\alpha\\) increases, the draws become increasingly uniform. For \\(\\alpha = 1000\\),\n 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10\n 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n\n\n\ntheta ~ dirichlet(alpha)\nIncrement target log probability density with dirichlet_lupdf(theta | alpha).\nAvailable since 2.0\n \n\n\n\n\nThe Dirichlet probability functions are overloaded to allow the simplex \\(\\theta\\) and prior counts (plus one) \\(\\alpha\\) to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal dirichlet_lpdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha\nAvailable since 2.12, vectorized in 2.21\n \n\nreal dirichlet_lupdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha dropping constant additive terms\nAvailable since 2.25\n \n\nvector dirichlet_rng(vector alpha) Generate a Dirichlet variate with prior counts (plus one) alpha; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "text": "Expressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.\n\n\nSquare brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression [ 1, 10, 100 ] denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.\nrow_vector[2] rv2 =  [ 1, 2 ];\nvector[3] v3 = [ 3, 4, 5 ]';\nThe vector expression values may be compound expressions or variable names, so it is legal to write [ 2 * 3, 1 + 4] or [ x, y ], providing that x and y are primitive variables.\n\n\n\nA matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.\nmatrix[3, 2] m1 = [ [ 1, 2 ], [ 3, 4 ], [5, 6 ] ];\nAny expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:\nvector[2] vX = [ 1, 10 ]';\nrow_vector[2] vY = [ 100, 1000 ];\nmatrix[3, 2] m2 = [ vX', vY, [ 1, 2 ]  ];\n\n\n\nComplex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.\ncomplex_vector[3] = [1 + 2i, 3 - 1.7i, 0]';\ncomplex_row_vector[2] = [12, -2i];\ncomplex_matrix[2, 3] = [[1 + 2i, 3 - 1.7i, 0],\n                        [3.9 - 1.234i, 176i, 1 + 1i]];\n\n\nThe empty expression [ ] is ambiguous and therefore is not allowed and similarly expressions such as [ [ ] ] or [ [ ], [ ] ] are not allowed.\n\n\n\n\nIf needed, it is possible to create an empty vector with\nrep_vector(e, 0)\nwhere the first expression e needs to scalar of type real.\nIf needed, it is possible to create an empty matrix with\nrep_matrix(e, 0, 0)\nwhere the first expression e needs to scalar of type real.\n\n\n\nCurly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression { 1, 10, 100 } denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.\narray[3] int a = { 1, 10, 100 };\nThe values may be compound expressions, so it is legal to write { 2 * 3, 1 + 4 }. It is also possible to write two dimensional arrays directly, as in the following example.\narray[2, 3] int b = { { 1, 2, 3 }, { 4, 5, 6 } };\nThis way, b[1] is { 1, 2, 3 } and b[2] is { 4, 5, 6 }.\nWhitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.\narray[2, 3] int b = { { 1, 2, 3 },\n                { 4, 5, 6 } };\n\n\n\nThe empty array expression ({ }) is not allowed. See more about restrictions on array expressions in subsection Restrictions on values.\nIf needed, it is possible to create an empty array with\nrep_array(e, 0)\nwhere the first expression e determines the type of the array. For example, rep_array(0.0, 0) returns an empty real array of type real[], whereas rep_array({123}, 0) returns an empty two dimensional integer array of type int[ , ]. Only the type of the first argument is used, so the integer arrays {123} and {0} produce equivalent values.\n\n\n\nAny type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.\nvector[3] b;\nvector[3] c;\n// ...\narray[2] vector[3] d = { b, c };\nThe elements may also be a mixture of int and real typed expressions, in which case the result is an array of real values.\narray[2] real b = { 1, 1.9 };\n\n\n\nStan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.\ntuple(int, vector[3]) xy = (42, [1, 2.9, -1.3]');\nThe expression 42 is of type int and the expression [1, 2.9, -1.3] is of type row_vector so that [1, 2.9, -1.3]' is of type vector and of size 3. The whole tuple expression (42, [1, 2.9, -1.3]') thus has a sized type of tuple(int, vector[3]) and an unsized type (e.g., for a function argument) of tuple(int, vector).\nA tuple of one element can be created using the same style as languages like Python, with a trailing comma, e.g., (3.14,). For longer tuples, Stan does not support trailing commas.\n\n\n\nThere are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.\n\n\nAlthough it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.\n{ { 1, 2, 3 }, { 4, 5 } }  // compile time error: size mismatch\nThis may appear to be OK, because it is creating a two-dimensional integer array (array[,] int) out of two one-dimensional array integer arrays (array[] int). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.\n{ { 1, 2, 3 }, m }  // runtime error if m not size 3\n\n\n\nBecause there is no way to infer the type of the result, the empty array expression ({ }) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.\narray[0] int a;   // a is fully defined as zero element array\n\n\n\nThere is no way to declare or construct a zero-tuple in Stan. Tuples must be at least one element long. The expression () does not pick out a zero-tuple—it is ill formed.\nOne-tuples need a trailing comma, like (1,). The expression (1), without the comma, is of type int rather than a tuple.",
     "crumbs": [
-      "Functions Reference",
-      "Continuous Distributions",
-      "Simplex Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/references.html",
-    "href": "functions-reference/references.html",
-    "title": "References",
-    "section": "",
-    "text": "References\n\n\n\n\n Back to top"
-  },
-  {
-    "objectID": "functions-reference/positive_lower-bounded_distributions.html",
-    "href": "functions-reference/positive_lower-bounded_distributions.html",
-    "title": "Positive Lower-Bounded Distributions",
+    "objectID": "reference-manual/expressions.html#parentheses-for-grouping",
+    "href": "reference-manual/expressions.html#parentheses-for-grouping",
+    "title": "Expressions",
     "section": "",
-    "text": "The positive lower-bounded probabilities have support on real values above some positive minimum value.\n\n\n\n\nIf \\(y_{\\text{min}} \\in \\mathbb{R}^+\\) and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\) with \\(y \\geq y_{\\text{min}}\\), \\[\\begin{equation*}\n\\text{Pareto}(y|y_{\\text{min}},\\alpha) = \\frac{\\displaystyle \\alpha\\,y_{\\text{min}}^\\alpha}{\\displaystyle y^{\\alpha+1}}.\n\\end{equation*}\\]\n\n\n\ny ~ pareto(y_min, alpha)\nIncrement target log probability density with pareto_lupdf(y | y_min, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal pareto_lpdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lupdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_cdf(reals y | reals y_min, reals alpha) The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.0\n \n\nreal pareto_lcdf(reals y | reals y_min, reals alpha) The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lccdf(reals y | reals y_min, reals alpha) The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nR pareto_rng(reals y_min, reals alpha) Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\lambda \\in \\mathbb{R}^+\\), and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\geq \\mu\\), \\[\\begin{equation*}\n\\mathrm{Pareto\\_Type\\_2}(y|\\mu,\\lambda,\\alpha) =\n\\ \\frac{\\alpha}{\\lambda} \\, \\left( 1+\\frac{y-\\mu}{\\lambda} \\right)^{-(\\alpha+1)} \\! .\n\\end{equation*}\\]\nNote that the Lomax distribution is a Pareto Type 2 distribution with \\(\\mu=0\\).\n\n\n\ny ~ pareto_type_2(mu, lambda, alpha)\nIncrement target log probability density with pareto_type_2_lupdf(y | mu, lambda, alpha).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal pareto_type_2_lpdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lupdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_type_2_cdf(reals y | reals mu, reals lambda, reals alpha) The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.5\n \n\nreal pareto_type_2_lcdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lccdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nR pareto_type_2_rng(reals mu, reals lambda, reals alpha) Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nFor an extended explanation of how to use the wiener_lpdf and wiener_l[c]cdf_unnorm functions, see Henrich et al. (2024).\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), \\(\\tau \\in \\mathbb{R}^+\\), \\(\\beta \\in (0, 1)\\), \\(\\delta \\in \\mathbb{R}\\), \\(s_{\\delta} \\in \\mathbb{R}^{\\geq 0}\\), \\(s_{\\beta} \\in [0, 1)\\), and \\(s_{\\tau} \\in \\mathbb{R}^{\\geq 0}\\) then for \\(y &gt; \\tau\\),\n\\[\\begin{equation*}\n\\begin{split}\n&\\text{Wiener}(y\\mid \\alpha,\\tau,\\beta,\\delta,s_{\\delta},s_{\\beta},s_{\\tau}) =\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}}\\int_{-\\infty}^{\\infty} p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega)\n\\\\\n&\\times \\frac{1}{\\sqrt{2\\pi s_{\\delta}^2}}\\exp\\Bigl(-\\frac{(\\nu-\\delta)^2}{2s_{\\delta}^2}\\Bigr) \\,d\\nu \\,d\\omega \\,d{\\tau_0}=\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}} M\\times p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega) \\,d\\omega \\,d{\\tau_0},\n\\end{split}\n\\end{equation*}\\]\nwhere \\(p()\\) denotes the density function, and \\(M\\) and \\(p_3()\\) are defined, by using \\(t:=y-{\\tau_0}\\), as\n\\[\\begin{equation*}\nM \\coloneqq \\frac{1}{\\sqrt{1+s_{\\delta}^2t}}\\exp\\Bigl(\\alpha{\\delta}\\omega+\\frac{\\delta^2t}{2}+\\frac{s_{\\delta}^2\\alpha^2\\omega^2-2\\alpha{\\delta}\\omega-\\delta^2t}{2(1+s_{\\delta}^2t)}\\Bigr)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\np_3(t\\mid \\alpha,\\delta,\\beta) \\coloneqq \\frac{1}{\\alpha^2}\\exp\\Bigl(-\\alpha\\delta\\beta-\\frac{\\delta^2t}{2}\\Bigr)f(\\frac{t}{\\alpha^2}\\mid 0,1,\\beta),\n\\end{equation*}\\]\nwhere \\(f(t^*=\\frac{t}{\\alpha^2}\\mid0,1,\\beta)\\) can be specified in two ways:\n\\[\\begin{equation*}\nf_l(t^*\\mid 0,1,\\beta) = \\sum_{k=1}^\\infty k\\pi \\exp\\Bigl(-\\frac{k^2\\pi^2t^*}{2}\\Bigr)\\sin(k\\pi \\beta)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\nf_s(t^*\\mid0,1,\\beta) = \\sum_{k=-\\infty}^\\infty \\frac{1}{\\sqrt{2\\pi(t^*)^3}}(\\beta+2k) \\exp\\Bigl(-\\frac{(\\beta+2k)^2}{2t^*}\\Bigr).\n\\end{equation*}\\]\nWhich of these is used in the computations depends on which expression requires the smaller number of components \\(k\\) to guarantee a pre-specified precision\nIn the case where \\(s_{\\delta}\\), \\(s_{\\beta}\\), and \\(s_{\\tau}\\) are all \\(0\\), this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\alpha}{(y-\\tau)^{3/2}} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = - \\infty}^{\\infty} (2k +\n\\beta) \\phi \\! \\left(\\frac{(2k + \\beta)\\alpha }{\\sqrt{y - \\tau}}\\right),\n\\end{equation*}\\] where \\(\\phi(x)\\) denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\pi}{\\alpha^2} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = 1}^{\\infty} k \\exp \\! \\left(-\\frac{k^2\\pi^2(y-\\tau)};\n{2\\alpha^2}\\right) \\sin \\!(k\\pi\\beta)\n\\end{equation*}\\] see (Feller 1968), (Navarro and Fuss 2009).\n\n\n\nFor the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.\nLet \\(\\alpha\\), \\(\\tau\\), \\(\\beta\\), \\(\\delta\\), \\(s_{\\delta}\\), \\(s_{\\beta}\\), \\(s_{\\tau}\\) and \\(y\\) be as above.\nThe formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in \\(\\tau\\) for the time being) at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha, \\beta, \\delta) = P(\\alpha, \\beta, \\delta) -\n    \\exp\\left(\\delta\\alpha(1-\\beta)-\\frac{\\delta^2 y}{2}\\right)F_l(y|\\alpha,\\beta,\\delta),\n\\end{equation}\\] where \\(P(\\alpha,\\beta,\\delta)\\) is the probability to hit the upper boundary, defined as\n\\[\\begin{equation}\nP(\\alpha, \\beta, \\delta) =\n\\begin{cases}\n    \\frac{1-\\exp(2\\delta \\alpha \\beta)}{\\exp(-2\\delta \\alpha(1-\\beta)) - \\exp(2\\delta \\alpha \\beta)}, & \\text{for } \\delta\\neq 0 \\\\\n    \\beta, & \\text{for } \\delta=0,\n\\end{cases}\n\\end{equation}\\]\nand\n\\[\\begin{equation}\nF_l(y|\\alpha, \\beta, \\delta) =\n    \\frac{2\\pi}{\\alpha^2}\\sum_{k=1}^{\\infty}{\\frac{k\\sin{k\\pi(1-\\beta)}}{\\delta^2+(k\\pi)^2/\\alpha^2}}\\exp(-\\frac{k^2\\pi^2y}{2\\alpha^2}).\n\\end{equation}\\]\nThe formula for the small-time cdf at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha,\\beta,\\delta) = \\exp\\left(\\delta \\alpha(1-\\beta) -\\frac{\\delta^2y}{2}\\right)F_s(y|\\alpha, \\beta,\\delta),\n\\end{equation}\\] where\n\\[\\begin{equation}\nF_s(y|\\alpha,\\beta,\\delta) = \\sum_{k=0}^{\\infty}(-1)^k\\phi\\left(\\frac{\\alpha(k+\\beta^{*}_k)}\n    {\\sqrt{y}} \\right) \\times \\left( R \\left(\\frac{\\alpha(k+\\beta^{*}_k)+\\delta y}{\\sqrt{y}} \\right)\n        + R \\left(\\frac{\\alpha(k+\\beta^*_k)-\\delta y}{\\sqrt{y}} \\right)\\right),\n\\end{equation}\\]\nwhere \\(\\beta^*_k=(1-\\beta)\\) for \\(k\\) even, \\(\\beta^*_k=\\beta\\) for \\(k\\) odd, and \\(R\\) is Mill’s ratio.\nThe cdf for the lower boundary is \\(F(y|\\alpha,1-\\beta,-\\delta)\\)\n\n\n\ny ~ wiener(alpha, tau, beta, delta)\nIncrement target log probability density with wiener_lupdf(y | alpha, tau, beta, delta).\nAvailable since 2.7\ny ~ wiener(alpha, tau, beta, delta, var_delta) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta).\nAvailable since 2.35\ny ~ wiener(alpha, tau, beta, delta, var_delta, var_beta, var_tau) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau).\nAvailable since 2.35\n \n\n\n\n\n \n\nreal wiener_lpdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.18\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta, dropping constant additive terms\nAvailable since 2.25\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta, dropping constant additive terms.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau, dropping constant additive terms.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n\nNote: The lcdf and lccdf functions for the wiener distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper boundary.\nSimilarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical \\(1 - \\textrm{cdf}\\).\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n\n\n\nStan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use \\[\\begin{equation*}\n\\text{Wiener}(y | \\alpha, \\tau, 1 - \\beta, - \\delta)\n\\end{equation*}\\] For more details, see the appendix of Vandekerckhove and Wabersich (2014).\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions (listed above as recieving only real) are implemented in such a way where they can be fully vectorized, but currently only versions that accept all real and all vector arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by opening an issue.\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions can also accept an additional data real argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to 1e-4 for the density and 1e-8 for the cdf functions.",
+    "text": "Any expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, ( and ), are allowed. The square brackets [ and ] are reserved for array indexing and the curly braces { and } for grouping statements.\nWith parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression 1 + 2 * 3 has a subexpression 2 * 3 and evaluates to 7. With parentheses, this grouping may be made explicit with the expression 1 + (2 * 3). More importantly, the expression (1 + 2) *   3 has 1 + 2 as a subexpression and evaluates to 9.",
     "crumbs": [
-      "Functions Reference",
-      "Continuous Distributions",
-      "Positive Lower-Bounded Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_lower-bounded_distributions.html#pareto-distribution",
-    "href": "functions-reference/positive_lower-bounded_distributions.html#pareto-distribution",
-    "title": "Positive Lower-Bounded Distributions",
+    "objectID": "reference-manual/expressions.html#arithmetic-expressions.section",
+    "href": "reference-manual/expressions.html#arithmetic-expressions.section",
+    "title": "Expressions",
     "section": "",
-    "text": "If \\(y_{\\text{min}} \\in \\mathbb{R}^+\\) and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\) with \\(y \\geq y_{\\text{min}}\\), \\[\\begin{equation*}\n\\text{Pareto}(y|y_{\\text{min}},\\alpha) = \\frac{\\displaystyle \\alpha\\,y_{\\text{min}}^\\alpha}{\\displaystyle y^{\\alpha+1}}.\n\\end{equation*}\\]\n\n\n\ny ~ pareto(y_min, alpha)\nIncrement target log probability density with pareto_lupdf(y | y_min, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal pareto_lpdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lupdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_cdf(reals y | reals y_min, reals alpha) The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.0\n \n\nreal pareto_lcdf(reals y | reals y_min, reals alpha) The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lccdf(reals y | reals y_min, reals alpha) The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nR pareto_rng(reals y_min, reals alpha) Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "For integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (+), subtraction (-), multiplication (*) and division (/) in the usual ways.\nFor integer expressions, Stan supports the modulus (%) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3.0 + 0.14\n-15\n2 * 3 + 1\n(x - y) / 2.0\n(n * (n + 1)) / 2\nx / n\nm % n\nThe negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (') is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.\nFor example, if y and mu are variables of type vector and Sigma is a variable of type matrix, then (y - mu)' * Sigma * (y - mu) is a well-formed expression of type real. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) y - mu are of type vector because the variables y and mu are of type vector. The transpose of this expression, the subexpression (y - mu)' is of type row_vector. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form (((y - mu)') * Sigma) * (y - mu).\nThe type of subexpression (y - mu)' * Sigma is inferred to be row_vector, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a real value.\nStan provides elementwise matrix multiplication (e.g., a .* b) and division (e.g., a ./ b) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,\nvector[N] a;\nvector[N] b;\nvector[N] c;\nthe assignment,\nc = a .* b;\nproduces the same result with roughly the same efficiency as the loop\nfor (n in 1:N) {\n  c[n] = a[n] * b[n];\n}\nStan supports exponentiation (^) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3 ^ 2\n3.0 ^ -2\n3.0 ^ 0.14\nx ^ n\nn ^ x\nn ^ m\nx ^ y\nExponentiation is right associative, so the expression 2 ^ 3 ^ 4 is equivalent to the fully specified form 2 ^ (3 ^ 4).\n\n\nThe precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.\n\n\nStan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e., operator*(int, int):int indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.\n\n\n\n\n\n\n\n\n\n\nOp.\nPrec.\nAssoc.\nPlacement\nDescription\n\n\n\n\n? ~ :\n10\nright\nternary infix\nconditional\n\n\n||\n9\nleft\nbinary infix\nlogical or\n\n\n&&\n8\nleft\nbinary infix\nlogical and\n\n\n==\n7\nleft\nbinary infix\nequality\n\n\n!=\n7\nleft\nbinary infix\ninequality\n\n\n&lt;\n6\nleft\nbinary infix\nless than\n\n\n&lt;=\n6\nleft\nbinary infix\nless than or equal\n\n\n&gt;\n6\nleft\nbinary infix\ngreater than\n\n\n&gt;=\n6\nleft\nbinary infix\ngreater than or equal\n\n\n+\n5\nleft\nbinary infix\naddition\n\n\n-\n5\nleft\nbinary infix\nsubtraction\n\n\n*\n4\nleft\nbinary infix\nmultiplication\n\n\n.*\n4\nleft\nbinary infix\nelementwise multiplication\n\n\n/\n4\nleft\nbinary infix\n(right) division\n\n\n./\n4\nleft\nbinary infix\nelementwise division\n\n\n%\n4\nleft\nbinary infix\nmodulus\n\n\n\\\n3\nleft\nbinary infix\nleft division\n\n\n%/%\n3\nleft\nbinary infix\ninteger division\n\n\n!\n2\nn/a\nunary prefix\nlogical negation\n\n\n-\n2\nn/a\nunary prefix\nnegation\n\n\n+\n2\nn/a\nunary prefix\npromotion (no-op in Stan)\n\n\n^\n1\nright\nbinary infix\nexponentiation\n\n\n.^\n1\nright\nbinary infix\nelementwise exponentiation\n\n\n'\n0\nn/a\nunary postfix\ntransposition\n\n\n()\n0\nn/a\nprefix, wrap\nfunction application\n\n\n[]\n0\nleft\nprefix, wrap\narray, matrix indexing\n\n\n\nOther expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.\nThe precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression a + b + c is interpreted as (a + b) + c. Similarly, a / b * c is interpreted as (a / b) * c.\nBecause multiplication has higher precedence than addition, the expression a * b + c is interpreted as (a * b) + c and the expression a + b * c is interpreted as a + (b * c). Similarly, 2 * x + 3 * - y is interpreted as (2 * x) + (3 * (-y)).\nTransposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, -u' is interpreted as -(u'), u * v' as u* (v'), and u' * v as (u') * v. For integer and reals, -n ^ 3 is interpreted as -(n ^ 3).",
     "crumbs": [
-      "Functions Reference",
-      "Continuous Distributions",
-      "Positive Lower-Bounded Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_lower-bounded_distributions.html#pareto-type-2-distribution",
-    "href": "functions-reference/positive_lower-bounded_distributions.html#pareto-type-2-distribution",
-    "title": "Positive Lower-Bounded Distributions",
+    "objectID": "reference-manual/expressions.html#conditional-operator.section",
+    "href": "reference-manual/expressions.html#conditional-operator.section",
+    "title": "Expressions",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\), \\(\\lambda \\in \\mathbb{R}^+\\), and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\geq \\mu\\), \\[\\begin{equation*}\n\\mathrm{Pareto\\_Type\\_2}(y|\\mu,\\lambda,\\alpha) =\n\\ \\frac{\\alpha}{\\lambda} \\, \\left( 1+\\frac{y-\\mu}{\\lambda} \\right)^{-(\\alpha+1)} \\! .\n\\end{equation*}\\]\nNote that the Lomax distribution is a Pareto Type 2 distribution with \\(\\mu=0\\).\n\n\n\ny ~ pareto_type_2(mu, lambda, alpha)\nIncrement target log probability density with pareto_type_2_lupdf(y | mu, lambda, alpha).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal pareto_type_2_lpdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lupdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_type_2_cdf(reals y | reals mu, reals lambda, reals alpha) The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.5\n \n\nreal pareto_type_2_lcdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lccdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nR pareto_type_2_rng(reals mu, reals lambda, reals alpha) Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "The ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If a is an expression of type int and b and c are expressions that can be converted to one another (e.g., compared with ==), then\na ? b : c\nis an expression of the promoted type of b and c. The result of this expression is b if condition a is true, c otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g. if one argument is of type int and the other of type real, the conditional expression as a whole is of type real. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.\n\n\nThe conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,\na &gt; 0 || b &lt; 0 ? c + d : e - f\nis equivalent to the explicitly grouped version\n(a &gt; 0 || b &lt; 0) ? (c + d) : (e - f)\nThe latter is easier to read even if the parentheses are not strictly necessary.\n\n\n\nThe conditional operator is right associative, so that\na ? b : c ? d : e\nparses as if explicitly grouped as\na ? b : (c ? d : e)\nAgain, the explicitly grouped version is easier to read.\n\n\n\n\nStan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable a in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value b is returned if the condition evaluates to a non-zero value and c is returned if the condition evaluates to zero.\n\n\nThe key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.\n\n\n\nIf one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full if-then conditional statement. For example,\ndata {\n  array[10] real x;\n  // ...\n}\nparameters {\n  array[10] real z;\n  // ...\n}\nmodel {\n  y ~ normal(cond ? x : z, sigma);\n  // ...\n}\nwould be more efficiently (if not more transparently) coded as\nif (cond) {\n  y ~ normal(x, sigma);\n} else {\n  y ~ normal(z, sigma);\n}\nThe conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable x will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.",
     "crumbs": [
-      "Functions Reference",
-      "Continuous Distributions",
-      "Positive Lower-Bounded Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_lower-bounded_distributions.html#wiener-first-passage-time-distribution",
-    "href": "functions-reference/positive_lower-bounded_distributions.html#wiener-first-passage-time-distribution",
-    "title": "Positive Lower-Bounded Distributions",
+    "objectID": "reference-manual/expressions.html#language-indexing.section",
+    "href": "reference-manual/expressions.html#language-indexing.section",
+    "title": "Expressions",
     "section": "",
-    "text": "For an extended explanation of how to use the wiener_lpdf and wiener_l[c]cdf_unnorm functions, see Henrich et al. (2024).\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), \\(\\tau \\in \\mathbb{R}^+\\), \\(\\beta \\in (0, 1)\\), \\(\\delta \\in \\mathbb{R}\\), \\(s_{\\delta} \\in \\mathbb{R}^{\\geq 0}\\), \\(s_{\\beta} \\in [0, 1)\\), and \\(s_{\\tau} \\in \\mathbb{R}^{\\geq 0}\\) then for \\(y &gt; \\tau\\),\n\\[\\begin{equation*}\n\\begin{split}\n&\\text{Wiener}(y\\mid \\alpha,\\tau,\\beta,\\delta,s_{\\delta},s_{\\beta},s_{\\tau}) =\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}}\\int_{-\\infty}^{\\infty} p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega)\n\\\\\n&\\times \\frac{1}{\\sqrt{2\\pi s_{\\delta}^2}}\\exp\\Bigl(-\\frac{(\\nu-\\delta)^2}{2s_{\\delta}^2}\\Bigr) \\,d\\nu \\,d\\omega \\,d{\\tau_0}=\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}} M\\times p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega) \\,d\\omega \\,d{\\tau_0},\n\\end{split}\n\\end{equation*}\\]\nwhere \\(p()\\) denotes the density function, and \\(M\\) and \\(p_3()\\) are defined, by using \\(t:=y-{\\tau_0}\\), as\n\\[\\begin{equation*}\nM \\coloneqq \\frac{1}{\\sqrt{1+s_{\\delta}^2t}}\\exp\\Bigl(\\alpha{\\delta}\\omega+\\frac{\\delta^2t}{2}+\\frac{s_{\\delta}^2\\alpha^2\\omega^2-2\\alpha{\\delta}\\omega-\\delta^2t}{2(1+s_{\\delta}^2t)}\\Bigr)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\np_3(t\\mid \\alpha,\\delta,\\beta) \\coloneqq \\frac{1}{\\alpha^2}\\exp\\Bigl(-\\alpha\\delta\\beta-\\frac{\\delta^2t}{2}\\Bigr)f(\\frac{t}{\\alpha^2}\\mid 0,1,\\beta),\n\\end{equation*}\\]\nwhere \\(f(t^*=\\frac{t}{\\alpha^2}\\mid0,1,\\beta)\\) can be specified in two ways:\n\\[\\begin{equation*}\nf_l(t^*\\mid 0,1,\\beta) = \\sum_{k=1}^\\infty k\\pi \\exp\\Bigl(-\\frac{k^2\\pi^2t^*}{2}\\Bigr)\\sin(k\\pi \\beta)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\nf_s(t^*\\mid0,1,\\beta) = \\sum_{k=-\\infty}^\\infty \\frac{1}{\\sqrt{2\\pi(t^*)^3}}(\\beta+2k) \\exp\\Bigl(-\\frac{(\\beta+2k)^2}{2t^*}\\Bigr).\n\\end{equation*}\\]\nWhich of these is used in the computations depends on which expression requires the smaller number of components \\(k\\) to guarantee a pre-specified precision\nIn the case where \\(s_{\\delta}\\), \\(s_{\\beta}\\), and \\(s_{\\tau}\\) are all \\(0\\), this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\alpha}{(y-\\tau)^{3/2}} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = - \\infty}^{\\infty} (2k +\n\\beta) \\phi \\! \\left(\\frac{(2k + \\beta)\\alpha }{\\sqrt{y - \\tau}}\\right),\n\\end{equation*}\\] where \\(\\phi(x)\\) denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\pi}{\\alpha^2} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = 1}^{\\infty} k \\exp \\! \\left(-\\frac{k^2\\pi^2(y-\\tau)};\n{2\\alpha^2}\\right) \\sin \\!(k\\pi\\beta)\n\\end{equation*}\\] see (Feller 1968), (Navarro and Fuss 2009).\n\n\n\nFor the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.\nLet \\(\\alpha\\), \\(\\tau\\), \\(\\beta\\), \\(\\delta\\), \\(s_{\\delta}\\), \\(s_{\\beta}\\), \\(s_{\\tau}\\) and \\(y\\) be as above.\nThe formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in \\(\\tau\\) for the time being) at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha, \\beta, \\delta) = P(\\alpha, \\beta, \\delta) -\n    \\exp\\left(\\delta\\alpha(1-\\beta)-\\frac{\\delta^2 y}{2}\\right)F_l(y|\\alpha,\\beta,\\delta),\n\\end{equation}\\] where \\(P(\\alpha,\\beta,\\delta)\\) is the probability to hit the upper boundary, defined as\n\\[\\begin{equation}\nP(\\alpha, \\beta, \\delta) =\n\\begin{cases}\n    \\frac{1-\\exp(2\\delta \\alpha \\beta)}{\\exp(-2\\delta \\alpha(1-\\beta)) - \\exp(2\\delta \\alpha \\beta)}, & \\text{for } \\delta\\neq 0 \\\\\n    \\beta, & \\text{for } \\delta=0,\n\\end{cases}\n\\end{equation}\\]\nand\n\\[\\begin{equation}\nF_l(y|\\alpha, \\beta, \\delta) =\n    \\frac{2\\pi}{\\alpha^2}\\sum_{k=1}^{\\infty}{\\frac{k\\sin{k\\pi(1-\\beta)}}{\\delta^2+(k\\pi)^2/\\alpha^2}}\\exp(-\\frac{k^2\\pi^2y}{2\\alpha^2}).\n\\end{equation}\\]\nThe formula for the small-time cdf at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha,\\beta,\\delta) = \\exp\\left(\\delta \\alpha(1-\\beta) -\\frac{\\delta^2y}{2}\\right)F_s(y|\\alpha, \\beta,\\delta),\n\\end{equation}\\] where\n\\[\\begin{equation}\nF_s(y|\\alpha,\\beta,\\delta) = \\sum_{k=0}^{\\infty}(-1)^k\\phi\\left(\\frac{\\alpha(k+\\beta^{*}_k)}\n    {\\sqrt{y}} \\right) \\times \\left( R \\left(\\frac{\\alpha(k+\\beta^{*}_k)+\\delta y}{\\sqrt{y}} \\right)\n        + R \\left(\\frac{\\alpha(k+\\beta^*_k)-\\delta y}{\\sqrt{y}} \\right)\\right),\n\\end{equation}\\]\nwhere \\(\\beta^*_k=(1-\\beta)\\) for \\(k\\) even, \\(\\beta^*_k=\\beta\\) for \\(k\\) odd, and \\(R\\) is Mill’s ratio.\nThe cdf for the lower boundary is \\(F(y|\\alpha,1-\\beta,-\\delta)\\)\n\n\n\ny ~ wiener(alpha, tau, beta, delta)\nIncrement target log probability density with wiener_lupdf(y | alpha, tau, beta, delta).\nAvailable since 2.7\ny ~ wiener(alpha, tau, beta, delta, var_delta) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta).\nAvailable since 2.35\ny ~ wiener(alpha, tau, beta, delta, var_delta, var_beta, var_tau) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau).\nAvailable since 2.35\n \n\n\n\n\n \n\nreal wiener_lpdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.18\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta, dropping constant additive terms\nAvailable since 2.25\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta, dropping constant additive terms.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau, dropping constant additive terms.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n\nNote: The lcdf and lccdf functions for the wiener distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper boundary.\nSimilarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical \\(1 - \\textrm{cdf}\\).\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n\n\n\nStan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use \\[\\begin{equation*}\n\\text{Wiener}(y | \\alpha, \\tau, 1 - \\beta, - \\delta)\n\\end{equation*}\\] For more details, see the appendix of Vandekerckhove and Wabersich (2014).\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions (listed above as recieving only real) are implemented in such a way where they can be fully vectorized, but currently only versions that accept all real and all vector arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by opening an issue.\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions can also accept an additional data real argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to 1e-4 for the density and 1e-8 for the cdf functions.",
+    "text": "Stan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if x is a variable of type array [] real (a one-dimensional array of reals) then x[1] is the value of the first element of the array.\nSubscripting has higher precedence than any of the arithmetic operations. For example, alpha * x[1] is equivalent to alpha * (x[1]).\nMultiple subscripts may be provided within a single pair of square brackets. If x is of type array[,] real, a two-dimensional array, then x[2, 501] is of type real.\n\n\nThe subscripting operator also returns subarrays of arrays. For example, if x is of type array[,,] real, then x[2] is of type array[,] real, and x[2, 3] is of type array[] real. As a result, the expressions x[2, 3] and x[2][3] have the same meaning.\n\n\n\nIf Sigma is a variable of type matrix, then Sigma[1] denotes the first row of Sigma and has the type row_vector.\n\n\n\nStan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if m is of type matrix[ , ], a two-dimensional array of matrices, then m[1] refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that m[1, 2] is of type matrix and denotes the matrix in the first row and second column of the array. Continuing to add indices, m[1, 2, 3] is of type row_vector and denotes the third row of the matrix denoted by m[1, 2]. Finally, m[1, 2, 3, 4] is of type real and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array m.",
     "crumbs": [
-      "Functions Reference",
-      "Continuous Distributions",
-      "Positive Lower-Bounded Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/multivariate_discrete_distributions.html",
-    "href": "functions-reference/multivariate_discrete_distributions.html",
-    "title": "Multivariate Discrete Distributions",
-    "section": "",
-    "text": "The multivariate discrete distributions are over multiple integer values, which are expressed in Stan as arrays.\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\theta \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\text{Multinomial}(y|\\theta) = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K \\theta_k^{y_k},\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial(theta)\nIncrement target log probability density with multinomial_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multinomial_lpmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y)\nAvailable since 2.12\n \n\nreal multinomial_lupmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_rng(vector theta, int N) Generate a multinomial variate with simplex distribution parameter theta and total count \\(N\\); may only be used in transformed data and generated quantities blocks\nAvailable since 2.8\n\n\n\n\nStan also provides a version of the multinomial probability mass function distribution with the \\(\\text{$K$-simplex}\\) for the event count probabilities per category given on the unconstrained logistic scale.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\text{softmax}(\\theta) \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{MultinomialLogit}(y \\mid \\gamma) & = \\text{Multinomial}(y \\mid \\text{softmax}(\\gamma)) \\\\\n& = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K [\\text{softmax}(\\gamma_k)]^{y_k},\n\\end{split}\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial_logit(gamma)\nIncrement target log probability density with multinomial_logit_lupmf(y | gamma).\nAvailable since 2.24\n \n\n\n\n\n \n\nreal multinomial_logit_lpmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y)\nAvailable since 2.24\n \n\nreal multinomial_logit_lupmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_logit_rng(vector gamma, int N) Generate a variate from a multinomial distribution with probabilities softmax(gamma) and total count N; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.24\n\n\n\n\nStan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\alpha \\in\n\\mathbb{R}_{+}^K\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), the PMF of the Dirichlet-multinomial distribution is defined as \\[\\begin{equation*}\n\\text{DirMult}(y|\\theta) =\n\\frac{\\Gamma(\\alpha_0)\\Gamma(N+1)}{\\Gamma(N+\\alpha_0)} \\prod_{k=1}^K \\frac{\\Gamma(y_k + \\alpha_k)}{\\Gamma(\\alpha_k)\\Gamma(y_k+1)},\n\\end{equation*}\\] where \\(\\alpha_0\\) is defined as \\(\\alpha_0 = \\sum_{k=1}^K \\alpha_k\\).\n\n\n\ny ~ dirichlet_multinomial(alpha)\nIncrement target log probability density with dirichlet_multinomial_lupmf(y | alpha).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal dirichlet_multinomial_lpmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y).\nAvailable since 2.34\n \n\nreal dirichlet_multinomial_lupmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements, given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y) dropping constant additive terms.\nAvailable since 2.34\n \n\narray[] int dirichlet_multinomial_rng(vector alpha, int N) Generate a multinomial variate with positive vector distribution parameter alpha and total count N; may only be used in transformed data and generated quantities blocks. This is equivalent to multinomial_rng(dirichlet_rng(alpha), N).\nAvailable since 2.34",
+    "objectID": "reference-manual/expressions.html#language-multi-indexing.section",
+    "href": "reference-manual/expressions.html#language-multi-indexing.section",
+    "title": "Expressions",
+    "section": "",
+    "text": "In addition to single integer indexes, as described in the language indexing section, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.\n\n\nTypes of indexes and examples with one-dimensional containers of size N and an integer array ii of type array [] real size K.\n\n\n\nindex type\nexample\nvalue\n\n\n\n\ninteger\na[11]\nvalue of a at index 11\n\n\ninteger array\na[ii]\na[ii[1]], …, a[ii[K]]\n\n\nlower bound\na[3:]\na[3], …, a[N]\n\n\nupper bound\na[:5]\na[1], …, a[5]\n\n\nrange\na[2:7]\na[2], …, a[7]\n\n\nrange\na[7:2]\n[]\n\n\nrange\na[5-3:5+2]\na[2], …, a[7]\n\n\nall\na[:]\na[1], …, a[N]\n\n\nall\na[]\na[1], …, a[N]\n\n\n\nThe range indexing with : allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:\n  array[6] int ii = reverse(linspaced_int_array(6, 2, 7));\nThen a[ii] evaluates to a[7], …, a[2].\n\n\n\nThe fundamental semantic rule for dealing with multiple indexes is the following. If idxs is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.\na[idxs, ...][i, ...] = a[idxs[i], ...][...]\nOn the other hand, if idx is a single index, it reduces the dimensionality of the output, so that\na[idx, ...] = a[idx][...]\nThe only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.\n\n\nSpecial rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix a, with integer single indexes i and j and integer array multiple indexes is and js. The same typing rules apply for all multiple indexes.\n\n\n\nexample\nrow index\ncolumn index\nresult type\n\n\n\n\na[i]\nsingle\nn/a\nrow vector\n\n\na[is]\nmultiple\nn/a\nmatrix\n\n\na[i, j]\nsingle\nsingle\nreal\n\n\na[i, js]\nsingle\nmultiple\nrow vector\n\n\na[is, j]\nmultiple\nsingle\nvector\n\n\na[is, js]\nmultiple\nmultiple\nmatrix\n\n\n\nEvaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.\nm[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]\nm[idxs, idx][j] = m[idxs[j], idx]\nm[idx, idxs][j] = m[idx, idxs[j]]\nEvaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Multivariate Discrete Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution",
-    "href": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution",
-    "title": "Multivariate Discrete Distributions",
+    "objectID": "reference-manual/expressions.html#function-application.section",
+    "href": "reference-manual/expressions.html#function-application.section",
+    "title": "Expressions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\theta \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\text{Multinomial}(y|\\theta) = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K \\theta_k^{y_k},\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial(theta)\nIncrement target log probability density with multinomial_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multinomial_lpmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y)\nAvailable since 2.12\n \n\nreal multinomial_lupmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_rng(vector theta, int N) Generate a multinomial variate with simplex distribution parameter theta and total count \\(N\\); may only be used in transformed data and generated quantities blocks\nAvailable since 2.8",
+    "text": "Stan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.\nExpressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, log(2.0) is the expression of type real denoting the result of applying the natural logarithm to the value of the real literal 2.0.\nSyntactically, function application has higher precedence than any of the other operators, so that y + log(x) is interpreted as y + (log(x)).\n\n\nEach function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as\nreal log(real);\nand the signature for the lmultiply function is\nreal lmultiply(real, real);\nA function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.\nreal mean(array [] real);\nreal mean(vector);\nThe first applies to a one-dimensional array of real values and the second to a vector.\nThe identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.\n\n\n\nConstants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants \\(\\pi\\) and \\(e\\) are represented as nullary functions named pi() and e(). See the Stan Functions Reference built-in constants section for a list of built-in constants.\n\n\n\nBecause of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.\nFor example, consider a situation in which the following two function signatures have been registered for foo.\nreal foo(real, real);\nint foo(int, int);\nThe use of foo in the expression foo(1.0, 1.0) resolves to foo(real, real), and thus the expression foo(1.0, 1.0) itself is assigned a type of real.\nBecause integers may be promoted to real values, the expression foo(1, 1) could potentially match either foo(real, real) or foo(int, int). The former requires two type promotions and the latter requires none, so foo(1, 1) is resolved to function foo(int, int) and is thus assigned the type int.\nThe expression foo(1, 1.0) has argument types (int, real) and thus does not explicitly match either function signature. By promoting the integer expression 1 to type real, it is able to match foo(real, real), and hence the type of the function expression foo(1, 1.0) is real.\nIn some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named bar with the following signatures.\nreal bar(real, int);\nreal bar(int, real);\nWith these signatures, the expression bar(1.0, 1) and bar(1, 1.0) resolve to the first and second of the above functions, respectively. The expression bar(1.0, 1.0) is illegal because real values may not be demoted to integers. The expression bar(1, 1) is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name bar is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with bar(1, 1) given the two declarations above, the Stan compiler will flag the expression as illegal.\n\n\n\nFor most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix _rng. For example, a univariate normal random number can be generated by normal_rng(0, 1); only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.\n\n\nThe use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in _rng.\nThis allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.\n\n\n\nPosterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of (Gelman et al. 2013) for more information on posterior predictive checks.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Multivariate Discrete Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution-logit-parameterization",
-    "href": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution-logit-parameterization",
-    "title": "Multivariate Discrete Distributions",
+    "objectID": "reference-manual/expressions.html#type-inference",
+    "href": "reference-manual/expressions.html#type-inference",
+    "title": "Expressions",
     "section": "",
-    "text": "Stan also provides a version of the multinomial probability mass function distribution with the \\(\\text{$K$-simplex}\\) for the event count probabilities per category given on the unconstrained logistic scale.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\text{softmax}(\\theta) \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{MultinomialLogit}(y \\mid \\gamma) & = \\text{Multinomial}(y \\mid \\text{softmax}(\\gamma)) \\\\\n& = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K [\\text{softmax}(\\gamma_k)]^{y_k},\n\\end{split}\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial_logit(gamma)\nIncrement target log probability density with multinomial_logit_lupmf(y | gamma).\nAvailable since 2.24\n \n\n\n\n\n \n\nreal multinomial_logit_lpmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y)\nAvailable since 2.24\n \n\nreal multinomial_logit_lupmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_logit_rng(vector gamma, int N) Generate a variate from a multinomial distribution with probabilities softmax(gamma) and total count N; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.24",
+    "text": "Stan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.\n\n\nThe primitive implementation types for Stan are\nint, real, complex, vector, row_vector,  matrix, complex_vector,\ncomplex_row_vector, complex_matrix\nEvery basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.\n\n\nThe table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.\n\n\n\ntype\nprimitive type\n\n\n\n\nint\nint\n\n\nreal\nreal\n\n\nvector\nvector\n\n\nsimplex\nvector\n\n\nunit_vector\nvector\n\n\nsum_to_zero_vector\nvector\n\n\nordered\nvector\n\n\npositive_ordered\nvector\n\n\nrow_vector\nrow_vector\n\n\nmatrix\nmatrix\n\n\ncov_matrix\nmatrix\n\n\ncorr_matrix\nmatrix\n\n\ncholesky_factor_cov\nmatrix\n\n\ncholesky_factor_corr\nmatrix\n\n\ncolumn_stochastic_matrix\nmatrix\n\n\nrow_stochastic_matrix\nmatrix\n\n\nsum_to_zero_matrix\nmatrix\n\n\ncomplex_vector\ncomplex_vector\n\n\ncomplex_row_vector\ncomplex_row_vector\n\n\ncomplex_matrix\ncomplex_matrix\n\n\n\nA full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, array [] real has an array dimensionality of 1, int an array dimensionality of 0, and array [,,] int an array dimensionality of 3. The implementation type matrix[ , , ] has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.\nRecall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.\narray[I, J, K] matrix[M, N] a;\nThe matrix a is indexed as a[i, j, k, m, n] with the array indices first, followed by the matrix indices, with a[i, j, k] being a matrix and a[i, j, k, m] being a row vector.\n\n\n\n\nStan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.\n\n\n\nThere are two basic promotion rules,\n\nint types may be promoted to real, and\nreal types may be promoted to complex.\n\nPlus, promotion is transitive, so that\n\nif type U can be promoted to type V and type V can be promoted to type T, then U can be promoted to T.\n\nThe first rule means that expressions of type int may be used anywhere an expression of type real is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.\nThe remaining rules have to do with covariant typing rules, which say that a container of type U may be promoted to a container of the same shape of type T if U can be promoted to T. For vector and matrix types, this induces three rules,\n\nvector may be promoted to complex_vector,\nrow_vector may be promoted to complex_row_vector\nmatrix may be promoted to complex_matrix.\n\nFor array types, there’s a single rule\n\narray[...] U may be promoted to array[...] T if U can be promoted to T.\n\nFor example, this means array[,] int may be used where array [,] real or array [,] complex is required; as another example, array[] real may be used anywhere array[] complex is required.\nTuples have the natural extension of the above rules, applied to all sub-types at once\n\nA tuple(U1, ..., UN) may be promoted to a tuple(T1, ..., TN) if every Un can be promoted to Tn for n in 1:N\n\n\n\nAn integer literal expression such as 42 is of type int. Real literals such as 42.0 are of type real. Imaginary literals such as -17i are of type complex. the expression 7 - 2i acts like a complex literal, but technically it combines a real literal 7 and an imaginary literal 2i through subtraction.\n\n\n\nThe type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is int.\nThere is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.1\n\n\n\nIf x is an expression of total dimensionality greater than or equal to \\(N\\), then the type of expression e[i1, i2, ..., iN] is the same as that of e[i1][i2]...[iN], so it suffices to define the type of a singly-indexed function. Suppose e is an expression and i is an expression of primitive type int. Then\n\nif e is an expression of type array[i1, i2, ..., iN] T and k, i1, …, iN are expressions of type int, then e[k] is an expression of type array[i2, ..., iN] T,\nif e is an expression of type array[i] T with i and k expressions of type int, then e[k] is of type T,\nif e has implementation type vector or row_vector, dimensionality 0, then e[i] has implementation type real,\nif e has implementation type matrix, then e[i] has type row_vector,\nif e has implementation type complex_vector or complex_row_vector and i is an expression of type int, then e[i] is an expression of type complex, and\nif e has implementation type complex_matrix, and i is an expression of type int, then e[i] is an expression of type complex_row_vector.\n\n\n\n\nIf f is the name of a function and e1,...,eN are expressions for \\(N \\geq 0\\), then f(e1,...,eN) is an expression whose type is determined by the return type in the function signature for f given e1 through eN. Recall that a function signature is a declaration of the argument types and the result type.\nIn looking up functions, binary operators like real * real are defined as operator*(real, real) in the documentation and index.\nIn matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type int may be promoted to type real or complex if necessary (see the subsection on type promotion in the function application section, a real argument will be promoted to complex if necessary, a vector will be promoted to complex_vector if necessary, and so on.\nIn general, matrix operations return the lowest inferable type. For example, row_vector * vector returns a value of type real, which is declared in the function documentation and index as real operator*(row_vector, vector).",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Multivariate Discrete Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/multivariate_discrete_distributions.html#dirichlet-multinomial-distribution",
-    "href": "functions-reference/multivariate_discrete_distributions.html#dirichlet-multinomial-distribution",
-    "title": "Multivariate Discrete Distributions",
+    "objectID": "reference-manual/expressions.html#higher-order-functions",
+    "href": "reference-manual/expressions.html#higher-order-functions",
+    "title": "Expressions",
     "section": "",
-    "text": "Stan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\alpha \\in\n\\mathbb{R}_{+}^K\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), the PMF of the Dirichlet-multinomial distribution is defined as \\[\\begin{equation*}\n\\text{DirMult}(y|\\theta) =\n\\frac{\\Gamma(\\alpha_0)\\Gamma(N+1)}{\\Gamma(N+\\alpha_0)} \\prod_{k=1}^K \\frac{\\Gamma(y_k + \\alpha_k)}{\\Gamma(\\alpha_k)\\Gamma(y_k+1)},\n\\end{equation*}\\] where \\(\\alpha_0\\) is defined as \\(\\alpha_0 = \\sum_{k=1}^K \\alpha_k\\).\n\n\n\ny ~ dirichlet_multinomial(alpha)\nIncrement target log probability density with dirichlet_multinomial_lupmf(y | alpha).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal dirichlet_multinomial_lpmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y).\nAvailable since 2.34\n \n\nreal dirichlet_multinomial_lupmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements, given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y) dropping constant additive terms.\nAvailable since 2.34\n \n\narray[] int dirichlet_multinomial_rng(vector alpha, int N) Generate a multinomial variate with positive vector distribution parameter alpha and total count N; may only be used in transformed data and generated quantities blocks. This is equivalent to multinomial_rng(dirichlet_rng(alpha), N).\nAvailable since 2.34",
+    "text": "There are several expression constructions in Stan that act as higher-order functions.2\nThe higher-order functions and the signature of their argument functions are listed in the following pair of tables.\n\n\nHigher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.\n\n\n\n\n\n\n\n\n\nfunction\nparameter or data args\ndata args\nreturn type\n\n\n\n\nalgebra_solver\nvector, vector\narray [] real, array [] real\nvector\n\n\nalgebra_solver_newton\nvector, vector\narray [] real, array [] real\nvector\n\n\nintegrate_1d,\nreal, real, array [] real\narray [] real, array [] real\nreal\n\n\nintegrate_ode_X,\nreal, array [] real, array [] real\narray [] real, array [] real\narray [] real\n\n\nmap_rect\nvector, vector\narray [] real, array [] real\nvector\n\n\n\nFor example, the integrate_ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  array [] real foo(real t,\n                    array [] real y,\n                    array [] real theta,\n                    array [] real x_r,\n                    array [] real x_i) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\narray[2] real y0;\nreal t0;\narray[T] real ts;\narray[1] real theta;\narray[0] real x_r;\narray[0] int x_i;\n// ...\narray[T, 2] real y_hat = integrate_ode_rk45(foo, y0, t0,\n                                              ts, theta, x_r, x_i);\nThe function argument is foo, the name of the user-defined function; as shown in the higher-order functions table, integrate_ode_rk45 takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.\n\n\n\nVariadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.\n\n\n\nfunction\nrestricted args\nreturn type\n\n\n\n\nsolve_X\nvector\nvector\n\n\node_X,\nvector, real, array [] real\nvector[]\n\n\nreduce_sum\narray[] T, T1, T2\nreal\n\n\n\nT, T1, and T2 can be any Stan type.\nFor example, the ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  vector foo(real t, vector y, real theta, vector beta,\n            array [] real x_i, int index) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\nvector[2] y0;\nreal t0;\narray[T] real ts;\nreal theta;\nvector[7] beta;\narray[10] int x_i;\nint index;\n// ...\nvector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,\n                              beta, x_i, index);\nThe function argument is foo, the name of the user-defined function. As shown in the variadic higher-order functions table, ode_rk45 takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of foo and returns an array of vectors.\n\n\n\nThe function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.\n\n\n\nSome of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.\nFor user-defined functions the qualifier data may be prepended to the type to restrict the argument to data-only variables.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Multivariate Discrete Distributions"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html",
-    "href": "functions-reference/matrix_operations.html",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/expressions.html#chain-rule-and-derivatives",
+    "href": "reference-manual/expressions.html#chain-rule-and-derivatives",
+    "title": "Expressions",
     "section": "",
-    "text": "int num_elements(vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.5\n \n\nint num_elements(row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.5\n \n\nint num_elements(matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.5\n \n\nint rows(vector x) The number of rows in the vector x\nAvailable since 2.0\n \n\nint rows(row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.0\n \n\nint rows(matrix x) The number of rows in the matrix x\nAvailable since 2.0\n \n\nint cols(vector x) The number of columns in the vector x, namely 1\nAvailable since 2.0\n \n\nint cols(row_vector x) The number of columns in the row vector x\nAvailable since 2.0\n \n\nint cols(matrix x) The number of columns in the matrix x\nAvailable since 2.0\n \n\nint size(vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15\nAvailable since 2.26\n\n\n\nStan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\nvector operator-(vector x) The negation of the vector x.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x) The negation of the row vector x.\nAvailable since 2.0\n \n\nmatrix operator-(matrix x) The negation of the matrix x.\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\nvector operator+(vector x, vector y) The sum of the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, row_vector y) The sum of the row vectors x and y.\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, matrix y) The sum of the matrices x and y\nAvailable since 2.0\n \n\nvector operator-(vector x, vector y) The difference between the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, row_vector y) The difference between the row vectors x and y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, matrix y) The difference between the matrices x and y\nAvailable since 2.0\n \n\nvector operator*(real x, vector y) The product of the scalar x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(real x, row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.0\n \n\nmatrix operator*(real x, matrix y) The product of the scalar x and the matrix y\nAvailable since 2.0\n \n\nvector operator*(vector x, real y) The product of the scalar y and vector x\nAvailable since 2.0\n \n\nmatrix operator*(vector x, row_vector y) The product of the vector x and row vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, real y) The product of the scalar y and row vector x\nAvailable since 2.0\n \n\nreal operator*(row_vector x, vector y) The product of the row vector x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, matrix y) The product of the row vector x and matrix y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, real y) The product of the scalar y and matrix x\nAvailable since 2.0\n \n\nvector operator*(matrix x, vector y) The product of the matrix x and vector y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, matrix y) The product of the matrices x and y\nAvailable since 2.0\n\n\n\n \n\nvector operator+(vector x, real y) The result of adding y to every entry in the vector x\nAvailable since 2.0\n \n\nvector operator+(real x, vector y) The result of adding x to every entry in the vector y\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, real y) The result of adding y to every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator+(real x, row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, real y) The result of adding y to every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator+(real x, matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.0\n \n\nvector operator-(vector x, real y) The result of subtracting y from every entry in the vector x\nAvailable since 2.0\n \n\nvector operator-(real x, vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, real y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator-(real x, row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, real y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator-(real x, matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.0\n \n\nvector operator/(vector x, real y) The result of dividing each entry in the vector x by y\nAvailable since 2.0\n \n\nrow_vector operator/(row_vector x, real y) The result of dividing each entry in the row vector x by y\nAvailable since 2.0\n \n\nmatrix operator/(matrix x, real y) The result of dividing each entry in the matrix x by y\nAvailable since 2.0\n\n\n\n\nMatrix transposition is represented using a postfix operator.\n \n\nmatrix operator'(matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.0\n \n\nrow_vector operator'(vector x) The transpose of the vector x, written as x'\nAvailable since 2.0\n \n\nvector operator'(row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.0\n\n\n\nElementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section function vectorization for the general cases.\n \n\nvector operator.*(vector x, vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nrow_vector operator.*(row_vector x, row_vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nmatrix operator.*(matrix x, matrix y) The elementwise product of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator./(real x, vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(row_vector x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nrow_vector operator./(row_vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(real x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(matrix x, matrix y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nmatrix operator./(matrix x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(real x, matrix y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator.^(vector x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(real x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(real x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(real x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n\n\n\n \n\nreal dot_product(vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(vector x, vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(row_vector x, row_vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(matrix x, matrix y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(vector x, vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(row_vector x, row_vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(matrix x, matrix y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nreal dot_self(vector x) The dot product of the vector x with itself\nAvailable since 2.0\n \n\nreal dot_self(row_vector x) The dot product of the row vector x with itself\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(matrix x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(matrix x) The dot product of the rows of x with themselves\nAvailable since 2.0\n\n\n \n\nmatrix tcrossprod(matrix x) The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}\\,\\text{x}^{\\top}\\).\nAvailable since 2.0\n \n\nmatrix crossprod(matrix x) The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}^{\\top}\\,\\text{x}\\).\nAvailable since 2.0\nThe following functions all provide shorthand forms for common expressions, which are also much more efficient.\n \n\nmatrix quad_form(matrix A, matrix B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nreal quad_form(matrix A, vector B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nmatrix quad_form_diag(matrix m, vector v) The quadratic form using the column vector v as a diagonal matrix, i.e., diag_matrix(v) * m * diag_matrix(v).\nAvailable since 2.3\n \n\nmatrix quad_form_diag(matrix m, row_vector rv) The quadratic form using the row vector rv as a diagonal matrix, i.e., diag_matrix(rv) * m * diag_matrix(rv).\nAvailable since 2.3\n \n\nmatrix quad_form_sym(matrix A, matrix B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal quad_form_sym(matrix A, vector B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal trace_quad_form(matrix A, matrix B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_quad_form(matrix A, vector B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_gen_quad_form(matrix D, matrix A, matrix B) The trace of a generalized quadratic form, i.e., trace(D * B' * A * B).\nAvailable since 2.0\n \n\nmatrix multiply_lower_tri_self_transpose(matrix x) The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if L is a matrix of the same dimensions as x with L(m,n) equal to x(m,n) for \\(\\text{n}\n\\leq \\text{m}\\) and L(m,n) equal to 0 if \\(\\text{n} &gt; \\text{m}\\), the result is the symmetric matrix \\(\\text{L}\\,\\text{L}^{\\top}\\). This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(vector v, matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(row_vector rv, matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, row_vector rv) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.0\n\n\n\n\n\n\n \n\nreal log_sum_exp(vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(row_vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(matrix x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n\n\n\n \n\nreal min(vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(row_vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(matrix x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(row_vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(matrix x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n\n\n\n \n\nreal sum(vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal prod(vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n\n\n\nFull definitions are provided for sample moments in section array reductions.\n \n\nreal mean(vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(row_vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(matrix x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(row_vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(matrix x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(row_vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(matrix x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n \n\nreal quantile(data row_vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data row_vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n\n\n\n\nThe following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\nvector rep_vector(real x, int m) Return the size m (column) vector consisting of copies of x.\nAvailable since 2.0\n \n\nrow_vector rep_row_vector(real x, int n) Return the size n row vector consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(real x, int m, int n) Return the m by n matrix consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.0\n \n\nmatrix rep_matrix(row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.0\nUnlike the situation with array broadcasting (see section array broadcasting), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.\n vector[3] x;\n x = rep_vector(1, 3);\n x = rep_vector(1.0, 3);\nThere are no integer vector or matrix types, so integer values are automatically promoted.\n\n\n \n\nmatrix symmetrize_from_lower_tri(matrix A)\nConstruct a symmetric matrix from the lower triangle of A.\nAvailable since 2.26\n\n\n\n\n \n\nmatrix add_diag(matrix m, row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, real d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.21\n \n\nvector diagonal(matrix x) The diagonal of the matrix x\nAvailable since 2.0\n \n\nmatrix diag_matrix(vector x) The diagonal matrix with diagonal x\nAvailable since 2.0\nAlthough the diag_matrix function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,\n y ~ multi_normal(mu, diag_matrix(square(sigma)));\nit is much more efficient to just use a univariate normal, which produces the same density,\n y ~ normal(mu, sigma);\nRather than writing m * diag_matrix(v) where m is a matrix and v is a vector, it is much more efficient to write diag_post_multiply(m, v) (and similarly for pre-multiplication). By the same token, it is better to use quad_form_diag(m, v) rather than quad_form(m, diag_matrix(v)).\n \n\nmatrix identity_matrix(int k) Create an identity matrix of size \\(k \\times k\\)\nAvailable since 2.26\n\n\n\n \n\narray[] real linspaced_array(int n, data real lower, data real upper) Create a real array of length n of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int linspaced_int_array(int n, int lower, int upper) Create a regularly spaced, increasing integer array of length n between lower and upper, inclusively. If (upper - lower) / (n - 1) is less than one, repeat each output (n - 1) / (upper - lower) times. If neither (upper - lower) / (n - 1) or (n - 1) / (upper - lower) are integers, upper is reduced until one of these is true.\nAvailable since 2.26\n \n\nvector linspaced_vector(int n, data real lower, data real upper) Create an n-dimensional vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\nrow_vector linspaced_row_vector(int n, data real lower, data real upper) Create an n-dimensional row-vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int one_hot_int_array(int n, int k) Create a one-hot encoded int array of length n with array[k] = 1\nAvailable since 2.26\n \n\narray[] real one_hot_array(int n, int k) Create a one-hot encoded real array of length n with array[k] = 1\nAvailable since 2.24\n \n\nvector one_hot_vector(int n, int k) Create an n-dimensional one-hot encoded vector with vector[k] = 1\nAvailable since 2.24\n \n\nrow_vector one_hot_row_vector(int n, int k) Create an n-dimensional one-hot encoded row-vector with row_vector[k] = 1\nAvailable since 2.24\n \n\narray[] int ones_int_array(int n) Create an int array of length n of all ones\nAvailable since 2.26\n \n\narray[] real ones_array(int n) Create a real array of length n of all ones\nAvailable since 2.26\n \n\nvector ones_vector(int n) Create an n-dimensional vector of all ones\nAvailable since 2.26\n \n\nrow_vector ones_row_vector(int n) Create an n-dimensional row-vector of all ones\nAvailable since 2.26\n \n\narray[] int zeros_int_array(int n) Create an int array of length n of all zeros\nAvailable since 2.26\n \n\narray[] real zeros_array(int n) Create a real array of length n of all zeros\nAvailable since 2.24\n \n\nvector zeros_vector(int n) Create an n-dimensional vector of all zeros\nAvailable since 2.24\n \n\nrow_vector zeros_row_vector(int n) Create an n-dimensional row-vector of all zeros\nAvailable since 2.24\n \n\nvector uniform_simplex(int n) Create an n-dimensional simplex with elements vector[i] = 1 / n for all \\(i \\in 1, \\dots, n\\)\nAvailable since 2.24\n\n\n\nStan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\nvector col(matrix x, int n) The n-th column of matrix x\nAvailable since 2.0\n \n\nrow_vector row(matrix x, int m) The m-th row of matrix x\nAvailable since 2.0\nThe row function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation x[m] is just an alternative way of writing row(x,m). The col function may not, be used as an lvalue, nor is there an indexing based shorthand for it.\n\n\n\n\n\nBlock operations may be used to extract a sub-block of a matrix.\n \n\nmatrix block(matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.0\nThe sub-row and sub-column operations may be used to extract a slice of row or column from a matrix\n \n\nvector sub_col(matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.0\n \n\nrow_vector sub_row(matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.0\n\n\n\nThe head operation extracts the first \\(n\\) elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.\n \n\nvector head(vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.0\n \n\nrow_vector head(row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.0\n \n\narray[] T head(array[] T sv, int n) Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector tail(vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.0\n \n\nrow_vector tail(row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.0\n \n\narray[] T tail(array[] T sv, int n) Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector segment(vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.0\n \n\nrow_vector segment(row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.10\n \n\narray[] T segment(array[] T sv, int i, int n) Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n\n\n\n\n\nStan’s matrix and vector concatenation operations append_col and append_row are like the operations cbind and rbind in R.\n\n\n \n\nmatrix append_col(matrix x, matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(matrix x, vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.5\n \n\nrow_vector append_col(row_vector x, row_vector y) Combine row vectors x and y of any size into another row vector by appending y to the end of x.\nAvailable since 2.5\n \n\nrow_vector append_col(real x, row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.12\n \n\nrow_vector append_col(row_vector x, real y) Append y to the end of x, returning another row vector.\nAvailable since 2.12\n\n\n\n \n\nmatrix append_row(matrix x, matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(matrix x, row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.5\n \n\nvector append_row(vector x, vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.5\n \n\nvector append_row(real x, vector y) Append x to the top of y, returning another vector.\nAvailable since 2.12\n \n\nvector append_row(vector x, real y) Append y to the bottom of x, returning another vector.\nAvailable since 2.12\n\n\n\n\n\n\nThe softmax function maps1 \\(y \\in \\mathbb{R}^K\\) to the \\(K\\)-simplex by \\[\\begin{equation*} \\text{softmax}(y)  = \\frac{\\exp(y)}\n{\\sum_{k=1}^K \\exp(y_k)}, \\end{equation*}\\] where \\(\\exp(y)\\) is the componentwise exponentiation of \\(y\\). Softmax is usually calculated on the log scale, \\[\\begin{eqnarray*} \\log \\text{softmax}(y) & = & \\ y - \\log \\sum_{k=1}^K\n\\exp(y_k) \\\\[4pt] & = & y - \\mathrm{log\\_sum\\_exp}(y). \\end{eqnarray*}\\] where the vector \\(y\\) minus the scalar \\(\\mathrm{log\\_sum\\_exp}(y)\\) subtracts the scalar from each component of \\(y\\).\nStan provides the following functions for softmax and its log.\n \n\nvector softmax(vector x) The softmax of x\nAvailable since 2.0\n \n\nvector log_softmax(vector x) The natural logarithm of the softmax of x\nAvailable since 2.0\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] int cumulative_sum(array[] int x) The cumulative sum of x\nAvailable since 2.30\n \n\narray[] real cumulative_sum(array[] real x) The cumulative sum of x\nAvailable since 2.0\n \n\nvector cumulative_sum(vector v) The cumulative sum of v\nAvailable since 2.0\n \n\nrow_vector cumulative_sum(row_vector rv) The cumulative sum of rv\nAvailable since 2.0\n\n\n\n\nThe Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.\nFor one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by \\(K_{ij} = k(x_i, x_j)\\) (where \\(x_i\\) is the \\(i^{th}\\) element of the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(x_i, y_j)\\).\nFor multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{x}_j)\\) (where \\(\\mathbf{x}_i\\) is the \\(i^{th}\\) vector in the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{y}_j)\\).\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|^2}{2l^2} \\right)\n\\]\n \n\nmatrix gp_exp_quad_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith bias \\(\\sigma_0\\) the dot product kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma_0^2 + \\mathbf{x}_i^T \\mathbf{x}_j\n\\]\n \n\nmatrix gp_dot_prod_cov(array[] real x, real sigma)\nGaussian process covariance with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(array[] real x1, array[] real x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x, real sigma)\nGaussian process covariance with dot product kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x1, vectors x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in multiple dimensions.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponential kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_exponential_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 3/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right) \\exp \\left( -\\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern32_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 5/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{5}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} + \\frac{5 |\\mathbf{x}_i - \\mathbf{x}_j|^2}{3l^2} \\right)\n\\exp \\left( -\\frac{\\sqrt{5} |\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern52_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\), length scale \\(l\\), and period \\(p\\), the periodic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left(-\\frac{2 \\sin^2 \\left( \\pi \\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{p} \\right) }{l^2} \\right)\n\\]\n \n\nmatrix gp_periodic_cov(array[] real x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(array[] real x1, array[] real x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x1, vectors x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\n\n\n\nIn general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.\n\n\n \n\nrow_vector operator/(row_vector b, matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.0\n \n\nmatrix operator/(matrix B, matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.5\n \n\nvector operator\\(matrix A, vector b) The left division of A by b; equivalently inverse(A) * b\nAvailable since 2.18\n \n\nmatrix operator\\(matrix A, matrix B) The left division of A by B; equivalently inverse(A) * B\nAvailable since 2.18\n\n\n\nThere are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix \\(\\text{tri}(A)\\) is used in the definitions and defined by \\[\\begin{equation*} \\text{tri}(A)[m,n] = \\left\\{\n\\begin{array}{ll} A[m,n] & \\text{if } m \\geq n, \\text{ and} \\\\[4pt] 0\n& \\text{otherwise}. \\end{array} \\right. \\end{equation*}\\] When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.\n \n\nvector mdivide_left_tri_low(matrix A, vector b) The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * b, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_left_tri_low(matrix A, matrix B) The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * B, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n \n\nrow_vector mdivide_right_tri_low(row_vector b, matrix A) The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form b * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_right_tri_low(matrix B, matrix A) The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form B * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n\n\n\n\nThere are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.\n \n\nmatrix mdivide_left_spd(matrix A, vector b) The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * b.\nAvailable since 2.12\n \n\nvector mdivide_left_spd(matrix A, matrix B) The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * B.\nAvailable since 2.12\n \n\nrow_vector mdivide_right_spd(row_vector b, matrix A) The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form b *inverse(A).\nAvailable since 2.12\n \n\nmatrix mdivide_right_spd(matrix B, matrix A) The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form B * inverse(A).\nAvailable since 2.12\n\n\n\nThe exponential of the matrix \\(A\\) is formally defined by the convergent power series: \\[\\begin{equation*} e^A = \\sum_{n=0}^{\\infty} \\dfrac{A^n}{n!}\n\\end{equation*}\\]\n \n\nmatrix matrix_exp(matrix A) The matrix exponential of A\nAvailable since 2.13\n \n\nmatrix matrix_exp_multiply(matrix A, matrix B) The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form matrix_exp(A) * B.\nAvailable since 2.18\n \n\nmatrix scale_matrix_exp_multiply(real t, matrix A, matrix B) The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form matrix_exp(t * A) * B.\nAvailable since 2.18\n\n\n\nReturns the nth power of the specific matrix: \\[\\begin{equation*} M^n = M_1 * ... * M_n \\end{equation*}\\]\n \n\nmatrix matrix_power(matrix A, int B) Matrix A raised to the power B.\nAvailable since 2.24\n\n\n\n\n\n \n\nreal trace(matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.0\n\n\n\n \n\nreal determinant(matrix A) The determinant of A\nAvailable since 2.0\n \n\nreal log_determinant(matrix A) The log of the absolute value of the determinant of A\nAvailable since 2.0\nreal log_determinant_spd(matrix A) The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.\nAvailable since 2.30\n\n\n\nIt is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix m and post-multiplying by a vector or matrix a, as in inverse(m) * a, it is better to code this using matrix division, as in m \\ a. The pre-multiplication case is similar, with b * inverse(m) being more efficiently coded as as b / m. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.\nWarning: The function inv(m) is the elementwise inverse function, which returns 1 / m[i, j] for each element.\n \n\nmatrix inverse(matrix A) Compute the inverse of A\nAvailable since 2.0\n \n\nmatrix inverse_spd(matrix A) Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.\nAvailable since 2.0\n \n\nmatrix chol2inv(matrix L) Compute the inverse of the matrix whose cholesky factorization is L. That is, for \\(A = L L^T\\), return \\(A^{-1}\\).\nAvailable since 2.26\n\n\n\nThe generalized inverse \\(M^+\\) of a matrix \\(M\\) is a matrix that satisfies \\(M M^+ M = M\\). For an invertible, square matrix \\(M\\), \\(M^+\\) is equivalent to \\(M^{-1}\\). The dimensions of \\(M^+\\) are equivalent to the dimensions of \\(M^T\\). The generalized inverse exists for any matrix, so the \\(M\\) may be singular or less than full rank.\nEven though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank (Golub and Pereyra 1973), meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix \\(A\\) as a function of \\(\\epsilon\\):\n\\[\nA = \\left(\n    \\begin{array}{cccc}\n    1 + \\epsilon & 2 & 1 \\\\\n    2 & 4 & 2\n    \\end{array}\n    \\right)\n\\]\nWhen \\(\\epsilon = 0\\), \\(A\\) is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If \\(\\epsilon \\neq 0\\), the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at \\(\\epsilon = 0\\), and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.\n \n\nmatrix generalized_inverse(matrix A) The generalized inverse of A\nAvailable since 2.26\n\n\n\n \n\ncomplex_vector eigenvalues(matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors(matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nvector eigenvalues_sym(matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.0\n \n\nmatrix eigenvectors_sym(matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.0\n \n\ntuple(matrix, vector) eigendecompose_sym(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\n \n\nmatrix qr_thin_Q(matrix A) The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A\nAvailable since 2.18\n \n\nmatrix qr_thin_R(matrix A) The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A\nAvailable since 2.18\n \n\ntuple(matrix, matrix) qr_thin(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_thin_Q(A), qr_thin_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nmatrix qr_Q(matrix A) The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A\nAvailable since 2.3\n \n\nmatrix qr_R(matrix A) The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A\nAvailable since 2.3\n \n tuple(matrix, matrix) qr(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_Q(A), qr_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nThe thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as \\[\\begin{equation*} A = Q \\, R. \\end{equation*}\\] Multiplying a column of an orthogonal matrix by \\(-1\\) still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by \\(-1\\) without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.\n\n\n\nEvery symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If \\(\\Sigma\\) is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector \\(L\\) such that \\[\\begin{equation*} \\Sigma = L \\, L^{\\top}. \\end{equation*}\\]\n \n\nmatrix cholesky_decompose(matrix A) The lower-triangular Cholesky factor of the symmetric positive-definite matrix A\nAvailable since 2.0\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(matrix A) The singular values of A in descending order\nAvailable since 2.0\n \n\nmatrix svd_U(matrix A) The left-singular vectors of A\nAvailable since 2.26\n \n\nmatrix svd_V(matrix A) The right-singular vectors of A\nAvailable since 2.26\n \n\ntuple(matrix, vector, matrix) svd(matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\n\n\nSee the sorting functions section for examples of how the functions work.\n \n\nvector sort_asc(vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nrow_vector sort_asc(row_vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nvector sort_desc(vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\nrow_vector sort_desc(row_vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(row_vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n\n\n\n \n\nvector reverse(vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.23\n \n\nrow_vector reverse(row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "text": "Derivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.\n\n\nUnlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(sqrt(x - x), 1);\n}\nAlgebraically, the distribution statement in the model could be reduced to\n  x ~ normal(0, 1);\nand it would seem the model should produce unit normal draws for x. But rather than canceling, the expression sqrt(x -   x) causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,\n\\[\n\\begin{array}{rcl}\n\\frac{d}{dx} \\sqrt{x - x}\n& = &\n\\frac{1}{2 \\sqrt{x - x}} \\times \\frac{d}{dx} (x - x)\n\\\\[4pt]\n& = &\n\\frac{1}{0} \\times (1 - 1)\n\\\\[4pt]\n& = &\n\\infty \\times 0\n\\\\[4pt]\n& = & \\mathrm{NaN}.\n\\end{array}\n\\]\nRather than the \\(x - x\\) canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.\nThe only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like sqrt(x - x) for which the chain rule produces not-a-number values.\n\n\n\nThe best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).\nFor example, compiling the above model to an executable sqrt-x-minus-x in CmdStan, the test can be run as\n&gt; ./sqrt-x-minus-x diagnose test=gradient\nwhich produces\n...\nTEST GRADIENT MODE\n\n Log probability=-0.393734\n\n param idx           value           model     finite diff           error\n         0       -0.887393             nan               0             nan\nEven though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#integer-valued-matrix-size-functions",
-    "href": "functions-reference/matrix_operations.html#integer-valued-matrix-size-functions",
-    "title": "Matrix Operations",
-    "section": "",
-    "text": "int num_elements(vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.5\n \n\nint num_elements(row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.5\n \n\nint num_elements(matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.5\n \n\nint rows(vector x) The number of rows in the vector x\nAvailable since 2.0\n \n\nint rows(row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.0\n \n\nint rows(matrix x) The number of rows in the matrix x\nAvailable since 2.0\n \n\nint cols(vector x) The number of columns in the vector x, namely 1\nAvailable since 2.0\n \n\nint cols(row_vector x) The number of columns in the row vector x\nAvailable since 2.0\n \n\nint cols(matrix x) The number of columns in the matrix x\nAvailable since 2.0\n \n\nint size(vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15\nAvailable since 2.26",
+    "objectID": "reference-manual/expressions.html#footnotes",
+    "href": "reference-manual/expressions.html#footnotes",
+    "title": "Expressions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nLanguages such as C++ and R allow the declaration of a variable of a given name in a narrower scope to hide (take precedence over for evaluation) a variable defined in a containing scope.↩︎\nInternally, they are implemented as their own expression types because Stan doesn’t have object-level functional types (yet).↩︎",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Expressions"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#matrix-arithmetic-operators",
-    "href": "functions-reference/matrix_operations.html#matrix-arithmetic-operators",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/encoding.html",
+    "href": "reference-manual/encoding.html",
+    "title": "Character Encoding",
     "section": "",
-    "text": "Stan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\nvector operator-(vector x) The negation of the vector x.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x) The negation of the row vector x.\nAvailable since 2.0\n \n\nmatrix operator-(matrix x) The negation of the matrix x.\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\nvector operator+(vector x, vector y) The sum of the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, row_vector y) The sum of the row vectors x and y.\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, matrix y) The sum of the matrices x and y\nAvailable since 2.0\n \n\nvector operator-(vector x, vector y) The difference between the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, row_vector y) The difference between the row vectors x and y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, matrix y) The difference between the matrices x and y\nAvailable since 2.0\n \n\nvector operator*(real x, vector y) The product of the scalar x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(real x, row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.0\n \n\nmatrix operator*(real x, matrix y) The product of the scalar x and the matrix y\nAvailable since 2.0\n \n\nvector operator*(vector x, real y) The product of the scalar y and vector x\nAvailable since 2.0\n \n\nmatrix operator*(vector x, row_vector y) The product of the vector x and row vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, real y) The product of the scalar y and row vector x\nAvailable since 2.0\n \n\nreal operator*(row_vector x, vector y) The product of the row vector x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, matrix y) The product of the row vector x and matrix y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, real y) The product of the scalar y and matrix x\nAvailable since 2.0\n \n\nvector operator*(matrix x, vector y) The product of the matrix x and vector y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, matrix y) The product of the matrices x and y\nAvailable since 2.0\n\n\n\n \n\nvector operator+(vector x, real y) The result of adding y to every entry in the vector x\nAvailable since 2.0\n \n\nvector operator+(real x, vector y) The result of adding x to every entry in the vector y\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, real y) The result of adding y to every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator+(real x, row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, real y) The result of adding y to every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator+(real x, matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.0\n \n\nvector operator-(vector x, real y) The result of subtracting y from every entry in the vector x\nAvailable since 2.0\n \n\nvector operator-(real x, vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, real y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator-(real x, row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, real y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator-(real x, matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.0\n \n\nvector operator/(vector x, real y) The result of dividing each entry in the vector x by y\nAvailable since 2.0\n \n\nrow_vector operator/(row_vector x, real y) The result of dividing each entry in the row vector x by y\nAvailable since 2.0\n \n\nmatrix operator/(matrix x, real y) The result of dividing each entry in the matrix x by y\nAvailable since 2.0",
+    "text": "The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.\n\n\nThe UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.\n\n\n\n\nAny bytes on a line after a line-comment sequence (// or #) are ignored up until the ASCII newline character (\\n). They may thus be written in any character encoding which is convenient.\nAny content after a block comment open sequence in ASCII (/*) up to the closing block comment (*/) is ignored, and thus may also be written in whatever character set is convenient.\n\n\n\nThe raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A print, reject, or fatal_error statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (\") when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.\nASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#transposition-operator",
-    "href": "functions-reference/matrix_operations.html#transposition-operator",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/encoding.html#content-characters",
+    "href": "reference-manual/encoding.html#content-characters",
+    "title": "Character Encoding",
     "section": "",
-    "text": "Matrix transposition is represented using a postfix operator.\n \n\nmatrix operator'(matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.0\n \n\nrow_vector operator'(vector x) The transpose of the vector x, written as x'\nAvailable since 2.0\n \n\nvector operator'(row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.0",
+    "text": "The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.\n\n\nThe UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#elementwise-functions",
-    "href": "functions-reference/matrix_operations.html#elementwise-functions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/encoding.html#comment-characters",
+    "href": "reference-manual/encoding.html#comment-characters",
+    "title": "Character Encoding",
     "section": "",
-    "text": "Elementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section function vectorization for the general cases.\n \n\nvector operator.*(vector x, vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nrow_vector operator.*(row_vector x, row_vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nmatrix operator.*(matrix x, matrix y) The elementwise product of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator./(real x, vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(row_vector x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nrow_vector operator./(row_vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(real x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(matrix x, matrix y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nmatrix operator./(matrix x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(real x, matrix y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator.^(vector x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(real x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(real x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(real x, matrix y) The elementwise power of y and x\nAvailable since 2.24",
+    "text": "Any bytes on a line after a line-comment sequence (// or #) are ignored up until the ASCII newline character (\\n). They may thus be written in any character encoding which is convenient.\nAny content after a block comment open sequence in ASCII (/*) up to the closing block comment (*/) is ignored, and thus may also be written in whatever character set is convenient.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#dot-products-and-specialized-products",
-    "href": "functions-reference/matrix_operations.html#dot-products-and-specialized-products",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/encoding.html#string-literals",
+    "href": "reference-manual/encoding.html#string-literals",
+    "title": "Character Encoding",
     "section": "",
-    "text": "real dot_product(vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(vector x, vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(row_vector x, row_vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(matrix x, matrix y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(vector x, vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(row_vector x, row_vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(matrix x, matrix y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nreal dot_self(vector x) The dot product of the vector x with itself\nAvailable since 2.0\n \n\nreal dot_self(row_vector x) The dot product of the row vector x with itself\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(matrix x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(matrix x) The dot product of the rows of x with themselves\nAvailable since 2.0\n\n\n \n\nmatrix tcrossprod(matrix x) The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}\\,\\text{x}^{\\top}\\).\nAvailable since 2.0\n \n\nmatrix crossprod(matrix x) The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}^{\\top}\\,\\text{x}\\).\nAvailable since 2.0\nThe following functions all provide shorthand forms for common expressions, which are also much more efficient.\n \n\nmatrix quad_form(matrix A, matrix B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nreal quad_form(matrix A, vector B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nmatrix quad_form_diag(matrix m, vector v) The quadratic form using the column vector v as a diagonal matrix, i.e., diag_matrix(v) * m * diag_matrix(v).\nAvailable since 2.3\n \n\nmatrix quad_form_diag(matrix m, row_vector rv) The quadratic form using the row vector rv as a diagonal matrix, i.e., diag_matrix(rv) * m * diag_matrix(rv).\nAvailable since 2.3\n \n\nmatrix quad_form_sym(matrix A, matrix B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal quad_form_sym(matrix A, vector B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal trace_quad_form(matrix A, matrix B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_quad_form(matrix A, vector B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_gen_quad_form(matrix D, matrix A, matrix B) The trace of a generalized quadratic form, i.e., trace(D * B' * A * B).\nAvailable since 2.0\n \n\nmatrix multiply_lower_tri_self_transpose(matrix x) The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if L is a matrix of the same dimensions as x with L(m,n) equal to x(m,n) for \\(\\text{n}\n\\leq \\text{m}\\) and L(m,n) equal to 0 if \\(\\text{n} &gt; \\text{m}\\), the result is the symmetric matrix \\(\\text{L}\\,\\text{L}^{\\top}\\). This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(vector v, matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(row_vector rv, matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, row_vector rv) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.0",
+    "text": "The raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A print, reject, or fatal_error statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (\") when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.\nASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Character Encoding"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#reductions",
-    "href": "functions-reference/matrix_operations.html#reductions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/deprecations.html",
+    "href": "reference-manual/deprecations.html",
+    "title": "Deprecated Features",
     "section": "",
-    "text": "real log_sum_exp(vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(row_vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(matrix x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n\n\n\n \n\nreal min(vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(row_vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(matrix x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(row_vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(matrix x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n\n\n\n \n\nreal sum(vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal prod(vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n\n\n\nFull definitions are provided for sample moments in section array reductions.\n \n\nreal mean(vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(row_vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(matrix x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(row_vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(matrix x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(row_vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(matrix x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n \n\nreal quantile(data row_vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data row_vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27",
+    "text": "This appendix lists currently deprecated functionality along with how to replace it.\nStarting with Stan 2.29, minor (syntax-level) deprecations can be removed 3 versions after release; e.g., syntax deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in Removed Features. The Stan compiler can automatically update many of these on the behalf of the user for at least one version after they are removed.\nAny feature which changes semantic meaning (such as the upgraded ODE solver interface) will not be removed until a major version change (e.g., Stan 3.0).\n\n\nDeprecated:The distribution lkj_cov is deprecated.\nReplacement: Replace lkj_cov_lpdf(...) with an lkj_corr distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace\ncov_matrix[K] Sigma;\n// ...\nSigma ~ lkj_cov(mu, tau, eta);\nwith\ncorr_matrix[K] Omega;\nvector&lt;lower=0&gt;[K] sigma;\n// ...\nOmega ~ lkj_corr(eta);\nsigma ~ lognormal(mu, tau);\n// ...\ncov_matrix[K] Sigma;\nSigma &lt;- quad_form_diag(Omega, sigma);\nThe variable Sigma may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.\nScheduled Removal: Stan 3.0 or later.\n\n\n\nDeprecated: Using functions that end in _lp in the transformed parameters block.\nReplacement: Use _jacobian functions and the jacobian += statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.\n\n\n\nSeveral built-in Stan functions have been deprecated. Consult the functions reference for more information.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#matrix-broadcast",
-    "href": "functions-reference/matrix_operations.html#matrix-broadcast",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/deprecations.html#lkj_cov-distribution",
+    "href": "reference-manual/deprecations.html#lkj_cov-distribution",
+    "title": "Deprecated Features",
     "section": "",
-    "text": "The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\nvector rep_vector(real x, int m) Return the size m (column) vector consisting of copies of x.\nAvailable since 2.0\n \n\nrow_vector rep_row_vector(real x, int n) Return the size n row vector consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(real x, int m, int n) Return the m by n matrix consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.0\n \n\nmatrix rep_matrix(row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.0\nUnlike the situation with array broadcasting (see section array broadcasting), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.\n vector[3] x;\n x = rep_vector(1, 3);\n x = rep_vector(1.0, 3);\nThere are no integer vector or matrix types, so integer values are automatically promoted.\n\n\n \n\nmatrix symmetrize_from_lower_tri(matrix A)\nConstruct a symmetric matrix from the lower triangle of A.\nAvailable since 2.26",
+    "text": "Deprecated:The distribution lkj_cov is deprecated.\nReplacement: Replace lkj_cov_lpdf(...) with an lkj_corr distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace\ncov_matrix[K] Sigma;\n// ...\nSigma ~ lkj_cov(mu, tau, eta);\nwith\ncorr_matrix[K] Omega;\nvector&lt;lower=0&gt;[K] sigma;\n// ...\nOmega ~ lkj_corr(eta);\nsigma ~ lognormal(mu, tau);\n// ...\ncov_matrix[K] Sigma;\nSigma &lt;- quad_form_diag(Omega, sigma);\nThe variable Sigma may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.\nScheduled Removal: Stan 3.0 or later.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#diagonal-matrix-functions",
-    "href": "functions-reference/matrix_operations.html#diagonal-matrix-functions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/deprecations.html#use-of-_lp-functions-in-transformed-parameters",
+    "href": "reference-manual/deprecations.html#use-of-_lp-functions-in-transformed-parameters",
+    "title": "Deprecated Features",
     "section": "",
-    "text": "matrix add_diag(matrix m, row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, real d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.21\n \n\nvector diagonal(matrix x) The diagonal of the matrix x\nAvailable since 2.0\n \n\nmatrix diag_matrix(vector x) The diagonal matrix with diagonal x\nAvailable since 2.0\nAlthough the diag_matrix function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,\n y ~ multi_normal(mu, diag_matrix(square(sigma)));\nit is much more efficient to just use a univariate normal, which produces the same density,\n y ~ normal(mu, sigma);\nRather than writing m * diag_matrix(v) where m is a matrix and v is a vector, it is much more efficient to write diag_post_multiply(m, v) (and similarly for pre-multiplication). By the same token, it is better to use quad_form_diag(m, v) rather than quad_form(m, diag_matrix(v)).\n \n\nmatrix identity_matrix(int k) Create an identity matrix of size \\(k \\times k\\)\nAvailable since 2.26",
+    "text": "Deprecated: Using functions that end in _lp in the transformed parameters block.\nReplacement: Use _jacobian functions and the jacobian += statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#container-construction",
-    "href": "functions-reference/matrix_operations.html#container-construction",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/deprecations.html#deprecated-functions",
+    "href": "reference-manual/deprecations.html#deprecated-functions",
+    "title": "Deprecated Features",
     "section": "",
-    "text": "array[] real linspaced_array(int n, data real lower, data real upper) Create a real array of length n of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int linspaced_int_array(int n, int lower, int upper) Create a regularly spaced, increasing integer array of length n between lower and upper, inclusively. If (upper - lower) / (n - 1) is less than one, repeat each output (n - 1) / (upper - lower) times. If neither (upper - lower) / (n - 1) or (n - 1) / (upper - lower) are integers, upper is reduced until one of these is true.\nAvailable since 2.26\n \n\nvector linspaced_vector(int n, data real lower, data real upper) Create an n-dimensional vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\nrow_vector linspaced_row_vector(int n, data real lower, data real upper) Create an n-dimensional row-vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int one_hot_int_array(int n, int k) Create a one-hot encoded int array of length n with array[k] = 1\nAvailable since 2.26\n \n\narray[] real one_hot_array(int n, int k) Create a one-hot encoded real array of length n with array[k] = 1\nAvailable since 2.24\n \n\nvector one_hot_vector(int n, int k) Create an n-dimensional one-hot encoded vector with vector[k] = 1\nAvailable since 2.24\n \n\nrow_vector one_hot_row_vector(int n, int k) Create an n-dimensional one-hot encoded row-vector with row_vector[k] = 1\nAvailable since 2.24\n \n\narray[] int ones_int_array(int n) Create an int array of length n of all ones\nAvailable since 2.26\n \n\narray[] real ones_array(int n) Create a real array of length n of all ones\nAvailable since 2.26\n \n\nvector ones_vector(int n) Create an n-dimensional vector of all ones\nAvailable since 2.26\n \n\nrow_vector ones_row_vector(int n) Create an n-dimensional row-vector of all ones\nAvailable since 2.26\n \n\narray[] int zeros_int_array(int n) Create an int array of length n of all zeros\nAvailable since 2.26\n \n\narray[] real zeros_array(int n) Create a real array of length n of all zeros\nAvailable since 2.24\n \n\nvector zeros_vector(int n) Create an n-dimensional vector of all zeros\nAvailable since 2.24\n \n\nrow_vector zeros_row_vector(int n) Create an n-dimensional row-vector of all zeros\nAvailable since 2.24\n \n\nvector uniform_simplex(int n) Create an n-dimensional simplex with elements vector[i] = 1 / n for all \\(i \\in 1, \\dots, n\\)\nAvailable since 2.24",
+    "text": "Several built-in Stan functions have been deprecated. Consult the functions reference for more information.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Deprecated Features"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#slicing-and-blocking-functions",
-    "href": "functions-reference/matrix_operations.html#slicing-and-blocking-functions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html",
+    "href": "reference-manual/blocks.html",
+    "title": "Program Blocks",
     "section": "",
-    "text": "Stan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\nvector col(matrix x, int n) The n-th column of matrix x\nAvailable since 2.0\n \n\nrow_vector row(matrix x, int m) The m-th row of matrix x\nAvailable since 2.0\nThe row function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation x[m] is just an alternative way of writing row(x,m). The col function may not, be used as an lvalue, nor is there an indexing based shorthand for it.\n\n\n\n\n\nBlock operations may be used to extract a sub-block of a matrix.\n \n\nmatrix block(matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.0\nThe sub-row and sub-column operations may be used to extract a slice of row or column from a matrix\n \n\nvector sub_col(matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.0\n \n\nrow_vector sub_row(matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.0\n\n\n\nThe head operation extracts the first \\(n\\) elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.\n \n\nvector head(vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.0\n \n\nrow_vector head(row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.0\n \n\narray[] T head(array[] T sv, int n) Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector tail(vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.0\n \n\nrow_vector tail(row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.0\n \n\narray[] T tail(array[] T sv, int n) Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector segment(vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.0\n \n\nrow_vector segment(row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.10\n \n\narray[] T segment(array[] T sv, int i, int n) Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0",
+    "text": "A Stan program is organized into a sequence of named blocks, the bodies of which consist of variable declarations, followed in the case of some blocks with statements.\n\n\nThe full set of named program blocks is exemplified in the following skeletal Stan program.\nfunctions {\n  // ... function declarations and definitions ...\n}\ndata {\n  // ... declarations ...\n}\ntransformed data {\n   // ... declarations ... statements ...\n}\nparameters {\n   // ... declarations ...\n}\ntransformed parameters {\n   // ... declarations ... statements ...\n}\nmodel {\n   // ... declarations ... statements ...\n}\ngenerated quantities {\n   // ... declarations ... statements ...\n}\nThe function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.\n\n\nAll of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.\n\n\n\nThe variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.\nVariables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).\n\n\n\nFunctions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).\nRandom-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in _rng; such functions are suffixed with _rng. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with _lp.\nDensity functions defined in the program may be used in distribution statements.\n\n\n\nThe variables declared in the data and parameters block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.\nThe variables in the data block are read from an external input source such as a file or a designated R data structure. The variables in the parameters block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.\n\n\n\nThe transformed data and transformed parameters block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.\nBefore generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.1\nTransformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.\n\n\n\nThe generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.\n\n\n\nA table summarizing the point at which variables are read, written, and defined is given in the block actions table.\nBlock Actions Table.  The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.\n\n\n\nblock\nstatement\naction / period\n\n\n\n\ndata\nno\nread / chain\n\n\ntransformed data\nyes\nevaluate / chain\n\n\nparameters\nno\ninv. transform, Jacobian / leapfrog\n\n\n \n \ninv. transform, write / sample\n\n\ntransformed parameters\nyes\nevaluate / leapfrog\n\n\n \n \nwrite / sample\n\n\nmodel\nyes\nevaluate / leapfrog step\n\n\ngenerated quantities\nyes\neval / sample\n\n\n \n \nwrite / sample\n\n\n(initialization)\nn/a\nread, transform / chain\n\n\n\nVariable Declaration Table.  This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (\\(*\\)) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.\n\n\n\nparam depend\nin target\nsave\ndeclare in\n\n\n\n\n+\n+\n+\ntransformed parameters\n\n\n+\n+\n-\nmodel (local)\n\n\n+\n-\n+\ngenerated quantities\n\n\n+\n-\n-\ngenerated quantities (local)\n\n\n-\n+\n+\ntransformed data   and generated quantities\n\n\n-\n+\n-\ntransformed data\n\n\n-\n-\n+\ngenerated quantities\n\n\n-\n-\n-\ntransformed data (local)\n\n\n\nAnother way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the variable declaration table. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.2\nThe rest of this chapter provides full details on when and how the variables and statements in each block are executed.\n\n\n\n\nStatistical Variable Taxonomy Table.  Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.\n\n\n\n\n\n\n\nvariable kind\ndeclaration block\n\n\n\n\nconstants\ndata, transformed data\n\n\nunmodeled data\ndata, transformed data\n\n\nmodeled data\ndata, transformed data\n\n\nmissing data\nparameters, transformed parameters\n\n\nmodeled parameters\nparameters, transformed parameters\n\n\nunmodeled parameters\ndata, transformed data\n\n\nderived quantities\ntransformed data, transformed parameters, generated quantities\n\n\nloop indices\nloop statement\n\n\n\nPage 366 of (Gelman and Hill 2007) provides a taxonomy of the kinds of variables used in Bayesian models. The table of kinds of variables contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.\nConstants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the data block.\nThe following program illustrates various variables kinds, listing the kind of each variable next to its declaration.\ndata {\n  int&lt;lower=0&gt; N;           // unmodeled data\n  array[N] real y;          // modeled data\n  real mu_mu;               // config. unmodeled param\n  real&lt;lower=0&gt; sigma_mu;   // config. unmodeled param\n}\ntransformed data {\n  real&lt;lower=0&gt; alpha;      // const. unmodeled param\n  real&lt;lower=0&gt; beta;       // const. unmodeled param\n  alpha = 0.1;\n  beta = 0.1;\n}\nparameters {\n  real mu_y;                // modeled param\n  real&lt;lower=0&gt; tau_y;      // modeled param\n}\ntransformed parameters {\n  real&lt;lower=0&gt; sigma_y;    // derived quantity (param)\n  sigma_y = pow(tau_y, -0.5);\n}\nmodel {\n  tau_y ~ gamma(alpha, beta);\n  mu_y ~ normal(mu_mu, sigma_mu);\n  for (n in 1:N) {\n    y[n] ~ normal(mu_y, sigma_y);\n  }\n}\ngenerated quantities {\n  real variance_y;       // derived quantity (transform)\n  variance_y = sigma_y * sigma_y;\n}\nIn this example, y is an array of modeled data. Although it is specified in the data block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.\nThe variable N is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.\nThe other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the data block may be specified on a per-model execution basis as part of the data read. In the above model, mu_mu and sigma_mu are configurable unmodeled parameters.\nUnmodeled parameters that are hard coded in the model must be declared in the transformed data block. For example, the unmodeled parameters alpha and beta are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the data block, like the variables mu_mu and sigma_mu.\nThis program declares two modeled parameters, mu_y and tau_y. These are the location and precision used in the normal model of the values in y. The heart of the model will be sampling the values of these parameters from their posterior distribution.\nThe modeled parameter tau_y is transformed from a precision to a scale parameter and assigned to the variable sigma_y in the transformed parameters block. Thus the variable sigma_y is considered a derived quantity — its value is entirely determined by the values of other variables.\nThe generated quantities block defines a value variance_y, which is defined as a transform of the scale or deviation parameter sigma_y. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).\nIn later versions of Stan which have random number generators for the distributions, the generated quantities block will be usable to generate replicated data for model checking.\nFinally, the variable n is used as a loop index in the model block.\n\n\n\nThe rest of this chapter will lay out the details of each block in order, starting with the data block in this section.\n\n\nThe data block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.3\nData variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.\n\n\n\nThe data block does not allow statements.\n\n\n\nEach variable’s value is validated against its declaration as it is read. For example, if a variable sigma is declared as real&lt;lower=0&gt;, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.\n\n\n\n\nThe transformed data block is for declaring and defining variables that do not need to be changed when running the program.\n\n\nFor the transformed data block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.\nVariables declared in the data block may be used to declare transformed variables.\n\n\n\nThe statements in a transformed data block are used to define (provide values for) variables declared in the transformed data block. Assignments are only allowed to variables declared in the transformed data block.\nThese statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.\nVariables declared in the data block may be used in statements in the transformed data block.\n\n\nThe statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and distribution statements may not be used.\n\n\n\n\nAny constraints on variables declared in the transformed data block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.\n\n\n\n\nThe variables declared in the parameters program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block are the parameters being sampled by Stan.\nVariables declared as parameters cannot be directly assigned values. So there is no block of statements in the parameters program block. Variable quantities derived from parameters may be declared in the transformed parameters or generated quantities blocks, or may be defined as local variables in any statement blocks following their declaration.\nThere is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.\n\n\nStan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of \\(\\mathbb{R}^n\\). To do this, the parameters defined in the parameters block must be transformed so they are unconstrained.\nIn practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s parameters program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.\nIn some cases, the number of parameters is reduced in the unconstrained space. For instance, a \\(K\\)-simplex only requires \\(K-1\\) unconstrained parameters, and a \\(K\\)-correlation matrix only requires \\(\\binom{K}{2}\\) unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the parameters program block.\nThe probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the parameters program block. The details of the specific transformations are provided in the variable transforms chapter.\n\n\n\nHamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.\nGradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.\nThe amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.\n\n\n\nIn the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the parameters block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.\n\n\n\n\nThe transformed parameters program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.\nAny variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.\n\n\nLike the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the section on reject statements for further discussion of the behavior of reject statements.\n\n\n\n\nThe model program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.\nLocal variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.\nThe statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.\n\n\n\nThe generated quantities program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.\nAmong the applications of posterior inference that can be coded in the generated quantities block are\n\nforward sampling to generate simulated data for model testing,\ngenerating predictions for new data,\ncalculating posterior event probabilities, including multiple comparisons, sign tests, etc.,\ncalculating posterior expectations,\ntransforming parameters for reporting,\napplying full Bayesian decision theory,\ncalculating log likelihoods, deviances, etc. for model comparison.\n\nParameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.\nWithin the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.\nIt is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.\nAfter the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.\nAll variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:\ngenerated quantities {\n  int a; // added to the output\n\n  {\n    int b; // not added to the output\n  }\n}",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#matrix-concatenation",
-    "href": "functions-reference/matrix_operations.html#matrix-concatenation",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html#overview-of-stans-program-blocks",
+    "href": "reference-manual/blocks.html#overview-of-stans-program-blocks",
+    "title": "Program Blocks",
     "section": "",
-    "text": "Stan’s matrix and vector concatenation operations append_col and append_row are like the operations cbind and rbind in R.\n\n\n \n\nmatrix append_col(matrix x, matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(matrix x, vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.5\n \n\nrow_vector append_col(row_vector x, row_vector y) Combine row vectors x and y of any size into another row vector by appending y to the end of x.\nAvailable since 2.5\n \n\nrow_vector append_col(real x, row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.12\n \n\nrow_vector append_col(row_vector x, real y) Append y to the end of x, returning another row vector.\nAvailable since 2.12\n\n\n\n \n\nmatrix append_row(matrix x, matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(matrix x, row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.5\n \n\nvector append_row(vector x, vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.5\n \n\nvector append_row(real x, vector y) Append x to the top of y, returning another vector.\nAvailable since 2.12\n \n\nvector append_row(vector x, real y) Append y to the bottom of x, returning another vector.\nAvailable since 2.12",
+    "text": "The full set of named program blocks is exemplified in the following skeletal Stan program.\nfunctions {\n  // ... function declarations and definitions ...\n}\ndata {\n  // ... declarations ...\n}\ntransformed data {\n   // ... declarations ... statements ...\n}\nparameters {\n   // ... declarations ...\n}\ntransformed parameters {\n   // ... declarations ... statements ...\n}\nmodel {\n   // ... declarations ... statements ...\n}\ngenerated quantities {\n   // ... declarations ... statements ...\n}\nThe function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.\n\n\nAll of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.\n\n\n\nThe variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.\nVariables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).\n\n\n\nFunctions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).\nRandom-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in _rng; such functions are suffixed with _rng. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with _lp.\nDensity functions defined in the program may be used in distribution statements.\n\n\n\nThe variables declared in the data and parameters block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.\nThe variables in the data block are read from an external input source such as a file or a designated R data structure. The variables in the parameters block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.\n\n\n\nThe transformed data and transformed parameters block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.\nBefore generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.1\nTransformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.\n\n\n\nThe generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.\n\n\n\nA table summarizing the point at which variables are read, written, and defined is given in the block actions table.\nBlock Actions Table.  The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.\n\n\n\nblock\nstatement\naction / period\n\n\n\n\ndata\nno\nread / chain\n\n\ntransformed data\nyes\nevaluate / chain\n\n\nparameters\nno\ninv. transform, Jacobian / leapfrog\n\n\n \n \ninv. transform, write / sample\n\n\ntransformed parameters\nyes\nevaluate / leapfrog\n\n\n \n \nwrite / sample\n\n\nmodel\nyes\nevaluate / leapfrog step\n\n\ngenerated quantities\nyes\neval / sample\n\n\n \n \nwrite / sample\n\n\n(initialization)\nn/a\nread, transform / chain\n\n\n\nVariable Declaration Table.  This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (\\(*\\)) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.\n\n\n\nparam depend\nin target\nsave\ndeclare in\n\n\n\n\n+\n+\n+\ntransformed parameters\n\n\n+\n+\n-\nmodel (local)\n\n\n+\n-\n+\ngenerated quantities\n\n\n+\n-\n-\ngenerated quantities (local)\n\n\n-\n+\n+\ntransformed data   and generated quantities\n\n\n-\n+\n-\ntransformed data\n\n\n-\n-\n+\ngenerated quantities\n\n\n-\n-\n-\ntransformed data (local)\n\n\n\nAnother way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the variable declaration table. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.2\nThe rest of this chapter provides full details on when and how the variables and statements in each block are executed.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#softmax",
-    "href": "functions-reference/matrix_operations.html#softmax",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html#statistical-variable-taxonomy",
+    "href": "reference-manual/blocks.html#statistical-variable-taxonomy",
+    "title": "Program Blocks",
     "section": "",
-    "text": "The softmax function maps1 \\(y \\in \\mathbb{R}^K\\) to the \\(K\\)-simplex by \\[\\begin{equation*} \\text{softmax}(y)  = \\frac{\\exp(y)}\n{\\sum_{k=1}^K \\exp(y_k)}, \\end{equation*}\\] where \\(\\exp(y)\\) is the componentwise exponentiation of \\(y\\). Softmax is usually calculated on the log scale, \\[\\begin{eqnarray*} \\log \\text{softmax}(y) & = & \\ y - \\log \\sum_{k=1}^K\n\\exp(y_k) \\\\[4pt] & = & y - \\mathrm{log\\_sum\\_exp}(y). \\end{eqnarray*}\\] where the vector \\(y\\) minus the scalar \\(\\mathrm{log\\_sum\\_exp}(y)\\) subtracts the scalar from each component of \\(y\\).\nStan provides the following functions for softmax and its log.\n \n\nvector softmax(vector x) The softmax of x\nAvailable since 2.0\n \n\nvector log_softmax(vector x) The natural logarithm of the softmax of x\nAvailable since 2.0\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] int cumulative_sum(array[] int x) The cumulative sum of x\nAvailable since 2.30\n \n\narray[] real cumulative_sum(array[] real x) The cumulative sum of x\nAvailable since 2.0\n \n\nvector cumulative_sum(vector v) The cumulative sum of v\nAvailable since 2.0\n \n\nrow_vector cumulative_sum(row_vector rv) The cumulative sum of rv\nAvailable since 2.0",
+    "text": "Statistical Variable Taxonomy Table.  Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.\n\n\n\n\n\n\n\nvariable kind\ndeclaration block\n\n\n\n\nconstants\ndata, transformed data\n\n\nunmodeled data\ndata, transformed data\n\n\nmodeled data\ndata, transformed data\n\n\nmissing data\nparameters, transformed parameters\n\n\nmodeled parameters\nparameters, transformed parameters\n\n\nunmodeled parameters\ndata, transformed data\n\n\nderived quantities\ntransformed data, transformed parameters, generated quantities\n\n\nloop indices\nloop statement\n\n\n\nPage 366 of (Gelman and Hill 2007) provides a taxonomy of the kinds of variables used in Bayesian models. The table of kinds of variables contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.\nConstants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the data block.\nThe following program illustrates various variables kinds, listing the kind of each variable next to its declaration.\ndata {\n  int&lt;lower=0&gt; N;           // unmodeled data\n  array[N] real y;          // modeled data\n  real mu_mu;               // config. unmodeled param\n  real&lt;lower=0&gt; sigma_mu;   // config. unmodeled param\n}\ntransformed data {\n  real&lt;lower=0&gt; alpha;      // const. unmodeled param\n  real&lt;lower=0&gt; beta;       // const. unmodeled param\n  alpha = 0.1;\n  beta = 0.1;\n}\nparameters {\n  real mu_y;                // modeled param\n  real&lt;lower=0&gt; tau_y;      // modeled param\n}\ntransformed parameters {\n  real&lt;lower=0&gt; sigma_y;    // derived quantity (param)\n  sigma_y = pow(tau_y, -0.5);\n}\nmodel {\n  tau_y ~ gamma(alpha, beta);\n  mu_y ~ normal(mu_mu, sigma_mu);\n  for (n in 1:N) {\n    y[n] ~ normal(mu_y, sigma_y);\n  }\n}\ngenerated quantities {\n  real variance_y;       // derived quantity (transform)\n  variance_y = sigma_y * sigma_y;\n}\nIn this example, y is an array of modeled data. Although it is specified in the data block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.\nThe variable N is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.\nThe other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the data block may be specified on a per-model execution basis as part of the data read. In the above model, mu_mu and sigma_mu are configurable unmodeled parameters.\nUnmodeled parameters that are hard coded in the model must be declared in the transformed data block. For example, the unmodeled parameters alpha and beta are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the data block, like the variables mu_mu and sigma_mu.\nThis program declares two modeled parameters, mu_y and tau_y. These are the location and precision used in the normal model of the values in y. The heart of the model will be sampling the values of these parameters from their posterior distribution.\nThe modeled parameter tau_y is transformed from a precision to a scale parameter and assigned to the variable sigma_y in the transformed parameters block. Thus the variable sigma_y is considered a derived quantity — its value is entirely determined by the values of other variables.\nThe generated quantities block defines a value variance_y, which is defined as a transform of the scale or deviation parameter sigma_y. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).\nIn later versions of Stan which have random number generators for the distributions, the generated quantities block will be usable to generate replicated data for model checking.\nFinally, the variable n is used as a loop index in the model block.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#gaussian-process-covariance-functions",
-    "href": "functions-reference/matrix_operations.html#gaussian-process-covariance-functions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html#program-block-data",
+    "href": "reference-manual/blocks.html#program-block-data",
+    "title": "Program Blocks",
     "section": "",
-    "text": "The Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.\nFor one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by \\(K_{ij} = k(x_i, x_j)\\) (where \\(x_i\\) is the \\(i^{th}\\) element of the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(x_i, y_j)\\).\nFor multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{x}_j)\\) (where \\(\\mathbf{x}_i\\) is the \\(i^{th}\\) vector in the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{y}_j)\\).\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|^2}{2l^2} \\right)\n\\]\n \n\nmatrix gp_exp_quad_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith bias \\(\\sigma_0\\) the dot product kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma_0^2 + \\mathbf{x}_i^T \\mathbf{x}_j\n\\]\n \n\nmatrix gp_dot_prod_cov(array[] real x, real sigma)\nGaussian process covariance with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(array[] real x1, array[] real x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x, real sigma)\nGaussian process covariance with dot product kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x1, vectors x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in multiple dimensions.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponential kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_exponential_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 3/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right) \\exp \\left( -\\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern32_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 5/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{5}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} + \\frac{5 |\\mathbf{x}_i - \\mathbf{x}_j|^2}{3l^2} \\right)\n\\exp \\left( -\\frac{\\sqrt{5} |\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern52_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\), length scale \\(l\\), and period \\(p\\), the periodic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left(-\\frac{2 \\sin^2 \\left( \\pi \\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{p} \\right) }{l^2} \\right)\n\\]\n \n\nmatrix gp_periodic_cov(array[] real x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(array[] real x1, array[] real x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x1, vectors x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20",
+    "text": "The rest of this chapter will lay out the details of each block in order, starting with the data block in this section.\n\n\nThe data block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.3\nData variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.\n\n\n\nThe data block does not allow statements.\n\n\n\nEach variable’s value is validated against its declaration as it is read. For example, if a variable sigma is declared as real&lt;lower=0&gt;, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#linear-algebra-functions-and-solvers",
-    "href": "functions-reference/matrix_operations.html#linear-algebra-functions-and-solvers",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html#program-block-transformed-data",
+    "href": "reference-manual/blocks.html#program-block-transformed-data",
+    "title": "Program Blocks",
     "section": "",
-    "text": "In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.\n\n\n \n\nrow_vector operator/(row_vector b, matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.0\n \n\nmatrix operator/(matrix B, matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.5\n \n\nvector operator\\(matrix A, vector b) The left division of A by b; equivalently inverse(A) * b\nAvailable since 2.18\n \n\nmatrix operator\\(matrix A, matrix B) The left division of A by B; equivalently inverse(A) * B\nAvailable since 2.18\n\n\n\nThere are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix \\(\\text{tri}(A)\\) is used in the definitions and defined by \\[\\begin{equation*} \\text{tri}(A)[m,n] = \\left\\{\n\\begin{array}{ll} A[m,n] & \\text{if } m \\geq n, \\text{ and} \\\\[4pt] 0\n& \\text{otherwise}. \\end{array} \\right. \\end{equation*}\\] When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.\n \n\nvector mdivide_left_tri_low(matrix A, vector b) The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * b, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_left_tri_low(matrix A, matrix B) The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * B, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n \n\nrow_vector mdivide_right_tri_low(row_vector b, matrix A) The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form b * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_right_tri_low(matrix B, matrix A) The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form B * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n\n\n\n\nThere are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.\n \n\nmatrix mdivide_left_spd(matrix A, vector b) The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * b.\nAvailable since 2.12\n \n\nvector mdivide_left_spd(matrix A, matrix B) The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * B.\nAvailable since 2.12\n \n\nrow_vector mdivide_right_spd(row_vector b, matrix A) The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form b *inverse(A).\nAvailable since 2.12\n \n\nmatrix mdivide_right_spd(matrix B, matrix A) The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form B * inverse(A).\nAvailable since 2.12\n\n\n\nThe exponential of the matrix \\(A\\) is formally defined by the convergent power series: \\[\\begin{equation*} e^A = \\sum_{n=0}^{\\infty} \\dfrac{A^n}{n!}\n\\end{equation*}\\]\n \n\nmatrix matrix_exp(matrix A) The matrix exponential of A\nAvailable since 2.13\n \n\nmatrix matrix_exp_multiply(matrix A, matrix B) The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form matrix_exp(A) * B.\nAvailable since 2.18\n \n\nmatrix scale_matrix_exp_multiply(real t, matrix A, matrix B) The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form matrix_exp(t * A) * B.\nAvailable since 2.18\n\n\n\nReturns the nth power of the specific matrix: \\[\\begin{equation*} M^n = M_1 * ... * M_n \\end{equation*}\\]\n \n\nmatrix matrix_power(matrix A, int B) Matrix A raised to the power B.\nAvailable since 2.24\n\n\n\n\n\n \n\nreal trace(matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.0\n\n\n\n \n\nreal determinant(matrix A) The determinant of A\nAvailable since 2.0\n \n\nreal log_determinant(matrix A) The log of the absolute value of the determinant of A\nAvailable since 2.0\nreal log_determinant_spd(matrix A) The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.\nAvailable since 2.30\n\n\n\nIt is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix m and post-multiplying by a vector or matrix a, as in inverse(m) * a, it is better to code this using matrix division, as in m \\ a. The pre-multiplication case is similar, with b * inverse(m) being more efficiently coded as as b / m. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.\nWarning: The function inv(m) is the elementwise inverse function, which returns 1 / m[i, j] for each element.\n \n\nmatrix inverse(matrix A) Compute the inverse of A\nAvailable since 2.0\n \n\nmatrix inverse_spd(matrix A) Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.\nAvailable since 2.0\n \n\nmatrix chol2inv(matrix L) Compute the inverse of the matrix whose cholesky factorization is L. That is, for \\(A = L L^T\\), return \\(A^{-1}\\).\nAvailable since 2.26\n\n\n\nThe generalized inverse \\(M^+\\) of a matrix \\(M\\) is a matrix that satisfies \\(M M^+ M = M\\). For an invertible, square matrix \\(M\\), \\(M^+\\) is equivalent to \\(M^{-1}\\). The dimensions of \\(M^+\\) are equivalent to the dimensions of \\(M^T\\). The generalized inverse exists for any matrix, so the \\(M\\) may be singular or less than full rank.\nEven though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank (Golub and Pereyra 1973), meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix \\(A\\) as a function of \\(\\epsilon\\):\n\\[\nA = \\left(\n    \\begin{array}{cccc}\n    1 + \\epsilon & 2 & 1 \\\\\n    2 & 4 & 2\n    \\end{array}\n    \\right)\n\\]\nWhen \\(\\epsilon = 0\\), \\(A\\) is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If \\(\\epsilon \\neq 0\\), the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at \\(\\epsilon = 0\\), and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.\n \n\nmatrix generalized_inverse(matrix A) The generalized inverse of A\nAvailable since 2.26\n\n\n\n \n\ncomplex_vector eigenvalues(matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors(matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nvector eigenvalues_sym(matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.0\n \n\nmatrix eigenvectors_sym(matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.0\n \n\ntuple(matrix, vector) eigendecompose_sym(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\n \n\nmatrix qr_thin_Q(matrix A) The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A\nAvailable since 2.18\n \n\nmatrix qr_thin_R(matrix A) The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A\nAvailable since 2.18\n \n\ntuple(matrix, matrix) qr_thin(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_thin_Q(A), qr_thin_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nmatrix qr_Q(matrix A) The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A\nAvailable since 2.3\n \n\nmatrix qr_R(matrix A) The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A\nAvailable since 2.3\n \n tuple(matrix, matrix) qr(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_Q(A), qr_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nThe thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as \\[\\begin{equation*} A = Q \\, R. \\end{equation*}\\] Multiplying a column of an orthogonal matrix by \\(-1\\) still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by \\(-1\\) without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.\n\n\n\nEvery symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If \\(\\Sigma\\) is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector \\(L\\) such that \\[\\begin{equation*} \\Sigma = L \\, L^{\\top}. \\end{equation*}\\]\n \n\nmatrix cholesky_decompose(matrix A) The lower-triangular Cholesky factor of the symmetric positive-definite matrix A\nAvailable since 2.0\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(matrix A) The singular values of A in descending order\nAvailable since 2.0\n \n\nmatrix svd_U(matrix A) The left-singular vectors of A\nAvailable since 2.26\n \n\nmatrix svd_V(matrix A) The right-singular vectors of A\nAvailable since 2.26\n \n\ntuple(matrix, vector, matrix) svd(matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33",
+    "text": "The transformed data block is for declaring and defining variables that do not need to be changed when running the program.\n\n\nFor the transformed data block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.\nVariables declared in the data block may be used to declare transformed variables.\n\n\n\nThe statements in a transformed data block are used to define (provide values for) variables declared in the transformed data block. Assignments are only allowed to variables declared in the transformed data block.\nThese statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.\nVariables declared in the data block may be used in statements in the transformed data block.\n\n\nThe statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and distribution statements may not be used.\n\n\n\n\nAny constraints on variables declared in the transformed data block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#sort-functions",
-    "href": "functions-reference/matrix_operations.html#sort-functions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html#program-block-parameters",
+    "href": "reference-manual/blocks.html#program-block-parameters",
+    "title": "Program Blocks",
     "section": "",
-    "text": "See the sorting functions section for examples of how the functions work.\n \n\nvector sort_asc(vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nrow_vector sort_asc(row_vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nvector sort_desc(vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\nrow_vector sort_desc(row_vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(row_vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0",
+    "text": "The variables declared in the parameters program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block are the parameters being sampled by Stan.\nVariables declared as parameters cannot be directly assigned values. So there is no block of statements in the parameters program block. Variable quantities derived from parameters may be declared in the transformed parameters or generated quantities blocks, or may be defined as local variables in any statement blocks following their declaration.\nThere is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.\n\n\nStan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of \\(\\mathbb{R}^n\\). To do this, the parameters defined in the parameters block must be transformed so they are unconstrained.\nIn practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s parameters program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.\nIn some cases, the number of parameters is reduced in the unconstrained space. For instance, a \\(K\\)-simplex only requires \\(K-1\\) unconstrained parameters, and a \\(K\\)-correlation matrix only requires \\(\\binom{K}{2}\\) unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the parameters program block.\nThe probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the parameters program block. The details of the specific transformations are provided in the variable transforms chapter.\n\n\n\nHamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.\nGradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.\nThe amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.\n\n\n\nIn the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the parameters block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#reverse-functions",
-    "href": "functions-reference/matrix_operations.html#reverse-functions",
-    "title": "Matrix Operations",
+    "objectID": "reference-manual/blocks.html#program-block-transformed-parameters",
+    "href": "reference-manual/blocks.html#program-block-transformed-parameters",
+    "title": "Program Blocks",
     "section": "",
-    "text": "vector reverse(vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.23\n \n\nrow_vector reverse(row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "text": "The transformed parameters program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.\nAny variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.\n\n\nLike the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the section on reject statements for further discussion of the behavior of reject statements.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/matrix_operations.html#footnotes",
-    "href": "functions-reference/matrix_operations.html#footnotes",
-    "title": "Matrix Operations",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe softmax function is so called because in the limit as \\(y_n \\rightarrow \\infty\\) with \\(y_m\\) for \\(m \\neq n\\) held constant, the result tends toward the “one-hot” vector \\(\\theta\\) with \\(\\theta_n\n= 1\\) and \\(\\theta_m = 0\\) for \\(m \\neq n\\), thus providing a “soft” version of the maximum function.↩︎",
+    "objectID": "reference-manual/blocks.html#program-block-model",
+    "href": "reference-manual/blocks.html#program-block-model",
+    "title": "Program Blocks",
+    "section": "",
+    "text": "The model program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.\nLocal variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.\nThe statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Matrix Operations"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/integer-valued_basic_functions.html",
-    "href": "functions-reference/integer-valued_basic_functions.html",
-    "title": "Integer-Valued Basic Functions",
+    "objectID": "reference-manual/blocks.html#program-block-generated-quantities",
+    "href": "reference-manual/blocks.html#program-block-generated-quantities",
+    "title": "Program Blocks",
     "section": "",
-    "text": "This chapter describes Stan’s built-in function that take various types of arguments and return integer values.\n\n\nStan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in 2 + 2, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in 2.0 + 2 and 2 + 2.0, then the integer is promoted to a floating point value and floating-point arithmetic is used.\nInteger arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.\nSecondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, 1 / 2 evaluates to 0 and 5 / 3 evaluates to 1.\nIf one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have 1 / 2 and -1 / 2 evaluate to 0, -7 / 2 evaluate to -3, and 7 / 2 evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.\nUnlike floating point division, where 1.0 / 0.0 produces the special positive infinite value, integer division by zero, as in 1 / 0, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.\n\n\nOperators are described using the C++ syntax. For instance, the binary operator of addition, written X + Y, would have the Stan signature int operator+(int, int) indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.\n \n\nint operator+(int x, int y) The sum of the addends x and y \\[\\begin{equation*} \\text{operator+}(x,y) = (x + y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator-(int x, int y) The difference between the minuend x and subtrahend y \\[\\begin{equation*}\n\\text{operator-}(x,y) = (x - y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator*(int x, int y) The product of the factors x and y \\[\\begin{equation*} \\text{operator*}(x,y) = (x\n\\times y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator/(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator/}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\] deprecated; - use operator%/% instead.\nAvailable since 2.0, deprecated in 2.24\n \n\nint operator%/%(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator\\%/\\%}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\]\nAvailable since 2.24\n \n\nint operator%(int x, int y) x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. \\[\\begin{equation*} \\mathrm{operator\\%}(x, y) \\ =\n\\ x \\ \\text{mod} \\ y \\ = \\ x - y * \\lfloor x / y \\rfloor \\end{equation*}\\]\nAvailable since 2.13\n\n\n\n \n\nint operator-(int x) The negation of the subtrahend x \\[\\begin{equation*} \\text{operator-}(x) = -x \\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of integers, -x is the same shape array where each individual integer is negated.\nAvailable since 2.31\n \n\nint operator+(int x) This is a no-op. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is int, or an array of ints, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nint int_step(int x)\n \n\nint int_step(real x) Return the step function of x as an integer, \\[\\begin{equation*} \\mathrm{int\\_step}(x)\n= \\begin{cases} 1 & \\text{if } x &gt; 0 \\\\ 0 & \\text{if } x \\leq 0 \\text{\nor } x \\text{ is } NaN \\end{cases} \\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nSee the warning in section step functions about the dangers of step functions applied to anything other than data.\nAvailable since 2.0\n\n\n\n \n\nint min(int x, int y) Return the minimum of x and y. \\[\\begin{equation*} \\text{min}(x, y) = \\begin{cases} x &\n\\text{if } x &lt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint max(int x, int y) Return the maximum of x and y. \\[\\begin{equation*} \\text{max}(x, y) = \\begin{cases} x &\n\\text{if } x &gt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n \n\nint size(int x)\n \n\nint size(real x)\nReturn the size of x which for scalar-valued x is 1\nAvailable since 2.26\n\n\n\nIt is possible to cast real numbers to integers as long as the real value is data. See data only qualifiers in the Stan Reference Manual.\n \n\nint to_int(data real x)\nReturn the value x truncated to an integer. This will throw an error if the value of x is too big to represent as a 32-bit signed integer.\nThis is similar to trunc (see Rounding functions) but the return type is of type int. For example, to_int(3.9) is 3, and to_int(-3.9) is -3.\nAvailable since 2.31\n \n\nI to_int(data T x)\nThe vectorized version of to_int. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.\nAvailable since 2.31",
+    "text": "The generated quantities program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.\nAmong the applications of posterior inference that can be coded in the generated quantities block are\n\nforward sampling to generate simulated data for model testing,\ngenerating predictions for new data,\ncalculating posterior event probabilities, including multiple comparisons, sign tests, etc.,\ncalculating posterior expectations,\ntransforming parameters for reporting,\napplying full Bayesian decision theory,\ncalculating log likelihoods, deviances, etc. for model comparison.\n\nParameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.\nWithin the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.\nIt is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.\nAfter the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.\nAll variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:\ngenerated quantities {\n  int a; // added to the output\n\n  {\n    int b; // not added to the output\n  }\n}",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Integer-Valued Basic Functions"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/integer-valued_basic_functions.html#int-arithmetic",
-    "href": "functions-reference/integer-valued_basic_functions.html#int-arithmetic",
-    "title": "Integer-Valued Basic Functions",
-    "section": "",
-    "text": "Stan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in 2 + 2, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in 2.0 + 2 and 2 + 2.0, then the integer is promoted to a floating point value and floating-point arithmetic is used.\nInteger arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.\nSecondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, 1 / 2 evaluates to 0 and 5 / 3 evaluates to 1.\nIf one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have 1 / 2 and -1 / 2 evaluate to 0, -7 / 2 evaluate to -3, and 7 / 2 evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.\nUnlike floating point division, where 1.0 / 0.0 produces the special positive infinite value, integer division by zero, as in 1 / 0, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.\n\n\nOperators are described using the C++ syntax. For instance, the binary operator of addition, written X + Y, would have the Stan signature int operator+(int, int) indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.\n \n\nint operator+(int x, int y) The sum of the addends x and y \\[\\begin{equation*} \\text{operator+}(x,y) = (x + y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator-(int x, int y) The difference between the minuend x and subtrahend y \\[\\begin{equation*}\n\\text{operator-}(x,y) = (x - y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator*(int x, int y) The product of the factors x and y \\[\\begin{equation*} \\text{operator*}(x,y) = (x\n\\times y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator/(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator/}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\] deprecated; - use operator%/% instead.\nAvailable since 2.0, deprecated in 2.24\n \n\nint operator%/%(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator\\%/\\%}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\]\nAvailable since 2.24\n \n\nint operator%(int x, int y) x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. \\[\\begin{equation*} \\mathrm{operator\\%}(x, y) \\ =\n\\ x \\ \\text{mod} \\ y \\ = \\ x - y * \\lfloor x / y \\rfloor \\end{equation*}\\]\nAvailable since 2.13\n\n\n\n \n\nint operator-(int x) The negation of the subtrahend x \\[\\begin{equation*} \\text{operator-}(x) = -x \\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of integers, -x is the same shape array where each individual integer is negated.\nAvailable since 2.31\n \n\nint operator+(int x) This is a no-op. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0",
+    "objectID": "reference-manual/blocks.html#footnotes",
+    "href": "reference-manual/blocks.html#footnotes",
+    "title": "Program Blocks",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nIf the C++ code is configured for concurrent threads, the data and transformed data blocks can be executed once and reused for multiple chains.↩︎\nIt is possible to print a variable every iteration that does not depend on parameters—just define it (or redefine it if it is transformed data) in the generated quantities block.↩︎\nWith multiple threads, or even running chains sequentially in a single thread, data could be read only once per set of chains. Stan was designed to be thread safe and future versions will provide a multithreading option for Markov chains.↩︎",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Integer-Valued Basic Functions"
+      "Reference Manual",
+      "Language",
+      "Program Blocks"
     ]
   },
   {
-    "objectID": "functions-reference/integer-valued_basic_functions.html#absolute-functions",
-    "href": "functions-reference/integer-valued_basic_functions.html#absolute-functions",
-    "title": "Integer-Valued Basic Functions",
+    "objectID": "functions-reference/void_functions.html",
+    "href": "functions-reference/void_functions.html",
+    "title": "Void Functions",
     "section": "",
-    "text": "T abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is int, or an array of ints, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nint int_step(int x)\n \n\nint int_step(real x) Return the step function of x as an integer, \\[\\begin{equation*} \\mathrm{int\\_step}(x)\n= \\begin{cases} 1 & \\text{if } x &gt; 0 \\\\ 0 & \\text{if } x \\leq 0 \\text{\nor } x \\text{ is } NaN \\end{cases} \\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nSee the warning in section step functions about the dangers of step functions applied to anything other than data.\nAvailable since 2.0",
+    "text": "Stan supports a few special statements for printing or for signaling an issue with the program.\nAlthough print, reject, and fatal_error appear to have the syntax of functions, they are actually special kinds of statements with slightly different form and behavior than other functions. First, they are the constructs that allow a variable number of arguments. Second, they are the the only constructs to accept string literals (e.g., \"hello world\") as arguments. Third, they have no effect on the log density function and operate solely through side effects.\nThe special keyword void is used for their return type because they behave like variadic functions with void return type, even though they are special kinds of statements built in to the language.\n\n\nPrinting has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).\n \n\nvoid print(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.1\n\n\n\nThe reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).\n \n\nvoid reject(T1 x1,..., TN xN) Reject the current iteration and print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.18\n\n\n\nThe fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.\n \n\nvoid fatal_error(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.35",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Integer-Valued Basic Functions"
+      "Void Functions"
     ]
   },
   {
-    "objectID": "functions-reference/integer-valued_basic_functions.html#bound-functions",
-    "href": "functions-reference/integer-valued_basic_functions.html#bound-functions",
-    "title": "Integer-Valued Basic Functions",
+    "objectID": "functions-reference/void_functions.html#print-statement",
+    "href": "functions-reference/void_functions.html#print-statement",
+    "title": "Void Functions",
     "section": "",
-    "text": "int min(int x, int y) Return the minimum of x and y. \\[\\begin{equation*} \\text{min}(x, y) = \\begin{cases} x &\n\\text{if } x &lt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint max(int x, int y) Return the maximum of x and y. \\[\\begin{equation*} \\text{max}(x, y) = \\begin{cases} x &\n\\text{if } x &gt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0",
+    "text": "Printing has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).\n \n\nvoid print(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.1",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Integer-Valued Basic Functions"
+      "Void Functions"
     ]
   },
   {
-    "objectID": "functions-reference/integer-valued_basic_functions.html#size-functions",
-    "href": "functions-reference/integer-valued_basic_functions.html#size-functions",
-    "title": "Integer-Valued Basic Functions",
+    "objectID": "functions-reference/void_functions.html#reject-statement",
+    "href": "functions-reference/void_functions.html#reject-statement",
+    "title": "Void Functions",
     "section": "",
-    "text": "int size(int x)\n \n\nint size(real x)\nReturn the size of x which for scalar-valued x is 1\nAvailable since 2.26",
+    "text": "The reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).\n \n\nvoid reject(T1 x1,..., TN xN) Reject the current iteration and print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Integer-Valued Basic Functions"
+      "Void Functions"
     ]
   },
   {
-    "objectID": "functions-reference/integer-valued_basic_functions.html#casting-functions",
-    "href": "functions-reference/integer-valued_basic_functions.html#casting-functions",
-    "title": "Integer-Valued Basic Functions",
+    "objectID": "functions-reference/void_functions.html#fatal-error-statement",
+    "href": "functions-reference/void_functions.html#fatal-error-statement",
+    "title": "Void Functions",
     "section": "",
-    "text": "It is possible to cast real numbers to integers as long as the real value is data. See data only qualifiers in the Stan Reference Manual.\n \n\nint to_int(data real x)\nReturn the value x truncated to an integer. This will throw an error if the value of x is too big to represent as a 32-bit signed integer.\nThis is similar to trunc (see Rounding functions) but the return type is of type int. For example, to_int(3.9) is 3, and to_int(-3.9) is -3.\nAvailable since 2.31\n \n\nI to_int(data T x)\nThe vectorized version of to_int. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.\nAvailable since 2.31",
+    "text": "The fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.\n \n\nvoid fatal_error(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.35",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Integer-Valued Basic Functions"
+      "Void Functions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html",
-    "href": "functions-reference/higher-order_functions.html",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html",
+    "href": "functions-reference/unbounded_continuous_distributions.html",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides a few higher-order functions that act on other functions. In all cases, the function arguments to the higher-order functions are defined as functions within the Stan language and passed by name to the higher-order functions.\n\n\nStan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.\nAn algebraic solver is a higher-order function, i.e. it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section Ordinary Differential Equation (ODE) Solvers and Differential Algebraic Equation (DAE) solver). Ordinary Stan functions do not allow functions as arguments.\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The function must return a vector and takes in, as its first argument, the unknowns \\(y\\) we wish to solve for, also passed as a vector. This argument is followed by additional arguments as specified by the user; we call such arguments variadic arguments and denote them .... The signature of the algebraic system is then:\n vector algebra_system (vector y, ...)\nThere is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword data. In the below example, the last variadic argument, \\(x\\), is restricted to being data:\n  vector algebra_system (vector y, vector theta, data vector x)\nDistinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.\n\n\n\n \n vector solve_newton(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.31\n \n vector solve_newton_tol(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.31\n \n vector solve_powell(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method.\nAvailable since 2.31\n \n vector solve_powell_tol(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.\nAvailable since 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, ...):vector. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\n...: variadic arguments.\n\nThe algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:\n\nscaling_step: for the Newton solver only, the scaled-step stopping tolerance, type real, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is \\(10^{-3}\\).\nrel_tol: for the Powell solver only, the relative tolerance, type real, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is \\(10^{-10}\\).\nfunction_tol: function tolerance for the algebraic solver, type real, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is \\(10^{-6}\\).\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is \\(200\\).\n\nThe difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0 (approximately, within the specified function tolerance).\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\n\n\n\nStan offers two methods to solve algebraic equations. solve_newton and solve_newton_tol use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite (Hindmarsh et al. 2005). For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.\nsolve_powell and solve_powell_tol are based on the Powell hybrid method (Powell 1970), which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library (Guennebaud, Jacob, et al. 2010). This solver is in turn based on the algorithm developed for the package MINPACK-1 (Jorge J. More 1980).\nFor both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see (Gaebler 2021) and (Margossian and Betancourt 2022).\n\n\n\n\n\nStan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).\nSolving an initial value ODE means given a set of differential equations \\(y'(t, \\theta) = f(t, y, \\theta)\\) and initial conditions \\(y(t_0, \\theta)\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 &lt; t_2, \\cdots &lt; t_n\\). \\(f(t, y, \\theta)\\) is referred to here as the ODE system function.\n\\(f(t, y, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.\nTo make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as .... The types of the arguments represented by ... in the ODE solve function call must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector ode_rk45(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.24\n \n\narray[] vector ode_rk45_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.24\n \n\narray[] vector ode_ckrk(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.\nAvailable since 2.27\n \n\narray[] vector ode_ckrk_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.27\n \n\narray[] vector ode_adams(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.24\n \n\narray[] vector ode_adams_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_bdf(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.24\n \n\narray[] vector ode_bdf_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_adjoint_tol_ctl(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)\nSolves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.\nAvailable since 2.27\n\n\n\nThe first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a vector return type, and the first two arguments must be a real and vector in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:\n vector ode(real time, vector state, ...)\nThe ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\n..., sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the ... arguments of the ODE solve function call.\n\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see Arguments to the adjoint ODE solver.\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nFor the versions of the ode solver functions ending in _tol, these three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the ODE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\n\nThe arguments to the adjoint ODE solver are different from those for the other functions (for those see Arguments to the ODE solvers).\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\ndata rel_tol_forward: Relative tolerance for forward solve, type real, data only,\ndata abs_tol_forward: Absolute tolerance vector for each state for forward solve, type vector, data only,\ndata rel_tol_backward: Relative tolerance for backward solve, type real, data only,\ndata abs_tol_backward: Absolute tolerance vector for each state for backward solve, type vector, data only,\ndata rel_tol_quadrature: Relative tolerance for backward quadrature, type real, data only,\ndata abs_tol_quadrature: Absolute tolerance for backward quadrature, type real, data only,\ndata max_num_steps: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type int, data only,\nnum_steps_between_checkpoints: number of steps between checkpointing forward solution, type int, data only,\ninterpolation_polynomial: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type int, data only,\nsolver_forward: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only,\nsolver_backward: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only.\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe return value for the ODE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time in specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.\n\n\n\n\n\n\nStan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 (Serban et al. 2021).\nSolving an initial value DAE means given a set of residual functions \\(r(y'(t, \\theta), y(t, \\theta), t)\\) and initial conditions \\((y(t_0, \\theta), y'(t_0, \\theta))\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 \\leq t_2, \\cdots \\leq t_n\\). The residual function \\(r(y', y, t, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.\nSimilar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as ..., and the types of these arguments, also represented by ... in the DAE solver call, must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector dae(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) Solves the DAE system using the backward differentiation formula (BDF) method (Serban et al. 2021).\nAvailable since 2.29\n \n\narray[] vector dae_tol(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.29\n\n\n\nThe first argument to the DAE solver is the DAE residual function. The DAE residual function must have a vector return type, and the first three arguments must be a real, vector, and vector, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:\n  vector residual(real time, vector state, vector state_derivative, ...)\nThe DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the DAE system\nstate, the state of the DAE system at the time specified\nstate_derivative, the time derivatives of the state of the DAE system at the time specified\n..., sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the ... arguments of the DAE solve function call.\n\n\n\n\nThe arguments to the DAE solver are\n\nresidual: DAE residual function,\ninitial_state: initial state, type vector,\ninitial_state_derivative: time derivative of the initial state, type vector,\ninitial_time: initial time, type data real,\ntimes: solution times, type data array[] real,\n...: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the ... arguments of the DAE residual function.\n\nFor dae_tol, the following three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the DAE solver, type real, data only,\ndata abs_tol: absolute tolerance for the DAE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the DAE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe user is responsible to ensure the residual function becomes zero at the initial time, t0, when the arguments initial_state and initial_state_derivative are introduced as state and state_derivative, respectively.\n\n\n\nThe return value for the DAE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.\n\n\n\n\n\n\nStan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.\nIt operates similarly to the algebraic solver and the ordinary differential equations solver in that it allows as an argument a function.\nLike both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.\n\n\nPerforming a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:\nreal integrand(real x, real xc, array[] real theta,\n               array[] real x_r, array[] int x_i)\nThe function should return the value of the integrand evaluated at the point x.\nThe argument of this function are:\n\nx, the independent variable being integrated over\nxc, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section Precision Loss).\ntheta, parameter values used to evaluate the integral\nx_r, data values used to evaluate the integral\nx_i, integer data used to evaluate the integral\n\nLike algebraic solver and the differential equations solver, the 1D integrator separates parameter values, theta, from data values, x_r.\n\n\n\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) Integrates the integrand from a to b.\nAvailable since 2.23\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance) Integrates the integrand from a to b with the given relative tolerance.\nAvailable since 2.23\n\n\nThe arguments to the 1D integrator are as follows:\n\nintegrand: function literal referring to a function specifying the integrand with signature (real, real, array[] real, array[] real, array[] int):real The arguments represent\n\n\nwhere integrand is evaluated,\n\n\ndistance from evaluation point to integration limit for definite integrals,\n\n\nparameters,\n\n\nreal data\n\n\ninteger data, and the return value is the integrand evaluated at the given point,\n\n\na: left limit of integration, may be negative infinity, type real,\nb: right limit of integration, may be positive infinity, type real,\ntheta: parameters only, type array[] real,\nx_r: real data only, type array[] real,\nx_i: integer data only, type array[] int.\n\nA relative_tolerance argument can optionally be provided for more control over the algorithm:\n\nrelative_tolerance: relative tolerance for the 1d integrator, type real, data only.\n\n\n\n\nThe return value for the 1D integrator is a real, the value of the integral.\n\n\n\nFor numeric stability, integrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). Each integral is separately integrated to the given relative_tolerance.\n\n\n\nWhen integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, 1.0 - x). xc (as passed to the integrand) is a high-precision version of the distance between x and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, xc will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, xc is set to NaN.\n\n\n\nInternally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in (Takahasi and Mori 1974), (Mori 1978), (Bailey, Jeyabalan, and Li 2005), and (Tanaka et al. 2009).\nThe gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.\n\n\n\n\n\nStan provides a higher-order reduce function for summation. A function which returns a scalar g: U -&gt; real is mapped to every element of a list of type array[] U, { x1, x2, ... } and all the results are accumulated,\ng(x1) + g(x2) + ...\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g itself, but instead works through evaluating partial sums, f: array[] U -&gt; real, where:\nf({ x1 }) = g(x1)\nf({ x1, x2 }) = g(x1) + g(x2)\nf({ x1, x2, ... }) = g(x1) + g(x2) + ...\nMathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load). This should result in stable numerical evaluations.\n\n\n\nThe higher-order reduce function takes a partial sum function f, an array argument x (with one array element for each term in the sum), a recommended grainsize, and a set of shared arguments. This representation allows parallelization of the resultant sum.\n \n\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n \n real reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nReturns the equivalent of f(x, 1, size(x), s1, s2, ...), but computes the result in parallel by breaking the array x into independent partial sums. s1, s2, ... are shared between all terms in the sum.\nAvailable since 2.23\n\nf: function literal referring to a function specifying the partial sum operation. Refer to the partial sum function.\nx: array of T, one for each term of the reduction, T can be any type,\ngrainsize: For reduce_sum, grainsize is the recommended size of the partial sum (grainsize = 1 means pick totally automatically). For reduce_sum_static, grainsize determines the maximum size of the partial sums, type int,\ns1: first (optional) shared argument, type T1, where T1 can be any type\ns2: second (optional) shared argument, type T2, where T2 can be any type,\n...: remainder of shared arguments, each of which can be any type.\n\n\n\n\nThe partial sum function must have the following signature where the type T, and the types of all the shared arguments (T1, T2, …) match those of the original reduce_sum (reduce_sum_static) call.\n(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real\nThe partial sum function returns the sum of the start to end terms (inclusive) of the overall calculations. The arguments to the partial sum function are:\n\nx_subset, the subset of x a given partial sum is responsible for computing, type array[] T, where T matches the type of x in reduce_sum (reduce_sum_static)\nstart, the index of the first term of the partial sum, type int\nend, the index of the last term of the partial sum (inclusive), type int\ns1, first shared argument, type T1, matching type of s1 in reduce_sum (reduce_sum_static)\ns2, second shared argument, type T2, matching type of s2 in reduce_sum (reduce_sum_static)\n..., remainder of shared arguments, with types matching those in reduce_sum (reduce_sum_static)\n\n\n\n\n\nStan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.\n\n\nThe function being mapped must have a signature identical to that of the function f in the following declaration.\n vector f(vector phi, vector theta,\n          data array[] real x_r, data array[] int x_i);\nThe map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:\n\nphi, the sequence of parameters shared across shards\ntheta, the sequence of parameters specific to this shard\nx_r, sequence of real-valued data\nx_i, sequence of integer data\n\nAll input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.\n\n\n\nThe rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.\n \n\nvector map_rect(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) Return the concatenation of the results of applying the function f, of type (vector, vector, array[] real, array[] int):vector elementwise, i.e., f(phi, theta[n], x_r[n], x_i[n]) for each n in 1:N, where N is the size of the parallel arrays of job-specific/local parameters theta, real data x_r, and integer data x_r. The shared/global parameters phi are passed to each invocation of f.\nAvailable since 2.18",
+    "text": "The unbounded univariate continuous probability distributions have support on all real numbers.\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Normal}(y|\\mu,\\sigma) =\n\\frac{1}{\\sqrt{2 \\pi} \\ \\sigma} \\exp\\left( - \\, \\frac{1}{2} \\left(\\frac{y - \\mu}{\\sigma} \\right)^2 \\right) \\!.\n\\end{equation*}\\]\n\n\n\ny ~ normal(mu, sigma)\nIncrement target log probability density with normal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal normal_lpdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal normal_lupdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_cdf(reals y | reals mu, reals sigma) The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 1 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; the function Phi_approx is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.0\n \n\nreal normal_lcdf(reals y | reals mu, reals sigma) The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.12\n \n\nreal normal_lccdf(reals y | reals mu, reals sigma) The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.15\n \n\nR normal_rng(reals mu, reals sigma) Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nThe standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. \\[\\begin{equation*}\n\\text{StdNormal}(y) \\ = \\ \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{1}{\\sqrt{2 \\pi}} \\, \\exp \\left( \\frac{-y^2}{2} \\right)\\!.\n\\end{equation*}\\] Up to a proportion on the log scale, where Stan computes, \\[\\begin{equation*}\n\\log \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{-y^2}{2} + \\text{const}.\n\\end{equation*}\\] With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location \\(0\\) and scale \\(1\\).\n\n\n\ny ~ std_normal()\nIncrement target log probability density with std_normal_lupdf(y).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal std_normal_lpdf(reals y) The standard normal (location zero, scale one) log probability density of y.\nAvailable since 2.18\n \n\nreal std_normal_lupdf(reals y) The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.\nAvailable since 2.25\n \n\nreal std_normal_cdf(reals y) The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for \\(y\\) below -37.5 and overflow to 1 for \\(y\\) above 8.25; the function Phi_approx is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lcdf(reals y) The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to \\(-\\infty\\) for \\(y\\) below -37.5 and overflow to 0 for \\(y\\) above 8.25; log(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lccdf(reals y) The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for \\(y\\) below -37.5 and underflow to \\(-\\infty\\) for \\(y\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nR std_normal_qf(T x) Returns the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The std_normal_qf is equivalent to the inv_Phi function.\nAvailable since 2.31\n \n\nR std_normal_log_qf(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) evaluated at the log of the specified quantile x. This function is equivalent to std_normal_qf(exp(x)) but is more numerically stable.\nAvailable since 2.31\n \n\nreal std_normal_rng() Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.21\n\n\n\n\nStan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e. a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\sigma\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^n\\), \\[\\begin{equation*} \\text{NormalIdGLM}(y|x, \\alpha, \\beta, \\sigma) = \\prod_{1\\leq i\n\\leq n}\\text{Normal}(y_i|\\alpha_i + x_i\\cdot \\beta, \\sigma). \\end{equation*}\\]\n\n\n\ny ~ normal_id_glm(x, alpha, beta, sigma)\nIncrement target log probability density with normal_id_glm_lupdf(y | x, alpha, beta, sigma).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n\n\n\n\nExponentially modified Gaussian describes the distribution of \\(Z = X + Y\\) when \\(X\\) and \\(Y\\) are independent and \\(X\\) is normally distributed (with mean \\(\\mu\\) and standard deviation \\(\\sigma\\)) and \\(Y\\) is exponentially distributed (with rate \\(\\lambda\\)).\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\), and \\(\\lambda \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{ExpModNormal}(y|\\mu,\\sigma,\\lambda) = \\frac{\\lambda}{2} \\ \\exp\n\\left(\\frac{\\lambda}{2} \\left(2\\mu + \\lambda \\sigma^2 -\n2y\\right)\\right) \\text{erfc}\\left(\\frac{\\mu + \\lambda\\sigma^2 -\ny}{\\sqrt{2}\\sigma}\\right) . \\end{equation*}\\]\n\n\n\ny ~ exp_mod_normal(mu, sigma, lambda)\nIncrement target log probability density with exp_mod_normal_lupdf(y | mu, sigma, lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exp_mod_normal_lpdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lupdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal exp_mod_normal_cdf(reals y | reals mu, reals sigma, reals lambda) The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.0\n \n\nreal exp_mod_normal_lcdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lccdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nR exp_mod_normal_rng(reals mu, reals sigma, reals lambda) Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\xi \\in \\mathbb{R}\\), \\(\\omega \\in \\mathbb{R}^+\\), and \\(\\alpha \\in\n\\mathbb{R}\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*} \\text{SkewNormal}(y \\mid\n\\xi, \\omega, \\alpha) = \\frac{1}{\\omega\\sqrt{2\\pi}} \\ \\exp\\left( - \\,\n\\frac{1}{2}            \\left(  \\frac{y - \\xi}{\\omega} \\right)^2\n\\right) \\ \\left(1 + \\text{erf}\\left( \\alpha\\left(\\frac{y -\n\\xi}{\\omega\\sqrt{2}}\\right)\\right)\\right) . \\end{equation*}\\]\n\n\n\ny ~ skew_normal(xi, omega, alpha)\nIncrement target log probability density with skew_normal_lupdf(y | xi, omega, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal skew_normal_lpdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lupdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal skew_normal_cdf(reals y | reals xi, reals omega, reals alpha) The skew normal distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lcdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nreal skew_normal_lccdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nR skew_normal_rng(reals xi, reals omega, real alpha) Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}\\), and \\(\\sigma \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{StudentT}(y|\\nu,\\mu,\\sigma) = \\frac{\\Gamma\\left((\\nu +\n1)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\nu \\pi} \\ \\sigma}\n\\ \\left( 1 + \\frac{1}{\\nu} \\left(\\frac{y - \\mu}{\\sigma}\\right)^2\n\\right)^{-(\\nu + 1)/2} \\! . \\end{equation*}\\]\n\n\n\ny ~ student_t(nu, mu, sigma)\nIncrement target log probability density with student_t_lupdf(y | nu, mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal student_t_lpdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lupdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal student_t_cdf(reals y | reals nu, reals mu, reals sigma) The Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.0\n \n\nreal student_t_lcdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lccdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nR student_t_rng(reals nu, reals mu, reals sigma) Generate a Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Cauchy}(y|\\mu,\\sigma) = \\frac{1}{\\pi \\sigma} \\ \\frac{1}{1 + \\left((y - \\mu)/\\sigma\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ cauchy(mu, sigma)\nIncrement target log probability density with cauchy_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal cauchy_lpdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lupdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal cauchy_cdf(reals y | reals mu, reals sigma) The Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal cauchy_lcdf(reals y | reals mu, reals sigma) The log of the Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lccdf(reals y | reals mu, reals sigma) The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR cauchy_rng(reals mu, reals sigma) Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{DoubleExponential}(y|\\mu,\\sigma) =\n\\frac{1}{2\\sigma}   \\exp \\left( - \\, \\frac{|y - \\mu|}{\\sigma} \\right)\n. \\end{equation*}\\] Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section exponential distribution), which is parameterized in terms of inverse scale.\nThe double-exponential distribution can be defined as a compound exponential-normal distribution (Ding and Blitzstein 2018). Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write \\[\\begin{equation*} \\alpha \\sim\n\\mathsf{Exponential}\\left( \\frac{1}{2 \\sigma^2} \\right) \\end{equation*}\\] and \\[\\begin{equation*} \\beta \\mid\n\\alpha \\sim \\mathsf{Normal}(\\mu, \\sqrt{\\alpha}), \\end{equation*}\\] then \\[\\begin{equation*} \\beta \\sim\n\\mathsf{DoubleExponential}(\\mu, \\sigma ). \\end{equation*}\\] This may be used to code a non-centered parameterization by taking \\[\\begin{equation*} \\beta^{\\text{raw}} \\sim\n\\mathsf{Normal}(0, 1) \\end{equation*}\\] and defining \\[\\begin{equation*} \\beta = \\mu + \\sqrt{\\alpha} \\,\n\\beta^{\\text{raw}}. \\end{equation*}\\]\n\n\n\ny ~ double_exponential(mu, sigma)\nIncrement target log probability density with double_exponential_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal double_exponential_lpdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lupdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal double_exponential_cdf(reals y | reals mu, reals sigma) The double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal double_exponential_lcdf(reals y | reals mu, reals sigma) The log of the double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lccdf(reals y | reals mu, reals sigma) The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR double_exponential_rng(reals mu, reals sigma) Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Logistic}(y|\\mu,\\sigma) = \\frac{1}{\\sigma} \\ \\exp\\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right)\n\\ \\left(1 + \\exp \\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right) \\right)^{\\!-2} \\! .\n\\end{equation*}\\]\n\n\n\ny ~ logistic(mu, sigma)\nIncrement target log probability density with logistic_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal logistic_lpdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lupdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal logistic_cdf(reals y | reals mu, reals sigma) The logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal logistic_lcdf(reals y | reals mu, reals sigma) The log of the logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lccdf(reals y | reals mu, reals sigma) The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR logistic_rng(reals mu, reals sigma) Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Gumbel}(y|\\mu,\\beta) = \\frac{1}{\\beta} \\ \\exp\\left(-\\frac{y-\\mu}{\\beta}-\\exp\\left(-\\frac{y-\\mu}{\\beta}\\right)\\right) .\n\\end{equation*}\\]\n\n\n\ny ~ gumbel(mu, beta)\nIncrement target log probability density with gumbel_lupdf(y | mu, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gumbel_lpdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lupdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gumbel_cdf(reals y | reals mu, reals beta) The gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.0\n \n\nreal gumbel_lcdf(reals y | reals mu, reals beta) The log of the gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lccdf(reals y | reals mu, reals beta) The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nR gumbel_rng(reals mu, reals beta) Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\) and \\(\\tau \\in [0, 1]\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{aligned}\n& \\text{SkewDoubleExponential}  (y|\\mu,\\sigma, \\tau) = \\\\\n& \\qquad \\qquad \\frac{2 \\tau (1 - \\tau) }{\\sigma} \\exp \\left[ - \\frac{2}{\\sigma} \\left[ \\left(1 - \\tau \\right) I(y &lt; \\mu)  (\\mu - y) + \\tau I(y &gt; \\mu)(y-\\mu)  \\right]    \\right]\n\\end{aligned}\\]\n\n\n\ny ~ skew_double_exponential(mu, sigma, tau)\nIncrement target log probability density with skew_double_exponential(y | mu, sigma, tau)\nAvailable since 2.28\n \n\n\n\n\n \n\nreal skew_double_exponential_lpdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lupdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms\nAvailable since 2.28\n \n\nreal skew_double_exponential_cdf(reals y | reals mu, reals sigma, reals tau) The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lcdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lccdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nR skew_double_exponential_rng(reals mu, reals sigma, reals tau) Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html#functions-algebraic-solver",
-    "href": "functions-reference/higher-order_functions.html#functions-algebraic-solver",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#normal-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#normal-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.\nAn algebraic solver is a higher-order function, i.e. it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section Ordinary Differential Equation (ODE) Solvers and Differential Algebraic Equation (DAE) solver). Ordinary Stan functions do not allow functions as arguments.\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The function must return a vector and takes in, as its first argument, the unknowns \\(y\\) we wish to solve for, also passed as a vector. This argument is followed by additional arguments as specified by the user; we call such arguments variadic arguments and denote them .... The signature of the algebraic system is then:\n vector algebra_system (vector y, ...)\nThere is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword data. In the below example, the last variadic argument, \\(x\\), is restricted to being data:\n  vector algebra_system (vector y, vector theta, data vector x)\nDistinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.\n\n\n\n \n vector solve_newton(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.31\n \n vector solve_newton_tol(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.31\n \n vector solve_powell(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method.\nAvailable since 2.31\n \n vector solve_powell_tol(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.\nAvailable since 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, ...):vector. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\n...: variadic arguments.\n\nThe algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:\n\nscaling_step: for the Newton solver only, the scaled-step stopping tolerance, type real, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is \\(10^{-3}\\).\nrel_tol: for the Powell solver only, the relative tolerance, type real, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is \\(10^{-10}\\).\nfunction_tol: function tolerance for the algebraic solver, type real, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is \\(10^{-6}\\).\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is \\(200\\).\n\nThe difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0 (approximately, within the specified function tolerance).\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\n\n\n\nStan offers two methods to solve algebraic equations. solve_newton and solve_newton_tol use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite (Hindmarsh et al. 2005). For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.\nsolve_powell and solve_powell_tol are based on the Powell hybrid method (Powell 1970), which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library (Guennebaud, Jacob, et al. 2010). This solver is in turn based on the algorithm developed for the package MINPACK-1 (Jorge J. More 1980).\nFor both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see (Gaebler 2021) and (Margossian and Betancourt 2022).",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Normal}(y|\\mu,\\sigma) =\n\\frac{1}{\\sqrt{2 \\pi} \\ \\sigma} \\exp\\left( - \\, \\frac{1}{2} \\left(\\frac{y - \\mu}{\\sigma} \\right)^2 \\right) \\!.\n\\end{equation*}\\]\n\n\n\ny ~ normal(mu, sigma)\nIncrement target log probability density with normal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal normal_lpdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal normal_lupdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_cdf(reals y | reals mu, reals sigma) The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 1 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; the function Phi_approx is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.0\n \n\nreal normal_lcdf(reals y | reals mu, reals sigma) The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.12\n \n\nreal normal_lccdf(reals y | reals mu, reals sigma) The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.15\n \n\nR normal_rng(reals mu, reals sigma) Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nThe standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. \\[\\begin{equation*}\n\\text{StdNormal}(y) \\ = \\ \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{1}{\\sqrt{2 \\pi}} \\, \\exp \\left( \\frac{-y^2}{2} \\right)\\!.\n\\end{equation*}\\] Up to a proportion on the log scale, where Stan computes, \\[\\begin{equation*}\n\\log \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{-y^2}{2} + \\text{const}.\n\\end{equation*}\\] With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location \\(0\\) and scale \\(1\\).\n\n\n\ny ~ std_normal()\nIncrement target log probability density with std_normal_lupdf(y).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal std_normal_lpdf(reals y) The standard normal (location zero, scale one) log probability density of y.\nAvailable since 2.18\n \n\nreal std_normal_lupdf(reals y) The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.\nAvailable since 2.25\n \n\nreal std_normal_cdf(reals y) The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for \\(y\\) below -37.5 and overflow to 1 for \\(y\\) above 8.25; the function Phi_approx is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lcdf(reals y) The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to \\(-\\infty\\) for \\(y\\) below -37.5 and overflow to 0 for \\(y\\) above 8.25; log(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lccdf(reals y) The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for \\(y\\) below -37.5 and underflow to \\(-\\infty\\) for \\(y\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nR std_normal_qf(T x) Returns the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The std_normal_qf is equivalent to the inv_Phi function.\nAvailable since 2.31\n \n\nR std_normal_log_qf(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) evaluated at the log of the specified quantile x. This function is equivalent to std_normal_qf(exp(x)) but is more numerically stable.\nAvailable since 2.31\n \n\nreal std_normal_rng() Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.21",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html#functions-ode-solver",
-    "href": "functions-reference/higher-order_functions.html#functions-ode-solver",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#normal-id-glm",
+    "href": "functions-reference/unbounded_continuous_distributions.html#normal-id-glm",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).\nSolving an initial value ODE means given a set of differential equations \\(y'(t, \\theta) = f(t, y, \\theta)\\) and initial conditions \\(y(t_0, \\theta)\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 &lt; t_2, \\cdots &lt; t_n\\). \\(f(t, y, \\theta)\\) is referred to here as the ODE system function.\n\\(f(t, y, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.\nTo make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as .... The types of the arguments represented by ... in the ODE solve function call must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector ode_rk45(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.24\n \n\narray[] vector ode_rk45_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.24\n \n\narray[] vector ode_ckrk(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.\nAvailable since 2.27\n \n\narray[] vector ode_ckrk_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.27\n \n\narray[] vector ode_adams(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.24\n \n\narray[] vector ode_adams_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_bdf(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.24\n \n\narray[] vector ode_bdf_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_adjoint_tol_ctl(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)\nSolves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.\nAvailable since 2.27\n\n\n\nThe first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a vector return type, and the first two arguments must be a real and vector in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:\n vector ode(real time, vector state, ...)\nThe ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\n..., sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the ... arguments of the ODE solve function call.\n\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see Arguments to the adjoint ODE solver.\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nFor the versions of the ode solver functions ending in _tol, these three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the ODE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\n\nThe arguments to the adjoint ODE solver are different from those for the other functions (for those see Arguments to the ODE solvers).\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\ndata rel_tol_forward: Relative tolerance for forward solve, type real, data only,\ndata abs_tol_forward: Absolute tolerance vector for each state for forward solve, type vector, data only,\ndata rel_tol_backward: Relative tolerance for backward solve, type real, data only,\ndata abs_tol_backward: Absolute tolerance vector for each state for backward solve, type vector, data only,\ndata rel_tol_quadrature: Relative tolerance for backward quadrature, type real, data only,\ndata abs_tol_quadrature: Absolute tolerance for backward quadrature, type real, data only,\ndata max_num_steps: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type int, data only,\nnum_steps_between_checkpoints: number of steps between checkpointing forward solution, type int, data only,\ninterpolation_polynomial: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type int, data only,\nsolver_forward: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only,\nsolver_backward: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only.\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe return value for the ODE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time in specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.",
+    "text": "Stan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e. a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\sigma\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^n\\), \\[\\begin{equation*} \\text{NormalIdGLM}(y|x, \\alpha, \\beta, \\sigma) = \\prod_{1\\leq i\n\\leq n}\\text{Normal}(y_i|\\alpha_i + x_i\\cdot \\beta, \\sigma). \\end{equation*}\\]\n\n\n\ny ~ normal_id_glm(x, alpha, beta, sigma)\nIncrement target log probability density with normal_id_glm_lupdf(y | x, alpha, beta, sigma).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html#functions-dae-solver",
-    "href": "functions-reference/higher-order_functions.html#functions-dae-solver",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#exponentially-modified-normal-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#exponentially-modified-normal-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 (Serban et al. 2021).\nSolving an initial value DAE means given a set of residual functions \\(r(y'(t, \\theta), y(t, \\theta), t)\\) and initial conditions \\((y(t_0, \\theta), y'(t_0, \\theta))\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 \\leq t_2, \\cdots \\leq t_n\\). The residual function \\(r(y', y, t, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.\nSimilar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as ..., and the types of these arguments, also represented by ... in the DAE solver call, must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector dae(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) Solves the DAE system using the backward differentiation formula (BDF) method (Serban et al. 2021).\nAvailable since 2.29\n \n\narray[] vector dae_tol(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.29\n\n\n\nThe first argument to the DAE solver is the DAE residual function. The DAE residual function must have a vector return type, and the first three arguments must be a real, vector, and vector, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:\n  vector residual(real time, vector state, vector state_derivative, ...)\nThe DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the DAE system\nstate, the state of the DAE system at the time specified\nstate_derivative, the time derivatives of the state of the DAE system at the time specified\n..., sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the ... arguments of the DAE solve function call.\n\n\n\n\nThe arguments to the DAE solver are\n\nresidual: DAE residual function,\ninitial_state: initial state, type vector,\ninitial_state_derivative: time derivative of the initial state, type vector,\ninitial_time: initial time, type data real,\ntimes: solution times, type data array[] real,\n...: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the ... arguments of the DAE residual function.\n\nFor dae_tol, the following three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the DAE solver, type real, data only,\ndata abs_tol: absolute tolerance for the DAE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the DAE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe user is responsible to ensure the residual function becomes zero at the initial time, t0, when the arguments initial_state and initial_state_derivative are introduced as state and state_derivative, respectively.\n\n\n\nThe return value for the DAE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.",
+    "text": "Exponentially modified Gaussian describes the distribution of \\(Z = X + Y\\) when \\(X\\) and \\(Y\\) are independent and \\(X\\) is normally distributed (with mean \\(\\mu\\) and standard deviation \\(\\sigma\\)) and \\(Y\\) is exponentially distributed (with rate \\(\\lambda\\)).\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\), and \\(\\lambda \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{ExpModNormal}(y|\\mu,\\sigma,\\lambda) = \\frac{\\lambda}{2} \\ \\exp\n\\left(\\frac{\\lambda}{2} \\left(2\\mu + \\lambda \\sigma^2 -\n2y\\right)\\right) \\text{erfc}\\left(\\frac{\\mu + \\lambda\\sigma^2 -\ny}{\\sqrt{2}\\sigma}\\right) . \\end{equation*}\\]\n\n\n\ny ~ exp_mod_normal(mu, sigma, lambda)\nIncrement target log probability density with exp_mod_normal_lupdf(y | mu, sigma, lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exp_mod_normal_lpdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lupdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal exp_mod_normal_cdf(reals y | reals mu, reals sigma, reals lambda) The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.0\n \n\nreal exp_mod_normal_lcdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lccdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nR exp_mod_normal_rng(reals mu, reals sigma, reals lambda) Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html#functions-1d-integrator",
-    "href": "functions-reference/higher-order_functions.html#functions-1d-integrator",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#skew-normal-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#skew-normal-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.\nIt operates similarly to the algebraic solver and the ordinary differential equations solver in that it allows as an argument a function.\nLike both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.\n\n\nPerforming a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:\nreal integrand(real x, real xc, array[] real theta,\n               array[] real x_r, array[] int x_i)\nThe function should return the value of the integrand evaluated at the point x.\nThe argument of this function are:\n\nx, the independent variable being integrated over\nxc, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section Precision Loss).\ntheta, parameter values used to evaluate the integral\nx_r, data values used to evaluate the integral\nx_i, integer data used to evaluate the integral\n\nLike algebraic solver and the differential equations solver, the 1D integrator separates parameter values, theta, from data values, x_r.\n\n\n\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) Integrates the integrand from a to b.\nAvailable since 2.23\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance) Integrates the integrand from a to b with the given relative tolerance.\nAvailable since 2.23\n\n\nThe arguments to the 1D integrator are as follows:\n\nintegrand: function literal referring to a function specifying the integrand with signature (real, real, array[] real, array[] real, array[] int):real The arguments represent\n\n\nwhere integrand is evaluated,\n\n\ndistance from evaluation point to integration limit for definite integrals,\n\n\nparameters,\n\n\nreal data\n\n\ninteger data, and the return value is the integrand evaluated at the given point,\n\n\na: left limit of integration, may be negative infinity, type real,\nb: right limit of integration, may be positive infinity, type real,\ntheta: parameters only, type array[] real,\nx_r: real data only, type array[] real,\nx_i: integer data only, type array[] int.\n\nA relative_tolerance argument can optionally be provided for more control over the algorithm:\n\nrelative_tolerance: relative tolerance for the 1d integrator, type real, data only.\n\n\n\n\nThe return value for the 1D integrator is a real, the value of the integral.\n\n\n\nFor numeric stability, integrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). Each integral is separately integrated to the given relative_tolerance.\n\n\n\nWhen integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, 1.0 - x). xc (as passed to the integrand) is a high-precision version of the distance between x and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, xc will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, xc is set to NaN.\n\n\n\nInternally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in (Takahasi and Mori 1974), (Mori 1978), (Bailey, Jeyabalan, and Li 2005), and (Tanaka et al. 2009).\nThe gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.",
+    "text": "If \\(\\xi \\in \\mathbb{R}\\), \\(\\omega \\in \\mathbb{R}^+\\), and \\(\\alpha \\in\n\\mathbb{R}\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*} \\text{SkewNormal}(y \\mid\n\\xi, \\omega, \\alpha) = \\frac{1}{\\omega\\sqrt{2\\pi}} \\ \\exp\\left( - \\,\n\\frac{1}{2}            \\left(  \\frac{y - \\xi}{\\omega} \\right)^2\n\\right) \\ \\left(1 + \\text{erf}\\left( \\alpha\\left(\\frac{y -\n\\xi}{\\omega\\sqrt{2}}\\right)\\right)\\right) . \\end{equation*}\\]\n\n\n\ny ~ skew_normal(xi, omega, alpha)\nIncrement target log probability density with skew_normal_lupdf(y | xi, omega, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal skew_normal_lpdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lupdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal skew_normal_cdf(reals y | reals xi, reals omega, reals alpha) The skew normal distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lcdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nreal skew_normal_lccdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nR skew_normal_rng(reals xi, reals omega, real alpha) Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html#functions-reduce",
-    "href": "functions-reference/higher-order_functions.html#functions-reduce",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#student-t-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#student-t-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides a higher-order reduce function for summation. A function which returns a scalar g: U -&gt; real is mapped to every element of a list of type array[] U, { x1, x2, ... } and all the results are accumulated,\ng(x1) + g(x2) + ...\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g itself, but instead works through evaluating partial sums, f: array[] U -&gt; real, where:\nf({ x1 }) = g(x1)\nf({ x1, x2 }) = g(x1) + g(x2)\nf({ x1, x2, ... }) = g(x1) + g(x2) + ...\nMathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load). This should result in stable numerical evaluations.\n\n\n\nThe higher-order reduce function takes a partial sum function f, an array argument x (with one array element for each term in the sum), a recommended grainsize, and a set of shared arguments. This representation allows parallelization of the resultant sum.\n \n\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n \n real reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nReturns the equivalent of f(x, 1, size(x), s1, s2, ...), but computes the result in parallel by breaking the array x into independent partial sums. s1, s2, ... are shared between all terms in the sum.\nAvailable since 2.23\n\nf: function literal referring to a function specifying the partial sum operation. Refer to the partial sum function.\nx: array of T, one for each term of the reduction, T can be any type,\ngrainsize: For reduce_sum, grainsize is the recommended size of the partial sum (grainsize = 1 means pick totally automatically). For reduce_sum_static, grainsize determines the maximum size of the partial sums, type int,\ns1: first (optional) shared argument, type T1, where T1 can be any type\ns2: second (optional) shared argument, type T2, where T2 can be any type,\n...: remainder of shared arguments, each of which can be any type.\n\n\n\n\nThe partial sum function must have the following signature where the type T, and the types of all the shared arguments (T1, T2, …) match those of the original reduce_sum (reduce_sum_static) call.\n(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real\nThe partial sum function returns the sum of the start to end terms (inclusive) of the overall calculations. The arguments to the partial sum function are:\n\nx_subset, the subset of x a given partial sum is responsible for computing, type array[] T, where T matches the type of x in reduce_sum (reduce_sum_static)\nstart, the index of the first term of the partial sum, type int\nend, the index of the last term of the partial sum (inclusive), type int\ns1, first shared argument, type T1, matching type of s1 in reduce_sum (reduce_sum_static)\ns2, second shared argument, type T2, matching type of s2 in reduce_sum (reduce_sum_static)\n..., remainder of shared arguments, with types matching those in reduce_sum (reduce_sum_static)",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}\\), and \\(\\sigma \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{StudentT}(y|\\nu,\\mu,\\sigma) = \\frac{\\Gamma\\left((\\nu +\n1)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\nu \\pi} \\ \\sigma}\n\\ \\left( 1 + \\frac{1}{\\nu} \\left(\\frac{y - \\mu}{\\sigma}\\right)^2\n\\right)^{-(\\nu + 1)/2} \\! . \\end{equation*}\\]\n\n\n\ny ~ student_t(nu, mu, sigma)\nIncrement target log probability density with student_t_lupdf(y | nu, mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal student_t_lpdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lupdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal student_t_cdf(reals y | reals nu, reals mu, reals sigma) The Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.0\n \n\nreal student_t_lcdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lccdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nR student_t_rng(reals nu, reals mu, reals sigma) Generate a Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/higher-order_functions.html#functions-map",
-    "href": "functions-reference/higher-order_functions.html#functions-map",
-    "title": "Higher-Order Functions",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#cauchy-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#cauchy-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "Stan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.\n\n\nThe function being mapped must have a signature identical to that of the function f in the following declaration.\n vector f(vector phi, vector theta,\n          data array[] real x_r, data array[] int x_i);\nThe map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:\n\nphi, the sequence of parameters shared across shards\ntheta, the sequence of parameters specific to this shard\nx_r, sequence of real-valued data\nx_i, sequence of integer data\n\nAll input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.\n\n\n\nThe rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.\n \n\nvector map_rect(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) Return the concatenation of the results of applying the function f, of type (vector, vector, array[] real, array[] int):vector elementwise, i.e., f(phi, theta[n], x_r[n], x_i[n]) for each n in 1:N, where N is the size of the parallel arrays of job-specific/local parameters theta, real data x_r, and integer data x_r. The shared/global parameters phi are passed to each invocation of f.\nAvailable since 2.18",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Cauchy}(y|\\mu,\\sigma) = \\frac{1}{\\pi \\sigma} \\ \\frac{1}{1 + \\left((y - \\mu)/\\sigma\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ cauchy(mu, sigma)\nIncrement target log probability density with cauchy_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal cauchy_lpdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lupdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal cauchy_cdf(reals y | reals mu, reals sigma) The Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal cauchy_lcdf(reals y | reals mu, reals sigma) The log of the Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lccdf(reals y | reals mu, reals sigma) The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR cauchy_rng(reals mu, reals sigma) Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Higher-Order Functions"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html",
-    "href": "functions-reference/functions_index.html",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#double-exponential-laplace-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#double-exponential-laplace-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "abs:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\nacos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nacosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nadd_diag:\n\n\n(complex_matrix m, complex_real d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_row_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, real d) : matrix (matrix_operations.html)\n\n\n(matrix m, row_vector d) : matrix (matrix_operations.html)\n\n\n(matrix m, vector d) : matrix (matrix_operations.html)\n\n\nalgebra_solver:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nalgebra_solver_newton:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector (deprecated_functions.html)\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nappend_array:\n\n\n(T x, T y) : T (array_operations.html)\n\n\nappend_col:\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(vector x, vector y) : matrix (matrix_operations.html)\n\n\nappend_row:\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, row_vector y) : matrix (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : matrix (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\narg:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nasin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nasinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan2:\n\n\n(T y, T x) : R (real-valued_basic_functions.html)\n\n\natanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\n\n\n\nbernoulli:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_cdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lccdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lcdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_logit:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_rng:\n\n\n(matrix x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\n(row_vector x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\nbernoulli_logit_lpmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_lupmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_rng:\n\n\n(reals alpha) : R (binary_distributions.html)\n\n\nbernoulli_lpmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lupmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_rng:\n\n\n(reals theta) : R (binary_distributions.html)\n\n\nbessel_first_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbessel_second_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_binomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbeta_binomial_cdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lccdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lcdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lpmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lupmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_rng:\n\n\n(ints N, reals alpha, reals beta) : R (bounded_discrete_distributions.html)\n\n\nbeta_cdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lccdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lcdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lpdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lupdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_cdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lccdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lcdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lpmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lupmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_rng:\n\n\n(reals r, reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nbeta_proportion:\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lccdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lcdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lpdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lupdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_rng:\n\n\n(reals mu, reals kappa) : R (continuous_distributions_on_0_1.html)\n\n\nbeta_rng:\n\n\n(reals alpha, reals beta) : R (continuous_distributions_on_0_1.html)\n\n\nbinary_log_loss:\n\n\n(int y, real y_hat) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbinomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_cdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lccdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lcdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lpmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lupmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lpmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lupmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lpmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lupmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_rng:\n\n\n(ints N, reals theta) : R (bounded_discrete_distributions.html)\n\n\nblock:\n\n\n(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows, int n_cols) : matrix (matrix_operations.html)\n\n\n\n\n\ncategorical:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lpmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lupmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_rng:\n\n\n(vector beta) : int (bounded_discrete_distributions.html)\n\n\ncategorical_lpmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_lupmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_rng:\n\n\n(vector theta) : int (bounded_discrete_distributions.html)\n\n\ncauchy:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ncauchy_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\ncbrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nceil:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nchi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nchi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\nchol2inv:\n\n\n(matrix L) : matrix (matrix_operations.html)\n\n\ncholesky_decompose:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ncholesky_factor_corr_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncholesky_factor_cov_constrain:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_jacobian:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\nchoose:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ncol:\n\n\n(complex_matrix x, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int n) : vector (matrix_operations.html)\n\n\ncols:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\ncolumns_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : row_vector (matrix_operations.html)\n\n\ncolumns_dot_self:\n\n\n(complex_matrix x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : row_vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\ncomplex_schur_decompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_t:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_u:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\nconj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(Z z) : Z (complex-valued_basic_functions.html)\n\n\ncorr_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncov_exp_quad:\n\n\n(array[] real x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(array[] real x1, array[] real x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x1, vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\ncov_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ncsr_extract:\n\n\n(matrix a) : tuple(vector, array[] int, array[] int) (sparse_matrix_operations.html)\n\n\ncsr_extract_u:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_v:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_w:\n\n\n(matrix a) : vector (sparse_matrix_operations.html)\n\n\ncsr_matrix_times_vector:\n\n\n(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector (sparse_matrix_operations.html)\n\n\ncsr_to_dense_matrix:\n\n\n(int m, int n, vector w, array[] int v, array[] int u) : matrix (sparse_matrix_operations.html)\n\n\ncumulative_sum:\n\n\n(array[] complex x) : array[] complex (complex_matrix_operations.html)\n\n\n(array[] int x) : array[] int (matrix_operations.html)\n\n\n(array[] real x) : array[] real (matrix_operations.html)\n\n\n(complex_row_vector rv) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\n\n\n\ndae:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\ndae_tol:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\ndeterminant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ndiag_matrix:\n\n\n(complex_vector x) : complex_matrix (complex_matrix_operations.html)\n\n\n(vector x) : matrix (matrix_operations.html)\n\n\ndiag_post_multiply:\n\n\n(complex_matrix m, complex_row_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\ndiag_pre_multiply:\n\n\n(complex_row_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(row_vector rv, matrix m) : matrix (matrix_operations.html)\n\n\n(vector v, matrix m) : matrix (matrix_operations.html)\n\n\ndiagonal:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\ndigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ndims:\n\n\n(T x) : array[] int (array_operations.html)\n\n\ndirichlet:\n\n\ndistribution statement (simplex_distributions.html)\n\n\ndirichlet_lpdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_lupdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_multinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lpmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lupmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_rng:\n\n\n(vector alpha, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\ndirichlet_rng:\n\n\n(vector alpha) : vector (simplex_distributions.html)\n\n\ndiscrete_range:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ndiscrete_range_cdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lccdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lcdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lpmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lupmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_rng:\n\n\n(ints l, ints u) : ints (bounded_discrete_distributions.html)\n\n\ndistance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\ndot_product:\n\n\n(complex_row_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(row_vector x, row_vector y) : real (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, row_vector y) : real (matrix_operations.html)\n\n\n(vector x, vector y) : real (matrix_operations.html)\n\n\ndot_self:\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\ndouble_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\n\n\n\ne:\n\n\n() : real (real-valued_basic_functions.html)\n\n\neigendecompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_vector) (matrix_operations.html)\n\n\neigendecompose_sym:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector) (matrix_operations.html)\n\n\neigenvalues:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : complex_vector (matrix_operations.html)\n\n\neigenvalues_sym:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\neigenvectors:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (matrix_operations.html)\n\n\neigenvectors_sym:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nerf:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nerfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp2:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp_mod_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_cdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_rng:\n\n\n(reals mu, reals sigma, reals lambda) : R (unbounded_continuous_distributions.html)\n\n\nexpm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexponential:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nexponential_cdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lccdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lcdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lpdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lupdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_rng:\n\n\n(reals beta) : R (positive_continuous_distributions.html)\n\n\n\n\n\nfalling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfatal_error:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nfdim:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfft:\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\nfft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\nfloor:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nfma:\n\n\n(real x, real y, real z) : real (real-valued_basic_functions.html)\n\n\nfmax:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmin:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmod:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfrechet:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nfrechet_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\n\n\n\ngamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ngamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_p:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_q:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ngaussian_dlm_obs:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lpdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lupdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngeneralized_inverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nget_imag:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\nget_real:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\ngp_dot_prod_cov:\n\n\n(array[] real x, real sigma) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma) : matrix (matrix_operations.html)\n\n\ngp_exp_quad_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\ngp_exponential_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern23_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern52_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_periodic_cov:\n\n\n(array[] real x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\ngumbel:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ngumbel_cdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lccdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lcdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lpdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lupdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_rng:\n\n\n(reals mu, reals beta) : R (unbounded_continuous_distributions.html)\n\n\n\n\n\nhead:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\nhmm_hidden_state_prob:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : matrix (hidden_markov_models.html)\n\n\nhmm_latent_rng:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : array[] int (hidden_markov_models.html)\n\n\nhmm_marginal:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : real (hidden_markov_models.html)\n\n\nhypergeometric:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nhypergeometric_1F0:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_2F1:\n\n\n(real a1, real a2, real b1, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_3F2:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_lpmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_lupmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_pFq:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_rng:\n\n\n(int N, int a, int2 b) : int (bounded_discrete_distributions.html)\n\n\nhypot:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n\n\n\nidentity_matrix:\n\n\n(int k) : matrix (matrix_operations.html)\n\n\ninc_beta:\n\n\n(real alpha, real beta, real x) : real (real-valued_basic_functions.html)\n\n\nint_step:\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\nintegrate_1d:\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real (higher-order_functions.html)\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real (higher-order_functions.html)\n\n\nintegrate_ode:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_adams:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_bdf:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_rk45:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\ninv:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_chi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\ninv_cloglog:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_erfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_fft:\n\n\n(complex_vector u) : complex_vector (complex_matrix_operations.html)\n\n\ninv_fft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\ninv_gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_gamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ninv_inc_beta:\n\n\n(real alpha, real beta, real p) : real (real-valued_basic_functions.html)\n\n\ninv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_Phi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_sqrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_square:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_wishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\ninv_wishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\ninverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ninverse_spd:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nis_inf:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\nis_nan:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\n\n\n\nlambert_w0:\n\n\n(reals x) : R (real-valued_basic_functions.html)\n\n\nlambert_wm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlchoose:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nldexp:\n\n\n(real x, int y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlinspaced_array:\n\n\n(int n, data real lower, data real upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_int_array:\n\n\n(int n, int lower, int upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_row_vector:\n\n\n(int n, data real lower, data real upper) : row_vector (matrix_operations.html)\n\n\nlinspaced_vector:\n\n\n(int n, data real lower, data real upper) : vector (matrix_operations.html)\n\n\nlkj_corr:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lpdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lupdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlkj_corr_lpdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_lupdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlmgamma:\n\n\n(int n, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlmultiply:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog10:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_determinant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\nlog_diff_exp:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\nlog_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_inv_logit_diff:\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_mix:\n\n\n(real theta, real lp1, real lp2) : real (real-valued_basic_functions.html)\n\n\n(T1 thetas, T2 lps) : R (real-valued_basic_functions.html)\n\n\nlog_modified_bessel_first_kind:\n\n\n(real v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_rising_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_softmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nlog_sum_exp:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nlogistic:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nlogistic_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nlogit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nloglogistic:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nloglogistic_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlognormal:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nlognormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlower_bound_constrain:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_jacobian:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_unconstrain:\n\n\n(reals x, reals lb) : reals (transform_functions.html)\n\n\nlower_upper_bound_constrain:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_jacobian:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_unconstrain:\n\n\n(reals x, reals lb, reals ub) : reals (transform_functions.html)\n\n\n\n\n\nmachine_precision:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nmap_rect:\n\n\n(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector (higher-order_functions.html)\n\n\nmatrix_exp:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nmatrix_exp_multiply:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nmatrix_power:\n\n\n(matrix A, int B) : matrix (matrix_operations.html)\n\n\nmax:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmdivide_left_spd:\n\n\n(matrix A, matrix B) : vector (matrix_operations.html)\n\n\n(matrix A, vector b) : matrix (matrix_operations.html)\n\n\nmdivide_left_tri_low:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\nmdivide_right_spd:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmdivide_right_tri_low:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmean:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmin:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmodified_bessel_first_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmodified_bessel_second_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmulti_gp:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lpdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lupdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lpdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lupdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_rng:\n\n\n(row_vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_rng:\n\n\n(row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_cholesky_t_rng:\n\n\n(real nu, vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lpdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lupdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_rng:\n\n\n(real nu, array[] row_vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, array[] vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lpdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lupdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_rng:\n\n\n(real nu, row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmultinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lpmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lupmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_rng:\n\n\n(vector gamma, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultinomial_lpmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_lupmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_rng:\n\n\n(vector theta, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultiply_lower_tri_self_transpose:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n\n\n\nneg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_cdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lccdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lcdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lpmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lupmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_rng:\n\n\n(reals eta, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lpmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lupmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_rng:\n\n\n(reals mu, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_cdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lccdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lcdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lpmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lupmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_rng:\n\n\n(reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nnegative_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnorm:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nnorm1:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnorm2:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnormal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lpdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lupdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nnot_a_number:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnum_elements:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\n\n\n\node_adams:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_adams_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_adjoint_tol_ctl:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector (higher-order_functions.html)\n\n\node_bdf:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_bdf_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\noffset_multiplier_constrain:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_jacobian:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_unconstrain:\n\n\n(reals x, reals offset, reals mult) : reals (transform_functions.html)\n\n\none_hot_array:\n\n\n(int n, int k) : array[] real (matrix_operations.html)\n\n\none_hot_int_array:\n\n\n(int n, int k) : array[] int (matrix_operations.html)\n\n\none_hot_row_vector:\n\n\n(int n, int k) : row_vector (matrix_operations.html)\n\n\none_hot_vector:\n\n\n(int K, int k) : vector (matrix_operations.html)\n\n\nones_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nones_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nones_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nones_vector:\n\n\n(int n) : vector (matrix_operations.html)\n\n\noperator!:\n\n\n(int x) : int (real-valued_basic_functions.html)\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\noperator!=:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator%/%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator&&:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\n&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;operator’:\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\noperator*:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : vector (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, row_vector y) : matrix (matrix_operations.html)\n\n\noperator*=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator+:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator+=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator-:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(T x) : T (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (matrix_operations.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator-=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.*:\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator.*=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator./:\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator./=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.^:\n\n\n( complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n( complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator/:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix B, complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector b, complex_matrix A) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\noperator/=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator&lt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&lt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\noperator==:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator\\:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\noperator^:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\noperator||:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\nordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_logistic:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lpmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lupmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lpmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lupmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_probit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_probit_lpmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_lupmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nowens_t:\n\n\n(real h, real a) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n\n\n\npareto:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_cdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lccdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lcdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lpdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lupdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_rng:\n\n\n(reals y_min, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\npareto_type_2:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_cdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lccdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lcdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lpdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lupdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_rng:\n\n\n(reals mu, reals lambda, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\nPhi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nPhi_approx:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\npi:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npoisson:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_cdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lccdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lcdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\npoisson_lpmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lupmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_rng:\n\n\n(reals lambda) : R (unbounded_discrete_distributions.html)\n\n\npolar:\n\n\n(real r, real theta) : complex (complex-valued_basic_functions.html)\n\n\npositive_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npositive_ordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\npow:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : Z (complex-valued_basic_functions.html)\n\n\nprint:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nprod:\n\n\n(array[] int x) : real (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nproj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n\n\n\nqr:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_thin_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nquad_form:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquad_form_diag:\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\nquad_form_sym:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquantile:\n\n\n(data array[] real x, data array[] real p) : array[] real (array_operations.html)\n\n\n(data array[] real x, data real p) : real (array_operations.html)\n\n\n(data row_vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data row_vector x, data real p) : real (matrix_operations.html)\n\n\n(data vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data vector x, data real p) : real (matrix_operations.html)\n\n\n\n\n\nrank:\n\n\n(array[] int v, int s) : int (array_operations.html)\n\n\n(array[] real v, int s) : int (array_operations.html)\n\n\n(row_vector v, int s) : int (matrix_operations.html)\n\n\n(vector v, int s) : int (matrix_operations.html)\n\n\nrayleigh:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nrayleigh_cdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lccdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lcdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lpdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lupdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_rng:\n\n\n(reals sigma) : R (positive_continuous_distributions.html)\n\n\nreduce_sum:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreduce_sum_static:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreject:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nrep_array:\n\n\n(T x, int k, int m, int n) : array[,,] T (array_operations.html)\n\n\n(T x, int m, int n) : array[,] T (array_operations.html)\n\n\n(T x, int n) : array[] T (array_operations.html)\n\n\nrep_matrix:\n\n\n(complex z, int m, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector rv, int m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(real x, int m, int n) : matrix (matrix_operations.html)\n\n\n(row_vector rv, int m) : matrix (matrix_operations.html)\n\n\n(vector v, int n) : matrix (matrix_operations.html)\n\n\nrep_row_vector:\n\n\n(complex z, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(real x, int n) : row_vector (matrix_operations.html)\n\n\nrep_vector:\n\n\n(complex z, int m) : complex_vector (complex_matrix_operations.html)\n\n\n(real x, int m) : vector (matrix_operations.html)\n\n\nreverse:\n\n\n(array[] T v) : array[] T (array_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nrising_factorial:\n\n\n(real x, int n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nround:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nrow:\n\n\n(complex_matrix x, int m) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int m) : row_vector (matrix_operations.html)\n\n\nrows:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nrows_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\nrows_dot_self:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n\n\n\nscale_matrix_exp_multiply:\n\n\n(real t, matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nscaled_inv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_cdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lccdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lcdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lpdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lupdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_rng:\n\n\n(reals nu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nsd:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsegment:\n\n\n(array[] T sv, int i, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int i, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int i, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int i, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int i, int n) : vector (matrix_operations.html)\n\n\nsimplex_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsingular_values:\n\n\n(complex_matrix A) : vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\nsinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsize:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nskew_double_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nskew_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_normal_cdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lccdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lcdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lpdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lupdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_rng:\n\n\n(reals xi, reals omega, real alpha) : R (unbounded_continuous_distributions.html)\n\n\nsoftmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nsolve_newton:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_newton_tol:\n\n\n(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell_tol:\n\n\n(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsort_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_indices_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsort_indices_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsqrt:\n\n\n(complex x) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsqrt2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nsquare:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsquared_distance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\nstd_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstd_normal_cdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lccdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lcdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_log_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_lpdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lupdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_rng:\n\n\n() : real (unbounded_continuous_distributions.html)\n\n\nstep:\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\nstochastic_column_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstochastic_row_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstudent_t:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstudent_t_cdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lccdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lcdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lpdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lupdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_rng:\n\n\n(reals nu, reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nsub_col:\n\n\n(complex_matrix x, int i, int j, int n_rows) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows) : vector (matrix_operations.html)\n\n\nsub_row:\n\n\n(complex_matrix x, int i, int j, int n_cols) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_cols) : row_vector (matrix_operations.html)\n\n\nsum:\n\n\n(array[] complex x) : complex (array_operations.html)\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsum_to_zero_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsvd:\n\n\n(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector, matrix) (matrix_operations.html)\n\n\nsvd_U:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsvd_V:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsymmetrize_from_lower_tri:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\n\n\n\ntail:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\ntan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntarget:\n\n\n() : real (real-valued_basic_functions.html)\n\n\ntcrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ntgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nto_array_1d:\n\n\n(array[...] complex a) : array[] complex (mixed_operations.html)\n\n\n(array[...] int a) : array[] int (mixed_operations.html)\n\n\n(array[...] real a) : array[] real (mixed_operations.html)\n\n\n(complex_matrix m) : array[] complex (mixed_operations.html)\n\n\n(complex_row_vector v) : array[] complex (mixed_operations.html)\n\n\n(complex_vector v) : array[] real (mixed_operations.html)\n\n\n(matrix m) : array[] real (mixed_operations.html)\n\n\n(row_vector v) : array[] real (mixed_operations.html)\n\n\n(vector v) : array[] real (mixed_operations.html)\n\n\nto_array_2d:\n\n\n(complex_matrix m) : array[,] real (mixed_operations.html)\n\n\n(matrix m) : array[,] real (mixed_operations.html)\n\n\nto_complex:\n\n\n() : complex (complex-valued_basic_functions.html)\n\n\n(real re) : complex (complex-valued_basic_functions.html)\n\n\n(real re, real im) : complex (complex-valued_basic_functions.html)\n\n\n(T1 re, T2 im) : Z (complex-valued_basic_functions.html)\n\n\nto_int:\n\n\n(data real x) : int (integer-valued_basic_functions.html)\n\n\nto_matrix:\n\n\n(array[,] complex a ) : complex_matrix (mixed_operations.html)\n\n\n(array[,] int a) : matrix (mixed_operations.html)\n\n\n(array[,] real a) : matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex_row_vector vs) : complex_matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] row_vector vs) : matrix (mixed_operations.html)\n\n\n(complex_matrix A, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix m) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix M, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(matrix A, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(matrix m) : matrix (mixed_operations.html)\n\n\n(matrix M, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(vector v) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\nto_row_vector:\n\n\n(array[] complex a) : complex_row_vector (mixed_operations.html)\n\n\n(array[] int a) : row_vector (mixed_operations.html)\n\n\n(array[] real a) : row_vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_row_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(matrix m) : row_vector (mixed_operations.html)\n\n\n(row_vector v) : row_vector (mixed_operations.html)\n\n\n(vector v) : row_vector (mixed_operations.html)\n\n\nto_vector:\n\n\n(array[] complex a) : complex_vector (mixed_operations.html)\n\n\n(array[] int a) : vector (mixed_operations.html)\n\n\n(array[] real a) : vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_vector (mixed_operations.html)\n\n\n(matrix m) : vector (mixed_operations.html)\n\n\n(row_vector v) : vector (mixed_operations.html)\n\n\n(vector v) : vector (mixed_operations.html)\n\n\ntrace:\n\n\n(complex_matrix A) : complex (complex_matrix_operations.html)\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ntrace_gen_quad_form:\n\n\n(matrix D ,matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_quad_form:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\ntrigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntrunc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\n\n\n\nuniform:\n\n\ndistribution statement (bounded_continuous_distributions.html)\n\n\nuniform_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_rng:\n\n\n(reals alpha, reals beta) : R (bounded_continuous_distributions.html)\n\n\nuniform_simplex:\n\n\n(int n) : vector (matrix_operations.html)\n\n\nunit_vectors_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nupper_bound_constrain:\n\n\n(reals y, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_jacobian:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_unconstrain:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\n\n\n\nvariance:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nvon_mises:\n\n\ndistribution statement (circular_distributions.html)\n\n\nvon_mises_cdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lccdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lcdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lpdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lupdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_rng:\n\n\n(reals mu, reals kappa) : R (circular_distributions.html)\n\n\n\n\n\nweibull:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nweibull_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\nwiener:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\nwiener_lccdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lcdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lpdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lupdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\nwishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\n\n\n\nzeros_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nzeros_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nzeros_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nzeros_vector:\n\n\n(int n) : vector (matrix_operations.html)",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{DoubleExponential}(y|\\mu,\\sigma) =\n\\frac{1}{2\\sigma}   \\exp \\left( - \\, \\frac{|y - \\mu|}{\\sigma} \\right)\n. \\end{equation*}\\] Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section exponential distribution), which is parameterized in terms of inverse scale.\nThe double-exponential distribution can be defined as a compound exponential-normal distribution (Ding and Blitzstein 2018). Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write \\[\\begin{equation*} \\alpha \\sim\n\\mathsf{Exponential}\\left( \\frac{1}{2 \\sigma^2} \\right) \\end{equation*}\\] and \\[\\begin{equation*} \\beta \\mid\n\\alpha \\sim \\mathsf{Normal}(\\mu, \\sqrt{\\alpha}), \\end{equation*}\\] then \\[\\begin{equation*} \\beta \\sim\n\\mathsf{DoubleExponential}(\\mu, \\sigma ). \\end{equation*}\\] This may be used to code a non-centered parameterization by taking \\[\\begin{equation*} \\beta^{\\text{raw}} \\sim\n\\mathsf{Normal}(0, 1) \\end{equation*}\\] and defining \\[\\begin{equation*} \\beta = \\mu + \\sqrt{\\alpha} \\,\n\\beta^{\\text{raw}}. \\end{equation*}\\]\n\n\n\ny ~ double_exponential(mu, sigma)\nIncrement target log probability density with double_exponential_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal double_exponential_lpdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lupdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal double_exponential_cdf(reals y | reals mu, reals sigma) The double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal double_exponential_lcdf(reals y | reals mu, reals sigma) The log of the double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lccdf(reals y | reals mu, reals sigma) The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR double_exponential_rng(reals mu, reals sigma) Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#a",
-    "href": "functions-reference/functions_index.html#a",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#logistic-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#logistic-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "abs:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\nacos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nacosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nadd_diag:\n\n\n(complex_matrix m, complex_real d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_row_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, real d) : matrix (matrix_operations.html)\n\n\n(matrix m, row_vector d) : matrix (matrix_operations.html)\n\n\n(matrix m, vector d) : matrix (matrix_operations.html)\n\n\nalgebra_solver:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nalgebra_solver_newton:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector (deprecated_functions.html)\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nappend_array:\n\n\n(T x, T y) : T (array_operations.html)\n\n\nappend_col:\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(vector x, vector y) : matrix (matrix_operations.html)\n\n\nappend_row:\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, row_vector y) : matrix (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : matrix (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\narg:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nasin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nasinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan2:\n\n\n(T y, T x) : R (real-valued_basic_functions.html)\n\n\natanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Logistic}(y|\\mu,\\sigma) = \\frac{1}{\\sigma} \\ \\exp\\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right)\n\\ \\left(1 + \\exp \\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right) \\right)^{\\!-2} \\! .\n\\end{equation*}\\]\n\n\n\ny ~ logistic(mu, sigma)\nIncrement target log probability density with logistic_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal logistic_lpdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lupdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal logistic_cdf(reals y | reals mu, reals sigma) The logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal logistic_lcdf(reals y | reals mu, reals sigma) The log of the logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lccdf(reals y | reals mu, reals sigma) The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR logistic_rng(reals mu, reals sigma) Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#b",
-    "href": "functions-reference/functions_index.html#b",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#gumbel-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#gumbel-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "bernoulli:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_cdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lccdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lcdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_logit:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_rng:\n\n\n(matrix x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\n(row_vector x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\nbernoulli_logit_lpmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_lupmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_rng:\n\n\n(reals alpha) : R (binary_distributions.html)\n\n\nbernoulli_lpmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lupmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_rng:\n\n\n(reals theta) : R (binary_distributions.html)\n\n\nbessel_first_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbessel_second_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_binomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbeta_binomial_cdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lccdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lcdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lpmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lupmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_rng:\n\n\n(ints N, reals alpha, reals beta) : R (bounded_discrete_distributions.html)\n\n\nbeta_cdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lccdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lcdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lpdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lupdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_cdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lccdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lcdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lpmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lupmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_rng:\n\n\n(reals r, reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nbeta_proportion:\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lccdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lcdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lpdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lupdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_rng:\n\n\n(reals mu, reals kappa) : R (continuous_distributions_on_0_1.html)\n\n\nbeta_rng:\n\n\n(reals alpha, reals beta) : R (continuous_distributions_on_0_1.html)\n\n\nbinary_log_loss:\n\n\n(int y, real y_hat) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbinomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_cdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lccdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lcdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lpmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lupmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lpmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lupmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lpmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lupmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_rng:\n\n\n(ints N, reals theta) : R (bounded_discrete_distributions.html)\n\n\nblock:\n\n\n(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows, int n_cols) : matrix (matrix_operations.html)",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Gumbel}(y|\\mu,\\beta) = \\frac{1}{\\beta} \\ \\exp\\left(-\\frac{y-\\mu}{\\beta}-\\exp\\left(-\\frac{y-\\mu}{\\beta}\\right)\\right) .\n\\end{equation*}\\]\n\n\n\ny ~ gumbel(mu, beta)\nIncrement target log probability density with gumbel_lupdf(y | mu, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gumbel_lpdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lupdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gumbel_cdf(reals y | reals mu, reals beta) The gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.0\n \n\nreal gumbel_lcdf(reals y | reals mu, reals beta) The log of the gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lccdf(reals y | reals mu, reals beta) The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nR gumbel_rng(reals mu, reals beta) Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#c",
-    "href": "functions-reference/functions_index.html#c",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/unbounded_continuous_distributions.html#skew-double-exponential-distribution",
+    "href": "functions-reference/unbounded_continuous_distributions.html#skew-double-exponential-distribution",
+    "title": "Unbounded Continuous Distributions",
     "section": "",
-    "text": "categorical:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lpmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lupmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_rng:\n\n\n(vector beta) : int (bounded_discrete_distributions.html)\n\n\ncategorical_lpmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_lupmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_rng:\n\n\n(vector theta) : int (bounded_discrete_distributions.html)\n\n\ncauchy:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ncauchy_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\ncbrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nceil:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nchi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nchi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\nchol2inv:\n\n\n(matrix L) : matrix (matrix_operations.html)\n\n\ncholesky_decompose:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ncholesky_factor_corr_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncholesky_factor_cov_constrain:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_jacobian:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\nchoose:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ncol:\n\n\n(complex_matrix x, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int n) : vector (matrix_operations.html)\n\n\ncols:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\ncolumns_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : row_vector (matrix_operations.html)\n\n\ncolumns_dot_self:\n\n\n(complex_matrix x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : row_vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\ncomplex_schur_decompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_t:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_u:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\nconj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(Z z) : Z (complex-valued_basic_functions.html)\n\n\ncorr_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncov_exp_quad:\n\n\n(array[] real x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(array[] real x1, array[] real x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x1, vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\ncov_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ncsr_extract:\n\n\n(matrix a) : tuple(vector, array[] int, array[] int) (sparse_matrix_operations.html)\n\n\ncsr_extract_u:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_v:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_w:\n\n\n(matrix a) : vector (sparse_matrix_operations.html)\n\n\ncsr_matrix_times_vector:\n\n\n(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector (sparse_matrix_operations.html)\n\n\ncsr_to_dense_matrix:\n\n\n(int m, int n, vector w, array[] int v, array[] int u) : matrix (sparse_matrix_operations.html)\n\n\ncumulative_sum:\n\n\n(array[] complex x) : array[] complex (complex_matrix_operations.html)\n\n\n(array[] int x) : array[] int (matrix_operations.html)\n\n\n(array[] real x) : array[] real (matrix_operations.html)\n\n\n(complex_row_vector rv) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\) and \\(\\tau \\in [0, 1]\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{aligned}\n& \\text{SkewDoubleExponential}  (y|\\mu,\\sigma, \\tau) = \\\\\n& \\qquad \\qquad \\frac{2 \\tau (1 - \\tau) }{\\sigma} \\exp \\left[ - \\frac{2}{\\sigma} \\left[ \\left(1 - \\tau \\right) I(y &lt; \\mu)  (\\mu - y) + \\tau I(y &gt; \\mu)(y-\\mu)  \\right]    \\right]\n\\end{aligned}\\]\n\n\n\ny ~ skew_double_exponential(mu, sigma, tau)\nIncrement target log probability density with skew_double_exponential(y | mu, sigma, tau)\nAvailable since 2.28\n \n\n\n\n\n \n\nreal skew_double_exponential_lpdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lupdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms\nAvailable since 2.28\n \n\nreal skew_double_exponential_cdf(reals y | reals mu, reals sigma, reals tau) The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lcdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lccdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nR skew_double_exponential_rng(reals mu, reals sigma, reals tau) Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Continuous Distributions",
+      "Unbounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#d",
-    "href": "functions-reference/functions_index.html#d",
-    "title": "Alphabetical Index",
-    "section": "",
-    "text": "dae:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\ndae_tol:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\ndeterminant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ndiag_matrix:\n\n\n(complex_vector x) : complex_matrix (complex_matrix_operations.html)\n\n\n(vector x) : matrix (matrix_operations.html)\n\n\ndiag_post_multiply:\n\n\n(complex_matrix m, complex_row_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\ndiag_pre_multiply:\n\n\n(complex_row_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(row_vector rv, matrix m) : matrix (matrix_operations.html)\n\n\n(vector v, matrix m) : matrix (matrix_operations.html)\n\n\ndiagonal:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\ndigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ndims:\n\n\n(T x) : array[] int (array_operations.html)\n\n\ndirichlet:\n\n\ndistribution statement (simplex_distributions.html)\n\n\ndirichlet_lpdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_lupdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_multinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lpmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lupmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_rng:\n\n\n(vector alpha, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\ndirichlet_rng:\n\n\n(vector alpha) : vector (simplex_distributions.html)\n\n\ndiscrete_range:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ndiscrete_range_cdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lccdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lcdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lpmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lupmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_rng:\n\n\n(ints l, ints u) : ints (bounded_discrete_distributions.html)\n\n\ndistance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\ndot_product:\n\n\n(complex_row_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(row_vector x, row_vector y) : real (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, row_vector y) : real (matrix_operations.html)\n\n\n(vector x, vector y) : real (matrix_operations.html)\n\n\ndot_self:\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\ndouble_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)",
+    "objectID": "functions-reference/sparse_matrix_operations.html",
+    "href": "functions-reference/sparse_matrix_operations.html",
+    "title": "Sparse Matrix Operations",
+    "section": "",
+    "text": "For sparse matrices, for which many elements are zero, it is more efficient to use specialized representations to save memory and speed up matrix arithmetic (including derivative calculations). Given Stan’s implementation, there is substantial space (memory) savings by using sparse matrices. Because of the ease of optimizing dense matrix operations, speed improvements only arise at 90% or even greater sparsity; below that level, dense matrices are faster but use more memory.\nBecause of this speedup and space savings, it may even be useful to read in a dense matrix and convert it to a sparse matrix before multiplying it by a vector. This chapter covers a very specific form of sparsity consisting of a sparse matrix multiplied by a dense vector.\n\n\nSparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix \\[\\begin{equation*}\nA = \\begin{bmatrix} 19 & 27 & 0 & 0 \\\\ 0 & 0 & 0 & 0 \\\\ 0 & 0 & 0 & 52 \\\\ 81 & 0 & 95 & 33 \\end{bmatrix}\n\\end{equation*}\\] is translated into a vector of the non-zero real values, read by row from the matrix \\(A\\), \\[\\begin{equation*}\nw(A) = \\begin{bmatrix} 19 & 27 & 52 & 81 & 95 & 33 \\end{bmatrix}^{\\top} \\! \\! \\! ,\n\\end{equation*}\\] an array of integer column indices for the values, \\[\\begin{equation*}\nv(A) = \\begin{bmatrix} 1 & 2 & 4 & 1 & 3 & 4 \\end{bmatrix} \\! ,\n\\end{equation*}\\] and an array of integer indices indicating where in \\(w(A)\\) a given row’s values start, \\[\\begin{equation*} u(A) = \\begin{bmatrix} 1 & 3 & 3 & 4 & 7 \\end{bmatrix} \\! ,\n\\end{equation*}\\] with a padded value at the end to guarantee that \\[\\begin{equation*} u(A)[n+1] - u(A)[n] \\end{equation*}\\] is the number of non-zero elements in row \\(n\\) of the matrix (here \\(2\\), \\(0\\), \\(1\\), and \\(3\\)). Note that because the second row has no non-zero elements both the second and third elements of \\(u(A)\\) correspond to the third element of \\(w(A)\\), which is \\(52\\). The values \\((w(A), \\, v(A), \\, u(A))\\) are sufficient to reconstruct \\(A\\).\nThe values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly \\(12 K + M\\) bytes plus a small constant overhead, which is often considerably fewer bytes than the \\(M \\times\nN\\) required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.\n\n\n\nConversion functions between dense and sparse matrices are provided.\n\n\nConverting a dense matrix \\(m\\) to a sparse representation produces a vector \\(w\\) and two integer arrays, \\(u\\) and \\(v\\).\n \n\nvector csr_extract_w(matrix a) Return non-zero values in matrix a; see section compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_v(matrix a) Return column indices for values in csr_extract_w(a); see compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_u(matrix a) Return array of row starting indices for entries in csr_extract_w(a) followed by the size of csr_extract_w(a) plus one; see section compressed row storage.\nAvailable since 2.8\n \n\ntuple(vector, array[] int, array[] int) csr_extract(matrix a) Return all three components of the CSR representation of the matrix a; see section compressed row storage. This function is equivalent to (csr_extract_w(a), csr_extract_v(a), csr_extract_u(a)).\nAvailable since 2.33\n\n\n\nTo convert a sparse matrix representation to a dense matrix, there is a single function.\n \n\nmatrix csr_to_dense_matrix(int m, int n, vector w, array[] int v, array[] int u) Return dense \\(\\text{m} \\times \\text{n}\\) matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section compressed row storage for more details.\nAvailable since 2.10\n\n\n\n\n\n\nThe only supported operation is the multiplication of a sparse matrix \\(A\\) and a dense vector \\(b\\) to produce a dense vector \\(A\\,b\\). Multiplying a dense row vector \\(b\\) and a sparse matrix \\(A\\) can be coded using transposition as \\[\\begin{equation*}\nb \\, A = (A^{\\top} \\, b^{\\top})^{\\top},\n\\end{equation*}\\] but care must be taken to represent \\(A^{\\top}\\) rather than \\(A\\) as a sparse matrix.\n \n\nvector csr_matrix_times_vector(int m, int n, vector w, array[] int v, array[] int u, vector b) Multiply the \\(\\text{m} \\times \\text{n}\\) matrix represented by values w, column indices v, and row start indices u by the vector b; see compressed row storage.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Sparse Matrix Operations"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#e",
-    "href": "functions-reference/functions_index.html#e",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/sparse_matrix_operations.html#CSR",
+    "href": "functions-reference/sparse_matrix_operations.html#CSR",
+    "title": "Sparse Matrix Operations",
     "section": "",
-    "text": "e:\n\n\n() : real (real-valued_basic_functions.html)\n\n\neigendecompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_vector) (matrix_operations.html)\n\n\neigendecompose_sym:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector) (matrix_operations.html)\n\n\neigenvalues:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : complex_vector (matrix_operations.html)\n\n\neigenvalues_sym:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\neigenvectors:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (matrix_operations.html)\n\n\neigenvectors_sym:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nerf:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nerfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp2:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp_mod_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_cdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_rng:\n\n\n(reals mu, reals sigma, reals lambda) : R (unbounded_continuous_distributions.html)\n\n\nexpm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexponential:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nexponential_cdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lccdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lcdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lpdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lupdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_rng:\n\n\n(reals beta) : R (positive_continuous_distributions.html)",
+    "text": "Sparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix \\[\\begin{equation*}\nA = \\begin{bmatrix} 19 & 27 & 0 & 0 \\\\ 0 & 0 & 0 & 0 \\\\ 0 & 0 & 0 & 52 \\\\ 81 & 0 & 95 & 33 \\end{bmatrix}\n\\end{equation*}\\] is translated into a vector of the non-zero real values, read by row from the matrix \\(A\\), \\[\\begin{equation*}\nw(A) = \\begin{bmatrix} 19 & 27 & 52 & 81 & 95 & 33 \\end{bmatrix}^{\\top} \\! \\! \\! ,\n\\end{equation*}\\] an array of integer column indices for the values, \\[\\begin{equation*}\nv(A) = \\begin{bmatrix} 1 & 2 & 4 & 1 & 3 & 4 \\end{bmatrix} \\! ,\n\\end{equation*}\\] and an array of integer indices indicating where in \\(w(A)\\) a given row’s values start, \\[\\begin{equation*} u(A) = \\begin{bmatrix} 1 & 3 & 3 & 4 & 7 \\end{bmatrix} \\! ,\n\\end{equation*}\\] with a padded value at the end to guarantee that \\[\\begin{equation*} u(A)[n+1] - u(A)[n] \\end{equation*}\\] is the number of non-zero elements in row \\(n\\) of the matrix (here \\(2\\), \\(0\\), \\(1\\), and \\(3\\)). Note that because the second row has no non-zero elements both the second and third elements of \\(u(A)\\) correspond to the third element of \\(w(A)\\), which is \\(52\\). The values \\((w(A), \\, v(A), \\, u(A))\\) are sufficient to reconstruct \\(A\\).\nThe values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly \\(12 K + M\\) bytes plus a small constant overhead, which is often considerably fewer bytes than the \\(M \\times\nN\\) required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Sparse Matrix Operations"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#f",
-    "href": "functions-reference/functions_index.html#f",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/sparse_matrix_operations.html#conversion-functions",
+    "href": "functions-reference/sparse_matrix_operations.html#conversion-functions",
+    "title": "Sparse Matrix Operations",
     "section": "",
-    "text": "falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfatal_error:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nfdim:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfft:\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\nfft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\nfloor:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nfma:\n\n\n(real x, real y, real z) : real (real-valued_basic_functions.html)\n\n\nfmax:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmin:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmod:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfrechet:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nfrechet_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)",
+    "text": "Conversion functions between dense and sparse matrices are provided.\n\n\nConverting a dense matrix \\(m\\) to a sparse representation produces a vector \\(w\\) and two integer arrays, \\(u\\) and \\(v\\).\n \n\nvector csr_extract_w(matrix a) Return non-zero values in matrix a; see section compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_v(matrix a) Return column indices for values in csr_extract_w(a); see compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_u(matrix a) Return array of row starting indices for entries in csr_extract_w(a) followed by the size of csr_extract_w(a) plus one; see section compressed row storage.\nAvailable since 2.8\n \n\ntuple(vector, array[] int, array[] int) csr_extract(matrix a) Return all three components of the CSR representation of the matrix a; see section compressed row storage. This function is equivalent to (csr_extract_w(a), csr_extract_v(a), csr_extract_u(a)).\nAvailable since 2.33\n\n\n\nTo convert a sparse matrix representation to a dense matrix, there is a single function.\n \n\nmatrix csr_to_dense_matrix(int m, int n, vector w, array[] int v, array[] int u) Return dense \\(\\text{m} \\times \\text{n}\\) matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section compressed row storage for more details.\nAvailable since 2.10",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Sparse Matrix Operations"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#g",
-    "href": "functions-reference/functions_index.html#g",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/sparse_matrix_operations.html#sparse-matrix-arithmetic",
+    "href": "functions-reference/sparse_matrix_operations.html#sparse-matrix-arithmetic",
+    "title": "Sparse Matrix Operations",
     "section": "",
-    "text": "gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ngamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_p:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_q:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ngaussian_dlm_obs:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lpdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lupdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngeneralized_inverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nget_imag:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\nget_real:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\ngp_dot_prod_cov:\n\n\n(array[] real x, real sigma) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma) : matrix (matrix_operations.html)\n\n\ngp_exp_quad_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\ngp_exponential_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern23_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern52_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_periodic_cov:\n\n\n(array[] real x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\ngumbel:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ngumbel_cdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lccdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lcdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lpdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lupdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_rng:\n\n\n(reals mu, reals beta) : R (unbounded_continuous_distributions.html)",
+    "text": "The only supported operation is the multiplication of a sparse matrix \\(A\\) and a dense vector \\(b\\) to produce a dense vector \\(A\\,b\\). Multiplying a dense row vector \\(b\\) and a sparse matrix \\(A\\) can be coded using transposition as \\[\\begin{equation*}\nb \\, A = (A^{\\top} \\, b^{\\top})^{\\top},\n\\end{equation*}\\] but care must be taken to represent \\(A^{\\top}\\) rather than \\(A\\) as a sparse matrix.\n \n\nvector csr_matrix_times_vector(int m, int n, vector w, array[] int v, array[] int u, vector b) Multiply the \\(\\text{m} \\times \\text{n}\\) matrix represented by values w, column indices v, and row start indices u by the vector b; see compressed row storage.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Sparse Matrix Operations"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#h",
-    "href": "functions-reference/functions_index.html#h",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/removed_functions.html",
+    "href": "functions-reference/removed_functions.html",
+    "title": "Removed Functions",
     "section": "",
-    "text": "head:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\nhmm_hidden_state_prob:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : matrix (hidden_markov_models.html)\n\n\nhmm_latent_rng:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : array[] int (hidden_markov_models.html)\n\n\nhmm_marginal:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : real (hidden_markov_models.html)\n\n\nhypergeometric:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nhypergeometric_1F0:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_2F1:\n\n\n(real a1, real a2, real b1, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_3F2:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_lpmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_lupmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_pFq:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_rng:\n\n\n(int N, int a, int2 b) : int (bounded_discrete_distributions.html)\n\n\nhypot:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)",
+    "text": "Functions which once existed in the Stan language and have since been replaced or removed are listed here.\n\n\nRemoved: Currently two non-conforming functions ending in suffix _log.\nReplacement: Replace multiply_log(...) with lmultiply(...). Replace binomial_coefficient_log(...) with lchoose(...).\nRemoved In: Stan 2.33\n\n\n\nRemoved: The built-in no-argument function get_lp() is deprecated.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33\n\n\n\nRemoved: The unary function fabs is deprecated.\nReplacement: Use the unary function abs instead. Note that the return type for abs is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.\nRemoved In: Stan 2.33\n\n\n\nThese covariance functions have been replaced by those described in Gaussian Process Covariance Functions\nWith magnitude \\(\\alpha\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(x_i, x_j) = \\alpha^2 \\exp \\left(-\\dfrac{1}{2\\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \\right)\n\\]\n \n\nmatrix cov_exp_quad(row_vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(row_vectors x1, row_vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x1, vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x1, array[] real x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n\n\n\nRemoved: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.\nReplacement: Explicit x != 0 comparison is preferred instead.\nRemoved In: Stan 2.34",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Removed Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#i",
-    "href": "functions-reference/functions_index.html#i",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/removed_functions.html#multiply_log-and-binomial_coefficient_log-functions",
+    "href": "functions-reference/removed_functions.html#multiply_log-and-binomial_coefficient_log-functions",
+    "title": "Removed Functions",
     "section": "",
-    "text": "identity_matrix:\n\n\n(int k) : matrix (matrix_operations.html)\n\n\ninc_beta:\n\n\n(real alpha, real beta, real x) : real (real-valued_basic_functions.html)\n\n\nint_step:\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\nintegrate_1d:\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real (higher-order_functions.html)\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real (higher-order_functions.html)\n\n\nintegrate_ode:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_adams:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_bdf:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_rk45:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\ninv:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_chi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\ninv_cloglog:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_erfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_fft:\n\n\n(complex_vector u) : complex_vector (complex_matrix_operations.html)\n\n\ninv_fft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\ninv_gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_gamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ninv_inc_beta:\n\n\n(real alpha, real beta, real p) : real (real-valued_basic_functions.html)\n\n\ninv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_Phi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_sqrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_square:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_wishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\ninv_wishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\ninverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ninverse_spd:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nis_inf:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\nis_nan:\n\n\n(real x) : int (real-valued_basic_functions.html)",
+    "text": "Removed: Currently two non-conforming functions ending in suffix _log.\nReplacement: Replace multiply_log(...) with lmultiply(...). Replace binomial_coefficient_log(...) with lchoose(...).\nRemoved In: Stan 2.33",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Removed Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#l",
-    "href": "functions-reference/functions_index.html#l",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/removed_functions.html#get_lp-function",
+    "href": "functions-reference/removed_functions.html#get_lp-function",
+    "title": "Removed Functions",
     "section": "",
-    "text": "lambert_w0:\n\n\n(reals x) : R (real-valued_basic_functions.html)\n\n\nlambert_wm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlchoose:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nldexp:\n\n\n(real x, int y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlinspaced_array:\n\n\n(int n, data real lower, data real upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_int_array:\n\n\n(int n, int lower, int upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_row_vector:\n\n\n(int n, data real lower, data real upper) : row_vector (matrix_operations.html)\n\n\nlinspaced_vector:\n\n\n(int n, data real lower, data real upper) : vector (matrix_operations.html)\n\n\nlkj_corr:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lpdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lupdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlkj_corr_lpdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_lupdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlmgamma:\n\n\n(int n, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlmultiply:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog10:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_determinant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\nlog_diff_exp:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\nlog_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_inv_logit_diff:\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_mix:\n\n\n(real theta, real lp1, real lp2) : real (real-valued_basic_functions.html)\n\n\n(T1 thetas, T2 lps) : R (real-valued_basic_functions.html)\n\n\nlog_modified_bessel_first_kind:\n\n\n(real v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_rising_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_softmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nlog_sum_exp:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nlogistic:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nlogistic_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nlogit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nloglogistic:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nloglogistic_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlognormal:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nlognormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlower_bound_constrain:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_jacobian:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_unconstrain:\n\n\n(reals x, reals lb) : reals (transform_functions.html)\n\n\nlower_upper_bound_constrain:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_jacobian:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_unconstrain:\n\n\n(reals x, reals lb, reals ub) : reals (transform_functions.html)",
+    "text": "Removed: The built-in no-argument function get_lp() is deprecated.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Removed Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#m",
-    "href": "functions-reference/functions_index.html#m",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/removed_functions.html#fabs-function",
+    "href": "functions-reference/removed_functions.html#fabs-function",
+    "title": "Removed Functions",
     "section": "",
-    "text": "machine_precision:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nmap_rect:\n\n\n(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector (higher-order_functions.html)\n\n\nmatrix_exp:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nmatrix_exp_multiply:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nmatrix_power:\n\n\n(matrix A, int B) : matrix (matrix_operations.html)\n\n\nmax:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmdivide_left_spd:\n\n\n(matrix A, matrix B) : vector (matrix_operations.html)\n\n\n(matrix A, vector b) : matrix (matrix_operations.html)\n\n\nmdivide_left_tri_low:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\nmdivide_right_spd:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmdivide_right_tri_low:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmean:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmin:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmodified_bessel_first_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmodified_bessel_second_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmulti_gp:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lpdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lupdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lpdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lupdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_rng:\n\n\n(row_vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_rng:\n\n\n(row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_cholesky_t_rng:\n\n\n(real nu, vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lpdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lupdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_rng:\n\n\n(real nu, array[] row_vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, array[] vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lpdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lupdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_rng:\n\n\n(real nu, row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmultinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lpmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lupmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_rng:\n\n\n(vector gamma, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultinomial_lpmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_lupmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_rng:\n\n\n(vector theta, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultiply_lower_tri_self_transpose:\n\n\n(matrix x) : matrix (matrix_operations.html)",
+    "text": "Removed: The unary function fabs is deprecated.\nReplacement: Use the unary function abs instead. Note that the return type for abs is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.\nRemoved In: Stan 2.33",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Removed Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#n",
-    "href": "functions-reference/functions_index.html#n",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/removed_functions.html#cov_exp_quad",
+    "href": "functions-reference/removed_functions.html#cov_exp_quad",
+    "title": "Removed Functions",
     "section": "",
-    "text": "neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_cdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lccdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lcdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lpmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lupmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_rng:\n\n\n(reals eta, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lpmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lupmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_rng:\n\n\n(reals mu, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_cdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lccdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lcdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lpmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lupmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_rng:\n\n\n(reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nnegative_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnorm:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nnorm1:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnorm2:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnormal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lpdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lupdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nnot_a_number:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnum_elements:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)",
+    "text": "These covariance functions have been replaced by those described in Gaussian Process Covariance Functions\nWith magnitude \\(\\alpha\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(x_i, x_j) = \\alpha^2 \\exp \\left(-\\dfrac{1}{2\\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \\right)\n\\]\n \n\nmatrix cov_exp_quad(row_vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(row_vectors x1, row_vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x1, vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x1, array[] real x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Removed Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#o",
-    "href": "functions-reference/functions_index.html#o",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/removed_functions.html#real-arguments-to-logical-operators-operator-operator-and-operator",
+    "href": "functions-reference/removed_functions.html#real-arguments-to-logical-operators-operator-operator-and-operator",
+    "title": "Removed Functions",
     "section": "",
-    "text": "ode_adams:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_adams_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_adjoint_tol_ctl:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector (higher-order_functions.html)\n\n\node_bdf:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_bdf_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\noffset_multiplier_constrain:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_jacobian:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_unconstrain:\n\n\n(reals x, reals offset, reals mult) : reals (transform_functions.html)\n\n\none_hot_array:\n\n\n(int n, int k) : array[] real (matrix_operations.html)\n\n\none_hot_int_array:\n\n\n(int n, int k) : array[] int (matrix_operations.html)\n\n\none_hot_row_vector:\n\n\n(int n, int k) : row_vector (matrix_operations.html)\n\n\none_hot_vector:\n\n\n(int K, int k) : vector (matrix_operations.html)\n\n\nones_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nones_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nones_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nones_vector:\n\n\n(int n) : vector (matrix_operations.html)\n\n\noperator!:\n\n\n(int x) : int (real-valued_basic_functions.html)\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\noperator!=:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator%/%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator&&:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\n&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;operator’:\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\noperator*:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : vector (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, row_vector y) : matrix (matrix_operations.html)\n\n\noperator*=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator+:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator+=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator-:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(T x) : T (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (matrix_operations.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator-=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.*:\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator.*=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator./:\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator./=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.^:\n\n\n( complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n( complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator/:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix B, complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector b, complex_matrix A) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\noperator/=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator&lt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&lt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\noperator==:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator\\:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\noperator^:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\noperator||:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\nordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_logistic:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lpmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lupmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lpmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lupmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_probit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_probit_lpmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_lupmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nowens_t:\n\n\n(real h, real a) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)",
+    "text": "Removed: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.\nReplacement: Explicit x != 0 comparison is preferred instead.\nRemoved In: Stan 2.34",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Removed Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#p",
-    "href": "functions-reference/functions_index.html#p",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html",
+    "href": "functions-reference/real-valued_basic_functions.html",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "pareto:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_cdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lccdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lcdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lpdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lupdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_rng:\n\n\n(reals y_min, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\npareto_type_2:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_cdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lccdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lcdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lpdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lupdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_rng:\n\n\n(reals mu, reals lambda, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\nPhi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nPhi_approx:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\npi:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npoisson:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_cdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lccdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lcdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\npoisson_lpmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lupmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_rng:\n\n\n(reals lambda) : R (unbounded_discrete_distributions.html)\n\n\npolar:\n\n\n(real r, real theta) : complex (complex-valued_basic_functions.html)\n\n\npositive_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npositive_ordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\npow:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : Z (complex-valued_basic_functions.html)\n\n\nprint:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nprod:\n\n\n(array[] int x) : real (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nproj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)",
+    "text": "This chapter describes built-in functions that take zero or more real or integer arguments and return real values.\n\n\nAlthough listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.\n\n\nMany of Stan’s unary functions can be applied to any argument type. For example, the exponential function, exp, can be applied to real arguments or arrays of real arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.\n\n\nWhen applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like exp() are defined elementwise. For example,\n // declare some variables for arguments\n real x0;\n array[5] real x1;\n array[4, 7] real x2;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = exp(x0);\n y1 = exp(x1);\n y2 = exp(x2);\nWhen exp is applied to an array, it applies elementwise. For example, the statement above,\n y2 = exp(x2);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = exp(x2[i, j]);\n  }\n}\n\n\n\nVectorized functions also apply elementwise to vectors and matrices. For example,\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = exp(xv);\n yrv = exp(xrv);\n ym = exp(xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = exp(u);\nAfter this has been executed, z[i, j, k] will be equal to exp(u[i, j, k]).\n\n\n\nInteger arguments are promoted to real values in vectorized unary functions. Thus if n is of type int, exp(n) is of type real. Arrays work the same way, so that if n2 is a one dimensional array of integers, then exp(n2) will be a one-dimensional array of reals with the same number of elements as n2. For example,\n array[23] int n1;\n array[23] real z1;\n z1 = exp(n1);\nIt would be illegal to try to assign exp(n1) to an array of integers; the return type is a real array.\n\n\n\n\nLike the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.\n\n\nWhen applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like pow() are defined elementwise. For example,\n // declare some variables for arguments\n real x00;\n real x01;\n array[5] real x10;\n array[5]real x11;\n array[4, 7] real x20;\n array[4, 7] real x21;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = pow(x00, x01);\n y1 = pow(x10, x11);\n y2 = pow(x20, x21);\nWhen pow is applied to two arrays, it applies elementwise. For example, the statement above,\n y2 = pow(x20, x21);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x21[i, j]);\n  }\n}\nAlternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:\ny2 = pow(x20, x00);\nproduces the same result for y2 as the explicit loop:\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x00);\n  }\n}\n\n\n\nVectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,\n real x00;\n vector[5] xv00;\n vector[5] xv01;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = pow(xv00, xv01);\n yrv = pow(xrv, x00);\n ym = pow(x00, xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = pow(u, x00);\nAfter this has been executed, z[i, j, k] will be equal to pow(u[i, j, k], x00).\n\n\n\nVectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv = pow(xv, xv)\n row_vector[7] yrv = pow(xrv, xrv)\n matrix[10, 20] = pow(xm, xm)\nBut the following statements are not:\n vector[5] xv;\n vector[7] xv2;\n row_vector[5] xrv;\n\n // Cannot mix different types\n vector[5] yv = pow(xv, xrv)\n\n // Cannot mix different sizes of the same type\n vector[5] yv = pow(xv, xv2)\nWhile the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., bessel_first_kind). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:\n vector[5] xv;\n matrix[5, 5] xm;\n array[5] int xi;\n array[5, 5] int xii;\n\n vector[5] yv = bessel_first_kind(xi, xv);\n matrix[5, 5] ym = bessel_first_kind(xii, xm);\nWhereas these are not:\n vector[5] xv;\n matrix[5, 5] xm;\n array[7] int xi;\n\n // Dimensions of containers do not match\n vector[5] yv = bessel_first_kind(xi, xv);\n\n // Function requires first argument be an integer type\n matrix[5, 5] ym = bessel_first_kind(xm, xm);\n\n\n\n\n\nConstants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant \\(\\pi\\) must be written in a Stan program as pi().\n \n\nreal pi() \\(\\pi\\), the ratio of a circle’s circumference to its diameter\nAvailable since 2.0\n \n\nreal e() \\(e\\), the base of the natural logarithm\nAvailable since 2.0\n \n\nreal sqrt2() The square root of 2\nAvailable since 2.0\n \n\nreal log2() The natural logarithm of 2\nAvailable since 2.0\n \n\nreal log10() The natural logarithm of 10\nAvailable since 2.0\n\n\n\n \n\nreal not_a_number() Not-a-number, a special non-finite real value returned to signal an error\nAvailable since 2.0\n \n\nreal positive_infinity() Positive infinity, a special non-finite real value larger than all finite numbers\nAvailable since 2.0\n \n\nreal negative_infinity() Negative infinity, a special non-finite real value smaller than all finite numbers\nAvailable since 2.0\n \n\nreal machine_precision() The smallest number \\(x\\) such that \\((x + 1) \\neq 1\\) in floating-point arithmetic on the current hardware platform\nAvailable since 2.0\n\n\n\nThe basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (~) and log probability increment statement (increment_log_prob) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.\nStan provides a special built-in function target() that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.\n \n\nreal target() Return the current value of the log probability accumulator.\nAvailable since 2.10\ntarget acts like a function ending in _lp, meaning that it may only may only be used in the model block.\n\n\n\nLike C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.\n\n\nAll comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.\n \n\nint operator&lt;(int x, int y)\n \n\nint operator&lt;(real x, real y) Return 1 if x is less than y and 0 otherwise. \\[\\begin{equation*} \\text{operator&lt;}(x,y)\n= \\begin{cases} 1 & \\text{if $x &lt; y$} \\\\ 0 & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&lt;=(int x, int y)\n \n\nint operator&lt;=(real x, real y) Return 1 if x is less than or equal y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&lt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\leq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;(int x, int y)\n \n\nint operator&gt;(real x, real y) Return 1 if x is greater than y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;}(x,y) = \\begin{cases} 1 & \\text{if $x &gt; y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;=(int x, int y)\n \n\nint operator&gt;=(real x, real y) Return 1 if x is greater than or equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\geq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator==(int x, int y)\n \n\nint operator==(real x, real y) Return 1 if x is equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator==}(x,y) = \\begin{cases} 1 & \\text{if $x = y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!=(int x, int y)\n \n\nint operator!=(real x, real y) Return 1 if x is not equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator!=}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq y$} \\\\ 0 &\n\\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\nBoolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as !a && b is interpreted as (!a) && b, and a &lt; b || c &gt;= d && e != f as (a &lt; b) || (((c &gt;= d) && (e != f))).\n \n\nint operator!(int x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0$} \\\\ 1 & \\text{if $x = 0$}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!(real x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0.0$} \\\\ 1 & \\text{if $x = 0.0$}\n\\end{cases} \\end{equation*}\\] deprecated; - use operator== instead.\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator&&(int x, int y)\nReturn 1 if x is unequal to 0 and y is unequal to 0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\text{ and } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&&(real x, real y) Return 1 if x is unequal to 0.0 and y is unequal to 0.0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\text{ and } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator||(int x, int y) Return 1 if x is unequal to 0 or y is unequal to 0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\textrm{ or } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator||(real x, real y) Return 1 if x is unequal to 0.0 or y is unequal to 0.0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\textrm{ or } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n\n\nLike in C++, the boolean operators && and || and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating a || b, if a evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression b. Similarly, evaluating a && b first evaluates a, and if the result is zero, returns 0 without evaluating b.\n\n\n\n\nThe logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.\n \n\nreal step(real x) Return 1 if x is positive and 0 otherwise. \\[\\begin{equation*} \\text{step}(x) =\n\\begin{cases} 0 & \\text{if } x &lt; 0 \\\\ 1 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nThe step function is often used in BUGS to perform conditional operations. For instance, step(a-b) evaluates to 1 if a is greater than b and evaluates to 0 otherwise. step is a step-like functions; see the warning in section step functions applied to expressions dependent on parameters.\nAvailable since 2.0\n \n\nint is_inf(real x) Return 1 if x is infinite (positive or negative) and 0 otherwise.\nAvailable since 2.5\n \n\nint is_nan(real x) Return 1 if x is NaN and 0 otherwise.\nAvailable since 2.5\nCare must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section step-like functions for details.\n\n\n\n\nThe arithmetic operators are presented using C++ notation. For instance operator+(x,y) refers to the binary addition operator and operator-(x) to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as x + y and -x, respectively.\n\n\n \n\nreal operator+(real x, real y) Return the sum of x and y. \\[\\begin{equation*} (x + y) = \\text{operator+}(x,y) = x+y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator-(real x, real y) Return the difference between x and y. \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x,y) = x - y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator*(real x, real y) Return the product of x and y. \\[\\begin{equation*} (x * y) = \\text{operator*}(x,y) = xy\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator/(real x, real y) Return the quotient of x and y. \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator^(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} (x^\\mathrm{\\wedge}y) =\n\\text{operator}^\\mathrm{\\wedge}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.5\n\n\n\n \n\nreal operator-(real x) Return the negation of the subtrahend x. \\[\\begin{equation*} \\text{operator-}(x) = (-x)\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of reals, -x is the same shape array where each individual number is negated.\nAvailable since 2.31\n \n\nreal operator+(real x) Return the value of x. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n\nWarning: These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is real vector, row_vector, matrix, or an array of those types, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nreal fdim(real x, real y) Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. \\[\\begin{equation*} \\text{fdim}(x,y) = \\begin{cases} x-y &\n\\text{if } x \\geq y \\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fdim(T1 x, T2 y) Vectorized implementation of the fdim function\nAvailable since 2.25\n\n\n\n \n\nreal fmin(real x, real y) Return the minimum of x and y; see warning above. \\[\\begin{equation*} \\text{fmin}(x,y) = \\begin{cases} x &\n\\text{if } x \\leq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmin(T1 x, T2 y) Vectorized implementation of the fmin function\nAvailable since 2.25\n \n\nreal fmax(real x, real y) Return the maximum of x and y; see warning above. \\[\\begin{equation*} \\text{fmax}(x,y) = \\begin{cases} x &\n\\text{if } x \\geq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmax(T1 x, T2 y) Vectorized implementation of the fmax function\nAvailable since 2.25\n\n\n\n \n\nreal fmod(real x, real y) Return the real value remainder after dividing x by y; see warning above. \\[\\begin{equation*} \\text{fmod}(x,y) = x - \\left\\lfloor \\frac{x}{y} \\right\\rfloor \\, y \\end{equation*}\\] The operator \\(\\lfloor u \\rfloor\\) is the floor operation; see below.\nAvailable since 2.0\n \n\nR fmod(T1 x, T2 y) Vectorized implementation of the fmod function\nAvailable since 2.25\n\n\n\nWarning: Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.\nThe rounding functions cannot be used as indices to arrays because they return real values. For operations over data or in the generated quantities block, the to_int() function can be used.\n \n\nR floor(T x) The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR ceil(T x) The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR round(T x) The nearest integer to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR trunc(T x) The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n\n\n\n\n \n\nR sqrt(T x) The square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR cbrt(T x) The cube root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR square(T x) The square of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp(T x) The natural exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp2(T x) The base-2 exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log(T x) The natural logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log2(T x) The base-2 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log10(T x) The base-10 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal pow(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} \\text{pow}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.0\n \n\nR pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.25\n \n\nR inv(T x) The inverse of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_sqrt(T x) The inverse of the square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_square(T x) The inverse of the square of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal hypot(real x, real y) Return the length of the hypotenuse of a right triangle with sides of length x and y. \\[\\begin{equation*} \\text{hypot}(x,y) = \\begin{cases} \\sqrt{x^2+y^2} &\n\\text{if } x,y\\geq 0 \\\\ \\textrm{NaN} & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR hypot(T1 x, T2 y) Vectorized implementation of the hypot function\nAvailable since 2.25\n \n\nR cos(T x) The cosine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sin(T x) The sine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tan(T x) The tangent of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acos(T x) The principal arc (inverse) cosine (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR asin(T x) The principal arc (inverse) sine (in radians) of x\nAvailable since 2.0\n \n\nR atan(T x) The principal arc (inverse) tangent (in radians) of x, with values from \\(-\\pi/2\\) to \\(\\pi/2\\)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atan2(T y, T x) Return the principal arc (inverse) tangent (in radians) of y divided by x, \\[\\begin{equation*} \\text{atan2}(y, x) = \\arctan\\left(\\frac{y}{x}\\right) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.34\n\n\n\n \n\nR cosh(T x) The hyperbolic cosine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sinh(T x) The hyperbolic sine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tanh(T x) The hyperbolic tangent of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acosh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR asinh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atanh(T x) The inverse hyperbolic tangent (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\nThe following functions are commonly used as link functions in generalized linear models. The function \\(\\Phi\\) is also commonly used as a link function (see section probability-related functions).\n \n\nR logit(T x) The log odds, or logit, function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_logit(T x) The logistic sigmoid function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_cloglog(T x) The inverse of the complementary log-log function applied to x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n\n\nThe error function erf is related to the standard normal cumulative distribution function \\(\\Phi\\) by scaling. See section normal distribution for the general normal cumulative distribution function (and its complement).\n \n\nR erf(T x) The error function, also known as the Gauss error function, of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR erfc(T x) The complementary error function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_erfc(T x) The inverse of the complementary error function of x\nAvailable since 2.29, vectorized in 2.29\n \n\nR Phi(T x) The standard normal cumulative distribution function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_Phi(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The details of the algorithm can be found in (Wichura 1988). Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.\nAvailable since 2.0, vectorized in 2.13\n \n\nR Phi_approx(T x) The fast approximation of the unit (may replace Phi for probit regression with maximum absolute error of 0.00014, see (Bowling et al. 2009) for details)\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal binary_log_loss(int y, real y_hat) Return the log loss function for for predicting \\(\\hat{y} \\in [0,1]\\) for boolean outcome \\(y \\in \\{0,1\\}\\). \\[\\begin{equation*}\n\\mathrm{binary\\_log\\_loss}(y,\\hat{y}) = \\begin{cases} -\\log \\hat{y} &\n\\text{if } y = 1\\\\ -\\log (1 - \\hat{y}) & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR binary_log_loss(T1 x, T2 y) Vectorized implementation of the binary_log_loss function\nAvailable since 2.25\n \n\nreal owens_t(real h, real a) Return the Owen’s T function for the probability of the event \\(X &gt; h\\) and \\(0&lt;Y&lt;aX\\) where X and Y are independent standard normal random variables. \\[\\begin{equation*} \\mathrm{owens\\_t}(h,a) = \\frac{1}{2\\pi} \\int_0^a\n\\frac{\\exp(-\\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \\end{equation*}\\]\nAvailable since 2.25\n \n\nR owens_t(T1 x, T2 y) Vectorized implementation of the owens_t function\nAvailable since 2.25\n\n\n\n\n \n\nreal beta(real alpha, real beta) Return the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.25\n \n\nR beta(T1 x, T2 y) Vectorized implementation of the beta function\nAvailable since 2.25\n \n\nreal inc_beta(real alpha, real beta, real x) Return the regularized incomplete beta function up to x applied to alpha and beta. See section appendix for a definition.\nAvailable since 2.10\n \n\nreal inv_inc_beta(real alpha, real beta, real p) Return the inverse of the regularized incomplete beta function. The return value x is the value that solves p = inc_beta(alpha, beta, x). See section appendix for a definition of the inc_beta.\nAvailable since 2.30\n \n\nreal lbeta(real alpha, real beta) Return the natural logarithm of the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). \\[\\begin{equation*}\n\\text{lbeta}(\\alpha,\\beta) = \\log \\Gamma(\\alpha) + \\log \\Gamma(\\beta) - \\log \\Gamma(\\alpha+\\beta)\n\\end{equation*}\\] See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.0\n \n\nR lbeta(T1 x, T2 y) Vectorized implementation of the lbeta function\nAvailable since 2.25\n \n\nR tgamma(T x) The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that \\(\\Gamma(n+1) = n!\\). See for a full definition of \\(\\Gamma(x)\\). The function is defined for positive numbers and non-integral negative numbers,\nAvailable since 2.0, vectorized in 2.13\n \n\nR lgamma(T x) The natural logarithm of the gamma function applied to x,\nAvailable since 2.0, vectorized in 2.15\n \n\nR digamma(T x) The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers\nAvailable since 2.0, vectorized in 2.13\n \n\nR trigamma(T x) The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function\nAvailable since 2.0, vectorized in 2.13\n \n\nreal lmgamma(int n, real x) Return the natural logarithm of the multivariate gamma function \\(\\Gamma_n\\) with n dimensions applied to x. \\[\\begin{equation*}\n\\text{lmgamma}(n,x) =\n\\begin{cases} \\frac{n(n-1)}{4} \\log \\pi + \\sum_{j=1}^n \\log \\Gamma\\left(x + \\frac{1 - j}{2}\\right)\n& \\text{if } x\\not\\in \\{\\dots,-3,-2,-1,0\\}\\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR lmgamma(T1 x, T2 y) Vectorized implementation of the lmgamma function\nAvailable since 2.25\n \n\nreal gamma_p(real a, real z) Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_p}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_0^zt^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_p(T1 x, T2 y) Vectorized implementation of the gamma_p function\nAvailable since 2.25\n \n\nreal gamma_q(real a, real z) Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_q}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_z^\\infty t^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\[6pt] \\textrm{error} & \\text{otherwise}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_q(T1 x, T2 y) Vectorized implementation of the gamma_q function\nAvailable since 2.25\n \n\nint choose(int x, int y) Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” In its the antilog of the lchoose function but returns an integer rather than a real number with no non-zero decimal places. For \\(0 \\leq y \\leq x\\), the binomial coefficient function can be defined via the factorial function \\[\\begin{equation*}\n\\text{choose}(x,y) = \\frac{x!}{\\left(y!\\right)\\left(x - y\\right)!}.\n\\end{equation*}\\]\nAvailable since 2.14\n \n\nR choose(T1 x, T2 y) Vectorized implementation of the choose function\nAvailable since 2.25\n \n\nreal bessel_first_kind(int v, real x) Return the Bessel function of the first kind with order v applied to x. \\[\\begin{equation*}\n\\mathrm{bessel\\_first\\_kind}(v,x) = J_v(x),\n\\end{equation*}\\] where \\[\\begin{equation*}\nJ_v(x)=\\left(\\frac{1}{2}x\\right)^v \\sum_{k=0}^\\infty\n\\frac{\\left(-\\frac{1}{4}x^2\\right)^k}{k!\\, \\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_first_kind(T1 x, T2 y) Vectorized implementation of the bessel_first_kind function\nAvailable since 2.25\n \n\nreal bessel_second_kind(int v, real x) Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For \\(x,v &gt; 0\\), \\[\\begin{equation*}\n\\mathrm{bessel\\_second\\_kind}(v,x) =\n\\begin{cases} Y_v(x) & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\nY_v(x)=\\frac{J_v(x)\\cos(v\\pi)-J_{-v}(x)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_second_kind(T1 x, T2 y) Vectorized implementation of the bessel_second_kind function\nAvailable since 2.25\n \n\nreal modified_bessel_first_kind(int v, real z) Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_first\\_kind}(v,z) = I_v(z)\n\\end{equation*}\\] where \\[\\begin{equation*}\n{I_v}(z) = \\left(\\frac{1}{2}z\\right)^v\\sum_{k=0}^\\infty \\frac{\\left(\\frac{1}{4}z^2\\right)^k}{k!\\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_first_kind function\nAvailable since 2.25\n \n\nreal log_modified_bessel_first_kind(real v, real z) Return the log of the modified Bessel function of the first kind. v does not have to be an integer.\nAvailable since 2.26\n \n\nR log_modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the log_modified_bessel_first_kind function\nAvailable since 2.26\n \n\nreal modified_bessel_second_kind(int v, real z) Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_second\\_kind}(v,z) =\n\\begin{cases} K_v(z) &  \\text{if } z &gt; 0 \\\\ \\textrm{error} & \\text{if } z \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} {K_v}(z) = \\frac{\\pi}{2}\\cdot\\frac{I_{-v}(z) - I_{v}(z)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_second_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_second_kind function\nAvailable since 2.25\n \n\nreal falling_factorial(real x, real n) Return the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*}\n\\mathrm{falling\\_factorial}(x,n) =\n\\begin{cases} (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\n(x)_n=\\frac{\\Gamma(x+1)}{\\Gamma(x-n+1)}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR falling_factorial(T1 x, T2 y) Vectorized implementation of the falling_factorial function\nAvailable since 2.25\n \n\nreal lchoose(real x, real y) Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For \\(0 \\leq y \\leq x\\), \\[\\begin{equation*} \\mathrm{binomial\\_coefficient\\_log}(x,y) =\n\\log\\Gamma(x+1) - \\log\\Gamma(y+1) - \\log\\Gamma(x-y+1). \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lchoose(T1 x, T2 y) Vectorized implementation of the lchoose function\nAvailable since 2.29\n \n\nreal log_falling_factorial(real x, real n) Return the log of the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_falling\\_factorial}(x,n) =\n\\begin{cases} \\log (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal rising_factorial(real x, int n) Return the rising factorial of x with power n defined for positive x and integer n. \\[\\begin{equation*}\n\\mathrm{rising\\_factorial}(x,n) = \\begin{cases} x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} x^{(n)}=\\frac{\\Gamma(x+n)}{\\Gamma(x)} \\end{equation*}\\]\nAvailable since 2.20\n \n\nR rising_factorial(T1 x, T2 y) Vectorized implementation of the rising_factorial function\nAvailable since 2.25\n \n\nreal log_rising_factorial(real x, real n) Return the log of the rising factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_rising\\_factorial}(x,n) =\n\\begin{cases} \\log x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR log_rising_factorial(T1 x, T2 y) Vectorized implementation of the log_rising_factorial function\nAvailable since 2.25\n\n\n\nThe functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.\n \n\nR expm1(T x) The natural exponential of x minus 1\nAvailable since 2.0, vectorized in 2.13\n \n\nreal fma(real x, real y, real z) Return z plus the result of x multiplied by y. \\[\\begin{equation*} \\text{fma}(x,y,z) =\n(x \\times y) + z \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal ldexp(real x, int y) Return the product of x and two raised to the y power. \\[\\begin{equation*}\n\\text{ldexp}(x,y) = x 2^y  \\end{equation*}\\]\nAvailable since 2.25\n \n\nR ldexp(T1 x, T2 y) Vectorized implementation of the ldexp function\nAvailable since 2.25\n \n\nreal lmultiply(real x, real y) Return the product of x and the natural logarithm of y. \\[\\begin{equation*}\n\\text{lmultiply}(x,y) = \\begin{cases} 0 & \\text{if } x = y = 0 \\\\ x\n\\log y & \\text{if } x, y \\neq 0 \\\\ \\text{NaN} & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lmultiply(T1 x, T2 y) Vectorized implementation of the lmultiply function\nAvailable since 2.25\n \n\nR log1p(T x) The natural logarithm of 1 plus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m(T x) The natural logarithm of 1 minus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1p_exp(T x) The natural logarithm of one plus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m_exp(T x) The logarithm of one minus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal log_diff_exp(real x, real y) Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_diff\\_exp}(x,y) = \\begin{cases} \\log(\\exp(x)-\\exp(y)) &\n\\text{if } +\\infty &gt; x \\ge y \\\\[6pt]\n\\textrm{NaN} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nWhen x is equal to y, log_diff_exp(x, y) returns \\(-\\infty\\), consistent with log(0) returning \\(-\\infty\\). This includes the case in which x and y are both equal to \\(-\\infty\\), which corresponds to log(0 - 0) because exp(negative_infinity()) returns 0.\nAvailable since 2.0\n \n\nR log_diff_exp(T1 x, T2 y) Vectorized implementation of the log_diff_exp function\nAvailable since 2.25\n \n\nreal log_mix(real theta, real lp1, real lp2) Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by \\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda_1, \\lambda_2) & = & \\log \\!\\left(\n\\theta \\exp(\\lambda_1) + \\left( 1 - \\theta \\right) \\exp(\\lambda_2)\n\\right) \\\\[3pt] & = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) +\n\\lambda_1, \\ \\log(1 - \\theta) + \\lambda_2\\right). \\end{eqnarray*}\\]\nAvailable since 2.6\n \n\nR log_mix(T1 thetas, T2 lps)\nCalculates the log mixture density given thetas, mixing proportions which should be between 0 and 1 and sum to 1, and lps, log densities. The lps variable must be either a 1-d container of the same length as thetas, or an array of such.\n\\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda)\n& = & \\log \\!\\left( \\sum_{n=1}^N \\theta_n * \\exp(\\lambda_n) \\right) \\\\[3pt]\n& = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) + \\lambda\\right).\n\\end{eqnarray*}\\]\nThis is a generalization of the above signature of three arguments to more than two densities. For example, log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2}).\nAvailable since 2.26\n \n\nR log_sum_exp(T1 x, T2 y) Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_sum\\_exp}(x,y) = \\log(\\exp(x)+\\exp(y)) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.33\n \n\nR log_inv_logit(T x) The natural logarithm of the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log_inv_logit_diff(T1 x, T2 y) The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y\nAvailable since 2.25\n \n\nR log1m_inv_logit(T x) The natural logarithm of 1 minus the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nR lambert_w0(T x) Implementation of the \\(W_0\\) branch of the Lambert W function, i.e., solution to the function \\(W_0(x) \\exp^{ W_0(x)} = x\\)\nAvailable since 2.25\n \n\nR lambert_wm1(T x) Implementation of the \\(W_{-1}\\) branch of the Lambert W function, i.e., solution to the function \\(W_{-1}(x) \\exp^{W_{-1}(x)} = x\\)\nAvailable since 2.25\n\n\n\nHypergeometric functions refer to a power series of the form \\[\\begin{equation*}\n_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \\sum_{n=0}^\\infty \\frac{(a_1)_n\\cdot\\cdot\\cdot(a_p)_n}{(b_1)_n\\cdot\\cdot\\cdot(b_q)_n} \\frac{z^n}{n!}\n\\end{equation*}\\] where \\((a)_n\\) is the Pochhammer symbol defined as \\((a)_n = \\frac{\\Gamma(a+n)}{\\Gamma(a)}\\).\nThe gradients of the hypergeometric function are given by: \\[\\begin{equation*}\n\\frac{\\partial }{\\partial a_1} =\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+a_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n    - \\psi\\left(a_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)\n\\end{equation*}\\] \\[\\begin{equation*}\n\\frac{\\partial }{\\partial b_1} =\n  \\psi\\left(b_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+b_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n\\end{equation*}\\] \\[\\begin{equation*}\n  \\frac{\\partial }{\\partial z} =\n  \\frac{\\prod_{j=1}^{p}a_j}{\\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)\n\\end{equation*}\\]\nStan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.\n \n\nreal hypergeometric_1F0(real a, real z) Special case of the hypergeometric function with \\(p=1\\) and \\(q=0\\).\nAvailable since 2.37\n \n\nreal hypergeometric_2F1(real a1, real a2, real b1, real z) Special case of the hypergeometric function with \\(p=2\\) and \\(q=1\\). If the function does not meet convergence criteria for given inputs, the function will attempt to apply Euler’s transformation to improve convergence: \\[\\begin{equation*}\n{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \\frac{z}{z-1})\\cdot(1-z)^{-a_2}\n\\end{equation*}\\]\nAvailable since 2.37\n \n\nreal hypergeometric_3F2(T1 a, T2 b, real z) Special case of the hypergeometric function with \\(p=3\\) and \\(q=2\\), where a and b are vectors of length 3 and 2, respectively.\nAvailable since 2.37\n \n\nreal hypergeometric_pFq(T1 a, T2 b, real z) Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.\nAvailable since 2.37",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#q",
-    "href": "functions-reference/functions_index.html#q",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#fun-vectorization",
+    "href": "functions-reference/real-valued_basic_functions.html#fun-vectorization",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "qr:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_thin_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nquad_form:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquad_form_diag:\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\nquad_form_sym:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquantile:\n\n\n(data array[] real x, data array[] real p) : array[] real (array_operations.html)\n\n\n(data array[] real x, data real p) : real (array_operations.html)\n\n\n(data row_vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data row_vector x, data real p) : real (matrix_operations.html)\n\n\n(data vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data vector x, data real p) : real (matrix_operations.html)",
+    "text": "Although listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.\n\n\nMany of Stan’s unary functions can be applied to any argument type. For example, the exponential function, exp, can be applied to real arguments or arrays of real arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.\n\n\nWhen applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like exp() are defined elementwise. For example,\n // declare some variables for arguments\n real x0;\n array[5] real x1;\n array[4, 7] real x2;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = exp(x0);\n y1 = exp(x1);\n y2 = exp(x2);\nWhen exp is applied to an array, it applies elementwise. For example, the statement above,\n y2 = exp(x2);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = exp(x2[i, j]);\n  }\n}\n\n\n\nVectorized functions also apply elementwise to vectors and matrices. For example,\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = exp(xv);\n yrv = exp(xrv);\n ym = exp(xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = exp(u);\nAfter this has been executed, z[i, j, k] will be equal to exp(u[i, j, k]).\n\n\n\nInteger arguments are promoted to real values in vectorized unary functions. Thus if n is of type int, exp(n) is of type real. Arrays work the same way, so that if n2 is a one dimensional array of integers, then exp(n2) will be a one-dimensional array of reals with the same number of elements as n2. For example,\n array[23] int n1;\n array[23] real z1;\n z1 = exp(n1);\nIt would be illegal to try to assign exp(n1) to an array of integers; the return type is a real array.\n\n\n\n\nLike the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.\n\n\nWhen applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like pow() are defined elementwise. For example,\n // declare some variables for arguments\n real x00;\n real x01;\n array[5] real x10;\n array[5]real x11;\n array[4, 7] real x20;\n array[4, 7] real x21;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = pow(x00, x01);\n y1 = pow(x10, x11);\n y2 = pow(x20, x21);\nWhen pow is applied to two arrays, it applies elementwise. For example, the statement above,\n y2 = pow(x20, x21);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x21[i, j]);\n  }\n}\nAlternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:\ny2 = pow(x20, x00);\nproduces the same result for y2 as the explicit loop:\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x00);\n  }\n}\n\n\n\nVectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,\n real x00;\n vector[5] xv00;\n vector[5] xv01;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = pow(xv00, xv01);\n yrv = pow(xrv, x00);\n ym = pow(x00, xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = pow(u, x00);\nAfter this has been executed, z[i, j, k] will be equal to pow(u[i, j, k], x00).\n\n\n\nVectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv = pow(xv, xv)\n row_vector[7] yrv = pow(xrv, xrv)\n matrix[10, 20] = pow(xm, xm)\nBut the following statements are not:\n vector[5] xv;\n vector[7] xv2;\n row_vector[5] xrv;\n\n // Cannot mix different types\n vector[5] yv = pow(xv, xrv)\n\n // Cannot mix different sizes of the same type\n vector[5] yv = pow(xv, xv2)\nWhile the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., bessel_first_kind). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:\n vector[5] xv;\n matrix[5, 5] xm;\n array[5] int xi;\n array[5, 5] int xii;\n\n vector[5] yv = bessel_first_kind(xi, xv);\n matrix[5, 5] ym = bessel_first_kind(xii, xm);\nWhereas these are not:\n vector[5] xv;\n matrix[5, 5] xm;\n array[7] int xi;\n\n // Dimensions of containers do not match\n vector[5] yv = bessel_first_kind(xi, xv);\n\n // Function requires first argument be an integer type\n matrix[5, 5] ym = bessel_first_kind(xm, xm);",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#r",
-    "href": "functions-reference/functions_index.html#r",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#built-in-constants",
+    "href": "functions-reference/real-valued_basic_functions.html#built-in-constants",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "rank:\n\n\n(array[] int v, int s) : int (array_operations.html)\n\n\n(array[] real v, int s) : int (array_operations.html)\n\n\n(row_vector v, int s) : int (matrix_operations.html)\n\n\n(vector v, int s) : int (matrix_operations.html)\n\n\nrayleigh:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nrayleigh_cdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lccdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lcdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lpdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lupdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_rng:\n\n\n(reals sigma) : R (positive_continuous_distributions.html)\n\n\nreduce_sum:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreduce_sum_static:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreject:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nrep_array:\n\n\n(T x, int k, int m, int n) : array[,,] T (array_operations.html)\n\n\n(T x, int m, int n) : array[,] T (array_operations.html)\n\n\n(T x, int n) : array[] T (array_operations.html)\n\n\nrep_matrix:\n\n\n(complex z, int m, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector rv, int m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(real x, int m, int n) : matrix (matrix_operations.html)\n\n\n(row_vector rv, int m) : matrix (matrix_operations.html)\n\n\n(vector v, int n) : matrix (matrix_operations.html)\n\n\nrep_row_vector:\n\n\n(complex z, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(real x, int n) : row_vector (matrix_operations.html)\n\n\nrep_vector:\n\n\n(complex z, int m) : complex_vector (complex_matrix_operations.html)\n\n\n(real x, int m) : vector (matrix_operations.html)\n\n\nreverse:\n\n\n(array[] T v) : array[] T (array_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nrising_factorial:\n\n\n(real x, int n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nround:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nrow:\n\n\n(complex_matrix x, int m) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int m) : row_vector (matrix_operations.html)\n\n\nrows:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nrows_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\nrows_dot_self:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : vector (matrix_operations.html)",
+    "text": "Constants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant \\(\\pi\\) must be written in a Stan program as pi().\n \n\nreal pi() \\(\\pi\\), the ratio of a circle’s circumference to its diameter\nAvailable since 2.0\n \n\nreal e() \\(e\\), the base of the natural logarithm\nAvailable since 2.0\n \n\nreal sqrt2() The square root of 2\nAvailable since 2.0\n \n\nreal log2() The natural logarithm of 2\nAvailable since 2.0\n \n\nreal log10() The natural logarithm of 10\nAvailable since 2.0",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#s",
-    "href": "functions-reference/functions_index.html#s",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#special-values",
+    "href": "functions-reference/real-valued_basic_functions.html#special-values",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "scale_matrix_exp_multiply:\n\n\n(real t, matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nscaled_inv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_cdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lccdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lcdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lpdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lupdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_rng:\n\n\n(reals nu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nsd:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsegment:\n\n\n(array[] T sv, int i, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int i, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int i, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int i, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int i, int n) : vector (matrix_operations.html)\n\n\nsimplex_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsingular_values:\n\n\n(complex_matrix A) : vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\nsinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsize:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nskew_double_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nskew_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_normal_cdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lccdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lcdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lpdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lupdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_rng:\n\n\n(reals xi, reals omega, real alpha) : R (unbounded_continuous_distributions.html)\n\n\nsoftmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nsolve_newton:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_newton_tol:\n\n\n(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell_tol:\n\n\n(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsort_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_indices_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsort_indices_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsqrt:\n\n\n(complex x) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsqrt2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nsquare:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsquared_distance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\nstd_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstd_normal_cdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lccdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lcdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_log_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_lpdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lupdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_rng:\n\n\n() : real (unbounded_continuous_distributions.html)\n\n\nstep:\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\nstochastic_column_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstochastic_row_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstudent_t:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstudent_t_cdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lccdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lcdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lpdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lupdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_rng:\n\n\n(reals nu, reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nsub_col:\n\n\n(complex_matrix x, int i, int j, int n_rows) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows) : vector (matrix_operations.html)\n\n\nsub_row:\n\n\n(complex_matrix x, int i, int j, int n_cols) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_cols) : row_vector (matrix_operations.html)\n\n\nsum:\n\n\n(array[] complex x) : complex (array_operations.html)\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsum_to_zero_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsvd:\n\n\n(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector, matrix) (matrix_operations.html)\n\n\nsvd_U:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsvd_V:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsymmetrize_from_lower_tri:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)",
+    "text": "real not_a_number() Not-a-number, a special non-finite real value returned to signal an error\nAvailable since 2.0\n \n\nreal positive_infinity() Positive infinity, a special non-finite real value larger than all finite numbers\nAvailable since 2.0\n \n\nreal negative_infinity() Negative infinity, a special non-finite real value smaller than all finite numbers\nAvailable since 2.0\n \n\nreal machine_precision() The smallest number \\(x\\) such that \\((x + 1) \\neq 1\\) in floating-point arithmetic on the current hardware platform\nAvailable since 2.0",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#t",
-    "href": "functions-reference/functions_index.html#t",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#get-log-prob",
+    "href": "functions-reference/real-valued_basic_functions.html#get-log-prob",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "tail:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\ntan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntarget:\n\n\n() : real (real-valued_basic_functions.html)\n\n\ntcrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ntgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nto_array_1d:\n\n\n(array[...] complex a) : array[] complex (mixed_operations.html)\n\n\n(array[...] int a) : array[] int (mixed_operations.html)\n\n\n(array[...] real a) : array[] real (mixed_operations.html)\n\n\n(complex_matrix m) : array[] complex (mixed_operations.html)\n\n\n(complex_row_vector v) : array[] complex (mixed_operations.html)\n\n\n(complex_vector v) : array[] real (mixed_operations.html)\n\n\n(matrix m) : array[] real (mixed_operations.html)\n\n\n(row_vector v) : array[] real (mixed_operations.html)\n\n\n(vector v) : array[] real (mixed_operations.html)\n\n\nto_array_2d:\n\n\n(complex_matrix m) : array[,] real (mixed_operations.html)\n\n\n(matrix m) : array[,] real (mixed_operations.html)\n\n\nto_complex:\n\n\n() : complex (complex-valued_basic_functions.html)\n\n\n(real re) : complex (complex-valued_basic_functions.html)\n\n\n(real re, real im) : complex (complex-valued_basic_functions.html)\n\n\n(T1 re, T2 im) : Z (complex-valued_basic_functions.html)\n\n\nto_int:\n\n\n(data real x) : int (integer-valued_basic_functions.html)\n\n\nto_matrix:\n\n\n(array[,] complex a ) : complex_matrix (mixed_operations.html)\n\n\n(array[,] int a) : matrix (mixed_operations.html)\n\n\n(array[,] real a) : matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex_row_vector vs) : complex_matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] row_vector vs) : matrix (mixed_operations.html)\n\n\n(complex_matrix A, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix m) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix M, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(matrix A, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(matrix m) : matrix (mixed_operations.html)\n\n\n(matrix M, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(vector v) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\nto_row_vector:\n\n\n(array[] complex a) : complex_row_vector (mixed_operations.html)\n\n\n(array[] int a) : row_vector (mixed_operations.html)\n\n\n(array[] real a) : row_vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_row_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(matrix m) : row_vector (mixed_operations.html)\n\n\n(row_vector v) : row_vector (mixed_operations.html)\n\n\n(vector v) : row_vector (mixed_operations.html)\n\n\nto_vector:\n\n\n(array[] complex a) : complex_vector (mixed_operations.html)\n\n\n(array[] int a) : vector (mixed_operations.html)\n\n\n(array[] real a) : vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_vector (mixed_operations.html)\n\n\n(matrix m) : vector (mixed_operations.html)\n\n\n(row_vector v) : vector (mixed_operations.html)\n\n\n(vector v) : vector (mixed_operations.html)\n\n\ntrace:\n\n\n(complex_matrix A) : complex (complex_matrix_operations.html)\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ntrace_gen_quad_form:\n\n\n(matrix D ,matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_quad_form:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\ntrigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntrunc:\n\n\n(T x) : R (real-valued_basic_functions.html)",
+    "text": "The basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (~) and log probability increment statement (increment_log_prob) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.\nStan provides a special built-in function target() that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.\n \n\nreal target() Return the current value of the log probability accumulator.\nAvailable since 2.10\ntarget acts like a function ending in _lp, meaning that it may only may only be used in the model block.",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#u",
-    "href": "functions-reference/functions_index.html#u",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#logical-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#logical-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "uniform:\n\n\ndistribution statement (bounded_continuous_distributions.html)\n\n\nuniform_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_rng:\n\n\n(reals alpha, reals beta) : R (bounded_continuous_distributions.html)\n\n\nuniform_simplex:\n\n\n(int n) : vector (matrix_operations.html)\n\n\nunit_vectors_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nupper_bound_constrain:\n\n\n(reals y, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_jacobian:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_unconstrain:\n\n\n(reals x, reals ub) : reals (transform_functions.html)",
+    "text": "Like C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.\n\n\nAll comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.\n \n\nint operator&lt;(int x, int y)\n \n\nint operator&lt;(real x, real y) Return 1 if x is less than y and 0 otherwise. \\[\\begin{equation*} \\text{operator&lt;}(x,y)\n= \\begin{cases} 1 & \\text{if $x &lt; y$} \\\\ 0 & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&lt;=(int x, int y)\n \n\nint operator&lt;=(real x, real y) Return 1 if x is less than or equal y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&lt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\leq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;(int x, int y)\n \n\nint operator&gt;(real x, real y) Return 1 if x is greater than y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;}(x,y) = \\begin{cases} 1 & \\text{if $x &gt; y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;=(int x, int y)\n \n\nint operator&gt;=(real x, real y) Return 1 if x is greater than or equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\geq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator==(int x, int y)\n \n\nint operator==(real x, real y) Return 1 if x is equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator==}(x,y) = \\begin{cases} 1 & \\text{if $x = y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!=(int x, int y)\n \n\nint operator!=(real x, real y) Return 1 if x is not equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator!=}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq y$} \\\\ 0 &\n\\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\nBoolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as !a && b is interpreted as (!a) && b, and a &lt; b || c &gt;= d && e != f as (a &lt; b) || (((c &gt;= d) && (e != f))).\n \n\nint operator!(int x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0$} \\\\ 1 & \\text{if $x = 0$}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!(real x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0.0$} \\\\ 1 & \\text{if $x = 0.0$}\n\\end{cases} \\end{equation*}\\] deprecated; - use operator== instead.\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator&&(int x, int y)\nReturn 1 if x is unequal to 0 and y is unequal to 0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\text{ and } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&&(real x, real y) Return 1 if x is unequal to 0.0 and y is unequal to 0.0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\text{ and } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator||(int x, int y) Return 1 if x is unequal to 0 or y is unequal to 0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\textrm{ or } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator||(real x, real y) Return 1 if x is unequal to 0.0 or y is unequal to 0.0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\textrm{ or } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n\n\nLike in C++, the boolean operators && and || and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating a || b, if a evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression b. Similarly, evaluating a && b first evaluates a, and if the result is zero, returns 0 without evaluating b.\n\n\n\n\nThe logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.\n \n\nreal step(real x) Return 1 if x is positive and 0 otherwise. \\[\\begin{equation*} \\text{step}(x) =\n\\begin{cases} 0 & \\text{if } x &lt; 0 \\\\ 1 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nThe step function is often used in BUGS to perform conditional operations. For instance, step(a-b) evaluates to 1 if a is greater than b and evaluates to 0 otherwise. step is a step-like functions; see the warning in section step functions applied to expressions dependent on parameters.\nAvailable since 2.0\n \n\nint is_inf(real x) Return 1 if x is infinite (positive or negative) and 0 otherwise.\nAvailable since 2.5\n \n\nint is_nan(real x) Return 1 if x is NaN and 0 otherwise.\nAvailable since 2.5\nCare must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section step-like functions for details.",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#v",
-    "href": "functions-reference/functions_index.html#v",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#real-valued-arithmetic-operators",
+    "href": "functions-reference/real-valued_basic_functions.html#real-valued-arithmetic-operators",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "variance:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nvon_mises:\n\n\ndistribution statement (circular_distributions.html)\n\n\nvon_mises_cdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lccdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lcdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lpdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lupdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_rng:\n\n\n(reals mu, reals kappa) : R (circular_distributions.html)",
+    "text": "The arithmetic operators are presented using C++ notation. For instance operator+(x,y) refers to the binary addition operator and operator-(x) to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as x + y and -x, respectively.\n\n\n \n\nreal operator+(real x, real y) Return the sum of x and y. \\[\\begin{equation*} (x + y) = \\text{operator+}(x,y) = x+y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator-(real x, real y) Return the difference between x and y. \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x,y) = x - y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator*(real x, real y) Return the product of x and y. \\[\\begin{equation*} (x * y) = \\text{operator*}(x,y) = xy\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator/(real x, real y) Return the quotient of x and y. \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator^(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} (x^\\mathrm{\\wedge}y) =\n\\text{operator}^\\mathrm{\\wedge}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.5\n\n\n\n \n\nreal operator-(real x) Return the negation of the subtrahend x. \\[\\begin{equation*} \\text{operator-}(x) = (-x)\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of reals, -x is the same shape array where each individual number is negated.\nAvailable since 2.31\n \n\nreal operator+(real x) Return the value of x. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#w",
-    "href": "functions-reference/functions_index.html#w",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#step-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#step-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "weibull:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nweibull_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\nwiener:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\nwiener_lccdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lcdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lpdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lupdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\nwishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)",
+    "text": "Warning: These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is real vector, row_vector, matrix, or an array of those types, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nreal fdim(real x, real y) Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. \\[\\begin{equation*} \\text{fdim}(x,y) = \\begin{cases} x-y &\n\\text{if } x \\geq y \\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fdim(T1 x, T2 y) Vectorized implementation of the fdim function\nAvailable since 2.25\n\n\n\n \n\nreal fmin(real x, real y) Return the minimum of x and y; see warning above. \\[\\begin{equation*} \\text{fmin}(x,y) = \\begin{cases} x &\n\\text{if } x \\leq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmin(T1 x, T2 y) Vectorized implementation of the fmin function\nAvailable since 2.25\n \n\nreal fmax(real x, real y) Return the maximum of x and y; see warning above. \\[\\begin{equation*} \\text{fmax}(x,y) = \\begin{cases} x &\n\\text{if } x \\geq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmax(T1 x, T2 y) Vectorized implementation of the fmax function\nAvailable since 2.25\n\n\n\n \n\nreal fmod(real x, real y) Return the real value remainder after dividing x by y; see warning above. \\[\\begin{equation*} \\text{fmod}(x,y) = x - \\left\\lfloor \\frac{x}{y} \\right\\rfloor \\, y \\end{equation*}\\] The operator \\(\\lfloor u \\rfloor\\) is the floor operation; see below.\nAvailable since 2.0\n \n\nR fmod(T1 x, T2 y) Vectorized implementation of the fmod function\nAvailable since 2.25\n\n\n\nWarning: Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.\nThe rounding functions cannot be used as indices to arrays because they return real values. For operations over data or in the generated quantities block, the to_int() function can be used.\n \n\nR floor(T x) The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR ceil(T x) The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR round(T x) The nearest integer to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR trunc(T x) The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/functions_index.html#z",
-    "href": "functions-reference/functions_index.html#z",
-    "title": "Alphabetical Index",
+    "objectID": "functions-reference/real-valued_basic_functions.html#power-and-logarithm-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#power-and-logarithm-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "zeros_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nzeros_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nzeros_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nzeros_vector:\n\n\n(int n) : vector (matrix_operations.html)",
+    "text": "R sqrt(T x) The square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR cbrt(T x) The cube root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR square(T x) The square of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp(T x) The natural exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp2(T x) The base-2 exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log(T x) The natural logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log2(T x) The base-2 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log10(T x) The base-10 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal pow(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} \\text{pow}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.0\n \n\nR pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.25\n \n\nR inv(T x) The inverse of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_sqrt(T x) The inverse of the square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_square(T x) The inverse of the square of x\nAvailable since 2.0, vectorized in 2.13",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Alphabetical Index"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/deprecated_functions.html",
-    "href": "functions-reference/deprecated_functions.html",
-    "title": "Deprecated Functions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#trigonometric-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#trigonometric-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "This appendix lists currently deprecated functionality along with how to replace it.\nStarting in Stan 2.29, deprecated functions with drop in replacements (such as the renaming of get_lp or multiply_log) will be removed 3 versions later e.g., functions deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in Removed Functions. The Stan compiler can automatically update these on the behalf of the user for the entire deprecation window and at least one version following the removal.\n\n\nDeprecated: Using / with two integer arguments is interpreted as integer floor division, such that\n\\[ 1 / 2 = 0 \\]\nThis is deprecated due to its confusion with real-valued division, where\n\\[ 1.0 / 2.0 = 0.5 \\]\nReplacement: Use the integer division operator operator%/% instead.\n\n\n\nThese ODE integrator functions have been replaced by those described in Ordinary Differential Equation (ODE) Solvers.\n\n\nA system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:\narray[] real ode(real time, array[] real state, array[] real theta,\n                 array[] real x_r, array[] int x_i);\nThe ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\ntheta, parameter values used to evaluate the ODE system\nx_r, data values used to evaluate the ODE system\nx_i, integer data values used to evaluate the ODE system.\n\nThe ODE system function separates parameter values, theta, from data values, x_r, for efficiency in computing the gradients of the ODE.\n\n\n\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.23, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.23, deprecated in 2.24\n\n\n\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.\n\node: function literal referring to a function specifying the system of differential equations with signature:\n\n(real, array[] real, array[] real, data array[] real, data array[] int):array[] real\nThe arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,\n\ninitial_state: initial state, type array[] real,\ninitial_time: initial time, type int or real,\ntimes: solution times, type array[] real,\ntheta: parameters, type array[] real,\ndata x_r: real data, type array[] real, data only, and\ndata x_i: integer data, type array[] int, data only.\n\nFor more fine-grained control of the ODE solvers, these parameters can also be provided:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\ndata max_num_steps: maximum number of steps to take in the ODE solver, type int, data only.\n\n\n\nThe return value for the ODE solvers is an array of type array[,] real, with values consisting of solutions at the specified times.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,\nsolution times and number of rows of the return value of the solver,\nparameters, real data and integer data passed to the solver will be passed to the system function\n\n\n\n\n\n\nThese algebraic solver functions have been replaced by those described in Algebraic Equation Solvers..\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:\n vector algebra_system(vector y, vector theta,\n                              data array[] real x_r, array[] int x_i)\nThe algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.\nThe argument of this function are:\n\ny, the unknowns we wish to solve for\ntheta, parameter values used to evaluate the algebraic system\nx_r, data values used to evaluate the algebraic system\nx_i, integer data used to evaluate the algebraic system\n\nThe algebraic system function separates parameter values, theta, from data values, x_r, for efficiency in propagating the derivatives through the algebraic system.\n\n\n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.\nAvailable since 2.17, deprecated in 2.31\n \n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.\nAvailable since 2.17, deprecated in 2.31\nNote: In future releases, the function algebra_solver will be deprecated and replaced with algebra_solver_powell.\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.24, deprecated in 2.31\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.24, deprecated in 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, vector, array[] real, array[] int):vector. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\ntheta: parameters only, type vector,\nx_r: real data only, type array[] real, and\nx_i: integer data only, type array[] int.\n\nFor more fine-grained control of the algebraic solver, these parameters can also be provided:\n\nrel_tol: relative tolerance for the algebraic solver, type real, data only,\nfunction_tol: function tolerance for the algebraic solver, type real, data only,\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only.\n\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0.\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\nThe parameters, real data, and integer data will be passed from the solver directly to the system function.",
+    "text": "real hypot(real x, real y) Return the length of the hypotenuse of a right triangle with sides of length x and y. \\[\\begin{equation*} \\text{hypot}(x,y) = \\begin{cases} \\sqrt{x^2+y^2} &\n\\text{if } x,y\\geq 0 \\\\ \\textrm{NaN} & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR hypot(T1 x, T2 y) Vectorized implementation of the hypot function\nAvailable since 2.25\n \n\nR cos(T x) The cosine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sin(T x) The sine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tan(T x) The tangent of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acos(T x) The principal arc (inverse) cosine (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR asin(T x) The principal arc (inverse) sine (in radians) of x\nAvailable since 2.0\n \n\nR atan(T x) The principal arc (inverse) tangent (in radians) of x, with values from \\(-\\pi/2\\) to \\(\\pi/2\\)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atan2(T y, T x) Return the principal arc (inverse) tangent (in radians) of y divided by x, \\[\\begin{equation*} \\text{atan2}(y, x) = \\arctan\\left(\\frac{y}{x}\\right) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.34",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Deprecated Functions"
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/deprecated_functions.html#integer-division-with-operator",
-    "href": "functions-reference/deprecated_functions.html#integer-division-with-operator",
-    "title": "Deprecated Functions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#hyperbolic-trigonometric-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#hyperbolic-trigonometric-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "Deprecated: Using / with two integer arguments is interpreted as integer floor division, such that\n\\[ 1 / 2 = 0 \\]\nThis is deprecated due to its confusion with real-valued division, where\n\\[ 1.0 / 2.0 = 0.5 \\]\nReplacement: Use the integer division operator operator%/% instead.",
+    "text": "R cosh(T x) The hyperbolic cosine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sinh(T x) The hyperbolic sine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tanh(T x) The hyperbolic tangent of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acosh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR asinh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atanh(T x) The inverse hyperbolic tangent (in radians) of x\nAvailable since 2.0, vectorized in 2.13",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Deprecated Functions"
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/deprecated_functions.html#functions-old-ode-solver",
-    "href": "functions-reference/deprecated_functions.html#functions-old-ode-solver",
-    "title": "Deprecated Functions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#link-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#link-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "These ODE integrator functions have been replaced by those described in Ordinary Differential Equation (ODE) Solvers.\n\n\nA system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:\narray[] real ode(real time, array[] real state, array[] real theta,\n                 array[] real x_r, array[] int x_i);\nThe ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\ntheta, parameter values used to evaluate the ODE system\nx_r, data values used to evaluate the ODE system\nx_i, integer data values used to evaluate the ODE system.\n\nThe ODE system function separates parameter values, theta, from data values, x_r, for efficiency in computing the gradients of the ODE.\n\n\n\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.23, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.23, deprecated in 2.24\n\n\n\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.\n\node: function literal referring to a function specifying the system of differential equations with signature:\n\n(real, array[] real, array[] real, data array[] real, data array[] int):array[] real\nThe arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,\n\ninitial_state: initial state, type array[] real,\ninitial_time: initial time, type int or real,\ntimes: solution times, type array[] real,\ntheta: parameters, type array[] real,\ndata x_r: real data, type array[] real, data only, and\ndata x_i: integer data, type array[] int, data only.\n\nFor more fine-grained control of the ODE solvers, these parameters can also be provided:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\ndata max_num_steps: maximum number of steps to take in the ODE solver, type int, data only.\n\n\n\nThe return value for the ODE solvers is an array of type array[,] real, with values consisting of solutions at the specified times.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,\nsolution times and number of rows of the return value of the solver,\nparameters, real data and integer data passed to the solver will be passed to the system function",
+    "text": "The following functions are commonly used as link functions in generalized linear models. The function \\(\\Phi\\) is also commonly used as a link function (see section probability-related functions).\n \n\nR logit(T x) The log odds, or logit, function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_logit(T x) The logistic sigmoid function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_cloglog(T x) The inverse of the complementary log-log function applied to x\nAvailable since 2.0, vectorized in 2.13",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Deprecated Functions"
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/deprecated_functions.html#functions-old-algebra-solver",
-    "href": "functions-reference/deprecated_functions.html#functions-old-algebra-solver",
-    "title": "Deprecated Functions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#Phi-function",
+    "href": "functions-reference/real-valued_basic_functions.html#Phi-function",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "These algebraic solver functions have been replaced by those described in Algebraic Equation Solvers..\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:\n vector algebra_system(vector y, vector theta,\n                              data array[] real x_r, array[] int x_i)\nThe algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.\nThe argument of this function are:\n\ny, the unknowns we wish to solve for\ntheta, parameter values used to evaluate the algebraic system\nx_r, data values used to evaluate the algebraic system\nx_i, integer data used to evaluate the algebraic system\n\nThe algebraic system function separates parameter values, theta, from data values, x_r, for efficiency in propagating the derivatives through the algebraic system.\n\n\n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.\nAvailable since 2.17, deprecated in 2.31\n \n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.\nAvailable since 2.17, deprecated in 2.31\nNote: In future releases, the function algebra_solver will be deprecated and replaced with algebra_solver_powell.\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.24, deprecated in 2.31\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.24, deprecated in 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, vector, array[] real, array[] int):vector. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\ntheta: parameters only, type vector,\nx_r: real data only, type array[] real, and\nx_i: integer data only, type array[] int.\n\nFor more fine-grained control of the algebraic solver, these parameters can also be provided:\n\nrel_tol: relative tolerance for the algebraic solver, type real, data only,\nfunction_tol: function tolerance for the algebraic solver, type real, data only,\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only.\n\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0.\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\nThe parameters, real data, and integer data will be passed from the solver directly to the system function.",
+    "text": "The error function erf is related to the standard normal cumulative distribution function \\(\\Phi\\) by scaling. See section normal distribution for the general normal cumulative distribution function (and its complement).\n \n\nR erf(T x) The error function, also known as the Gauss error function, of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR erfc(T x) The complementary error function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_erfc(T x) The inverse of the complementary error function of x\nAvailable since 2.29, vectorized in 2.29\n \n\nR Phi(T x) The standard normal cumulative distribution function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_Phi(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The details of the algorithm can be found in (Wichura 1988). Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.\nAvailable since 2.0, vectorized in 2.13\n \n\nR Phi_approx(T x) The fast approximation of the unit (may replace Phi for probit regression with maximum absolute error of 0.00014, see (Bowling et al. 2009) for details)\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal binary_log_loss(int y, real y_hat) Return the log loss function for for predicting \\(\\hat{y} \\in [0,1]\\) for boolean outcome \\(y \\in \\{0,1\\}\\). \\[\\begin{equation*}\n\\mathrm{binary\\_log\\_loss}(y,\\hat{y}) = \\begin{cases} -\\log \\hat{y} &\n\\text{if } y = 1\\\\ -\\log (1 - \\hat{y}) & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR binary_log_loss(T1 x, T2 y) Vectorized implementation of the binary_log_loss function\nAvailable since 2.25\n \n\nreal owens_t(real h, real a) Return the Owen’s T function for the probability of the event \\(X &gt; h\\) and \\(0&lt;Y&lt;aX\\) where X and Y are independent standard normal random variables. \\[\\begin{equation*} \\mathrm{owens\\_t}(h,a) = \\frac{1}{2\\pi} \\int_0^a\n\\frac{\\exp(-\\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \\end{equation*}\\]\nAvailable since 2.25\n \n\nR owens_t(T1 x, T2 y) Vectorized implementation of the owens_t function\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Deprecated Functions"
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/correlation_matrix_distributions.html",
-    "href": "functions-reference/correlation_matrix_distributions.html",
-    "title": "Correlation Matrix Distributions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#betafun",
+    "href": "functions-reference/real-valued_basic_functions.html#betafun",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "The correlation matrix distributions have support on the (Cholesky factors of) correlation matrices. A Cholesky factor \\(L\\) for a \\(K\n\\times K\\) correlation matrix \\(\\Sigma\\) of dimension \\(K\\) has rows of unit length so that the diagonal of \\(L L^{\\top}\\) is the unit \\(K\\)-vector. Even though models are usually conceptualized in terms of correlation matrices, it is better to operationalize them in terms of their Cholesky factors. If you are interested in the posterior distribution of the correlations, you can recover them in the generated quantities block via\n generated quantities {\n   corr_matrix[K] Sigma;\n   Sigma = multiply_lower_tri_self_transpose(L);\n }\n\n\n\n\nFor \\(\\eta &gt; 0\\), if \\(\\Sigma\\) a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*}\n\\text{LkjCorr}(\\Sigma|\\eta) \\propto \\det \\left( \\Sigma \\right)^{(\\eta\n- 1)}. \\end{equation*}\\] The expectation is the identity matrix for any positive value of the shape parameter \\(\\eta\\), which can be interpreted like the shape parameter of a symmetric beta distribution:\n\nif \\(\\eta = 1\\), then the density is uniform over correlation matrices of order \\(K\\);\nif \\(\\eta &gt; 1\\), the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger \\(\\eta\\); and\nfor \\(0 &lt; \\eta &lt; 1\\), the density has a trough at the identity matrix.\nif \\(\\eta\\) were an unknown parameter, the Jeffreys prior is proportional to \\(\\sqrt{2\\sum_{k=1}^{K-1}\\left(\n\\psi_1\\left(\\eta+\\frac{K-k-1}{2}\\right) -   2\\psi_1\\left(2\\eta+K-k-1\n\\right)\\right)}\\), where \\(\\psi_1()\\) is the trigamma function\n\nSee (Lewandowski, Kurowicka, and Joe 2009) for definitions. However, it is much better computationally to work directly with the Cholesky factor of \\(\\Sigma\\), so this distribution should never be explicitly used in practice.\n\n\n\ny ~ lkj_corr(eta)\nIncrement target log probability density with lkj_corr_lupdf(y | eta).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal lkj_corr_lpdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta. lkj_corr_cholesky_lpdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.12\n \n\nreal lkj_corr_lupdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. lkj_corr_cholesky_lupdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.25\n \n\nmatrix lkj_corr_rng(int K, real eta) Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n\n\n\n\nStan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if L is a Cholesky factor of a correlation matrix, then\n L ~ lkj_corr_cholesky(2.0); # implies L * L' ~ lkj_corr(2.0);\nBecause Stan requires models to have support on all valid constrained parameters, L will almost always1 be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,\n parameters {   cholesky_factor_corr[K] L;   # rather than corr_matrix[K] Sigma;   // ...\n\n\nFor \\(\\eta &gt; 0\\), if \\(L\\) is a \\(K \\times K\\) lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*} \\text{LkjCholesky}(L|\\eta)\n\\propto \\left|J\\right|\\det(L L^\\top)^{(\\eta - 1)} = \\prod_{k=2}^K\nL_{kk}^{K-k+2\\eta-2}. \\end{equation*}\\] See the previous section for details on interpreting the shape parameter \\(\\eta\\). Note that even if \\(\\eta=1\\), it is still essential to evaluate the density function because the density of \\(L\\) is not constant, regardless of the value of \\(\\eta\\), even though the density of \\(LL^\\top\\) is constant iff \\(\\eta=1\\).\nA lower triangular \\(L\\) is a Cholesky factor for a correlation matrix if and only if \\(L_{k,k} &gt; 0\\) for \\(k \\in 1{:}K\\) and each row \\(L_k\\) has unit Euclidean length.\n\n\n\nL ~ lkj_corr_cholesky(eta)\nIncrement target log probability density with lkj_corr_cholesky_lupdf(L | eta).\nAvailable since 2.4\n \n\n\n\n\n \n\nreal lkj_corr_cholesky_lpdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta\nAvailable since 2.12\n \n\nreal lkj_corr_cholesky_lupdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms\nAvailable since 2.25\n \n\nmatrix lkj_corr_cholesky_rng(int K, real eta) Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.4",
+    "text": "real beta(real alpha, real beta) Return the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.25\n \n\nR beta(T1 x, T2 y) Vectorized implementation of the beta function\nAvailable since 2.25\n \n\nreal inc_beta(real alpha, real beta, real x) Return the regularized incomplete beta function up to x applied to alpha and beta. See section appendix for a definition.\nAvailable since 2.10\n \n\nreal inv_inc_beta(real alpha, real beta, real p) Return the inverse of the regularized incomplete beta function. The return value x is the value that solves p = inc_beta(alpha, beta, x). See section appendix for a definition of the inc_beta.\nAvailable since 2.30\n \n\nreal lbeta(real alpha, real beta) Return the natural logarithm of the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). \\[\\begin{equation*}\n\\text{lbeta}(\\alpha,\\beta) = \\log \\Gamma(\\alpha) + \\log \\Gamma(\\beta) - \\log \\Gamma(\\alpha+\\beta)\n\\end{equation*}\\] See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.0\n \n\nR lbeta(T1 x, T2 y) Vectorized implementation of the lbeta function\nAvailable since 2.25\n \n\nR tgamma(T x) The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that \\(\\Gamma(n+1) = n!\\). See for a full definition of \\(\\Gamma(x)\\). The function is defined for positive numbers and non-integral negative numbers,\nAvailable since 2.0, vectorized in 2.13\n \n\nR lgamma(T x) The natural logarithm of the gamma function applied to x,\nAvailable since 2.0, vectorized in 2.15\n \n\nR digamma(T x) The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers\nAvailable since 2.0, vectorized in 2.13\n \n\nR trigamma(T x) The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function\nAvailable since 2.0, vectorized in 2.13\n \n\nreal lmgamma(int n, real x) Return the natural logarithm of the multivariate gamma function \\(\\Gamma_n\\) with n dimensions applied to x. \\[\\begin{equation*}\n\\text{lmgamma}(n,x) =\n\\begin{cases} \\frac{n(n-1)}{4} \\log \\pi + \\sum_{j=1}^n \\log \\Gamma\\left(x + \\frac{1 - j}{2}\\right)\n& \\text{if } x\\not\\in \\{\\dots,-3,-2,-1,0\\}\\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR lmgamma(T1 x, T2 y) Vectorized implementation of the lmgamma function\nAvailable since 2.25\n \n\nreal gamma_p(real a, real z) Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_p}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_0^zt^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_p(T1 x, T2 y) Vectorized implementation of the gamma_p function\nAvailable since 2.25\n \n\nreal gamma_q(real a, real z) Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_q}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_z^\\infty t^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\[6pt] \\textrm{error} & \\text{otherwise}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_q(T1 x, T2 y) Vectorized implementation of the gamma_q function\nAvailable since 2.25\n \n\nint choose(int x, int y) Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” In its the antilog of the lchoose function but returns an integer rather than a real number with no non-zero decimal places. For \\(0 \\leq y \\leq x\\), the binomial coefficient function can be defined via the factorial function \\[\\begin{equation*}\n\\text{choose}(x,y) = \\frac{x!}{\\left(y!\\right)\\left(x - y\\right)!}.\n\\end{equation*}\\]\nAvailable since 2.14\n \n\nR choose(T1 x, T2 y) Vectorized implementation of the choose function\nAvailable since 2.25\n \n\nreal bessel_first_kind(int v, real x) Return the Bessel function of the first kind with order v applied to x. \\[\\begin{equation*}\n\\mathrm{bessel\\_first\\_kind}(v,x) = J_v(x),\n\\end{equation*}\\] where \\[\\begin{equation*}\nJ_v(x)=\\left(\\frac{1}{2}x\\right)^v \\sum_{k=0}^\\infty\n\\frac{\\left(-\\frac{1}{4}x^2\\right)^k}{k!\\, \\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_first_kind(T1 x, T2 y) Vectorized implementation of the bessel_first_kind function\nAvailable since 2.25\n \n\nreal bessel_second_kind(int v, real x) Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For \\(x,v &gt; 0\\), \\[\\begin{equation*}\n\\mathrm{bessel\\_second\\_kind}(v,x) =\n\\begin{cases} Y_v(x) & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\nY_v(x)=\\frac{J_v(x)\\cos(v\\pi)-J_{-v}(x)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_second_kind(T1 x, T2 y) Vectorized implementation of the bessel_second_kind function\nAvailable since 2.25\n \n\nreal modified_bessel_first_kind(int v, real z) Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_first\\_kind}(v,z) = I_v(z)\n\\end{equation*}\\] where \\[\\begin{equation*}\n{I_v}(z) = \\left(\\frac{1}{2}z\\right)^v\\sum_{k=0}^\\infty \\frac{\\left(\\frac{1}{4}z^2\\right)^k}{k!\\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_first_kind function\nAvailable since 2.25\n \n\nreal log_modified_bessel_first_kind(real v, real z) Return the log of the modified Bessel function of the first kind. v does not have to be an integer.\nAvailable since 2.26\n \n\nR log_modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the log_modified_bessel_first_kind function\nAvailable since 2.26\n \n\nreal modified_bessel_second_kind(int v, real z) Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_second\\_kind}(v,z) =\n\\begin{cases} K_v(z) &  \\text{if } z &gt; 0 \\\\ \\textrm{error} & \\text{if } z \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} {K_v}(z) = \\frac{\\pi}{2}\\cdot\\frac{I_{-v}(z) - I_{v}(z)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_second_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_second_kind function\nAvailable since 2.25\n \n\nreal falling_factorial(real x, real n) Return the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*}\n\\mathrm{falling\\_factorial}(x,n) =\n\\begin{cases} (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\n(x)_n=\\frac{\\Gamma(x+1)}{\\Gamma(x-n+1)}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR falling_factorial(T1 x, T2 y) Vectorized implementation of the falling_factorial function\nAvailable since 2.25\n \n\nreal lchoose(real x, real y) Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For \\(0 \\leq y \\leq x\\), \\[\\begin{equation*} \\mathrm{binomial\\_coefficient\\_log}(x,y) =\n\\log\\Gamma(x+1) - \\log\\Gamma(y+1) - \\log\\Gamma(x-y+1). \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lchoose(T1 x, T2 y) Vectorized implementation of the lchoose function\nAvailable since 2.29\n \n\nreal log_falling_factorial(real x, real n) Return the log of the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_falling\\_factorial}(x,n) =\n\\begin{cases} \\log (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal rising_factorial(real x, int n) Return the rising factorial of x with power n defined for positive x and integer n. \\[\\begin{equation*}\n\\mathrm{rising\\_factorial}(x,n) = \\begin{cases} x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} x^{(n)}=\\frac{\\Gamma(x+n)}{\\Gamma(x)} \\end{equation*}\\]\nAvailable since 2.20\n \n\nR rising_factorial(T1 x, T2 y) Vectorized implementation of the rising_factorial function\nAvailable since 2.25\n \n\nreal log_rising_factorial(real x, real n) Return the log of the rising factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_rising\\_factorial}(x,n) =\n\\begin{cases} \\log x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR log_rising_factorial(T1 x, T2 y) Vectorized implementation of the log_rising_factorial function\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Correlation Matrix Distributions"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/correlation_matrix_distributions.html#lkj-correlation",
-    "href": "functions-reference/correlation_matrix_distributions.html#lkj-correlation",
-    "title": "Correlation Matrix Distributions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#composed-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#composed-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "For \\(\\eta &gt; 0\\), if \\(\\Sigma\\) a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*}\n\\text{LkjCorr}(\\Sigma|\\eta) \\propto \\det \\left( \\Sigma \\right)^{(\\eta\n- 1)}. \\end{equation*}\\] The expectation is the identity matrix for any positive value of the shape parameter \\(\\eta\\), which can be interpreted like the shape parameter of a symmetric beta distribution:\n\nif \\(\\eta = 1\\), then the density is uniform over correlation matrices of order \\(K\\);\nif \\(\\eta &gt; 1\\), the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger \\(\\eta\\); and\nfor \\(0 &lt; \\eta &lt; 1\\), the density has a trough at the identity matrix.\nif \\(\\eta\\) were an unknown parameter, the Jeffreys prior is proportional to \\(\\sqrt{2\\sum_{k=1}^{K-1}\\left(\n\\psi_1\\left(\\eta+\\frac{K-k-1}{2}\\right) -   2\\psi_1\\left(2\\eta+K-k-1\n\\right)\\right)}\\), where \\(\\psi_1()\\) is the trigamma function\n\nSee (Lewandowski, Kurowicka, and Joe 2009) for definitions. However, it is much better computationally to work directly with the Cholesky factor of \\(\\Sigma\\), so this distribution should never be explicitly used in practice.\n\n\n\ny ~ lkj_corr(eta)\nIncrement target log probability density with lkj_corr_lupdf(y | eta).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal lkj_corr_lpdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta. lkj_corr_cholesky_lpdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.12\n \n\nreal lkj_corr_lupdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. lkj_corr_cholesky_lupdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.25\n \n\nmatrix lkj_corr_rng(int K, real eta) Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "text": "The functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.\n \n\nR expm1(T x) The natural exponential of x minus 1\nAvailable since 2.0, vectorized in 2.13\n \n\nreal fma(real x, real y, real z) Return z plus the result of x multiplied by y. \\[\\begin{equation*} \\text{fma}(x,y,z) =\n(x \\times y) + z \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal ldexp(real x, int y) Return the product of x and two raised to the y power. \\[\\begin{equation*}\n\\text{ldexp}(x,y) = x 2^y  \\end{equation*}\\]\nAvailable since 2.25\n \n\nR ldexp(T1 x, T2 y) Vectorized implementation of the ldexp function\nAvailable since 2.25\n \n\nreal lmultiply(real x, real y) Return the product of x and the natural logarithm of y. \\[\\begin{equation*}\n\\text{lmultiply}(x,y) = \\begin{cases} 0 & \\text{if } x = y = 0 \\\\ x\n\\log y & \\text{if } x, y \\neq 0 \\\\ \\text{NaN} & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lmultiply(T1 x, T2 y) Vectorized implementation of the lmultiply function\nAvailable since 2.25\n \n\nR log1p(T x) The natural logarithm of 1 plus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m(T x) The natural logarithm of 1 minus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1p_exp(T x) The natural logarithm of one plus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m_exp(T x) The logarithm of one minus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal log_diff_exp(real x, real y) Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_diff\\_exp}(x,y) = \\begin{cases} \\log(\\exp(x)-\\exp(y)) &\n\\text{if } +\\infty &gt; x \\ge y \\\\[6pt]\n\\textrm{NaN} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nWhen x is equal to y, log_diff_exp(x, y) returns \\(-\\infty\\), consistent with log(0) returning \\(-\\infty\\). This includes the case in which x and y are both equal to \\(-\\infty\\), which corresponds to log(0 - 0) because exp(negative_infinity()) returns 0.\nAvailable since 2.0\n \n\nR log_diff_exp(T1 x, T2 y) Vectorized implementation of the log_diff_exp function\nAvailable since 2.25\n \n\nreal log_mix(real theta, real lp1, real lp2) Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by \\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda_1, \\lambda_2) & = & \\log \\!\\left(\n\\theta \\exp(\\lambda_1) + \\left( 1 - \\theta \\right) \\exp(\\lambda_2)\n\\right) \\\\[3pt] & = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) +\n\\lambda_1, \\ \\log(1 - \\theta) + \\lambda_2\\right). \\end{eqnarray*}\\]\nAvailable since 2.6\n \n\nR log_mix(T1 thetas, T2 lps)\nCalculates the log mixture density given thetas, mixing proportions which should be between 0 and 1 and sum to 1, and lps, log densities. The lps variable must be either a 1-d container of the same length as thetas, or an array of such.\n\\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda)\n& = & \\log \\!\\left( \\sum_{n=1}^N \\theta_n * \\exp(\\lambda_n) \\right) \\\\[3pt]\n& = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) + \\lambda\\right).\n\\end{eqnarray*}\\]\nThis is a generalization of the above signature of three arguments to more than two densities. For example, log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2}).\nAvailable since 2.26\n \n\nR log_sum_exp(T1 x, T2 y) Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_sum\\_exp}(x,y) = \\log(\\exp(x)+\\exp(y)) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.33\n \n\nR log_inv_logit(T x) The natural logarithm of the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log_inv_logit_diff(T1 x, T2 y) The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y\nAvailable since 2.25\n \n\nR log1m_inv_logit(T x) The natural logarithm of 1 minus the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Correlation Matrix Distributions"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/correlation_matrix_distributions.html#cholesky-lkj-correlation-distribution",
-    "href": "functions-reference/correlation_matrix_distributions.html#cholesky-lkj-correlation-distribution",
-    "title": "Correlation Matrix Distributions",
+    "objectID": "functions-reference/real-valued_basic_functions.html#special-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#special-functions",
+    "title": "Real-Valued Basic Functions",
     "section": "",
-    "text": "Stan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if L is a Cholesky factor of a correlation matrix, then\n L ~ lkj_corr_cholesky(2.0); # implies L * L' ~ lkj_corr(2.0);\nBecause Stan requires models to have support on all valid constrained parameters, L will almost always1 be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,\n parameters {   cholesky_factor_corr[K] L;   # rather than corr_matrix[K] Sigma;   // ...\n\n\nFor \\(\\eta &gt; 0\\), if \\(L\\) is a \\(K \\times K\\) lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*} \\text{LkjCholesky}(L|\\eta)\n\\propto \\left|J\\right|\\det(L L^\\top)^{(\\eta - 1)} = \\prod_{k=2}^K\nL_{kk}^{K-k+2\\eta-2}. \\end{equation*}\\] See the previous section for details on interpreting the shape parameter \\(\\eta\\). Note that even if \\(\\eta=1\\), it is still essential to evaluate the density function because the density of \\(L\\) is not constant, regardless of the value of \\(\\eta\\), even though the density of \\(LL^\\top\\) is constant iff \\(\\eta=1\\).\nA lower triangular \\(L\\) is a Cholesky factor for a correlation matrix if and only if \\(L_{k,k} &gt; 0\\) for \\(k \\in 1{:}K\\) and each row \\(L_k\\) has unit Euclidean length.\n\n\n\nL ~ lkj_corr_cholesky(eta)\nIncrement target log probability density with lkj_corr_cholesky_lupdf(L | eta).\nAvailable since 2.4\n \n\n\n\n\n \n\nreal lkj_corr_cholesky_lpdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta\nAvailable since 2.12\n \n\nreal lkj_corr_cholesky_lupdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms\nAvailable since 2.25\n \n\nmatrix lkj_corr_cholesky_rng(int K, real eta) Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.4",
+    "text": "R lambert_w0(T x) Implementation of the \\(W_0\\) branch of the Lambert W function, i.e., solution to the function \\(W_0(x) \\exp^{ W_0(x)} = x\\)\nAvailable since 2.25\n \n\nR lambert_wm1(T x) Implementation of the \\(W_{-1}\\) branch of the Lambert W function, i.e., solution to the function \\(W_{-1}(x) \\exp^{W_{-1}(x)} = x\\)\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Correlation Matrix Distributions"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/correlation_matrix_distributions.html#footnotes",
-    "href": "functions-reference/correlation_matrix_distributions.html#footnotes",
-    "title": "Correlation Matrix Distributions",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nIt is possible to build up a valid L within Stan, but that would then require Jacobian adjustments to imply the intended posterior.↩︎",
+    "objectID": "functions-reference/real-valued_basic_functions.html#hypergeometric-functions",
+    "href": "functions-reference/real-valued_basic_functions.html#hypergeometric-functions",
+    "title": "Real-Valued Basic Functions",
+    "section": "",
+    "text": "Hypergeometric functions refer to a power series of the form \\[\\begin{equation*}\n_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \\sum_{n=0}^\\infty \\frac{(a_1)_n\\cdot\\cdot\\cdot(a_p)_n}{(b_1)_n\\cdot\\cdot\\cdot(b_q)_n} \\frac{z^n}{n!}\n\\end{equation*}\\] where \\((a)_n\\) is the Pochhammer symbol defined as \\((a)_n = \\frac{\\Gamma(a+n)}{\\Gamma(a)}\\).\nThe gradients of the hypergeometric function are given by: \\[\\begin{equation*}\n\\frac{\\partial }{\\partial a_1} =\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+a_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n    - \\psi\\left(a_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)\n\\end{equation*}\\] \\[\\begin{equation*}\n\\frac{\\partial }{\\partial b_1} =\n  \\psi\\left(b_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+b_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n\\end{equation*}\\] \\[\\begin{equation*}\n  \\frac{\\partial }{\\partial z} =\n  \\frac{\\prod_{j=1}^{p}a_j}{\\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)\n\\end{equation*}\\]\nStan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.\n \n\nreal hypergeometric_1F0(real a, real z) Special case of the hypergeometric function with \\(p=1\\) and \\(q=0\\).\nAvailable since 2.37\n \n\nreal hypergeometric_2F1(real a1, real a2, real b1, real z) Special case of the hypergeometric function with \\(p=2\\) and \\(q=1\\). If the function does not meet convergence criteria for given inputs, the function will attempt to apply Euler’s transformation to improve convergence: \\[\\begin{equation*}\n{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \\frac{z}{z-1})\\cdot(1-z)^{-a_2}\n\\end{equation*}\\]\nAvailable since 2.37\n \n\nreal hypergeometric_3F2(T1 a, T2 b, real z) Special case of the hypergeometric function with \\(p=3\\) and \\(q=2\\), where a and b are vectors of length 3 and 2, respectively.\nAvailable since 2.37\n \n\nreal hypergeometric_pFq(T1 a, T2 b, real z) Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.\nAvailable since 2.37",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Correlation Matrix Distributions"
+      "Built-in Functions",
+      "Real-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/continuous_distributions_on_0_1.html",
-    "href": "functions-reference/continuous_distributions_on_0_1.html",
-    "title": "Continuous Distributions on [0, 1]",
+    "objectID": "functions-reference/positive_continuous_distributions.html",
+    "href": "functions-reference/positive_continuous_distributions.html",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "The continuous distributions with outcomes in the interval \\([0,1]\\) are used to characterized bounded quantities, including probabilities.\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(\\theta \\in (0,1)\\), \\[\\begin{equation*} \\text{Beta}(\\theta|\\alpha,\\beta) =\n\\frac{1}{\\mathrm{B}(\\alpha,\\beta)} \\, \\theta^{\\alpha - 1} \\, (1 -\n\\theta)^{\\beta - 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, \\(\\alpha, \\beta &gt;\n0\\).\n\n\n\ntheta ~ beta(alpha, beta)\nIncrement target log probability density with beta_lupdf(theta | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_lpdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lupdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_cdf(reals theta | reals alpha, reals beta) The beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.0\n \n\nreal beta_lcdf(reals theta | reals alpha, reals beta) The log of the beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lccdf(reals theta | reals alpha, reals beta) The log of the beta complementary cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nR beta_rng(reals alpha, reals beta) Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in (0, 1)\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(\\theta\n\\in (0,1)\\), \\[\\begin{equation*} \\mathrm{Beta\\_Proportion}(\\theta|\\mu,\\kappa) =\n\\frac{1}{\\mathrm{B}(\\mu \\kappa, (1 - \\mu) \\kappa)} \\,\n\\theta^{\\mu\\kappa - 1} \\, (1 - \\theta)^{(1 - \\mu)\\kappa- 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires \\(\\mu \\in (0, 1)\\) and strictly positive parameter, \\(\\kappa &gt; 0\\).\n\n\n\ntheta ~ beta_proportion(mu, kappa)\nIncrement target log probability density with beta_proportion_lupdf(theta | mu, kappa).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal beta_proportion_lpdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.19\n \n\nreal beta_proportion_lupdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_proportion_lcdf(reals theta | reals mu, reals kappa) The log of the beta_proportion cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nreal beta_proportion_lccdf(reals theta | reals mu, reals kappa) The log of the beta_proportion complementary cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nR beta_proportion_rng(reals mu, reals kappa) Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "The positive continuous probability functions have support on the positive real numbers.\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}^+\\), \\[\\begin{equation*} \\text{LogNormal}(y|\\mu,\\sigma) = \\frac{1}{\\sqrt{2\n\\pi} \\ \\sigma} \\, \\frac{1}{y} \\ \\exp \\! \\left(        - \\, \\frac{1}{2}\n\\, \\left( \\frac{\\log y - \\mu}{\\sigma} \\right)^2      \\right) . \\end{equation*}\\]\n\n\n\ny ~ lognormal(mu, sigma)\nIncrement target log probability density with lognormal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal lognormal_lpdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lupdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal lognormal_cdf(reals y | reals mu, reals sigma) The cumulative lognormal distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal lognormal_lcdf(reals y | reals mu, reals sigma) The log of the lognormal cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lccdf(reals y | reals mu, reals sigma) The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR lognormal_rng(reals mu, reals sigma) Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.22\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{ChiSquare}(y|\\nu) = \\frac{2^{-\\nu/2}}     {\\Gamma(\\nu / 2)} \\,\ny^{\\nu/2 - 1} \\, \\exp \\! \\left( -\\, \\frac{1}{2} \\, y \\right) . \\end{equation*}\\]\n\n\n\ny ~ chi_square(nu)\nIncrement target log probability density with chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal chi_square_lpdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lupdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal chi_square_cdf(reals y | reals nu) The Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal chi_square_lcdf(reals y | reals nu) The log of the Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lccdf(reals y | reals nu) The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR chi_square_rng(reals nu) Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{InvChiSquare}(y \\, | \\, \\nu) = \\frac{2^{-\\nu/2}}    {\\Gamma(\\nu\n/ 2)} \\, y^{-\\nu/2 - 1} \\, \\exp\\! \\left( \\! - \\, \\frac{1}{2} \\,\n\\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_chi_square(nu)\nIncrement target log probability density with inv_chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_chi_square_lpdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lupdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_chi_square_cdf(reals y | reals nu) The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal inv_chi_square_lcdf(reals y | reals nu) The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lccdf(reals y | reals nu) The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR inv_chi_square_rng(reals nu) Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{ScaledInvChiSquare}(y|\\nu,\\sigma) =\n\\frac{(\\nu / 2)^{\\nu/2}}      {\\Gamma(\\nu / 2)} \\, \\sigma^{\\nu} \\,\ny^{-(\\nu/2 + 1)} \\, \\exp \\! \\left( \\!    - \\, \\frac{1}{2} \\, \\nu \\,\n\\sigma^2 \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ scaled_inv_chi_square(nu, sigma)\nIncrement target log probability density with scaled_inv_chi_square_lupdf(y | nu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal scaled_inv_chi_square_lpdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lupdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal scaled_inv_chi_square_cdf(reals y | reals nu, reals sigma) The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.0\n \n\nreal scaled_inv_chi_square_lcdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lccdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nR scaled_inv_chi_square_rng(reals nu, reals sigma) Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf inverse scale (rate) \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Exponential}(y|\\beta) = \\beta \\, \\exp ( - \\beta \\, y ) . \\end{equation*}\\]\n\n\n\ny ~ exponential(beta)\nIncrement target log probability density with exponential_lupdf(y | beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exponential_lpdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lupdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal exponential_cdf(reals y | reals beta) The exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.0\n \n\nreal exponential_lcdf(reals y | reals beta) The log of the exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lccdf(reals y | reals beta) The log of the exponential complementary cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nR exponential_rng(reals beta) Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf the shape parameter \\(\\alpha \\in \\mathbb{R}^+\\) and the rate (or inverse scale) parameter \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Gamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\, y^{\\alpha - 1}\n\\exp(-\\beta \\, y) . \\end{equation*}\\]\nUnder the shape and rate formulation of the Gamma distribution, \\(\\mathbb{E}[y] = \\alpha / \\beta\\) and \\(\\textrm{var}[y] = \\alpha / \\beta^2\\).\n\n\n\ny ~ gamma(alpha, beta)\nIncrement target log probability density with gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gamma_lpdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lupdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gamma_cdf(reals y | reals alpha, reals beta) The cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal gamma_lcdf(reals y | reals alpha, reals beta) The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lccdf(reals y | reals alpha, reals beta) The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR gamma_rng(reals alpha, reals beta) Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{InvGamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\ y^{-(\\alpha + 1)} \\,\n\\exp \\! \\left( \\! - \\beta \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_gamma(alpha, beta)\nIncrement target log probability density with inv_gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_gamma_lpdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lupdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_gamma_cdf(reals y | reals alpha, reals beta) The inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.0\n \n\nreal inv_gamma_lcdf(reals y | reals alpha, reals beta) The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lccdf(reals y | reals alpha, reals beta) The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nR inv_gamma_rng(reals alpha, reals beta) Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*} \\text{Weibull}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Weibull}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Frechet}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ weibull(alpha, sigma)\nIncrement target log probability density with weibull_lupdf(y | alpha, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal weibull_lpdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lupdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal weibull_cdf(reals y | reals alpha, reals sigma) The Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.0\n \n\nreal weibull_lcdf(reals y | reals alpha, reals sigma) The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lccdf(reals y | reals alpha, reals sigma) The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR weibull_rng(reals alpha, reals sigma) Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Frechet}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Frechet}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Weibull}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ frechet(alpha, sigma)\nIncrement target log probability density with frechet_lupdf(y | alpha, sigma).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal frechet_lpdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lupdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal frechet_cdf(reals y | reals alpha, reals sigma) The Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.5\n \n\nreal frechet_lcdf(reals y | reals alpha, reals sigma) The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lccdf(reals y | reals alpha, reals sigma) The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR frechet_rng(reals alpha, reals sigma) Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*}\n\\text{Rayleigh}(y|\\sigma) = \\frac{y}{\\sigma^2} \\exp(-y^2 / 2\\sigma^2)\n\\!. \\end{equation*}\\]\n\n\n\ny ~ rayleigh(sigma)\nIncrement target log probability density with rayleigh_lupdf(y | sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal rayleigh_lpdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lupdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal rayleigh_cdf(real y | real sigma) The Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.0\n \n\nreal rayleigh_lcdf(real y | real sigma) The log of the Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lccdf(real y | real sigma) The log of the Rayleigh complementary cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nR rayleigh_rng(reals sigma) Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha, \\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Log-Logistic}(y|\\alpha,\\beta) =\n\\frac{\\ \\left(\\frac{\\beta}{\\alpha}\\right) \\left(\\frac{y}{\\alpha}\\right)^{\\beta-1}\\ }{\\left(1 + \\left(\\frac{y}{\\alpha}\\right)^\\beta\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ loglogistic(alpha, beta)\nIncrement target log probability density with unnormalized version of loglogistic_lpdf(y | alpha, beta)\nAvailable since 2.29\n \n\n\n\n\n \n\nreal loglogistic_lpdf(reals y | reals alpha, reals beta) The log of the log-logistic density of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nreal loglogistic_cdf(reals y | reals alpha, reals beta) The log-logistic cumulative distribution function of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nR loglogistic_rng(reals alpha, reals beta) Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.29",
     "crumbs": [
       "Functions Reference",
       "Continuous Distributions",
-      "Continuous Distributions on [0, 1]"
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/continuous_distributions_on_0_1.html#beta-distribution",
-    "href": "functions-reference/continuous_distributions_on_0_1.html#beta-distribution",
-    "title": "Continuous Distributions on [0, 1]",
+    "objectID": "functions-reference/positive_continuous_distributions.html#lognormal",
+    "href": "functions-reference/positive_continuous_distributions.html#lognormal",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(\\theta \\in (0,1)\\), \\[\\begin{equation*} \\text{Beta}(\\theta|\\alpha,\\beta) =\n\\frac{1}{\\mathrm{B}(\\alpha,\\beta)} \\, \\theta^{\\alpha - 1} \\, (1 -\n\\theta)^{\\beta - 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, \\(\\alpha, \\beta &gt;\n0\\).\n\n\n\ntheta ~ beta(alpha, beta)\nIncrement target log probability density with beta_lupdf(theta | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_lpdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lupdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_cdf(reals theta | reals alpha, reals beta) The beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.0\n \n\nreal beta_lcdf(reals theta | reals alpha, reals beta) The log of the beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lccdf(reals theta | reals alpha, reals beta) The log of the beta complementary cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nR beta_rng(reals alpha, reals beta) Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}^+\\), \\[\\begin{equation*} \\text{LogNormal}(y|\\mu,\\sigma) = \\frac{1}{\\sqrt{2\n\\pi} \\ \\sigma} \\, \\frac{1}{y} \\ \\exp \\! \\left(        - \\, \\frac{1}{2}\n\\, \\left( \\frac{\\log y - \\mu}{\\sigma} \\right)^2      \\right) . \\end{equation*}\\]\n\n\n\ny ~ lognormal(mu, sigma)\nIncrement target log probability density with lognormal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal lognormal_lpdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lupdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal lognormal_cdf(reals y | reals mu, reals sigma) The cumulative lognormal distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal lognormal_lcdf(reals y | reals mu, reals sigma) The log of the lognormal cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lccdf(reals y | reals mu, reals sigma) The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR lognormal_rng(reals mu, reals sigma) Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.22",
     "crumbs": [
       "Functions Reference",
       "Continuous Distributions",
-      "Continuous Distributions on [0, 1]"
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/continuous_distributions_on_0_1.html#beta-proportion-distribution",
-    "href": "functions-reference/continuous_distributions_on_0_1.html#beta-proportion-distribution",
-    "title": "Continuous Distributions on [0, 1]",
+    "objectID": "functions-reference/positive_continuous_distributions.html#chi-square-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#chi-square-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "If \\(\\mu \\in (0, 1)\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(\\theta\n\\in (0,1)\\), \\[\\begin{equation*} \\mathrm{Beta\\_Proportion}(\\theta|\\mu,\\kappa) =\n\\frac{1}{\\mathrm{B}(\\mu \\kappa, (1 - \\mu) \\kappa)} \\,\n\\theta^{\\mu\\kappa - 1} \\, (1 - \\theta)^{(1 - \\mu)\\kappa- 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires \\(\\mu \\in (0, 1)\\) and strictly positive parameter, \\(\\kappa &gt; 0\\).\n\n\n\ntheta ~ beta_proportion(mu, kappa)\nIncrement target log probability density with beta_proportion_lupdf(theta | mu, kappa).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal beta_proportion_lpdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.19\n \n\nreal beta_proportion_lupdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_proportion_lcdf(reals theta | reals mu, reals kappa) The log of the beta_proportion cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nreal beta_proportion_lccdf(reals theta | reals mu, reals kappa) The log of the beta_proportion complementary cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nR beta_proportion_rng(reals mu, reals kappa) Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{ChiSquare}(y|\\nu) = \\frac{2^{-\\nu/2}}     {\\Gamma(\\nu / 2)} \\,\ny^{\\nu/2 - 1} \\, \\exp \\! \\left( -\\, \\frac{1}{2} \\, y \\right) . \\end{equation*}\\]\n\n\n\ny ~ chi_square(nu)\nIncrement target log probability density with chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal chi_square_lpdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lupdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal chi_square_cdf(reals y | reals nu) The Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal chi_square_lcdf(reals y | reals nu) The log of the Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lccdf(reals y | reals nu) The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR chi_square_rng(reals nu) Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
       "Continuous Distributions",
-      "Continuous Distributions on [0, 1]"
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html",
-    "href": "functions-reference/complex_matrix_operations.html",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#inverse-chi-square-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#inverse-chi-square-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.\n\n\nIf an expression e can be assigned to a variable of type T, then it can be used as an argument to a function that is specified to take arguments of type T. For instance, sqrt(real) is specified to take a real argument, but an integer expression such as 2 + 2 of type int can be passed to sqrt, so that sqrt(2 + 2) is well defined. This works by promoting the integer expression 2 + 2 to be of real type.\nThe rules for promotion in Stan are simple:\n\nint may be promoted to real,\nreal may be promoted to complex,\nvector can be promoted to complex_vector,\nrow_vector can be promoted to complex_row_vector,\nmatrix can be promoted to complex_matrix,\nif T can be promoted to U and U can be promoted to V, then T can be promoted to V (transitive), and\nif T can be promoted to U, then T[] can be promoted to U[] (covariant).\n\n\n\n\nWhen a function is called, the definition requiring the fewest number of promotions is used. For example, when calling vector + vector, the real-valued signature is used. When calling any of complex_vector + vector, vector + complex_vector, or complex_vector + complex_vector, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.\n\n\n\nComplex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, complex operator+(complex_vector, complex_vector). Through promotion, operator+ may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures operator+(complex_vector, vector) and operator+(vector, complex_vector).\n\n\n\nGeneric functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions append_array, dims, head, num_elements, rep_array, reverse, segment, size, and tail.\n\n\n\n\n \n\nint num_elements(complex_vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.30\n \n\nint num_elements(complex_row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.30\n \n\nint num_elements(complex_matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.30\n \n\nint rows(complex_vector x) The number of rows in the vector x\nAvailable since 2.30\n \n\nint rows(complex_row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.30\n \n\nint rows(complex_matrix x) The number of rows in the matrix x\nAvailable since 2.30\n \n\nint cols(complex_vector x) The number of columns in the vector x, namely 1\nAvailable since 2.30\n \n\nint cols(complex_row_vector x) The number of columns in the row vector x\nAvailable since 2.30\n \n\nint cols(complex_matrix x) The number of columns in the matrix x\nAvailable since 2.30\n \n\nint size(complex_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(complex_row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15.\nAvailable since 2.30\n\n\n\nStan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\ncomplex_vector operator-(complex_vector x) The negation of the vector x.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x) The negation of the row vector x.\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x) The negation of the matrix x.\nAvailable since 2.30\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex_vector y) The sum of the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex_row_vector y) The sum of the row vectors x and y.\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex_matrix y) The sum of the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex_vector y) The difference between the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex_row_vector y) The difference between the row vectors x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex_matrix y) The difference between the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex x, complex_vector y) The product of the scalar x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex x, complex_row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex x, complex_matrix y) The product of the scalar x and the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_vector x, complex y) The product of the scalar y and vector x\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_vector x, complex_row_vector y) The product of the vector x and row vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex y) The product of the scalar y and row vector x\nAvailable since 2.30\n \n\ncomplex operator*(complex_row_vector x, complex_vector y) The product of the row vector x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex_matrix y) The product of the row vector x and matrix y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex y) The product of the scalar y and matrix x\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_matrix x, complex_vector y) The product of the matrix x and vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex_matrix y) The product of the matrices x and y\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex y) The result of adding y to every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator+(complex x, complex_vector y) The result of adding x to every entry in the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex y) The result of adding y to every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex x, complex_row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex y) The result of adding y to every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex x, complex_matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex y) The result of subtracting y from every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex x, complex_vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex x, complex_row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex x, complex_matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator/(complex_vector x, complex y) The result of dividing each entry in the vector x by y\nAvailable since 2.30\n \n\ncomplex_row_vector operator/(complex_row_vector x, complex y) The result of dividing each entry in the row vector x by y\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix x, complex y) The result of dividing each entry in the matrix x by y\nAvailable since 2.30\n\n\n\n\nComplex complex_matrix transposition is represented using a postfix operator.\n \n\ncomplex_matrix operator'(complex_matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.30\n \n\ncomplex_row_vector operator'(complex_vector x) The transpose of the vector x, written as x'\nAvailable since 2.30\n \n\ncomplex_vector operator'(complex_row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.30\n\n\n\nAs in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.\n \n\ncomplex_vector operator.*(complex_vector x, complex_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator.*(complex_row_vector x, complex_row_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator.*(complex_matrix x, complex_matrix y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^(complex x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n\n\n\n \n\ncomplex dot_product(complex_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_vector x, complex_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_matrix x, complex_matrix y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_vector x, complex_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_matrix x, complex_matrix y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex dot_self(complex_vector x) The dot product of the vector x with itself\nAvailable since 2.30\n \n\ncomplex dot_self(complex_row_vector x) The dot product of the row vector x with itself\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_matrix x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_matrix x) The dot product of the rows of x with themselves\nAvailable since 2.30\n\n\n \n\ncomplex_matrix diag_pre_multiply(complex_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_pre_multiply(complex_row_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_row_vector v) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex sum(complex_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n\n\n\n\nMuch like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.\n\n\nThese functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a real type. For example, get_real(complex_matrix M) yields a matrix containing the real component of each value in M.\nThe following table contains examples of what this notation can mean:\n\n\n\nType T\nType T_demoted\n\n\n\n\ncomplex\nreal\n\n\ncomplex_vector\nvector\n\n\ncomplex_row_vector\nrow_vector\n\n\ncomplex_matrix\nmatrix\n\n\narray[] complex\narray[] real\n\n\narray[,,] complex\narray[,,] real\n\n\n\n\n\n\n \n\nT_demoted get_real(T x) Given an object of complex type T, return the same shape object but of type real by getting the real component of each element of x.\nAvailable since 2.30\n \n\nT_demoted get_imag(T x) Given an object of complex type T, return the same shape object but of type real by getting the imaginary component of each element of x.\nAvailable since 2.30\nFor example, given the Stan declaration\n  complex_vector[2] z = [3+4i, 5+6i]';\nA call get_real(z) will yield the vector [3, 5]', and a call get_imag(z) will yield the vector [4, 6]'.\n\n\n\n\nThe following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\ncomplex_vector rep_vector(complex z, int m) Return the size m (column) vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_row_vector rep_row_vector(complex z, int n) Return the size n row vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex z, int m, int n) Return the m by n matrix consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.30\n\n\n \n\ncomplex_matrix symmetrize_from_lower_tri(complex_matrix A) Construct a symmetric matrix from the lower triangle of A.\nAvailable since 2.30\n\n\n\n\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.30\n \n\ncomplex_vector diagonal(complex_matrix x) The diagonal of the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix diag_matrix(complex_vector x) The diagonal matrix with diagonal x\nAvailable since 2.30\n\n\n\nStan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\ncomplex_vector col(complex_matrix x, int n) The n-th column of matrix x\nAvailable since 2.30\n \n\ncomplex_row_vector row(complex_matrix x, int m) The m-th row of matrix x\nAvailable since 2.30\n\n\n\n\n\n \n\ncomplex_matrix block(complex_matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.30\n \n\ncomplex_vector sub_col(complex_matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.30\n \n\ncomplex_row_vector sub_row(complex_matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector head(complex_vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector head(complex_row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector tail(complex_vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector tail(complex_row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector segment(complex_vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n \n\ncomplex_row_vector segment(complex_row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n\n\n\n\n\n\n\n \n\ncomplex_matrix append_col(complex_matrix x, complex_matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_matrix x, complex_vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex_row_vector y) Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex x, complex_row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex y) Append y to the end of x, returning another row vector.\nAvailable since 2.30\n\n\n\n \n\ncomplex_matrix append_row(complex_matrix x, complex_matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_matrix x, complex_row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex_vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex x, complex_vector y) Append x to the top of y, returning another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex y) Append y to the bottom of x, returning another vector.\nAvailable since 2.30\n\n\n\n\n\n\nStan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors u and v,\n    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v\nand in the 2-dimensional case for complex matrices A and B,\n    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B\nAlthough the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.\n \n\ncomplex_vector fft(complex_vector v) Return the discrete Fourier transform of the specified complex vector v. If \\(v \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(u =\n\\textrm{fft}(v)\\), then \\[\\begin{equation*}\nu_n = \\sum_{m &lt; n}\nv_m \\cdot\n\\exp\\left(\\frac{-n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\]\nAvailable since 2.30\n \n\ncomplex_matrix fft2(complex_matrix m) Return the 2D discrete Fourier transform of the specified complex matrix m. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.\nAvailable since 2.30\n \n\ncomplex_vector inv_fft(complex_vector u) Return the inverse of the discrete Fourier transform of the specified complex vector u. The inverse FFT (this function) is scaled so that fft(inv_fft(u)) == u. If \\(u \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(v = \\textrm{fft}^{-1}(u)\\), then \\[\\begin{equation*}\nv_n =  \\frac{1}{N} \\sum_{m &lt; n}\nu_m \\cdot\n\\exp\\left(\\frac{n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\] This only differs from the FFT by the sign inside the exponential and the scaling. The \\(\\frac{1}{N}\\) scaling ensures that fft(inv_fft(u)) == u and inv_fft(fft(v)) == v for complex vectors u and v.\nAvailable since 2.30\n \n\ncomplex_matrix inv_fft2(complex_matrix m) Return the inverse of the 2D discrete Fourier transform of the specified complex matrix m. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures fft2(inv_fft2(A)) == A and inv_fft2(fft2(B)) == B.\nAvailable since 2.30\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] complex cumulative_sum(array[] complex x) The cumulative sum of x\nAvailable since 2.30\n \n\ncomplex_vector cumulative_sum(complex_vector v) The cumulative sum of v\nAvailable since 2.30\n \n\ncomplex_row_vector cumulative_sum(complex_row_vector rv) The cumulative sum of rv\nAvailable since 2.30\n\n\n\n\n\n\nIn general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.\n\n\n \n\ncomplex_row_vector operator/(complex_row_vector b, complex_matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix B, complex_matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex trace(complex_matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector eigenvalues(complex_matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.32\n \n\ncomplex_matrix eigenvectors(complex_matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.32\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\ncomplex_vector eigenvalues_sym(complex_matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors_sym(complex_matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose_sym(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(complex_matrix A) The singular values of A in descending order\nAvailable since 2.30\n \n\ncomplex_matrix svd_U(complex_matrix A) The left-singular vectors of A\nAvailable since 2.30\n \n\ncomplex_matrix svd_V(complex_matrix A) The right-singular vectors of A\nAvailable since 2.30\n \n\ntuple(complex_matrix, vector, complex_matrix) svd(complex_matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\nThe complex Schur decomposition of a square matrix \\(A\\) produces a complex unitary matrix \\(U\\) and a complex upper-triangular Schur form matrix \\(T\\) such that \\[A = U \\cdot T \\cdot U^{-1}\\]\nSince \\(U\\) is unitary, its inverse is also its conjugate transpose, \\(U^{-1} = U^*\\), \\(U^*(i, j) = \\mathrm{conj}(U(j, i))\\)\n \n\ncomplex_matrix complex_schur_decompose_t(matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_t(complex_matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(complex_matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results. This overload is equivalent to complex_schur_decompose(to_complex(A,0)) but is more efficient.\nAvailable since 2.33\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(complex_matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n\n\n\n\n\n \n\ncomplex_vector reverse(complex_vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.30\n \n\ncomplex_row_vector reverse(complex_row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.30",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{InvChiSquare}(y \\, | \\, \\nu) = \\frac{2^{-\\nu/2}}    {\\Gamma(\\nu\n/ 2)} \\, y^{-\\nu/2 - 1} \\, \\exp\\! \\left( \\! - \\, \\frac{1}{2} \\,\n\\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_chi_square(nu)\nIncrement target log probability density with inv_chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_chi_square_lpdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lupdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_chi_square_cdf(reals y | reals nu) The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal inv_chi_square_lcdf(reals y | reals nu) The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lccdf(reals y | reals nu) The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR inv_chi_square_rng(reals nu) Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-promotion",
-    "href": "functions-reference/complex_matrix_operations.html#complex-promotion",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#scaled-inverse-chi-square-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#scaled-inverse-chi-square-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.\n\n\nIf an expression e can be assigned to a variable of type T, then it can be used as an argument to a function that is specified to take arguments of type T. For instance, sqrt(real) is specified to take a real argument, but an integer expression such as 2 + 2 of type int can be passed to sqrt, so that sqrt(2 + 2) is well defined. This works by promoting the integer expression 2 + 2 to be of real type.\nThe rules for promotion in Stan are simple:\n\nint may be promoted to real,\nreal may be promoted to complex,\nvector can be promoted to complex_vector,\nrow_vector can be promoted to complex_row_vector,\nmatrix can be promoted to complex_matrix,\nif T can be promoted to U and U can be promoted to V, then T can be promoted to V (transitive), and\nif T can be promoted to U, then T[] can be promoted to U[] (covariant).\n\n\n\n\nWhen a function is called, the definition requiring the fewest number of promotions is used. For example, when calling vector + vector, the real-valued signature is used. When calling any of complex_vector + vector, vector + complex_vector, or complex_vector + complex_vector, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.\n\n\n\nComplex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, complex operator+(complex_vector, complex_vector). Through promotion, operator+ may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures operator+(complex_vector, vector) and operator+(vector, complex_vector).\n\n\n\nGeneric functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions append_array, dims, head, num_elements, rep_array, reverse, segment, size, and tail.",
+    "text": "If \\(\\nu \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{ScaledInvChiSquare}(y|\\nu,\\sigma) =\n\\frac{(\\nu / 2)^{\\nu/2}}      {\\Gamma(\\nu / 2)} \\, \\sigma^{\\nu} \\,\ny^{-(\\nu/2 + 1)} \\, \\exp \\! \\left( \\!    - \\, \\frac{1}{2} \\, \\nu \\,\n\\sigma^2 \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ scaled_inv_chi_square(nu, sigma)\nIncrement target log probability density with scaled_inv_chi_square_lupdf(y | nu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal scaled_inv_chi_square_lpdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lupdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal scaled_inv_chi_square_cdf(reals y | reals nu, reals sigma) The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.0\n \n\nreal scaled_inv_chi_square_lcdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lccdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nR scaled_inv_chi_square_rng(reals nu, reals sigma) Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#integer-valued-complex-matrix-size-functions",
-    "href": "functions-reference/complex_matrix_operations.html#integer-valued-complex-matrix-size-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#exponential-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#exponential-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "int num_elements(complex_vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.30\n \n\nint num_elements(complex_row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.30\n \n\nint num_elements(complex_matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.30\n \n\nint rows(complex_vector x) The number of rows in the vector x\nAvailable since 2.30\n \n\nint rows(complex_row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.30\n \n\nint rows(complex_matrix x) The number of rows in the matrix x\nAvailable since 2.30\n \n\nint cols(complex_vector x) The number of columns in the vector x, namely 1\nAvailable since 2.30\n \n\nint cols(complex_row_vector x) The number of columns in the row vector x\nAvailable since 2.30\n \n\nint cols(complex_matrix x) The number of columns in the matrix x\nAvailable since 2.30\n \n\nint size(complex_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(complex_row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15.\nAvailable since 2.30",
+    "text": "If inverse scale (rate) \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Exponential}(y|\\beta) = \\beta \\, \\exp ( - \\beta \\, y ) . \\end{equation*}\\]\n\n\n\ny ~ exponential(beta)\nIncrement target log probability density with exponential_lupdf(y | beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exponential_lpdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lupdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal exponential_cdf(reals y | reals beta) The exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.0\n \n\nreal exponential_lcdf(reals y | reals beta) The log of the exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lccdf(reals y | reals beta) The log of the exponential complementary cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nR exponential_rng(reals beta) Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-matrix-arithmetic-operators",
-    "href": "functions-reference/complex_matrix_operations.html#complex-matrix-arithmetic-operators",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#gamma-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#gamma-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "Stan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\ncomplex_vector operator-(complex_vector x) The negation of the vector x.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x) The negation of the row vector x.\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x) The negation of the matrix x.\nAvailable since 2.30\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex_vector y) The sum of the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex_row_vector y) The sum of the row vectors x and y.\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex_matrix y) The sum of the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex_vector y) The difference between the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex_row_vector y) The difference between the row vectors x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex_matrix y) The difference between the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex x, complex_vector y) The product of the scalar x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex x, complex_row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex x, complex_matrix y) The product of the scalar x and the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_vector x, complex y) The product of the scalar y and vector x\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_vector x, complex_row_vector y) The product of the vector x and row vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex y) The product of the scalar y and row vector x\nAvailable since 2.30\n \n\ncomplex operator*(complex_row_vector x, complex_vector y) The product of the row vector x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex_matrix y) The product of the row vector x and matrix y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex y) The product of the scalar y and matrix x\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_matrix x, complex_vector y) The product of the matrix x and vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex_matrix y) The product of the matrices x and y\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex y) The result of adding y to every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator+(complex x, complex_vector y) The result of adding x to every entry in the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex y) The result of adding y to every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex x, complex_row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex y) The result of adding y to every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex x, complex_matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex y) The result of subtracting y from every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex x, complex_vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex x, complex_row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex x, complex_matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator/(complex_vector x, complex y) The result of dividing each entry in the vector x by y\nAvailable since 2.30\n \n\ncomplex_row_vector operator/(complex_row_vector x, complex y) The result of dividing each entry in the row vector x by y\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix x, complex y) The result of dividing each entry in the matrix x by y\nAvailable since 2.30",
+    "text": "If the shape parameter \\(\\alpha \\in \\mathbb{R}^+\\) and the rate (or inverse scale) parameter \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Gamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\, y^{\\alpha - 1}\n\\exp(-\\beta \\, y) . \\end{equation*}\\]\nUnder the shape and rate formulation of the Gamma distribution, \\(\\mathbb{E}[y] = \\alpha / \\beta\\) and \\(\\textrm{var}[y] = \\alpha / \\beta^2\\).\n\n\n\ny ~ gamma(alpha, beta)\nIncrement target log probability density with gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gamma_lpdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lupdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gamma_cdf(reals y | reals alpha, reals beta) The cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal gamma_lcdf(reals y | reals alpha, reals beta) The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lccdf(reals y | reals alpha, reals beta) The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR gamma_rng(reals alpha, reals beta) Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-transposition-operator",
-    "href": "functions-reference/complex_matrix_operations.html#complex-transposition-operator",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#inverse-gamma-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#inverse-gamma-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "Complex complex_matrix transposition is represented using a postfix operator.\n \n\ncomplex_matrix operator'(complex_matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.30\n \n\ncomplex_row_vector operator'(complex_vector x) The transpose of the vector x, written as x'\nAvailable since 2.30\n \n\ncomplex_vector operator'(complex_row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.30",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{InvGamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\ y^{-(\\alpha + 1)} \\,\n\\exp \\! \\left( \\! - \\beta \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_gamma(alpha, beta)\nIncrement target log probability density with inv_gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_gamma_lpdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lupdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_gamma_cdf(reals y | reals alpha, reals beta) The inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.0\n \n\nreal inv_gamma_lcdf(reals y | reals alpha, reals beta) The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lccdf(reals y | reals alpha, reals beta) The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nR inv_gamma_rng(reals alpha, reals beta) Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-elementwise-functions",
-    "href": "functions-reference/complex_matrix_operations.html#complex-elementwise-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#weibull-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#weibull-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "As in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.\n \n\ncomplex_vector operator.*(complex_vector x, complex_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator.*(complex_row_vector x, complex_row_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator.*(complex_matrix x, complex_matrix y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^(complex x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*} \\text{Weibull}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Weibull}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Frechet}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ weibull(alpha, sigma)\nIncrement target log probability density with weibull_lupdf(y | alpha, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal weibull_lpdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lupdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal weibull_cdf(reals y | reals alpha, reals sigma) The Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.0\n \n\nreal weibull_lcdf(reals y | reals alpha, reals sigma) The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lccdf(reals y | reals alpha, reals sigma) The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR weibull_rng(reals alpha, reals sigma) Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#dot-products-and-specialized-products-for-complex-matrices",
-    "href": "functions-reference/complex_matrix_operations.html#dot-products-and-specialized-products-for-complex-matrices",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#frechet-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#frechet-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "complex dot_product(complex_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_vector x, complex_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_matrix x, complex_matrix y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_vector x, complex_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_matrix x, complex_matrix y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex dot_self(complex_vector x) The dot product of the vector x with itself\nAvailable since 2.30\n \n\ncomplex dot_self(complex_row_vector x) The dot product of the row vector x with itself\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_matrix x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_matrix x) The dot product of the rows of x with themselves\nAvailable since 2.30\n\n\n \n\ncomplex_matrix diag_pre_multiply(complex_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_pre_multiply(complex_row_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_row_vector v) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.30",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Frechet}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Frechet}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Weibull}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ frechet(alpha, sigma)\nIncrement target log probability density with frechet_lupdf(y | alpha, sigma).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal frechet_lpdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lupdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal frechet_cdf(reals y | reals alpha, reals sigma) The Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.5\n \n\nreal frechet_lcdf(reals y | reals alpha, reals sigma) The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lccdf(reals y | reals alpha, reals sigma) The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR frechet_rng(reals alpha, reals sigma) Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-reductions",
-    "href": "functions-reference/complex_matrix_operations.html#complex-reductions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#rayleigh-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#rayleigh-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "complex sum(complex_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30",
+    "text": "If \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*}\n\\text{Rayleigh}(y|\\sigma) = \\frac{y}{\\sigma^2} \\exp(-y^2 / 2\\sigma^2)\n\\!. \\end{equation*}\\]\n\n\n\ny ~ rayleigh(sigma)\nIncrement target log probability density with rayleigh_lupdf(y | sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal rayleigh_lpdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lupdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal rayleigh_cdf(real y | real sigma) The Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.0\n \n\nreal rayleigh_lcdf(real y | real sigma) The log of the Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lccdf(real y | real sigma) The log of the Rayleigh complementary cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nR rayleigh_rng(reals sigma) Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#vectorized-accessor-functions",
-    "href": "functions-reference/complex_matrix_operations.html#vectorized-accessor-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/positive_continuous_distributions.html#log-logistic-distribution",
+    "href": "functions-reference/positive_continuous_distributions.html#log-logistic-distribution",
+    "title": "Positive Continuous Distributions",
     "section": "",
-    "text": "Much like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.\n\n\nThese functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a real type. For example, get_real(complex_matrix M) yields a matrix containing the real component of each value in M.\nThe following table contains examples of what this notation can mean:\n\n\n\nType T\nType T_demoted\n\n\n\n\ncomplex\nreal\n\n\ncomplex_vector\nvector\n\n\ncomplex_row_vector\nrow_vector\n\n\ncomplex_matrix\nmatrix\n\n\narray[] complex\narray[] real\n\n\narray[,,] complex\narray[,,] real\n\n\n\n\n\n\n \n\nT_demoted get_real(T x) Given an object of complex type T, return the same shape object but of type real by getting the real component of each element of x.\nAvailable since 2.30\n \n\nT_demoted get_imag(T x) Given an object of complex type T, return the same shape object but of type real by getting the imaginary component of each element of x.\nAvailable since 2.30\nFor example, given the Stan declaration\n  complex_vector[2] z = [3+4i, 5+6i]';\nA call get_real(z) will yield the vector [3, 5]', and a call get_imag(z) will yield the vector [4, 6]'.",
+    "text": "If \\(\\alpha, \\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Log-Logistic}(y|\\alpha,\\beta) =\n\\frac{\\ \\left(\\frac{\\beta}{\\alpha}\\right) \\left(\\frac{y}{\\alpha}\\right)^{\\beta-1}\\ }{\\left(1 + \\left(\\frac{y}{\\alpha}\\right)^\\beta\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ loglogistic(alpha, beta)\nIncrement target log probability density with unnormalized version of loglogistic_lpdf(y | alpha, beta)\nAvailable since 2.29\n \n\n\n\n\n \n\nreal loglogistic_lpdf(reals y | reals alpha, reals beta) The log of the log-logistic density of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nreal loglogistic_cdf(reals y | reals alpha, reals beta) The log-logistic cumulative distribution function of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nR loglogistic_rng(reals alpha, reals beta) Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.29",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Continuous Distributions",
+      "Positive Continuous Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-broadcast-functions",
-    "href": "functions-reference/complex_matrix_operations.html#complex-broadcast-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/mixed_operations.html",
+    "href": "functions-reference/mixed_operations.html",
+    "title": "Mixed Operations",
     "section": "",
-    "text": "The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\ncomplex_vector rep_vector(complex z, int m) Return the size m (column) vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_row_vector rep_row_vector(complex z, int n) Return the size n row vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex z, int m, int n) Return the m by n matrix consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.30\n\n\n \n\ncomplex_matrix symmetrize_from_lower_tri(complex_matrix A) Construct a symmetric matrix from the lower triangle of A.\nAvailable since 2.30",
+    "text": "Mixed Operations\nThese functions perform conversions between Stan containers matrix, vector, row vector and arrays.\n \n\nmatrix to_matrix(matrix m) Return the matrix m itself.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_matrix m) Return the matrix m itself.\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v) Convert the column vector v to a size(v) by 1 matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_vector v) Convert the column vector v to a size(v) by 1 matrix.\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v) Convert the row vector v to a 1 by size(v) matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_row_vector v) Convert the row vector v to a 1 by size(v) matrix.\nAvailable since 2.30\n \n\nmatrix to_matrix(matrix M, int m, int n) Convert a matrix A to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_matrix M, int m, int n) Convert a matrix A to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v, int m, int n) Convert a vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_vector v, int m, int n) Convert a vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v, int m, int n) Convert a row_vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_row_vector v, int m, int n) Convert a row vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(matrix A, int m, int n, int col_major) Convert a matrix A to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_matrix A, int m, int n, int col_major) Convert a matrix A to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v, int m, int n, int col_major) Convert a vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_vector v, int m, int n, int col_major) Convert a vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v, int m, int n, int col_major) Convert a row vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_row_vector v, int m, int n, int col_major) Convert a row vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] real a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\nmatrix to_matrix(array[] int a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(array[] complex a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] real a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\nmatrix to_matrix(array[] int a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(array[] complex a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] row_vector vs) Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.\nAvailable since 2.28\n \n\ncomplex_matrix to_matrix(array[] complex_row_vector vs) Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.\nAvailable since 2.30\n \n\nmatrix to_matrix(array[,] real a) Convert the two dimensional array a to a matrix with the same dimensions and indexing order.\nAvailable since 2.3\n \n\nmatrix to_matrix(array[,] int a) Convert the two dimensional array a to a matrix with the same dimensions and indexing order. If any of the dimensions of a are zero, the result will be a \\(0 \\times 0\\) matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(array[,] complex a ) Convert the two dimensional array a to a matrix with the same dimensions and indexing order.\nAvailable since 2.30\n \n\nvector to_vector(matrix m) Convert the matrix m to a column vector in column-major order.\nAvailable since 2.0\n \n\ncomplex_vector to_vector(complex_matrix m) Convert the matrix m to a column vector in column-major order.\nAvailable since 2.30\n \n\nvector to_vector(vector v) Return the column vector v itself.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(complex_vector v) Return the column vector v itself.\nAvailable since 2.30\n \n\nvector to_vector(row_vector v) Convert the row vector v to a column vector.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(complex_row_vector v) Convert the row vector v to a column vector.\nAvailable since 2.30\n \n\nvector to_vector(array[] real a) Convert the one-dimensional array a to a column vector.\nAvailable since 2.3\n \n\nvector to_vector(array[] int a) Convert the one-dimensional integer array a to a column vector.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(array[] complex a) Convert the one-dimensional complex array a to a column vector.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(matrix m) Convert the matrix m to a row vector in column-major order.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_matrix m) Convert the matrix m to a row vector in column-major order.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(vector v) Convert the column vector v to a row vector.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_vector v) Convert the column vector v to a row vector.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(row_vector v) Return the row vector v itself.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_row_vector v) Return the row vector v itself.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(array[] real a) Convert the one-dimensional array a to a row vector.\nAvailable since 2.3\n \n\nrow_vector to_row_vector(array[] int a) Convert the one-dimensional array a to a row vector.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(array[] complex a) Convert the one-dimensional complex array a to a row vector.\nAvailable since 2.30\n \n\narray[,] real to_array_2d(matrix m) Convert the matrix m to a two dimensional array with the same dimensions and indexing order.\nAvailable since 2.3\n \n\narray[,] complex to_array_2d(complex_matrix m) Convert the matrix m to a two dimensional array with the same dimensions and indexing order.\nAvailable since 2.30\n \n\narray[] real to_array_1d(vector v) Convert the column vector v to a one-dimensional array.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(complex_vector v) Convert the column vector v to a one-dimensional array.\nAvailable since 2.30\n \n\narray[] real to_array_1d(row_vector v) Convert the row vector v to a one-dimensional array.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(complex_row_vector v) Convert the row vector v to a one-dimensional array.\nAvailable since 2.30\n \n\narray[] real to_array_1d(matrix m) Convert the matrix m to a one-dimensional array in column-major order.\nAvailable since 2.3\n \n\narray[] real to_array_1d(complex_matrix m) Convert the matrix m to a one-dimensional array in column-major order.\nAvailable since 2.30\n \n\narray[] real to_array_1d(array[...] real a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.3\n \n\narray[] int to_array_1d(array[...] int a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(array[...] complex a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.30\n\n\n\n\n Back to top",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Complex Matrix Operations"
+      "Mixed Operations"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#diagonal-complex-matrix-functions",
-    "href": "functions-reference/complex_matrix_operations.html#diagonal-complex-matrix-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/mathematical_functions.html",
+    "href": "functions-reference/mathematical_functions.html",
+    "title": "Mathematical Functions",
     "section": "",
-    "text": "complex_matrix add_diag(complex_matrix m, complex_row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.30\n \n\ncomplex_vector diagonal(complex_matrix x) The diagonal of the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix diag_matrix(complex_vector x) The diagonal matrix with diagonal x\nAvailable since 2.30",
+    "text": "This appendix provides the definition of several mathematical functions used throughout the manual.\n\n\nThe beta function, \\(\\text{B}(a, b)\\), computes the normalizing constant for the beta distribution, and is defined for \\(a &gt; 0\\) and \\(b\n&gt; 0\\) by \\[\\begin{equation*} \\text{B}(a,b) \\ = \\ \\int_0^1 u^{a - 1} (1 - u)^{b - 1} \\,\ndu \\ = \\ \\frac{\\Gamma(a) \\, \\Gamma(b)}{\\Gamma(a+b)} \\, , \\end{equation*}\\] where \\(\\Gamma(x)\\) is the Gamma function.\n\n\n\nThe incomplete beta function, \\(\\text{B}(x; a, b)\\), is defined for \\(x\n\\in [0, 1]\\) and \\(a, b \\geq 0\\) such that \\(a + b \\neq 0\\) by \\[\\begin{equation*}\n\\text{B}(x; \\, a, b) \\ = \\ \\int_0^x u^{a - 1} \\, (1 - u)^{b - 1} \\,\ndu, \\end{equation*}\\] where \\(\\text{B}(a, b)\\) is the beta function defined in appendix. If \\(x = 1\\), the incomplete beta function reduces to the beta function, \\(\\text{B}(1; a, b) = \\text{B}(a, b)\\).\nThe regularized incomplete beta function divides the incomplete beta function by the beta function, \\[\\begin{equation*} I_x(a, b) \\ = \\ \\frac{\\text{B}(x; \\,\na, b)}{B(a, b)} \\, . \\end{equation*}\\]\n\n\n\nThe gamma function, \\(\\Gamma(x)\\), is the generalization of the factorial function to continuous variables, defined so that for positive integers \\(n\\), \\[\\begin{equation*} \\Gamma(n+1) = n! \\end{equation*}\\] Generalizing to all positive numbers and non-integer negative numbers, \\[\\begin{equation*} \\Gamma(x) =\n\\int_0^{\\infty} u^{x - 1} \\exp(-u) \\, du. \\end{equation*}\\]\n\n\n\nThe digamma function \\(\\Psi\\) is the derivative of the \\(\\log \\Gamma\\) function, \\[\\begin{equation*}\n\\Psi(u) \\ = \\ \\frac{d}{d u} \\log \\Gamma(u) \\ = \\ \\frac{1}{\\Gamma(u)} \\ \\frac{d}{d u} \\Gamma(u).\n\\end{equation*}\\]",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Appendix",
+      "Mathematical Functions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#slicing-and-blocking-functions-for-complex-matrices",
-    "href": "functions-reference/complex_matrix_operations.html#slicing-and-blocking-functions-for-complex-matrices",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/mathematical_functions.html#beta-appendix",
+    "href": "functions-reference/mathematical_functions.html#beta-appendix",
+    "title": "Mathematical Functions",
     "section": "",
-    "text": "Stan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\ncomplex_vector col(complex_matrix x, int n) The n-th column of matrix x\nAvailable since 2.30\n \n\ncomplex_row_vector row(complex_matrix x, int m) The m-th row of matrix x\nAvailable since 2.30\n\n\n\n\n\n \n\ncomplex_matrix block(complex_matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.30\n \n\ncomplex_vector sub_col(complex_matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.30\n \n\ncomplex_row_vector sub_row(complex_matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector head(complex_vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector head(complex_row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector tail(complex_vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector tail(complex_row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector segment(complex_vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n \n\ncomplex_row_vector segment(complex_row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30",
+    "text": "The beta function, \\(\\text{B}(a, b)\\), computes the normalizing constant for the beta distribution, and is defined for \\(a &gt; 0\\) and \\(b\n&gt; 0\\) by \\[\\begin{equation*} \\text{B}(a,b) \\ = \\ \\int_0^1 u^{a - 1} (1 - u)^{b - 1} \\,\ndu \\ = \\ \\frac{\\Gamma(a) \\, \\Gamma(b)}{\\Gamma(a+b)} \\, , \\end{equation*}\\] where \\(\\Gamma(x)\\) is the Gamma function.",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Appendix",
+      "Mathematical Functions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-matrix-concatenation",
-    "href": "functions-reference/complex_matrix_operations.html#complex-matrix-concatenation",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/mathematical_functions.html#inc-beta-appendix",
+    "href": "functions-reference/mathematical_functions.html#inc-beta-appendix",
+    "title": "Mathematical Functions",
     "section": "",
-    "text": "complex_matrix append_col(complex_matrix x, complex_matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_matrix x, complex_vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex_row_vector y) Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex x, complex_row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex y) Append y to the end of x, returning another row vector.\nAvailable since 2.30\n\n\n\n \n\ncomplex_matrix append_row(complex_matrix x, complex_matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_matrix x, complex_row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex_vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex x, complex_vector y) Append x to the top of y, returning another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex y) Append y to the bottom of x, returning another vector.\nAvailable since 2.30",
+    "text": "The incomplete beta function, \\(\\text{B}(x; a, b)\\), is defined for \\(x\n\\in [0, 1]\\) and \\(a, b \\geq 0\\) such that \\(a + b \\neq 0\\) by \\[\\begin{equation*}\n\\text{B}(x; \\, a, b) \\ = \\ \\int_0^x u^{a - 1} \\, (1 - u)^{b - 1} \\,\ndu, \\end{equation*}\\] where \\(\\text{B}(a, b)\\) is the beta function defined in appendix. If \\(x = 1\\), the incomplete beta function reduces to the beta function, \\(\\text{B}(1; a, b) = \\text{B}(a, b)\\).\nThe regularized incomplete beta function divides the incomplete beta function by the beta function, \\[\\begin{equation*} I_x(a, b) \\ = \\ \\frac{\\text{B}(x; \\,\na, b)}{B(a, b)} \\, . \\end{equation*}\\]",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Appendix",
+      "Mathematical Functions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-special-matrix-functions",
-    "href": "functions-reference/complex_matrix_operations.html#complex-special-matrix-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/mathematical_functions.html#gamma-appendix",
+    "href": "functions-reference/mathematical_functions.html#gamma-appendix",
+    "title": "Mathematical Functions",
     "section": "",
-    "text": "Stan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors u and v,\n    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v\nand in the 2-dimensional case for complex matrices A and B,\n    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B\nAlthough the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.\n \n\ncomplex_vector fft(complex_vector v) Return the discrete Fourier transform of the specified complex vector v. If \\(v \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(u =\n\\textrm{fft}(v)\\), then \\[\\begin{equation*}\nu_n = \\sum_{m &lt; n}\nv_m \\cdot\n\\exp\\left(\\frac{-n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\]\nAvailable since 2.30\n \n\ncomplex_matrix fft2(complex_matrix m) Return the 2D discrete Fourier transform of the specified complex matrix m. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.\nAvailable since 2.30\n \n\ncomplex_vector inv_fft(complex_vector u) Return the inverse of the discrete Fourier transform of the specified complex vector u. The inverse FFT (this function) is scaled so that fft(inv_fft(u)) == u. If \\(u \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(v = \\textrm{fft}^{-1}(u)\\), then \\[\\begin{equation*}\nv_n =  \\frac{1}{N} \\sum_{m &lt; n}\nu_m \\cdot\n\\exp\\left(\\frac{n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\] This only differs from the FFT by the sign inside the exponential and the scaling. The \\(\\frac{1}{N}\\) scaling ensures that fft(inv_fft(u)) == u and inv_fft(fft(v)) == v for complex vectors u and v.\nAvailable since 2.30\n \n\ncomplex_matrix inv_fft2(complex_matrix m) Return the inverse of the 2D discrete Fourier transform of the specified complex matrix m. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures fft2(inv_fft2(A)) == A and inv_fft2(fft2(B)) == B.\nAvailable since 2.30\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] complex cumulative_sum(array[] complex x) The cumulative sum of x\nAvailable since 2.30\n \n\ncomplex_vector cumulative_sum(complex_vector v) The cumulative sum of v\nAvailable since 2.30\n \n\ncomplex_row_vector cumulative_sum(complex_row_vector rv) The cumulative sum of rv\nAvailable since 2.30",
+    "text": "The gamma function, \\(\\Gamma(x)\\), is the generalization of the factorial function to continuous variables, defined so that for positive integers \\(n\\), \\[\\begin{equation*} \\Gamma(n+1) = n! \\end{equation*}\\] Generalizing to all positive numbers and non-integer negative numbers, \\[\\begin{equation*} \\Gamma(x) =\n\\int_0^{\\infty} u^{x - 1} \\exp(-u) \\, du. \\end{equation*}\\]",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Appendix",
+      "Mathematical Functions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#complex-linear-algebra-functions",
-    "href": "functions-reference/complex_matrix_operations.html#complex-linear-algebra-functions",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/mathematical_functions.html#digamma-appendix",
+    "href": "functions-reference/mathematical_functions.html#digamma-appendix",
+    "title": "Mathematical Functions",
     "section": "",
-    "text": "In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.\n\n\n \n\ncomplex_row_vector operator/(complex_row_vector b, complex_matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix B, complex_matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex trace(complex_matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector eigenvalues(complex_matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.32\n \n\ncomplex_matrix eigenvectors(complex_matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.32\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\ncomplex_vector eigenvalues_sym(complex_matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors_sym(complex_matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose_sym(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(complex_matrix A) The singular values of A in descending order\nAvailable since 2.30\n \n\ncomplex_matrix svd_U(complex_matrix A) The left-singular vectors of A\nAvailable since 2.30\n \n\ncomplex_matrix svd_V(complex_matrix A) The right-singular vectors of A\nAvailable since 2.30\n \n\ntuple(complex_matrix, vector, complex_matrix) svd(complex_matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\nThe complex Schur decomposition of a square matrix \\(A\\) produces a complex unitary matrix \\(U\\) and a complex upper-triangular Schur form matrix \\(T\\) such that \\[A = U \\cdot T \\cdot U^{-1}\\]\nSince \\(U\\) is unitary, its inverse is also its conjugate transpose, \\(U^{-1} = U^*\\), \\(U^*(i, j) = \\mathrm{conj}(U(j, i))\\)\n \n\ncomplex_matrix complex_schur_decompose_t(matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_t(complex_matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(complex_matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results. This overload is equivalent to complex_schur_decompose(to_complex(A,0)) but is more efficient.\nAvailable since 2.33\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(complex_matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33",
+    "text": "The digamma function \\(\\Psi\\) is the derivative of the \\(\\log \\Gamma\\) function, \\[\\begin{equation*}\n\\Psi(u) \\ = \\ \\frac{d}{d u} \\log \\Gamma(u) \\ = \\ \\frac{1}{\\Gamma(u)} \\ \\frac{d}{d u} \\Gamma(u).\n\\end{equation*}\\]",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Appendix",
+      "Mathematical Functions"
     ]
   },
   {
-    "objectID": "functions-reference/complex_matrix_operations.html#reverse-functions-for-complex-matrices",
-    "href": "functions-reference/complex_matrix_operations.html#reverse-functions-for-complex-matrices",
-    "title": "Complex Matrix Operations",
+    "objectID": "functions-reference/index.html",
+    "href": "functions-reference/index.html",
+    "title": "Stan Functions Reference",
     "section": "",
-    "text": "complex_vector reverse(complex_vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.30\n \n\ncomplex_row_vector reverse(complex_row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.30",
+    "text": "This is the reference for the functions defined in the Stan math library and available in the Stan programming language.\nFor more information the Stan language and inference engines and how to use Stan for Bayesian inference, see\n\nthe Stan User’s Guide. The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:\nthe Stan Reference Manual. Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.\n\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Complex Matrix Operations"
+      "Stan Functions Reference"
     ]
   },
   {
-    "objectID": "functions-reference/circular_distributions.html",
-    "href": "functions-reference/circular_distributions.html",
-    "title": "Circular Distributions",
+    "objectID": "functions-reference/hidden_markov_models.html",
+    "href": "functions-reference/hidden_markov_models.html",
+    "title": "Hidden Markov Models",
     "section": "",
-    "text": "Circular distributions are defined for finite values y in any interval of length \\(2\\pi\\).\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{VonMises}(y|\\mu,\\kappa) =\n\\frac{\\exp(\\kappa\\cos(y-\\mu))}{2\\pi I_0(\\kappa)} \\!. \\end{equation*}\\] In order for this density to properly normalize, \\(y\\) must be restricted to some interval \\((c, c + 2\\pi)\\) of length \\(2 \\pi\\), because \\[\\begin{equation*} \\int_{c}^{c +\n2\\pi} \\text{VonMises}(y|\\mu,\\kappa) dy = 1. \\end{equation*}\\] Similarly, if \\(\\mu\\) is a parameter, it will typically be restricted to the same range as \\(y\\).\nIf \\(\\kappa &gt; 0\\), a von Mises distribution with its \\(2 \\pi\\) interval of support centered around its location \\(\\mu\\) will have a single mode at \\(\\mu\\); for example, restricting \\(y\\) to \\((-\\pi,\\pi)\\) and taking \\(\\mu = 0\\) leads to a single local optimum at the mode \\(\\mu\\). If the location \\(\\mu\\) is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean \\(\\mu\\).\nFor \\(\\kappa = 0\\), the Von Mises distribution corresponds to the circular uniform distribution with density \\(1 / (2 \\pi)\\) (independently of the values of \\(y\\) or \\(\\mu\\)).\n\n\n\ny ~ von_mises(mu, kappa)\nIncrement target log probability density with von_mises_lupdf(y | mu, kappa).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal von_mises_lpdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa.\nAvailable since 2.18\n \n\nreal von_mises_lupdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.\nAvailable since 2.25\n \n\nreal von_mises_cdf(reals y | reals mu, reals kappa) The von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lcdf(reals y | reals mu, reals kappa) The log of the von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lccdf(reals y | reals mu, reals kappa) The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nR von_mises_rng(reals mu, reals kappa) Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval \\([(\\mu \\mod 2\\pi)-\\pi,(\\mu \\mod\n2\\pi)+\\pi]\\)); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nEvaluating the Von Mises distribution for \\(\\kappa &gt; 100\\) is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as \\(\\kappa \\rightarrow \\infty\\), \\[\\begin{equation*}\n\\text{VonMises}(y|\\mu,\\kappa) \\rightarrow \\text{Normal}(\\mu, \\sqrt{1 /\n\\kappa}). \\end{equation*}\\] The workaround is to replace y ~ von_mises(mu,kappa) with\nif (kappa &lt; 100) {\n  y ~ von_mises(mu, kappa);\n} else {\n  y ~ normal(mu, sqrt(1 / kappa));\n}",
+    "text": "An elementary first-order Hidden Markov model is a probabilistic model over \\(N\\) observations, \\(y_n\\), and \\(N\\) hidden states, \\(x_n\\), which can be fully defined by the conditional distributions \\(p(y_n \\mid x_n, \\phi)\\) and \\(p(x_n \\mid x_{n - 1}, \\phi)\\). Here we make the dependency on additional model parameters, \\(\\phi\\), explicit. When \\(x\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(x\\) out.\nWhen each state \\(x\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can take advantage of the dependency structure to marginalize \\(x\\) and compute \\(p(y \\mid \\phi)\\). We start by defining the conditional observational distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid x_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n  \\Gamma_{ij} = p(x_n = j \\mid x_{n - 1} = i, \\phi).\n\\] Each row defines a probability distribution and must therefore be a simplex (i.e. its components must add to 1). Currently, Stan only supports stationary transitions where a single transition matrix is used for all transitions. Finally we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n  \\rho_k = p(x_0 = k \\mid \\phi).\n\\]\nThe Stan functions that support this type of model are special in that the user does not explicitly pass \\(y\\) and \\(\\phi\\) as arguments. Instead, the user passes \\(\\log \\omega\\), \\(\\Gamma\\), and \\(\\rho\\), which in turn depend on \\(y\\) and \\(\\phi\\).\n\n\n \n\nreal hmm_marginal(matrix log_omega, matrix Gamma, vector rho) Returns the log probability density of \\(y\\), with \\(x_n\\) integrated out at each iteration.\nAvailable since 2.24\nThe arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.\n\nlog_omega: \\(\\log \\omega_{kn} = \\log p(y_n \\mid x_n = k, \\phi)\\), log density of each output,\nGamma: \\(\\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \\phi)\\), the transition matrix,\nrho: \\(\\rho_k = p(x_0 = k \\mid \\phi)\\), the initial state probability.\n\n \n\narray[] int hmm_latent_rng(matrix log_omega, matrix Gamma, vector rho) Returns a length \\(N\\) array of integers over \\(\\{1, ..., K\\}\\), sampled from the joint posterior distribution of the hidden states, \\(p(x \\mid \\phi, y)\\). May be only used in transformed data and generated quantities.\nAvailable since 2.24\n \n\nmatrix hmm_hidden_state_prob(matrix log_omega, matrix Gamma, vector rho) Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a \\(K \\times N\\) matrix.    The \\(n^\\mathrm{th}\\) column is a simplex of probabilities for the \\(n^\\mathrm{th}\\) variable. Moreover, let \\(A\\) be the output. Then \\(A_{ij} = p(x_j = i \\mid \\phi, y)\\).  This function may only be used in transformed data and generated quantities.\nAvailable since 2.24",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Circular Distributions"
+      "Additional Distributions",
+      "Hidden Markov Models"
     ]
   },
   {
-    "objectID": "functions-reference/circular_distributions.html#von-mises-distribution",
-    "href": "functions-reference/circular_distributions.html#von-mises-distribution",
-    "title": "Circular Distributions",
+    "objectID": "functions-reference/hidden_markov_models.html#hmm-stan-functions",
+    "href": "functions-reference/hidden_markov_models.html#hmm-stan-functions",
+    "title": "Hidden Markov Models",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{VonMises}(y|\\mu,\\kappa) =\n\\frac{\\exp(\\kappa\\cos(y-\\mu))}{2\\pi I_0(\\kappa)} \\!. \\end{equation*}\\] In order for this density to properly normalize, \\(y\\) must be restricted to some interval \\((c, c + 2\\pi)\\) of length \\(2 \\pi\\), because \\[\\begin{equation*} \\int_{c}^{c +\n2\\pi} \\text{VonMises}(y|\\mu,\\kappa) dy = 1. \\end{equation*}\\] Similarly, if \\(\\mu\\) is a parameter, it will typically be restricted to the same range as \\(y\\).\nIf \\(\\kappa &gt; 0\\), a von Mises distribution with its \\(2 \\pi\\) interval of support centered around its location \\(\\mu\\) will have a single mode at \\(\\mu\\); for example, restricting \\(y\\) to \\((-\\pi,\\pi)\\) and taking \\(\\mu = 0\\) leads to a single local optimum at the mode \\(\\mu\\). If the location \\(\\mu\\) is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean \\(\\mu\\).\nFor \\(\\kappa = 0\\), the Von Mises distribution corresponds to the circular uniform distribution with density \\(1 / (2 \\pi)\\) (independently of the values of \\(y\\) or \\(\\mu\\)).\n\n\n\ny ~ von_mises(mu, kappa)\nIncrement target log probability density with von_mises_lupdf(y | mu, kappa).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal von_mises_lpdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa.\nAvailable since 2.18\n \n\nreal von_mises_lupdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.\nAvailable since 2.25\n \n\nreal von_mises_cdf(reals y | reals mu, reals kappa) The von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lcdf(reals y | reals mu, reals kappa) The log of the von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lccdf(reals y | reals mu, reals kappa) The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nR von_mises_rng(reals mu, reals kappa) Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval \\([(\\mu \\mod 2\\pi)-\\pi,(\\mu \\mod\n2\\pi)+\\pi]\\)); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nEvaluating the Von Mises distribution for \\(\\kappa &gt; 100\\) is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as \\(\\kappa \\rightarrow \\infty\\), \\[\\begin{equation*}\n\\text{VonMises}(y|\\mu,\\kappa) \\rightarrow \\text{Normal}(\\mu, \\sqrt{1 /\n\\kappa}). \\end{equation*}\\] The workaround is to replace y ~ von_mises(mu,kappa) with\nif (kappa &lt; 100) {\n  y ~ von_mises(mu, kappa);\n} else {\n  y ~ normal(mu, sqrt(1 / kappa));\n}",
+    "text": "real hmm_marginal(matrix log_omega, matrix Gamma, vector rho) Returns the log probability density of \\(y\\), with \\(x_n\\) integrated out at each iteration.\nAvailable since 2.24\nThe arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.\n\nlog_omega: \\(\\log \\omega_{kn} = \\log p(y_n \\mid x_n = k, \\phi)\\), log density of each output,\nGamma: \\(\\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \\phi)\\), the transition matrix,\nrho: \\(\\rho_k = p(x_0 = k \\mid \\phi)\\), the initial state probability.\n\n \n\narray[] int hmm_latent_rng(matrix log_omega, matrix Gamma, vector rho) Returns a length \\(N\\) array of integers over \\(\\{1, ..., K\\}\\), sampled from the joint posterior distribution of the hidden states, \\(p(x \\mid \\phi, y)\\). May be only used in transformed data and generated quantities.\nAvailable since 2.24\n \n\nmatrix hmm_hidden_state_prob(matrix log_omega, matrix Gamma, vector rho) Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a \\(K \\times N\\) matrix.    The \\(n^\\mathrm{th}\\) column is a simplex of probabilities for the \\(n^\\mathrm{th}\\) variable. Moreover, let \\(A\\) be the output. Then \\(A_{ij} = p(x_j = i \\mid \\phi, y)\\).  This function may only be used in transformed data and generated quantities.\nAvailable since 2.24",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Circular Distributions"
+      "Additional Distributions",
+      "Hidden Markov Models"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_continuous_distributions.html",
-    "href": "functions-reference/bounded_continuous_distributions.html",
-    "title": "Bounded Continuous Distributions",
+    "objectID": "functions-reference/embedded_laplace.html",
+    "href": "functions-reference/embedded_laplace.html",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "The bounded continuous probabilities have support on a finite interval of real numbers.\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in (\\alpha,\\infty)\\), then for \\(y \\in [\\alpha,\\beta]\\), \\[\\begin{equation*} \\text{Uniform}(y|\\alpha,\\beta) =\n\\frac{1}{\\beta - \\alpha} . \\end{equation*}\\]\n\n\n\ny ~ uniform(alpha, beta)\nIncrement target log probability density with uniform_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal uniform_lpdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lupdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal uniform_cdf(reals y | reals alpha, reals beta) The uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.0\n \n\nreal uniform_lcdf(reals y | reals alpha, reals beta) The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lccdf(reals y | reals alpha, reals beta) The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nR uniform_rng(reals alpha, reals beta) Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "The embedded Laplace approximation can be used to approximate certain marginal and conditional distributions that arise in latent Gaussian models. Embedded Laplace replaces explicit sampling of (high-dimensional) Gaussian latent variables with a local Gaussian approximation. In doing so, it marginalizes out the latent Gaussian variables. Inference can then be performed on the remaining, often low-dimensional, parameters. The embedded Laplace approximation in Stan is best suited for latent Gaussian models when jointly sampling over all model parameters is expensive and the conditional posterior of the Gaussian latent variables is reasonably close to Gaussian.\nFor observed data \\(y\\), latent Gaussian variables \\(\\theta\\), and hyperparameters \\(\\phi\\), a latent Gaussian model observes the following hierarchical structure: \\[\\begin{eqnarray}\n  \\phi &\\sim& p(\\phi), \\\\\n  \\theta &\\sim& \\text{MultiNormal}(0, K(\\phi)), \\\\\n  y &\\sim& p(y \\mid \\theta, \\phi).\n\\end{eqnarray}\\] In this formulation, \\(p(y \\mid \\theta, \\phi)\\) is the data model that specifies how observations are generated conditional on \\(\\theta\\) and \\(\\phi\\). \\(K(\\phi)\\) denotes the prior covariance matrix for the latent Gaussian variables \\(\\theta\\) and is parameterized by \\(\\phi\\). The prior on \\(\\theta\\) is centered at 0, however an offset can always be added when specifying the data model \\(p(y \\mid \\theta, \\phi)\\).\nConditioning on observations \\(y\\) we obtain the joint posterior \\(p(\\phi, \\theta \\mid y) \\propto p(y \\mid \\theta, \\phi) p(\\theta |\n\\phi) p(\\phi)\\), where \\(p(y \\mid \\theta, \\phi)\\) as function of \\(\\theta\\) and \\(\\phi\\) is the likelihood function. To sample from the joint posterior, we can either use a standard method, such as Markov chain Monte Carlo, or we can follow a two-step procedure:\n\nsample from the marginal posterior \\(p(\\phi \\mid y)\\),\nsample from the conditional posterior \\(p(\\theta \\mid y, \\phi)\\).\n\nIn the above procedure, neither the marginal posterior nor the conditional posterior are typically available in closed form and so they must be approximated. The marginal posterior can be written as \\(p(\\phi \\mid y) \\propto p(y \\mid \\phi) p(\\phi)\\), where \\(p(y \\mid \\phi) = \\int p(y \\mid \\phi, \\theta) p(\\theta) \\text{d}\\theta\\) is called the marginal likelihood. The Laplace method approximates \\(p(y \\mid \\phi, \\theta) p(\\theta)\\) with a normal distribution centered at the mode, \\[\n  \\theta^* = \\underset{\\theta}{\\text{argmax}} \\ \\log p(\\theta \\mid y, \\phi),\n\\] and \\(\\theta^*\\) is obtained using a numerical optimizer. The resulting Gaussian integral can be evaluated analytically to obtain an approximation to the log marginal likelihood \\(\\log \\hat p(y \\mid \\phi) \\approx \\log p(y \\mid \\phi)\\). Specifically: \\[\n  \\hat p(y \\mid \\phi) = \\frac{p(\\theta^* \\mid \\phi) p(y \\mid \\theta^*, \\phi)}{\\hat p (\\theta^* \\mid \\phi, y)}.\n\\]\nCombining this marginal likelihood with the prior in the model block, we can then sample from the marginal posterior \\(p(\\phi \\mid y)\\) using one of Stan’s algorithms. The marginal posterior is lower dimensional and likely to have a simpler geometry leading to more efficient inference. On the other hand each marginal likelihood computation is more costly, and the combined change in efficiency depends on the application.\nTo obtain posterior draws for \\(\\theta\\), we sample from the normal approximation to \\(p(\\theta \\mid y, \\phi)\\) in generated quantities. The process of iteratively sampling from \\(p(\\phi \\mid y)\\) (say, with MCMC) and then \\(p(\\theta \\mid y, \\phi)\\) produces posterior draws from the joint posterior \\(p(\\theta, \\phi \\mid y)\\).\nThe Laplace approximation is especially useful if \\(p(y \\mid \\phi, \\theta)\\) as function of \\(\\theta\\) is log-concave, e.g., in case of Poisson, binomial, negative-binomial, and Bernoulli. (The likelihood of normal model is also log concave, however when the likelihood is normal, marginalization can be performed exactly and does not required an approximation.) Stan’s embedded Laplace approximation is restricted to the case where the prior \\(p(\\theta \\mid \\phi)\\) is multivariate normal. Furthermore, the likelihood \\(p(y \\mid \\phi, \\theta)\\) must be computed using only operations which support higher-order derivatives (see section specifying the likelihood function).\nThe Laplace approximation can also be useful in generated quantities to marginalize out latent variables even if the sampling had been done using the full joint posterior.\n\n\nIn the model block, we increment target with laplace_marginal, a function that approximates the log marginal likelihood \\(\\log p(y \\mid \\phi)\\). The signature of the function is:\n \n\nreal laplace_marginal(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nwhich returns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\).\nThe embedded Laplace functions accept two functors whose user defined arguments are passed in as tuples to laplace_marginal.\n\nlikelihood_function - user-specified log likelihood whose first argument is the vector of latent Gaussian variables \\(\\theta\\). The subsequent arguments are user defined.\n\n\nreal likelihood_function(vector theta, likelihood_arguments_1, likelihood_arguments_2, ...).\n\n\nlikelihood_arguments - A tuple of arguments whose internal members are be passed to the log likelihood function. This tuple does NOT include the latent variable \\(\\theta\\).\nhessian_block_size - the block size of the Hessian of the log likelihood, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\).\ncovariance_function - A function that returns the covariance matrix of the multivariate normal prior on \\(\\theta\\).\n\n\nmatrix covariance_function(covariance_argument_1, covariance_argument_2, ...).\n\n\ncovariance_arguments A tuple of the arguments whose internal members will be passed to the the covariance function.\n\nAvailable since 2.39\nBelow we go over each argument in more detail.\n\n\n\nThe first step to use the embedded Laplace approximation is to write down a function in the functions block which returns the log likelihood \\(\\log p(y \\mid \\theta, \\phi)\\).\nThere are a few constraints on this function:\n\nThe function return type must be real.\nThe first argument must be the latent Gaussian variable \\(\\theta\\) and must have type vector.\nThe operations in the function must support higher-order automatic differentiation (AD). Most functions in Stan support higher-order AD. The exceptions are functions with specialized calls for reverse-mode AD, and these are higher-order functions (algebraic solvers, differential equation solvers, and integrators), the marginalization function for hidden Markov models (HMM) function, and the embedded Laplace approximation itself.\n\nThe base signature of the function is\nreal likelihood_function(vector theta, ...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter.\nThe tuple after likelihood_function contains the arguments that get passed to likelihood_function excluding \\(\\theta\\). For instance, if a user defined likelihood uses a real and a matrix, the likelihood function’s signature would first have a vector and then a real and matrix argument.\nreal likelihood_fun(vector theta, real a, matrix X)\nThe call to the laplace marginal would start with this likelihood and tuple holding the other likelihood arguments. We do not need to pass theta, since it is marginalized out and therefore does not appear explicitly as a passed parameter.\nreal val = laplace_marginal(likelihood_fun, (a, X), hessian_block_size, ...);\nIf the likelihood_function has only one argument, the tuple syntax is (a, ).\nAs always, users should use parameter arguments only when necessary in order to speed up differentiation. In general, we recommend marking data only arguments with the keyword data, for example,\nreal likelihood_function(vector theta, data vector x, ...)\nIn addition to the likelihood function, users must specify the block size of the Hessian, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\). The Hessian is often block diagonal and this structure can be taken advantage of for fast computation. For example, if \\(y_i\\) only depends on \\(\\theta_i\\), then the Hessian is diagonal and hessian_block_size=1,\nreal val = laplace_marginal(likelihood_fun, (a, X), 1, ...);\nOn the other hand, if the Hessian is not block diagonal, we can always set hessian_block_size=n where \\(n\\) is the size of \\(\\theta\\).\n\n\n\nThe argument covariance_function returns the prior covariance matrix \\(K\\). The signature for this function is the same as a standard stan function. It’s return type must be a matrix of size \\(n \\times n\\) where \\(n\\) is the size of \\(\\theta\\).\nmatrix covariance_function(...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter. The variables \\(\\phi\\) is implicitly defined as the collection of all non-data arguments passed to likelihood_function (excluding \\(\\theta\\)) and covariance_function.\nThe tuple after covariance_function contains the arguments that get passed to covariance_function. For instance, if a user defined covariance function uses two vectors\nmatrix cov_fun(real b, matrix Z)\nthe call to the Laplace marginal would include the covariance function and a tuple holding the covariance function arguments.\nreal val = laplace_marginal(likelihood_fun, (a, X), cov_fun, (b, Z), ...);\nIf the covariance_function has only one argument, the tuple syntax is (b, ).\n\n\n\nIt also possible to specify control parameters, which can help improve the optimization that underlies the Laplace approximation, using laplace_marginal_tol with the following signature:\n \n\nreal laplace_marginal_tol(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...),  tuple(vector, real, int, int, int, int) tolerances)\nThe final argument, tolerances, is a tuple with the following elements\ntuple(vector theta_init, real tol, int max_steps, int solver,\n      int max_steps_linesearch, int allow_fallback)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) and allows the user to tune the control parameters of the approximation.\n\ntheta_init: the initial guess for a Newton solver when finding the mode of \\(p(\\theta \\mid y, \\phi)\\). By default, it is a zero-vector.\ntol: the tolerance \\(\\epsilon\\) of the optimizer. Specifically, the optimizer stops when \\(||\\nabla \\log p(\\theta \\mid y, \\phi)|| \\le \\epsilon\\). By default, the value is \\(\\epsilon \\approx 1.49 \\times 10^{-8}\\), which is the square-root of machine precision.\nmax_num_steps: the maximum number of steps taken by the optimizer before it gives up (in which case the Metropolis proposal gets rejected). The default is 500 steps.\nsolver: choice of Newton solver. The optimizer underlying the Laplace approximation does one of three matrix decompositions to compute a Newton step. The problem determines which decomposition is numerically stable. By default (solver=1), the solver attempts a Cholesky decomposition of the negative Hessian of the log likelihood, \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\). This operation is legal if the negative Hessian is positive-definite, which will always be true when the likelihood as function of \\(\\theta\\) is log concave. If solver=2, the solver makes a Cholesky decomposition of the covariance matrix \\(K(\\phi)\\). Since a covariance matrix is always positive-definite, computing its Cholesky decomposition is always a legal operation, at least in theory. In practice, we may not be able to compute the Cholesky decomposition of the negative Hessian nor of the covariance matrix, either because it does not exist or because of numerical issues. In that case, we can use solver=3 which uses a more expensive but less specialized approach to compute a Newton step.\nmax_steps_linesearch: maximum number of steps in linesearch. The linesearch adjusts to step size to ensure that a Newton step leads to an increase in the objective function (i.e., \\(f(\\theta) = p(\\theta \\mid \\phi, y)\\)). If a standard Newton step does not improve the objective function, the step is adjusted iteratively until the objective function increases or the maximum number of steps in the linesearch is reached. By default, max_steps_linesearch=1000. Setting max_steps_linesearch=0 results in no linesearch.\nallow_fallback: If user set solver fails, this flag determines whether to fallback to the next solver. For example, if the user specifies solver=1 but the Cholesky decomposition of the negative Hessian \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\) fails, the optimizer will try solver=2 instead. By default, allow_fallback = 1 (TRUE).\n\nAvailable since 2.39\nThe embedded Laplace approximation’s options have a helper callable generate_laplace_options(int theta_size) that will generate the tuple for the user. This can be useful for quickly setting up the control parameters in the transformed data block to reuse within the model.\ntuple(vector[theta_size], real, int, int, int, int, int) laplace_ops = generate_laplace_options(theta_size);\n// Modify solver type\nlaplace_ops.5 = 2;\n// Turn off fallthrough\nlaplace_ops.7 = 0;\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(int dimension)\nCreate a default laplace options tuple for a theta_init of size dimension.\nAvailable since 2.39\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(vector theta_init)\nCreate a default Laplace options tuple containing theta_init.\nAvailable since 2.39\n\n\n\nIn generated quantities, it is possible to sample from the Laplace approximation of \\(p(\\theta \\mid \\phi, y)\\) using laplace_latent_rng. The signature for laplace_latent_rng follows closely the signature for laplace_marginal:\n \n\nvector laplace_latent_rng(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nSamples from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\).\nAvailable since 2.39\nOnce again, it is possible to specify control parameters:\n \n\nvector laplace_latent_tol_rng(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...), tuple(vector, real, int, int, int, int) tolerances) Samples from the approximate conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nStan provides convenient wrappers for the embedded Laplace approximation when applied to latent Gaussian models with certain likelihoods arising from some common data models. With this wrapper, the likelihood is pre-specified and does not need to be specified by the user. The selection of supported likelihoods is currently narrow and expected to grow. The wrappers exist for the user’s convenience but are not more computationally efficient than specifying log likelihoods in the functions block.\n\n\nGiven count data, with each observed count \\(y_i\\) associated with a group \\(g(i)\\) and a corresponding latent variable \\(\\theta_{g(i)}\\), and a Poisson model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Poisson} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)})),\n\\] where \\(m_{g(i)}\\) acts as an offset for \\(\\theta_{g(i)}\\). This can also be interpreted as a prior mean on \\(\\theta_{g(i)}\\). The arguments required to compute this likelihood are:\n\ny: an array of counts.\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments.\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covaraince_arguments, tolerances).\nAvailable since 2.39\nThe signatures for the embedded Laplace approximation function with a Poisson likelihood are\n \n real laplace_marginal_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n real laplace_marginal_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n vector laplace_latent_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) Returns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nThe negative Binomial distribution generalizes the Poisson distribution by introducing the dispersion parameter \\(\\eta\\). The corresponding likelihood is then \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{NegBinomial2} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)}), \\eta).\n\\] Here we use the alternative parameterization implemented in Stan, meaning that \\[\n\\mathbb E(y_i) = \\exp (\\theta_{g(i)} + m_{g(i)}), \\\\\n\\text{Var}(y_i) = \\mathbb E(y_i) + \\frac{(\\mathbb E(y_i))^2}{\\eta}.\n\\] The arguments for the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\neta: the overdispersion parameter.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a negative Binomial likelihood are\n \n\nreal laplace_marginal_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n\nreal laplace_marginal_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, function covariance_function, data int hessian_block_size, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nGiven binary outcome \\(y_i \\in \\{0, 1\\}\\) and Bernoulli model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Bernoulli} (y_i \\mid \\text{logit}^{-1}(\\theta_{g(i)} + m_{g(i)})).\n\\] The arguments of the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a Bernoulli likelihood are\n \n\nreal laplace_marginal_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters.\nAvailable since 2.39\n \n\nreal laplace_marginal_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link, and lets the user tune the control parameters of the approximation.\nAvailable since 2.39",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Bounded Continuous Distributions"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_continuous_distributions.html#uniform-distribution",
-    "href": "functions-reference/bounded_continuous_distributions.html#uniform-distribution",
-    "title": "Bounded Continuous Distributions",
+    "objectID": "functions-reference/embedded_laplace.html#approximating-the-log-marginal-likelihood-log-py-mid-phi",
+    "href": "functions-reference/embedded_laplace.html#approximating-the-log-marginal-likelihood-log-py-mid-phi",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in (\\alpha,\\infty)\\), then for \\(y \\in [\\alpha,\\beta]\\), \\[\\begin{equation*} \\text{Uniform}(y|\\alpha,\\beta) =\n\\frac{1}{\\beta - \\alpha} . \\end{equation*}\\]\n\n\n\ny ~ uniform(alpha, beta)\nIncrement target log probability density with uniform_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal uniform_lpdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lupdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal uniform_cdf(reals y | reals alpha, reals beta) The uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.0\n \n\nreal uniform_lcdf(reals y | reals alpha, reals beta) The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lccdf(reals y | reals alpha, reals beta) The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nR uniform_rng(reals alpha, reals beta) Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "In the model block, we increment target with laplace_marginal, a function that approximates the log marginal likelihood \\(\\log p(y \\mid \\phi)\\). The signature of the function is:\n \n\nreal laplace_marginal(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nwhich returns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\).\nThe embedded Laplace functions accept two functors whose user defined arguments are passed in as tuples to laplace_marginal.\n\nlikelihood_function - user-specified log likelihood whose first argument is the vector of latent Gaussian variables \\(\\theta\\). The subsequent arguments are user defined.\n\n\nreal likelihood_function(vector theta, likelihood_arguments_1, likelihood_arguments_2, ...).\n\n\nlikelihood_arguments - A tuple of arguments whose internal members are be passed to the log likelihood function. This tuple does NOT include the latent variable \\(\\theta\\).\nhessian_block_size - the block size of the Hessian of the log likelihood, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\).\ncovariance_function - A function that returns the covariance matrix of the multivariate normal prior on \\(\\theta\\).\n\n\nmatrix covariance_function(covariance_argument_1, covariance_argument_2, ...).\n\n\ncovariance_arguments A tuple of the arguments whose internal members will be passed to the the covariance function.\n\nAvailable since 2.39\nBelow we go over each argument in more detail.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Bounded Continuous Distributions"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html",
-    "href": "functions-reference/array_operations.html",
-    "title": "Array Operations",
+    "objectID": "functions-reference/embedded_laplace.html#laplace-likelihood_spec",
+    "href": "functions-reference/embedded_laplace.html#laplace-likelihood_spec",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).\n\n\n \n\nreal min(array[] real x) The minimum value in x, or \\(+\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint min(array[] int x) The minimum value in x, or error if x is size 0.\nAvailable since 2.0\n \n\nreal max(array[] real x) The maximum value in x, or \\(-\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint max(array[] int x) The maximum value in x, or error if x is size 0.\nAvailable since 2.0\n\n\n\n \n\nint sum(array[] int x) The sum of the elements in x, or 0 if the array is empty.\nAvailable since 2.1\n \n\nreal sum(array[] real x) The sum of the elements in x; see definition above.\nAvailable since 2.0\n \n\ncomplex sum(array[] complex x) The sum of the elements in x; see definition above.\nAvailable since 2.30\n \n\nreal prod(array[] real x) The product of the elements in x, or 1 if x is size 0.\nAvailable since 2.0\n \n\nreal prod(array[] int x) The product of the elements in x, \\[\\begin{equation*}\n\\text{product}(x) = \\begin{cases}\n\\prod_{n=1}^N x_n & \\text{if} N &gt; 0 \\\\[4pt] 1 & \\text{if} N = 0\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal log_sum_exp(array[] real x) The natural logarithm of the sum of the exponentials of the elements in x, or \\(-\\infty\\) if the array is empty.\nAvailable since 2.0\n\n\n\nThe sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.1 The sample deviation is defined as the square root of the sample deviation, but is not unbiased.\n \n\nreal mean(array[] real x) The sample mean of the elements in x. For an array \\(x\\) of size \\(N &gt;\n0\\), \\[\\begin{equation*}\n\\text{mean}(x) \\ = \\ \\bar{x} \\ = \\ \\frac{1}{N} \\sum_{n=1}^N\nx_n.\n\\end{equation*}\\] It is an error to the call the mean function with an array of size \\(0\\).\nAvailable since 2.0\n \n\nreal variance(array[] real x) The sample variance of the elements in x. For \\(N &gt; 0\\), \\[\\begin{equation*}\n\\text{variance}(x) \\ = \\ \\begin{cases} \\frac{1}{N-1} \\sum_{n=1}^N (x_n\n- \\bar{x})^2 & \\text{if } N &gt; 1 \\\\[4pt] 0 & \\text{if } N = 1\n\\end{cases}\n\\end{equation*}\\] It is an error to call the variance function with an array of size 0.\nAvailable since 2.0\n \n\nreal sd(array[] real x) The sample standard deviation of elements in x. \\[\\begin{equation*}\n\\text{sd}(x) =\n\\begin{cases} \\sqrt{\\, \\text{variance}(x)} & \\text{if } N &gt; 1 \\\\[4pt]\n0 & \\text{if } N = 0 \\end{cases}\n\\end{equation*}\\] It is an error to call the sd function with an array of size 0.\nAvailable since 2.0\n\n\n\n \n\nreal norm1(vector x) The L1 norm of x, defined by \\[\\begin{equation*}\n\\text{norm1}(x) \\ = \\ \\textstyle \\sum_{n=1}^N (|x_n|)\n\\end{equation*}\\] where N is the size of x.\nAvailable since 2.30\n \n\nreal norm1(row_vector x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm1(array[] real x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm2(vector x) The L2 norm of x, defined by \\[\\begin{equation*}\n\\text{norm2}(x) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n)^2}\n\\end{equation*}\\] where N is the size of x\nAvailable since 2.30\n \n\nreal norm2(row_vector x) The L2 norm of x\nAvailable since 2.30\n \n\nreal norm2(array[] real x) The L2 norm of x\nAvailable since 2.30\n\n\n\n \n\nreal distance(vector x, vector y) The Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\text{distance}(x,y) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n - y_n)^2}\n\\end{equation*}\\] where N is the size of x and y. It is an error to call distance with arguments of unequal size.\nAvailable since 2.2\n \n\nreal distance(vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal squared_distance(vector x, vector y) The squared Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\mathrm{squared\\_distance}(x,y) \\ = \\ \\text{distance}(x,y)^2 \\ = \\ \\textstyle \\sum_{n=1}^N (x_n - y_n)^2,\n\\end{equation*}\\] where N is the size of x and y. It is an error to call squared_distance with arguments of unequal size.\nAvailable since 2.7\n \n\nreal squared_distance(vector x, row_vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.26\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data array[] real x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data array[] real x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n\n\n\n\nThe size of an array or matrix can be obtained using the dims() function. The dims() function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,\n array[7, 8, 9] real x;\n array[7] matrix[8, 9] y;\nthen calling dims(x) or dims(y) returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.\nThe size() function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as\n array[M, N] real a;\nthe size of a is M.\nThe function num_elements, on the other hand, measures all of the elements, so that the array a above has \\(M \\times N\\) elements.\nThe specialized functions rows() and cols() should be used to extract the dimensions of vectors and matrices.\n \n\narray[] int dims(T x) Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.\nAvailable since 2.0\n \n\nint num_elements(array[] T x) Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if x is of type array[4, 3] real then num_elements(x) is 12, and if y is declared as array[5] matrix[3, 4] y, then size(y) evaluates to 60.\nAvailable since 2.5\n \n\nint size(array[] T x) Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if x is of type array[4, 3] real then size(x) is 4.\nAvailable since 2.0\n\n\n\nThe following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.\n \n\narray[] T rep_array(T x, int n) Return the n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray [,] T rep_array(T x, int m, int n) Return the m by n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray[,,] T rep_array(T x, int k, int m, int n) Return the k by m by n array with every entry assigned to x.\nAvailable since 2.0\nFor example, rep_array(1.0,5) produces a real array (type array[] real) of size 5 with all values set to 1.0. On the other hand, rep_array(1,5) produces an integer array (type array[] int) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment\n array[5] real y;\n array[5] int x;\n\n x = rep_array(1, 5);     // ok\n y = rep_array(1.0, 5);   // ok\n\n x = rep_array(1.0, 5);   // illegal\n y = rep_array(1, 5);     // illegal\n\n x = y;                  // illegal\n y = x;                  // illegal\nIf the value being repeated v is a vector (i.e., T is vector), then rep_array(v, 27) is a size 27 array consisting of 27 copies of the vector v.\n vector[5] v;\n array[3] vector[5] a;\n\n a = rep_array(v, 3);  // fill a with copies of v\n a[2, 4] = 9.0;        // v[4], a[1, 4], a[3, 4] unchanged\nIf the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the rep_array functions is called. For instance, consider the following legal code snippet.\n array[5, 6] real a;\n array[3, 4, 5, 6] real b;\n\n b = rep_array(a, 3, 4); //  make (3 x 4) copies of a\n b[1, 1, 1, 1] = 27.9;    //  a[1, 1] unchanged\nAfter the assignment to b, the value for b[j, k, m, n] is equal to a[m, n] where it is defined, for j in 1:3, k in 1:4, m in 1:5, and n in 1:6.\n\n\n\n \n\nT append_array(T x, T y) Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.\nAvailable since 2.18\nFor example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.\n array[2, 1, 7] matrix[4, 6] x1;\n array[3, 1, 7] matrix[4, 6] x2;\n array[5, 1, 7] matrix[4, 6] x3;\n\n x3 = append_array(x1, x2);\n\n\n\nSorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1, -10.3,\n20.987),\n\\end{equation*}\\] then the various sort routines produce \\[\\begin{eqnarray*}\n\\mathrm{sort\\_asc(v)} & = &  (-10.3,1,20.987) \\\\[4pt]\n\\mathrm{sort\\_desc(v)} & = &  (20.987,1,-10.3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_asc(v)} & = &  (2,1,3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_desc(v)} & = &  (3,1,2)\n\\end{eqnarray*}\\]\n \n\narray[] real sort_asc(array[] real v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] int sort_asc(array[] int v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] real sort_desc(array[] real v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_desc(array[] int v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(array[] real v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(array[] int v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n\n\n\nStan provides functions to create a new array by reversing the order of elements in an existing array. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1,\\, -10.3,\\, 20.987),\n\\end{equation*}\\] then \\[\\begin{equation*}\n\\mathrm{reverse(v)} = (20.987,\\, -10.3,\\, 1).\n\\end{equation*}\\]\n \n\narray[] T reverse(array[] T v) Return a new array containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "text": "The first step to use the embedded Laplace approximation is to write down a function in the functions block which returns the log likelihood \\(\\log p(y \\mid \\theta, \\phi)\\).\nThere are a few constraints on this function:\n\nThe function return type must be real.\nThe first argument must be the latent Gaussian variable \\(\\theta\\) and must have type vector.\nThe operations in the function must support higher-order automatic differentiation (AD). Most functions in Stan support higher-order AD. The exceptions are functions with specialized calls for reverse-mode AD, and these are higher-order functions (algebraic solvers, differential equation solvers, and integrators), the marginalization function for hidden Markov models (HMM) function, and the embedded Laplace approximation itself.\n\nThe base signature of the function is\nreal likelihood_function(vector theta, ...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter.\nThe tuple after likelihood_function contains the arguments that get passed to likelihood_function excluding \\(\\theta\\). For instance, if a user defined likelihood uses a real and a matrix, the likelihood function’s signature would first have a vector and then a real and matrix argument.\nreal likelihood_fun(vector theta, real a, matrix X)\nThe call to the laplace marginal would start with this likelihood and tuple holding the other likelihood arguments. We do not need to pass theta, since it is marginalized out and therefore does not appear explicitly as a passed parameter.\nreal val = laplace_marginal(likelihood_fun, (a, X), hessian_block_size, ...);\nIf the likelihood_function has only one argument, the tuple syntax is (a, ).\nAs always, users should use parameter arguments only when necessary in order to speed up differentiation. In general, we recommend marking data only arguments with the keyword data, for example,\nreal likelihood_function(vector theta, data vector x, ...)\nIn addition to the likelihood function, users must specify the block size of the Hessian, \\(\\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial \\theta^2\\). The Hessian is often block diagonal and this structure can be taken advantage of for fast computation. For example, if \\(y_i\\) only depends on \\(\\theta_i\\), then the Hessian is diagonal and hessian_block_size=1,\nreal val = laplace_marginal(likelihood_fun, (a, X), 1, ...);\nOn the other hand, if the Hessian is not block diagonal, we can always set hessian_block_size=n where \\(n\\) is the size of \\(\\theta\\).",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Array Operations"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#array-reductions",
-    "href": "functions-reference/array_operations.html#array-reductions",
-    "title": "Array Operations",
+    "objectID": "functions-reference/embedded_laplace.html#specifying-the-covariance-function",
+    "href": "functions-reference/embedded_laplace.html#specifying-the-covariance-function",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).\n\n\n \n\nreal min(array[] real x) The minimum value in x, or \\(+\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint min(array[] int x) The minimum value in x, or error if x is size 0.\nAvailable since 2.0\n \n\nreal max(array[] real x) The maximum value in x, or \\(-\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint max(array[] int x) The maximum value in x, or error if x is size 0.\nAvailable since 2.0\n\n\n\n \n\nint sum(array[] int x) The sum of the elements in x, or 0 if the array is empty.\nAvailable since 2.1\n \n\nreal sum(array[] real x) The sum of the elements in x; see definition above.\nAvailable since 2.0\n \n\ncomplex sum(array[] complex x) The sum of the elements in x; see definition above.\nAvailable since 2.30\n \n\nreal prod(array[] real x) The product of the elements in x, or 1 if x is size 0.\nAvailable since 2.0\n \n\nreal prod(array[] int x) The product of the elements in x, \\[\\begin{equation*}\n\\text{product}(x) = \\begin{cases}\n\\prod_{n=1}^N x_n & \\text{if} N &gt; 0 \\\\[4pt] 1 & \\text{if} N = 0\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal log_sum_exp(array[] real x) The natural logarithm of the sum of the exponentials of the elements in x, or \\(-\\infty\\) if the array is empty.\nAvailable since 2.0\n\n\n\nThe sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.1 The sample deviation is defined as the square root of the sample deviation, but is not unbiased.\n \n\nreal mean(array[] real x) The sample mean of the elements in x. For an array \\(x\\) of size \\(N &gt;\n0\\), \\[\\begin{equation*}\n\\text{mean}(x) \\ = \\ \\bar{x} \\ = \\ \\frac{1}{N} \\sum_{n=1}^N\nx_n.\n\\end{equation*}\\] It is an error to the call the mean function with an array of size \\(0\\).\nAvailable since 2.0\n \n\nreal variance(array[] real x) The sample variance of the elements in x. For \\(N &gt; 0\\), \\[\\begin{equation*}\n\\text{variance}(x) \\ = \\ \\begin{cases} \\frac{1}{N-1} \\sum_{n=1}^N (x_n\n- \\bar{x})^2 & \\text{if } N &gt; 1 \\\\[4pt] 0 & \\text{if } N = 1\n\\end{cases}\n\\end{equation*}\\] It is an error to call the variance function with an array of size 0.\nAvailable since 2.0\n \n\nreal sd(array[] real x) The sample standard deviation of elements in x. \\[\\begin{equation*}\n\\text{sd}(x) =\n\\begin{cases} \\sqrt{\\, \\text{variance}(x)} & \\text{if } N &gt; 1 \\\\[4pt]\n0 & \\text{if } N = 0 \\end{cases}\n\\end{equation*}\\] It is an error to call the sd function with an array of size 0.\nAvailable since 2.0\n\n\n\n \n\nreal norm1(vector x) The L1 norm of x, defined by \\[\\begin{equation*}\n\\text{norm1}(x) \\ = \\ \\textstyle \\sum_{n=1}^N (|x_n|)\n\\end{equation*}\\] where N is the size of x.\nAvailable since 2.30\n \n\nreal norm1(row_vector x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm1(array[] real x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm2(vector x) The L2 norm of x, defined by \\[\\begin{equation*}\n\\text{norm2}(x) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n)^2}\n\\end{equation*}\\] where N is the size of x\nAvailable since 2.30\n \n\nreal norm2(row_vector x) The L2 norm of x\nAvailable since 2.30\n \n\nreal norm2(array[] real x) The L2 norm of x\nAvailable since 2.30\n\n\n\n \n\nreal distance(vector x, vector y) The Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\text{distance}(x,y) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n - y_n)^2}\n\\end{equation*}\\] where N is the size of x and y. It is an error to call distance with arguments of unequal size.\nAvailable since 2.2\n \n\nreal distance(vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal squared_distance(vector x, vector y) The squared Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\mathrm{squared\\_distance}(x,y) \\ = \\ \\text{distance}(x,y)^2 \\ = \\ \\textstyle \\sum_{n=1}^N (x_n - y_n)^2,\n\\end{equation*}\\] where N is the size of x and y. It is an error to call squared_distance with arguments of unequal size.\nAvailable since 2.7\n \n\nreal squared_distance(vector x, row_vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.26\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data array[] real x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data array[] real x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27",
+    "text": "The argument covariance_function returns the prior covariance matrix \\(K\\). The signature for this function is the same as a standard stan function. It’s return type must be a matrix of size \\(n \\times n\\) where \\(n\\) is the size of \\(\\theta\\).\nmatrix covariance_function(...)\nThe ... represents a set of optional variadic arguments. There is no type restrictions for the variadic arguments ... and each argument can be passed as data or parameter. The variables \\(\\phi\\) is implicitly defined as the collection of all non-data arguments passed to likelihood_function (excluding \\(\\theta\\)) and covariance_function.\nThe tuple after covariance_function contains the arguments that get passed to covariance_function. For instance, if a user defined covariance function uses two vectors\nmatrix cov_fun(real b, matrix Z)\nthe call to the Laplace marginal would include the covariance function and a tuple holding the covariance function arguments.\nreal val = laplace_marginal(likelihood_fun, (a, X), cov_fun, (b, Z), ...);\nIf the covariance_function has only one argument, the tuple syntax is (b, ).",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Array Operations"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#array-size-and-dimension-function",
-    "href": "functions-reference/array_operations.html#array-size-and-dimension-function",
-    "title": "Array Operations",
+    "objectID": "functions-reference/embedded_laplace.html#control-parameters",
+    "href": "functions-reference/embedded_laplace.html#control-parameters",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "The size of an array or matrix can be obtained using the dims() function. The dims() function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,\n array[7, 8, 9] real x;\n array[7] matrix[8, 9] y;\nthen calling dims(x) or dims(y) returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.\nThe size() function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as\n array[M, N] real a;\nthe size of a is M.\nThe function num_elements, on the other hand, measures all of the elements, so that the array a above has \\(M \\times N\\) elements.\nThe specialized functions rows() and cols() should be used to extract the dimensions of vectors and matrices.\n \n\narray[] int dims(T x) Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.\nAvailable since 2.0\n \n\nint num_elements(array[] T x) Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if x is of type array[4, 3] real then num_elements(x) is 12, and if y is declared as array[5] matrix[3, 4] y, then size(y) evaluates to 60.\nAvailable since 2.5\n \n\nint size(array[] T x) Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if x is of type array[4, 3] real then size(x) is 4.\nAvailable since 2.0",
+    "text": "It also possible to specify control parameters, which can help improve the optimization that underlies the Laplace approximation, using laplace_marginal_tol with the following signature:\n \n\nreal laplace_marginal_tol(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...),  tuple(vector, real, int, int, int, int) tolerances)\nThe final argument, tolerances, is a tuple with the following elements\ntuple(vector theta_init, real tol, int max_steps, int solver,\n      int max_steps_linesearch, int allow_fallback)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) and allows the user to tune the control parameters of the approximation.\n\ntheta_init: the initial guess for a Newton solver when finding the mode of \\(p(\\theta \\mid y, \\phi)\\). By default, it is a zero-vector.\ntol: the tolerance \\(\\epsilon\\) of the optimizer. Specifically, the optimizer stops when \\(||\\nabla \\log p(\\theta \\mid y, \\phi)|| \\le \\epsilon\\). By default, the value is \\(\\epsilon \\approx 1.49 \\times 10^{-8}\\), which is the square-root of machine precision.\nmax_num_steps: the maximum number of steps taken by the optimizer before it gives up (in which case the Metropolis proposal gets rejected). The default is 500 steps.\nsolver: choice of Newton solver. The optimizer underlying the Laplace approximation does one of three matrix decompositions to compute a Newton step. The problem determines which decomposition is numerically stable. By default (solver=1), the solver attempts a Cholesky decomposition of the negative Hessian of the log likelihood, \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\). This operation is legal if the negative Hessian is positive-definite, which will always be true when the likelihood as function of \\(\\theta\\) is log concave. If solver=2, the solver makes a Cholesky decomposition of the covariance matrix \\(K(\\phi)\\). Since a covariance matrix is always positive-definite, computing its Cholesky decomposition is always a legal operation, at least in theory. In practice, we may not be able to compute the Cholesky decomposition of the negative Hessian nor of the covariance matrix, either because it does not exist or because of numerical issues. In that case, we can use solver=3 which uses a more expensive but less specialized approach to compute a Newton step.\nmax_steps_linesearch: maximum number of steps in linesearch. The linesearch adjusts to step size to ensure that a Newton step leads to an increase in the objective function (i.e., \\(f(\\theta) = p(\\theta \\mid \\phi, y)\\)). If a standard Newton step does not improve the objective function, the step is adjusted iteratively until the objective function increases or the maximum number of steps in the linesearch is reached. By default, max_steps_linesearch=1000. Setting max_steps_linesearch=0 results in no linesearch.\nallow_fallback: If user set solver fails, this flag determines whether to fallback to the next solver. For example, if the user specifies solver=1 but the Cholesky decomposition of the negative Hessian \\(- \\partial^2 \\log p(y \\mid \\theta, \\phi) / \\partial^2 \\theta\\) fails, the optimizer will try solver=2 instead. By default, allow_fallback = 1 (TRUE).\n\nAvailable since 2.39\nThe embedded Laplace approximation’s options have a helper callable generate_laplace_options(int theta_size) that will generate the tuple for the user. This can be useful for quickly setting up the control parameters in the transformed data block to reuse within the model.\ntuple(vector[theta_size], real, int, int, int, int, int) laplace_ops = generate_laplace_options(theta_size);\n// Modify solver type\nlaplace_ops.5 = 2;\n// Turn off fallthrough\nlaplace_ops.7 = 0;\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(int dimension)\nCreate a default laplace options tuple for a theta_init of size dimension.\nAvailable since 2.39\n \n\ntuple(vector, real, int, int, int, int) generate_laplace_options(vector theta_init)\nCreate a default Laplace options tuple containing theta_init.\nAvailable since 2.39",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Array Operations"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#array-broadcasting",
-    "href": "functions-reference/array_operations.html#array-broadcasting",
-    "title": "Array Operations",
+    "objectID": "functions-reference/embedded_laplace.html#sample-from-the-approximate-conditional-hatptheta-mid-y-phi",
+    "href": "functions-reference/embedded_laplace.html#sample-from-the-approximate-conditional-hatptheta-mid-y-phi",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "The following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.\n \n\narray[] T rep_array(T x, int n) Return the n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray [,] T rep_array(T x, int m, int n) Return the m by n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray[,,] T rep_array(T x, int k, int m, int n) Return the k by m by n array with every entry assigned to x.\nAvailable since 2.0\nFor example, rep_array(1.0,5) produces a real array (type array[] real) of size 5 with all values set to 1.0. On the other hand, rep_array(1,5) produces an integer array (type array[] int) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment\n array[5] real y;\n array[5] int x;\n\n x = rep_array(1, 5);     // ok\n y = rep_array(1.0, 5);   // ok\n\n x = rep_array(1.0, 5);   // illegal\n y = rep_array(1, 5);     // illegal\n\n x = y;                  // illegal\n y = x;                  // illegal\nIf the value being repeated v is a vector (i.e., T is vector), then rep_array(v, 27) is a size 27 array consisting of 27 copies of the vector v.\n vector[5] v;\n array[3] vector[5] a;\n\n a = rep_array(v, 3);  // fill a with copies of v\n a[2, 4] = 9.0;        // v[4], a[1, 4], a[3, 4] unchanged\nIf the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the rep_array functions is called. For instance, consider the following legal code snippet.\n array[5, 6] real a;\n array[3, 4, 5, 6] real b;\n\n b = rep_array(a, 3, 4); //  make (3 x 4) copies of a\n b[1, 1, 1, 1] = 27.9;    //  a[1, 1] unchanged\nAfter the assignment to b, the value for b[j, k, m, n] is equal to a[m, n] where it is defined, for j in 1:3, k in 1:4, m in 1:5, and n in 1:6.",
+    "text": "In generated quantities, it is possible to sample from the Laplace approximation of \\(p(\\theta \\mid \\phi, y)\\) using laplace_latent_rng. The signature for laplace_latent_rng follows closely the signature for laplace_marginal:\n \n\nvector laplace_latent_rng(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nSamples from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\).\nAvailable since 2.39\nOnce again, it is possible to specify control parameters:\n \n\nvector laplace_latent_tol_rng(function likelihood_function, tuple(...), int hessian_block_size, function covariance_function, tuple(...), tuple(vector, real, int, int, int, int) tolerances) Samples from the approximate conditional posterior \\(p(\\theta \\mid y, \\phi)\\) and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Array Operations"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#array-concatenation",
-    "href": "functions-reference/array_operations.html#array-concatenation",
-    "title": "Array Operations",
+    "objectID": "functions-reference/embedded_laplace.html#built-in-laplace-marginal-likelihood-functions",
+    "href": "functions-reference/embedded_laplace.html#built-in-laplace-marginal-likelihood-functions",
+    "title": "Embedded Laplace Approximation",
     "section": "",
-    "text": "T append_array(T x, T y) Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.\nAvailable since 2.18\nFor example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.\n array[2, 1, 7] matrix[4, 6] x1;\n array[3, 1, 7] matrix[4, 6] x2;\n array[5, 1, 7] matrix[4, 6] x3;\n\n x3 = append_array(x1, x2);",
+    "text": "Stan provides convenient wrappers for the embedded Laplace approximation when applied to latent Gaussian models with certain likelihoods arising from some common data models. With this wrapper, the likelihood is pre-specified and does not need to be specified by the user. The selection of supported likelihoods is currently narrow and expected to grow. The wrappers exist for the user’s convenience but are not more computationally efficient than specifying log likelihoods in the functions block.\n\n\nGiven count data, with each observed count \\(y_i\\) associated with a group \\(g(i)\\) and a corresponding latent variable \\(\\theta_{g(i)}\\), and a Poisson model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Poisson} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)})),\n\\] where \\(m_{g(i)}\\) acts as an offset for \\(\\theta_{g(i)}\\). This can also be interpreted as a prior mean on \\(\\theta_{g(i)}\\). The arguments required to compute this likelihood are:\n\ny: an array of counts.\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments.\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_poisson_log(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_poisson_log_lupmf(y | y_index, m, hessian_block_size, covariance_function, covaraince_arguments, tolerances).\nAvailable since 2.39\nThe signatures for the embedded Laplace approximation function with a Poisson likelihood are\n \n real laplace_marginal_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n real laplace_marginal_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n real laplace_marginal_tol_poisson_log_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n vector laplace_latent_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) Returns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_poisson_log_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Poisson distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nThe negative Binomial distribution generalizes the Poisson distribution by introducing the dispersion parameter \\(\\eta\\). The corresponding likelihood is then \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{NegBinomial2} (y_i \\mid \\exp(\\theta_{g(i)} + m_{g(i)}), \\eta).\n\\] Here we use the alternative parameterization implemented in Stan, meaning that \\[\n\\mathbb E(y_i) = \\exp (\\theta_{g(i)} + m_{g(i)}), \\\\\n\\text{Var}(y_i) = \\mathbb E(y_i) + \\frac{(\\mathbb E(y_i))^2}{\\eta}.\n\\] The arguments for the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\neta: the overdispersion parameter.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_neg_binomial_2_log(y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_neg_binomial_2_log_lupmf(y | y_index, eta, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a negative Binomial likelihood are\n \n\nreal laplace_marginal_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lpmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n \n\nreal laplace_marginal_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, function covariance_function, data int hessian_block_size, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_neg_binomial_2_log_lupmf(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative Binomial distribution with a log link, and allows the user to tune the control parameters of the approximation, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_neg_binomial_2_log_rng(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi, \\eta)\\) in the special case where the likelihood \\(p(y \\mid \\theta, \\eta)\\) is a Negative binomial distribution with a log link and allows the user to tune the control parameters of the approximation.\nAvailable since 2.39\n\n\n\nGiven binary outcome \\(y_i \\in \\{0, 1\\}\\) and Bernoulli model, the likelihood is \\[\np(y \\mid \\theta, \\phi) = \\prod_i\\text{Bernoulli} (y_i \\mid \\text{logit}^{-1}(\\theta_{g(i)} + m_{g(i)})).\n\\] The arguments of the likelihood function are:\n\ny: the observed counts\ny_index: an array whose \\(i^\\text{th}\\) element indicates to which group the \\(i^\\text{th}\\) observation belongs to.\nm: a vector of offsets or prior means for \\(\\theta\\).\n\n \n\ny ~ laplace_marginal_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments)\nIncrement target log probability density with laplace_marginal_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments).\nAvailable since 2.39\n \n\ny ~ laplace_marginal_tol_bernoulli_logit(y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances)\nIncrement target log probability density with laplace_marginal_tol_bernoulli_logit_lupmf(y | y_index, m, hessian_block_size, covariance_function, covariance_arguments, tolerances).\nAvailable since 2.39\nThe function signatures for the embedded Laplace approximation with a Bernoulli likelihood are\n \n\nreal laplace_marginal_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lpmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters.\nAvailable since 2.39\n \n\nreal laplace_marginal_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link, dropping constant terms.\nAvailable since 2.39\n \n\nreal laplace_marginal_tol_bernoulli_logit_lupmf(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns an approximation to the log marginal likelihood \\(p(y \\mid \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a bernoulli distribution with a logit link and allows the user to tune the control parameters, dropping constant terms.\nAvailable since 2.39\n \n\nvector laplace_latent_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link.\nAvailable since 2.39\n \n\nvector laplace_latent_tol_bernoulli_logit_rng(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances)\nReturns a draw from the Laplace approximation to the conditional posterior \\(p(\\theta \\mid y, \\phi)\\) in the special case where the likelihood \\(p(y \\mid \\theta)\\) is a Bernoulli distribution with a logit link, and lets the user tune the control parameters of the approximation.\nAvailable since 2.39",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Array Operations"
+      "Additional Distributions",
+      "Embedded Laplace Approximation"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#sorting-functions",
-    "href": "functions-reference/array_operations.html#sorting-functions",
-    "title": "Array Operations",
+    "objectID": "functions-reference/deprecated_functions.html",
+    "href": "functions-reference/deprecated_functions.html",
+    "title": "Deprecated Functions",
     "section": "",
-    "text": "Sorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1, -10.3,\n20.987),\n\\end{equation*}\\] then the various sort routines produce \\[\\begin{eqnarray*}\n\\mathrm{sort\\_asc(v)} & = &  (-10.3,1,20.987) \\\\[4pt]\n\\mathrm{sort\\_desc(v)} & = &  (20.987,1,-10.3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_asc(v)} & = &  (2,1,3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_desc(v)} & = &  (3,1,2)\n\\end{eqnarray*}\\]\n \n\narray[] real sort_asc(array[] real v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] int sort_asc(array[] int v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] real sort_desc(array[] real v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_desc(array[] int v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(array[] real v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(array[] int v, int s) Number of components of v less than v[s]\nAvailable since 2.0",
+    "text": "This appendix lists currently deprecated functionality along with how to replace it.\nStarting in Stan 2.29, deprecated functions with drop in replacements (such as the renaming of get_lp or multiply_log) will be removed 3 versions later e.g., functions deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in Removed Functions. The Stan compiler can automatically update these on the behalf of the user for the entire deprecation window and at least one version following the removal.\n\n\nDeprecated: Using / with two integer arguments is interpreted as integer floor division, such that\n\\[ 1 / 2 = 0 \\]\nThis is deprecated due to its confusion with real-valued division, where\n\\[ 1.0 / 2.0 = 0.5 \\]\nReplacement: Use the integer division operator operator%/% instead.\n\n\n\nThese ODE integrator functions have been replaced by those described in Ordinary Differential Equation (ODE) Solvers.\n\n\nA system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:\narray[] real ode(real time, array[] real state, array[] real theta,\n                 array[] real x_r, array[] int x_i);\nThe ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\ntheta, parameter values used to evaluate the ODE system\nx_r, data values used to evaluate the ODE system\nx_i, integer data values used to evaluate the ODE system.\n\nThe ODE system function separates parameter values, theta, from data values, x_r, for efficiency in computing the gradients of the ODE.\n\n\n\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.23, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.23, deprecated in 2.24\n\n\n\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.\n\node: function literal referring to a function specifying the system of differential equations with signature:\n\n(real, array[] real, array[] real, data array[] real, data array[] int):array[] real\nThe arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,\n\ninitial_state: initial state, type array[] real,\ninitial_time: initial time, type int or real,\ntimes: solution times, type array[] real,\ntheta: parameters, type array[] real,\ndata x_r: real data, type array[] real, data only, and\ndata x_i: integer data, type array[] int, data only.\n\nFor more fine-grained control of the ODE solvers, these parameters can also be provided:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\ndata max_num_steps: maximum number of steps to take in the ODE solver, type int, data only.\n\n\n\nThe return value for the ODE solvers is an array of type array[,] real, with values consisting of solutions at the specified times.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,\nsolution times and number of rows of the return value of the solver,\nparameters, real data and integer data passed to the solver will be passed to the system function\n\n\n\n\n\n\nThese algebraic solver functions have been replaced by those described in Algebraic Equation Solvers..\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:\n vector algebra_system(vector y, vector theta,\n                              data array[] real x_r, array[] int x_i)\nThe algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.\nThe argument of this function are:\n\ny, the unknowns we wish to solve for\ntheta, parameter values used to evaluate the algebraic system\nx_r, data values used to evaluate the algebraic system\nx_i, integer data used to evaluate the algebraic system\n\nThe algebraic system function separates parameter values, theta, from data values, x_r, for efficiency in propagating the derivatives through the algebraic system.\n\n\n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.\nAvailable since 2.17, deprecated in 2.31\n \n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.\nAvailable since 2.17, deprecated in 2.31\nNote: In future releases, the function algebra_solver will be deprecated and replaced with algebra_solver_powell.\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.24, deprecated in 2.31\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.24, deprecated in 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, vector, array[] real, array[] int):vector. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\ntheta: parameters only, type vector,\nx_r: real data only, type array[] real, and\nx_i: integer data only, type array[] int.\n\nFor more fine-grained control of the algebraic solver, these parameters can also be provided:\n\nrel_tol: relative tolerance for the algebraic solver, type real, data only,\nfunction_tol: function tolerance for the algebraic solver, type real, data only,\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only.\n\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0.\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\nThe parameters, real data, and integer data will be passed from the solver directly to the system function.",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Array Operations"
+      "Deprecated Functions"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#reversing-functions",
-    "href": "functions-reference/array_operations.html#reversing-functions",
-    "title": "Array Operations",
+    "objectID": "functions-reference/deprecated_functions.html#integer-division-with-operator",
+    "href": "functions-reference/deprecated_functions.html#integer-division-with-operator",
+    "title": "Deprecated Functions",
     "section": "",
-    "text": "Stan provides functions to create a new array by reversing the order of elements in an existing array. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1,\\, -10.3,\\, 20.987),\n\\end{equation*}\\] then \\[\\begin{equation*}\n\\mathrm{reverse(v)} = (20.987,\\, -10.3,\\, 1).\n\\end{equation*}\\]\n \n\narray[] T reverse(array[] T v) Return a new array containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "text": "Deprecated: Using / with two integer arguments is interpreted as integer floor division, such that\n\\[ 1 / 2 = 0 \\]\nThis is deprecated due to its confusion with real-valued division, where\n\\[ 1.0 / 2.0 = 0.5 \\]\nReplacement: Use the integer division operator operator%/% instead.",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Array Operations"
+      "Deprecated Functions"
     ]
   },
   {
-    "objectID": "functions-reference/array_operations.html#footnotes",
-    "href": "functions-reference/array_operations.html#footnotes",
-    "title": "Array Operations",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nDividing by \\(N\\) rather than \\((N-1)\\) produces a maximum likelihood estimate of variance, which is biased to underestimate variance.↩︎",
+    "objectID": "functions-reference/deprecated_functions.html#functions-old-ode-solver",
+    "href": "functions-reference/deprecated_functions.html#functions-old-ode-solver",
+    "title": "Deprecated Functions",
+    "section": "",
+    "text": "These ODE integrator functions have been replaced by those described in Ordinary Differential Equation (ODE) Solvers.\n\n\nA system of ODEs is specified as an ordinary function in Stan within the functions block. The ODE system function must have this function signature:\narray[] real ode(real time, array[] real state, array[] real theta,\n                 array[] real x_r, array[] int x_i);\nThe ODE system function should return the derivative of the state with respect to time at the time provided. The length of the returned real array must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\ntheta, parameter values used to evaluate the ODE system\nx_r, data values used to evaluate the ODE system\nx_i, integer data values used to evaluate the ODE system.\n\nThe ODE system function separates parameter values, theta, from data values, x_r, for efficiency in computing the gradients of the ODE.\n\n\n\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_rk45(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.23, deprecated in 2.24\n \n\narray[,] real integrate_ode_adams(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.23, deprecated in 2.24\n\n\n\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.10, deprecated in 2.24\n \n\narray[,] real integrate_ode_bdf(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, data array[] real x_r, data array[] int x_i, data real rel_tol, data real abs_tol, data int max_num_steps) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.10, deprecated in 2.24\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff cases are as follows.\n\node: function literal referring to a function specifying the system of differential equations with signature:\n\n(real, array[] real, array[] real, data array[] real, data array[] int):array[] real\nThe arguments represent (1) time, (2) system state, (3) parameters, (4) real data, and (5) integer data, and the return value contains the derivatives with respect to time of the state,\n\ninitial_state: initial state, type array[] real,\ninitial_time: initial time, type int or real,\ntimes: solution times, type array[] real,\ntheta: parameters, type array[] real,\ndata x_r: real data, type array[] real, data only, and\ndata x_i: integer data, type array[] int, data only.\n\nFor more fine-grained control of the ODE solvers, these parameters can also be provided:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\ndata max_num_steps: maximum number of steps to take in the ODE solver, type int, data only.\n\n\n\nThe return value for the ODE solvers is an array of type array[,] real, with values consisting of solutions at the specified times.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and rows of the return value of the solver,\nsolution times and number of rows of the return value of the solver,\nparameters, real data and integer data passed to the solver will be passed to the system function",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Array Operations"
+      "Deprecated Functions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stansummary.html",
-    "href": "cmdstan-guide/stansummary.html",
-    "title": "stansummary: MCMC Output Analysis",
+    "objectID": "functions-reference/deprecated_functions.html#functions-old-algebra-solver",
+    "href": "functions-reference/deprecated_functions.html#functions-old-algebra-solver",
+    "title": "Deprecated Functions",
     "section": "",
-    "text": "The CmdStan stansummary program reports statistics for one or more sampler chains over all sampler and model parameters and quantities of interest. The statistics reported include both summary statistics of the estimates and diagnostic statistics on the sampler chains, reported in the following order:\n\nMean - sample mean\nMCSE - Monte Carlo Standard Error, a measure of the amount of noise in the sample\nStdDev - sample standard deviation - the standard deviation around the sample mean.\nMAD - Median Absolute Deviation - the median absolute deviation around the sample median.\nQuantiles - default 5%, 50%, 95%\nESS_bulk\nESS_tail\nESS_bulk/s - Bulk ESS per second\nR_hat - \\(\\hat{R}\\) statistic, a MCMC convergence diagnostic\n\nWhen reviewing the stansummary output, it is important to check the final three output columns first - these are the diagnostic statistics on MCMC convergence and effective sample size. A \\(\\hat{R}\\) statistic of greater than \\(1\\) indicates potential convergence problems and that the sample is not presentative of the target posterior, thus the estimates of the mean and all other summary statistics are likely to be invalid. A value \\(1.01\\) can be used as generic threshold to decide whether more iterations or further convergence analysis is needed, but other thresholds can be used depending on the specific use case.\nEstimation by sampling produces an approximate value for the model parameters; the MCSE statistic indicates the amount of uncertainty in the estimate. Therefore MCSE column is placed next to the sample mean column, in order to make it easy to compare this sample with others.\nFor more information, see the Posterior Analysis chapter of the Stan Reference Manual which describes both the theory and practice of MCMC estimation techniques.\nThe statistics - Mean, StdDev, MAD, and Quantiles - are computed directly from all draws across all chains. The diagnostic statistics - ESS_bulk, ESS_tail, and R_hat are computed from the rank-normalized, folded, and splitted chains according to the definitions by Vehtari et al. (2021). the MCSE statistic is computed using split chain R_hat and autocorrelations. The summary statistics and the algorithms used to compute them are described in sections Notation for draws and Effective sample size.\n\n\nThe CmdStan makefile task build compiles the stansummary utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/stansummary\n\n\n\nThe stansummary utility processes one or more output files from a set of chains from one run of the HMC sampler. To run stansummary on the output file or files generated by a run of the sampler, on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nFor example, after running 4 chains to fit the example model eight_schools.stan to the supplied example data file, we run stansummary on the resulting Stan CSV output files to get the following report:\n&gt; bin/stansummary eight_*.csv\nInference for Stan model: eight_schools_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total\nSampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total\n\n                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0\naccept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0\nstepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan\ntreedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0\nn_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0\ndivergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan\nenergy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0\n\nmu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0\ntheta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0\ntheta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0\ntheta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0\ntheta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0\ntheta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0\ntheta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0\ntheta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0\ntheta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0\ntau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nThe console output information consists of\n\nModel, chains, and timing summaries\nSampler parameter statistics\nModel parameter statistics\nSampling algorithm - either nuts (shown here) or static HMC.\n\nThere is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g. if there are no divergent transitions.\n\n\nThe initial Stan CSV columns provide information on the sampler state for each draw:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause we ran the NUTS sampler, the above summary reports sampler parameters treedepth__, n_leapfrog__, and divergent__; the static HMC sampler would report int_time__ instead.\n\n\n\nThe remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The eight_schools.stan program parameters block contains the following parameter variable declarations:\n  real mu;\n  array[J] real theta;\n  real&lt;lower=0&gt; tau;\nIn the example data, J is \\(8\\); therefore the stansummary listing reports on theta[1] through theta[8].\n\n\n\n\nThe stansummary command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the -h or --help option, the program prints the usage message to the console and exits.\nReport statistics for one or more Stan CSV files from a HMC sampler run.\nExample:  stansummary model_chain_1.csv model_chain_2.csv\nOptions:\n  -a, --autocorr [n]          Display the chain autocorrelation for the n-th\n                              input file, in addition to statistics.\n  -c, --csv_filename [file]   Write statistics to a CSV file.\n  -h, --help                  Produce help message, then exit.\n  -p, --percentiles [values]  Percentiles to report as ordered set of\n                              comma-separated numbers from (0.1,99.9), inclusive.\n                              Default is 5,50,95.\n  -s, --sig_figs [n]          Significant figures reported. Default is 2.\n                              Must be an integer from (1, 18), inclusive.\n  -i, --include_param [name]  Include the named parameter in the summary output.\n                              By default, all parameters in the file are summarized,\n                              passing this argument one or more times will filter\n                              the output down to just the requested arguments.\nBoth short an long option names are allowed. Short names are specified as -&lt;o&gt; &lt;value&gt;; long option names can be specified either as --&lt;option&gt;=&lt;value&gt; or --&lt;option&gt; &lt;value&gt;.\nThe --percentiles argument can also be passed an empty string \"\", which results in no percentiles being displayed in the output of the command.\nThe amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument sig_figs. The default sampler output precision is 8. The --sig_figs argument to the stansummary program should not exceed the sig_figs argument to the sampler.",
+    "text": "These algebraic solver functions have been replaced by those described in Algebraic Equation Solvers..\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The algebraic system function must have this signature:\n vector algebra_system(vector y, vector theta,\n                              data array[] real x_r, array[] int x_i)\nThe algebraic system function should return the value of the algebraic function which goes to 0, when we plug in the solution to the algebraic system.\nThe argument of this function are:\n\ny, the unknowns we wish to solve for\ntheta, parameter values used to evaluate the algebraic system\nx_r, data values used to evaluate the algebraic system\nx_i, integer data used to evaluate the algebraic system\n\nThe algebraic system function separates parameter values, theta, from data values, x_r, for efficiency in propagating the derivatives through the algebraic system.\n\n\n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm.\nAvailable since 2.17, deprecated in 2.31\n \n\nvector algebra_solver(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using the Powell hybrid algorithm with additional control parameters for the solver.\nAvailable since 2.17, deprecated in 2.31\nNote: In future releases, the function algebra_solver will be deprecated and replaced with algebra_solver_powell.\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.24, deprecated in 2.31\n \n\nvector algebra_solver_newton(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.24, deprecated in 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, vector, array[] real, array[] int):vector. The arguments represent (1) unknowns, (2) parameters, (3) real data, and (4) integer data, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\ntheta: parameters only, type vector,\nx_r: real data only, type array[] real, and\nx_i: integer data only, type array[] int.\n\nFor more fine-grained control of the algebraic solver, these parameters can also be provided:\n\nrel_tol: relative tolerance for the algebraic solver, type real, data only,\nfunction_tol: function tolerance for the algebraic solver, type real, data only,\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only.\n\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0.\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\nThe parameters, real data, and integer data will be passed from the solver directly to the system function.",
     "crumbs": [
-      "Tools and Utilities",
-      "`stansummary`: MCMC Output Analysis"
+      "Functions Reference",
+      "Built-in Functions",
+      "Deprecated Functions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stansummary.html#building-the-stansummary-command",
-    "href": "cmdstan-guide/stansummary.html#building-the-stansummary-command",
-    "title": "stansummary: MCMC Output Analysis",
+    "objectID": "functions-reference/correlation_matrix_distributions.html",
+    "href": "functions-reference/correlation_matrix_distributions.html",
+    "title": "Correlation Matrix Distributions",
     "section": "",
-    "text": "The CmdStan makefile task build compiles the stansummary utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/stansummary",
+    "text": "The correlation matrix distributions have support on the (Cholesky factors of) correlation matrices. A Cholesky factor \\(L\\) for a \\(K\n\\times K\\) correlation matrix \\(\\Sigma\\) of dimension \\(K\\) has rows of unit length so that the diagonal of \\(L L^{\\top}\\) is the unit \\(K\\)-vector. Even though models are usually conceptualized in terms of correlation matrices, it is better to operationalize them in terms of their Cholesky factors. If you are interested in the posterior distribution of the correlations, you can recover them in the generated quantities block via\n generated quantities {\n   corr_matrix[K] Sigma;\n   Sigma = multiply_lower_tri_self_transpose(L);\n }\n\n\n\n\nFor \\(\\eta &gt; 0\\), if \\(\\Sigma\\) a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*}\n\\text{LkjCorr}(\\Sigma|\\eta) \\propto \\det \\left( \\Sigma \\right)^{(\\eta\n- 1)}. \\end{equation*}\\] The expectation is the identity matrix for any positive value of the shape parameter \\(\\eta\\), which can be interpreted like the shape parameter of a symmetric beta distribution:\n\nif \\(\\eta = 1\\), then the density is uniform over correlation matrices of order \\(K\\);\nif \\(\\eta &gt; 1\\), the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger \\(\\eta\\); and\nfor \\(0 &lt; \\eta &lt; 1\\), the density has a trough at the identity matrix.\nif \\(\\eta\\) were an unknown parameter, the Jeffreys prior is proportional to \\(\\sqrt{2\\sum_{k=1}^{K-1}\\left(\n\\psi_1\\left(\\eta+\\frac{K-k-1}{2}\\right) -   2\\psi_1\\left(2\\eta+K-k-1\n\\right)\\right)}\\), where \\(\\psi_1()\\) is the trigamma function\n\nSee (Lewandowski, Kurowicka, and Joe 2009) for definitions. However, it is much better computationally to work directly with the Cholesky factor of \\(\\Sigma\\), so this distribution should never be explicitly used in practice.\n\n\n\ny ~ lkj_corr(eta)\nIncrement target log probability density with lkj_corr_lupdf(y | eta).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal lkj_corr_lpdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta. lkj_corr_cholesky_lpdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.12\n \n\nreal lkj_corr_lupdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. lkj_corr_cholesky_lupdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.25\n \n\nmatrix lkj_corr_rng(int K, real eta) Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n\n\n\n\nStan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if L is a Cholesky factor of a correlation matrix, then\n L ~ lkj_corr_cholesky(2.0); # implies L * L' ~ lkj_corr(2.0);\nBecause Stan requires models to have support on all valid constrained parameters, L will almost always1 be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,\n parameters {   cholesky_factor_corr[K] L;   # rather than corr_matrix[K] Sigma;   // ...\n\n\nFor \\(\\eta &gt; 0\\), if \\(L\\) is a \\(K \\times K\\) lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*} \\text{LkjCholesky}(L|\\eta)\n\\propto \\left|J\\right|\\det(L L^\\top)^{(\\eta - 1)} = \\prod_{k=2}^K\nL_{kk}^{K-k+2\\eta-2}. \\end{equation*}\\] See the previous section for details on interpreting the shape parameter \\(\\eta\\). Note that even if \\(\\eta=1\\), it is still essential to evaluate the density function because the density of \\(L\\) is not constant, regardless of the value of \\(\\eta\\), even though the density of \\(LL^\\top\\) is constant iff \\(\\eta=1\\).\nA lower triangular \\(L\\) is a Cholesky factor for a correlation matrix if and only if \\(L_{k,k} &gt; 0\\) for \\(k \\in 1{:}K\\) and each row \\(L_k\\) has unit Euclidean length.\n\n\n\nL ~ lkj_corr_cholesky(eta)\nIncrement target log probability density with lkj_corr_cholesky_lupdf(L | eta).\nAvailable since 2.4\n \n\n\n\n\n \n\nreal lkj_corr_cholesky_lpdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta\nAvailable since 2.12\n \n\nreal lkj_corr_cholesky_lupdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms\nAvailable since 2.25\n \n\nmatrix lkj_corr_cholesky_rng(int K, real eta) Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.4",
     "crumbs": [
-      "Tools and Utilities",
-      "`stansummary`: MCMC Output Analysis"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stansummary.html#running-the-stansummary-program",
-    "href": "cmdstan-guide/stansummary.html#running-the-stansummary-program",
-    "title": "stansummary: MCMC Output Analysis",
+    "objectID": "functions-reference/correlation_matrix_distributions.html#lkj-correlation",
+    "href": "functions-reference/correlation_matrix_distributions.html#lkj-correlation",
+    "title": "Correlation Matrix Distributions",
     "section": "",
-    "text": "The stansummary utility processes one or more output files from a set of chains from one run of the HMC sampler. To run stansummary on the output file or files generated by a run of the sampler, on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nFor example, after running 4 chains to fit the example model eight_schools.stan to the supplied example data file, we run stansummary on the resulting Stan CSV output files to get the following report:\n&gt; bin/stansummary eight_*.csv\nInference for Stan model: eight_schools_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total\nSampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total\n\n                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0\naccept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0\nstepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan\ntreedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0\nn_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0\ndivergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan\nenergy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0\n\nmu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0\ntheta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0\ntheta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0\ntheta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0\ntheta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0\ntheta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0\ntheta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0\ntheta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0\ntheta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0\ntau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nThe console output information consists of\n\nModel, chains, and timing summaries\nSampler parameter statistics\nModel parameter statistics\nSampling algorithm - either nuts (shown here) or static HMC.\n\nThere is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g. if there are no divergent transitions.\n\n\nThe initial Stan CSV columns provide information on the sampler state for each draw:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause we ran the NUTS sampler, the above summary reports sampler parameters treedepth__, n_leapfrog__, and divergent__; the static HMC sampler would report int_time__ instead.\n\n\n\nThe remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The eight_schools.stan program parameters block contains the following parameter variable declarations:\n  real mu;\n  array[J] real theta;\n  real&lt;lower=0&gt; tau;\nIn the example data, J is \\(8\\); therefore the stansummary listing reports on theta[1] through theta[8].",
+    "text": "For \\(\\eta &gt; 0\\), if \\(\\Sigma\\) a positive-definite, symmetric matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*}\n\\text{LkjCorr}(\\Sigma|\\eta) \\propto \\det \\left( \\Sigma \\right)^{(\\eta\n- 1)}. \\end{equation*}\\] The expectation is the identity matrix for any positive value of the shape parameter \\(\\eta\\), which can be interpreted like the shape parameter of a symmetric beta distribution:\n\nif \\(\\eta = 1\\), then the density is uniform over correlation matrices of order \\(K\\);\nif \\(\\eta &gt; 1\\), the identity matrix is the modal correlation matrix, with a sharper peak in the density at the identity matrix for larger \\(\\eta\\); and\nfor \\(0 &lt; \\eta &lt; 1\\), the density has a trough at the identity matrix.\nif \\(\\eta\\) were an unknown parameter, the Jeffreys prior is proportional to \\(\\sqrt{2\\sum_{k=1}^{K-1}\\left(\n\\psi_1\\left(\\eta+\\frac{K-k-1}{2}\\right) -   2\\psi_1\\left(2\\eta+K-k-1\n\\right)\\right)}\\), where \\(\\psi_1()\\) is the trigamma function\n\nSee (Lewandowski, Kurowicka, and Joe 2009) for definitions. However, it is much better computationally to work directly with the Cholesky factor of \\(\\Sigma\\), so this distribution should never be explicitly used in practice.\n\n\n\ny ~ lkj_corr(eta)\nIncrement target log probability density with lkj_corr_lupdf(y | eta).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal lkj_corr_lpdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta. lkj_corr_cholesky_lpdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.12\n \n\nreal lkj_corr_lupdf(matrix y | real eta) The log of the LKJ density for the correlation matrix y given nonnegative shape eta dropping constant additive terms. lkj_corr_cholesky_lupdf is faster, more numerically stable, uses less memory, and should be preferred to this.\nAvailable since 2.25\n \n\nmatrix lkj_corr_rng(int K, real eta) Generate a LKJ random correlation matrix of order K with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
     "crumbs": [
-      "Tools and Utilities",
-      "`stansummary`: MCMC Output Analysis"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stansummary.html#command-line-options",
-    "href": "cmdstan-guide/stansummary.html#command-line-options",
-    "title": "stansummary: MCMC Output Analysis",
+    "objectID": "functions-reference/correlation_matrix_distributions.html#cholesky-lkj-correlation-distribution",
+    "href": "functions-reference/correlation_matrix_distributions.html#cholesky-lkj-correlation-distribution",
+    "title": "Correlation Matrix Distributions",
     "section": "",
-    "text": "The stansummary command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the -h or --help option, the program prints the usage message to the console and exits.\nReport statistics for one or more Stan CSV files from a HMC sampler run.\nExample:  stansummary model_chain_1.csv model_chain_2.csv\nOptions:\n  -a, --autocorr [n]          Display the chain autocorrelation for the n-th\n                              input file, in addition to statistics.\n  -c, --csv_filename [file]   Write statistics to a CSV file.\n  -h, --help                  Produce help message, then exit.\n  -p, --percentiles [values]  Percentiles to report as ordered set of\n                              comma-separated numbers from (0.1,99.9), inclusive.\n                              Default is 5,50,95.\n  -s, --sig_figs [n]          Significant figures reported. Default is 2.\n                              Must be an integer from (1, 18), inclusive.\n  -i, --include_param [name]  Include the named parameter in the summary output.\n                              By default, all parameters in the file are summarized,\n                              passing this argument one or more times will filter\n                              the output down to just the requested arguments.\nBoth short an long option names are allowed. Short names are specified as -&lt;o&gt; &lt;value&gt;; long option names can be specified either as --&lt;option&gt;=&lt;value&gt; or --&lt;option&gt; &lt;value&gt;.\nThe --percentiles argument can also be passed an empty string \"\", which results in no percentiles being displayed in the output of the command.\nThe amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument sig_figs. The default sampler output precision is 8. The --sig_figs argument to the stansummary program should not exceed the sig_figs argument to the sampler.",
+    "text": "Stan provides an implicit parameterization of the LKJ correlation matrix density in terms of its Cholesky factor, which you should use rather than the explicit parameterization in the previous section. For example, if L is a Cholesky factor of a correlation matrix, then\n L ~ lkj_corr_cholesky(2.0); # implies L * L' ~ lkj_corr(2.0);\nBecause Stan requires models to have support on all valid constrained parameters, L will almost always1 be a parameter declared with the type of a Cholesky factor for a correlation matrix; for example,\n parameters {   cholesky_factor_corr[K] L;   # rather than corr_matrix[K] Sigma;   // ...\n\n\nFor \\(\\eta &gt; 0\\), if \\(L\\) is a \\(K \\times K\\) lower-triangular Cholesky factor of a symmetric positive-definite matrix with unit diagonal (i.e., a correlation matrix), then \\[\\begin{equation*} \\text{LkjCholesky}(L|\\eta)\n\\propto \\left|J\\right|\\det(L L^\\top)^{(\\eta - 1)} = \\prod_{k=2}^K\nL_{kk}^{K-k+2\\eta-2}. \\end{equation*}\\] See the previous section for details on interpreting the shape parameter \\(\\eta\\). Note that even if \\(\\eta=1\\), it is still essential to evaluate the density function because the density of \\(L\\) is not constant, regardless of the value of \\(\\eta\\), even though the density of \\(LL^\\top\\) is constant iff \\(\\eta=1\\).\nA lower triangular \\(L\\) is a Cholesky factor for a correlation matrix if and only if \\(L_{k,k} &gt; 0\\) for \\(k \\in 1{:}K\\) and each row \\(L_k\\) has unit Euclidean length.\n\n\n\nL ~ lkj_corr_cholesky(eta)\nIncrement target log probability density with lkj_corr_cholesky_lupdf(L | eta).\nAvailable since 2.4\n \n\n\n\n\n \n\nreal lkj_corr_cholesky_lpdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta\nAvailable since 2.12\n \n\nreal lkj_corr_cholesky_lupdf(matrix L | real eta) The log of the LKJ density for the lower-triangular Cholesky factor L of a correlation matrix given shape eta dropping constant additive terms\nAvailable since 2.25\n \n\nmatrix lkj_corr_cholesky_rng(int K, real eta) Generate a random Cholesky factor of a correlation matrix of order K that is distributed LKJ with shape eta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.4",
     "crumbs": [
-      "Tools and Utilities",
-      "`stansummary`: MCMC Output Analysis"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html",
-    "href": "cmdstan-guide/stan_csv_apdx.html",
-    "title": "Stan CSV File Format",
-    "section": "",
-    "text": "The output from all CmdStan methods is in CSV format. A Stan CSV file is a data table where the columns are the method and model parameters and quantities of interest. Each row contains one record’s worth of data in plain-text format using the comma character (‘,’) as the field delimiter (hence the name).\nFor the Stan CSV files, data is strictly numerical, however, possible values include both positive and negative infinity and “Not-a-Number” which are represented as the strings NaN, inf, +inf, -inf. All other values are written in decimal notation by default with at most 8 digits of precision. The number of significant digits written can be controlled with argument sig_figs=&lt;int&gt;. See more in Output control arguments section.\nStan CSV files have a header row containing the column names. They also make extensive use of CSV comments, i.e., lines which begin with the # character. In addition to initial and final comment rows, some methods also put comment rows in the middle of the data table, which makes it difficult to use many of the commonly used CSV parser packages.\n\n\nThe data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.\nStan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is\n\nThe column name consists of the variable name followed by the element indices.\nIndices are delimited by periods (‘.’).\nIndexing is 1-based, i.e., given a dimension of size \\(N\\), the first element index is \\(1\\) and the last element index is \\(N\\).\nTuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).\n\nContainer variables are serialized in column major order, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.\nTo see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:\n array[2, 3, 4] real foo;\nThe Stan CSV file will require 24 columns to output the elements of foo. The first 6 columns will be labeled:\nfoo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1\nThe final 6 columns will be labeled:\nfoo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4\nTo see how a tuple would be laid out, consider the following variable:\ntuple(real, array[3] real) bar;\nThis will correspond to 4 columns in the CSV file, which are labeled\nbar:1,bar:2.1,bar:2.2,bar:2.3\n\n\n\nThe sample method produces both a Stan CSV output file and a diagnostic file which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.\nTo see how this works, we show snippets of the output file resulting from the following command:\n./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \\\n            data file=bernoulli.data.json \\\n            output file=bernoulli_samples.csv\n\n\nThe sampler output file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nComment rows for adaptation listing step size and metric used for sampling\nSampling draws\nComment rows giving timing information\n\nInitial comments rows: argument configuration\nAll configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as (Default).\nIn the above example the num_samples, num_warmup, and save_warmup arguments were specified, whereas subargument thin is left at its default value, as seen in the initial comment rows:\n# stan_version_major = 2\n# stan_version_minor = 24\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 100\n#     num_warmup = 200\n#     save_warmup = 1\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000000000000003 (Default)\n#       delta = 0.80000000000000004 (Default)\n#       kappa = 0.75 (Default)\n#       t0 = 10 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1 (Default)\n#         stepsize_jitter = 0 (Default)\n# id = 0 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2991989946 (Default)\n# output\n#   file = bernoulli_samples.csv\n#   diagnostic_file = bernoulli_diagnostics.csv\n#   refresh = 100 (Default)\nNote that when running multi-threaded programs which use reduce_sum for high-level parallelization, the number of threads used will also be included in this initial comment header.\nColumn headers\nThe CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the output file section of the chapter on MCMC Sampling. The example model bernoulli.stan only contains one parameter theta, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nAs a second example, we show the output of the eight_schools.stan model on run on example dataset. This model has 3 parameters: mu, theta a vector whose length is dependent on the input data, here N = 8, and tau. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:\nmu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau\nData rows containing warmup draws\nWhen run with option save_warmup=1, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is save_warmup=0, this section is usually not present in the output file.\nHere we specified num_warmup=200 and left thin at the default value \\(1\\), therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.\n-6.74827,1,1,1,1,0,6.75348,0.247195\n-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195\n-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195\n-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619\n-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517\nComment rows for adaptation\nDuring warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.\n# Adaptation terminated\n# Step size = 0.813694\n# Diagonal elements of inverse mass matrix:\n# 0.592879\nAs the example bernoulli model only contains a single parameter, and as the default metric is diag_e, the inverse mass matrix is a \\(1 \\times 1\\) matrix, and the length of the diagonal vector is also \\(1\\).\nIn contrast, if we run the eight schools example model with metric dense_e, the adaptation comments section lists both the stepsize and the full \\(10 \\times 10\\) inverse mass matrix:\n# Adaptation terminated\n# Step size = 0.211252\n# Elements of inverse mass matrix:\n# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757\n# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028\n# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353\n# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097\n# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245\n# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507\n# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946\n# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661\n# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104\n# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433\nNote that when the sampler is run with arguments algorithm=fixed_param, this section will be missing.\nData rows containing sampling draws\nThe output file contains the values for the thinned set draws during sampling. Here we specified num_sampling=100 and left thin at the default value \\(1\\), therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.\n-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093\n-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431\n-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341\nTiming information\nUpon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:\n#\n#  Elapsed Time: 0.005 seconds (Warm-up)\n#                0.002 seconds (Sampling)\n#                0.007 seconds (Total)\n#\n\n\n\nThe diagnostic file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nSampling draws\nComment rows giving timing information\n\nThe columns in this file contain, in order:\n\nall sampler parameters\nall model parameter estimates (on the unconstrained scale)\nthe latent Hamiltonian for each parameter\nthe gradient for each parameters\n\nThe labels for the latent Hamiltonian columns are the parameter column label with prefix p_ and the labels for the gradient columns are the parameter column label with prefix g_.\nThese are the column labels from the file bernoulli_diagnostic.csv:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta\n\n\n\nThe profiling information is stored in a plain CSV format with no meta information in the comments.\nEach row represents timing information collected in a profile statement for a given thread. It is possible that some profile statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).\nThe columns are as follows:\n\nname, The name of the profile statement that is being timed\nthread_id, The thread that executed the profile statement\ntotal_time, The combined time spent executing statements inside the profile which includes calculation with and without automatic differentiation\nforward_time, The time spent in the profile statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation\nreverse_time, The time spent in the profile statement during the reverse (backward) pass of reverse mode automatic differentiation\nchain_stack, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass\nno_chain_stack, The number of objects allocated on the non-chaining automatic differentiation stack\nautodiff_calls, The total number of times the profile statement was executed with automatic differentiation\nno_autodiff_calls - The total number of times the profile statement was executed without automatic differentiation\n\n\n\n\n\n\nConfig as comments\nHeader row\nPenalized maximum likelihood estimate\n\n\n\n\n\nConfig as comments\nHeader row\nAdaptation as comments\nVariational estimate\nSample draws from estimate of the posterior\n\n\n\n\n\nHeader row\nQuantities of interest\n\n\n\n\n\nHeader row\nGradients",
+    "objectID": "functions-reference/correlation_matrix_distributions.html#footnotes",
+    "href": "functions-reference/correlation_matrix_distributions.html#footnotes",
+    "title": "Correlation Matrix Distributions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nIt is possible to build up a valid L within Stan, but that would then require Jacobian adjustments to imply the intended posterior.↩︎",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Correlation Matrix Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html#csv-column-names-and-order",
-    "href": "cmdstan-guide/stan_csv_apdx.html#csv-column-names-and-order",
-    "title": "Stan CSV File Format",
+    "objectID": "functions-reference/continuous_distributions_on_0_1.html",
+    "href": "functions-reference/continuous_distributions_on_0_1.html",
+    "title": "Continuous Distributions on [0, 1]",
     "section": "",
-    "text": "The data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.\nStan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is\n\nThe column name consists of the variable name followed by the element indices.\nIndices are delimited by periods (‘.’).\nIndexing is 1-based, i.e., given a dimension of size \\(N\\), the first element index is \\(1\\) and the last element index is \\(N\\).\nTuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).\n\nContainer variables are serialized in column major order, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.\nTo see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:\n array[2, 3, 4] real foo;\nThe Stan CSV file will require 24 columns to output the elements of foo. The first 6 columns will be labeled:\nfoo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1\nThe final 6 columns will be labeled:\nfoo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4\nTo see how a tuple would be laid out, consider the following variable:\ntuple(real, array[3] real) bar;\nThis will correspond to 4 columns in the CSV file, which are labeled\nbar:1,bar:2.1,bar:2.2,bar:2.3",
+    "text": "The continuous distributions with outcomes in the interval \\([0,1]\\) are used to characterized bounded quantities, including probabilities.\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(\\theta \\in (0,1)\\), \\[\\begin{equation*} \\text{Beta}(\\theta|\\alpha,\\beta) =\n\\frac{1}{\\mathrm{B}(\\alpha,\\beta)} \\, \\theta^{\\alpha - 1} \\, (1 -\n\\theta)^{\\beta - 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, \\(\\alpha, \\beta &gt;\n0\\).\n\n\n\ntheta ~ beta(alpha, beta)\nIncrement target log probability density with beta_lupdf(theta | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_lpdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lupdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_cdf(reals theta | reals alpha, reals beta) The beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.0\n \n\nreal beta_lcdf(reals theta | reals alpha, reals beta) The log of the beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lccdf(reals theta | reals alpha, reals beta) The log of the beta complementary cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nR beta_rng(reals alpha, reals beta) Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in (0, 1)\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(\\theta\n\\in (0,1)\\), \\[\\begin{equation*} \\mathrm{Beta\\_Proportion}(\\theta|\\mu,\\kappa) =\n\\frac{1}{\\mathrm{B}(\\mu \\kappa, (1 - \\mu) \\kappa)} \\,\n\\theta^{\\mu\\kappa - 1} \\, (1 - \\theta)^{(1 - \\mu)\\kappa- 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires \\(\\mu \\in (0, 1)\\) and strictly positive parameter, \\(\\kappa &gt; 0\\).\n\n\n\ntheta ~ beta_proportion(mu, kappa)\nIncrement target log probability density with beta_proportion_lupdf(theta | mu, kappa).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal beta_proportion_lpdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.19\n \n\nreal beta_proportion_lupdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_proportion_lcdf(reals theta | reals mu, reals kappa) The log of the beta_proportion cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nreal beta_proportion_lccdf(reals theta | reals mu, reals kappa) The log of the beta_proportion complementary cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nR beta_proportion_rng(reals mu, reals kappa) Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Continuous Distributions on [0, 1]"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html#mcmc-sampler-csv-output",
-    "href": "cmdstan-guide/stan_csv_apdx.html#mcmc-sampler-csv-output",
-    "title": "Stan CSV File Format",
+    "objectID": "functions-reference/continuous_distributions_on_0_1.html#beta-distribution",
+    "href": "functions-reference/continuous_distributions_on_0_1.html#beta-distribution",
+    "title": "Continuous Distributions on [0, 1]",
     "section": "",
-    "text": "The sample method produces both a Stan CSV output file and a diagnostic file which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.\nTo see how this works, we show snippets of the output file resulting from the following command:\n./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \\\n            data file=bernoulli.data.json \\\n            output file=bernoulli_samples.csv\n\n\nThe sampler output file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nComment rows for adaptation listing step size and metric used for sampling\nSampling draws\nComment rows giving timing information\n\nInitial comments rows: argument configuration\nAll configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as (Default).\nIn the above example the num_samples, num_warmup, and save_warmup arguments were specified, whereas subargument thin is left at its default value, as seen in the initial comment rows:\n# stan_version_major = 2\n# stan_version_minor = 24\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 100\n#     num_warmup = 200\n#     save_warmup = 1\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000000000000003 (Default)\n#       delta = 0.80000000000000004 (Default)\n#       kappa = 0.75 (Default)\n#       t0 = 10 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1 (Default)\n#         stepsize_jitter = 0 (Default)\n# id = 0 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2991989946 (Default)\n# output\n#   file = bernoulli_samples.csv\n#   diagnostic_file = bernoulli_diagnostics.csv\n#   refresh = 100 (Default)\nNote that when running multi-threaded programs which use reduce_sum for high-level parallelization, the number of threads used will also be included in this initial comment header.\nColumn headers\nThe CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the output file section of the chapter on MCMC Sampling. The example model bernoulli.stan only contains one parameter theta, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nAs a second example, we show the output of the eight_schools.stan model on run on example dataset. This model has 3 parameters: mu, theta a vector whose length is dependent on the input data, here N = 8, and tau. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:\nmu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau\nData rows containing warmup draws\nWhen run with option save_warmup=1, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is save_warmup=0, this section is usually not present in the output file.\nHere we specified num_warmup=200 and left thin at the default value \\(1\\), therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.\n-6.74827,1,1,1,1,0,6.75348,0.247195\n-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195\n-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195\n-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619\n-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517\nComment rows for adaptation\nDuring warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.\n# Adaptation terminated\n# Step size = 0.813694\n# Diagonal elements of inverse mass matrix:\n# 0.592879\nAs the example bernoulli model only contains a single parameter, and as the default metric is diag_e, the inverse mass matrix is a \\(1 \\times 1\\) matrix, and the length of the diagonal vector is also \\(1\\).\nIn contrast, if we run the eight schools example model with metric dense_e, the adaptation comments section lists both the stepsize and the full \\(10 \\times 10\\) inverse mass matrix:\n# Adaptation terminated\n# Step size = 0.211252\n# Elements of inverse mass matrix:\n# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757\n# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028\n# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353\n# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097\n# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245\n# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507\n# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946\n# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661\n# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104\n# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433\nNote that when the sampler is run with arguments algorithm=fixed_param, this section will be missing.\nData rows containing sampling draws\nThe output file contains the values for the thinned set draws during sampling. Here we specified num_sampling=100 and left thin at the default value \\(1\\), therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.\n-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093\n-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431\n-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341\nTiming information\nUpon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:\n#\n#  Elapsed Time: 0.005 seconds (Warm-up)\n#                0.002 seconds (Sampling)\n#                0.007 seconds (Total)\n#\n\n\n\nThe diagnostic file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nSampling draws\nComment rows giving timing information\n\nThe columns in this file contain, in order:\n\nall sampler parameters\nall model parameter estimates (on the unconstrained scale)\nthe latent Hamiltonian for each parameter\nthe gradient for each parameters\n\nThe labels for the latent Hamiltonian columns are the parameter column label with prefix p_ and the labels for the gradient columns are the parameter column label with prefix g_.\nThese are the column labels from the file bernoulli_diagnostic.csv:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta\n\n\n\nThe profiling information is stored in a plain CSV format with no meta information in the comments.\nEach row represents timing information collected in a profile statement for a given thread. It is possible that some profile statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).\nThe columns are as follows:\n\nname, The name of the profile statement that is being timed\nthread_id, The thread that executed the profile statement\ntotal_time, The combined time spent executing statements inside the profile which includes calculation with and without automatic differentiation\nforward_time, The time spent in the profile statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation\nreverse_time, The time spent in the profile statement during the reverse (backward) pass of reverse mode automatic differentiation\nchain_stack, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass\nno_chain_stack, The number of objects allocated on the non-chaining automatic differentiation stack\nautodiff_calls, The total number of times the profile statement was executed with automatic differentiation\nno_autodiff_calls - The total number of times the profile statement was executed without automatic differentiation",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(\\theta \\in (0,1)\\), \\[\\begin{equation*} \\text{Beta}(\\theta|\\alpha,\\beta) =\n\\frac{1}{\\mathrm{B}(\\alpha,\\beta)} \\, \\theta^{\\alpha - 1} \\, (1 -\n\\theta)^{\\beta - 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, \\(\\alpha, \\beta &gt;\n0\\).\n\n\n\ntheta ~ beta(alpha, beta)\nIncrement target log probability density with beta_lupdf(theta | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_lpdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lupdf(reals theta | reals alpha, reals beta) The log of the beta density of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_cdf(reals theta | reals alpha, reals beta) The beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.0\n \n\nreal beta_lcdf(reals theta | reals alpha, reals beta) The log of the beta cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nreal beta_lccdf(reals theta | reals alpha, reals beta) The log of the beta complementary cumulative distribution function of theta in \\([0,1]\\) given positive prior successes (plus one) alpha and prior failures (plus one) beta\nAvailable since 2.12\n \n\nR beta_rng(reals alpha, reals beta) Generate a beta variate with positive prior successes (plus one) alpha and prior failures (plus one) beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Continuous Distributions on [0, 1]"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html#optimization-output",
-    "href": "cmdstan-guide/stan_csv_apdx.html#optimization-output",
-    "title": "Stan CSV File Format",
+    "objectID": "functions-reference/continuous_distributions_on_0_1.html#beta-proportion-distribution",
+    "href": "functions-reference/continuous_distributions_on_0_1.html#beta-proportion-distribution",
+    "title": "Continuous Distributions on [0, 1]",
     "section": "",
-    "text": "Config as comments\nHeader row\nPenalized maximum likelihood estimate",
+    "text": "If \\(\\mu \\in (0, 1)\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(\\theta\n\\in (0,1)\\), \\[\\begin{equation*} \\mathrm{Beta\\_Proportion}(\\theta|\\mu,\\kappa) =\n\\frac{1}{\\mathrm{B}(\\mu \\kappa, (1 - \\mu) \\kappa)} \\,\n\\theta^{\\mu\\kappa - 1} \\, (1 - \\theta)^{(1 - \\mu)\\kappa- 1} , \\end{equation*}\\] where the beta function \\(\\mathrm{B}()\\) is as defined in section combinatorial functions.\nWarning: If \\(\\theta = 0\\) or \\(\\theta = 1\\), then the probability is 0 and the log probability is \\(-\\infty\\). Similarly, the distribution requires \\(\\mu \\in (0, 1)\\) and strictly positive parameter, \\(\\kappa &gt; 0\\).\n\n\n\ntheta ~ beta_proportion(mu, kappa)\nIncrement target log probability density with beta_proportion_lupdf(theta | mu, kappa).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal beta_proportion_lpdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.19\n \n\nreal beta_proportion_lupdf(reals theta | reals mu, reals kappa) The log of the beta_proportion density of theta in \\((0,1)\\) given mean mu and precision kappa dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_proportion_lcdf(reals theta | reals mu, reals kappa) The log of the beta_proportion cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nreal beta_proportion_lccdf(reals theta | reals mu, reals kappa) The log of the beta_proportion complementary cumulative distribution function of theta in \\((0,1)\\) given mean mu and precision kappa\nAvailable since 2.18\n \n\nR beta_proportion_rng(reals mu, reals kappa) Generate a beta_proportion variate with mean mu and precision kappa; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Continuous Distributions on [0, 1]"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html#variational-inference-output",
-    "href": "cmdstan-guide/stan_csv_apdx.html#variational-inference-output",
-    "title": "Stan CSV File Format",
+    "objectID": "functions-reference/complex_matrix_operations.html",
+    "href": "functions-reference/complex_matrix_operations.html",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "Config as comments\nHeader row\nAdaptation as comments\nVariational estimate\nSample draws from estimate of the posterior",
+    "text": "This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.\n\n\nIf an expression e can be assigned to a variable of type T, then it can be used as an argument to a function that is specified to take arguments of type T. For instance, sqrt(real) is specified to take a real argument, but an integer expression such as 2 + 2 of type int can be passed to sqrt, so that sqrt(2 + 2) is well defined. This works by promoting the integer expression 2 + 2 to be of real type.\nThe rules for promotion in Stan are simple:\n\nint may be promoted to real,\nreal may be promoted to complex,\nvector can be promoted to complex_vector,\nrow_vector can be promoted to complex_row_vector,\nmatrix can be promoted to complex_matrix,\nif T can be promoted to U and U can be promoted to V, then T can be promoted to V (transitive), and\nif T can be promoted to U, then T[] can be promoted to U[] (covariant).\n\n\n\n\nWhen a function is called, the definition requiring the fewest number of promotions is used. For example, when calling vector + vector, the real-valued signature is used. When calling any of complex_vector + vector, vector + complex_vector, or complex_vector + complex_vector, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.\n\n\n\nComplex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, complex operator+(complex_vector, complex_vector). Through promotion, operator+ may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures operator+(complex_vector, vector) and operator+(vector, complex_vector).\n\n\n\nGeneric functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions append_array, dims, head, num_elements, rep_array, reverse, segment, size, and tail.\n\n\n\n\n \n\nint num_elements(complex_vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.30\n \n\nint num_elements(complex_row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.30\n \n\nint num_elements(complex_matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.30\n \n\nint rows(complex_vector x) The number of rows in the vector x\nAvailable since 2.30\n \n\nint rows(complex_row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.30\n \n\nint rows(complex_matrix x) The number of rows in the matrix x\nAvailable since 2.30\n \n\nint cols(complex_vector x) The number of columns in the vector x, namely 1\nAvailable since 2.30\n \n\nint cols(complex_row_vector x) The number of columns in the row vector x\nAvailable since 2.30\n \n\nint cols(complex_matrix x) The number of columns in the matrix x\nAvailable since 2.30\n \n\nint size(complex_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(complex_row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15.\nAvailable since 2.30\n\n\n\nStan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\ncomplex_vector operator-(complex_vector x) The negation of the vector x.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x) The negation of the row vector x.\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x) The negation of the matrix x.\nAvailable since 2.30\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex_vector y) The sum of the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex_row_vector y) The sum of the row vectors x and y.\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex_matrix y) The sum of the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex_vector y) The difference between the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex_row_vector y) The difference between the row vectors x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex_matrix y) The difference between the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex x, complex_vector y) The product of the scalar x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex x, complex_row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex x, complex_matrix y) The product of the scalar x and the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_vector x, complex y) The product of the scalar y and vector x\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_vector x, complex_row_vector y) The product of the vector x and row vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex y) The product of the scalar y and row vector x\nAvailable since 2.30\n \n\ncomplex operator*(complex_row_vector x, complex_vector y) The product of the row vector x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex_matrix y) The product of the row vector x and matrix y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex y) The product of the scalar y and matrix x\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_matrix x, complex_vector y) The product of the matrix x and vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex_matrix y) The product of the matrices x and y\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex y) The result of adding y to every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator+(complex x, complex_vector y) The result of adding x to every entry in the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex y) The result of adding y to every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex x, complex_row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex y) The result of adding y to every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex x, complex_matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex y) The result of subtracting y from every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex x, complex_vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex x, complex_row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex x, complex_matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator/(complex_vector x, complex y) The result of dividing each entry in the vector x by y\nAvailable since 2.30\n \n\ncomplex_row_vector operator/(complex_row_vector x, complex y) The result of dividing each entry in the row vector x by y\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix x, complex y) The result of dividing each entry in the matrix x by y\nAvailable since 2.30\n\n\n\n\nComplex complex_matrix transposition is represented using a postfix operator.\n \n\ncomplex_matrix operator'(complex_matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.30\n \n\ncomplex_row_vector operator'(complex_vector x) The transpose of the vector x, written as x'\nAvailable since 2.30\n \n\ncomplex_vector operator'(complex_row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.30\n\n\n\nAs in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.\n \n\ncomplex_vector operator.*(complex_vector x, complex_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator.*(complex_row_vector x, complex_row_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator.*(complex_matrix x, complex_matrix y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^(complex x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n\n\n\n \n\ncomplex dot_product(complex_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_vector x, complex_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_matrix x, complex_matrix y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_vector x, complex_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_matrix x, complex_matrix y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex dot_self(complex_vector x) The dot product of the vector x with itself\nAvailable since 2.30\n \n\ncomplex dot_self(complex_row_vector x) The dot product of the row vector x with itself\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_matrix x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_matrix x) The dot product of the rows of x with themselves\nAvailable since 2.30\n\n\n \n\ncomplex_matrix diag_pre_multiply(complex_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_pre_multiply(complex_row_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_row_vector v) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex sum(complex_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n\n\n\n\nMuch like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.\n\n\nThese functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a real type. For example, get_real(complex_matrix M) yields a matrix containing the real component of each value in M.\nThe following table contains examples of what this notation can mean:\n\n\n\nType T\nType T_demoted\n\n\n\n\ncomplex\nreal\n\n\ncomplex_vector\nvector\n\n\ncomplex_row_vector\nrow_vector\n\n\ncomplex_matrix\nmatrix\n\n\narray[] complex\narray[] real\n\n\narray[,,] complex\narray[,,] real\n\n\n\n\n\n\n \n\nT_demoted get_real(T x) Given an object of complex type T, return the same shape object but of type real by getting the real component of each element of x.\nAvailable since 2.30\n \n\nT_demoted get_imag(T x) Given an object of complex type T, return the same shape object but of type real by getting the imaginary component of each element of x.\nAvailable since 2.30\nFor example, given the Stan declaration\n  complex_vector[2] z = [3+4i, 5+6i]';\nA call get_real(z) will yield the vector [3, 5]', and a call get_imag(z) will yield the vector [4, 6]'.\n\n\n\n\nThe following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\ncomplex_vector rep_vector(complex z, int m) Return the size m (column) vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_row_vector rep_row_vector(complex z, int n) Return the size n row vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex z, int m, int n) Return the m by n matrix consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.30\n\n\n \n\ncomplex_matrix symmetrize_from_lower_tri(complex_matrix A) Construct a symmetric matrix from the lower triangle of A.\nAvailable since 2.30\n\n\n\n\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.30\n \n\ncomplex_vector diagonal(complex_matrix x) The diagonal of the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix diag_matrix(complex_vector x) The diagonal matrix with diagonal x\nAvailable since 2.30\n\n\n\nStan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\ncomplex_vector col(complex_matrix x, int n) The n-th column of matrix x\nAvailable since 2.30\n \n\ncomplex_row_vector row(complex_matrix x, int m) The m-th row of matrix x\nAvailable since 2.30\n\n\n\n\n\n \n\ncomplex_matrix block(complex_matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.30\n \n\ncomplex_vector sub_col(complex_matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.30\n \n\ncomplex_row_vector sub_row(complex_matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector head(complex_vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector head(complex_row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector tail(complex_vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector tail(complex_row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector segment(complex_vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n \n\ncomplex_row_vector segment(complex_row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n\n\n\n\n\n\n\n \n\ncomplex_matrix append_col(complex_matrix x, complex_matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_matrix x, complex_vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex_row_vector y) Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex x, complex_row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex y) Append y to the end of x, returning another row vector.\nAvailable since 2.30\n\n\n\n \n\ncomplex_matrix append_row(complex_matrix x, complex_matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_matrix x, complex_row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex_vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex x, complex_vector y) Append x to the top of y, returning another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex y) Append y to the bottom of x, returning another vector.\nAvailable since 2.30\n\n\n\n\n\n\nStan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors u and v,\n    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v\nand in the 2-dimensional case for complex matrices A and B,\n    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B\nAlthough the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.\n \n\ncomplex_vector fft(complex_vector v) Return the discrete Fourier transform of the specified complex vector v. If \\(v \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(u =\n\\textrm{fft}(v)\\), then \\[\\begin{equation*}\nu_n = \\sum_{m &lt; n}\nv_m \\cdot\n\\exp\\left(\\frac{-n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\]\nAvailable since 2.30\n \n\ncomplex_matrix fft2(complex_matrix m) Return the 2D discrete Fourier transform of the specified complex matrix m. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.\nAvailable since 2.30\n \n\ncomplex_vector inv_fft(complex_vector u) Return the inverse of the discrete Fourier transform of the specified complex vector u. The inverse FFT (this function) is scaled so that fft(inv_fft(u)) == u. If \\(u \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(v = \\textrm{fft}^{-1}(u)\\), then \\[\\begin{equation*}\nv_n =  \\frac{1}{N} \\sum_{m &lt; n}\nu_m \\cdot\n\\exp\\left(\\frac{n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\] This only differs from the FFT by the sign inside the exponential and the scaling. The \\(\\frac{1}{N}\\) scaling ensures that fft(inv_fft(u)) == u and inv_fft(fft(v)) == v for complex vectors u and v.\nAvailable since 2.30\n \n\ncomplex_matrix inv_fft2(complex_matrix m) Return the inverse of the 2D discrete Fourier transform of the specified complex matrix m. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures fft2(inv_fft2(A)) == A and inv_fft2(fft2(B)) == B.\nAvailable since 2.30\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] complex cumulative_sum(array[] complex x) The cumulative sum of x\nAvailable since 2.30\n \n\ncomplex_vector cumulative_sum(complex_vector v) The cumulative sum of v\nAvailable since 2.30\n \n\ncomplex_row_vector cumulative_sum(complex_row_vector rv) The cumulative sum of rv\nAvailable since 2.30\n\n\n\n\n\n\nIn general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.\n\n\n \n\ncomplex_row_vector operator/(complex_row_vector b, complex_matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix B, complex_matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex trace(complex_matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector eigenvalues(complex_matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.32\n \n\ncomplex_matrix eigenvectors(complex_matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.32\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\ncomplex_vector eigenvalues_sym(complex_matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors_sym(complex_matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose_sym(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(complex_matrix A) The singular values of A in descending order\nAvailable since 2.30\n \n\ncomplex_matrix svd_U(complex_matrix A) The left-singular vectors of A\nAvailable since 2.30\n \n\ncomplex_matrix svd_V(complex_matrix A) The right-singular vectors of A\nAvailable since 2.30\n \n\ntuple(complex_matrix, vector, complex_matrix) svd(complex_matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\nThe complex Schur decomposition of a square matrix \\(A\\) produces a complex unitary matrix \\(U\\) and a complex upper-triangular Schur form matrix \\(T\\) such that \\[A = U \\cdot T \\cdot U^{-1}\\]\nSince \\(U\\) is unitary, its inverse is also its conjugate transpose, \\(U^{-1} = U^*\\), \\(U^*(i, j) = \\mathrm{conj}(U(j, i))\\)\n \n\ncomplex_matrix complex_schur_decompose_t(matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_t(complex_matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(complex_matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results. This overload is equivalent to complex_schur_decompose(to_complex(A,0)) but is more efficient.\nAvailable since 2.33\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(complex_matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n\n\n\n\n\n \n\ncomplex_vector reverse(complex_vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.30\n \n\ncomplex_row_vector reverse(complex_row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.30",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html#generate-quantities-outputs",
-    "href": "cmdstan-guide/stan_csv_apdx.html#generate-quantities-outputs",
-    "title": "Stan CSV File Format",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-promotion",
+    "href": "functions-reference/complex_matrix_operations.html#complex-promotion",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "Header row\nQuantities of interest",
+    "text": "This chapter provides the details of functions that operate over complex matrices, vectors, and row vectors. These mirror the operations over real complex_matrix types and are defined in the usual way for complex numbers.\n\n\nIf an expression e can be assigned to a variable of type T, then it can be used as an argument to a function that is specified to take arguments of type T. For instance, sqrt(real) is specified to take a real argument, but an integer expression such as 2 + 2 of type int can be passed to sqrt, so that sqrt(2 + 2) is well defined. This works by promoting the integer expression 2 + 2 to be of real type.\nThe rules for promotion in Stan are simple:\n\nint may be promoted to real,\nreal may be promoted to complex,\nvector can be promoted to complex_vector,\nrow_vector can be promoted to complex_row_vector,\nmatrix can be promoted to complex_matrix,\nif T can be promoted to U and U can be promoted to V, then T can be promoted to V (transitive), and\nif T can be promoted to U, then T[] can be promoted to U[] (covariant).\n\n\n\n\nWhen a function is called, the definition requiring the fewest number of promotions is used. For example, when calling vector + vector, the real-valued signature is used. When calling any of complex_vector + vector, vector + complex_vector, or complex_vector + complex_vector, the complex signature is used. If more than one signature matches with a the minimal number of promotions, the call is ambiguous, and an error will be raised by the compiler. Promotion ambiguity leading to ill-defined calls should never happen with Stan built-in functions.\n\n\n\nComplex function signatures will only list the fully complex type. For example, with complex vector addition, we will list a single signature, complex operator+(complex_vector, complex_vector). Through promotion, operator+ may be called with one complex vector and one real vector as well, but the documentation elides the implied signatures operator+(complex_vector, vector) and operator+(vector, complex_vector).\n\n\n\nGeneric functions work for arrays containing complex, complex matrix, complex vector, or complex row vector types. This includes the functions append_array, dims, head, num_elements, rep_array, reverse, segment, size, and tail.",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
-    ]
-  },
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
+    ]
+  },
   {
-    "objectID": "cmdstan-guide/stan_csv_apdx.html#diagnose-method-outputs",
-    "href": "cmdstan-guide/stan_csv_apdx.html#diagnose-method-outputs",
-    "title": "Stan CSV File Format",
+    "objectID": "functions-reference/complex_matrix_operations.html#integer-valued-complex-matrix-size-functions",
+    "href": "functions-reference/complex_matrix_operations.html#integer-valued-complex-matrix-size-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "Header row\nGradients",
+    "text": "int num_elements(complex_vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.30\n \n\nint num_elements(complex_row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.30\n \n\nint num_elements(complex_matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.30\n \n\nint rows(complex_vector x) The number of rows in the vector x\nAvailable since 2.30\n \n\nint rows(complex_row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.30\n \n\nint rows(complex_matrix x) The number of rows in the matrix x\nAvailable since 2.30\n \n\nint cols(complex_vector x) The number of columns in the vector x, namely 1\nAvailable since 2.30\n \n\nint cols(complex_row_vector x) The number of columns in the row vector x\nAvailable since 2.30\n \n\nint cols(complex_matrix x) The number of columns in the matrix x\nAvailable since 2.30\n \n\nint size(complex_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(complex_row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.30\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15.\nAvailable since 2.30",
     "crumbs": [
-      "Appendices",
-      "Stan CSV File Format"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/print.html",
-    "href": "cmdstan-guide/print.html",
-    "title": "print (deprecated): MCMC Output Analysis",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-matrix-arithmetic-operators",
+    "href": "functions-reference/complex_matrix_operations.html#complex-matrix-arithmetic-operators",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "print (deprecated): MCMC Output Analysis\nThe print utility is deprecated, but is still available until CmdStan v3.0. It has been replaced by the stansummary utility.\n\n\n\n\n Back to top",
+    "text": "Stan supports all basic complex arithmetic operators using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\ncomplex_vector operator-(complex_vector x) The negation of the vector x.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x) The negation of the row vector x.\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x) The negation of the matrix x.\nAvailable since 2.30\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex_vector y) The sum of the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex_row_vector y) The sum of the row vectors x and y.\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex_matrix y) The sum of the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex_vector y) The difference between the vectors x and y.\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex_row_vector y) The difference between the row vectors x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex_matrix y) The difference between the matrices x and y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex x, complex_vector y) The product of the scalar x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex x, complex_row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex x, complex_matrix y) The product of the scalar x and the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_vector x, complex y) The product of the scalar y and vector x\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_vector x, complex_row_vector y) The product of the vector x and row vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex y) The product of the scalar y and row vector x\nAvailable since 2.30\n \n\ncomplex operator*(complex_row_vector x, complex_vector y) The product of the row vector x and vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator*(complex_row_vector x, complex_matrix y) The product of the row vector x and matrix y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex y) The product of the scalar y and matrix x\nAvailable since 2.30\n \n\ncomplex_vector operator*(complex_matrix x, complex_vector y) The product of the matrix x and vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator*(complex_matrix x, complex_matrix y) The product of the matrices x and y\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector operator+(complex_vector x, complex y) The result of adding y to every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator+(complex x, complex_vector y) The result of adding x to every entry in the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex_row_vector x, complex y) The result of adding y to every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator+(complex x, complex_row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex_matrix x, complex y) The result of adding y to every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator+(complex x, complex_matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex_vector x, complex y) The result of subtracting y from every entry in the vector x\nAvailable since 2.30\n \n\ncomplex_vector operator-(complex x, complex_vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex_row_vector x, complex y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.30\n \n\ncomplex_row_vector operator-(complex x, complex_row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex_matrix x, complex y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix operator-(complex x, complex_matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.30\n \n\ncomplex_vector operator/(complex_vector x, complex y) The result of dividing each entry in the vector x by y\nAvailable since 2.30\n \n\ncomplex_row_vector operator/(complex_row_vector x, complex y) The result of dividing each entry in the row vector x by y\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix x, complex y) The result of dividing each entry in the matrix x by y\nAvailable since 2.30",
     "crumbs": [
-      "Tools and Utilities",
-      "`print` (deprecated): MCMC Output Analysis"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/parallelization.html",
-    "href": "cmdstan-guide/parallelization.html",
-    "title": "Parallelization",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-transposition-operator",
+    "href": "functions-reference/complex_matrix_operations.html#complex-transposition-operator",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "Stan provides three ways of parallelizing execution of a Stan model:\n\nmulti-threading with Intel Threading Building Blocks (TBB),\nmulti-processing with Message Passing Interface (MPI) and\nmanycore processing with OpenCL.\n\n\n\nIn order to exploit multi-threading in a Stan model, the models must be rewritten to use the reduce_sum and map_rect functions. For instructions on how to rewrite Stan models to use these functions see Stan’s User guide chapter on parallelization, the reduce_sum case study or the Multithreading and Map-Reduce tutorial.\n\n\nOnce a model is rewritten to use the above-mentioned functions, the model must be compiled with the STAN_THREADS makefile flag. The flag can be supplied in the make call but we recommend writing the flag to the make/local file. If the STAN_THREADS flag is defined/non-empty, threads will be enabled.\nAn example of the contents of make/local to enable threading with TBB:\nSTAN_THREADS=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nBefore running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the num_threads argument. Valid values for num_threads are positive integers and -1. If num_threads is set to -1, all available cores will be used.\nGenerally, this number should not exceed the number of available cores for best performance.\nExample:\n./model sample data file=data.json num_threads=4 ...\nWhen the model is compiled with STAN_THREADS we can sample with multiple chains with a single executable (see section running multiple chains for cases when this is available). When running multiple chains num_threads is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:\n./model sample num_chains=2 data file=data.json num_threads=8 ...\n\n\n\n\nIn order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the map_rect function. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.\n\n\nCompiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are MPICH and OpenMPI.\n\n\n\nOnce a model is rewritten to use map_rect, additional makefile flags must be written to the make/local. These are:\n\nSTAN_MPI: Enables the use of MPI with Stan if defined.\nCXX: The name of the MPI C++ compiler wrapper. Typically mpicxx.\nTBB_CXX_TYPE: The C++ compiler the MPI wrapper wraps. Typically gcc on Linux and clang on macOS.\n\nAn example of make/local on Linux:\nSTAN_MPI=true\nCXX=mpicxx\nTBB_CXX_TYPE=gcc\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_MPI is run using an MPI launcher. The MPI standard suggests using mpiexec, but a vendor wrapper for the launcher like mpirun can also be used. The launcher is supplied the path to the built executable and the number of processes to start: -n X for mpiexec or -np X for mpirun where X is replaced by the integer representing the number of processes.\nExample for running a model with six processes:\nmpiexec -n 6 path/to/model sample data file=data.json ...\n\n\n\n\n\n\nOpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.\nIn order to check if any OpenCL-enabled device and its runtime is already present use the clinfo tool. On Linux, clinfo can typically be installed with the default package manager (for example sudo apt-get install clinfo on Ubuntu). For Windows, pre-built clinfo binary can be found here.\nAlso use clinfo to verify successful installation of OpenCL runtimes.\n\n\n\nLinux:\nInstall the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:\nsudo apt update\nsudo apt install nvidia-driver-460 nvidia-cuda-toolkit\nReplace the driver version (460 in the above case) with the lastest number at the time of installation.\nWindows:\nInstall the NVIDIA GPU Driver and CUDA Toolkit.\n\n\n\n\n\nLinux:\nInstall Radeon Software for Linux available here.\nWindows:\nWe recommend installing the open source OCL-SDK.\n\n\n\n\nInstall the open source PoCL.\n\n\n\nFollow Intel’s install instructions given here (requires registration).\n\n\n\n\nIn order to enable the OpenCL backend the model must be compiled with the STAN_OPENCL makefile flag defined/non-empty. The flag can be supplied in the make call but we recommend writing the flag to the make/local file.\nAn example of the contents of make/local to enable parallelization with OpenCL:\nSTAN_OPENCL=true\nIf you are using OpenCL with an integrated GPU you also need to define the INTEGRATED_OPENCL flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:\nINTEGRATED_OPENCL=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_OPENCL can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the clinfo program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically 0. In that case you can also omit the OpenCL IDs as the default 0 IDs are used in that case.\nWe supply these IDs when starting the executable as shown below:\npath/to/model sample data file=data.json opencl platform=0 device=1",
+    "text": "Complex complex_matrix transposition is represented using a postfix operator.\n \n\ncomplex_matrix operator'(complex_matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.30\n \n\ncomplex_row_vector operator'(complex_vector x) The transpose of the vector x, written as x'\nAvailable since 2.30\n \n\ncomplex_vector operator'(complex_row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.30",
     "crumbs": [
-      "Getting Started",
-      "Parallelization"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/parallelization.html#multi-threading-with-tbb",
-    "href": "cmdstan-guide/parallelization.html#multi-threading-with-tbb",
-    "title": "Parallelization",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-elementwise-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-elementwise-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "In order to exploit multi-threading in a Stan model, the models must be rewritten to use the reduce_sum and map_rect functions. For instructions on how to rewrite Stan models to use these functions see Stan’s User guide chapter on parallelization, the reduce_sum case study or the Multithreading and Map-Reduce tutorial.\n\n\nOnce a model is rewritten to use the above-mentioned functions, the model must be compiled with the STAN_THREADS makefile flag. The flag can be supplied in the make call but we recommend writing the flag to the make/local file. If the STAN_THREADS flag is defined/non-empty, threads will be enabled.\nAn example of the contents of make/local to enable threading with TBB:\nSTAN_THREADS=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nBefore running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the num_threads argument. Valid values for num_threads are positive integers and -1. If num_threads is set to -1, all available cores will be used.\nGenerally, this number should not exceed the number of available cores for best performance.\nExample:\n./model sample data file=data.json num_threads=4 ...\nWhen the model is compiled with STAN_THREADS we can sample with multiple chains with a single executable (see section running multiple chains for cases when this is available). When running multiple chains num_threads is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:\n./model sample num_chains=2 data file=data.json num_threads=8 ...",
+    "text": "As in the real case, elementwise complex functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument.\n \n\ncomplex_vector operator.*(complex_vector x, complex_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator.*(complex_row_vector x, complex_row_vector y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator.*(complex_matrix x, complex_matrix y) The elementwise product of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex x, complex_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_vector operator./(complex_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex x, complex_row_vector y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector operator./(complex_row_vector x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex x, complex_matrix y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\ncomplex_matrix operator./(complex_matrix x, complex y) The elementwise quotient of x and y\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nvector operator.^(complex x, complex_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex_row_vector x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nrow_vector operator.^(complex x, complex_row_vector y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^( complex_matrix x, complex y) The elementwise power of y and x\nAvailable since 2.30\n \n\nmatrix operator.^(complex x, complex_matrix y) The elementwise power of y and x\nAvailable since 2.30",
     "crumbs": [
-      "Getting Started",
-      "Parallelization"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/parallelization.html#multi-processing-with-mpi",
-    "href": "cmdstan-guide/parallelization.html#multi-processing-with-mpi",
-    "title": "Parallelization",
+    "objectID": "functions-reference/complex_matrix_operations.html#dot-products-and-specialized-products-for-complex-matrices",
+    "href": "functions-reference/complex_matrix_operations.html#dot-products-and-specialized-products-for-complex-matrices",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "In order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the map_rect function. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.\n\n\nCompiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are MPICH and OpenMPI.\n\n\n\nOnce a model is rewritten to use map_rect, additional makefile flags must be written to the make/local. These are:\n\nSTAN_MPI: Enables the use of MPI with Stan if defined.\nCXX: The name of the MPI C++ compiler wrapper. Typically mpicxx.\nTBB_CXX_TYPE: The C++ compiler the MPI wrapper wraps. Typically gcc on Linux and clang on macOS.\n\nAn example of make/local on Linux:\nSTAN_MPI=true\nCXX=mpicxx\nTBB_CXX_TYPE=gcc\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_MPI is run using an MPI launcher. The MPI standard suggests using mpiexec, but a vendor wrapper for the launcher like mpirun can also be used. The launcher is supplied the path to the built executable and the number of processes to start: -n X for mpiexec or -np X for mpirun where X is replaced by the integer representing the number of processes.\nExample for running a model with six processes:\nmpiexec -n 6 path/to/model sample data file=data.json ...",
+    "text": "complex dot_product(complex_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex dot_product(complex_row_vector x, complex_row_vector y) The dot product of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_vector x, complex_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_product(complex_matrix x, complex_matrix y) The dot product of the columns of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_vector x, complex_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_row_vector x, complex_row_vector y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_product(complex_matrix x, complex_matrix y) The dot product of the rows of x and y\nAvailable since 2.30\n \n\ncomplex dot_self(complex_vector x) The dot product of the vector x with itself\nAvailable since 2.30\n \n\ncomplex dot_self(complex_row_vector x) The dot product of the row vector x with itself\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_row_vector columns_dot_self(complex_matrix x) The dot product of the columns of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.30\n \n\ncomplex_vector rows_dot_self(complex_matrix x) The dot product of the rows of x with themselves\nAvailable since 2.30\n\n\n \n\ncomplex_matrix diag_pre_multiply(complex_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_pre_multiply(complex_row_vector v, complex_matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.30\n \n\ncomplex_matrix diag_post_multiply(complex_matrix m, complex_row_vector v) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.30",
     "crumbs": [
-      "Getting Started",
-      "Parallelization"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/parallelization.html#opencl",
-    "href": "cmdstan-guide/parallelization.html#opencl",
-    "title": "Parallelization",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-reductions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-reductions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "OpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.\nIn order to check if any OpenCL-enabled device and its runtime is already present use the clinfo tool. On Linux, clinfo can typically be installed with the default package manager (for example sudo apt-get install clinfo on Ubuntu). For Windows, pre-built clinfo binary can be found here.\nAlso use clinfo to verify successful installation of OpenCL runtimes.\n\n\n\nLinux:\nInstall the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:\nsudo apt update\nsudo apt install nvidia-driver-460 nvidia-cuda-toolkit\nReplace the driver version (460 in the above case) with the lastest number at the time of installation.\nWindows:\nInstall the NVIDIA GPU Driver and CUDA Toolkit.\n\n\n\n\n\nLinux:\nInstall Radeon Software for Linux available here.\nWindows:\nWe recommend installing the open source OCL-SDK.\n\n\n\n\nInstall the open source PoCL.\n\n\n\nFollow Intel’s install instructions given here (requires registration).\n\n\n\n\nIn order to enable the OpenCL backend the model must be compiled with the STAN_OPENCL makefile flag defined/non-empty. The flag can be supplied in the make call but we recommend writing the flag to the make/local file.\nAn example of the contents of make/local to enable parallelization with OpenCL:\nSTAN_OPENCL=true\nIf you are using OpenCL with an integrated GPU you also need to define the INTEGRATED_OPENCL flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:\nINTEGRATED_OPENCL=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_OPENCL can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the clinfo program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically 0. In that case you can also omit the OpenCL IDs as the default 0 IDs are used in that case.\nWe supply these IDs when starting the executable as shown below:\npath/to/model sample data file=data.json opencl platform=0 device=1",
+    "text": "complex sum(complex_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex sum(complex_matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30\n \n\ncomplex prod(complex_matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.30",
     "crumbs": [
-      "Getting Started",
-      "Parallelization"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html",
-    "href": "cmdstan-guide/mcmc_config.html",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#vectorized-accessor-functions",
+    "href": "functions-reference/complex_matrix_operations.html#vectorized-accessor-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "The sample method provides Bayesian inference over the model conditioned on data using Hamiltonian Monte Carlo (HMC) sampling. By default, the inference engine used is the No-U-Turn sampler (NUTS), an adaptive form of Hamiltonian Monte Carlo sampling. For details on HMC and NUTS, see the Stan Reference Manual chapter on MCMC Sampling.\n\n\nTo generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument sample or method=sample together with the input data. The executable can be run from any directory.\nThe full set of configuration options available for the sample method is available by using the sample help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.\nHere, we run it in the directory which contains the Stan program and input data, &lt;cmdstan-home&gt;/examples/bernoulli:\n&gt; cd examples/bernoulli\n&gt; ls\n  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan\nTo execute sampling of the model under Linux or Mac, use:\n&gt; ./bernoulli sample data file=bernoulli.data.json\nIn Windows, the ./ prefix is not needed:\n&gt; bernoulli.exe sample data file=bernoulli.data.json\nThe output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:\nmethod = sample (Default)\n  sample\n    num_samples = 1000 (Default)\n    num_warmup = 1000 (Default)\n    save_warmup = false (Default)\n    thin = 1 (Default)\n    adapt\n      engaged = true (Default)\n      gamma = 0.050000000000000003 (Default)\n      delta = 0.80000000000000004 (Default)\n      kappa = 0.75 (Default)\n      t0 = 10 (Default)\n      init_buffer = 75 (Default)\n      term_buffer = 50 (Default)\n      window = 25 (Default)\n      save_metric = false (Default)\n    algorithm = hmc (Default)\n      hmc\n        engine = nuts (Default)\n          nuts\n            max_depth = 10 (Default)\n        metric = diag_e (Default)\n        metric_file =  (Default)\n        stepsize = 1 (Default)\n        stepsize_jitter = 0 (Default)\n    num_chains = 1 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 3252652196 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\nAfter the configuration has been displayed, a short timing message is given.\nGradient evaluation took 1.2e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.\nAdjust your expectations accordingly!\nNext, the sampler reports the iteration number, reporting the percentage complete.\nIteration:    1 / 2000 [  0%]  (Warmup)\n...\nIteration: 2000 / 2000 [100%]  (Sampling)\nFinally, the sampler reports timing information:\n Elapsed Time: 0.007 seconds (Warm-up)\n               0.017 seconds (Sampling)\n               0.024 seconds (Total)\n\n\n\nEach execution of the model results in draws from a single Markov chain being written to a file in comma-separated value (CSV) format. The default name of the output file is output.csv.\nThe first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (#)).\nWhen the example model bernoulli.stan is run via the command line with all default arguments, the following configuration is displayed:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 1000 (Default)\n#     num_warmup = 1000 (Default)\n#     save_warmup = false (Default)\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000 (Default)\n#       delta = 0.800000 (Default)\n#       kappa = 0.750000 (Default)\n#       t0 = 10.000000 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#       save_metric = false (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1.000000 (Default)\n#         stepsize_jitter = 0.000000 (Default)\n#     num_chains = 1 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\nThis is followed by a CSV header indicating the names of the values sampled.\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nThe first output columns report the HMC sampler information:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause the above header is from the NUTS sampler, it has columns treedepth__, n_leapfrog__, and divergent__ and doesn’t have column int_time__. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, theta.\nThe header line is written to the output file before warmup begins. If option save_warmup is set to true, the warmup draws are output directly after the header. The total number of warmup draws saved is num_warmup divided by thin, rounded up (i.e., ceiling).\nFollowing the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:\n# Adaptation terminated\n# Step size = 0.884484\n# Diagonal elements of inverse mass matrix:\n# 0.535006\nThe default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter theta.\nDraws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.\n-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853\n-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295\n-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299\n-6.88712,1,0.884484,1,1,0,7.02101,0.188229\n-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596\n...\nThe output ends with timing details:\n#  Elapsed Time: 0.007 seconds (Warm-up)\n#                0.017 seconds (Sampling)\n#                0.024 seconds (Total)\n\n\n\nAt every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters metric and stepsize in order to efficiently sample from typical set, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.\nThe following keyword-value arguments control the total number of iterations:\n\nnum_samples\nnum_warmup\nsave_warmup\nthin\n\nThe values for arguments num_samples and num_warmup must be a non-negative integer. The default value for both is \\(1000\\).\nFor well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.\nThe number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is \\(\\sqrt{N}\\); therefore every additional decimal place of accuracy increases this by a factor of 10.\nArgument save_warmup takes values false or true. The default value is false, i.e., warmup draws are not saved to the output file. When the value is true, the warmup draws are written to the CSV output file directly after the CSV header line.\nArgument thin controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce anticorrelated draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.\nThe value of argument thin must be a positive integer. When thin is set to value \\(N\\), every \\(N^{th}\\) iteration is written to the output file. Should the value of thin exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of thin equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When num_sampling=M and thin=N, the number of iterations written to the output CSV file will be ceiling(M/N). If save_warmup=true, thinning is applied to the warmup iterations as well.\n\n\n\nThe adapt keyword is used to specify non-default options for the sampler adaptation schedule and settings.\nAdaptation can be turned off by setting sub-argument engaged to value false. If engaged=false, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is engaged=1, this keyword-value pair can be omitted from the command.\nThere are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section Automatic Parameter Tuning.\nThe boolean sub-argument save_metric was added in Stan version 2.34. When save_metric=true, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix _metric.json, e.g., if using the default output filename output.csv, the saved metric file will be output_metric.json. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument metric_file.\n\n\nThe Stan User’s Guide section on model conditioning and curvature provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on HMC algorithm parameters explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.\nThe following keyword-value arguments control the settings used to optimize the step size:\n\ndelta - The target Metropolis acceptance rate. The default value is \\(0.8\\). Its value must be strictly between \\(0\\) and \\(1\\). Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of delta will also allow some models that would otherwise get stuck to overcome their blockages.  Models with difficult posterior geometries may required increasing the delta argument closer to \\(1\\); we recommend first trying to raise it to \\(0.9\\) or at most \\(0.95\\). Values about \\(0.95\\) are strong indication of bad geometry; the better solution is to change the model geometry through reparameterization which could yield both more efficient and faster sampling.\ngamma - Adaptation regularization scale. Must be a positive real number, default value is \\(0.05\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nkappa - Adaptation relaxation exponent. Must be a positive real number, default value is \\(0.75\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nt_0 - Adaptation iteration offset. Must be a positive real number, default value is \\(10\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\n\n\n\n\nWhen adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:\n\ninit_buffer - The number of iterations spent tuning the step size at the outset of adaptation.\nwindow - The initial number of iterations devoted to tune the metric, will be doubled successively.\nterm_buffer - The number of iterations used to re-tune the step size once the metric has been tuned.\n\nThe specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.\nThe following figure is taken from the Stan Reference Manual, where label “I” correspond to init_buffer, the initial “II” corresponds to window, and the final “III” corresponds to term_buffer:\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\n\n\n\n\nThe algorithm keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: hmc, which generates from an HMC-driven Markov chain; and fixed_param which generates a new sample without changing the state of the Markov chain. The default argument is algorithm=hmc.\n\n\nIf a model doesn’t specify any parameters, then argument algorithm=fixed_param is mandatory.\nThe fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.\n\n\n\nAll HMC algorithms have three parameters:\n\nstep size\nmetric\nintegration time - the number of steps taken along the Hamiltonian trajectory\n\nSee the Stan Reference Manual section on HMC algorithm parameters for further details.\n\n\nThe HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.\nThe following keyword-value arguments control the step size:\n\nstepsize - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is \\(1\\).\nstepsize_jitter - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between \\(0\\) and \\(1\\). The default value is \\(0\\). Setting stepsize_jitter to \\(1\\) causes step sizes to be selected in the range of \\(0\\) to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.\n\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nThe metric argument specifies the choice of Euclidean HMC implementations:\n\nmetric=unit specifies unit metric (diagonal matrix of ones).\nmetric=diag_e specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.\nmetric=dense_e specifies a dense metric (a dense, symmetric positive definite matrix).\n\nBy default, the metric is estimated during warmup. However, when metric=diag_e or metric=dense_e, an initial guess for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a single variable inv_metric. For a diag_e metric the inv_metric value must be a vector of positive values, one for each parameter in the system. For a dense_e metric, inv_metric value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.\nThe metric_file option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n\n\n\nThe total integration time is determined by the argument engine which take possible values:\n\nnuts - the No-U-Turn Sampler which dynamically determines the optimal integration time.\nstatic - an HMC sampler which uses a user-specified integration time.\n\nThe default argument is engine=nuts.\nThe NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.\nWhen engine=nuts, the subargument max_depth can be used to control the depth of the tree. The default argument is max_depth=10. In the case where a model has a difficult posterior from which to sample, max_depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\nWhen the argument engine=static is specified, the user must specify the integration time via keyword int_time which takes as a value a positive number. The default value is \\(2\\pi\\).\n\n\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.\n\n\n\nA Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.\nThe preferred way of using multiple chains is to run them all from the same executable using the num_chains argument. There is also the option to use the Unix or DOS shell to run multiple executables.\n\n\nThe num_chains argument can be used for all of Stan’s samplers with the exception of the static HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with STAN_THREADS=true, these will either run in parallel or one after the other.\nWhen num_chains is greather than 1 (the default), arguments related to filenames (e.g. output file=, init=) can accept a comma separated list of values, one per each chain.\nFor example, sample will specify the names of the three chain’s output files.\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv\nThis will write the output in output_1.csv, output_2.csv, output_3.csv.\nIf the model was not compiled with STAN_THREADS=true, the above command will run 3 chains sequentially.\nIf the model was compiled with STAN_THREADS=true, the chains can run in parallel, with the num_threads argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (map_rect or reduce_sum calls), the below command will run 3 chains in parallel, provided there are cores available:\n./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4\nIf the model uses within-chain parallelization (map_rect or reduce_sum calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the Threading Building Blocks scheduler.\n./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16\n\n\nIf a comma separated list is not used, the num_chains argument changes the normal meanings of filename arguments when it is greater than 1 (the default). They are now interpreted as a “template” which is used for each chain.\nFor example, when num_chains=2, the argument output file=foo.csv no longer produces a file foo.csv, but instead produces two files, foo_1.csv and foo_2.csv. If you also supply id=5, the files produced will be foo_5.csv and foo_6.csv – id=5 gives the id of the first chain, and the remaining chains are sequential from there.\nThis also applies to input files, like those used for initialization. For example, if num_chains=3 and init=bar.json will first look for bar_1.json. If it exists, it will use bar_1.json for the first chain, bar_2.json for the second, and so on. If bar_1.json does not exist, it falls back to looking for bar.json, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the id argument, which defaults to 1.\nFor example, this shorthand is equivalent to the example given above:\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv\nA suffix with the chain id is appended to the provided output filename (output.csv in the above command), so this will also produce files output_1.csv, output_2.csv, output_3.csv.\n\n\n\n\n\nThe stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\nTo run stansummary on the output files generated by the for loop above, by the above run of the bernoulli model on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary output_*.csv\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe output_*.csv\nThe stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable theta:\nInference for Stan model: bernoulli_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total\nSampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total\n\n                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0\naccept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0\nstepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan\ntreedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0\nn_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0\ndivergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan\nenergy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0\n\ntheta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nIn this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for theta reflect the uncertainty in our estimate, due to the small amount of data, given the prior of beta(1, 1)\n\n\n\nNote: Many of these examples can be simplified by using the num_chains argument.\nWhen the num_chains argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.\nTo run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell for loop to set up index variables needed to identify each chain and its outputs.\nOn MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:\nfor NAME [in LIST]; do COMMANDS; done\nThe list can be a simple sequence of numbers, or you can use the shell expansion syntax {1..N} which expands to the sequence from \\(1\\) to \\(N\\), e.g. {1..4} expands to 1 2 3 4. Note that the expression {1..N} cannot contain spaces.\nTo run 4 chains for the example bernoulli model on MacOS or Linux:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv\n    done\nThe backslash (\\) indicates a line continuation in Unix. The expression ${i} substitutes in the value of loop index variable i. To run chains in parallel, put an ampersand (&) at the end of the nested sampler command:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv &\n    done\nThis pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.\nOn Windows, the DOS for-loop syntax is one of:\nfor %i in (SET) do COMMAND COMMAND-ARGUMENTS\nfor /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS\nTo run 4 chains in parallel on Windows:\n&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^\n                                    data file=bernoulli.data.json my_data ^\n                                    output file=output_%i.csv\nThe caret (^) indicates a line continuation in DOS. The expression %i is the loop index.\nIn the following extended examples, we focus on just the nested sampler command for Unix.\n\n\nFor reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.\nThe RNG seed is specified by random seed=&lt;int&gt; and the offset is specified by id=&lt;loop index&gt;, so the call to the sampler is:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            random seed=12345 id=${i}\n\n\n\nThe warmup and sampling iteration keyword-value arguments must follow the sample keyword. The call to the sampler which overrides the default warmup and sampling iterations is:\n./my_model sample num_warmup=500 num_sampling=500 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nTo save warmup draws as part of the Stan CSV output file, use the keyword-value argument save_warmup=true. This must be grouped with the other sample keyword sub-arguments.\n./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nBy default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range \\([{-2}, 2]\\). To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument init=&lt;filepath&gt;:\n./my_model sample init=my_param_inits.json data file=my_model.data.json \\\n           output file=output_${i}.csv\nTo verify that the specified values will be used by the sampler, you can run the sampler with option algorithm=fixed_param, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting num_warmp=0 and num_samples=1 saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.\nFor example, if we run the example Bernoulli model with specified initial value for parameter “theta”:\n{ \"theta\" : 0.5 }\nvia command:\n./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \\\n            init=bernoulli.init.json data file=bernoulli.data.json\nThe resulting output CSV file contains a single draw:\nlp__,accept_stat__,theta\n0,0,0.5\n#\n#  Elapsed Time: 0 seconds (Warm-up)\n#                0 seconds (Sampling)\n#                0 seconds (Total)\n#\n\n\n\nAn initial estimate for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a variable inv_metric. The metric_file option can be used with and without adaptation enabled.\nBy default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file bernoulli.diag_e.json, contents\n{ \"inv_metric\" : [0.296291] }\ncan be used as the initial metric as follows:\n../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nThe resulting output CSV file will contain the following set of comment lines:\n# Adaptation terminated\n# Step size = 0.9\n# Diagonal elements of inverse mass matrix:\n# 0.296291\nAs of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument save_metric, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the stepsize argument as well as the metric_file arguments; the former is the value of the stepsize element in the saved metric file, and the later is the metric file path.\n\n\n\nThe keyword-value arguments for these settings are grouped together under the adapt keyword which itself is a sub-argument of the sample keyword.\nModels with difficult posterior geometries may required increasing the delta argument closer to \\(1\\).\n./my_model sample adapt delta=0.95 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nTo skip adaptation altogether, use the keyword-value argument engaged=false. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nEven with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the typical set.\nTo skip warmup altogether requires specifying both num_warmup=0 and adapt engaged=false.\n../my_model sample num_warmup=0 adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nModels with difficult posterior geometries may required increasing the max_depth argument from its default value \\(10\\). This requires specifying a series of keyword-argument pairs:\n./my_model sample adapt delta=0.95 \\\n            algorithm=hmc engine=nuts max_depth=15 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; write the sampler parameters and gradients of all model parameters for each draw to a CSV file:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            diagnostic_file=diagnostics_${i}.csv\n\n\n\nThe output keyword sub-argument refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is \\(100\\) iterations. The progress updates look like:\nIteration:    1 / 2000 [  0%]  (Warmup)\nIteration:  100 / 2000 [  5%]  (Warmup)\nIteration:  200 / 2000 [ 10%]  (Warmup)\nIteration:  300 / 2000 [ 15%]  (Warmup)\nFor simple models which fit quickly, such updates can be annoying; to suppress them altogether, set refresh=0. This only turns off the Iteration: messages; the configuration and timing information are still written to the terminal.\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            refresh=0\nFor complicated models which take a long time to fit, setting the refresh rate to a low number, e.g. \\(10\\) or even \\(1\\), provides a way to more closely monitor the sampler.\n\n\n\nThe CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:\n./bernoulli sample data file=bernoulli.data.json adapt delta=0.95\nadapt is either mistyped or misplaced.\nPerhaps you meant one of the following valid configurations?\n  method=sample sample adapt\n  method=variational variational adapt\nFailed to parse arguments, terminating Stan\nThe following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.\n./my_model sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100 \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           random seed=12345 id=${i}\nThe keywords sample, data, output, and random are the top-level argument groups. Within the sample config arguments, the keyword adapt groups the adaptation algorithm parameters and the keyword-value algorithm=hmc groups the NUTS-HMC parameters.\nThe top-level groups can be freely ordered with respect to one another. The following is also a valid command:\n./my_model random seed=12345 id=${i} \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100",
+    "text": "Much like with complex scalars, two functions are defined to get the real and imaginary components of complex-valued objects.\n\n\nThese functions return the same shape (e.g., matrix, vector, row vector, or array) object as their input, but demoted to a real type. For example, get_real(complex_matrix M) yields a matrix containing the real component of each value in M.\nThe following table contains examples of what this notation can mean:\n\n\n\nType T\nType T_demoted\n\n\n\n\ncomplex\nreal\n\n\ncomplex_vector\nvector\n\n\ncomplex_row_vector\nrow_vector\n\n\ncomplex_matrix\nmatrix\n\n\narray[] complex\narray[] real\n\n\narray[,,] complex\narray[,,] real\n\n\n\n\n\n\n \n\nT_demoted get_real(T x) Given an object of complex type T, return the same shape object but of type real by getting the real component of each element of x.\nAvailable since 2.30\n \n\nT_demoted get_imag(T x) Given an object of complex type T, return the same shape object but of type real by getting the imaginary component of each element of x.\nAvailable since 2.30\nFor example, given the Stan declaration\n  complex_vector[2] z = [3+4i, 5+6i]';\nA call get_real(z) will yield the vector [3, 5]', and a call get_imag(z) will yield the vector [4, 6]'.",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#running-the-sampler",
-    "href": "cmdstan-guide/mcmc_config.html#running-the-sampler",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-broadcast-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-broadcast-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "To generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument sample or method=sample together with the input data. The executable can be run from any directory.\nThe full set of configuration options available for the sample method is available by using the sample help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.\nHere, we run it in the directory which contains the Stan program and input data, &lt;cmdstan-home&gt;/examples/bernoulli:\n&gt; cd examples/bernoulli\n&gt; ls\n  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan\nTo execute sampling of the model under Linux or Mac, use:\n&gt; ./bernoulli sample data file=bernoulli.data.json\nIn Windows, the ./ prefix is not needed:\n&gt; bernoulli.exe sample data file=bernoulli.data.json\nThe output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:\nmethod = sample (Default)\n  sample\n    num_samples = 1000 (Default)\n    num_warmup = 1000 (Default)\n    save_warmup = false (Default)\n    thin = 1 (Default)\n    adapt\n      engaged = true (Default)\n      gamma = 0.050000000000000003 (Default)\n      delta = 0.80000000000000004 (Default)\n      kappa = 0.75 (Default)\n      t0 = 10 (Default)\n      init_buffer = 75 (Default)\n      term_buffer = 50 (Default)\n      window = 25 (Default)\n      save_metric = false (Default)\n    algorithm = hmc (Default)\n      hmc\n        engine = nuts (Default)\n          nuts\n            max_depth = 10 (Default)\n        metric = diag_e (Default)\n        metric_file =  (Default)\n        stepsize = 1 (Default)\n        stepsize_jitter = 0 (Default)\n    num_chains = 1 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 3252652196 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\nAfter the configuration has been displayed, a short timing message is given.\nGradient evaluation took 1.2e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.\nAdjust your expectations accordingly!\nNext, the sampler reports the iteration number, reporting the percentage complete.\nIteration:    1 / 2000 [  0%]  (Warmup)\n...\nIteration: 2000 / 2000 [100%]  (Sampling)\nFinally, the sampler reports timing information:\n Elapsed Time: 0.007 seconds (Warm-up)\n               0.017 seconds (Sampling)\n               0.024 seconds (Total)",
+    "text": "The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\ncomplex_vector rep_vector(complex z, int m) Return the size m (column) vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_row_vector rep_row_vector(complex z, int n) Return the size n row vector consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex z, int m, int n) Return the m by n matrix consisting of copies of z.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.30\n \n\ncomplex_matrix rep_matrix(complex_row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.30\n\n\n \n\ncomplex_matrix symmetrize_from_lower_tri(complex_matrix A) Construct a symmetric matrix from the lower triangle of A.\nAvailable since 2.30",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#mcmc_output_csv",
-    "href": "cmdstan-guide/mcmc_config.html#mcmc_output_csv",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#diagonal-complex-matrix-functions",
+    "href": "functions-reference/complex_matrix_operations.html#diagonal-complex-matrix-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "Each execution of the model results in draws from a single Markov chain being written to a file in comma-separated value (CSV) format. The default name of the output file is output.csv.\nThe first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (#)).\nWhen the example model bernoulli.stan is run via the command line with all default arguments, the following configuration is displayed:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 1000 (Default)\n#     num_warmup = 1000 (Default)\n#     save_warmup = false (Default)\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000 (Default)\n#       delta = 0.800000 (Default)\n#       kappa = 0.750000 (Default)\n#       t0 = 10.000000 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#       save_metric = false (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1.000000 (Default)\n#         stepsize_jitter = 0.000000 (Default)\n#     num_chains = 1 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\nThis is followed by a CSV header indicating the names of the values sampled.\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nThe first output columns report the HMC sampler information:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause the above header is from the NUTS sampler, it has columns treedepth__, n_leapfrog__, and divergent__ and doesn’t have column int_time__. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, theta.\nThe header line is written to the output file before warmup begins. If option save_warmup is set to true, the warmup draws are output directly after the header. The total number of warmup draws saved is num_warmup divided by thin, rounded up (i.e., ceiling).\nFollowing the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:\n# Adaptation terminated\n# Step size = 0.884484\n# Diagonal elements of inverse mass matrix:\n# 0.535006\nThe default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter theta.\nDraws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.\n-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853\n-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295\n-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299\n-6.88712,1,0.884484,1,1,0,7.02101,0.188229\n-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596\n...\nThe output ends with timing details:\n#  Elapsed Time: 0.007 seconds (Warm-up)\n#                0.017 seconds (Sampling)\n#                0.024 seconds (Total)",
+    "text": "complex_matrix add_diag(complex_matrix m, complex_row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex_vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.30\n \n\ncomplex_matrix add_diag(complex_matrix m, complex d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.30\n \n\ncomplex_vector diagonal(complex_matrix x) The diagonal of the matrix x\nAvailable since 2.30\n \n\ncomplex_matrix diag_matrix(complex_vector x) The diagonal matrix with diagonal x\nAvailable since 2.30",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#iterations",
-    "href": "cmdstan-guide/mcmc_config.html#iterations",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#slicing-and-blocking-functions-for-complex-matrices",
+    "href": "functions-reference/complex_matrix_operations.html#slicing-and-blocking-functions-for-complex-matrices",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "At every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters metric and stepsize in order to efficiently sample from typical set, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.\nThe following keyword-value arguments control the total number of iterations:\n\nnum_samples\nnum_warmup\nsave_warmup\nthin\n\nThe values for arguments num_samples and num_warmup must be a non-negative integer. The default value for both is \\(1000\\).\nFor well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.\nThe number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is \\(\\sqrt{N}\\); therefore every additional decimal place of accuracy increases this by a factor of 10.\nArgument save_warmup takes values false or true. The default value is false, i.e., warmup draws are not saved to the output file. When the value is true, the warmup draws are written to the CSV output file directly after the CSV header line.\nArgument thin controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce anticorrelated draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.\nThe value of argument thin must be a positive integer. When thin is set to value \\(N\\), every \\(N^{th}\\) iteration is written to the output file. Should the value of thin exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of thin equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When num_sampling=M and thin=N, the number of iterations written to the output CSV file will be ceiling(M/N). If save_warmup=true, thinning is applied to the warmup iterations as well.",
+    "text": "Stan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\ncomplex_vector col(complex_matrix x, int n) The n-th column of matrix x\nAvailable since 2.30\n \n\ncomplex_row_vector row(complex_matrix x, int m) The m-th row of matrix x\nAvailable since 2.30\n\n\n\n\n\n \n\ncomplex_matrix block(complex_matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.30\n \n\ncomplex_vector sub_col(complex_matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.30\n \n\ncomplex_row_vector sub_row(complex_matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector head(complex_vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector head(complex_row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector tail(complex_vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.30\n \n\ncomplex_row_vector tail(complex_row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.30\n \n\ncomplex_vector segment(complex_vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30\n \n\ncomplex_row_vector segment(complex_row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.30",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#adaptation",
-    "href": "cmdstan-guide/mcmc_config.html#adaptation",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-matrix-concatenation",
+    "href": "functions-reference/complex_matrix_operations.html#complex-matrix-concatenation",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "The adapt keyword is used to specify non-default options for the sampler adaptation schedule and settings.\nAdaptation can be turned off by setting sub-argument engaged to value false. If engaged=false, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is engaged=1, this keyword-value pair can be omitted from the command.\nThere are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section Automatic Parameter Tuning.\nThe boolean sub-argument save_metric was added in Stan version 2.34. When save_metric=true, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix _metric.json, e.g., if using the default output filename output.csv, the saved metric file will be output_metric.json. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument metric_file.\n\n\nThe Stan User’s Guide section on model conditioning and curvature provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on HMC algorithm parameters explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.\nThe following keyword-value arguments control the settings used to optimize the step size:\n\ndelta - The target Metropolis acceptance rate. The default value is \\(0.8\\). Its value must be strictly between \\(0\\) and \\(1\\). Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of delta will also allow some models that would otherwise get stuck to overcome their blockages.  Models with difficult posterior geometries may required increasing the delta argument closer to \\(1\\); we recommend first trying to raise it to \\(0.9\\) or at most \\(0.95\\). Values about \\(0.95\\) are strong indication of bad geometry; the better solution is to change the model geometry through reparameterization which could yield both more efficient and faster sampling.\ngamma - Adaptation regularization scale. Must be a positive real number, default value is \\(0.05\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nkappa - Adaptation relaxation exponent. Must be a positive real number, default value is \\(0.75\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nt_0 - Adaptation iteration offset. Must be a positive real number, default value is \\(10\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\n\n\n\n\nWhen adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:\n\ninit_buffer - The number of iterations spent tuning the step size at the outset of adaptation.\nwindow - The initial number of iterations devoted to tune the metric, will be doubled successively.\nterm_buffer - The number of iterations used to re-tune the step size once the metric has been tuned.\n\nThe specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.\nThe following figure is taken from the Stan Reference Manual, where label “I” correspond to init_buffer, the initial “II” corresponds to window, and the final “III” corresponds to term_buffer:\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.",
+    "text": "complex_matrix append_col(complex_matrix x, complex_matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_matrix x, complex_vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_matrix append_col(complex_vector x, complex_vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex_row_vector y) Combine row vectors x and y (of any size) into another row vector by appending y to the end of x.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex x, complex_row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.30\n \n\ncomplex_row_vector append_col(complex_row_vector x, complex y) Append y to the end of x, returning another row vector.\nAvailable since 2.30\n\n\n\n \n\ncomplex_matrix append_row(complex_matrix x, complex_matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_matrix x, complex_row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_matrix append_row(complex_row_vector x, complex_row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex_vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex x, complex_vector y) Append x to the top of y, returning another vector.\nAvailable since 2.30\n \n\ncomplex_vector append_row(complex_vector x, complex y) Append y to the bottom of x, returning another vector.\nAvailable since 2.30",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#algorithm",
-    "href": "cmdstan-guide/mcmc_config.html#algorithm",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-special-matrix-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-special-matrix-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "The algorithm keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: hmc, which generates from an HMC-driven Markov chain; and fixed_param which generates a new sample without changing the state of the Markov chain. The default argument is algorithm=hmc.\n\n\nIf a model doesn’t specify any parameters, then argument algorithm=fixed_param is mandatory.\nThe fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.\n\n\n\nAll HMC algorithms have three parameters:\n\nstep size\nmetric\nintegration time - the number of steps taken along the Hamiltonian trajectory\n\nSee the Stan Reference Manual section on HMC algorithm parameters for further details.\n\n\nThe HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.\nThe following keyword-value arguments control the step size:\n\nstepsize - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is \\(1\\).\nstepsize_jitter - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between \\(0\\) and \\(1\\). The default value is \\(0\\). Setting stepsize_jitter to \\(1\\) causes step sizes to be selected in the range of \\(0\\) to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.\n\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nThe metric argument specifies the choice of Euclidean HMC implementations:\n\nmetric=unit specifies unit metric (diagonal matrix of ones).\nmetric=diag_e specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.\nmetric=dense_e specifies a dense metric (a dense, symmetric positive definite matrix).\n\nBy default, the metric is estimated during warmup. However, when metric=diag_e or metric=dense_e, an initial guess for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a single variable inv_metric. For a diag_e metric the inv_metric value must be a vector of positive values, one for each parameter in the system. For a dense_e metric, inv_metric value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.\nThe metric_file option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n\n\n\nThe total integration time is determined by the argument engine which take possible values:\n\nnuts - the No-U-Turn Sampler which dynamically determines the optimal integration time.\nstatic - an HMC sampler which uses a user-specified integration time.\n\nThe default argument is engine=nuts.\nThe NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.\nWhen engine=nuts, the subargument max_depth can be used to control the depth of the tree. The default argument is max_depth=10. In the case where a model has a difficult posterior from which to sample, max_depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\nWhen the argument engine=static is specified, the user must specify the integration time via keyword int_time which takes as a value a positive number. The default value is \\(2\\pi\\).",
+    "text": "Stan’s fast Fourier transform functions take the standard definition of the discrete Fourier transform (see the definitions below for specifics) and scale the inverse transform by one over dimensionality so that the following identities hold for complex vectors u and v,\n    fft(inv_fft(u)) == u        inv_fft(fft(v)) == v\nand in the 2-dimensional case for complex matrices A and B,\n    fft2(inv_fft2(A)) == A      inv_fft2(fft2(B)) == B\nAlthough the FFT functions only accept complex inputs, real vectors and matrices will be promoted to their complex counterparts before applying the FFT functions.\n \n\ncomplex_vector fft(complex_vector v) Return the discrete Fourier transform of the specified complex vector v. If \\(v \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(u =\n\\textrm{fft}(v)\\), then \\[\\begin{equation*}\nu_n = \\sum_{m &lt; n}\nv_m \\cdot\n\\exp\\left(\\frac{-n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\]\nAvailable since 2.30\n \n\ncomplex_matrix fft2(complex_matrix m) Return the 2D discrete Fourier transform of the specified complex matrix m. The 2D FFT is defined as the result of applying the FFT to each row and then to each column.\nAvailable since 2.30\n \n\ncomplex_vector inv_fft(complex_vector u) Return the inverse of the discrete Fourier transform of the specified complex vector u. The inverse FFT (this function) is scaled so that fft(inv_fft(u)) == u. If \\(u \\in \\mathbb{C}^N\\) is a complex vector with \\(N\\) elements and \\(v = \\textrm{fft}^{-1}(u)\\), then \\[\\begin{equation*}\nv_n =  \\frac{1}{N} \\sum_{m &lt; n}\nu_m \\cdot\n\\exp\\left(\\frac{n \\cdot m \\cdot 2 \\cdot \\pi \\cdot \\sqrt{-1}}{N}\\right).\n\\end{equation*}\\] This only differs from the FFT by the sign inside the exponential and the scaling. The \\(\\frac{1}{N}\\) scaling ensures that fft(inv_fft(u)) == u and inv_fft(fft(v)) == v for complex vectors u and v.\nAvailable since 2.30\n \n\ncomplex_matrix inv_fft2(complex_matrix m) Return the inverse of the 2D discrete Fourier transform of the specified complex matrix m. The 2D inverse FFT is defined as the result of applying the inverse FFT to each row and then to each column. The invertible scaling of the inverse FFT ensures fft2(inv_fft2(A)) == A and inv_fft2(fft2(B)) == B.\nAvailable since 2.30\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] complex cumulative_sum(array[] complex x) The cumulative sum of x\nAvailable since 2.30\n \n\ncomplex_vector cumulative_sum(complex_vector v) The cumulative sum of v\nAvailable since 2.30\n \n\ncomplex_row_vector cumulative_sum(complex_row_vector rv) The cumulative sum of rv\nAvailable since 2.30",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#sampler-diag-file",
-    "href": "cmdstan-guide/mcmc_config.html#sampler-diag-file",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#complex-linear-algebra-functions",
+    "href": "functions-reference/complex_matrix_operations.html#complex-linear-algebra-functions",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "The output keyword sub-argument diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.",
+    "text": "In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse.\n\n\n \n\ncomplex_row_vector operator/(complex_row_vector b, complex_matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.30\n \n\ncomplex_matrix operator/(complex_matrix B, complex_matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.30\n\n\n\n\n\n\n \n\ncomplex trace(complex_matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.30\n\n\n\n \n\ncomplex_vector eigenvalues(complex_matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.32\n \n\ncomplex_matrix eigenvectors(complex_matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.32\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\ncomplex_vector eigenvalues_sym(complex_matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors_sym(complex_matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose_sym(complex_matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(complex_matrix A) The singular values of A in descending order\nAvailable since 2.30\n \n\ncomplex_matrix svd_U(complex_matrix A) The left-singular vectors of A\nAvailable since 2.30\n \n\ncomplex_matrix svd_V(complex_matrix A) The right-singular vectors of A\nAvailable since 2.30\n \n\ntuple(complex_matrix, vector, complex_matrix) svd(complex_matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\nThe complex Schur decomposition of a square matrix \\(A\\) produces a complex unitary matrix \\(U\\) and a complex upper-triangular Schur form matrix \\(T\\) such that \\[A = U \\cdot T \\cdot U^{-1}\\]\nSince \\(U\\) is unitary, its inverse is also its conjugate transpose, \\(U^{-1} = U^*\\), \\(U^*(i, j) = \\mathrm{conj}(U(j, i))\\)\n \n\ncomplex_matrix complex_schur_decompose_t(matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_t(complex_matrix A) Compute the upper-triangular Schur form matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ncomplex_matrix complex_schur_decompose_u(complex_matrix A) Compute the unitary matrix of the complex Schur decomposition of A.\nAvailable since 2.31\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results. This overload is equivalent to complex_schur_decompose(to_complex(A,0)) but is more efficient.\nAvailable since 2.33\n \n\ntuple(complex_matrix, complex_matrix) complex_schur_decompose(complex_matrix A) Returns the unitary matrix and the upper-triangular Schur form matrix of the complex Schur decomposition of A. This function is equivalent to (complex_schur_decompose_u(A), complex_schur_decompose_t(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#multi-chain-sampling",
-    "href": "cmdstan-guide/mcmc_config.html#multi-chain-sampling",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/complex_matrix_operations.html#reverse-functions-for-complex-matrices",
+    "href": "functions-reference/complex_matrix_operations.html#reverse-functions-for-complex-matrices",
+    "title": "Complex Matrix Operations",
     "section": "",
-    "text": "A Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.\nThe preferred way of using multiple chains is to run them all from the same executable using the num_chains argument. There is also the option to use the Unix or DOS shell to run multiple executables.\n\n\nThe num_chains argument can be used for all of Stan’s samplers with the exception of the static HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with STAN_THREADS=true, these will either run in parallel or one after the other.\nWhen num_chains is greather than 1 (the default), arguments related to filenames (e.g. output file=, init=) can accept a comma separated list of values, one per each chain.\nFor example, sample will specify the names of the three chain’s output files.\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv\nThis will write the output in output_1.csv, output_2.csv, output_3.csv.\nIf the model was not compiled with STAN_THREADS=true, the above command will run 3 chains sequentially.\nIf the model was compiled with STAN_THREADS=true, the chains can run in parallel, with the num_threads argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (map_rect or reduce_sum calls), the below command will run 3 chains in parallel, provided there are cores available:\n./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4\nIf the model uses within-chain parallelization (map_rect or reduce_sum calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the Threading Building Blocks scheduler.\n./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16\n\n\nIf a comma separated list is not used, the num_chains argument changes the normal meanings of filename arguments when it is greater than 1 (the default). They are now interpreted as a “template” which is used for each chain.\nFor example, when num_chains=2, the argument output file=foo.csv no longer produces a file foo.csv, but instead produces two files, foo_1.csv and foo_2.csv. If you also supply id=5, the files produced will be foo_5.csv and foo_6.csv – id=5 gives the id of the first chain, and the remaining chains are sequential from there.\nThis also applies to input files, like those used for initialization. For example, if num_chains=3 and init=bar.json will first look for bar_1.json. If it exists, it will use bar_1.json for the first chain, bar_2.json for the second, and so on. If bar_1.json does not exist, it falls back to looking for bar.json, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the id argument, which defaults to 1.\nFor example, this shorthand is equivalent to the example given above:\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv\nA suffix with the chain id is appended to the provided output filename (output.csv in the above command), so this will also produce files output_1.csv, output_2.csv, output_3.csv.",
+    "text": "complex_vector reverse(complex_vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.30\n \n\ncomplex_row_vector reverse(complex_row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.30",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Built-in Functions",
+      "Complex Matrix Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#summarizing-sampler-outputs-with-stansummary",
-    "href": "cmdstan-guide/mcmc_config.html#summarizing-sampler-outputs-with-stansummary",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/circular_distributions.html",
+    "href": "functions-reference/circular_distributions.html",
+    "title": "Circular Distributions",
     "section": "",
-    "text": "The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\nTo run stansummary on the output files generated by the for loop above, by the above run of the bernoulli model on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary output_*.csv\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe output_*.csv\nThe stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable theta:\nInference for Stan model: bernoulli_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total\nSampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total\n\n                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0\naccept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0\nstepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan\ntreedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0\nn_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0\ndivergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan\nenergy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0\n\ntheta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nIn this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for theta reflect the uncertainty in our estimate, due to the small amount of data, given the prior of beta(1, 1)",
+    "text": "Circular distributions are defined for finite values y in any interval of length \\(2\\pi\\).\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{VonMises}(y|\\mu,\\kappa) =\n\\frac{\\exp(\\kappa\\cos(y-\\mu))}{2\\pi I_0(\\kappa)} \\!. \\end{equation*}\\] In order for this density to properly normalize, \\(y\\) must be restricted to some interval \\((c, c + 2\\pi)\\) of length \\(2 \\pi\\), because \\[\\begin{equation*} \\int_{c}^{c +\n2\\pi} \\text{VonMises}(y|\\mu,\\kappa) dy = 1. \\end{equation*}\\] Similarly, if \\(\\mu\\) is a parameter, it will typically be restricted to the same range as \\(y\\).\nIf \\(\\kappa &gt; 0\\), a von Mises distribution with its \\(2 \\pi\\) interval of support centered around its location \\(\\mu\\) will have a single mode at \\(\\mu\\); for example, restricting \\(y\\) to \\((-\\pi,\\pi)\\) and taking \\(\\mu = 0\\) leads to a single local optimum at the mode \\(\\mu\\). If the location \\(\\mu\\) is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean \\(\\mu\\).\nFor \\(\\kappa = 0\\), the Von Mises distribution corresponds to the circular uniform distribution with density \\(1 / (2 \\pi)\\) (independently of the values of \\(y\\) or \\(\\mu\\)).\n\n\n\ny ~ von_mises(mu, kappa)\nIncrement target log probability density with von_mises_lupdf(y | mu, kappa).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal von_mises_lpdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa.\nAvailable since 2.18\n \n\nreal von_mises_lupdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.\nAvailable since 2.25\n \n\nreal von_mises_cdf(reals y | reals mu, reals kappa) The von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lcdf(reals y | reals mu, reals kappa) The log of the von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lccdf(reals y | reals mu, reals kappa) The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nR von_mises_rng(reals mu, reals kappa) Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval \\([(\\mu \\mod 2\\pi)-\\pi,(\\mu \\mod\n2\\pi)+\\pi]\\)); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nEvaluating the Von Mises distribution for \\(\\kappa &gt; 100\\) is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as \\(\\kappa \\rightarrow \\infty\\), \\[\\begin{equation*}\n\\text{VonMises}(y|\\mu,\\kappa) \\rightarrow \\text{Normal}(\\mu, \\sqrt{1 /\n\\kappa}). \\end{equation*}\\] The workaround is to replace y ~ von_mises(mu,kappa) with\nif (kappa &lt; 100) {\n  y ~ von_mises(mu, kappa);\n} else {\n  y ~ normal(mu, sqrt(1 / kappa));\n}",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Circular Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/mcmc_config.html#old-multichain",
-    "href": "cmdstan-guide/mcmc_config.html#old-multichain",
-    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
+    "objectID": "functions-reference/circular_distributions.html#von-mises-distribution",
+    "href": "functions-reference/circular_distributions.html#von-mises-distribution",
+    "title": "Circular Distributions",
     "section": "",
-    "text": "Note: Many of these examples can be simplified by using the num_chains argument.\nWhen the num_chains argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.\nTo run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell for loop to set up index variables needed to identify each chain and its outputs.\nOn MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:\nfor NAME [in LIST]; do COMMANDS; done\nThe list can be a simple sequence of numbers, or you can use the shell expansion syntax {1..N} which expands to the sequence from \\(1\\) to \\(N\\), e.g. {1..4} expands to 1 2 3 4. Note that the expression {1..N} cannot contain spaces.\nTo run 4 chains for the example bernoulli model on MacOS or Linux:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv\n    done\nThe backslash (\\) indicates a line continuation in Unix. The expression ${i} substitutes in the value of loop index variable i. To run chains in parallel, put an ampersand (&) at the end of the nested sampler command:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv &\n    done\nThis pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.\nOn Windows, the DOS for-loop syntax is one of:\nfor %i in (SET) do COMMAND COMMAND-ARGUMENTS\nfor /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS\nTo run 4 chains in parallel on Windows:\n&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^\n                                    data file=bernoulli.data.json my_data ^\n                                    output file=output_%i.csv\nThe caret (^) indicates a line continuation in DOS. The expression %i is the loop index.\nIn the following extended examples, we focus on just the nested sampler command for Unix.\n\n\nFor reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.\nThe RNG seed is specified by random seed=&lt;int&gt; and the offset is specified by id=&lt;loop index&gt;, so the call to the sampler is:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            random seed=12345 id=${i}\n\n\n\nThe warmup and sampling iteration keyword-value arguments must follow the sample keyword. The call to the sampler which overrides the default warmup and sampling iterations is:\n./my_model sample num_warmup=500 num_sampling=500 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nTo save warmup draws as part of the Stan CSV output file, use the keyword-value argument save_warmup=true. This must be grouped with the other sample keyword sub-arguments.\n./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nBy default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range \\([{-2}, 2]\\). To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument init=&lt;filepath&gt;:\n./my_model sample init=my_param_inits.json data file=my_model.data.json \\\n           output file=output_${i}.csv\nTo verify that the specified values will be used by the sampler, you can run the sampler with option algorithm=fixed_param, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting num_warmp=0 and num_samples=1 saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.\nFor example, if we run the example Bernoulli model with specified initial value for parameter “theta”:\n{ \"theta\" : 0.5 }\nvia command:\n./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \\\n            init=bernoulli.init.json data file=bernoulli.data.json\nThe resulting output CSV file contains a single draw:\nlp__,accept_stat__,theta\n0,0,0.5\n#\n#  Elapsed Time: 0 seconds (Warm-up)\n#                0 seconds (Sampling)\n#                0 seconds (Total)\n#\n\n\n\nAn initial estimate for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a variable inv_metric. The metric_file option can be used with and without adaptation enabled.\nBy default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file bernoulli.diag_e.json, contents\n{ \"inv_metric\" : [0.296291] }\ncan be used as the initial metric as follows:\n../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nThe resulting output CSV file will contain the following set of comment lines:\n# Adaptation terminated\n# Step size = 0.9\n# Diagonal elements of inverse mass matrix:\n# 0.296291\nAs of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument save_metric, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the stepsize argument as well as the metric_file arguments; the former is the value of the stepsize element in the saved metric file, and the later is the metric file path.\n\n\n\nThe keyword-value arguments for these settings are grouped together under the adapt keyword which itself is a sub-argument of the sample keyword.\nModels with difficult posterior geometries may required increasing the delta argument closer to \\(1\\).\n./my_model sample adapt delta=0.95 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nTo skip adaptation altogether, use the keyword-value argument engaged=false. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nEven with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the typical set.\nTo skip warmup altogether requires specifying both num_warmup=0 and adapt engaged=false.\n../my_model sample num_warmup=0 adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nModels with difficult posterior geometries may required increasing the max_depth argument from its default value \\(10\\). This requires specifying a series of keyword-argument pairs:\n./my_model sample adapt delta=0.95 \\\n            algorithm=hmc engine=nuts max_depth=15 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; write the sampler parameters and gradients of all model parameters for each draw to a CSV file:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            diagnostic_file=diagnostics_${i}.csv\n\n\n\nThe output keyword sub-argument refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is \\(100\\) iterations. The progress updates look like:\nIteration:    1 / 2000 [  0%]  (Warmup)\nIteration:  100 / 2000 [  5%]  (Warmup)\nIteration:  200 / 2000 [ 10%]  (Warmup)\nIteration:  300 / 2000 [ 15%]  (Warmup)\nFor simple models which fit quickly, such updates can be annoying; to suppress them altogether, set refresh=0. This only turns off the Iteration: messages; the configuration and timing information are still written to the terminal.\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            refresh=0\nFor complicated models which take a long time to fit, setting the refresh rate to a low number, e.g. \\(10\\) or even \\(1\\), provides a way to more closely monitor the sampler.\n\n\n\nThe CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:\n./bernoulli sample data file=bernoulli.data.json adapt delta=0.95\nadapt is either mistyped or misplaced.\nPerhaps you meant one of the following valid configurations?\n  method=sample sample adapt\n  method=variational variational adapt\nFailed to parse arguments, terminating Stan\nThe following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.\n./my_model sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100 \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           random seed=12345 id=${i}\nThe keywords sample, data, output, and random are the top-level argument groups. Within the sample config arguments, the keyword adapt groups the adaptation algorithm parameters and the keyword-value algorithm=hmc groups the NUTS-HMC parameters.\nThe top-level groups can be freely ordered with respect to one another. The following is also a valid command:\n./my_model random seed=12345 id=${i} \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\kappa \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{VonMises}(y|\\mu,\\kappa) =\n\\frac{\\exp(\\kappa\\cos(y-\\mu))}{2\\pi I_0(\\kappa)} \\!. \\end{equation*}\\] In order for this density to properly normalize, \\(y\\) must be restricted to some interval \\((c, c + 2\\pi)\\) of length \\(2 \\pi\\), because \\[\\begin{equation*} \\int_{c}^{c +\n2\\pi} \\text{VonMises}(y|\\mu,\\kappa) dy = 1. \\end{equation*}\\] Similarly, if \\(\\mu\\) is a parameter, it will typically be restricted to the same range as \\(y\\).\nIf \\(\\kappa &gt; 0\\), a von Mises distribution with its \\(2 \\pi\\) interval of support centered around its location \\(\\mu\\) will have a single mode at \\(\\mu\\); for example, restricting \\(y\\) to \\((-\\pi,\\pi)\\) and taking \\(\\mu = 0\\) leads to a single local optimum at the mode \\(\\mu\\). If the location \\(\\mu\\) is not in the center of the support, the density is circularly translated and there will be a second local maximum at the boundary furthest from the mode. Ideally, the parameterization and support will be set up so that the bulk of the probability mass is in a continuous interval around the mean \\(\\mu\\).\nFor \\(\\kappa = 0\\), the Von Mises distribution corresponds to the circular uniform distribution with density \\(1 / (2 \\pi)\\) (independently of the values of \\(y\\) or \\(\\mu\\)).\n\n\n\ny ~ von_mises(mu, kappa)\nIncrement target log probability density with von_mises_lupdf(y | mu, kappa).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal von_mises_lpdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa.\nAvailable since 2.18\n \n\nreal von_mises_lupdf(reals y | reals mu, reals kappa) The log of the von mises density of y given location mu and scale kappa dropping constant additive terms.\nAvailable since 2.25\n \n\nreal von_mises_cdf(reals y | reals mu, reals kappa) The von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lcdf(reals y | reals mu, reals kappa) The log of the von mises cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nreal von_mises_lccdf(reals y | reals mu, reals kappa) The log of the von mises complementary cumulative distribution function of y given location mu and scale kappa.\nAvailable since 2.29\n \n\nR von_mises_rng(reals mu, reals kappa) Generate a Von Mises variate with location mu and scale kappa (i.e. returns values in the interval \\([(\\mu \\mod 2\\pi)-\\pi,(\\mu \\mod\n2\\pi)+\\pi]\\)); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nEvaluating the Von Mises distribution for \\(\\kappa &gt; 100\\) is numerically unstable in the current implementation. Nathanael I. Lichti suggested the following workaround on the Stan users group, based on the fact that as \\(\\kappa \\rightarrow \\infty\\), \\[\\begin{equation*}\n\\text{VonMises}(y|\\mu,\\kappa) \\rightarrow \\text{Normal}(\\mu, \\sqrt{1 /\n\\kappa}). \\end{equation*}\\] The workaround is to replace y ~ von_mises(mu,kappa) with\nif (kappa &lt; 100) {\n  y ~ von_mises(mu, kappa);\n} else {\n  y ~ normal(mu, sqrt(1 / kappa));\n}",
     "crumbs": [
-      "Running CmdStan",
-      "MCMC Sampling using Hamiltonian Monte Carlo"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Circular Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/laplace_sample_config.html",
-    "href": "cmdstan-guide/laplace_sample_config.html",
-    "title": "Laplace sampling",
+    "objectID": "functions-reference/bounded_continuous_distributions.html",
+    "href": "functions-reference/bounded_continuous_distributions.html",
+    "title": "Bounded Continuous Distributions",
     "section": "",
-    "text": "The laplace method produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood. In general, the posterior mode in the unconstrained space doesn’t correspond to the mean (nor mode) in the constrained space, and thus the sample is needed to infer the mean as well as the standard deviation. (See this case study for a visual illustration.)\nThis is computationally inexpensive compared to exact Bayesian inference with MCMC. The goodness of this estimate depends on both the estimate of the mode and how much the true posterior in the unconstrained space resembles a Gaussian.\n\n\nThis method takes several arguments:\n\nmode - Input file of parameters values on the constrained scale. When Stan’s optimize method is used to estimate the modal values, the value of boolean argument jacobian should be false if optimize was run with default settings, i.e., the input is the MLE estimate; if optimize was run with argument jacobian=true, then the laplace method default setting, jacobian=true, should be used.\njacobian - Whether or not the Jacobian adjustment should be included in the gradient. The default value is true (include adjustment). (Note: in optimization, the default value is false, for historical reasons.)\ndraws - How many total draws to return. The default is \\(1000\\).\ncalculate_lp - Whether to calculate the log probability of the model at each draw. If this is false, the log_p__ column of the output will be entirely nan. The default value is true.\n\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the laplace method is reported at the beginning of the output file as CSV comments.\nOutput columns log_p__ and log_q__, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.\nOutput columns for all model parameters on the constrained scale.\n\n\n\n\nIf requested with output diagnostic_file=, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.\n\n\n\nTo get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:\n\nfind the MAP estimate by running optimization with argument jacobian=true\nrun the Laplace estimator using the MAP estimate as the mode argument.\n\nBecause the default output file name from all methods is output.csv, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.\n./examples/bernoulli/bernoulli optimize jacobian=1 \\\n  data file=examples/bernoulli/bernoulli.data.json \\\n  output file=bernoulli_optimize_lbfgs.csv random seed=1234\n\n\n./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \\\n data file=examples/bernoulli/bernoulli.data.json random seed=1234\nThe header and first few data rows of the output sample are shown below.\n# method = laplace\n#   laplace\n#     mode = bernoulli_lbfgs.csv\n#     jacobian = true (Default)\n#     draws = 1000 (Default)\n#     calculate_lp = true (default)\n# id = 1 (Default)\n# data\n#   file = examples/bernoulli/bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 875960551 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlog_p__,log_q__,theta\n-9.4562,-2.33997,0.0498545\n-6.9144,-0.0117349,0.182898\n-7.18171,-0.746034,0.376428\n...",
+    "text": "The bounded continuous probabilities have support on a finite interval of real numbers.\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in (\\alpha,\\infty)\\), then for \\(y \\in [\\alpha,\\beta]\\), \\[\\begin{equation*} \\text{Uniform}(y|\\alpha,\\beta) =\n\\frac{1}{\\beta - \\alpha} . \\end{equation*}\\]\n\n\n\ny ~ uniform(alpha, beta)\nIncrement target log probability density with uniform_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal uniform_lpdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lupdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal uniform_cdf(reals y | reals alpha, reals beta) The uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.0\n \n\nreal uniform_lcdf(reals y | reals alpha, reals beta) The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lccdf(reals y | reals alpha, reals beta) The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nR uniform_rng(reals alpha, reals beta) Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
-      "Running CmdStan",
-      "Laplace sampling"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Bounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/laplace_sample_config.html#configuration",
-    "href": "cmdstan-guide/laplace_sample_config.html#configuration",
-    "title": "Laplace sampling",
+    "objectID": "functions-reference/bounded_continuous_distributions.html#uniform-distribution",
+    "href": "functions-reference/bounded_continuous_distributions.html#uniform-distribution",
+    "title": "Bounded Continuous Distributions",
     "section": "",
-    "text": "This method takes several arguments:\n\nmode - Input file of parameters values on the constrained scale. When Stan’s optimize method is used to estimate the modal values, the value of boolean argument jacobian should be false if optimize was run with default settings, i.e., the input is the MLE estimate; if optimize was run with argument jacobian=true, then the laplace method default setting, jacobian=true, should be used.\njacobian - Whether or not the Jacobian adjustment should be included in the gradient. The default value is true (include adjustment). (Note: in optimization, the default value is false, for historical reasons.)\ndraws - How many total draws to return. The default is \\(1000\\).\ncalculate_lp - Whether to calculate the log probability of the model at each draw. If this is false, the log_p__ column of the output will be entirely nan. The default value is true.",
+    "text": "If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in (\\alpha,\\infty)\\), then for \\(y \\in [\\alpha,\\beta]\\), \\[\\begin{equation*} \\text{Uniform}(y|\\alpha,\\beta) =\n\\frac{1}{\\beta - \\alpha} . \\end{equation*}\\]\n\n\n\ny ~ uniform(alpha, beta)\nIncrement target log probability density with uniform_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal uniform_lpdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lupdf(reals y | reals alpha, reals beta) The log of the uniform density of y given lower bound alpha and upper bound beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal uniform_cdf(reals y | reals alpha, reals beta) The uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.0\n \n\nreal uniform_lcdf(reals y | reals alpha, reals beta) The log of the uniform cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nreal uniform_lccdf(reals y | reals alpha, reals beta) The log of the uniform complementary cumulative distribution function of y given lower bound alpha and upper bound beta\nAvailable since 2.12\n \n\nR uniform_rng(reals alpha, reals beta) Generate a uniform variate with lower bound alpha and upper bound beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
-      "Running CmdStan",
-      "Laplace sampling"
+      "Functions Reference",
+      "Continuous Distributions",
+      "Bounded Continuous Distributions"
     ]
   },
   {
-    "objectID": "cmdstan-guide/laplace_sample_config.html#csv-output",
-    "href": "cmdstan-guide/laplace_sample_config.html#csv-output",
-    "title": "Laplace sampling",
+    "objectID": "functions-reference/array_operations.html",
+    "href": "functions-reference/array_operations.html",
+    "title": "Array Operations",
     "section": "",
-    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the laplace method is reported at the beginning of the output file as CSV comments.\nOutput columns log_p__ and log_q__, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.\nOutput columns for all model parameters on the constrained scale.",
+    "text": "The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).\n\n\n \n\nreal min(array[] real x) The minimum value in x, or \\(+\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint min(array[] int x) The minimum value in x, or error if x is size 0.\nAvailable since 2.0\n \n\nreal max(array[] real x) The maximum value in x, or \\(-\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint max(array[] int x) The maximum value in x, or error if x is size 0.\nAvailable since 2.0\n\n\n\n \n\nint sum(array[] int x) The sum of the elements in x, or 0 if the array is empty.\nAvailable since 2.1\n \n\nreal sum(array[] real x) The sum of the elements in x; see definition above.\nAvailable since 2.0\n \n\ncomplex sum(array[] complex x) The sum of the elements in x; see definition above.\nAvailable since 2.30\n \n\nreal prod(array[] real x) The product of the elements in x, or 1 if x is size 0.\nAvailable since 2.0\n \n\nreal prod(array[] int x) The product of the elements in x, \\[\\begin{equation*}\n\\text{product}(x) = \\begin{cases}\n\\prod_{n=1}^N x_n & \\text{if} N &gt; 0 \\\\[4pt] 1 & \\text{if} N = 0\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal log_sum_exp(array[] real x) The natural logarithm of the sum of the exponentials of the elements in x, or \\(-\\infty\\) if the array is empty.\nAvailable since 2.0\n\n\n\nThe sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.1 The sample deviation is defined as the square root of the sample deviation, but is not unbiased.\n \n\nreal mean(array[] real x) The sample mean of the elements in x. For an array \\(x\\) of size \\(N &gt;\n0\\), \\[\\begin{equation*}\n\\text{mean}(x) \\ = \\ \\bar{x} \\ = \\ \\frac{1}{N} \\sum_{n=1}^N\nx_n.\n\\end{equation*}\\] It is an error to the call the mean function with an array of size \\(0\\).\nAvailable since 2.0\n \n\nreal variance(array[] real x) The sample variance of the elements in x. For \\(N &gt; 0\\), \\[\\begin{equation*}\n\\text{variance}(x) \\ = \\ \\begin{cases} \\frac{1}{N-1} \\sum_{n=1}^N (x_n\n- \\bar{x})^2 & \\text{if } N &gt; 1 \\\\[4pt] 0 & \\text{if } N = 1\n\\end{cases}\n\\end{equation*}\\] It is an error to call the variance function with an array of size 0.\nAvailable since 2.0\n \n\nreal sd(array[] real x) The sample standard deviation of elements in x. \\[\\begin{equation*}\n\\text{sd}(x) =\n\\begin{cases} \\sqrt{\\, \\text{variance}(x)} & \\text{if } N &gt; 1 \\\\[4pt]\n0 & \\text{if } N = 0 \\end{cases}\n\\end{equation*}\\] It is an error to call the sd function with an array of size 0.\nAvailable since 2.0\n\n\n\n \n\nreal norm1(vector x) The L1 norm of x, defined by \\[\\begin{equation*}\n\\text{norm1}(x) \\ = \\ \\textstyle \\sum_{n=1}^N (|x_n|)\n\\end{equation*}\\] where N is the size of x.\nAvailable since 2.30\n \n\nreal norm1(row_vector x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm1(array[] real x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm2(vector x) The L2 norm of x, defined by \\[\\begin{equation*}\n\\text{norm2}(x) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n)^2}\n\\end{equation*}\\] where N is the size of x\nAvailable since 2.30\n \n\nreal norm2(row_vector x) The L2 norm of x\nAvailable since 2.30\n \n\nreal norm2(array[] real x) The L2 norm of x\nAvailable since 2.30\n\n\n\n \n\nreal distance(vector x, vector y) The Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\text{distance}(x,y) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n - y_n)^2}\n\\end{equation*}\\] where N is the size of x and y. It is an error to call distance with arguments of unequal size.\nAvailable since 2.2\n \n\nreal distance(vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal squared_distance(vector x, vector y) The squared Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\mathrm{squared\\_distance}(x,y) \\ = \\ \\text{distance}(x,y)^2 \\ = \\ \\textstyle \\sum_{n=1}^N (x_n - y_n)^2,\n\\end{equation*}\\] where N is the size of x and y. It is an error to call squared_distance with arguments of unequal size.\nAvailable since 2.7\n \n\nreal squared_distance(vector x, row_vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.26\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data array[] real x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data array[] real x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n\n\n\n\nThe size of an array or matrix can be obtained using the dims() function. The dims() function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,\n array[7, 8, 9] real x;\n array[7] matrix[8, 9] y;\nthen calling dims(x) or dims(y) returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.\nThe size() function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as\n array[M, N] real a;\nthe size of a is M.\nThe function num_elements, on the other hand, measures all of the elements, so that the array a above has \\(M \\times N\\) elements.\nThe specialized functions rows() and cols() should be used to extract the dimensions of vectors and matrices.\n \n\narray[] int dims(T x) Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.\nAvailable since 2.0\n \n\nint num_elements(array[] T x) Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if x is of type array[4, 3] real then num_elements(x) is 12, and if y is declared as array[5] matrix[3, 4] y, then size(y) evaluates to 60.\nAvailable since 2.5\n \n\nint size(array[] T x) Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if x is of type array[4, 3] real then size(x) is 4.\nAvailable since 2.0\n\n\n\nThe following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.\n \n\narray[] T rep_array(T x, int n) Return the n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray [,] T rep_array(T x, int m, int n) Return the m by n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray[,,] T rep_array(T x, int k, int m, int n) Return the k by m by n array with every entry assigned to x.\nAvailable since 2.0\nFor example, rep_array(1.0,5) produces a real array (type array[] real) of size 5 with all values set to 1.0. On the other hand, rep_array(1,5) produces an integer array (type array[] int) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment\n array[5] real y;\n array[5] int x;\n\n x = rep_array(1, 5);     // ok\n y = rep_array(1.0, 5);   // ok\n\n x = rep_array(1.0, 5);   // illegal\n y = rep_array(1, 5);     // illegal\n\n x = y;                  // illegal\n y = x;                  // illegal\nIf the value being repeated v is a vector (i.e., T is vector), then rep_array(v, 27) is a size 27 array consisting of 27 copies of the vector v.\n vector[5] v;\n array[3] vector[5] a;\n\n a = rep_array(v, 3);  // fill a with copies of v\n a[2, 4] = 9.0;        // v[4], a[1, 4], a[3, 4] unchanged\nIf the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the rep_array functions is called. For instance, consider the following legal code snippet.\n array[5, 6] real a;\n array[3, 4, 5, 6] real b;\n\n b = rep_array(a, 3, 4); //  make (3 x 4) copies of a\n b[1, 1, 1, 1] = 27.9;    //  a[1, 1] unchanged\nAfter the assignment to b, the value for b[j, k, m, n] is equal to a[m, n] where it is defined, for j in 1:3, k in 1:4, m in 1:5, and n in 1:6.\n\n\n\n \n\nT append_array(T x, T y) Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.\nAvailable since 2.18\nFor example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.\n array[2, 1, 7] matrix[4, 6] x1;\n array[3, 1, 7] matrix[4, 6] x2;\n array[5, 1, 7] matrix[4, 6] x3;\n\n x3 = append_array(x1, x2);\n\n\n\nSorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1, -10.3,\n20.987),\n\\end{equation*}\\] then the various sort routines produce \\[\\begin{eqnarray*}\n\\mathrm{sort\\_asc(v)} & = &  (-10.3,1,20.987) \\\\[4pt]\n\\mathrm{sort\\_desc(v)} & = &  (20.987,1,-10.3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_asc(v)} & = &  (2,1,3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_desc(v)} & = &  (3,1,2)\n\\end{eqnarray*}\\]\n \n\narray[] real sort_asc(array[] real v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] int sort_asc(array[] int v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] real sort_desc(array[] real v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_desc(array[] int v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(array[] real v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(array[] int v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n\n\n\nStan provides functions to create a new array by reversing the order of elements in an existing array. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1,\\, -10.3,\\, 20.987),\n\\end{equation*}\\] then \\[\\begin{equation*}\n\\mathrm{reverse(v)} = (20.987,\\, -10.3,\\, 1).\n\\end{equation*}\\]\n \n\narray[] T reverse(array[] T v) Return a new array containing the elements of the argument in reverse order.\nAvailable since 2.23",
     "crumbs": [
-      "Running CmdStan",
-      "Laplace sampling"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/laplace_sample_config.html#diagnostic-file-outputs",
-    "href": "cmdstan-guide/laplace_sample_config.html#diagnostic-file-outputs",
-    "title": "Laplace sampling",
-    "section": "",
-    "text": "If requested with output diagnostic_file=, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.",
+    "objectID": "functions-reference/array_operations.html#array-reductions",
+    "href": "functions-reference/array_operations.html#array-reductions",
+    "title": "Array Operations",
+    "section": "",
+    "text": "The following operations take arrays as input and produce single output values. The boundary values for size 0 arrays are the unit with respect to the combination operation (min, max, sum, or product).\n\n\n \n\nreal min(array[] real x) The minimum value in x, or \\(+\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint min(array[] int x) The minimum value in x, or error if x is size 0.\nAvailable since 2.0\n \n\nreal max(array[] real x) The maximum value in x, or \\(-\\infty\\) if x is size 0.\nAvailable since 2.0\n \n\nint max(array[] int x) The maximum value in x, or error if x is size 0.\nAvailable since 2.0\n\n\n\n \n\nint sum(array[] int x) The sum of the elements in x, or 0 if the array is empty.\nAvailable since 2.1\n \n\nreal sum(array[] real x) The sum of the elements in x; see definition above.\nAvailable since 2.0\n \n\ncomplex sum(array[] complex x) The sum of the elements in x; see definition above.\nAvailable since 2.30\n \n\nreal prod(array[] real x) The product of the elements in x, or 1 if x is size 0.\nAvailable since 2.0\n \n\nreal prod(array[] int x) The product of the elements in x, \\[\\begin{equation*}\n\\text{product}(x) = \\begin{cases}\n\\prod_{n=1}^N x_n & \\text{if} N &gt; 0 \\\\[4pt] 1 & \\text{if} N = 0\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal log_sum_exp(array[] real x) The natural logarithm of the sum of the exponentials of the elements in x, or \\(-\\infty\\) if the array is empty.\nAvailable since 2.0\n\n\n\nThe sample mean, variance, and standard deviation are calculated in the usual way. For i.i.d. draws from a distribution of finite mean, the sample mean is an unbiased estimate of the mean of the distribution. Similarly, for i.i.d. draws from a distribution of finite variance, the sample variance is an unbiased estimate of the variance.1 The sample deviation is defined as the square root of the sample deviation, but is not unbiased.\n \n\nreal mean(array[] real x) The sample mean of the elements in x. For an array \\(x\\) of size \\(N &gt;\n0\\), \\[\\begin{equation*}\n\\text{mean}(x) \\ = \\ \\bar{x} \\ = \\ \\frac{1}{N} \\sum_{n=1}^N\nx_n.\n\\end{equation*}\\] It is an error to the call the mean function with an array of size \\(0\\).\nAvailable since 2.0\n \n\nreal variance(array[] real x) The sample variance of the elements in x. For \\(N &gt; 0\\), \\[\\begin{equation*}\n\\text{variance}(x) \\ = \\ \\begin{cases} \\frac{1}{N-1} \\sum_{n=1}^N (x_n\n- \\bar{x})^2 & \\text{if } N &gt; 1 \\\\[4pt] 0 & \\text{if } N = 1\n\\end{cases}\n\\end{equation*}\\] It is an error to call the variance function with an array of size 0.\nAvailable since 2.0\n \n\nreal sd(array[] real x) The sample standard deviation of elements in x. \\[\\begin{equation*}\n\\text{sd}(x) =\n\\begin{cases} \\sqrt{\\, \\text{variance}(x)} & \\text{if } N &gt; 1 \\\\[4pt]\n0 & \\text{if } N = 0 \\end{cases}\n\\end{equation*}\\] It is an error to call the sd function with an array of size 0.\nAvailable since 2.0\n\n\n\n \n\nreal norm1(vector x) The L1 norm of x, defined by \\[\\begin{equation*}\n\\text{norm1}(x) \\ = \\ \\textstyle \\sum_{n=1}^N (|x_n|)\n\\end{equation*}\\] where N is the size of x.\nAvailable since 2.30\n \n\nreal norm1(row_vector x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm1(array[] real x) The L1 norm of x\nAvailable since 2.30\n \n\nreal norm2(vector x) The L2 norm of x, defined by \\[\\begin{equation*}\n\\text{norm2}(x) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n)^2}\n\\end{equation*}\\] where N is the size of x\nAvailable since 2.30\n \n\nreal norm2(row_vector x) The L2 norm of x\nAvailable since 2.30\n \n\nreal norm2(array[] real x) The L2 norm of x\nAvailable since 2.30\n\n\n\n \n\nreal distance(vector x, vector y) The Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\text{distance}(x,y) \\ = \\ \\sqrt{\\textstyle \\sum_{n=1}^N (x_n - y_n)^2}\n\\end{equation*}\\] where N is the size of x and y. It is an error to call distance with arguments of unequal size.\nAvailable since 2.2\n \n\nreal distance(vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.2\n \n\nreal squared_distance(vector x, vector y) The squared Euclidean distance between x and y, defined by \\[\\begin{equation*}\n\\mathrm{squared\\_distance}(x,y) \\ = \\ \\text{distance}(x,y)^2 \\ = \\ \\textstyle \\sum_{n=1}^N (x_n - y_n)^2,\n\\end{equation*}\\] where N is the size of x and y. It is an error to call squared_distance with arguments of unequal size.\nAvailable since 2.7\n \n\nreal squared_distance(vector x, row_vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, vector y) The squared Euclidean distance between x and y\nAvailable since 2.26\n \n\nreal squared_distance(row_vector x, row_vector y) The Euclidean distance between x and y\nAvailable since 2.26\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data array[] real x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data array[] real x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27",
     "crumbs": [
-      "Running CmdStan",
-      "Laplace sampling"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/laplace_sample_config.html#example",
-    "href": "cmdstan-guide/laplace_sample_config.html#example",
-    "title": "Laplace sampling",
+    "objectID": "functions-reference/array_operations.html#array-size-and-dimension-function",
+    "href": "functions-reference/array_operations.html#array-size-and-dimension-function",
+    "title": "Array Operations",
     "section": "",
-    "text": "To get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:\n\nfind the MAP estimate by running optimization with argument jacobian=true\nrun the Laplace estimator using the MAP estimate as the mode argument.\n\nBecause the default output file name from all methods is output.csv, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.\n./examples/bernoulli/bernoulli optimize jacobian=1 \\\n  data file=examples/bernoulli/bernoulli.data.json \\\n  output file=bernoulli_optimize_lbfgs.csv random seed=1234\n\n\n./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \\\n data file=examples/bernoulli/bernoulli.data.json random seed=1234\nThe header and first few data rows of the output sample are shown below.\n# method = laplace\n#   laplace\n#     mode = bernoulli_lbfgs.csv\n#     jacobian = true (Default)\n#     draws = 1000 (Default)\n#     calculate_lp = true (default)\n# id = 1 (Default)\n# data\n#   file = examples/bernoulli/bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 875960551 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlog_p__,log_q__,theta\n-9.4562,-2.33997,0.0498545\n-6.9144,-0.0117349,0.182898\n-7.18171,-0.746034,0.376428\n...",
+    "text": "The size of an array or matrix can be obtained using the dims() function. The dims() function is defined to take an argument consisting of any variable with up to 8 array dimensions (and up to 2 additional matrix dimensions) and returns an array of integers with the dimensions. For example, if two variables are declared as follows,\n array[7, 8, 9] real x;\n array[7] matrix[8, 9] y;\nthen calling dims(x) or dims(y) returns an integer array of size 3 containing the elements 7, 8, and 9 in that order.\nThe size() function extracts the number of elements in an array. This is just the top-level elements, so if the array is declared as\n array[M, N] real a;\nthe size of a is M.\nThe function num_elements, on the other hand, measures all of the elements, so that the array a above has \\(M \\times N\\) elements.\nThe specialized functions rows() and cols() should be used to extract the dimensions of vectors and matrices.\n \n\narray[] int dims(T x) Return an integer array containing the dimensions of x; the type of the argument T can be any Stan type with up to 8 array dimensions.\nAvailable since 2.0\n \n\nint num_elements(array[] T x) Return the total number of elements in the array x including all elements in contained arrays, vectors, and matrices. T can be any array type. For example, if x is of type array[4, 3] real then num_elements(x) is 12, and if y is declared as array[5] matrix[3, 4] y, then size(y) evaluates to 60.\nAvailable since 2.5\n \n\nint size(array[] T x) Return the number of elements in the array x; the type of the array T can be any type, but the size is just the size of the top level array, not the total number of elements contained. For example, if x is of type array[4, 3] real then size(x) is 4.\nAvailable since 2.0",
     "crumbs": [
-      "Running CmdStan",
-      "Laplace sampling"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html",
-    "href": "cmdstan-guide/installation.html",
-    "title": "CmdStan Installation",
+    "objectID": "functions-reference/array_operations.html#array-broadcasting",
+    "href": "functions-reference/array_operations.html#array-broadcasting",
+    "title": "Array Operations",
     "section": "",
-    "text": "There are a few ways that you can install CmdStan. Depending on your operating system and your level of expertise, you can either:\n\nUse the conda package management system to install a pre-built version of CmdStan along with the required dependencies. Recommended for Windows users.\nInstall the source code from GitHub CmdStan repository. This requires a modern C++ compiler and toolchain. See the C++ Toolchain section for further details.\n\n\n\nWith conda, you can install CmdStan from the conda-forge channel. This will install a pre-built version of CmdStan along with the required dependencies (i.e. a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command cmdstan_model to activate the CmdStan makefile from anywhere.\nNote: This requires that conda has been installed already on your machine. We recommend using the miniforge distribution.\nWe recommend installing CmdStan in a new conda environment:\n conda create -n stan -c conda-forge cmdstan\nThis command creates a new conda environment named stan and downloads and installs the cmdstan package as well as CmdStan and the required C++ toolchain.\nTo install into an existing conda environment, use the conda install command instead of create:\n conda install -c conda-forge cmdstan\nWhichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment stan, run the command\n conda activate stan\nBy default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and newer can be installed by specifying cmdstan==VERSION in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment\nconda install -c conda-forge cmdstan=2.27.0\n\n\nA Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command\n conda info -e\nThe shell environment variable CONDA_PREFIX points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the bin subdirectory of the conda environment directory, i.e., $CONDA_PREFIX/bin/cmdstan (Linux, MacOS), %CONDA_PREFIX%\\bin\\cmdstan (Windows).\nPlease report conda-specific install problems directly to the conda-forge issue tracker, here.\n\n\n\n\nInstallation from GitHub consists of the following steps:\n\nVerify that you have a modern C++ toolchain. See the C++ Toolchain section for details.\nDownload the CmdStan source code from GitHub\nBuild the CmdStan libraries and executables\nCheck the installation by compiling and running the CmdStan example model bernoulli.stan.\n\n\n\nThe GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the cmdstan module; the Stan inference engine algorithms and Stan math library functions are specified as submodules and stored in the GitHub repositories stan and math, respectively.\nA CmdStan release is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do not contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-”, e.g. “cmdstan-2.29.2”.\nBy cloning the CmdStan repository with argument --recursive, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named cmdstan:\n&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive\nThroughout this manual, we refer to this top-level CmdStan source directory as &lt;cmdstan-home&gt;. This directory contains the following subdirectories:\n\ndirectory cmdstan/stan contains the sub-module stan (https://github.com/stan-dev/stan)\ndirectory cmdstan/stan/lib/stan_math contains the sub-module math (https://github.com/stan-dev/math)\n\n\n\n\nBuilding CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:\n\nstanc: the Stan compiler (translates Stan language to C++).\nstansummary: a basic posterior analysis tool. The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\ndiagnose: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.\n\nCmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): bin/linux-stanc, bin/mac-stanc and bin/windows-stanc. The CmdStan makefile build task copies the appropriate binary to bin/stanc. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.\nSteps to build CmdStan:\n\nOpen a command-line terminal window and change directories to the CmdStan home directory.\nRun the makefile target build which instantiates the CmdStan utilities and compiles all necessary C++ libraries.\n\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make build\nIf your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the -j option. For example, to build on 4 cores, type:\n&gt; make -j4 build\nWhen make build is successful, the directory &lt;cmdstan-home&gt;/bin/ will contain the executables stanc, stansummary, and diagnose (on Windows, corresponding .exe files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:\n--- CmdStan v2.29.2 built ---\nWarning: The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.\nWindows only: CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory &lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb is part of the PATH environment variable. See these instructions for details on changing the PATH. To permanently make this setting for the current user, you may execute:\n&gt; make install-tbb\nAfter changing the PATH environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)\n\n\n\n\nTo check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.\nOn Linux and macOS:\n# compile the example\n&gt; make examples/bernoulli/bernoulli\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli sample\\\n  data file=examples/bernoulli/bernoulli.data.json\n\n# default output written to file `output.csv`,\n# default num_samples is 1000, output file should have approx. 1050 lines\n&gt; wc -l output.csv\n\n# run the `bin/stansummary utility to summarize parameter estimates\n&gt; bin/stansummary output.csv\nOn Windows:\n# compile the example\n&gt; make examples/bernoulli/bernoulli.exe\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json\n\n# run the `bin/stansummary.exe utility to summarize parameter estimates\n&gt; bin/stansummary.exe output.csv\nThe sample data in file bernoulli.json.data specifies 2 out of 10 successes, therefore the range mean(theta)\\(\\pm\\)sd(theta) should include 0.2.\n\n\n\nUpdates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the &lt;cmdstan-home&gt; directory:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make clean-all\n&gt; make build\n\n\nThis section contains solutions to problems reported on https://discourse.mc-stan.org\nCompiler error message about PCH file\nTo speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:\nerror: PCH file uses an older PCH format that is no longer supported\nIn this case, clean and rebuild CmdStan, as shown in the previous section.\nWindows: ‘g++’, ‘make’, or ‘cut’ is not recognized\nThe CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:\n'cut' is not recognized as an internal or external command,\noperable program or batch file.\nTo fix this, ensure you have followed the steps for adding the toolchain to your PATH and installing the additional utilities covered in the configuration instructions\nSpaces in paths to CmdStan or model\nmake can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the One Drive folder.\nUnfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:\nmake: *** INTERNAL: readdir: Invalid argument\nmake: *** [make/program:50: x.hpp] Error 2\nIf the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.\n\n\n\n\nCompiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.\n\n\nThe required C++ compiler is g++ 4.9 3. On most systems the GNU Make utility is pre-installed and is the default make utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:\ng++ --version\nmake --version\nIf these are at least at g++ version 4.9.3 or later and make version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler g++, because later versions are faster.\nTo install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:\nsudo apt install g++\nsudo apt install make\nIf you can’t run sudo, you will need to ask your sysadmin or cluster administrator to install these tools for you.\n\n\n\nTo check if you already already have an appropriate toolchain installed, open the Terminal application and enter:\nclang++ --version\nmake --version\nIf either of these commands prints the message command not found, you will need to install Xcode’s command line tools.\nOpen the Terminal application and enter:\nxcode-select --install\nSelect “Install” in the window that opens.\nAfter the installation completes, you can double check that installation was successful by reopening the Terminal and running:\nclang++ --version\nmake --version\nYou can read more about Xcode on its site: https://developer.apple.com/xcode/\nWe don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.\n\n\n\nThe Windows toolchain consists of programs g++, the C++ compiler, and make, the GNU Make utility. To check if these are present, open a command shell [^1] and type:\ng++ --version\nmake --version\nCmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your PATH variable, See these instructions for details on changing the PATH if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.\n\n\nAll required utilities (e.g., make, g++) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the R Project documentation.\nThe R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.\nNote that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a Microsoft update.\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45\\usr\\bin\nC:\\rtools45\\x86_64-w64-mingw32.static.posix\\bin\n\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45 - ARM64\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45-aarch64\\usr\\bin\nC:\\rtools45-aarch64\\aarch64-w64-mingw32.static.posix\\bin\n\n\n\n\n\n\nCmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.\nGNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed only when there are changes to those other files. Thus Make provides an efficient way to manage complex software.\nThe CmdStan Makefile is in the &lt;cmdstan-home&gt; directory and is named makefile. This is one of the default GNU Makefile names, which allows you to omit the -f makefile argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, Make only works properly when invoked from the &lt;cmdstan-home&gt; directory; attempts to use this Makefile from another directory by specifying the full path to the file makefile won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:\nmake -f ~/github/stan-dev/cmdstan/makefile\n/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory\nmake: *** No rule to make target `make/command'.  Stop.\nThe conda-forge cmdstan package provides a solution to this problem via cmdstan_model command which lets you run the CmdStan makefile from anywhere to compile a Stan model.\nMakefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix .stan; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program my_program.stan in directory ../my_dir/, the make target is ../my_dir/my_program or ../my_dir/my_program.exe (on Windows).\nTo call Make, you invoke the utility name, make, followed by, in order:\n\nzero or more Make program options, then specify any Make variables as a series of\nzero of more Make variables, described below\nzero or more target names; the set of names is determined by the Makefile rules.\n\nmake &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;\nMakefile Variables\nMake targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile ../my_dir/my_program.stan for an OpenCL (GPU) machine, set the makefile variable STAN_OPENCL to TRUE:\n&gt; make STAN_OPENCL=TRUE ../my_dir/my_program\nMakefile variables can also be set by creating a file named local in the CmdStan make subdirectory which contains a list of &lt;VARIABLE&gt;=&lt;VALUE&gt; pairs, one per line. For example, to get the same effect as the above command every time, you would put the line STAN_OPENCL=TRUE into the file &lt;cmdstan_home&gt;/make/local.\nThe complete set of Makefile variables can be found in file &lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags.\nMake Targets\nWhen invoked without any arguments at all, Make prints a help message:\n&gt; make\n--------------------------------------------------------------------------------\nCmdStan v2.33.1 help\n\n  Build CmdStan utilities:\n    &gt; make build\n\n    This target will:\n    1. Install the Stan compiler bin/stanc from stanc3 binaries.\n    2. Build the print utility bin/print (deprecated; will be removed in v3.0)\n    3. Build the stansummary utility bin/stansummary\n    4. Build the diagnose utility bin/diagnose\n    5. Build all libraries and object files compile and link an executable Stan program\n\n    Note: to build using multiple cores, use the -j option to make, e.g.,\n    for 4 cores:\n    &gt; make build -j4\n\n\n  Build a Stan program:\n\n    Given a Stan program at foo/bar.stan, build an executable by typing:\n    &gt; make foo/bar\n\n    This target will:\n    1. Install the Stan compiler (bin/stanc), as needed.\n    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.\n    3. Compile the C++ code using cc . to generate foo/bar\n\n  Additional make options:\n    STANCFLAGS: defaults to \"\". These are extra options passed to bin/stanc\n      when generating C++ code. If you want to allow undefined functions in the\n      Stan program, either add this to make/local or the command line:\n          STANCFLAGS = --allow_undefined\n    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the\n      header file that is included. This defaults to \"user_header.hpp\" in the\n      directory of the Stan program.\n    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads\n      the nightly version.\n    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.\n    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.\n\n\n  Example - bernoulli model: examples/bernoulli/bernoulli.stan\n\n    1. Build the model:\n       &gt; make examples/bernoulli/bernoulli\n    2. Run the sampling algorithm given the model and data:\n       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R\n    3. Look at the posterior sample:\n       &gt; bin/stansummary output.csv\n\n\n  Clean CmdStan:\n\n    Remove the built CmdStan tools:\n    &gt; make clean-all\n\n--------------------------------------------------------------------------------",
+    "text": "The following operations create arrays by repeating elements to fill an array of a specified size. These operations work for all input types T, including reals, integers, vectors, row vectors, matrices, or arrays.\n \n\narray[] T rep_array(T x, int n) Return the n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray [,] T rep_array(T x, int m, int n) Return the m by n array with every entry assigned to x.\nAvailable since 2.0\n \n\narray[,,] T rep_array(T x, int k, int m, int n) Return the k by m by n array with every entry assigned to x.\nAvailable since 2.0\nFor example, rep_array(1.0,5) produces a real array (type array[] real) of size 5 with all values set to 1.0. On the other hand, rep_array(1,5) produces an integer array (type array[] int) of size 5 with all values set to 1. This distinction is important because it is not possible to assign an integer array to a real array. For example, the following example contrasts legal with illegal array creation and assignment\n array[5] real y;\n array[5] int x;\n\n x = rep_array(1, 5);     // ok\n y = rep_array(1.0, 5);   // ok\n\n x = rep_array(1.0, 5);   // illegal\n y = rep_array(1, 5);     // illegal\n\n x = y;                  // illegal\n y = x;                  // illegal\nIf the value being repeated v is a vector (i.e., T is vector), then rep_array(v, 27) is a size 27 array consisting of 27 copies of the vector v.\n vector[5] v;\n array[3] vector[5] a;\n\n a = rep_array(v, 3);  // fill a with copies of v\n a[2, 4] = 9.0;        // v[4], a[1, 4], a[3, 4] unchanged\nIf the type T of x is itself an array type, then the result will be an array with one, two, or three added dimensions, depending on which of the rep_array functions is called. For instance, consider the following legal code snippet.\n array[5, 6] real a;\n array[3, 4, 5, 6] real b;\n\n b = rep_array(a, 3, 4); //  make (3 x 4) copies of a\n b[1, 1, 1, 1] = 27.9;    //  a[1, 1] unchanged\nAfter the assignment to b, the value for b[j, k, m, n] is equal to a[m, n] where it is defined, for j in 1:3, k in 1:4, m in 1:5, and n in 1:6.",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html#conda-install",
-    "href": "cmdstan-guide/installation.html#conda-install",
-    "title": "CmdStan Installation",
+    "objectID": "functions-reference/array_operations.html#array-concatenation",
+    "href": "functions-reference/array_operations.html#array-concatenation",
+    "title": "Array Operations",
     "section": "",
-    "text": "With conda, you can install CmdStan from the conda-forge channel. This will install a pre-built version of CmdStan along with the required dependencies (i.e. a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command cmdstan_model to activate the CmdStan makefile from anywhere.\nNote: This requires that conda has been installed already on your machine. We recommend using the miniforge distribution.\nWe recommend installing CmdStan in a new conda environment:\n conda create -n stan -c conda-forge cmdstan\nThis command creates a new conda environment named stan and downloads and installs the cmdstan package as well as CmdStan and the required C++ toolchain.\nTo install into an existing conda environment, use the conda install command instead of create:\n conda install -c conda-forge cmdstan\nWhichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment stan, run the command\n conda activate stan\nBy default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and newer can be installed by specifying cmdstan==VERSION in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment\nconda install -c conda-forge cmdstan=2.27.0\n\n\nA Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command\n conda info -e\nThe shell environment variable CONDA_PREFIX points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the bin subdirectory of the conda environment directory, i.e., $CONDA_PREFIX/bin/cmdstan (Linux, MacOS), %CONDA_PREFIX%\\bin\\cmdstan (Windows).\nPlease report conda-specific install problems directly to the conda-forge issue tracker, here.",
+    "text": "T append_array(T x, T y) Return the concatenation of two arrays in the order of the arguments. T must be an N-dimensional array of any Stan type (with a maximum N of 7). All dimensions but the first must match.\nAvailable since 2.18\nFor example, the following code appends two three dimensional arrays of matrices together. Note that all dimensions except the first match. Any mismatches will cause an error to be thrown.\n array[2, 1, 7] matrix[4, 6] x1;\n array[3, 1, 7] matrix[4, 6] x2;\n array[5, 1, 7] matrix[4, 6] x3;\n\n x3 = append_array(x1, x2);",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html#installation-from-github",
-    "href": "cmdstan-guide/installation.html#installation-from-github",
-    "title": "CmdStan Installation",
+    "objectID": "functions-reference/array_operations.html#sorting-functions",
+    "href": "functions-reference/array_operations.html#sorting-functions",
+    "title": "Array Operations",
     "section": "",
-    "text": "Installation from GitHub consists of the following steps:\n\nVerify that you have a modern C++ toolchain. See the C++ Toolchain section for details.\nDownload the CmdStan source code from GitHub\nBuild the CmdStan libraries and executables\nCheck the installation by compiling and running the CmdStan example model bernoulli.stan.\n\n\n\nThe GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the cmdstan module; the Stan inference engine algorithms and Stan math library functions are specified as submodules and stored in the GitHub repositories stan and math, respectively.\nA CmdStan release is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do not contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-”, e.g. “cmdstan-2.29.2”.\nBy cloning the CmdStan repository with argument --recursive, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named cmdstan:\n&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive\nThroughout this manual, we refer to this top-level CmdStan source directory as &lt;cmdstan-home&gt;. This directory contains the following subdirectories:\n\ndirectory cmdstan/stan contains the sub-module stan (https://github.com/stan-dev/stan)\ndirectory cmdstan/stan/lib/stan_math contains the sub-module math (https://github.com/stan-dev/math)\n\n\n\n\nBuilding CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:\n\nstanc: the Stan compiler (translates Stan language to C++).\nstansummary: a basic posterior analysis tool. The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\ndiagnose: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.\n\nCmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): bin/linux-stanc, bin/mac-stanc and bin/windows-stanc. The CmdStan makefile build task copies the appropriate binary to bin/stanc. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.\nSteps to build CmdStan:\n\nOpen a command-line terminal window and change directories to the CmdStan home directory.\nRun the makefile target build which instantiates the CmdStan utilities and compiles all necessary C++ libraries.\n\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make build\nIf your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the -j option. For example, to build on 4 cores, type:\n&gt; make -j4 build\nWhen make build is successful, the directory &lt;cmdstan-home&gt;/bin/ will contain the executables stanc, stansummary, and diagnose (on Windows, corresponding .exe files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:\n--- CmdStan v2.29.2 built ---\nWarning: The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.\nWindows only: CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory &lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb is part of the PATH environment variable. See these instructions for details on changing the PATH. To permanently make this setting for the current user, you may execute:\n&gt; make install-tbb\nAfter changing the PATH environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)",
+    "text": "Sorting can be used to sort values or the indices of those values in either ascending or descending order. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1, -10.3,\n20.987),\n\\end{equation*}\\] then the various sort routines produce \\[\\begin{eqnarray*}\n\\mathrm{sort\\_asc(v)} & = &  (-10.3,1,20.987) \\\\[4pt]\n\\mathrm{sort\\_desc(v)} & = &  (20.987,1,-10.3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_asc(v)} & = &  (2,1,3) \\\\[4pt]\n\\mathrm{sort\\_indices\\_desc(v)} & = &  (3,1,2)\n\\end{eqnarray*}\\]\n \n\narray[] real sort_asc(array[] real v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] int sort_asc(array[] int v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\narray[] real sort_desc(array[] real v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_desc(array[] int v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] real v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(array[] int v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(array[] real v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(array[] int v, int s) Number of components of v less than v[s]\nAvailable since 2.0",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html#checking-the-stan-compiler",
-    "href": "cmdstan-guide/installation.html#checking-the-stan-compiler",
-    "title": "CmdStan Installation",
+    "objectID": "functions-reference/array_operations.html#reversing-functions",
+    "href": "functions-reference/array_operations.html#reversing-functions",
+    "title": "Array Operations",
     "section": "",
-    "text": "To check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.\nOn Linux and macOS:\n# compile the example\n&gt; make examples/bernoulli/bernoulli\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli sample\\\n  data file=examples/bernoulli/bernoulli.data.json\n\n# default output written to file `output.csv`,\n# default num_samples is 1000, output file should have approx. 1050 lines\n&gt; wc -l output.csv\n\n# run the `bin/stansummary utility to summarize parameter estimates\n&gt; bin/stansummary output.csv\nOn Windows:\n# compile the example\n&gt; make examples/bernoulli/bernoulli.exe\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json\n\n# run the `bin/stansummary.exe utility to summarize parameter estimates\n&gt; bin/stansummary.exe output.csv\nThe sample data in file bernoulli.json.data specifies 2 out of 10 successes, therefore the range mean(theta)\\(\\pm\\)sd(theta) should include 0.2.",
+    "text": "Stan provides functions to create a new array by reversing the order of elements in an existing array. For example, if v is declared as a real array of size 3, with values \\[\\begin{equation*}\n\\text{v} = (1,\\, -10.3,\\, 20.987),\n\\end{equation*}\\] then \\[\\begin{equation*}\n\\mathrm{reverse(v)} = (20.987,\\, -10.3,\\, 1).\n\\end{equation*}\\]\n \n\narray[] T reverse(array[] T v) Return a new array containing the elements of the argument in reverse order.\nAvailable since 2.23",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html#troubleshooting-the-installation",
-    "href": "cmdstan-guide/installation.html#troubleshooting-the-installation",
-    "title": "CmdStan Installation",
-    "section": "",
-    "text": "Updates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the &lt;cmdstan-home&gt; directory:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make clean-all\n&gt; make build\n\n\nThis section contains solutions to problems reported on https://discourse.mc-stan.org\nCompiler error message about PCH file\nTo speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:\nerror: PCH file uses an older PCH format that is no longer supported\nIn this case, clean and rebuild CmdStan, as shown in the previous section.\nWindows: ‘g++’, ‘make’, or ‘cut’ is not recognized\nThe CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:\n'cut' is not recognized as an internal or external command,\noperable program or batch file.\nTo fix this, ensure you have followed the steps for adding the toolchain to your PATH and installing the additional utilities covered in the configuration instructions\nSpaces in paths to CmdStan or model\nmake can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the One Drive folder.\nUnfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:\nmake: *** INTERNAL: readdir: Invalid argument\nmake: *** [make/program:50: x.hpp] Error 2\nIf the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.",
+    "objectID": "functions-reference/array_operations.html#footnotes",
+    "href": "functions-reference/array_operations.html#footnotes",
+    "title": "Array Operations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nDividing by \\(N\\) rather than \\((N-1)\\) produces a maximum likelihood estimate of variance, which is biased to underestimate variance.↩︎",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Functions Reference",
+      "Built-in Functions",
+      "Array Operations"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html#cpp-toolchain",
-    "href": "cmdstan-guide/installation.html#cpp-toolchain",
-    "title": "CmdStan Installation",
+    "objectID": "cmdstan-guide/stansummary.html",
+    "href": "cmdstan-guide/stansummary.html",
+    "title": "stansummary: MCMC Output Analysis",
     "section": "",
-    "text": "Compiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.\n\n\nThe required C++ compiler is g++ 4.9 3. On most systems the GNU Make utility is pre-installed and is the default make utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:\ng++ --version\nmake --version\nIf these are at least at g++ version 4.9.3 or later and make version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler g++, because later versions are faster.\nTo install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:\nsudo apt install g++\nsudo apt install make\nIf you can’t run sudo, you will need to ask your sysadmin or cluster administrator to install these tools for you.\n\n\n\nTo check if you already already have an appropriate toolchain installed, open the Terminal application and enter:\nclang++ --version\nmake --version\nIf either of these commands prints the message command not found, you will need to install Xcode’s command line tools.\nOpen the Terminal application and enter:\nxcode-select --install\nSelect “Install” in the window that opens.\nAfter the installation completes, you can double check that installation was successful by reopening the Terminal and running:\nclang++ --version\nmake --version\nYou can read more about Xcode on its site: https://developer.apple.com/xcode/\nWe don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.\n\n\n\nThe Windows toolchain consists of programs g++, the C++ compiler, and make, the GNU Make utility. To check if these are present, open a command shell [^1] and type:\ng++ --version\nmake --version\nCmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your PATH variable, See these instructions for details on changing the PATH if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.\n\n\nAll required utilities (e.g., make, g++) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the R Project documentation.\nThe R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.\nNote that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a Microsoft update.\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45\\usr\\bin\nC:\\rtools45\\x86_64-w64-mingw32.static.posix\\bin\n\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45 - ARM64\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45-aarch64\\usr\\bin\nC:\\rtools45-aarch64\\aarch64-w64-mingw32.static.posix\\bin",
+    "text": "The CmdStan stansummary program reports statistics for one or more sampler chains over all sampler and model parameters and quantities of interest. The statistics reported include both summary statistics of the estimates and diagnostic statistics on the sampler chains, reported in the following order:\n\nMean - sample mean\nMCSE - Monte Carlo Standard Error, a measure of the amount of noise in the sample\nStdDev - sample standard deviation - the standard deviation around the sample mean.\nMAD - Median Absolute Deviation - the median absolute deviation around the sample median.\nQuantiles - default 5%, 50%, 95%\nESS_bulk\nESS_tail\nESS_bulk/s - Bulk ESS per second\nR_hat - \\(\\hat{R}\\) statistic, a MCMC convergence diagnostic\n\nWhen reviewing the stansummary output, it is important to check the final three output columns first - these are the diagnostic statistics on MCMC convergence and effective sample size. A \\(\\hat{R}\\) statistic of greater than \\(1\\) indicates potential convergence problems and that the sample is not presentative of the target posterior, thus the estimates of the mean and all other summary statistics are likely to be invalid. A value \\(1.01\\) can be used as generic threshold to decide whether more iterations or further convergence analysis is needed, but other thresholds can be used depending on the specific use case.\nEstimation by sampling produces an approximate value for the model parameters; the MCSE statistic indicates the amount of uncertainty in the estimate. Therefore MCSE column is placed next to the sample mean column, in order to make it easy to compare this sample with others.\nFor more information, see the Posterior Analysis chapter of the Stan Reference Manual which describes both the theory and practice of MCMC estimation techniques.\nThe statistics - Mean, StdDev, MAD, and Quantiles - are computed directly from all draws across all chains. The diagnostic statistics - ESS_bulk, ESS_tail, and R_hat are computed from the rank-normalized, folded, and splitted chains according to the definitions by Vehtari et al. (2021). the MCSE statistic is computed using split chain R_hat and autocorrelations. The summary statistics and the algorithms used to compute them are described in sections Notation for draws and Effective sample size.\n\n\nThe CmdStan makefile task build compiles the stansummary utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/stansummary\n\n\n\nThe stansummary utility processes one or more output files from a set of chains from one run of the HMC sampler. To run stansummary on the output file or files generated by a run of the sampler, on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nFor example, after running 4 chains to fit the example model eight_schools.stan to the supplied example data file, we run stansummary on the resulting Stan CSV output files to get the following report:\n&gt; bin/stansummary eight_*.csv\nInference for Stan model: eight_schools_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total\nSampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total\n\n                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0\naccept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0\nstepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan\ntreedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0\nn_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0\ndivergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan\nenergy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0\n\nmu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0\ntheta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0\ntheta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0\ntheta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0\ntheta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0\ntheta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0\ntheta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0\ntheta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0\ntheta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0\ntau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nThe console output information consists of\n\nModel, chains, and timing summaries\nSampler parameter statistics\nModel parameter statistics\nSampling algorithm - either nuts (shown here) or static HMC.\n\nThere is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g. if there are no divergent transitions.\n\n\nThe initial Stan CSV columns provide information on the sampler state for each draw:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause we ran the NUTS sampler, the above summary reports sampler parameters treedepth__, n_leapfrog__, and divergent__; the static HMC sampler would report int_time__ instead.\n\n\n\nThe remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The eight_schools.stan program parameters block contains the following parameter variable declarations:\n  real mu;\n  array[J] real theta;\n  real&lt;lower=0&gt; tau;\nIn the example data, J is \\(8\\); therefore the stansummary listing reports on theta[1] through theta[8].\n\n\n\n\nThe stansummary command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the -h or --help option, the program prints the usage message to the console and exits.\nReport statistics for one or more Stan CSV files from a HMC sampler run.\nExample:  stansummary model_chain_1.csv model_chain_2.csv\nOptions:\n  -a, --autocorr [n]          Display the chain autocorrelation for the n-th\n                              input file, in addition to statistics.\n  -c, --csv_filename [file]   Write statistics to a CSV file.\n  -h, --help                  Produce help message, then exit.\n  -p, --percentiles [values]  Percentiles to report as ordered set of\n                              comma-separated numbers from (0.1,99.9), inclusive.\n                              Default is 5,50,95.\n  -s, --sig_figs [n]          Significant figures reported. Default is 2.\n                              Must be an integer from (1, 18), inclusive.\n  -i, --include_param [name]  Include the named parameter in the summary output.\n                              By default, all parameters in the file are summarized,\n                              passing this argument one or more times will filter\n                              the output down to just the requested arguments.\nBoth short an long option names are allowed. Short names are specified as -&lt;o&gt; &lt;value&gt;; long option names can be specified either as --&lt;option&gt;=&lt;value&gt; or --&lt;option&gt; &lt;value&gt;.\nThe --percentiles argument can also be passed an empty string \"\", which results in no percentiles being displayed in the output of the command.\nThe amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument sig_figs. The default sampler output precision is 8. The --sig_figs argument to the stansummary program should not exceed the sig_figs argument to the sampler.",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
     ]
   },
   {
-    "objectID": "cmdstan-guide/installation.html#gnu-make",
-    "href": "cmdstan-guide/installation.html#gnu-make",
-    "title": "CmdStan Installation",
+    "objectID": "cmdstan-guide/stansummary.html#building-the-stansummary-command",
+    "href": "cmdstan-guide/stansummary.html#building-the-stansummary-command",
+    "title": "stansummary: MCMC Output Analysis",
     "section": "",
-    "text": "CmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.\nGNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed only when there are changes to those other files. Thus Make provides an efficient way to manage complex software.\nThe CmdStan Makefile is in the &lt;cmdstan-home&gt; directory and is named makefile. This is one of the default GNU Makefile names, which allows you to omit the -f makefile argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, Make only works properly when invoked from the &lt;cmdstan-home&gt; directory; attempts to use this Makefile from another directory by specifying the full path to the file makefile won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:\nmake -f ~/github/stan-dev/cmdstan/makefile\n/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory\nmake: *** No rule to make target `make/command'.  Stop.\nThe conda-forge cmdstan package provides a solution to this problem via cmdstan_model command which lets you run the CmdStan makefile from anywhere to compile a Stan model.\nMakefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix .stan; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program my_program.stan in directory ../my_dir/, the make target is ../my_dir/my_program or ../my_dir/my_program.exe (on Windows).\nTo call Make, you invoke the utility name, make, followed by, in order:\n\nzero or more Make program options, then specify any Make variables as a series of\nzero of more Make variables, described below\nzero or more target names; the set of names is determined by the Makefile rules.\n\nmake &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;\nMakefile Variables\nMake targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile ../my_dir/my_program.stan for an OpenCL (GPU) machine, set the makefile variable STAN_OPENCL to TRUE:\n&gt; make STAN_OPENCL=TRUE ../my_dir/my_program\nMakefile variables can also be set by creating a file named local in the CmdStan make subdirectory which contains a list of &lt;VARIABLE&gt;=&lt;VALUE&gt; pairs, one per line. For example, to get the same effect as the above command every time, you would put the line STAN_OPENCL=TRUE into the file &lt;cmdstan_home&gt;/make/local.\nThe complete set of Makefile variables can be found in file &lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags.\nMake Targets\nWhen invoked without any arguments at all, Make prints a help message:\n&gt; make\n--------------------------------------------------------------------------------\nCmdStan v2.33.1 help\n\n  Build CmdStan utilities:\n    &gt; make build\n\n    This target will:\n    1. Install the Stan compiler bin/stanc from stanc3 binaries.\n    2. Build the print utility bin/print (deprecated; will be removed in v3.0)\n    3. Build the stansummary utility bin/stansummary\n    4. Build the diagnose utility bin/diagnose\n    5. Build all libraries and object files compile and link an executable Stan program\n\n    Note: to build using multiple cores, use the -j option to make, e.g.,\n    for 4 cores:\n    &gt; make build -j4\n\n\n  Build a Stan program:\n\n    Given a Stan program at foo/bar.stan, build an executable by typing:\n    &gt; make foo/bar\n\n    This target will:\n    1. Install the Stan compiler (bin/stanc), as needed.\n    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.\n    3. Compile the C++ code using cc . to generate foo/bar\n\n  Additional make options:\n    STANCFLAGS: defaults to \"\". These are extra options passed to bin/stanc\n      when generating C++ code. If you want to allow undefined functions in the\n      Stan program, either add this to make/local or the command line:\n          STANCFLAGS = --allow_undefined\n    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the\n      header file that is included. This defaults to \"user_header.hpp\" in the\n      directory of the Stan program.\n    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads\n      the nightly version.\n    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.\n    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.\n\n\n  Example - bernoulli model: examples/bernoulli/bernoulli.stan\n\n    1. Build the model:\n       &gt; make examples/bernoulli/bernoulli\n    2. Run the sampling algorithm given the model and data:\n       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R\n    3. Look at the posterior sample:\n       &gt; bin/stansummary output.csv\n\n\n  Clean CmdStan:\n\n    Remove the built CmdStan tools:\n    &gt; make clean-all\n\n--------------------------------------------------------------------------------",
+    "text": "The CmdStan makefile task build compiles the stansummary utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/stansummary",
     "crumbs": [
-      "Getting Started",
-      "CmdStan Installation"
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
     ]
   },
   {
-    "objectID": "cmdstan-guide/generate_quantities_config.html",
-    "href": "cmdstan-guide/generate_quantities_config.html",
-    "title": "Generating Quantities of Interest from a Fitted Model",
+    "objectID": "cmdstan-guide/stansummary.html#running-the-stansummary-program",
+    "href": "cmdstan-guide/stansummary.html#running-the-stansummary-program",
+    "title": "stansummary: MCMC Output Analysis",
     "section": "",
-    "text": "The generate_quantities method allows you to generate additional quantities of interest from a fitted model without re-running the sampler. Instead, you write a modified version of the original Stan program and add a generated quantities block or modify the existing one which specifies how to compute the new quantities of interest. Running the generate_quantities method on the new program together with sampler outputs (i.e., a set of draws) from the fitted model runs the generated quantities block of the new program using the the existing sample by plugging in the per-draw parameter estimates for the computations in the generated quantities block.\nThis method requires sub-argument fitted_params which takes as its value an existing Stan CSV file that contains a parameter values from an equivalent model, i.e., a model with the same parameters block, conditioned on the same data.\nThe generated quantities block computes quantities of interest (QOIs) based on the data, transformed data, parameters, and transformed parameters. It can be used to:\n\ngenerate simulated data for model testing by forward sampling\ngenerate predictions for new data\ncalculate posterior event probabilities, including multiple comparisons, sign tests, etc.\ncalculate posterior expectations\ntransform parameters for reporting\napply full Bayesian decision theory\ncalculate log likelihoods, deviances, etc. for model comparison\n\nFor an overview of the uses of this feature, see the Stan User’s Guide section on Stand-alone generated quantities and ongoing prediction.\n\n\nTo illustrate how this works we use the generate_quantities method to do posterior predictive checks using the estimate of theta given the example bernoulli model and data, following the posterior predictive simulation procedure in the Stan User’s Guide.\nWe write a program bernoulli_ppc.stan which contains the following generated quantities block, with comments to explain the procedure:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nThe rest of the program is the same as in bernoulli.stan.\nThe generate_method requires the sub-argument fitted_params which takes as its value the name of a Stan CSV file. The per-draw parameter values from the fitted_params file will be used to run the generated quantities block.\nIf we run the bernoulli.stan program for a single chain to generate a sample in file bernoulli_fit.csv:\n&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv\nThen we can run the bernoulli_ppc.stan to carry out the posterior predictive checks:\n&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \\\n                  data file=bernoulli.data.json \\\n                  output file=bernoulli_ppc.csv\nThe output file bernoulli_ppc.csv contains only the values for the variables declared in the generated quantities block, i.e., theta_rep and the elements of y_sim:\n# model = bernoulli_ppc_model\n# method = generate_quantities\n#   generate_quantities\n#     fitted_params = bernoulli_fit.csv\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2983956445 (Default)\n# output\n#   file = output.csv (Default)\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep\n1,1,1,0,0,0,1,1,0,1,0.6\n1,1,0,1,0,0,1,0,1,0,0.5\n1,0,1,1,1,1,1,1,0,1,0.8\n0,1,0,1,0,1,0,1,0,0,0.4\n1,0,0,0,0,0,0,0,0,0,0.1\n0,0,0,0,0,1,1,1,0,0,0.3\n0,0,1,0,1,0,0,0,0,0,0.2\n1,0,1,0,1,1,0,1,1,0,0.6\n...\nGiven the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  real&lt;lower=0, upper=1&gt; theta_cp = theta;\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nNow the output is slightly more interpretable: theta_cp is the same as the theta used to generate the values y_sim[1] through y_sim[1]. Comparing columns theta_cp and theta_rep allows us to see how the uncertainty in our estimate of theta is carried forward into our predictions:\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep\n0,1,1,0,1,0,0,1,1,0,0.545679,0.5\n1,1,1,1,1,1,0,1,1,0,0.527164,0.8\n1,1,1,1,0,1,1,1,1,0,0.529116,0.8\n1,0,1,1,1,1,0,0,1,0,0.478844,0.6\n0,1,0,0,0,0,1,0,1,0,0.238793,0.3\n0,0,0,0,0,1,1,0,0,0,0.258294,0.2\n1,1,1,0,0,0,0,0,0,0,0.258465,0.3\n\n\n\nThe fitted_params file must be a Stan CSV file; attempts to use a regular CSV file will result an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe fitted_params file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe parameter values of the fitted_params are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in bernoulli_fit.csv so that the value of theta is outside the declared bounds real&lt;lower=0, upper=1&gt;, the program will return the following error message:\nException: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \\\n(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)",
+    "text": "The stansummary utility processes one or more output files from a set of chains from one run of the HMC sampler. To run stansummary on the output file or files generated by a run of the sampler, on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe &lt;file_1.csv&gt; ... &lt;file_N.csv&gt;\nFor example, after running 4 chains to fit the example model eight_schools.stan to the supplied example data file, we run stansummary on the resulting Stan CSV output files to get the following report:\n&gt; bin/stansummary eight_*.csv\nInference for Stan model: eight_schools_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.065, 0.078, 0.080, 0.086) seconds, 0.31 seconds total\nSampling took (0.047, 0.044, 0.045, 0.053) seconds, 0.19 seconds total\n\n                 Mean   MCSE  StdDev    MAD       5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__              -19   0.31     4.9    5.0      -27   -19   -11       264       275        1396    1.0\naccept_stat__    0.77  0.024    0.31  0.096  6.5e-03  0.93  1.00       243       273        1287    1.0\nstepsize__       0.25    nan   0.016  0.016  2.2e-01  0.25  0.26       nan       nan         nan    nan\ntreedepth__       3.4  0.048    0.76   0.00  2.0e+00   4.0   4.0       285       295        1507    1.0\nn_leapfrog__       13   0.80     7.1   0.00  3.0e+00    15    31       220       274        1165    1.0\ndivergent__     0.015    nan    0.12   0.00  0.0e+00  0.00  0.00       nan       nan         nan    nan\nenergy__           24   0.32     5.4    5.5  1.5e+01    24    33       289       488        1527    1.0\n\nmu                7.8   0.20     5.5    4.9     -1.3   7.7    17       688       915        3641    1.0\ntheta[1]           12   0.28     8.7    7.4    -0.36    11    28       908       763        4802    1.0\ntheta[2]          7.7   0.19     6.8    6.1     -3.4   7.8    19      1194      2011        6320    1.0\ntheta[3]          5.6   0.23     8.5    7.0     -9.1   6.2    18      1260      1723        6669    1.0\ntheta[4]          7.5   0.20     7.0    6.5     -4.1   7.6    19      1171      1744        6197    1.0\ntheta[5]          4.6   0.21     6.7    6.3     -7.0   4.9    15      1045      1513        5530    1.0\ntheta[6]          5.7   0.23     7.2    6.4     -6.8   6.0    17      1012      1626        5354    1.0\ntheta[7]           11   0.24     7.1    6.6    0.025    11    24       885       473        4682    1.0\ntheta[8]          8.4   0.23     8.5    7.3     -4.8   8.1    23      1280      1848        6773    1.0\ntau               7.8   0.26     5.9    4.5      1.8   6.3    18       248       178        1310    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nThe console output information consists of\n\nModel, chains, and timing summaries\nSampler parameter statistics\nModel parameter statistics\nSampling algorithm - either nuts (shown here) or static HMC.\n\nThere is one row per parameter and the row order in the summary report corresponds to the column order in the Stan CSV output file. NaN values for some columns are expected if the value doesn’t change, e.g. if there are no divergent transitions.\n\n\nThe initial Stan CSV columns provide information on the sampler state for each draw:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause we ran the NUTS sampler, the above summary reports sampler parameters treedepth__, n_leapfrog__, and divergent__; the static HMC sampler would report int_time__ instead.\n\n\n\nThe remaining Stan CSV columns report the values of all parameters, transformed parameters, and generated quantities in the order in which these variables are declared in the Stan program. For container variables, i.e., vector, row_vector, matrix, and array variables, the statistics for each element are reported separately, in row-major order. The eight_schools.stan program parameters block contains the following parameter variable declarations:\n  real mu;\n  array[J] real theta;\n  real&lt;lower=0&gt; tau;\nIn the example data, J is \\(8\\); therefore the stansummary listing reports on theta[1] through theta[8].",
     "crumbs": [
-      "Running CmdStan",
-      "Generating Quantities of Interest from a Fitted Model"
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
     ]
   },
   {
-    "objectID": "cmdstan-guide/generate_quantities_config.html#example",
-    "href": "cmdstan-guide/generate_quantities_config.html#example",
-    "title": "Generating Quantities of Interest from a Fitted Model",
+    "objectID": "cmdstan-guide/stansummary.html#command-line-options",
+    "href": "cmdstan-guide/stansummary.html#command-line-options",
+    "title": "stansummary: MCMC Output Analysis",
     "section": "",
-    "text": "To illustrate how this works we use the generate_quantities method to do posterior predictive checks using the estimate of theta given the example bernoulli model and data, following the posterior predictive simulation procedure in the Stan User’s Guide.\nWe write a program bernoulli_ppc.stan which contains the following generated quantities block, with comments to explain the procedure:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nThe rest of the program is the same as in bernoulli.stan.\nThe generate_method requires the sub-argument fitted_params which takes as its value the name of a Stan CSV file. The per-draw parameter values from the fitted_params file will be used to run the generated quantities block.\nIf we run the bernoulli.stan program for a single chain to generate a sample in file bernoulli_fit.csv:\n&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv\nThen we can run the bernoulli_ppc.stan to carry out the posterior predictive checks:\n&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \\\n                  data file=bernoulli.data.json \\\n                  output file=bernoulli_ppc.csv\nThe output file bernoulli_ppc.csv contains only the values for the variables declared in the generated quantities block, i.e., theta_rep and the elements of y_sim:\n# model = bernoulli_ppc_model\n# method = generate_quantities\n#   generate_quantities\n#     fitted_params = bernoulli_fit.csv\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2983956445 (Default)\n# output\n#   file = output.csv (Default)\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep\n1,1,1,0,0,0,1,1,0,1,0.6\n1,1,0,1,0,0,1,0,1,0,0.5\n1,0,1,1,1,1,1,1,0,1,0.8\n0,1,0,1,0,1,0,1,0,0,0.4\n1,0,0,0,0,0,0,0,0,0,0.1\n0,0,0,0,0,1,1,1,0,0,0.3\n0,0,1,0,1,0,0,0,0,0,0.2\n1,0,1,0,1,1,0,1,1,0,0.6\n...\nGiven the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  real&lt;lower=0, upper=1&gt; theta_cp = theta;\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nNow the output is slightly more interpretable: theta_cp is the same as the theta used to generate the values y_sim[1] through y_sim[1]. Comparing columns theta_cp and theta_rep allows us to see how the uncertainty in our estimate of theta is carried forward into our predictions:\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep\n0,1,1,0,1,0,0,1,1,0,0.545679,0.5\n1,1,1,1,1,1,0,1,1,0,0.527164,0.8\n1,1,1,1,0,1,1,1,1,0,0.529116,0.8\n1,0,1,1,1,1,0,0,1,0,0.478844,0.6\n0,1,0,0,0,0,1,0,1,0,0.238793,0.3\n0,0,0,0,0,1,1,0,0,0,0.258294,0.2\n1,1,1,0,0,0,0,0,0,0,0.258465,0.3",
+    "text": "The stansummary command syntax provides a set of flags to customize the output which must precede the list of filenames. When invoked with no arguments or with the -h or --help option, the program prints the usage message to the console and exits.\nReport statistics for one or more Stan CSV files from a HMC sampler run.\nExample:  stansummary model_chain_1.csv model_chain_2.csv\nOptions:\n  -a, --autocorr [n]          Display the chain autocorrelation for the n-th\n                              input file, in addition to statistics.\n  -c, --csv_filename [file]   Write statistics to a CSV file.\n  -h, --help                  Produce help message, then exit.\n  -p, --percentiles [values]  Percentiles to report as ordered set of\n                              comma-separated numbers from (0.1,99.9), inclusive.\n                              Default is 5,50,95.\n  -s, --sig_figs [n]          Significant figures reported. Default is 2.\n                              Must be an integer from (1, 18), inclusive.\n  -i, --include_param [name]  Include the named parameter in the summary output.\n                              By default, all parameters in the file are summarized,\n                              passing this argument one or more times will filter\n                              the output down to just the requested arguments.\nBoth short an long option names are allowed. Short names are specified as -&lt;o&gt; &lt;value&gt;; long option names can be specified either as --&lt;option&gt;=&lt;value&gt; or --&lt;option&gt; &lt;value&gt;.\nThe --percentiles argument can also be passed an empty string \"\", which results in no percentiles being displayed in the output of the command.\nThe amount of precision in the sampler output limits the amount of real precision in the summary report. CmdStan’s command line interface also has output argument sig_figs. The default sampler output precision is 8. The --sig_figs argument to the stansummary program should not exceed the sig_figs argument to the sampler.",
     "crumbs": [
-      "Running CmdStan",
-      "Generating Quantities of Interest from a Fitted Model"
+      "Tools and Utilities",
+      "`stansummary`: MCMC Output Analysis"
     ]
   },
   {
-    "objectID": "cmdstan-guide/generate_quantities_config.html#errors",
-    "href": "cmdstan-guide/generate_quantities_config.html#errors",
-    "title": "Generating Quantities of Interest from a Fitted Model",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html",
+    "href": "cmdstan-guide/stan_csv_apdx.html",
+    "title": "Stan CSV File Format",
     "section": "",
-    "text": "The fitted_params file must be a Stan CSV file; attempts to use a regular CSV file will result an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe fitted_params file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe parameter values of the fitted_params are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in bernoulli_fit.csv so that the value of theta is outside the declared bounds real&lt;lower=0, upper=1&gt;, the program will return the following error message:\nException: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \\\n(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)",
+    "text": "The output from all CmdStan methods is in CSV format. A Stan CSV file is a data table where the columns are the method and model parameters and quantities of interest. Each row contains one record’s worth of data in plain-text format using the comma character (‘,’) as the field delimiter (hence the name).\nFor the Stan CSV files, data is strictly numerical, however, possible values include both positive and negative infinity and “Not-a-Number” which are represented as the strings NaN, inf, +inf, -inf. All other values are written in decimal notation by default with at most 8 digits of precision. The number of significant digits written can be controlled with argument sig_figs=&lt;int&gt;. See more in Output control arguments section.\nStan CSV files have a header row containing the column names. They also make extensive use of CSV comments, i.e., lines which begin with the # character. In addition to initial and final comment rows, some methods also put comment rows in the middle of the data table, which makes it difficult to use many of the commonly used CSV parser packages.\n\n\nThe data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.\nStan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is\n\nThe column name consists of the variable name followed by the element indices.\nIndices are delimited by periods (‘.’).\nIndexing is 1-based, i.e., given a dimension of size \\(N\\), the first element index is \\(1\\) and the last element index is \\(N\\).\nTuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).\n\nContainer variables are serialized in column major order, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.\nTo see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:\n array[2, 3, 4] real foo;\nThe Stan CSV file will require 24 columns to output the elements of foo. The first 6 columns will be labeled:\nfoo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1\nThe final 6 columns will be labeled:\nfoo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4\nTo see how a tuple would be laid out, consider the following variable:\ntuple(real, array[3] real) bar;\nThis will correspond to 4 columns in the CSV file, which are labeled\nbar:1,bar:2.1,bar:2.2,bar:2.3\n\n\n\nThe sample method produces both a Stan CSV output file and a diagnostic file which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.\nTo see how this works, we show snippets of the output file resulting from the following command:\n./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \\\n            data file=bernoulli.data.json \\\n            output file=bernoulli_samples.csv\n\n\nThe sampler output file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nComment rows for adaptation listing step size and metric used for sampling\nSampling draws\nComment rows giving timing information\n\nInitial comments rows: argument configuration\nAll configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as (Default).\nIn the above example the num_samples, num_warmup, and save_warmup arguments were specified, whereas subargument thin is left at its default value, as seen in the initial comment rows:\n# stan_version_major = 2\n# stan_version_minor = 24\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 100\n#     num_warmup = 200\n#     save_warmup = 1\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000000000000003 (Default)\n#       delta = 0.80000000000000004 (Default)\n#       kappa = 0.75 (Default)\n#       t0 = 10 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1 (Default)\n#         stepsize_jitter = 0 (Default)\n# id = 0 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2991989946 (Default)\n# output\n#   file = bernoulli_samples.csv\n#   diagnostic_file = bernoulli_diagnostics.csv\n#   refresh = 100 (Default)\nNote that when running multi-threaded programs which use reduce_sum for high-level parallelization, the number of threads used will also be included in this initial comment header.\nColumn headers\nThe CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the output file section of the chapter on MCMC Sampling. The example model bernoulli.stan only contains one parameter theta, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nAs a second example, we show the output of the eight_schools.stan model on run on example dataset. This model has 3 parameters: mu, theta a vector whose length is dependent on the input data, here N = 8, and tau. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:\nmu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau\nData rows containing warmup draws\nWhen run with option save_warmup=1, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is save_warmup=0, this section is usually not present in the output file.\nHere we specified num_warmup=200 and left thin at the default value \\(1\\), therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.\n-6.74827,1,1,1,1,0,6.75348,0.247195\n-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195\n-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195\n-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619\n-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517\nComment rows for adaptation\nDuring warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.\n# Adaptation terminated\n# Step size = 0.813694\n# Diagonal elements of inverse mass matrix:\n# 0.592879\nAs the example bernoulli model only contains a single parameter, and as the default metric is diag_e, the inverse mass matrix is a \\(1 \\times 1\\) matrix, and the length of the diagonal vector is also \\(1\\).\nIn contrast, if we run the eight schools example model with metric dense_e, the adaptation comments section lists both the stepsize and the full \\(10 \\times 10\\) inverse mass matrix:\n# Adaptation terminated\n# Step size = 0.211252\n# Elements of inverse mass matrix:\n# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757\n# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028\n# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353\n# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097\n# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245\n# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507\n# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946\n# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661\n# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104\n# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433\nNote that when the sampler is run with arguments algorithm=fixed_param, this section will be missing.\nData rows containing sampling draws\nThe output file contains the values for the thinned set draws during sampling. Here we specified num_sampling=100 and left thin at the default value \\(1\\), therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.\n-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093\n-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431\n-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341\nTiming information\nUpon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:\n#\n#  Elapsed Time: 0.005 seconds (Warm-up)\n#                0.002 seconds (Sampling)\n#                0.007 seconds (Total)\n#\n\n\n\nThe diagnostic file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nSampling draws\nComment rows giving timing information\n\nThe columns in this file contain, in order:\n\nall sampler parameters\nall model parameter estimates (on the unconstrained scale)\nthe latent Hamiltonian for each parameter\nthe gradient for each parameters\n\nThe labels for the latent Hamiltonian columns are the parameter column label with prefix p_ and the labels for the gradient columns are the parameter column label with prefix g_.\nThese are the column labels from the file bernoulli_diagnostic.csv:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta\n\n\n\nThe profiling information is stored in a plain CSV format with no meta information in the comments.\nEach row represents timing information collected in a profile statement for a given thread. It is possible that some profile statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).\nThe columns are as follows:\n\nname, The name of the profile statement that is being timed\nthread_id, The thread that executed the profile statement\ntotal_time, The combined time spent executing statements inside the profile which includes calculation with and without automatic differentiation\nforward_time, The time spent in the profile statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation\nreverse_time, The time spent in the profile statement during the reverse (backward) pass of reverse mode automatic differentiation\nchain_stack, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass\nno_chain_stack, The number of objects allocated on the non-chaining automatic differentiation stack\nautodiff_calls, The total number of times the profile statement was executed with automatic differentiation\nno_autodiff_calls - The total number of times the profile statement was executed without automatic differentiation\n\n\n\n\n\n\nConfig as comments\nHeader row\nPenalized maximum likelihood estimate\n\n\n\n\n\nConfig as comments\nHeader row\nAdaptation as comments\nVariational estimate\nSample draws from estimate of the posterior\n\n\n\n\n\nHeader row\nQuantities of interest\n\n\n\n\n\nHeader row\nGradients",
     "crumbs": [
-      "Running CmdStan",
-      "Generating Quantities of Interest from a Fitted Model"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/example_model_data.html",
-    "href": "cmdstan-guide/example_model_data.html",
-    "title": "Example Model and Data",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#csv-column-names-and-order",
+    "href": "cmdstan-guide/stan_csv_apdx.html#csv-column-names-and-order",
+    "title": "Stan CSV File Format",
     "section": "",
-    "text": "The following is a simple, complete Stan program for a Bernoulli model of binary data.1 The model assumes the binary observed data y[1],...,y[N] are i.i.d. with Bernoulli chance-of-success theta.\ndata { \n  int&lt;lower=0&gt; N; \n  array[N] int&lt;lower=0, upper=1&gt; y;\n} \nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n} \nmodel {\n  theta ~ beta(1, 1);  // uniform prior on interval 0,1\n  y ~ bernoulli(theta);\n}\nThe input data file contains definitions for the two variables N and y which are specified in the data block of program bernoulli.stan (above).\nA data set of N=10 observations is included in the example Bernoulli model directory in both JSON notation and Rdump data format where 8 out of 10 trials had outcome 0 (failure) and 2 trials had outcome 1 (success). In JSON, this data is:\n{\n    \"N\" : 10,\n    \"y\" : [0,1,0,0,0,0,0,0,0,1]\n}",
+    "text": "The data table is laid out with zero or more method-specific columns followed by the Stan program variables declared in the parameter block, then the variables in the transformed parameters block, finally variables declared in the generated quantities, in declaration order.\nStan provides three types of container objects: arrays, vectors, and matrices. In order to output all elements of a container object, it is necessary to choose an indexing notation and a serialization order. The Stan CSV file indexing notation is\n\nThe column name consists of the variable name followed by the element indices.\nIndices are delimited by periods (‘.’).\nIndexing is 1-based, i.e., given a dimension of size \\(N\\), the first element index is \\(1\\) and the last element index is \\(N\\).\nTuples are laid out element-by-element, with each tuple slot being delimited by a colon (‘:’).\n\nContainer variables are serialized in column major order, a.k.a. “Fortran” order. In column major-order for a 2-D container, all elements of column 1 are listed in ascending order, followed by all elements of column 2, thus the column index changes the slowest and the row index changes the fastest. For higher dimensions, this generalizes to the last index changing the slowest and first index changing the fastest.\nTo see how this works, consider a 3-dimensional variable with dimension sizes 2, 3, and 4, e.g., an array of matrices, a 2-D array of vectors or row_vectors, or a 3-D array of scalars. Given a Stan program with model parameter variable:\n array[2, 3, 4] real foo;\nThe Stan CSV file will require 24 columns to output the elements of foo. The first 6 columns will be labeled:\nfoo.1.1.1,foo.2.1.1,foo.1.2.1,foo.2.2.1,foo.1.3.1,foo.2.3.1\nThe final 6 columns will be labeled:\nfoo.1.1.4,foo.2.1.4,foo.1.2.4,foo.2.2.4,foo.1.3.4,foo.2.3.4\nTo see how a tuple would be laid out, consider the following variable:\ntuple(real, array[3] real) bar;\nThis will correspond to 4 columns in the CSV file, which are labeled\nbar:1,bar:2.1,bar:2.2,bar:2.3",
     "crumbs": [
-      "Getting Started",
-      "Example Model and Data"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/example_model_data.html#footnotes",
-    "href": "cmdstan-guide/example_model_data.html#footnotes",
-    "title": "Example Model and Data",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe model is available with the CmdStan distribution at the path &lt;cmdstan-home&gt;/examples/bernoulli/bernoulli.stan.↩︎",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#mcmc-sampler-csv-output",
+    "href": "cmdstan-guide/stan_csv_apdx.html#mcmc-sampler-csv-output",
+    "title": "Stan CSV File Format",
+    "section": "",
+    "text": "The sample method produces both a Stan CSV output file and a diagnostic file which contains the sampler parameters together with the gradients on the unconstrained scale and log probabilities for all parameters in the model.\nTo see how this works, we show snippets of the output file resulting from the following command:\n./bernoulli sample save_warmup=1 num_warmup=200 num_samples=100 \\\n            data file=bernoulli.data.json \\\n            output file=bernoulli_samples.csv\n\n\nThe sampler output file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nComment rows for adaptation listing step size and metric used for sampling\nSampling draws\nComment rows giving timing information\n\nInitial comments rows: argument configuration\nAll configuration arguments are listed, one per line, indented according to CmdStan’s hierarchy of arguments and sub-arguments. Arguments not overtly specified on the command line are annotated as (Default).\nIn the above example the num_samples, num_warmup, and save_warmup arguments were specified, whereas subargument thin is left at its default value, as seen in the initial comment rows:\n# stan_version_major = 2\n# stan_version_minor = 24\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 100\n#     num_warmup = 200\n#     save_warmup = 1\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000000000000003 (Default)\n#       delta = 0.80000000000000004 (Default)\n#       kappa = 0.75 (Default)\n#       t0 = 10 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1 (Default)\n#         stepsize_jitter = 0 (Default)\n# id = 0 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2991989946 (Default)\n# output\n#   file = bernoulli_samples.csv\n#   diagnostic_file = bernoulli_diagnostics.csv\n#   refresh = 100 (Default)\nNote that when running multi-threaded programs which use reduce_sum for high-level parallelization, the number of threads used will also be included in this initial comment header.\nColumn headers\nThe CSV header row lists all sampler parameters, model parameters, transformed parameters, and quantities of interest. The sampler parameters are described in detail in the output file section of the chapter on MCMC Sampling. The example model bernoulli.stan only contains one parameter theta, therefore the CSV file data table consists of 7 sampler parameter columns and one column for the model parameter:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nAs a second example, we show the output of the eight_schools.stan model on run on example dataset. This model has 3 parameters: mu, theta a vector whose length is dependent on the input data, here N = 8, and tau. The initial columns are for the 7 sampler parameters, as before. The column headers for the model parameters are:\nmu,theta.1,theta.2,theta.3,theta.4,theta.5,theta.6,theta.7,theta.8,tau\nData rows containing warmup draws\nWhen run with option save_warmup=1, the thinned warmup draws are written to the CSV output file directly after the CSV header line. Since the default option is save_warmup=0, this section is usually not present in the output file.\nHere we specified num_warmup=200 and left thin at the default value \\(1\\), therefore the next 200 lines are data rows containing the sampler and model parameter values for each warmup draw.\n-6.74827,1,1,1,1,0,6.75348,0.247195\n-6.74827,4.1311e-103,14.3855,1,1,0,6.95087,0.247195\n-6.74827,1.74545e-21,2.43117,1,1,0,7.67546,0.247195\n-6.77655,0.99873,0.239791,2,7,0,6.81982,0.280619\n-6.7552,0.999392,0.323158,1,3,0,6.79175,0.26517\nComment rows for adaptation\nDuring warmup, the sampler adjusts the stepsize and the metric. At the end warmup, the sampler outputs this information as comments.\n# Adaptation terminated\n# Step size = 0.813694\n# Diagonal elements of inverse mass matrix:\n# 0.592879\nAs the example bernoulli model only contains a single parameter, and as the default metric is diag_e, the inverse mass matrix is a \\(1 \\times 1\\) matrix, and the length of the diagonal vector is also \\(1\\).\nIn contrast, if we run the eight schools example model with metric dense_e, the adaptation comments section lists both the stepsize and the full \\(10 \\times 10\\) inverse mass matrix:\n# Adaptation terminated\n# Step size = 0.211252\n# Elements of inverse mass matrix:\n# 25.6389, 17.3379, 13.9455, 15.9036, 15.1953, 8.73729, 16.9486, 14.4231, 17.4969, 0.518757\n# 17.3379, 79.8719, 12.2989, -1.28006, 9.92895, -3.51622, 10.073, 22.0196, 19.8151, 4.71028\n# 13.9455, 12.2989, 36.1572, 12.8734, 11.9446, 9.09582, 9.74519, 10.9539, 12.1204, 0.211353\n# 15.9036, -1.28006, 12.8734, 59.9998, 10.245, 8.03461, 16.9754, 3.13443, 9.68292, -1.36097\n# 15.1953, 9.92895, 11.9446, 10.245, 43.548, 15.3403, 13.0537, 7.69818, 10.1093, 0.155245\n# 8.73729, -3.51622, 9.09582, 8.03461, 15.3403, 39.981, 12.7695, 1.16248, 6.13749, -2.08507\n# 16.9486, 10.073, 9.74519, 16.9754, 13.0537, 12.7695, 45.8884, 11.6074, 8.96413, -1.15946\n# 14.4231, 22.0196, 10.9539, 3.13443, 7.69818, 1.16248, 11.6074, 49.4083, 18.9169, 3.15661\n# 17.4969, 19.8151, 12.1204, 9.68292, 10.1093, 6.13749, 8.96413, 18.9169, 68.0228, 1.74104\n# 0.518757, 4.71028, 0.211353, -1.36097, 0.155245, -2.08507, -1.15946, 3.15661, 1.74104, 1.50433\nNote that when the sampler is run with arguments algorithm=fixed_param, this section will be missing.\nData rows containing sampling draws\nThe output file contains the values for the thinned set draws during sampling. Here we specified num_sampling=100 and left thin at the default value \\(1\\), therefore the next 100 lines are data rows containing the sampler and model parameter values for each sampling iteration.\n-8.76921,0.796814,0.813694,1,1,0,9.75854,0.535093\n-6.79143,0.979604,0.813694,1,3,0,9.13092,0.214431\n-6.79451,0.955359,0.813694,2,3,0,7.19149,0.289341\nTiming information\nUpon successful completion, the sampler writes timing information to the output CSV file as a series of final comment lines:\n#\n#  Elapsed Time: 0.005 seconds (Warm-up)\n#                0.002 seconds (Sampling)\n#                0.007 seconds (Total)\n#\n\n\n\nThe diagnostic file contains the following:\n\nInitial comment rows listing full CmdStan argument configuration.\nHeader row\nData rows containing warmup draws, if run with option save_warmup=1\nSampling draws\nComment rows giving timing information\n\nThe columns in this file contain, in order:\n\nall sampler parameters\nall model parameter estimates (on the unconstrained scale)\nthe latent Hamiltonian for each parameter\nthe gradient for each parameters\n\nThe labels for the latent Hamiltonian columns are the parameter column label with prefix p_ and the labels for the gradient columns are the parameter column label with prefix g_.\nThese are the column labels from the file bernoulli_diagnostic.csv:\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta,p_theta,g_theta\n\n\n\nThe profiling information is stored in a plain CSV format with no meta information in the comments.\nEach row represents timing information collected in a profile statement for a given thread. It is possible that some profile statements have only one entry (if they were only executed by one thread) and others have multiple entries (if they were executed by multiple threads).\nThe columns are as follows:\n\nname, The name of the profile statement that is being timed\nthread_id, The thread that executed the profile statement\ntotal_time, The combined time spent executing statements inside the profile which includes calculation with and without automatic differentiation\nforward_time, The time spent in the profile statement during the forward pass of a reverse mode automatic differentiation calculation or during a calculation without automatic differentiation\nreverse_time, The time spent in the profile statement during the reverse (backward) pass of reverse mode automatic differentiation\nchain_stack, The number of objects allocated on the chaining automatic differentiation stack. There is a function call for each of these objects in the reverse pass\nno_chain_stack, The number of objects allocated on the non-chaining automatic differentiation stack\nautodiff_calls, The total number of times the profile statement was executed with automatic differentiation\nno_autodiff_calls - The total number of times the profile statement was executed without automatic differentiation",
     "crumbs": [
-      "Getting Started",
-      "Example Model and Data"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/diagnose_config.html",
-    "href": "cmdstan-guide/diagnose_config.html",
-    "title": "Diagnosing HMC by Comparison of Gradients",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#optimization-output",
+    "href": "cmdstan-guide/stan_csv_apdx.html#optimization-output",
+    "title": "Stan CSV File Format",
     "section": "",
-    "text": "Diagnosing HMC by Comparison of Gradients\nCmdStan has a basic diagnostic feature that will calculate the gradients of the initial state and compare them with gradients calculated by finite differences. Discrepancies between the two indicate that there is a problem with the model or initial states or else there is a bug in Stan.\nTo allow for the possibility of adding other kinds of diagnostic tests, the diagnose method argument configuration has subargument test which currently only takes value gradient. There are two available gradient test configuration arguments:\n\nepsilon - The finite difference step size. Must be a positive real number. Default value is \\(1^{-6}\\)\nerror - The error threshold. Must be a positive real number. Default value is \\(1^{-6}\\)\n\nTo run on the different platforms with the default configuration, use one of the following.\nMac OS and Linux\n&gt; ./my_model diagnose data file=my_data\nWindows\n&gt; my_model diagnose data file=my_data\nTo relax the test threshold, specify the error argument as follows:\n&gt; ./my_model diagnose test=gradient error=0.0001 data file=my_data\nTo see how this works, we run diagnostics on the example bernoulli model:\n&gt; ./bernoulli diagnose data file=bernoulli.data.R\nExecuting this command prints output to the console and as a series of comment lines to the output csv file. The console output is:\nmethod = diagnose\n  diagnose\n    test = gradient (Default)\n      gradient\n        epsilon = 9.9999999999999995e-07 (Default)\n        error = 9.9999999999999995e-07 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2152196153 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n\nTEST GRADIENT MODE\n\n Log probability=-8.42814\n\n param idx           value           model     finite diff           error\n         0       0.0361376         -3.1084         -3.1084    -2.37554e-10\nThe same information is printed to the output file as csv comments, i.e., each line is prefixed with a pound sign #.\n\n\n\n\n Back to top",
+    "text": "Config as comments\nHeader row\nPenalized maximum likelihood estimate",
     "crumbs": [
-      "Running CmdStan",
-      "Diagnosing HMC by Comparison of Gradients"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html",
-    "href": "cmdstan-guide/command_line_options.html",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#variational-inference-output",
+    "href": "cmdstan-guide/stan_csv_apdx.html#variational-inference-output",
+    "title": "Stan CSV File Format",
     "section": "",
-    "text": "A CmdStan executable is built from the Stan model concept and the CmdStan command line parser. The command line argument syntax consists of sets of keywords and keyword-value pairs. Arguments are grouped by the following keywords:\n\nmethod - specifies the kind of inference done on the model. Each kind of inference requires further configuration via sub-arguments. The method argument is required. It can be specified overtly as the a keyword-value pair method=&lt;inference&gt; or implicitly as one of the following:\n\nsample - obtain a sample (set of draws) from the posterior using HMC\noptimize - penalized maximum likelihood estimation\nvariational - automatic variational inference\ngenerate_quantities - run model’s generated quantities block on existing sample to obtain new quantities of interest.\nlog_prob - compute the log probability and gradient of the model for one set of parameters.\ndiagnose - compute and compare sampler gradient calculations to finite differences.\n\ndata - specifies the input data file, if any.\noutput - specifies program outputs, both disk files and terminal window outputs.\ninit - specifies initial values for the model parameters, if any.\nrandom - specifies the seed for the pseudo-random number.\n\nThe remainder of this chapter covers the general configuration options used for all processing. The following chapters cover the per-inference configuration options.\n\n\nThe values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:\ndata file=&lt;filepath&gt;\nThe keyword data must be followed directly by the keyword-value pair file=&lt;filepath&gt;. If the model doesn’t declare any data variables, this argument is ignored.\nThe input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model bernoulli.stan defines two data variables N and y. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.\nFor example if the input data file doesn’t include the definition for variable y, the executable exits with the following message:\nException: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)\n\n\n\nThe output keyword is used to specify non-default options for output files and messages written to the terminal window. The output keyword takes several keyword-value pair sub-arguments.\nThe keyword value pair file=&lt;filepath&gt; specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named output.csv in the current working directory.\nThe keyword value pair diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms sample and variational.\nThe keyword value pair refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.\nThe keyword value pair sig_figs=&lt;int&gt; specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8.   Note: increasing sig_figs above the default will increase the size of the output CSV files accordingly.\nThe keyword value pair profile_file=&lt;filepath&gt; specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and profile_file is unspecified, the profiling data is written to a file named profile.csv in the current working directory.\nThe keyword value pair save_cmdstan_config=&lt;boolean&gt; specifies whether to save the configuration options used to run the program to a file named &lt;output file&gt;_config.json alongside the other output files. The default value is false, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.\n\n\n\nInitialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range \\([-2, 2]\\). These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\nThe initialization argument is specified as keyword-value pair with keyword init. The value can be one of the following:\n\npositive real number \\(x\\). All parameters will be initialized to random draws from a uniform distribution over the range \\([-x, x]\\).\n\\(0\\) - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: \\(0\\) for unconstrained parameters; \\(1\\) for parameters constrained to be positive; \\(0.5\\) for variables to constrained to lie between \\(0\\) and \\(1\\); a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.\nfilepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range \\([-2, 2]\\).\n\n\n\n\nThe random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nThe syntax for the random seed argument is:\nrandom seed=&lt;int&gt;\nThe keyword random must be followed directly by the keyword-value pair seed=&lt;int&gt;.\n\n\n\nThe chain identifier argument is used in conjunction with the random seed argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.\nThe syntax for the random seed argument is:\nid=&lt;int&gt;\nThe default value is 1.\nWhen running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.\nFor complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the Stan Reference Manual Reproducibility chapter for further details.\n\n\n\nCmdStan provides a help and help-all mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword help:\n./bernoulli help\n\n\n\nCmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.\nAll program executables provide a return code between 0 and 255:\n\n0 - Program ran to termination as expected.\nvalue in range [1 : 125] - Method invoked could not run due to problems with model or data.\nvalue &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g. return code 139 results from termination signal 11 (segmentation violation).\n\nA non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.",
+    "text": "Config as comments\nHeader row\nAdaptation as comments\nVariational estimate\nSample draws from estimate of the posterior",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#input-data-argument",
-    "href": "cmdstan-guide/command_line_options.html#input-data-argument",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#generate-quantities-outputs",
+    "href": "cmdstan-guide/stan_csv_apdx.html#generate-quantities-outputs",
+    "title": "Stan CSV File Format",
     "section": "",
-    "text": "The values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:\ndata file=&lt;filepath&gt;\nThe keyword data must be followed directly by the keyword-value pair file=&lt;filepath&gt;. If the model doesn’t declare any data variables, this argument is ignored.\nThe input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model bernoulli.stan defines two data variables N and y. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.\nFor example if the input data file doesn’t include the definition for variable y, the executable exits with the following message:\nException: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)",
+    "text": "Header row\nQuantities of interest",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#output-control.section",
-    "href": "cmdstan-guide/command_line_options.html#output-control.section",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/stan_csv_apdx.html#diagnose-method-outputs",
+    "href": "cmdstan-guide/stan_csv_apdx.html#diagnose-method-outputs",
+    "title": "Stan CSV File Format",
     "section": "",
-    "text": "The output keyword is used to specify non-default options for output files and messages written to the terminal window. The output keyword takes several keyword-value pair sub-arguments.\nThe keyword value pair file=&lt;filepath&gt; specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named output.csv in the current working directory.\nThe keyword value pair diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms sample and variational.\nThe keyword value pair refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.\nThe keyword value pair sig_figs=&lt;int&gt; specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8.   Note: increasing sig_figs above the default will increase the size of the output CSV files accordingly.\nThe keyword value pair profile_file=&lt;filepath&gt; specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and profile_file is unspecified, the profiling data is written to a file named profile.csv in the current working directory.\nThe keyword value pair save_cmdstan_config=&lt;boolean&gt; specifies whether to save the configuration options used to run the program to a file named &lt;output file&gt;_config.json alongside the other output files. The default value is false, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.",
+    "text": "Header row\nGradients",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Appendices",
+      "Stan CSV File Format"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#initialize-model-parameters-argument",
-    "href": "cmdstan-guide/command_line_options.html#initialize-model-parameters-argument",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/print.html",
+    "href": "cmdstan-guide/print.html",
+    "title": "print (deprecated): MCMC Output Analysis",
     "section": "",
-    "text": "Initialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range \\([-2, 2]\\). These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\nThe initialization argument is specified as keyword-value pair with keyword init. The value can be one of the following:\n\npositive real number \\(x\\). All parameters will be initialized to random draws from a uniform distribution over the range \\([-x, x]\\).\n\\(0\\) - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: \\(0\\) for unconstrained parameters; \\(1\\) for parameters constrained to be positive; \\(0.5\\) for variables to constrained to lie between \\(0\\) and \\(1\\); a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.\nfilepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range \\([-2, 2]\\).",
+    "text": "print (deprecated): MCMC Output Analysis\nThe print utility is deprecated, but is still available until CmdStan v3.0. It has been replaced by the stansummary utility.\n\n\n\n\n Back to top",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Tools and Utilities",
+      "`print` (deprecated): MCMC Output Analysis"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#random-number-generator-arguments",
-    "href": "cmdstan-guide/command_line_options.html#random-number-generator-arguments",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/parallelization.html",
+    "href": "cmdstan-guide/parallelization.html",
+    "title": "Parallelization",
     "section": "",
-    "text": "The random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nThe syntax for the random seed argument is:\nrandom seed=&lt;int&gt;\nThe keyword random must be followed directly by the keyword-value pair seed=&lt;int&gt;.",
+    "text": "Stan provides three ways of parallelizing execution of a Stan model:\n\nmulti-threading with Intel Threading Building Blocks (TBB),\nmulti-processing with Message Passing Interface (MPI) and\nmanycore processing with OpenCL.\n\n\n\nIn order to exploit multi-threading in a Stan model, the models must be rewritten to use the reduce_sum and map_rect functions. For instructions on how to rewrite Stan models to use these functions see Stan’s User guide chapter on parallelization, the reduce_sum case study or the Multithreading and Map-Reduce tutorial.\n\n\nOnce a model is rewritten to use the above-mentioned functions, the model must be compiled with the STAN_THREADS makefile flag. The flag can be supplied in the make call but we recommend writing the flag to the make/local file. If the STAN_THREADS flag is defined/non-empty, threads will be enabled.\nAn example of the contents of make/local to enable threading with TBB:\nSTAN_THREADS=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nBefore running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the num_threads argument. Valid values for num_threads are positive integers and -1. If num_threads is set to -1, all available cores will be used.\nGenerally, this number should not exceed the number of available cores for best performance.\nExample:\n./model sample data file=data.json num_threads=4 ...\nWhen the model is compiled with STAN_THREADS we can sample with multiple chains with a single executable (see section running multiple chains for cases when this is available). When running multiple chains num_threads is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:\n./model sample num_chains=2 data file=data.json num_threads=8 ...\n\n\n\n\nIn order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the map_rect function. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.\n\n\nCompiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are MPICH and OpenMPI.\n\n\n\nOnce a model is rewritten to use map_rect, additional makefile flags must be written to the make/local. These are:\n\nSTAN_MPI: Enables the use of MPI with Stan if defined.\nCXX: The name of the MPI C++ compiler wrapper. Typically mpicxx.\nTBB_CXX_TYPE: The C++ compiler the MPI wrapper wraps. Typically gcc on Linux and clang on macOS.\n\nAn example of make/local on Linux:\nSTAN_MPI=true\nCXX=mpicxx\nTBB_CXX_TYPE=gcc\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_MPI is run using an MPI launcher. The MPI standard suggests using mpiexec, but a vendor wrapper for the launcher like mpirun can also be used. The launcher is supplied the path to the built executable and the number of processes to start: -n X for mpiexec or -np X for mpirun where X is replaced by the integer representing the number of processes.\nExample for running a model with six processes:\nmpiexec -n 6 path/to/model sample data file=data.json ...\n\n\n\n\n\n\nOpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.\nIn order to check if any OpenCL-enabled device and its runtime is already present use the clinfo tool. On Linux, clinfo can typically be installed with the default package manager (for example sudo apt-get install clinfo on Ubuntu). For Windows, pre-built clinfo binary can be found here.\nAlso use clinfo to verify successful installation of OpenCL runtimes.\n\n\n\nLinux:\nInstall the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:\nsudo apt update\nsudo apt install nvidia-driver-460 nvidia-cuda-toolkit\nReplace the driver version (460 in the above case) with the lastest number at the time of installation.\nWindows:\nInstall the NVIDIA GPU Driver and CUDA Toolkit.\n\n\n\n\n\nLinux:\nInstall Radeon Software for Linux available here.\nWindows:\nWe recommend installing the open source OCL-SDK.\n\n\n\n\nInstall the open source PoCL.\n\n\n\nFollow Intel’s install instructions given here (requires registration).\n\n\n\n\nIn order to enable the OpenCL backend the model must be compiled with the STAN_OPENCL makefile flag defined/non-empty. The flag can be supplied in the make call but we recommend writing the flag to the make/local file.\nAn example of the contents of make/local to enable parallelization with OpenCL:\nSTAN_OPENCL=true\nIf you are using OpenCL with an integrated GPU you also need to define the INTEGRATED_OPENCL flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:\nINTEGRATED_OPENCL=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_OPENCL can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the clinfo program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically 0. In that case you can also omit the OpenCL IDs as the default 0 IDs are used in that case.\nWe supply these IDs when starting the executable as shown below:\npath/to/model sample data file=data.json opencl platform=0 device=1",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Getting Started",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#chain-identifier-argument-id",
-    "href": "cmdstan-guide/command_line_options.html#chain-identifier-argument-id",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/parallelization.html#multi-threading-with-tbb",
+    "href": "cmdstan-guide/parallelization.html#multi-threading-with-tbb",
+    "title": "Parallelization",
     "section": "",
-    "text": "The chain identifier argument is used in conjunction with the random seed argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.\nThe syntax for the random seed argument is:\nid=&lt;int&gt;\nThe default value is 1.\nWhen running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.\nFor complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the Stan Reference Manual Reproducibility chapter for further details.",
+    "text": "In order to exploit multi-threading in a Stan model, the models must be rewritten to use the reduce_sum and map_rect functions. For instructions on how to rewrite Stan models to use these functions see Stan’s User guide chapter on parallelization, the reduce_sum case study or the Multithreading and Map-Reduce tutorial.\n\n\nOnce a model is rewritten to use the above-mentioned functions, the model must be compiled with the STAN_THREADS makefile flag. The flag can be supplied in the make call but we recommend writing the flag to the make/local file. If the STAN_THREADS flag is defined/non-empty, threads will be enabled.\nAn example of the contents of make/local to enable threading with TBB:\nSTAN_THREADS=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nBefore running a multi-threaded model, we need to specify the maximum number of threads the program can run (total threads for all chains). This is done by setting the num_threads argument. Valid values for num_threads are positive integers and -1. If num_threads is set to -1, all available cores will be used.\nGenerally, this number should not exceed the number of available cores for best performance.\nExample:\n./model sample data file=data.json num_threads=4 ...\nWhen the model is compiled with STAN_THREADS we can sample with multiple chains with a single executable (see section running multiple chains for cases when this is available). When running multiple chains num_threads is the maximum number of threads that can be used by all the chains combined. The exact number of threads that will be used for each chain at a given point in time is determined by the TBB scheduler. The following example start 2 chains with 8 total threads available:\n./model sample num_chains=2 data file=data.json num_threads=8 ...",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Getting Started",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#command-line-help",
-    "href": "cmdstan-guide/command_line_options.html#command-line-help",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/parallelization.html#multi-processing-with-mpi",
+    "href": "cmdstan-guide/parallelization.html#multi-processing-with-mpi",
+    "title": "Parallelization",
     "section": "",
-    "text": "CmdStan provides a help and help-all mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword help:\n./bernoulli help",
+    "text": "In order to use multi-processing with MPI in a Stan model, the models must be rewritten to use the map_rect function. By using MPI, the model can be parallelized across multiple cores or a cluster. MPI with Stan is supported on MacOS and Linux.\n\n\nCompiling and running Stan models with MPI requires that the system has an MPI implementation installed. For Unix systems the most commonly used implementations are MPICH and OpenMPI.\n\n\n\nOnce a model is rewritten to use map_rect, additional makefile flags must be written to the make/local. These are:\n\nSTAN_MPI: Enables the use of MPI with Stan if defined.\nCXX: The name of the MPI C++ compiler wrapper. Typically mpicxx.\nTBB_CXX_TYPE: The C++ compiler the MPI wrapper wraps. Typically gcc on Linux and clang on macOS.\n\nAn example of make/local on Linux:\nSTAN_MPI=true\nCXX=mpicxx\nTBB_CXX_TYPE=gcc\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_MPI is run using an MPI launcher. The MPI standard suggests using mpiexec, but a vendor wrapper for the launcher like mpirun can also be used. The launcher is supplied the path to the built executable and the number of processes to start: -n X for mpiexec or -np X for mpirun where X is replaced by the integer representing the number of processes.\nExample for running a model with six processes:\nmpiexec -n 6 path/to/model sample data file=data.json ...",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Getting Started",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "cmdstan-guide/command_line_options.html#error-messages-and-return-codes",
-    "href": "cmdstan-guide/command_line_options.html#error-messages-and-return-codes",
-    "title": "Command-Line Interface Overview",
+    "objectID": "cmdstan-guide/parallelization.html#opencl",
+    "href": "cmdstan-guide/parallelization.html#opencl",
+    "title": "Parallelization",
     "section": "",
-    "text": "CmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.\nAll program executables provide a return code between 0 and 255:\n\n0 - Program ran to termination as expected.\nvalue in range [1 : 125] - Method invoked could not run due to problems with model or data.\nvalue &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g. return code 139 results from termination signal 11 (segmentation violation).\n\nA non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.",
+    "text": "OpenCL is supported on most modern CPUs and GPUs. In order to run OpenCL-enabled Stan models, an OpenCL runtime for the target device must be installed. This subsection lists installation instructions for OpenCL runtimes of the commonly-found devices.\nIn order to check if any OpenCL-enabled device and its runtime is already present use the clinfo tool. On Linux, clinfo can typically be installed with the default package manager (for example sudo apt-get install clinfo on Ubuntu). For Windows, pre-built clinfo binary can be found here.\nAlso use clinfo to verify successful installation of OpenCL runtimes.\n\n\n\nLinux:\nInstall the NVIDIA GPU driver and the NVIDIA CUDA Toolkit. On Ubuntu the commands to install both is:\nsudo apt update\nsudo apt install nvidia-driver-460 nvidia-cuda-toolkit\nReplace the driver version (460 in the above case) with the lastest number at the time of installation.\nWindows:\nInstall the NVIDIA GPU Driver and CUDA Toolkit.\n\n\n\n\n\nLinux:\nInstall Radeon Software for Linux available here.\nWindows:\nWe recommend installing the open source OCL-SDK.\n\n\n\n\nInstall the open source PoCL.\n\n\n\nFollow Intel’s install instructions given here (requires registration).\n\n\n\n\nIn order to enable the OpenCL backend the model must be compiled with the STAN_OPENCL makefile flag defined/non-empty. The flag can be supplied in the make call but we recommend writing the flag to the make/local file.\nAn example of the contents of make/local to enable parallelization with OpenCL:\nSTAN_OPENCL=true\nIf you are using OpenCL with an integrated GPU you also need to define the INTEGRATED_OPENCL flag, as the sharing of memory between CPU and GPU is slightly different with integrated graphics:\nINTEGRATED_OPENCL=true\nThe model is then compiled as normal:\nmake path/to/model\n\n\n\nThe Stan model compiled with STAN_OPENCL can also be supplied the OpenCL platform and device IDs of the target device. These IDs determine the device on which to run the OpenCL-supported functions on. You can list the devices on your system using the clinfo program. If the system has one GPU and no OpenCL CPU runtime, the platform and device IDs of the GPU are typically 0. In that case you can also omit the OpenCL IDs as the default 0 IDs are used in that case.\nWe supply these IDs when starting the executable as shown below:\npath/to/model sample data file=data.json opencl platform=0 device=1",
     "crumbs": [
-      "Running CmdStan",
-      "Command-Line Interface Overview"
+      "Getting Started",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "index.html",
-    "href": "index.html",
-    "title": "Stan Documentation",
+    "objectID": "cmdstan-guide/mcmc_config.html",
+    "href": "cmdstan-guide/mcmc_config.html",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "This is the official documentation for Stan.\n\nThe Stan User’s Guide (pdf) provides example models and programming techniques for coding statistical models in Stan.\nThe Stan Reference Manual (pdf) specifies the Stan programming language and inference algorithms.\nThe Stan Functions Reference (pdf) specifies the functions built into the Stan programming language.\n\nThere are also separate installation and getting started guides for CmdStan (pdf), the command-line interface to the Stan inference engine, and the R, Python, and Julia interfaces.\n\nOlder Versions\nThis documentation is for Stan 2.38. Older versions of each of the documents linked above can be found in the table below:\n\n\n\nVersion\nStan Reference Manual\nStan Users Guide\nCmdStan Guide\nStan Functions Reference\n\n\n\n\n2.38\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.37\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.36\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.35\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.34\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.33\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.32\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.31\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.30\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.29\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.28\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.27\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.26\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.25\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.24\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.23\nhtml pdf\nhtml pdf\nhtml\nhtml pdf\n\n\n2.22\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.21\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.20\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.19\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.18\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n\nPrior to version 2.18, all documentation was part of a single document called the Stan User’s Guide and Reference Manual. These versions are still available for download as PDFs:\n\n\n\nVersion\nStan User’s Guide and Reference Manual\n\n\n\n\n2.17\npdf\n\n\n2.16\npdf\n\n\n2.15\npdf\n\n\n2.14\npdf\n\n\n2.13\npdf\n\n\n2.12\npdf\n\n\n2.11\npdf\n\n\n2.10\npdf\n\n\n2.9\npdf\n\n\n2.8\npdf\n\n\n2.7\npdf\n\n\n2.6\npdf\n\n\n2.5\npdf\n\n\n2.4\npdf\n\n\n2.3\npdf\n\n\n2.2\npdf\n\n\n2.1\npdf\n\n\n2.0\npdf\n\n\n1.3\npdf\n\n\n1.2\npdf\n\n\n1.1\npdf\n\n\n1.0\npdf\n\n\n\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n\n Back to top"
+    "text": "The sample method provides Bayesian inference over the model conditioned on data using Hamiltonian Monte Carlo (HMC) sampling. By default, the inference engine used is the No-U-Turn sampler (NUTS), an adaptive form of Hamiltonian Monte Carlo sampling. For details on HMC and NUTS, see the Stan Reference Manual chapter on MCMC Sampling.\n\n\nTo generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument sample or method=sample together with the input data. The executable can be run from any directory.\nThe full set of configuration options available for the sample method is available by using the sample help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.\nHere, we run it in the directory which contains the Stan program and input data, &lt;cmdstan-home&gt;/examples/bernoulli:\n&gt; cd examples/bernoulli\n&gt; ls\n  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan\nTo execute sampling of the model under Linux or Mac, use:\n&gt; ./bernoulli sample data file=bernoulli.data.json\nIn Windows, the ./ prefix is not needed:\n&gt; bernoulli.exe sample data file=bernoulli.data.json\nThe output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:\nmethod = sample (Default)\n  sample\n    num_samples = 1000 (Default)\n    num_warmup = 1000 (Default)\n    save_warmup = false (Default)\n    thin = 1 (Default)\n    adapt\n      engaged = true (Default)\n      gamma = 0.050000000000000003 (Default)\n      delta = 0.80000000000000004 (Default)\n      kappa = 0.75 (Default)\n      t0 = 10 (Default)\n      init_buffer = 75 (Default)\n      term_buffer = 50 (Default)\n      window = 25 (Default)\n      save_metric = false (Default)\n    algorithm = hmc (Default)\n      hmc\n        engine = nuts (Default)\n          nuts\n            max_depth = 10 (Default)\n        metric = diag_e (Default)\n        metric_file =  (Default)\n        stepsize = 1 (Default)\n        stepsize_jitter = 0 (Default)\n    num_chains = 1 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 3252652196 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\nAfter the configuration has been displayed, a short timing message is given.\nGradient evaluation took 1.2e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.\nAdjust your expectations accordingly!\nNext, the sampler reports the iteration number, reporting the percentage complete.\nIteration:    1 / 2000 [  0%]  (Warmup)\n...\nIteration: 2000 / 2000 [100%]  (Sampling)\nFinally, the sampler reports timing information:\n Elapsed Time: 0.007 seconds (Warm-up)\n               0.017 seconds (Sampling)\n               0.024 seconds (Total)\n\n\n\nEach execution of the model results in draws from a single Markov chain being written to a file in comma-separated value (CSV) format. The default name of the output file is output.csv.\nThe first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (#)).\nWhen the example model bernoulli.stan is run via the command line with all default arguments, the following configuration is displayed:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 1000 (Default)\n#     num_warmup = 1000 (Default)\n#     save_warmup = false (Default)\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000 (Default)\n#       delta = 0.800000 (Default)\n#       kappa = 0.750000 (Default)\n#       t0 = 10.000000 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#       save_metric = false (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1.000000 (Default)\n#         stepsize_jitter = 0.000000 (Default)\n#     num_chains = 1 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\nThis is followed by a CSV header indicating the names of the values sampled.\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nThe first output columns report the HMC sampler information:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause the above header is from the NUTS sampler, it has columns treedepth__, n_leapfrog__, and divergent__ and doesn’t have column int_time__. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, theta.\nThe header line is written to the output file before warmup begins. If option save_warmup is set to true, the warmup draws are output directly after the header. The total number of warmup draws saved is num_warmup divided by thin, rounded up (i.e., ceiling).\nFollowing the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:\n# Adaptation terminated\n# Step size = 0.884484\n# Diagonal elements of inverse mass matrix:\n# 0.535006\nThe default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter theta.\nDraws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.\n-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853\n-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295\n-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299\n-6.88712,1,0.884484,1,1,0,7.02101,0.188229\n-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596\n...\nThe output ends with timing details:\n#  Elapsed Time: 0.007 seconds (Warm-up)\n#                0.017 seconds (Sampling)\n#                0.024 seconds (Total)\n\n\n\nAt every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters metric and stepsize in order to efficiently sample from typical set, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.\nThe following keyword-value arguments control the total number of iterations:\n\nnum_samples\nnum_warmup\nsave_warmup\nthin\n\nThe values for arguments num_samples and num_warmup must be a non-negative integer. The default value for both is \\(1000\\).\nFor well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.\nThe number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is \\(\\sqrt{N}\\); therefore every additional decimal place of accuracy increases this by a factor of 10.\nArgument save_warmup takes values false or true. The default value is false, i.e., warmup draws are not saved to the output file. When the value is true, the warmup draws are written to the CSV output file directly after the CSV header line.\nArgument thin controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce anticorrelated draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.\nThe value of argument thin must be a positive integer. When thin is set to value \\(N\\), every \\(N^{th}\\) iteration is written to the output file. Should the value of thin exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of thin equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When num_sampling=M and thin=N, the number of iterations written to the output CSV file will be ceiling(M/N). If save_warmup=true, thinning is applied to the warmup iterations as well.\n\n\n\nThe adapt keyword is used to specify non-default options for the sampler adaptation schedule and settings.\nAdaptation can be turned off by setting sub-argument engaged to value false. If engaged=false, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is engaged=1, this keyword-value pair can be omitted from the command.\nThere are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section Automatic Parameter Tuning.\nThe boolean sub-argument save_metric was added in Stan version 2.34. When save_metric=true, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix _metric.json, e.g., if using the default output filename output.csv, the saved metric file will be output_metric.json. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument metric_file.\n\n\nThe Stan User’s Guide section on model conditioning and curvature provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on HMC algorithm parameters explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.\nThe following keyword-value arguments control the settings used to optimize the step size:\n\ndelta - The target Metropolis acceptance rate. The default value is \\(0.8\\). Its value must be strictly between \\(0\\) and \\(1\\). Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of delta will also allow some models that would otherwise get stuck to overcome their blockages.  Models with difficult posterior geometries may required increasing the delta argument closer to \\(1\\); we recommend first trying to raise it to \\(0.9\\) or at most \\(0.95\\). Values about \\(0.95\\) are strong indication of bad geometry; the better solution is to change the model geometry through reparameterization which could yield both more efficient and faster sampling.\ngamma - Adaptation regularization scale. Must be a positive real number, default value is \\(0.05\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nkappa - Adaptation relaxation exponent. Must be a positive real number, default value is \\(0.75\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nt_0 - Adaptation iteration offset. Must be a positive real number, default value is \\(10\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\n\n\n\n\nWhen adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:\n\ninit_buffer - The number of iterations spent tuning the step size at the outset of adaptation.\nwindow - The initial number of iterations devoted to tune the metric, will be doubled successively.\nterm_buffer - The number of iterations used to re-tune the step size once the metric has been tuned.\n\nThe specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.\nThe following figure is taken from the Stan Reference Manual, where label “I” correspond to init_buffer, the initial “II” corresponds to window, and the final “III” corresponds to term_buffer:\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.\n\n\n\n\n\nThe algorithm keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: hmc, which generates from an HMC-driven Markov chain; and fixed_param which generates a new sample without changing the state of the Markov chain. The default argument is algorithm=hmc.\n\n\nIf a model doesn’t specify any parameters, then argument algorithm=fixed_param is mandatory.\nThe fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.\n\n\n\nAll HMC algorithms have three parameters:\n\nstep size\nmetric\nintegration time - the number of steps taken along the Hamiltonian trajectory\n\nSee the Stan Reference Manual section on HMC algorithm parameters for further details.\n\n\nThe HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.\nThe following keyword-value arguments control the step size:\n\nstepsize - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is \\(1\\).\nstepsize_jitter - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between \\(0\\) and \\(1\\). The default value is \\(0\\). Setting stepsize_jitter to \\(1\\) causes step sizes to be selected in the range of \\(0\\) to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.\n\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nThe metric argument specifies the choice of Euclidean HMC implementations:\n\nmetric=unit specifies unit metric (diagonal matrix of ones).\nmetric=diag_e specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.\nmetric=dense_e specifies a dense metric (a dense, symmetric positive definite matrix).\n\nBy default, the metric is estimated during warmup. However, when metric=diag_e or metric=dense_e, an initial guess for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a single variable inv_metric. For a diag_e metric the inv_metric value must be a vector of positive values, one for each parameter in the system. For a dense_e metric, inv_metric value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.\nThe metric_file option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n\n\n\nThe total integration time is determined by the argument engine which take possible values:\n\nnuts - the No-U-Turn Sampler which dynamically determines the optimal integration time.\nstatic - an HMC sampler which uses a user-specified integration time.\n\nThe default argument is engine=nuts.\nThe NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.\nWhen engine=nuts, the subargument max_depth can be used to control the depth of the tree. The default argument is max_depth=10. In the case where a model has a difficult posterior from which to sample, max_depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\nWhen the argument engine=static is specified, the user must specify the integration time via keyword int_time which takes as a value a positive number. The default value is \\(2\\pi\\).\n\n\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.\n\n\n\nA Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.\nThe preferred way of using multiple chains is to run them all from the same executable using the num_chains argument. There is also the option to use the Unix or DOS shell to run multiple executables.\n\n\nThe num_chains argument can be used for all of Stan’s samplers with the exception of the static HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with STAN_THREADS=true, these will either run in parallel or one after the other.\nWhen num_chains is greather than 1 (the default), arguments related to filenames (e.g. output file=, init=) can accept a comma separated list of values, one per each chain.\nFor example, sample will specify the names of the three chain’s output files.\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv\nThis will write the output in output_1.csv, output_2.csv, output_3.csv.\nIf the model was not compiled with STAN_THREADS=true, the above command will run 3 chains sequentially.\nIf the model was compiled with STAN_THREADS=true, the chains can run in parallel, with the num_threads argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (map_rect or reduce_sum calls), the below command will run 3 chains in parallel, provided there are cores available:\n./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4\nIf the model uses within-chain parallelization (map_rect or reduce_sum calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the Threading Building Blocks scheduler.\n./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16\n\n\nIf a comma separated list is not used, the num_chains argument changes the normal meanings of filename arguments when it is greater than 1 (the default). They are now interpreted as a “template” which is used for each chain.\nFor example, when num_chains=2, the argument output file=foo.csv no longer produces a file foo.csv, but instead produces two files, foo_1.csv and foo_2.csv. If you also supply id=5, the files produced will be foo_5.csv and foo_6.csv – id=5 gives the id of the first chain, and the remaining chains are sequential from there.\nThis also applies to input files, like those used for initialization. For example, if num_chains=3 and init=bar.json will first look for bar_1.json. If it exists, it will use bar_1.json for the first chain, bar_2.json for the second, and so on. If bar_1.json does not exist, it falls back to looking for bar.json, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the id argument, which defaults to 1.\nFor example, this shorthand is equivalent to the example given above:\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv\nA suffix with the chain id is appended to the provided output filename (output.csv in the above command), so this will also produce files output_1.csv, output_2.csv, output_3.csv.\n\n\n\n\n\nThe stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\nTo run stansummary on the output files generated by the for loop above, by the above run of the bernoulli model on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary output_*.csv\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe output_*.csv\nThe stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable theta:\nInference for Stan model: bernoulli_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total\nSampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total\n\n                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0\naccept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0\nstepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan\ntreedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0\nn_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0\ndivergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan\nenergy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0\n\ntheta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nIn this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for theta reflect the uncertainty in our estimate, due to the small amount of data, given the prior of beta(1, 1)\n\n\n\nNote: Many of these examples can be simplified by using the num_chains argument.\nWhen the num_chains argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.\nTo run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell for loop to set up index variables needed to identify each chain and its outputs.\nOn MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:\nfor NAME [in LIST]; do COMMANDS; done\nThe list can be a simple sequence of numbers, or you can use the shell expansion syntax {1..N} which expands to the sequence from \\(1\\) to \\(N\\), e.g. {1..4} expands to 1 2 3 4. Note that the expression {1..N} cannot contain spaces.\nTo run 4 chains for the example bernoulli model on MacOS or Linux:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv\n    done\nThe backslash (\\) indicates a line continuation in Unix. The expression ${i} substitutes in the value of loop index variable i. To run chains in parallel, put an ampersand (&) at the end of the nested sampler command:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv &\n    done\nThis pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.\nOn Windows, the DOS for-loop syntax is one of:\nfor %i in (SET) do COMMAND COMMAND-ARGUMENTS\nfor /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS\nTo run 4 chains in parallel on Windows:\n&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^\n                                    data file=bernoulli.data.json my_data ^\n                                    output file=output_%i.csv\nThe caret (^) indicates a line continuation in DOS. The expression %i is the loop index.\nIn the following extended examples, we focus on just the nested sampler command for Unix.\n\n\nFor reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.\nThe RNG seed is specified by random seed=&lt;int&gt; and the offset is specified by id=&lt;loop index&gt;, so the call to the sampler is:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            random seed=12345 id=${i}\n\n\n\nThe warmup and sampling iteration keyword-value arguments must follow the sample keyword. The call to the sampler which overrides the default warmup and sampling iterations is:\n./my_model sample num_warmup=500 num_sampling=500 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nTo save warmup draws as part of the Stan CSV output file, use the keyword-value argument save_warmup=true. This must be grouped with the other sample keyword sub-arguments.\n./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nBy default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range \\([{-2}, 2]\\). To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument init=&lt;filepath&gt;:\n./my_model sample init=my_param_inits.json data file=my_model.data.json \\\n           output file=output_${i}.csv\nTo verify that the specified values will be used by the sampler, you can run the sampler with option algorithm=fixed_param, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting num_warmp=0 and num_samples=1 saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.\nFor example, if we run the example Bernoulli model with specified initial value for parameter “theta”:\n{ \"theta\" : 0.5 }\nvia command:\n./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \\\n            init=bernoulli.init.json data file=bernoulli.data.json\nThe resulting output CSV file contains a single draw:\nlp__,accept_stat__,theta\n0,0,0.5\n#\n#  Elapsed Time: 0 seconds (Warm-up)\n#                0 seconds (Sampling)\n#                0 seconds (Total)\n#\n\n\n\nAn initial estimate for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a variable inv_metric. The metric_file option can be used with and without adaptation enabled.\nBy default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file bernoulli.diag_e.json, contents\n{ \"inv_metric\" : [0.296291] }\ncan be used as the initial metric as follows:\n../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nThe resulting output CSV file will contain the following set of comment lines:\n# Adaptation terminated\n# Step size = 0.9\n# Diagonal elements of inverse mass matrix:\n# 0.296291\nAs of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument save_metric, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the stepsize argument as well as the metric_file arguments; the former is the value of the stepsize element in the saved metric file, and the later is the metric file path.\n\n\n\nThe keyword-value arguments for these settings are grouped together under the adapt keyword which itself is a sub-argument of the sample keyword.\nModels with difficult posterior geometries may required increasing the delta argument closer to \\(1\\).\n./my_model sample adapt delta=0.95 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nTo skip adaptation altogether, use the keyword-value argument engaged=false. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nEven with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the typical set.\nTo skip warmup altogether requires specifying both num_warmup=0 and adapt engaged=false.\n../my_model sample num_warmup=0 adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nModels with difficult posterior geometries may required increasing the max_depth argument from its default value \\(10\\). This requires specifying a series of keyword-argument pairs:\n./my_model sample adapt delta=0.95 \\\n            algorithm=hmc engine=nuts max_depth=15 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; write the sampler parameters and gradients of all model parameters for each draw to a CSV file:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            diagnostic_file=diagnostics_${i}.csv\n\n\n\nThe output keyword sub-argument refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is \\(100\\) iterations. The progress updates look like:\nIteration:    1 / 2000 [  0%]  (Warmup)\nIteration:  100 / 2000 [  5%]  (Warmup)\nIteration:  200 / 2000 [ 10%]  (Warmup)\nIteration:  300 / 2000 [ 15%]  (Warmup)\nFor simple models which fit quickly, such updates can be annoying; to suppress them altogether, set refresh=0. This only turns off the Iteration: messages; the configuration and timing information are still written to the terminal.\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            refresh=0\nFor complicated models which take a long time to fit, setting the refresh rate to a low number, e.g. \\(10\\) or even \\(1\\), provides a way to more closely monitor the sampler.\n\n\n\nThe CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:\n./bernoulli sample data file=bernoulli.data.json adapt delta=0.95\nadapt is either mistyped or misplaced.\nPerhaps you meant one of the following valid configurations?\n  method=sample sample adapt\n  method=variational variational adapt\nFailed to parse arguments, terminating Stan\nThe following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.\n./my_model sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100 \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           random seed=12345 id=${i}\nThe keywords sample, data, output, and random are the top-level argument groups. Within the sample config arguments, the keyword adapt groups the adaptation algorithm parameters and the keyword-value algorithm=hmc groups the NUTS-HMC parameters.\nThe top-level groups can be freely ordered with respect to one another. The following is also a valid command:\n./my_model random seed=12345 id=${i} \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
   },
   {
-    "objectID": "cmdstan-guide/bib.html",
-    "href": "cmdstan-guide/bib.html",
-    "title": "References",
+    "objectID": "cmdstan-guide/mcmc_config.html#running-the-sampler",
+    "href": "cmdstan-guide/mcmc_config.html#running-the-sampler",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "References\n\n\n\n\n Back to top"
+    "text": "To generate a sample from the posterior distribution of the model conditioned on the data, we run the executable program with the argument sample or method=sample together with the input data. The executable can be run from any directory.\nThe full set of configuration options available for the sample method is available by using the sample help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the sampler console output and in the output CSV file’s comments.\nHere, we run it in the directory which contains the Stan program and input data, &lt;cmdstan-home&gt;/examples/bernoulli:\n&gt; cd examples/bernoulli\n&gt; ls\n  bernoulli  bernoulli.data.json  bernoulli.data.R  bernoulli.stan\nTo execute sampling of the model under Linux or Mac, use:\n&gt; ./bernoulli sample data file=bernoulli.data.json\nIn Windows, the ./ prefix is not needed:\n&gt; bernoulli.exe sample data file=bernoulli.data.json\nThe output is the same across all supported platforms. First, the configuration of the program is echoed to the standard output:\nmethod = sample (Default)\n  sample\n    num_samples = 1000 (Default)\n    num_warmup = 1000 (Default)\n    save_warmup = false (Default)\n    thin = 1 (Default)\n    adapt\n      engaged = true (Default)\n      gamma = 0.050000000000000003 (Default)\n      delta = 0.80000000000000004 (Default)\n      kappa = 0.75 (Default)\n      t0 = 10 (Default)\n      init_buffer = 75 (Default)\n      term_buffer = 50 (Default)\n      window = 25 (Default)\n      save_metric = false (Default)\n    algorithm = hmc (Default)\n      hmc\n        engine = nuts (Default)\n          nuts\n            max_depth = 10 (Default)\n        metric = diag_e (Default)\n        metric_file =  (Default)\n        stepsize = 1 (Default)\n        stepsize_jitter = 0 (Default)\n    num_chains = 1 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 3252652196 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\nAfter the configuration has been displayed, a short timing message is given.\nGradient evaluation took 1.2e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.12 seconds.\nAdjust your expectations accordingly!\nNext, the sampler reports the iteration number, reporting the percentage complete.\nIteration:    1 / 2000 [  0%]  (Warmup)\n...\nIteration: 2000 / 2000 [100%]  (Sampling)\nFinally, the sampler reports timing information:\n Elapsed Time: 0.007 seconds (Warm-up)\n               0.017 seconds (Sampling)\n               0.024 seconds (Total)",
+    "crumbs": [
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
+    ]
   },
   {
-    "objectID": "cmdstan-guide/compiling_stan_programs.html",
-    "href": "cmdstan-guide/compiling_stan_programs.html",
-    "title": "Compiling a Stan Program",
+    "objectID": "cmdstan-guide/mcmc_config.html#mcmc_output_csv",
+    "href": "cmdstan-guide/mcmc_config.html#mcmc_output_csv",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "A Stan program must be in a file with extension .stan. The CmdStan makefile rules specify all necessary steps to translate files with suffix .stan to a CmdStan executable program. This is a two-stage process:\n\nfirst the Stan program is translated to C++ by the stanc compiler\nthen the C++ compiler compiles all C++ sources and links them together with the CmdStan interface program and the Stan and math libraries.\n\n\n\nTo compile Stan programs, you must invoke the Make program from the &lt;cmdstan-home&gt; directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.\n&gt; cd &lt;cmdstan_home&gt;\nIn the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:\n&gt; make examples/bernoulli/bernoulli\nOn Windows, the command is the same with the addition of .exe at the end of the target (note the use of forward slashes):\n&gt; make examples/bernoulli/bernoulli.exe\nThe generated C++ code (bernoulli.hpp), object file (bernoulli.o) and the compiled executable will be placed in the same directory as the Stan program.\nThe compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.\n\n\n\nWhen executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target bernoulli twice in a row, without any editing bernoulli.stan or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:\n&gt; make examples/bernoulli/bernoulli\nmake: `examples/bernoulli/bernoulli' is up to date.\nIf the file containing the Stan program is updated, the next call to make will rebuild the CmdStan executable.\n\n\n\nThe Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.\nThe simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the bernoulli.stan program, we introduce a typo on line \\(9\\) by writing thata instead of theta, the Make command fails with the following\n--- Translating Stan model to C++ code ---\nbin/stanc  --o=bernoulli.hpp bernoulli.stan\n\nSemantic error in 'bernoulli.stan', line 9, column 2 to column 7:\n   -------------------------------------------------\n     7:  }\n     8:  model {\n     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1\n           ^\n    10:    y ~ bernoulli(theta);\n    11:  }\n   -------------------------------------------------\n\nIdentifier 'thata' not in scope.\n\nmake: *** [bernoulli.hpp] Error 1\nStan is a strongly-typed language; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program foo.stan contains an illegal assignment statement:\ndata {\n  real x;\n}\ntransformed data {\n  int y = x;\n}\nThe Make command fails with the following:\nSemantic error in 'foo.stan', line 5, column 2 to column 12:\n   -------------------------------------------------\n     3:  }\n     4:  transformed data {\n     5:    int y = x;\n           ^\n     6:  }\n   -------------------------------------------------\n\nIll-typed arguments supplied to assignment operator =:\n lhs has type int and rhs has type real\nThe Stan Reference Manual provides a complete specification of the Stan programming language. The Stan User’s Guide also contains a full description of the errors and warnings stanc can emit.\n\n\n\nIf the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.\nThe Make command prints the following message to the terminal at the point when it compiles and links the C++ file:\n--- Compiling, linking C++ code ---\nIf the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.\nIf this happens, please report the error, together with the Stan program on either the Stan Forums or on the Stan compiler GitHub issues tracker.\n\n\n\nUsers can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.\nThe CXXFLAGS and LDFLAGS makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the make/local file.\nFor example:\nCXXFLAGS = -O2\nA recommend a set of CXXFLAGS and LDFLAGS flags can be turned on by setting STAN_CPP_OPTIMS=true in the make/local file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.\n\n\nWhen assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.\nFor some models these checks can represent a significant part of the models execution time. By setting the STAN_NO_RANGE_CHECKS=true makefile flag in the make/local file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.",
+    "text": "Each execution of the model results in draws from a single Markov chain being written to a file in comma-separated value (CSV) format. The default name of the output file is output.csv.\nThe first part of the output file records the version of the underlying Stan library and the configuration as comments (i.e., lines beginning with the pound sign (#)).\nWhen the example model bernoulli.stan is run via the command line with all default arguments, the following configuration is displayed:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = sample (Default)\n#   sample\n#     num_samples = 1000 (Default)\n#     num_warmup = 1000 (Default)\n#     save_warmup = false (Default)\n#     thin = 1 (Default)\n#     adapt\n#       engaged = 1 (Default)\n#       gamma = 0.050000 (Default)\n#       delta = 0.800000 (Default)\n#       kappa = 0.750000 (Default)\n#       t0 = 10.000000 (Default)\n#       init_buffer = 75 (Default)\n#       term_buffer = 50 (Default)\n#       window = 25 (Default)\n#       save_metric = false (Default)\n#     algorithm = hmc (Default)\n#       hmc\n#         engine = nuts (Default)\n#           nuts\n#             max_depth = 10 (Default)\n#         metric = diag_e (Default)\n#         metric_file =  (Default)\n#         stepsize = 1.000000 (Default)\n#         stepsize_jitter = 0.000000 (Default)\n#     num_chains = 1 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\nThis is followed by a CSV header indicating the names of the values sampled.\nlp__,accept_stat__,stepsize__,treedepth__,n_leapfrog__,divergent__,energy__,theta\nThe first output columns report the HMC sampler information:\n\nlp__ - the total log probability density (up to an additive constant) at each sample\naccept_stat__ - the average Metropolis acceptance probability over each simulated Hamiltonian trajectory\nstepsize__ - integrator step size\ntreedepth__ - depth of tree used by NUTS (NUTS sampler)\nn_leapfrog__ - number of leapfrog calculations (NUTS sampler)\ndivergent__ - has value 1 if trajectory diverged, otherwise 0. (NUTS sampler)\nenergy__ - value of the Hamiltonian\nint_time__ - total integration time (static HMC sampler)\n\nBecause the above header is from the NUTS sampler, it has columns treedepth__, n_leapfrog__, and divergent__ and doesn’t have column int_time__. The remaining columns correspond to model parameters. For the Bernoulli model, it is just the final column, theta.\nThe header line is written to the output file before warmup begins. If option save_warmup is set to true, the warmup draws are output directly after the header. The total number of warmup draws saved is num_warmup divided by thin, rounded up (i.e., ceiling).\nFollowing the warmup draws (if any), are comments which record the results of adaptation: the stepsize, and inverse mass metric used during sampling:\n# Adaptation terminated\n# Step size = 0.884484\n# Diagonal elements of inverse mass matrix:\n# 0.535006\nThe default sampler is NUTS with an adapted step size and a diagonal inverse mass matrix. For this example, the step size is 0.884484, and the inverse mass contains the single entry 0.535006 corresponding to the parameter theta.\nDraws from the posterior distribution are printed out next, each line containing a single draw with the columns corresponding to the header.\n-6.84097,0.974135,0.884484,1,3,0,6.89299,0.198853\n-6.91767,0.985167,0.884484,1,1,0,6.92236,0.182295\n-7.04879,0.976609,0.884484,1,1,0,7.05641,0.162299\n-6.88712,1,0.884484,1,1,0,7.02101,0.188229\n-7.22917,0.899446,0.884484,1,3,0,7.73663,0.383596\n...\nThe output ends with timing details:\n#  Elapsed Time: 0.007 seconds (Warm-up)\n#                0.017 seconds (Sampling)\n#                0.024 seconds (Total)",
     "crumbs": [
-      "Getting Started",
-      "Compiling a Stan Program"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/compiling_stan_programs.html#invoking-the-make-utility",
-    "href": "cmdstan-guide/compiling_stan_programs.html#invoking-the-make-utility",
-    "title": "Compiling a Stan Program",
+    "objectID": "cmdstan-guide/mcmc_config.html#iterations",
+    "href": "cmdstan-guide/mcmc_config.html#iterations",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "To compile Stan programs, you must invoke the Make program from the &lt;cmdstan-home&gt; directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.\n&gt; cd &lt;cmdstan_home&gt;\nIn the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:\n&gt; make examples/bernoulli/bernoulli\nOn Windows, the command is the same with the addition of .exe at the end of the target (note the use of forward slashes):\n&gt; make examples/bernoulli/bernoulli.exe\nThe generated C++ code (bernoulli.hpp), object file (bernoulli.o) and the compiled executable will be placed in the same directory as the Stan program.\nThe compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.",
+    "text": "At every sampler iteration, the sampler returns a set of estimates for all parameters and quantities of interest in the model. During warmup, the NUTS algorithm adjusts the HMC algorithm parameters metric and stepsize in order to efficiently sample from typical set, the neighborhood substantial posterior probability mass through which the Markov chain will travel in equilibrium. After warmup, the fixed metric and stepsize are used to produce a set of draws.\nThe following keyword-value arguments control the total number of iterations:\n\nnum_samples\nnum_warmup\nsave_warmup\nthin\n\nThe values for arguments num_samples and num_warmup must be a non-negative integer. The default value for both is \\(1000\\).\nFor well-specified models and data, the sampler may converge faster and this many warmup iterations may be overkill. Conversely, complex models which have difficult posterior geometries may require more warmup iterations in order to arrive at good values for the step size and metric.\nThe number of sampling iterations to runs depends on the effective sample size (EFF) reported for each parameter and the desired precision of your estimates. An EFF of at least 100 is required to make a viable estimate. The precision of your estimate is \\(\\sqrt{N}\\); therefore every additional decimal place of accuracy increases this by a factor of 10.\nArgument save_warmup takes values false or true. The default value is false, i.e., warmup draws are not saved to the output file. When the value is true, the warmup draws are written to the CSV output file directly after the CSV header line.\nArgument thin controls the number of draws from the posterior written to the output file. Some users familiar with older approaches to MCMC sampling might be used to thinning to eliminate an expected autocorrelation in the draws. HMC is not nearly as susceptible to this autocorrelation problem and thus thinning is generally not required nor advised, as HMC can produce anticorrelated draws, which increase the effective sample size beyond the number of draws from the posterior. Thinning should only be used in circumstances where storage of the draws is limited and/or RAM for later processing the draws is limited.\nThe value of argument thin must be a positive integer. When thin is set to value \\(N\\), every \\(N^{th}\\) iteration is written to the output file. Should the value of thin exceed the specified number of iterations, the first iteration is saved to the output. This is because the iteration counter starts from zero and whenever the counter modulo the value of thin equals zero, the iteration is saved to the output file. Since zero modulo any positive integer is zero, the first iteration is always saved. When num_sampling=M and thin=N, the number of iterations written to the output CSV file will be ceiling(M/N). If save_warmup=true, thinning is applied to the warmup iterations as well.",
     "crumbs": [
-      "Getting Started",
-      "Compiling a Stan Program"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/compiling_stan_programs.html#dependencies",
-    "href": "cmdstan-guide/compiling_stan_programs.html#dependencies",
-    "title": "Compiling a Stan Program",
+    "objectID": "cmdstan-guide/mcmc_config.html#adaptation",
+    "href": "cmdstan-guide/mcmc_config.html#adaptation",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "When executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target bernoulli twice in a row, without any editing bernoulli.stan or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:\n&gt; make examples/bernoulli/bernoulli\nmake: `examples/bernoulli/bernoulli' is up to date.\nIf the file containing the Stan program is updated, the next call to make will rebuild the CmdStan executable.",
+    "text": "The adapt keyword is used to specify non-default options for the sampler adaptation schedule and settings.\nAdaptation can be turned off by setting sub-argument engaged to value false. If engaged=false, no adaptation will be done, and all other adaptation sub-arguments will be ignored. Since the default argument is engaged=1, this keyword-value pair can be omitted from the command.\nThere are two sets of adaptation sub-arguments: step size optimization parameters and the warmup schedule. These are described in detail in the Reference Manual section Automatic Parameter Tuning.\nThe boolean sub-argument save_metric was added in Stan version 2.34. When save_metric=true, the adapted stepsize and metric are output as JSON at the end of adaptation. The saved metric file name is the output file basename with the suffix _metric.json, e.g., if using the default output filename output.csv, the saved metric file will be output_metric.json. This metric file can be reused in subsequent sampler runs as the initial metric, via sampler argument metric_file.\n\n\nThe Stan User’s Guide section on model conditioning and curvature provides a discussion of adaptation and stepsize issues. The Stan Reference Manual section on HMC algorithm parameters explains the NUTS-HMC adaptation schedule and the tuning parameters for setting the step size.\nThe following keyword-value arguments control the settings used to optimize the step size:\n\ndelta - The target Metropolis acceptance rate. The default value is \\(0.8\\). Its value must be strictly between \\(0\\) and \\(1\\). Increasing the default value forces the algorithm to use smaller step sizes. This can improve sampling efficiency (effective sample size per iteration) at the cost of increased iteration times. Raising the value of delta will also allow some models that would otherwise get stuck to overcome their blockages.  Models with difficult posterior geometries may required increasing the delta argument closer to \\(1\\); we recommend first trying to raise it to \\(0.9\\) or at most \\(0.95\\). Values about \\(0.95\\) are strong indication of bad geometry; the better solution is to change the model geometry through reparameterization which could yield both more efficient and faster sampling.\ngamma - Adaptation regularization scale. Must be a positive real number, default value is \\(0.05\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nkappa - Adaptation relaxation exponent. Must be a positive real number, default value is \\(0.75\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\nt_0 - Adaptation iteration offset. Must be a positive real number, default value is \\(10\\). This is a parameter of the Nesterov dual-averaging algorithm. We recommend always using the default value.\n\n\n\n\nWhen adaptation is engaged, the warmup schedule is specified by sub-arguments, all of which take positive integers as values:\n\ninit_buffer - The number of iterations spent tuning the step size at the outset of adaptation.\nwindow - The initial number of iterations devoted to tune the metric, will be doubled successively.\nterm_buffer - The number of iterations used to re-tune the step size once the metric has been tuned.\n\nThe specified values may be modified slightly in order to ensure alignment between the warmup schedule and total number of warmup iterations.\nThe following figure is taken from the Stan Reference Manual, where label “I” correspond to init_buffer, the initial “II” corresponds to window, and the final “III” corresponds to term_buffer:\nWarmup Epochs Figure.  Adaptation during warmup occurs in three stages: an initial fast adaptation interval (I), a series of expanding slow adaptation intervals (II), and a final fast adaptation interval (III). For HMC, both the fast and slow intervals are used for adapting the step size, while the slow intervals are used for learning the (co)variance necessitated by the metric. Iteration numbering starts at 1 on the left side of the figure and increases to the right.",
     "crumbs": [
-      "Getting Started",
-      "Compiling a Stan Program"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/compiling_stan_programs.html#compiler-errors",
-    "href": "cmdstan-guide/compiling_stan_programs.html#compiler-errors",
-    "title": "Compiling a Stan Program",
+    "objectID": "cmdstan-guide/mcmc_config.html#algorithm",
+    "href": "cmdstan-guide/mcmc_config.html#algorithm",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "The Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.\nThe simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the bernoulli.stan program, we introduce a typo on line \\(9\\) by writing thata instead of theta, the Make command fails with the following\n--- Translating Stan model to C++ code ---\nbin/stanc  --o=bernoulli.hpp bernoulli.stan\n\nSemantic error in 'bernoulli.stan', line 9, column 2 to column 7:\n   -------------------------------------------------\n     7:  }\n     8:  model {\n     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1\n           ^\n    10:    y ~ bernoulli(theta);\n    11:  }\n   -------------------------------------------------\n\nIdentifier 'thata' not in scope.\n\nmake: *** [bernoulli.hpp] Error 1\nStan is a strongly-typed language; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program foo.stan contains an illegal assignment statement:\ndata {\n  real x;\n}\ntransformed data {\n  int y = x;\n}\nThe Make command fails with the following:\nSemantic error in 'foo.stan', line 5, column 2 to column 12:\n   -------------------------------------------------\n     3:  }\n     4:  transformed data {\n     5:    int y = x;\n           ^\n     6:  }\n   -------------------------------------------------\n\nIll-typed arguments supplied to assignment operator =:\n lhs has type int and rhs has type real\nThe Stan Reference Manual provides a complete specification of the Stan programming language. The Stan User’s Guide also contains a full description of the errors and warnings stanc can emit.",
+    "text": "The algorithm keyword-value pair specifies the algorithm used to generate the sample. There are two possible values: hmc, which generates from an HMC-driven Markov chain; and fixed_param which generates a new sample without changing the state of the Markov chain. The default argument is algorithm=hmc.\n\n\nIf a model doesn’t specify any parameters, then argument algorithm=fixed_param is mandatory.\nThe fixed parameter sampler generates a new sample without changing the current state of the Markov chain. This can be used to write models which generate pseudo-data via calls to RNG functions in the transformed data and generated quantities blocks.\n\n\n\nAll HMC algorithms have three parameters:\n\nstep size\nmetric\nintegration time - the number of steps taken along the Hamiltonian trajectory\n\nSee the Stan Reference Manual section on HMC algorithm parameters for further details.\n\n\nThe HMC algorithm simulates the evolution of a Hamiltonian system. The step size parameter controls the resolution of the sampler. Low step sizes can get HMC samplers unstuck that would otherwise get stuck with higher step sizes.\nThe following keyword-value arguments control the step size:\n\nstepsize - How far to move each time the Hamiltonian system evolves forward. Must be a positive real number, default value is \\(1\\).\nstepsize_jitter - Allows step size to be “jittered” randomly during sampling to avoid any poor interactions with a fixed step size and regions of high curvature. Must be a real value between \\(0\\) and \\(1\\). The default value is \\(0\\). Setting stepsize_jitter to \\(1\\) causes step sizes to be selected in the range of \\(0\\) to twice the adapted step size. Jittering below the adapted value will increase the number of steps required and will slow down sampling, while jittering above the adapted value can cause premature rejection due to simulation error in the Hamiltonian dynamics calculation. We strongly recommend always using the default value.\n\n\n\n\nAll HMC implementations in Stan utilize quadratic kinetic energy functions which are specified up to the choice of a symmetric, positive-definite matrix known as a mass matrix or, more formally, a metric Betancourt (2017).\nThe metric argument specifies the choice of Euclidean HMC implementations:\n\nmetric=unit specifies unit metric (diagonal matrix of ones).\nmetric=diag_e specifies a diagonal metric (diagonal matrix with positive diagonal entries). This is the default value.\nmetric=dense_e specifies a dense metric (a dense, symmetric positive definite matrix).\n\nBy default, the metric is estimated during warmup. However, when metric=diag_e or metric=dense_e, an initial guess for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a single variable inv_metric. For a diag_e metric the inv_metric value must be a vector of positive values, one for each parameter in the system. For a dense_e metric, inv_metric value must be a positive-definite square matrix with number of rows and columns equal to the number of parameters in the model.\nThe metric_file option can be used with and without adaptation enabled. If adaptation is enabled, the provided metric will be used as the initial guess in the adaptation process. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess.\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n\n\n\nThe total integration time is determined by the argument engine which take possible values:\n\nnuts - the No-U-Turn Sampler which dynamically determines the optimal integration time.\nstatic - an HMC sampler which uses a user-specified integration time.\n\nThe default argument is engine=nuts.\nThe NUTS sampler generates a proposal by starting at an initial position determined by the parameters drawn in the last iteration. It then evolves the initial system both forwards and backwards in time to form a balanced binary tree. The algorithm is iterative; at each iteration the tree depth is increased by one, doubling the number of leapfrog steps thus effectively doubling the computation time. The algorithm terminates in one of two ways: either the NUTS criterion (i.e., a U-turn in Euclidean space on a subtree) is satisfied for a new subtree or the completed tree; or the depth of the completed tree hits the maximum depth allowed.\nWhen engine=nuts, the subargument max_depth can be used to control the depth of the tree. The default argument is max_depth=10. In the case where a model has a difficult posterior from which to sample, max_depth should be increased to ensure that that the NUTS tree can grow as large as necessary.\nWhen the argument engine=static is specified, the user must specify the integration time via keyword int_time which takes as a value a positive number. The default value is \\(2\\pi\\).",
     "crumbs": [
-      "Getting Started",
-      "Compiling a Stan Program"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/compiling_stan_programs.html#troubleshooting-c-compiler-or-linker-errors",
-    "href": "cmdstan-guide/compiling_stan_programs.html#troubleshooting-c-compiler-or-linker-errors",
-    "title": "Compiling a Stan Program",
+    "objectID": "cmdstan-guide/mcmc_config.html#sampler-diag-file",
+    "href": "cmdstan-guide/mcmc_config.html#sampler-diag-file",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "If the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.\nThe Make command prints the following message to the terminal at the point when it compiles and links the C++ file:\n--- Compiling, linking C++ code ---\nIf the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.\nIf this happens, please report the error, together with the Stan program on either the Stan Forums or on the Stan compiler GitHub issues tracker.",
+    "text": "The output keyword sub-argument diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file which contains sampler information for each draw, and the gradients on the unconstrained scale and log probabilities for all parameters in the model. By default, no auxiliary output file is produced.",
     "crumbs": [
-      "Getting Started",
-      "Compiling a Stan Program"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/compiling_stan_programs.html#c-compilation-and-linking-flags",
-    "href": "cmdstan-guide/compiling_stan_programs.html#c-compilation-and-linking-flags",
-    "title": "Compiling a Stan Program",
+    "objectID": "cmdstan-guide/mcmc_config.html#multi-chain-sampling",
+    "href": "cmdstan-guide/mcmc_config.html#multi-chain-sampling",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "Users can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.\nThe CXXFLAGS and LDFLAGS makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the make/local file.\nFor example:\nCXXFLAGS = -O2\nA recommend a set of CXXFLAGS and LDFLAGS flags can be turned on by setting STAN_CPP_OPTIMS=true in the make/local file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.\n\n\nWhen assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.\nFor some models these checks can represent a significant part of the models execution time. By setting the STAN_NO_RANGE_CHECKS=true makefile flag in the make/local file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.",
+    "text": "A Markov chain generates draws from the target distribution only after it has converged to equilibrium. In theory, convergence is only guaranteed asymptotically as the number of draws grows without bound. In practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available. One way to monitor whether a chain has approximately converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. For robust diagnostics, we recommend running 4 chains.\nThe preferred way of using multiple chains is to run them all from the same executable using the num_chains argument. There is also the option to use the Unix or DOS shell to run multiple executables.\n\n\nThe num_chains argument can be used for all of Stan’s samplers with the exception of the static HMC engine. This will run multiple chains of MCMC from the same executable, which can save on memory usage due to only needing one copy of the model and data. Depending on whether the model was compiled with STAN_THREADS=true, these will either run in parallel or one after the other.\nWhen num_chains is greather than 1 (the default), arguments related to filenames (e.g. output file=, init=) can accept a comma separated list of values, one per each chain.\nFor example, sample will specify the names of the three chain’s output files.\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output_1.csv,output_2.csv,output_3.csv\nThis will write the output in output_1.csv, output_2.csv, output_3.csv.\nIf the model was not compiled with STAN_THREADS=true, the above command will run 3 chains sequentially.\nIf the model was compiled with STAN_THREADS=true, the chains can run in parallel, with the num_threads argument defining the maximum number of threads used to run the chains. If the model uses no within-chain parallelization (map_rect or reduce_sum calls), the below command will run 3 chains in parallel, provided there are cores available:\n./bernoulli sample num_chains=4 data file=bernoulli.data.json num_threads=4\nIf the model uses within-chain parallelization (map_rect or reduce_sum calls), the threads are automatically scheduled to run the parallel parts of a single chain or run the sequential parts of another chains. The below call starts 4 chains that can use 16 threads. At a given moment a single chain may use all 16 threads, 1 thread, anything in between, or can wait for a thread to be available. The scheduling is left to the Threading Building Blocks scheduler.\n./bernoulli_par sample num_chains=4 data file=bernoulli.data.json num_threads=16\n\n\nIf a comma separated list is not used, the num_chains argument changes the normal meanings of filename arguments when it is greater than 1 (the default). They are now interpreted as a “template” which is used for each chain.\nFor example, when num_chains=2, the argument output file=foo.csv no longer produces a file foo.csv, but instead produces two files, foo_1.csv and foo_2.csv. If you also supply id=5, the files produced will be foo_5.csv and foo_6.csv – id=5 gives the id of the first chain, and the remaining chains are sequential from there.\nThis also applies to input files, like those used for initialization. For example, if num_chains=3 and init=bar.json will first look for bar_1.json. If it exists, it will use bar_1.json for the first chain, bar_2.json for the second, and so on. If bar_1.json does not exist, it falls back to looking for bar.json, and if it exists, uses the same initial values for each chain. The numbers in these filenames are also based on the id argument, which defaults to 1.\nFor example, this shorthand is equivalent to the example given above:\n./bernoulli sample num_chains=3 data file=bernoulli.data.json output file=output.csv\nA suffix with the chain id is appended to the provided output filename (output.csv in the above command), so this will also produce files output_1.csv, output_2.csv, output_3.csv.",
     "crumbs": [
-      "Getting Started",
-      "Compiling a Stan Program"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/diagnose_utility.html",
-    "href": "cmdstan-guide/diagnose_utility.html",
-    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "objectID": "cmdstan-guide/mcmc_config.html#summarizing-sampler-outputs-with-stansummary",
+    "href": "cmdstan-guide/mcmc_config.html#summarizing-sampler-outputs-with-stansummary",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "CmdStan is distributed with a utility that is able to read in and analyze the output of one or more Markov chains to check for the following potential problems:\n\nDivergent transitions\nTransitions that hit the maximum treedepth\nLow E-BFMI values\nLow effective sample sizes\nHigh \\(\\hat{R}\\) values\n\nThe meanings of several of these problems are discussed in https://arxiv.org/abs/1701.02434.\n\n\nThe CmdStan makefile task build compiles the diagnose utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/diagnose\n\n\n\nThe diagnose command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to diagnose, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.\nTo fully exercise the diagnose command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the Stan User’s Guide reparameterization section. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y / 2));\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan\nStan has trouble sampling from the region where y is small and thus x is constrained to be near 0. This is due to the fact that the density’s scale changes with y, so that a step size that works well when y is large is inefficient when y is small and vice-versa.\nRunning 4 chains produces output files output_1.csv, …, output_4.csv. We run diagnose command on this fileset:\n&gt; bin/diagnose output_*.csv\nThe output is printed to the terminal window:\nChecking sampler transitions treedepth.\n18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.\nTrajectories that are prematurely terminated due to this limit will result in slow exploration.\nFor optimal performance, increase this limit.\n\nChecking sampler transitions for divergences.\n11 of 4000 (0.28%) transitions ended with a divergence.\nThese divergent transitions indicate that HMC is not fully able to explore the posterior distribution.\nTry increasing adapt delta closer to 1.\nIf this doesn't remove all divergences, try to reparameterize the model.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nThe E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.\nIf possible, try to reparameterize the model.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nThe following parameters had rank-normalized split R-hat greater than 1.01:\n  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]\nSuch high values indicate incomplete mixing and biased estimation.\nYou should consider regularizing your model with additional prior information or a more effective parameterization.\n\nProcessing complete.\nIn this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y / 2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y / 2))\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan\nWe compile the program and run 4 chains, as before. Now the diagnose command doesn’t detect any problems:\nChecking sampler transitions treedepth.\nTreedepth satisfactory for all transitions.\n\nChecking sampler transitions for divergences.\nNo divergent transitions found.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nE-BFMI satisfactory.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nRank-normalized split R-hat values satisfactory for all parameters.\n\nProcessing complete, no problems detected.\n\n\n\n\n\nStan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a Hamiltonian system. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the step size controls the resolution of the sampler.\nUnfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as divergences which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.\nIf the divergent transitions cannot be eliminated by increasing the adapt_delta parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the reparameterization section of the Stan User’s Guide.\n\n\n\nWarnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a validity concern, hitting the maximum treedepth is an efficiency concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the Hamiltonian Monte Carlo Sampling chapter in the Stan Reference Manual). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.\nThis is controlled through the max_depth argument. If the number of transitions which exceed maximum treedepth is low, increasing max_depth may correct this problem.\n\n\n\nThe sampler csv output column energy__ is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of energy is much larger than \\(\\sqrt{D / 2}\\), where \\(D\\) is the number of unconstrained parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.\nThe warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.\n\n\n\nRoughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses \\(\\hat{R}\\) adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.\nBulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g. mean and median).\nOften quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.\n\n\n\n\\(\\hat{R}\\) (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), \\(\\hat{R}\\) is larger than 1. We recommend running at least four chains by default and only using the sample if \\(\\hat{R}\\) is less than 1.01. Stan reports \\(\\hat{R}\\) which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.\nThere is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.",
+    "text": "The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler given a model and data. For all columns in the Stan CSV output file stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\nTo run stansummary on the output files generated by the for loop above, by the above run of the bernoulli model on Mac or Linux:\n&lt;cmdstan-home&gt;/bin/stansummary output_*.csv\nOn Windows, use backslashes to call the stansummary.exe.\n&lt;cmdstan-home&gt;\\bin\\stansummary.exe output_*.csv\nThe stansummary output consists of one row of statistics per column in the Stan CSV output file. Therefore, the first rows in the stansummary report statistics over the sampler state. The final row of output summarizes the estimates of the model variable theta:\nInference for Stan model: bernoulli_model\n4 chains: each with iter=1000; warmup=1000; thin=1; 1000 iterations saved.\n\nWarmup took (0.0060, 0.0040, 0.0050, 0.0050) seconds, 0.020 seconds total\nSampling took (0.0080, 0.010, 0.010, 0.010) seconds, 0.038 seconds total\n\n                Mean     MCSE  StdDev    MAD     5%   50%   95%  ESS_bulk  ESS_tail  ESS_bulk/s  R_hat\n\nlp__            -7.3  1.9e-02    0.72   0.34   -8.7  -7.0  -6.8      1731      1610       45546    1.0\naccept_stat__   0.93  2.7e-03    0.12  0.041   0.68  0.97   1.0      5078      3437     1.3e+05    1.0\nstepsize__      0.90      nan    0.10  0.046   0.82  0.86   1.1       nan       nan         nan    nan\ntreedepth__      1.4  9.3e-03    0.51   0.00    1.0   1.0   2.0      3167      3441     8.3e+04    1.0\nn_leapfrog__     2.7  1.7e-01     1.6   0.00    1.0   3.0   7.0       494      2000     1.3e+04    1.0\ndivergent__     0.00      nan    0.00   0.00   0.00  0.00  0.00       nan       nan         nan    nan\nenergy__         7.8  2.6e-02     1.0   0.70    6.8   7.4   9.8      1598      2069     4.2e+04    1.0\n\ntheta           0.26  2.9e-03    0.12   0.12  0.084  0.24  0.47      1658      1490       43629    1.0\n\nSamples were drawn using hmc with nuts.\nFor each parameter, ESS_bulk and ESS_tail measure the effective sample size for the entire sample (bulk)\nand for the .05 and .95 tails (tail), and R_hat measures the potential scale reduction on split chains.\nAt convergence R_hat will be very close to 1.00.\nIn this example, we conditioned the model on data consisting of the outcomes of 10 bernoulli trials, where only 2 trials reported success. The 5%, 50%, and 95% percentile values for theta reflect the uncertainty in our estimate, due to the small amount of data, given the prior of beta(1, 1)",
     "crumbs": [
-      "Tools and Utilities",
-      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/diagnose_utility.html#building-the-diagnose-command",
-    "href": "cmdstan-guide/diagnose_utility.html#building-the-diagnose-command",
-    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "objectID": "cmdstan-guide/mcmc_config.html#old-multichain",
+    "href": "cmdstan-guide/mcmc_config.html#old-multichain",
+    "title": "MCMC Sampling using Hamiltonian Monte Carlo",
     "section": "",
-    "text": "The CmdStan makefile task build compiles the diagnose utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/diagnose",
+    "text": "Note: Many of these examples can be simplified by using the num_chains argument.\nWhen the num_chains argument is not available or is undesirable for whatever reason, built-in tools in the system shell can be used.\nTo run multiple chains given a model and data, either sequentially or in parallel, we can also use the Unix or DOS shell for loop to set up index variables needed to identify each chain and its outputs.\nOn MacOS or Linux, the for-loop syntax for both the bash and zsh interpreters is:\nfor NAME [in LIST]; do COMMANDS; done\nThe list can be a simple sequence of numbers, or you can use the shell expansion syntax {1..N} which expands to the sequence from \\(1\\) to \\(N\\), e.g. {1..4} expands to 1 2 3 4. Note that the expression {1..N} cannot contain spaces.\nTo run 4 chains for the example bernoulli model on MacOS or Linux:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv\n    done\nThe backslash (\\) indicates a line continuation in Unix. The expression ${i} substitutes in the value of loop index variable i. To run chains in parallel, put an ampersand (&) at the end of the nested sampler command:\n&gt; for i in {1..4}\n    do\n      ./bernoulli sample data file=bernoulli.data.json \\\n      output file=output_${i}.csv &\n    done\nThis pushes each process into the background which allows the loop to continue without waiting for the current chain to finish.\nOn Windows, the DOS for-loop syntax is one of:\nfor %i in (SET) do COMMAND COMMAND-ARGUMENTS\nfor /l %i in (START, STEP, END) do COMMAND COMMAND-ARGUMENTS\nTo run 4 chains in parallel on Windows:\n&gt;for /l %i in (1, 1, 4) do start /b bernoulli.exe sample ^\n                                    data file=bernoulli.data.json my_data ^\n                                    output file=output_%i.csv\nThe caret (^) indicates a line continuation in DOS. The expression %i is the loop index.\nIn the following extended examples, we focus on just the nested sampler command for Unix.\n\n\nFor reproducibility, we specify the same RNG seed across all chains and use the chain id argument to specify the RNG offset.\nThe RNG seed is specified by random seed=&lt;int&gt; and the offset is specified by id=&lt;loop index&gt;, so the call to the sampler is:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            random seed=12345 id=${i}\n\n\n\nThe warmup and sampling iteration keyword-value arguments must follow the sample keyword. The call to the sampler which overrides the default warmup and sampling iterations is:\n./my_model sample num_warmup=500 num_sampling=500 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nTo save warmup draws as part of the Stan CSV output file, use the keyword-value argument save_warmup=true. This must be grouped with the other sample keyword sub-arguments.\n./my_model sample num_warmup=500 num_sampling=500 save_warmup=true \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nBy default, all parameters are initialized on an unconstrained scale to random draws from a uniform distribution over the range \\([{-2}, 2]\\). To initialize some or all parameters to good starting points on the constrained scale from a data file in JSON or Rdump format, use the keyword-value argument init=&lt;filepath&gt;:\n./my_model sample init=my_param_inits.json data file=my_model.data.json \\\n           output file=output_${i}.csv\nTo verify that the specified values will be used by the sampler, you can run the sampler with option algorithm=fixed_param, so that the initial values are used to generate the sample. Since this generates a set of identical draws, setting num_warmp=0 and num_samples=1 saves unnecessary iterations. As the output values are also on the constrained scale, the set of reported values will match the set of specified initial values.\nFor example, if we run the example Bernoulli model with specified initial value for parameter “theta”:\n{ \"theta\" : 0.5 }\nvia command:\n./bernoulli sample algorithm=fixed_param num_warmup=0 num_samples=1 \\\n            init=bernoulli.init.json data file=bernoulli.data.json\nThe resulting output CSV file contains a single draw:\nlp__,accept_stat__,theta\n0,0,0.5\n#\n#  Elapsed Time: 0 seconds (Warm-up)\n#                0 seconds (Sampling)\n#                0 seconds (Total)\n#\n\n\n\nAn initial estimate for the metric can be specified with the metric_file argument whose value is the filepath to a JSON or Rdump file which contains a variable inv_metric. The metric_file option can be used with and without adaptation enabled.\nBy default, the metric is estimated during warmup adaptation. If the initial guess is good, then adaptation should not change it much. If the metric is no good, then the adaptation will override the initial guess. For example, the JSON file bernoulli.diag_e.json, contents\n{ \"inv_metric\" : [0.296291] }\ncan be used as the initial metric as follows:\n../my_model sample algorithm=hmc metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nIf adaptation is disabled, both the metric_file and stepsize arguments should be specified.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nThe resulting output CSV file will contain the following set of comment lines:\n# Adaptation terminated\n# Step size = 0.9\n# Diagonal elements of inverse mass matrix:\n# 0.296291\nAs of Stan versione 2.34, the adapted metric can be saved in JSON format, via sub-argument save_metric, described above. This allows for no or minimal adaptation starting from this file. It is still necessary to specify the stepsize argument as well as the metric_file arguments; the former is the value of the stepsize element in the saved metric file, and the later is the metric file path.\n\n\n\nThe keyword-value arguments for these settings are grouped together under the adapt keyword which itself is a sub-argument of the sample keyword.\nModels with difficult posterior geometries may required increasing the delta argument closer to \\(1\\).\n./my_model sample adapt delta=0.95 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nTo skip adaptation altogether, use the keyword-value argument engaged=false. Disabling adaptation disables both metric and stepsize adaptation, so a stepsize should be provided along with a metric to enable efficient sampling.\n../my_model sample adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\nEven with adaptation disabled, it is still advisable to run warmup iterations in order to allow the initial parameter values to be adjusted to estimates which fall within the typical set.\nTo skip warmup altogether requires specifying both num_warmup=0 and adapt engaged=false.\n../my_model sample num_warmup=0 adapt engaged=false \\\n            algorithm=hmc stepsize=0.9 \\\n            metric_file=bernoulli.diag_e.json \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nModels with difficult posterior geometries may required increasing the max_depth argument from its default value \\(10\\). This requires specifying a series of keyword-argument pairs:\n./my_model sample adapt delta=0.95 \\\n            algorithm=hmc engine=nuts max_depth=15 \\\n            data file=my_model.data.json \\\n            output file=output_${i}.csv\n\n\n\nThe output keyword sub-argument diagnostic_file=&lt;filepath&gt; write the sampler parameters and gradients of all model parameters for each draw to a CSV file:\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            diagnostic_file=diagnostics_${i}.csv\n\n\n\nThe output keyword sub-argument refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is \\(100\\) iterations. The progress updates look like:\nIteration:    1 / 2000 [  0%]  (Warmup)\nIteration:  100 / 2000 [  5%]  (Warmup)\nIteration:  200 / 2000 [ 10%]  (Warmup)\nIteration:  300 / 2000 [ 15%]  (Warmup)\nFor simple models which fit quickly, such updates can be annoying; to suppress them altogether, set refresh=0. This only turns off the Iteration: messages; the configuration and timing information are still written to the terminal.\n./my_model sample data file=my_model.data.json \\\n            output file=output_${i}.csv \\\n            refresh=0\nFor complicated models which take a long time to fit, setting the refresh rate to a low number, e.g. \\(10\\) or even \\(1\\), provides a way to more closely monitor the sampler.\n\n\n\nThe CmdStan argument parser requires keeping sampler config sub-arguments together; interleaving sampler config with the inputs, outputs, inits, RNG seed and chain id config results in an error message such as the following:\n./bernoulli sample data file=bernoulli.data.json adapt delta=0.95\nadapt is either mistyped or misplaced.\nPerhaps you meant one of the following valid configurations?\n  method=sample sample adapt\n  method=variational variational adapt\nFailed to parse arguments, terminating Stan\nThe following example provides a template for a call to the sampler which specifies input data, initial parameters, initial step-size and metric, adaptation, output, and RNG initialization.\n./my_model sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100 \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           random seed=12345 id=${i}\nThe keywords sample, data, output, and random are the top-level argument groups. Within the sample config arguments, the keyword adapt groups the adaptation algorithm parameters and the keyword-value algorithm=hmc groups the NUTS-HMC parameters.\nThe top-level groups can be freely ordered with respect to one another. The following is also a valid command:\n./my_model random seed=12345 id=${i} \\\n           data file=my_model.data.json \\\n           output file=output_${i}.csv refresh=10 \\\n           sample num_warmup=2000 \\\n           init=my_param_inits.json \\\n           algorithm=hmc engine=nuts max_depth=15 \\\n           metric=dense_e metric_file=my_metric.json \\\n           stepsize=0.6555 \\\n           adapt delta=0.95 init_buffer=100 \\\n           window=50 term_buffer=100",
     "crumbs": [
-      "Tools and Utilities",
-      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+      "Running CmdStan",
+      "MCMC Sampling using Hamiltonian Monte Carlo"
     ]
   },
   {
-    "objectID": "cmdstan-guide/diagnose_utility.html#running-the-diagnose-command",
-    "href": "cmdstan-guide/diagnose_utility.html#running-the-diagnose-command",
-    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "objectID": "cmdstan-guide/laplace_sample_config.html",
+    "href": "cmdstan-guide/laplace_sample_config.html",
+    "title": "Laplace sampling",
     "section": "",
-    "text": "The diagnose command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to diagnose, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.\nTo fully exercise the diagnose command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the Stan User’s Guide reparameterization section. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y / 2));\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan\nStan has trouble sampling from the region where y is small and thus x is constrained to be near 0. This is due to the fact that the density’s scale changes with y, so that a step size that works well when y is large is inefficient when y is small and vice-versa.\nRunning 4 chains produces output files output_1.csv, …, output_4.csv. We run diagnose command on this fileset:\n&gt; bin/diagnose output_*.csv\nThe output is printed to the terminal window:\nChecking sampler transitions treedepth.\n18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.\nTrajectories that are prematurely terminated due to this limit will result in slow exploration.\nFor optimal performance, increase this limit.\n\nChecking sampler transitions for divergences.\n11 of 4000 (0.28%) transitions ended with a divergence.\nThese divergent transitions indicate that HMC is not fully able to explore the posterior distribution.\nTry increasing adapt delta closer to 1.\nIf this doesn't remove all divergences, try to reparameterize the model.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nThe E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.\nIf possible, try to reparameterize the model.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nThe following parameters had rank-normalized split R-hat greater than 1.01:\n  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]\nSuch high values indicate incomplete mixing and biased estimation.\nYou should consider regularizing your model with additional prior information or a more effective parameterization.\n\nProcessing complete.\nIn this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y / 2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y / 2))\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan\nWe compile the program and run 4 chains, as before. Now the diagnose command doesn’t detect any problems:\nChecking sampler transitions treedepth.\nTreedepth satisfactory for all transitions.\n\nChecking sampler transitions for divergences.\nNo divergent transitions found.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nE-BFMI satisfactory.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nRank-normalized split R-hat values satisfactory for all parameters.\n\nProcessing complete, no problems detected.",
+    "text": "The laplace method produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood. In general, the posterior mode in the unconstrained space doesn’t correspond to the mean (nor mode) in the constrained space, and thus the sample is needed to infer the mean as well as the standard deviation. (See this case study for a visual illustration.)\nThis is computationally inexpensive compared to exact Bayesian inference with MCMC. The goodness of this estimate depends on both the estimate of the mode and how much the true posterior in the unconstrained space resembles a Gaussian.\n\n\nThis method takes several arguments:\n\nmode - Input file of parameters values on the constrained scale. When Stan’s optimize method is used to estimate the modal values, the value of boolean argument jacobian should be false if optimize was run with default settings, i.e., the input is the MLE estimate; if optimize was run with argument jacobian=true, then the laplace method default setting, jacobian=true, should be used.\njacobian - Whether or not the Jacobian adjustment should be included in the gradient. The default value is true (include adjustment). (Note: in optimization, the default value is false, for historical reasons.)\ndraws - How many total draws to return. The default is \\(1000\\).\ncalculate_lp - Whether to calculate the log probability of the model at each draw. If this is false, the log_p__ column of the output will be entirely nan. The default value is true.\n\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the laplace method is reported at the beginning of the output file as CSV comments.\nOutput columns log_p__ and log_q__, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.\nOutput columns for all model parameters on the constrained scale.\n\n\n\n\nIf requested with output diagnostic_file=, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.\n\n\n\nTo get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:\n\nfind the MAP estimate by running optimization with argument jacobian=true\nrun the Laplace estimator using the MAP estimate as the mode argument.\n\nBecause the default output file name from all methods is output.csv, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.\n./examples/bernoulli/bernoulli optimize jacobian=1 \\\n  data file=examples/bernoulli/bernoulli.data.json \\\n  output file=bernoulli_optimize_lbfgs.csv random seed=1234\n\n\n./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \\\n data file=examples/bernoulli/bernoulli.data.json random seed=1234\nThe header and first few data rows of the output sample are shown below.\n# method = laplace\n#   laplace\n#     mode = bernoulli_lbfgs.csv\n#     jacobian = true (Default)\n#     draws = 1000 (Default)\n#     calculate_lp = true (default)\n# id = 1 (Default)\n# data\n#   file = examples/bernoulli/bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 875960551 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlog_p__,log_q__,theta\n-9.4562,-2.33997,0.0498545\n-6.9144,-0.0117349,0.182898\n-7.18171,-0.746034,0.376428\n...",
     "crumbs": [
-      "Tools and Utilities",
-      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+      "Running CmdStan",
+      "Laplace sampling"
     ]
   },
   {
-    "objectID": "cmdstan-guide/diagnose_utility.html#diagnose-warnings-and-recommendations",
-    "href": "cmdstan-guide/diagnose_utility.html#diagnose-warnings-and-recommendations",
-    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
+    "objectID": "cmdstan-guide/laplace_sample_config.html#configuration",
+    "href": "cmdstan-guide/laplace_sample_config.html#configuration",
+    "title": "Laplace sampling",
     "section": "",
-    "text": "Stan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a Hamiltonian system. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the step size controls the resolution of the sampler.\nUnfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as divergences which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.\nIf the divergent transitions cannot be eliminated by increasing the adapt_delta parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the reparameterization section of the Stan User’s Guide.\n\n\n\nWarnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a validity concern, hitting the maximum treedepth is an efficiency concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the Hamiltonian Monte Carlo Sampling chapter in the Stan Reference Manual). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.\nThis is controlled through the max_depth argument. If the number of transitions which exceed maximum treedepth is low, increasing max_depth may correct this problem.\n\n\n\nThe sampler csv output column energy__ is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of energy is much larger than \\(\\sqrt{D / 2}\\), where \\(D\\) is the number of unconstrained parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.\nThe warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.\n\n\n\nRoughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses \\(\\hat{R}\\) adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.\nBulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g. mean and median).\nOften quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.\n\n\n\n\\(\\hat{R}\\) (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), \\(\\hat{R}\\) is larger than 1. We recommend running at least four chains by default and only using the sample if \\(\\hat{R}\\) is less than 1.01. Stan reports \\(\\hat{R}\\) which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.\nThere is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.",
+    "text": "This method takes several arguments:\n\nmode - Input file of parameters values on the constrained scale. When Stan’s optimize method is used to estimate the modal values, the value of boolean argument jacobian should be false if optimize was run with default settings, i.e., the input is the MLE estimate; if optimize was run with argument jacobian=true, then the laplace method default setting, jacobian=true, should be used.\njacobian - Whether or not the Jacobian adjustment should be included in the gradient. The default value is true (include adjustment). (Note: in optimization, the default value is false, for historical reasons.)\ndraws - How many total draws to return. The default is \\(1000\\).\ncalculate_lp - Whether to calculate the log probability of the model at each draw. If this is false, the log_p__ column of the output will be entirely nan. The default value is true.",
     "crumbs": [
-      "Tools and Utilities",
-      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
+      "Running CmdStan",
+      "Laplace sampling"
     ]
   },
   {
-    "objectID": "cmdstan-guide/external_code.html",
-    "href": "cmdstan-guide/external_code.html",
-    "title": "Using external C++ code",
+    "objectID": "cmdstan-guide/laplace_sample_config.html#csv-output",
+    "href": "cmdstan-guide/laplace_sample_config.html#csv-output",
+    "title": "Laplace sampling",
     "section": "",
-    "text": "The --allow-undefined flag can be passed to the call to stanc, which will allow undefined functions in the Stan language to be parsed without an error. We can then include a definition of the function in a C++ header file.\nThis requires specifying two makefile variables:\n\nSTANCFLAGS=--allow-undefined\nUSER_HEADER=&lt;header_file.hpp&gt;, where &lt;header_file.hpp&gt; is the name of a header file that defines a function with the same name and a compatible signature. This function can appear in the global namespace or in the model namespace, which is defined as the name of the model (either the file name, or the --name argument to stanc) followed by _namespace.\n\nThis is an advanced feature which is only recommended to users familiar with the internals of Stan’s Math library. Most existing C++ code will need to be modified to work with Stan, to varying degrees.\nAs an example, consider the following variant of the Bernoulli example\nfunctions {\n  real make_odds(data real theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(1, 1); // uniform prior on interval 0, 1\n  y ~ bernoulli(theta);\n}\ngenerated quantities {\n  real odds;\n  odds = make_odds(theta);\n}\nHere the make_odds function is declared but not defined, which would ordinarily result in a parser error. However, if you put STANCFLAGS = --allow-undefined into the make/local file or into the stanc call, then the stanc compiler will translate this program to C++, but the generated C++ code will not compile unless you write a file such as examples/bernoulli/make_odds.hpp with the following lines\n#include &lt;ostream&gt;\n\ndouble make_odds(const double& theta, std::ostream *pstream__) {\n  return theta / (1 - theta);\n}\nThe signature for this function needs to fulfill all the usages in the C++ class emitted by stanc. The pstream__ argument is mandatory in the signature but need not be used if your function does not print any output. Because make_odds was declared with a data argument and only used in generated quantites, a signature which accepts and returns double is acceptable. Functions which will have parameters passed as input in the transformed parameters or model blocks will require the ability to accept Stan’s autodiff types. If you wish to autodiff through this function, the simplest option is to make it a template, like\ntemplate &lt;typename T&gt;\nT make_odds(const T &theta, std::ostream *pstream__)\n{\n    return theta / (1 - theta);\n}\nGiven the above, the following make invocation should work\n&gt; make STANCFLAGS=--allow-undefined USER_HEADER=examples/bernoulli/make_odds.hpp examples/bernoulli/bernoulli # on Windows add .exe\nAlternatively, you could put STANCFLAGS and USER_HEADER into the make/local file instead of specifying them on the command-line.\nIf the function were more complicated and involved functions in the Stan Math Library, then you would need to add #include &lt;stan/model/model_header.hpp&gt; and prefix the function calls with stan::math::.\n\n\nExternal C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the Stan Math documentation.\nSuppose you have the following (nonsensical) model which relies on a function called my_dot_self. We will implement this as a copy of the built-in dot_self function.\nfunctions {\n  // both overloads end up using the same C++ template\n  real my_dot_self(vector theta);\n  real my_dot_self(row_vector theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] input_data;\n}\ntransformed data {\n  // no autodiff for data - will call using doubles\n  real ds = my_dot_self(input_data);\n}\nparameters {\n  row_vector[N] thetas;\n}\nmodel {\n  thetas ~ normal(0,1);\n  // autodiff - will call using stan::math::var types\n  input_data ~ normal(thetas, my_dot_self(thetas));\n}\nIf you wanted to autodiff through this function, the following header would suffice1:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr&gt;\ninline stan::value_type_t&lt;EigVec&gt; my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    const auto &x_ref = stan::math::to_ref(x);\n    stan::value_type_t&lt;EigVec&gt; sum_x = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum_x += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum_x;\n}\nHowever, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr,\n          stan::require_not_st_var&lt;EigVec&gt; * = nullptr&gt;\ninline double my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    auto x_ref = stan::math::to_ref(x);\n    double sum = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum;\n}\n\ntemplate &lt;typename EigVec, stan::require_eigen_vt&lt;stan::is_var, EigVec&gt; * = nullptr&gt;\ninline stan::math::var my_dot_self(const EigVec &v, std::ostream *pstream__)\n{\n    // (1) put v into our memory arena\n    stan::arena_t&lt;EigVec&gt; arena_v(v);\n    // (2) calculate forward pass using\n    // (3) the .val() method for matrices of var types\n    stan::math::var res = my_dot_self(arena_v.val(), pstream__);\n    // (4) Place a callback for the reverse pass on the callback stack.\n    stan::math::reverse_pass_callback(\n        [res, arena_v]() mutable\n        { arena_v.adj() += 2.0 * res.adj() * arena_v.val(); });\n    return res;\n}\nFor more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.\n\n\n\nSome functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.\n\nRNGs must end with _rng. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a stan::rng_t object (a type alias to boost::rng::mixmax).\nFunctions which edit the target directly must end with _lp and will be passed a reference to lp__ and a reference to a stan::math::accumulator object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter propto__ which controls whether or not constant terms can be dropped.\nProbability distributions must end with _lpdf or _lpmf and will be passed a boolean template parameter propto__ which controls whether or not constant terms can be dropped.",
+    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the laplace method is reported at the beginning of the output file as CSV comments.\nOutput columns log_p__ and log_q__, the unnormalized log density and the unnormalized density of the Laplace approximation, respectively. These can be used for diagnostics and importance sampling.\nOutput columns for all model parameters on the constrained scale.",
     "crumbs": [
-      "Appendices",
-      "Using external C++ code"
+      "Running CmdStan",
+      "Laplace sampling"
     ]
   },
   {
-    "objectID": "cmdstan-guide/external_code.html#derivative-specializations",
-    "href": "cmdstan-guide/external_code.html#derivative-specializations",
-    "title": "Using external C++ code",
+    "objectID": "cmdstan-guide/laplace_sample_config.html#diagnostic-file-outputs",
+    "href": "cmdstan-guide/laplace_sample_config.html#diagnostic-file-outputs",
+    "title": "Laplace sampling",
     "section": "",
-    "text": "External C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the Stan Math documentation.\nSuppose you have the following (nonsensical) model which relies on a function called my_dot_self. We will implement this as a copy of the built-in dot_self function.\nfunctions {\n  // both overloads end up using the same C++ template\n  real my_dot_self(vector theta);\n  real my_dot_self(row_vector theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] input_data;\n}\ntransformed data {\n  // no autodiff for data - will call using doubles\n  real ds = my_dot_self(input_data);\n}\nparameters {\n  row_vector[N] thetas;\n}\nmodel {\n  thetas ~ normal(0,1);\n  // autodiff - will call using stan::math::var types\n  input_data ~ normal(thetas, my_dot_self(thetas));\n}\nIf you wanted to autodiff through this function, the following header would suffice1:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr&gt;\ninline stan::value_type_t&lt;EigVec&gt; my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    const auto &x_ref = stan::math::to_ref(x);\n    stan::value_type_t&lt;EigVec&gt; sum_x = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum_x += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum_x;\n}\nHowever, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr,\n          stan::require_not_st_var&lt;EigVec&gt; * = nullptr&gt;\ninline double my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    auto x_ref = stan::math::to_ref(x);\n    double sum = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum;\n}\n\ntemplate &lt;typename EigVec, stan::require_eigen_vt&lt;stan::is_var, EigVec&gt; * = nullptr&gt;\ninline stan::math::var my_dot_self(const EigVec &v, std::ostream *pstream__)\n{\n    // (1) put v into our memory arena\n    stan::arena_t&lt;EigVec&gt; arena_v(v);\n    // (2) calculate forward pass using\n    // (3) the .val() method for matrices of var types\n    stan::math::var res = my_dot_self(arena_v.val(), pstream__);\n    // (4) Place a callback for the reverse pass on the callback stack.\n    stan::math::reverse_pass_callback(\n        [res, arena_v]() mutable\n        { arena_v.adj() += 2.0 * res.adj() * arena_v.val(); });\n    return res;\n}\nFor more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.",
+    "text": "If requested with output diagnostic_file=, a JSON file will be created which contains the log density, the gradient, and the Hessian of the log density evaluated at the mode.",
     "crumbs": [
-      "Appendices",
-      "Using external C++ code"
+      "Running CmdStan",
+      "Laplace sampling"
     ]
   },
   {
-    "objectID": "cmdstan-guide/external_code.html#special-functions-rngs-distributions-editing-target",
-    "href": "cmdstan-guide/external_code.html#special-functions-rngs-distributions-editing-target",
-    "title": "Using external C++ code",
+    "objectID": "cmdstan-guide/laplace_sample_config.html#example",
+    "href": "cmdstan-guide/laplace_sample_config.html#example",
+    "title": "Laplace sampling",
     "section": "",
-    "text": "Some functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.\n\nRNGs must end with _rng. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a stan::rng_t object (a type alias to boost::rng::mixmax).\nFunctions which edit the target directly must end with _lp and will be passed a reference to lp__ and a reference to a stan::math::accumulator object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter propto__ which controls whether or not constant terms can be dropped.\nProbability distributions must end with _lpdf or _lpmf and will be passed a boolean template parameter propto__ which controls whether or not constant terms can be dropped.",
+    "text": "To get an approximate estimate of the mode and standard deviation of the example Bernoulli model given the example dataset:\n\nfind the MAP estimate by running optimization with argument jacobian=true\nrun the Laplace estimator using the MAP estimate as the mode argument.\n\nBecause the default output file name from all methods is output.csv, a more informative name is used for the output of optimization. We run the commands from the CmdStan home directory. This results in a sample with mean 2.7 and standard deviation 0.12. In comparison, running the NUTS-HMC sampler results in mean 2.6 and standard deviation 0.12.\n./examples/bernoulli/bernoulli optimize jacobian=1 \\\n  data file=examples/bernoulli/bernoulli.data.json \\\n  output file=bernoulli_optimize_lbfgs.csv random seed=1234\n\n\n./examples/bernoulli/bernoulli laplace mode=bernoulli_optimize_lbfgs.csv \\\n data file=examples/bernoulli/bernoulli.data.json random seed=1234\nThe header and first few data rows of the output sample are shown below.\n# method = laplace\n#   laplace\n#     mode = bernoulli_lbfgs.csv\n#     jacobian = true (Default)\n#     draws = 1000 (Default)\n#     calculate_lp = true (default)\n# id = 1 (Default)\n# data\n#   file = examples/bernoulli/bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 875960551 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlog_p__,log_q__,theta\n-9.4562,-2.33997,0.0498545\n-6.9144,-0.0117349,0.182898\n-7.18171,-0.746034,0.376428\n...",
     "crumbs": [
-      "Appendices",
-      "Using external C++ code"
+      "Running CmdStan",
+      "Laplace sampling"
     ]
   },
   {
-    "objectID": "cmdstan-guide/external_code.html#footnotes",
-    "href": "cmdstan-guide/external_code.html#footnotes",
-    "title": "Using external C++ code",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nDetails of programming in the Stan Math style are omitted from this section, it is presented only as an example↩︎",
+    "objectID": "cmdstan-guide/installation.html",
+    "href": "cmdstan-guide/installation.html",
+    "title": "CmdStan Installation",
+    "section": "",
+    "text": "There are a few ways that you can install CmdStan. Depending on your operating system and your level of expertise, you can either:\n\nUse the conda package management system to install a pre-built version of CmdStan along with the required dependencies. Recommended for Windows users.\nInstall the source code from GitHub CmdStan repository. This requires a modern C++ compiler and toolchain. See the C++ Toolchain section for further details.\n\n\n\nWith conda, you can install CmdStan from the conda-forge channel. This will install a pre-built version of CmdStan along with the required dependencies (i.e. a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command cmdstan_model to activate the CmdStan makefile from anywhere.\nNote: This requires that conda has been installed already on your machine. We recommend using the miniforge distribution.\nWe recommend installing CmdStan in a new conda environment:\n conda create -n stan -c conda-forge cmdstan\nThis command creates a new conda environment named stan and downloads and installs the cmdstan package as well as CmdStan and the required C++ toolchain.\nTo install into an existing conda environment, use the conda install command instead of create:\n conda install -c conda-forge cmdstan\nWhichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment stan, run the command\n conda activate stan\nBy default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and newer can be installed by specifying cmdstan==VERSION in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment\nconda install -c conda-forge cmdstan=2.27.0\n\n\nA Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command\n conda info -e\nThe shell environment variable CONDA_PREFIX points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the bin subdirectory of the conda environment directory, i.e., $CONDA_PREFIX/bin/cmdstan (Linux, MacOS), %CONDA_PREFIX%\\bin\\cmdstan (Windows).\nPlease report conda-specific install problems directly to the conda-forge issue tracker, here.\n\n\n\n\nInstallation from GitHub consists of the following steps:\n\nVerify that you have a modern C++ toolchain. See the C++ Toolchain section for details.\nDownload the CmdStan source code from GitHub\nBuild the CmdStan libraries and executables\nCheck the installation by compiling and running the CmdStan example model bernoulli.stan.\n\n\n\nThe GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the cmdstan module; the Stan inference engine algorithms and Stan math library functions are specified as submodules and stored in the GitHub repositories stan and math, respectively.\nA CmdStan release is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do not contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-”, e.g. “cmdstan-2.29.2”.\nBy cloning the CmdStan repository with argument --recursive, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named cmdstan:\n&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive\nThroughout this manual, we refer to this top-level CmdStan source directory as &lt;cmdstan-home&gt;. This directory contains the following subdirectories:\n\ndirectory cmdstan/stan contains the sub-module stan (https://github.com/stan-dev/stan)\ndirectory cmdstan/stan/lib/stan_math contains the sub-module math (https://github.com/stan-dev/math)\n\n\n\n\nBuilding CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:\n\nstanc: the Stan compiler (translates Stan language to C++).\nstansummary: a basic posterior analysis tool. The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\ndiagnose: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.\n\nCmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): bin/linux-stanc, bin/mac-stanc and bin/windows-stanc. The CmdStan makefile build task copies the appropriate binary to bin/stanc. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.\nSteps to build CmdStan:\n\nOpen a command-line terminal window and change directories to the CmdStan home directory.\nRun the makefile target build which instantiates the CmdStan utilities and compiles all necessary C++ libraries.\n\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make build\nIf your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the -j option. For example, to build on 4 cores, type:\n&gt; make -j4 build\nWhen make build is successful, the directory &lt;cmdstan-home&gt;/bin/ will contain the executables stanc, stansummary, and diagnose (on Windows, corresponding .exe files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:\n--- CmdStan v2.29.2 built ---\nWarning: The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.\nWindows only: CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory &lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb is part of the PATH environment variable. See these instructions for details on changing the PATH. To permanently make this setting for the current user, you may execute:\n&gt; make install-tbb\nAfter changing the PATH environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)\n\n\n\n\nTo check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.\nOn Linux and macOS:\n# compile the example\n&gt; make examples/bernoulli/bernoulli\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli sample\\\n  data file=examples/bernoulli/bernoulli.data.json\n\n# default output written to file `output.csv`,\n# default num_samples is 1000, output file should have approx. 1050 lines\n&gt; wc -l output.csv\n\n# run the `bin/stansummary utility to summarize parameter estimates\n&gt; bin/stansummary output.csv\nOn Windows:\n# compile the example\n&gt; make examples/bernoulli/bernoulli.exe\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json\n\n# run the `bin/stansummary.exe utility to summarize parameter estimates\n&gt; bin/stansummary.exe output.csv\nThe sample data in file bernoulli.json.data specifies 2 out of 10 successes, therefore the range mean(theta)\\(\\pm\\)sd(theta) should include 0.2.\n\n\n\nUpdates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the &lt;cmdstan-home&gt; directory:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make clean-all\n&gt; make build\n\n\nThis section contains solutions to problems reported on https://discourse.mc-stan.org\nCompiler error message about PCH file\nTo speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:\nerror: PCH file uses an older PCH format that is no longer supported\nIn this case, clean and rebuild CmdStan, as shown in the previous section.\nWindows: ‘g++’, ‘make’, or ‘cut’ is not recognized\nThe CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:\n'cut' is not recognized as an internal or external command,\noperable program or batch file.\nTo fix this, ensure you have followed the steps for adding the toolchain to your PATH and installing the additional utilities covered in the configuration instructions\nSpaces in paths to CmdStan or model\nmake can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the One Drive folder.\nUnfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:\nmake: *** INTERNAL: readdir: Invalid argument\nmake: *** [make/program:50: x.hpp] Error 2\nIf the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.\n\n\n\n\nCompiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.\n\n\nThe required C++ compiler is g++ 4.9 3. On most systems the GNU Make utility is pre-installed and is the default make utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:\ng++ --version\nmake --version\nIf these are at least at g++ version 4.9.3 or later and make version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler g++, because later versions are faster.\nTo install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:\nsudo apt install g++\nsudo apt install make\nIf you can’t run sudo, you will need to ask your sysadmin or cluster administrator to install these tools for you.\n\n\n\nTo check if you already already have an appropriate toolchain installed, open the Terminal application and enter:\nclang++ --version\nmake --version\nIf either of these commands prints the message command not found, you will need to install Xcode’s command line tools.\nOpen the Terminal application and enter:\nxcode-select --install\nSelect “Install” in the window that opens.\nAfter the installation completes, you can double check that installation was successful by reopening the Terminal and running:\nclang++ --version\nmake --version\nYou can read more about Xcode on its site: https://developer.apple.com/xcode/\nWe don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.\n\n\n\nThe Windows toolchain consists of programs g++, the C++ compiler, and make, the GNU Make utility. To check if these are present, open a command shell [^1] and type:\ng++ --version\nmake --version\nCmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your PATH variable, See these instructions for details on changing the PATH if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.\n\n\nAll required utilities (e.g., make, g++) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the R Project documentation.\nThe R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.\nNote that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a Microsoft update.\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45\\usr\\bin\nC:\\rtools45\\x86_64-w64-mingw32.static.posix\\bin\n\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45 - ARM64\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45-aarch64\\usr\\bin\nC:\\rtools45-aarch64\\aarch64-w64-mingw32.static.posix\\bin\n\n\n\n\n\n\nCmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.\nGNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed only when there are changes to those other files. Thus Make provides an efficient way to manage complex software.\nThe CmdStan Makefile is in the &lt;cmdstan-home&gt; directory and is named makefile. This is one of the default GNU Makefile names, which allows you to omit the -f makefile argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, Make only works properly when invoked from the &lt;cmdstan-home&gt; directory; attempts to use this Makefile from another directory by specifying the full path to the file makefile won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:\nmake -f ~/github/stan-dev/cmdstan/makefile\n/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory\nmake: *** No rule to make target `make/command'.  Stop.\nThe conda-forge cmdstan package provides a solution to this problem via cmdstan_model command which lets you run the CmdStan makefile from anywhere to compile a Stan model.\nMakefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix .stan; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program my_program.stan in directory ../my_dir/, the make target is ../my_dir/my_program or ../my_dir/my_program.exe (on Windows).\nTo call Make, you invoke the utility name, make, followed by, in order:\n\nzero or more Make program options, then specify any Make variables as a series of\nzero of more Make variables, described below\nzero or more target names; the set of names is determined by the Makefile rules.\n\nmake &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;\nMakefile Variables\nMake targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile ../my_dir/my_program.stan for an OpenCL (GPU) machine, set the makefile variable STAN_OPENCL to TRUE:\n&gt; make STAN_OPENCL=TRUE ../my_dir/my_program\nMakefile variables can also be set by creating a file named local in the CmdStan make subdirectory which contains a list of &lt;VARIABLE&gt;=&lt;VALUE&gt; pairs, one per line. For example, to get the same effect as the above command every time, you would put the line STAN_OPENCL=TRUE into the file &lt;cmdstan_home&gt;/make/local.\nThe complete set of Makefile variables can be found in file &lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags.\nMake Targets\nWhen invoked without any arguments at all, Make prints a help message:\n&gt; make\n--------------------------------------------------------------------------------\nCmdStan v2.33.1 help\n\n  Build CmdStan utilities:\n    &gt; make build\n\n    This target will:\n    1. Install the Stan compiler bin/stanc from stanc3 binaries.\n    2. Build the print utility bin/print (deprecated; will be removed in v3.0)\n    3. Build the stansummary utility bin/stansummary\n    4. Build the diagnose utility bin/diagnose\n    5. Build all libraries and object files compile and link an executable Stan program\n\n    Note: to build using multiple cores, use the -j option to make, e.g.,\n    for 4 cores:\n    &gt; make build -j4\n\n\n  Build a Stan program:\n\n    Given a Stan program at foo/bar.stan, build an executable by typing:\n    &gt; make foo/bar\n\n    This target will:\n    1. Install the Stan compiler (bin/stanc), as needed.\n    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.\n    3. Compile the C++ code using cc . to generate foo/bar\n\n  Additional make options:\n    STANCFLAGS: defaults to \"\". These are extra options passed to bin/stanc\n      when generating C++ code. If you want to allow undefined functions in the\n      Stan program, either add this to make/local or the command line:\n          STANCFLAGS = --allow_undefined\n    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the\n      header file that is included. This defaults to \"user_header.hpp\" in the\n      directory of the Stan program.\n    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads\n      the nightly version.\n    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.\n    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.\n\n\n  Example - bernoulli model: examples/bernoulli/bernoulli.stan\n\n    1. Build the model:\n       &gt; make examples/bernoulli/bernoulli\n    2. Run the sampling algorithm given the model and data:\n       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R\n    3. Look at the posterior sample:\n       &gt; bin/stansummary output.csv\n\n\n  Clean CmdStan:\n\n    Remove the built CmdStan tools:\n    &gt; make clean-all\n\n--------------------------------------------------------------------------------",
     "crumbs": [
-      "Appendices",
-      "Using external C++ code"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/index.html",
-    "href": "cmdstan-guide/index.html",
-    "title": "CmdStan User’s Guide",
+    "objectID": "cmdstan-guide/installation.html#conda-install",
+    "href": "cmdstan-guide/installation.html#conda-install",
+    "title": "CmdStan Installation",
     "section": "",
-    "text": "This document is a user’s guide for CmdStan, the command-line interface to the Stan statistical modeling language. CmdStan provides the programs and tools to compile Stan programs into C++ executables that can be run directly from the command line, together with a few utilities to check and summarize the resulting outputs.\nIn CmdStan, statistical models written in the Stan probabilistic programming language are translated into a C++ program which is then compiled together with the CmdStan routines that provide the logic needed to manage all user inputs and program outputs and the Stan inference algorithms and math library. The resulting command line executable program can be used to\n\ndo inference on data, producing an exact or approximate estimate of the posterior;\ngenerate new quantities of interest from an existing estimate;\ngenerate data from the model according to a given set of parameters.\n\nThe packages CmdStanR and CmdStanPy provide interfaces to CmdStan from R and Python, respectively, similarly, JuliaStan also interfaces with CmdStan.\nDownload the pdf version of this manual.\n\nBenefits of CmdStan\n\nWith every new Stan release, there is a corresponding CmdStan release, therefore CmdStan provides access to the latest version of Stan, and can be used to run the development version of Stan as well.\nOf the Stan interfaces, CmdStan has the lightest memory footprint, therefore it can fit larger and more complex models. It has has the fewest dependencies, which makes it easier to run in limited environments such as clusters.\nThe output generated is in CSV format and can be post-processed using other Stan interfaces or general tools.\n\n\n\nStan documentation\n\nStan User’s Guide The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:\nStan Reference Manual Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.\nStan Functions Reference The Stan Functions Reference provides definitions and examples for all the functions defined in the Stan math library and available in the Stan programming language, including all probability distributions.\n\n\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "text": "With conda, you can install CmdStan from the conda-forge channel. This will install a pre-built version of CmdStan along with the required dependencies (i.e. a C++ compiler, a version of Make, and required libraries). The conda installation is designed so one can use the R or Python bindings to CmdStan seamlessly. Additionally, it provides the command cmdstan_model to activate the CmdStan makefile from anywhere.\nNote: This requires that conda has been installed already on your machine. We recommend using the miniforge distribution.\nWe recommend installing CmdStan in a new conda environment:\n conda create -n stan -c conda-forge cmdstan\nThis command creates a new conda environment named stan and downloads and installs the cmdstan package as well as CmdStan and the required C++ toolchain.\nTo install into an existing conda environment, use the conda install command instead of create:\n conda install -c conda-forge cmdstan\nWhichever installation method you use, afterwards you must activate the new environment or deactivate/activate the existing one. For example, if you installed cmdstan into a new environment stan, run the command\n conda activate stan\nBy default, the latest release of CmdStan is installed. If you require a specific release of CmdStan, CmdStan versions 2.26.1 and newer can be installed by specifying cmdstan==VERSION in the install command. For example to install an earlier version of CmdStan into your current conda environment, run the following command, then re-activate the environment\nconda install -c conda-forge cmdstan=2.27.0\n\n\nA Conda environment is a directory that contains a specific collection of Conda packages. To see the locations of your conda environments, use the command\n conda info -e\nThe shell environment variable CONDA_PREFIX points to the active conda environment (if any). Both CmdStan and the C++ toolchain are installed into the bin subdirectory of the conda environment directory, i.e., $CONDA_PREFIX/bin/cmdstan (Linux, MacOS), %CONDA_PREFIX%\\bin\\cmdstan (Windows).\nPlease report conda-specific install problems directly to the conda-forge issue tracker, here.",
     "crumbs": [
-      "CmdStan User's Guide"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/json_apdx.html",
-    "href": "cmdstan-guide/json_apdx.html",
-    "title": "JSON Format for CmdStan",
+    "objectID": "cmdstan-guide/installation.html#installation-from-github",
+    "href": "cmdstan-guide/installation.html#installation-from-github",
+    "title": "CmdStan Installation",
     "section": "",
-    "text": "CmdStan can use JSON format for input data for both model data and parameters. Model data is read in by the model constructor. Model parameters are used to initialize the sampler and optimizer.\n\n\nYou can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the write_stan_json() function provided by the CmdStanR interface.\n\n\n\nJSON is a data interchange notation, defined by an EMCA standard. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:\n\nStructural tokens are the left and right curly bracket {}, left and right square bracket [], the semicolon ;, and the comma ,.\nLiteral tokens must always be in lowercase. There are three literal tokens: true, false, null.\nA primitive value is a single token which is either a literal, a string, or a number.\nA string consists of zero or more Unicode characters enclosed in double quotes, e.g. \"foo\". A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g. \"\\\\uHHHH\" where HHHH is the Unicode code point in hex.\nNumbers are represented using either decimal notation or scientific notation. The following are examples of numbers: 17, 17.2, -17.2, -17.2e8, 17.2e-8.  There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.\nThe special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string \"Inf\", the string \"Infinity\", or the atom Infinity. Negative infinity can be represented as the string \"-Inf\", the string \"-Infinity\", or the atom -Infinity. Not-a-number can be represented as the string \"NaN\" or the atom NaN. These values may be mixed with other numerical types.\nA complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number \\(2.3 - 1.83i\\) would be represented in JSON as the two-element array [2.3, -1.83].\nA JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: [], [1], [0.2, \"-inf\", true].\nVectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector \\([1 \\quad 2]^{\\top}\\) and the row vector \\([1 \\quad 2]\\) are represented by the JSON array [1, 2].\nComplex vectors are represented as arrays of two-element arrays. For example, the complex vector \\([2.3 - 1.83i \\quad -4.8 +\n2i]^{\\top}\\) is represented as [[2.3, -1.83], [-4.8, 2]] in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).\nMatrices are represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 & 2.7 & -9.8 \\\\\n4.2 & 1.8 & -7.3\n\\end{bmatrix}\n\\end{equation*}\\] is represented in JSON as [[1, 2.7, -9.8],   [4.2, 1.8, -7.3]].\nComplex matrices are also represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) complex matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 + 2i & 3 - 4.2i & 13.1 + 2.7i \\\\\n3.1 & -5i & 0\n\\end{bmatrix}\n\\end{equation*}\\] would be represented in JSON as [[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]].\nTuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple (1.5, 3.4) is represented in JSON as {\"1\": 1.5, \"2\": 3.4}.\nA name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.\nA JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: { }, {\"foo\": null}, {\"bar\" : 17, \"baz\" : [14,15,16.6] }.\n\n\n\n\nStan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:\n{ \"N\" : 10, \"y\" : [0,1,0,0,0,0,0,0,0,1] }\nMatrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:\ndata {\n  int d1;\n  int d2;\n  int d3;\n  array[d1, d2, d3] int ar;\n}\nthe following JSON input would be valid:\n{ \"d1\" : 2,\n  \"d2\" : 3,\n  \"d3\" : 4,\n  \"ar\" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],\n          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]\n}\nJSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.\nAll data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.\n\n\n\n\n\n\n\nStan declaration\nJSON encoding\n\n\n\n\nint i\n\"i\": 17\n\n\n\n\n\n\nreal a\n\"a\" : 17\n\n\n\n\"a\" : 17.2\n\n\n\n\"a\" : \"NaN\"\n\n\n\n\"a\" : \"+inf\"\n\n\n\n\"a\" : \"-inf\"\n\n\n\n\n\n\ncomplex z\n\"z\": [1, -2.3]\n\n\narray[5] int\n\"a\" : [1, 2, 3, 4, 5]\n\n\narray[5] real a\n\"a\" : [ 1, 2, 3.3, \"NaN\", 5 ]\n\n\narray[2] complex b\n\"b\" : [[1, -2.3], [4.9, 0]]\n\n\nvector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nrow_vector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nmatrix[2, 3] a\n\"a\" : [[ 1, 2, 3 ], [ 4, 5, 6]]\n\n\ncomplex_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_row_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_matrix[2, 3] d\n\"d\" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]\n\n\ntuple(real, array[2] int) t\n\"t\" : { \"1\": 1.4, \"2\": [1, 2]}\n\n\n\n\n\nJSON notation is not able to distinguish between multi-dimensional arrays where any dimension is \\(0\\), e.g., a 2-D array with dimensions \\((1,0)\\), i.e., an array which contains a single array which is empty, has JSON representation [ ]. To see how this works, consider the following Stan program data block:\ndata {\n  int d;\n  array[d] int ar_1d;\n  array[d, d] int ar_2d;\n  array[d, d, d] int ar_3d;\n}\nIn the case where variable d is 1, all arrays will contain a single value. If array variable ar_d1 contains value 7, 2-D array variable ar_d2 contains (an array which contains) value 8, and 3-D array variable ar_d3 contains (an array which contains an array which contains) value 9, the JSON representation is:\n{ \"ar_d1\" : [7],\n  \"ar_d2\" : [[8]],\n  \"ar_d3\" : [[[9]]]\n}\nHowever, in the case where variable d is 0, ar_d1 is empty, i.e., it contains no values, as is ar_d2, ar_d3, and the JSON representation is\n{ \"d\" : 0,\n  \"ar_d1\" : [ ],\n  \"ar_d2\" : [ ],\n  \"ar_d3\" : [ ]\n}",
+    "text": "Installation from GitHub consists of the following steps:\n\nVerify that you have a modern C++ toolchain. See the C++ Toolchain section for details.\nDownload the CmdStan source code from GitHub\nBuild the CmdStan libraries and executables\nCheck the installation by compiling and running the CmdStan example model bernoulli.stan.\n\n\n\nThe GitHub source code is divided into sub-modules, each in its own repository. The CmdStan repo contains just the cmdstan module; the Stan inference engine algorithms and Stan math library functions are specified as submodules and stored in the GitHub repositories stan and math, respectively.\nA CmdStan release is compressed tarfile which contains CmdStan and the Stan and math library submodules. The most recent CmdStan release is always available as https://github.com/stan-dev/cmdstan/releases/latest. A CmdStan release is versioned by major, minor, patch numbers, e.g., “2.29.2”. Please ensure you download a tarfile which is named “cmdstan-&lt;version-number” rather than using the “Source Code” links at the bottom of the release. These are automatically generated by GitHub and do not contain the required submodules. The release tarfile unpacks into a directory named “cmdstan-”, e.g. “cmdstan-2.29.2”.\nBy cloning the CmdStan repository with argument --recursive, Git automatically initializes and updates each submodule in the repository, including nested submodules if any of the submodules in the repository have submodules themselves. The following command will download the source code from the current development branch of CmdStan into a directory named cmdstan:\n&gt; git clone https://github.com/stan-dev/cmdstan.git --recursive\nThroughout this manual, we refer to this top-level CmdStan source directory as &lt;cmdstan-home&gt;. This directory contains the following subdirectories:\n\ndirectory cmdstan/stan contains the sub-module stan (https://github.com/stan-dev/stan)\ndirectory cmdstan/stan/lib/stan_math contains the sub-module math (https://github.com/stan-dev/math)\n\n\n\n\nBuilding CmdStan involves preparing a set of executable programs and compiling the command line interface and supporting libraries. The CmdStan tools are:\n\nstanc: the Stan compiler (translates Stan language to C++).\nstansummary: a basic posterior analysis tool. The stansummary utility processes one or more output files from a run or set of runs of Stan’s HMC sampler. For all parameters and quantities of interest in the Stan program, stansummary reports a set of statistics including mean, standard deviation, percentiles, effective sample size, and \\(\\hat{R}\\) values.\ndiagnose: a basic sampler diagnostic tool which checks for indications that the HMC sampler was unable to sample from the full posterior.\n\nCmdStan releases include pre-built binaries of the Stan language compiler (https://github.com/stan-dev/stanc3): bin/linux-stanc, bin/mac-stanc and bin/windows-stanc. The CmdStan makefile build task copies the appropriate binary to bin/stanc. For CmdStan installations which have been cloned of downloaded from the CmdStan GitHub repository, the makefile task will download the appropriate OS-specific binary from the stanc3 repository’s nightly release.\nSteps to build CmdStan:\n\nOpen a command-line terminal window and change directories to the CmdStan home directory.\nRun the makefile target build which instantiates the CmdStan utilities and compiles all necessary C++ libraries.\n\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make build\nIf your computer has multiple cores and sufficient ram, the build process can be parallelized by providing the -j option. For example, to build on 4 cores, type:\n&gt; make -j4 build\nWhen make build is successful, the directory &lt;cmdstan-home&gt;/bin/ will contain the executables stanc, stansummary, and diagnose (on Windows, corresponding .exe files) and the final lines of console output will show the version of CmdStan that has just been built, e.g.:\n--- CmdStan v2.29.2 built ---\nWarning: The Make program may take 10+ minutes and consume 2+ GB of memory to build CmdStan.\nWindows only: CmdStan requires that the Intel TBB library, which is built by the above command, can be found by the Windows system. This requires that the directory &lt;cmdstan-home&gt;/stan/lib/stan_math/lib/tbb is part of the PATH environment variable. See these instructions for details on changing the PATH. To permanently make this setting for the current user, you may execute:\n&gt; make install-tbb\nAfter changing the PATH environment variable, you must open an new shell in order for the new environment variable settings to take effect. (This is not necessary on Mac and Linux systems because they can use the absolute path to the Intel TBB library when linking into Stan programs.)",
     "crumbs": [
-      "Appendices",
-      "JSON Format for CmdStan"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/json_apdx.html#creating-json-files",
-    "href": "cmdstan-guide/json_apdx.html#creating-json-files",
-    "title": "JSON Format for CmdStan",
+    "objectID": "cmdstan-guide/installation.html#checking-the-stan-compiler",
+    "href": "cmdstan-guide/installation.html#checking-the-stan-compiler",
+    "title": "CmdStan Installation",
     "section": "",
-    "text": "You can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the write_stan_json() function provided by the CmdStanR interface.",
+    "text": "To check that the CmdStan installation is complete and in working order, run the following series of commands from the folder which CmdStan was installed.\nOn Linux and macOS:\n# compile the example\n&gt; make examples/bernoulli/bernoulli\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli sample\\\n  data file=examples/bernoulli/bernoulli.data.json\n\n# default output written to file `output.csv`,\n# default num_samples is 1000, output file should have approx. 1050 lines\n&gt; wc -l output.csv\n\n# run the `bin/stansummary utility to summarize parameter estimates\n&gt; bin/stansummary output.csv\nOn Windows:\n# compile the example\n&gt; make examples/bernoulli/bernoulli.exe\n\n# fit to provided data (results of 10 trials, 2 out of 10 successes)\n&gt; ./examples/bernoulli/bernoulli.exe sample data file=examples/bernoulli/bernoulli.data.json\n\n# run the `bin/stansummary.exe utility to summarize parameter estimates\n&gt; bin/stansummary.exe output.csv\nThe sample data in file bernoulli.json.data specifies 2 out of 10 successes, therefore the range mean(theta)\\(\\pm\\)sd(theta) should include 0.2.",
     "crumbs": [
-      "Appendices",
-      "JSON Format for CmdStan"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/json_apdx.html#json-syntax-summary",
-    "href": "cmdstan-guide/json_apdx.html#json-syntax-summary",
-    "title": "JSON Format for CmdStan",
+    "objectID": "cmdstan-guide/installation.html#troubleshooting-the-installation",
+    "href": "cmdstan-guide/installation.html#troubleshooting-the-installation",
+    "title": "CmdStan Installation",
     "section": "",
-    "text": "JSON is a data interchange notation, defined by an EMCA standard. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:\n\nStructural tokens are the left and right curly bracket {}, left and right square bracket [], the semicolon ;, and the comma ,.\nLiteral tokens must always be in lowercase. There are three literal tokens: true, false, null.\nA primitive value is a single token which is either a literal, a string, or a number.\nA string consists of zero or more Unicode characters enclosed in double quotes, e.g. \"foo\". A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g. \"\\\\uHHHH\" where HHHH is the Unicode code point in hex.\nNumbers are represented using either decimal notation or scientific notation. The following are examples of numbers: 17, 17.2, -17.2, -17.2e8, 17.2e-8.  There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.\nThe special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string \"Inf\", the string \"Infinity\", or the atom Infinity. Negative infinity can be represented as the string \"-Inf\", the string \"-Infinity\", or the atom -Infinity. Not-a-number can be represented as the string \"NaN\" or the atom NaN. These values may be mixed with other numerical types.\nA complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number \\(2.3 - 1.83i\\) would be represented in JSON as the two-element array [2.3, -1.83].\nA JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: [], [1], [0.2, \"-inf\", true].\nVectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector \\([1 \\quad 2]^{\\top}\\) and the row vector \\([1 \\quad 2]\\) are represented by the JSON array [1, 2].\nComplex vectors are represented as arrays of two-element arrays. For example, the complex vector \\([2.3 - 1.83i \\quad -4.8 +\n2i]^{\\top}\\) is represented as [[2.3, -1.83], [-4.8, 2]] in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).\nMatrices are represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 & 2.7 & -9.8 \\\\\n4.2 & 1.8 & -7.3\n\\end{bmatrix}\n\\end{equation*}\\] is represented in JSON as [[1, 2.7, -9.8],   [4.2, 1.8, -7.3]].\nComplex matrices are also represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) complex matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 + 2i & 3 - 4.2i & 13.1 + 2.7i \\\\\n3.1 & -5i & 0\n\\end{bmatrix}\n\\end{equation*}\\] would be represented in JSON as [[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]].\nTuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple (1.5, 3.4) is represented in JSON as {\"1\": 1.5, \"2\": 3.4}.\nA name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.\nA JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: { }, {\"foo\": null}, {\"bar\" : 17, \"baz\" : [14,15,16.6] }.",
+    "text": "Updates to CmdStan, changes in compiler options, or updates to the C++ toolchain may result in errors when trying to compile a Stan program. Often, these problems can be resolved by removing the existing CmdStan binaries and recompiling. To do this, you must run the makefile commands from the &lt;cmdstan-home&gt; directory:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make clean-all\n&gt; make build\n\n\nThis section contains solutions to problems reported on https://discourse.mc-stan.org\nCompiler error message about PCH file\nTo speed up compilation, the Stan makefile pre-compiles parts of the core Stan library. If these pre-compiled files are out of sync with the compiled model, the compiler will complain, e.g.:\nerror: PCH file uses an older PCH format that is no longer supported\nIn this case, clean and rebuild CmdStan, as shown in the previous section.\nWindows: ‘g++’, ‘make’, or ‘cut’ is not recognized\nThe CmdStan makefile uses a few shell utilities which might not be present in Windows, resulting in the error message:\n'cut' is not recognized as an internal or external command,\noperable program or batch file.\nTo fix this, ensure you have followed the steps for adding the toolchain to your PATH and installing the additional utilities covered in the configuration instructions\nSpaces in paths to CmdStan or model\nmake can fail when dealing with files in folders with a space somewhere in their file path. Particularly on Windows, this can be an issue when CmdStan, or the models you are trying to build, are placed in the One Drive folder.\nUnfortunately, the errors created by this situation are not alwas informative. Some errors you may see are:\nmake: *** INTERNAL: readdir: Invalid argument\nmake: *** [make/program:50: x.hpp] Error 2\nIf the (fully-expanded) folder path to CmdStan or the model you are trying to build contains a space, we recommend trying a different location if you encounter any issues during building.",
     "crumbs": [
-      "Appendices",
-      "JSON Format for CmdStan"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/json_apdx.html#stan-data-types-in-json-notation",
-    "href": "cmdstan-guide/json_apdx.html#stan-data-types-in-json-notation",
-    "title": "JSON Format for CmdStan",
+    "objectID": "cmdstan-guide/installation.html#cpp-toolchain",
+    "href": "cmdstan-guide/installation.html#cpp-toolchain",
+    "title": "CmdStan Installation",
     "section": "",
-    "text": "Stan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:\n{ \"N\" : 10, \"y\" : [0,1,0,0,0,0,0,0,0,1] }\nMatrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:\ndata {\n  int d1;\n  int d2;\n  int d3;\n  array[d1, d2, d3] int ar;\n}\nthe following JSON input would be valid:\n{ \"d1\" : 2,\n  \"d2\" : 3,\n  \"d3\" : 4,\n  \"ar\" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],\n          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]\n}\nJSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.\nAll data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.\n\n\n\n\n\n\n\nStan declaration\nJSON encoding\n\n\n\n\nint i\n\"i\": 17\n\n\n\n\n\n\nreal a\n\"a\" : 17\n\n\n\n\"a\" : 17.2\n\n\n\n\"a\" : \"NaN\"\n\n\n\n\"a\" : \"+inf\"\n\n\n\n\"a\" : \"-inf\"\n\n\n\n\n\n\ncomplex z\n\"z\": [1, -2.3]\n\n\narray[5] int\n\"a\" : [1, 2, 3, 4, 5]\n\n\narray[5] real a\n\"a\" : [ 1, 2, 3.3, \"NaN\", 5 ]\n\n\narray[2] complex b\n\"b\" : [[1, -2.3], [4.9, 0]]\n\n\nvector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nrow_vector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nmatrix[2, 3] a\n\"a\" : [[ 1, 2, 3 ], [ 4, 5, 6]]\n\n\ncomplex_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_row_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_matrix[2, 3] d\n\"d\" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]\n\n\ntuple(real, array[2] int) t\n\"t\" : { \"1\": 1.4, \"2\": [1, 2]}\n\n\n\n\n\nJSON notation is not able to distinguish between multi-dimensional arrays where any dimension is \\(0\\), e.g., a 2-D array with dimensions \\((1,0)\\), i.e., an array which contains a single array which is empty, has JSON representation [ ]. To see how this works, consider the following Stan program data block:\ndata {\n  int d;\n  array[d] int ar_1d;\n  array[d, d] int ar_2d;\n  array[d, d, d] int ar_3d;\n}\nIn the case where variable d is 1, all arrays will contain a single value. If array variable ar_d1 contains value 7, 2-D array variable ar_d2 contains (an array which contains) value 8, and 3-D array variable ar_d3 contains (an array which contains an array which contains) value 9, the JSON representation is:\n{ \"ar_d1\" : [7],\n  \"ar_d2\" : [[8]],\n  \"ar_d3\" : [[[9]]]\n}\nHowever, in the case where variable d is 0, ar_d1 is empty, i.e., it contains no values, as is ar_d2, ar_d3, and the JSON representation is\n{ \"d\" : 0,\n  \"ar_d1\" : [ ],\n  \"ar_d2\" : [ ],\n  \"ar_d3\" : [ ]\n}",
+    "text": "Compiling a Stan program requires a modern C++ compiler and the GNU Make build utility (a.k.a. “gmake”). These vary by operating system.\n\n\nThe required C++ compiler is g++ 4.9 3. On most systems the GNU Make utility is pre-installed and is the default make utility. There is usually a pre-installed C++ compiler as well, however, it may not be new enough. To check, run commands:\ng++ --version\nmake --version\nIf these are at least at g++ version 4.9.3 or later and make version 3.81 or later, no additional installations are necessary. It may still be desirable to update the C++ compiler g++, because later versions are faster.\nTo install the latest version of these tools (or upgrade an older version), use the following commands or their equivalent for your distribution, install via the commands:\nsudo apt install g++\nsudo apt install make\nIf you can’t run sudo, you will need to ask your sysadmin or cluster administrator to install these tools for you.\n\n\n\nTo check if you already already have an appropriate toolchain installed, open the Terminal application and enter:\nclang++ --version\nmake --version\nIf either of these commands prints the message command not found, you will need to install Xcode’s command line tools.\nOpen the Terminal application and enter:\nxcode-select --install\nSelect “Install” in the window that opens.\nAfter the installation completes, you can double check that installation was successful by reopening the Terminal and running:\nclang++ --version\nmake --version\nYou can read more about Xcode on its site: https://developer.apple.com/xcode/\nWe don’t recommend trying to use the GNU C++ compiler, available via Homebrew, based on the number of reports of installation difficulties from Mac users on GitHub as well as the Stan forums.\n\n\n\nThe Windows toolchain consists of programs g++, the C++ compiler, and make, the GNU Make utility. To check if these are present, open a command shell [^1] and type:\ng++ --version\nmake --version\nCmdStan is known compatible with the RTools45 toolchain. The toolchain will require updating your PATH variable, See these instructions for details on changing the PATH if you are unfamiliar. The following instructions will assume that the default installation directory was used, so be sure to update the paths accordingly if you have chosen a different directory.\n\n\nAll required utilities (e.g., make, g++) for compiling and running CmdStan models on Windows are provided by the RTools45 toolchain from the R Project. Installation steps are provided below, and for more technical details on the toolchain refer to the R Project documentation.\nThe R Project provides RTools45 for both Intel/AMD 64-bit (x86_64) and ARM 64-bit (aarch64) systems. If you are unsure which to use, then you can check by going to the Windows Settings, selecting the ‘System’ menu and then the ‘About’ option. If the ‘System Type’ field lists ‘ARM-based processor’, then you should follow the ARM64 instructions below.\nNote that the toolchain is only available for 64-bit systems, and uses the new Universal C Runtime (UCRT). UCRT is only natively supported on Windows 10 and newer, older systems will require a Microsoft update.\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45\\usr\\bin\nC:\\rtools45\\x86_64-w64-mingw32.static.posix\\bin\n\n\n\nDownload the installer and complete the prompts for installation:\n\nRTools45 - ARM64\n\nNext, you need to add the toolchain directory to your PATH variable:\nC:\\rtools45-aarch64\\usr\\bin\nC:\\rtools45-aarch64\\aarch64-w64-mingw32.static.posix\\bin",
     "crumbs": [
-      "Appendices",
-      "JSON Format for CmdStan"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/log_prob_config.html",
-    "href": "cmdstan-guide/log_prob_config.html",
-    "title": "Extracting log probabilities and gradients for diagnostics",
+    "objectID": "cmdstan-guide/installation.html#gnu-make",
+    "href": "cmdstan-guide/installation.html#gnu-make",
+    "title": "CmdStan Installation",
     "section": "",
-    "text": "CmdStan can return the computed log probability and the gradient with respect to a set of parameters.\nThis is similar to the diagnose subcommand, but the output format differs and the results here are not compared with those from finite differences.\nNote: Startup and data initialization costs mean that this method is not an efficient way to calculate these quantities. It is provided only for convenience and should not be used for serious computation.\n\n\nThis method takes 3 arguments:\n\njacobian - Whether or not the Jacobian adjustment for constrained parameters should be included in the gradient. Default value is true (include adjustment).\nconstrained_params - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using JSON format. Alternatively, the input file can be set of draws in StanCSV format.\nunconstrained_params - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called params_r, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.\n\nOnly one of constrained_params and unconstrained_params can be specified.\nFor more on the differences between constrained and unconstrained parameters, see the Stan reference manual section on variable transforms.\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the log_prob method is reported at the beginning of the output file as CSV comments.\nColumn headers, the first column is labelled lp__, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.\nValues which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).\n\nFor example, if we have a file called params.json:\n{\n    \"theta\" : 0.1\n}\nWe can run the example model:\n/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json\nThis yields\n# method = log_prob\n#   log_prob\n#     unconstrained_params =  (Default)\n#     constrained_params = params.json\n#     jacobian = true (Default)\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2390820139 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlp_,theta\n-7.856,1.8",
+    "text": "CmdStan relies on the GNU Make utility to build both the Stan model executables and the CmdStan tools.\nGNU Make builds executable programs and libraries from source code by reading files called Makefiles which specify how to derive the target program. A Makefile consists of a set of recursive rules where each rule specifies a target, its dependencies, and the specific operations required to build the target. Specifying dependencies for a target provides a way to control the build process so that targets which depend on other files will be updated as needed only when there are changes to those other files. Thus Make provides an efficient way to manage complex software.\nThe CmdStan Makefile is in the &lt;cmdstan-home&gt; directory and is named makefile. This is one of the default GNU Makefile names, which allows you to omit the -f makefile argument to the Make command. Because the CmdStan Makefile includes several other Makefiles, Make only works properly when invoked from the &lt;cmdstan-home&gt; directory; attempts to use this Makefile from another directory by specifying the full path to the file makefile won’t work. For example, trying to call Make from another directory by specifying the full path the the makefile results in the following set of error messages:\nmake -f ~/github/stan-dev/cmdstan/makefile\n/Users/mitzi/github/stan-dev/cmdstan/makefile:58: make/stanc: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:59: make/program: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:60: make/tests: No such file or directory\n/Users/mitzi/github/stan-dev/cmdstan/makefile:61: make/command: No such file or directory\nmake: *** No rule to make target `make/command'.  Stop.\nThe conda-forge cmdstan package provides a solution to this problem via cmdstan_model command which lets you run the CmdStan makefile from anywhere to compile a Stan model.\nMakefile syntax allows general pattern rules based on file suffixes. Stan programs must be stored in files with suffix .stan; the CmdStan makefile rules specify how to transform the Stan source code into a binary executable. For example, to compile the Stan program my_program.stan in directory ../my_dir/, the make target is ../my_dir/my_program or ../my_dir/my_program.exe (on Windows).\nTo call Make, you invoke the utility name, make, followed by, in order:\n\nzero or more Make program options, then specify any Make variables as a series of\nzero of more Make variables, described below\nzero or more target names; the set of names is determined by the Makefile rules.\n\nmake &lt;flags&gt; &lt;variables&gt; &lt;targets&gt;\nMakefile Variables\nMake targets can be preceded by any number of Makefile variable name=value pairs. For example, to compile ../my_dir/my_program.stan for an OpenCL (GPU) machine, set the makefile variable STAN_OPENCL to TRUE:\n&gt; make STAN_OPENCL=TRUE ../my_dir/my_program\nMakefile variables can also be set by creating a file named local in the CmdStan make subdirectory which contains a list of &lt;VARIABLE&gt;=&lt;VALUE&gt; pairs, one per line. For example, to get the same effect as the above command every time, you would put the line STAN_OPENCL=TRUE into the file &lt;cmdstan_home&gt;/make/local.\nThe complete set of Makefile variables can be found in file &lt;cmdstan-home&gt;/cmdstan/stan/lib/stan_math/make/compiler_flags.\nMake Targets\nWhen invoked without any arguments at all, Make prints a help message:\n&gt; make\n--------------------------------------------------------------------------------\nCmdStan v2.33.1 help\n\n  Build CmdStan utilities:\n    &gt; make build\n\n    This target will:\n    1. Install the Stan compiler bin/stanc from stanc3 binaries.\n    2. Build the print utility bin/print (deprecated; will be removed in v3.0)\n    3. Build the stansummary utility bin/stansummary\n    4. Build the diagnose utility bin/diagnose\n    5. Build all libraries and object files compile and link an executable Stan program\n\n    Note: to build using multiple cores, use the -j option to make, e.g.,\n    for 4 cores:\n    &gt; make build -j4\n\n\n  Build a Stan program:\n\n    Given a Stan program at foo/bar.stan, build an executable by typing:\n    &gt; make foo/bar\n\n    This target will:\n    1. Install the Stan compiler (bin/stanc), as needed.\n    2. Use the Stan compiler to generate C++ code, foo/bar.hpp.\n    3. Compile the C++ code using cc . to generate foo/bar\n\n  Additional make options:\n    STANCFLAGS: defaults to \"\". These are extra options passed to bin/stanc\n      when generating C++ code. If you want to allow undefined functions in the\n      Stan program, either add this to make/local or the command line:\n          STANCFLAGS = --allow_undefined\n    USER_HEADER: when STANCFLAGS has --allow_undefined, this is the name of the\n      header file that is included. This defaults to \"user_header.hpp\" in the\n      directory of the Stan program.\n    STANC3_VERSION: When set, uses that tagged version specified; otherwise, downloads\n      the nightly version.\n    STAN_CPP_OPTIMS: Turns on additonal compiler flags for performance.\n    STAN_NO_RANGE_CHECKS: Removes the range checks from the model for performance.\n\n\n  Example - bernoulli model: examples/bernoulli/bernoulli.stan\n\n    1. Build the model:\n       &gt; make examples/bernoulli/bernoulli\n    2. Run the sampling algorithm given the model and data:\n       &gt; examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.R\n    3. Look at the posterior sample:\n       &gt; bin/stansummary output.csv\n\n\n  Clean CmdStan:\n\n    Remove the built CmdStan tools:\n    &gt; make clean-all\n\n--------------------------------------------------------------------------------",
     "crumbs": [
-      "Running CmdStan",
-      "Extracting log probabilities and gradients for diagnostics"
+      "Getting Started",
+      "CmdStan Installation"
     ]
   },
   {
-    "objectID": "cmdstan-guide/log_prob_config.html#configuration",
-    "href": "cmdstan-guide/log_prob_config.html#configuration",
-    "title": "Extracting log probabilities and gradients for diagnostics",
+    "objectID": "cmdstan-guide/generate_quantities_config.html",
+    "href": "cmdstan-guide/generate_quantities_config.html",
+    "title": "Generating Quantities of Interest from a Fitted Model",
     "section": "",
-    "text": "This method takes 3 arguments:\n\njacobian - Whether or not the Jacobian adjustment for constrained parameters should be included in the gradient. Default value is true (include adjustment).\nconstrained_params - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using JSON format. Alternatively, the input file can be set of draws in StanCSV format.\nunconstrained_params - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called params_r, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.\n\nOnly one of constrained_params and unconstrained_params can be specified.\nFor more on the differences between constrained and unconstrained parameters, see the Stan reference manual section on variable transforms.",
+    "text": "The generate_quantities method allows you to generate additional quantities of interest from a fitted model without re-running the sampler. Instead, you write a modified version of the original Stan program and add a generated quantities block or modify the existing one which specifies how to compute the new quantities of interest. Running the generate_quantities method on the new program together with sampler outputs (i.e., a set of draws) from the fitted model runs the generated quantities block of the new program using the the existing sample by plugging in the per-draw parameter estimates for the computations in the generated quantities block.\nThis method requires sub-argument fitted_params which takes as its value an existing Stan CSV file that contains a parameter values from an equivalent model, i.e., a model with the same parameters block, conditioned on the same data.\nThe generated quantities block computes quantities of interest (QOIs) based on the data, transformed data, parameters, and transformed parameters. It can be used to:\n\ngenerate simulated data for model testing by forward sampling\ngenerate predictions for new data\ncalculate posterior event probabilities, including multiple comparisons, sign tests, etc.\ncalculate posterior expectations\ntransform parameters for reporting\napply full Bayesian decision theory\ncalculate log likelihoods, deviances, etc. for model comparison\n\nFor an overview of the uses of this feature, see the Stan User’s Guide section on Stand-alone generated quantities and ongoing prediction.\n\n\nTo illustrate how this works we use the generate_quantities method to do posterior predictive checks using the estimate of theta given the example bernoulli model and data, following the posterior predictive simulation procedure in the Stan User’s Guide.\nWe write a program bernoulli_ppc.stan which contains the following generated quantities block, with comments to explain the procedure:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nThe rest of the program is the same as in bernoulli.stan.\nThe generate_method requires the sub-argument fitted_params which takes as its value the name of a Stan CSV file. The per-draw parameter values from the fitted_params file will be used to run the generated quantities block.\nIf we run the bernoulli.stan program for a single chain to generate a sample in file bernoulli_fit.csv:\n&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv\nThen we can run the bernoulli_ppc.stan to carry out the posterior predictive checks:\n&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \\\n                  data file=bernoulli.data.json \\\n                  output file=bernoulli_ppc.csv\nThe output file bernoulli_ppc.csv contains only the values for the variables declared in the generated quantities block, i.e., theta_rep and the elements of y_sim:\n# model = bernoulli_ppc_model\n# method = generate_quantities\n#   generate_quantities\n#     fitted_params = bernoulli_fit.csv\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2983956445 (Default)\n# output\n#   file = output.csv (Default)\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep\n1,1,1,0,0,0,1,1,0,1,0.6\n1,1,0,1,0,0,1,0,1,0,0.5\n1,0,1,1,1,1,1,1,0,1,0.8\n0,1,0,1,0,1,0,1,0,0,0.4\n1,0,0,0,0,0,0,0,0,0,0.1\n0,0,0,0,0,1,1,1,0,0,0.3\n0,0,1,0,1,0,0,0,0,0,0.2\n1,0,1,0,1,1,0,1,1,0,0.6\n...\nGiven the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  real&lt;lower=0, upper=1&gt; theta_cp = theta;\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nNow the output is slightly more interpretable: theta_cp is the same as the theta used to generate the values y_sim[1] through y_sim[1]. Comparing columns theta_cp and theta_rep allows us to see how the uncertainty in our estimate of theta is carried forward into our predictions:\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep\n0,1,1,0,1,0,0,1,1,0,0.545679,0.5\n1,1,1,1,1,1,0,1,1,0,0.527164,0.8\n1,1,1,1,0,1,1,1,1,0,0.529116,0.8\n1,0,1,1,1,1,0,0,1,0,0.478844,0.6\n0,1,0,0,0,0,1,0,1,0,0.238793,0.3\n0,0,0,0,0,1,1,0,0,0,0.258294,0.2\n1,1,1,0,0,0,0,0,0,0,0.258465,0.3\n\n\n\nThe fitted_params file must be a Stan CSV file; attempts to use a regular CSV file will result an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe fitted_params file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe parameter values of the fitted_params are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in bernoulli_fit.csv so that the value of theta is outside the declared bounds real&lt;lower=0, upper=1&gt;, the program will return the following error message:\nException: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \\\n(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)",
     "crumbs": [
       "Running CmdStan",
-      "Extracting log probabilities and gradients for diagnostics"
+      "Generating Quantities of Interest from a Fitted Model"
     ]
   },
   {
-    "objectID": "cmdstan-guide/log_prob_config.html#csv-output",
-    "href": "cmdstan-guide/log_prob_config.html#csv-output",
-    "title": "Extracting log probabilities and gradients for diagnostics",
+    "objectID": "cmdstan-guide/generate_quantities_config.html#example",
+    "href": "cmdstan-guide/generate_quantities_config.html#example",
+    "title": "Generating Quantities of Interest from a Fitted Model",
     "section": "",
-    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the log_prob method is reported at the beginning of the output file as CSV comments.\nColumn headers, the first column is labelled lp__, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.\nValues which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).\n\nFor example, if we have a file called params.json:\n{\n    \"theta\" : 0.1\n}\nWe can run the example model:\n/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json\nThis yields\n# method = log_prob\n#   log_prob\n#     unconstrained_params =  (Default)\n#     constrained_params = params.json\n#     jacobian = true (Default)\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2390820139 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlp_,theta\n-7.856,1.8",
+    "text": "To illustrate how this works we use the generate_quantities method to do posterior predictive checks using the estimate of theta given the example bernoulli model and data, following the posterior predictive simulation procedure in the Stan User’s Guide.\nWe write a program bernoulli_ppc.stan which contains the following generated quantities block, with comments to explain the procedure:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nThe rest of the program is the same as in bernoulli.stan.\nThe generate_method requires the sub-argument fitted_params which takes as its value the name of a Stan CSV file. The per-draw parameter values from the fitted_params file will be used to run the generated quantities block.\nIf we run the bernoulli.stan program for a single chain to generate a sample in file bernoulli_fit.csv:\n&gt; ./bernoulli sample data file=bernoulli.data.json output file=bernoulli_fit.csv\nThen we can run the bernoulli_ppc.stan to carry out the posterior predictive checks:\n&gt; ./bernoulli_ppc generate_quantities fitted_params=bernoulli_fit.csv \\\n                  data file=bernoulli.data.json \\\n                  output file=bernoulli_ppc.csv\nThe output file bernoulli_ppc.csv contains only the values for the variables declared in the generated quantities block, i.e., theta_rep and the elements of y_sim:\n# model = bernoulli_ppc_model\n# method = generate_quantities\n#   generate_quantities\n#     fitted_params = bernoulli_fit.csv\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2983956445 (Default)\n# output\n#   file = output.csv (Default)\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_rep\n1,1,1,0,0,0,1,1,0,1,0.6\n1,1,0,1,0,0,1,0,1,0,0.5\n1,0,1,1,1,1,1,1,0,1,0.8\n0,1,0,1,0,1,0,1,0,0,0.4\n1,0,0,0,0,0,0,0,0,0,0.1\n0,0,0,0,0,1,1,1,0,0,0.3\n0,0,1,0,1,0,0,0,0,0,0.2\n1,0,1,0,1,1,0,1,1,0,0.6\n...\nGiven the current implementation, to see the fitted parameter values for each draw, create a copy variable in the generated quantities block, e.g.:\ngenerated quantities {\n  array[N] int y_sim;\n  // use current estimate of theta to generate new sample\n  for (n in 1:N) {\n    y_sim[n] = bernoulli_rng(theta);\n  }\n  real&lt;lower=0, upper=1&gt; theta_cp = theta;\n  // estimate theta_rep from new sample\n  real&lt;lower=0, upper=1&gt; theta_rep = sum(y_sim) * 1.0 / N;\n}\nNow the output is slightly more interpretable: theta_cp is the same as the theta used to generate the values y_sim[1] through y_sim[1]. Comparing columns theta_cp and theta_rep allows us to see how the uncertainty in our estimate of theta is carried forward into our predictions:\ny_sim.1,y_sim.2,y_sim.3,y_sim.4,y_sim.5,y_sim.6,y_sim.7,y_sim.8,y_sim.9,y_sim.10,theta_cp,theta_rep\n0,1,1,0,1,0,0,1,1,0,0.545679,0.5\n1,1,1,1,1,1,0,1,1,0,0.527164,0.8\n1,1,1,1,0,1,1,1,1,0,0.529116,0.8\n1,0,1,1,1,1,0,0,1,0,0.478844,0.6\n0,1,0,0,0,0,1,0,1,0,0.238793,0.3\n0,0,0,0,0,1,1,0,0,0,0.258294,0.2\n1,1,1,0,0,0,0,0,0,0,0.258465,0.3",
     "crumbs": [
       "Running CmdStan",
-      "Extracting log probabilities and gradients for diagnostics"
+      "Generating Quantities of Interest from a Fitted Model"
     ]
   },
   {
-    "objectID": "cmdstan-guide/optimize_config.html",
-    "href": "cmdstan-guide/optimize_config.html",
-    "title": "Optimization",
+    "objectID": "cmdstan-guide/generate_quantities_config.html#errors",
+    "href": "cmdstan-guide/generate_quantities_config.html#errors",
+    "title": "Generating Quantities of Interest from a Fitted Model",
     "section": "",
-    "text": "The CmdStan executable can run Stan’s optimization algorithms, which provide a deterministic method to find the posterior mode. If the posterior is not convex, there is no guarantee Stan will be able to find the global optimum as opposed to a local optimum of log probability.\nThe full set of configuration options available for the optimize method is available by using the optimize help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the optimizer console output and in the output CSV file’s comments.\nThe executable does not need to be recompiled in order to switch from sampling to optimization, and the data input format is the same. The following is a minimal call to Stan’s optimizer using defaults for everything but the location of the data file.\n&gt; ./bernoulli optimize data file=bernoulli.data.json\nExecuting this command prints both output to the console and to a CSV file.\nThe first part of the console output reports on the configuration used. The above command uses all default configurations, therefore the optimizer used is the L-BFGS optimizer and its default initial stepsize and tolerances for monitoring convergence:\n ./bernoulli optimize data file=bernoulli.data.json\nmethod = optimize\n  optimize\n    algorithm = lbfgs (Default)\n      lbfgs\n        init_alpha = 0.001 (Default)\n        tol_obj = 1e-12 (Default)\n        tol_rel_obj = 10000 (Default)\n        tol_grad = 1e-08 (Default)\n        tol_rel_grad = 1e+07 (Default)\n        tol_param = 1e-08 (Default)\n        history_size = 5 (Default)\n    jacobian = false (Default)\n    iter = 2000 (Default)\n    save_iterations = false (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 87122538 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nThe second part of the output indicates how well the algorithm fared, here converging and terminating normally. The numbers reported indicate that it took 5 iterations and 8 gradient evaluations. This is, not surprisingly, far fewer iterations than required for sampling; even fewer iterations would be used with less stringent user-specified convergence tolerances. The alpha value is for step size used. In the final state the change in parameters was roughly \\(0.002\\) and the length of the gradient roughly 3e-05 (\\(0.00003\\)).\nInitial log joint probability = -6.85653\n    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes\n       5      -5.00402    0.00184936   3.35074e-05           1           1        8\nOptimization terminated normally:\n  Convergence detected: relative gradient magnitude is below tolerance\nThe output from optimization is written into the file output.csv by default. The output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as comment lines:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = optimize\n#   optimize\n#     algorithm = lbfgs (Default)\n#       lbfgs\n#         init_alpha = 0.001 (Default)\n#         tol_obj = 1e-12 (Default)\n#         tol_rel_obj = 10000 (Default)\n#         tol_grad = 1e-08 (Default)\n#         tol_rel_grad = 1e+07 (Default)\n#         tol_param = 1e-08 (Default)\n#         history_size = 5 (Default)\n#     jacobian = false (Default)\n#     iter = 2000 (Default)\n#     save_iterations = false (Default)\nFollowing the config information are two lines of output, the CSV headers and the recorded values:\nlp__,theta\n-5.00402,0.200003\nNote that everything is a comment other than a line for the header, and a line for the values. Here, the header indicates the unnormalized log probability with lp__ and the model parameter theta. The maximum log probability is -5.0 and the posterior mode for theta is 0.20. The mode exactly matches what we would expect from the data. Because the prior was uniform, the result 0.20 represents the maximum likelihood estimate (MLE) for the very simple Bernoulli model. Note that no uncertainty is reported.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value is false, i.e., intermediate iterations are not saved to the output file. Running the optimizer with save_iterations=true writes both the initial log joint probability and values for all iterations to the output CSV file.\nRunning the example model with option save_iterations=true, i.e., the command\n&gt; ./bernoulli optimize save_iterations=1 data file=bernoulli.data.json\nproduces CSV file output rows:\nlp__,theta\n-6.85653,0.493689\n-6.10128,0.420936\n-5.02953,0.22956\n-5.00517,0.206107\n-5.00403,0.200299\n-5.00402,0.200003\n\n\nIf there are constrained parameters, Stan makes a transformation to an unconstrained space and runs the optimization algorithm in the unconstrained space.\nThe jacobian argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of inverse parameter transforms.\nIf the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.\nIf the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\n\n\n\nThe algorithm argument specifies the optimization algorithm. This argument takes one of the following three values:\n\nlbfgs A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.\nbfgs A quasi-Newton optimizer.\nnewton A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.\n\nSee the Stan Reference Manual’s Optimization chapter for a description of these algorithms.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value isfalse`, i.e., intermediate iterations are not saved to the output file.\n\n\n\nFor both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the BFGS and L-BFGS configuration section for details on the convergence tests.\nBoth BFGS and L-BFGS have the following configuration arguments:\n\ninit_alpha - The initial step size parameter. Must be a positive real number. Default value is \\(0.001\\)\ntol_obj - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is \\(1^{-12}\\).\ntol_rel_obj - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is \\(1^{4}\\).\ntol_grad - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is \\(1^{-8}\\).\ntol_rel_grad - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is \\(1^{7}\\).\ntol_param - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is \\(1^{-8}\\).\n\nThe init_alpha argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set init_alpha to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.\nIn addition to the above, the L-BFGS algorithm has argument history_size which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (\\(5\\)-\\(10\\)) are sufficient; the default value is \\(5\\).\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.\n\n\n\nThere are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.",
+    "text": "The fitted_params file must be a Stan CSV file; attempts to use a regular CSV file will result an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe fitted_params file must contain columns corresponding to legal values for all parameters defined in the model. If any parameters are missing, the program will exit with an error message of the form:\nError reading fitted param names from sample csv file &lt;filename.csv&gt;\nThe parameter values of the fitted_params are on the constrained scale and must obey all constraints. For example, if we modify the contents of the first reported draw in bernoulli_fit.csv so that the value of theta is outside the declared bounds real&lt;lower=0, upper=1&gt;, the program will return the following error message:\nException: lub_free: Bounded variable is 1.21397, but must be in the interval [0, 1] \\\n(in 'bernoulli_ppc.stan', line 5, column 2 to column 30)",
     "crumbs": [
       "Running CmdStan",
-      "Optimization"
+      "Generating Quantities of Interest from a Fitted Model"
     ]
   },
   {
-    "objectID": "cmdstan-guide/optimize_config.html#jacobian-adjustments",
-    "href": "cmdstan-guide/optimize_config.html#jacobian-adjustments",
-    "title": "Optimization",
+    "objectID": "cmdstan-guide/example_model_data.html",
+    "href": "cmdstan-guide/example_model_data.html",
+    "title": "Example Model and Data",
     "section": "",
-    "text": "If there are constrained parameters, Stan makes a transformation to an unconstrained space and runs the optimization algorithm in the unconstrained space.\nThe jacobian argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of inverse parameter transforms.\nIf the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.\nIf the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.",
+    "text": "The following is a simple, complete Stan program for a Bernoulli model of binary data.1 The model assumes the binary observed data y[1],...,y[N] are i.i.d. with Bernoulli chance-of-success theta.\ndata { \n  int&lt;lower=0&gt; N; \n  array[N] int&lt;lower=0, upper=1&gt; y;\n} \nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n} \nmodel {\n  theta ~ beta(1, 1);  // uniform prior on interval 0,1\n  y ~ bernoulli(theta);\n}\nThe input data file contains definitions for the two variables N and y which are specified in the data block of program bernoulli.stan (above).\nA data set of N=10 observations is included in the example Bernoulli model directory in both JSON notation and Rdump data format where 8 out of 10 trials had outcome 0 (failure) and 2 trials had outcome 1 (success). In JSON, this data is:\n{\n    \"N\" : 10,\n    \"y\" : [0,1,0,0,0,0,0,0,0,1]\n}",
     "crumbs": [
-      "Running CmdStan",
-      "Optimization"
+      "Getting Started",
+      "Example Model and Data"
     ]
   },
   {
-    "objectID": "cmdstan-guide/optimize_config.html#optimization-algorithms",
-    "href": "cmdstan-guide/optimize_config.html#optimization-algorithms",
-    "title": "Optimization",
-    "section": "",
-    "text": "The algorithm argument specifies the optimization algorithm. This argument takes one of the following three values:\n\nlbfgs A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.\nbfgs A quasi-Newton optimizer.\nnewton A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.\n\nSee the Stan Reference Manual’s Optimization chapter for a description of these algorithms.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value isfalse`, i.e., intermediate iterations are not saved to the output file.",
+    "objectID": "cmdstan-guide/example_model_data.html#footnotes",
+    "href": "cmdstan-guide/example_model_data.html#footnotes",
+    "title": "Example Model and Data",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe model is available with the CmdStan distribution at the path &lt;cmdstan-home&gt;/examples/bernoulli/bernoulli.stan.↩︎",
     "crumbs": [
-      "Running CmdStan",
-      "Optimization"
+      "Getting Started",
+      "Example Model and Data"
     ]
   },
   {
-    "objectID": "cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers",
-    "href": "cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers",
-    "title": "Optimization",
+    "objectID": "cmdstan-guide/diagnose_config.html",
+    "href": "cmdstan-guide/diagnose_config.html",
+    "title": "Diagnosing HMC by Comparison of Gradients",
     "section": "",
-    "text": "For both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the BFGS and L-BFGS configuration section for details on the convergence tests.\nBoth BFGS and L-BFGS have the following configuration arguments:\n\ninit_alpha - The initial step size parameter. Must be a positive real number. Default value is \\(0.001\\)\ntol_obj - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is \\(1^{-12}\\).\ntol_rel_obj - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is \\(1^{4}\\).\ntol_grad - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is \\(1^{-8}\\).\ntol_rel_grad - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is \\(1^{7}\\).\ntol_param - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is \\(1^{-8}\\).\n\nThe init_alpha argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set init_alpha to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.\nIn addition to the above, the L-BFGS algorithm has argument history_size which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (\\(5\\)-\\(10\\)) are sufficient; the default value is \\(5\\).\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.",
+    "text": "Diagnosing HMC by Comparison of Gradients\nCmdStan has a basic diagnostic feature that will calculate the gradients of the initial state and compare them with gradients calculated by finite differences. Discrepancies between the two indicate that there is a problem with the model or initial states or else there is a bug in Stan.\nTo allow for the possibility of adding other kinds of diagnostic tests, the diagnose method argument configuration has subargument test which currently only takes value gradient. There are two available gradient test configuration arguments:\n\nepsilon - The finite difference step size. Must be a positive real number. Default value is \\(1^{-6}\\)\nerror - The error threshold. Must be a positive real number. Default value is \\(1^{-6}\\)\n\nTo run on the different platforms with the default configuration, use one of the following.\nMac OS and Linux\n&gt; ./my_model diagnose data file=my_data\nWindows\n&gt; my_model diagnose data file=my_data\nTo relax the test threshold, specify the error argument as follows:\n&gt; ./my_model diagnose test=gradient error=0.0001 data file=my_data\nTo see how this works, we run diagnostics on the example bernoulli model:\n&gt; ./bernoulli diagnose data file=bernoulli.data.R\nExecuting this command prints output to the console and as a series of comment lines to the output csv file. The console output is:\nmethod = diagnose\n  diagnose\n    test = gradient (Default)\n      gradient\n        epsilon = 9.9999999999999995e-07 (Default)\n        error = 9.9999999999999995e-07 (Default)\nid = 0 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2152196153 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n\nTEST GRADIENT MODE\n\n Log probability=-8.42814\n\n param idx           value           model     finite diff           error\n         0       0.0361376         -3.1084         -3.1084    -2.37554e-10\nThe same information is printed to the output file as csv comments, i.e., each line is prefixed with a pound sign #.\n\n\n\n\n Back to top",
     "crumbs": [
       "Running CmdStan",
-      "Optimization"
+      "Diagnosing HMC by Comparison of Gradients"
     ]
   },
   {
-    "objectID": "cmdstan-guide/optimize_config.html#the-newton-optimizer",
-    "href": "cmdstan-guide/optimize_config.html#the-newton-optimizer",
-    "title": "Optimization",
+    "objectID": "cmdstan-guide/command_line_options.html",
+    "href": "cmdstan-guide/command_line_options.html",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "There are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.",
+    "text": "A CmdStan executable is built from the Stan model concept and the CmdStan command line parser. The command line argument syntax consists of sets of keywords and keyword-value pairs. Arguments are grouped by the following keywords:\n\nmethod - specifies the kind of inference done on the model. Each kind of inference requires further configuration via sub-arguments. The method argument is required. It can be specified overtly as the a keyword-value pair method=&lt;inference&gt; or implicitly as one of the following:\n\nsample - obtain a sample (set of draws) from the posterior using HMC\noptimize - penalized maximum likelihood estimation\npathfinder - quasi-Newton variational inference\nvariational - automatic differentation variational inference (ADVI)\nlaplace - sample from a normal approximation centered at the mode\ngenerate_quantities - run model’s generated quantities block on existing sample to obtain new quantities of interest.\nlog_prob - compute the log probability and gradient of the model for one set of parameters.\ndiagnose - compute and compare sampler gradient calculations to finite differences.\n\ndata - specifies the input data file, if any.\noutput - specifies program outputs, both disk files and terminal window outputs.\ninit - specifies initial values for the model parameters, if any.\nrandom - specifies the seed for the pseudo-random number.\n\nThe remainder of this chapter covers the general configuration options used for all processing. The following chapters cover the per-inference configuration options.\n\n\nThe values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:\ndata file=&lt;filepath&gt;\nThe keyword data must be followed directly by the keyword-value pair file=&lt;filepath&gt;. If the model doesn’t declare any data variables, this argument is ignored.\nThe input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model bernoulli.stan defines two data variables N and y. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.\nFor example if the input data file doesn’t include the definition for variable y, the executable exits with the following message:\nException: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)\n\n\n\nThe output keyword is used to specify non-default options for output files and messages written to the terminal window. The output keyword takes several keyword-value pair sub-arguments.\nThe keyword value pair file=&lt;filepath&gt; specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named output.csv in the current working directory.\nThe keyword value pair diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms sample and variational.\nThe keyword value pair refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.\nThe keyword value pair sig_figs=&lt;int&gt; specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8.   Note: increasing sig_figs above the default will increase the size of the output CSV files accordingly.\nThe keyword value pair profile_file=&lt;filepath&gt; specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and profile_file is unspecified, the profiling data is written to a file named profile.csv in the current working directory.\nThe keyword value pair save_cmdstan_config=&lt;boolean&gt; specifies whether to save the configuration options used to run the program to a file named &lt;output file&gt;_config.json alongside the other output files. The default value is false, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.\n\n\n\nInitialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range \\([-2, 2]\\). These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\nThe initialization argument is specified as keyword-value pair with keyword init. The value can be one of the following:\n\npositive real number \\(x\\). All parameters will be initialized to random draws from a uniform distribution over the range \\([-x, x]\\).\n\\(0\\) - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: \\(0\\) for unconstrained parameters; \\(1\\) for parameters constrained to be positive; \\(0.5\\) for variables to constrained to lie between \\(0\\) and \\(1\\); a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.\nfilepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range \\([-2, 2]\\).\n\n\n\n\nThe random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nThe syntax for the random seed argument is:\nrandom seed=&lt;int&gt;\nThe keyword random must be followed directly by the keyword-value pair seed=&lt;int&gt;.\n\n\n\nThe chain identifier argument is used in conjunction with the random seed argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.\nThe syntax for the random seed argument is:\nid=&lt;int&gt;\nThe default value is 1.\nWhen running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.\nFor complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the Stan Reference Manual Reproducibility chapter for further details.\n\n\n\nCmdStan provides a help and help-all mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword help:\n./bernoulli help\n\n\n\nCmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.\nAll program executables provide a return code between 0 and 255:\n\n0 - Program ran to termination as expected.\nvalue in range [1 : 125] - Method invoked could not run due to problems with model or data.\nvalue &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g. return code 139 results from termination signal 11 (segmentation violation).\n\nA non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.",
     "crumbs": [
       "Running CmdStan",
-      "Optimization"
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/pathfinder_config.html",
-    "href": "cmdstan-guide/pathfinder_config.html",
-    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "objectID": "cmdstan-guide/command_line_options.html#input-data-argument",
+    "href": "cmdstan-guide/command_line_options.html#input-data-argument",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "The CmdStan method pathfinder uses the Pathfinder algorithm of Zhang et al. (2022), which is further described in the Stan Reference Manual.\nA single run of the Pathfinder algorithm generates a set of approximate draws. Inference is improved by running multiple Pathfinder instances and using Pareto-smoothed importance resampling (PSIS) of the resulting sets of draws. This better matches non-normal target densities and also eliminates minor modes.\nThe pathfinder method runs multi-path Pathfinder by default, which returns a PSIS sample over the draws from several individual (“single-path”) Pathfinder runs. Argument num_paths specifies the number of single-path Pathfinders, the default is \\(4\\). If num_paths is set to 1, then only one individual Pathfinder is run without the PSIS reweighting of the sample.\nThe full set of configuration options available for the pathfinder method is available by using the pathfinder help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.\nThe following is a minimal call to the Pathfinder algorithm using defaults for everything but the location of the data file.\n&gt; ./bernoulli pathfinder data file=bernoulli.data.R\nExecuting this command prints output both to the console and to csv files.\nThe first part of the console output reports on the configuration used.\nmethod = pathfinder\n  pathfinder\n    init_alpha = 0.001 (Default)\n    tol_obj = 1e-12 (Default)\n    tol_rel_obj = 10000 (Default)\n    tol_grad = 1e-08 (Default)\n    tol_rel_grad = 1e+07 (Default)\n    tol_param = 1e-08 (Default)\n    history_size = 5 (Default)\n    num_psis_draws = 1000 (Default)\n    num_paths = 4 (Default)\n    save_single_paths = false (Default)\n    psis_resample = true (Default)\n    calculate_lp = true (Default)\n    max_lbfgs_iters = 1000 (Default)\n    num_draws = 1000 (Default)\n    num_elbo_draws = 25 (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2790476610 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nThe rest of the output describes the progression of the algorithm.\nBy default, the Pathfinder algorithm runs 4 single-path Pathfinders in parallel, then uses importance resampling on the set of returned draws to produce the specified number of draws.\nPath [1] :Initial log joint density = -11.543343\nPath [1] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      1.070e-03   1.707e-05    1.000e+00  1.000e+00       126 -6.220e+00 -6.220e+00\nPath [1] :Best Iter: [5] ELBO (-6.219833) evaluations: (126)\nPath [2] :Initial log joint density = -7.443345\nPath [2] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      9.936e-05   3.738e-07    1.000e+00  1.000e+00       126 -6.164e+00 -6.164e+00\nPath [2] :Best Iter: [5] ELBO (-6.164015) evaluations: (126)\nPath [3] :Initial log joint density = -18.986308\nPath [3] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      2.996e-04   4.018e-06    1.000e+00  1.000e+00       126 -6.201e+00 -6.201e+00\nPath [3] :Best Iter: [5] ELBO (-6.200559) evaluations: (126)\nPath [4] :Initial log joint density = -8.304453\nPath [4] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      2.814e-04   2.034e-06    1.000e+00  1.000e+00       126 -6.221e+00 -6.221e+00\nPath [4] :Best Iter: [3] ELBO (-6.161276) evaluations: (126)\nTotal log probability function evaluations:8404\n\n\n\nnum_psis_draws - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is \\(1000\\).\nnum_paths - Number of single pathfinders. Must be a positive integer. Default value is \\(4\\).\nsave_single_paths - When true, save outputs from single pathfinders. Valid values: [true, false]. Default is false.\nmax_lbfgs_iters - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is \\(1000\\).\nnum_draws - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is \\(1000\\). Can differ from num_psis_draws.\nnum_elbo_draws - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is \\(25\\).\npsis_resample - If true, perform PSIS resampling on draws returned from individual pathfinders. If false, returns all num_paths * num_draws draws from the individual pathfinders. Valid values: [true, false]. Default is true.\ncalculate_lp - If true, log probabilities of the approximate draws are calculated and returned with the output. If false, each pathfinder will only calculate the lp values needed for the ELBO calculation. If false, PSIS resampling cannot be performed and the algorithm returns num_paths * num_draws samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: [true, false]. Default is true.\n\n\n\n\nArguments init_alpha through history_size are the full set of arguments to the L-BFGS optimizer and have the same defaults for optimization.\n\n\n\nBy default, the pathfinder method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a StanCSV file.\nThe CSV files have the following structure:\nThe initial CSV comment rows contain the complete set of CmdStan configuration options.\n...\n# method = pathfinder\n#   pathfinder\n#     init_alpha = 0.001 (Default)\n#     tol_obj = 9.9999999999999998e-13 (Default)\n#     tol_rel_obj = 10000 (Default)\n#     tol_grad = 1e-08 (Default)\n#     tol_rel_grad = 10000000 (Default)\n#     tol_param = 1e-08 (Default)\n#     history_size = 5 (Default)\n#     num_psis_draws = 1000 (Default)\n#     num_paths = 4 (Default)\n#     psis_resample = 1 (Default)\n#     calculate_lp = 1 (Default)\n#     save_single_paths = 0 (Default)\n#     max_lbfgs_iters = 1000 (Default)\n#     num_draws = 1000 (Default)\n#     num_elbo_draws = 25 (Default)\n...\nNext is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs lp_approx__, the log density in the approximating distribution, and lp__, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.\nlp_approx__,lp__,theta\n-2.4973, -8.2951, 0.0811852\n-0.87445, -7.06526, 0.160207\n-0.812285, -7.07124, 0.35819\n...\nThe final lines are comment lines which give timing information.\n# Elapsed Time: 0.016000 seconds (Pathfinders)\n#               0.003000 seconds (PSIS)\n#               0.019000 seconds (Total)\nPathfinder provides option save_single_paths which will save output from the single-path Pathfinder runs.\n\n\n\nThe boolean option save_single_paths is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When save_single_paths is true, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name output.csv and default number of runs (4), the resulting CSV files are\noutput.csv\noutput_path_1.csv\noutput_path_1.json\noutput_path_2.csv\noutput_path_2.json\noutput_path_3.csv\noutput_path_3.json\noutput_path_4.csv\noutput_path_4.json\nThe individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.\nTo see how this works, we run Pathfinder on the centered-parameterization of the eight-schools model, where the posterior distribution has a funnel shape:\n&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json\nEach JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.\n{\n  \"0\": {\n    \"iter\": 0,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]\n  },\n  \"1\": {\n    \"iter\": 1,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],\n    \"history_size\": 1,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],\n    \"logDetCholHk\": -2.6839,\n    \"L_approx\": [[-0.0630456, -0.0187959], [0, 1.08328]],\n    \"Qk\": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],\n    \"alpha\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n    \"full\": false,\n    \"lbfgs_note\": \"\"\n  },\n  ...,\n  \"171\": {\n    \"iter\": 171,\n    \"unconstrained_parameters\": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],\n    \"grads\": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],\n    \"history_size\": 5,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],\n    \"logDetCholHk\": 299.023,\n    \"L_approx\": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],\n    \"Qk\": [],\n    \"alpha\": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],\n    \"full\": true,\n    \"lbfgs_note\": \"\"\n  },\n  \"172\": {\n    \"iter\": 172,\n    \"unconstrained_parameters\": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],\n    \"grads\": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],\n    \"history_size\": 5,\n    \"lbfgs_success\": false,\n    \"pathfinder_success\": false,\n    \"lbfgs_note\": \"\"\n  }\n}\nOption num_paths=1 runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments num_paths=1 save_single_paths=true creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.",
+    "text": "The values for all variables declared in the data block of the model are read in from an input data file in either JSON or Rdump format. The syntax for the input data argument is:\ndata file=&lt;filepath&gt;\nThe keyword data must be followed directly by the keyword-value pair file=&lt;filepath&gt;. If the model doesn’t declare any data variables, this argument is ignored.\nThe input data file must contain definitions for all data variables declared in the data block. If one or more data block variables are missing from the input data file, the program prints an error message to stderr and returns a non-zero return code. For example, the model bernoulli.stan defines two data variables N and y. If the input data file doesn’t include both variables, or if the data variable doesn’t match the declared type and dimensions, the program will exit with an error message at the point where it first encounters missing data.\nFor example if the input data file doesn’t include the definition for variable y, the executable exits with the following message:\nException: variable does not exist; processing stage=data initialization; variable name=y; base type=int (in 'examples/bernoulli/bernoulli.stan', line 3, column 2 to column 28)",
     "crumbs": [
       "Running CmdStan",
-      "Pathfinder Method for Approximate Bayesian Inference"
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/pathfinder_config.html#pathfinder-configuration",
-    "href": "cmdstan-guide/pathfinder_config.html#pathfinder-configuration",
-    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "objectID": "cmdstan-guide/command_line_options.html#output-control.section",
+    "href": "cmdstan-guide/command_line_options.html#output-control.section",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "num_psis_draws - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is \\(1000\\).\nnum_paths - Number of single pathfinders. Must be a positive integer. Default value is \\(4\\).\nsave_single_paths - When true, save outputs from single pathfinders. Valid values: [true, false]. Default is false.\nmax_lbfgs_iters - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is \\(1000\\).\nnum_draws - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is \\(1000\\). Can differ from num_psis_draws.\nnum_elbo_draws - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is \\(25\\).\npsis_resample - If true, perform PSIS resampling on draws returned from individual pathfinders. If false, returns all num_paths * num_draws draws from the individual pathfinders. Valid values: [true, false]. Default is true.\ncalculate_lp - If true, log probabilities of the approximate draws are calculated and returned with the output. If false, each pathfinder will only calculate the lp values needed for the ELBO calculation. If false, PSIS resampling cannot be performed and the algorithm returns num_paths * num_draws samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: [true, false]. Default is true.",
+    "text": "The output keyword is used to specify non-default options for output files and messages written to the terminal window. The output keyword takes several keyword-value pair sub-arguments.\nThe keyword value pair file=&lt;filepath&gt; specifies the location of the Stan CSV output file. If unspecified, the output file is written to a file named output.csv in the current working directory.\nThe keyword value pair diagnostic_file=&lt;filepath&gt; specifies the location of the auxiliary output file. By default, no auxiliary output file is produced. This option is only valid for the iterative algorithms sample and variational.\nThe keyword value pair refresh=&lt;int&gt; specifies the number of iterations between progress messages written to the terminal window. The default value is 100 iterations.\nThe keyword value pair sig_figs=&lt;int&gt; specifies the number of significant digits for all numerical values in the output files. Allowable values are between 1 and 18, which is the maximum amount of precision available for 64-bit floating point arithmetic. The default value is 8.   Note: increasing sig_figs above the default will increase the size of the output CSV files accordingly.\nThe keyword value pair profile_file=&lt;filepath&gt; specifies the location of the output file for profiling data. If the model uses no profiling, the output profile file is not produced. If the model uses profiling and profile_file is unspecified, the profiling data is written to a file named profile.csv in the current working directory.\nThe keyword value pair save_cmdstan_config=&lt;boolean&gt; specifies whether to save the configuration options used to run the program to a file named &lt;output file&gt;_config.json alongside the other output files. The default value is false, which means the configuration file is not saved. The contents of this file are similar to the comments in the Stan CSV file, but should be more portable across versions and easier to parse.",
     "crumbs": [
       "Running CmdStan",
-      "Pathfinder Method for Approximate Bayesian Inference"
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/pathfinder_config.html#l-bfgs-configuration",
-    "href": "cmdstan-guide/pathfinder_config.html#l-bfgs-configuration",
-    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "objectID": "cmdstan-guide/command_line_options.html#initialize-model-parameters-argument",
+    "href": "cmdstan-guide/command_line_options.html#initialize-model-parameters-argument",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "Arguments init_alpha through history_size are the full set of arguments to the L-BFGS optimizer and have the same defaults for optimization.",
+    "text": "Initialization is only applied to parameters defined in the parameters block. By default, all parameters are initialized to random draws from a uniform distribution over the range \\([-2, 2]\\). These values are on the unconstrained scale, so must be inverse transformed back to satisfy the constraints declared for parameters. Because zero is chosen to be a reasonable default initial value for most parameters, the interval around zero provides a fairly diffuse starting point. For instance, unconstrained variables are initialized randomly in \\((-2, 2)\\), variables constrained to be positive are initialized roughly in \\((0.14, 7.4)\\), variables constrained to fall between 0 and 1 are initialized with values roughly in \\((0.12, 0.88)\\).\nThe initialization argument is specified as keyword-value pair with keyword init. The value can be one of the following:\n\npositive real number \\(x\\). All parameters will be initialized to random draws from a uniform distribution over the range \\([-x, x]\\).\n\\(0\\) - All parameters will be initialized to zero values on the unconstrained scale. The transforms are arranged in such a way that zero initialization provides reasonable variable initializations: \\(0\\) for unconstrained parameters; \\(1\\) for parameters constrained to be positive; \\(0.5\\) for variables to constrained to lie between \\(0\\) and \\(1\\); a symmetric (uniform) vector for simplexes; unit matrices for both correlation and covariance matrices; and so on.\nfilepath - A data file in JSON or Rdump format containing initial parameters values for some or all of the model parameters. User specified initial values must satisfy the constraints declared in the model (i.e., they are on the constrained scale). Parameters which aren’t explicitly initialized will be initialized randomly over the range \\([-2, 2]\\).",
     "crumbs": [
       "Running CmdStan",
-      "Pathfinder Method for Approximate Bayesian Inference"
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/pathfinder_config.html#pathfinder_csv",
-    "href": "cmdstan-guide/pathfinder_config.html#pathfinder_csv",
-    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "objectID": "cmdstan-guide/command_line_options.html#random-number-generator-arguments",
+    "href": "cmdstan-guide/command_line_options.html#random-number-generator-arguments",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "By default, the pathfinder method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a StanCSV file.\nThe CSV files have the following structure:\nThe initial CSV comment rows contain the complete set of CmdStan configuration options.\n...\n# method = pathfinder\n#   pathfinder\n#     init_alpha = 0.001 (Default)\n#     tol_obj = 9.9999999999999998e-13 (Default)\n#     tol_rel_obj = 10000 (Default)\n#     tol_grad = 1e-08 (Default)\n#     tol_rel_grad = 10000000 (Default)\n#     tol_param = 1e-08 (Default)\n#     history_size = 5 (Default)\n#     num_psis_draws = 1000 (Default)\n#     num_paths = 4 (Default)\n#     psis_resample = 1 (Default)\n#     calculate_lp = 1 (Default)\n#     save_single_paths = 0 (Default)\n#     max_lbfgs_iters = 1000 (Default)\n#     num_draws = 1000 (Default)\n#     num_elbo_draws = 25 (Default)\n...\nNext is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs lp_approx__, the log density in the approximating distribution, and lp__, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.\nlp_approx__,lp__,theta\n-2.4973, -8.2951, 0.0811852\n-0.87445, -7.06526, 0.160207\n-0.812285, -7.07124, 0.35819\n...\nThe final lines are comment lines which give timing information.\n# Elapsed Time: 0.016000 seconds (Pathfinders)\n#               0.003000 seconds (PSIS)\n#               0.019000 seconds (Total)\nPathfinder provides option save_single_paths which will save output from the single-path Pathfinder runs.",
+    "text": "The random-number generator’s behavior is determined by the unsigned seed (positive integer) it is started with. If a seed is not specified, or a seed of 0 or less is specified, the system time is used to generate a seed. The seed is recorded and included with Stan’s output regardless of whether it was specified or generated randomly from the system time.\nThe syntax for the random seed argument is:\nrandom seed=&lt;int&gt;\nThe keyword random must be followed directly by the keyword-value pair seed=&lt;int&gt;.",
     "crumbs": [
       "Running CmdStan",
-      "Pathfinder Method for Approximate Bayesian Inference"
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/pathfinder_config.html#single-path-pathfinder-outputs",
-    "href": "cmdstan-guide/pathfinder_config.html#single-path-pathfinder-outputs",
-    "title": "Pathfinder Method for Approximate Bayesian Inference",
+    "objectID": "cmdstan-guide/command_line_options.html#chain-identifier-argument-id",
+    "href": "cmdstan-guide/command_line_options.html#chain-identifier-argument-id",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "The boolean option save_single_paths is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When save_single_paths is true, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name output.csv and default number of runs (4), the resulting CSV files are\noutput.csv\noutput_path_1.csv\noutput_path_1.json\noutput_path_2.csv\noutput_path_2.json\noutput_path_3.csv\noutput_path_3.json\noutput_path_4.csv\noutput_path_4.json\nThe individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.\nTo see how this works, we run Pathfinder on the centered-parameterization of the eight-schools model, where the posterior distribution has a funnel shape:\n&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json\nEach JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.\n{\n  \"0\": {\n    \"iter\": 0,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]\n  },\n  \"1\": {\n    \"iter\": 1,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],\n    \"history_size\": 1,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],\n    \"logDetCholHk\": -2.6839,\n    \"L_approx\": [[-0.0630456, -0.0187959], [0, 1.08328]],\n    \"Qk\": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],\n    \"alpha\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n    \"full\": false,\n    \"lbfgs_note\": \"\"\n  },\n  ...,\n  \"171\": {\n    \"iter\": 171,\n    \"unconstrained_parameters\": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],\n    \"grads\": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],\n    \"history_size\": 5,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],\n    \"logDetCholHk\": 299.023,\n    \"L_approx\": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],\n    \"Qk\": [],\n    \"alpha\": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],\n    \"full\": true,\n    \"lbfgs_note\": \"\"\n  },\n  \"172\": {\n    \"iter\": 172,\n    \"unconstrained_parameters\": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],\n    \"grads\": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],\n    \"history_size\": 5,\n    \"lbfgs_success\": false,\n    \"pathfinder_success\": false,\n    \"lbfgs_note\": \"\"\n  }\n}\nOption num_paths=1 runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments num_paths=1 save_single_paths=true creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.",
+    "text": "The chain identifier argument is used in conjunction with the random seed argument when running multiple Markov chains for sampling. The chain identifier is used to advance the random number generator a very large number of random variates so that two chains with the same seed and different identifiers draw from non-overlapping subsequences of the random-number sequence determined by the seed. Together, the seed and chain identifier determine the behavior of the random number generator.\nThe syntax for the random seed argument is:\nid=&lt;int&gt;\nThe default value is 1.\nWhen running a set of chains from the command line with a specified seed, this argument should be set to the chain index. E.g., when running 4 chains, the value should be 1,..,4, successively. When running multiple chains from a single command, Stan’s interfaces manage the chain identifier arguments automatically.\nFor complete reproducibility, every aspect of the environment needs to be locked down from the OS and version to the C++ compiler and version to the version of Stan and all dependent libraries. See the Stan Reference Manual Reproducibility chapter for further details.",
     "crumbs": [
       "Running CmdStan",
-      "Pathfinder Method for Approximate Bayesian Inference"
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html",
-    "href": "cmdstan-guide/rdump_apdx.html",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/command_line_options.html#command-line-help",
+    "href": "cmdstan-guide/command_line_options.html#command-line-help",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "NOTE: Although the RDump format is still supported, I/O with JSON is faster and recommended. See the chapter on JSON for more details.\nRDump format can be used to represent values for Stan variables. This format was introduced in SPLUS and is used in R, JAGS, and in BUGS (but with a different ordering).\nA dump file is structured as a sequence of variable definitions. Each variable is defined in terms of its dimensionality and its values. There are three kinds of variable declarations: - scalars - sequences - general arrays\n\n\nDump files can be created from R using RStan, via the rstan package function stan_rdump. Stan RDump files must be created via stan_rdump and not by R’s native dump function because R’s dump function uses a richer syntax than is supported by the underlying Stan i/o libraries.\n\n\n\nA simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable y with value \\(17.2\\):\ny &lt;- 17.2\n\n\n\nOne-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.\nn &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)\nArrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.\nSequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as\nn &lt;- 1:3\nThe sequence denoted by 1:3 is of length \\(3\\), running from \\(1\\) to \\(3\\) inclusive. The colon notation allows sequences going from high to low. The following are equivalent:\nn &lt;- 2:-2\nn &lt;- c(2,1,0,-1,-2)\nAs a special case, a sequence of zeros can also be represented in the dump format by integer(x) and double(x), for type int and double, respectively. Here x is a non-negative integer to specify the length. If x is \\(0\\), it can be omitted. The following are some examples.\nx1 &lt;- integer()\nx2 &lt;- integer(0)\nx3 &lt;- integer(2)\ny1 &lt;- double()\ny2 &lt;- double(0)\ny3 &lt;- double(2)\n\n\n\nFor more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a \\(2 \\times 3\\) array:\ny &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))\nData is stored column-major, thus the values for y will be:\ny[1, 1] = 1\ny[1, 2] = 3\ny[1, 3] = 5\ny[2, 1] = 2\ny[2, 2] = 4\ny[2, 3] = 6\nThe structure keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.\nIf the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.\ny &lt;- structure(1:6, .Dim = c(2,3))\nSequence notation can be used within any call to the generic c() function in R. In the above example, c(2,3) could be written as c(2:3).\nThe generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,\nz &lt;- structure(1:24, .Dim = c(2,3,4))\nproduces a three-dimensional int (assignable to real) array z with values:\nz[1, 1, 1] = 1\nz[2, 1, 1] = 2\nz[1, 2, 1] = 3\nz[2, 2, 1] = 4\nz[1, 3, 1] = 5\nz[2, 3, 1] = 6\nz[1, 1, 2] = 7\nz[2, 1, 2] = 8\nz[1, 2, 2] = 9\nz[2, 2, 2] = 10\nz[1, 3, 2] = 11\nz[2, 3, 2] = 12\nz[1, 1, 3] = 13\nz[2, 1, 3] = 14\nz[1, 2, 3] = 15\nz[2, 2, 3] = 16\nz[1, 3, 3] = 17\nz[2, 3, 3] = 18\nz[1, 1, 4] = 19\nz[2, 1, 4] = 20\nz[1, 2, 4] = 21\nz[2, 2, 4] = 22\nz[1, 3, 4] = 23\nz[2, 3, 4] = 24\nIf the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.\nThe sequence of values inside structure can also be integer(x) or double(x). In particular, if one or more dimensions is zero, integer() can be put inside structure. For instance, the following example is supported by the dump format.\ny &lt;- structure(integer(), .Dim = c(2,0))\n\n\n\nThe dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.\n\n\nThe following three declarations have the same dump format for their data.\narray[K] real a;\nvector[K] b;\nrow_vector[K] c;\n\n\n\nThe following declarations have the same dump format.\narray[M, N] real a;\nmatrix[M, N] b;\n\n\n\nThe key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.\nFor the dump data format, the following declarations have the same arrangement.\narray[M, N] real a;\nmatrix[M, N] b;\narray[M] vector[N] c;\narray[M] row_vector[N] d;\nSimilarly, the following also have the same dump format.\narray[P, M, N] real a;\narray[P] matrix[M, N] b;\narray[P, M] vector[N] c;\narray[P, M] row_vector[N] d;\n\n\n\n\nAt this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.\n\n\n\nThere is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., \\(132.3\\)) or uses scientific notation (e.g., \\(1.323e2\\)), Stan assumes that the values are real.\nFor a single value, if there is no decimal point, it may be assigned to an int or real variable in Stan. An array value may only be assigned to an int array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.\nThe following dump file declares an integer value for y.\ny &lt;- 2\nThis definition can be used for a Stan variable y declared as real or as int. Assigning an integer value to a real variable automatically promotes the integer value to a real value.\nInteger values may optionally be followed by L or l, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.\ny &lt;- 2L\nThe following dump file provides a real value for y.\ny &lt;- 2.0\nEven though this is a round value, the occurrence of the decimal point in the value, \\(2.0\\), causes Stan to infer that y is real valued. This dump file may only be used for variables y declared as real in Stan.\n\n\nNumbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as \\(1e+06\\).\n\n\n\nStan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.\n\n\n\nValue\nPreferred Form\nAlternative Forms\n\n\n\n\npositive infinity\nInf\nInfinity, infinity\n\n\nnegative infinity\n-Inf\n-Infinity, -infinity\n\n\nnot a number\nNaN\n\n\n\n\nThese strings are not case sensitive, so inf may also be used for positive infinity, or NAN for not-a-number.\n\n\n\n\nIn order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.\n\"y\" &lt;- c(1,2,3) \\end{Verbatim}\n\n\n\nThe line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.\ny &lt;- 2\ny &lt;-\n3\nAlso following R, breaking before the assignment arrow are not allowed, so the following is invalid.\ny\n&lt;- 2 # Syntax Error\nLines may also be broken in the middle of sequences declared using the c(...) notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a \\(2 \\times 2 \\times 3\\) array is valid.\ny &lt;-\nstructure(c(1,2,3,\n4,5,6,7,8,9,10,11,\n12), .Dim = c(2,2,\n3))\nBecause there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as int or real.\n\n\n\nA more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.\ndefinition ::= name &lt;- value optional_semicolon\n\nname ::= char*     | ''' char* '''     | '\"' char* '\"'\n\nvalue ::= value&lt;int&gt; | value&lt;double&gt;\n\nvalue&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |\n'structure' '(' seq&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'       | 'structure'\n'(' zero_array&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'\n\nseq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;\n\nzero_array&lt;int&gt; ::= \"integer\" '(' &lt;non-negative int&gt;? ')'\n\nzero_array&lt;real&gt; ::= \"double\" '(' &lt;non-negative int&gt;? ')'\n\nseq&lt;real&gt; ::= cseq&lt;real&gt;\n\ncseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'\n\nvseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;\nThe template parameters T will be set to either int or real. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.",
+    "text": "CmdStan provides a help and help-all mechanism that displays either the available top-level or keyword-specific key-value argument pairs. To display top-level help, call the CmdStan executable with keyword help:\n./bernoulli help",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Running CmdStan",
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#creating-dump-files",
-    "href": "cmdstan-guide/rdump_apdx.html#creating-dump-files",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/command_line_options.html#error-messages-and-return-codes",
+    "href": "cmdstan-guide/command_line_options.html#error-messages-and-return-codes",
+    "title": "Command-Line Interface Overview",
     "section": "",
-    "text": "Dump files can be created from R using RStan, via the rstan package function stan_rdump. Stan RDump files must be created via stan_rdump and not by R’s native dump function because R’s dump function uses a richer syntax than is supported by the underlying Stan i/o libraries.",
+    "text": "CmdStan executables and utility programs use streams standard output (stdout) and standard error (stderr) to report information and error messages, respectively. Some methods also generate warning messages when the algorithm detects potential problems with the inference. Depending on the method, these messages are sent to either standard out or standard error.\nAll program executables provide a return code between 0 and 255:\n\n0 - Program ran to termination as expected.\nvalue in range [1 : 125] - Method invoked could not run due to problems with model or data.\nvalue &gt; 128 - Fatal error during execution, process terminated by signal. To determine the signal number, subtract 128 from the return value, e.g. return code 139 results from termination signal 11 (segmentation violation).\n\nA non-zero return code or outputs sent to stderr indicate problems with the inference. However, a return code of zero and absence of error messages doesn’t necessarily mean that the inference is valid, it is still necessary to validate the inferences using all available summary and diagnostic techniques.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Running CmdStan",
+      "Command-Line Interface Overview"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#scalar-variables",
-    "href": "cmdstan-guide/rdump_apdx.html#scalar-variables",
-    "title": "RDump Format for CmdStan",
+    "objectID": "index.html",
+    "href": "index.html",
+    "title": "Stan Documentation",
     "section": "",
-    "text": "A simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable y with value \\(17.2\\):\ny &lt;- 17.2",
-    "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
-    ]
+    "text": "This is the official documentation for Stan.\n\nThe Stan User’s Guide (pdf) provides example models and programming techniques for coding statistical models in Stan.\nThe Stan Reference Manual (pdf) specifies the Stan programming language and inference algorithms.\nThe Stan Functions Reference (pdf) specifies the functions built into the Stan programming language.\n\nThere are also separate installation and getting started guides for CmdStan (pdf), the command-line interface to the Stan inference engine, and the R, Python, and Julia interfaces.\n\nOlder Versions\nThis documentation is for Stan 2.39. Older versions of each of the documents linked above can be found in the table below:\n\n\n\nVersion\nStan Reference Manual\nStan Users Guide\nCmdStan Guide\nStan Functions Reference\n\n\n\n\n2.39\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.38\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.37\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.36\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.35\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.34\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.33\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.32\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.31\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.30\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.29\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.28\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.27\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.26\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.25\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.24\nhtml pdf\nhtml pdf\nhtml pdf\nhtml pdf\n\n\n2.23\nhtml pdf\nhtml pdf\nhtml\nhtml pdf\n\n\n2.22\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.21\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.20\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.19\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n2.18\nhtml pdf\nhtml pdf\n\nhtml pdf\n\n\n\nPrior to version 2.18, all documentation was part of a single document called the Stan User’s Guide and Reference Manual. These versions are still available for download as PDFs:\n\n\n\nVersion\nStan User’s Guide and Reference Manual\n\n\n\n\n2.17\npdf\n\n\n2.16\npdf\n\n\n2.15\npdf\n\n\n2.14\npdf\n\n\n2.13\npdf\n\n\n2.12\npdf\n\n\n2.11\npdf\n\n\n2.10\npdf\n\n\n2.9\npdf\n\n\n2.8\npdf\n\n\n2.7\npdf\n\n\n2.6\npdf\n\n\n2.5\npdf\n\n\n2.4\npdf\n\n\n2.3\npdf\n\n\n2.2\npdf\n\n\n2.1\npdf\n\n\n2.0\npdf\n\n\n1.3\npdf\n\n\n1.2\npdf\n\n\n1.1\npdf\n\n\n1.0\npdf\n\n\n\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n\n Back to top"
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#sequence-variables",
-    "href": "cmdstan-guide/rdump_apdx.html#sequence-variables",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/bib.html",
+    "href": "cmdstan-guide/bib.html",
+    "title": "References",
     "section": "",
-    "text": "One-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.\nn &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)\nArrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.\nSequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as\nn &lt;- 1:3\nThe sequence denoted by 1:3 is of length \\(3\\), running from \\(1\\) to \\(3\\) inclusive. The colon notation allows sequences going from high to low. The following are equivalent:\nn &lt;- 2:-2\nn &lt;- c(2,1,0,-1,-2)\nAs a special case, a sequence of zeros can also be represented in the dump format by integer(x) and double(x), for type int and double, respectively. Here x is a non-negative integer to specify the length. If x is \\(0\\), it can be omitted. The following are some examples.\nx1 &lt;- integer()\nx2 &lt;- integer(0)\nx3 &lt;- integer(2)\ny1 &lt;- double()\ny2 &lt;- double(0)\ny3 &lt;- double(2)",
-    "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
-    ]
+    "text": "References\n\n\n\n\n Back to top"
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#array-variables",
-    "href": "cmdstan-guide/rdump_apdx.html#array-variables",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/compiling_stan_programs.html",
+    "href": "cmdstan-guide/compiling_stan_programs.html",
+    "title": "Compiling a Stan Program",
     "section": "",
-    "text": "For more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a \\(2 \\times 3\\) array:\ny &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))\nData is stored column-major, thus the values for y will be:\ny[1, 1] = 1\ny[1, 2] = 3\ny[1, 3] = 5\ny[2, 1] = 2\ny[2, 2] = 4\ny[2, 3] = 6\nThe structure keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.\nIf the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.\ny &lt;- structure(1:6, .Dim = c(2,3))\nSequence notation can be used within any call to the generic c() function in R. In the above example, c(2,3) could be written as c(2:3).\nThe generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,\nz &lt;- structure(1:24, .Dim = c(2,3,4))\nproduces a three-dimensional int (assignable to real) array z with values:\nz[1, 1, 1] = 1\nz[2, 1, 1] = 2\nz[1, 2, 1] = 3\nz[2, 2, 1] = 4\nz[1, 3, 1] = 5\nz[2, 3, 1] = 6\nz[1, 1, 2] = 7\nz[2, 1, 2] = 8\nz[1, 2, 2] = 9\nz[2, 2, 2] = 10\nz[1, 3, 2] = 11\nz[2, 3, 2] = 12\nz[1, 1, 3] = 13\nz[2, 1, 3] = 14\nz[1, 2, 3] = 15\nz[2, 2, 3] = 16\nz[1, 3, 3] = 17\nz[2, 3, 3] = 18\nz[1, 1, 4] = 19\nz[2, 1, 4] = 20\nz[1, 2, 4] = 21\nz[2, 2, 4] = 22\nz[1, 3, 4] = 23\nz[2, 3, 4] = 24\nIf the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.\nThe sequence of values inside structure can also be integer(x) or double(x). In particular, if one or more dimensions is zero, integer() can be put inside structure. For instance, the following example is supported by the dump format.\ny &lt;- structure(integer(), .Dim = c(2,0))",
+    "text": "A Stan program must be in a file with extension .stan. The CmdStan makefile rules specify all necessary steps to translate files with suffix .stan to a CmdStan executable program. This is a two-stage process:\n\nfirst the Stan program is translated to C++ by the stanc compiler\nthen the C++ compiler compiles all C++ sources and links them together with the CmdStan interface program and the Stan and math libraries.\n\n\n\nTo compile Stan programs, you must invoke the Make program from the &lt;cmdstan-home&gt; directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.\n&gt; cd &lt;cmdstan_home&gt;\nIn the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:\n&gt; make examples/bernoulli/bernoulli\nOn Windows, the command is the same with the addition of .exe at the end of the target (note the use of forward slashes):\n&gt; make examples/bernoulli/bernoulli.exe\nThe generated C++ code (bernoulli.hpp), object file (bernoulli.o) and the compiled executable will be placed in the same directory as the Stan program.\nThe compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.\n\n\n\nWhen executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target bernoulli twice in a row, without any editing bernoulli.stan or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:\n&gt; make examples/bernoulli/bernoulli\nmake: `examples/bernoulli/bernoulli' is up to date.\nIf the file containing the Stan program is updated, the next call to make will rebuild the CmdStan executable.\n\n\n\nThe Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.\nThe simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the bernoulli.stan program, we introduce a typo on line \\(9\\) by writing thata instead of theta, the Make command fails with the following\n--- Translating Stan model to C++ code ---\nbin/stanc  --o=bernoulli.hpp bernoulli.stan\n\nSemantic error in 'bernoulli.stan', line 9, column 2 to column 7:\n   -------------------------------------------------\n     7:  }\n     8:  model {\n     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1\n           ^\n    10:    y ~ bernoulli(theta);\n    11:  }\n   -------------------------------------------------\n\nIdentifier 'thata' not in scope.\n\nmake: *** [bernoulli.hpp] Error 1\nStan is a strongly-typed language; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program foo.stan contains an illegal assignment statement:\ndata {\n  real x;\n}\ntransformed data {\n  int y = x;\n}\nThe Make command fails with the following:\nSemantic error in 'foo.stan', line 5, column 2 to column 12:\n   -------------------------------------------------\n     3:  }\n     4:  transformed data {\n     5:    int y = x;\n           ^\n     6:  }\n   -------------------------------------------------\n\nIll-typed arguments supplied to assignment operator =:\n lhs has type int and rhs has type real\nThe Stan Reference Manual provides a complete specification of the Stan programming language. The Stan User’s Guide also contains a full description of the errors and warnings stanc can emit.\n\n\n\nIf the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.\nThe Make command prints the following message to the terminal at the point when it compiles and links the C++ file:\n--- Compiling, linking C++ code ---\nIf the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.\nIf this happens, please report the error, together with the Stan program on either the Stan Forums or on the Stan compiler GitHub issues tracker.\n\n\n\nUsers can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.\nThe CXXFLAGS and LDFLAGS makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the make/local file.\nFor example:\nCXXFLAGS = -O2\nA recommend a set of CXXFLAGS and LDFLAGS flags can be turned on by setting STAN_CPP_OPTIMS=true in the make/local file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.\n\n\nWhen assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.\nFor some models these checks can represent a significant part of the models execution time. By setting the STAN_NO_RANGE_CHECKS=true makefile flag in the make/local file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Getting Started",
+      "Compiling a Stan Program"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#matrix--and-vector-valued-variables",
-    "href": "cmdstan-guide/rdump_apdx.html#matrix--and-vector-valued-variables",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#invoking-the-make-utility",
+    "href": "cmdstan-guide/compiling_stan_programs.html#invoking-the-make-utility",
+    "title": "Compiling a Stan Program",
     "section": "",
-    "text": "The dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.\n\n\nThe following three declarations have the same dump format for their data.\narray[K] real a;\nvector[K] b;\nrow_vector[K] c;\n\n\n\nThe following declarations have the same dump format.\narray[M, N] real a;\nmatrix[M, N] b;\n\n\n\nThe key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.\nFor the dump data format, the following declarations have the same arrangement.\narray[M, N] real a;\nmatrix[M, N] b;\narray[M] vector[N] c;\narray[M] row_vector[N] d;\nSimilarly, the following also have the same dump format.\narray[P, M, N] real a;\narray[P] matrix[M, N] b;\narray[P, M] vector[N] c;\narray[P, M] row_vector[N] d;",
+    "text": "To compile Stan programs, you must invoke the Make program from the &lt;cmdstan-home&gt; directory. The Stan program can be in a different directory, but the directory path names cannot contain spaces - this limitation is imposed by Make.\n&gt; cd &lt;cmdstan_home&gt;\nIn the call to the Make program, the target is name of the CmdStan executable corresponding to the Stan program file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. To build the Bernoulli example, on Mac and Linux:\n&gt; make examples/bernoulli/bernoulli\nOn Windows, the command is the same with the addition of .exe at the end of the target (note the use of forward slashes):\n&gt; make examples/bernoulli/bernoulli.exe\nThe generated C++ code (bernoulli.hpp), object file (bernoulli.o) and the compiled executable will be placed in the same directory as the Stan program.\nThe compiled executable consists of the Stan model and the CmdStan command line interface which provides inference algorithms to do MCMC sampling, optimization, and variational inference. The following sections provide examples of doing inference using each method on the example model and data file.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Getting Started",
+      "Compiling a Stan Program"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#complex-valued-variables",
-    "href": "cmdstan-guide/rdump_apdx.html#complex-valued-variables",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#dependencies",
+    "href": "cmdstan-guide/compiling_stan_programs.html#dependencies",
+    "title": "Compiling a Stan Program",
     "section": "",
-    "text": "At this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.",
+    "text": "When executing a Make target, all its dependencies are checked to see if they are up to date, and if they are not, they are rebuilt. If the you call Make with target bernoulli twice in a row, without any editing bernoulli.stan or otherwise changing the system, on the second invocation, Make will determine that the executable is already newer than the Stan source file and will not recompile the program:\n&gt; make examples/bernoulli/bernoulli\nmake: `examples/bernoulli/bernoulli' is up to date.\nIf the file containing the Stan program is updated, the next call to make will rebuild the CmdStan executable.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Getting Started",
+      "Compiling a Stan Program"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#integer--and-real-valued-variables",
-    "href": "cmdstan-guide/rdump_apdx.html#integer--and-real-valued-variables",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#compiler-errors",
+    "href": "cmdstan-guide/compiling_stan_programs.html#compiler-errors",
+    "title": "Compiling a Stan Program",
     "section": "",
-    "text": "There is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., \\(132.3\\)) or uses scientific notation (e.g., \\(1.323e2\\)), Stan assumes that the values are real.\nFor a single value, if there is no decimal point, it may be assigned to an int or real variable in Stan. An array value may only be assigned to an int array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.\nThe following dump file declares an integer value for y.\ny &lt;- 2\nThis definition can be used for a Stan variable y declared as real or as int. Assigning an integer value to a real variable automatically promotes the integer value to a real value.\nInteger values may optionally be followed by L or l, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.\ny &lt;- 2L\nThe following dump file provides a real value for y.\ny &lt;- 2.0\nEven though this is a round value, the occurrence of the decimal point in the value, \\(2.0\\), causes Stan to infer that y is real valued. This dump file may only be used for variables y declared as real in Stan.\n\n\nNumbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as \\(1e+06\\).\n\n\n\nStan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.\n\n\n\nValue\nPreferred Form\nAlternative Forms\n\n\n\n\npositive infinity\nInf\nInfinity, infinity\n\n\nnegative infinity\n-Inf\n-Infinity, -infinity\n\n\nnot a number\nNaN\n\n\n\n\nThese strings are not case sensitive, so inf may also be used for positive infinity, or NAN for not-a-number.",
+    "text": "The Stan probabilistic programming language is a programming language with a rich syntax, as such, it is often the case that a carefully written program contains errors.\nThe simplest class of errors are simple syntax errors such as forgetting the semi-colon statement termination marker at the end of a line, or typos such as a misspelled variable name. For example, if in the bernoulli.stan program, we introduce a typo on line \\(9\\) by writing thata instead of theta, the Make command fails with the following\n--- Translating Stan model to C++ code ---\nbin/stanc  --o=bernoulli.hpp bernoulli.stan\n\nSemantic error in 'bernoulli.stan', line 9, column 2 to column 7:\n   -------------------------------------------------\n     7:  }\n     8:  model {\n     9:    thata ~ beta(1, 1);  // uniform prior on interval 0, 1\n           ^\n    10:    y ~ bernoulli(theta);\n    11:  }\n   -------------------------------------------------\n\nIdentifier 'thata' not in scope.\n\nmake: *** [bernoulli.hpp] Error 1\nStan is a strongly-typed language; and the compiler will throw an error if statements or expressions violate the type rules. The following trivial program foo.stan contains an illegal assignment statement:\ndata {\n  real x;\n}\ntransformed data {\n  int y = x;\n}\nThe Make command fails with the following:\nSemantic error in 'foo.stan', line 5, column 2 to column 12:\n   -------------------------------------------------\n     3:  }\n     4:  transformed data {\n     5:    int y = x;\n           ^\n     6:  }\n   -------------------------------------------------\n\nIll-typed arguments supplied to assignment operator =:\n lhs has type int and rhs has type real\nThe Stan Reference Manual provides a complete specification of the Stan programming language. The Stan User’s Guide also contains a full description of the errors and warnings stanc can emit.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Getting Started",
+      "Compiling a Stan Program"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#quoted-variable-names",
-    "href": "cmdstan-guide/rdump_apdx.html#quoted-variable-names",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#troubleshooting-c-compiler-or-linker-errors",
+    "href": "cmdstan-guide/compiling_stan_programs.html#troubleshooting-c-compiler-or-linker-errors",
+    "title": "Compiling a Stan Program",
     "section": "",
-    "text": "In order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.\n\"y\" &lt;- c(1,2,3) \\end{Verbatim}",
+    "text": "If the stanc compiler successfully translates a Stan program to C++, the resulting C++ code should be valid C++ which can be compiled into an executable. The stanc compiler is also a program, and while it has been extensively tested, it may still contain errors such that the generated C++ code fails to compile.\nThe Make command prints the following message to the terminal at the point when it compiles and links the C++ file:\n--- Compiling, linking C++ code ---\nIf the program fails to compile for any reason, the C++ compiler and linker will most likely print a long series of error messages to the console.\nIf this happens, please report the error, together with the Stan program on either the Stan Forums or on the Stan compiler GitHub issues tracker.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Getting Started",
+      "Compiling a Stan Program"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#line-breaks",
-    "href": "cmdstan-guide/rdump_apdx.html#line-breaks",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/compiling_stan_programs.html#c-compilation-and-linking-flags",
+    "href": "cmdstan-guide/compiling_stan_programs.html#c-compilation-and-linking-flags",
+    "title": "Compiling a Stan Program",
     "section": "",
-    "text": "The line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.\ny &lt;- 2\ny &lt;-\n3\nAlso following R, breaking before the assignment arrow are not allowed, so the following is invalid.\ny\n&lt;- 2 # Syntax Error\nLines may also be broken in the middle of sequences declared using the c(...) notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a \\(2 \\times 2 \\times 3\\) array is valid.\ny &lt;-\nstructure(c(1,2,3,\n4,5,6,7,8,9,10,11,\n12), .Dim = c(2,2,\n3))\nBecause there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as int or real.",
+    "text": "Users can set flags for the C++ compiler and linker and compiler to optimize their executables. We advise users to only do this once they are sure their basic setup of Cmdstan without flags works.\nThe CXXFLAGS and LDFLAGS makefile variables can be used to set compiler and linker flags respectively. We recommend setting these in the make/local file.\nFor example:\nCXXFLAGS = -O2\nA recommend a set of CXXFLAGS and LDFLAGS flags can be turned on by setting STAN_CPP_OPTIMS=true in the make/local file. These are tested compiler and link-time optimizations that can speed up execution of certain models. We have observed speedups up to 15 percent, but this depends on the model, operating system and hardware used. The use of these flags does considerably slow down compilation, so they are not used by default.\n\n\nWhen assigning or reading from with vectors, row_vectors, matrices or arrays using indexing, Stan checks that a supplied index is valid (not out of range), which avoids segmentation faults and other difficult-to-debug runtime errors.\nFor some models these checks can represent a significant part of the models execution time. By setting the STAN_NO_RANGE_CHECKS=true makefile flag in the make/local file the range checks can be removed. Use this flag with caution (only once the indexing has been validated). In case of any unexpected behavior remove the flag for easier debugging.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Getting Started",
+      "Compiling a Stan Program"
     ]
   },
   {
-    "objectID": "cmdstan-guide/rdump_apdx.html#bnf-grammar-for-dump-data",
-    "href": "cmdstan-guide/rdump_apdx.html#bnf-grammar-for-dump-data",
-    "title": "RDump Format for CmdStan",
+    "objectID": "cmdstan-guide/diagnose_utility.html",
+    "href": "cmdstan-guide/diagnose_utility.html",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
     "section": "",
-    "text": "A more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.\ndefinition ::= name &lt;- value optional_semicolon\n\nname ::= char*     | ''' char* '''     | '\"' char* '\"'\n\nvalue ::= value&lt;int&gt; | value&lt;double&gt;\n\nvalue&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |\n'structure' '(' seq&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'       | 'structure'\n'(' zero_array&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'\n\nseq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;\n\nzero_array&lt;int&gt; ::= \"integer\" '(' &lt;non-negative int&gt;? ')'\n\nzero_array&lt;real&gt; ::= \"double\" '(' &lt;non-negative int&gt;? ')'\n\nseq&lt;real&gt; ::= cseq&lt;real&gt;\n\ncseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'\n\nvseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;\nThe template parameters T will be set to either int or real. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.",
+    "text": "CmdStan is distributed with a utility that is able to read in and analyze the output of one or more Markov chains to check for the following potential problems:\n\nDivergent transitions\nTransitions that hit the maximum treedepth\nLow E-BFMI values\nLow effective sample sizes\nHigh \\(\\hat{R}\\) values\n\nThe meanings of several of these problems are discussed in https://arxiv.org/abs/1701.02434.\n\n\nThe CmdStan makefile task build compiles the diagnose utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/diagnose\n\n\n\nThe diagnose command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to diagnose, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.\nTo fully exercise the diagnose command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the Stan User’s Guide reparameterization section. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y / 2));\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan\nStan has trouble sampling from the region where y is small and thus x is constrained to be near 0. This is due to the fact that the density’s scale changes with y, so that a step size that works well when y is large is inefficient when y is small and vice-versa.\nRunning 4 chains produces output files output_1.csv, …, output_4.csv. We run diagnose command on this fileset:\n&gt; bin/diagnose output_*.csv\nThe output is printed to the terminal window:\nChecking sampler transitions treedepth.\n18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.\nTrajectories that are prematurely terminated due to this limit will result in slow exploration.\nFor optimal performance, increase this limit.\n\nChecking sampler transitions for divergences.\n11 of 4000 (0.28%) transitions ended with a divergence.\nThese divergent transitions indicate that HMC is not fully able to explore the posterior distribution.\nTry increasing adapt delta closer to 1.\nIf this doesn't remove all divergences, try to reparameterize the model.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nThe E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.\nIf possible, try to reparameterize the model.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nThe following parameters had rank-normalized split R-hat greater than 1.01:\n  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]\nSuch high values indicate incomplete mixing and biased estimation.\nYou should consider regularizing your model with additional prior information or a more effective parameterization.\n\nProcessing complete.\nIn this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y / 2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y / 2))\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan\nWe compile the program and run 4 chains, as before. Now the diagnose command doesn’t detect any problems:\nChecking sampler transitions treedepth.\nTreedepth satisfactory for all transitions.\n\nChecking sampler transitions for divergences.\nNo divergent transitions found.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nE-BFMI satisfactory.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nRank-normalized split R-hat values satisfactory for all parameters.\n\nProcessing complete, no problems detected.\n\n\n\n\n\nStan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a Hamiltonian system. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the step size controls the resolution of the sampler.\nUnfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as divergences which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.\nIf the divergent transitions cannot be eliminated by increasing the adapt_delta parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the reparameterization section of the Stan User’s Guide.\n\n\n\nWarnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a validity concern, hitting the maximum treedepth is an efficiency concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the Hamiltonian Monte Carlo Sampling chapter in the Stan Reference Manual). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.\nThis is controlled through the max_depth argument. If the number of transitions which exceed maximum treedepth is low, increasing max_depth may correct this problem.\n\n\n\nThe sampler csv output column energy__ is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of energy is much larger than \\(\\sqrt{D / 2}\\), where \\(D\\) is the number of unconstrained parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.\nThe warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.\n\n\n\nRoughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses \\(\\hat{R}\\) adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.\nBulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g. mean and median).\nOften quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.\n\n\n\n\\(\\hat{R}\\) (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), \\(\\hat{R}\\) is larger than 1. We recommend running at least four chains by default and only using the sample if \\(\\hat{R}\\) is less than 1.01. Stan reports \\(\\hat{R}\\) which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.\nThere is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.",
     "crumbs": [
-      "Appendices",
-      "RDump Format for CmdStan"
+      "Tools and Utilities",
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stanc.html",
-    "href": "cmdstan-guide/stanc.html",
-    "title": "stanc: Translating Stan to C++",
+    "objectID": "cmdstan-guide/diagnose_utility.html#building-the-diagnose-command",
+    "href": "cmdstan-guide/diagnose_utility.html#building-the-diagnose-command",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
     "section": "",
-    "text": "CmdStan translates Stan programs to C++ using the Stan compiler program which is included in the CmdStan release bin directory as program stanc. One can view the complete stanc documentation in the Stan User’s Guide.\nAs of release 2.22, the CmdStan Stan to C++ compiler is written in OCaml. This compiler is called “stanc3” and has has its own repository https://github.com/stan-dev/stanc3, from which pre-built binaries for Linux, Mac, and Windows can be downloaded.\n\n\nBefore the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:\nmake bin/stanc\nFor Windows:\nmake bin/stanc.exe\nThis is also done as part of the make build command.\n\n\n\nThe Stan compiler program stanc converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model bernoulli.stan. For Linux and Mac:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nFor Windows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nThe base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (a--z or A--Z) and contain only alphanumeric characters (a--z, A--Z, and 0--9) and underscores (_) and should not conflict with any C++ reserved keyword.\nThe C++ code implementing the class is written to the file bernoulli.hpp in the current directory. The final argument, bernoulli.stan, is the file from which to read the Stan program.\nIn practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via make, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:\n&gt; make examples/bernoulli/bernoulli\nFor Windows:\n&gt; make examples/bernoulli/bernoulli.exe\nThe makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable STANCFLAGS can be used to to override the default arguments to stanc, e.g.,\n&gt; make STANCFLAGS=\"--include-paths=~/foo\" examples/bernoulli/bernoulli",
+    "text": "The CmdStan makefile task build compiles the diagnose utility into the bin directory. It can be compiled directly using the makefile as follows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; make bin/diagnose",
     "crumbs": [
       "Tools and Utilities",
-      "`stanc`: Translating Stan to C++"
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stanc.html#instantiating-the-stanc-binary",
-    "href": "cmdstan-guide/stanc.html#instantiating-the-stanc-binary",
-    "title": "stanc: Translating Stan to C++",
+    "objectID": "cmdstan-guide/diagnose_utility.html#running-the-diagnose-command",
+    "href": "cmdstan-guide/diagnose_utility.html#running-the-diagnose-command",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
     "section": "",
-    "text": "Before the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:\nmake bin/stanc\nFor Windows:\nmake bin/stanc.exe\nThis is also done as part of the make build command.",
+    "text": "The diagnose command is executed on one or more output files, which are provided as command-line arguments separated by spaces. If there are no apparent problems with the output files passed to diagnose, it outputs a message that all transitions are within treedepth limit and that no divergent transitions were found. It problems are detected, it outputs a summary of the problem along with possible ways to mitigate it.\nTo fully exercise the diagnose command, we run 4 chains to sample from the Neal’s funnel distribution, discussed in the Stan User’s Guide reparameterization section. This program defines a distribution which exemplifies the difficulties of sampling from some hierarchical models:\nparameters {\n  real y;\n  vector[9] x;\n}\nmodel {\n  y ~ normal(0, 3);\n  x ~ normal(0, exp(y / 2));\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel.stan\nStan has trouble sampling from the region where y is small and thus x is constrained to be near 0. This is due to the fact that the density’s scale changes with y, so that a step size that works well when y is large is inefficient when y is small and vice-versa.\nRunning 4 chains produces output files output_1.csv, …, output_4.csv. We run diagnose command on this fileset:\n&gt; bin/diagnose output_*.csv\nThe output is printed to the terminal window:\nChecking sampler transitions treedepth.\n18 of 4000 (0.45%) transitions hit the maximum treedepth limit of 10, or 2^10 leapfrog steps.\nTrajectories that are prematurely terminated due to this limit will result in slow exploration.\nFor optimal performance, increase this limit.\n\nChecking sampler transitions for divergences.\n11 of 4000 (0.28%) transitions ended with a divergence.\nThese divergent transitions indicate that HMC is not fully able to explore the posterior distribution.\nTry increasing adapt delta closer to 1.\nIf this doesn't remove all divergences, try to reparameterize the model.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nThe E-BFMI, 0.06, is below the nominal threshold of 0.30 which suggests that HMC may have trouble exploring the target distribution.\nIf possible, try to reparameterize the model.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nThe following parameters had rank-normalized split R-hat greater than 1.01:\n  y, x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9]\nSuch high values indicate incomplete mixing and biased estimation.\nYou should consider regularizing your model with additional prior information or a more effective parameterization.\n\nProcessing complete.\nIn this example, changing the model to use a non-centered parameterization is the only way to correct these problems. In this second model, the parameters x_raw and y_raw are sampled as independent standard normals, which is easy for Stan.\nparameters {\n  real y_raw;\n  vector[9] x_raw;\n}\ntransformed parameters {\n  real y;\n  vector[9] x;\n\n  y = 3.0 * y_raw;\n  x = exp(y / 2) * x_raw;\n}\nmodel {\n  y_raw ~ std_normal(); // implies y ~ normal(0, 3)\n  x_raw ~ std_normal(); // implies x ~ normal(0, exp(y / 2))\n}\nThis program is available on GitHub: https://github.com/stan-dev/example-models/blob/master/misc/funnel/funnel_reparam.stan\nWe compile the program and run 4 chains, as before. Now the diagnose command doesn’t detect any problems:\nChecking sampler transitions treedepth.\nTreedepth satisfactory for all transitions.\n\nChecking sampler transitions for divergences.\nNo divergent transitions found.\n\nChecking E-BFMI - sampler transitions HMC potential energy.\nE-BFMI satisfactory.\n\nRank-normalized split effective sample size satisfactory for all parameters.\n\nRank-normalized split R-hat values satisfactory for all parameters.\n\nProcessing complete, no problems detected.",
     "crumbs": [
       "Tools and Utilities",
-      "`stanc`: Translating Stan to C++"
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
     ]
   },
   {
-    "objectID": "cmdstan-guide/stanc.html#the-stan-compiler-program",
-    "href": "cmdstan-guide/stanc.html#the-stan-compiler-program",
-    "title": "stanc: Translating Stan to C++",
+    "objectID": "cmdstan-guide/diagnose_utility.html#diagnose-warnings-and-recommendations",
+    "href": "cmdstan-guide/diagnose_utility.html#diagnose-warnings-and-recommendations",
+    "title": "diagnose: Diagnosing Biased Hamiltonian Monte Carlo Inferences",
     "section": "",
-    "text": "The Stan compiler program stanc converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model bernoulli.stan. For Linux and Mac:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nFor Windows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nThe base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (a--z or A--Z) and contain only alphanumeric characters (a--z, A--Z, and 0--9) and underscores (_) and should not conflict with any C++ reserved keyword.\nThe C++ code implementing the class is written to the file bernoulli.hpp in the current directory. The final argument, bernoulli.stan, is the file from which to read the Stan program.\nIn practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via make, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:\n&gt; make examples/bernoulli/bernoulli\nFor Windows:\n&gt; make examples/bernoulli/bernoulli.exe\nThe makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable STANCFLAGS can be used to to override the default arguments to stanc, e.g.,\n&gt; make STANCFLAGS=\"--include-paths=~/foo\" examples/bernoulli/bernoulli",
+    "text": "Stan uses Hamiltonian Monte Carlo (HMC) to explore the target distribution — the posterior defined by a Stan program + data — by simulating the evolution of a Hamiltonian system. In order to approximate the exact solution of the Hamiltonian dynamics we need to choose a step size governing how far we move each time we evolve the system forward. That is, the step size controls the resolution of the sampler.\nUnfortunately, for particularly hard problems there are features of the target distribution that are too small for this resolution. Consequently the sampler misses those features and returns biased estimates. Fortunately, this mismatch of scales manifests as divergences which provide a practical diagnostic. If there are any divergences after warmup, then the sample based estimates may be biased.\nIf the divergent transitions cannot be eliminated by increasing the adapt_delta parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution. This problem occurs frequently with hierarchical models and one of the simplest examples is Neal’s Funnel, which is discussed in the reparameterization section of the Stan User’s Guide.\n\n\n\nWarnings about hitting the maximum treedepth are not as serious as warnings about divergent transitions. While divergent transitions are a validity concern, hitting the maximum treedepth is an efficiency concern. Configuring the No-U-Turn-Sampler (the variant of HMC used by Stan) requires putting a cap on the depth of the trees that it evaluates during each iteration (for details on this see the Hamiltonian Monte Carlo Sampling chapter in the Stan Reference Manual). When the maximum allowed tree depth is reached it indicates that NUTS is terminating prematurely to avoid excessively long execution time.\nThis is controlled through the max_depth argument. If the number of transitions which exceed maximum treedepth is low, increasing max_depth may correct this problem.\n\n\n\nThe sampler csv output column energy__ is used to diagnose the accuracy of any Hamiltonian Monte Carlo sampler. If the standard deviation of energy is much larger than \\(\\sqrt{D / 2}\\), where \\(D\\) is the number of unconstrained parameters, then the sampler is unlikely to be able to explore the posterior adequately. This is usually due to heavy-tailed posteriors and can sometimes be remedied by reparameterizing the model.\nThe warning that some number of chains had an estimated Bayesian Fraction of Missing Information (BFMI) that was too low implies that the adaptation phase of the Markov Chains did not turn out well and those chains likely did not explore the posterior distribution efficiently. For more details on this diagnostic, see https://arxiv.org/abs/1604.00695. Should this occur, you can either run the sampler for more iterations, or consider reparameterizing your model.\n\n\n\nRoughly speaking, the effective sample size (ESS) of a quantity of interest captures how many independent draws contain the same amount of information as the dependent sample obtained by the MCMC algorithm. Clearly, the higher the ESS the better. Stan uses \\(\\hat{R}\\) adjustment to use the between-chain information in computing the ESS. For example, in case of multimodal distributions with well-separated modes, this leads to an ESS estimate that is close to the number of distinct modes that are found.\nBulk-ESS refers to the effective sample size based on the rank normalized draws. This does not directly compute the ESS relevant for computing the mean of the parameter, but instead computes a quantity that is well defined even if the chains do not have finite mean or variance. Overall bulk-ESS estimates the sampling efficiency for the location of the distribution (e.g. mean and median).\nOften quite smaller ESS would be sufficient for the desired estimation accuracy, but the estimation of ESS and convergence diagnostics themselves require higher ESS. We recommend requiring that the bulk-ESS is greater than 100 times the number of chains. For example, when running four chains, this corresponds to having a rank-normalized effective sample size of at least 400.\n\n\n\n\\(\\hat{R}\\) (R-hat) convergence diagnostic compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don’t agree), \\(\\hat{R}\\) is larger than 1. We recommend running at least four chains by default and only using the sample if \\(\\hat{R}\\) is less than 1.01. Stan reports \\(\\hat{R}\\) which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale. For more details on this diagnostic, see https://arxiv.org/abs/1903.08008.\nThere is further discussion in https://arxiv.org/abs/1701.02434; however the correct resolution is necessarily model specific, hence all suggestions general guidelines only.",
     "crumbs": [
       "Tools and Utilities",
-      "`stanc`: Translating Stan to C++"
+      "`diagnose`: Diagnosing Biased Hamiltonian Monte Carlo Inferences"
     ]
   },
   {
-    "objectID": "cmdstan-guide/variational_config.html",
-    "href": "cmdstan-guide/variational_config.html",
-    "title": "Variational Inference using ADVI",
+    "objectID": "cmdstan-guide/external_code.html",
+    "href": "cmdstan-guide/external_code.html",
+    "title": "Using external C++ code",
     "section": "",
-    "text": "Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) Kucukelbir et al. (2017). ADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO (evidence lower bound). ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. The measures of convergence are similar to the relative tolerance scheme of Stan’s optimization algorithms.\nThe algorithm progression consists of an adaptation phase followed by a sampling phase. The adaptation phase finds a good value for the step size scaling parameter eta. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. The algorithm runs until the mean change in ELBO drops below the specified tolerance.\nThe full set of configuration options available for the variational method is available by using the variational help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.\nThe following is a minimal call to Stan’s variational inference algorithm using defaults for everything but the location of the data file.\n&gt; ./bernoulli variational data file=bernoulli.data.R\nExecuting this command prints both output to the console and to a csv file.\nThe first part of the console output reports on the configuration used: the default option algorithm=meanfield and the default tolerances for monitoring the algorithm’s convergence.\nmethod = variational\n  variational\n    algorithm = meanfield (Default)\n      meanfield\n    iter = 10000 (Default)\n    grad_samples = 1 (Default)\n    elbo_samples = 100 (Default)\n    eta = 1 (Default)\n    adapt\n      engaged = true (Default)\n      iter = 50 (Default)\n    tol_rel_obj = 0.01 (Default)\n    eval_elbo = 100 (Default)\n    output_samples = 1000 (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2790599354 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nAfter the configuration has been displayed, informational and timing messages are output:\n------------------------------------------------------------\nEXPERIMENTAL ALGORITHM:\n  This procedure has not been thoroughly tested and may be unstable\n  or buggy. The interface is subject to change.\n------------------------------------------------------------\n\nGradient evaluation took 2.1e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.21 seconds.\nAdjust your expectations accordingly!\nThe rest of the output describes the progression of the algorithm. An adaptation phase finds a good value for the step size scaling parameter eta. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. That is, while the algorithm appears to have converged in \\(\\sim\\) 250 iterations, the algorithm runs for another few thousand iterations until mean change in ELBO drops below the default tolerance of 0.01.\nBegin eta adaptation.\nIteration:   1 / 250 [  0%]  (Adaptation)\nIteration:  50 / 250 [ 20%]  (Adaptation)\nIteration: 100 / 250 [ 40%]  (Adaptation)\nIteration: 150 / 250 [ 60%]  (Adaptation)\nIteration: 200 / 250 [ 80%]  (Adaptation)\nSuccess! Found best value [eta = 1] earlier than expected.\n\nBegin stochastic gradient ascent.\n  iter             ELBO   delta_ELBO_mean   delta_ELBO_med   notes\n   100           -6.131             1.000            1.000\n   200           -6.458             0.525            1.000\n   300           -6.300             0.359            0.051\n   400           -6.137             0.276            0.051\n   500           -6.243             0.224            0.027\n   600           -6.305             0.188            0.027\n   700           -6.289             0.162            0.025\n   800           -6.402             0.144            0.025\n   900           -6.103             0.133            0.025\n  1000           -6.314             0.123            0.027\n  1100           -6.348             0.024            0.025\n  1200           -6.244             0.020            0.018\n  1300           -6.293             0.019            0.017\n  1400           -6.250             0.017            0.017\n  1500           -6.241             0.015            0.010   MEDIAN ELBO CONVERGED\n\nDrawing a sample of size 1000 from the approximate posterior...\nCOMPLETED.\n\n\nStan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option algorithm=meanfield. The algorithm argument specifies the variational algorithm.\n\nalgorithm=meanfield - Use a fully factorized Gaussian for the approximation. This is the default algorithm.\nalgorithm=fullrank Use a Gaussian with a full-rank covariance matrix for the approximation.\n\n\n\n\n\niter=&lt;int&gt; Maximum number of iterations. Must be \\(&gt; 0\\). Default is \\(10000\\).\ngrad_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of gradients. Must be \\(&gt; 0\\). Default is \\(1\\).\nelbo_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of ELBO (objective function). Must be \\(&gt; 0\\). Default is \\(100\\).\neta=&lt;double&gt; Stepsize weighting parameter for adaptive stepsize sequence. Must be \\(&gt; 0\\). Default is \\(1.0\\).\nadapt Warmup Adaptation keyword, takes sub-arguments:\n\nengaged=&lt;boolean&gt; Adaptation engaged? Valid values: [true, false]. Default is true.\niter=&lt;int&gt; Maximum number of adaptation iterations. Must be \\(&gt; 0\\). Default is \\(50\\).\n\ntol_rel_obj=&lt;double&gt; Convergence tolerance on the relative norm of the objective. Must be \\(&gt; 0\\). Default is \\(0.01\\).\neval_elbo=&lt;int&gt; Evaluate ELBO every Nth iteration. Must be \\(&gt; 0\\). Default is 100.\noutput_samples=&lt;int&gt; Number of posterior samples to draw and save. Must be \\(&gt; 0\\). Default is 1000.\n\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the variational method is reported at the beginning of the sampler output file as CSV comments.\nThe first three output columns are labelled lp__, log_p__, log_g__, the rest are the model parameters.\nThe stepsize adaptation information is output as CSV comments following column header row.\nThe following line contains the mean of the variational approximation.\nThe rest of the output contains output_samples number of draws sampled from the variational approximation.\n\nTo illustrate, we call Stan’s variational inference on the example model and data:\n&gt; ./bernoulli variational data file=bernoulli.data.R\nBy default, the output file is output.csv.\nThe output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = variational\n#   variational\n#     algorithm = meanfield (Default)\n#       meanfield\n#     iter = 10000 (Default)\n#     grad_samples = 1 (Default)\n#     elbo_samples = 100 (Default)\n#     eta = 1 (Default)\n#     adapt\n#       engaged = true (Default)\n#       iter = 50 (Default)\n#     tol_rel_obj = 0.01 (Default)\n#     eval_elbo = 100 (Default)\n#     output_samples = 1000 (Default)\n...\nNext, the column header row:\nlp__,log_p__,log_g__,theta\nAdditional comments provide stepsize adaptation information:\n# Stepsize adaptation complete.\n# eta = 1\nFollowed by the data rows. The first line is special — it is the mean of the variational approximation.\n0,0,0,0.214911\nThat is, the estimate for theta given the data is 0.2.\nThe rest of the output contains output_samples number of draws samples from the variational approximation.\nThe following is a sample based on this approximation:\n0,-14.0252,-5.21718,0.770397\n0,-7.05063,-0.10025,0.162061\n0,-6.75031,-0.0191099,0.241606\n...\nThe header indicates the unnormalized log probability with lp__. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.",
+    "text": "The --allow-undefined flag can be passed to the call to stanc, which will allow undefined functions in the Stan language to be parsed without an error. We can then include a definition of the function in a C++ header file.\nThis requires specifying two makefile variables:\n\nSTANCFLAGS=--allow-undefined\nUSER_HEADER=&lt;header_file.hpp&gt;, where &lt;header_file.hpp&gt; is the name of a header file that defines a function with the same name and a compatible signature. This function can appear in the global namespace or in the model namespace, which is defined as the name of the model (either the file name, or the --name argument to stanc) followed by _namespace.\n\nThis is an advanced feature which is only recommended to users familiar with the internals of Stan’s Math library. Most existing C++ code will need to be modified to work with Stan, to varying degrees.\nAs an example, consider the following variant of the Bernoulli example\nfunctions {\n  real make_odds(data real theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n}\nmodel {\n  theta ~ beta(1, 1); // uniform prior on interval 0, 1\n  y ~ bernoulli(theta);\n}\ngenerated quantities {\n  real odds;\n  odds = make_odds(theta);\n}\nHere the make_odds function is declared but not defined, which would ordinarily result in a parser error. However, if you put STANCFLAGS = --allow-undefined into the make/local file or into the stanc call, then the stanc compiler will translate this program to C++, but the generated C++ code will not compile unless you write a file such as examples/bernoulli/make_odds.hpp with the following lines\n#include &lt;ostream&gt;\n\ndouble make_odds(const double& theta, std::ostream *pstream__) {\n  return theta / (1 - theta);\n}\nThe signature for this function needs to fulfill all the usages in the C++ class emitted by stanc. The pstream__ argument is mandatory in the signature but need not be used if your function does not print any output. Because make_odds was declared with a data argument and only used in generated quantites, a signature which accepts and returns double is acceptable. Functions which will have parameters passed as input in the transformed parameters or model blocks will require the ability to accept Stan’s autodiff types. If you wish to autodiff through this function, the simplest option is to make it a template, like\ntemplate &lt;typename T&gt;\nT make_odds(const T &theta, std::ostream *pstream__)\n{\n    return theta / (1 - theta);\n}\nGiven the above, the following make invocation should work\n&gt; make STANCFLAGS=--allow-undefined USER_HEADER=examples/bernoulli/make_odds.hpp examples/bernoulli/bernoulli # on Windows add .exe\nAlternatively, you could put STANCFLAGS and USER_HEADER into the make/local file instead of specifying them on the command-line.\nIf the function were more complicated and involved functions in the Stan Math Library, then you would need to add #include &lt;stan/model/model_header.hpp&gt; and prefix the function calls with stan::math::.\n\n\nExternal C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the Stan Math documentation.\nSuppose you have the following (nonsensical) model which relies on a function called my_dot_self. We will implement this as a copy of the built-in dot_self function.\nfunctions {\n  // both overloads end up using the same C++ template\n  real my_dot_self(vector theta);\n  real my_dot_self(row_vector theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] input_data;\n}\ntransformed data {\n  // no autodiff for data - will call using doubles\n  real ds = my_dot_self(input_data);\n}\nparameters {\n  row_vector[N] thetas;\n}\nmodel {\n  thetas ~ normal(0,1);\n  // autodiff - will call using stan::math::var types\n  input_data ~ normal(thetas, my_dot_self(thetas));\n}\nIf you wanted to autodiff through this function, the following header would suffice1:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr&gt;\ninline stan::value_type_t&lt;EigVec&gt; my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    const auto &x_ref = stan::math::to_ref(x);\n    stan::value_type_t&lt;EigVec&gt; sum_x = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum_x += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum_x;\n}\nHowever, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr,\n          stan::require_not_st_var&lt;EigVec&gt; * = nullptr&gt;\ninline double my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    auto x_ref = stan::math::to_ref(x);\n    double sum = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum;\n}\n\ntemplate &lt;typename EigVec, stan::require_eigen_vt&lt;stan::is_var, EigVec&gt; * = nullptr&gt;\ninline stan::math::var my_dot_self(const EigVec &v, std::ostream *pstream__)\n{\n    // (1) put v into our memory arena\n    stan::arena_t&lt;EigVec&gt; arena_v(v);\n    // (2) calculate forward pass using\n    // (3) the .val() method for matrices of var types\n    stan::math::var res = my_dot_self(arena_v.val(), pstream__);\n    // (4) Place a callback for the reverse pass on the callback stack.\n    stan::math::reverse_pass_callback(\n        [res, arena_v]() mutable\n        { arena_v.adj() += 2.0 * res.adj() * arena_v.val(); });\n    return res;\n}\nFor more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.\n\n\n\nSome functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.\n\nRNGs must end with _rng. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a stan::rng_t object (a type alias to boost::rng::mixmax).\nFunctions which edit the target directly must end with _lp and will be passed a reference to lp__ and a reference to a stan::math::accumulator object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter propto__ which controls whether or not constant terms can be dropped.\nProbability distributions must end with _lpdf or _lpmf and will be passed a boolean template parameter propto__ which controls whether or not constant terms can be dropped.",
     "crumbs": [
-      "Running CmdStan",
-      "Variational Inference using ADVI"
+      "Appendices",
+      "Using external C++ code"
     ]
   },
   {
-    "objectID": "cmdstan-guide/variational_config.html#variational-algorithms",
-    "href": "cmdstan-guide/variational_config.html#variational-algorithms",
-    "title": "Variational Inference using ADVI",
+    "objectID": "cmdstan-guide/external_code.html#derivative-specializations",
+    "href": "cmdstan-guide/external_code.html#derivative-specializations",
+    "title": "Using external C++ code",
     "section": "",
-    "text": "Stan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option algorithm=meanfield. The algorithm argument specifies the variational algorithm.\n\nalgorithm=meanfield - Use a fully factorized Gaussian for the approximation. This is the default algorithm.\nalgorithm=fullrank Use a Gaussian with a full-rank covariance matrix for the approximation.",
+    "text": "External C++ functions are currently the only way to encode a function with a known analytic gradient outside the Stan Math Library. This is done very similarly to how a function would be added to the Math library with a reverse-mode specialization. The following code is adapted from the Stan Math documentation.\nSuppose you have the following (nonsensical) model which relies on a function called my_dot_self. We will implement this as a copy of the built-in dot_self function.\nfunctions {\n  // both overloads end up using the same C++ template\n  real my_dot_self(vector theta);\n  real my_dot_self(row_vector theta);\n}\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] input_data;\n}\ntransformed data {\n  // no autodiff for data - will call using doubles\n  real ds = my_dot_self(input_data);\n}\nparameters {\n  row_vector[N] thetas;\n}\nmodel {\n  thetas ~ normal(0,1);\n  // autodiff - will call using stan::math::var types\n  input_data ~ normal(thetas, my_dot_self(thetas));\n}\nIf you wanted to autodiff through this function, the following header would suffice1:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr&gt;\ninline stan::value_type_t&lt;EigVec&gt; my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    const auto &x_ref = stan::math::to_ref(x);\n    stan::value_type_t&lt;EigVec&gt; sum_x = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum_x += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum_x;\n}\nHowever, we know the derivative of this function directly. To leverage this, we could use a more complicated form which has two function templates that differentiate themselves based on whether or not derivatives are required:\n#include &lt;stan/model/model_header.hpp&gt;\n#include &lt;ostream&gt;\n\ntemplate &lt;typename EigVec, stan::require_eigen_vector_t&lt;EigVec&gt; * = nullptr,\n          stan::require_not_st_var&lt;EigVec&gt; * = nullptr&gt;\ninline double my_dot_self(const EigVec &x, std::ostream *pstream__)\n{\n    auto x_ref = stan::math::to_ref(x);\n    double sum = 0.0;\n    for (int i = 0; i &lt; x.size(); ++i)\n    {\n        sum += x_ref.coeff(i) * x_ref.coeff(i);\n    }\n    return sum;\n}\n\ntemplate &lt;typename EigVec, stan::require_eigen_vt&lt;stan::is_var, EigVec&gt; * = nullptr&gt;\ninline stan::math::var my_dot_self(const EigVec &v, std::ostream *pstream__)\n{\n    // (1) put v into our memory arena\n    stan::arena_t&lt;EigVec&gt; arena_v(v);\n    // (2) calculate forward pass using\n    // (3) the .val() method for matrices of var types\n    stan::math::var res = my_dot_self(arena_v.val(), pstream__);\n    // (4) Place a callback for the reverse pass on the callback stack.\n    stan::math::reverse_pass_callback(\n        [res, arena_v]() mutable\n        { arena_v.adj() += 2.0 * res.adj() * arena_v.val(); });\n    return res;\n}\nFor more details about how to write C++ code using the Stan Math Library, see the Math library documentation at https://mc-stan.org/math/ or the paper at https://arxiv.org/abs/1509.07164.",
     "crumbs": [
-      "Running CmdStan",
-      "Variational Inference using ADVI"
+      "Appendices",
+      "Using external C++ code"
     ]
   },
   {
-    "objectID": "cmdstan-guide/variational_config.html#configuration",
-    "href": "cmdstan-guide/variational_config.html#configuration",
-    "title": "Variational Inference using ADVI",
+    "objectID": "cmdstan-guide/external_code.html#special-functions-rngs-distributions-editing-target",
+    "href": "cmdstan-guide/external_code.html#special-functions-rngs-distributions-editing-target",
+    "title": "Using external C++ code",
     "section": "",
-    "text": "iter=&lt;int&gt; Maximum number of iterations. Must be \\(&gt; 0\\). Default is \\(10000\\).\ngrad_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of gradients. Must be \\(&gt; 0\\). Default is \\(1\\).\nelbo_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of ELBO (objective function). Must be \\(&gt; 0\\). Default is \\(100\\).\neta=&lt;double&gt; Stepsize weighting parameter for adaptive stepsize sequence. Must be \\(&gt; 0\\). Default is \\(1.0\\).\nadapt Warmup Adaptation keyword, takes sub-arguments:\n\nengaged=&lt;boolean&gt; Adaptation engaged? Valid values: [true, false]. Default is true.\niter=&lt;int&gt; Maximum number of adaptation iterations. Must be \\(&gt; 0\\). Default is \\(50\\).\n\ntol_rel_obj=&lt;double&gt; Convergence tolerance on the relative norm of the objective. Must be \\(&gt; 0\\). Default is \\(0.01\\).\neval_elbo=&lt;int&gt; Evaluate ELBO every Nth iteration. Must be \\(&gt; 0\\). Default is 100.\noutput_samples=&lt;int&gt; Number of posterior samples to draw and save. Must be \\(&gt; 0\\). Default is 1000.",
+    "text": "Some functions have special meanings in Stan and place additional requirements on their signatures if used in external C++.\n\nRNGs must end with _rng. They will be passed a “base RNG object” as the second to last argument, before the pointer to the ostream. We recommend making this a template, since it may change. This is currently a stan::rng_t object (a type alias to boost::rng::mixmax).\nFunctions which edit the target directly must end with _lp and will be passed a reference to lp__ and a reference to a stan::math::accumulator object as the final parameters before the ostream pointer. They are also expected to have a boolean template parameter propto__ which controls whether or not constant terms can be dropped.\nProbability distributions must end with _lpdf or _lpmf and will be passed a boolean template parameter propto__ which controls whether or not constant terms can be dropped.",
     "crumbs": [
-      "Running CmdStan",
-      "Variational Inference using ADVI"
+      "Appendices",
+      "Using external C++ code"
     ]
   },
   {
-    "objectID": "cmdstan-guide/variational_config.html#csv-output",
-    "href": "cmdstan-guide/variational_config.html#csv-output",
-    "title": "Variational Inference using ADVI",
-    "section": "",
-    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the variational method is reported at the beginning of the sampler output file as CSV comments.\nThe first three output columns are labelled lp__, log_p__, log_g__, the rest are the model parameters.\nThe stepsize adaptation information is output as CSV comments following column header row.\nThe following line contains the mean of the variational approximation.\nThe rest of the output contains output_samples number of draws sampled from the variational approximation.\n\nTo illustrate, we call Stan’s variational inference on the example model and data:\n&gt; ./bernoulli variational data file=bernoulli.data.R\nBy default, the output file is output.csv.\nThe output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = variational\n#   variational\n#     algorithm = meanfield (Default)\n#       meanfield\n#     iter = 10000 (Default)\n#     grad_samples = 1 (Default)\n#     elbo_samples = 100 (Default)\n#     eta = 1 (Default)\n#     adapt\n#       engaged = true (Default)\n#       iter = 50 (Default)\n#     tol_rel_obj = 0.01 (Default)\n#     eval_elbo = 100 (Default)\n#     output_samples = 1000 (Default)\n...\nNext, the column header row:\nlp__,log_p__,log_g__,theta\nAdditional comments provide stepsize adaptation information:\n# Stepsize adaptation complete.\n# eta = 1\nFollowed by the data rows. The first line is special — it is the mean of the variational approximation.\n0,0,0,0.214911\nThat is, the estimate for theta given the data is 0.2.\nThe rest of the output contains output_samples number of draws samples from the variational approximation.\nThe following is a sample based on this approximation:\n0,-14.0252,-5.21718,0.770397\n0,-7.05063,-0.10025,0.162061\n0,-6.75031,-0.0191099,0.241606\n...\nThe header indicates the unnormalized log probability with lp__. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.",
+    "objectID": "cmdstan-guide/external_code.html#footnotes",
+    "href": "cmdstan-guide/external_code.html#footnotes",
+    "title": "Using external C++ code",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nDetails of programming in the Stan Math style are omitted from this section, it is presented only as an example↩︎",
     "crumbs": [
-      "Running CmdStan",
-      "Variational Inference using ADVI"
+      "Appendices",
+      "Using external C++ code"
     ]
   },
   {
-    "objectID": "functions-reference/binary_distributions.html",
-    "href": "functions-reference/binary_distributions.html",
-    "title": "Binary Distributions",
+    "objectID": "cmdstan-guide/index.html",
+    "href": "cmdstan-guide/index.html",
+    "title": "CmdStan User’s Guide",
     "section": "",
-    "text": "Binary probability distributions have support on \\(\\{0,1\\}\\), where 1 represents the value true and 0 the value false.\n\n\n\n\nIf \\(\\theta \\in [0,1]\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{Bernoulli}(y~|~\\theta) = \\left\\{ \\begin{array}{ll} \\theta &\n\\text{if } y = 1, \\text{ and} \\\\ 1 - \\theta & \\text{if } y = 0.\n\\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli(theta)\nIncrement target log probability density with bernoulli_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_lpmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lupmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal bernoulli_cdf(ints y | reals theta) The Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.0\n \n\nreal bernoulli_lcdf(ints y | reals theta) The log of the Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lccdf(ints y | reals theta) The log of the Bernoulli complementary cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nints bernoulli_rng(reals theta) Generate a Bernoulli variate with chance of success theta or an array of Bernoulli variates given an array of thetas of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{BernoulliLogit}(y~|~\\alpha) = \\text{Bernoulli}(y |\n\\text{logit}^{-1}(\\alpha)) = \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha) & \\text{if } y = 1, \\text{ and} \\\\ 1 -\n\\text{logit}^{-1}(\\alpha) &  \\text{if } y = 0. \\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli_logit(alpha)\nIncrement target log probability density with bernoulli_logit_lupmf(y | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_logit_lpmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha)\nAvailable since 2.12\n \n\nreal bernoulli_logit_lupmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha) dropping constant additive terms\nAvailable since 2.25\n \n\nR bernoulli_logit_rng(reals alpha) Generate a Bernoulli variate with chance of success \\(\\text{logit}^{-1}(\\alpha)\\); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e. a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in {\\{0,1\\}}^n\\), \\[\\begin{align*}\n&\\text{BernoulliLogitGLM}(y~|~x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Bernoulli}(y_i~|~\\text{logit}^{-1}(\\alpha_i + x_i\\cdot\n\\beta))\\\\ &= \\prod_{1\\leq i \\leq n} \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j)\n& \\text{if } y_i = 1, \\text{ and} \\\\ 1 - \\text{logit}^{-1}(\\alpha_i +\n\\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) & \\text{if } y_i = 0.\n\\end{array} \\right. \\end{align*}\\]\n\n\n\ny ~ bernoulli_logit_glm(x, alpha, beta)\nIncrement target log probability density with bernoulli_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.25\n \n\n\n\n\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\narray[] int bernoulli_logit_glm_rng(matrix x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29\n \n\narray[] int bernoulli_logit_glm_rng(row_vector x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29",
+    "text": "This document is a user’s guide for CmdStan, the command-line interface to the Stan statistical modeling language. CmdStan provides the programs and tools to compile Stan programs into C++ executables that can be run directly from the command line, together with a few utilities to check and summarize the resulting outputs.\nIn CmdStan, statistical models written in the Stan probabilistic programming language are translated into a C++ program which is then compiled together with the CmdStan routines that provide the logic needed to manage all user inputs and program outputs and the Stan inference algorithms and math library. The resulting command line executable program can be used to\n\ndo inference on data, producing an exact or approximate estimate of the posterior;\ngenerate new quantities of interest from an existing estimate;\ngenerate data from the model according to a given set of parameters.\n\nThe packages CmdStanR and CmdStanPy provide interfaces to CmdStan from R and Python, respectively, similarly, JuliaStan also interfaces with CmdStan.\nDownload the pdf version of this manual.\n\nBenefits of CmdStan\n\nWith every new Stan release, there is a corresponding CmdStan release, therefore CmdStan provides access to the latest version of Stan, and can be used to run the development version of Stan as well.\nOf the Stan interfaces, CmdStan has the lightest memory footprint, therefore it can fit larger and more complex models. It has has the fewest dependencies, which makes it easier to run in limited environments such as clusters.\nThe output generated is in CSV format and can be post-processed using other Stan interfaces or general tools.\n\n\n\nStan documentation\n\nStan User’s Guide The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:\nStan Reference Manual Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.\nStan Functions Reference The Stan Functions Reference provides definitions and examples for all the functions defined in the Stan math library and available in the Stan programming language, including all probability distributions.\n\n\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Binary Distributions"
+      "CmdStan User's Guide"
     ]
   },
   {
-    "objectID": "functions-reference/binary_distributions.html#bernoulli-distribution",
-    "href": "functions-reference/binary_distributions.html#bernoulli-distribution",
-    "title": "Binary Distributions",
+    "objectID": "cmdstan-guide/json_apdx.html",
+    "href": "cmdstan-guide/json_apdx.html",
+    "title": "JSON Format for CmdStan",
     "section": "",
-    "text": "If \\(\\theta \\in [0,1]\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{Bernoulli}(y~|~\\theta) = \\left\\{ \\begin{array}{ll} \\theta &\n\\text{if } y = 1, \\text{ and} \\\\ 1 - \\theta & \\text{if } y = 0.\n\\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli(theta)\nIncrement target log probability density with bernoulli_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_lpmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lupmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal bernoulli_cdf(ints y | reals theta) The Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.0\n \n\nreal bernoulli_lcdf(ints y | reals theta) The log of the Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lccdf(ints y | reals theta) The log of the Bernoulli complementary cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nints bernoulli_rng(reals theta) Generate a Bernoulli variate with chance of success theta or an array of Bernoulli variates given an array of thetas of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "CmdStan can use JSON format for input data for both model data and parameters. Model data is read in by the model constructor. Model parameters are used to initialize the sampler and optimizer.\n\n\nYou can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the write_stan_json() function provided by the CmdStanR interface.\n\n\n\nJSON is a data interchange notation, defined by an EMCA standard. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:\n\nStructural tokens are the left and right curly bracket {}, left and right square bracket [], the semicolon ;, and the comma ,.\nLiteral tokens must always be in lowercase. There are three literal tokens: true, false, null.\nA primitive value is a single token which is either a literal, a string, or a number.\nA string consists of zero or more Unicode characters enclosed in double quotes, e.g. \"foo\". A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g. \"\\\\uHHHH\" where HHHH is the Unicode code point in hex.\nNumbers are represented using either decimal notation or scientific notation. The following are examples of numbers: 17, 17.2, -17.2, -17.2e8, 17.2e-8.  There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.\nThe special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string \"Inf\", the string \"Infinity\", or the atom Infinity. Negative infinity can be represented as the string \"-Inf\", the string \"-Infinity\", or the atom -Infinity. Not-a-number can be represented as the string \"NaN\" or the atom NaN. These values may be mixed with other numerical types.\nA complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number \\(2.3 - 1.83i\\) would be represented in JSON as the two-element array [2.3, -1.83].\nA JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: [], [1], [0.2, \"-inf\", true].\nVectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector \\([1 \\quad 2]^{\\top}\\) and the row vector \\([1 \\quad 2]\\) are represented by the JSON array [1, 2].\nComplex vectors are represented as arrays of two-element arrays. For example, the complex vector \\([2.3 - 1.83i \\quad -4.8 +\n2i]^{\\top}\\) is represented as [[2.3, -1.83], [-4.8, 2]] in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).\nMatrices are represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 & 2.7 & -9.8 \\\\\n4.2 & 1.8 & -7.3\n\\end{bmatrix}\n\\end{equation*}\\] is represented in JSON as [[1, 2.7, -9.8],   [4.2, 1.8, -7.3]].\nComplex matrices are also represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) complex matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 + 2i & 3 - 4.2i & 13.1 + 2.7i \\\\\n3.1 & -5i & 0\n\\end{bmatrix}\n\\end{equation*}\\] would be represented in JSON as [[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]].\nTuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple (1.5, 3.4) is represented in JSON as {\"1\": 1.5, \"2\": 3.4}.\nA name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.\nA JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: { }, {\"foo\": null}, {\"bar\" : 17, \"baz\" : [14,15,16.6] }.\n\n\n\n\nStan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:\n{ \"N\" : 10, \"y\" : [0,1,0,0,0,0,0,0,0,1] }\nMatrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:\ndata {\n  int d1;\n  int d2;\n  int d3;\n  array[d1, d2, d3] int ar;\n}\nthe following JSON input would be valid:\n{ \"d1\" : 2,\n  \"d2\" : 3,\n  \"d3\" : 4,\n  \"ar\" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],\n          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]\n}\nJSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.\nAll data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.\n\n\n\n\n\n\n\nStan declaration\nJSON encoding\n\n\n\n\nint i\n\"i\": 17\n\n\n\n\n\n\nreal a\n\"a\" : 17\n\n\n\n\"a\" : 17.2\n\n\n\n\"a\" : \"NaN\"\n\n\n\n\"a\" : \"+inf\"\n\n\n\n\"a\" : \"-inf\"\n\n\n\n\n\n\ncomplex z\n\"z\": [1, -2.3]\n\n\narray[5] int\n\"a\" : [1, 2, 3, 4, 5]\n\n\narray[5] real a\n\"a\" : [ 1, 2, 3.3, \"NaN\", 5 ]\n\n\narray[2] complex b\n\"b\" : [[1, -2.3], [4.9, 0]]\n\n\nvector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nrow_vector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nmatrix[2, 3] a\n\"a\" : [[ 1, 2, 3 ], [ 4, 5, 6]]\n\n\ncomplex_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_row_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_matrix[2, 3] d\n\"d\" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]\n\n\ntuple(real, array[2] int) t\n\"t\" : { \"1\": 1.4, \"2\": [1, 2]}\n\n\n\n\n\nJSON notation is not able to distinguish between multi-dimensional arrays where any dimension is \\(0\\), e.g., a 2-D array with dimensions \\((1,0)\\), i.e., an array which contains a single array which is empty, has JSON representation [ ]. To see how this works, consider the following Stan program data block:\ndata {\n  int d;\n  array[d] int ar_1d;\n  array[d, d] int ar_2d;\n  array[d, d, d] int ar_3d;\n}\nIn the case where variable d is 1, all arrays will contain a single value. If array variable ar_d1 contains value 7, 2-D array variable ar_d2 contains (an array which contains) value 8, and 3-D array variable ar_d3 contains (an array which contains an array which contains) value 9, the JSON representation is:\n{ \"ar_d1\" : [7],\n  \"ar_d2\" : [[8]],\n  \"ar_d3\" : [[[9]]]\n}\nHowever, in the case where variable d is 0, ar_d1 is empty, i.e., it contains no values, as is ar_d2, ar_d3, and the JSON representation is\n{ \"d\" : 0,\n  \"ar_d1\" : [ ],\n  \"ar_d2\" : [ ],\n  \"ar_d3\" : [ ]\n}",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Binary Distributions"
+      "Appendices",
+      "JSON Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/binary_distributions.html#bernoulli-logit-distribution",
-    "href": "functions-reference/binary_distributions.html#bernoulli-logit-distribution",
-    "title": "Binary Distributions",
+    "objectID": "cmdstan-guide/json_apdx.html#creating-json-files",
+    "href": "cmdstan-guide/json_apdx.html#creating-json-files",
+    "title": "JSON Format for CmdStan",
     "section": "",
-    "text": "Stan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{BernoulliLogit}(y~|~\\alpha) = \\text{Bernoulli}(y |\n\\text{logit}^{-1}(\\alpha)) = \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha) & \\text{if } y = 1, \\text{ and} \\\\ 1 -\n\\text{logit}^{-1}(\\alpha) &  \\text{if } y = 0. \\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli_logit(alpha)\nIncrement target log probability density with bernoulli_logit_lupmf(y | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_logit_lpmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha)\nAvailable since 2.12\n \n\nreal bernoulli_logit_lupmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha) dropping constant additive terms\nAvailable since 2.25\n \n\nR bernoulli_logit_rng(reals alpha) Generate a Bernoulli variate with chance of success \\(\\text{logit}^{-1}(\\alpha)\\); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "You can create the JSON file yourself using the guidelines below, but a more convenient way to create a JSON file for use with CmdStan is to use the write_stan_json() function provided by the CmdStanR interface.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Binary Distributions"
+      "Appendices",
+      "JSON Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/binary_distributions.html#bernoulli-logit-glm",
-    "href": "functions-reference/binary_distributions.html#bernoulli-logit-glm",
-    "title": "Binary Distributions",
+    "objectID": "cmdstan-guide/json_apdx.html#json-syntax-summary",
+    "href": "cmdstan-guide/json_apdx.html#json-syntax-summary",
+    "title": "JSON Format for CmdStan",
     "section": "",
-    "text": "Stan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e. a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in {\\{0,1\\}}^n\\), \\[\\begin{align*}\n&\\text{BernoulliLogitGLM}(y~|~x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Bernoulli}(y_i~|~\\text{logit}^{-1}(\\alpha_i + x_i\\cdot\n\\beta))\\\\ &= \\prod_{1\\leq i \\leq n} \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j)\n& \\text{if } y_i = 1, \\text{ and} \\\\ 1 - \\text{logit}^{-1}(\\alpha_i +\n\\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) & \\text{if } y_i = 0.\n\\end{array} \\right. \\end{align*}\\]\n\n\n\ny ~ bernoulli_logit_glm(x, alpha, beta)\nIncrement target log probability density with bernoulli_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.25\n \n\n\n\n\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\narray[] int bernoulli_logit_glm_rng(matrix x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29\n \n\narray[] int bernoulli_logit_glm_rng(row_vector x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29",
+    "text": "JSON is a data interchange notation, defined by an EMCA standard. JSON data files must in Unicode. JSON data is a series of structural tokens, literal tokens, and values:\n\nStructural tokens are the left and right curly bracket {}, left and right square bracket [], the semicolon ;, and the comma ,.\nLiteral tokens must always be in lowercase. There are three literal tokens: true, false, null.\nA primitive value is a single token which is either a literal, a string, or a number.\nA string consists of zero or more Unicode characters enclosed in double quotes, e.g. \"foo\". A backslash is used to escape the double quote character as well as the backslash itself. JSON allows the use of Unicode character escapes, e.g. \"\\\\uHHHH\" where HHHH is the Unicode code point in hex.\nNumbers are represented using either decimal notation or scientific notation. The following are examples of numbers: 17, 17.2, -17.2, -17.2e8, 17.2e-8.  There is no distinction between integer and real numbers in the JSON format other than whether they have periods or scientific notation.\nThe special floating point values for positive infinity, negative infinity, and not-a-number can be represented in multiple ways. Positive infinity can be represented as the string \"Inf\", the string \"Infinity\", or the atom Infinity. Negative infinity can be represented as the string \"-Inf\", the string \"-Infinity\", or the atom -Infinity. Not-a-number can be represented as the string \"NaN\" or the atom NaN. These values may be mixed with other numerical types.\nA complex scalar is represented as a two-element array consisting of its real component followed by its imaginary component. For example, the complex number \\(2.3 - 1.83i\\) would be represented in JSON as the two-element array [2.3, -1.83].\nA JSON array is an ordered, comma-separated list of zero or more JSON values enclosed in square brackets. The elements of an array can be of any type. The following are examples of arrays: [], [1], [0.2, \"-inf\", true].\nVectors and row vectors in JSON are representing as arrays of their elements. For example, both the vector \\([1 \\quad 2]^{\\top}\\) and the row vector \\([1 \\quad 2]\\) are represented by the JSON array [1, 2].\nComplex vectors are represented as arrays of two-element arrays. For example, the complex vector \\([2.3 - 1.83i \\quad -4.8 +\n2i]^{\\top}\\) is represented as [[2.3, -1.83], [-4.8, 2]] in JSON. A complex row vector has the same representation as its transpose (the vector with the same elements).\nMatrices are represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 & 2.7 & -9.8 \\\\\n4.2 & 1.8 & -7.3\n\\end{bmatrix}\n\\end{equation*}\\] is represented in JSON as [[1, 2.7, -9.8],   [4.2, 1.8, -7.3]].\nComplex matrices are also represented as arrays of their row vectors. For example, the \\(2 \\times 3\\) complex matrix \\[\\begin{equation*}\n\\begin{bmatrix}\n1 + 2i & 3 - 4.2i & 13.1 + 2.7i \\\\\n3.1 & -5i & 0\n\\end{bmatrix}\n\\end{equation*}\\] would be represented in JSON as [[[1, 2], [3, -4.2], [13.1, 2.7]],  [[3.1,  0], [0, -5], [0, 0]]].\nTuples are written as nested JSON objects where the keys are strings for the numbered slots in the tuple. For example, the tuple (1.5, 3.4) is represented in JSON as {\"1\": 1.5, \"2\": 3.4}.\nA name-value pair consists of a string followed by a colon followed by a value, either primitive or compound.\nA JSON object is a comma-separated series of zero or more name-value pairs enclosed in curly brackets. Each name-value pair is a member of the object. Membership is unordered. Member names are not required to be unique. The following are examples of objects: { }, {\"foo\": null}, {\"bar\" : 17, \"baz\" : [14,15,16.6] }.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Binary Distributions"
+      "Appendices",
+      "JSON Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html",
-    "href": "functions-reference/bounded_discrete_distributions.html",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/json_apdx.html#stan-data-types-in-json-notation",
+    "href": "cmdstan-guide/json_apdx.html#stan-data-types-in-json-notation",
+    "title": "JSON Format for CmdStan",
     "section": "",
-    "text": "Bounded discrete probability functions have support on \\(\\{ 0, \\ldots,\nN \\}\\) for some upper bound \\(N\\).\n\n\n\n\nSuppose \\(N \\in \\mathbb{N}\\) and \\(\\theta \\in [0,1]\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). \\[\\begin{equation*} \\text{Binomial}(n~|~N,\\theta) = \\binom{N}{n}\n\\theta^n (1 - \\theta)^{N - n}. \\end{equation*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{Binomial}(n~|~N,\\theta) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt] & & {\n} + n \\log \\theta + (N - n) \\log (1 - \\theta), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\theta} \\log \\text{Binomial}(n~|~N,\\theta)\n= \\frac{n}{\\theta} - \\frac{N - n}{1 - \\theta} \\end{equation*}\\]\n\n\n\nn ~ binomial(N, theta)\nIncrement target log probability density with binomial_lupmf(n | N, theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_lpmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lupmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal binomial_cdf(ints n | ints N, reals theta) The binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.0\n \n\nreal binomial_lcdf(ints n | ints N, reals theta) The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lccdf(ints n | ints N, reals theta) The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nR binomial_rng(ints N, reals theta) Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{eqnarray*}\n\\text{BinomialLogit}(n~|~N,\\alpha) & = &\n\\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha)) \\\\[6pt] & = &\n\\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha) \\right)^{n}  \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right)^{N - n}.  \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{BinomialLogit}(n~|~N,\\alpha) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt]   & &\n{ } + n \\log \\text{logit}^{-1}(\\alpha) + (N - n) \\log \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\alpha} \\log\n\\text{BinomialLogit}(n~|~N,\\alpha) =\n\\frac{n}{\\text{logit}^{-1}(-\\alpha)} - \\frac{N -\nn}{\\text{logit}^{-1}(\\alpha)} \\end{equation*}\\]\n\n\n\nn ~ binomial_logit(N, alpha)\nIncrement target log probability density with binomial_logit_lupmf(n | N, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_logit_lpmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha\nAvailable since 2.12\n \n\nreal binomial_logit_lupmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms\nAvailable since 2.25\n\n\n\n\nStan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta \\in \\mathbb{R}^m\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{align*}\n  &\\text{BinomialLogitGLM}(n~|~N, x, \\alpha, \\beta) = \\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha_i + x_i \\cdot \\beta)) \\\\\n  &= \\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{n}  \\left( 1 - \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{N - n}.\n\\end{align*}\\]\n\n\n\nn ~ binomial_logit_glm(N, x, alpha, beta)\nIncrement target log probability density with binomial_logit_glm_lupmf(n | N, x, alpha, beta).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n\n\n\n\n\n\nIf \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in\n\\mathbb{R}^+\\), then for \\(n \\in {0,\\ldots,N}\\), \\[\\begin{equation*}\n\\text{BetaBinomial}(n~|~N,\\alpha,\\beta) = \\binom{N}{n}\n\\frac{\\mathrm{B}(n+\\alpha, N -n +   \\beta)}{\\mathrm{B}(\\alpha,\\beta)},\n\\end{equation*}\\] where the beta function \\(\\mathrm{B}(u,v)\\) is defined for \\(u \\in\n\\mathbb{R}^+\\) and \\(v \\in \\mathbb{R}^+\\) by \\[\\begin{equation*} \\mathrm{B}(u,v) =\n\\frac{\\Gamma(u) \\ \\Gamma(v)}{\\Gamma(u + v)}. \\end{equation*}\\]\n\n\n\nn ~ beta_binomial(N, alpha, beta)\nIncrement target log probability density with beta_binomial_lupmf(n | N, alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_binomial_lpmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lupmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_binomial_cdf(ints n | ints N, reals alpha, reals beta) The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.0\n \n\nreal beta_binomial_lcdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lccdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nR beta_binomial_rng(ints N, reals alpha, reals beta) Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(a \\in \\mathbb{N}\\), \\(b \\in \\mathbb{N}\\), and \\(N \\in\n\\{0,\\ldots,a+b\\}\\), then for \\(n \\in \\{\\max(0,N-b),\\ldots,\\min(a,N)\\}\\), \\[\\begin{equation*} \\text{Hypergeometric}(n~|~N,a,b) = \\frac{\\normalsize{\\binom{a}{n}\n\\binom{b}{N - n}}}      {\\normalsize{\\binom{a + b}{N}}}. \\end{equation*}\\]\n\n\n\nn ~ hypergeometric(N, a, b)\nIncrement target log probability density with hypergeometric_lupmf(n | N, a, b).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal hypergeometric_lpmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b\nAvailable since 2.12\n \n\nreal hypergeometric_lupmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms\nAvailable since 2.25\n \n\nint hypergeometric_rng(int N, int a, int b) Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(N \\in \\mathbb{N}\\), \\(N &gt; 0\\), and if \\(\\theta \\in \\mathbb{R}^N\\) forms an \\(N\\)-simplex (i.e., has nonnegative entries summing to one), then for \\(y \\in \\{1,\\ldots,N\\}\\), \\[\\begin{equation*} \\text{Categorical}(y~|~\\theta) =\n\\theta_y. \\end{equation*}\\] In addition, Stan provides a log-odds scaled categorical distribution, \\[\\begin{equation*} \\text{CategoricalLogit}(y~|~\\beta) =\n\\text{Categorical}(y~|~\\text{softmax}(\\beta)). \\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical(theta)\nIncrement target log probability density with categorical_lupmf(y | theta) dropping constant additive terms.\nAvailable since 2.0\n \n\n\n\n\ny ~ categorical_logit(beta)\nIncrement target log probability density with categorical_logit_lupmf(y | beta).\nAvailable since 2.4\n \n\n\n\n\nAll of the categorical distributions are vectorized so that the outcome y can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal categorical_lpmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.12\n \n\nreal categorical_lupmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.25\n \n\nreal categorical_logit_lpmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta.\nAvailable since 2.12\n \n\nreal categorical_logit_lupmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta dropping constant additive terms.\nAvailable since 2.25\n \n\nint categorical_rng(vector theta) Generate a categorical variate with \\(N\\)-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nint categorical_logit_rng(vector beta) Generate a categorical variate with outcome in range \\(1:N\\) from log-odds vector beta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.16\n\n\n\n\nStan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e. a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.\nNote that the implementation does not put any restrictions on the coefficient matrix \\(\\beta\\). It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the Stan User’s Guide.\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\), \\(N,M,K &gt; 0\\), and if \\(x\\in \\mathbb{R}^{M\\times K}, \\alpha \\in \\mathbb{R}^N, \\beta\\in \\mathbb{R}^{K\\cdot N}\\), then for \\(y \\in \\{1,\\ldots,N\\}^M\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{CategoricalLogitGLM}(y~|~x,\\alpha,\\beta)\n& =  \\prod_{1\\leq i \\leq M}\\text{CategoricalLogit}(y_i~|~\\alpha+x_i\\cdot\\beta)  \\\\[8pt]\n& =  \\prod_{1\\leq i \\leq M}\\text{Categorical}(y_i~|~softmax(\\alpha+x_i\\cdot\\beta)).\n\\end{split}\n\\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical_logit_glm(x, alpha, beta)\nIncrement target log probability density with categorical_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal categorical_logit_glm_lpmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(l, u \\in \\mathbb{Z}\\) are lower and upper bounds (\\(l \\le u\\)), then for any integer \\(y \\in \\{l,\\ldots,u\\}\\), \\[\\begin{equation*} \\text{DiscreteRange}(y ~|~ l, u) =\n\\frac{1}{u - l + 1}. \\end{equation*}\\]\n\n\n\ny ~ discrete_range(l, u)\nIncrement the target log probability density with discrete_range_lupmf(y | l, u) dropping constant additive terms.\nAvailable since 2.26\n \n\n\n\n\nAll of the discrete range distributions are vectorized so that the outcome y and the bounds l, u can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal discrete_range_lpmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\).\nAvailable since 2.26\n \n\nreal discrete_range_lupmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\) dropping constant additive terms.\nAvailable since 2.26\n \n\nreal discrete_range_cdf(ints y | ints l, ints u) The discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lcdf(ints y | ints l, ints u) The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lccdf(ints y | ints l, ints u) The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nints discrete_range_rng(ints l, ints u) Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.26\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedLogistic}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\text{logit}^{-1}(\\eta - c_1)  &  \\text{if } k = 1, \\\\[4pt]\n\\text{logit}^{-1}(\\eta - c_{k-1}) - \\text{logit}^{-1}(\\eta - c_{k})  &\n\\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt] \\text{logit}^{-1}(\\eta -\nc_{K-1}) - 0  &  \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\nk ~ ordered_logistic(eta, c)\nIncrement target log probability density with ordered_logistic_lupmf(k | eta, c).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal ordered_logistic_lpmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_logistic_lupmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nint ordered_logistic_rng(real eta, vector c) Generate an ordered logistic variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n\n\n\n\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\) with \\(N, M &gt; 0\\), \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(x\\in \\mathbb{R}^{N\\times M}, \\beta\\in \\mathbb{R}^M\\), then for \\(y \\in \\{1,\\ldots,K\\}^N\\), \\[\\begin{equation*}\n\\begin{split}\n\\\\\n& \\text{OrderedLogisticGLM}(y~|~x,\\beta,c) \\\\[8pt]\n& = \\prod_{1\\leq i \\leq N}\\text{OrderedLogistic}(y_i~|~x_i\\cdot \\beta,c) \\\\\n& = \\prod_{1\\leq i \\leq N}  \\left\\{ \\begin{array}{ll}\n1 - \\text{logit}^{-1}(x_i\\cdot \\beta - c_1)  &  \\text{if } y = 1, \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{y-1}) - \\text{logit}^{-1}(x_i\\cdot \\beta - c_{y}) & \\text{if } 1 &lt; y &lt; K, \\text{and} \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{K-1}) - 0  &  \\text{if } y = K.\n\\end{array} \\right.\n\\end{split}\n\\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(y=1\\) and \\(y=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\ny ~ ordered_logistic_glm(x, beta, c)\nIncrement target log probability density with ordered_logistic_lupmf(y | x, beta, c).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal ordered_logistic_glm_lpmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedProbit}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\Phi(\\eta - c_1) & \\text{if } k = 1, \\\\[4pt] \\Phi(\\eta - c_{k-1}) -\n\\Phi(\\eta - c_{k})  & \\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt]\n\\Phi(\\eta - c_{K-1}) - 0 & \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 =\n-\\infty\\) and \\(c_K = +\\infty\\) with \\(\\Phi(-\\infty) = 0\\) and \\(\\Phi(\\infty) = 1\\).\n\n\n\nk ~ ordered_probit(eta, c)\nIncrement target log probability density with ordered_probit_lupmf(k | eta, c).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal ordered_probit_lpmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_probit_lupmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nreal ordered_probit_lpmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.\nAvailable since 2.19\n \n\nreal ordered_probit_lupmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.19\n \n\nint ordered_probit_rng(real eta, vector c) Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "text": "Stan follows the JSON standard. A Stan input file in JSON notation consists of single JSON object which contains zero or more name-value pairs. This structure corresponds to a Python data dictionary object. The following is an example of JSON data for the simple Bernoulli example model:\n{ \"N\" : 10, \"y\" : [0,1,0,0,0,0,0,0,0,1] }\nMatrix data and multi-dimensional arrays are indexed in row-major order. For a Stan program which has data block:\ndata {\n  int d1;\n  int d2;\n  int d3;\n  array[d1, d2, d3] int ar;\n}\nthe following JSON input would be valid:\n{ \"d1\" : 2,\n  \"d2\" : 3,\n  \"d3\" : 4,\n  \"ar\" : [[[0,1,2,3], [4,5,6,7], [8,9,10,11]],\n          [[12,13,14,15], [16,17,18,19], [20,21,22,23]]]\n}\nJSON ignores whitespace. In the above examples, the spaces and newlines are only used to improve readability and can be omitted.\nAll data inputs are encoded as name-value pairs. The following table provides more examples of JSON data. The left column contains a Stan data variable declaration and the right column contains valid JSON data inputs.\n\n\n\n\n\n\n\nStan declaration\nJSON encoding\n\n\n\n\nint i\n\"i\": 17\n\n\n\n\n\n\nreal a\n\"a\" : 17\n\n\n\n\"a\" : 17.2\n\n\n\n\"a\" : \"NaN\"\n\n\n\n\"a\" : \"+inf\"\n\n\n\n\"a\" : \"-inf\"\n\n\n\n\n\n\ncomplex z\n\"z\": [1, -2.3]\n\n\narray[5] int\n\"a\" : [1, 2, 3, 4, 5]\n\n\narray[5] real a\n\"a\" : [ 1, 2, 3.3, \"NaN\", 5 ]\n\n\narray[2] complex b\n\"b\" : [[1, -2.3], [4.9, 0]]\n\n\nvector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nrow_vector[5] a\n\"a\" : [1, 2, 3.3, \"NaN\", 5]\n\n\nmatrix[2, 3] a\n\"a\" : [[ 1, 2, 3 ], [ 4, 5, 6]]\n\n\ncomplex_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_row_vector[2] c\n\"c\" : [[-1.2, 3.3], [4.8, 1.9], [2.3, 0]]\n\n\ncomplex_matrix[2, 3] d\n\"d\" : [[[1, 1], [2, 2], [3, 3]], [4, 4], [5, 5], [6, 6]]]\n\n\ntuple(real, array[2] int) t\n\"t\" : { \"1\": 1.4, \"2\": [1, 2]}\n\n\n\n\n\nJSON notation is not able to distinguish between multi-dimensional arrays where any dimension is \\(0\\), e.g., a 2-D array with dimensions \\((1,0)\\), i.e., an array which contains a single array which is empty, has JSON representation [ ]. To see how this works, consider the following Stan program data block:\ndata {\n  int d;\n  array[d] int ar_1d;\n  array[d, d] int ar_2d;\n  array[d, d, d] int ar_3d;\n}\nIn the case where variable d is 1, all arrays will contain a single value. If array variable ar_d1 contains value 7, 2-D array variable ar_d2 contains (an array which contains) value 8, and 3-D array variable ar_d3 contains (an array which contains an array which contains) value 9, the JSON representation is:\n{ \"ar_d1\" : [7],\n  \"ar_d2\" : [[8]],\n  \"ar_d3\" : [[[9]]]\n}\nHowever, in the case where variable d is 0, ar_d1 is empty, i.e., it contains no values, as is ar_d2, ar_d3, and the JSON representation is\n{ \"d\" : 0,\n  \"ar_d1\" : [ ],\n  \"ar_d2\" : [ ],\n  \"ar_d3\" : [ ]\n}",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Appendices",
+      "JSON Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#binomial-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/log_prob_config.html",
+    "href": "cmdstan-guide/log_prob_config.html",
+    "title": "Extracting log probabilities and gradients for diagnostics",
     "section": "",
-    "text": "Suppose \\(N \\in \\mathbb{N}\\) and \\(\\theta \\in [0,1]\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). \\[\\begin{equation*} \\text{Binomial}(n~|~N,\\theta) = \\binom{N}{n}\n\\theta^n (1 - \\theta)^{N - n}. \\end{equation*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{Binomial}(n~|~N,\\theta) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt] & & {\n} + n \\log \\theta + (N - n) \\log (1 - \\theta), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\theta} \\log \\text{Binomial}(n~|~N,\\theta)\n= \\frac{n}{\\theta} - \\frac{N - n}{1 - \\theta} \\end{equation*}\\]\n\n\n\nn ~ binomial(N, theta)\nIncrement target log probability density with binomial_lupmf(n | N, theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_lpmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lupmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal binomial_cdf(ints n | ints N, reals theta) The binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.0\n \n\nreal binomial_lcdf(ints n | ints N, reals theta) The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lccdf(ints n | ints N, reals theta) The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nR binomial_rng(ints N, reals theta) Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "CmdStan can return the computed log probability and the gradient with respect to a set of parameters.\nThis is similar to the diagnose subcommand, but the output format differs and the results here are not compared with those from finite differences.\nNote: Startup and data initialization costs mean that this method is not an efficient way to calculate these quantities. It is provided only for convenience and should not be used for serious computation.\n\n\nThis method takes 3 arguments:\n\njacobian - Whether or not the Jacobian adjustment for constrained parameters should be included in the gradient. Default value is true (include adjustment).\nconstrained_params - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using JSON format. Alternatively, the input file can be set of draws in StanCSV format.\nunconstrained_params - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called params_r, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.\n\nOnly one of constrained_params and unconstrained_params can be specified.\nFor more on the differences between constrained and unconstrained parameters, see the Stan reference manual section on variable transforms.\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the log_prob method is reported at the beginning of the output file as CSV comments.\nColumn headers, the first column is labelled lp__, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.\nValues which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).\n\nFor example, if we have a file called params.json:\n{\n    \"theta\" : 0.1\n}\nWe can run the example model:\n/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json\nThis yields\n# method = log_prob\n#   log_prob\n#     unconstrained_params =  (Default)\n#     constrained_params = params.json\n#     jacobian = true (Default)\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2390820139 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlp_,theta\n-7.856,1.8",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Extracting log probabilities and gradients for diagnostics"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-distribution-logit-parameterization",
-    "href": "functions-reference/bounded_discrete_distributions.html#binomial-distribution-logit-parameterization",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/log_prob_config.html#configuration",
+    "href": "cmdstan-guide/log_prob_config.html#configuration",
+    "title": "Extracting log probabilities and gradients for diagnostics",
     "section": "",
-    "text": "Stan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{eqnarray*}\n\\text{BinomialLogit}(n~|~N,\\alpha) & = &\n\\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha)) \\\\[6pt] & = &\n\\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha) \\right)^{n}  \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right)^{N - n}.  \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{BinomialLogit}(n~|~N,\\alpha) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt]   & &\n{ } + n \\log \\text{logit}^{-1}(\\alpha) + (N - n) \\log \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\alpha} \\log\n\\text{BinomialLogit}(n~|~N,\\alpha) =\n\\frac{n}{\\text{logit}^{-1}(-\\alpha)} - \\frac{N -\nn}{\\text{logit}^{-1}(\\alpha)} \\end{equation*}\\]\n\n\n\nn ~ binomial_logit(N, alpha)\nIncrement target log probability density with binomial_logit_lupmf(n | N, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_logit_lpmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha\nAvailable since 2.12\n \n\nreal binomial_logit_lupmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms\nAvailable since 2.25",
+    "text": "This method takes 3 arguments:\n\njacobian - Whether or not the Jacobian adjustment for constrained parameters should be included in the gradient. Default value is true (include adjustment).\nconstrained_params - Input file of parameters values on the constrained scale. A single set of constrained parameters can be specified using JSON format. Alternatively, the input file can be set of draws in StanCSV format.\nunconstrained_params - Input file (JSON or R dump) of parameter values on unconstrained scale. These files should contain a single variable, called params_r, which is a flattened vector of all unconstrained parameters. If this object is two dimensional, each entry should be a vector of the same form and the output will feature multiple rows.\n\nOnly one of constrained_params and unconstrained_params can be specified.\nFor more on the differences between constrained and unconstrained parameters, see the Stan reference manual section on variable transforms.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Extracting log probabilities and gradients for diagnostics"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-logit-glm",
-    "href": "functions-reference/bounded_discrete_distributions.html#binomial-logit-glm",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/log_prob_config.html#csv-output",
+    "href": "cmdstan-guide/log_prob_config.html#csv-output",
+    "title": "Extracting log probabilities and gradients for diagnostics",
     "section": "",
-    "text": "Stan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta \\in \\mathbb{R}^m\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{align*}\n  &\\text{BinomialLogitGLM}(n~|~N, x, \\alpha, \\beta) = \\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha_i + x_i \\cdot \\beta)) \\\\\n  &= \\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{n}  \\left( 1 - \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{N - n}.\n\\end{align*}\\]\n\n\n\nn ~ binomial_logit_glm(N, x, alpha, beta)\nIncrement target log probability density with binomial_logit_glm_lupmf(n | N, x, alpha, beta).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34",
+    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the log_prob method is reported at the beginning of the output file as CSV comments.\nColumn headers, the first column is labelled lp__, and the rest are named after parameters. These will be the unconstrained parameters, regardless of whether constrained or unconstrained parameters were supplied as input.\nValues which correspond to the value of the log density (column 1) and the gradient with respect to each parameter (remaining columns).\n\nFor example, if we have a file called params.json:\n{\n    \"theta\" : 0.1\n}\nWe can run the example model:\n/bernoulli log_prob constrained_params=params.json data file=bernoulli.data.json\nThis yields\n# method = log_prob\n#   log_prob\n#     unconstrained_params =  (Default)\n#     constrained_params = params.json\n#     jacobian = true (Default)\n# id = 1 (Default)\n# data\n#   file = bernoulli.data.json\n# init = 2 (Default)\n# random\n#   seed = 2390820139 (Default)\n# output\n#   file = output.csv (Default)\n#   diagnostic_file =  (Default)\n#   refresh = 100 (Default)\n#   sig_figs = 8 (Default)\n#   profile_file = profile.csv (Default)\n# num_threads = 1 (Default)\nlp_,theta\n-7.856,1.8",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Extracting log probabilities and gradients for diagnostics"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#beta-binomial-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#beta-binomial-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/optimize_config.html",
+    "href": "cmdstan-guide/optimize_config.html",
+    "title": "Optimization",
     "section": "",
-    "text": "If \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in\n\\mathbb{R}^+\\), then for \\(n \\in {0,\\ldots,N}\\), \\[\\begin{equation*}\n\\text{BetaBinomial}(n~|~N,\\alpha,\\beta) = \\binom{N}{n}\n\\frac{\\mathrm{B}(n+\\alpha, N -n +   \\beta)}{\\mathrm{B}(\\alpha,\\beta)},\n\\end{equation*}\\] where the beta function \\(\\mathrm{B}(u,v)\\) is defined for \\(u \\in\n\\mathbb{R}^+\\) and \\(v \\in \\mathbb{R}^+\\) by \\[\\begin{equation*} \\mathrm{B}(u,v) =\n\\frac{\\Gamma(u) \\ \\Gamma(v)}{\\Gamma(u + v)}. \\end{equation*}\\]\n\n\n\nn ~ beta_binomial(N, alpha, beta)\nIncrement target log probability density with beta_binomial_lupmf(n | N, alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_binomial_lpmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lupmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_binomial_cdf(ints n | ints N, reals alpha, reals beta) The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.0\n \n\nreal beta_binomial_lcdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lccdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nR beta_binomial_rng(ints N, reals alpha, reals beta) Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "The CmdStan executable can run Stan’s optimization algorithms, which provide a deterministic method to find the posterior mode. If the posterior is not convex, there is no guarantee Stan will be able to find the global optimum as opposed to a local optimum of log probability.\nThe full set of configuration options available for the optimize method is available by using the optimize help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the optimizer console output and in the output CSV file’s comments.\nThe executable does not need to be recompiled in order to switch from sampling to optimization, and the data input format is the same. The following is a minimal call to Stan’s optimizer using defaults for everything but the location of the data file.\n&gt; ./bernoulli optimize data file=bernoulli.data.json\nExecuting this command prints both output to the console and to a CSV file.\nThe first part of the console output reports on the configuration used. The above command uses all default configurations, therefore the optimizer used is the L-BFGS optimizer and its default initial stepsize and tolerances for monitoring convergence:\n ./bernoulli optimize data file=bernoulli.data.json\nmethod = optimize\n  optimize\n    algorithm = lbfgs (Default)\n      lbfgs\n        init_alpha = 0.001 (Default)\n        tol_obj = 1e-12 (Default)\n        tol_rel_obj = 10000 (Default)\n        tol_grad = 1e-08 (Default)\n        tol_rel_grad = 1e+07 (Default)\n        tol_param = 1e-08 (Default)\n        history_size = 5 (Default)\n    jacobian = false (Default)\n    iter = 2000 (Default)\n    save_iterations = false (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 87122538 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nThe second part of the output indicates how well the algorithm fared, here converging and terminating normally. The numbers reported indicate that it took 5 iterations and 8 gradient evaluations. This is, not surprisingly, far fewer iterations than required for sampling; even fewer iterations would be used with less stringent user-specified convergence tolerances. The alpha value is for step size used. In the final state the change in parameters was roughly \\(0.002\\) and the length of the gradient roughly 3e-05 (\\(0.00003\\)).\nInitial log joint probability = -6.85653\n    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes\n       5      -5.00402    0.00184936   3.35074e-05           1           1        8\nOptimization terminated normally:\n  Convergence detected: relative gradient magnitude is below tolerance\nThe output from optimization is written into the file output.csv by default. The output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as comment lines:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = optimize\n#   optimize\n#     algorithm = lbfgs (Default)\n#       lbfgs\n#         init_alpha = 0.001 (Default)\n#         tol_obj = 1e-12 (Default)\n#         tol_rel_obj = 10000 (Default)\n#         tol_grad = 1e-08 (Default)\n#         tol_rel_grad = 1e+07 (Default)\n#         tol_param = 1e-08 (Default)\n#         history_size = 5 (Default)\n#     jacobian = false (Default)\n#     iter = 2000 (Default)\n#     save_iterations = false (Default)\nFollowing the config information are two lines of output, the CSV headers and the recorded values:\nlp__,converged__,theta\n-5.00402,31,0.200003\nNote that everything is a comment other than a line for the header, and a line for the values. Here, the header indicates the unnormalized log probability with lp__, algorithm status in converged__, and the model parameter theta. The maximum log probability is -5.0 and the posterior mode for theta is 0.20. The mode exactly matches what we would expect from the data. Because the prior was uniform, the result 0.20 represents the maximum likelihood estimate (MLE) for the very simple Bernoulli model. Note that no uncertainty is reported.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value is false, i.e., intermediate iterations are not saved to the output file. Running the optimizer with save_iterations=true writes both the initial log joint probability and values for all iterations to the output CSV file.\nRunning the example model with option save_iterations=true, i.e., the command\n&gt; ./bernoulli optimize save_iterations=1 data file=bernoulli.data.json\nproduces CSV file output rows:\nlp__,converged__,theta\n-6.85653,0,0.493689\n-6.10128,0,0.420936\n-5.02953,0,0.22956\n-5.00517,0,0.206107\n-5.00403,0,0.200299\n-5.00402,31,0.200003\n\n\nThe converged__ column is used to indicate the status of the algorithm. It can take on the following values.\n\n\n\n\n\n\n\nNumeric value\nMeaning\n\n\n\n\n-1\nLine search failed to achieve a sufficient decrease, no more progress can be made\n\n\n0\nSuccessful step completed\n\n\n10\nConvergence detected: absolute parameter change was below tolerance\n\n\n20\nConvergence detected: absolute change in objective function was below tolerance\n\n\n21\nConvergence detected: relative change in objective function was below tolerance\n\n\n30\nConvergence detected: gradient norm is below tolerance\n\n\n31\nConvergence detected: relative gradient magnitude is below tolerance\n\n\n40\nMaximum number of iterations hit, may not be at an optima\n\n\n\nNote that not all algorithms can produce all codes, and some, such as 0, will only be observed if save_iterations is used.\nA comment explaining the final value (with the same contents as the above table) is added at the end of the CSV file.\n\n\n\nIf there are constrained parameters, Stan makes a transformation to an unconstrained space and runs the optimization algorithm in the unconstrained space.\nThe jacobian argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of inverse parameter transforms.\nIf the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.\nIf the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\n\n\n\nThe algorithm argument specifies the optimization algorithm. This argument takes one of the following three values:\n\nlbfgs A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.\nbfgs A quasi-Newton optimizer.\nnewton A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.\n\nSee the Stan Reference Manual’s Optimization chapter for a description of these algorithms.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value isfalse`, i.e., intermediate iterations are not saved to the output file.\n\n\n\nFor both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the BFGS and L-BFGS configuration section for details on the convergence tests.\nBoth BFGS and L-BFGS have the following configuration arguments:\n\ninit_alpha - The initial step size parameter. Must be a positive real number. Default value is \\(0.001\\)\ntol_obj - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is \\(1^{-12}\\).\ntol_rel_obj - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is \\(1^{4}\\).\ntol_grad - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is \\(1^{-8}\\).\ntol_rel_grad - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is \\(1^{7}\\).\ntol_param - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is \\(1^{-8}\\).\n\nThe init_alpha argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set init_alpha to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.\nIn addition to the above, the L-BFGS algorithm has argument history_size which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (\\(5\\)-\\(10\\)) are sufficient; the default value is \\(5\\).\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.\n\n\n\nThere are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Optimization"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#hypergeometric-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#hypergeometric-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/optimize_config.html#meaning-of-the-converged__-column",
+    "href": "cmdstan-guide/optimize_config.html#meaning-of-the-converged__-column",
+    "title": "Optimization",
     "section": "",
-    "text": "If \\(a \\in \\mathbb{N}\\), \\(b \\in \\mathbb{N}\\), and \\(N \\in\n\\{0,\\ldots,a+b\\}\\), then for \\(n \\in \\{\\max(0,N-b),\\ldots,\\min(a,N)\\}\\), \\[\\begin{equation*} \\text{Hypergeometric}(n~|~N,a,b) = \\frac{\\normalsize{\\binom{a}{n}\n\\binom{b}{N - n}}}      {\\normalsize{\\binom{a + b}{N}}}. \\end{equation*}\\]\n\n\n\nn ~ hypergeometric(N, a, b)\nIncrement target log probability density with hypergeometric_lupmf(n | N, a, b).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal hypergeometric_lpmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b\nAvailable since 2.12\n \n\nreal hypergeometric_lupmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms\nAvailable since 2.25\n \n\nint hypergeometric_rng(int N, int a, int b) Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "text": "The converged__ column is used to indicate the status of the algorithm. It can take on the following values.\n\n\n\n\n\n\n\nNumeric value\nMeaning\n\n\n\n\n-1\nLine search failed to achieve a sufficient decrease, no more progress can be made\n\n\n0\nSuccessful step completed\n\n\n10\nConvergence detected: absolute parameter change was below tolerance\n\n\n20\nConvergence detected: absolute change in objective function was below tolerance\n\n\n21\nConvergence detected: relative change in objective function was below tolerance\n\n\n30\nConvergence detected: gradient norm is below tolerance\n\n\n31\nConvergence detected: relative gradient magnitude is below tolerance\n\n\n40\nMaximum number of iterations hit, may not be at an optima\n\n\n\nNote that not all algorithms can produce all codes, and some, such as 0, will only be observed if save_iterations is used.\nA comment explaining the final value (with the same contents as the above table) is added at the end of the CSV file.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Optimization"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#categorical-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#categorical-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/optimize_config.html#jacobian-adjustments",
+    "href": "cmdstan-guide/optimize_config.html#jacobian-adjustments",
+    "title": "Optimization",
     "section": "",
-    "text": "If \\(N \\in \\mathbb{N}\\), \\(N &gt; 0\\), and if \\(\\theta \\in \\mathbb{R}^N\\) forms an \\(N\\)-simplex (i.e., has nonnegative entries summing to one), then for \\(y \\in \\{1,\\ldots,N\\}\\), \\[\\begin{equation*} \\text{Categorical}(y~|~\\theta) =\n\\theta_y. \\end{equation*}\\] In addition, Stan provides a log-odds scaled categorical distribution, \\[\\begin{equation*} \\text{CategoricalLogit}(y~|~\\beta) =\n\\text{Categorical}(y~|~\\text{softmax}(\\beta)). \\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical(theta)\nIncrement target log probability density with categorical_lupmf(y | theta) dropping constant additive terms.\nAvailable since 2.0\n \n\n\n\n\ny ~ categorical_logit(beta)\nIncrement target log probability density with categorical_logit_lupmf(y | beta).\nAvailable since 2.4\n \n\n\n\n\nAll of the categorical distributions are vectorized so that the outcome y can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal categorical_lpmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.12\n \n\nreal categorical_lupmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.25\n \n\nreal categorical_logit_lpmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta.\nAvailable since 2.12\n \n\nreal categorical_logit_lupmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta dropping constant additive terms.\nAvailable since 2.25\n \n\nint categorical_rng(vector theta) Generate a categorical variate with \\(N\\)-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nint categorical_logit_rng(vector beta) Generate a categorical variate with outcome in range \\(1:N\\) from log-odds vector beta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.16",
+    "text": "If there are constrained parameters, Stan makes a transformation to an unconstrained space and runs the optimization algorithm in the unconstrained space.\nThe jacobian argument specifies whether or not the call to the model’s log probability function should include the log absolute Jacobian determinant of inverse parameter transforms.\nIf the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space.\nIf the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Optimization"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#categorical-logit-glm",
-    "href": "functions-reference/bounded_discrete_distributions.html#categorical-logit-glm",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/optimize_config.html#optimization-algorithms",
+    "href": "cmdstan-guide/optimize_config.html#optimization-algorithms",
+    "title": "Optimization",
     "section": "",
-    "text": "Stan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e. a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.\nNote that the implementation does not put any restrictions on the coefficient matrix \\(\\beta\\). It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the Stan User’s Guide.\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\), \\(N,M,K &gt; 0\\), and if \\(x\\in \\mathbb{R}^{M\\times K}, \\alpha \\in \\mathbb{R}^N, \\beta\\in \\mathbb{R}^{K\\cdot N}\\), then for \\(y \\in \\{1,\\ldots,N\\}^M\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{CategoricalLogitGLM}(y~|~x,\\alpha,\\beta)\n& =  \\prod_{1\\leq i \\leq M}\\text{CategoricalLogit}(y_i~|~\\alpha+x_i\\cdot\\beta)  \\\\[8pt]\n& =  \\prod_{1\\leq i \\leq M}\\text{Categorical}(y_i~|~softmax(\\alpha+x_i\\cdot\\beta)).\n\\end{split}\n\\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical_logit_glm(x, alpha, beta)\nIncrement target log probability density with categorical_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal categorical_logit_glm_lpmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25",
+    "text": "The algorithm argument specifies the optimization algorithm. This argument takes one of the following three values:\n\nlbfgs A quasi-Newton optimizer. This is the default optimizer and also much faster than the other optimizers.\nbfgs A quasi-Newton optimizer.\nnewton A Newton optimizer. This is the least efficient optimization algorithm, but has the advantage of setting its own stepsize.\n\nSee the Stan Reference Manual’s Optimization chapter for a description of these algorithms.\nAll of the optimizers stream per-iteration intermediate approximations to the command line console. The sub-argument save_iterations specifies whether or not to save the intermediate iterations to the output file. Allowed values are true or false. The default value isfalse`, i.e., intermediate iterations are not saved to the output file.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Optimization"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#discrete-range-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#discrete-range-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers",
+    "href": "cmdstan-guide/optimize_config.html#the-quasi-newton-optimizers",
+    "title": "Optimization",
     "section": "",
-    "text": "If \\(l, u \\in \\mathbb{Z}\\) are lower and upper bounds (\\(l \\le u\\)), then for any integer \\(y \\in \\{l,\\ldots,u\\}\\), \\[\\begin{equation*} \\text{DiscreteRange}(y ~|~ l, u) =\n\\frac{1}{u - l + 1}. \\end{equation*}\\]\n\n\n\ny ~ discrete_range(l, u)\nIncrement the target log probability density with discrete_range_lupmf(y | l, u) dropping constant additive terms.\nAvailable since 2.26\n \n\n\n\n\nAll of the discrete range distributions are vectorized so that the outcome y and the bounds l, u can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal discrete_range_lpmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\).\nAvailable since 2.26\n \n\nreal discrete_range_lupmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\) dropping constant additive terms.\nAvailable since 2.26\n \n\nreal discrete_range_cdf(ints y | ints l, ints u) The discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lcdf(ints y | ints l, ints u) The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lccdf(ints y | ints l, ints u) The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nints discrete_range_rng(ints l, ints u) Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.26",
+    "text": "For both BFGS and L-BFGS optimizers, convergence monitoring is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. See the BFGS and L-BFGS configuration section for details on the convergence tests.\nBoth BFGS and L-BFGS have the following configuration arguments:\n\ninit_alpha - The initial step size parameter. Must be a positive real number. Default value is \\(0.001\\)\ntol_obj - Convergence tolerance on changes in objective function value. Must be a positive real number. Default value is \\(1^{-12}\\).\ntol_rel_obj - Convergence tolerance on relative changes in objective function value. Must be a positive real number. Default value is \\(1^{4}\\).\ntol_grad - Convergence tolerance on the norm of the gradient. Must be a positive real number. Default value is \\(1^{-8}\\).\ntol_rel_grad - Convergence tolerance on the relative norm of the gradient. Must be a positive real number. Default value is \\(1^{7}\\).\ntol_param - Convergence tolerance on changes in parameter value. Must be a positive real number. Default value is \\(1^{-8}\\).\n\nThe init_alpha argument specifies the first step size to try on the initial iteration. If the first iteration takes a long time (and requires a lot of function evaluations), set init_alpha to be the roughly equal to the alpha used in that first iteration. The default value is very small, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length.\nIn addition to the above, the L-BFGS algorithm has argument history_size which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (\\(5\\)-\\(10\\)) are sufficient; the default value is \\(5\\).\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Optimization"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/optimize_config.html#the-newton-optimizer",
+    "href": "cmdstan-guide/optimize_config.html#the-newton-optimizer",
+    "title": "Optimization",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedLogistic}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\text{logit}^{-1}(\\eta - c_1)  &  \\text{if } k = 1, \\\\[4pt]\n\\text{logit}^{-1}(\\eta - c_{k-1}) - \\text{logit}^{-1}(\\eta - c_{k})  &\n\\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt] \\text{logit}^{-1}(\\eta -\nc_{K-1}) - 0  &  \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\nk ~ ordered_logistic(eta, c)\nIncrement target log probability density with ordered_logistic_lupmf(k | eta, c).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal ordered_logistic_lpmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_logistic_lupmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nint ordered_logistic_rng(real eta, vector c) Generate an ordered logistic variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "text": "There are no configuration parameters for the Newton optimizer. It is not recommended because of the slow Hessian calculation involving finite differences.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Optimization"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-generalized-linear-model-ordinal-regression",
-    "href": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-generalized-linear-model-ordinal-regression",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/pathfinder_config.html",
+    "href": "cmdstan-guide/pathfinder_config.html",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
     "section": "",
-    "text": "If \\(N,M,K \\in \\mathbb{N}\\) with \\(N, M &gt; 0\\), \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(x\\in \\mathbb{R}^{N\\times M}, \\beta\\in \\mathbb{R}^M\\), then for \\(y \\in \\{1,\\ldots,K\\}^N\\), \\[\\begin{equation*}\n\\begin{split}\n\\\\\n& \\text{OrderedLogisticGLM}(y~|~x,\\beta,c) \\\\[8pt]\n& = \\prod_{1\\leq i \\leq N}\\text{OrderedLogistic}(y_i~|~x_i\\cdot \\beta,c) \\\\\n& = \\prod_{1\\leq i \\leq N}  \\left\\{ \\begin{array}{ll}\n1 - \\text{logit}^{-1}(x_i\\cdot \\beta - c_1)  &  \\text{if } y = 1, \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{y-1}) - \\text{logit}^{-1}(x_i\\cdot \\beta - c_{y}) & \\text{if } 1 &lt; y &lt; K, \\text{and} \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{K-1}) - 0  &  \\text{if } y = K.\n\\end{array} \\right.\n\\end{split}\n\\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(y=1\\) and \\(y=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\ny ~ ordered_logistic_glm(x, beta, c)\nIncrement target log probability density with ordered_logistic_lupmf(y | x, beta, c).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal ordered_logistic_glm_lpmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25",
+    "text": "The CmdStan method pathfinder uses the Pathfinder algorithm of Zhang et al. (2022), which is further described in the Stan Reference Manual.\nA single run of the Pathfinder algorithm generates a set of approximate draws. Inference is improved by running multiple Pathfinder instances and using Pareto-smoothed importance resampling (PSIS) of the resulting sets of draws. This better matches non-normal target densities and also eliminates minor modes.\nThe pathfinder method runs multi-path Pathfinder by default, which returns a PSIS sample over the draws from several individual (“single-path”) Pathfinder runs. Argument num_paths specifies the number of single-path Pathfinders, the default is \\(4\\). If num_paths is set to 1, then only one individual Pathfinder is run without the PSIS reweighting of the sample.\nThe full set of configuration options available for the pathfinder method is available by using the pathfinder help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.\nThe following is a minimal call to the Pathfinder algorithm using defaults for everything but the location of the data file.\n&gt; ./bernoulli pathfinder data file=bernoulli.data.R\nExecuting this command prints output both to the console and to csv files.\nThe first part of the console output reports on the configuration used.\nmethod = pathfinder\n  pathfinder\n    init_alpha = 0.001 (Default)\n    tol_obj = 1e-12 (Default)\n    tol_rel_obj = 10000 (Default)\n    tol_grad = 1e-08 (Default)\n    tol_rel_grad = 1e+07 (Default)\n    tol_param = 1e-08 (Default)\n    history_size = 5 (Default)\n    num_psis_draws = 1000 (Default)\n    num_paths = 4 (Default)\n    save_single_paths = false (Default)\n    psis_resample = true (Default)\n    calculate_lp = true (Default)\n    max_lbfgs_iters = 1000 (Default)\n    num_draws = 1000 (Default)\n    num_elbo_draws = 25 (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2790476610 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nThe rest of the output describes the progression of the algorithm.\nBy default, the Pathfinder algorithm runs 4 single-path Pathfinders in parallel, then uses importance resampling on the set of returned draws to produce the specified number of draws.\nPath [1] :Initial log joint density = -11.543343\nPath [1] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      1.070e-03   1.707e-05    1.000e+00  1.000e+00       126 -6.220e+00 -6.220e+00\nPath [1] :Best Iter: [5] ELBO (-6.219833) evaluations: (126)\nPath [2] :Initial log joint density = -7.443345\nPath [2] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      9.936e-05   3.738e-07    1.000e+00  1.000e+00       126 -6.164e+00 -6.164e+00\nPath [2] :Best Iter: [5] ELBO (-6.164015) evaluations: (126)\nPath [3] :Initial log joint density = -18.986308\nPath [3] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      2.996e-04   4.018e-06    1.000e+00  1.000e+00       126 -6.201e+00 -6.201e+00\nPath [3] :Best Iter: [5] ELBO (-6.200559) evaluations: (126)\nPath [4] :Initial log joint density = -8.304453\nPath [4] : Iter      log prob        ||dx||      ||grad||     alpha      alpha0      # evals       ELBO    Best ELBO        Notes\n              5      -6.748e+00      2.814e-04   2.034e-06    1.000e+00  1.000e+00       126 -6.221e+00 -6.221e+00\nPath [4] :Best Iter: [3] ELBO (-6.161276) evaluations: (126)\nTotal log probability function evaluations:8404\n\n\n\nnum_psis_draws - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is \\(1000\\).\nnum_paths - Number of single pathfinders. Must be a positive integer. Default value is \\(4\\).\nsave_single_paths - When true, save outputs from single pathfinders. Valid values: [true, false]. Default is false.\nmax_lbfgs_iters - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is \\(1000\\).\nnum_draws - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is \\(1000\\). Can differ from num_psis_draws.\nnum_elbo_draws - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is \\(25\\).\npsis_resample - If true, perform PSIS resampling on draws returned from individual pathfinders. If false, returns all num_paths * num_draws draws from the individual pathfinders. Valid values: [true, false]. Default is true.\ncalculate_lp - If true, log probabilities of the approximate draws are calculated and returned with the output. If false, each pathfinder will only calculate the lp values needed for the ELBO calculation. If false, PSIS resampling cannot be performed and the algorithm returns num_paths * num_draws samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: [true, false]. Default is true.\n\n\n\n\nArguments init_alpha through history_size are the full set of arguments to the L-BFGS optimizer and have the same defaults for optimization.\n\n\n\nBy default, the pathfinder method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a StanCSV file.\nThe CSV files have the following structure:\nThe initial CSV comment rows contain the complete set of CmdStan configuration options.\n...\n# method = pathfinder\n#   pathfinder\n#     init_alpha = 0.001 (Default)\n#     tol_obj = 9.9999999999999998e-13 (Default)\n#     tol_rel_obj = 10000 (Default)\n#     tol_grad = 1e-08 (Default)\n#     tol_rel_grad = 10000000 (Default)\n#     tol_param = 1e-08 (Default)\n#     history_size = 5 (Default)\n#     num_psis_draws = 1000 (Default)\n#     num_paths = 4 (Default)\n#     psis_resample = 1 (Default)\n#     calculate_lp = 1 (Default)\n#     save_single_paths = 0 (Default)\n#     max_lbfgs_iters = 1000 (Default)\n#     num_draws = 1000 (Default)\n#     num_elbo_draws = 25 (Default)\n...\nNext is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs lp_approx__, the log density in the approximating distribution, and lp__, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.\nlp_approx__,lp__,theta\n-2.4973, -8.2951, 0.0811852\n-0.87445, -7.06526, 0.160207\n-0.812285, -7.07124, 0.35819\n...\nThe final lines are comment lines which give timing information.\n# Elapsed Time: 0.016000 seconds (Pathfinders)\n#               0.003000 seconds (PSIS)\n#               0.019000 seconds (Total)\nPathfinder provides option save_single_paths which will save output from the single-path Pathfinder runs.\n\n\n\nThe boolean option save_single_paths is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When save_single_paths is true, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name output.csv and default number of runs (4), the resulting CSV files are\noutput.csv\noutput_path_1.csv\noutput_path_1.json\noutput_path_2.csv\noutput_path_2.json\noutput_path_3.csv\noutput_path_3.json\noutput_path_4.csv\noutput_path_4.json\nThe individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.\nTo see how this works, we run Pathfinder on the centered-parameterization of the eight-schools model, where the posterior distribution has a funnel shape:\n&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json\nEach JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.\n{\n  \"0\": {\n    \"iter\": 0,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]\n  },\n  \"1\": {\n    \"iter\": 1,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],\n    \"history_size\": 1,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],\n    \"logDetCholHk\": -2.6839,\n    \"L_approx\": [[-0.0630456, -0.0187959], [0, 1.08328]],\n    \"Qk\": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],\n    \"alpha\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n    \"full\": false,\n    \"lbfgs_note\": \"\"\n  },\n  ...,\n  \"171\": {\n    \"iter\": 171,\n    \"unconstrained_parameters\": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],\n    \"grads\": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],\n    \"history_size\": 5,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],\n    \"logDetCholHk\": 299.023,\n    \"L_approx\": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],\n    \"Qk\": [],\n    \"alpha\": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],\n    \"full\": true,\n    \"lbfgs_note\": \"\"\n  },\n  \"172\": {\n    \"iter\": 172,\n    \"unconstrained_parameters\": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],\n    \"grads\": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],\n    \"history_size\": 5,\n    \"lbfgs_success\": false,\n    \"pathfinder_success\": false,\n    \"lbfgs_note\": \"\"\n  }\n}\nOption num_paths=1 runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments num_paths=1 save_single_paths=true creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
     ]
   },
   {
-    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-probit-distribution",
-    "href": "functions-reference/bounded_discrete_distributions.html#ordered-probit-distribution",
-    "title": "Bounded Discrete Distributions",
+    "objectID": "cmdstan-guide/pathfinder_config.html#pathfinder-configuration",
+    "href": "cmdstan-guide/pathfinder_config.html#pathfinder-configuration",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedProbit}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\Phi(\\eta - c_1) & \\text{if } k = 1, \\\\[4pt] \\Phi(\\eta - c_{k-1}) -\n\\Phi(\\eta - c_{k})  & \\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt]\n\\Phi(\\eta - c_{K-1}) - 0 & \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 =\n-\\infty\\) and \\(c_K = +\\infty\\) with \\(\\Phi(-\\infty) = 0\\) and \\(\\Phi(\\infty) = 1\\).\n\n\n\nk ~ ordered_probit(eta, c)\nIncrement target log probability density with ordered_probit_lupmf(k | eta, c).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal ordered_probit_lpmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_probit_lupmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nreal ordered_probit_lpmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.\nAvailable since 2.19\n \n\nreal ordered_probit_lupmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.19\n \n\nint ordered_probit_rng(real eta, vector c) Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "text": "num_psis_draws - Final number of draws from multi-path pathfinder. Must be a positive integer. Default value is \\(1000\\).\nnum_paths - Number of single pathfinders. Must be a positive integer. Default value is \\(4\\).\nsave_single_paths - When true, save outputs from single pathfinders. Valid values: [true, false]. Default is false.\nmax_lbfgs_iters - Maximum number of L-BFGS iterations. Must be a positive integer. Default value is \\(1000\\).\nnum_draws - Number of approximate posterior draws for each single pathfinder. Must be a positive integer. Default value is \\(1000\\). Can differ from num_psis_draws.\nnum_elbo_draws - Number of Monte Carlo draws to evaluate ELBO. Must be a positive integer. Default value is \\(25\\).\npsis_resample - If true, perform PSIS resampling on draws returned from individual pathfinders. If false, returns all num_paths * num_draws draws from the individual pathfinders. Valid values: [true, false]. Default is true.\ncalculate_lp - If true, log probabilities of the approximate draws are calculated and returned with the output. If false, each pathfinder will only calculate the lp values needed for the ELBO calculation. If false, PSIS resampling cannot be performed and the algorithm returns num_paths * num_draws samples. The output will still contain any lp values used when calculating ELBO scores within L-BFGS iterations. Valid values: [true, false]. Default is true.",
     "crumbs": [
-      "Functions Reference",
-      "Discrete Distributions",
-      "Bounded Discrete Distributions"
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html",
-    "href": "functions-reference/complex-valued_basic_functions.html",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/pathfinder_config.html#l-bfgs-configuration",
+    "href": "cmdstan-guide/pathfinder_config.html#l-bfgs-configuration",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
     "section": "",
-    "text": "This chapter describes built-in functions that operate on complex numbers, either as an argument type or a return type. This includes the arithmetic operators generalized to complex numbers.\n\n\nJust as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.\nint n = 5;       // n = 5\nreal x = a;      // x = 5.0\ncomplex z1 = n;  // z = 5.0 + 0.0i\ncomplex z2 = x;  // z = 5.0 + 0.0i\n\n\nFunction arguments of type int or real may be promoted to type complex. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if z is complex, then pow(z, 2) will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where z + 2 will be legal and produce a complex result. Functions such as arg and conj that are only available for complex numbers can accept integer or real arguments, promoting them to complex before applying the function.\n\n\n\n\n\n\nVariables and constants of type complex are constructed from zero, one, or two real numbers.\ncomplex z1 = to_complex();        // z1 = 0.0 + 0.0i\nreal re = -2.9;\ncomplex z2 = to_complex(re);      // z2 = -2.9 + 0.0i\nreal im = 1.3;\ncomplex z3 = to_complex(re, im);  // z3 = -2.9 + 1.3i\n \n\ncomplex to_complex() Return complex number with real part 0.0 and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re) Return complex number with real part re and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re, real im) Return complex number with real part re and imaginary part im.\nAvailable since 2.28\n \n\nZ to_complex(T1 re, T2 im) Vectorized implementation of the to_complex function.\nT1 and T2 can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.\nAvailable since 2.30\n\n\n\nGiven a complex number, its real and imaginary parts can be extracted with the following functions.\n \n\nreal get_real(complex z) Return the real part of the complex number z.\nAvailable since 2.28\n \n\nreal get_imag(complex z) Return the imaginary part of the complex number z.\nAvailable since 2.28\n\n\n\n\nThe arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type complex. If there are two arguments and only one is of type complex, then the other will be promoted to type complex before performing the operation.\n\n\n \n\ncomplex operator+(complex z) Return the complex argument z, \\[\\begin{equation*} +z = z. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex z) Return the negation of the complex argument z, which for \\(z = x + yi\\) is \\[\\begin{equation*} -z = -x - yi. \\end{equation*}\\]\nAvailable since 2.28\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of complex numbers, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex operator+(complex x, complex y) Return the sum of x and y, \\[\\begin{equation*} (x + y) = \\text{operator+}(x, y) = x + y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex x, complex y) Return the difference between x and y, \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x, y) = x - y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator*(complex x, complex y) Return the product of x and y, \\[\\begin{equation*} (x \\, * \\, y) = \\text{operator*}(x, y) = x\n\\times y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator/(complex x, complex y) Return the quotient of x and y, \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator^(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n(x^\\mathrm{\\wedge}y)= \\text{operator}^\\mathrm{\\wedge}(x,y)\n= \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\n\nComplex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional\nz1 == z2\nis equivalent to\nget_real(z1) == get_real(z2) && get_imag(z1) == get_imag(z2)\nAs with other complex functions, if one of the arguments is of type real or int, it will be promoted to type complex before comparison. For example, if z is of type complex, then z == 0 will be true if z has real component equal to 0.0 and complex component equal to 0.0.\nWarning: As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.\n \n\nint operator==(complex x, complex y) Return 1 if x is equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{==}\\, y)\n\\ = \\ \\text{operator==}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x = y$}, \\ \\text{and} \\\\ 0 & \\text{otherwise.}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nint operator!=(complex x, complex y) Return 1 if x is not equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{!=}\\, y)\n\\ = \\ \\text{operator!=}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x \\neq y$}, \\ \\text{and} \\\\ 0 &\n\\text{otherwise.} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\nThe assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.\nAssignment of complex numbers works elementwise. If an expression of type int or real is assigned to a complex number, it will be promoted before assignment as if calling to_complex(), so that the imaginary component is 0.0.\n \n\nvoid operator=(complex x, complex y) y = x; assigns a (copy of) the value of y to x.\nAvailable since 2.28\n \n\nvoid operator+=(complex x, complex y) x += y; is equivalent to x = x + y;.\nAvailable since 2.28\n \n\nvoid operator-=(complex x, complex y) x -= y; is equivalent to x = x - y;.\nAvailable since 2.28\n \n\nvoid operator*=(complex x, complex y) x *= y; is equivalent to x = x * y;.\nAvailable since 2.28\n \n\nvoid operator/=(complex x, complex y) x /= y; is equivalent to x = x / y;.\nAvailable since 2.28\n\n\n\nThe following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.\n \n\nreal abs(complex z) Return the absolute value of z, also known as the modulus or magnitude, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{abs}(z) = \\sqrt{x^2 + y^2}.\n\\end{equation*}\\]\nThis function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a complex_vector[n] input will return a vector[n] output, with each element transformed by the above equation.\nAvailable since 2.28, vectorized in 2.30\n \n\nreal arg(complex z) Return the phase angle (in radians) of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{arg}(z) = \\textrm{atan2}(y, x) = \\textrm{atan}(y / x).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nreal norm(complex z) Return the Euclidean norm of z, which is its absolute value squared, and which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{norm}(z) = \\textrm{abs}^2(z) = x^2 + y^2.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex conj(complex z) Return the complex conjugate of z, which negates the imaginary component, so that if \\(z = x + yi\\), \\[\\begin{equation*}\n\\textrm{conj}(z) = x - yi.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ conj(Z z) Vectorized version of conj. This will apply the conj function to each element of a complex array, vector, or matrix.\nAvailable since 2.31\n \n\ncomplex proj(complex z) Return the projection of z onto the Riemann sphere, which for \\(z = x\n+ yi\\) is \\[\\begin{equation*}\n\\textrm{proj}(z)\n= \\begin{cases}\n    z & \\textrm{if} \\ z \\ \\textrm{is finite, and} \\\\\n    0 + \\textrm{sign}(y)i & \\textrm{otherwise,}\n\\end{cases}\n\\end{equation*}\\] where \\(\\textrm{sign}(y)\\) is -1 if \\(y\\) is negative and 1 otherwise.\nAvailable since 2.28\n \n\ncomplex polar(real r, real theta) Return the complex number with magnitude (absolute value) r and phase angle theta.\nAvailable since 2.28\n\n\n\nThe exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.\n \n\ncomplex exp(complex z) Return the complex natural exponential of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\exp z = \\exp(x) \\textrm{cis}(y) = \\exp(x) (\\cos(y) + i \\sin(y)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log(complex z) Return the complex natural logarithm of z, which for \\(z = \\textrm{polar}(r,\n\\theta)\\) is \\[\\begin{equation*}\n\\log z = \\log r + \\theta i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log10(complex z) Return the complex common logarithm of z, \\[\\begin{equation*}\n\\log_{10} z = \\frac{\\log z}{\\log 10}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex pow(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n\\text{pow}(x,y) = \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.30\n \n\ncomplex sqrt(complex x) Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.\nAvailable since 2.28\n\n\n\nThe standard trigonometric functions are supported for complex numbers.\n \n\ncomplex cos(complex z) Return the complex cosine of z, which is \\[\\begin{equation*}\n\\cos(z)\n= \\textrm{cosh}(z \\, i)\n= \\frac{\\displaystyle \\exp(z \\, i) + \\exp(-z \\, i)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sin(complex z) Return the complex sine of z, \\[\\begin{equation*}\n\\sin(z)\n= -\\textrm{sinh}(z \\, i) \\, i\n= \\frac{\\displaystyle \\exp(z \\, i) - \\exp(-z \\, i)}\n       {\\displaystyle 2 \\, i}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tan(complex z) Return the complex tangent of z, \\[\\begin{equation*}\n\\tan(z)\n= -\\textrm{tanh}(z \\, i) \\, i\n= \\frac{(\\exp(-z \\, i) - \\exp(z \\, i)) \\, i}\n       {\\exp(-z \\, i) + \\exp(z \\, i)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acos(complex z) Return the complex arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acos}(z)\n= \\frac{1}{2} \\pi + \\log (z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asin(complex z) Return the complex arc (inverse) sine of z, \\[\\begin{equation*}\n\\text{asin}(z)\n= -\\log(z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atan(complex z) Return the complex arc (inverse) tangent of z, \\[\\begin{equation*}\n\\text{atan}(z)\n= - \\frac{1}{2} (\\log(1 - z \\, i) - \\log(1 + z \\, i)) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\nThe standard hyperbolic trigonometric functions are supported for complex numbers.\n \n\ncomplex cosh(complex z) Return the complex hyperbolic cosine of z, \\[\\begin{equation*}\n\\textrm{cosh}(z)\n= \\frac{\\exp(z) + \\exp(-z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sinh(complex z) Return the complex hyperbolic sine of z, \\[\\begin{equation*}\n\\textrm{sinh}(z)\n= \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tanh(complex z) Return the complex hyperbolic tangent of z, \\[\\begin{equation*}\n\\textrm{tanh}(z)\n\\ = \\ \\frac{\\textrm{sinh}(z)}\n           {\\textrm{cosh}(z)}\n\\ = \\ \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n           {\\displaystyle \\exp(z) + \\exp(-z)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acosh(complex z) Return the complex hyperbolic arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acosh}(z)\n= \\log(z + \\sqrt{(z + 1)(z - 1)}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asinh(complex z) Return the complex hyperbolic arc (inverse) sine of z, \\[\\begin{equation*}\n\\textrm{asinh}(z)\n= \\log(z + \\sqrt{1 + z^2}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atanh(complex z) Return the complex hyperbolic arc (inverse) tangent of z, \\[\\begin{equation*}\n\\textrm{atanh}(z)\n= \\frac{\\log(1 + z) - \\log(1 - z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28",
+    "text": "Arguments init_alpha through history_size are the full set of arguments to the L-BFGS optimizer and have the same defaults for optimization.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-assign-promote",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-assign-promote",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/pathfinder_config.html#pathfinder_csv",
+    "href": "cmdstan-guide/pathfinder_config.html#pathfinder_csv",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
     "section": "",
-    "text": "Just as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.\nint n = 5;       // n = 5\nreal x = a;      // x = 5.0\ncomplex z1 = n;  // z = 5.0 + 0.0i\ncomplex z2 = x;  // z = 5.0 + 0.0i\n\n\nFunction arguments of type int or real may be promoted to type complex. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if z is complex, then pow(z, 2) will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where z + 2 will be legal and produce a complex result. Functions such as arg and conj that are only available for complex numbers can accept integer or real arguments, promoting them to complex before applying the function.",
+    "text": "By default, the pathfinder method uses 4 independent Pathfinder runs, each of which produces 1000 approximate draws, which are then importance resampled down to 1000 final draws. The importance resampled draws are output as a StanCSV file.\nThe CSV files have the following structure:\nThe initial CSV comment rows contain the complete set of CmdStan configuration options.\n...\n# method = pathfinder\n#   pathfinder\n#     init_alpha = 0.001 (Default)\n#     tol_obj = 9.9999999999999998e-13 (Default)\n#     tol_rel_obj = 10000 (Default)\n#     tol_grad = 1e-08 (Default)\n#     tol_rel_grad = 10000000 (Default)\n#     tol_param = 1e-08 (Default)\n#     history_size = 5 (Default)\n#     num_psis_draws = 1000 (Default)\n#     num_paths = 4 (Default)\n#     psis_resample = 1 (Default)\n#     calculate_lp = 1 (Default)\n#     save_single_paths = 0 (Default)\n#     max_lbfgs_iters = 1000 (Default)\n#     num_draws = 1000 (Default)\n#     num_elbo_draws = 25 (Default)\n...\nNext is the column header line, followed the set of approximate draws. The Pathfinder algorithm first outputs lp_approx__, the log density in the approximating distribution, and lp__, the log density in the target distribution, followed by estimates of the model parameters, transformed parameters, and generated quantities.\nlp_approx__,lp__,theta\n-2.4973, -8.2951, 0.0811852\n-0.87445, -7.06526, 0.160207\n-0.812285, -7.07124, 0.35819\n...\nThe final lines are comment lines which give timing information.\n# Elapsed Time: 0.016000 seconds (Pathfinders)\n#               0.003000 seconds (PSIS)\n#               0.019000 seconds (Total)\nPathfinder provides option save_single_paths which will save output from the single-path Pathfinder runs.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-constructors",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-constructors",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/pathfinder_config.html#single-path-pathfinder-outputs",
+    "href": "cmdstan-guide/pathfinder_config.html#single-path-pathfinder-outputs",
+    "title": "Pathfinder Method for Approximate Bayesian Inference",
     "section": "",
-    "text": "Variables and constants of type complex are constructed from zero, one, or two real numbers.\ncomplex z1 = to_complex();        // z1 = 0.0 + 0.0i\nreal re = -2.9;\ncomplex z2 = to_complex(re);      // z2 = -2.9 + 0.0i\nreal im = 1.3;\ncomplex z3 = to_complex(re, im);  // z3 = -2.9 + 1.3i\n \n\ncomplex to_complex() Return complex number with real part 0.0 and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re) Return complex number with real part re and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re, real im) Return complex number with real part re and imaginary part im.\nAvailable since 2.28\n \n\nZ to_complex(T1 re, T2 im) Vectorized implementation of the to_complex function.\nT1 and T2 can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.\nAvailable since 2.30\n\n\n\nGiven a complex number, its real and imaginary parts can be extracted with the following functions.\n \n\nreal get_real(complex z) Return the real part of the complex number z.\nAvailable since 2.28\n \n\nreal get_imag(complex z) Return the imaginary part of the complex number z.\nAvailable since 2.28",
+    "text": "The boolean option save_single_paths is used to save both the draws and the ELBO iterations from the individual Pathfinder runs. When save_single_paths is true, the draws from each are saved to StanCSV files with the same format as the PSIS sample and the ELBO evaluations along the L-BFGS trajectory for each are saved as JSON. Given an output file name, CmdStan adds suffixes to the base filename to distinguish between the output files. For the default output file name output.csv and default number of runs (4), the resulting CSV files are\noutput.csv\noutput_path_1.csv\noutput_path_1.json\noutput_path_2.csv\noutput_path_2.json\noutput_path_3.csv\noutput_path_3.json\noutput_path_4.csv\noutput_path_4.json\nThe individual sample CSV files have the same structure as the PSIS sample CSV file. The JSON files contain information from each ELBO iteration.\nTo see how this works, we run Pathfinder on the centered-parameterization of the eight-schools model, where the posterior distribution has a funnel shape:\n&gt; ./eight_schools pathfinder save_single_paths=true data file=eight_schools.data.json\nEach JSON file records the approximations to the target density at each point along the trajectory of the L-BFGS optimization algorithms.\n{\n  \"0\": {\n    \"iter\": 0,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445]\n  },\n  \"1\": {\n    \"iter\": 1,\n    \"unconstrained_parameters\": [1.00595, -0.503687, 1.79367, 0.99083, 0.498077, -0.65816, 1.49176, -1.22647, 1.62911, 0.767445],\n    \"grads\": [-0.868919, 0.45198, -0.107675, -0.0123304, 0.163172, 0.354362, -0.108746, 0.673306, -0.102268, -4.51445],\n    \"history_size\": 1,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [0.126047, -0.065048, 1.55708, 0.958509, 0.628075, -0.217041, 1.32032, -0.561338, 1.42988, 1.23213],\n    \"logDetCholHk\": -2.6839,\n    \"L_approx\": [[-0.0630456, -0.0187959], [0, 1.08328]],\n    \"Qk\": [[-0.361073, 0.5624], [0.183922, -0.279474], [-0.0708175, 0.15715], [-0.00917823, 0.0215802], [0.0606019, -0.0814513], [0.164071, -0.285769], [-0.057723, 0.112428], [0.276376, -0.424348], [-0.0620524, 0.131786], [-0.846488, -0.531094]],\n    \"alpha\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n    \"full\": false,\n    \"lbfgs_note\": \"\"\n  },\n  ...,\n  \"171\": {\n    \"iter\": 171,\n    \"unconstrained_parameters\": [1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, 1.60479, -35.7821],\n    \"grads\": [2.66927e+15, -0.117312, -0.0639521, -2.66927e+15, -0.0445885, 0.0321579, 0.00499827, -0.163952, -0.032084, 6.4073],\n    \"history_size\": 5,\n    \"lbfgs_success\": true,\n    \"pathfinder_success\": true,\n    \"x_center\": [5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, 5.58876e+15, -2.02979e+17],\n    \"logDetCholHk\": 299.023,\n    \"L_approx\": [[4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, 4.6852e+06, -1.70162e+08], [0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 2.19511e+13, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 2.19511e+13, -7.97244e+14], [0, 0, 0, 0, 0, 0, 0, 0, 0, 2.89552e+16]],\n    \"Qk\": [],\n    \"alpha\": [1.11027e-12, 2.24669e-12, 2.05603e-12, 3.71177e-12, 5.7855e-12, 1.80169e-12, 3.40291e-12, 2.29699e-12, 3.43423e-12, 1.25815e-08],\n    \"full\": true,\n    \"lbfgs_note\": \"\"\n  },\n  \"172\": {\n    \"iter\": 172,\n    \"unconstrained_parameters\": [1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, 1.60531, -35.801],\n    \"grads\": [-0, -0.11731, -0.0639469, 0.0179895, -0.0445842, 0.0321643, 0.00500256, -0.163947, -0.0320824, 7],\n    \"history_size\": 5,\n    \"lbfgs_success\": false,\n    \"pathfinder_success\": false,\n    \"lbfgs_note\": \"\"\n  }\n}\nOption num_paths=1 runs one single-path Pathfinder and the output CSV file contains the draws from that run without PSIS reweighting. The combination of arguments num_paths=1 save_single_paths=true creates just two output files, the CSV sample and the set of ELBO iterations. In this case, the default output file name is “output.csv” and the default diagnostic file name is “output.json”.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Running CmdStan",
+      "Pathfinder Method for Approximate Bayesian Inference"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-arithmetic",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-arithmetic",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html",
+    "href": "cmdstan-guide/rdump_apdx.html",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "The arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type complex. If there are two arguments and only one is of type complex, then the other will be promoted to type complex before performing the operation.\n\n\n \n\ncomplex operator+(complex z) Return the complex argument z, \\[\\begin{equation*} +z = z. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex z) Return the negation of the complex argument z, which for \\(z = x + yi\\) is \\[\\begin{equation*} -z = -x - yi. \\end{equation*}\\]\nAvailable since 2.28\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of complex numbers, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex operator+(complex x, complex y) Return the sum of x and y, \\[\\begin{equation*} (x + y) = \\text{operator+}(x, y) = x + y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex x, complex y) Return the difference between x and y, \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x, y) = x - y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator*(complex x, complex y) Return the product of x and y, \\[\\begin{equation*} (x \\, * \\, y) = \\text{operator*}(x, y) = x\n\\times y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator/(complex x, complex y) Return the quotient of x and y, \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator^(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n(x^\\mathrm{\\wedge}y)= \\text{operator}^\\mathrm{\\wedge}(x,y)\n= \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28",
+    "text": "NOTE: Although the RDump format is still supported, I/O with JSON is faster and recommended. See the chapter on JSON for more details.\nRDump format can be used to represent values for Stan variables. This format was introduced in SPLUS and is used in R, JAGS, and in BUGS (but with a different ordering).\nA dump file is structured as a sequence of variable definitions. Each variable is defined in terms of its dimensionality and its values. There are three kinds of variable declarations: - scalars - sequences - general arrays\n\n\nDump files can be created from R using RStan, via the rstan package function stan_rdump. Stan RDump files must be created via stan_rdump and not by R’s native dump function because R’s dump function uses a richer syntax than is supported by the underlying Stan i/o libraries.\n\n\n\nA simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable y with value \\(17.2\\):\ny &lt;- 17.2\n\n\n\nOne-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.\nn &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)\nArrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.\nSequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as\nn &lt;- 1:3\nThe sequence denoted by 1:3 is of length \\(3\\), running from \\(1\\) to \\(3\\) inclusive. The colon notation allows sequences going from high to low. The following are equivalent:\nn &lt;- 2:-2\nn &lt;- c(2,1,0,-1,-2)\nAs a special case, a sequence of zeros can also be represented in the dump format by integer(x) and double(x), for type int and double, respectively. Here x is a non-negative integer to specify the length. If x is \\(0\\), it can be omitted. The following are some examples.\nx1 &lt;- integer()\nx2 &lt;- integer(0)\nx3 &lt;- integer(2)\ny1 &lt;- double()\ny2 &lt;- double(0)\ny3 &lt;- double(2)\n\n\n\nFor more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a \\(2 \\times 3\\) array:\ny &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))\nData is stored column-major, thus the values for y will be:\ny[1, 1] = 1\ny[1, 2] = 3\ny[1, 3] = 5\ny[2, 1] = 2\ny[2, 2] = 4\ny[2, 3] = 6\nThe structure keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.\nIf the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.\ny &lt;- structure(1:6, .Dim = c(2,3))\nSequence notation can be used within any call to the generic c() function in R. In the above example, c(2,3) could be written as c(2:3).\nThe generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,\nz &lt;- structure(1:24, .Dim = c(2,3,4))\nproduces a three-dimensional int (assignable to real) array z with values:\nz[1, 1, 1] = 1\nz[2, 1, 1] = 2\nz[1, 2, 1] = 3\nz[2, 2, 1] = 4\nz[1, 3, 1] = 5\nz[2, 3, 1] = 6\nz[1, 1, 2] = 7\nz[2, 1, 2] = 8\nz[1, 2, 2] = 9\nz[2, 2, 2] = 10\nz[1, 3, 2] = 11\nz[2, 3, 2] = 12\nz[1, 1, 3] = 13\nz[2, 1, 3] = 14\nz[1, 2, 3] = 15\nz[2, 2, 3] = 16\nz[1, 3, 3] = 17\nz[2, 3, 3] = 18\nz[1, 1, 4] = 19\nz[2, 1, 4] = 20\nz[1, 2, 4] = 21\nz[2, 2, 4] = 22\nz[1, 3, 4] = 23\nz[2, 3, 4] = 24\nIf the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.\nThe sequence of values inside structure can also be integer(x) or double(x). In particular, if one or more dimensions is zero, integer() can be put inside structure. For instance, the following example is supported by the dump format.\ny &lt;- structure(integer(), .Dim = c(2,0))\n\n\n\nThe dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.\n\n\nThe following three declarations have the same dump format for their data.\narray[K] real a;\nvector[K] b;\nrow_vector[K] c;\n\n\n\nThe following declarations have the same dump format.\narray[M, N] real a;\nmatrix[M, N] b;\n\n\n\nThe key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.\nFor the dump data format, the following declarations have the same arrangement.\narray[M, N] real a;\nmatrix[M, N] b;\narray[M] vector[N] c;\narray[M] row_vector[N] d;\nSimilarly, the following also have the same dump format.\narray[P, M, N] real a;\narray[P] matrix[M, N] b;\narray[P, M] vector[N] c;\narray[P, M] row_vector[N] d;\n\n\n\n\nAt this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.\n\n\n\nThere is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., \\(132.3\\)) or uses scientific notation (e.g., \\(1.323e2\\)), Stan assumes that the values are real.\nFor a single value, if there is no decimal point, it may be assigned to an int or real variable in Stan. An array value may only be assigned to an int array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.\nThe following dump file declares an integer value for y.\ny &lt;- 2\nThis definition can be used for a Stan variable y declared as real or as int. Assigning an integer value to a real variable automatically promotes the integer value to a real value.\nInteger values may optionally be followed by L or l, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.\ny &lt;- 2L\nThe following dump file provides a real value for y.\ny &lt;- 2.0\nEven though this is a round value, the occurrence of the decimal point in the value, \\(2.0\\), causes Stan to infer that y is real valued. This dump file may only be used for variables y declared as real in Stan.\n\n\nNumbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as \\(1e+06\\).\n\n\n\nStan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.\n\n\n\nValue\nPreferred Form\nAlternative Forms\n\n\n\n\npositive infinity\nInf\nInfinity, infinity\n\n\nnegative infinity\n-Inf\n-Infinity, -infinity\n\n\nnot a number\nNaN\n\n\n\n\nThese strings are not case sensitive, so inf may also be used for positive infinity, or NAN for not-a-number.\n\n\n\n\nIn order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.\n\"y\" &lt;- c(1,2,3) \\end{Verbatim}\n\n\n\nThe line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.\ny &lt;- 2\ny &lt;-\n3\nAlso following R, breaking before the assignment arrow are not allowed, so the following is invalid.\ny\n&lt;- 2 # Syntax Error\nLines may also be broken in the middle of sequences declared using the c(...) notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a \\(2 \\times 2 \\times 3\\) array is valid.\ny &lt;-\nstructure(c(1,2,3,\n4,5,6,7,8,9,10,11,\n12), .Dim = c(2,2,\n3))\nBecause there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as int or real.\n\n\n\nA more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.\ndefinition ::= name &lt;- value optional_semicolon\n\nname ::= char*     | ''' char* '''     | '\"' char* '\"'\n\nvalue ::= value&lt;int&gt; | value&lt;double&gt;\n\nvalue&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |\n'structure' '(' seq&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'       | 'structure'\n'(' zero_array&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'\n\nseq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;\n\nzero_array&lt;int&gt; ::= \"integer\" '(' &lt;non-negative int&gt;? ')'\n\nzero_array&lt;real&gt; ::= \"double\" '(' &lt;non-negative int&gt;? ')'\n\nseq&lt;real&gt; ::= cseq&lt;real&gt;\n\ncseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'\n\nvseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;\nThe template parameters T will be set to either int or real. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-comparison",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-comparison",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html#creating-dump-files",
+    "href": "cmdstan-guide/rdump_apdx.html#creating-dump-files",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "Complex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional\nz1 == z2\nis equivalent to\nget_real(z1) == get_real(z2) && get_imag(z1) == get_imag(z2)\nAs with other complex functions, if one of the arguments is of type real or int, it will be promoted to type complex before comparison. For example, if z is of type complex, then z == 0 will be true if z has real component equal to 0.0 and complex component equal to 0.0.\nWarning: As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.\n \n\nint operator==(complex x, complex y) Return 1 if x is equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{==}\\, y)\n\\ = \\ \\text{operator==}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x = y$}, \\ \\text{and} \\\\ 0 & \\text{otherwise.}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nint operator!=(complex x, complex y) Return 1 if x is not equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{!=}\\, y)\n\\ = \\ \\text{operator!=}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x \\neq y$}, \\ \\text{and} \\\\ 0 &\n\\text{otherwise.} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28",
+    "text": "Dump files can be created from R using RStan, via the rstan package function stan_rdump. Stan RDump files must be created via stan_rdump and not by R’s native dump function because R’s dump function uses a richer syntax than is supported by the underlying Stan i/o libraries.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-assignment",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-assignment",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html#scalar-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#scalar-variables",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "The assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.\nAssignment of complex numbers works elementwise. If an expression of type int or real is assigned to a complex number, it will be promoted before assignment as if calling to_complex(), so that the imaginary component is 0.0.\n \n\nvoid operator=(complex x, complex y) y = x; assigns a (copy of) the value of y to x.\nAvailable since 2.28\n \n\nvoid operator+=(complex x, complex y) x += y; is equivalent to x = x + y;.\nAvailable since 2.28\n \n\nvoid operator-=(complex x, complex y) x -= y; is equivalent to x = x - y;.\nAvailable since 2.28\n \n\nvoid operator*=(complex x, complex y) x *= y; is equivalent to x = x * y;.\nAvailable since 2.28\n \n\nvoid operator/=(complex x, complex y) x /= y; is equivalent to x = x / y;.\nAvailable since 2.28",
+    "text": "A simple scalar value can be thought of as having an empty list of dimensions. Its declaration in the dump format follows the SPLUS assignment syntax. For example, the following would constitute a valid dump file defining a single scalar variable y with value \\(17.2\\):\ny &lt;- 17.2",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-special",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-special",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html#sequence-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#sequence-variables",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "The following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.\n \n\nreal abs(complex z) Return the absolute value of z, also known as the modulus or magnitude, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{abs}(z) = \\sqrt{x^2 + y^2}.\n\\end{equation*}\\]\nThis function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a complex_vector[n] input will return a vector[n] output, with each element transformed by the above equation.\nAvailable since 2.28, vectorized in 2.30\n \n\nreal arg(complex z) Return the phase angle (in radians) of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{arg}(z) = \\textrm{atan2}(y, x) = \\textrm{atan}(y / x).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nreal norm(complex z) Return the Euclidean norm of z, which is its absolute value squared, and which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{norm}(z) = \\textrm{abs}^2(z) = x^2 + y^2.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex conj(complex z) Return the complex conjugate of z, which negates the imaginary component, so that if \\(z = x + yi\\), \\[\\begin{equation*}\n\\textrm{conj}(z) = x - yi.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ conj(Z z) Vectorized version of conj. This will apply the conj function to each element of a complex array, vector, or matrix.\nAvailable since 2.31\n \n\ncomplex proj(complex z) Return the projection of z onto the Riemann sphere, which for \\(z = x\n+ yi\\) is \\[\\begin{equation*}\n\\textrm{proj}(z)\n= \\begin{cases}\n    z & \\textrm{if} \\ z \\ \\textrm{is finite, and} \\\\\n    0 + \\textrm{sign}(y)i & \\textrm{otherwise,}\n\\end{cases}\n\\end{equation*}\\] where \\(\\textrm{sign}(y)\\) is -1 if \\(y\\) is negative and 1 otherwise.\nAvailable since 2.28\n \n\ncomplex polar(real r, real theta) Return the complex number with magnitude (absolute value) r and phase angle theta.\nAvailable since 2.28",
+    "text": "One-dimensional arrays may be specified directly using the SPLUS sequence notation. The following example defines an integer-value and a real-valued sequence.\nn &lt;- c(1,2,3) y &lt;- c(2.0,3.0,9.7)\nArrays are provided without a declaration of dimensionality because the reader just counts the number of entries to determine the size of the array.\nSequence variables may alternatively be represented with R’s colon-based notation. For instance, the first example above could equivalently be written as\nn &lt;- 1:3\nThe sequence denoted by 1:3 is of length \\(3\\), running from \\(1\\) to \\(3\\) inclusive. The colon notation allows sequences going from high to low. The following are equivalent:\nn &lt;- 2:-2\nn &lt;- c(2,1,0,-1,-2)\nAs a special case, a sequence of zeros can also be represented in the dump format by integer(x) and double(x), for type int and double, respectively. Here x is a non-negative integer to specify the length. If x is \\(0\\), it can be omitted. The following are some examples.\nx1 &lt;- integer()\nx2 &lt;- integer(0)\nx3 &lt;- integer(2)\ny1 &lt;- double()\ny2 &lt;- double(0)\ny3 &lt;- double(2)",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-exponential",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-exponential",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html#array-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#array-variables",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "The exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.\n \n\ncomplex exp(complex z) Return the complex natural exponential of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\exp z = \\exp(x) \\textrm{cis}(y) = \\exp(x) (\\cos(y) + i \\sin(y)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log(complex z) Return the complex natural logarithm of z, which for \\(z = \\textrm{polar}(r,\n\\theta)\\) is \\[\\begin{equation*}\n\\log z = \\log r + \\theta i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log10(complex z) Return the complex common logarithm of z, \\[\\begin{equation*}\n\\log_{10} z = \\frac{\\log z}{\\log 10}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex pow(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n\\text{pow}(x,y) = \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.30\n \n\ncomplex sqrt(complex x) Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.\nAvailable since 2.28",
+    "text": "For more than one dimension, the dump format uses a dimensionality specification. For example, the following defines a \\(2 \\times 3\\) array:\ny &lt;- structure(c(1,2,3,4,5,6), .Dim = c(2,3))\nData is stored column-major, thus the values for y will be:\ny[1, 1] = 1\ny[1, 2] = 3\ny[1, 3] = 5\ny[2, 1] = 2\ny[2, 2] = 4\ny[2, 3] = 6\nThe structure keyword just wraps a sequence of values and a dimensionality declaration, which is itself just a sequence of non-negative integer values. The product of the dimensions must equal the length of the array.\nIf the values happen to form a contiguous sequence of integers, they may be written with colon notation. Thus the example above is equivalent to the following.\ny &lt;- structure(1:6, .Dim = c(2,3))\nSequence notation can be used within any call to the generic c() function in R. In the above example, c(2,3) could be written as c(2:3).\nThe generalization of column-major indexing is last-index major indexing. Arrays of more than two dimensions are written in a last-index major form. For example,\nz &lt;- structure(1:24, .Dim = c(2,3,4))\nproduces a three-dimensional int (assignable to real) array z with values:\nz[1, 1, 1] = 1\nz[2, 1, 1] = 2\nz[1, 2, 1] = 3\nz[2, 2, 1] = 4\nz[1, 3, 1] = 5\nz[2, 3, 1] = 6\nz[1, 1, 2] = 7\nz[2, 1, 2] = 8\nz[1, 2, 2] = 9\nz[2, 2, 2] = 10\nz[1, 3, 2] = 11\nz[2, 3, 2] = 12\nz[1, 1, 3] = 13\nz[2, 1, 3] = 14\nz[1, 2, 3] = 15\nz[2, 2, 3] = 16\nz[1, 3, 3] = 17\nz[2, 3, 3] = 18\nz[1, 1, 4] = 19\nz[2, 1, 4] = 20\nz[1, 2, 4] = 21\nz[2, 2, 4] = 22\nz[1, 3, 4] = 23\nz[2, 3, 4] = 24\nIf the underlying 3-D array is stored as a 1-D array in last-index major format, the innermost array elements will be contiguous.\nThe sequence of values inside structure can also be integer(x) or double(x). In particular, if one or more dimensions is zero, integer() can be put inside structure. For instance, the following example is supported by the dump format.\ny &lt;- structure(integer(), .Dim = c(2,0))",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-trig",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-trig",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html#matrix--and-vector-valued-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#matrix--and-vector-valued-variables",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "The standard trigonometric functions are supported for complex numbers.\n \n\ncomplex cos(complex z) Return the complex cosine of z, which is \\[\\begin{equation*}\n\\cos(z)\n= \\textrm{cosh}(z \\, i)\n= \\frac{\\displaystyle \\exp(z \\, i) + \\exp(-z \\, i)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sin(complex z) Return the complex sine of z, \\[\\begin{equation*}\n\\sin(z)\n= -\\textrm{sinh}(z \\, i) \\, i\n= \\frac{\\displaystyle \\exp(z \\, i) - \\exp(-z \\, i)}\n       {\\displaystyle 2 \\, i}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tan(complex z) Return the complex tangent of z, \\[\\begin{equation*}\n\\tan(z)\n= -\\textrm{tanh}(z \\, i) \\, i\n= \\frac{(\\exp(-z \\, i) - \\exp(z \\, i)) \\, i}\n       {\\exp(-z \\, i) + \\exp(z \\, i)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acos(complex z) Return the complex arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acos}(z)\n= \\frac{1}{2} \\pi + \\log (z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asin(complex z) Return the complex arc (inverse) sine of z, \\[\\begin{equation*}\n\\text{asin}(z)\n= -\\log(z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atan(complex z) Return the complex arc (inverse) tangent of z, \\[\\begin{equation*}\n\\text{atan}(z)\n= - \\frac{1}{2} (\\log(1 - z \\, i) - \\log(1 + z \\, i)) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28",
+    "text": "The dump format for matrices and vectors, including arrays of matrices and vectors, is the same as that for arrays of the same shape.\n\n\nThe following three declarations have the same dump format for their data.\narray[K] real a;\nvector[K] b;\nrow_vector[K] c;\n\n\n\nThe following declarations have the same dump format.\narray[M, N] real a;\nmatrix[M, N] b;\n\n\n\nThe key to understanding arrays is that the array indexing comes before any of the container indexing. That is, an array of vectors is just that: each array element is a vector. See the chapter on array and matrix types in the user’s guide section of the language manual for more information.\nFor the dump data format, the following declarations have the same arrangement.\narray[M, N] real a;\nmatrix[M, N] b;\narray[M] vector[N] c;\narray[M] row_vector[N] d;\nSimilarly, the following also have the same dump format.\narray[P, M, N] real a;\narray[P] matrix[M, N] b;\narray[P, M] vector[N] c;\narray[P, M] row_vector[N] d;",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-h-trig",
-    "href": "functions-reference/complex-valued_basic_functions.html#complex-h-trig",
-    "title": "Complex-Valued Basic Functions",
+    "objectID": "cmdstan-guide/rdump_apdx.html#complex-valued-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#complex-valued-variables",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "The standard hyperbolic trigonometric functions are supported for complex numbers.\n \n\ncomplex cosh(complex z) Return the complex hyperbolic cosine of z, \\[\\begin{equation*}\n\\textrm{cosh}(z)\n= \\frac{\\exp(z) + \\exp(-z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sinh(complex z) Return the complex hyperbolic sine of z, \\[\\begin{equation*}\n\\textrm{sinh}(z)\n= \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tanh(complex z) Return the complex hyperbolic tangent of z, \\[\\begin{equation*}\n\\textrm{tanh}(z)\n\\ = \\ \\frac{\\textrm{sinh}(z)}\n           {\\textrm{cosh}(z)}\n\\ = \\ \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n           {\\displaystyle \\exp(z) + \\exp(-z)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acosh(complex z) Return the complex hyperbolic arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acosh}(z)\n= \\log(z + \\sqrt{(z + 1)(z - 1)}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asinh(complex z) Return the complex hyperbolic arc (inverse) sine of z, \\[\\begin{equation*}\n\\textrm{asinh}(z)\n= \\log(z + \\sqrt{1 + z^2}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atanh(complex z) Return the complex hyperbolic arc (inverse) tangent of z, \\[\\begin{equation*}\n\\textrm{atanh}(z)\n= \\frac{\\log(1 + z) - \\log(1 - z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28",
+    "text": "At this time, there is no support for complex number input through the R dump format. As an alternative, the JSON input format supports complex numbers.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Complex-Valued Basic Functions"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/rdump_apdx.html#integer--and-real-valued-variables",
+    "href": "cmdstan-guide/rdump_apdx.html#integer--and-real-valued-variables",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "Compound arithmetic and assignment statements combine an arithmetic operation and assignment, replacing a statement such as\n x = x op y;\nwith the more compact compound form\n x op= y;\nFor example, x = x + 1; may be replaced with x += 1;. This works for all types that support arithmetic, including the scalar types int, real, complex, the real matrix types vector, row_vector, and matrix, and the complex matrix types, complex_vector, complex_row_vector, and complex_matrix.\n\n\nCompound addition and assignment works wherever the corresponding addition and assignment would be well formed.\n \n\nvoid operator+=(T x, U y) x += y is equivalent to x = x + y. Defined for all types T and U where T = T + U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.\n \n\nvoid operator-=(T x, U y) x -= y is equivalent to x = x - y. Defined for all types T and U where T = T - U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator*=(T x, U y) x *= y is equivalent to x = x * y. Defined for all types T and U where T = T * U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator/=(T x, U y) x /= y is equivalent to x = x / y. Defined for all types T and U where T = T / U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator.*=(T x, U y) x .*= y is equivalent to x = x .* y. Defined for all types T and U where T = T .* U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator./=(T x, U y) x ./= y is equivalent to x = x ./ y. Defined for all types T and U where T = T ./ U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "There is no declaration in a dump file that distinguishes integer versus continuous values. If a value in a dump file’s definition of a variable contains a decimal point (e.g., \\(132.3\\)) or uses scientific notation (e.g., \\(1.323e2\\)), Stan assumes that the values are real.\nFor a single value, if there is no decimal point, it may be assigned to an int or real variable in Stan. An array value may only be assigned to an int array if there is no decimal point or scientific notation in any of the values. This convention is compatible with the way R writes data.\nThe following dump file declares an integer value for y.\ny &lt;- 2\nThis definition can be used for a Stan variable y declared as real or as int. Assigning an integer value to a real variable automatically promotes the integer value to a real value.\nInteger values may optionally be followed by L or l, denoting long integer values. The following example, where the type is explicit, is equivalent to the above.\ny &lt;- 2L\nThe following dump file provides a real value for y.\ny &lt;- 2.0\nEven though this is a round value, the occurrence of the decimal point in the value, \\(2.0\\), causes Stan to infer that y is real valued. This dump file may only be used for variables y declared as real in Stan.\n\n\nNumbers written in scientific notation may only be used for real values in Stan. R will write out the integer one million as \\(1e+06\\).\n\n\n\nStan’s reader supports infinite and not-a-number values for scalar quantities (see the section of the reference manual section of the language manual for more information on Stan’s numerical data types). Both infinite and not-a-number values are supported by Stan’s dump-format readers.\n\n\n\nValue\nPreferred Form\nAlternative Forms\n\n\n\n\npositive infinity\nInf\nInfinity, infinity\n\n\nnegative infinity\n-Inf\n-Infinity, -infinity\n\n\nnot a number\nNaN\n\n\n\n\nThese strings are not case sensitive, so inf may also be used for positive infinity, or NAN for not-a-number.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-addition-and-assignment",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-addition-and-assignment",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/rdump_apdx.html#quoted-variable-names",
+    "href": "cmdstan-guide/rdump_apdx.html#quoted-variable-names",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "Compound addition and assignment works wherever the corresponding addition and assignment would be well formed.\n \n\nvoid operator+=(T x, U y) x += y is equivalent to x = x + y. Defined for all types T and U where T = T + U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "In order to support JAGS data files, variables may be double quoted. For instance, the following definition is legal in a dump file.\n\"y\" &lt;- c(1,2,3) \\end{Verbatim}",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-subtraction-and-assignment",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-subtraction-and-assignment",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/rdump_apdx.html#line-breaks",
+    "href": "cmdstan-guide/rdump_apdx.html#line-breaks",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "Compound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.\n \n\nvoid operator-=(T x, U y) x -= y is equivalent to x = x - y. Defined for all types T and U where T = T - U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "The line breaks in a dump file are required to be consistent with the way R reads in data. Both of the following declarations are legal.\ny &lt;- 2\ny &lt;-\n3\nAlso following R, breaking before the assignment arrow are not allowed, so the following is invalid.\ny\n&lt;- 2 # Syntax Error\nLines may also be broken in the middle of sequences declared using the c(...) notation., as well as between the comma following a sequence definition and the dimensionality declaration. For example, the following declaration of a \\(2 \\times 2 \\times 3\\) array is valid.\ny &lt;-\nstructure(c(1,2,3,\n4,5,6,7,8,9,10,11,\n12), .Dim = c(2,2,\n3))\nBecause there are no decimal points in the values, the resulting dump file may be used for three-dimensional array variables declared as int or real.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-multiplication-and-assignment",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-multiplication-and-assignment",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/rdump_apdx.html#bnf-grammar-for-dump-data",
+    "href": "cmdstan-guide/rdump_apdx.html#bnf-grammar-for-dump-data",
+    "title": "RDump Format for CmdStan",
     "section": "",
-    "text": "Compound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator*=(T x, U y) x *= y is equivalent to x = x * y. Defined for all types T and U where T = T * U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "A more precise definition of the dump data format is provided by the following (mildly templated) Backus-Naur form grammar.\ndefinition ::= name &lt;- value optional_semicolon\n\nname ::= char*     | ''' char* '''     | '\"' char* '\"'\n\nvalue ::= value&lt;int&gt; | value&lt;double&gt;\n\nvalue&lt;T&gt; ::= T       | seq&lt;T&gt;       | zero_array&lt;T&gt;       |\n'structure' '(' seq&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'       | 'structure'\n'(' zero_array&lt;T&gt; ',' \".Dim\" '=' seq&lt;int&gt; ')'\n\nseq&lt;int&gt; ::= int ':' int       | cseq&lt;int&gt;\n\nzero_array&lt;int&gt; ::= \"integer\" '(' &lt;non-negative int&gt;? ')'\n\nzero_array&lt;real&gt; ::= \"double\" '(' &lt;non-negative int&gt;? ')'\n\nseq&lt;real&gt; ::= cseq&lt;real&gt;\n\ncseq&lt;T&gt; ::= 'c' '(' vseq&lt;T&gt; ')'\n\nvseq&lt;T&gt; ::= T      | T ',' vseq&lt;T&gt;\nThe template parameters T will be set to either int or real. Because Stan allows promotion of integer values to real values, an integer sequence specification in the dump data format may be assigned to either an integer- or real-based variable in Stan.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
+      "Appendices",
+      "RDump Format for CmdStan"
     ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-division-and-assignment",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-division-and-assignment",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/stanc.html",
+    "href": "cmdstan-guide/stanc.html",
+    "title": "stanc: Translating Stan to C++",
     "section": "",
-    "text": "Compound division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator/=(T x, U y) x /= y is equivalent to x = x / y. Defined for all types T and U where T = T / U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "CmdStan translates Stan programs to C++ using the Stan compiler program which is included in the CmdStan release bin directory as program stanc. One can view the complete stanc documentation in the Stan User’s Guide.\nAs of release 2.22, the CmdStan Stan to C++ compiler is written in OCaml. This compiler is called “stanc3” and has has its own repository https://github.com/stan-dev/stanc3, from which pre-built binaries for Linux, Mac, and Windows can be downloaded.\n\n\nBefore the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:\nmake bin/stanc\nFor Windows:\nmake bin/stanc.exe\nThis is also done as part of the make build command.\n\n\n\nThe Stan compiler program stanc converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model bernoulli.stan. For Linux and Mac:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nFor Windows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nThe base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (a--z or A--Z) and contain only alphanumeric characters (a--z, A--Z, and 0--9) and underscores (_) and should not conflict with any C++ reserved keyword.\nThe C++ code implementing the class is written to the file bernoulli.hpp in the current directory. The final argument, bernoulli.stan, is the file from which to read the Stan program.\nIn practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via make, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:\n&gt; make examples/bernoulli/bernoulli\nFor Windows:\n&gt; make examples/bernoulli/bernoulli.exe\nThe makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable STANCFLAGS can be used to to override the default arguments to stanc, e.g.,\n&gt; make STANCFLAGS=\"--include-paths=~/foo\" examples/bernoulli/bernoulli",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
-    ]
+      "Tools and Utilities",
+      "`stanc`: Translating Stan to C++"
+    ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-multiplication-and-assignment",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-multiplication-and-assignment",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/stanc.html#instantiating-the-stanc-binary",
+    "href": "cmdstan-guide/stanc.html#instantiating-the-stanc-binary",
+    "title": "stanc: Translating Stan to C++",
     "section": "",
-    "text": "Compound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator.*=(T x, U y) x .*= y is equivalent to x = x .* y. Defined for all types T and U where T = T .* U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "Before the Stan compiler can be used, the binary stanc must be created. This can be done using the makefile as follows. For Mac and Linux:\nmake bin/stanc\nFor Windows:\nmake bin/stanc.exe\nThis is also done as part of the make build command.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
+      "Tools and Utilities",
+      "`stanc`: Translating Stan to C++"
     ]
   },
   {
-    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-division-and-assignment",
-    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-division-and-assignment",
-    "title": "Compound Arithmetic and Assignment",
+    "objectID": "cmdstan-guide/stanc.html#the-stan-compiler-program",
+    "href": "cmdstan-guide/stanc.html#the-stan-compiler-program",
+    "title": "stanc: Translating Stan to C++",
     "section": "",
-    "text": "Compound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator./=(T x, U y) x ./= y is equivalent to x = x ./ y. Defined for all types T and U where T = T ./ U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
+    "text": "The Stan compiler program stanc converts Stan programs to C++ concepts. If the compiler encounters syntax errors in the program, it will provide an error message indicating the location in the input where the failure occurred and reason for the failure. The following example illustrates a fully qualified call to stanc to generate the C++ translation of the example model bernoulli.stan. For Linux and Mac:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nFor Windows:\n&gt; cd &lt;cmdstan-home&gt;\n&gt; bin/stanc.exe --o=bernoulli.hpp examples/bernoulli/bernoulli.stan\nThe base name of the Stan program file determines the name of the C++ model class. Because this name is the name of a C++ class, it must start with an alphabetic character (a--z or A--Z) and contain only alphanumeric characters (a--z, A--Z, and 0--9) and underscores (_) and should not conflict with any C++ reserved keyword.\nThe C++ code implementing the class is written to the file bernoulli.hpp in the current directory. The final argument, bernoulli.stan, is the file from which to read the Stan program.\nIn practice, stanc is invoked indirectly, via the GNU Make utility, which contains rules that compile a Stan program to its corresponding executable. To build the simple Bernoulli model via make, we specify the name of the target executable file. On Mac and Linux, this is the name of the Stan program with the .stan omitted. On Windows, replace .stan with .exe, and make sure that the path is given with slashes and not backslashes. For Linux and Mac:\n&gt; make examples/bernoulli/bernoulli\nFor Windows:\n&gt; make examples/bernoulli/bernoulli.exe\nThe makefile rules first invoke the stanc compiler to translate the Stan model to C++ , then compiles and links the C++ code to a binary executable. The makefile variable STANCFLAGS can be used to to override the default arguments to stanc, e.g.,\n&gt; make STANCFLAGS=\"--include-paths=~/foo\" examples/bernoulli/bernoulli",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Compound Arithmetic and Assignment"
+      "Tools and Utilities",
+      "`stanc`: Translating Stan to C++"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html",
-    "href": "functions-reference/conventions_for_probability_functions.html",
-    "title": "Conventions for Probability Functions",
+    "objectID": "cmdstan-guide/variational_config.html",
+    "href": "cmdstan-guide/variational_config.html",
+    "title": "Variational Inference using ADVI",
     "section": "",
-    "text": "Functions associated with distributions are set up to follow the same naming conventions for both built-in distributions and for user-defined distributions.\n\n\nThe suffix is determined by the type of function according to the following table.\n\n\n\n\n\n\n\n\nfunction\noutcome\nsuffix\n\n\n\n\nlog probability mass function\ndiscrete\n_lpmf\n\n\nlog probability density function\ncontinuous\n_lpdf\n\n\nlog cumulative distribution function\nany\n_lcdf\n\n\nlog complementary cumulative distribution function\nany\n_lccdf\n\n\nrandom number generator\nany\n_rng\n\n\n\nFor example, normal_lpdf is the log of the normal probability density function (pdf) and bernoulli_lpmf is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, normal_lcdf and bernoulli_lcdf.\n\n\n\nEach probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, normal_lpdf(y | mu, sigma) returns the value of mathematical formula \\(\\log \\text{Normal}(y \\, | \\, \\mu, \\sigma)\\). Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., normal_lcdf(y_low | mu, sigma)\n\n\n\nThe notation\n y ~ normal(mu, sigma);\nprovides the same (proportional) contribution to the model log density as the explicit target density increment,\n target += normal_lpdf(y | mu, sigma);\nIn both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (~) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.\n\n\n\nAll of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.\n\n\n\nMany distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter \\(\\sigma &gt; 0\\). If \\(\\sigma = 0\\), the probability function will throw an exception.\nThis is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.\n\n\n\nFor most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix _rng. For example, the function normal_rng(real, real) accepts two real arguments, an unconstrained location \\(\\mu\\) and positive scale \\(\\sigma &gt; 0\\), and returns an unconstrained pseudorandom value drawn from \\(\\text{Normal}(\\mu,\\sigma)\\). There are also vectorized forms of random number generators which return more than one random variate at a time.\n\n\nUnlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.\n\n\n\nUnlike the probability functions, only some of the PRNG functions are vectorized.\n\n\n\n\nFor most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.\nFor a univariate random variable \\(Y\\) with probability function \\(p_Y(y \\, | \\, \\theta)\\), the cumulative distribution function (CDF) \\(F_Y\\) is defined by \\[\\begin{equation*}\nF_Y(y) \\ = \\ \\text{Pr}[Y \\le y] \\ = \\ \\int_{-\\infty}^y p(y\\, | \\, \\theta) \\ \\text{d}y.\n\\end{equation*}\\] The complementary cumulative distribution function (CCDF) is defined as \\[\\begin{equation*}\n\\text{Pr}[Y &gt; y] \\ = \\ 1 - F_Y(y).\n\\end{equation*}\\] The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly \\(10^{-300}\\)), but not numbers very close to 1 (the closest you can get is roughly \\(1 - 10^{-15}\\)).\nIn Stan, there is a cumulative distribution function for each probability function. For instance, normal_cdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\ \\text{d}y.\n\\end{equation*}\\] There are also log forms of the CDF and CCDF for most univariate distributions. For example, normal_lcdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right)\n\\end{equation*}\\] and normal_lccdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( 1 - \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right).\n\\end{equation*}\\]\n\n\n\nStan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section vectorized PRNG functions for more details.\nIn all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.\nStan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.\nStan also overloads some scalar functions, such as log and exp, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.\n\n\n\n\nThe normal probability function is specified with the signature\n normal_lpdf(reals | reals, reals);\nThe pseudotype reals is used to indicate that an argument position may be vectorized. Argument positions declared as reals may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use normal_lpdf(row_vector | vector, real) as long as the vector and row vector have the same size.\n\n\n\nThe multivariate normal distribution accepting vector or array of vector arguments is written as\n multi_normal_lpdf(vectors | vectors, matrix);\nThese arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.\n\n\n\nThe pseudotype ints is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.\n\n\n\n\nThe result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely real or int, is repeated. For instance, if y is a vector of size N, mu is a vector of size N, and sigma is a scalar, then\n ll = normal_lpdf(y | mu, sigma);\nis just a more efficient way to write\n ll = 0;\n for (n in 1:N) {\n   ll = ll + normal_lpdf(y[n] | mu[n], sigma);\n }\nWith the same arguments, the vectorized sampling statement\n y ~ normal(mu, sigma);\nhas the same effect on the total log probability as\n for (n in 1:N) {\n   y[n] ~ normal(mu[n], sigma);\n }\n\n\n\nSome PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes reals or ints. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.\n vector[3] mu = // ...\n array[3] real x = normal_rng(mu, 3);\n\n\nIn the case of PRNG functions, arguments marked ints may be integers or integer arrays, whereas arguments marked reals may be integers or reals, integer or real arrays, vectors, or row vectors.\n\n\n\n\n\n\n\npseudotype\nallowable PRNG arguments\n\n\n\n\nints\nint, array[] int\n\n\nreals\nint, array[] int, real, array[] real, vector, row_vector\n\n\n\n\n\n\nIn general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the normal_rng function may be called with one vector argument and one real array argument as long as they have the same number of elements.\n vector[3] mu = // ...\n array[3] real sigma = // ...\n array[3] real x = normal_rng(mu, sigma);\n\n\n\nThe result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (array[] real) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return ints and continuous distributions return reals, each of appropriate size. The symbol R denotes such a return type.",
+    "text": "Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) Kucukelbir et al. (2017). ADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO (evidence lower bound). ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. The measures of convergence are similar to the relative tolerance scheme of Stan’s optimization algorithms.\nThe algorithm progression consists of an adaptation phase followed by a sampling phase. The adaptation phase finds a good value for the step size scaling parameter eta. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. The algorithm runs until the mean change in ELBO drops below the specified tolerance.\nThe full set of configuration options available for the variational method is available by using the variational help-all subcommand. The arguments with their requested values or defaults are also reported at the beginning of the algorithm’s console output and in the output CSV file’s comments.\nThe following is a minimal call to Stan’s variational inference algorithm using defaults for everything but the location of the data file.\n&gt; ./bernoulli variational data file=bernoulli.data.R\nExecuting this command prints both output to the console and to a csv file.\nThe first part of the console output reports on the configuration used: the default option algorithm=meanfield and the default tolerances for monitoring the algorithm’s convergence.\nmethod = variational\n  variational\n    algorithm = meanfield (Default)\n      meanfield\n    iter = 10000 (Default)\n    grad_samples = 1 (Default)\n    elbo_samples = 100 (Default)\n    eta = 1 (Default)\n    adapt\n      engaged = true (Default)\n      iter = 50 (Default)\n    tol_rel_obj = 0.01 (Default)\n    eval_elbo = 100 (Default)\n    output_samples = 1000 (Default)\nid = 1 (Default)\ndata\n  file = bernoulli.data.json\ninit = 2 (Default)\nrandom\n  seed = 2790599354 (Default)\noutput\n  file = output.csv (Default)\n  diagnostic_file =  (Default)\n  refresh = 100 (Default)\n  sig_figs = 8 (Default)\n  profile_file = profile.csv (Default)\n  save_cmdstan_config = false (Default)\nnum_threads = 1 (Default)\nAfter the configuration has been displayed, informational and timing messages are output:\n------------------------------------------------------------\nEXPERIMENTAL ALGORITHM:\n  This procedure has not been thoroughly tested and may be unstable\n  or buggy. The interface is subject to change.\n------------------------------------------------------------\n\nGradient evaluation took 2.1e-05 seconds\n1000 transitions using 10 leapfrog steps per transition would take 0.21 seconds.\nAdjust your expectations accordingly!\nThe rest of the output describes the progression of the algorithm. An adaptation phase finds a good value for the step size scaling parameter eta. The evidence lower bound (ELBO) is the variational objective function and is evaluated based on a Monte Carlo estimate. The variational inference algorithm in Stan is stochastic, which makes it challenging to assess convergence. That is, while the algorithm appears to have converged in \\(\\sim\\) 250 iterations, the algorithm runs for another few thousand iterations until mean change in ELBO drops below the default tolerance of 0.01.\nBegin eta adaptation.\nIteration:   1 / 250 [  0%]  (Adaptation)\nIteration:  50 / 250 [ 20%]  (Adaptation)\nIteration: 100 / 250 [ 40%]  (Adaptation)\nIteration: 150 / 250 [ 60%]  (Adaptation)\nIteration: 200 / 250 [ 80%]  (Adaptation)\nSuccess! Found best value [eta = 1] earlier than expected.\n\nBegin stochastic gradient ascent.\n  iter             ELBO   delta_ELBO_mean   delta_ELBO_med   notes\n   100           -6.131             1.000            1.000\n   200           -6.458             0.525            1.000\n   300           -6.300             0.359            0.051\n   400           -6.137             0.276            0.051\n   500           -6.243             0.224            0.027\n   600           -6.305             0.188            0.027\n   700           -6.289             0.162            0.025\n   800           -6.402             0.144            0.025\n   900           -6.103             0.133            0.025\n  1000           -6.314             0.123            0.027\n  1100           -6.348             0.024            0.025\n  1200           -6.244             0.020            0.018\n  1300           -6.293             0.019            0.017\n  1400           -6.250             0.017            0.017\n  1500           -6.241             0.015            0.010   MEDIAN ELBO CONVERGED\n\nDrawing a sample of size 1000 from the approximate posterior...\nCOMPLETED.\n\n\nStan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option algorithm=meanfield. The algorithm argument specifies the variational algorithm.\n\nalgorithm=meanfield - Use a fully factorized Gaussian for the approximation. This is the default algorithm.\nalgorithm=fullrank Use a Gaussian with a full-rank covariance matrix for the approximation.\n\n\n\n\n\niter=&lt;int&gt; Maximum number of iterations. Must be \\(&gt; 0\\). Default is \\(10000\\).\ngrad_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of gradients. Must be \\(&gt; 0\\). Default is \\(1\\).\nelbo_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of ELBO (objective function). Must be \\(&gt; 0\\). Default is \\(100\\).\neta=&lt;double&gt; Stepsize weighting parameter for adaptive stepsize sequence. Must be \\(&gt; 0\\). Default is \\(1.0\\).\nadapt Warmup Adaptation keyword, takes sub-arguments:\n\nengaged=&lt;boolean&gt; Adaptation engaged? Valid values: [true, false]. Default is true.\niter=&lt;int&gt; Maximum number of adaptation iterations. Must be \\(&gt; 0\\). Default is \\(50\\).\n\ntol_rel_obj=&lt;double&gt; Convergence tolerance on the relative norm of the objective. Must be \\(&gt; 0\\). Default is \\(0.01\\).\neval_elbo=&lt;int&gt; Evaluate ELBO every Nth iteration. Must be \\(&gt; 0\\). Default is 100.\noutput_samples=&lt;int&gt; Number of posterior samples to draw and save. Must be \\(&gt; 0\\). Default is 1000.\n\n\n\n\nThe output file consists of the following pieces of information:\n\nThe full set of configuration options available for the variational method is reported at the beginning of the sampler output file as CSV comments.\nThe first three output columns are labelled lp__, log_p__, log_g__, the rest are the model parameters.\nThe stepsize adaptation information is output as CSV comments following column header row.\nThe following line contains the mean of the variational approximation.\nThe rest of the output contains output_samples number of draws sampled from the variational approximation.\n\nTo illustrate, we call Stan’s variational inference on the example model and data:\n&gt; ./bernoulli variational data file=bernoulli.data.R\nBy default, the output file is output.csv.\nThe output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = variational\n#   variational\n#     algorithm = meanfield (Default)\n#       meanfield\n#     iter = 10000 (Default)\n#     grad_samples = 1 (Default)\n#     elbo_samples = 100 (Default)\n#     eta = 1 (Default)\n#     adapt\n#       engaged = true (Default)\n#       iter = 50 (Default)\n#     tol_rel_obj = 0.01 (Default)\n#     eval_elbo = 100 (Default)\n#     output_samples = 1000 (Default)\n...\nNext, the column header row:\nlp__,log_p__,log_g__,theta\nAdditional comments provide stepsize adaptation information:\n# Stepsize adaptation complete.\n# eta = 1\nFollowed by the data rows. The first line is special — it is the mean of the variational approximation.\n0,0,0,0.214911\nThat is, the estimate for theta given the data is 0.2.\nThe rest of the output contains output_samples number of draws samples from the variational approximation.\nThe following is a sample based on this approximation:\n0,-14.0252,-5.21718,0.770397\n0,-7.05063,-0.10025,0.162061\n0,-6.75031,-0.0191099,0.241606\n...\nThe header indicates the unnormalized log probability with lp__. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Running CmdStan",
+      "Variational Inference using ADVI"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#suffix-marks-type-of-function",
-    "href": "functions-reference/conventions_for_probability_functions.html#suffix-marks-type-of-function",
-    "title": "Conventions for Probability Functions",
+    "objectID": "cmdstan-guide/variational_config.html#variational-algorithms",
+    "href": "cmdstan-guide/variational_config.html#variational-algorithms",
+    "title": "Variational Inference using ADVI",
     "section": "",
-    "text": "The suffix is determined by the type of function according to the following table.\n\n\n\n\n\n\n\n\nfunction\noutcome\nsuffix\n\n\n\n\nlog probability mass function\ndiscrete\n_lpmf\n\n\nlog probability density function\ncontinuous\n_lpdf\n\n\nlog cumulative distribution function\nany\n_lcdf\n\n\nlog complementary cumulative distribution function\nany\n_lccdf\n\n\nrandom number generator\nany\n_rng\n\n\n\nFor example, normal_lpdf is the log of the normal probability density function (pdf) and bernoulli_lpmf is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, normal_lcdf and bernoulli_lcdf.",
+    "text": "Stan implements two variational algorithms. They differ in the approximating distribution used in the unconstrained variable space. By default, ADVI uses option algorithm=meanfield. The algorithm argument specifies the variational algorithm.\n\nalgorithm=meanfield - Use a fully factorized Gaussian for the approximation. This is the default algorithm.\nalgorithm=fullrank Use a Gaussian with a full-rank covariance matrix for the approximation.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Running CmdStan",
+      "Variational Inference using ADVI"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#argument-order-and-the-vertical-bar",
-    "href": "functions-reference/conventions_for_probability_functions.html#argument-order-and-the-vertical-bar",
-    "title": "Conventions for Probability Functions",
+    "objectID": "cmdstan-guide/variational_config.html#configuration",
+    "href": "cmdstan-guide/variational_config.html#configuration",
+    "title": "Variational Inference using ADVI",
     "section": "",
-    "text": "Each probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, normal_lpdf(y | mu, sigma) returns the value of mathematical formula \\(\\log \\text{Normal}(y \\, | \\, \\mu, \\sigma)\\). Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., normal_lcdf(y_low | mu, sigma)",
+    "text": "iter=&lt;int&gt; Maximum number of iterations. Must be \\(&gt; 0\\). Default is \\(10000\\).\ngrad_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of gradients. Must be \\(&gt; 0\\). Default is \\(1\\).\nelbo_samples=&lt;int&gt; Number of samples for Monte Carlo estimate of ELBO (objective function). Must be \\(&gt; 0\\). Default is \\(100\\).\neta=&lt;double&gt; Stepsize weighting parameter for adaptive stepsize sequence. Must be \\(&gt; 0\\). Default is \\(1.0\\).\nadapt Warmup Adaptation keyword, takes sub-arguments:\n\nengaged=&lt;boolean&gt; Adaptation engaged? Valid values: [true, false]. Default is true.\niter=&lt;int&gt; Maximum number of adaptation iterations. Must be \\(&gt; 0\\). Default is \\(50\\).\n\ntol_rel_obj=&lt;double&gt; Convergence tolerance on the relative norm of the objective. Must be \\(&gt; 0\\). Default is \\(0.01\\).\neval_elbo=&lt;int&gt; Evaluate ELBO every Nth iteration. Must be \\(&gt; 0\\). Default is 100.\noutput_samples=&lt;int&gt; Number of posterior samples to draw and save. Must be \\(&gt; 0\\). Default is 1000.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Running CmdStan",
+      "Variational Inference using ADVI"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#sampling-notation",
-    "href": "functions-reference/conventions_for_probability_functions.html#sampling-notation",
-    "title": "Conventions for Probability Functions",
+    "objectID": "cmdstan-guide/variational_config.html#csv-output",
+    "href": "cmdstan-guide/variational_config.html#csv-output",
+    "title": "Variational Inference using ADVI",
     "section": "",
-    "text": "The notation\n y ~ normal(mu, sigma);\nprovides the same (proportional) contribution to the model log density as the explicit target density increment,\n target += normal_lpdf(y | mu, sigma);\nIn both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (~) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.",
+    "text": "The output file consists of the following pieces of information:\n\nThe full set of configuration options available for the variational method is reported at the beginning of the sampler output file as CSV comments.\nThe first three output columns are labelled lp__, log_p__, log_g__, the rest are the model parameters.\nThe stepsize adaptation information is output as CSV comments following column header row.\nThe following line contains the mean of the variational approximation.\nThe rest of the output contains output_samples number of draws sampled from the variational approximation.\n\nTo illustrate, we call Stan’s variational inference on the example model and data:\n&gt; ./bernoulli variational data file=bernoulli.data.R\nBy default, the output file is output.csv.\nThe output follows the same pattern as the output for sampling, first dumping the entire set of parameters used as CSV comments:\n# stan_version_major = 2\n# stan_version_minor = 23\n# stan_version_patch = 0\n# model = bernoulli_model\n# method = variational\n#   variational\n#     algorithm = meanfield (Default)\n#       meanfield\n#     iter = 10000 (Default)\n#     grad_samples = 1 (Default)\n#     elbo_samples = 100 (Default)\n#     eta = 1 (Default)\n#     adapt\n#       engaged = true (Default)\n#       iter = 50 (Default)\n#     tol_rel_obj = 0.01 (Default)\n#     eval_elbo = 100 (Default)\n#     output_samples = 1000 (Default)\n...\nNext, the column header row:\nlp__,log_p__,log_g__,theta\nAdditional comments provide stepsize adaptation information:\n# Stepsize adaptation complete.\n# eta = 1\nFollowed by the data rows. The first line is special — it is the mean of the variational approximation.\n0,0,0,0.214911\nThat is, the estimate for theta given the data is 0.2.\nThe rest of the output contains output_samples number of draws samples from the variational approximation.\nThe following is a sample based on this approximation:\n0,-14.0252,-5.21718,0.770397\n0,-7.05063,-0.10025,0.162061\n0,-6.75031,-0.0191099,0.241606\n...\nThe header indicates the unnormalized log probability with lp__. This is a legacy feature that we do not use for variational inference. The ELBO is not stored unless a diagnostic option is given.",
     "crumbs": [
-      "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Running CmdStan",
+      "Variational Inference using ADVI"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#finite-inputs",
-    "href": "functions-reference/conventions_for_probability_functions.html#finite-inputs",
-    "title": "Conventions for Probability Functions",
+    "objectID": "functions-reference/binary_distributions.html",
+    "href": "functions-reference/binary_distributions.html",
+    "title": "Binary Distributions",
     "section": "",
-    "text": "All of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.",
+    "text": "Binary probability distributions have support on \\(\\{0,1\\}\\), where 1 represents the value true and 0 the value false.\n\n\n\n\nIf \\(\\theta \\in [0,1]\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{Bernoulli}(y~|~\\theta) = \\left\\{ \\begin{array}{ll} \\theta &\n\\text{if } y = 1, \\text{ and} \\\\ 1 - \\theta & \\text{if } y = 0.\n\\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli(theta)\nIncrement target log probability density with bernoulli_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_lpmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lupmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal bernoulli_cdf(ints y | reals theta) The Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.0\n \n\nreal bernoulli_lcdf(ints y | reals theta) The log of the Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lccdf(ints y | reals theta) The log of the Bernoulli complementary cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nints bernoulli_rng(reals theta) Generate a Bernoulli variate with chance of success theta or an array of Bernoulli variates given an array of thetas of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{BernoulliLogit}(y~|~\\alpha) = \\text{Bernoulli}(y |\n\\text{logit}^{-1}(\\alpha)) = \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha) & \\text{if } y = 1, \\text{ and} \\\\ 1 -\n\\text{logit}^{-1}(\\alpha) &  \\text{if } y = 0. \\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli_logit(alpha)\nIncrement target log probability density with bernoulli_logit_lupmf(y | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_logit_lpmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha)\nAvailable since 2.12\n \n\nreal bernoulli_logit_lupmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha) dropping constant additive terms\nAvailable since 2.25\n \n\nR bernoulli_logit_rng(reals alpha) Generate a Bernoulli variate with chance of success \\(\\text{logit}^{-1}(\\alpha)\\); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e. a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in {\\{0,1\\}}^n\\), \\[\\begin{align*}\n&\\text{BernoulliLogitGLM}(y~|~x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Bernoulli}(y_i~|~\\text{logit}^{-1}(\\alpha_i + x_i\\cdot\n\\beta))\\\\ &= \\prod_{1\\leq i \\leq n} \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j)\n& \\text{if } y_i = 1, \\text{ and} \\\\ 1 - \\text{logit}^{-1}(\\alpha_i +\n\\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) & \\text{if } y_i = 0.\n\\end{array} \\right. \\end{align*}\\]\n\n\n\ny ~ bernoulli_logit_glm(x, alpha, beta)\nIncrement target log probability density with bernoulli_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.25\n \n\n\n\n\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\narray[] int bernoulli_logit_glm_rng(matrix x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29\n \n\narray[] int bernoulli_logit_glm_rng(row_vector x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Discrete Distributions",
+      "Binary Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#boundary-conditions",
-    "href": "functions-reference/conventions_for_probability_functions.html#boundary-conditions",
-    "title": "Conventions for Probability Functions",
+    "objectID": "functions-reference/binary_distributions.html#bernoulli-distribution",
+    "href": "functions-reference/binary_distributions.html#bernoulli-distribution",
+    "title": "Binary Distributions",
     "section": "",
-    "text": "Many distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter \\(\\sigma &gt; 0\\). If \\(\\sigma = 0\\), the probability function will throw an exception.\nThis is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.",
+    "text": "If \\(\\theta \\in [0,1]\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{Bernoulli}(y~|~\\theta) = \\left\\{ \\begin{array}{ll} \\theta &\n\\text{if } y = 1, \\text{ and} \\\\ 1 - \\theta & \\text{if } y = 0.\n\\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli(theta)\nIncrement target log probability density with bernoulli_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_lpmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lupmf(ints y | reals theta) The log Bernoulli probability mass of y given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal bernoulli_cdf(ints y | reals theta) The Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.0\n \n\nreal bernoulli_lcdf(ints y | reals theta) The log of the Bernoulli cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nreal bernoulli_lccdf(ints y | reals theta) The log of the Bernoulli complementary cumulative distribution function of y given chance of success theta\nAvailable since 2.12\n \n\nints bernoulli_rng(reals theta) Generate a Bernoulli variate with chance of success theta or an array of Bernoulli variates given an array of thetas of the same dimensions; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Discrete Distributions",
+      "Binary Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#distributions-prng",
-    "href": "functions-reference/conventions_for_probability_functions.html#distributions-prng",
-    "title": "Conventions for Probability Functions",
+    "objectID": "functions-reference/binary_distributions.html#bernoulli-logit-distribution",
+    "href": "functions-reference/binary_distributions.html#bernoulli-logit-distribution",
+    "title": "Binary Distributions",
     "section": "",
-    "text": "For most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix _rng. For example, the function normal_rng(real, real) accepts two real arguments, an unconstrained location \\(\\mu\\) and positive scale \\(\\sigma &gt; 0\\), and returns an unconstrained pseudorandom value drawn from \\(\\text{Normal}(\\mu,\\sigma)\\). There are also vectorized forms of random number generators which return more than one random variate at a time.\n\n\nUnlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.\n\n\n\nUnlike the probability functions, only some of the PRNG functions are vectorized.",
+    "text": "Stan also supplies a direct parameterization in terms of a logit-transformed chance-of-success parameter. This parameterization is more numerically stable if the chance-of-success parameter is on the logit scale, as with the linear predictor in a logistic regression.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(y \\in \\{0,1\\}\\), \\[\\begin{equation*}\n\\text{BernoulliLogit}(y~|~\\alpha) = \\text{Bernoulli}(y |\n\\text{logit}^{-1}(\\alpha)) = \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha) & \\text{if } y = 1, \\text{ and} \\\\ 1 -\n\\text{logit}^{-1}(\\alpha) &  \\text{if } y = 0. \\end{array} \\right.\n\\end{equation*}\\]\n\n\n\ny ~ bernoulli_logit(alpha)\nIncrement target log probability density with bernoulli_logit_lupmf(y | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal bernoulli_logit_lpmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha)\nAvailable since 2.12\n \n\nreal bernoulli_logit_lupmf(ints y | reals alpha) The log Bernoulli probability mass of y given chance of success inv_logit(alpha) dropping constant additive terms\nAvailable since 2.25\n \n\nR bernoulli_logit_rng(reals alpha) Generate a Bernoulli variate with chance of success \\(\\text{logit}^{-1}(\\alpha)\\); may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Discrete Distributions",
+      "Binary Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#cumulative-distribution-functions",
-    "href": "functions-reference/conventions_for_probability_functions.html#cumulative-distribution-functions",
-    "title": "Conventions for Probability Functions",
+    "objectID": "functions-reference/binary_distributions.html#bernoulli-logit-glm",
+    "href": "functions-reference/binary_distributions.html#bernoulli-logit-glm",
+    "title": "Binary Distributions",
     "section": "",
-    "text": "For most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.\nFor a univariate random variable \\(Y\\) with probability function \\(p_Y(y \\, | \\, \\theta)\\), the cumulative distribution function (CDF) \\(F_Y\\) is defined by \\[\\begin{equation*}\nF_Y(y) \\ = \\ \\text{Pr}[Y \\le y] \\ = \\ \\int_{-\\infty}^y p(y\\, | \\, \\theta) \\ \\text{d}y.\n\\end{equation*}\\] The complementary cumulative distribution function (CCDF) is defined as \\[\\begin{equation*}\n\\text{Pr}[Y &gt; y] \\ = \\ 1 - F_Y(y).\n\\end{equation*}\\] The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly \\(10^{-300}\\)), but not numbers very close to 1 (the closest you can get is roughly \\(1 - 10^{-15}\\)).\nIn Stan, there is a cumulative distribution function for each probability function. For instance, normal_cdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\ \\text{d}y.\n\\end{equation*}\\] There are also log forms of the CDF and CCDF for most univariate distributions. For example, normal_lcdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right)\n\\end{equation*}\\] and normal_lccdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( 1 - \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right).\n\\end{equation*}\\]",
+    "text": "Stan also supplies a single function for a generalized linear model with Bernoulli distribution and logit link function, i.e. a function for a logistic regression. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a Bernoulli distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in {\\{0,1\\}}^n\\), \\[\\begin{align*}\n&\\text{BernoulliLogitGLM}(y~|~x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Bernoulli}(y_i~|~\\text{logit}^{-1}(\\alpha_i + x_i\\cdot\n\\beta))\\\\ &= \\prod_{1\\leq i \\leq n} \\left\\{ \\begin{array}{ll}\n\\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j)\n& \\text{if } y_i = 1, \\text{ and} \\\\ 1 - \\text{logit}^{-1}(\\alpha_i +\n\\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) & \\text{if } y_i = 0.\n\\end{array} \\right. \\end{align*}\\]\n\n\n\ny ~ bernoulli_logit_glm(x, alpha, beta)\nIncrement target log probability density with bernoulli_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.25\n \n\n\n\n\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.23\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\nreal bernoulli_logit_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta).\nAvailable since 2.18\n \n\nreal bernoulli_logit_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Bernoulli probability mass of y given chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.25\n \n\narray[] int bernoulli_logit_glm_rng(matrix x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29\n \n\narray[] int bernoulli_logit_glm_rng(row_vector x, vector alpha, vector beta) Generate an array of Bernoulli variates with chances of success inv_logit(alpha + x * beta); may only be used in transformed data and generated quantities blocks.\nAvailable since 2.29",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Discrete Distributions",
+      "Binary Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/conventions_for_probability_functions.html#vectorization",
-    "href": "functions-reference/conventions_for_probability_functions.html#vectorization",
-    "title": "Conventions for Probability Functions",
+    "objectID": "functions-reference/bounded_discrete_distributions.html",
+    "href": "functions-reference/bounded_discrete_distributions.html",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "Stan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section vectorized PRNG functions for more details.\nIn all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.\nStan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.\nStan also overloads some scalar functions, such as log and exp, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.\n\n\n\n\nThe normal probability function is specified with the signature\n normal_lpdf(reals | reals, reals);\nThe pseudotype reals is used to indicate that an argument position may be vectorized. Argument positions declared as reals may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use normal_lpdf(row_vector | vector, real) as long as the vector and row vector have the same size.\n\n\n\nThe multivariate normal distribution accepting vector or array of vector arguments is written as\n multi_normal_lpdf(vectors | vectors, matrix);\nThese arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.\n\n\n\nThe pseudotype ints is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.\n\n\n\n\nThe result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely real or int, is repeated. For instance, if y is a vector of size N, mu is a vector of size N, and sigma is a scalar, then\n ll = normal_lpdf(y | mu, sigma);\nis just a more efficient way to write\n ll = 0;\n for (n in 1:N) {\n   ll = ll + normal_lpdf(y[n] | mu[n], sigma);\n }\nWith the same arguments, the vectorized sampling statement\n y ~ normal(mu, sigma);\nhas the same effect on the total log probability as\n for (n in 1:N) {\n   y[n] ~ normal(mu[n], sigma);\n }\n\n\n\nSome PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes reals or ints. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.\n vector[3] mu = // ...\n array[3] real x = normal_rng(mu, 3);\n\n\nIn the case of PRNG functions, arguments marked ints may be integers or integer arrays, whereas arguments marked reals may be integers or reals, integer or real arrays, vectors, or row vectors.\n\n\n\n\n\n\n\npseudotype\nallowable PRNG arguments\n\n\n\n\nints\nint, array[] int\n\n\nreals\nint, array[] int, real, array[] real, vector, row_vector\n\n\n\n\n\n\nIn general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the normal_rng function may be called with one vector argument and one real array argument as long as they have the same number of elements.\n vector[3] mu = // ...\n array[3] real sigma = // ...\n array[3] real x = normal_rng(mu, sigma);\n\n\n\nThe result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (array[] real) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return ints and continuous distributions return reals, each of appropriate size. The symbol R denotes such a return type.",
+    "text": "Bounded discrete probability functions have support on \\(\\{ 0, \\ldots,\nN \\}\\) for some upper bound \\(N\\).\n\n\n\n\nSuppose \\(N \\in \\mathbb{N}\\) and \\(\\theta \\in [0,1]\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). \\[\\begin{equation*} \\text{Binomial}(n~|~N,\\theta) = \\binom{N}{n}\n\\theta^n (1 - \\theta)^{N - n}. \\end{equation*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{Binomial}(n~|~N,\\theta) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt] & & {\n} + n \\log \\theta + (N - n) \\log (1 - \\theta), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\theta} \\log \\text{Binomial}(n~|~N,\\theta)\n= \\frac{n}{\\theta} - \\frac{N - n}{1 - \\theta} \\end{equation*}\\]\n\n\n\nn ~ binomial(N, theta)\nIncrement target log probability density with binomial_lupmf(n | N, theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_lpmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lupmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal binomial_cdf(ints n | ints N, reals theta) The binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.0\n \n\nreal binomial_lcdf(ints n | ints N, reals theta) The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lccdf(ints n | ints N, reals theta) The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nR binomial_rng(ints N, reals theta) Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nStan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{eqnarray*}\n\\text{BinomialLogit}(n~|~N,\\alpha) & = &\n\\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha)) \\\\[6pt] & = &\n\\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha) \\right)^{n}  \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right)^{N - n}.  \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{BinomialLogit}(n~|~N,\\alpha) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt]   & &\n{ } + n \\log \\text{logit}^{-1}(\\alpha) + (N - n) \\log \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\alpha} \\log\n\\text{BinomialLogit}(n~|~N,\\alpha) =\n\\frac{n}{\\text{logit}^{-1}(-\\alpha)} - \\frac{N -\nn}{\\text{logit}^{-1}(\\alpha)} \\end{equation*}\\]\n\n\n\nn ~ binomial_logit(N, alpha)\nIncrement target log probability density with binomial_logit_lupmf(n | N, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_logit_lpmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha\nAvailable since 2.12\n \n\nreal binomial_logit_lupmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms\nAvailable since 2.25\n\n\n\n\nStan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta \\in \\mathbb{R}^m\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{align*}\n  &\\text{BinomialLogitGLM}(n~|~N, x, \\alpha, \\beta) = \\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha_i + x_i \\cdot \\beta)) \\\\\n  &= \\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{n}  \\left( 1 - \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{N - n}.\n\\end{align*}\\]\n\n\n\nn ~ binomial_logit_glm(N, x, alpha, beta)\nIncrement target log probability density with binomial_logit_glm_lupmf(n | N, x, alpha, beta).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n\n\n\n\n\n\nIf \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in\n\\mathbb{R}^+\\), then for \\(n \\in {0,\\ldots,N}\\), \\[\\begin{equation*}\n\\text{BetaBinomial}(n~|~N,\\alpha,\\beta) = \\binom{N}{n}\n\\frac{\\mathrm{B}(n+\\alpha, N -n +   \\beta)}{\\mathrm{B}(\\alpha,\\beta)},\n\\end{equation*}\\] where the beta function \\(\\mathrm{B}(u,v)\\) is defined for \\(u \\in\n\\mathbb{R}^+\\) and \\(v \\in \\mathbb{R}^+\\) by \\[\\begin{equation*} \\mathrm{B}(u,v) =\n\\frac{\\Gamma(u) \\ \\Gamma(v)}{\\Gamma(u + v)}. \\end{equation*}\\]\n\n\n\nn ~ beta_binomial(N, alpha, beta)\nIncrement target log probability density with beta_binomial_lupmf(n | N, alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_binomial_lpmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lupmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_binomial_cdf(ints n | ints N, reals alpha, reals beta) The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.0\n \n\nreal beta_binomial_lcdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lccdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nR beta_binomial_rng(ints N, reals alpha, reals beta) Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(a \\in \\mathbb{N}\\), \\(b \\in \\mathbb{N}\\), and \\(N \\in\n\\{0,\\ldots,a+b\\}\\), then for \\(n \\in \\{\\max(0,N-b),\\ldots,\\min(a,N)\\}\\), \\[\\begin{equation*} \\text{Hypergeometric}(n~|~N,a,b) = \\frac{\\normalsize{\\binom{a}{n}\n\\binom{b}{N - n}}}      {\\normalsize{\\binom{a + b}{N}}}. \\end{equation*}\\]\n\n\n\nn ~ hypergeometric(N, a, b)\nIncrement target log probability density with hypergeometric_lupmf(n | N, a, b).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal hypergeometric_lpmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b\nAvailable since 2.12\n \n\nreal hypergeometric_lupmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms\nAvailable since 2.25\n \n\nint hypergeometric_rng(int N, int a, int b) Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(N \\in \\mathbb{N}\\), \\(N &gt; 0\\), and if \\(\\theta \\in \\mathbb{R}^N\\) forms an \\(N\\)-simplex (i.e., has nonnegative entries summing to one), then for \\(y \\in \\{1,\\ldots,N\\}\\), \\[\\begin{equation*} \\text{Categorical}(y~|~\\theta) =\n\\theta_y. \\end{equation*}\\] In addition, Stan provides a log-odds scaled categorical distribution, \\[\\begin{equation*} \\text{CategoricalLogit}(y~|~\\beta) =\n\\text{Categorical}(y~|~\\text{softmax}(\\beta)). \\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical(theta)\nIncrement target log probability density with categorical_lupmf(y | theta) dropping constant additive terms.\nAvailable since 2.0\n \n\n\n\n\ny ~ categorical_logit(beta)\nIncrement target log probability density with categorical_logit_lupmf(y | beta).\nAvailable since 2.4\n \n\n\n\n\nAll of the categorical distributions are vectorized so that the outcome y can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal categorical_lpmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.12\n \n\nreal categorical_lupmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.25\n \n\nreal categorical_logit_lpmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta.\nAvailable since 2.12\n \n\nreal categorical_logit_lupmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta dropping constant additive terms.\nAvailable since 2.25\n \n\nint categorical_rng(vector theta) Generate a categorical variate with \\(N\\)-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nint categorical_logit_rng(vector beta) Generate a categorical variate with outcome in range \\(1:N\\) from log-odds vector beta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.16\n\n\n\n\nStan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e. a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.\nNote that the implementation does not put any restrictions on the coefficient matrix \\(\\beta\\). It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the Stan User’s Guide.\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\), \\(N,M,K &gt; 0\\), and if \\(x\\in \\mathbb{R}^{M\\times K}, \\alpha \\in \\mathbb{R}^N, \\beta\\in \\mathbb{R}^{K\\cdot N}\\), then for \\(y \\in \\{1,\\ldots,N\\}^M\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{CategoricalLogitGLM}(y~|~x,\\alpha,\\beta)\n& =  \\prod_{1\\leq i \\leq M}\\text{CategoricalLogit}(y_i~|~\\alpha+x_i\\cdot\\beta)  \\\\[8pt]\n& =  \\prod_{1\\leq i \\leq M}\\text{Categorical}(y_i~|~softmax(\\alpha+x_i\\cdot\\beta)).\n\\end{split}\n\\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical_logit_glm(x, alpha, beta)\nIncrement target log probability density with categorical_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal categorical_logit_glm_lpmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(l, u \\in \\mathbb{Z}\\) are lower and upper bounds (\\(l \\le u\\)), then for any integer \\(y \\in \\{l,\\ldots,u\\}\\), \\[\\begin{equation*} \\text{DiscreteRange}(y ~|~ l, u) =\n\\frac{1}{u - l + 1}. \\end{equation*}\\]\n\n\n\ny ~ discrete_range(l, u)\nIncrement the target log probability density with discrete_range_lupmf(y | l, u) dropping constant additive terms.\nAvailable since 2.26\n \n\n\n\n\nAll of the discrete range distributions are vectorized so that the outcome y and the bounds l, u can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal discrete_range_lpmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\).\nAvailable since 2.26\n \n\nreal discrete_range_lupmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\) dropping constant additive terms.\nAvailable since 2.26\n \n\nreal discrete_range_cdf(ints y | ints l, ints u) The discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lcdf(ints y | ints l, ints u) The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lccdf(ints y | ints l, ints u) The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nints discrete_range_rng(ints l, ints u) Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.26\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedLogistic}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\text{logit}^{-1}(\\eta - c_1)  &  \\text{if } k = 1, \\\\[4pt]\n\\text{logit}^{-1}(\\eta - c_{k-1}) - \\text{logit}^{-1}(\\eta - c_{k})  &\n\\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt] \\text{logit}^{-1}(\\eta -\nc_{K-1}) - 0  &  \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\nk ~ ordered_logistic(eta, c)\nIncrement target log probability density with ordered_logistic_lupmf(k | eta, c).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal ordered_logistic_lpmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_logistic_lupmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nint ordered_logistic_rng(real eta, vector c) Generate an ordered logistic variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n\n\n\n\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\) with \\(N, M &gt; 0\\), \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(x\\in \\mathbb{R}^{N\\times M}, \\beta\\in \\mathbb{R}^M\\), then for \\(y \\in \\{1,\\ldots,K\\}^N\\), \\[\\begin{equation*}\n\\begin{split}\n\\\\\n& \\text{OrderedLogisticGLM}(y~|~x,\\beta,c) \\\\[8pt]\n& = \\prod_{1\\leq i \\leq N}\\text{OrderedLogistic}(y_i~|~x_i\\cdot \\beta,c) \\\\\n& = \\prod_{1\\leq i \\leq N}  \\left\\{ \\begin{array}{ll}\n1 - \\text{logit}^{-1}(x_i\\cdot \\beta - c_1)  &  \\text{if } y = 1, \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{y-1}) - \\text{logit}^{-1}(x_i\\cdot \\beta - c_{y}) & \\text{if } 1 &lt; y &lt; K, \\text{and} \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{K-1}) - 0  &  \\text{if } y = K.\n\\end{array} \\right.\n\\end{split}\n\\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(y=1\\) and \\(y=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\ny ~ ordered_logistic_glm(x, beta, c)\nIncrement target log probability density with ordered_logistic_lupmf(y | x, beta, c).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal ordered_logistic_glm_lpmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedProbit}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\Phi(\\eta - c_1) & \\text{if } k = 1, \\\\[4pt] \\Phi(\\eta - c_{k-1}) -\n\\Phi(\\eta - c_{k})  & \\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt]\n\\Phi(\\eta - c_{K-1}) - 0 & \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 =\n-\\infty\\) and \\(c_K = +\\infty\\) with \\(\\Phi(-\\infty) = 0\\) and \\(\\Phi(\\infty) = 1\\).\n\n\n\nk ~ ordered_probit(eta, c)\nIncrement target log probability density with ordered_probit_lupmf(k | eta, c).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal ordered_probit_lpmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_probit_lupmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nreal ordered_probit_lpmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.\nAvailable since 2.19\n \n\nreal ordered_probit_lupmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.19\n \n\nint ordered_probit_rng(real eta, vector c) Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Conventions for Probability Functions"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/covariance_matrix_distributions.html",
-    "href": "functions-reference/covariance_matrix_distributions.html",
-    "title": "Covariance Matrix Distributions",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#binomial-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "The covariance matrix distributions have support on symmetric, positive-definite \\(K \\times K\\) matrices or their Cholesky factors (square, lower triangular matrices with positive diagonal elements).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{Wishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}}\n\\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{-\\nu/2} \\ \\left| W \\right|^{(\\nu - K - 1)/2}\n\\ \\exp \\! \\left(- \\frac{1}{2} \\ \\text{tr}\\left( S^{-1} W \\right) \\right) \\! ,\n\\end{equation*}\\] where \\(\\text{tr}()\\) is the matrix trace function, and \\(\\Gamma_K()\\) is the multivariate Gamma function, \\[\\begin{equation*}\n\\Gamma_K(x) = \\frac{1}{\\pi^{K(K-1)/4}} \\ \\prod_{k=1}^K \\Gamma \\left( x + \\frac{1 - k}{2} \\right) \\!.\n\\end{equation*}\\]\n\n\n\nW ~ wishart(nu, Sigma)\nIncrement target log probability density with wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix wishart_rng(real nu, matrix Sigma) Generate a Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0\n\n\n\n\nThe Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{WishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{Wishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in \\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is \\[\\begin{equation*}\n\\text{WishartCholesky}(L_W \\mid \\nu,L_S)\n= \\text{Wishart}(L_W L_W^{\\top} \\mid \\nu,L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of a Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks\nAvailable since 2.30\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{InvWishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}} \\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{\\nu/2} \\ \\left| W \\right|^{-(\\nu + K + 1)/2}\n\\ \\exp \\! \\left( - \\frac{1}{2} \\ \\text{tr}(SW^{-1}) \\right) \\! .\n\\end{equation*}\\]\n\n\n\nW ~ inv_wishart(nu, Sigma)\nIncrement target log probability density with inv_wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal inv_wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix inv_wishart_rng(real nu, matrix Sigma) Generate an inverse Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0\n\n\n\n\nThe Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{InvWishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{InvWishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in\n\\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is \\[\\begin{equation*}\n\\text{InvWishartCholesky}(L_W \\mid \\nu,L_S) =\n\\text{InvWishart}(L_WL_W^{\\top} \\mid \\nu, L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal inv_wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal inv_wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix inv_wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
+    "text": "Suppose \\(N \\in \\mathbb{N}\\) and \\(\\theta \\in [0,1]\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). \\[\\begin{equation*} \\text{Binomial}(n~|~N,\\theta) = \\binom{N}{n}\n\\theta^n (1 - \\theta)^{N - n}. \\end{equation*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{Binomial}(n~|~N,\\theta) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt] & & {\n} + n \\log \\theta + (N - n) \\log (1 - \\theta), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\theta} \\log \\text{Binomial}(n~|~N,\\theta)\n= \\frac{n}{\\theta} - \\frac{N - n}{1 - \\theta} \\end{equation*}\\]\n\n\n\nn ~ binomial(N, theta)\nIncrement target log probability density with binomial_lupmf(n | N, theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_lpmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lupmf(ints n | ints N, reals theta) The log binomial probability mass of n successes in N trials given chance of success theta dropping constant additive terms\nAvailable since 2.25\n \n\nreal binomial_cdf(ints n | ints N, reals theta) The binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.0\n \n\nreal binomial_lcdf(ints n | ints N, reals theta) The log of the binomial cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nreal binomial_lccdf(ints n | ints N, reals theta) The log of the binomial complementary cumulative distribution function of n successes in N trials given chance of success theta\nAvailable since 2.12\n \n\nR binomial_rng(ints N, reals theta) Generate a binomial variate with N trials and chance of success theta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Covariance Matrix Distributions"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/covariance_matrix_distributions.html#wishart-distribution",
-    "href": "functions-reference/covariance_matrix_distributions.html#wishart-distribution",
-    "title": "Covariance Matrix Distributions",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-distribution-logit-parameterization",
+    "href": "functions-reference/bounded_discrete_distributions.html#binomial-distribution-logit-parameterization",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{Wishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}}\n\\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{-\\nu/2} \\ \\left| W \\right|^{(\\nu - K - 1)/2}\n\\ \\exp \\! \\left(- \\frac{1}{2} \\ \\text{tr}\\left( S^{-1} W \\right) \\right) \\! ,\n\\end{equation*}\\] where \\(\\text{tr}()\\) is the matrix trace function, and \\(\\Gamma_K()\\) is the multivariate Gamma function, \\[\\begin{equation*}\n\\Gamma_K(x) = \\frac{1}{\\pi^{K(K-1)/4}} \\ \\prod_{k=1}^K \\Gamma \\left( x + \\frac{1 - k}{2} \\right) \\!.\n\\end{equation*}\\]\n\n\n\nW ~ wishart(nu, Sigma)\nIncrement target log probability density with wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix wishart_rng(real nu, matrix Sigma) Generate a Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0",
+    "text": "Stan also provides a version of the binomial probability mass function distribution with the chance of success parameterized on the unconstrained logistic scale.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{eqnarray*}\n\\text{BinomialLogit}(n~|~N,\\alpha) & = &\n\\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha)) \\\\[6pt] & = &\n\\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha) \\right)^{n}  \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right)^{N - n}.  \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{eqnarray*} \\log \\text{BinomialLogit}(n~|~N,\\alpha) & = & \\log\n\\Gamma(N+1) - \\log \\Gamma(n + 1) - \\log \\Gamma(N- n + 1) \\\\[4pt]   & &\n{ } + n \\log \\text{logit}^{-1}(\\alpha) + (N - n) \\log \\left( 1 -\n\\text{logit}^{-1}(\\alpha) \\right), \\end{eqnarray*}\\]\n\n\n\n\\[\\begin{equation*} \\frac{\\partial}{\\partial \\alpha} \\log\n\\text{BinomialLogit}(n~|~N,\\alpha) =\n\\frac{n}{\\text{logit}^{-1}(-\\alpha)} - \\frac{N -\nn}{\\text{logit}^{-1}(\\alpha)} \\end{equation*}\\]\n\n\n\nn ~ binomial_logit(N, alpha)\nIncrement target log probability density with binomial_logit_lupmf(n | N, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal binomial_logit_lpmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha\nAvailable since 2.12\n \n\nreal binomial_logit_lupmf(ints n | ints N, reals alpha) The log binomial probability mass of n successes in N trials given logit-scaled chance of success alpha dropping constant additive terms\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Covariance Matrix Distributions"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/covariance_matrix_distributions.html#wishart-cholesky-distribution",
-    "href": "functions-reference/covariance_matrix_distributions.html#wishart-cholesky-distribution",
-    "title": "Covariance Matrix Distributions",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#binomial-logit-glm",
+    "href": "functions-reference/bounded_discrete_distributions.html#binomial-logit-glm",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "The Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{WishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{Wishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in \\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is \\[\\begin{equation*}\n\\text{WishartCholesky}(L_W \\mid \\nu,L_S)\n= \\text{Wishart}(L_W L_W^{\\top} \\mid \\nu,L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of a Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks\nAvailable since 2.30",
+    "text": "Stan also supplies a single function for a generalized linear model with binomial distribution and logit link function, i.e., a function for logistic regression with aggregated outcomes. This provides a more efficient implementation of logistic regression than a manually written regression in terms of a binomial distribution and matrix multiplication.\n\n\nSuppose \\(N \\in \\mathbb{N}\\), \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta \\in \\mathbb{R}^m\\), and \\(n \\in\n\\{0,\\ldots,N\\}\\). Then \\[\\begin{align*}\n  &\\text{BinomialLogitGLM}(n~|~N, x, \\alpha, \\beta) = \\text{Binomial}(n~|~N,\\text{logit}^{-1}(\\alpha_i + x_i \\cdot \\beta)) \\\\\n  &= \\binom{N}{n} \\left( \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{n}  \\left( 1 - \\text{logit}^{-1}(\\alpha_i + \\sum_{1\\leq j\\leq m}x_{ij}\\cdot \\beta_j) \\right)^{N - n}.\n\\end{align*}\\]\n\n\n\nn ~ binomial_logit_glm(N, x, alpha, beta)\nIncrement target log probability density with binomial_logit_glm_lupmf(n | N, x, alpha, beta).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(int n | int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, real alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lpmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta).\nAvailable since 2.34\n \n\nreal binomial_logit_glm_lupmf(array[] int n | array[] int N, matrix x, vector alpha, vector beta) The log binomial probability mass of n given N trials and chance of success inv_logit(alpha + x * beta) dropping constant additive terms.\nAvailable since 2.34",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Covariance Matrix Distributions"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-distribution",
-    "href": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-distribution",
-    "title": "Covariance Matrix Distributions",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#beta-binomial-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#beta-binomial-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{InvWishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}} \\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{\\nu/2} \\ \\left| W \\right|^{-(\\nu + K + 1)/2}\n\\ \\exp \\! \\left( - \\frac{1}{2} \\ \\text{tr}(SW^{-1}) \\right) \\! .\n\\end{equation*}\\]\n\n\n\nW ~ inv_wishart(nu, Sigma)\nIncrement target log probability density with inv_wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal inv_wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix inv_wishart_rng(real nu, matrix Sigma) Generate an inverse Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0",
+    "text": "If \\(N \\in \\mathbb{N}\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in\n\\mathbb{R}^+\\), then for \\(n \\in {0,\\ldots,N}\\), \\[\\begin{equation*}\n\\text{BetaBinomial}(n~|~N,\\alpha,\\beta) = \\binom{N}{n}\n\\frac{\\mathrm{B}(n+\\alpha, N -n +   \\beta)}{\\mathrm{B}(\\alpha,\\beta)},\n\\end{equation*}\\] where the beta function \\(\\mathrm{B}(u,v)\\) is defined for \\(u \\in\n\\mathbb{R}^+\\) and \\(v \\in \\mathbb{R}^+\\) by \\[\\begin{equation*} \\mathrm{B}(u,v) =\n\\frac{\\Gamma(u) \\ \\Gamma(v)}{\\Gamma(u + v)}. \\end{equation*}\\]\n\n\n\nn ~ beta_binomial(N, alpha, beta)\nIncrement target log probability density with beta_binomial_lupmf(n | N, alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal beta_binomial_lpmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lupmf(ints n | ints N, reals alpha, reals beta) The log beta-binomial probability mass of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal beta_binomial_cdf(ints n | ints N, reals alpha, reals beta) The beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.0\n \n\nreal beta_binomial_lcdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nreal beta_binomial_lccdf(ints n | ints N, reals alpha, reals beta) The log of the beta-binomial complementary cumulative distribution function of n successes in N trials given prior success count (plus one) of alpha and prior failure count (plus one) of beta\nAvailable since 2.12\n \n\nR beta_binomial_rng(ints N, reals alpha, reals beta) Generate a beta-binomial variate with N trials, prior success count (plus one) of alpha, and prior failure count (plus one) of beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Covariance Matrix Distributions"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-cholesky-distribution",
-    "href": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-cholesky-distribution",
-    "title": "Covariance Matrix Distributions",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#hypergeometric-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#hypergeometric-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "The Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{InvWishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{InvWishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in\n\\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is \\[\\begin{equation*}\n\\text{InvWishartCholesky}(L_W \\mid \\nu,L_S) =\n\\text{InvWishart}(L_WL_W^{\\top} \\mid \\nu, L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal inv_wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal inv_wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix inv_wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
+    "text": "If \\(a \\in \\mathbb{N}\\), \\(b \\in \\mathbb{N}\\), and \\(N \\in\n\\{0,\\ldots,a+b\\}\\), then for \\(n \\in \\{\\max(0,N-b),\\ldots,\\min(a,N)\\}\\), \\[\\begin{equation*} \\text{Hypergeometric}(n~|~N,a,b) = \\frac{\\normalsize{\\binom{a}{n}\n\\binom{b}{N - n}}}      {\\normalsize{\\binom{a + b}{N}}}. \\end{equation*}\\]\n\n\n\nn ~ hypergeometric(N, a, b)\nIncrement target log probability density with hypergeometric_lupmf(n | N, a, b).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal hypergeometric_lpmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b\nAvailable since 2.12\n \n\nreal hypergeometric_lupmf(int n | int N, int a, int b) The log hypergeometric probability mass of n successes in N trials given total success count of a and total failure count of b dropping constant additive terms\nAvailable since 2.25\n \n\nint hypergeometric_rng(int N, int a, int b) Generate a hypergeometric variate with N trials, total success count of a, and total failure count of b; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Covariance Matrix Distributions"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#categorical-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#categorical-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "The unbounded vector probability distributions have support on all of \\(\\mathbb{R}^K\\) for some fixed \\(K\\).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormal}(y|\\mu,\\Sigma) =\n\\frac{1}{\\left( 2 \\pi \\right)^{K/2}} \\ \\frac{1}{\\sqrt{|\\Sigma|}}\n\\  \\exp \\! \\left( \\! - \\frac{1}{2} (y - \\mu)^{\\top} \\, \\Sigma^{-1} \\, (y - \\mu) \\right) \\! ,\n\\end{equation*}\\] where \\(|\\Sigma|\\) is the absolute determinant of \\(\\Sigma\\).\n\n\n\ny ~ multi_normal(mu, Sigma)\nIncrement target log probability density with multi_normal_lupdf(y | mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\nThe multivariate normal probability function is overloaded to allow the variate vector \\(y\\) and location vector \\(\\mu\\) to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal multi_normal_lpdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\nAlthough there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call multi_normal_cholesky_rng; see section multi-variate normal, cholesky parameterization.\n \n\nvector multi_normal_rng(vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_normal_rng(row_vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(row_vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Omega \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\text{MultiNormalPrecision}(y|\\mu,\\Omega)\n= \\text{MultiNormal}(y|\\mu,\\Omega^{-1}) \\end{equation*}\\]\n\n\n\ny ~ multi_normal_prec(mu, Omega)\nIncrement target log probability density with multi_normal_prec_lupdf(y | mu, Omega).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_normal_prec_lpdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L \\in\n\\mathbb{R}^{K \\times K}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormalCholesky}(y|\\mu,L) =\n\\text{MultiNormal}(y|\\mu,LL^{\\top}). \\end{equation*}\\] If \\(L\\) is lower triangular and \\(LL^{top}\\) is a \\(K \\times K\\) positive definite matrix, then \\(L_{k,k}\\) must be strictly positive for \\(k \\in 1{:}K\\). If an \\(L\\) is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.\n\n\n\ny ~ multi_normal_cholesky(mu, L)\nIncrement target log probability density with multi_normal_cholesky_lupdf(y | mu, L).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_normal_cholesky_lpdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_normal_cholesky_rng(vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.3\n \n\nvector multi_normal_cholesky_rng(row_vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(row_vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(K,N \\in \\mathbb{N}\\), \\(\\Sigma \\in \\mathbb{R}^{N \\times N}\\) is symmetric, positive definite kernel matrix and \\(w \\in \\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y \\in \\mathbb{R}^{K\n\\times N}\\), \\[\\begin{equation*} \\text{MultiGP}(y|\\Sigma,w) = \\prod_{i=1}^{K}\n\\text{MultiNormal}(y_i|0,w_i^{-1} \\Sigma), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp(Sigma, w)\nIncrement target log probability density with multi_gp_lupdf(y | Sigma, w).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_gp_lpdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_lupdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K,N \\in \\mathbb{N}\\), \\(L \\in \\mathbb{R}^{N \\times N}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite kernel matrix (implying \\(L_{n,n} &gt; 0\\) for \\(n \\in 1{:}N\\)), and \\(w \\in\n\\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y\n\\in \\mathbb{R}^{K \\times N}\\), \\[\\begin{equation*} \\text{MultiGPCholesky}(y \\, | \\ L,w)\n= \\prod_{i=1}^{K} \\text{MultiNormal}(y_i|0,w_i^{-1} LL^{\\top}), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than \\(\\text{MultiGP}()\\). Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp_cholesky(L, w)\nIncrement target log probability density with multi_gp_cholesky_lupdf(y | L, w).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal multi_gp_cholesky_lpdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_cholesky_lupdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in \\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\begin{array}{l}\n\\text{MultiStudentT}(y\\,|\\,\\nu,\\,\\mu,\\,\\Sigma) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\left| \\Sigma\n\\right|}} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\n\\Sigma^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array} \\end{equation*}\\]\n\n\n\ny ~ multi_student_t(nu, mu, Sigma)\nIncrement target log probability density with multi_student_t_lupdf(y | nu, mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_student_t_lpdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_student_t_rng(real nu, vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_student_t_rng(real nu, row_vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, row_vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nLet \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L\\) a \\(K \\times K\\) lower-triangular matrix with strictly positive, finite diagonal then \\[\\begin{equation*}\n\\begin{array}{l}\n\\text{MultiStudentTCholesky}(y\\,\\mid \\nu,\\,\\mu,\\,L) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\left| L\n\\right|} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\nL^{-T}L^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array}\n\\end{equation*}\\]\n\n\n\ny ~ multi_student_t_cholesky(nu, mu, L)\nIncrement target log probability density with multi_student_t_cholesky_lupdf(y | nu, mu, L).\nAvailable since 2.30\n \n\n\n\n\n \n\nreal multi_student_t_cholesky_lpdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or array of vectors y given degrees of freedom nu, location vector or array of vectors mu, and Cholesky factor of the scale matrix L. For a definition of the arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nreal multi_student_t_cholesky_lupdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or vector array y given degrees of freedom nu, location vector or vector array mu, and Cholesky factor of the scale matrix L, dropping constant additive terms. For a definition of arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nvector multi_student_t_cholesky_rng(real nu, vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] row_vector mu, matrix L) Generate an array of multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n\n\n\n\nA Gaussian Dynamic Linear model is defined as follows, For \\(t \\in 1,\n\\dots, T\\), \\[\\begin{equation*}   \\begin{aligned}[t]     y_{t} &\\sim N(F' \\theta_{t}, V)\n\\\\     \\theta_{t} &\\sim N(G \\theta_{t - 1}, W) \\\\     \\theta_{0} &\\sim\nN(m_{0}, C_{0})   \\end{aligned} \\end{equation*}\\] where \\(y\\) is \\(n \\times T\\) matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (\\(p(y | F, G, V, W, m_{0}, C_{0})\\)). This log-density is a system that is calculated using the Kalman Filter. If \\(V\\) is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used (Durbin and Koopman 2001, sec. 6.4).\n\n\ny ~ gaussian_dlm_obs(F, G, V, W, m0, C0)\nIncrement target log probability density with gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0).\nAvailable since 2.0\n \n\n\n\n\nThe following two functions differ in the type of their V, the first taking a full observation covariance matrix V and the second a vector V representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25",
+    "text": "If \\(N \\in \\mathbb{N}\\), \\(N &gt; 0\\), and if \\(\\theta \\in \\mathbb{R}^N\\) forms an \\(N\\)-simplex (i.e., has nonnegative entries summing to one), then for \\(y \\in \\{1,\\ldots,N\\}\\), \\[\\begin{equation*} \\text{Categorical}(y~|~\\theta) =\n\\theta_y. \\end{equation*}\\] In addition, Stan provides a log-odds scaled categorical distribution, \\[\\begin{equation*} \\text{CategoricalLogit}(y~|~\\beta) =\n\\text{Categorical}(y~|~\\text{softmax}(\\beta)). \\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical(theta)\nIncrement target log probability density with categorical_lupmf(y | theta) dropping constant additive terms.\nAvailable since 2.0\n \n\n\n\n\ny ~ categorical_logit(beta)\nIncrement target log probability density with categorical_logit_lupmf(y | beta).\nAvailable since 2.4\n \n\n\n\n\nAll of the categorical distributions are vectorized so that the outcome y can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal categorical_lpmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.12\n \n\nreal categorical_lupmf(ints y | vector theta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given \\(N\\)-vector of outcome probabilities theta dropping constant additive terms. The parameter theta must have non-negative entries that sum to one, but it need not be a variable declared as a simplex.\nAvailable since 2.25\n \n\nreal categorical_logit_lpmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta.\nAvailable since 2.12\n \n\nreal categorical_logit_lupmf(ints y | vector beta) The log categorical probability mass function with outcome(s) y in \\(1:N\\) given log-odds of outcomes beta dropping constant additive terms.\nAvailable since 2.25\n \n\nint categorical_rng(vector theta) Generate a categorical variate with \\(N\\)-simplex distribution parameter theta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nint categorical_logit_rng(vector beta) Generate a categorical variate with outcome in range \\(1:N\\) from log-odds vector beta; may only be used in transformed data and generated quantities blocks\nAvailable since 2.16",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#categorical-logit-glm",
+    "href": "functions-reference/bounded_discrete_distributions.html#categorical-logit-glm",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormal}(y|\\mu,\\Sigma) =\n\\frac{1}{\\left( 2 \\pi \\right)^{K/2}} \\ \\frac{1}{\\sqrt{|\\Sigma|}}\n\\  \\exp \\! \\left( \\! - \\frac{1}{2} (y - \\mu)^{\\top} \\, \\Sigma^{-1} \\, (y - \\mu) \\right) \\! ,\n\\end{equation*}\\] where \\(|\\Sigma|\\) is the absolute determinant of \\(\\Sigma\\).\n\n\n\ny ~ multi_normal(mu, Sigma)\nIncrement target log probability density with multi_normal_lupdf(y | mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\nThe multivariate normal probability function is overloaded to allow the variate vector \\(y\\) and location vector \\(\\mu\\) to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal multi_normal_lpdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\nAlthough there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call multi_normal_cholesky_rng; see section multi-variate normal, cholesky parameterization.\n \n\nvector multi_normal_rng(vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_normal_rng(row_vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(row_vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "text": "Stan also supplies a single function for a generalized linear model with categorical distribution and logit link function, i.e. a function for a softmax regression. This provides a more efficient implementation of softmax regression than a manually written regression in terms of a categorical distribution and matrix multiplication.\nNote that the implementation does not put any restrictions on the coefficient matrix \\(\\beta\\). It is up to the user to use a reference category, a suitable prior or some other means of identifiability. See Multi-logit in the Stan User’s Guide.\n\n\nIf \\(N,M,K \\in \\mathbb{N}\\), \\(N,M,K &gt; 0\\), and if \\(x\\in \\mathbb{R}^{M\\times K}, \\alpha \\in \\mathbb{R}^N, \\beta\\in \\mathbb{R}^{K\\cdot N}\\), then for \\(y \\in \\{1,\\ldots,N\\}^M\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{CategoricalLogitGLM}(y~|~x,\\alpha,\\beta)\n& =  \\prod_{1\\leq i \\leq M}\\text{CategoricalLogit}(y_i~|~\\alpha+x_i\\cdot\\beta)  \\\\[8pt]\n& =  \\prod_{1\\leq i \\leq M}\\text{Categorical}(y_i~|~softmax(\\alpha+x_i\\cdot\\beta)).\n\\end{split}\n\\end{equation*}\\] See the definition of softmax for the definition of the softmax function.\n\n\n\ny ~ categorical_logit_glm(x, alpha, beta)\nIncrement target log probability density with categorical_logit_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal categorical_logit_glm_lpmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcome y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | row_vector x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal categorical_logit_glm_lpmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta.\nAvailable since 2.23\n \n\nreal categorical_logit_glm_lupmf(array[] int y | matrix x, vector alpha, matrix beta) The log categorical probability mass function with outcomes y in \\(1:N\\) given \\(N\\)-vector of log-odds of outcomes alpha + x * beta dropping constant additive terms.\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution-precision-parameterization",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution-precision-parameterization",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#discrete-range-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#discrete-range-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Omega \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\text{MultiNormalPrecision}(y|\\mu,\\Omega)\n= \\text{MultiNormal}(y|\\mu,\\Omega^{-1}) \\end{equation*}\\]\n\n\n\ny ~ multi_normal_prec(mu, Omega)\nIncrement target log probability density with multi_normal_prec_lupdf(y | mu, Omega).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_normal_prec_lpdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25",
+    "text": "If \\(l, u \\in \\mathbb{Z}\\) are lower and upper bounds (\\(l \\le u\\)), then for any integer \\(y \\in \\{l,\\ldots,u\\}\\), \\[\\begin{equation*} \\text{DiscreteRange}(y ~|~ l, u) =\n\\frac{1}{u - l + 1}. \\end{equation*}\\]\n\n\n\ny ~ discrete_range(l, u)\nIncrement the target log probability density with discrete_range_lupmf(y | l, u) dropping constant additive terms.\nAvailable since 2.26\n \n\n\n\n\nAll of the discrete range distributions are vectorized so that the outcome y and the bounds l, u can be a single integer (type int) or an array of integers (type array[] int).\n \n\nreal discrete_range_lpmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\).\nAvailable since 2.26\n \n\nreal discrete_range_lupmf(ints y | ints l, ints u) The log probability mass function with outcome(s) y in \\(l:u\\) dropping constant additive terms.\nAvailable since 2.26\n \n\nreal discrete_range_cdf(ints y | ints l, ints u) The discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lcdf(ints y | ints l, ints u) The log of the discrete range cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nreal discrete_range_lccdf(ints y | ints l, ints u) The log of the discrete range complementary cumulative distribution function for the given y, lower and upper bounds.\nAvailable since 2.26\n \n\nints discrete_range_rng(ints l, ints u) Generate a discrete variate between the given lower and upper bounds; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.26",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multi-normal-cholesky-fun",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multi-normal-cholesky-fun",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L \\in\n\\mathbb{R}^{K \\times K}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormalCholesky}(y|\\mu,L) =\n\\text{MultiNormal}(y|\\mu,LL^{\\top}). \\end{equation*}\\] If \\(L\\) is lower triangular and \\(LL^{top}\\) is a \\(K \\times K\\) positive definite matrix, then \\(L_{k,k}\\) must be strictly positive for \\(k \\in 1{:}K\\). If an \\(L\\) is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.\n\n\n\ny ~ multi_normal_cholesky(mu, L)\nIncrement target log probability density with multi_normal_cholesky_lupdf(y | mu, L).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_normal_cholesky_lpdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_normal_cholesky_rng(vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.3\n \n\nvector multi_normal_cholesky_rng(row_vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(row_vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
+    "text": "If \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedLogistic}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\text{logit}^{-1}(\\eta - c_1)  &  \\text{if } k = 1, \\\\[4pt]\n\\text{logit}^{-1}(\\eta - c_{k-1}) - \\text{logit}^{-1}(\\eta - c_{k})  &\n\\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt] \\text{logit}^{-1}(\\eta -\nc_{K-1}) - 0  &  \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\nk ~ ordered_logistic(eta, c)\nIncrement target log probability density with ordered_logistic_lupmf(k | eta, c).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal ordered_logistic_lpmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_logistic_lupmf(ints k | vector eta, vectors c) The log ordered logistic probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nint ordered_logistic_rng(real eta, vector c) Generate an ordered logistic variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-generalized-linear-model-ordinal-regression",
+    "href": "functions-reference/bounded_discrete_distributions.html#ordered-logistic-generalized-linear-model-ordinal-regression",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K,N \\in \\mathbb{N}\\), \\(\\Sigma \\in \\mathbb{R}^{N \\times N}\\) is symmetric, positive definite kernel matrix and \\(w \\in \\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y \\in \\mathbb{R}^{K\n\\times N}\\), \\[\\begin{equation*} \\text{MultiGP}(y|\\Sigma,w) = \\prod_{i=1}^{K}\n\\text{MultiNormal}(y_i|0,w_i^{-1} \\Sigma), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp(Sigma, w)\nIncrement target log probability density with multi_gp_lupdf(y | Sigma, w).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_gp_lpdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_lupdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms\nAvailable since 2.25",
+    "text": "If \\(N,M,K \\in \\mathbb{N}\\) with \\(N, M &gt; 0\\), \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(x\\in \\mathbb{R}^{N\\times M}, \\beta\\in \\mathbb{R}^M\\), then for \\(y \\in \\{1,\\ldots,K\\}^N\\), \\[\\begin{equation*}\n\\begin{split}\n\\\\\n& \\text{OrderedLogisticGLM}(y~|~x,\\beta,c) \\\\[8pt]\n& = \\prod_{1\\leq i \\leq N}\\text{OrderedLogistic}(y_i~|~x_i\\cdot \\beta,c) \\\\\n& = \\prod_{1\\leq i \\leq N}  \\left\\{ \\begin{array}{ll}\n1 - \\text{logit}^{-1}(x_i\\cdot \\beta - c_1)  &  \\text{if } y = 1, \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{y-1}) - \\text{logit}^{-1}(x_i\\cdot \\beta - c_{y}) & \\text{if } 1 &lt; y &lt; K, \\text{and} \\\\[4pt]\n\\text{logit}^{-1}(x_i\\cdot \\beta - c_{K-1}) - 0  &  \\text{if } y = K.\n\\end{array} \\right.\n\\end{split}\n\\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(y=1\\) and \\(y=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 = -\\infty\\) and \\(c_K = +\\infty\\) with \\(\\text{logit}^{-1}(-\\infty) = 0\\) and \\(\\text{logit}^{-1}(\\infty) = 1\\).\n\n\n\ny ~ ordered_logistic_glm(x, beta, c)\nIncrement target log probability density with ordered_logistic_lupmf(y | x, beta, c).\nAvailable since 2.23\n \n\n\n\n\n \n\nreal ordered_logistic_glm_lpmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | row_vector x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25\n \n\nreal ordered_logistic_glm_lpmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c. The cutpoints c must be ordered.\nAvailable since 2.23\n \n\nreal ordered_logistic_glm_lupmf(array[] int y | matrix x, vector beta, vector c) The log ordered logistic probability mass of y, given linear predictors x * beta, and cutpoints c dropping constant additive terms. The cutpoints c must be ordered.\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution-cholesky-parameterization",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution-cholesky-parameterization",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/bounded_discrete_distributions.html#ordered-probit-distribution",
+    "href": "functions-reference/bounded_discrete_distributions.html#ordered-probit-distribution",
+    "title": "Bounded Discrete Distributions",
     "section": "",
-    "text": "If \\(K,N \\in \\mathbb{N}\\), \\(L \\in \\mathbb{R}^{N \\times N}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite kernel matrix (implying \\(L_{n,n} &gt; 0\\) for \\(n \\in 1{:}N\\)), and \\(w \\in\n\\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y\n\\in \\mathbb{R}^{K \\times N}\\), \\[\\begin{equation*} \\text{MultiGPCholesky}(y \\, | \\ L,w)\n= \\prod_{i=1}^{K} \\text{MultiNormal}(y_i|0,w_i^{-1} LL^{\\top}), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than \\(\\text{MultiGP}()\\). Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp_cholesky(L, w)\nIncrement target log probability density with multi_gp_cholesky_lupdf(y | L, w).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal multi_gp_cholesky_lpdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_cholesky_lupdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms\nAvailable since 2.25",
+    "text": "If \\(K \\in \\mathbb{N}\\) with \\(K &gt; 2\\), \\(c \\in \\mathbb{R}^{K-1}\\) such that \\(c_k &lt; c_{k+1}\\) for \\(k \\in \\{1,\\ldots,K-2\\}\\), and \\(\\eta \\in\n\\mathbb{R}\\), then for \\(k \\in \\{1,\\ldots,K\\}\\), \\[\\begin{equation*}\n\\text{OrderedProbit}(k~|~\\eta,c) = \\left\\{ \\begin{array}{ll} 1 -\n\\Phi(\\eta - c_1) & \\text{if } k = 1, \\\\[4pt] \\Phi(\\eta - c_{k-1}) -\n\\Phi(\\eta - c_{k})  & \\text{if } 1 &lt; k &lt; K, \\text{and} \\\\[4pt]\n\\Phi(\\eta - c_{K-1}) - 0 & \\text{if } k = K. \\end{array} \\right. \\end{equation*}\\] The \\(k=K\\) case is written with the redundant subtraction of zero to illustrate the parallelism of the cases; the \\(k=1\\) and \\(k=K\\) edge cases can be subsumed into the general definition by setting \\(c_0 =\n-\\infty\\) and \\(c_K = +\\infty\\) with \\(\\Phi(-\\infty) = 0\\) and \\(\\Phi(\\infty) = 1\\).\n\n\n\nk ~ ordered_probit(eta, c)\nIncrement target log probability density with ordered_probit_lupmf(k | eta, c).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal ordered_probit_lpmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c.\nAvailable since 2.18\n \n\nreal ordered_probit_lupmf(ints k | vector eta, vectors c) The log ordered probit probability mass of k given linear predictors eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.25\n \n\nreal ordered_probit_lpmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c.\nAvailable since 2.19\n \n\nreal ordered_probit_lupmf(ints k | real eta, vectors c) The log ordered probit probability mass of k given linear predictor eta, and cutpoints c dropping constant additive terms.\nAvailable since 2.19\n \n\nint ordered_probit_rng(real eta, vector c) Generate an ordered probit variate with linear predictor eta and cutpoints c; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Discrete Distributions",
+      "Bounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-student-t-distribution",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-student-t-distribution",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/complex-valued_basic_functions.html",
+    "href": "functions-reference/complex-valued_basic_functions.html",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in \\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\begin{array}{l}\n\\text{MultiStudentT}(y\\,|\\,\\nu,\\,\\mu,\\,\\Sigma) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\left| \\Sigma\n\\right|}} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\n\\Sigma^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array} \\end{equation*}\\]\n\n\n\ny ~ multi_student_t(nu, mu, Sigma)\nIncrement target log probability density with multi_student_t_lupdf(y | nu, mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_student_t_lpdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_student_t_rng(real nu, vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_student_t_rng(real nu, row_vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, row_vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks\nAvailable since 2.18",
+    "text": "This chapter describes built-in functions that operate on complex numbers, either as an argument type or a return type. This includes the arithmetic operators generalized to complex numbers.\n\n\nJust as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.\nint n = 5;       // n = 5\nreal x = a;      // x = 5.0\ncomplex z1 = n;  // z = 5.0 + 0.0i\ncomplex z2 = x;  // z = 5.0 + 0.0i\n\n\nFunction arguments of type int or real may be promoted to type complex. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if z is complex, then pow(z, 2) will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where z + 2 will be legal and produce a complex result. Functions such as arg and conj that are only available for complex numbers can accept integer or real arguments, promoting them to complex before applying the function.\n\n\n\n\n\n\nVariables and constants of type complex are constructed from zero, one, or two real numbers.\ncomplex z1 = to_complex();        // z1 = 0.0 + 0.0i\nreal re = -2.9;\ncomplex z2 = to_complex(re);      // z2 = -2.9 + 0.0i\nreal im = 1.3;\ncomplex z3 = to_complex(re, im);  // z3 = -2.9 + 1.3i\n \n\ncomplex to_complex() Return complex number with real part 0.0 and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re) Return complex number with real part re and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re, real im) Return complex number with real part re and imaginary part im.\nAvailable since 2.28\n \n\nZ to_complex(T1 re, T2 im) Vectorized implementation of the to_complex function.\nT1 and T2 can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.\nAvailable since 2.30\n\n\n\nGiven a complex number, its real and imaginary parts can be extracted with the following functions.\n \n\nreal get_real(complex z) Return the real part of the complex number z.\nAvailable since 2.28\n \n\nreal get_imag(complex z) Return the imaginary part of the complex number z.\nAvailable since 2.28\n\n\n\n\nThe arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type complex. If there are two arguments and only one is of type complex, then the other will be promoted to type complex before performing the operation.\n\n\n \n\ncomplex operator+(complex z) Return the complex argument z, \\[\\begin{equation*} +z = z. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex z) Return the negation of the complex argument z, which for \\(z = x + yi\\) is \\[\\begin{equation*} -z = -x - yi. \\end{equation*}\\]\nAvailable since 2.28\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of complex numbers, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex operator+(complex x, complex y) Return the sum of x and y, \\[\\begin{equation*} (x + y) = \\text{operator+}(x, y) = x + y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex x, complex y) Return the difference between x and y, \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x, y) = x - y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator*(complex x, complex y) Return the product of x and y, \\[\\begin{equation*} (x \\, * \\, y) = \\text{operator*}(x, y) = x\n\\times y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator/(complex x, complex y) Return the quotient of x and y, \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator^(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n(x^\\mathrm{\\wedge}y)= \\text{operator}^\\mathrm{\\wedge}(x,y)\n= \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\n\nComplex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional\nz1 == z2\nis equivalent to\nget_real(z1) == get_real(z2) && get_imag(z1) == get_imag(z2)\nAs with other complex functions, if one of the arguments is of type real or int, it will be promoted to type complex before comparison. For example, if z is of type complex, then z == 0 will be true if z has real component equal to 0.0 and complex component equal to 0.0.\nWarning: As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.\n \n\nint operator==(complex x, complex y) Return 1 if x is equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{==}\\, y)\n\\ = \\ \\text{operator==}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x = y$}, \\ \\text{and} \\\\ 0 & \\text{otherwise.}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nint operator!=(complex x, complex y) Return 1 if x is not equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{!=}\\, y)\n\\ = \\ \\text{operator!=}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x \\neq y$}, \\ \\text{and} \\\\ 0 &\n\\text{otherwise.} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\nThe assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.\nAssignment of complex numbers works elementwise. If an expression of type int or real is assigned to a complex number, it will be promoted before assignment as if calling to_complex(), so that the imaginary component is 0.0.\n \n\nvoid operator=(complex x, complex y) y = x; assigns a (copy of) the value of y to x.\nAvailable since 2.28\n \n\nvoid operator+=(complex x, complex y) x += y; is equivalent to x = x + y;.\nAvailable since 2.28\n \n\nvoid operator-=(complex x, complex y) x -= y; is equivalent to x = x - y;.\nAvailable since 2.28\n \n\nvoid operator*=(complex x, complex y) x *= y; is equivalent to x = x * y;.\nAvailable since 2.28\n \n\nvoid operator/=(complex x, complex y) x /= y; is equivalent to x = x / y;.\nAvailable since 2.28\n\n\n\nThe following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.\n \n\nreal abs(complex z) Return the absolute value of z, also known as the modulus or magnitude, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{abs}(z) = \\sqrt{x^2 + y^2}.\n\\end{equation*}\\]\nThis function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a complex_vector[n] input will return a vector[n] output, with each element transformed by the above equation.\nAvailable since 2.28, vectorized in 2.30\n \n\nreal arg(complex z) Return the phase angle (in radians) of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{arg}(z) = \\textrm{atan2}(y, x) = \\textrm{atan}(y / x).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nreal norm(complex z) Return the Euclidean norm of z, which is its absolute value squared, and which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{norm}(z) = \\textrm{abs}^2(z) = x^2 + y^2.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex conj(complex z) Return the complex conjugate of z, which negates the imaginary component, so that if \\(z = x + yi\\), \\[\\begin{equation*}\n\\textrm{conj}(z) = x - yi.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ conj(Z z) Vectorized version of conj. This will apply the conj function to each element of a complex array, vector, or matrix.\nAvailable since 2.31\n \n\ncomplex proj(complex z) Return the projection of z onto the Riemann sphere, which for \\(z = x\n+ yi\\) is \\[\\begin{equation*}\n\\textrm{proj}(z)\n= \\begin{cases}\n    z & \\textrm{if} \\ z \\ \\textrm{is finite, and} \\\\\n    0 + \\textrm{sign}(y)i & \\textrm{otherwise,}\n\\end{cases}\n\\end{equation*}\\] where \\(\\textrm{sign}(y)\\) is -1 if \\(y\\) is negative and 1 otherwise.\nAvailable since 2.28\n \n\ncomplex polar(real r, real theta) Return the complex number with magnitude (absolute value) r and phase angle theta.\nAvailable since 2.28\n\n\n\nThe exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.\n \n\ncomplex exp(complex z) Return the complex natural exponential of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\exp z = \\exp(x) \\textrm{cis}(y) = \\exp(x) (\\cos(y) + i \\sin(y)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log(complex z) Return the complex natural logarithm of z, which for \\(z = \\textrm{polar}(r,\n\\theta)\\) is \\[\\begin{equation*}\n\\log z = \\log r + \\theta i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log10(complex z) Return the complex common logarithm of z, \\[\\begin{equation*}\n\\log_{10} z = \\frac{\\log z}{\\log 10}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex pow(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n\\text{pow}(x,y) = \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.30\n \n\ncomplex sqrt(complex x) Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.\nAvailable since 2.28\n\n\n\nThe standard trigonometric functions are supported for complex numbers.\n \n\ncomplex cos(complex z) Return the complex cosine of z, which is \\[\\begin{equation*}\n\\cos(z)\n= \\textrm{cosh}(z \\, i)\n= \\frac{\\displaystyle \\exp(z \\, i) + \\exp(-z \\, i)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sin(complex z) Return the complex sine of z, \\[\\begin{equation*}\n\\sin(z)\n= -\\textrm{sinh}(z \\, i) \\, i\n= \\frac{\\displaystyle \\exp(z \\, i) - \\exp(-z \\, i)}\n       {\\displaystyle 2 \\, i}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tan(complex z) Return the complex tangent of z, \\[\\begin{equation*}\n\\tan(z)\n= -\\textrm{tanh}(z \\, i) \\, i\n= \\frac{(\\exp(-z \\, i) - \\exp(z \\, i)) \\, i}\n       {\\exp(-z \\, i) + \\exp(z \\, i)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acos(complex z) Return the complex arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acos}(z)\n= \\frac{1}{2} \\pi + \\log (z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asin(complex z) Return the complex arc (inverse) sine of z, \\[\\begin{equation*}\n\\text{asin}(z)\n= -\\log(z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atan(complex z) Return the complex arc (inverse) tangent of z, \\[\\begin{equation*}\n\\text{atan}(z)\n= - \\frac{1}{2} (\\log(1 - z \\, i) - \\log(1 + z \\, i)) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n\n\n\nThe standard hyperbolic trigonometric functions are supported for complex numbers.\n \n\ncomplex cosh(complex z) Return the complex hyperbolic cosine of z, \\[\\begin{equation*}\n\\textrm{cosh}(z)\n= \\frac{\\exp(z) + \\exp(-z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sinh(complex z) Return the complex hyperbolic sine of z, \\[\\begin{equation*}\n\\textrm{sinh}(z)\n= \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tanh(complex z) Return the complex hyperbolic tangent of z, \\[\\begin{equation*}\n\\textrm{tanh}(z)\n\\ = \\ \\frac{\\textrm{sinh}(z)}\n           {\\textrm{cosh}(z)}\n\\ = \\ \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n           {\\displaystyle \\exp(z) + \\exp(-z)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acosh(complex z) Return the complex hyperbolic arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acosh}(z)\n= \\log(z + \\sqrt{(z + 1)(z - 1)}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asinh(complex z) Return the complex hyperbolic arc (inverse) sine of z, \\[\\begin{equation*}\n\\textrm{asinh}(z)\n= \\log(z + \\sqrt{1 + z^2}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atanh(complex z) Return the complex hyperbolic arc (inverse) tangent of z, \\[\\begin{equation*}\n\\textrm{atanh}(z)\n= \\frac{\\log(1 + z) - \\log(1 - z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multi-student-t-cholesky-fun",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#multi-student-t-cholesky-fun",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-assign-promote",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-assign-promote",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "Let \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L\\) a \\(K \\times K\\) lower-triangular matrix with strictly positive, finite diagonal then \\[\\begin{equation*}\n\\begin{array}{l}\n\\text{MultiStudentTCholesky}(y\\,\\mid \\nu,\\,\\mu,\\,L) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\left| L\n\\right|} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\nL^{-T}L^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array}\n\\end{equation*}\\]\n\n\n\ny ~ multi_student_t_cholesky(nu, mu, L)\nIncrement target log probability density with multi_student_t_cholesky_lupdf(y | nu, mu, L).\nAvailable since 2.30\n \n\n\n\n\n \n\nreal multi_student_t_cholesky_lpdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or array of vectors y given degrees of freedom nu, location vector or array of vectors mu, and Cholesky factor of the scale matrix L. For a definition of the arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nreal multi_student_t_cholesky_lupdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or vector array y given degrees of freedom nu, location vector or vector array mu, and Cholesky factor of the scale matrix L, dropping constant additive terms. For a definition of arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nvector multi_student_t_cholesky_rng(real nu, vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] row_vector mu, matrix L) Generate an array of multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
+    "text": "Just as integers may be assigned to real variables, real variables may be assigned to complex numbers, with the result being a zero imaginary component.\nint n = 5;       // n = 5\nreal x = a;      // x = 5.0\ncomplex z1 = n;  // z = 5.0 + 0.0i\ncomplex z2 = x;  // z = 5.0 + 0.0i\n\n\nFunction arguments of type int or real may be promoted to type complex. The complex version of functions in this chapter are only used if one of the arguments is complex. For example, if z is complex, then pow(z, 2) will call the complex version of the power function and the integer 2 will be promoted to a complex number with a real component of 2 and an imaginary component of 0. The same goes for binary operators like addition and subtraction, where z + 2 will be legal and produce a complex result. Functions such as arg and conj that are only available for complex numbers can accept integer or real arguments, promoting them to complex before applying the function.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#gaussian-dynamic-linear-models",
-    "href": "functions-reference/distributions_over_unbounded_vectors.html#gaussian-dynamic-linear-models",
-    "title": "Distributions over Unbounded Vectors",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-constructors",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-constructors",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "A Gaussian Dynamic Linear model is defined as follows, For \\(t \\in 1,\n\\dots, T\\), \\[\\begin{equation*}   \\begin{aligned}[t]     y_{t} &\\sim N(F' \\theta_{t}, V)\n\\\\     \\theta_{t} &\\sim N(G \\theta_{t - 1}, W) \\\\     \\theta_{0} &\\sim\nN(m_{0}, C_{0})   \\end{aligned} \\end{equation*}\\] where \\(y\\) is \\(n \\times T\\) matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (\\(p(y | F, G, V, W, m_{0}, C_{0})\\)). This log-density is a system that is calculated using the Kalman Filter. If \\(V\\) is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used (Durbin and Koopman 2001, sec. 6.4).\n\n\ny ~ gaussian_dlm_obs(F, G, V, W, m0, C0)\nIncrement target log probability density with gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0).\nAvailable since 2.0\n \n\n\n\n\nThe following two functions differ in the type of their V, the first taking a full observation covariance matrix V and the second a vector V representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25",
+    "text": "Variables and constants of type complex are constructed from zero, one, or two real numbers.\ncomplex z1 = to_complex();        // z1 = 0.0 + 0.0i\nreal re = -2.9;\ncomplex z2 = to_complex(re);      // z2 = -2.9 + 0.0i\nreal im = 1.3;\ncomplex z3 = to_complex(re, im);  // z3 = -2.9 + 1.3i\n \n\ncomplex to_complex() Return complex number with real part 0.0 and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re) Return complex number with real part re and imaginary part 0.0.\nAvailable since 2.28\n \n\ncomplex to_complex(real re, real im) Return complex number with real part re and imaginary part im.\nAvailable since 2.28\n \n\nZ to_complex(T1 re, T2 im) Vectorized implementation of the to_complex function.\nT1 and T2 can either be real containers of the same size, or a real container and a real, in which case the real value is used for the corresponding component in all elements of the output.\nAvailable since 2.30\n\n\n\nGiven a complex number, its real and imaginary parts can be extracted with the following functions.\n \n\nreal get_real(complex z) Return the real part of the complex number z.\nAvailable since 2.28\n \n\nreal get_imag(complex z) Return the imaginary part of the complex number z.\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Distributions over Unbounded Vectors"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/hidden_markov_models.html",
-    "href": "functions-reference/hidden_markov_models.html",
-    "title": "Hidden Markov Models",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-arithmetic",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-arithmetic",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "An elementary first-order Hidden Markov model is a probabilistic model over \\(N\\) observations, \\(y_n\\), and \\(N\\) hidden states, \\(x_n\\), which can be fully defined by the conditional distributions \\(p(y_n \\mid x_n, \\phi)\\) and \\(p(x_n \\mid x_{n - 1}, \\phi)\\). Here we make the dependency on additional model parameters, \\(\\phi\\), explicit. When \\(x\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(x\\) out.\nWhen each state \\(x\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can take advantage of the dependency structure to marginalize \\(x\\) and compute \\(p(y \\mid \\phi)\\). We start by defining the conditional observational distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid x_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n  \\Gamma_{ij} = p(x_n = j \\mid x_{n - 1} = i, \\phi).\n\\] Each row defines a probability distribution and must therefore be a simplex (i.e. its components must add to 1). Currently, Stan only supports stationary transitions where a single transition matrix is used for all transitions. Finally we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n  \\rho_k = p(x_0 = k \\mid \\phi).\n\\]\nThe Stan functions that support this type of model are special in that the user does not explicitly pass \\(y\\) and \\(\\phi\\) as arguments. Instead, the user passes \\(\\log \\omega\\), \\(\\Gamma\\), and \\(\\rho\\), which in turn depend on \\(y\\) and \\(\\phi\\).\n\n\n \n\nreal hmm_marginal(matrix log_omega, matrix Gamma, vector rho) Returns the log probability density of \\(y\\), with \\(x_n\\) integrated out at each iteration.\nAvailable since 2.24\nThe arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.\n\nlog_omega: \\(\\log \\omega_{kn} = \\log p(y_n \\mid x_n = k, \\phi)\\), log density of each output,\nGamma: \\(\\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \\phi)\\), the transition matrix,\nrho: \\(\\rho_k = p(x_0 = k \\mid \\phi)\\), the initial state probability.\n\n \n\narray[] int hmm_latent_rng(matrix log_omega, matrix Gamma, vector rho) Returns a length \\(N\\) array of integers over \\(\\{1, ..., K\\}\\), sampled from the joint posterior distribution of the hidden states, \\(p(x \\mid \\phi, y)\\). May be only used in transformed data and generated quantities.\nAvailable since 2.24\n \n\nmatrix hmm_hidden_state_prob(matrix log_omega, matrix Gamma, vector rho) Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a \\(K \\times N\\) matrix.    The \\(n^\\mathrm{th}\\) column is a simplex of probabilities for the \\(n^\\mathrm{th}\\) variable. Moreover, let \\(A\\) be the output. Then \\(A_{ij} = p(x_j = i \\mid \\phi, y)\\).  This function may only be used in transformed data and generated quantities.\nAvailable since 2.24",
+    "text": "The arithmetic operators have the same precedence for complex and real arguments. The complex form of an operator will be selected if at least one of its argument is of type complex. If there are two arguments and only one is of type complex, then the other will be promoted to type complex before performing the operation.\n\n\n \n\ncomplex operator+(complex z) Return the complex argument z, \\[\\begin{equation*} +z = z. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex z) Return the negation of the complex argument z, which for \\(z = x + yi\\) is \\[\\begin{equation*} -z = -x - yi. \\end{equation*}\\]\nAvailable since 2.28\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of complex numbers, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\ncomplex operator+(complex x, complex y) Return the sum of x and y, \\[\\begin{equation*} (x + y) = \\text{operator+}(x, y) = x + y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator-(complex x, complex y) Return the difference between x and y, \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x, y) = x - y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator*(complex x, complex y) Return the product of x and y, \\[\\begin{equation*} (x \\, * \\, y) = \\text{operator*}(x, y) = x\n\\times y. \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator/(complex x, complex y) Return the quotient of x and y, \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex operator^(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n(x^\\mathrm{\\wedge}y)= \\text{operator}^\\mathrm{\\wedge}(x,y)\n= \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Additional Distributions",
-      "Hidden Markov Models"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/hidden_markov_models.html#hmm-stan-functions",
-    "href": "functions-reference/hidden_markov_models.html#hmm-stan-functions",
-    "title": "Hidden Markov Models",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-comparison",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-comparison",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "real hmm_marginal(matrix log_omega, matrix Gamma, vector rho) Returns the log probability density of \\(y\\), with \\(x_n\\) integrated out at each iteration.\nAvailable since 2.24\nThe arguments represent (1) the log density of each output, (2) the transition matrix, and (3) the initial state vector.\n\nlog_omega: \\(\\log \\omega_{kn} = \\log p(y_n \\mid x_n = k, \\phi)\\), log density of each output,\nGamma: \\(\\Gamma_{ij} = p(x_n = j | x_{n - 1} = i, \\phi)\\), the transition matrix,\nrho: \\(\\rho_k = p(x_0 = k \\mid \\phi)\\), the initial state probability.\n\n \n\narray[] int hmm_latent_rng(matrix log_omega, matrix Gamma, vector rho) Returns a length \\(N\\) array of integers over \\(\\{1, ..., K\\}\\), sampled from the joint posterior distribution of the hidden states, \\(p(x \\mid \\phi, y)\\). May be only used in transformed data and generated quantities.\nAvailable since 2.24\n \n\nmatrix hmm_hidden_state_prob(matrix log_omega, matrix Gamma, vector rho) Returns the matrix of marginal posterior probabilities of each hidden state value. This will be a \\(K \\times N\\) matrix.    The \\(n^\\mathrm{th}\\) column is a simplex of probabilities for the \\(n^\\mathrm{th}\\) variable. Moreover, let \\(A\\) be the output. Then \\(A_{ij} = p(x_j = i \\mid \\phi, y)\\).  This function may only be used in transformed data and generated quantities.\nAvailable since 2.24",
+    "text": "Complex numbers are equal if and only if both their real and imaginary components are equal. That is, the conditional\nz1 == z2\nis equivalent to\nget_real(z1) == get_real(z2) && get_imag(z1) == get_imag(z2)\nAs with other complex functions, if one of the arguments is of type real or int, it will be promoted to type complex before comparison. For example, if z is of type complex, then z == 0 will be true if z has real component equal to 0.0 and complex component equal to 0.0.\nWarning: As with real values, it is usually a mistake to compare complex numbers for equality because their parts are implemented using floating-point arithmetic, which suffers from precision errors, rendering algebraically equivalent expressions not equal after evaluation.\n \n\nint operator==(complex x, complex y) Return 1 if x is equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{==}\\, y)\n\\ = \\ \\text{operator==}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x = y$}, \\ \\text{and} \\\\ 0 & \\text{otherwise.}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nint operator!=(complex x, complex y) Return 1 if x is not equal to y and 0 otherwise, \\[\\begin{equation*}\n(x \\,\\text{!=}\\, y)\n\\ = \\ \\text{operator!=}(x,y)\n\\ = \\ \\begin{cases} 1 & \\text{if $x \\neq y$}, \\ \\text{and} \\\\ 0 &\n\\text{otherwise.} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Additional Distributions",
-      "Hidden Markov Models"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/index.html",
-    "href": "functions-reference/index.html",
-    "title": "Stan Functions Reference",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-assignment",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-assignment",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "This is the reference for the functions defined in the Stan math library and available in the Stan programming language.\nFor more information the Stan language and inference engines and how to use Stan for Bayesian inference, see\n\nthe Stan User’s Guide. The Stan user’s guide provides example models and programming techniques for coding statistical models in Stan. It also serves as an example-driven introduction to Bayesian modeling and inference:\nthe Stan Reference Manual. Stan’s modeling language is shared across all of its interfaces. The Stan Language Reference Manual provides a concise definition of the language syntax for all elements in the language together with an overview of the inference algorithms and posterior inference tools.\n\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "text": "The assignment operator only serves as a component in the assignment statement and is thus not technically a function in the Stan language. With that caveat, it is documented here for completeness.\nAssignment of complex numbers works elementwise. If an expression of type int or real is assigned to a complex number, it will be promoted before assignment as if calling to_complex(), so that the imaginary component is 0.0.\n \n\nvoid operator=(complex x, complex y) y = x; assigns a (copy of) the value of y to x.\nAvailable since 2.28\n \n\nvoid operator+=(complex x, complex y) x += y; is equivalent to x = x + y;.\nAvailable since 2.28\n \n\nvoid operator-=(complex x, complex y) x -= y; is equivalent to x = x - y;.\nAvailable since 2.28\n \n\nvoid operator*=(complex x, complex y) x *= y; is equivalent to x = x * y;.\nAvailable since 2.28\n \n\nvoid operator/=(complex x, complex y) x /= y; is equivalent to x = x / y;.\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Stan Functions Reference"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/mathematical_functions.html",
-    "href": "functions-reference/mathematical_functions.html",
-    "title": "Mathematical Functions",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-special",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-special",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "This appendix provides the definition of several mathematical functions used throughout the manual.\n\n\nThe beta function, \\(\\text{B}(a, b)\\), computes the normalizing constant for the beta distribution, and is defined for \\(a &gt; 0\\) and \\(b\n&gt; 0\\) by \\[\\begin{equation*} \\text{B}(a,b) \\ = \\ \\int_0^1 u^{a - 1} (1 - u)^{b - 1} \\,\ndu \\ = \\ \\frac{\\Gamma(a) \\, \\Gamma(b)}{\\Gamma(a+b)} \\, , \\end{equation*}\\] where \\(\\Gamma(x)\\) is the Gamma function.\n\n\n\nThe incomplete beta function, \\(\\text{B}(x; a, b)\\), is defined for \\(x\n\\in [0, 1]\\) and \\(a, b \\geq 0\\) such that \\(a + b \\neq 0\\) by \\[\\begin{equation*}\n\\text{B}(x; \\, a, b) \\ = \\ \\int_0^x u^{a - 1} \\, (1 - u)^{b - 1} \\,\ndu, \\end{equation*}\\] where \\(\\text{B}(a, b)\\) is the beta function defined in appendix. If \\(x = 1\\), the incomplete beta function reduces to the beta function, \\(\\text{B}(1; a, b) = \\text{B}(a, b)\\).\nThe regularized incomplete beta function divides the incomplete beta function by the beta function, \\[\\begin{equation*} I_x(a, b) \\ = \\ \\frac{\\text{B}(x; \\,\na, b)}{B(a, b)} \\, . \\end{equation*}\\]\n\n\n\nThe gamma function, \\(\\Gamma(x)\\), is the generalization of the factorial function to continuous variables, defined so that for positive integers \\(n\\), \\[\\begin{equation*} \\Gamma(n+1) = n! \\end{equation*}\\] Generalizing to all positive numbers and non-integer negative numbers, \\[\\begin{equation*} \\Gamma(x) =\n\\int_0^{\\infty} u^{x - 1} \\exp(-u) \\, du. \\end{equation*}\\]\n\n\n\nThe digamma function \\(\\Psi\\) is the derivative of the \\(\\log \\Gamma\\) function, \\[\\begin{equation*}\n\\Psi(u) \\ = \\ \\frac{d}{d u} \\log \\Gamma(u) \\ = \\ \\frac{1}{\\Gamma(u)} \\ \\frac{d}{d u} \\Gamma(u).\n\\end{equation*}\\]",
+    "text": "The following functions are specific to complex numbers other than absolute value, which has a specific meaning for complex numbers.\n \n\nreal abs(complex z) Return the absolute value of z, also known as the modulus or magnitude, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{abs}(z) = \\sqrt{x^2 + y^2}.\n\\end{equation*}\\]\nThis function works elementwise over containers, returning the same shape and kind of the input container but holding reals. For example, a complex_vector[n] input will return a vector[n] output, with each element transformed by the above equation.\nAvailable since 2.28, vectorized in 2.30\n \n\nreal arg(complex z) Return the phase angle (in radians) of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{arg}(z) = \\textrm{atan2}(y, x) = \\textrm{atan}(y / x).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nreal norm(complex z) Return the Euclidean norm of z, which is its absolute value squared, and which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\textrm{norm}(z) = \\textrm{abs}^2(z) = x^2 + y^2.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex conj(complex z) Return the complex conjugate of z, which negates the imaginary component, so that if \\(z = x + yi\\), \\[\\begin{equation*}\n\\textrm{conj}(z) = x - yi.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ conj(Z z) Vectorized version of conj. This will apply the conj function to each element of a complex array, vector, or matrix.\nAvailable since 2.31\n \n\ncomplex proj(complex z) Return the projection of z onto the Riemann sphere, which for \\(z = x\n+ yi\\) is \\[\\begin{equation*}\n\\textrm{proj}(z)\n= \\begin{cases}\n    z & \\textrm{if} \\ z \\ \\textrm{is finite, and} \\\\\n    0 + \\textrm{sign}(y)i & \\textrm{otherwise,}\n\\end{cases}\n\\end{equation*}\\] where \\(\\textrm{sign}(y)\\) is -1 if \\(y\\) is negative and 1 otherwise.\nAvailable since 2.28\n \n\ncomplex polar(real r, real theta) Return the complex number with magnitude (absolute value) r and phase angle theta.\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Mathematical Functions"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/mathematical_functions.html#beta-appendix",
-    "href": "functions-reference/mathematical_functions.html#beta-appendix",
-    "title": "Mathematical Functions",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-exponential",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-exponential",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "The beta function, \\(\\text{B}(a, b)\\), computes the normalizing constant for the beta distribution, and is defined for \\(a &gt; 0\\) and \\(b\n&gt; 0\\) by \\[\\begin{equation*} \\text{B}(a,b) \\ = \\ \\int_0^1 u^{a - 1} (1 - u)^{b - 1} \\,\ndu \\ = \\ \\frac{\\Gamma(a) \\, \\Gamma(b)}{\\Gamma(a+b)} \\, , \\end{equation*}\\] where \\(\\Gamma(x)\\) is the Gamma function.",
+    "text": "The exponential, log, and power functions may be supplied with complex arguments with specialized meanings that generalize their real counterparts. These versions are only called when the argument is complex.\n \n\ncomplex exp(complex z) Return the complex natural exponential of z, which for \\(z = x + yi\\) is \\[\\begin{equation*}\n\\exp z = \\exp(x) \\textrm{cis}(y) = \\exp(x) (\\cos(y) + i \\sin(y)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log(complex z) Return the complex natural logarithm of z, which for \\(z = \\textrm{polar}(r,\n\\theta)\\) is \\[\\begin{equation*}\n\\log z = \\log r + \\theta i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex log10(complex z) Return the complex common logarithm of z, \\[\\begin{equation*}\n\\log_{10} z = \\frac{\\log z}{\\log 10}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex pow(complex x, complex y) Return x raised to the power of y, \\[\\begin{equation*}\n\\text{pow}(x,y) = \\textrm{exp}(y \\, \\log(x)).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\nZ pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.30\n \n\ncomplex sqrt(complex x) Return the complex square root of x with branch cut along the negative real axis. For finite inputs, the result will be in the right half-plane.\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Mathematical Functions"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/mathematical_functions.html#inc-beta-appendix",
-    "href": "functions-reference/mathematical_functions.html#inc-beta-appendix",
-    "title": "Mathematical Functions",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-trig",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-trig",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "The incomplete beta function, \\(\\text{B}(x; a, b)\\), is defined for \\(x\n\\in [0, 1]\\) and \\(a, b \\geq 0\\) such that \\(a + b \\neq 0\\) by \\[\\begin{equation*}\n\\text{B}(x; \\, a, b) \\ = \\ \\int_0^x u^{a - 1} \\, (1 - u)^{b - 1} \\,\ndu, \\end{equation*}\\] where \\(\\text{B}(a, b)\\) is the beta function defined in appendix. If \\(x = 1\\), the incomplete beta function reduces to the beta function, \\(\\text{B}(1; a, b) = \\text{B}(a, b)\\).\nThe regularized incomplete beta function divides the incomplete beta function by the beta function, \\[\\begin{equation*} I_x(a, b) \\ = \\ \\frac{\\text{B}(x; \\,\na, b)}{B(a, b)} \\, . \\end{equation*}\\]",
+    "text": "The standard trigonometric functions are supported for complex numbers.\n \n\ncomplex cos(complex z) Return the complex cosine of z, which is \\[\\begin{equation*}\n\\cos(z)\n= \\textrm{cosh}(z \\, i)\n= \\frac{\\displaystyle \\exp(z \\, i) + \\exp(-z \\, i)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sin(complex z) Return the complex sine of z, \\[\\begin{equation*}\n\\sin(z)\n= -\\textrm{sinh}(z \\, i) \\, i\n= \\frac{\\displaystyle \\exp(z \\, i) - \\exp(-z \\, i)}\n       {\\displaystyle 2 \\, i}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tan(complex z) Return the complex tangent of z, \\[\\begin{equation*}\n\\tan(z)\n= -\\textrm{tanh}(z \\, i) \\, i\n= \\frac{(\\exp(-z \\, i) - \\exp(z \\, i)) \\, i}\n       {\\exp(-z \\, i) + \\exp(z \\, i)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acos(complex z) Return the complex arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acos}(z)\n= \\frac{1}{2} \\pi + \\log (z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asin(complex z) Return the complex arc (inverse) sine of z, \\[\\begin{equation*}\n\\text{asin}(z)\n= -\\log(z \\, i + \\sqrt{1 - z^2}) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atan(complex z) Return the complex arc (inverse) tangent of z, \\[\\begin{equation*}\n\\text{atan}(z)\n= - \\frac{1}{2} (\\log(1 - z \\, i) - \\log(1 + z \\, i)) \\, i.\n\\end{equation*}\\]\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Mathematical Functions"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/mathematical_functions.html#gamma-appendix",
-    "href": "functions-reference/mathematical_functions.html#gamma-appendix",
-    "title": "Mathematical Functions",
+    "objectID": "functions-reference/complex-valued_basic_functions.html#complex-h-trig",
+    "href": "functions-reference/complex-valued_basic_functions.html#complex-h-trig",
+    "title": "Complex-Valued Basic Functions",
     "section": "",
-    "text": "The gamma function, \\(\\Gamma(x)\\), is the generalization of the factorial function to continuous variables, defined so that for positive integers \\(n\\), \\[\\begin{equation*} \\Gamma(n+1) = n! \\end{equation*}\\] Generalizing to all positive numbers and non-integer negative numbers, \\[\\begin{equation*} \\Gamma(x) =\n\\int_0^{\\infty} u^{x - 1} \\exp(-u) \\, du. \\end{equation*}\\]",
+    "text": "The standard hyperbolic trigonometric functions are supported for complex numbers.\n \n\ncomplex cosh(complex z) Return the complex hyperbolic cosine of z, \\[\\begin{equation*}\n\\textrm{cosh}(z)\n= \\frac{\\exp(z) + \\exp(-z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex sinh(complex z) Return the complex hyperbolic sine of z, \\[\\begin{equation*}\n\\textrm{sinh}(z)\n= \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n       {\\displaystyle 2}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex tanh(complex z) Return the complex hyperbolic tangent of z, \\[\\begin{equation*}\n\\textrm{tanh}(z)\n\\ = \\ \\frac{\\textrm{sinh}(z)}\n           {\\textrm{cosh}(z)}\n\\ = \\ \\frac{\\displaystyle \\exp(z) - \\exp(-z)}\n           {\\displaystyle \\exp(z) + \\exp(-z)}.\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex acosh(complex z) Return the complex hyperbolic arc (inverse) cosine of z, \\[\\begin{equation*}\n\\textrm{acosh}(z)\n= \\log(z + \\sqrt{(z + 1)(z - 1)}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex asinh(complex z) Return the complex hyperbolic arc (inverse) sine of z, \\[\\begin{equation*}\n\\textrm{asinh}(z)\n= \\log(z + \\sqrt{1 + z^2}).\n\\end{equation*}\\]\nAvailable since 2.28\n \n\ncomplex atanh(complex z) Return the complex hyperbolic arc (inverse) tangent of z, \\[\\begin{equation*}\n\\textrm{atanh}(z)\n= \\frac{\\log(1 + z) - \\log(1 - z)}\n       {2}.\n\\end{equation*}\\]\nAvailable since 2.28",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Mathematical Functions"
+      "Built-in Functions",
+      "Complex-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "functions-reference/mathematical_functions.html#digamma-appendix",
-    "href": "functions-reference/mathematical_functions.html#digamma-appendix",
-    "title": "Mathematical Functions",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "The digamma function \\(\\Psi\\) is the derivative of the \\(\\log \\Gamma\\) function, \\[\\begin{equation*}\n\\Psi(u) \\ = \\ \\frac{d}{d u} \\log \\Gamma(u) \\ = \\ \\frac{1}{\\Gamma(u)} \\ \\frac{d}{d u} \\Gamma(u).\n\\end{equation*}\\]",
+    "text": "Compound arithmetic and assignment statements combine an arithmetic operation and assignment, replacing a statement such as\n x = x op y;\nwith the more compact compound form\n x op= y;\nFor example, x = x + 1; may be replaced with x += 1;. This works for all types that support arithmetic, including the scalar types int, real, complex, the real matrix types vector, row_vector, and matrix, and the complex matrix types, complex_vector, complex_row_vector, and complex_matrix.\n\n\nCompound addition and assignment works wherever the corresponding addition and assignment would be well formed.\n \n\nvoid operator+=(T x, U y) x += y is equivalent to x = x + y. Defined for all types T and U where T = T + U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.\n \n\nvoid operator-=(T x, U y) x -= y is equivalent to x = x - y. Defined for all types T and U where T = T - U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator*=(T x, U y) x *= y is equivalent to x = x * y. Defined for all types T and U where T = T * U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator/=(T x, U y) x /= y is equivalent to x = x / y. Defined for all types T and U where T = T / U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator.*=(T x, U y) x .*= y is equivalent to x = x .* y. Defined for all types T and U where T = T .* U is well formed.\nAvailable since 2.17, complex signatures added in 2.30\n\n\n\nCompound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator./=(T x, U y) x ./= y is equivalent to x = x ./ y. Defined for all types T and U where T = T ./ U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
-      "Appendix",
-      "Mathematical Functions"
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/mixed_operations.html",
-    "href": "functions-reference/mixed_operations.html",
-    "title": "Mixed Operations",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-addition-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-addition-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "Mixed Operations\nThese functions perform conversions between Stan containers matrix, vector, row vector and arrays.\n \n\nmatrix to_matrix(matrix m) Return the matrix m itself.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_matrix m) Return the matrix m itself.\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v) Convert the column vector v to a size(v) by 1 matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_vector v) Convert the column vector v to a size(v) by 1 matrix.\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v) Convert the row vector v to a 1 by size(v) matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(complex_row_vector v) Convert the row vector v to a 1 by size(v) matrix.\nAvailable since 2.30\n \n\nmatrix to_matrix(matrix M, int m, int n) Convert a matrix A to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_matrix M, int m, int n) Convert a matrix A to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v, int m, int n) Convert a vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_vector v, int m, int n) Convert a vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v, int m, int n) Convert a row_vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_row_vector v, int m, int n) Convert a row vector v to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(matrix A, int m, int n, int col_major) Convert a matrix A to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_matrix A, int m, int n, int col_major) Convert a matrix A to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(vector v, int m, int n, int col_major) Convert a vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_vector v, int m, int n, int col_major) Convert a vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(row_vector v, int m, int n, int col_major) Convert a row vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(complex_row_vector v, int m, int n, int col_major) Convert a row vector v to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] real a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\nmatrix to_matrix(array[] int a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(array[] complex a, int m, int n) Convert a one-dimensional array a to a matrix with m rows and n columns filled in column-major order.\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] real a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\nmatrix to_matrix(array[] int a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.15\n \n\ncomplex_matrix to_matrix(array[] complex a, int m, int n, int col_major) Convert a one-dimensional array a to a matrix with m rows and n columns filled in row-major order if col_major equals 0 (otherwise, they get filled in column-major order).\nAvailable since 2.30\n \n\nmatrix to_matrix(array[] row_vector vs) Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.\nAvailable since 2.28\n \n\ncomplex_matrix to_matrix(array[] complex_row_vector vs) Convert a one-dimensional array of row vectors to a matrix, where the size of the array is the number of rows of the resulting matrix and the length of row vectors is the number of columns.\nAvailable since 2.30\n \n\nmatrix to_matrix(array[,] real a) Convert the two dimensional array a to a matrix with the same dimensions and indexing order.\nAvailable since 2.3\n \n\nmatrix to_matrix(array[,] int a) Convert the two dimensional array a to a matrix with the same dimensions and indexing order. If any of the dimensions of a are zero, the result will be a \\(0 \\times 0\\) matrix.\nAvailable since 2.3\n \n\ncomplex_matrix to_matrix(array[,] complex a ) Convert the two dimensional array a to a matrix with the same dimensions and indexing order.\nAvailable since 2.30\n \n\nvector to_vector(matrix m) Convert the matrix m to a column vector in column-major order.\nAvailable since 2.0\n \n\ncomplex_vector to_vector(complex_matrix m) Convert the matrix m to a column vector in column-major order.\nAvailable since 2.30\n \n\nvector to_vector(vector v) Return the column vector v itself.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(complex_vector v) Return the column vector v itself.\nAvailable since 2.30\n \n\nvector to_vector(row_vector v) Convert the row vector v to a column vector.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(complex_row_vector v) Convert the row vector v to a column vector.\nAvailable since 2.30\n \n\nvector to_vector(array[] real a) Convert the one-dimensional array a to a column vector.\nAvailable since 2.3\n \n\nvector to_vector(array[] int a) Convert the one-dimensional integer array a to a column vector.\nAvailable since 2.3\n \n\ncomplex_vector to_vector(array[] complex a) Convert the one-dimensional complex array a to a column vector.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(matrix m) Convert the matrix m to a row vector in column-major order.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_matrix m) Convert the matrix m to a row vector in column-major order.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(vector v) Convert the column vector v to a row vector.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_vector v) Convert the column vector v to a row vector.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(row_vector v) Return the row vector v itself.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(complex_row_vector v) Return the row vector v itself.\nAvailable since 2.30\n \n\nrow_vector to_row_vector(array[] real a) Convert the one-dimensional array a to a row vector.\nAvailable since 2.3\n \n\nrow_vector to_row_vector(array[] int a) Convert the one-dimensional array a to a row vector.\nAvailable since 2.3\n \n\ncomplex_row_vector to_row_vector(array[] complex a) Convert the one-dimensional complex array a to a row vector.\nAvailable since 2.30\n \n\narray[,] real to_array_2d(matrix m) Convert the matrix m to a two dimensional array with the same dimensions and indexing order.\nAvailable since 2.3\n \n\narray[,] complex to_array_2d(complex_matrix m) Convert the matrix m to a two dimensional array with the same dimensions and indexing order.\nAvailable since 2.30\n \n\narray[] real to_array_1d(vector v) Convert the column vector v to a one-dimensional array.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(complex_vector v) Convert the column vector v to a one-dimensional array.\nAvailable since 2.30\n \n\narray[] real to_array_1d(row_vector v) Convert the row vector v to a one-dimensional array.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(complex_row_vector v) Convert the row vector v to a one-dimensional array.\nAvailable since 2.30\n \n\narray[] real to_array_1d(matrix m) Convert the matrix m to a one-dimensional array in column-major order.\nAvailable since 2.3\n \n\narray[] real to_array_1d(complex_matrix m) Convert the matrix m to a one-dimensional array in column-major order.\nAvailable since 2.30\n \n\narray[] real to_array_1d(array[...] real a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.3\n \n\narray[] int to_array_1d(array[...] int a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.3\n \n\narray[] complex to_array_1d(array[...] complex a) Convert the array a (of any dimension up to 10) to a one-dimensional array in row-major order.\nAvailable since 2.30\n\n\n\n\n Back to top",
+    "text": "Compound addition and assignment works wherever the corresponding addition and assignment would be well formed.\n \n\nvoid operator+=(T x, U y) x += y is equivalent to x = x + y. Defined for all types T and U where T = T + U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Mixed Operations"
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html",
-    "href": "functions-reference/positive_continuous_distributions.html",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-subtraction-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-subtraction-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "The positive continuous probability functions have support on the positive real numbers.\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}^+\\), \\[\\begin{equation*} \\text{LogNormal}(y|\\mu,\\sigma) = \\frac{1}{\\sqrt{2\n\\pi} \\ \\sigma} \\, \\frac{1}{y} \\ \\exp \\! \\left(        - \\, \\frac{1}{2}\n\\, \\left( \\frac{\\log y - \\mu}{\\sigma} \\right)^2      \\right) . \\end{equation*}\\]\n\n\n\ny ~ lognormal(mu, sigma)\nIncrement target log probability density with lognormal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal lognormal_lpdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lupdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal lognormal_cdf(reals y | reals mu, reals sigma) The cumulative lognormal distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal lognormal_lcdf(reals y | reals mu, reals sigma) The log of the lognormal cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lccdf(reals y | reals mu, reals sigma) The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR lognormal_rng(reals mu, reals sigma) Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.22\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{ChiSquare}(y|\\nu) = \\frac{2^{-\\nu/2}}     {\\Gamma(\\nu / 2)} \\,\ny^{\\nu/2 - 1} \\, \\exp \\! \\left( -\\, \\frac{1}{2} \\, y \\right) . \\end{equation*}\\]\n\n\n\ny ~ chi_square(nu)\nIncrement target log probability density with chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal chi_square_lpdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lupdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal chi_square_cdf(reals y | reals nu) The Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal chi_square_lcdf(reals y | reals nu) The log of the Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lccdf(reals y | reals nu) The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR chi_square_rng(reals nu) Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{InvChiSquare}(y \\, | \\, \\nu) = \\frac{2^{-\\nu/2}}    {\\Gamma(\\nu\n/ 2)} \\, y^{-\\nu/2 - 1} \\, \\exp\\! \\left( \\! - \\, \\frac{1}{2} \\,\n\\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_chi_square(nu)\nIncrement target log probability density with inv_chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_chi_square_lpdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lupdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_chi_square_cdf(reals y | reals nu) The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal inv_chi_square_lcdf(reals y | reals nu) The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lccdf(reals y | reals nu) The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR inv_chi_square_rng(reals nu) Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{ScaledInvChiSquare}(y|\\nu,\\sigma) =\n\\frac{(\\nu / 2)^{\\nu/2}}      {\\Gamma(\\nu / 2)} \\, \\sigma^{\\nu} \\,\ny^{-(\\nu/2 + 1)} \\, \\exp \\! \\left( \\!    - \\, \\frac{1}{2} \\, \\nu \\,\n\\sigma^2 \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ scaled_inv_chi_square(nu, sigma)\nIncrement target log probability density with scaled_inv_chi_square_lupdf(y | nu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal scaled_inv_chi_square_lpdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lupdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal scaled_inv_chi_square_cdf(reals y | reals nu, reals sigma) The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.0\n \n\nreal scaled_inv_chi_square_lcdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lccdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nR scaled_inv_chi_square_rng(reals nu, reals sigma) Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf inverse scale (rate) \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Exponential}(y|\\beta) = \\beta \\, \\exp ( - \\beta \\, y ) . \\end{equation*}\\]\n\n\n\ny ~ exponential(beta)\nIncrement target log probability density with exponential_lupdf(y | beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exponential_lpdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lupdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal exponential_cdf(reals y | reals beta) The exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.0\n \n\nreal exponential_lcdf(reals y | reals beta) The log of the exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lccdf(reals y | reals beta) The log of the exponential complementary cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nR exponential_rng(reals beta) Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf the shape parameter \\(\\alpha \\in \\mathbb{R}^+\\) and the rate (or inverse scale) parameter \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Gamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\, y^{\\alpha - 1}\n\\exp(-\\beta \\, y) . \\end{equation*}\\]\nUnder the shape and rate formulation of the Gamma distribution, \\(\\mathbb{E}[y] = \\alpha / \\beta\\) and \\(\\textrm{var}[y] = \\alpha / \\beta^2\\).\n\n\n\ny ~ gamma(alpha, beta)\nIncrement target log probability density with gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gamma_lpdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lupdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gamma_cdf(reals y | reals alpha, reals beta) The cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal gamma_lcdf(reals y | reals alpha, reals beta) The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lccdf(reals y | reals alpha, reals beta) The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR gamma_rng(reals alpha, reals beta) Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{InvGamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\ y^{-(\\alpha + 1)} \\,\n\\exp \\! \\left( \\! - \\beta \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_gamma(alpha, beta)\nIncrement target log probability density with inv_gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_gamma_lpdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lupdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_gamma_cdf(reals y | reals alpha, reals beta) The inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.0\n \n\nreal inv_gamma_lcdf(reals y | reals alpha, reals beta) The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lccdf(reals y | reals alpha, reals beta) The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nR inv_gamma_rng(reals alpha, reals beta) Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*} \\text{Weibull}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Weibull}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Frechet}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ weibull(alpha, sigma)\nIncrement target log probability density with weibull_lupdf(y | alpha, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal weibull_lpdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lupdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal weibull_cdf(reals y | reals alpha, reals sigma) The Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.0\n \n\nreal weibull_lcdf(reals y | reals alpha, reals sigma) The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lccdf(reals y | reals alpha, reals sigma) The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR weibull_rng(reals alpha, reals sigma) Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Frechet}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Frechet}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Weibull}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ frechet(alpha, sigma)\nIncrement target log probability density with frechet_lupdf(y | alpha, sigma).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal frechet_lpdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lupdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal frechet_cdf(reals y | reals alpha, reals sigma) The Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.5\n \n\nreal frechet_lcdf(reals y | reals alpha, reals sigma) The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lccdf(reals y | reals alpha, reals sigma) The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR frechet_rng(reals alpha, reals sigma) Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*}\n\\text{Rayleigh}(y|\\sigma) = \\frac{y}{\\sigma^2} \\exp(-y^2 / 2\\sigma^2)\n\\!. \\end{equation*}\\]\n\n\n\ny ~ rayleigh(sigma)\nIncrement target log probability density with rayleigh_lupdf(y | sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal rayleigh_lpdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lupdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal rayleigh_cdf(real y | real sigma) The Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.0\n \n\nreal rayleigh_lcdf(real y | real sigma) The log of the Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lccdf(real y | real sigma) The log of the Rayleigh complementary cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nR rayleigh_rng(reals sigma) Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\alpha, \\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Log-Logistic}(y|\\alpha,\\beta) =\n\\frac{\\ \\left(\\frac{\\beta}{\\alpha}\\right) \\left(\\frac{y}{\\alpha}\\right)^{\\beta-1}\\ }{\\left(1 + \\left(\\frac{y}{\\alpha}\\right)^\\beta\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ loglogistic(alpha, beta)\nIncrement target log probability density with unnormalized version of loglogistic_lpdf(y | alpha, beta)\nAvailable since 2.29\n \n\n\n\n\n \n\nreal loglogistic_lpdf(reals y | reals alpha, reals beta) The log of the log-logistic density of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nreal loglogistic_cdf(reals y | reals alpha, reals beta) The log-logistic cumulative distribution function of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nR loglogistic_rng(reals alpha, reals beta) Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.29",
+    "text": "Compound addition and assignment works wherever the corresponding subtraction and assignment would be well formed.\n \n\nvoid operator-=(T x, U y) x -= y is equivalent to x = x - y. Defined for all types T and U where T = T - U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#lognormal",
-    "href": "functions-reference/positive_continuous_distributions.html#lognormal",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-multiplication-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-multiplication-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}^+\\), \\[\\begin{equation*} \\text{LogNormal}(y|\\mu,\\sigma) = \\frac{1}{\\sqrt{2\n\\pi} \\ \\sigma} \\, \\frac{1}{y} \\ \\exp \\! \\left(        - \\, \\frac{1}{2}\n\\, \\left( \\frac{\\log y - \\mu}{\\sigma} \\right)^2      \\right) . \\end{equation*}\\]\n\n\n\ny ~ lognormal(mu, sigma)\nIncrement target log probability density with lognormal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal lognormal_lpdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lupdf(reals y | reals mu, reals sigma) The log of the lognormal density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal lognormal_cdf(reals y | reals mu, reals sigma) The cumulative lognormal distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal lognormal_lcdf(reals y | reals mu, reals sigma) The log of the lognormal cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal lognormal_lccdf(reals y | reals mu, reals sigma) The log of the lognormal complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR lognormal_rng(reals mu, reals sigma) Generate a lognormal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.22",
+    "text": "Compound multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator*=(T x, U y) x *= y is equivalent to x = x * y. Defined for all types T and U where T = T * U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#chi-square-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#chi-square-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-division-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-division-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{ChiSquare}(y|\\nu) = \\frac{2^{-\\nu/2}}     {\\Gamma(\\nu / 2)} \\,\ny^{\\nu/2 - 1} \\, \\exp \\! \\left( -\\, \\frac{1}{2} \\, y \\right) . \\end{equation*}\\]\n\n\n\ny ~ chi_square(nu)\nIncrement target log probability density with chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal chi_square_lpdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lupdf(reals y | reals nu) The log of the Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal chi_square_cdf(reals y | reals nu) The Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal chi_square_lcdf(reals y | reals nu) The log of the Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal chi_square_lccdf(reals y | reals nu) The log of the complementary Chi-square cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR chi_square_rng(reals nu) Generate a Chi-square variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "Compound division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator/=(T x, U y) x /= y is equivalent to x = x / y. Defined for all types T and U where T = T / U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#inverse-chi-square-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#inverse-chi-square-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-multiplication-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-multiplication-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{InvChiSquare}(y \\, | \\, \\nu) = \\frac{2^{-\\nu/2}}    {\\Gamma(\\nu\n/ 2)} \\, y^{-\\nu/2 - 1} \\, \\exp\\! \\left( \\! - \\, \\frac{1}{2} \\,\n\\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_chi_square(nu)\nIncrement target log probability density with inv_chi_square_lupdf(y | nu).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_chi_square_lpdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lupdf(reals y | reals nu) The log of the inverse Chi-square density of y given degrees of freedom nu dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_chi_square_cdf(reals y | reals nu) The inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.0\n \n\nreal inv_chi_square_lcdf(reals y | reals nu) The log of the inverse Chi-squared cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nreal inv_chi_square_lccdf(reals y | reals nu) The log of the inverse Chi-squared complementary cumulative distribution function of y given degrees of freedom nu\nAvailable since 2.12\n \n\nR inv_chi_square_rng(reals nu) Generate an inverse Chi-squared variate with degrees of freedom nu; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "Compound elementwise multiplication and assignment works wherever the corresponding multiplication and assignment would be well formed.\n \n\nvoid operator.*=(T x, U y) x .*= y is equivalent to x = x .* y. Defined for all types T and U where T = T .* U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#scaled-inverse-chi-square-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#scaled-inverse-chi-square-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-division-and-assignment",
+    "href": "functions-reference/compound_arithmetic_and_assignment.html#compound-elementwise-division-and-assignment",
+    "title": "Compound Arithmetic and Assignment",
     "section": "",
-    "text": "If \\(\\nu \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{ScaledInvChiSquare}(y|\\nu,\\sigma) =\n\\frac{(\\nu / 2)^{\\nu/2}}      {\\Gamma(\\nu / 2)} \\, \\sigma^{\\nu} \\,\ny^{-(\\nu/2 + 1)} \\, \\exp \\! \\left( \\!    - \\, \\frac{1}{2} \\, \\nu \\,\n\\sigma^2 \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ scaled_inv_chi_square(nu, sigma)\nIncrement target log probability density with scaled_inv_chi_square_lupdf(y | nu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal scaled_inv_chi_square_lpdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lupdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square density of y given degrees of freedom nu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal scaled_inv_chi_square_cdf(reals y | reals nu, reals sigma) The scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.0\n \n\nreal scaled_inv_chi_square_lcdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nreal scaled_inv_chi_square_lccdf(reals y | reals nu, reals sigma) The log of the scaled inverse Chi-square complementary cumulative distribution function of y given degrees of freedom nu and scale sigma\nAvailable since 2.12\n \n\nR scaled_inv_chi_square_rng(reals nu, reals sigma) Generate a scaled inverse Chi-squared variate with degrees of freedom nu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "Compound elementwise division and assignment works wherever the corresponding division and assignment would be well formed.\n \n\nvoid operator./=(T x, U y) x ./= y is equivalent to x = x ./ y. Defined for all types T and U where T = T ./ U is well formed.\nAvailable since 2.17, complex signatures added in 2.30",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Compound Arithmetic and Assignment"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#exponential-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#exponential-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html",
+    "href": "functions-reference/conventions_for_probability_functions.html",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If inverse scale (rate) \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Exponential}(y|\\beta) = \\beta \\, \\exp ( - \\beta \\, y ) . \\end{equation*}\\]\n\n\n\ny ~ exponential(beta)\nIncrement target log probability density with exponential_lupdf(y | beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exponential_lpdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lupdf(reals y | reals beta) The log of the exponential density of y given inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal exponential_cdf(reals y | reals beta) The exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.0\n \n\nreal exponential_lcdf(reals y | reals beta) The log of the exponential cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nreal exponential_lccdf(reals y | reals beta) The log of the exponential complementary cumulative distribution function of y given inverse scale beta\nAvailable since 2.12\n \n\nR exponential_rng(reals beta) Generate an exponential variate with inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "Functions associated with distributions are set up to follow the same naming conventions for both built-in distributions and for user-defined distributions.\n\n\nThe suffix is determined by the type of function according to the following table.\n\n\n\n\n\n\n\n\nfunction\noutcome\nsuffix\n\n\n\n\nlog probability mass function\ndiscrete\n_lpmf\n\n\nlog probability density function\ncontinuous\n_lpdf\n\n\nlog cumulative distribution function\nany\n_lcdf\n\n\nlog complementary cumulative distribution function\nany\n_lccdf\n\n\nrandom number generator\nany\n_rng\n\n\n\nFor example, normal_lpdf is the log of the normal probability density function (pdf) and bernoulli_lpmf is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, normal_lcdf and bernoulli_lcdf.\n\n\n\nEach probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, normal_lpdf(y | mu, sigma) returns the value of mathematical formula \\(\\log \\text{Normal}(y \\, | \\, \\mu, \\sigma)\\). Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., normal_lcdf(y_low | mu, sigma)\n\n\n\nThe notation\n y ~ normal(mu, sigma);\nprovides the same (proportional) contribution to the model log density as the explicit target density increment,\n target += normal_lpdf(y | mu, sigma);\nIn both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (~) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.\n\n\n\nAll of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.\n\n\n\nMany distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter \\(\\sigma &gt; 0\\). If \\(\\sigma = 0\\), the probability function will throw an exception.\nThis is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.\n\n\n\nFor most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix _rng. For example, the function normal_rng(real, real) accepts two real arguments, an unconstrained location \\(\\mu\\) and positive scale \\(\\sigma &gt; 0\\), and returns an unconstrained pseudorandom value drawn from \\(\\text{Normal}(\\mu,\\sigma)\\). There are also vectorized forms of random number generators which return more than one random variate at a time.\n\n\nUnlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.\n\n\n\nUnlike the probability functions, only some of the PRNG functions are vectorized.\n\n\n\n\nFor most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.\nFor a univariate random variable \\(Y\\) with probability function \\(p_Y(y \\, | \\, \\theta)\\), the cumulative distribution function (CDF) \\(F_Y\\) is defined by \\[\\begin{equation*}\nF_Y(y) \\ = \\ \\text{Pr}[Y \\le y] \\ = \\ \\int_{-\\infty}^y p(y\\, | \\, \\theta) \\ \\text{d}y.\n\\end{equation*}\\] The complementary cumulative distribution function (CCDF) is defined as \\[\\begin{equation*}\n\\text{Pr}[Y &gt; y] \\ = \\ 1 - F_Y(y).\n\\end{equation*}\\] The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly \\(10^{-300}\\)), but not numbers very close to 1 (the closest you can get is roughly \\(1 - 10^{-15}\\)).\nIn Stan, there is a cumulative distribution function for each probability function. For instance, normal_cdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\ \\text{d}y.\n\\end{equation*}\\] There are also log forms of the CDF and CCDF for most univariate distributions. For example, normal_lcdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right)\n\\end{equation*}\\] and normal_lccdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( 1 - \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right).\n\\end{equation*}\\]\n\n\n\nStan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section vectorized PRNG functions for more details.\nIn all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.\nStan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.\nStan also overloads some scalar functions, such as log and exp, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.\n\n\n\n\nThe normal probability function is specified with the signature\n normal_lpdf(reals | reals, reals);\nThe pseudotype reals is used to indicate that an argument position may be vectorized. Argument positions declared as reals may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use normal_lpdf(row_vector | vector, real) as long as the vector and row vector have the same size.\n\n\n\nThe multivariate normal distribution accepting vector or array of vector arguments is written as\n multi_normal_lpdf(vectors | vectors, matrix);\nThese arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.\n\n\n\nThe pseudotype ints is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.\n\n\n\n\nThe result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely real or int, is repeated. For instance, if y is a vector of size N, mu is a vector of size N, and sigma is a scalar, then\n ll = normal_lpdf(y | mu, sigma);\nis just a more efficient way to write\n ll = 0;\n for (n in 1:N) {\n   ll = ll + normal_lpdf(y[n] | mu[n], sigma);\n }\nWith the same arguments, the vectorized sampling statement\n y ~ normal(mu, sigma);\nhas the same effect on the total log probability as\n for (n in 1:N) {\n   y[n] ~ normal(mu[n], sigma);\n }\n\n\n\nSome PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes reals or ints. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.\n vector[3] mu = // ...\n array[3] real x = normal_rng(mu, 3);\n\n\nIn the case of PRNG functions, arguments marked ints may be integers or integer arrays, whereas arguments marked reals may be integers or reals, integer or real arrays, vectors, or row vectors.\n\n\n\n\n\n\n\npseudotype\nallowable PRNG arguments\n\n\n\n\nints\nint, array[] int\n\n\nreals\nint, array[] int, real, array[] real, vector, row_vector\n\n\n\n\n\n\nIn general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the normal_rng function may be called with one vector argument and one real array argument as long as they have the same number of elements.\n vector[3] mu = // ...\n array[3] real sigma = // ...\n array[3] real x = normal_rng(mu, sigma);\n\n\n\nThe result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (array[] real) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return ints and continuous distributions return reals, each of appropriate size. The symbol R denotes such a return type.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#gamma-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#gamma-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#suffix-marks-type-of-function",
+    "href": "functions-reference/conventions_for_probability_functions.html#suffix-marks-type-of-function",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If the shape parameter \\(\\alpha \\in \\mathbb{R}^+\\) and the rate (or inverse scale) parameter \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Gamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\, y^{\\alpha - 1}\n\\exp(-\\beta \\, y) . \\end{equation*}\\]\nUnder the shape and rate formulation of the Gamma distribution, \\(\\mathbb{E}[y] = \\alpha / \\beta\\) and \\(\\textrm{var}[y] = \\alpha / \\beta^2\\).\n\n\n\ny ~ gamma(alpha, beta)\nIncrement target log probability density with gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gamma_lpdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lupdf(reals y | reals alpha, reals beta) The log of the gamma density of y given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gamma_cdf(reals y | reals alpha, reals beta) The cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal gamma_lcdf(reals y | reals alpha, reals beta) The log of the cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal gamma_lccdf(reals y | reals alpha, reals beta) The log of the complementary cumulative gamma distribution function of y given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR gamma_rng(reals alpha, reals beta) Generate a gamma variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "The suffix is determined by the type of function according to the following table.\n\n\n\n\n\n\n\n\nfunction\noutcome\nsuffix\n\n\n\n\nlog probability mass function\ndiscrete\n_lpmf\n\n\nlog probability density function\ncontinuous\n_lpdf\n\n\nlog cumulative distribution function\nany\n_lcdf\n\n\nlog complementary cumulative distribution function\nany\n_lccdf\n\n\nrandom number generator\nany\n_rng\n\n\n\nFor example, normal_lpdf is the log of the normal probability density function (pdf) and bernoulli_lpmf is the log of the bernoulli probability mass function (pmf). The log of the corresponding cumulative distribution functions (cdf) use the same suffix, normal_lcdf and bernoulli_lcdf.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#inverse-gamma-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#inverse-gamma-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#argument-order-and-the-vertical-bar",
+    "href": "functions-reference/conventions_for_probability_functions.html#argument-order-and-the-vertical-bar",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y\n\\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{InvGamma}(y|\\alpha,\\beta) =\n\\frac{\\beta^{\\alpha}}      {\\Gamma(\\alpha)} \\ y^{-(\\alpha + 1)} \\,\n\\exp \\! \\left( \\! - \\beta \\, \\frac{1}{y} \\right) . \\end{equation*}\\]\n\n\n\ny ~ inv_gamma(alpha, beta)\nIncrement target log probability density with inv_gamma_lupdf(y | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_gamma_lpdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lupdf(reals y | reals alpha, reals beta) The log of the inverse gamma density of y given shape alpha and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal inv_gamma_cdf(reals y | reals alpha, reals beta) The inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.0\n \n\nreal inv_gamma_lcdf(reals y | reals alpha, reals beta) The log of the inverse gamma cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nreal inv_gamma_lccdf(reals y | reals alpha, reals beta) The log of the inverse gamma complementary cumulative distribution function of y given shape alpha and scale beta\nAvailable since 2.12\n \n\nR inv_gamma_rng(reals alpha, reals beta) Generate an inverse gamma variate with shape alpha and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "Each probability function has a specific outcome value and a number of parameters. Following conditional probability notation, probability density and mass functions use a vertical bar to separate the outcome from the parameters of the distribution. For example, normal_lpdf(y | mu, sigma) returns the value of mathematical formula \\(\\log \\text{Normal}(y \\, | \\, \\mu, \\sigma)\\). Cumulative distribution functions separate the outcome from the parameters in the same way (e.g., normal_lcdf(y_low | mu, sigma)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#weibull-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#weibull-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#sampling-notation",
+    "href": "functions-reference/conventions_for_probability_functions.html#sampling-notation",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*} \\text{Weibull}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Weibull}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Frechet}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ weibull(alpha, sigma)\nIncrement target log probability density with weibull_lupdf(y | alpha, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal weibull_lpdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lupdf(reals y | reals alpha, reals sigma) The log of the Weibull density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal weibull_cdf(reals y | reals alpha, reals sigma) The Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.0\n \n\nreal weibull_lcdf(reals y | reals alpha, reals sigma) The log of the Weibull cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal weibull_lccdf(reals y | reals alpha, reals sigma) The log of the Weibull complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR weibull_rng(reals alpha, reals sigma) Generate a weibull variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "The notation\n y ~ normal(mu, sigma);\nprovides the same (proportional) contribution to the model log density as the explicit target density increment,\n target += normal_lpdf(y | mu, sigma);\nIn both cases, the effect is to add terms to the target log density. The only difference is that the example with the sampling (~) notation drops all additive constants in the log density; the constants are not necessary for any of Stan’s sampling, approximation, or optimization algorithms.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#frechet-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#frechet-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#finite-inputs",
+    "href": "functions-reference/conventions_for_probability_functions.html#finite-inputs",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*} \\text{Frechet}(y|\\alpha,\\sigma) =\n\\frac{\\alpha}{\\sigma} \\, \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha - 1}\n\\, \\exp \\! \\left( \\! - \\left( \\frac{y}{\\sigma} \\right)^{-\\alpha}\n\\right) . \\end{equation*}\\]\nNote that if \\(Y \\propto \\text{Frechet}(\\alpha,\\sigma)\\), then \\(Y^{-1}\n\\propto \\text{Weibull}(\\alpha,\\sigma^{-1})\\).\n\n\n\ny ~ frechet(alpha, sigma)\nIncrement target log probability density with frechet_lupdf(y | alpha, sigma).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal frechet_lpdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lupdf(reals y | reals alpha, reals sigma) The log of the Frechet density of y given shape alpha and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal frechet_cdf(reals y | reals alpha, reals sigma) The Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.5\n \n\nreal frechet_lcdf(reals y | reals alpha, reals sigma) The log of the Frechet cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nreal frechet_lccdf(reals y | reals alpha, reals sigma) The log of the Frechet complementary cumulative distribution function of y given shape alpha and scale sigma\nAvailable since 2.12\n \n\nR frechet_rng(reals alpha, reals sigma) Generate a Frechet variate with shape alpha and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "All of the distribution functions are configured to throw exceptions (effectively rejecting iterations or optimization steps) when they are supplied with non-finite arguments. The two cases of non-finite arguments are the infinite values and not-a-number value—these are standard in floating-point arithmetic.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#rayleigh-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#rayleigh-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#boundary-conditions",
+    "href": "functions-reference/conventions_for_probability_functions.html#boundary-conditions",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in [0,\\infty)\\), \\[\\begin{equation*}\n\\text{Rayleigh}(y|\\sigma) = \\frac{y}{\\sigma^2} \\exp(-y^2 / 2\\sigma^2)\n\\!. \\end{equation*}\\]\n\n\n\ny ~ rayleigh(sigma)\nIncrement target log probability density with rayleigh_lupdf(y | sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal rayleigh_lpdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lupdf(reals y | reals sigma) The log of the Rayleigh density of y given scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal rayleigh_cdf(real y | real sigma) The Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.0\n \n\nreal rayleigh_lcdf(real y | real sigma) The log of the Rayleigh cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nreal rayleigh_lccdf(real y | real sigma) The log of the Rayleigh complementary cumulative distribution of y given scale sigma\nAvailable since 2.12\n \n\nR rayleigh_rng(reals sigma) Generate a Rayleigh variate with scale sigma; may only be used in generated quantities block. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "Many distributions are defined with support or constraints on parameters forming an open interval. For example, the normal density function accepts a scale parameter \\(\\sigma &gt; 0\\). If \\(\\sigma = 0\\), the probability function will throw an exception.\nThis is true even for (complementary) cumulative distribution functions, which will throw exceptions when given input that is out of the support.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/positive_continuous_distributions.html#log-logistic-distribution",
-    "href": "functions-reference/positive_continuous_distributions.html#log-logistic-distribution",
-    "title": "Positive Continuous Distributions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#distributions-prng",
+    "href": "functions-reference/conventions_for_probability_functions.html#distributions-prng",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "If \\(\\alpha, \\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\), \\[\\begin{equation*}\n\\text{Log-Logistic}(y|\\alpha,\\beta) =\n\\frac{\\ \\left(\\frac{\\beta}{\\alpha}\\right) \\left(\\frac{y}{\\alpha}\\right)^{\\beta-1}\\ }{\\left(1 + \\left(\\frac{y}{\\alpha}\\right)^\\beta\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ loglogistic(alpha, beta)\nIncrement target log probability density with unnormalized version of loglogistic_lpdf(y | alpha, beta)\nAvailable since 2.29\n \n\n\n\n\n \n\nreal loglogistic_lpdf(reals y | reals alpha, reals beta) The log of the log-logistic density of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nreal loglogistic_cdf(reals y | reals alpha, reals beta) The log-logistic cumulative distribution function of y given scale alpha and shape beta\nAvailable since 2.29\n \n\nR loglogistic_rng(reals alpha, reals beta) Generate a log-logistic variate with scale alpha and shape beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.29",
+    "text": "For most of the probability functions, there is a matching pseudorandom number generator (PRNG) with the suffix _rng. For example, the function normal_rng(real, real) accepts two real arguments, an unconstrained location \\(\\mu\\) and positive scale \\(\\sigma &gt; 0\\), and returns an unconstrained pseudorandom value drawn from \\(\\text{Normal}(\\mu,\\sigma)\\). There are also vectorized forms of random number generators which return more than one random variate at a time.\n\n\nUnlike regular functions, the PRNG functions may only be used in the transformed data or generated quantities blocks.\n\n\n\nUnlike the probability functions, only some of the PRNG functions are vectorized.",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Positive Continuous Distributions"
+      "Built-in Functions",
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html",
-    "href": "functions-reference/real-valued_basic_functions.html",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#cumulative-distribution-functions",
+    "href": "functions-reference/conventions_for_probability_functions.html#cumulative-distribution-functions",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "This chapter describes built-in functions that take zero or more real or integer arguments and return real values.\n\n\nAlthough listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.\n\n\nMany of Stan’s unary functions can be applied to any argument type. For example, the exponential function, exp, can be applied to real arguments or arrays of real arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.\n\n\nWhen applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like exp() are defined elementwise. For example,\n // declare some variables for arguments\n real x0;\n array[5] real x1;\n array[4, 7] real x2;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = exp(x0);\n y1 = exp(x1);\n y2 = exp(x2);\nWhen exp is applied to an array, it applies elementwise. For example, the statement above,\n y2 = exp(x2);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = exp(x2[i, j]);\n  }\n}\n\n\n\nVectorized functions also apply elementwise to vectors and matrices. For example,\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = exp(xv);\n yrv = exp(xrv);\n ym = exp(xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = exp(u);\nAfter this has been executed, z[i, j, k] will be equal to exp(u[i, j, k]).\n\n\n\nInteger arguments are promoted to real values in vectorized unary functions. Thus if n is of type int, exp(n) is of type real. Arrays work the same way, so that if n2 is a one dimensional array of integers, then exp(n2) will be a one-dimensional array of reals with the same number of elements as n2. For example,\n array[23] int n1;\n array[23] real z1;\n z1 = exp(n1);\nIt would be illegal to try to assign exp(n1) to an array of integers; the return type is a real array.\n\n\n\n\nLike the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.\n\n\nWhen applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like pow() are defined elementwise. For example,\n // declare some variables for arguments\n real x00;\n real x01;\n array[5] real x10;\n array[5]real x11;\n array[4, 7] real x20;\n array[4, 7] real x21;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = pow(x00, x01);\n y1 = pow(x10, x11);\n y2 = pow(x20, x21);\nWhen pow is applied to two arrays, it applies elementwise. For example, the statement above,\n y2 = pow(x20, x21);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x21[i, j]);\n  }\n}\nAlternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:\ny2 = pow(x20, x00);\nproduces the same result for y2 as the explicit loop:\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x00);\n  }\n}\n\n\n\nVectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,\n real x00;\n vector[5] xv00;\n vector[5] xv01;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = pow(xv00, xv01);\n yrv = pow(xrv, x00);\n ym = pow(x00, xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = pow(u, x00);\nAfter this has been executed, z[i, j, k] will be equal to pow(u[i, j, k], x00).\n\n\n\nVectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv = pow(xv, xv)\n row_vector[7] yrv = pow(xrv, xrv)\n matrix[10, 20] = pow(xm, xm)\nBut the following statements are not:\n vector[5] xv;\n vector[7] xv2;\n row_vector[5] xrv;\n\n // Cannot mix different types\n vector[5] yv = pow(xv, xrv)\n\n // Cannot mix different sizes of the same type\n vector[5] yv = pow(xv, xv2)\nWhile the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., bessel_first_kind). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:\n vector[5] xv;\n matrix[5, 5] xm;\n array[5] int xi;\n array[5, 5] int xii;\n\n vector[5] yv = bessel_first_kind(xi, xv);\n matrix[5, 5] ym = bessel_first_kind(xii, xm);\nWhereas these are not:\n vector[5] xv;\n matrix[5, 5] xm;\n array[7] int xi;\n\n // Dimensions of containers do not match\n vector[5] yv = bessel_first_kind(xi, xv);\n\n // Function requires first argument be an integer type\n matrix[5, 5] ym = bessel_first_kind(xm, xm);\n\n\n\n\n\nConstants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant \\(\\pi\\) must be written in a Stan program as pi().\n \n\nreal pi() \\(\\pi\\), the ratio of a circle’s circumference to its diameter\nAvailable since 2.0\n \n\nreal e() \\(e\\), the base of the natural logarithm\nAvailable since 2.0\n \n\nreal sqrt2() The square root of 2\nAvailable since 2.0\n \n\nreal log2() The natural logarithm of 2\nAvailable since 2.0\n \n\nreal log10() The natural logarithm of 10\nAvailable since 2.0\n\n\n\n \n\nreal not_a_number() Not-a-number, a special non-finite real value returned to signal an error\nAvailable since 2.0\n \n\nreal positive_infinity() Positive infinity, a special non-finite real value larger than all finite numbers\nAvailable since 2.0\n \n\nreal negative_infinity() Negative infinity, a special non-finite real value smaller than all finite numbers\nAvailable since 2.0\n \n\nreal machine_precision() The smallest number \\(x\\) such that \\((x + 1) \\neq 1\\) in floating-point arithmetic on the current hardware platform\nAvailable since 2.0\n\n\n\nThe basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (~) and log probability increment statement (increment_log_prob) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.\nStan provides a special built-in function target() that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.\n \n\nreal target() Return the current value of the log probability accumulator.\nAvailable since 2.10\ntarget acts like a function ending in _lp, meaning that it may only may only be used in the model block.\n\n\n\nLike C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.\n\n\nAll comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.\n \n\nint operator&lt;(int x, int y)\n \n\nint operator&lt;(real x, real y) Return 1 if x is less than y and 0 otherwise. \\[\\begin{equation*} \\text{operator&lt;}(x,y)\n= \\begin{cases} 1 & \\text{if $x &lt; y$} \\\\ 0 & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&lt;=(int x, int y)\n \n\nint operator&lt;=(real x, real y) Return 1 if x is less than or equal y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&lt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\leq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;(int x, int y)\n \n\nint operator&gt;(real x, real y) Return 1 if x is greater than y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;}(x,y) = \\begin{cases} 1 & \\text{if $x &gt; y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;=(int x, int y)\n \n\nint operator&gt;=(real x, real y) Return 1 if x is greater than or equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\geq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator==(int x, int y)\n \n\nint operator==(real x, real y) Return 1 if x is equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator==}(x,y) = \\begin{cases} 1 & \\text{if $x = y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!=(int x, int y)\n \n\nint operator!=(real x, real y) Return 1 if x is not equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator!=}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq y$} \\\\ 0 &\n\\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\nBoolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as !a && b is interpreted as (!a) && b, and a &lt; b || c &gt;= d && e != f as (a &lt; b) || (((c &gt;= d) && (e != f))).\n \n\nint operator!(int x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0$} \\\\ 1 & \\text{if $x = 0$}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!(real x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0.0$} \\\\ 1 & \\text{if $x = 0.0$}\n\\end{cases} \\end{equation*}\\] deprecated; - use operator== instead.\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator&&(int x, int y)\nReturn 1 if x is unequal to 0 and y is unequal to 0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\text{ and } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&&(real x, real y) Return 1 if x is unequal to 0.0 and y is unequal to 0.0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\text{ and } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator||(int x, int y) Return 1 if x is unequal to 0 or y is unequal to 0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\textrm{ or } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator||(real x, real y) Return 1 if x is unequal to 0.0 or y is unequal to 0.0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\textrm{ or } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n\n\nLike in C++, the boolean operators && and || and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating a || b, if a evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression b. Similarly, evaluating a && b first evaluates a, and if the result is zero, returns 0 without evaluating b.\n\n\n\n\nThe logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.\n \n\nreal step(real x) Return 1 if x is positive and 0 otherwise. \\[\\begin{equation*} \\text{step}(x) =\n\\begin{cases} 0 & \\text{if } x &lt; 0 \\\\ 1 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nThe step function is often used in BUGS to perform conditional operations. For instance, step(a-b) evaluates to 1 if a is greater than b and evaluates to 0 otherwise. step is a step-like functions; see the warning in section step functions applied to expressions dependent on parameters.\nAvailable since 2.0\n \n\nint is_inf(real x) Return 1 if x is infinite (positive or negative) and 0 otherwise.\nAvailable since 2.5\n \n\nint is_nan(real x) Return 1 if x is NaN and 0 otherwise.\nAvailable since 2.5\nCare must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section step-like functions for details.\n\n\n\n\nThe arithmetic operators are presented using C++ notation. For instance operator+(x,y) refers to the binary addition operator and operator-(x) to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as x + y and -x, respectively.\n\n\n \n\nreal operator+(real x, real y) Return the sum of x and y. \\[\\begin{equation*} (x + y) = \\text{operator+}(x,y) = x+y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator-(real x, real y) Return the difference between x and y. \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x,y) = x - y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator*(real x, real y) Return the product of x and y. \\[\\begin{equation*} (x * y) = \\text{operator*}(x,y) = xy\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator/(real x, real y) Return the quotient of x and y. \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator^(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} (x^\\mathrm{\\wedge}y) =\n\\text{operator}^\\mathrm{\\wedge}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.5\n\n\n\n \n\nreal operator-(real x) Return the negation of the subtrahend x. \\[\\begin{equation*} \\text{operator-}(x) = (-x)\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of reals, -x is the same shape array where each individual number is negated.\nAvailable since 2.31\n \n\nreal operator+(real x) Return the value of x. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n\nWarning: These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is real vector, row_vector, matrix, or an array of those types, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nreal fdim(real x, real y) Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. \\[\\begin{equation*} \\text{fdim}(x,y) = \\begin{cases} x-y &\n\\text{if } x \\geq y \\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fdim(T1 x, T2 y) Vectorized implementation of the fdim function\nAvailable since 2.25\n\n\n\n \n\nreal fmin(real x, real y) Return the minimum of x and y; see warning above. \\[\\begin{equation*} \\text{fmin}(x,y) = \\begin{cases} x &\n\\text{if } x \\leq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmin(T1 x, T2 y) Vectorized implementation of the fmin function\nAvailable since 2.25\n \n\nreal fmax(real x, real y) Return the maximum of x and y; see warning above. \\[\\begin{equation*} \\text{fmax}(x,y) = \\begin{cases} x &\n\\text{if } x \\geq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmax(T1 x, T2 y) Vectorized implementation of the fmax function\nAvailable since 2.25\n\n\n\n \n\nreal fmod(real x, real y) Return the real value remainder after dividing x by y; see warning above. \\[\\begin{equation*} \\text{fmod}(x,y) = x - \\left\\lfloor \\frac{x}{y} \\right\\rfloor \\, y \\end{equation*}\\] The operator \\(\\lfloor u \\rfloor\\) is the floor operation; see below.\nAvailable since 2.0\n \n\nR fmod(T1 x, T2 y) Vectorized implementation of the fmod function\nAvailable since 2.25\n\n\n\nWarning: Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.\nThe rounding functions cannot be used as indices to arrays because they return real values. For operations over data or in the generated quantities block, the to_int() function can be used.\n \n\nR floor(T x) The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR ceil(T x) The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR round(T x) The nearest integer to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR trunc(T x) The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n\n\n\n\n \n\nR sqrt(T x) The square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR cbrt(T x) The cube root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR square(T x) The square of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp(T x) The natural exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp2(T x) The base-2 exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log(T x) The natural logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log2(T x) The base-2 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log10(T x) The base-10 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal pow(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} \\text{pow}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.0\n \n\nR pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.25\n \n\nR inv(T x) The inverse of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_sqrt(T x) The inverse of the square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_square(T x) The inverse of the square of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal hypot(real x, real y) Return the length of the hypotenuse of a right triangle with sides of length x and y. \\[\\begin{equation*} \\text{hypot}(x,y) = \\begin{cases} \\sqrt{x^2+y^2} &\n\\text{if } x,y\\geq 0 \\\\ \\textrm{NaN} & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR hypot(T1 x, T2 y) Vectorized implementation of the hypot function\nAvailable since 2.25\n \n\nR cos(T x) The cosine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sin(T x) The sine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tan(T x) The tangent of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acos(T x) The principal arc (inverse) cosine (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR asin(T x) The principal arc (inverse) sine (in radians) of x\nAvailable since 2.0\n \n\nR atan(T x) The principal arc (inverse) tangent (in radians) of x, with values from \\(-\\pi/2\\) to \\(\\pi/2\\)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atan2(T y, T x) Return the principal arc (inverse) tangent (in radians) of y divided by x, \\[\\begin{equation*} \\text{atan2}(y, x) = \\arctan\\left(\\frac{y}{x}\\right) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.34\n\n\n\n \n\nR cosh(T x) The hyperbolic cosine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sinh(T x) The hyperbolic sine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tanh(T x) The hyperbolic tangent of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acosh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR asinh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atanh(T x) The inverse hyperbolic tangent (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\nThe following functions are commonly used as link functions in generalized linear models. The function \\(\\Phi\\) is also commonly used as a link function (see section probability-related functions).\n \n\nR logit(T x) The log odds, or logit, function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_logit(T x) The logistic sigmoid function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_cloglog(T x) The inverse of the complementary log-log function applied to x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n\n\nThe error function erf is related to the standard normal cumulative distribution function \\(\\Phi\\) by scaling. See section normal distribution for the general normal cumulative distribution function (and its complement).\n \n\nR erf(T x) The error function, also known as the Gauss error function, of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR erfc(T x) The complementary error function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_erfc(T x) The inverse of the complementary error function of x\nAvailable since 2.29, vectorized in 2.29\n \n\nR Phi(T x) The standard normal cumulative distribution function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_Phi(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The details of the algorithm can be found in (Wichura 1988). Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.\nAvailable since 2.0, vectorized in 2.13\n \n\nR Phi_approx(T x) The fast approximation of the unit (may replace Phi for probit regression with maximum absolute error of 0.00014, see (Bowling et al. 2009) for details)\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal binary_log_loss(int y, real y_hat) Return the log loss function for for predicting \\(\\hat{y} \\in [0,1]\\) for boolean outcome \\(y \\in \\{0,1\\}\\). \\[\\begin{equation*}\n\\mathrm{binary\\_log\\_loss}(y,\\hat{y}) = \\begin{cases} -\\log \\hat{y} &\n\\text{if } y = 1\\\\ -\\log (1 - \\hat{y}) & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR binary_log_loss(T1 x, T2 y) Vectorized implementation of the binary_log_loss function\nAvailable since 2.25\n \n\nreal owens_t(real h, real a) Return the Owen’s T function for the probability of the event \\(X &gt; h\\) and \\(0&lt;Y&lt;aX\\) where X and Y are independent standard normal random variables. \\[\\begin{equation*} \\mathrm{owens\\_t}(h,a) = \\frac{1}{2\\pi} \\int_0^a\n\\frac{\\exp(-\\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \\end{equation*}\\]\nAvailable since 2.25\n \n\nR owens_t(T1 x, T2 y) Vectorized implementation of the owens_t function\nAvailable since 2.25\n\n\n\n\n \n\nreal beta(real alpha, real beta) Return the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.25\n \n\nR beta(T1 x, T2 y) Vectorized implementation of the beta function\nAvailable since 2.25\n \n\nreal inc_beta(real alpha, real beta, real x) Return the regularized incomplete beta function up to x applied to alpha and beta. See section appendix for a definition.\nAvailable since 2.10\n \n\nreal inv_inc_beta(real alpha, real beta, real p) Return the inverse of the regularized incomplete beta function. The return value x is the value that solves p = inc_beta(alpha, beta, x). See section appendix for a definition of the inc_beta.\nAvailable since 2.30\n \n\nreal lbeta(real alpha, real beta) Return the natural logarithm of the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). \\[\\begin{equation*}\n\\text{lbeta}(\\alpha,\\beta) = \\log \\Gamma(\\alpha) + \\log \\Gamma(\\beta) - \\log \\Gamma(\\alpha+\\beta)\n\\end{equation*}\\] See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.0\n \n\nR lbeta(T1 x, T2 y) Vectorized implementation of the lbeta function\nAvailable since 2.25\n \n\nR tgamma(T x) The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that \\(\\Gamma(n+1) = n!\\). See for a full definition of \\(\\Gamma(x)\\). The function is defined for positive numbers and non-integral negative numbers,\nAvailable since 2.0, vectorized in 2.13\n \n\nR lgamma(T x) The natural logarithm of the gamma function applied to x,\nAvailable since 2.0, vectorized in 2.15\n \n\nR digamma(T x) The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers\nAvailable since 2.0, vectorized in 2.13\n \n\nR trigamma(T x) The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function\nAvailable since 2.0, vectorized in 2.13\n \n\nreal lmgamma(int n, real x) Return the natural logarithm of the multivariate gamma function \\(\\Gamma_n\\) with n dimensions applied to x. \\[\\begin{equation*}\n\\text{lmgamma}(n,x) =\n\\begin{cases} \\frac{n(n-1)}{4} \\log \\pi + \\sum_{j=1}^n \\log \\Gamma\\left(x + \\frac{1 - j}{2}\\right)\n& \\text{if } x\\not\\in \\{\\dots,-3,-2,-1,0\\}\\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR lmgamma(T1 x, T2 y) Vectorized implementation of the lmgamma function\nAvailable since 2.25\n \n\nreal gamma_p(real a, real z) Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_p}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_0^zt^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_p(T1 x, T2 y) Vectorized implementation of the gamma_p function\nAvailable since 2.25\n \n\nreal gamma_q(real a, real z) Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_q}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_z^\\infty t^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\[6pt] \\textrm{error} & \\text{otherwise}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_q(T1 x, T2 y) Vectorized implementation of the gamma_q function\nAvailable since 2.25\n \n\nint choose(int x, int y) Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” In its the antilog of the lchoose function but returns an integer rather than a real number with no non-zero decimal places. For \\(0 \\leq y \\leq x\\), the binomial coefficient function can be defined via the factorial function \\[\\begin{equation*}\n\\text{choose}(x,y) = \\frac{x!}{\\left(y!\\right)\\left(x - y\\right)!}.\n\\end{equation*}\\]\nAvailable since 2.14\n \n\nR choose(T1 x, T2 y) Vectorized implementation of the choose function\nAvailable since 2.25\n \n\nreal bessel_first_kind(int v, real x) Return the Bessel function of the first kind with order v applied to x. \\[\\begin{equation*}\n\\mathrm{bessel\\_first\\_kind}(v,x) = J_v(x),\n\\end{equation*}\\] where \\[\\begin{equation*}\nJ_v(x)=\\left(\\frac{1}{2}x\\right)^v \\sum_{k=0}^\\infty\n\\frac{\\left(-\\frac{1}{4}x^2\\right)^k}{k!\\, \\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_first_kind(T1 x, T2 y) Vectorized implementation of the bessel_first_kind function\nAvailable since 2.25\n \n\nreal bessel_second_kind(int v, real x) Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For \\(x,v &gt; 0\\), \\[\\begin{equation*}\n\\mathrm{bessel\\_second\\_kind}(v,x) =\n\\begin{cases} Y_v(x) & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\nY_v(x)=\\frac{J_v(x)\\cos(v\\pi)-J_{-v}(x)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_second_kind(T1 x, T2 y) Vectorized implementation of the bessel_second_kind function\nAvailable since 2.25\n \n\nreal modified_bessel_first_kind(int v, real z) Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_first\\_kind}(v,z) = I_v(z)\n\\end{equation*}\\] where \\[\\begin{equation*}\n{I_v}(z) = \\left(\\frac{1}{2}z\\right)^v\\sum_{k=0}^\\infty \\frac{\\left(\\frac{1}{4}z^2\\right)^k}{k!\\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_first_kind function\nAvailable since 2.25\n \n\nreal log_modified_bessel_first_kind(real v, real z) Return the log of the modified Bessel function of the first kind. v does not have to be an integer.\nAvailable since 2.26\n \n\nR log_modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the log_modified_bessel_first_kind function\nAvailable since 2.26\n \n\nreal modified_bessel_second_kind(int v, real z) Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_second\\_kind}(v,z) =\n\\begin{cases} K_v(z) &  \\text{if } z &gt; 0 \\\\ \\textrm{error} & \\text{if } z \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} {K_v}(z) = \\frac{\\pi}{2}\\cdot\\frac{I_{-v}(z) - I_{v}(z)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_second_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_second_kind function\nAvailable since 2.25\n \n\nreal falling_factorial(real x, real n) Return the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*}\n\\mathrm{falling\\_factorial}(x,n) =\n\\begin{cases} (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\n(x)_n=\\frac{\\Gamma(x+1)}{\\Gamma(x-n+1)}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR falling_factorial(T1 x, T2 y) Vectorized implementation of the falling_factorial function\nAvailable since 2.25\n \n\nreal lchoose(real x, real y) Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For \\(0 \\leq y \\leq x\\), \\[\\begin{equation*} \\mathrm{binomial\\_coefficient\\_log}(x,y) =\n\\log\\Gamma(x+1) - \\log\\Gamma(y+1) - \\log\\Gamma(x-y+1). \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lchoose(T1 x, T2 y) Vectorized implementation of the lchoose function\nAvailable since 2.29\n \n\nreal log_falling_factorial(real x, real n) Return the log of the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_falling\\_factorial}(x,n) =\n\\begin{cases} \\log (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal rising_factorial(real x, int n) Return the rising factorial of x with power n defined for positive x and integer n. \\[\\begin{equation*}\n\\mathrm{rising\\_factorial}(x,n) = \\begin{cases} x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} x^{(n)}=\\frac{\\Gamma(x+n)}{\\Gamma(x)} \\end{equation*}\\]\nAvailable since 2.20\n \n\nR rising_factorial(T1 x, T2 y) Vectorized implementation of the rising_factorial function\nAvailable since 2.25\n \n\nreal log_rising_factorial(real x, real n) Return the log of the rising factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_rising\\_factorial}(x,n) =\n\\begin{cases} \\log x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR log_rising_factorial(T1 x, T2 y) Vectorized implementation of the log_rising_factorial function\nAvailable since 2.25\n\n\n\nThe functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.\n \n\nR expm1(T x) The natural exponential of x minus 1\nAvailable since 2.0, vectorized in 2.13\n \n\nreal fma(real x, real y, real z) Return z plus the result of x multiplied by y. \\[\\begin{equation*} \\text{fma}(x,y,z) =\n(x \\times y) + z \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal ldexp(real x, int y) Return the product of x and two raised to the y power. \\[\\begin{equation*}\n\\text{ldexp}(x,y) = x 2^y  \\end{equation*}\\]\nAvailable since 2.25\n \n\nR ldexp(T1 x, T2 y) Vectorized implementation of the ldexp function\nAvailable since 2.25\n \n\nreal lmultiply(real x, real y) Return the product of x and the natural logarithm of y. \\[\\begin{equation*}\n\\text{lmultiply}(x,y) = \\begin{cases} 0 & \\text{if } x = y = 0 \\\\ x\n\\log y & \\text{if } x, y \\neq 0 \\\\ \\text{NaN} & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lmultiply(T1 x, T2 y) Vectorized implementation of the lmultiply function\nAvailable since 2.25\n \n\nR log1p(T x) The natural logarithm of 1 plus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m(T x) The natural logarithm of 1 minus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1p_exp(T x) The natural logarithm of one plus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m_exp(T x) The logarithm of one minus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal log_diff_exp(real x, real y) Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_diff\\_exp}(x,y) = \\begin{cases} \\log(\\exp(x)-\\exp(y)) &\n\\text{if } +\\infty &gt; x \\ge y \\\\[6pt]\n\\textrm{NaN} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nWhen x is equal to y, log_diff_exp(x, y) returns \\(-\\infty\\), consistent with log(0) returning \\(-\\infty\\). This includes the case in which x and y are both equal to \\(-\\infty\\), which corresponds to log(0 - 0) because exp(negative_infinity()) returns 0.\nAvailable since 2.0\n \n\nR log_diff_exp(T1 x, T2 y) Vectorized implementation of the log_diff_exp function\nAvailable since 2.25\n \n\nreal log_mix(real theta, real lp1, real lp2) Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by \\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda_1, \\lambda_2) & = & \\log \\!\\left(\n\\theta \\exp(\\lambda_1) + \\left( 1 - \\theta \\right) \\exp(\\lambda_2)\n\\right) \\\\[3pt] & = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) +\n\\lambda_1, \\ \\log(1 - \\theta) + \\lambda_2\\right). \\end{eqnarray*}\\]\nAvailable since 2.6\n \n\nR log_mix(T1 thetas, T2 lps)\nCalculates the log mixture density given thetas, mixing proportions which should be between 0 and 1 and sum to 1, and lps, log densities. The lps variable must be either a 1-d container of the same length as thetas, or an array of such.\n\\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda)\n& = & \\log \\!\\left( \\sum_{n=1}^N \\theta_n * \\exp(\\lambda_n) \\right) \\\\[3pt]\n& = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) + \\lambda\\right).\n\\end{eqnarray*}\\]\nThis is a generalization of the above signature of three arguments to more than two densities. For example, log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2}).\nAvailable since 2.26\n \n\nR log_sum_exp(T1 x, T2 y) Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_sum\\_exp}(x,y) = \\log(\\exp(x)+\\exp(y)) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.33\n \n\nR log_inv_logit(T x) The natural logarithm of the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log_inv_logit_diff(T1 x, T2 y) The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y\nAvailable since 2.25\n \n\nR log1m_inv_logit(T x) The natural logarithm of 1 minus the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nR lambert_w0(T x) Implementation of the \\(W_0\\) branch of the Lambert W function, i.e., solution to the function \\(W_0(x) \\exp^{ W_0(x)} = x\\)\nAvailable since 2.25\n \n\nR lambert_wm1(T x) Implementation of the \\(W_{-1}\\) branch of the Lambert W function, i.e., solution to the function \\(W_{-1}(x) \\exp^{W_{-1}(x)} = x\\)\nAvailable since 2.25\n\n\n\nHypergeometric functions refer to a power series of the form \\[\\begin{equation*}\n_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \\sum_{n=0}^\\infty \\frac{(a_1)_n\\cdot\\cdot\\cdot(a_p)_n}{(b_1)_n\\cdot\\cdot\\cdot(b_q)_n} \\frac{z^n}{n!}\n\\end{equation*}\\] where \\((a)_n\\) is the Pochhammer symbol defined as \\((a)_n = \\frac{\\Gamma(a+n)}{\\Gamma(a)}\\).\nThe gradients of the hypergeometric function are given by: \\[\\begin{equation*}\n\\frac{\\partial }{\\partial a_1} =\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+a_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n    - \\psi\\left(a_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)\n\\end{equation*}\\] \\[\\begin{equation*}\n\\frac{\\partial }{\\partial b_1} =\n  \\psi\\left(b_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+b_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n\\end{equation*}\\] \\[\\begin{equation*}\n  \\frac{\\partial }{\\partial z} =\n  \\frac{\\prod_{j=1}^{p}a_j}{\\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)\n\\end{equation*}\\]\nStan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.\n \n\nreal hypergeometric_1F0(real a, real z) Special case of the hypergeometric function with \\(p=1\\) and \\(q=0\\).\nAvailable since 2.37\n \n\nreal hypergeometric_2F1(real a1, real a2, real b1, real z) Special case of the hypergeometric function with \\(p=2\\) and \\(q=1\\). If the function does not meet convergence criteria for given inputs, the function will attempt to apply Euler’s transformation to improve convergence: \\[\\begin{equation*}\n{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \\frac{z}{z-1})\\cdot(1-z)^{-a_2}\n\\end{equation*}\\]\nAvailable since 2.37\n \n\nreal hypergeometric_3F2(T1 a, T2 b, real z) Special case of the hypergeometric function with \\(p=3\\) and \\(q=2\\), where a and b are vectors of length 3 and 2, respectively.\nAvailable since 2.37\n \n\nreal hypergeometric_pFq(T1 a, T2 b, real z) Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.\nAvailable since 2.37",
+    "text": "For most of the univariate probability functions, there is a corresponding cumulative distribution function, log cumulative distribution function, and log complementary cumulative distribution function.\nFor a univariate random variable \\(Y\\) with probability function \\(p_Y(y \\, | \\, \\theta)\\), the cumulative distribution function (CDF) \\(F_Y\\) is defined by \\[\\begin{equation*}\nF_Y(y) \\ = \\ \\text{Pr}[Y \\le y] \\ = \\ \\int_{-\\infty}^y p(y\\, | \\, \\theta) \\ \\text{d}y.\n\\end{equation*}\\] The complementary cumulative distribution function (CCDF) is defined as \\[\\begin{equation*}\n\\text{Pr}[Y &gt; y] \\ = \\ 1 - F_Y(y).\n\\end{equation*}\\] The reason to use CCDFs instead of CDFs in floating-point arithmetic is that it is possible to represent numbers very close to 0 (the closest you can get is roughly \\(10^{-300}\\)), but not numbers very close to 1 (the closest you can get is roughly \\(1 - 10^{-15}\\)).\nIn Stan, there is a cumulative distribution function for each probability function. For instance, normal_cdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\ \\text{d}y.\n\\end{equation*}\\] There are also log forms of the CDF and CCDF for most univariate distributions. For example, normal_lcdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right)\n\\end{equation*}\\] and normal_lccdf(y | mu, sigma) is defined by \\[\\begin{equation*}\n\\log \\left( 1 - \\int_{-\\infty}^y \\text{Normal}(y \\, | \\, \\mu, \\sigma) \\   \\text{d}y \\right).\n\\end{equation*}\\]",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#fun-vectorization",
-    "href": "functions-reference/real-valued_basic_functions.html#fun-vectorization",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/conventions_for_probability_functions.html#vectorization",
+    "href": "functions-reference/conventions_for_probability_functions.html#vectorization",
+    "title": "Conventions for Probability Functions",
     "section": "",
-    "text": "Although listed in this chapter, many of Stan’s built-in functions are vectorized so that they may be applied to any argument type. The vectorized form of these functions is not any faster than writing an explicit loop that iterates over the elements applying the function—it’s just easier to read and write and less error prone.\n\n\nMany of Stan’s unary functions can be applied to any argument type. For example, the exponential function, exp, can be applied to real arguments or arrays of real arguments. Other than for integer arguments, the result type is the same as the argument type, including dimensionality and size. Integer arguments are first promoted to real values, but the result will still have the same dimensionality and size as the argument.\n\n\nWhen applied to a simple real value, the result is a real value. When applied to arrays, vectorized functions like exp() are defined elementwise. For example,\n // declare some variables for arguments\n real x0;\n array[5] real x1;\n array[4, 7] real x2;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = exp(x0);\n y1 = exp(x1);\n y2 = exp(x2);\nWhen exp is applied to an array, it applies elementwise. For example, the statement above,\n y2 = exp(x2);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = exp(x2[i, j]);\n  }\n}\n\n\n\nVectorized functions also apply elementwise to vectors and matrices. For example,\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = exp(xv);\n yrv = exp(xrv);\n ym = exp(xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = exp(u);\nAfter this has been executed, z[i, j, k] will be equal to exp(u[i, j, k]).\n\n\n\nInteger arguments are promoted to real values in vectorized unary functions. Thus if n is of type int, exp(n) is of type real. Arrays work the same way, so that if n2 is a one dimensional array of integers, then exp(n2) will be a one-dimensional array of reals with the same number of elements as n2. For example,\n array[23] int n1;\n array[23] real z1;\n z1 = exp(n1);\nIt would be illegal to try to assign exp(n1) to an array of integers; the return type is a real array.\n\n\n\n\nLike the unary functions, many of Stan’s binary functions have been vectorized, and can be applied elementwise to combinations of both scalars or container types.\n\n\nWhen applied to two scalar values, the result is a scalar value. When applied to two arrays, or combination of a scalar value and an array, vectorized functions like pow() are defined elementwise. For example,\n // declare some variables for arguments\n real x00;\n real x01;\n array[5] real x10;\n array[5]real x11;\n array[4, 7] real x20;\n array[4, 7] real x21;\n // ...\n // declare some variables for results\n real y0;\n array[5] real y1;\n array[4, 7] real y2;\n // ...\n // calculate and assign results\n y0 = pow(x00, x01);\n y1 = pow(x10, x11);\n y2 = pow(x20, x21);\nWhen pow is applied to two arrays, it applies elementwise. For example, the statement above,\n y2 = pow(x20, x21);\nproduces the same result for y2 as the explicit loop\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x21[i, j]);\n  }\n}\nAlternatively, if a combination of an array and a scalar are provided, the scalar value is broadcast to be applied to each value of the array. For example, the following statement:\ny2 = pow(x20, x00);\nproduces the same result for y2 as the explicit loop:\nfor (i in 1:4) {\n  for (j in 1:7) {\n    y2[i, j] = pow(x20[i, j], x00);\n  }\n}\n\n\n\nVectorized binary functions also apply elementwise to vectors and matrices, and to combinations of these with scalar values. For example,\n real x00;\n vector[5] xv00;\n vector[5] xv01;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv;\n row_vector[7] yrv;\n matrix[10, 20] ym;\n\n yv = pow(xv00, xv01);\n yrv = pow(xrv, x00);\n ym = pow(x00, xm);\nArrays of vectors and matrices work the same way. For example,\n array[12] matrix[17, 93] u;\n\n array[12] matrix[17, 93] z;\n\n z = pow(u, x00);\nAfter this has been executed, z[i, j, k] will be equal to pow(u[i, j, k], x00).\n\n\n\nVectorised binary functions require that both inputs, unless one is a real, be containers of the same type and size. For example, the following statements are legal:\n vector[5] xv;\n row_vector[7] xrv;\n matrix[10, 20] xm;\n\n vector[5] yv = pow(xv, xv)\n row_vector[7] yrv = pow(xrv, xrv)\n matrix[10, 20] = pow(xm, xm)\nBut the following statements are not:\n vector[5] xv;\n vector[7] xv2;\n row_vector[5] xrv;\n\n // Cannot mix different types\n vector[5] yv = pow(xv, xrv)\n\n // Cannot mix different sizes of the same type\n vector[5] yv = pow(xv, xv2)\nWhile the vectorized binary functions generally require the same input types, the only exception to this is for binary functions that require one input to be an integer and the other to be a real (e.g., bessel_first_kind). For these functions, one argument can be a container of any type while the other can be an integer array, as long as the dimensions of both are the same. For example, the following statements are legal:\n vector[5] xv;\n matrix[5, 5] xm;\n array[5] int xi;\n array[5, 5] int xii;\n\n vector[5] yv = bessel_first_kind(xi, xv);\n matrix[5, 5] ym = bessel_first_kind(xii, xm);\nWhereas these are not:\n vector[5] xv;\n matrix[5, 5] xm;\n array[7] int xi;\n\n // Dimensions of containers do not match\n vector[5] yv = bessel_first_kind(xi, xv);\n\n // Function requires first argument be an integer type\n matrix[5, 5] ym = bessel_first_kind(xm, xm);",
+    "text": "Stan’s univariate log probability functions, including the log density functions, log mass functions, log CDFs, and log CCDFs, all support vectorized function application, with results defined to be the sum of the elementwise application of the function. Some of the PRNG functions support vectorization, see section vectorized PRNG functions for more details.\nIn all cases, matrix operations are at least as fast and usually faster than loops and vectorized log probability functions are faster than their equivalent form defined with loops. This isn’t because loops are slow in Stan, but because more efficient automatic differentiation can be used. The efficiency comes from the fact that a vectorized log probability function only introduces one new node into the expression graph, thus reducing the number of virtual function calls required to compute gradients in C++, as well as from allowing caching of repeated computations.\nStan also overloads the multivariate normal distribution, including the Cholesky-factor form, allowing arrays of row vectors or vectors for the variate and location parameter. This is a huge savings in speed because the work required to solve the linear system for the covariance matrix is only done once.\nStan also overloads some scalar functions, such as log and exp, to apply to vectors (arrays) and return vectors (arrays). These vectorizations are defined elementwise and unlike the probability functions, provide only minimal efficiency speedups over repeated application and assignment in a loop.\n\n\n\n\nThe normal probability function is specified with the signature\n normal_lpdf(reals | reals, reals);\nThe pseudotype reals is used to indicate that an argument position may be vectorized. Argument positions declared as reals may be filled with a real, a one-dimensional array, a vector, or a row-vector. If there is more than one array or vector argument, their types can be anything but their size must match. For instance, it is legal to use normal_lpdf(row_vector | vector, real) as long as the vector and row vector have the same size.\n\n\n\nThe multivariate normal distribution accepting vector or array of vector arguments is written as\n multi_normal_lpdf(vectors | vectors, matrix);\nThese arguments may be row vectors, column vectors, or arrays of row vectors or column vectors.\n\n\n\nThe pseudotype ints is used for vectorized integer arguments. Where it appears either an integer or array of integers may be used.\n\n\n\n\nThe result of a vectorized log probability function is equivalent to the sum of the evaluations on each element. Any non-vector argument, namely real or int, is repeated. For instance, if y is a vector of size N, mu is a vector of size N, and sigma is a scalar, then\n ll = normal_lpdf(y | mu, sigma);\nis just a more efficient way to write\n ll = 0;\n for (n in 1:N) {\n   ll = ll + normal_lpdf(y[n] | mu[n], sigma);\n }\nWith the same arguments, the vectorized sampling statement\n y ~ normal(mu, sigma);\nhas the same effect on the total log probability as\n for (n in 1:N) {\n   y[n] ~ normal(mu[n], sigma);\n }\n\n\n\nSome PRNG functions accept sequences as well as scalars as arguments. Such functions are indicated by argument pseudotypes reals or ints. In cases of sequence arguments, the output will also be a sequence. For example, the following is allowed in the transformed data and generated quantities blocks.\n vector[3] mu = // ...\n array[3] real x = normal_rng(mu, 3);\n\n\nIn the case of PRNG functions, arguments marked ints may be integers or integer arrays, whereas arguments marked reals may be integers or reals, integer or real arrays, vectors, or row vectors.\n\n\n\n\n\n\n\npseudotype\nallowable PRNG arguments\n\n\n\n\nints\nint, array[] int\n\n\nreals\nint, array[] int, real, array[] real, vector, row_vector\n\n\n\n\n\n\nIn general, if there are multiple non-scalar arguments, they must all have the same dimensions, but need not have the same type. For example, the normal_rng function may be called with one vector argument and one real array argument as long as they have the same number of elements.\n vector[3] mu = // ...\n array[3] real sigma = // ...\n array[3] real x = normal_rng(mu, sigma);\n\n\n\nThe result of a vectorized PRNG function depends on the size of the arguments and the distribution’s support. If all arguments are scalars, then the return type is a scalar. For a continuous distribution, if there are any non-scalar arguments, the return type is a real array (array[] real) matching the size of any of the non-scalar arguments, as all non-scalar arguments must have matching size. Discrete distributions return ints and continuous distributions return reals, each of appropriate size. The symbol R denotes such a return type.",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Conventions for Probability Functions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#built-in-constants",
-    "href": "functions-reference/real-valued_basic_functions.html#built-in-constants",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/covariance_matrix_distributions.html",
+    "href": "functions-reference/covariance_matrix_distributions.html",
+    "title": "Covariance Matrix Distributions",
     "section": "",
-    "text": "Constants are represented as functions with no arguments and must be called as such. For instance, the mathematical constant \\(\\pi\\) must be written in a Stan program as pi().\n \n\nreal pi() \\(\\pi\\), the ratio of a circle’s circumference to its diameter\nAvailable since 2.0\n \n\nreal e() \\(e\\), the base of the natural logarithm\nAvailable since 2.0\n \n\nreal sqrt2() The square root of 2\nAvailable since 2.0\n \n\nreal log2() The natural logarithm of 2\nAvailable since 2.0\n \n\nreal log10() The natural logarithm of 10\nAvailable since 2.0",
+    "text": "The covariance matrix distributions have support on symmetric, positive-definite \\(K \\times K\\) matrices or their Cholesky factors (square, lower triangular matrices with positive diagonal elements).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{Wishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}}\n\\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{-\\nu/2} \\ \\left| W \\right|^{(\\nu - K - 1)/2}\n\\ \\exp \\! \\left(- \\frac{1}{2} \\ \\text{tr}\\left( S^{-1} W \\right) \\right) \\! ,\n\\end{equation*}\\] where \\(\\text{tr}()\\) is the matrix trace function, and \\(\\Gamma_K()\\) is the multivariate Gamma function, \\[\\begin{equation*}\n\\Gamma_K(x) = \\frac{1}{\\pi^{K(K-1)/4}} \\ \\prod_{k=1}^K \\Gamma \\left( x + \\frac{1 - k}{2} \\right) \\!.\n\\end{equation*}\\]\n\n\n\nW ~ wishart(nu, Sigma)\nIncrement target log probability density with wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix wishart_rng(real nu, matrix Sigma) Generate a Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0\n\n\n\n\nThe Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{WishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{Wishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in \\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is \\[\\begin{equation*}\n\\text{WishartCholesky}(L_W \\mid \\nu,L_S)\n= \\text{Wishart}(L_W L_W^{\\top} \\mid \\nu,L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of a Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks\nAvailable since 2.30\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{InvWishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}} \\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{\\nu/2} \\ \\left| W \\right|^{-(\\nu + K + 1)/2}\n\\ \\exp \\! \\left( - \\frac{1}{2} \\ \\text{tr}(SW^{-1}) \\right) \\! .\n\\end{equation*}\\]\n\n\n\nW ~ inv_wishart(nu, Sigma)\nIncrement target log probability density with inv_wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal inv_wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix inv_wishart_rng(real nu, matrix Sigma) Generate an inverse Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0\n\n\n\n\nThe Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{InvWishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{InvWishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in\n\\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is \\[\\begin{equation*}\n\\text{InvWishartCholesky}(L_W \\mid \\nu,L_S) =\n\\text{InvWishart}(L_WL_W^{\\top} \\mid \\nu, L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal inv_wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal inv_wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix inv_wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#special-values",
-    "href": "functions-reference/real-valued_basic_functions.html#special-values",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/covariance_matrix_distributions.html#wishart-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#wishart-distribution",
+    "title": "Covariance Matrix Distributions",
     "section": "",
-    "text": "real not_a_number() Not-a-number, a special non-finite real value returned to signal an error\nAvailable since 2.0\n \n\nreal positive_infinity() Positive infinity, a special non-finite real value larger than all finite numbers\nAvailable since 2.0\n \n\nreal negative_infinity() Negative infinity, a special non-finite real value smaller than all finite numbers\nAvailable since 2.0\n \n\nreal machine_precision() The smallest number \\(x\\) such that \\((x + 1) \\neq 1\\) in floating-point arithmetic on the current hardware platform\nAvailable since 2.0",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{Wishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}}\n\\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{-\\nu/2} \\ \\left| W \\right|^{(\\nu - K - 1)/2}\n\\ \\exp \\! \\left(- \\frac{1}{2} \\ \\text{tr}\\left( S^{-1} W \\right) \\right) \\! ,\n\\end{equation*}\\] where \\(\\text{tr}()\\) is the matrix trace function, and \\(\\Gamma_K()\\) is the multivariate Gamma function, \\[\\begin{equation*}\n\\Gamma_K(x) = \\frac{1}{\\pi^{K(K-1)/4}} \\ \\prod_{k=1}^K \\Gamma \\left( x + \\frac{1 - k}{2} \\right) \\!.\n\\end{equation*}\\]\n\n\n\nW ~ wishart(nu, Sigma)\nIncrement target log probability density with wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix wishart_rng(real nu, matrix Sigma) Generate a Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#get-log-prob",
-    "href": "functions-reference/real-valued_basic_functions.html#get-log-prob",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/covariance_matrix_distributions.html#wishart-cholesky-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#wishart-cholesky-distribution",
+    "title": "Covariance Matrix Distributions",
     "section": "",
-    "text": "The basic purpose of a Stan program is to compute a log probability function and its derivatives. The log probability function in a Stan model outputs the log density on the unconstrained scale. A log probability accumulator starts at zero and is then incremented in various ways by a Stan program. The variables are first transformed from unconstrained to constrained, and the log Jacobian determinant added to the log probability accumulator. Then the model block is executed on the constrained parameters, with each sampling statement (~) and log probability increment statement (increment_log_prob) adding to the accumulator. At the end of the model block execution, the value of the log probability accumulator is the log probability value returned by the Stan program.\nStan provides a special built-in function target() that takes no arguments and returns the current value of the log probability accumulator. This function is primarily useful for debugging purposes, where for instance, it may be used with a print statement to display the log probability accumulator at various stages of execution to see where it becomes ill defined.\n \n\nreal target() Return the current value of the log probability accumulator.\nAvailable since 2.10\ntarget acts like a function ending in _lp, meaning that it may only may only be used in the model block.",
+    "text": "The Cholesky parameterization of the Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{WishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{Wishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in \\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized Wishart density is \\[\\begin{equation*}\n\\text{WishartCholesky}(L_W \\mid \\nu,L_S)\n= \\text{Wishart}(L_W L_W^{\\top} \\mid \\nu,L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of a Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks\nAvailable since 2.30",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#logical-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#logical-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-distribution",
+    "title": "Covariance Matrix Distributions",
     "section": "",
-    "text": "Like C++, BUGS, and R, Stan uses 0 to encode false, and 1 to encode true. Stan supports the usual boolean comparison operations and boolean operators. These all have the same syntax and precedence as in C++; for the full list of operators and precedences, see the reference manual.\n\n\nAll comparison operators return boolean values, either 0 or 1. Each operator has two signatures, one for integer comparisons and one for floating-point comparisons. Comparing an integer and real value is carried out by first promoting the integer value.\n \n\nint operator&lt;(int x, int y)\n \n\nint operator&lt;(real x, real y) Return 1 if x is less than y and 0 otherwise. \\[\\begin{equation*} \\text{operator&lt;}(x,y)\n= \\begin{cases} 1 & \\text{if $x &lt; y$} \\\\ 0 & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&lt;=(int x, int y)\n \n\nint operator&lt;=(real x, real y) Return 1 if x is less than or equal y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&lt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\leq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;(int x, int y)\n \n\nint operator&gt;(real x, real y) Return 1 if x is greater than y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;}(x,y) = \\begin{cases} 1 & \\text{if $x &gt; y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&gt;=(int x, int y)\n \n\nint operator&gt;=(real x, real y) Return 1 if x is greater than or equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator&gt;=}(x,y) = \\begin{cases} 1 & \\text{if $x \\geq y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator==(int x, int y)\n \n\nint operator==(real x, real y) Return 1 if x is equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator==}(x,y) = \\begin{cases} 1 & \\text{if $x = y$} \\\\ 0 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!=(int x, int y)\n \n\nint operator!=(real x, real y) Return 1 if x is not equal to y and 0 otherwise. \\[\\begin{equation*}\n\\text{operator!=}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq y$} \\\\ 0 &\n\\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\nBoolean operators return either 0 for false or 1 for true. Inputs may be any real or integer values, with non-zero values being treated as true and zero values treated as false. These operators have the usual precedences, with negation (not) binding the most tightly, conjunction the next and disjunction the weakest; all of the operators bind more tightly than the comparisons. Thus an expression such as !a && b is interpreted as (!a) && b, and a &lt; b || c &gt;= d && e != f as (a &lt; b) || (((c &gt;= d) && (e != f))).\n \n\nint operator!(int x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0$} \\\\ 1 & \\text{if $x = 0$}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator!(real x) Return 1 if x is zero and 0 otherwise. \\[\\begin{equation*} \\text{operator!}(x) =\n\\begin{cases} 0 & \\text{if $x \\neq 0.0$} \\\\ 1 & \\text{if $x = 0.0$}\n\\end{cases} \\end{equation*}\\] deprecated; - use operator== instead.\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator&&(int x, int y)\nReturn 1 if x is unequal to 0 and y is unequal to 0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\text{ and } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator&&(real x, real y) Return 1 if x is unequal to 0.0 and y is unequal to 0.0. \\[\\begin{equation*}\n\\mathrm{operator\\&\\&}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\text{ and } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n \n\nint operator||(int x, int y) Return 1 if x is unequal to 0 or y is unequal to 0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0$}\n\\textrm{ or } y \\neq 0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator||(real x, real y) Return 1 if x is unequal to 0.0 or y is unequal to 0.0. \\[\\begin{equation*}\n\\text{operator||}(x,y) = \\begin{cases} 1 & \\text{if $x \\neq 0.0$}\n\\textrm{ or } y \\neq 0.0\\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\] deprecated\nAvailable since 2.0, deprecated in 2.31\n\n\nLike in C++, the boolean operators && and || and are implemented to short circuit directly to a return value after evaluating the first argument if it is sufficient to resolve the result. In evaluating a || b, if a evaluates to a value other than zero, the expression returns the value 1 without evaluating the expression b. Similarly, evaluating a && b first evaluates a, and if the result is zero, returns 0 without evaluating b.\n\n\n\n\nThe logical functions introduce conditional behavior functionally and are primarily provided for compatibility with BUGS and JAGS.\n \n\nreal step(real x) Return 1 if x is positive and 0 otherwise. \\[\\begin{equation*} \\text{step}(x) =\n\\begin{cases} 0 & \\text{if } x &lt; 0 \\\\ 1 & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nThe step function is often used in BUGS to perform conditional operations. For instance, step(a-b) evaluates to 1 if a is greater than b and evaluates to 0 otherwise. step is a step-like functions; see the warning in section step functions applied to expressions dependent on parameters.\nAvailable since 2.0\n \n\nint is_inf(real x) Return 1 if x is infinite (positive or negative) and 0 otherwise.\nAvailable since 2.5\n \n\nint is_nan(real x) Return 1 if x is NaN and 0 otherwise.\nAvailable since 2.5\nCare must be taken because both of these indicator functions are step-like and thus can cause discontinuities in gradients when applied to parameters; see section step-like functions for details.",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1,\\infty)\\), and \\(S \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for symmetric and positive-definite \\(W \\in \\mathbb{R}^{K \\times K}\\), \\[\\begin{equation*}\n\\text{InvWishart}(W \\mid \\nu,S) = \\frac{1}{2^{\\nu K / 2}} \\ \\frac{1}{\\Gamma_K \\! \\left( \\frac{\\nu}{2} \\right)}\n\\ \\left| S \\right|^{\\nu/2} \\ \\left| W \\right|^{-(\\nu + K + 1)/2}\n\\ \\exp \\! \\left( - \\frac{1}{2} \\ \\text{tr}(SW^{-1}) \\right) \\! .\n\\end{equation*}\\]\n\n\n\nW ~ inv_wishart(nu, Sigma)\nIncrement target log probability density with inv_wishart_lupdf(W | nu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal inv_wishart_lpdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma.\nAvailable since 2.12\n \n\nreal inv_wishart_lupdf(matrix W | real nu, matrix Sigma) Return the log of the inverse Wishart density for symmetric and positive-definite matrix W given degrees of freedom nu and symmetric and positive-definite scale matrix Sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nmatrix inv_wishart_rng(real nu, matrix Sigma) Generate an inverse Wishart variate with degrees of freedom nu and symmetric and positive-definite scale matrix Sigma; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.0",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#real-valued-arithmetic-operators",
-    "href": "functions-reference/real-valued_basic_functions.html#real-valued-arithmetic-operators",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-cholesky-distribution",
+    "href": "functions-reference/covariance_matrix_distributions.html#inverse-wishart-cholesky-distribution",
+    "title": "Covariance Matrix Distributions",
     "section": "",
-    "text": "The arithmetic operators are presented using C++ notation. For instance operator+(x,y) refers to the binary addition operator and operator-(x) to the unary negation operator. In Stan programs, these are written using the usual infix and prefix notations as x + y and -x, respectively.\n\n\n \n\nreal operator+(real x, real y) Return the sum of x and y. \\[\\begin{equation*} (x + y) = \\text{operator+}(x,y) = x+y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator-(real x, real y) Return the difference between x and y. \\[\\begin{equation*} (x - y) =\n\\text{operator-}(x,y) = x - y \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator*(real x, real y) Return the product of x and y. \\[\\begin{equation*} (x * y) = \\text{operator*}(x,y) = xy\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator/(real x, real y) Return the quotient of x and y. \\[\\begin{equation*} (x / y) = \\text{operator/}(x,y) =\n\\frac{x}{y} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal operator^(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} (x^\\mathrm{\\wedge}y) =\n\\text{operator}^\\mathrm{\\wedge}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.5\n\n\n\n \n\nreal operator-(real x) Return the negation of the subtrahend x. \\[\\begin{equation*} \\text{operator-}(x) = (-x)\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of reals, -x is the same shape array where each individual number is negated.\nAvailable since 2.31\n \n\nreal operator+(real x) Return the value of x. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0",
+    "text": "The Cholesky parameterization of the inverse Wishart distribution uses a Cholesky factor for both the variate and the parameter. If \\(S\\) and \\(W\\) are positive definite matrices with Cholesky factors \\(L_S\\) and \\(L_W\\) (i.e., \\(S = L_S L_S^{\\top}\\) and \\(W = L_W L_W^{\\top}\\)), then the Cholesky parameterization is defined so that \\[\\begin{equation*}\nL_W \\sim \\textrm{InvWishartCholesky}(\\nu, L_S)\n\\end{equation*}\\] if and only if \\[\\begin{equation*}\nW \\sim \\textrm{InvWishart}(\\nu, S).\n\\end{equation*}\\]\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in (K-1, \\infty)\\), and \\(L_S, L_W \\in\n\\mathbb{R}^{K \\times K}\\) are lower triangular matrixes with positive diagonal elements, then the Cholesky parameterized inverse Wishart density is \\[\\begin{equation*}\n\\text{InvWishartCholesky}(L_W \\mid \\nu,L_S) =\n\\text{InvWishart}(L_WL_W^{\\top} \\mid \\nu, L_S L_S^{\\top}) \\, \\left| J_{f^{-1}} \\right|,\n\\end{equation*}\\] where \\(J_{f^{-1}}\\) is the Jacobian of the (inverse) transform of the variate, \\(f^{-1}(L_W) = L_W L_W^{\\top}\\). The log absolute determinant is \\[\\begin{equation*}\n\\log \\left| J_{f^{-1}} \\right|\n= K \\log(2) + \\sum_{k=1}^K (K - k + 1) \\log {(L_W)_{k,\\, k}}.\n\\end{equation*}\\]\nThe probability functions will raise errors if \\(\\nu \\leq K - 1\\) or if \\(L_S\\) and \\(L_W\\) are not Cholesky factors (square, lower-triangular matrices with positive diagonal elements) of the same size.\n\n\n\n \n\nreal inv_wishart_cholesky_lpdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S.\nAvailable since 2.30\n \n\nreal inv_wishart_cholesky_lupdf(matrix L_W | real nu, matrix L_S) Return the log of the inverse Wishart density for lower-triangular Cholesky factor of L_W given degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S dropping constant additive terms.\nAvailable since 2.30\n \n\nmatrix inv_wishart_cholesky_rng(real nu, matrix L_S) Generate the Cholesky factor of an inverse Wishart variate with degrees of freedom nu and lower-triangular Cholesky factor of the scale matrix L_S; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Covariance Matrix Distributions"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#step-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#step-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "Warning: These functions can seriously hinder sampling and optimization efficiency for gradient-based methods (e.g., NUTS, HMC, BFGS) if applied to parameters (including transformed parameters and local variables in the transformed parameters or model block). The problem is that they break gradients due to discontinuities coupled with zero gradients elsewhere. They do not hinder sampling when used in the data, transformed data, or generated quantities blocks.\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is real vector, row_vector, matrix, or an array of those types, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nreal fdim(real x, real y) Return the positive difference between x and y, which is x - y if x is greater than y and 0 otherwise; see warning above. \\[\\begin{equation*} \\text{fdim}(x,y) = \\begin{cases} x-y &\n\\text{if } x \\geq y \\\\ 0 & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fdim(T1 x, T2 y) Vectorized implementation of the fdim function\nAvailable since 2.25\n\n\n\n \n\nreal fmin(real x, real y) Return the minimum of x and y; see warning above. \\[\\begin{equation*} \\text{fmin}(x,y) = \\begin{cases} x &\n\\text{if } x \\leq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmin(T1 x, T2 y) Vectorized implementation of the fmin function\nAvailable since 2.25\n \n\nreal fmax(real x, real y) Return the maximum of x and y; see warning above. \\[\\begin{equation*} \\text{fmax}(x,y) = \\begin{cases} x &\n\\text{if } x \\geq y \\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR fmax(T1 x, T2 y) Vectorized implementation of the fmax function\nAvailable since 2.25\n\n\n\n \n\nreal fmod(real x, real y) Return the real value remainder after dividing x by y; see warning above. \\[\\begin{equation*} \\text{fmod}(x,y) = x - \\left\\lfloor \\frac{x}{y} \\right\\rfloor \\, y \\end{equation*}\\] The operator \\(\\lfloor u \\rfloor\\) is the floor operation; see below.\nAvailable since 2.0\n \n\nR fmod(T1 x, T2 y) Vectorized implementation of the fmod function\nAvailable since 2.25\n\n\n\nWarning: Rounding functions convert real values to integers. Because the output is an integer, any gradient information resulting from functions applied to the integer is not passed to the real value it was derived from. With MCMC sampling using HMC or NUTS, the MCMC acceptance procedure will correct for any error due to poor gradient calculations, but the result is likely to be reduced acceptance probabilities and less efficient sampling.\nThe rounding functions cannot be used as indices to arrays because they return real values. For operations over data or in the generated quantities block, the to_int() function can be used.\n \n\nR floor(T x) The floor of x, which is the largest integer less than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR ceil(T x) The ceiling of x, which is the smallest integer greater than or equal to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR round(T x) The nearest integer to x, converted to a real value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13\n \n\nR trunc(T x) The integer nearest to but no larger in magnitude than x, converted to a double value; see warning at start of section step-like functions\nAvailable since 2.0, vectorized in 2.13",
+    "text": "The unbounded vector probability distributions have support on all of \\(\\mathbb{R}^K\\) for some fixed \\(K\\).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormal}(y|\\mu,\\Sigma) =\n\\frac{1}{\\left( 2 \\pi \\right)^{K/2}} \\ \\frac{1}{\\sqrt{|\\Sigma|}}\n\\  \\exp \\! \\left( \\! - \\frac{1}{2} (y - \\mu)^{\\top} \\, \\Sigma^{-1} \\, (y - \\mu) \\right) \\! ,\n\\end{equation*}\\] where \\(|\\Sigma|\\) is the absolute determinant of \\(\\Sigma\\).\n\n\n\ny ~ multi_normal(mu, Sigma)\nIncrement target log probability density with multi_normal_lupdf(y | mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\nThe multivariate normal probability function is overloaded to allow the variate vector \\(y\\) and location vector \\(\\mu\\) to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal multi_normal_lpdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\nAlthough there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call multi_normal_cholesky_rng; see section multi-variate normal, cholesky parameterization.\n \n\nvector multi_normal_rng(vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_normal_rng(row_vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(row_vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Omega \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\text{MultiNormalPrecision}(y|\\mu,\\Omega)\n= \\text{MultiNormal}(y|\\mu,\\Omega^{-1}) \\end{equation*}\\]\n\n\n\ny ~ multi_normal_prec(mu, Omega)\nIncrement target log probability density with multi_normal_prec_lupdf(y | mu, Omega).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_normal_prec_lpdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L \\in\n\\mathbb{R}^{K \\times K}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormalCholesky}(y|\\mu,L) =\n\\text{MultiNormal}(y|\\mu,LL^{\\top}). \\end{equation*}\\] If \\(L\\) is lower triangular and \\(LL^{top}\\) is a \\(K \\times K\\) positive definite matrix, then \\(L_{k,k}\\) must be strictly positive for \\(k \\in 1{:}K\\). If an \\(L\\) is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.\n\n\n\ny ~ multi_normal_cholesky(mu, L)\nIncrement target log probability density with multi_normal_cholesky_lupdf(y | mu, L).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_normal_cholesky_lpdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_normal_cholesky_rng(vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.3\n \n\nvector multi_normal_cholesky_rng(row_vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(row_vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(K,N \\in \\mathbb{N}\\), \\(\\Sigma \\in \\mathbb{R}^{N \\times N}\\) is symmetric, positive definite kernel matrix and \\(w \\in \\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y \\in \\mathbb{R}^{K\n\\times N}\\), \\[\\begin{equation*} \\text{MultiGP}(y|\\Sigma,w) = \\prod_{i=1}^{K}\n\\text{MultiNormal}(y_i|0,w_i^{-1} \\Sigma), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp(Sigma, w)\nIncrement target log probability density with multi_gp_lupdf(y | Sigma, w).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_gp_lpdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_lupdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K,N \\in \\mathbb{N}\\), \\(L \\in \\mathbb{R}^{N \\times N}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite kernel matrix (implying \\(L_{n,n} &gt; 0\\) for \\(n \\in 1{:}N\\)), and \\(w \\in\n\\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y\n\\in \\mathbb{R}^{K \\times N}\\), \\[\\begin{equation*} \\text{MultiGPCholesky}(y \\, | \\ L,w)\n= \\prod_{i=1}^{K} \\text{MultiNormal}(y_i|0,w_i^{-1} LL^{\\top}), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than \\(\\text{MultiGP}()\\). Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp_cholesky(L, w)\nIncrement target log probability density with multi_gp_cholesky_lupdf(y | L, w).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal multi_gp_cholesky_lpdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_cholesky_lupdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in \\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\begin{array}{l}\n\\text{MultiStudentT}(y\\,|\\,\\nu,\\,\\mu,\\,\\Sigma) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\left| \\Sigma\n\\right|}} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\n\\Sigma^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array} \\end{equation*}\\]\n\n\n\ny ~ multi_student_t(nu, mu, Sigma)\nIncrement target log probability density with multi_student_t_lupdf(y | nu, mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_student_t_lpdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_student_t_rng(real nu, vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_student_t_rng(real nu, row_vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, row_vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks\nAvailable since 2.18\n\n\n\n\n\n\nLet \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L\\) a \\(K \\times K\\) lower-triangular matrix with strictly positive, finite diagonal then \\[\\begin{equation*}\n\\begin{array}{l}\n\\text{MultiStudentTCholesky}(y\\,\\mid \\nu,\\,\\mu,\\,L) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\left| L\n\\right|} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\nL^{-T}L^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array}\n\\end{equation*}\\]\n\n\n\ny ~ multi_student_t_cholesky(nu, mu, L)\nIncrement target log probability density with multi_student_t_cholesky_lupdf(y | nu, mu, L).\nAvailable since 2.30\n \n\n\n\n\n \n\nreal multi_student_t_cholesky_lpdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or array of vectors y given degrees of freedom nu, location vector or array of vectors mu, and Cholesky factor of the scale matrix L. For a definition of the arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nreal multi_student_t_cholesky_lupdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or vector array y given degrees of freedom nu, location vector or vector array mu, and Cholesky factor of the scale matrix L, dropping constant additive terms. For a definition of arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nvector multi_student_t_cholesky_rng(real nu, vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] row_vector mu, matrix L) Generate an array of multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n\n\n\n\nA Gaussian Dynamic Linear model is defined as follows, For \\(t \\in 1,\n\\dots, T\\), \\[\\begin{equation*}   \\begin{aligned}[t]     y_{t} &\\sim N(F' \\theta_{t}, V)\n\\\\     \\theta_{t} &\\sim N(G \\theta_{t - 1}, W) \\\\     \\theta_{0} &\\sim\nN(m_{0}, C_{0})   \\end{aligned} \\end{equation*}\\] where \\(y\\) is \\(n \\times T\\) matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (\\(p(y | F, G, V, W, m_{0}, C_{0})\\)). This log-density is a system that is calculated using the Kalman Filter. If \\(V\\) is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used (Durbin and Koopman 2001, sec. 6.4).\n\n\ny ~ gaussian_dlm_obs(F, G, V, W, m0, C0)\nIncrement target log probability density with gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0).\nAvailable since 2.0\n \n\n\n\n\nThe following two functions differ in the type of their V, the first taking a full observation covariance matrix V and the second a vector V representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#power-and-logarithm-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#power-and-logarithm-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "R sqrt(T x) The square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR cbrt(T x) The cube root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR square(T x) The square of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp(T x) The natural exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR exp2(T x) The base-2 exponential of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log(T x) The natural logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log2(T x) The base-2 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log10(T x) The base-10 logarithm of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal pow(real x, real y) Return x raised to the power of y. \\[\\begin{equation*} \\text{pow}(x,y) = x^y \\end{equation*}\\]\nAvailable since 2.0\n \n\nR pow(T1 x, T2 y) Vectorized implementation of the pow function\nAvailable since 2.25\n \n\nR inv(T x) The inverse of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_sqrt(T x) The inverse of the square root of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_square(T x) The inverse of the square of x\nAvailable since 2.0, vectorized in 2.13",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormal}(y|\\mu,\\Sigma) =\n\\frac{1}{\\left( 2 \\pi \\right)^{K/2}} \\ \\frac{1}{\\sqrt{|\\Sigma|}}\n\\  \\exp \\! \\left( \\! - \\frac{1}{2} (y - \\mu)^{\\top} \\, \\Sigma^{-1} \\, (y - \\mu) \\right) \\! ,\n\\end{equation*}\\] where \\(|\\Sigma|\\) is the absolute determinant of \\(\\Sigma\\).\n\n\n\ny ~ multi_normal(mu, Sigma)\nIncrement target log probability density with multi_normal_lupdf(y | mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\nThe multivariate normal probability function is overloaded to allow the variate vector \\(y\\) and location vector \\(\\mu\\) to be vectors or row vectors (or to mix the two types). The density function is also vectorized, so it allows arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal multi_normal_lpdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_lpdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma\nAvailable since 2.12\n \n\nreal multi_normal_lupdf(row_vectors y | row_vectors mu, matrix Sigma) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and covariance matrix Sigma dropping constant additive terms\nAvailable since 2.25\nAlthough there is a direct multi-normal RNG function, if more than one result is required, it’s much more efficient to Cholesky factor the covariance matrix and call multi_normal_cholesky_rng; see section multi-variate normal, cholesky parameterization.\n \n\nvector multi_normal_rng(vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_normal_rng(row_vector mu, matrix Sigma) Generate a multivariate normal variate with location mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_rng(row_vectors mu, matrix Sigma) Generate an array of multivariate normal variates with locations mu and covariance matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#trigonometric-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#trigonometric-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution-precision-parameterization",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-normal-distribution-precision-parameterization",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "real hypot(real x, real y) Return the length of the hypotenuse of a right triangle with sides of length x and y. \\[\\begin{equation*} \\text{hypot}(x,y) = \\begin{cases} \\sqrt{x^2+y^2} &\n\\text{if } x,y\\geq 0 \\\\ \\textrm{NaN} & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR hypot(T1 x, T2 y) Vectorized implementation of the hypot function\nAvailable since 2.25\n \n\nR cos(T x) The cosine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sin(T x) The sine of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tan(T x) The tangent of the angle x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acos(T x) The principal arc (inverse) cosine (in radians) of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR asin(T x) The principal arc (inverse) sine (in radians) of x\nAvailable since 2.0\n \n\nR atan(T x) The principal arc (inverse) tangent (in radians) of x, with values from \\(-\\pi/2\\) to \\(\\pi/2\\)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atan2(T y, T x) Return the principal arc (inverse) tangent (in radians) of y divided by x, \\[\\begin{equation*} \\text{atan2}(y, x) = \\arctan\\left(\\frac{y}{x}\\right) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.34",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Omega \\in\n\\mathbb{R}^{K \\times   K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\text{MultiNormalPrecision}(y|\\mu,\\Omega)\n= \\text{MultiNormal}(y|\\mu,\\Omega^{-1}) \\end{equation*}\\]\n\n\n\ny ~ multi_normal_prec(mu, Omega)\nIncrement target log probability density with multi_normal_prec_lupdf(y | mu, Omega).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_normal_prec_lpdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_prec_lpdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega\nAvailable since 2.18\n \n\nreal multi_normal_prec_lupdf(row_vectors y | row_vectors mu, matrix Omega) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and positive definite precision matrix Omega dropping constant additive terms\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#hyperbolic-trigonometric-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#hyperbolic-trigonometric-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multi-normal-cholesky-fun",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multi-normal-cholesky-fun",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "R cosh(T x) The hyperbolic cosine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR sinh(T x) The hyperbolic sine of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR tanh(T x) The hyperbolic tangent of x (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR acosh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR asinh(T x) The inverse hyperbolic cosine (in radians)\nAvailable since 2.0, vectorized in 2.13\n \n\nR atanh(T x) The inverse hyperbolic tangent (in radians) of x\nAvailable since 2.0, vectorized in 2.13",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L \\in\n\\mathbb{R}^{K \\times K}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*}\n\\text{MultiNormalCholesky}(y|\\mu,L) =\n\\text{MultiNormal}(y|\\mu,LL^{\\top}). \\end{equation*}\\] If \\(L\\) is lower triangular and \\(LL^{top}\\) is a \\(K \\times K\\) positive definite matrix, then \\(L_{k,k}\\) must be strictly positive for \\(k \\in 1{:}K\\). If an \\(L\\) is provided that is not the Cholesky factor of a positive-definite matrix, the probability functions will raise errors.\n\n\n\ny ~ multi_normal_cholesky(mu, L)\nIncrement target log probability density with multi_normal_cholesky_lupdf(y | mu, L).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_normal_cholesky_lpdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_normal_cholesky_lpdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L\nAvailable since 2.18\n \n\nreal multi_normal_cholesky_lupdf(row_vectors y | row_vectors mu, matrix L) The log of the multivariate normal density of row vector(s) y given location row vector(s) mu and lower-triangular Cholesky factor of the covariance matrix L dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_normal_cholesky_rng(vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.3\n \n\nvector multi_normal_cholesky_rng(row_vector mu, matrix L) Generate a multivariate normal variate with location mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_normal_cholesky_rng(row_vectors mu, matrix L) Generate an array of multivariate normal variates with locations mu and lower-triangular Cholesky factor of the covariance matrix L; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#link-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#link-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "The following functions are commonly used as link functions in generalized linear models. The function \\(\\Phi\\) is also commonly used as a link function (see section probability-related functions).\n \n\nR logit(T x) The log odds, or logit, function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_logit(T x) The logistic sigmoid function applied to x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_cloglog(T x) The inverse of the complementary log-log function applied to x\nAvailable since 2.0, vectorized in 2.13",
+    "text": "If \\(K,N \\in \\mathbb{N}\\), \\(\\Sigma \\in \\mathbb{R}^{N \\times N}\\) is symmetric, positive definite kernel matrix and \\(w \\in \\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y \\in \\mathbb{R}^{K\n\\times N}\\), \\[\\begin{equation*} \\text{MultiGP}(y|\\Sigma,w) = \\prod_{i=1}^{K}\n\\text{MultiNormal}(y_i|0,w_i^{-1} \\Sigma), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp(Sigma, w)\nIncrement target log probability density with multi_gp_lupdf(y | Sigma, w).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal multi_gp_lpdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_lupdf(matrix y | matrix Sigma, vector w) The log of the multivariate GP density of matrix y given kernel matrix Sigma and inverses scales w dropping constant additive terms\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#Phi-function",
-    "href": "functions-reference/real-valued_basic_functions.html#Phi-function",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution-cholesky-parameterization",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-gaussian-process-distribution-cholesky-parameterization",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "The error function erf is related to the standard normal cumulative distribution function \\(\\Phi\\) by scaling. See section normal distribution for the general normal cumulative distribution function (and its complement).\n \n\nR erf(T x) The error function, also known as the Gauss error function, of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR erfc(T x) The complementary error function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_erfc(T x) The inverse of the complementary error function of x\nAvailable since 2.29, vectorized in 2.29\n \n\nR Phi(T x) The standard normal cumulative distribution function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR inv_Phi(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The details of the algorithm can be found in (Wichura 1988). Quantile arguments below 1e-16 are untested; quantiles above 0.999999999 result in increasingly large errors.\nAvailable since 2.0, vectorized in 2.13\n \n\nR Phi_approx(T x) The fast approximation of the unit (may replace Phi for probit regression with maximum absolute error of 0.00014, see (Bowling et al. 2009) for details)\nAvailable since 2.0, vectorized in 2.13\n\n\n\n \n\nreal binary_log_loss(int y, real y_hat) Return the log loss function for for predicting \\(\\hat{y} \\in [0,1]\\) for boolean outcome \\(y \\in \\{0,1\\}\\). \\[\\begin{equation*}\n\\mathrm{binary\\_log\\_loss}(y,\\hat{y}) = \\begin{cases} -\\log \\hat{y} &\n\\text{if } y = 1\\\\ -\\log (1 - \\hat{y}) & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR binary_log_loss(T1 x, T2 y) Vectorized implementation of the binary_log_loss function\nAvailable since 2.25\n \n\nreal owens_t(real h, real a) Return the Owen’s T function for the probability of the event \\(X &gt; h\\) and \\(0&lt;Y&lt;aX\\) where X and Y are independent standard normal random variables. \\[\\begin{equation*} \\mathrm{owens\\_t}(h,a) = \\frac{1}{2\\pi} \\int_0^a\n\\frac{\\exp(-\\frac{1}{2}h^2(1+x^2))}{1+x^2}dx \\end{equation*}\\]\nAvailable since 2.25\n \n\nR owens_t(T1 x, T2 y) Vectorized implementation of the owens_t function\nAvailable since 2.25",
+    "text": "If \\(K,N \\in \\mathbb{N}\\), \\(L \\in \\mathbb{R}^{N \\times N}\\) is lower triangular and such that \\(LL^{\\top}\\) is positive definite kernel matrix (implying \\(L_{n,n} &gt; 0\\) for \\(n \\in 1{:}N\\)), and \\(w \\in\n\\mathbb{R}^{K}\\) is a vector of positive inverse scales, then for \\(y\n\\in \\mathbb{R}^{K \\times N}\\), \\[\\begin{equation*} \\text{MultiGPCholesky}(y \\, | \\ L,w)\n= \\prod_{i=1}^{K} \\text{MultiNormal}(y_i|0,w_i^{-1} LL^{\\top}), \\end{equation*}\\] where \\(y_i\\) is the \\(i\\)th row of \\(y\\). This is used to efficiently handle Gaussian Processes with multi-variate outputs where only the output dimensions share a kernel function but vary based on their scale. If the model allows parameterization in terms of Cholesky factor of the kernel matrix, this distribution is also more efficient than \\(\\text{MultiGP}()\\). Note that this function does not take into account the mean prediction.\n\n\n\ny ~ multi_gp_cholesky(L, w)\nIncrement target log probability density with multi_gp_cholesky_lupdf(y | L, w).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal multi_gp_cholesky_lpdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w\nAvailable since 2.12\n \n\nreal multi_gp_cholesky_lupdf(matrix y | matrix L, vector w) The log of the multivariate GP density of matrix y given lower-triangular Cholesky factor of the kernel matrix L and inverses scales w dropping constant additive terms\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#betafun",
-    "href": "functions-reference/real-valued_basic_functions.html#betafun",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-student-t-distribution",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multivariate-student-t-distribution",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "real beta(real alpha, real beta) Return the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.25\n \n\nR beta(T1 x, T2 y) Vectorized implementation of the beta function\nAvailable since 2.25\n \n\nreal inc_beta(real alpha, real beta, real x) Return the regularized incomplete beta function up to x applied to alpha and beta. See section appendix for a definition.\nAvailable since 2.10\n \n\nreal inv_inc_beta(real alpha, real beta, real p) Return the inverse of the regularized incomplete beta function. The return value x is the value that solves p = inc_beta(alpha, beta, x). See section appendix for a definition of the inc_beta.\nAvailable since 2.30\n \n\nreal lbeta(real alpha, real beta) Return the natural logarithm of the beta function applied to alpha and beta. The beta function, \\(\\text{B}(\\alpha,\\beta)\\), computes the normalizing constant for the beta distribution, and is defined for \\(\\alpha &gt; 0\\) and \\(\\beta &gt; 0\\). \\[\\begin{equation*}\n\\text{lbeta}(\\alpha,\\beta) = \\log \\Gamma(\\alpha) + \\log \\Gamma(\\beta) - \\log \\Gamma(\\alpha+\\beta)\n\\end{equation*}\\] See section appendix for definition of \\(\\text{B}(\\alpha, \\beta)\\).\nAvailable since 2.0\n \n\nR lbeta(T1 x, T2 y) Vectorized implementation of the lbeta function\nAvailable since 2.25\n \n\nR tgamma(T x) The gamma function applied to x. The gamma function is the generalization of the factorial function to continuous variables, defined so that \\(\\Gamma(n+1) = n!\\). See for a full definition of \\(\\Gamma(x)\\). The function is defined for positive numbers and non-integral negative numbers,\nAvailable since 2.0, vectorized in 2.13\n \n\nR lgamma(T x) The natural logarithm of the gamma function applied to x,\nAvailable since 2.0, vectorized in 2.15\n \n\nR digamma(T x) The digamma function applied to x. The digamma function is the derivative of the natural logarithm of the Gamma function. The function is defined for positive numbers and non-integral negative numbers\nAvailable since 2.0, vectorized in 2.13\n \n\nR trigamma(T x) The trigamma function applied to x. The trigamma function is the second derivative of the natural logarithm of the Gamma function\nAvailable since 2.0, vectorized in 2.13\n \n\nreal lmgamma(int n, real x) Return the natural logarithm of the multivariate gamma function \\(\\Gamma_n\\) with n dimensions applied to x. \\[\\begin{equation*}\n\\text{lmgamma}(n,x) =\n\\begin{cases} \\frac{n(n-1)}{4} \\log \\pi + \\sum_{j=1}^n \\log \\Gamma\\left(x + \\frac{1 - j}{2}\\right)\n& \\text{if } x\\not\\in \\{\\dots,-3,-2,-1,0\\}\\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR lmgamma(T1 x, T2 y) Vectorized implementation of the lmgamma function\nAvailable since 2.25\n \n\nreal gamma_p(real a, real z) Return the normalized lower incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_p}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_0^zt^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_p(T1 x, T2 y) Vectorized implementation of the gamma_p function\nAvailable since 2.25\n \n\nreal gamma_q(real a, real z) Return the normalized upper incomplete gamma function of a and z defined for positive a and nonnegative z. \\[\\begin{equation*}\n\\mathrm{gamma\\_q}(a,z) =\n\\begin{cases} \\frac{1}{\\Gamma(a)}\\int_z^\\infty t^{a-1}e^{-t}dt &\n\\text{if } a &gt; 0, z \\geq 0 \\\\[6pt] \\textrm{error} & \\text{otherwise}\n\\end{cases}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR gamma_q(T1 x, T2 y) Vectorized implementation of the gamma_q function\nAvailable since 2.25\n \n\nint choose(int x, int y) Return the binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” In its the antilog of the lchoose function but returns an integer rather than a real number with no non-zero decimal places. For \\(0 \\leq y \\leq x\\), the binomial coefficient function can be defined via the factorial function \\[\\begin{equation*}\n\\text{choose}(x,y) = \\frac{x!}{\\left(y!\\right)\\left(x - y\\right)!}.\n\\end{equation*}\\]\nAvailable since 2.14\n \n\nR choose(T1 x, T2 y) Vectorized implementation of the choose function\nAvailable since 2.25\n \n\nreal bessel_first_kind(int v, real x) Return the Bessel function of the first kind with order v applied to x. \\[\\begin{equation*}\n\\mathrm{bessel\\_first\\_kind}(v,x) = J_v(x),\n\\end{equation*}\\] where \\[\\begin{equation*}\nJ_v(x)=\\left(\\frac{1}{2}x\\right)^v \\sum_{k=0}^\\infty\n\\frac{\\left(-\\frac{1}{4}x^2\\right)^k}{k!\\, \\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_first_kind(T1 x, T2 y) Vectorized implementation of the bessel_first_kind function\nAvailable since 2.25\n \n\nreal bessel_second_kind(int v, real x) Return the Bessel function of the second kind with order v applied to x defined for positive x and v. For \\(x,v &gt; 0\\), \\[\\begin{equation*}\n\\mathrm{bessel\\_second\\_kind}(v,x) =\n\\begin{cases} Y_v(x) & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\nY_v(x)=\\frac{J_v(x)\\cos(v\\pi)-J_{-v}(x)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.5\n \n\nR bessel_second_kind(T1 x, T2 y) Vectorized implementation of the bessel_second_kind function\nAvailable since 2.25\n \n\nreal modified_bessel_first_kind(int v, real z) Return the modified Bessel function of the first kind with order v applied to z defined for all z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_first\\_kind}(v,z) = I_v(z)\n\\end{equation*}\\] where \\[\\begin{equation*}\n{I_v}(z) = \\left(\\frac{1}{2}z\\right)^v\\sum_{k=0}^\\infty \\frac{\\left(\\frac{1}{4}z^2\\right)^k}{k!\\Gamma(v+k+1)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_first_kind function\nAvailable since 2.25\n \n\nreal log_modified_bessel_first_kind(real v, real z) Return the log of the modified Bessel function of the first kind. v does not have to be an integer.\nAvailable since 2.26\n \n\nR log_modified_bessel_first_kind(T1 x, T2 y) Vectorized implementation of the log_modified_bessel_first_kind function\nAvailable since 2.26\n \n\nreal modified_bessel_second_kind(int v, real z) Return the modified Bessel function of the second kind with order v applied to z defined for positive z and integer v. \\[\\begin{equation*}\n\\mathrm{modified\\_bessel\\_second\\_kind}(v,z) =\n\\begin{cases} K_v(z) &  \\text{if } z &gt; 0 \\\\ \\textrm{error} & \\text{if } z \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} {K_v}(z) = \\frac{\\pi}{2}\\cdot\\frac{I_{-v}(z) - I_{v}(z)}{\\sin(v\\pi)}\n\\end{equation*}\\]\nAvailable since 2.1\n \n\nR modified_bessel_second_kind(T1 x, T2 y) Vectorized implementation of the modified_bessel_second_kind function\nAvailable since 2.25\n \n\nreal falling_factorial(real x, real n) Return the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*}\n\\mathrm{falling\\_factorial}(x,n) =\n\\begin{cases} (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*}\n(x)_n=\\frac{\\Gamma(x+1)}{\\Gamma(x-n+1)}\n\\end{equation*}\\]\nAvailable since 2.0\n \n\nR falling_factorial(T1 x, T2 y) Vectorized implementation of the falling_factorial function\nAvailable since 2.25\n \n\nreal lchoose(real x, real y) Return the natural logarithm of the generalized binomial coefficient of x and y. For non-negative integer inputs, the binomial coefficient function is written as \\(\\binom{x}{y}\\) and pronounced “x choose y.” This function generalizes to real numbers using the gamma function. For \\(0 \\leq y \\leq x\\), \\[\\begin{equation*} \\mathrm{binomial\\_coefficient\\_log}(x,y) =\n\\log\\Gamma(x+1) - \\log\\Gamma(y+1) - \\log\\Gamma(x-y+1). \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lchoose(T1 x, T2 y) Vectorized implementation of the lchoose function\nAvailable since 2.29\n \n\nreal log_falling_factorial(real x, real n) Return the log of the falling factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_falling\\_factorial}(x,n) =\n\\begin{cases} \\log (x)_n & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal rising_factorial(real x, int n) Return the rising factorial of x with power n defined for positive x and integer n. \\[\\begin{equation*}\n\\mathrm{rising\\_factorial}(x,n) = \\begin{cases} x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} & \\text{if } x \\leq 0 \\end{cases}\n\\end{equation*}\\] where \\[\\begin{equation*} x^{(n)}=\\frac{\\Gamma(x+n)}{\\Gamma(x)} \\end{equation*}\\]\nAvailable since 2.20\n \n\nR rising_factorial(T1 x, T2 y) Vectorized implementation of the rising_factorial function\nAvailable since 2.25\n \n\nreal log_rising_factorial(real x, real n) Return the log of the rising factorial of x with power n defined for positive x and real n. \\[\\begin{equation*} \\mathrm{log\\_rising\\_factorial}(x,n) =\n\\begin{cases} \\log x^{(n)} & \\text{if } x &gt; 0 \\\\ \\textrm{error} &\n\\text{if } x \\leq 0 \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nR log_rising_factorial(T1 x, T2 y) Vectorized implementation of the log_rising_factorial function\nAvailable since 2.25",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(\\Sigma \\in \\mathbb{R}^{K \\times K}\\) is symmetric and positive definite, then for \\(y \\in \\mathbb{R}^K\\), \\[\\begin{equation*} \\begin{array}{l}\n\\text{MultiStudentT}(y\\,|\\,\\nu,\\,\\mu,\\,\\Sigma) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\left| \\Sigma\n\\right|}} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\n\\Sigma^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array} \\end{equation*}\\]\n\n\n\ny ~ multi_student_t(nu, mu, Sigma)\nIncrement target log probability density with multi_student_t_lupdf(y | nu, mu, Sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multi_student_t_lpdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal multi_student_t_lpdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma\nAvailable since 2.18\n \n\nreal multi_student_t_lupdf(row_vectors y | real nu, row_vectors mu, matrix Sigma) The log of the multivariate Student-\\(t\\) density of row vector(s) y given degrees of freedom nu, location row vector(s) mu, and scale matrix Sigma dropping constant additive terms\nAvailable since 2.25\n \n\nvector multi_student_t_rng(real nu, vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0\n \n\nvector multi_student_t_rng(real nu, row_vector mu, matrix Sigma) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data and generated quantities blocks\nAvailable since 2.18\n \n\nvectors multi_student_t_rng(real nu, row_vectors mu, matrix Sigma) Generate an array of multivariate Student-\\(t\\) variates with degrees of freedom nu, locations mu, and scale matrix Sigma; may only be used in transformed data andgenerated quantities blocks\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#composed-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#composed-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#multi-student-t-cholesky-fun",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#multi-student-t-cholesky-fun",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "The functions in this section are equivalent in theory to combinations of other functions. In practice, they are implemented to be more efficient and more numerically stable than defining them directly using more basic Stan functions.\n \n\nR expm1(T x) The natural exponential of x minus 1\nAvailable since 2.0, vectorized in 2.13\n \n\nreal fma(real x, real y, real z) Return z plus the result of x multiplied by y. \\[\\begin{equation*} \\text{fma}(x,y,z) =\n(x \\times y) + z \\end{equation*}\\]\nAvailable since 2.0\n \n\nreal ldexp(real x, int y) Return the product of x and two raised to the y power. \\[\\begin{equation*}\n\\text{ldexp}(x,y) = x 2^y  \\end{equation*}\\]\nAvailable since 2.25\n \n\nR ldexp(T1 x, T2 y) Vectorized implementation of the ldexp function\nAvailable since 2.25\n \n\nreal lmultiply(real x, real y) Return the product of x and the natural logarithm of y. \\[\\begin{equation*}\n\\text{lmultiply}(x,y) = \\begin{cases} 0 & \\text{if } x = y = 0 \\\\ x\n\\log y & \\text{if } x, y \\neq 0 \\\\ \\text{NaN} & \\text{otherwise}\n\\end{cases} \\end{equation*}\\]\nAvailable since 2.10\n \n\nR lmultiply(T1 x, T2 y) Vectorized implementation of the lmultiply function\nAvailable since 2.25\n \n\nR log1p(T x) The natural logarithm of 1 plus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m(T x) The natural logarithm of 1 minus x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1p_exp(T x) The natural logarithm of one plus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log1m_exp(T x) The logarithm of one minus the natural exponentiation of x\nAvailable since 2.0, vectorized in 2.13\n \n\nreal log_diff_exp(real x, real y) Return the natural logarithm of the difference of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_diff\\_exp}(x,y) = \\begin{cases} \\log(\\exp(x)-\\exp(y)) &\n\\text{if } +\\infty &gt; x \\ge y \\\\[6pt]\n\\textrm{NaN} & \\text{otherwise} \\end{cases}\n\\end{equation*}\\]\nWhen x is equal to y, log_diff_exp(x, y) returns \\(-\\infty\\), consistent with log(0) returning \\(-\\infty\\). This includes the case in which x and y are both equal to \\(-\\infty\\), which corresponds to log(0 - 0) because exp(negative_infinity()) returns 0.\nAvailable since 2.0\n \n\nR log_diff_exp(T1 x, T2 y) Vectorized implementation of the log_diff_exp function\nAvailable since 2.25\n \n\nreal log_mix(real theta, real lp1, real lp2) Return the log mixture of the log densities lp1 and lp2 with mixing proportion theta, defined by \\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda_1, \\lambda_2) & = & \\log \\!\\left(\n\\theta \\exp(\\lambda_1) + \\left( 1 - \\theta \\right) \\exp(\\lambda_2)\n\\right) \\\\[3pt] & = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) +\n\\lambda_1, \\ \\log(1 - \\theta) + \\lambda_2\\right). \\end{eqnarray*}\\]\nAvailable since 2.6\n \n\nR log_mix(T1 thetas, T2 lps)\nCalculates the log mixture density given thetas, mixing proportions which should be between 0 and 1 and sum to 1, and lps, log densities. The lps variable must be either a 1-d container of the same length as thetas, or an array of such.\n\\[\\begin{eqnarray*}\n\\mathrm{log\\_mix}(\\theta, \\lambda)\n& = & \\log \\!\\left( \\sum_{n=1}^N \\theta_n * \\exp(\\lambda_n) \\right) \\\\[3pt]\n& = & \\mathrm{log\\_sum\\_exp}\\!\\left(\\log(\\theta) + \\lambda\\right).\n\\end{eqnarray*}\\]\nThis is a generalization of the above signature of three arguments to more than two densities. For example, log_mix(lambda, lp1, lp2) == log_mix({lambda, 1 - lambda}, {lp1, lp2}).\nAvailable since 2.26\n \n\nR log_sum_exp(T1 x, T2 y) Return the natural logarithm of the sum of the natural exponentiation of x and the natural exponentiation of y. \\[\\begin{equation*}\n\\mathrm{log\\_sum\\_exp}(x,y) = \\log(\\exp(x)+\\exp(y)) \\end{equation*}\\]\nAvailable since 2.0, vectorized in 2.33\n \n\nR log_inv_logit(T x) The natural logarithm of the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13\n \n\nR log_inv_logit_diff(T1 x, T2 y) The natural logarithm of the difference of the inverse logit function of x and the inverse logit function of y\nAvailable since 2.25\n \n\nR log1m_inv_logit(T x) The natural logarithm of 1 minus the inverse logit function of x\nAvailable since 2.0, vectorized in 2.13",
+    "text": "Let \\(K \\in \\mathbb{N}\\), \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}^K\\), and \\(L\\) a \\(K \\times K\\) lower-triangular matrix with strictly positive, finite diagonal then \\[\\begin{equation*}\n\\begin{array}{l}\n\\text{MultiStudentTCholesky}(y\\,\\mid \\nu,\\,\\mu,\\,L) \\\\  =\n\\frac{1}{\\pi^{K/2}} \\ \\frac{1}{\\nu^{K/2}} \\ \\frac{\\Gamma\\!\\left((\\nu +\nK)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\left| L\n\\right|} \\ \\left( 1 + \\frac{1}{\\nu} \\, \\left(y - \\mu\\right)^{\\top} \\,\nL^{-T}L^{-1} \\, \\left(y - \\mu\\right) \\right)^{-(\\nu + K)/2} \\! .\n\\end{array}\n\\end{equation*}\\]\n\n\n\ny ~ multi_student_t_cholesky(nu, mu, L)\nIncrement target log probability density with multi_student_t_cholesky_lupdf(y | nu, mu, L).\nAvailable since 2.30\n \n\n\n\n\n \n\nreal multi_student_t_cholesky_lpdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or array of vectors y given degrees of freedom nu, location vector or array of vectors mu, and Cholesky factor of the scale matrix L. For a definition of the arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nreal multi_student_t_cholesky_lupdf(vectors y | real nu, vectors mu, matrix L) The log of the multivariate Student-\\(t\\) density of vector or vector array y given degrees of freedom nu, location vector or vector array mu, and Cholesky factor of the scale matrix L, dropping constant additive terms. For a definition of arguments compatible with the vectors type, see the probability vectorization section.\nAvailable since 2.30\n \n\nvector multi_student_t_cholesky_rng(real nu, vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] vector mu, matrix L) Generate a multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30\n \n\narray[] vector multi_student_t_cholesky_rng(real nu, array[] row_vector mu, matrix L) Generate an array of multivariate Student-\\(t\\) variate with degrees of freedom nu, location array mu, and Cholesky factor of the scale matrix L; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.30",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#special-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#special-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/distributions_over_unbounded_vectors.html#gaussian-dynamic-linear-models",
+    "href": "functions-reference/distributions_over_unbounded_vectors.html#gaussian-dynamic-linear-models",
+    "title": "Distributions over Unbounded Vectors",
     "section": "",
-    "text": "R lambert_w0(T x) Implementation of the \\(W_0\\) branch of the Lambert W function, i.e., solution to the function \\(W_0(x) \\exp^{ W_0(x)} = x\\)\nAvailable since 2.25\n \n\nR lambert_wm1(T x) Implementation of the \\(W_{-1}\\) branch of the Lambert W function, i.e., solution to the function \\(W_{-1}(x) \\exp^{W_{-1}(x)} = x\\)\nAvailable since 2.25",
+    "text": "A Gaussian Dynamic Linear model is defined as follows, For \\(t \\in 1,\n\\dots, T\\), \\[\\begin{equation*}   \\begin{aligned}[t]     y_{t} &\\sim N(F' \\theta_{t}, V)\n\\\\     \\theta_{t} &\\sim N(G \\theta_{t - 1}, W) \\\\     \\theta_{0} &\\sim\nN(m_{0}, C_{0})   \\end{aligned} \\end{equation*}\\] where \\(y\\) is \\(n \\times T\\) matrix where rows are variables and columns are observations. These functions calculate the log-density of the observations marginalizing over the latent states (\\(p(y | F, G, V, W, m_{0}, C_{0})\\)). This log-density is a system that is calculated using the Kalman Filter. If \\(V\\) is diagonal, then a more efficient algorithm which sequentially processes observations and avoids a matrix inversions can be used (Durbin and Koopman 2001, sec. 6.4).\n\n\ny ~ gaussian_dlm_obs(F, G, V, W, m0, C0)\nIncrement target log probability density with gaussian_dlm_obs_lupdf(y | F, G, V, W, m0, C0).\nAvailable since 2.0\n \n\n\n\n\nThe following two functions differ in the type of their V, the first taking a full observation covariance matrix V and the second a vector V representing the diagonal of the observation covariance matrix. The sampling statement defined in the previous section works with either type of observation V.\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25\n \n\nreal gaussian_dlm_obs_lpdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0.\nAvailable since 2.12\n \n\nreal gaussian_dlm_obs_lupdf(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) The log of the density of the Gaussian Dynamic Linear model with observation matrix y in which rows are variables and columns are observations, design matrix F, transition matrix G, observation covariance matrix with diagonal V, system covariance matrix W, and the initial state is distributed normal with mean m0 and covariance C0. This function drops constant additive terms.\nAvailable since 2.25",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Continuous Distributions",
+      "Distributions over Unbounded Vectors"
     ]
   },
   {
-    "objectID": "functions-reference/real-valued_basic_functions.html#hypergeometric-functions",
-    "href": "functions-reference/real-valued_basic_functions.html#hypergeometric-functions",
-    "title": "Real-Valued Basic Functions",
+    "objectID": "functions-reference/functions_index.html",
+    "href": "functions-reference/functions_index.html",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Hypergeometric functions refer to a power series of the form \\[\\begin{equation*}\n_pF_q(a_1,...,a_p;b_1,...,b_q;z) = \\sum_{n=0}^\\infty \\frac{(a_1)_n\\cdot\\cdot\\cdot(a_p)_n}{(b_1)_n\\cdot\\cdot\\cdot(b_q)_n} \\frac{z^n}{n!}\n\\end{equation*}\\] where \\((a)_n\\) is the Pochhammer symbol defined as \\((a)_n = \\frac{\\Gamma(a+n)}{\\Gamma(a)}\\).\nThe gradients of the hypergeometric function are given by: \\[\\begin{equation*}\n\\frac{\\partial }{\\partial a_1} =\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+a_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n    - \\psi\\left(a_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z)\n\\end{equation*}\\] \\[\\begin{equation*}\n\\frac{\\partial }{\\partial b_1} =\n  \\psi\\left(b_1\\right){}_pF_q(a_1,...,a_p;b_1,...,b_q;z) -\n  \\sum_{k=0}^{\\infty}{\n    \\frac\n      {\\psi\\left(k+b_1\\right)\\left(\\prod_{j=1}^p\\left(a_j\\right)_k\\right)z^k}\n      {k!\\prod_{j=1}^q\\left(b_j\\right)_k}}\n\\end{equation*}\\] \\[\\begin{equation*}\n  \\frac{\\partial }{\\partial z} =\n  \\frac{\\prod_{j=1}^{p}a_j}{\\prod_{j=1}^{q} b_j}{}_pF_q(a_1+1,...,a_p+1;b_1+1,...,b_q+1;z)\n\\end{equation*}\\]\nStan provides both the generalized hypergeometric function as well as several special cases for particular values of p and q.\n \n\nreal hypergeometric_1F0(real a, real z) Special case of the hypergeometric function with \\(p=1\\) and \\(q=0\\).\nAvailable since 2.37\n \n\nreal hypergeometric_2F1(real a1, real a2, real b1, real z) Special case of the hypergeometric function with \\(p=2\\) and \\(q=1\\). If the function does not meet convergence criteria for given inputs, the function will attempt to apply Euler’s transformation to improve convergence: \\[\\begin{equation*}\n{}_2F_1(a_1,a_2, b_1, z)={}_2F_1(b_1 - a_1,a_2, b_1, \\frac{z}{z-1})\\cdot(1-z)^{-a_2}\n\\end{equation*}\\]\nAvailable since 2.37\n \n\nreal hypergeometric_3F2(T1 a, T2 b, real z) Special case of the hypergeometric function with \\(p=3\\) and \\(q=2\\), where a and b are vectors of length 3 and 2, respectively.\nAvailable since 2.37\n \n\nreal hypergeometric_pFq(T1 a, T2 b, real z) Generalized hypergeometric function, where a and b are vectors of length p and q, respectively.\nAvailable since 2.37",
+    "text": "abs:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\nacos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nacosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nadd_diag:\n\n\n(complex_matrix m, complex_real d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_row_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, real d) : matrix (matrix_operations.html)\n\n\n(matrix m, row_vector d) : matrix (matrix_operations.html)\n\n\n(matrix m, vector d) : matrix (matrix_operations.html)\n\n\nalgebra_solver:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nalgebra_solver_newton:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector (deprecated_functions.html)\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nappend_array:\n\n\n(T x, T y) : T (array_operations.html)\n\n\nappend_col:\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(vector x, vector y) : matrix (matrix_operations.html)\n\n\nappend_row:\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, row_vector y) : matrix (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : matrix (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\narg:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nasin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nasinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan2:\n\n\n(T y, T x) : R (real-valued_basic_functions.html)\n\n\natanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\n\n\n\nbernoulli:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_cdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lccdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lcdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_logit:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_rng:\n\n\n(matrix x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\n(row_vector x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\nbernoulli_logit_lpmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_lupmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_rng:\n\n\n(reals alpha) : R (binary_distributions.html)\n\n\nbernoulli_lpmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lupmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_rng:\n\n\n(reals theta) : R (binary_distributions.html)\n\n\nbessel_first_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbessel_second_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_binomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbeta_binomial_cdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lccdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lcdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lpmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lupmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_rng:\n\n\n(ints N, reals alpha, reals beta) : R (bounded_discrete_distributions.html)\n\n\nbeta_cdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lccdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lcdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lpdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lupdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_cdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lccdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lcdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lpmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lupmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_rng:\n\n\n(reals r, reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nbeta_proportion:\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lccdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lcdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lpdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lupdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_rng:\n\n\n(reals mu, reals kappa) : R (continuous_distributions_on_0_1.html)\n\n\nbeta_rng:\n\n\n(reals alpha, reals beta) : R (continuous_distributions_on_0_1.html)\n\n\nbinary_log_loss:\n\n\n(int y, real y_hat) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbinomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_cdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lccdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lcdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lpmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lupmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lpmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lupmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lpmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lupmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_rng:\n\n\n(ints N, reals theta) : R (bounded_discrete_distributions.html)\n\n\nblock:\n\n\n(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows, int n_cols) : matrix (matrix_operations.html)\n\n\n\n\n\ncategorical:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lpmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lupmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_rng:\n\n\n(vector beta) : int (bounded_discrete_distributions.html)\n\n\ncategorical_lpmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_lupmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_rng:\n\n\n(vector theta) : int (bounded_discrete_distributions.html)\n\n\ncauchy:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ncauchy_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\ncbrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nceil:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nchi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nchi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\nchol2inv:\n\n\n(matrix L) : matrix (matrix_operations.html)\n\n\ncholesky_decompose:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ncholesky_factor_corr_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncholesky_factor_cov_constrain:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_jacobian:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\nchoose:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ncol:\n\n\n(complex_matrix x, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int n) : vector (matrix_operations.html)\n\n\ncols:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\ncolumns_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : row_vector (matrix_operations.html)\n\n\ncolumns_dot_self:\n\n\n(complex_matrix x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : row_vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\ncomplex_schur_decompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_t:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_u:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\nconj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(Z z) : Z (complex-valued_basic_functions.html)\n\n\ncorr_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncov_exp_quad:\n\n\n(array[] real x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(array[] real x1, array[] real x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x1, vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\ncov_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ncsr_extract:\n\n\n(matrix a) : tuple(vector, array[] int, array[] int) (sparse_matrix_operations.html)\n\n\ncsr_extract_u:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_v:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_w:\n\n\n(matrix a) : vector (sparse_matrix_operations.html)\n\n\ncsr_matrix_times_vector:\n\n\n(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector (sparse_matrix_operations.html)\n\n\ncsr_to_dense_matrix:\n\n\n(int m, int n, vector w, array[] int v, array[] int u) : matrix (sparse_matrix_operations.html)\n\n\ncumulative_sum:\n\n\n(array[] complex x) : array[] complex (complex_matrix_operations.html)\n\n\n(array[] int x) : array[] int (matrix_operations.html)\n\n\n(array[] real x) : array[] real (matrix_operations.html)\n\n\n(complex_row_vector rv) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\n\n\n\ndae:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\ndae_tol:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\ndeterminant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ndiag_matrix:\n\n\n(complex_vector x) : complex_matrix (complex_matrix_operations.html)\n\n\n(vector x) : matrix (matrix_operations.html)\n\n\ndiag_post_multiply:\n\n\n(complex_matrix m, complex_row_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\ndiag_pre_multiply:\n\n\n(complex_row_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(row_vector rv, matrix m) : matrix (matrix_operations.html)\n\n\n(vector v, matrix m) : matrix (matrix_operations.html)\n\n\ndiagonal:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\ndigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ndims:\n\n\n(T x) : array[] int (array_operations.html)\n\n\ndirichlet:\n\n\ndistribution statement (simplex_distributions.html)\n\n\ndirichlet_lpdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_lupdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_multinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lpmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lupmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_rng:\n\n\n(vector alpha, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\ndirichlet_rng:\n\n\n(vector alpha) : vector (simplex_distributions.html)\n\n\ndiscrete_range:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ndiscrete_range_cdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lccdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lcdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lpmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lupmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_rng:\n\n\n(ints l, ints u) : ints (bounded_discrete_distributions.html)\n\n\ndistance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\ndot_product:\n\n\n(complex_row_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(row_vector x, row_vector y) : real (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, row_vector y) : real (matrix_operations.html)\n\n\n(vector x, vector y) : real (matrix_operations.html)\n\n\ndot_self:\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\ndouble_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\n\n\n\ne:\n\n\n() : real (real-valued_basic_functions.html)\n\n\neigendecompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_vector) (matrix_operations.html)\n\n\neigendecompose_sym:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector) (matrix_operations.html)\n\n\neigenvalues:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : complex_vector (matrix_operations.html)\n\n\neigenvalues_sym:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\neigenvectors:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (matrix_operations.html)\n\n\neigenvectors_sym:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nerf:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nerfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp2:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp_mod_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_cdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_rng:\n\n\n(reals mu, reals sigma, reals lambda) : R (unbounded_continuous_distributions.html)\n\n\nexpm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexponential:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nexponential_cdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lccdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lcdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lpdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lupdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_rng:\n\n\n(reals beta) : R (positive_continuous_distributions.html)\n\n\n\n\n\nfalling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfatal_error:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nfdim:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfft:\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\nfft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\nfloor:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nfma:\n\n\n(real x, real y, real z) : real (real-valued_basic_functions.html)\n\n\nfmax:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmin:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmod:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfrechet:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nfrechet_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\n\n\n\ngamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ngamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_p:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_q:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ngaussian_dlm_obs:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lpdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lupdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngeneralized_inverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ngenerate_laplace_options:\n\n\n(int dimension) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\n(vector theta_init) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\nget_imag:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\nget_real:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\ngp_dot_prod_cov:\n\n\n(array[] real x, real sigma) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma) : matrix (matrix_operations.html)\n\n\ngp_exp_quad_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\ngp_exponential_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern23_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern52_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_periodic_cov:\n\n\n(array[] real x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\ngumbel:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ngumbel_cdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lccdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lcdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lpdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lupdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_rng:\n\n\n(reals mu, reals beta) : R (unbounded_continuous_distributions.html)\n\n\n\n\n\nhead:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\nhmm_hidden_state_prob:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : matrix (hidden_markov_models.html)\n\n\nhmm_latent_rng:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : array[] int (hidden_markov_models.html)\n\n\nhmm_marginal:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : real (hidden_markov_models.html)\n\n\nhypergeometric:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nhypergeometric_1F0:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_2F1:\n\n\n(real a1, real a2, real b1, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_3F2:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_lpmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_lupmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_pFq:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_rng:\n\n\n(int N, int a, int2 b) : int (bounded_discrete_distributions.html)\n\n\nhypot:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n\n\n\nidentity_matrix:\n\n\n(int k) : matrix (matrix_operations.html)\n\n\ninc_beta:\n\n\n(real alpha, real beta, real x) : real (real-valued_basic_functions.html)\n\n\nint_step:\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\nintegrate_1d:\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real (higher-order_functions.html)\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real (higher-order_functions.html)\n\n\nintegrate_ode:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_adams:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_bdf:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_rk45:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\ninv:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_chi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\ninv_cloglog:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_erfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_fft:\n\n\n(complex_vector u) : complex_vector (complex_matrix_operations.html)\n\n\ninv_fft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\ninv_gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_gamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ninv_inc_beta:\n\n\n(real alpha, real beta, real p) : real (real-valued_basic_functions.html)\n\n\ninv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_Phi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_sqrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_square:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_wishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\ninv_wishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\ninverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ninverse_spd:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nis_inf:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\nis_nan:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\n\n\n\nlambert_w0:\n\n\n(reals x) : R (real-valued_basic_functions.html)\n\n\nlambert_wm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlaplace_latent_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_marginal:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlchoose:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nldexp:\n\n\n(real x, int y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlinspaced_array:\n\n\n(int n, data real lower, data real upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_int_array:\n\n\n(int n, int lower, int upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_row_vector:\n\n\n(int n, data real lower, data real upper) : row_vector (matrix_operations.html)\n\n\nlinspaced_vector:\n\n\n(int n, data real lower, data real upper) : vector (matrix_operations.html)\n\n\nlkj_corr:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lpdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lupdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlkj_corr_lpdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_lupdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlmgamma:\n\n\n(int n, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlmultiply:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog10:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_determinant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\nlog_diff_exp:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\nlog_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_inv_logit_diff:\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_mix:\n\n\n(real theta, real lp1, real lp2) : real (real-valued_basic_functions.html)\n\n\n(T1 thetas, T2 lps) : R (real-valued_basic_functions.html)\n\n\nlog_modified_bessel_first_kind:\n\n\n(real v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_rising_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_softmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nlog_sum_exp:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nlogistic:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nlogistic_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nlogit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nloglogistic:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nloglogistic_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlognormal:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nlognormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlower_bound_constrain:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_jacobian:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_unconstrain:\n\n\n(reals x, reals lb) : reals (transform_functions.html)\n\n\nlower_upper_bound_constrain:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_jacobian:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_unconstrain:\n\n\n(reals x, reals lb, reals ub) : reals (transform_functions.html)\n\n\n\n\n\nmachine_precision:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nmap_rect:\n\n\n(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector (higher-order_functions.html)\n\n\nmatrix_exp:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nmatrix_exp_multiply:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nmatrix_power:\n\n\n(matrix A, int B) : matrix (matrix_operations.html)\n\n\nmax:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmdivide_left_spd:\n\n\n(matrix A, matrix B) : vector (matrix_operations.html)\n\n\n(matrix A, vector b) : matrix (matrix_operations.html)\n\n\nmdivide_left_tri_low:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\nmdivide_right_spd:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmdivide_right_tri_low:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmean:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmin:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmodified_bessel_first_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmodified_bessel_second_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmulti_gp:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lpdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lupdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lpdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lupdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_rng:\n\n\n(row_vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_rng:\n\n\n(row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_cholesky_t_rng:\n\n\n(real nu, vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lpdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lupdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_rng:\n\n\n(real nu, array[] row_vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, array[] vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lpdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lupdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_rng:\n\n\n(real nu, row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmultinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lpmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lupmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_rng:\n\n\n(vector gamma, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultinomial_lpmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_lupmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_rng:\n\n\n(vector theta, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultiply_lower_tri_self_transpose:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n\n\n\nneg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_cdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lccdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lcdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lpmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lupmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_rng:\n\n\n(reals eta, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lpmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lupmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_rng:\n\n\n(reals mu, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_cdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lccdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lcdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lpmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lupmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_rng:\n\n\n(reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nnegative_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnorm:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nnorm1:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnorm2:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnormal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lpdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lupdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nnot_a_number:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnum_elements:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\n\n\n\node_adams:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_adams_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_adjoint_tol_ctl:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector (higher-order_functions.html)\n\n\node_bdf:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_bdf_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\noffset_multiplier_constrain:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_jacobian:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_unconstrain:\n\n\n(reals x, reals offset, reals mult) : reals (transform_functions.html)\n\n\none_hot_array:\n\n\n(int n, int k) : array[] real (matrix_operations.html)\n\n\none_hot_int_array:\n\n\n(int n, int k) : array[] int (matrix_operations.html)\n\n\none_hot_row_vector:\n\n\n(int n, int k) : row_vector (matrix_operations.html)\n\n\none_hot_vector:\n\n\n(int K, int k) : vector (matrix_operations.html)\n\n\nones_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nones_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nones_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nones_vector:\n\n\n(int n) : vector (matrix_operations.html)\n\n\noperator!:\n\n\n(int x) : int (real-valued_basic_functions.html)\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\noperator!=:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator%/%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator&&:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\n&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;operator’:\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\noperator*:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : vector (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, row_vector y) : matrix (matrix_operations.html)\n\n\noperator*=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator+:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator+=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator-:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(T x) : T (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (matrix_operations.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator-=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.*:\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator.*=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator./:\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator./=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.^:\n\n\n( complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n( complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator/:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix B, complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector b, complex_matrix A) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\noperator/=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator&lt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&lt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\noperator==:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator\\:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\noperator^:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\noperator||:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\nordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_logistic:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lpmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lupmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lpmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lupmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_probit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_probit_lpmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_lupmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nowens_t:\n\n\n(real h, real a) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n\n\n\npareto:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_cdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lccdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lcdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lpdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lupdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_rng:\n\n\n(reals y_min, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\npareto_type_2:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_cdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lccdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lcdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lpdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lupdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_rng:\n\n\n(reals mu, reals lambda, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\nPhi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nPhi_approx:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\npi:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npoisson:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_cdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lccdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lcdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\npoisson_lpmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lupmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_rng:\n\n\n(reals lambda) : R (unbounded_discrete_distributions.html)\n\n\npolar:\n\n\n(real r, real theta) : complex (complex-valued_basic_functions.html)\n\n\npositive_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npositive_ordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\npow:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : Z (complex-valued_basic_functions.html)\n\n\nprint:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nprod:\n\n\n(array[] int x) : real (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nproj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n\n\n\nqr:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_thin_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nquad_form:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquad_form_diag:\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\nquad_form_sym:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquantile:\n\n\n(data array[] real x, data array[] real p) : array[] real (array_operations.html)\n\n\n(data array[] real x, data real p) : real (array_operations.html)\n\n\n(data row_vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data row_vector x, data real p) : real (matrix_operations.html)\n\n\n(data vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data vector x, data real p) : real (matrix_operations.html)\n\n\n\n\n\nrank:\n\n\n(array[] int v, int s) : int (array_operations.html)\n\n\n(array[] real v, int s) : int (array_operations.html)\n\n\n(row_vector v, int s) : int (matrix_operations.html)\n\n\n(vector v, int s) : int (matrix_operations.html)\n\n\nrayleigh:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nrayleigh_cdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lccdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lcdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lpdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lupdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_rng:\n\n\n(reals sigma) : R (positive_continuous_distributions.html)\n\n\nreduce_sum:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreduce_sum_static:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreject:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nrep_array:\n\n\n(T x, int k, int m, int n) : array[,,] T (array_operations.html)\n\n\n(T x, int m, int n) : array[,] T (array_operations.html)\n\n\n(T x, int n) : array[] T (array_operations.html)\n\n\nrep_matrix:\n\n\n(complex z, int m, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector rv, int m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(real x, int m, int n) : matrix (matrix_operations.html)\n\n\n(row_vector rv, int m) : matrix (matrix_operations.html)\n\n\n(vector v, int n) : matrix (matrix_operations.html)\n\n\nrep_row_vector:\n\n\n(complex z, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(real x, int n) : row_vector (matrix_operations.html)\n\n\nrep_vector:\n\n\n(complex z, int m) : complex_vector (complex_matrix_operations.html)\n\n\n(real x, int m) : vector (matrix_operations.html)\n\n\nreverse:\n\n\n(array[] T v) : array[] T (array_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nrising_factorial:\n\n\n(real x, int n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nround:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nrow:\n\n\n(complex_matrix x, int m) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int m) : row_vector (matrix_operations.html)\n\n\nrows:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nrows_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\nrows_dot_self:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n\n\n\nscale_matrix_exp_multiply:\n\n\n(real t, matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nscaled_inv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_cdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lccdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lcdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lpdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lupdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_rng:\n\n\n(reals nu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nsd:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsegment:\n\n\n(array[] T sv, int i, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int i, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int i, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int i, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int i, int n) : vector (matrix_operations.html)\n\n\nsimplex_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsingular_values:\n\n\n(complex_matrix A) : vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\nsinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsize:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nskew_double_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nskew_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_normal_cdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lccdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lcdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lpdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lupdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_rng:\n\n\n(reals xi, reals omega, real alpha) : R (unbounded_continuous_distributions.html)\n\n\nsoftmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nsolve_newton:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_newton_tol:\n\n\n(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell_tol:\n\n\n(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsort_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_indices_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsort_indices_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsqrt:\n\n\n(complex x) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsqrt2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nsquare:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsquared_distance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\nstd_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstd_normal_cdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lccdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lcdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_log_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_lpdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lupdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_rng:\n\n\n() : real (unbounded_continuous_distributions.html)\n\n\nstep:\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\nstochastic_column_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstochastic_row_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstudent_t:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstudent_t_cdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lccdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lcdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lpdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lupdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_rng:\n\n\n(reals nu, reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nsub_col:\n\n\n(complex_matrix x, int i, int j, int n_rows) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows) : vector (matrix_operations.html)\n\n\nsub_row:\n\n\n(complex_matrix x, int i, int j, int n_cols) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_cols) : row_vector (matrix_operations.html)\n\n\nsum:\n\n\n(array[] complex x) : complex (array_operations.html)\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsum_to_zero_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsvd:\n\n\n(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector, matrix) (matrix_operations.html)\n\n\nsvd_U:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsvd_V:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsymmetrize_from_lower_tri:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\n\n\n\ntail:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\ntan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntarget:\n\n\n() : real (real-valued_basic_functions.html)\n\n\ntcrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ntgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nto_array_1d:\n\n\n(array[...] complex a) : array[] complex (mixed_operations.html)\n\n\n(array[...] int a) : array[] int (mixed_operations.html)\n\n\n(array[...] real a) : array[] real (mixed_operations.html)\n\n\n(complex_matrix m) : array[] complex (mixed_operations.html)\n\n\n(complex_row_vector v) : array[] complex (mixed_operations.html)\n\n\n(complex_vector v) : array[] real (mixed_operations.html)\n\n\n(matrix m) : array[] real (mixed_operations.html)\n\n\n(row_vector v) : array[] real (mixed_operations.html)\n\n\n(vector v) : array[] real (mixed_operations.html)\n\n\nto_array_2d:\n\n\n(complex_matrix m) : array[,] real (mixed_operations.html)\n\n\n(matrix m) : array[,] real (mixed_operations.html)\n\n\nto_complex:\n\n\n() : complex (complex-valued_basic_functions.html)\n\n\n(real re) : complex (complex-valued_basic_functions.html)\n\n\n(real re, real im) : complex (complex-valued_basic_functions.html)\n\n\n(T1 re, T2 im) : Z (complex-valued_basic_functions.html)\n\n\nto_int:\n\n\n(data real x) : int (integer-valued_basic_functions.html)\n\n\nto_matrix:\n\n\n(array[,] complex a ) : complex_matrix (mixed_operations.html)\n\n\n(array[,] int a) : matrix (mixed_operations.html)\n\n\n(array[,] real a) : matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex_row_vector vs) : complex_matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] row_vector vs) : matrix (mixed_operations.html)\n\n\n(complex_matrix A, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix m) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix M, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(matrix A, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(matrix m) : matrix (mixed_operations.html)\n\n\n(matrix M, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(vector v) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\nto_row_vector:\n\n\n(array[] complex a) : complex_row_vector (mixed_operations.html)\n\n\n(array[] int a) : row_vector (mixed_operations.html)\n\n\n(array[] real a) : row_vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_row_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(matrix m) : row_vector (mixed_operations.html)\n\n\n(row_vector v) : row_vector (mixed_operations.html)\n\n\n(vector v) : row_vector (mixed_operations.html)\n\n\nto_vector:\n\n\n(array[] complex a) : complex_vector (mixed_operations.html)\n\n\n(array[] int a) : vector (mixed_operations.html)\n\n\n(array[] real a) : vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_vector (mixed_operations.html)\n\n\n(matrix m) : vector (mixed_operations.html)\n\n\n(row_vector v) : vector (mixed_operations.html)\n\n\n(vector v) : vector (mixed_operations.html)\n\n\ntrace:\n\n\n(complex_matrix A) : complex (complex_matrix_operations.html)\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ntrace_dot:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_gen_quad_form:\n\n\n(matrix D ,matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_quad_form:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\ntrigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntrunc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\n\n\n\nuniform:\n\n\ndistribution statement (bounded_continuous_distributions.html)\n\n\nuniform_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_rng:\n\n\n(reals alpha, reals beta) : R (bounded_continuous_distributions.html)\n\n\nuniform_simplex:\n\n\n(int n) : vector (matrix_operations.html)\n\n\nunit_vectors_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nupper_bound_constrain:\n\n\n(reals y, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_jacobian:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_unconstrain:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\n\n\n\nvariance:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nvon_mises:\n\n\ndistribution statement (circular_distributions.html)\n\n\nvon_mises_cdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lccdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lcdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lpdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lupdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_rng:\n\n\n(reals mu, reals kappa) : R (circular_distributions.html)\n\n\n\n\n\nweibull:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nweibull_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\nwiener:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\nwiener_lccdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lcdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lpdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lupdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\nwishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\n\n\n\nyule_simon:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nyule_simon_cdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lccdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lcdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\n\n\n\nzeros_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nzeros_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nzeros_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nzeros_vector:\n\n\n(int n) : vector (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Real-Valued Basic Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/removed_functions.html",
-    "href": "functions-reference/removed_functions.html",
-    "title": "Removed Functions",
+    "objectID": "functions-reference/functions_index.html#a",
+    "href": "functions-reference/functions_index.html#a",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Functions which once existed in the Stan language and have since been replaced or removed are listed here.\n\n\nRemoved: Currently two non-conforming functions ending in suffix _log.\nReplacement: Replace multiply_log(...) with lmultiply(...). Replace binomial_coefficient_log(...) with lchoose(...).\nRemoved In: Stan 2.33\n\n\n\nRemoved: The built-in no-argument function get_lp() is deprecated.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33\n\n\n\nRemoved: The unary function fabs is deprecated.\nReplacement: Use the unary function abs instead. Note that the return type for abs is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.\nRemoved In: Stan 2.33\n\n\n\nThese covariance functions have been replaced by those described in Gaussian Process Covariance Functions\nWith magnitude \\(\\alpha\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(x_i, x_j) = \\alpha^2 \\exp \\left(-\\dfrac{1}{2\\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \\right)\n\\]\n \n\nmatrix cov_exp_quad(row_vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(row_vectors x1, row_vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x1, vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x1, array[] real x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n\n\n\nRemoved: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.\nReplacement: Explicit x != 0 comparison is preferred instead.\nRemoved In: Stan 2.34",
+    "text": "abs:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\nacos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nacosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nadd_diag:\n\n\n(complex_matrix m, complex_real d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_row_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector d) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, real d) : matrix (matrix_operations.html)\n\n\n(matrix m, row_vector d) : matrix (matrix_operations.html)\n\n\n(matrix m, vector d) : matrix (matrix_operations.html)\n\n\nalgebra_solver:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nalgebra_solver_newton:\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i) : vector (deprecated_functions.html)\n\n\n(function algebra_system, vector y_guess, vector theta, data array[] real x_r, array[] int x_i, data real rel_tol, data real f_tol, int max_steps) : vector (deprecated_functions.html)\n\n\nappend_array:\n\n\n(T x, T y) : T (array_operations.html)\n\n\nappend_col:\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(vector x, vector y) : matrix (matrix_operations.html)\n\n\nappend_row:\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, row_vector y) : matrix (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : matrix (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\narg:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nasin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nasinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\natan2:\n\n\n(T y, T x) : R (real-valued_basic_functions.html)\n\n\natanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Removed Functions"
-    ]
+      "Appendix",
+      "Alphabetical Index"
+    ]
   },
   {
-    "objectID": "functions-reference/removed_functions.html#multiply_log-and-binomial_coefficient_log-functions",
-    "href": "functions-reference/removed_functions.html#multiply_log-and-binomial_coefficient_log-functions",
-    "title": "Removed Functions",
+    "objectID": "functions-reference/functions_index.html#b",
+    "href": "functions-reference/functions_index.html#b",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Removed: Currently two non-conforming functions ending in suffix _log.\nReplacement: Replace multiply_log(...) with lmultiply(...). Replace binomial_coefficient_log(...) with lchoose(...).\nRemoved In: Stan 2.33",
+    "text": "bernoulli:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_cdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lccdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lcdf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_logit:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm:\n\n\ndistribution statement (binary_distributions.html)\n\n\nbernoulli_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (binary_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (binary_distributions.html)\n\n\nbernoulli_logit_glm_rng:\n\n\n(matrix x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\n(row_vector x, vector alpha, vector beta) : array[] int (binary_distributions.html)\n\n\nbernoulli_logit_lpmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_lupmf:\n\n\n(ints y | reals alpha) : real (binary_distributions.html)\n\n\nbernoulli_logit_rng:\n\n\n(reals alpha) : R (binary_distributions.html)\n\n\nbernoulli_lpmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_lupmf:\n\n\n(ints y | reals theta) : real (binary_distributions.html)\n\n\nbernoulli_rng:\n\n\n(reals theta) : R (binary_distributions.html)\n\n\nbessel_first_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbessel_second_kind:\n\n\n(int v, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_binomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbeta_binomial_cdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lccdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lcdf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lpmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_lupmf:\n\n\n(ints n | ints N, reals alpha, reals beta) : real (bounded_discrete_distributions.html)\n\n\nbeta_binomial_rng:\n\n\n(ints N, reals alpha, reals beta) : R (bounded_discrete_distributions.html)\n\n\nbeta_cdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lccdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lcdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lpdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_lupdf:\n\n\n(reals theta | reals alpha, reals beta) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_cdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lccdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lcdf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lpmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_lupmf:\n\n\n(ints n | reals r, reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nbeta_neg_binomial_rng:\n\n\n(reals r, reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nbeta_proportion:\n\n\ndistribution statement (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lccdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lcdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lpdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_lupdf:\n\n\n(reals theta | reals mu, reals kappa) : real (continuous_distributions_on_0_1.html)\n\n\nbeta_proportion_rng:\n\n\n(reals mu, reals kappa) : R (continuous_distributions_on_0_1.html)\n\n\nbeta_rng:\n\n\n(reals alpha, reals beta) : R (continuous_distributions_on_0_1.html)\n\n\nbinary_log_loss:\n\n\n(int y, real y_hat) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nbinomial:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_cdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lccdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lcdf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lpmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_glm_lupmf:\n\n\n(array[] int n | array[] int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int n | array[] int N, row_vector x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, real alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\n(int n | int N, matrix x, vector alpha, vector beta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lpmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_logit_lupmf:\n\n\n(ints n | ints N, reals alpha) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lpmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_lupmf:\n\n\n(ints n | ints N, reals theta) : real (bounded_discrete_distributions.html)\n\n\nbinomial_rng:\n\n\n(ints N, reals theta) : R (bounded_discrete_distributions.html)\n\n\nblock:\n\n\n(complex_matrix x, int i, int j, int n_rows, int n_cols) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows, int n_cols) : matrix (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Removed Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/removed_functions.html#get_lp-function",
-    "href": "functions-reference/removed_functions.html#get_lp-function",
-    "title": "Removed Functions",
+    "objectID": "functions-reference/functions_index.html#c",
+    "href": "functions-reference/functions_index.html#c",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Removed: The built-in no-argument function get_lp() is deprecated.\nReplacement: Use the no-argument function target() instead.\nRemoved In: Stan 2.33",
+    "text": "categorical:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lpmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_glm_lupmf:\n\n\n(array[] int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector alpha, matrix beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lpmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_lupmf:\n\n\n(ints y | vector beta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_logit_rng:\n\n\n(vector beta) : int (bounded_discrete_distributions.html)\n\n\ncategorical_lpmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_lupmf:\n\n\n(ints y | vector theta) : real (bounded_discrete_distributions.html)\n\n\ncategorical_rng:\n\n\n(vector theta) : int (bounded_discrete_distributions.html)\n\n\ncauchy:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ncauchy_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ncauchy_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\ncbrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nceil:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nchi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nchi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\nchi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\nchol2inv:\n\n\n(matrix L) : matrix (matrix_operations.html)\n\n\ncholesky_decompose:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ncholesky_factor_corr_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncholesky_factor_corr_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncholesky_factor_cov_constrain:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_jacobian:\n\n\n(vectors y, int M, int N) : matrices (transform_functions.html)\n\n\ncholesky_factor_cov_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\nchoose:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ncol:\n\n\n(complex_matrix x, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int n) : vector (matrix_operations.html)\n\n\ncols:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\ncolumns_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : row_vector (matrix_operations.html)\n\n\ncolumns_dot_self:\n\n\n(complex_matrix x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : row_vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\ncomplex_schur_decompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_matrix) (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_t:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\ncomplex_schur_decompose_u:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\nconj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(Z z) : Z (complex-valued_basic_functions.html)\n\n\ncorr_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncorr_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncos:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncosh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ncov_exp_quad:\n\n\n(array[] real x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(array[] real x1, array[] real x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(row_vectors x1, row_vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x, real alpha, real rho) : matrix (removed_functions.html)\n\n\n(vectors x1, vectors x2, real alpha, real rho) : matrix (removed_functions.html)\n\n\ncov_matrix_constrain:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_jacobian:\n\n\n(vectors y, int K) : matrices (transform_functions.html)\n\n\ncov_matrix_unconstrain:\n\n\n(matrices x) : vectors (transform_functions.html)\n\n\ncrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ncsr_extract:\n\n\n(matrix a) : tuple(vector, array[] int, array[] int) (sparse_matrix_operations.html)\n\n\ncsr_extract_u:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_v:\n\n\n(matrix a) : array[] int (sparse_matrix_operations.html)\n\n\ncsr_extract_w:\n\n\n(matrix a) : vector (sparse_matrix_operations.html)\n\n\ncsr_matrix_times_vector:\n\n\n(int m, int n, vector w, array[] int v, array[] int u, vector b) : vector (sparse_matrix_operations.html)\n\n\ncsr_to_dense_matrix:\n\n\n(int m, int n, vector w, array[] int v, array[] int u) : matrix (sparse_matrix_operations.html)\n\n\ncumulative_sum:\n\n\n(array[] complex x) : array[] complex (complex_matrix_operations.html)\n\n\n(array[] int x) : array[] int (matrix_operations.html)\n\n\n(array[] real x) : array[] real (matrix_operations.html)\n\n\n(complex_row_vector rv) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Removed Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/removed_functions.html#fabs-function",
-    "href": "functions-reference/removed_functions.html#fabs-function",
-    "title": "Removed Functions",
+    "objectID": "functions-reference/functions_index.html#d",
+    "href": "functions-reference/functions_index.html#d",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Removed: The unary function fabs is deprecated.\nReplacement: Use the unary function abs instead. Note that the return type for abs is different for integer overloads, but this replacement is safe due to Stan’s type promotion rules.\nRemoved In: Stan 2.33",
+    "text": "dae:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\ndae_tol:\n\n\n(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\ndeterminant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ndiag_matrix:\n\n\n(complex_vector x) : complex_matrix (complex_matrix_operations.html)\n\n\n(vector x) : matrix (matrix_operations.html)\n\n\ndiag_post_multiply:\n\n\n(complex_matrix m, complex_row_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix m, complex_vector v) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\ndiag_pre_multiply:\n\n\n(complex_row_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\n(row_vector rv, matrix m) : matrix (matrix_operations.html)\n\n\n(vector v, matrix m) : matrix (matrix_operations.html)\n\n\ndiagonal:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\ndigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ndims:\n\n\n(T x) : array[] int (array_operations.html)\n\n\ndirichlet:\n\n\ndistribution statement (simplex_distributions.html)\n\n\ndirichlet_lpdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_lupdf:\n\n\n(vectors theta | vectors alpha) : real (simplex_distributions.html)\n\n\ndirichlet_multinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lpmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_lupmf:\n\n\n(array[] int y | vector alpha) : real (multivariate_discrete_distributions.html)\n\n\ndirichlet_multinomial_rng:\n\n\n(vector alpha, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\ndirichlet_rng:\n\n\n(vector alpha) : vector (simplex_distributions.html)\n\n\ndiscrete_range:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\ndiscrete_range_cdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lccdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lcdf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lpmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_lupmf:\n\n\n(ints y | ints l, ints u) : real (bounded_discrete_distributions.html)\n\n\ndiscrete_range_rng:\n\n\n(ints l, ints u) : ints (bounded_discrete_distributions.html)\n\n\ndistance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\ndot_product:\n\n\n(complex_row_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(row_vector x, row_vector y) : real (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, row_vector y) : real (matrix_operations.html)\n\n\n(vector x, vector y) : real (matrix_operations.html)\n\n\ndot_self:\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\ndouble_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\ndouble_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Removed Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/removed_functions.html#cov_exp_quad",
-    "href": "functions-reference/removed_functions.html#cov_exp_quad",
-    "title": "Removed Functions",
+    "objectID": "functions-reference/functions_index.html#e",
+    "href": "functions-reference/functions_index.html#e",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "These covariance functions have been replaced by those described in Gaussian Process Covariance Functions\nWith magnitude \\(\\alpha\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(x_i, x_j) = \\alpha^2 \\exp \\left(-\\dfrac{1}{2\\rho^2} \\sum_{d=1}^D (x_{i,d} - x_{j,d})^2 \\right)\n\\]\n \n\nmatrix cov_exp_quad(row_vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x.\nAvailable since 2.16, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(row_vectors x1, row_vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(vectors x1, vectors x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33\n \n\nmatrix cov_exp_quad(array[] real x1, array[] real x2, real alpha, real rho) The covariance matrix with an exponentiated quadratic kernel of x1 and x2.\nAvailable since 2.18, deprecated since 2.20, removed in in 2.33",
+    "text": "e:\n\n\n() : real (real-valued_basic_functions.html)\n\n\neigendecompose:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(complex_matrix, complex_vector) (matrix_operations.html)\n\n\neigendecompose_sym:\n\n\n(complex_matrix A) : tuple(complex_matrix, complex_vector) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector) (matrix_operations.html)\n\n\neigenvalues:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : complex_vector (matrix_operations.html)\n\n\neigenvalues_sym:\n\n\n(complex_matrix A) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\neigenvectors:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : complex_matrix (matrix_operations.html)\n\n\neigenvectors_sym:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nerf:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nerfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp2:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexp_mod_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_cdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals lambda) : real (unbounded_continuous_distributions.html)\n\n\nexp_mod_normal_rng:\n\n\n(reals mu, reals sigma, reals lambda) : R (unbounded_continuous_distributions.html)\n\n\nexpm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nexponential:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nexponential_cdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lccdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lcdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lpdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_lupdf:\n\n\n(reals y | reals beta) : real (positive_continuous_distributions.html)\n\n\nexponential_rng:\n\n\n(reals beta) : R (positive_continuous_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Removed Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/removed_functions.html#real-arguments-to-logical-operators-operator-operator-and-operator",
-    "href": "functions-reference/removed_functions.html#real-arguments-to-logical-operators-operator-operator-and-operator",
-    "title": "Removed Functions",
+    "objectID": "functions-reference/functions_index.html#f",
+    "href": "functions-reference/functions_index.html#f",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Removed: A nonzero real number (even NaN) was interpreted as true and a zero was interpreted as false.\nReplacement: Explicit x != 0 comparison is preferred instead.\nRemoved In: Stan 2.34",
+    "text": "falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfatal_error:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nfdim:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfft:\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\nfft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\nfloor:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nfma:\n\n\n(real x, real y, real z) : real (real-valued_basic_functions.html)\n\n\nfmax:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmin:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfmod:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nfrechet:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nfrechet_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nfrechet_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Removed Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/sparse_matrix_operations.html",
-    "href": "functions-reference/sparse_matrix_operations.html",
-    "title": "Sparse Matrix Operations",
+    "objectID": "functions-reference/functions_index.html#g",
+    "href": "functions-reference/functions_index.html#g",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "For sparse matrices, for which many elements are zero, it is more efficient to use specialized representations to save memory and speed up matrix arithmetic (including derivative calculations). Given Stan’s implementation, there is substantial space (memory) savings by using sparse matrices. Because of the ease of optimizing dense matrix operations, speed improvements only arise at 90% or even greater sparsity; below that level, dense matrices are faster but use more memory.\nBecause of this speedup and space savings, it may even be useful to read in a dense matrix and convert it to a sparse matrix before multiplying it by a vector. This chapter covers a very specific form of sparsity consisting of a sparse matrix multiplied by a dense vector.\n\n\nSparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix \\[\\begin{equation*}\nA = \\begin{bmatrix} 19 & 27 & 0 & 0 \\\\ 0 & 0 & 0 & 0 \\\\ 0 & 0 & 0 & 52 \\\\ 81 & 0 & 95 & 33 \\end{bmatrix}\n\\end{equation*}\\] is translated into a vector of the non-zero real values, read by row from the matrix \\(A\\), \\[\\begin{equation*}\nw(A) = \\begin{bmatrix} 19 & 27 & 52 & 81 & 95 & 33 \\end{bmatrix}^{\\top} \\! \\! \\! ,\n\\end{equation*}\\] an array of integer column indices for the values, \\[\\begin{equation*}\nv(A) = \\begin{bmatrix} 1 & 2 & 4 & 1 & 3 & 4 \\end{bmatrix} \\! ,\n\\end{equation*}\\] and an array of integer indices indicating where in \\(w(A)\\) a given row’s values start, \\[\\begin{equation*} u(A) = \\begin{bmatrix} 1 & 3 & 3 & 4 & 7 \\end{bmatrix} \\! ,\n\\end{equation*}\\] with a padded value at the end to guarantee that \\[\\begin{equation*} u(A)[n+1] - u(A)[n] \\end{equation*}\\] is the number of non-zero elements in row \\(n\\) of the matrix (here \\(2\\), \\(0\\), \\(1\\), and \\(3\\)). Note that because the second row has no non-zero elements both the second and third elements of \\(u(A)\\) correspond to the third element of \\(w(A)\\), which is \\(52\\). The values \\((w(A), \\, v(A), \\, u(A))\\) are sufficient to reconstruct \\(A\\).\nThe values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly \\(12 K + M\\) bytes plus a small constant overhead, which is often considerably fewer bytes than the \\(M \\times\nN\\) required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.\n\n\n\nConversion functions between dense and sparse matrices are provided.\n\n\nConverting a dense matrix \\(m\\) to a sparse representation produces a vector \\(w\\) and two integer arrays, \\(u\\) and \\(v\\).\n \n\nvector csr_extract_w(matrix a) Return non-zero values in matrix a; see section compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_v(matrix a) Return column indices for values in csr_extract_w(a); see compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_u(matrix a) Return array of row starting indices for entries in csr_extract_w(a) followed by the size of csr_extract_w(a) plus one; see section compressed row storage.\nAvailable since 2.8\n \n\ntuple(vector, array[] int, array[] int) csr_extract(matrix a) Return all three components of the CSR representation of the matrix a; see section compressed row storage. This function is equivalent to (csr_extract_w(a), csr_extract_v(a), csr_extract_u(a)).\nAvailable since 2.33\n\n\n\nTo convert a sparse matrix representation to a dense matrix, there is a single function.\n \n\nmatrix csr_to_dense_matrix(int m, int n, vector w, array[] int v, array[] int u) Return dense \\(\\text{m} \\times \\text{n}\\) matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section compressed row storage for more details.\nAvailable since 2.10\n\n\n\n\n\n\nThe only supported operation is the multiplication of a sparse matrix \\(A\\) and a dense vector \\(b\\) to produce a dense vector \\(A\\,b\\). Multiplying a dense row vector \\(b\\) and a sparse matrix \\(A\\) can be coded using transposition as \\[\\begin{equation*}\nb \\, A = (A^{\\top} \\, b^{\\top})^{\\top},\n\\end{equation*}\\] but care must be taken to represent \\(A^{\\top}\\) rather than \\(A\\) as a sparse matrix.\n \n\nvector csr_matrix_times_vector(int m, int n, vector w, array[] int v, array[] int u, vector b) Multiply the \\(\\text{m} \\times \\text{n}\\) matrix represented by values w, column indices v, and row start indices u by the vector b; see compressed row storage.\nAvailable since 2.18",
+    "text": "gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ngamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ngamma_p:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_q:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\ngamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ngaussian_dlm_obs:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lpdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngaussian_dlm_obs_lupdf:\n\n\n(matrix y | matrix F, matrix G, matrix V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\n(matrix y | matrix F, matrix G, vector V, matrix W, vector m0, matrix C0) : real (distributions_over_unbounded_vectors.html)\n\n\ngeneralized_inverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ngenerate_laplace_options:\n\n\n(int dimension) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\n(vector theta_init) : tuple(vector, real, int, int, int, int) (embedded_laplace.html)\n\n\nget_imag:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\nget_real:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\ngp_dot_prod_cov:\n\n\n(array[] real x, real sigma) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma) : matrix (matrix_operations.html)\n\n\ngp_exp_quad_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\ngp_exponential_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern23_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_matern52_cov:\n\n\n(array[] real x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, array[] real length_scale) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale) : matrix (matrix_operations.html)\n\n\ngp_periodic_cov:\n\n\n(array[] real x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(array[] real x1, array[] real x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\n(vectors x1, vectors x2, real sigma, real length_scale, real period) : matrix (matrix_operations.html)\n\n\ngumbel:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\ngumbel_cdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lccdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lcdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lpdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_lupdf:\n\n\n(reals y | reals mu, reals beta) : real (unbounded_continuous_distributions.html)\n\n\ngumbel_rng:\n\n\n(reals mu, reals beta) : R (unbounded_continuous_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Sparse Matrix Operations"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/sparse_matrix_operations.html#CSR",
-    "href": "functions-reference/sparse_matrix_operations.html#CSR",
-    "title": "Sparse Matrix Operations",
+    "objectID": "functions-reference/functions_index.html#h",
+    "href": "functions-reference/functions_index.html#h",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Sparse matrices are represented in Stan using compressed row storage (CSR). For example, the matrix \\[\\begin{equation*}\nA = \\begin{bmatrix} 19 & 27 & 0 & 0 \\\\ 0 & 0 & 0 & 0 \\\\ 0 & 0 & 0 & 52 \\\\ 81 & 0 & 95 & 33 \\end{bmatrix}\n\\end{equation*}\\] is translated into a vector of the non-zero real values, read by row from the matrix \\(A\\), \\[\\begin{equation*}\nw(A) = \\begin{bmatrix} 19 & 27 & 52 & 81 & 95 & 33 \\end{bmatrix}^{\\top} \\! \\! \\! ,\n\\end{equation*}\\] an array of integer column indices for the values, \\[\\begin{equation*}\nv(A) = \\begin{bmatrix} 1 & 2 & 4 & 1 & 3 & 4 \\end{bmatrix} \\! ,\n\\end{equation*}\\] and an array of integer indices indicating where in \\(w(A)\\) a given row’s values start, \\[\\begin{equation*} u(A) = \\begin{bmatrix} 1 & 3 & 3 & 4 & 7 \\end{bmatrix} \\! ,\n\\end{equation*}\\] with a padded value at the end to guarantee that \\[\\begin{equation*} u(A)[n+1] - u(A)[n] \\end{equation*}\\] is the number of non-zero elements in row \\(n\\) of the matrix (here \\(2\\), \\(0\\), \\(1\\), and \\(3\\)). Note that because the second row has no non-zero elements both the second and third elements of \\(u(A)\\) correspond to the third element of \\(w(A)\\), which is \\(52\\). The values \\((w(A), \\, v(A), \\, u(A))\\) are sufficient to reconstruct \\(A\\).\nThe values are structured so that there is a real value and integer column index for each non-zero entry in the array, plus one integer for each row of the matrix, plus one for padding. There is also underlying storage for internal container pointers and sizes. The total memory usage is roughly \\(12 K + M\\) bytes plus a small constant overhead, which is often considerably fewer bytes than the \\(M \\times\nN\\) required to store a dense matrix. Even more importantly, zero values do not introduce derivatives under multiplication or addition, so many storage and evaluation steps are saved when sparse matrices are multiplied.",
+    "text": "head:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\nhmm_hidden_state_prob:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : matrix (hidden_markov_models.html)\n\n\nhmm_latent_rng:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : array[] int (hidden_markov_models.html)\n\n\nhmm_marginal:\n\n\n(matrix log_omega, matrix Gamma, vector rho) : real (hidden_markov_models.html)\n\n\nhypergeometric:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nhypergeometric_1F0:\n\n\n(real a, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_2F1:\n\n\n(real a1, real a2, real b1, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_3F2:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_lpmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_lupmf:\n\n\n(int n | int N, int a, int b) : real (bounded_discrete_distributions.html)\n\n\nhypergeometric_pFq:\n\n\n(T1 a, T2 b, real z) : real (real-valued_basic_functions.html)\n\n\nhypergeometric_rng:\n\n\n(int N, int a, int2 b) : int (bounded_discrete_distributions.html)\n\n\nhypot:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Sparse Matrix Operations"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/sparse_matrix_operations.html#conversion-functions",
-    "href": "functions-reference/sparse_matrix_operations.html#conversion-functions",
-    "title": "Sparse Matrix Operations",
+    "objectID": "functions-reference/functions_index.html#i",
+    "href": "functions-reference/functions_index.html#i",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Conversion functions between dense and sparse matrices are provided.\n\n\nConverting a dense matrix \\(m\\) to a sparse representation produces a vector \\(w\\) and two integer arrays, \\(u\\) and \\(v\\).\n \n\nvector csr_extract_w(matrix a) Return non-zero values in matrix a; see section compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_v(matrix a) Return column indices for values in csr_extract_w(a); see compressed row storage.\nAvailable since 2.8\n \n\narray[] int csr_extract_u(matrix a) Return array of row starting indices for entries in csr_extract_w(a) followed by the size of csr_extract_w(a) plus one; see section compressed row storage.\nAvailable since 2.8\n \n\ntuple(vector, array[] int, array[] int) csr_extract(matrix a) Return all three components of the CSR representation of the matrix a; see section compressed row storage. This function is equivalent to (csr_extract_w(a), csr_extract_v(a), csr_extract_u(a)).\nAvailable since 2.33\n\n\n\nTo convert a sparse matrix representation to a dense matrix, there is a single function.\n \n\nmatrix csr_to_dense_matrix(int m, int n, vector w, array[] int v, array[] int u) Return dense \\(\\text{m} \\times \\text{n}\\) matrix with non-zero matrix entries w, column indices v, and row starting indices u; the vector w and array v must be the same size (corresponding to the total number of nonzero entries in the matrix), array v must have index values bounded by m, array u must have length equal to m + 1 and contain index values bounded by the number of nonzeros (except for the last entry, which must be equal to the number of nonzeros plus one). See section compressed row storage for more details.\nAvailable since 2.10",
+    "text": "identity_matrix:\n\n\n(int k) : matrix (matrix_operations.html)\n\n\ninc_beta:\n\n\n(real alpha, real beta, real x) : real (real-valued_basic_functions.html)\n\n\nint_step:\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\nintegrate_1d:\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) : real (higher-order_functions.html)\n\n\n(function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i), real relative_tolerance) : real (higher-order_functions.html)\n\n\nintegrate_ode:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_adams:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_bdf:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\nintegrate_ode_rk45:\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i) : array[,] real (deprecated_functions.html)\n\n\n(function ode, array[] real initial_state, real initial_time, array[] real times, array[] real theta, array[] real x_r, array[] int x_i, real rel_tol, real abs_tol, int max_num_steps) : array[,] real (deprecated_functions.html)\n\n\ninv:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_chi_square_cdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lccdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lcdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lpdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_lupdf:\n\n\n(reals y | reals nu) : real (positive_continuous_distributions.html)\n\n\ninv_chi_square_rng:\n\n\n(reals nu) : R (positive_continuous_distributions.html)\n\n\ninv_cloglog:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_erfc:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_fft:\n\n\n(complex_vector u) : complex_vector (complex_matrix_operations.html)\n\n\ninv_fft2:\n\n\n(complex_matrix m) : complex_matrix (complex_matrix_operations.html)\n\n\ninv_gamma:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\ninv_gamma_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\ninv_gamma_rng:\n\n\n(reals alpha, reals beta) : R (positive_continuous_distributions.html)\n\n\ninv_inc_beta:\n\n\n(real alpha, real beta, real p) : real (real-valued_basic_functions.html)\n\n\ninv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_Phi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_sqrt:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_square:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ninv_wishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\ninv_wishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\ninv_wishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)\n\n\ninverse:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\ninverse_spd:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nis_inf:\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\nis_nan:\n\n\n(real x) : int (real-valued_basic_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Sparse Matrix Operations"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/sparse_matrix_operations.html#sparse-matrix-arithmetic",
-    "href": "functions-reference/sparse_matrix_operations.html#sparse-matrix-arithmetic",
-    "title": "Sparse Matrix Operations",
+    "objectID": "functions-reference/functions_index.html#l",
+    "href": "functions-reference/functions_index.html#l",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "The only supported operation is the multiplication of a sparse matrix \\(A\\) and a dense vector \\(b\\) to produce a dense vector \\(A\\,b\\). Multiplying a dense row vector \\(b\\) and a sparse matrix \\(A\\) can be coded using transposition as \\[\\begin{equation*}\nb \\, A = (A^{\\top} \\, b^{\\top})^{\\top},\n\\end{equation*}\\] but care must be taken to represent \\(A^{\\top}\\) rather than \\(A\\) as a sparse matrix.\n \n\nvector csr_matrix_times_vector(int m, int n, vector w, array[] int v, array[] int u, vector b) Multiply the \\(\\text{m} \\times \\text{n}\\) matrix represented by values w, column indices v, and row start indices u by the vector b; see compressed row storage.\nAvailable since 2.18",
+    "text": "lambert_w0:\n\n\n(reals x) : R (real-valued_basic_functions.html)\n\n\nlambert_wm1:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlaplace_latent_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : vector (embedded_laplace.html)\n\n\nlaplace_latent_rng_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_bernoulli_logit_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_neg_binomial_2_log_rng:\n\n\n(array[] int y, array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_latent_tol_poisson_log_rng:\n\n\n(array[] int y, array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : vector (embedded_laplace.html)\n\n\nlaplace_marginal:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments) : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol:\n\n\n(function likelihood_function, tuple(...) likelihood_arguments, int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_bernoulli_logit_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lpmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_neg_binomial_2_log_lupmf:\n\n\n(array[] int y | array[] int y_index, real eta, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log:\n\n\ndistribution statement (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lpmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlaplace_marginal_tol_poisson_log_lupmf:\n\n\n(array[] int y | array[] int y_index, vector m, data int hessian_block_size, function covariance_function, tuple(...) covariance_arguments, tuple(vector, real, int, int, int, int) tolerances) : real (embedded_laplace.html)\n\n\nlbeta:\n\n\n(real alpha, real beta) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlchoose:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nldexp:\n\n\n(real x, int y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlinspaced_array:\n\n\n(int n, data real lower, data real upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_int_array:\n\n\n(int n, int lower, int upper) : array[] real (matrix_operations.html)\n\n\nlinspaced_row_vector:\n\n\n(int n, data real lower, data real upper) : row_vector (matrix_operations.html)\n\n\nlinspaced_vector:\n\n\n(int n, data real lower, data real upper) : vector (matrix_operations.html)\n\n\nlkj_corr:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky:\n\n\ndistribution statement (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lpdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_lupdf:\n\n\n(matrix L | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_cholesky_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlkj_corr_lpdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_lupdf:\n\n\n(matrix y | real eta) : real (correlation_matrix_distributions.html)\n\n\nlkj_corr_rng:\n\n\n(int K, real eta) : matrix (correlation_matrix_distributions.html)\n\n\nlmgamma:\n\n\n(int n, real x) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlmultiply:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog10:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1m_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog1p_exp:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_determinant:\n\n\n(matrix A) : real (matrix_operations.html)\n\n\nlog_diff_exp:\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_falling_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\nlog_inv_logit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nlog_inv_logit_diff:\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_mix:\n\n\n(real theta, real lp1, real lp2) : real (real-valued_basic_functions.html)\n\n\n(T1 thetas, T2 lps) : R (real-valued_basic_functions.html)\n\n\nlog_modified_bessel_first_kind:\n\n\n(real v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_rising_factorial:\n\n\n(real x, real n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nlog_softmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nlog_sum_exp:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nlogistic:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nlogistic_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nlogistic_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nlogit:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nloglogistic:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nloglogistic_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (positive_continuous_distributions.html)\n\n\nloglogistic_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlognormal:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nlognormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nlognormal_rng:\n\n\n(reals mu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nlower_bound_constrain:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_jacobian:\n\n\n(reals y, reals lb) : reals (transform_functions.html)\n\n\nlower_bound_unconstrain:\n\n\n(reals x, reals lb) : reals (transform_functions.html)\n\n\nlower_upper_bound_constrain:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_jacobian:\n\n\n(reals y, reals lb, reals ub) : reals (transform_functions.html)\n\n\nlower_upper_bound_unconstrain:\n\n\n(reals x, reals lb, reals ub) : reals (transform_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Sparse Matrix Operations"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html",
-    "href": "functions-reference/unbounded_continuous_distributions.html",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#m",
+    "href": "functions-reference/functions_index.html#m",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "The unbounded univariate continuous probability distributions have support on all real numbers.\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Normal}(y|\\mu,\\sigma) =\n\\frac{1}{\\sqrt{2 \\pi} \\ \\sigma} \\exp\\left( - \\, \\frac{1}{2} \\left(\\frac{y - \\mu}{\\sigma} \\right)^2 \\right) \\!.\n\\end{equation*}\\]\n\n\n\ny ~ normal(mu, sigma)\nIncrement target log probability density with normal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal normal_lpdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal normal_lupdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_cdf(reals y | reals mu, reals sigma) The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 1 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; the function Phi_approx is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.0\n \n\nreal normal_lcdf(reals y | reals mu, reals sigma) The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.12\n \n\nreal normal_lccdf(reals y | reals mu, reals sigma) The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.15\n \n\nR normal_rng(reals mu, reals sigma) Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nThe standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. \\[\\begin{equation*}\n\\text{StdNormal}(y) \\ = \\ \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{1}{\\sqrt{2 \\pi}} \\, \\exp \\left( \\frac{-y^2}{2} \\right)\\!.\n\\end{equation*}\\] Up to a proportion on the log scale, where Stan computes, \\[\\begin{equation*}\n\\log \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{-y^2}{2} + \\text{const}.\n\\end{equation*}\\] With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location \\(0\\) and scale \\(1\\).\n\n\n\ny ~ std_normal()\nIncrement target log probability density with std_normal_lupdf(y).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal std_normal_lpdf(reals y) The standard normal (location zero, scale one) log probability density of y.\nAvailable since 2.18\n \n\nreal std_normal_lupdf(reals y) The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.\nAvailable since 2.25\n \n\nreal std_normal_cdf(reals y) The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for \\(y\\) below -37.5 and overflow to 1 for \\(y\\) above 8.25; the function Phi_approx is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lcdf(reals y) The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to \\(-\\infty\\) for \\(y\\) below -37.5 and overflow to 0 for \\(y\\) above 8.25; log(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lccdf(reals y) The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for \\(y\\) below -37.5 and underflow to \\(-\\infty\\) for \\(y\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nR std_normal_qf(T x) Returns the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The std_normal_qf is equivalent to the inv_Phi function.\nAvailable since 2.31\n \n\nR std_normal_log_qf(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) evaluated at the log of the specified quantile x. This function is equivalent to std_normal_qf(exp(x)) but is more numerically stable.\nAvailable since 2.31\n \n\nreal std_normal_rng() Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.21\n\n\n\n\nStan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e. a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\sigma\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^n\\), \\[\\begin{equation*} \\text{NormalIdGLM}(y|x, \\alpha, \\beta, \\sigma) = \\prod_{1\\leq i\n\\leq n}\\text{Normal}(y_i|\\alpha_i + x_i\\cdot \\beta, \\sigma). \\end{equation*}\\]\n\n\n\ny ~ normal_id_glm(x, alpha, beta, sigma)\nIncrement target log probability density with normal_id_glm_lupdf(y | x, alpha, beta, sigma).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n\n\n\n\nExponentially modified Gaussian describes the distribution of \\(Z = X + Y\\) when \\(X\\) and \\(Y\\) are independent and \\(X\\) is normally distributed (with mean \\(\\mu\\) and standard deviation \\(\\sigma\\)) and \\(Y\\) is exponentially distributed (with rate \\(\\lambda\\)).\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\), and \\(\\lambda \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{ExpModNormal}(y|\\mu,\\sigma,\\lambda) = \\frac{\\lambda}{2} \\ \\exp\n\\left(\\frac{\\lambda}{2} \\left(2\\mu + \\lambda \\sigma^2 -\n2y\\right)\\right) \\text{erfc}\\left(\\frac{\\mu + \\lambda\\sigma^2 -\ny}{\\sqrt{2}\\sigma}\\right) . \\end{equation*}\\]\n\n\n\ny ~ exp_mod_normal(mu, sigma, lambda)\nIncrement target log probability density with exp_mod_normal_lupdf(y | mu, sigma, lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exp_mod_normal_lpdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lupdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal exp_mod_normal_cdf(reals y | reals mu, reals sigma, reals lambda) The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.0\n \n\nreal exp_mod_normal_lcdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lccdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nR exp_mod_normal_rng(reals mu, reals sigma, reals lambda) Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\xi \\in \\mathbb{R}\\), \\(\\omega \\in \\mathbb{R}^+\\), and \\(\\alpha \\in\n\\mathbb{R}\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*} \\text{SkewNormal}(y \\mid\n\\xi, \\omega, \\alpha) = \\frac{1}{\\omega\\sqrt{2\\pi}} \\ \\exp\\left( - \\,\n\\frac{1}{2}            \\left(  \\frac{y - \\xi}{\\omega} \\right)^2\n\\right) \\ \\left(1 + \\text{erf}\\left( \\alpha\\left(\\frac{y -\n\\xi}{\\omega\\sqrt{2}}\\right)\\right)\\right) . \\end{equation*}\\]\n\n\n\ny ~ skew_normal(xi, omega, alpha)\nIncrement target log probability density with skew_normal_lupdf(y | xi, omega, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal skew_normal_lpdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lupdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal skew_normal_cdf(reals y | reals xi, reals omega, reals alpha) The skew normal distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lcdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nreal skew_normal_lccdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nR skew_normal_rng(reals xi, reals omega, real alpha) Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}\\), and \\(\\sigma \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{StudentT}(y|\\nu,\\mu,\\sigma) = \\frac{\\Gamma\\left((\\nu +\n1)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\nu \\pi} \\ \\sigma}\n\\ \\left( 1 + \\frac{1}{\\nu} \\left(\\frac{y - \\mu}{\\sigma}\\right)^2\n\\right)^{-(\\nu + 1)/2} \\! . \\end{equation*}\\]\n\n\n\ny ~ student_t(nu, mu, sigma)\nIncrement target log probability density with student_t_lupdf(y | nu, mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal student_t_lpdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lupdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal student_t_cdf(reals y | reals nu, reals mu, reals sigma) The Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.0\n \n\nreal student_t_lcdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lccdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nR student_t_rng(reals nu, reals mu, reals sigma) Generate a Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Cauchy}(y|\\mu,\\sigma) = \\frac{1}{\\pi \\sigma} \\ \\frac{1}{1 + \\left((y - \\mu)/\\sigma\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ cauchy(mu, sigma)\nIncrement target log probability density with cauchy_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal cauchy_lpdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lupdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal cauchy_cdf(reals y | reals mu, reals sigma) The Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal cauchy_lcdf(reals y | reals mu, reals sigma) The log of the Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lccdf(reals y | reals mu, reals sigma) The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR cauchy_rng(reals mu, reals sigma) Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{DoubleExponential}(y|\\mu,\\sigma) =\n\\frac{1}{2\\sigma}   \\exp \\left( - \\, \\frac{|y - \\mu|}{\\sigma} \\right)\n. \\end{equation*}\\] Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section exponential distribution), which is parameterized in terms of inverse scale.\nThe double-exponential distribution can be defined as a compound exponential-normal distribution (Ding and Blitzstein 2018). Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write \\[\\begin{equation*} \\alpha \\sim\n\\mathsf{Exponential}\\left( \\frac{1}{2 \\sigma^2} \\right) \\end{equation*}\\] and \\[\\begin{equation*} \\beta \\mid\n\\alpha \\sim \\mathsf{Normal}(\\mu, \\sqrt{\\alpha}), \\end{equation*}\\] then \\[\\begin{equation*} \\beta \\sim\n\\mathsf{DoubleExponential}(\\mu, \\sigma ). \\end{equation*}\\] This may be used to code a non-centered parameterization by taking \\[\\begin{equation*} \\beta^{\\text{raw}} \\sim\n\\mathsf{Normal}(0, 1) \\end{equation*}\\] and defining \\[\\begin{equation*} \\beta = \\mu + \\sqrt{\\alpha} \\,\n\\beta^{\\text{raw}}. \\end{equation*}\\]\n\n\n\ny ~ double_exponential(mu, sigma)\nIncrement target log probability density with double_exponential_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal double_exponential_lpdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lupdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal double_exponential_cdf(reals y | reals mu, reals sigma) The double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal double_exponential_lcdf(reals y | reals mu, reals sigma) The log of the double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lccdf(reals y | reals mu, reals sigma) The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR double_exponential_rng(reals mu, reals sigma) Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Logistic}(y|\\mu,\\sigma) = \\frac{1}{\\sigma} \\ \\exp\\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right)\n\\ \\left(1 + \\exp \\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right) \\right)^{\\!-2} \\! .\n\\end{equation*}\\]\n\n\n\ny ~ logistic(mu, sigma)\nIncrement target log probability density with logistic_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal logistic_lpdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lupdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal logistic_cdf(reals y | reals mu, reals sigma) The logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal logistic_lcdf(reals y | reals mu, reals sigma) The log of the logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lccdf(reals y | reals mu, reals sigma) The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR logistic_rng(reals mu, reals sigma) Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Gumbel}(y|\\mu,\\beta) = \\frac{1}{\\beta} \\ \\exp\\left(-\\frac{y-\\mu}{\\beta}-\\exp\\left(-\\frac{y-\\mu}{\\beta}\\right)\\right) .\n\\end{equation*}\\]\n\n\n\ny ~ gumbel(mu, beta)\nIncrement target log probability density with gumbel_lupdf(y | mu, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gumbel_lpdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lupdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gumbel_cdf(reals y | reals mu, reals beta) The gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.0\n \n\nreal gumbel_lcdf(reals y | reals mu, reals beta) The log of the gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lccdf(reals y | reals mu, reals beta) The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nR gumbel_rng(reals mu, reals beta) Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\) and \\(\\tau \\in [0, 1]\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{aligned}\n& \\text{SkewDoubleExponential}  (y|\\mu,\\sigma, \\tau) = \\\\\n& \\qquad \\qquad \\frac{2 \\tau (1 - \\tau) }{\\sigma} \\exp \\left[ - \\frac{2}{\\sigma} \\left[ \\left(1 - \\tau \\right) I(y &lt; \\mu)  (\\mu - y) + \\tau I(y &gt; \\mu)(y-\\mu)  \\right]    \\right]\n\\end{aligned}\\]\n\n\n\ny ~ skew_double_exponential(mu, sigma, tau)\nIncrement target log probability density with skew_double_exponential(y | mu, sigma, tau)\nAvailable since 2.28\n \n\n\n\n\n \n\nreal skew_double_exponential_lpdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lupdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms\nAvailable since 2.28\n \n\nreal skew_double_exponential_cdf(reals y | reals mu, reals sigma, reals tau) The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lcdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lccdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nR skew_double_exponential_rng(reals mu, reals sigma, reals tau) Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.28",
+    "text": "machine_precision:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nmap_rect:\n\n\n(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) : vector (higher-order_functions.html)\n\n\nmatrix_exp:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nmatrix_exp_multiply:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nmatrix_power:\n\n\n(matrix A, int B) : matrix (matrix_operations.html)\n\n\nmax:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmdivide_left_spd:\n\n\n(matrix A, matrix B) : vector (matrix_operations.html)\n\n\n(matrix A, vector b) : matrix (matrix_operations.html)\n\n\nmdivide_left_tri_low:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\nmdivide_right_spd:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmdivide_right_tri_low:\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\nmean:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmin:\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nmodified_bessel_first_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmodified_bessel_second_kind:\n\n\n(int v, real z) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nmulti_gp:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lpdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_cholesky_lupdf:\n\n\n(matrix y | matrix L, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lpdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_gp_lupdf:\n\n\n(matrix y | matrix Sigma, vector w) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_cholesky_rng:\n\n\n(row_vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix L) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lpdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_prec_lupdf:\n\n\n(row_vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | row_vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | vectors mu, matrix Omega) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_normal_rng:\n\n\n(row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_cholesky_t_rng:\n\n\n(real nu, vector mu, matrix L) : vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky:\n\n\ndistribution statement (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lpdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_lupdf:\n\n\n(vectors y | real nu, vectors mu, matrix L) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_cholesky_rng:\n\n\n(real nu, array[] row_vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, array[] vector mu, matrix L) : array[] vector (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lpdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_lupdf:\n\n\n(row_vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(row_vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, row_vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\n(vectors y | real nu, vectors mu, matrix Sigma) : real (distributions_over_unbounded_vectors.html)\n\n\nmulti_student_t_rng:\n\n\n(real nu, row_vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, row_vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vector mu, matrix Sigma) : vector (distributions_over_unbounded_vectors.html)\n\n\n(real nu, vectors mu, matrix Sigma) : vectors (distributions_over_unbounded_vectors.html)\n\n\nmultinomial:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit:\n\n\ndistribution statement (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lpmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_lupmf:\n\n\n(array[] int y | vector gamma) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_logit_rng:\n\n\n(vector gamma, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultinomial_lpmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_lupmf:\n\n\n(array[] int y | vector theta) : real (multivariate_discrete_distributions.html)\n\n\nmultinomial_rng:\n\n\n(vector theta, int N) : array[] int (multivariate_discrete_distributions.html)\n\n\nmultiply_lower_tri_self_transpose:\n\n\n(matrix x) : matrix (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#normal-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#normal-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#n",
+    "href": "functions-reference/functions_index.html#n",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Normal}(y|\\mu,\\sigma) =\n\\frac{1}{\\sqrt{2 \\pi} \\ \\sigma} \\exp\\left( - \\, \\frac{1}{2} \\left(\\frac{y - \\mu}{\\sigma} \\right)^2 \\right) \\!.\n\\end{equation*}\\]\n\n\n\ny ~ normal(mu, sigma)\nIncrement target log probability density with normal_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal normal_lpdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal normal_lupdf(reals y | reals mu, reals sigma) The log of the normal density of y given location mu and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_cdf(reals y | reals mu, reals sigma) The cumulative normal distribution of y given location mu and scale sigma; normal_cdf will underflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 1 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; the function Phi_approx is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.0\n \n\nreal normal_lcdf(reals y | reals mu, reals sigma) The log of the cumulative normal distribution of y given location mu and scale sigma; normal_lcdf will underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.12\n \n\nreal normal_lccdf(reals y | reals mu, reals sigma) The log of the complementary cumulative normal distribution of y given location mu and scale sigma; normal_lccdf will overflow to 0 for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) below -37.5 and underflow to \\(-\\infty\\) for \\(\\frac{{y}-{\\mu}}{{\\sigma}}\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails, but must be scaled and translated for anything other than a standard normal.\nAvailable since 2.15\n \n\nR normal_rng(reals mu, reals sigma) Generate a normal variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\nThe standard normal distribution is so-called because its parameters are the units for their respective operations—the location (mean) is zero and the scale (standard deviation) one. The standard normal is parameter-free, and the unit parameters allow considerable simplification of the expression for the density. \\[\\begin{equation*}\n\\text{StdNormal}(y) \\ = \\ \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{1}{\\sqrt{2 \\pi}} \\, \\exp \\left( \\frac{-y^2}{2} \\right)\\!.\n\\end{equation*}\\] Up to a proportion on the log scale, where Stan computes, \\[\\begin{equation*}\n\\log \\text{Normal}(y \\mid 0, 1) \\ = \\ \\frac{-y^2}{2} + \\text{const}.\n\\end{equation*}\\] With no logarithm, no subtraction, and no division by a parameter, the standard normal log density is much more efficient to compute than the normal log density with constant location \\(0\\) and scale \\(1\\).\n\n\n\ny ~ std_normal()\nIncrement target log probability density with std_normal_lupdf(y).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal std_normal_lpdf(reals y) The standard normal (location zero, scale one) log probability density of y.\nAvailable since 2.18\n \n\nreal std_normal_lupdf(reals y) The standard normal (location zero, scale one) log probability density of y dropping constant additive terms.\nAvailable since 2.25\n \n\nreal std_normal_cdf(reals y) The cumulative standard normal distribution of y; std_normal_cdf will underflow to 0 for \\(y\\) below -37.5 and overflow to 1 for \\(y\\) above 8.25; the function Phi_approx is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lcdf(reals y) The log of the cumulative standard normal distribution of y; std_normal_lcdf will underflow to \\(-\\infty\\) for \\(y\\) below -37.5 and overflow to 0 for \\(y\\) above 8.25; log(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nreal std_normal_lccdf(reals y) The log of the complementary cumulative standard normal distribution of y; std_normal_lccdf will overflow to 0 for \\(y\\) below -37.5 and underflow to \\(-\\infty\\) for \\(y\\) above 8.25; log1m(Phi_approx(...)) is more robust in the tails.\nAvailable since 2.21\n \n\nR std_normal_qf(T x) Returns the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) at the specified quantile x. The std_normal_qf is equivalent to the inv_Phi function.\nAvailable since 2.31\n \n\nR std_normal_log_qf(T x) Return the value of the inverse standard normal cdf \\(\\Phi^{-1}\\) evaluated at the log of the specified quantile x. This function is equivalent to std_normal_qf(exp(x)) but is more numerically stable.\nAvailable since 2.31\n \n\nreal std_normal_rng() Generate a normal variate with location zero and scale one; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.21",
+    "text": "neg_binomial:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_cdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lccdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lcdf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta, real phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lpmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_lupmf:\n\n\n(ints n | reals eta, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_log_rng:\n\n\n(reals eta, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lpmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_lupmf:\n\n\n(ints n | reals mu, reals phi) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_2_rng:\n\n\n(reals mu, reals phi) : R (unbounded_discrete_distributions.html)\n\n\nneg_binomial_cdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lccdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lcdf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lpmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_lupmf:\n\n\n(ints n | reals alpha, reals beta) : real (unbounded_discrete_distributions.html)\n\n\nneg_binomial_rng:\n\n\n(reals alpha, reals beta) : R (unbounded_discrete_distributions.html)\n\n\nnegative_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnorm:\n\n\n(complex z) : real (complex-valued_basic_functions.html)\n\n\nnorm1:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnorm2:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(row_vector x) : real (array_operations.html)\n\n\n(vector x) : real (array_operations.html)\n\n\nnormal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_cdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lpdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_id_glm_lupdf:\n\n\n(real y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(real y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, real alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | matrix x, vector alpha, vector beta, vector sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, real alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\n(vector y | row_vector x, vector alpha, vector beta, real sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lccdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lcdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lpdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_lupdf:\n\n\n(reals y | reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nnormal_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nnot_a_number:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nnum_elements:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#normal-id-glm",
-    "href": "functions-reference/unbounded_continuous_distributions.html#normal-id-glm",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#o",
+    "href": "functions-reference/functions_index.html#o",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Stan also supplies a single function for a generalized linear model with normal distribution and identity link function, i.e. a function for a linear regression. This provides a more efficient implementation of linear regression than a manually written regression in terms of a normal distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\sigma\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^n\\), \\[\\begin{equation*} \\text{NormalIdGLM}(y|x, \\alpha, \\beta, \\sigma) = \\prod_{1\\leq i\n\\leq n}\\text{Normal}(y_i|\\alpha_i + x_i\\cdot \\beta, \\sigma). \\end{equation*}\\]\n\n\n\ny ~ normal_id_glm(x, alpha, beta, sigma)\nIncrement target log probability density with normal_id_glm_lupdf(y | x, alpha, beta, sigma).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(real y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.25\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.29\n \n\nreal normal_id_glm_lupdf(vector y | row_vector x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.29\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.23\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, real sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.23\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, real alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30\n \n\nreal normal_id_glm_lpdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma.\nAvailable since 2.30\n \n\nreal normal_id_glm_lupdf(vector y | matrix x, vector alpha, vector beta, vector sigma) The log normal probability density of y given location alpha + x * beta and scale sigma dropping constant additive terms.\nAvailable since 2.30",
+    "text": "ode_adams:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_adams_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_adjoint_tol_ctl:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, data real rel_tol_quadrature, data real abs_tol_qudrature, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward,...) : array[] vector (higher-order_functions.html)\n\n\node_bdf:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_bdf_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_ckrk_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, ...) : array[] vector (higher-order_functions.html)\n\n\node_rk45_tol:\n\n\n(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) : array[] vector (higher-order_functions.html)\n\n\noffset_multiplier_constrain:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_jacobian:\n\n\n(reals y, reals offset, reals mult) : reals (transform_functions.html)\n\n\noffset_multiplier_unconstrain:\n\n\n(reals x, reals offset, reals mult) : reals (transform_functions.html)\n\n\none_hot_array:\n\n\n(int n, int k) : array[] real (matrix_operations.html)\n\n\none_hot_int_array:\n\n\n(int n, int k) : array[] int (matrix_operations.html)\n\n\none_hot_row_vector:\n\n\n(int n, int k) : row_vector (matrix_operations.html)\n\n\none_hot_vector:\n\n\n(int K, int k) : vector (matrix_operations.html)\n\n\nones_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nones_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nones_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nones_vector:\n\n\n(int n) : vector (matrix_operations.html)\n\n\noperator!:\n\n\n(int x) : int (real-valued_basic_functions.html)\n\n\n(real x) : int (real-valued_basic_functions.html)\n\n\noperator!=:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator%/%:\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\noperator&&:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\n&lt;a id=‘operator’’ href=‘#operator’’ class=‘anchored unlink’&gt;operator’:\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : row_vector (matrix_operations.html)\n\n\noperator*:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_matrix y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_vector y) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_row_vector y) : complex_matrix (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(matrix x, vector y) : vector (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, matrix y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, vector y) : real (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, row_vector y) : matrix (matrix_operations.html)\n\n\noperator*=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator+:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator+=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator-:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix x) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(T x) : T (complex-valued_basic_functions.html)\n\n\n(T x) : T (complex_matrix_operations.html)\n\n\n(T x) : T (integer-valued_basic_functions.html)\n\n\n(T x) : T (matrix_operations.html)\n\n\n(T x) : T (real-valued_basic_functions.html)\n\n\n(vector x) : vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator-=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.*:\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator.*=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator./:\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator./=:\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator.^:\n\n\n( complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n( complex_matrix x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_matrix y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, matrix y) : matrix (matrix_operations.html)\n\n\n(real x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(real x, vector y) : vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\noperator/:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(complex_matrix B, complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_matrix x, complex y) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector b, complex_matrix A) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex y) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex y) : complex_vector (complex_matrix_operations.html)\n\n\n(int x, int y) : int (integer-valued_basic_functions.html)\n\n\n(matrix B, matrix A) : matrix (matrix_operations.html)\n\n\n(matrix x, real y) : matrix (matrix_operations.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(row_vector b, matrix A) : row_vector (matrix_operations.html)\n\n\n(row_vector x, real y) : row_vector (matrix_operations.html)\n\n\n(vector x, real y) : vector (matrix_operations.html)\n\n\noperator/=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\n(T x, U y) : void (compound_arithmetic_and_assignment.html)\n\n\noperator&lt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&lt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator=:\n\n\n(complex x, complex y) : void (complex-valued_basic_functions.html)\n\n\noperator==:\n\n\n(complex x, complex y) : int (complex-valued_basic_functions.html)\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator&gt;=:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\noperator\\:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector b) : vector (matrix_operations.html)\n\n\noperator^:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\noperator||:\n\n\n(int x, int y) : int (real-valued_basic_functions.html)\n\n\n(real x, real y) : int (real-valued_basic_functions.html)\n\n\nordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nordered_logistic:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lpmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_glm_lupmf:\n\n\n(array[] int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\n(int y | row_vector x, vector beta, vector c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lpmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_lupmf:\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_logistic_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_probit:\n\n\ndistribution statement (bounded_discrete_distributions.html)\n\n\nordered_probit_lpmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_lupmf:\n\n\n(ints k | real eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\n(ints k | vector eta, vectors c) : real (bounded_discrete_distributions.html)\n\n\nordered_probit_rng:\n\n\n(real eta, vector c) : int (bounded_discrete_distributions.html)\n\n\nordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nowens_t:\n\n\n(real h, real a) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#exponentially-modified-normal-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#exponentially-modified-normal-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#p",
+    "href": "functions-reference/functions_index.html#p",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Exponentially modified Gaussian describes the distribution of \\(Z = X + Y\\) when \\(X\\) and \\(Y\\) are independent and \\(X\\) is normally distributed (with mean \\(\\mu\\) and standard deviation \\(\\sigma\\)) and \\(Y\\) is exponentially distributed (with rate \\(\\lambda\\)).\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\), and \\(\\lambda \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{ExpModNormal}(y|\\mu,\\sigma,\\lambda) = \\frac{\\lambda}{2} \\ \\exp\n\\left(\\frac{\\lambda}{2} \\left(2\\mu + \\lambda \\sigma^2 -\n2y\\right)\\right) \\text{erfc}\\left(\\frac{\\mu + \\lambda\\sigma^2 -\ny}{\\sqrt{2}\\sigma}\\right) . \\end{equation*}\\]\n\n\n\ny ~ exp_mod_normal(mu, sigma, lambda)\nIncrement target log probability density with exp_mod_normal_lupdf(y | mu, sigma, lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal exp_mod_normal_lpdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lupdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal density of y given location mu, scale sigma, and rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal exp_mod_normal_cdf(reals y | reals mu, reals sigma, reals lambda) The exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.0\n \n\nreal exp_mod_normal_lcdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nreal exp_mod_normal_lccdf(reals y | reals mu, reals sigma, reals lambda) The log of the exponentially modified normal complementary cumulative distribution function of y given location mu, scale sigma, and rate lambda\nAvailable since 2.18\n \n\nR exp_mod_normal_rng(reals mu, reals sigma, reals lambda) Generate a exponentially modified normal variate with location mu, scale sigma, and rate lambda; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "pareto:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_cdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lccdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lcdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lpdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_lupdf:\n\n\n(reals y | reals y_min, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_rng:\n\n\n(reals y_min, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\npareto_type_2:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_cdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lccdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lcdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lpdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_lupdf:\n\n\n(reals y | reals mu, reals lambda, reals alpha) : real (positive_lower-bounded_distributions.html)\n\n\npareto_type_2_rng:\n\n\n(reals mu, reals lambda, reals alpha) : R (positive_lower-bounded_distributions.html)\n\n\nPhi:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nPhi_approx:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\npi:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npoisson:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_cdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lccdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lcdf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lpmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_glm_lupmf:\n\n\n(array[] int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(array[] int y | row_vector x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, real alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\n(int y | matrix x, vector alpha, vector beta) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\npoisson_log_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)\n\n\npoisson_lpmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_lupmf:\n\n\n(ints n | reals lambda) : real (unbounded_discrete_distributions.html)\n\n\npoisson_rng:\n\n\n(reals lambda) : R (unbounded_discrete_distributions.html)\n\n\npolar:\n\n\n(real r, real theta) : complex (complex-valued_basic_functions.html)\n\n\npositive_infinity:\n\n\n() : real (real-valued_basic_functions.html)\n\n\npositive_ordered_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\npositive_ordered_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\npow:\n\n\n(complex x, complex y) : complex (complex-valued_basic_functions.html)\n\n\n(real x, real y) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : Z (complex-valued_basic_functions.html)\n\n\nprint:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nprod:\n\n\n(array[] int x) : real (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nproj:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#skew-normal-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#skew-normal-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#q",
+    "href": "functions-reference/functions_index.html#q",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\xi \\in \\mathbb{R}\\), \\(\\omega \\in \\mathbb{R}^+\\), and \\(\\alpha \\in\n\\mathbb{R}\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*} \\text{SkewNormal}(y \\mid\n\\xi, \\omega, \\alpha) = \\frac{1}{\\omega\\sqrt{2\\pi}} \\ \\exp\\left( - \\,\n\\frac{1}{2}            \\left(  \\frac{y - \\xi}{\\omega} \\right)^2\n\\right) \\ \\left(1 + \\text{erf}\\left( \\alpha\\left(\\frac{y -\n\\xi}{\\omega\\sqrt{2}}\\right)\\right)\\right) . \\end{equation*}\\]\n\n\n\ny ~ skew_normal(xi, omega, alpha)\nIncrement target log probability density with skew_normal_lupdf(y | xi, omega, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal skew_normal_lpdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lupdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal density of y given location xi, scale omega, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal skew_normal_cdf(reals y | reals xi, reals omega, reals alpha) The skew normal distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.16\n \n\nreal skew_normal_lcdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nreal skew_normal_lccdf(reals y | reals xi, reals omega, reals alpha) The log of the skew normal complementary cumulative distribution function of y given location xi, scale omega, and shape alpha\nAvailable since 2.18\n \n\nR skew_normal_rng(reals xi, reals omega, real alpha) Generate a skew normal variate with location xi, scale omega, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "qr:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin:\n\n\n(matrix A) : tuple(matrix, matrix) (matrix_operations.html)\n\n\nqr_thin_Q:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nqr_thin_R:\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nquad_form:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquad_form_diag:\n\n\n(matrix m, row_vector rv) : matrix (matrix_operations.html)\n\n\n(matrix m, vector v) : matrix (matrix_operations.html)\n\n\nquad_form_sym:\n\n\n(matrix A, matrix B) : matrix (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\nquantile:\n\n\n(data array[] real x, data array[] real p) : array[] real (array_operations.html)\n\n\n(data array[] real x, data real p) : real (array_operations.html)\n\n\n(data row_vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data row_vector x, data real p) : real (matrix_operations.html)\n\n\n(data vector x, data array[] real p) : array[] real (matrix_operations.html)\n\n\n(data vector x, data real p) : real (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#student-t-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#student-t-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#r",
+    "href": "functions-reference/functions_index.html#r",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\nu \\in \\mathbb{R}^+\\), \\(\\mu \\in \\mathbb{R}\\), and \\(\\sigma \\in\n\\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{StudentT}(y|\\nu,\\mu,\\sigma) = \\frac{\\Gamma\\left((\\nu +\n1)/2\\right)}      {\\Gamma(\\nu/2)} \\ \\frac{1}{\\sqrt{\\nu \\pi} \\ \\sigma}\n\\ \\left( 1 + \\frac{1}{\\nu} \\left(\\frac{y - \\mu}{\\sigma}\\right)^2\n\\right)^{-(\\nu + 1)/2} \\! . \\end{equation*}\\]\n\n\n\ny ~ student_t(nu, mu, sigma)\nIncrement target log probability density with student_t_lupdf(y | nu, mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal student_t_lpdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lupdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) density of y given degrees of freedom nu, location mu, and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal student_t_cdf(reals y | reals nu, reals mu, reals sigma) The Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.0\n \n\nreal student_t_lcdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nreal student_t_lccdf(reals y | reals nu, reals mu, reals sigma) The log of the Student-\\(t\\) complementary cumulative distribution function of y given degrees of freedom nu, location mu, and scale sigma\nAvailable since 2.12\n \n\nR student_t_rng(reals nu, reals mu, reals sigma) Generate a Student-\\(t\\) variate with degrees of freedom nu, location mu, and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "rank:\n\n\n(array[] int v, int s) : int (array_operations.html)\n\n\n(array[] real v, int s) : int (array_operations.html)\n\n\n(row_vector v, int s) : int (matrix_operations.html)\n\n\n(vector v, int s) : int (matrix_operations.html)\n\n\nrayleigh:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nrayleigh_cdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lccdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lcdf:\n\n\n(real y | real sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lpdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_lupdf:\n\n\n(reals y | reals sigma) : real (positive_continuous_distributions.html)\n\n\nrayleigh_rng:\n\n\n(reals sigma) : R (positive_continuous_distributions.html)\n\n\nreduce_sum:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreduce_sum_static:\n\n\n(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...) : real (higher-order_functions.html)\n\n\nreject:\n\n\n(T1 x1,..., TN xN) : void (void_functions.html)\n\n\nrep_array:\n\n\n(T x, int k, int m, int n) : array[,,] T (array_operations.html)\n\n\n(T x, int m, int n) : array[,] T (array_operations.html)\n\n\n(T x, int n) : array[] T (array_operations.html)\n\n\nrep_matrix:\n\n\n(complex z, int m, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_row_vector rv, int m) : complex_matrix (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_matrix (complex_matrix_operations.html)\n\n\n(real x, int m, int n) : matrix (matrix_operations.html)\n\n\n(row_vector rv, int m) : matrix (matrix_operations.html)\n\n\n(vector v, int n) : matrix (matrix_operations.html)\n\n\nrep_row_vector:\n\n\n(complex z, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(real x, int n) : row_vector (matrix_operations.html)\n\n\nrep_vector:\n\n\n(complex z, int m) : complex_vector (complex_matrix_operations.html)\n\n\n(real x, int m) : vector (matrix_operations.html)\n\n\nreverse:\n\n\n(array[] T v) : array[] T (array_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nrising_factorial:\n\n\n(real x, int n) : real (real-valued_basic_functions.html)\n\n\n(T1 x, T2 y) : R (real-valued_basic_functions.html)\n\n\nround:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nrow:\n\n\n(complex_matrix x, int m) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int m) : row_vector (matrix_operations.html)\n\n\nrows:\n\n\n(complex_matrix x) : int (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nrows_dot_product:\n\n\n(complex_matrix x, complex_matrix y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x, complex_row_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x, complex_vector y) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, matrix y) : vector (matrix_operations.html)\n\n\n(row_vector x, row_vector y) : vector (matrix_operations.html)\n\n\n(vector x, vector y) : vector (matrix_operations.html)\n\n\nrows_dot_self:\n\n\n(complex_matrix x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x) : vector (matrix_operations.html)\n\n\n(row_vector x) : vector (matrix_operations.html)\n\n\n(vector x) : vector (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#cauchy-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#cauchy-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#s",
+    "href": "functions-reference/functions_index.html#s",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Cauchy}(y|\\mu,\\sigma) = \\frac{1}{\\pi \\sigma} \\ \\frac{1}{1 + \\left((y - \\mu)/\\sigma\\right)^2} .\n\\end{equation*}\\]\n\n\n\ny ~ cauchy(mu, sigma)\nIncrement target log probability density with cauchy_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal cauchy_lpdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lupdf(reals y | reals mu, reals sigma) The log of the Cauchy density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal cauchy_cdf(reals y | reals mu, reals sigma) The Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal cauchy_lcdf(reals y | reals mu, reals sigma) The log of the Cauchy cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal cauchy_lccdf(reals y | reals mu, reals sigma) The log of the Cauchy complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR cauchy_rng(reals mu, reals sigma) Generate a Cauchy variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "scale_matrix_exp_multiply:\n\n\n(real t, matrix A, matrix B) : matrix (matrix_operations.html)\n\n\nscaled_inv_chi_square:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_cdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lccdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lcdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lpdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_lupdf:\n\n\n(reals y | reals nu, reals sigma) : real (positive_continuous_distributions.html)\n\n\nscaled_inv_chi_square_rng:\n\n\n(reals nu, reals sigma) : R (positive_continuous_distributions.html)\n\n\nsd:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsegment:\n\n\n(array[] T sv, int i, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int i, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int i, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int i, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int i, int n) : vector (matrix_operations.html)\n\n\nsimplex_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsimplex_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsin:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsingular_values:\n\n\n(complex_matrix A) : vector (complex_matrix_operations.html)\n\n\n(matrix A) : vector (matrix_operations.html)\n\n\nsinh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsize:\n\n\n(array[] T x) : int (array_operations.html)\n\n\n(complex_row_vector x) : int (complex_matrix_operations.html)\n\n\n(complex_vector x) : int (complex_matrix_operations.html)\n\n\n(int x) : int (integer-valued_basic_functions.html)\n\n\n(matrix x) : int (complex_matrix_operations.html)\n\n\n(matrix x) : int (matrix_operations.html)\n\n\n(real x) : int (integer-valued_basic_functions.html)\n\n\n(row_vector x) : int (matrix_operations.html)\n\n\n(vector x) : int (matrix_operations.html)\n\n\nskew_double_exponential:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_cdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lccdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lcdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lpdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_lupdf:\n\n\n(reals y | reals mu, reals sigma, reals tau) : real (unbounded_continuous_distributions.html)\n\n\nskew_double_exponential_rng:\n\n\n(reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nskew_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nskew_normal_cdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lccdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lcdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lpdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_lupdf:\n\n\n(reals y | reals xi, reals omega, reals alpha) : real (unbounded_continuous_distributions.html)\n\n\nskew_normal_rng:\n\n\n(reals xi, reals omega, real alpha) : R (unbounded_continuous_distributions.html)\n\n\nsoftmax:\n\n\n(vector x) : vector (matrix_operations.html)\n\n\nsolve_newton:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_newton_tol:\n\n\n(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell:\n\n\n(function algebra_system, vector y_guess, ...) : vector (higher-order_functions.html)\n\n\nsolve_powell_tol:\n\n\n(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) : vector (higher-order_functions.html)\n\n\nsort_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] real (array_operations.html)\n\n\n(row_vector v) : row_vector (matrix_operations.html)\n\n\n(vector v) : vector (matrix_operations.html)\n\n\nsort_indices_asc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsort_indices_desc:\n\n\n(array[] int v) : array[] int (array_operations.html)\n\n\n(array[] real v) : array[] int (array_operations.html)\n\n\n(row_vector v) : array[] int (matrix_operations.html)\n\n\n(vector v) : array[] int (matrix_operations.html)\n\n\nsqrt:\n\n\n(complex x) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsqrt2:\n\n\n() : real (real-valued_basic_functions.html)\n\n\nsquare:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nsquared_distance:\n\n\n(row_vector x, row_vector y) : real (array_operations.html)\n\n\n(row_vector x, vector y) : real (array_operations.html)\n\n\n(vector x, row_vector y) : real (array_operations.html)\n\n\n(vector x, vector y) : real (array_operations.html)\n\n\nstd_normal:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstd_normal_cdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lccdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lcdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_log_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_lpdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_lupdf:\n\n\n(reals y) : real (unbounded_continuous_distributions.html)\n\n\nstd_normal_qf:\n\n\n(T x) : R (unbounded_continuous_distributions.html)\n\n\nstd_normal_rng:\n\n\n() : real (unbounded_continuous_distributions.html)\n\n\nstep:\n\n\n(real x) : real (real-valued_basic_functions.html)\n\n\nstochastic_column_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_column_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstochastic_row_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\nstochastic_row_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\nstudent_t:\n\n\ndistribution statement (unbounded_continuous_distributions.html)\n\n\nstudent_t_cdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lccdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lcdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lpdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_lupdf:\n\n\n(reals y | reals nu, reals mu, reals sigma) : real (unbounded_continuous_distributions.html)\n\n\nstudent_t_rng:\n\n\n(reals nu, reals mu, reals sigma) : R (unbounded_continuous_distributions.html)\n\n\nsub_col:\n\n\n(complex_matrix x, int i, int j, int n_rows) : complex_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_rows) : vector (matrix_operations.html)\n\n\nsub_row:\n\n\n(complex_matrix x, int i, int j, int n_cols) : complex_row_vector (complex_matrix_operations.html)\n\n\n(matrix x, int i, int j, int n_cols) : row_vector (matrix_operations.html)\n\n\nsum:\n\n\n(array[] complex x) : complex (array_operations.html)\n\n\n(array[] int x) : int (array_operations.html)\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(complex_matrix x) : complex (complex_matrix_operations.html)\n\n\n(complex_row_vector x) : complex (complex_matrix_operations.html)\n\n\n(complex_vector x) : complex (complex_matrix_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nsum_to_zero_constrain:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_jacobian:\n\n\n(matrices y) : matrices (transform_functions.html)\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nsum_to_zero_unconstrain:\n\n\n(matrices x) : matrices (transform_functions.html)\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nsvd:\n\n\n(complex_matrix A) : tuple(complex_matrix, vector, complex_matrix) (complex_matrix_operations.html)\n\n\n(matrix A) : tuple(matrix, vector, matrix) (matrix_operations.html)\n\n\nsvd_U:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsvd_V:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)\n\n\nsymmetrize_from_lower_tri:\n\n\n(complex_matrix A) : complex_matrix (complex_matrix_operations.html)\n\n\n(matrix A) : matrix (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#double-exponential-laplace-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#double-exponential-laplace-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#t",
+    "href": "functions-reference/functions_index.html#t",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in\n\\mathbb{R}\\), \\[\\begin{equation*} \\text{DoubleExponential}(y|\\mu,\\sigma) =\n\\frac{1}{2\\sigma}   \\exp \\left( - \\, \\frac{|y - \\mu|}{\\sigma} \\right)\n. \\end{equation*}\\] Note that the double exponential distribution is parameterized in terms of the scale, in contrast to the exponential distribution (see section exponential distribution), which is parameterized in terms of inverse scale.\nThe double-exponential distribution can be defined as a compound exponential-normal distribution (Ding and Blitzstein 2018). Using the inverse scale parameterization for the exponential distribution, and the standard deviation parameterization for the normal distribution, one can write \\[\\begin{equation*} \\alpha \\sim\n\\mathsf{Exponential}\\left( \\frac{1}{2 \\sigma^2} \\right) \\end{equation*}\\] and \\[\\begin{equation*} \\beta \\mid\n\\alpha \\sim \\mathsf{Normal}(\\mu, \\sqrt{\\alpha}), \\end{equation*}\\] then \\[\\begin{equation*} \\beta \\sim\n\\mathsf{DoubleExponential}(\\mu, \\sigma ). \\end{equation*}\\] This may be used to code a non-centered parameterization by taking \\[\\begin{equation*} \\beta^{\\text{raw}} \\sim\n\\mathsf{Normal}(0, 1) \\end{equation*}\\] and defining \\[\\begin{equation*} \\beta = \\mu + \\sqrt{\\alpha} \\,\n\\beta^{\\text{raw}}. \\end{equation*}\\]\n\n\n\ny ~ double_exponential(mu, sigma)\nIncrement target log probability density with double_exponential_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal double_exponential_lpdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lupdf(reals y | reals mu, reals sigma) The log of the double exponential density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal double_exponential_cdf(reals y | reals mu, reals sigma) The double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal double_exponential_lcdf(reals y | reals mu, reals sigma) The log of the double exponential cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal double_exponential_lccdf(reals y | reals mu, reals sigma) The log of the double exponential complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR double_exponential_rng(reals mu, reals sigma) Generate a double exponential variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "tail:\n\n\n(array[] T sv, int n) : array[] T (matrix_operations.html)\n\n\n(complex_row_vector rv, int n) : complex_row_vector (complex_matrix_operations.html)\n\n\n(complex_vector v, int n) : complex_vector (complex_matrix_operations.html)\n\n\n(row_vector rv, int n) : row_vector (matrix_operations.html)\n\n\n(vector v, int n) : vector (matrix_operations.html)\n\n\ntan:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntanh:\n\n\n(complex z) : complex (complex-valued_basic_functions.html)\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntarget:\n\n\n() : real (real-valued_basic_functions.html)\n\n\ntcrossprod:\n\n\n(matrix x) : matrix (matrix_operations.html)\n\n\ntgamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\nto_array_1d:\n\n\n(array[...] complex a) : array[] complex (mixed_operations.html)\n\n\n(array[...] int a) : array[] int (mixed_operations.html)\n\n\n(array[...] real a) : array[] real (mixed_operations.html)\n\n\n(complex_matrix m) : array[] complex (mixed_operations.html)\n\n\n(complex_row_vector v) : array[] complex (mixed_operations.html)\n\n\n(complex_vector v) : array[] real (mixed_operations.html)\n\n\n(matrix m) : array[] real (mixed_operations.html)\n\n\n(row_vector v) : array[] real (mixed_operations.html)\n\n\n(vector v) : array[] real (mixed_operations.html)\n\n\nto_array_2d:\n\n\n(complex_matrix m) : array[,] real (mixed_operations.html)\n\n\n(matrix m) : array[,] real (mixed_operations.html)\n\n\nto_complex:\n\n\n() : complex (complex-valued_basic_functions.html)\n\n\n(real re) : complex (complex-valued_basic_functions.html)\n\n\n(real re, real im) : complex (complex-valued_basic_functions.html)\n\n\n(T1 re, T2 im) : Z (complex-valued_basic_functions.html)\n\n\nto_int:\n\n\n(data real x) : int (integer-valued_basic_functions.html)\n\n\nto_matrix:\n\n\n(array[,] complex a ) : complex_matrix (mixed_operations.html)\n\n\n(array[,] int a) : matrix (mixed_operations.html)\n\n\n(array[,] real a) : matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex a, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(array[] complex_row_vector vs) : complex_matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] int a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n) : matrix (mixed_operations.html)\n\n\n(array[] real a, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(array[] row_vector vs) : matrix (mixed_operations.html)\n\n\n(complex_matrix A, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix m) : complex_matrix (mixed_operations.html)\n\n\n(complex_matrix M, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_row_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n) : complex_matrix (mixed_operations.html)\n\n\n(complex_vector v, int m, int n, int col_major) : complex_matrix (mixed_operations.html)\n\n\n(matrix A, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(matrix m) : matrix (mixed_operations.html)\n\n\n(matrix M, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(row_vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\n(vector v) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n) : matrix (mixed_operations.html)\n\n\n(vector v, int m, int n, int col_major) : matrix (mixed_operations.html)\n\n\nto_row_vector:\n\n\n(array[] complex a) : complex_row_vector (mixed_operations.html)\n\n\n(array[] int a) : row_vector (mixed_operations.html)\n\n\n(array[] real a) : row_vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_row_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_row_vector (mixed_operations.html)\n\n\n(matrix m) : row_vector (mixed_operations.html)\n\n\n(row_vector v) : row_vector (mixed_operations.html)\n\n\n(vector v) : row_vector (mixed_operations.html)\n\n\nto_vector:\n\n\n(array[] complex a) : complex_vector (mixed_operations.html)\n\n\n(array[] int a) : vector (mixed_operations.html)\n\n\n(array[] real a) : vector (mixed_operations.html)\n\n\n(complex_matrix m) : complex_vector (mixed_operations.html)\n\n\n(complex_row_vector v) : complex_vector (mixed_operations.html)\n\n\n(complex_vector v) : complex_vector (mixed_operations.html)\n\n\n(matrix m) : vector (mixed_operations.html)\n\n\n(row_vector v) : vector (mixed_operations.html)\n\n\n(vector v) : vector (mixed_operations.html)\n\n\ntrace:\n\n\n(complex_matrix A) : complex (complex_matrix_operations.html)\n\n\n(matrix A) : real (matrix_operations.html)\n\n\ntrace_dot:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_gen_quad_form:\n\n\n(matrix D ,matrix A, matrix B) : real (matrix_operations.html)\n\n\ntrace_quad_form:\n\n\n(matrix A, matrix B) : real (matrix_operations.html)\n\n\n(matrix A, vector B) : real (matrix_operations.html)\n\n\ntrigamma:\n\n\n(T x) : R (real-valued_basic_functions.html)\n\n\ntrunc:\n\n\n(T x) : R (real-valued_basic_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#logistic-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#logistic-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#u",
+    "href": "functions-reference/functions_index.html#u",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\sigma \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Logistic}(y|\\mu,\\sigma) = \\frac{1}{\\sigma} \\ \\exp\\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right)\n\\ \\left(1 + \\exp \\!\\left( - \\, \\frac{y - \\mu}{\\sigma} \\right) \\right)^{\\!-2} \\! .\n\\end{equation*}\\]\n\n\n\ny ~ logistic(mu, sigma)\nIncrement target log probability density with logistic_lupdf(y | mu, sigma).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal logistic_lpdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lupdf(reals y | reals mu, reals sigma) The log of the logistic density of y given location mu and scale sigma dropping constant additive terms\nAvailable since 2.25\n \n\nreal logistic_cdf(reals y | reals mu, reals sigma) The logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.0\n \n\nreal logistic_lcdf(reals y | reals mu, reals sigma) The log of the logistic cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nreal logistic_lccdf(reals y | reals mu, reals sigma) The log of the logistic complementary cumulative distribution function of y given location mu and scale sigma\nAvailable since 2.12\n \n\nR logistic_rng(reals mu, reals sigma) Generate a logistic variate with location mu and scale sigma; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "uniform:\n\n\ndistribution statement (bounded_continuous_distributions.html)\n\n\nuniform_cdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lccdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lcdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lpdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_lupdf:\n\n\n(reals y | reals alpha, reals beta) : real (bounded_continuous_distributions.html)\n\n\nuniform_rng:\n\n\n(reals alpha, reals beta) : R (bounded_continuous_distributions.html)\n\n\nuniform_simplex:\n\n\n(int n) : vector (matrix_operations.html)\n\n\nunit_vectors_constrain:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_jacobian:\n\n\n(vectors y) : vectors (transform_functions.html)\n\n\nunit_vectors_unconstrain:\n\n\n(vectors x) : vectors (transform_functions.html)\n\n\nupper_bound_constrain:\n\n\n(reals y, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_jacobian:\n\n\n(reals x, reals ub) : reals (transform_functions.html)\n\n\nupper_bound_unconstrain:\n\n\n(reals x, reals ub) : reals (transform_functions.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#gumbel-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#gumbel-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#v",
+    "href": "functions-reference/functions_index.html#v",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{equation*}\n\\text{Gumbel}(y|\\mu,\\beta) = \\frac{1}{\\beta} \\ \\exp\\left(-\\frac{y-\\mu}{\\beta}-\\exp\\left(-\\frac{y-\\mu}{\\beta}\\right)\\right) .\n\\end{equation*}\\]\n\n\n\ny ~ gumbel(mu, beta)\nIncrement target log probability density with gumbel_lupdf(y | mu, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal gumbel_lpdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lupdf(reals y | reals mu, reals beta) The log of the gumbel density of y given location mu and scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal gumbel_cdf(reals y | reals mu, reals beta) The gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.0\n \n\nreal gumbel_lcdf(reals y | reals mu, reals beta) The log of the gumbel cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nreal gumbel_lccdf(reals y | reals mu, reals beta) The log of the gumbel complementary cumulative distribution function of y given location mu and scale beta\nAvailable since 2.12\n \n\nR gumbel_rng(reals mu, reals beta) Generate a gumbel variate with location mu and scale beta; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "text": "variance:\n\n\n(array[] real x) : real (array_operations.html)\n\n\n(matrix x) : real (matrix_operations.html)\n\n\n(row_vector x) : real (matrix_operations.html)\n\n\n(vector x) : real (matrix_operations.html)\n\n\nvon_mises:\n\n\ndistribution statement (circular_distributions.html)\n\n\nvon_mises_cdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lccdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lcdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lpdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_lupdf:\n\n\n(reals y | reals mu, reals kappa) : real (circular_distributions.html)\n\n\nvon_mises_rng:\n\n\n(reals mu, reals kappa) : R (circular_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/unbounded_continuous_distributions.html#skew-double-exponential-distribution",
-    "href": "functions-reference/unbounded_continuous_distributions.html#skew-double-exponential-distribution",
-    "title": "Unbounded Continuous Distributions",
+    "objectID": "functions-reference/functions_index.html#w",
+    "href": "functions-reference/functions_index.html#w",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "If \\(\\mu \\in \\mathbb{R}\\), \\(\\sigma \\in \\mathbb{R}^+\\) and \\(\\tau \\in [0, 1]\\), then for \\(y \\in \\mathbb{R}\\), \\[\\begin{aligned}\n& \\text{SkewDoubleExponential}  (y|\\mu,\\sigma, \\tau) = \\\\\n& \\qquad \\qquad \\frac{2 \\tau (1 - \\tau) }{\\sigma} \\exp \\left[ - \\frac{2}{\\sigma} \\left[ \\left(1 - \\tau \\right) I(y &lt; \\mu)  (\\mu - y) + \\tau I(y &gt; \\mu)(y-\\mu)  \\right]    \\right]\n\\end{aligned}\\]\n\n\n\ny ~ skew_double_exponential(mu, sigma, tau)\nIncrement target log probability density with skew_double_exponential(y | mu, sigma, tau)\nAvailable since 2.28\n \n\n\n\n\n \n\nreal skew_double_exponential_lpdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lupdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential density of y given location mu, scale sigma and skewness tau dropping constant additive terms\nAvailable since 2.28\n \n\nreal skew_double_exponential_cdf(reals y | reals mu, reals sigma, reals tau) The skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lcdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nreal skew_double_exponential_lccdf(reals y | reals mu, reals sigma, reals tau) The log of the skew double exponential complementary cumulative distribution function of y given location mu, scale sigma and skewness tau\nAvailable since 2.28\n \n\nR skew_double_exponential_rng(reals mu, reals sigma, reals tau) Generate a skew double exponential variate with location mu, scale sigma and skewness tau; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.28",
+    "text": "weibull:\n\n\ndistribution statement (positive_continuous_distributions.html)\n\n\nweibull_cdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lccdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lcdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lpdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_lupdf:\n\n\n(reals y | reals alpha, reals sigma) : real (positive_continuous_distributions.html)\n\n\nweibull_rng:\n\n\n(reals alpha, reals sigma) : R (positive_continuous_distributions.html)\n\n\nwiener:\n\n\ndistribution statement (positive_lower-bounded_distributions.html)\n\n\nwiener_lccdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lcdf_unnorm:\n\n\n(real y, real alpha, real tau, real beta, real delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lpdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwiener_lupdf:\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta) : real (positive_lower-bounded_distributions.html)\n\n\n(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) : real (positive_lower-bounded_distributions.html)\n\n\n(reals y | reals alpha, reals tau, reals beta, reals delta) : real (positive_lower-bounded_distributions.html)\n\n\nwishart:\n\n\ndistribution statement (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lpdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_lupdf:\n\n\n(matrix L_W | real nu, matrix L_S) : real (covariance_matrix_distributions.html)\n\n\nwishart_cholesky_rng:\n\n\n(real nu, matrix L_S) : matrix (covariance_matrix_distributions.html)\n\n\nwishart_lpdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_lupdf:\n\n\n(matrix W | real nu, matrix Sigma) : real (covariance_matrix_distributions.html)\n\n\nwishart_rng:\n\n\n(real nu, matrix Sigma) : matrix (covariance_matrix_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Continuous Distributions",
-      "Unbounded Continuous Distributions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/void_functions.html",
-    "href": "functions-reference/void_functions.html",
-    "title": "Void Functions",
+    "objectID": "functions-reference/functions_index.html#y",
+    "href": "functions-reference/functions_index.html#y",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Stan supports a few special statements for printing or for signaling an issue with the program.\nAlthough print, reject, and fatal_error appear to have the syntax of functions, they are actually special kinds of statements with slightly different form and behavior than other functions. First, they are the constructs that allow a variable number of arguments. Second, they are the the only constructs to accept string literals (e.g., \"hello world\") as arguments. Third, they have no effect on the log density function and operate solely through side effects.\nThe special keyword void is used for their return type because they behave like variadic functions with void return type, even though they are special kinds of statements built in to the language.\n\n\nPrinting has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).\n \n\nvoid print(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.1\n\n\n\nThe reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).\n \n\nvoid reject(T1 x1,..., TN xN) Reject the current iteration and print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.18\n\n\n\nThe fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.\n \n\nvoid fatal_error(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.35",
+    "text": "yule_simon:\n\n\ndistribution statement (unbounded_discrete_distributions.html)\n\n\nyule_simon_cdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lccdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lcdf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lpmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_lupmf:\n\n\n(ints n | reals alpha) : real (unbounded_discrete_distributions.html)\n\n\nyule_simon_rng:\n\n\n(reals alpha) : R (unbounded_discrete_distributions.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Void Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/void_functions.html#print-statement",
-    "href": "functions-reference/void_functions.html#print-statement",
-    "title": "Void Functions",
+    "objectID": "functions-reference/functions_index.html#z",
+    "href": "functions-reference/functions_index.html#z",
+    "title": "Alphabetical Index",
     "section": "",
-    "text": "Printing has no effect on the model’s log probability function. Its sole purpose is the side effect (i.e., an effect not represented in a return value) of arguments being printed to whatever the standard output stream is connected to (e.g., the terminal in command-line Stan or the R console in RStan).\n \n\nvoid print(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.1",
+    "text": "zeros_array:\n\n\n(int n) : array[] real (matrix_operations.html)\n\n\nzeros_int_array:\n\n\n(int n) : array[] int (matrix_operations.html)\n\n\nzeros_row_vector:\n\n\n(int n) : row_vector (matrix_operations.html)\n\n\nzeros_vector:\n\n\n(int n) : vector (matrix_operations.html)",
     "crumbs": [
       "Functions Reference",
-      "Built-in Functions",
-      "Void Functions"
+      "Appendix",
+      "Alphabetical Index"
     ]
   },
   {
-    "objectID": "functions-reference/void_functions.html#reject-statement",
-    "href": "functions-reference/void_functions.html#reject-statement",
-    "title": "Void Functions",
+    "objectID": "functions-reference/higher-order_functions.html",
+    "href": "functions-reference/higher-order_functions.html",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "The reject statement has the same syntax as the print statement, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a reject statement is to throw an exception internally that terminates the current iteration with a rejection (the behavior of which will depend on the algorithmic context in which it occurs).\n \n\nvoid reject(T1 x1,..., TN xN) Reject the current iteration and print the values denoted by the arguments x1 through xN on the output message stream. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.18",
+    "text": "Stan provides a few higher-order functions that act on other functions. In all cases, the function arguments to the higher-order functions are defined as functions within the Stan language and passed by name to the higher-order functions.\n\n\nStan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.\nAn algebraic solver is a higher-order function, i.e. it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section Ordinary Differential Equation (ODE) Solvers and Differential Algebraic Equation (DAE) solver). Ordinary Stan functions do not allow functions as arguments.\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The function must return a vector and takes in, as its first argument, the unknowns \\(y\\) we wish to solve for, also passed as a vector. This argument is followed by additional arguments as specified by the user; we call such arguments variadic arguments and denote them .... The signature of the algebraic system is then:\n vector algebra_system (vector y, ...)\nThere is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword data. In the below example, the last variadic argument, \\(x\\), is restricted to being data:\n  vector algebra_system (vector y, vector theta, data vector x)\nDistinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.\n\n\n\n \n vector solve_newton(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.31\n \n vector solve_newton_tol(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.31\n \n vector solve_powell(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method.\nAvailable since 2.31\n \n vector solve_powell_tol(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.\nAvailable since 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, ...):vector. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\n...: variadic arguments.\n\nThe algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:\n\nscaling_step: for the Newton solver only, the scaled-step stopping tolerance, type real, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is \\(10^{-3}\\).\nrel_tol: for the Powell solver only, the relative tolerance, type real, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is \\(10^{-10}\\).\nfunction_tol: function tolerance for the algebraic solver, type real, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is \\(10^{-6}\\).\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is \\(200\\).\n\nThe difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0 (approximately, within the specified function tolerance).\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\n\n\n\nStan offers two methods to solve algebraic equations. solve_newton and solve_newton_tol use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite (Hindmarsh et al. 2005). For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.\nsolve_powell and solve_powell_tol are based on the Powell hybrid method (Powell 1970), which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library (Guennebaud, Jacob, et al. 2010). This solver is in turn based on the algorithm developed for the package MINPACK-1 (Jorge J. More 1980).\nFor both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see (Gaebler 2021) and (Margossian and Betancourt 2022).\n\n\n\n\n\nStan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).\nSolving an initial value ODE means given a set of differential equations \\(y'(t, \\theta) = f(t, y, \\theta)\\) and initial conditions \\(y(t_0, \\theta)\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 &lt; t_2, \\cdots &lt; t_n\\). \\(f(t, y, \\theta)\\) is referred to here as the ODE system function.\n\\(f(t, y, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.\nTo make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as .... The types of the arguments represented by ... in the ODE solve function call must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector ode_rk45(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.24\n \n\narray[] vector ode_rk45_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.24\n \n\narray[] vector ode_ckrk(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.\nAvailable since 2.27\n \n\narray[] vector ode_ckrk_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.27\n \n\narray[] vector ode_adams(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.24\n \n\narray[] vector ode_adams_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_bdf(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.24\n \n\narray[] vector ode_bdf_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_adjoint_tol_ctl(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)\nSolves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.\nAvailable since 2.27\n\n\n\nThe first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a vector return type, and the first two arguments must be a real and vector in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:\n vector ode(real time, vector state, ...)\nThe ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\n..., sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the ... arguments of the ODE solve function call.\n\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see Arguments to the adjoint ODE solver.\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nFor the versions of the ode solver functions ending in _tol, these three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the ODE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\n\nThe arguments to the adjoint ODE solver are different from those for the other functions (for those see Arguments to the ODE solvers).\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\ndata rel_tol_forward: Relative tolerance for forward solve, type real, data only,\ndata abs_tol_forward: Absolute tolerance vector for each state for forward solve, type vector, data only,\ndata rel_tol_backward: Relative tolerance for backward solve, type real, data only,\ndata abs_tol_backward: Absolute tolerance vector for each state for backward solve, type vector, data only,\ndata rel_tol_quadrature: Relative tolerance for backward quadrature, type real, data only,\ndata abs_tol_quadrature: Absolute tolerance for backward quadrature, type real, data only,\ndata max_num_steps: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type int, data only,\nnum_steps_between_checkpoints: number of steps between checkpointing forward solution, type int, data only,\ninterpolation_polynomial: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type int, data only,\nsolver_forward: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only,\nsolver_backward: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only.\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe return value for the ODE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time in specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.\n\n\n\n\n\n\nStan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 (Serban et al. 2021).\nSolving an initial value DAE means given a set of residual functions \\(r(y'(t, \\theta), y(t, \\theta), t)\\) and initial conditions \\((y(t_0, \\theta), y'(t_0, \\theta))\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 \\leq t_2, \\cdots \\leq t_n\\). The residual function \\(r(y', y, t, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.\nSimilar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as ..., and the types of these arguments, also represented by ... in the DAE solver call, must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector dae(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) Solves the DAE system using the backward differentiation formula (BDF) method (Serban et al. 2021).\nAvailable since 2.29\n \n\narray[] vector dae_tol(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.29\n\n\n\nThe first argument to the DAE solver is the DAE residual function. The DAE residual function must have a vector return type, and the first three arguments must be a real, vector, and vector, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:\n  vector residual(real time, vector state, vector state_derivative, ...)\nThe DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the DAE system\nstate, the state of the DAE system at the time specified\nstate_derivative, the time derivatives of the state of the DAE system at the time specified\n..., sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the ... arguments of the DAE solve function call.\n\n\n\n\nThe arguments to the DAE solver are\n\nresidual: DAE residual function,\ninitial_state: initial state, type vector,\ninitial_state_derivative: time derivative of the initial state, type vector,\ninitial_time: initial time, type data real,\ntimes: solution times, type data array[] real,\n...: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the ... arguments of the DAE residual function.\n\nFor dae_tol, the following three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the DAE solver, type real, data only,\ndata abs_tol: absolute tolerance for the DAE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the DAE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe user is responsible to ensure the residual function becomes zero at the initial time, t0, when the arguments initial_state and initial_state_derivative are introduced as state and state_derivative, respectively.\n\n\n\nThe return value for the DAE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.\n\n\n\n\n\n\nStan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.\nIt operates similarly to the algebraic solver and the ordinary differential equations solver in that it allows as an argument a function.\nLike both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.\n\n\nPerforming a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:\nreal integrand(real x, real xc, array[] real theta,\n               array[] real x_r, array[] int x_i)\nThe function should return the value of the integrand evaluated at the point x.\nThe argument of this function are:\n\nx, the independent variable being integrated over\nxc, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section Precision Loss).\ntheta, parameter values used to evaluate the integral\nx_r, data values used to evaluate the integral\nx_i, integer data used to evaluate the integral\n\nLike algebraic solver and the differential equations solver, the 1D integrator separates parameter values, theta, from data values, x_r.\n\n\n\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) Integrates the integrand from a to b.\nAvailable since 2.23\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance) Integrates the integrand from a to b with the given relative tolerance.\nAvailable since 2.23\n\n\nThe arguments to the 1D integrator are as follows:\n\nintegrand: function literal referring to a function specifying the integrand with signature (real, real, array[] real, array[] real, array[] int):real The arguments represent\n\n\nwhere integrand is evaluated,\n\n\ndistance from evaluation point to integration limit for definite integrals,\n\n\nparameters,\n\n\nreal data\n\n\ninteger data, and the return value is the integrand evaluated at the given point,\n\n\na: left limit of integration, may be negative infinity, type real,\nb: right limit of integration, may be positive infinity, type real,\ntheta: parameters only, type array[] real,\nx_r: real data only, type array[] real,\nx_i: integer data only, type array[] int.\n\nA relative_tolerance argument can optionally be provided for more control over the algorithm:\n\nrelative_tolerance: relative tolerance for the 1d integrator, type real, data only.\n\n\n\n\nThe return value for the 1D integrator is a real, the value of the integral.\n\n\n\nFor numeric stability, integrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). Each integral is separately integrated to the given relative_tolerance.\n\n\n\nWhen integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, 1.0 - x). xc (as passed to the integrand) is a high-precision version of the distance between x and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, xc will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, xc is set to NaN.\n\n\n\nInternally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in (Takahasi and Mori 1974), (Mori 1978), (Bailey, Jeyabalan, and Li 2005), and (Tanaka et al. 2009).\nThe gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.\n\n\n\n\n\nStan provides a higher-order reduce function for summation. A function which returns a scalar g: U -&gt; real is mapped to every element of a list of type array[] U, { x1, x2, ... } and all the results are accumulated,\ng(x1) + g(x2) + ...\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g itself, but instead works through evaluating partial sums, f: array[] U -&gt; real, where:\nf({ x1 }) = g(x1)\nf({ x1, x2 }) = g(x1) + g(x2)\nf({ x1, x2, ... }) = g(x1) + g(x2) + ...\nMathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load). This should result in stable numerical evaluations.\n\n\n\nThe higher-order reduce function takes a partial sum function f, an array argument x (with one array element for each term in the sum), a recommended grainsize, and a set of shared arguments. This representation allows parallelization of the resultant sum.\n \n\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n \n real reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nReturns the equivalent of f(x, 1, size(x), s1, s2, ...), but computes the result in parallel by breaking the array x into independent partial sums. s1, s2, ... are shared between all terms in the sum.\nAvailable since 2.23\n\nf: function literal referring to a function specifying the partial sum operation. Refer to the partial sum function.\nx: array of T, one for each term of the reduction, T can be any type,\ngrainsize: For reduce_sum, grainsize is the recommended size of the partial sum (grainsize = 1 means pick totally automatically). For reduce_sum_static, grainsize determines the maximum size of the partial sums, type int,\ns1: first (optional) shared argument, type T1, where T1 can be any type\ns2: second (optional) shared argument, type T2, where T2 can be any type,\n...: remainder of shared arguments, each of which can be any type.\n\n\n\n\nThe partial sum function must have the following signature where the type T, and the types of all the shared arguments (T1, T2, …) match those of the original reduce_sum (reduce_sum_static) call.\n(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real\nThe partial sum function returns the sum of the start to end terms (inclusive) of the overall calculations. The arguments to the partial sum function are:\n\nx_subset, the subset of x a given partial sum is responsible for computing, type array[] T, where T matches the type of x in reduce_sum (reduce_sum_static)\nstart, the index of the first term of the partial sum, type int\nend, the index of the last term of the partial sum (inclusive), type int\ns1, first shared argument, type T1, matching type of s1 in reduce_sum (reduce_sum_static)\ns2, second shared argument, type T2, matching type of s2 in reduce_sum (reduce_sum_static)\n..., remainder of shared arguments, with types matching those in reduce_sum (reduce_sum_static)\n\n\n\n\n\nStan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.\n\n\nThe function being mapped must have a signature identical to that of the function f in the following declaration.\n vector f(vector phi, vector theta,\n          data array[] real x_r, data array[] int x_i);\nThe map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:\n\nphi, the sequence of parameters shared across shards\ntheta, the sequence of parameters specific to this shard\nx_r, sequence of real-valued data\nx_i, sequence of integer data\n\nAll input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.\n\n\n\nThe rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.\n \n\nvector map_rect(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) Return the concatenation of the results of applying the function f, of type (vector, vector, array[] real, array[] int):vector elementwise, i.e., f(phi, theta[n], x_r[n], x_i[n]) for each n in 1:N, where N is the size of the parallel arrays of job-specific/local parameters theta, real data x_r, and integer data x_r. The shared/global parameters phi are passed to each invocation of f.\nAvailable since 2.18",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Void Functions"
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "functions-reference/void_functions.html#fatal-error-statement",
-    "href": "functions-reference/void_functions.html#fatal-error-statement",
-    "title": "Void Functions",
+    "objectID": "functions-reference/higher-order_functions.html#functions-algebraic-solver",
+    "href": "functions-reference/higher-order_functions.html#functions-algebraic-solver",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "The fatal error statement has the same syntax as the print and reject statements, accepting an arbitrary number of arguments of any type (including string literals). The effect of executing a fatal_error statement is to throw an exception internally that terminates the algorithm completely. It can be viewed as an unrecoverable version of reject, and as such should be used only when exiting the algorithm is the only option.\n \n\nvoid fatal_error(T1 x1,..., TN xN) Print the values denoted by the arguments x1 through xN on the output message stream and then exit the currently running algorithm. There are no spaces between items in the print, but a line feed (LF; Unicode U+000A; C++ literal '\\n') is inserted at the end of the printed line. The types T1 through TN can be any of Stan’s built-in numerical types or double quoted strings of characters (bytes).\nAvailable since 2.35",
+    "text": "Stan provides two built-in algebraic equation solvers, respectively based on the Newton method and the Powell “dog leg” hybrid method. Empirically the Newton method is found to be faster and its use is recommended for most problems.\nAn algebraic solver is a higher-order function, i.e. it takes another function as one of its arguments. Other functions in Stan which share this feature are the differential equation solvers (see section Ordinary Differential Equation (ODE) Solvers and Differential Algebraic Equation (DAE) solver). Ordinary Stan functions do not allow functions as arguments.\n\n\nAn algebraic system is specified as an ordinary function in Stan within the function block. The function must return a vector and takes in, as its first argument, the unknowns \\(y\\) we wish to solve for, also passed as a vector. This argument is followed by additional arguments as specified by the user; we call such arguments variadic arguments and denote them .... The signature of the algebraic system is then:\n vector algebra_system (vector y, ...)\nThere is no type restriction for the variadic arguments and each argument can be passed as data or parameter. However users should use parameter arguments only when necessary and mark data arguments with the keyword data. In the below example, the last variadic argument, \\(x\\), is restricted to being data:\n  vector algebra_system (vector y, vector theta, data vector x)\nDistinguishing data and parameter is important for computational reasons. Augmenting the total number of parameters increases the cost of propagating derivatives through the solution to the algebraic equation, and ultimately the computational cost of evaluating the gradients.\n\n\n\n \n vector solve_newton(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Newton’s method.\nAvailable since 2.31\n \n vector solve_newton_tol(function algebra_system, vector y_guess, data real scaling_step, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Newton’s method with additional control parameters for the solver.\nAvailable since 2.31\n \n vector solve_powell(function algebra_system, vector y_guess, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method.\nAvailable since 2.31\n \n vector solve_powell_tol(function algebra_system, vector y_guess, data real rel_tol, data real f_tol, int max_steps, ...) Solves the algebraic system, given an initial guess, using Powell’s hybrid method with additional control parameters for the solver.\nAvailable since 2.31\n\n\nThe arguments to the algebraic solvers are as follows:\n\nalgebra_system: function literal referring to a function specifying the system of algebraic equations with signature (vector, ...):vector. The arguments represent (1) unknowns, (2) additional parameter and/or data arguments, and the return value contains the value of the algebraic function, which goes to 0 when we plug in the solution to the algebraic system,\ny_guess: initial guess for the solution, type vector,\n...: variadic arguments.\n\nThe algebraic solvers admit control parameters. While Stan provides default values, the user should be prepared to adjust the control parameters. The following controls are available:\n\nscaling_step: for the Newton solver only, the scaled-step stopping tolerance, type real, data only. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. Default value is \\(10^{-3}\\).\nrel_tol: for the Powell solver only, the relative tolerance, type real, data only. The relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. Default value is \\(10^{-10}\\).\nfunction_tol: function tolerance for the algebraic solver, type real, data only. After convergence of the solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. Default value is \\(10^{-6}\\).\nmax_num_steps: maximum number of steps to take in the algebraic solver, type int, data only. If the solver reaches this number of steps, it breaks and returns an error message. Default value is \\(200\\).\n\nThe difference in which control parameters are available has to do with the underlying implementations for the solvers and the control parameters these implementations support. The Newton solver is based on KINSOL from the SUNDIAL suites, while the Powell solver uses a module from the Eigen library.\n\n\n\nThe return value for the algebraic solver is an object of type vector, with values which, when plugged in as y make the algebraic function go to 0 (approximately, within the specified function tolerance).\n\n\n\nCertain sizes have to be consistent. The initial guess, return value of the solver, and return value of the algebraic function must all be the same size.\n\n\n\nStan offers two methods to solve algebraic equations. solve_newton and solve_newton_tol use the Newton method, a first-order derivative based numerical solver. The Stan code builds on the implementation in KINSOL from the SUNDIALS suite (Hindmarsh et al. 2005). For many problems, we find that the Newton method is faster than the Powell method. If however Newton’s method performs poorly, either failing to or requiring an excessively long time to converge, the user should be prepared to switch to the Powell method.\nsolve_powell and solve_powell_tol are based on the Powell hybrid method (Powell 1970), which also uses first-order derivatives. The Stan code builds on the implementation of the hybrid solver in the unsupported module for nonlinear optimization problems of the Eigen library (Guennebaud, Jacob, et al. 2010). This solver is in turn based on the algorithm developed for the package MINPACK-1 (Jorge J. More 1980).\nFor both solvers, derivatives are propagated through the solution to the algebraic solution using the implicit function theorem and an adjoint method of automatic differentiation; for a discussion on this topic, see (Gaebler 2021) and (Margossian and Betancourt 2022).",
     "crumbs": [
       "Functions Reference",
       "Built-in Functions",
-      "Void Functions"
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html",
-    "href": "reference-manual/blocks.html",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/higher-order_functions.html#functions-ode-solver",
+    "href": "functions-reference/higher-order_functions.html#functions-ode-solver",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "A Stan program is organized into a sequence of named blocks, the bodies of which consist of variable declarations, followed in the case of some blocks with statements.\n\n\nThe full set of named program blocks is exemplified in the following skeletal Stan program.\nfunctions {\n  // ... function declarations and definitions ...\n}\ndata {\n  // ... declarations ...\n}\ntransformed data {\n   // ... declarations ... statements ...\n}\nparameters {\n   // ... declarations ...\n}\ntransformed parameters {\n   // ... declarations ... statements ...\n}\nmodel {\n   // ... declarations ... statements ...\n}\ngenerated quantities {\n   // ... declarations ... statements ...\n}\nThe function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.\n\n\nAll of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.\n\n\n\nThe variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.\nVariables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).\n\n\n\nFunctions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).\nRandom-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in _rng; such functions are suffixed with _rng. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with _lp.\nDensity functions defined in the program may be used in distribution statements.\n\n\n\nThe variables declared in the data and parameters block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.\nThe variables in the data block are read from an external input source such as a file or a designated R data structure. The variables in the parameters block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.\n\n\n\nThe transformed data and transformed parameters block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.\nBefore generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.1\nTransformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.\n\n\n\nThe generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.\n\n\n\nA table summarizing the point at which variables are read, written, and defined is given in the block actions table.\nBlock Actions Table.  The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.\n\n\n\nblock\nstatement\naction / period\n\n\n\n\ndata\nno\nread / chain\n\n\ntransformed data\nyes\nevaluate / chain\n\n\nparameters\nno\ninv. transform, Jacobian / leapfrog\n\n\n \n \ninv. transform, write / sample\n\n\ntransformed parameters\nyes\nevaluate / leapfrog\n\n\n \n \nwrite / sample\n\n\nmodel\nyes\nevaluate / leapfrog step\n\n\ngenerated quantities\nyes\neval / sample\n\n\n \n \nwrite / sample\n\n\n(initialization)\nn/a\nread, transform / chain\n\n\n\nVariable Declaration Table.  This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (\\(*\\)) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.\n\n\n\nparam depend\nin target\nsave\ndeclare in\n\n\n\n\n+\n+\n+\ntransformed parameters\n\n\n+\n+\n-\nmodel (local)\n\n\n+\n-\n+\ngenerated quantities\n\n\n+\n-\n-\ngenerated quantities (local)\n\n\n-\n+\n+\ntransformed data   and generated quantities\n\n\n-\n+\n-\ntransformed data\n\n\n-\n-\n+\ngenerated quantities\n\n\n-\n-\n-\ntransformed data (local)\n\n\n\nAnother way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the variable declaration table. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.2\nThe rest of this chapter provides full details on when and how the variables and statements in each block are executed.\n\n\n\n\nStatistical Variable Taxonomy Table.  Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.\n\n\n\n\n\n\n\nvariable kind\ndeclaration block\n\n\n\n\nconstants\ndata, transformed data\n\n\nunmodeled data\ndata, transformed data\n\n\nmodeled data\ndata, transformed data\n\n\nmissing data\nparameters, transformed parameters\n\n\nmodeled parameters\nparameters, transformed parameters\n\n\nunmodeled parameters\ndata, transformed data\n\n\nderived quantities\ntransformed data, transformed parameters, generated quantities\n\n\nloop indices\nloop statement\n\n\n\nPage 366 of (Gelman and Hill 2007) provides a taxonomy of the kinds of variables used in Bayesian models. The table of kinds of variables contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.\nConstants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the data block.\nThe following program illustrates various variables kinds, listing the kind of each variable next to its declaration.\ndata {\n  int&lt;lower=0&gt; N;           // unmodeled data\n  array[N] real y;          // modeled data\n  real mu_mu;               // config. unmodeled param\n  real&lt;lower=0&gt; sigma_mu;   // config. unmodeled param\n}\ntransformed data {\n  real&lt;lower=0&gt; alpha;      // const. unmodeled param\n  real&lt;lower=0&gt; beta;       // const. unmodeled param\n  alpha = 0.1;\n  beta = 0.1;\n}\nparameters {\n  real mu_y;                // modeled param\n  real&lt;lower=0&gt; tau_y;      // modeled param\n}\ntransformed parameters {\n  real&lt;lower=0&gt; sigma_y;    // derived quantity (param)\n  sigma_y = pow(tau_y, -0.5);\n}\nmodel {\n  tau_y ~ gamma(alpha, beta);\n  mu_y ~ normal(mu_mu, sigma_mu);\n  for (n in 1:N) {\n    y[n] ~ normal(mu_y, sigma_y);\n  }\n}\ngenerated quantities {\n  real variance_y;       // derived quantity (transform)\n  variance_y = sigma_y * sigma_y;\n}\nIn this example, y is an array of modeled data. Although it is specified in the data block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.\nThe variable N is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.\nThe other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the data block may be specified on a per-model execution basis as part of the data read. In the above model, mu_mu and sigma_mu are configurable unmodeled parameters.\nUnmodeled parameters that are hard coded in the model must be declared in the transformed data block. For example, the unmodeled parameters alpha and beta are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the data block, like the variables mu_mu and sigma_mu.\nThis program declares two modeled parameters, mu_y and tau_y. These are the location and precision used in the normal model of the values in y. The heart of the model will be sampling the values of these parameters from their posterior distribution.\nThe modeled parameter tau_y is transformed from a precision to a scale parameter and assigned to the variable sigma_y in the transformed parameters block. Thus the variable sigma_y is considered a derived quantity — its value is entirely determined by the values of other variables.\nThe generated quantities block defines a value variance_y, which is defined as a transform of the scale or deviation parameter sigma_y. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).\nIn later versions of Stan which have random number generators for the distributions, the generated quantities block will be usable to generate replicated data for model checking.\nFinally, the variable n is used as a loop index in the model block.\n\n\n\nThe rest of this chapter will lay out the details of each block in order, starting with the data block in this section.\n\n\nThe data block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.3\nData variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.\n\n\n\nThe data block does not allow statements.\n\n\n\nEach variable’s value is validated against its declaration as it is read. For example, if a variable sigma is declared as real&lt;lower=0&gt;, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.\n\n\n\n\nThe transformed data block is for declaring and defining variables that do not need to be changed when running the program.\n\n\nFor the transformed data block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.\nVariables declared in the data block may be used to declare transformed variables.\n\n\n\nThe statements in a transformed data block are used to define (provide values for) variables declared in the transformed data block. Assignments are only allowed to variables declared in the transformed data block.\nThese statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.\nVariables declared in the data block may be used in statements in the transformed data block.\n\n\nThe statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and distribution statements may not be used.\n\n\n\n\nAny constraints on variables declared in the transformed data block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.\n\n\n\n\nThe variables declared in the parameters program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block are the parameters being sampled by Stan.\nVariables declared as parameters cannot be directly assigned values. So there is no block of statements in the parameters program block. Variable quantities derived from parameters may be declared in the transformed parameters or generated quantities blocks, or may be defined as local variables in any statement blocks following their declaration.\nThere is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.\n\n\nStan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of \\(\\mathbb{R}^n\\). To do this, the parameters defined in the parameters block must be transformed so they are unconstrained.\nIn practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s parameters program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.\nIn some cases, the number of parameters is reduced in the unconstrained space. For instance, a \\(K\\)-simplex only requires \\(K-1\\) unconstrained parameters, and a \\(K\\)-correlation matrix only requires \\(\\binom{K}{2}\\) unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the parameters program block.\nThe probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the parameters program block. The details of the specific transformations are provided in the variable transforms chapter.\n\n\n\nHamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.\nGradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.\nThe amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.\n\n\n\nIn the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the parameters block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.\n\n\n\n\nThe transformed parameters program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.\nAny variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.\n\n\nLike the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the section on reject statements for further discussion of the behavior of reject statements.\n\n\n\n\nThe model program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.\nLocal variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.\nThe statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.\n\n\n\nThe generated quantities program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.\nAmong the applications of posterior inference that can be coded in the generated quantities block are\n\nforward sampling to generate simulated data for model testing,\ngenerating predictions for new data,\ncalculating posterior event probabilities, including multiple comparisons, sign tests, etc.,\ncalculating posterior expectations,\ntransforming parameters for reporting,\napplying full Bayesian decision theory,\ncalculating log likelihoods, deviances, etc. for model comparison.\n\nParameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.\nWithin the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.\nIt is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.\nAfter the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.\nAll variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:\ngenerated quantities {\n  int a; // added to the output\n\n  {\n    int b; // not added to the output\n  }\n}",
+    "text": "Stan provides several higher order functions for solving initial value problems specified as Ordinary Differential Equations (ODEs).\nSolving an initial value ODE means given a set of differential equations \\(y'(t, \\theta) = f(t, y, \\theta)\\) and initial conditions \\(y(t_0, \\theta)\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 &lt; t_2, \\cdots &lt; t_n\\). \\(f(t, y, \\theta)\\) is referred to here as the ODE system function.\n\\(f(t, y, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the ODE solver functions.\nTo make it easier to write ODEs, the solve functions take extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as .... The types of the arguments represented by ... in the ODE solve function call must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector ode_rk45(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method.\nAvailable since 2.24\n \n\narray[] vector ode_rk45_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Dormand-Prince algorithm, a 4th/5th order Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.24\n \n\narray[] vector ode_ckrk(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method.\nAvailable since 2.27\n \n\narray[] vector ode_ckrk_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Cash-Karp algorithm, a 4th/5th order explicit Runge-Kutta method with additional control parameters for the solver.\nAvailable since 2.27\n \n\narray[] vector ode_adams(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the Adams-Moulton method.\nAvailable since 2.24\n \n\narray[] vector ode_adams_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the Adams-Moulton method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_bdf(function ode, vector initial_state, real initial_time, array[] real times, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method.\nAvailable since 2.24\n \n\narray[] vector ode_bdf_tol(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the ODE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.24\n\n\n\n \n\narray[] vector ode_adjoint_tol_ctl(function ode, vector initial_state, real initial_time, array[] real times, data real rel_tol_forward, data vector abs_tol_forward, data real rel_tol_backward, data vector abs_tol_backward, int max_num_steps, int num_steps_between_checkpoints, int interpolation_polynomial, int solver_forward, int solver_backward, ...)\nSolves the ODE system for the times provided using the adjoint ODE solver method from CVODES. The adjoint ODE solver requires a checkpointed forward in time ODE integration, a backwards in time integration that makes uses of an interpolated version of the forward solution, and the solution of a quadrature problem (the number of which depends on the number of parameters passed to the solve). The tolerances and numeric methods used for the forward solve, backward solve, quadratures, and interpolation can all be configured.\nAvailable since 2.27\n\n\n\nThe first argument to one of the ODE solvers is always the ODE system function. The ODE system function must have a vector return type, and the first two arguments must be a real and vector in that order. These two arguments are followed by the variadic arguments that are passed through from the ODE solve function call:\n vector ode(real time, vector state, ...)\nThe ODE system function should return the derivative of the state with respect to time at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the ODE system\nstate, the state of the ODE system at the time specified\n..., sequence of arguments passed unmodified from the ODE solve function call. The types here must match the types in the ... arguments of the ODE solve function call.\n\n\n\n\nThe arguments to the ODE solvers in both the stiff and non-stiff solvers are the same. The arguments to the adjoint ODE solver are different; see Arguments to the adjoint ODE solver.\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nFor the versions of the ode solver functions ending in _tol, these three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the ODE solver, type real, data only,\ndata abs_tol: absolute tolerance for the ODE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the ODE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\n\nThe arguments to the adjoint ODE solver are different from those for the other functions (for those see Arguments to the ODE solvers).\n\node: ODE system function,\ninitial_state: initial state, type vector,\ninitial_time: initial time, type real,\ntimes: solution times, type array[] real,\ndata rel_tol_forward: Relative tolerance for forward solve, type real, data only,\ndata abs_tol_forward: Absolute tolerance vector for each state for forward solve, type vector, data only,\ndata rel_tol_backward: Relative tolerance for backward solve, type real, data only,\ndata abs_tol_backward: Absolute tolerance vector for each state for backward solve, type vector, data only,\ndata rel_tol_quadrature: Relative tolerance for backward quadrature, type real, data only,\ndata abs_tol_quadrature: Absolute tolerance for backward quadrature, type real, data only,\ndata max_num_steps: Maximum number of time-steps to take in integrating the ODE solution between output time points for forward and backward solve, type int, data only,\nnum_steps_between_checkpoints: number of steps between checkpointing forward solution, type int, data only,\ninterpolation_polynomial: can be 1 for hermite or 2 for polynomial interpolation method of CVODES, type int, data only,\nsolver_forward: solver used for forward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only,\nsolver_backward: solver used for backward ODE problem: 1=Adams (non-stiff), 2=BDF (stiff), type int, data only.\n...: sequence of arguments that will be passed through unmodified to the ODE system function. The types here must match the types in the ... arguments of the ODE system function.\n\nBecause the tolerances are data arguments, they must be defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe return value for the ODE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time in specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables passed into the system function, derivatives returned by the system function, initial state passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#overview-of-stans-program-blocks",
-    "href": "reference-manual/blocks.html#overview-of-stans-program-blocks",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/higher-order_functions.html#functions-dae-solver",
+    "href": "functions-reference/higher-order_functions.html#functions-dae-solver",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "The full set of named program blocks is exemplified in the following skeletal Stan program.\nfunctions {\n  // ... function declarations and definitions ...\n}\ndata {\n  // ... declarations ...\n}\ntransformed data {\n   // ... declarations ... statements ...\n}\nparameters {\n   // ... declarations ...\n}\ntransformed parameters {\n   // ... declarations ... statements ...\n}\nmodel {\n   // ... declarations ... statements ...\n}\ngenerated quantities {\n   // ... declarations ... statements ...\n}\nThe function-definition block contains user-defined functions. The data block declares the required data for the model. The transformed data block allows the definition of constants and transforms of the data. The parameters block declares the model’s parameters — the unconstrained version of the parameters is what’s sampled or optimized. The transformed parameters block allows variables to be defined in terms of data and parameters that may be used later and will be saved. The model block is where the log probability function is defined. The generated quantities block allows derived quantities based on parameters, data, and optionally (pseudo) random number generation.\n\n\nAll of the blocks are optional. A consequence of this is that the empty string is a valid Stan program, although it will trigger a warning message from the Stan compiler. The Stan program blocks that occur must occur in the order presented in the skeletal program above. Within each block, both declarations and statements are optional, subject to the restriction that the declarations come before the statements.\n\n\n\nThe variables declared in each block have scope over all subsequent statements. Thus a variable declared in the transformed data block may be used in the model block. But a variable declared in the generated quantities block may not be used in any earlier block, including the model block. The exception to this rule is that variables declared in the model block are always local to the model block and may not be accessed in the generated quantities block; to make a variable accessible in the model and generated quantities block, it must be declared as a transformed parameter.\nVariables declared as function parameters have scope only within that function definition’s body, and may not be assigned to (they are constant).\n\n\n\nFunctions defined in the function block may be used in any appropriate block. Most functions can be used in any block and applied to a mixture of parameters and data (including constants or program literals).\nRandom-number-generating functions are restricted to transformed data and generated quantities blocks, and within user-defined functions ending in _rng; such functions are suffixed with _rng. Log-probability modifying functions to blocks where the log probability accumulator is in scope (transformed parameters and model); such functions are suffixed with _lp.\nDensity functions defined in the program may be used in distribution statements.\n\n\n\nThe variables declared in the data and parameters block are treated differently than other variables in that they are automatically defined by the context in which they are used. This is why there are no statements allowed in the data or parameters block.\nThe variables in the data block are read from an external input source such as a file or a designated R data structure. The variables in the parameters block are read from the sampler’s current parameter values (either standard HMC or NUTS). The initial values may be provided through an external input source, which is also typically a file or a designated R data structure. In each case, the parameters are instantiated to the values for which the model defines a log probability function.\n\n\n\nThe transformed data and transformed parameters block behave similarly to each other. Both allow new variables to be declared and then defined through a sequence of statements. Because variables scope over every statement that follows them, transformed data variables may be defined in terms of the data variables.\nBefore generating any draws, data variables are read in, then the transformed data variables are declared and the associated statements executed to define them. This means the statements in the transformed data block are only ever evaluated once.1\nTransformed parameters work the same way, being defined in terms of the parameters, transformed data, and data variables. The difference is the frequency of evaluation. Parameters are read in and (inverse) transformed to constrained representations on their natural scales once per log probability and gradient evaluation. This means the inverse transforms and their log absolute Jacobian determinants are evaluated once per leapfrog step. Transformed parameters are then declared and their defining statements executed once per leapfrog step.\n\n\n\nThe generated quantity variables are defined once per sample after all the leapfrog steps have been completed. These may be random quantities, so the block must be rerun even if the Metropolis adjustment of HMC or NUTS rejects the update proposal.\n\n\n\nA table summarizing the point at which variables are read, written, and defined is given in the block actions table.\nBlock Actions Table.  The read, write, transform, and evaluate actions and periodicities listed in the last column correspond to the Stan program blocks in the first column. The middle column indicates whether the block allows statements. The last row indicates that parameter initialization requires a read and transform operation applied once per chain.\n\n\n\nblock\nstatement\naction / period\n\n\n\n\ndata\nno\nread / chain\n\n\ntransformed data\nyes\nevaluate / chain\n\n\nparameters\nno\ninv. transform, Jacobian / leapfrog\n\n\n \n \ninv. transform, write / sample\n\n\ntransformed parameters\nyes\nevaluate / leapfrog\n\n\n \n \nwrite / sample\n\n\nmodel\nyes\nevaluate / leapfrog step\n\n\ngenerated quantities\nyes\neval / sample\n\n\n \n \nwrite / sample\n\n\n(initialization)\nn/a\nread, transform / chain\n\n\n\nVariable Declaration Table.  This table indicates where variables that are not basic data or parameters should be declared, based on whether it is defined in terms of parameters, whether it is used in the log probability function defined in the model block, and whether it is printed. The two lines marked with asterisks (\\(*\\)) should not be used as there is no need to print a variable every iteration that does not depend on the value of any parameters.\n\n\n\nparam depend\nin target\nsave\ndeclare in\n\n\n\n\n+\n+\n+\ntransformed parameters\n\n\n+\n+\n-\nmodel (local)\n\n\n+\n-\n+\ngenerated quantities\n\n\n+\n-\n-\ngenerated quantities (local)\n\n\n-\n+\n+\ntransformed data   and generated quantities\n\n\n-\n+\n-\ntransformed data\n\n\n-\n-\n+\ngenerated quantities\n\n\n-\n-\n-\ntransformed data (local)\n\n\n\nAnother way to look at the variables is in terms of their function. To decide which variable to use, consult the charts in the variable declaration table. The last line has no corresponding location, as there is no need to print a variable every iteration that does not depend on parameters.2\nThe rest of this chapter provides full details on when and how the variables and statements in each block are executed.",
+    "text": "Stan provides two higher order functions for solving initial value problems specified as Differential-Algebraic Equations (DAEs) with index-1 (Serban et al. 2021).\nSolving an initial value DAE means given a set of residual functions \\(r(y'(t, \\theta), y(t, \\theta), t)\\) and initial conditions \\((y(t_0, \\theta), y'(t_0, \\theta))\\), solving for \\(y\\) at a sequence of times \\(t_0 &lt; t_1 \\leq t_2, \\cdots \\leq t_n\\). The residual function \\(r(y', y, t, \\theta)\\) will be defined as a function with a certain signature and provided along with the initial conditions and output times to one of the DAE solver functions.\nSimilar to ODE solvers, the DAE solver function takes extra arguments that are passed along unmodified to the user-supplied system function. Because there can be any number of these arguments and they can be of different types, they are denoted below as ..., and the types of these arguments, also represented by ... in the DAE solver call, must match the types of the arguments represented by ... in the user-supplied system function.\n\n\n \n\narray[] vector dae(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, ...) Solves the DAE system using the backward differentiation formula (BDF) method (Serban et al. 2021).\nAvailable since 2.29\n \n\narray[] vector dae_tol(function residual, vector initial_state, vector initial_state_derivative, data real initial_time, data array[] real times, data real rel_tol, data real abs_tol, int max_num_steps, ...) Solves the DAE system for the times provided using the backward differentiation formula (BDF) method with additional control parameters for the solver.\nAvailable since 2.29\n\n\n\nThe first argument to the DAE solver is the DAE residual function. The DAE residual function must have a vector return type, and the first three arguments must be a real, vector, and vector, in that order. These three arguments are followed by the variadic arguments that are passed through from the DAE solver function call:\n  vector residual(real time, vector state, vector state_derivative, ...)\nThe DAE residual function should return the residuals at the time and state provided. The length of the returned vector must match the length of the state input into the function.\nThe arguments to this function are:\n\ntime, the time to evaluate the DAE system\nstate, the state of the DAE system at the time specified\nstate_derivative, the time derivatives of the state of the DAE system at the time specified\n..., sequence of arguments passed unmodified from the DAE solve function call. The types here must match the types in the ... arguments of the DAE solve function call.\n\n\n\n\nThe arguments to the DAE solver are\n\nresidual: DAE residual function,\ninitial_state: initial state, type vector,\ninitial_state_derivative: time derivative of the initial state, type vector,\ninitial_time: initial time, type data real,\ntimes: solution times, type data array[] real,\n...: sequence of arguments that will be passed through unmodified to the DAE residual function. The types here must match the types in the ... arguments of the DAE residual function.\n\nFor dae_tol, the following three parameters must be provided after times and before the ... arguments:\n\ndata rel_tol: relative tolerance for the DAE solver, type real, data only,\ndata abs_tol: absolute tolerance for the DAE solver, type real, data only, and\nmax_num_steps: maximum number of steps to take between output times in the DAE solver, type int, data only.\n\nBecause the tolerances are data arguments, they must be supplied as primitive numerics or defined in either the data or transformed data blocks. They cannot be parameters, transformed parameters or functions of parameters or transformed parameters.\n\n\nThe user is responsible to ensure the residual function becomes zero at the initial time, t0, when the arguments initial_state and initial_state_derivative are introduced as state and state_derivative, respectively.\n\n\n\nThe return value for the DAE solvers is an array of vectors (type array[] vector), one vector representing the state of the system at every time specified in the times argument.\n\n\n\nThe sizes must match, and in particular, the following groups are of the same size:\n\nstate variables and state derivatives passed into the residual function, the residual returned by the residual function, initial state and initial state derivatives passed into the solver, and length of each vector in the output,\nnumber of solution times and number of vectors in the output.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#statistical-variable-taxonomy",
-    "href": "reference-manual/blocks.html#statistical-variable-taxonomy",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/higher-order_functions.html#functions-1d-integrator",
+    "href": "functions-reference/higher-order_functions.html#functions-1d-integrator",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "Statistical Variable Taxonomy Table.  Variables of the kind indicated in the left column must be declared in one of the blocks declared in the right column.\n\n\n\n\n\n\n\nvariable kind\ndeclaration block\n\n\n\n\nconstants\ndata, transformed data\n\n\nunmodeled data\ndata, transformed data\n\n\nmodeled data\ndata, transformed data\n\n\nmissing data\nparameters, transformed parameters\n\n\nmodeled parameters\nparameters, transformed parameters\n\n\nunmodeled parameters\ndata, transformed data\n\n\nderived quantities\ntransformed data, transformed parameters, generated quantities\n\n\nloop indices\nloop statement\n\n\n\nPage 366 of (Gelman and Hill 2007) provides a taxonomy of the kinds of variables used in Bayesian models. The table of kinds of variables contains Gelman and Hill’s taxonomy along with a missing-data kind along with the corresponding locations of declarations and definitions in Stan.\nConstants can be built into a model as literals, data variables, or as transformed data variables. If specified as variables, their definition must be included in data files. If they are specified as transformed data variables, they cannot be used to specify the sizes of elements in the data block.\nThe following program illustrates various variables kinds, listing the kind of each variable next to its declaration.\ndata {\n  int&lt;lower=0&gt; N;           // unmodeled data\n  array[N] real y;          // modeled data\n  real mu_mu;               // config. unmodeled param\n  real&lt;lower=0&gt; sigma_mu;   // config. unmodeled param\n}\ntransformed data {\n  real&lt;lower=0&gt; alpha;      // const. unmodeled param\n  real&lt;lower=0&gt; beta;       // const. unmodeled param\n  alpha = 0.1;\n  beta = 0.1;\n}\nparameters {\n  real mu_y;                // modeled param\n  real&lt;lower=0&gt; tau_y;      // modeled param\n}\ntransformed parameters {\n  real&lt;lower=0&gt; sigma_y;    // derived quantity (param)\n  sigma_y = pow(tau_y, -0.5);\n}\nmodel {\n  tau_y ~ gamma(alpha, beta);\n  mu_y ~ normal(mu_mu, sigma_mu);\n  for (n in 1:N) {\n    y[n] ~ normal(mu_y, sigma_y);\n  }\n}\ngenerated quantities {\n  real variance_y;       // derived quantity (transform)\n  variance_y = sigma_y * sigma_y;\n}\nIn this example, y is an array of modeled data. Although it is specified in the data block, and thus must have a known value before the program may be run, it is modeled as if it were generated randomly as described by the model.\nThe variable N is a typical example of unmodeled data. It is used to indicate a size that is not part of the model itself.\nThe other variables declared in the data and transformed data block are examples of unmodeled parameters, also known as hyperparameters. Unmodeled parameters are parameters to probability densities that are not themselves modeled probabilistically. In Stan, unmodeled parameters that appear in the data block may be specified on a per-model execution basis as part of the data read. In the above model, mu_mu and sigma_mu are configurable unmodeled parameters.\nUnmodeled parameters that are hard coded in the model must be declared in the transformed data block. For example, the unmodeled parameters alpha and beta are both hard coded to the value 0.1. To allow such variables to be configurable based on data supplied to the program at run time, they must be declared in the data block, like the variables mu_mu and sigma_mu.\nThis program declares two modeled parameters, mu_y and tau_y. These are the location and precision used in the normal model of the values in y. The heart of the model will be sampling the values of these parameters from their posterior distribution.\nThe modeled parameter tau_y is transformed from a precision to a scale parameter and assigned to the variable sigma_y in the transformed parameters block. Thus the variable sigma_y is considered a derived quantity — its value is entirely determined by the values of other variables.\nThe generated quantities block defines a value variance_y, which is defined as a transform of the scale or deviation parameter sigma_y. It is defined in the generated quantities block because it is not used in the model. Making it a generated quantity allows it to be monitored for convergence (being a non-linear transform, it will have different autocorrelation and hence convergence properties than the deviation itself).\nIn later versions of Stan which have random number generators for the distributions, the generated quantities block will be usable to generate replicated data for model checking.\nFinally, the variable n is used as a loop index in the model block.",
+    "text": "Stan provides a built-in mechanism to perform 1D integration of a function via quadrature methods.\nIt operates similarly to the algebraic solver and the ordinary differential equations solver in that it allows as an argument a function.\nLike both of those utilities, some of the arguments are limited to data only expressions. These expressions must not contain variables other than those declared in the data or transformed data blocks.\n\n\nPerforming a 1D integration requires the integrand to be specified somehow. This is done by defining a function in the Stan functions block with the special signature:\nreal integrand(real x, real xc, array[] real theta,\n               array[] real x_r, array[] int x_i)\nThe function should return the value of the integrand evaluated at the point x.\nThe argument of this function are:\n\nx, the independent variable being integrated over\nxc, a high precision version of the distance from x to the nearest endpoint in a definite integral (for more into see section Precision Loss).\ntheta, parameter values used to evaluate the integral\nx_r, data values used to evaluate the integral\nx_i, integer data used to evaluate the integral\n\nLike algebraic solver and the differential equations solver, the 1D integrator separates parameter values, theta, from data values, x_r.\n\n\n\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i) Integrates the integrand from a to b.\nAvailable since 2.23\n \n\nreal integrate_1d (function integrand, real a, real b, array[] real theta, array[] real x_r, array[] int x_i, real relative_tolerance) Integrates the integrand from a to b with the given relative tolerance.\nAvailable since 2.23\n\n\nThe arguments to the 1D integrator are as follows:\n\nintegrand: function literal referring to a function specifying the integrand with signature (real, real, array[] real, array[] real, array[] int):real The arguments represent\n\n\nwhere integrand is evaluated,\n\n\ndistance from evaluation point to integration limit for definite integrals,\n\n\nparameters,\n\n\nreal data\n\n\ninteger data, and the return value is the integrand evaluated at the given point,\n\n\na: left limit of integration, may be negative infinity, type real,\nb: right limit of integration, may be positive infinity, type real,\ntheta: parameters only, type array[] real,\nx_r: real data only, type array[] real,\nx_i: integer data only, type array[] int.\n\nA relative_tolerance argument can optionally be provided for more control over the algorithm:\n\nrelative_tolerance: relative tolerance for the 1d integrator, type real, data only.\n\n\n\n\nThe return value for the 1D integrator is a real, the value of the integral.\n\n\n\nFor numeric stability, integrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). Each integral is separately integrated to the given relative_tolerance.\n\n\n\nWhen integrating certain definite integrals, there can be significant precision loss in evaluating the integrand near the endpoints. This has to do with the breakdown in precision of double precision floating point values when adding or subtracting a small number from a number much larger than it in magnitude (for instance, 1.0 - x). xc (as passed to the integrand) is a high-precision version of the distance between x and the definite integral endpoints and can be used to address this issue. More information (and an example where this is useful) is given in the User’s Guide. For zero crossing integrals, xc will be a high precision version of the distance to the endpoints of the two smaller integrals. For any integral with an endpoint at negative infinity or positive infinity, xc is set to NaN.\n\n\n\nInternally the 1D integrator uses the double-exponential methods in the Boost 1D quadrature library. Boost in turn makes use of quadrature methods developed in (Takahasi and Mori 1974), (Mori 1978), (Bailey, Jeyabalan, and Li 2005), and (Tanaka et al. 2009).\nThe gradients of the integral are computed in accordance with the Leibniz integral rule. Gradients of the integrand are computed internally with Stan’s automatic differentiation.",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#program-block-data",
-    "href": "reference-manual/blocks.html#program-block-data",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/higher-order_functions.html#functions-reduce",
+    "href": "functions-reference/higher-order_functions.html#functions-reduce",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "The rest of this chapter will lay out the details of each block in order, starting with the data block in this section.\n\n\nThe data block is for the declaration of variables that are read in as data. With the current model executable, each Markov chain of draws will be executed in a different process, and each such process will read the data exactly once.3\nData variables are not transformed in any way. The format for data files or data in memory depends on the interface; see the user’s guides and interface documentation for PyStan, RStan, and CmdStan for details.\n\n\n\nThe data block does not allow statements.\n\n\n\nEach variable’s value is validated against its declaration as it is read. For example, if a variable sigma is declared as real&lt;lower=0&gt;, then trying to assign it a negative value will raise an error. As a result, data type errors will be caught as early as possible. Similarly, attempts to provide data of the wrong size for a compound data structure will also raise an error.",
+    "text": "Stan provides a higher-order reduce function for summation. A function which returns a scalar g: U -&gt; real is mapped to every element of a list of type array[] U, { x1, x2, ... } and all the results are accumulated,\ng(x1) + g(x2) + ...\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g itself, but instead works through evaluating partial sums, f: array[] U -&gt; real, where:\nf({ x1 }) = g(x1)\nf({ x1, x2 }) = g(x1) + g(x2)\nf({ x1, x2, ... }) = g(x1) + g(x2) + ...\nMathematically the summation reduction is associative and forming arbitrary partial sums in an arbitrary order will not change the result. However, floating point numerics on computers only have a limited precision such that associativity does not hold exactly. This implies that the order of summation determines the exact numerical result. For this reason, the higher-order reduce function is available in two variants:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load). This should result in stable numerical evaluations.\n\n\n\nThe higher-order reduce function takes a partial sum function f, an array argument x (with one array element for each term in the sum), a recommended grainsize, and a set of shared arguments. This representation allows parallelization of the resultant sum.\n \n\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n \n real reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nReturns the equivalent of f(x, 1, size(x), s1, s2, ...), but computes the result in parallel by breaking the array x into independent partial sums. s1, s2, ... are shared between all terms in the sum.\nAvailable since 2.23\n\nf: function literal referring to a function specifying the partial sum operation. Refer to the partial sum function.\nx: array of T, one for each term of the reduction, T can be any type,\ngrainsize: For reduce_sum, grainsize is the recommended size of the partial sum (grainsize = 1 means pick totally automatically). For reduce_sum_static, grainsize determines the maximum size of the partial sums, type int,\ns1: first (optional) shared argument, type T1, where T1 can be any type\ns2: second (optional) shared argument, type T2, where T2 can be any type,\n...: remainder of shared arguments, each of which can be any type.\n\n\n\n\nThe partial sum function must have the following signature where the type T, and the types of all the shared arguments (T1, T2, …) match those of the original reduce_sum (reduce_sum_static) call.\n(array[] T x_subset, int start, int end, T1 s1, T2 s2, ...):real\nThe partial sum function returns the sum of the start to end terms (inclusive) of the overall calculations. The arguments to the partial sum function are:\n\nx_subset, the subset of x a given partial sum is responsible for computing, type array[] T, where T matches the type of x in reduce_sum (reduce_sum_static)\nstart, the index of the first term of the partial sum, type int\nend, the index of the last term of the partial sum (inclusive), type int\ns1, first shared argument, type T1, matching type of s1 in reduce_sum (reduce_sum_static)\ns2, second shared argument, type T2, matching type of s2 in reduce_sum (reduce_sum_static)\n..., remainder of shared arguments, with types matching those in reduce_sum (reduce_sum_static)",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#program-block-transformed-data",
-    "href": "reference-manual/blocks.html#program-block-transformed-data",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/higher-order_functions.html#functions-map",
+    "href": "functions-reference/higher-order_functions.html#functions-map",
+    "title": "Higher-Order Functions",
     "section": "",
-    "text": "The transformed data block is for declaring and defining variables that do not need to be changed when running the program.\n\n\nFor the transformed data block, variables are all declared in the variable declarations and defined in the statements. There is no reading from external sources and no transformations performed.\nVariables declared in the data block may be used to declare transformed variables.\n\n\n\nThe statements in a transformed data block are used to define (provide values for) variables declared in the transformed data block. Assignments are only allowed to variables declared in the transformed data block.\nThese statements are executed once, in order, right after the data is read into the data variables. This means they are executed once per chain.\nVariables declared in the data block may be used in statements in the transformed data block.\n\n\nThe statements in the transformed data block are designed to be executed once and have a deterministic result. Therefore, log probability is not accumulated and distribution statements may not be used.\n\n\n\n\nAny constraints on variables declared in the transformed data block are checked after the statements are executed. If any defined variable violates its constraints, Stan will halt with a diagnostic error message.",
+    "text": "Stan provides a higher-order map function. This allows map-reduce functionality to be coded in Stan as described in the user’s guide.\n\n\nThe function being mapped must have a signature identical to that of the function f in the following declaration.\n vector f(vector phi, vector theta,\n          data array[] real x_r, data array[] int x_i);\nThe map function returns the sequence of results for the particular shard being evaluated. The arguments to the mapped function are:\n\nphi, the sequence of parameters shared across shards\ntheta, the sequence of parameters specific to this shard\nx_r, sequence of real-valued data\nx_i, sequence of integer data\n\nAll input for the mapped function must be packed into these sequences and all output from the mapped function must be packed into a single vector. The vector of output from each mapped function is concatenated into the final result.\n\n\n\nThe rectangular map function operates on rectangular (not ragged) data structures, with parallel data structures for job-specific parameters, job-specific real data, and job-specific integer data.\n \n\nvector map_rect(F f, vector phi, array[] vector theta, data array[,] real x_r, data array[,] int x_i) Return the concatenation of the results of applying the function f, of type (vector, vector, array[] real, array[] int):vector elementwise, i.e., f(phi, theta[n], x_r[n], x_i[n]) for each n in 1:N, where N is the size of the parallel arrays of job-specific/local parameters theta, real data x_r, and integer data x_r. The shared/global parameters phi are passed to each invocation of f.\nAvailable since 2.18",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Higher-Order Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#program-block-parameters",
-    "href": "reference-manual/blocks.html#program-block-parameters",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/integer-valued_basic_functions.html",
+    "href": "functions-reference/integer-valued_basic_functions.html",
+    "title": "Integer-Valued Basic Functions",
     "section": "",
-    "text": "The variables declared in the parameters program block correspond directly to the variables being sampled by Stan’s samplers (HMC and NUTS). From a user’s perspective, the parameters in the program block are the parameters being sampled by Stan.\nVariables declared as parameters cannot be directly assigned values. So there is no block of statements in the parameters program block. Variable quantities derived from parameters may be declared in the transformed parameters or generated quantities blocks, or may be defined as local variables in any statement blocks following their declaration.\nThere is a substantial amount of computation involved for parameter variables in a Stan program at each leapfrog step within the HMC or NUTS samplers, and a bit more computation along with writes involved for saving the parameter values corresponding to a sample.\n\n\nStan’s two samplers, standard Hamiltonian Monte Carlo (HMC) and the adaptive No-U-Turn sampler (NUTS), are most easily (and often most effectively) implemented over a multivariate probability density that has support on all of \\(\\mathbb{R}^n\\). To do this, the parameters defined in the parameters block must be transformed so they are unconstrained.\nIn practice, the samplers keep an unconstrained parameter vector in memory representing the current state of the sampler. The model defined by the compiled Stan program defines an (unnormalized) log probability function over the unconstrained parameters. In order to do this, the log probability function must apply the inverse transform to the unconstrained parameters to calculate the constrained parameters defined in Stan’s parameters program block. The log Jacobian of the inverse transform is then added to the accumulated log probability function. This then allows the Stan model to be defined in terms of the constrained parameters.\nIn some cases, the number of parameters is reduced in the unconstrained space. For instance, a \\(K\\)-simplex only requires \\(K-1\\) unconstrained parameters, and a \\(K\\)-correlation matrix only requires \\(\\binom{K}{2}\\) unconstrained parameters. This means that the probability function defined by the compiled Stan program may have fewer parameters than it would appear from looking at the declarations in the parameters program block.\nThe probability function on the unconstrained parameters is defined in such a way that the order of the parameters in the vector corresponds to the order of the variables defined in the parameters program block. The details of the specific transformations are provided in the variable transforms chapter.\n\n\n\nHamiltonian Monte Carlo requires the gradient of the (unnormalized) log probability function with respect to the unconstrained parameters to be evaluated during every leapfrog step. There may be one leapfrog step per sample or hundreds, with more being required for models with complex posterior distribution geometries.\nGradients are calculated behind the scenes using Stan’s algorithmic differentiation library. The time to compute the gradient does not depend directly on the number of parameters, only on the number of subexpressions in the calculation of the log probability. This includes the expressions added from the transforms’ Jacobians.\nThe amount of work done by the sampler does depend on the number of unconstrained parameters, but this is usually dwarfed by the gradient calculations.\n\n\n\nIn the basic Stan compiled program, there is a file to which the values of variables are written for each draw. The constrained versions of the variables are written in the order they are defined in the parameters block. In order to do this, the transformed parameter, model, and generated quantities statements must also be executed.",
+    "text": "This chapter describes Stan’s built-in function that take various types of arguments and return integer values.\n\n\nStan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in 2 + 2, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in 2.0 + 2 and 2 + 2.0, then the integer is promoted to a floating point value and floating-point arithmetic is used.\nInteger arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.\nSecondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, 1 / 2 evaluates to 0 and 5 / 3 evaluates to 1.\nIf one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have 1 / 2 and -1 / 2 evaluate to 0, -7 / 2 evaluate to -3, and 7 / 2 evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.\nUnlike floating point division, where 1.0 / 0.0 produces the special positive infinite value, integer division by zero, as in 1 / 0, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.\n\n\nOperators are described using the C++ syntax. For instance, the binary operator of addition, written X + Y, would have the Stan signature int operator+(int, int) indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.\n \n\nint operator+(int x, int y) The sum of the addends x and y \\[\\begin{equation*} \\text{operator+}(x,y) = (x + y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator-(int x, int y) The difference between the minuend x and subtrahend y \\[\\begin{equation*}\n\\text{operator-}(x,y) = (x - y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator*(int x, int y) The product of the factors x and y \\[\\begin{equation*} \\text{operator*}(x,y) = (x\n\\times y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator/(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator/}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\] deprecated; - use operator%/% instead.\nAvailable since 2.0, deprecated in 2.24\n \n\nint operator%/%(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator\\%/\\%}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\]\nAvailable since 2.24\n \n\nint operator%(int x, int y) x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. \\[\\begin{equation*} \\mathrm{operator\\%}(x, y) \\ =\n\\ x \\ \\text{mod} \\ y \\ = \\ x - y * \\lfloor x / y \\rfloor \\end{equation*}\\]\nAvailable since 2.13\n\n\n\n \n\nint operator-(int x) The negation of the subtrahend x \\[\\begin{equation*} \\text{operator-}(x) = -x \\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of integers, -x is the same shape array where each individual integer is negated.\nAvailable since 2.31\n \n\nint operator+(int x) This is a no-op. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n\n \n\nT abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is int, or an array of ints, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nint int_step(int x)\n \n\nint int_step(real x) Return the step function of x as an integer, \\[\\begin{equation*} \\mathrm{int\\_step}(x)\n= \\begin{cases} 1 & \\text{if } x &gt; 0 \\\\ 0 & \\text{if } x \\leq 0 \\text{\nor } x \\text{ is } NaN \\end{cases} \\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nSee the warning in section step functions about the dangers of step functions applied to anything other than data.\nAvailable since 2.0\n\n\n\n \n\nint min(int x, int y) Return the minimum of x and y. \\[\\begin{equation*} \\text{min}(x, y) = \\begin{cases} x &\n\\text{if } x &lt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint max(int x, int y) Return the maximum of x and y. \\[\\begin{equation*} \\text{max}(x, y) = \\begin{cases} x &\n\\text{if } x &gt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n\n\n\n \n\nint size(int x)\n \n\nint size(real x)\nReturn the size of x which for scalar-valued x is 1\nAvailable since 2.26\n\n\n\nIt is possible to cast real numbers to integers as long as the real value is data. See data only qualifiers in the Stan Reference Manual.\n \n\nint to_int(data real x)\nReturn the value x truncated to an integer. This will throw an error if the value of x is too big to represent as a 32-bit signed integer.\nThis is similar to trunc (see Rounding functions) but the return type is of type int. For example, to_int(3.9) is 3, and to_int(-3.9) is -3.\nAvailable since 2.31\n \n\nI to_int(data T x)\nThe vectorized version of to_int. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.\nAvailable since 2.31",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#program-block-transformed-parameters",
-    "href": "reference-manual/blocks.html#program-block-transformed-parameters",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/integer-valued_basic_functions.html#int-arithmetic",
+    "href": "functions-reference/integer-valued_basic_functions.html#int-arithmetic",
+    "title": "Integer-Valued Basic Functions",
     "section": "",
-    "text": "The transformed parameters program block consists of optional variable declarations followed by statements. After the statements are executed, the constraints on the transformed parameters are validated. Any variable declared as a transformed parameter is part of the output produced for draws.\nAny variable that is defined wholly in terms of data or transformed data should be declared and defined in the transformed data block. Defining such quantities in the transformed parameters block is legal, but much less efficient than defining them as transformed data.\n\n\nLike the constraints on data, the constraints on transformed parameters is meant to catch programming errors as well as convey programmer intent. They are not automatically transformed in such a way as to be satisfied. What will happen if a transformed parameter does not match its constraint is that the current parameter values will be rejected. This can cause Stan’s algorithms to hang or to devolve to random walks. It is not intended to be a way to enforce ad hoc constraints in Stan programs. See the section on reject statements for further discussion of the behavior of reject statements.",
+    "text": "Stan’s arithmetic is based on standard double-precision C++ integer and floating-point arithmetic. If the arguments to an arithmetic operator are both integers, as in 2 + 2, integer arithmetic is used. If one argument is an integer and the other a floating-point value, as in 2.0 + 2 and 2 + 2.0, then the integer is promoted to a floating point value and floating-point arithmetic is used.\nInteger arithmetic behaves slightly differently than floating point arithmetic. The first difference is how overflow is treated. If the sum or product of two integers overflows the maximum integer representable, the result is an undesirable wraparound behavior at the bit level. If the integers were first promoted to real numbers, they would not overflow a floating-point representation. There are no extra checks in Stan to flag overflows, so it is up to the user to make sure it does not occur.\nSecondly, because the set of integers is not closed under division and there is no special infinite value for integers, integer division implicitly rounds the result. If both arguments are positive, the result is rounded down. For example, 1 / 2 evaluates to 0 and 5 / 3 evaluates to 1.\nIf one of the integer arguments to division is negative, the latest C++ specification ( C++11), requires rounding toward zero. This would have 1 / 2 and -1 / 2 evaluate to 0, -7 / 2 evaluate to -3, and 7 / 2 evaluate to 3. Before the C++11 specification, the behavior was platform dependent, allowing rounding up or down. All compilers recent enough to be able to deal with Stan’s templating should follow the C++11 specification, but it may be worth testing if you are not sure and plan to use integer division with negative values.\nUnlike floating point division, where 1.0 / 0.0 produces the special positive infinite value, integer division by zero, as in 1 / 0, has undefined behavior in the C++ standard. For example, the clang++ compiler on Mac OS X returns 3764, whereas the g++ compiler throws an exception and aborts the program with a warning. As with overflow, it is up to the user to make sure integer divide-by-zero does not occur.\n\n\nOperators are described using the C++ syntax. For instance, the binary operator of addition, written X + Y, would have the Stan signature int operator+(int, int) indicating it takes two real arguments and returns a real value. As noted previously, the value of integer division is platform-dependent when rounding is platform dependent before C++11; the descriptions below provide the C++11 definition.\n \n\nint operator+(int x, int y) The sum of the addends x and y \\[\\begin{equation*} \\text{operator+}(x,y) = (x + y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator-(int x, int y) The difference between the minuend x and subtrahend y \\[\\begin{equation*}\n\\text{operator-}(x,y) = (x - y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator*(int x, int y) The product of the factors x and y \\[\\begin{equation*} \\text{operator*}(x,y) = (x\n\\times y) \\end{equation*}\\]\nAvailable since 2.0\n \n\nint operator/(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator/}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\] deprecated; - use operator%/% instead.\nAvailable since 2.0, deprecated in 2.24\n \n\nint operator%/%(int x, int y) The integer quotient of the dividend x and divisor y \\[\\begin{equation*}\n\\text{operator\\%/\\%}(x,y) = \\begin{cases} \\lfloor x / y \\rfloor & \\text{if\n} x / y \\geq 0 \\\\ - \\lfloor \\text{floor}(-x / y) \\rfloor & \\text{if }\nx / y &lt; 0. \\end{cases} \\end{equation*}\\]\nAvailable since 2.24\n \n\nint operator%(int x, int y) x modulo y, which is the positive remainder after dividing x by y. If both x and y are non-negative, so is the result; otherwise, the sign of the result is platform dependent. \\[\\begin{equation*} \\mathrm{operator\\%}(x, y) \\ =\n\\ x \\ \\text{mod} \\ y \\ = \\ x - y * \\lfloor x / y \\rfloor \\end{equation*}\\]\nAvailable since 2.13\n\n\n\n \n\nint operator-(int x) The negation of the subtrahend x \\[\\begin{equation*} \\text{operator-}(x) = -x \\end{equation*}\\]\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of integers, -x is the same shape array where each individual integer is negated.\nAvailable since 2.31\n \n\nint operator+(int x) This is a no-op. \\[\\begin{equation*} \\text{operator+}(x) = x \\end{equation*}\\]\nAvailable since 2.0",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#program-block-model",
-    "href": "reference-manual/blocks.html#program-block-model",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/integer-valued_basic_functions.html#absolute-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#absolute-functions",
+    "title": "Integer-Valued Basic Functions",
     "section": "",
-    "text": "The model program block consists of optional variable declarations followed by statements. The variables in the model block are local variables and are not written as part of the output.\nLocal variables may not be defined with constraints because there is no well-defined way to have them be both flexible and easy to validate.\nThe statements in the model block typically define the model. This is the block in which probability (distribution notation) statements are allowed. These are typically used when programming in the BUGS idiom to define the probability model.",
+    "text": "T abs(T x) The absolute value of x.\nThis function works elementwise over containers such as vectors. Given a type T which is int, or an array of ints, abs returns the same type where each element has had its absolute value taken.\nAvailable since 2.0, vectorized in 2.30\n \n\nint int_step(int x)\n \n\nint int_step(real x) Return the step function of x as an integer, \\[\\begin{equation*} \\mathrm{int\\_step}(x)\n= \\begin{cases} 1 & \\text{if } x &gt; 0 \\\\ 0 & \\text{if } x \\leq 0 \\text{\nor } x \\text{ is } NaN \\end{cases} \\end{equation*}\\] Warning: int_step(0) and int_step(NaN) return 0 whereas step(0) and step(NaN) return 1.\nSee the warning in section step functions about the dangers of step functions applied to anything other than data.\nAvailable since 2.0",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#program-block-generated-quantities",
-    "href": "reference-manual/blocks.html#program-block-generated-quantities",
-    "title": "Program Blocks",
+    "objectID": "functions-reference/integer-valued_basic_functions.html#bound-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#bound-functions",
+    "title": "Integer-Valued Basic Functions",
     "section": "",
-    "text": "The generated quantities program block is rather different than the other blocks. Nothing in the generated quantities block affects the sampled parameter values. The block is executed only after a sample has been generated.\nAmong the applications of posterior inference that can be coded in the generated quantities block are\n\nforward sampling to generate simulated data for model testing,\ngenerating predictions for new data,\ncalculating posterior event probabilities, including multiple comparisons, sign tests, etc.,\ncalculating posterior expectations,\ntransforming parameters for reporting,\napplying full Bayesian decision theory,\ncalculating log likelihoods, deviances, etc. for model comparison.\n\nParameter estimates, predictions, statistics, and event probabilities calculated directly using plug-in estimates. Stan automatically provides full Bayesian inference by producing draws from the posterior distribution of any calculated event probabilities, predictions, or statistics.\nWithin the generated quantities block, the values of all other variables declared in earlier program blocks (other than local variables) are available for use in the generated quantities block.\nIt is more efficient to define a variable in the generated quantities block instead of the transformed parameters block. Therefore, if a quantity does not play a role in the model, it should be defined in the generated quantities block.\nAfter the generated quantities statements are executed, the constraints on the declared generated quantity variables are validated.\nAll variables declared as generated quantities are printed as part of the output. Variables declared in nested blocks are local variables, not generated quantities, and thus won’t be printed. For example:\ngenerated quantities {\n  int a; // added to the output\n\n  {\n    int b; // not added to the output\n  }\n}",
+    "text": "int min(int x, int y) Return the minimum of x and y. \\[\\begin{equation*} \\text{min}(x, y) = \\begin{cases} x &\n\\text{if } x &lt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0\n \n\nint max(int x, int y) Return the maximum of x and y. \\[\\begin{equation*} \\text{max}(x, y) = \\begin{cases} x &\n\\text{if } x &gt; y\\\\ y & \\text{otherwise} \\end{cases} \\end{equation*}\\]\nAvailable since 2.0",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "reference-manual/blocks.html#footnotes",
-    "href": "reference-manual/blocks.html#footnotes",
-    "title": "Program Blocks",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nIf the C++ code is configured for concurrent threads, the data and transformed data blocks can be executed once and reused for multiple chains.↩︎\nIt is possible to print a variable every iteration that does not depend on parameters—just define it (or redefine it if it is transformed data) in the generated quantities block.↩︎\nWith multiple threads, or even running chains sequentially in a single thread, data could be read only once per set of chains. Stan was designed to be thread safe and future versions will provide a multithreading option for Markov chains.↩︎",
+    "objectID": "functions-reference/integer-valued_basic_functions.html#size-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#size-functions",
+    "title": "Integer-Valued Basic Functions",
+    "section": "",
+    "text": "int size(int x)\n \n\nint size(real x)\nReturn the size of x which for scalar-valued x is 1\nAvailable since 2.26",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Program Blocks"
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "reference-manual/deprecations.html",
-    "href": "reference-manual/deprecations.html",
-    "title": "Deprecated Features",
+    "objectID": "functions-reference/integer-valued_basic_functions.html#casting-functions",
+    "href": "functions-reference/integer-valued_basic_functions.html#casting-functions",
+    "title": "Integer-Valued Basic Functions",
     "section": "",
-    "text": "This appendix lists currently deprecated functionality along with how to replace it.\nStarting with Stan 2.29, minor (syntax-level) deprecations can be removed 3 versions after release; e.g., syntax deprecated in Stan 2.20 will be removed in Stan 2.23 and placed in Removed Features. The Stan compiler can automatically update many of these on the behalf of the user for at least one version after they are removed.\nAny feature which changes semantic meaning (such as the upgraded ODE solver interface) will not be removed until a major version change (e.g., Stan 3.0).\n\n\nDeprecated:The distribution lkj_cov is deprecated.\nReplacement: Replace lkj_cov_lpdf(...) with an lkj_corr distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace\ncov_matrix[K] Sigma;\n// ...\nSigma ~ lkj_cov(mu, tau, eta);\nwith\ncorr_matrix[K] Omega;\nvector&lt;lower=0&gt;[K] sigma;\n// ...\nOmega ~ lkj_corr(eta);\nsigma ~ lognormal(mu, tau);\n// ...\ncov_matrix[K] Sigma;\nSigma &lt;- quad_form_diag(Omega, sigma);\nThe variable Sigma may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.\nScheduled Removal: Stan 3.0 or later.\n\n\n\nDeprecated: Using functions that end in _lp in the transformed parameters block.\nReplacement: Use _jacobian functions and the jacobian += statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.\n\n\n\nSeveral built-in Stan functions have been deprecated. Consult the functions reference for more information.",
+    "text": "It is possible to cast real numbers to integers as long as the real value is data. See data only qualifiers in the Stan Reference Manual.\n \n\nint to_int(data real x)\nReturn the value x truncated to an integer. This will throw an error if the value of x is too big to represent as a 32-bit signed integer.\nThis is similar to trunc (see Rounding functions) but the return type is of type int. For example, to_int(3.9) is 3, and to_int(-3.9) is -3.\nAvailable since 2.31\n \n\nI to_int(data T x)\nThe vectorized version of to_int. This function accepts a (possibly nested) array of reals and returns an array of the same shape where each element has been truncated to an integer.\nAvailable since 2.31",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Deprecated Features"
+      "Functions Reference",
+      "Built-in Functions",
+      "Integer-Valued Basic Functions"
     ]
   },
   {
-    "objectID": "reference-manual/deprecations.html#lkj_cov-distribution",
-    "href": "reference-manual/deprecations.html#lkj_cov-distribution",
-    "title": "Deprecated Features",
+    "objectID": "functions-reference/matrix_operations.html",
+    "href": "functions-reference/matrix_operations.html",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "Deprecated:The distribution lkj_cov is deprecated.\nReplacement: Replace lkj_cov_lpdf(...) with an lkj_corr distribution on the correlation matrix and independent lognormal distributions on the scales. That is, replace\ncov_matrix[K] Sigma;\n// ...\nSigma ~ lkj_cov(mu, tau, eta);\nwith\ncorr_matrix[K] Omega;\nvector&lt;lower=0&gt;[K] sigma;\n// ...\nOmega ~ lkj_corr(eta);\nsigma ~ lognormal(mu, tau);\n// ...\ncov_matrix[K] Sigma;\nSigma &lt;- quad_form_diag(Omega, sigma);\nThe variable Sigma may be defined as a local variable in the model block or as a transformed parameter. An even more efficient transform would use Cholesky factors rather than full correlation matrix types.\nScheduled Removal: Stan 3.0 or later.",
+    "text": "int num_elements(vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.5\n \n\nint num_elements(row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.5\n \n\nint num_elements(matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.5\n \n\nint rows(vector x) The number of rows in the vector x\nAvailable since 2.0\n \n\nint rows(row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.0\n \n\nint rows(matrix x) The number of rows in the matrix x\nAvailable since 2.0\n \n\nint cols(vector x) The number of columns in the vector x, namely 1\nAvailable since 2.0\n \n\nint cols(row_vector x) The number of columns in the row vector x\nAvailable since 2.0\n \n\nint cols(matrix x) The number of columns in the matrix x\nAvailable since 2.0\n \n\nint size(vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15\nAvailable since 2.26\n\n\n\nStan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\nvector operator-(vector x) The negation of the vector x.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x) The negation of the row vector x.\nAvailable since 2.0\n \n\nmatrix operator-(matrix x) The negation of the matrix x.\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\nvector operator+(vector x, vector y) The sum of the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, row_vector y) The sum of the row vectors x and y.\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, matrix y) The sum of the matrices x and y\nAvailable since 2.0\n \n\nvector operator-(vector x, vector y) The difference between the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, row_vector y) The difference between the row vectors x and y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, matrix y) The difference between the matrices x and y\nAvailable since 2.0\n \n\nvector operator*(real x, vector y) The product of the scalar x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(real x, row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.0\n \n\nmatrix operator*(real x, matrix y) The product of the scalar x and the matrix y\nAvailable since 2.0\n \n\nvector operator*(vector x, real y) The product of the scalar y and vector x\nAvailable since 2.0\n \n\nmatrix operator*(vector x, row_vector y) The product of the vector x and row vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, real y) The product of the scalar y and row vector x\nAvailable since 2.0\n \n\nreal operator*(row_vector x, vector y) The product of the row vector x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, matrix y) The product of the row vector x and matrix y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, real y) The product of the scalar y and matrix x\nAvailable since 2.0\n \n\nvector operator*(matrix x, vector y) The product of the matrix x and vector y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, matrix y) The product of the matrices x and y\nAvailable since 2.0\n\n\n\n \n\nvector operator+(vector x, real y) The result of adding y to every entry in the vector x\nAvailable since 2.0\n \n\nvector operator+(real x, vector y) The result of adding x to every entry in the vector y\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, real y) The result of adding y to every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator+(real x, row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, real y) The result of adding y to every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator+(real x, matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.0\n \n\nvector operator-(vector x, real y) The result of subtracting y from every entry in the vector x\nAvailable since 2.0\n \n\nvector operator-(real x, vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, real y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator-(real x, row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, real y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator-(real x, matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.0\n \n\nvector operator/(vector x, real y) The result of dividing each entry in the vector x by y\nAvailable since 2.0\n \n\nrow_vector operator/(row_vector x, real y) The result of dividing each entry in the row vector x by y\nAvailable since 2.0\n \n\nmatrix operator/(matrix x, real y) The result of dividing each entry in the matrix x by y\nAvailable since 2.0\n\n\n\n\nMatrix transposition is represented using a postfix operator.\n \n\nmatrix operator'(matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.0\n \n\nrow_vector operator'(vector x) The transpose of the vector x, written as x'\nAvailable since 2.0\n \n\nvector operator'(row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.0\n\n\n\nElementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section function vectorization for the general cases.\n \n\nvector operator.*(vector x, vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nrow_vector operator.*(row_vector x, row_vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nmatrix operator.*(matrix x, matrix y) The elementwise product of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator./(real x, vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(row_vector x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nrow_vector operator./(row_vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(real x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(matrix x, matrix y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nmatrix operator./(matrix x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(real x, matrix y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator.^(vector x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(real x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(real x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(real x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n\n\n\n \n\nreal dot_product(vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(vector x, vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(row_vector x, row_vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(matrix x, matrix y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(vector x, vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(row_vector x, row_vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(matrix x, matrix y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nreal dot_self(vector x) The dot product of the vector x with itself\nAvailable since 2.0\n \n\nreal dot_self(row_vector x) The dot product of the row vector x with itself\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(matrix x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(matrix x) The dot product of the rows of x with themselves\nAvailable since 2.0\n\n\n \n\nmatrix tcrossprod(matrix x) The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}\\,\\text{x}^{\\top}\\).\nAvailable since 2.0\n \n\nmatrix crossprod(matrix x) The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}^{\\top}\\,\\text{x}\\).\nAvailable since 2.0\nThe following functions all provide shorthand forms for common expressions, which are also much more efficient.\n \n\nmatrix quad_form(matrix A, matrix B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nreal quad_form(matrix A, vector B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nmatrix quad_form_diag(matrix m, vector v) The quadratic form using the column vector v as a diagonal matrix, i.e., diag_matrix(v) * m * diag_matrix(v).\nAvailable since 2.3\n \n\nmatrix quad_form_diag(matrix m, row_vector rv) The quadratic form using the row vector rv as a diagonal matrix, i.e., diag_matrix(rv) * m * diag_matrix(rv).\nAvailable since 2.3\n \n\nmatrix quad_form_sym(matrix A, matrix B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal quad_form_sym(matrix A, vector B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal trace_dot(matrix A, matrix B) The trace of the matrix product, i.e., trace(A * B).\nAvailable since 2.39\n \n\nreal trace_quad_form(matrix A, matrix B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_quad_form(matrix A, vector B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_gen_quad_form(matrix D, matrix A, matrix B) The trace of a generalized quadratic form, i.e., trace(D * B' * A * B).\nAvailable since 2.0\n \n\nmatrix multiply_lower_tri_self_transpose(matrix x) The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if L is a matrix of the same dimensions as x with L(m,n) equal to x(m,n) for \\(\\text{n}\n\\leq \\text{m}\\) and L(m,n) equal to 0 if \\(\\text{n} &gt; \\text{m}\\), the result is the symmetric matrix \\(\\text{L}\\,\\text{L}^{\\top}\\). This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(vector v, matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(row_vector rv, matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, row_vector rv) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.0\n\n\n\n\n\n\n \n\nreal log_sum_exp(vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(row_vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(matrix x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n\n\n\n \n\nreal min(vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(row_vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(matrix x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(row_vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(matrix x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n\n\n\n \n\nreal sum(vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal prod(vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n\n\n\nFull definitions are provided for sample moments in section array reductions.\n \n\nreal mean(vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(row_vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(matrix x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(row_vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(matrix x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(row_vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(matrix x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n \n\nreal quantile(data row_vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data row_vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n\n\n\n\nThe following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\nvector rep_vector(real x, int m) Return the size m (column) vector consisting of copies of x.\nAvailable since 2.0\n \n\nrow_vector rep_row_vector(real x, int n) Return the size n row vector consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(real x, int m, int n) Return the m by n matrix consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.0\n \n\nmatrix rep_matrix(row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.0\nUnlike the situation with array broadcasting (see section array broadcasting), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.\n vector[3] x;\n x = rep_vector(1, 3);\n x = rep_vector(1.0, 3);\nThere are no integer vector or matrix types, so integer values are automatically promoted.\n\n\n \n\nmatrix symmetrize_from_lower_tri(matrix A)\nConstruct a symmetric matrix from the lower triangle of A.\nAvailable since 2.26\n\n\n\n\n \n\nmatrix add_diag(matrix m, row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, real d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.21\n \n\nvector diagonal(matrix x) The diagonal of the matrix x\nAvailable since 2.0\n \n\nmatrix diag_matrix(vector x) The diagonal matrix with diagonal x\nAvailable since 2.0\nAlthough the diag_matrix function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,\n y ~ multi_normal(mu, diag_matrix(square(sigma)));\nit is much more efficient to just use a univariate normal, which produces the same density,\n y ~ normal(mu, sigma);\nRather than writing m * diag_matrix(v) where m is a matrix and v is a vector, it is much more efficient to write diag_post_multiply(m, v) (and similarly for pre-multiplication). By the same token, it is better to use quad_form_diag(m, v) rather than quad_form(m, diag_matrix(v)).\n \n\nmatrix identity_matrix(int k) Create an identity matrix of size \\(k \\times k\\)\nAvailable since 2.26\n\n\n\n \n\narray[] real linspaced_array(int n, data real lower, data real upper) Create a real array of length n of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int linspaced_int_array(int n, int lower, int upper) Create a regularly spaced, increasing integer array of length n between lower and upper, inclusively. If (upper - lower) / (n - 1) is less than one, repeat each output (n - 1) / (upper - lower) times. If neither (upper - lower) / (n - 1) or (n - 1) / (upper - lower) are integers, upper is reduced until one of these is true.\nAvailable since 2.26\n \n\nvector linspaced_vector(int n, data real lower, data real upper) Create an n-dimensional vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\nrow_vector linspaced_row_vector(int n, data real lower, data real upper) Create an n-dimensional row-vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int one_hot_int_array(int n, int k) Create a one-hot encoded int array of length n with array[k] = 1\nAvailable since 2.26\n \n\narray[] real one_hot_array(int n, int k) Create a one-hot encoded real array of length n with array[k] = 1\nAvailable since 2.24\n \n\nvector one_hot_vector(int n, int k) Create an n-dimensional one-hot encoded vector with vector[k] = 1\nAvailable since 2.24\n \n\nrow_vector one_hot_row_vector(int n, int k) Create an n-dimensional one-hot encoded row-vector with row_vector[k] = 1\nAvailable since 2.24\n \n\narray[] int ones_int_array(int n) Create an int array of length n of all ones\nAvailable since 2.26\n \n\narray[] real ones_array(int n) Create a real array of length n of all ones\nAvailable since 2.26\n \n\nvector ones_vector(int n) Create an n-dimensional vector of all ones\nAvailable since 2.26\n \n\nrow_vector ones_row_vector(int n) Create an n-dimensional row-vector of all ones\nAvailable since 2.26\n \n\narray[] int zeros_int_array(int n) Create an int array of length n of all zeros\nAvailable since 2.26\n \n\narray[] real zeros_array(int n) Create a real array of length n of all zeros\nAvailable since 2.24\n \n\nvector zeros_vector(int n) Create an n-dimensional vector of all zeros\nAvailable since 2.24\n \n\nrow_vector zeros_row_vector(int n) Create an n-dimensional row-vector of all zeros\nAvailable since 2.24\n \n\nvector uniform_simplex(int n) Create an n-dimensional simplex with elements vector[i] = 1 / n for all \\(i \\in 1, \\dots, n\\)\nAvailable since 2.24\n\n\n\nStan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\nvector col(matrix x, int n) The n-th column of matrix x\nAvailable since 2.0\n \n\nrow_vector row(matrix x, int m) The m-th row of matrix x\nAvailable since 2.0\nThe row function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation x[m] is just an alternative way of writing row(x,m). The col function may not, be used as an lvalue, nor is there an indexing based shorthand for it.\n\n\n\n\n\nBlock operations may be used to extract a sub-block of a matrix.\n \n\nmatrix block(matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.0\nThe sub-row and sub-column operations may be used to extract a slice of row or column from a matrix\n \n\nvector sub_col(matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.0\n \n\nrow_vector sub_row(matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.0\n\n\n\nThe head operation extracts the first \\(n\\) elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.\n \n\nvector head(vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.0\n \n\nrow_vector head(row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.0\n \n\narray[] T head(array[] T sv, int n) Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector tail(vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.0\n \n\nrow_vector tail(row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.0\n \n\narray[] T tail(array[] T sv, int n) Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector segment(vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.0\n \n\nrow_vector segment(row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.10\n \n\narray[] T segment(array[] T sv, int i, int n) Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n\n\n\n\n\nStan’s matrix and vector concatenation operations append_col and append_row are like the operations cbind and rbind in R.\n\n\n \n\nmatrix append_col(matrix x, matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(matrix x, vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.5\n \n\nrow_vector append_col(row_vector x, row_vector y) Combine row vectors x and y of any size into another row vector by appending y to the end of x.\nAvailable since 2.5\n \n\nrow_vector append_col(real x, row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.12\n \n\nrow_vector append_col(row_vector x, real y) Append y to the end of x, returning another row vector.\nAvailable since 2.12\n\n\n\n \n\nmatrix append_row(matrix x, matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(matrix x, row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.5\n \n\nvector append_row(vector x, vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.5\n \n\nvector append_row(real x, vector y) Append x to the top of y, returning another vector.\nAvailable since 2.12\n \n\nvector append_row(vector x, real y) Append y to the bottom of x, returning another vector.\nAvailable since 2.12\n\n\n\n\n\n\nThe softmax function maps1 \\(y \\in \\mathbb{R}^K\\) to the \\(K\\)-simplex by \\[\\begin{equation*} \\text{softmax}(y)  = \\frac{\\exp(y)}\n{\\sum_{k=1}^K \\exp(y_k)}, \\end{equation*}\\] where \\(\\exp(y)\\) is the componentwise exponentiation of \\(y\\). Softmax is usually calculated on the log scale, \\[\\begin{eqnarray*} \\log \\text{softmax}(y) & = & \\ y - \\log \\sum_{k=1}^K\n\\exp(y_k) \\\\[4pt] & = & y - \\mathrm{log\\_sum\\_exp}(y). \\end{eqnarray*}\\] where the vector \\(y\\) minus the scalar \\(\\mathrm{log\\_sum\\_exp}(y)\\) subtracts the scalar from each component of \\(y\\).\nStan provides the following functions for softmax and its log.\n \n\nvector softmax(vector x) The softmax of x\nAvailable since 2.0\n \n\nvector log_softmax(vector x) The natural logarithm of the softmax of x\nAvailable since 2.0\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] int cumulative_sum(array[] int x) The cumulative sum of x\nAvailable since 2.30\n \n\narray[] real cumulative_sum(array[] real x) The cumulative sum of x\nAvailable since 2.0\n \n\nvector cumulative_sum(vector v) The cumulative sum of v\nAvailable since 2.0\n \n\nrow_vector cumulative_sum(row_vector rv) The cumulative sum of rv\nAvailable since 2.0\n\n\n\n\nThe Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.\nFor one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by \\(K_{ij} = k(x_i, x_j)\\) (where \\(x_i\\) is the \\(i^{th}\\) element of the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(x_i, y_j)\\).\nFor multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{x}_j)\\) (where \\(\\mathbf{x}_i\\) is the \\(i^{th}\\) vector in the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{y}_j)\\).\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|^2}{2l^2} \\right)\n\\]\n \n\nmatrix gp_exp_quad_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith bias \\(\\sigma_0\\) the dot product kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma_0^2 + \\mathbf{x}_i^T \\mathbf{x}_j\n\\]\n \n\nmatrix gp_dot_prod_cov(array[] real x, real sigma)\nGaussian process covariance with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(array[] real x1, array[] real x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x, real sigma)\nGaussian process covariance with dot product kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x1, vectors x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in multiple dimensions.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponential kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_exponential_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 3/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right) \\exp \\left( -\\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern32_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 5/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{5}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} + \\frac{5 |\\mathbf{x}_i - \\mathbf{x}_j|^2}{3l^2} \\right)\n\\exp \\left( -\\frac{\\sqrt{5} |\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern52_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\), length scale \\(l\\), and period \\(p\\), the periodic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left(-\\frac{2 \\sin^2 \\left( \\pi \\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{p} \\right) }{l^2} \\right)\n\\]\n \n\nmatrix gp_periodic_cov(array[] real x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(array[] real x1, array[] real x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x1, vectors x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\n\n\n\nIn general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.\n\n\n \n\nrow_vector operator/(row_vector b, matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.0\n \n\nmatrix operator/(matrix B, matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.5\n \n\nvector operator\\(matrix A, vector b) The left division of A by b; equivalently inverse(A) * b\nAvailable since 2.18\n \n\nmatrix operator\\(matrix A, matrix B) The left division of A by B; equivalently inverse(A) * B\nAvailable since 2.18\n\n\n\nThere are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix \\(\\text{tri}(A)\\) is used in the definitions and defined by \\[\\begin{equation*} \\text{tri}(A)[m,n] = \\left\\{\n\\begin{array}{ll} A[m,n] & \\text{if } m \\geq n, \\text{ and} \\\\[4pt] 0\n& \\text{otherwise}. \\end{array} \\right. \\end{equation*}\\] When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.\n \n\nvector mdivide_left_tri_low(matrix A, vector b) The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * b, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_left_tri_low(matrix A, matrix B) The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * B, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n \n\nrow_vector mdivide_right_tri_low(row_vector b, matrix A) The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form b * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_right_tri_low(matrix B, matrix A) The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form B * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n\n\n\n\nThere are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.\n \n\nmatrix mdivide_left_spd(matrix A, vector b) The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * b.\nAvailable since 2.12\n \n\nvector mdivide_left_spd(matrix A, matrix B) The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * B.\nAvailable since 2.12\n \n\nrow_vector mdivide_right_spd(row_vector b, matrix A) The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form b *inverse(A).\nAvailable since 2.12\n \n\nmatrix mdivide_right_spd(matrix B, matrix A) The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form B * inverse(A).\nAvailable since 2.12\n\n\n\nThe exponential of the matrix \\(A\\) is formally defined by the convergent power series: \\[\\begin{equation*} e^A = \\sum_{n=0}^{\\infty} \\dfrac{A^n}{n!}\n\\end{equation*}\\]\n \n\nmatrix matrix_exp(matrix A) The matrix exponential of A\nAvailable since 2.13\n \n\nmatrix matrix_exp_multiply(matrix A, matrix B) The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form matrix_exp(A) * B.\nAvailable since 2.18\n \n\nmatrix scale_matrix_exp_multiply(real t, matrix A, matrix B) The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form matrix_exp(t * A) * B.\nAvailable since 2.18\n\n\n\nReturns the nth power of the specific matrix: \\[\\begin{equation*} M^n = M_1 * ... * M_n \\end{equation*}\\]\n \n\nmatrix matrix_power(matrix A, int B) Matrix A raised to the power B.\nAvailable since 2.24\n\n\n\n\n\n \n\nreal trace(matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.0\n\n\n\n \n\nreal determinant(matrix A) The determinant of A\nAvailable since 2.0\n \n\nreal log_determinant(matrix A) The log of the absolute value of the determinant of A\nAvailable since 2.0\nreal log_determinant_spd(matrix A) The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.\nAvailable since 2.30\n\n\n\nIt is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix m and post-multiplying by a vector or matrix a, as in inverse(m) * a, it is better to code this using matrix division, as in m \\ a. The pre-multiplication case is similar, with b * inverse(m) being more efficiently coded as as b / m. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.\nWarning: The function inv(m) is the elementwise inverse function, which returns 1 / m[i, j] for each element.\n \n\nmatrix inverse(matrix A) Compute the inverse of A\nAvailable since 2.0\n \n\nmatrix inverse_spd(matrix A) Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.\nAvailable since 2.0\n \n\nmatrix chol2inv(matrix L) Compute the inverse of the matrix whose cholesky factorization is L. That is, for \\(A = L L^T\\), return \\(A^{-1}\\).\nAvailable since 2.26\n\n\n\nThe generalized inverse \\(M^+\\) of a matrix \\(M\\) is a matrix that satisfies \\(M M^+ M = M\\). For an invertible, square matrix \\(M\\), \\(M^+\\) is equivalent to \\(M^{-1}\\). The dimensions of \\(M^+\\) are equivalent to the dimensions of \\(M^T\\). The generalized inverse exists for any matrix, so the \\(M\\) may be singular or less than full rank.\nEven though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank (Golub and Pereyra 1973), meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix \\(A\\) as a function of \\(\\epsilon\\):\n\\[\nA = \\left(\n    \\begin{array}{cccc}\n    1 + \\epsilon & 2 & 1 \\\\\n    2 & 4 & 2\n    \\end{array}\n    \\right)\n\\]\nWhen \\(\\epsilon = 0\\), \\(A\\) is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If \\(\\epsilon \\neq 0\\), the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at \\(\\epsilon = 0\\), and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.\n \n\nmatrix generalized_inverse(matrix A) The generalized inverse of A\nAvailable since 2.26\n\n\n\n \n\ncomplex_vector eigenvalues(matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors(matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nvector eigenvalues_sym(matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.0\n \n\nmatrix eigenvectors_sym(matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.0\n \n\ntuple(matrix, vector) eigendecompose_sym(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\n \n\nmatrix qr_thin_Q(matrix A) The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A\nAvailable since 2.18\n \n\nmatrix qr_thin_R(matrix A) The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A\nAvailable since 2.18\n \n\ntuple(matrix, matrix) qr_thin(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_thin_Q(A), qr_thin_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nmatrix qr_Q(matrix A) The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A\nAvailable since 2.3\n \n\nmatrix qr_R(matrix A) The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A\nAvailable since 2.3\n \n tuple(matrix, matrix) qr(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthogonal matrix in the fat QR decomposition and the second element (“R”) is upper trapezoidal. This function is equivalent to (qr_Q(A), qr_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nThe thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as \\[\\begin{equation*} A = Q \\, R. \\end{equation*}\\] Multiplying a column of an orthogonal matrix by \\(-1\\) still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by \\(-1\\) without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.\n\n\n\nEvery symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If \\(\\Sigma\\) is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector \\(L\\) such that \\[\\begin{equation*} \\Sigma = L \\, L^{\\top}. \\end{equation*}\\]\n \n\nmatrix cholesky_decompose(matrix A) The lower-triangular Cholesky factor of the symmetric positive-definite matrix A\nAvailable since 2.0\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(matrix A) The singular values of A in descending order\nAvailable since 2.0\n \n\nmatrix svd_U(matrix A) The left-singular vectors of A\nAvailable since 2.26\n \n\nmatrix svd_V(matrix A) The right-singular vectors of A\nAvailable since 2.26\n \n\ntuple(matrix, vector, matrix) svd(matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33\n\n\n\n\n\nSee the sorting functions section for examples of how the functions work.\n \n\nvector sort_asc(vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nrow_vector sort_asc(row_vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nvector sort_desc(vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\nrow_vector sort_desc(row_vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(row_vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n\n\n\n \n\nvector reverse(vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.23\n \n\nrow_vector reverse(row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.23",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Deprecated Features"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/deprecations.html#use-of-_lp-functions-in-transformed-parameters",
-    "href": "reference-manual/deprecations.html#use-of-_lp-functions-in-transformed-parameters",
-    "title": "Deprecated Features",
+    "objectID": "functions-reference/matrix_operations.html#integer-valued-matrix-size-functions",
+    "href": "functions-reference/matrix_operations.html#integer-valued-matrix-size-functions",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "Deprecated: Using functions that end in _lp in the transformed parameters block.\nReplacement: Use _jacobian functions and the jacobian += statement instead. These allow for change-of-variable adjustments which can be conditionally enabled by Stan’s algorithms.",
+    "text": "int num_elements(vector x) The total number of elements in the vector x (same as function rows)\nAvailable since 2.5\n \n\nint num_elements(row_vector x) The total number of elements in the vector x (same as function cols)\nAvailable since 2.5\n \n\nint num_elements(matrix x) The total number of elements in the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then num_elements(x) is 15\nAvailable since 2.5\n \n\nint rows(vector x) The number of rows in the vector x\nAvailable since 2.0\n \n\nint rows(row_vector x) The number of rows in the row vector x, namely 1\nAvailable since 2.0\n \n\nint rows(matrix x) The number of rows in the matrix x\nAvailable since 2.0\n \n\nint cols(vector x) The number of columns in the vector x, namely 1\nAvailable since 2.0\n \n\nint cols(row_vector x) The number of columns in the row vector x\nAvailable since 2.0\n \n\nint cols(matrix x) The number of columns in the matrix x\nAvailable since 2.0\n \n\nint size(vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(row_vector x) The size of x, i.e., the number of elements\nAvailable since 2.26\n \n\nint size(matrix x) The size of the matrix x. For example, if x is a \\(5 \\times 3\\) matrix, then size(x) is 15\nAvailable since 2.26",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Deprecated Features"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/deprecations.html#deprecated-functions",
-    "href": "reference-manual/deprecations.html#deprecated-functions",
-    "title": "Deprecated Features",
+    "objectID": "functions-reference/matrix_operations.html#matrix-arithmetic-operators",
+    "href": "functions-reference/matrix_operations.html#matrix-arithmetic-operators",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "Several built-in Stan functions have been deprecated. Consult the functions reference for more information.",
+    "text": "Stan supports the basic matrix operations using infix, prefix and postfix operations. This section lists the operations supported by Stan along with their argument and result types.\n\n\n \n\nvector operator-(vector x) The negation of the vector x.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x) The negation of the row vector x.\nAvailable since 2.0\n \n\nmatrix operator-(matrix x) The negation of the matrix x.\nAvailable since 2.0\n \n\nT operator-(T x) Vectorized version of operator-. If T x is a (possibly nested) array of matrix types, -x is the same shape array where each individual value is negated.\nAvailable since 2.31\n\n\n\n \n\nvector operator+(vector x, vector y) The sum of the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, row_vector y) The sum of the row vectors x and y.\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, matrix y) The sum of the matrices x and y\nAvailable since 2.0\n \n\nvector operator-(vector x, vector y) The difference between the vectors x and y.\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, row_vector y) The difference between the row vectors x and y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, matrix y) The difference between the matrices x and y\nAvailable since 2.0\n \n\nvector operator*(real x, vector y) The product of the scalar x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(real x, row_vector y) The product of the scalar x and the row vector y\nAvailable since 2.0\n \n\nmatrix operator*(real x, matrix y) The product of the scalar x and the matrix y\nAvailable since 2.0\n \n\nvector operator*(vector x, real y) The product of the scalar y and vector x\nAvailable since 2.0\n \n\nmatrix operator*(vector x, row_vector y) The product of the vector x and row vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, real y) The product of the scalar y and row vector x\nAvailable since 2.0\n \n\nreal operator*(row_vector x, vector y) The product of the row vector x and vector y\nAvailable since 2.0\n \n\nrow_vector operator*(row_vector x, matrix y) The product of the row vector x and matrix y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, real y) The product of the scalar y and matrix x\nAvailable since 2.0\n \n\nvector operator*(matrix x, vector y) The product of the matrix x and vector y\nAvailable since 2.0\n \n\nmatrix operator*(matrix x, matrix y) The product of the matrices x and y\nAvailable since 2.0\n\n\n\n \n\nvector operator+(vector x, real y) The result of adding y to every entry in the vector x\nAvailable since 2.0\n \n\nvector operator+(real x, vector y) The result of adding x to every entry in the vector y\nAvailable since 2.0\n \n\nrow_vector operator+(row_vector x, real y) The result of adding y to every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator+(real x, row_vector y) The result of adding x to every entry in the row vector y\nAvailable since 2.0\n \n\nmatrix operator+(matrix x, real y) The result of adding y to every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator+(real x, matrix y) The result of adding x to every entry in the matrix y\nAvailable since 2.0\n \n\nvector operator-(vector x, real y) The result of subtracting y from every entry in the vector x\nAvailable since 2.0\n \n\nvector operator-(real x, vector y) The result of adding x to every entry in the negation of the vector y\nAvailable since 2.0\n \n\nrow_vector operator-(row_vector x, real y) The result of subtracting y from every entry in the row vector x\nAvailable since 2.0\n \n\nrow_vector operator-(real x, row_vector y) The result of adding x to every entry in the negation of the row vector y\nAvailable since 2.0\n \n\nmatrix operator-(matrix x, real y) The result of subtracting y from every entry in the matrix x\nAvailable since 2.0\n \n\nmatrix operator-(real x, matrix y) The result of adding x to every entry in negation of the matrix y\nAvailable since 2.0\n \n\nvector operator/(vector x, real y) The result of dividing each entry in the vector x by y\nAvailable since 2.0\n \n\nrow_vector operator/(row_vector x, real y) The result of dividing each entry in the row vector x by y\nAvailable since 2.0\n \n\nmatrix operator/(matrix x, real y) The result of dividing each entry in the matrix x by y\nAvailable since 2.0",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Deprecated Features"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/encoding.html",
-    "href": "reference-manual/encoding.html",
-    "title": "Character Encoding",
+    "objectID": "functions-reference/matrix_operations.html#transposition-operator",
+    "href": "functions-reference/matrix_operations.html#transposition-operator",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.\n\n\nThe UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.\n\n\n\n\nAny bytes on a line after a line-comment sequence (// or #) are ignored up until the ASCII newline character (\\n). They may thus be written in any character encoding which is convenient.\nAny content after a block comment open sequence in ASCII (/*) up to the closing block comment (*/) is ignored, and thus may also be written in whatever character set is convenient.\n\n\n\nThe raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A print, reject, or fatal_error statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (\") when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.\nASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.",
+    "text": "Matrix transposition is represented using a postfix operator.\n \n\nmatrix operator'(matrix x) The transpose of the matrix x, written as x'\nAvailable since 2.0\n \n\nrow_vector operator'(vector x) The transpose of the vector x, written as x'\nAvailable since 2.0\n \n\nvector operator'(row_vector x) The transpose of the row vector x, written as x'\nAvailable since 2.0",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Character Encoding"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/encoding.html#content-characters",
-    "href": "reference-manual/encoding.html#content-characters",
-    "title": "Character Encoding",
+    "objectID": "functions-reference/matrix_operations.html#elementwise-functions",
+    "href": "functions-reference/matrix_operations.html#elementwise-functions",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "The content of a Stan program must be coded in ASCII. All identifiers must consist of only ASCII alpha-numeric characters and the underscore character. All arithmetic operators and punctuation must be coded in ASCII.\n\n\nThe UTF-8 encoding of Unicode and the Latin-1 (ISO-8859-1) encoding share the first 128 code points with ASCII and thus cannot be distinguished from ASCII. That means you can set editors, etc., to use UTF-8 or Latin-1 (or the other Latin-n variants) without worrying that the content of a Stan program will be destroyed.",
+    "text": "Elementwise functions apply a function to each element of a vector or matrix, returning a result of the same shape as the argument. There are many functions that are vectorized in addition to the ad hoc cases listed in this section; see section function vectorization for the general cases.\n \n\nvector operator.*(vector x, vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nrow_vector operator.*(row_vector x, row_vector y) The elementwise product of y and x\nAvailable since 2.0\n \n\nmatrix operator.*(matrix x, matrix y) The elementwise product of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nvector operator./(vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator./(real x, vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(row_vector x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nrow_vector operator./(row_vector x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nrow_vector operator./(real x, row_vector y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(matrix x, matrix y) The elementwise quotient of y and x\nAvailable since 2.0\n \n\nmatrix operator./(matrix x, real y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nmatrix operator./(real x, matrix y) The elementwise quotient of y and x\nAvailable since 2.4\n \n\nvector operator.^(vector x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nvector operator.^(real x, vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(row_vector x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nrow_vector operator.^(real x, row_vector y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, matrix y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(matrix x, real y) The elementwise power of y and x\nAvailable since 2.24\n \n\nmatrix operator.^(real x, matrix y) The elementwise power of y and x\nAvailable since 2.24",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Character Encoding"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/encoding.html#comment-characters",
-    "href": "reference-manual/encoding.html#comment-characters",
-    "title": "Character Encoding",
+    "objectID": "functions-reference/matrix_operations.html#dot-products-and-specialized-products",
+    "href": "functions-reference/matrix_operations.html#dot-products-and-specialized-products",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "Any bytes on a line after a line-comment sequence (// or #) are ignored up until the ASCII newline character (\\n). They may thus be written in any character encoding which is convenient.\nAny content after a block comment open sequence in ASCII (/*) up to the closing block comment (*/) is ignored, and thus may also be written in whatever character set is convenient.",
+    "text": "real dot_product(vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, vector y) The dot product of x and y\nAvailable since 2.0\n \n\nreal dot_product(row_vector x, row_vector y) The dot product of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(vector x, vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(row_vector x, row_vector y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nrow_vector columns_dot_product(matrix x, matrix y) The dot product of the columns of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(vector x, vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(row_vector x, row_vector y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nvector rows_dot_product(matrix x, matrix y) The dot product of the rows of x and y\nAvailable since 2.0\n \n\nreal dot_self(vector x) The dot product of the vector x with itself\nAvailable since 2.0\n \n\nreal dot_self(row_vector x) The dot product of the row vector x with itself\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(row_vector x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nrow_vector columns_dot_self(matrix x) The dot product of the columns of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(row_vector x) The dot product of the rows of x with themselves\nAvailable since 2.0\n \n\nvector rows_dot_self(matrix x) The dot product of the rows of x with themselves\nAvailable since 2.0\n\n\n \n\nmatrix tcrossprod(matrix x) The product of x postmultiplied by its own transpose, similar to the tcrossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}\\,\\text{x}^{\\top}\\).\nAvailable since 2.0\n \n\nmatrix crossprod(matrix x) The product of x premultiplied by its own transpose, similar to the crossprod(x) function in R. The result is a symmetric matrix \\(\\text{x}^{\\top}\\,\\text{x}\\).\nAvailable since 2.0\nThe following functions all provide shorthand forms for common expressions, which are also much more efficient.\n \n\nmatrix quad_form(matrix A, matrix B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nreal quad_form(matrix A, vector B) The quadratic form, i.e., B' * A * B.\nAvailable since 2.0\n \n\nmatrix quad_form_diag(matrix m, vector v) The quadratic form using the column vector v as a diagonal matrix, i.e., diag_matrix(v) * m * diag_matrix(v).\nAvailable since 2.3\n \n\nmatrix quad_form_diag(matrix m, row_vector rv) The quadratic form using the row vector rv as a diagonal matrix, i.e., diag_matrix(rv) * m * diag_matrix(rv).\nAvailable since 2.3\n \n\nmatrix quad_form_sym(matrix A, matrix B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal quad_form_sym(matrix A, vector B) Similarly to quad_form, gives B' * A * B, but additionally checks if A is symmetric and ensures that the result is also symmetric.\nAvailable since 2.3\n \n\nreal trace_dot(matrix A, matrix B) The trace of the matrix product, i.e., trace(A * B).\nAvailable since 2.39\n \n\nreal trace_quad_form(matrix A, matrix B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_quad_form(matrix A, vector B) The trace of the quadratic form, i.e., trace(B' * A * B).\nAvailable since 2.0\n \n\nreal trace_gen_quad_form(matrix D, matrix A, matrix B) The trace of a generalized quadratic form, i.e., trace(D * B' * A * B).\nAvailable since 2.0\n \n\nmatrix multiply_lower_tri_self_transpose(matrix x) The product of the lower triangular portion of x (including the diagonal) times its own transpose; that is, if L is a matrix of the same dimensions as x with L(m,n) equal to x(m,n) for \\(\\text{n}\n\\leq \\text{m}\\) and L(m,n) equal to 0 if \\(\\text{n} &gt; \\text{m}\\), the result is the symmetric matrix \\(\\text{L}\\,\\text{L}^{\\top}\\). This is a specialization of tcrossprod(x) for lower-triangular matrices. The input matrix does not need to be square.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(vector v, matrix m) Return the product of the diagonal matrix formed from the vector v and the matrix m, i.e., diag_matrix(v) * m.\nAvailable since 2.0\n \n\nmatrix diag_pre_multiply(row_vector rv, matrix m) Return the product of the diagonal matrix formed from the vector rv and the matrix m, i.e., diag_matrix(rv) * m.\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, vector v) Return the product of the matrix m and the diagonal matrix formed from the vector v, i.e., m * diag_matrix(v).\nAvailable since 2.0\n \n\nmatrix diag_post_multiply(matrix m, row_vector rv) Return the product of the matrix m and the diagonal matrix formed from the the row vector rv, i.e., m * diag_matrix(rv).\nAvailable since 2.0",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Character Encoding"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/encoding.html#string-literals",
-    "href": "reference-manual/encoding.html#string-literals",
-    "title": "Character Encoding",
+    "objectID": "functions-reference/matrix_operations.html#reductions",
+    "href": "functions-reference/matrix_operations.html#reductions",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "The raw byte sequence within a string literal is escaped according to the C++ standard. In particular, this means that UTF-8 encoded strings are supported, however they are not tested for invalid byte sequences. A print, reject, or fatal_error statement should properly display Unicode characters if your terminal supports the encoding used in the input. In other words, Stan simply preserves any string of bytes between two double quotes (\") when passing to C++. On compliant terminals, this allows the use of glyphs and other characters from encodings such as UTF-8 that fall outside the ASCII-compatible range.\nASCII is the recommended encoding for maximum portability, because it encodes the ASCII characters (Unicode code points 0–127) using the same sequence of bytes as the UTF-8 encoding of Unicode and common ISO-8859 extensions of Latin.",
+    "text": "real log_sum_exp(vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(row_vector x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n \n\nreal log_sum_exp(matrix x) The natural logarithm of the sum of the exponentials of the elements in x\nAvailable since 2.0\n\n\n\n \n\nreal min(vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(row_vector x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal min(matrix x) The minimum value in x, or \\(+\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(row_vector x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n \n\nreal max(matrix x) The maximum value in x, or \\(-\\infty\\) if x is empty\nAvailable since 2.0\n\n\n\n \n\nreal sum(vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(row_vector x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal sum(matrix x) The sum of the values in x, or 0 if x is empty\nAvailable since 2.0\n \n\nreal prod(vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(row_vector x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n \n\nreal prod(matrix x) The product of the values in x, or 1 if x is empty\nAvailable since 2.0\n\n\n\nFull definitions are provided for sample moments in section array reductions.\n \n\nreal mean(vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(row_vector x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal mean(matrix x) The sample mean of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(row_vector x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal variance(matrix x) The sample variance of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(row_vector x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n \n\nreal sd(matrix x) The sample standard deviation of the values in x; see section array reductions for details.\nAvailable since 2.0\n\n\n\nProduces sample quantiles corresponding to the given probabilities. The smallest observation corresponds to a probability of 0 and the largest to a probability of 1.\nImplements algorithm 7 from Hyndman, R. J. and Fan, Y., Sample quantiles in Statistical Packages (R’s default quantile function).\n \n\nreal quantile(data vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27\n \n\nreal quantile(data row_vector x, data real p) The p-th quantile of x\nAvailable since 2.27\n \n\narray[] real quantile(data row_vector x, data array[] real p) An array containing the quantiles of x given by the array of probabilities p\nAvailable since 2.27",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Character Encoding"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html",
-    "href": "reference-manual/expressions.html",
-    "title": "Expressions",
+    "objectID": "functions-reference/matrix_operations.html#matrix-broadcast",
+    "href": "functions-reference/matrix_operations.html#matrix-broadcast",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "An expression is the syntactic unit in a Stan program that denotes a value. Every expression in a well-formed Stan program has a type that is determined statically (at compile time), based only on the type of its variables and the types of the functions used in it. If an expressions type cannot be determined statically, the Stan compiler will report the location of the problem.\nThis chapter covers the syntax, typing, and usage of the various forms of expressions in Stan.\n\n\nThe simplest form of expression is a literal that denotes a primitive numerical value.\n\n\nInteger literals represent integers of type int. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression --1 is interpreted as the negation of the literal -1.)\nThe following list contains well-formed integer literals.\n0, 1, -1, 256, -127098, 24567898765\nInteger literals must have values that fall within the bounds for integer values (see the section on numerical data types).\nInteger literals may not contain decimal points (.). Thus the expressions 1. and 1.0 are of type real and may not be used where a value of type int is required.\n\n\n\nA number written with a period or with scientific notation is assigned to a the continuous numeric type real. Real literals are written in base 10 with a period (.) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.\n0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.\nThe notation e or E followed by a positive or negative integer denotes a power of 10 to multiply. For instance, 2.7e3 and 2.7e+3 denote \\(2.7 \\times 10^3\\), whereas -2E-5 denotes \\(-2 \\times\n10^{-5}\\).\n\n\n\nA number followed by the character i denotes an imaginary number and is assigned to the numeric type complex. The number preceding i may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.\n1i, 2i, -325.786i, 1e10i, 2.87e-10i.\nNote that the character i by itself is not a well-formed imaginary literal. The unit imaginary number must be written as 1i.\n\n\n\nStan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.\n1 + 2i, -3.2e9 + 1e10i\nThese will be assigned the type complex, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.\n\n\n\n\nA variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (_) character. Variables must start with a letter (a--z and A--Z) and may not end with two underscores (__).\nExamples of legal variable identifiers are as follows.\na, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable\nUnlike in R and BUGS, variable identifiers in Stan may not contain a period character.\n\n\nStan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the stanc translator halting with an error message indicating which reserved name was used and its location in the model code.\n\n\nThe name of the model cannot be used as a variable within the model. This is usually not a problem because the default in bin/stanc is to append _model to the name of the file containing the model specification. For example, if the model is in file foo.stan, it would not be legal to have a variable named foo_model when using the default model name through bin/stanc. With user-specified model names, variables cannot match the model.\n\n\n\nThe following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.\nfor, in, while, repeat, until, if, then, else,\ntrue, false, target, struct, typedef, export,\nauto, extern, var, static, lower, upper, offset,\nmultiplier\nVariables should not be named after types, either, and thus may not be any of the following.\nint, real, complex, vector, simplex, unit_vector,\nsum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,\nrow_vector, matrix, cholesky_factor_corr,\ncolumn_stochastic_matrix,row_stochastic_matrix,\ncholesky_factor_cov, corr_matrix, cov_matrix, array\nThe following keywords or built-in functions are also reserved and cannot be used as variable names:\nprint, reject, profile, fatal_error, target, jacobian\nThe following block identifiers are reserved and cannot be used as variable names:\nfunctions, model, data, parameters, quantities,\ntransformed, generated\n\n\n\nVariable names will also conflict with the names of distributions suffixed with _lpdf, _lpmf, _lcdf, and _lccdf, _cdf, and _ccdf, such as normal_lcdf_log. No user-defined variable can take a name ending in _lupdf or _lupmf even if a corresponding _lpdf or _lpmf is not defined.\nUsing any of these variable names causes the stanc translator to halt and report the name and location of the variable causing the conflict.\n\n\n\nStan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed _stan_NAME (e.g. _stan_public) behind the scenes (in the generated C++ code).\n\nalignas, alignof, and, and_eq, asm, bitand, bitor, bool,\ncase, catch, char, char16_t, char32_t, class, compl, const,\nconstexpr, const_cast, decltype, default, delete, do,\ndouble, dynamic_cast, enum, explicit, float, friend, goto,\ninline, long, mutable, namespace, new, noexcept, not, not_eq,\nnullptr, operator, or, or_eq, private, protected, public,\nregister, reinterpret_cast, short, signed, sizeof,\nstatic_assert, static_cast, switch, template, this, thread_local,\nthrow, try, typeid, typename, union, unsigned, using, virtual,\nvolatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,\nSTAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH\n\n\n\n\nThe legal characters for variable identifiers are given in the following table.\n\n\nThe alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.\n\n\n\ncharacters\nASCII code points\n\n\n\n\na -- z\n97 – 122\n\n\nA -- Z\n65 – 90\n\n\n0 -- 9\n48 – 57\n\n\n_\n95\n\n\n\nAlthough not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.\n\n\n\nWithin comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.\n\n\n\n\n\nExpressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.\n\n\nSquare brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression [ 1, 10, 100 ] denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.\nrow_vector[2] rv2 =  [ 1, 2 ];\nvector[3] v3 = [ 3, 4, 5 ]';\nThe vector expression values may be compound expressions or variable names, so it is legal to write [ 2 * 3, 1 + 4] or [ x, y ], providing that x and y are primitive variables.\n\n\n\nA matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.\nmatrix[3, 2] m1 = [ [ 1, 2 ], [ 3, 4 ], [5, 6 ] ];\nAny expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:\nvector[2] vX = [ 1, 10 ]';\nrow_vector[2] vY = [ 100, 1000 ];\nmatrix[3, 2] m2 = [ vX', vY, [ 1, 2 ]  ];\n\n\n\nComplex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.\ncomplex_vector[3] = [1 + 2i, 3 - 1.7i, 0]';\ncomplex_row_vector[2] = [12, -2i];\ncomplex_matrix[2, 3] = [[1 + 2i, 3 - 1.7i, 0],\n                        [3.9 - 1.234i, 176i, 1 + 1i]];\n\n\nThe empty expression [ ] is ambiguous and therefore is not allowed and similarly expressions such as [ [ ] ] or [ [ ], [ ] ] are not allowed.\n\n\n\n\nIf needed, it is possible to create an empty vector with\nrep_vector(e, 0)\nwhere the first expression e needs to scalar of type real.\nIf needed, it is possible to create an empty matrix with\nrep_matrix(e, 0, 0)\nwhere the first expression e needs to scalar of type real.\n\n\n\nCurly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression { 1, 10, 100 } denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.\narray[3] int a = { 1, 10, 100 };\nThe values may be compound expressions, so it is legal to write { 2 * 3, 1 + 4 }. It is also possible to write two dimensional arrays directly, as in the following example.\narray[2, 3] int b = { { 1, 2, 3 }, { 4, 5, 6 } };\nThis way, b[1] is { 1, 2, 3 } and b[2] is { 4, 5, 6 }.\nWhitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.\narray[2, 3] int b = { { 1, 2, 3 },\n                { 4, 5, 6 } };\n\n\n\nThe empty array expression ({ }) is not allowed. See more about restrictions on array expressions in subsection Restrictions on values.\nIf needed, it is possible to create an empty array with\nrep_array(e, 0)\nwhere the first expression e determines the type of the array. For example, rep_array(0.0, 0) returns an empty real array of type real[], whereas rep_array({123}, 0) returns an empty two dimensional integer array of type int[ , ]. Only the type of the first argument is used, so the integer arrays {123} and {0} produce equivalent values.\n\n\n\nAny type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.\nvector[3] b;\nvector[3] c;\n// ...\narray[2] vector[3] d = { b, c };\nThe elements may also be a mixture of int and real typed expressions, in which case the result is an array of real values.\narray[2] real b = { 1, 1.9 };\n\n\n\nStan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.\ntuple(int, vector[3]) xy = (42, [1, 2.9, -1.3]');\nThe expression 42 is of type int and the expression [1, 2.9, -1.3] is of type row_vector so that [1, 2.9, -1.3]' is of type vector and of size 3. The whole tuple expression (42, [1, 2.9, -1.3]') thus has a sized type of tuple(int, vector[3]) and an unsized type (e.g., for a function argument) of tuple(int, vector).\nStan does not support the Python notation with trailing commas, such as (1, 2, 3, ) for a 3-tuple.\n\n\n\nThere are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.\n\n\nAlthough it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.\n{ { 1, 2, 3 }, { 4, 5 } }  // compile time error: size mismatch\nThis may appear to be OK, because it is creating a two-dimensional integer array (array[,] int) out of two one-dimensional array integer arrays (array[] int). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.\n{ { 1, 2, 3 }, m }  // runtime error if m not size 3\n\n\n\nBecause there is no way to infer the type of the result, the empty array expression ({ }) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.\narray[0] int a;   // a is fully defined as zero element array\n\n\n\nThere is no way to declare or construct a zero-tuple or one-tuple in Stan. Tuples must be at least two elements long. The expression () does not pick out a zero-tuple—it is ill formed. Similarly, the expression (1) is of type int rather than a tuple.\n\n\n\n\n\nAny expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, ( and ), are allowed. The square brackets [ and ] are reserved for array indexing and the curly braces { and } for grouping statements.\nWith parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression 1 + 2 * 3 has a subexpression 2 * 3 and evaluates to 7. With parentheses, this grouping may be made explicit with the expression 1 + (2 * 3). More importantly, the expression (1 + 2) *   3 has 1 + 2 as a subexpression and evaluates to 9.\n\n\n\nFor integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (+), subtraction (-), multiplication (*) and division (/) in the usual ways.\nFor integer expressions, Stan supports the modulus (%) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3.0 + 0.14\n-15\n2 * 3 + 1\n(x - y) / 2.0\n(n * (n + 1)) / 2\nx / n\nm % n\nThe negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (') is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.\nFor example, if y and mu are variables of type vector and Sigma is a variable of type matrix, then (y - mu)' * Sigma * (y - mu) is a well-formed expression of type real. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) y - mu are of type vector because the variables y and mu are of type vector. The transpose of this expression, the subexpression (y - mu)' is of type row_vector. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form (((y - mu)') * Sigma) * (y - mu).\nThe type of subexpression (y - mu)' * Sigma is inferred to be row_vector, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a real value.\nStan provides elementwise matrix multiplication (e.g., a .* b) and division (e.g., a ./ b) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,\nvector[N] a;\nvector[N] b;\nvector[N] c;\nthe assignment,\nc = a .* b;\nproduces the same result with roughly the same efficiency as the loop\nfor (n in 1:N) {\n  c[n] = a[n] * b[n];\n}\nStan supports exponentiation (^) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3 ^ 2\n3.0 ^ -2\n3.0 ^ 0.14\nx ^ n\nn ^ x\nn ^ m\nx ^ y\nExponentiation is right associative, so the expression 2 ^ 3 ^ 4 is equivalent to the fully specified form 2 ^ (3 ^ 4).\n\n\nThe precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.\n\n\nStan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e., operator*(int, int):int indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.\n\n\n\n\n\n\n\n\n\n\nOp.\nPrec.\nAssoc.\nPlacement\nDescription\n\n\n\n\n? ~ :\n10\nright\nternary infix\nconditional\n\n\n||\n9\nleft\nbinary infix\nlogical or\n\n\n&&\n8\nleft\nbinary infix\nlogical and\n\n\n==\n7\nleft\nbinary infix\nequality\n\n\n!=\n7\nleft\nbinary infix\ninequality\n\n\n&lt;\n6\nleft\nbinary infix\nless than\n\n\n&lt;=\n6\nleft\nbinary infix\nless than or equal\n\n\n&gt;\n6\nleft\nbinary infix\ngreater than\n\n\n&gt;=\n6\nleft\nbinary infix\ngreater than or equal\n\n\n+\n5\nleft\nbinary infix\naddition\n\n\n-\n5\nleft\nbinary infix\nsubtraction\n\n\n*\n4\nleft\nbinary infix\nmultiplication\n\n\n.*\n4\nleft\nbinary infix\nelementwise multiplication\n\n\n/\n4\nleft\nbinary infix\n(right) division\n\n\n./\n4\nleft\nbinary infix\nelementwise division\n\n\n%\n4\nleft\nbinary infix\nmodulus\n\n\n\\\n3\nleft\nbinary infix\nleft division\n\n\n%/%\n3\nleft\nbinary infix\ninteger division\n\n\n!\n2\nn/a\nunary prefix\nlogical negation\n\n\n-\n2\nn/a\nunary prefix\nnegation\n\n\n+\n2\nn/a\nunary prefix\npromotion (no-op in Stan)\n\n\n^\n1\nright\nbinary infix\nexponentiation\n\n\n.^\n1\nright\nbinary infix\nelementwise exponentiation\n\n\n'\n0\nn/a\nunary postfix\ntransposition\n\n\n()\n0\nn/a\nprefix, wrap\nfunction application\n\n\n[]\n0\nleft\nprefix, wrap\narray, matrix indexing\n\n\n\nOther expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.\nThe precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression a + b + c is interpreted as (a + b) + c. Similarly, a / b * c is interpreted as (a / b) * c.\nBecause multiplication has higher precedence than addition, the expression a * b + c is interpreted as (a * b) + c and the expression a + b * c is interpreted as a + (b * c). Similarly, 2 * x + 3 * - y is interpreted as (2 * x) + (3 * (-y)).\nTransposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, -u' is interpreted as -(u'), u * v' as u* (v'), and u' * v as (u') * v. For integer and reals, -n ^ 3 is interpreted as -(n ^ 3).\n\n\n\n\n\n\n\nThe ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If a is an expression of type int and b and c are expressions that can be converted to one another (e.g., compared with ==), then\na ? b : c\nis an expression of the promoted type of b and c. The result of this expression is b if condition a is true, c otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g. if one argument is of type int and the other of type real, the conditional expression as a whole is of type real. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.\n\n\nThe conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,\na &gt; 0 || b &lt; 0 ? c + d : e - f\nis equivalent to the explicitly grouped version\n(a &gt; 0 || b &lt; 0) ? (c + d) : (e - f)\nThe latter is easier to read even if the parentheses are not strictly necessary.\n\n\n\nThe conditional operator is right associative, so that\na ? b : c ? d : e\nparses as if explicitly grouped as\na ? b : (c ? d : e)\nAgain, the explicitly grouped version is easier to read.\n\n\n\n\nStan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable a in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value b is returned if the condition evaluates to a non-zero value and c is returned if the condition evaluates to zero.\n\n\nThe key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.\n\n\n\nIf one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full if-then conditional statement. For example,\ndata {\n  array[10] real x;\n  // ...\n}\nparameters {\n  array[10] real z;\n  // ...\n}\nmodel {\n  y ~ normal(cond ? x : z, sigma);\n  // ...\n}\nwould be more efficiently (if not more transparently) coded as\nif (cond) {\n  y ~ normal(x, sigma);\n} else {\n  y ~ normal(z, sigma);\n}\nThe conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable x will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.\n\n\n\n\n\nStan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if x is a variable of type array [] real (a one-dimensional array of reals) then x[1] is the value of the first element of the array.\nSubscripting has higher precedence than any of the arithmetic operations. For example, alpha * x[1] is equivalent to alpha * (x[1]).\nMultiple subscripts may be provided within a single pair of square brackets. If x is of type array[,] real, a two-dimensional array, then x[2, 501] is of type real.\n\n\nThe subscripting operator also returns subarrays of arrays. For example, if x is of type array[,,] real, then x[2] is of type array[,] real, and x[2, 3] is of type array[] real. As a result, the expressions x[2, 3] and x[2][3] have the same meaning.\n\n\n\nIf Sigma is a variable of type matrix, then Sigma[1] denotes the first row of Sigma and has the type row_vector.\n\n\n\nStan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if m is of type matrix[ , ], a two-dimensional array of matrices, then m[1] refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that m[1, 2] is of type matrix and denotes the matrix in the first row and second column of the array. Continuing to add indices, m[1, 2, 3] is of type row_vector and denotes the third row of the matrix denoted by m[1, 2]. Finally, m[1, 2, 3, 4] is of type real and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array m.\n\n\n\n\nIn addition to single integer indexes, as described in the language indexing section, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.\n\n\nTypes of indexes and examples with one-dimensional containers of size N and an integer array ii of type array [] real size K.\n\n\n\nindex type\nexample\nvalue\n\n\n\n\ninteger\na[11]\nvalue of a at index 11\n\n\ninteger array\na[ii]\na[ii[1]], …, a[ii[K]]\n\n\nlower bound\na[3:]\na[3], …, a[N]\n\n\nupper bound\na[:5]\na[1], …, a[5]\n\n\nrange\na[2:7]\na[2], …, a[7]\n\n\nrange\na[7:2]\n[]\n\n\nrange\na[5-3:5+2]\na[2], …, a[7]\n\n\nall\na[:]\na[1], …, a[N]\n\n\nall\na[]\na[1], …, a[N]\n\n\n\nThe range indexing with : allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:\n  array[6] int ii = reverse(linspaced_int_array(6, 2, 7));\nThen a[ii] evaluates to a[7], …, a[2].\n\n\n\nThe fundamental semantic rule for dealing with multiple indexes is the following. If idxs is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.\na[idxs, ...][i, ...] = a[idxs[i], ...][...]\nOn the other hand, if idx is a single index, it reduces the dimensionality of the output, so that\na[idx, ...] = a[idx][...]\nThe only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.\n\n\nSpecial rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix a, with integer single indexes i and j and integer array multiple indexes is and js. The same typing rules apply for all multiple indexes.\n\n\n\nexample\nrow index\ncolumn index\nresult type\n\n\n\n\na[i]\nsingle\nn/a\nrow vector\n\n\na[is]\nmultiple\nn/a\nmatrix\n\n\na[i, j]\nsingle\nsingle\nreal\n\n\na[i, js]\nsingle\nmultiple\nrow vector\n\n\na[is, j]\nmultiple\nsingle\nvector\n\n\na[is, js]\nmultiple\nmultiple\nmatrix\n\n\n\nEvaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.\nm[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]\nm[idxs, idx][j] = m[idxs[j], idx]\nm[idx, idxs][j] = m[idx, idxs[j]]\nEvaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.\n\n\n\n\n\nStan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.\nExpressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, log(2.0) is the expression of type real denoting the result of applying the natural logarithm to the value of the real literal 2.0.\nSyntactically, function application has higher precedence than any of the other operators, so that y + log(x) is interpreted as y + (log(x)).\n\n\nEach function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as\nreal log(real);\nand the signature for the lmultiply function is\nreal lmultiply(real, real);\nA function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.\nreal mean(array [] real);\nreal mean(vector);\nThe first applies to a one-dimensional array of real values and the second to a vector.\nThe identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.\n\n\n\nConstants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants \\(\\pi\\) and \\(e\\) are represented as nullary functions named pi() and e(). See the Stan Functions Reference built-in constants section for a list of built-in constants.\n\n\n\nBecause of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.\nFor example, consider a situation in which the following two function signatures have been registered for foo.\nreal foo(real, real);\nint foo(int, int);\nThe use of foo in the expression foo(1.0, 1.0) resolves to foo(real, real), and thus the expression foo(1.0, 1.0) itself is assigned a type of real.\nBecause integers may be promoted to real values, the expression foo(1, 1) could potentially match either foo(real, real) or foo(int, int). The former requires two type promotions and the latter requires none, so foo(1, 1) is resolved to function foo(int, int) and is thus assigned the type int.\nThe expression foo(1, 1.0) has argument types (int, real) and thus does not explicitly match either function signature. By promoting the integer expression 1 to type real, it is able to match foo(real, real), and hence the type of the function expression foo(1, 1.0) is real.\nIn some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named bar with the following signatures.\nreal bar(real, int);\nreal bar(int, real);\nWith these signatures, the expression bar(1.0, 1) and bar(1, 1.0) resolve to the first and second of the above functions, respectively. The expression bar(1.0, 1.0) is illegal because real values may not be demoted to integers. The expression bar(1, 1) is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name bar is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with bar(1, 1) given the two declarations above, the Stan compiler will flag the expression as illegal.\n\n\n\nFor most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix _rng. For example, a univariate normal random number can be generated by normal_rng(0, 1); only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.\n\n\nThe use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in _rng.\nThis allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.\n\n\n\nPosterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of (Gelman et al. 2013) for more information on posterior predictive checks.\n\n\n\n\n\nStan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.\n\n\nThe primitive implementation types for Stan are\nint, real, complex, vector, row_vector,  matrix, complex_vector,\ncomplex_row_vector, complex_matrix\nEvery basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.\n\n\nThe table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.\n\n\n\ntype\nprimitive type\n\n\n\n\nint\nint\n\n\nreal\nreal\n\n\nvector\nvector\n\n\nsimplex\nvector\n\n\nunit_vector\nvector\n\n\nsum_to_zero_vector\nvector\n\n\nordered\nvector\n\n\npositive_ordered\nvector\n\n\nrow_vector\nrow_vector\n\n\nmatrix\nmatrix\n\n\ncov_matrix\nmatrix\n\n\ncorr_matrix\nmatrix\n\n\ncholesky_factor_cov\nmatrix\n\n\ncholesky_factor_corr\nmatrix\n\n\ncolumn_stochastic_matrix\nmatrix\n\n\nrow_stochastic_matrix\nmatrix\n\n\nsum_to_zero_matrix\nmatrix\n\n\ncomplex_vector\ncomplex_vector\n\n\ncomplex_row_vector\ncomplex_row_vector\n\n\ncomplex_matrix\ncomplex_matrix\n\n\n\nA full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, array [] real has an array dimensionality of 1, int an array dimensionality of 0, and array [,,] int an array dimensionality of 3. The implementation type matrix[ , , ] has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.\nRecall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.\narray[I, J, K] matrix[M, N] a;\nThe matrix a is indexed as a[i, j, k, m, n] with the array indices first, followed by the matrix indices, with a[i, j, k] being a matrix and a[i, j, k, m] being a row vector.\n\n\n\n\nStan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.\n\n\n\nThere are two basic promotion rules,\n\nint types may be promoted to real, and\nreal types may be promoted to complex.\n\nPlus, promotion is transitive, so that\n\nif type U can be promoted to type V and type V can be promoted to type T, then U can be promoted to T.\n\nThe first rule means that expressions of type int may be used anywhere an expression of type real is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.\nThe remaining rules have to do with covariant typing rules, which say that a container of type U may be promoted to a container of the same shape of type T if U can be promoted to T. For vector and matrix types, this induces three rules,\n\nvector may be promoted to complex_vector,\nrow_vector may be promoted to complex_row_vector\nmatrix may be promoted to complex_matrix.\n\nFor array types, there’s a single rule\n\narray[...] U may be promoted to array[...] T if U can be promoted to T.\n\nFor example, this means array[,] int may be used where array [,] real or array [,] complex is required; as another example, array[] real may be used anywhere array[] complex is required.\nTuples have the natural extension of the above rules, applied to all sub-types at once\n\nA tuple(U1, ..., UN) may be promoted to a tuple(T1, ..., TN) if every Un can be promoted to Tn for n in 1:N\n\n\n\nAn integer literal expression such as 42 is of type int. Real literals such as 42.0 are of type real. Imaginary literals such as -17i are of type complex. the expression 7 - 2i acts like a complex literal, but technically it combines a real literal 7 and an imaginary literal 2i through subtraction.\n\n\n\nThe type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is int.\nThere is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.1\n\n\n\nIf x is an expression of total dimensionality greater than or equal to \\(N\\), then the type of expression e[i1, i2, ..., iN] is the same as that of e[i1][i2]...[iN], so it suffices to define the type of a singly-indexed function. Suppose e is an expression and i is an expression of primitive type int. Then\n\nif e is an expression of type array[i1, i2, ..., iN] T and k, i1, …, iN are expressions of type int, then e[k] is an expression of type array[i2, ..., iN] T,\nif e is an expression of type array[i] T with i and k expressions of type int, then e[k] is of type T,\nif e has implementation type vector or row_vector, dimensionality 0, then e[i] has implementation type real,\nif e has implementation type matrix, then e[i] has type row_vector,\nif e has implementation type complex_vector or complex_row_vector and i is an expression of type int, then e[i] is an expression of type complex, and\nif e has implementation type complex_matrix, and i is an expression of type int, then e[i] is an expression of type complex_row_vector.\n\n\n\n\nIf f is the name of a function and e1,...,eN are expressions for \\(N \\geq 0\\), then f(e1,...,eN) is an expression whose type is determined by the return type in the function signature for f given e1 through eN. Recall that a function signature is a declaration of the argument types and the result type.\nIn looking up functions, binary operators like real * real are defined as operator*(real, real) in the documentation and index.\nIn matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type int may be promoted to type real or complex if necessary (see the subsection on type promotion in the function application section, a real argument will be promoted to complex if necessary, a vector will be promoted to complex_vector if necessary, and so on.\nIn general, matrix operations return the lowest inferable type. For example, row_vector * vector returns a value of type real, which is declared in the function documentation and index as real operator*(row_vector, vector).\n\n\n\n\n\nThere are several expression constructions in Stan that act as higher-order functions.2\nThe higher-order functions and the signature of their argument functions are listed in the following pair of tables.\n\n\nHigher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.\n\n\n\n\n\n\n\n\n\nfunction\nparameter or data args\ndata args\nreturn type\n\n\n\n\nalgebra_solver\nvector, vector\narray [] real, array [] real\nvector\n\n\nalgebra_solver_newton\nvector, vector\narray [] real, array [] real\nvector\n\n\nintegrate_1d,\nreal, real, array [] real\narray [] real, array [] real\nreal\n\n\nintegrate_ode_X,\nreal, array [] real, array [] real\narray [] real, array [] real\narray [] real\n\n\nmap_rect\nvector, vector\narray [] real, array [] real\nvector\n\n\n\nFor example, the integrate_ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  array [] real foo(real t,\n                    array [] real y,\n                    array [] real theta,\n                    array [] real x_r,\n                    array [] real x_i) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\narray[2] real y0;\nreal t0;\narray[T] real ts;\narray[1] real theta;\narray[0] real x_r;\narray[0] int x_i;\n// ...\narray[T, 2] real y_hat = integrate_ode_rk45(foo, y0, t0,\n                                              ts, theta, x_r, x_i);\nThe function argument is foo, the name of the user-defined function; as shown in the higher-order functions table, integrate_ode_rk45 takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.\n\n\n\nVariadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.\n\n\n\nfunction\nrestricted args\nreturn type\n\n\n\n\nsolve_X\nvector\nvector\n\n\node_X,\nvector, real, array [] real\nvector[]\n\n\nreduce_sum\narray[] T, T1, T2\nreal\n\n\n\nT, T1, and T2 can be any Stan type.\nFor example, the ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  vector foo(real t, vector y, real theta, vector beta,\n            array [] real x_i, int index) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\nvector[2] y0;\nreal t0;\narray[T] real ts;\nreal theta;\nvector[7] beta;\narray[10] int x_i;\nint index;\n// ...\nvector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,\n                              beta, x_i, index);\nThe function argument is foo, the name of the user-defined function. As shown in the variadic higher-order functions table, ode_rk45 takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of foo and returns an array of vectors.\n\n\n\nThe function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.\n\n\n\nSome of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.\nFor user-defined functions the qualifier data may be prepended to the type to restrict the argument to data-only variables.\n\n\n\n\nDerivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.\n\n\nUnlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(sqrt(x - x), 1);\n}\nAlgebraically, the distribution statement in the model could be reduced to\n  x ~ normal(0, 1);\nand it would seem the model should produce unit normal draws for x. But rather than canceling, the expression sqrt(x -   x) causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,\n\\[\n\\begin{array}{rcl}\n\\frac{d}{dx} \\sqrt{x - x}\n& = &\n\\frac{1}{2 \\sqrt{x - x}} \\times \\frac{d}{dx} (x - x)\n\\\\[4pt]\n& = &\n\\frac{1}{0} \\times (1 - 1)\n\\\\[4pt]\n& = &\n\\infty \\times 0\n\\\\[4pt]\n& = & \\mathrm{NaN}.\n\\end{array}\n\\]\nRather than the \\(x - x\\) canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.\nThe only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like sqrt(x - x) for which the chain rule produces not-a-number values.\n\n\n\nThe best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).\nFor example, compiling the above model to an executable sqrt-x-minus-x in CmdStan, the test can be run as\n&gt; ./sqrt-x-minus-x diagnose test=gradient\nwhich produces\n...\nTEST GRADIENT MODE\n\n Log probability=-0.393734\n\n param idx           value           model     finite diff           error\n         0       -0.887393             nan               0             nan\nEven though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.",
+    "text": "The following broadcast functions allow vectors, row vectors and matrices to be created by copying a single element into all of their cells. Matrices may also be created by stacking copies of row vectors vertically or stacking copies of column vectors horizontally.\n \n\nvector rep_vector(real x, int m) Return the size m (column) vector consisting of copies of x.\nAvailable since 2.0\n \n\nrow_vector rep_row_vector(real x, int n) Return the size n row vector consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(real x, int m, int n) Return the m by n matrix consisting of copies of x.\nAvailable since 2.0\n \n\nmatrix rep_matrix(vector v, int n) Return the m by n matrix consisting of n copies of the (column) vector v of size m.\nAvailable since 2.0\n \n\nmatrix rep_matrix(row_vector rv, int m) Return the m by n matrix consisting of m copies of the row vector rv of size n.\nAvailable since 2.0\nUnlike the situation with array broadcasting (see section array broadcasting), where there is a distinction between integer and real arguments, the following two statements produce the same result for vector broadcasting; row vector and matrix broadcasting behave similarly.\n vector[3] x;\n x = rep_vector(1, 3);\n x = rep_vector(1.0, 3);\nThere are no integer vector or matrix types, so integer values are automatically promoted.\n\n\n \n\nmatrix symmetrize_from_lower_tri(matrix A)\nConstruct a symmetric matrix from the lower triangle of A.\nAvailable since 2.26",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#numeric-literals",
-    "href": "reference-manual/expressions.html#numeric-literals",
-    "title": "Expressions",
+    "objectID": "functions-reference/matrix_operations.html#diagonal-matrix-functions",
+    "href": "functions-reference/matrix_operations.html#diagonal-matrix-functions",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "The simplest form of expression is a literal that denotes a primitive numerical value.\n\n\nInteger literals represent integers of type int. Integer literals are written in base 10 without any separators. Integer literals may contain a single negative sign. (The expression --1 is interpreted as the negation of the literal -1.)\nThe following list contains well-formed integer literals.\n0, 1, -1, 256, -127098, 24567898765\nInteger literals must have values that fall within the bounds for integer values (see the section on numerical data types).\nInteger literals may not contain decimal points (.). Thus the expressions 1. and 1.0 are of type real and may not be used where a value of type int is required.\n\n\n\nA number written with a period or with scientific notation is assigned to a the continuous numeric type real. Real literals are written in base 10 with a period (.) as a separator and optionally an exponent with optional sign. Examples of well-formed real literals include the following.\n0.0, 1.0, 3.14, -217.9387, 2.7e3, -2E-5, 1.23e+3.\nThe notation e or E followed by a positive or negative integer denotes a power of 10 to multiply. For instance, 2.7e3 and 2.7e+3 denote \\(2.7 \\times 10^3\\), whereas -2E-5 denotes \\(-2 \\times\n10^{-5}\\).\n\n\n\nA number followed by the character i denotes an imaginary number and is assigned to the numeric type complex. The number preceding i may be either a real or integer literal and determines the magnitude of the imaginary number. Examples of well-formed imaginary literals include the following.\n1i, 2i, -325.786i, 1e10i, 2.87e-10i.\nNote that the character i by itself is not a well-formed imaginary literal. The unit imaginary number must be written as 1i.\n\n\n\nStan does not include complex literals directly, but a real or integer literal can be added to an imaginary literal to derive an expression that behaves like a complex literal. Examples include the following.\n1 + 2i, -3.2e9 + 1e10i\nThese will be assigned the type complex, which is the result of adding a real or integer and a complex number. They will also function like literals in the sense that the C++ compiler is able to reduce them to a single complex constant at compile time.",
+    "text": "matrix add_diag(matrix m, row_vector d) Add row_vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, vector d) Add vector d to the diagonal of matrix m.\nAvailable since 2.21\n \n\nmatrix add_diag(matrix m, real d) Add scalar d to every diagonal element of matrix m.\nAvailable since 2.21\n \n\nvector diagonal(matrix x) The diagonal of the matrix x\nAvailable since 2.0\n \n\nmatrix diag_matrix(vector x) The diagonal matrix with diagonal x\nAvailable since 2.0\nAlthough the diag_matrix function is available, it is unlikely to ever show up in an efficient Stan program. For example, rather than converting a diagonal to a full matrix for use as a covariance matrix,\n y ~ multi_normal(mu, diag_matrix(square(sigma)));\nit is much more efficient to just use a univariate normal, which produces the same density,\n y ~ normal(mu, sigma);\nRather than writing m * diag_matrix(v) where m is a matrix and v is a vector, it is much more efficient to write diag_post_multiply(m, v) (and similarly for pre-multiplication). By the same token, it is better to use quad_form_diag(m, v) rather than quad_form(m, diag_matrix(v)).\n \n\nmatrix identity_matrix(int k) Create an identity matrix of size \\(k \\times k\\)\nAvailable since 2.26",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#variables.section",
-    "href": "reference-manual/expressions.html#variables.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/matrix_operations.html#container-construction",
+    "href": "functions-reference/matrix_operations.html#container-construction",
+    "title": "Matrix Operations",
     "section": "",
-    "text": "A variable by itself is a well-formed expression of the same type as the variable. Variables in Stan consist of ASCII strings containing only the basic lower-case and upper-case Roman letters, digits, and the underscore (_) character. Variables must start with a letter (a--z and A--Z) and may not end with two underscores (__).\nExamples of legal variable identifiers are as follows.\na, a3, a_3, Sigma, my_cpp_style_variable, myCamelCaseVariable\nUnlike in R and BUGS, variable identifiers in Stan may not contain a period character.\n\n\nStan reserves many strings for internal use and these may not be used as the name of a variable. An attempt to name a variable after an internal string results in the stanc translator halting with an error message indicating which reserved name was used and its location in the model code.\n\n\nThe name of the model cannot be used as a variable within the model. This is usually not a problem because the default in bin/stanc is to append _model to the name of the file containing the model specification. For example, if the model is in file foo.stan, it would not be legal to have a variable named foo_model when using the default model name through bin/stanc. With user-specified model names, variables cannot match the model.\n\n\n\nThe following list contains reserved words for Stan’s programming language. Not all of these features are implemented in Stan yet, but the tokens are reserved for future use.\nfor, in, while, repeat, until, if, then, else,\ntrue, false, target, struct, typedef, export,\nauto, extern, var, static, lower, upper, offset,\nmultiplier\nVariables should not be named after types, either, and thus may not be any of the following.\nint, real, complex, vector, simplex, unit_vector,\nsum_to_zero_vector, sum_to_zero_matrix, ordered, positive_ordered,\nrow_vector, matrix, cholesky_factor_corr,\ncolumn_stochastic_matrix,row_stochastic_matrix,\ncholesky_factor_cov, corr_matrix, cov_matrix, array\nThe following keywords or built-in functions are also reserved and cannot be used as variable names:\nprint, reject, profile, fatal_error, target, jacobian\nThe following block identifiers are reserved and cannot be used as variable names:\nfunctions, model, data, parameters, quantities,\ntransformed, generated\n\n\n\nVariable names will also conflict with the names of distributions suffixed with _lpdf, _lpmf, _lcdf, and _lccdf, _cdf, and _ccdf, such as normal_lcdf_log. No user-defined variable can take a name ending in _lupdf or _lupmf even if a corresponding _lpdf or _lpmf is not defined.\nUsing any of these variable names causes the stanc translator to halt and report the name and location of the variable causing the conflict.\n\n\n\nStan primarily generates code in C++, which features its own reserved words. It is legal to name a variable any of the following names, however doing so will lead to it being renamed _stan_NAME (e.g. _stan_public) behind the scenes (in the generated C++ code).\n\nalignas, alignof, and, and_eq, asm, bitand, bitor, bool,\ncase, catch, char, char16_t, char32_t, class, compl, const,\nconstexpr, const_cast, decltype, default, delete, do,\ndouble, dynamic_cast, enum, explicit, float, friend, goto,\ninline, long, mutable, namespace, new, noexcept, not, not_eq,\nnullptr, operator, or, or_eq, private, protected, public,\nregister, reinterpret_cast, short, signed, sizeof,\nstatic_assert, static_cast, switch, template, this, thread_local,\nthrow, try, typeid, typename, union, unsigned, using, virtual,\nvolatile, wchar_t, xor, xor_eq, fvar, STAN_MAJOR, STAN_MINOR,\nSTAN_PATCH, STAN_MATH_MAJOR, STAN_MATH_MINOR, STAN_MATH_PATCH\n\n\n\n\nThe legal characters for variable identifiers are given in the following table.\n\n\nThe alphanumeric characters and underscore in base ASCII are the only legal characters in Stan identifiers.\n\n\n\ncharacters\nASCII code points\n\n\n\n\na -- z\n97 – 122\n\n\nA -- Z\n65 – 90\n\n\n0 -- 9\n48 – 57\n\n\n_\n95\n\n\n\nAlthough not the most expressive character set, ASCII is the most portable and least prone to corruption through improper character encodings or decodings. Sticking to this range of ASCII makes Stan compatible with Latin-1 or UTF-8 encodings of these characters, which are byte-for-byte identical to ASCII.\n\n\n\nWithin comments, Stan can work with any ASCII-compatible character encoding, such as ASCII itself, UTF-8, or Latin1. It is up to user shells and editors to display them properly.",
+    "text": "array[] real linspaced_array(int n, data real lower, data real upper) Create a real array of length n of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int linspaced_int_array(int n, int lower, int upper) Create a regularly spaced, increasing integer array of length n between lower and upper, inclusively. If (upper - lower) / (n - 1) is less than one, repeat each output (n - 1) / (upper - lower) times. If neither (upper - lower) / (n - 1) or (n - 1) / (upper - lower) are integers, upper is reduced until one of these is true.\nAvailable since 2.26\n \n\nvector linspaced_vector(int n, data real lower, data real upper) Create an n-dimensional vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\nrow_vector linspaced_row_vector(int n, data real lower, data real upper) Create an n-dimensional row-vector of equidistantly-spaced elements between lower and upper\nAvailable since 2.24\n \n\narray[] int one_hot_int_array(int n, int k) Create a one-hot encoded int array of length n with array[k] = 1\nAvailable since 2.26\n \n\narray[] real one_hot_array(int n, int k) Create a one-hot encoded real array of length n with array[k] = 1\nAvailable since 2.24\n \n\nvector one_hot_vector(int n, int k) Create an n-dimensional one-hot encoded vector with vector[k] = 1\nAvailable since 2.24\n \n\nrow_vector one_hot_row_vector(int n, int k) Create an n-dimensional one-hot encoded row-vector with row_vector[k] = 1\nAvailable since 2.24\n \n\narray[] int ones_int_array(int n) Create an int array of length n of all ones\nAvailable since 2.26\n \n\narray[] real ones_array(int n) Create a real array of length n of all ones\nAvailable since 2.26\n \n\nvector ones_vector(int n) Create an n-dimensional vector of all ones\nAvailable since 2.26\n \n\nrow_vector ones_row_vector(int n) Create an n-dimensional row-vector of all ones\nAvailable since 2.26\n \n\narray[] int zeros_int_array(int n) Create an int array of length n of all zeros\nAvailable since 2.26\n \n\narray[] real zeros_array(int n) Create a real array of length n of all zeros\nAvailable since 2.24\n \n\nvector zeros_vector(int n) Create an n-dimensional vector of all zeros\nAvailable since 2.24\n \n\nrow_vector zeros_row_vector(int n) Create an n-dimensional row-vector of all zeros\nAvailable since 2.24\n \n\nvector uniform_simplex(int n) Create an n-dimensional simplex with elements vector[i] = 1 / n for all \\(i \\in 1, \\dots, n\\)\nAvailable since 2.24",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#slicing-and-blocking-functions",
+    "href": "functions-reference/matrix_operations.html#slicing-and-blocking-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Stan provides several functions for generating slices or blocks or diagonal entries for matrices.\n\n\n \n\nvector col(matrix x, int n) The n-th column of matrix x\nAvailable since 2.0\n \n\nrow_vector row(matrix x, int m) The m-th row of matrix x\nAvailable since 2.0\nThe row function is special in that it may be used as an lvalue in an assignment statement (i.e., something to which a value may be assigned). The row function is also special in that the indexing notation x[m] is just an alternative way of writing row(x,m). The col function may not, be used as an lvalue, nor is there an indexing based shorthand for it.\n\n\n\n\n\nBlock operations may be used to extract a sub-block of a matrix.\n \n\nmatrix block(matrix x, int i, int j, int n_rows, int n_cols) Return the submatrix of x that starts at row i and column j and extends n_rows rows and n_cols columns.\nAvailable since 2.0\nThe sub-row and sub-column operations may be used to extract a slice of row or column from a matrix\n \n\nvector sub_col(matrix x, int i, int j, int n_rows) Return the sub-column of x that starts at row i and column j and extends n_rows rows and 1 column.\nAvailable since 2.0\n \n\nrow_vector sub_row(matrix x, int i, int j, int n_cols) Return the sub-row of x that starts at row i and column j and extends 1 row and n_cols columns.\nAvailable since 2.0\n\n\n\nThe head operation extracts the first \\(n\\) elements of a vector and the tail operation the last. The segment operation extracts an arbitrary subvector.\n \n\nvector head(vector v, int n) Return the vector consisting of the first n elements of v.\nAvailable since 2.0\n \n\nrow_vector head(row_vector rv, int n) Return the row vector consisting of the first n elements of rv.\nAvailable since 2.0\n \n\narray[] T head(array[] T sv, int n) Return the array consisting of the first n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector tail(vector v, int n) Return the vector consisting of the last n elements of v.\nAvailable since 2.0\n \n\nrow_vector tail(row_vector rv, int n) Return the row vector consisting of the last n elements of rv.\nAvailable since 2.0\n \n\narray[] T tail(array[] T sv, int n) Return the array consisting of the last n elements of sv; applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0\n \n\nvector segment(vector v, int i, int n) Return the vector consisting of the n elements of v starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.0\n \n\nrow_vector segment(row_vector rv, int i, int n) Return the row vector consisting of the n elements of rv starting at i; i.e., elements i through through i + n - 1.\nAvailable since 2.10\n \n\narray[] T segment(array[] T sv, int i, int n) Return the array consisting of the n elements of sv starting at i; i.e., elements i through through i + n - 1. Applies to up to three-dimensional arrays containing any type of elements T.\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#matrix-concatenation",
+    "href": "functions-reference/matrix_operations.html#matrix-concatenation",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "Stan’s matrix and vector concatenation operations append_col and append_row are like the operations cbind and rbind in R.\n\n\n \n\nmatrix append_col(matrix x, matrix y) Combine matrices x and y by column. The matrices must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(matrix x, vector y) Combine matrix x and vector y by column. The matrix and the vector must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, matrix y) Combine vector x and matrix y by column. The vector and the matrix must have the same number of rows.\nAvailable since 2.5\n \n\nmatrix append_col(vector x, vector y) Combine vectors x and y by column. The vectors must have the same number of rows.\nAvailable since 2.5\n \n\nrow_vector append_col(row_vector x, row_vector y) Combine row vectors x and y of any size into another row vector by appending y to the end of x.\nAvailable since 2.5\n \n\nrow_vector append_col(real x, row_vector y) Append x to the front of y, returning another row vector.\nAvailable since 2.12\n \n\nrow_vector append_col(row_vector x, real y) Append y to the end of x, returning another row vector.\nAvailable since 2.12\n\n\n\n \n\nmatrix append_row(matrix x, matrix y) Combine matrices x and y by row. The matrices must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(matrix x, row_vector y) Combine matrix x and row vector y by row. The matrix and the row vector must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, matrix y) Combine row vector x and matrix y by row. The row vector and the matrix must have the same number of columns.\nAvailable since 2.5\n \n\nmatrix append_row(row_vector x, row_vector y) Combine row vectors x and y by row. The row vectors must have the same number of columns.\nAvailable since 2.5\n \n\nvector append_row(vector x, vector y) Concatenate vectors x and y of any size into another vector.\nAvailable since 2.5\n \n\nvector append_row(real x, vector y) Append x to the top of y, returning another vector.\nAvailable since 2.12\n \n\nvector append_row(vector x, real y) Append y to the bottom of x, returning another vector.\nAvailable since 2.12",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#softmax",
+    "href": "functions-reference/matrix_operations.html#softmax",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "The softmax function maps1 \\(y \\in \\mathbb{R}^K\\) to the \\(K\\)-simplex by \\[\\begin{equation*} \\text{softmax}(y)  = \\frac{\\exp(y)}\n{\\sum_{k=1}^K \\exp(y_k)}, \\end{equation*}\\] where \\(\\exp(y)\\) is the componentwise exponentiation of \\(y\\). Softmax is usually calculated on the log scale, \\[\\begin{eqnarray*} \\log \\text{softmax}(y) & = & \\ y - \\log \\sum_{k=1}^K\n\\exp(y_k) \\\\[4pt] & = & y - \\mathrm{log\\_sum\\_exp}(y). \\end{eqnarray*}\\] where the vector \\(y\\) minus the scalar \\(\\mathrm{log\\_sum\\_exp}(y)\\) subtracts the scalar from each component of \\(y\\).\nStan provides the following functions for softmax and its log.\n \n\nvector softmax(vector x) The softmax of x\nAvailable since 2.0\n \n\nvector log_softmax(vector x) The natural logarithm of the softmax of x\nAvailable since 2.0\n\n\n\nThe cumulative sum of a sequence \\(x_1,\\ldots,x_N\\) is the sequence \\(y_1,\\ldots,y_N\\), where \\[\\begin{equation*} y_n = \\sum_{m = 1}^{n} x_m. \\end{equation*}\\]\n \n\narray[] int cumulative_sum(array[] int x) The cumulative sum of x\nAvailable since 2.30\n \n\narray[] real cumulative_sum(array[] real x) The cumulative sum of x\nAvailable since 2.0\n \n\nvector cumulative_sum(vector v) The cumulative sum of v\nAvailable since 2.0\n \n\nrow_vector cumulative_sum(row_vector rv) The cumulative sum of rv\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#gaussian-process-covariance-functions",
+    "href": "functions-reference/matrix_operations.html#gaussian-process-covariance-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "The Gaussian process covariance functions compute the covariance between observations in an input data set or the cross-covariance between two input data sets.\nFor one dimensional GPs, the input data sets are arrays of scalars. The covariance matrix is given by \\(K_{ij} = k(x_i, x_j)\\) (where \\(x_i\\) is the \\(i^{th}\\) element of the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(x_i, y_j)\\).\nFor multi-dimensional GPs, the input data sets are arrays of vectors. The covariance matrix is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{x}_j)\\) (where \\(\\mathbf{x}_i\\) is the \\(i^{th}\\) vector in the array \\(x\\)) and the cross-covariance is given by \\(K_{ij} = k(\\mathbf{x}_i, \\mathbf{y}_j)\\).\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponentiated quadratic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|^2}{2l^2} \\right)\n\\]\n \n\nmatrix gp_exp_quad_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exp_quad_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponentiated quadratic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith bias \\(\\sigma_0\\) the dot product kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma_0^2 + \\mathbf{x}_i^T \\mathbf{x}_j\n\\]\n \n\nmatrix gp_dot_prod_cov(array[] real x, real sigma)\nGaussian process covariance with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(array[] real x1, array[] real x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x, real sigma)\nGaussian process covariance with dot product kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_dot_prod_cov(vectors x1, vectors x2, real sigma)\nGaussian process cross-covariance of x1 and x2 with dot product kernel in multiple dimensions.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the exponential kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left( -\\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_exponential_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_exponential_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with exponential kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 3/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right) \\exp \\left( -\\frac{\\sqrt{3}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern32_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern32_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 3/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\) and length scale \\(l\\), the Matern 5/2 kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\left( 1 + \\frac{\\sqrt{5}|\\mathbf{x}_i - \\mathbf{x}_j|}{l} + \\frac{5 |\\mathbf{x}_i - \\mathbf{x}_j|^2}{3l^2} \\right)\n\\exp \\left( -\\frac{\\sqrt{5} |\\mathbf{x}_i - \\mathbf{x}_j|}{l} \\right)\n\\]\n \n\nmatrix gp_matern52_cov(array[] real x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(array[] real x1, array[] real x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x, real sigma, array[] real length_scale)\nGaussian process covariance with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_matern52_cov(vectors x1, vectors x2, real sigma, array[] real length_scale)\nGaussian process cross-covariance of x1 and x2 with Matern 5/2 kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20\n\n\n\nWith magnitude \\(\\sigma\\), length scale \\(l\\), and period \\(p\\), the periodic kernel is:\n\\[\nk(\\mathbf{x}_i, \\mathbf{x}_j) = \\sigma^2 \\exp \\left(-\\frac{2 \\sin^2 \\left( \\pi \\frac{|\\mathbf{x}_i - \\mathbf{x}_j|}{p} \\right) }{l^2} \\right)\n\\]\n \n\nmatrix gp_periodic_cov(array[] real x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(array[] real x1, array[] real x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in one dimension.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x, real sigma, real length_scale, real period)\nGaussian process covariance with periodic kernel in multiple dimensions.\nAvailable since 2.20\n \n\nmatrix gp_periodic_cov(vectors x1, vectors x2, real sigma, real length_scale, real period)\nGaussian process cross-covariance of x1 and x2 with periodic kernel in multiple dimensions with a length scale for each dimension.\nAvailable since 2.20",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#linear-algebra-functions-and-solvers",
+    "href": "functions-reference/matrix_operations.html#linear-algebra-functions-and-solvers",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "In general, it is much more efficient and also more arithmetically stable to use matrix division than to multiply by an inverse. There are specialized forms for lower triangular matrices and for symmetric, positive-definite matrices.\n\n\n \n\nrow_vector operator/(row_vector b, matrix A) The right division of b by A; equivalently b * inverse(A)\nAvailable since 2.0\n \n\nmatrix operator/(matrix B, matrix A) The right division of B by A; equivalently B * inverse(A)\nAvailable since 2.5\n \n\nvector operator\\(matrix A, vector b) The left division of A by b; equivalently inverse(A) * b\nAvailable since 2.18\n \n\nmatrix operator\\(matrix A, matrix B) The left division of A by B; equivalently inverse(A) * B\nAvailable since 2.18\n\n\n\nThere are four division functions which use lower triangular views of a matrix. The lower triangular view of a matrix \\(\\text{tri}(A)\\) is used in the definitions and defined by \\[\\begin{equation*} \\text{tri}(A)[m,n] = \\left\\{\n\\begin{array}{ll} A[m,n] & \\text{if } m \\geq n, \\text{ and} \\\\[4pt] 0\n& \\text{otherwise}. \\end{array} \\right. \\end{equation*}\\] When a lower triangular view of a matrix is used, the elements above the diagonal are ignored.\n \n\nvector mdivide_left_tri_low(matrix A, vector b) The left division of b by a lower-triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * b, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_left_tri_low(matrix A, matrix B) The left division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form inverse(tri(A)) * B, where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n \n\nrow_vector mdivide_right_tri_low(row_vector b, matrix A) The right division of b by a triangular view of A; algebraically equivalent to the less efficient and stable form b * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.12\n \n\nmatrix mdivide_right_tri_low(matrix B, matrix A) The right division of B by a triangular view of A; algebraically equivalent to the less efficient and stable form B * inverse(tri(A)), where tri(A) is the lower-triangular portion of A with the above-diagonal entries set to zero.\nAvailable since 2.5\n\n\n\n\nThere are four division functions which are specialized for efficiency and stability for symmetric positive-definite matrix dividends. If the matrix dividend argument is not symmetric and positive definite, these will reject and print warnings.\n \n\nmatrix mdivide_left_spd(matrix A, vector b) The left division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * b.\nAvailable since 2.12\n \n\nvector mdivide_left_spd(matrix A, matrix B) The left division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form inverse(A) * B.\nAvailable since 2.12\n \n\nrow_vector mdivide_right_spd(row_vector b, matrix A) The right division of b by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form b *inverse(A).\nAvailable since 2.12\n \n\nmatrix mdivide_right_spd(matrix B, matrix A) The right division of B by the symmetric, positive-definite matrix A; algebraically equivalent to the less efficient and stable form B * inverse(A).\nAvailable since 2.12\n\n\n\nThe exponential of the matrix \\(A\\) is formally defined by the convergent power series: \\[\\begin{equation*} e^A = \\sum_{n=0}^{\\infty} \\dfrac{A^n}{n!}\n\\end{equation*}\\]\n \n\nmatrix matrix_exp(matrix A) The matrix exponential of A\nAvailable since 2.13\n \n\nmatrix matrix_exp_multiply(matrix A, matrix B) The multiplication of matrix exponential of A and matrix B; algebraically equivalent to the less efficient form matrix_exp(A) * B.\nAvailable since 2.18\n \n\nmatrix scale_matrix_exp_multiply(real t, matrix A, matrix B) The multiplication of matrix exponential of tA and matrix B; algebraically equivalent to the less efficient form matrix_exp(t * A) * B.\nAvailable since 2.18\n\n\n\nReturns the nth power of the specific matrix: \\[\\begin{equation*} M^n = M_1 * ... * M_n \\end{equation*}\\]\n \n\nmatrix matrix_power(matrix A, int B) Matrix A raised to the power B.\nAvailable since 2.24\n\n\n\n\n\n \n\nreal trace(matrix A) The trace of A, or 0 if A is empty; A is not required to be diagonal\nAvailable since 2.0\n\n\n\n \n\nreal determinant(matrix A) The determinant of A\nAvailable since 2.0\n \n\nreal log_determinant(matrix A) The log of the absolute value of the determinant of A\nAvailable since 2.0\nreal log_determinant_spd(matrix A) The log of the absolute value of the determinant of the symmetric, positive-definite matrix A.\nAvailable since 2.30\n\n\n\nIt is almost never a good idea to use matrix inverses directly because they are both inefficient and arithmetically unstable compared to the alternatives. Rather than inverting a matrix m and post-multiplying by a vector or matrix a, as in inverse(m) * a, it is better to code this using matrix division, as in m \\ a. The pre-multiplication case is similar, with b * inverse(m) being more efficiently coded as as b / m. There are also useful special cases for triangular and symmetric, positive-definite matrices that use more efficient solvers.\nWarning: The function inv(m) is the elementwise inverse function, which returns 1 / m[i, j] for each element.\n \n\nmatrix inverse(matrix A) Compute the inverse of A\nAvailable since 2.0\n \n\nmatrix inverse_spd(matrix A) Compute the inverse of A where A is symmetric, positive definite. This version is faster and more arithmetically stable when the input is symmetric and positive definite.\nAvailable since 2.0\n \n\nmatrix chol2inv(matrix L) Compute the inverse of the matrix whose cholesky factorization is L. That is, for \\(A = L L^T\\), return \\(A^{-1}\\).\nAvailable since 2.26\n\n\n\nThe generalized inverse \\(M^+\\) of a matrix \\(M\\) is a matrix that satisfies \\(M M^+ M = M\\). For an invertible, square matrix \\(M\\), \\(M^+\\) is equivalent to \\(M^{-1}\\). The dimensions of \\(M^+\\) are equivalent to the dimensions of \\(M^T\\). The generalized inverse exists for any matrix, so the \\(M\\) may be singular or less than full rank.\nEven though the generalized inverse exists for any arbitrary matrix, the derivatives of this function only exist on matrices of locally constant rank (Golub and Pereyra 1973), meaning, the derivatives do not exist if small perturbations make the matrix change rank. For example, considered the rank of the matrix \\(A\\) as a function of \\(\\epsilon\\):\n\\[\nA = \\left(\n    \\begin{array}{cccc}\n    1 + \\epsilon & 2 & 1 \\\\\n    2 & 4 & 2\n    \\end{array}\n    \\right)\n\\]\nWhen \\(\\epsilon = 0\\), \\(A\\) is rank 1 because the second row is twice the first (and so there is only one linearly independent row). If \\(\\epsilon \\neq 0\\), the rows are no longer linearly dependent, and the matrix is rank 2. This matrix does not have locally constant rank at \\(\\epsilon = 0\\), and so the derivatives do not exist at zero. Because HMC depends on the derivatives existing, this lack of differentiability creates undefined behavior.\n \n\nmatrix generalized_inverse(matrix A) The generalized inverse of A\nAvailable since 2.26\n\n\n\n \n\ncomplex_vector eigenvalues(matrix A) The complex-valued vector of eigenvalues of the matrix A. The eigenvalues are repeated according to their algebraic multiplicity, so there are as many eigenvalues as rows in the matrix. The eigenvalues are not sorted in any particular order.\nAvailable since 2.30\n \n\ncomplex_matrix eigenvectors(matrix A) The matrix with the complex-valued (column) eigenvectors of the matrix A in the same order as returned by the function eigenvalues\nAvailable since 2.30\n \n\ntuple(complex_matrix, complex_vector) eigendecompose(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the matrix A. This function is equivalent to (eigenvectors(A), eigenvalues(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nvector eigenvalues_sym(matrix A) The vector of eigenvalues of a symmetric matrix A in ascending order\nAvailable since 2.0\n \n\nmatrix eigenvectors_sym(matrix A) The matrix with the (column) eigenvectors of symmetric matrix A in the same order as returned by the function eigenvalues_sym\nAvailable since 2.0\n \n\ntuple(matrix, vector) eigendecompose_sym(matrix A) Return the matrix of (column) eigenvectors and vector of eigenvalues of the symmetric matrix A. This function is equivalent to (eigenvectors_sym(A), eigenvalues_sym(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nBecause multiplying an eigenvector by \\(-1\\) results in an eigenvector, eigenvectors returned by a decomposition are only identified up to a sign change. In order to compare the eigenvectors produced by Stan’s eigendecomposition to others, signs may need to be normalized in some way, such as by fixing the sign of a component, or doing comparisons allowing a multiplication by \\(-1\\).\nThe condition number of a symmetric matrix is defined to be the ratio of the largest eigenvalue to the smallest eigenvalue. Large condition numbers lead to difficulty in numerical algorithms such as computing inverses, and thus known as “ill conditioned.” The ratio can even be infinite in the case of singular matrices (i.e., those with eigenvalues of 0).\n\n\n\n \n\nmatrix qr_thin_Q(matrix A) The orthogonal matrix in the thin QR decomposition of A, which implies that the resulting matrix has the same dimensions as A\nAvailable since 2.18\n \n\nmatrix qr_thin_R(matrix A) The upper triangular matrix in the thin QR decomposition of A, which implies that the resulting matrix is square with the same number of columns as A\nAvailable since 2.18\n \n\ntuple(matrix, matrix) qr_thin(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthonormal matrix in the thin QR decomposition and the second element (“R”) is upper triangular. This function is equivalent to (qr_thin_Q(A), qr_thin_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\n \n\nmatrix qr_Q(matrix A) The orthogonal matrix in the fat QR decomposition of A, which implies that the resulting matrix is square with the same number of rows as A\nAvailable since 2.3\n \n\nmatrix qr_R(matrix A) The upper trapezoidal matrix in the fat QR decomposition of A, which implies that the resulting matrix will be rectangular with the same dimensions as A\nAvailable since 2.3\n \n tuple(matrix, matrix) qr(matrix A) Returns both portions of the QR decomposition of A. The first element (“Q”) is the orthogonal matrix in the fat QR decomposition and the second element (“R”) is upper trapezoidal. This function is equivalent to (qr_Q(A), qr_R(A)) but with a lower computational cost due to the shared work between the two results.\nAvailable since 2.33\nThe thin QR decomposition is always preferable because it will consume much less memory when the input matrix is large than will the fat QR decomposition. Both versions of the decomposition represent the input matrix as \\[\\begin{equation*} A = Q \\, R. \\end{equation*}\\] Multiplying a column of an orthogonal matrix by \\(-1\\) still results in an orthogonal matrix, and you can multiply the corresponding row of the upper trapezoidal matrix by \\(-1\\) without changing the product. Thus, Stan adopts the normalization that the diagonal elements of the upper trapezoidal matrix are strictly positive and the columns of the orthogonal matrix are reflected if necessary. Also, these QR decomposition algorithms do not utilize pivoting and thus may be numerically unstable on input matrices that have less than full rank.\n\n\n\nEvery symmetric, positive-definite matrix (such as a correlation or covariance matrix) has a Cholesky decomposition. If \\(\\Sigma\\) is a symmetric, positive-definite matrix, its Cholesky decomposition is the lower-triangular vector \\(L\\) such that \\[\\begin{equation*} \\Sigma = L \\, L^{\\top}. \\end{equation*}\\]\n \n\nmatrix cholesky_decompose(matrix A) The lower-triangular Cholesky factor of the symmetric positive-definite matrix A\nAvailable since 2.0\n\n\n\nThe matrix A can be decomposed into a diagonal matrix of singular values, D, and matrices of its left and right singular vectors, U and V, \\[\\begin{equation*} A = U D V^T. \\end{equation*}\\] The matrices of singular vectors here are thin. That is for an \\(N\\) by \\(P\\) input A, \\(M = min(N, P)\\), U is size \\(N\\) by \\(M\\) and V is size \\(P\\) by \\(M\\).\n \n\nvector singular_values(matrix A) The singular values of A in descending order\nAvailable since 2.0\n \n\nmatrix svd_U(matrix A) The left-singular vectors of A\nAvailable since 2.26\n \n\nmatrix svd_V(matrix A) The right-singular vectors of A\nAvailable since 2.26\n \n\ntuple(matrix, vector, matrix) svd(matrix A) Returns a tuple containing the left-singular vectors of A, the singular values of A in descending order, and the right-singular values of A. This function is equivalent to (svd_U(A), singular_values(A), svd_V(A)) but with a lower computational cost due to the shared work between the different components.\nAvailable since 2.33",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#sort-functions",
+    "href": "functions-reference/matrix_operations.html#sort-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "See the sorting functions section for examples of how the functions work.\n \n\nvector sort_asc(vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nrow_vector sort_asc(row_vector v) Sort the elements of v in ascending order\nAvailable since 2.0\n \n\nvector sort_desc(vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\nrow_vector sort_desc(row_vector v) Sort the elements of v in descending order\nAvailable since 2.0\n \n\narray[] int sort_indices_asc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_asc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in ascending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\narray[] int sort_indices_desc(row_vector v) Return an array of indices between 1 and the size of v, sorted to index v in descending order.\nAvailable since 2.3\n \n\nint rank(vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0\n \n\nint rank(row_vector v, int s) Number of components of v less than v[s]\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#reverse-functions",
+    "href": "functions-reference/matrix_operations.html#reverse-functions",
+    "title": "Matrix Operations",
+    "section": "",
+    "text": "vector reverse(vector v) Return a new vector containing the elements of the argument in reverse order.\nAvailable since 2.23\n \n\nrow_vector reverse(row_vector v) Return a new row vector containing the elements of the argument in reverse order.\nAvailable since 2.23",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/matrix_operations.html#footnotes",
+    "href": "functions-reference/matrix_operations.html#footnotes",
+    "title": "Matrix Operations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe softmax function is so called because in the limit as \\(y_n \\rightarrow \\infty\\) with \\(y_m\\) for \\(m \\neq n\\) held constant, the result tends toward the “one-hot” vector \\(\\theta\\) with \\(\\theta_n\n= 1\\) and \\(\\theta_m = 0\\) for \\(m \\neq n\\), thus providing a “soft” version of the maximum function.↩︎",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Matrix Operations"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html",
+    "href": "functions-reference/multivariate_discrete_distributions.html",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "The multivariate discrete distributions are over multiple integer values, which are expressed in Stan as arrays.\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\theta \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\text{Multinomial}(y|\\theta) = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K \\theta_k^{y_k},\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial(theta)\nIncrement target log probability density with multinomial_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multinomial_lpmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y)\nAvailable since 2.12\n \n\nreal multinomial_lupmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_rng(vector theta, int N) Generate a multinomial variate with simplex distribution parameter theta and total count \\(N\\); may only be used in transformed data and generated quantities blocks\nAvailable since 2.8\n\n\n\n\nStan also provides a version of the multinomial probability mass function distribution with the \\(\\text{$K$-simplex}\\) for the event count probabilities per category given on the unconstrained logistic scale.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\text{softmax}(\\theta) \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{MultinomialLogit}(y \\mid \\gamma) & = \\text{Multinomial}(y \\mid \\text{softmax}(\\gamma)) \\\\\n& = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K [\\text{softmax}(\\gamma_k)]^{y_k},\n\\end{split}\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial_logit(gamma)\nIncrement target log probability density with multinomial_logit_lupmf(y | gamma).\nAvailable since 2.24\n \n\n\n\n\n \n\nreal multinomial_logit_lpmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y)\nAvailable since 2.24\n \n\nreal multinomial_logit_lupmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_logit_rng(vector gamma, int N) Generate a variate from a multinomial distribution with probabilities softmax(gamma) and total count N; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.24\n\n\n\n\nStan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\alpha \\in\n\\mathbb{R}_{+}^K\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), the PMF of the Dirichlet-multinomial distribution is defined as \\[\\begin{equation*}\n\\text{DirMult}(y|\\theta) =\n\\frac{\\Gamma(\\alpha_0)\\Gamma(N+1)}{\\Gamma(N+\\alpha_0)} \\prod_{k=1}^K \\frac{\\Gamma(y_k + \\alpha_k)}{\\Gamma(\\alpha_k)\\Gamma(y_k+1)},\n\\end{equation*}\\] where \\(\\alpha_0\\) is defined as \\(\\alpha_0 = \\sum_{k=1}^K \\alpha_k\\).\n\n\n\ny ~ dirichlet_multinomial(alpha)\nIncrement target log probability density with dirichlet_multinomial_lupmf(y | alpha).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal dirichlet_multinomial_lpmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y).\nAvailable since 2.34\n \n\nreal dirichlet_multinomial_lupmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements, given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y) dropping constant additive terms.\nAvailable since 2.34\n \n\narray[] int dirichlet_multinomial_rng(vector alpha, int N) Generate a multinomial variate with positive vector distribution parameter alpha and total count N; may only be used in transformed data and generated quantities blocks. This is equivalent to multinomial_rng(dirichlet_rng(alpha), N).\nAvailable since 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution",
+    "href": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\theta \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\text{Multinomial}(y|\\theta) = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K \\theta_k^{y_k},\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial(theta)\nIncrement target log probability density with multinomial_lupmf(y | theta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal multinomial_lpmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y)\nAvailable since 2.12\n \n\nreal multinomial_lupmf(array[] int y | vector theta) The log multinomial probability mass function with outcome array y of size \\(K\\) given the \\(K\\)-simplex distribution parameter theta and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_rng(vector theta, int N) Generate a multinomial variate with simplex distribution parameter theta and total count \\(N\\); may only be used in transformed data and generated quantities blocks\nAvailable since 2.8",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution-logit-parameterization",
+    "href": "functions-reference/multivariate_discrete_distributions.html#multinomial-distribution-logit-parameterization",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides a version of the multinomial probability mass function distribution with the \\(\\text{$K$-simplex}\\) for the event count probabilities per category given on the unconstrained logistic scale.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\text{softmax}(\\theta) \\in \\text{$K$-simplex}\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), \\[\\begin{equation*}\n\\begin{split}\n\\text{MultinomialLogit}(y \\mid \\gamma) & = \\text{Multinomial}(y \\mid \\text{softmax}(\\gamma)) \\\\\n& = \\binom{N}{y_1,\\ldots,y_K} \\prod_{k=1}^K [\\text{softmax}(\\gamma_k)]^{y_k},\n\\end{split}\n\\end{equation*}\\] where the multinomial coefficient is defined by \\[\\begin{equation*}\n\\binom{N}{y_1,\\ldots,y_k} = \\frac{N!}{\\prod_{k=1}^K y_k!}.\n\\end{equation*}\\]\n\n\n\ny ~ multinomial_logit(gamma)\nIncrement target log probability density with multinomial_logit_lupmf(y | gamma).\nAvailable since 2.24\n \n\n\n\n\n \n\nreal multinomial_logit_lpmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y)\nAvailable since 2.24\n \n\nreal multinomial_logit_lupmf(array[] int y | vector gamma) The log multinomial probability mass function with outcome array y of size \\(K\\) given the log \\(K\\)-simplex distribution parameter \\(\\gamma\\) and (implicit) total count N = sum(y) dropping constant additive terms\nAvailable since 2.25\n \n\narray[] int multinomial_logit_rng(vector gamma, int N) Generate a variate from a multinomial distribution with probabilities softmax(gamma) and total count N; may only be used in transformed data and generated quantities blocks.\nAvailable since 2.24",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/multivariate_discrete_distributions.html#dirichlet-multinomial-distribution",
+    "href": "functions-reference/multivariate_discrete_distributions.html#dirichlet-multinomial-distribution",
+    "title": "Multivariate Discrete Distributions",
+    "section": "",
+    "text": "Stan also provides the Dirichlet-multinomial distribution, which generalizes the Beta-binomial distribution to more than two categories. As such, it is an overdispersed version of the multinomial distribution.\n\n\nIf \\(K \\in \\mathbb{N}\\), \\(N \\in \\mathbb{N}\\), and \\(\\alpha \\in\n\\mathbb{R}_{+}^K\\), then for \\(y \\in \\mathbb{N}^K\\) such that \\(\\sum_{k=1}^K y_k = N\\), the PMF of the Dirichlet-multinomial distribution is defined as \\[\\begin{equation*}\n\\text{DirMult}(y|\\theta) =\n\\frac{\\Gamma(\\alpha_0)\\Gamma(N+1)}{\\Gamma(N+\\alpha_0)} \\prod_{k=1}^K \\frac{\\Gamma(y_k + \\alpha_k)}{\\Gamma(\\alpha_k)\\Gamma(y_k+1)},\n\\end{equation*}\\] where \\(\\alpha_0\\) is defined as \\(\\alpha_0 = \\sum_{k=1}^K \\alpha_k\\).\n\n\n\ny ~ dirichlet_multinomial(alpha)\nIncrement target log probability density with dirichlet_multinomial_lupmf(y | alpha).\nAvailable since 2.34\n \n\n\n\n\n \n\nreal dirichlet_multinomial_lpmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y).\nAvailable since 2.34\n \n\nreal dirichlet_multinomial_lupmf(array[] int y | vector alpha) The log multinomial probability mass function with outcome array y with \\(K\\) elements, given the positive \\(K\\)-vector distribution parameter alpha and (implicit) total count N = sum(y) dropping constant additive terms.\nAvailable since 2.34\n \n\narray[] int dirichlet_multinomial_rng(vector alpha, int N) Generate a multinomial variate with positive vector distribution parameter alpha and total count N; may only be used in transformed data and generated quantities blocks. This is equivalent to multinomial_rng(dirichlet_rng(alpha), N).\nAvailable since 2.34",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Multivariate Discrete Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html",
+    "href": "functions-reference/positive_lower-bounded_distributions.html",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "The positive lower-bounded probabilities have support on real values above some positive minimum value.\n\n\n\n\nIf \\(y_{\\text{min}} \\in \\mathbb{R}^+\\) and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\) with \\(y \\geq y_{\\text{min}}\\), \\[\\begin{equation*}\n\\text{Pareto}(y|y_{\\text{min}},\\alpha) = \\frac{\\displaystyle \\alpha\\,y_{\\text{min}}^\\alpha}{\\displaystyle y^{\\alpha+1}}.\n\\end{equation*}\\]\n\n\n\ny ~ pareto(y_min, alpha)\nIncrement target log probability density with pareto_lupdf(y | y_min, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal pareto_lpdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lupdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_cdf(reals y | reals y_min, reals alpha) The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.0\n \n\nreal pareto_lcdf(reals y | reals y_min, reals alpha) The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lccdf(reals y | reals y_min, reals alpha) The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nR pareto_rng(reals y_min, reals alpha) Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\n\n\nIf \\(\\mu \\in \\mathbb{R}\\), \\(\\lambda \\in \\mathbb{R}^+\\), and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\geq \\mu\\), \\[\\begin{equation*}\n\\mathrm{Pareto\\_Type\\_2}(y|\\mu,\\lambda,\\alpha) =\n\\ \\frac{\\alpha}{\\lambda} \\, \\left( 1+\\frac{y-\\mu}{\\lambda} \\right)^{-(\\alpha+1)} \\! .\n\\end{equation*}\\]\nNote that the Lomax distribution is a Pareto Type 2 distribution with \\(\\mu=0\\).\n\n\n\ny ~ pareto_type_2(mu, lambda, alpha)\nIncrement target log probability density with pareto_type_2_lupdf(y | mu, lambda, alpha).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal pareto_type_2_lpdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lupdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_type_2_cdf(reals y | reals mu, reals lambda, reals alpha) The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.5\n \n\nreal pareto_type_2_lcdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lccdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nR pareto_type_2_rng(reals mu, reals lambda, reals alpha) Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18\n\n\n\n\nFor an extended explanation of how to use the wiener_lpdf and wiener_l[c]cdf_unnorm functions, see Henrich et al. (2024).\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), \\(\\tau \\in \\mathbb{R}^+\\), \\(\\beta \\in (0, 1)\\), \\(\\delta \\in \\mathbb{R}\\), \\(s_{\\delta} \\in \\mathbb{R}^{\\geq 0}\\), \\(s_{\\beta} \\in [0, 1)\\), and \\(s_{\\tau} \\in \\mathbb{R}^{\\geq 0}\\) then for \\(y &gt; \\tau\\),\n\\[\\begin{equation*}\n\\begin{split}\n&\\text{Wiener}(y\\mid \\alpha,\\tau,\\beta,\\delta,s_{\\delta},s_{\\beta},s_{\\tau}) =\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}}\\int_{-\\infty}^{\\infty} p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega)\n\\\\\n&\\times \\frac{1}{\\sqrt{2\\pi s_{\\delta}^2}}\\exp\\Bigl(-\\frac{(\\nu-\\delta)^2}{2s_{\\delta}^2}\\Bigr) \\,d\\nu \\,d\\omega \\,d{\\tau_0}=\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}} M\\times p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega) \\,d\\omega \\,d{\\tau_0},\n\\end{split}\n\\end{equation*}\\]\nwhere \\(p()\\) denotes the density function, and \\(M\\) and \\(p_3()\\) are defined, by using \\(t:=y-{\\tau_0}\\), as\n\\[\\begin{equation*}\nM \\coloneqq \\frac{1}{\\sqrt{1+s_{\\delta}^2t}}\\exp\\Bigl(\\alpha{\\delta}\\omega+\\frac{\\delta^2t}{2}+\\frac{s_{\\delta}^2\\alpha^2\\omega^2-2\\alpha{\\delta}\\omega-\\delta^2t}{2(1+s_{\\delta}^2t)}\\Bigr)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\np_3(t\\mid \\alpha,\\delta,\\beta) \\coloneqq \\frac{1}{\\alpha^2}\\exp\\Bigl(-\\alpha\\delta\\beta-\\frac{\\delta^2t}{2}\\Bigr)f(\\frac{t}{\\alpha^2}\\mid 0,1,\\beta),\n\\end{equation*}\\]\nwhere \\(f(t^*=\\frac{t}{\\alpha^2}\\mid0,1,\\beta)\\) can be specified in two ways:\n\\[\\begin{equation*}\nf_l(t^*\\mid 0,1,\\beta) = \\sum_{k=1}^\\infty k\\pi \\exp\\Bigl(-\\frac{k^2\\pi^2t^*}{2}\\Bigr)\\sin(k\\pi \\beta)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\nf_s(t^*\\mid0,1,\\beta) = \\sum_{k=-\\infty}^\\infty \\frac{1}{\\sqrt{2\\pi(t^*)^3}}(\\beta+2k) \\exp\\Bigl(-\\frac{(\\beta+2k)^2}{2t^*}\\Bigr).\n\\end{equation*}\\]\nWhich of these is used in the computations depends on which expression requires the smaller number of components \\(k\\) to guarantee a pre-specified precision\nIn the case where \\(s_{\\delta}\\), \\(s_{\\beta}\\), and \\(s_{\\tau}\\) are all \\(0\\), this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\alpha}{(y-\\tau)^{3/2}} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = - \\infty}^{\\infty} (2k +\n\\beta) \\phi \\! \\left(\\frac{(2k + \\beta)\\alpha }{\\sqrt{y - \\tau}}\\right),\n\\end{equation*}\\] where \\(\\phi(x)\\) denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\pi}{\\alpha^2} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = 1}^{\\infty} k \\exp \\! \\left(-\\frac{k^2\\pi^2(y-\\tau)};\n{2\\alpha^2}\\right) \\sin \\!(k\\pi\\beta)\n\\end{equation*}\\] see (Feller 1968), (Navarro and Fuss 2009).\n\n\n\nFor the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.\nLet \\(\\alpha\\), \\(\\tau\\), \\(\\beta\\), \\(\\delta\\), \\(s_{\\delta}\\), \\(s_{\\beta}\\), \\(s_{\\tau}\\) and \\(y\\) be as above.\nThe formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in \\(\\tau\\) for the time being) at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha, \\beta, \\delta) = P(\\alpha, \\beta, \\delta) -\n    \\exp\\left(\\delta\\alpha(1-\\beta)-\\frac{\\delta^2 y}{2}\\right)F_l(y|\\alpha,\\beta,\\delta),\n\\end{equation}\\] where \\(P(\\alpha,\\beta,\\delta)\\) is the probability to hit the upper boundary, defined as\n\\[\\begin{equation}\nP(\\alpha, \\beta, \\delta) =\n\\begin{cases}\n    \\frac{1-\\exp(2\\delta \\alpha \\beta)}{\\exp(-2\\delta \\alpha(1-\\beta)) - \\exp(2\\delta \\alpha \\beta)}, & \\text{for } \\delta\\neq 0 \\\\\n    \\beta, & \\text{for } \\delta=0,\n\\end{cases}\n\\end{equation}\\]\nand\n\\[\\begin{equation}\nF_l(y|\\alpha, \\beta, \\delta) =\n    \\frac{2\\pi}{\\alpha^2}\\sum_{k=1}^{\\infty}{\\frac{k\\sin{k\\pi(1-\\beta)}}{\\delta^2+(k\\pi)^2/\\alpha^2}}\\exp(-\\frac{k^2\\pi^2y}{2\\alpha^2}).\n\\end{equation}\\]\nThe formula for the small-time cdf at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha,\\beta,\\delta) = \\exp\\left(\\delta \\alpha(1-\\beta) -\\frac{\\delta^2y}{2}\\right)F_s(y|\\alpha, \\beta,\\delta),\n\\end{equation}\\] where\n\\[\\begin{equation}\nF_s(y|\\alpha,\\beta,\\delta) = \\sum_{k=0}^{\\infty}(-1)^k\\phi\\left(\\frac{\\alpha(k+\\beta^{*}_k)}\n    {\\sqrt{y}} \\right) \\times \\left( R \\left(\\frac{\\alpha(k+\\beta^{*}_k)+\\delta y}{\\sqrt{y}} \\right)\n        + R \\left(\\frac{\\alpha(k+\\beta^*_k)-\\delta y}{\\sqrt{y}} \\right)\\right),\n\\end{equation}\\]\nwhere \\(\\beta^*_k=(1-\\beta)\\) for \\(k\\) even, \\(\\beta^*_k=\\beta\\) for \\(k\\) odd, and \\(R\\) is Mill’s ratio.\nThe cdf for the lower boundary is \\(F(y|\\alpha,1-\\beta,-\\delta)\\)\n\n\n\ny ~ wiener(alpha, tau, beta, delta)\nIncrement target log probability density with wiener_lupdf(y | alpha, tau, beta, delta).\nAvailable since 2.7\ny ~ wiener(alpha, tau, beta, delta, var_delta) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta).\nAvailable since 2.35\ny ~ wiener(alpha, tau, beta, delta, var_delta, var_beta, var_tau) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau).\nAvailable since 2.35\n \n\n\n\n\n \n\nreal wiener_lpdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.18\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta, dropping constant additive terms\nAvailable since 2.25\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta, dropping constant additive terms.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau, dropping constant additive terms.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n\nNote: The lcdf and lccdf functions for the wiener distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper boundary.\nSimilarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical \\(1 - \\textrm{cdf}\\).\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n\n\n\nStan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use \\[\\begin{equation*}\n\\text{Wiener}(y | \\alpha, \\tau, 1 - \\beta, - \\delta)\n\\end{equation*}\\] For more details, see the appendix of Vandekerckhove and Wabersich (2014).\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions (listed above as recieving only real) are implemented in such a way where they can be fully vectorized, but currently only versions that accept all real and all vector arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by opening an issue.\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions can also accept an additional data real argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to 1e-4 for the density and 1e-8 for the cdf functions.",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html#pareto-distribution",
+    "href": "functions-reference/positive_lower-bounded_distributions.html#pareto-distribution",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "If \\(y_{\\text{min}} \\in \\mathbb{R}^+\\) and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{R}^+\\) with \\(y \\geq y_{\\text{min}}\\), \\[\\begin{equation*}\n\\text{Pareto}(y|y_{\\text{min}},\\alpha) = \\frac{\\displaystyle \\alpha\\,y_{\\text{min}}^\\alpha}{\\displaystyle y^{\\alpha+1}}.\n\\end{equation*}\\]\n\n\n\ny ~ pareto(y_min, alpha)\nIncrement target log probability density with pareto_lupdf(y | y_min, alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal pareto_lpdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lupdf(reals y | reals y_min, reals alpha) The log of the Pareto density of y given positive minimum value y_min and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_cdf(reals y | reals y_min, reals alpha) The Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.0\n \n\nreal pareto_lcdf(reals y | reals y_min, reals alpha) The log of the Pareto cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nreal pareto_lccdf(reals y | reals y_min, reals alpha) The log of the Pareto complementary cumulative distribution function of y given positive minimum value y_min and shape alpha\nAvailable since 2.12\n \n\nR pareto_rng(reals y_min, reals alpha) Generate a Pareto variate with positive minimum value y_min and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html#pareto-type-2-distribution",
+    "href": "functions-reference/positive_lower-bounded_distributions.html#pareto-type-2-distribution",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "If \\(\\mu \\in \\mathbb{R}\\), \\(\\lambda \\in \\mathbb{R}^+\\), and \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(y \\geq \\mu\\), \\[\\begin{equation*}\n\\mathrm{Pareto\\_Type\\_2}(y|\\mu,\\lambda,\\alpha) =\n\\ \\frac{\\alpha}{\\lambda} \\, \\left( 1+\\frac{y-\\mu}{\\lambda} \\right)^{-(\\alpha+1)} \\! .\n\\end{equation*}\\]\nNote that the Lomax distribution is a Pareto Type 2 distribution with \\(\\mu=0\\).\n\n\n\ny ~ pareto_type_2(mu, lambda, alpha)\nIncrement target log probability density with pareto_type_2_lupdf(y | mu, lambda, alpha).\nAvailable since 2.5\n \n\n\n\n\n \n\nreal pareto_type_2_lpdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lupdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 density of y given location mu, scale lambda, and shape alpha dropping constant additive terms\nAvailable since 2.25\n \n\nreal pareto_type_2_cdf(reals y | reals mu, reals lambda, reals alpha) The Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.5\n \n\nreal pareto_type_2_lcdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nreal pareto_type_2_lccdf(reals y | reals mu, reals lambda, reals alpha) The log of the Pareto Type 2 complementary cumulative distribution function of y given location mu, scale lambda, and shape alpha\nAvailable since 2.18\n \n\nR pareto_type_2_rng(reals mu, reals lambda, reals alpha) Generate a Pareto Type 2 variate with location mu, scale lambda, and shape alpha; may only be used in transformed data and generated quantities blocks. For a description of argument and return types, see section vectorized PRNG functions.\nAvailable since 2.18",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/positive_lower-bounded_distributions.html#wiener-first-passage-time-distribution",
+    "href": "functions-reference/positive_lower-bounded_distributions.html#wiener-first-passage-time-distribution",
+    "title": "Positive Lower-Bounded Distributions",
+    "section": "",
+    "text": "For an extended explanation of how to use the wiener_lpdf and wiener_l[c]cdf_unnorm functions, see Henrich et al. (2024).\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), \\(\\tau \\in \\mathbb{R}^+\\), \\(\\beta \\in (0, 1)\\), \\(\\delta \\in \\mathbb{R}\\), \\(s_{\\delta} \\in \\mathbb{R}^{\\geq 0}\\), \\(s_{\\beta} \\in [0, 1)\\), and \\(s_{\\tau} \\in \\mathbb{R}^{\\geq 0}\\) then for \\(y &gt; \\tau\\),\n\\[\\begin{equation*}\n\\begin{split}\n&\\text{Wiener}(y\\mid \\alpha,\\tau,\\beta,\\delta,s_{\\delta},s_{\\beta},s_{\\tau}) =\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}}\\int_{-\\infty}^{\\infty} p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega)\n\\\\\n&\\times \\frac{1}{\\sqrt{2\\pi s_{\\delta}^2}}\\exp\\Bigl(-\\frac{(\\nu-\\delta)^2}{2s_{\\delta}^2}\\Bigr) \\,d\\nu \\,d\\omega \\,d{\\tau_0}=\n\\\\\n&\\frac{1}{s_{\\tau}}\\int_{\\tau}^{\\tau+s_{\\tau}}\\frac{1}{s_{\\beta}}\\int_{\\beta-\\frac{1}{2}s_{\\beta}}^{\\beta+\\frac{1}{2}s_{\\beta}} M\\times p_3(y-{\\tau_0}\\mid \\alpha,\\nu,\\omega) \\,d\\omega \\,d{\\tau_0},\n\\end{split}\n\\end{equation*}\\]\nwhere \\(p()\\) denotes the density function, and \\(M\\) and \\(p_3()\\) are defined, by using \\(t:=y-{\\tau_0}\\), as\n\\[\\begin{equation*}\nM \\coloneqq \\frac{1}{\\sqrt{1+s_{\\delta}^2t}}\\exp\\Bigl(\\alpha{\\delta}\\omega+\\frac{\\delta^2t}{2}+\\frac{s_{\\delta}^2\\alpha^2\\omega^2-2\\alpha{\\delta}\\omega-\\delta^2t}{2(1+s_{\\delta}^2t)}\\Bigr)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\np_3(t\\mid \\alpha,\\delta,\\beta) \\coloneqq \\frac{1}{\\alpha^2}\\exp\\Bigl(-\\alpha\\delta\\beta-\\frac{\\delta^2t}{2}\\Bigr)f(\\frac{t}{\\alpha^2}\\mid 0,1,\\beta),\n\\end{equation*}\\]\nwhere \\(f(t^*=\\frac{t}{\\alpha^2}\\mid0,1,\\beta)\\) can be specified in two ways:\n\\[\\begin{equation*}\nf_l(t^*\\mid 0,1,\\beta) = \\sum_{k=1}^\\infty k\\pi \\exp\\Bigl(-\\frac{k^2\\pi^2t^*}{2}\\Bigr)\\sin(k\\pi \\beta)\\text{ and}\n\\end{equation*}\\]\n\\[\\begin{equation*}\nf_s(t^*\\mid0,1,\\beta) = \\sum_{k=-\\infty}^\\infty \\frac{1}{\\sqrt{2\\pi(t^*)^3}}(\\beta+2k) \\exp\\Bigl(-\\frac{(\\beta+2k)^2}{2t^*}\\Bigr).\n\\end{equation*}\\]\nWhich of these is used in the computations depends on which expression requires the smaller number of components \\(k\\) to guarantee a pre-specified precision\nIn the case where \\(s_{\\delta}\\), \\(s_{\\beta}\\), and \\(s_{\\tau}\\) are all \\(0\\), this simplifies to one representation that converges fast for small reaction-time values (“small time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\alpha}{(y-\\tau)^{3/2}} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = - \\infty}^{\\infty} (2k +\n\\beta) \\phi \\! \\left(\\frac{(2k + \\beta)\\alpha }{\\sqrt{y - \\tau}}\\right),\n\\end{equation*}\\] where \\(\\phi(x)\\) denotes the standard normal density function, and one representation that converges fast for large reaction-time values (“large time expansion”): \\[\\begin{equation*}\n\\text{Wiener}(y|\\alpha, \\tau, \\beta, \\delta) =\n\\frac{\\pi}{\\alpha^2} \\exp \\! \\left(- \\delta \\alpha \\beta -\n\\frac{\\delta^2(y-\\tau)}{2}\\right) \\sum_{k = 1}^{\\infty} k \\exp \\! \\left(-\\frac{k^2\\pi^2(y-\\tau)};\n{2\\alpha^2}\\right) \\sin \\!(k\\pi\\beta)\n\\end{equation*}\\] see (Feller 1968), (Navarro and Fuss 2009).\n\n\n\nFor the cumulative distribution function (cdf) there also exist two expressions depending on the reaction time.\nLet \\(\\alpha\\), \\(\\tau\\), \\(\\beta\\), \\(\\delta\\), \\(s_{\\delta}\\), \\(s_{\\beta}\\), \\(s_{\\tau}\\) and \\(y\\) be as above.\nThe formula for the large-time cdf of decision times (excluding the additive reaction time components summarized in \\(\\tau\\) for the time being) at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha, \\beta, \\delta) = P(\\alpha, \\beta, \\delta) -\n    \\exp\\left(\\delta\\alpha(1-\\beta)-\\frac{\\delta^2 y}{2}\\right)F_l(y|\\alpha,\\beta,\\delta),\n\\end{equation}\\] where \\(P(\\alpha,\\beta,\\delta)\\) is the probability to hit the upper boundary, defined as\n\\[\\begin{equation}\nP(\\alpha, \\beta, \\delta) =\n\\begin{cases}\n    \\frac{1-\\exp(2\\delta \\alpha \\beta)}{\\exp(-2\\delta \\alpha(1-\\beta)) - \\exp(2\\delta \\alpha \\beta)}, & \\text{for } \\delta\\neq 0 \\\\\n    \\beta, & \\text{for } \\delta=0,\n\\end{cases}\n\\end{equation}\\]\nand\n\\[\\begin{equation}\nF_l(y|\\alpha, \\beta, \\delta) =\n    \\frac{2\\pi}{\\alpha^2}\\sum_{k=1}^{\\infty}{\\frac{k\\sin{k\\pi(1-\\beta)}}{\\delta^2+(k\\pi)^2/\\alpha^2}}\\exp(-\\frac{k^2\\pi^2y}{2\\alpha^2}).\n\\end{equation}\\]\nThe formula for the small-time cdf at the upper boundary is stated as follows:\n\\[\\begin{equation}\nF(y|\\alpha,\\beta,\\delta) = \\exp\\left(\\delta \\alpha(1-\\beta) -\\frac{\\delta^2y}{2}\\right)F_s(y|\\alpha, \\beta,\\delta),\n\\end{equation}\\] where\n\\[\\begin{equation}\nF_s(y|\\alpha,\\beta,\\delta) = \\sum_{k=0}^{\\infty}(-1)^k\\phi\\left(\\frac{\\alpha(k+\\beta^{*}_k)}\n    {\\sqrt{y}} \\right) \\times \\left( R \\left(\\frac{\\alpha(k+\\beta^{*}_k)+\\delta y}{\\sqrt{y}} \\right)\n        + R \\left(\\frac{\\alpha(k+\\beta^*_k)-\\delta y}{\\sqrt{y}} \\right)\\right),\n\\end{equation}\\]\nwhere \\(\\beta^*_k=(1-\\beta)\\) for \\(k\\) even, \\(\\beta^*_k=\\beta\\) for \\(k\\) odd, and \\(R\\) is Mill’s ratio.\nThe cdf for the lower boundary is \\(F(y|\\alpha,1-\\beta,-\\delta)\\)\n\n\n\ny ~ wiener(alpha, tau, beta, delta)\nIncrement target log probability density with wiener_lupdf(y | alpha, tau, beta, delta).\nAvailable since 2.7\ny ~ wiener(alpha, tau, beta, delta, var_delta) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta).\nAvailable since 2.35\ny ~ wiener(alpha, tau, beta, delta, var_delta, var_beta, var_tau) Increment target log probability density with wiener_lupdf(y | alpha, tau, beta, delta, var_delta, var_beta, var_tau).\nAvailable since 2.35\n \n\n\n\n\n \n\nreal wiener_lpdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.18\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lpdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(reals y | reals alpha, reals tau, reals beta, reals delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta, dropping constant additive terms\nAvailable since 2.25\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, and inter-trial drift rate variability var_delta, dropping constant additive terms.\nSetting var_delta to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n \n\nreal wiener_lupdf(real y | real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau) The log of the Wiener first passage time density of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau, dropping constant additive terms.\nSetting var_delta, var_beta, and var_tau to 0 recovers the 4-parameter signature above.\nAvailable since 2.35\n\nNote: The lcdf and lccdf functions for the wiener distribution are conditional and unnormalized, meaning that the cdf does not asymptote at 1, but rather at the probability to hit the upper boundary.\nSimilarly, the ccdf is defined as the probability to hit the upper boundary less the value of the cdf, as opposed to the more typical \\(1 - \\textrm{cdf}\\).\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, and drift rate delta.\nAvailable since 2.38\n \n\nreal wiener_lcdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the cumulative distribution function (cdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n \n\nreal wiener_lccdf_unnorm(real y, real alpha, real tau, real beta, real delta, real var_delta, real var_beta, real var_tau)\nThe log of the complementary cumulative distribution function (ccdf) of the Wiener distribution of y given boundary separation alpha, non-decision time tau, starting point beta, drift rate delta, inter-trial drift rate variability var_delta, inter-trial variability of the starting point (bias) var_beta, and inter-trial variability of the non-decision time var_tau.\nAvailable since 2.38\n\n\n\nStan returns the first passage time of the accumulation process over the upper boundary only. To get the result for the lower boundary, use \\[\\begin{equation*}\n\\text{Wiener}(y | \\alpha, \\tau, 1 - \\beta, - \\delta)\n\\end{equation*}\\] For more details, see the appendix of Vandekerckhove and Wabersich (2014).\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions (listed above as recieving only real) are implemented in such a way where they can be fully vectorized, but currently only versions that accept all real and all vector arguments are exposed by Stan. If there are additional signatures that would prove useful, please request them by opening an issue.\n\n\n\nThe 5- and 7-argument forms of the wiener distribution functions can also accept an additional data real argument controlling the required precision of the gradient calculation of the function. If omitted, this defaults to 1e-4 for the density and 1e-8 for the cdf functions.",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Positive Lower-Bounded Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/references.html",
+    "href": "functions-reference/references.html",
+    "title": "References",
+    "section": "",
+    "text": "References\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "functions-reference/simplex_distributions.html",
+    "href": "functions-reference/simplex_distributions.html",
+    "title": "Simplex Distributions",
+    "section": "",
+    "text": "The simplex probabilities have support on the unit \\(K\\)-simplex for a specified \\(K\\). A \\(K\\)-dimensional vector \\(\\theta\\) is a unit \\(K\\)-simplex if \\(\\theta_k \\geq 0\\) for \\(k \\in \\{1,\\ldots,K\\}\\) and \\(\\sum_{k = 1}^K \\theta_k = 1\\).\n\n\n\n\nIf \\(K \\in \\mathbb{N}\\) and \\(\\alpha \\in (\\mathbb{R}^+)^{K}\\), then for \\(\\theta \\in \\text{$K$-simplex}\\),\n\\[\\begin{equation*}\n\\text{Dirichlet}(\\theta|\\alpha) =\n\\frac{\\Gamma \\! \\left( \\sum_{k=1}^K \\alpha_k \\right)}{\\prod_{k=1}^K \\Gamma(\\alpha_k)}  \\prod_{k=1}^K \\theta_k^{\\alpha_k -1}\n\\end{equation*}\\]\nWarning: If any of the components of \\(\\theta\\) satisfies \\(\\theta_i = 0\\) or \\(\\theta_i = 1\\), then the probability is \\(0\\) and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, with \\(\\alpha_i &gt; 0\\) for each \\(i\\).\n\n\n\nA symmetric Dirichlet prior is \\([\\alpha, \\ldots, \\alpha]^{\\top}\\). To code this in Stan,\n data {\n   int&lt;lower=1&gt; K;\n   real&lt;lower=0&gt; alpha;\n }\n generated quantities {\n   vector[K] theta = dirichlet_rng(rep_vector(alpha, K));\n }\nTaking \\(K = 10\\), here are the first five draws for \\(\\alpha = 1\\). For \\(\\alpha = 1\\), the distribution is uniform over simplexes.\n 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05\n 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06\n 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03\n 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18\n 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01\nThat does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) \\(1/K\\). When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on \\(10\\)-simplexes, the probability that a component is greater than the mean of \\(1/10\\) is only 39%. Most of the posterior marginal probability mass for each component is in the interval \\((0, 0.1)\\).\nWhen the \\(\\alpha\\) value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for \\(\\alpha = 0.001\\).\n 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279\n 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000\n 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112\n 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000\n 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000\nEach row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.\nAs \\(\\alpha\\) increases, the draws become increasingly uniform. For \\(\\alpha = 1000\\),\n 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10\n 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n\n\n\ntheta ~ dirichlet(alpha)\nIncrement target log probability density with dirichlet_lupdf(theta | alpha).\nAvailable since 2.0\n \n\n\n\n\nThe Dirichlet probability functions are overloaded to allow the simplex \\(\\theta\\) and prior counts (plus one) \\(\\alpha\\) to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal dirichlet_lpdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha\nAvailable since 2.12, vectorized in 2.21\n \n\nreal dirichlet_lupdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha dropping constant additive terms\nAvailable since 2.25\n \n\nvector dirichlet_rng(vector alpha) Generate a Dirichlet variate with prior counts (plus one) alpha; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Simplex Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/simplex_distributions.html#dirichlet-distribution",
+    "href": "functions-reference/simplex_distributions.html#dirichlet-distribution",
+    "title": "Simplex Distributions",
+    "section": "",
+    "text": "If \\(K \\in \\mathbb{N}\\) and \\(\\alpha \\in (\\mathbb{R}^+)^{K}\\), then for \\(\\theta \\in \\text{$K$-simplex}\\),\n\\[\\begin{equation*}\n\\text{Dirichlet}(\\theta|\\alpha) =\n\\frac{\\Gamma \\! \\left( \\sum_{k=1}^K \\alpha_k \\right)}{\\prod_{k=1}^K \\Gamma(\\alpha_k)}  \\prod_{k=1}^K \\theta_k^{\\alpha_k -1}\n\\end{equation*}\\]\nWarning: If any of the components of \\(\\theta\\) satisfies \\(\\theta_i = 0\\) or \\(\\theta_i = 1\\), then the probability is \\(0\\) and the log probability is \\(-\\infty\\). Similarly, the distribution requires strictly positive parameters, with \\(\\alpha_i &gt; 0\\) for each \\(i\\).\n\n\n\nA symmetric Dirichlet prior is \\([\\alpha, \\ldots, \\alpha]^{\\top}\\). To code this in Stan,\n data {\n   int&lt;lower=1&gt; K;\n   real&lt;lower=0&gt; alpha;\n }\n generated quantities {\n   vector[K] theta = dirichlet_rng(rep_vector(alpha, K));\n }\nTaking \\(K = 10\\), here are the first five draws for \\(\\alpha = 1\\). For \\(\\alpha = 1\\), the distribution is uniform over simplexes.\n 1) 0.17 0.05 0.07 0.17 0.03 0.13 0.03 0.03 0.27 0.05\n 2) 0.08 0.02 0.12 0.07 0.52 0.01 0.07 0.04 0.01 0.06\n 3) 0.02 0.03 0.22 0.29 0.17 0.10 0.09 0.00 0.05 0.03\n 4) 0.04 0.03 0.21 0.13 0.04 0.01 0.10 0.04 0.22 0.18\n 5) 0.11 0.22 0.02 0.01 0.06 0.18 0.33 0.04 0.01 0.01\nThat does not mean it’s uniform over the marginal probabilities of each element. As the size of the simplex grows, the marginal draws become more and more concentrated below (not around) \\(1/K\\). When one component of the simplex is large, the others must all be relatively small to compensate. For example, in a uniform distribution on \\(10\\)-simplexes, the probability that a component is greater than the mean of \\(1/10\\) is only 39%. Most of the posterior marginal probability mass for each component is in the interval \\((0, 0.1)\\).\nWhen the \\(\\alpha\\) value is small, the draws gravitate to the corners of the simplex. Here are the first five draws for \\(\\alpha = 0.001\\).\n 1) 3e-203 0e+00 2e-298 9e-106 1e+000 0e+00 0e+000 1e-047 0e+00 4e-279\n 2) 1e+000 0e+00 5e-279 2e-014 1e-275 0e+00 3e-285 9e-147 0e+00 0e+000\n 3) 1e-308 0e+00 1e-213 0e+000 0e+000 8e-75 0e+000 1e+000 4e-58 7e-112\n 4) 6e-166 5e-65 3e-068 3e-147 0e+000 1e+00 3e-249 0e+000 0e+00 0e+000\n 5) 2e-091 0e+00 0e+000 0e+000 1e-060 0e+00 4e-312 1e+000 0e+00 0e+000\nEach row denotes a draw. Each draw has a single value that rounds to one and other values that are very close to zero or rounded down to zero.\nAs \\(\\alpha\\) increases, the draws become increasingly uniform. For \\(\\alpha = 1000\\),\n 1) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 2) 0.10 0.10 0.09 0.10 0.10 0.10 0.11 0.10 0.10 0.10\n 3) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 4) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n 5) 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10\n\n\n\ntheta ~ dirichlet(alpha)\nIncrement target log probability density with dirichlet_lupdf(theta | alpha).\nAvailable since 2.0\n \n\n\n\n\nThe Dirichlet probability functions are overloaded to allow the simplex \\(\\theta\\) and prior counts (plus one) \\(\\alpha\\) to be vectors or row vectors (or to mix the two types). The density functions are also vectorized, so they allow arrays of row vectors or vectors as arguments; see section vectorized function signatures for a description of vectorization.\n \n\nreal dirichlet_lpdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha\nAvailable since 2.12, vectorized in 2.21\n \n\nreal dirichlet_lupdf(vectors theta | vectors alpha) The log of the Dirichlet density for simplex(es) theta given prior counts (plus one) alpha dropping constant additive terms\nAvailable since 2.25\n \n\nvector dirichlet_rng(vector alpha) Generate a Dirichlet variate with prior counts (plus one) alpha; may only be used in transformed data and generated quantities blocks\nAvailable since 2.0",
+    "crumbs": [
+      "Functions Reference",
+      "Continuous Distributions",
+      "Simplex Distributions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html",
+    "href": "functions-reference/transform_functions.html",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "Variable transformation functions provide implementations of the built-in constraining and unconstraining transforms defined in Stan Reference Manual.\nFor each of the built-in variable transforms there are three functions named after the transform with differing suffixes. A _unconstrain function that maps from the constrained space back to free variables (the “transform”), A _constrain function that maps from free variables to constrained variables (the “inverse transform”), and a _jacobian function, which computes the same value as the _constrain function while also incrementing the Jacobian accumulator with the log Jacobian determinant.\nFor this page, variables named y are unconstrained, while variables named x are in the constrained space. The unconstraining functions will reject if their input does not satisfy the declared constraint.\n\n\nThese transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.\nThe functions are all overloaded to apply to containers elementwise. If the y argument is a container, the others must be either scalars or containers of exactly the same size.\n\n\nThese functions perform the transform and inverse transform described in the Lower Bounded Scalar section.\n \n\nreals lower_bound_constrain(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_bound_jacobian(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_bound_unconstrain(reals x, reals lb) Takes a value x which is greater than lb and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Upper Bounded Scalar section.\n \n\nreals upper_bound_constrain(reals y, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals upper_bound_jacobian(reals x, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals upper_bound_unconstrain(reals x, reals ub) Takes a value x which is less than ub and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Lower and Upper Bounded Scalar section.\n \n\nreals lower_upper_bound_constrain(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_upper_bound_jacobian(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_upper_bound_unconstrain(reals x, reals lb, reals ub) Takes a value x which is bounded between lb and ub and returns returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Affinely Transformed Scalar section.\n \n\nreals offset_multiplier_constrain(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nAvailable since 2.37\n \n\nreals offset_multiplier_jacobian(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals offset_multiplier_unconstrain(reals x, reals offset, reals mult) Takes a value x, shift offset, and scale mult and a value which has been un-scaled and un-shifted.\nAvailable since 2.37\n\n\n\n\nThese functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.\nWhere vectors is used, this indicates that either a vector or a (possibly multidimensional) array of vectors may be provided. The array will be processed element by element.\n\n\nThese functions perform the transform and inverse transform described in the Ordered Vector section.\n \n\nvectors ordered_constrain(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nAvailable since 2.37\n \n\nvectors ordered_jacobian(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors ordered_unconstrain(vectors x) Takes an ordered vector x and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Positive Ordered Vector section.\n \n\nvectors positive_ordered_constrain(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nAvailable since 2.37\n \n\nvectors positive_ordered_jacobian(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors positive_ordered_unconstrain(vectors x) Takes an ordered vector x with positive entries and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Simplex section.\n \n\nvectors simplex_constrain(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors simplex_jacobian(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors simplex_unconstrain(vectors x) Takes a simplex x and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Zero Sum Vector section.\n \n\nvectors sum_to_zero_constrain(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors sum_to_zero_jacobian(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThe returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors sum_to_zero_unconstrain(vectors x) Takes a vector x with elements that sum to 0 and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Vector section.\n \n\nvectors unit_vectors_constrain(vectors y) Takes a free vector y and returns a vector with unit length, i.e., norm2(unit_vectors_constrain(y)) == 1 for any y that has a positive and finite norm itself (if y does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.\nAvailable since 2.37\n \n\nvectors unit_vectors_jacobian(vectors y) Takes a free vector y and returns a vector with unit length. This function rejects if given a vector of all zeros.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors unit_vectors_unconstrain(vectors x) Takes a vector x of unit length and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\n\nSimilarly to the above, vectors means a vector or array thereof, and matrices means a matrix or array thereof.\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Correlation Matrices section.\n \n\nmatrices cholesky_factor_corr_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.\nAvailable since 2.37\n \n Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment. matrices cholesky_factor_corr_jacobian(vectors y, int K)\nAvailable since 2.37\n \n\nvectors cholesky_factor_corr_unconstrain(matrices x) Takes x, a (\\(K \\times K\\)) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Covariance Matrices section.\n \n\nmatrices cholesky_factor_cov_constrain(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nAvailable since 2.37\n \n\nmatrices cholesky_factor_cov_jacobian(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cholesky_factor_cov_unconstrain(matrices x) Takes a \\(M \\times N\\) matrix x which is a Cholesky factor of a covariance matrix (a matrix \\(L\\) such that \\(L^T L\\) is positive definite) and returns the corresponding free vector of length \\(N + \\binom{N}{2} + (M - N)N\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Correlation Matrices section.\n \n\nmatrices corr_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nAvailable since 2.37\n \n\nmatrices corr_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors corr_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) matrix x which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size \\(\\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Covariance Matrices section.\n \n\nmatrices cov_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nAvailable since 2.37\n \n\nmatrices cov_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cov_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) positive definite matrix x and returns the corresponding free vector of size \\(K + \\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for column (left) stochastic matrices.\n \n\nmatrices stochastic_column_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nAvailable since 2.37\n \n\nmatrices stochastic_column_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_column_unconstrain(matrices x) Takes a left stochastic matrix x of size \\(N+1 \\times M\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for row (right) stochastic matrices.\n \n\nmatrices stochastic_row_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nAvailable since 2.37\n \n\nmatrices stochastic_row_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_row_unconstrain(matrices x) Takes a right stochastic matrix x of size \\(N \\times M+1\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThe sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the sum-to-zero matrix section of the Reference Manual.\n \n\nmatrices sum_to_zero_constrain(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_jacobian(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to sum_to_zero_constrain.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_unconstrain(matrices x) This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html#transforms-for-scalars",
+    "href": "functions-reference/transform_functions.html#transforms-for-scalars",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "These transformations take unconstrained values on the real number line and either constrain the, to a subset of the real line with a lower bound, upper bound, or both, or provide an affine map that does not constrain values but can help with shifting and scaling them so they are more standardized.\nThe functions are all overloaded to apply to containers elementwise. If the y argument is a container, the others must be either scalars or containers of exactly the same size.\n\n\nThese functions perform the transform and inverse transform described in the Lower Bounded Scalar section.\n \n\nreals lower_bound_constrain(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_bound_jacobian(reals y, reals lb) Takes a value y and lower bound lb and returns the corresponding value which is greater than lb (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_bound_unconstrain(reals x, reals lb) Takes a value x which is greater than lb and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Upper Bounded Scalar section.\n \n\nreals upper_bound_constrain(reals y, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals upper_bound_jacobian(reals x, reals ub) Takes a value y and upper bound ub and returns the corresponding value which is less than ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals upper_bound_unconstrain(reals x, reals ub) Takes a value x which is less than ub and returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Lower and Upper Bounded Scalar section.\n \n\nreals lower_upper_bound_constrain(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nAvailable since 2.37\n \n\nreals lower_upper_bound_jacobian(reals y, reals lb, reals ub) Takes a value y, lower bound lb, and upper bound ub and returns the corresponding value which is bounded between lb and ub (except for the possibility of rounding due to numeric precision issues, in which case it will be equal to the bound).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals lower_upper_bound_unconstrain(reals x, reals lb, reals ub) Takes a value x which is bounded between lb and ub and returns returns the corresponding unconstrained value.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Affinely Transformed Scalar section.\n \n\nreals offset_multiplier_constrain(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nAvailable since 2.37\n \n\nreals offset_multiplier_jacobian(reals y, reals offset, reals mult) Takes a value y, shift offset, and scale mult and returns a rescaled and shifted value.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nreals offset_multiplier_unconstrain(reals x, reals offset, reals mult) Takes a value x, shift offset, and scale mult and a value which has been un-scaled and un-shifted.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html#transforms-for-constrained-vectors",
+    "href": "functions-reference/transform_functions.html#transforms-for-constrained-vectors",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "These functions constrain entire vectors hollistically. Some transforms also change the length of the vector, as noted in the documentation.\nWhere vectors is used, this indicates that either a vector or a (possibly multidimensional) array of vectors may be provided. The array will be processed element by element.\n\n\nThese functions perform the transform and inverse transform described in the Ordered Vector section.\n \n\nvectors ordered_constrain(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nAvailable since 2.37\n \n\nvectors ordered_jacobian(vectors y) Takes a free vector y and returns a vector with elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors ordered_unconstrain(vectors x) Takes an ordered vector x and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Positive Ordered Vector section.\n \n\nvectors positive_ordered_constrain(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nAvailable since 2.37\n \n\nvectors positive_ordered_jacobian(vectors y) Takes a free vector y and returns a vector with positive elements in ascending order.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors positive_ordered_unconstrain(vectors x) Takes an ordered vector x with positive entries and returns the corresponding free vector.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Simplex section.\n \n\nvectors simplex_constrain(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors simplex_jacobian(vectors y) Takes a free vector y and returns a simplex (a vector such that each element is between 0 and 1, and the sum of the elements is 1, up to rounding errors).\nThis returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors simplex_unconstrain(vectors x) Takes a simplex x and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Zero Sum Vector section.\n \n\nvectors sum_to_zero_constrain(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThis returned vector will have one extra element compared to the input y.\nAvailable since 2.37\n \n\nvectors sum_to_zero_jacobian(vectors y) Takes a free vector y and returns a vector such that the elements sum to 0.\nThe returned vector will have one extra element compared to the input y.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors sum_to_zero_unconstrain(vectors x) Takes a vector x with elements that sum to 0 and returns the corresponding free vector.\nThis returned vector will have one fewer elements compared to the input x.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Unit Vector section.\n \n\nvectors unit_vectors_constrain(vectors y) Takes a free vector y and returns a vector with unit length, i.e., norm2(unit_vectors_constrain(y)) == 1 for any y that has a positive and finite norm itself (if y does not, the function rejects). Note that, in particular, this implies the function rejects if given a vector of all zeros.\nAvailable since 2.37\n \n\nvectors unit_vectors_jacobian(vectors y) Takes a free vector y and returns a vector with unit length. This function rejects if given a vector of all zeros.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors unit_vectors_unconstrain(vectors x) Takes a vector x of unit length and returns the corresponding free vector.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/transform_functions.html#transforms-for-constrained-matrices",
+    "href": "functions-reference/transform_functions.html#transforms-for-constrained-matrices",
+    "title": "Variable Transformation Functions",
+    "section": "",
+    "text": "Similarly to the above, vectors means a vector or array thereof, and matrices means a matrix or array thereof.\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Correlation Matrices section.\n \n\nmatrices cholesky_factor_corr_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix. This matrix is a Cholesky factor of a covariance matrix (i.e., a lower triangular matrix with a strictly positive diagonal), but with the additional constraint that each row is of unit length.\nAvailable since 2.37\n \n Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K Cholesky factor of a correlation matrix.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment. matrices cholesky_factor_corr_jacobian(vectors y, int K)\nAvailable since 2.37\n \n\nvectors cholesky_factor_corr_unconstrain(matrices x) Takes x, a (\\(K \\times K\\)) matrix which is the Cholesky factor of a correlation matrix (a lower triangular matrix with a strictly positive diagonal and each row having unit length), and returns the corresponding free vector of length $ imes $.\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Cholesky Factors of Covariance Matrices section.\n \n\nmatrices cholesky_factor_cov_constrain(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nAvailable since 2.37\n \n\nmatrices cholesky_factor_cov_jacobian(vectors y, int M, int N) Takes a free vector y and integers M and N and returns the M by N Cholesky factor of a covariance matrix. This matrix is a lower triangular matrix \\(L\\), with a strictly positive diagonal, such that \\(L^T L\\) is positive definite.\nNote that y must have length N + choose(N, 2) + (M - N) * N, and M must be greater than or equal to N.\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cholesky_factor_cov_unconstrain(matrices x) Takes a \\(M \\times N\\) matrix x which is a Cholesky factor of a covariance matrix (a matrix \\(L\\) such that \\(L^T L\\) is positive definite) and returns the corresponding free vector of length \\(N + \\binom{N}{2} + (M - N)N\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Correlation Matrices section.\n \n\nmatrices corr_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nAvailable since 2.37\n \n\nmatrices corr_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == choose(K, 2), and returns a K by K correlation matrix (a positive definite matrix with a unit diagonal).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors corr_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) matrix x which is a correlation matrix (a positive definite matrix with a unit diagonal) and returns the corresponding free vector of size \\(\\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Covariance Matrices section.\n \n\nmatrices cov_matrix_constrain(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nAvailable since 2.37\n \n\nmatrices cov_matrix_jacobian(vectors y, int K) Takes a vector y and integer K, where length(y) == K + choose(K, 2), and returns a K by K covariance matrix (a positive definite matrix).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nvectors cov_matrix_unconstrain(matrices x) Takes a \\(K \\times K\\) positive definite matrix x and returns the corresponding free vector of size \\(K + \\binom{K}{2}\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for column (left) stochastic matrices.\n \n\nmatrices stochastic_column_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nAvailable since 2.37\n \n\nmatrices stochastic_column_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a left stochastic matrix (a matrix where each column is a simplex) of size \\(N+1 \\times M\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_column_unconstrain(matrices x) Takes a left stochastic matrix x of size \\(N+1 \\times M\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThese functions perform the transform and inverse transform described in the Stochastic Matrix section for row (right) stochastic matrices.\n \n\nmatrices stochastic_row_constrain(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nAvailable since 2.37\n \n\nmatrices stochastic_row_jacobian(matrices y) Takes a free matrix y of size \\(N \\times M\\) and returns a right stochastic matrix (a matrix where each row is a simplex) of size \\(N \\times M+1\\).\nThis function also increments the Jacobian accumulator with the corresponding change of variables adjustment.\nAvailable since 2.37\n \n\nmatrices stochastic_row_unconstrain(matrices x) Takes a right stochastic matrix x of size \\(N \\times M+1\\) and returns the corresponding free matrix of size \\(N \\times M\\).\nAvailable since 2.37\n\n\n\nThe sum-to-zero matrix transforms map between unconstrained values and matrices whose rows and columns sum to zero; full definitions of the function and Jacobian can be found in the sum-to-zero matrix section of the Reference Manual.\n \n\nmatrices sum_to_zero_constrain(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. This function covers the incrementation of the log Jacobian because the incrementation is zero.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_jacobian(matrices y) The constraining function maps an unconstrained N x M matrix to an (N + 1) x (M + 1) matrix for which the rows and columns all sum to zero. Because the log Jacobian incrementation is zero, this is identical to sum_to_zero_constrain.\nThis returned matrix will have one extra row and column compared to the input y.\nAvailable since 2.37\n \n\nmatrices sum_to_zero_unconstrain(matrices x) This function maps a matrix with rows that sum to zero and columns that sum to zero to an unconstrained matrix with one fewer row and and one fewer column.\nAvailable since 2.37",
+    "crumbs": [
+      "Functions Reference",
+      "Built-in Functions",
+      "Variable Transformation Functions"
+    ]
+  },
+  {
+    "objectID": "functions-reference/unbounded_discrete_distributions.html",
+    "href": "functions-reference/unbounded_discrete_distributions.html",
+    "title": "Unbounded Discrete Distributions",
+    "section": "",
+    "text": "The unbounded discrete distributions have support over the natural numbers (i.e., the non-negative integers).\n\n\nFor the negative binomial distribution Stan uses the parameterization described in Gelman et al. (2013). For alternative parameterizations, see section negative binomial glm.\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n\n\\in \\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial}(n~|~\\alpha,\\beta) = \\binom{n +\n\\alpha - 1}{\\alpha - 1} \\, \\left( \\frac{\\beta}{\\beta+1}\n\\right)^{\\!\\alpha} \\, \\left( \\frac{1}{\\beta + 1} \\right)^{\\!n} \\!. \\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial}(\\alpha,\\beta)\\) are given by \\[\\begin{equation*} \\mathbb{E}[n] =\n\\frac{\\alpha}{\\beta} \\ \\ \\text{ and } \\ \\ \\text{Var}[n] =\n\\frac{\\alpha}{\\beta^2} (\\beta + 1). \\end{equation*}\\]\n\n\n\nn ~ neg_binomial(alpha, beta)\nIncrement target log probability density with neg_binomial_lupmf(n | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal neg_binomial_lpmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lupmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal neg_binomial_cdf(ints n | reals alpha, reals beta) The negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal neg_binomial_lcdf(ints n | reals alpha, reals beta) The log of the negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lccdf(ints n | reals alpha, reals beta) The log of the negative binomial complementary cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR neg_binomial_rng(reals alpha, reals beta) Generate a negative binomial variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) beta must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section combinatorial functions, below, provides a second alternative parameterization directly in terms of the log mean.\n\n\nThe first parameterization is for \\(\\mu \\in \\mathbb{R}^+\\) and \\(\\phi \\in\n\\mathbb{R}^+\\), which for \\(n \\in \\mathbb{N}\\) is defined as \\[\\begin{equation*}\n\\text{NegBinomial2}(n \\, | \\, \\mu, \\phi)\n= \\binom{n + \\phi - 1}{n} \\, \\left( \\frac{\\mu}{\\mu+\\phi} \\right)^{\\!n} \\, \\left( \\frac{\\phi}{\\mu+\\phi} \\right)^{\\!\\phi} \\!.\n\\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial2}(n~|~\\mu,\\phi)\\) are \\[\\begin{equation*}\n\\mathbb{E}[n] = \\mu \\ \\ \\ \\text{ and } \\ \\ \\ \\text{Var}[n] = \\mu + \\frac{\\mu^2}{\\phi}.\n\\end{equation*}\\] Recall that \\(\\text{Poisson}(\\mu)\\) has variance \\(\\mu\\), so \\(\\mu^2 / \\phi &gt; 0\\) is the additional variance of the negative binomial above that of the Poisson with mean \\(\\mu\\). So the inverse of parameter \\(\\phi\\) controls the overdispersion, scaled by the square of the mean, \\(\\mu^2\\).\n\n\n\nn ~ neg_binomial_2(mu, phi)\nIncrement target log probability density with neg_binomial_2_lupmf(n | mu, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_lpmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_lupmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_cdf(ints n | reals mu, reals phi) The negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.6\n \n\nreal neg_binomial_2_lcdf(ints n | reals mu, reals phi) The log of the negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nreal neg_binomial_2_lccdf(ints n | reals mu, reals phi) The log of the negative binomial complementary cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nR neg_binomial_2_rng(reals mu, reals phi) Generate a negative binomial variate with location mu and precision phi; may only be used in transformed data and generated quantities blocks. mu must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nRelated to the parameterization in section negative binomial, alternative parameterization, the following parameterization uses a log mean parameter \\(\\eta = \\log(\\mu)\\), defined for \\(\\eta \\in \\mathbb{R}\\), \\(\\phi \\in \\mathbb{R}^+\\), so that for \\(n \\in\n\\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial2Log}(n \\, | \\, \\eta, \\phi) =\n\\text{NegBinomial2}(n | \\exp(\\eta), \\phi). \\end{equation*}\\] This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.\n\n\nn ~ neg_binomial_2_log(eta, phi)\nIncrement target log probability density with neg_binomial_2_log_lupmf(n | eta, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_log_lpmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_log_lupmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nR neg_binomial_2_log_rng(reals eta, reals phi) Generate a negative binomial variate with log-location eta and inverse overdispersion control phi; may only be used in transformed data and generated quantities blocks. eta must be less than \\(29 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e. a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\phi\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{NegBinomial2LogGLM}(y~|~x, \\alpha, \\beta, \\phi) = \\prod_{1\\leq i\n\\leq n}\\text{NegBinomial2}(y_i~|~\\exp(\\alpha_i + x_i\\cdot \\beta),\n\\phi). \\end{equation*}\\]\n\n\n\ny ~ neg_binomial_2_log_glm(x, alpha, beta, phi)\nIncrement target log probability density with neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(\\lambda \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{Poisson}(n|\\lambda) = \\frac{1}{n!} \\, \\lambda^n \\,\n\\exp(-\\lambda). \\end{equation*}\\]\n\n\n\nn ~ poisson(lambda)\nIncrement target log probability density with poisson_lupmf(n | lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_lpmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lupmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal poisson_cdf(ints n | reals lambda) The Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.0\n \n\nreal poisson_lcdf(ints n | reals lambda) The log of the Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lccdf(ints n | reals lambda) The log of the Poisson complementary cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nR poisson_rng(reals lambda) Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than \\(2^{30}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also provides a parameterization of the Poisson using the log rate \\(\\alpha = \\log \\lambda\\) as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{PoissonLog}(n|\\alpha) = \\frac{1}{n!} \\, \\exp \\left(n\\alpha -\n\\exp(\\alpha) \\right). \\end{equation*}\\]\n\n\n\nn ~ poisson_log(alpha)\nIncrement target log probability density with poisson_log_lupmf(n | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_log_lpmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha\nAvailable since 2.12\n \n\nreal poisson_log_lupmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha dropping constant additive terms\nAvailable since 2.25\n \n\nR poisson_log_rng(reals alpha) Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than \\(30 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18\n\n\n\n\nStan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e. a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{PoissonLogGLM}(y|x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Poisson}(y_i|\\exp(\\alpha_i + x_i\\cdot \\beta)). \\end{equation*}\\]\n\n\n\ny ~ poisson_log_glm(x, alpha, beta)\nIncrement target log probability density with poisson_log_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n\n\n\n\n\n\nIf \\(r \\in \\mathbb{R}^+\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{BetaNegBinomial}(n|r,\\alpha,\\beta) = \\frac {\\Gamma (n+r )}{n!\\;\\Gamma (r )}\n\\frac {\\mathrm {B} (\\beta+n,\\alpha +r )}{\\mathrm {B} (\\beta,\\alpha )}. \\end{equation*}\\]\n\n\n\nn ~ beta_neg_binomial(r,alpha,beta)\nIncrement target log probability density with beta_neg_binomial_lupmf(n | r, alpha, beta).\nAvailable since 2.36\n \n\n\n\n\n \n\nreal beta_neg_binomial_lpmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lupmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta dropping constant additive terms.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_cdf(ints n | reals r, reals alpha, reals beta) The beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lcdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lccdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial complementary cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nR beta_neg_binomial_rng(reals r, reals alpha, reals beta) Generate a beta negative binomial variate with parameters r, alpha and beta; may only be used in transformed data and generated quantities blocks. r \\(\\cdot\\) beta \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.36\n\n\n\n\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}^+=\\{1,2,...\\}\\), \\[\\begin{equation*}\n\\text{YuleSimon}(n|\\alpha) = \\alpha \\, \\mathrm{B}(\\alpha + 1, n) = \\alpha \\, \\frac{\\Gamma(n) \\, \\Gamma(\\alpha + 1)}{\\Gamma(n + \\alpha + 1)}.\n\\end{equation*}\\]\n\n\n\nn ~ yule_simon(alpha)\nIncrement target log probability density with yule_simon_lupmf(n | alpha).\nAvailable since 2.39\n \n\n\n\n\n \n\nreal yule_simon_lpmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lupmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha dropping constant additive terms.\nAvailable since 2.39\n \n\nreal yule_simon_cdf(ints n | reals alpha) The Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lcdf(ints n | reals alpha) The log of the Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lccdf(ints n | reals alpha) The log of the Yule Simon complementary cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nR yule_simon_rng(reals alpha) Generate a Yule Simon variate with parameter alpha; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.39",
+    "crumbs": [
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#vector-matrix-array-expressions.section",
-    "href": "reference-manual/expressions.html#vector-matrix-array-expressions.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#negative-binomial-distribution",
+    "href": "functions-reference/unbounded_discrete_distributions.html#negative-binomial-distribution",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "Expressions for the Stan container objects, namely arrays, vectors, row vectors, matrices, and tuples, can all be constructed using expressions.\n\n\nSquare brackets may be wrapped around a sequence of comma separated primitive expressions to produce a row vector expression. For example, the expression [ 1, 10, 100 ] denotes a row vector of three elements with real values 1.0, 10.0, and 100.0. Applying the transpose operator to a row vector expression produces a vector expression. This syntax provides a way declare and define small vectors a single line, as follows.\nrow_vector[2] rv2 =  [ 1, 2 ];\nvector[3] v3 = [ 3, 4, 5 ]';\nThe vector expression values may be compound expressions or variable names, so it is legal to write [ 2 * 3, 1 + 4] or [ x, y ], providing that x and y are primitive variables.\n\n\n\nA matrix expression consists of square brackets wrapped around a sequence of comma separated row vector expressions. This syntax provides a way declare and define a matrix in a single line, as follows.\nmatrix[3, 2] m1 = [ [ 1, 2 ], [ 3, 4 ], [5, 6 ] ];\nAny expression denoting a row vector can be used in a matrix expression. For example, the following code is valid:\nvector[2] vX = [ 1, 10 ]';\nrow_vector[2] vY = [ 100, 1000 ];\nmatrix[3, 2] m2 = [ vX', vY, [ 1, 2 ]  ];\n\n\n\nComplex vector expressions work the same way as real vector expressions. For example, the following are all legal Stan expressions and assignments.\ncomplex_vector[3] = [1 + 2i, 3 - 1.7i, 0]';\ncomplex_row_vector[2] = [12, -2i];\ncomplex_matrix[2, 3] = [[1 + 2i, 3 - 1.7i, 0],\n                        [3.9 - 1.234i, 176i, 1 + 1i]];\n\n\nThe empty expression [ ] is ambiguous and therefore is not allowed and similarly expressions such as [ [ ] ] or [ [ ], [ ] ] are not allowed.\n\n\n\n\nIf needed, it is possible to create an empty vector with\nrep_vector(e, 0)\nwhere the first expression e needs to scalar of type real.\nIf needed, it is possible to create an empty matrix with\nrep_matrix(e, 0, 0)\nwhere the first expression e needs to scalar of type real.\n\n\n\nCurly braces may be wrapped around a sequence of expressions to produce an array expression. For example, the expression { 1, 10, 100 } denotes an integer array of three elements with values 1, 10, and 100. This syntax is particularly convenient to define small arrays in a single line, as follows.\narray[3] int a = { 1, 10, 100 };\nThe values may be compound expressions, so it is legal to write { 2 * 3, 1 + 4 }. It is also possible to write two dimensional arrays directly, as in the following example.\narray[2, 3] int b = { { 1, 2, 3 }, { 4, 5, 6 } };\nThis way, b[1] is { 1, 2, 3 } and b[2] is { 4, 5, 6 }.\nWhitespace is always interchangeable in Stan, so the above can be laid out as follows to more clearly indicate the row and column structure of the resulting two dimensional array.\narray[2, 3] int b = { { 1, 2, 3 },\n                { 4, 5, 6 } };\n\n\n\nThe empty array expression ({ }) is not allowed. See more about restrictions on array expressions in subsection Restrictions on values.\nIf needed, it is possible to create an empty array with\nrep_array(e, 0)\nwhere the first expression e determines the type of the array. For example, rep_array(0.0, 0) returns an empty real array of type real[], whereas rep_array({123}, 0) returns an empty two dimensional integer array of type int[ , ]. Only the type of the first argument is used, so the integer arrays {123} and {0} produce equivalent values.\n\n\n\nAny type of expression may be used within braces to form an array expression. In the simplest case, all of the elements will be of the same type and the result will be an array of elements of that type. For example, the elements of the array can be vectors, in which case the result is an array of vectors.\nvector[3] b;\nvector[3] c;\n// ...\narray[2] vector[3] d = { b, c };\nThe elements may also be a mixture of int and real typed expressions, in which case the result is an array of real values.\narray[2] real b = { 1, 1.9 };\n\n\n\nStan uses parentheses around a comma-separated sequence of expressions to construct a tuple. For example, we can construct a 2-tuple as follows.\ntuple(int, vector[3]) xy = (42, [1, 2.9, -1.3]');\nThe expression 42 is of type int and the expression [1, 2.9, -1.3] is of type row_vector so that [1, 2.9, -1.3]' is of type vector and of size 3. The whole tuple expression (42, [1, 2.9, -1.3]') thus has a sized type of tuple(int, vector[3]) and an unsized type (e.g., for a function argument) of tuple(int, vector).\nStan does not support the Python notation with trailing commas, such as (1, 2, 3, ) for a 3-tuple.\n\n\n\nThere are some restrictions on how array expressions may be used that arise from their types being calculated bottom up and the basic data type and assignment rules of Stan.\n\n\nAlthough it is tempting to try to define a ragged array expression, all Stan data types are rectangular (or boxes or other higher-dimensional generalizations). Thus the following nested array expression will cause an error when it tries to create a non-rectangular array.\n{ { 1, 2, 3 }, { 4, 5 } }  // compile time error: size mismatch\nThis may appear to be OK, because it is creating a two-dimensional integer array (array[,] int) out of two one-dimensional array integer arrays (array[] int). But it is not allowed because the two one-dimensional arrays are not the same size. If the elements are array expressions, this can be diagnosed at compile time. If one or both expressions is a variable, then that won’t be caught until runtime.\n{ { 1, 2, 3 }, m }  // runtime error if m not size 3\n\n\n\nBecause there is no way to infer the type of the result, the empty array expression ({ }) is not allowed. This does not sacrifice expressive power, because a declaration is sufficient to initialize a zero-element array.\narray[0] int a;   // a is fully defined as zero element array\n\n\n\nThere is no way to declare or construct a zero-tuple or one-tuple in Stan. Tuples must be at least two elements long. The expression () does not pick out a zero-tuple—it is ill formed. Similarly, the expression (1) is of type int rather than a tuple.",
+    "text": "For the negative binomial distribution Stan uses the parameterization described in Gelman et al. (2013). For alternative parameterizations, see section negative binomial glm.\n\n\nIf \\(\\alpha \\in \\mathbb{R}^+\\) and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n\n\\in \\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial}(n~|~\\alpha,\\beta) = \\binom{n +\n\\alpha - 1}{\\alpha - 1} \\, \\left( \\frac{\\beta}{\\beta+1}\n\\right)^{\\!\\alpha} \\, \\left( \\frac{1}{\\beta + 1} \\right)^{\\!n} \\!. \\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial}(\\alpha,\\beta)\\) are given by \\[\\begin{equation*} \\mathbb{E}[n] =\n\\frac{\\alpha}{\\beta} \\ \\ \\text{ and } \\ \\ \\text{Var}[n] =\n\\frac{\\alpha}{\\beta^2} (\\beta + 1). \\end{equation*}\\]\n\n\n\nn ~ neg_binomial(alpha, beta)\nIncrement target log probability density with neg_binomial_lupmf(n | alpha, beta).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal neg_binomial_lpmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lupmf(ints n | reals alpha, reals beta) The log negative binomial probability mass of n given shape alpha and inverse scale beta dropping constant additive terms\nAvailable since 2.25\n \n\nreal neg_binomial_cdf(ints n | reals alpha, reals beta) The negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.0\n \n\nreal neg_binomial_lcdf(ints n | reals alpha, reals beta) The log of the negative binomial cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nreal neg_binomial_lccdf(ints n | reals alpha, reals beta) The log of the negative binomial complementary cumulative distribution function of n given shape alpha and inverse scale beta\nAvailable since 2.12\n \n\nR neg_binomial_rng(reals alpha, reals beta) Generate a negative binomial variate with shape alpha and inverse scale beta; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) beta must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#parentheses-for-grouping",
-    "href": "reference-manual/expressions.html#parentheses-for-grouping",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#nbalt",
+    "href": "functions-reference/unbounded_discrete_distributions.html#nbalt",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "Any expression wrapped in parentheses is also an expression. Like in C++, but unlike in R, only the round parentheses, ( and ), are allowed. The square brackets [ and ] are reserved for array indexing and the curly braces { and } for grouping statements.\nWith parentheses it is possible to explicitly group subexpressions with operators. Without parentheses, the expression 1 + 2 * 3 has a subexpression 2 * 3 and evaluates to 7. With parentheses, this grouping may be made explicit with the expression 1 + (2 * 3). More importantly, the expression (1 + 2) *   3 has 1 + 2 as a subexpression and evaluates to 9.",
+    "text": "Stan also provides an alternative parameterization of the negative binomial distribution directly using a mean (i.e., location) parameter and a parameter that controls overdispersion relative to the square of the mean. Section combinatorial functions, below, provides a second alternative parameterization directly in terms of the log mean.\n\n\nThe first parameterization is for \\(\\mu \\in \\mathbb{R}^+\\) and \\(\\phi \\in\n\\mathbb{R}^+\\), which for \\(n \\in \\mathbb{N}\\) is defined as \\[\\begin{equation*}\n\\text{NegBinomial2}(n \\, | \\, \\mu, \\phi)\n= \\binom{n + \\phi - 1}{n} \\, \\left( \\frac{\\mu}{\\mu+\\phi} \\right)^{\\!n} \\, \\left( \\frac{\\phi}{\\mu+\\phi} \\right)^{\\!\\phi} \\!.\n\\end{equation*}\\]\nThe mean and variance of a random variable \\(n \\sim\n\\text{NegBinomial2}(n~|~\\mu,\\phi)\\) are \\[\\begin{equation*}\n\\mathbb{E}[n] = \\mu \\ \\ \\ \\text{ and } \\ \\ \\ \\text{Var}[n] = \\mu + \\frac{\\mu^2}{\\phi}.\n\\end{equation*}\\] Recall that \\(\\text{Poisson}(\\mu)\\) has variance \\(\\mu\\), so \\(\\mu^2 / \\phi &gt; 0\\) is the additional variance of the negative binomial above that of the Poisson with mean \\(\\mu\\). So the inverse of parameter \\(\\phi\\) controls the overdispersion, scaled by the square of the mean, \\(\\mu^2\\).\n\n\n\nn ~ neg_binomial_2(mu, phi)\nIncrement target log probability density with neg_binomial_2_lupmf(n | mu, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_lpmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_lupmf(ints n | reals mu, reals phi) The log negative binomial probability mass of n given location mu and precision phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_cdf(ints n | reals mu, reals phi) The negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.6\n \n\nreal neg_binomial_2_lcdf(ints n | reals mu, reals phi) The log of the negative binomial cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nreal neg_binomial_2_lccdf(ints n | reals mu, reals phi) The log of the negative binomial complementary cumulative distribution function of n given location mu and precision phi.\nAvailable since 2.12\n \n\nR neg_binomial_2_rng(reals mu, reals phi) Generate a negative binomial variate with location mu and precision phi; may only be used in transformed data and generated quantities blocks. mu must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#arithmetic-expressions.section",
-    "href": "reference-manual/expressions.html#arithmetic-expressions.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log",
+    "href": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "For integer and real-valued expressions, Stan supports the basic binary arithmetic operations of addition (+), subtraction (-), multiplication (*) and division (/) in the usual ways.\nFor integer expressions, Stan supports the modulus (%) binary arithmetic operation. Stan also supports the unary operation of negation for integer and real-valued expressions. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3.0 + 0.14\n-15\n2 * 3 + 1\n(x - y) / 2.0\n(n * (n + 1)) / 2\nx / n\nm % n\nThe negation, addition, subtraction, and multiplication operations are extended to matrices, vectors, and row vectors. The transpose operation, written using an apostrophe (') is also supported for vectors, row vectors, and matrices. Return types for matrix operations are the smallest types that can be statically guaranteed to contain the result. The full set of allowable input types and corresponding return types is detailed in the list of functions.\nFor example, if y and mu are variables of type vector and Sigma is a variable of type matrix, then (y - mu)' * Sigma * (y - mu) is a well-formed expression of type real. The type of the complete expression is inferred working outward from the subexpressions. The subexpression(s) y - mu are of type vector because the variables y and mu are of type vector. The transpose of this expression, the subexpression (y - mu)' is of type row_vector. Multiplication is left associative and transpose has higher precedence than multiplication, so the above expression is equivalent to the following fully specified form (((y - mu)') * Sigma) * (y - mu).\nThe type of subexpression (y - mu)' * Sigma is inferred to be row_vector, being the result of multiplying a row vector by a matrix. The whole expression’s type is thus the type of a row vector multiplied by a (column) vector, which produces a real value.\nStan provides elementwise matrix multiplication (e.g., a .* b) and division (e.g., a ./ b) operations. These provide a shorthand to replace loops, but are not intrinsically more efficient than a version programmed with an elementwise calculations and assignments in a loop. For example, given declarations,\nvector[N] a;\nvector[N] b;\nvector[N] c;\nthe assignment,\nc = a .* b;\nproduces the same result with roughly the same efficiency as the loop\nfor (n in 1:N) {\n  c[n] = a[n] * b[n];\n}\nStan supports exponentiation (^) of integer and real-valued expressions. The return type of exponentiation is always a real-value. For example, assuming n and m are integer variables and x and y real variables, the following expressions are legal.\n3 ^ 2\n3.0 ^ -2\n3.0 ^ 0.14\nx ^ n\nn ^ x\nn ^ m\nx ^ y\nExponentiation is right associative, so the expression 2 ^ 3 ^ 4 is equivalent to the fully specified form 2 ^ (3 ^ 4).\n\n\nThe precedence and associativity of operators, as well as built-in syntax such as array indexing and function application is given in tabular form in the following table.\n\n\nStan’s unary, binary, and ternary operators, with their precedences, associativities, place in an expression, and a description. The last two lines list the precedence of function application and array, matrix, and vector indexing. The operators are listed in order of precedence, from least tightly binding to most tightly binding. The full set of legal arguments and corresponding result types are provided in the function documentation for the operators (i.e., operator*(int, int):int indicates the application of the multiplication operator to two integers, which returns an integer). Parentheses may be used to group expressions explicitly rather than relying on precedence and associativity.\n\n\n\n\n\n\n\n\n\n\nOp.\nPrec.\nAssoc.\nPlacement\nDescription\n\n\n\n\n? ~ :\n10\nright\nternary infix\nconditional\n\n\n||\n9\nleft\nbinary infix\nlogical or\n\n\n&&\n8\nleft\nbinary infix\nlogical and\n\n\n==\n7\nleft\nbinary infix\nequality\n\n\n!=\n7\nleft\nbinary infix\ninequality\n\n\n&lt;\n6\nleft\nbinary infix\nless than\n\n\n&lt;=\n6\nleft\nbinary infix\nless than or equal\n\n\n&gt;\n6\nleft\nbinary infix\ngreater than\n\n\n&gt;=\n6\nleft\nbinary infix\ngreater than or equal\n\n\n+\n5\nleft\nbinary infix\naddition\n\n\n-\n5\nleft\nbinary infix\nsubtraction\n\n\n*\n4\nleft\nbinary infix\nmultiplication\n\n\n.*\n4\nleft\nbinary infix\nelementwise multiplication\n\n\n/\n4\nleft\nbinary infix\n(right) division\n\n\n./\n4\nleft\nbinary infix\nelementwise division\n\n\n%\n4\nleft\nbinary infix\nmodulus\n\n\n\\\n3\nleft\nbinary infix\nleft division\n\n\n%/%\n3\nleft\nbinary infix\ninteger division\n\n\n!\n2\nn/a\nunary prefix\nlogical negation\n\n\n-\n2\nn/a\nunary prefix\nnegation\n\n\n+\n2\nn/a\nunary prefix\npromotion (no-op in Stan)\n\n\n^\n1\nright\nbinary infix\nexponentiation\n\n\n.^\n1\nright\nbinary infix\nelementwise exponentiation\n\n\n'\n0\nn/a\nunary postfix\ntransposition\n\n\n()\n0\nn/a\nprefix, wrap\nfunction application\n\n\n[]\n0\nleft\nprefix, wrap\narray, matrix indexing\n\n\n\nOther expression-forming operations, such as function application and subscripting bind more tightly than any of the arithmetic operations.\nThe precedence and associativity determine how expressions are interpreted. Because addition is left associative, the expression a + b + c is interpreted as (a + b) + c. Similarly, a / b * c is interpreted as (a / b) * c.\nBecause multiplication has higher precedence than addition, the expression a * b + c is interpreted as (a * b) + c and the expression a + b * c is interpreted as a + (b * c). Similarly, 2 * x + 3 * - y is interpreted as (2 * x) + (3 * (-y)).\nTransposition and exponentiation bind more tightly than any other arithmetic or logical operation. For vectors, row vectors, and matrices, -u' is interpreted as -(u'), u * v' as u* (v'), and u' * v as (u') * v. For integer and reals, -n ^ 3 is interpreted as -(n ^ 3).",
+    "text": "Related to the parameterization in section negative binomial, alternative parameterization, the following parameterization uses a log mean parameter \\(\\eta = \\log(\\mu)\\), defined for \\(\\eta \\in \\mathbb{R}\\), \\(\\phi \\in \\mathbb{R}^+\\), so that for \\(n \\in\n\\mathbb{N}\\), \\[\\begin{equation*} \\text{NegBinomial2Log}(n \\, | \\, \\eta, \\phi) =\n\\text{NegBinomial2}(n | \\exp(\\eta), \\phi). \\end{equation*}\\] This alternative may be used for sampling, as a function, and for random number generation, but as of yet, there are no CDFs implemented for it. This is especially useful for log-linear negative binomial regressions.\n\n\nn ~ neg_binomial_2_log(eta, phi)\nIncrement target log probability density with neg_binomial_2_log_lupmf(n | eta, phi).\nAvailable since 2.3\n \n\n\n\n\n \n\nreal neg_binomial_2_log_lpmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi.\nAvailable since 2.20\n \n\nreal neg_binomial_2_log_lupmf(ints n | reals eta, reals phi) The log negative binomial probability mass of n given log-location eta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nR neg_binomial_2_log_rng(reals eta, reals phi) Generate a negative binomial variate with log-location eta and inverse overdispersion control phi; may only be used in transformed data and generated quantities blocks. eta must be less than \\(29 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#conditional-operator.section",
-    "href": "reference-manual/expressions.html#conditional-operator.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log-glm",
+    "href": "functions-reference/unbounded_discrete_distributions.html#neg-binom-2-log-glm",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "The ternary conditional operator is unique in that it takes three arguments and uses a mixed syntax. If a is an expression of type int and b and c are expressions that can be converted to one another (e.g., compared with ==), then\na ? b : c\nis an expression of the promoted type of b and c. The result of this expression is b if condition a is true, c otherwise. The only promotion allowed in Stan is integer -&gt; real -&gt; complex; e.g. if one argument is of type int and the other of type real, the conditional expression as a whole is of type real. In other cases, the arguments have to be of the same underlying Stan type (i.e., constraints don’t count, only the shape) and the conditional expression is of that type.\n\n\nThe conditional operator is the most loosely binding operator, so its arguments rarely require parentheses for disambiguation. For example,\na &gt; 0 || b &lt; 0 ? c + d : e - f\nis equivalent to the explicitly grouped version\n(a &gt; 0 || b &lt; 0) ? (c + d) : (e - f)\nThe latter is easier to read even if the parentheses are not strictly necessary.\n\n\n\nThe conditional operator is right associative, so that\na ? b : c ? d : e\nparses as if explicitly grouped as\na ? b : (c ? d : e)\nAgain, the explicitly grouped version is easier to read.\n\n\n\n\nStan’s conditional operator works very much like its C++ analogue. The first argument must be an expression denoting an integer. Typically this is a variable or a relation operator, as in the variable a in the example above. Then there are two resulting arguments, the first being the result returned if the condition evaluates to true (i.e., non-zero) and the second if the condition evaluates to false (i.e., zero). In the example above, the value b is returned if the condition evaluates to a non-zero value and c is returned if the condition evaluates to zero.\n\n\nThe key property of the conditional operator that makes it so useful in high-performance computing is that it only evaluates the returned subexpression, not the alternative expression. In other words, it is not like a typical function that evaluates its argument expressions eagerly in order to pass their values to the function. As usual, the saving is mostly in the derivatives that do not get computed rather than the unnecessary function evaluation itself.\n\n\n\nIf one return expression is a data value (an expression involving only constants and variables defined in the data or transformed data block), and the other is not, then the ternary operator will promote the data value to a parameter value. This can cause needless work calculating derivatives in some cases and be less efficient than a full if-then conditional statement. For example,\ndata {\n  array[10] real x;\n  // ...\n}\nparameters {\n  array[10] real z;\n  // ...\n}\nmodel {\n  y ~ normal(cond ? x : z, sigma);\n  // ...\n}\nwould be more efficiently (if not more transparently) coded as\nif (cond) {\n  y ~ normal(x, sigma);\n} else {\n  y ~ normal(z, sigma);\n}\nThe conditional statement, like the conditional operator, only evaluates one of the result statements. In this case, the variable x will not be promoted to a parameter and thus not cause any needless work to be carried out when propagating the chain rule during derivative calculations.",
+    "text": "Stan also supplies a single function for a generalized linear model with negative binomial distribution and log link function, i.e. a function for a negative binomial regression. This provides a more efficient implementation of negative binomial regression than a manually written regression in terms of a negative binomial distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m, \\phi\\in \\mathbb{R}^+\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{NegBinomial2LogGLM}(y~|~x, \\alpha, \\beta, \\phi) = \\prod_{1\\leq i\n\\leq n}\\text{NegBinomial2}(y_i~|~\\exp(\\alpha_i + x_i\\cdot \\beta),\n\\phi). \\end{equation*}\\]\n\n\n\ny ~ neg_binomial_2_log_glm(x, alpha, beta, phi)\nIncrement target log probability density with neg_binomial_2_log_glm_lupmf(y | x, alpha, beta, phi).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.23\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25\n \n\nreal neg_binomial_2_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi.\nAvailable since 2.18\n \n\nreal neg_binomial_2_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta, real phi) The log negative binomial probability mass of y given log-location alpha + x * beta and inverse overdispersion parameter phi dropping constant additive terms.\nAvailable since 2.25",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#language-indexing.section",
-    "href": "reference-manual/expressions.html#language-indexing.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson",
+    "href": "functions-reference/unbounded_discrete_distributions.html#poisson",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "Stan arrays, matrices, vectors, and row vectors are all accessed using the same array-like notation. For instance, if x is a variable of type array [] real (a one-dimensional array of reals) then x[1] is the value of the first element of the array.\nSubscripting has higher precedence than any of the arithmetic operations. For example, alpha * x[1] is equivalent to alpha * (x[1]).\nMultiple subscripts may be provided within a single pair of square brackets. If x is of type array[,] real, a two-dimensional array, then x[2, 501] is of type real.\n\n\nThe subscripting operator also returns subarrays of arrays. For example, if x is of type array[,,] real, then x[2] is of type array[,] real, and x[2, 3] is of type array[] real. As a result, the expressions x[2, 3] and x[2][3] have the same meaning.\n\n\n\nIf Sigma is a variable of type matrix, then Sigma[1] denotes the first row of Sigma and has the type row_vector.\n\n\n\nStan supports mixed indexing of arrays and their vector, row vector or matrix values. For example, if m is of type matrix[ , ], a two-dimensional array of matrices, then m[1] refers to the first row of the array, which is a one-dimensional array of matrices. More than one index may be used, so that m[1, 2] is of type matrix and denotes the matrix in the first row and second column of the array. Continuing to add indices, m[1, 2, 3] is of type row_vector and denotes the third row of the matrix denoted by m[1, 2]. Finally, m[1, 2, 3, 4] is of type real and denotes the value in the third row and fourth column of the matrix that is found at the first row and second column of the array m.",
+    "text": "If \\(\\lambda \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{Poisson}(n|\\lambda) = \\frac{1}{n!} \\, \\lambda^n \\,\n\\exp(-\\lambda). \\end{equation*}\\]\n\n\n\nn ~ poisson(lambda)\nIncrement target log probability density with poisson_lupmf(n | lambda).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_lpmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lupmf(ints n | reals lambda) The log Poisson probability mass of n given rate lambda dropping constant additive terms\nAvailable since 2.25\n \n\nreal poisson_cdf(ints n | reals lambda) The Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.0\n \n\nreal poisson_lcdf(ints n | reals lambda) The log of the Poisson cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nreal poisson_lccdf(ints n | reals lambda) The log of the Poisson complementary cumulative distribution function of n given rate lambda\nAvailable since 2.12\n \n\nR poisson_rng(reals lambda) Generate a Poisson variate with rate lambda; may only be used in transformed data and generated quantities blocks. lambda must be less than \\(2^{30}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#language-multi-indexing.section",
-    "href": "reference-manual/expressions.html#language-multi-indexing.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson-distribution-log-parameterization",
+    "href": "functions-reference/unbounded_discrete_distributions.html#poisson-distribution-log-parameterization",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "In addition to single integer indexes, as described in the language indexing section, Stan supports multiple indexing. Multiple indexes can be integer arrays of indexes, lower bounds, upper bounds, lower and upper bounds, or simply shorthand for all of the indexes. If the upper bound is smaller than the lower bound, the range is empty (unlike, e.g., in R). The upper bound and lower bound can be expressions that evaluate to integer. A complete list of index types is given in the following table.\n\n\nTypes of indexes and examples with one-dimensional containers of size N and an integer array ii of type array [] real size K.\n\n\n\nindex type\nexample\nvalue\n\n\n\n\ninteger\na[11]\nvalue of a at index 11\n\n\ninteger array\na[ii]\na[ii[1]], …, a[ii[K]]\n\n\nlower bound\na[3:]\na[3], …, a[N]\n\n\nupper bound\na[:5]\na[1], …, a[5]\n\n\nrange\na[2:7]\na[2], …, a[7]\n\n\nrange\na[7:2]\n[]\n\n\nrange\na[5-3:5+2]\na[2], …, a[7]\n\n\nall\na[:]\na[1], …, a[N]\n\n\nall\na[]\na[1], …, a[N]\n\n\n\nThe range indexing with : allows only increasing sequences. Indexing with a decereasing sequence can be made by creating an integer array in the following way:\n  array[6] int ii = reverse(linspaced_int_array(6, 2, 7));\nThen a[ii] evaluates to a[7], …, a[2].\n\n\n\nThe fundamental semantic rule for dealing with multiple indexes is the following. If idxs is a multiple index, then it produces an indexable position in the result. To evaluate that index position in the result, the index is first passed to the multiple index, and the resulting index used.\na[idxs, ...][i, ...] = a[idxs[i], ...][...]\nOn the other hand, if idx is a single index, it reduces the dimensionality of the output, so that\na[idx, ...] = a[idx][...]\nThe only issue is what happens with matrices and vectors. Vectors work just like arrays. Matrices with multiple row indexes and multiple column indexes produce matrices. Matrices with multiple row indexes and a single column index become (column) vectors. Matrices with a single row index and multiple column indexes become row vectors. The types are summarized in the following table.\n\n\nSpecial rules for reducing matrices based on whether the argument is a single or multiple index. Examples are for a matrix a, with integer single indexes i and j and integer array multiple indexes is and js. The same typing rules apply for all multiple indexes.\n\n\n\nexample\nrow index\ncolumn index\nresult type\n\n\n\n\na[i]\nsingle\nn/a\nrow vector\n\n\na[is]\nmultiple\nn/a\nmatrix\n\n\na[i, j]\nsingle\nsingle\nreal\n\n\na[i, js]\nsingle\nmultiple\nrow vector\n\n\na[is, j]\nmultiple\nsingle\nvector\n\n\na[is, js]\nmultiple\nmultiple\nmatrix\n\n\n\nEvaluation of matrices with multiple indexes is defined to respect the following distributivity conditions.\nm[idxs1, idxs2][i, j] = m[idxs1[i], idxs2[j]]\nm[idxs, idx][j] = m[idxs[j], idx]\nm[idx, idxs][j] = m[idx, idxs[j]]\nEvaluation of arrays of matrices and arrays of vectors or row vectors is defined recursively, beginning with the array dimensions.",
+    "text": "Stan also provides a parameterization of the Poisson using the log rate \\(\\alpha = \\log \\lambda\\) as a parameter. This is useful for log-linear Poisson regressions so that the predictor does not need to be exponentiated and passed into the standard Poisson probability function.\n\n\nIf \\(\\alpha \\in \\mathbb{R}\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{PoissonLog}(n|\\alpha) = \\frac{1}{n!} \\, \\exp \\left(n\\alpha -\n\\exp(\\alpha) \\right). \\end{equation*}\\]\n\n\n\nn ~ poisson_log(alpha)\nIncrement target log probability density with poisson_log_lupmf(n | alpha).\nAvailable since 2.0\n \n\n\n\n\n \n\nreal poisson_log_lpmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha\nAvailable since 2.12\n \n\nreal poisson_log_lupmf(ints n | reals alpha) The log Poisson probability mass of n given log rate alpha dropping constant additive terms\nAvailable since 2.25\n \n\nR poisson_log_rng(reals alpha) Generate a Poisson variate with log rate alpha; may only be used in transformed data and generated quantities blocks. alpha must be less than \\(30 \\log 2\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.18",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#function-application.section",
-    "href": "reference-manual/expressions.html#function-application.section",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#poisson-log-glm",
+    "href": "functions-reference/unbounded_discrete_distributions.html#poisson-log-glm",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "Stan provides a range of built in mathematical and statistical functions, which are documented in the built-in function documentation.\nExpressions in Stan may consist of the name of function followed by a sequence of zero or more argument expressions. For instance, log(2.0) is the expression of type real denoting the result of applying the natural logarithm to the value of the real literal 2.0.\nSyntactically, function application has higher precedence than any of the other operators, so that y + log(x) is interpreted as y + (log(x)).\n\n\nEach function has a type signature which determines the allowable type of its arguments and its return type. For instance, the function signature for the logarithm function can be expressed as\nreal log(real);\nand the signature for the lmultiply function is\nreal lmultiply(real, real);\nA function is uniquely determined by its name and its sequence of argument types. For instance, the following two functions are different functions.\nreal mean(array [] real);\nreal mean(vector);\nThe first applies to a one-dimensional array of real values and the second to a vector.\nThe identity conditions for functions explicitly forbids having two functions with the same name and argument types but different return types. This restriction also makes it possible to infer the type of a function expression compositionally by only examining the type of its subexpressions.\n\n\n\nConstants in Stan are nothing more than nullary (no-argument) functions. For instance, the mathematical constants \\(\\pi\\) and \\(e\\) are represented as nullary functions named pi() and e(). See the Stan Functions Reference built-in constants section for a list of built-in constants.\n\n\n\nBecause of integer to real type promotion, rules must be established for which function is called given a sequence of argument types. The scheme employed by Stan is the same as that used by C++, which resolves a function call to the function requiring the minimum number of type promotions.\nFor example, consider a situation in which the following two function signatures have been registered for foo.\nreal foo(real, real);\nint foo(int, int);\nThe use of foo in the expression foo(1.0, 1.0) resolves to foo(real, real), and thus the expression foo(1.0, 1.0) itself is assigned a type of real.\nBecause integers may be promoted to real values, the expression foo(1, 1) could potentially match either foo(real, real) or foo(int, int). The former requires two type promotions and the latter requires none, so foo(1, 1) is resolved to function foo(int, int) and is thus assigned the type int.\nThe expression foo(1, 1.0) has argument types (int, real) and thus does not explicitly match either function signature. By promoting the integer expression 1 to type real, it is able to match foo(real, real), and hence the type of the function expression foo(1, 1.0) is real.\nIn some cases (though not for any built-in Stan functions), a situation may arise in which the function referred to by an expression remains ambiguous. For example, consider a situation in which there are exactly two functions named bar with the following signatures.\nreal bar(real, int);\nreal bar(int, real);\nWith these signatures, the expression bar(1.0, 1) and bar(1, 1.0) resolve to the first and second of the above functions, respectively. The expression bar(1.0, 1.0) is illegal because real values may not be demoted to integers. The expression bar(1, 1) is illegal for a different reason. If the first argument is promoted to a real value, it matches the first signature, whereas if the second argument is promoted to a real value, it matches the second signature. The problem is that these both require one promotion, so the function name bar is ambiguous. If there is not a unique function requiring fewer promotions than all others, as with bar(1, 1) given the two declarations above, the Stan compiler will flag the expression as illegal.\n\n\n\nFor most of the distributions supported by Stan, there is a corresponding random-number generating function. These random number generators are named by the distribution with the suffix _rng. For example, a univariate normal random number can be generated by normal_rng(0, 1); only the parameters of the distribution, here a location (0) and scale (1) are specified because the variate is generated.\n\n\nThe use of random-number generating functions is restricted to the transformed data and generated quantities blocks; attempts to use them elsewhere will result in a parsing error with a diagnostic message. They may also be used in the bodies of user-defined functions whose names end in _rng.\nThis allows the random number generating functions to be used for simulation in general, and for Bayesian posterior predictive checking in particular.\n\n\n\nPosterior predictive checks typically use the parameters of the model to generate simulated data (at the individual and optionally at the group level for hierarchical models), which can then be compared informally using plots and formally by means of test statistics, to the actual data in order to assess the suitability of the model; see Chapter 6 of (Gelman et al. 2013) for more information on posterior predictive checks.",
+    "text": "Stan also supplies a single function for a generalized linear model with Poisson distribution and log link function, i.e. a function for a Poisson regression. This provides a more efficient implementation of Poisson regression than a manually written regression in terms of a Poisson distribution and matrix multiplication.\n\n\nIf \\(x\\in \\mathbb{R}^{n\\cdot m}, \\alpha \\in \\mathbb{R}^n, \\beta\\in\n\\mathbb{R}^m\\), then for \\(y \\in \\mathbb{N}^n\\), \\[\\begin{equation*}\n\\text{PoissonLogGLM}(y|x, \\alpha, \\beta) = \\prod_{1\\leq i \\leq\nn}\\text{Poisson}(y_i|\\exp(\\alpha_i + x_i\\cdot \\beta)). \\end{equation*}\\]\n\n\n\ny ~ poisson_log_glm(x, alpha, beta)\nIncrement target log probability density with poisson_log_glm_lupmf(y | x, alpha, beta).\nAvailable since 2.19\n \n\n\n\n\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.23\n \n\nreal poisson_log_glm_lupmf(array[] int y | row_vector x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, real alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25\n \n\nreal poisson_log_glm_lpmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta.\nAvailable since 2.18\n \n\nreal poisson_log_glm_lupmf(array[] int y | matrix x, vector alpha, vector beta) The log Poisson probability mass of y given the log-rate alpha + x * beta dropping constant additive terms.\nAvailable since 2.25",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#type-inference",
-    "href": "reference-manual/expressions.html#type-inference",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#beta-neg-binomial",
+    "href": "functions-reference/unbounded_discrete_distributions.html#beta-neg-binomial",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "Stan is strongly statically typed, meaning that the implementation type of an expression can be resolved at compile time.\n\n\nThe primitive implementation types for Stan are\nint, real, complex, vector, row_vector,  matrix, complex_vector,\ncomplex_row_vector, complex_matrix\nEvery basic declared type corresponds to a primitive type; the following table shows the mapping from types to their primitive types.\n\n\nThe table shows the variable declaration types of Stan and their corresponding primitive implementation type. Stan functions, operators, and probability functions have argument and result types declared in terms of primitive types plus array dimensionality.\n\n\n\ntype\nprimitive type\n\n\n\n\nint\nint\n\n\nreal\nreal\n\n\nvector\nvector\n\n\nsimplex\nvector\n\n\nunit_vector\nvector\n\n\nsum_to_zero_vector\nvector\n\n\nordered\nvector\n\n\npositive_ordered\nvector\n\n\nrow_vector\nrow_vector\n\n\nmatrix\nmatrix\n\n\ncov_matrix\nmatrix\n\n\ncorr_matrix\nmatrix\n\n\ncholesky_factor_cov\nmatrix\n\n\ncholesky_factor_corr\nmatrix\n\n\ncolumn_stochastic_matrix\nmatrix\n\n\nrow_stochastic_matrix\nmatrix\n\n\nsum_to_zero_matrix\nmatrix\n\n\ncomplex_vector\ncomplex_vector\n\n\ncomplex_row_vector\ncomplex_row_vector\n\n\ncomplex_matrix\ncomplex_matrix\n\n\n\nA full implementation type consists of a primitive implementation type and an integer array dimensionality greater than or equal to zero. These will be written to emphasize their array-like nature. For example, array [] real has an array dimensionality of 1, int an array dimensionality of 0, and array [,,] int an array dimensionality of 3. The implementation type matrix[ , , ] has a total of five dimensions and takes up to five indices, three from the array and two from the matrix.\nRecall that the array dimensions come before the matrix or vector dimensions in an expression such as the following declaration of a three-dimensional array of matrices.\narray[I, J, K] matrix[M, N] a;\nThe matrix a is indexed as a[i, j, k, m, n] with the array indices first, followed by the matrix indices, with a[i, j, k] being a matrix and a[i, j, k, m] being a row vector.\n\n\n\n\nStan’s type inference rules define the implementation type of an expression based on a background set of variable declarations. The rules work bottom up from primitive literal and variable expressions to complex expressions.\n\n\n\nThere are two basic promotion rules,\n\nint types may be promoted to real, and\nreal types may be promoted to complex.\n\nPlus, promotion is transitive, so that\n\nif type U can be promoted to type V and type V can be promoted to type T, then U can be promoted to T.\n\nThe first rule means that expressions of type int may be used anywhere an expression of type real is specified, namely in assignment or function argument passing. An integer is promoted to real by casting it in the underlying C++ code.\nThe remaining rules have to do with covariant typing rules, which say that a container of type U may be promoted to a container of the same shape of type T if U can be promoted to T. For vector and matrix types, this induces three rules,\n\nvector may be promoted to complex_vector,\nrow_vector may be promoted to complex_row_vector\nmatrix may be promoted to complex_matrix.\n\nFor array types, there’s a single rule\n\narray[...] U may be promoted to array[...] T if U can be promoted to T.\n\nFor example, this means array[,] int may be used where array [,] real or array [,] complex is required; as another example, array[] real may be used anywhere array[] complex is required.\nTuples have the natural extension of the above rules, applied to all sub-types at once\n\nA tuple(U1, ..., UN) may be promoted to a tuple(T1, ..., TN) if every Un can be promoted to Tn for n in 1:N\n\n\n\nAn integer literal expression such as 42 is of type int. Real literals such as 42.0 are of type real. Imaginary literals such as -17i are of type complex. the expression 7 - 2i acts like a complex literal, but technically it combines a real literal 7 and an imaginary literal 2i through subtraction.\n\n\n\nThe type of a variable declared locally or in a previous block is determined by its declaration. The type of a loop variable is int.\nThere is always a unique declaration for each variable in each scope because Stan prohibits the redeclaration of an already-declared variables.1\n\n\n\nIf x is an expression of total dimensionality greater than or equal to \\(N\\), then the type of expression e[i1, i2, ..., iN] is the same as that of e[i1][i2]...[iN], so it suffices to define the type of a singly-indexed function. Suppose e is an expression and i is an expression of primitive type int. Then\n\nif e is an expression of type array[i1, i2, ..., iN] T and k, i1, …, iN are expressions of type int, then e[k] is an expression of type array[i2, ..., iN] T,\nif e is an expression of type array[i] T with i and k expressions of type int, then e[k] is of type T,\nif e has implementation type vector or row_vector, dimensionality 0, then e[i] has implementation type real,\nif e has implementation type matrix, then e[i] has type row_vector,\nif e has implementation type complex_vector or complex_row_vector and i is an expression of type int, then e[i] is an expression of type complex, and\nif e has implementation type complex_matrix, and i is an expression of type int, then e[i] is an expression of type complex_row_vector.\n\n\n\n\nIf f is the name of a function and e1,...,eN are expressions for \\(N \\geq 0\\), then f(e1,...,eN) is an expression whose type is determined by the return type in the function signature for f given e1 through eN. Recall that a function signature is a declaration of the argument types and the result type.\nIn looking up functions, binary operators like real * real are defined as operator*(real, real) in the documentation and index.\nIn matching a function definition, all of the promotion rules are in play (integers may be promoted to reals, reals to complex, and containers may be promoted if their types are promoted). For example, arguments of type int may be promoted to type real or complex if necessary (see the subsection on type promotion in the function application section, a real argument will be promoted to complex if necessary, a vector will be promoted to complex_vector if necessary, and so on.\nIn general, matrix operations return the lowest inferable type. For example, row_vector * vector returns a value of type real, which is declared in the function documentation and index as real operator*(row_vector, vector).",
+    "text": "If \\(r \\in \\mathbb{R}^+\\), \\(\\alpha \\in \\mathbb{R}^+\\), and \\(\\beta \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}\\), \\[\\begin{equation*}\n\\text{BetaNegBinomial}(n|r,\\alpha,\\beta) = \\frac {\\Gamma (n+r )}{n!\\;\\Gamma (r )}\n\\frac {\\mathrm {B} (\\beta+n,\\alpha +r )}{\\mathrm {B} (\\beta,\\alpha )}. \\end{equation*}\\]\n\n\n\nn ~ beta_neg_binomial(r,alpha,beta)\nIncrement target log probability density with beta_neg_binomial_lupmf(n | r, alpha, beta).\nAvailable since 2.36\n \n\n\n\n\n \n\nreal beta_neg_binomial_lpmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lupmf(ints n | reals r, reals alpha, reals beta) The log beta negative binomial probability mass of n given parameters r, alpha and beta dropping constant additive terms.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_cdf(ints n | reals r, reals alpha, reals beta) The beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lcdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nreal beta_neg_binomial_lccdf(ints n | reals r, reals alpha, reals beta) The log of the beta negative binomial complementary cumulative distribution function of n given parameters r, alpha and beta.\nAvailable since 2.36\n \n\nR beta_neg_binomial_rng(reals r, reals alpha, reals beta) Generate a beta negative binomial variate with parameters r, alpha and beta; may only be used in transformed data and generated quantities blocks. r \\(\\cdot\\) beta \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.36",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#higher-order-functions",
-    "href": "reference-manual/expressions.html#higher-order-functions",
-    "title": "Expressions",
+    "objectID": "functions-reference/unbounded_discrete_distributions.html#yule-simon",
+    "href": "functions-reference/unbounded_discrete_distributions.html#yule-simon",
+    "title": "Unbounded Discrete Distributions",
     "section": "",
-    "text": "There are several expression constructions in Stan that act as higher-order functions.2\nThe higher-order functions and the signature of their argument functions are listed in the following pair of tables.\n\n\nHigher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.\n\n\n\n\n\n\n\n\n\nfunction\nparameter or data args\ndata args\nreturn type\n\n\n\n\nalgebra_solver\nvector, vector\narray [] real, array [] real\nvector\n\n\nalgebra_solver_newton\nvector, vector\narray [] real, array [] real\nvector\n\n\nintegrate_1d,\nreal, real, array [] real\narray [] real, array [] real\nreal\n\n\nintegrate_ode_X,\nreal, array [] real, array [] real\narray [] real, array [] real\narray [] real\n\n\nmap_rect\nvector, vector\narray [] real, array [] real\nvector\n\n\n\nFor example, the integrate_ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  array [] real foo(real t,\n                    array [] real y,\n                    array [] real theta,\n                    array [] real x_r,\n                    array [] real x_i) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\narray[2] real y0;\nreal t0;\narray[T] real ts;\narray[1] real theta;\narray[0] real x_r;\narray[0] int x_i;\n// ...\narray[T, 2] real y_hat = integrate_ode_rk45(foo, y0, t0,\n                                              ts, theta, x_r, x_i);\nThe function argument is foo, the name of the user-defined function; as shown in the higher-order functions table, integrate_ode_rk45 takes a real array, a real, three more real arrays, and an integer array as arguments and returns 2D real array.\n\n\n\nVariadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.\n\n\n\nfunction\nrestricted args\nreturn type\n\n\n\n\nsolve_X\nvector\nvector\n\n\node_X,\nvector, real, array [] real\nvector[]\n\n\nreduce_sum\narray[] T, T1, T2\nreal\n\n\n\nT, T1, and T2 can be any Stan type.\nFor example, the ode_rk45 function can be used to integrate differential equations in Stan:\nfunctions {\n  vector foo(real t, vector y, real theta, vector beta,\n            array [] real x_i, int index) {\n    // ...\n  }\n}\n// ...\nint&lt;lower=1&gt; T;\nvector[2] y0;\nreal t0;\narray[T] real ts;\nreal theta;\nvector[7] beta;\narray[10] int x_i;\nint index;\n// ...\nvector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta,\n                              beta, x_i, index);\nThe function argument is foo, the name of the user-defined function. As shown in the variadic higher-order functions table, ode_rk45 takes a real, a vector, a real, a real array, and a sequence of arguments whose types match those at the end of foo and returns an array of vectors.\n\n\n\nThe function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.\n\n\n\nSome of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.\nFor user-defined functions the qualifier data may be prepended to the type to restrict the argument to data-only variables.",
+    "text": "If \\(\\alpha \\in \\mathbb{R}^+\\), then for \\(n \\in \\mathbb{N}^+=\\{1,2,...\\}\\), \\[\\begin{equation*}\n\\text{YuleSimon}(n|\\alpha) = \\alpha \\, \\mathrm{B}(\\alpha + 1, n) = \\alpha \\, \\frac{\\Gamma(n) \\, \\Gamma(\\alpha + 1)}{\\Gamma(n + \\alpha + 1)}.\n\\end{equation*}\\]\n\n\n\nn ~ yule_simon(alpha)\nIncrement target log probability density with yule_simon_lupmf(n | alpha).\nAvailable since 2.39\n \n\n\n\n\n \n\nreal yule_simon_lpmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lupmf(ints n | reals alpha) The log Yule Simon probability mass of n given parameter alpha dropping constant additive terms.\nAvailable since 2.39\n \n\nreal yule_simon_cdf(ints n | reals alpha) The Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lcdf(ints n | reals alpha) The log of the Yule Simon cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nreal yule_simon_lccdf(ints n | reals alpha) The log of the Yule Simon complementary cumulative distribution function of n given parameter alpha.\nAvailable since 2.39\n \n\nR yule_simon_rng(reals alpha) Generate a Yule Simon variate with parameter alpha; may only be used in transformed data and generated quantities blocks. alpha \\(/\\) (alpha\\(-1\\)) must be less than \\(2 ^ {29}\\). For a description of argument and return types, see section vectorized function signatures.\nAvailable since 2.39",
     "crumbs": [
-      "Reference Manual",
-      "Language",
-      "Expressions"
+      "Functions Reference",
+      "Discrete Distributions",
+      "Unbounded Discrete Distributions"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#chain-rule-and-derivatives",
-    "href": "reference-manual/expressions.html#chain-rule-and-derivatives",
-    "title": "Expressions",
+    "objectID": "reference-manual/analysis.html",
+    "href": "reference-manual/analysis.html",
+    "title": "Posterior Analysis",
     "section": "",
-    "text": "Derivatives of the log probability function defined by a model are used in several ways by Stan. The Hamiltonian Monte Carlo samplers, including NUTS, use gradients to guide updates. The BFGS optimizers also use gradients to guide search for posterior modes.\n\n\nUnlike evaluations in pure mathematics, evaluation of derivatives in Stan is done by applying the chain rule on an expression-by-expression basis, evaluating using floating-point arithmetic. As a result, models such as the following are problematic for inference involving derivatives.\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(sqrt(x - x), 1);\n}\nAlgebraically, the distribution statement in the model could be reduced to\n  x ~ normal(0, 1);\nand it would seem the model should produce unit normal draws for x. But rather than canceling, the expression sqrt(x -   x) causes a problem for derivatives. The cause is the mechanistic evaluation of the chain rule,\n\\[\n\\begin{array}{rcl}\n\\frac{d}{dx} \\sqrt{x - x}\n& = &\n\\frac{1}{2 \\sqrt{x - x}} \\times \\frac{d}{dx} (x - x)\n\\\\[4pt]\n& = &\n\\frac{1}{0} \\times (1 - 1)\n\\\\[4pt]\n& = &\n\\infty \\times 0\n\\\\[4pt]\n& = & \\mathrm{NaN}.\n\\end{array}\n\\]\nRather than the \\(x - x\\) canceling out, it introduces a 0 into the numerator and denominator of the chain-rule evaluation.\nThe only way to avoid this kind problem is to be careful to do the necessary algebraic reductions as part of the model and not introduce expressions like sqrt(x - x) for which the chain rule produces not-a-number values.\n\n\n\nThe best way to diagnose whether something is going wrong with the derivatives is to use the test-gradient option to the sampler or optimizer inputs; this option is available in both Stan and RStan (though it may be slow, because it relies on finite differences to make a comparison to the built-in automatic differentiation).\nFor example, compiling the above model to an executable sqrt-x-minus-x in CmdStan, the test can be run as\n&gt; ./sqrt-x-minus-x diagnose test=gradient\nwhich produces\n...\nTEST GRADIENT MODE\n\n Log probability=-0.393734\n\n param idx           value           model     finite diff           error\n         0       -0.887393             nan               0             nan\nEven though finite differences calculates the right gradient of 0, automatic differentiation follows the chain rule and produces a not-a-number output.",
+    "text": "Stan uses Markov chain Monte Carlo (MCMC) techniques to generate draws from the posterior distribution for full Bayesian inference. Markov chain Monte Carlo (MCMC) methods were developed for situations in which it is not straightforward to make independent draws Metropolis et al. (1953).\nStan’s variational inference algorithm provides draws from the variational approximation to the posterior which may be analyzed just as any other MCMC output, despite the fact that it is not actually a Markov chain.\nStan’s Laplace algorithm produces a sample from a normal approximation centered at the mode of a distribution in the unconstrained space. If the mode is a maximum a posteriori (MAP) estimate, the sample provides an estimate of the mean and standard deviation of the posterior distribution. If the mode is a maximum likelihood estimate (MLE), the sample provides an estimate of the standard error of the likelihood.\n\n\nA Markov chain is a sequence of random variables \\(\\theta^{(1)},\n\\theta^{(2)},\\ldots\\) where each variable is conditionally independent of all other variables given the value of the previous value. Thus if \\(\\theta = \\theta^{(1)}, \\theta^{(2)},\\ldots, \\theta^{(N)}\\), then\n\\[\np(\\theta) = p(\\theta^{(1)}) \\prod_{n=2}^N p(\\theta^{(n)}|\\theta^{(n-1)}).\n\\]\nStan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the Hamiltonian Monte Carlo chapter.\nThe Markov chains Stan and other MCMC samplers generate are ergodic in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of \\(\\theta\\) from another. The Markov chains are also stationary, meaning that the transition probabilities do not change at different positions in the chain, so that for \\(n, n' \\geq 0\\), the probability function \\(p(\\theta^{(n+1)}|\\theta^{(n)})\\) is the same as \\(p(\\theta^{(n'+1)}|\\theta^{(n')})\\) (following the convention of overloading random and bound variables and picking out a probability function by its arguments).\nStationary Markov chains have an equilibrium distribution on states in which each has the same marginal probability function, so that \\(p(\\theta^{(n)})\\) is the same probability function as \\(p(\\theta^{(n+1)})\\). In Stan, this equilibrium distribution \\(p(\\theta^{(n)})\\) is the target density \\(p(\\theta)\\) defined by a Stan program, which is typically a proper Bayesian posterior density \\(p(\\theta | y)\\) defined on the log scale up to a constant.\nUsing MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.\nStan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.\n\n\n\nBy definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when \\(p(\\theta^{(n)})\\) is the target density). The following point cannot be expressed strongly enough:\n\nIn theory, convergence is only guaranteed asymptotically as the number of draws grows without bound.\nIn practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available.\n\n\n\n\nTo establish basic notation, suppose a target Bayesian posterior density \\(p(\\theta | y)\\) given real-valued vectors of parameters \\(\\theta\\) and real- and discrete-valued data \\(y\\).1\nAn MCMC sample consists of a set of a sequence of \\(M\\) Markov chains, each consisting of an ordered sequence of \\(N\\) draws from the posterior.2 The sample thus consists of \\(M \\times N\\) draws from the posterior.\n\n\nOne way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the Gelman and Rubin (1992) potential scale reduction statistic, \\(\\hat{R}\\). The \\(\\hat{R}\\) statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and \\(\\hat{R}\\) will be one. If the chains have not converged to a common distribution, the \\(\\hat{R}\\) statistic will be greater than one.\nGelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for \\(\\hat{R}\\) are below some threshold. Vehtari et al. (2021) suggest in general to use a threshold \\(1.01\\), but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.\nThe \\(\\hat{R}\\) statistic is defined for a set of \\(M\\) Markov chains, \\(\\theta_m\\), each of which has \\(N\\) draws \\(\\theta^{(n)}_m\\). The between-chain variance estimate is\n\\[\nB\n=\n\\frac{N}{M-1}\n\\,\n\\sum_{m=1}^M (\\bar{\\theta}^{(\\bullet)}_{m}\n                - \\bar{\\theta}^{(\\bullet)}_{\\bullet})^2,\n\\]\nwhere\n\\[\n\\bar{\\theta}_m^{(\\bullet)}\n= \\frac{1}{N} \\sum_{n = 1}^N \\theta_m^{(n)}\n\\]\nand\n\\[\n\\bar{\\theta}^{(\\bullet)}_{\\bullet}\n= \\frac{1}{M} \\, \\sum_{m=1}^M \\bar{\\theta}_m^{(\\bullet)}.\n\\]\nThe within-chain variance is averaged over the chains,\n\\[\nW = \\frac{1}{M} \\, \\sum_{m=1}^M s_m^2,\n\\]\nwhere\n\\[\ns_m^2\n=\n\\frac{1}{N-1}\n\\, \\sum_{n=1}^N (\\theta^{(n)}_m - \\bar{\\theta}^{(\\bullet)}_m)^2.\n\\]\nThe variance estimator is a mixture of the within-chain and cross-chain sample variances,\n\\[\n\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)\n= \\frac{N-1}{N}\\, W \\, + \\, \\frac{1}{N} \\, B.\n\\]\nFinally, the potential scale reduction statistic is defined by\n\\[\n\\hat{R}\n\\, = \\,\n\\sqrt{\\frac{\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)}{W}}.\n\\]\n\n\n\nBefore Stan calculating the potential-scale-reduction statistic \\(\\hat{R}\\), each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have \\(\\hat{R}\\) values near unity. In this case, splitting each chain into two parts leads to \\(\\hat{R}\\) values substantially greater than 1 because the first half of each chain has not mixed with the second half.\n\n\n\nSplit R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following Vehtari et al. (2021), we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.\nRank normalization proceeds as follows:\n\nFirst, replace each value \\(\\theta^{(nm)}\\) by its rank \\(r^{(nm)}\\) within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.\nSecond, transform ranks to normal scores using the inverse normal transformation and a fractional offset:\n\n\\[\nz_{(nm)} = \\Phi^{-1} \\left( \\frac{r_{(nm)} - 3/8}{S - 1/4} \\right)\n\\]\nTo further improve sensitivity to chains having different scales,\nrank normalized R-hat is computed also for the for the corresponding folded draws \\(\\zeta^{(mn)}\\), absolute deviations from the median, \\[\n\\label{zeta}\n\\zeta^{(mn)} = \\left|\\theta^{(nm)}-{\\rm median}(\\theta)\\right|.\n\\] The rank normalized split-\\(\\widehat{R}\\) measure computed on the \\(\\zeta^{(mn)}\\) values is called -\\(\\widehat{R}\\). This measures convergence in the tails rather than in the bulk of the distribution.\nTo obtain a single conservative \\(\\widehat{R}\\) estimate, we propose to report the maximum of rank normalized split-\\(\\widehat{R}\\) and rank normalized folded-split-\\(\\widehat{R}\\) for each parameter.\nBulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See Effective Sample Size section for details on how ESS is estimated.\n\n\n\nA question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.\nFor example, consider the value lp__, which is the log posterior density (up to a constant).3\nIt is thus a mistake to declare convergence in any practical sense if lp__ has not converged, because different chains are really in different parts of the space. Yet measuring convergence for lp__ is particularly tricky, as noted below.\n\n\nMarkov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.\n\n\n\nThe \\(\\hat{R}\\) statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.\nThe transformation from unconstrained space to constrained space is just another function, so does not effect convergence.\nDifferent functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, lp__ and other measured quantities should at least be consistent.\nThe obvious difference in lp__ is that it tends to vary quickly with position and is consequently susceptible to outliers.\n\n\n\nThe question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.\nSo what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.\n\n\n\nFirstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better \\(\\hat{R}\\). Functions vulnerable to such outliers may show fragility.\nSecondly, use of the \\(\\hat{R}\\) statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (lp__) almost never looks Gaussian, instead it features long tails that can lead to large \\(\\hat{R}\\) even in the large \\(N\\) limit. Tweaks to \\(\\hat{R}\\), such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the \\(\\hat{R}\\) statistic more accurate.\n\n\n\n“Convergence” is a global property and holds for all integrable functions at once, but employing the \\(\\hat{R}\\) statistic requires additional assumptions and thus may not work for all functions equally well.\nNote that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.\n\n\n\n\n\nThe second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see Charles J. Geyer (2011).\nStan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).\nUnlike most packages, the particular calculations used by Stan follow those for split-\\(\\hat{R}\\), which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see Gelman et al. (2013) and Vehtari et al. (2021).\n\n\nThe amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size \\(N\\). Given dependent sample, the sample size is replaced with the effective sample size \\(N_{\\mathrm{eff}}\\).\nFor example, Monte Carlo standard error (MCSE) is proportional to \\(1 / \\sqrt{N_{\\mathrm{eff}}}\\) rather than \\(1/\\sqrt{N}\\).\nThe effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation \\(\\rho_t\\) at lag \\(t \\geq 0\\) for a chain with joint probability function \\(p(\\theta)\\) with mean \\(\\mu\\) and variance \\(\\sigma^2\\) is defined to be\n\\[\n\\rho_t\n=\n\\frac{1}{\\sigma^2} \\, \\int_{\\Theta} (\\theta^{(n)} - \\mu)\n(\\theta^{(n+t)} - \\mu) \\, p(\\theta) \\, d\\theta.\n\\]\nThis is the correlation between the two chains offset by \\(t\\) positions (i.e., a lag in time-series terminology). Because we know \\(\\theta^{(n)}\\) and \\(\\theta^{(n+t)}\\) have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields\n\\[\n\\rho_t\n= \\frac{1}{\\sigma^2}\n\\, \\int_{\\Theta}\n      \\theta^{(n)} \\, \\theta^{(n+t)} \\, p(\\theta)\n   \\, d\\theta - \\frac{\\mu^2}{\\sigma^2}.\n\\]\nThe effective sample size of \\(N\\) draws generated by a process with autocorrelations \\(\\rho_t\\) is defined by \\[\nN_{\\mathrm{eff}}\n\\ = \\\n\\frac{N}{\\sum_{t = -\\infty}^{\\infty} \\rho_t}\n\\ = \\\n\\frac{N}{1 + 2 \\sum_{t = 1}^{\\infty} \\rho_t}.\n\\]\nFor independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.\n\n\n\nIn practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-\\(\\hat{R}\\) based effective sample size estimator, based on multiple chains. As before, each chain \\(\\theta_m\\) will be assumed to be of length \\(N\\).\nStan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see Charles J. Geyer (2011) for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates \\(\\hat{\\rho}_{t,m}\\) at lag \\(t\\) from multiple chains \\(m \\in\n(1,\\ldots,M)\\) are combined with within-sample variance estimate \\(W\\) and multi-chain variance estimate \\(\\widehat{\\mbox{var}}^{+}\\) introduced in the previous section to compute the combined autocorrelation at lag \\(t\\) as\n\\[\n\\hat{\\rho}_t\n= 1 - \\frac{\\displaystyle W\n              - \\textstyle \\frac{1}{M}\\sum_{m=1}^M s_m^2 \\hat{\\rho}_{t,m}}\n        {\\widehat{\\mbox{var}}^{+}}.\n\\]\nIf the chains have not converged, the variance estimator \\(\\widehat{\\mbox{var}}^{+}\\) will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.\nBecause of the noise in the correlation estimates \\(\\hat{\\rho}_t\\) as \\(t\\) increases, a typical truncated sum of \\(\\hat{\\rho}_t\\) is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise Charles J. Geyer (1992), Charles J. Geyer (2011). Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as\n\\[\n\\hat{N}_{\\mathrm{eff}} = \\frac{M \\cdot N}{\\hat{\\tau}},\n\\]\nwhere\n\\[\n\\hat{\\tau} = 1 + 2 \\sum_{t=1}^{2m+1} \\hat{\\rho}_t = -1 + 2 \\sum_{t'=0}^{m} \\hat{P}_{t'},\n\\]\nwhere \\(\\hat{P}_{t'}=\\hat{\\rho}_{2t'}+\\hat{\\rho}_{2t'+1}\\). Initial positive sequence estimators is obtained by choosing the largest \\(m\\) such that \\(\\hat{P}_{t'}&gt;0, \\quad t' = 1,\\ldots,m\\). The initial monotone sequence is obtained by further reducing \\(\\hat{P}_{t'}\\) to the minimum of the preceding ones so that the estimated sequence is monotone.\n\n\n\nThe posterior standard deviation of a parameter \\(\\theta_n\\) conditioned on observed data \\(y\\) is just the standard deviation of the posterior density \\(p(\\theta_n | y)\\). This is estimated by the standard deviation of the combined posterior draws across chains,\n\\[\n\\hat{\\sigma}_n = \\mathrm{sd}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n).\n\\]\nThe previous section showed how to estimate \\(N_{\\mathrm{eff}}\\) for a parameter \\(\\theta_n\\) based on multiple chains of posterior draws.\nThe mean of the posterior draws of \\(\\theta_n\\) \\[\n\\hat{\\theta}_n\n= \\mathrm{mean}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n)\n\\]\nis treated as an estimator of the true posterior mean,\n\\[\n\\mathbb{E}[\\theta_n \\mid y]\n\\ = \\\n\\int_{-\\infty}^{\\infty}\n    \\, \\theta \\, p(\\theta | y)\n\\, \\mathrm{d}\\theta_n,\n\\]\nbased the observed data \\(y\\).\nThe standard error for the estimator \\(\\hat{\\theta}_n\\) is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as \\(\\hat{\\sigma}_n / \\sqrt{N_{\\mathrm{eff}}}\\). The smaller the standard error, the closer the estimate \\(\\hat{\\theta}_n\\) is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see Charles J. Geyer (2011) for more details of the MCMC central limit theorem.\n\n\n\nIn complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag \\(t\\), \\(\\rho_t\\), decreases as the lag, \\(t\\), increases. In this situation, thinning the sample by keeping only every \\(N\\)-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.\nFor instance, consider generating one thousand posterior draws in one of the following two ways.\n\nGenerate 1000 draws after convergence and save all of them.\nGenerate 10,000 draws after convergence and save every tenth draw.\n\nEven though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation \\(\\rho_t\\) for the thinned sequence is equivalent to \\(\\rho_{10t}\\) in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.\nNow contrast the second approach above with the unthinned alternative,\n\nGenerate 10,000 draws after convergence and save every draw.\n\nThis will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, the only reason to thin a sample is to reduce memory requirements.\nIf draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Expressions"
+      "Algorithms",
+      "Posterior Analysis"
     ]
   },
   {
-    "objectID": "reference-manual/expressions.html#footnotes",
-    "href": "reference-manual/expressions.html#footnotes",
-    "title": "Expressions",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nLanguages such as C++ and R allow the declaration of a variable of a given name in a narrower scope to hide (take precedence over for evaluation) a variable defined in a containing scope.↩︎\nInternally, they are implemented as their own expression types because Stan doesn’t have object-level functional types (yet).↩︎",
+    "objectID": "reference-manual/analysis.html#markov-chains",
+    "href": "reference-manual/analysis.html#markov-chains",
+    "title": "Posterior Analysis",
+    "section": "",
+    "text": "A Markov chain is a sequence of random variables \\(\\theta^{(1)},\n\\theta^{(2)},\\ldots\\) where each variable is conditionally independent of all other variables given the value of the previous value. Thus if \\(\\theta = \\theta^{(1)}, \\theta^{(2)},\\ldots, \\theta^{(N)}\\), then\n\\[\np(\\theta) = p(\\theta^{(1)}) \\prod_{n=2}^N p(\\theta^{(n)}|\\theta^{(n-1)}).\n\\]\nStan uses Hamiltonian Monte Carlo to generate a next state in a manner described in the Hamiltonian Monte Carlo chapter.\nThe Markov chains Stan and other MCMC samplers generate are ergodic in the sense required by the Markov chain central limit theorem, meaning roughly that there is a reasonable chance of reaching one value of \\(\\theta\\) from another. The Markov chains are also stationary, meaning that the transition probabilities do not change at different positions in the chain, so that for \\(n, n' \\geq 0\\), the probability function \\(p(\\theta^{(n+1)}|\\theta^{(n)})\\) is the same as \\(p(\\theta^{(n'+1)}|\\theta^{(n')})\\) (following the convention of overloading random and bound variables and picking out a probability function by its arguments).\nStationary Markov chains have an equilibrium distribution on states in which each has the same marginal probability function, so that \\(p(\\theta^{(n)})\\) is the same probability function as \\(p(\\theta^{(n+1)})\\). In Stan, this equilibrium distribution \\(p(\\theta^{(n)})\\) is the target density \\(p(\\theta)\\) defined by a Stan program, which is typically a proper Bayesian posterior density \\(p(\\theta | y)\\) defined on the log scale up to a constant.\nUsing MCMC methods introduces two difficulties that are not faced by independent sample Monte Carlo methods. The first problem is determining when a randomly initialized Markov chain has converged to its equilibrium distribution. The second problem is that the draws from a Markov chain may be correlated or even anti-correlated, and thus the central limit theorem’s bound on estimation error no longer applies. These problems are addressed in the next two sections.\nStan’s posterior analysis tools compute a number of summary statistics, estimates, and diagnostics for Markov chain Monte Carlo (MCMC) sample. Stan’s estimators and diagnostics are more robust in the face of non-convergence, antithetical sampling, and long-term Markov chain correlations than most of the other tools available. The algorithms Stan uses to achieve this are described in this chapter.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Expressions"
+      "Algorithms",
+      "Posterior Analysis"
     ]
   },
   {
-    "objectID": "reference-manual/index.html",
-    "href": "reference-manual/index.html",
-    "title": "Stan Reference Manual",
+    "objectID": "reference-manual/analysis.html#convergence",
+    "href": "reference-manual/analysis.html#convergence",
+    "title": "Posterior Analysis",
     "section": "",
-    "text": "This is the official reference manual for Stan’s programming language for coding probability models, inference algorithms for fitting models and making predictions, and posterior analysis tools for evaluating the results. This manual applies to all Stan interfaces.\nThe first part of the reference manual provides a full specification of the Stan programming language. The language is responsible for defining a log density function conditioned on data. Typically, this is a Bayesian posterior, but it may also be a penalized likelihood function. The second part of the manual specifies the inference algorithms and posterior inference tools. The third part provides auxiliary information about the use of Stan.\nDownload the pdf version of this manual.\n\nCopyright and trademark\n\nCopyright 2011–2025, Stan Development Team and their assignees.\nThe Stan name and logo are registered trademarks of NumFOCUS.\n\n\n\nLicensing\n\nText content: CC-BY ND 4.0 license\nComputer code: BSD 3-clause license\nLogo: Stan logo usage guidelines\n\n\n\n\n\n Back to top",
+    "text": "By definition, a Markov chain samples from the target distribution only after it has converged to equilibrium (i.e., equilibrium is defined as being achieved when \\(p(\\theta^{(n)})\\) is the target density). The following point cannot be expressed strongly enough:\n\nIn theory, convergence is only guaranteed asymptotically as the number of draws grows without bound.\nIn practice, diagnostics must be applied to monitor convergence for the finite number of draws actually available.",
     "crumbs": [
       "Reference Manual",
-      "Stan Reference Manual"
+      "Algorithms",
+      "Posterior Analysis"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html",
-    "href": "reference-manual/licenses.html",
-    "title": "Licenses and Dependencies",
+    "objectID": "reference-manual/analysis.html#notation-for-samples-chains-and-draws",
+    "href": "reference-manual/analysis.html#notation-for-samples-chains-and-draws",
+    "title": "Posterior Analysis",
     "section": "",
-    "text": "Stan and its dependent libraries, are distributed under generous, freedom-respecting licenses approved by the Open Source Initiative.\nIn particular, the licenses for Stan and its dependent libraries have no “copyleft” provisions requiring applications of Stan to be open source if they are redistributed.\nThis chapter specifies the licenses for the libraries on which Stan’s math library, language, and algorithms depend. The last tool mentioned, Google Test, is only used for testing and is not needed to run Stan.\n\n\nStan is distributed under\n\nBSD 3-clause license (BSD New)\n\nThe copyright holder of each contribution is the developer or his or her assignee.1\n\n\n\nStan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the\n\nBoost Software License version 1.0\n\nThe copyright for each Boost package is held by its developers or their assignees.\n\n\n\nStan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the\n\nMozilla Public License, version 2\n\nThe copyright of Eigen is owned jointly by its developers or their assignees.\n\n\n\nStan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.\n\n\n\nStan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the\n\nApache License, version 2\n\nThe copyright of TBB is owned by Intel Corporation.\n\n\n\nStan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of Google Test is owned by Google, Inc.",
+    "text": "To establish basic notation, suppose a target Bayesian posterior density \\(p(\\theta | y)\\) given real-valued vectors of parameters \\(\\theta\\) and real- and discrete-valued data \\(y\\).1\nAn MCMC sample consists of a set of a sequence of \\(M\\) Markov chains, each consisting of an ordered sequence of \\(N\\) draws from the posterior.2 The sample thus consists of \\(M \\times N\\) draws from the posterior.\n\n\nOne way to monitor whether a chain has converged to the equilibrium distribution is to compare its behavior to other randomly initialized chains. This is the motivation for the Gelman and Rubin (1992) potential scale reduction statistic, \\(\\hat{R}\\). The \\(\\hat{R}\\) statistic measures the ratio of the average variance of drawss within each chain to the variance of the pooled draws across chains; if all chains are at equilibrium, these will be the same and \\(\\hat{R}\\) will be one. If the chains have not converged to a common distribution, the \\(\\hat{R}\\) statistic will be greater than one.\nGelman and Rubin’s recommendation is that the independent Markov chains be initialized with diffuse starting values for the parameters and sampled until all values for \\(\\hat{R}\\) are below some threshold. Vehtari et al. (2021) suggest in general to use a threshold \\(1.01\\), but othe thresholds can be used depending on the use case. Stan allows users to specify initial values for parameters and it is also able to draw diffuse random initializations automatically satisfying the declared parameter constraints.\nThe \\(\\hat{R}\\) statistic is defined for a set of \\(M\\) Markov chains, \\(\\theta_m\\), each of which has \\(N\\) draws \\(\\theta^{(n)}_m\\). The between-chain variance estimate is\n\\[\nB\n=\n\\frac{N}{M-1}\n\\,\n\\sum_{m=1}^M (\\bar{\\theta}^{(\\bullet)}_{m}\n                - \\bar{\\theta}^{(\\bullet)}_{\\bullet})^2,\n\\]\nwhere\n\\[\n\\bar{\\theta}_m^{(\\bullet)}\n= \\frac{1}{N} \\sum_{n = 1}^N \\theta_m^{(n)}\n\\]\nand\n\\[\n\\bar{\\theta}^{(\\bullet)}_{\\bullet}\n= \\frac{1}{M} \\, \\sum_{m=1}^M \\bar{\\theta}_m^{(\\bullet)}.\n\\]\nThe within-chain variance is averaged over the chains,\n\\[\nW = \\frac{1}{M} \\, \\sum_{m=1}^M s_m^2,\n\\]\nwhere\n\\[\ns_m^2\n=\n\\frac{1}{N-1}\n\\, \\sum_{n=1}^N (\\theta^{(n)}_m - \\bar{\\theta}^{(\\bullet)}_m)^2.\n\\]\nThe variance estimator is a mixture of the within-chain and cross-chain sample variances,\n\\[\n\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)\n= \\frac{N-1}{N}\\, W \\, + \\, \\frac{1}{N} \\, B.\n\\]\nFinally, the potential scale reduction statistic is defined by\n\\[\n\\hat{R}\n\\, = \\,\n\\sqrt{\\frac{\\widehat{\\mbox{var}}^{+}\\!(\\theta|y)}{W}}.\n\\]\n\n\n\nBefore Stan calculating the potential-scale-reduction statistic \\(\\hat{R}\\), each chain is split into two halves. This provides an additional means to detect non-stationarity in the individual chains. If one chain involves gradually increasing values and one involves gradually decreasing values, they have not mixed well, but they can have \\(\\hat{R}\\) values near unity. In this case, splitting each chain into two parts leads to \\(\\hat{R}\\) values substantially greater than 1 because the first half of each chain has not mixed with the second half.\n\n\n\nSplit R-hat and the effective sample size (ESS) are well defined only if the marginal posteriors have finite mean and variance. Therefore, following Vehtari et al. (2021), we compute the rank normalized parameter values and then feed them into the formulas for split R-hat and ESS.\nRank normalization proceeds as follows:\n\nFirst, replace each value \\(\\theta^{(nm)}\\) by its rank \\(r^{(nm)}\\) within the pooled draws from all chains. Average rank for ties are used to conserve the number of unique values of discrete quantities.\nSecond, transform ranks to normal scores using the inverse normal transformation and a fractional offset:\n\n\\[\nz_{(nm)} = \\Phi^{-1} \\left( \\frac{r_{(nm)} - 3/8}{S - 1/4} \\right)\n\\]\nTo further improve sensitivity to chains having different scales,\nrank normalized R-hat is computed also for the for the corresponding folded draws \\(\\zeta^{(mn)}\\), absolute deviations from the median, \\[\n\\label{zeta}\n\\zeta^{(mn)} = \\left|\\theta^{(nm)}-{\\rm median}(\\theta)\\right|.\n\\] The rank normalized split-\\(\\widehat{R}\\) measure computed on the \\(\\zeta^{(mn)}\\) values is called -\\(\\widehat{R}\\). This measures convergence in the tails rather than in the bulk of the distribution.\nTo obtain a single conservative \\(\\widehat{R}\\) estimate, we propose to report the maximum of rank normalized split-\\(\\widehat{R}\\) and rank normalized folded-split-\\(\\widehat{R}\\) for each parameter.\nBulk-ESS is defined as ESS for rank normalized split chains. Tail-ESS is defined as the minimum ESS for the 5% and 95% quantiles. See Effective Sample Size section for details on how ESS is estimated.\n\n\n\nA question that often arises is whether it is acceptable to monitor convergence of only a subset of the parameters or generated quantities. The short answer is “no,” but this is elaborated further in this section.\nFor example, consider the value lp__, which is the log posterior density (up to a constant).3\nIt is thus a mistake to declare convergence in any practical sense if lp__ has not converged, because different chains are really in different parts of the space. Yet measuring convergence for lp__ is particularly tricky, as noted below.\n\n\nMarkov chain convergence is a global property in the sense that it does not depend on the choice of function of the parameters that is monitored. There is no hard cutoff between pre-convergence “transience” and post-convergence “equilibrium.” What happens is that as the number of states in the chain approaches infinity, the distribution of possible states in the chain approaches the target distribution and in that limit the expected value of the Monte Carlo estimator of any integrable function converges to the true expectation. There is nothing like warmup here, because in the limit, the effects of initial state are completely washed out.\n\n\n\nThe \\(\\hat{R}\\) statistic considers the composition of a Markov chain and a function, and if the Markov chain has converged then each Markov chain and function composition will have converged. Multivariate functions converge when all of their margins have converged by the Cramer-Wold theorem.\nThe transformation from unconstrained space to constrained space is just another function, so does not effect convergence.\nDifferent functions may have different autocorrelations, but if the Markov chain has equilibrated then all Markov chain plus function compositions should be consistent with convergence. Formally, any function that appears inconsistent is of concern and although it would be unreasonable to test every function, lp__ and other measured quantities should at least be consistent.\nThe obvious difference in lp__ is that it tends to vary quickly with position and is consequently susceptible to outliers.\n\n\n\nThe question is what happens for finite numbers of states? If we can prove a strong geometric ergodicity property (which depends on the sampler and the target distribution), then one can show that there exists a finite time after which the chain forgets its initial state with a large probability. This is both the autocorrelation time and the warmup time. But even if you can show it exists and is finite (which is nigh impossible) you can’t compute an actual value analytically.\nSo what we do in practice is hope that the finite number of draws is large enough for the expectations to be reasonably accurate. Removing warmup iterations improves the accuracy of the expectations but there is no guarantee that removing any finite number of draws will be enough.\n\n\n\nFirstly, as noted above, for any finite number of draws, there will always be some residual effect of the initial state, which typically manifests as some small (or large if the autocorrelation time is huge) probability of having a large outlier. Functions robust to such outliers (say, quantiles) will appear more stable and have better \\(\\hat{R}\\). Functions vulnerable to such outliers may show fragility.\nSecondly, use of the \\(\\hat{R}\\) statistic makes very strong assumptions. In particular, it assumes that the functions being considered are Gaussian or it only uses the first two moments and assumes some kind of independence. The point is that strong assumptions are made that do not always hold. In particular, the distribution for the log posterior density (lp__) almost never looks Gaussian, instead it features long tails that can lead to large \\(\\hat{R}\\) even in the large \\(N\\) limit. Tweaks to \\(\\hat{R}\\), such as using quantiles in place of raw values, have the flavor of making the sample of interest more Gaussian and hence the \\(\\hat{R}\\) statistic more accurate.\n\n\n\n“Convergence” is a global property and holds for all integrable functions at once, but employing the \\(\\hat{R}\\) statistic requires additional assumptions and thus may not work for all functions equally well.\nNote that if you just compare the expectations between chains then we can rely on the Markov chain asymptotics for Gaussian distributions and can apply the standard tests.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Algorithms",
+      "Posterior Analysis"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#stan-license",
-    "href": "reference-manual/licenses.html#stan-license",
-    "title": "Licenses and Dependencies",
+    "objectID": "reference-manual/analysis.html#effective-sample-size.section",
+    "href": "reference-manual/analysis.html#effective-sample-size.section",
+    "title": "Posterior Analysis",
     "section": "",
-    "text": "Stan is distributed under\n\nBSD 3-clause license (BSD New)\n\nThe copyright holder of each contribution is the developer or his or her assignee.1",
+    "text": "The second technical difficulty posed by MCMC methods is that the draws will typically be autocorrelated (or anticorrelated) within a chain. This increases (or reduces) the uncertainty of the estimation of posterior quantities of interest, such as means, variances, or quantiles; see Charles J. Geyer (2011).\nStan estimates an effective sample size for each parameter, which plays the role in the Markov chain Monte Carlo central limit theorem (MCMC CLT) as the number of independent draws plays in the standard central limit theorem (CLT).\nUnlike most packages, the particular calculations used by Stan follow those for split-\\(\\hat{R}\\), which involve both cross-chain (mean) and within-chain calculations (autocorrelation); see Gelman et al. (2013) and Vehtari et al. (2021).\n\n\nThe amount by which autocorrelation within the chains increases uncertainty in estimates can be measured by effective sample size (ESS). Given independent sample (with finite variance), the central limit theorem bounds uncertainty in estimates based on the sample size \\(N\\). Given dependent sample, the sample size is replaced with the effective sample size \\(N_{\\mathrm{eff}}\\).\nFor example, Monte Carlo standard error (MCSE) is proportional to \\(1 / \\sqrt{N_{\\mathrm{eff}}}\\) rather than \\(1/\\sqrt{N}\\).\nThe effective sample size of a sequence is defined in terms of the autocorrelations within the sequence at different lags. The autocorrelation \\(\\rho_t\\) at lag \\(t \\geq 0\\) for a chain with joint probability function \\(p(\\theta)\\) with mean \\(\\mu\\) and variance \\(\\sigma^2\\) is defined to be\n\\[\n\\rho_t\n=\n\\frac{1}{\\sigma^2} \\, \\int_{\\Theta} (\\theta^{(n)} - \\mu)\n(\\theta^{(n+t)} - \\mu) \\, p(\\theta) \\, d\\theta.\n\\]\nThis is the correlation between the two chains offset by \\(t\\) positions (i.e., a lag in time-series terminology). Because we know \\(\\theta^{(n)}\\) and \\(\\theta^{(n+t)}\\) have the same marginal distribution in an MCMC setting, multiplying the two difference terms and reducing yields\n\\[\n\\rho_t\n= \\frac{1}{\\sigma^2}\n\\, \\int_{\\Theta}\n      \\theta^{(n)} \\, \\theta^{(n+t)} \\, p(\\theta)\n   \\, d\\theta - \\frac{\\mu^2}{\\sigma^2}.\n\\]\nThe effective sample size of \\(N\\) draws generated by a process with autocorrelations \\(\\rho_t\\) is defined by \\[\nN_{\\mathrm{eff}}\n\\ = \\\n\\frac{N}{\\sum_{t = -\\infty}^{\\infty} \\rho_t}\n\\ = \\\n\\frac{N}{1 + 2 \\sum_{t = 1}^{\\infty} \\rho_t}.\n\\]\nFor independent draws, the effective sample size is just the number of iterations. For correlated draws, the effective sample size is usually lower than the number of iterations, but in case of anticorrelated draws, the effective sample size can be larger than the number of iterations. In this latter case, MCMC can work better than independent sampling for some estimation problems. Hamiltonian Monte Carlo, including the no-U-turn sampler used by default in Stan, can produce anticorrelated draws if the posterior is close to Gaussian with little posterior correlation.\n\n\n\nIn practice, the probability function in question cannot be tractably integrated and thus the autocorrelation cannot be calculated, nor the effective sample size. Instead, these quantities must be estimated from the draws themselves. The rest of this section describes a autocorrelations and split-\\(\\hat{R}\\) based effective sample size estimator, based on multiple chains. As before, each chain \\(\\theta_m\\) will be assumed to be of length \\(N\\).\nStan carries out the autocorrelation computations for all lags simultaneously using Eigen’s fast Fourier transform (FFT) package with appropriate padding; see Charles J. Geyer (2011) for more detail on using FFT for autocorrelation calculations. The autocorrelation estimates \\(\\hat{\\rho}_{t,m}\\) at lag \\(t\\) from multiple chains \\(m \\in\n(1,\\ldots,M)\\) are combined with within-sample variance estimate \\(W\\) and multi-chain variance estimate \\(\\widehat{\\mbox{var}}^{+}\\) introduced in the previous section to compute the combined autocorrelation at lag \\(t\\) as\n\\[\n\\hat{\\rho}_t\n= 1 - \\frac{\\displaystyle W\n              - \\textstyle \\frac{1}{M}\\sum_{m=1}^M s_m^2 \\hat{\\rho}_{t,m}}\n        {\\widehat{\\mbox{var}}^{+}}.\n\\]\nIf the chains have not converged, the variance estimator \\(\\widehat{\\mbox{var}}^{+}\\) will overestimate variance, leading to an overestimate of autocorrelation and an underestimate effective sample size.\nBecause of the noise in the correlation estimates \\(\\hat{\\rho}_t\\) as \\(t\\) increases, a typical truncated sum of \\(\\hat{\\rho}_t\\) is used. Negative autocorrelations may occur only on odd lags and by summing over pairs starting from lag 0, the paired autocorrelation is guaranteed to be positive, monotone and convex modulo estimator noise Charles J. Geyer (1992), Charles J. Geyer (2011). Stan uses Geyer’s initial monotone sequence criterion. The effective sample size estimator is defined as\n\\[\n\\hat{N}_{\\mathrm{eff}} = \\frac{M \\cdot N}{\\hat{\\tau}},\n\\]\nwhere\n\\[\n\\hat{\\tau} = 1 + 2 \\sum_{t=1}^{2m+1} \\hat{\\rho}_t = -1 + 2 \\sum_{t'=0}^{m} \\hat{P}_{t'},\n\\]\nwhere \\(\\hat{P}_{t'}=\\hat{\\rho}_{2t'}+\\hat{\\rho}_{2t'+1}\\). Initial positive sequence estimators is obtained by choosing the largest \\(m\\) such that \\(\\hat{P}_{t'}&gt;0, \\quad t' = 1,\\ldots,m\\). The initial monotone sequence is obtained by further reducing \\(\\hat{P}_{t'}\\) to the minimum of the preceding ones so that the estimated sequence is monotone.\n\n\n\nThe posterior standard deviation of a parameter \\(\\theta_n\\) conditioned on observed data \\(y\\) is just the standard deviation of the posterior density \\(p(\\theta_n | y)\\). This is estimated by the standard deviation of the combined posterior draws across chains,\n\\[\n\\hat{\\sigma}_n = \\mathrm{sd}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n).\n\\]\nThe previous section showed how to estimate \\(N_{\\mathrm{eff}}\\) for a parameter \\(\\theta_n\\) based on multiple chains of posterior draws.\nThe mean of the posterior draws of \\(\\theta_n\\) \\[\n\\hat{\\theta}_n\n= \\mathrm{mean}(\\theta^{(1)}_n, \\ldots, \\theta^{(m)}_n)\n\\]\nis treated as an estimator of the true posterior mean,\n\\[\n\\mathbb{E}[\\theta_n \\mid y]\n\\ = \\\n\\int_{-\\infty}^{\\infty}\n    \\, \\theta \\, p(\\theta | y)\n\\, \\mathrm{d}\\theta_n,\n\\]\nbased the observed data \\(y\\).\nThe standard error for the estimator \\(\\hat{\\theta}_n\\) is given by the posterior standard deviation divided by the square root of the effective sample size. This standard error is itself estimated as \\(\\hat{\\sigma}_n / \\sqrt{N_{\\mathrm{eff}}}\\). The smaller the standard error, the closer the estimate \\(\\hat{\\theta}_n\\) is expected to be to the true value. This is just the MCMC CLT applied to an estimator; see Charles J. Geyer (2011) for more details of the MCMC central limit theorem.\n\n\n\nIn complex posteriors, draws are almost always positively correlated. In these situations, the autocorrelation at lag \\(t\\), \\(\\rho_t\\), decreases as the lag, \\(t\\), increases. In this situation, thinning the sample by keeping only every \\(N\\)-th draw will reduce the autocorrelation of the resulting chain. This is particularly useful if we need to save storage or re-use the draws for inference.\nFor instance, consider generating one thousand posterior draws in one of the following two ways.\n\nGenerate 1000 draws after convergence and save all of them.\nGenerate 10,000 draws after convergence and save every tenth draw.\n\nEven though both produce a sample consisting one thousand draws, the second approach with thinning can produce a higher effective sample size when the draws are positively correlated. That’s because the autocorrelation \\(\\rho_t\\) for the thinned sequence is equivalent to \\(\\rho_{10t}\\) in the unthinned sequence, so the sum of the autocorrelations usually will be lower and thus the effective sample size higher.\nNow contrast the second approach above with the unthinned alternative,\n\nGenerate 10,000 draws after convergence and save every draw.\n\nThis will typically have a higher effective sample than the thinned sample consisting of every tenth drawn. But the gap might not be very large. To summarize, the only reason to thin a sample is to reduce memory requirements.\nIf draws are anticorrelated, then thinning will increase correlation and further reduce the overall effective sample size.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Algorithms",
+      "Posterior Analysis"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#boost-license",
-    "href": "reference-manual/licenses.html#boost-license",
-    "title": "Licenses and Dependencies",
-    "section": "",
-    "text": "Stan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the\n\nBoost Software License version 1.0\n\nThe copyright for each Boost package is held by its developers or their assignees.",
+    "objectID": "reference-manual/analysis.html#footnotes",
+    "href": "reference-manual/analysis.html#footnotes",
+    "title": "Posterior Analysis",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nUsing vectors simplifies high level exposition at the expense of collapsing structure.↩︎\nThe structure is assumed to be rectangular; in the future, this needs to be generalized to ragged samples.↩︎\nThe lp__ value also represents the potential energy in the Hamiltonian system and is rate bounded by the randomly supplied kinetic energy each iteration, which follows a Chi-square distribution in the number of parameters.↩︎",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Algorithms",
+      "Posterior Analysis"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#eigen-license",
-    "href": "reference-manual/licenses.html#eigen-license",
-    "title": "Licenses and Dependencies",
+    "objectID": "reference-manual/comments.html",
+    "href": "reference-manual/comments.html",
+    "title": "Comments",
     "section": "",
-    "text": "Stan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the\n\nMozilla Public License, version 2\n\nThe copyright of Eigen is owned jointly by its developers or their assignees.",
+    "text": "Stan supports C++-style line-based and bracketed comments. Comments may be used anywhere whitespace is allowed in a Stan program.\n\n\nAny characters on a line following two forward slashes (//) is ignored along with the slashes. These may be used, for example, to document variables,\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  array[N] real y;  // observations\n}\n\n\n\nFor bracketed comments, any text between a forward-slash and asterisk pair (/*) and an asterisk and forward-slash pair (*/) is ignored.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Language",
+      "Comments"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#sundials-license",
-    "href": "reference-manual/licenses.html#sundials-license",
-    "title": "Licenses and Dependencies",
+    "objectID": "reference-manual/comments.html#line-based-comments",
+    "href": "reference-manual/comments.html#line-based-comments",
+    "title": "Comments",
     "section": "",
-    "text": "Stan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.",
+    "text": "Any characters on a line following two forward slashes (//) is ignored along with the slashes. These may be used, for example, to document variables,\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  array[N] real y;  // observations\n}",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Language",
+      "Comments"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#threaded-building-blocks-tbb-license",
-    "href": "reference-manual/licenses.html#threaded-building-blocks-tbb-license",
-    "title": "Licenses and Dependencies",
+    "objectID": "reference-manual/comments.html#bracketed-comments",
+    "href": "reference-manual/comments.html#bracketed-comments",
+    "title": "Comments",
     "section": "",
-    "text": "Stan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the\n\nApache License, version 2\n\nThe copyright of TBB is owned by Intel Corporation.",
+    "text": "For bracketed comments, any text between a forward-slash and asterisk pair (/*) and an asterisk and forward-slash pair (*/) is ignored.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Language",
+      "Comments"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#google-test-license",
-    "href": "reference-manual/licenses.html#google-test-license",
-    "title": "Licenses and Dependencies",
+    "objectID": "reference-manual/diagnostics.html",
+    "href": "reference-manual/diagnostics.html",
+    "title": "Diagnostic Mode",
     "section": "",
-    "text": "Stan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of Google Test is owned by Google, Inc.",
+    "text": "Stan’s diagnostic mode runs a Stan program with data, initializing parameters either randomly or with user-specified initial values, and then evaluates the log probability and its gradients. The gradients computed by the Stan program are compared to values calculated by finite differences.\nDiagnostic mode may be configured with two parameters.\nDiagnostic Mode Configuration Table. The diagnostic model configuration parameters, constraints, and default values.\n\n\n\n\n\n\n\n\n\nparameter\ndescription\nconstraints\ndefault\n\n\n\n\nepsilon\nfinite difference size\n(0, infty)\n1e–6\n\n\nerror\nerror threshold for matching\n(0, infty)\n1e–6\n\n\n\nIf the difference between the Stan program’s gradient value and that calculated by finite difference is higher than the specified threshold, the argument will be flagged.\n\n\nDiagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.\n\n\nThe output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an \\(N \\times N\\) correlation matrix, requires \\(\\binom{N}{2}\\) unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.\n\n\n\nThe log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.\n\n\n\n\nThe general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.\n\n\n\nDue to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Algorithms",
+      "Diagnostic Mode"
     ]
   },
   {
-    "objectID": "reference-manual/licenses.html#footnotes",
-    "href": "reference-manual/licenses.html#footnotes",
-    "title": "Licenses and Dependencies",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nUniversities or companies often own the copyright of computer programs developed by their employees.↩︎",
+    "objectID": "reference-manual/diagnostics.html#diagnostic-mode-output",
+    "href": "reference-manual/diagnostics.html#diagnostic-mode-output",
+    "title": "Diagnostic Mode",
+    "section": "",
+    "text": "Diagnostic mode prints the log posterior density (up to a proportion) calculated by the Stan program for the specified initial values. For each parameter, it prints the gradient at the initial parameter values calculated by Stan’s program and by finite differences over Stan’s program for the log probability.\n\n\nThe output is for the variable values and their gradients are on the unconstrained scale, which means each variable is a vector of size corresponding to the number of unconstrained variables required to define it. For example, an \\(N \\times N\\) correlation matrix, requires \\(\\binom{N}{2}\\) unconstrained parameters. The transformations from constrained to unconstrained parameters are based on the constraints in the parameter declarations and described in the reference manual chapter on transforms.\n\n\n\nThe log density includes the Jacobian adjustment implied by the constraints declared on variables. The Jacobian adjustment for constrained parameter transforms may be turned off for optimization, but there is as of yet no way to turn it off in diagnostic mode.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Licenses and Dependencies"
+      "Algorithms",
+      "Diagnostic Mode"
     ]
   },
   {
-    "objectID": "reference-manual/optimization.html",
-    "href": "reference-manual/optimization.html",
-    "title": "Optimization",
+    "objectID": "reference-manual/diagnostics.html#configuration-options",
+    "href": "reference-manual/diagnostics.html#configuration-options",
+    "title": "Diagnostic Mode",
     "section": "",
-    "text": "Stan provides optimization algorithms which find modes of the density specified by a Stan program. Such modes may be used as parameter estimates or as the basis of approximations to a Bayesian posterior.\nStan provides three different optimizers, a Newton optimizer, and two related quasi-Newton algorithms, BFGS and L-BFGS; see Nocedal and Wright (2006) for thorough description and analysis of all of these algorithms. The L-BFGS algorithm is the default optimizer. Newton’s method is the least efficient of the three, but has the advantage of setting its own stepsize.\n\n\nAll of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\nAll of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.\nAll of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.\n\n\n\n\n\nConvergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.\n\n\nThe parameters \\(\\theta_i\\) in iteration \\(i\\) are considered to have converged with respect to tolerance tol_param if\n\\[\n|| \\theta_{i} - \\theta_{i-1} || &lt; \\mathtt{tol\\_param}.\n\\]\n\n\n\nThe (unnormalized) log density \\(\\log p(\\theta_{i}|y)\\) for the parameters \\(\\theta_i\\) in iteration \\(i\\) given data \\(y\\) is considered to have converged with respect to tolerance tol_obj if\n\\[\n\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right| &lt;\n\\mathtt{tol\\_obj}.\n\\]\nThe log density is considered to have converged to within relative tolerance tol_rel_obj if\n\\[\n\\frac{\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right|}{\\\n  \\max\\left(\\left| \\log p(\\theta_{i}|y)\\right|,\\left| \\log\n      p(\\theta_{i-1}|y)\\right|,1.0\\right)}\n&lt; \\mathtt{tol\\_rel\\_obj} * \\epsilon.\n\\]\n\n\n\nThe gradient is considered to have converged to 0 relative to a specified tolerance tol_grad if\n\\[\n|| g_{i} || &lt; \\mathtt{tol\\_grad},\n\\] where \\(\\nabla_{\\theta}\\) is the gradient operator with respect to \\(\\theta\\) and \\(g_{i} = \\nabla_{\\theta} \\log p(\\theta | y)\\) is the gradient at iteration \\(i\\) evaluated at \\(\\theta^{(i)}\\), the value on the \\(i\\)-th posterior iteration.\nThe gradient is considered to have converged to 0 relative to a specified relative tolerance tol_rel_grad if\n\\[\n\\frac{g_{i}^T \\hat{H}_{i}^{-1} g_{i} }{ \\max\\left(\\left|\\log\np(\\theta_{i}|y)\\right|,1.0\\right) }\n\\ &lt; \\\n\\mathtt{tol\\_rel\\_grad} * \\epsilon,\n\\]\nwhere \\(\\hat{H}_{i}\\) is the estimate of the Hessian at iteration \\(i\\), \\(|u|\\) is the absolute value (L1 norm) of \\(u\\), \\(||u||\\) is the vector length (L2 norm) of \\(u\\), and \\(\\epsilon \\approx 2e-16\\) is machine precision.\n\n\n\n\nThe initial step size parameter \\(\\alpha\\) for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize \\(\\alpha\\) to be the roughly equal to the \\(\\alpha\\) used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune \\(\\alpha\\) can save some real time.\n\n\n\nL-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.\n\n\n\n\n\n\nFor constrained optimization problems, for instance, with a standard deviation parameter \\(\\sigma\\) constrained so that \\(\\sigma &gt; 0\\), it can be much more efficient to declare a parameter sigma with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward \\(-\\infty\\) on the \\(\\log \\sigma\\) scale.\nWith unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector\nvector[M] sigma;\nand using the default random initialization which is \\(\\mathsf{Uniform}(-2, 2)\\) on the unconstrained scale means that there is only a \\(2^{-M}\\) chance that the initialization will be within support.\nFor any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.",
+    "text": "The general configuration options for diagnostics are the same as those for MCMC. Initial values may be specified, or they may be drawn at random. Setting the random number generator will only have an effect if a random initialization is specified.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Optimization"
+      "Diagnostic Mode"
     ]
   },
   {
-    "objectID": "reference-manual/optimization.html#general-configuration",
-    "href": "reference-manual/optimization.html#general-configuration",
-    "title": "Optimization",
+    "objectID": "reference-manual/diagnostics.html#speed-warning-and-data-trimming",
+    "href": "reference-manual/diagnostics.html#speed-warning-and-data-trimming",
+    "title": "Diagnostic Mode",
     "section": "",
-    "text": "All of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\nAll of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.\nAll of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.",
+    "text": "Due to the application of finite differences, the computation time grows linearly with the number of parameters. This can be require a very long time, especially in models with latent parameters that grow with the data size. It can be helpful to diagnose a model with smaller data sizes in such cases.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Optimization"
+      "Diagnostic Mode"
     ]
   },
   {
-    "objectID": "reference-manual/optimization.html#bfgs-and-l-bfgs-configuration",
-    "href": "reference-manual/optimization.html#bfgs-and-l-bfgs-configuration",
-    "title": "Optimization",
+    "objectID": "reference-manual/execution.html",
+    "href": "reference-manual/execution.html",
+    "title": "Program Execution",
     "section": "",
-    "text": "Convergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.\n\n\nThe parameters \\(\\theta_i\\) in iteration \\(i\\) are considered to have converged with respect to tolerance tol_param if\n\\[\n|| \\theta_{i} - \\theta_{i-1} || &lt; \\mathtt{tol\\_param}.\n\\]\n\n\n\nThe (unnormalized) log density \\(\\log p(\\theta_{i}|y)\\) for the parameters \\(\\theta_i\\) in iteration \\(i\\) given data \\(y\\) is considered to have converged with respect to tolerance tol_obj if\n\\[\n\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right| &lt;\n\\mathtt{tol\\_obj}.\n\\]\nThe log density is considered to have converged to within relative tolerance tol_rel_obj if\n\\[\n\\frac{\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right|}{\\\n  \\max\\left(\\left| \\log p(\\theta_{i}|y)\\right|,\\left| \\log\n      p(\\theta_{i-1}|y)\\right|,1.0\\right)}\n&lt; \\mathtt{tol\\_rel\\_obj} * \\epsilon.\n\\]\n\n\n\nThe gradient is considered to have converged to 0 relative to a specified tolerance tol_grad if\n\\[\n|| g_{i} || &lt; \\mathtt{tol\\_grad},\n\\] where \\(\\nabla_{\\theta}\\) is the gradient operator with respect to \\(\\theta\\) and \\(g_{i} = \\nabla_{\\theta} \\log p(\\theta | y)\\) is the gradient at iteration \\(i\\) evaluated at \\(\\theta^{(i)}\\), the value on the \\(i\\)-th posterior iteration.\nThe gradient is considered to have converged to 0 relative to a specified relative tolerance tol_rel_grad if\n\\[\n\\frac{g_{i}^T \\hat{H}_{i}^{-1} g_{i} }{ \\max\\left(\\left|\\log\np(\\theta_{i}|y)\\right|,1.0\\right) }\n\\ &lt; \\\n\\mathtt{tol\\_rel\\_grad} * \\epsilon,\n\\]\nwhere \\(\\hat{H}_{i}\\) is the estimate of the Hessian at iteration \\(i\\), \\(|u|\\) is the absolute value (L1 norm) of \\(u\\), \\(||u||\\) is the vector length (L2 norm) of \\(u\\), and \\(\\epsilon \\approx 2e-16\\) is machine precision.\n\n\n\n\nThe initial step size parameter \\(\\alpha\\) for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize \\(\\alpha\\) to be the roughly equal to the \\(\\alpha\\) used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune \\(\\alpha\\) can save some real time.\n\n\n\nL-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.",
+    "text": "This chapter provides a sketch of how a compiled Stan model is executed using sampling. Optimization shares the same data reading and initialization steps, but then does optimization rather than sampling.\nThis sketch is elaborated in the following chapters of this part, which cover variable declarations, expressions, statements, and blocks in more detail.\n\n\nThe reading and transforming data steps are the same for sampling, optimization and diagnostics.\n\n\nThe first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.1\nAll of the variables declared in the data block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.\nAfter each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable N is declared as int&lt;lower=0&gt;, after N is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.\n\n\n\nAfter data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.\nTransformed data variables are initialized with real values set to NaN and integer values set to the smallest integer (large absolute value negative number).\nAfter the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.\n\n\n\n\nInitialization is the same for all of Stan’s algorithms.\n\n\nIf there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.\nAfter being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.\n\n\nBecause of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nan initial value of 0 for theta leads to an unconstrained value of \\(-\\infty\\), whereas a value of 1 leads to an unconstrained value of \\(+\\infty\\). While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.\n\n\n\n\nIf there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval \\((-2,2)\\). The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.\nAn unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.\nFor parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to \\(\\exp(0) = 1\\) on the constrained scale. A value of -2 corresponds to \\(\\exp(-2) = .13\\) and a value of 2 corresponds to \\(\\exp(2) = 7.4\\).\nFor parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.\nSimplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is \\(1/K\\) in a \\(K\\)-simplex).\nCholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.\nThe initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the chapter on variable transforms.\n\n\n\nThe initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.\n\n\n\n\nSampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.\n\n\nFor each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.\nFor simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.\nIf the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively (Hoffman and Gelman 2014).\n\n\n\nDuring each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.\nThe first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.\nBecause the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.\nNext, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.\nThe final step in the log probability function calculation is to execute the statements defined in the model block.\nAs the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.\n\n\n\nA standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.\nIf step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).\nIf the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.\n\n\n\n\nOptimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.\n\n\n\nVariational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.\n\n\n\nModel diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.\n\n\n\nFor each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.\n\n\nBefore generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.\nAfter the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.\n\n\n\nThe final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.\nIn the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).2",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Optimization"
+      "Language",
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/optimization.html#writing-models-for-optimization",
-    "href": "reference-manual/optimization.html#writing-models-for-optimization",
-    "title": "Optimization",
+    "objectID": "reference-manual/execution.html#reading-and-transforming-data",
+    "href": "reference-manual/execution.html#reading-and-transforming-data",
+    "title": "Program Execution",
     "section": "",
-    "text": "For constrained optimization problems, for instance, with a standard deviation parameter \\(\\sigma\\) constrained so that \\(\\sigma &gt; 0\\), it can be much more efficient to declare a parameter sigma with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward \\(-\\infty\\) on the \\(\\log \\sigma\\) scale.\nWith unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector\nvector[M] sigma;\nand using the default random initialization which is \\(\\mathsf{Uniform}(-2, 2)\\) on the unconstrained scale means that there is only a \\(2^{-M}\\) chance that the initialization will be within support.\nFor any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.",
+    "text": "The reading and transforming data steps are the same for sampling, optimization and diagnostics.\n\n\nThe first step of execution is to read data into memory. Data may be read in through file (in CmdStan) or through memory (RStan and PyStan); see their respective manuals for details.1\nAll of the variables declared in the data block will be read. If a variable cannot be read, the program will halt with a message indicating which data variable is missing.\nAfter each variable is read, if it has a declared constraint, the constraint is validated. For example, if a variable N is declared as int&lt;lower=0&gt;, after N is read, it will be tested to make sure it is greater than or equal to zero. If a variable violates its declared constraint, the program will halt with a warning message indicating which variable contains an illegal value, the value that was read, and the constraint that was declared.\n\n\n\nAfter data is read into the model, the transformed data variable statements are executed in order to define the transformed data variables. As the statements execute, declared constraints on variables are not enforced.\nTransformed data variables are initialized with real values set to NaN and integer values set to the smallest integer (large absolute value negative number).\nAfter the statements are executed, all declared constraints on transformed data variables are validated. If the validation fails, execution halts and the variable’s name, value and constraints are displayed.",
     "crumbs": [
       "Reference Manual",
-      "Algorithms",
-      "Optimization"
+      "Language",
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/references.html",
-    "href": "reference-manual/references.html",
-    "title": "References",
+    "objectID": "reference-manual/execution.html#initialization",
+    "href": "reference-manual/execution.html#initialization",
+    "title": "Program Execution",
     "section": "",
-    "text": "References\n\n\n\n\n Back to top"
+    "text": "Initialization is the same for all of Stan’s algorithms.\n\n\nIf there are user-supplied initial values for parameters, these are read using the same input mechanism and same file format as data reads. Any constraints declared on the parameters are validated for the initial values. If a variable’s value violates its declared constraint, the program halts and a diagnostic message is printed.\nAfter being read, initial values are transformed to unconstrained values that will be used to initialize the sampler.\n\n\nBecause of the way Stan defines its transforms from the constrained to the unconstrained space, initializing parameters on the boundaries of their constraints is usually problematic. For instance, with a constraint\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nan initial value of 0 for theta leads to an unconstrained value of \\(-\\infty\\), whereas a value of 1 leads to an unconstrained value of \\(+\\infty\\). While this will be inverse transformed back correctly given the behavior of floating point arithmetic, the Jacobian will be infinite and the log probability function will fail and raise an exception.\n\n\n\n\nIf there are no user-supplied initial values, the default initialization strategy is to initialize the unconstrained parameters directly with values drawn uniformly from the interval \\((-2,2)\\). The bounds of this initialization can be changed but it is always symmetric around 0. The value of 0 is special in that it represents the median of the initialization. An unconstrained value of 0 corresponds to different parameter values depending on the constraints declared on the parameters.\nAn unconstrained real does not involve any transform, so an initial value of 0 for the unconstrained parameters is also a value of 0 for the constrained parameters.\nFor parameters that are bounded below at 0, the initial value of 0 on the unconstrained scale corresponds to \\(\\exp(0) = 1\\) on the constrained scale. A value of -2 corresponds to \\(\\exp(-2) = .13\\) and a value of 2 corresponds to \\(\\exp(2) = 7.4\\).\nFor parameters bounded above and below, the initial value of 0 on the unconstrained scale corresponds to a value at the midpoint of the constraint interval. For probability parameters, bounded below by 0 and above by 1, the transform is the inverse logit, so that an initial unconstrained value of 0 corresponds to a constrained value of 0.5, -2 corresponds to 0.12 and 2 to 0.88. Bounds other than 0 and 1 are just scaled and translated.\nSimplexes with initial values of 0 on the unconstrained basis correspond to symmetric values on the constrained values (i.e., each value is \\(1/K\\) in a \\(K\\)-simplex).\nCholesky factors for positive-definite matrices are initialized to 1 on the diagonal and 0 elsewhere; this is because the diagonal is log transformed and the below-diagonal values are unconstrained.\nThe initial values for other parameters can be determined from the transform that is applied. The transforms are all described in full detail in the chapter on variable transforms.\n\n\n\nThe initial values may all be set to 0 on the unconstrained scale. This can be helpful for diagnosis, and may also be a good starting point for sampling. Once a model is running, multiple chains with more diffuse starting points can help diagnose problems with convergence; see the user’s guide for more information on convergence monitoring.",
+    "crumbs": [
+      "Reference Manual",
+      "Language",
+      "Program Execution"
+    ]
   },
   {
-    "objectID": "reference-manual/reproducibility.html",
-    "href": "reference-manual/reproducibility.html",
-    "title": "Reproducibility",
+    "objectID": "reference-manual/execution.html#sampling",
+    "href": "reference-manual/execution.html#sampling",
+    "title": "Program Execution",
     "section": "",
-    "text": "Floating point operations on modern computers are notoriously difficult to replicate because the fundamental arithmetic operations, right down to the IEEE 754 encoding level, are not fully specified. The primary problem is that the precision of operations varies across different hardware platforms and software implementations.\nStan is designed to allow full reproducibility. However, this is only possible up to the external constraints imposed by floating point arithmetic.\nStan results will only be exactly reproducible if all of the following components are identical:\n\nStan version\nStan interface (RStan, PyStan, CmdStan) and version, plus version of interface language (R, Python, shell)\nversions of included libraries (Boost and Eigen)\noperating system version\ncomputer hardware including CPU, motherboard and memory\nC++ compiler, including version, compiler flags, and linked libraries\nsame configuration of call to Stan, including random seed, chain ID, initialization and data\n\nIt doesn’t matter if you use a stable release version of Stan or the version with a particular Git hash tag. The same goes for all of the interfaces, compilers, and so on. The point is that if any of these moving parts changes in some way, floating point results may change.\nConcretely, if you compile a single Stan program using the same CmdStan code base, but changed the optimization flag (-O3 vs. -O2 or -O0), the two programs may not return the identical stream of results. Thus it is very hard to guarantee reproducibility on externally managed hardware, like in a cluster or even a desktop managed by an IT department or with automatic updates turned on.\nIf, however, you compiled a Stan program today using one set of flags, took the computer away from the internet and didn’t allow it to update anything, then came back in a decade and recompiled the Stan program in the same way, you should get the same results.\nThe data needs to be the same down to the bit level. For example, if you are running in RStan, Rcpp handles the conversion between R’s floating point numbers and C++ doubles. If Rcpp changes the conversion process or use different types, the results are not guaranteed to be the same down to the bit level.\nThe compiler and compiler settings can also be an issue. There is a nice discussion of the issues and how to control reproducibility in Intel’s proprietary compiler by Corden and Kreitzer (2014).\n\n\nAs noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.\nHowever, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still does not guarantee exact reproducibility between it and other versions.\n\nStan 2.28 changed the default chain ID for MCMC from 0 to 1. Users who had set a seed but not a chain ID would observe completely different outputs.\nStan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.",
+    "text": "Sampling is based on simulating the Hamiltonian of a particle with a starting position equal to the current parameter values and an initial momentum (kinetic energy) generated randomly. The potential energy at work on the particle is taken to be the negative log (unnormalized) total probability function defined by the model. In the usual approach to implementing HMC, the Hamiltonian dynamics of the particle is simulated using the leapfrog integrator, which discretizes the smooth path of the particle into a number of small time steps called leapfrog steps.\n\n\nFor each leapfrog step, the negative log probability function and its gradient need to be evaluated at the position corresponding to the current parameter values (a more detailed sketch is provided in the next section). These are used to update the momentum based on the gradient and the position based on the momentum.\nFor simple models, only a few leapfrog steps with large step sizes are needed. For models with complex posterior geometries, many small leapfrog steps may be needed to accurately model the path of the parameters.\nIf the user specifies the number of leapfrog steps (i.e., chooses to use standard HMC), that number of leapfrog steps are simulated. If the user has not specified the number of leapfrog steps, the No-U-Turn sampler (NUTS) will determine the number of leapfrog steps adaptively (Hoffman and Gelman 2014).\n\n\n\nDuring each leapfrog step, the log probability function and its gradient must be calculated. This is where most of the time in the Stan algorithm is spent. This log probability function, which is used by the sampling algorithm, is defined over the unconstrained parameters.\nThe first step of the calculation requires the inverse transform of the unconstrained parameter values back to the constrained parameters in terms of which the model is defined. There is no error checking required because the inverse transform is a total function on every point in whose range satisfies the constraints.\nBecause the probability statements in the model are defined in terms of constrained parameters, the log Jacobian of the inverse transform must be added to the accumulated log probability.\nNext, the transformed parameter statements are executed. After they complete, any constraints declared for the transformed parameters are checked. If the constraints are violated, the model will halt with a diagnostic error message.\nThe final step in the log probability function calculation is to execute the statements defined in the model block.\nAs the log probability function executes, it accumulates an in-memory representation of the expression tree used to calculate the log probability. This includes all of the transformed parameter operations and all of the Jacobian adjustments. This tree is then used to evaluate the gradients by propagating partial derivatives backward along the expression graph. The gradient calculations account for the majority of the cycles consumed by a Stan program.\n\n\n\nA standard Metropolis accept/reject step is required to retain detailed balance and ensure draws are marginally distributed according to the probability function defined by the model. This Metropolis adjustment is based on comparing log probabilities, here defined by the Hamiltonian, which is the sum of the potential (negative log probability) and kinetic (squared momentum) energies. In theory, the Hamiltonian is invariant over the path of the particle and rejection should never occur. In practice, the probability of rejection is determined by the accuracy of the leapfrog approximation to the true trajectory of the parameters.\nIf step sizes are small, very few updates will be rejected, but many steps will be required to move the same distance. If step sizes are large, more updates will be rejected, but fewer steps will be required to move the same distance. Thus a balance between effort and rejection rate is required. If the user has not specified a step size, Stan will tune the step size during warmup sampling to achieve a desired rejection rate (thus balancing rejection versus number of steps).\nIf the proposal is accepted, the parameters are updated to their new values. Otherwise, the sample is the current set of parameter values.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Reproducibility"
+      "Language",
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/reproducibility.html#notable-changes-across-versions",
-    "href": "reference-manual/reproducibility.html#notable-changes-across-versions",
-    "title": "Reproducibility",
+    "objectID": "reference-manual/execution.html#optimization",
+    "href": "reference-manual/execution.html#optimization",
+    "title": "Program Execution",
     "section": "",
-    "text": "As noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.\nHowever, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still does not guarantee exact reproducibility between it and other versions.\n\nStan 2.28 changed the default chain ID for MCMC from 0 to 1. Users who had set a seed but not a chain ID would observe completely different outputs.\nStan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.",
+    "text": "Optimization runs very much like sampling in that it starts by reading the data and then initializing parameters. Unlike sampling, it produces a deterministic output which requires no further analysis other than to verify that the optimizer itself converged to a posterior mode. The output for optimization is also similar to that for sampling.",
     "crumbs": [
       "Reference Manual",
-      "Usage",
-      "Reproducibility"
+      "Language",
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/syntax.html",
-    "href": "reference-manual/syntax.html",
-    "title": "Language Syntax",
+    "objectID": "reference-manual/execution.html#variational-inference",
+    "href": "reference-manual/execution.html#variational-inference",
+    "title": "Program Execution",
     "section": "",
-    "text": "This chapter defines the basic syntax of the Stan modeling language using a Backus-Naur form (BNF) grammar plus extra-grammatical constraints on function typing and operator precedence and associativity.\n\n\n\n\nIn the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by &lt; and &gt;. A square brackets ([A]) indicates optionality of A. A postfixed Kleene star (A*) indicates zero or more occurrences of A. Parenthesis can be used to group symbols together in productions.\nFinally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library Menhir. A rule like &lt;list(x)&gt; ::= x (COMMA x)* declares a generic list rule, which can later be applied to others by the symbol &lt;list(&lt;expression&gt;)&gt;.\nThe following representation is constructed directly from the OCaml reference parser using a tool called Obelisk. The raw output is available here.\n\n\n\n\n&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]\n              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]\n              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF\n\n&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF\n\n&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE\n\n&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE\n\n&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE\n                             &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*\n                       RBRACE\n\n&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE\n                                   &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE\n                                 &lt;top_vardecl_or_statement&gt;* RBRACE\n\n\n\n&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA\n                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;\n\n&lt;return_type&gt; ::= VOID\n                | &lt;unsized_type&gt;\n\n&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;\n\n&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;\n                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;\n                 | &lt;basic_type&gt;\n                 | &lt;unsized_tuple_type&gt;\n\n&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA &lt;unsized_type&gt;\n                         (COMMA &lt;unsized_type&gt;)* RPAREN\n\n&lt;basic_type&gt; ::= INT\n               | REAL\n               | COMPLEX\n               | VECTOR\n               | ROWVECTOR\n               | MATRIX\n               | COMPLEXVECTOR\n               | COMPLEXROWVECTOR\n               | COMPLEXMATRIX\n\n&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK\n\n\n\n\n&lt;identifier&gt; ::= IDENTIFIER\n               | TRUNCATE\n               | JACOBIAN\n\n&lt;decl_identifier&gt; ::= &lt;identifier&gt;\n\n&lt;no_assign&gt; ::= UNREACHABLE\n\n&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]\n\n&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;\n                                      &lt;optional_assignment(rhs)&gt;\n\n&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;\n                           &lt;optional_assignment(rhs)&gt; SEMICOLON\n                         | &lt;higher_type(type_rule)&gt;\n                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA\n                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON\n\n&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;\n                           | &lt;tuple_type(type_rule)&gt;\n                           | type_rule\n\n&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule\n                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;\n\n&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA\n                            &lt;higher_type(type_rule)&gt; (COMMA\n                            &lt;higher_type(type_rule)&gt;)* RPAREN\n\n&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;\n                           | SEMICOLON\n\n&lt;sized_basic_type&gt; ::= INT\n                     | REAL\n                     | COMPLEX\n                     | VECTOR LBRACK &lt;expression&gt; RBRACK\n                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                       RBRACK\n\n&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]\n                 | REAL &lt;type_constraint&gt;                  | TUPLE\n\n                 | COMPLEX &lt;type_constraint&gt;\n                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;\n                   RBRACK\n                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | ORDERED LBRACK &lt;expression&gt; RBRACK\n                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK\n                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK\n                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZERO LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]\n                   RBRACK\n                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                   RBRACK\n\n&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]\n                    | LABRACK &lt;offset_mult&gt; RABRACK\n\n&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN\n            &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN\n            &lt;constr_expression&gt;\n          | LOWER ASSIGN &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt;\n\n&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN\n                  &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN\n                  &lt;constr_expression&gt;\n                | OFFSET ASSIGN &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt;\n\n&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK\n\n\n\n\n&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;\n               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;\n               | &lt;prefixOp&gt; &lt;expression&gt;\n               | &lt;expression&gt; &lt;postfixOp&gt;\n               | &lt;common_expression&gt;\n\n&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;\n                        &lt;constr_expression&gt;\n                      | &lt;prefixOp&gt; &lt;constr_expression&gt;\n                      | &lt;constr_expression&gt; &lt;postfixOp&gt;\n                      | &lt;common_expression&gt;\n\n&lt;common_expression&gt; ::= &lt;identifier&gt;\n                      | INTNUMERAL\n                      | REALNUMERAL\n                      | DOTNUMERAL\n                      | IMAGNUMERAL\n                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE\n                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK\n                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | TARGET LPAREN RPAREN\n                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;\n                        (COMMA &lt;expression&gt;)*] RPAREN\n                      | LPAREN &lt;expression&gt; COMMA &lt;expression&gt; (COMMA\n                        &lt;expression&gt;)* RPAREN\n                      | &lt;common_expression&gt; DOTNUMERAL\n                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK\n                      | LPAREN &lt;expression&gt; RPAREN\n\n&lt;prefixOp&gt; ::= BANG\n             | MINUS\n             | PLUS\n\n&lt;postfixOp&gt; ::= TRANSPOSE\n\n&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;\n            | &lt;logicalBinOp&gt;\n\n&lt;arithmeticBinOp&gt; ::= PLUS\n                    | MINUS\n                    | TIMES\n                    | DIVIDE\n                    | IDIVIDE\n                    | MODULO\n                    | LDIVIDE\n                    | ELTTIMES\n                    | ELTDIVIDE\n                    | HAT\n                    | ELTPOW\n\n&lt;logicalBinOp&gt; ::= OR\n                 | AND\n                 | EQUALS\n                 | NEQUALS\n                 | LABRACK\n                 | LEQ\n                 | RABRACK\n                 | GEQ\n\n&lt;indexes&gt; ::= epsilon\n            | COLON\n            | &lt;expression&gt;\n            | &lt;expression&gt; COLON\n            | COLON &lt;expression&gt;\n            | &lt;expression&gt; COLON &lt;expression&gt;\n            | &lt;indexes&gt; COMMA &lt;indexes&gt;\n\n&lt;printables&gt; ::= &lt;expression&gt;\n               | &lt;string_literal&gt;\n               | &lt;printables&gt; COMMA &lt;printables&gt;\n\n\n\n&lt;statement&gt; ::= &lt;atomic_statement&gt;\n              | &lt;nested_statement&gt;\n\n&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;\n                       SEMICOLON\n                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                       &lt;expression&gt;)*] RPAREN SEMICOLON\n                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;\n                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON\n                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | BREAK SEMICOLON\n                     | CONTINUE SEMICOLON\n                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | RETURN &lt;expression&gt; SEMICOLON\n                     | RETURN SEMICOLON\n                     | SEMICOLON\n\n&lt;assignment_op&gt; ::= ASSIGN\n                  | PLUSASSIGN\n                  | MINUSASSIGN\n                  | TIMESASSIGN\n                  | DIVIDEASSIGN\n                  | ELTTIMESASSIGN\n                  | ELTDIVIDEASSIGN\n\n&lt;string_literal&gt; ::= STRINGLITERAL\n\n&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK\n\n&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                       ELSE &lt;vardecl_or_statement&gt;\n                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | WHILE LPAREN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON\n                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE\n                       &lt;vardecl_or_statement&gt;* RBRACE\n                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;\n                         | &lt;var_decl&gt;\n\n&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;\n                             | &lt;top_var_decl&gt;\n\n\n\n\nMany of the tokens used in the BNF grammars follow obviously from their names: DATABLOCK is the literal string ‘data’, COMMA is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the operator precedence table.\nA few tokens are not so obvious, and are defined here in regular expressions:\nIDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*\n\nSTRINGLITERAL = \".*\"\n\nINTNUMERAL = [0-9]+ (_ [0-9]+)*\n\nEXPLITERAL = [eE] [+-]? INTNUMERAL\n\nREALNUMERAL = INTNUMERAL \\. INTNUMERAL? EXPLITERAL?\n            | \\. INTNUMERAL EXPLITERAL\n            | INTNUMERAL EXPLITERAL\n\nIMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i\n\nDOTNUMERAL = \\. INTNUMERAL\n\n\n\n\n\nA well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.\n\n\n\nIn the Stan grammar provided in this chapter, the expression 1 + 2 * 3 has two parses. As described in the operator precedence table, Stan disambiguates between the meaning \\(1\n+ (2 \\times 3)\\) and the meaning \\((1 + 2) \\times 3\\) based on operator precedences and associativities.\n\n\n\nIn a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.\n\n\n\nThe types of expressions used for elements in array expressions ('{' expressions '}') must all be of the same type or a mixture of scalar (int, real and complex) types (in which case the result is promoted to be of the highest type on the int -&gt; real -&gt; complex hierarchy).\n\n\n\nInteger literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.\n\n\n\nBoth the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.\n\n\n\nThe for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).\n\n\n\nThe arguments to a print statement cannot be void.\n\n\n\nThe break and continue statements may only be used within the body of a for-loop or while-loop.\n\n\n\nSome constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.\n\n\nFunctions ending in _rng may only be called in the transformed data and generated quantities block, and within the bodies of user-defined functions with names ending in _rng.\n\n\n\nUnnormalized distributions (with suffixes _lupmf or _lupdf) may only be called in the model block, user-defined probability functions, or within the bodies of user defined functions which end in _lp.\n\n\n\ntarget += statements can only be used inside of the model block or user-defined functions which end in _lp.\nUser defined functions which end in _lp and the target() function can only be used in the model block, transformed parameters block, and in the bodies of other user defined functions which end in _lp.\nSampling statements (using ~) can only be used in the model block or in the bodies of user-defined functions which end in _lp.\njacobian += statements can only be used inside of the transformed parameters block or in functions that end with _jacobian.\n\n\n\n\nA probability function literal must have one of the following suffixes: _lpdf, _lpmf, _lcdf, or _lccdf.\n\n\n\nStandalone expressions used as indexes must denote either an integer (int) or an integer array (array[] int). Expressions participating in range indexes (e.g., a and b in a : b) must denote integers (int).\nA second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type array[ , , ] matrix, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.",
+    "text": "Variational inference also runs similar to sampling. It begins by reading the data and initializing the algorithm. The initial variational approximation is a random draw from the standard normal distribution in the unconstrained (real-coordinate) space. Again, similar to sampling, it outputs draws from the approximate posterior once the algorithm has decided that it has converged. Thus, the tools we use for analyzing the result of Stan’s sampling routines can also be used for variational inference.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Language Syntax"
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/syntax.html#bnf-grammars",
-    "href": "reference-manual/syntax.html#bnf-grammars",
-    "title": "Language Syntax",
+    "objectID": "reference-manual/execution.html#model-diagnostics",
+    "href": "reference-manual/execution.html#model-diagnostics",
+    "title": "Program Execution",
     "section": "",
-    "text": "In the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by &lt; and &gt;. A square brackets ([A]) indicates optionality of A. A postfixed Kleene star (A*) indicates zero or more occurrences of A. Parenthesis can be used to group symbols together in productions.\nFinally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library Menhir. A rule like &lt;list(x)&gt; ::= x (COMMA x)* declares a generic list rule, which can later be applied to others by the symbol &lt;list(&lt;expression&gt;)&gt;.\nThe following representation is constructed directly from the OCaml reference parser using a tool called Obelisk. The raw output is available here.\n\n\n\n\n&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]\n              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]\n              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF\n\n&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF\n\n&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE\n\n&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE\n\n&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE\n                             &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*\n                       RBRACE\n\n&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE\n                                   &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE\n                                 &lt;top_vardecl_or_statement&gt;* RBRACE\n\n\n\n&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA\n                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;\n\n&lt;return_type&gt; ::= VOID\n                | &lt;unsized_type&gt;\n\n&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;\n\n&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;\n                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;\n                 | &lt;basic_type&gt;\n                 | &lt;unsized_tuple_type&gt;\n\n&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA &lt;unsized_type&gt;\n                         (COMMA &lt;unsized_type&gt;)* RPAREN\n\n&lt;basic_type&gt; ::= INT\n               | REAL\n               | COMPLEX\n               | VECTOR\n               | ROWVECTOR\n               | MATRIX\n               | COMPLEXVECTOR\n               | COMPLEXROWVECTOR\n               | COMPLEXMATRIX\n\n&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK\n\n\n\n\n&lt;identifier&gt; ::= IDENTIFIER\n               | TRUNCATE\n               | JACOBIAN\n\n&lt;decl_identifier&gt; ::= &lt;identifier&gt;\n\n&lt;no_assign&gt; ::= UNREACHABLE\n\n&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]\n\n&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;\n                                      &lt;optional_assignment(rhs)&gt;\n\n&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;\n                           &lt;optional_assignment(rhs)&gt; SEMICOLON\n                         | &lt;higher_type(type_rule)&gt;\n                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA\n                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON\n\n&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;\n                           | &lt;tuple_type(type_rule)&gt;\n                           | type_rule\n\n&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule\n                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;\n\n&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA\n                            &lt;higher_type(type_rule)&gt; (COMMA\n                            &lt;higher_type(type_rule)&gt;)* RPAREN\n\n&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;\n                           | SEMICOLON\n\n&lt;sized_basic_type&gt; ::= INT\n                     | REAL\n                     | COMPLEX\n                     | VECTOR LBRACK &lt;expression&gt; RBRACK\n                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                       RBRACK\n\n&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]\n                 | REAL &lt;type_constraint&gt;                  | TUPLE\n\n                 | COMPLEX &lt;type_constraint&gt;\n                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;\n                   RBRACK\n                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | ORDERED LBRACK &lt;expression&gt; RBRACK\n                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK\n                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK\n                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZERO LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]\n                   RBRACK\n                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                   RBRACK\n\n&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]\n                    | LABRACK &lt;offset_mult&gt; RABRACK\n\n&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN\n            &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN\n            &lt;constr_expression&gt;\n          | LOWER ASSIGN &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt;\n\n&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN\n                  &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN\n                  &lt;constr_expression&gt;\n                | OFFSET ASSIGN &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt;\n\n&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK\n\n\n\n\n&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;\n               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;\n               | &lt;prefixOp&gt; &lt;expression&gt;\n               | &lt;expression&gt; &lt;postfixOp&gt;\n               | &lt;common_expression&gt;\n\n&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;\n                        &lt;constr_expression&gt;\n                      | &lt;prefixOp&gt; &lt;constr_expression&gt;\n                      | &lt;constr_expression&gt; &lt;postfixOp&gt;\n                      | &lt;common_expression&gt;\n\n&lt;common_expression&gt; ::= &lt;identifier&gt;\n                      | INTNUMERAL\n                      | REALNUMERAL\n                      | DOTNUMERAL\n                      | IMAGNUMERAL\n                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE\n                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK\n                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | TARGET LPAREN RPAREN\n                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;\n                        (COMMA &lt;expression&gt;)*] RPAREN\n                      | LPAREN &lt;expression&gt; COMMA &lt;expression&gt; (COMMA\n                        &lt;expression&gt;)* RPAREN\n                      | &lt;common_expression&gt; DOTNUMERAL\n                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK\n                      | LPAREN &lt;expression&gt; RPAREN\n\n&lt;prefixOp&gt; ::= BANG\n             | MINUS\n             | PLUS\n\n&lt;postfixOp&gt; ::= TRANSPOSE\n\n&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;\n            | &lt;logicalBinOp&gt;\n\n&lt;arithmeticBinOp&gt; ::= PLUS\n                    | MINUS\n                    | TIMES\n                    | DIVIDE\n                    | IDIVIDE\n                    | MODULO\n                    | LDIVIDE\n                    | ELTTIMES\n                    | ELTDIVIDE\n                    | HAT\n                    | ELTPOW\n\n&lt;logicalBinOp&gt; ::= OR\n                 | AND\n                 | EQUALS\n                 | NEQUALS\n                 | LABRACK\n                 | LEQ\n                 | RABRACK\n                 | GEQ\n\n&lt;indexes&gt; ::= epsilon\n            | COLON\n            | &lt;expression&gt;\n            | &lt;expression&gt; COLON\n            | COLON &lt;expression&gt;\n            | &lt;expression&gt; COLON &lt;expression&gt;\n            | &lt;indexes&gt; COMMA &lt;indexes&gt;\n\n&lt;printables&gt; ::= &lt;expression&gt;\n               | &lt;string_literal&gt;\n               | &lt;printables&gt; COMMA &lt;printables&gt;\n\n\n\n&lt;statement&gt; ::= &lt;atomic_statement&gt;\n              | &lt;nested_statement&gt;\n\n&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;\n                       SEMICOLON\n                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                       &lt;expression&gt;)*] RPAREN SEMICOLON\n                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;\n                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON\n                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | BREAK SEMICOLON\n                     | CONTINUE SEMICOLON\n                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | RETURN &lt;expression&gt; SEMICOLON\n                     | RETURN SEMICOLON\n                     | SEMICOLON\n\n&lt;assignment_op&gt; ::= ASSIGN\n                  | PLUSASSIGN\n                  | MINUSASSIGN\n                  | TIMESASSIGN\n                  | DIVIDEASSIGN\n                  | ELTTIMESASSIGN\n                  | ELTDIVIDEASSIGN\n\n&lt;string_literal&gt; ::= STRINGLITERAL\n\n&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK\n\n&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                       ELSE &lt;vardecl_or_statement&gt;\n                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | WHILE LPAREN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON\n                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE\n                       &lt;vardecl_or_statement&gt;* RBRACE\n                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;\n                         | &lt;var_decl&gt;\n\n&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;\n                             | &lt;top_var_decl&gt;",
+    "text": "Model diagnostics are like sampling and optimization in that they depend on a model’s data being read and its parameters being initialized. The user’s guides for the interfaces (RStan, PyStan, CmdStan) provide more details on the diagnostics available; as of Stan 2.0, that’s just gradients on the unconstrained scale and log probabilities.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Language Syntax"
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/syntax.html#tokenizing-rules",
-    "href": "reference-manual/syntax.html#tokenizing-rules",
-    "title": "Language Syntax",
+    "objectID": "reference-manual/execution.html#output",
+    "href": "reference-manual/execution.html#output",
+    "title": "Program Execution",
     "section": "",
-    "text": "Many of the tokens used in the BNF grammars follow obviously from their names: DATABLOCK is the literal string ‘data’, COMMA is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the operator precedence table.\nA few tokens are not so obvious, and are defined here in regular expressions:\nIDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*\n\nSTRINGLITERAL = \".*\"\n\nINTNUMERAL = [0-9]+ (_ [0-9]+)*\n\nEXPLITERAL = [eE] [+-]? INTNUMERAL\n\nREALNUMERAL = INTNUMERAL \\. INTNUMERAL? EXPLITERAL?\n            | \\. INTNUMERAL EXPLITERAL\n            | INTNUMERAL EXPLITERAL\n\nIMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i\n\nDOTNUMERAL = \\. INTNUMERAL",
+    "text": "For each final draw (not counting draws during warmup or draws that are thinned), there is an output stage of writing the draw.\n\n\nBefore generating any output, the statements in the generated quantities block are executed. This can be used for any forward simulation based on parameters of the model. Or it may be used to transform parameters to an appropriate form for output.\nAfter the generated quantities statements execute, the constraints declared on generated quantities variables are validated. If these constraints are violated, the program will terminate with a diagnostic message.\n\n\n\nThe final step is to write the actual values. The values of all variables declared as parameters, transformed parameters, or generated quantities are written. Local variables are not written, nor is the data or transformed data. All values are written in their constrained forms, that is the form that is used in the model definitions.\nIn the executable form of a Stan models, parameters, transformed parameters, and generated quantities are written to a file in comma-separated value (CSV) notation with a header defining the names of the parameters (including indices for multivariate parameters).2",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Language Syntax"
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/syntax.html#extra-grammatical-constraints",
-    "href": "reference-manual/syntax.html#extra-grammatical-constraints",
-    "title": "Language Syntax",
-    "section": "",
-    "text": "A well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.\n\n\n\nIn the Stan grammar provided in this chapter, the expression 1 + 2 * 3 has two parses. As described in the operator precedence table, Stan disambiguates between the meaning \\(1\n+ (2 \\times 3)\\) and the meaning \\((1 + 2) \\times 3\\) based on operator precedences and associativities.\n\n\n\nIn a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.\n\n\n\nThe types of expressions used for elements in array expressions ('{' expressions '}') must all be of the same type or a mixture of scalar (int, real and complex) types (in which case the result is promoted to be of the highest type on the int -&gt; real -&gt; complex hierarchy).\n\n\n\nInteger literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.\n\n\n\nBoth the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.\n\n\n\nThe for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).\n\n\n\nThe arguments to a print statement cannot be void.\n\n\n\nThe break and continue statements may only be used within the body of a for-loop or while-loop.\n\n\n\nSome constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.\n\n\nFunctions ending in _rng may only be called in the transformed data and generated quantities block, and within the bodies of user-defined functions with names ending in _rng.\n\n\n\nUnnormalized distributions (with suffixes _lupmf or _lupdf) may only be called in the model block, user-defined probability functions, or within the bodies of user defined functions which end in _lp.\n\n\n\ntarget += statements can only be used inside of the model block or user-defined functions which end in _lp.\nUser defined functions which end in _lp and the target() function can only be used in the model block, transformed parameters block, and in the bodies of other user defined functions which end in _lp.\nSampling statements (using ~) can only be used in the model block or in the bodies of user-defined functions which end in _lp.\njacobian += statements can only be used inside of the transformed parameters block or in functions that end with _jacobian.\n\n\n\n\nA probability function literal must have one of the following suffixes: _lpdf, _lpmf, _lcdf, or _lccdf.\n\n\n\nStandalone expressions used as indexes must denote either an integer (int) or an integer array (array[] int). Expressions participating in range indexes (e.g., a and b in a : b) must denote integers (int).\nA second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type array[ , , ] matrix, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.",
+    "objectID": "reference-manual/execution.html#footnotes",
+    "href": "reference-manual/execution.html#footnotes",
+    "title": "Program Execution",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe C++ code underlying Stan is flexible enough to allow data to be read from memory or file. Calls from R, for instance, can be configured to read data from file or directly from R’s memory.↩︎\nIn the R version of Stan, the values may either be written to a CSV file or directly back to R’s memory.↩︎",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Language Syntax"
+      "Program Execution"
     ]
   },
   {
-    "objectID": "reference-manual/types.html",
-    "href": "reference-manual/types.html",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/includes.html",
+    "href": "reference-manual/includes.html",
+    "title": "Includes",
     "section": "",
-    "text": "This chapter covers the data types for expressions in Stan. Every variable used in a Stan program must have a declared data type. Only values of that type will be assignable to the variable (except for temporary states of transformed data and transformed parameter values). This follows the convention of programming languages like C++, not the conventions of scripting languages like Python or statistical languages such as R or BUGS.\nThe motivation for strong, static typing is threefold.\n\nStrong typing forces the programmer’s intent to be declared with the variable, making programs easier to comprehend and hence easier to debug and maintain.\nStrong typing allows programming errors relative to the declared intent to be caught sooner (at compile time) rather than later (at run time). The Stan compiler (called through an interface such as CmdStan, RStan, or PyStan) will flag any type errors and indicate the offending expressions quickly when the program is compiled.\nConstrained types will catch runtime data, initialization, and intermediate value errors as soon as they occur rather than allowing them to propagate and potentially pollute final results.\n\nStrong typing disallows assigning the same variable to objects of different types at different points in the program or in different invocations of the program.\n\n\nArguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.\nPassing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.\n\n\nStan provides two primitive data types, real for continuous values and int for integer values. These are both considered scalar types.\n\n\n\nStan provides a complex number data type complex, where a complex number contains both a real and an imaginary component, both of which are of type real. Complex types are considered scalar types.\n\n\n\nStan provides three real-valued matrix data types, vector for column vectors, row_vector for row vectors, and matrix for matrices.\nStan also provides three complex-valued matrix data types, complex_vector for column vectors, complex_row_vector for row vectors, and complex_matrix for matrices.\n\n\n\nAny type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,\narray[10] real x;\narray[6, 7] matrix[3, 3] m;\narray[12, 8, 15] complex z;\ndeclares x to be a one-dimensional array of size 10 containing real values, declares m to be a two-dimensional array of size \\(6 \\times 7\\) containing values that are \\(3 \\times 3\\) matrices, and declares z to be a \\(12 \\times 8 \\times 15\\) array of complex numbers.\nPrior to 2.26 Stan models used a different syntax which has since been removed. See the Removed Features chapter for more details.\n\n\n\nFor any sequence of types, Stan provides a tuple data type. For example,\ntuple(real, array[5] int) xi;\ndeclares xi to be a tuple holding two values, the first of which is of type type real and the second of which a 5-dimensional array of type int.\n\n\n\nDeclarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.\nAll of the basic data types other than complex may be given lower and upper bounds using syntax such as\nint&lt;lower=1&gt; N;\nreal&lt;upper=0&gt; log_p;\nvector&lt;lower=-1, upper=1&gt;[3] rho;\nThere are also special data types for structured vectors and matrices. There are five constrained vector data types, simplex for unit simplexes, unit_vector for unit-length vectors, sum_to_zero_vector for vectors that sum to zero, ordered for ordered vectors of scalars, and positive_ordered for vectors of positive ordered scalars. There are specialized matrix data types corr_matrix and cov_matrix for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type cholesky_factor_cov is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type cholesky_factor_corr is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type sum_to_zero_matrix is for matrices that sum to zero across rows and columns.\nConstraints provide error checking for variables defined in the data, transformed data, transformed parameters, and generated quantities blocks. Constraints are critical for variables declared in the parameters block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of \\(\\mathbb{R}^n\\)).\nIt is worth calling out the most important aspect of constrained data types:\n\nThe model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.\n\nIf this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.\n\n\n\n\nUnfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.\n\n\nStan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is \\(2^{31}-1\\); the minimum value is \\(-(2^{31})\\).\nWhen integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.\nInteger arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see the Stan Functions Reference integer-valued arithmetic operators section for more information).\n\n\n\nStan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly1 follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly \\(\\pm 2^{1022}\\), which is slightly larger than \\(\\pm 10^{307}\\). It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.\n64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.\nThere are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.\n\n\nThe not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).\n\n\n\nPositive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.\n\n\n\n\nStan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.\nUnlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.\n\n\n\n\nThe complex data type is a scalar, but unlike real and int types, it contains two components, a real and imaginary component, both of which are of type real. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.\n\n\nImaginary literals are written in mathematical notation using a numeral followed by the suffix i. For example, the following example constructs a complex number \\(2 - 1.3i\\) and assigns it to the variable z.\ncomplex z = 2 - 1.3i;\nreal re = get_real(z);  // re has value 2.0\nreal im = get_imag(z);  // im has value -1.3\nThe getter functions then extract the real and imaginary components of z and assign them to re and im respectively.\nThe function to_complex constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.\nvector[K] re;\nvector[K] im;\n// ...\nfor (k in 1:K) {\n  complex z = to_complex(re[k], im[k]);\n  // ...\n}\n\n\n\nExpressions of type real may be assigned to variables of type complex. For example, the following is a valid sequence of Stan statements.\nreal x = 5.0;\ncomplex z = x;  // get_real(z) == 5.0, get_imag(z) == 0\nThe real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.\nAssignability is transitive, so that expressions of type int may also be assigned to variables of type complex, as in the following example.\nint n = 2;\ncomplex z = n;\nFunction arguments also support promotion of integer or real typed expressions to type complex.\n\n\n\n\nAll variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.\n\n\nUnconstrained integers are declared using the int keyword. For example, the variable N is declared to be an integer as follows.\nint N;\n\n\n\nInteger data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare N to be a positive integer, use the following.\nint&lt;lower=1&gt; N;\nThis illustrates that the bounds are inclusive for integers.\nTo declare an integer variable cond to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.\nint&lt;lower=0, upper=1&gt; cond;\n\n\n\nUnconstrained real variables are declared using the keyword real. The following example declares theta to be an unconstrained continuous value.\nreal theta;\n\n\n\nUnconstrained complex numbers are declared using the keyword complex. The following example declares z to be an unconstrained complex variable.\ncomplex z;\n\n\n\nReal variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.\nThe variable sigma may be declared to be non-negative as follows.\nreal&lt;lower=0&gt; sigma;\nThe following declares the variable x to be less than or equal to \\(-1\\).\nreal&lt;upper=-1&gt; x;\nTo ensure rho takes on values between \\(-1\\) and \\(1\\), use the following declaration.\nreal&lt;lower=-1, upper=1&gt; rho;\n\n\nLower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants positive_infinity() and negative_infinity() which may be used for this purpose, or they may be supplied as data.\n\n\n\n\nReal variables may be declared on a space that has been transformed using an affine transformation \\(x\\mapsto \\mu + \\sigma * x\\) with offset \\(\\mu\\) and (positive) multiplier \\(\\sigma\\), using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.\nThe variable x may be declared to have offset \\(1\\) as follows.\nreal&lt;offset=1&gt; x;\nSimilarly, it can be declared to have multiplier \\(2\\) as follows.\nreal&lt;multiplier=2&gt; x;\nFinally, we can combine both declarations to declare a variable with offset \\(1\\) and multiplier \\(2\\).\nreal&lt;offset=1, multiplier=2&gt; x;\nAs an example, we can give x a normal distribution with non-centered parameterization as follows.\nparameters {\n  real&lt;offset=mu, multiplier=sigma&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nRecall that the centered parameterization is achieved with the code\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nor equivalently\nparameters {\n  real&lt;offset=0, multiplier=1&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\n\n\n\nBounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.\nFor example, it is acceptable to have the following declarations.\ndata {\n real lb;\n}\nparameters {\n   real&lt;lower=lb&gt; phi;\n}\nThis declares a real-valued parameter phi to take values greater than the value of the real-valued data variable lb. Constraints may be arbitrary expressions, but must be of type int for integer variables and of type real for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,\ndata {\n   int&lt;lower=1&gt; N;\n   array[N] real y;\n}\nparameters {\n   real&lt;lower=min(y), upper=max(y)&gt; phi;\n}\nThis declares a positive integer data variable N, an array y of real-valued data of length N, and then a parameter ranging between the minimum and maximum value of y. As shown in the example code, the functions min() and max() may be applied to containers such as arrays.\nA more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;  // enforces a &lt; b\n}\ntransformed parameters {\n  real c;\n  real&lt;lower=c&gt; d;\n  c = a;\n  d = b;\n}\nThe parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though c isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.\n\n\n\nA variable may be declared with a size that depends on a boolean constant. For example, consider the definition of alpha in the following program fragment.\ndata {\n  int&lt;lower=0, upper=1&gt; include_alpha;\n  // ...\n}\nparameters {\n  vector[include_alpha ? N : 0] alpha;\n  // ...\n}\nIf include_alpha is true, the model will include the vector alpha; if the flag is false, the model will not include alpha (technically, it will include alpha of size 0, which means it won’t contain any values and won’t be included in any output).\nThis technique is not just useful for containers. If the value of N is set to 1, then the vector alpha will contain a single element and thus alpha[1] behaves like an optional scalar, the existence of which is controlled by include_alpha.\nThis coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the RStanArm package.\n\n\n\n\nStan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A \\(3 \\times 4\\) matrix is a different kind of object in Stan than a \\(3\n\\times 4\\) array.\nThe intention of using matrix types is to call out their usage in the code. There are three situations in Stan where only vectors and matrices may be used,\n\nmatrix arithmetic operations (e.g., matrix multiplication)\nlinear algebra functions (e.g., eigenvalues and determinants), and\nmultivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).\n\nVectors and matrices cannot be typed to return integer values. They are restricted to real and complex values.\nFor constructing vectors and matrices in Stan, see Vector, Matrix, and Array Expressions.\n\n\nVectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.\n\n\n\nVectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword vector, as follows.\nvector[3] u;\nVectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.\nvector&lt;lower=0&gt;[3] u;\nSimilarly, they may be declared with a offset and/or multiplier, as in the following example\nvector&lt;offset=42, multiplier=3&gt;[3] u;\n\n\n\nLike real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword complex_vector, as follows.\ncomplex_vector[3] v;\nComplex vector declarations do not support any constraints.\n\n\n\nA unit simplex is a vector with non-negative values whose entries sum to 1. For instance, \\([0.2,0.3,0.4,0.1]^{\\top}\\) is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-simplex by\nsimplex[5] theta;\nUnit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\nIn high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.\n\n\n\nA stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a \\(3 \\times 4\\) column-stochastic matrix.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.3 \\\\\n0.3 & 0.3 & 0.6 & 0.4 \\\\\n0.5 & 0.2 & 0.3 & 0.3\n\\end{bmatrix}\n\\]\nAn example of a \\(3 \\times 4\\) row-stochastic matrix is the following.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.2 \\\\\n0.2 & 0.1 & 0.6 & 0.1 \\\\\n0.5 & 0.2 & 0.2 & 0.1\n\\end{bmatrix}\n\\]\nIn the examples above, each column (or row) sums to 1, making the matrices valid column_stochastic_matrix and row_stochastic_matrix types.\nColumn-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.\nThe column_stochastic_matrix and row_stochastic_matrix types are declared with row and column sizes. For instance, a matrix theta with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.\ncolumn_stochastic_matrix[3, 4] theta;\nA matrix theta with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.\nrow_stochastic_matrix[3, 4] theta;\nAs with simplexes, column_stochastic_matrix and row_stochastic_matrix variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold \\(\\epsilon\\).\n\n\nIn high-dimensional settings, column_stochastic_matrix and row_stochastic_matrix types may require careful tuning of the inference algorithms. To ensure stability:\n\nSmaller Step Sizes: In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.\nHigher Target Acceptance Rates: Setting higher target acceptance rates can improve the robustness of the sampling process.\nLonger Warmup Periods: Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.\nTighter Optimization Tolerances: For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.\nCustom Initialization: If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.\n\n\n\n\n\nA unit vector is a vector with a norm of one. For instance, \\([0.5,\n0.5, 0.5, 0.5]^{\\top}\\) is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-vector by\nunit_vector[5] theta;\nUnit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nA zero-sum vector is constrained such that the sum of its elements is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\(N - 1\\) degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, beta is declared to be a zero-sum \\(5\\)-vector (4 DoF) by\nsum_to_zero_vector[5] beta;\nZero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nAn ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, \\((-1.3,2.7,2.71)^{\\top}\\) is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see section).\nThe variable c is declared as an ordered 5-vector by\nordered[5] c;\nAfter their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nThere is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, \\((2,3.7,4,12.9)\\) is a positive, ordered 4-vector.\nThe variable d is declared as a positive, ordered 5-vector by\npositive_ordered[5] d;\nLike ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nRow vectors are declared with the keyword row_vector. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector u would be declared as\nrow_vector[1093] u;\nConstraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.\nrow_vector&lt;lower=-1, upper=1&gt;[10] u;\nOffset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.\nrow_vector&lt;offset=-42, multiplier=3&gt;[3] u;\nRow vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.\n\n\n\nComplex row vectors are declared with the keyword complex_row_vector and given a size in basic declarations. For example, a 12-dimensional complex row vector v would be declared as\ncomplex_row_vector[12] v;\nComplex row vectors do not allow constraints.\n\n\n\nMatrices are declared with the keyword matrix along with a number of rows and number of columns. For example,\nmatrix[3, 3] A;\nmatrix[M, N] B;\ndeclares A to be a \\(3 \\times 3\\) matrix and B to be a \\(M\n\\times N\\) matrix. For the second declaration to be well formed, the variables M and N must be declared as integers in either the data or transformed data block and before the matrix declaration.\nMatrices may also be declared with constraints, as in this (\\(3 \\times 4\\)) matrix of non-positive values.\nmatrix&lt;upper=0&gt;[3, 4] B;\nSimilarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.\nmatrix&lt;multiplier=5&gt;[3, 4] B;\n\n\nRows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.\nmatrix[M, N] a;\nrow_vector[N] b;\n// ...\na[1] = b;\nThis copies the values from row vector b to a[1], which is the first row of the matrix a. If the number of columns in a is not the same as the size of b, a run-time error is raised; the number of columns of a is N, which is also the number of columns of b.\nAssignment works by copying values in Stan. That means any subsequent assignment to a[1] does not affect b, nor does an assignment to b affect a.\n\n\n\n\nComplex matrices are declared with the keyword complex_matrix and a number of rows and columns. For example,\ncomplex_matrix[3, 3] C;\nComplex matrices do not allow constraints.\n\n\n\nMatrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,\ncov_matrix[K] Sigma;\ndeclares Sigma to be a \\(K \\times K\\) covariance matrix, where \\(K\\) is the value of the data variable K.\n\n\n\nMatrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between \\(-1\\) and \\(1\\), and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,\ncorr_matrix[3] Omega;\ndeclares Omega to be a \\(3 \\times 3\\) correlation matrix.\nCorrelation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.\nA Cholesky factor \\(L\\) is an \\(M \\times N\\) lower-triangular matrix (if \\(m &lt; n\\) then \\(L[m, n] =0\\)) with a strictly positive diagonal (\\(L[k, k]\n&gt; 0\\)) and \\(M \\geq N\\). If \\(L\\) is a Cholesky factor, then \\(\\Sigma = L\n\\, L^{\\top}\\) is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.\nThe typical case of a square Cholesky factor may be declared with a single dimension,\ncholesky_factor_cov[4] L;\n\n\nIn general, two dimensions may be declared, with the above being equal to cholesky_factor_cov[4, 4]. The type cholesky_factor_cov[M, N] may be used for the general \\(M \\times N\\) case to produce positive semi-definite matrices of rank \\(M\\).\n\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a correlation matrix.\nA Cholesky factor for a correlation matrix \\(L\\) is a \\(K \\times K\\) lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., \\(\\sum_{n=1}^K L_{m,n}^2 = 1\\)). If \\(L\\) is a Cholesky factor for a correlation matrix, then \\(L\\,L^{\\top}\\) is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).\nTo declare the variable L to be a K by K Cholesky factor of a correlation matrix, the following code may be used.\ncholesky_factor_corr[K] L;\n\n\n\nA sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\((N - 1) \\times (M - 1)\\) degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, beta is declared to be a sum-to-zero \\(5 \\times 4\\)-matrix (20 degrees of freedom) by\nsum_to_zero_matrix[5, 4] beta;\nSum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nConstrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be real&lt;lower=0, upper=1&gt; could be assigned to a variable declared as real and vice-versa. Similarly, a variable declared as matrix[3, 3] may be assigned to a variable declared as cov_matrix[3] or cholesky_factor_cov[3], and vice-versa.\nChecks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.\n\n\n\nReal-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.\nvector[N] v = ...;\ncomplex_vector[N] u = 2 * v;\nRow vectors and matrices work the same way.\n\n\n\nVariables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.\ndata {\n  int&lt;lower=0&gt; N_observed, N_missing;\n  // ...\ntransformed parameters {\n  vector[N_observed + N_missing] y;\n  // ...\n\n\n\nIf v is a column vector or row vector, then v[2] is the second element in the vector. If m is a matrix, then m[2, 3] is the value in the second row and third column.\nProviding a matrix with a single index returns the specified row. For instance, if m is a matrix, then m[2] is the second row. This allows Stan blocks such as\nmatrix[M, N] m;\nrow_vector[N] v;\nreal x;\n// ...\nv = m[2];\nx = v[3];   // x == m[2][3] == m[2, 3]\nThe type of m[2] is row_vector because it is the second row of m. Thus it is possible to write m[2][3] instead of m[2, 3] to access the third element in the second row. When given a choice, the form m[2, 3] is preferred.\nComplex versions work the same way,\ncomplex_matrix[M, N] m = ...;\ncomplex_row_vector[N] u = m[3];\ncomplex_vector[M] v = m[ , 2];\n\n\nThe form m[2, 3] is more efficient because it does not require the creation and use of an intermediate expression template for m[2]. In later versions, explicit calls to m[2][3] may be optimized to be as efficient as m[2, 3] by the Stan compiler.\n\n\n\n\nAn integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size M + N using\nvector[M + N] y;\nAny integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.\n\n\n\n\nStan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.\nA two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if a is a two-dimensional array, then a[m, n] is just a convenient shorthand for a[m][n].\nVectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.\nFor constructing arrays in Stan, see Vector, Matrix, and Array Expressions.\n\n\nArrays are declared with the keyword array followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.\nThe variable n is declared as an array of five integers as follows.\narray[5] int n;\nA two-dimensional array of complex values with three rows and four columns is declared as follows.\narray[3, 4] complex a;\nA three-dimensional array z of positive reals with five rows, four columns, and two shelves can be declared as follows.\narray[5, 4, 2] real&lt;lower=0&gt; z;\nArrays may also be declared to contain vectors. For example,\narray[3] vector[7] mu;\ndeclares mu to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example\narray[15, 12] complex_matrix[7, 2] mu;\ndeclares a 15 by 12 array of \\(7 \\times 2\\) complex matrices. Any of the constrained types may also be used in arrays, as in the declaration\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nof a \\(2 \\times 3 \\times 4\\) array of \\(5 \\times 6\\) Cholesky factors of covariance matrices.\n\n\n\nIf x is a 1-dimensional array of length 5, then x[1] is the first element in the array and x[5] is the last. For a \\(3\n\\times 4\\) array y of two dimensions, y[1, 1] is the first element and y[3, 4] the last element. For a three-dimensional array z, the first element is z[1, 1, 1], and so on.\nSubarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose y is a two-dimensional array with three rows and four columns. Then y[3] is one-dimensional array of length four. This means that y[3][1] may be used instead of y[3, 1] to access the value of the first column of the third row of y. The form y[3, 1] is the preferred form (see note in this chapter).\n\n\n\nSubarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as\narray[9, 10, 11] real w;\narray[10, 11] real x;\narray[11] real y;\nreal z;\n// ...\nx = w[5];\ny = x[4];  // y == w[5][4] == w[5, 4]\nz = y[3];  // z == w[5][4][3] == w[5, 4, 3]\nComplex-valued arrays work the same way.\n\n\n\nArrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.\narray[3, 4] vector[5] a;\narray[4] vector[5] b;\nvector[5] c;\nreal x;\nWith these declarations, the following assignments are legal.\nb = a[1];       // result is array of vectors\nc = a[1, 3];    // result is vector\nc = b[3];       //   same result as above\nx = a[1, 3, 5]; // result is scalar\nx = b[3, 5];    //   same result as above\nx = c[5];       //   same result as above\nRow vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.\nConsider the following matrix, vector and scalar declarations.\narray[3, 4] matrix[6, 5] d;\narray[4] matrix[6, 5] e;\nmatrix[6, 5] f;\nrow_vector[5] g;\nreal x;\nWith these declarations, the following definitions are legal.\ne = d[1];           // result is array of matrices\nf = d[1, 3];        // result is matrix\nf = e[3];           //   same result as above\ng = d[1, 3, 2];     // result is row vector\ng = e[3, 2];        //   same result as above\ng = f[2];           //   same result as above\nx = d[1, 3, 5, 2];  // result is scalar\nx = e[3, 5, 2];     //   same result as above\nx = f[5, 2];        //   same result as above\nx = g[2];           //   same result as above\nAs shown, the result f[2] of supplying a single index to a matrix is the indexed row, here row 2 of matrix f.\n\n\n\nSubarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.\narray[I, J, K] real x;\narray[J, K] real y;\narray[K] real z;\n// ...\nx[1] = y;\nx[1, 1] = z;\nThe sizes must match. Here, x[1] is a J by K array, as is y.\nPartial array assignment also works for arrays of matrices, vectors, and row vectors.\n\n\n\nArrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).\n\n\nFor example, vectors cannot be assigned to arrays or vice-versa.\narray[4] real a;\nvector[4] b;\nrow_vector[4] c;\n// ...\na = b; // illegal assignment of vector to array\nb = a; // illegal assignment of array to vector\na = c; // illegal assignment of row vector to array\nc = a; // illegal assignment of array to row vector\n\n\n\nIt is not even legal to assign row vectors to column vectors or vice versa.\nvector[4] b;\nrow_vector[4] c;\n// ...\nb = c; // illegal assignment of row vector to column vector\nc = b; // illegal assignment of column vector to row vector\n\n\n\nThe same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.\narray[3, 4] real a;\nmatrix[3, 4] b;\n// ...\na = b;  // illegal assignment of matrix to array\nb = a;  // illegal assignment of array to matrix\n\n\n\nA \\(1 \\times N\\) matrix cannot be assigned a row vector or vice versa.\nmatrix[1, 4] a;\nrow_vector[4] b;\n// ...\na = b;  // illegal assignment of row vector to matrix\nb = a;  // illegal assignment of matrix to row vector\nSimilarly, an \\(M \\times 1\\) matrix may not be assigned to a column vector.\nmatrix[4, 1] a;\nvector[4] b;\n// ...\na = b;  // illegal assignment of column vector to matrix\nb = a;  // illegal assignment of matrix to column vector\n\n\n\n\nAn integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.\n\n\n\nIf any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare\narray[3, 0] real a;\nthen the resulting size of a is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, a[1] will be a size-zero one-dimensional array. For comparison, declaring\narray[0, 3] real b;\nalso produces an array with an overall size of zero, but in this case, there is no way to index legally into b, because b[0] is undefined. The array will behave at run time as if it’s a \\(0 \\times\n0\\) array. For example, the result of to_matrix(b) will be a \\(0 \\times 0\\) matrix, not a \\(0 \\times 3\\) matrix.\n\n\n\n\nStan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.\n\n\nTuples are declared with the keyword tuple followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as\ntuple(int, vector[3], complex) abc;\nTuples must have at least two entries, so the following declarations are illegal.\ntuple() nil;  // ILLEGAL\ntuple(int) n;  // ILLEGAL\nTuples can be assigned as a whole if their elements can be assigned individually. For example, a can be assigned to b in the following example because int can be promoted to complex.\ntuple(int, real) a;\n...\ntuple(complex, real) b = a;\nTuple types may have elements which are declared as tuples, such as the following example.\ntuple(int, tuple(real, complex)) x;\nIn this case, it would probably be simpler to use a 3-tuple type, tuple(int, real, complex).\nTuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration\nreal&lt;lower=0&gt; sigma;\nreal&lt;lower=0, upper=1&gt; theta;\nit is legal to have a tuple with constraints such as\ntuple(real&lt;lower=0&gt;, real&lt;lower=0, upper=1&gt;) sigma_theta;\n\n\n\nTuple elements may be accessed directly. For example, with our declaration of abc from the last section, Stan uses abc.1 for the first element, abc.2 for the second, and abc.3 for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that abc.1 is of type int, abc.2 of type vector[3] and abc.3 of type complex.\n\n\n\nTuple elements can be assigned individually, allowing, e.g.,\ntuple(int, real) ab;\nab.1 = 123;\nab.2 = 12.9;\nAs with other assignments, promotions will happen if necessary (of int to real and of real to complex, along with the corresponding container type promotions).\n\n\n\nFor convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.\nGiven a tuple t of type tuple(T1, ..., Tn) and a sequence of assignable expressions of types v1, …, vn, where each vi has a type which is assignable from type Ti, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement\n(v1, /*...*/, vn) = t;\nNote that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).\nThese unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.\nFor example, if T is a tuple of type tuple(int, (real, real), complex), then the program\nint i;\nreal x, y;\ncomplex z;\n\n(i, (x, y), z) = T;\nAssigns the result of T.1 to i, the result of T.2.1 to x, the result of T.2.2 to y, and the result of T.3 to z.\nThe left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.\n\n\n\n\nThe type information associated with a variable only contains the underlying type and dimensionality of the variable.\n\n\nThe size associated with a given variable is not part of its data type. For example, declaring a variable using\narray[3] real a;\ndeclares the variable a to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.\n\n\nSizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an \\(N \\times M\\) by a \\(J \\times K\\) matrix will raise a run-time error if \\(M \\neq J\\).\n\n\n\n\nLike sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.\nSimilarly, a bounded real may be assigned to an unconstrained real and vice-versa.\n\n\nFor arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.\n\n\n\nFor data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.\nFor parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.\nFor all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to NaN and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).\nFor generated quantities, constraints are enforced after the statements in the generated quantities block have executed.\n\n\n\n\nIn order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.\nBecause size information is not part of a data type, data types will be written without size information. For instance, array[] real is the type of one-dimensional array of reals and matrix is the type of matrices. The three-dimensional integer array type is written as array[,,] int, indicating the number slots available for indexing. Similarly, array[,] vector is the type of a two-dimensional array of vectors.\n\n\n\n\nVariables in Stan are declared by giving a type and a name. For example\nint N;\nvector[N] y;\narray[5] matrix[3, 4] A;\ndeclares a variable N that is an integer, a variable y that is a vector of length N (the previously declared variable), and a variable A, which is a length-5 array where each element is a 3 by 4 matrix.\nThe size of top-level variables in the parameters, transformed parameters, and generated quantities must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.\n// illegal and will be flagged by the compiler:\ngenerated quantities {\n  int N = 10;\n  array[N] int foo;\nDepending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.\n// valid block variables, but not locals or function parameters\nvector&lt;lower=0&gt;[N] u;\n\n// valid as a block or local variable, but not a function parameter\narray[3] int is;\n\n// function parameters exclude sizes and cannot be constrained\nvoid pretty_print_tri_lower(matrix x) { ... }\nTop-level variables can have constraints and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as function parameters are not constrained types and exclude sizes.\nIn the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables M and N indicate number of columns and rows, respectively. The variable K is used for square matrices, i.e., K denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types cov_matrix[K], corr_matrix[K], cholesky_factor_cov[K], and cholesky_factor_corr[K] are only assignable to matrices of dimensions matrix[K, K] types.\n\n\n\n\n\n\n\n\nFunction Argument (unsized)\n   Local\n(unconstrained)\n              Block\n          (constrained)\n\n\n\n\nint\nint\nint\n\n\n\n\nint&lt;lower=L&gt;\n\n\n\n\nint&lt;upper=U&gt;\n\n\n\n\nint&lt;lower=L, upper=U&gt;\n\n\n\n\nint&lt;offset=O&gt;\n\n\n\n\nint&lt;multiplier=M&gt;\n\n\n\n\nint&lt;offset=O, multiplier=M&gt;\n\n\nreal\nreal\nreal\n\n\n\n\nreal&lt;lower=L&gt;\n\n\n\n\nreal&lt;upper=U&gt;\n\n\n\n\nreal&lt;lower=L, upper=U&gt;\n\n\n\n\nreal&lt;offset=O&gt;\n\n\n\n\nreal&lt;multiplier=M&gt;\n\n\n\n\nreal&lt;offset=O, multiplier=M&gt;\n\n\ncomplex\ncomplex\ncomplex\n\n\nvector\nvector[N]\nvector[N]\n\n\n\n\nvector[N]&lt;lower=L&gt;\n\n\n\n\nvector[N]&lt;upper=U&gt;\n\n\n\n\nvector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nvector[N]&lt;offset=O&gt;\n\n\n\n\nvector[N]&lt;multiplier=M&gt;\n\n\n\n\nvector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\nordered[N]\n\n\n\n\npositive_ordered[N]\n\n\n\n\nsimplex[N]\n\n\n\n\nunit_vector[N]\n\n\n\n\nsum_to_zero_vector[N]\n\n\nrow_vector\nrow_vector[N]\nrow_vector[N]\n\n\n\n\nrow_vector[N]&lt;lower=L&gt;\n\n\n\n\nrow_vector[N]&lt;upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O&gt;\n\n\n\n\nrow_vector[N]&lt;multiplier=M&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O, multiplier=M&gt;\n\n\nmatrix\nmatrix[M, N]\nmatrix[M, N]\n\n\n\n\nmatrix[M, N]&lt;lower=L&gt;\n\n\n\n\nmatrix[M, N]&lt;upper=U&gt;\n\n\n\n\nmatrix[M, N]&lt;lower=L, upper=U&gt; |\n\n\n\n\nmatrix[M, N]&lt;offset=O&gt;\n\n\n\n\nmatrix[M, N]&lt;multiplier=M&gt;\n\n\n\n\nmatrix[M, N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\ncolumn_stochastic_matrix[M, N]\n\n\n\n\nrow_stochastic_matrix[M, N]\n\n\n\n\nsum_to_zero_matrix[M, N]\n\n\n\nmatrix[K, K]\ncorr_matrix[K]\n\n\n\nmatrix[K, K]\ncov_matrix[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_corr[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_cov[K]\n\n\ncomplex_vector\ncomplex_vector[M]\ncomplex_vector[M]\n\n\ncomplex_row_vector\ncomplex_row_vector[N]\ncomplex_row_vector[N]\n\n\ncomplex_matrix\ncomplex_matrix[M, N]\ncomplex_matrix[M,N]\n\n\narray[] vector\narray[M] vector[N]\narray[M] vector[N]\n\n\n\n\narray[M] vector[N]&lt;lower=L&gt;\n\n\n\n\narray[M] vector[N]&lt;upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O&gt;\n\n\n\n\narray[M] vector[N]&lt;multiplier=M&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\narray[M] ordered[N]\n\n\n\n\narray[M] positive_ordered[N]\n\n\n\n\narray[M] simplex[N]\n\n\n\n\narray[M] unit_vector[N]\n\n\n\n\narray[M] sum_to_zero_vector[N]\n\n\n\n\nAdditional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g. array[ , ] is a 2-D array.\nFor more on how function arguments and return types are declared, consult the User’s Guide chapter on functions.\n\n\n\nStan allows assignable variables to be declared and defined in a single statement. Assignable variables are\n\nlocal variables, and\nvariables declared in the transformed data, transformed parameters, or generated quantities blocks.\n\nFor example, the statement\nint N = 5;\ndeclares the variable N to be an integer scalar type and at the same time defines it to be the value of the expression 5.\n\n\nThe type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have\nreal sum = 0;\neven though 0 is of type int and sum is of type real, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.\nVariables of any type may have values assigned to them. For example,\nmatrix[3, 2] a = b;\ndeclares a \\(3 \\times 2\\) matrix variable a and assigns a copy of the value of b to the variable a. The variable b must be of type matrix for the statement to be well formed. For the code to execute successfully, b must be the same shape as a, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to a do not affect b and subsequent changes to b do not affect a.\n\n\n\nThe right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,\nmatrix[3, 2] a = 0.5 * (b + c);\nassigns the matrix variable a to half of the sum of b and c. The only requirement on b and c is that the expression b + c be of type matrix. For example, b could be of type matrix and c of type real, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.\nSimilarly,\ncomplex z = 2 + 3i;\nassigns the the complex number \\(2 + 3i\\) to the complex scalar z. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).\n\n\n\nAny variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.\n\n\n\n\nStan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.\n\n\nThe code:\nreal x, y;\nis equivalent to\nreal x;\nreal y;\nAs a result, all declarations on the same line must be of the same type.\n\n\n\nThe ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in the section introducing compound declarations and definitions :\nreal x = 3, y = 5.6;\nConstrained data types can also be declared together, so long as the constraint for each variable is the same:\nreal&lt;lower=0&gt; x, y;",
+    "text": "Stan allows one file to be included within another file using a syntax similar to that from C++. For example, suppose the file my-std-normal.stan defines the standard normal log probability density function (up to an additive constant).\nfunctions {\n  real my_std_normal_lpdf(vector y) {\n    return -0.5 * y' * y;\n  }\n}\nSuppose we also have a file containing a Stan program with an include statement.\n#include my-std-normal.stan\nparameters {\n  real y;\n}\nmodel {\n  y ~ my_std_normal();\n}\nThis Stan program behaves as if the contents of the file my-std-normal.stan replace the line with the #include statement, behaving as if a single Stan program were provided.\nfunctions {\n  real my_std_normal_lpdf(vector y) {\n    return -0.5 * y' * y;\n  }\n}\nparameters {\n  real y;\n}\nmodel {\n  y ~ my_std_normal();\n}\nThere are no restrictions on where include statements may be placed within a file or what the contents are of the replaced file.\n\n\nIt is possible to use includes on a line non-initially. For example, the previous example could’ve included space before the # in the include line:\n    #include my-std-normal.stan\nparameters {\n// ...\nIf there is initial space before an include, it will be discarded.\n\n\n\nIt is also possible to include line-based comments after the include. For example, the previous example can be coded as:\n#include my-std-normal.stan  // definition of standard normal\nparameters {\n// ...\nLine comments are discarded when the entire line is replaced with the contents of the included file.\n\n\n\nRecursive includes will lead to a compiler error. For example, suppose a.stan contains\n#include b.stan\nand b.stan contains\n#include a.stan\nThis will result in an error explaining the circular dependency:\nSyntax error in './b.stan', line 1, column 0, included from\n'./a.stan', line 1, column 0, included from\n'./b.stan', line 1, column 0, included from\n'a.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include a.stan\n         ^\n   -------------------------------------------------\n\nFile a.stan recursively included itself.\n\n\n\nThe Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.\n\n\nIf there is not a final / or \\ in the path, a / will be appended between the path and the included file name.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Data Types and Declarations"
+      "Includes"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#overview-of-data-types",
-    "href": "reference-manual/types.html#overview-of-data-types",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/includes.html#recursive-includes",
+    "href": "reference-manual/includes.html#recursive-includes",
+    "title": "Includes",
     "section": "",
-    "text": "Arguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.\nPassing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.\n\n\nStan provides two primitive data types, real for continuous values and int for integer values. These are both considered scalar types.\n\n\n\nStan provides a complex number data type complex, where a complex number contains both a real and an imaginary component, both of which are of type real. Complex types are considered scalar types.\n\n\n\nStan provides three real-valued matrix data types, vector for column vectors, row_vector for row vectors, and matrix for matrices.\nStan also provides three complex-valued matrix data types, complex_vector for column vectors, complex_row_vector for row vectors, and complex_matrix for matrices.\n\n\n\nAny type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,\narray[10] real x;\narray[6, 7] matrix[3, 3] m;\narray[12, 8, 15] complex z;\ndeclares x to be a one-dimensional array of size 10 containing real values, declares m to be a two-dimensional array of size \\(6 \\times 7\\) containing values that are \\(3 \\times 3\\) matrices, and declares z to be a \\(12 \\times 8 \\times 15\\) array of complex numbers.\nPrior to 2.26 Stan models used a different syntax which has since been removed. See the Removed Features chapter for more details.\n\n\n\nFor any sequence of types, Stan provides a tuple data type. For example,\ntuple(real, array[5] int) xi;\ndeclares xi to be a tuple holding two values, the first of which is of type type real and the second of which a 5-dimensional array of type int.\n\n\n\nDeclarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.\nAll of the basic data types other than complex may be given lower and upper bounds using syntax such as\nint&lt;lower=1&gt; N;\nreal&lt;upper=0&gt; log_p;\nvector&lt;lower=-1, upper=1&gt;[3] rho;\nThere are also special data types for structured vectors and matrices. There are five constrained vector data types, simplex for unit simplexes, unit_vector for unit-length vectors, sum_to_zero_vector for vectors that sum to zero, ordered for ordered vectors of scalars, and positive_ordered for vectors of positive ordered scalars. There are specialized matrix data types corr_matrix and cov_matrix for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type cholesky_factor_cov is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type cholesky_factor_corr is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type sum_to_zero_matrix is for matrices that sum to zero across rows and columns.\nConstraints provide error checking for variables defined in the data, transformed data, transformed parameters, and generated quantities blocks. Constraints are critical for variables declared in the parameters block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of \\(\\mathbb{R}^n\\)).\nIt is worth calling out the most important aspect of constrained data types:\n\nThe model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.\n\nIf this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.",
+    "text": "Recursive includes will lead to a compiler error. For example, suppose a.stan contains\n#include b.stan\nand b.stan contains\n#include a.stan\nThis will result in an error explaining the circular dependency:\nSyntax error in './b.stan', line 1, column 0, included from\n'./a.stan', line 1, column 0, included from\n'./b.stan', line 1, column 0, included from\n'a.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include a.stan\n         ^\n   -------------------------------------------------\n\nFile a.stan recursively included itself.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Data Types and Declarations"
+      "Includes"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#numerical-data-types",
-    "href": "reference-manual/types.html#numerical-data-types",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/includes.html#include-paths",
+    "href": "reference-manual/includes.html#include-paths",
+    "title": "Includes",
     "section": "",
-    "text": "Unfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.\n\n\nStan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is \\(2^{31}-1\\); the minimum value is \\(-(2^{31})\\).\nWhen integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.\nInteger arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see the Stan Functions Reference integer-valued arithmetic operators section for more information).\n\n\n\nStan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly1 follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly \\(\\pm 2^{1022}\\), which is slightly larger than \\(\\pm 10^{307}\\). It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.\n64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.\nThere are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.\n\n\nThe not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).\n\n\n\nPositive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.\n\n\n\n\nStan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.\nUnlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.",
+    "text": "The Stan interfaces may provide a mechanism for specifying a sequence of system paths in which to search for include files. The file included is the first one that is found in the sequence.\n\n\nIf there is not a final / or \\ in the path, a / will be appended between the path and the included file name.",
     "crumbs": [
       "Reference Manual",
       "Language",
-      "Data Types and Declarations"
+      "Includes"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#complex-numerical-data-type",
-    "href": "reference-manual/types.html#complex-numerical-data-type",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/laplace.html",
+    "href": "reference-manual/laplace.html",
+    "title": "Laplace Approximation",
     "section": "",
-    "text": "The complex data type is a scalar, but unlike real and int types, it contains two components, a real and imaginary component, both of which are of type real. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.\n\n\nImaginary literals are written in mathematical notation using a numeral followed by the suffix i. For example, the following example constructs a complex number \\(2 - 1.3i\\) and assigns it to the variable z.\ncomplex z = 2 - 1.3i;\nreal re = get_real(z);  // re has value 2.0\nreal im = get_imag(z);  // im has value -1.3\nThe getter functions then extract the real and imaginary components of z and assign them to re and im respectively.\nThe function to_complex constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.\nvector[K] re;\nvector[K] im;\n// ...\nfor (k in 1:K) {\n  complex z = to_complex(re[k], im[k]);\n  // ...\n}\n\n\n\nExpressions of type real may be assigned to variables of type complex. For example, the following is a valid sequence of Stan statements.\nreal x = 5.0;\ncomplex z = x;  // get_real(z) == 5.0, get_imag(z) == 0\nThe real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.\nAssignability is transitive, so that expressions of type int may also be assigned to variables of type complex, as in the following example.\nint n = 2;\ncomplex z = n;\nFunction arguments also support promotion of integer or real typed expressions to type complex.",
+    "text": "Laplace Approximation\nStan provides a Laplace approximation algorithm which can be used to obtain draws from an approximated posterior. The Laplace approximation works in the unconstrained space, so that if there are constrained parameters, the normal approximation is centered at the mode in the unconstrained space and then the implemented method transforms the normal approximation sample to the constrained space before outputting them.\nGiven the estimate of the mode \\(\\widehat{\\theta}\\), the Hessian \\(H(\\widehat{\\theta})\\) is computed using central finite differences of the model functor. Next the algorithm computes the Cholesky factor of the negative inverse Hessian:\n\\(R^{-1} = \\textrm{chol}(-H(\\widehat{\\theta})) \\backslash \\mathbf{1}\\).\nEach draw is generated on the unconstrained scale by sampling\n\\(\\theta^{\\textrm{std}(m)} \\sim \\textrm{normal}(0, \\textrm{I})\\)\nand defining draw \\(m\\) to be\n\\(\\theta^{(m)} = \\widehat{\\theta} + R^{-1} \\cdot \\theta^{\\textrm{std}(m)}\\)\nFinally, each \\(\\theta^{(m)}\\) is transformed back to the constrained scale.\nThe one-time computation of the Cholesky factor incurs a high constant overhead of \\(\\mathcal{O}(N^3)\\) in \\(N\\) dimensions. It also requires \\(2N\\) gradient calculations to use as the basis, which scales at best as \\(\\mathcal{O}(N^2)\\) and is worse for models whose gradient calculation is super-linear in dimension. The algorithm also has a high per-draw overhead, requiring \\(N\\) standard normal pseudorandom numbers and \\(\\mathcal{O}(N^2)\\) per draw (to multiply by the Cholesky factor). For \\(M\\) draws, the total cost is proportional to \\(\\mathcal{O}(N^3 + M \\cdot N^2)\\).\n\n\n\n\n Back to top",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Algorithms",
+      "Laplace Approximation"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#scalar-data-types-and-variable-declarations",
-    "href": "reference-manual/types.html#scalar-data-types-and-variable-declarations",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html",
+    "href": "reference-manual/licenses.html",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "All variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.\n\n\nUnconstrained integers are declared using the int keyword. For example, the variable N is declared to be an integer as follows.\nint N;\n\n\n\nInteger data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare N to be a positive integer, use the following.\nint&lt;lower=1&gt; N;\nThis illustrates that the bounds are inclusive for integers.\nTo declare an integer variable cond to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.\nint&lt;lower=0, upper=1&gt; cond;\n\n\n\nUnconstrained real variables are declared using the keyword real. The following example declares theta to be an unconstrained continuous value.\nreal theta;\n\n\n\nUnconstrained complex numbers are declared using the keyword complex. The following example declares z to be an unconstrained complex variable.\ncomplex z;\n\n\n\nReal variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.\nThe variable sigma may be declared to be non-negative as follows.\nreal&lt;lower=0&gt; sigma;\nThe following declares the variable x to be less than or equal to \\(-1\\).\nreal&lt;upper=-1&gt; x;\nTo ensure rho takes on values between \\(-1\\) and \\(1\\), use the following declaration.\nreal&lt;lower=-1, upper=1&gt; rho;\n\n\nLower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants positive_infinity() and negative_infinity() which may be used for this purpose, or they may be supplied as data.\n\n\n\n\nReal variables may be declared on a space that has been transformed using an affine transformation \\(x\\mapsto \\mu + \\sigma * x\\) with offset \\(\\mu\\) and (positive) multiplier \\(\\sigma\\), using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.\nThe variable x may be declared to have offset \\(1\\) as follows.\nreal&lt;offset=1&gt; x;\nSimilarly, it can be declared to have multiplier \\(2\\) as follows.\nreal&lt;multiplier=2&gt; x;\nFinally, we can combine both declarations to declare a variable with offset \\(1\\) and multiplier \\(2\\).\nreal&lt;offset=1, multiplier=2&gt; x;\nAs an example, we can give x a normal distribution with non-centered parameterization as follows.\nparameters {\n  real&lt;offset=mu, multiplier=sigma&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nRecall that the centered parameterization is achieved with the code\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nor equivalently\nparameters {\n  real&lt;offset=0, multiplier=1&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\n\n\n\nBounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.\nFor example, it is acceptable to have the following declarations.\ndata {\n real lb;\n}\nparameters {\n   real&lt;lower=lb&gt; phi;\n}\nThis declares a real-valued parameter phi to take values greater than the value of the real-valued data variable lb. Constraints may be arbitrary expressions, but must be of type int for integer variables and of type real for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,\ndata {\n   int&lt;lower=1&gt; N;\n   array[N] real y;\n}\nparameters {\n   real&lt;lower=min(y), upper=max(y)&gt; phi;\n}\nThis declares a positive integer data variable N, an array y of real-valued data of length N, and then a parameter ranging between the minimum and maximum value of y. As shown in the example code, the functions min() and max() may be applied to containers such as arrays.\nA more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;  // enforces a &lt; b\n}\ntransformed parameters {\n  real c;\n  real&lt;lower=c&gt; d;\n  c = a;\n  d = b;\n}\nThe parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though c isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.\n\n\n\nA variable may be declared with a size that depends on a boolean constant. For example, consider the definition of alpha in the following program fragment.\ndata {\n  int&lt;lower=0, upper=1&gt; include_alpha;\n  // ...\n}\nparameters {\n  vector[include_alpha ? N : 0] alpha;\n  // ...\n}\nIf include_alpha is true, the model will include the vector alpha; if the flag is false, the model will not include alpha (technically, it will include alpha of size 0, which means it won’t contain any values and won’t be included in any output).\nThis technique is not just useful for containers. If the value of N is set to 1, then the vector alpha will contain a single element and thus alpha[1] behaves like an optional scalar, the existence of which is controlled by include_alpha.\nThis coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the RStanArm package.",
+    "text": "Stan and its dependent libraries, are distributed under generous, freedom-respecting licenses approved by the Open Source Initiative.\nIn particular, the licenses for Stan and its dependent libraries have no “copyleft” provisions requiring applications of Stan to be open source if they are redistributed.\nThis chapter specifies the licenses for the libraries on which Stan’s math library, language, and algorithms depend. The last tool mentioned, Google Test, is only used for testing and is not needed to run Stan.\n\n\nStan is distributed under\n\nBSD 3-clause license (BSD New)\n\nThe copyright holder of each contribution is the developer or his or her assignee.1\n\n\n\nStan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the\n\nBoost Software License version 1.0\n\nThe copyright for each Boost package is held by its developers or their assignees.\n\n\n\nStan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the\n\nMozilla Public License, version 2\n\nThe copyright of Eigen is owned jointly by its developers or their assignees.\n\n\n\nStan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.\n\n\n\nStan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the\n\nApache License, version 2\n\nThe copyright of TBB is owned by Intel Corporation.\n\n\n\nStan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of Google Test is owned by Google, Inc.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#vector-and-matrix-data-types",
-    "href": "reference-manual/types.html#vector-and-matrix-data-types",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html#stan-license",
+    "href": "reference-manual/licenses.html#stan-license",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "Stan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A \\(3 \\times 4\\) matrix is a different kind of object in Stan than a \\(3\n\\times 4\\) array.\nThe intention of using matrix types is to call out their usage in the code. There are three situations in Stan where only vectors and matrices may be used,\n\nmatrix arithmetic operations (e.g., matrix multiplication)\nlinear algebra functions (e.g., eigenvalues and determinants), and\nmultivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).\n\nVectors and matrices cannot be typed to return integer values. They are restricted to real and complex values.\nFor constructing vectors and matrices in Stan, see Vector, Matrix, and Array Expressions.\n\n\nVectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.\n\n\n\nVectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword vector, as follows.\nvector[3] u;\nVectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.\nvector&lt;lower=0&gt;[3] u;\nSimilarly, they may be declared with a offset and/or multiplier, as in the following example\nvector&lt;offset=42, multiplier=3&gt;[3] u;\n\n\n\nLike real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword complex_vector, as follows.\ncomplex_vector[3] v;\nComplex vector declarations do not support any constraints.\n\n\n\nA unit simplex is a vector with non-negative values whose entries sum to 1. For instance, \\([0.2,0.3,0.4,0.1]^{\\top}\\) is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-simplex by\nsimplex[5] theta;\nUnit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\nIn high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.\n\n\n\nA stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a \\(3 \\times 4\\) column-stochastic matrix.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.3 \\\\\n0.3 & 0.3 & 0.6 & 0.4 \\\\\n0.5 & 0.2 & 0.3 & 0.3\n\\end{bmatrix}\n\\]\nAn example of a \\(3 \\times 4\\) row-stochastic matrix is the following.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.2 \\\\\n0.2 & 0.1 & 0.6 & 0.1 \\\\\n0.5 & 0.2 & 0.2 & 0.1\n\\end{bmatrix}\n\\]\nIn the examples above, each column (or row) sums to 1, making the matrices valid column_stochastic_matrix and row_stochastic_matrix types.\nColumn-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.\nThe column_stochastic_matrix and row_stochastic_matrix types are declared with row and column sizes. For instance, a matrix theta with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.\ncolumn_stochastic_matrix[3, 4] theta;\nA matrix theta with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.\nrow_stochastic_matrix[3, 4] theta;\nAs with simplexes, column_stochastic_matrix and row_stochastic_matrix variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold \\(\\epsilon\\).\n\n\nIn high-dimensional settings, column_stochastic_matrix and row_stochastic_matrix types may require careful tuning of the inference algorithms. To ensure stability:\n\nSmaller Step Sizes: In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.\nHigher Target Acceptance Rates: Setting higher target acceptance rates can improve the robustness of the sampling process.\nLonger Warmup Periods: Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.\nTighter Optimization Tolerances: For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.\nCustom Initialization: If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.\n\n\n\n\n\nA unit vector is a vector with a norm of one. For instance, \\([0.5,\n0.5, 0.5, 0.5]^{\\top}\\) is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-vector by\nunit_vector[5] theta;\nUnit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nA zero-sum vector is constrained such that the sum of its elements is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\(N - 1\\) degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, beta is declared to be a zero-sum \\(5\\)-vector (4 DoF) by\nsum_to_zero_vector[5] beta;\nZero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nAn ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, \\((-1.3,2.7,2.71)^{\\top}\\) is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see section).\nThe variable c is declared as an ordered 5-vector by\nordered[5] c;\nAfter their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nThere is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, \\((2,3.7,4,12.9)\\) is a positive, ordered 4-vector.\nThe variable d is declared as a positive, ordered 5-vector by\npositive_ordered[5] d;\nLike ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nRow vectors are declared with the keyword row_vector. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector u would be declared as\nrow_vector[1093] u;\nConstraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.\nrow_vector&lt;lower=-1, upper=1&gt;[10] u;\nOffset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.\nrow_vector&lt;offset=-42, multiplier=3&gt;[3] u;\nRow vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.\n\n\n\nComplex row vectors are declared with the keyword complex_row_vector and given a size in basic declarations. For example, a 12-dimensional complex row vector v would be declared as\ncomplex_row_vector[12] v;\nComplex row vectors do not allow constraints.\n\n\n\nMatrices are declared with the keyword matrix along with a number of rows and number of columns. For example,\nmatrix[3, 3] A;\nmatrix[M, N] B;\ndeclares A to be a \\(3 \\times 3\\) matrix and B to be a \\(M\n\\times N\\) matrix. For the second declaration to be well formed, the variables M and N must be declared as integers in either the data or transformed data block and before the matrix declaration.\nMatrices may also be declared with constraints, as in this (\\(3 \\times 4\\)) matrix of non-positive values.\nmatrix&lt;upper=0&gt;[3, 4] B;\nSimilarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.\nmatrix&lt;multiplier=5&gt;[3, 4] B;\n\n\nRows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.\nmatrix[M, N] a;\nrow_vector[N] b;\n// ...\na[1] = b;\nThis copies the values from row vector b to a[1], which is the first row of the matrix a. If the number of columns in a is not the same as the size of b, a run-time error is raised; the number of columns of a is N, which is also the number of columns of b.\nAssignment works by copying values in Stan. That means any subsequent assignment to a[1] does not affect b, nor does an assignment to b affect a.\n\n\n\n\nComplex matrices are declared with the keyword complex_matrix and a number of rows and columns. For example,\ncomplex_matrix[3, 3] C;\nComplex matrices do not allow constraints.\n\n\n\nMatrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,\ncov_matrix[K] Sigma;\ndeclares Sigma to be a \\(K \\times K\\) covariance matrix, where \\(K\\) is the value of the data variable K.\n\n\n\nMatrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between \\(-1\\) and \\(1\\), and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,\ncorr_matrix[3] Omega;\ndeclares Omega to be a \\(3 \\times 3\\) correlation matrix.\nCorrelation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.\nA Cholesky factor \\(L\\) is an \\(M \\times N\\) lower-triangular matrix (if \\(m &lt; n\\) then \\(L[m, n] =0\\)) with a strictly positive diagonal (\\(L[k, k]\n&gt; 0\\)) and \\(M \\geq N\\). If \\(L\\) is a Cholesky factor, then \\(\\Sigma = L\n\\, L^{\\top}\\) is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.\nThe typical case of a square Cholesky factor may be declared with a single dimension,\ncholesky_factor_cov[4] L;\n\n\nIn general, two dimensions may be declared, with the above being equal to cholesky_factor_cov[4, 4]. The type cholesky_factor_cov[M, N] may be used for the general \\(M \\times N\\) case to produce positive semi-definite matrices of rank \\(M\\).\n\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a correlation matrix.\nA Cholesky factor for a correlation matrix \\(L\\) is a \\(K \\times K\\) lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., \\(\\sum_{n=1}^K L_{m,n}^2 = 1\\)). If \\(L\\) is a Cholesky factor for a correlation matrix, then \\(L\\,L^{\\top}\\) is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).\nTo declare the variable L to be a K by K Cholesky factor of a correlation matrix, the following code may be used.\ncholesky_factor_corr[K] L;\n\n\n\nA sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\((N - 1) \\times (M - 1)\\) degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, beta is declared to be a sum-to-zero \\(5 \\times 4\\)-matrix (20 degrees of freedom) by\nsum_to_zero_matrix[5, 4] beta;\nSum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nConstrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be real&lt;lower=0, upper=1&gt; could be assigned to a variable declared as real and vice-versa. Similarly, a variable declared as matrix[3, 3] may be assigned to a variable declared as cov_matrix[3] or cholesky_factor_cov[3], and vice-versa.\nChecks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.\n\n\n\nReal-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.\nvector[N] v = ...;\ncomplex_vector[N] u = 2 * v;\nRow vectors and matrices work the same way.\n\n\n\nVariables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.\ndata {\n  int&lt;lower=0&gt; N_observed, N_missing;\n  // ...\ntransformed parameters {\n  vector[N_observed + N_missing] y;\n  // ...\n\n\n\nIf v is a column vector or row vector, then v[2] is the second element in the vector. If m is a matrix, then m[2, 3] is the value in the second row and third column.\nProviding a matrix with a single index returns the specified row. For instance, if m is a matrix, then m[2] is the second row. This allows Stan blocks such as\nmatrix[M, N] m;\nrow_vector[N] v;\nreal x;\n// ...\nv = m[2];\nx = v[3];   // x == m[2][3] == m[2, 3]\nThe type of m[2] is row_vector because it is the second row of m. Thus it is possible to write m[2][3] instead of m[2, 3] to access the third element in the second row. When given a choice, the form m[2, 3] is preferred.\nComplex versions work the same way,\ncomplex_matrix[M, N] m = ...;\ncomplex_row_vector[N] u = m[3];\ncomplex_vector[M] v = m[ , 2];\n\n\nThe form m[2, 3] is more efficient because it does not require the creation and use of an intermediate expression template for m[2]. In later versions, explicit calls to m[2][3] may be optimized to be as efficient as m[2, 3] by the Stan compiler.\n\n\n\n\nAn integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size M + N using\nvector[M + N] y;\nAny integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.",
+    "text": "Stan is distributed under\n\nBSD 3-clause license (BSD New)\n\nThe copyright holder of each contribution is the developer or his or her assignee.1",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#array-data-types.section",
-    "href": "reference-manual/types.html#array-data-types.section",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html#boost-license",
+    "href": "reference-manual/licenses.html#boost-license",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "Stan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.\nA two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if a is a two-dimensional array, then a[m, n] is just a convenient shorthand for a[m][n].\nVectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.\nFor constructing arrays in Stan, see Vector, Matrix, and Array Expressions.\n\n\nArrays are declared with the keyword array followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.\nThe variable n is declared as an array of five integers as follows.\narray[5] int n;\nA two-dimensional array of complex values with three rows and four columns is declared as follows.\narray[3, 4] complex a;\nA three-dimensional array z of positive reals with five rows, four columns, and two shelves can be declared as follows.\narray[5, 4, 2] real&lt;lower=0&gt; z;\nArrays may also be declared to contain vectors. For example,\narray[3] vector[7] mu;\ndeclares mu to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example\narray[15, 12] complex_matrix[7, 2] mu;\ndeclares a 15 by 12 array of \\(7 \\times 2\\) complex matrices. Any of the constrained types may also be used in arrays, as in the declaration\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nof a \\(2 \\times 3 \\times 4\\) array of \\(5 \\times 6\\) Cholesky factors of covariance matrices.\n\n\n\nIf x is a 1-dimensional array of length 5, then x[1] is the first element in the array and x[5] is the last. For a \\(3\n\\times 4\\) array y of two dimensions, y[1, 1] is the first element and y[3, 4] the last element. For a three-dimensional array z, the first element is z[1, 1, 1], and so on.\nSubarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose y is a two-dimensional array with three rows and four columns. Then y[3] is one-dimensional array of length four. This means that y[3][1] may be used instead of y[3, 1] to access the value of the first column of the third row of y. The form y[3, 1] is the preferred form (see note in this chapter).\n\n\n\nSubarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as\narray[9, 10, 11] real w;\narray[10, 11] real x;\narray[11] real y;\nreal z;\n// ...\nx = w[5];\ny = x[4];  // y == w[5][4] == w[5, 4]\nz = y[3];  // z == w[5][4][3] == w[5, 4, 3]\nComplex-valued arrays work the same way.\n\n\n\nArrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.\narray[3, 4] vector[5] a;\narray[4] vector[5] b;\nvector[5] c;\nreal x;\nWith these declarations, the following assignments are legal.\nb = a[1];       // result is array of vectors\nc = a[1, 3];    // result is vector\nc = b[3];       //   same result as above\nx = a[1, 3, 5]; // result is scalar\nx = b[3, 5];    //   same result as above\nx = c[5];       //   same result as above\nRow vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.\nConsider the following matrix, vector and scalar declarations.\narray[3, 4] matrix[6, 5] d;\narray[4] matrix[6, 5] e;\nmatrix[6, 5] f;\nrow_vector[5] g;\nreal x;\nWith these declarations, the following definitions are legal.\ne = d[1];           // result is array of matrices\nf = d[1, 3];        // result is matrix\nf = e[3];           //   same result as above\ng = d[1, 3, 2];     // result is row vector\ng = e[3, 2];        //   same result as above\ng = f[2];           //   same result as above\nx = d[1, 3, 5, 2];  // result is scalar\nx = e[3, 5, 2];     //   same result as above\nx = f[5, 2];        //   same result as above\nx = g[2];           //   same result as above\nAs shown, the result f[2] of supplying a single index to a matrix is the indexed row, here row 2 of matrix f.\n\n\n\nSubarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.\narray[I, J, K] real x;\narray[J, K] real y;\narray[K] real z;\n// ...\nx[1] = y;\nx[1, 1] = z;\nThe sizes must match. Here, x[1] is a J by K array, as is y.\nPartial array assignment also works for arrays of matrices, vectors, and row vectors.\n\n\n\nArrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).\n\n\nFor example, vectors cannot be assigned to arrays or vice-versa.\narray[4] real a;\nvector[4] b;\nrow_vector[4] c;\n// ...\na = b; // illegal assignment of vector to array\nb = a; // illegal assignment of array to vector\na = c; // illegal assignment of row vector to array\nc = a; // illegal assignment of array to row vector\n\n\n\nIt is not even legal to assign row vectors to column vectors or vice versa.\nvector[4] b;\nrow_vector[4] c;\n// ...\nb = c; // illegal assignment of row vector to column vector\nc = b; // illegal assignment of column vector to row vector\n\n\n\nThe same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.\narray[3, 4] real a;\nmatrix[3, 4] b;\n// ...\na = b;  // illegal assignment of matrix to array\nb = a;  // illegal assignment of array to matrix\n\n\n\nA \\(1 \\times N\\) matrix cannot be assigned a row vector or vice versa.\nmatrix[1, 4] a;\nrow_vector[4] b;\n// ...\na = b;  // illegal assignment of row vector to matrix\nb = a;  // illegal assignment of matrix to row vector\nSimilarly, an \\(M \\times 1\\) matrix may not be assigned to a column vector.\nmatrix[4, 1] a;\nvector[4] b;\n// ...\na = b;  // illegal assignment of column vector to matrix\nb = a;  // illegal assignment of matrix to column vector\n\n\n\n\nAn integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.\n\n\n\nIf any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare\narray[3, 0] real a;\nthen the resulting size of a is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, a[1] will be a size-zero one-dimensional array. For comparison, declaring\narray[0, 3] real b;\nalso produces an array with an overall size of zero, but in this case, there is no way to index legally into b, because b[0] is undefined. The array will behave at run time as if it’s a \\(0 \\times\n0\\) array. For example, the result of to_matrix(b) will be a \\(0 \\times 0\\) matrix, not a \\(0 \\times 3\\) matrix.",
+    "text": "Stan uses the Boost library for template metaprograms, traits programs, the parser, and various numerical libraries for special functions, probability functions, and random number generators. Boost is distributed under the\n\nBoost Software License version 1.0\n\nThe copyright for each Boost package is held by its developers or their assignees.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#tuple-data-type",
-    "href": "reference-manual/types.html#tuple-data-type",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html#eigen-license",
+    "href": "reference-manual/licenses.html#eigen-license",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "Stan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.\n\n\nTuples are declared with the keyword tuple followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as\ntuple(int, vector[3], complex) abc;\nTuples must have at least two entries, so the following declarations are illegal.\ntuple() nil;  // ILLEGAL\ntuple(int) n;  // ILLEGAL\nTuples can be assigned as a whole if their elements can be assigned individually. For example, a can be assigned to b in the following example because int can be promoted to complex.\ntuple(int, real) a;\n...\ntuple(complex, real) b = a;\nTuple types may have elements which are declared as tuples, such as the following example.\ntuple(int, tuple(real, complex)) x;\nIn this case, it would probably be simpler to use a 3-tuple type, tuple(int, real, complex).\nTuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration\nreal&lt;lower=0&gt; sigma;\nreal&lt;lower=0, upper=1&gt; theta;\nit is legal to have a tuple with constraints such as\ntuple(real&lt;lower=0&gt;, real&lt;lower=0, upper=1&gt;) sigma_theta;\n\n\n\nTuple elements may be accessed directly. For example, with our declaration of abc from the last section, Stan uses abc.1 for the first element, abc.2 for the second, and abc.3 for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that abc.1 is of type int, abc.2 of type vector[3] and abc.3 of type complex.\n\n\n\nTuple elements can be assigned individually, allowing, e.g.,\ntuple(int, real) ab;\nab.1 = 123;\nab.2 = 12.9;\nAs with other assignments, promotions will happen if necessary (of int to real and of real to complex, along with the corresponding container type promotions).\n\n\n\nFor convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.\nGiven a tuple t of type tuple(T1, ..., Tn) and a sequence of assignable expressions of types v1, …, vn, where each vi has a type which is assignable from type Ti, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement\n(v1, /*...*/, vn) = t;\nNote that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).\nThese unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.\nFor example, if T is a tuple of type tuple(int, (real, real), complex), then the program\nint i;\nreal x, y;\ncomplex z;\n\n(i, (x, y), z) = T;\nAssigns the result of T.1 to i, the result of T.2.1 to x, the result of T.2.2 to y, and the result of T.3 to z.\nThe left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.",
+    "text": "Stan uses the Eigen library for matrix arithmetic and linear algebra. Eigen is distributed under the\n\nMozilla Public License, version 2\n\nThe copyright of Eigen is owned jointly by its developers or their assignees.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#variable-types-vs.-constraints-and-sizes",
-    "href": "reference-manual/types.html#variable-types-vs.-constraints-and-sizes",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html#sundials-license",
+    "href": "reference-manual/licenses.html#sundials-license",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "The type information associated with a variable only contains the underlying type and dimensionality of the variable.\n\n\nThe size associated with a given variable is not part of its data type. For example, declaring a variable using\narray[3] real a;\ndeclares the variable a to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.\n\n\nSizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an \\(N \\times M\\) by a \\(J \\times K\\) matrix will raise a run-time error if \\(M \\neq J\\).\n\n\n\n\nLike sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.\nSimilarly, a bounded real may be assigned to an unconstrained real and vice-versa.\n\n\nFor arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.\n\n\n\nFor data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.\nFor parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.\nFor all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to NaN and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).\nFor generated quantities, constraints are enforced after the statements in the generated quantities block have executed.\n\n\n\n\nIn order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.\nBecause size information is not part of a data type, data types will be written without size information. For instance, array[] real is the type of one-dimensional array of reals and matrix is the type of matrices. The three-dimensional integer array type is written as array[,,] int, indicating the number slots available for indexing. Similarly, array[,] vector is the type of a two-dimensional array of vectors.",
+    "text": "Stan uses the SUNDIALS package for solving differential equations. SUNDIALS is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of SUNDIALS is owned by Lawrence Livermore National Security Lab.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#variable-declaration.section",
-    "href": "reference-manual/types.html#variable-declaration.section",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html#threaded-building-blocks-tbb-license",
+    "href": "reference-manual/licenses.html#threaded-building-blocks-tbb-license",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "Variables in Stan are declared by giving a type and a name. For example\nint N;\nvector[N] y;\narray[5] matrix[3, 4] A;\ndeclares a variable N that is an integer, a variable y that is a vector of length N (the previously declared variable), and a variable A, which is a length-5 array where each element is a 3 by 4 matrix.\nThe size of top-level variables in the parameters, transformed parameters, and generated quantities must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.\n// illegal and will be flagged by the compiler:\ngenerated quantities {\n  int N = 10;\n  array[N] int foo;\nDepending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.\n// valid block variables, but not locals or function parameters\nvector&lt;lower=0&gt;[N] u;\n\n// valid as a block or local variable, but not a function parameter\narray[3] int is;\n\n// function parameters exclude sizes and cannot be constrained\nvoid pretty_print_tri_lower(matrix x) { ... }\nTop-level variables can have constraints and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as function parameters are not constrained types and exclude sizes.\nIn the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables M and N indicate number of columns and rows, respectively. The variable K is used for square matrices, i.e., K denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types cov_matrix[K], corr_matrix[K], cholesky_factor_cov[K], and cholesky_factor_corr[K] are only assignable to matrices of dimensions matrix[K, K] types.\n\n\n\n\n\n\n\n\nFunction Argument (unsized)\n   Local\n(unconstrained)\n              Block\n          (constrained)\n\n\n\n\nint\nint\nint\n\n\n\n\nint&lt;lower=L&gt;\n\n\n\n\nint&lt;upper=U&gt;\n\n\n\n\nint&lt;lower=L, upper=U&gt;\n\n\n\n\nint&lt;offset=O&gt;\n\n\n\n\nint&lt;multiplier=M&gt;\n\n\n\n\nint&lt;offset=O, multiplier=M&gt;\n\n\nreal\nreal\nreal\n\n\n\n\nreal&lt;lower=L&gt;\n\n\n\n\nreal&lt;upper=U&gt;\n\n\n\n\nreal&lt;lower=L, upper=U&gt;\n\n\n\n\nreal&lt;offset=O&gt;\n\n\n\n\nreal&lt;multiplier=M&gt;\n\n\n\n\nreal&lt;offset=O, multiplier=M&gt;\n\n\ncomplex\ncomplex\ncomplex\n\n\nvector\nvector[N]\nvector[N]\n\n\n\n\nvector[N]&lt;lower=L&gt;\n\n\n\n\nvector[N]&lt;upper=U&gt;\n\n\n\n\nvector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nvector[N]&lt;offset=O&gt;\n\n\n\n\nvector[N]&lt;multiplier=M&gt;\n\n\n\n\nvector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\nordered[N]\n\n\n\n\npositive_ordered[N]\n\n\n\n\nsimplex[N]\n\n\n\n\nunit_vector[N]\n\n\n\n\nsum_to_zero_vector[N]\n\n\nrow_vector\nrow_vector[N]\nrow_vector[N]\n\n\n\n\nrow_vector[N]&lt;lower=L&gt;\n\n\n\n\nrow_vector[N]&lt;upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O&gt;\n\n\n\n\nrow_vector[N]&lt;multiplier=M&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O, multiplier=M&gt;\n\n\nmatrix\nmatrix[M, N]\nmatrix[M, N]\n\n\n\n\nmatrix[M, N]&lt;lower=L&gt;\n\n\n\n\nmatrix[M, N]&lt;upper=U&gt;\n\n\n\n\nmatrix[M, N]&lt;lower=L, upper=U&gt; |\n\n\n\n\nmatrix[M, N]&lt;offset=O&gt;\n\n\n\n\nmatrix[M, N]&lt;multiplier=M&gt;\n\n\n\n\nmatrix[M, N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\ncolumn_stochastic_matrix[M, N]\n\n\n\n\nrow_stochastic_matrix[M, N]\n\n\n\n\nsum_to_zero_matrix[M, N]\n\n\n\nmatrix[K, K]\ncorr_matrix[K]\n\n\n\nmatrix[K, K]\ncov_matrix[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_corr[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_cov[K]\n\n\ncomplex_vector\ncomplex_vector[M]\ncomplex_vector[M]\n\n\ncomplex_row_vector\ncomplex_row_vector[N]\ncomplex_row_vector[N]\n\n\ncomplex_matrix\ncomplex_matrix[M, N]\ncomplex_matrix[M,N]\n\n\narray[] vector\narray[M] vector[N]\narray[M] vector[N]\n\n\n\n\narray[M] vector[N]&lt;lower=L&gt;\n\n\n\n\narray[M] vector[N]&lt;upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O&gt;\n\n\n\n\narray[M] vector[N]&lt;multiplier=M&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\narray[M] ordered[N]\n\n\n\n\narray[M] positive_ordered[N]\n\n\n\n\narray[M] simplex[N]\n\n\n\n\narray[M] unit_vector[N]\n\n\n\n\narray[M] sum_to_zero_vector[N]\n\n\n\n\nAdditional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g. array[ , ] is a 2-D array.\nFor more on how function arguments and return types are declared, consult the User’s Guide chapter on functions.",
+    "text": "Stan uses the Threaded Building Blocks (TBB) library for parallel computations. TBB is distributed under the\n\nApache License, version 2\n\nThe copyright of TBB is owned by Intel Corporation.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#compound-variable-declaration-and-definition",
-    "href": "reference-manual/types.html#compound-variable-declaration-and-definition",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/licenses.html#google-test-license",
+    "href": "reference-manual/licenses.html#google-test-license",
+    "title": "Licenses and Dependencies",
     "section": "",
-    "text": "Stan allows assignable variables to be declared and defined in a single statement. Assignable variables are\n\nlocal variables, and\nvariables declared in the transformed data, transformed parameters, or generated quantities blocks.\n\nFor example, the statement\nint N = 5;\ndeclares the variable N to be an integer scalar type and at the same time defines it to be the value of the expression 5.\n\n\nThe type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have\nreal sum = 0;\neven though 0 is of type int and sum is of type real, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.\nVariables of any type may have values assigned to them. For example,\nmatrix[3, 2] a = b;\ndeclares a \\(3 \\times 2\\) matrix variable a and assigns a copy of the value of b to the variable a. The variable b must be of type matrix for the statement to be well formed. For the code to execute successfully, b must be the same shape as a, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to a do not affect b and subsequent changes to b do not affect a.\n\n\n\nThe right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,\nmatrix[3, 2] a = 0.5 * (b + c);\nassigns the matrix variable a to half of the sum of b and c. The only requirement on b and c is that the expression b + c be of type matrix. For example, b could be of type matrix and c of type real, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.\nSimilarly,\ncomplex z = 2 + 3i;\nassigns the the complex number \\(2 + 3i\\) to the complex scalar z. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).\n\n\n\nAny variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.",
+    "text": "Stan uses Google Test for unit testing; it is not required to compile or execute models. Google Test is distributed under the\n\nBSD 3-clause license (BSD New)\n\nThe copyright of Google Test is owned by Google, Inc.",
+    "crumbs": [
+      "Reference Manual",
+      "Usage",
+      "Licenses and Dependencies"
+    ]
+  },
+  {
+    "objectID": "reference-manual/licenses.html#footnotes",
+    "href": "reference-manual/licenses.html#footnotes",
+    "title": "Licenses and Dependencies",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nUniversities or companies often own the copyright of computer programs developed by their employees.↩︎",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Usage",
+      "Licenses and Dependencies"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#declaring-multiple-variables-at-once",
-    "href": "reference-manual/types.html#declaring-multiple-variables-at-once",
-    "title": "Data Types and Declarations",
+    "objectID": "reference-manual/optimization.html",
+    "href": "reference-manual/optimization.html",
+    "title": "Optimization",
     "section": "",
-    "text": "Stan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.\n\n\nThe code:\nreal x, y;\nis equivalent to\nreal x;\nreal y;\nAs a result, all declarations on the same line must be of the same type.\n\n\n\nThe ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in the section introducing compound declarations and definitions :\nreal x = 3, y = 5.6;\nConstrained data types can also be declared together, so long as the constraint for each variable is the same:\nreal&lt;lower=0&gt; x, y;",
+    "text": "Stan provides optimization algorithms which find modes of the density specified by a Stan program. Such modes may be used as parameter estimates or as the basis of approximations to a Bayesian posterior.\nStan provides three different optimizers, a Newton optimizer, and two related quasi-Newton algorithms, BFGS and L-BFGS; see Nocedal and Wright (2006) for thorough description and analysis of all of these algorithms. The L-BFGS algorithm is the default optimizer. Newton’s method is the least efficient of the three, but has the advantage of setting its own stepsize.\n\n\nAll of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\nAll of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.\nAll of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.\n\n\n\n\n\nConvergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.\n\n\nThe parameters \\(\\theta_i\\) in iteration \\(i\\) are considered to have converged with respect to tolerance tol_param if\n\\[\n|| \\theta_{i} - \\theta_{i-1} || &lt; \\mathtt{tol\\_param}.\n\\]\n\n\n\nThe (unnormalized) log density \\(\\log p(\\theta_{i}|y)\\) for the parameters \\(\\theta_i\\) in iteration \\(i\\) given data \\(y\\) is considered to have converged with respect to tolerance tol_obj if\n\\[\n\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right| &lt;\n\\mathtt{tol\\_obj}.\n\\]\nThe log density is considered to have converged to within relative tolerance tol_rel_obj if\n\\[\n\\frac{\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right|}{\\\n  \\max\\left(\\left| \\log p(\\theta_{i}|y)\\right|,\\left| \\log\n      p(\\theta_{i-1}|y)\\right|,1.0\\right)}\n&lt; \\mathtt{tol\\_rel\\_obj} * \\epsilon.\n\\]\n\n\n\nThe gradient is considered to have converged to 0 relative to a specified tolerance tol_grad if\n\\[\n|| g_{i} || &lt; \\mathtt{tol\\_grad},\n\\] where \\(\\nabla_{\\theta}\\) is the gradient operator with respect to \\(\\theta\\) and \\(g_{i} = \\nabla_{\\theta} \\log p(\\theta | y)\\) is the gradient at iteration \\(i\\) evaluated at \\(\\theta^{(i)}\\), the value on the \\(i\\)-th posterior iteration.\nThe gradient is considered to have converged to 0 relative to a specified relative tolerance tol_rel_grad if\n\\[\n\\frac{g_{i}^T \\hat{H}_{i}^{-1} g_{i} }{ \\max\\left(\\left|\\log\np(\\theta_{i}|y)\\right|,1.0\\right) }\n\\ &lt; \\\n\\mathtt{tol\\_rel\\_grad} * \\epsilon,\n\\]\nwhere \\(\\hat{H}_{i}\\) is the estimate of the Hessian at iteration \\(i\\), \\(|u|\\) is the absolute value (L1 norm) of \\(u\\), \\(||u||\\) is the vector length (L2 norm) of \\(u\\), and \\(\\epsilon \\approx 2e-16\\) is machine precision.\n\n\n\n\nThe initial step size parameter \\(\\alpha\\) for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize \\(\\alpha\\) to be the roughly equal to the \\(\\alpha\\) used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune \\(\\alpha\\) can save some real time.\n\n\n\nL-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.\n\n\n\n\n\n\nFor constrained optimization problems, for instance, with a standard deviation parameter \\(\\sigma\\) constrained so that \\(\\sigma &gt; 0\\), it can be much more efficient to declare a parameter sigma with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward \\(-\\infty\\) on the \\(\\log \\sigma\\) scale.\nWith unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector\nvector[M] sigma;\nand using the default random initialization which is \\(\\mathsf{Uniform}(-2, 2)\\) on the unconstrained scale means that there is only a \\(2^{-M}\\) chance that the initialization will be within support.\nFor any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Algorithms",
+      "Optimization"
     ]
   },
   {
-    "objectID": "reference-manual/types.html#footnotes",
-    "href": "reference-manual/types.html#footnotes",
-    "title": "Data Types and Declarations",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nStan compiles integers to int and reals to double types in C++. Precise details of rounding will depend on the compiler and hardware architecture on which the code is run.↩︎",
+    "objectID": "reference-manual/optimization.html#general-configuration",
+    "href": "reference-manual/optimization.html#general-configuration",
+    "title": "Optimization",
+    "section": "",
+    "text": "All of the optimizers have the option of including the the log absolute Jacobian determinant of inverse parameter transforms in the log probability computation. If the Jacobian adjustment is not included (the default), the optimization returns parameter values that correspond to a mode of the target in the constrained space (if such mode exists). Thus this option is useful for any optimization where we want to find the mode in the original constrained parameter space. If the Jacobian adjustment is included, the optimization returns parameter values that correspond to a mode in the unconstrained space. This is useful, for example, if we want to make a distributional approximation of the posterior at the mode (see, Laplace sampling, as then Jacobian adjustment needs to be included for correct results.\nAll of the optimizers are iterative and allow the maximum number of iterations to be specified; the default maximum number of iterations is 2000.\nAll of the optimizers are able to stream intermediate output reporting on their progress. Whether or not to save the intermediate iterations and stream progress is configurable.",
     "crumbs": [
       "Reference Manual",
-      "Language",
-      "Data Types and Declarations"
+      "Algorithms",
+      "Optimization"
     ]
   },
   {
-    "objectID": "reference-manual/variational.html",
-    "href": "reference-manual/variational.html",
-    "title": "Variational Inference",
+    "objectID": "reference-manual/optimization.html#bfgs-and-l-bfgs-configuration",
+    "href": "reference-manual/optimization.html#bfgs-and-l-bfgs-configuration",
+    "title": "Optimization",
     "section": "",
-    "text": "Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) Kucukelbir et al. (2017). In this chapter, we describe the specifics of how ADVI maximizes the variational objective.\n\n\nADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.\n\n\nADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by elbo_samples. We recommend a default value of \\(100\\), as we only evaluate the ELBO every eval_elbo iterations, which also defaults to \\(100\\).\n\n\n\nADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by grad_samples. We recommend a default value of \\(1\\), as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.\n\n\n\nADVI uses a finite-memory version of adaGrad Duchi, Hazan, and Singer (2011). This has a single parameter that we expose, denoted eta. We now have a warmup adaptation phase that selects a good value for eta. The procedure does a heuristic search over eta values that span 5 orders of magnitude.\n\n\n\nADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by tol_rel_obj, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.",
+    "text": "Convergence monitoring in (L-)BFGS is controlled by a number of tolerance values, any one of which being satisfied causes the algorithm to terminate with a solution. Any of the convergence tests can be disabled by setting its corresponding tolerance parameter to zero. The tests for convergence are as follows.\n\n\nThe parameters \\(\\theta_i\\) in iteration \\(i\\) are considered to have converged with respect to tolerance tol_param if\n\\[\n|| \\theta_{i} - \\theta_{i-1} || &lt; \\mathtt{tol\\_param}.\n\\]\n\n\n\nThe (unnormalized) log density \\(\\log p(\\theta_{i}|y)\\) for the parameters \\(\\theta_i\\) in iteration \\(i\\) given data \\(y\\) is considered to have converged with respect to tolerance tol_obj if\n\\[\n\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right| &lt;\n\\mathtt{tol\\_obj}.\n\\]\nThe log density is considered to have converged to within relative tolerance tol_rel_obj if\n\\[\n\\frac{\\left| \\log p(\\theta_{i}|y) - \\log p(\\theta_{i-1}|y) \\right|}{\\\n  \\max\\left(\\left| \\log p(\\theta_{i}|y)\\right|,\\left| \\log\n      p(\\theta_{i-1}|y)\\right|,1.0\\right)}\n&lt; \\mathtt{tol\\_rel\\_obj} * \\epsilon.\n\\]\n\n\n\nThe gradient is considered to have converged to 0 relative to a specified tolerance tol_grad if\n\\[\n|| g_{i} || &lt; \\mathtt{tol\\_grad},\n\\] where \\(\\nabla_{\\theta}\\) is the gradient operator with respect to \\(\\theta\\) and \\(g_{i} = \\nabla_{\\theta} \\log p(\\theta | y)\\) is the gradient at iteration \\(i\\) evaluated at \\(\\theta^{(i)}\\), the value on the \\(i\\)-th posterior iteration.\nThe gradient is considered to have converged to 0 relative to a specified relative tolerance tol_rel_grad if\n\\[\n\\frac{g_{i}^T \\hat{H}_{i}^{-1} g_{i} }{ \\max\\left(\\left|\\log\np(\\theta_{i}|y)\\right|,1.0\\right) }\n\\ &lt; \\\n\\mathtt{tol\\_rel\\_grad} * \\epsilon,\n\\]\nwhere \\(\\hat{H}_{i}\\) is the estimate of the Hessian at iteration \\(i\\), \\(|u|\\) is the absolute value (L1 norm) of \\(u\\), \\(||u||\\) is the vector length (L2 norm) of \\(u\\), and \\(\\epsilon \\approx 2e-16\\) is machine precision.\n\n\n\n\nThe initial step size parameter \\(\\alpha\\) for BFGS-style optimizers may be specified. If the first iteration takes a long time (and requires a lot of function evaluations) initialize \\(\\alpha\\) to be the roughly equal to the \\(\\alpha\\) used in that first iteration. The default value is intentionally small, 0.001, which is reasonable for many problems but might be too large or too small depending on the objective function and initialization. Being too big or too small just means that the first iteration will take longer (i.e., require more gradient evaluations) before the line search finds a good step length. It’s not a critical parameter, but for optimizing the same model multiple times (as you tweak things or with different data), being able to tune \\(\\alpha\\) can save some real time.\n\n\n\nL-BFGS has a command-line argument which controls the size of the history it uses to approximate the Hessian. The value should be less than the dimensionality of the parameter space and, in general, relatively small values (5–10) are sufficient; the default value is 5.\nIf L-BFGS performs poorly but BFGS performs well, consider increasing the history size. Increasing history size will increase the memory usage, although this is unlikely to be an issue for typical Stan models.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Variational Inference"
+      "Optimization"
     ]
   },
   {
-    "objectID": "reference-manual/variational.html#stochastic-gradient-ascent",
-    "href": "reference-manual/variational.html#stochastic-gradient-ascent",
-    "title": "Variational Inference",
+    "objectID": "reference-manual/optimization.html#writing-models-for-optimization",
+    "href": "reference-manual/optimization.html#writing-models-for-optimization",
+    "title": "Optimization",
     "section": "",
-    "text": "ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.\n\n\nADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by elbo_samples. We recommend a default value of \\(100\\), as we only evaluate the ELBO every eval_elbo iterations, which also defaults to \\(100\\).\n\n\n\nADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by grad_samples. We recommend a default value of \\(1\\), as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.\n\n\n\nADVI uses a finite-memory version of adaGrad Duchi, Hazan, and Singer (2011). This has a single parameter that we expose, denoted eta. We now have a warmup adaptation phase that selects a good value for eta. The procedure does a heuristic search over eta values that span 5 orders of magnitude.\n\n\n\nADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by tol_rel_obj, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.",
+    "text": "For constrained optimization problems, for instance, with a standard deviation parameter \\(\\sigma\\) constrained so that \\(\\sigma &gt; 0\\), it can be much more efficient to declare a parameter sigma with no constraints. This allows the optimizer to easily get close to 0 without having to tend toward \\(-\\infty\\) on the \\(\\log \\sigma\\) scale.\nWith unconstrained parameterizations of parameters with constrained support, it is important to provide a custom initialization that is within the support. For example, declaring a vector\nvector[M] sigma;\nand using the default random initialization which is \\(\\mathsf{Uniform}(-2, 2)\\) on the unconstrained scale means that there is only a \\(2^{-M}\\) chance that the initialization will be within support.\nFor any given optimization problem, it is probably worthwhile trying the program both ways, with and without the constraint, to see which one is more efficient.",
     "crumbs": [
       "Reference Manual",
       "Algorithms",
-      "Variational Inference"
+      "Optimization"
     ]
   },
   {
-    "objectID": "stan-users-guide/algebraic-equations.html",
-    "href": "stan-users-guide/algebraic-equations.html",
-    "title": "Solving Algebraic Equations",
+    "objectID": "reference-manual/references.html",
+    "href": "reference-manual/references.html",
+    "title": "References",
     "section": "",
-    "text": "Stan provides a built-in mechanism for specifying systems of algebraic equations. These systems can be solved either with the Newton method, as implemented in the Kinsol package (Hindmarsh et al. 2005), or with the Powell hybrid method (Powell 1970). The function signatures for Stan’s algebraic solvers are fully described in the algebraic solver section of the reference manual.\nSolving any system of algebraic equations can be translated into a root-finding problem, that is, given a function \\(f\\), we wish to find \\(y\\) such that \\(f(y) = 0\\).\n\n\nFor systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.\nAs an illustrative example, we consider the following nonlinear system of two equations with two unknowns: \\[\\begin{align*}\nz_1 &= y_1 - \\theta_1 \\\\\nz_2 &= y_1 y_2 + \\theta_2\n\\end{align*}\\]\nOur goal is to simultaneously solve all equations for \\(y_1\\) and \\(y_2\\), such that the vector \\(z\\) goes to 0.\n\n\n\nA system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the user-defined functions section for more information on coding user-defined functions).\nvector system(vector y,              // unknowns\n              vector theta,          // parameters\n              data array[] real x_r, // data (real)\n              array[] int x_i) {     // data (integer)\n  vector[2] z;\n  z[1] = y[1] - theta[1];\n  z[2] = y[1] * y[2] - theta[2];\n  return z;\n}\nThe function takes the unknowns we wish to solve for in y (a vector), the system parameters in theta (a vector), the real data in x_r (a real array) and the integer data in x_i (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.\nThe body of the system function here could also be coded using a row vector constructor and transposition,\nreturn [ y[1] - theta[1],\n         y[1] * y[2] - theta[2] ]';\nAs systems get more complicated, naming the intermediate expressions goes a long way toward readability.\n\n\nThe function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.\n\n\n\n\nLet’s suppose \\(\\theta = (3, 6)\\). To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.\nThe solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about algebraic solvers with control parameters.\nThe following code returns the solution to our nonlinear algebraic system:\ntransformed data {\n  vector[2] y_guess = [1, 1]';\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\ntransformed parameters {\n  vector[2] theta = [3, 6]';\n  vector[2] y;\n\n  y = solve_newton(system, y_guess, theta, x_r, x_i);\n}\nwhich returns \\(y = (3, -2)\\).\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, y_guess, which may be a data or a parameter vector.\n\n\n\nThe Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function \\(f\\) with respect to the unknowns \\(x\\) must be invertible. This requires the Jacobian to be square, meaning \\(f(y)\\) and \\(y\\) have the same length or, in other words the number of equations in the system is the same as the number of unknowns.\n\n\n\nCertain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.\nOn the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.\n\n\n\n\nThe call to the algebraic solver shown previously uses the default control settings. The _tol variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.\ny = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,\n                     theta, x_r, x_i);\nFor the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.\nThe default values for the control arguments are respectively scaling_step = 1e-3 (\\(10^{-3}\\)), rel_tol = 1e-10 (\\(10^{-10}\\)), f_tol = 1e-6 (\\(10^{-6}\\)), and max_steps = 200 (\\(200\\)).\n\n\nThe relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.\nSmaller relative tolerances produce more accurate solutions but require more computational time.\n\n\nThe tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by Timonen et al. (2023).\n\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.",
-    "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Solving Algebraic Equations"
-    ]
+    "text": "References\n\n\n\n\n Back to top"
   },
   {
-    "objectID": "stan-users-guide/algebraic-equations.html#example-system-of-nonlinear-algebraic-equations",
-    "href": "stan-users-guide/algebraic-equations.html#example-system-of-nonlinear-algebraic-equations",
-    "title": "Solving Algebraic Equations",
+    "objectID": "reference-manual/reproducibility.html",
+    "href": "reference-manual/reproducibility.html",
+    "title": "Reproducibility",
     "section": "",
-    "text": "For systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.\nAs an illustrative example, we consider the following nonlinear system of two equations with two unknowns: \\[\\begin{align*}\nz_1 &= y_1 - \\theta_1 \\\\\nz_2 &= y_1 y_2 + \\theta_2\n\\end{align*}\\]\nOur goal is to simultaneously solve all equations for \\(y_1\\) and \\(y_2\\), such that the vector \\(z\\) goes to 0.",
+    "text": "Floating point operations on modern computers are notoriously difficult to replicate because the fundamental arithmetic operations, right down to the IEEE 754 encoding level, are not fully specified. The primary problem is that the precision of operations varies across different hardware platforms and software implementations.\nStan is designed to allow full reproducibility. However, this is only possible up to the external constraints imposed by floating point arithmetic.\nStan results will only be exactly reproducible if all of the following components are identical:\n\nStan version\nStan interface (RStan, PyStan, CmdStan) and version, plus version of interface language (R, Python, shell)\nversions of included libraries (Boost and Eigen)\noperating system version\ncomputer hardware including CPU, motherboard and memory\nC++ compiler, including version, compiler flags, and linked libraries\nsame configuration of call to Stan, including random seed, chain ID, initialization and data\n\nIt doesn’t matter if you use a stable release version of Stan or the version with a particular Git hash tag. The same goes for all of the interfaces, compilers, and so on. The point is that if any of these moving parts changes in some way, floating point results may change.\nConcretely, if you compile a single Stan program using the same CmdStan code base, but changed the optimization flag (-O3 vs. -O2 or -O0), the two programs may not return the identical stream of results. Thus it is very hard to guarantee reproducibility on externally managed hardware, like in a cluster or even a desktop managed by an IT department or with automatic updates turned on.\nIf, however, you compiled a Stan program today using one set of flags, took the computer away from the internet and didn’t allow it to update anything, then came back in a decade and recompiled the Stan program in the same way, you should get the same results.\nThe data needs to be the same down to the bit level. For example, if you are running in RStan, Rcpp handles the conversion between R’s floating point numbers and C++ doubles. If Rcpp changes the conversion process or use different types, the results are not guaranteed to be the same down to the bit level.\nThe compiler and compiler settings can also be an issue. There is a nice discussion of the issues and how to control reproducibility in Intel’s proprietary compiler by Corden and Kreitzer (2014).\n\n\nAs noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.\nHowever, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still does not guarantee exact reproducibility between it and other versions.\n\nStan 2.28 changed the default chain ID for MCMC from 0 to 1. Users who had set a seed but not a chain ID would observe completely different outputs.\nStan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Solving Algebraic Equations"
+      "Reference Manual",
+      "Usage",
+      "Reproducibility"
     ]
   },
   {
-    "objectID": "stan-users-guide/algebraic-equations.html#coding-an-algebraic-system",
-    "href": "stan-users-guide/algebraic-equations.html#coding-an-algebraic-system",
-    "title": "Solving Algebraic Equations",
+    "objectID": "reference-manual/reproducibility.html#notable-changes-across-versions",
+    "href": "reference-manual/reproducibility.html#notable-changes-across-versions",
+    "title": "Reproducibility",
     "section": "",
-    "text": "A system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the user-defined functions section for more information on coding user-defined functions).\nvector system(vector y,              // unknowns\n              vector theta,          // parameters\n              data array[] real x_r, // data (real)\n              array[] int x_i) {     // data (integer)\n  vector[2] z;\n  z[1] = y[1] - theta[1];\n  z[2] = y[1] * y[2] - theta[2];\n  return z;\n}\nThe function takes the unknowns we wish to solve for in y (a vector), the system parameters in theta (a vector), the real data in x_r (a real array) and the integer data in x_i (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.\nThe body of the system function here could also be coded using a row vector constructor and transposition,\nreturn [ y[1] - theta[1],\n         y[1] * y[2] - theta[2] ]';\nAs systems get more complicated, naming the intermediate expressions goes a long way toward readability.\n\n\nThe function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.",
+    "text": "As noted above, there is no guarantee that the same results will be reproducible between two different versions of Stan, even if the same settings and environment are used.\nHowever, there are occassionally notable changes which would affect many if not all users, and these are noted here. The absence of a version from this list still does not guarantee exact reproducibility between it and other versions.\n\nStan 2.28 changed the default chain ID for MCMC from 0 to 1. Users who had set a seed but not a chain ID would observe completely different outputs.\nStan 2.35 changed the default pseudo-random number generator used by the Stan algorithms. There is no relationship between seeds in versions pre-2.35 and version 2.35 and on.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Solving Algebraic Equations"
+      "Reference Manual",
+      "Usage",
+      "Reproducibility"
     ]
   },
   {
-    "objectID": "stan-users-guide/algebraic-equations.html#calling-the-algebraic-solver",
-    "href": "stan-users-guide/algebraic-equations.html#calling-the-algebraic-solver",
-    "title": "Solving Algebraic Equations",
+    "objectID": "reference-manual/syntax.html",
+    "href": "reference-manual/syntax.html",
+    "title": "Language Syntax",
     "section": "",
-    "text": "Let’s suppose \\(\\theta = (3, 6)\\). To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.\nThe solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about algebraic solvers with control parameters.\nThe following code returns the solution to our nonlinear algebraic system:\ntransformed data {\n  vector[2] y_guess = [1, 1]';\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\ntransformed parameters {\n  vector[2] theta = [3, 6]';\n  vector[2] y;\n\n  y = solve_newton(system, y_guess, theta, x_r, x_i);\n}\nwhich returns \\(y = (3, -2)\\).\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, y_guess, which may be a data or a parameter vector.\n\n\n\nThe Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function \\(f\\) with respect to the unknowns \\(x\\) must be invertible. This requires the Jacobian to be square, meaning \\(f(y)\\) and \\(y\\) have the same length or, in other words the number of equations in the system is the same as the number of unknowns.\n\n\n\nCertain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.\nOn the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.",
+    "text": "This chapter defines the basic syntax of the Stan modeling language using a Backus-Naur form (BNF) grammar plus extra-grammatical constraints on function typing and operator precedence and associativity.\n\n\n\n\nIn the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by &lt; and &gt;. A square brackets ([A]) indicates optionality of A. A postfixed Kleene star (A*) indicates zero or more occurrences of A. Parenthesis can be used to group symbols together in productions.\nFinally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library Menhir. A rule like &lt;list(x)&gt; ::= x (COMMA x)* declares a generic list rule, which can later be applied to others by the symbol &lt;list(&lt;expression&gt;)&gt;.\nThe following representation is constructed directly from the OCaml reference parser using a tool called Obelisk. The raw output is available here.\n\n\n\n\n&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]\n              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]\n              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF\n\n&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF\n\n&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE\n\n&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE\n\n&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE\n                             &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*\n                       RBRACE\n\n&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE\n                                   &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE\n                                 &lt;top_vardecl_or_statement&gt;* RBRACE\n\n\n\n&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA\n                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;\n\n&lt;return_type&gt; ::= VOID\n                | &lt;unsized_type&gt;\n\n&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;\n\n&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;\n                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;\n                 | &lt;basic_type&gt;\n                 | &lt;unsized_tuple_type&gt;\n\n&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA [&lt;unsized_type&gt;\n                         (COMMA &lt;unsized_type&gt;)*] RPAREN\n\n&lt;basic_type&gt; ::= INT\n               | REAL\n               | COMPLEX\n               | VECTOR\n               | ROWVECTOR\n               | MATRIX\n               | COMPLEXVECTOR\n               | COMPLEXROWVECTOR\n               | COMPLEXMATRIX\n\n&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK\n\n\n\n\n&lt;identifier&gt; ::= IDENTIFIER\n               | TRUNCATE\n\n&lt;decl_identifier&gt; ::= &lt;identifier&gt;\n\n&lt;no_assign&gt; ::= UNREACHABLE\n\n&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]\n\n&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;\n                                      &lt;optional_assignment(rhs)&gt;\n\n&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;\n                           &lt;optional_assignment(rhs)&gt; SEMICOLON\n                         | &lt;higher_type(type_rule)&gt;\n                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA\n                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON\n\n&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;\n                           | &lt;tuple_type(type_rule)&gt;\n                           | type_rule\n\n&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule\n                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;\n\n&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA\n                            [&lt;higher_type(type_rule)&gt; (COMMA\n                            &lt;higher_type(type_rule)&gt;)*] RPAREN\n\n&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;\n                           | SEMICOLON\n\n&lt;sized_basic_type&gt; ::= INT\n                     | REAL\n                     | COMPLEX\n                     | VECTOR LBRACK &lt;expression&gt; RBRACK\n                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                       RBRACK\n\n&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]\n                 | REAL &lt;type_constraint&gt;\n                 | COMPLEX &lt;type_constraint&gt;\n                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;\n                   RBRACK\n                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | ORDERED LBRACK &lt;expression&gt; RBRACK\n                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK\n                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK\n                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROVEC LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]\n                   RBRACK\n                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                   RBRACK\n\n&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]\n                    | LABRACK &lt;offset_mult&gt; RABRACK\n\n&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN\n            &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN\n            &lt;constr_expression&gt;\n          | LOWER ASSIGN &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt;\n\n&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN\n                  &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN\n                  &lt;constr_expression&gt;\n                | OFFSET ASSIGN &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt;\n\n&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK\n\n\n\n\n&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;\n               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;\n               | &lt;prefixOp&gt; &lt;expression&gt;\n               | &lt;expression&gt; &lt;postfixOp&gt;\n               | &lt;common_expression&gt;\n\n&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;\n                        &lt;constr_expression&gt;\n                      | &lt;prefixOp&gt; &lt;constr_expression&gt;\n                      | &lt;constr_expression&gt; &lt;postfixOp&gt;\n                      | &lt;common_expression&gt;\n\n&lt;common_expression&gt; ::= &lt;identifier&gt;\n                      | INTNUMERAL\n                      | REALNUMERAL\n                      | DOTNUMERAL\n                      | IMAGNUMERAL\n                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE\n                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK\n                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | TARGET LPAREN RPAREN\n                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;\n                        (COMMA &lt;expression&gt;)*] RPAREN\n                      | LPAREN &lt;expression&gt; COMMA [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | &lt;common_expression&gt; DOTNUMERAL\n                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK\n                      | LPAREN &lt;expression&gt; RPAREN\n\n&lt;prefixOp&gt; ::= BANG\n             | MINUS\n             | PLUS\n\n&lt;postfixOp&gt; ::= TRANSPOSE\n\n&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;\n            | &lt;logicalBinOp&gt;\n\n&lt;arithmeticBinOp&gt; ::= PLUS\n                    | MINUS\n                    | TIMES\n                    | DIVIDE\n                    | IDIVIDE\n                    | MODULO\n                    | LDIVIDE\n                    | ELTTIMES\n                    | ELTDIVIDE\n                    | HAT\n                    | ELTPOW\n\n&lt;logicalBinOp&gt; ::= OR\n                 | AND\n                 | EQUALS\n                 | NEQUALS\n                 | LABRACK\n                 | LEQ\n                 | RABRACK\n                 | GEQ\n\n&lt;indexes&gt; ::= epsilon\n            | COLON\n            | &lt;expression&gt;\n            | &lt;expression&gt; COLON\n            | COLON &lt;expression&gt;\n            | &lt;expression&gt; COLON &lt;expression&gt;\n            | &lt;indexes&gt; COMMA &lt;indexes&gt;\n\n&lt;printables&gt; ::= &lt;expression&gt;\n               | &lt;string_literal&gt;\n               | &lt;printables&gt; COMMA &lt;printables&gt;\n\n\n\n&lt;statement&gt; ::= &lt;atomic_statement&gt;\n              | &lt;nested_statement&gt;\n\n&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;\n                       SEMICOLON\n                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                       &lt;expression&gt;)*] RPAREN SEMICOLON\n                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;\n                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON\n                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | BREAK SEMICOLON\n                     | CONTINUE SEMICOLON\n                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | RETURN &lt;expression&gt; SEMICOLON\n                     | RETURN SEMICOLON\n                     | SEMICOLON\n\n&lt;assignment_op&gt; ::= ASSIGN\n                  | PLUSASSIGN\n                  | MINUSASSIGN\n                  | TIMESASSIGN\n                  | DIVIDEASSIGN\n                  | ELTTIMESASSIGN\n                  | ELTDIVIDEASSIGN\n\n&lt;string_literal&gt; ::= STRINGLITERAL\n\n&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK\n\n&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                       ELSE &lt;vardecl_or_statement&gt;\n                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | WHILE LPAREN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON\n                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE\n                       &lt;vardecl_or_statement&gt;* RBRACE\n                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;\n                         | &lt;var_decl&gt;\n\n&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;\n                             | &lt;top_var_decl&gt;\n\n\n\n\nMany of the tokens used in the BNF grammars follow obviously from their names: DATABLOCK is the literal string ‘data’, COMMA is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the operator precedence table.\nA few tokens are not so obvious, and are defined here in regular expressions:\nIDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*\n\nSTRINGLITERAL = \".*\"\n\nINTNUMERAL = [0-9]+ (_ [0-9]+)*\n\nEXPLITERAL = [eE] [+-]? INTNUMERAL\n\nREALNUMERAL = INTNUMERAL \\. INTNUMERAL? EXPLITERAL?\n            | \\. INTNUMERAL EXPLITERAL\n            | INTNUMERAL EXPLITERAL\n\nIMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i\n\nDOTNUMERAL = \\. INTNUMERAL\n\n\n\n\n\nA well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.\n\n\n\nIn the Stan grammar provided in this chapter, the expression 1 + 2 * 3 has two parses. As described in the operator precedence table, Stan disambiguates between the meaning \\(1\n+ (2 \\times 3)\\) and the meaning \\((1 + 2) \\times 3\\) based on operator precedences and associativities.\n\n\n\nIn a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.\n\n\n\nThe types of expressions used for elements in array expressions ('{' expressions '}') must all be of the same type or a mixture of scalar (int, real and complex) types (in which case the result is promoted to be of the highest type on the int -&gt; real -&gt; complex hierarchy).\n\n\n\nInteger literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.\n\n\n\nBoth the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.\n\n\n\nThe for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).\n\n\n\nThe arguments to a print statement cannot be void.\n\n\n\nThe break and continue statements may only be used within the body of a for-loop or while-loop.\n\n\n\nSome constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.\n\n\nFunctions ending in _rng may only be called in the transformed data and generated quantities block, and within the bodies of user-defined functions with names ending in _rng.\n\n\n\nUnnormalized distributions (with suffixes _lupmf or _lupdf) may only be called in the model block, user-defined probability functions, or within the bodies of user defined functions which end in _lp.\n\n\n\ntarget += statements can only be used inside of the model block or user-defined functions which end in _lp.\nUser defined functions which end in _lp and the target() function can only be used in the model block, transformed parameters block, and in the bodies of other user defined functions which end in _lp.\nSampling statements (using ~) can only be used in the model block or in the bodies of user-defined functions which end in _lp.\njacobian += statements can only be used inside of the transformed parameters block or in functions that end with _jacobian.\n\n\n\n\nA probability function literal must have one of the following suffixes: _lpdf, _lpmf, _lcdf, or _lccdf.\n\n\n\nStandalone expressions used as indexes must denote either an integer (int) or an integer array (array[] int). Expressions participating in range indexes (e.g., a and b in a : b) must denote integers (int).\nA second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type array[ , , ] matrix, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Solving Algebraic Equations"
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
     ]
   },
   {
-    "objectID": "stan-users-guide/algebraic-equations.html#algebra-control.section",
-    "href": "stan-users-guide/algebraic-equations.html#algebra-control.section",
-    "title": "Solving Algebraic Equations",
+    "objectID": "reference-manual/syntax.html#bnf-grammars",
+    "href": "reference-manual/syntax.html#bnf-grammars",
+    "title": "Language Syntax",
     "section": "",
-    "text": "The call to the algebraic solver shown previously uses the default control settings. The _tol variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.\ny = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,\n                     theta, x_r, x_i);\nFor the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.\nThe default values for the control arguments are respectively scaling_step = 1e-3 (\\(10^{-3}\\)), rel_tol = 1e-10 (\\(10^{-10}\\)), f_tol = 1e-6 (\\(10^{-6}\\)), and max_steps = 200 (\\(200\\)).\n\n\nThe relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.\nSmaller relative tolerances produce more accurate solutions but require more computational time.\n\n\nThe tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by Timonen et al. (2023).\n\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.",
+    "text": "In the following BNF grammars, tokens are represented in ALLCAPS. Grammar non-terminals are surrounded by &lt; and &gt;. A square brackets ([A]) indicates optionality of A. A postfixed Kleene star (A*) indicates zero or more occurrences of A. Parenthesis can be used to group symbols together in productions.\nFinally, this grammar uses the concept of “parameterized nonterminals” as used in the parsing library Menhir. A rule like &lt;list(x)&gt; ::= x (COMMA x)* declares a generic list rule, which can later be applied to others by the symbol &lt;list(&lt;expression&gt;)&gt;.\nThe following representation is constructed directly from the OCaml reference parser using a tool called Obelisk. The raw output is available here.\n\n\n\n\n&lt;program&gt; ::= [&lt;function_block&gt;] [&lt;data_block&gt;] [&lt;transformed_data_block&gt;]\n              [&lt;parameters_block&gt;] [&lt;transformed_parameters_block&gt;]\n              [&lt;model_block&gt;] [&lt;generated_quantities_block&gt;] EOF\n\n&lt;functions_only&gt; ::= &lt;function_def&gt;* EOF\n\n&lt;function_block&gt; ::= FUNCTIONBLOCK LBRACE &lt;function_def&gt;* RBRACE\n\n&lt;data_block&gt; ::= DATABLOCK LBRACE &lt;top_var_decl_no_assign&gt;* RBRACE\n\n&lt;transformed_data_block&gt; ::= TRANSFORMEDDATABLOCK LBRACE\n                             &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;parameters_block&gt; ::= PARAMETERSBLOCK LBRACE &lt;top_var_decl_no_assign&gt;*\n                       RBRACE\n\n&lt;transformed_parameters_block&gt; ::= TRANSFORMEDPARAMETERSBLOCK LBRACE\n                                   &lt;top_vardecl_or_statement&gt;* RBRACE\n\n&lt;model_block&gt; ::= MODELBLOCK LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;generated_quantities_block&gt; ::= GENERATEDQUANTITIESBLOCK LBRACE\n                                 &lt;top_vardecl_or_statement&gt;* RBRACE\n\n\n\n&lt;function_def&gt; ::= &lt;return_type&gt; &lt;decl_identifier&gt; LPAREN [&lt;arg_decl&gt; (COMMA\n                   &lt;arg_decl&gt;)*] RPAREN &lt;statement&gt;\n\n&lt;return_type&gt; ::= VOID\n                | &lt;unsized_type&gt;\n\n&lt;arg_decl&gt; ::= [DATABLOCK] &lt;unsized_type&gt; &lt;decl_identifier&gt;\n\n&lt;unsized_type&gt; ::= ARRAY &lt;unsized_dims&gt; &lt;basic_type&gt;\n                 | ARRAY &lt;unsized_dims&gt; &lt;unsized_tuple_type&gt;\n                 | &lt;basic_type&gt;\n                 | &lt;unsized_tuple_type&gt;\n\n&lt;unsized_tuple_type&gt; ::= TUPLE LPAREN &lt;unsized_type&gt; COMMA [&lt;unsized_type&gt;\n                         (COMMA &lt;unsized_type&gt;)*] RPAREN\n\n&lt;basic_type&gt; ::= INT\n               | REAL\n               | COMPLEX\n               | VECTOR\n               | ROWVECTOR\n               | MATRIX\n               | COMPLEXVECTOR\n               | COMPLEXROWVECTOR\n               | COMPLEXMATRIX\n\n&lt;unsized_dims&gt; ::= LBRACK COMMA* RBRACK\n\n\n\n\n&lt;identifier&gt; ::= IDENTIFIER\n               | TRUNCATE\n\n&lt;decl_identifier&gt; ::= &lt;identifier&gt;\n\n&lt;no_assign&gt; ::= UNREACHABLE\n\n&lt;optional_assignment(rhs)&gt; ::= [ASSIGN rhs]\n\n&lt;id_and_optional_assignment(rhs)&gt; ::= &lt;decl_identifier&gt;\n                                      &lt;optional_assignment(rhs)&gt;\n\n&lt;decl(type_rule, rhs)&gt; ::= type_rule &lt;decl_identifier&gt; &lt;dims&gt;\n                           &lt;optional_assignment(rhs)&gt; SEMICOLON\n                         | &lt;higher_type(type_rule)&gt;\n                           &lt;id_and_optional_assignment(rhs)&gt; (COMMA\n                           &lt;id_and_optional_assignment(rhs)&gt;)* SEMICOLON\n\n&lt;higher_type(type_rule)&gt; ::= &lt;array_type(type_rule)&gt;\n                           | &lt;tuple_type(type_rule)&gt;\n                           | type_rule\n\n&lt;array_type(type_rule)&gt; ::= &lt;arr_dims&gt; type_rule\n                          | &lt;arr_dims&gt; &lt;tuple_type(type_rule)&gt;\n\n&lt;tuple_type(type_rule)&gt; ::= TUPLE LPAREN &lt;higher_type(type_rule)&gt; COMMA\n                            [&lt;higher_type(type_rule)&gt; (COMMA\n                            &lt;higher_type(type_rule)&gt;)*] RPAREN\n\n&lt;var_decl&gt; ::= &lt;decl(&lt;sized_basic_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;expression&gt;)&gt;\n\n&lt;top_var_decl_no_assign&gt; ::= &lt;decl(&lt;top_var_type&gt;, &lt;no_assign&gt;)&gt;\n                           | SEMICOLON\n\n&lt;sized_basic_type&gt; ::= INT\n                     | REAL\n                     | COMPLEX\n                     | VECTOR LBRACK &lt;expression&gt; RBRACK\n                     | ROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | MATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                     | COMPLEXVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXROWVECTOR LBRACK &lt;expression&gt; RBRACK\n                     | COMPLEXMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                       RBRACK\n\n&lt;top_var_type&gt; ::= INT [LABRACK &lt;range&gt; RABRACK]\n                 | REAL &lt;type_constraint&gt;\n                 | COMPLEX &lt;type_constraint&gt;\n                 | VECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | ROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | MATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | COMPLEXVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt; RBRACK\n                 | COMPLEXROWVECTOR &lt;type_constraint&gt; LBRACK &lt;expression&gt;\n                   RBRACK\n                 | COMPLEXMATRIX &lt;type_constraint&gt; LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | ORDERED LBRACK &lt;expression&gt; RBRACK\n                 | POSITIVEORDERED LBRACK &lt;expression&gt; RBRACK\n                 | SIMPLEX LBRACK &lt;expression&gt; RBRACK\n                 | UNITVECTOR LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROVEC LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCORR LBRACK &lt;expression&gt; RBRACK\n                 | CHOLESKYFACTORCOV LBRACK &lt;expression&gt; [COMMA &lt;expression&gt;]\n                   RBRACK\n                 | CORRMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | COVMATRIX LBRACK &lt;expression&gt; RBRACK\n                 | SUMTOZEROMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt; RBRACK\n                 | STOCHASTICCOLUMNMATRIX LBRACK &lt;expression&gt; COMMA\n                   &lt;expression&gt; RBRACK\n                 | STOCHASTICROWMATRIX LBRACK &lt;expression&gt; COMMA &lt;expression&gt;\n                   RBRACK\n\n&lt;type_constraint&gt; ::= [LABRACK &lt;range&gt; RABRACK]\n                    | LABRACK &lt;offset_mult&gt; RABRACK\n\n&lt;range&gt; ::= LOWER ASSIGN &lt;constr_expression&gt; COMMA UPPER ASSIGN\n            &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt; COMMA LOWER ASSIGN\n            &lt;constr_expression&gt;\n          | LOWER ASSIGN &lt;constr_expression&gt;\n          | UPPER ASSIGN &lt;constr_expression&gt;\n\n&lt;offset_mult&gt; ::= OFFSET ASSIGN &lt;constr_expression&gt; COMMA MULTIPLIER ASSIGN\n                  &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt; COMMA OFFSET ASSIGN\n                  &lt;constr_expression&gt;\n                | OFFSET ASSIGN &lt;constr_expression&gt;\n                | MULTIPLIER ASSIGN &lt;constr_expression&gt;\n\n&lt;arr_dims&gt; ::= ARRAY LBRACK &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACK\n\n\n\n\n&lt;expression&gt; ::= &lt;expression&gt; QMARK &lt;expression&gt; COLON &lt;expression&gt;\n               | &lt;expression&gt; &lt;infixOp&gt; &lt;expression&gt;\n               | &lt;prefixOp&gt; &lt;expression&gt;\n               | &lt;expression&gt; &lt;postfixOp&gt;\n               | &lt;common_expression&gt;\n\n&lt;constr_expression&gt; ::= &lt;constr_expression&gt; &lt;arithmeticBinOp&gt;\n                        &lt;constr_expression&gt;\n                      | &lt;prefixOp&gt; &lt;constr_expression&gt;\n                      | &lt;constr_expression&gt; &lt;postfixOp&gt;\n                      | &lt;common_expression&gt;\n\n&lt;common_expression&gt; ::= &lt;identifier&gt;\n                      | INTNUMERAL\n                      | REALNUMERAL\n                      | DOTNUMERAL\n                      | IMAGNUMERAL\n                      | LBRACE &lt;expression&gt; (COMMA &lt;expression&gt;)* RBRACE\n                      | LBRACK [&lt;expression&gt; (COMMA &lt;expression&gt;)*] RBRACK\n                      | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | TARGET LPAREN RPAREN\n                      | &lt;identifier&gt; LPAREN &lt;expression&gt; BAR [&lt;expression&gt;\n                        (COMMA &lt;expression&gt;)*] RPAREN\n                      | LPAREN &lt;expression&gt; COMMA [&lt;expression&gt; (COMMA\n                        &lt;expression&gt;)*] RPAREN\n                      | &lt;common_expression&gt; DOTNUMERAL\n                      | &lt;common_expression&gt; LBRACK &lt;indexes&gt; RBRACK\n                      | LPAREN &lt;expression&gt; RPAREN\n\n&lt;prefixOp&gt; ::= BANG\n             | MINUS\n             | PLUS\n\n&lt;postfixOp&gt; ::= TRANSPOSE\n\n&lt;infixOp&gt; ::= &lt;arithmeticBinOp&gt;\n            | &lt;logicalBinOp&gt;\n\n&lt;arithmeticBinOp&gt; ::= PLUS\n                    | MINUS\n                    | TIMES\n                    | DIVIDE\n                    | IDIVIDE\n                    | MODULO\n                    | LDIVIDE\n                    | ELTTIMES\n                    | ELTDIVIDE\n                    | HAT\n                    | ELTPOW\n\n&lt;logicalBinOp&gt; ::= OR\n                 | AND\n                 | EQUALS\n                 | NEQUALS\n                 | LABRACK\n                 | LEQ\n                 | RABRACK\n                 | GEQ\n\n&lt;indexes&gt; ::= epsilon\n            | COLON\n            | &lt;expression&gt;\n            | &lt;expression&gt; COLON\n            | COLON &lt;expression&gt;\n            | &lt;expression&gt; COLON &lt;expression&gt;\n            | &lt;indexes&gt; COMMA &lt;indexes&gt;\n\n&lt;printables&gt; ::= &lt;expression&gt;\n               | &lt;string_literal&gt;\n               | &lt;printables&gt; COMMA &lt;printables&gt;\n\n\n\n&lt;statement&gt; ::= &lt;atomic_statement&gt;\n              | &lt;nested_statement&gt;\n\n&lt;atomic_statement&gt; ::= &lt;common_expression&gt; &lt;assignment_op&gt; &lt;expression&gt;\n                       SEMICOLON\n                     | &lt;identifier&gt; LPAREN [&lt;expression&gt; (COMMA\n                       &lt;expression&gt;)*] RPAREN SEMICOLON\n                     | &lt;expression&gt; TILDE &lt;identifier&gt; LPAREN [&lt;expression&gt;\n                       (COMMA &lt;expression&gt;)*] RPAREN [&lt;truncation&gt;] SEMICOLON\n                     | TARGET PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | JACOBIAN PLUSASSIGN &lt;expression&gt; SEMICOLON\n                     | BREAK SEMICOLON\n                     | CONTINUE SEMICOLON\n                     | PRINT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | REJECT LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | FATAL_ERROR LPAREN &lt;printables&gt; RPAREN SEMICOLON\n                     | RETURN &lt;expression&gt; SEMICOLON\n                     | RETURN SEMICOLON\n                     | SEMICOLON\n\n&lt;assignment_op&gt; ::= ASSIGN\n                  | PLUSASSIGN\n                  | MINUSASSIGN\n                  | TIMESASSIGN\n                  | DIVIDEASSIGN\n                  | ELTTIMESASSIGN\n                  | ELTDIVIDEASSIGN\n\n&lt;string_literal&gt; ::= STRINGLITERAL\n\n&lt;truncation&gt; ::= TRUNCATE LBRACK [&lt;expression&gt;] COMMA [&lt;expression&gt;] RBRACK\n\n&lt;nested_statement&gt; ::= IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                       ELSE &lt;vardecl_or_statement&gt;\n                     | IF LPAREN &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | WHILE LPAREN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; COLON\n                       &lt;expression&gt; RPAREN &lt;vardecl_or_statement&gt;\n                     | FOR LPAREN &lt;identifier&gt; IN &lt;expression&gt; RPAREN\n                       &lt;vardecl_or_statement&gt;\n                     | PROFILE LPAREN &lt;string_literal&gt; RPAREN LBRACE\n                       &lt;vardecl_or_statement&gt;* RBRACE\n                     | LBRACE &lt;vardecl_or_statement&gt;* RBRACE\n\n&lt;vardecl_or_statement&gt; ::= &lt;statement&gt;\n                         | &lt;var_decl&gt;\n\n&lt;top_vardecl_or_statement&gt; ::= &lt;statement&gt;\n                             | &lt;top_var_decl&gt;",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Solving Algebraic Equations"
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html",
-    "href": "stan-users-guide/clustering.html",
-    "title": "Clustering Models",
+    "objectID": "reference-manual/syntax.html#tokenizing-rules",
+    "href": "reference-manual/syntax.html#tokenizing-rules",
+    "title": "Language Syntax",
     "section": "",
-    "text": "Unsupervised methods for organizing data into groups are collectively referred to as clustering. This chapter describes the implementation in Stan of two widely used statistical clustering models, soft \\(K\\)-means and latent Dirichlet allocation (LDA). In addition, this chapter includes naive Bayesian classification, which can be viewed as a form of clustering which may be supervised. These models are typically expressed using discrete parameters for cluster assignments. Nevertheless, they can be implemented in Stan like any other mixture model by marginalizing out the discrete parameters (see the mixture modeling chapter).\n\n\nAs mentioned in the clustering section, clustering models and finite mixture models are really just two sides of the same coin. The “soft” \\(K\\)-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of \\(K\\)-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.\n\n\n\n\\(K\\)-means clustering is a method of clustering data represented as \\(D\\)-dimensional vectors. Specifically, there will be \\(N\\) items to be clustered, each represented as a vector \\(y_n \\in \\mathbb{R}^D\\). In the “soft” version of \\(K\\)-means, the assignments to clusters will be probabilistic.\n\n\n\\(K\\)-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters \\(K\\) and data vectors \\(y\\) as input.\n\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), randomly assign vector \\(y_n\\) to a cluster in \\(\\{1,\\dotsc,K\\}\\);\nRepeat\n\nFor each cluster \\(k\\) in \\(\\{1,\\dotsc,K\\}\\), compute the cluster centroid \\(\\mu_k\\) by averaging the vectors assigned to that cluster;\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), reassign \\(y_n\\) to the cluster \\(k\\) for which the (Euclidean) distance from \\(y_n\\) to \\(\\mu_k\\) is smallest;\nIf no vectors changed cluster, return the cluster assignments.\n\n\nThis algorithm is guaranteed to terminate.\n\n\n\nSoft \\(K\\)-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft \\(K\\)-means can be expressed (and coded in Stan) as a multivariate normal mixture model.\nIn the full generative model, each data point \\(n\\) in \\(\\{1,\\dotsc,N\\}\\) is assigned a cluster \\(z_n \\in \\{1,\\dotsc,K\\}\\) with symmetric uniform probability, \\[\nz_n \\sim \\textsf{categorical}(1/K),\n\\] where \\(1\\) is the unit vector of \\(K\\) dimensions, so that \\(1/K\\) is the symmetric \\(K\\)-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.\nThe data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment \\(z_n\\), \\[\ny_n \\sim  \\textsf{normal}(\\mu_{z[n]},\\Sigma_{z[n]})\n\\]\nThe sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters \\(k\\), \\[\n\\Sigma_k = \\mathrm{diag\\_matrix}({\\bf 1}),\n\\] so that the log multivariate normal can be implemented directly up to a proportion by \\[\n\\mathrm{normal}\\left( y_n \\mid \\mu_k, \\mathrm{diag\\_matrix}({\\bf 1}) \\right)\n\\propto \\exp \\left (- \\frac{1}{2} \\sum_{d=1}^D \\left( \\mu_{k,d} - y_{n,d}\n  \\right)^2 \\right).\n\\] The spatial perspective on \\(K\\)-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean \\(\\mu_k\\) to the data point \\(y_n\\).\n\n\n\nConsider the following Stan program for implementing \\(K\\)-means clustering.\ndata {\n  int&lt;lower=0&gt; N;        // number of data points\n  int&lt;lower=1&gt; D;        // number of dimensions\n  int&lt;lower=1&gt; K;        // number of clusters\n  array[N] vector[D] y;  // observations\n}\ntransformed data {\n  real&lt;upper=0&gt; neg_log_K;\n  neg_log_K = -log(K);\n}\nparameters {\n  array[K] vector[D] mu; // cluster means\n}\ntransformed parameters {\n  array[N, K] real&lt;upper=0&gt; soft_z; // log unnormalized clusters\n  for (n in 1:N) {\n    for (k in 1:K) {\n      soft_z[n, k] = neg_log_K\n                     - 0.5 * dot_self(mu[k] - y[n]);\n    }\n  }\n}\nmodel {\n  // prior\n  for (k in 1:K) {\n    mu[k] ~ std_normal();\n  }\n\n  // likelihood\n  for (n in 1:N) {\n    target += log_sum_exp(soft_z[n]);\n  }\n}\nThere is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.\nThe only parameter is mu, where mu[k] is the centroid for cluster \\(k\\). The transformed parameters soft_z[n] contain the log of the unnormalized cluster assignment probabilities. The vector soft_z[n] can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.\n\n\n\nThe multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., \\(L_2\\) distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on \\(L_1\\) distance (i.e., Manhattan or taxicab distance).\nWithin the multivariate normal version of \\(K\\)-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.\nAlthough there is no global spatial analog, it is common to see soft \\(K\\)-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.\n\n\n\n\nTwo problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the label switching section.\n\n\nCluster assignments are not identified—permuting the cluster mean vectors mu leads to a model with identical likelihoods. For instance, permuting the first two indexes in mu and the first two indexes in each soft_z[n] leads to an identical likelihood (and prior).\nThe lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft \\(K\\)-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.\n\n\n\nThe other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.\nBayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.\nIn light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.\n\n\n\n\nNaive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.1\nMultinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.\nNaive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.\nThe observed data consists of a sequence of \\(M\\) documents made up of bags of words drawn from a vocabulary of \\(V\\) distinct words. A document \\(m\\) has \\(N_m\\) words, which are indexed as \\(w_{m,1}, \\dotsc,\nw_{m,N[m]} \\in \\{1,\\dotsc,V\\}\\). Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) \\(K\\) is fixed.\nThe multinomial mixture model generates a single category \\(z_m \\in\n\\{1,\\dotsc,K\\}\\) for each document \\(m \\in \\{1,\\dotsc,M\\}\\) according to a categorical distribution, \\[\nz_m \\sim \\textsf{categorical}(\\theta).\n\\] The \\(K\\)-simplex parameter \\(\\theta\\) represents the prevalence of each category in the data.\nNext, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word \\(n\\) of document \\(m\\) being generated as \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m]}).\n\\] The parameter \\(\\phi_{z[m]}\\) is a \\(V\\)-simplex representing the probability of each word in the vocabulary in documents of category \\(z_m\\).\nThe parameters \\(\\theta\\) and \\(\\phi\\) are typically given symmetric Dirichlet priors. The prevalence \\(\\theta\\) is sometimes fixed to produce equal probabilities for each category \\(k \\in \\{1,\\dotsc,K\\}\\).\n\n\nThe specification for naive Bayes in the previous sections have used a ragged array notation for the words \\(w\\). Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.\n\\[\n\\begin{array}{lll}\n\\hline\n\\mathrm{n} \\qquad\\qquad\\qquad\\qquad & \\mathrm{w[n]} \\qquad & \\mathrm{doc[n]} \\\\\n\\hline\n1         & w_{1,1}    & 1 \\\\\n2         & w_{1,2}    & 1 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1     & w_{1,N[1]} & 1 \\\\\nN_1 + 1 & w_{2,1}    & 2 \\\\\nN_1 + 2 & w_{2,2}    & 2 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1 + N_2     & w_{2,N[2]} & 2 \\\\\nN_1 + N_2 + 1 & w_{3,1}    & 3 \\\\\n\\vdots       & \\vdots    & \\vdots \\\\\nN = \\sum_{m=1}^M N_m & w_{M,N[M]} & M \\\\\n\\hline\n\\end{array}\n\\]\nThe relevant variables for the program are N, the total number of words in all the documents, the word array w, and the document identity array doc.\n\n\n\nA naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows\ndata {\n  // training data\n  int&lt;lower=1&gt; K;               // num topics\n  int&lt;lower=1&gt; V;               // num words\n  int&lt;lower=0&gt; M;               // num docs\n  int&lt;lower=0&gt; N;               // total word instances\n  array[M] int&lt;lower=1, upper=K&gt; z;    // topic for doc m\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  // hyperparameters\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  simplex[K] theta;             // topic prevalence\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    z[m] ~ categorical(theta);\n  }\n  for (n in 1:N) {\n    w[n] ~ categorical(phi[z[doc[n]]]);\n  }\n}\nThe topic identifiers \\(z_m\\) are declared as data and the latent category assignments are included as part of the likelihood function.\n\n\n\nNaive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number \\(K\\) of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels z. Because z is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability\n\\[\\begin{align*}\n\\log\\, &p(w_{m,1},\\dotsc,w_{m,N_m} \\mid \\theta,\\phi) \\\\\n&= \\log \\sum_{k=1}^K\n    \\left( \\textsf{categorical}(k \\mid \\theta)\n           \\times \\prod_{n=1}^{N_m} \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right) \\\\\n&= \\log \\sum_{k=1}^K \\exp \\left(\n    \\log \\textsf{categorical}(k \\mid \\theta)\n     + \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right).\n\\end{align*}\\]\nThe last step shows how the log_sum_exp function can be used to stabilize the numerical calculation and return a result on the log scale.\nmodel {\n  array[M, K] real gamma;\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    for (k in 1:K) {\n      gamma[m, k] = categorical_lpmf(k | theta);\n    }\n  }\n  for (n in 1:N) {\n    for (k in 1:K) {\n      gamma[doc[n], k] = gamma[doc[n], k]\n                         + categorical_lpmf(w[n] | phi[k]);\n    }\n  }\n  for (m in 1:M) {\n    target += log_sum_exp(gamma[m]);\n  }\n}\nThe local variable gamma[m, k] represents the value \\[\n\\gamma_{m,k} = \\log \\textsf{categorical}(k \\mid \\theta)\n+ \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k).\n\\]\nGiven \\(\\gamma\\), the posterior probability that document \\(m\\) is assigned category \\(k\\) is \\[\n\\Pr[z_m = k \\mid w,\\alpha,\\beta]\n=\n\\exp \\left(\n\\gamma_{m,k}\n- \\log \\sum_{k=1}^K \\exp \\left( \\gamma_{m,k} \\right)\n\\right).\n\\]\nIf the variable gamma were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.\n\n\n\nFull Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see Gelman et al. (2013) and Gelman and Hill (2007) for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.\nTo specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables M2 N2, w2, and doc2. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.\n\n\n\nAn alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of gamma for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.\n\n\n\n\nLatent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model (Blei, Ng, and Jordan 2003) that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.\n\n\nThe basic model assumes each document is generated independently based on fixed hyperparameters. For document \\(m\\), the first step is to draw a topic distribution simplex \\(\\theta_m\\) over the \\(K\\) topics, \\[\n\\theta_m \\sim \\textsf{Dirichlet}(\\alpha).\n\\]\nThe prior hyperparameter \\(\\alpha\\) is fixed to a \\(K\\)-vector of positive values. Each word in the document is generated independently conditional on the distribution \\(\\theta_m\\). First, a topic \\(z_{m,n} \\in \\{1,\\dotsc,K\\}\\) is drawn for the word based on the document-specific topic-distribution, \\[\nz_{m,n} \\sim \\textsf{categorical}(\\theta_m).\n\\]\nFinally, the word \\(w_{m,n}\\) is drawn according to the word distribution for topic \\(z_{m,n}\\), \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m,n]}).\n\\] The distributions \\(\\phi_k\\) over words for topic \\(k\\) are also given a Dirichlet prior, \\[\n\\phi_k \\sim \\textsf{Dirichlet}(\\beta)\n\\]\nwhere \\(\\beta\\) is a fixed \\(V\\)-vector of positive values.\n\n\n\nAlthough Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is \\[\\begin{align*}\np(\\theta,\\phi \\mid w,\\alpha,\\beta)\n&\\propto p(\\theta \\mid \\alpha) \\, p(\\phi \\mid \\beta) \\, p(w \\mid \\theta,\\phi) \\\\\n&= \\prod_{m=1}^M p(\\theta_m \\mid \\alpha)\n    \\times \\prod_{k=1}^K p(\\phi_k \\mid \\beta)\n    \\times \\prod_{m=1}^M \\prod_{n=1}^{M[n]} p(w_{m,n} \\mid \\theta_m,\\phi).\n\\end{align*}\\]\nThe inner word-probability term is defined by summing out the topic assignments, \\[\\begin{align*}\np(w_{m,n} \\mid \\theta_m,\\phi)\n&= \\sum_{z=1}^K p(z,w_{m,n} \\mid \\theta_m,\\phi) \\\\\n&= \\sum_{z=1}^K p(z \\mid \\theta_m) \\, p(w_{m,n} \\mid \\phi_z).\n\\end{align*}\\]\nPlugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, \\[\\begin{align*}\n\\log\\, &p(\\theta,\\phi \\mid w,\\alpha,\\beta) \\\\\n&= \\sum_{m=1}^M \\log \\textsf{Dirichlet}(\\theta_m \\mid \\alpha)\n    + \\sum_{k=1}^K \\log \\textsf{Dirichlet}(\\phi_k \\mid \\beta) \\\\\n&\\qquad + \\sum_{m=1}^M \\sum_{n=1}^{N[m]} \\log \\left(\n    \\sum_{z=1}^K \\textsf{categorical}(z \\mid \\theta_m)\n    \\times \\textsf{categorical}(w_{m,n} \\mid \\phi_z)\n  \\right)\n\\end{align*}\\]\n\n\n\nApplying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.\ndata {\n  int&lt;lower=2&gt; K;               // num topics\n  int&lt;lower=2&gt; V;               // num words\n  int&lt;lower=1&gt; M;               // num docs\n  int&lt;lower=1&gt; N;               // total word instances\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  array[M] simplex[K] theta;    // topic dist for doc m\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  for (m in 1:M) {\n    theta[m] ~ dirichlet(alpha);  // prior\n  }\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);     // prior\n  }\n  for (n in 1:N) {\n    array[K] real gamma;\n    for (k in 1:K) {\n      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);\n    }\n    target += log_sum_exp(gamma);  // likelihood;\n  }\n}\nAs in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.\n\n\n\nTo account for correlations in the distribution of topics for documents, Blei and Lafferty (2007) introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.\nThe authors treat the prior as a fixed hyperparameter. They use an \\(L_1\\)-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.\n\n\nThe Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior alpha in the data declaration with the mean and covariance of the multivariate logistic normal prior.\ndata {\n  // ... data as before without alpha ...\n  vector[K] mu;          // topic mean\n  cov_matrix[K] Sigma;   // topic covariance\n}\nRather than drawing the simplex parameter theta from a Dirichlet, a parameter eta is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.\nparameters {\n  array[K] simplex[V] phi;     // word dist for topic k\n  array[M] vector[K] eta;      // topic dist for doc m\n}\ntransformed parameters {\n  array[M] simplex[K] theta;\n  for (m in 1:M) {\n    theta[m] = softmax(eta[m]);\n  }\n}\nmodel {\n  for (m in 1:M) {\n    eta[m] ~ multi_normal(mu, Sigma);\n  }\n  // ... model as before w/o prior for theta ...\n}\n\n\n\nBy adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean mu and covariance Sigma from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix Sigma may be encoded as follows.\n// ... data block as before, but without alpha ...\nparameters {\n  vector[K] mu;              // topic mean\n  corr_matrix[K] Omega;      // correlation matrix\n  vector&lt;lower=0&gt;[K] sigma;  // scales\n  array[M] vector[K] eta;    // logit topic dist for doc m\n  array[K] simplex[V] phi;   // word dist for topic k\n}\ntransformed parameters {\n  // ... eta as above ...\n  cov_matrix[K] Sigma;       // covariance matrix\n  for (m in 1:K) {\n    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];\n  }\n  for (m in 1:(K-1)) {\n    for (n in (m+1):K) {\n      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];\n      Sigma[n, m] = Sigma[m, n];\n    }\n  }\n}\nmodel {\n  mu ~ normal(0, 5);      // vectorized, diffuse\n  Omega ~ lkj_corr(2.0);  // regularize to unit correlation\n  sigma ~ cauchy(0, 5);   // half-Cauchy due to constraint\n  // ... words sampled as above ...\n}\nThe \\(\\textsf{LKJCorr}\\) distribution with shape \\(\\alpha &gt; 0\\) has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by \\[\n\\mathsf{LkjCorr}(\\Omega\\mid\\alpha) \\propto \\mathrm{det}(\\Omega)^{\\alpha - 1}\n\\] With a scale of \\(\\alpha = 2\\), the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix \\(\\Sigma\\) for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on sigma.",
+    "text": "Many of the tokens used in the BNF grammars follow obviously from their names: DATABLOCK is the literal string ‘data’, COMMA is a single ‘,’ character, etc. The literal representation of each operator is additionally provided in the operator precedence table.\nA few tokens are not so obvious, and are defined here in regular expressions:\nIDENTIFIER = [a-zA-Z] [a-zA-Z0-9_]*\n\nSTRINGLITERAL = \".*\"\n\nINTNUMERAL = [0-9]+ (_ [0-9]+)*\n\nEXPLITERAL = [eE] [+-]? INTNUMERAL\n\nREALNUMERAL = INTNUMERAL \\. INTNUMERAL? EXPLITERAL?\n            | \\. INTNUMERAL EXPLITERAL\n            | INTNUMERAL EXPLITERAL\n\nIMAGNUMERAL = (REALNUMERAL | INTNUMERAL) i\n\nDOTNUMERAL = \\. INTNUMERAL",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html#relation-to-finite-mixture-models",
-    "href": "stan-users-guide/clustering.html#relation-to-finite-mixture-models",
-    "title": "Clustering Models",
+    "objectID": "reference-manual/syntax.html#extra-grammatical-constraints",
+    "href": "reference-manual/syntax.html#extra-grammatical-constraints",
+    "title": "Language Syntax",
     "section": "",
-    "text": "As mentioned in the clustering section, clustering models and finite mixture models are really just two sides of the same coin. The “soft” \\(K\\)-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of \\(K\\)-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.",
+    "text": "A well-formed Stan program must satisfy the type constraints imposed by functions and distributions. For example, the binomial distribution requires an integer total count parameter and integer variate and when truncated would require integer truncation points. If these constraints are violated, the program will be rejected during compilation with an error message indicating the location of the problem.\n\n\n\nIn the Stan grammar provided in this chapter, the expression 1 + 2 * 3 has two parses. As described in the operator precedence table, Stan disambiguates between the meaning \\(1\n+ (2 \\times 3)\\) and the meaning \\((1 + 2) \\times 3\\) based on operator precedences and associativities.\n\n\n\nIn a compound variable declaration and definition, the type of the right-hand side expression must be assignable to the variable being declared. The assignability constraint restricts compound declarations and definitions to local variables and variables declared in the transformed data, transformed parameters, and generated quantities blocks.\n\n\n\nThe types of expressions used for elements in array expressions ('{' expressions '}') must all be of the same type or a mixture of scalar (int, real and complex) types (in which case the result is promoted to be of the highest type on the int -&gt; real -&gt; complex hierarchy).\n\n\n\nInteger literals longer than one digit may not start with 0 and real literals cannot consist of only a period or only an exponent.\n\n\n\nBoth the conditional if-then-else statement and while-loop statement require the expression denoting the condition to be a primitive type, integer or real.\n\n\n\nThe for loop statement requires that we specify in addition to the loop identifier, either a range consisting of two expressions denoting an integer, separated by ‘:’, or a single expression denoting a container. The loop variable will be of type integer in the former case and of the contained type in the latter case. Furthermore, the loop variable must not be in scope (i.e., there is no masking of variables).\n\n\n\nThe arguments to a print statement cannot be void.\n\n\n\nThe break and continue statements may only be used within the body of a for-loop or while-loop.\n\n\n\nSome constructs in the Stan language are only allowed in certain blocks or in certain kinds of user-defined functions.\n\n\nFunctions ending in _rng may only be called in the transformed data and generated quantities block, and within the bodies of user-defined functions with names ending in _rng.\n\n\n\nUnnormalized distributions (with suffixes _lupmf or _lupdf) may only be called in the model block, user-defined probability functions, or within the bodies of user defined functions which end in _lp.\n\n\n\ntarget += statements can only be used inside of the model block or user-defined functions which end in _lp.\nUser defined functions which end in _lp and the target() function can only be used in the model block, transformed parameters block, and in the bodies of other user defined functions which end in _lp.\nSampling statements (using ~) can only be used in the model block or in the bodies of user-defined functions which end in _lp.\njacobian += statements can only be used inside of the transformed parameters block or in functions that end with _jacobian.\n\n\n\n\nA probability function literal must have one of the following suffixes: _lpdf, _lpmf, _lcdf, or _lccdf.\n\n\n\nStandalone expressions used as indexes must denote either an integer (int) or an integer array (array[] int). Expressions participating in range indexes (e.g., a and b in a : b) must denote integers (int).\nA second condition is that there not be more indexes provided than dimensions of the underlying expression (in general) or variable (on the left side of assignments) being indexed. A vector or row vector adds 1 to the array dimension and a matrix adds 2. That is, the type array[ , , ] matrix, a three-dimensional array of matrices, has five index positions: three for the array, one for the row of the matrix and one for the column.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Language Syntax"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html#soft-k-means",
-    "href": "stan-users-guide/clustering.html#soft-k-means",
-    "title": "Clustering Models",
+    "objectID": "reference-manual/types.html",
+    "href": "reference-manual/types.html",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "\\(K\\)-means clustering is a method of clustering data represented as \\(D\\)-dimensional vectors. Specifically, there will be \\(N\\) items to be clustered, each represented as a vector \\(y_n \\in \\mathbb{R}^D\\). In the “soft” version of \\(K\\)-means, the assignments to clusters will be probabilistic.\n\n\n\\(K\\)-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters \\(K\\) and data vectors \\(y\\) as input.\n\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), randomly assign vector \\(y_n\\) to a cluster in \\(\\{1,\\dotsc,K\\}\\);\nRepeat\n\nFor each cluster \\(k\\) in \\(\\{1,\\dotsc,K\\}\\), compute the cluster centroid \\(\\mu_k\\) by averaging the vectors assigned to that cluster;\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), reassign \\(y_n\\) to the cluster \\(k\\) for which the (Euclidean) distance from \\(y_n\\) to \\(\\mu_k\\) is smallest;\nIf no vectors changed cluster, return the cluster assignments.\n\n\nThis algorithm is guaranteed to terminate.\n\n\n\nSoft \\(K\\)-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft \\(K\\)-means can be expressed (and coded in Stan) as a multivariate normal mixture model.\nIn the full generative model, each data point \\(n\\) in \\(\\{1,\\dotsc,N\\}\\) is assigned a cluster \\(z_n \\in \\{1,\\dotsc,K\\}\\) with symmetric uniform probability, \\[\nz_n \\sim \\textsf{categorical}(1/K),\n\\] where \\(1\\) is the unit vector of \\(K\\) dimensions, so that \\(1/K\\) is the symmetric \\(K\\)-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.\nThe data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment \\(z_n\\), \\[\ny_n \\sim  \\textsf{normal}(\\mu_{z[n]},\\Sigma_{z[n]})\n\\]\nThe sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters \\(k\\), \\[\n\\Sigma_k = \\mathrm{diag\\_matrix}({\\bf 1}),\n\\] so that the log multivariate normal can be implemented directly up to a proportion by \\[\n\\mathrm{normal}\\left( y_n \\mid \\mu_k, \\mathrm{diag\\_matrix}({\\bf 1}) \\right)\n\\propto \\exp \\left (- \\frac{1}{2} \\sum_{d=1}^D \\left( \\mu_{k,d} - y_{n,d}\n  \\right)^2 \\right).\n\\] The spatial perspective on \\(K\\)-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean \\(\\mu_k\\) to the data point \\(y_n\\).\n\n\n\nConsider the following Stan program for implementing \\(K\\)-means clustering.\ndata {\n  int&lt;lower=0&gt; N;        // number of data points\n  int&lt;lower=1&gt; D;        // number of dimensions\n  int&lt;lower=1&gt; K;        // number of clusters\n  array[N] vector[D] y;  // observations\n}\ntransformed data {\n  real&lt;upper=0&gt; neg_log_K;\n  neg_log_K = -log(K);\n}\nparameters {\n  array[K] vector[D] mu; // cluster means\n}\ntransformed parameters {\n  array[N, K] real&lt;upper=0&gt; soft_z; // log unnormalized clusters\n  for (n in 1:N) {\n    for (k in 1:K) {\n      soft_z[n, k] = neg_log_K\n                     - 0.5 * dot_self(mu[k] - y[n]);\n    }\n  }\n}\nmodel {\n  // prior\n  for (k in 1:K) {\n    mu[k] ~ std_normal();\n  }\n\n  // likelihood\n  for (n in 1:N) {\n    target += log_sum_exp(soft_z[n]);\n  }\n}\nThere is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.\nThe only parameter is mu, where mu[k] is the centroid for cluster \\(k\\). The transformed parameters soft_z[n] contain the log of the unnormalized cluster assignment probabilities. The vector soft_z[n] can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.\n\n\n\nThe multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., \\(L_2\\) distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on \\(L_1\\) distance (i.e., Manhattan or taxicab distance).\nWithin the multivariate normal version of \\(K\\)-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.\nAlthough there is no global spatial analog, it is common to see soft \\(K\\)-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.",
+    "text": "This chapter covers the data types for expressions in Stan. Every variable used in a Stan program must have a declared data type. Only values of that type will be assignable to the variable (except for temporary states of transformed data and transformed parameter values). This follows the convention of programming languages like C++, not the conventions of scripting languages like Python or statistical languages such as R or BUGS.\nThe motivation for strong, static typing is threefold.\n\nStrong typing forces the programmer’s intent to be declared with the variable, making programs easier to comprehend and hence easier to debug and maintain.\nStrong typing allows programming errors relative to the declared intent to be caught sooner (at compile time) rather than later (at run time). The Stan compiler (called through an interface such as CmdStan, RStan, or PyStan) will flag any type errors and indicate the offending expressions quickly when the program is compiled.\nConstrained types will catch runtime data, initialization, and intermediate value errors as soon as they occur rather than allowing them to propagate and potentially pollute final results.\n\nStrong typing disallows assigning the same variable to objects of different types at different points in the program or in different invocations of the program.\n\n\nArguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.\nPassing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.\n\n\nStan provides two primitive data types, real for continuous values and int for integer values. These are both considered scalar types.\n\n\n\nStan provides a complex number data type complex, where a complex number contains both a real and an imaginary component, both of which are of type real. Complex types are considered scalar types.\n\n\n\nStan provides three real-valued matrix data types, vector for column vectors, row_vector for row vectors, and matrix for matrices.\nStan also provides three complex-valued matrix data types, complex_vector for column vectors, complex_row_vector for row vectors, and complex_matrix for matrices.\n\n\n\nAny type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,\narray[10] real x;\narray[6, 7] matrix[3, 3] m;\narray[12, 8, 15] complex z;\ndeclares x to be a one-dimensional array of size 10 containing real values, declares m to be a two-dimensional array of size \\(6 \\times 7\\) containing values that are \\(3 \\times 3\\) matrices, and declares z to be a \\(12 \\times 8 \\times 15\\) array of complex numbers.\nPrior to 2.26 Stan models used a different syntax which has since been removed. See the Removed Features chapter for more details.\n\n\n\nFor any sequence of types, Stan provides a tuple data type. For example,\ntuple(real, array[5] int) xi;\ndeclares xi to be a tuple holding two values, the first of which is of type type real and the second of which a 5-dimensional array of type int.\n\n\n\nDeclarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.\nAll of the basic data types other than complex may be given lower and upper bounds using syntax such as\nint&lt;lower=1&gt; N;\nreal&lt;upper=0&gt; log_p;\nvector&lt;lower=-1, upper=1&gt;[3] rho;\nThere are also special data types for structured vectors and matrices. There are five constrained vector data types, simplex for unit simplexes, unit_vector for unit-length vectors, sum_to_zero_vector for vectors that sum to zero, ordered for ordered vectors of scalars, and positive_ordered for vectors of positive ordered scalars. There are specialized matrix data types corr_matrix and cov_matrix for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type cholesky_factor_cov is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type cholesky_factor_corr is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type sum_to_zero_matrix is for matrices that sum to zero across rows and columns.\nConstraints provide error checking for variables defined in the data, transformed data, transformed parameters, and generated quantities blocks. Constraints are critical for variables declared in the parameters block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of \\(\\mathbb{R}^n\\)).\nIt is worth calling out the most important aspect of constrained data types:\n\nThe model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.\n\nIf this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.\n\n\n\n\nUnfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.\n\n\nStan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is \\(2^{31}-1\\); the minimum value is \\(-(2^{31})\\).\nWhen integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.\nInteger arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see the Stan Functions Reference integer-valued arithmetic operators section for more information).\n\n\n\nStan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly1 follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly \\(\\pm 2^{1022}\\), which is slightly larger than \\(\\pm 10^{307}\\). It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.\n64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.\nThere are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.\n\n\nThe not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).\n\n\n\nPositive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.\n\n\n\n\nStan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.\nUnlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.\n\n\n\n\nThe complex data type is a scalar, but unlike real and int types, it contains two components, a real and imaginary component, both of which are of type real. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.\n\n\nImaginary literals are written in mathematical notation using a numeral followed by the suffix i. For example, the following example constructs a complex number \\(2 - 1.3i\\) and assigns it to the variable z.\ncomplex z = 2 - 1.3i;\nreal re = get_real(z);  // re has value 2.0\nreal im = get_imag(z);  // im has value -1.3\nThe getter functions then extract the real and imaginary components of z and assign them to re and im respectively.\nThe function to_complex constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.\nvector[K] re;\nvector[K] im;\n// ...\nfor (k in 1:K) {\n  complex z = to_complex(re[k], im[k]);\n  // ...\n}\n\n\n\nExpressions of type real may be assigned to variables of type complex. For example, the following is a valid sequence of Stan statements.\nreal x = 5.0;\ncomplex z = x;  // get_real(z) == 5.0, get_imag(z) == 0\nThe real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.\nAssignability is transitive, so that expressions of type int may also be assigned to variables of type complex, as in the following example.\nint n = 2;\ncomplex z = n;\nFunction arguments also support promotion of integer or real typed expressions to type complex.\n\n\n\n\nAll variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.\n\n\nUnconstrained integers are declared using the int keyword. For example, the variable N is declared to be an integer as follows.\nint N;\n\n\n\nInteger data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare N to be a positive integer, use the following.\nint&lt;lower=1&gt; N;\nThis illustrates that the bounds are inclusive for integers.\nTo declare an integer variable cond to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.\nint&lt;lower=0, upper=1&gt; cond;\n\n\n\nUnconstrained real variables are declared using the keyword real. The following example declares theta to be an unconstrained continuous value.\nreal theta;\n\n\n\nUnconstrained complex numbers are declared using the keyword complex. The following example declares z to be an unconstrained complex variable.\ncomplex z;\n\n\n\nReal variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.\nThe variable sigma may be declared to be non-negative as follows.\nreal&lt;lower=0&gt; sigma;\nThe following declares the variable x to be less than or equal to \\(-1\\).\nreal&lt;upper=-1&gt; x;\nTo ensure rho takes on values between \\(-1\\) and \\(1\\), use the following declaration.\nreal&lt;lower=-1, upper=1&gt; rho;\n\n\nLower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants positive_infinity() and negative_infinity() which may be used for this purpose, or they may be supplied as data.\n\n\n\n\nReal variables may be declared on a space that has been transformed using an affine transformation \\(x\\mapsto \\mu + \\sigma * x\\) with offset \\(\\mu\\) and (positive) multiplier \\(\\sigma\\), using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.\nThe variable x may be declared to have offset \\(1\\) as follows.\nreal&lt;offset=1&gt; x;\nSimilarly, it can be declared to have multiplier \\(2\\) as follows.\nreal&lt;multiplier=2&gt; x;\nFinally, we can combine both declarations to declare a variable with offset \\(1\\) and multiplier \\(2\\).\nreal&lt;offset=1, multiplier=2&gt; x;\nAs an example, we can give x a normal distribution with non-centered parameterization as follows.\nparameters {\n  real&lt;offset=mu, multiplier=sigma&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nRecall that the centered parameterization is achieved with the code\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nor equivalently\nparameters {\n  real&lt;offset=0, multiplier=1&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\n\n\n\nBounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.\nFor example, it is acceptable to have the following declarations.\ndata {\n real lb;\n}\nparameters {\n   real&lt;lower=lb&gt; phi;\n}\nThis declares a real-valued parameter phi to take values greater than the value of the real-valued data variable lb. Constraints may be arbitrary expressions, but must be of type int for integer variables and of type real for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,\ndata {\n   int&lt;lower=1&gt; N;\n   array[N] real y;\n}\nparameters {\n   real&lt;lower=min(y), upper=max(y)&gt; phi;\n}\nThis declares a positive integer data variable N, an array y of real-valued data of length N, and then a parameter ranging between the minimum and maximum value of y. As shown in the example code, the functions min() and max() may be applied to containers such as arrays.\nA more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;  // enforces a &lt; b\n}\ntransformed parameters {\n  real c;\n  real&lt;lower=c&gt; d;\n  c = a;\n  d = b;\n}\nThe parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though c isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.\n\n\n\nA variable may be declared with a size that depends on a boolean constant. For example, consider the definition of alpha in the following program fragment.\ndata {\n  int&lt;lower=0, upper=1&gt; include_alpha;\n  // ...\n}\nparameters {\n  vector[include_alpha ? N : 0] alpha;\n  // ...\n}\nIf include_alpha is true, the model will include the vector alpha; if the flag is false, the model will not include alpha (technically, it will include alpha of size 0, which means it won’t contain any values and won’t be included in any output).\nThis technique is not just useful for containers. If the value of N is set to 1, then the vector alpha will contain a single element and thus alpha[1] behaves like an optional scalar, the existence of which is controlled by include_alpha.\nThis coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the RStanArm package.\n\n\n\n\nStan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A \\(3 \\times 4\\) matrix is a different kind of object in Stan than a \\(3\n\\times 4\\) array.\nThe intention of using matrix types is to call out their usage in the code. There are three situations in Stan where only vectors and matrices may be used,\n\nmatrix arithmetic operations (e.g., matrix multiplication)\nlinear algebra functions (e.g., eigenvalues and determinants), and\nmultivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).\n\nVectors and matrices cannot be typed to return integer values. They are restricted to real and complex values.\nFor constructing vectors and matrices in Stan, see Vector, Matrix, and Array Expressions.\n\n\nVectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.\n\n\n\nVectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword vector, as follows.\nvector[3] u;\nVectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.\nvector&lt;lower=0&gt;[3] u;\nSimilarly, they may be declared with a offset and/or multiplier, as in the following example\nvector&lt;offset=42, multiplier=3&gt;[3] u;\n\n\n\nLike real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword complex_vector, as follows.\ncomplex_vector[3] v;\nComplex vector declarations do not support any constraints.\n\n\n\nA unit simplex is a vector with non-negative values whose entries sum to 1. For instance, \\([0.2,0.3,0.4,0.1]^{\\top}\\) is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-simplex by\nsimplex[5] theta;\nUnit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\nIn high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.\n\n\n\nA stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a \\(3 \\times 4\\) column-stochastic matrix.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.3 \\\\\n0.3 & 0.3 & 0.6 & 0.4 \\\\\n0.5 & 0.2 & 0.3 & 0.3\n\\end{bmatrix}\n\\]\nAn example of a \\(3 \\times 4\\) row-stochastic matrix is the following.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.2 \\\\\n0.2 & 0.1 & 0.6 & 0.1 \\\\\n0.5 & 0.2 & 0.2 & 0.1\n\\end{bmatrix}\n\\]\nIn the examples above, each column (or row) sums to 1, making the matrices valid column_stochastic_matrix and row_stochastic_matrix types.\nColumn-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.\nThe column_stochastic_matrix and row_stochastic_matrix types are declared with row and column sizes. For instance, a matrix theta with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.\ncolumn_stochastic_matrix[3, 4] theta;\nA matrix theta with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.\nrow_stochastic_matrix[3, 4] theta;\nAs with simplexes, column_stochastic_matrix and row_stochastic_matrix variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold \\(\\epsilon\\).\n\n\nIn high-dimensional settings, column_stochastic_matrix and row_stochastic_matrix types may require careful tuning of the inference algorithms. To ensure stability:\n\nSmaller Step Sizes: In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.\nHigher Target Acceptance Rates: Setting higher target acceptance rates can improve the robustness of the sampling process.\nLonger Warmup Periods: Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.\nTighter Optimization Tolerances: For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.\nCustom Initialization: If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.\n\n\n\n\n\nA unit vector is a vector with a norm of one. For instance, \\([0.5,\n0.5, 0.5, 0.5]^{\\top}\\) is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-vector by\nunit_vector[5] theta;\nUnit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nA zero-sum vector is constrained such that the sum of its elements is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\(N - 1\\) degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, beta is declared to be a zero-sum \\(5\\)-vector (4 DoF) by\nsum_to_zero_vector[5] beta;\nZero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nAn ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, \\((-1.3,2.7,2.71)^{\\top}\\) is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see section).\nThe variable c is declared as an ordered 5-vector by\nordered[5] c;\nAfter their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nThere is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, \\((2,3.7,4,12.9)\\) is a positive, ordered 4-vector.\nThe variable d is declared as a positive, ordered 5-vector by\npositive_ordered[5] d;\nLike ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nRow vectors are declared with the keyword row_vector. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector u would be declared as\nrow_vector[1093] u;\nConstraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.\nrow_vector&lt;lower=-1, upper=1&gt;[10] u;\nOffset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.\nrow_vector&lt;offset=-42, multiplier=3&gt;[3] u;\nRow vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.\n\n\n\nComplex row vectors are declared with the keyword complex_row_vector and given a size in basic declarations. For example, a 12-dimensional complex row vector v would be declared as\ncomplex_row_vector[12] v;\nComplex row vectors do not allow constraints.\n\n\n\nMatrices are declared with the keyword matrix along with a number of rows and number of columns. For example,\nmatrix[3, 3] A;\nmatrix[M, N] B;\ndeclares A to be a \\(3 \\times 3\\) matrix and B to be a \\(M\n\\times N\\) matrix. For the second declaration to be well formed, the variables M and N must be declared as integers in either the data or transformed data block and before the matrix declaration.\nMatrices may also be declared with constraints, as in this (\\(3 \\times 4\\)) matrix of non-positive values.\nmatrix&lt;upper=0&gt;[3, 4] B;\nSimilarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.\nmatrix&lt;multiplier=5&gt;[3, 4] B;\n\n\nRows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.\nmatrix[M, N] a;\nrow_vector[N] b;\n// ...\na[1] = b;\nThis copies the values from row vector b to a[1], which is the first row of the matrix a. If the number of columns in a is not the same as the size of b, a run-time error is raised; the number of columns of a is N, which is also the number of columns of b.\nAssignment works by copying values in Stan. That means any subsequent assignment to a[1] does not affect b, nor does an assignment to b affect a.\n\n\n\n\nComplex matrices are declared with the keyword complex_matrix and a number of rows and columns. For example,\ncomplex_matrix[3, 3] C;\nComplex matrices do not allow constraints.\n\n\n\nMatrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,\ncov_matrix[K] Sigma;\ndeclares Sigma to be a \\(K \\times K\\) covariance matrix, where \\(K\\) is the value of the data variable K.\n\n\n\nMatrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between \\(-1\\) and \\(1\\), and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,\ncorr_matrix[3] Omega;\ndeclares Omega to be a \\(3 \\times 3\\) correlation matrix.\nCorrelation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.\nA Cholesky factor \\(L\\) is an \\(M \\times N\\) lower-triangular matrix (if \\(m &lt; n\\) then \\(L[m, n] =0\\)) with a strictly positive diagonal (\\(L[k, k]\n&gt; 0\\)) and \\(M \\geq N\\). If \\(L\\) is a Cholesky factor, then \\(\\Sigma = L\n\\, L^{\\top}\\) is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.\nThe typical case of a square Cholesky factor may be declared with a single dimension,\ncholesky_factor_cov[4] L;\n\n\nIn general, two dimensions may be declared, with the above being equal to cholesky_factor_cov[4, 4]. The type cholesky_factor_cov[M, N] may be used for the general \\(M \\times N\\) case to produce positive semi-definite matrices of rank \\(M\\).\n\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a correlation matrix.\nA Cholesky factor for a correlation matrix \\(L\\) is a \\(K \\times K\\) lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., \\(\\sum_{n=1}^K L_{m,n}^2 = 1\\)). If \\(L\\) is a Cholesky factor for a correlation matrix, then \\(L\\,L^{\\top}\\) is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).\nTo declare the variable L to be a K by K Cholesky factor of a correlation matrix, the following code may be used.\ncholesky_factor_corr[K] L;\n\n\n\nA sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\((N - 1) \\times (M - 1)\\) degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, beta is declared to be a sum-to-zero \\(5 \\times 4\\)-matrix (20 degrees of freedom) by\nsum_to_zero_matrix[5, 4] beta;\nSum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nConstrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be real&lt;lower=0, upper=1&gt; could be assigned to a variable declared as real and vice-versa. Similarly, a variable declared as matrix[3, 3] may be assigned to a variable declared as cov_matrix[3] or cholesky_factor_cov[3], and vice-versa.\nChecks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.\n\n\n\nReal-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.\nvector[N] v = ...;\ncomplex_vector[N] u = 2 * v;\nRow vectors and matrices work the same way.\n\n\n\nVariables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.\ndata {\n  int&lt;lower=0&gt; N_observed, N_missing;\n  // ...\ntransformed parameters {\n  vector[N_observed + N_missing] y;\n  // ...\n\n\n\nIf v is a column vector or row vector, then v[2] is the second element in the vector. If m is a matrix, then m[2, 3] is the value in the second row and third column.\nProviding a matrix with a single index returns the specified row. For instance, if m is a matrix, then m[2] is the second row. This allows Stan blocks such as\nmatrix[M, N] m;\nrow_vector[N] v;\nreal x;\n// ...\nv = m[2];\nx = v[3];   // x == m[2][3] == m[2, 3]\nThe type of m[2] is row_vector because it is the second row of m. Thus it is possible to write m[2][3] instead of m[2, 3] to access the third element in the second row. When given a choice, the form m[2, 3] is preferred.\nComplex versions work the same way,\ncomplex_matrix[M, N] m = ...;\ncomplex_row_vector[N] u = m[3];\ncomplex_vector[M] v = m[ , 2];\n\n\nThe form m[2, 3] is more efficient because it does not require the creation and use of an intermediate expression template for m[2]. In later versions, explicit calls to m[2][3] may be optimized to be as efficient as m[2, 3] by the Stan compiler.\n\n\n\n\nAn integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size M + N using\nvector[M + N] y;\nAny integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.\n\n\n\n\nStan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.\nA two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if a is a two-dimensional array, then a[m, n] is just a convenient shorthand for a[m][n].\nVectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.\nFor constructing arrays in Stan, see Vector, Matrix, and Array Expressions.\n\n\nArrays are declared with the keyword array followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.\nThe variable n is declared as an array of five integers as follows.\narray[5] int n;\nA two-dimensional array of complex values with three rows and four columns is declared as follows.\narray[3, 4] complex a;\nA three-dimensional array z of positive reals with five rows, four columns, and two shelves can be declared as follows.\narray[5, 4, 2] real&lt;lower=0&gt; z;\nArrays may also be declared to contain vectors. For example,\narray[3] vector[7] mu;\ndeclares mu to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example\narray[15, 12] complex_matrix[7, 2] mu;\ndeclares a 15 by 12 array of \\(7 \\times 2\\) complex matrices. Any of the constrained types may also be used in arrays, as in the declaration\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nof a \\(2 \\times 3 \\times 4\\) array of \\(5 \\times 6\\) Cholesky factors of covariance matrices.\n\n\n\nIf x is a 1-dimensional array of length 5, then x[1] is the first element in the array and x[5] is the last. For a \\(3\n\\times 4\\) array y of two dimensions, y[1, 1] is the first element and y[3, 4] the last element. For a three-dimensional array z, the first element is z[1, 1, 1], and so on.\nSubarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose y is a two-dimensional array with three rows and four columns. Then y[3] is one-dimensional array of length four. This means that y[3][1] may be used instead of y[3, 1] to access the value of the first column of the third row of y. The form y[3, 1] is the preferred form (see note in this chapter).\n\n\n\nSubarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as\narray[9, 10, 11] real w;\narray[10, 11] real x;\narray[11] real y;\nreal z;\n// ...\nx = w[5];\ny = x[4];  // y == w[5][4] == w[5, 4]\nz = y[3];  // z == w[5][4][3] == w[5, 4, 3]\nComplex-valued arrays work the same way.\n\n\n\nArrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.\narray[3, 4] vector[5] a;\narray[4] vector[5] b;\nvector[5] c;\nreal x;\nWith these declarations, the following assignments are legal.\nb = a[1];       // result is array of vectors\nc = a[1, 3];    // result is vector\nc = b[3];       //   same result as above\nx = a[1, 3, 5]; // result is scalar\nx = b[3, 5];    //   same result as above\nx = c[5];       //   same result as above\nRow vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.\nConsider the following matrix, vector and scalar declarations.\narray[3, 4] matrix[6, 5] d;\narray[4] matrix[6, 5] e;\nmatrix[6, 5] f;\nrow_vector[5] g;\nreal x;\nWith these declarations, the following definitions are legal.\ne = d[1];           // result is array of matrices\nf = d[1, 3];        // result is matrix\nf = e[3];           //   same result as above\ng = d[1, 3, 2];     // result is row vector\ng = e[3, 2];        //   same result as above\ng = f[2];           //   same result as above\nx = d[1, 3, 5, 2];  // result is scalar\nx = e[3, 5, 2];     //   same result as above\nx = f[5, 2];        //   same result as above\nx = g[2];           //   same result as above\nAs shown, the result f[2] of supplying a single index to a matrix is the indexed row, here row 2 of matrix f.\n\n\n\nSubarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.\narray[I, J, K] real x;\narray[J, K] real y;\narray[K] real z;\n// ...\nx[1] = y;\nx[1, 1] = z;\nThe sizes must match. Here, x[1] is a J by K array, as is y.\nPartial array assignment also works for arrays of matrices, vectors, and row vectors.\n\n\n\nArrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).\n\n\nFor example, vectors cannot be assigned to arrays or vice-versa.\narray[4] real a;\nvector[4] b;\nrow_vector[4] c;\n// ...\na = b; // illegal assignment of vector to array\nb = a; // illegal assignment of array to vector\na = c; // illegal assignment of row vector to array\nc = a; // illegal assignment of array to row vector\n\n\n\nIt is not even legal to assign row vectors to column vectors or vice versa.\nvector[4] b;\nrow_vector[4] c;\n// ...\nb = c; // illegal assignment of row vector to column vector\nc = b; // illegal assignment of column vector to row vector\n\n\n\nThe same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.\narray[3, 4] real a;\nmatrix[3, 4] b;\n// ...\na = b;  // illegal assignment of matrix to array\nb = a;  // illegal assignment of array to matrix\n\n\n\nA \\(1 \\times N\\) matrix cannot be assigned a row vector or vice versa.\nmatrix[1, 4] a;\nrow_vector[4] b;\n// ...\na = b;  // illegal assignment of row vector to matrix\nb = a;  // illegal assignment of matrix to row vector\nSimilarly, an \\(M \\times 1\\) matrix may not be assigned to a column vector.\nmatrix[4, 1] a;\nvector[4] b;\n// ...\na = b;  // illegal assignment of column vector to matrix\nb = a;  // illegal assignment of matrix to column vector\n\n\n\n\nAn integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.\n\n\n\nIf any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare\narray[3, 0] real a;\nthen the resulting size of a is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, a[1] will be a size-zero one-dimensional array. For comparison, declaring\narray[0, 3] real b;\nalso produces an array with an overall size of zero, but in this case, there is no way to index legally into b, because b[0] is undefined. The array will behave at run time as if it’s a \\(0 \\times\n0\\) array. For example, the result of to_matrix(b) will be a \\(0 \\times 0\\) matrix, not a \\(0 \\times 3\\) matrix.\n\n\n\n\nStan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.\n\n\nTuples are declared with the keyword tuple followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as\ntuple(int, vector[3], complex) abc;\nTuples must have at least one entry, so the following declaration is illegal.\ntuple() nil;  // ILLEGAL\nTuples of length one must use a trailing comma, to align with the expression syntax.\ntuple(int,) m; // CORRECT\ntuple(int) n;  // ILLEGAL\nTuples can be assigned as a whole if their elements can be assigned individually. For example, a can be assigned to b in the following example because int can be promoted to complex.\ntuple(int, real) a;\n...\ntuple(complex, real) b = a;\nTuple types may have elements which are declared as tuples, such as the following example.\ntuple(int, tuple(real, complex)) x;\nIn this case, it would probably be simpler to use a 3-tuple type, tuple(int, real, complex).\nTuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration\nreal&lt;lower=0&gt; sigma;\nreal&lt;lower=0, upper=1&gt; theta;\nit is legal to have a tuple with constraints such as\ntuple(real&lt;lower=0&gt;, real&lt;lower=0, upper=1&gt;) sigma_theta;\n\n\n\nTuple elements may be accessed directly. For example, with our declaration of abc from the last section, Stan uses abc.1 for the first element, abc.2 for the second, and abc.3 for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that abc.1 is of type int, abc.2 of type vector[3] and abc.3 of type complex.\n\n\n\nTuple elements can be assigned individually, allowing, e.g.,\ntuple(int, real) ab;\nab.1 = 123;\nab.2 = 12.9;\nAs with other assignments, promotions will happen if necessary (of int to real and of real to complex, along with the corresponding container type promotions).\n\n\n\nFor convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.\nGiven a tuple t of type tuple(T1, ..., Tn) and a sequence of assignable expressions of types v1, …, vn, where each vi has a type which is assignable from type Ti, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement\n(v1, /*...*/, vn) = t;\nNote that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).\nThese unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.\nFor example, if T is a tuple of type tuple(int, (real, real), complex), then the program\nint i;\nreal x, y;\ncomplex z;\n\n(i, (x, y), z) = T;\nAssigns the result of T.1 to i, the result of T.2.1 to x, the result of T.2.2 to y, and the result of T.3 to z.\nThe left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.\n\n\n\n\nThe type information associated with a variable only contains the underlying type and dimensionality of the variable.\n\n\nThe size associated with a given variable is not part of its data type. For example, declaring a variable using\narray[3] real a;\ndeclares the variable a to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.\n\n\nSizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an \\(N \\times M\\) by a \\(J \\times K\\) matrix will raise a run-time error if \\(M \\neq J\\).\n\n\n\n\nLike sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.\nSimilarly, a bounded real may be assigned to an unconstrained real and vice-versa.\n\n\nFor arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.\n\n\n\nFor data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.\nFor parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.\nFor all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to NaN and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).\nFor generated quantities, constraints are enforced after the statements in the generated quantities block have executed.\n\n\n\n\nIn order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.\nBecause size information is not part of a data type, data types will be written without size information. For instance, array[] real is the type of one-dimensional array of reals and matrix is the type of matrices. The three-dimensional integer array type is written as array[,,] int, indicating the number slots available for indexing. Similarly, array[,] vector is the type of a two-dimensional array of vectors.\n\n\n\n\nVariables in Stan are declared by giving a type and a name. For example\nint N;\nvector[N] y;\narray[5] matrix[3, 4] A;\ndeclares a variable N that is an integer, a variable y that is a vector of length N (the previously declared variable), and a variable A, which is a length-5 array where each element is a 3 by 4 matrix.\nThe size of top-level variables in the parameters, transformed parameters, and generated quantities must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.\n// illegal and will be flagged by the compiler:\ngenerated quantities {\n  int N = 10;\n  array[N] int foo;\nDepending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.\n// valid block variables, but not locals or function parameters\nvector&lt;lower=0&gt;[N] u;\n\n// valid as a block or local variable, but not a function parameter\narray[3] int is;\n\n// function parameters exclude sizes and cannot be constrained\nvoid pretty_print_tri_lower(matrix x) { ... }\nTop-level variables can have constraints and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as function parameters are not constrained types and exclude sizes.\nIn the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables M and N indicate number of columns and rows, respectively. The variable K is used for square matrices, i.e., K denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types cov_matrix[K], corr_matrix[K], cholesky_factor_cov[K], and cholesky_factor_corr[K] are only assignable to matrices of dimensions matrix[K, K] types.\n\n\n\n\n\n\n\n\nFunction Argument (unsized)\n   Local\n(unconstrained)\n              Block\n          (constrained)\n\n\n\n\nint\nint\nint\n\n\n\n\nint&lt;lower=L&gt;\n\n\n\n\nint&lt;upper=U&gt;\n\n\n\n\nint&lt;lower=L, upper=U&gt;\n\n\n\n\nint&lt;offset=O&gt;\n\n\n\n\nint&lt;multiplier=M&gt;\n\n\n\n\nint&lt;offset=O, multiplier=M&gt;\n\n\nreal\nreal\nreal\n\n\n\n\nreal&lt;lower=L&gt;\n\n\n\n\nreal&lt;upper=U&gt;\n\n\n\n\nreal&lt;lower=L, upper=U&gt;\n\n\n\n\nreal&lt;offset=O&gt;\n\n\n\n\nreal&lt;multiplier=M&gt;\n\n\n\n\nreal&lt;offset=O, multiplier=M&gt;\n\n\ncomplex\ncomplex\ncomplex\n\n\nvector\nvector[N]\nvector[N]\n\n\n\n\nvector[N]&lt;lower=L&gt;\n\n\n\n\nvector[N]&lt;upper=U&gt;\n\n\n\n\nvector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nvector[N]&lt;offset=O&gt;\n\n\n\n\nvector[N]&lt;multiplier=M&gt;\n\n\n\n\nvector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\nordered[N]\n\n\n\n\npositive_ordered[N]\n\n\n\n\nsimplex[N]\n\n\n\n\nunit_vector[N]\n\n\n\n\nsum_to_zero_vector[N]\n\n\nrow_vector\nrow_vector[N]\nrow_vector[N]\n\n\n\n\nrow_vector[N]&lt;lower=L&gt;\n\n\n\n\nrow_vector[N]&lt;upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O&gt;\n\n\n\n\nrow_vector[N]&lt;multiplier=M&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O, multiplier=M&gt;\n\n\nmatrix\nmatrix[M, N]\nmatrix[M, N]\n\n\n\n\nmatrix[M, N]&lt;lower=L&gt;\n\n\n\n\nmatrix[M, N]&lt;upper=U&gt;\n\n\n\n\nmatrix[M, N]&lt;lower=L, upper=U&gt; |\n\n\n\n\nmatrix[M, N]&lt;offset=O&gt;\n\n\n\n\nmatrix[M, N]&lt;multiplier=M&gt;\n\n\n\n\nmatrix[M, N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\ncolumn_stochastic_matrix[M, N]\n\n\n\n\nrow_stochastic_matrix[M, N]\n\n\n\n\nsum_to_zero_matrix[M, N]\n\n\n\nmatrix[K, K]\ncorr_matrix[K]\n\n\n\nmatrix[K, K]\ncov_matrix[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_corr[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_cov[K]\n\n\ncomplex_vector\ncomplex_vector[M]\ncomplex_vector[M]\n\n\ncomplex_row_vector\ncomplex_row_vector[N]\ncomplex_row_vector[N]\n\n\ncomplex_matrix\ncomplex_matrix[M, N]\ncomplex_matrix[M,N]\n\n\narray[] vector\narray[M] vector[N]\narray[M] vector[N]\n\n\n\n\narray[M] vector[N]&lt;lower=L&gt;\n\n\n\n\narray[M] vector[N]&lt;upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O&gt;\n\n\n\n\narray[M] vector[N]&lt;multiplier=M&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\narray[M] ordered[N]\n\n\n\n\narray[M] positive_ordered[N]\n\n\n\n\narray[M] simplex[N]\n\n\n\n\narray[M] unit_vector[N]\n\n\n\n\narray[M] sum_to_zero_vector[N]\n\n\n\n\nAdditional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g. array[ , ] is a 2-D array.\nFor more on how function arguments and return types are declared, consult the User’s Guide chapter on functions.\n\n\n\nStan allows assignable variables to be declared and defined in a single statement. Assignable variables are\n\nlocal variables, and\nvariables declared in the transformed data, transformed parameters, or generated quantities blocks.\n\nFor example, the statement\nint N = 5;\ndeclares the variable N to be an integer scalar type and at the same time defines it to be the value of the expression 5.\n\n\nThe type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have\nreal sum = 0;\neven though 0 is of type int and sum is of type real, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.\nVariables of any type may have values assigned to them. For example,\nmatrix[3, 2] a = b;\ndeclares a \\(3 \\times 2\\) matrix variable a and assigns a copy of the value of b to the variable a. The variable b must be of type matrix for the statement to be well formed. For the code to execute successfully, b must be the same shape as a, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to a do not affect b and subsequent changes to b do not affect a.\n\n\n\nThe right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,\nmatrix[3, 2] a = 0.5 * (b + c);\nassigns the matrix variable a to half of the sum of b and c. The only requirement on b and c is that the expression b + c be of type matrix. For example, b could be of type matrix and c of type real, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.\nSimilarly,\ncomplex z = 2 + 3i;\nassigns the the complex number \\(2 + 3i\\) to the complex scalar z. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).\n\n\n\nAny variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.\n\n\n\n\nStan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.\n\n\nThe code:\nreal x, y;\nis equivalent to\nreal x;\nreal y;\nAs a result, all declarations on the same line must be of the same type.\n\n\n\nThe ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in the section introducing compound declarations and definitions :\nreal x = 3, y = 5.6;\nConstrained data types can also be declared together, so long as the constraint for each variable is the same:\nreal&lt;lower=0&gt; x, y;",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html#the-difficulty-of-bayesian-inference-for-clustering",
-    "href": "stan-users-guide/clustering.html#the-difficulty-of-bayesian-inference-for-clustering",
-    "title": "Clustering Models",
+    "objectID": "reference-manual/types.html#overview-of-data-types",
+    "href": "reference-manual/types.html#overview-of-data-types",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "Two problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the label switching section.\n\n\nCluster assignments are not identified—permuting the cluster mean vectors mu leads to a model with identical likelihoods. For instance, permuting the first two indexes in mu and the first two indexes in each soft_z[n] leads to an identical likelihood (and prior).\nThe lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft \\(K\\)-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.\n\n\n\nThe other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.\nBayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.\nIn light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.",
+    "text": "Arguments for built-in and user-defined functions and local variables are required to be basic data types, meaning an unconstrained scalar, vector, or matrix type, or an array of such.\nPassing arguments to functions in Stan works just like assignment to basic types. Stan functions are only specified for the basic data types of their arguments, including array dimensionality, but not for sizes or constraints. Of course, functions often check constraints as part of their behavior.\n\n\nStan provides two primitive data types, real for continuous values and int for integer values. These are both considered scalar types.\n\n\n\nStan provides a complex number data type complex, where a complex number contains both a real and an imaginary component, both of which are of type real. Complex types are considered scalar types.\n\n\n\nStan provides three real-valued matrix data types, vector for column vectors, row_vector for row vectors, and matrix for matrices.\nStan also provides three complex-valued matrix data types, complex_vector for column vectors, complex_row_vector for row vectors, and complex_matrix for matrices.\n\n\n\nAny type (including the constrained types discussed in the next section) can be made into an array type by declaring array arguments. For example,\narray[10] real x;\narray[6, 7] matrix[3, 3] m;\narray[12, 8, 15] complex z;\ndeclares x to be a one-dimensional array of size 10 containing real values, declares m to be a two-dimensional array of size \\(6 \\times 7\\) containing values that are \\(3 \\times 3\\) matrices, and declares z to be a \\(12 \\times 8 \\times 15\\) array of complex numbers.\nPrior to 2.26 Stan models used a different syntax which has since been removed. See the Removed Features chapter for more details.\n\n\n\nFor any sequence of types, Stan provides a tuple data type. For example,\ntuple(real, array[5] int) xi;\ndeclares xi to be a tuple holding two values, the first of which is of type type real and the second of which a 5-dimensional array of type int.\n\n\n\nDeclarations of variables other than local variables may be provided with constraints. These constraints are not part of the underlying data type for a variable, but determine error checking in the transformed data, transformed parameter, and generated quantities block, and the transform from unconstrained to constrained space in the parameters block.\nAll of the basic data types other than complex may be given lower and upper bounds using syntax such as\nint&lt;lower=1&gt; N;\nreal&lt;upper=0&gt; log_p;\nvector&lt;lower=-1, upper=1&gt;[3] rho;\nThere are also special data types for structured vectors and matrices. There are five constrained vector data types, simplex for unit simplexes, unit_vector for unit-length vectors, sum_to_zero_vector for vectors that sum to zero, ordered for ordered vectors of scalars, and positive_ordered for vectors of positive ordered scalars. There are specialized matrix data types corr_matrix and cov_matrix for correlation matrices (symmetric, positive definite, unit diagonal) and covariance matrices (symmetric, positive definite). The type cholesky_factor_cov is for Cholesky factors of covariance matrices (lower triangular, positive diagonal, product with own transpose is a covariance matrix). The type cholesky_factor_corr is for Cholesky factors of correlation matrices (lower triangular, positive diagonal, unit-length rows). The type sum_to_zero_matrix is for matrices that sum to zero across rows and columns.\nConstraints provide error checking for variables defined in the data, transformed data, transformed parameters, and generated quantities blocks. Constraints are critical for variables declared in the parameters block, where they determine the transformation from constrained variables (those satisfying the declared constraint) to unconstrained variables (those ranging over all of \\(\\mathbb{R}^n\\)).\nIt is worth calling out the most important aspect of constrained data types:\n\nThe model must have support (non-zero density, equivalently finite log density) at parameter values that satisfy the declared constraints.\n\nIf this condition is violated with parameter values that satisfy declared constraints but do not have finite log density, then the samplers and optimizers may have any of a number of pathologies including just getting stuck, failure to initialize, excessive Metropolis rejection, or biased draws due to inability to explore the tails of the distribution.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html#naive-bayes-classification-and-clustering",
-    "href": "stan-users-guide/clustering.html#naive-bayes-classification-and-clustering",
-    "title": "Clustering Models",
+    "objectID": "reference-manual/types.html#numerical-data-types",
+    "href": "reference-manual/types.html#numerical-data-types",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "Naive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.1\nMultinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.\nNaive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.\nThe observed data consists of a sequence of \\(M\\) documents made up of bags of words drawn from a vocabulary of \\(V\\) distinct words. A document \\(m\\) has \\(N_m\\) words, which are indexed as \\(w_{m,1}, \\dotsc,\nw_{m,N[m]} \\in \\{1,\\dotsc,V\\}\\). Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) \\(K\\) is fixed.\nThe multinomial mixture model generates a single category \\(z_m \\in\n\\{1,\\dotsc,K\\}\\) for each document \\(m \\in \\{1,\\dotsc,M\\}\\) according to a categorical distribution, \\[\nz_m \\sim \\textsf{categorical}(\\theta).\n\\] The \\(K\\)-simplex parameter \\(\\theta\\) represents the prevalence of each category in the data.\nNext, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word \\(n\\) of document \\(m\\) being generated as \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m]}).\n\\] The parameter \\(\\phi_{z[m]}\\) is a \\(V\\)-simplex representing the probability of each word in the vocabulary in documents of category \\(z_m\\).\nThe parameters \\(\\theta\\) and \\(\\phi\\) are typically given symmetric Dirichlet priors. The prevalence \\(\\theta\\) is sometimes fixed to produce equal probabilities for each category \\(k \\in \\{1,\\dotsc,K\\}\\).\n\n\nThe specification for naive Bayes in the previous sections have used a ragged array notation for the words \\(w\\). Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.\n\\[\n\\begin{array}{lll}\n\\hline\n\\mathrm{n} \\qquad\\qquad\\qquad\\qquad & \\mathrm{w[n]} \\qquad & \\mathrm{doc[n]} \\\\\n\\hline\n1         & w_{1,1}    & 1 \\\\\n2         & w_{1,2}    & 1 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1     & w_{1,N[1]} & 1 \\\\\nN_1 + 1 & w_{2,1}    & 2 \\\\\nN_1 + 2 & w_{2,2}    & 2 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1 + N_2     & w_{2,N[2]} & 2 \\\\\nN_1 + N_2 + 1 & w_{3,1}    & 3 \\\\\n\\vdots       & \\vdots    & \\vdots \\\\\nN = \\sum_{m=1}^M N_m & w_{M,N[M]} & M \\\\\n\\hline\n\\end{array}\n\\]\nThe relevant variables for the program are N, the total number of words in all the documents, the word array w, and the document identity array doc.\n\n\n\nA naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows\ndata {\n  // training data\n  int&lt;lower=1&gt; K;               // num topics\n  int&lt;lower=1&gt; V;               // num words\n  int&lt;lower=0&gt; M;               // num docs\n  int&lt;lower=0&gt; N;               // total word instances\n  array[M] int&lt;lower=1, upper=K&gt; z;    // topic for doc m\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  // hyperparameters\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  simplex[K] theta;             // topic prevalence\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    z[m] ~ categorical(theta);\n  }\n  for (n in 1:N) {\n    w[n] ~ categorical(phi[z[doc[n]]]);\n  }\n}\nThe topic identifiers \\(z_m\\) are declared as data and the latent category assignments are included as part of the likelihood function.\n\n\n\nNaive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number \\(K\\) of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels z. Because z is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability\n\\[\\begin{align*}\n\\log\\, &p(w_{m,1},\\dotsc,w_{m,N_m} \\mid \\theta,\\phi) \\\\\n&= \\log \\sum_{k=1}^K\n    \\left( \\textsf{categorical}(k \\mid \\theta)\n           \\times \\prod_{n=1}^{N_m} \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right) \\\\\n&= \\log \\sum_{k=1}^K \\exp \\left(\n    \\log \\textsf{categorical}(k \\mid \\theta)\n     + \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right).\n\\end{align*}\\]\nThe last step shows how the log_sum_exp function can be used to stabilize the numerical calculation and return a result on the log scale.\nmodel {\n  array[M, K] real gamma;\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    for (k in 1:K) {\n      gamma[m, k] = categorical_lpmf(k | theta);\n    }\n  }\n  for (n in 1:N) {\n    for (k in 1:K) {\n      gamma[doc[n], k] = gamma[doc[n], k]\n                         + categorical_lpmf(w[n] | phi[k]);\n    }\n  }\n  for (m in 1:M) {\n    target += log_sum_exp(gamma[m]);\n  }\n}\nThe local variable gamma[m, k] represents the value \\[\n\\gamma_{m,k} = \\log \\textsf{categorical}(k \\mid \\theta)\n+ \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k).\n\\]\nGiven \\(\\gamma\\), the posterior probability that document \\(m\\) is assigned category \\(k\\) is \\[\n\\Pr[z_m = k \\mid w,\\alpha,\\beta]\n=\n\\exp \\left(\n\\gamma_{m,k}\n- \\log \\sum_{k=1}^K \\exp \\left( \\gamma_{m,k} \\right)\n\\right).\n\\]\nIf the variable gamma were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.\n\n\n\nFull Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see Gelman et al. (2013) and Gelman and Hill (2007) for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.\nTo specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables M2 N2, w2, and doc2. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.\n\n\n\nAn alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of gamma for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.",
+    "text": "Unfortunately, the lovely mathematical abstraction of integers and real numbers is only partially supported by finite-precision computer arithmetic.\n\n\nStan uses 32-bit (4-byte) integers for all of its integer representations. The maximum value that can be represented as an integer is \\(2^{31}-1\\); the minimum value is \\(-(2^{31})\\).\nWhen integers overflow, their value is determined by the underlying architecture. On most, their values wrap, but this cannot be guaranteed. Thus it is up to the Stan programmer to make sure the integer values in their programs stay in range. In particular, every intermediate expression must have an integer value that is in range.\nInteger arithmetic works in the expected way for addition, subtraction, and multiplication, but truncates the result of division (see the Stan Functions Reference integer-valued arithmetic operators section for more information).\n\n\n\nStan uses 64-bit (8-byte) floating point representations of real numbers. Stan roughly1 follows the IEEE 754 standard for floating-point computation. The range of a 64-bit number is roughly \\(\\pm 2^{1022}\\), which is slightly larger than \\(\\pm 10^{307}\\). It is a good idea to stay well away from such extreme values in Stan models as they are prone to cause overflow.\n64-bit floating point representations have roughly 15 decimal digits of accuracy. But when they are combined, the result often has less accuracy. In some cases, the difference in accuracy between two operands and their result is large.\nThere are three special real values used to represent (1) not-a-number value for error conditions, (2) positive infinity for overflow, and (3) negative infinity for overflow. The behavior of these special numbers follows standard IEEE 754 behavior.\n\n\nThe not-a-number value propagates. If an argument to a real-valued function is not-a-number, it either rejects (an exception in the underlying C++) or returns not-a-number itself. For boolean-valued comparison operators, if one of the arguments is not-a-number, the return value is always zero (i.e., false).\n\n\n\nPositive infinity is greater than all numbers other than itself and not-a-number; negative infinity is similarly smaller. Adding an infinite value to a finite value returns the infinite value. Dividing a finite number by an infinite value returns zero; dividing an infinite number by a finite number returns the infinite number of appropriate sign. Dividing a finite number by zero returns positive infinity. Dividing two infinite numbers produces a not-a-number value as does subtracting two infinite numbers. Some functions are sensitive to infinite values; for example, the exponential function returns zero if given negative infinity and positive infinity if given positive infinity. Often the gradients will break down when values are infinite, making these boundary conditions less useful than they may appear at first.\n\n\n\n\nStan automatically promotes integer values to real values if necessary, but does not automatically demote real values to integers. For very large integers, this will cause a rounding error to fewer significant digits in the floating point representation than in the integer representation.\nUnlike in C++, real values are never demoted to integers. Therefore, real values may only be assigned to real variables. Integer values may be assigned to either integer variables or real variables. Internally, the integer representation is cast to a floating-point representation. This operation is not without overhead and should thus be avoided where possible.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html#latent-dirichlet-allocation",
-    "href": "stan-users-guide/clustering.html#latent-dirichlet-allocation",
-    "title": "Clustering Models",
+    "objectID": "reference-manual/types.html#complex-numerical-data-type",
+    "href": "reference-manual/types.html#complex-numerical-data-type",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "Latent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model (Blei, Ng, and Jordan 2003) that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.\n\n\nThe basic model assumes each document is generated independently based on fixed hyperparameters. For document \\(m\\), the first step is to draw a topic distribution simplex \\(\\theta_m\\) over the \\(K\\) topics, \\[\n\\theta_m \\sim \\textsf{Dirichlet}(\\alpha).\n\\]\nThe prior hyperparameter \\(\\alpha\\) is fixed to a \\(K\\)-vector of positive values. Each word in the document is generated independently conditional on the distribution \\(\\theta_m\\). First, a topic \\(z_{m,n} \\in \\{1,\\dotsc,K\\}\\) is drawn for the word based on the document-specific topic-distribution, \\[\nz_{m,n} \\sim \\textsf{categorical}(\\theta_m).\n\\]\nFinally, the word \\(w_{m,n}\\) is drawn according to the word distribution for topic \\(z_{m,n}\\), \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m,n]}).\n\\] The distributions \\(\\phi_k\\) over words for topic \\(k\\) are also given a Dirichlet prior, \\[\n\\phi_k \\sim \\textsf{Dirichlet}(\\beta)\n\\]\nwhere \\(\\beta\\) is a fixed \\(V\\)-vector of positive values.\n\n\n\nAlthough Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is \\[\\begin{align*}\np(\\theta,\\phi \\mid w,\\alpha,\\beta)\n&\\propto p(\\theta \\mid \\alpha) \\, p(\\phi \\mid \\beta) \\, p(w \\mid \\theta,\\phi) \\\\\n&= \\prod_{m=1}^M p(\\theta_m \\mid \\alpha)\n    \\times \\prod_{k=1}^K p(\\phi_k \\mid \\beta)\n    \\times \\prod_{m=1}^M \\prod_{n=1}^{M[n]} p(w_{m,n} \\mid \\theta_m,\\phi).\n\\end{align*}\\]\nThe inner word-probability term is defined by summing out the topic assignments, \\[\\begin{align*}\np(w_{m,n} \\mid \\theta_m,\\phi)\n&= \\sum_{z=1}^K p(z,w_{m,n} \\mid \\theta_m,\\phi) \\\\\n&= \\sum_{z=1}^K p(z \\mid \\theta_m) \\, p(w_{m,n} \\mid \\phi_z).\n\\end{align*}\\]\nPlugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, \\[\\begin{align*}\n\\log\\, &p(\\theta,\\phi \\mid w,\\alpha,\\beta) \\\\\n&= \\sum_{m=1}^M \\log \\textsf{Dirichlet}(\\theta_m \\mid \\alpha)\n    + \\sum_{k=1}^K \\log \\textsf{Dirichlet}(\\phi_k \\mid \\beta) \\\\\n&\\qquad + \\sum_{m=1}^M \\sum_{n=1}^{N[m]} \\log \\left(\n    \\sum_{z=1}^K \\textsf{categorical}(z \\mid \\theta_m)\n    \\times \\textsf{categorical}(w_{m,n} \\mid \\phi_z)\n  \\right)\n\\end{align*}\\]\n\n\n\nApplying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.\ndata {\n  int&lt;lower=2&gt; K;               // num topics\n  int&lt;lower=2&gt; V;               // num words\n  int&lt;lower=1&gt; M;               // num docs\n  int&lt;lower=1&gt; N;               // total word instances\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  array[M] simplex[K] theta;    // topic dist for doc m\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  for (m in 1:M) {\n    theta[m] ~ dirichlet(alpha);  // prior\n  }\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);     // prior\n  }\n  for (n in 1:N) {\n    array[K] real gamma;\n    for (k in 1:K) {\n      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);\n    }\n    target += log_sum_exp(gamma);  // likelihood;\n  }\n}\nAs in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.\n\n\n\nTo account for correlations in the distribution of topics for documents, Blei and Lafferty (2007) introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.\nThe authors treat the prior as a fixed hyperparameter. They use an \\(L_1\\)-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.\n\n\nThe Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior alpha in the data declaration with the mean and covariance of the multivariate logistic normal prior.\ndata {\n  // ... data as before without alpha ...\n  vector[K] mu;          // topic mean\n  cov_matrix[K] Sigma;   // topic covariance\n}\nRather than drawing the simplex parameter theta from a Dirichlet, a parameter eta is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.\nparameters {\n  array[K] simplex[V] phi;     // word dist for topic k\n  array[M] vector[K] eta;      // topic dist for doc m\n}\ntransformed parameters {\n  array[M] simplex[K] theta;\n  for (m in 1:M) {\n    theta[m] = softmax(eta[m]);\n  }\n}\nmodel {\n  for (m in 1:M) {\n    eta[m] ~ multi_normal(mu, Sigma);\n  }\n  // ... model as before w/o prior for theta ...\n}\n\n\n\nBy adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean mu and covariance Sigma from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix Sigma may be encoded as follows.\n// ... data block as before, but without alpha ...\nparameters {\n  vector[K] mu;              // topic mean\n  corr_matrix[K] Omega;      // correlation matrix\n  vector&lt;lower=0&gt;[K] sigma;  // scales\n  array[M] vector[K] eta;    // logit topic dist for doc m\n  array[K] simplex[V] phi;   // word dist for topic k\n}\ntransformed parameters {\n  // ... eta as above ...\n  cov_matrix[K] Sigma;       // covariance matrix\n  for (m in 1:K) {\n    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];\n  }\n  for (m in 1:(K-1)) {\n    for (n in (m+1):K) {\n      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];\n      Sigma[n, m] = Sigma[m, n];\n    }\n  }\n}\nmodel {\n  mu ~ normal(0, 5);      // vectorized, diffuse\n  Omega ~ lkj_corr(2.0);  // regularize to unit correlation\n  sigma ~ cauchy(0, 5);   // half-Cauchy due to constraint\n  // ... words sampled as above ...\n}\nThe \\(\\textsf{LKJCorr}\\) distribution with shape \\(\\alpha &gt; 0\\) has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by \\[\n\\mathsf{LkjCorr}(\\Omega\\mid\\alpha) \\propto \\mathrm{det}(\\Omega)^{\\alpha - 1}\n\\] With a scale of \\(\\alpha = 2\\), the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix \\(\\Sigma\\) for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on sigma.",
+    "text": "The complex data type is a scalar, but unlike real and int types, it contains two components, a real and imaginary component, both of which are of type real. That is, the real and imaginary components of a complex number are 64-bit, IEEE 754-complaint floating point numbers.\n\n\nImaginary literals are written in mathematical notation using a numeral followed by the suffix i. For example, the following example constructs a complex number \\(2 - 1.3i\\) and assigns it to the variable z.\ncomplex z = 2 - 1.3i;\nreal re = get_real(z);  // re has value 2.0\nreal im = get_imag(z);  // im has value -1.3\nThe getter functions then extract the real and imaginary components of z and assign them to re and im respectively.\nThe function to_complex constructs a complex number from its real and imaginary components. The functional form needs to be used whenever the components are not literal numerals, as in the following example.\nvector[K] re;\nvector[K] im;\n// ...\nfor (k in 1:K) {\n  complex z = to_complex(re[k], im[k]);\n  // ...\n}\n\n\n\nExpressions of type real may be assigned to variables of type complex. For example, the following is a valid sequence of Stan statements.\nreal x = 5.0;\ncomplex z = x;  // get_real(z) == 5.0, get_imag(z) == 0\nThe real number assigned to a complex number determine’s the complex number’s real component, with the imaginary component set to zero.\nAssignability is transitive, so that expressions of type int may also be assigned to variables of type complex, as in the following example.\nint n = 2;\ncomplex z = n;\nFunction arguments also support promotion of integer or real typed expressions to type complex.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/clustering.html#footnotes",
-    "href": "stan-users-guide/clustering.html#footnotes",
-    "title": "Clustering Models",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nFor clustering, the non-identifiability problems for all mixture models present a problem, whereas there is no such problem for classification. Despite the difficulties with full Bayesian inference for clustering, researchers continue to use it, often in an exploratory data analysis setting rather than for predictive modeling.↩︎",
+    "objectID": "reference-manual/types.html#scalar-data-types-and-variable-declarations",
+    "href": "reference-manual/types.html#scalar-data-types-and-variable-declarations",
+    "title": "Data Types and Declarations",
+    "section": "",
+    "text": "All variables used in a Stan program must have an explicitly declared data type. The form of a declaration includes the type and the name of a variable. This section covers scalar types, namely integer, real, and complex. The next section covers vector and matrix types, and the following section array types.\n\n\nUnconstrained integers are declared using the int keyword. For example, the variable N is declared to be an integer as follows.\nint N;\n\n\n\nInteger data types may be constrained to allow values only in a specified interval by providing a lower bound, an upper bound, or both. For instance, to declare N to be a positive integer, use the following.\nint&lt;lower=1&gt; N;\nThis illustrates that the bounds are inclusive for integers.\nTo declare an integer variable cond to take only binary values, that is zero or one, a lower and upper bound must be provided, as in the following example.\nint&lt;lower=0, upper=1&gt; cond;\n\n\n\nUnconstrained real variables are declared using the keyword real. The following example declares theta to be an unconstrained continuous value.\nreal theta;\n\n\n\nUnconstrained complex numbers are declared using the keyword complex. The following example declares z to be an unconstrained complex variable.\ncomplex z;\n\n\n\nReal variables may be bounded using the same syntax as integers. In theory (that is, with arbitrary-precision arithmetic), the bounds on real values would be exclusive. Unfortunately, finite-precision arithmetic rounding errors will often lead to values on the boundaries, so they are allowed in Stan.\nThe variable sigma may be declared to be non-negative as follows.\nreal&lt;lower=0&gt; sigma;\nThe following declares the variable x to be less than or equal to \\(-1\\).\nreal&lt;upper=-1&gt; x;\nTo ensure rho takes on values between \\(-1\\) and \\(1\\), use the following declaration.\nreal&lt;lower=-1, upper=1&gt; rho;\n\n\nLower bounds that are negative infinity or upper bounds that are positive infinity are ignored. Stan provides constants positive_infinity() and negative_infinity() which may be used for this purpose, or they may be supplied as data.\n\n\n\n\nReal variables may be declared on a space that has been transformed using an affine transformation \\(x\\mapsto \\mu + \\sigma * x\\) with offset \\(\\mu\\) and (positive) multiplier \\(\\sigma\\), using a syntax similar to that for bounds. While these transforms do not change the asymptotic sampling behaviour of the resulting Stan program (in a sense, the model the program implements), they can be useful for making the sampling process more efficient by transforming the geometry of the problem to a more natural multiplier and to a more natural offset for the sampling process, for instance by facilitating a non-centered parameterisation. While these affine transformation declarations do not impose a hard constraint on variables, they behave like the bounds constraints in many ways and could perhaps be viewed as acting as a sort of soft constraint.\nThe variable x may be declared to have offset \\(1\\) as follows.\nreal&lt;offset=1&gt; x;\nSimilarly, it can be declared to have multiplier \\(2\\) as follows.\nreal&lt;multiplier=2&gt; x;\nFinally, we can combine both declarations to declare a variable with offset \\(1\\) and multiplier \\(2\\).\nreal&lt;offset=1, multiplier=2&gt; x;\nAs an example, we can give x a normal distribution with non-centered parameterization as follows.\nparameters {\n  real&lt;offset=mu, multiplier=sigma&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nRecall that the centered parameterization is achieved with the code\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nor equivalently\nparameters {\n  real&lt;offset=0, multiplier=1&gt; x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\n\n\n\nBounds (and offset and multiplier) for integer or real variables may be arbitrary expressions. The only requirement is that they only include variables that have been declared (though not necessarily defined) before the declaration. If the bounds themselves are parameters, the behind-the-scenes variable transform accounts for them in the log Jacobian.\nFor example, it is acceptable to have the following declarations.\ndata {\n real lb;\n}\nparameters {\n   real&lt;lower=lb&gt; phi;\n}\nThis declares a real-valued parameter phi to take values greater than the value of the real-valued data variable lb. Constraints may be arbitrary expressions, but must be of type int for integer variables and of type real for real variables (including constraints on vectors, row vectors, and matrices). Variables used in constraints can be any variable that has been defined at the point the constraint is used. For instance,\ndata {\n   int&lt;lower=1&gt; N;\n   array[N] real y;\n}\nparameters {\n   real&lt;lower=min(y), upper=max(y)&gt; phi;\n}\nThis declares a positive integer data variable N, an array y of real-valued data of length N, and then a parameter ranging between the minimum and maximum value of y. As shown in the example code, the functions min() and max() may be applied to containers such as arrays.\nA more subtle case involves declarations of parameters or transformed parameters based on parameters declared previously. For example, the following program will work as intended.\nparameters {\n  real a;\n  real&lt;lower=a&gt; b;  // enforces a &lt; b\n}\ntransformed parameters {\n  real c;\n  real&lt;lower=c&gt; d;\n  c = a;\n  d = b;\n}\nThe parameters instance works because all parameters are defined externally before the block is executed. The transformed parameters case works even though c isn’t defined at the point it is used, because constraints on transformed parameters are only validated at the end of the block. Data variables work like parameter variables, whereas transformed data and generated quantity variables work like transformed parameter variables.\n\n\n\nA variable may be declared with a size that depends on a boolean constant. For example, consider the definition of alpha in the following program fragment.\ndata {\n  int&lt;lower=0, upper=1&gt; include_alpha;\n  // ...\n}\nparameters {\n  vector[include_alpha ? N : 0] alpha;\n  // ...\n}\nIf include_alpha is true, the model will include the vector alpha; if the flag is false, the model will not include alpha (technically, it will include alpha of size 0, which means it won’t contain any values and won’t be included in any output).\nThis technique is not just useful for containers. If the value of N is set to 1, then the vector alpha will contain a single element and thus alpha[1] behaves like an optional scalar, the existence of which is controlled by include_alpha.\nThis coding pattern allows a single Stan program to define different models based on the data provided as input. This strategy is used extensively in the implementation of the RStanArm package.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Clustering Models"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html",
-    "href": "stan-users-guide/copulas.html",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#vector-and-matrix-data-types",
+    "href": "reference-manual/types.html#vector-and-matrix-data-types",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "Copulas provide a flexible way to model multivariate distributions by separating the marginal cumulative distribution functions from the dependence structure. This chapter introduces copulas in Stan, focusing on implementation techniques and practical examples. This chapter was derived from Brynjólfur Gauti Guðrúnar Jónsson’s A gentle introduction: the Gaussian copula.\n\n\nAccording to Sklar’s theorem (Sklar 1959), any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.\nFor a multivariate random variable \\(\\mathbf{X} = [X_1 \\cdots X_D]^\\top\\) with marginal cumulative distribution functions \\(F_i\\), the joint cumulative distribution function can be written as:\n\\[\nF_{\\mathbf{X}}(\\mathbf{x}) = C(F_1(x_1), \\ldots, F_D(x_D)) = \\Pr[X_1 \\leq x_1, \\ldots, X_D \\leq x_D]\n\\]\nwhere \\(C\\) is the copula function, \\(F_{\\mathbf{X}}\\) is the joint cumulative distribution function, and \\(F_i\\) are the marginal cumulative distribution functions. The copula function \\(C\\) must be a joint cumulative distribution function over the unit hypercube \\([0, 1]^D\\).\n\n\n\nThis section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.\nThe log density of a multivariate distribution using a copula can be written as:\n\\[\n\\log h(\\mathbf{x}) = \\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right) + \\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\n\\]\nwhere:\n\n\\(u_i = F_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) are the probability integral transforms of the data\n\\(\\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right)\\) is the log density of the copula\n\\(\\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) is the sum of the log densities of the marginals\n\\(\\boldsymbol{\\alpha}\\) represents the parameters describing the parametric form of the copula\n\\(\\boldsymbol{\\beta}_i\\) represents the parameters describing the parametric form of the \\(i\\)-th marginal distribution\n\nThe implementation of copulas in Stan has two key requirements:\n\nBoth the probability density functions and cumulative distribution functions of the marginal distributions must be available\nA function that computes the log density of the copula for the transformed data must be implemented\n\nMost copula implementations in Stan follow a three-step process:\n\nAccumulate marginal log likelihoods: Calculate and add the log density of each marginal distribution to the target log density\nTransform to uniform variables: Apply the marginal CDFs to transform the data to uniform variables on the unit interval\nCalculate copula density: Compute the log density of the copula based on these uniform variables and add it to the target log density\n\nThis process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.\nIn a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where \\(\\log c(\\mathbf{u}) = 0\\), resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\nMost parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).\n\n\n\nThe Gaussian copula is constructed using the multivariate normal distribution. For a \\(D\\)-dimensional random vector \\(\\mathbf{X}\\) with marginals \\(F_i\\), the log Gaussian copula density is given by:\n\\[\n\\begin{aligned}\n\\log c(\\mathbf{u}) &=\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top (\\boldsymbol{\\Omega}^{-1} - \\mathbf{I}) \\mathbf{z} \\\\\n& =\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top \\boldsymbol{\\Omega}^{-1} \\mathbf{z} + \\frac{1}{2} \\mathbf{z}^\\top \\mathbf{z} \\\\\n&= \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\boldsymbol{\\Omega}) - \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\mathbf{I})\n\\end{aligned}\n\\]\nwhere \\(\\mathbf{z} = [\\Phi^{-1}(u_1), \\ldots, \\Phi^{-1}(u_D)]^\\top\\) are the inverse normal CDF transforms of the uniform marginals, \\(\\boldsymbol{\\Omega}\\) is the correlation matrix, and \\(\\mathbf{I}\\) is the identity matrix. The joint log density is then:\n\\[\n\\log h(\\mathbf{x}) = \\log c(F_1(x_1), \\ldots, F_D(x_D)) + \\sum_{i=1}^D \\log f_i(x_i)\n\\]\nFollowing the three-step process for implementing copulas in Stan:\n\nAccumulate marginal log likelihoods: The exponential log densities are added to the target in the line target += exponential_lpdf(y[n] | lambda)\nTransform to uniform variables: The exponential CDF transforms the data to uniform variables: exponential_cdf(y[n, d] | lambda[d])\nCalculate copula density: The transformed variables are converted to normal scale using inv_Phi and the multivariate normal log density is computed: z ~ multi_normal_cholesky(zeros, L_Omega)\n\nThe following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  int&lt;lower=0&gt; D;  // number of dimensions\n  vector&lt;lower=0&gt;[D] y[N];  // data\n}\n\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\n\nparameters {\n  // Parameters for exponential marginal distributions\n  vector&lt;lower=0&gt;[D] lambda;  // rate parameters\n  \n  // Correlation matrix for Gaussian copula\n  cholesky_factor_corr[D] L_Omega;\n}\n\nmodel {\n  // Priors\n  lambda ~ gamma(2, 1);  // prior for rate parameters\n  L_Omega ~ lkj_corr_cholesky(2);\n  \n  // Likelihood using Gaussian copula with exponential marginals\n  for (n in 1:N) {\n    // Add exponential log density to target\n    target += exponential_lpdf(y[n] | lambda);\n    \n    vector[D] z;\n    for (d in 1:D) {\n      // Transform to uniform using exponential CDF\n      real u_d = exponential_cdf(y[n, d] | lambda[d]);\n      \n      // Transform to standard normal\n      z[d] = inv_Phi(u_d);\n    }\n    // Multivariate normal log density with correlation matrix\n    z ~ multi_normal_cholesky(zeros, L_Omega);\n  }\n}\n\ngenerated quantities {\n  // Optional: Recover correlation matrix from Cholesky factor\n  matrix[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\n\n\n\nCopulas offer several advantages in statistical modeling:\n\nFlexibility: They allow combining any marginal distributions with various dependence structures. For example:\n\nModeling financial returns with heavy-tailed marginals and complex dependence structures\nCombining different types of distributions (e.g., normal and gamma) in a single model\nCapturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic\nModeling different types of tail dependence in different parts of the distribution\n\nFactorability: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.\nFor example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.\nTail dependence: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.\nUniversal Framework: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\n\n\n\n\nWhen implementing copulas in Stan, several considerations should be kept in mind:\n\nComputational efficiency: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.\nParameter identifiability: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.\nModel selection: The choice of copula family should be guided by the specific dependence structure of the data. For example:\n\nThe Gaussian copula may underestimate the probability of joint extreme events in financial data\nThe Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications\nArchimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions\n\nNumerical stability: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.\nSymmetry considerations: Many copula families exhibit strong symmetries that may not match the data:\n\nRadial symmetry: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns\nExchangeability: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.\n\nTail dependence: Understanding and choosing appropriate tail dependence is crucial:\n\nThe upper (lower) tail dependence coefficient \\(\\lambda_U (\\lambda_L)\\) is the probability that one variable is extremely large (small) given that another is extremely large (small).\nDifferent copula families exhibit different tail dependence properties:\n\nSome copulas (like Gaussian) have zero tail dependence\nOthers can model symmetric tail dependence (\\(\\lambda_U = \\lambda_L\\))\nSome can capture asymmetric tail dependence (\\(\\lambda_U \\neq \\lambda_L\\))\nCertain copulas allow for tail dependence even with zero correlation\n\nThe choice of copula should be guided by the expected tail behavior in the application:\n\nFinancial data often requires modeling joint lower extreme events\nRisk management applications may need asymmetric tail dependence\nSome applications may require different tail behavior in different parts of the distribution\n\n\nHigh-dimensional modeling: As dimensionality increases:\n\nThe number of dependence parameters grows\nSome copula families become less flexible\nVine copulas or factor copulas may be more appropriate\n\n\n\n\n\nSeveral copula families are available for modeling different dependence structures in the correlation component:\n\nGaussian copula: Based on the multivariate normal distribution, offering symmetric dependence\nStudent-t copula: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula\nArchimedean copulas: A class of copulas defined through generator functions, including:\n\nClayton copula: Stronger lower tail dependence\nGumbel copula: Stronger upper tail dependence\nFrank copula: Symmetric dependence\n\nVine copulas: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas\n\n\n\n\n\nJónsson’s three part blog series, Copulas in Stan:\n\nPart I: If it bleeds, we can kill it\nPart II: A gentle introduction: the Gaussian copula\nPart III: It was the best of tails, it was the worst of tails: The T-Copula\n\nBrynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, Copulas in Stan: Modeling Spatial Dependence\nSean Pinkney’s Helpful Stan functions: copula functions",
+    "text": "Stan provides three types of container objects: arrays, vectors, and matrices. Vectors and matrices are more limited kinds of data structures than arrays. Vectors are intrinsically one-dimensional collections of real or complex values, whereas matrices are intrinsically two dimensional. Vectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical. A \\(3 \\times 4\\) matrix is a different kind of object in Stan than a \\(3\n\\times 4\\) array.\nThe intention of using matrix types is to call out their usage in the code. There are three situations in Stan where only vectors and matrices may be used,\n\nmatrix arithmetic operations (e.g., matrix multiplication)\nlinear algebra functions (e.g., eigenvalues and determinants), and\nmultivariate function parameters and outcomes (e.g., multivariate normal distribution arguments).\n\nVectors and matrices cannot be typed to return integer values. They are restricted to real and complex values.\nFor constructing vectors and matrices in Stan, see Vector, Matrix, and Array Expressions.\n\n\nVectors and matrices, as well as arrays, are indexed starting from one (1) in Stan. This follows the convention in statistics and linear algebra as well as their implementations in the statistical software packages R, MATLAB, BUGS, and JAGS. General computer programming languages, on the other hand, such as C++ and Python, index arrays starting from zero.\n\n\n\nVectors in Stan are column vectors; see below for information on row vectors. Vectors are declared with a size (i.e., a dimensionality). For example, a 3-dimensional real vector is declared with the keyword vector, as follows.\nvector[3] u;\nVectors may also be declared with constraints, as in the following declaration of a 3-vector of non-negative values.\nvector&lt;lower=0&gt;[3] u;\nSimilarly, they may be declared with a offset and/or multiplier, as in the following example\nvector&lt;offset=42, multiplier=3&gt;[3] u;\n\n\n\nLike real vectors, complex vectors are column vectors and are declared with a size. For example, a 3-dimensional complex vector is declared with the keyword complex_vector, as follows.\ncomplex_vector[3] v;\nComplex vector declarations do not support any constraints.\n\n\n\nA unit simplex is a vector with non-negative values whose entries sum to 1. For instance, \\([0.2,0.3,0.4,0.1]^{\\top}\\) is a unit 4-simplex. Unit simplexes are most often used as parameters in categorical or multinomial distributions, and they are also the sampled variate in a Dirichlet distribution. Simplexes are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-simplex by\nsimplex[5] theta;\nUnit simplexes are implemented as vectors and may be assigned to other vectors and vice-versa. Simplex variables, like other constrained variables, are validated to ensure they contain simplex values; for simplexes, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\nIn high dimensional problems, simplexes may require smaller step sizes in the inference algorithms in order to remain stable; this can be achieved through higher target acceptance rates for samplers and longer warmup periods, tighter tolerances for optimization with more iterations, and in either case, with less dispersed parameter initialization or custom initialization if there are informative priors for some parameters.\n\n\n\nA stochastic matrix is a matrix where each column or row is a unit simplex, meaning that each column (row) vector has non-negative values that sum to 1. The following example is a \\(3 \\times 4\\) column-stochastic matrix.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.3 \\\\\n0.3 & 0.3 & 0.6 & 0.4 \\\\\n0.5 & 0.2 & 0.3 & 0.3\n\\end{bmatrix}\n\\]\nAn example of a \\(3 \\times 4\\) row-stochastic matrix is the following.\n\\[\n\\begin{bmatrix}\n0.2 & 0.5 & 0.1 & 0.2 \\\\\n0.2 & 0.1 & 0.6 & 0.1 \\\\\n0.5 & 0.2 & 0.2 & 0.1\n\\end{bmatrix}\n\\]\nIn the examples above, each column (or row) sums to 1, making the matrices valid column_stochastic_matrix and row_stochastic_matrix types.\nColumn-stochastic matrices are often used in models where each column represents a probability distribution across a set of categories such as in multiple multinomial distributions, factor models, transition matrices in Markov models, or compositional data analysis. They can also be used in situations where you need multiple simplexes of the same dimensionality.\nThe column_stochastic_matrix and row_stochastic_matrix types are declared with row and column sizes. For instance, a matrix theta with 3 rows and 4 columns, where each column is a 3-simplex, is declared like a matrix with 3 rows and 4 columns.\ncolumn_stochastic_matrix[3, 4] theta;\nA matrix theta with 3 rows and 4 columns, where each row is a 4-simplex, is similarly declared as a matrix with 3 rows and 4 columns.\nrow_stochastic_matrix[3, 4] theta;\nAs with simplexes, column_stochastic_matrix and row_stochastic_matrix variables are subject to validation, ensuring that each column (row) satisfies the simplex constraints. This validation accounts for floating-point imprecision, with checks performed up to a statically specified accuracy threshold \\(\\epsilon\\).\n\n\nIn high-dimensional settings, column_stochastic_matrix and row_stochastic_matrix types may require careful tuning of the inference algorithms. To ensure stability:\n\nSmaller Step Sizes: In samplers like Hamiltonian Monte Carlo (HMC), smaller step sizes can help maintain stability, especially in high dimensions.\nHigher Target Acceptance Rates: Setting higher target acceptance rates can improve the robustness of the sampling process.\nLonger Warmup Periods: Increasing the warmup period allows the sampler to better explore the parameter space before the actual sampling begins.\nTighter Optimization Tolerances: For optimization-based inference, tighter tolerances with more iterations can yield more accurate results.\nCustom Initialization: If prior information about the parameters is available, custom initialization or less dispersed initialization can lead to more efficient inference.\n\n\n\n\n\nA unit vector is a vector with a norm of one. For instance, \\([0.5,\n0.5, 0.5, 0.5]^{\\top}\\) is a unit 4-vector. Unit vectors are sometimes used in directional statistics. Unit vectors are declared with their full dimensionality. For instance, theta is declared to be a unit \\(5\\)-vector by\nunit_vector[5] theta;\nUnit vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Unit vector variables, like other constrained variables, are validated to ensure that they are indeed unit length; for unit vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nA zero-sum vector is constrained such that the sum of its elements is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\(N - 1\\) degrees of freedom, zero sum vectors are declared with their full dimensionality. For instance, beta is declared to be a zero-sum \\(5\\)-vector (4 DoF) by\nsum_to_zero_vector[5] beta;\nZero sum vectors are implemented as vectors and may be assigned to other vectors and vice-versa. Zero sum vector variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum vectors, this is only done up to a statically specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nAn ordered vector type in Stan represents a vector whose entries are sorted in ascending order. For instance, \\((-1.3,2.7,2.71)^{\\top}\\) is an ordered 3-vector. Ordered vectors are most often employed as cut points in ordered logistic regression models (see section).\nThe variable c is declared as an ordered 5-vector by\nordered[5] c;\nAfter their declaration, ordered vectors, like unit simplexes, may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nThere is also a positive, ordered vector type which operates similarly to ordered vectors, but all entries are constrained to be positive. For instance, \\((2,3.7,4,12.9)\\) is a positive, ordered 4-vector.\nThe variable d is declared as a positive, ordered 5-vector by\npositive_ordered[5] d;\nLike ordered vectors, after their declaration, positive ordered vectors may be assigned to other vectors and other vectors may be assigned to them. Constraints will be checked after executing the block in which the variables were declared.\n\n\n\nRow vectors are declared with the keyword row_vector. Like (column) vectors, they are declared with a size. For example, a 1093-dimensional row vector u would be declared as\nrow_vector[1093] u;\nConstraints are declared as for vectors, as in the following example of a 10-vector with values between -1 and 1.\nrow_vector&lt;lower=-1, upper=1&gt;[10] u;\nOffset and multiplier are also similar as for the following 3-row-vector with offset -42 and multiplier 3.\nrow_vector&lt;offset=-42, multiplier=3&gt;[3] u;\nRow vectors may not be assigned to column vectors, nor may column vectors be assigned to row vectors. If assignments are required, they may be accommodated through the transposition operator.\n\n\n\nComplex row vectors are declared with the keyword complex_row_vector and given a size in basic declarations. For example, a 12-dimensional complex row vector v would be declared as\ncomplex_row_vector[12] v;\nComplex row vectors do not allow constraints.\n\n\n\nMatrices are declared with the keyword matrix along with a number of rows and number of columns. For example,\nmatrix[3, 3] A;\nmatrix[M, N] B;\ndeclares A to be a \\(3 \\times 3\\) matrix and B to be a \\(M\n\\times N\\) matrix. For the second declaration to be well formed, the variables M and N must be declared as integers in either the data or transformed data block and before the matrix declaration.\nMatrices may also be declared with constraints, as in this (\\(3 \\times 4\\)) matrix of non-positive values.\nmatrix&lt;upper=0&gt;[3, 4] B;\nSimilarly, matrices can be declared to have a set offset and/or multiplier, as in this matrix with multiplier 5.\nmatrix&lt;multiplier=5&gt;[3, 4] B;\n\n\nRows of a matrix can be assigned by indexing the left-hand side of an assignment statement. For example, this is possible.\nmatrix[M, N] a;\nrow_vector[N] b;\n// ...\na[1] = b;\nThis copies the values from row vector b to a[1], which is the first row of the matrix a. If the number of columns in a is not the same as the size of b, a run-time error is raised; the number of columns of a is N, which is also the number of columns of b.\nAssignment works by copying values in Stan. That means any subsequent assignment to a[1] does not affect b, nor does an assignment to b affect a.\n\n\n\n\nComplex matrices are declared with the keyword complex_matrix and a number of rows and columns. For example,\ncomplex_matrix[3, 3] C;\nComplex matrices do not allow constraints.\n\n\n\nMatrix variables may be constrained to represent covariance matrices. A matrix is a covariance matrix if it is symmetric and positive definite. Like correlation matrices, covariance matrices only need a single dimension in their declaration. For instance,\ncov_matrix[K] Sigma;\ndeclares Sigma to be a \\(K \\times K\\) covariance matrix, where \\(K\\) is the value of the data variable K.\n\n\n\nMatrix variables may be constrained to represent correlation matrices. A matrix is a correlation matrix if it is symmetric and positive definite, has entries between \\(-1\\) and \\(1\\), and has a unit diagonal. Because correlation matrices are square, only one dimension needs to be declared. For example,\ncorr_matrix[3] Omega;\ndeclares Omega to be a \\(3 \\times 3\\) correlation matrix.\nCorrelation matrices may be assigned to other matrices, including unconstrained matrices, if their dimensions match, and vice-versa.\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a covariance matrix. This is often more convenient or more efficient than representing covariance matrices directly.\nA Cholesky factor \\(L\\) is an \\(M \\times N\\) lower-triangular matrix (if \\(m &lt; n\\) then \\(L[m, n] =0\\)) with a strictly positive diagonal (\\(L[k, k]\n&gt; 0\\)) and \\(M \\geq N\\). If \\(L\\) is a Cholesky factor, then \\(\\Sigma = L\n\\, L^{\\top}\\) is a covariance matrix (i.e., it is positive definite). The mapping between positive definite matrices and their Cholesky factors is bijective—every covariance matrix has a unique Cholesky factorization.\nThe typical case of a square Cholesky factor may be declared with a single dimension,\ncholesky_factor_cov[4] L;\n\n\nIn general, two dimensions may be declared, with the above being equal to cholesky_factor_cov[4, 4]. The type cholesky_factor_cov[M, N] may be used for the general \\(M \\times N\\) case to produce positive semi-definite matrices of rank \\(M\\).\n\n\n\n\nMatrix variables may be constrained to represent the Cholesky factors of a correlation matrix.\nA Cholesky factor for a correlation matrix \\(L\\) is a \\(K \\times K\\) lower-triangular matrix with positive diagonal entries and rows that are of length 1 (i.e., \\(\\sum_{n=1}^K L_{m,n}^2 = 1\\)). If \\(L\\) is a Cholesky factor for a correlation matrix, then \\(L\\,L^{\\top}\\) is a correlation matrix (i.e., symmetric positive definite with a unit diagonal).\nTo declare the variable L to be a K by K Cholesky factor of a correlation matrix, the following code may be used.\ncholesky_factor_corr[K] L;\n\n\n\nA sum-to-zero matrix is constrained such that the sum of rows and the sum of the columns is always \\(0\\). These are sometimes useful for resolving identifiability issues in regression models. While the underlying vector has only \\((N - 1) \\times (M - 1)\\) degrees of freedom, zero sum matrices are declared with their full dimensionality. For instance, beta is declared to be a sum-to-zero \\(5 \\times 4\\)-matrix (20 degrees of freedom) by\nsum_to_zero_matrix[5, 4] beta;\nSum-to-zero matrices are implemented as matrices and may be assigned to other matrices and vice-versa. Zero sum matrix variables, like other constrained variables, are validated to ensure that they are indeed sum to zero; for zero sum matrices, this is only done up to an internally specified accuracy threshold \\(\\epsilon\\) to account for errors arising from floating-point imprecision.\n\n\n\nConstrained variables of all types may be assigned to other variables of the same unconstrained type and vice-versa. Matching is interpreted strictly as having the same basic type and number of array dimensions. Constraints are not considered, but basic data types are. For instance, a variable declared to be real&lt;lower=0, upper=1&gt; could be assigned to a variable declared as real and vice-versa. Similarly, a variable declared as matrix[3, 3] may be assigned to a variable declared as cov_matrix[3] or cholesky_factor_cov[3], and vice-versa.\nChecks are carried out at the end of each relevant block of statements to ensure constraints are enforced. This includes run-time size checks. The Stan compiler isn’t able to catch the fact that an attempt may be made to assign a matrix of one dimensionality to a matrix of mismatching dimensionality.\n\n\n\nReal-valued vectors, row vectors and matrices may be assigned to complex-valued vectors, row vectors and matrices, respectively. For example, the following is legal.\nvector[N] v = ...;\ncomplex_vector[N] u = 2 * v;\nRow vectors and matrices work the same way.\n\n\n\nVariables may be declared with sizes given by expressions. Such expressions are constrained to only contain data or transformed data variables. This ensures that all sizes are determined once the data is read in and transformed data variables defined by their statements. For example, the following is legal.\ndata {\n  int&lt;lower=0&gt; N_observed, N_missing;\n  // ...\ntransformed parameters {\n  vector[N_observed + N_missing] y;\n  // ...\n\n\n\nIf v is a column vector or row vector, then v[2] is the second element in the vector. If m is a matrix, then m[2, 3] is the value in the second row and third column.\nProviding a matrix with a single index returns the specified row. For instance, if m is a matrix, then m[2] is the second row. This allows Stan blocks such as\nmatrix[M, N] m;\nrow_vector[N] v;\nreal x;\n// ...\nv = m[2];\nx = v[3];   // x == m[2][3] == m[2, 3]\nThe type of m[2] is row_vector because it is the second row of m. Thus it is possible to write m[2][3] instead of m[2, 3] to access the third element in the second row. When given a choice, the form m[2, 3] is preferred.\nComplex versions work the same way,\ncomplex_matrix[M, N] m = ...;\ncomplex_row_vector[N] u = m[3];\ncomplex_vector[M] v = m[ , 2];\n\n\nThe form m[2, 3] is more efficient because it does not require the creation and use of an intermediate expression template for m[2]. In later versions, explicit calls to m[2][3] may be optimized to be as efficient as m[2, 3] by the Stan compiler.\n\n\n\n\nAn integer expression is used to pick out the sizes of vectors, matrices, and arrays. For instance, we can declare a vector of size M + N using\nvector[M + N] y;\nAny integer-denoting expression may be used for the size declaration, providing all variables involved are either data, transformed data, or local variables. That is, expressions used for size declarations may not include parameters or transformed parameters or generated quantities.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#what-are-copulas",
-    "href": "stan-users-guide/copulas.html#what-are-copulas",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#array-data-types.section",
+    "href": "reference-manual/types.html#array-data-types.section",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "According to Sklar’s theorem (Sklar 1959), any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.\nFor a multivariate random variable \\(\\mathbf{X} = [X_1 \\cdots X_D]^\\top\\) with marginal cumulative distribution functions \\(F_i\\), the joint cumulative distribution function can be written as:\n\\[\nF_{\\mathbf{X}}(\\mathbf{x}) = C(F_1(x_1), \\ldots, F_D(x_D)) = \\Pr[X_1 \\leq x_1, \\ldots, X_D \\leq x_D]\n\\]\nwhere \\(C\\) is the copula function, \\(F_{\\mathbf{X}}\\) is the joint cumulative distribution function, and \\(F_i\\) are the marginal cumulative distribution functions. The copula function \\(C\\) must be a joint cumulative distribution function over the unit hypercube \\([0, 1]^D\\).",
+    "text": "Stan supports arrays of arbitrary dimension. The values in an array can be any type, so that arrays may contain values that are simple reals or integers, vectors, matrices, or other arrays. Arrays are the only way to store sequences of integers, and some functions in Stan, such as discrete distributions, require integer arguments.\nA two-dimensional array is just an array of arrays, both conceptually and in terms of current implementation. When an index is supplied to an array, it returns the value at that index. When more than one index is supplied, this indexing operation is chained. For example, if a is a two-dimensional array, then a[m, n] is just a convenient shorthand for a[m][n].\nVectors, matrices, and arrays are not assignable to one another, even if their dimensions are identical.\nFor constructing arrays in Stan, see Vector, Matrix, and Array Expressions.\n\n\nArrays are declared with the keyword array followed by the dimensions enclosed in square brackets, the element type, and the name of the variable.\nThe variable n is declared as an array of five integers as follows.\narray[5] int n;\nA two-dimensional array of complex values with three rows and four columns is declared as follows.\narray[3, 4] complex a;\nA three-dimensional array z of positive reals with five rows, four columns, and two shelves can be declared as follows.\narray[5, 4, 2] real&lt;lower=0&gt; z;\nArrays may also be declared to contain vectors. For example,\narray[3] vector[7] mu;\ndeclares mu to be an array of size 3 containing vectors with 7 elements. Arrays may also contain matrices. The example\narray[15, 12] complex_matrix[7, 2] mu;\ndeclares a 15 by 12 array of \\(7 \\times 2\\) complex matrices. Any of the constrained types may also be used in arrays, as in the declaration\narray[2, 3, 4] cholesky_factor_cov[5, 6] mu;\nof a \\(2 \\times 3 \\times 4\\) array of \\(5 \\times 6\\) Cholesky factors of covariance matrices.\n\n\n\nIf x is a 1-dimensional array of length 5, then x[1] is the first element in the array and x[5] is the last. For a \\(3\n\\times 4\\) array y of two dimensions, y[1, 1] is the first element and y[3, 4] the last element. For a three-dimensional array z, the first element is z[1, 1, 1], and so on.\nSubarrays of arrays may be accessed by providing fewer than the full number of indexes. For example, suppose y is a two-dimensional array with three rows and four columns. Then y[3] is one-dimensional array of length four. This means that y[3][1] may be used instead of y[3, 1] to access the value of the first column of the third row of y. The form y[3, 1] is the preferred form (see note in this chapter).\n\n\n\nSubarrays may be manipulated and assigned just like any other variables. Similar to the behavior of matrices, Stan allows blocks such as\narray[9, 10, 11] real w;\narray[10, 11] real x;\narray[11] real y;\nreal z;\n// ...\nx = w[5];\ny = x[4];  // y == w[5][4] == w[5, 4]\nz = y[3];  // z == w[5][4][3] == w[5, 4, 3]\nComplex-valued arrays work the same way.\n\n\n\nArrays of vectors and matrices are accessed in the same way as arrays of doubles. Consider the following vector and scalar declarations.\narray[3, 4] vector[5] a;\narray[4] vector[5] b;\nvector[5] c;\nreal x;\nWith these declarations, the following assignments are legal.\nb = a[1];       // result is array of vectors\nc = a[1, 3];    // result is vector\nc = b[3];       //   same result as above\nx = a[1, 3, 5]; // result is scalar\nx = b[3, 5];    //   same result as above\nx = c[5];       //   same result as above\nRow vectors and other derived vector types (simplex and ordered) behave the same way in terms of indexing.\nConsider the following matrix, vector and scalar declarations.\narray[3, 4] matrix[6, 5] d;\narray[4] matrix[6, 5] e;\nmatrix[6, 5] f;\nrow_vector[5] g;\nreal x;\nWith these declarations, the following definitions are legal.\ne = d[1];           // result is array of matrices\nf = d[1, 3];        // result is matrix\nf = e[3];           //   same result as above\ng = d[1, 3, 2];     // result is row vector\ng = e[3, 2];        //   same result as above\ng = f[2];           //   same result as above\nx = d[1, 3, 5, 2];  // result is scalar\nx = e[3, 5, 2];     //   same result as above\nx = f[5, 2];        //   same result as above\nx = g[2];           //   same result as above\nAs shown, the result f[2] of supplying a single index to a matrix is the indexed row, here row 2 of matrix f.\n\n\n\nSubarrays of arrays may be assigned by indexing on the left-hand side of an assignment statement. For example, the following is legal.\narray[I, J, K] real x;\narray[J, K] real y;\narray[K] real z;\n// ...\nx[1] = y;\nx[1, 1] = z;\nThe sizes must match. Here, x[1] is a J by K array, as is y.\nPartial array assignment also works for arrays of matrices, vectors, and row vectors.\n\n\n\nArrays, row vectors, column vectors and matrices are not interchangeable in Stan. Thus a variable of any one of these fundamental types is not assignable to any of the others, nor may it be used as an argument where the other is required (use as arguments follows the assignment rules).\n\n\nFor example, vectors cannot be assigned to arrays or vice-versa.\narray[4] real a;\nvector[4] b;\nrow_vector[4] c;\n// ...\na = b; // illegal assignment of vector to array\nb = a; // illegal assignment of array to vector\na = c; // illegal assignment of row vector to array\nc = a; // illegal assignment of array to row vector\n\n\n\nIt is not even legal to assign row vectors to column vectors or vice versa.\nvector[4] b;\nrow_vector[4] c;\n// ...\nb = c; // illegal assignment of row vector to column vector\nc = b; // illegal assignment of column vector to row vector\n\n\n\nThe same holds for matrices, where 2-dimensional arrays may not be assigned to matrices or vice-versa.\narray[3, 4] real a;\nmatrix[3, 4] b;\n// ...\na = b;  // illegal assignment of matrix to array\nb = a;  // illegal assignment of array to matrix\n\n\n\nA \\(1 \\times N\\) matrix cannot be assigned a row vector or vice versa.\nmatrix[1, 4] a;\nrow_vector[4] b;\n// ...\na = b;  // illegal assignment of row vector to matrix\nb = a;  // illegal assignment of matrix to row vector\nSimilarly, an \\(M \\times 1\\) matrix may not be assigned to a column vector.\nmatrix[4, 1] a;\nvector[4] b;\n// ...\na = b;  // illegal assignment of column vector to matrix\nb = a;  // illegal assignment of matrix to column vector\n\n\n\n\nAn integer expression is used to pick out the sizes of arrays. The same restrictions as for vector and matrix sizes apply, namely that the size is declared with an integer-denoting expression that does not contain any parameters, transformed parameters, or generated quantities.\n\n\n\nIf any of an array’s dimensions is size zero, the entire array will be of size zero. That is, if we declare\narray[3, 0] real a;\nthen the resulting size of a is zero and querying any of its dimensions at run time will result in the value zero. Declared as above, a[1] will be a size-zero one-dimensional array. For comparison, declaring\narray[0, 3] real b;\nalso produces an array with an overall size of zero, but in this case, there is no way to index legally into b, because b[0] is undefined. The array will behave at run time as if it’s a \\(0 \\times\n0\\) array. For example, the result of to_matrix(b) will be a \\(0 \\times 0\\) matrix, not a \\(0 \\times 3\\) matrix.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#general-structure-of-copula-models-in-stan",
-    "href": "stan-users-guide/copulas.html#general-structure-of-copula-models-in-stan",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#tuple-data-type",
+    "href": "reference-manual/types.html#tuple-data-type",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "This section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.\nThe log density of a multivariate distribution using a copula can be written as:\n\\[\n\\log h(\\mathbf{x}) = \\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right) + \\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\n\\]\nwhere:\n\n\\(u_i = F_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) are the probability integral transforms of the data\n\\(\\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right)\\) is the log density of the copula\n\\(\\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) is the sum of the log densities of the marginals\n\\(\\boldsymbol{\\alpha}\\) represents the parameters describing the parametric form of the copula\n\\(\\boldsymbol{\\beta}_i\\) represents the parameters describing the parametric form of the \\(i\\)-th marginal distribution\n\nThe implementation of copulas in Stan has two key requirements:\n\nBoth the probability density functions and cumulative distribution functions of the marginal distributions must be available\nA function that computes the log density of the copula for the transformed data must be implemented\n\nMost copula implementations in Stan follow a three-step process:\n\nAccumulate marginal log likelihoods: Calculate and add the log density of each marginal distribution to the target log density\nTransform to uniform variables: Apply the marginal CDFs to transform the data to uniform variables on the unit interval\nCalculate copula density: Compute the log density of the copula based on these uniform variables and add it to the target log density\n\nThis process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.\nIn a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where \\(\\log c(\\mathbf{u}) = 0\\), resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\nMost parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).",
+    "text": "Stan supports tuples of arbitrary size. The values in a tuple can be of arbitrary type, but the component types must be declared along with the declaration of the tuple. Tuples can be manipulated as a whole, or their elements may be accessed and set individually.\n\n\nTuples are declared with the keyword tuple followed by a parenthesized sequence of types, which determine the types of the respective tuple entries. For example, a tuple with three elements may be declared as\ntuple(int, vector[3], complex) abc;\nTuples must have at least one entry, so the following declaration is illegal.\ntuple() nil;  // ILLEGAL\nTuples of length one must use a trailing comma, to align with the expression syntax.\ntuple(int,) m; // CORRECT\ntuple(int) n;  // ILLEGAL\nTuples can be assigned as a whole if their elements can be assigned individually. For example, a can be assigned to b in the following example because int can be promoted to complex.\ntuple(int, real) a;\n...\ntuple(complex, real) b = a;\nTuple types may have elements which are declared as tuples, such as the following example.\ntuple(int, tuple(real, complex)) x;\nIn this case, it would probably be simpler to use a 3-tuple type, tuple(int, real, complex).\nTuples can be declared with constraints anywhere that ordinary variables can (i.e., as top-level block variables). That means any context in which it is legal to have a declaration\nreal&lt;lower=0&gt; sigma;\nreal&lt;lower=0, upper=1&gt; theta;\nit is legal to have a tuple with constraints such as\ntuple(real&lt;lower=0&gt;, real&lt;lower=0, upper=1&gt;) sigma_theta;\n\n\n\nTuple elements may be accessed directly. For example, with our declaration of abc from the last section, Stan uses abc.1 for the first element, abc.2 for the second, and abc.3 for the third. These numbers must be integer literals (i.e., they cannot be variables), and must be within the size of the number of elements of tuples. The types of elements are as declared, so that abc.1 is of type int, abc.2 of type vector[3] and abc.3 of type complex.\n\n\n\nTuple elements can be assigned individually, allowing, e.g.,\ntuple(int, real) ab;\nab.1 = 123;\nab.2 = 12.9;\nAs with other assignments, promotions will happen if necessary (of int to real and of real to complex, along with the corresponding container type promotions).\n\n\n\nFor convenience of using values stored in tuples, Stan supports “unpacking” (or “destructuring”) of tuples in an assignment statement.\nGiven a tuple t of type tuple(T1, ..., Tn) and a sequence of assignable expressions of types v1, …, vn, where each vi has a type which is assignable from type Ti, individual elements of the tuple may be assigned to the corresponding variables in the sequence by the statement\n(v1, /*...*/, vn) = t;\nNote that the above parenthesis are required, unlike in some other languages with similar features (e.g., Python).\nThese unpacking assignments can be nested if the tuple on the right hand side contains nested tuples.\nFor example, if T is a tuple of type tuple(int, (real, real), complex), then the program\nint i;\nreal x, y;\ncomplex z;\n\n(i, (x, y), z) = T;\nAssigns the result of T.1 to i, the result of T.2.1 to x, the result of T.2.2 to y, and the result of T.3 to z.\nThe left hand side must match in size the tuple on the right. Additionally, the same variable may not appear more than once in the left hand side of an unpacking assignment.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#gaussian-copula-example",
-    "href": "stan-users-guide/copulas.html#gaussian-copula-example",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#variable-types-vs.-constraints-and-sizes",
+    "href": "reference-manual/types.html#variable-types-vs.-constraints-and-sizes",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "The Gaussian copula is constructed using the multivariate normal distribution. For a \\(D\\)-dimensional random vector \\(\\mathbf{X}\\) with marginals \\(F_i\\), the log Gaussian copula density is given by:\n\\[\n\\begin{aligned}\n\\log c(\\mathbf{u}) &=\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top (\\boldsymbol{\\Omega}^{-1} - \\mathbf{I}) \\mathbf{z} \\\\\n& =\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top \\boldsymbol{\\Omega}^{-1} \\mathbf{z} + \\frac{1}{2} \\mathbf{z}^\\top \\mathbf{z} \\\\\n&= \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\boldsymbol{\\Omega}) - \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\mathbf{I})\n\\end{aligned}\n\\]\nwhere \\(\\mathbf{z} = [\\Phi^{-1}(u_1), \\ldots, \\Phi^{-1}(u_D)]^\\top\\) are the inverse normal CDF transforms of the uniform marginals, \\(\\boldsymbol{\\Omega}\\) is the correlation matrix, and \\(\\mathbf{I}\\) is the identity matrix. The joint log density is then:\n\\[\n\\log h(\\mathbf{x}) = \\log c(F_1(x_1), \\ldots, F_D(x_D)) + \\sum_{i=1}^D \\log f_i(x_i)\n\\]\nFollowing the three-step process for implementing copulas in Stan:\n\nAccumulate marginal log likelihoods: The exponential log densities are added to the target in the line target += exponential_lpdf(y[n] | lambda)\nTransform to uniform variables: The exponential CDF transforms the data to uniform variables: exponential_cdf(y[n, d] | lambda[d])\nCalculate copula density: The transformed variables are converted to normal scale using inv_Phi and the multivariate normal log density is computed: z ~ multi_normal_cholesky(zeros, L_Omega)\n\nThe following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  int&lt;lower=0&gt; D;  // number of dimensions\n  vector&lt;lower=0&gt;[D] y[N];  // data\n}\n\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\n\nparameters {\n  // Parameters for exponential marginal distributions\n  vector&lt;lower=0&gt;[D] lambda;  // rate parameters\n  \n  // Correlation matrix for Gaussian copula\n  cholesky_factor_corr[D] L_Omega;\n}\n\nmodel {\n  // Priors\n  lambda ~ gamma(2, 1);  // prior for rate parameters\n  L_Omega ~ lkj_corr_cholesky(2);\n  \n  // Likelihood using Gaussian copula with exponential marginals\n  for (n in 1:N) {\n    // Add exponential log density to target\n    target += exponential_lpdf(y[n] | lambda);\n    \n    vector[D] z;\n    for (d in 1:D) {\n      // Transform to uniform using exponential CDF\n      real u_d = exponential_cdf(y[n, d] | lambda[d]);\n      \n      // Transform to standard normal\n      z[d] = inv_Phi(u_d);\n    }\n    // Multivariate normal log density with correlation matrix\n    z ~ multi_normal_cholesky(zeros, L_Omega);\n  }\n}\n\ngenerated quantities {\n  // Optional: Recover correlation matrix from Cholesky factor\n  matrix[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);\n}",
+    "text": "The type information associated with a variable only contains the underlying type and dimensionality of the variable.\n\n\nThe size associated with a given variable is not part of its data type. For example, declaring a variable using\narray[3] real a;\ndeclares the variable a to be an array. The fact that it was declared to have size 3 is part of its declaration, but not part of its underlying type.\n\n\nSizes are determined dynamically (at run time) and thus cannot be type-checked statically when the program is compiled. As a result, any conformance error on size will raise a run-time error. For example, trying to assign an array of size 5 to an array of size 6 will cause a run-time error. Similarly, multiplying an \\(N \\times M\\) by a \\(J \\times K\\) matrix will raise a run-time error if \\(M \\neq J\\).\n\n\n\n\nLike sizes, constraints are not treated as part of a variable’s type in Stan when it comes to the compile-time check of operations it may participate in. Anywhere Stan accepts a matrix as an argument, it will syntactically accept a correlation matrix or covariance matrix or Cholesky factor. Thus a covariance matrix may be assigned to a matrix and vice-versa.\nSimilarly, a bounded real may be assigned to an unconstrained real and vice-versa.\n\n\nFor arguments to functions, constraints are sometimes, but not always checked when the function is called. Exclusions include C++ standard library functions. All probability functions and cumulative distribution functions check that their arguments are appropriate at run time as the function is called.\n\n\n\nFor data variables, constraints are checked after the variable is read from a data file or other source. For transformed data variables, the check is done after the statements in the transformed data block have executed. Thus it is legal for intermediate values of variables to not satisfy declared constraints.\nFor parameters, constraints are enforced by the transform applied and do not need to be checked. For transformed parameters, the check is done after the statements in the transformed parameter block have executed.\nFor all blocks defining variables (transformed data, transformed parameters, generated quantities), real values are initialized to NaN and integer values are initialized to the smallest legal integer (i.e., a large absolute value negative number).\nFor generated quantities, constraints are enforced after the statements in the generated quantities block have executed.\n\n\n\n\nIn order to refer to data types, it is convenient to have a way to refer to them. The type naming notation outlined in this section is not part of the Stan programming language, but rather a convention adopted in this document to enable a concise description of a type.\nBecause size information is not part of a data type, data types will be written without size information. For instance, array[] real is the type of one-dimensional array of reals and matrix is the type of matrices. The three-dimensional integer array type is written as array[,,] int, indicating the number slots available for indexing. Similarly, array[,] vector is the type of a two-dimensional array of vectors.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#advantages-of-copulas",
-    "href": "stan-users-guide/copulas.html#advantages-of-copulas",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#variable-declaration.section",
+    "href": "reference-manual/types.html#variable-declaration.section",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "Copulas offer several advantages in statistical modeling:\n\nFlexibility: They allow combining any marginal distributions with various dependence structures. For example:\n\nModeling financial returns with heavy-tailed marginals and complex dependence structures\nCombining different types of distributions (e.g., normal and gamma) in a single model\nCapturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic\nModeling different types of tail dependence in different parts of the distribution\n\nFactorability: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.\nFor example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.\nTail dependence: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.\nUniversal Framework: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.",
+    "text": "Variables in Stan are declared by giving a type and a name. For example\nint N;\nvector[N] y;\narray[5] matrix[3, 4] A;\ndeclares a variable N that is an integer, a variable y that is a vector of length N (the previously declared variable), and a variable A, which is a length-5 array where each element is a 3 by 4 matrix.\nThe size of top-level variables in the parameters, transformed parameters, and generated quantities must remain constant across all iterations, therefore only data variables can be used in top-level size declarations.\n// illegal and will be flagged by the compiler:\ngenerated quantities {\n  int N = 10;\n  array[N] int foo;\nDepending on where the variable is declared in the Stan program, it either must or cannot have size information, and constraints are either optional or not allowed.\n// valid block variables, but not locals or function parameters\nvector&lt;lower=0&gt;[N] u;\n\n// valid as a block or local variable, but not a function parameter\narray[3] int is;\n\n// function parameters exclude sizes and cannot be constrained\nvoid pretty_print_tri_lower(matrix x) { ... }\nTop-level variables can have constraints and must include sizes for their types, as in the above examples. Local variables, like those defined inside loops or local blocks cannot be constrained, but still include sizes. Finally, variables declared as function parameters are not constrained types and exclude sizes.\nIn the following table, the leftmost column is a list of the unconstrained and undimensioned basic types; these are used as function return types and argument types. The middle column is of unconstrained types with dimensions; these are used as local variable types. The variables M and N indicate number of columns and rows, respectively. The variable K is used for square matrices, i.e., K denotes both the number of rows and columns. The rightmost column lists the corresponding constrained types. An expression of any right-hand column type may be assigned to its corresponding left-hand column basic type. At runtime, dimensions are checked for consistency for all variables; containers of any sizes may be assigned to function arguments. The constrained matrix types cov_matrix[K], corr_matrix[K], cholesky_factor_cov[K], and cholesky_factor_corr[K] are only assignable to matrices of dimensions matrix[K, K] types.\n\n\n\n\n\n\n\n\nFunction Argument (unsized)\n   Local\n(unconstrained)\n              Block\n          (constrained)\n\n\n\n\nint\nint\nint\n\n\n\n\nint&lt;lower=L&gt;\n\n\n\n\nint&lt;upper=U&gt;\n\n\n\n\nint&lt;lower=L, upper=U&gt;\n\n\n\n\nint&lt;offset=O&gt;\n\n\n\n\nint&lt;multiplier=M&gt;\n\n\n\n\nint&lt;offset=O, multiplier=M&gt;\n\n\nreal\nreal\nreal\n\n\n\n\nreal&lt;lower=L&gt;\n\n\n\n\nreal&lt;upper=U&gt;\n\n\n\n\nreal&lt;lower=L, upper=U&gt;\n\n\n\n\nreal&lt;offset=O&gt;\n\n\n\n\nreal&lt;multiplier=M&gt;\n\n\n\n\nreal&lt;offset=O, multiplier=M&gt;\n\n\ncomplex\ncomplex\ncomplex\n\n\nvector\nvector[N]\nvector[N]\n\n\n\n\nvector[N]&lt;lower=L&gt;\n\n\n\n\nvector[N]&lt;upper=U&gt;\n\n\n\n\nvector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nvector[N]&lt;offset=O&gt;\n\n\n\n\nvector[N]&lt;multiplier=M&gt;\n\n\n\n\nvector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\nordered[N]\n\n\n\n\npositive_ordered[N]\n\n\n\n\nsimplex[N]\n\n\n\n\nunit_vector[N]\n\n\n\n\nsum_to_zero_vector[N]\n\n\nrow_vector\nrow_vector[N]\nrow_vector[N]\n\n\n\n\nrow_vector[N]&lt;lower=L&gt;\n\n\n\n\nrow_vector[N]&lt;upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O&gt;\n\n\n\n\nrow_vector[N]&lt;multiplier=M&gt;\n\n\n\n\nrow_vector[N]&lt;offset=O, multiplier=M&gt;\n\n\nmatrix\nmatrix[M, N]\nmatrix[M, N]\n\n\n\n\nmatrix[M, N]&lt;lower=L&gt;\n\n\n\n\nmatrix[M, N]&lt;upper=U&gt;\n\n\n\n\nmatrix[M, N]&lt;lower=L, upper=U&gt; |\n\n\n\n\nmatrix[M, N]&lt;offset=O&gt;\n\n\n\n\nmatrix[M, N]&lt;multiplier=M&gt;\n\n\n\n\nmatrix[M, N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\ncolumn_stochastic_matrix[M, N]\n\n\n\n\nrow_stochastic_matrix[M, N]\n\n\n\n\nsum_to_zero_matrix[M, N]\n\n\n\nmatrix[K, K]\ncorr_matrix[K]\n\n\n\nmatrix[K, K]\ncov_matrix[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_corr[K]\n\n\n\nmatrix[K, K]\ncholesky_factor_cov[K]\n\n\ncomplex_vector\ncomplex_vector[M]\ncomplex_vector[M]\n\n\ncomplex_row_vector\ncomplex_row_vector[N]\ncomplex_row_vector[N]\n\n\ncomplex_matrix\ncomplex_matrix[M, N]\ncomplex_matrix[M,N]\n\n\narray[] vector\narray[M] vector[N]\narray[M] vector[N]\n\n\n\n\narray[M] vector[N]&lt;lower=L&gt;\n\n\n\n\narray[M] vector[N]&lt;upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;lower=L, upper=U&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O&gt;\n\n\n\n\narray[M] vector[N]&lt;multiplier=M&gt;\n\n\n\n\narray[M] vector[N]&lt;offset=O, multiplier=M&gt;\n\n\n\n\narray[M] ordered[N]\n\n\n\n\narray[M] positive_ordered[N]\n\n\n\n\narray[M] simplex[N]\n\n\n\n\narray[M] unit_vector[N]\n\n\n\n\narray[M] sum_to_zero_vector[N]\n\n\n\n\nAdditional array types follow the same basic template as the final example in the table and can contain any of the previous types. The unsized version of arrays with more than one dimension is specified by using commas, e.g. array[ , ] is a 2-D array.\nFor more on how function arguments and return types are declared, consult the User’s Guide chapter on functions.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#common-pitfalls-and-considerations",
-    "href": "stan-users-guide/copulas.html#common-pitfalls-and-considerations",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#compound-variable-declaration-and-definition",
+    "href": "reference-manual/types.html#compound-variable-declaration-and-definition",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "When implementing copulas in Stan, several considerations should be kept in mind:\n\nComputational efficiency: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.\nParameter identifiability: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.\nModel selection: The choice of copula family should be guided by the specific dependence structure of the data. For example:\n\nThe Gaussian copula may underestimate the probability of joint extreme events in financial data\nThe Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications\nArchimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions\n\nNumerical stability: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.\nSymmetry considerations: Many copula families exhibit strong symmetries that may not match the data:\n\nRadial symmetry: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns\nExchangeability: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.\n\nTail dependence: Understanding and choosing appropriate tail dependence is crucial:\n\nThe upper (lower) tail dependence coefficient \\(\\lambda_U (\\lambda_L)\\) is the probability that one variable is extremely large (small) given that another is extremely large (small).\nDifferent copula families exhibit different tail dependence properties:\n\nSome copulas (like Gaussian) have zero tail dependence\nOthers can model symmetric tail dependence (\\(\\lambda_U = \\lambda_L\\))\nSome can capture asymmetric tail dependence (\\(\\lambda_U \\neq \\lambda_L\\))\nCertain copulas allow for tail dependence even with zero correlation\n\nThe choice of copula should be guided by the expected tail behavior in the application:\n\nFinancial data often requires modeling joint lower extreme events\nRisk management applications may need asymmetric tail dependence\nSome applications may require different tail behavior in different parts of the distribution\n\n\nHigh-dimensional modeling: As dimensionality increases:\n\nThe number of dependence parameters grows\nSome copula families become less flexible\nVine copulas or factor copulas may be more appropriate",
+    "text": "Stan allows assignable variables to be declared and defined in a single statement. Assignable variables are\n\nlocal variables, and\nvariables declared in the transformed data, transformed parameters, or generated quantities blocks.\n\nFor example, the statement\nint N = 5;\ndeclares the variable N to be an integer scalar type and at the same time defines it to be the value of the expression 5.\n\n\nThe type of the expression on the right-hand side of the assignment must be assignable to the type of the variable being declared. For example, it is legal to have\nreal sum = 0;\neven though 0 is of type int and sum is of type real, because integer-typed scalar expressions can be assigned to real-valued scalar variables. In all other cases, the type of the expression on the right-hand side of the assignment must be identical to the type of the variable being declared.\nVariables of any type may have values assigned to them. For example,\nmatrix[3, 2] a = b;\ndeclares a \\(3 \\times 2\\) matrix variable a and assigns a copy of the value of b to the variable a. The variable b must be of type matrix for the statement to be well formed. For the code to execute successfully, b must be the same shape as a, but this cannot be validated until run time. Because a copy is assigned, subsequent changes to a do not affect b and subsequent changes to b do not affect a.\n\n\n\nThe right-hand side may be any expression which has a type which is assignable to the variable being declared. For example,\nmatrix[3, 2] a = 0.5 * (b + c);\nassigns the matrix variable a to half of the sum of b and c. The only requirement on b and c is that the expression b + c be of type matrix. For example, b could be of type matrix and c of type real, because adding a matrix to a scalar produces a matrix, and the multiplying by a scalar produces another matrix.\nSimilarly,\ncomplex z = 2 + 3i;\nassigns the the complex number \\(2 + 3i\\) to the complex scalar z. The right-hand side expression can be a call to a user defined function, allowing general algorithms to be applied that might not be otherwise expressible as simple expressions (e.g., iterative or recursive algorithms).\n\n\n\nAny variable that is in scope and any function that is available in the block in which the compound declaration and definition appears may be used in the expression on the right-hand side of the compound declaration and definition statement.",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#common-copula-families",
-    "href": "stan-users-guide/copulas.html#common-copula-families",
-    "title": "Copulas",
+    "objectID": "reference-manual/types.html#declaring-multiple-variables-at-once",
+    "href": "reference-manual/types.html#declaring-multiple-variables-at-once",
+    "title": "Data Types and Declarations",
     "section": "",
-    "text": "Several copula families are available for modeling different dependence structures in the correlation component:\n\nGaussian copula: Based on the multivariate normal distribution, offering symmetric dependence\nStudent-t copula: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula\nArchimedean copulas: A class of copulas defined through generator functions, including:\n\nClayton copula: Stronger lower tail dependence\nGumbel copula: Stronger upper tail dependence\nFrank copula: Symmetric dependence\n\nVine copulas: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas",
+    "text": "Stan will interpret multiple comma-separated variable names following a single type as declaring multiple new variables. This is available for all variable declarations in all blocks.\n\n\nThe code:\nreal x, y;\nis equivalent to\nreal x;\nreal y;\nAs a result, all declarations on the same line must be of the same type.\n\n\n\nThe ability to declare multiple variables can be combined with assignments whenever a declare-define is valid, as documented in the section introducing compound declarations and definitions :\nreal x = 3, y = 5.6;\nConstrained data types can also be declared together, so long as the constraint for each variable is the same:\nreal&lt;lower=0&gt; x, y;",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/copulas.html#further-readingviewing",
-    "href": "stan-users-guide/copulas.html#further-readingviewing",
-    "title": "Copulas",
-    "section": "",
-    "text": "Jónsson’s three part blog series, Copulas in Stan:\n\nPart I: If it bleeds, we can kill it\nPart II: A gentle introduction: the Gaussian copula\nPart III: It was the best of tails, it was the worst of tails: The T-Copula\n\nBrynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, Copulas in Stan: Modeling Spatial Dependence\nSean Pinkney’s Helpful Stan functions: copula functions",
+    "objectID": "reference-manual/types.html#footnotes",
+    "href": "reference-manual/types.html#footnotes",
+    "title": "Data Types and Declarations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nStan compiles integers to int and reals to double types in C++. Precise details of rounding will depend on the compiler and hardware architecture on which the code is run.↩︎",
     "crumbs": [
-      "Stan Users Guide",
-      "Example Models",
-      "Copulas"
+      "Reference Manual",
+      "Language",
+      "Data Types and Declarations"
     ]
   },
   {
-    "objectID": "stan-users-guide/custom-probability.html",
-    "href": "stan-users-guide/custom-probability.html",
-    "title": "Custom Probability Functions",
+    "objectID": "reference-manual/variational.html",
+    "href": "reference-manual/variational.html",
+    "title": "Variational Inference",
     "section": "",
-    "text": "Custom distributions may also be implemented directly within Stan’s programming language. The only thing that is needed is to increment the total log probability. The rest of the chapter provides examples.\n\n\n\n\nA simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}\\) are the bounds, with \\(\\alpha &lt; \\beta\\), then \\(y \\in (\\alpha,\\beta)\\) has a density defined as \\[\n\\textsf{triangle}(y \\mid \\alpha,\\beta)\n=\n\\frac{1}{(\\beta - \\alpha)^2}\n\\cdot\n\\textrm{min}(y - \\alpha, \\beta - y).\n\\]\nThe general form of triangle can be coded in Stan as follows.\ndata {\n  real alpha;\n  real&lt;lower=alpha&gt; beta;\n}\nparameters {\n  real&lt;lower=alpha, upper=beta&gt; y;\n}\nmodel {\n  target += -2 * log(beta - alpha)\n    + log(fmin(y - alpha, beta - y));\n}\nBecause the bounds are specified as data here, the term -2 * log(beta - alpha) could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.\nIf \\(\\alpha = -1\\), \\(\\beta = 1\\), and \\(y \\in (-1,1)\\), then fmin(y - alpha, beta - y) is fmin(y + 1, -1 - y), which is fmin(y + 1, -(y + 1)), which reduces to 1 - abs(y). Therfore, the density, dropping constants, reduces to \\[\n\\textsf{triangle}(y \\mid -1, 1) \\propto 1 - |y|.\n\\] Consider the following Stan implementation of \\(\\textsf{triangle}(-1,1)\\) for sampling.\nparameters {\n  real&lt;lower=-1, upper=1&gt; y;\n}\nmodel {\n  target += log1m(abs(y));\n}\nThe single scalar parameter y is declared as lying in the interval (-1, 1). The total log probability is incremented with the joint log probability of all parameters, i.e., \\(\\log \\mathsf{Triangle}(y \\mid -1, 1)\\). This value is coded in Stan as log1m(abs(y)). The function log1m is defined so that log1m(x) has the same value as \\(\\log(1 - x)\\), but the computation is faster, more accurate, and more stable.\nThe constrained type real&lt;lower=-1, upper=1&gt; declared for y is critical for correct sampling behavior. If the constraint on y is removed from the program, say by declaring y as having the unconstrained scalar type real, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of y outside of \\((-1,1)\\).\nNow suppose the log probability function were extended to all of \\(\\mathbb{R}\\) as follows by defining the probability to be log(0.0), i.e., \\(-\\infty\\), for values outside of \\((-1, 1)\\).\ntarget += log(fmax(0.0,1 - abs(y)));\nWith the constraint on y in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on y is removed, the model will compile and run without arithmetic errors, but will not sample properly.1\n\n\n\nIf Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where lambda is the inverse scale and y the sampled variate.\ntarget += log(lambda) - y * lambda;\nThis encoding will work for any lambda and y; they can be parameters, data, or one of each, or even local variables.\nThe assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.\ny ~ exponential(lambda);\nThere are two notable differences. First, the distribution statement will check the inputs to make sure both lambda is positive and y is non-negative (which includes checking that neither is the special not-a-number value).\nThe second difference is that if lambda is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the log(lambda) term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If lambda and y are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).\n\n\n\nFor another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation rho. That is, it computes \\[\n\\texttt{binormal}\\mathtt{\\_}\\texttt{cdf}(z_1, z_2, \\rho) = \\Pr[Z_1 \\leq z_1 \\text{ and } Z_2 \\leq z_2]\n\\] where the random 2-vector \\(Z\\) has the distribution \\[\nZ \\sim \\textsf{multivariate normal}\\left(\n\\begin{bmatrix}\n0 \\\\\n0\n\\end{bmatrix}, \\\n\\begin{bmatrix}\n1 & \\rho\n\\\\\n\\rho & 1\n\\end{bmatrix}\n\\right).\n\\]\nThe following Stan program implements this function,\nreal binormal_cdf(tuple(real, real) z, real rho) {\n  real z1 = z.1;\n  real z2 = z.2;\n  if (z1 == 0 && z2 == 0) {\n    return 0.25 + asin(rho) / (2 * pi());\n  }\n  real denom = sqrt((1 + rho) * (1 - rho));\n  real term1 = z1 == 0\n    ? (z2 &gt; 0 ? 0.25 : -0.25)\n    :  owens_t(z1, (z2 / z1 - rho) / denom);\n  real term2 = z2 == 0\n    ? (z1 &gt; 0 ? 0.25 : -0.25)\n    : owens_t(z2, (z1 / z2 - rho) / denom);\n  real z1z2 = z1 * z2;\n  real delta = z1z2 &lt; 0 || (z1z2 == 0 && (z1 + z2) &lt; 0);\n  return 0.5 * (Phi(z1) + Phi(z2) - delta) - term1 - term2;\n}\nIt is written using a tuple argument so that it may be called as binormal_cdf((z1, z2) | rho). The two ternary operators defining term1 and term2 are derived by taking the limit of the owens_t function when the second argument goes to infinity.\nOne way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.\ntransformed data {\n  for (zzr in {[0, 0, 0.5],\n               [0, 1, 0.5],\n               [1, 0, -0.2],\n               [1, 3, 0.9]}) {\n    real z1  = zzr[1];\n    real z2 = zzr[2];\n    real rho = zzr[3];\n    print(\"binomial_cdf((\",  z1, \", \", z2, \") | \", rho, \")\",\n          \"=\", binormal_cdf((z1, z2) | rho));\n  }\n}\nIn this case, we verified that the results match those of the pbivnorm package in R.",
+    "text": "Stan implements an automatic variational inference algorithm, called Automatic Differentiation Variational Inference (ADVI) Kucukelbir et al. (2017). In this chapter, we describe the specifics of how ADVI maximizes the variational objective.\n\n\nADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.\n\n\nADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by elbo_samples. We recommend a default value of \\(100\\), as we only evaluate the ELBO every eval_elbo iterations, which also defaults to \\(100\\).\n\n\n\nADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by grad_samples. We recommend a default value of \\(1\\), as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.\n\n\n\nADVI uses a finite-memory version of adaGrad Duchi, Hazan, and Singer (2011). This has a single parameter that we expose, denoted eta. We now have a warmup adaptation phase that selects a good value for eta. The procedure does a heuristic search over eta values that span 5 orders of magnitude.\n\n\n\nADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by tol_rel_obj, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.",
     "crumbs": [
-      "Stan Users Guide",
-      "Programming Techniques",
-      "Custom Probability Functions"
+      "Reference Manual",
+      "Algorithms",
+      "Variational Inference"
     ]
   },
   {
-    "objectID": "stan-users-guide/custom-probability.html#examples",
-    "href": "stan-users-guide/custom-probability.html#examples",
-    "title": "Custom Probability Functions",
+    "objectID": "reference-manual/variational.html#stochastic-gradient-ascent",
+    "href": "reference-manual/variational.html#stochastic-gradient-ascent",
+    "title": "Variational Inference",
     "section": "",
-    "text": "A simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}\\) are the bounds, with \\(\\alpha &lt; \\beta\\), then \\(y \\in (\\alpha,\\beta)\\) has a density defined as \\[\n\\textsf{triangle}(y \\mid \\alpha,\\beta)\n=\n\\frac{1}{(\\beta - \\alpha)^2}\n\\cdot\n\\textrm{min}(y - \\alpha, \\beta - y).\n\\]\nThe general form of triangle can be coded in Stan as follows.\ndata {\n  real alpha;\n  real&lt;lower=alpha&gt; beta;\n}\nparameters {\n  real&lt;lower=alpha, upper=beta&gt; y;\n}\nmodel {\n  target += -2 * log(beta - alpha)\n    + log(fmin(y - alpha, beta - y));\n}\nBecause the bounds are specified as data here, the term -2 * log(beta - alpha) could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.\nIf \\(\\alpha = -1\\), \\(\\beta = 1\\), and \\(y \\in (-1,1)\\), then fmin(y - alpha, beta - y) is fmin(y + 1, -1 - y), which is fmin(y + 1, -(y + 1)), which reduces to 1 - abs(y). Therfore, the density, dropping constants, reduces to \\[\n\\textsf{triangle}(y \\mid -1, 1) \\propto 1 - |y|.\n\\] Consider the following Stan implementation of \\(\\textsf{triangle}(-1,1)\\) for sampling.\nparameters {\n  real&lt;lower=-1, upper=1&gt; y;\n}\nmodel {\n  target += log1m(abs(y));\n}\nThe single scalar parameter y is declared as lying in the interval (-1, 1). The total log probability is incremented with the joint log probability of all parameters, i.e., \\(\\log \\mathsf{Triangle}(y \\mid -1, 1)\\). This value is coded in Stan as log1m(abs(y)). The function log1m is defined so that log1m(x) has the same value as \\(\\log(1 - x)\\), but the computation is faster, more accurate, and more stable.\nThe constrained type real&lt;lower=-1, upper=1&gt; declared for y is critical for correct sampling behavior. If the constraint on y is removed from the program, say by declaring y as having the unconstrained scalar type real, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of y outside of \\((-1,1)\\).\nNow suppose the log probability function were extended to all of \\(\\mathbb{R}\\) as follows by defining the probability to be log(0.0), i.e., \\(-\\infty\\), for values outside of \\((-1, 1)\\).\ntarget += log(fmax(0.0,1 - abs(y)));\nWith the constraint on y in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on y is removed, the model will compile and run without arithmetic errors, but will not sample properly.1\n\n\n\nIf Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where lambda is the inverse scale and y the sampled variate.\ntarget += log(lambda) - y * lambda;\nThis encoding will work for any lambda and y; they can be parameters, data, or one of each, or even local variables.\nThe assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.\ny ~ exponential(lambda);\nThere are two notable differences. First, the distribution statement will check the inputs to make sure both lambda is positive and y is non-negative (which includes checking that neither is the special not-a-number value).\nThe second difference is that if lambda is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the log(lambda) term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If lambda and y are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).\n\n\n\nFor another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation rho. That is, it computes \\[\n\\texttt{binormal}\\mathtt{\\_}\\texttt{cdf}(z_1, z_2, \\rho) = \\Pr[Z_1 \\leq z_1 \\text{ and } Z_2 \\leq z_2]\n\\] where the random 2-vector \\(Z\\) has the distribution \\[\nZ \\sim \\textsf{multivariate normal}\\left(\n\\begin{bmatrix}\n0 \\\\\n0\n\\end{bmatrix}, \\\n\\begin{bmatrix}\n1 & \\rho\n\\\\\n\\rho & 1\n\\end{bmatrix}\n\\right).\n\\]\nThe following Stan program implements this function,\nreal binormal_cdf(tuple(real, real) z, real rho) {\n  real z1 = z.1;\n  real z2 = z.2;\n  if (z1 == 0 && z2 == 0) {\n    return 0.25 + asin(rho) / (2 * pi());\n  }\n  real denom = sqrt((1 + rho) * (1 - rho));\n  real term1 = z1 == 0\n    ? (z2 &gt; 0 ? 0.25 : -0.25)\n    :  owens_t(z1, (z2 / z1 - rho) / denom);\n  real term2 = z2 == 0\n    ? (z1 &gt; 0 ? 0.25 : -0.25)\n    : owens_t(z2, (z1 / z2 - rho) / denom);\n  real z1z2 = z1 * z2;\n  real delta = z1z2 &lt; 0 || (z1z2 == 0 && (z1 + z2) &lt; 0);\n  return 0.5 * (Phi(z1) + Phi(z2) - delta) - term1 - term2;\n}\nIt is written using a tuple argument so that it may be called as binormal_cdf((z1, z2) | rho). The two ternary operators defining term1 and term2 are derived by taking the limit of the owens_t function when the second argument goes to infinity.\nOne way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.\ntransformed data {\n  for (zzr in {[0, 0, 0.5],\n               [0, 1, 0.5],\n               [1, 0, -0.2],\n               [1, 3, 0.9]}) {\n    real z1  = zzr[1];\n    real z2 = zzr[2];\n    real rho = zzr[3];\n    print(\"binomial_cdf((\",  z1, \", \", z2, \") | \", rho, \")\",\n          \"=\", binormal_cdf((z1, z2) | rho));\n  }\n}\nIn this case, we verified that the results match those of the pbivnorm package in R.",
+    "text": "ADVI optimizes the ELBO in the real-coordinate space using stochastic gradient ascent. We obtain noisy (yet unbiased) gradients of the variational objective using automatic differentiation and Monte Carlo integration. The algorithm ascends these gradients using an adaptive stepsize sequence. We evaluate the ELBO also using Monte Carlo integration and measure convergence similar to the relative tolerance scheme in Stan’s optimization feature.\n\n\nADVI uses Monte Carlo integration to approximate the variational objective function, the ELBO. The number of draws used to approximate the ELBO is denoted by elbo_samples. We recommend a default value of \\(100\\), as we only evaluate the ELBO every eval_elbo iterations, which also defaults to \\(100\\).\n\n\n\nADVI uses Monte Carlo integration to approximate the gradients of the ELBO. The number of draws used to approximate the gradients is denoted by grad_samples. We recommend a default value of \\(1\\), as this is the most efficient. It also a very noisy estimate of the gradient, but stochastic gradient ascent is capable of following such gradients.\n\n\n\nADVI uses a finite-memory version of adaGrad Duchi, Hazan, and Singer (2011). This has a single parameter that we expose, denoted eta. We now have a warmup adaptation phase that selects a good value for eta. The procedure does a heuristic search over eta values that span 5 orders of magnitude.\n\n\n\nADVI tracks the progression of the ELBO through the stochastic optimization. Specifically, ADVI heuristically determines a rolling window over which it computes the average and the median change of the ELBO. Should either number fall below a threshold, denoted by tol_rel_obj, we consider the algorithm to have converged. The change in ELBO is calculated the same way as in Stan’s optimization module.",
     "crumbs": [
-      "Stan Users Guide",
-      "Programming Techniques",
-      "Custom Probability Functions"
+      "Reference Manual",
+      "Algorithms",
+      "Variational Inference"
     ]
   },
   {
-    "objectID": "stan-users-guide/custom-probability.html#footnotes",
-    "href": "stan-users-guide/custom-probability.html#footnotes",
-    "title": "Custom Probability Functions",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe problem is the (extremely!) light tails of the triangle distribution. The standard HMC and NUTS samplers can’t get into the corners of the triangle properly. Because the Stan code declares y to be of type real&lt;lower=-1, upper=1&gt;, the inverse logit transform is applied to the unconstrained variable and its log absolute derivative added to the log probability. The resulting distribution on the logit-transformed y is well behaved.↩︎",
+    "objectID": "stan-users-guide/algebraic-equations.html",
+    "href": "stan-users-guide/algebraic-equations.html",
+    "title": "Solving Algebraic Equations",
+    "section": "",
+    "text": "Stan provides a built-in mechanism for specifying systems of algebraic equations. These systems can be solved either with the Newton method, as implemented in the Kinsol package (Hindmarsh et al. 2005), or with the Powell hybrid method (Powell 1970). The function signatures for Stan’s algebraic solvers are fully described in the algebraic solver section of the reference manual.\nSolving any system of algebraic equations can be translated into a root-finding problem, that is, given a function \\(f\\), we wish to find \\(y\\) such that \\(f(y) = 0\\).\n\n\nFor systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.\nAs an illustrative example, we consider the following nonlinear system of two equations with two unknowns: \\[\\begin{align*}\nz_1 &= y_1 - \\theta_1 \\\\\nz_2 &= y_1 y_2 + \\theta_2\n\\end{align*}\\]\nOur goal is to simultaneously solve all equations for \\(y_1\\) and \\(y_2\\), such that the vector \\(z\\) goes to 0.\n\n\n\nA system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the user-defined functions section for more information on coding user-defined functions).\nvector system(vector y,              // unknowns\n              vector theta,          // parameters\n              data array[] real x_r, // data (real)\n              array[] int x_i) {     // data (integer)\n  vector[2] z;\n  z[1] = y[1] - theta[1];\n  z[2] = y[1] * y[2] - theta[2];\n  return z;\n}\nThe function takes the unknowns we wish to solve for in y (a vector), the system parameters in theta (a vector), the real data in x_r (a real array) and the integer data in x_i (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.\nThe body of the system function here could also be coded using a row vector constructor and transposition,\nreturn [ y[1] - theta[1],\n         y[1] * y[2] - theta[2] ]';\nAs systems get more complicated, naming the intermediate expressions goes a long way toward readability.\n\n\nThe function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.\n\n\n\n\nLet’s suppose \\(\\theta = (3, 6)\\). To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.\nThe solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about algebraic solvers with control parameters.\nThe following code returns the solution to our nonlinear algebraic system:\ntransformed data {\n  vector[2] y_guess = [1, 1]';\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\ntransformed parameters {\n  vector[2] theta = [3, 6]';\n  vector[2] y;\n\n  y = solve_newton(system, y_guess, theta, x_r, x_i);\n}\nwhich returns \\(y = (3, -2)\\).\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, y_guess, which may be a data or a parameter vector.\n\n\n\nThe Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function \\(f\\) with respect to the unknowns \\(x\\) must be invertible. This requires the Jacobian to be square, meaning \\(f(y)\\) and \\(y\\) have the same length or, in other words the number of equations in the system is the same as the number of unknowns.\n\n\n\nCertain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.\nOn the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.\n\n\n\n\nThe call to the algebraic solver shown previously uses the default control settings. The _tol variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.\ny = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,\n                     theta, x_r, x_i);\nFor the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.\nThe default values for the control arguments are respectively scaling_step = 1e-3 (\\(10^{-3}\\)), rel_tol = 1e-10 (\\(10^{-10}\\)), f_tol = 1e-6 (\\(10^{-6}\\)), and max_steps = 200 (\\(200\\)).\n\n\nThe relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.\nSmaller relative tolerances produce more accurate solutions but require more computational time.\n\n\nThe tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by Timonen et al. (2023).\n\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Custom Probability Functions"
+      "Example Models",
+      "Solving Algebraic Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/decision-analysis.html",
-    "href": "stan-users-guide/decision-analysis.html",
-    "title": "Decision Analysis",
+    "objectID": "stan-users-guide/algebraic-equations.html#example-system-of-nonlinear-algebraic-equations",
+    "href": "stan-users-guide/algebraic-equations.html#example-system-of-nonlinear-algebraic-equations",
+    "title": "Solving Algebraic Equations",
     "section": "",
-    "text": "Statistical decision analysis is about making decisions under uncertainty. In order to make decisions, outcomes must have some notion of “utility” associated with them. The so-called “Bayes optimal” decision is the one that maximizes expected utility (or equivalently, minimizes expected loss). This chapter shows how Stan can be used to simultaneously estimate the distribution of outcomes based on decisions and compute the required expected utilities.\n\n\nFollowing Gelman et al. (2013), Bayesian decision analysis can be factored into the following four steps.\n\nDefine a set \\(X\\) of possible outcomes and a set \\(D\\) of possible decisions.\nDefine a probability distribution of outcomes conditional on decisions through a conditional density function \\(p(x \\mid d)\\) for \\(x \\in X\\) and \\(d \\in D.\\)\nDefine a utility function \\(U : X \\rightarrow \\mathbb{R}\\) mapping outcomes to their utility.\nChoose action \\(d^* \\in D\\) with highest expected utility, \\[\nd^* = \\textrm{arg max}_d \\ \\mathbb{E}[U(x) \\mid d].\n\\]\n\nThe outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).\n\n\n\nThis section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.\n\n\nA decision consists of the choice of commute mode and the outcome is a time and cost. More formally,\n\nthe set of decisions is \\(D = 1:4\\), corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and\nthe set of outcomes \\(X = \\mathbb{R} \\times \\mathbb{R}_+\\) contains pairs of numbers \\(x = (c, t)\\) consisting of a cost \\(c\\) and time \\(t \\geq 0\\).\n\n\n\n\nThe density required is \\(p(x \\mid d),\\) where \\(d \\in D\\) is a decision and \\(x = (c, t) \\in X\\) is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters \\(\\theta,\\) \\[\np(x \\mid d, x^{\\textrm{obs}}, d^{\\textrm{obs}})\n=\n\\int\n  p(x \\mid d, \\theta)\n  \\cdot p(\\theta \\mid x^{\\textrm{obs}}, d^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta.\n\\] The observed data for a year of commutes consists of choice of the chosen commute mode \\(d^{\\textrm{obs}}_n\\) and observed costs and times \\(x^{\\textrm{obs}}_n = (c^{\\textrm{obs}}_n, t^{\\textrm{obs}}_n)\\) for \\(n\n\\in 1:200.\\)\nFor simplicity, commute time \\(t_n\\) for trip \\(n\\) will be modeled as lognormal for a given choice of transportation \\(d_n \\in 1:4,\\) \\[\nt_n \\sim \\textrm{lognormal}(\\mu_{d[n]}, \\sigma_{d[n]}).\n\\] To understand the notation, \\(d_n\\), also written \\(d[n]\\), is the mode of transportation used for trip \\(n\\). For example if trip \\(n\\) was by bicycle, then \\(t_n \\sim \\textrm{lognormal}(\\mu_2, \\sigma_2),\\) where \\(\\mu_2\\) and \\(\\sigma_2\\) are the lognormal parameters for bicycling.\nSimple fixed priors are used for each mode of transportation \\(k \\in 1:4,\\) \\[\\begin{eqnarray*}\n\\mu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\sigma_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.\nCost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, \\[\nc_n \\sim \\textrm{lognormal}(\\nu_{d[n]}, \\tau_{d[n]}).\n\\] Again, the priors are fixed for the modes of transportation, \\[\\begin{eqnarray*}\n\\nu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\tau_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.\nThe full set of parameters that are marginalized in the posterior predictive distribution is \\[\n\\theta = (\\mu_{1:4}, \\sigma_{1:4}, \\nu_{1:4}, \\tau_{1:4}).\n\\]\n\n\n\nFor the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as\n\\[\nU(c, t) = -(c + 25 \\cdot t)\n\\]\nThe sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.\n\n\n\nAt this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.\nfunctions {\n  real U(real c, real t) {\n    return -(c + 25 * t);\n  }\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; d;\n  array[N] real c;\n  array[N] real&lt;lower=0&gt; t;\n}\nparameters {\n  vector[4] mu;\n  vector&lt;lower=0&gt;[4] sigma;\n  array[4] real nu;\n  array[4] real&lt;lower=0&gt; tau;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.25);\n  nu ~ normal(0, 20);\n  tau ~ lognormal(0, 0.25);\n  t ~ lognormal(mu[d], sigma[d]);\n  c ~ lognormal(nu[d], tau[d]);\n}\ngenerated quantities {\n  array[4] real util;\n  for (k in 1:4) {\n    util[k] = U(lognormal_rng(nu[k], tau[k]),\n                lognormal_rng(mu[k], sigma[k]));\n  }\n}\nThe generated quantities block defines an array variable util where util[k], which will hold the utility derived from a random commute for choice k generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.\nFor simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, nu and tau would be declared as univariate, and the RNG for cost would only be employed when k == 4.\nDefining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.\nAll that is left is to run Stan. The posterior mean for util[k] is the expected utility, which written out with full conditioning, is \\[\\begin{eqnarray*}\n\\mathbb{E}\\!\\left[U(x) \\mid d = k, d^{\\textrm{obs}}, x^{\\textrm{obs}}\\right]\n& = &\n\\int\n  U(x)\n  \\cdot p(x \\mid d = k, \\theta)\n  \\cdot p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M  U(x^{(m)} ),\n\\end{eqnarray*}\\] where \\[\nx^{(m)} \\sim p(x \\mid d = k, \\theta^{(m)} )\n\\] and \\[\n\\theta^{(m)}\n\\sim p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}}).\n\\]\nIn terms of Stan’s execution, the random generation of \\(x^{(m)}\\) is carried out with the lognormal_rng operations after \\(\\theta^{(m)}\\) is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.\nIt only remains to make the decision k with highest expected utility, which will correspond to the choice with the highest posterior mean for util[k]. This can be read off of the mean column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.\n\n\n\n\nMany choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.",
+    "text": "For systems of linear algebraic equations, we recommend solving the system using matrix division. The algebraic solver becomes handy when we want to solve nonlinear equations.\nAs an illustrative example, we consider the following nonlinear system of two equations with two unknowns: \\[\\begin{align*}\nz_1 &= y_1 - \\theta_1 \\\\\nz_2 &= y_1 y_2 + \\theta_2\n\\end{align*}\\]\nOur goal is to simultaneously solve all equations for \\(y_1\\) and \\(y_2\\), such that the vector \\(z\\) goes to 0.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Decision Analysis"
+      "Example Models",
+      "Solving Algebraic Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/decision-analysis.html#outline-of-decision-analysis",
-    "href": "stan-users-guide/decision-analysis.html#outline-of-decision-analysis",
-    "title": "Decision Analysis",
+    "objectID": "stan-users-guide/algebraic-equations.html#coding-an-algebraic-system",
+    "href": "stan-users-guide/algebraic-equations.html#coding-an-algebraic-system",
+    "title": "Solving Algebraic Equations",
     "section": "",
-    "text": "Following Gelman et al. (2013), Bayesian decision analysis can be factored into the following four steps.\n\nDefine a set \\(X\\) of possible outcomes and a set \\(D\\) of possible decisions.\nDefine a probability distribution of outcomes conditional on decisions through a conditional density function \\(p(x \\mid d)\\) for \\(x \\in X\\) and \\(d \\in D.\\)\nDefine a utility function \\(U : X \\rightarrow \\mathbb{R}\\) mapping outcomes to their utility.\nChoose action \\(d^* \\in D\\) with highest expected utility, \\[\nd^* = \\textrm{arg max}_d \\ \\mathbb{E}[U(x) \\mid d].\n\\]\n\nThe outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).",
+    "text": "A system of algebraic equations is coded directly in Stan as a function with a strictly specified signature. For example, the nonlinear system given above can be coded using the following function in Stan (see the user-defined functions section for more information on coding user-defined functions).\nvector system(vector y,              // unknowns\n              vector theta,          // parameters\n              data array[] real x_r, // data (real)\n              array[] int x_i) {     // data (integer)\n  vector[2] z;\n  z[1] = y[1] - theta[1];\n  z[2] = y[1] * y[2] - theta[2];\n  return z;\n}\nThe function takes the unknowns we wish to solve for in y (a vector), the system parameters in theta (a vector), the real data in x_r (a real array) and the integer data in x_i (an integer array). The system function returns the value of the function (a vector), for which we want to compute the roots. Our example does not use real or integer data. Nevertheless, these unused arguments must be included in the system function with exactly the signature above.\nThe body of the system function here could also be coded using a row vector constructor and transposition,\nreturn [ y[1] - theta[1],\n         y[1] * y[2] - theta[2] ]';\nAs systems get more complicated, naming the intermediate expressions goes a long way toward readability.\n\n\nThe function defining the system must have exactly these argument types and return type. This may require passing in zero-length arrays for data or a zero-length vector for parameters if the system does not involve data or parameters.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Decision Analysis"
+      "Example Models",
+      "Solving Algebraic Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/decision-analysis.html#example-decision-analysis",
-    "href": "stan-users-guide/decision-analysis.html#example-decision-analysis",
-    "title": "Decision Analysis",
+    "objectID": "stan-users-guide/algebraic-equations.html#calling-the-algebraic-solver",
+    "href": "stan-users-guide/algebraic-equations.html#calling-the-algebraic-solver",
+    "title": "Solving Algebraic Equations",
     "section": "",
-    "text": "This section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.\n\n\nA decision consists of the choice of commute mode and the outcome is a time and cost. More formally,\n\nthe set of decisions is \\(D = 1:4\\), corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and\nthe set of outcomes \\(X = \\mathbb{R} \\times \\mathbb{R}_+\\) contains pairs of numbers \\(x = (c, t)\\) consisting of a cost \\(c\\) and time \\(t \\geq 0\\).\n\n\n\n\nThe density required is \\(p(x \\mid d),\\) where \\(d \\in D\\) is a decision and \\(x = (c, t) \\in X\\) is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters \\(\\theta,\\) \\[\np(x \\mid d, x^{\\textrm{obs}}, d^{\\textrm{obs}})\n=\n\\int\n  p(x \\mid d, \\theta)\n  \\cdot p(\\theta \\mid x^{\\textrm{obs}}, d^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta.\n\\] The observed data for a year of commutes consists of choice of the chosen commute mode \\(d^{\\textrm{obs}}_n\\) and observed costs and times \\(x^{\\textrm{obs}}_n = (c^{\\textrm{obs}}_n, t^{\\textrm{obs}}_n)\\) for \\(n\n\\in 1:200.\\)\nFor simplicity, commute time \\(t_n\\) for trip \\(n\\) will be modeled as lognormal for a given choice of transportation \\(d_n \\in 1:4,\\) \\[\nt_n \\sim \\textrm{lognormal}(\\mu_{d[n]}, \\sigma_{d[n]}).\n\\] To understand the notation, \\(d_n\\), also written \\(d[n]\\), is the mode of transportation used for trip \\(n\\). For example if trip \\(n\\) was by bicycle, then \\(t_n \\sim \\textrm{lognormal}(\\mu_2, \\sigma_2),\\) where \\(\\mu_2\\) and \\(\\sigma_2\\) are the lognormal parameters for bicycling.\nSimple fixed priors are used for each mode of transportation \\(k \\in 1:4,\\) \\[\\begin{eqnarray*}\n\\mu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\sigma_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.\nCost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, \\[\nc_n \\sim \\textrm{lognormal}(\\nu_{d[n]}, \\tau_{d[n]}).\n\\] Again, the priors are fixed for the modes of transportation, \\[\\begin{eqnarray*}\n\\nu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\tau_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.\nThe full set of parameters that are marginalized in the posterior predictive distribution is \\[\n\\theta = (\\mu_{1:4}, \\sigma_{1:4}, \\nu_{1:4}, \\tau_{1:4}).\n\\]\n\n\n\nFor the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as\n\\[\nU(c, t) = -(c + 25 \\cdot t)\n\\]\nThe sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.\n\n\n\nAt this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.\nfunctions {\n  real U(real c, real t) {\n    return -(c + 25 * t);\n  }\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; d;\n  array[N] real c;\n  array[N] real&lt;lower=0&gt; t;\n}\nparameters {\n  vector[4] mu;\n  vector&lt;lower=0&gt;[4] sigma;\n  array[4] real nu;\n  array[4] real&lt;lower=0&gt; tau;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.25);\n  nu ~ normal(0, 20);\n  tau ~ lognormal(0, 0.25);\n  t ~ lognormal(mu[d], sigma[d]);\n  c ~ lognormal(nu[d], tau[d]);\n}\ngenerated quantities {\n  array[4] real util;\n  for (k in 1:4) {\n    util[k] = U(lognormal_rng(nu[k], tau[k]),\n                lognormal_rng(mu[k], sigma[k]));\n  }\n}\nThe generated quantities block defines an array variable util where util[k], which will hold the utility derived from a random commute for choice k generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.\nFor simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, nu and tau would be declared as univariate, and the RNG for cost would only be employed when k == 4.\nDefining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.\nAll that is left is to run Stan. The posterior mean for util[k] is the expected utility, which written out with full conditioning, is \\[\\begin{eqnarray*}\n\\mathbb{E}\\!\\left[U(x) \\mid d = k, d^{\\textrm{obs}}, x^{\\textrm{obs}}\\right]\n& = &\n\\int\n  U(x)\n  \\cdot p(x \\mid d = k, \\theta)\n  \\cdot p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M  U(x^{(m)} ),\n\\end{eqnarray*}\\] where \\[\nx^{(m)} \\sim p(x \\mid d = k, \\theta^{(m)} )\n\\] and \\[\n\\theta^{(m)}\n\\sim p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}}).\n\\]\nIn terms of Stan’s execution, the random generation of \\(x^{(m)}\\) is carried out with the lognormal_rng operations after \\(\\theta^{(m)}\\) is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.\nIt only remains to make the decision k with highest expected utility, which will correspond to the choice with the highest posterior mean for util[k]. This can be read off of the mean column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.",
+    "text": "Let’s suppose \\(\\theta = (3, 6)\\). To call the algebraic solver, we need to provide an initial guess. This varies on a case-by-case basis, but in general a good guess will speed up the solver and, in pathological cases, even determine whether the solver converges or not. If the solver does not converge, the Metropolis proposal gets rejected and a warning message, stating no acceptable solution was found, is issued.\nThe solver has three tuning parameters to determine convergence: the relative tolerance, the function tolerance, and the maximum number of steps. Their behavior is explained in the section about algebraic solvers with control parameters.\nThe following code returns the solution to our nonlinear algebraic system:\ntransformed data {\n  vector[2] y_guess = [1, 1]';\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\ntransformed parameters {\n  vector[2] theta = [3, 6]';\n  vector[2] y;\n\n  y = solve_newton(system, y_guess, theta, x_r, x_i);\n}\nwhich returns \\(y = (3, -2)\\).\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, must only involve parameters. Note there are no restrictions on the initial guess, y_guess, which may be a data or a parameter vector.\n\n\n\nThe Jacobian of the solution with respect to the parameters is computed using the implicit function theorem, which imposes certain restrictions. In particular, the Jacobian of the algebraic function \\(f\\) with respect to the unknowns \\(x\\) must be invertible. This requires the Jacobian to be square, meaning \\(f(y)\\) and \\(y\\) have the same length or, in other words the number of equations in the system is the same as the number of unknowns.\n\n\n\nCertain systems may be degenerate, meaning they have multiple solutions. The algebraic solver will not report these cases, as the algorithm stops once it has found an acceptable solution. The initial guess will often determine which solution gets found first. The degeneracy may be broken by putting additional constraints on the solution. For instance, it might make “physical sense” for a solution to be positive or negative.\nOn the other hand, a system may not have a solution (for a given point in the parameter space). In that case, the solver will not converge to a solution. When the solver fails to do so, the current Metropolis proposal gets rejected.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Decision Analysis"
+      "Example Models",
+      "Solving Algebraic Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/decision-analysis.html#continuous-choices",
-    "href": "stan-users-guide/decision-analysis.html#continuous-choices",
-    "title": "Decision Analysis",
+    "objectID": "stan-users-guide/algebraic-equations.html#algebra-control.section",
+    "href": "stan-users-guide/algebraic-equations.html#algebra-control.section",
+    "title": "Solving Algebraic Equations",
     "section": "",
-    "text": "Many choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.",
+    "text": "The call to the algebraic solver shown previously uses the default control settings. The _tol variant of the solver function allows three additional parameters, all of which must be supplied before the variadic arguments.\ny = solve_newton_tol(system, y_guess,  scaling_step, f_tol, max_steps,\n                     theta, x_r, x_i);\nFor the Newton solver the three control arguments are scaling step, function tolerance, and maximum number of steps. For the Powell’s hybrid method the three control arguments are relative tolerance, function tolerance, and maximum number of steps. If a Newton step is smaller than the scaling step tolerance, the code breaks, assuming the solver is no longer making significant progress. If set to 0, this constraint is ignored. For Powell’s hybrid method the relative tolerance is the estimated relative error of the solver and serves to test if a satisfactory solution has been found. After convergence of the either solver, the proposed solution is plugged into the algebraic system and its norm is compared to the function tolerance. If the norm is below the function tolerance, the solution is deemed acceptable. If the solver solver reaches the maximum number of steps, it stops and returns an error message. If one of the criteria is not met, the Metropolis proposal gets rejected with a warning message explaining which criterion was not satisfied.\nThe default values for the control arguments are respectively scaling_step = 1e-3 (\\(10^{-3}\\)), rel_tol = 1e-10 (\\(10^{-10}\\)), f_tol = 1e-6 (\\(10^{-6}\\)), and max_steps = 200 (\\(200\\)).\n\n\nThe relative and function tolerances control the accuracy of the solution generated by the solver. Relative tolerances are relative to the solution value. The function tolerance is the norm of the algebraic function, once we plug in the proposed solution. This norm should go to 0 (equivalently, all elements of the vector function are 0). It helps to think about this geometrically. Ideally the output of the algebraic function is at the origin; the norm measures deviations from this ideal. As the length of the return vector increases, a certain function tolerance becomes an increasingly difficult criterion to meet, given each individual element of the vector contribute to the norm.\nSmaller relative tolerances produce more accurate solutions but require more computational time.\n\n\nThe tolerances should be set low enough that setting them lower does not change the statistical properties of posterior sample generated by the Stan program. The sensitivity can be analysed using importance sampling without need to re-run MCMC with different tolerances as shown by Timonen et al. (2023).\n\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in MCMC when a bad jump is taken, particularly during warmup. If the limit is hit, the current Metropolis proposal gets rejected. Users will see a warning message stating the maximum number of steps has been exceeded.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Decision Analysis"
+      "Example Models",
+      "Solving Algebraic Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html",
-    "href": "stan-users-guide/finite-mixtures.html",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/clustering.html",
+    "href": "stan-users-guide/clustering.html",
+    "title": "Clustering Models",
     "section": "",
-    "text": "Finite mixture models of an outcome assume that the outcome is drawn from one of several distributions, the identity of which is controlled by a categorical mixing distribution. Mixture models typically have multimodal densities with modes near the modes of the mixture components. Mixture models may be parameterized in several ways, as described in the following sections. Mixture models may be used directly for modeling data with multimodal distributions, or they may be used as priors for other parameters.\n\n\nClustering models, as discussed in the clustering chapter, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the \\(K\\)-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.\n\n\n\nOne way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider \\(K\\) normal distributions with locations \\(\\mu_k \\in \\mathbb{R}\\) and scales \\(\\sigma_k \\in (0,\\infty)\\). Now consider mixing them in proportion \\(\\lambda\\), where \\(\\lambda_k \\geq 0\\) and \\(\\sum_{k=1}^K \\lambda_k = 1\\) (i.e., \\(\\lambda\\) lies in the unit \\(K\\)-simplex). For each outcome \\(y_n\\) there is a latent variable \\(z_n\\) in \\(\\{ 1,\\dotsc,K \\}\\) with a categorical distribution parameterized by \\(\\lambda\\), \\[\nz_n \\sim \\textsf{categorical}(\\lambda).\n\\]\nThe variable \\(y_n\\) is distributed according to the parameters of the mixture component \\(z_n\\), \\[\ny_n \\sim \\textsf{normal}(\\mu_{z[n]},\\sigma_{z[n]}).\n\\]\nThis model is not directly supported by Stan because it involves discrete parameters \\(z_n\\), but Stan can sample \\(\\mu\\) and \\(\\sigma\\) by summing out the \\(z\\) parameter as described in the next section.\n\n\n\nTo implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If \\(Y\\) is a mixture of \\(K\\) normal distributions with locations \\(\\mu_k\\) and scales \\(\\sigma_k\\) with mixing proportions \\(\\lambda\\) in the unit \\(K\\)-simplex, then \\[\np_Y\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n=\n\\sum_{k=1}^K \\lambda_k \\, \\textsf{normal}\\left(y \\mid \\mu_k, \\sigma_k\\right).\n\\]\n\n\nThe log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b) = \\log \\left(\\exp(a) + \\exp(b)\\right).\n\\]\nIf \\(a\\) and \\(b\\) are probabilities on the log scale, then \\(\\exp(a) +\n\\exp(b)\\) is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in log_sum_exp function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as \\[\n\\log \\left( \\exp(a) + \\exp(b)\\right)\n= c + \\log \\left( \\exp(a - c) + \\exp(b - c) \\right),\n\\] where \\(c = \\max(a, b)\\). In this evaluation, one of the terms, \\(a - c\\) or \\(b - c\\), is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the \\(\\max(a, b)\\) out of the log-exp round trip.\nFor example, the mixture of \\(\\textsf{normal}(-1, 2)\\) with \\(\\textsf{normal}(3, 1)\\), with mixing proportion \\(\\lambda =\n[0.3,0.7]^{\\top}\\), can be implemented in Stan as follows.\nparameters {\n  real y;\n}\nmodel {\n  target += log_sum_exp(log(0.3) + normal_lpdf(y | -1, 2),\n                        log(0.7) + normal_lpdf(y | 3, 1));\n}\nThe log probability term is derived by taking \\[\\begin{align*}\n\\log\\, &p\\left(y \\mid \\lambda,\\mu,\\sigma \\right) \\\\\n&= \\log\\big( 0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\n            + 0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big) \\\\\n&= \\log\\bigg( \\exp\\Big(\\log\\big(0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\\big)\\Big)\n             + \\exp\\Big(\\log\\big(0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right)\\big)\\Big) \\bigg) \\\\\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}\\big(\n                         \\log(0.3) + \\log \\textsf{normal}\\left(y \\mid -1,2 \\right),\n                         \\log(0.7) + \\log \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big).\n\\end{align*}\\]\n\n\n\nIf a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because\nlog_half = log(0.5);\nfor (n in 1:N) {\n    target +=\n        log_sum_exp(log_half + normal_lpdf(y[n] | mu[1], sigma[1]),\n                    log_half + normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nthen the \\(\\log 0.5\\) term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version\nfor (n in 1:N) {\n  target += log_sum_exp(normal_lpdf(y[n] | mu[1], sigma[1]),\n                        normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nThe same result holds if there are \\(K\\) components and the mixing simplex \\(\\lambda\\) is symmetric, i.e., \\[\n\\lambda = \\left( \\frac{1}{K}, \\dotsc, \\frac{1}{K} \\right).\n\\]\nThe result follows from the identity \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(c + a, c + b)\n=\nc + \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b)\n\\] and the fact that adding a constant \\(c\\) to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.\n\n\n\nThe posterior \\(p(z_n \\mid y_n, \\mu, \\sigma)\\) over the mixture indicator \\(z_n\n\\in 1:K\\) is often of interest as \\(p(z_n = k \\mid y, \\mu, \\sigma)\\) is the posterior probability that that observation \\(y_n\\) was generated by mixture component \\(k\\). The posterior can be computed via Bayes’s rule, \\[\\begin{align*}\n\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right]\n  &\\propto p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right)\\, p\\left(z_n = k \\mid \\lambda\\right) \\\\\n  &= \\textsf{normal}\\left(y_n \\mid \\mu_k, \\sigma_k\\right) \\cdot \\lambda_k.\n\\end{align*}\\]\nThe normalization can be done via summation, because \\(z_n \\in 1{:}K\\) only takes on finitely many values. In detail, \\[\np\\left(z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right) =\n\\frac{p\\left(y_n \\mid z_n = k, \\mu, \\sigma \\right) \\cdot p\\left(z_n = k \\mid \\lambda \\right)}\n     {\\sum_{k' = 1}^K p\\left(y_n \\mid z_n = k', \\mu, \\sigma \\right)\n                    \\cdot p\\left(z_n = k' \\mid \\lambda \\right)}.\n\\]\nOn the log scale, the normalized probability is computed as \\[\\begin{align*}\n\\log\\,&\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda\\right] \\\\\n&= \\log p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right) + \\log \\Pr\\!\\left[z_n = k \\mid \\lambda\\right] \\\\\n&\\quad  - \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{k' = 1}^K \\big(\\log p\\left(y_n \\mid z_n = k', \\mu, \\sigma\\right)\n                                                 + \\log p\\left(z_n = k' \\mid \\lambda\\right)\\big).\n\\end{align*}\\] This can be coded up directly in Stan; the change-point model in the change point section provides an example.\n\n\n\nGiven the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.\ndata {\n  int&lt;lower=1&gt; K;          // number of mixture components\n  int&lt;lower=1&gt; N;          // number of data points\n  array[N] real y;         // observations\n}\nparameters {\n  simplex[K] theta;          // mixing proportions\n  ordered[K] mu;             // locations of mixture components\n  vector&lt;lower=0&gt;[K] sigma;  // scales of mixture components\n}\nmodel {\n  vector[K] log_theta = log(theta);  // cache log calculation\n  sigma ~ lognormal(0, 2);\n  mu ~ normal(0, 10);\n  for (n in 1:N) {\n    vector[K] lps = log_theta;\n    for (k in 1:K) {\n      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);\n    }\n    target += log_sum_exp(lps);\n  }\n}\nThe model involves K mixture components and N data points. The mixing proportion parameter theta is declared to be a unit \\(K\\)-simplex, whereas the component location parameter mu and scale parameter sigma are both defined to be K-vectors.\nThe location parameter mu is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components mu[k] is symmetric, as it is here (each component has an independent \\(\\textsf{normal}(0, 10)\\) prior). It would even be possible to include a hierarchical prior for the components.\nThe values in the scale array sigma are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.\nThe model declares a local array variable lps to be size K and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points n. For each such point, the log of \\(\\theta_k \\times\n\\textsf{normal}\\left(y_n \\mid \\mu_k,\\sigma_k\\right)\\) is calculated and added to the array lps. Then the log probability is incremented with the log sum of exponentials of those values.\n\n\n\n\nThere is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that lambda, y[n], mu[1], mu[2], and sigma[1], sigma[2] are all scalars and lambda is between 0 and 1.\nfor (n in 1:N) {\n  target += log_sum_exp(log(lambda)\n                          + normal_lpdf(y[n] | mu[1], sigma[1]),\n                        log1m(lambda)\n                          + normal_lpdf(y[n] | mu[2], sigma[2]));\nor equivalently\nfor (n in 1:N) {\n  target += log_mix(lambda,\n                    normal_lpdf(y[n] | mu[1], sigma[1]),\n                    normal_lpdf(y[n] | mu[2], sigma[2]))\n  };\nThis definition assumes that each observation \\(y_n\\) may have arisen from either of the mixture components. The density is \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma\\right)\n= \\prod_{n=1}^N \\big(\\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1 \\right)\n                 + (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2 \\right)\\big).\n\\]\nContrast the previous model with the following (erroneous) attempt to vectorize the model.\ntarget += log_sum_exp(log(lambda)\n                        + normal_lpdf(y | mu[1], sigma[1]),\n                      log1m(lambda)\n                        + normal_lpdf(y | mu[2], sigma[2]));\nor equivalently,\ntarget += log_mix(lambda,\n                  normal_lpdf(y | mu[1], sigma[1]),\n                  normal_lpdf(y | mu[2], sigma[2]));\nThis second definition implies that the entire sequence \\(y_1, \\dotsc, y_n\\) of observations comes form one component or the other, defining a different density, \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n= \\lambda \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right).\n\\]\n\n\n\nIn many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.\nWe have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs. female, or Democrat vs. Republican).\nIn other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.\n\n\nAs an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\), \\[\\begin{align*}\n\\mu_1, \\mu_2       &\\sim \\textsf{normal}(0, 10) \\\\\n\\sigma_1, \\sigma_2 &\\sim \\textsf{halfnormal}(0, 10) \\\\\n\\end{align*}\\]\nThe prior on the mixture ratio is uniform, \\[\n\\lambda \\sim \\textsf{uniform}(0, 1),\n\\] so that with the likelihood \\[\np\\left(y_n \\mid \\mu, \\sigma\\right)\n= \\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right),\n\\] the joint distribution \\(p(y, \\mu, \\sigma, \\lambda)\\) is exchangeable in the parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\) with \\(\\lambda\\) flipping to \\(1 - \\lambda\\).1\n\n\n\nIn cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.\n\n\nPosterior predictive distribution for a new observation \\(\\tilde{y}\\) given the complete parameter vector \\(\\theta\\) will be \\[\np(\\tilde{y} \\mid y)\n=\n\\int_{\\theta}\np(\\tilde{y} \\mid \\theta)\n\\, p(\\theta \\mid y)\n\\, \\textsf{d}\\theta.\n\\]\nThe normal mixture example from the previous section, with \\(\\theta =\n(\\mu, \\sigma, \\lambda)\\), shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing \\(p(\\tilde{y} \\mid y)\\), which is more efficient statistically in terms of effective sample size, or simulating draws of \\(\\tilde{y}\\), which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] y_tilde;\n  // ...\n}\ngenerated quantities {\n  vector[N_tilde] log_p_y_tilde;\n  for (n in 1:N_tilde) {\n    log_p_y_tilde[n]\n      = log_mix(lambda,\n                normal_lpdf(y_tilde[n] | mu[1], sigma[1])\n                normal_lpdf(y_tilde[n] | mu[2], sigma[2]));\n  }\n}\nIt is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply log_sum_exp of the posterior draws of log_p_y_tilde. The average log predictive density is then given by subtracting log(N_new).\n\n\n\nOften a mixture model will be applied to a clustering problem and there might be two data items \\(y_i\\) and \\(y_j\\) for which there is a question of whether they arose from the same mixture component. If we take \\(z_i\\) and \\(z_j\\) to be the component responsibility discrete variables, then the quantity of interest is \\(z_i = z_j\\), which can be summarized as an event probability \\[\n\\Pr[z_i = z_j \\mid y]\n=\n\\int_{\\theta}\n\\frac{\\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \\mid \\theta)}\n     {\\sum_{k=0}^1 \\sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \\mid \\theta)}\n\\,\np(\\theta \\mid y)\n\\,\n\\textsf{d}\\theta.\n\\]\nAs with other event probabilities, this can be calculated in the generated quantities block either by sampling \\(z_i\\) and \\(z_j\\) and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.\n\n\n\n\n\nZero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by Lambert (1992), add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.\nZero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.\n\n\nConsider the following example for zero-inflated Poisson distributions. There is a probability \\(\\theta\\) of observing a zero, and a probability \\(1 - \\theta\\) of observing a count with a \\(\\textsf{Poisson}(\\lambda)\\) distribution (now \\(\\theta\\) is being used for mixing proportions because \\(\\lambda\\) is the traditional notation for a Poisson mean parameter). Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as \\[\\begin{align*}\ny_n & = 0 & \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}(y_n \\mid \\lambda) & \\quad\\text{with probability } 1-\\theta.\n\\end{align*}\\]\nStan does not support conditional distribution statements (with ~) conditional on some parameter, and we need to consider the corresponding likelihood \\[\np(y_n \\mid \\theta,\\lambda)\n=\n\\begin{cases}\n\\theta + (1 - \\theta) \\times \\textsf{Poisson}(0 \\mid \\lambda) & \\quad\\text{if } y_n = 0, \\text{ and}\\\\\n(1-\\theta) \\times \\textsf{Poisson}(y_n \\mid \\lambda) &\\quad\\text{if } y_n &gt; 0.\n\\end{cases}\n\\] The log likelihood can be coded directly in Stan (with target +=) as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  for (n in 1:N) {\n    if (y[n] == 0) {\n      target += log_sum_exp(log(theta),\n                            log1m(theta)\n                              + poisson_lpmf(y[n] | lambda));\n    } else {\n      target += log1m(theta)\n                  + poisson_lpmf(y[n] | lambda);\n    }\n  }\n}\nThe log1m(theta) computes log(1-theta), but is more computationally stable. The log_sum_exp(lp1,lp2) function adds the log probabilities on the linear scale; it is defined to be equal to log(exp(lp1) + exp(lp2)), but is more computationally stable and faster.\n\n\nThe code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also poisson_lpmf(0 | lambda) every time the first condition body executes. The use of the redundant terms is conditioned on y, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.\nThe number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare y_nonzero, so it must be computed in a function.\nfunctions {\n  int num_zeros(array[] int y) {\n    int sum = 0;\n    for (n in 1:size(y)) {\n      sum += (y[n] == 0);\n    }\n    return sum;\n  }\n}\n// ...\ntransformed data {\n  int&lt;lower=0&gt; N_zero = num_zeros(y);\n  array[N - N_zero] int&lt;lower=1&gt; y_nonzero;\n  int N_nonzero = 0;\n  for (n in 1:N) {\n    if (y[n] == 0) continue;\n    N_nonzero += 1;\n    y_nonzero[N_nonzero] = y[n];\n  }\n}\n// ...\nmodel {\n  // ...\n   target\n     += N_zero\n          * log_sum_exp(log(theta),\n                        log1m(theta)\n                          + poisson_lpmf(0 | lambda));\n   target += N_nonzero * log1m(theta);\n   target += poisson_lpmf(y_nonzero | lambda);\n  // ...\n}\nThe boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if y_nonzero is empty, N_nonzero will be zero, and the last two target increment terms will add zeros.\n\n\n\n\nThe hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as [ \\[\\begin{align*}\ny_n & = 0 \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}_{x\\neq 0}(y_n \\mid \\lambda) \\quad\\text{with probability } 1-\\theta,\n\\end{align*}\\] ] Where \\(\\textsf{Poisson}_{x\\neq 0}\\) is a truncated Poisson distribution, truncated at \\(0\\).\nThe corresponding likelihood function for the hurdle model is defined by \\[\np(y\\mid\\theta,\\lambda)\n=\n\\begin{cases}\n\\theta &\\quad\\text{if } y = 0, \\text{ and}\\\\\n(1 - \\theta)\n   \\frac{\\displaystyle \\textsf{Poisson}(y \\mid \\lambda)}\n        {\\displaystyle  1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)}\n&\\quad\\text{if } y &gt; 0,\n\\end{cases}\n\\] where \\(\\textsf{PoissonCDF}\\) is the cumulative distribution function for the Poisson distribution and and \\(1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)\\) is the relative normalization term for the truncated Poisson (truncated at \\(0\\)).\nThe hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.\nif (y[n] == 0) {\n  target += log(theta);\n} else {\n  target += log1m(theta) + poisson_lpmf(y[n] | lambda)\n            - poisson_lccdf(0 | lambda));\n}\nJulian King pointed out that because \\[\\begin{align*}\n\\log \\left( 1 - \\textsf{PoissonCDF}(0 \\mid \\lambda) \\right)\n  &= \\log \\left( 1 - \\textsf{Poisson}(0 \\mid \\lambda) \\right) \\\\\n  &= \\log(1 - \\exp(-\\lambda))\n\\end{align*}\\] the CCDF in the else clause can be replaced with a simpler expression.\ntarget += log1m(theta) + poisson_lpmf(y[n] | lambda)\n          - log1m_exp(-lambda));\nThe resulting code is about 15% faster than the code with the CCDF.\nThis is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size \\(N=200\\) and parameters \\(\\theta=0.3\\) and \\(\\lambda = 8\\), the speedup is a factor of 10; it will be lower for smaller \\(N\\) and greater for larger \\(N\\); it will also be greater for larger \\(\\theta\\).\nTo achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,\nfunctions {\n  int num_zero(array[] int y) {\n    int nz = 0;\n    for (n in 1:size(y)) {\n      if (y[n] == 0) {\n        nz += 1;\n      }\n    }\n    return nz;\n  }\n}\nThen a transformed data block can be used to store the sufficient statistics,\ntransformed data {\n  int&lt;lower=0, upper=N&gt; N0 = num_zero(y);\n  int&lt;lower=0, upper=N&gt; Ngt0 = N - N0;\n  array[N - num_zero(y)] int&lt;lower=1&gt; y_nz;\n  {\n    int pos = 1;\n    for (n in 1:N) {\n      if (y[n] != 0) {\n        y_nz[pos] = y[n];\n        pos += 1;\n      }\n    }\n  }\n}\nThe model block is then reduced to three statements.\nmodel {\n  N0 ~ binomial(N, theta);\n  y_nz ~ poisson(lambda);\n  target += -Ngt0 * log1m_exp(-lambda);\n}\nThe first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant Ngt0; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.\n\n\n\n\nSuppose we have a two-component mixture model with mixing rate \\(\\lambda \\in (0, 1)\\). Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are \\(N\\) observations, the mixture components will be estimated with effective data sizes of \\(\\theta \\, N\\) and \\((1 - \\theta)\n\\, N\\) for the two components for some \\(\\theta \\in (0, 1)\\). The effective weighting size is determined by posterior responsibility, not simply by the mixing rate \\(\\lambda\\).\n\n\nIn contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data \\(y\\).\nIf different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see Hoeting et al. (1999) and Gelman et al. (2013) for theoretical and computational details).",
+    "text": "Unsupervised methods for organizing data into groups are collectively referred to as clustering. This chapter describes the implementation in Stan of two widely used statistical clustering models, soft \\(K\\)-means and latent Dirichlet allocation (LDA). In addition, this chapter includes naive Bayesian classification, which can be viewed as a form of clustering which may be supervised. These models are typically expressed using discrete parameters for cluster assignments. Nevertheless, they can be implemented in Stan like any other mixture model by marginalizing out the discrete parameters (see the mixture modeling chapter).\n\n\nAs mentioned in the clustering section, clustering models and finite mixture models are really just two sides of the same coin. The “soft” \\(K\\)-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of \\(K\\)-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.\n\n\n\n\\(K\\)-means clustering is a method of clustering data represented as \\(D\\)-dimensional vectors. Specifically, there will be \\(N\\) items to be clustered, each represented as a vector \\(y_n \\in \\mathbb{R}^D\\). In the “soft” version of \\(K\\)-means, the assignments to clusters will be probabilistic.\n\n\n\\(K\\)-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters \\(K\\) and data vectors \\(y\\) as input.\n\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), randomly assign vector \\(y_n\\) to a cluster in \\(\\{1,\\dotsc,K\\}\\);\nRepeat\n\nFor each cluster \\(k\\) in \\(\\{1,\\dotsc,K\\}\\), compute the cluster centroid \\(\\mu_k\\) by averaging the vectors assigned to that cluster;\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), reassign \\(y_n\\) to the cluster \\(k\\) for which the (Euclidean) distance from \\(y_n\\) to \\(\\mu_k\\) is smallest;\nIf no vectors changed cluster, return the cluster assignments.\n\n\nThis algorithm is guaranteed to terminate.\n\n\n\nSoft \\(K\\)-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft \\(K\\)-means can be expressed (and coded in Stan) as a multivariate normal mixture model.\nIn the full generative model, each data point \\(n\\) in \\(\\{1,\\dotsc,N\\}\\) is assigned a cluster \\(z_n \\in \\{1,\\dotsc,K\\}\\) with symmetric uniform probability, \\[\nz_n \\sim \\textsf{categorical}(1/K),\n\\] where \\(1\\) is the unit vector of \\(K\\) dimensions, so that \\(1/K\\) is the symmetric \\(K\\)-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.\nThe data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment \\(z_n\\), \\[\ny_n \\sim  \\textsf{normal}(\\mu_{z[n]},\\Sigma_{z[n]})\n\\]\nThe sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters \\(k\\), \\[\n\\Sigma_k = \\mathrm{diag\\_matrix}({\\bf 1}),\n\\] so that the log multivariate normal can be implemented directly up to a proportion by \\[\n\\mathrm{normal}\\left( y_n \\mid \\mu_k, \\mathrm{diag\\_matrix}({\\bf 1}) \\right)\n\\propto \\exp \\left (- \\frac{1}{2} \\sum_{d=1}^D \\left( \\mu_{k,d} - y_{n,d}\n  \\right)^2 \\right).\n\\] The spatial perspective on \\(K\\)-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean \\(\\mu_k\\) to the data point \\(y_n\\).\n\n\n\nConsider the following Stan program for implementing \\(K\\)-means clustering.\ndata {\n  int&lt;lower=0&gt; N;        // number of data points\n  int&lt;lower=1&gt; D;        // number of dimensions\n  int&lt;lower=1&gt; K;        // number of clusters\n  array[N] vector[D] y;  // observations\n}\ntransformed data {\n  real&lt;upper=0&gt; neg_log_K;\n  neg_log_K = -log(K);\n}\nparameters {\n  array[K] vector[D] mu; // cluster means\n}\ntransformed parameters {\n  array[N, K] real&lt;upper=0&gt; soft_z; // log unnormalized clusters\n  for (n in 1:N) {\n    for (k in 1:K) {\n      soft_z[n, k] = neg_log_K\n                     - 0.5 * dot_self(mu[k] - y[n]);\n    }\n  }\n}\nmodel {\n  // prior\n  for (k in 1:K) {\n    mu[k] ~ std_normal();\n  }\n\n  // likelihood\n  for (n in 1:N) {\n    target += log_sum_exp(soft_z[n]);\n  }\n}\nThere is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.\nThe only parameter is mu, where mu[k] is the centroid for cluster \\(k\\). The transformed parameters soft_z[n] contain the log of the unnormalized cluster assignment probabilities. The vector soft_z[n] can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.\n\n\n\nThe multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., \\(L_2\\) distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on \\(L_1\\) distance (i.e., Manhattan or taxicab distance).\nWithin the multivariate normal version of \\(K\\)-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.\nAlthough there is no global spatial analog, it is common to see soft \\(K\\)-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.\n\n\n\n\nTwo problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the label switching section.\n\n\nCluster assignments are not identified—permuting the cluster mean vectors mu leads to a model with identical likelihoods. For instance, permuting the first two indexes in mu and the first two indexes in each soft_z[n] leads to an identical likelihood (and prior).\nThe lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft \\(K\\)-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.\n\n\n\nThe other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.\nBayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.\nIn light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.\n\n\n\n\nNaive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.1\nMultinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.\nNaive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.\nThe observed data consists of a sequence of \\(M\\) documents made up of bags of words drawn from a vocabulary of \\(V\\) distinct words. A document \\(m\\) has \\(N_m\\) words, which are indexed as \\(w_{m,1}, \\dotsc,\nw_{m,N[m]} \\in \\{1,\\dotsc,V\\}\\). Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) \\(K\\) is fixed.\nThe multinomial mixture model generates a single category \\(z_m \\in\n\\{1,\\dotsc,K\\}\\) for each document \\(m \\in \\{1,\\dotsc,M\\}\\) according to a categorical distribution, \\[\nz_m \\sim \\textsf{categorical}(\\theta).\n\\] The \\(K\\)-simplex parameter \\(\\theta\\) represents the prevalence of each category in the data.\nNext, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word \\(n\\) of document \\(m\\) being generated as \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m]}).\n\\] The parameter \\(\\phi_{z[m]}\\) is a \\(V\\)-simplex representing the probability of each word in the vocabulary in documents of category \\(z_m\\).\nThe parameters \\(\\theta\\) and \\(\\phi\\) are typically given symmetric Dirichlet priors. The prevalence \\(\\theta\\) is sometimes fixed to produce equal probabilities for each category \\(k \\in \\{1,\\dotsc,K\\}\\).\n\n\nThe specification for naive Bayes in the previous sections have used a ragged array notation for the words \\(w\\). Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.\n\\[\n\\begin{array}{lll}\n\\hline\n\\mathrm{n} \\qquad\\qquad\\qquad\\qquad & \\mathrm{w[n]} \\qquad & \\mathrm{doc[n]} \\\\\n\\hline\n1         & w_{1,1}    & 1 \\\\\n2         & w_{1,2}    & 1 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1     & w_{1,N[1]} & 1 \\\\\nN_1 + 1 & w_{2,1}    & 2 \\\\\nN_1 + 2 & w_{2,2}    & 2 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1 + N_2     & w_{2,N[2]} & 2 \\\\\nN_1 + N_2 + 1 & w_{3,1}    & 3 \\\\\n\\vdots       & \\vdots    & \\vdots \\\\\nN = \\sum_{m=1}^M N_m & w_{M,N[M]} & M \\\\\n\\hline\n\\end{array}\n\\]\nThe relevant variables for the program are N, the total number of words in all the documents, the word array w, and the document identity array doc.\n\n\n\nA naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows\ndata {\n  // training data\n  int&lt;lower=1&gt; K;               // num topics\n  int&lt;lower=1&gt; V;               // num words\n  int&lt;lower=0&gt; M;               // num docs\n  int&lt;lower=0&gt; N;               // total word instances\n  array[M] int&lt;lower=1, upper=K&gt; z;    // topic for doc m\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  // hyperparameters\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  simplex[K] theta;             // topic prevalence\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    z[m] ~ categorical(theta);\n  }\n  for (n in 1:N) {\n    w[n] ~ categorical(phi[z[doc[n]]]);\n  }\n}\nThe topic identifiers \\(z_m\\) are declared as data and the latent category assignments are included as part of the likelihood function.\n\n\n\nNaive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number \\(K\\) of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels z. Because z is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability\n\\[\\begin{align*}\n\\log\\, &p(w_{m,1},\\dotsc,w_{m,N_m} \\mid \\theta,\\phi) \\\\\n&= \\log \\sum_{k=1}^K\n    \\left( \\textsf{categorical}(k \\mid \\theta)\n           \\times \\prod_{n=1}^{N_m} \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right) \\\\\n&= \\log \\sum_{k=1}^K \\exp \\left(\n    \\log \\textsf{categorical}(k \\mid \\theta)\n     + \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right).\n\\end{align*}\\]\nThe last step shows how the log_sum_exp function can be used to stabilize the numerical calculation and return a result on the log scale.\nmodel {\n  array[M, K] real gamma;\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    for (k in 1:K) {\n      gamma[m, k] = categorical_lpmf(k | theta);\n    }\n  }\n  for (n in 1:N) {\n    for (k in 1:K) {\n      gamma[doc[n], k] = gamma[doc[n], k]\n                         + categorical_lpmf(w[n] | phi[k]);\n    }\n  }\n  for (m in 1:M) {\n    target += log_sum_exp(gamma[m]);\n  }\n}\nThe local variable gamma[m, k] represents the value \\[\n\\gamma_{m,k} = \\log \\textsf{categorical}(k \\mid \\theta)\n+ \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k).\n\\]\nGiven \\(\\gamma\\), the posterior probability that document \\(m\\) is assigned category \\(k\\) is \\[\n\\Pr[z_m = k \\mid w,\\alpha,\\beta]\n=\n\\exp \\left(\n\\gamma_{m,k}\n- \\log \\sum_{k=1}^K \\exp \\left( \\gamma_{m,k} \\right)\n\\right).\n\\]\nIf the variable gamma were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.\n\n\n\nFull Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see Gelman et al. (2013) and Gelman and Hill (2007) for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.\nTo specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables M2 N2, w2, and doc2. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.\n\n\n\nAn alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of gamma for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.\n\n\n\n\nLatent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model (Blei, Ng, and Jordan 2003) that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.\n\n\nThe basic model assumes each document is generated independently based on fixed hyperparameters. For document \\(m\\), the first step is to draw a topic distribution simplex \\(\\theta_m\\) over the \\(K\\) topics, \\[\n\\theta_m \\sim \\textsf{Dirichlet}(\\alpha).\n\\]\nThe prior hyperparameter \\(\\alpha\\) is fixed to a \\(K\\)-vector of positive values. Each word in the document is generated independently conditional on the distribution \\(\\theta_m\\). First, a topic \\(z_{m,n} \\in \\{1,\\dotsc,K\\}\\) is drawn for the word based on the document-specific topic-distribution, \\[\nz_{m,n} \\sim \\textsf{categorical}(\\theta_m).\n\\]\nFinally, the word \\(w_{m,n}\\) is drawn according to the word distribution for topic \\(z_{m,n}\\), \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m,n]}).\n\\] The distributions \\(\\phi_k\\) over words for topic \\(k\\) are also given a Dirichlet prior, \\[\n\\phi_k \\sim \\textsf{Dirichlet}(\\beta)\n\\]\nwhere \\(\\beta\\) is a fixed \\(V\\)-vector of positive values.\n\n\n\nAlthough Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is \\[\\begin{align*}\np(\\theta,\\phi \\mid w,\\alpha,\\beta)\n&\\propto p(\\theta \\mid \\alpha) \\, p(\\phi \\mid \\beta) \\, p(w \\mid \\theta,\\phi) \\\\\n&= \\prod_{m=1}^M p(\\theta_m \\mid \\alpha)\n    \\times \\prod_{k=1}^K p(\\phi_k \\mid \\beta)\n    \\times \\prod_{m=1}^M \\prod_{n=1}^{M[n]} p(w_{m,n} \\mid \\theta_m,\\phi).\n\\end{align*}\\]\nThe inner word-probability term is defined by summing out the topic assignments, \\[\\begin{align*}\np(w_{m,n} \\mid \\theta_m,\\phi)\n&= \\sum_{z=1}^K p(z,w_{m,n} \\mid \\theta_m,\\phi) \\\\\n&= \\sum_{z=1}^K p(z \\mid \\theta_m) \\, p(w_{m,n} \\mid \\phi_z).\n\\end{align*}\\]\nPlugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, \\[\\begin{align*}\n\\log\\, &p(\\theta,\\phi \\mid w,\\alpha,\\beta) \\\\\n&= \\sum_{m=1}^M \\log \\textsf{Dirichlet}(\\theta_m \\mid \\alpha)\n    + \\sum_{k=1}^K \\log \\textsf{Dirichlet}(\\phi_k \\mid \\beta) \\\\\n&\\qquad + \\sum_{m=1}^M \\sum_{n=1}^{N[m]} \\log \\left(\n    \\sum_{z=1}^K \\textsf{categorical}(z \\mid \\theta_m)\n    \\times \\textsf{categorical}(w_{m,n} \\mid \\phi_z)\n  \\right)\n\\end{align*}\\]\n\n\n\nApplying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.\ndata {\n  int&lt;lower=2&gt; K;               // num topics\n  int&lt;lower=2&gt; V;               // num words\n  int&lt;lower=1&gt; M;               // num docs\n  int&lt;lower=1&gt; N;               // total word instances\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  array[M] simplex[K] theta;    // topic dist for doc m\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  for (m in 1:M) {\n    theta[m] ~ dirichlet(alpha);  // prior\n  }\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);     // prior\n  }\n  for (n in 1:N) {\n    array[K] real gamma;\n    for (k in 1:K) {\n      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);\n    }\n    target += log_sum_exp(gamma);  // likelihood;\n  }\n}\nAs in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.\n\n\n\nTo account for correlations in the distribution of topics for documents, Blei and Lafferty (2007) introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.\nThe authors treat the prior as a fixed hyperparameter. They use an \\(L_1\\)-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.\n\n\nThe Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior alpha in the data declaration with the mean and covariance of the multivariate logistic normal prior.\ndata {\n  // ... data as before without alpha ...\n  vector[K] mu;          // topic mean\n  cov_matrix[K] Sigma;   // topic covariance\n}\nRather than drawing the simplex parameter theta from a Dirichlet, a parameter eta is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.\nparameters {\n  array[K] simplex[V] phi;     // word dist for topic k\n  array[M] vector[K] eta;      // topic dist for doc m\n}\ntransformed parameters {\n  array[M] simplex[K] theta;\n  for (m in 1:M) {\n    theta[m] = softmax(eta[m]);\n  }\n}\nmodel {\n  for (m in 1:M) {\n    eta[m] ~ multi_normal(mu, Sigma);\n  }\n  // ... model as before w/o prior for theta ...\n}\n\n\n\nBy adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean mu and covariance Sigma from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix Sigma may be encoded as follows.\n// ... data block as before, but without alpha ...\nparameters {\n  vector[K] mu;              // topic mean\n  corr_matrix[K] Omega;      // correlation matrix\n  vector&lt;lower=0&gt;[K] sigma;  // scales\n  array[M] vector[K] eta;    // logit topic dist for doc m\n  array[K] simplex[V] phi;   // word dist for topic k\n}\ntransformed parameters {\n  // ... eta as above ...\n  cov_matrix[K] Sigma;       // covariance matrix\n  for (m in 1:K) {\n    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];\n  }\n  for (m in 1:(K-1)) {\n    for (n in (m+1):K) {\n      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];\n      Sigma[n, m] = Sigma[m, n];\n    }\n  }\n}\nmodel {\n  mu ~ normal(0, 5);      // vectorized, diffuse\n  Omega ~ lkj_corr(2.0);  // regularize to unit correlation\n  sigma ~ cauchy(0, 5);   // half-Cauchy due to constraint\n  // ... words sampled as above ...\n}\nThe \\(\\textsf{LKJCorr}\\) distribution with shape \\(\\alpha &gt; 0\\) has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by \\[\n\\mathsf{LkjCorr}(\\Omega\\mid\\alpha) \\propto \\mathrm{det}(\\Omega)^{\\alpha - 1}\n\\] With a scale of \\(\\alpha = 2\\), the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix \\(\\Sigma\\) for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on sigma.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#clustering-mixture.section",
-    "href": "stan-users-guide/finite-mixtures.html#clustering-mixture.section",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/clustering.html#relation-to-finite-mixture-models",
+    "href": "stan-users-guide/clustering.html#relation-to-finite-mixture-models",
+    "title": "Clustering Models",
     "section": "",
-    "text": "Clustering models, as discussed in the clustering chapter, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the \\(K\\)-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.",
+    "text": "As mentioned in the clustering section, clustering models and finite mixture models are really just two sides of the same coin. The “soft” \\(K\\)-means model described in the next section is a normal mixture model (with varying assumptions about covariance in higher dimensions leading to variants of \\(K\\)-means). Latent Dirichlet allocation is a mixed-membership multinomial mixture.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#latent-discrete-parameterization",
-    "href": "stan-users-guide/finite-mixtures.html#latent-discrete-parameterization",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/clustering.html#soft-k-means",
+    "href": "stan-users-guide/clustering.html#soft-k-means",
+    "title": "Clustering Models",
     "section": "",
-    "text": "One way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider \\(K\\) normal distributions with locations \\(\\mu_k \\in \\mathbb{R}\\) and scales \\(\\sigma_k \\in (0,\\infty)\\). Now consider mixing them in proportion \\(\\lambda\\), where \\(\\lambda_k \\geq 0\\) and \\(\\sum_{k=1}^K \\lambda_k = 1\\) (i.e., \\(\\lambda\\) lies in the unit \\(K\\)-simplex). For each outcome \\(y_n\\) there is a latent variable \\(z_n\\) in \\(\\{ 1,\\dotsc,K \\}\\) with a categorical distribution parameterized by \\(\\lambda\\), \\[\nz_n \\sim \\textsf{categorical}(\\lambda).\n\\]\nThe variable \\(y_n\\) is distributed according to the parameters of the mixture component \\(z_n\\), \\[\ny_n \\sim \\textsf{normal}(\\mu_{z[n]},\\sigma_{z[n]}).\n\\]\nThis model is not directly supported by Stan because it involves discrete parameters \\(z_n\\), but Stan can sample \\(\\mu\\) and \\(\\sigma\\) by summing out the \\(z\\) parameter as described in the next section.",
+    "text": "\\(K\\)-means clustering is a method of clustering data represented as \\(D\\)-dimensional vectors. Specifically, there will be \\(N\\) items to be clustered, each represented as a vector \\(y_n \\in \\mathbb{R}^D\\). In the “soft” version of \\(K\\)-means, the assignments to clusters will be probabilistic.\n\n\n\\(K\\)-means clustering is typically described geometrically in terms of the following algorithm, which assumes the number of clusters \\(K\\) and data vectors \\(y\\) as input.\n\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), randomly assign vector \\(y_n\\) to a cluster in \\(\\{1,\\dotsc,K\\}\\);\nRepeat\n\nFor each cluster \\(k\\) in \\(\\{1,\\dotsc,K\\}\\), compute the cluster centroid \\(\\mu_k\\) by averaging the vectors assigned to that cluster;\nFor each \\(n\\) in \\(\\{1,\\dotsc,N\\}\\), reassign \\(y_n\\) to the cluster \\(k\\) for which the (Euclidean) distance from \\(y_n\\) to \\(\\mu_k\\) is smallest;\nIf no vectors changed cluster, return the cluster assignments.\n\n\nThis algorithm is guaranteed to terminate.\n\n\n\nSoft \\(K\\)-means clustering treats the cluster assignments as probability distributions over the clusters. Because of the connection between Euclidean distance and multivariate normal models with a fixed covariance, soft \\(K\\)-means can be expressed (and coded in Stan) as a multivariate normal mixture model.\nIn the full generative model, each data point \\(n\\) in \\(\\{1,\\dotsc,N\\}\\) is assigned a cluster \\(z_n \\in \\{1,\\dotsc,K\\}\\) with symmetric uniform probability, \\[\nz_n \\sim \\textsf{categorical}(1/K),\n\\] where \\(1\\) is the unit vector of \\(K\\) dimensions, so that \\(1/K\\) is the symmetric \\(K\\)-simplex. Thus the model assumes that each data point is drawn from a hard decision about cluster membership. The softness arises only from the uncertainty about which cluster generated a data point.\nThe data points themselves are generated from a multivariate normal distribution whose parameters are determined by the cluster assignment \\(z_n\\), \\[\ny_n \\sim  \\textsf{normal}(\\mu_{z[n]},\\Sigma_{z[n]})\n\\]\nThe sample implementation in this section assumes a fixed unit covariance matrix shared by all clusters \\(k\\), \\[\n\\Sigma_k = \\mathrm{diag\\_matrix}({\\bf 1}),\n\\] so that the log multivariate normal can be implemented directly up to a proportion by \\[\n\\mathrm{normal}\\left( y_n \\mid \\mu_k, \\mathrm{diag\\_matrix}({\\bf 1}) \\right)\n\\propto \\exp \\left (- \\frac{1}{2} \\sum_{d=1}^D \\left( \\mu_{k,d} - y_{n,d}\n  \\right)^2 \\right).\n\\] The spatial perspective on \\(K\\)-means arises by noting that the inner term is just half the negative Euclidean distance from the cluster mean \\(\\mu_k\\) to the data point \\(y_n\\).\n\n\n\nConsider the following Stan program for implementing \\(K\\)-means clustering.\ndata {\n  int&lt;lower=0&gt; N;        // number of data points\n  int&lt;lower=1&gt; D;        // number of dimensions\n  int&lt;lower=1&gt; K;        // number of clusters\n  array[N] vector[D] y;  // observations\n}\ntransformed data {\n  real&lt;upper=0&gt; neg_log_K;\n  neg_log_K = -log(K);\n}\nparameters {\n  array[K] vector[D] mu; // cluster means\n}\ntransformed parameters {\n  array[N, K] real&lt;upper=0&gt; soft_z; // log unnormalized clusters\n  for (n in 1:N) {\n    for (k in 1:K) {\n      soft_z[n, k] = neg_log_K\n                     - 0.5 * dot_self(mu[k] - y[n]);\n    }\n  }\n}\nmodel {\n  // prior\n  for (k in 1:K) {\n    mu[k] ~ std_normal();\n  }\n\n  // likelihood\n  for (n in 1:N) {\n    target += log_sum_exp(soft_z[n]);\n  }\n}\nThere is an independent standard normal prior on the centroid parameters; this prior could be swapped with other priors, or even a hierarchical model to fit an overall problem scale and location.\nThe only parameter is mu, where mu[k] is the centroid for cluster \\(k\\). The transformed parameters soft_z[n] contain the log of the unnormalized cluster assignment probabilities. The vector soft_z[n] can be converted back to a normalized simplex using the softmax function (see the functions reference manual), either externally or within the model’s generated quantities block.\n\n\n\nThe multivariate normal distribution with unit covariance matrix produces a log probability density proportional to Euclidean distance (i.e., \\(L_2\\) distance). Other distributions relate to other geometries. For instance, replacing the normal distribution with the double exponential (Laplace) distribution produces a clustering model based on \\(L_1\\) distance (i.e., Manhattan or taxicab distance).\nWithin the multivariate normal version of \\(K\\)-means, replacing the unit covariance matrix with a shared covariance matrix amounts to working with distances defined in a space transformed by the inverse covariance matrix.\nAlthough there is no global spatial analog, it is common to see soft \\(K\\)-means specified with a per-cluster covariance matrix. In this situation, a hierarchical prior may be used for the covariance matrices.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#summing-out-the-responsibility-parameter",
-    "href": "stan-users-guide/finite-mixtures.html#summing-out-the-responsibility-parameter",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/clustering.html#the-difficulty-of-bayesian-inference-for-clustering",
+    "href": "stan-users-guide/clustering.html#the-difficulty-of-bayesian-inference-for-clustering",
+    "title": "Clustering Models",
     "section": "",
-    "text": "To implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If \\(Y\\) is a mixture of \\(K\\) normal distributions with locations \\(\\mu_k\\) and scales \\(\\sigma_k\\) with mixing proportions \\(\\lambda\\) in the unit \\(K\\)-simplex, then \\[\np_Y\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n=\n\\sum_{k=1}^K \\lambda_k \\, \\textsf{normal}\\left(y \\mid \\mu_k, \\sigma_k\\right).\n\\]\n\n\nThe log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b) = \\log \\left(\\exp(a) + \\exp(b)\\right).\n\\]\nIf \\(a\\) and \\(b\\) are probabilities on the log scale, then \\(\\exp(a) +\n\\exp(b)\\) is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in log_sum_exp function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as \\[\n\\log \\left( \\exp(a) + \\exp(b)\\right)\n= c + \\log \\left( \\exp(a - c) + \\exp(b - c) \\right),\n\\] where \\(c = \\max(a, b)\\). In this evaluation, one of the terms, \\(a - c\\) or \\(b - c\\), is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the \\(\\max(a, b)\\) out of the log-exp round trip.\nFor example, the mixture of \\(\\textsf{normal}(-1, 2)\\) with \\(\\textsf{normal}(3, 1)\\), with mixing proportion \\(\\lambda =\n[0.3,0.7]^{\\top}\\), can be implemented in Stan as follows.\nparameters {\n  real y;\n}\nmodel {\n  target += log_sum_exp(log(0.3) + normal_lpdf(y | -1, 2),\n                        log(0.7) + normal_lpdf(y | 3, 1));\n}\nThe log probability term is derived by taking \\[\\begin{align*}\n\\log\\, &p\\left(y \\mid \\lambda,\\mu,\\sigma \\right) \\\\\n&= \\log\\big( 0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\n            + 0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big) \\\\\n&= \\log\\bigg( \\exp\\Big(\\log\\big(0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\\big)\\Big)\n             + \\exp\\Big(\\log\\big(0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right)\\big)\\Big) \\bigg) \\\\\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}\\big(\n                         \\log(0.3) + \\log \\textsf{normal}\\left(y \\mid -1,2 \\right),\n                         \\log(0.7) + \\log \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big).\n\\end{align*}\\]\n\n\n\nIf a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because\nlog_half = log(0.5);\nfor (n in 1:N) {\n    target +=\n        log_sum_exp(log_half + normal_lpdf(y[n] | mu[1], sigma[1]),\n                    log_half + normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nthen the \\(\\log 0.5\\) term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version\nfor (n in 1:N) {\n  target += log_sum_exp(normal_lpdf(y[n] | mu[1], sigma[1]),\n                        normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nThe same result holds if there are \\(K\\) components and the mixing simplex \\(\\lambda\\) is symmetric, i.e., \\[\n\\lambda = \\left( \\frac{1}{K}, \\dotsc, \\frac{1}{K} \\right).\n\\]\nThe result follows from the identity \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(c + a, c + b)\n=\nc + \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b)\n\\] and the fact that adding a constant \\(c\\) to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.\n\n\n\nThe posterior \\(p(z_n \\mid y_n, \\mu, \\sigma)\\) over the mixture indicator \\(z_n\n\\in 1:K\\) is often of interest as \\(p(z_n = k \\mid y, \\mu, \\sigma)\\) is the posterior probability that that observation \\(y_n\\) was generated by mixture component \\(k\\). The posterior can be computed via Bayes’s rule, \\[\\begin{align*}\n\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right]\n  &\\propto p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right)\\, p\\left(z_n = k \\mid \\lambda\\right) \\\\\n  &= \\textsf{normal}\\left(y_n \\mid \\mu_k, \\sigma_k\\right) \\cdot \\lambda_k.\n\\end{align*}\\]\nThe normalization can be done via summation, because \\(z_n \\in 1{:}K\\) only takes on finitely many values. In detail, \\[\np\\left(z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right) =\n\\frac{p\\left(y_n \\mid z_n = k, \\mu, \\sigma \\right) \\cdot p\\left(z_n = k \\mid \\lambda \\right)}\n     {\\sum_{k' = 1}^K p\\left(y_n \\mid z_n = k', \\mu, \\sigma \\right)\n                    \\cdot p\\left(z_n = k' \\mid \\lambda \\right)}.\n\\]\nOn the log scale, the normalized probability is computed as \\[\\begin{align*}\n\\log\\,&\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda\\right] \\\\\n&= \\log p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right) + \\log \\Pr\\!\\left[z_n = k \\mid \\lambda\\right] \\\\\n&\\quad  - \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{k' = 1}^K \\big(\\log p\\left(y_n \\mid z_n = k', \\mu, \\sigma\\right)\n                                                 + \\log p\\left(z_n = k' \\mid \\lambda\\right)\\big).\n\\end{align*}\\] This can be coded up directly in Stan; the change-point model in the change point section provides an example.\n\n\n\nGiven the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.\ndata {\n  int&lt;lower=1&gt; K;          // number of mixture components\n  int&lt;lower=1&gt; N;          // number of data points\n  array[N] real y;         // observations\n}\nparameters {\n  simplex[K] theta;          // mixing proportions\n  ordered[K] mu;             // locations of mixture components\n  vector&lt;lower=0&gt;[K] sigma;  // scales of mixture components\n}\nmodel {\n  vector[K] log_theta = log(theta);  // cache log calculation\n  sigma ~ lognormal(0, 2);\n  mu ~ normal(0, 10);\n  for (n in 1:N) {\n    vector[K] lps = log_theta;\n    for (k in 1:K) {\n      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);\n    }\n    target += log_sum_exp(lps);\n  }\n}\nThe model involves K mixture components and N data points. The mixing proportion parameter theta is declared to be a unit \\(K\\)-simplex, whereas the component location parameter mu and scale parameter sigma are both defined to be K-vectors.\nThe location parameter mu is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components mu[k] is symmetric, as it is here (each component has an independent \\(\\textsf{normal}(0, 10)\\) prior). It would even be possible to include a hierarchical prior for the components.\nThe values in the scale array sigma are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.\nThe model declares a local array variable lps to be size K and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points n. For each such point, the log of \\(\\theta_k \\times\n\\textsf{normal}\\left(y_n \\mid \\mu_k,\\sigma_k\\right)\\) is calculated and added to the array lps. Then the log probability is incremented with the log sum of exponentials of those values.",
+    "text": "Two problems make it pretty much impossible to perform full Bayesian inference for clustering models, the lack of parameter identifiability and the extreme multimodality of the posteriors. There is additional discussion related to the non-identifiability due to label switching in the label switching section.\n\n\nCluster assignments are not identified—permuting the cluster mean vectors mu leads to a model with identical likelihoods. For instance, permuting the first two indexes in mu and the first two indexes in each soft_z[n] leads to an identical likelihood (and prior).\nThe lack of identifiability means that the cluster parameters cannot be compared across multiple Markov chains. In fact, the only parameter in soft \\(K\\)-means is not identified, leading to problems in monitoring convergence. Clusters can even fail to be identified within a single chain, with indices swapping if the chain is long enough or the data are not cleanly separated.\n\n\n\nThe other problem with clustering models is that their posteriors are highly multimodal. One form of multimodality is the non-identifiability leading to index swapping. But even without the index problems the posteriors are highly multimodal.\nBayesian inference fails in cases of high multimodality because there is no way to visit all of the modes in the posterior in appropriate proportions and thus no way to evaluate integrals involved in posterior predictive inference.\nIn light of these two problems, the advice often given in fitting clustering models is to try many different initializations and select the sample with the highest overall probability. It is also popular to use optimization-based point estimators such as expectation maximization or variational Bayes, which can be much more efficient than sampling-based approaches.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#vectorizing-mixtures",
-    "href": "stan-users-guide/finite-mixtures.html#vectorizing-mixtures",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/clustering.html#naive-bayes-classification-and-clustering",
+    "href": "stan-users-guide/clustering.html#naive-bayes-classification-and-clustering",
+    "title": "Clustering Models",
     "section": "",
-    "text": "There is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that lambda, y[n], mu[1], mu[2], and sigma[1], sigma[2] are all scalars and lambda is between 0 and 1.\nfor (n in 1:N) {\n  target += log_sum_exp(log(lambda)\n                          + normal_lpdf(y[n] | mu[1], sigma[1]),\n                        log1m(lambda)\n                          + normal_lpdf(y[n] | mu[2], sigma[2]));\nor equivalently\nfor (n in 1:N) {\n  target += log_mix(lambda,\n                    normal_lpdf(y[n] | mu[1], sigma[1]),\n                    normal_lpdf(y[n] | mu[2], sigma[2]))\n  };\nThis definition assumes that each observation \\(y_n\\) may have arisen from either of the mixture components. The density is \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma\\right)\n= \\prod_{n=1}^N \\big(\\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1 \\right)\n                 + (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2 \\right)\\big).\n\\]\nContrast the previous model with the following (erroneous) attempt to vectorize the model.\ntarget += log_sum_exp(log(lambda)\n                        + normal_lpdf(y | mu[1], sigma[1]),\n                      log1m(lambda)\n                        + normal_lpdf(y | mu[2], sigma[2]));\nor equivalently,\ntarget += log_mix(lambda,\n                  normal_lpdf(y | mu[1], sigma[1]),\n                  normal_lpdf(y | mu[2], sigma[2]));\nThis second definition implies that the entire sequence \\(y_1, \\dotsc, y_n\\) of observations comes form one component or the other, defining a different density, \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n= \\lambda \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right).\n\\]",
+    "text": "Naive Bayes is a kind of mixture model that can be used for classification or for clustering (or a mix of both), depending on which labels for items are observed.1\nMultinomial mixture models are referred to as “naive Bayes” because they are often applied to classification problems where the multinomial independence assumptions are clearly false.\nNaive Bayes classification and clustering can be applied to any data with multinomial structure. A typical example of this is natural language text classification and clustering, which is used an example in what follows.\nThe observed data consists of a sequence of \\(M\\) documents made up of bags of words drawn from a vocabulary of \\(V\\) distinct words. A document \\(m\\) has \\(N_m\\) words, which are indexed as \\(w_{m,1}, \\dotsc,\nw_{m,N[m]} \\in \\{1,\\dotsc,V\\}\\). Despite the ordered indexing of words in a document, this order is not part of the model, which is clearly defective for natural human language data. A number of topics (or categories) \\(K\\) is fixed.\nThe multinomial mixture model generates a single category \\(z_m \\in\n\\{1,\\dotsc,K\\}\\) for each document \\(m \\in \\{1,\\dotsc,M\\}\\) according to a categorical distribution, \\[\nz_m \\sim \\textsf{categorical}(\\theta).\n\\] The \\(K\\)-simplex parameter \\(\\theta\\) represents the prevalence of each category in the data.\nNext, the words in each document are generated conditionally independently of each other and the words in other documents based on the category of the document, with word \\(n\\) of document \\(m\\) being generated as \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m]}).\n\\] The parameter \\(\\phi_{z[m]}\\) is a \\(V\\)-simplex representing the probability of each word in the vocabulary in documents of category \\(z_m\\).\nThe parameters \\(\\theta\\) and \\(\\phi\\) are typically given symmetric Dirichlet priors. The prevalence \\(\\theta\\) is sometimes fixed to produce equal probabilities for each category \\(k \\in \\{1,\\dotsc,K\\}\\).\n\n\nThe specification for naive Bayes in the previous sections have used a ragged array notation for the words \\(w\\). Because Stan does not support ragged arrays, the models are coded using an alternative strategy that provides an index for each word in a global list of words. The data is organized as follows, with the word arrays laid out in a column and each assigned to its document in a second column.\n\\[\n\\begin{array}{lll}\n\\hline\n\\mathrm{n} \\qquad\\qquad\\qquad\\qquad & \\mathrm{w[n]} \\qquad & \\mathrm{doc[n]} \\\\\n\\hline\n1         & w_{1,1}    & 1 \\\\\n2         & w_{1,2}    & 1 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1     & w_{1,N[1]} & 1 \\\\\nN_1 + 1 & w_{2,1}    & 2 \\\\\nN_1 + 2 & w_{2,2}    & 2 \\\\\n\\vdots & \\vdots    & \\vdots \\\\\nN_1 + N_2     & w_{2,N[2]} & 2 \\\\\nN_1 + N_2 + 1 & w_{3,1}    & 3 \\\\\n\\vdots       & \\vdots    & \\vdots \\\\\nN = \\sum_{m=1}^M N_m & w_{M,N[M]} & M \\\\\n\\hline\n\\end{array}\n\\]\nThe relevant variables for the program are N, the total number of words in all the documents, the word array w, and the document identity array doc.\n\n\n\nA naive Bayes model for estimating the simplex parameters given training data with documents of known categories can be coded in Stan as follows\ndata {\n  // training data\n  int&lt;lower=1&gt; K;               // num topics\n  int&lt;lower=1&gt; V;               // num words\n  int&lt;lower=0&gt; M;               // num docs\n  int&lt;lower=0&gt; N;               // total word instances\n  array[M] int&lt;lower=1, upper=K&gt; z;    // topic for doc m\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  // hyperparameters\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  simplex[K] theta;             // topic prevalence\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    z[m] ~ categorical(theta);\n  }\n  for (n in 1:N) {\n    w[n] ~ categorical(phi[z[doc[n]]]);\n  }\n}\nThe topic identifiers \\(z_m\\) are declared as data and the latent category assignments are included as part of the likelihood function.\n\n\n\nNaive Bayes models can be used in an unsupervised fashion to cluster multinomial-structured data into a fixed number \\(K\\) of categories. The data declaration includes the same variables as the model in the previous section excluding the topic labels z. Because z is discrete, it needs to be summed out of the model calculation. This is done for naive Bayes as for other mixture models. The parameters are the same up to the priors, but the likelihood is now computed as the marginal document probability\n\\[\\begin{align*}\n\\log\\, &p(w_{m,1},\\dotsc,w_{m,N_m} \\mid \\theta,\\phi) \\\\\n&= \\log \\sum_{k=1}^K\n    \\left( \\textsf{categorical}(k \\mid \\theta)\n           \\times \\prod_{n=1}^{N_m} \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right) \\\\\n&= \\log \\sum_{k=1}^K \\exp \\left(\n    \\log \\textsf{categorical}(k \\mid \\theta)\n     + \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k)\n    \\right).\n\\end{align*}\\]\nThe last step shows how the log_sum_exp function can be used to stabilize the numerical calculation and return a result on the log scale.\nmodel {\n  array[M, K] real gamma;\n  theta ~ dirichlet(alpha);\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);\n  }\n  for (m in 1:M) {\n    for (k in 1:K) {\n      gamma[m, k] = categorical_lpmf(k | theta);\n    }\n  }\n  for (n in 1:N) {\n    for (k in 1:K) {\n      gamma[doc[n], k] = gamma[doc[n], k]\n                         + categorical_lpmf(w[n] | phi[k]);\n    }\n  }\n  for (m in 1:M) {\n    target += log_sum_exp(gamma[m]);\n  }\n}\nThe local variable gamma[m, k] represents the value \\[\n\\gamma_{m,k} = \\log \\textsf{categorical}(k \\mid \\theta)\n+ \\sum_{n=1}^{N_m} \\log \\textsf{categorical}(w_{m,n} \\mid \\phi_k).\n\\]\nGiven \\(\\gamma\\), the posterior probability that document \\(m\\) is assigned category \\(k\\) is \\[\n\\Pr[z_m = k \\mid w,\\alpha,\\beta]\n=\n\\exp \\left(\n\\gamma_{m,k}\n- \\log \\sum_{k=1}^K \\exp \\left( \\gamma_{m,k} \\right)\n\\right).\n\\]\nIf the variable gamma were declared and defined in the transformed parameter block, its sampled values would be saved by Stan. The normalized posterior probabilities could also be defined as generated quantities.\n\n\n\nFull Bayesian posterior predictive inference for the naive Bayes model can be implemented in Stan by combining the models for labeled and unlabeled data. The estimands include both the model parameters and the posterior distribution over categories for the unlabeled data. The model is essentially a missing data model assuming the unknown category labels are missing completely at random; see Gelman et al. (2013) and Gelman and Hill (2007) for more information on missing data imputation. The model is also an instance of semisupervised learning because the unlabeled data contributes to the parameter estimations.\nTo specify a Stan model for performing full Bayesian inference, the model for labeled data is combined with the model for unlabeled data. A second document collection is declared as data, but without the category labels, leading to new variables M2 N2, w2, and doc2. The number of categories and number of words, as well as the hyperparameters are shared and only declared once. Similarly, there is only one set of parameters. Then the model contains a single set of statements for the prior, a set of statements for the labeled data, and a set of statements for the unlabeled data.\n\n\n\nAn alternative to full Bayesian inference involves estimating a model using labeled data, then applying it to unlabeled data without updating the parameter estimates based on the unlabeled data. This behavior can be implemented by moving the definition of gamma for the unlabeled documents to the generated quantities block. Because the variables no longer contribute to the log probability, they no longer jointly contribute to the estimation of the model parameters.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#mixture-inference.section",
-    "href": "stan-users-guide/finite-mixtures.html#mixture-inference.section",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/clustering.html#latent-dirichlet-allocation",
+    "href": "stan-users-guide/clustering.html#latent-dirichlet-allocation",
+    "title": "Clustering Models",
     "section": "",
-    "text": "In many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.\nWe have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs. female, or Democrat vs. Republican).\nIn other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.\n\n\nAs an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\), \\[\\begin{align*}\n\\mu_1, \\mu_2       &\\sim \\textsf{normal}(0, 10) \\\\\n\\sigma_1, \\sigma_2 &\\sim \\textsf{halfnormal}(0, 10) \\\\\n\\end{align*}\\]\nThe prior on the mixture ratio is uniform, \\[\n\\lambda \\sim \\textsf{uniform}(0, 1),\n\\] so that with the likelihood \\[\np\\left(y_n \\mid \\mu, \\sigma\\right)\n= \\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right),\n\\] the joint distribution \\(p(y, \\mu, \\sigma, \\lambda)\\) is exchangeable in the parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\) with \\(\\lambda\\) flipping to \\(1 - \\lambda\\).1\n\n\n\nIn cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.\n\n\nPosterior predictive distribution for a new observation \\(\\tilde{y}\\) given the complete parameter vector \\(\\theta\\) will be \\[\np(\\tilde{y} \\mid y)\n=\n\\int_{\\theta}\np(\\tilde{y} \\mid \\theta)\n\\, p(\\theta \\mid y)\n\\, \\textsf{d}\\theta.\n\\]\nThe normal mixture example from the previous section, with \\(\\theta =\n(\\mu, \\sigma, \\lambda)\\), shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing \\(p(\\tilde{y} \\mid y)\\), which is more efficient statistically in terms of effective sample size, or simulating draws of \\(\\tilde{y}\\), which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] y_tilde;\n  // ...\n}\ngenerated quantities {\n  vector[N_tilde] log_p_y_tilde;\n  for (n in 1:N_tilde) {\n    log_p_y_tilde[n]\n      = log_mix(lambda,\n                normal_lpdf(y_tilde[n] | mu[1], sigma[1])\n                normal_lpdf(y_tilde[n] | mu[2], sigma[2]));\n  }\n}\nIt is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply log_sum_exp of the posterior draws of log_p_y_tilde. The average log predictive density is then given by subtracting log(N_new).\n\n\n\nOften a mixture model will be applied to a clustering problem and there might be two data items \\(y_i\\) and \\(y_j\\) for which there is a question of whether they arose from the same mixture component. If we take \\(z_i\\) and \\(z_j\\) to be the component responsibility discrete variables, then the quantity of interest is \\(z_i = z_j\\), which can be summarized as an event probability \\[\n\\Pr[z_i = z_j \\mid y]\n=\n\\int_{\\theta}\n\\frac{\\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \\mid \\theta)}\n     {\\sum_{k=0}^1 \\sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \\mid \\theta)}\n\\,\np(\\theta \\mid y)\n\\,\n\\textsf{d}\\theta.\n\\]\nAs with other event probabilities, this can be calculated in the generated quantities block either by sampling \\(z_i\\) and \\(z_j\\) and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.",
+    "text": "Latent Dirichlet allocation (LDA) is a mixed-membership multinomial clustering model (Blei, Ng, and Jordan 2003) that generalizes naive Bayes. Using the topic and document terminology common in discussions of LDA, each document is modeled as having a mixture of topics, with each word drawn from a topic based on the mixing proportions.\n\n\nThe basic model assumes each document is generated independently based on fixed hyperparameters. For document \\(m\\), the first step is to draw a topic distribution simplex \\(\\theta_m\\) over the \\(K\\) topics, \\[\n\\theta_m \\sim \\textsf{Dirichlet}(\\alpha).\n\\]\nThe prior hyperparameter \\(\\alpha\\) is fixed to a \\(K\\)-vector of positive values. Each word in the document is generated independently conditional on the distribution \\(\\theta_m\\). First, a topic \\(z_{m,n} \\in \\{1,\\dotsc,K\\}\\) is drawn for the word based on the document-specific topic-distribution, \\[\nz_{m,n} \\sim \\textsf{categorical}(\\theta_m).\n\\]\nFinally, the word \\(w_{m,n}\\) is drawn according to the word distribution for topic \\(z_{m,n}\\), \\[\nw_{m,n} \\sim \\textsf{categorical}(\\phi_{z[m,n]}).\n\\] The distributions \\(\\phi_k\\) over words for topic \\(k\\) are also given a Dirichlet prior, \\[\n\\phi_k \\sim \\textsf{Dirichlet}(\\beta)\n\\]\nwhere \\(\\beta\\) is a fixed \\(V\\)-vector of positive values.\n\n\n\nAlthough Stan does not (yet) support discrete sampling, it is possible to calculate the marginal distribution over the continuous parameters by summing out the discrete parameters as in other mixture models. The marginal posterior of the topic and word variables is \\[\\begin{align*}\np(\\theta,\\phi \\mid w,\\alpha,\\beta)\n&\\propto p(\\theta \\mid \\alpha) \\, p(\\phi \\mid \\beta) \\, p(w \\mid \\theta,\\phi) \\\\\n&= \\prod_{m=1}^M p(\\theta_m \\mid \\alpha)\n    \\times \\prod_{k=1}^K p(\\phi_k \\mid \\beta)\n    \\times \\prod_{m=1}^M \\prod_{n=1}^{M[n]} p(w_{m,n} \\mid \\theta_m,\\phi).\n\\end{align*}\\]\nThe inner word-probability term is defined by summing out the topic assignments, \\[\\begin{align*}\np(w_{m,n} \\mid \\theta_m,\\phi)\n&= \\sum_{z=1}^K p(z,w_{m,n} \\mid \\theta_m,\\phi) \\\\\n&= \\sum_{z=1}^K p(z \\mid \\theta_m) \\, p(w_{m,n} \\mid \\phi_z).\n\\end{align*}\\]\nPlugging the distributions in and converting to the log scale provides a formula that can be implemented directly in Stan, \\[\\begin{align*}\n\\log\\, &p(\\theta,\\phi \\mid w,\\alpha,\\beta) \\\\\n&= \\sum_{m=1}^M \\log \\textsf{Dirichlet}(\\theta_m \\mid \\alpha)\n    + \\sum_{k=1}^K \\log \\textsf{Dirichlet}(\\phi_k \\mid \\beta) \\\\\n&\\qquad + \\sum_{m=1}^M \\sum_{n=1}^{N[m]} \\log \\left(\n    \\sum_{z=1}^K \\textsf{categorical}(z \\mid \\theta_m)\n    \\times \\textsf{categorical}(w_{m,n} \\mid \\phi_z)\n  \\right)\n\\end{align*}\\]\n\n\n\nApplying the marginal derived in the last section to the data structure described in this section leads to the following Stan program for LDA.\ndata {\n  int&lt;lower=2&gt; K;               // num topics\n  int&lt;lower=2&gt; V;               // num words\n  int&lt;lower=1&gt; M;               // num docs\n  int&lt;lower=1&gt; N;               // total word instances\n  array[N] int&lt;lower=1, upper=V&gt; w;    // word n\n  array[N] int&lt;lower=1, upper=M&gt; doc;  // doc ID for word n\n  vector&lt;lower=0&gt;[K] alpha;     // topic prior\n  vector&lt;lower=0&gt;[V] beta;      // word prior\n}\nparameters {\n  array[M] simplex[K] theta;    // topic dist for doc m\n  array[K] simplex[V] phi;      // word dist for topic k\n}\nmodel {\n  for (m in 1:M) {\n    theta[m] ~ dirichlet(alpha);  // prior\n  }\n  for (k in 1:K) {\n    phi[k] ~ dirichlet(beta);     // prior\n  }\n  for (n in 1:N) {\n    array[K] real gamma;\n    for (k in 1:K) {\n      gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);\n    }\n    target += log_sum_exp(gamma);  // likelihood;\n  }\n}\nAs in the other mixture models, the log-sum-of-exponents function is used to stabilize the numerical arithmetic.\n\n\n\nTo account for correlations in the distribution of topics for documents, Blei and Lafferty (2007) introduced a variant of LDA in which the Dirichlet prior on the per-document topic distribution is replaced with a multivariate logistic normal distribution.\nThe authors treat the prior as a fixed hyperparameter. They use an \\(L_1\\)-regularized estimate of covariance, which is equivalent to the maximum a posteriori estimate given a double-exponential prior. Here the mean and covariance of the multivariate logistic normal are specified as data.\n\n\nThe Stan model in the previous section can be modified to implement the correlated topic model by replacing the Dirichlet topic prior alpha in the data declaration with the mean and covariance of the multivariate logistic normal prior.\ndata {\n  // ... data as before without alpha ...\n  vector[K] mu;          // topic mean\n  cov_matrix[K] Sigma;   // topic covariance\n}\nRather than drawing the simplex parameter theta from a Dirichlet, a parameter eta is drawn from a multivariate normal distribution and then transformed using softmax into a simplex.\nparameters {\n  array[K] simplex[V] phi;     // word dist for topic k\n  array[M] vector[K] eta;      // topic dist for doc m\n}\ntransformed parameters {\n  array[M] simplex[K] theta;\n  for (m in 1:M) {\n    theta[m] = softmax(eta[m]);\n  }\n}\nmodel {\n  for (m in 1:M) {\n    eta[m] ~ multi_normal(mu, Sigma);\n  }\n  // ... model as before w/o prior for theta ...\n}\n\n\n\nBy adding a prior for the mean and covariance, Stan supports full Bayesian inference for the correlated topic model. This requires moving the declarations of topic mean mu and covariance Sigma from the data block to the parameters block and providing them with priors in the model. A relatively efficient and interpretable prior for the covariance matrix Sigma may be encoded as follows.\n// ... data block as before, but without alpha ...\nparameters {\n  vector[K] mu;              // topic mean\n  corr_matrix[K] Omega;      // correlation matrix\n  vector&lt;lower=0&gt;[K] sigma;  // scales\n  array[M] vector[K] eta;    // logit topic dist for doc m\n  array[K] simplex[V] phi;   // word dist for topic k\n}\ntransformed parameters {\n  // ... eta as above ...\n  cov_matrix[K] Sigma;       // covariance matrix\n  for (m in 1:K) {\n    Sigma[m, m] = sigma[m] * sigma[m] * Omega[m, m];\n  }\n  for (m in 1:(K-1)) {\n    for (n in (m+1):K) {\n      Sigma[m, n] = sigma[m] * sigma[n] * Omega[m, n];\n      Sigma[n, m] = Sigma[m, n];\n    }\n  }\n}\nmodel {\n  mu ~ normal(0, 5);      // vectorized, diffuse\n  Omega ~ lkj_corr(2.0);  // regularize to unit correlation\n  sigma ~ cauchy(0, 5);   // half-Cauchy due to constraint\n  // ... words sampled as above ...\n}\nThe \\(\\textsf{LKJCorr}\\) distribution with shape \\(\\alpha &gt; 0\\) has support on correlation matrices (i.e., symmetric positive definite with unit diagonal). Its density is defined by \\[\n\\mathsf{LkjCorr}(\\Omega\\mid\\alpha) \\propto \\mathrm{det}(\\Omega)^{\\alpha - 1}\n\\] With a scale of \\(\\alpha = 2\\), the weakly informative prior favors a unit correlation matrix. Thus the compound effect of this prior on the covariance matrix \\(\\Sigma\\) for the multivariate logistic normal is a slight concentration around diagonal covariance matrices with scales determined by the prior on sigma.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#zero-inflated.section",
-    "href": "stan-users-guide/finite-mixtures.html#zero-inflated.section",
-    "title": "Finite Mixtures",
-    "section": "",
-    "text": "Zero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by Lambert (1992), add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.\nZero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.\n\n\nConsider the following example for zero-inflated Poisson distributions. There is a probability \\(\\theta\\) of observing a zero, and a probability \\(1 - \\theta\\) of observing a count with a \\(\\textsf{Poisson}(\\lambda)\\) distribution (now \\(\\theta\\) is being used for mixing proportions because \\(\\lambda\\) is the traditional notation for a Poisson mean parameter). Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as \\[\\begin{align*}\ny_n & = 0 & \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}(y_n \\mid \\lambda) & \\quad\\text{with probability } 1-\\theta.\n\\end{align*}\\]\nStan does not support conditional distribution statements (with ~) conditional on some parameter, and we need to consider the corresponding likelihood \\[\np(y_n \\mid \\theta,\\lambda)\n=\n\\begin{cases}\n\\theta + (1 - \\theta) \\times \\textsf{Poisson}(0 \\mid \\lambda) & \\quad\\text{if } y_n = 0, \\text{ and}\\\\\n(1-\\theta) \\times \\textsf{Poisson}(y_n \\mid \\lambda) &\\quad\\text{if } y_n &gt; 0.\n\\end{cases}\n\\] The log likelihood can be coded directly in Stan (with target +=) as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  for (n in 1:N) {\n    if (y[n] == 0) {\n      target += log_sum_exp(log(theta),\n                            log1m(theta)\n                              + poisson_lpmf(y[n] | lambda));\n    } else {\n      target += log1m(theta)\n                  + poisson_lpmf(y[n] | lambda);\n    }\n  }\n}\nThe log1m(theta) computes log(1-theta), but is more computationally stable. The log_sum_exp(lp1,lp2) function adds the log probabilities on the linear scale; it is defined to be equal to log(exp(lp1) + exp(lp2)), but is more computationally stable and faster.\n\n\nThe code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also poisson_lpmf(0 | lambda) every time the first condition body executes. The use of the redundant terms is conditioned on y, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.\nThe number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare y_nonzero, so it must be computed in a function.\nfunctions {\n  int num_zeros(array[] int y) {\n    int sum = 0;\n    for (n in 1:size(y)) {\n      sum += (y[n] == 0);\n    }\n    return sum;\n  }\n}\n// ...\ntransformed data {\n  int&lt;lower=0&gt; N_zero = num_zeros(y);\n  array[N - N_zero] int&lt;lower=1&gt; y_nonzero;\n  int N_nonzero = 0;\n  for (n in 1:N) {\n    if (y[n] == 0) continue;\n    N_nonzero += 1;\n    y_nonzero[N_nonzero] = y[n];\n  }\n}\n// ...\nmodel {\n  // ...\n   target\n     += N_zero\n          * log_sum_exp(log(theta),\n                        log1m(theta)\n                          + poisson_lpmf(0 | lambda));\n   target += N_nonzero * log1m(theta);\n   target += poisson_lpmf(y_nonzero | lambda);\n  // ...\n}\nThe boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if y_nonzero is empty, N_nonzero will be zero, and the last two target increment terms will add zeros.\n\n\n\n\nThe hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as [ \\[\\begin{align*}\ny_n & = 0 \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}_{x\\neq 0}(y_n \\mid \\lambda) \\quad\\text{with probability } 1-\\theta,\n\\end{align*}\\] ] Where \\(\\textsf{Poisson}_{x\\neq 0}\\) is a truncated Poisson distribution, truncated at \\(0\\).\nThe corresponding likelihood function for the hurdle model is defined by \\[\np(y\\mid\\theta,\\lambda)\n=\n\\begin{cases}\n\\theta &\\quad\\text{if } y = 0, \\text{ and}\\\\\n(1 - \\theta)\n   \\frac{\\displaystyle \\textsf{Poisson}(y \\mid \\lambda)}\n        {\\displaystyle  1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)}\n&\\quad\\text{if } y &gt; 0,\n\\end{cases}\n\\] where \\(\\textsf{PoissonCDF}\\) is the cumulative distribution function for the Poisson distribution and and \\(1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)\\) is the relative normalization term for the truncated Poisson (truncated at \\(0\\)).\nThe hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.\nif (y[n] == 0) {\n  target += log(theta);\n} else {\n  target += log1m(theta) + poisson_lpmf(y[n] | lambda)\n            - poisson_lccdf(0 | lambda));\n}\nJulian King pointed out that because \\[\\begin{align*}\n\\log \\left( 1 - \\textsf{PoissonCDF}(0 \\mid \\lambda) \\right)\n  &= \\log \\left( 1 - \\textsf{Poisson}(0 \\mid \\lambda) \\right) \\\\\n  &= \\log(1 - \\exp(-\\lambda))\n\\end{align*}\\] the CCDF in the else clause can be replaced with a simpler expression.\ntarget += log1m(theta) + poisson_lpmf(y[n] | lambda)\n          - log1m_exp(-lambda));\nThe resulting code is about 15% faster than the code with the CCDF.\nThis is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size \\(N=200\\) and parameters \\(\\theta=0.3\\) and \\(\\lambda = 8\\), the speedup is a factor of 10; it will be lower for smaller \\(N\\) and greater for larger \\(N\\); it will also be greater for larger \\(\\theta\\).\nTo achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,\nfunctions {\n  int num_zero(array[] int y) {\n    int nz = 0;\n    for (n in 1:size(y)) {\n      if (y[n] == 0) {\n        nz += 1;\n      }\n    }\n    return nz;\n  }\n}\nThen a transformed data block can be used to store the sufficient statistics,\ntransformed data {\n  int&lt;lower=0, upper=N&gt; N0 = num_zero(y);\n  int&lt;lower=0, upper=N&gt; Ngt0 = N - N0;\n  array[N - num_zero(y)] int&lt;lower=1&gt; y_nz;\n  {\n    int pos = 1;\n    for (n in 1:N) {\n      if (y[n] != 0) {\n        y_nz[pos] = y[n];\n        pos += 1;\n      }\n    }\n  }\n}\nThe model block is then reduced to three statements.\nmodel {\n  N0 ~ binomial(N, theta);\n  y_nz ~ poisson(lambda);\n  target += -Ngt0 * log1m_exp(-lambda);\n}\nThe first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant Ngt0; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.",
+    "objectID": "stan-users-guide/clustering.html#footnotes",
+    "href": "stan-users-guide/clustering.html#footnotes",
+    "title": "Clustering Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nFor clustering, the non-identifiability problems for all mixture models present a problem, whereas there is no such problem for classification. Despite the difficulties with full Bayesian inference for clustering, researchers continue to use it, often in an exploratory data analysis setting rather than for predictive modeling.↩︎",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Clustering Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#priors-and-effective-data-size-in-mixture-models",
-    "href": "stan-users-guide/finite-mixtures.html#priors-and-effective-data-size-in-mixture-models",
-    "title": "Finite Mixtures",
+    "objectID": "stan-users-guide/copulas.html",
+    "href": "stan-users-guide/copulas.html",
+    "title": "Copulas",
     "section": "",
-    "text": "Suppose we have a two-component mixture model with mixing rate \\(\\lambda \\in (0, 1)\\). Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are \\(N\\) observations, the mixture components will be estimated with effective data sizes of \\(\\theta \\, N\\) and \\((1 - \\theta)\n\\, N\\) for the two components for some \\(\\theta \\in (0, 1)\\). The effective weighting size is determined by posterior responsibility, not simply by the mixing rate \\(\\lambda\\).\n\n\nIn contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data \\(y\\).\nIf different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see Hoeting et al. (1999) and Gelman et al. (2013) for theoretical and computational details).",
+    "text": "Copulas provide a flexible way to model multivariate distributions by separating the marginal cumulative distribution functions from the dependence structure. This chapter introduces copulas in Stan, focusing on implementation techniques and practical examples. This chapter was derived from Brynjólfur Gauti Guðrúnar Jónsson’s A gentle introduction: the Gaussian copula.\n\n\nAccording to Sklar’s theorem (Sklar 1959), any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.\nFor a multivariate random variable \\(\\mathbf{X} = [X_1 \\cdots X_D]^\\top\\) with marginal cumulative distribution functions \\(F_i\\), the joint cumulative distribution function can be written as:\n\\[\nF_{\\mathbf{X}}(\\mathbf{x}) = C(F_1(x_1), \\ldots, F_D(x_D)) = \\Pr[X_1 \\leq x_1, \\ldots, X_D \\leq x_D]\n\\]\nwhere \\(C\\) is the copula function, \\(F_{\\mathbf{X}}\\) is the joint cumulative distribution function, and \\(F_i\\) are the marginal cumulative distribution functions. The copula function \\(C\\) must be a joint cumulative distribution function over the unit hypercube \\([0, 1]^D\\).\n\n\n\nThis section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.\nThe log density of a multivariate distribution using a copula can be written as:\n\\[\n\\log h(\\mathbf{x}) = \\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right) + \\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\n\\]\nwhere:\n\n\\(u_i = F_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) are the probability integral transforms of the data\n\\(\\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right)\\) is the log density of the copula\n\\(\\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) is the sum of the log densities of the marginals\n\\(\\boldsymbol{\\alpha}\\) represents the parameters describing the parametric form of the copula\n\\(\\boldsymbol{\\beta}_i\\) represents the parameters describing the parametric form of the \\(i\\)-th marginal distribution\n\nThe implementation of copulas in Stan has two key requirements:\n\nBoth the probability density functions and cumulative distribution functions of the marginal distributions must be available\nA function that computes the log density of the copula for the transformed data must be implemented\n\nMost copula implementations in Stan follow a three-step process:\n\nAccumulate marginal log likelihoods: Calculate and add the log density of each marginal distribution to the target log density\nTransform to uniform variables: Apply the marginal CDFs to transform the data to uniform variables on the unit interval\nCalculate copula density: Compute the log density of the copula based on these uniform variables and add it to the target log density\n\nThis process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.\nIn a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where \\(\\log c(\\mathbf{u}) = 0\\), resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\nMost parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).\n\n\n\nThe Gaussian copula is constructed using the multivariate normal distribution. For a \\(D\\)-dimensional random vector \\(\\mathbf{X}\\) with marginals \\(F_i\\), the log Gaussian copula density is given by:\n\\[\n\\begin{aligned}\n\\log c(\\mathbf{u}) &=\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top (\\boldsymbol{\\Omega}^{-1} - \\mathbf{I}) \\mathbf{z} \\\\\n& =\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top \\boldsymbol{\\Omega}^{-1} \\mathbf{z} + \\frac{1}{2} \\mathbf{z}^\\top \\mathbf{z} \\\\\n&= \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\boldsymbol{\\Omega}) - \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\mathbf{I})\n\\end{aligned}\n\\]\nwhere \\(\\mathbf{z} = [\\Phi^{-1}(u_1), \\ldots, \\Phi^{-1}(u_D)]^\\top\\) are the inverse normal CDF transforms of the uniform marginals, \\(\\boldsymbol{\\Omega}\\) is the correlation matrix, and \\(\\mathbf{I}\\) is the identity matrix. The joint log density is then:\n\\[\n\\log h(\\mathbf{x}) = \\log c(F_1(x_1), \\ldots, F_D(x_D)) + \\sum_{i=1}^D \\log f_i(x_i)\n\\]\nFollowing the three-step process for implementing copulas in Stan:\n\nAccumulate marginal log likelihoods: The exponential log densities are added to the target in the line target += exponential_lpdf(y[n] | lambda)\nTransform to uniform variables: The exponential CDF transforms the data to uniform variables: exponential_cdf(y[n, d] | lambda[d])\nCalculate copula density: The transformed variables are converted to normal scale using inv_Phi and the multivariate normal log density is computed: z ~ multi_normal_cholesky(zeros, L_Omega)\n\nThe following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  int&lt;lower=0&gt; D;  // number of dimensions\n  vector&lt;lower=0&gt;[D] y[N];  // data\n}\n\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\n\nparameters {\n  // Parameters for exponential marginal distributions\n  vector&lt;lower=0&gt;[D] lambda;  // rate parameters\n  \n  // Correlation matrix for Gaussian copula\n  cholesky_factor_corr[D] L_Omega;\n}\n\nmodel {\n  // Priors\n  lambda ~ gamma(2, 1);  // prior for rate parameters\n  L_Omega ~ lkj_corr_cholesky(2);\n  \n  // Likelihood using Gaussian copula with exponential marginals\n  for (n in 1:N) {\n    // Add exponential log density to target\n    target += exponential_lpdf(y[n] | lambda);\n    \n    vector[D] z;\n    for (d in 1:D) {\n      // Transform to uniform using exponential CDF\n      real u_d = exponential_cdf(y[n, d] | lambda[d]);\n      \n      // Transform to standard normal\n      z[d] = inv_Phi(u_d);\n    }\n    // Multivariate normal log density with correlation matrix\n    z ~ multi_normal_cholesky(zeros, L_Omega);\n  }\n}\n\ngenerated quantities {\n  // Optional: Recover correlation matrix from Cholesky factor\n  matrix[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\n\n\n\nCopulas offer several advantages in statistical modeling:\n\nFlexibility: They allow combining any marginal distributions with various dependence structures. For example:\n\nModeling financial returns with heavy-tailed marginals and complex dependence structures\nCombining different types of distributions (e.g., normal and gamma) in a single model\nCapturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic\nModeling different types of tail dependence in different parts of the distribution\n\nFactorability: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.\nFor example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.\nTail dependence: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.\nUniversal Framework: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\n\n\n\n\nWhen implementing copulas in Stan, several considerations should be kept in mind:\n\nComputational efficiency: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.\nParameter identifiability: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.\nModel selection: The choice of copula family should be guided by the specific dependence structure of the data. For example:\n\nThe Gaussian copula may underestimate the probability of joint extreme events in financial data\nThe Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications\nArchimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions\n\nNumerical stability: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.\nSymmetry considerations: Many copula families exhibit strong symmetries that may not match the data:\n\nRadial symmetry: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns\nExchangeability: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.\n\nTail dependence: Understanding and choosing appropriate tail dependence is crucial:\n\nThe upper (lower) tail dependence coefficient \\(\\lambda_U (\\lambda_L)\\) is the probability that one variable is extremely large (small) given that another is extremely large (small).\nDifferent copula families exhibit different tail dependence properties:\n\nSome copulas (like Gaussian) have zero tail dependence\nOthers can model symmetric tail dependence (\\(\\lambda_U = \\lambda_L\\))\nSome can capture asymmetric tail dependence (\\(\\lambda_U \\neq \\lambda_L\\))\nCertain copulas allow for tail dependence even with zero correlation\n\nThe choice of copula should be guided by the expected tail behavior in the application:\n\nFinancial data often requires modeling joint lower extreme events\nRisk management applications may need asymmetric tail dependence\nSome applications may require different tail behavior in different parts of the distribution\n\n\nHigh-dimensional modeling: As dimensionality increases:\n\nThe number of dependence parameters grows\nSome copula families become less flexible\nVine copulas or factor copulas may be more appropriate\n\n\n\n\n\nSeveral copula families are available for modeling different dependence structures in the correlation component:\n\nGaussian copula: Based on the multivariate normal distribution, offering symmetric dependence\nStudent-t copula: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula\nArchimedean copulas: A class of copulas defined through generator functions, including:\n\nClayton copula: Stronger lower tail dependence\nGumbel copula: Stronger upper tail dependence\nFrank copula: Symmetric dependence\n\nVine copulas: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas\n\n\n\n\n\nJónsson’s three part blog series, Copulas in Stan:\n\nPart I: If it bleeds, we can kill it\nPart II: A gentle introduction: the Gaussian copula\nPart III: It was the best of tails, it was the worst of tails: The T-Copula\n\nBrynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, Copulas in Stan: Modeling Spatial Dependence\nSean Pinkney’s Helpful Stan functions: copula functions",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/finite-mixtures.html#footnotes",
-    "href": "stan-users-guide/finite-mixtures.html#footnotes",
-    "title": "Finite Mixtures",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nImposing a constraint such as \\(\\theta &lt; 0.5\\) will resolve the symmetry, but fundamentally changes the model and its posterior inferences.↩︎",
+    "objectID": "stan-users-guide/copulas.html#what-are-copulas",
+    "href": "stan-users-guide/copulas.html#what-are-copulas",
+    "title": "Copulas",
+    "section": "",
+    "text": "According to Sklar’s theorem (Sklar 1959), any multivariate distribution can be expressed in terms of its marginals and a copula that captures the dependence structure. Copulas are functions that join univariate marginal cumulative distribution functions to form multivariate distributions.\nFor a multivariate random variable \\(\\mathbf{X} = [X_1 \\cdots X_D]^\\top\\) with marginal cumulative distribution functions \\(F_i\\), the joint cumulative distribution function can be written as:\n\\[\nF_{\\mathbf{X}}(\\mathbf{x}) = C(F_1(x_1), \\ldots, F_D(x_D)) = \\Pr[X_1 \\leq x_1, \\ldots, X_D \\leq x_D]\n\\]\nwhere \\(C\\) is the copula function, \\(F_{\\mathbf{X}}\\) is the joint cumulative distribution function, and \\(F_i\\) are the marginal cumulative distribution functions. The copula function \\(C\\) must be a joint cumulative distribution function over the unit hypercube \\([0, 1]^D\\).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Finite Mixtures"
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html",
-    "href": "stan-users-guide/for-bugs-users.html",
-    "title": "Transitioning from BUGS",
+    "objectID": "stan-users-guide/copulas.html#general-structure-of-copula-models-in-stan",
+    "href": "stan-users-guide/copulas.html#general-structure-of-copula-models-in-stan",
+    "title": "Copulas",
     "section": "",
-    "text": "From the outside, Stan and BUGS1 are similar—they use statistically-themed modeling languages (which are similar but with some differences; see below), they can be called from R, running some specified number of chains to some specified length, producing posterior simulations that can be assessed using standard convergence diagnostics. This is not a coincidence: in designing Stan: we wanted to keep many of the useful features of Bugs.\n\n\n\n\nStan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).\n\n\n\nBUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.\n\n\n\nBUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).\n\n\n\nThe BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).\n\n\n\nIn BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets mu before using it to sample y:\nmu = a + b * x;\ny ~ normal(mu, sigma);\ntranslates to the following C++ code:\nmu = a + b * x;\ntarget += normal_lpdf(y | mu, sigma);\nContrast this with the following Stan program:\ny ~ normal(mu, sigma);\nmu = a + b * x;\nThis program is well formed, but is almost certainly a coding error, because it attempts to use mu before it is set. The direct translation to C++ code highlights the potential error of using mu in the first statement:\ntarget += normal_lpdf(y | mu, sigma);\nmu = a + b * x;\nTo trap these kinds of errors, variables are initialized to the special not-a-number (NaN) value. If NaN is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.\n\n\n\nStan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.\n\n\n\nBoth BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (\\(\\hat{R}\\)) and the effective sample size to judge when to stop (stopping when \\(\\hat{R}\\) values do not counter-indicate convergence and when high enough effective sample size have been obtained).\n\n\n\nWinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.\n\n\n\nLike WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.\n\n\n\nLike OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.\n\n\n\n\nThe BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:\ny ~ normal(mu, sigma);\nand\nfor (i in 1:n) y[i] ~ normal(mu, sigma);\nOr, equivalently (recall that a line break is just another form of whitespace),\nfor (i in 1:n)\n  y[i] ~ normal(mu, sigma);\nand also equivalently,\nfor (i in 1:n) {\n  y[i] ~ normal(mu, sigma);\n}\nThere’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.\nIn Stan, variables can have names constructed using letters, numbers, and the underscore (_) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (.) symbol.\nIn Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in Bayesian Data Analysis) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is normal(1,2), not normal(1,4) or normal(1,0.25).\nSimilarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.\nThe distributions have slightly different names:\n\n\n\nBUGS\nStan\n\n\n\n\ndnorm\nnormal\n\n\ndbinom\nbinomial\n\n\ndpois\npoisson\n\n\n…\n…\n\n\n\nStan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.\n{\n  total = 0;\n  for (i in 1:n) {\n    theta[i] ~ normal(total, sigma);\n    total = total + theta[i];\n  }\n}\nIn BUGS, the above model would not be legal because the variable total is defined more than once. But in Stan, the loop is executed in order, so total is overwritten in each step.\nStan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.\nIn Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.\nStan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,2 allowing\nfor (i in 1:n) {\n  y[i] ~ normal(mu[i], sigma[i]);\n}\nto be expressed more compactly as\ny ~ normal(mu, sigma);\nThe vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.\nStan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of K parameters for each of J groups; this can be declared using\narray[J] vector[K] theta;\nThen theta[j] is an expression denoting a K-vector and may be used in the code just like any other vector variable.\nAn alternative encoding would be with a two-dimensional array, as in\narray[J, K] real theta;\nThe vector version can have some advantages, both in convenience and in computational speed for some operations.\nA third encoding would use a matrix:\nmatrix[J, K] theta;\nbut in this case, theta[j] is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in theta[j]', may be used to convert the row vector theta[j] to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.\nStan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the zero inflation section.\nStan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop\nmodel {\n    // ...\n    for (n in 1:N) {\n        // ... do something with n ....\n    }\n}\nmay be recoded using the following while loop.\nmodel {\n    int n;\n    // ...\n    n = 1;\n    while (n &lt;= N) {\n        // ... do something with n ...\n        n = n + 1;\n    }\n}\n\n\n\nStan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.\nStan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in Lewandowski, Kurowicka, and Joe (2009). In particular, the Lewandowski et al. prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.\nIn BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter p declared as\nreal&lt;lower=0, upper=1&gt; p;\nand then have no distribution statement for p in the model block, then you are implicitly assigning a uniform \\([0,1]\\) prior on p.\nOn the other hand, if you have a parameter theta declared with\nreal theta;\nand have no distribution statement for theta in the model block, then you are implicitly assigning an improper uniform prior on \\((-\\infty,\\infty)\\) to theta.\nBUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:\nparameters {\n  real theta;\n}\nmodel { }\nAlthough parameters in Stan models may have improper priors, we do not want improper posterior distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.\nEach statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:\ny ~ normal(0,1);\ny ~ normal(2,3);\nThe second line here does not simply overwrite the first; rather, both statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, \\(\\textsf{normal}(y \\mid 0,1) * \\textsf{normal}(y \\mid 2,3)\\), into the density function.\nFor a perhaps more confusing example, consider the following two lines in a Stan model:\nx ~ normal(0.8 * y, sigma);\ny ~ normal(0.8 * x, sigma);\nAt first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are not interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, \\(\\textsf{normal}(x \\mid 0.8y,\\sigma)\n\\times \\textsf{normal}(y \\mid 0.8x,\\sigma)\\), which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.\nWith censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in Gelman et al. (2013) and Gelman and Hill (2007) may be implemented in Stan directly as written.\nStan, like BUGS, can benefit from human intervention in the form of reparameterization.\n\n\n\nStan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.\nIn practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.\nWhen you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)\nWhen running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable x from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.\nAs in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using plot() and print(). You can access the simulations themselves using various extractor functions, as described in the RStan documentation.\nVarious information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.\n\n\n\nStan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the Stan web site for information on subscribing and posting and to look at archives.",
+    "text": "This section describes the general structure of copula models in Stan. The next sections will provide specific examples of copula implementations, but first, let’s understand the general pattern that separates the marginal distributions from the dependence structure.\nThe log density of a multivariate distribution using a copula can be written as:\n\\[\n\\log h(\\mathbf{x}) = \\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right) + \\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\n\\]\nwhere:\n\n\\(u_i = F_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) are the probability integral transforms of the data\n\\(\\log c\\left(u_1, \\dots, u_D \\vert \\boldsymbol{\\alpha}\\right)\\) is the log density of the copula\n\\(\\sum_{i=1}^D \\log f_i(x_i \\vert \\boldsymbol{\\beta}_i)\\) is the sum of the log densities of the marginals\n\\(\\boldsymbol{\\alpha}\\) represents the parameters describing the parametric form of the copula\n\\(\\boldsymbol{\\beta}_i\\) represents the parameters describing the parametric form of the \\(i\\)-th marginal distribution\n\nThe implementation of copulas in Stan has two key requirements:\n\nBoth the probability density functions and cumulative distribution functions of the marginal distributions must be available\nA function that computes the log density of the copula for the transformed data must be implemented\n\nMost copula implementations in Stan follow a three-step process:\n\nAccumulate marginal log likelihoods: Calculate and add the log density of each marginal distribution to the target log density\nTransform to uniform variables: Apply the marginal CDFs to transform the data to uniform variables on the unit interval\nCalculate copula density: Compute the log density of the copula based on these uniform variables and add it to the target log density\n\nThis process is reflected in the general form of the log density shown above, where the first term represents the copula density and the second term represents the sum of marginal log densities.\nIn a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula, where \\(\\log c(\\mathbf{u}) = 0\\), resulting in the familiar sum of marginal log densities. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.\nMost parametric copula families include independence as a special case, either as a subset of their parameter space (e.g., when correlation parameters are zero) or as a limit when parameters approach specific values (e.g., when the dependence parameter approaches zero in Archimedean copulas).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Example Models",
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-how-bugs-and-stan-work",
-    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-how-bugs-and-stan-work",
-    "title": "Transitioning from BUGS",
+    "objectID": "stan-users-guide/copulas.html#gaussian-copula-example",
+    "href": "stan-users-guide/copulas.html#gaussian-copula-example",
+    "title": "Copulas",
     "section": "",
-    "text": "Stan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).\n\n\n\nBUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.\n\n\n\nBUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).\n\n\n\nThe BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).\n\n\n\nIn BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets mu before using it to sample y:\nmu = a + b * x;\ny ~ normal(mu, sigma);\ntranslates to the following C++ code:\nmu = a + b * x;\ntarget += normal_lpdf(y | mu, sigma);\nContrast this with the following Stan program:\ny ~ normal(mu, sigma);\nmu = a + b * x;\nThis program is well formed, but is almost certainly a coding error, because it attempts to use mu before it is set. The direct translation to C++ code highlights the potential error of using mu in the first statement:\ntarget += normal_lpdf(y | mu, sigma);\nmu = a + b * x;\nTo trap these kinds of errors, variables are initialized to the special not-a-number (NaN) value. If NaN is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.\n\n\n\nStan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.\n\n\n\nBoth BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (\\(\\hat{R}\\)) and the effective sample size to judge when to stop (stopping when \\(\\hat{R}\\) values do not counter-indicate convergence and when high enough effective sample size have been obtained).\n\n\n\nWinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.\n\n\n\nLike WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.\n\n\n\nLike OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.",
+    "text": "The Gaussian copula is constructed using the multivariate normal distribution. For a \\(D\\)-dimensional random vector \\(\\mathbf{X}\\) with marginals \\(F_i\\), the log Gaussian copula density is given by:\n\\[\n\\begin{aligned}\n\\log c(\\mathbf{u}) &=\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top (\\boldsymbol{\\Omega}^{-1} - \\mathbf{I}) \\mathbf{z} \\\\\n& =\n-\\frac{1}{2} \\log |\\boldsymbol{\\Omega}| -\\frac{1}{2} \\mathbf{z}^\\top \\boldsymbol{\\Omega}^{-1} \\mathbf{z} + \\frac{1}{2} \\mathbf{z}^\\top \\mathbf{z} \\\\\n&= \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\boldsymbol{\\Omega}) - \\log \\mathcal{N}(\\mathbf{z} \\mid \\mathbf{0}, \\mathbf{I})\n\\end{aligned}\n\\]\nwhere \\(\\mathbf{z} = [\\Phi^{-1}(u_1), \\ldots, \\Phi^{-1}(u_D)]^\\top\\) are the inverse normal CDF transforms of the uniform marginals, \\(\\boldsymbol{\\Omega}\\) is the correlation matrix, and \\(\\mathbf{I}\\) is the identity matrix. The joint log density is then:\n\\[\n\\log h(\\mathbf{x}) = \\log c(F_1(x_1), \\ldots, F_D(x_D)) + \\sum_{i=1}^D \\log f_i(x_i)\n\\]\nFollowing the three-step process for implementing copulas in Stan:\n\nAccumulate marginal log likelihoods: The exponential log densities are added to the target in the line target += exponential_lpdf(y[n] | lambda)\nTransform to uniform variables: The exponential CDF transforms the data to uniform variables: exponential_cdf(y[n, d] | lambda[d])\nCalculate copula density: The transformed variables are converted to normal scale using inv_Phi and the multivariate normal log density is computed: z ~ multi_normal_cholesky(zeros, L_Omega)\n\nThe following example demonstrates a Gaussian copula with exponential marginal distributions. Note that while the copula is Gaussian, the marginals are exponential.\ndata {\n  int&lt;lower=0&gt; N;  // number of observations\n  int&lt;lower=0&gt; D;  // number of dimensions\n  vector&lt;lower=0&gt;[D] y[N];  // data\n}\n\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\n\nparameters {\n  // Parameters for exponential marginal distributions\n  vector&lt;lower=0&gt;[D] lambda;  // rate parameters\n  \n  // Correlation matrix for Gaussian copula\n  cholesky_factor_corr[D] L_Omega;\n}\n\nmodel {\n  // Priors\n  lambda ~ gamma(2, 1);  // prior for rate parameters\n  L_Omega ~ lkj_corr_cholesky(2);\n  \n  // Likelihood using Gaussian copula with exponential marginals\n  for (n in 1:N) {\n    // Add exponential log density to target\n    target += exponential_lpdf(y[n] | lambda);\n    \n    vector[D] z;\n    for (d in 1:D) {\n      // Transform to uniform using exponential CDF\n      real u_d = exponential_cdf(y[n, d] | lambda[d]);\n      \n      // Transform to standard normal\n      z[d] = inv_Phi(u_d);\n    }\n    // Multivariate normal log density with correlation matrix\n    z ~ multi_normal_cholesky(zeros, L_Omega);\n  }\n}\n\ngenerated quantities {\n  // Optional: Recover correlation matrix from Cholesky factor\n  matrix[D, D] Omega = multiply_lower_tri_self_transpose(L_Omega);\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Example Models",
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-the-modeling-languages",
-    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-the-modeling-languages",
-    "title": "Transitioning from BUGS",
+    "objectID": "stan-users-guide/copulas.html#advantages-of-copulas",
+    "href": "stan-users-guide/copulas.html#advantages-of-copulas",
+    "title": "Copulas",
     "section": "",
-    "text": "The BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:\ny ~ normal(mu, sigma);\nand\nfor (i in 1:n) y[i] ~ normal(mu, sigma);\nOr, equivalently (recall that a line break is just another form of whitespace),\nfor (i in 1:n)\n  y[i] ~ normal(mu, sigma);\nand also equivalently,\nfor (i in 1:n) {\n  y[i] ~ normal(mu, sigma);\n}\nThere’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.\nIn Stan, variables can have names constructed using letters, numbers, and the underscore (_) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (.) symbol.\nIn Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in Bayesian Data Analysis) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is normal(1,2), not normal(1,4) or normal(1,0.25).\nSimilarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.\nThe distributions have slightly different names:\n\n\n\nBUGS\nStan\n\n\n\n\ndnorm\nnormal\n\n\ndbinom\nbinomial\n\n\ndpois\npoisson\n\n\n…\n…\n\n\n\nStan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.\n{\n  total = 0;\n  for (i in 1:n) {\n    theta[i] ~ normal(total, sigma);\n    total = total + theta[i];\n  }\n}\nIn BUGS, the above model would not be legal because the variable total is defined more than once. But in Stan, the loop is executed in order, so total is overwritten in each step.\nStan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.\nIn Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.\nStan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,2 allowing\nfor (i in 1:n) {\n  y[i] ~ normal(mu[i], sigma[i]);\n}\nto be expressed more compactly as\ny ~ normal(mu, sigma);\nThe vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.\nStan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of K parameters for each of J groups; this can be declared using\narray[J] vector[K] theta;\nThen theta[j] is an expression denoting a K-vector and may be used in the code just like any other vector variable.\nAn alternative encoding would be with a two-dimensional array, as in\narray[J, K] real theta;\nThe vector version can have some advantages, both in convenience and in computational speed for some operations.\nA third encoding would use a matrix:\nmatrix[J, K] theta;\nbut in this case, theta[j] is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in theta[j]', may be used to convert the row vector theta[j] to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.\nStan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the zero inflation section.\nStan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop\nmodel {\n    // ...\n    for (n in 1:N) {\n        // ... do something with n ....\n    }\n}\nmay be recoded using the following while loop.\nmodel {\n    int n;\n    // ...\n    n = 1;\n    while (n &lt;= N) {\n        // ... do something with n ...\n        n = n + 1;\n    }\n}",
+    "text": "Copulas offer several advantages in statistical modeling:\n\nFlexibility: They allow combining any marginal distributions with various dependence structures. For example:\n\nModeling financial returns with heavy-tailed marginals and complex dependence structures\nCombining different types of distributions (e.g., normal and gamma) in a single model\nCapturing asymmetric dependencies between variables, such as in financial markets where joint negative returns are more common than joint positive returns due to macro-events affecting multiple stocks simultaneously, while positive returns tend to be more idiosyncratic\nModeling different types of tail dependence in different parts of the distribution\n\nFactorability: The marginal distributions and dependence structure can be modeled separately, allowing for different prior knowledge about each component. This is similar to the common practice of factoring scale and correlation in multivariate normal priors.\nFor example, when modeling the survival times of two components in a system, we can separately specify exponential or gamma marginal distributions based on historical failure data for each component, and a Gaussian copula (or asymmetrical Archimedean copula) capturing how the failure of one component affects the other, making it easier to incorporate prior knowledge about each aspect independently.\nTail dependence: Different copulas can capture different types of tail dependence, which is crucial in applications like risk management and extreme value analysis where joint extreme scenarios need to be quantified.\nUniversal Framework: In a way, we are always modeling with copulas, as the independence assumption can be viewed as a special case using the independence copula. This perspective highlights that traditional independent modeling is just a specific case within the broader copula framework.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Example Models",
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-the-statistical-models-that-are-allowed",
-    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-the-statistical-models-that-are-allowed",
-    "title": "Transitioning from BUGS",
+    "objectID": "stan-users-guide/copulas.html#common-pitfalls-and-considerations",
+    "href": "stan-users-guide/copulas.html#common-pitfalls-and-considerations",
+    "title": "Copulas",
     "section": "",
-    "text": "Stan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.\nStan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in Lewandowski, Kurowicka, and Joe (2009). In particular, the Lewandowski et al. prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.\nIn BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter p declared as\nreal&lt;lower=0, upper=1&gt; p;\nand then have no distribution statement for p in the model block, then you are implicitly assigning a uniform \\([0,1]\\) prior on p.\nOn the other hand, if you have a parameter theta declared with\nreal theta;\nand have no distribution statement for theta in the model block, then you are implicitly assigning an improper uniform prior on \\((-\\infty,\\infty)\\) to theta.\nBUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:\nparameters {\n  real theta;\n}\nmodel { }\nAlthough parameters in Stan models may have improper priors, we do not want improper posterior distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.\nEach statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:\ny ~ normal(0,1);\ny ~ normal(2,3);\nThe second line here does not simply overwrite the first; rather, both statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, \\(\\textsf{normal}(y \\mid 0,1) * \\textsf{normal}(y \\mid 2,3)\\), into the density function.\nFor a perhaps more confusing example, consider the following two lines in a Stan model:\nx ~ normal(0.8 * y, sigma);\ny ~ normal(0.8 * x, sigma);\nAt first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are not interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, \\(\\textsf{normal}(x \\mid 0.8y,\\sigma)\n\\times \\textsf{normal}(y \\mid 0.8x,\\sigma)\\), which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.\nWith censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in Gelman et al. (2013) and Gelman and Hill (2007) may be implemented in Stan directly as written.\nStan, like BUGS, can benefit from human intervention in the form of reparameterization.",
+    "text": "When implementing copulas in Stan, several considerations should be kept in mind:\n\nComputational efficiency: The probability integral transform and inverse transform steps can be computationally intensive, especially for complex marginal distributions.\nParameter identifiability: Care must be taken to ensure that the parameters of the marginal distributions and the copula are identifiable.\nModel selection: The choice of copula family should be guided by the specific dependence structure of the data. For example:\n\nThe Gaussian copula may underestimate the probability of joint extreme events in financial data\nThe Student-t copula, while offering tail dependence, maintains symmetric tail behavior that may not match all applications\nArchimedean copulas can model asymmetric tail dependence but may be less flexible and harder to estimate in high dimensions\n\nNumerical stability: The transformations between different scales (original, uniform, and normal/Student-t/calculations using Archimedian copulas) require careful implementation to maintain numerical stability.\nSymmetry considerations: Many copula families exhibit strong symmetries that may not match the data:\n\nRadial symmetry: Some copulas (like Gaussian and Student-t) treat positive and negative extremes equally, which may not match financial data where joint negative returns are more common than joint positive returns\nExchangeability: Some copulas are invariant under permutations of their arguments, which can lead to unintuitive results when combined with inhomogeneous marginals. For example, when modeling time-to-event scenarios with different marginal distributions (e.g., exponential distributions with different parameters), perfect dependence in the copula does not imply simultaneous events. Instead, one event triggers the other at a later time corresponding to the same quantile, which can lead to incorrect modeling of joint events.\n\nTail dependence: Understanding and choosing appropriate tail dependence is crucial:\n\nThe upper (lower) tail dependence coefficient \\(\\lambda_U (\\lambda_L)\\) is the probability that one variable is extremely large (small) given that another is extremely large (small).\nDifferent copula families exhibit different tail dependence properties:\n\nSome copulas (like Gaussian) have zero tail dependence\nOthers can model symmetric tail dependence (\\(\\lambda_U = \\lambda_L\\))\nSome can capture asymmetric tail dependence (\\(\\lambda_U \\neq \\lambda_L\\))\nCertain copulas allow for tail dependence even with zero correlation\n\nThe choice of copula should be guided by the expected tail behavior in the application:\n\nFinancial data often requires modeling joint lower extreme events\nRisk management applications may need asymmetric tail dependence\nSome applications may require different tail behavior in different parts of the distribution\n\n\nHigh-dimensional modeling: As dimensionality increases:\n\nThe number of dependence parameters grows\nSome copula families become less flexible\nVine copulas or factor copulas may be more appropriate",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Example Models",
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-when-running-from-r",
-    "href": "stan-users-guide/for-bugs-users.html#some-differences-when-running-from-r",
-    "title": "Transitioning from BUGS",
+    "objectID": "stan-users-guide/copulas.html#common-copula-families",
+    "href": "stan-users-guide/copulas.html#common-copula-families",
+    "title": "Copulas",
     "section": "",
-    "text": "Stan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.\nIn practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.\nWhen you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)\nWhen running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable x from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.\nAs in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using plot() and print(). You can access the simulations themselves using various extractor functions, as described in the RStan documentation.\nVarious information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.",
+    "text": "Several copula families are available for modeling different dependence structures in the correlation component:\n\nGaussian copula: Based on the multivariate normal distribution, offering symmetric dependence\nStudent-t copula: Based on the multivariate Student-t distribution, providing more flexibility in tail dependence than the Gaussian copula\nArchimedean copulas: A class of copulas defined through generator functions, including:\n\nClayton copula: Stronger lower tail dependence\nGumbel copula: Stronger upper tail dependence\nFrank copula: Symmetric dependence\n\nVine copulas: A flexible approach for modeling high-dimensional dependencies by decomposing the joint distribution into a series of bivariate copulas",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Example Models",
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html#the-stan-community",
-    "href": "stan-users-guide/for-bugs-users.html#the-stan-community",
-    "title": "Transitioning from BUGS",
+    "objectID": "stan-users-guide/copulas.html#further-readingviewing",
+    "href": "stan-users-guide/copulas.html#further-readingviewing",
+    "title": "Copulas",
     "section": "",
-    "text": "Stan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the Stan web site for information on subscribing and posting and to look at archives.",
+    "text": "Jónsson’s three part blog series, Copulas in Stan:\n\nPart I: If it bleeds, we can kill it\nPart II: A gentle introduction: the Gaussian copula\nPart III: It was the best of tails, it was the worst of tails: The T-Copula\n\nBrynjólfur Gauti Guðrúnar Jónsson’s StanCon 2024 presentation, Copulas in Stan: Modeling Spatial Dependence\nSean Pinkney’s Helpful Stan functions: copula functions",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Example Models",
+      "Copulas"
     ]
   },
   {
-    "objectID": "stan-users-guide/for-bugs-users.html#footnotes",
-    "href": "stan-users-guide/for-bugs-users.html#footnotes",
-    "title": "Transitioning from BUGS",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nExcept where otherwise noted, we use “BUGS” to refer to WinBUGS, OpenBUGS, and JAGS, indiscriminately.↩︎\nMost distributions have been vectorized, but currently the truncated versions may not exist.↩︎",
+    "objectID": "stan-users-guide/custom-probability.html",
+    "href": "stan-users-guide/custom-probability.html",
+    "title": "Custom Probability Functions",
+    "section": "",
+    "text": "Custom distributions may also be implemented directly within Stan’s programming language. The only thing that is needed is to increment the total log probability. The rest of the chapter provides examples.\n\n\n\n\nA simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}\\) are the bounds, with \\(\\alpha &lt; \\beta\\), then \\(y \\in (\\alpha,\\beta)\\) has a density defined as \\[\n\\textsf{triangle}(y \\mid \\alpha,\\beta)\n=\n\\frac{1}{(\\beta - \\alpha)^2}\n\\cdot\n\\textrm{min}(y - \\alpha, \\beta - y).\n\\]\nThe general form of triangle can be coded in Stan as follows.\ndata {\n  real alpha;\n  real&lt;lower=alpha&gt; beta;\n}\nparameters {\n  real&lt;lower=alpha, upper=beta&gt; y;\n}\nmodel {\n  target += -2 * log(beta - alpha)\n    + log(fmin(y - alpha, beta - y));\n}\nBecause the bounds are specified as data here, the term -2 * log(beta - alpha) could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.\nIf \\(\\alpha = -1\\), \\(\\beta = 1\\), and \\(y \\in (-1,1)\\), then fmin(y - alpha, beta - y) is fmin(y + 1, -1 - y), which is fmin(y + 1, -(y + 1)), which reduces to 1 - abs(y). Therfore, the density, dropping constants, reduces to \\[\n\\textsf{triangle}(y \\mid -1, 1) \\propto 1 - |y|.\n\\] Consider the following Stan implementation of \\(\\textsf{triangle}(-1,1)\\) for sampling.\nparameters {\n  real&lt;lower=-1, upper=1&gt; y;\n}\nmodel {\n  target += log1m(abs(y));\n}\nThe single scalar parameter y is declared as lying in the interval (-1, 1). The total log probability is incremented with the joint log probability of all parameters, i.e., \\(\\log \\mathsf{Triangle}(y \\mid -1, 1)\\). This value is coded in Stan as log1m(abs(y)). The function log1m is defined so that log1m(x) has the same value as \\(\\log(1 - x)\\), but the computation is faster, more accurate, and more stable.\nThe constrained type real&lt;lower=-1, upper=1&gt; declared for y is critical for correct sampling behavior. If the constraint on y is removed from the program, say by declaring y as having the unconstrained scalar type real, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of y outside of \\((-1,1)\\).\nNow suppose the log probability function were extended to all of \\(\\mathbb{R}\\) as follows by defining the probability to be log(0.0), i.e., \\(-\\infty\\), for values outside of \\((-1, 1)\\).\ntarget += log(fmax(0.0,1 - abs(y)));\nWith the constraint on y in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on y is removed, the model will compile and run without arithmetic errors, but will not sample properly.1\n\n\n\nIf Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where lambda is the inverse scale and y the sampled variate.\ntarget += log(lambda) - y * lambda;\nThis encoding will work for any lambda and y; they can be parameters, data, or one of each, or even local variables.\nThe assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.\ny ~ exponential(lambda);\nThere are two notable differences. First, the distribution statement will check the inputs to make sure both lambda is positive and y is non-negative (which includes checking that neither is the special not-a-number value).\nThe second difference is that if lambda is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the log(lambda) term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If lambda and y are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).\n\n\n\nFor another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation rho. That is, it computes \\[\n\\texttt{binormal}\\mathtt{\\_}\\texttt{cdf}(z_1, z_2, \\rho) = \\Pr[Z_1 \\leq z_1 \\text{ and } Z_2 \\leq z_2]\n\\] where the random 2-vector \\(Z\\) has the distribution \\[\nZ \\sim \\textsf{multivariate normal}\\left(\n\\begin{bmatrix}\n0 \\\\\n0\n\\end{bmatrix}, \\\n\\begin{bmatrix}\n1 & \\rho\n\\\\\n\\rho & 1\n\\end{bmatrix}\n\\right).\n\\]\nThe following Stan program implements this function,\nreal binormal_cdf(tuple(real, real) z, real rho) {\n  real z1 = z.1;\n  real z2 = z.2;\n  if (z1 == 0 && z2 == 0) {\n    return 0.25 + asin(rho) / (2 * pi());\n  }\n  real denom = sqrt((1 + rho) * (1 - rho));\n  real term1 = z1 == 0\n    ? (z2 &gt; 0 ? 0.25 : -0.25)\n    :  owens_t(z1, (z2 / z1 - rho) / denom);\n  real term2 = z2 == 0\n    ? (z1 &gt; 0 ? 0.25 : -0.25)\n    : owens_t(z2, (z1 / z2 - rho) / denom);\n  real z1z2 = z1 * z2;\n  real delta = z1z2 &lt; 0 || (z1z2 == 0 && (z1 + z2) &lt; 0);\n  return 0.5 * (Phi(z1) + Phi(z2) - delta) - term1 - term2;\n}\nIt is written using a tuple argument so that it may be called as binormal_cdf((z1, z2) | rho). The two ternary operators defining term1 and term2 are derived by taking the limit of the owens_t function when the second argument goes to infinity.\nOne way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.\ntransformed data {\n  for (zzr in {[0, 0, 0.5],\n               [0, 1, 0.5],\n               [1, 0, -0.2],\n               [1, 3, 0.9]}) {\n    real z1  = zzr[1];\n    real z2 = zzr[2];\n    real rho = zzr[3];\n    print(\"binomial_cdf((\",  z1, \", \", z2, \") | \", rho, \")\",\n          \"=\", binormal_cdf((z1, z2) | rho));\n  }\n}\nIn this case, we verified that the results match those of the pbivnorm package in R.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Transitioning from BUGS"
+      "Programming Techniques",
+      "Custom Probability Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/hyperspherical-models.html",
-    "href": "stan-users-guide/hyperspherical-models.html",
-    "title": "Directions, Rotations, and Hyperspheres",
+    "objectID": "stan-users-guide/custom-probability.html#examples",
+    "href": "stan-users-guide/custom-probability.html#examples",
+    "title": "Custom Probability Functions",
     "section": "",
-    "text": "Directional statistics involve data and/or parameters that are constrained to be directions. The set of directions forms a sphere, the geometry of which is not smoothly mappable to that of a Euclidean space because you can move around a sphere and come back to where you started. This is why it is impossible to make a map of the globe on a flat piece of paper where all points that are close to each other on the globe are close to each other on the flat map. The fundamental problem is easy to visualize in two dimensions, because as you move around a circle, you wind up back where you started. In other words, 0 degrees and 360 degrees (equivalently, 0 and \\(2 \\pi\\) radians) pick out the same point, and the distance between 359 degrees and 2 degrees is the same as the distance between 137 and 140 degrees.\nStan supports directional statistics by providing a unit-vector data type, the values of which determine points on a hypersphere (circle in two dimensions, sphere in three dimensions).\n\n\nThe length of a vector \\(x \\in \\mathbb{R}^K\\) is given by \\[\n\\Vert x \\Vert\n= \\sqrt{x^{\\top}\\,x}\n= \\sqrt{x_1^2 + x_2^2 + \\cdots + x_K^2}.\n\\] Unit vectors are defined to be vectors of unit length (i.e., length one).\nWith a variable declaration such as\nunit_vector[K] x;\nthe value of x will be constrained to be a vector of size K with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.\nWarning: An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (lp__ in some interfaces). The underlying source of the problem is that a unit vector of size \\(K\\) has only \\(K - 1\\) degrees of freedom. But there is no way to map those \\(K - 1\\) degrees of freedom continuously to \\(\\mathbb{R}^N\\)—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses \\(K\\) unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.\n\n\n\nAn \\(n\\)-sphere, written \\(S^{n}\\), is defined as the set of \\((n +\n1)\\)-dimensional unit vectors, \\[\nS^{n} = \\left\\{ x \\in \\mathbb{R}^{n+1} \\: : \\: \\Vert x \\Vert = 1 \\right\\}.\n\\]\nEven though \\(S^n\\) is made up of points in \\((n+1)\\) dimensions, it is only an \\(n\\)-dimensional manifold. For example, \\(S^2\\) is defined as a set of points in \\(\\mathbb{R}^3\\), but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by \\(S^2\\) in \\(\\mathbb{R}^3\\) behaves locally like a plane, i.e., \\(\\mathbb{R}^2\\). However, the overall shape of \\(S^2\\) is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (\\(S^2\\)) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.\nEven though \\(S^{n-1}\\) behaves locally like \\(\\mathbb{R}^{n-1}\\), there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at \\(2\\pi\\) radians in natural units), they do not produce a smooth map.\nLike a bounded interval \\((a, b)\\), in geometric terms, a sphere is compact in that the distance between any two points is bounded.\n\n\n\nStan (inverse) transforms arbitrary points in \\(\\mathbb{R}^{K+1}\\) to points in \\(S^K\\) using the auxiliary variable approach of Muller (1959). A point \\(y \\in \\mathbb{R}^K\\) is transformed to a point \\(x \\in S^{K-1}\\) by \\[\nx = \\frac{y}{\\sqrt{y^{\\top} y}}.\n\\]\nThe problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. Muller (1959) introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nUnit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle \\(\\theta\\). Given an angle \\(\\theta\\), a matrix can be defined, the pre-multiplication by which rotates a point by an angle of \\(\\theta\\). For angle \\(\\theta\\) (in two dimensions), the \\(2 \\times 2\\) rotation matrix is defined by \\[\nR_{\\theta}\n=\n\\begin{bmatrix}\n\\cos \\theta & -\\sin \\theta \\\\\n\\sin \\theta &  \\cos \\theta\n\\end{bmatrix}.\n\\] Given a two-dimensional vector \\(x\\), \\(R_{\\theta} \\, x\\) is the rotation of \\(x\\) (around the origin) by \\(\\theta\\) degrees.\n\n\nAngles can be calculated from unit vectors. For example, a random variable theta representing an angle in \\((-\\pi, \\pi)\\) radians can be declared as a two-dimensional unit vector then transformed to an angle.\nparameters {\n  unit_vector[2] xy;\n}\ntransformed parameters {\n  real&lt;lower=-pi(), upper=pi()&gt; theta = atan2(xy[2], xy[1]);\n}\nIf the distribution of \\((x, y)\\) is uniform over a circle, then the distribution of \\(\\arctan \\frac{y}{x}\\) is uniform over \\((-\\pi, \\pi)\\).\nIt might be tempting to try to just declare theta directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values \\(-\\pi\\) and \\(\\pi\\) are at \\(-\\infty\\) and \\(\\infty\\) on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.\nWith a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.\n\n\n\n\nA 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.\nIn human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.",
+    "text": "A simple example is the symmetric triangle distribution, whose density is shaped like an isosceles triangle with corners at specified bounds and height determined by the constraint that a density integrate to 1. If \\(\\alpha \\in \\mathbb{R}\\) and \\(\\beta \\in \\mathbb{R}\\) are the bounds, with \\(\\alpha &lt; \\beta\\), then \\(y \\in (\\alpha,\\beta)\\) has a density defined as \\[\n\\textsf{triangle}(y \\mid \\alpha,\\beta)\n=\n\\frac{1}{(\\beta - \\alpha)^2}\n\\cdot\n\\textrm{min}(y - \\alpha, \\beta - y).\n\\]\nThe general form of triangle can be coded in Stan as follows.\ndata {\n  real alpha;\n  real&lt;lower=alpha&gt; beta;\n}\nparameters {\n  real&lt;lower=alpha, upper=beta&gt; y;\n}\nmodel {\n  target += -2 * log(beta - alpha)\n    + log(fmin(y - alpha, beta - y));\n}\nBecause the bounds are specified as data here, the term -2 * log(beta - alpha) could be dropped from the log density. If either of the bounds depends on a parameter, then this term must be included.\nIf \\(\\alpha = -1\\), \\(\\beta = 1\\), and \\(y \\in (-1,1)\\), then fmin(y - alpha, beta - y) is fmin(y + 1, -1 - y), which is fmin(y + 1, -(y + 1)), which reduces to 1 - abs(y). Therfore, the density, dropping constants, reduces to \\[\n\\textsf{triangle}(y \\mid -1, 1) \\propto 1 - |y|.\n\\] Consider the following Stan implementation of \\(\\textsf{triangle}(-1,1)\\) for sampling.\nparameters {\n  real&lt;lower=-1, upper=1&gt; y;\n}\nmodel {\n  target += log1m(abs(y));\n}\nThe single scalar parameter y is declared as lying in the interval (-1, 1). The total log probability is incremented with the joint log probability of all parameters, i.e., \\(\\log \\mathsf{Triangle}(y \\mid -1, 1)\\). This value is coded in Stan as log1m(abs(y)). The function log1m is defined so that log1m(x) has the same value as \\(\\log(1 - x)\\), but the computation is faster, more accurate, and more stable.\nThe constrained type real&lt;lower=-1, upper=1&gt; declared for y is critical for correct sampling behavior. If the constraint on y is removed from the program, say by declaring y as having the unconstrained scalar type real, the program would compile, but it would produce arithmetic exceptions at run time when the sampler explored values of y outside of \\((-1,1)\\).\nNow suppose the log probability function were extended to all of \\(\\mathbb{R}\\) as follows by defining the probability to be log(0.0), i.e., \\(-\\infty\\), for values outside of \\((-1, 1)\\).\ntarget += log(fmax(0.0,1 - abs(y)));\nWith the constraint on y in place, this is just a less efficient, slower, and less arithmetically stable version of the original program. But if the constraint on y is removed, the model will compile and run without arithmetic errors, but will not sample properly.1\n\n\n\nIf Stan didn’t happen to include the exponential distribution, it could be coded directly using the following assignment statement, where lambda is the inverse scale and y the sampled variate.\ntarget += log(lambda) - y * lambda;\nThis encoding will work for any lambda and y; they can be parameters, data, or one of each, or even local variables.\nThe assignment statement in the previous paragraph generates C++ code that is similar to that generated by the following distribution statement.\ny ~ exponential(lambda);\nThere are two notable differences. First, the distribution statement will check the inputs to make sure both lambda is positive and y is non-negative (which includes checking that neither is the special not-a-number value).\nThe second difference is that if lambda is not a parameter, transformed parameter, or local model variable, the distribution statement is clever enough to drop the log(lambda) term. This results in the same posterior because Stan only needs the log probability up to an additive constant. If lambda and y are both constants, the distribution statement will drop both terms (but still check for out-of-domain errors on the inputs).\n\n\n\nFor another example of user-defined functions, consider the following definition of the bivariate normal cumulative distribution function (CDF) with location zero, unit variance, and correlation rho. That is, it computes \\[\n\\texttt{binormal}\\mathtt{\\_}\\texttt{cdf}(z_1, z_2, \\rho) = \\Pr[Z_1 \\leq z_1 \\text{ and } Z_2 \\leq z_2]\n\\] where the random 2-vector \\(Z\\) has the distribution \\[\nZ \\sim \\textsf{multivariate normal}\\left(\n\\begin{bmatrix}\n0 \\\\\n0\n\\end{bmatrix}, \\\n\\begin{bmatrix}\n1 & \\rho\n\\\\\n\\rho & 1\n\\end{bmatrix}\n\\right).\n\\]\nThe following Stan program implements this function,\nreal binormal_cdf(tuple(real, real) z, real rho) {\n  real z1 = z.1;\n  real z2 = z.2;\n  if (z1 == 0 && z2 == 0) {\n    return 0.25 + asin(rho) / (2 * pi());\n  }\n  real denom = sqrt((1 + rho) * (1 - rho));\n  real term1 = z1 == 0\n    ? (z2 &gt; 0 ? 0.25 : -0.25)\n    :  owens_t(z1, (z2 / z1 - rho) / denom);\n  real term2 = z2 == 0\n    ? (z1 &gt; 0 ? 0.25 : -0.25)\n    : owens_t(z2, (z1 / z2 - rho) / denom);\n  real z1z2 = z1 * z2;\n  real delta = z1z2 &lt; 0 || (z1z2 == 0 && (z1 + z2) &lt; 0);\n  return 0.5 * (Phi(z1) + Phi(z2) - delta) - term1 - term2;\n}\nIt is written using a tuple argument so that it may be called as binormal_cdf((z1, z2) | rho). The two ternary operators defining term1 and term2 are derived by taking the limit of the owens_t function when the second argument goes to infinity.\nOne way to test a user-defined function is to have it operate on transformed data. That way, when it’s run, the output of the functions is printed before sampling begins.\ntransformed data {\n  for (zzr in {[0, 0, 0.5],\n               [0, 1, 0.5],\n               [1, 0, -0.2],\n               [1, 3, 0.9]}) {\n    real z1  = zzr[1];\n    real z2 = zzr[2];\n    real rho = zzr[3];\n    print(\"binomial_cdf((\",  z1, \", \", z2, \") | \", rho, \")\",\n          \"=\", binormal_cdf((z1, z2) | rho));\n  }\n}\nIn this case, we verified that the results match those of the pbivnorm package in R.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Directions, Rotations, and Hyperspheres"
+      "Programming Techniques",
+      "Custom Probability Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/hyperspherical-models.html#unit-vectors",
-    "href": "stan-users-guide/hyperspherical-models.html#unit-vectors",
-    "title": "Directions, Rotations, and Hyperspheres",
-    "section": "",
-    "text": "The length of a vector \\(x \\in \\mathbb{R}^K\\) is given by \\[\n\\Vert x \\Vert\n= \\sqrt{x^{\\top}\\,x}\n= \\sqrt{x_1^2 + x_2^2 + \\cdots + x_K^2}.\n\\] Unit vectors are defined to be vectors of unit length (i.e., length one).\nWith a variable declaration such as\nunit_vector[K] x;\nthe value of x will be constrained to be a vector of size K with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.\nWarning: An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (lp__ in some interfaces). The underlying source of the problem is that a unit vector of size \\(K\\) has only \\(K - 1\\) degrees of freedom. But there is no way to map those \\(K - 1\\) degrees of freedom continuously to \\(\\mathbb{R}^N\\)—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses \\(K\\) unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.",
+    "objectID": "stan-users-guide/custom-probability.html#footnotes",
+    "href": "stan-users-guide/custom-probability.html#footnotes",
+    "title": "Custom Probability Functions",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe problem is the (extremely!) light tails of the triangle distribution. The standard HMC and NUTS samplers can’t get into the corners of the triangle properly. Because the Stan code declares y to be of type real&lt;lower=-1, upper=1&gt;, the inverse logit transform is applied to the unconstrained variable and its log absolute derivative added to the log probability. The resulting distribution on the logit-transformed y is well behaved.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Directions, Rotations, and Hyperspheres"
+      "Programming Techniques",
+      "Custom Probability Functions"
     ]
   },
   {
-    "objectID": "stan-users-guide/hyperspherical-models.html#circles-spheres-and-hyperspheres",
-    "href": "stan-users-guide/hyperspherical-models.html#circles-spheres-and-hyperspheres",
-    "title": "Directions, Rotations, and Hyperspheres",
+    "objectID": "stan-users-guide/decision-analysis.html",
+    "href": "stan-users-guide/decision-analysis.html",
+    "title": "Decision Analysis",
     "section": "",
-    "text": "An \\(n\\)-sphere, written \\(S^{n}\\), is defined as the set of \\((n +\n1)\\)-dimensional unit vectors, \\[\nS^{n} = \\left\\{ x \\in \\mathbb{R}^{n+1} \\: : \\: \\Vert x \\Vert = 1 \\right\\}.\n\\]\nEven though \\(S^n\\) is made up of points in \\((n+1)\\) dimensions, it is only an \\(n\\)-dimensional manifold. For example, \\(S^2\\) is defined as a set of points in \\(\\mathbb{R}^3\\), but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by \\(S^2\\) in \\(\\mathbb{R}^3\\) behaves locally like a plane, i.e., \\(\\mathbb{R}^2\\). However, the overall shape of \\(S^2\\) is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (\\(S^2\\)) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.\nEven though \\(S^{n-1}\\) behaves locally like \\(\\mathbb{R}^{n-1}\\), there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at \\(2\\pi\\) radians in natural units), they do not produce a smooth map.\nLike a bounded interval \\((a, b)\\), in geometric terms, a sphere is compact in that the distance between any two points is bounded.",
+    "text": "Statistical decision analysis is about making decisions under uncertainty. In order to make decisions, outcomes must have some notion of “utility” associated with them. The so-called “Bayes optimal” decision is the one that maximizes expected utility (or equivalently, minimizes expected loss). This chapter shows how Stan can be used to simultaneously estimate the distribution of outcomes based on decisions and compute the required expected utilities.\n\n\nFollowing Gelman et al. (2013), Bayesian decision analysis can be factored into the following four steps.\n\nDefine a set \\(X\\) of possible outcomes and a set \\(D\\) of possible decisions.\nDefine a probability distribution of outcomes conditional on decisions through a conditional density function \\(p(x \\mid d)\\) for \\(x \\in X\\) and \\(d \\in D.\\)\nDefine a utility function \\(U : X \\rightarrow \\mathbb{R}\\) mapping outcomes to their utility.\nChoose action \\(d^* \\in D\\) with highest expected utility, \\[\nd^* = \\textrm{arg max}_d \\ \\mathbb{E}[U(x) \\mid d].\n\\]\n\nThe outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).\n\n\n\nThis section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.\n\n\nA decision consists of the choice of commute mode and the outcome is a time and cost. More formally,\n\nthe set of decisions is \\(D = 1:4\\), corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and\nthe set of outcomes \\(X = \\mathbb{R} \\times \\mathbb{R}_+\\) contains pairs of numbers \\(x = (c, t)\\) consisting of a cost \\(c\\) and time \\(t \\geq 0\\).\n\n\n\n\nThe density required is \\(p(x \\mid d),\\) where \\(d \\in D\\) is a decision and \\(x = (c, t) \\in X\\) is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters \\(\\theta,\\) \\[\np(x \\mid d, x^{\\textrm{obs}}, d^{\\textrm{obs}})\n=\n\\int\n  p(x \\mid d, \\theta)\n  \\cdot p(\\theta \\mid x^{\\textrm{obs}}, d^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta.\n\\] The observed data for a year of commutes consists of choice of the chosen commute mode \\(d^{\\textrm{obs}}_n\\) and observed costs and times \\(x^{\\textrm{obs}}_n = (c^{\\textrm{obs}}_n, t^{\\textrm{obs}}_n)\\) for \\(n\n\\in 1:200.\\)\nFor simplicity, commute time \\(t_n\\) for trip \\(n\\) will be modeled as lognormal for a given choice of transportation \\(d_n \\in 1:4,\\) \\[\nt_n \\sim \\textrm{lognormal}(\\mu_{d[n]}, \\sigma_{d[n]}).\n\\] To understand the notation, \\(d_n\\), also written \\(d[n]\\), is the mode of transportation used for trip \\(n\\). For example if trip \\(n\\) was by bicycle, then \\(t_n \\sim \\textrm{lognormal}(\\mu_2, \\sigma_2),\\) where \\(\\mu_2\\) and \\(\\sigma_2\\) are the lognormal parameters for bicycling.\nSimple fixed priors are used for each mode of transportation \\(k \\in 1:4,\\) \\[\\begin{eqnarray*}\n\\mu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\sigma_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.\nCost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, \\[\nc_n \\sim \\textrm{lognormal}(\\nu_{d[n]}, \\tau_{d[n]}).\n\\] Again, the priors are fixed for the modes of transportation, \\[\\begin{eqnarray*}\n\\nu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\tau_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.\nThe full set of parameters that are marginalized in the posterior predictive distribution is \\[\n\\theta = (\\mu_{1:4}, \\sigma_{1:4}, \\nu_{1:4}, \\tau_{1:4}).\n\\]\n\n\n\nFor the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as\n\\[\nU(c, t) = -(c + 25 \\cdot t)\n\\]\nThe sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.\n\n\n\nAt this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.\nfunctions {\n  real U(real c, real t) {\n    return -(c + 25 * t);\n  }\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; d;\n  array[N] real c;\n  array[N] real&lt;lower=0&gt; t;\n}\nparameters {\n  vector[4] mu;\n  vector&lt;lower=0&gt;[4] sigma;\n  array[4] real nu;\n  array[4] real&lt;lower=0&gt; tau;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.25);\n  nu ~ normal(0, 20);\n  tau ~ lognormal(0, 0.25);\n  t ~ lognormal(mu[d], sigma[d]);\n  c ~ lognormal(nu[d], tau[d]);\n}\ngenerated quantities {\n  array[4] real util;\n  for (k in 1:4) {\n    util[k] = U(lognormal_rng(nu[k], tau[k]),\n                lognormal_rng(mu[k], sigma[k]));\n  }\n}\nThe generated quantities block defines an array variable util where util[k], which will hold the utility derived from a random commute for choice k generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.\nFor simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, nu and tau would be declared as univariate, and the RNG for cost would only be employed when k == 4.\nDefining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.\nAll that is left is to run Stan. The posterior mean for util[k] is the expected utility, which written out with full conditioning, is \\[\\begin{eqnarray*}\n\\mathbb{E}\\!\\left[U(x) \\mid d = k, d^{\\textrm{obs}}, x^{\\textrm{obs}}\\right]\n& = &\n\\int\n  U(x)\n  \\cdot p(x \\mid d = k, \\theta)\n  \\cdot p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M  U(x^{(m)} ),\n\\end{eqnarray*}\\] where \\[\nx^{(m)} \\sim p(x \\mid d = k, \\theta^{(m)} )\n\\] and \\[\n\\theta^{(m)}\n\\sim p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}}).\n\\]\nIn terms of Stan’s execution, the random generation of \\(x^{(m)}\\) is carried out with the lognormal_rng operations after \\(\\theta^{(m)}\\) is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.\nIt only remains to make the decision k with highest expected utility, which will correspond to the choice with the highest posterior mean for util[k]. This can be read off of the mean column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.\n\n\n\n\nMany choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Directions, Rotations, and Hyperspheres"
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/hyperspherical-models.html#transforming-to-unconstrained-parameters",
-    "href": "stan-users-guide/hyperspherical-models.html#transforming-to-unconstrained-parameters",
-    "title": "Directions, Rotations, and Hyperspheres",
+    "objectID": "stan-users-guide/decision-analysis.html#outline-of-decision-analysis",
+    "href": "stan-users-guide/decision-analysis.html#outline-of-decision-analysis",
+    "title": "Decision Analysis",
     "section": "",
-    "text": "Stan (inverse) transforms arbitrary points in \\(\\mathbb{R}^{K+1}\\) to points in \\(S^K\\) using the auxiliary variable approach of Muller (1959). A point \\(y \\in \\mathbb{R}^K\\) is transformed to a point \\(x \\in S^{K-1}\\) by \\[\nx = \\frac{y}{\\sqrt{y^{\\top} y}}.\n\\]\nThe problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. Muller (1959) introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.",
+    "text": "Following Gelman et al. (2013), Bayesian decision analysis can be factored into the following four steps.\n\nDefine a set \\(X\\) of possible outcomes and a set \\(D\\) of possible decisions.\nDefine a probability distribution of outcomes conditional on decisions through a conditional density function \\(p(x \\mid d)\\) for \\(x \\in X\\) and \\(d \\in D.\\)\nDefine a utility function \\(U : X \\rightarrow \\mathbb{R}\\) mapping outcomes to their utility.\nChoose action \\(d^* \\in D\\) with highest expected utility, \\[\nd^* = \\textrm{arg max}_d \\ \\mathbb{E}[U(x) \\mid d].\n\\]\n\nThe outcomes should represent as much information as possible that is relevant to utility. In Bayesian decision analysis, the distribution of outcomes will typically be a posterior predictive distribution conditioned on observed data. There is a large literature in psychology and economics related to defining utility functions. For example, the utility of money is usually assumed to be strictly concave rather than linear (i.e., the marginal utility of getting another unit of money decreases the more money one has).",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Directions, Rotations, and Hyperspheres"
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/hyperspherical-models.html#unit-vectors-and-rotations",
-    "href": "stan-users-guide/hyperspherical-models.html#unit-vectors-and-rotations",
-    "title": "Directions, Rotations, and Hyperspheres",
+    "objectID": "stan-users-guide/decision-analysis.html#example-decision-analysis",
+    "href": "stan-users-guide/decision-analysis.html#example-decision-analysis",
+    "title": "Decision Analysis",
     "section": "",
-    "text": "Unit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle \\(\\theta\\). Given an angle \\(\\theta\\), a matrix can be defined, the pre-multiplication by which rotates a point by an angle of \\(\\theta\\). For angle \\(\\theta\\) (in two dimensions), the \\(2 \\times 2\\) rotation matrix is defined by \\[\nR_{\\theta}\n=\n\\begin{bmatrix}\n\\cos \\theta & -\\sin \\theta \\\\\n\\sin \\theta &  \\cos \\theta\n\\end{bmatrix}.\n\\] Given a two-dimensional vector \\(x\\), \\(R_{\\theta} \\, x\\) is the rotation of \\(x\\) (around the origin) by \\(\\theta\\) degrees.\n\n\nAngles can be calculated from unit vectors. For example, a random variable theta representing an angle in \\((-\\pi, \\pi)\\) radians can be declared as a two-dimensional unit vector then transformed to an angle.\nparameters {\n  unit_vector[2] xy;\n}\ntransformed parameters {\n  real&lt;lower=-pi(), upper=pi()&gt; theta = atan2(xy[2], xy[1]);\n}\nIf the distribution of \\((x, y)\\) is uniform over a circle, then the distribution of \\(\\arctan \\frac{y}{x}\\) is uniform over \\((-\\pi, \\pi)\\).\nIt might be tempting to try to just declare theta directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values \\(-\\pi\\) and \\(\\pi\\) are at \\(-\\infty\\) and \\(\\infty\\) on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.\nWith a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.",
+    "text": "This section outlines a very simple decision analysis for a commuter deciding among modes of transportation to get to work: walk, bike share, public transportation, or cab. Suppose the commuter has been taking various modes of transportation for the previous year and the transportation conditions and costs have not changed during that time. Over the year, such a commuter might accumulate two hundred observations of the time it takes to get to work given a choice of commute mode.\n\n\nA decision consists of the choice of commute mode and the outcome is a time and cost. More formally,\n\nthe set of decisions is \\(D = 1:4\\), corresponding to the commute types walking, bicycling, public transportation, and cab, respectively, and\nthe set of outcomes \\(X = \\mathbb{R} \\times \\mathbb{R}_+\\) contains pairs of numbers \\(x = (c, t)\\) consisting of a cost \\(c\\) and time \\(t \\geq 0\\).\n\n\n\n\nThe density required is \\(p(x \\mid d),\\) where \\(d \\in D\\) is a decision and \\(x = (c, t) \\in X\\) is an outcome. Being a statistical decision problem, this density will the a posterior predictive distribution conditioned on previously observed outcome and decision pairs, based on a parameter model with parameters \\(\\theta,\\) \\[\np(x \\mid d, x^{\\textrm{obs}}, d^{\\textrm{obs}})\n=\n\\int\n  p(x \\mid d, \\theta)\n  \\cdot p(\\theta \\mid x^{\\textrm{obs}}, d^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta.\n\\] The observed data for a year of commutes consists of choice of the chosen commute mode \\(d^{\\textrm{obs}}_n\\) and observed costs and times \\(x^{\\textrm{obs}}_n = (c^{\\textrm{obs}}_n, t^{\\textrm{obs}}_n)\\) for \\(n\n\\in 1:200.\\)\nFor simplicity, commute time \\(t_n\\) for trip \\(n\\) will be modeled as lognormal for a given choice of transportation \\(d_n \\in 1:4,\\) \\[\nt_n \\sim \\textrm{lognormal}(\\mu_{d[n]}, \\sigma_{d[n]}).\n\\] To understand the notation, \\(d_n\\), also written \\(d[n]\\), is the mode of transportation used for trip \\(n\\). For example if trip \\(n\\) was by bicycle, then \\(t_n \\sim \\textrm{lognormal}(\\mu_2, \\sigma_2),\\) where \\(\\mu_2\\) and \\(\\sigma_2\\) are the lognormal parameters for bicycling.\nSimple fixed priors are used for each mode of transportation \\(k \\in 1:4,\\) \\[\\begin{eqnarray*}\n\\mu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\sigma_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] These priors are consistent with a broad range of commute times; in a more realistic model each commute mode would have its own prior based on knowledge of the city and the time of day would be used as a covariate; here the commutes are taken to be exchangeable.\nCost is usually a constant function for public transportation, walking, and bicycling. Nevertheless, for simplicity, all costs will be modeled as lognormal, \\[\nc_n \\sim \\textrm{lognormal}(\\nu_{d[n]}, \\tau_{d[n]}).\n\\] Again, the priors are fixed for the modes of transportation, \\[\\begin{eqnarray*}\n\\nu_k & \\sim & \\textrm{normal}(0, 5)\n\\\\[2pt]\n\\tau_k & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\] A more realistic approach would model cost conditional on time, because the cost of a cab depends on route chosen and the time it takes.\nThe full set of parameters that are marginalized in the posterior predictive distribution is \\[\n\\theta = (\\mu_{1:4}, \\sigma_{1:4}, \\nu_{1:4}, \\tau_{1:4}).\n\\]\n\n\n\nFor the sake of concreteness, the utility function will be assumed to be a simple function of cost and time. Further suppose the commuter values their commute time at $25 per hour and has a utility function that is linear in the commute cost and time. Then the utility function may be defined as\n\\[\nU(c, t) = -(c + 25 \\cdot t)\n\\]\nThe sign is negative because high cost is undesirable. A better utility function might have a step function or increasing costs for being late, different costs for different modes of transportation because of their comfort and environmental impact, and non-linearity of utility in cost.\n\n\n\nAt this point, all that is left is to calculate expected utility for each decision and choose the optimum. If the decisions consist of a small set of discrete choices, expected utility can be easily coded in Stan. The utility function is coded as a function, the observed data is coded as data, the model parameters coded as parameters, and the model block itself coded to follow the sampling distributions of each parameter.\nfunctions {\n  real U(real c, real t) {\n    return -(c + 25 * t);\n  }\n}\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; d;\n  array[N] real c;\n  array[N] real&lt;lower=0&gt; t;\n}\nparameters {\n  vector[4] mu;\n  vector&lt;lower=0&gt;[4] sigma;\n  array[4] real nu;\n  array[4] real&lt;lower=0&gt; tau;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.25);\n  nu ~ normal(0, 20);\n  tau ~ lognormal(0, 0.25);\n  t ~ lognormal(mu[d], sigma[d]);\n  c ~ lognormal(nu[d], tau[d]);\n}\ngenerated quantities {\n  array[4] real util;\n  for (k in 1:4) {\n    util[k] = U(lognormal_rng(nu[k], tau[k]),\n                lognormal_rng(mu[k], sigma[k]));\n  }\n}\nThe generated quantities block defines an array variable util where util[k], which will hold the utility derived from a random commute for choice k generated according to the model parameters for that choice. This randomness is required to appropriately characterize the posterior predictive distribution of utility.\nFor simplicity in this initial formulation, all four commute options have their costs estimated, even though cost is fixed for three of the options. To deal with the fact that some costs are fixed, the costs would have to be hardcoded or read in as data, nu and tau would be declared as univariate, and the RNG for cost would only be employed when k == 4.\nDefining the utility function for pairs of vectors would allow the random number generation in the generated quantities block to be vectorized.\nAll that is left is to run Stan. The posterior mean for util[k] is the expected utility, which written out with full conditioning, is \\[\\begin{eqnarray*}\n\\mathbb{E}\\!\\left[U(x) \\mid d = k, d^{\\textrm{obs}}, x^{\\textrm{obs}}\\right]\n& = &\n\\int\n  U(x)\n  \\cdot p(x \\mid d = k, \\theta)\n  \\cdot p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}})\n  \\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M  U(x^{(m)} ),\n\\end{eqnarray*}\\] where \\[\nx^{(m)} \\sim p(x \\mid d = k, \\theta^{(m)} )\n\\] and \\[\n\\theta^{(m)}\n\\sim p(\\theta \\mid d^{\\textrm{obs}}, x^{\\textrm{obs}}).\n\\]\nIn terms of Stan’s execution, the random generation of \\(x^{(m)}\\) is carried out with the lognormal_rng operations after \\(\\theta^{(m)}\\) is drawn from the model posterior. The average is then calculated after multiple chains are run and combined.\nIt only remains to make the decision k with highest expected utility, which will correspond to the choice with the highest posterior mean for util[k]. This can be read off of the mean column of the Stan’s summary statistics or accessed programmatically through Stan’s interfaces.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Directions, Rotations, and Hyperspheres"
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/hyperspherical-models.html#circular-representations-of-days-and-years",
-    "href": "stan-users-guide/hyperspherical-models.html#circular-representations-of-days-and-years",
-    "title": "Directions, Rotations, and Hyperspheres",
+    "objectID": "stan-users-guide/decision-analysis.html#continuous-choices",
+    "href": "stan-users-guide/decision-analysis.html#continuous-choices",
+    "title": "Decision Analysis",
     "section": "",
-    "text": "A 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.\nIn human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.",
+    "text": "Many choices, such as how much to invest for retirement or how long to spend at the gym are not discrete, but continuous. In these cases, the continuous choice can be coded as data in the Stan program. Then the expected utilities may be calculated. In other words, Stan can be used as a function from a choice to expected utilities. Then an external optimizer can call that function. This optimization can be difficult without gradient information. Gradients could be supplied by automatic differentiation, but Stan is not currently instrumented to calculate those derivatives.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Directions, Rotations, and Hyperspheres"
+      "Posterior Inference & Model Checking",
+      "Decision Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html",
-    "href": "stan-users-guide/latent-discrete.html",
-    "title": "Latent Discrete Parameters",
+    "objectID": "stan-users-guide/finite-mixtures.html",
+    "href": "stan-users-guide/finite-mixtures.html",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "Stan does not support sampling discrete parameters. So it is not possible to directly translate BUGS or JAGS models with discrete parameters (i.e., discrete stochastic nodes). Nevertheless, it is possible to code many models that involve bounded discrete parameters by marginalizing out the discrete parameters.1\nThis chapter shows how to code several widely-used models involving latent discrete parameters. The next chapter, the clustering chapter, on clustering models, considers further models involving latent discrete parameters.\n\n\nAlthough it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.\nStandard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.\n\n\n\nThe first example is a model of coal mining disasters in the U.K. for the years 1851–1962.2\n\n\nFonnesbeck et al. (2013, sec. 3.1) provides a Poisson model of disaster \\(D_t\\) in year \\(t\\) with two rate parameters, an early rate (\\(e\\)) and late rate (\\(l\\)), that change at a given point in time \\(s\\). The full model expressed using a latent discrete parameter \\(s\\) is \\[\\begin{align*}\ne   &\\sim  \\textsf{exponential}(r_e) \\\\\nl   &\\sim  \\textsf{exponential}(r_l) \\\\\ns   &\\sim  \\textsf{uniform}(1, T) \\\\\nD_t &\\sim  \\textsf{Poisson}(t &lt; s \\; ? \\; e \\: : \\: l)\n\\end{align*}\\]\nThe last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.3\nIt uses a compact notation involving separating its three arguments by a question mark (?) and a colon (:). The conditional operator is defined by \\[\nc \\; ? \\; x_1 \\: : \\: x_2\n=\n\\begin{cases}\n\\ x_1 & \\quad\\text{if } c \\text{ is true (i.e., non-zero), and} \\\\\n\\ x_2 & \\quad\\text{if } c \\text{ is false (i.e., zero).}\n\\end{cases}\n\\]\n\n\n\nTo code this model in Stan, the discrete parameter \\(s\\) must be marginalized out to produce a model defining the log of the probability function \\(p(e,l,D_t)\\). The full joint probability factors as \\[\\begin{align*}\np(e,l,s,D) &=  p(e) \\, p(l) \\, p(s) \\, p(D \\mid s, e, l) \\\\\n&= \\textsf{exponential}(e \\mid r_e) \\ \\textsf{exponential}(l \\mid r_l) \\,\n    \\textsf{uniform}(s \\mid 1, T) \\\\\n& \\qquad \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nTo marginalize, an alternative factorization into prior and likelihood is used, \\[\np(e,l,D) = p(e,l) \\, p(D \\mid e,l),\n\\]\nwhere the likelihood is defined by marginalizing \\(s\\) as \\[\\begin{align*}\np(D \\mid e,l) &= \\sum_{s=1}^T p(s, D \\mid e,l) \\\\\n&= \\sum_{s=1}^T p(s) \\, p(D \\mid s,e,l) \\\\\n&= \\sum_{s=1}^T \\textsf{uniform}(s \\mid 1,T) \\,\n    \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nStan operates on the log scale and thus requires the log likelihood, \\[\\begin{align*}\n\\log p(D \\mid e,l)\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{s=1}^T\n    \\left( \\log \\textsf{uniform}(s \\mid 1, T) \\vphantom{\\sum_{t=1}^T}\\right. \\\\\n&\\qquad \\left.\n    + \\sum_{t=1}^T \\log \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l)\n\\right),\n\\end{align*}\\] where the log sum of exponents function is defined by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{n=1}^N \\, \\alpha_n =\n\\log \\sum_{n=1}^N \\exp(\\alpha_n).\n\\]\nThe log sum of exponents function allows the model to be coded directly in Stan using the built-in function log_sum_exp, which provides both arithmetic stability and efficiency for mixture model calculations.\n\n\n\nThe Stan program for the change point model is shown in the figure below. The transformed parameter lp[s] stores the quantity \\(\\log p(s, D \\mid e, l)\\).\ndata {\n  real&lt;lower=0&gt; r_e;\n  real&lt;lower=0&gt; r_l;\n\n  int&lt;lower=1&gt; T;\n  array[T] int&lt;lower=0&gt; D;\n}\ntransformed data {\n  real log_unif;\n  log_unif = -log(T);\n}\nparameters {\n  real&lt;lower=0&gt; e;\n  real&lt;lower=0&gt; l;\n}\ntransformed parameters {\n  vector[T] lp;\n  lp = rep_vector(log_unif, T);\n  for (s in 1:T) {\n    for (t in 1:T) {\n      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);\n    }\n  }\n}\nmodel {\n  e ~ exponential(r_e);\n  l ~ exponential(r_l);\n  target += log_sum_exp(lp);\n}\nA change point model in which disaster rates D[t] have one rate, e, before the change point and a different rate, l, after the change point. The change point itself, s, is marginalized out as described in the text.\nAlthough the change-point model is coded directly, the doubly nested loop used for s and t is quadratic in T. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.\ntransformed parameters {\n    vector[T] lp;\n    {\n      vector[T + 1] lp_e;\n      vector[T + 1] lp_l;\n      lp_e[1] = 0;\n      lp_l[1] = 0;\n      for (t in 1:T) {\n        lp_e[t + 1] = lp_e[t] + poisson_lpmf(D[t] | e);\n        lp_l[t + 1] = lp_l[t] + poisson_lpmf(D[t] | l);\n      }\n      lp = rep_vector(log_unif + lp_l[T + 1], T)\n           + head(lp_e, T) - head(lp_l, T);\n    }\n  }\nAs should be obvious from looking at it, it has linear complexity in T rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger T.\nThe key to understanding Wiklendt’s dynamic programming version is to see that head(lp_e) holds the forward values, whereas lp_l[T + 1] - head(lp_l, T) holds the backward values; the clever use of subtraction allows lp_l to be accumulated naturally in the forward direction.\n\n\n\nThis model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.\nThe value of lp for each iteration for each change point is available because it is declared as a transformed parameter. If the value of lp were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.\n\n\n\nThe value of lp[s] in a given iteration is given by \\(\\log\np(s,D \\mid e,l)\\) for the values of the early and late rates, \\(e\\) and \\(l\\), in the iteration. In each iteration after convergence, the early and late disaster rates, \\(e\\) and \\(l\\), are drawn from the posterior \\(p(e,l \\mid D)\\) by MCMC sampling and the associated lp calculated. The value of lp may be normalized to calculate \\(p(s \\mid e,l,D)\\) in each iteration, based on on the current values of \\(e\\) and \\(l\\). Averaging over iterations provides an unnormalized probability estimate of the change point being \\(s\\) (see below for the normalizing constant), \\[\\begin{align*}\np(s \\mid D) &\\propto q(s \\mid D) \\\\\n&= \\frac{1}{M} \\sum_{m=1}^{M} \\exp(\\texttt{lp}[m,s]).\n\\end{align*}\\] where \\(\\texttt{lp}[m,s]\\) represents the value of lp in posterior draw \\(m\\) for change point \\(s\\). By averaging over draws, \\(e\\) and \\(l\\) are themselves marginalized out, and the result has no dependence on a given iteration’s value for \\(e\\) and \\(l\\). A final normalization then produces the quantity of interest, the posterior probability of the change point being \\(s\\) conditioned on the data \\(D\\), \\[\np(s \\mid D) = \\frac{q(s \\mid D)}{\\sum_{s'=1}^T q(s' \\mid D)}.\n\\]\nA plot of the values of \\(\\log p(s \\mid D)\\) computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.\nLog probability of change point being in year, calculated analytically.\n\n\n\nAnalytical change-point posterior\n\n\nThe frequency of change points generated by sampling the discrete change points.\n\n\n\nSampled change-point posterior\n\n\nIn order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of \\(s\\) is roughly 1891.\n\n\n\nThe generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with lp defined as above, the following program draws a random value for s at every iteration.\ngenerated quantities {\n  int&lt;lower=1, upper=T&gt; s;\n  s = categorical_logit_rng(lp);\n}\nA posterior histogram of draws for \\(s\\) is shown on the second change point posterior figure above.\nCompared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for s given by softmax(lp).\n\n\n\nThe discrete sample generated for \\(s\\) can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using lp.\n\n\n\nThere is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, e, m, and l, and two loops for the change point and then one over time, with log densities being stored in a matrix.\nmatrix[T, T] lp;\nlp = rep_matrix(log_unif, T);\nfor (s1 in 1:T) {\n  for (s2 in 1:T) {\n    for (t in 1:T) {\n      lp[s1,s2] = lp[s1,s2]\n        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));\n    }\n  }\n}\nThe matrix can then be converted back to a vector using to_vector before being passed to log_sum_exp.\n\n\n\n\nA widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.\nThe first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.\n\n\nIn the simplest case, a one-stage mark-recapture study produces the following data\n\n\\(M\\) : number of animals marked in first capture,\n\\(C\\) : number animals in second capture, and\n\\(R\\) : number of marked animals in second capture.\n\nThe estimand of interest is\n\n\\(N\\) : number of animals in the population.\n\nDespite the notation, the model will take \\(N\\) to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.\nThe Lincoln-Petersen (Lincoln 1930; Petersen 1896) method for estimating population size is \\[\n\\hat{N} = \\frac{M C}{R}.\n\\]\nThis population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, \\[\nR \\sim \\textsf{binomial}(C, M / N)\n\\] given the total number of animals captured in the second round (\\(C\\)) with a recapture probability of \\(M/N\\), the fraction of the total population \\(N\\) marked in the first round.\ndata {\n  int&lt;lower=0&gt; M;\n  int&lt;lower=0&gt; C;\n  int&lt;lower=0, upper=min(M, C)&gt; R;\n}\nparameters {\n  real&lt;lower=(C - R + M)&gt; N;\n}\nmodel {\n  R ~ binomial(C, M / N);\n}\nA probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on \\(N\\) is necessary to efficiently eliminate impossible values.\nThe probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.\nTo ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for \\(N\\) is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.\nThe one tricky part of the model is the lower bound \\(C - R + M\\) placed on the population size \\(N\\). Values below this bound are impossible because it is otherwise not possible to draw \\(R\\) samples out of the \\(C\\) animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for \\(C\\) implies a variable transform \\(f : (C-R+M,\\infty) \\rightarrow (-\\infty,+\\infty)\\) defined by \\(f(N) = \\log(N - (C - R + M))\\); the reference manual contains full details of all constrained parameter transforms.\n\n\n\nThe Cormack-Jolly-Seber (CJS) model (Cormack 1964; Jolly 1965; Seber 1965) is an open-population model in which the population may change over time due to death; the presentation here draws heavily on Schofield (2007).\nThe basic data are\n\n\\(I\\): number of individuals,\n\\(T\\): number of capture periods, and\n\\(y_{i,t}\\): Boolean indicating if individual \\(i\\) was captured at time \\(t\\).\n\nEach individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.\nThere are two Bernoulli parameters in the model,\n\n\\(\\phi_t\\) : probability that animal alive at time \\(t\\) survives until \\(t + 1\\) and\n\\(p_t\\) : probability that animal alive at time \\(t\\) is captured at time \\(t\\).\n\nThese parameters will both be given uniform priors, but information should be used to tighten these priors in practice.\nThe CJS model also employs a latent discrete parameter \\(z_{i,t}\\) indicating for each individual \\(i\\) whether it is alive at time \\(t\\), distributed as \\[\nz_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t-1} \\; ? \\; 0 \\: : \\: \\phi_{t-1}).\n\\]\nThe conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on \\(z\\) as \\[\ny_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t} \\; ? \\; 0 \\: : \\: p_t).\n\\]\nThe conditional enforces the constraint that dead animals cannot be captured.\n\n\n\nThis subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.\nIn order to ease the marginalization of the latent discrete parameter \\(z_{i,t}\\), the Stan models rely on a derived quantity \\(\\chi_t\\) for the probability that an individual is never captured again if it is alive at time \\(t\\) (if it is dead, the recapture probability is zero). this quantity is defined recursively by \\[\n\\chi_t\n=\n\\begin{cases}\n1 & \\quad\\text{if } t = T \\\\\n(1 - \\phi_t) + \\phi_t (1 - p_{t+1}) \\chi_{t+1}\n  & \\quad\\text{if } t &lt; T\n\\end{cases}\n\\]\nThe base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining \\(\\chi_{t}\\) in terms of \\(\\chi_{t+1}\\) involves two possibilities: (1) not surviving to the next time period, with probability \\((1 - \\phi_t)\\), or (2) surviving to the next time period with probability \\(\\phi_t\\), not being captured in the next time period with probability \\((1 - p_{t+1})\\), and not being captured again after being alive in period \\(t+1\\) with probability \\(\\chi_{t+1}\\).\nWith three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.\n\\[\n\\begin{array}{crclc}\n\\hline\n& \\qquad\\qquad & captures & \\qquad\\qquad &  \\\\\n\\mathrm{profile} & 1 & 2 & 3 & \\mathrm{probability} \\\\\n\\hline\n0 & - & - & - & n/a \\\\\n1 & - & - & + & n/a \\\\\n2 & - & + & - & \\chi_2 \\\\\n3 & - & + & + & \\phi_2 \\, p_3 \\\\\n4 & + & - & - & \\chi_1 \\\\\n5 & + & - & + & \\phi_1 \\, (1 - p_2) \\, \\phi_2 \\, p_3 \\\\\n6 & + & + & - & \\phi_1 \\, p_2 \\, \\chi_2 \\\\\n7 & + & + & + & \\phi_1 \\, p_2 \\, \\phi_2 \\, p_3 \\\\\n\\hline\n\\end{array}\n\\]\nHistory 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.\nBy defining these probabilities in terms of \\(\\chi\\) directly, there is no need for a latent binary parameter indicating whether an animal is alive at time \\(t\\) or not. The definition of \\(\\chi\\) is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model (Schofield 2007).\nThe Stan model defines \\(\\chi\\) as a transformed parameter based on parameters \\(\\phi\\) and \\(p\\). In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using target += rather than being part of a built-in probability function.\nThe following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods\ndata {\n  array[7] int&lt;lower=0&gt; history;\n}\nparameters {\n  array[2] real&lt;lower=0, upper=1&gt; phi;\n  array[3] real&lt;lower=0, upper=1&gt; p;\n}\ntransformed parameters {\n  array[2] real&lt;lower=0, upper=1&gt; chi;\n  chi[2] = (1 - phi[2]) + phi[2] * (1 - p[3]);\n  chi[1] = (1 - phi[1]) + phi[1] * (1 - p[2]) * chi[2];\n}\nmodel {\n  target += history[2] * log(chi[2]);\n  target += history[3] * (log(phi[2]) + log(p[3]));\n  target += history[4] * (log(chi[1]));\n  target += history[5] * (log(phi[1]) + log1m(p[2])\n                            + log(phi[2]) + log(p[3]));\n  target += history[6] * (log(phi[1]) + log(p[2])\n                            + log(chi[2]));\n  target += history[7] * (log(phi[1]) + log(p[2])\n                            + log(phi[2]) + log(p[3]));\n}\ngenerated quantities {\n  real&lt;lower=0, upper=1&gt; beta3;\n  beta3 = phi[2] * p[3];\n}\n\n\nThe parameters \\(\\phi_2\\) and \\(p_3\\), the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, \\(\\beta_3 = \\phi_2 \\, p_3\\), is identified. The Stan model defines beta3 as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the \\(p\\) and \\(\\phi\\) parameters.\n\n\n\n\nThis section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number \\(T\\) of capture events, the number \\(I\\) of individuals, and a boolean flag \\(y_{i,t}\\) indicating if individual \\(i\\) was observed at time \\(t\\). In Stan,\ndata {\n  int&lt;lower=2&gt; T;\n  int&lt;lower=0&gt; I;\n  array[I, T] int&lt;lower=0, upper=1&gt; y;\n}\nThe advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding \\(2^T\\) observation histories for \\(T\\) capture times.\n\n\nThe individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.4\nfunctions {\n  int first_capture(array[] int y_i) {\n    for (k in 1:size(y_i)) {\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  int last_capture(array[] int y_i) {\n    for (k_rev in 0:(size(y_i) - 1)) {\n      int k;\n      k = size(y_i) - k_rev;\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  // ...\n}\nThese two functions are used to define the first and last capture time for each individual in the transformed data block.5\ntransformed data {\n  array[I] int&lt;lower=0, upper=T&gt; first;\n  array[I] int&lt;lower=0, upper=T&gt; last;\n  vector&lt;lower=0, upper=I&gt;[T] n_captured;\n  for (i in 1:I) {\n    first[i] = first_capture(y[i]);\n  }\n  for (i in 1:I) {\n    last[i] = last_capture(y[i]);\n  }\n  n_captured = rep_vector(0, T);\n  for (t in 1:T) {\n    for (i in 1:I) {\n      if (y[i, t]) {\n        n_captured[t] = n_captured[t] + 1;\n      }\n    }\n  }\n}\nThe transformed data block also defines n_captured[t], which is the total number of captures at time t. The variable n_captured is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.\nThe parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector chi, the probability that if an individual is alive at t that it will never be captured again.\nparameters {\n  vector&lt;lower=0, upper=1&gt;[T - 1] phi;\n  vector&lt;lower=0, upper=1&gt;[T] p;\n}\ntransformed parameters {\n  vector&lt;lower=0, upper=1&gt;[T] chi;\n  chi = prob_uncaptured(T, p, phi);\n}\nThe definition of prob_uncaptured, from the functions block, is\nfunctions {\n  // ...\n  vector prob_uncaptured(int T, vector p, vector phi) {\n    vector[T] chi;\n    chi[T] = 1.0;\n    for (t in 1:(T - 1)) {\n      int t_curr;\n      int t_next;\n      t_curr = T - t;\n      t_next = t_curr + 1;\n      chi[t_curr] = (1 - phi[t_curr])\n                     + phi[t_curr]\n                       * (1 - p[t_next])\n                       * chi[t_next];\n    }\n    return chi;\n  }\n}\nThe function definition directly follows the mathematical definition of \\(\\chi_t\\), unrolling the recursion into an iteration and defining the elements of chi from T down to 1.\n\n\n\nGiven the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations q given the parameters p and phi as well as the transformed parameter chi defined in terms of p and phi.\nmodel {\n  for (i in 1:I) {\n    if (first[i] &gt; 0) {\n      for (t in (first[i]+1):last[i]) {\n        1 ~ bernoulli(phi[t - 1]);\n        y[i, t] ~ bernoulli(p[t]);\n      }\n      1 ~ bernoulli(chi[last[i]]);\n    }\n  }\n}\nThe outer loop is over individuals, conditional skipping individuals i which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for first if an individual is never captured.\nThe inner loop for individual i first increments the log probability based on the survival of the individual with probability phi[t - 1]. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.\nIn the inner loop, the observed capture status y[i, t] for individual i at time t has a Bernoulli distribution based on the capture probability p[t] at time t.\nAfter the inner loop, the probability of an animal never being seen again after being observed at time last[i] is included, because last[i] was defined to be the last time period in which animal i was observed.\n\n\n\nAs with the collective model described in the previous subsection, this model does not identify phi[T - 1] and p[T], but does identify their product, beta. Thus beta is defined as a generated quantity to monitor convergence and report.\ngenerated quantities {\n  real beta;\n  // ...\n\n  beta = phi[T - 1] * p[T];\n  // ...\n}\nThe parameter p[1] is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case p[1] would be an unknown initial condition and both phi[T - 1] and p[T] could be identified.\n\n\n\nThe generated quantities also calculates an estimate of the population mean at each time t in the same way as in the simple mark-recapture model as the number of individuals captured at time t divided by the probability of capture at time t. This is done with the elementwise division operation for vectors (./) in the generated quantities block.\ngenerated quantities {\n  // ...\n  vector&lt;lower=0&gt;[T] pop;\n  // ...\n  pop = n_captured ./ p;\n  pop[1] = -1;\n}\n\n\n\nAll individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.\n\n\n\n\n\nAlthough seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.\n\n\nSuppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.\nThere are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.\n\n\n\nHumans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.\n\n\n\nIn this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.\nDawid and Skene (1979) introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.\n\n\nThe data for the model consists of \\(J\\) raters (diagnostic tests), \\(I\\) items (patients), and \\(K\\) categories (condition statuses) to annotate, with \\(y_{i, j} \\in \\{1, \\dotsc, K\\}\\) being the rating provided by rater \\(j\\) for item \\(i\\). In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often \\(K=2\\), with values signaling the presence or absence of the condition.6\nIt is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.\n\n\n\n\nThe model is based on three parameters, the first of which is discrete:\n\n\\(z_i\\) : a value in \\(\\{1, \\dotsc, K\\}\\) indicating the true category of item \\(i\\),\n\\(\\pi\\) : a \\(K\\)-simplex for the prevalence of the \\(K\\) categories in the population, and\n\\(\\theta_{j,k}\\) : a \\(K\\)-simplex for the response of annotator \\(j\\) to an item of true category \\(k\\).\n\n\n\n\nThe true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, \\[\nz_i \\sim \\textsf{categorical}(\\pi).\n\\]\nThe rating \\(y_{i, j}\\) provided for item \\(i\\) by rater \\(j\\) is modeled as a categorical response of rater \\(i\\) to an item of category \\(z_i\\),7 \\[\ny_{i, j} \\sim \\textsf{categorical}(\\theta_{j,\\pi_{z[i]}}).\n\\]\n\n\nDawid and Skene provided maximum likelihood estimates for \\(\\theta\\) and \\(\\pi\\), which allows them to generate probability estimates for each \\(z_i\\).\nTo mimic Dawid and Skene’s maximum likelihood model, the parameters \\(\\theta_{j,k}\\) and \\(\\pi\\) can be given uniform priors over \\(K\\)-simplexes. It is straightforward to generalize to Dirichlet priors, \\[\n\\pi \\sim \\textsf{Dirichlet}(\\alpha)\n\\] and \\[\n\\theta_{j,k} \\sim \\textsf{Dirichlet}(\\beta_k)\n\\] with fixed hyperparameters \\(\\alpha\\) (a vector) and \\(\\beta\\) (a matrix or array of vectors). The prior for \\(\\theta_{j,k}\\) must be allowed to vary in \\(k\\), so that, for instance, \\(\\beta_{k,k}\\) is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.\nBecause there are \\(J\\) coders, it would be natural to extend the model to include a hierarchical prior for \\(\\beta\\) and to partially pool the estimates of coder accuracy and bias.\n\n\n\nBecause the true category parameter \\(z\\) is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as \\[\np(y, \\theta, \\pi) = p(y \\mid \\theta,\\pi) \\, p(\\pi) \\, p(\\theta),\n\\] where \\(p(y \\mid \\theta,\\pi)\\) is derived by marginalizing \\(z\\) out of \\[\np(z, y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\left( \\textsf{categorical}(z_i \\mid \\pi)\n                     \\prod_{j=1}^J\n                     \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, z[i]})\n              \\right).\n\\]\nThis can be done item by item, with \\[\np(y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\sum_{k=1}^K\n  \\left( \\textsf{categorical}(k \\mid \\pi)\n         \\prod_{j=1}^J\n         \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n  \\right).\n\\]\nIn the missing data model, only the observed labels would be used in the inner product.\nDawid and Skene (1979) derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, \\[\\begin{align*}\n\\log p(y \\mid \\theta, \\pi)\n&= \\sum_{i=1}^I \\log \\left( \\sum_{k=1}^K \\exp\n    \\left(\\log \\textsf{categorical}(k \\mid \\pi) \\vphantom{\\sum_{j=1}^J}\\right.\\right.\n    \\left.\\left. + \\ \\sum_{j=1}^J\n           \\log \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n    \\right) \\right),\n\\end{align*}\\] which can be directly coded using Stan’s built-in log_sum_exp function.\n\n\n\n\nThe Stan program for the Dawid and Skene model is provided below (Dawid and Skene 1979).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=1&gt; I;\n  int&lt;lower=1&gt; J;\n\n  array[I, J] int&lt;lower=1, upper=K&gt; y;\n\n  vector&lt;lower=0&gt;[K] alpha;\n  vector&lt;lower=0&gt;[K] beta[K];\n}\nparameters {\n  simplex[K] pi;\n  array[J, K] simplex[K] theta;\n}\ntransformed parameters {\n  array[I] vector[K] log_q_z;\n  for (i in 1:I) {\n    log_q_z[i] = log(pi);\n    for (j in 1:J) {\n      for (k in 1:K) {\n        log_q_z[i, k] = log_q_z[i, k]\n                         + log(theta[j, k, y[i, j]]);\n      }\n    }\n  }\n}\nmodel {\n  pi ~ dirichlet(alpha);\n  for (j in 1:J) {\n    for (k in 1:K) {\n      theta[j, k] ~ dirichlet(beta[k]);\n    }\n  }\n\n  for (i in 1:I) {\n    target += log_sum_exp(log_q_z[i]);\n  }\n}\nThe model marginalizes out the discrete parameter \\(z\\), storing the unnormalized conditional probability \\(\\log q(z_i=k|\\theta,\\pi)\\) in log_q_z[i, k].\nThe Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are \\(\\alpha_k = 3\\) and \\(\\beta_{k,k} = 2.5 K\\) and \\(\\beta_{k,k'} = 1\\) if \\(k \\neq k'\\). Taking \\(\\alpha\\) and \\(\\beta_k\\) to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of Dawid and Skene (1979).\n\n\nThe quantity log_q_z[i] is defined as a transformed parameter. It encodes the (unnormalized) log of \\(p(z_i \\mid \\theta,\n\\pi)\\). Each iteration provides a value conditioned on that iteration’s values for \\(\\theta\\) and \\(\\pi\\). Applying the softmax function to log_q_z[i] provides a simplex corresponding to the probability mass function of \\(z_i\\) in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each \\(z_i\\).\n\n\n\n\n\n\n\nThis section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is not necessary to fit models with discrete parameters using Stan.\nThroughout, the model under consideration consists of both continuous parameters, \\(\\Theta\\), and discrete parameters, \\(Z\\). It is also assumed that \\(Z\\) can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.\n\n\n\nWhen performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating \\(\\mu\\), say, where \\(\\mu = \\mathbb{E}[g(\\Theta, Z)]\\) and \\(g(\\cdot)\\) is an arbitrary function. An example of such a quantity is \\(\\mathbb{E}[\\Theta]\\), the posterior mean of the continuous parameters, where we would take \\(g(\\theta, z) = \\theta\\). To estimate \\(\\mu\\) the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.\nIf both \\(\\Theta\\) and \\(Z\\) were continuous, Stan could be used to sample \\(M\\) draws from the joint posterior \\(p_{\\Theta, Z}(\\theta, z)\\) and then estimate \\(\\mu\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M {g(\\theta^{(i)}, z^{(i)})}.\n\\] Given \\(Z\\) is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out \\(Z\\) from the joint posterior to give the marginalized posterior \\(p_\\Theta(\\theta)\\). This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws \\(\\{\\theta^{(i)}\\}_{i = 1}^M\\) from the marginalized posterior.\nUsing only these draws, how can we estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\)? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function \\(g\\) of interest, we compute \\[\nh(\\Theta) = \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n\\] and then estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M h(\\theta^{(i)}).\n\\] This estimator is justified by the law of iterated expectation, the fact that \\[\n\\mathbb{E}[h(\\Theta)] = \\mathbb{E}[\\mathbb{E}[g(\\Theta, Z)] \\mid \\Theta] = \\mathbb{E}[g(\\Theta, Z)] = \\mu.\n\\] Using this marginalized estimator provides a way to estimate the expectation of any function \\(g(\\cdot)\\) for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\).\n\n\n\nFortunately, the discrete nature of \\(Z\\) makes evaluating \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\) easy. The function \\(h(\\Theta)\\) can be written as: \\[\nh(\\Theta)\n= \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n= \\sum_{k} g(\\Theta, k) \\Pr[Z = k \\mid \\Theta],\n\\] where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, \\(\\Pr[Z = k \\mid \\Theta]\\). More detail on how this quantity can be calculated is included below. Note that if \\(Z\\) takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.\nWhen \\(g(\\theta, z)\\) is a function of either \\(\\theta\\) or \\(z\\) only, the above formula simplifies further.\nIn the first case, where \\(g(\\theta, z) = g(\\theta)\\), we have: \\[\\begin{align*}\nh(\\Theta)\n&= \\sum_{k} g(\\Theta) \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta) \\sum_{k} \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta).\n\\end{align*}\\] This means that we can estimate \\(\\mathbb{E}[g(\\Theta)]\\) with the standard, seemingly unconditional, estimator: \\[\n\\frac{1}{M} \\sum_{i = 1}^M g(\\theta^{(i)}).\n\\] Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.\nIn the second case, where \\(g(\\theta, z) = g(z)\\), the conditional expectation instead simplifies as follows: \\[\nh(\\Theta) = \\sum_{k} g(k) \\Pr[Z = k \\mid \\Theta].\n\\] An important special case of this result is when \\(g(\\theta, z) = \\textrm{I}(z = k)\\), where \\(\\textrm{I}\\) is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable \\(Z\\), since \\(\\mathbb{E}[\\textrm{I}(Z = k)] = \\Pr[Z = k]\\). In this case, \\[\nh(\\Theta)\n= \\sum_{k} \\textrm{I}(z = k) \\Pr[Z = k \\mid \\Theta]\n= \\Pr[Z = k \\mid \\Theta].\n\\] The quantity \\(\\Pr[Z = k]\\) can therefore be estimated with: \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}].\n\\] When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, \\(Y\\). That is, we are really estimating \\(\\Pr[Z = k \\mid Y]\\) with \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y].\n\\] This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us \\[\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y]\n= \\frac{\\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}\n{\\sum_{k = 1}^K \\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}.\n\\] Here, \\(\\Pr[Y \\mid \\Theta = \\theta^{(i)}, Z = k]\\) is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.\nOther than the use of Bayes’s theorem, \\(\\Pr[Z = k \\mid \\theta = \\theta^{(i)}, Y]\\) can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the Dawid–Skene model).\nFor a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see Pullin, Gurrin, and Vukcevic (2021).",
+    "text": "Finite mixture models of an outcome assume that the outcome is drawn from one of several distributions, the identity of which is controlled by a categorical mixing distribution. Mixture models typically have multimodal densities with modes near the modes of the mixture components. Mixture models may be parameterized in several ways, as described in the following sections. Mixture models may be used directly for modeling data with multimodal distributions, or they may be used as priors for other parameters.\n\n\nClustering models, as discussed in the clustering chapter, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the \\(K\\)-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.\n\n\n\nOne way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider \\(K\\) normal distributions with locations \\(\\mu_k \\in \\mathbb{R}\\) and scales \\(\\sigma_k \\in (0,\\infty)\\). Now consider mixing them in proportion \\(\\lambda\\), where \\(\\lambda_k \\geq 0\\) and \\(\\sum_{k=1}^K \\lambda_k = 1\\) (i.e., \\(\\lambda\\) lies in the unit \\(K\\)-simplex). For each outcome \\(y_n\\) there is a latent variable \\(z_n\\) in \\(\\{ 1,\\dotsc,K \\}\\) with a categorical distribution parameterized by \\(\\lambda\\), \\[\nz_n \\sim \\textsf{categorical}(\\lambda).\n\\]\nThe variable \\(y_n\\) is distributed according to the parameters of the mixture component \\(z_n\\), \\[\ny_n \\sim \\textsf{normal}(\\mu_{z[n]},\\sigma_{z[n]}).\n\\]\nThis model is not directly supported by Stan because it involves discrete parameters \\(z_n\\), but Stan can sample \\(\\mu\\) and \\(\\sigma\\) by summing out the \\(z\\) parameter as described in the next section.\n\n\n\nTo implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If \\(Y\\) is a mixture of \\(K\\) normal distributions with locations \\(\\mu_k\\) and scales \\(\\sigma_k\\) with mixing proportions \\(\\lambda\\) in the unit \\(K\\)-simplex, then \\[\np_Y\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n=\n\\sum_{k=1}^K \\lambda_k \\, \\textsf{normal}\\left(y \\mid \\mu_k, \\sigma_k\\right).\n\\]\n\n\nThe log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b) = \\log \\left(\\exp(a) + \\exp(b)\\right).\n\\]\nIf \\(a\\) and \\(b\\) are probabilities on the log scale, then \\(\\exp(a) +\n\\exp(b)\\) is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in log_sum_exp function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as \\[\n\\log \\left( \\exp(a) + \\exp(b)\\right)\n= c + \\log \\left( \\exp(a - c) + \\exp(b - c) \\right),\n\\] where \\(c = \\max(a, b)\\). In this evaluation, one of the terms, \\(a - c\\) or \\(b - c\\), is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the \\(\\max(a, b)\\) out of the log-exp round trip.\nFor example, the mixture of \\(\\textsf{normal}(-1, 2)\\) with \\(\\textsf{normal}(3, 1)\\), with mixing proportion \\(\\lambda =\n[0.3,0.7]^{\\top}\\), can be implemented in Stan as follows.\nparameters {\n  real y;\n}\nmodel {\n  target += log_sum_exp(log(0.3) + normal_lpdf(y | -1, 2),\n                        log(0.7) + normal_lpdf(y | 3, 1));\n}\nThe log probability term is derived by taking \\[\\begin{align*}\n\\log\\, &p\\left(y \\mid \\lambda,\\mu,\\sigma \\right) \\\\\n&= \\log\\big( 0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\n            + 0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big) \\\\\n&= \\log\\bigg( \\exp\\Big(\\log\\big(0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\\big)\\Big)\n             + \\exp\\Big(\\log\\big(0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right)\\big)\\Big) \\bigg) \\\\\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}\\big(\n                         \\log(0.3) + \\log \\textsf{normal}\\left(y \\mid -1,2 \\right),\n                         \\log(0.7) + \\log \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big).\n\\end{align*}\\]\n\n\n\nIf a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because\nlog_half = log(0.5);\nfor (n in 1:N) {\n    target +=\n        log_sum_exp(log_half + normal_lpdf(y[n] | mu[1], sigma[1]),\n                    log_half + normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nthen the \\(\\log 0.5\\) term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version\nfor (n in 1:N) {\n  target += log_sum_exp(normal_lpdf(y[n] | mu[1], sigma[1]),\n                        normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nThe same result holds if there are \\(K\\) components and the mixing simplex \\(\\lambda\\) is symmetric, i.e., \\[\n\\lambda = \\left( \\frac{1}{K}, \\dotsc, \\frac{1}{K} \\right).\n\\]\nThe result follows from the identity \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(c + a, c + b)\n=\nc + \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b)\n\\] and the fact that adding a constant \\(c\\) to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.\n\n\n\nThe posterior \\(p(z_n \\mid y_n, \\mu, \\sigma)\\) over the mixture indicator \\(z_n\n\\in 1:K\\) is often of interest as \\(p(z_n = k \\mid y, \\mu, \\sigma)\\) is the posterior probability that that observation \\(y_n\\) was generated by mixture component \\(k\\). The posterior can be computed via Bayes’s rule, \\[\\begin{align*}\n\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right]\n  &\\propto p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right)\\, p\\left(z_n = k \\mid \\lambda\\right) \\\\\n  &= \\textsf{normal}\\left(y_n \\mid \\mu_k, \\sigma_k\\right) \\cdot \\lambda_k.\n\\end{align*}\\]\nThe normalization can be done via summation, because \\(z_n \\in 1{:}K\\) only takes on finitely many values. In detail, \\[\np\\left(z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right) =\n\\frac{p\\left(y_n \\mid z_n = k, \\mu, \\sigma \\right) \\cdot p\\left(z_n = k \\mid \\lambda \\right)}\n     {\\sum_{k' = 1}^K p\\left(y_n \\mid z_n = k', \\mu, \\sigma \\right)\n                    \\cdot p\\left(z_n = k' \\mid \\lambda \\right)}.\n\\]\nOn the log scale, the normalized probability is computed as \\[\\begin{align*}\n\\log\\,&\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda\\right] \\\\\n&= \\log p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right) + \\log \\Pr\\!\\left[z_n = k \\mid \\lambda\\right] \\\\\n&\\quad  - \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{k' = 1}^K \\big(\\log p\\left(y_n \\mid z_n = k', \\mu, \\sigma\\right)\n                                                 + \\log p\\left(z_n = k' \\mid \\lambda\\right)\\big).\n\\end{align*}\\] This can be coded up directly in Stan; the change-point model in the change point section provides an example.\n\n\n\nGiven the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.\ndata {\n  int&lt;lower=1&gt; K;          // number of mixture components\n  int&lt;lower=1&gt; N;          // number of data points\n  array[N] real y;         // observations\n}\nparameters {\n  simplex[K] theta;          // mixing proportions\n  ordered[K] mu;             // locations of mixture components\n  vector&lt;lower=0&gt;[K] sigma;  // scales of mixture components\n}\nmodel {\n  vector[K] log_theta = log(theta);  // cache log calculation\n  sigma ~ lognormal(0, 2);\n  mu ~ normal(0, 10);\n  for (n in 1:N) {\n    vector[K] lps = log_theta;\n    for (k in 1:K) {\n      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);\n    }\n    target += log_sum_exp(lps);\n  }\n}\nThe model involves K mixture components and N data points. The mixing proportion parameter theta is declared to be a unit \\(K\\)-simplex, whereas the component location parameter mu and scale parameter sigma are both defined to be K-vectors.\nThe location parameter mu is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components mu[k] is symmetric, as it is here (each component has an independent \\(\\textsf{normal}(0, 10)\\) prior). It would even be possible to include a hierarchical prior for the components.\nThe values in the scale array sigma are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.\nThe model declares a local array variable lps to be size K and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points n. For each such point, the log of \\(\\theta_k \\times\n\\textsf{normal}\\left(y_n \\mid \\mu_k,\\sigma_k\\right)\\) is calculated and added to the array lps. Then the log probability is incremented with the log sum of exponentials of those values.\n\n\n\n\nThere is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that lambda, y[n], mu[1], mu[2], and sigma[1], sigma[2] are all scalars and lambda is between 0 and 1.\nfor (n in 1:N) {\n  target += log_sum_exp(log(lambda)\n                          + normal_lpdf(y[n] | mu[1], sigma[1]),\n                        log1m(lambda)\n                          + normal_lpdf(y[n] | mu[2], sigma[2]));\nor equivalently\nfor (n in 1:N) {\n  target += log_mix(lambda,\n                    normal_lpdf(y[n] | mu[1], sigma[1]),\n                    normal_lpdf(y[n] | mu[2], sigma[2]))\n  };\nThis definition assumes that each observation \\(y_n\\) may have arisen from either of the mixture components. The density is \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma\\right)\n= \\prod_{n=1}^N \\big(\\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1 \\right)\n                 + (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2 \\right)\\big).\n\\]\nContrast the previous model with the following (erroneous) attempt to vectorize the model.\ntarget += log_sum_exp(log(lambda)\n                        + normal_lpdf(y | mu[1], sigma[1]),\n                      log1m(lambda)\n                        + normal_lpdf(y | mu[2], sigma[2]));\nor equivalently,\ntarget += log_mix(lambda,\n                  normal_lpdf(y | mu[1], sigma[1]),\n                  normal_lpdf(y | mu[2], sigma[2]));\nThis second definition implies that the entire sequence \\(y_1, \\dotsc, y_n\\) of observations comes form one component or the other, defining a different density, \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n= \\lambda \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right).\n\\]\n\n\n\nIn many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.\nWe have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs. female, or Democrat vs. Republican).\nIn other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.\n\n\nAs an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\), \\[\\begin{align*}\n\\mu_1, \\mu_2       &\\sim \\textsf{normal}(0, 10) \\\\\n\\sigma_1, \\sigma_2 &\\sim \\textsf{halfnormal}(0, 10) \\\\\n\\end{align*}\\]\nThe prior on the mixture ratio is uniform, \\[\n\\lambda \\sim \\textsf{uniform}(0, 1),\n\\] so that with the likelihood \\[\np\\left(y_n \\mid \\mu, \\sigma\\right)\n= \\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right),\n\\] the joint distribution \\(p(y, \\mu, \\sigma, \\lambda)\\) is exchangeable in the parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\) with \\(\\lambda\\) flipping to \\(1 - \\lambda\\).1\n\n\n\nIn cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.\n\n\nPosterior predictive distribution for a new observation \\(\\tilde{y}\\) given the complete parameter vector \\(\\theta\\) will be \\[\np(\\tilde{y} \\mid y)\n=\n\\int_{\\theta}\np(\\tilde{y} \\mid \\theta)\n\\, p(\\theta \\mid y)\n\\, \\textsf{d}\\theta.\n\\]\nThe normal mixture example from the previous section, with \\(\\theta =\n(\\mu, \\sigma, \\lambda)\\), shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing \\(p(\\tilde{y} \\mid y)\\), which is more efficient statistically in terms of effective sample size, or simulating draws of \\(\\tilde{y}\\), which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] y_tilde;\n  // ...\n}\ngenerated quantities {\n  vector[N_tilde] log_p_y_tilde;\n  for (n in 1:N_tilde) {\n    log_p_y_tilde[n]\n      = log_mix(lambda,\n                normal_lpdf(y_tilde[n] | mu[1], sigma[1])\n                normal_lpdf(y_tilde[n] | mu[2], sigma[2]));\n  }\n}\nIt is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply log_sum_exp of the posterior draws of log_p_y_tilde. The average log predictive density is then given by subtracting log(N_new).\n\n\n\nOften a mixture model will be applied to a clustering problem and there might be two data items \\(y_i\\) and \\(y_j\\) for which there is a question of whether they arose from the same mixture component. If we take \\(z_i\\) and \\(z_j\\) to be the component responsibility discrete variables, then the quantity of interest is \\(z_i = z_j\\), which can be summarized as an event probability \\[\n\\Pr[z_i = z_j \\mid y]\n=\n\\int_{\\theta}\n\\frac{\\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \\mid \\theta)}\n     {\\sum_{k=0}^1 \\sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \\mid \\theta)}\n\\,\np(\\theta \\mid y)\n\\,\n\\textsf{d}\\theta.\n\\]\nAs with other event probabilities, this can be calculated in the generated quantities block either by sampling \\(z_i\\) and \\(z_j\\) and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.\n\n\n\n\n\nZero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by Lambert (1992), add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.\nZero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.\n\n\nConsider the following example for zero-inflated Poisson distributions. There is a probability \\(\\theta\\) of observing a zero, and a probability \\(1 - \\theta\\) of observing a count with a \\(\\textsf{Poisson}(\\lambda)\\) distribution (now \\(\\theta\\) is being used for mixing proportions because \\(\\lambda\\) is the traditional notation for a Poisson mean parameter). Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as \\[\\begin{align*}\ny_n & = 0 & \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}(y_n \\mid \\lambda) & \\quad\\text{with probability } 1-\\theta.\n\\end{align*}\\]\nStan does not support conditional distribution statements (with ~) conditional on some parameter, and we need to consider the corresponding likelihood \\[\np(y_n \\mid \\theta,\\lambda)\n=\n\\begin{cases}\n\\theta + (1 - \\theta) \\times \\textsf{Poisson}(0 \\mid \\lambda) & \\quad\\text{if } y_n = 0, \\text{ and}\\\\\n(1-\\theta) \\times \\textsf{Poisson}(y_n \\mid \\lambda) &\\quad\\text{if } y_n &gt; 0.\n\\end{cases}\n\\] The log likelihood can be coded directly in Stan (with target +=) as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  for (n in 1:N) {\n    if (y[n] == 0) {\n      target += log_sum_exp(log(theta),\n                            log1m(theta)\n                              + poisson_lpmf(y[n] | lambda));\n    } else {\n      target += log1m(theta)\n                  + poisson_lpmf(y[n] | lambda);\n    }\n  }\n}\nThe log1m(theta) computes log(1-theta), but is more computationally stable. The log_sum_exp(lp1,lp2) function adds the log probabilities on the linear scale; it is defined to be equal to log(exp(lp1) + exp(lp2)), but is more computationally stable and faster.\n\n\nThe code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also poisson_lpmf(0 | lambda) every time the first condition body executes. The use of the redundant terms is conditioned on y, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.\nThe number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare y_nonzero, so it must be computed in a function.\nfunctions {\n  int num_zeros(array[] int y) {\n    int sum = 0;\n    for (n in 1:size(y)) {\n      sum += (y[n] == 0);\n    }\n    return sum;\n  }\n}\n// ...\ntransformed data {\n  int&lt;lower=0&gt; N_zero = num_zeros(y);\n  array[N - N_zero] int&lt;lower=1&gt; y_nonzero;\n  int N_nonzero = 0;\n  for (n in 1:N) {\n    if (y[n] == 0) continue;\n    N_nonzero += 1;\n    y_nonzero[N_nonzero] = y[n];\n  }\n}\n// ...\nmodel {\n  // ...\n   target\n     += N_zero\n          * log_sum_exp(log(theta),\n                        log1m(theta)\n                          + poisson_lpmf(0 | lambda));\n   target += N_nonzero * log1m(theta);\n   target += poisson_lpmf(y_nonzero | lambda);\n  // ...\n}\nThe boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if y_nonzero is empty, N_nonzero will be zero, and the last two target increment terms will add zeros.\n\n\n\n\nThe hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as [ \\[\\begin{align*}\ny_n & = 0 \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}_{x\\neq 0}(y_n \\mid \\lambda) \\quad\\text{with probability } 1-\\theta,\n\\end{align*}\\] ] Where \\(\\textsf{Poisson}_{x\\neq 0}\\) is a truncated Poisson distribution, truncated at \\(0\\).\nThe corresponding likelihood function for the hurdle model is defined by \\[\np(y\\mid\\theta,\\lambda)\n=\n\\begin{cases}\n\\theta &\\quad\\text{if } y = 0, \\text{ and}\\\\\n(1 - \\theta)\n   \\frac{\\displaystyle \\textsf{Poisson}(y \\mid \\lambda)}\n        {\\displaystyle  1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)}\n&\\quad\\text{if } y &gt; 0,\n\\end{cases}\n\\] where \\(\\textsf{PoissonCDF}\\) is the cumulative distribution function for the Poisson distribution and and \\(1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)\\) is the relative normalization term for the truncated Poisson (truncated at \\(0\\)).\nThe hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.\nif (y[n] == 0) {\n  target += log(theta);\n} else {\n  target += log1m(theta) + poisson_lpmf(y[n] | lambda)\n            - poisson_lccdf(0 | lambda));\n}\nJulian King pointed out that because \\[\\begin{align*}\n\\log \\left( 1 - \\textsf{PoissonCDF}(0 \\mid \\lambda) \\right)\n  &= \\log \\left( 1 - \\textsf{Poisson}(0 \\mid \\lambda) \\right) \\\\\n  &= \\log(1 - \\exp(-\\lambda))\n\\end{align*}\\] the CCDF in the else clause can be replaced with a simpler expression.\ntarget += log1m(theta) + poisson_lpmf(y[n] | lambda)\n          - log1m_exp(-lambda));\nThe resulting code is about 15% faster than the code with the CCDF.\nThis is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size \\(N=200\\) and parameters \\(\\theta=0.3\\) and \\(\\lambda = 8\\), the speedup is a factor of 10; it will be lower for smaller \\(N\\) and greater for larger \\(N\\); it will also be greater for larger \\(\\theta\\).\nTo achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,\nfunctions {\n  int num_zero(array[] int y) {\n    int nz = 0;\n    for (n in 1:size(y)) {\n      if (y[n] == 0) {\n        nz += 1;\n      }\n    }\n    return nz;\n  }\n}\nThen a transformed data block can be used to store the sufficient statistics,\ntransformed data {\n  int&lt;lower=0, upper=N&gt; N0 = num_zero(y);\n  int&lt;lower=0, upper=N&gt; Ngt0 = N - N0;\n  array[N - num_zero(y)] int&lt;lower=1&gt; y_nz;\n  {\n    int pos = 1;\n    for (n in 1:N) {\n      if (y[n] != 0) {\n        y_nz[pos] = y[n];\n        pos += 1;\n      }\n    }\n  }\n}\nThe model block is then reduced to three statements.\nmodel {\n  N0 ~ binomial(N, theta);\n  y_nz ~ poisson(lambda);\n  target += -Ngt0 * log1m_exp(-lambda);\n}\nThe first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant Ngt0; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.\n\n\n\n\nSuppose we have a two-component mixture model with mixing rate \\(\\lambda \\in (0, 1)\\). Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are \\(N\\) observations, the mixture components will be estimated with effective data sizes of \\(\\theta \\, N\\) and \\((1 - \\theta)\n\\, N\\) for the two components for some \\(\\theta \\in (0, 1)\\). The effective weighting size is determined by posterior responsibility, not simply by the mixing rate \\(\\lambda\\).\n\n\nIn contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data \\(y\\).\nIf different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see Hoeting et al. (1999) and Gelman et al. (2013) for theoretical and computational details).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html#rao-blackwell.section",
-    "href": "stan-users-guide/latent-discrete.html#rao-blackwell.section",
-    "title": "Latent Discrete Parameters",
+    "objectID": "stan-users-guide/finite-mixtures.html#clustering-mixture.section",
+    "href": "stan-users-guide/finite-mixtures.html#clustering-mixture.section",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "Although it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.\nStandard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.",
+    "text": "Clustering models, as discussed in the clustering chapter, are just a particular class of mixture models that have been widely applied to clustering in the engineering and machine-learning literature. The normal mixture model discussed in this chapter reappears in multivariate form as the statistical basis for the \\(K\\)-means algorithm; the latent Dirichlet allocation model, usually applied to clustering problems, can be viewed as a mixed-membership multinomial mixture model.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html#change-point.section",
-    "href": "stan-users-guide/latent-discrete.html#change-point.section",
-    "title": "Latent Discrete Parameters",
+    "objectID": "stan-users-guide/finite-mixtures.html#latent-discrete-parameterization",
+    "href": "stan-users-guide/finite-mixtures.html#latent-discrete-parameterization",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "The first example is a model of coal mining disasters in the U.K. for the years 1851–1962.2\n\n\nFonnesbeck et al. (2013, sec. 3.1) provides a Poisson model of disaster \\(D_t\\) in year \\(t\\) with two rate parameters, an early rate (\\(e\\)) and late rate (\\(l\\)), that change at a given point in time \\(s\\). The full model expressed using a latent discrete parameter \\(s\\) is \\[\\begin{align*}\ne   &\\sim  \\textsf{exponential}(r_e) \\\\\nl   &\\sim  \\textsf{exponential}(r_l) \\\\\ns   &\\sim  \\textsf{uniform}(1, T) \\\\\nD_t &\\sim  \\textsf{Poisson}(t &lt; s \\; ? \\; e \\: : \\: l)\n\\end{align*}\\]\nThe last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.3\nIt uses a compact notation involving separating its three arguments by a question mark (?) and a colon (:). The conditional operator is defined by \\[\nc \\; ? \\; x_1 \\: : \\: x_2\n=\n\\begin{cases}\n\\ x_1 & \\quad\\text{if } c \\text{ is true (i.e., non-zero), and} \\\\\n\\ x_2 & \\quad\\text{if } c \\text{ is false (i.e., zero).}\n\\end{cases}\n\\]\n\n\n\nTo code this model in Stan, the discrete parameter \\(s\\) must be marginalized out to produce a model defining the log of the probability function \\(p(e,l,D_t)\\). The full joint probability factors as \\[\\begin{align*}\np(e,l,s,D) &=  p(e) \\, p(l) \\, p(s) \\, p(D \\mid s, e, l) \\\\\n&= \\textsf{exponential}(e \\mid r_e) \\ \\textsf{exponential}(l \\mid r_l) \\,\n    \\textsf{uniform}(s \\mid 1, T) \\\\\n& \\qquad \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nTo marginalize, an alternative factorization into prior and likelihood is used, \\[\np(e,l,D) = p(e,l) \\, p(D \\mid e,l),\n\\]\nwhere the likelihood is defined by marginalizing \\(s\\) as \\[\\begin{align*}\np(D \\mid e,l) &= \\sum_{s=1}^T p(s, D \\mid e,l) \\\\\n&= \\sum_{s=1}^T p(s) \\, p(D \\mid s,e,l) \\\\\n&= \\sum_{s=1}^T \\textsf{uniform}(s \\mid 1,T) \\,\n    \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nStan operates on the log scale and thus requires the log likelihood, \\[\\begin{align*}\n\\log p(D \\mid e,l)\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{s=1}^T\n    \\left( \\log \\textsf{uniform}(s \\mid 1, T) \\vphantom{\\sum_{t=1}^T}\\right. \\\\\n&\\qquad \\left.\n    + \\sum_{t=1}^T \\log \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l)\n\\right),\n\\end{align*}\\] where the log sum of exponents function is defined by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{n=1}^N \\, \\alpha_n =\n\\log \\sum_{n=1}^N \\exp(\\alpha_n).\n\\]\nThe log sum of exponents function allows the model to be coded directly in Stan using the built-in function log_sum_exp, which provides both arithmetic stability and efficiency for mixture model calculations.\n\n\n\nThe Stan program for the change point model is shown in the figure below. The transformed parameter lp[s] stores the quantity \\(\\log p(s, D \\mid e, l)\\).\ndata {\n  real&lt;lower=0&gt; r_e;\n  real&lt;lower=0&gt; r_l;\n\n  int&lt;lower=1&gt; T;\n  array[T] int&lt;lower=0&gt; D;\n}\ntransformed data {\n  real log_unif;\n  log_unif = -log(T);\n}\nparameters {\n  real&lt;lower=0&gt; e;\n  real&lt;lower=0&gt; l;\n}\ntransformed parameters {\n  vector[T] lp;\n  lp = rep_vector(log_unif, T);\n  for (s in 1:T) {\n    for (t in 1:T) {\n      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);\n    }\n  }\n}\nmodel {\n  e ~ exponential(r_e);\n  l ~ exponential(r_l);\n  target += log_sum_exp(lp);\n}\nA change point model in which disaster rates D[t] have one rate, e, before the change point and a different rate, l, after the change point. The change point itself, s, is marginalized out as described in the text.\nAlthough the change-point model is coded directly, the doubly nested loop used for s and t is quadratic in T. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.\ntransformed parameters {\n    vector[T] lp;\n    {\n      vector[T + 1] lp_e;\n      vector[T + 1] lp_l;\n      lp_e[1] = 0;\n      lp_l[1] = 0;\n      for (t in 1:T) {\n        lp_e[t + 1] = lp_e[t] + poisson_lpmf(D[t] | e);\n        lp_l[t + 1] = lp_l[t] + poisson_lpmf(D[t] | l);\n      }\n      lp = rep_vector(log_unif + lp_l[T + 1], T)\n           + head(lp_e, T) - head(lp_l, T);\n    }\n  }\nAs should be obvious from looking at it, it has linear complexity in T rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger T.\nThe key to understanding Wiklendt’s dynamic programming version is to see that head(lp_e) holds the forward values, whereas lp_l[T + 1] - head(lp_l, T) holds the backward values; the clever use of subtraction allows lp_l to be accumulated naturally in the forward direction.\n\n\n\nThis model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.\nThe value of lp for each iteration for each change point is available because it is declared as a transformed parameter. If the value of lp were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.\n\n\n\nThe value of lp[s] in a given iteration is given by \\(\\log\np(s,D \\mid e,l)\\) for the values of the early and late rates, \\(e\\) and \\(l\\), in the iteration. In each iteration after convergence, the early and late disaster rates, \\(e\\) and \\(l\\), are drawn from the posterior \\(p(e,l \\mid D)\\) by MCMC sampling and the associated lp calculated. The value of lp may be normalized to calculate \\(p(s \\mid e,l,D)\\) in each iteration, based on on the current values of \\(e\\) and \\(l\\). Averaging over iterations provides an unnormalized probability estimate of the change point being \\(s\\) (see below for the normalizing constant), \\[\\begin{align*}\np(s \\mid D) &\\propto q(s \\mid D) \\\\\n&= \\frac{1}{M} \\sum_{m=1}^{M} \\exp(\\texttt{lp}[m,s]).\n\\end{align*}\\] where \\(\\texttt{lp}[m,s]\\) represents the value of lp in posterior draw \\(m\\) for change point \\(s\\). By averaging over draws, \\(e\\) and \\(l\\) are themselves marginalized out, and the result has no dependence on a given iteration’s value for \\(e\\) and \\(l\\). A final normalization then produces the quantity of interest, the posterior probability of the change point being \\(s\\) conditioned on the data \\(D\\), \\[\np(s \\mid D) = \\frac{q(s \\mid D)}{\\sum_{s'=1}^T q(s' \\mid D)}.\n\\]\nA plot of the values of \\(\\log p(s \\mid D)\\) computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.\nLog probability of change point being in year, calculated analytically.\n\n\n\nAnalytical change-point posterior\n\n\nThe frequency of change points generated by sampling the discrete change points.\n\n\n\nSampled change-point posterior\n\n\nIn order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of \\(s\\) is roughly 1891.\n\n\n\nThe generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with lp defined as above, the following program draws a random value for s at every iteration.\ngenerated quantities {\n  int&lt;lower=1, upper=T&gt; s;\n  s = categorical_logit_rng(lp);\n}\nA posterior histogram of draws for \\(s\\) is shown on the second change point posterior figure above.\nCompared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for s given by softmax(lp).\n\n\n\nThe discrete sample generated for \\(s\\) can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using lp.\n\n\n\nThere is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, e, m, and l, and two loops for the change point and then one over time, with log densities being stored in a matrix.\nmatrix[T, T] lp;\nlp = rep_matrix(log_unif, T);\nfor (s1 in 1:T) {\n  for (s2 in 1:T) {\n    for (t in 1:T) {\n      lp[s1,s2] = lp[s1,s2]\n        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));\n    }\n  }\n}\nThe matrix can then be converted back to a vector using to_vector before being passed to log_sum_exp.",
+    "text": "One way to parameterize a mixture model is with a latent categorical variable indicating which mixture component was responsible for the outcome. For example, consider \\(K\\) normal distributions with locations \\(\\mu_k \\in \\mathbb{R}\\) and scales \\(\\sigma_k \\in (0,\\infty)\\). Now consider mixing them in proportion \\(\\lambda\\), where \\(\\lambda_k \\geq 0\\) and \\(\\sum_{k=1}^K \\lambda_k = 1\\) (i.e., \\(\\lambda\\) lies in the unit \\(K\\)-simplex). For each outcome \\(y_n\\) there is a latent variable \\(z_n\\) in \\(\\{ 1,\\dotsc,K \\}\\) with a categorical distribution parameterized by \\(\\lambda\\), \\[\nz_n \\sim \\textsf{categorical}(\\lambda).\n\\]\nThe variable \\(y_n\\) is distributed according to the parameters of the mixture component \\(z_n\\), \\[\ny_n \\sim \\textsf{normal}(\\mu_{z[n]},\\sigma_{z[n]}).\n\\]\nThis model is not directly supported by Stan because it involves discrete parameters \\(z_n\\), but Stan can sample \\(\\mu\\) and \\(\\sigma\\) by summing out the \\(z\\) parameter as described in the next section.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html#mark-recapture-models",
-    "href": "stan-users-guide/latent-discrete.html#mark-recapture-models",
-    "title": "Latent Discrete Parameters",
+    "objectID": "stan-users-guide/finite-mixtures.html#summing-out-the-responsibility-parameter",
+    "href": "stan-users-guide/finite-mixtures.html#summing-out-the-responsibility-parameter",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "A widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.\nThe first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.\n\n\nIn the simplest case, a one-stage mark-recapture study produces the following data\n\n\\(M\\) : number of animals marked in first capture,\n\\(C\\) : number animals in second capture, and\n\\(R\\) : number of marked animals in second capture.\n\nThe estimand of interest is\n\n\\(N\\) : number of animals in the population.\n\nDespite the notation, the model will take \\(N\\) to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.\nThe Lincoln-Petersen (Lincoln 1930; Petersen 1896) method for estimating population size is \\[\n\\hat{N} = \\frac{M C}{R}.\n\\]\nThis population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, \\[\nR \\sim \\textsf{binomial}(C, M / N)\n\\] given the total number of animals captured in the second round (\\(C\\)) with a recapture probability of \\(M/N\\), the fraction of the total population \\(N\\) marked in the first round.\ndata {\n  int&lt;lower=0&gt; M;\n  int&lt;lower=0&gt; C;\n  int&lt;lower=0, upper=min(M, C)&gt; R;\n}\nparameters {\n  real&lt;lower=(C - R + M)&gt; N;\n}\nmodel {\n  R ~ binomial(C, M / N);\n}\nA probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on \\(N\\) is necessary to efficiently eliminate impossible values.\nThe probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.\nTo ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for \\(N\\) is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.\nThe one tricky part of the model is the lower bound \\(C - R + M\\) placed on the population size \\(N\\). Values below this bound are impossible because it is otherwise not possible to draw \\(R\\) samples out of the \\(C\\) animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for \\(C\\) implies a variable transform \\(f : (C-R+M,\\infty) \\rightarrow (-\\infty,+\\infty)\\) defined by \\(f(N) = \\log(N - (C - R + M))\\); the reference manual contains full details of all constrained parameter transforms.\n\n\n\nThe Cormack-Jolly-Seber (CJS) model (Cormack 1964; Jolly 1965; Seber 1965) is an open-population model in which the population may change over time due to death; the presentation here draws heavily on Schofield (2007).\nThe basic data are\n\n\\(I\\): number of individuals,\n\\(T\\): number of capture periods, and\n\\(y_{i,t}\\): Boolean indicating if individual \\(i\\) was captured at time \\(t\\).\n\nEach individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.\nThere are two Bernoulli parameters in the model,\n\n\\(\\phi_t\\) : probability that animal alive at time \\(t\\) survives until \\(t + 1\\) and\n\\(p_t\\) : probability that animal alive at time \\(t\\) is captured at time \\(t\\).\n\nThese parameters will both be given uniform priors, but information should be used to tighten these priors in practice.\nThe CJS model also employs a latent discrete parameter \\(z_{i,t}\\) indicating for each individual \\(i\\) whether it is alive at time \\(t\\), distributed as \\[\nz_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t-1} \\; ? \\; 0 \\: : \\: \\phi_{t-1}).\n\\]\nThe conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on \\(z\\) as \\[\ny_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t} \\; ? \\; 0 \\: : \\: p_t).\n\\]\nThe conditional enforces the constraint that dead animals cannot be captured.\n\n\n\nThis subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.\nIn order to ease the marginalization of the latent discrete parameter \\(z_{i,t}\\), the Stan models rely on a derived quantity \\(\\chi_t\\) for the probability that an individual is never captured again if it is alive at time \\(t\\) (if it is dead, the recapture probability is zero). this quantity is defined recursively by \\[\n\\chi_t\n=\n\\begin{cases}\n1 & \\quad\\text{if } t = T \\\\\n(1 - \\phi_t) + \\phi_t (1 - p_{t+1}) \\chi_{t+1}\n  & \\quad\\text{if } t &lt; T\n\\end{cases}\n\\]\nThe base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining \\(\\chi_{t}\\) in terms of \\(\\chi_{t+1}\\) involves two possibilities: (1) not surviving to the next time period, with probability \\((1 - \\phi_t)\\), or (2) surviving to the next time period with probability \\(\\phi_t\\), not being captured in the next time period with probability \\((1 - p_{t+1})\\), and not being captured again after being alive in period \\(t+1\\) with probability \\(\\chi_{t+1}\\).\nWith three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.\n\\[\n\\begin{array}{crclc}\n\\hline\n& \\qquad\\qquad & captures & \\qquad\\qquad &  \\\\\n\\mathrm{profile} & 1 & 2 & 3 & \\mathrm{probability} \\\\\n\\hline\n0 & - & - & - & n/a \\\\\n1 & - & - & + & n/a \\\\\n2 & - & + & - & \\chi_2 \\\\\n3 & - & + & + & \\phi_2 \\, p_3 \\\\\n4 & + & - & - & \\chi_1 \\\\\n5 & + & - & + & \\phi_1 \\, (1 - p_2) \\, \\phi_2 \\, p_3 \\\\\n6 & + & + & - & \\phi_1 \\, p_2 \\, \\chi_2 \\\\\n7 & + & + & + & \\phi_1 \\, p_2 \\, \\phi_2 \\, p_3 \\\\\n\\hline\n\\end{array}\n\\]\nHistory 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.\nBy defining these probabilities in terms of \\(\\chi\\) directly, there is no need for a latent binary parameter indicating whether an animal is alive at time \\(t\\) or not. The definition of \\(\\chi\\) is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model (Schofield 2007).\nThe Stan model defines \\(\\chi\\) as a transformed parameter based on parameters \\(\\phi\\) and \\(p\\). In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using target += rather than being part of a built-in probability function.\nThe following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods\ndata {\n  array[7] int&lt;lower=0&gt; history;\n}\nparameters {\n  array[2] real&lt;lower=0, upper=1&gt; phi;\n  array[3] real&lt;lower=0, upper=1&gt; p;\n}\ntransformed parameters {\n  array[2] real&lt;lower=0, upper=1&gt; chi;\n  chi[2] = (1 - phi[2]) + phi[2] * (1 - p[3]);\n  chi[1] = (1 - phi[1]) + phi[1] * (1 - p[2]) * chi[2];\n}\nmodel {\n  target += history[2] * log(chi[2]);\n  target += history[3] * (log(phi[2]) + log(p[3]));\n  target += history[4] * (log(chi[1]));\n  target += history[5] * (log(phi[1]) + log1m(p[2])\n                            + log(phi[2]) + log(p[3]));\n  target += history[6] * (log(phi[1]) + log(p[2])\n                            + log(chi[2]));\n  target += history[7] * (log(phi[1]) + log(p[2])\n                            + log(phi[2]) + log(p[3]));\n}\ngenerated quantities {\n  real&lt;lower=0, upper=1&gt; beta3;\n  beta3 = phi[2] * p[3];\n}\n\n\nThe parameters \\(\\phi_2\\) and \\(p_3\\), the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, \\(\\beta_3 = \\phi_2 \\, p_3\\), is identified. The Stan model defines beta3 as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the \\(p\\) and \\(\\phi\\) parameters.\n\n\n\n\nThis section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number \\(T\\) of capture events, the number \\(I\\) of individuals, and a boolean flag \\(y_{i,t}\\) indicating if individual \\(i\\) was observed at time \\(t\\). In Stan,\ndata {\n  int&lt;lower=2&gt; T;\n  int&lt;lower=0&gt; I;\n  array[I, T] int&lt;lower=0, upper=1&gt; y;\n}\nThe advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding \\(2^T\\) observation histories for \\(T\\) capture times.\n\n\nThe individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.4\nfunctions {\n  int first_capture(array[] int y_i) {\n    for (k in 1:size(y_i)) {\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  int last_capture(array[] int y_i) {\n    for (k_rev in 0:(size(y_i) - 1)) {\n      int k;\n      k = size(y_i) - k_rev;\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  // ...\n}\nThese two functions are used to define the first and last capture time for each individual in the transformed data block.5\ntransformed data {\n  array[I] int&lt;lower=0, upper=T&gt; first;\n  array[I] int&lt;lower=0, upper=T&gt; last;\n  vector&lt;lower=0, upper=I&gt;[T] n_captured;\n  for (i in 1:I) {\n    first[i] = first_capture(y[i]);\n  }\n  for (i in 1:I) {\n    last[i] = last_capture(y[i]);\n  }\n  n_captured = rep_vector(0, T);\n  for (t in 1:T) {\n    for (i in 1:I) {\n      if (y[i, t]) {\n        n_captured[t] = n_captured[t] + 1;\n      }\n    }\n  }\n}\nThe transformed data block also defines n_captured[t], which is the total number of captures at time t. The variable n_captured is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.\nThe parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector chi, the probability that if an individual is alive at t that it will never be captured again.\nparameters {\n  vector&lt;lower=0, upper=1&gt;[T - 1] phi;\n  vector&lt;lower=0, upper=1&gt;[T] p;\n}\ntransformed parameters {\n  vector&lt;lower=0, upper=1&gt;[T] chi;\n  chi = prob_uncaptured(T, p, phi);\n}\nThe definition of prob_uncaptured, from the functions block, is\nfunctions {\n  // ...\n  vector prob_uncaptured(int T, vector p, vector phi) {\n    vector[T] chi;\n    chi[T] = 1.0;\n    for (t in 1:(T - 1)) {\n      int t_curr;\n      int t_next;\n      t_curr = T - t;\n      t_next = t_curr + 1;\n      chi[t_curr] = (1 - phi[t_curr])\n                     + phi[t_curr]\n                       * (1 - p[t_next])\n                       * chi[t_next];\n    }\n    return chi;\n  }\n}\nThe function definition directly follows the mathematical definition of \\(\\chi_t\\), unrolling the recursion into an iteration and defining the elements of chi from T down to 1.\n\n\n\nGiven the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations q given the parameters p and phi as well as the transformed parameter chi defined in terms of p and phi.\nmodel {\n  for (i in 1:I) {\n    if (first[i] &gt; 0) {\n      for (t in (first[i]+1):last[i]) {\n        1 ~ bernoulli(phi[t - 1]);\n        y[i, t] ~ bernoulli(p[t]);\n      }\n      1 ~ bernoulli(chi[last[i]]);\n    }\n  }\n}\nThe outer loop is over individuals, conditional skipping individuals i which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for first if an individual is never captured.\nThe inner loop for individual i first increments the log probability based on the survival of the individual with probability phi[t - 1]. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.\nIn the inner loop, the observed capture status y[i, t] for individual i at time t has a Bernoulli distribution based on the capture probability p[t] at time t.\nAfter the inner loop, the probability of an animal never being seen again after being observed at time last[i] is included, because last[i] was defined to be the last time period in which animal i was observed.\n\n\n\nAs with the collective model described in the previous subsection, this model does not identify phi[T - 1] and p[T], but does identify their product, beta. Thus beta is defined as a generated quantity to monitor convergence and report.\ngenerated quantities {\n  real beta;\n  // ...\n\n  beta = phi[T - 1] * p[T];\n  // ...\n}\nThe parameter p[1] is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case p[1] would be an unknown initial condition and both phi[T - 1] and p[T] could be identified.\n\n\n\nThe generated quantities also calculates an estimate of the population mean at each time t in the same way as in the simple mark-recapture model as the number of individuals captured at time t divided by the probability of capture at time t. This is done with the elementwise division operation for vectors (./) in the generated quantities block.\ngenerated quantities {\n  // ...\n  vector&lt;lower=0&gt;[T] pop;\n  // ...\n  pop = n_captured ./ p;\n  pop[1] = -1;\n}\n\n\n\nAll individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.",
+    "text": "To implement the normal mixture model outlined in the previous section in Stan, the discrete parameters can be summed out of the model. If \\(Y\\) is a mixture of \\(K\\) normal distributions with locations \\(\\mu_k\\) and scales \\(\\sigma_k\\) with mixing proportions \\(\\lambda\\) in the unit \\(K\\)-simplex, then \\[\np_Y\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n=\n\\sum_{k=1}^K \\lambda_k \\, \\textsf{normal}\\left(y \\mid \\mu_k, \\sigma_k\\right).\n\\]\n\n\nThe log sum of exponentials function is used to define mixtures on the log scale. It is defined for two inputs by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b) = \\log \\left(\\exp(a) + \\exp(b)\\right).\n\\]\nIf \\(a\\) and \\(b\\) are probabilities on the log scale, then \\(\\exp(a) +\n\\exp(b)\\) is their sum on the linear scale, and the outer log converts the result back to the log scale; to summarize, log_sum_exp does linear addition on the log scale. The reason to use Stan’s built-in log_sum_exp function is that it can prevent underflow and overflow in the exponentiation, by calculating the result as \\[\n\\log \\left( \\exp(a) + \\exp(b)\\right)\n= c + \\log \\left( \\exp(a - c) + \\exp(b - c) \\right),\n\\] where \\(c = \\max(a, b)\\). In this evaluation, one of the terms, \\(a - c\\) or \\(b - c\\), is zero and the other is negative, thus eliminating the possibility of overflow or underflow in the leading term while extracting the most arithmetic precision possible by pulling the \\(\\max(a, b)\\) out of the log-exp round trip.\nFor example, the mixture of \\(\\textsf{normal}(-1, 2)\\) with \\(\\textsf{normal}(3, 1)\\), with mixing proportion \\(\\lambda =\n[0.3,0.7]^{\\top}\\), can be implemented in Stan as follows.\nparameters {\n  real y;\n}\nmodel {\n  target += log_sum_exp(log(0.3) + normal_lpdf(y | -1, 2),\n                        log(0.7) + normal_lpdf(y | 3, 1));\n}\nThe log probability term is derived by taking \\[\\begin{align*}\n\\log\\, &p\\left(y \\mid \\lambda,\\mu,\\sigma \\right) \\\\\n&= \\log\\big( 0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\n            + 0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big) \\\\\n&= \\log\\bigg( \\exp\\Big(\\log\\big(0.3 \\times \\textsf{normal}\\left(y \\mid -1,2 \\right)\\big)\\Big)\n             + \\exp\\Big(\\log\\big(0.7 \\times \\textsf{normal}\\left(y \\mid 3,1 \\right)\\big)\\Big) \\bigg) \\\\\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}\\big(\n                         \\log(0.3) + \\log \\textsf{normal}\\left(y \\mid -1,2 \\right),\n                         \\log(0.7) + \\log \\textsf{normal}\\left(y \\mid 3,1 \\right) \\big).\n\\end{align*}\\]\n\n\n\nIf a two-component mixture has a mixing ratio of 0.5, then the mixing ratios can be dropped, because\nlog_half = log(0.5);\nfor (n in 1:N) {\n    target +=\n        log_sum_exp(log_half + normal_lpdf(y[n] | mu[1], sigma[1]),\n                    log_half + normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nthen the \\(\\log 0.5\\) term isn’t contributing to the proportional density, and the above can be replaced with the more efficient version\nfor (n in 1:N) {\n  target += log_sum_exp(normal_lpdf(y[n] | mu[1], sigma[1]),\n                        normal_lpdf(y[n] | mu[2], sigma[2]));\n}\nThe same result holds if there are \\(K\\) components and the mixing simplex \\(\\lambda\\) is symmetric, i.e., \\[\n\\lambda = \\left( \\frac{1}{K}, \\dotsc, \\frac{1}{K} \\right).\n\\]\nThe result follows from the identity \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(c + a, c + b)\n=\nc + \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}(a, b)\n\\] and the fact that adding a constant \\(c\\) to the log density accumulator has no effect because the log density is only specified up to an additive constant in the first place. There is nothing specific to the normal distribution here; constants may always be dropped from the target.\n\n\n\nThe posterior \\(p(z_n \\mid y_n, \\mu, \\sigma)\\) over the mixture indicator \\(z_n\n\\in 1:K\\) is often of interest as \\(p(z_n = k \\mid y, \\mu, \\sigma)\\) is the posterior probability that that observation \\(y_n\\) was generated by mixture component \\(k\\). The posterior can be computed via Bayes’s rule, \\[\\begin{align*}\n\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right]\n  &\\propto p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right)\\, p\\left(z_n = k \\mid \\lambda\\right) \\\\\n  &= \\textsf{normal}\\left(y_n \\mid \\mu_k, \\sigma_k\\right) \\cdot \\lambda_k.\n\\end{align*}\\]\nThe normalization can be done via summation, because \\(z_n \\in 1{:}K\\) only takes on finitely many values. In detail, \\[\np\\left(z_n = k \\mid y_n, \\mu, \\sigma, \\lambda \\right) =\n\\frac{p\\left(y_n \\mid z_n = k, \\mu, \\sigma \\right) \\cdot p\\left(z_n = k \\mid \\lambda \\right)}\n     {\\sum_{k' = 1}^K p\\left(y_n \\mid z_n = k', \\mu, \\sigma \\right)\n                    \\cdot p\\left(z_n = k' \\mid \\lambda \\right)}.\n\\]\nOn the log scale, the normalized probability is computed as \\[\\begin{align*}\n\\log\\,&\\Pr\\!\\left[z_n = k \\mid y_n, \\mu, \\sigma, \\lambda\\right] \\\\\n&= \\log p\\left(y_n \\mid z_n = k, \\mu, \\sigma\\right) + \\log \\Pr\\!\\left[z_n = k \\mid \\lambda\\right] \\\\\n&\\quad  - \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{k' = 1}^K \\big(\\log p\\left(y_n \\mid z_n = k', \\mu, \\sigma\\right)\n                                                 + \\log p\\left(z_n = k' \\mid \\lambda\\right)\\big).\n\\end{align*}\\] This can be coded up directly in Stan; the change-point model in the change point section provides an example.\n\n\n\nGiven the scheme for representing mixtures, it may be moved to an estimation setting, where the locations, scales, and mixture components are unknown. Further generalizing to a number of mixture components specified as data yields the following model.\ndata {\n  int&lt;lower=1&gt; K;          // number of mixture components\n  int&lt;lower=1&gt; N;          // number of data points\n  array[N] real y;         // observations\n}\nparameters {\n  simplex[K] theta;          // mixing proportions\n  ordered[K] mu;             // locations of mixture components\n  vector&lt;lower=0&gt;[K] sigma;  // scales of mixture components\n}\nmodel {\n  vector[K] log_theta = log(theta);  // cache log calculation\n  sigma ~ lognormal(0, 2);\n  mu ~ normal(0, 10);\n  for (n in 1:N) {\n    vector[K] lps = log_theta;\n    for (k in 1:K) {\n      lps[k] += normal_lpdf(y[n] | mu[k], sigma[k]);\n    }\n    target += log_sum_exp(lps);\n  }\n}\nThe model involves K mixture components and N data points. The mixing proportion parameter theta is declared to be a unit \\(K\\)-simplex, whereas the component location parameter mu and scale parameter sigma are both defined to be K-vectors.\nThe location parameter mu is declared to be an ordered vector in order to identify the model. This will not affect inferences that do not depend on the ordering of the components as long as the prior for the components mu[k] is symmetric, as it is here (each component has an independent \\(\\textsf{normal}(0, 10)\\) prior). It would even be possible to include a hierarchical prior for the components.\nThe values in the scale array sigma are constrained to be non-negative, and have a weakly informative prior given in the model chosen to avoid zero values and thus collapsing components.\nThe model declares a local array variable lps to be size K and uses it to accumulate the log contributions from the mixture components. The main action is in the loop over data points n. For each such point, the log of \\(\\theta_k \\times\n\\textsf{normal}\\left(y_n \\mid \\mu_k,\\sigma_k\\right)\\) is calculated and added to the array lps. Then the log probability is incremented with the log sum of exponentials of those values.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html#data-coding-models.section",
-    "href": "stan-users-guide/latent-discrete.html#data-coding-models.section",
-    "title": "Latent Discrete Parameters",
+    "objectID": "stan-users-guide/finite-mixtures.html#vectorizing-mixtures",
+    "href": "stan-users-guide/finite-mixtures.html#vectorizing-mixtures",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "Although seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.\n\n\nSuppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.\nThere are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.\n\n\n\nHumans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.\n\n\n\nIn this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.\nDawid and Skene (1979) introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.\n\n\nThe data for the model consists of \\(J\\) raters (diagnostic tests), \\(I\\) items (patients), and \\(K\\) categories (condition statuses) to annotate, with \\(y_{i, j} \\in \\{1, \\dotsc, K\\}\\) being the rating provided by rater \\(j\\) for item \\(i\\). In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often \\(K=2\\), with values signaling the presence or absence of the condition.6\nIt is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.\n\n\n\n\nThe model is based on three parameters, the first of which is discrete:\n\n\\(z_i\\) : a value in \\(\\{1, \\dotsc, K\\}\\) indicating the true category of item \\(i\\),\n\\(\\pi\\) : a \\(K\\)-simplex for the prevalence of the \\(K\\) categories in the population, and\n\\(\\theta_{j,k}\\) : a \\(K\\)-simplex for the response of annotator \\(j\\) to an item of true category \\(k\\).\n\n\n\n\nThe true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, \\[\nz_i \\sim \\textsf{categorical}(\\pi).\n\\]\nThe rating \\(y_{i, j}\\) provided for item \\(i\\) by rater \\(j\\) is modeled as a categorical response of rater \\(i\\) to an item of category \\(z_i\\),7 \\[\ny_{i, j} \\sim \\textsf{categorical}(\\theta_{j,\\pi_{z[i]}}).\n\\]\n\n\nDawid and Skene provided maximum likelihood estimates for \\(\\theta\\) and \\(\\pi\\), which allows them to generate probability estimates for each \\(z_i\\).\nTo mimic Dawid and Skene’s maximum likelihood model, the parameters \\(\\theta_{j,k}\\) and \\(\\pi\\) can be given uniform priors over \\(K\\)-simplexes. It is straightforward to generalize to Dirichlet priors, \\[\n\\pi \\sim \\textsf{Dirichlet}(\\alpha)\n\\] and \\[\n\\theta_{j,k} \\sim \\textsf{Dirichlet}(\\beta_k)\n\\] with fixed hyperparameters \\(\\alpha\\) (a vector) and \\(\\beta\\) (a matrix or array of vectors). The prior for \\(\\theta_{j,k}\\) must be allowed to vary in \\(k\\), so that, for instance, \\(\\beta_{k,k}\\) is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.\nBecause there are \\(J\\) coders, it would be natural to extend the model to include a hierarchical prior for \\(\\beta\\) and to partially pool the estimates of coder accuracy and bias.\n\n\n\nBecause the true category parameter \\(z\\) is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as \\[\np(y, \\theta, \\pi) = p(y \\mid \\theta,\\pi) \\, p(\\pi) \\, p(\\theta),\n\\] where \\(p(y \\mid \\theta,\\pi)\\) is derived by marginalizing \\(z\\) out of \\[\np(z, y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\left( \\textsf{categorical}(z_i \\mid \\pi)\n                     \\prod_{j=1}^J\n                     \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, z[i]})\n              \\right).\n\\]\nThis can be done item by item, with \\[\np(y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\sum_{k=1}^K\n  \\left( \\textsf{categorical}(k \\mid \\pi)\n         \\prod_{j=1}^J\n         \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n  \\right).\n\\]\nIn the missing data model, only the observed labels would be used in the inner product.\nDawid and Skene (1979) derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, \\[\\begin{align*}\n\\log p(y \\mid \\theta, \\pi)\n&= \\sum_{i=1}^I \\log \\left( \\sum_{k=1}^K \\exp\n    \\left(\\log \\textsf{categorical}(k \\mid \\pi) \\vphantom{\\sum_{j=1}^J}\\right.\\right.\n    \\left.\\left. + \\ \\sum_{j=1}^J\n           \\log \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n    \\right) \\right),\n\\end{align*}\\] which can be directly coded using Stan’s built-in log_sum_exp function.\n\n\n\n\nThe Stan program for the Dawid and Skene model is provided below (Dawid and Skene 1979).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=1&gt; I;\n  int&lt;lower=1&gt; J;\n\n  array[I, J] int&lt;lower=1, upper=K&gt; y;\n\n  vector&lt;lower=0&gt;[K] alpha;\n  vector&lt;lower=0&gt;[K] beta[K];\n}\nparameters {\n  simplex[K] pi;\n  array[J, K] simplex[K] theta;\n}\ntransformed parameters {\n  array[I] vector[K] log_q_z;\n  for (i in 1:I) {\n    log_q_z[i] = log(pi);\n    for (j in 1:J) {\n      for (k in 1:K) {\n        log_q_z[i, k] = log_q_z[i, k]\n                         + log(theta[j, k, y[i, j]]);\n      }\n    }\n  }\n}\nmodel {\n  pi ~ dirichlet(alpha);\n  for (j in 1:J) {\n    for (k in 1:K) {\n      theta[j, k] ~ dirichlet(beta[k]);\n    }\n  }\n\n  for (i in 1:I) {\n    target += log_sum_exp(log_q_z[i]);\n  }\n}\nThe model marginalizes out the discrete parameter \\(z\\), storing the unnormalized conditional probability \\(\\log q(z_i=k|\\theta,\\pi)\\) in log_q_z[i, k].\nThe Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are \\(\\alpha_k = 3\\) and \\(\\beta_{k,k} = 2.5 K\\) and \\(\\beta_{k,k'} = 1\\) if \\(k \\neq k'\\). Taking \\(\\alpha\\) and \\(\\beta_k\\) to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of Dawid and Skene (1979).\n\n\nThe quantity log_q_z[i] is defined as a transformed parameter. It encodes the (unnormalized) log of \\(p(z_i \\mid \\theta,\n\\pi)\\). Each iteration provides a value conditioned on that iteration’s values for \\(\\theta\\) and \\(\\pi\\). Applying the softmax function to log_q_z[i] provides a simplex corresponding to the probability mass function of \\(z_i\\) in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each \\(z_i\\).",
+    "text": "There is (currently) no way to vectorize mixture models at the observation level in Stan. This section is to warn users away from attempting to vectorize naively, as it results in a different model. A proper mixture at the observation level is defined as follows, where we assume that lambda, y[n], mu[1], mu[2], and sigma[1], sigma[2] are all scalars and lambda is between 0 and 1.\nfor (n in 1:N) {\n  target += log_sum_exp(log(lambda)\n                          + normal_lpdf(y[n] | mu[1], sigma[1]),\n                        log1m(lambda)\n                          + normal_lpdf(y[n] | mu[2], sigma[2]));\nor equivalently\nfor (n in 1:N) {\n  target += log_mix(lambda,\n                    normal_lpdf(y[n] | mu[1], sigma[1]),\n                    normal_lpdf(y[n] | mu[2], sigma[2]))\n  };\nThis definition assumes that each observation \\(y_n\\) may have arisen from either of the mixture components. The density is \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma\\right)\n= \\prod_{n=1}^N \\big(\\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1 \\right)\n                 + (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2 \\right)\\big).\n\\]\nContrast the previous model with the following (erroneous) attempt to vectorize the model.\ntarget += log_sum_exp(log(lambda)\n                        + normal_lpdf(y | mu[1], sigma[1]),\n                      log1m(lambda)\n                        + normal_lpdf(y | mu[2], sigma[2]));\nor equivalently,\ntarget += log_mix(lambda,\n                  normal_lpdf(y | mu[1], sigma[1]),\n                  normal_lpdf(y | mu[2], sigma[2]));\nThis second definition implies that the entire sequence \\(y_1, \\dotsc, y_n\\) of observations comes form one component or the other, defining a different density, \\[\np\\left(y \\mid \\lambda, \\mu, \\sigma \\right)\n= \\lambda \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\prod_{n=1}^N \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right).\n\\]",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html#marginalization-mathematics.section",
-    "href": "stan-users-guide/latent-discrete.html#marginalization-mathematics.section",
-    "title": "Latent Discrete Parameters",
+    "objectID": "stan-users-guide/finite-mixtures.html#mixture-inference.section",
+    "href": "stan-users-guide/finite-mixtures.html#mixture-inference.section",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "This section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is not necessary to fit models with discrete parameters using Stan.\nThroughout, the model under consideration consists of both continuous parameters, \\(\\Theta\\), and discrete parameters, \\(Z\\). It is also assumed that \\(Z\\) can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.\n\n\n\nWhen performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating \\(\\mu\\), say, where \\(\\mu = \\mathbb{E}[g(\\Theta, Z)]\\) and \\(g(\\cdot)\\) is an arbitrary function. An example of such a quantity is \\(\\mathbb{E}[\\Theta]\\), the posterior mean of the continuous parameters, where we would take \\(g(\\theta, z) = \\theta\\). To estimate \\(\\mu\\) the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.\nIf both \\(\\Theta\\) and \\(Z\\) were continuous, Stan could be used to sample \\(M\\) draws from the joint posterior \\(p_{\\Theta, Z}(\\theta, z)\\) and then estimate \\(\\mu\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M {g(\\theta^{(i)}, z^{(i)})}.\n\\] Given \\(Z\\) is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out \\(Z\\) from the joint posterior to give the marginalized posterior \\(p_\\Theta(\\theta)\\). This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws \\(\\{\\theta^{(i)}\\}_{i = 1}^M\\) from the marginalized posterior.\nUsing only these draws, how can we estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\)? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function \\(g\\) of interest, we compute \\[\nh(\\Theta) = \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n\\] and then estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M h(\\theta^{(i)}).\n\\] This estimator is justified by the law of iterated expectation, the fact that \\[\n\\mathbb{E}[h(\\Theta)] = \\mathbb{E}[\\mathbb{E}[g(\\Theta, Z)] \\mid \\Theta] = \\mathbb{E}[g(\\Theta, Z)] = \\mu.\n\\] Using this marginalized estimator provides a way to estimate the expectation of any function \\(g(\\cdot)\\) for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\).\n\n\n\nFortunately, the discrete nature of \\(Z\\) makes evaluating \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\) easy. The function \\(h(\\Theta)\\) can be written as: \\[\nh(\\Theta)\n= \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n= \\sum_{k} g(\\Theta, k) \\Pr[Z = k \\mid \\Theta],\n\\] where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, \\(\\Pr[Z = k \\mid \\Theta]\\). More detail on how this quantity can be calculated is included below. Note that if \\(Z\\) takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.\nWhen \\(g(\\theta, z)\\) is a function of either \\(\\theta\\) or \\(z\\) only, the above formula simplifies further.\nIn the first case, where \\(g(\\theta, z) = g(\\theta)\\), we have: \\[\\begin{align*}\nh(\\Theta)\n&= \\sum_{k} g(\\Theta) \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta) \\sum_{k} \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta).\n\\end{align*}\\] This means that we can estimate \\(\\mathbb{E}[g(\\Theta)]\\) with the standard, seemingly unconditional, estimator: \\[\n\\frac{1}{M} \\sum_{i = 1}^M g(\\theta^{(i)}).\n\\] Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.\nIn the second case, where \\(g(\\theta, z) = g(z)\\), the conditional expectation instead simplifies as follows: \\[\nh(\\Theta) = \\sum_{k} g(k) \\Pr[Z = k \\mid \\Theta].\n\\] An important special case of this result is when \\(g(\\theta, z) = \\textrm{I}(z = k)\\), where \\(\\textrm{I}\\) is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable \\(Z\\), since \\(\\mathbb{E}[\\textrm{I}(Z = k)] = \\Pr[Z = k]\\). In this case, \\[\nh(\\Theta)\n= \\sum_{k} \\textrm{I}(z = k) \\Pr[Z = k \\mid \\Theta]\n= \\Pr[Z = k \\mid \\Theta].\n\\] The quantity \\(\\Pr[Z = k]\\) can therefore be estimated with: \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}].\n\\] When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, \\(Y\\). That is, we are really estimating \\(\\Pr[Z = k \\mid Y]\\) with \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y].\n\\] This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us \\[\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y]\n= \\frac{\\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}\n{\\sum_{k = 1}^K \\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}.\n\\] Here, \\(\\Pr[Y \\mid \\Theta = \\theta^{(i)}, Z = k]\\) is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.\nOther than the use of Bayes’s theorem, \\(\\Pr[Z = k \\mid \\theta = \\theta^{(i)}, Y]\\) can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the Dawid–Skene model).\nFor a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see Pullin, Gurrin, and Vukcevic (2021).",
+    "text": "In many mixture models, the mixture components are underlyingly exchangeable in the model and thus not identifiable. This arises if the parameters of the mixture components have exchangeable priors and the mixture ratio gets a uniform prior so that the parameters of the mixture components are also exchangeable in the likelihood.\nWe have finessed this basic problem by ordering the parameters. This will allow us in some cases to pick out mixture components either ahead of time or after fitting (e.g., male vs. female, or Democrat vs. Republican).\nIn other cases, we do not care about the actual identities of the mixture components and want to consider inferences that are independent of indexes. For example, we might only be interested in posterior predictions for new observations.\n\n\nAs an example, consider the normal mixture from the previous section, which provides an exchangeable prior on the pairs of parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\), \\[\\begin{align*}\n\\mu_1, \\mu_2       &\\sim \\textsf{normal}(0, 10) \\\\\n\\sigma_1, \\sigma_2 &\\sim \\textsf{halfnormal}(0, 10) \\\\\n\\end{align*}\\]\nThe prior on the mixture ratio is uniform, \\[\n\\lambda \\sim \\textsf{uniform}(0, 1),\n\\] so that with the likelihood \\[\np\\left(y_n \\mid \\mu, \\sigma\\right)\n= \\lambda \\times \\textsf{normal}\\left(y_n \\mid \\mu_1, \\sigma_1\\right)\n+ (1 - \\lambda) \\times \\textsf{normal}\\left(y_n \\mid \\mu_2, \\sigma_2\\right),\n\\] the joint distribution \\(p(y, \\mu, \\sigma, \\lambda)\\) is exchangeable in the parameters \\((\\mu_1, \\sigma_1)\\) and \\((\\mu_2, \\sigma_2)\\) with \\(\\lambda\\) flipping to \\(1 - \\lambda\\).1\n\n\n\nIn cases where the mixture components are not identifiable, it can be difficult to diagnose convergence of sampling or optimization algorithms because the labels will switch, or be permuted, in different MCMC chains or different optimization runs. Luckily, posterior inferences which do not refer to specific component labels are invariant under label switching and may be used directly. This subsection considers a pair of examples.\n\n\nPosterior predictive distribution for a new observation \\(\\tilde{y}\\) given the complete parameter vector \\(\\theta\\) will be \\[\np(\\tilde{y} \\mid y)\n=\n\\int_{\\theta}\np(\\tilde{y} \\mid \\theta)\n\\, p(\\theta \\mid y)\n\\, \\textsf{d}\\theta.\n\\]\nThe normal mixture example from the previous section, with \\(\\theta =\n(\\mu, \\sigma, \\lambda)\\), shows that the model returns the same density under label switching and thus the predictive inference is sound. In Stan, that predictive inference can be done either by computing \\(p(\\tilde{y} \\mid y)\\), which is more efficient statistically in terms of effective sample size, or simulating draws of \\(\\tilde{y}\\), which is easier to plug into other inferences. Both approaches can be coded directly in the generated quantities block of the program. Here’s an example of the direct (non-sampling) approach.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] y_tilde;\n  // ...\n}\ngenerated quantities {\n  vector[N_tilde] log_p_y_tilde;\n  for (n in 1:N_tilde) {\n    log_p_y_tilde[n]\n      = log_mix(lambda,\n                normal_lpdf(y_tilde[n] | mu[1], sigma[1])\n                normal_lpdf(y_tilde[n] | mu[2], sigma[2]));\n  }\n}\nIt is a bit of a bother afterwards, because the logarithm function isn’t linear and hence doesn’t distribute through averages (Jensen’s inequality shows which way the inequality goes). The right thing to do is to apply log_sum_exp of the posterior draws of log_p_y_tilde. The average log predictive density is then given by subtracting log(N_new).\n\n\n\nOften a mixture model will be applied to a clustering problem and there might be two data items \\(y_i\\) and \\(y_j\\) for which there is a question of whether they arose from the same mixture component. If we take \\(z_i\\) and \\(z_j\\) to be the component responsibility discrete variables, then the quantity of interest is \\(z_i = z_j\\), which can be summarized as an event probability \\[\n\\Pr[z_i = z_j \\mid y]\n=\n\\int_{\\theta}\n\\frac{\\sum_{k=0}^1 p(z_i=k, z_j = k, y_i, y_j \\mid \\theta)}\n     {\\sum_{k=0}^1 \\sum_{m=0}^1 p(z_i = k, z_j = m, y_i, y_j \\mid \\theta)}\n\\,\np(\\theta \\mid y)\n\\,\n\\textsf{d}\\theta.\n\\]\nAs with other event probabilities, this can be calculated in the generated quantities block either by sampling \\(z_i\\) and \\(z_j\\) and using the indicator function on their equality, or by computing the term inside the integral as a generated quantity. As with posterior predictive distribute, working in expectation is more statistically efficient than sampling.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/latent-discrete.html#footnotes",
-    "href": "stan-users-guide/latent-discrete.html#footnotes",
-    "title": "Latent Discrete Parameters",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe computations are similar to those involved in expectation maximization (EM) algorithms (Dempster, Laird, and Rubin 1977).↩︎\nThe source of the data is (Jarrett 1979), which itself is a note correcting an earlier data collection.↩︎\nThe R counterpart, ifelse, is slightly different in that it is typically used in a vectorized situation. The conditional operator is not (yet) vectorized in Stan.↩︎\nAn alternative would be to compute this on the outside and feed it into the Stan model as preprocessed data. Yet another alternative encoding would be a sparse one recording only the capture events along with their time and identifying the individual captured.↩︎\nBoth functions return 0 if the individual represented by the input array was never captured. Individuals with no captures are not relevant for estimating the model because all probability statements are conditional on earlier captures. Typically they would be removed from the data, but the program allows them to be included even though they make not contribution to the log probability function.↩︎\nDiagnostic procedures are often ordinal, as in stages of cancer in oncological diagnosis or the severity of a cavity in dental diagnosis. Dawid and Skene’s model may be used as is or naturally generalized for ordinal ratings using a latent continuous rating and cutpoints as in ordinal logistic regression.↩︎\nIn the subscript, \\(z_i\\) is written as \\(z[i]\\) to improve legibility.↩︎",
+    "objectID": "stan-users-guide/finite-mixtures.html#zero-inflated.section",
+    "href": "stan-users-guide/finite-mixtures.html#zero-inflated.section",
+    "title": "Finite Mixtures",
+    "section": "",
+    "text": "Zero-inflated and hurdle models both provide mixtures of a Poisson and Bernoulli probability mass function to allow more flexibility in modeling the probability of a zero outcome. Zero-inflated models, as defined by Lambert (1992), add additional probability mass to the outcome of zero. Hurdle models, on the other hand, are formulated as pure mixtures of zero and non-zero outcomes.\nZero inflation and hurdle models can be formulated for discrete distributions other than the Poisson. Zero inflation does not work for continuous distributions in Stan because of issues with derivatives; in particular, there is no way to add a point mass to a continuous distribution, such as zero-inflating a normal as a regression coefficient prior. Hurdle models can be formulated as combination of point mass at zero and continuous distribution for positive values.\n\n\nConsider the following example for zero-inflated Poisson distributions. There is a probability \\(\\theta\\) of observing a zero, and a probability \\(1 - \\theta\\) of observing a count with a \\(\\textsf{Poisson}(\\lambda)\\) distribution (now \\(\\theta\\) is being used for mixing proportions because \\(\\lambda\\) is the traditional notation for a Poisson mean parameter). Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as \\[\\begin{align*}\ny_n & = 0 & \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}(y_n \\mid \\lambda) & \\quad\\text{with probability } 1-\\theta.\n\\end{align*}\\]\nStan does not support conditional distribution statements (with ~) conditional on some parameter, and we need to consider the corresponding likelihood \\[\np(y_n \\mid \\theta,\\lambda)\n=\n\\begin{cases}\n\\theta + (1 - \\theta) \\times \\textsf{Poisson}(0 \\mid \\lambda) & \\quad\\text{if } y_n = 0, \\text{ and}\\\\\n(1-\\theta) \\times \\textsf{Poisson}(y_n \\mid \\lambda) &\\quad\\text{if } y_n &gt; 0.\n\\end{cases}\n\\] The log likelihood can be coded directly in Stan (with target +=) as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  for (n in 1:N) {\n    if (y[n] == 0) {\n      target += log_sum_exp(log(theta),\n                            log1m(theta)\n                              + poisson_lpmf(y[n] | lambda));\n    } else {\n      target += log1m(theta)\n                  + poisson_lpmf(y[n] | lambda);\n    }\n  }\n}\nThe log1m(theta) computes log(1-theta), but is more computationally stable. The log_sum_exp(lp1,lp2) function adds the log probabilities on the linear scale; it is defined to be equal to log(exp(lp1) + exp(lp2)), but is more computationally stable and faster.\n\n\nThe code given above to compute the zero-inflated Poisson redundantly calculates all of the Bernoulli terms and also poisson_lpmf(0 | lambda) every time the first condition body executes. The use of the redundant terms is conditioned on y, which is known when the data are read in. This allows the transformed data block to be used to compute some more convenient terms for expressing the log density each iteration.\nThe number of zero cases is computed and handled separately. Then the nonzero cases are collected into their own array for vectorization. The number of zeros is required to declare y_nonzero, so it must be computed in a function.\nfunctions {\n  int num_zeros(array[] int y) {\n    int sum = 0;\n    for (n in 1:size(y)) {\n      sum += (y[n] == 0);\n    }\n    return sum;\n  }\n}\n// ...\ntransformed data {\n  int&lt;lower=0&gt; N_zero = num_zeros(y);\n  array[N - N_zero] int&lt;lower=1&gt; y_nonzero;\n  int N_nonzero = 0;\n  for (n in 1:N) {\n    if (y[n] == 0) continue;\n    N_nonzero += 1;\n    y_nonzero[N_nonzero] = y[n];\n  }\n}\n// ...\nmodel {\n  // ...\n   target\n     += N_zero\n          * log_sum_exp(log(theta),\n                        log1m(theta)\n                          + poisson_lpmf(0 | lambda));\n   target += N_nonzero * log1m(theta);\n   target += poisson_lpmf(y_nonzero | lambda);\n  // ...\n}\nThe boundary conditions of all zeros and no zero outcomes is handled appropriately; in the vectorized case, if y_nonzero is empty, N_nonzero will be zero, and the last two target increment terms will add zeros.\n\n\n\n\nThe hurdle model is similar to the zero-inflated model, but more flexible in that the zero outcomes can be deflated as well as inflated. Given the probability \\(\\theta\\) and the intensity \\(\\lambda\\), the distribution for \\(y_n\\) can be written as [ \\[\\begin{align*}\ny_n & = 0 \\quad\\text{with probability } \\theta, \\text{ and}\\\\\ny_n & \\sim \\textsf{Poisson}_{x\\neq 0}(y_n \\mid \\lambda) \\quad\\text{with probability } 1-\\theta,\n\\end{align*}\\] ] Where \\(\\textsf{Poisson}_{x\\neq 0}\\) is a truncated Poisson distribution, truncated at \\(0\\).\nThe corresponding likelihood function for the hurdle model is defined by \\[\np(y\\mid\\theta,\\lambda)\n=\n\\begin{cases}\n\\theta &\\quad\\text{if } y = 0, \\text{ and}\\\\\n(1 - \\theta)\n   \\frac{\\displaystyle \\textsf{Poisson}(y \\mid \\lambda)}\n        {\\displaystyle  1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)}\n&\\quad\\text{if } y &gt; 0,\n\\end{cases}\n\\] where \\(\\textsf{PoissonCDF}\\) is the cumulative distribution function for the Poisson distribution and and \\(1 - \\textsf{PoissonCDF}(0 \\mid \\lambda)\\) is the relative normalization term for the truncated Poisson (truncated at \\(0\\)).\nThe hurdle model is even more straightforward to program in Stan, as it does not require an explicit mixture.\nif (y[n] == 0) {\n  target += log(theta);\n} else {\n  target += log1m(theta) + poisson_lpmf(y[n] | lambda)\n            - poisson_lccdf(0 | lambda));\n}\nJulian King pointed out that because \\[\\begin{align*}\n\\log \\left( 1 - \\textsf{PoissonCDF}(0 \\mid \\lambda) \\right)\n  &= \\log \\left( 1 - \\textsf{Poisson}(0 \\mid \\lambda) \\right) \\\\\n  &= \\log(1 - \\exp(-\\lambda))\n\\end{align*}\\] the CCDF in the else clause can be replaced with a simpler expression.\ntarget += log1m(theta) + poisson_lpmf(y[n] | lambda)\n          - log1m_exp(-lambda));\nThe resulting code is about 15% faster than the code with the CCDF.\nThis is an example where collecting counts ahead of time can also greatly speed up the execution speed without changing the density. For data size \\(N=200\\) and parameters \\(\\theta=0.3\\) and \\(\\lambda = 8\\), the speedup is a factor of 10; it will be lower for smaller \\(N\\) and greater for larger \\(N\\); it will also be greater for larger \\(\\theta\\).\nTo achieve this speedup, it helps to have a function to count the number of non-zero entries in an array of integers,\nfunctions {\n  int num_zero(array[] int y) {\n    int nz = 0;\n    for (n in 1:size(y)) {\n      if (y[n] == 0) {\n        nz += 1;\n      }\n    }\n    return nz;\n  }\n}\nThen a transformed data block can be used to store the sufficient statistics,\ntransformed data {\n  int&lt;lower=0, upper=N&gt; N0 = num_zero(y);\n  int&lt;lower=0, upper=N&gt; Ngt0 = N - N0;\n  array[N - num_zero(y)] int&lt;lower=1&gt; y_nz;\n  {\n    int pos = 1;\n    for (n in 1:N) {\n      if (y[n] != 0) {\n        y_nz[pos] = y[n];\n        pos += 1;\n      }\n    }\n  }\n}\nThe model block is then reduced to three statements.\nmodel {\n  N0 ~ binomial(N, theta);\n  y_nz ~ poisson(lambda);\n  target += -Ngt0 * log1m_exp(-lambda);\n}\nThe first statement accounts for the Bernoulli contribution to both the zero and non-zero counts. The second line is the Poisson contribution from the non-zero counts, which is now vectorized. Finally, the normalization for the truncation is a single line, so that the expression for the log CCDF at 0 isn’t repeated. Also note that the negation is applied to the constant Ngt0; whenever possible, leave subexpressions constant because then gradients need not be propagated until a non-constant term is encountered.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Latent Discrete Parameters"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/measurement-error.html",
-    "href": "stan-users-guide/measurement-error.html",
-    "title": "Measurement Error and Meta-Analysis",
+    "objectID": "stan-users-guide/finite-mixtures.html#priors-and-effective-data-size-in-mixture-models",
+    "href": "stan-users-guide/finite-mixtures.html#priors-and-effective-data-size-in-mixture-models",
+    "title": "Finite Mixtures",
     "section": "",
-    "text": "Most quantities used in statistical models arise from measurements. Most of these measurements are taken with some error. When the measurement error is small relative to the quantity being measured, its effect on a model is usually small. When measurement error is large relative to the quantity being measured, or when precise relations can be estimated being measured quantities, it is useful to introduce an explicit model of measurement error. One kind of measurement error is rounding.\nMeta-analysis plays out statistically much like measurement error models, where the inferences drawn from multiple data sets are combined to do inference over all of them. Inferences for each data set are treated as providing a kind of measurement error with respect to true parameter values.\n\n\nA Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data (Clayton 1992; Richardson and Gilks 1993). This requires a model of how the measurements are derived from the true values.\n\n\nBefore considering regression with measurement error, first consider a linear regression model where the observed data for \\(N\\) cases includes a predictor \\(x_n\\) and outcome \\(y_n\\). In Stan, a linear regression for \\(y\\) based on \\(x\\) with a slope and intercept is modeled as follows.\ndata {\n  int&lt;lower=0&gt; N;       // number of cases\n  vector[N] x;          // predictor (covariate)\n  vector[N] y;          // outcome (variate)\n}\nparameters {\n  real alpha;           // intercept\n  real beta;            // slope\n  real&lt;lower=0&gt; sigma;  // outcome noise\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ cauchy(0, 5);\n}\nNow suppose that the true values of the predictors \\(x_n\\) are not known, but for each \\(n\\), a measurement \\(x^{\\textrm{meas}}_n\\) of \\(x_n\\) is available. If the error in measurement can be modeled, the measured value \\(x^{\\textrm{meas}}_n\\) can be modeled in terms of the true value \\(x_n\\) plus measurement noise. The true value \\(x_n\\) is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation \\(\\tau\\). This leads to the following regression model with constant measurement error.\ndata {\n  // ...\n  array[N] real x_meas;   // measurement of x\n  real&lt;lower=0&gt; tau;     // measurement noise\n}\nparameters {\n  array[N] real x;    // unknown true value\n  real mu_x;          // prior location\n  real sigma_x;       // prior scale\n  // ...\n}\nmodel {\n  x ~ normal(mu_x, sigma_x);  // prior\n  x_meas ~ normal(x, tau);    // measurement model\n  y ~ normal(alpha + beta * x, sigma);\n  // ...\n}\nThe regression coefficients alpha and beta and regression noise scale sigma are the same as before, but now x is declared as a parameter rather than as data. The data are now x_meas, which is a measurement of the true x value with noise scale tau. The model then specifies that the measurement error for x_meas[n] given true value x[n] is normal with deviation tau. Furthermore, the true values x are given a hierarchical prior here.\nIn cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, Clayton (1992) introduces an exposure model for the unknown (but noisily measured) risk factors \\(x\\) in terms of known (without measurement error) risk factors \\(c\\). A simple model would regress \\(x_n\\) on the covariates \\(c_n\\) with noise term \\(\\upsilon\\), \\[\nx_n \\sim \\textsf{normal}(\\gamma^{\\top}c, \\upsilon).\n\\] This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.\n\n\n\nA common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.\nExercise 3.5(b) by Gelman et al. (2013) provides an example.\n\n3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on \\(\\mu\\) and \\(\\sigma^2\\).\n\nGive the correct posterior distribution for \\((\\mu, \\sigma^2)\\), treating the measurements as rounded.\n\n\nLetting \\(z_n\\) be the unrounded measurement for \\(y_n\\), the problem as stated assumes \\[\nz_n \\sim \\textsf{normal}(\\mu, \\sigma).\n\\]\nThe rounding process entails that \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\)1. The probability mass function for the discrete observation \\(y\\) is then given by marginalizing out the unrounded measurement, producing the likelihood \\[\\begin{align*}\np(y_n \\mid \\mu, \\sigma)\n&= \\int_{y_n - 0.5}^{y_n + 0.5} \\textsf{normal}(z_n \\mid \\mu, \\sigma) \\,\\textsf{d}z_n \\\\\n&= \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n   -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right).\n\\end{align*}\\] Gelman’s answer for this problem took the noninformative prior to be uniform in the variance \\(\\sigma^2\\) on the log scale, but we replace it with more recently recommended half-normal prior on \\(\\sigma\\) \\[\n\\sigma \\sim \\textsf{normal}^+(0, 1).\n\\] The posterior after observing \\(y = (10, 10, 12, 11, 9)\\) can be calculated by Bayes’s rule as \\[\\begin{align*}\np(\\mu, \\sigma \\mid y)\n&\\propto p(\\mu, \\sigma) \\ p(y \\mid \\mu, \\sigma) \\\\\n&\\propto \\textsf{normal}^+(\\sigma \\mid 0, 1)\\prod_{n=1}^5\n  \\left( \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n        -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right)\n  \\right).\n\\end{align*}\\]\nThe Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  for (n in 1:N) {\n    target += log_diff_exp(normal_lcdf(y[n] + 0.5 | mu, sigma),\n                           normal_lcdf(y[n] - 0.5 | mu, sigma));\n  }\n}\nwhere normal_lcdf(y[n]+0.5 | mu, sigma) is equal to log(Phi((y[n] + 0.5 - mu) / sigma)), and log_diff_exp(a, b) computes log(exp(a) - exp(b)) in numerically more stable way.\nAlternatively, the model may be defined with latent parameters for the unrounded measurements \\(z_n\\). The Stan code in this case uses a distribution statement for \\(z_n\\) directly while respecting the constraint \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;lower=y-0.5, upper=y+0.5&gt;[N] z;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  z ~ normal(mu, sigma);\n}\nThis explicit model for the unrounded measurements \\(z\\) produces the same posterior for \\(\\mu\\) and \\(\\sigma\\) as the previous model that marginalizes \\(z\\) out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.\n\n\n\n\nMeta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.\nThe Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.\n\n\nSuppose the data in question arise from a total of \\(M\\) studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen (Warn, Thompson, and Spiegelhalter 2002) or mortality after myocardial infarction under a treatment of beta blockers (Gelman et al. 2013, sec. 5.6).\n\n\nThe clinical data consists of \\(J\\) trials, each with \\(n^t\\) treatment cases, \\(n^c\\) control cases, \\(r^t\\) successful outcomes among those treated and \\(r^c\\) successful outcomes among those in the control group. This data can be declared in Stan as follows.2\ndata {\n  int&lt;lower=0&gt; J;\n  array[J] int&lt;lower=0&gt; n_t;  // num cases, treatment\n  array[J] int&lt;lower=0&gt; r_t;  // num successes, treatment\n  array[J] int&lt;lower=0&gt; n_c;  // num cases, control\n  array[J] int&lt;lower=0&gt; r_c;  // num successes, control\n}\n\n\n\nAlthough the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio \\[\\begin{align*}\ny_j &= \\log \\left( \\frac{r^t_j / (n^t_j - r^t_j)}\n                        {r^c_j / (n^c_j - r^c_j)}\n        \\right) \\\\\n&= \\log \\left( \\frac{r^t_j}{n^t_j - r^t_j} \\right)\n   -\\log \\left( \\frac{r^c_j}{n^c_j - r^c_j} \\right)\n\\end{align*}\\] and corresponding standard errors \\[\n\\sigma_j = \\sqrt{\n  \\frac{1}{r^T_j}\n+ \\frac{1}{n^T_j - r^T_j}\n+ \\frac{1}{r^C_j}\n+ \\frac{1}{n^C_j - r^C_j}\n}.\n\\]\nThe log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.3\ntransformed data {\n  array[J] real y;\n  array[J] real&lt;lower=0&gt; sigma;\n  for (j in 1:J) {\n    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])\n            - (log(r_c[j]) - log(n_c[j] - r_c[j]));\n  }\n  for (j in 1:J) {\n    sigma[j] = sqrt(1 / r_t[j] + 1 / (n_t[j] - r_t[j])\n                     + 1 / r_c[j] + 1 / (n_c[j] - r_c[j]));\n  }\n}\nThis definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.\n\n\n\nWith the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.\nparameters {\n  real theta;  // global treatment effect, log odds\n}\nmodel {\n  y ~ normal(theta, sigma);\n}\nThe distribution statement for y is vectorized; it has the same effect as the following.\n  for (j in 1:J) {\n    y[j] ~ normal(theta, sigma[j]);\n  }\nIt is common to include a prior for theta in this model, but it is not strictly necessary for the model to be proper because y is fixed and \\(\\textsf{normal}(y \\mid \\mu,\\sigma) =\n\\textsf{normal}(\\mu \\mid y,\\sigma)\\).\n\n\n\nTo model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.\nparameters {\n  array[J] real theta;  // per-trial treatment effect\n  real mu;              // mean treatment effect\n  real&lt;lower=0&gt; tau;    // deviation of treatment effects\n}\nmodel {\n  y ~ normal(theta, sigma);\n  theta ~ normal(mu, tau);\n  mu ~ normal(0, 10);\n  tau ~ cauchy(0, 5);\n}\nAlthough the vectorized distribution statement for y appears unchanged, the parameter theta is now a vector. The distribution statement for theta is also vectorized, with the hyperparameters mu and tau themselves being given wide priors compared to the scale of the data.\nRubin (1981) provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.\n\n\n\nSmith, Spiegelhalter, and Thomas (1995) and Gelman et al. (2013, sec. 19.4) provide meta-analyses based directly on binomial data. Warn, Thompson, and Spiegelhalter (2002) consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.\nIf trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects \\(\\theta_j\\).",
+    "text": "Suppose we have a two-component mixture model with mixing rate \\(\\lambda \\in (0, 1)\\). Because the likelihood for the mixture components is proportionally weighted by the mixture weights, the effective data size used to estimate each of the mixture components will also be weighted as a fraction of the overall data size. Thus although there are \\(N\\) observations, the mixture components will be estimated with effective data sizes of \\(\\theta \\, N\\) and \\((1 - \\theta)\n\\, N\\) for the two components for some \\(\\theta \\in (0, 1)\\). The effective weighting size is determined by posterior responsibility, not simply by the mixing rate \\(\\lambda\\).\n\n\nIn contrast to mixture models, which create mixtures at the observation level, model averaging creates mixtures over the posteriors of models separately fit with the entire data set. In this situation, the priors work as expected when fitting the models independently, with the posteriors being based on the complete observed data \\(y\\).\nIf different models are expected to account for different observations, we recommend building mixture models directly. If the models being mixed are similar, often a single expanded model will capture the features of both and may be used on its own for inferential purposes (estimation, decision making, prediction, etc.). For example, rather than fitting an intercept-only regression and a slope-only regression and averaging their predictions, even as a mixture model, we would recommend building a single regression with both a slope and an intercept. Model complexity, such as having more predictors than data points, can be tamed using appropriately regularizing priors. If computation becomes a bottleneck, the only recourse can be model averaging, which can be calculated after fitting each model independently (see Hoeting et al. (1999) and Gelman et al. (2013) for theoretical and computational details).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Measurement Error and Meta-Analysis"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/measurement-error.html#bayesian-measurement-error-model",
-    "href": "stan-users-guide/measurement-error.html#bayesian-measurement-error-model",
-    "title": "Measurement Error and Meta-Analysis",
-    "section": "",
-    "text": "A Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data (Clayton 1992; Richardson and Gilks 1993). This requires a model of how the measurements are derived from the true values.\n\n\nBefore considering regression with measurement error, first consider a linear regression model where the observed data for \\(N\\) cases includes a predictor \\(x_n\\) and outcome \\(y_n\\). In Stan, a linear regression for \\(y\\) based on \\(x\\) with a slope and intercept is modeled as follows.\ndata {\n  int&lt;lower=0&gt; N;       // number of cases\n  vector[N] x;          // predictor (covariate)\n  vector[N] y;          // outcome (variate)\n}\nparameters {\n  real alpha;           // intercept\n  real beta;            // slope\n  real&lt;lower=0&gt; sigma;  // outcome noise\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ cauchy(0, 5);\n}\nNow suppose that the true values of the predictors \\(x_n\\) are not known, but for each \\(n\\), a measurement \\(x^{\\textrm{meas}}_n\\) of \\(x_n\\) is available. If the error in measurement can be modeled, the measured value \\(x^{\\textrm{meas}}_n\\) can be modeled in terms of the true value \\(x_n\\) plus measurement noise. The true value \\(x_n\\) is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation \\(\\tau\\). This leads to the following regression model with constant measurement error.\ndata {\n  // ...\n  array[N] real x_meas;   // measurement of x\n  real&lt;lower=0&gt; tau;     // measurement noise\n}\nparameters {\n  array[N] real x;    // unknown true value\n  real mu_x;          // prior location\n  real sigma_x;       // prior scale\n  // ...\n}\nmodel {\n  x ~ normal(mu_x, sigma_x);  // prior\n  x_meas ~ normal(x, tau);    // measurement model\n  y ~ normal(alpha + beta * x, sigma);\n  // ...\n}\nThe regression coefficients alpha and beta and regression noise scale sigma are the same as before, but now x is declared as a parameter rather than as data. The data are now x_meas, which is a measurement of the true x value with noise scale tau. The model then specifies that the measurement error for x_meas[n] given true value x[n] is normal with deviation tau. Furthermore, the true values x are given a hierarchical prior here.\nIn cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, Clayton (1992) introduces an exposure model for the unknown (but noisily measured) risk factors \\(x\\) in terms of known (without measurement error) risk factors \\(c\\). A simple model would regress \\(x_n\\) on the covariates \\(c_n\\) with noise term \\(\\upsilon\\), \\[\nx_n \\sim \\textsf{normal}(\\gamma^{\\top}c, \\upsilon).\n\\] This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.\n\n\n\nA common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.\nExercise 3.5(b) by Gelman et al. (2013) provides an example.\n\n3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on \\(\\mu\\) and \\(\\sigma^2\\).\n\nGive the correct posterior distribution for \\((\\mu, \\sigma^2)\\), treating the measurements as rounded.\n\n\nLetting \\(z_n\\) be the unrounded measurement for \\(y_n\\), the problem as stated assumes \\[\nz_n \\sim \\textsf{normal}(\\mu, \\sigma).\n\\]\nThe rounding process entails that \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\)1. The probability mass function for the discrete observation \\(y\\) is then given by marginalizing out the unrounded measurement, producing the likelihood \\[\\begin{align*}\np(y_n \\mid \\mu, \\sigma)\n&= \\int_{y_n - 0.5}^{y_n + 0.5} \\textsf{normal}(z_n \\mid \\mu, \\sigma) \\,\\textsf{d}z_n \\\\\n&= \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n   -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right).\n\\end{align*}\\] Gelman’s answer for this problem took the noninformative prior to be uniform in the variance \\(\\sigma^2\\) on the log scale, but we replace it with more recently recommended half-normal prior on \\(\\sigma\\) \\[\n\\sigma \\sim \\textsf{normal}^+(0, 1).\n\\] The posterior after observing \\(y = (10, 10, 12, 11, 9)\\) can be calculated by Bayes’s rule as \\[\\begin{align*}\np(\\mu, \\sigma \\mid y)\n&\\propto p(\\mu, \\sigma) \\ p(y \\mid \\mu, \\sigma) \\\\\n&\\propto \\textsf{normal}^+(\\sigma \\mid 0, 1)\\prod_{n=1}^5\n  \\left( \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n        -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right)\n  \\right).\n\\end{align*}\\]\nThe Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  for (n in 1:N) {\n    target += log_diff_exp(normal_lcdf(y[n] + 0.5 | mu, sigma),\n                           normal_lcdf(y[n] - 0.5 | mu, sigma));\n  }\n}\nwhere normal_lcdf(y[n]+0.5 | mu, sigma) is equal to log(Phi((y[n] + 0.5 - mu) / sigma)), and log_diff_exp(a, b) computes log(exp(a) - exp(b)) in numerically more stable way.\nAlternatively, the model may be defined with latent parameters for the unrounded measurements \\(z_n\\). The Stan code in this case uses a distribution statement for \\(z_n\\) directly while respecting the constraint \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;lower=y-0.5, upper=y+0.5&gt;[N] z;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  z ~ normal(mu, sigma);\n}\nThis explicit model for the unrounded measurements \\(z\\) produces the same posterior for \\(\\mu\\) and \\(\\sigma\\) as the previous model that marginalizes \\(z\\) out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.",
+    "objectID": "stan-users-guide/finite-mixtures.html#footnotes",
+    "href": "stan-users-guide/finite-mixtures.html#footnotes",
+    "title": "Finite Mixtures",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nImposing a constraint such as \\(\\theta &lt; 0.5\\) will resolve the symmetry, but fundamentally changes the model and its posterior inferences.↩︎",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Measurement Error and Meta-Analysis"
+      "Finite Mixtures"
     ]
   },
   {
-    "objectID": "stan-users-guide/measurement-error.html#meta-analysis",
-    "href": "stan-users-guide/measurement-error.html#meta-analysis",
-    "title": "Measurement Error and Meta-Analysis",
+    "objectID": "stan-users-guide/for-bugs-users.html",
+    "href": "stan-users-guide/for-bugs-users.html",
+    "title": "Transitioning from BUGS",
     "section": "",
-    "text": "Meta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.\nThe Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.\n\n\nSuppose the data in question arise from a total of \\(M\\) studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen (Warn, Thompson, and Spiegelhalter 2002) or mortality after myocardial infarction under a treatment of beta blockers (Gelman et al. 2013, sec. 5.6).\n\n\nThe clinical data consists of \\(J\\) trials, each with \\(n^t\\) treatment cases, \\(n^c\\) control cases, \\(r^t\\) successful outcomes among those treated and \\(r^c\\) successful outcomes among those in the control group. This data can be declared in Stan as follows.2\ndata {\n  int&lt;lower=0&gt; J;\n  array[J] int&lt;lower=0&gt; n_t;  // num cases, treatment\n  array[J] int&lt;lower=0&gt; r_t;  // num successes, treatment\n  array[J] int&lt;lower=0&gt; n_c;  // num cases, control\n  array[J] int&lt;lower=0&gt; r_c;  // num successes, control\n}\n\n\n\nAlthough the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio \\[\\begin{align*}\ny_j &= \\log \\left( \\frac{r^t_j / (n^t_j - r^t_j)}\n                        {r^c_j / (n^c_j - r^c_j)}\n        \\right) \\\\\n&= \\log \\left( \\frac{r^t_j}{n^t_j - r^t_j} \\right)\n   -\\log \\left( \\frac{r^c_j}{n^c_j - r^c_j} \\right)\n\\end{align*}\\] and corresponding standard errors \\[\n\\sigma_j = \\sqrt{\n  \\frac{1}{r^T_j}\n+ \\frac{1}{n^T_j - r^T_j}\n+ \\frac{1}{r^C_j}\n+ \\frac{1}{n^C_j - r^C_j}\n}.\n\\]\nThe log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.3\ntransformed data {\n  array[J] real y;\n  array[J] real&lt;lower=0&gt; sigma;\n  for (j in 1:J) {\n    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])\n            - (log(r_c[j]) - log(n_c[j] - r_c[j]));\n  }\n  for (j in 1:J) {\n    sigma[j] = sqrt(1 / r_t[j] + 1 / (n_t[j] - r_t[j])\n                     + 1 / r_c[j] + 1 / (n_c[j] - r_c[j]));\n  }\n}\nThis definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.\n\n\n\nWith the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.\nparameters {\n  real theta;  // global treatment effect, log odds\n}\nmodel {\n  y ~ normal(theta, sigma);\n}\nThe distribution statement for y is vectorized; it has the same effect as the following.\n  for (j in 1:J) {\n    y[j] ~ normal(theta, sigma[j]);\n  }\nIt is common to include a prior for theta in this model, but it is not strictly necessary for the model to be proper because y is fixed and \\(\\textsf{normal}(y \\mid \\mu,\\sigma) =\n\\textsf{normal}(\\mu \\mid y,\\sigma)\\).\n\n\n\nTo model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.\nparameters {\n  array[J] real theta;  // per-trial treatment effect\n  real mu;              // mean treatment effect\n  real&lt;lower=0&gt; tau;    // deviation of treatment effects\n}\nmodel {\n  y ~ normal(theta, sigma);\n  theta ~ normal(mu, tau);\n  mu ~ normal(0, 10);\n  tau ~ cauchy(0, 5);\n}\nAlthough the vectorized distribution statement for y appears unchanged, the parameter theta is now a vector. The distribution statement for theta is also vectorized, with the hyperparameters mu and tau themselves being given wide priors compared to the scale of the data.\nRubin (1981) provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.\n\n\n\nSmith, Spiegelhalter, and Thomas (1995) and Gelman et al. (2013, sec. 19.4) provide meta-analyses based directly on binomial data. Warn, Thompson, and Spiegelhalter (2002) consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.\nIf trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects \\(\\theta_j\\).",
+    "text": "From the outside, Stan and BUGS1 are similar—they use statistically-themed modeling languages (which are similar but with some differences; see below), they can be called from R, running some specified number of chains to some specified length, producing posterior simulations that can be assessed using standard convergence diagnostics. This is not a coincidence: in designing Stan: we wanted to keep many of the useful features of Bugs.\n\n\n\n\nStan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).\n\n\n\nBUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.\n\n\n\nBUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).\n\n\n\nThe BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).\n\n\n\nIn BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets mu before using it to sample y:\nmu = a + b * x;\ny ~ normal(mu, sigma);\ntranslates to the following C++ code:\nmu = a + b * x;\ntarget += normal_lpdf(y | mu, sigma);\nContrast this with the following Stan program:\ny ~ normal(mu, sigma);\nmu = a + b * x;\nThis program is well formed, but is almost certainly a coding error, because it attempts to use mu before it is set. The direct translation to C++ code highlights the potential error of using mu in the first statement:\ntarget += normal_lpdf(y | mu, sigma);\nmu = a + b * x;\nTo trap these kinds of errors, variables are initialized to the special not-a-number (NaN) value. If NaN is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.\n\n\n\nStan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.\n\n\n\nBoth BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (\\(\\hat{R}\\)) and the effective sample size to judge when to stop (stopping when \\(\\hat{R}\\) values do not counter-indicate convergence and when high enough effective sample size have been obtained).\n\n\n\nWinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.\n\n\n\nLike WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.\n\n\n\nLike OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.\n\n\n\n\nThe BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:\ny ~ normal(mu, sigma);\nand\nfor (i in 1:n) y[i] ~ normal(mu, sigma);\nOr, equivalently (recall that a line break is just another form of whitespace),\nfor (i in 1:n)\n  y[i] ~ normal(mu, sigma);\nand also equivalently,\nfor (i in 1:n) {\n  y[i] ~ normal(mu, sigma);\n}\nThere’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.\nIn Stan, variables can have names constructed using letters, numbers, and the underscore (_) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (.) symbol.\nIn Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in Bayesian Data Analysis) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is normal(1,2), not normal(1,4) or normal(1,0.25).\nSimilarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.\nThe distributions have slightly different names:\n\n\n\nBUGS\nStan\n\n\n\n\ndnorm\nnormal\n\n\ndbinom\nbinomial\n\n\ndpois\npoisson\n\n\n…\n…\n\n\n\nStan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.\n{\n  total = 0;\n  for (i in 1:n) {\n    theta[i] ~ normal(total, sigma);\n    total = total + theta[i];\n  }\n}\nIn BUGS, the above model would not be legal because the variable total is defined more than once. But in Stan, the loop is executed in order, so total is overwritten in each step.\nStan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.\nIn Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.\nStan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,2 allowing\nfor (i in 1:n) {\n  y[i] ~ normal(mu[i], sigma[i]);\n}\nto be expressed more compactly as\ny ~ normal(mu, sigma);\nThe vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.\nStan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of K parameters for each of J groups; this can be declared using\narray[J] vector[K] theta;\nThen theta[j] is an expression denoting a K-vector and may be used in the code just like any other vector variable.\nAn alternative encoding would be with a two-dimensional array, as in\narray[J, K] real theta;\nThe vector version can have some advantages, both in convenience and in computational speed for some operations.\nA third encoding would use a matrix:\nmatrix[J, K] theta;\nbut in this case, theta[j] is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in theta[j]', may be used to convert the row vector theta[j] to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.\nStan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the zero inflation section.\nStan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop\nmodel {\n    // ...\n    for (n in 1:N) {\n        // ... do something with n ....\n    }\n}\nmay be recoded using the following while loop.\nmodel {\n    int n;\n    // ...\n    n = 1;\n    while (n &lt;= N) {\n        // ... do something with n ...\n        n = n + 1;\n    }\n}\n\n\n\nStan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.\nStan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in Lewandowski, Kurowicka, and Joe (2009). In particular, the Lewandowski et al. prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.\nIn BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter p declared as\nreal&lt;lower=0, upper=1&gt; p;\nand then have no distribution statement for p in the model block, then you are implicitly assigning a uniform \\([0,1]\\) prior on p.\nOn the other hand, if you have a parameter theta declared with\nreal theta;\nand have no distribution statement for theta in the model block, then you are implicitly assigning an improper uniform prior on \\((-\\infty,\\infty)\\) to theta.\nBUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:\nparameters {\n  real theta;\n}\nmodel { }\nAlthough parameters in Stan models may have improper priors, we do not want improper posterior distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.\nEach statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:\ny ~ normal(0,1);\ny ~ normal(2,3);\nThe second line here does not simply overwrite the first; rather, both statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, \\(\\textsf{normal}(y \\mid 0,1) * \\textsf{normal}(y \\mid 2,3)\\), into the density function.\nFor a perhaps more confusing example, consider the following two lines in a Stan model:\nx ~ normal(0.8 * y, sigma);\ny ~ normal(0.8 * x, sigma);\nAt first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are not interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, \\(\\textsf{normal}(x \\mid 0.8y,\\sigma)\n\\times \\textsf{normal}(y \\mid 0.8x,\\sigma)\\), which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.\nWith censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in Gelman et al. (2013) and Gelman and Hill (2007) may be implemented in Stan directly as written.\nStan, like BUGS, can benefit from human intervention in the form of reparameterization.\n\n\n\nStan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.\nIn practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.\nWhen you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)\nWhen running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable x from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.\nAs in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using plot() and print(). You can access the simulations themselves using various extractor functions, as described in the RStan documentation.\nVarious information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.\n\n\n\nStan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the Stan web site for information on subscribing and posting and to look at archives.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Measurement Error and Meta-Analysis"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/measurement-error.html#footnotes",
-    "href": "stan-users-guide/measurement-error.html#footnotes",
-    "title": "Measurement Error and Meta-Analysis",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThere are several different rounding rules (see, e.g., Wikipedia: Rounding), which affect which interval ends are open and which are closed, but these do not matter here as for continuous \\(z_n\\) \\(p(z_n=y_n-0.5)=p(z_n=y_n+0.5)=0\\).↩︎\nStan’s integer constraints are not powerful enough to express the constraint that \\(\\texttt{r}\\mathtt{\\_}\\texttt{t[j]} \\leq \\texttt{n}\\mathtt{\\_}\\texttt{t[j]}\\), but this constraint could be checked in the transformed data block.↩︎\nWhen dividing two integers, the result type is an integer and rounding will ensue if the result is not exact. See the discussion of primitive arithmetic types in the reference manual for more information.↩︎",
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-how-bugs-and-stan-work",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-how-bugs-and-stan-work",
+    "title": "Transitioning from BUGS",
+    "section": "",
+    "text": "Stan is compiled in two steps, first a model is translated to templated C++ and then to a platform-specific executable. Stan, unlike BUGS, allows the user to directly program in C++, but we do not describe how to do this in this Stan manual (see the getting started with C++ section of https://mc-stan.org for more information on using Stan directly from C++).\n\n\n\nBUGS performs MCMC updating one scalar parameter at a time, (with some exceptions such as JAGS’s implementation of regression and generalized linear models and some conjugate multivariate parameters), using conditional distributions (Gibbs sampling) where possible and otherwise using adaptive rejection sampling, slice sampling, and Metropolis jumping. BUGS figures out the dependence structure of the joint distribution as specified in its modeling language and uses this information to compute only what it needs at each step. Stan moves in the entire space of all the parameters using Hamiltonian Monte Carlo (more precisely, the no-U-turn sampler), thus avoiding some difficulties that occur with one-dimension-at-a-time sampling in high dimensions but at the cost of requiring the computation of the entire log density at each step.\n\n\n\nBUGS tunes its adaptive jumping (if necessary) during its warmup phase (traditionally referred to as “burn-in”). Stan uses its warmup phase to tune the no-U-turn sampler (NUTS).\n\n\n\nThe BUGS modeling language is not directly executable. Rather, BUGS parses its model to determine the posterior density and then decides on a sampling scheme. In contrast, the statements in a Stan model are directly executable: they translate exactly into C++ code that is used to compute the log posterior density (which in turn is used to compute the gradient).\n\n\n\nIn BUGS, statements are executed according to the directed graphical model so that variables are always defined when needed. A side effect of the direct execution of Stan’s modeling language is that statements execute in the order in which they are written. For instance, the following Stan program, which sets mu before using it to sample y:\nmu = a + b * x;\ny ~ normal(mu, sigma);\ntranslates to the following C++ code:\nmu = a + b * x;\ntarget += normal_lpdf(y | mu, sigma);\nContrast this with the following Stan program:\ny ~ normal(mu, sigma);\nmu = a + b * x;\nThis program is well formed, but is almost certainly a coding error, because it attempts to use mu before it is set. The direct translation to C++ code highlights the potential error of using mu in the first statement:\ntarget += normal_lpdf(y | mu, sigma);\nmu = a + b * x;\nTo trap these kinds of errors, variables are initialized to the special not-a-number (NaN) value. If NaN is passed to a log probability function, it will raise a domain exception, which will in turn be reported by the sampler. The sampler will reject the sample out of hand as if it had zero probability.\n\n\n\nStan uses its own C++ algorithmic differentiation packages to compute the gradient of the log density (up to a proportion). Gradients are required during the Hamiltonian dynamics simulations within the leapfrog algorithm of the Hamiltonian Monte Carlo and NUTS samplers.\n\n\n\nBoth BUGS and Stan are semi-automatic in that they run by themselves with no outside tuning required. Nevertheless, the user needs to pick the number of chains and number of iterations per chain. We usually pick 4 chains and start with 10 iterations per chain (to make sure there are no major bugs and to approximately check the timing), then go to 100, 1000, or more iterations as necessary. Compared to Gibbs or Metropolis, Hamiltonian Monte Carlo can take longer per iteration (as it typically takes many “leapfrog steps” within each iteration), but the iterations typically have lower autocorrelation. So Stan might work fine with 1000 iterations in an example where BUGS would require 100,000 for good mixing. We recommend monitoring potential scale reduction statistics (\\(\\hat{R}\\)) and the effective sample size to judge when to stop (stopping when \\(\\hat{R}\\) values do not counter-indicate convergence and when high enough effective sample size have been obtained).\n\n\n\nWinBUGS is closed source. OpenBUGS and JAGS are both licensed under the Gnu Public License (GPL), otherwise known as copyleft due to the restrictions it places on derivative works. Stan is licensed under the much more liberal new BSD license.\n\n\n\nLike WinBUGS, OpenBUGS and JAGS, Stan can be run directly from the command line or through common analytics platforms like R, Python, Julia, MATLAB, Mathematica, and the command line.\n\n\n\nLike OpenBUGS and JAGS, Stan can be run on Linux, Mac, and Windows platforms.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Measurement Error and Meta-Analysis"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/multi-indexing.html",
-    "href": "stan-users-guide/multi-indexing.html",
-    "title": "Multiple Indexing and Range Indexing",
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-the-modeling-languages",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-the-modeling-languages",
+    "title": "Transitioning from BUGS",
     "section": "",
-    "text": "Stan allows multiple indexes to be provided for containers (i.e., arrays, vectors, and matrices) in a single position, using either an array of integer indexes or range bounds. In many cases, there are functions that provide similar behavior.\nAllowing multiple indexes supports inline vectorization of models. For instance, consider the data model for a varying-slope, varying-intercept hierarchical linear regression, which could be coded as\nfor (n in 1:N) {\n  y[n] ~ normal(alpha[ii[n]] + beta[ii[n]] * x[n], sigma);\n}\nWith multiple indexing, this can be coded in one line, leading to more efficient vectorized code.\ny ~ normal(alpha[ii] + rows_dot_product(beta[ii], x), sigma);\nThis latter version is faster than the loop version; it is equivalent in speed to the clunky assignment to a local variable.\n{\n  vector[N] mu;\n  for (n in 1:N) {\n    mu[n] = alpha[ii[n]] + beta[ii[n]] * x[n];\n  }\n  y ~ normal(mu, sigma);\n}\nThe boost in speed compared to the original version is because the single call to the normal log density in the distribution statement will be much more memory efficient than the original version.\n\n\nThe following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.\narray[3] int c;\n// ... define: c == (5, 9, 7)\narray[4] int idxs;\n// ... define: idxs == (3, 3, 1, 2)\narray[4] int d;\nd = c[idxs];    // result: d == (7, 7, 5, 9)\nIn general, the multiple indexed expression c[idxs] is defined as follows, assuming idxs is of size K.\nc[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )\nThus c[idxs] is of the same size as idxs, which is K in this example.\nMultiple indexing can also be used with multi-dimensional arrays. For example, consider the following.\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[4] int idxs;\n// ... define: idxs = (2, 2, 1, 2)\narray[4, 3] int d\nd = c[idxs];    // result: d = ((7, 11, 13), (7, 11, 13),\n                //              (1, 3, 5), (7, 11, 13))\nThat is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by c[idxs][j] ==   c[idxs[j]].\nMultiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.\narray[4] int e;\ne = c[2, idxs]; // result:  c[2] = (7, 11, 13)\n                // result:  e = (11, 11, 7, 11)\nThe single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, c[2,idxs] evaluates to the same value as c[2][idxs].\nMultiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[3] int idxs1;\n// ... define: idxs1 = (2, 2, 1)\narray[2] int idxs2;\n// ... define: idxs2 = (1, 3)\narray[3, 2] int d;\nd = c[idxs1, idxs2];  // result: d = ((7, 13), (7, 13), (1, 5))\nWith multiple indexes, we no longer have c[idxs1, idxs2] being the same as c[idxs1][idxs2]. Rather, the entry d[i, j] after executing the above is given by\nd[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]\nThis example illustrates the operation of multiple indexing in the general case: a multiple index like idxs1 converts an index i used on the result (here, c[idxs1, idxs2]) to index idxs1[i] in the variable being indexed (here, c). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.\n\n\n\nSlicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.\n\n\nFor instance, consider supplying an upper and lower bound for an index.\narray[7] int c;\n// ...\narray[4] int d;\nd = c[3:6];  // result: d == (c[3], c[4], c[5], c[6])\nThe range index 3:6 behaves semantically just like the multiple index (3, 4, 5, 6). In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.\n\n\n\nIt is also possible to supply just a lower bound, or just an upper bound. Writing c[3:] is just shorthand for c[3:size(c)]. Writing c[:5] is just shorthand for c[1:5].\n\n\n\nFinally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (:) as the index or leaving the index position empty. In both cases, c[] and c[:] are equal to c[1:size(c)], which in turn is just equal to c.\n\n\n\nStan provides head and tail functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,\nvector[M] a = ...;\nvector[N] b = head(a, N);\nassigns b to be a vector equivalent to the first N elements of the vector a. The function tail works the same way for suffixes, with\narray[M] a = ...;\narray[N] b = tail(a, N);\nFinally, there is a segment function, which specifies a first element and number of elements. For example,\narray[15] a = ...;\narray[3] b = segment(a, 5, 3);\nwill set b to be equal to { a[5], a[6], a[7] }, so that it starts at element 5 of a and includes a total of 3 elements.\n\n\n\n\nMultiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.\narray[3] int a;\narray[2] int c;\narray[2] int idxs;\n// ... define: a == (1, 2, 3);  c == (5, 9)\n               //         idxs = (3,2)\na[idxs] = c;   // result: a == (1, 9, 5)\nThe result above can be worked out by noting that the assignment sets a[idxs[1]] (a[3]) to c[1] (5) and a[idxs[2]] (a[2]) to c[2] (9).\nThe same principle applies when there are many multiple indexes, as in the following example.\narray[5, 7] int a;\narray[2, 2] int c;\n// ...\na[2:3, 5:6] = c;  // result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]\n                  //         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]\nAs in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.\nUsage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.\narray[10, 13] int a;\narray[2] int c;\n// ...\na[4, 2:3] = c;  // result:  a[4, 2] == c[1];  a[4, 3] == c[2]\n\n\nAliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array a in the following code fragment.\narray[3] int a;\n// ... define: a == (5, 6, 7)\na[2:3] = a[1:2];\n// ... result: a == (5, 5, 6)\nThe reason the value of a after the assignment is \\((5,5,6)\\) rather than \\((5,5,5)\\) is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.\narray[3] int a;\narray[3] int idxs;\n// ... define idxs = (2, 1, 3)\na[idxs] = a;\nIn this case, it is evident why the right-hand side needs to be copied before the assignment.\nIt is tempting (but wrong) to think of the assignment a[2:3] =   a[1:2] as executing the following assignments.\n// ... define: a = (5, 6, 7)\na[2] = a[1];      // result: a = (5, 5, 7)\na[3] = a[2];      // result: a = (5, 5, 5)!\nThis produces a different result than executing the assignment because a[2]’s value changes before it is used.\n\n\n\n\nMultiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.\n\n\nVectors and row vectors behave exactly the same way as arrays with multiple indexes. If v is a vector, then v[3] is a scalar real value, whereas v[2:4] is a vector of size 3 containing the elements v[2], v[3], and v[4].\nThe only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.\narray[3] vector[5] v;\narray[7] int idxs;\n// ...\nvector[7] u;\nu = v[2, idxs];\n\narray[7] real w;\nw = v[idxs, 2];\nThe key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because v is an array of vectors, v[2, idxs] reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, v[idxs, 2] does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for w is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.\n\n\n\nMatrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.\nmatrix[5, 7] m;\n// ...\nrow_vector[3] rv;\nrv = m[4, 3:5];    // result is 1 x 3\n// ...\nvector[4] v;\nv = m[2:5, 3];     // result is 3 x 1\n// ...\nmatrix[3, 4] m2;\nm2 = m[1:3, 2:5];  // result is 3 x 4\nThe key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.\n\n\n\nIf matrices receive a single multiple index, the result is a matrix. So if m is a matrix, so is m[2:4]. In contrast, supplying a single index, m[3], produces a row vector result. That is, m[3] produces the same result as m[3, ] or m[3, 1:cols(m)].\n\n\n\nWith arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.\narray[5, 7] matrix[3, 4] m;\n// ...\narray[2] matrix[3, 4] a;\na = m[1, 2:3];  // knock off first array dimension\na = m[3:4, 5];  // knock off second array dimension\nIn both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, a[i] ==   m[1, i + 1], whereas in the second case, a[i] == m[i + 2,   5].\nContinuing the previous example, consider the following.\n// ...\nvector[2] b;\nb = a[1, 3, 2:3, 2];\nHere, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, b[j] == a[1, 3, 1 + j, 2].\nThis last example illustrates an important point: if there is a lower-bounded index, such as 2:3, with lower bound 2, then the lower bound minus one is added to the index, as seen in the 1 + j expression above.\nContinuing further, consider continuing with the following.\n// ...\narray[2] row_vector[3] c;\nc = a[4:5, 3, 1, 2: ];\nHere, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by c[i, j] == a[i + 3, 3, 1, j + 1]\n\n\n\nMatrix slicing can also be performed using the block function. For example,\nmatrix[20, 20] a = ...;\nmatrix[3, 2] b = block(a, 5, 9, 3, 2);\nwill set b equal to the submatrix of a starting at index [5, 9] and extending 3 rows and 2 columns. Thus block(a, 5, 9, 3, 2) is equivalent to b[5:7, 9:10].\nThe sub_col function extracts a slice of a column of a matrix as a vector. For example,\nmatrix[10, 10] a = ...;\nvector b = sub_col(a, 2, 3, 5);\nwill set b equal to the vector a[2:6, 3], taking the element starting at [2, 3], then extending for a total of 5 rows. The function sub_row works the same way for extracting a slice of a row as a row vector. For example, sub_row(a, 2, 3, 5) is equal to the row vector a[2, 3:7], which also starts at position [2, 3] then extends for a total of 5 columns.\n\n\n\n\nSuppose you have a \\(3 \\times 3\\) matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.\nSuppose a \\(3 \\times 3\\) matrix is known to be zero at indexes \\([1,2]\\) and \\([1,3]\\). The indexes for parameters are included in a “melted” data-frame or database format.\ntransformed data {\n  array[7, 2] int&lt;lower=1, upper=3&gt; idxs\n    = { {1, 1},\n        {2, 1}, {2, 2}, {2, 3},\n        {3, 1}, {3, 2}, {3, 3} };\n  // ...\nThe seven remaining parameters are declared as a vector.\nparameters {\n  vector[7] A_raw;\n  // ...\n}\nThen the full matrix A is constructed in the model block as a local variable.\nmodel {\n  matrix[3, 3] A;\n  for (i in 1:7) {\n    A[idxs[i, 1], idxs[i, 2]] = A_raw[i];\n  }\n  A[1, 2] = 0;\n  A[1, 3] = 0;\n  // ...\n}\nThis may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the idxs array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.",
+    "text": "The BUGS modeling language follows an R-like syntax in which line breaks are meaningful. Stan follows the rules of C, in which line breaks are equivalent to spaces, and each statement ends in a semicolon. For example:\ny ~ normal(mu, sigma);\nand\nfor (i in 1:n) y[i] ~ normal(mu, sigma);\nOr, equivalently (recall that a line break is just another form of whitespace),\nfor (i in 1:n)\n  y[i] ~ normal(mu, sigma);\nand also equivalently,\nfor (i in 1:n) {\n  y[i] ~ normal(mu, sigma);\n}\nThere’s a semicolon after the model statement but not after the brackets indicating the body of the for loop.\nIn Stan, variables can have names constructed using letters, numbers, and the underscore (_) symbol, but nothing else (and a variable name cannot begin with a number). BUGS variables can also include the dot, or period (.) symbol.\nIn Stan, the second argument to the “normal” function is the standard deviation (i.e., the scale), not the variance (as in Bayesian Data Analysis) and not the inverse-variance (i.e., precision) (as in BUGS). Thus a normal with mean 1 and standard deviation 2 is normal(1,2), not normal(1,4) or normal(1,0.25).\nSimilarly, the second argument to the “multivariate normal” function is the covariance matrix and not the inverse covariance matrix (i.e., the precision matrix) (as in BUGS). The same is true for the “multivariate student” distribution.\nThe distributions have slightly different names:\n\n\n\nBUGS\nStan\n\n\n\n\ndnorm\nnormal\n\n\ndbinom\nbinomial\n\n\ndpois\npoisson\n\n\n…\n…\n\n\n\nStan, unlike BUGS, allows intermediate quantities, in the form of local variables, to be reassigned. For example, the following is legal and meaningful (if possibly inefficient) Stan code.\n{\n  total = 0;\n  for (i in 1:n) {\n    theta[i] ~ normal(total, sigma);\n    total = total + theta[i];\n  }\n}\nIn BUGS, the above model would not be legal because the variable total is defined more than once. But in Stan, the loop is executed in order, so total is overwritten in each step.\nStan uses explicit declarations. Variables are declared with base type integer or real, and vectors, matrices, and arrays have specified dimensions. When variables are bounded, we give that information also. For data and transformed parameters, the bounds are used for error checking. For parameters, the constraints are critical to sampling as they determine the geometry over which the Hamiltonian is simulated.\nIn Stan, variables can be declared as data, transformed data, parameters, transformed parameters, or generated quantities. They can also be declared as local variables within blocks. For more information, see the part of this manual devoted to the Stan programming language and examine at the example models.\nStan allows all sorts of tricks with vector and matrix operations which can make Stan models more compact. For example, arguments to probability functions may be vectorized,2 allowing\nfor (i in 1:n) {\n  y[i] ~ normal(mu[i], sigma[i]);\n}\nto be expressed more compactly as\ny ~ normal(mu, sigma);\nThe vectorized form is also more efficient because Stan can unfold the computation of the chain rule during algorithmic differentiation.\nStan also allows for arrays of vectors and matrices. For example, in a hierarchical model might have a vector of K parameters for each of J groups; this can be declared using\narray[J] vector[K] theta;\nThen theta[j] is an expression denoting a K-vector and may be used in the code just like any other vector variable.\nAn alternative encoding would be with a two-dimensional array, as in\narray[J, K] real theta;\nThe vector version can have some advantages, both in convenience and in computational speed for some operations.\nA third encoding would use a matrix:\nmatrix[J, K] theta;\nbut in this case, theta[j] is a row vector, not a vector, and accessing it as a vector is less efficient than with an array of vectors. The transposition operator, as in theta[j]', may be used to convert the row vector theta[j] to a (column) vector. Column vector and row vector types are not interchangeable everywhere in Stan; see the function signature declarations in the programming language section of this manual.\nStan supports general conditional statements using a standard if-else syntax. For example, a zero-inflated (or -deflated) Poisson mixture model is defined using the if-else syntax as described in the zero inflation section.\nStan supports general while loops using a standard syntax. While loops give Stan full Turing equivalent computational power. They are useful for defining iterative functions with complex termination conditions. As an illustration of their syntax, the for-loop\nmodel {\n    // ...\n    for (n in 1:N) {\n        // ... do something with n ....\n    }\n}\nmay be recoded using the following while loop.\nmodel {\n    int n;\n    // ...\n    n = 1;\n    while (n &lt;= N) {\n        // ... do something with n ...\n        n = n + 1;\n    }\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Multiple Indexing and Range Indexing"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexing",
-    "href": "stan-users-guide/multi-indexing.html#multiple-indexing",
-    "title": "Multiple Indexing and Range Indexing",
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-in-the-statistical-models-that-are-allowed",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-in-the-statistical-models-that-are-allowed",
+    "title": "Transitioning from BUGS",
     "section": "",
-    "text": "The following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.\narray[3] int c;\n// ... define: c == (5, 9, 7)\narray[4] int idxs;\n// ... define: idxs == (3, 3, 1, 2)\narray[4] int d;\nd = c[idxs];    // result: d == (7, 7, 5, 9)\nIn general, the multiple indexed expression c[idxs] is defined as follows, assuming idxs is of size K.\nc[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )\nThus c[idxs] is of the same size as idxs, which is K in this example.\nMultiple indexing can also be used with multi-dimensional arrays. For example, consider the following.\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[4] int idxs;\n// ... define: idxs = (2, 2, 1, 2)\narray[4, 3] int d\nd = c[idxs];    // result: d = ((7, 11, 13), (7, 11, 13),\n                //              (1, 3, 5), (7, 11, 13))\nThat is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by c[idxs][j] ==   c[idxs[j]].\nMultiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.\narray[4] int e;\ne = c[2, idxs]; // result:  c[2] = (7, 11, 13)\n                // result:  e = (11, 11, 7, 11)\nThe single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, c[2,idxs] evaluates to the same value as c[2][idxs].\nMultiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[3] int idxs1;\n// ... define: idxs1 = (2, 2, 1)\narray[2] int idxs2;\n// ... define: idxs2 = (1, 3)\narray[3, 2] int d;\nd = c[idxs1, idxs2];  // result: d = ((7, 13), (7, 13), (1, 5))\nWith multiple indexes, we no longer have c[idxs1, idxs2] being the same as c[idxs1][idxs2]. Rather, the entry d[i, j] after executing the above is given by\nd[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]\nThis example illustrates the operation of multiple indexing in the general case: a multiple index like idxs1 converts an index i used on the result (here, c[idxs1, idxs2]) to index idxs1[i] in the variable being indexed (here, c). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.",
+    "text": "Stan does not yet support declaration of discrete parameters. Discrete data variables are supported. Inference is supported for discrete parameters as described in the mixture and latent discrete parameters chapters of the manual.\nStan has some distributions on covariance matrices that do not exist in BUGS, including a uniform distribution over correlation matrices which may be rescaled, and the priors based on C-vines defined in Lewandowski, Kurowicka, and Joe (2009). In particular, the Lewandowski et al. prior allows the correlation matrix to be shrunk toward the unit matrix while the scales are given independent priors.\nIn BUGS you need to define all variables. In Stan, if you declare but don’t define a parameter it implicitly has a flat prior (on the scale in which the parameter is defined). For example, if you have a parameter p declared as\nreal&lt;lower=0, upper=1&gt; p;\nand then have no distribution statement for p in the model block, then you are implicitly assigning a uniform \\([0,1]\\) prior on p.\nOn the other hand, if you have a parameter theta declared with\nreal theta;\nand have no distribution statement for theta in the model block, then you are implicitly assigning an improper uniform prior on \\((-\\infty,\\infty)\\) to theta.\nBUGS models are always proper (being constructed as a product of proper marginal and conditional densities). Stan models can be improper. Here is the simplest improper Stan model:\nparameters {\n  real theta;\n}\nmodel { }\nAlthough parameters in Stan models may have improper priors, we do not want improper posterior distributions, as we are trying to use these distributions for Bayesian inference. There is no general way to check if a posterior distribution is improper. But if all the priors are proper, the posterior will be proper also.\nEach statement in a Stan model is directly translated into the C++ code for computing the log posterior. Thus, for example, the following pair of statements is legal in a Stan model:\ny ~ normal(0,1);\ny ~ normal(2,3);\nThe second line here does not simply overwrite the first; rather, both statements contribute to the density function that is evaluated. The above two lines have the effect of including the product, \\(\\textsf{normal}(y \\mid 0,1) * \\textsf{normal}(y \\mid 2,3)\\), into the density function.\nFor a perhaps more confusing example, consider the following two lines in a Stan model:\nx ~ normal(0.8 * y, sigma);\ny ~ normal(0.8 * x, sigma);\nAt first, this might look like a joint normal distribution with a correlation of 0.8. But it is not. The above are not interpreted as conditional entities; rather, they are factors in the joint density. Multiplying them gives, \\(\\textsf{normal}(x \\mid 0.8y,\\sigma)\n\\times \\textsf{normal}(y \\mid 0.8x,\\sigma)\\), which is what it is (you can work out the algebra) but it is not the joint distribution where the conditionals have regressions with slope 0.8.\nWith censoring and truncation, Stan uses the censored-data or truncated-data likelihood—this is not always done in BUGS. All of the approaches to censoring and truncation discussed in Gelman et al. (2013) and Gelman and Hill (2007) may be implemented in Stan directly as written.\nStan, like BUGS, can benefit from human intervention in the form of reparameterization.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Multiple Indexing and Range Indexing"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/multi-indexing.html#slicing-with-range-indexes",
-    "href": "stan-users-guide/multi-indexing.html#slicing-with-range-indexes",
-    "title": "Multiple Indexing and Range Indexing",
+    "objectID": "stan-users-guide/for-bugs-users.html#some-differences-when-running-from-r",
+    "href": "stan-users-guide/for-bugs-users.html#some-differences-when-running-from-r",
+    "title": "Transitioning from BUGS",
     "section": "",
-    "text": "Slicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.\n\n\nFor instance, consider supplying an upper and lower bound for an index.\narray[7] int c;\n// ...\narray[4] int d;\nd = c[3:6];  // result: d == (c[3], c[4], c[5], c[6])\nThe range index 3:6 behaves semantically just like the multiple index (3, 4, 5, 6). In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.\n\n\n\nIt is also possible to supply just a lower bound, or just an upper bound. Writing c[3:] is just shorthand for c[3:size(c)]. Writing c[:5] is just shorthand for c[1:5].\n\n\n\nFinally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (:) as the index or leaving the index position empty. In both cases, c[] and c[:] are equal to c[1:size(c)], which in turn is just equal to c.\n\n\n\nStan provides head and tail functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,\nvector[M] a = ...;\nvector[N] b = head(a, N);\nassigns b to be a vector equivalent to the first N elements of the vector a. The function tail works the same way for suffixes, with\narray[M] a = ...;\narray[N] b = tail(a, N);\nFinally, there is a segment function, which specifies a first element and number of elements. For example,\narray[15] a = ...;\narray[3] b = segment(a, 5, 3);\nwill set b to be equal to { a[5], a[6], a[7] }, so that it starts at element 5 of a and includes a total of 3 elements.",
+    "text": "Stan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.\nIn practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.\nWhen you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)\nWhen running BUGS from R, you need to send exactly the data that the model needs. When running RStan, you can include extra data, which can be helpful when playing around with models. For example, if you remove a variable x from the model, you can keep it in the data sent from R, thus allowing you to quickly alter the Stan model without having to also change the calling information in your R script.\nAs in R2WinBUGS and R2jags, after running the Stan model, you can quickly summarize using plot() and print(). You can access the simulations themselves using various extractor functions, as described in the RStan documentation.\nVarious information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Multiple Indexing and Range Indexing"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexing-on-the-left-of-assignments",
-    "href": "stan-users-guide/multi-indexing.html#multiple-indexing-on-the-left-of-assignments",
-    "title": "Multiple Indexing and Range Indexing",
+    "objectID": "stan-users-guide/for-bugs-users.html#the-stan-community",
+    "href": "stan-users-guide/for-bugs-users.html#the-stan-community",
+    "title": "Transitioning from BUGS",
     "section": "",
-    "text": "Multiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.\narray[3] int a;\narray[2] int c;\narray[2] int idxs;\n// ... define: a == (1, 2, 3);  c == (5, 9)\n               //         idxs = (3,2)\na[idxs] = c;   // result: a == (1, 9, 5)\nThe result above can be worked out by noting that the assignment sets a[idxs[1]] (a[3]) to c[1] (5) and a[idxs[2]] (a[2]) to c[2] (9).\nThe same principle applies when there are many multiple indexes, as in the following example.\narray[5, 7] int a;\narray[2, 2] int c;\n// ...\na[2:3, 5:6] = c;  // result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]\n                  //         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]\nAs in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.\nUsage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.\narray[10, 13] int a;\narray[2] int c;\n// ...\na[4, 2:3] = c;  // result:  a[4, 2] == c[1];  a[4, 3] == c[2]\n\n\nAliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array a in the following code fragment.\narray[3] int a;\n// ... define: a == (5, 6, 7)\na[2:3] = a[1:2];\n// ... result: a == (5, 5, 6)\nThe reason the value of a after the assignment is \\((5,5,6)\\) rather than \\((5,5,5)\\) is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.\narray[3] int a;\narray[3] int idxs;\n// ... define idxs = (2, 1, 3)\na[idxs] = a;\nIn this case, it is evident why the right-hand side needs to be copied before the assignment.\nIt is tempting (but wrong) to think of the assignment a[2:3] =   a[1:2] as executing the following assignments.\n// ... define: a = (5, 6, 7)\na[2] = a[1];      // result: a = (5, 5, 7)\na[3] = a[2];      // result: a = (5, 5, 5)!\nThis produces a different result than executing the assignment because a[2]’s value changes before it is used.",
+    "text": "Stan, like WinBUGS, OpenBUGS, and JAGS, has an active community, which you can access via the user’s mailing list and the developer’s mailing list; see the Stan web site for information on subscribing and posting and to look at archives.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Multiple Indexing and Range Indexing"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexes-with-vectors-and-matrices",
-    "href": "stan-users-guide/multi-indexing.html#multiple-indexes-with-vectors-and-matrices",
-    "title": "Multiple Indexing and Range Indexing",
-    "section": "",
-    "text": "Multiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.\n\n\nVectors and row vectors behave exactly the same way as arrays with multiple indexes. If v is a vector, then v[3] is a scalar real value, whereas v[2:4] is a vector of size 3 containing the elements v[2], v[3], and v[4].\nThe only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.\narray[3] vector[5] v;\narray[7] int idxs;\n// ...\nvector[7] u;\nu = v[2, idxs];\n\narray[7] real w;\nw = v[idxs, 2];\nThe key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because v is an array of vectors, v[2, idxs] reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, v[idxs, 2] does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for w is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.\n\n\n\nMatrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.\nmatrix[5, 7] m;\n// ...\nrow_vector[3] rv;\nrv = m[4, 3:5];    // result is 1 x 3\n// ...\nvector[4] v;\nv = m[2:5, 3];     // result is 3 x 1\n// ...\nmatrix[3, 4] m2;\nm2 = m[1:3, 2:5];  // result is 3 x 4\nThe key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.\n\n\n\nIf matrices receive a single multiple index, the result is a matrix. So if m is a matrix, so is m[2:4]. In contrast, supplying a single index, m[3], produces a row vector result. That is, m[3] produces the same result as m[3, ] or m[3, 1:cols(m)].\n\n\n\nWith arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.\narray[5, 7] matrix[3, 4] m;\n// ...\narray[2] matrix[3, 4] a;\na = m[1, 2:3];  // knock off first array dimension\na = m[3:4, 5];  // knock off second array dimension\nIn both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, a[i] ==   m[1, i + 1], whereas in the second case, a[i] == m[i + 2,   5].\nContinuing the previous example, consider the following.\n// ...\nvector[2] b;\nb = a[1, 3, 2:3, 2];\nHere, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, b[j] == a[1, 3, 1 + j, 2].\nThis last example illustrates an important point: if there is a lower-bounded index, such as 2:3, with lower bound 2, then the lower bound minus one is added to the index, as seen in the 1 + j expression above.\nContinuing further, consider continuing with the following.\n// ...\narray[2] row_vector[3] c;\nc = a[4:5, 3, 1, 2: ];\nHere, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by c[i, j] == a[i + 3, 3, 1, j + 1]\n\n\n\nMatrix slicing can also be performed using the block function. For example,\nmatrix[20, 20] a = ...;\nmatrix[3, 2] b = block(a, 5, 9, 3, 2);\nwill set b equal to the submatrix of a starting at index [5, 9] and extending 3 rows and 2 columns. Thus block(a, 5, 9, 3, 2) is equivalent to b[5:7, 9:10].\nThe sub_col function extracts a slice of a column of a matrix as a vector. For example,\nmatrix[10, 10] a = ...;\nvector b = sub_col(a, 2, 3, 5);\nwill set b equal to the vector a[2:6, 3], taking the element starting at [2, 3], then extending for a total of 5 rows. The function sub_row works the same way for extracting a slice of a row as a row vector. For example, sub_row(a, 2, 3, 5) is equal to the row vector a[2, 3:7], which also starts at position [2, 3] then extends for a total of 5 columns.",
+    "objectID": "stan-users-guide/for-bugs-users.html#footnotes",
+    "href": "stan-users-guide/for-bugs-users.html#footnotes",
+    "title": "Transitioning from BUGS",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nExcept where otherwise noted, we use “BUGS” to refer to WinBUGS, OpenBUGS, and JAGS, indiscriminately.↩︎\nMost distributions have been vectorized, but currently the truncated versions may not exist.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Multiple Indexing and Range Indexing"
+      "Appendices",
+      "Transitioning from BUGS"
     ]
   },
   {
-    "objectID": "stan-users-guide/multi-indexing.html#matrices-with-parameters-and-constants",
-    "href": "stan-users-guide/multi-indexing.html#matrices-with-parameters-and-constants",
-    "title": "Multiple Indexing and Range Indexing",
+    "objectID": "stan-users-guide/hyperspherical-models.html",
+    "href": "stan-users-guide/hyperspherical-models.html",
+    "title": "Directions, Rotations, and Hyperspheres",
     "section": "",
-    "text": "Suppose you have a \\(3 \\times 3\\) matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.\nSuppose a \\(3 \\times 3\\) matrix is known to be zero at indexes \\([1,2]\\) and \\([1,3]\\). The indexes for parameters are included in a “melted” data-frame or database format.\ntransformed data {\n  array[7, 2] int&lt;lower=1, upper=3&gt; idxs\n    = { {1, 1},\n        {2, 1}, {2, 2}, {2, 3},\n        {3, 1}, {3, 2}, {3, 3} };\n  // ...\nThe seven remaining parameters are declared as a vector.\nparameters {\n  vector[7] A_raw;\n  // ...\n}\nThen the full matrix A is constructed in the model block as a local variable.\nmodel {\n  matrix[3, 3] A;\n  for (i in 1:7) {\n    A[idxs[i, 1], idxs[i, 2]] = A_raw[i];\n  }\n  A[1, 2] = 0;\n  A[1, 3] = 0;\n  // ...\n}\nThis may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the idxs array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.",
+    "text": "Directional statistics involve data and/or parameters that are constrained to be directions. The set of directions forms a sphere, the geometry of which is not smoothly mappable to that of a Euclidean space because you can move around a sphere and come back to where you started. This is why it is impossible to make a map of the globe on a flat piece of paper where all points that are close to each other on the globe are close to each other on the flat map. The fundamental problem is easy to visualize in two dimensions, because as you move around a circle, you wind up back where you started. In other words, 0 degrees and 360 degrees (equivalently, 0 and \\(2 \\pi\\) radians) pick out the same point, and the distance between 359 degrees and 2 degrees is the same as the distance between 137 and 140 degrees.\nStan supports directional statistics by providing a unit-vector data type, the values of which determine points on a hypersphere (circle in two dimensions, sphere in three dimensions).\n\n\nThe length of a vector \\(x \\in \\mathbb{R}^K\\) is given by \\[\n\\Vert x \\Vert\n= \\sqrt{x^{\\top}\\,x}\n= \\sqrt{x_1^2 + x_2^2 + \\cdots + x_K^2}.\n\\] Unit vectors are defined to be vectors of unit length (i.e., length one).\nWith a variable declaration such as\nunit_vector[K] x;\nthe value of x will be constrained to be a vector of size K with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.\nWarning: An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (lp__ in some interfaces). The underlying source of the problem is that a unit vector of size \\(K\\) has only \\(K - 1\\) degrees of freedom. But there is no way to map those \\(K - 1\\) degrees of freedom continuously to \\(\\mathbb{R}^N\\)—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses \\(K\\) unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.\n\n\n\nAn \\(n\\)-sphere, written \\(S^{n}\\), is defined as the set of \\((n +\n1)\\)-dimensional unit vectors, \\[\nS^{n} = \\left\\{ x \\in \\mathbb{R}^{n+1} \\: : \\: \\Vert x \\Vert = 1 \\right\\}.\n\\]\nEven though \\(S^n\\) is made up of points in \\((n+1)\\) dimensions, it is only an \\(n\\)-dimensional manifold. For example, \\(S^2\\) is defined as a set of points in \\(\\mathbb{R}^3\\), but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by \\(S^2\\) in \\(\\mathbb{R}^3\\) behaves locally like a plane, i.e., \\(\\mathbb{R}^2\\). However, the overall shape of \\(S^2\\) is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (\\(S^2\\)) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.\nEven though \\(S^{n-1}\\) behaves locally like \\(\\mathbb{R}^{n-1}\\), there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at \\(2\\pi\\) radians in natural units), they do not produce a smooth map.\nLike a bounded interval \\((a, b)\\), in geometric terms, a sphere is compact in that the distance between any two points is bounded.\n\n\n\nStan (inverse) transforms arbitrary points in \\(\\mathbb{R}^{K+1}\\) to points in \\(S^K\\) using the auxiliary variable approach of Muller (1959). A point \\(y \\in \\mathbb{R}^K\\) is transformed to a point \\(x \\in S^{K-1}\\) by \\[\nx = \\frac{y}{\\sqrt{y^{\\top} y}}.\n\\]\nThe problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. Muller (1959) introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.\n\n\n\n\nUnit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle \\(\\theta\\). Given an angle \\(\\theta\\), a matrix can be defined, the pre-multiplication by which rotates a point by an angle of \\(\\theta\\). For angle \\(\\theta\\) (in two dimensions), the \\(2 \\times 2\\) rotation matrix is defined by \\[\nR_{\\theta}\n=\n\\begin{bmatrix}\n\\cos \\theta & -\\sin \\theta \\\\\n\\sin \\theta &  \\cos \\theta\n\\end{bmatrix}.\n\\] Given a two-dimensional vector \\(x\\), \\(R_{\\theta} \\, x\\) is the rotation of \\(x\\) (around the origin) by \\(\\theta\\) degrees.\n\n\nAngles can be calculated from unit vectors. For example, a random variable theta representing an angle in \\((-\\pi, \\pi)\\) radians can be declared as a two-dimensional unit vector then transformed to an angle.\nparameters {\n  unit_vector[2] xy;\n}\ntransformed parameters {\n  real&lt;lower=-pi(), upper=pi()&gt; theta = atan2(xy[2], xy[1]);\n}\nIf the distribution of \\((x, y)\\) is uniform over a circle, then the distribution of \\(\\arctan \\frac{y}{x}\\) is uniform over \\((-\\pi, \\pi)\\).\nIt might be tempting to try to just declare theta directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values \\(-\\pi\\) and \\(\\pi\\) are at \\(-\\infty\\) and \\(\\infty\\) on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.\nWith a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.\n\n\n\n\nA 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.\nIn human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Multiple Indexing and Range Indexing"
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
     ]
   },
   {
-    "objectID": "stan-users-guide/one-dimensional-integrals.html",
-    "href": "stan-users-guide/one-dimensional-integrals.html",
-    "title": "Computing One Dimensional Integrals",
+    "objectID": "stan-users-guide/hyperspherical-models.html#unit-vectors",
+    "href": "stan-users-guide/hyperspherical-models.html#unit-vectors",
+    "title": "Directions, Rotations, and Hyperspheres",
     "section": "",
-    "text": "Definite and indefinite one dimensional integrals can be performed in Stan using the integrate_1d function\nAs an example, the normalizing constant of a left-truncated normal distribution is\n\\[\n  \\int_a^\\infty \\frac{1}{\\sqrt{2 \\pi \\sigma^2}} e^{-\\frac{1}{2}\\frac{(x - \\mu)^2}{\\sigma^2}}\n\\]\nTo compute this integral in Stan, the integrand must first be defined as a Stan function (see the Stan Reference Manual chapter on User-Defined Functions for more information on coding user-defined functions).\nreal normal_density(real x,             // Function argument\n                    real xc,            // Complement of function argument\n                                        //  on the domain (defined later)\n                    array[] real theta, // parameters\n                    array[] real x_r,   // data (real)\n                    array[] int x_i) {  // data (integer)\n  real mu = theta[1];\n  real sigma = theta[2];\n\n  return 1 / (sqrt(2 * pi()) * sigma) * exp(-0.5 * ((x - mu) / sigma)^2);\n}\nThis function is expected to return the value of the integrand evaluated at point x. The argument xc is used in definite integrals to avoid loss of precision near the limits of integration and is set to NaN when either limit is infinite (see the section on precision/loss in the chapter on Higher-Order Functions of the Stan Functions Reference for details on how to use this). The argument theta is used to pass in arguments of the integral that are a function of the parameters in our model. The arguments x_r and x_i are used to pass in real and integer arguments of the integral that are not a function of our parameters.\nThe function defining the integrand must have exactly the argument types and return type of normal_density above, though argument naming is not important. Even if x_r and x_i are unused in the integrand, they must be included in the function signature. Even if the integral does not involve some of these, they must still be supplied some value. The most efficient will be a zero-length array or vector, which can be created with rep_array(0, 0) and rep_vector(0, 0), respectively. Other options include an uninitialized variable declared with size 0, which is equivalent to the above, or any easy value, such as size 1 array created with {0}.\n\n\nSuppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.\nSuch a model might look like (include the function defined at the beginning of this chapter to make this code compile):\ndata {\n  int N;\n  array[N] real y;\n}\n\ntransformed data {\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\nparameters {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n  real left_limit;\n}\n\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  left_limit ~ normal(0, 1);\n  target += normal_lpdf(y | mu, sigma);\n  target += -log(integrate_1d(normal_density,\n                              left_limit,\n                              positive_infinity(),\n                              { mu, sigma }, x_r, x_i));\n}\n\n\nThe limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls negative_infinity() and positive_infinity().\nIf both limits are either negative_infinity() or positive_infinity(), the integral and its gradients are set to zero.\n\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.\nThe endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).\n\n\n\n\nThe integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library (Agrawal et al. 2017). If the \\(n\\)th estimate of the integral is denoted \\(I_n\\) and the \\(n\\)th estimate of the norm of the integral is denoted \\(|I|_n\\), the iteration is terminated when\n\\[\n  \\frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \\text{relative tolerance}.\n\\]\nThe relative_tolerance parameter can be optionally specified as the last argument to integrate_1d. By default, integrate_1d follows the Boost library recommendation of setting relative_tolerance to the square root of the machine epsilon of double precision floating point numbers (about 1e-8). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If integrate_1d causes an exception in transformed parameters block or model block, the result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If integrate_1d causes an exception in generated quantities block, the returned output from integrate_1d is NaN. In these cases, a bigger relative_tolerance value can be specified.\n\n\nIntegrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). This is because the quadrature methods employed internally can have difficulty near zero.\nIn this case, each integral is separately integrated to the given relative_tolerance.\n\n\n\nIf care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.\nFor instance, in integrating the pdf of a beta distribution when the values of \\(\\alpha\\) and \\(\\beta\\) are small, most of the probability mass is lumped near zero and one.\nThe pdf of a beta distribution is proportional to\n\\[\np(x) \\propto x^{\\alpha - 1}(1 - x)^{\\beta - 1}\n\\]\nNormalizing this distribution requires computing the integral of \\(p(x)\\) from zero to one. In Stan code, the integrand might look like:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n\n  return x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n}\nThe issue is that there will be numerical breakdown in the precision of 1.0 - x as x gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of alpha and beta much less than one.\nThis is where xc is useful. It is defined, for definite integrals, as a high precision version of the distance from x to the nearest endpoint — a - x or b - x for a lower endpoint a and an upper endpoint b. To make use of this for the beta integral, the integrand can be re-coded:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n  real v;\n\n  if(x &gt; 0.5) {\n    v = x^(alpha - 1.0) * xc^(beta - 1.0);\n  } else {\n    v = x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n  }\n\n  return v;\n}\nIn this case, as we approach the upper limit of integration \\(a = 1\\), xc will take on the value of \\(a - x = 1 - x\\). This version of the integrand will converge for much smaller values of alpha and beta than otherwise possible.\nConsider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount \\(\\delta\\), and truncated at some value \\(b\\). If we were interested in calculating the expectation of a variable \\(X\\) distributed in this way, we would need to calculate \\[\n\\int_a^b xf(x)\\,dx = \\int_{\\delta}^b xf(x)\\,dx\n\\] in the numerator, where \\(f(x)\\) is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:\nreal shift_lognormal_pdf(real x,\n                         real mu,\n                         real sigma,\n                         real delta) {\n  real p;\n\n  p = (1.0 / ((x - delta) * sigma * sqrt(2 * pi()))) *\n    exp(-1 * (log(x - delta) - mu)^2 / (2 * sigma^2));\n\n  return p;\n}\nTherefore, the function that we want to integrate is:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  p = shift_lognormal_pdf(x, mu, sigma, delta);\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhat happens here is that, given that the log-normal distribution is shifted by \\(\\delta\\), when we then try to integrate the numerator, our x starts at values just above delta. This, in turn, causes the x - delta term to be near zero, leading to a breakdown.\nWe can use xc, and define the integrand as:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  if (x &lt; delta + 1) {\n    p = shift_lognormal_pdf(xc, mu, sigma, delta);\n  } else {\n    p = shift_lognormal_pdf(x, mu, sigma, delta);\n  }\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhy does this work? When our values of x are less than delta + 1 (so, when they’re near delta, given that our lower bound of integration is equal to \\(\\delta\\)), we pass xc as an argument to our shift_lognormal_pdf function. This way, instead of dealing with x - delta in shift_lognormal_pdf, we are working with xc - delta which is equal to delta - x - delta, as delta is the lower endpoint in that case. The delta terms cancel out, and we are left with a high-precision version of x. We don’t encounter the same problem at the upper limit \\(b\\) so we don’t adjust the code for that case.\nNote, xc is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, xc will be NaN.\nFor zero-crossing definite integrals (see section Zero Crossing) the integrals are broken into two pieces (\\((a, 0)\\) and \\((0, b)\\) for endpoints \\(a &lt; 0\\) and \\(b &gt; 0\\)) and xc is a high precision version of the distance to the limits of each of the two integrals separately. This means xc will be a high precision version of a - x, x, or b - x, depending on the value of x and the endpoints.",
+    "text": "The length of a vector \\(x \\in \\mathbb{R}^K\\) is given by \\[\n\\Vert x \\Vert\n= \\sqrt{x^{\\top}\\,x}\n= \\sqrt{x_1^2 + x_2^2 + \\cdots + x_K^2}.\n\\] Unit vectors are defined to be vectors of unit length (i.e., length one).\nWith a variable declaration such as\nunit_vector[K] x;\nthe value of x will be constrained to be a vector of size K with unit length; the reference manual chapter on constrained parameter transforms provides precise definitions.\nWarning: An extra term gets added to the log density to ensure the distribution on unit vectors is proper. This is not a problem in practice, but it may lead to misunderstandings of the target log density output (lp__ in some interfaces). The underlying source of the problem is that a unit vector of size \\(K\\) has only \\(K - 1\\) degrees of freedom. But there is no way to map those \\(K - 1\\) degrees of freedom continuously to \\(\\mathbb{R}^N\\)—for example, the circle can’t be mapped continuously to a line so the limits work out, nor can a sphere be mapped to a plane. A workaround is needed instead. Stan’s unit vector transform uses \\(K\\) unconstrained variables, then projects down to the unit hypersphere. Even though the hypersphere is compact, the result would be an improper distribution. To ensure the unit vector distribution is proper, each unconstrained variable is given a “Jacobian” adjustment equal to an independent standard normal distribution. Effectively, each dimension is drawn standard normal, then they are together projected down to the hypersphere to produce a unit vector. The result is a proper uniform distribution over the hypersphere.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Computing One Dimensional Integrals"
+      "Directions, Rotations, and Hyperspheres"
     ]
   },
   {
-    "objectID": "stan-users-guide/one-dimensional-integrals.html#calling-the-integrator",
-    "href": "stan-users-guide/one-dimensional-integrals.html#calling-the-integrator",
-    "title": "Computing One Dimensional Integrals",
+    "objectID": "stan-users-guide/hyperspherical-models.html#circles-spheres-and-hyperspheres",
+    "href": "stan-users-guide/hyperspherical-models.html#circles-spheres-and-hyperspheres",
+    "title": "Directions, Rotations, and Hyperspheres",
     "section": "",
-    "text": "Suppose that our model requires evaluating the lpdf of a left-truncated normal, but the truncation limit is to be estimated as a parameter. Because the truncation point is a parameter, we must include the normalization term of the truncated pdf when computing our model’s log density. Note this is just an example of how to use the 1D integrator. The more efficient way to perform the correct normalization in Stan is described in the chapter on Truncated or Censored Data of this guide.\nSuch a model might look like (include the function defined at the beginning of this chapter to make this code compile):\ndata {\n  int N;\n  array[N] real y;\n}\n\ntransformed data {\n  array[0] real x_r;\n  array[0] int x_i;\n}\n\nparameters {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n  real left_limit;\n}\n\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  left_limit ~ normal(0, 1);\n  target += normal_lpdf(y | mu, sigma);\n  target += -log(integrate_1d(normal_density,\n                              left_limit,\n                              positive_infinity(),\n                              { mu, sigma }, x_r, x_i));\n}\n\n\nThe limits of integration can be finite or infinite. The infinite limits are made available via the Stan calls negative_infinity() and positive_infinity().\nIf both limits are either negative_infinity() or positive_infinity(), the integral and its gradients are set to zero.\n\n\n\nThe arguments for the real data x_r and the integer data x_i must be expressions that only involve data or transformed data variables. theta, on the other hand, can be a function of data, transformed data, parameters, or transformed parameters.\nThe endpoints of integration can be data or parameters (and internally the derivatives of the integral with respect to the endpoints are handled with the Leibniz integral rule).",
+    "text": "An \\(n\\)-sphere, written \\(S^{n}\\), is defined as the set of \\((n +\n1)\\)-dimensional unit vectors, \\[\nS^{n} = \\left\\{ x \\in \\mathbb{R}^{n+1} \\: : \\: \\Vert x \\Vert = 1 \\right\\}.\n\\]\nEven though \\(S^n\\) is made up of points in \\((n+1)\\) dimensions, it is only an \\(n\\)-dimensional manifold. For example, \\(S^2\\) is defined as a set of points in \\(\\mathbb{R}^3\\), but each such point may be described uniquely by a latitude and longitude. Geometrically, the surface defined by \\(S^2\\) in \\(\\mathbb{R}^3\\) behaves locally like a plane, i.e., \\(\\mathbb{R}^2\\). However, the overall shape of \\(S^2\\) is not like a plane in that it is compact (i.e., there is a maximum distance between points). If you set off around the globe in a “straight line” (i.e., a geodesic), you wind up back where you started eventually; that is why the geodesics on the sphere (\\(S^2\\)) are called “great circles,” and why we need to use some clever representations to do circular or spherical statistics.\nEven though \\(S^{n-1}\\) behaves locally like \\(\\mathbb{R}^{n-1}\\), there is no way to smoothly map between them. For example, because latitude and longitude work on a modular basis (wrapping at \\(2\\pi\\) radians in natural units), they do not produce a smooth map.\nLike a bounded interval \\((a, b)\\), in geometric terms, a sphere is compact in that the distance between any two points is bounded.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Computing One Dimensional Integrals"
+      "Directions, Rotations, and Hyperspheres"
     ]
   },
   {
-    "objectID": "stan-users-guide/one-dimensional-integrals.html#integrator-convergence",
-    "href": "stan-users-guide/one-dimensional-integrals.html#integrator-convergence",
-    "title": "Computing One Dimensional Integrals",
+    "objectID": "stan-users-guide/hyperspherical-models.html#transforming-to-unconstrained-parameters",
+    "href": "stan-users-guide/hyperspherical-models.html#transforming-to-unconstrained-parameters",
+    "title": "Directions, Rotations, and Hyperspheres",
     "section": "",
-    "text": "The integral is performed with the iterative 1D double exponential quadrature methods implemented in the Boost library (Agrawal et al. 2017). If the \\(n\\)th estimate of the integral is denoted \\(I_n\\) and the \\(n\\)th estimate of the norm of the integral is denoted \\(|I|_n\\), the iteration is terminated when\n\\[\n  \\frac{{|I_{n + 1} - I_n|}}{{|I|_{n + 1}}} &lt; \\text{relative tolerance}.\n\\]\nThe relative_tolerance parameter can be optionally specified as the last argument to integrate_1d. By default, integrate_1d follows the Boost library recommendation of setting relative_tolerance to the square root of the machine epsilon of double precision floating point numbers (about 1e-8). If the Boost integrator is not able to reach the relative tolerance an exception is raised with a message somehing like “Exception: integrate: error estimate of integral 4.25366e-13 exceeds the given relative tolerance times norm of integral”. If integrate_1d causes an exception in transformed parameters block or model block, the result has the same effect as assigning a \\(-\\infty\\) log probability, which causes rejection of the current proposal in MCMC samplers and adjustment of search parameters in optimization. If integrate_1d causes an exception in generated quantities block, the returned output from integrate_1d is NaN. In these cases, a bigger relative_tolerance value can be specified.\n\n\nIntegrals on the (possibly infinite) interval \\((a, b)\\) that cross zero are split into two integrals, one from \\((a, 0)\\) and one from \\((0, b)\\). This is because the quadrature methods employed internally can have difficulty near zero.\nIn this case, each integral is separately integrated to the given relative_tolerance.\n\n\n\nIf care is not taken, the quadrature can suffer from numerical loss of precision near the endpoints of definite integrals.\nFor instance, in integrating the pdf of a beta distribution when the values of \\(\\alpha\\) and \\(\\beta\\) are small, most of the probability mass is lumped near zero and one.\nThe pdf of a beta distribution is proportional to\n\\[\np(x) \\propto x^{\\alpha - 1}(1 - x)^{\\beta - 1}\n\\]\nNormalizing this distribution requires computing the integral of \\(p(x)\\) from zero to one. In Stan code, the integrand might look like:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n\n  return x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n}\nThe issue is that there will be numerical breakdown in the precision of 1.0 - x as x gets close to one. This is because of the limited precision of double precision floating numbers. This integral will fail to converge for values of alpha and beta much less than one.\nThis is where xc is useful. It is defined, for definite integrals, as a high precision version of the distance from x to the nearest endpoint — a - x or b - x for a lower endpoint a and an upper endpoint b. To make use of this for the beta integral, the integrand can be re-coded:\nreal beta(real x, real xc, array[] real theta, array[] real x_r, array[] int x_i) {\n  real alpha = theta[1];\n  real beta = theta[2];\n  real v;\n\n  if(x &gt; 0.5) {\n    v = x^(alpha - 1.0) * xc^(beta - 1.0);\n  } else {\n    v = x^(alpha - 1.0) * (1.0 - x)^(beta - 1.0);\n  }\n\n  return v;\n}\nIn this case, as we approach the upper limit of integration \\(a = 1\\), xc will take on the value of \\(a - x = 1 - x\\). This version of the integrand will converge for much smaller values of alpha and beta than otherwise possible.\nConsider another example: let’s say we have a log-normal distribution that is both shifted away from zero by some amount \\(\\delta\\), and truncated at some value \\(b\\). If we were interested in calculating the expectation of a variable \\(X\\) distributed in this way, we would need to calculate \\[\n\\int_a^b xf(x)\\,dx = \\int_{\\delta}^b xf(x)\\,dx\n\\] in the numerator, where \\(f(x)\\) is the probability density function for the shifted log-normal distribution. This probability density function can be coded in Stan as:\nreal shift_lognormal_pdf(real x,\n                         real mu,\n                         real sigma,\n                         real delta) {\n  real p;\n\n  p = (1.0 / ((x - delta) * sigma * sqrt(2 * pi()))) *\n    exp(-1 * (log(x - delta) - mu)^2 / (2 * sigma^2));\n\n  return p;\n}\nTherefore, the function that we want to integrate is:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  p = shift_lognormal_pdf(x, mu, sigma, delta);\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhat happens here is that, given that the log-normal distribution is shifted by \\(\\delta\\), when we then try to integrate the numerator, our x starts at values just above delta. This, in turn, causes the x - delta term to be near zero, leading to a breakdown.\nWe can use xc, and define the integrand as:\nreal integrand(real x,\n               real xc,\n               array[] real theta,\n               array[] real x_r,\n               array[] int x_i) {\n  real numerator;\n  real p;\n\n  real mu = theta[1];\n  real sigma = theta[2];\n  real delta = theta[3];\n  real b = theta[4];\n\n  if (x &lt; delta + 1) {\n    p = shift_lognormal_pdf(xc, mu, sigma, delta);\n  } else {\n    p = shift_lognormal_pdf(x, mu, sigma, delta);\n  }\n\n  numerator = x * p;\n\n  return numerator;\n}\nWhy does this work? When our values of x are less than delta + 1 (so, when they’re near delta, given that our lower bound of integration is equal to \\(\\delta\\)), we pass xc as an argument to our shift_lognormal_pdf function. This way, instead of dealing with x - delta in shift_lognormal_pdf, we are working with xc - delta which is equal to delta - x - delta, as delta is the lower endpoint in that case. The delta terms cancel out, and we are left with a high-precision version of x. We don’t encounter the same problem at the upper limit \\(b\\) so we don’t adjust the code for that case.\nNote, xc is only used for definite integrals. If either the left endpoint is at negative infinity or the right endpoint is at positive infinity, xc will be NaN.\nFor zero-crossing definite integrals (see section Zero Crossing) the integrals are broken into two pieces (\\((a, 0)\\) and \\((0, b)\\) for endpoints \\(a &lt; 0\\) and \\(b &gt; 0\\)) and xc is a high precision version of the distance to the limits of each of the two integrals separately. This means xc will be a high precision version of a - x, x, or b - x, depending on the value of x and the endpoints.",
+    "text": "Stan (inverse) transforms arbitrary points in \\(\\mathbb{R}^{K+1}\\) to points in \\(S^K\\) using the auxiliary variable approach of Muller (1959). A point \\(y \\in \\mathbb{R}^K\\) is transformed to a point \\(x \\in S^{K-1}\\) by \\[\nx = \\frac{y}{\\sqrt{y^{\\top} y}}.\n\\]\nThe problem with this mapping is that it’s many to one; any point lying on a vector out of the origin is projected to the same point on the surface of the sphere. Muller (1959) introduced an auxiliary variable interpretation of this mapping that provides the desired properties of uniformity; the reference manual contains the precise definitions used in the chapter on constrained parameter transforms.\n\n\nThe above mapping from \\(\\mathbb{R}^n\\) to \\(S^n\\) is not defined at zero. While this point outcome has measure zero during sampling, and may thus be ignored, it is the default initialization point and thus unit vector parameters cannot be initialized at zero. A simple workaround is to initialize from a small interval around zero, which is an option built into all of the Stan interfaces.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Computing One Dimensional Integrals"
+      "Directions, Rotations, and Hyperspheres"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html",
-    "href": "stan-users-guide/posterior-prediction.html",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/hyperspherical-models.html#unit-vectors-and-rotations",
+    "href": "stan-users-guide/hyperspherical-models.html#unit-vectors-and-rotations",
+    "title": "Directions, Rotations, and Hyperspheres",
     "section": "",
-    "text": "The goal of inference is often posterior prediction, that is evaluating or sampling from the posterior predictive distribution \\(p(\\tilde{y} \\mid y),\\) where \\(y\\) is observed data and \\(\\tilde{y}\\) is yet to be observed data. Often there are unmodeled predictors \\(x\\) and \\(\\tilde{x}\\) for the observed data \\(y\\) and unobserved data \\(\\tilde{y}\\). With predictors, the posterior predictive density is \\(p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\\) All of these variables may represent multivariate quantities.\nThis chapter explains how to sample from the posterior predictive distribution in Stan, including applications to posterior predictive simulation and calculating event probabilities. These techniques can be coded in Stan using random number generation in the generated quantities block. Further, a technique for fitting and performing inference in two stages is presented in a section on stand-alone generated quantities in Stan\n\n\nGiven a full Bayesian model \\(p(y, \\theta)\\), the posterior predictive density for new data \\(\\tilde{y}\\) given observed data \\(y\\) is \\[\np(\\tilde{y} \\mid y)\n=\n\\int p(\\tilde{y} \\mid \\theta) \\cdot p(\\theta \\mid y)\n\\, \\textrm{d}\\theta.\n\\] The product under the integral reduces to the joint posterior density \\(p(\\tilde{y}, \\theta \\mid y),\\) so that the integral is simply marginalizing out the parameters \\(\\theta,\\) leaving the predictive density \\(p(\\tilde{y} \\mid y)\\) of future observations given past observations.\n\n\n\nThe posterior predictive density (or mass) of a prediction \\(\\tilde{y}\\) given observed data \\(y\\) can be computed using \\(M\\) Monte Carlo draws\n\\[\n\\theta^{(m)} \\sim p(\\theta \\mid y)\n\\] from the posterior as \\[\np(\\tilde{y} \\mid y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\]\nComputing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, \\(\\log\np(\\tilde{y} \\mid y)\\) may be computed using the stable log sum of exponents function as \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\\\[4pt]\n& = &\n- \\log M\n+ \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\theta^{(m)}),\n\\end{eqnarray*}\\] where \\[\n\\textrm{log-sum-exp}_{m = 1}^M v_m\n= \\log \\sum_{m = 1}^M \\exp v_m\n\\] is used to maintain arithmetic precision. See the section on log sum of exponentials for more details.\n\n\n\nGiven draws from the posterior \\(\\theta^{(m)} \\sim p(\\theta \\mid y),\\) draws from the posterior predictive \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid y)\\) can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\theta^{(m)}).\n\\]\nRandomly drawing \\(\\tilde{y}\\) from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because \\(\\theta\\) is unknown and estimated based only on a finite sample of data \\(y\\). Aleatoric uncertainty arises because even a known value of \\(\\theta\\) leads to uncertainty about new \\(\\tilde{y}\\) as described by the data model \\(p(\\tilde{y} \\mid \\theta)\\). Both forms of uncertainty show up in the factored form of the posterior predictive distribution, \\[\np(\\tilde{y} \\mid y)\n=\n\\int\n\\underbrace{p(\\tilde{y} \\mid \\theta)}_{\\begin{array}{l}\n                                         \\textrm{aleatoric}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\cdot \\underbrace{p(\\theta \\mid y)}_{\\begin{array}{l}\n                                         \\textrm{epistemic}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\, \\textrm{d}\\theta.\n\\]\n\n\n\nPosterior predictive quantities can be coded in Stan using the generated quantities block.\n\n\nFor example, consider a simple Poisson model for count data with a rate parameter \\(\\lambda &gt; 0\\) having a gamma-distributed prior, \\[\n\\lambda \\sim \\textrm{gamma}(1, 1).\n\\] The \\(N\\) observations \\(y_1, \\ldots, y_N\\) are modeled as Poisson distributed, \\[\ny_n \\sim \\textrm{poisson}(\\lambda).\n\\]\n\n\n\nThe following Stan program defines a variable for \\(\\tilde{y}\\) by random number generation in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  lambda ~ gamma(1, 1);\n  y ~ poisson(lambda);\n}\ngenerated quantities {\n  int&lt;lower=0&gt; y_tilde = poisson_rng(lambda);\n}\nThe random draw from the data model for \\(\\tilde{y}\\) is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new \\(\\tilde{y}^{(m)} \\sim p(y \\mid\n\\lambda^{(m)})\\) for each posterior draw \\(\\lambda^{(m)} \\sim p(\\theta\n\\mid y).\\)\nThe posterior draws \\(\\tilde{y}^{(m)}\\) may be used to estimate the expected value of \\(\\tilde{y}\\) or any of its quantiles or posterior intervals, as well as event probabilities involving \\(\\tilde{y}\\). In general, \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\) may be evaluated as \\[\n\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\n\\approx \\frac{1}{M} \\sum_{m=1}^M f(\\tilde{y}^{(m)}, \\theta^{(m)}),\n\\] which is just the posterior mean of \\(f(\\tilde{y}, \\theta).\\) This quantity is computed by Stan if the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to a variable in the generated quantities block. That is, if we have\ngenerated quantities {\n  real f_val = f(y_tilde, theta);\n  // ...\n}\nwhere the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to variable f_val, then the posterior mean of f_val will be the expectation \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\).\n\n\n\nThe gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density \\(p(\\lambda \\mid y)\\) will also follow a gamma distribution.\nBecause the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive \\(p(\\tilde{y} \\mid\ny)\\) will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, \\(y \\sim\n\\textrm{negative-binomial}(\\alpha, \\beta)\\) if \\(\\lambda \\sim\n\\textrm{gamma}(\\alpha, \\beta)\\) and \\(y \\sim \\textrm{poisson}(\\lambda).\\) Rather than marginalizing out the rate parameter \\(\\lambda\\) analytically as can be done to define the negative binomial probability mass function, the rate \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is sampled from the posterior and then used to generate a draw of \\(\\tilde{y}^{(m)} \\sim p(y \\mid \\lambda^{(m)}).\\)\n\n\n\n\n\n\nConsider a regression with a single predictor \\(x_n\\) for the training outcome \\(y_n\\) and \\(\\tilde{x}_n\\) for the test outcome \\(\\tilde{y}_n.\\) Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in \\[\\begin{eqnarray}\np(\\tilde{y} \\mid \\tilde{x}, y, x)\n& = & \\int p(\\tilde{y} \\mid \\tilde{x}, \\theta) \\cdot p(\\theta \\mid y, x) \\,\n\\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m=1}^M \\, p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\end{eqnarray}\\] where \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y).\\)\n\n\n\nThe following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed \\(y\\) values are coded as data. The predictive quantities \\(\\tilde{y}\\) appear in the generated quantities block, where they are generated by random number generation.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0&gt; y;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ poisson_log(alpha + beta * x);\n  { alpha, beta } ~ normal(0, 1);\n}\ngenerated quantities {\n  array[N_tilde] int&lt;lower=0&gt; y_tilde\n    = poisson_log_rng(alpha + beta * x_tilde);\n}\nThe Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s poisson_log vs. poisson, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept alpha and slope beta, are given standard normal priors.\nIn the model block, the log rate for the Poisson is a linear function of the training data \\(x\\), whereas in the generated quantities block it is a function of the test data \\(\\tilde{x}\\). Because the generated quantities block does not affect the posterior draws, the model fits \\(\\alpha\\) and \\(\\beta\\) using only the training data, reserving \\(\\tilde{x}\\) to generate \\(\\tilde{y}.\\)\nThe result from running Stan is a predictive sample \\(\\tilde{y}^{(1)},\n\\ldots \\tilde{y}^{(M)}\\) where each \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid \\tilde{x}, x, y).\\)\nThe mean of the posterior predictive distribution is the expected value \\[\\begin{align}\n\\mathbb{E}[\\tilde{y} \\mid \\tilde{x}, x, y]\n& =\n\\int\n\\tilde{y}\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n\\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}^{(m)},\n\\end{align}\\] where the \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y)\\) are drawn from the posterior predictive distribution. Thus the posterior mean of y_tilde[n] after running Stan is the expected value of \\(\\tilde{y}_n\\) conditioned on the training data \\(x, y\\) and predictor \\(\\tilde{x}_n.\\) This is the Bayesian estimate for \\(\\tilde{y}\\) with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for \\(\\tilde{y}\\), as well as covariance of the \\(\\tilde{y_n}.\\) The posterior draws \\(\\tilde{y}^{(m)}\\) may also be used to estimate predictive event probabilities, such as \\(\\Pr[\\tilde{y}_1 &gt; 0]\\) or \\(\\Pr[\\prod_{n =\n1}^{\\tilde{N}}(\\tilde{y_n}) &gt; 1],\\) as expectations of indicator functions.\nAll of this can be carried out by running Stan only a single time to draw a single sample of \\(M\\) draws, \\[\n\\tilde{y}^{(1)}, \\ldots, \\tilde{y}^{(M)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\n\\] It’s only when moving to cross-validation where multiple runs are required.\n\n\n\n\nEvent probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate \\(\\Pr[\\lambda &gt; 5 \\mid y]\\) in the simple Poisson example with only a rate parameter \\(\\lambda\\), it suffices to define a generated quantity\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; lambda_gt_5 = lambda &gt; 5;\n  // ...\n}\nThe value of the expression lambda &gt; 5 is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = &\n\\int \\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}[\\lambda^{(m)} &gt; 5],\n\\end{eqnarray*}\\] where each \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter lambda_gt_5.\nIn general, event probabilities may be expressed as expectations of indicator functions. For example, \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = & \\mathbb{E}[\\textrm{I}[\\lambda &gt; 5] \\mid y]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}(\\lambda^{(m)} &gt; 5).\n\\end{eqnarray*}\\] The last line above is the posterior mean of the indicator function as coded in Stan.\nEvent probabilities involving posterior predictive quantities \\(\\tilde{y}\\) work exactly the same way as those for parameters. For example, if \\(\\tilde{y}_n\\) is the prediction for the \\(n\\)-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then \\[\\begin{eqnarray*}\n\\Pr[\\tilde{y}_3 &gt; \\tilde{y}_7 \\mid \\tilde{x}, x, y]\n& = &\n\\mathbb{E}\\!\\left[I[\\tilde{y}_3 &gt; \\tilde{y}_7] \\mid \\tilde{x}, x, y\\right]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\tilde{y}_3 &gt; \\tilde{y}_7)\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\, \\textrm{d}\\tilde{y}\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n\\textrm{I}(\\tilde{y}^{(m)}_3 &gt; \\tilde{y}^{(m)}_7),\n\\end{eqnarray*}\\] where \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y).\\)\n\n\n\nStan’s sampling algorithms take a Stan program representing a posterior \\(p(\\theta \\mid y, x)\\) along with actual data \\(x\\) and \\(y\\) to produce a set of draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)}\\) from the posterior. Posterior predictive draws \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y)\\) can be generated by drawing \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\tilde{x}, \\theta^{(m)})\n\\] from the data model. Note that drawing \\(\\tilde{y}^{(m)}\\) only depends on the new predictors \\(\\tilde{x}\\) and the posterior draws \\(\\theta^{(m)}\\). Most importantly, neither the original data or the model density is required.\nBy saving the posterior draws, predictions for new data items \\(\\tilde{x}\\) may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 5);\n  beta ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.5);\n}\nA second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde\n    = normal_rng(alpha + beta * x_tilde, sigma);\n}\nWhen running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for \\(\\tilde{y}\\) or derived quantities such as event probabilities.\nAny posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.",
+    "text": "Unit vectors correspond directly to angles and thus to rotations. This is easy to see in two dimensions, where a point on a circle determines a compass direction, or equivalently, an angle \\(\\theta\\). Given an angle \\(\\theta\\), a matrix can be defined, the pre-multiplication by which rotates a point by an angle of \\(\\theta\\). For angle \\(\\theta\\) (in two dimensions), the \\(2 \\times 2\\) rotation matrix is defined by \\[\nR_{\\theta}\n=\n\\begin{bmatrix}\n\\cos \\theta & -\\sin \\theta \\\\\n\\sin \\theta &  \\cos \\theta\n\\end{bmatrix}.\n\\] Given a two-dimensional vector \\(x\\), \\(R_{\\theta} \\, x\\) is the rotation of \\(x\\) (around the origin) by \\(\\theta\\) degrees.\n\n\nAngles can be calculated from unit vectors. For example, a random variable theta representing an angle in \\((-\\pi, \\pi)\\) radians can be declared as a two-dimensional unit vector then transformed to an angle.\nparameters {\n  unit_vector[2] xy;\n}\ntransformed parameters {\n  real&lt;lower=-pi(), upper=pi()&gt; theta = atan2(xy[2], xy[1]);\n}\nIf the distribution of \\((x, y)\\) is uniform over a circle, then the distribution of \\(\\arctan \\frac{y}{x}\\) is uniform over \\((-\\pi, \\pi)\\).\nIt might be tempting to try to just declare theta directly as a parameter with the lower and upper bound constraint as given above. The drawback to this approach is that the values \\(-\\pi\\) and \\(\\pi\\) are at \\(-\\infty\\) and \\(\\infty\\) on the unconstrained scale, which can produce multimodal posterior distributions when the true distribution on the circle is unimodal.\nWith a little additional work on the trigonometric front, the same conversion back to angles may be accomplished in more dimensions.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#posterior-predictive-distribution",
-    "href": "stan-users-guide/posterior-prediction.html#posterior-predictive-distribution",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/hyperspherical-models.html#circular-representations-of-days-and-years",
+    "href": "stan-users-guide/hyperspherical-models.html#circular-representations-of-days-and-years",
+    "title": "Directions, Rotations, and Hyperspheres",
     "section": "",
-    "text": "Given a full Bayesian model \\(p(y, \\theta)\\), the posterior predictive density for new data \\(\\tilde{y}\\) given observed data \\(y\\) is \\[\np(\\tilde{y} \\mid y)\n=\n\\int p(\\tilde{y} \\mid \\theta) \\cdot p(\\theta \\mid y)\n\\, \\textrm{d}\\theta.\n\\] The product under the integral reduces to the joint posterior density \\(p(\\tilde{y}, \\theta \\mid y),\\) so that the integral is simply marginalizing out the parameters \\(\\theta,\\) leaving the predictive density \\(p(\\tilde{y} \\mid y)\\) of future observations given past observations.",
+    "text": "A 24-hour clock naturally represents the progression of time through the day, moving from midnight to noon and back again in one rotation. A point on a circle divided into 24 hours is thus a natural representation for the time of day. Similarly, years cycle through the seasons and return to the season from which they started.\nIn human affairs, temporal effects often arise by convention. These can be modeled directly with ad-hoc predictors for holidays and weekends, or with data normalization back to natural scales for daylight savings time.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Directions, Rotations, and Hyperspheres"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#computing-the-posterior-predictive-distribution",
-    "href": "stan-users-guide/posterior-prediction.html#computing-the-posterior-predictive-distribution",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/latent-discrete.html",
+    "href": "stan-users-guide/latent-discrete.html",
+    "title": "Latent Discrete Parameters",
     "section": "",
-    "text": "The posterior predictive density (or mass) of a prediction \\(\\tilde{y}\\) given observed data \\(y\\) can be computed using \\(M\\) Monte Carlo draws\n\\[\n\\theta^{(m)} \\sim p(\\theta \\mid y)\n\\] from the posterior as \\[\np(\\tilde{y} \\mid y)\n\\approx\n\\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\]\nComputing directly using this formula will lead to underflow in many situations, but the log posterior predictive density, \\(\\log\np(\\tilde{y} \\mid y)\\) may be computed using the stable log sum of exponents function as \\[\\begin{eqnarray*}\n\\log p(\\tilde{y} \\mid y)\n& \\approx &\n\\log \\frac{1}{M} \\sum_{m = 1}^M p(\\tilde{y} \\mid \\theta^{(m)}).\n\\\\[4pt]\n& = &\n- \\log M\n+ \\textrm{log-sum-exp}_{m = 1}^M \\log p(\\tilde{y} \\mid \\theta^{(m)}),\n\\end{eqnarray*}\\] where \\[\n\\textrm{log-sum-exp}_{m = 1}^M v_m\n= \\log \\sum_{m = 1}^M \\exp v_m\n\\] is used to maintain arithmetic precision. See the section on log sum of exponentials for more details.",
+    "text": "Stan does not support sampling discrete parameters. So it is not possible to directly translate BUGS or JAGS models with discrete parameters (i.e., discrete stochastic nodes). Nevertheless, it is possible to code many models that involve bounded discrete parameters by marginalizing out the discrete parameters.1\nThis chapter shows how to code several widely-used models involving latent discrete parameters. The next chapter, the clustering chapter, on clustering models, considers further models involving latent discrete parameters.\n\n\nAlthough it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.\nStandard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.\n\n\n\nThe first example is a model of coal mining disasters in the U.K. for the years 1851–1962.2\n\n\nFonnesbeck et al. (2013, sec. 3.1) provides a Poisson model of disaster \\(D_t\\) in year \\(t\\) with two rate parameters, an early rate (\\(e\\)) and late rate (\\(l\\)), that change at a given point in time \\(s\\). The full model expressed using a latent discrete parameter \\(s\\) is \\[\\begin{align*}\ne   &\\sim  \\textsf{exponential}(r_e) \\\\\nl   &\\sim  \\textsf{exponential}(r_l) \\\\\ns   &\\sim  \\textsf{uniform}(1, T) \\\\\nD_t &\\sim  \\textsf{Poisson}(t &lt; s \\; ? \\; e \\: : \\: l)\n\\end{align*}\\]\nThe last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.3\nIt uses a compact notation involving separating its three arguments by a question mark (?) and a colon (:). The conditional operator is defined by \\[\nc \\; ? \\; x_1 \\: : \\: x_2\n=\n\\begin{cases}\n\\ x_1 & \\quad\\text{if } c \\text{ is true (i.e., non-zero), and} \\\\\n\\ x_2 & \\quad\\text{if } c \\text{ is false (i.e., zero).}\n\\end{cases}\n\\]\n\n\n\nTo code this model in Stan, the discrete parameter \\(s\\) must be marginalized out to produce a model defining the log of the probability function \\(p(e,l,D_t)\\). The full joint probability factors as \\[\\begin{align*}\np(e,l,s,D) &=  p(e) \\, p(l) \\, p(s) \\, p(D \\mid s, e, l) \\\\\n&= \\textsf{exponential}(e \\mid r_e) \\ \\textsf{exponential}(l \\mid r_l) \\,\n    \\textsf{uniform}(s \\mid 1, T) \\\\\n& \\qquad \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nTo marginalize, an alternative factorization into prior and likelihood is used, \\[\np(e,l,D) = p(e,l) \\, p(D \\mid e,l),\n\\]\nwhere the likelihood is defined by marginalizing \\(s\\) as \\[\\begin{align*}\np(D \\mid e,l) &= \\sum_{s=1}^T p(s, D \\mid e,l) \\\\\n&= \\sum_{s=1}^T p(s) \\, p(D \\mid s,e,l) \\\\\n&= \\sum_{s=1}^T \\textsf{uniform}(s \\mid 1,T) \\,\n    \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nStan operates on the log scale and thus requires the log likelihood, \\[\\begin{align*}\n\\log p(D \\mid e,l)\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{s=1}^T\n    \\left( \\log \\textsf{uniform}(s \\mid 1, T) \\vphantom{\\sum_{t=1}^T}\\right. \\\\\n&\\qquad \\left.\n    + \\sum_{t=1}^T \\log \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l)\n\\right),\n\\end{align*}\\] where the log sum of exponents function is defined by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{n=1}^N \\, \\alpha_n =\n\\log \\sum_{n=1}^N \\exp(\\alpha_n).\n\\]\nThe log sum of exponents function allows the model to be coded directly in Stan using the built-in function log_sum_exp, which provides both arithmetic stability and efficiency for mixture model calculations.\n\n\n\nThe Stan program for the change point model is shown in the figure below. The transformed parameter lp[s] stores the quantity \\(\\log p(s, D \\mid e, l)\\).\ndata {\n  real&lt;lower=0&gt; r_e;\n  real&lt;lower=0&gt; r_l;\n\n  int&lt;lower=1&gt; T;\n  array[T] int&lt;lower=0&gt; D;\n}\ntransformed data {\n  real log_unif;\n  log_unif = -log(T);\n}\nparameters {\n  real&lt;lower=0&gt; e;\n  real&lt;lower=0&gt; l;\n}\ntransformed parameters {\n  vector[T] lp;\n  lp = rep_vector(log_unif, T);\n  for (s in 1:T) {\n    for (t in 1:T) {\n      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);\n    }\n  }\n}\nmodel {\n  e ~ exponential(r_e);\n  l ~ exponential(r_l);\n  target += log_sum_exp(lp);\n}\nA change point model in which disaster rates D[t] have one rate, e, before the change point and a different rate, l, after the change point. The change point itself, s, is marginalized out as described in the text.\nAlthough the change-point model is coded directly, the doubly nested loop used for s and t is quadratic in T. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.\ntransformed parameters {\n    vector[T] lp;\n    {\n      vector[T + 1] lp_e;\n      vector[T + 1] lp_l;\n      lp_e[1] = 0;\n      lp_l[1] = 0;\n      for (t in 1:T) {\n        lp_e[t + 1] = lp_e[t] + poisson_lpmf(D[t] | e);\n        lp_l[t + 1] = lp_l[t] + poisson_lpmf(D[t] | l);\n      }\n      lp = rep_vector(log_unif + lp_l[T + 1], T)\n           + head(lp_e, T) - head(lp_l, T);\n    }\n  }\nAs should be obvious from looking at it, it has linear complexity in T rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger T.\nThe key to understanding Wiklendt’s dynamic programming version is to see that head(lp_e) holds the forward values, whereas lp_l[T + 1] - head(lp_l, T) holds the backward values; the clever use of subtraction allows lp_l to be accumulated naturally in the forward direction.\n\n\n\nThis model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.\nThe value of lp for each iteration for each change point is available because it is declared as a transformed parameter. If the value of lp were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.\n\n\n\nThe value of lp[s] in a given iteration is given by \\(\\log\np(s,D \\mid e,l)\\) for the values of the early and late rates, \\(e\\) and \\(l\\), in the iteration. In each iteration after convergence, the early and late disaster rates, \\(e\\) and \\(l\\), are drawn from the posterior \\(p(e,l \\mid D)\\) by MCMC sampling and the associated lp calculated. The value of lp may be normalized to calculate \\(p(s \\mid e,l,D)\\) in each iteration, based on on the current values of \\(e\\) and \\(l\\). Averaging over iterations provides an unnormalized probability estimate of the change point being \\(s\\) (see below for the normalizing constant), \\[\\begin{align*}\np(s \\mid D) &\\propto q(s \\mid D) \\\\\n&= \\frac{1}{M} \\sum_{m=1}^{M} \\exp(\\texttt{lp}[m,s]).\n\\end{align*}\\] where \\(\\texttt{lp}[m,s]\\) represents the value of lp in posterior draw \\(m\\) for change point \\(s\\). By averaging over draws, \\(e\\) and \\(l\\) are themselves marginalized out, and the result has no dependence on a given iteration’s value for \\(e\\) and \\(l\\). A final normalization then produces the quantity of interest, the posterior probability of the change point being \\(s\\) conditioned on the data \\(D\\), \\[\np(s \\mid D) = \\frac{q(s \\mid D)}{\\sum_{s'=1}^T q(s' \\mid D)}.\n\\]\nA plot of the values of \\(\\log p(s \\mid D)\\) computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.\nLog probability of change point being in year, calculated analytically.\n\n\n\nAnalytical change-point posterior\n\n\nThe frequency of change points generated by sampling the discrete change points.\n\n\n\nSampled change-point posterior\n\n\nIn order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of \\(s\\) is roughly 1891.\n\n\n\nThe generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with lp defined as above, the following program draws a random value for s at every iteration.\ngenerated quantities {\n  int&lt;lower=1, upper=T&gt; s;\n  s = categorical_logit_rng(lp);\n}\nA posterior histogram of draws for \\(s\\) is shown on the second change point posterior figure above.\nCompared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for s given by softmax(lp).\n\n\n\nThe discrete sample generated for \\(s\\) can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using lp.\n\n\n\nThere is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, e, m, and l, and two loops for the change point and then one over time, with log densities being stored in a matrix.\nmatrix[T, T] lp;\nlp = rep_matrix(log_unif, T);\nfor (s1 in 1:T) {\n  for (s2 in 1:T) {\n    for (t in 1:T) {\n      lp[s1,s2] = lp[s1,s2]\n        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));\n    }\n  }\n}\nThe matrix can then be converted back to a vector using to_vector before being passed to log_sum_exp.\n\n\n\n\nA widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.\nThe first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.\n\n\nIn the simplest case, a one-stage mark-recapture study produces the following data\n\n\\(M\\) : number of animals marked in first capture,\n\\(C\\) : number animals in second capture, and\n\\(R\\) : number of marked animals in second capture.\n\nThe estimand of interest is\n\n\\(N\\) : number of animals in the population.\n\nDespite the notation, the model will take \\(N\\) to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.\nThe Lincoln-Petersen (Lincoln 1930; Petersen 1896) method for estimating population size is \\[\n\\hat{N} = \\frac{M C}{R}.\n\\]\nThis population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, \\[\nR \\sim \\textsf{binomial}(C, M / N)\n\\] given the total number of animals captured in the second round (\\(C\\)) with a recapture probability of \\(M/N\\), the fraction of the total population \\(N\\) marked in the first round.\ndata {\n  int&lt;lower=0&gt; M;\n  int&lt;lower=0&gt; C;\n  int&lt;lower=0, upper=min(M, C)&gt; R;\n}\nparameters {\n  real&lt;lower=(C - R + M)&gt; N;\n}\nmodel {\n  R ~ binomial(C, M / N);\n}\nA probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on \\(N\\) is necessary to efficiently eliminate impossible values.\nThe probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.\nTo ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for \\(N\\) is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.\nThe one tricky part of the model is the lower bound \\(C - R + M\\) placed on the population size \\(N\\). Values below this bound are impossible because it is otherwise not possible to draw \\(R\\) samples out of the \\(C\\) animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for \\(C\\) implies a variable transform \\(f : (C-R+M,\\infty) \\rightarrow (-\\infty,+\\infty)\\) defined by \\(f(N) = \\log(N - (C - R + M))\\); the reference manual contains full details of all constrained parameter transforms.\n\n\n\nThe Cormack-Jolly-Seber (CJS) model (Cormack 1964; Jolly 1965; Seber 1965) is an open-population model in which the population may change over time due to death; the presentation here draws heavily on Schofield (2007).\nThe basic data are\n\n\\(I\\): number of individuals,\n\\(T\\): number of capture periods, and\n\\(y_{i,t}\\): Boolean indicating if individual \\(i\\) was captured at time \\(t\\).\n\nEach individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.\nThere are two Bernoulli parameters in the model,\n\n\\(\\phi_t\\) : probability that animal alive at time \\(t\\) survives until \\(t + 1\\) and\n\\(p_t\\) : probability that animal alive at time \\(t\\) is captured at time \\(t\\).\n\nThese parameters will both be given uniform priors, but information should be used to tighten these priors in practice.\nThe CJS model also employs a latent discrete parameter \\(z_{i,t}\\) indicating for each individual \\(i\\) whether it is alive at time \\(t\\), distributed as \\[\nz_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t-1} \\; ? \\; 0 \\: : \\: \\phi_{t-1}).\n\\]\nThe conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on \\(z\\) as \\[\ny_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t} \\; ? \\; 0 \\: : \\: p_t).\n\\]\nThe conditional enforces the constraint that dead animals cannot be captured.\n\n\n\nThis subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.\nIn order to ease the marginalization of the latent discrete parameter \\(z_{i,t}\\), the Stan models rely on a derived quantity \\(\\chi_t\\) for the probability that an individual is never captured again if it is alive at time \\(t\\) (if it is dead, the recapture probability is zero). this quantity is defined recursively by \\[\n\\chi_t\n=\n\\begin{cases}\n1 & \\quad\\text{if } t = T \\\\\n(1 - \\phi_t) + \\phi_t (1 - p_{t+1}) \\chi_{t+1}\n  & \\quad\\text{if } t &lt; T\n\\end{cases}\n\\]\nThe base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining \\(\\chi_{t}\\) in terms of \\(\\chi_{t+1}\\) involves two possibilities: (1) not surviving to the next time period, with probability \\((1 - \\phi_t)\\), or (2) surviving to the next time period with probability \\(\\phi_t\\), not being captured in the next time period with probability \\((1 - p_{t+1})\\), and not being captured again after being alive in period \\(t+1\\) with probability \\(\\chi_{t+1}\\).\nWith three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.\n\\[\n\\begin{array}{crclc}\n\\hline\n& \\qquad\\qquad & captures & \\qquad\\qquad &  \\\\\n\\mathrm{profile} & 1 & 2 & 3 & \\mathrm{probability} \\\\\n\\hline\n0 & - & - & - & n/a \\\\\n1 & - & - & + & n/a \\\\\n2 & - & + & - & \\chi_2 \\\\\n3 & - & + & + & \\phi_2 \\, p_3 \\\\\n4 & + & - & - & \\chi_1 \\\\\n5 & + & - & + & \\phi_1 \\, (1 - p_2) \\, \\phi_2 \\, p_3 \\\\\n6 & + & + & - & \\phi_1 \\, p_2 \\, \\chi_2 \\\\\n7 & + & + & + & \\phi_1 \\, p_2 \\, \\phi_2 \\, p_3 \\\\\n\\hline\n\\end{array}\n\\]\nHistory 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.\nBy defining these probabilities in terms of \\(\\chi\\) directly, there is no need for a latent binary parameter indicating whether an animal is alive at time \\(t\\) or not. The definition of \\(\\chi\\) is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model (Schofield 2007).\nThe Stan model defines \\(\\chi\\) as a transformed parameter based on parameters \\(\\phi\\) and \\(p\\). In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using target += rather than being part of a built-in probability function.\nThe following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods\ndata {\n  array[7] int&lt;lower=0&gt; history;\n}\nparameters {\n  array[2] real&lt;lower=0, upper=1&gt; phi;\n  array[3] real&lt;lower=0, upper=1&gt; p;\n}\ntransformed parameters {\n  array[2] real&lt;lower=0, upper=1&gt; chi;\n  chi[2] = (1 - phi[2]) + phi[2] * (1 - p[3]);\n  chi[1] = (1 - phi[1]) + phi[1] * (1 - p[2]) * chi[2];\n}\nmodel {\n  target += history[2] * log(chi[2]);\n  target += history[3] * (log(phi[2]) + log(p[3]));\n  target += history[4] * (log(chi[1]));\n  target += history[5] * (log(phi[1]) + log1m(p[2])\n                            + log(phi[2]) + log(p[3]));\n  target += history[6] * (log(phi[1]) + log(p[2])\n                            + log(chi[2]));\n  target += history[7] * (log(phi[1]) + log(p[2])\n                            + log(phi[2]) + log(p[3]));\n}\ngenerated quantities {\n  real&lt;lower=0, upper=1&gt; beta3;\n  beta3 = phi[2] * p[3];\n}\n\n\nThe parameters \\(\\phi_2\\) and \\(p_3\\), the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, \\(\\beta_3 = \\phi_2 \\, p_3\\), is identified. The Stan model defines beta3 as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the \\(p\\) and \\(\\phi\\) parameters.\n\n\n\n\nThis section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number \\(T\\) of capture events, the number \\(I\\) of individuals, and a boolean flag \\(y_{i,t}\\) indicating if individual \\(i\\) was observed at time \\(t\\). In Stan,\ndata {\n  int&lt;lower=2&gt; T;\n  int&lt;lower=0&gt; I;\n  array[I, T] int&lt;lower=0, upper=1&gt; y;\n}\nThe advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding \\(2^T\\) observation histories for \\(T\\) capture times.\n\n\nThe individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.4\nfunctions {\n  int first_capture(array[] int y_i) {\n    for (k in 1:size(y_i)) {\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  int last_capture(array[] int y_i) {\n    for (k_rev in 0:(size(y_i) - 1)) {\n      int k;\n      k = size(y_i) - k_rev;\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  // ...\n}\nThese two functions are used to define the first and last capture time for each individual in the transformed data block.5\ntransformed data {\n  array[I] int&lt;lower=0, upper=T&gt; first;\n  array[I] int&lt;lower=0, upper=T&gt; last;\n  vector&lt;lower=0, upper=I&gt;[T] n_captured;\n  for (i in 1:I) {\n    first[i] = first_capture(y[i]);\n  }\n  for (i in 1:I) {\n    last[i] = last_capture(y[i]);\n  }\n  n_captured = rep_vector(0, T);\n  for (t in 1:T) {\n    for (i in 1:I) {\n      if (y[i, t]) {\n        n_captured[t] = n_captured[t] + 1;\n      }\n    }\n  }\n}\nThe transformed data block also defines n_captured[t], which is the total number of captures at time t. The variable n_captured is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.\nThe parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector chi, the probability that if an individual is alive at t that it will never be captured again.\nparameters {\n  vector&lt;lower=0, upper=1&gt;[T - 1] phi;\n  vector&lt;lower=0, upper=1&gt;[T] p;\n}\ntransformed parameters {\n  vector&lt;lower=0, upper=1&gt;[T] chi;\n  chi = prob_uncaptured(T, p, phi);\n}\nThe definition of prob_uncaptured, from the functions block, is\nfunctions {\n  // ...\n  vector prob_uncaptured(int T, vector p, vector phi) {\n    vector[T] chi;\n    chi[T] = 1.0;\n    for (t in 1:(T - 1)) {\n      int t_curr;\n      int t_next;\n      t_curr = T - t;\n      t_next = t_curr + 1;\n      chi[t_curr] = (1 - phi[t_curr])\n                     + phi[t_curr]\n                       * (1 - p[t_next])\n                       * chi[t_next];\n    }\n    return chi;\n  }\n}\nThe function definition directly follows the mathematical definition of \\(\\chi_t\\), unrolling the recursion into an iteration and defining the elements of chi from T down to 1.\n\n\n\nGiven the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations q given the parameters p and phi as well as the transformed parameter chi defined in terms of p and phi.\nmodel {\n  for (i in 1:I) {\n    if (first[i] &gt; 0) {\n      for (t in (first[i]+1):last[i]) {\n        1 ~ bernoulli(phi[t - 1]);\n        y[i, t] ~ bernoulli(p[t]);\n      }\n      1 ~ bernoulli(chi[last[i]]);\n    }\n  }\n}\nThe outer loop is over individuals, conditional skipping individuals i which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for first if an individual is never captured.\nThe inner loop for individual i first increments the log probability based on the survival of the individual with probability phi[t - 1]. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.\nIn the inner loop, the observed capture status y[i, t] for individual i at time t has a Bernoulli distribution based on the capture probability p[t] at time t.\nAfter the inner loop, the probability of an animal never being seen again after being observed at time last[i] is included, because last[i] was defined to be the last time period in which animal i was observed.\n\n\n\nAs with the collective model described in the previous subsection, this model does not identify phi[T - 1] and p[T], but does identify their product, beta. Thus beta is defined as a generated quantity to monitor convergence and report.\ngenerated quantities {\n  real beta;\n  // ...\n\n  beta = phi[T - 1] * p[T];\n  // ...\n}\nThe parameter p[1] is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case p[1] would be an unknown initial condition and both phi[T - 1] and p[T] could be identified.\n\n\n\nThe generated quantities also calculates an estimate of the population mean at each time t in the same way as in the simple mark-recapture model as the number of individuals captured at time t divided by the probability of capture at time t. This is done with the elementwise division operation for vectors (./) in the generated quantities block.\ngenerated quantities {\n  // ...\n  vector&lt;lower=0&gt;[T] pop;\n  // ...\n  pop = n_captured ./ p;\n  pop[1] = -1;\n}\n\n\n\nAll individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.\n\n\n\n\n\nAlthough seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.\n\n\nSuppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.\nThere are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.\n\n\n\nHumans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.\n\n\n\nIn this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.\nDawid and Skene (1979) introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.\n\n\nThe data for the model consists of \\(J\\) raters (diagnostic tests), \\(I\\) items (patients), and \\(K\\) categories (condition statuses) to annotate, with \\(y_{i, j} \\in \\{1, \\dotsc, K\\}\\) being the rating provided by rater \\(j\\) for item \\(i\\). In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often \\(K=2\\), with values signaling the presence or absence of the condition.6\nIt is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.\n\n\n\n\nThe model is based on three parameters, the first of which is discrete:\n\n\\(z_i\\) : a value in \\(\\{1, \\dotsc, K\\}\\) indicating the true category of item \\(i\\),\n\\(\\pi\\) : a \\(K\\)-simplex for the prevalence of the \\(K\\) categories in the population, and\n\\(\\theta_{j,k}\\) : a \\(K\\)-simplex for the response of annotator \\(j\\) to an item of true category \\(k\\).\n\n\n\n\nThe true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, \\[\nz_i \\sim \\textsf{categorical}(\\pi).\n\\]\nThe rating \\(y_{i, j}\\) provided for item \\(i\\) by rater \\(j\\) is modeled as a categorical response of rater \\(i\\) to an item of category \\(z_i\\),7 \\[\ny_{i, j} \\sim \\textsf{categorical}(\\theta_{j,\\pi_{z[i]}}).\n\\]\n\n\nDawid and Skene provided maximum likelihood estimates for \\(\\theta\\) and \\(\\pi\\), which allows them to generate probability estimates for each \\(z_i\\).\nTo mimic Dawid and Skene’s maximum likelihood model, the parameters \\(\\theta_{j,k}\\) and \\(\\pi\\) can be given uniform priors over \\(K\\)-simplexes. It is straightforward to generalize to Dirichlet priors, \\[\n\\pi \\sim \\textsf{Dirichlet}(\\alpha)\n\\] and \\[\n\\theta_{j,k} \\sim \\textsf{Dirichlet}(\\beta_k)\n\\] with fixed hyperparameters \\(\\alpha\\) (a vector) and \\(\\beta\\) (a matrix or array of vectors). The prior for \\(\\theta_{j,k}\\) must be allowed to vary in \\(k\\), so that, for instance, \\(\\beta_{k,k}\\) is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.\nBecause there are \\(J\\) coders, it would be natural to extend the model to include a hierarchical prior for \\(\\beta\\) and to partially pool the estimates of coder accuracy and bias.\n\n\n\nBecause the true category parameter \\(z\\) is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as \\[\np(y, \\theta, \\pi) = p(y \\mid \\theta,\\pi) \\, p(\\pi) \\, p(\\theta),\n\\] where \\(p(y \\mid \\theta,\\pi)\\) is derived by marginalizing \\(z\\) out of \\[\np(z, y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\left( \\textsf{categorical}(z_i \\mid \\pi)\n                     \\prod_{j=1}^J\n                     \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, z[i]})\n              \\right).\n\\]\nThis can be done item by item, with \\[\np(y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\sum_{k=1}^K\n  \\left( \\textsf{categorical}(k \\mid \\pi)\n         \\prod_{j=1}^J\n         \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n  \\right).\n\\]\nIn the missing data model, only the observed labels would be used in the inner product.\nDawid and Skene (1979) derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, \\[\\begin{align*}\n\\log p(y \\mid \\theta, \\pi)\n&= \\sum_{i=1}^I \\log \\left( \\sum_{k=1}^K \\exp\n    \\left(\\log \\textsf{categorical}(k \\mid \\pi) \\vphantom{\\sum_{j=1}^J}\\right.\\right.\n    \\left.\\left. + \\ \\sum_{j=1}^J\n           \\log \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n    \\right) \\right),\n\\end{align*}\\] which can be directly coded using Stan’s built-in log_sum_exp function.\n\n\n\n\nThe Stan program for the Dawid and Skene model is provided below (Dawid and Skene 1979).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=1&gt; I;\n  int&lt;lower=1&gt; J;\n\n  array[I, J] int&lt;lower=1, upper=K&gt; y;\n\n  vector&lt;lower=0&gt;[K] alpha;\n  vector&lt;lower=0&gt;[K] beta[K];\n}\nparameters {\n  simplex[K] pi;\n  array[J, K] simplex[K] theta;\n}\ntransformed parameters {\n  array[I] vector[K] log_q_z;\n  for (i in 1:I) {\n    log_q_z[i] = log(pi);\n    for (j in 1:J) {\n      for (k in 1:K) {\n        log_q_z[i, k] = log_q_z[i, k]\n                         + log(theta[j, k, y[i, j]]);\n      }\n    }\n  }\n}\nmodel {\n  pi ~ dirichlet(alpha);\n  for (j in 1:J) {\n    for (k in 1:K) {\n      theta[j, k] ~ dirichlet(beta[k]);\n    }\n  }\n\n  for (i in 1:I) {\n    target += log_sum_exp(log_q_z[i]);\n  }\n}\nThe model marginalizes out the discrete parameter \\(z\\), storing the unnormalized conditional probability \\(\\log q(z_i=k|\\theta,\\pi)\\) in log_q_z[i, k].\nThe Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are \\(\\alpha_k = 3\\) and \\(\\beta_{k,k} = 2.5 K\\) and \\(\\beta_{k,k'} = 1\\) if \\(k \\neq k'\\). Taking \\(\\alpha\\) and \\(\\beta_k\\) to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of Dawid and Skene (1979).\n\n\nThe quantity log_q_z[i] is defined as a transformed parameter. It encodes the (unnormalized) log of \\(p(z_i \\mid \\theta,\n\\pi)\\). Each iteration provides a value conditioned on that iteration’s values for \\(\\theta\\) and \\(\\pi\\). Applying the softmax function to log_q_z[i] provides a simplex corresponding to the probability mass function of \\(z_i\\) in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each \\(z_i\\).\n\n\n\n\n\n\n\nThis section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is not necessary to fit models with discrete parameters using Stan.\nThroughout, the model under consideration consists of both continuous parameters, \\(\\Theta\\), and discrete parameters, \\(Z\\). It is also assumed that \\(Z\\) can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.\n\n\n\nWhen performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating \\(\\mu\\), say, where \\(\\mu = \\mathbb{E}[g(\\Theta, Z)]\\) and \\(g(\\cdot)\\) is an arbitrary function. An example of such a quantity is \\(\\mathbb{E}[\\Theta]\\), the posterior mean of the continuous parameters, where we would take \\(g(\\theta, z) = \\theta\\). To estimate \\(\\mu\\) the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.\nIf both \\(\\Theta\\) and \\(Z\\) were continuous, Stan could be used to sample \\(M\\) draws from the joint posterior \\(p_{\\Theta, Z}(\\theta, z)\\) and then estimate \\(\\mu\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M {g(\\theta^{(i)}, z^{(i)})}.\n\\] Given \\(Z\\) is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out \\(Z\\) from the joint posterior to give the marginalized posterior \\(p_\\Theta(\\theta)\\). This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws \\(\\{\\theta^{(i)}\\}_{i = 1}^M\\) from the marginalized posterior.\nUsing only these draws, how can we estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\)? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function \\(g\\) of interest, we compute \\[\nh(\\Theta) = \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n\\] and then estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M h(\\theta^{(i)}).\n\\] This estimator is justified by the law of iterated expectation, the fact that \\[\n\\mathbb{E}[h(\\Theta)] = \\mathbb{E}[\\mathbb{E}[g(\\Theta, Z)] \\mid \\Theta] = \\mathbb{E}[g(\\Theta, Z)] = \\mu.\n\\] Using this marginalized estimator provides a way to estimate the expectation of any function \\(g(\\cdot)\\) for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\).\n\n\n\nFortunately, the discrete nature of \\(Z\\) makes evaluating \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\) easy. The function \\(h(\\Theta)\\) can be written as: \\[\nh(\\Theta)\n= \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n= \\sum_{k} g(\\Theta, k) \\Pr[Z = k \\mid \\Theta],\n\\] where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, \\(\\Pr[Z = k \\mid \\Theta]\\). More detail on how this quantity can be calculated is included below. Note that if \\(Z\\) takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.\nWhen \\(g(\\theta, z)\\) is a function of either \\(\\theta\\) or \\(z\\) only, the above formula simplifies further.\nIn the first case, where \\(g(\\theta, z) = g(\\theta)\\), we have: \\[\\begin{align*}\nh(\\Theta)\n&= \\sum_{k} g(\\Theta) \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta) \\sum_{k} \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta).\n\\end{align*}\\] This means that we can estimate \\(\\mathbb{E}[g(\\Theta)]\\) with the standard, seemingly unconditional, estimator: \\[\n\\frac{1}{M} \\sum_{i = 1}^M g(\\theta^{(i)}).\n\\] Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.\nIn the second case, where \\(g(\\theta, z) = g(z)\\), the conditional expectation instead simplifies as follows: \\[\nh(\\Theta) = \\sum_{k} g(k) \\Pr[Z = k \\mid \\Theta].\n\\] An important special case of this result is when \\(g(\\theta, z) = \\textrm{I}(z = k)\\), where \\(\\textrm{I}\\) is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable \\(Z\\), since \\(\\mathbb{E}[\\textrm{I}(Z = k)] = \\Pr[Z = k]\\). In this case, \\[\nh(\\Theta)\n= \\sum_{k} \\textrm{I}(z = k) \\Pr[Z = k \\mid \\Theta]\n= \\Pr[Z = k \\mid \\Theta].\n\\] The quantity \\(\\Pr[Z = k]\\) can therefore be estimated with: \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}].\n\\] When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, \\(Y\\). That is, we are really estimating \\(\\Pr[Z = k \\mid Y]\\) with \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y].\n\\] This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us \\[\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y]\n= \\frac{\\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}\n{\\sum_{k = 1}^K \\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}.\n\\] Here, \\(\\Pr[Y \\mid \\Theta = \\theta^{(i)}, Z = k]\\) is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.\nOther than the use of Bayes’s theorem, \\(\\Pr[Z = k \\mid \\theta = \\theta^{(i)}, Y]\\) can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the Dawid–Skene model).\nFor a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see Pullin, Gurrin, and Vukcevic (2021).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#sampling-from-the-posterior-predictive-distribution",
-    "href": "stan-users-guide/posterior-prediction.html#sampling-from-the-posterior-predictive-distribution",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/latent-discrete.html#rao-blackwell.section",
+    "href": "stan-users-guide/latent-discrete.html#rao-blackwell.section",
+    "title": "Latent Discrete Parameters",
     "section": "",
-    "text": "Given draws from the posterior \\(\\theta^{(m)} \\sim p(\\theta \\mid y),\\) draws from the posterior predictive \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid y)\\) can be generated by randomly generating from the sampling distribution with the parameter draw plugged in, \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\theta^{(m)}).\n\\]\nRandomly drawing \\(\\tilde{y}\\) from the data model is critical because there are two forms of uncertainty in posterior predictive quantities, aleatoric uncertainty and epistemic uncertainty. Epistemic uncertainty arises because \\(\\theta\\) is unknown and estimated based only on a finite sample of data \\(y\\). Aleatoric uncertainty arises because even a known value of \\(\\theta\\) leads to uncertainty about new \\(\\tilde{y}\\) as described by the data model \\(p(\\tilde{y} \\mid \\theta)\\). Both forms of uncertainty show up in the factored form of the posterior predictive distribution, \\[\np(\\tilde{y} \\mid y)\n=\n\\int\n\\underbrace{p(\\tilde{y} \\mid \\theta)}_{\\begin{array}{l}\n                                         \\textrm{aleatoric}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\cdot \\underbrace{p(\\theta \\mid y)}_{\\begin{array}{l}\n                                         \\textrm{epistemic}\n                                         \\\\[-2pt] \\textrm{uncertainty}\n                                       \\end{array}}\n\\, \\textrm{d}\\theta.\n\\]",
+    "text": "Although it requires some algebra on the joint probability function, a pleasant byproduct of the required calculations is the posterior expectation of the marginalized variable, which is often the quantity of interest for a model. This allows far greater exploration of the tails of the distribution as well as more efficient sampling on an iteration-by-iteration basis because the expectation at all possible values is being used rather than itself being estimated through sampling a discrete parameter.\nStandard optimization algorithms, including expectation maximization (EM), are often provided in applied statistics papers to describe maximum likelihood estimation algorithms. Such derivations provide exactly the marginalization needed for coding the model in Stan.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan",
-    "href": "stan-users-guide/posterior-prediction.html#posterior-predictive-simulation-in-stan",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/latent-discrete.html#change-point.section",
+    "href": "stan-users-guide/latent-discrete.html#change-point.section",
+    "title": "Latent Discrete Parameters",
     "section": "",
-    "text": "Posterior predictive quantities can be coded in Stan using the generated quantities block.\n\n\nFor example, consider a simple Poisson model for count data with a rate parameter \\(\\lambda &gt; 0\\) having a gamma-distributed prior, \\[\n\\lambda \\sim \\textrm{gamma}(1, 1).\n\\] The \\(N\\) observations \\(y_1, \\ldots, y_N\\) are modeled as Poisson distributed, \\[\ny_n \\sim \\textrm{poisson}(\\lambda).\n\\]\n\n\n\nThe following Stan program defines a variable for \\(\\tilde{y}\\) by random number generation in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  lambda ~ gamma(1, 1);\n  y ~ poisson(lambda);\n}\ngenerated quantities {\n  int&lt;lower=0&gt; y_tilde = poisson_rng(lambda);\n}\nThe random draw from the data model for \\(\\tilde{y}\\) is coded using Stan’s Poisson random number generator in the generated quantities block. This accounts for the aleatoric component of the uncertainty; Stan’s posterior sampler will account for the epistemic uncertainty, generating a new \\(\\tilde{y}^{(m)} \\sim p(y \\mid\n\\lambda^{(m)})\\) for each posterior draw \\(\\lambda^{(m)} \\sim p(\\theta\n\\mid y).\\)\nThe posterior draws \\(\\tilde{y}^{(m)}\\) may be used to estimate the expected value of \\(\\tilde{y}\\) or any of its quantiles or posterior intervals, as well as event probabilities involving \\(\\tilde{y}\\). In general, \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\) may be evaluated as \\[\n\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\n\\approx \\frac{1}{M} \\sum_{m=1}^M f(\\tilde{y}^{(m)}, \\theta^{(m)}),\n\\] which is just the posterior mean of \\(f(\\tilde{y}, \\theta).\\) This quantity is computed by Stan if the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to a variable in the generated quantities block. That is, if we have\ngenerated quantities {\n  real f_val = f(y_tilde, theta);\n  // ...\n}\nwhere the value of \\(f(\\tilde{y}, \\theta)\\) is assigned to variable f_val, then the posterior mean of f_val will be the expectation \\(\\mathbb{E}[f(\\tilde{y}, \\theta) \\mid y]\\).\n\n\n\nThe gamma distribution is the conjugate prior distribution for the Poisson distribution, so the posterior density \\(p(\\lambda \\mid y)\\) will also follow a gamma distribution.\nBecause the posterior follows a gamma distribution and the sampling distribution is Poisson, the posterior predictive \\(p(\\tilde{y} \\mid\ny)\\) will follow a negative binomial distribution, because the negative binomial is defined as a compound gamma-Poisson. That is, \\(y \\sim\n\\textrm{negative-binomial}(\\alpha, \\beta)\\) if \\(\\lambda \\sim\n\\textrm{gamma}(\\alpha, \\beta)\\) and \\(y \\sim \\textrm{poisson}(\\lambda).\\) Rather than marginalizing out the rate parameter \\(\\lambda\\) analytically as can be done to define the negative binomial probability mass function, the rate \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is sampled from the posterior and then used to generate a draw of \\(\\tilde{y}^{(m)} \\sim p(y \\mid \\lambda^{(m)}).\\)",
+    "text": "The first example is a model of coal mining disasters in the U.K. for the years 1851–1962.2\n\n\nFonnesbeck et al. (2013, sec. 3.1) provides a Poisson model of disaster \\(D_t\\) in year \\(t\\) with two rate parameters, an early rate (\\(e\\)) and late rate (\\(l\\)), that change at a given point in time \\(s\\). The full model expressed using a latent discrete parameter \\(s\\) is \\[\\begin{align*}\ne   &\\sim  \\textsf{exponential}(r_e) \\\\\nl   &\\sim  \\textsf{exponential}(r_l) \\\\\ns   &\\sim  \\textsf{uniform}(1, T) \\\\\nD_t &\\sim  \\textsf{Poisson}(t &lt; s \\; ? \\; e \\: : \\: l)\n\\end{align*}\\]\nThe last line uses the conditional operator (also known as the ternary operator), which is borrowed from C and related languages. The conditional operator has the same behavior as its counterpart in C++.3\nIt uses a compact notation involving separating its three arguments by a question mark (?) and a colon (:). The conditional operator is defined by \\[\nc \\; ? \\; x_1 \\: : \\: x_2\n=\n\\begin{cases}\n\\ x_1 & \\quad\\text{if } c \\text{ is true (i.e., non-zero), and} \\\\\n\\ x_2 & \\quad\\text{if } c \\text{ is false (i.e., zero).}\n\\end{cases}\n\\]\n\n\n\nTo code this model in Stan, the discrete parameter \\(s\\) must be marginalized out to produce a model defining the log of the probability function \\(p(e,l,D_t)\\). The full joint probability factors as \\[\\begin{align*}\np(e,l,s,D) &=  p(e) \\, p(l) \\, p(s) \\, p(D \\mid s, e, l) \\\\\n&= \\textsf{exponential}(e \\mid r_e) \\ \\textsf{exponential}(l \\mid r_l) \\,\n    \\textsf{uniform}(s \\mid 1, T) \\\\\n& \\qquad \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nTo marginalize, an alternative factorization into prior and likelihood is used, \\[\np(e,l,D) = p(e,l) \\, p(D \\mid e,l),\n\\]\nwhere the likelihood is defined by marginalizing \\(s\\) as \\[\\begin{align*}\np(D \\mid e,l) &= \\sum_{s=1}^T p(s, D \\mid e,l) \\\\\n&= \\sum_{s=1}^T p(s) \\, p(D \\mid s,e,l) \\\\\n&= \\sum_{s=1}^T \\textsf{uniform}(s \\mid 1,T) \\,\n    \\prod_{t=1}^T \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l).\n\\end{align*}\\]\nStan operates on the log scale and thus requires the log likelihood, \\[\\begin{align*}\n\\log p(D \\mid e,l)\n&= \\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{s=1}^T\n    \\left( \\log \\textsf{uniform}(s \\mid 1, T) \\vphantom{\\sum_{t=1}^T}\\right. \\\\\n&\\qquad \\left.\n    + \\sum_{t=1}^T \\log \\textsf{Poisson}(D_t \\mid t &lt; s \\; ? \\; e \\: : \\: l)\n\\right),\n\\end{align*}\\] where the log sum of exponents function is defined by \\[\n\\texttt{log}\\mathtt{\\_}\\texttt{sum}\\mathtt{\\_}\\texttt{exp}_{n=1}^N \\, \\alpha_n =\n\\log \\sum_{n=1}^N \\exp(\\alpha_n).\n\\]\nThe log sum of exponents function allows the model to be coded directly in Stan using the built-in function log_sum_exp, which provides both arithmetic stability and efficiency for mixture model calculations.\n\n\n\nThe Stan program for the change point model is shown in the figure below. The transformed parameter lp[s] stores the quantity \\(\\log p(s, D \\mid e, l)\\).\ndata {\n  real&lt;lower=0&gt; r_e;\n  real&lt;lower=0&gt; r_l;\n\n  int&lt;lower=1&gt; T;\n  array[T] int&lt;lower=0&gt; D;\n}\ntransformed data {\n  real log_unif;\n  log_unif = -log(T);\n}\nparameters {\n  real&lt;lower=0&gt; e;\n  real&lt;lower=0&gt; l;\n}\ntransformed parameters {\n  vector[T] lp;\n  lp = rep_vector(log_unif, T);\n  for (s in 1:T) {\n    for (t in 1:T) {\n      lp[s] = lp[s] + poisson_lpmf(D[t] | t &lt; s ? e : l);\n    }\n  }\n}\nmodel {\n  e ~ exponential(r_e);\n  l ~ exponential(r_l);\n  target += log_sum_exp(lp);\n}\nA change point model in which disaster rates D[t] have one rate, e, before the change point and a different rate, l, after the change point. The change point itself, s, is marginalized out as described in the text.\nAlthough the change-point model is coded directly, the doubly nested loop used for s and t is quadratic in T. Luke Wiklendt pointed out that a linear alternative can be achieved by the use of dynamic programming similar to the forward-backward algorithm for Hidden Markov models; he submitted a slight variant of the following code to replace the transformed parameters block of the above Stan program.\ntransformed parameters {\n    vector[T] lp;\n    {\n      vector[T + 1] lp_e;\n      vector[T + 1] lp_l;\n      lp_e[1] = 0;\n      lp_l[1] = 0;\n      for (t in 1:T) {\n        lp_e[t + 1] = lp_e[t] + poisson_lpmf(D[t] | e);\n        lp_l[t + 1] = lp_l[t] + poisson_lpmf(D[t] | l);\n      }\n      lp = rep_vector(log_unif + lp_l[T + 1], T)\n           + head(lp_e, T) - head(lp_l, T);\n    }\n  }\nAs should be obvious from looking at it, it has linear complexity in T rather than quadratic. The result for the mining-disaster data is about 20 times faster; the improvement will be greater for larger T.\nThe key to understanding Wiklendt’s dynamic programming version is to see that head(lp_e) holds the forward values, whereas lp_l[T + 1] - head(lp_l, T) holds the backward values; the clever use of subtraction allows lp_l to be accumulated naturally in the forward direction.\n\n\n\nThis model is easy to fit using MCMC with NUTS in its default configuration. Convergence is fast and sampling produces roughly one effective sample every two iterations. Because it is a relatively small model (the inner double loop over time is roughly 20,000 steps), it is fast.\nThe value of lp for each iteration for each change point is available because it is declared as a transformed parameter. If the value of lp were not of interest, it could be coded as a local variable in the model block and thus avoid the I/O overhead of saving values every iteration.\n\n\n\nThe value of lp[s] in a given iteration is given by \\(\\log\np(s,D \\mid e,l)\\) for the values of the early and late rates, \\(e\\) and \\(l\\), in the iteration. In each iteration after convergence, the early and late disaster rates, \\(e\\) and \\(l\\), are drawn from the posterior \\(p(e,l \\mid D)\\) by MCMC sampling and the associated lp calculated. The value of lp may be normalized to calculate \\(p(s \\mid e,l,D)\\) in each iteration, based on on the current values of \\(e\\) and \\(l\\). Averaging over iterations provides an unnormalized probability estimate of the change point being \\(s\\) (see below for the normalizing constant), \\[\\begin{align*}\np(s \\mid D) &\\propto q(s \\mid D) \\\\\n&= \\frac{1}{M} \\sum_{m=1}^{M} \\exp(\\texttt{lp}[m,s]).\n\\end{align*}\\] where \\(\\texttt{lp}[m,s]\\) represents the value of lp in posterior draw \\(m\\) for change point \\(s\\). By averaging over draws, \\(e\\) and \\(l\\) are themselves marginalized out, and the result has no dependence on a given iteration’s value for \\(e\\) and \\(l\\). A final normalization then produces the quantity of interest, the posterior probability of the change point being \\(s\\) conditioned on the data \\(D\\), \\[\np(s \\mid D) = \\frac{q(s \\mid D)}{\\sum_{s'=1}^T q(s' \\mid D)}.\n\\]\nA plot of the values of \\(\\log p(s \\mid D)\\) computed using Stan 2.4’s default MCMC implementation is shown in the posterior plot.\nLog probability of change point being in year, calculated analytically.\n\n\n\nAnalytical change-point posterior\n\n\nThe frequency of change points generated by sampling the discrete change points.\n\n\n\nSampled change-point posterior\n\n\nIn order their range of estimates be visible, the first plot is on the log scale and the second plot on the linear scale; note the narrower range of years in the second plot resulting from sampling. The posterior mean of \\(s\\) is roughly 1891.\n\n\n\nThe generated quantities block may be used to draw discrete parameter values using the built-in pseudo-random number generators. For example, with lp defined as above, the following program draws a random value for s at every iteration.\ngenerated quantities {\n  int&lt;lower=1, upper=T&gt; s;\n  s = categorical_logit_rng(lp);\n}\nA posterior histogram of draws for \\(s\\) is shown on the second change point posterior figure above.\nCompared to working in terms of expectations, discrete sampling is highly inefficient, especially for tails of distributions, so this approach should only be used if draws from a distribution are explicitly required. Otherwise, expectations should be computed in the generated quantities block based on the posterior distribution for s given by softmax(lp).\n\n\n\nThe discrete sample generated for \\(s\\) can be used to calculate covariance with other parameters. Although the sampling approach is straightforward, it is more statistically efficient (in the sense of requiring far fewer iterations for the same degree of accuracy) to calculate these covariances in expectation using lp.\n\n\n\nThere is no obstacle in principle to allowing multiple change points. The only issue is that computation increases from linear to quadratic in marginalizing out two change points, cubic for three change points, and so on. There are three parameters, e, m, and l, and two loops for the change point and then one over time, with log densities being stored in a matrix.\nmatrix[T, T] lp;\nlp = rep_matrix(log_unif, T);\nfor (s1 in 1:T) {\n  for (s2 in 1:T) {\n    for (t in 1:T) {\n      lp[s1,s2] = lp[s1,s2]\n        + poisson_lpmf(D[t] | t &lt; s1 ? e : (t &lt; s2 ? m : l));\n    }\n  }\n}\nThe matrix can then be converted back to a vector using to_vector before being passed to log_sum_exp.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#posterior-prediction-for-regressions",
-    "href": "stan-users-guide/posterior-prediction.html#posterior-prediction-for-regressions",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/latent-discrete.html#mark-recapture-models",
+    "href": "stan-users-guide/latent-discrete.html#mark-recapture-models",
+    "title": "Latent Discrete Parameters",
     "section": "",
-    "text": "Consider a regression with a single predictor \\(x_n\\) for the training outcome \\(y_n\\) and \\(\\tilde{x}_n\\) for the test outcome \\(\\tilde{y}_n.\\) Without considering the parametric form of any of the distributions, the posterior predictive distribution for a general regression in \\[\\begin{eqnarray}\np(\\tilde{y} \\mid \\tilde{x}, y, x)\n& = & \\int p(\\tilde{y} \\mid \\tilde{x}, \\theta) \\cdot p(\\theta \\mid y, x) \\,\n\\textrm{d}\\theta\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m=1}^M \\, p(\\tilde{y} \\mid \\tilde{x}, \\theta^{(m)}),\n\\end{eqnarray}\\] where \\(\\theta^{(m)} \\sim p(\\theta \\mid x, y).\\)\n\n\n\nThe following program defines a Poisson regression with a single predictor. These predictors are all coded as data, as are their sizes. Only the observed \\(y\\) values are coded as data. The predictive quantities \\(\\tilde{y}\\) appear in the generated quantities block, where they are generated by random number generation.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0&gt; y;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ poisson_log(alpha + beta * x);\n  { alpha, beta } ~ normal(0, 1);\n}\ngenerated quantities {\n  array[N_tilde] int&lt;lower=0&gt; y_tilde\n    = poisson_log_rng(alpha + beta * x_tilde);\n}\nThe Poisson distributions in both the model and generated quantities block are coded using the log rate as a parameter (that’s poisson_log vs. poisson, with the suffixes defining the scale of the parameter). The regression coefficients, an intercept alpha and slope beta, are given standard normal priors.\nIn the model block, the log rate for the Poisson is a linear function of the training data \\(x\\), whereas in the generated quantities block it is a function of the test data \\(\\tilde{x}\\). Because the generated quantities block does not affect the posterior draws, the model fits \\(\\alpha\\) and \\(\\beta\\) using only the training data, reserving \\(\\tilde{x}\\) to generate \\(\\tilde{y}.\\)\nThe result from running Stan is a predictive sample \\(\\tilde{y}^{(1)},\n\\ldots \\tilde{y}^{(M)}\\) where each \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y}\n\\mid \\tilde{x}, x, y).\\)\nThe mean of the posterior predictive distribution is the expected value \\[\\begin{align}\n\\mathbb{E}[\\tilde{y} \\mid \\tilde{x}, x, y]\n& =\n\\int\n\\tilde{y}\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, \\theta)\n\\cdot p(\\theta \\mid x, y)\n\\, \\textrm{d}\\theta\n\\\\[4pt]\n& \\approx \\frac{1}{M} \\sum_{m = 1}^M \\tilde{y}^{(m)},\n\\end{align}\\] where the \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y)\\) are drawn from the posterior predictive distribution. Thus the posterior mean of y_tilde[n] after running Stan is the expected value of \\(\\tilde{y}_n\\) conditioned on the training data \\(x, y\\) and predictor \\(\\tilde{x}_n.\\) This is the Bayesian estimate for \\(\\tilde{y}\\) with minimum expected squared error. The posterior draws can also be used to estimate quantiles for the median and any posterior intervals of interest for \\(\\tilde{y}\\), as well as covariance of the \\(\\tilde{y_n}.\\) The posterior draws \\(\\tilde{y}^{(m)}\\) may also be used to estimate predictive event probabilities, such as \\(\\Pr[\\tilde{y}_1 &gt; 0]\\) or \\(\\Pr[\\prod_{n =\n1}^{\\tilde{N}}(\\tilde{y_n}) &gt; 1],\\) as expectations of indicator functions.\nAll of this can be carried out by running Stan only a single time to draw a single sample of \\(M\\) draws, \\[\n\\tilde{y}^{(1)}, \\ldots, \\tilde{y}^{(M)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y).\n\\] It’s only when moving to cross-validation where multiple runs are required.",
+    "text": "A widely applied field method in ecology is to capture (or sight) animals, mark them (e.g., by tagging), then release them. This process is then repeated one or more times, and is often done for populations on an ongoing basis. The resulting data may be used to estimate population size.\nThe first subsection describes a simple mark-recapture model that does not involve any latent discrete parameters. The following subsections describes the Cormack-Jolly-Seber model, which involves latent discrete parameters for animal death.\n\n\nIn the simplest case, a one-stage mark-recapture study produces the following data\n\n\\(M\\) : number of animals marked in first capture,\n\\(C\\) : number animals in second capture, and\n\\(R\\) : number of marked animals in second capture.\n\nThe estimand of interest is\n\n\\(N\\) : number of animals in the population.\n\nDespite the notation, the model will take \\(N\\) to be a continuous parameter; just because the population must be finite doesn’t mean the parameter representing it must be. The parameter will be used to produce a real-valued estimate of the population size.\nThe Lincoln-Petersen (Lincoln 1930; Petersen 1896) method for estimating population size is \\[\n\\hat{N} = \\frac{M C}{R}.\n\\]\nThis population estimate would arise from a probabilistic model in which the number of recaptured animals is distributed binomially, \\[\nR \\sim \\textsf{binomial}(C, M / N)\n\\] given the total number of animals captured in the second round (\\(C\\)) with a recapture probability of \\(M/N\\), the fraction of the total population \\(N\\) marked in the first round.\ndata {\n  int&lt;lower=0&gt; M;\n  int&lt;lower=0&gt; C;\n  int&lt;lower=0, upper=min(M, C)&gt; R;\n}\nparameters {\n  real&lt;lower=(C - R + M)&gt; N;\n}\nmodel {\n  R ~ binomial(C, M / N);\n}\nA probabilistic formulation of the Lincoln-Petersen estimator for population size based on data from a one-step mark-recapture study. The lower bound on \\(N\\) is necessary to efficiently eliminate impossible values.\nThe probabilistic variant of the Lincoln-Petersen estimator can be directly coded in Stan as shown in the Lincon-Petersen model figure. The Lincoln-Petersen estimate is the maximum likelihood estimate (MLE) for this model.\nTo ensure the MLE is the Lincoln-Petersen estimate, an improper uniform prior for \\(N\\) is used; this could (and should) be replaced with a more informative prior if possible, based on knowledge of the population under study.\nThe one tricky part of the model is the lower bound \\(C - R + M\\) placed on the population size \\(N\\). Values below this bound are impossible because it is otherwise not possible to draw \\(R\\) samples out of the \\(C\\) animals recaptured. Implementing this lower bound is necessary to ensure sampling and optimization can be carried out in an unconstrained manner with unbounded support for parameters on the transformed (unconstrained) space. The lower bound in the declaration for \\(C\\) implies a variable transform \\(f : (C-R+M,\\infty) \\rightarrow (-\\infty,+\\infty)\\) defined by \\(f(N) = \\log(N - (C - R + M))\\); the reference manual contains full details of all constrained parameter transforms.\n\n\n\nThe Cormack-Jolly-Seber (CJS) model (Cormack 1964; Jolly 1965; Seber 1965) is an open-population model in which the population may change over time due to death; the presentation here draws heavily on Schofield (2007).\nThe basic data are\n\n\\(I\\): number of individuals,\n\\(T\\): number of capture periods, and\n\\(y_{i,t}\\): Boolean indicating if individual \\(i\\) was captured at time \\(t\\).\n\nEach individual is assumed to have been captured at least once because an individual only contributes information conditionally after they have been captured the first time.\nThere are two Bernoulli parameters in the model,\n\n\\(\\phi_t\\) : probability that animal alive at time \\(t\\) survives until \\(t + 1\\) and\n\\(p_t\\) : probability that animal alive at time \\(t\\) is captured at time \\(t\\).\n\nThese parameters will both be given uniform priors, but information should be used to tighten these priors in practice.\nThe CJS model also employs a latent discrete parameter \\(z_{i,t}\\) indicating for each individual \\(i\\) whether it is alive at time \\(t\\), distributed as \\[\nz_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t-1} \\; ? \\; 0 \\: : \\: \\phi_{t-1}).\n\\]\nThe conditional prevents the model positing zombies; once an animal is dead, it stays dead. The data distribution is then simple to express conditional on \\(z\\) as \\[\ny_{i,t} \\sim \\mathsf{Bernoulli}(z_{i,t} \\; ? \\; 0 \\: : \\: p_t).\n\\]\nThe conditional enforces the constraint that dead animals cannot be captured.\n\n\n\nThis subsection presents an implementation of the model in terms of counts for different history profiles for individuals over three capture times. It assumes exchangeability of the animals in that each is assigned the same capture and survival probabilities.\nIn order to ease the marginalization of the latent discrete parameter \\(z_{i,t}\\), the Stan models rely on a derived quantity \\(\\chi_t\\) for the probability that an individual is never captured again if it is alive at time \\(t\\) (if it is dead, the recapture probability is zero). this quantity is defined recursively by \\[\n\\chi_t\n=\n\\begin{cases}\n1 & \\quad\\text{if } t = T \\\\\n(1 - \\phi_t) + \\phi_t (1 - p_{t+1}) \\chi_{t+1}\n  & \\quad\\text{if } t &lt; T\n\\end{cases}\n\\]\nThe base case arises because if an animal was captured in the last time period, the probability it is never captured again is 1 because there are no more capture periods. The recursive case defining \\(\\chi_{t}\\) in terms of \\(\\chi_{t+1}\\) involves two possibilities: (1) not surviving to the next time period, with probability \\((1 - \\phi_t)\\), or (2) surviving to the next time period with probability \\(\\phi_t\\), not being captured in the next time period with probability \\((1 - p_{t+1})\\), and not being captured again after being alive in period \\(t+1\\) with probability \\(\\chi_{t+1}\\).\nWith three capture times, there are eight captured/not-captured profiles an individual may have. These may be naturally coded as binary numbers as follows.\n\\[\n\\begin{array}{crclc}\n\\hline\n& \\qquad\\qquad & captures & \\qquad\\qquad &  \\\\\n\\mathrm{profile} & 1 & 2 & 3 & \\mathrm{probability} \\\\\n\\hline\n0 & - & - & - & n/a \\\\\n1 & - & - & + & n/a \\\\\n2 & - & + & - & \\chi_2 \\\\\n3 & - & + & + & \\phi_2 \\, p_3 \\\\\n4 & + & - & - & \\chi_1 \\\\\n5 & + & - & + & \\phi_1 \\, (1 - p_2) \\, \\phi_2 \\, p_3 \\\\\n6 & + & + & - & \\phi_1 \\, p_2 \\, \\chi_2 \\\\\n7 & + & + & + & \\phi_1 \\, p_2 \\, \\phi_2 \\, p_3 \\\\\n\\hline\n\\end{array}\n\\]\nHistory 0, for animals that are never captured, is unobservable because only animals that are captured are observed. History 1, for animals that are only captured in the last round, provides no information for the CJS model, because capture/non-capture status is only informative when conditioned on earlier captures. For the remaining cases, the contribution to the likelihood is provided in the final column.\nBy defining these probabilities in terms of \\(\\chi\\) directly, there is no need for a latent binary parameter indicating whether an animal is alive at time \\(t\\) or not. The definition of \\(\\chi\\) is typically used to define the likelihood (i.e., marginalize out the latent discrete parameter) for the CJS model (Schofield 2007).\nThe Stan model defines \\(\\chi\\) as a transformed parameter based on parameters \\(\\phi\\) and \\(p\\). In the model block, the log probability is incremented for each history based on its count. This second step is similar to collecting Bernoulli observations into a binomial or categorical observations into a multinomial, only it is coded directly in the Stan program using target += rather than being part of a built-in probability function.\nThe following is the Stan program for the Cormack-Jolly-Seber mark-recapture model that considers counts of individuals with observation histories of being observed or not in three capture periods\ndata {\n  array[7] int&lt;lower=0&gt; history;\n}\nparameters {\n  array[2] real&lt;lower=0, upper=1&gt; phi;\n  array[3] real&lt;lower=0, upper=1&gt; p;\n}\ntransformed parameters {\n  array[2] real&lt;lower=0, upper=1&gt; chi;\n  chi[2] = (1 - phi[2]) + phi[2] * (1 - p[3]);\n  chi[1] = (1 - phi[1]) + phi[1] * (1 - p[2]) * chi[2];\n}\nmodel {\n  target += history[2] * log(chi[2]);\n  target += history[3] * (log(phi[2]) + log(p[3]));\n  target += history[4] * (log(chi[1]));\n  target += history[5] * (log(phi[1]) + log1m(p[2])\n                            + log(phi[2]) + log(p[3]));\n  target += history[6] * (log(phi[1]) + log(p[2])\n                            + log(chi[2]));\n  target += history[7] * (log(phi[1]) + log(p[2])\n                            + log(phi[2]) + log(p[3]));\n}\ngenerated quantities {\n  real&lt;lower=0, upper=1&gt; beta3;\n  beta3 = phi[2] * p[3];\n}\n\n\nThe parameters \\(\\phi_2\\) and \\(p_3\\), the probability of death at time 2 and probability of capture at time 3 are not identifiable, because both may be used to account for lack of capture at time 3. Their product, \\(\\beta_3 = \\phi_2 \\, p_3\\), is identified. The Stan model defines beta3 as a generated quantity. Unidentified parameters pose a problem for Stan’s samplers’ adaptation. Although the problem posed for adaptation is mild here because the parameters are bounded and thus have proper uniform priors, it would be better to formulate an identified parameterization. One way to do this would be to formulate a hierarchical model for the \\(p\\) and \\(\\phi\\) parameters.\n\n\n\n\nThis section presents a version of the Cormack-Jolly-Seber (CJS) model cast at the individual level rather than collectively as in the previous subsection. It also extends the model to allow an arbitrary number of time periods. The data will consist of the number \\(T\\) of capture events, the number \\(I\\) of individuals, and a boolean flag \\(y_{i,t}\\) indicating if individual \\(i\\) was observed at time \\(t\\). In Stan,\ndata {\n  int&lt;lower=2&gt; T;\n  int&lt;lower=0&gt; I;\n  array[I, T] int&lt;lower=0, upper=1&gt; y;\n}\nThe advantages to the individual-level model is that it becomes possible to add individual “random effects” that affect survival or capture probability, as well as to avoid the combinatorics involved in unfolding \\(2^T\\) observation histories for \\(T\\) capture times.\n\n\nThe individual CJS model is written involves several function definitions. The first two are used in the transformed data block to compute the first and last time period in which an animal was captured.4\nfunctions {\n  int first_capture(array[] int y_i) {\n    for (k in 1:size(y_i)) {\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  int last_capture(array[] int y_i) {\n    for (k_rev in 0:(size(y_i) - 1)) {\n      int k;\n      k = size(y_i) - k_rev;\n      if (y_i[k]) {\n        return k;\n      }\n    }\n    return 0;\n  }\n  // ...\n}\nThese two functions are used to define the first and last capture time for each individual in the transformed data block.5\ntransformed data {\n  array[I] int&lt;lower=0, upper=T&gt; first;\n  array[I] int&lt;lower=0, upper=T&gt; last;\n  vector&lt;lower=0, upper=I&gt;[T] n_captured;\n  for (i in 1:I) {\n    first[i] = first_capture(y[i]);\n  }\n  for (i in 1:I) {\n    last[i] = last_capture(y[i]);\n  }\n  n_captured = rep_vector(0, T);\n  for (t in 1:T) {\n    for (i in 1:I) {\n      if (y[i, t]) {\n        n_captured[t] = n_captured[t] + 1;\n      }\n    }\n  }\n}\nThe transformed data block also defines n_captured[t], which is the total number of captures at time t. The variable n_captured is defined as a vector instead of an integer array so that it can be used in an elementwise vector operation in the generated quantities block to model the population estimates at each time point.\nThe parameters and transformed parameters are as before, but now there is a function definition for computing the entire vector chi, the probability that if an individual is alive at t that it will never be captured again.\nparameters {\n  vector&lt;lower=0, upper=1&gt;[T - 1] phi;\n  vector&lt;lower=0, upper=1&gt;[T] p;\n}\ntransformed parameters {\n  vector&lt;lower=0, upper=1&gt;[T] chi;\n  chi = prob_uncaptured(T, p, phi);\n}\nThe definition of prob_uncaptured, from the functions block, is\nfunctions {\n  // ...\n  vector prob_uncaptured(int T, vector p, vector phi) {\n    vector[T] chi;\n    chi[T] = 1.0;\n    for (t in 1:(T - 1)) {\n      int t_curr;\n      int t_next;\n      t_curr = T - t;\n      t_next = t_curr + 1;\n      chi[t_curr] = (1 - phi[t_curr])\n                     + phi[t_curr]\n                       * (1 - p[t_next])\n                       * chi[t_next];\n    }\n    return chi;\n  }\n}\nThe function definition directly follows the mathematical definition of \\(\\chi_t\\), unrolling the recursion into an iteration and defining the elements of chi from T down to 1.\n\n\n\nGiven the precomputed quantities, the model block directly encodes the CJS model’s log likelihood function. All parameters are left with their default uniform priors and the model simply encodes the log probability of the observations q given the parameters p and phi as well as the transformed parameter chi defined in terms of p and phi.\nmodel {\n  for (i in 1:I) {\n    if (first[i] &gt; 0) {\n      for (t in (first[i]+1):last[i]) {\n        1 ~ bernoulli(phi[t - 1]);\n        y[i, t] ~ bernoulli(p[t]);\n      }\n      1 ~ bernoulli(chi[last[i]]);\n    }\n  }\n}\nThe outer loop is over individuals, conditional skipping individuals i which are never captured. The never-captured check depends on the convention of the first-capture and last-capture functions returning 0 for first if an individual is never captured.\nThe inner loop for individual i first increments the log probability based on the survival of the individual with probability phi[t - 1]. The outcome of 1 is fixed because the individual must survive between the first and last capture (i.e., no zombies). The loop starts after the first capture, because all information in the CJS model is conditional on the first capture.\nIn the inner loop, the observed capture status y[i, t] for individual i at time t has a Bernoulli distribution based on the capture probability p[t] at time t.\nAfter the inner loop, the probability of an animal never being seen again after being observed at time last[i] is included, because last[i] was defined to be the last time period in which animal i was observed.\n\n\n\nAs with the collective model described in the previous subsection, this model does not identify phi[T - 1] and p[T], but does identify their product, beta. Thus beta is defined as a generated quantity to monitor convergence and report.\ngenerated quantities {\n  real beta;\n  // ...\n\n  beta = phi[T - 1] * p[T];\n  // ...\n}\nThe parameter p[1] is also not modeled and will just be uniform between 0 and 1. A more finely articulated model might have a hierarchical or time-series component, in which case p[1] would be an unknown initial condition and both phi[T - 1] and p[T] could be identified.\n\n\n\nThe generated quantities also calculates an estimate of the population mean at each time t in the same way as in the simple mark-recapture model as the number of individuals captured at time t divided by the probability of capture at time t. This is done with the elementwise division operation for vectors (./) in the generated quantities block.\ngenerated quantities {\n  // ...\n  vector&lt;lower=0&gt;[T] pop;\n  // ...\n  pop = n_captured ./ p;\n  pop[1] = -1;\n}\n\n\n\nAll individuals are modeled as having the same capture probability, but this model could be easily generalized to use a logistic regression here based on individual-level inputs to be used as predictors.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#estimating-event-probabilities",
-    "href": "stan-users-guide/posterior-prediction.html#estimating-event-probabilities",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/latent-discrete.html#data-coding-models.section",
+    "href": "stan-users-guide/latent-discrete.html#data-coding-models.section",
+    "title": "Latent Discrete Parameters",
     "section": "",
-    "text": "Event probabilities involving either parameters or predictions or both may be coded in the generated quantities block. For example, to evaluate \\(\\Pr[\\lambda &gt; 5 \\mid y]\\) in the simple Poisson example with only a rate parameter \\(\\lambda\\), it suffices to define a generated quantity\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; lambda_gt_5 = lambda &gt; 5;\n  // ...\n}\nThe value of the expression lambda &gt; 5 is 1 if the condition is true and 0 otherwise. The posterior mean of this parameter is the event probability \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = &\n\\int \\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}[\\lambda^{(m)} &gt; 5],\n\\end{eqnarray*}\\] where each \\(\\lambda^{(m)} \\sim p(\\lambda \\mid y)\\) is distributed according to the posterior. In Stan, this is recovered as the posterior mean of the parameter lambda_gt_5.\nIn general, event probabilities may be expressed as expectations of indicator functions. For example, \\[\\begin{eqnarray*}\n\\Pr[\\lambda &gt; 5 \\mid y]\n& = & \\mathbb{E}[\\textrm{I}[\\lambda &gt; 5] \\mid y]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\lambda &gt; 5) \\cdot p(\\lambda \\mid y)\n\\, \\textrm{d}\\lambda\n\\\\[4pt]\n& \\approx & \\frac{1}{M} \\sum_{m = 1}^M \\textrm{I}(\\lambda^{(m)} &gt; 5).\n\\end{eqnarray*}\\] The last line above is the posterior mean of the indicator function as coded in Stan.\nEvent probabilities involving posterior predictive quantities \\(\\tilde{y}\\) work exactly the same way as those for parameters. For example, if \\(\\tilde{y}_n\\) is the prediction for the \\(n\\)-th unobserved outcome (such as the score of a team in a game or a level of expression of a protein in a cell), then \\[\\begin{eqnarray*}\n\\Pr[\\tilde{y}_3 &gt; \\tilde{y}_7 \\mid \\tilde{x}, x, y]\n& = &\n\\mathbb{E}\\!\\left[I[\\tilde{y}_3 &gt; \\tilde{y}_7] \\mid \\tilde{x}, x, y\\right]\n\\\\[4pt]\n& = &\n\\int\n\\textrm{I}(\\tilde{y}_3 &gt; \\tilde{y}_7)\n\\cdot p(\\tilde{y} \\mid \\tilde{x}, x, y)\n\\, \\textrm{d}\\tilde{y}\n\\\\[4pt]\n& \\approx &\n\\frac{1}{M} \\sum_{m = 1}^M\n\\textrm{I}(\\tilde{y}^{(m)}_3 &gt; \\tilde{y}^{(m)}_7),\n\\end{eqnarray*}\\] where \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid \\tilde{x}, x, y).\\)",
+    "text": "Although seemingly disparate tasks, the rating/coding/annotation of items with categories and diagnostic testing for disease or other conditions, share several characteristics which allow their statistical properties to be modeled similarly.\n\n\nSuppose you have diagnostic tests for a condition of varying sensitivity and specificity. Sensitivity is the probability a test returns positive when the patient has the condition and specificity is the probability that a test returns negative when the patient does not have the condition. For example, mammograms and puncture biopsy tests both test for the presence of breast cancer. Mammograms have high sensitivity and low specificity, meaning lots of false positives, whereas puncture biopsies are the opposite, with low sensitivity and high specificity, meaning lots of false negatives.\nThere are several estimands of interest in such studies. An epidemiological study may be interested in the prevalence of a kind of infection, such as malaria, in a population. A test development study might be interested in the diagnostic accuracy of a new test. A health care worker performing tests might be interested in the disease status of a particular patient.\n\n\n\nHumans are often given the task of coding (equivalently rating or annotating) data. For example, journal or grant reviewers rate submissions, a political study may code campaign commercials as to whether they are attack ads or not, a natural language processing study might annotate Tweets as to whether they are positive or negative in overall sentiment, or a dentist looking at an X-ray classifies a patient as having a cavity or not. In all of these cases, the data coders play the role of the diagnostic tests and all of the same estimands are in play — data coder accuracy and bias, true categories of items being coded, or the prevalence of various categories of items in the data.\n\n\n\nIn this section, only categorical ratings are considered, and the challenge in the modeling for Stan is to marginalize out the discrete parameters.\nDawid and Skene (1979) introduce a noisy-measurement model for coding and apply it in the epidemiological setting of coding what doctors say about patient histories; the same model can be used for diagnostic procedures.\n\n\nThe data for the model consists of \\(J\\) raters (diagnostic tests), \\(I\\) items (patients), and \\(K\\) categories (condition statuses) to annotate, with \\(y_{i, j} \\in \\{1, \\dotsc, K\\}\\) being the rating provided by rater \\(j\\) for item \\(i\\). In a diagnostic test setting for a particular condition, the raters are diagnostic procedures and often \\(K=2\\), with values signaling the presence or absence of the condition.6\nIt is relatively straightforward to extend Dawid and Skene’s model to deal with the situation where not every rater rates each item exactly once.\n\n\n\n\nThe model is based on three parameters, the first of which is discrete:\n\n\\(z_i\\) : a value in \\(\\{1, \\dotsc, K\\}\\) indicating the true category of item \\(i\\),\n\\(\\pi\\) : a \\(K\\)-simplex for the prevalence of the \\(K\\) categories in the population, and\n\\(\\theta_{j,k}\\) : a \\(K\\)-simplex for the response of annotator \\(j\\) to an item of true category \\(k\\).\n\n\n\n\nThe true category of an item is assumed to be generated by a simple categorical distribution based on item prevalence, \\[\nz_i \\sim \\textsf{categorical}(\\pi).\n\\]\nThe rating \\(y_{i, j}\\) provided for item \\(i\\) by rater \\(j\\) is modeled as a categorical response of rater \\(i\\) to an item of category \\(z_i\\),7 \\[\ny_{i, j} \\sim \\textsf{categorical}(\\theta_{j,\\pi_{z[i]}}).\n\\]\n\n\nDawid and Skene provided maximum likelihood estimates for \\(\\theta\\) and \\(\\pi\\), which allows them to generate probability estimates for each \\(z_i\\).\nTo mimic Dawid and Skene’s maximum likelihood model, the parameters \\(\\theta_{j,k}\\) and \\(\\pi\\) can be given uniform priors over \\(K\\)-simplexes. It is straightforward to generalize to Dirichlet priors, \\[\n\\pi \\sim \\textsf{Dirichlet}(\\alpha)\n\\] and \\[\n\\theta_{j,k} \\sim \\textsf{Dirichlet}(\\beta_k)\n\\] with fixed hyperparameters \\(\\alpha\\) (a vector) and \\(\\beta\\) (a matrix or array of vectors). The prior for \\(\\theta_{j,k}\\) must be allowed to vary in \\(k\\), so that, for instance, \\(\\beta_{k,k}\\) is large enough to allow the prior to favor better-than-chance annotators over random or adversarial ones.\nBecause there are \\(J\\) coders, it would be natural to extend the model to include a hierarchical prior for \\(\\beta\\) and to partially pool the estimates of coder accuracy and bias.\n\n\n\nBecause the true category parameter \\(z\\) is discrete, it must be marginalized out of the joint posterior in order to carry out sampling or maximum likelihood estimation in Stan. The joint posterior factors as \\[\np(y, \\theta, \\pi) = p(y \\mid \\theta,\\pi) \\, p(\\pi) \\, p(\\theta),\n\\] where \\(p(y \\mid \\theta,\\pi)\\) is derived by marginalizing \\(z\\) out of \\[\np(z, y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\left( \\textsf{categorical}(z_i \\mid \\pi)\n                     \\prod_{j=1}^J\n                     \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, z[i]})\n              \\right).\n\\]\nThis can be done item by item, with \\[\np(y \\mid \\theta, \\pi) =\n\\prod_{i=1}^I \\sum_{k=1}^K\n  \\left( \\textsf{categorical}(k \\mid \\pi)\n         \\prod_{j=1}^J\n         \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n  \\right).\n\\]\nIn the missing data model, only the observed labels would be used in the inner product.\nDawid and Skene (1979) derive exactly the same equation in their Equation (2.7), required for the E-step in their expectation maximization (EM) algorithm. Stan requires the marginalized probability function on the log scale, \\[\\begin{align*}\n\\log p(y \\mid \\theta, \\pi)\n&= \\sum_{i=1}^I \\log \\left( \\sum_{k=1}^K \\exp\n    \\left(\\log \\textsf{categorical}(k \\mid \\pi) \\vphantom{\\sum_{j=1}^J}\\right.\\right.\n    \\left.\\left. + \\ \\sum_{j=1}^J\n           \\log \\textsf{categorical}(y_{i, j} \\mid \\theta_{j, k})\n    \\right) \\right),\n\\end{align*}\\] which can be directly coded using Stan’s built-in log_sum_exp function.\n\n\n\n\nThe Stan program for the Dawid and Skene model is provided below (Dawid and Skene 1979).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=1&gt; I;\n  int&lt;lower=1&gt; J;\n\n  array[I, J] int&lt;lower=1, upper=K&gt; y;\n\n  vector&lt;lower=0&gt;[K] alpha;\n  vector&lt;lower=0&gt;[K] beta[K];\n}\nparameters {\n  simplex[K] pi;\n  array[J, K] simplex[K] theta;\n}\ntransformed parameters {\n  array[I] vector[K] log_q_z;\n  for (i in 1:I) {\n    log_q_z[i] = log(pi);\n    for (j in 1:J) {\n      for (k in 1:K) {\n        log_q_z[i, k] = log_q_z[i, k]\n                         + log(theta[j, k, y[i, j]]);\n      }\n    }\n  }\n}\nmodel {\n  pi ~ dirichlet(alpha);\n  for (j in 1:J) {\n    for (k in 1:K) {\n      theta[j, k] ~ dirichlet(beta[k]);\n    }\n  }\n\n  for (i in 1:I) {\n    target += log_sum_exp(log_q_z[i]);\n  }\n}\nThe model marginalizes out the discrete parameter \\(z\\), storing the unnormalized conditional probability \\(\\log q(z_i=k|\\theta,\\pi)\\) in log_q_z[i, k].\nThe Stan model converges quickly and mixes well using NUTS starting at diffuse initial points, unlike the equivalent model implemented with Gibbs sampling over the discrete parameter. Reasonable weakly informative priors are \\(\\alpha_k = 3\\) and \\(\\beta_{k,k} = 2.5 K\\) and \\(\\beta_{k,k'} = 1\\) if \\(k \\neq k'\\). Taking \\(\\alpha\\) and \\(\\beta_k\\) to be unit vectors and applying optimization will produce the same answer as the expectation maximization (EM) algorithm of Dawid and Skene (1979).\n\n\nThe quantity log_q_z[i] is defined as a transformed parameter. It encodes the (unnormalized) log of \\(p(z_i \\mid \\theta,\n\\pi)\\). Each iteration provides a value conditioned on that iteration’s values for \\(\\theta\\) and \\(\\pi\\). Applying the softmax function to log_q_z[i] provides a simplex corresponding to the probability mass function of \\(z_i\\) in the posterior. These may be averaged across the iterations to provide the posterior probability distribution over each \\(z_i\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction",
-    "href": "stan-users-guide/posterior-prediction.html#stand-alone-generated-quantities-and-ongoing-prediction",
-    "title": "Posterior Predictive Sampling",
+    "objectID": "stan-users-guide/latent-discrete.html#marginalization-mathematics.section",
+    "href": "stan-users-guide/latent-discrete.html#marginalization-mathematics.section",
+    "title": "Latent Discrete Parameters",
     "section": "",
-    "text": "Stan’s sampling algorithms take a Stan program representing a posterior \\(p(\\theta \\mid y, x)\\) along with actual data \\(x\\) and \\(y\\) to produce a set of draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)}\\) from the posterior. Posterior predictive draws \\(\\tilde{y}^{(m)} \\sim p(\\tilde{y} \\mid\n\\tilde{x}, x, y)\\) can be generated by drawing \\[\n\\tilde{y}^{(m)} \\sim p(y \\mid \\tilde{x}, \\theta^{(m)})\n\\] from the data model. Note that drawing \\(\\tilde{y}^{(m)}\\) only depends on the new predictors \\(\\tilde{x}\\) and the posterior draws \\(\\theta^{(m)}\\). Most importantly, neither the original data or the model density is required.\nBy saving the posterior draws, predictions for new data items \\(\\tilde{x}\\) may be generated whenever needed. In Stan’s interfaces, this is done by writing a second Stan program that inputs the original program’s parameters and the new predictors. For example, for the linear regression case, the program to take posterior draws declares the data and parameters, and defines the model.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 5);\n  beta ~ normal(0, 1);\n  sigma ~ lognormal(0, 0.5);\n}\nA second program can be used to generate new observations. This follow-on program need only declare the parameters as they were originally defined. This may require defining constants in the data block such as sizes and hyperparameters that are involved in parameter size or constraint declarations. Then additional data is read in corresponding to predictors for new outcomes that have yet to be observed. There is no need to repeat the model or unneeded transformed parameters or generated quantities. The complete follow-on program for prediction just declares the predictors in the data, the original parameters, and then the predictions in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde\n    = normal_rng(alpha + beta * x_tilde, sigma);\n}\nWhen running stand-alone generated quantities, the inputs required are the original draws for the parameters and any predictors corresponding to new predictions, and the output will be draws for \\(\\tilde{y}\\) or derived quantities such as event probabilities.\nAny posterior predictive quantities desired may be generated this way. For example, event probabilities are estimated in the usual way by defining indicator variables in the generated quantities block.",
+    "text": "This section describes in more detail the mathematics of statistical inference using the output of marginalized Stan models, such as those presented in the last three sections. It provides a mathematical explanation of why and how certain manipulations of Stan’s output produce valid summaries of the posterior distribution when discrete parameters have been marginalized out of a statistical model. Ultimately, however, fully understanding the mathematics in this section is not necessary to fit models with discrete parameters using Stan.\nThroughout, the model under consideration consists of both continuous parameters, \\(\\Theta\\), and discrete parameters, \\(Z\\). It is also assumed that \\(Z\\) can only take finitely many values, as is the case for all the models described in this chapter of the User’s Guide. To simplify notation, any conditioning on data is suppressed in this section, except where specified. As with all Bayesian analyses, however, all inferences using models with marginalized parameters are made conditional on the observed data.\n\n\n\nWhen performing Bayesian inference, interest often centers on estimating some (constant) low-dimensional summary statistics of the posterior distribution. Mathematically, we are interested in estimating \\(\\mu\\), say, where \\(\\mu = \\mathbb{E}[g(\\Theta, Z)]\\) and \\(g(\\cdot)\\) is an arbitrary function. An example of such a quantity is \\(\\mathbb{E}[\\Theta]\\), the posterior mean of the continuous parameters, where we would take \\(g(\\theta, z) = \\theta\\). To estimate \\(\\mu\\) the most common approach is to sample a series of values, at least approximately, from the posterior distribution of the parameters of interest. The numerical values of these draws can then be used to calculate the quantities of interest. Often, this process of calculation is trivial, but more care is required when working with marginalized posteriors as we describe in this section.\nIf both \\(\\Theta\\) and \\(Z\\) were continuous, Stan could be used to sample \\(M\\) draws from the joint posterior \\(p_{\\Theta, Z}(\\theta, z)\\) and then estimate \\(\\mu\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M {g(\\theta^{(i)}, z^{(i)})}.\n\\] Given \\(Z\\) is discrete, however, Stan cannot be used to sample from the joint posterior (or even to do optimization). Instead, as outlined in the previous sections describing specific models, the user can first marginalize out \\(Z\\) from the joint posterior to give the marginalized posterior \\(p_\\Theta(\\theta)\\). This marginalized posterior can then be implemented in Stan as usual, and Stan will give draws \\(\\{\\theta^{(i)}\\}_{i = 1}^M\\) from the marginalized posterior.\nUsing only these draws, how can we estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\)? We can use a conditional estimator. We explain in more detail below, but at a high level the idea is that, for each function \\(g\\) of interest, we compute \\[\nh(\\Theta) = \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n\\] and then estimate \\(\\mathbb{E}[g(\\Theta, Z)]\\) with \\[\n\\hat{\\mu} = \\frac{1}{M} \\sum_{i = 1}^M h(\\theta^{(i)}).\n\\] This estimator is justified by the law of iterated expectation, the fact that \\[\n\\mathbb{E}[h(\\Theta)] = \\mathbb{E}[\\mathbb{E}[g(\\Theta, Z)] \\mid \\Theta] = \\mathbb{E}[g(\\Theta, Z)] = \\mu.\n\\] Using this marginalized estimator provides a way to estimate the expectation of any function \\(g(\\cdot)\\) for all combinations of discrete or continuous parameters in the model. However, it presents a possible new challenge: evaluating the conditional expectation \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\).\n\n\n\nFortunately, the discrete nature of \\(Z\\) makes evaluating \\(\\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\\) easy. The function \\(h(\\Theta)\\) can be written as: \\[\nh(\\Theta)\n= \\mathbb{E}[g(\\Theta, Z) \\mid \\Theta]\n= \\sum_{k} g(\\Theta, k) \\Pr[Z = k \\mid \\Theta],\n\\] where we sum over the possible values of the latent discrete parameters. An essential part of this formula is the probability of the discrete parameters conditional on the continuous parameters, \\(\\Pr[Z = k \\mid \\Theta]\\). More detail on how this quantity can be calculated is included below. Note that if \\(Z\\) takes infinitely many values then computing the infinite sums will involve, potentially computationally expensive, approximation.\nWhen \\(g(\\theta, z)\\) is a function of either \\(\\theta\\) or \\(z\\) only, the above formula simplifies further.\nIn the first case, where \\(g(\\theta, z) = g(\\theta)\\), we have: \\[\\begin{align*}\nh(\\Theta)\n&= \\sum_{k} g(\\Theta) \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta) \\sum_{k} \\Pr[Z = k \\mid \\Theta] \\\\\n&= g(\\Theta).\n\\end{align*}\\] This means that we can estimate \\(\\mathbb{E}[g(\\Theta)]\\) with the standard, seemingly unconditional, estimator: \\[\n\\frac{1}{M} \\sum_{i = 1}^M g(\\theta^{(i)}).\n\\] Even after marginalization, computing expectations of functions of the continuous parameters can be performed as if no marginalization had taken place.\nIn the second case, where \\(g(\\theta, z) = g(z)\\), the conditional expectation instead simplifies as follows: \\[\nh(\\Theta) = \\sum_{k} g(k) \\Pr[Z = k \\mid \\Theta].\n\\] An important special case of this result is when \\(g(\\theta, z) = \\textrm{I}(z = k)\\), where \\(\\textrm{I}\\) is the indicator function. This choice allows us to recover the probability mass function of the discrete random variable \\(Z\\), since \\(\\mathbb{E}[\\textrm{I}(Z = k)] = \\Pr[Z = k]\\). In this case, \\[\nh(\\Theta)\n= \\sum_{k} \\textrm{I}(z = k) \\Pr[Z = k \\mid \\Theta]\n= \\Pr[Z = k \\mid \\Theta].\n\\] The quantity \\(\\Pr[Z = k]\\) can therefore be estimated with: \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}].\n\\] When calculating this conditional probability it is important to remember that we are also conditioning on the observed data, \\(Y\\). That is, we are really estimating \\(\\Pr[Z = k \\mid Y]\\) with \\[\n\\frac{1}{M} \\sum_{i = 1}^M \\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y].\n\\] This point is important as it suggests one of the main ways of calculating the required conditional probability. Using Bayes’s theorem gives us \\[\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}, Y]\n= \\frac{\\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}\n{\\sum_{k = 1}^K \\Pr[Y \\mid Z = k, \\Theta = \\theta^{(i)}]\n\\Pr[Z = k \\mid \\Theta = \\theta^{(i)}]}.\n\\] Here, \\(\\Pr[Y \\mid \\Theta = \\theta^{(i)}, Z = k]\\) is the likelihood conditional on a particular value of the latent variables. Crucially, all elements of the expression can be calculated using the draws from the posterior of the continuous parameters and knowledge of the model structure.\nOther than the use of Bayes’s theorem, \\(\\Pr[Z = k \\mid \\theta = \\theta^{(i)}, Y]\\) can also be extracted by coding the Stan model to include the conditional probability explicitly (as is done for the Dawid–Skene model).\nFor a longer introduction to the mathematics of marginalization in Stan, which also covers the connections between Rao–Blackwellization and marginalization, see Pullin, Gurrin, and Vukcevic (2021).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Posterior Predictive Sampling"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html",
-    "href": "stan-users-guide/poststratification.html",
-    "title": "Poststratification",
-    "section": "",
-    "text": "Stratification is a technique developed for survey sampling in which a population is partitioned into subgroups (i.e., stratified) and each group (i.e., stratum) is sampled independently. If the subgroups are more homogeneous (i.e., lower variance) than the population as a whole, this can reduce variance in the estimate of a quantity of interest at the population level.\nPoststratification is a technique for adjusting a non-representative sample (i.e., a convenience sample or other observational data) for which there are demographic predictors characterizing the strata. It is carried out after a model is fit to the observed data, hence the name poststratification (Little 1993). Poststratification can be fruitfully combined with regression modeling (or more general parametric modeling), which provides estimates based on combinations of predictors (or general parameters) rather than raw counts in each stratum. Multilevel modeling is useful in determining how much partial pooling to apply in the regressions, leading to the popularity of the combination of multilevel regression and poststratification (MRP) (Park, Gelman, and Bafumi 2004).\n\n\n\n\nStratification and poststratification can be applied to many applications beyond survey sampling (Kennedy and Gelman 2019). For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid (Paustian et al. 1997). Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both (Field et al. 1998).\n\n\n\nSuppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.\nThe proportion of support in the poll among students in each group provides a simple maximum likelihood estimate \\(\\theta^* = (0.25, 0.5,\n0.8)\\) of support in each group for a simple Bernoulli model where student \\(n\\)’s vote is modeled as \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:3\\) is the subgroup to which the \\(n\\)-th student belongs.\nAn estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting \\(N = (20\\,000,\\, 5\\,000,\\, 2\\,000)\\) be the subgroup sizes, the poststratified estimate of support in the population \\(\\phi^*\\) is estimated by \\[\n\\phi^*\n= \\frac{\\displaystyle \\sum_{j = 1}^3 \\theta_j^* \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^3 N_j}.\n\\] Plugging in our estimates and population counts yields \\[\\begin{eqnarray*}\n\\phi*\n& = & \\frac{0.25 \\cdot 20\\,000 + 0.4 \\cdot 5\\,000 + 0.8 \\cdot 2\\,000}\n           {20\\,000 + 5\\,000 + 2\\,000}\n\\\\[4pt] & = & \\frac{8\\,600}{27\\,000}\n\\\\[4pt] & \\approx & 0.32.\n\\end{eqnarray*}\\]\n\n\n\n\nConsidering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:J\\) is the group to which item \\(n\\) belongs and \\(\\theta_j\\) is the proportion of support in group \\(j\\).\nThis can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting \\(A_j\\) be the number of respondents in group \\(j\\) and \\(a_j\\) be the number of positive responses in group \\(j\\), the data model may be reduced to the form \\[\na_j \\sim \\textrm{binomial}(A_j, \\theta_j).\n\\] A simple uniform prior on the proportion of support in each group completes the model, \\[\n\\theta_j \\sim \\textrm{beta}(1, 1).\n\\] A more informative prior could be used if there is prior information available about support among the student body.\nUsing sampling, draws \\(\\theta^{(m)} \\sim p(\\theta \\mid y)\\) from the posterior may be combined with the population sizes \\(N\\) to estimate \\(\\phi\\), the proportion of support in the population, \\[\n\\phi^{(m)}\n= \\frac{\\displaystyle \\sum_{j = 1}^J \\theta_j^{(m)} \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^J N_j}.\n\\] The posterior draws for \\(\\phi^{(m)}\\) characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the \\(\\phi^{(m)}\\)), posterior intervals (quantiles of the \\(\\phi^{(m)}\\)), or to plot a histogram.\n\n\n\nThe maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where \\(A_j\\) is the number of voters from group \\(j\\), \\(a_j\\) is the number of positive responses from group \\(j\\), and \\(N_j\\) is the size of group \\(j\\) in the population.\ndata {\n  int&lt;lower=1&gt; J;\n  array[J] int&lt;lower=0&gt; A; \n  array[J] int&lt;lower=0&gt; a;\n  vector&lt;lower=0&gt;[J] N;\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[J] theta;\n}\nmodel {\n  a ~ binomial(A, theta);\n}\ngenerated quantities {t\n  real&lt;lower=0, upper=1&gt; phi = dot(N, theta) / sum(N);\n}\nThe binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the \\(J\\) terms. The prior is implicitly uniform on \\((0, 1),\\) the support of \\(\\theta.\\) The summation is computed using a dot product and the sum function, which is why N was declared as a vector rather than as an array of integers.\n\n\n\nIn applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.\nThe problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to \\(5 \\cdot 2 \\cdot 5 \\cdot\n50 = 2000\\) strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter \\(\\theta_j\\) for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.\nA practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only \\(4 + 2 + 5 + 50 =\n61\\) regression coefficients to estimate. Now suppose that item \\(n\\) has demographic features \\(\\textrm{age}_n \\in 1:5\\), \\(\\textrm{sex}_n \\in 1:2\\), \\(\\textrm{income}_n \\in 1:5,\\) and \\(\\textrm{state}_n \\in 1:50\\). A logistic regression may be formulated as \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)),\n\\] where \\(\\textrm{age}[n]\\) is the age of the \\(n\\)-th respondent, \\(\\textrm{sex}[n]\\) is their sex, \\(\\textrm{income}[n]\\) their income and \\(\\textrm{state}[n]\\) their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.\nTo poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as \\[\n\\sum_{i = 1}^5 \\sum_{j = 1}^2 \\sum_{k = 1}^5 \\sum_{m = 1}^{50}\n\\textrm{logit}^{-1}(\\alpha + \\beta_i + \\gamma_j + \\delta_k + \\eta_m)\n\\cdot \\textrm{pop}_{i, j, k, m},\n\\] where \\(\\textrm{pop}_{i, j, k, m}\\) is the size of the subpopulation with age \\(i\\), sex \\(j\\), income level \\(k\\), and state of residence \\(m\\).\nAs formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes \\(y\\) given predictors \\(x\\).\n\n\n\nWith large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.\nThe model introduced in the previous section had the data model \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)).\n\\] The overall intercept can be given a broad fixed prior, \\[\n\\alpha \\sim \\textrm{normal}(0, 5).\n\\] The other regression parameters can be given hierarchical priors, \\[\\begin{eqnarray*}\n\\beta_{1:4} & \\sim & \\textrm{normal}(0, \\sigma^{\\beta})\n\\\\[2pt]\n\\gamma_{1:2} & \\sim & \\textrm{normal}(0, \\sigma^{\\gamma})\n\\\\[2pt]\n\\delta_{1:5} & \\sim & \\textrm{normal}(0, \\sigma^{\\delta})\n\\\\[2pt]\n\\epsilon_{1:50} & \\sim & \\textrm{normal}(0, \\sigma^{\\epsilon}).\n\\end{eqnarray*}\\]\nThe hyperparameters for scale of variation within a group can be given simple standard hyperpriors, \\[\n\\sigma^{\\beta}, \\sigma^{\\gamma}, \\sigma^{\\delta}, \\sigma^{\\epsilon}\n\\sim \\textrm{normal}(0, 1).\n\\] The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).\n\n\nThe multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.\nThis is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, \\(\\gamma_2 =\n-\\gamma_1.\\)\nTo implement sum-to-zero constraints,\nparameters {\n  vector[K - 1] alpha_raw;\n// ...\n}\ntransformed parameters {\n  vector&lt;multiplier=sigma_alpha&gt;[K] alpha\n    = append_row(alpha_raw, -sum(alpha_raw));\n// ...    \n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n}\nThis prior is hard to interpret in that there are K normal distributions, but only K - 1 free parameters. An alternative is to put the prior only on alpha_raw, but that is also difficult to interpret.\nSoft constraints can be more computationally tractable. They are also simpler to implement.\nparameters {\n  vector&lt;multiplier=alpha&gt;[K] alpha;\n// ...\n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n  sum(alpha) ~ normal(0, 0.001);\n}\nThis leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.\nNote that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept \\(\\alpha.\\) Even with a wide prior on the intercept, the hyperprior on \\(\\sigma^{\\gamma}\\) may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.\nNone of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.\n\n\n\n\nMultilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the multiplier transform is used on each of the parameters. The combination of\nvector&lt;multiplier=s&gt;[K] a;\n// ...\na ~ normal(0, s);\nimplements a non-centered parameterization for a; a centered parameterization would drop the multiplier specification. The prior scale s is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the multiplier.\nHere is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable P.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; age;\n  array[N] int&lt;lower=1, upper=5&gt; income;\n  array[N] int&lt;lower=1, upper=50&gt; state;\n  array[N] int&lt;lower=0&gt; y;\n  array[4, 5, 50] int&lt;lower=0&gt; P;\n}\nparameters {\n  real alpha;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;multiplier=sigma_beta&gt;[4] beta;\n  real&lt;lower=0&gt; sigma_gamma;\n  vector&lt;multiplier=sigma_gamma&gt;[5] gamma;\n  real&lt;lower=0&gt; sigma_delta;\n  vector&lt;multiplier=sigma_delta&gt;[50] delta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, sigma_beta);\n  gamma ~ normal(0, sigma_gamma);\n  delta ~ normal(0, sigma_delta);\n  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(0, 1);\n}\ngenerated quantities {\n  real expect_pos = 0;\n  int total = 0;\n  for (b in 1:4) {\n    for (c in 1:5) {\n      for (d in 1:50) {\n        total += P[b, c, d];\n        expect_pos\n          += P[b, c, d]\n             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);\n      }\n    }\n  }\n  real&lt;lower=0, upper=1&gt; phi = expect_pos / total;\n}\nUnlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).\n\n\nIn some cases, it can be more efficient to break the data down by group. Suppose there are \\(4 \\times 5 \\times 2 \\times 50 = 2000\\) groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group\nint&lt;lower=0&gt; G;\narray[G] int&lt;lower=1, upper=4&gt; age;\narray[G] int&lt;lower=1, upper=5&gt; income;\narray[G] int&lt;lower=1, upper=50&gt; state;\nThen the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.\narray[G] int&lt;lower=0&gt; pos_votes;\narray[G] int&lt;lower=0&gt; total_votes;\nFinally, the data model is converted to binomial.\npos_votes ~ binomial_logit(total_votes,\n                           alpha + beta[age] + ...);\nThe predictors look the same because of the way the age and other data items are coded.\n\n\n\nIn this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the N data points.\n  array[N] int&lt;lower=1, upper=2&gt; sex;\nThen add a single regression coefficient as a parameter,\n  real epsilon;\nIn the log odds calculation, introduce a new term\n[epsilon, -epsilon][sex]';\nThat is, the data model will now look like\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]\n                      + [epsilon, -epsilon][sex]');\nFor data point n, the expression [epsilon, -epsilon][sex] takes on value [epsilon, -epsilon][sex][n], which with Stan’s multi-indexing reduces to [epsilon, -epsilon][sex[n]]. This term evaluates to epsilon if sex[n] is 1 and to -epsilon if sex[n] is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The ' at the end transposes [epsilon, -epsilon][sex] which is a row_vector into a vector that can be added to the other variables.\nFinally, a prior is needed for the coefficient in the model block,\nepsilon ~ normal(0, 2);\nAs with other priors in multilevel models, the posterior for epsilon should be investigated to make sure it is not unrealistically wide.\n\n\n\n\nIf there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable\narray[50] real&lt;lower=0&gt; income;\nthen a regression coefficient psi can be added for the effect of average state income,\nreal psi;\nwith a fixed prior,\npsi ~ normal(0, 2);\nThis prior assumes the income predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,\ny ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]\n                    + income[state] * psi);\nAnd finally, the formula in the generated quantities block is also updated,\nexpect_pos\n  += P[b, c, d]\n     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]\n             + income[d] * psi);\nHere d is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.",
+    "objectID": "stan-users-guide/latent-discrete.html#footnotes",
+    "href": "stan-users-guide/latent-discrete.html#footnotes",
+    "title": "Latent Discrete Parameters",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe computations are similar to those involved in expectation maximization (EM) algorithms (Dempster, Laird, and Rubin 1977).↩︎\nThe source of the data is (Jarrett 1979), which itself is a note correcting an earlier data collection.↩︎\nThe R counterpart, ifelse, is slightly different in that it is typically used in a vectorized situation. The conditional operator is not (yet) vectorized in Stan.↩︎\nAn alternative would be to compute this on the outside and feed it into the Stan model as preprocessed data. Yet another alternative encoding would be a sparse one recording only the capture events along with their time and identifying the individual captured.↩︎\nBoth functions return 0 if the individual represented by the input array was never captured. Individuals with no captures are not relevant for estimating the model because all probability statements are conditional on earlier captures. Typically they would be removed from the data, but the program allows them to be included even though they make not contribution to the log probability function.↩︎\nDiagnostic procedures are often ordinal, as in stages of cancer in oncological diagnosis or the severity of a cavity in dental diagnosis. Dawid and Skene’s model may be used as is or naturally generalized for ordinal ratings using a latent continuous rating and cutpoints as in ordinal logistic regression.↩︎\nIn the subscript, \\(z_i\\) is written as \\(z[i]\\) to improve legibility.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Example Models",
+      "Latent Discrete Parameters"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html#some-examples",
-    "href": "stan-users-guide/poststratification.html#some-examples",
-    "title": "Poststratification",
+    "objectID": "stan-users-guide/measurement-error.html",
+    "href": "stan-users-guide/measurement-error.html",
+    "title": "Measurement Error and Meta-Analysis",
     "section": "",
-    "text": "Stratification and poststratification can be applied to many applications beyond survey sampling (Kennedy and Gelman 2019). For example, large-scale whole-earth soil-carbon models are fit with parametric models of how soil-carbon depends on features of an area such as soil composition, flora, fauna, temperature, humidity, etc. Given a model that predicts soil-carbon concentration given these features, a whole-earth model can be created by stratifying the earth into a grid of say 10km by 10km “squares” (they can’t literally be square because the earth’s surface is topologically a sphere). Each grid area has an estimated makeup of soil type, forestation, climate, etc. The global level of soil carbon is then estimated using poststratification by simply summing the expected soil carbon estimated for each square in the grid (Paustian et al. 1997). Dynamic models can then be constructed by layering a time-series component, varying the poststratification predictors over time, or both (Field et al. 1998).\n\n\n\nSuppose a university’s administration would like to estimate the support for a given proposal among its students. A poll is carried out in which 490 respondents are undergraduates, 112 are graduate students, and 47 are continuing education students. Now suppose that support for the issue among the poll respondents is is 25% among undergraduate students (subgroup 1), 40% among graduate students (subgroup 2), and 80% among continuing education students (subgroup 3). Now suppose that the student body is made up of 20,000 undergraduates, 5,000 graduate students, and 2,000 continuing education students. It is important that our subgroups are exclusive and exhaustive, i.e., they form a partition of the population.\nThe proportion of support in the poll among students in each group provides a simple maximum likelihood estimate \\(\\theta^* = (0.25, 0.5,\n0.8)\\) of support in each group for a simple Bernoulli model where student \\(n\\)’s vote is modeled as \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:3\\) is the subgroup to which the \\(n\\)-th student belongs.\nAn estimate of the population prevalence of support for the issue among students can be constructed by simply multiplying estimated support in each group by the size of each group. Letting \\(N = (20\\,000,\\, 5\\,000,\\, 2\\,000)\\) be the subgroup sizes, the poststratified estimate of support in the population \\(\\phi^*\\) is estimated by \\[\n\\phi^*\n= \\frac{\\displaystyle \\sum_{j = 1}^3 \\theta_j^* \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^3 N_j}.\n\\] Plugging in our estimates and population counts yields \\[\\begin{eqnarray*}\n\\phi*\n& = & \\frac{0.25 \\cdot 20\\,000 + 0.4 \\cdot 5\\,000 + 0.8 \\cdot 2\\,000}\n           {20\\,000 + 5\\,000 + 2\\,000}\n\\\\[4pt] & = & \\frac{8\\,600}{27\\,000}\n\\\\[4pt] & \\approx & 0.32.\n\\end{eqnarray*}\\]",
+    "text": "Most quantities used in statistical models arise from measurements. Most of these measurements are taken with some error. When the measurement error is small relative to the quantity being measured, its effect on a model is usually small. When measurement error is large relative to the quantity being measured, or when precise relations can be estimated being measured quantities, it is useful to introduce an explicit model of measurement error. One kind of measurement error is rounding.\nMeta-analysis plays out statistically much like measurement error models, where the inferences drawn from multiple data sets are combined to do inference over all of them. Inferences for each data set are treated as providing a kind of measurement error with respect to true parameter values.\n\n\nA Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data (Clayton 1992; Richardson and Gilks 1993). This requires a model of how the measurements are derived from the true values.\n\n\nBefore considering regression with measurement error, first consider a linear regression model where the observed data for \\(N\\) cases includes a predictor \\(x_n\\) and outcome \\(y_n\\). In Stan, a linear regression for \\(y\\) based on \\(x\\) with a slope and intercept is modeled as follows.\ndata {\n  int&lt;lower=0&gt; N;       // number of cases\n  vector[N] x;          // predictor (covariate)\n  vector[N] y;          // outcome (variate)\n}\nparameters {\n  real alpha;           // intercept\n  real beta;            // slope\n  real&lt;lower=0&gt; sigma;  // outcome noise\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ cauchy(0, 5);\n}\nNow suppose that the true values of the predictors \\(x_n\\) are not known, but for each \\(n\\), a measurement \\(x^{\\textrm{meas}}_n\\) of \\(x_n\\) is available. If the error in measurement can be modeled, the measured value \\(x^{\\textrm{meas}}_n\\) can be modeled in terms of the true value \\(x_n\\) plus measurement noise. The true value \\(x_n\\) is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation \\(\\tau\\). This leads to the following regression model with constant measurement error.\ndata {\n  // ...\n  array[N] real x_meas;   // measurement of x\n  real&lt;lower=0&gt; tau;     // measurement noise\n}\nparameters {\n  array[N] real x;    // unknown true value\n  real mu_x;          // prior location\n  real sigma_x;       // prior scale\n  // ...\n}\nmodel {\n  x ~ normal(mu_x, sigma_x);  // prior\n  x_meas ~ normal(x, tau);    // measurement model\n  y ~ normal(alpha + beta * x, sigma);\n  // ...\n}\nThe regression coefficients alpha and beta and regression noise scale sigma are the same as before, but now x is declared as a parameter rather than as data. The data are now x_meas, which is a measurement of the true x value with noise scale tau. The model then specifies that the measurement error for x_meas[n] given true value x[n] is normal with deviation tau. Furthermore, the true values x are given a hierarchical prior here.\nIn cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, Clayton (1992) introduces an exposure model for the unknown (but noisily measured) risk factors \\(x\\) in terms of known (without measurement error) risk factors \\(c\\). A simple model would regress \\(x_n\\) on the covariates \\(c_n\\) with noise term \\(\\upsilon\\), \\[\nx_n \\sim \\textsf{normal}(\\gamma^{\\top}c, \\upsilon).\n\\] This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.\n\n\n\nA common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.\nExercise 3.5(b) by Gelman et al. (2013) provides an example.\n\n3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on \\(\\mu\\) and \\(\\sigma^2\\).\n\nGive the correct posterior distribution for \\((\\mu, \\sigma^2)\\), treating the measurements as rounded.\n\n\nLetting \\(z_n\\) be the unrounded measurement for \\(y_n\\), the problem as stated assumes \\[\nz_n \\sim \\textsf{normal}(\\mu, \\sigma).\n\\]\nThe rounding process entails that \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\)1. The probability mass function for the discrete observation \\(y\\) is then given by marginalizing out the unrounded measurement, producing the likelihood \\[\\begin{align*}\np(y_n \\mid \\mu, \\sigma)\n&= \\int_{y_n - 0.5}^{y_n + 0.5} \\textsf{normal}(z_n \\mid \\mu, \\sigma) \\,\\textsf{d}z_n \\\\\n&= \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n   -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right).\n\\end{align*}\\] Gelman’s answer for this problem took the noninformative prior to be uniform in the variance \\(\\sigma^2\\) on the log scale, but we replace it with more recently recommended half-normal prior on \\(\\sigma\\) \\[\n\\sigma \\sim \\textsf{normal}^+(0, 1).\n\\] The posterior after observing \\(y = (10, 10, 12, 11, 9)\\) can be calculated by Bayes’s rule as \\[\\begin{align*}\np(\\mu, \\sigma \\mid y)\n&\\propto p(\\mu, \\sigma) \\ p(y \\mid \\mu, \\sigma) \\\\\n&\\propto \\textsf{normal}^+(\\sigma \\mid 0, 1)\\prod_{n=1}^5\n  \\left( \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n        -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right)\n  \\right).\n\\end{align*}\\]\nThe Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  for (n in 1:N) {\n    target += log_diff_exp(normal_lcdf(y[n] + 0.5 | mu, sigma),\n                           normal_lcdf(y[n] - 0.5 | mu, sigma));\n  }\n}\nwhere normal_lcdf(y[n]+0.5 | mu, sigma) is equal to log(Phi((y[n] + 0.5 - mu) / sigma)), and log_diff_exp(a, b) computes log(exp(a) - exp(b)) in numerically more stable way.\nAlternatively, the model may be defined with latent parameters for the unrounded measurements \\(z_n\\). The Stan code in this case uses a distribution statement for \\(z_n\\) directly while respecting the constraint \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;lower=y-0.5, upper=y+0.5&gt;[N] z;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  z ~ normal(mu, sigma);\n}\nThis explicit model for the unrounded measurements \\(z\\) produces the same posterior for \\(\\mu\\) and \\(\\sigma\\) as the previous model that marginalizes \\(z\\) out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.\n\n\n\n\nMeta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.\nThe Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.\n\n\nSuppose the data in question arise from a total of \\(M\\) studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen (Warn, Thompson, and Spiegelhalter 2002) or mortality after myocardial infarction under a treatment of beta blockers (Gelman et al. 2013, sec. 5.6).\n\n\nThe clinical data consists of \\(J\\) trials, each with \\(n^t\\) treatment cases, \\(n^c\\) control cases, \\(r^t\\) successful outcomes among those treated and \\(r^c\\) successful outcomes among those in the control group. This data can be declared in Stan as follows.2\ndata {\n  int&lt;lower=0&gt; J;\n  array[J] int&lt;lower=0&gt; n_t;  // num cases, treatment\n  array[J] int&lt;lower=0&gt; r_t;  // num successes, treatment\n  array[J] int&lt;lower=0&gt; n_c;  // num cases, control\n  array[J] int&lt;lower=0&gt; r_c;  // num successes, control\n}\n\n\n\nAlthough the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio \\[\\begin{align*}\ny_j &= \\log \\left( \\frac{r^t_j / (n^t_j - r^t_j)}\n                        {r^c_j / (n^c_j - r^c_j)}\n        \\right) \\\\\n&= \\log \\left( \\frac{r^t_j}{n^t_j - r^t_j} \\right)\n   -\\log \\left( \\frac{r^c_j}{n^c_j - r^c_j} \\right)\n\\end{align*}\\] and corresponding standard errors \\[\n\\sigma_j = \\sqrt{\n  \\frac{1}{r^T_j}\n+ \\frac{1}{n^T_j - r^T_j}\n+ \\frac{1}{r^C_j}\n+ \\frac{1}{n^C_j - r^C_j}\n}.\n\\]\nThe log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.3\ntransformed data {\n  array[J] real y;\n  array[J] real&lt;lower=0&gt; sigma;\n  for (j in 1:J) {\n    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])\n            - (log(r_c[j]) - log(n_c[j] - r_c[j]));\n  }\n  for (j in 1:J) {\n    sigma[j] = sqrt(1 / r_t[j] + 1 / (n_t[j] - r_t[j])\n                     + 1 / r_c[j] + 1 / (n_c[j] - r_c[j]));\n  }\n}\nThis definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.\n\n\n\nWith the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.\nparameters {\n  real theta;  // global treatment effect, log odds\n}\nmodel {\n  y ~ normal(theta, sigma);\n}\nThe distribution statement for y is vectorized; it has the same effect as the following.\n  for (j in 1:J) {\n    y[j] ~ normal(theta, sigma[j]);\n  }\nIt is common to include a prior for theta in this model, but it is not strictly necessary for the model to be proper because y is fixed and \\(\\textsf{normal}(y \\mid \\mu,\\sigma) =\n\\textsf{normal}(\\mu \\mid y,\\sigma)\\).\n\n\n\nTo model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.\nparameters {\n  array[J] real theta;  // per-trial treatment effect\n  real mu;              // mean treatment effect\n  real&lt;lower=0&gt; tau;    // deviation of treatment effects\n}\nmodel {\n  y ~ normal(theta, sigma);\n  theta ~ normal(mu, tau);\n  mu ~ normal(0, 10);\n  tau ~ cauchy(0, 5);\n}\nAlthough the vectorized distribution statement for y appears unchanged, the parameter theta is now a vector. The distribution statement for theta is also vectorized, with the hyperparameters mu and tau themselves being given wide priors compared to the scale of the data.\nRubin (1981) provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.\n\n\n\nSmith, Spiegelhalter, and Thomas (1995) and Gelman et al. (2013, sec. 19.4) provide meta-analyses based directly on binomial data. Warn, Thompson, and Spiegelhalter (2002) consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.\nIf trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects \\(\\theta_j\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
     ]
   },
-  {
-    "objectID": "stan-users-guide/poststratification.html#bayesian-poststratification",
-    "href": "stan-users-guide/poststratification.html#bayesian-poststratification",
-    "title": "Poststratification",
+  {
+    "objectID": "stan-users-guide/measurement-error.html#bayesian-measurement-error-model",
+    "href": "stan-users-guide/measurement-error.html#bayesian-measurement-error-model",
+    "title": "Measurement Error and Meta-Analysis",
     "section": "",
-    "text": "Considering the same polling data from the previous section in a Bayesian setting, the uncertainty in the estimation of subgroup support is pushed through predictive inference in order to get some idea of the uncertainty of estimated support. Continuing the example of the previous section, the data model remains the same, \\[\ny_n \\sim \\textrm{bernoulli}(\\theta_{jj[n]}),\n\\] where \\(jj[n] \\in 1:J\\) is the group to which item \\(n\\) belongs and \\(\\theta_j\\) is the proportion of support in group \\(j\\).\nThis can be reformulated from a Bernoulli model to a binomial model in the usual way. Letting \\(A_j\\) be the number of respondents in group \\(j\\) and \\(a_j\\) be the number of positive responses in group \\(j\\), the data model may be reduced to the form \\[\na_j \\sim \\textrm{binomial}(A_j, \\theta_j).\n\\] A simple uniform prior on the proportion of support in each group completes the model, \\[\n\\theta_j \\sim \\textrm{beta}(1, 1).\n\\] A more informative prior could be used if there is prior information available about support among the student body.\nUsing sampling, draws \\(\\theta^{(m)} \\sim p(\\theta \\mid y)\\) from the posterior may be combined with the population sizes \\(N\\) to estimate \\(\\phi\\), the proportion of support in the population, \\[\n\\phi^{(m)}\n= \\frac{\\displaystyle \\sum_{j = 1}^J \\theta_j^{(m)} \\cdot N_j}\n       {\\displaystyle \\sum_{j = 1}^J N_j}.\n\\] The posterior draws for \\(\\phi^{(m)}\\) characterize expected support for the issue in the entire population. These draws may be used to estimate expected support (the average of the \\(\\phi^{(m)}\\)), posterior intervals (quantiles of the \\(\\phi^{(m)}\\)), or to plot a histogram.",
+    "text": "A Bayesian approach to measurement error can be formulated directly by treating the true quantities being measured as missing data (Clayton 1992; Richardson and Gilks 1993). This requires a model of how the measurements are derived from the true values.\n\n\nBefore considering regression with measurement error, first consider a linear regression model where the observed data for \\(N\\) cases includes a predictor \\(x_n\\) and outcome \\(y_n\\). In Stan, a linear regression for \\(y\\) based on \\(x\\) with a slope and intercept is modeled as follows.\ndata {\n  int&lt;lower=0&gt; N;       // number of cases\n  vector[N] x;          // predictor (covariate)\n  vector[N] y;          // outcome (variate)\n}\nparameters {\n  real alpha;           // intercept\n  real beta;            // slope\n  real&lt;lower=0&gt; sigma;  // outcome noise\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n  alpha ~ normal(0, 10);\n  beta ~ normal(0, 10);\n  sigma ~ cauchy(0, 5);\n}\nNow suppose that the true values of the predictors \\(x_n\\) are not known, but for each \\(n\\), a measurement \\(x^{\\textrm{meas}}_n\\) of \\(x_n\\) is available. If the error in measurement can be modeled, the measured value \\(x^{\\textrm{meas}}_n\\) can be modeled in terms of the true value \\(x_n\\) plus measurement noise. The true value \\(x_n\\) is treated as missing data and estimated along with other quantities in the model. A simple approach is to assume the measurement error is normal with known deviation \\(\\tau\\). This leads to the following regression model with constant measurement error.\ndata {\n  // ...\n  array[N] real x_meas;   // measurement of x\n  real&lt;lower=0&gt; tau;     // measurement noise\n}\nparameters {\n  array[N] real x;    // unknown true value\n  real mu_x;          // prior location\n  real sigma_x;       // prior scale\n  // ...\n}\nmodel {\n  x ~ normal(mu_x, sigma_x);  // prior\n  x_meas ~ normal(x, tau);    // measurement model\n  y ~ normal(alpha + beta * x, sigma);\n  // ...\n}\nThe regression coefficients alpha and beta and regression noise scale sigma are the same as before, but now x is declared as a parameter rather than as data. The data are now x_meas, which is a measurement of the true x value with noise scale tau. The model then specifies that the measurement error for x_meas[n] given true value x[n] is normal with deviation tau. Furthermore, the true values x are given a hierarchical prior here.\nIn cases where the measurement errors are not normal, richer measurement error models may be specified. The prior on the true values may also be enriched. For instance, Clayton (1992) introduces an exposure model for the unknown (but noisily measured) risk factors \\(x\\) in terms of known (without measurement error) risk factors \\(c\\). A simple model would regress \\(x_n\\) on the covariates \\(c_n\\) with noise term \\(\\upsilon\\), \\[\nx_n \\sim \\textsf{normal}(\\gamma^{\\top}c, \\upsilon).\n\\] This can be coded in Stan just like any other regression. And, of course, other exposure models can be provided.\n\n\n\nA common form of measurement error arises from rounding measurements. Rounding may be done in many ways, such as rounding weights to the nearest milligram, or to the nearest pound; rounding may even be done by rounding down to the nearest integer.\nExercise 3.5(b) by Gelman et al. (2013) provides an example.\n\n3.5. Suppose we weigh an object five times and measure weights, rounded to the nearest pound, of 10, 10, 12, 11, 9. Assume the unrounded measurements are normally distributed with a noninformative prior distribution on \\(\\mu\\) and \\(\\sigma^2\\).\n\nGive the correct posterior distribution for \\((\\mu, \\sigma^2)\\), treating the measurements as rounded.\n\n\nLetting \\(z_n\\) be the unrounded measurement for \\(y_n\\), the problem as stated assumes \\[\nz_n \\sim \\textsf{normal}(\\mu, \\sigma).\n\\]\nThe rounding process entails that \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\)1. The probability mass function for the discrete observation \\(y\\) is then given by marginalizing out the unrounded measurement, producing the likelihood \\[\\begin{align*}\np(y_n \\mid \\mu, \\sigma)\n&= \\int_{y_n - 0.5}^{y_n + 0.5} \\textsf{normal}(z_n \\mid \\mu, \\sigma) \\,\\textsf{d}z_n \\\\\n&= \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n   -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right).\n\\end{align*}\\] Gelman’s answer for this problem took the noninformative prior to be uniform in the variance \\(\\sigma^2\\) on the log scale, but we replace it with more recently recommended half-normal prior on \\(\\sigma\\) \\[\n\\sigma \\sim \\textsf{normal}^+(0, 1).\n\\] The posterior after observing \\(y = (10, 10, 12, 11, 9)\\) can be calculated by Bayes’s rule as \\[\\begin{align*}\np(\\mu, \\sigma \\mid y)\n&\\propto p(\\mu, \\sigma) \\ p(y \\mid \\mu, \\sigma) \\\\\n&\\propto \\textsf{normal}^+(\\sigma \\mid 0, 1)\\prod_{n=1}^5\n  \\left( \\Phi\\!\\left(\\frac{y_n + 0.5 - \\mu}{\\sigma}\\right)\n        -\\Phi\\!\\left(\\frac{y_n - 0.5 - \\mu}{\\sigma}\\right)\n  \\right).\n\\end{align*}\\]\nThe Stan code simply follows the mathematical definition, providing an example of the direct definition of a probability function up to a proportion.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  for (n in 1:N) {\n    target += log_diff_exp(normal_lcdf(y[n] + 0.5 | mu, sigma),\n                           normal_lcdf(y[n] - 0.5 | mu, sigma));\n  }\n}\nwhere normal_lcdf(y[n]+0.5 | mu, sigma) is equal to log(Phi((y[n] + 0.5 - mu) / sigma)), and log_diff_exp(a, b) computes log(exp(a) - exp(b)) in numerically more stable way.\nAlternatively, the model may be defined with latent parameters for the unrounded measurements \\(z_n\\). The Stan code in this case uses a distribution statement for \\(z_n\\) directly while respecting the constraint \\(z_n \\in (y_n - 0.5, y_n + 0.5)\\).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;lower=y-0.5, upper=y+0.5&gt;[N] z;\n}\nmodel {\n  sigma ~ normal(0, 1);\n  z ~ normal(mu, sigma);\n}\nThis explicit model for the unrounded measurements \\(z\\) produces the same posterior for \\(\\mu\\) and \\(\\sigma\\) as the previous model that marginalizes \\(z\\) out. Both approaches mix well, but the latent parameter version is about twice as efficient in terms of effective sample size per iteration, as well as providing a posterior for the unrounded parameters.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html#poststratification-in-stan",
-    "href": "stan-users-guide/poststratification.html#poststratification-in-stan",
-    "title": "Poststratification",
+    "objectID": "stan-users-guide/measurement-error.html#meta-analysis",
+    "href": "stan-users-guide/measurement-error.html#meta-analysis",
+    "title": "Measurement Error and Meta-Analysis",
     "section": "",
-    "text": "The maximum likelihood and Bayesian estimates can be handled with the same Stan program. The model of individual votes is collapsed to a binomial, where \\(A_j\\) is the number of voters from group \\(j\\), \\(a_j\\) is the number of positive responses from group \\(j\\), and \\(N_j\\) is the size of group \\(j\\) in the population.\ndata {\n  int&lt;lower=1&gt; J;\n  array[J] int&lt;lower=0&gt; A; \n  array[J] int&lt;lower=0&gt; a;\n  vector&lt;lower=0&gt;[J] N;\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[J] theta;\n}\nmodel {\n  a ~ binomial(A, theta);\n}\ngenerated quantities {t\n  real&lt;lower=0, upper=1&gt; phi = dot(N, theta) / sum(N);\n}\nThe binomial distribution statement is vectorized, and implicitly generates the joint likelihood for the \\(J\\) terms. The prior is implicitly uniform on \\((0, 1),\\) the support of \\(\\theta.\\) The summation is computed using a dot product and the sum function, which is why N was declared as a vector rather than as an array of integers.",
+    "text": "Meta-analysis aims to pool the data from several studies, such as the application of a tutoring program in several schools or treatment using a drug in several clinical trials.\nThe Bayesian framework is particularly convenient for meta-analysis, because each previous study can be treated as providing a noisy measurement of some underlying quantity of interest. The model then follows directly from two components, a prior on the underlying quantities of interest and a measurement-error style model for each of the studies being analyzed.\n\n\nSuppose the data in question arise from a total of \\(M\\) studies providing paired binomial data for a treatment and control group. For instance, the data might be post-surgical pain reduction under a treatment of ibuprofen (Warn, Thompson, and Spiegelhalter 2002) or mortality after myocardial infarction under a treatment of beta blockers (Gelman et al. 2013, sec. 5.6).\n\n\nThe clinical data consists of \\(J\\) trials, each with \\(n^t\\) treatment cases, \\(n^c\\) control cases, \\(r^t\\) successful outcomes among those treated and \\(r^c\\) successful outcomes among those in the control group. This data can be declared in Stan as follows.2\ndata {\n  int&lt;lower=0&gt; J;\n  array[J] int&lt;lower=0&gt; n_t;  // num cases, treatment\n  array[J] int&lt;lower=0&gt; r_t;  // num successes, treatment\n  array[J] int&lt;lower=0&gt; n_c;  // num cases, control\n  array[J] int&lt;lower=0&gt; r_c;  // num successes, control\n}\n\n\n\nAlthough the clinical trial data are binomial in its raw format, it may be transformed to an unbounded scale by considering the log odds ratio \\[\\begin{align*}\ny_j &= \\log \\left( \\frac{r^t_j / (n^t_j - r^t_j)}\n                        {r^c_j / (n^c_j - r^c_j)}\n        \\right) \\\\\n&= \\log \\left( \\frac{r^t_j}{n^t_j - r^t_j} \\right)\n   -\\log \\left( \\frac{r^c_j}{n^c_j - r^c_j} \\right)\n\\end{align*}\\] and corresponding standard errors \\[\n\\sigma_j = \\sqrt{\n  \\frac{1}{r^T_j}\n+ \\frac{1}{n^T_j - r^T_j}\n+ \\frac{1}{r^C_j}\n+ \\frac{1}{n^C_j - r^C_j}\n}.\n\\]\nThe log odds and standard errors can be defined in a transformed data block, though care must be taken not to use integer division.3\ntransformed data {\n  array[J] real y;\n  array[J] real&lt;lower=0&gt; sigma;\n  for (j in 1:J) {\n    y[j] = log(r_t[j]) - log(n_t[j] - r_t[j])\n            - (log(r_c[j]) - log(n_c[j] - r_c[j]));\n  }\n  for (j in 1:J) {\n    sigma[j] = sqrt(1 / r_t[j] + 1 / (n_t[j] - r_t[j])\n                     + 1 / r_c[j] + 1 / (n_c[j] - r_c[j]));\n  }\n}\nThis definition will be problematic if any of the success counts is zero or equal to the number of trials. If that arises, a direct binomial model will be required or other transforms must be used than the unregularized sample log odds.\n\n\n\nWith the transformed data in hand, two standard forms of meta-analysis can be applied. The first is a so-called “fixed effects” model, which assumes a single parameter for the global odds ratio. This model is coded in Stan as follows.\nparameters {\n  real theta;  // global treatment effect, log odds\n}\nmodel {\n  y ~ normal(theta, sigma);\n}\nThe distribution statement for y is vectorized; it has the same effect as the following.\n  for (j in 1:J) {\n    y[j] ~ normal(theta, sigma[j]);\n  }\nIt is common to include a prior for theta in this model, but it is not strictly necessary for the model to be proper because y is fixed and \\(\\textsf{normal}(y \\mid \\mu,\\sigma) =\n\\textsf{normal}(\\mu \\mid y,\\sigma)\\).\n\n\n\nTo model so-called “random effects,” where the treatment effect may vary by clinical trial, a hierarchical model can be used. The parameters include per-trial treatment effects and the hierarchical prior parameters, which will be estimated along with other unknown quantities.\nparameters {\n  array[J] real theta;  // per-trial treatment effect\n  real mu;              // mean treatment effect\n  real&lt;lower=0&gt; tau;    // deviation of treatment effects\n}\nmodel {\n  y ~ normal(theta, sigma);\n  theta ~ normal(mu, tau);\n  mu ~ normal(0, 10);\n  tau ~ cauchy(0, 5);\n}\nAlthough the vectorized distribution statement for y appears unchanged, the parameter theta is now a vector. The distribution statement for theta is also vectorized, with the hyperparameters mu and tau themselves being given wide priors compared to the scale of the data.\nRubin (1981) provided a hierarchical Bayesian meta-analysis of the treatment effect of Scholastic Aptitude Test (SAT) coaching in eight schools based on the sample treatment effect and standard error in each school.\n\n\n\nSmith, Spiegelhalter, and Thomas (1995) and Gelman et al. (2013, sec. 19.4) provide meta-analyses based directly on binomial data. Warn, Thompson, and Spiegelhalter (2002) consider the modeling implications of using alternatives to the log-odds ratio in transforming the binomial data.\nIf trial-specific predictors are available, these can be included directly in a regression model for the per-trial treatment effects \\(\\theta_j\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html#regression-and-poststratification",
-    "href": "stan-users-guide/poststratification.html#regression-and-poststratification",
-    "title": "Poststratification",
-    "section": "",
-    "text": "In applications to polling, there are often numerous demographic features like age, gender, income, education, state of residence, etc. If each of these demographic features induces a partition on the population, then their product also induces a partition on the population. Often sources such as the census have matching (or at least matchable) demographic data; otherwise it must be estimated.\nThe problem facing poststratification by demographic feature is that the number of strata increases exponentially as a function of the number of features. For instance, 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence leads to \\(5 \\cdot 2 \\cdot 5 \\cdot\n50 = 2000\\) strata. Adding another 5-way distinction, say for education level, leads to 10,000 strata. A simple model like the one in the previous section that takes an independent parameter \\(\\theta_j\\) for support in each stratum is unworkable in that many groups will have zero respondents and almost all groups will have very few respondents.\nA practical approach to overcoming the problem of low data size per stratum is to use a regression model. Each demographic feature will require a regression coefficient for each of its subgroups, but now the parameters add to rather than multiply the total number of parameters. For example, with 4 age brackets, 2 sexes, 5 income brackets, and 50 states of residence, there are only \\(4 + 2 + 5 + 50 =\n61\\) regression coefficients to estimate. Now suppose that item \\(n\\) has demographic features \\(\\textrm{age}_n \\in 1:5\\), \\(\\textrm{sex}_n \\in 1:2\\), \\(\\textrm{income}_n \\in 1:5,\\) and \\(\\textrm{state}_n \\in 1:50\\). A logistic regression may be formulated as \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)),\n\\] where \\(\\textrm{age}[n]\\) is the age of the \\(n\\)-th respondent, \\(\\textrm{sex}[n]\\) is their sex, \\(\\textrm{income}[n]\\) their income and \\(\\textrm{state}[n]\\) their state of residence. These coefficients can be assigned priors, resulting in a Bayesian regression model.\nTo poststratify the results, the population size for each combination of predictors must still be known. Then the population estimate is constructed as \\[\n\\sum_{i = 1}^5 \\sum_{j = 1}^2 \\sum_{k = 1}^5 \\sum_{m = 1}^{50}\n\\textrm{logit}^{-1}(\\alpha + \\beta_i + \\gamma_j + \\delta_k + \\eta_m)\n\\cdot \\textrm{pop}_{i, j, k, m},\n\\] where \\(\\textrm{pop}_{i, j, k, m}\\) is the size of the subpopulation with age \\(i\\), sex \\(j\\), income level \\(k\\), and state of residence \\(m\\).\nAs formulated, it should be clear that any kind of prediction could be used as a basis for poststratification. For example, a Gaussian process or neural network could be used to produce a non-parametric model of outcomes \\(y\\) given predictors \\(x\\).",
+    "objectID": "stan-users-guide/measurement-error.html#footnotes",
+    "href": "stan-users-guide/measurement-error.html#footnotes",
+    "title": "Measurement Error and Meta-Analysis",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThere are several different rounding rules (see, e.g., Wikipedia: Rounding), which affect which interval ends are open and which are closed, but these do not matter here as for continuous \\(z_n\\) \\(p(z_n=y_n-0.5)=p(z_n=y_n+0.5)=0\\).↩︎\nStan’s integer constraints are not powerful enough to express the constraint that \\(\\texttt{r}\\mathtt{\\_}\\texttt{t[j]} \\leq \\texttt{n}\\mathtt{\\_}\\texttt{t[j]}\\), but this constraint could be checked in the transformed data block.↩︎\nWhen dividing two integers, the result type is an integer and rounding will ensue if the result is not exact. See the discussion of primitive arithmetic types in the reference manual for more information.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Example Models",
+      "Measurement Error and Meta-Analysis"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html#multilevel-regression-and-poststratification",
-    "href": "stan-users-guide/poststratification.html#multilevel-regression-and-poststratification",
-    "title": "Poststratification",
+    "objectID": "stan-users-guide/multi-indexing.html",
+    "href": "stan-users-guide/multi-indexing.html",
+    "title": "Multiple Indexing and Range Indexing",
     "section": "",
-    "text": "With large numbers of demographic features, each cell may have very few items in it with which to estimate regression coefficients. For example, even in a national-level poll of 10,000 respondents, if they are divided by the 50 states, that’s only 200 respondents per state on average. When data sizes are small, parameter estimation can be stabilized and sharpened by providing hierarchical priors. With hierarchical priors, the data determines the amount of partial pooling among the groups. The only drawback is that if the number of groups is small, it can be hard to fit these models without strong hyperpriors.\nThe model introduced in the previous section had the data model \\[\ny_n \\sim\n\\textrm{bernoulli}(\\textrm{logit}^{-1}(\n\\alpha + \\beta_{\\textrm{age}[n]}\n+ \\gamma_{\\textrm{sex}[n]}\n+ \\delta_{\\textrm{income}[n]}\n+ \\epsilon_{\\textrm{state}[n]}\n)).\n\\] The overall intercept can be given a broad fixed prior, \\[\n\\alpha \\sim \\textrm{normal}(0, 5).\n\\] The other regression parameters can be given hierarchical priors, \\[\\begin{eqnarray*}\n\\beta_{1:4} & \\sim & \\textrm{normal}(0, \\sigma^{\\beta})\n\\\\[2pt]\n\\gamma_{1:2} & \\sim & \\textrm{normal}(0, \\sigma^{\\gamma})\n\\\\[2pt]\n\\delta_{1:5} & \\sim & \\textrm{normal}(0, \\sigma^{\\delta})\n\\\\[2pt]\n\\epsilon_{1:50} & \\sim & \\textrm{normal}(0, \\sigma^{\\epsilon}).\n\\end{eqnarray*}\\]\nThe hyperparameters for scale of variation within a group can be given simple standard hyperpriors, \\[\n\\sigma^{\\beta}, \\sigma^{\\gamma}, \\sigma^{\\delta}, \\sigma^{\\epsilon}\n\\sim \\textrm{normal}(0, 1).\n\\] The scales of these fixed hyperpriors need to be determined on a problem-by-problem basis, though ideally they will be close to standard (mean zero, unit variance).\n\n\nThe multilevel structure of the models used for multilevel regression and poststratification consist of a sum of intercepts that vary by demographic feature. This immediately introduces non-identifiability. A constant added to each state coefficient and subtracted from each age coefficient leads to exactly the same likelihood.\nThis is non-identifiability that is only mitigated by the (hierarchical) priors. When demographic partitions are small, as they are with several categories in the example, it can be more computationally tractable to enforce a sum-to-zero constraint on the coefficients. Other values than zero will by necessity be absorbed into the intercept, which is why it typically gets a broader prior even with standardized data. With a sum to zero constraint, coefficients for binary features will be negations of each other. For example, because there are only two sex categories, \\(\\gamma_2 =\n-\\gamma_1.\\)\nTo implement sum-to-zero constraints,\nparameters {\n  vector[K - 1] alpha_raw;\n// ...\n}\ntransformed parameters {\n  vector&lt;multiplier=sigma_alpha&gt;[K] alpha\n    = append_row(alpha_raw, -sum(alpha_raw));\n// ...    \n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n}\nThis prior is hard to interpret in that there are K normal distributions, but only K - 1 free parameters. An alternative is to put the prior only on alpha_raw, but that is also difficult to interpret.\nSoft constraints can be more computationally tractable. They are also simpler to implement.\nparameters {\n  vector&lt;multiplier=alpha&gt;[K] alpha;\n// ...\n}\nmodel {\n  alpha ~ normal(0, sigma_alpha);\n  sum(alpha) ~ normal(0, 0.001);\n}\nThis leaves the regular prior, but adds a second prior that concentrates the sum near zero. The scale of the second prior will need to be established on a problem and data-set specific basis so that it doesn’t shrink the estimates beyond the shrinkage of the hierarchical scale parameters.\nNote that in the hierarchical model, the values of the coefficients when there are only two coefficients should be the same absolute value but opposite signs. Any other difference could be combined into the overall intercept \\(\\alpha.\\) Even with a wide prior on the intercept, the hyperprior on \\(\\sigma^{\\gamma}\\) may not be strong enough to enforce that, leading to a weak form non-identifiability in the posterior. Enforcing a (hard or soft) sum-to-zero constraint can help mitigate non-identifiability. Whatever prior is chosen, prior predictive checks can help diagnose problems with it.\nNone of this work to manage identifiability in multilevel regressions has anything to do with the poststratification; it’s just required to fit a large multilevel regression with multiple discrete categories. Having multiple intercepts always leads to weak non-identifiability, even with the priors on the intercepts all centered at zero.",
+    "text": "Stan allows multiple indexes to be provided for containers (i.e., arrays, vectors, and matrices) in a single position, using either an array of integer indexes or range bounds. In many cases, there are functions that provide similar behavior.\nAllowing multiple indexes supports inline vectorization of models. For instance, consider the data model for a varying-slope, varying-intercept hierarchical linear regression, which could be coded as\nfor (n in 1:N) {\n  y[n] ~ normal(alpha[ii[n]] + beta[ii[n]] * x[n], sigma);\n}\nWith multiple indexing, this can be coded in one line, leading to more efficient vectorized code.\ny ~ normal(alpha[ii] + rows_dot_product(beta[ii], x), sigma);\nThis latter version is faster than the loop version; it is equivalent in speed to the clunky assignment to a local variable.\n{\n  vector[N] mu;\n  for (n in 1:N) {\n    mu[n] = alpha[ii[n]] + beta[ii[n]] * x[n];\n  }\n  y ~ normal(mu, sigma);\n}\nThe boost in speed compared to the original version is because the single call to the normal log density in the distribution statement will be much more memory efficient than the original version.\n\n\nThe following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.\narray[3] int c;\n// ... define: c == (5, 9, 7)\narray[4] int idxs;\n// ... define: idxs == (3, 3, 1, 2)\narray[4] int d;\nd = c[idxs];    // result: d == (7, 7, 5, 9)\nIn general, the multiple indexed expression c[idxs] is defined as follows, assuming idxs is of size K.\nc[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )\nThus c[idxs] is of the same size as idxs, which is K in this example.\nMultiple indexing can also be used with multi-dimensional arrays. For example, consider the following.\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[4] int idxs;\n// ... define: idxs = (2, 2, 1, 2)\narray[4, 3] int d\nd = c[idxs];    // result: d = ((7, 11, 13), (7, 11, 13),\n                //              (1, 3, 5), (7, 11, 13))\nThat is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by c[idxs][j] ==   c[idxs[j]].\nMultiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.\narray[4] int e;\ne = c[2, idxs]; // result:  c[2] = (7, 11, 13)\n                // result:  e = (11, 11, 7, 11)\nThe single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, c[2,idxs] evaluates to the same value as c[2][idxs].\nMultiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[3] int idxs1;\n// ... define: idxs1 = (2, 2, 1)\narray[2] int idxs2;\n// ... define: idxs2 = (1, 3)\narray[3, 2] int d;\nd = c[idxs1, idxs2];  // result: d = ((7, 13), (7, 13), (1, 5))\nWith multiple indexes, we no longer have c[idxs1, idxs2] being the same as c[idxs1][idxs2]. Rather, the entry d[i, j] after executing the above is given by\nd[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]\nThis example illustrates the operation of multiple indexing in the general case: a multiple index like idxs1 converts an index i used on the result (here, c[idxs1, idxs2]) to index idxs1[i] in the variable being indexed (here, c). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.\n\n\n\nSlicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.\n\n\nFor instance, consider supplying an upper and lower bound for an index.\narray[7] int c;\n// ...\narray[4] int d;\nd = c[3:6];  // result: d == (c[3], c[4], c[5], c[6])\nThe range index 3:6 behaves semantically just like the multiple index (3, 4, 5, 6). In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.\n\n\n\nIt is also possible to supply just a lower bound, or just an upper bound. Writing c[3:] is just shorthand for c[3:size(c)]. Writing c[:5] is just shorthand for c[1:5].\n\n\n\nFinally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (:) as the index or leaving the index position empty. In both cases, c[] and c[:] are equal to c[1:size(c)], which in turn is just equal to c.\n\n\n\nStan provides head and tail functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,\nvector[M] a = ...;\nvector[N] b = head(a, N);\nassigns b to be a vector equivalent to the first N elements of the vector a. The function tail works the same way for suffixes, with\narray[M] a = ...;\narray[N] b = tail(a, N);\nFinally, there is a segment function, which specifies a first element and number of elements. For example,\narray[15] a = ...;\narray[3] b = segment(a, 5, 3);\nwill set b to be equal to { a[5], a[6], a[7] }, so that it starts at element 5 of a and includes a total of 3 elements.\n\n\n\n\nMultiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.\narray[3] int a;\narray[2] int c;\narray[2] int idxs;\n// ... define: a == (1, 2, 3);  c == (5, 9)\n               //         idxs = (3,2)\na[idxs] = c;   // result: a == (1, 9, 5)\nThe result above can be worked out by noting that the assignment sets a[idxs[1]] (a[3]) to c[1] (5) and a[idxs[2]] (a[2]) to c[2] (9).\nThe same principle applies when there are many multiple indexes, as in the following example.\narray[5, 7] int a;\narray[2, 2] int c;\n// ...\na[2:3, 5:6] = c;  // result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]\n                  //         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]\nAs in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.\nUsage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.\narray[10, 13] int a;\narray[2] int c;\n// ...\na[4, 2:3] = c;  // result:  a[4, 2] == c[1];  a[4, 3] == c[2]\n\n\nAliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array a in the following code fragment.\narray[3] int a;\n// ... define: a == (5, 6, 7)\na[2:3] = a[1:2];\n// ... result: a == (5, 5, 6)\nThe reason the value of a after the assignment is \\((5,5,6)\\) rather than \\((5,5,5)\\) is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.\narray[3] int a;\narray[3] int idxs;\n// ... define idxs = (2, 1, 3)\na[idxs] = a;\nIn this case, it is evident why the right-hand side needs to be copied before the assignment.\nIt is tempting (but wrong) to think of the assignment a[2:3] =   a[1:2] as executing the following assignments.\n// ... define: a = (5, 6, 7)\na[2] = a[1];      // result: a = (5, 5, 7)\na[3] = a[2];      // result: a = (5, 5, 5)!\nThis produces a different result than executing the assignment because a[2]’s value changes before it is used.\n\n\n\n\nMultiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.\n\n\nVectors and row vectors behave exactly the same way as arrays with multiple indexes. If v is a vector, then v[3] is a scalar real value, whereas v[2:4] is a vector of size 3 containing the elements v[2], v[3], and v[4].\nThe only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.\narray[3] vector[5] v;\narray[7] int idxs;\n// ...\nvector[7] u;\nu = v[2, idxs];\n\narray[7] real w;\nw = v[idxs, 2];\nThe key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because v is an array of vectors, v[2, idxs] reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, v[idxs, 2] does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for w is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.\n\n\n\nMatrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.\nmatrix[5, 7] m;\n// ...\nrow_vector[3] rv;\nrv = m[4, 3:5];    // result is 1 x 3\n// ...\nvector[4] v;\nv = m[2:5, 3];     // result is 3 x 1\n// ...\nmatrix[3, 4] m2;\nm2 = m[1:3, 2:5];  // result is 3 x 4\nThe key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.\n\n\n\nIf matrices receive a single multiple index, the result is a matrix. So if m is a matrix, so is m[2:4]. In contrast, supplying a single index, m[3], produces a row vector result. That is, m[3] produces the same result as m[3, ] or m[3, 1:cols(m)].\n\n\n\nWith arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.\narray[5, 7] matrix[3, 4] m;\n// ...\narray[2] matrix[3, 4] a;\na = m[1, 2:3];  // knock off first array dimension\na = m[3:4, 5];  // knock off second array dimension\nIn both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, a[i] ==   m[1, i + 1], whereas in the second case, a[i] == m[i + 2,   5].\nContinuing the previous example, consider the following.\n// ...\nvector[2] b;\nb = a[1, 3, 2:3, 2];\nHere, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, b[j] == a[1, 3, 1 + j, 2].\nThis last example illustrates an important point: if there is a lower-bounded index, such as 2:3, with lower bound 2, then the lower bound minus one is added to the index, as seen in the 1 + j expression above.\nContinuing further, consider continuing with the following.\n// ...\narray[2] row_vector[3] c;\nc = a[4:5, 3, 1, 2: ];\nHere, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by c[i, j] == a[i + 3, 3, 1, j + 1]\n\n\n\nMatrix slicing can also be performed using the block function. For example,\nmatrix[20, 20] a = ...;\nmatrix[3, 2] b = block(a, 5, 9, 3, 2);\nwill set b equal to the submatrix of a starting at index [5, 9] and extending 3 rows and 2 columns. Thus block(a, 5, 9, 3, 2) is equivalent to b[5:7, 9:10].\nThe sub_col function extracts a slice of a column of a matrix as a vector. For example,\nmatrix[10, 10] a = ...;\nvector b = sub_col(a, 2, 3, 5);\nwill set b equal to the vector a[2:6, 3], taking the element starting at [2, 3], then extending for a total of 5 rows. The function sub_row works the same way for extracting a slice of a row as a row vector. For example, sub_row(a, 2, 3, 5) is equal to the row vector a[2, 3:7], which also starts at position [2, 3] then extends for a total of 5 columns.\n\n\n\n\nSuppose you have a \\(3 \\times 3\\) matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.\nSuppose a \\(3 \\times 3\\) matrix is known to be zero at indexes \\([1,2]\\) and \\([1,3]\\). The indexes for parameters are included in a “melted” data-frame or database format.\ntransformed data {\n  array[7, 2] int&lt;lower=1, upper=3&gt; idxs\n    = { {1, 1},\n        {2, 1}, {2, 2}, {2, 3},\n        {3, 1}, {3, 2}, {3, 3} };\n  // ...\nThe seven remaining parameters are declared as a vector.\nparameters {\n  vector[7] A_raw;\n  // ...\n}\nThen the full matrix A is constructed in the model block as a local variable.\nmodel {\n  matrix[3, 3] A;\n  for (i in 1:7) {\n    A[idxs[i, 1], idxs[i, 2]] = A_raw[i];\n  }\n  A[1, 2] = 0;\n  A[1, 3] = 0;\n  // ...\n}\nThis may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the idxs array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html#coding-mrp-in-stan",
-    "href": "stan-users-guide/poststratification.html#coding-mrp-in-stan",
-    "title": "Poststratification",
+    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexing",
+    "href": "stan-users-guide/multi-indexing.html#multiple-indexing",
+    "title": "Multiple Indexing and Range Indexing",
     "section": "",
-    "text": "Multilevel regression and poststratification can be coded directly in Stan. To code the non-centered parameterization for each coefficient, which will be required for sampling efficiency, the multiplier transform is used on each of the parameters. The combination of\nvector&lt;multiplier=s&gt;[K] a;\n// ...\na ~ normal(0, s);\nimplements a non-centered parameterization for a; a centered parameterization would drop the multiplier specification. The prior scale s is being centered here. The prior location is fixed to zero in multilevel regressions because there is an overall intercept; introducing a location parameters in the prior would introduce non-identifiability with the overall intercept. The centered parameterization drops the multiplier.\nHere is the full Stan model, which performs poststratification in the generated quantities using population sizes made available through data variable P.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=4&gt; age;\n  array[N] int&lt;lower=1, upper=5&gt; income;\n  array[N] int&lt;lower=1, upper=50&gt; state;\n  array[N] int&lt;lower=0&gt; y;\n  array[4, 5, 50] int&lt;lower=0&gt; P;\n}\nparameters {\n  real alpha;\n  real&lt;lower=0&gt; sigma_beta;\n  vector&lt;multiplier=sigma_beta&gt;[4] beta;\n  real&lt;lower=0&gt; sigma_gamma;\n  vector&lt;multiplier=sigma_gamma&gt;[5] gamma;\n  real&lt;lower=0&gt; sigma_delta;\n  vector&lt;multiplier=sigma_delta&gt;[50] delta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]);\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, sigma_beta);\n  gamma ~ normal(0, sigma_gamma);\n  delta ~ normal(0, sigma_delta);\n  { sigma_beta, sigma_gamma, sigma_delta } ~ normal(0, 1);\n}\ngenerated quantities {\n  real expect_pos = 0;\n  int total = 0;\n  for (b in 1:4) {\n    for (c in 1:5) {\n      for (d in 1:50) {\n        total += P[b, c, d];\n        expect_pos\n          += P[b, c, d]\n             * inv_logit(alpha + beta[b] + gamma[c] + delta[d]);\n      }\n    }\n  }\n  real&lt;lower=0, upper=1&gt; phi = expect_pos / total;\n}\nUnlike in posterior predictive inference aimed at uncertainty, there is no need to introduce binomial sampling uncertainty into the estimate of expected positive votes. Instead, generated quantities are computed as expectations. In general, it is more efficient to work in expectation if possible (the Rao-Blackwell theorem says it’s at least as efficient to work in expectation, but in practice, it can be much much more efficient, especially for discrete quantities).\n\n\nIn some cases, it can be more efficient to break the data down by group. Suppose there are \\(4 \\times 5 \\times 2 \\times 50 = 2000\\) groups. The data can be broken down into a size-2000 array, with entries corresponding to total vote counts in that group\nint&lt;lower=0&gt; G;\narray[G] int&lt;lower=1, upper=4&gt; age;\narray[G] int&lt;lower=1, upper=5&gt; income;\narray[G] int&lt;lower=1, upper=50&gt; state;\nThen the number of positive votes and the number of total votes are collected into two parallel arrays indexed by group.\narray[G] int&lt;lower=0&gt; pos_votes;\narray[G] int&lt;lower=0&gt; total_votes;\nFinally, the data model is converted to binomial.\npos_votes ~ binomial_logit(total_votes,\n                           alpha + beta[age] + ...);\nThe predictors look the same because of the way the age and other data items are coded.\n\n\n\nIn this first model, sex is not included as a predictor. With only two categories, it needs to be modeled separately, because it is not feasible to build a hierarchical model with only two cases. A sex predictor is straightforward to add to the data block; it takes on values 1 or 2 for each of the N data points.\n  array[N] int&lt;lower=1, upper=2&gt; sex;\nThen add a single regression coefficient as a parameter,\n  real epsilon;\nIn the log odds calculation, introduce a new term\n[epsilon, -epsilon][sex]';\nThat is, the data model will now look like\n  y ~ bernoulli_logit(alpha + beta[age] + gamma[income] + delta[state]\n                      + [epsilon, -epsilon][sex]');\nFor data point n, the expression [epsilon, -epsilon][sex] takes on value [epsilon, -epsilon][sex][n], which with Stan’s multi-indexing reduces to [epsilon, -epsilon][sex[n]]. This term evaluates to epsilon if sex[n] is 1 and to -epsilon if sex[n] is 2. The result is effectively a sum-to-zero constraint on two sex coefficients. The ' at the end transposes [epsilon, -epsilon][sex] which is a row_vector into a vector that can be added to the other variables.\nFinally, a prior is needed for the coefficient in the model block,\nepsilon ~ normal(0, 2);\nAs with other priors in multilevel models, the posterior for epsilon should be investigated to make sure it is not unrealistically wide.",
+    "text": "The following is the simplest concrete example of multiple indexing with an array of integers; the ellipses stand for code defining the variables as indicated in the comments.\narray[3] int c;\n// ... define: c == (5, 9, 7)\narray[4] int idxs;\n// ... define: idxs == (3, 3, 1, 2)\narray[4] int d;\nd = c[idxs];    // result: d == (7, 7, 5, 9)\nIn general, the multiple indexed expression c[idxs] is defined as follows, assuming idxs is of size K.\nc[idxs] = ( c[idxs[1]], c[idxs[2]], ..., c[idxs[K]] )\nThus c[idxs] is of the same size as idxs, which is K in this example.\nMultiple indexing can also be used with multi-dimensional arrays. For example, consider the following.\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[4] int idxs;\n// ... define: idxs = (2, 2, 1, 2)\narray[4, 3] int d\nd = c[idxs];    // result: d = ((7, 11, 13), (7, 11, 13),\n                //              (1, 3, 5), (7, 11, 13))\nThat is, putting an index in the first position acts exactly the same way as defined above. The fact that the values are themselves arrays makes no difference—the result is still defined by c[idxs][j] ==   c[idxs[j]].\nMultiple indexing may also be used in the second position of a multi-dimensional array. Continuing the above example, consider a single index in the first position and a multiple index in the second.\narray[4] int e;\ne = c[2, idxs]; // result:  c[2] = (7, 11, 13)\n                // result:  e = (11, 11, 7, 11)\nThe single index is applied, the one-dimensional result is determined, then the multiple index is applied to the result. That is, c[2,idxs] evaluates to the same value as c[2][idxs].\nMultiple indexing can apply to more than one position of a multi-dimensional array. For instance, consider the following\narray[2, 3] int c;\n// ... define: c = ((1, 3, 5), (7, 11, 13))\narray[3] int idxs1;\n// ... define: idxs1 = (2, 2, 1)\narray[2] int idxs2;\n// ... define: idxs2 = (1, 3)\narray[3, 2] int d;\nd = c[idxs1, idxs2];  // result: d = ((7, 13), (7, 13), (1, 5))\nWith multiple indexes, we no longer have c[idxs1, idxs2] being the same as c[idxs1][idxs2]. Rather, the entry d[i, j] after executing the above is given by\nd[i, j] == c[idxs1, idxs2][i, j] = c[idxs1[i], idxs2[j]]\nThis example illustrates the operation of multiple indexing in the general case: a multiple index like idxs1 converts an index i used on the result (here, c[idxs1, idxs2]) to index idxs1[i] in the variable being indexed (here, c). In contrast, a single index just returns the value at that index, thus reducing dimensionality by one in the result.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
     ]
   },
   {
-    "objectID": "stan-users-guide/poststratification.html#adding-group-level-predictors",
-    "href": "stan-users-guide/poststratification.html#adding-group-level-predictors",
-    "title": "Poststratification",
+    "objectID": "stan-users-guide/multi-indexing.html#slicing-with-range-indexes",
+    "href": "stan-users-guide/multi-indexing.html#slicing-with-range-indexes",
+    "title": "Multiple Indexing and Range Indexing",
     "section": "",
-    "text": "If there are group-level predictors, such as average income in a state, or vote share in a previous election, these may be used as predictors in the regression. They will not pose an obstacle to poststratification because they are at the group level. For example, suppose the average income level in the state is available as the data variable\narray[50] real&lt;lower=0&gt; income;\nthen a regression coefficient psi can be added for the effect of average state income,\nreal psi;\nwith a fixed prior,\npsi ~ normal(0, 2);\nThis prior assumes the income predictor has been standardized. Finally, a term is added to the regression for the fixed predictor,\ny ~ bernoulli_logit(alpha + beta[age] + ... + delta[state]\n                    + income[state] * psi);\nAnd finally, the formula in the generated quantities block is also updated,\nexpect_pos\n  += P[b, c, d]\n     * inv_logit(alpha + beta[b] + gamma[c] + delta[d]\n             + income[d] * psi);\nHere d is the loop variable looping over states. This ensures that the poststratification formula matches the model formula.",
+    "text": "Slicing returns a contiguous slice of a one-dimensional array, a contiguous sub-block of a two-dimensional array, and so on. Semantically, it is just a special form of multiple indexing.\n\n\nFor instance, consider supplying an upper and lower bound for an index.\narray[7] int c;\n// ...\narray[4] int d;\nd = c[3:6];  // result: d == (c[3], c[4], c[5], c[6])\nThe range index 3:6 behaves semantically just like the multiple index (3, 4, 5, 6). In terms of implementation, the sliced upper and/or lower bounded indices are faster and use less memory because they do not explicitly create a multiple index, but rather use a direct loop. They are also easier to read, so should be preferred over multiple indexes where applicable.\n\n\n\nIt is also possible to supply just a lower bound, or just an upper bound. Writing c[3:] is just shorthand for c[3:size(c)]. Writing c[:5] is just shorthand for c[1:5].\n\n\n\nFinally, it is possible to write a range index that covers the entire range of an array, either by including just the range symbol (:) as the index or leaving the index position empty. In both cases, c[] and c[:] are equal to c[1:size(c)], which in turn is just equal to c.\n\n\n\nStan provides head and tail functions that pull out prefixes or suffixes of vectors, row vectors, and one-dimensional arrays. In each case, the return type is the same as the argument type. For example,\nvector[M] a = ...;\nvector[N] b = head(a, N);\nassigns b to be a vector equivalent to the first N elements of the vector a. The function tail works the same way for suffixes, with\narray[M] a = ...;\narray[N] b = tail(a, N);\nFinally, there is a segment function, which specifies a first element and number of elements. For example,\narray[15] a = ...;\narray[3] b = segment(a, 5, 3);\nwill set b to be equal to { a[5], a[6], a[7] }, so that it starts at element 5 of a and includes a total of 3 elements.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Poststratification"
+      "Programming Techniques",
+      "Multiple Indexing and Range Indexing"
     ]
   },
   {
-    "objectID": "stan-users-guide/proportionality-constants.html",
-    "href": "stan-users-guide/proportionality-constants.html",
-    "title": "Proportionality Constants",
+    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexing-on-the-left-of-assignments",
+    "href": "stan-users-guide/multi-indexing.html#multiple-indexing-on-the-left-of-assignments",
+    "title": "Multiple Indexing and Range Indexing",
     "section": "",
-    "text": "When evaluating a likelihood or prior as part of the log density computation in MCMC, variational inference, or optimization, it is usually only necessary to compute the functions up to a proportionality constant (or similarly compute log densities up to an additive constant). In MCMC this comes from the fact that the distribution being sampled does not need to be normalized (and so it is the normalization constant that is ignored). Similarly the distribution does not need normalized to perform variational inference or do optimizations. The advantage of working with unnormalized distributions is they can make computation quite a bit cheaper.\nThere are three different syntaxes to build the model in Stan. The way to select between them is by determining if the proportionality constants are necessary. If performance is not a problem, it is always safe to use the normalized densities.\nThe distribution statement (~) and log density increment statement (target +=) with _lupdf() use unnormalized densities for \\(x\\) (dropping proportionality constants):\nx ~ normal(0, 1);\ntarget += normal_lupdf(x | 0, 1); // the 'u' is for unnormalized\nThe log density increment statement (target +=) with _lpdf() uses the full normalized density for \\(x\\) (dropping no constants):\ntarget += normal_lpdf(x | 0, 1);\nFor discrete distributions, the target += syntax is using _lupmf and _lpmf instead:\ny ~ bernoulli(0.5);\ntarget += bernoulli_lupmf(y | 0.5);\ntarget += bernoulli_lpmf(y | 0.5);\n\n\nIf a density \\(p(\\theta)\\) can be factored into \\(K g(\\theta)\\) where \\(K\\) are all the factors that are a not a function of \\(\\theta\\) and \\(g(\\theta)\\) are all the terms that are a function of \\(\\theta\\), then it is said that \\(g(\\theta)\\) is proportional to \\(p(\\theta)\\) up to a constant.\nThe advantage of all this is that sometimes \\(K\\) is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.\nStan takes advantage of the proportionality constant fact with the ~ syntax. Take for instance the normal data model:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nSyntactically, this is just shorthand for the equivalent model that replaces the ~ syntax with a target += statement and a normal_lupdf function call:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  target += normal_lupdf(x | mu, sigma);\n}\nThe function normal_lupdf is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\nNow because the density here is only a function of \\(x\\), the additive terms in the log density that are not a function of \\(x\\) can be dropped. In this case it is enough to know only the quadratic term:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\n\n\n\nIn the case that the proportionality constants were needed for a normal log density the function normal_lpdf can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the ~ or target += normal_lupdf syntaxes if it is absolutely clear that the proportionality constants are not necessary.\n\n\n\nWhen a custom _lpdf or _lpmf function is defined, the compiler will automatically make available a _lupdf or _lupmf version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.\nThe difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.\nThe following code demonstrates the different behaviors:\nfunctions {\n  real custom1_lpdf(x) {\n    return normal_lupdf(x | 0.0, 1.0);\n  }\n  real custom2_lpdf(x) {\n    return normal_lpdf(x | 0.0, 1.0);\n  }\n}\nparameters {\n  real mu;\n}\nmodel {\n  mu ~ custom1(); // Normalization constants dropped\n  target += custom1_lupdf(mu); // Normalization constants dropped\n  target += custom1_lpdf(mu);  // Normalization constants kept\n\n  mu ~ custom2();  // Normalization constants kept\n  target += custom2_lupdf(mu);  // Normalization constants kept\n  target += custom2_lpdf(mu);  // Normalization constants kept\n}\n\n\n\nTo avoid ambiguities in how the normalization constants work, functions ending in _lupdf and _lupmf can only be used in the model block or user-defined probability functions (functions ending in _lpdf or _lpmf).",
+    "text": "Multiple expressions may be used on the left-hand side of an assignment statement, where they work exactly the same way as on the right-hand side in terms of picking out entries of a container. For example, consider the following.\narray[3] int a;\narray[2] int c;\narray[2] int idxs;\n// ... define: a == (1, 2, 3);  c == (5, 9)\n               //         idxs = (3,2)\na[idxs] = c;   // result: a == (1, 9, 5)\nThe result above can be worked out by noting that the assignment sets a[idxs[1]] (a[3]) to c[1] (5) and a[idxs[2]] (a[2]) to c[2] (9).\nThe same principle applies when there are many multiple indexes, as in the following example.\narray[5, 7] int a;\narray[2, 2] int c;\n// ...\na[2:3, 5:6] = c;  // result: a[2, 5] == c[1, 1];  a[2, 6] == c[1, 2]\n                  //         a[3, 5] == c[2, 1];  a[3, 6] == c[2, 2]\nAs in the one-dimensional case, the right-hand side is written into the slice, block, or general chunk picked out by the left-hand side.\nUsage on the left-hand side allows the full generality of multiple indexing, with single indexes reducing dimensionality and multiple indexes maintaining dimensionality while rearranging, slicing, or blocking. For example, it is valid to assign to a segment of a row of an array as follows.\narray[10, 13] int a;\narray[2] int c;\n// ...\na[4, 2:3] = c;  // result:  a[4, 2] == c[1];  a[4, 3] == c[2]\n\n\nAliasing issues arise when there are references to the same data structure on the right-hand and left-hand side of an assignment. For example, consider the array a in the following code fragment.\narray[3] int a;\n// ... define: a == (5, 6, 7)\na[2:3] = a[1:2];\n// ... result: a == (5, 5, 6)\nThe reason the value of a after the assignment is \\((5,5,6)\\) rather than \\((5,5,5)\\) is that Stan behaves as if the right-hand side expression is evaluated to a fresh copy. As another example, consider the following.\narray[3] int a;\narray[3] int idxs;\n// ... define idxs = (2, 1, 3)\na[idxs] = a;\nIn this case, it is evident why the right-hand side needs to be copied before the assignment.\nIt is tempting (but wrong) to think of the assignment a[2:3] =   a[1:2] as executing the following assignments.\n// ... define: a = (5, 6, 7)\na[2] = a[1];      // result: a = (5, 5, 7)\na[3] = a[2];      // result: a = (5, 5, 5)!\nThis produces a different result than executing the assignment because a[2]’s value changes before it is used.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Proportionality Constants"
+      "Multiple Indexing and Range Indexing"
     ]
   },
   {
-    "objectID": "stan-users-guide/proportionality-constants.html#dropping-proportionality-constants",
-    "href": "stan-users-guide/proportionality-constants.html#dropping-proportionality-constants",
-    "title": "Proportionality Constants",
+    "objectID": "stan-users-guide/multi-indexing.html#multiple-indexes-with-vectors-and-matrices",
+    "href": "stan-users-guide/multi-indexing.html#multiple-indexes-with-vectors-and-matrices",
+    "title": "Multiple Indexing and Range Indexing",
     "section": "",
-    "text": "If a density \\(p(\\theta)\\) can be factored into \\(K g(\\theta)\\) where \\(K\\) are all the factors that are a not a function of \\(\\theta\\) and \\(g(\\theta)\\) are all the terms that are a function of \\(\\theta\\), then it is said that \\(g(\\theta)\\) is proportional to \\(p(\\theta)\\) up to a constant.\nThe advantage of all this is that sometimes \\(K\\) is expensive to compute and if it is not a function of the distribution that is to be sampled (or optimized or approximated with variational inference), there is no need to compute it because it will not affect the results.\nStan takes advantage of the proportionality constant fact with the ~ syntax. Take for instance the normal data model:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  x ~ normal(mu, sigma);\n}\nSyntactically, this is just shorthand for the equivalent model that replaces the ~ syntax with a target += statement and a normal_lupdf function call:\ndata {\n  real mu;\n  real&lt;lower=0.0&gt; sigma;\n}\nparameters {\n  real x;\n}\nmodel {\n  target += normal_lupdf(x | mu, sigma);\n}\nThe function normal_lupdf is only guaranteed to return the log density of the normal distribution up to a proportionality constant density to be sampled. The proportionality constant itself is not defined. The full log density of the statement here is:\n\\[\n\\textsf{normal\\_lpdf}(x | \\mu, \\sigma) =\n-\\log \\left( \\sigma \\sqrt{2 \\pi} \\right)\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]\nNow because the density here is only a function of \\(x\\), the additive terms in the log density that are not a function of \\(x\\) can be dropped. In this case it is enough to know only the quadratic term:\n\\[\n\\textsf{normal\\_lupdf}(x | \\mu, \\sigma) =\n-\\frac{1}{2} \\left( \\frac{x - \\mu}{\\sigma} \\right)^2.\n\\]",
+    "text": "Multiple indexes can be supplied to vectors and matrices as well as arrays of vectors and matrices.\n\n\nVectors and row vectors behave exactly the same way as arrays with multiple indexes. If v is a vector, then v[3] is a scalar real value, whereas v[2:4] is a vector of size 3 containing the elements v[2], v[3], and v[4].\nThe only subtlety with vectors is in inferring the return type when there are multiple indexes. For example, consider the following minimal example.\narray[3] vector[5] v;\narray[7] int idxs;\n// ...\nvector[7] u;\nu = v[2, idxs];\n\narray[7] real w;\nw = v[idxs, 2];\nThe key is understanding that a single index always reduces dimensionality, whereas a multiple index never does. The dimensions with multiple indexes (and unindexed dimensions) determine the indexed expression’s type. In the example above, because v is an array of vectors, v[2, idxs] reduces the array dimension but doesn’t reduce the vector dimension, so the result is a vector. In contrast, v[idxs, 2] does not reduce the array dimension, but does reduce the vector dimension (to a scalar), so the result type for w is an array of reals. In both cases, the size of the multiple index (here, 7) determines the size of the result.\n\n\n\nMatrices are a bit trickier because they have two dimensions, but the underlying principle of type inference is the same—multiple indexes leave dimensions in place, whereas single indexes reduce them. The following code shows how this works for multiple indexing of matrices.\nmatrix[5, 7] m;\n// ...\nrow_vector[3] rv;\nrv = m[4, 3:5];    // result is 1 x 3\n// ...\nvector[4] v;\nv = m[2:5, 3];     // result is 3 x 1\n// ...\nmatrix[3, 4] m2;\nm2 = m[1:3, 2:5];  // result is 3 x 4\nThe key is realizing that any position with a multiple index or bounded index remains in play in the result, whereas any dimension with a single index is replaced with 1 in the resulting dimensions. Then the type of the result can be read off of the resulting dimensionality as indicated in the comments above.\n\n\n\nIf matrices receive a single multiple index, the result is a matrix. So if m is a matrix, so is m[2:4]. In contrast, supplying a single index, m[3], produces a row vector result. That is, m[3] produces the same result as m[3, ] or m[3, 1:cols(m)].\n\n\n\nWith arrays of matrices, vectors, and row vectors, the basic access rules remain exactly the same: single indexes reduce dimensionality and multiple indexes redirect indexes. For example, consider the following example.\narray[5, 7] matrix[3, 4] m;\n// ...\narray[2] matrix[3, 4] a;\na = m[1, 2:3];  // knock off first array dimension\na = m[3:4, 5];  // knock off second array dimension\nIn both assignments, the multiple index knocks off an array dimension, but it’s different in both cases. In the first case, a[i] ==   m[1, i + 1], whereas in the second case, a[i] == m[i + 2,   5].\nContinuing the previous example, consider the following.\n// ...\nvector[2] b;\nb = a[1, 3, 2:3, 2];\nHere, the two array dimensions are reduced as is the column dimension of the matrix, leaving only a row dimension index, hence the result is a vector. In this case, b[j] == a[1, 3, 1 + j, 2].\nThis last example illustrates an important point: if there is a lower-bounded index, such as 2:3, with lower bound 2, then the lower bound minus one is added to the index, as seen in the 1 + j expression above.\nContinuing further, consider continuing with the following.\n// ...\narray[2] row_vector[3] c;\nc = a[4:5, 3, 1, 2: ];\nHere, the first array dimension is reduced, leaving a single array dimension, and the row index of the matrix is reduced, leaving a row vector. For indexing, the values are given by c[i, j] == a[i + 3, 3, 1, j + 1]\n\n\n\nMatrix slicing can also be performed using the block function. For example,\nmatrix[20, 20] a = ...;\nmatrix[3, 2] b = block(a, 5, 9, 3, 2);\nwill set b equal to the submatrix of a starting at index [5, 9] and extending 3 rows and 2 columns. Thus block(a, 5, 9, 3, 2) is equivalent to b[5:7, 9:10].\nThe sub_col function extracts a slice of a column of a matrix as a vector. For example,\nmatrix[10, 10] a = ...;\nvector b = sub_col(a, 2, 3, 5);\nwill set b equal to the vector a[2:6, 3], taking the element starting at [2, 3], then extending for a total of 5 rows. The function sub_row works the same way for extracting a slice of a row as a row vector. For example, sub_row(a, 2, 3, 5) is equal to the row vector a[2, 3:7], which also starts at position [2, 3] then extends for a total of 5 columns.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Proportionality Constants"
+      "Multiple Indexing and Range Indexing"
     ]
   },
   {
-    "objectID": "stan-users-guide/proportionality-constants.html#keeping-proportionality-constants",
-    "href": "stan-users-guide/proportionality-constants.html#keeping-proportionality-constants",
-    "title": "Proportionality Constants",
+    "objectID": "stan-users-guide/multi-indexing.html#matrices-with-parameters-and-constants",
+    "href": "stan-users-guide/multi-indexing.html#matrices-with-parameters-and-constants",
+    "title": "Multiple Indexing and Range Indexing",
     "section": "",
-    "text": "In the case that the proportionality constants were needed for a normal log density the function normal_lpdf can be used. For clarity, if there is ever a situation where it is unclear if the normalization is necessary, it should always be safe to include it. Only use the ~ or target += normal_lupdf syntaxes if it is absolutely clear that the proportionality constants are not necessary.",
+    "text": "Suppose you have a \\(3 \\times 3\\) matrix and know that two entries are zero but the others are parameters. Such a situation arises in missing data situations and in problems with fixed structural parameters.\nSuppose a \\(3 \\times 3\\) matrix is known to be zero at indexes \\([1,2]\\) and \\([1,3]\\). The indexes for parameters are included in a “melted” data-frame or database format.\ntransformed data {\n  array[7, 2] int&lt;lower=1, upper=3&gt; idxs\n    = { {1, 1},\n        {2, 1}, {2, 2}, {2, 3},\n        {3, 1}, {3, 2}, {3, 3} };\n  // ...\nThe seven remaining parameters are declared as a vector.\nparameters {\n  vector[7] A_raw;\n  // ...\n}\nThen the full matrix A is constructed in the model block as a local variable.\nmodel {\n  matrix[3, 3] A;\n  for (i in 1:7) {\n    A[idxs[i, 1], idxs[i, 2]] = A_raw[i];\n  }\n  A[1, 2] = 0;\n  A[1, 3] = 0;\n  // ...\n}\nThis may seem like overkill in this setting, but in more general settings, the matrix size, vector size, and the idxs array will be too large to code directly. Similar techniques can be used to build up matrices with ad-hoc constraints, such as a handful of entries known to be positive.",
     "crumbs": [
       "Stan Users Guide",
       "Programming Techniques",
-      "Proportionality Constants"
+      "Multiple Indexing and Range Indexing"
     ]
   },
   {
-    "objectID": "stan-users-guide/proportionality-constants.html#user-defined-distributions",
-    "href": "stan-users-guide/proportionality-constants.html#user-defined-distributions",
-    "title": "Proportionality Constants",
+    "objectID": "stan-users-guide/odes.html",
+    "href": "stan-users-guide/odes.html",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "When a custom _lpdf or _lpmf function is defined, the compiler will automatically make available a _lupdf or _lupmf version of the function. It is only possible to define custom distributions in the normalized form in Stan. Any attempt to define an unnormalized distribution directly will result in an error.\nThe difference in the normalized and unnormalized versions of custom probability functions is how probability functions are treated inside these functions. Any internal unnormalized probability function call will be replaced with its normalized equivalent if the normalized version of the parent custom distribution is called.\nThe following code demonstrates the different behaviors:\nfunctions {\n  real custom1_lpdf(x) {\n    return normal_lupdf(x | 0.0, 1.0);\n  }\n  real custom2_lpdf(x) {\n    return normal_lpdf(x | 0.0, 1.0);\n  }\n}\nparameters {\n  real mu;\n}\nmodel {\n  mu ~ custom1(); // Normalization constants dropped\n  target += custom1_lupdf(mu); // Normalization constants dropped\n  target += custom1_lpdf(mu);  // Normalization constants kept\n\n  mu ~ custom2();  // Normalization constants kept\n  target += custom2_lupdf(mu);  // Normalization constants kept\n  target += custom2_lpdf(mu);  // Normalization constants kept\n}",
+    "text": "Stan provides a number of different methods for solving systems of ordinary differential equations (ODEs). All of these methods adaptively refine their solutions in order to satisfy given tolerances, but internally they handle calculations quite a bit differently.\nBecause Stan’s algorithms requires gradients of the log density, the ODE solvers must not only provide the solution to the ODE itself, but also the gradient of the ODE solution with respect to parameters (the sensitivities). Two fundamentally different approaches are available in Stan to solve this problem, each having very different computational cost depending on the number of ODE states \\(N\\) and the number of parameters \\(M\\) being used:\n\nA forward sensitivity solver expands the base ODE system with additional ODE equations for the gradients of the solution. For each parameter, an additional full set of \\(N\\) sensitivity states are added meaning that the full ODE solved has \\(N \\, + N \\cdot M\\) states.\nAn adjoint sensitivity solver starts by solving the base ODE system forward in time to get the ODE solution and then solves another ODE system (the adjoint) backward in time to get the gradients. The forward and reverse solves both have \\(N\\) states each. There is additionally one quadrature problem solved for every parameter.\n\nThe adjoint sensitivity approach scales much better than the forward sensitivity approach. Whereas the computational cost of the forward approach scales multiplicatively in the number of ODE states \\(N\\) and parameters \\(M\\), the adjoint sensitivity approach scales linear in states \\(N\\) and parameters \\(M\\). However, the adjoint problem is harder to configure and the overhead for small problems actually makes it slower than solving the full forward sensitivity system. With that in mind, the rest of this introduction focuses on the forward sensitivity interfaces. For information on the adjoint sensitivity interface see the Adjoint ODE solver\nTwo interfaces are provided for each forward sensitivity solver: one with default tolerances and default max number of steps, and one that allows these controls to be modified. Choosing tolerances is important for making any of the solvers work well – the defaults will not work everywhere. The tolerances should be chosen primarily with consideration to the scales of the solutions, the accuracy needed for the solutions, and how the solutions are used in the model. For instance, if a solution component slowly varies between 3.0 and 5.0 and measurements of the ODE state are noisy, then perhaps the tolerances do not need to be as tight as for a situation where the solutions vary between 3.0 and 3.1 and very high precision measurements of the ODE state are available. It is also often useful to reduce the absolute tolerance when a component of the solution is expected to approach zero. For information on choosing tolerances, see the control parameters section.\nThe advantage of adaptive solvers is that as long as reasonable tolerances are provided and an ODE solver well-suited to the problem is chosen the technical details of solving the ODE can be abstracted away. The catch is that it is not always clear from the outset what reasonable tolerances are or which ODE solver is best suited to a problem. In addition, as changes are made to an ODE model, the optimal solver and tolerances may change.\nWith this in mind, the four forward solvers are rk45, bdf, adams, and ckrk. If no other information about the ODE is available, start with the rk45 solver. The list below has information on when each solver is useful.\nIf there is any uncertainty about which solver is the best, it can be useful to measure the performance of all the interesting solvers using profile statements. It is difficult to always know exactly what solver is the best in all situations, but a profile can provide a quick check.\n\nrk45: a fourth and fifth order Runge-Kutta method for non-stiff systems (Dormand and Prince 1980; Ahnert and Mulansky 2011). rk45 is the most generic solver and should be tried first.\nbdf: a variable-step, variable-order, backward-differentiation formula implementation for stiff systems (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). bdf is often useful for ODEs modeling chemical reactions.\nadams: a variable-step, variable-order, Adams-Moulton formula implementation for non-stiff systems (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). The method has order up to 12, hence is commonly used when high-accuracy is desired for a very smooth solution, such as in modeling celestial mechanics and orbital dynamics (Montenbruck and Gill 2000).\nckrk: a fourth and fifth order explicit Runge-Kutta method for non-stiff and semi-stiff systems (Cash and Karp 1990; Mazzia, Cash, and Soetaert 2012). The difference between ckrk and rk45 is that ckrk should perform better for systems that exhibit rapidly varying solutions. Often in those situations the derivatives become large or even nearly discontinuous, and ckrk is designed to address such problems.\n\nFor a discussion of stiff ODE systems, see the stiff ODE section. For information on the adjoint sensitivity interface see the Adjoint ODE solver section. The function signatures for Stan’s ODE solvers can be found in the function reference manual section on ODE solvers.\n\n\nAn ODE is defined by a set of differential equations, \\(y(t, \\theta)' = f(t, y, \\theta)\\), and initial conditions, \\(y(t_0, \\theta) = y_0\\). The function \\(f(t, y, \\theta)\\) is called the system function. The \\(\\theta\\) dependence is included in the notation for \\(y(t, \\theta)\\) and \\(f(t, y, \\theta)\\) as a reminder that the solution is a function of any parameters used in the computation.\n\n\n\nAs an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair \\(y = (y_1, y_2)\\) representing position and speed. The change in the system with respect to time is given by the following differential equations.1\n\\[\\begin{align*}\n&\\frac{d}{dt} y_1 = y_2  \\\\\n&\\frac{d}{dt} y_2 = -y_1 - \\theta y_2\n\\end{align*}\\]\nThe state equations implicitly defines the state at future times as a function of an initial state and the system parameters.\n\n\n\nThe first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the system function is time, passed as a real; the second argument to the system function is the system state, passed as a vector, and the return value from the system function are the current time derivatives of the state defined as a vector. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.\nThe simple harmonic oscillator can be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector sho(real t,        // time\n           vector y,      // state\n           real theta) {  // friction parameter\n  vector[2] dydt;\n  dydt[1] = y[2];\n  dydt[2] = -y[1] - theta * y[2];\n  return dydt;\n}\nThe function takes in a time t (a real), the system state y (a vector), and the parameter theta (a real). The function returns a vector of time derivatives of the system state at time t, state y, and parameter theta. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, t does not show up in the definition of dydt, however it is still required.\n\n\nThe types in the ODE system function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible ODE signatures:\nvector myode1(real t, vector y, real a0);\nvector myode2(real t, vector y, array[] int a0, vector a1);\nvector myode3(real t, vector y, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector myode1(real t, array[] real y, real a0);\n// Second argument is not a vector\narray[] real myode2(real t, vector y, real a0);\n// Return type is not a vector\nvector myode3(vector y, real a0);\n// First argument is not a real and second is not a vector\n\n\n\n\nNoisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.\n\n\nAs an example, suppose the simple harmonic oscillator has a parameter value of \\(\\theta = 0.15\\) and an initial state \\(y(t = 0, \\theta = 0.15) = (1, 0)\\). Assume the system is measured at 10 time points, \\(t = 1, 2, \\cdots, 10\\), where each measurement of \\(y(t, \\theta)\\) has independent \\(\\textsf{normal}(0, 0.1)\\) error in both dimensions (\\(y_1(t, \\theta)\\) and \\(y_2(t, \\theta)\\)).\nThe following model can be used to generate data like this:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  real t0;\n  array[T] real ts;\n  real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThe system parameters theta and initial state y0 are read in as data along with the initial time t0 and observation times ts. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations y_sim. Because the system is not stiff, the ode_rk45 solver is used.\nThis program illustrates the way in which the ODE solver is called in a Stan program,\narray[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\nthis returns the solution of the ODE initial value problem defined by system function sho, initial state y0, initial time t0, and parameter theta at the times ts. The call explicitly specifies the non-stiff RK45 solver.\nThe parameter theta is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, \\(\\theta\\) and \\(\\beta\\), the system function definition would look like:\nvector sho(real t, vector y, real theta, real beta) { ... }\nand the appropriate ODE solver call would be:\node_rk45(sho, y0, t0, ts, theta, beta);\nAny number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).\nBecause all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the T outputs with normal_rng.\n\n\n\nTypical realization of harmonic oscillator trajectory.\n\n\n\n\n\nThese ten noisy observations of the state can be used to estimate the friction parameter, \\(\\theta\\), the initial conditions, \\(y(t_0, \\theta)\\), and the scale of the noise in the problem. The full Stan model is:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  array[T] vector[2] y;\n  real t0;\n  array[T] real ts;\n}\nparameters {\n  vector[2] y0;\n  vector&lt;lower=0&gt;[2] sigma;\n  real theta;\n}\nmodel {\n  array[T] vector[2] mu = ode_rk45(sho, y0, t0, ts, theta);\n  sigma ~ normal(0, 2.5);\n  theta ~ std_normal();\n  y0 ~ std_normal();\n  for (t in 1:T) {\n    y[t] ~ normal(mu[t], sigma);\n  }\n}\nBecause the solves are now a function of model parameters, the ode_rk45 call is now made in the model block. There are half-normal priors on the measurement error scales sigma, and standard normal priors on theta and the initial state vector y0. The solutions to the ODE are assigned to mu, which is used as the location for the normal observation model.\nAs with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).\n\n\n\n\nStiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.\nStan provides a specialized solver for stiff ODEs (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the rk45 suffix is replaced with bdf, as in\node_bdf(sho, y0, t0, ts, theta);\nUsing the stiff (bdf) solver on a system that is not stiff may be much slower than using the non-stiff (rk45) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.\nIf it is not known for sure that an ODE system is stiff, run the model with both the rk45 and bdf solvers and see which is faster. If the rk45 solver is faster, then the problem is probably non-stiff, and then it makes sense to try the adams solver as well. The adams solver uses higher order methods which can take larger timesteps than the rk45 solver, though similar to the bdf solver each of these steps is more expensive to compute.\n\n\n\nFor additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters. These are the same as the regular function names but with _tol appended to the end. All three control arguments must be supplied with this signature (there are no defaults).\narray[T] vector[2] y_sim = ode_bdf_tol(sho, y0, t0, ts,\n                                 relative_tolerance,\n                                 absolute_tolerance,\n                                 max_num_steps,\n                                 theta);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nFor the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both \\(10^{-6}\\) and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million.\n\n\nIf there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.\nNonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times \\(0 &lt; t &lt; t'\\) and then positive for \\(t \\geq t'\\). In this example example, we would use code in the system such as\nif (t &lt; t_lag) {\n  return [0, 0]';\n} else {\n  // ... return non-zero vector...\n}\nIn general it is better to integrate up to t_lag in one solve and then integrate from t_lag onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.\nIf the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.\n\n\n\nThe relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns \\(y=(y_1,\\dots,y_n)^T\\), at every step the solver controls estimated local error \\(e=(e_1,\\dots,e_n)^T\\) through its weighted root-mean-square norm (Serban and Hindmarsh (2005), Hairer, Nørsett, and Wanner (1993))\n\\[\\begin{equation*}\n\\sqrt{\\sum_{i=1}^n{\\frac{1}{n}\\frac{e_i^2}{(\\text{RTOL}\\times y_i + \\text{ATOL})^2}}} &lt; 1\n\\end{equation*}\\] by reducing the stepsize when the inequality is not satisfied.\nTo understand the roles of the two tolerances it helps to assume \\(y\\) at opposite scales in the above expression: on one hand the absolute tolerance has little effect when \\(y_i \\gg 1\\), on the other the relative tolerance can not affect the norm when \\(y_i = 0\\). Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow Brenan, Campbell, and Petzold (1995) for a rule of thumb: let \\(m\\) be the number of significant digits required for \\(y\\), set \\(\\text{RTOL}=10^{-(m+1)}\\), and set ATOL at which \\(y\\) becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in bdf and adams solvers, and the same tolerances are used to control forward sensitivity calculation. See Serban and Hindmarsh (2005) for details.\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.\n\n\n\n\nThe adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with \\(N\\) additional states for each parameter for which gradients are needed. If there are \\(M\\) parameters for which sensitivities are required, then the augmented ODE system has a total of \\(N \\cdot (M +\n1)\\) states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.\nIn contrast, the adjoint ODE solver integrates forward in time a system of \\(N\\) equations to compute the ODE solution and then integrates backwards in time another system of \\(N\\) equations to get the sensitivities. Additionally, for \\(M\\) parameters there are \\(M\\) additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the bdf and adams forward sensitivity interfaces).\nThe solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.\nThe solver type (either bdf or adams) can be individually set for both the forward and backward solves.\nThe tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:\narray[T] vector[2] y_sim\n    = ode_adjoint_tol_ctl(sho, y0, t0, ts,\n                          relative_tolerance/9.0,                // forward tolerance\n                          rep_vector(absolute_tolerance/9.0, 2), // forward tolerance\n                          relative_tolerance/3.0,                // backward tolerance\n                          rep_vector(absolute_tolerance/3.0, 2), // backward tolerance\n                          relative_tolerance,                    // quadrature tolerance\n                          absolute_tolerance,                    // quadrature tolerance\n                          max_num_steps,\n                          150,                                   // number of steps between checkpoints\n                          1,                                     // interpolation polynomial: 1=Hermite, 2=polynomial\n                          2,                                     // solver for forward phase: 1=Adams, 2=BDF\n                          2,                                     // solver for backward phase: 1=Adams, 2=BDF\n                          theta);\nFor a detailed information on each argument please see the Stan function reference manual.\n\n\n\nLinear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.\nThe solution to \\(\\frac{d}{dt} y = ay\\) is \\(y = y_0e^{at}\\), where the constant \\(y_0\\) is determined by boundary conditions. We can extend this solution to the vector case: \\[\n\\frac{d}{dt}y = A \\, y\n\\]  where \\(y\\) is now a vector of length \\(n\\) and \\(A\\) is an \\(n\\) by \\(n\\) matrix. The solution is then given by: \\[\ny = e^{tA} \\, y_0\n\\]  where the matrix exponential is formally defined by the convergent power series: \\[\ne^{tA} = \\sum_{n=0}^{\\infty} \\dfrac{tA^n}{n!} = I + tA + \\frac{t^2A^2}{2!} + \\dotsb\n\\] \nWe can apply this technique to the simple harmonic oscillator example, by setting \\[\ny = \\begin{bmatrix} y_1 \\\\ y_2 \\end{bmatrix} \\qquad\nA = \\begin{bmatrix} 0 & 1 \\\\ -1 & -\\theta \\end{bmatrix}\n\\] \nThe Stan model to simulate noisy observations using a matrix exponential function is given below.\nIn general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  array[T] real ts;\n  array[1] real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim;\n  matrix[2, 2] A = [[ 0,  1],\n                    [-1, -theta[1]]]\n  for (t in 1:T) {\n    y_sim[t] = matrix_exp((t - 1) * A) * y0;\n  }\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThis Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters theta and initial state y0 are read in as data along observation times ts. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations y_sim. Because the ODEs are linear, we can use the matrix_exp function to solve the system.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Proportionality Constants"
+      "Example Models",
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/proportionality-constants.html#limitations-on-using-_lupdf-and-_lupmf-functions",
-    "href": "stan-users-guide/proportionality-constants.html#limitations-on-using-_lupdf-and-_lupmf-functions",
-    "title": "Proportionality Constants",
+    "objectID": "stan-users-guide/odes.html#notation",
+    "href": "stan-users-guide/odes.html#notation",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "To avoid ambiguities in how the normalization constants work, functions ending in _lupdf and _lupmf can only be used in the model block or user-defined probability functions (functions ending in _lpdf or _lpmf).",
+    "text": "An ODE is defined by a set of differential equations, \\(y(t, \\theta)' = f(t, y, \\theta)\\), and initial conditions, \\(y(t_0, \\theta) = y_0\\). The function \\(f(t, y, \\theta)\\) is called the system function. The \\(\\theta\\) dependence is included in the notation for \\(y(t, \\theta)\\) and \\(f(t, y, \\theta)\\) as a reminder that the solution is a function of any parameters used in the computation.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "Proportionality Constants"
+      "Example Models",
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html",
-    "href": "stan-users-guide/regression.html",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#example-simple-harmonic-oscillator",
+    "href": "stan-users-guide/odes.html#example-simple-harmonic-oscillator",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "Stan supports regression models from simple linear regressions to multilevel generalized linear models.\n\n\nThe simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\quad\\text{where}\\quad\n\\epsilon_n \\sim \\operatorname{normal}(0,\\sigma).\n\\]\nThis is equivalent to the following sampling involving the residual, \\[\ny_n - (\\alpha + \\beta X_n) \\sim \\operatorname{normal}(0,\\sigma),\n\\] and reducing still further, to \\[\ny_n \\sim \\operatorname{normal}(\\alpha + \\beta X_n, \\, \\sigma).\n\\]\nThis latter form of the model is coded in Stan as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\nThere are N observations and for each observation, \\(n \\in N\\), we have predictor x[n] and outcome y[n]. The intercept and slope parameters are alpha and beta. The model assumes a normally distributed noise term with scale sigma. This model has improper priors for the two regression coefficients.\n\n\nThe distribution statement in the previous model is vectorized, with\ny ~ normal(alpha + beta * x, sigma);\nproviding the same model as the unvectorized version,\nfor (n in 1:N) {\n  y[n] ~ normal(alpha + beta * x[n], sigma);\n}\nIn addition to being more concise, the vectorized form is much faster.1\nIn general, Stan allows the arguments to distributions such as normal to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.\nThe other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because x is of type vector and beta of type real, the expression beta * x is of type vector. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] beta;       // coefficients for predictors\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(x * beta + alpha, sigma);  // data model\n}\nThe constraint lower=0 in the declaration of sigma constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.\nIn the model above, x is an \\(N \\times K\\) matrix of predictors and beta a \\(K\\)-vector of coefficients, so x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. These predictions line up with the outcomes in the \\(N\\)-vector y, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix x to remove the alpha parameter.\nThe distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, sigma);\n  }\n}\nWith Stan’s matrix indexing scheme, x[n] picks out row n of the matrix x; because beta is a column vector, the product x[n] * beta is a scalar of type real.\n\n\nIn the model formulation\ny ~ normal(x * beta, sigma);\nthere is no longer an intercept coefficient alpha. Instead, we have assumed that the first column of the input matrix x is a column of 1 values. This way, beta[1] plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.\n\n\n\n\n\nIn the previous example, the linear predictor can be written as \\(\\eta\n= x \\beta\\), where \\(\\eta\\) is a \\(N\\)-vector of predictions, \\(x\\) is a \\(N\n\\times K\\) matrix, and \\(\\beta\\) is a \\(K\\)-vector of coefficients. Presuming \\(N \\geq K\\), we can exploit the fact that any design matrix \\(x\\) can be decomposed using the thin QR decomposition into an orthogonal matrix \\(Q\\) and an upper-triangular matrix \\(R\\), i.e. \\(x = Q\nR\\).\nThe functions qr_thin_Q and qr_thin_R implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using qr_Q and qr_R, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the qr_thin_Q and qr_thin_R functions). In practice, it is best to write \\(x = Q^\\ast\nR^\\ast\\) where \\(Q^\\ast = Q * \\sqrt{n - 1}\\) and \\(R^\\ast =\n\\frac{1}{\\sqrt{n - 1}} R\\). Thus, we can equivalently write \\(\\eta = x\n\\beta = Q R \\beta = Q^\\ast R^\\ast \\beta\\). If we let \\(\\theta = R^\\ast\n\\beta\\), then we have \\(\\eta = Q^\\ast \\theta\\) and \\(\\beta = R^{\\ast^{-1}}\n\\theta\\). In that case, the previous Stan program becomes\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\ntransformed data {\n  matrix[N, K] Q_ast;\n  matrix[K, K] R_ast;\n  matrix[K, K] R_ast_inverse;\n  // thin and scale the QR decomposition\n  Q_ast = qr_thin_Q(x) * sqrt(N - 1);\n  R_ast = qr_thin_R(x) / sqrt(N - 1);\n  R_ast_inverse = inverse(R_ast);\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] theta;      // coefficients on Q_ast\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(Q_ast * theta + alpha, sigma);  // data model\n}\ngenerated quantities {\n  vector[K] beta;\n  beta = R_ast_inverse * theta; // coefficients on x\n}\nSince this Stan program generates equivalent predictions for \\(y\\) and the same posterior distribution for \\(\\alpha\\), \\(\\beta\\), and \\(\\sigma\\) as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:\n\nThe columns of \\(Q^\\ast\\) are orthogonal whereas the columns of \\(x\\) generally are not. Thus, it is easier for a Markov Chain to move around in \\(\\theta\\)-space than in \\(\\beta\\)-space.\nThe columns of \\(Q^\\ast\\) have the same scale whereas the columns of \\(x\\) generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps\nSince the covariance matrix for the columns of \\(Q^\\ast\\) is an identity matrix, \\(\\theta\\) typically has a reasonable scale if the units of \\(y\\) are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.\n\nConsequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever \\(K &gt; 1\\) and you do not have an informative prior on the location of \\(\\beta\\). It can also be worthwhile to subtract the mean from each column of \\(x\\) before obtaining the QR decomposition, which does not affect the posterior distribution of \\(\\theta\\) or \\(\\beta\\) but does affect \\(\\alpha\\) and allows you to interpret \\(\\alpha\\) as the expectation of \\(y\\) in a linear model.\n\n\n\nSee our general discussion of priors for tips on priors for parameters in regression models.\nLater sections discuss univariate hierarchical priors and multivariate hierarchical priors, as well as priors used to identify models.\nHowever, as described in QR-reparameterization section, if you do not have an informative prior on the location of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.\n\n\n\nThe standard approach to linear regression is to model the noise term \\(\\epsilon\\) as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-\\(t\\) distribution. To code this in Stan, the distribution distribution is changed to the following.\ndata {\n  // ...\n  real&lt;lower=0&gt; nu;\n}\n// ...\nmodel {\n  y ~ student_t(nu, alpha + beta * x, sigma);\n}\nThe degrees of freedom constant nu is specified as data.\n\n\n\nFor binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in \\((-\\infty,\\infty)\\) to probability values in \\((0,1)\\). Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both S-shaped).\nA logistic regression model with one predictor and an intercept is coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta * x);\n}\nThe noise parameter is built into the Bernoulli formulation here rather than specified directly.\nLogistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by \\[\n\\operatorname{logit}(v) = \\log \\left( \\frac{v}{1-v} \\right).\n\\]\nThe inverse of the link function appears in the model: \\[\n\\operatorname{logit}^{-1}(u) = \\texttt{inv}\\mathtt{\\_}\\texttt{logit}(u) = \\frac{1}{1 + \\exp(-u)}.\n\\]\nThe model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha \\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe formulation is also vectorized in the sense that alpha and beta are scalars and x is a vector, so that alpha   + beta * x is a vector. The vectorized formulation is equivalent to the less efficient version\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(alpha + beta * x[n]);\n}\nExpanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable\nfor (n in 1:N) {\n  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));\n}\nOther link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as\n\\[\n\\Phi(x) = \\int_{-\\infty}^x \\textsf{normal}\\left(y \\mid 0,1 \\right) \\,\\textrm{d}y.\n\\]\nThe cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.\ny[n] ~ bernoulli(Phi(alpha + beta * x[n]));\nA fast approximation to the cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi_approx.2 The approximate probit regression model may be coded with the following.\ny[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));\n\n\n\nMultiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are \\(K\\) possible outcomes for each output variable \\(y_n\\). Also suppose that there is a \\(D\\)-dimensional vector \\(x_n\\) of predictors for \\(y_n\\). The multi-logit model with \\(\\textsf{normal}(0,5)\\) priors on the coefficients is coded as follows.\ndata {\n  int K;\n  int N;\n  int D;\n  array[N] int y;\n  matrix[N, D] x;\n}\nparameters {\n  matrix[D, K] beta;\n}\nmodel {\n  matrix[N, K] x_beta = x * beta;\n\n  to_vector(beta) ~ normal(0, 5);\n\n  for (n in 1:N) {\n    y[n] ~ categorical_logit(x_beta[n]');\n\n  }\n}\nwhere x_beta[n]' is the transpose of x_beta[n]. The prior on beta is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, \\[\n\\texttt{categorical}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{categorical}\\left(y \\mid \\texttt{softmax}(\\alpha)\\right),\n\\] where \\[\n\\texttt{softmax}(u) = \\exp(u) / \\operatorname{sum}\\left(\\exp(u)\\right).\n\\]\nThe categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing\ny[n] ~ categorical(softmax(x[n] * beta));\n\n\nThe data block in the above model is defined without constraints on sizes K, N, and D or on the outcome array y. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to\nint&lt;lower=2&gt; K;\nint&lt;lower=0&gt; N;\nint&lt;lower=1&gt; D;\narray[N] int&lt;lower=1, upper=K&gt; y;\nThese constraints arise because the number of categories, K, must be at least two in order for a categorical model to be useful. The number of data items, N, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of 1:N when N is equal to zero ensures the loop’s body will not be executed. The number of predictors, D, must be at least one in order for beta * x[n] to produce an appropriate argument for softmax(). The categorical outcomes y[n] must be between 1 and K in order for the discrete sampling to be well defined.\nConstraints on data declarations are optional. Constraints on parameters declared in the parameters block, on the other hand, are not optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.\n\n\n\nBecause softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.\nAn alternative is to use \\((K-1)\\)-vectors by fixing one of them to be zero. The partially known parameters section discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use \\((K - 1)\\)-vectors\nparameters {\n  matrix[D, K - 1] beta_raw;\n}\nand then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\nwhich can then be appended to beta_raw to produce the coefficient matrix beta,\ntransformed parameters {\n  matrix[D, K] beta = append_col(beta_raw, zeros);\n}\nThe rep_vector(0, D) call creates a column vector of size D with all entries set to zero. The derived matrix beta is then defined to be the result of appending the vector zeros as a new column at the end of beta_raw; the vector zeros is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.\nThis is not the same model as using \\(K\\)-vectors as parameters, because now the prior only applies to \\((K-1)\\)-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.\n\n\n\n\nWhen there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept \\(\\alpha\\) and then a varying effect \\(\\beta_k\\) for age group \\(k\\) to make a linear predictor \\(\\alpha +\n\\beta_k\\). With this predictor, we can add a constant to \\(\\alpha\\) and subtract from each \\(\\beta_k\\) and get exactly the same likelihood.\nThe traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., \\(\\beta_1 = 0\\). With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to \\(\\beta_1\\).\nIn a Bayesian setting, a proper prior on each of the \\(\\beta\\) is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the collinearity section.\nAn alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., \\(\\sum_{k=1}^K \\beta_k = 0.\\)\nA parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the multi-logit section for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the item-response model section for details).\n\n\nAs of Stan 2.36, there is a built in sum_to_zero_vector type, which can be used as follows.\nparameters {\n  sum_to_zero_vector[K] beta;\n  // ...\n}\nThis produces a vector of size K such that sum(beta) = 0. In the unconstrained representation requires only K - 1 values because the last is determined by the first K - 1.\nPlacing a prior on beta in this parameterization, for example,\n  beta ~ normal(0, 1);\nleads to a subtly different posterior than what you would get with the same prior on an unconstrained size-K vector. As explained below, the variance is reduced.\nThe sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., \\(\\beta_K\n= -\\sum_{k=1}^{K-1} \\beta_k.\\) But that leads to high correlation among the \\(\\beta_k\\).\nThe transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; Seyboldt (2024) provides an explanation. The Stan Reference Manual provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.\n\n\nOn the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of beta,\nmodel {\n  beta ~ normal(0, inv(sqrt(1 - inv(K))));\n  // ...\n}\nThe scale component can be multiplied by sigma to produce a normal(0, sigma) prior marginally.\nTo generate distributions with marginals other than standard normal, the resulting beta may be scaled by some factor sigma and translated to some new location mu.\n\n\n\n\nAdding a prior such as \\(\\beta \\sim \\textsf{normal}(0,\\epsilon)\\) for a small \\(\\epsilon\\) will provide a kind of soft centering of a parameter vector \\(\\beta\\) by preferring, all else being equal, that \\(\\sum_{k=1}^K\n\\beta_k = 0\\). This approach is only guaranteed to roughly center if \\(\\beta\\) and the elementwise addition \\(\\beta + c\\) for a scalar constant \\(c\\) produce the same likelihood (perhaps by another vector \\(\\alpha\\) being transformed to \\(\\alpha - c\\), as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an \\(\\epsilon\\). If \\(\\epsilon\\) is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.\n\n\n\n\nOrdered regression for an outcome \\(y_n \\in \\{ 1, \\dotsc, k \\}\\) with predictors \\(x_n \\in \\mathbb{R}^D\\) is determined by a single coefficient vector \\(\\beta \\in \\mathbb{R}^D\\) along with a sequence of cutpoints \\(c \\in\n\\mathbb{R}^{K-1}\\) sorted so that \\(c_d &lt; c_{d+1}\\). The discrete output is \\(k\\) if the linear predictor \\(x_n \\beta\\) falls between \\(c_{k-1}\\) and \\(c_k\\), assuming \\(c_0 = -\\infty\\) and \\(c_K = \\infty\\). The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.\n\n\nThe ordered logistic model can be coded in Stan using the ordered data type for the cutpoints and the built-in ordered_logistic distribution.\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  array[N] row_vector[D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ ordered_logistic(x[n] * beta, c);\n  }\n}\nThe vector of cutpoints c is declared as ordered[K - 1], which guarantees that c[k] is less than c[k + 1].\nIf the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.\n\n\nAn ordered probit model could be coded in exactly the same way by swapping the cumulative logistic (inv_logit) for the cumulative normal (Phi).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  array[N] row_vector[D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  vector[K] theta;\n  for (n in 1:N) {\n    real eta;\n    eta = x[n] * beta;\n    theta[1] = 1 - Phi(eta - c[1]);\n    for (k in 2:(K - 1)) {\n      theta[k] = Phi(eta - c[k - 1]) - Phi(eta - c[k]);\n    }\n    theta[K] = Phi(eta - c[K - 1]);\n    y[n] ~ categorical(theta);\n  }\n}\nThe logistic model could also be coded this way by replacing Phi with inv_logit, though the built-in encoding based on the softmax transform is more efficient and more numerically stable. A small efficiency gain could be achieved by computing the values Phi(eta - c[k]) once and storing them for re-use.\n\n\n\n\n\nThe simplest multilevel model is a hierarchical model in which the data are grouped into \\(L\\) distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients \\(\\beta\\). At the other extreme, an approach with no pooling assigns each level \\(l\\) its own coefficient vector \\(\\beta_l\\) that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.\nSuppose each binary outcome \\(y_n \\in \\{ 0, 1 \\}\\) has an associated level, \\(ll_n \\in \\{ 1, \\dotsc, L \\}\\). Each outcome will also have an associated predictor vector \\(x_n \\in \\mathbb{R}^D\\). Each level \\(l\\) gets its own coefficient vector \\(\\beta_l \\in \\mathbb{R}^D\\). The hierarchical structure involves drawing the coefficients \\(\\beta_{l,d}\n\\in \\mathbb{R}\\) from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.\nThe following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.\ndata {\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; L;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  array[N] int&lt;lower=1, upper=L&gt; ll;\n  array[N] row_vector[D] x;\n}\nparameters {\n  array[D] real mu;\n  array[D] real&lt;lower=0&gt; sigma;\n  array[L] vector[D] beta;\n}\nmodel {\n  for (d in 1:D) {\n    mu[d] ~ normal(0, 100);\n    for (l in 1:L) {\n      beta[l, d] ~ normal(mu[d], sigma[d]);\n    }\n  }\n  for (n in 1:N) {\n    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));\n  }\n}\nThe standard deviation parameter sigma gets an implicit uniform prior on \\((0,\\infty)\\) because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the regression priors section for recommendations on priors for regression coefficients and scales.\n\n\nWhere possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.\nThe first optimization vectorizes the for-loop over D as\nmu ~ normal(0, 100);\nfor (l in 1:L) {\n  beta[l] ~ normal(mu, sigma);\n}\nThe declaration of beta as an array of vectors means that the expression beta[l] denotes a vector. Although beta could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the indexing efficiency section for more information on the efficiency tradeoffs among arrays, vectors, and matrices.\nThis model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,3\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);\n}\nUnlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.\n{\n  vector[N] x_beta_ll;\n  for (n in 1:N) {\n    x_beta_ll[n] = x[n] * beta[ll[n]];\n  }\n  y ~ bernoulli_logit(x_beta_ll);\n}\nThe brackets introduce a new scope for the local variable x_beta_ll; alternatively, the variable may be declared at the top of the model block.\nIn some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.\n\n\n\n\nPriors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.\n\n\nThe fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of \\(y_n \\in\n\\mathbb{R}\\) on \\(x_n \\in \\mathbb{R}^K\\), formulated as \\[\\begin{align*}\ny_n     & \\sim \\textsf{normal}(x_n \\beta, \\sigma) \\\\\n\\beta_k & \\sim \\textsf{normal}(0,\\tau) \\\\\n\\tau    & \\sim \\textsf{Cauchy}(0,2.5) \\\\\n\\end{align*}\\]\nIn this case, as \\(\\tau \\rightarrow 0\\) and \\(\\beta_k \\rightarrow 0\\), the posterior density \\[ p(\\beta,\\tau,\\sigma|y,x) \\propto p(y|x,\\beta,\\tau,\\sigma) \\] grows without bound. See the Neal’s funnel density, which has similar behavior.\nThere is obviously no MLE estimate for \\(\\beta,\\tau,\\sigma\\) in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by Chung et al. (2013) is to use a gamma distribution as a prior, such as \\[\n\\sigma \\sim \\textsf{Gamma}(2, 1/A),\n\\] for a reasonably large value of \\(A\\), such as \\(A = 10\\).\n\n\n\n\nItem-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see Gelman and Hill (2007, pps. 314–320) for a textbook introduction to hierarchical IRT models and Curtis (2010) for encodings of a range of IRT models in BUGS.\n\n\nThe data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.\ndata {\n  int&lt;lower=1&gt; J;                     // number of students\n  int&lt;lower=1&gt; K;                     // number of questions\n  int&lt;lower=1&gt; N;                     // number of observations\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // student for observation n\n  array[N] int&lt;lower=1, upper=K&gt; kk;  // question for observation n\n  array[N] int&lt;lower=0, upper=1&gt; y;   // correctness for observation n\n}\nThis declares a total of N student-question pairs in the data set, where each n in 1:N indexes a binary observation y[n] of the correctness of the answer of student jj[n] on question kk[n].\nThe prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.\n\n\n\nThe 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).\nThe model parameters are declared as follows.\nparameters {\n  real delta;            // mean student ability\n  array[J] real alpha;   // ability of student j - mean ability\n  array[K] real beta;    // difficulty of question k\n}\nThe parameter alpha[J] is the ability coefficient for student j and beta[k] is the difficulty coefficient for question k. The non-standard parameterization used here also includes an intercept term delta, which represents the average student’s response to the average question.4\nThe model itself is as follows.\nmodel {\n  alpha ~ std_normal();         // informative true prior\n  beta ~ std_normal();          // informative true prior\n  delta ~ normal(0.75, 1);      // informative true prior\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);\n  }\n}\nThis model uses the logit-parameterized Bernoulli distribution, where \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe key to understanding it is the term inside the bernoulli_logit distribution, from which it follows that \\[\n\\Pr[y_n = 1] = \\operatorname{logit}^{-1}\\left(\\alpha_{jj[n]} - \\beta_{kk[n]}\n+ \\delta\\right).\n\\]\nThe model suffers from additive identifiability issues without the priors. For example, adding a term \\(\\xi\\) to each \\(\\alpha_j\\) and \\(\\beta_k\\) results in the same predictions. The use of priors for \\(\\alpha\\) and \\(\\beta\\) located at 0 identifies the parameters; see Gelman and Hill (2007) for a discussion of identifiability issues and alternative approaches to identification.\nFor testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.\n\n\n\nThe simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.\nparameters {\n  real mu_beta;                // mean question difficulty\n  vector[J] alpha;             // ability for j - mean\n  vector[K] beta;              // difficulty for k\n  vector&lt;lower=0&gt;[K] gamma;    // discrimination of k\n  real&lt;lower=0&gt; sigma_beta;    // scale of difficulties\n  real&lt;lower=0&gt; sigma_gamma;   // scale of log discrimination\n}\nThe parameters should be clearer after the model definition.\nmodel {\n  alpha ~ std_normal();\n  beta ~ normal(0, sigma_beta);\n  gamma ~ lognormal(0, sigma_gamma);\n  mu_beta ~ cauchy(0, 5);\n  sigma_beta ~ cauchy(0, 5);\n  sigma_gamma ~ cauchy(0, 5);\n  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));\n}\nThe std_normal function is used here, defined by \\[\n\\texttt{std}\\mathtt{\\_}\\texttt{normal}(y)\n=\n\\textsf{normal}\\left(y \\mid 0, 1\\right).\n\\]\nThe distribution statement is also vectorized using elementwise multiplication; it is equivalent to\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(gamma[kk[n]]\n                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));\n}\nThe 2PL model is similar to the 1PL model, with the additional parameter gamma[k] modeling how discriminative question k is. If gamma[k] is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter gamma[k] is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.\nThe model is parameterized here with student abilities alpha being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the problematic posteriors chapter for further discussion of identifiability. The difficulty and discrimination parameters beta and gamma then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as\nbeta ~ normal(0, 5);\ngamma ~ lognormal(0, 2);\nThe point is that the alpha determines the scale and location and beta and gamma are allowed to float.\nThe beta parameter is here given a non-centered parameterization, with parameter mu_beta serving as the mean beta location. An alternative would’ve been to take:\nbeta ~ normal(mu_beta, sigma_beta);\nand\ny[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));\nNon-centered parameterizations tend to be more efficient in hierarchical models; see the reparameterization section for more information on non-centered reparameterizations.\nThe intercept term mu_beta can’t itself be modeled hierarchically, so it is given a weakly informative \\(\\textsf{Cauchy}(0,5)\\) prior. Similarly, the scale terms, sigma_beta, and sigma_gamma, are given half-Cauchy priors. As mentioned earlier, the scale and location for alpha are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to \\((0,\\infty)\\) by their declarations.\n\n\n\n\n\n\nOne application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the collinearity section for a simple example of the problems this poses for estimation.\nThe non-identifiability is resolved by providing a standard normal (i.e., \\(\\textsf{normal}(0,1)\\)) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.\n\n\n\nAnother case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.\n\n\n\nIn a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.\nSimilar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.\n\n\n\n\nIn hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.\nAs an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.\n\n\nGelman and Hill (2007, chap. 13, Chapter 17) provide a discussion of a hierarchical model with \\(N\\) individuals organized into \\(J\\) groups. Each individual has a predictor row vector \\(x_n\\) of size \\(K\\); to unify the notation, they assume that \\(x_{n,1} = 1\\) is a fixed “intercept” predictor. To encode group membership, they assume individual \\(n\\) belongs to group \\(jj[n] \\in \\{ 1, \\dotsc, J \\}\\). Each individual \\(n\\) also has an observed outcome \\(y_n\\) taking on real values.\n\n\nThe model is a linear regression with slope and intercept coefficients varying by group, so that \\(\\beta_j\\) is the coefficient \\(K\\)-vector for group \\(j\\). The data model for individual \\(n\\) is then just \\[\ny_n \\sim \\textsf{normal}(x_n \\, \\beta_{jj[n]}, \\, \\sigma)\n\\quad\\text{for}\\quad n \\in \\{ 1, \\dotsc, N \\}.\n\\]\n\n\n\nGelman and Hill model the coefficient vectors \\(\\beta_j\\) as being drawn from a multivariate distribution with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\), \\[\n\\beta_j \\sim \\textsf{multivariate normal}(\\mu_j, \\, \\Sigma)\n\\quad\\text{for}\\quad j \\in \\{ 1, \\dotsc, J \\}.\n\\]\nBelow, we discuss the full model of Gelman and Hill, which uses group-level predictors to model \\(\\mu\\); for now, we assume \\(\\mu\\) is a simple vector parameter.\n\n\n\nFor hierarchical modeling, the group-level mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\) must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as \\[\n\\mu_j \\sim \\textsf{normal}(0,5).\n\\] If more is known about the expected coefficient values \\(\\beta_{j, k}\\), this information can be incorporated into the prior for \\(\\mu_j\\).\nFor the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling\nIn Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define \\[\n\\Sigma = \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau) \\times \\Omega \\times \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau),\n\\] where \\(\\Omega\\) is a correlation matrix and \\(\\tau\\) is the vector of coefficient scales. This mapping from scale vector \\(\\tau\\) and correlation matrix \\(\\Omega\\) can be inverted, using \\[\n\\tau_k = \\sqrt{\\Sigma_{k,k}}\n\\quad\\textsf{and}\\quad\n\\Omega_{i, j} = \\frac{\\Sigma_{i, j}}{\\tau_i \\, \\tau_j}.\n\\]\nThe components of the scale vector \\(\\tau\\) can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as \\[\n\\tau_k \\sim \\textsf{Cauchy}(0, 2.5)\n\\quad\\text{for}\\quad k \\in \\{ 1, \\dotsc, K \\}\n\\quad\\text{constrained\\ by}\\quad \\tau_k &gt; 0.\n\\] As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for \\(\\tau\\). For large numbers of exchangeable coefficients, the components of \\(\\tau\\) itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.\nOur final recommendation is to give the correlation matrix \\(\\Omega\\) an LKJ prior with shape \\(\\eta \\geq 1\\),5\n\\[\n\\Omega \\sim \\textsf{LKJCorr}(\\eta).\n\\]\nThe LKJ correlation distribution is defined by \\[\n\\textsf{LKJCorr}\\left(\\Sigma \\mid \\eta\\right)\n\\propto\n\\operatorname{det}\\left(\\Sigma\\right)^{\\eta - 1}.\n\\]\nThe basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For \\(\\eta = 1\\), the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.\nFor \\(\\eta &gt; 1\\), the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For \\(\\eta &lt; 1\\), it increasingly concentrates mass in the other direction, i.e., favoring more correlation.\nThe LKJ prior may thus be used to control the expected amount of correlation among the parameters \\(\\beta_j\\). For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see Barnard, McCulloch, and Meng (2000).\n\n\n\nTo complete Gelman and Hill’s model, suppose each group \\(j \\in \\{ 1, \\dotsc, J \\}\\) is supplied with an \\(L\\)-dimensional row-vector of group-level predictors \\(u_j\\). The prior mean for the \\(\\beta_j\\) can then itself be modeled as a regression, using an \\(L\\)-dimensional coefficient vector \\(\\gamma\\). The prior for the group-level coefficients then becomes \\[\n\\beta_j \\sim \\textsf{multivariate normal}(u_j \\, \\gamma, \\Sigma)\n\\]\nThe group-level coefficients \\(\\gamma\\) may themselves be given independent weakly informative priors, such as \\[\n\\gamma_l \\sim \\textsf{normal}(0,5).\n\\] As usual, information about the group-level means should be incorporated into this prior.\n\n\n\nThe Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.\ndata {\n  int&lt;lower=0&gt; N;              // num individuals\n  int&lt;lower=1&gt; K;              // num ind predictors\n  int&lt;lower=1&gt; J;              // num groups\n  int&lt;lower=1&gt; L;              // num group predictors\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // group for individual\n  matrix[N, K] x;              // individual predictors\n  array[J] row_vector[L] u;    // group predictors\n  vector[N] y;                 // outcomes\n}\nparameters {\n  corr_matrix[K] Omega;        // prior correlation\n  vector&lt;lower=0&gt;[K] tau;      // prior scale\n  matrix[L, K] gamma;          // group coeffs\n  array[J] vector[K] beta;     // indiv coeffs by group\n  real&lt;lower=0&gt; sigma;         // prediction error scale\n}\nmodel {\n  tau ~ cauchy(0, 2.5);\n  Omega ~ lkj_corr(2);\n  to_vector(gamma) ~ normal(0, 5);\n  {\n    array[J] row_vector[K] u_gamma;\n    for (j in 1:J) {\n      u_gamma[j] = u[j] * gamma;\n    }\n    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));\n  }\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n  }\n}\nThe hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix Omega and scale vector tau are more natural to inspect in the output; to output Sigma, define it as a transformed parameter. The function quad_form_diag is defined so that quad_form_diag(Sigma, tau) is equivalent to diag_matrix(tau) * Sigma * diag_matrix(tau), where diag_matrix(tau) returns the matrix with tau on the diagonal and zeroes off diagonal; the version using quad_form_diag should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.\n\n\n\nThe code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:\nfor (n in 1:N) {\n  y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n}\nwith the vectorized distribution statement:\n{\n  vector[N] x_beta_jj;\n  for (n in 1:N) {\n    x_beta_jj[n] = x[n] * beta[jj[n]];\n  }\n  y ~ normal(x_beta_jj, sigma);\n}\nThe outer brackets create a local scope in which to define the variable x_beta_jj, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.\nAlthough it is tempting to redeclare beta and include a revised model block distribution statement,\nparameters {\n  matrix[J, K] beta;\n// ...\n}\nmodel {\n  y ~ normal(rows_dot_product(x, beta[jj]), sigma);\n  // ...\n}\nthis fails because it breaks the vectorization for beta,6\nbeta ~ multi_normal(...);\nwhich requires beta to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because rows_dot_product cannot do much optimization in and of itself because there are no shared computations.\nThe code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.\n{\n  matrix[K, K] Sigma_beta;\n  Sigma_beta = quad_form_diag(Omega, tau);\n  for (j in 1:J) {\n    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);\n  }\n}\nIn this example, the covariance matrix Sigma_beta is defined as a local variable so as not to have to repeat the quadratic form computation \\(J\\) times. This vectorization can be combined with the Cholesky-factor optimization in the next section.\n\n\n\nThe multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.\ndata {\n  matrix[L, J] u;              // group predictors transposed\n  // ...\n}\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  matrix[K, L] gamma;\n  // ...\n}\ntransformed parameters {\n  matrix[K, J] beta;\n  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  // ...\n}\nThe data variable u was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with gamma and beta, to minimize the number of transposition operations. The new parameter L_Omega is the Cholesky factor of the original correlation matrix Omega, so that\nOmega = L_Omega * L_Omega'\nThe prior scale vector tau is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,\nSigma_beta\n  = quad_form_diag(Omega, tau)\n  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'\nwhere the diagonal pre-multiply compound operation is defined by\ndiag_pre_multiply(a, b) = diag_matrix(a) * b\nThe new variable z is declared as a matrix, the entries of which are given independent standard normal priors; the to_vector operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of z being a \\(K\\)-variate normal random vector with the identity as covariance matrix. Therefore, multiplying z by the Cholesky factor of the covariance matrix and adding the mean (u * gamma)' produces a beta distributed as in the original model, where the variance is, letting \\(L = \\mathrm{diag}(\\tau)\\,\\Omega_L\\),\n\\[\n\\begin{aligned}\n\\mathbb{V}[\\beta] &= \\mathbb{E}\\big((L \\, z)(L \\, z)^\\top) \\\\\n&= \\mathbb{E}\\big((L \\, z \\, z^\\top \\, L^\\top) \\\\\n&= L \\, \\mathbb{E}(z \\, z^\\top) \\, L^\\top \\\\\n&= L \\, L^\\top =(\\mathrm{diag}(\\tau)\\,\\Omega_L)\\,(\\mathrm{diag}(\\tau)\\,\\Omega_L)^\\top \\\\\n&= \\mathrm{diag}(\\tau)\\,\\Omega\\,\\mathrm{diag}(\\tau) \\\\\n&= \\Sigma.\n\\end{aligned}\n\\] Where we have used the linearity of expectations (line 2 to 3), the definition of \\(\\Omega = \\Omega_L \\, \\Omega_L^\\top\\), and the fact that \\(\\mathbb{E}(z \\, z^\\top) = I\\) since \\(z \\sim \\mathcal{N}(0, I)\\).\nOmitting the remaining data declarations, which are the same as before with the exception of u, the optimized model is as follows.\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0, upper=pi() / 2&gt;[K] tau_unif;  // prior scale\n  matrix[K, L] gamma;                        // group coeffs\n  real&lt;lower=0&gt; sigma;                       // prediction error scale\n}\ntransformed parameters {\n  vector&lt;lower=0&gt;[K] tau = 2.5 * tan(tau_unif);\n  matrix[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  vector[N] mu;\n  for(n in 1:N) {\n    mu[n] = x[n, ] * beta[, jj[n]];\n  }\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  to_vector(gamma) ~ normal(0, 5);\n  y ~ normal(mu, sigma);\n}\nThis model also reparameterizes the prior scale tau to avoid potential problems with the heavy tails of the Cauchy distribution. The statement tau_unif ~ uniform(0, pi() / 2) can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.\n\n\n\n\n\nStan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).\n\n\nAs a simple example, the following linear regression provides the same setup for estimating the coefficients beta as in our very first example, using y for the N observations and x for the N predictor vectors. The model parameters and model for observations are exactly the same as before.\nTo make predictions, we need to be given the number of predictions, N_new, and their predictor matrix, x_new. The predictions themselves are modeled as a parameter y_new. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector y_new and prediction matrix x_new.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=0&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n\n  int&lt;lower=0&gt; N_new;\n  matrix[N_new, K] x_new;\n}\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n\n  vector[N_new] y_new;                  // predictions\n}\nmodel {\n  y ~ normal(x * beta, sigma);          // observed model\n\n  y_new ~ normal(x_new * beta, sigma);  // prediction model\n}\n\n\n\nWhere possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.\n// ...data as above...\n\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(x * beta, sigma);\n}\ngenerated quantities {\n  vector[N_new] y_new;\n  for (n in 1:N_new) {\n    y_new[n] = normal_rng(x_new[n] * beta, sigma);\n  }\n}\nNow the data are just as before, but the parameter y_new is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.\n\n\nIt is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of \\(2^{31}\\), must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by \\(2^{31} - 1\\) in Stan.\n\n\n\n\n\nMost regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.\nThis section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.\n\n\nThe first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors (Zellner 1962; Greene 2011).\nThe model is easy to write down as a regression, \\[\\begin{align*}\ny_n        &=  x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nwhere \\(x_n\\) is a \\(J\\)-row-vector of predictors (\\(x\\) is an \\((N \\times\nJ)\\) matrix), \\(y_n\\) is a \\(K\\)-vector of observations, \\(\\beta\\) is a \\((K\n\\times J)\\) matrix of regression coefficients (vector \\(\\beta_k\\) holds coefficients for outcome \\(k\\)), and \\(\\Sigma\\) is covariance matrix governing the error. As usual, the intercept can be rolled into \\(x\\) as a column of ones.\nThe basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; J;\n  int&lt;lower=0&gt; N;\n  array[N] vector[J] x;\n  array[N] vector[K] y;\n}\nparameters {\n  matrix[K, J] beta;\n  cov_matrix[K] Sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n  }\n  y ~ multi_normal(mu, Sigma);\n}\nFor efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.\nFollowing the advice in the multivariate hierarchical priors section, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.\n// ...\nparameters {\n  matrix[K, J] beta;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0&gt;[K] L_sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  matrix[K, K] L_Sigma;\n\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n\n  }\n\n  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);\n\n  to_vector(beta) ~ normal(0, 5);\n  L_Omega ~ lkj_corr_cholesky(4);\n  L_sigma ~ cauchy(0, 2.5);\n\n  y ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix beta to a vector.\nIf required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.\n\n\n\nThe multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.\nThe observations \\(y_n\\) are \\(D\\)-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations \\(y_n\\) are based on latent values \\(z_n\\) drawn from a seemingly unrelated regression model (see the previous section), \\[\\begin{align*}\nz_n        &=     x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim  \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nThese are then put through the step function to produce a \\(K\\)-vector \\(z_n\\) of boolean values with elements defined by \\[\ny_{n, k} = \\operatorname{I}\\left(z_{n, k} &gt; 0\\right),\n\\] where \\(\\operatorname{I}()\\) is the indicator function taking the value 1 if its argument is true and 0 otherwise.\nUnlike in the seemingly unrelated regressions case, here the covariance matrix \\(\\Sigma\\) has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see Greene (2011)).\nMultivariate probit regression can be coded in Stan using the trick introduced by Albert and Chib (1993), where the underlying continuous value vectors \\(y_n\\) are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector \\(z\\) in two parts, based on whether the corresponding value of \\(y\\) is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.\nFirst, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in \\(y\\).\nfunctions {\n  int sum2d(array[,] int a) {\n    int s = 0;\n    for (i in 1:size(a)) {\n      s += sum(a[i]);\n    }\n    return s;\n  }\n}\nThe function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.\nThe data declaration block is much like for the seemingly unrelated regressions, but the observations y are now integers constrained to be 0 or 1.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  array[N, D] int&lt;lower=0, upper=1&gt; y;\n  array[N] vector[K] x;\n}\nAfter declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array y into positive and negative components, keeping track of indexes so that z can be easily reassembled in the transformed parameters block.\ntransformed data {\n  int&lt;lower=0&gt; N_pos;\n  array[sum2d(y)] int&lt;lower=1, upper=N&gt; n_pos;\n  array[size(n_pos)] int&lt;lower=1, upper=D&gt; d_pos;\n  int&lt;lower=0&gt; N_neg;\n  array[(N * D) - size(n_pos)] int&lt;lower=1, upper=N&gt; n_neg;\n  array[size(n_neg)] int&lt;lower=1, upper=D&gt; d_neg;\n\n  N_pos = size(n_pos);\n  N_neg = size(n_neg);\n  {\n    int i;\n    int j;\n    i = 1;\n    j = 1;\n    for (n in 1:N) {\n      for (d in 1:D) {\n        if (y[n, d] == 1) {\n          n_pos[i] = n;\n          d_pos[i] = d;\n          i += 1;\n        } else {\n          n_neg[j] = n;\n          d_neg[j] = d;\n          j += 1;\n        }\n      }\n    }\n  }\n}\nThe variables N_pos and N_neg are set to the number of true (1) and number of false (0) observations in y. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.\nThe parameters are declared as follows.\nparameters {\n  matrix[D, K] beta;\n  cholesky_factor_corr[D] L_Omega;\n  vector&lt;lower=0&gt;[N_pos] z_pos;\n  vector&lt;upper=0&gt;[N_neg] z_neg;\n}\nThese include the regression coefficients beta and the Cholesky factor of the correlation matrix, L_Omega. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).\nThe critical part of the parameter declaration is that the latent real value \\(z\\) is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct \\(z\\).\ntransformed parameters {\n  array[N] vector[D] z;\n  for (n in 1:N_pos) {\n    z[n_pos[n], d_pos[n]] = z_pos[n];\n  }\n  for (n in 1:N_neg) {\n    z[n_neg[n], d_neg[n]] = z_neg[n];\n  }\n}\nAt this point, the model is simple, pretty much recreating the seemingly unrelated regression.\nmodel {\n  L_Omega ~ lkj_corr_cholesky(4);\n  to_vector(beta) ~ normal(0, 5);\n  {\n    array[N] vector[D] beta_x;\n    for (n in 1:N) {\n      beta_x[n] = beta * x[n];\n    }\n    z ~ multi_normal_cholesky(beta_x, L_Omega);\n  }\n}\nThis simple form of model is made possible by the Albert and Chib-style constraints on z.\nFinally, the correlation matrix itself can be put back together in the generated quantities block if desired.\ngenerated quantities {\n  corr_matrix[D] Omega;\n  Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\nThe same could be done for the seemingly unrelated regressions in the previous section.\n\n\n\n\nThe main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.\n\n\nConsider predicting unobserved outcomes using linear regression. Given predictors \\(x_1, \\dotsc, x_N\\) and observed outcomes \\(y_1, \\dotsc, y_N\\), and assuming a standard linear regression with intercept \\(\\alpha\\), slope \\(\\beta\\), and error scale \\(\\sigma\\), along with improper uniform priors, the posterior over the parameters given \\(x\\) and \\(y\\) is \\[\np\\left(\\alpha, \\beta, \\sigma \\mid x, y \\right)\n\\propto\n\\prod_{n=1}^N\n  \\textsf{normal}\\left(y_n \\mid \\alpha + \\beta x_n, \\sigma\\right).\n\\]\nFor this model, the posterior predictive inference for a new outcome \\(\\tilde{y}_m\\) given a predictor \\(\\tilde{x}_m\\), conditioned on the observed data \\(x\\) and \\(y\\), is \\[\np\\left(\\tilde{y}_n \\mid \\tilde{x}_n, x, y\\right)\n= \\int_{(\\alpha,\\beta,\\sigma)}\n  \\textsf{normal}\\left(\\tilde{y}_n \\mid \\alpha + \\beta \\tilde{x}_n, \\sigma\\right)\n  \\times\n  p\\left(\\alpha, \\beta, \\sigma \\mid x, y\\right)\n  \\,\\textrm{d}(\\alpha,\\beta,\\sigma).\n\\]\nTo code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde;\n  for (n in 1:N_tilde) {\n    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);\n  }\n}\nGiven observed predictors \\(x\\) and outcomes \\(y\\), y_tilde will be drawn according to \\(p\\left(\\tilde{y} \\mid \\tilde{x}, y, x\\right)\\). This means that, for example, the posterior mean for y_tilde is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).\n\n\n\nA good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors \\(x\\) are used in place of new predictors \\(\\tilde{x}\\) for unobserved outcomes. If the model fits the data well, the predictions for \\(\\tilde{y}\\) based on \\(x\\) should match the observed data \\(y\\).\nTo code posterior predictive checks in Stan requires only a slight modification of the prediction code to use \\(x\\) and \\(N\\) in place of \\(\\tilde{x}\\) and \\(\\tilde{N}\\),\ngenerated quantities {\n  vector[N] y_tilde;\n  for (n in 1:N) {\n    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);\n  }\n}\nGelman et al. (2013) recommend choosing several posterior draws \\(\\tilde{y}^{(1)}, \\dotsc, \\tilde{y}^{(M)}\\) and plotting each of them alongside the data \\(y\\) that was actually observed. If the model fits well, the simulated \\(\\tilde{y}\\) will look like the actual data \\(y\\).",
+    "text": "As an example of a system of ODEs, consider a harmonic oscillator. In a harmonic oscillator a particle disturbed from equilibrium is pulled back towards its equilibrium position by a force proportional to its displacement from equilibrium. The system here additionally has a friction force proportional to particle speed which points in the opposite direction of the particle velocity. The system state will be a pair \\(y = (y_1, y_2)\\) representing position and speed. The change in the system with respect to time is given by the following differential equations.1\n\\[\\begin{align*}\n&\\frac{d}{dt} y_1 = y_2  \\\\\n&\\frac{d}{dt} y_2 = -y_1 - \\theta y_2\n\\end{align*}\\]\nThe state equations implicitly defines the state at future times as a function of an initial state and the system parameters.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#linear-regression",
-    "href": "stan-users-guide/regression.html#linear-regression",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#coding-the-ode-system-function",
+    "href": "stan-users-guide/odes.html#coding-the-ode-system-function",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "The simplest linear regression model is the following, with a single predictor and a slope and intercept coefficient, and normally distributed noise. This model can be written using standard regression notation as \\[\ny_n = \\alpha + \\beta x_n + \\epsilon_n\n\\quad\\text{where}\\quad\n\\epsilon_n \\sim \\operatorname{normal}(0,\\sigma).\n\\]\nThis is equivalent to the following sampling involving the residual, \\[\ny_n - (\\alpha + \\beta X_n) \\sim \\operatorname{normal}(0,\\sigma),\n\\] and reducing still further, to \\[\ny_n \\sim \\operatorname{normal}(\\alpha + \\beta X_n, \\, \\sigma).\n\\]\nThis latter form of the model is coded in Stan as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\nThere are N observations and for each observation, \\(n \\in N\\), we have predictor x[n] and outcome y[n]. The intercept and slope parameters are alpha and beta. The model assumes a normally distributed noise term with scale sigma. This model has improper priors for the two regression coefficients.\n\n\nThe distribution statement in the previous model is vectorized, with\ny ~ normal(alpha + beta * x, sigma);\nproviding the same model as the unvectorized version,\nfor (n in 1:N) {\n  y[n] ~ normal(alpha + beta * x[n], sigma);\n}\nIn addition to being more concise, the vectorized form is much faster.1\nIn general, Stan allows the arguments to distributions such as normal to be vectors. If any of the other arguments are vectors or arrays, they have to be the same size. If any of the other arguments is a scalar, it is reused for each vector entry.\nThe other reason this works is that Stan’s arithmetic operators are overloaded to perform matrix arithmetic on matrices. In this case, because x is of type vector and beta of type real, the expression beta * x is of type vector. Because Stan supports vectorization, a regression model with more than one predictor can be written directly using matrix notation.\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] beta;       // coefficients for predictors\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(x * beta + alpha, sigma);  // data model\n}\nThe constraint lower=0 in the declaration of sigma constrains the value to be greater than or equal to 0. With no prior in the model block, the effect is an improper prior on non-negative real numbers. Although a more informative prior may be added, improper priors are acceptable as long as they lead to proper posteriors.\nIn the model above, x is an \\(N \\times K\\) matrix of predictors and beta a \\(K\\)-vector of coefficients, so x * beta is an \\(N\\)-vector of predictions, one for each of the \\(N\\) data items. These predictions line up with the outcomes in the \\(N\\)-vector y, so the entire model may be written using matrix arithmetic as shown. It would be possible to include a column of ones in the data matrix x to remove the alpha parameter.\nThe distribution statement in the model above is just a more efficient, vector-based approach to coding the model with a loop, as in the following statistically equivalent model.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta, sigma);\n  }\n}\nWith Stan’s matrix indexing scheme, x[n] picks out row n of the matrix x; because beta is a column vector, the product x[n] * beta is a scalar of type real.\n\n\nIn the model formulation\ny ~ normal(x * beta, sigma);\nthere is no longer an intercept coefficient alpha. Instead, we have assumed that the first column of the input matrix x is a column of 1 values. This way, beta[1] plays the role of the intercept. If the intercept gets a different prior than the slope terms, then it would be clearer to break it out. It is also slightly more efficient in its explicit form with the intercept variable singled out because there’s one fewer multiplications; it should not make that much of a difference to speed, though, so the choice should be based on clarity.",
+    "text": "The first step in coding an ODE system in Stan is defining the ODE system function. The system functions require a specific signature so that the solvers know how to use them properly.\nThe first argument to the system function is time, passed as a real; the second argument to the system function is the system state, passed as a vector, and the return value from the system function are the current time derivatives of the state defined as a vector. Additional arguments can be included in the system function to pass other information into the solve (these will be passed through the function that starts the ODE integration). These argument can be parameters (in this case, the friction coefficient), data, or any quantities that are needed to define the differential equation.\nThe simple harmonic oscillator can be coded using the following function in Stan (see the user-defined functions chapter for more information on coding user-defined functions).\nvector sho(real t,        // time\n           vector y,      // state\n           real theta) {  // friction parameter\n  vector[2] dydt;\n  dydt[1] = y[2];\n  dydt[2] = -y[1] - theta * y[2];\n  return dydt;\n}\nThe function takes in a time t (a real), the system state y (a vector), and the parameter theta (a real). The function returns a vector of time derivatives of the system state at time t, state y, and parameter theta. The simple harmonic oscillator coded here does not have time-sensitive equations; that is, t does not show up in the definition of dydt, however it is still required.\n\n\nThe types in the ODE system function are strict. The first argument is the time passed as a real, the second argument is the state passed as a vector, and the return type is a vector. A model that does not have this signature will fail to compile. The third argument onwards can be any type, granted all the argument types match the types of the respective arguments in the solver call.\nAll of these are possible ODE signatures:\nvector myode1(real t, vector y, real a0);\nvector myode2(real t, vector y, array[] int a0, vector a1);\nvector myode3(real t, vector y, matrix a0, array[] real a1, row_vector a2);\nbut these are not allowed:\nvector myode1(real t, array[] real y, real a0);\n// Second argument is not a vector\narray[] real myode2(real t, vector y, real a0);\n// Return type is not a vector\nvector myode3(vector y, real a0);\n// First argument is not a real and second is not a vector",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#QR-reparameterization.section",
-    "href": "stan-users-guide/regression.html#QR-reparameterization.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#measurement-error-models",
+    "href": "stan-users-guide/odes.html#measurement-error-models",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "In the previous example, the linear predictor can be written as \\(\\eta\n= x \\beta\\), where \\(\\eta\\) is a \\(N\\)-vector of predictions, \\(x\\) is a \\(N\n\\times K\\) matrix, and \\(\\beta\\) is a \\(K\\)-vector of coefficients. Presuming \\(N \\geq K\\), we can exploit the fact that any design matrix \\(x\\) can be decomposed using the thin QR decomposition into an orthogonal matrix \\(Q\\) and an upper-triangular matrix \\(R\\), i.e. \\(x = Q\nR\\).\nThe functions qr_thin_Q and qr_thin_R implement the thin QR decomposition, which is to be preferred to the fat QR decomposition that would be obtained by using qr_Q and qr_R, as the latter would more easily run out of memory (see the Stan Functions Reference for more information on the qr_thin_Q and qr_thin_R functions). In practice, it is best to write \\(x = Q^\\ast\nR^\\ast\\) where \\(Q^\\ast = Q * \\sqrt{n - 1}\\) and \\(R^\\ast =\n\\frac{1}{\\sqrt{n - 1}} R\\). Thus, we can equivalently write \\(\\eta = x\n\\beta = Q R \\beta = Q^\\ast R^\\ast \\beta\\). If we let \\(\\theta = R^\\ast\n\\beta\\), then we have \\(\\eta = Q^\\ast \\theta\\) and \\(\\beta = R^{\\ast^{-1}}\n\\theta\\). In that case, the previous Stan program becomes\ndata {\n  int&lt;lower=0&gt; N;   // number of data items\n  int&lt;lower=0&gt; K;   // number of predictors\n  matrix[N, K] x;   // predictor matrix\n  vector[N] y;      // outcome vector\n}\ntransformed data {\n  matrix[N, K] Q_ast;\n  matrix[K, K] R_ast;\n  matrix[K, K] R_ast_inverse;\n  // thin and scale the QR decomposition\n  Q_ast = qr_thin_Q(x) * sqrt(N - 1);\n  R_ast = qr_thin_R(x) / sqrt(N - 1);\n  R_ast_inverse = inverse(R_ast);\n}\nparameters {\n  real alpha;           // intercept\n  vector[K] theta;      // coefficients on Q_ast\n  real&lt;lower=0&gt; sigma;  // error scale\n}\nmodel {\n  y ~ normal(Q_ast * theta + alpha, sigma);  // data model\n}\ngenerated quantities {\n  vector[K] beta;\n  beta = R_ast_inverse * theta; // coefficients on x\n}\nSince this Stan program generates equivalent predictions for \\(y\\) and the same posterior distribution for \\(\\alpha\\), \\(\\beta\\), and \\(\\sigma\\) as the previous Stan program, many wonder why the version with this QR reparameterization performs so much better in practice, often both in terms of wall time and in terms of effective sample size. The reasoning is threefold:\n\nThe columns of \\(Q^\\ast\\) are orthogonal whereas the columns of \\(x\\) generally are not. Thus, it is easier for a Markov Chain to move around in \\(\\theta\\)-space than in \\(\\beta\\)-space.\nThe columns of \\(Q^\\ast\\) have the same scale whereas the columns of \\(x\\) generally do not. Thus, a Hamiltonian Monte Carlo algorithm can move around the parameter space with a smaller number of larger steps\nSince the covariance matrix for the columns of \\(Q^\\ast\\) is an identity matrix, \\(\\theta\\) typically has a reasonable scale if the units of \\(y\\) are also reasonable. This also helps HMC move efficiently without compromising numerical accuracy.\n\nConsequently, this QR reparameterization is recommended for linear and generalized linear models in Stan whenever \\(K &gt; 1\\) and you do not have an informative prior on the location of \\(\\beta\\). It can also be worthwhile to subtract the mean from each column of \\(x\\) before obtaining the QR decomposition, which does not affect the posterior distribution of \\(\\theta\\) or \\(\\beta\\) but does affect \\(\\alpha\\) and allows you to interpret \\(\\alpha\\) as the expectation of \\(y\\) in a linear model.",
+    "text": "Noisy observations of the ODE state can be used to estimate the parameters and/or the initial state of the system.\n\n\nAs an example, suppose the simple harmonic oscillator has a parameter value of \\(\\theta = 0.15\\) and an initial state \\(y(t = 0, \\theta = 0.15) = (1, 0)\\). Assume the system is measured at 10 time points, \\(t = 1, 2, \\cdots, 10\\), where each measurement of \\(y(t, \\theta)\\) has independent \\(\\textsf{normal}(0, 0.1)\\) error in both dimensions (\\(y_1(t, \\theta)\\) and \\(y_2(t, \\theta)\\)).\nThe following model can be used to generate data like this:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  real t0;\n  array[T] real ts;\n  real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThe system parameters theta and initial state y0 are read in as data along with the initial time t0 and observation times ts. The ODE is solved for the specified times, and then random measurement errors are added to produce simulated observations y_sim. Because the system is not stiff, the ode_rk45 solver is used.\nThis program illustrates the way in which the ODE solver is called in a Stan program,\narray[T] vector[2] y_sim = ode_rk45(sho, y0, t0, ts, theta);\nthis returns the solution of the ODE initial value problem defined by system function sho, initial state y0, initial time t0, and parameter theta at the times ts. The call explicitly specifies the non-stiff RK45 solver.\nThe parameter theta is passed unmodified to the ODE system function. If there were additional arguments that must be passed, they could be appended to the end of the ode call here. For instance, if the system function took two parameters, \\(\\theta\\) and \\(\\beta\\), the system function definition would look like:\nvector sho(real t, vector y, real theta, real beta) { ... }\nand the appropriate ODE solver call would be:\node_rk45(sho, y0, t0, ts, theta, beta);\nAny number of additional arguments can be added. They can be any Stan type (as long as the types match between the ODE system function and the solver call).\nBecause all none of the input arguments are a function of parameters, the ODE solver is called in the generated quantities block. The random measurement noise is added to each of the T outputs with normal_rng.\n\n\n\nTypical realization of harmonic oscillator trajectory.\n\n\n\n\n\nThese ten noisy observations of the state can be used to estimate the friction parameter, \\(\\theta\\), the initial conditions, \\(y(t_0, \\theta)\\), and the scale of the noise in the problem. The full Stan model is:\nfunctions {\n  vector sho(real t,\n             vector y,\n             real theta) {\n    vector[2] dydt;\n    dydt[1] = y[2];\n    dydt[2] = -y[1] - theta * y[2];\n    return dydt;\n  }\n}\ndata {\n  int&lt;lower=1&gt; T;\n  array[T] vector[2] y;\n  real t0;\n  array[T] real ts;\n}\nparameters {\n  vector[2] y0;\n  vector&lt;lower=0&gt;[2] sigma;\n  real theta;\n}\nmodel {\n  array[T] vector[2] mu = ode_rk45(sho, y0, t0, ts, theta);\n  sigma ~ normal(0, 2.5);\n  theta ~ std_normal();\n  y0 ~ std_normal();\n  for (t in 1:T) {\n    y[t] ~ normal(mu[t], sigma);\n  }\n}\nBecause the solves are now a function of model parameters, the ode_rk45 call is now made in the model block. There are half-normal priors on the measurement error scales sigma, and standard normal priors on theta and the initial state vector y0. The solutions to the ODE are assigned to mu, which is used as the location for the normal observation model.\nAs with other regression models, it’s easy to change the noise model to something with heavier tails (e.g., Student-t distributed), correlation in the state variables (e.g., with a multivariate normal distribution), or both heavy tails and correlation in the state variables (e.g., with a multivariate Student-t distribution).",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#regression-priors.section",
-    "href": "stan-users-guide/regression.html#regression-priors.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#stiff-ode.section",
+    "href": "stan-users-guide/odes.html#stiff-ode.section",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "See our general discussion of priors for tips on priors for parameters in regression models.\nLater sections discuss univariate hierarchical priors and multivariate hierarchical priors, as well as priors used to identify models.\nHowever, as described in QR-reparameterization section, if you do not have an informative prior on the location of the regression coefficients, then you are better off reparameterizing your model so that the regression coefficients are a generated quantity. In that case, it usually does not matter much what prior is used on on the reparameterized regression coefficients and almost any weakly informative prior that scales with the outcome will do.",
+    "text": "Stiffness is a numerical phenomena that causes some differential equation solvers difficulty, notably the Runge-Kutta RK45 solver used in the examples earlier. The phenomena is common in chemical reaction systems, which are often characterized by having multiple vastly different time-scales. The stiffness of a system can also vary between different parts of parameter space, and so a typically non-stiff system may exhibit stiffness occasionally. These sorts of difficulties can occur more frequently with loose priors or during warmup.\nStan provides a specialized solver for stiff ODEs (Cohen and Hindmarsh 1996; Serban and Hindmarsh 2005). An ODE system is specified exactly the same way with a function of exactly the same signature. The only difference is in the call to the solver the rk45 suffix is replaced with bdf, as in\node_bdf(sho, y0, t0, ts, theta);\nUsing the stiff (bdf) solver on a system that is not stiff may be much slower than using the non-stiff (rk45) solver because each step of the stiff solver takes more time to compute. On the other hand, attempting to use the non-stiff solver for a stiff system will cause the timestep to become very small, leading the non-stiff solver taking more time overall even if each step is easier to compute than for the stiff solver.\nIf it is not known for sure that an ODE system is stiff, run the model with both the rk45 and bdf solvers and see which is faster. If the rk45 solver is faster, then the problem is probably non-stiff, and then it makes sense to try the adams solver as well. The adams solver uses higher order methods which can take larger timesteps than the rk45 solver, though similar to the bdf solver each of these steps is more expensive to compute.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#robust-noise-models",
-    "href": "stan-users-guide/regression.html#robust-noise-models",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#control-ode.section",
+    "href": "stan-users-guide/odes.html#control-ode.section",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "The standard approach to linear regression is to model the noise term \\(\\epsilon\\) as having a normal distribution. From Stan’s perspective, there is nothing special about normally distributed noise. For instance, robust regression can be accommodated by giving the noise term a Student-\\(t\\) distribution. To code this in Stan, the distribution distribution is changed to the following.\ndata {\n  // ...\n  real&lt;lower=0&gt; nu;\n}\n// ...\nmodel {\n  y ~ student_t(nu, alpha + beta * x, sigma);\n}\nThe degrees of freedom constant nu is specified as data.",
+    "text": "For additional control of the solves, both the stiff and non-stiff forward ODE solvers have function signatures that makes it possible to specify the relative_tolerance, absolute_tolerance, and max_num_steps parameters. These are the same as the regular function names but with _tol appended to the end. All three control arguments must be supplied with this signature (there are no defaults).\narray[T] vector[2] y_sim = ode_bdf_tol(sho, y0, t0, ts,\n                                 relative_tolerance,\n                                 absolute_tolerance,\n                                 max_num_steps,\n                                 theta);\nrelative_tolerance and absolute_tolerance control accuracy the solver tries to achieve, and max_num_steps specifies the maximum number of steps the solver will take between output time points before throwing an error.\nThe control parameters must be data variables – they cannot be parameters or expressions that depend on parameters, including local variables in any block other than transformed data and generated quantities. User-defined function arguments may be qualified as only allowing data arguments using the data qualifier.\nFor the RK45 and Cash-Karp solvers, the default values for relative and absolute tolerance are both \\(10^{-6}\\) and the maximum number of steps between outputs is one million. For the BDF and Adams solvers, the relative and absolute tolerances are \\(10^{-10}\\) and the maximum number of steps between outputs is one hundred million.\n\n\nIf there are discontinuities in the ODE system function, it is best to integrate the ODE between the discontinuities, stopping the solver at each one, and restarting it on the other side.\nNonetheless, the ODE solvers will attempt to integrate over discontinuities they encounters in the state function. The accuracy of the solution near the discontinuity may be problematic (requiring many small steps). An example of such a discontinuity is a lag in a pharmacokinetic model, where a concentration is zero for times \\(0 &lt; t &lt; t'\\) and then positive for \\(t \\geq t'\\). In this example example, we would use code in the system such as\nif (t &lt; t_lag) {\n  return [0, 0]';\n} else {\n  // ... return non-zero vector...\n}\nIn general it is better to integrate up to t_lag in one solve and then integrate from t_lag onwards in another. Mathematically, the discontinuity can make the problem ill-defined and the numerical integrator may behave erratically around it.\nIf the location of the discontinuity cannot be controlled precisely, or there is some other rapidly change in ODE behavior, it can be useful to tell the ODE solver to produce output in the neighborhood. This can help the ODE solver avoid indiscriminately stepping over an important feature of the solution.\n\n\n\nThe relative tolerance RTOL and absolute tolerance ATOL control the accuracy of the numerical solution. Specifically, when solving an ODE with unknowns \\(y=(y_1,\\dots,y_n)^T\\), at every step the solver controls estimated local error \\(e=(e_1,\\dots,e_n)^T\\) through its weighted root-mean-square norm (Serban and Hindmarsh (2005), Hairer, Nørsett, and Wanner (1993))\n\\[\\begin{equation*}\n\\sqrt{\\sum_{i=1}^n{\\frac{1}{n}\\frac{e_i^2}{(\\text{RTOL}\\times y_i + \\text{ATOL})^2}}} &lt; 1\n\\end{equation*}\\] by reducing the stepsize when the inequality is not satisfied.\nTo understand the roles of the two tolerances it helps to assume \\(y\\) at opposite scales in the above expression: on one hand the absolute tolerance has little effect when \\(y_i \\gg 1\\), on the other the relative tolerance can not affect the norm when \\(y_i = 0\\). Users are strongly encouraged to carefully choose tolerance values according to the ODE and its application. One can follow Brenan, Campbell, and Petzold (1995) for a rule of thumb: let \\(m\\) be the number of significant digits required for \\(y\\), set \\(\\text{RTOL}=10^{-(m+1)}\\), and set ATOL at which \\(y\\) becomes insignificant. Note that the same weighted root-mean-square norm is used to control nonlinear solver convergence in bdf and adams solvers, and the same tolerances are used to control forward sensitivity calculation. See Serban and Hindmarsh (2005) for details.\n\n\n\nThe maximum number of steps can be used to stop a runaway simulation. This can arise in when MCMC moves to a part of parameter space very far from where a differential equation would typically be solved. In particular this can happen during warmup. With the non-stiff solver, this may happen when the sampler moves to stiff regions of parameter space, which will requires small step sizes.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#logistic-probit-regression.section",
-    "href": "stan-users-guide/regression.html#logistic-probit-regression.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#adjoint-ode.section",
+    "href": "stan-users-guide/odes.html#adjoint-ode.section",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "For binary outcomes, either of the closely related logistic or probit regression models may be used. These generalized linear models vary only in the link function they use to map linear predictions in \\((-\\infty,\\infty)\\) to probability values in \\((0,1)\\). Their respective link functions, the logistic function and the standard normal cumulative distribution function, are both sigmoid functions (i.e., they are both S-shaped).\nA logistic regression model with one predictor and an intercept is coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real alpha;\n  real beta;\n}\nmodel {\n  y ~ bernoulli_logit(alpha + beta * x);\n}\nThe noise parameter is built into the Bernoulli formulation here rather than specified directly.\nLogistic regression is a kind of generalized linear model with binary outcomes and the log odds (logit) link function, defined by \\[\n\\operatorname{logit}(v) = \\log \\left( \\frac{v}{1-v} \\right).\n\\]\nThe inverse of the link function appears in the model: \\[\n\\operatorname{logit}^{-1}(u) = \\texttt{inv}\\mathtt{\\_}\\texttt{logit}(u) = \\frac{1}{1 + \\exp(-u)}.\n\\]\nThe model formulation above uses the logit-parameterized version of the Bernoulli distribution, which is defined by \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha \\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe formulation is also vectorized in the sense that alpha and beta are scalars and x is a vector, so that alpha   + beta * x is a vector. The vectorized formulation is equivalent to the less efficient version\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(alpha + beta * x[n]);\n}\nExpanding out the Bernoulli logit, the model is equivalent to the more explicit, but less efficient and less arithmetically stable\nfor (n in 1:N) {\n  y[n] ~ bernoulli(inv_logit(alpha + beta * x[n]));\n}\nOther link functions may be used in the same way. For example, probit regression uses the cumulative normal distribution function, which is typically written as\n\\[\n\\Phi(x) = \\int_{-\\infty}^x \\textsf{normal}\\left(y \\mid 0,1 \\right) \\,\\textrm{d}y.\n\\]\nThe cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi. The probit regression model may be coded in Stan by replacing the logistic model’s distribution statement with the following.\ny[n] ~ bernoulli(Phi(alpha + beta * x[n]));\nA fast approximation to the cumulative standard normal distribution function \\(\\Phi\\) is implemented in Stan as the function Phi_approx.2 The approximate probit regression model may be coded with the following.\ny[n] ~ bernoulli(Phi_approx(alpha + beta * x[n]));",
+    "text": "The adjoint ODE solver method differs mathematically from the forward ODE solvers in the way gradients of the ODE solution are obtained. The forward ODE approach augments the original ODE system with \\(N\\) additional states for each parameter for which gradients are needed. If there are \\(M\\) parameters for which sensitivities are required, then the augmented ODE system has a total of \\(N \\cdot (M +\n1)\\) states. This can result in very large ODE systems through the multiplicative scaling of the computational effort needed.\nIn contrast, the adjoint ODE solver integrates forward in time a system of \\(N\\) equations to compute the ODE solution and then integrates backwards in time another system of \\(N\\) equations to get the sensitivities. Additionally, for \\(M\\) parameters there are \\(M\\) additional equations to integrate during the backwards solve. Because of this the adjoint sensitivity problem scales better in parameters than the forward sensitivity problem. The adjoint solver in Stan uses CVODES (the same as the bdf and adams forward sensitivity interfaces).\nThe solution computed in the forward integration is required during the backward integration. CVODES uses a checkpointing scheme that saves the forward solver state regularly. The number of steps between saving checkpoints is configurable in the interface. These checkpoints are then interpolated during the backward solve using one of two interpolation schemes.\nThe solver type (either bdf or adams) can be individually set for both the forward and backward solves.\nThe tolerances for each phase of the solve must be specified in the interface. Note that the absolute tolerance for the forward and backward ODE integration phase need to be set for each ODE state separately. The harmonic oscillator example call from above becomes:\narray[T] vector[2] y_sim\n    = ode_adjoint_tol_ctl(sho, y0, t0, ts,\n                          relative_tolerance/9.0,                // forward tolerance\n                          rep_vector(absolute_tolerance/9.0, 2), // forward tolerance\n                          relative_tolerance/3.0,                // backward tolerance\n                          rep_vector(absolute_tolerance/3.0, 2), // backward tolerance\n                          relative_tolerance,                    // quadrature tolerance\n                          absolute_tolerance,                    // quadrature tolerance\n                          max_num_steps,\n                          150,                                   // number of steps between checkpoints\n                          1,                                     // interpolation polynomial: 1=Hermite, 2=polynomial\n                          2,                                     // solver for forward phase: 1=Adams, 2=BDF\n                          2,                                     // solver for backward phase: 1=Adams, 2=BDF\n                          theta);\nFor a detailed information on each argument please see the Stan function reference manual.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#multi-logit.section",
-    "href": "stan-users-guide/regression.html#multi-logit.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/odes.html#solving-a-system-of-linear-odes-using-a-matrix-exponential",
+    "href": "stan-users-guide/odes.html#solving-a-system-of-linear-odes-using-a-matrix-exponential",
+    "title": "Ordinary Differential Equations",
     "section": "",
-    "text": "Multiple outcome forms of logistic regression can be coded directly in Stan. For instance, suppose there are \\(K\\) possible outcomes for each output variable \\(y_n\\). Also suppose that there is a \\(D\\)-dimensional vector \\(x_n\\) of predictors for \\(y_n\\). The multi-logit model with \\(\\textsf{normal}(0,5)\\) priors on the coefficients is coded as follows.\ndata {\n  int K;\n  int N;\n  int D;\n  array[N] int y;\n  matrix[N, D] x;\n}\nparameters {\n  matrix[D, K] beta;\n}\nmodel {\n  matrix[N, K] x_beta = x * beta;\n\n  to_vector(beta) ~ normal(0, 5);\n\n  for (n in 1:N) {\n    y[n] ~ categorical_logit(x_beta[n]');\n\n  }\n}\nwhere x_beta[n]' is the transpose of x_beta[n]. The prior on beta is coded in vectorized form. As of Stan 2.18, the categorical-logit distribution is not vectorized for parameter arguments, so the loop is required. The matrix multiplication is pulled out to define a local variable for all of the predictors for efficiency. Like the Bernoulli-logit, the categorical-logit distribution applies softmax internally to convert an arbitrary vector to a simplex, \\[\n\\texttt{categorical}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{categorical}\\left(y \\mid \\texttt{softmax}(\\alpha)\\right),\n\\] where \\[\n\\texttt{softmax}(u) = \\exp(u) / \\operatorname{sum}\\left(\\exp(u)\\right).\n\\]\nThe categorical distribution with log-odds (logit) scaled parameters used above is equivalent to writing\ny[n] ~ categorical(softmax(x[n] * beta));\n\n\nThe data block in the above model is defined without constraints on sizes K, N, and D or on the outcome array y. Constraints on data declarations provide error checking at the point data are read (or transformed data are defined), which is before sampling begins. Constraints on data declarations also make the model author’s intentions more explicit, which can help with readability. The above model’s declarations could be tightened to\nint&lt;lower=2&gt; K;\nint&lt;lower=0&gt; N;\nint&lt;lower=1&gt; D;\narray[N] int&lt;lower=1, upper=K&gt; y;\nThese constraints arise because the number of categories, K, must be at least two in order for a categorical model to be useful. The number of data items, N, can be zero, but not negative; unlike R, Stan’s for-loops always move forward, so that a loop extent of 1:N when N is equal to zero ensures the loop’s body will not be executed. The number of predictors, D, must be at least one in order for beta * x[n] to produce an appropriate argument for softmax(). The categorical outcomes y[n] must be between 1 and K in order for the discrete sampling to be well defined.\nConstraints on data declarations are optional. Constraints on parameters declared in the parameters block, on the other hand, are not optional—they are required to ensure support for all parameter values satisfying their constraints. Constraints on transformed data, transformed parameters, and generated quantities are also optional.\n\n\n\nBecause softmax is invariant under adding a constant to each component of its input, the model is typically only identified if there is a suitable prior on the coefficients.\nAn alternative is to use \\((K-1)\\)-vectors by fixing one of them to be zero. The partially known parameters section discusses how to mix constants and parameters in a vector. In the multi-logit case, the parameter block would be redefined to use \\((K - 1)\\)-vectors\nparameters {\n  matrix[D, K - 1] beta_raw;\n}\nand then these are transformed to parameters to use in the model. First, a transformed data block is added before the parameters block to define a vector of zero values,\ntransformed data {\n  vector[D] zeros = rep_vector(0, D);\n}\nwhich can then be appended to beta_raw to produce the coefficient matrix beta,\ntransformed parameters {\n  matrix[D, K] beta = append_col(beta_raw, zeros);\n}\nThe rep_vector(0, D) call creates a column vector of size D with all entries set to zero. The derived matrix beta is then defined to be the result of appending the vector zeros as a new column at the end of beta_raw; the vector zeros is defined as transformed data so that it doesn’t need to be constructed from scratch each time it is used.\nThis is not the same model as using \\(K\\)-vectors as parameters, because now the prior only applies to \\((K-1)\\)-vectors. In practice, this will cause the maximum likelihood solutions to be different and also the posteriors to be slightly different when taking priors centered around zero, as is typical for regression coefficients.",
+    "text": "Linear systems of ODEs can be solved using a matrix exponential. This can be considerably faster than using one of the ODE solvers.\nThe solution to \\(\\frac{d}{dt} y = ay\\) is \\(y = y_0e^{at}\\), where the constant \\(y_0\\) is determined by boundary conditions. We can extend this solution to the vector case: \\[\n\\frac{d}{dt}y = A \\, y\n\\]  where \\(y\\) is now a vector of length \\(n\\) and \\(A\\) is an \\(n\\) by \\(n\\) matrix. The solution is then given by: \\[\ny = e^{tA} \\, y_0\n\\]  where the matrix exponential is formally defined by the convergent power series: \\[\ne^{tA} = \\sum_{n=0}^{\\infty} \\dfrac{tA^n}{n!} = I + tA + \\frac{t^2A^2}{2!} + \\dotsb\n\\] \nWe can apply this technique to the simple harmonic oscillator example, by setting \\[\ny = \\begin{bmatrix} y_1 \\\\ y_2 \\end{bmatrix} \\qquad\nA = \\begin{bmatrix} 0 & 1 \\\\ -1 & -\\theta \\end{bmatrix}\n\\] \nThe Stan model to simulate noisy observations using a matrix exponential function is given below.\nIn general, computing a matrix exponential will be more efficient than using a numerical solver. We can however only apply this technique to systems of linear ODEs.\ndata {\n  int&lt;lower=1&gt; T;\n  vector[2] y0;\n  array[T] real ts;\n  array[1] real theta;\n}\nmodel {\n}\ngenerated quantities {\n  array[T] vector[2] y_sim;\n  matrix[2, 2] A = [[ 0,  1],\n                    [-1, -theta[1]]]\n  for (t in 1:T) {\n    y_sim[t] = matrix_exp((t - 1) * A) * y0;\n  }\n  // add measurement error\n  for (t in 1:T) {\n    y_sim[t, 1] += normal_rng(0, 0.1);\n    y_sim[t, 2] += normal_rng(0, 0.1);\n  }\n}\nThis Stan program simulates noisy measurements from a simple harmonic oscillator. The system of linear differential equations is coded as a matrix. The system parameters theta and initial state y0 are read in as data along observation times ts. The generated quantities block is used to solve the ODE for the specified times and then add random measurement error, producing observations y_sim. Because the ODEs are linear, we can use the matrix_exp function to solve the system.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#parameterizing-centered-vectors",
-    "href": "stan-users-guide/regression.html#parameterizing-centered-vectors",
-    "title": "Regression Models",
-    "section": "",
-    "text": "When there are varying effects in a regression, the resulting likelihood is not identified unless further steps are taken. For example, we might have a global intercept \\(\\alpha\\) and then a varying effect \\(\\beta_k\\) for age group \\(k\\) to make a linear predictor \\(\\alpha +\n\\beta_k\\). With this predictor, we can add a constant to \\(\\alpha\\) and subtract from each \\(\\beta_k\\) and get exactly the same likelihood.\nThe traditional approach to identifying such a model is to pin the first varying effect to zero, i.e., \\(\\beta_1 = 0\\). With one of the varying effects fixed, you can no longer add a constant to all of them and the model’s likelihood is identified. In addition to the difficulty in specifying such a model in Stan, it is awkward to formulate priors because the other coefficients are all interpreted relative to \\(\\beta_1\\).\nIn a Bayesian setting, a proper prior on each of the \\(\\beta\\) is enough to identify the model. Unfortunately, this can lead to inefficiency during sampling as the model is still only weakly identified through the prior—there is a very simple example of the difference in the discussion of collinearity in the collinearity section.\nAn alternative identification strategy that allows a symmetric prior is to enforce a sum-to-zero constraint on the varying effects, i.e., \\(\\sum_{k=1}^K \\beta_k = 0.\\)\nA parameter vector constrained to sum to zero may also be used to identify a multi-logit regression parameter vector (see the multi-logit section for details), or may be used for ability or difficulty parameters (but not both) in an IRT model (see the item-response model section for details).\n\n\nAs of Stan 2.36, there is a built in sum_to_zero_vector type, which can be used as follows.\nparameters {\n  sum_to_zero_vector[K] beta;\n  // ...\n}\nThis produces a vector of size K such that sum(beta) = 0. In the unconstrained representation requires only K - 1 values because the last is determined by the first K - 1.\nPlacing a prior on beta in this parameterization, for example,\n  beta ~ normal(0, 1);\nleads to a subtly different posterior than what you would get with the same prior on an unconstrained size-K vector. As explained below, the variance is reduced.\nThe sum-to-zero constraint can be implemented naively by setting the last element to the negative sum of the first elements, i.e., \\(\\beta_K\n= -\\sum_{k=1}^{K-1} \\beta_k.\\) But that leads to high correlation among the \\(\\beta_k\\).\nThe transform used in Stan eliminates these correlations by constructing an orthogonal basis and applying it to the zero-sum-constraint; Seyboldt (2024) provides an explanation. The Stan Reference Manual provides the details in the chapter on transforms. Although any orthogonal basis can be used, Stan uses the inverse isometric log transform because it is convenient to describe and the transform simplifies to efficient scalar operations rather than more expensive matrix operations.\n\n\nOn the Stan forums, Aaron Goodman provided the following code to produce a prior with standard normal marginals on the components of beta,\nmodel {\n  beta ~ normal(0, inv(sqrt(1 - inv(K))));\n  // ...\n}\nThe scale component can be multiplied by sigma to produce a normal(0, sigma) prior marginally.\nTo generate distributions with marginals other than standard normal, the resulting beta may be scaled by some factor sigma and translated to some new location mu.\n\n\n\n\nAdding a prior such as \\(\\beta \\sim \\textsf{normal}(0,\\epsilon)\\) for a small \\(\\epsilon\\) will provide a kind of soft centering of a parameter vector \\(\\beta\\) by preferring, all else being equal, that \\(\\sum_{k=1}^K\n\\beta_k = 0\\). This approach is only guaranteed to roughly center if \\(\\beta\\) and the elementwise addition \\(\\beta + c\\) for a scalar constant \\(c\\) produce the same likelihood (perhaps by another vector \\(\\alpha\\) being transformed to \\(\\alpha - c\\), as in the IRT models). This is another way of achieving a symmetric prior, though it requires choosing an \\(\\epsilon\\). If \\(\\epsilon\\) is too large, there won’t be a strong enough centering effect and if it is too small, it will add high curvature to the target density and impede sampling.",
+    "objectID": "stan-users-guide/odes.html#footnotes",
+    "href": "stan-users-guide/odes.html#footnotes",
+    "title": "Ordinary Differential Equations",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis example is drawn from the documentation for the Boost Numeric Odeint library (Ahnert and Mulansky 2011), which Stan uses to implement the rk45 and ckrk solver.↩︎",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Regression Models"
+      "Ordinary Differential Equations"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#ordered-logistic.section",
-    "href": "stan-users-guide/regression.html#ordered-logistic.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/parallelization.html",
+    "href": "stan-users-guide/parallelization.html",
+    "title": "Parallelization",
     "section": "",
-    "text": "Ordered regression for an outcome \\(y_n \\in \\{ 1, \\dotsc, k \\}\\) with predictors \\(x_n \\in \\mathbb{R}^D\\) is determined by a single coefficient vector \\(\\beta \\in \\mathbb{R}^D\\) along with a sequence of cutpoints \\(c \\in\n\\mathbb{R}^{K-1}\\) sorted so that \\(c_d &lt; c_{d+1}\\). The discrete output is \\(k\\) if the linear predictor \\(x_n \\beta\\) falls between \\(c_{k-1}\\) and \\(c_k\\), assuming \\(c_0 = -\\infty\\) and \\(c_K = \\infty\\). The noise term is fixed by the form of regression, with examples for ordered logistic and ordered probit models.\n\n\nThe ordered logistic model can be coded in Stan using the ordered data type for the cutpoints and the built-in ordered_logistic distribution.\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  array[N] row_vector[D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ ordered_logistic(x[n] * beta, c);\n  }\n}\nThe vector of cutpoints c is declared as ordered[K - 1], which guarantees that c[k] is less than c[k + 1].\nIf the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.\n\n\nAn ordered probit model could be coded in exactly the same way by swapping the cumulative logistic (inv_logit) for the cumulative normal (Phi).\ndata {\n  int&lt;lower=2&gt; K;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; D;\n  array[N] int&lt;lower=1, upper=K&gt; y;\n  array[N] row_vector[D] x;\n}\nparameters {\n  vector[D] beta;\n  ordered[K - 1] c;\n}\nmodel {\n  vector[K] theta;\n  for (n in 1:N) {\n    real eta;\n    eta = x[n] * beta;\n    theta[1] = 1 - Phi(eta - c[1]);\n    for (k in 2:(K - 1)) {\n      theta[k] = Phi(eta - c[k - 1]) - Phi(eta - c[k]);\n    }\n    theta[K] = Phi(eta - c[K - 1]);\n    y[n] ~ categorical(theta);\n  }\n}\nThe logistic model could also be coded this way by replacing Phi with inv_logit, though the built-in encoding based on the softmax transform is more efficient and more numerically stable. A small efficiency gain could be achieved by computing the values Phi(eta - c[k]) once and storing them for re-use.",
+    "text": "Stan has support for different types of parallelization: multi-threading with Intel Threading Building Blocks (TBB), multi-processing with Message Passing Interface (MPI) and manycore processing with OpenCL.\nMulti-threading in Stan can be used with two mechanisms: reduce with summation and rectangular map. The latter can also be used with multi-processing.\nThe advantages of reduce with summation are:\n\nMore flexible argument interface, avoiding the packing and unpacking that is necessary with rectangular map.\nPartitions data for parallelization automatically (this is done manually in rectangular map).\nIs easier to use.\n\nThe advantages of rectangular map are:\n\nReturns a list of vectors, while the reduce summation returns only a scalar.\nCan be parallelized across multiple cores and multiple computers, while reduce summation can only parallelized across multiple cores on a single machine.\n\nThe actual speedup gained from using these functions will depend on many details. It is strongly recommended to only parallelize the computationally most expensive operations in a Stan program. Oftentimes this is the evaluation of the log likelihood for the observed data. When it is not clear which parts of the model is the most computationally expensive, we recommend using profiling, which is available in Stan 2.26 and newer.\nSince only portions of a Stan program will run in parallel, the maximal speedup one can achieve is capped, a phenomen described by Amdahl’s law.\n\n\nIt is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If g: U -&gt; real is the function and { x1, x2, ... } is an array of inputs, then that sum looks like:\ng(x1) + g(x2) + ...\nreduce_sum and reduce_sum_static are tools for parallelizing these calculations.\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g, but instead the partial sum function f: U[] -&gt; real, where f computes the partial sum corresponding to a slice of the sequence x passed in. Due to the associativity of the sum reduction it holds that:\ng(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })\n                      = f({ x1, x2 }) + f({ x3 })\n                      = f({ x1 }) + f({ x2, x3 })\n                      = f({ x1 }) + f({ x2 }) + f({ x3 })\nWith the partial sum function f: U[] -&gt; real reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load).\n\ngrainsize is the one tuning parameter. For reduce_sum, grainsize is a suggested partial sum size. A grainsize of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using reduce_sum unless time is spent carefully picking grainsize. For picking a grainsize, see details below.\nFor reduce_sum_static, grainsize specifies the maximal partial sum size. With reduce_sum_static it is more important to choose grainsize carefully since it entirely determines the partitioning of work. See details below.\nFor efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array { x1, x2, ... } and the shared arguments s1, s2, ... the effective sum (with individual terms) looks like:\ng(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...\nwhich can be written equivalently with partial sums to look like:\nf({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)\nwhere the particular slicing of the x array can change.\nGiven this, the signatures are:\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nreal reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n\nf - User defined function that computes partial sums\nx - Array to slice, each element corresponds to a term in the summation\ngrainsize - Target for size of slices\ns1, s2, ... - Arguments shared in every term\n\nThe user-defined partial sum functions have the signature:\nreal f(array[] T x_slice, int start, int end, T1 s1, T2 s2, ...)\nand take the arguments:\n\nx_slice - The subset of x (from reduce_sum / reduce_sum_static) for which this partial sum is responsible (x_slice = x[start:end])\nstart - An integer specifying the first term in the partial sum\nend - An integer specifying the last term in the partial sum (inclusive)\ns1, s2, ... - Arguments shared in every term (passed on without modification from the reduce_sum / reduce_sum_static call)\n\nThe user-provided function f is expected to compute the partial sum with the terms start through end of the overall sum. The user function is passed the subset x[start:end] as x_slice. start and end are passed so that f can index any of the tailing sM arguments as necessary. The trailing sM arguments are passed without modification to every call of f.\nA reduce_sum (or reduce_sum_static) call:\nreal sum = reduce_sum(f, x, grainsize, s1, s2, ...);\ncan be replaced by either:\nreal sum = f(x, 1, size(x), s1, s2, ...);\nor the code:\nreal sum = 0.0;\nfor(i in 1:size(x)) {\n  sum += f({ x[i] }, i, i, s1, s2, ...);\n}\n\n\nLogistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * x);\n}\nIn this model predictions are made about the N outputs y using the covariate x. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:\ny ~ bernoulli_logit(beta[1] + beta[2] * x);\ncan be rewritten (up to a proportionality constant) as:\nfor(n in 1:N) {\n  target += bernoulli_logit_lpmf(y[n] | beta[1] + beta[2] * x[n])\n}\nNow it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in Reduce-Sum, such a function can be written like:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\nThe likelihood statement in the model can now be written:\ntarget += partial_sum(y, 1, N, x, beta); // Sum terms 1 to N of the likelihood\nIn this example, y was chosen to be sliced over because there is one term in the summation per value of y. Technically x would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g. slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because x is a shared argument, it is subset accordingly with start:end. With this function, reduce summation can be used to automatically parallelize the likelihood:\nint grainsize = 1;\ntarget += reduce_sum(partial_sum, y,\n                     grainsize,\n                     x, beta);\nThe reduce summation facility automatically breaks the sum into pieces and computes them in parallel. grainsize = 1 specifies that the grainsize should be estimated automatically. The final model is:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  int grainsize = 1;\n  beta ~ std_normal();\n  target += reduce_sum(partial_sum, y,\n                       grainsize,\n                       x, beta);\n}\n\n\n\nThe rational for choosing a sensible grainsize is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.\nIn reduce_sum, grainsize is a recommendation on how to partition the work in the partial sum into smaller pieces. A grainsize of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.\nIn reduce_sum_static, grainsize is an upper limit on the worksize. Work will be split until all partial sums are just smaller than grainsize (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible grainsize.\nIn order to figure out an optimal grainsize, if there are N terms and M cores, run a quick test model with grainsize set roughly to N / M. Record the time, cut the grainsize in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable grainsize for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.\nFor instance, in a model with N=10000 and M = 4, start with grainsize = 2500, and sequentially try grainsize = 1250, grainsize = 625, etc.\nIt is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller grainsize that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a grainsize of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.\n\n\n\n\nMap-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).\nA map function is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, \\(f(x) = x^2\\), over the vector \\([3, 5, 10]\\) produces the vector \\([9, 25, 100]\\). In other words, map applies the square function elementwise.\nThe output of mapping a sequence is often fed into a reduction. A reduction function takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, map-reduce.\n\n\nIn order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.\n\n\n\nStan’s map function has the following signature\nvector map_rect((vector, vector, array[] real, array[] int):vector f,\n                vector phi, array[] vector thetas,\n                data array[,] real x_rs, data array[,] int x_is);\nThe arrays thetas of parameters, x_rs of real data, and x_is of integer data have the suffix “s” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function f. The value of phi is reused in each mapped operation.\nThe _rect suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.\nThe last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the data qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.\nThe notation (vector, vector, array[] real, array[] int):vector indicates that the function argument f must have the following signature.\nvector f(vector phi, vector theta,\n         data array[] real x_r, data array[] int x_i);\nAlthough f will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.\n\n\nStan’s map function applies the function f to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments theta, x_r, and x_i must be arrays of the same size. If the arrays are all size N, the result is defined as follows.\nmap_rect(f, phi, thetas, xs, ns)\n= f(phi, thetas[1], xs[1], ns[1]) . f(phi, thetas[2], xs[2], ns[2])\n  . ... . f(phi, thetas[N], xs[N], ns[N])\nThe dot operators in the notation above are meant to indicate concatenation (implemented as append_row in Stan). The output of each application of f is a vector, and the sequence of N vectors is concatenated together to return a single vector.\n\n\n\n\nAn example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.\n\n\nConsider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.\ndata {\n  array[12] int y;\n  array[12] real x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * to_vector(x));\n}\nThe program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The bernoulli_logit distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.\n\n\n\nThe unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.\nfunctions {\n  vector lr(vector beta, vector theta, array[] real x, array[] int y) {\n    real lp = bernoulli_logit_lpmf(y | beta[1]\n                                       + to_vector(x) * beta[2]);\n    return [lp]';\n  }\n}\ndata {\n  array[12] int y;\n  array[12] real x;\n}\ntransformed data {\n  // K = 3 shards\n  array[3, 4] int ys = { y[1:4], y[5:8], y[9:12] };\n  array[3, 4] real xs = { x[1:4], x[5:8], x[9:12] };\n  array[3] vector[0] theta;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  target += sum(map_rect(lr, beta, theta, xs, ys));\n}\nThe first piece of the code is the actual function to compute the logistic regression. The argument beta will contain the regression coefficients (intercept and slope), as before. The second argument theta of job-specific parameters is not used, but nevertheless must be present. The modeled data y is passed as an array of integers and the predictors x as an array of real values. The function body then computes the log probability mass of y and assigns it to the local variable lp. This variable is then used in [lp]' to construct a row vector and then transpose it to a vector to return.\nThe data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.1\nThe value 3 is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array ys contains the modeled data variables; each element of the array ys is an array of size four. The second array xs is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors x stand in a one-to-one relationship with the modeled data y. The final array theta is also of size three; its elements are empty vectors, because there are no shard-specific parameters.\nThe parameters and the prior are as before. The likelihood is now coded using map-reduce. The function lr to compute the log probability mass is mapped over the data xs and ys, which contain the original predictors and outcomes broken into shards. The parameters beta are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters theta contains only empty vectors.\n\n\n\n\nConsider a hierarchical model of American presidential voting behavior based on state of residence.2\nEach of the fifty states \\(k \\in \\{1,\\dotsc,50\\}\\) will have its own slope \\(\\beta_k\\) and intercept \\(\\alpha_k\\) to model the log odds of voting for the Republican candidate as a function of income. Suppose there are \\(N\\) voters and with voter \\(n \\in 1{:}N\\) being in state \\(s[n]\\) with income \\(x_n\\). The data model for the vote \\(y_n \\in \\{ 0, 1 \\}\\) is \\[\ny_n \\sim \\textsf{Bernoulli}\n\\Big(\n  \\operatorname{logit}^{-1}\\left( \\alpha_{s[n]} + \\beta_{s[n]} \\, x_n \\right)\n\\Big).\n\\]\nThe slopes and intercepts get hierarchical priors, \\[\\begin{align*}\n\\alpha_k &\\sim \\textsf{normal}(\\mu_{\\alpha}, \\sigma_{\\alpha}) \\\\\n\\beta_k  &\\sim \\textsf{normal}(\\mu_{\\beta}, \\sigma_{\\beta})\n\\end{align*}\\]\n\n\nThis model can be coded up in Stan directly as follows.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  matrix[K, 2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  for (i in 1:2) {\n    beta[ , i] ~ normal(mu[i], sigma[i]);\n  }\n  y ~ bernoulli_logit(beta[kk, 1] + beta[kk, 2] .* x);\n}\nFor this model the vector of predictors x is coded as a vector, corresponding to how it is used in the model. The priors for mu and sigma are vectorized. The priors on the two components of beta (intercept and slope, respectively) are stored in a \\(K \\times 2\\) matrix.\nThe distribution statement is also vectorized using multi-indexing with index kk for the states and elementwise multiplication (.*) for the income x. The vectorized distribution statement works out to the same thing as the following less efficient looped form.\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(beta[kk[n], 1] + beta[kk[n], 2] * x[n]);\n}\n\n\n\nThe mapped version of the model will map over the states K. This means the group-level parameters, real data, and integer-data must be arrays of the same size.\nThe mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name bernoulli_logit_glm derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).\nfunctions {\n vector bl_glm(vector mu_sigma, vector beta,\n               array[] real x, array[] int y) {\n   vector[2] mu = mu_sigma[1:2];\n   vector[2] sigma = mu_sigma[3:4];\n   real lp = normal_lpdf(beta | mu, sigma);\n   real ll = bernoulli_logit_lpmf(y | beta[1] + beta[2] * to_vector(x));\n   return [lp + ll]';\n }\n}\nThe shared parameter mu_sigma contains the locations (mu_sigma[1:2]) and scales (mu_sigma[3:4]) of the priors, which are extracted in the first two lines of the program. The variable lp is assigned the log density of the prior on beta. The vector beta is of size two, as are the vectors mu and sigma, so everything lines up for the vectorization. Next, the variable ll is assigned to the log likelihood contribution for the group. Here beta[1] is the intercept of the regression and beta[2] the slope. The predictor array x needs to be converted to a vector allow the multiplication.\nThe data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\ntransformed data {\n  int&lt;lower=0&gt; J = N / K;\n  array[K, J] real x_r;\n  array[K, J] int&lt;lower=0, upper=1&gt; x_i;\n  {\n    int pos = 1;\n    for (k in 1:K) {\n      int end = pos + J - 1;\n      x_r[k] = to_array_1d(x[pos:end]);\n      x_i[k] = to_array_1d(y[pos:end]);\n      pos += J;\n    }\n  }\n}\nThe integer J is set to the number of observations per group.3\nThe real data array x_r holds the predictors and the integer data array x_i holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector x (and converting it to an array) and slicing the outcome array y.\nGiven the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters mu and sigma, but moves the prior for beta and the likelihood to the mapped function.\nparameters {\n  array[K] vector[2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  target += sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));\n                         \n}\nThe model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.\n\n\n\n\nThe previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by map_rect.\n\n\nIf each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.\n\n\n\nThe output of each mapped function is concatenated in order of inputs to produce the output of map_rect. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the to_matrix function if they are rectangular.\nIf the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.\n\n\n\n\n\nOpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.\nStan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the STAN_OPENCL makefile flag. Setting this flag means that the Stan-to-C++ translator (stanc3) will be supplied the --use-opencl flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.\nIn Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:\n\nbernoulli_lpmf\nbernoulli_logit_lpmf\nbernoulli_logit_glm_lpmf*\nbeta_lpdf\nbeta_proportion_lpdf\nbinomial_lpmf\ncategorical_logit_glm_lpmf*\ncauchy_lpdf\nchi_square_lpdf\ndouble_exponential_lpdf\nexp_mod_normal_lpdf\nexponential_lpdf\nfrechet_lpdf\ngamma_lpdf\ngumbel_lpdf\ninv_chi_square_lpdf\ninv_gamma_lpdf\nlogistic_lpdf\nlognormal_lpdf\nneg_binomial_lpmf\nneg_binomial_2_lpmf\nneg_binomial_2_log_lpmf\nneg_binomial_2_log_glm_lpmf*\nnormal_lpdf\nnormal_id_glm_lpdf*\nordered_logistic_glm_lpmf*\npareto_lpdf\npareto_type_2_lpdf\npoisson_lpmf\npoisson_log_lpmf\npoisson_log_glm_lpmf*\nrayleigh_lpdf\nscaled_inv_chi_square_lpdf\nskew_normal_lpdf\nstd_normal_lpdf\nstudent_t_lpdf\nuniform_lpdf\nweibull_lpdf\n\n* OpenCL is not used when the covariate argument to the GLM functions is a row_vector.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Programming Techniques",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#hierarchical-regression",
-    "href": "stan-users-guide/regression.html#hierarchical-regression",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/parallelization.html#reduce-sum",
+    "href": "stan-users-guide/parallelization.html#reduce-sum",
+    "title": "Parallelization",
     "section": "",
-    "text": "The simplest multilevel model is a hierarchical model in which the data are grouped into \\(L\\) distinct categories (or levels). An extreme approach would be to completely pool all the data and estimate a common vector of regression coefficients \\(\\beta\\). At the other extreme, an approach with no pooling assigns each level \\(l\\) its own coefficient vector \\(\\beta_l\\) that is estimated separately from the other levels. A hierarchical model is an intermediate solution where the degree of pooling is determined by the data and a prior on the amount of pooling.\nSuppose each binary outcome \\(y_n \\in \\{ 0, 1 \\}\\) has an associated level, \\(ll_n \\in \\{ 1, \\dotsc, L \\}\\). Each outcome will also have an associated predictor vector \\(x_n \\in \\mathbb{R}^D\\). Each level \\(l\\) gets its own coefficient vector \\(\\beta_l \\in \\mathbb{R}^D\\). The hierarchical structure involves drawing the coefficients \\(\\beta_{l,d}\n\\in \\mathbb{R}\\) from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.\nThe following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.\ndata {\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  int&lt;lower=1&gt; L;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  array[N] int&lt;lower=1, upper=L&gt; ll;\n  array[N] row_vector[D] x;\n}\nparameters {\n  array[D] real mu;\n  array[D] real&lt;lower=0&gt; sigma;\n  array[L] vector[D] beta;\n}\nmodel {\n  for (d in 1:D) {\n    mu[d] ~ normal(0, 100);\n    for (l in 1:L) {\n      beta[l, d] ~ normal(mu[d], sigma[d]);\n    }\n  }\n  for (n in 1:N) {\n    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));\n  }\n}\nThe standard deviation parameter sigma gets an implicit uniform prior on \\((0,\\infty)\\) because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the regression priors section for recommendations on priors for regression coefficients and scales.\n\n\nWhere possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.\nThe first optimization vectorizes the for-loop over D as\nmu ~ normal(0, 100);\nfor (l in 1:L) {\n  beta[l] ~ normal(mu, sigma);\n}\nThe declaration of beta as an array of vectors means that the expression beta[l] denotes a vector. Although beta could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the indexing efficiency section for more information on the efficiency tradeoffs among arrays, vectors, and matrices.\nThis model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,3\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);\n}\nUnlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.\n{\n  vector[N] x_beta_ll;\n  for (n in 1:N) {\n    x_beta_ll[n] = x[n] * beta[ll[n]];\n  }\n  y ~ bernoulli_logit(x_beta_ll);\n}\nThe brackets introduce a new scope for the local variable x_beta_ll; alternatively, the variable may be declared at the top of the model block.\nIn some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.",
+    "text": "It is often necessary in probabilistic modeling to compute the sum of a number of independent function evaluations. This occurs, for instance, when evaluating a number of conditionally independent terms in a log-likelihood. If g: U -&gt; real is the function and { x1, x2, ... } is an array of inputs, then that sum looks like:\ng(x1) + g(x2) + ...\nreduce_sum and reduce_sum_static are tools for parallelizing these calculations.\nFor efficiency reasons the reduce function doesn’t work with the element-wise evaluated function g, but instead the partial sum function f: U[] -&gt; real, where f computes the partial sum corresponding to a slice of the sequence x passed in. Due to the associativity of the sum reduction it holds that:\ng(x1) + g(x2) + g(x3) = f({ x1, x2, x3 })\n                      = f({ x1, x2 }) + f({ x3 })\n                      = f({ x1 }) + f({ x2, x3 })\n                      = f({ x1 }) + f({ x2 }) + f({ x3 })\nWith the partial sum function f: U[] -&gt; real reduction of a large number of terms can be evaluated in parallel automatically, since the overall sum can be partitioned into arbitrary smaller partial sums. The exact partitioning into the partial sums is not under the control of the user. However, since the exact numerical result will depend on the order of summation, Stan provides two versions of the reduce summation facility:\n\nreduce_sum: Automatically choose partial sums partitioning based on a dynamic scheduling algorithm.\nreduce_sum_static: Compute the same sum as reduce_sum, but partition the input in the same way for given data set (in reduce_sum this partitioning might change depending on computer load).\n\ngrainsize is the one tuning parameter. For reduce_sum, grainsize is a suggested partial sum size. A grainsize of 1 leaves the partitioning entirely up to the scheduler. This should be the default way of using reduce_sum unless time is spent carefully picking grainsize. For picking a grainsize, see details below.\nFor reduce_sum_static, grainsize specifies the maximal partial sum size. With reduce_sum_static it is more important to choose grainsize carefully since it entirely determines the partitioning of work. See details below.\nFor efficiency and convenience additional shared arguments can be passed to every term in the sum. So for the array { x1, x2, ... } and the shared arguments s1, s2, ... the effective sum (with individual terms) looks like:\ng(x1, s1, s2, ...) + g(x2, s1, s2, ...) + g(x3, s1, s2, ...) + ...\nwhich can be written equivalently with partial sums to look like:\nf({ x1, x2 }, s1, s2, ...) + f({ x3 }, s1, s2, ...)\nwhere the particular slicing of the x array can change.\nGiven this, the signatures are:\nreal reduce_sum(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\nreal reduce_sum_static(F f, array[] T x, int grainsize, T1 s1, T2 s2, ...)\n\nf - User defined function that computes partial sums\nx - Array to slice, each element corresponds to a term in the summation\ngrainsize - Target for size of slices\ns1, s2, ... - Arguments shared in every term\n\nThe user-defined partial sum functions have the signature:\nreal f(array[] T x_slice, int start, int end, T1 s1, T2 s2, ...)\nand take the arguments:\n\nx_slice - The subset of x (from reduce_sum / reduce_sum_static) for which this partial sum is responsible (x_slice = x[start:end])\nstart - An integer specifying the first term in the partial sum\nend - An integer specifying the last term in the partial sum (inclusive)\ns1, s2, ... - Arguments shared in every term (passed on without modification from the reduce_sum / reduce_sum_static call)\n\nThe user-provided function f is expected to compute the partial sum with the terms start through end of the overall sum. The user function is passed the subset x[start:end] as x_slice. start and end are passed so that f can index any of the tailing sM arguments as necessary. The trailing sM arguments are passed without modification to every call of f.\nA reduce_sum (or reduce_sum_static) call:\nreal sum = reduce_sum(f, x, grainsize, s1, s2, ...);\ncan be replaced by either:\nreal sum = f(x, 1, size(x), s1, s2, ...);\nor the code:\nreal sum = 0.0;\nfor(i in 1:size(x)) {\n  sum += f({ x[i] }, i, i, s1, s2, ...);\n}\n\n\nLogistic regression is a useful example to clarify both the syntax and semantics of reduce summation and how it can be used to speed up a typical model. A basic logistic regression can be coded in Stan as:\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * x);\n}\nIn this model predictions are made about the N outputs y using the covariate x. The intercept and slope of the linear equation are to be estimated. The key point to getting this calculation to use reduce summation, is recognizing that the statement:\ny ~ bernoulli_logit(beta[1] + beta[2] * x);\ncan be rewritten (up to a proportionality constant) as:\nfor(n in 1:N) {\n  target += bernoulli_logit_lpmf(y[n] | beta[1] + beta[2] * x[n])\n}\nNow it is clear that the calculation is the sum of a number of conditionally independent Bernoulli log probability statements, which is the condition where reduce summation is useful. To use the reduce summation, a function must be written that can be used to compute arbitrary partial sums of the total sum. Using the interface defined in Reduce-Sum, such a function can be written like:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\nThe likelihood statement in the model can now be written:\ntarget += partial_sum(y, 1, N, x, beta); // Sum terms 1 to N of the likelihood\nIn this example, y was chosen to be sliced over because there is one term in the summation per value of y. Technically x would have worked as well. Use whatever conceptually makes the most sense for a given model, e.g. slice over independent terms like conditionally independent observations or groups of observations as in hierarchical models. Because x is a shared argument, it is subset accordingly with start:end. With this function, reduce summation can be used to automatically parallelize the likelihood:\nint grainsize = 1;\ntarget += reduce_sum(partial_sum, y,\n                     grainsize,\n                     x, beta);\nThe reduce summation facility automatically breaks the sum into pieces and computes them in parallel. grainsize = 1 specifies that the grainsize should be estimated automatically. The final model is:\nfunctions {\n  real partial_sum(array[] int y_slice,\n                   int start, int end,\n                   vector x,\n                   vector beta) {\n    return bernoulli_logit_lpmf(y_slice | beta[1] + beta[2] * x[start:end]);\n  }\n}\ndata {\n  int N;\n  array[N] int y;\n  vector[N] x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  int grainsize = 1;\n  beta ~ std_normal();\n  target += reduce_sum(partial_sum, y,\n                       grainsize,\n                       x, beta);\n}\n\n\n\nThe rational for choosing a sensible grainsize is based on balancing the overhead implied by creating many small tasks versus creating fewer large tasks which limits the potential parallelism.\nIn reduce_sum, grainsize is a recommendation on how to partition the work in the partial sum into smaller pieces. A grainsize of 1 leaves this entirely up to the internal scheduler and should be chosen if no benchmarking of other grainsizes is done. Ideally this will be efficient, but there are no guarantees.\nIn reduce_sum_static, grainsize is an upper limit on the worksize. Work will be split until all partial sums are just smaller than grainsize (and the split will happen the same way every time for the same inputs). For the static version it is more important to select a sensible grainsize.\nIn order to figure out an optimal grainsize, if there are N terms and M cores, run a quick test model with grainsize set roughly to N / M. Record the time, cut the grainsize in half, and run the test again. Repeat this iteratively until the model runtime begins to increase. This is a suitable grainsize for the model, because this ensures the calculations can be carried out with the most parallelism without losing too much efficiency.\nFor instance, in a model with N=10000 and M = 4, start with grainsize = 2500, and sequentially try grainsize = 1250, grainsize = 625, etc.\nIt is important to repeat this process until performance gets worse. It is possible after many halvings nothing happens, but there might still be a smaller grainsize that performs better. Even if a sum has many tens of thousands of terms, depending on the internal calculations, a grainsize of thirty or forty or smaller might be the best, and it is difficult to predict this behavior. Without doing these halvings until performance actually gets worse, it is easy to miss this.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Programming Techniques",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#hierarchical-priors.section",
-    "href": "stan-users-guide/regression.html#hierarchical-priors.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/parallelization.html#map-rect",
+    "href": "stan-users-guide/parallelization.html#map-rect",
+    "title": "Parallelization",
     "section": "",
-    "text": "Priors on priors, also known as “hyperpriors,” should be treated the same way as priors on lower-level parameters in that as much prior information as is available should be brought to bear. Because hyperpriors often apply to only a handful of lower-level parameters, care must be taken to ensure the posterior is both proper and not overly sensitive either statistically or computationally to wide tails in the priors.\n\n\nThe fundamental problem with maximum likelihood estimation (MLE) in the hierarchical model setting is that as the hierarchical variance drops and the values cluster around the hierarchical mean, the overall density grows without bound. As an illustration, consider a simple hierarchical linear regression (with fixed prior mean) of \\(y_n \\in\n\\mathbb{R}\\) on \\(x_n \\in \\mathbb{R}^K\\), formulated as \\[\\begin{align*}\ny_n     & \\sim \\textsf{normal}(x_n \\beta, \\sigma) \\\\\n\\beta_k & \\sim \\textsf{normal}(0,\\tau) \\\\\n\\tau    & \\sim \\textsf{Cauchy}(0,2.5) \\\\\n\\end{align*}\\]\nIn this case, as \\(\\tau \\rightarrow 0\\) and \\(\\beta_k \\rightarrow 0\\), the posterior density \\[ p(\\beta,\\tau,\\sigma|y,x) \\propto p(y|x,\\beta,\\tau,\\sigma) \\] grows without bound. See the Neal’s funnel density, which has similar behavior.\nThere is obviously no MLE estimate for \\(\\beta,\\tau,\\sigma\\) in such a case, and therefore the model must be modified if posterior modes are to be used for inference. The approach recommended by Chung et al. (2013) is to use a gamma distribution as a prior, such as \\[\n\\sigma \\sim \\textsf{Gamma}(2, 1/A),\n\\] for a reasonably large value of \\(A\\), such as \\(A = 10\\).",
+    "text": "Map-reduce allows large calculations (e.g., log likelihoods) to be broken into components which may be calculated modularly (e.g., data blocks) and combined (e.g., by summation and incrementing the target log density).\nA map function is a higher-order function that applies an argument function to every member of some collection, returning a collection of the results. For example, mapping the square function, \\(f(x) = x^2\\), over the vector \\([3, 5, 10]\\) produces the vector \\([9, 25, 100]\\). In other words, map applies the square function elementwise.\nThe output of mapping a sequence is often fed into a reduction. A reduction function takes an arbitrarily long sequence of inputs and returns a single output. Examples of reduction functions are summation (with the return being a single value) or sorting (with the return being a sorted sequence). The combination of mapping and reducing is so common it has its own name, map-reduce.\n\n\nIn order to generalize the form of functions and results that are possible and accommodate both parameters (which need derivatives) and data values (which don’t), Stan’s map function operates on more than just a sequence of inputs.\n\n\n\nStan’s map function has the following signature\nvector map_rect((vector, vector, array[] real, array[] int):vector f,\n                vector phi, array[] vector thetas,\n                data array[,] real x_rs, data array[,] int x_is);\nThe arrays thetas of parameters, x_rs of real data, and x_is of integer data have the suffix “s” to indicate they are arrays. These arrays must all be the same size, as they will be mapped in parallel by the function f. The value of phi is reused in each mapped operation.\nThe _rect suffix in the name arises because the data structures it takes as arguments are rectangular. In order to deal with ragged inputs, ragged inputs must be padded out to rectangular form.\nThe last two arguments are two dimensional arrays of real and integer data values. These argument types are marked with the data qualifier to indicate that they must only contain variables originating in the data or transformed data blocks. This will allow such data to be pinned to a processor on which it is being processed to reduce communication overhead.\nThe notation (vector, vector, array[] real, array[] int):vector indicates that the function argument f must have the following signature.\nvector f(vector phi, vector theta,\n         data array[] real x_r, data array[] int x_i);\nAlthough f will often return a vector of size one, the built-in flexibility allows general multivariate functions to be mapped, even raggedly.\n\n\nStan’s map function applies the function f to the shared parameters along with one element each of the job parameters, real data, and integer data arrays. Each of the arguments theta, x_r, and x_i must be arrays of the same size. If the arrays are all size N, the result is defined as follows.\nmap_rect(f, phi, thetas, xs, ns)\n= f(phi, thetas[1], xs[1], ns[1]) . f(phi, thetas[2], xs[2], ns[2])\n  . ... . f(phi, thetas[N], xs[N], ns[N])\nThe dot operators in the notation above are meant to indicate concatenation (implemented as append_row in Stan). The output of each application of f is a vector, and the sequence of N vectors is concatenated together to return a single vector.\n\n\n\n\nAn example should help to clarify both the syntax and semantics of the mapping operation and how it may be combined with reductions built into Stan to provide a map-reduce implementation.\n\n\nConsider the following simple logistic regression model, which is coded unconventionally to accommodate direct translation to a mapped implementation.\ndata {\n  array[12] int y;\n  array[12] real x;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  y ~ bernoulli_logit(beta[1] + beta[2] * to_vector(x));\n}\nThe program is unusual in that it (a) hardcodes the data size, which is not required by the map function but is just used here for simplicity, (b) represents the predictors as a real array even though it needs to be used as a vector, and (c) represents the regression coefficients (intercept and slope) as a vector even though they’re used individually. The bernoulli_logit distribution is used because the argument is on the logit scale—it implicitly applies the inverse logit function to map the argument to a probability.\n\n\n\nThe unmapped logistic regression model described in the previous subsection may be implemented using Stan’s rectangular mapping functionality as follows.\nfunctions {\n  vector lr(vector beta, vector theta, array[] real x, array[] int y) {\n    real lp = bernoulli_logit_lpmf(y | beta[1]\n                                       + to_vector(x) * beta[2]);\n    return [lp]';\n  }\n}\ndata {\n  array[12] int y;\n  array[12] real x;\n}\ntransformed data {\n  // K = 3 shards\n  array[3, 4] int ys = { y[1:4], y[5:8], y[9:12] };\n  array[3, 4] real xs = { x[1:4], x[5:8], x[9:12] };\n  array[3] vector[0] theta;\n}\nparameters {\n  vector[2] beta;\n}\nmodel {\n  beta ~ std_normal();\n  target += sum(map_rect(lr, beta, theta, xs, ys));\n}\nThe first piece of the code is the actual function to compute the logistic regression. The argument beta will contain the regression coefficients (intercept and slope), as before. The second argument theta of job-specific parameters is not used, but nevertheless must be present. The modeled data y is passed as an array of integers and the predictors x as an array of real values. The function body then computes the log probability mass of y and assigns it to the local variable lp. This variable is then used in [lp]' to construct a row vector and then transpose it to a vector to return.\nThe data are taken in as before. There is an additional transformed data block that breaks the data up into three shards.1\nThe value 3 is also hard coded; a more practical program would allow the number of shards to be controlled. There are three parallel arrays defined here, each of size three, corresponding to the number of shards. The array ys contains the modeled data variables; each element of the array ys is an array of size four. The second array xs is for the predictors, and each element of it is also of size four. These contained arrays are the same size because the predictors x stand in a one-to-one relationship with the modeled data y. The final array theta is also of size three; its elements are empty vectors, because there are no shard-specific parameters.\nThe parameters and the prior are as before. The likelihood is now coded using map-reduce. The function lr to compute the log probability mass is mapped over the data xs and ys, which contain the original predictors and outcomes broken into shards. The parameters beta are in the first argument because they are shared across shards. There are no shard-specific parameters, so the array of job-specific parameters theta contains only empty vectors.\n\n\n\n\nConsider a hierarchical model of American presidential voting behavior based on state of residence.2\nEach of the fifty states \\(k \\in \\{1,\\dotsc,50\\}\\) will have its own slope \\(\\beta_k\\) and intercept \\(\\alpha_k\\) to model the log odds of voting for the Republican candidate as a function of income. Suppose there are \\(N\\) voters and with voter \\(n \\in 1{:}N\\) being in state \\(s[n]\\) with income \\(x_n\\). The data model for the vote \\(y_n \\in \\{ 0, 1 \\}\\) is \\[\ny_n \\sim \\textsf{Bernoulli}\n\\Big(\n  \\operatorname{logit}^{-1}\\left( \\alpha_{s[n]} + \\beta_{s[n]} \\, x_n \\right)\n\\Big).\n\\]\nThe slopes and intercepts get hierarchical priors, \\[\\begin{align*}\n\\alpha_k &\\sim \\textsf{normal}(\\mu_{\\alpha}, \\sigma_{\\alpha}) \\\\\n\\beta_k  &\\sim \\textsf{normal}(\\mu_{\\beta}, \\sigma_{\\beta})\n\\end{align*}\\]\n\n\nThis model can be coded up in Stan directly as follows.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  matrix[K, 2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  for (i in 1:2) {\n    beta[ , i] ~ normal(mu[i], sigma[i]);\n  }\n  y ~ bernoulli_logit(beta[kk, 1] + beta[kk, 2] .* x);\n}\nFor this model the vector of predictors x is coded as a vector, corresponding to how it is used in the model. The priors for mu and sigma are vectorized. The priors on the two components of beta (intercept and slope, respectively) are stored in a \\(K \\times 2\\) matrix.\nThe distribution statement is also vectorized using multi-indexing with index kk for the states and elementwise multiplication (.*) for the income x. The vectorized distribution statement works out to the same thing as the following less efficient looped form.\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(beta[kk[n], 1] + beta[kk[n], 2] * x[n]);\n}\n\n\n\nThe mapped version of the model will map over the states K. This means the group-level parameters, real data, and integer-data must be arrays of the same size.\nThe mapped implementation requires a function to be mapped. In this function we can’t use distribution statements, but need to accumulate the desired log prior and log likelihood terms to the return value. The following function evaluates both the likelihood for the data observed for a group as well as the prior for the group-specific parameters (the name bernoulli_logit_glm derives from the fact that it’s a generalized linear model with a Bernoulli data model and logistic link function).\nfunctions {\n vector bl_glm(vector mu_sigma, vector beta,\n               array[] real x, array[] int y) {\n   vector[2] mu = mu_sigma[1:2];\n   vector[2] sigma = mu_sigma[3:4];\n   real lp = normal_lpdf(beta | mu, sigma);\n   real ll = bernoulli_logit_lpmf(y | beta[1] + beta[2] * to_vector(x));\n   return [lp + ll]';\n }\n}\nThe shared parameter mu_sigma contains the locations (mu_sigma[1:2]) and scales (mu_sigma[3:4]) of the priors, which are extracted in the first two lines of the program. The variable lp is assigned the log density of the prior on beta. The vector beta is of size two, as are the vectors mu and sigma, so everything lines up for the vectorization. Next, the variable ll is assigned to the log likelihood contribution for the group. Here beta[1] is the intercept of the regression and beta[2] the slope. The predictor array x needs to be converted to a vector allow the multiplication.\nThe data block is identical to that of the previous program, but repeated here for convenience. A transformed data block computes the data structures needed for the mapping by organizing the data into arrays indexed by group.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  vector[N] x;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\ntransformed data {\n  int&lt;lower=0&gt; J = N / K;\n  array[K, J] real x_r;\n  array[K, J] int&lt;lower=0, upper=1&gt; x_i;\n  {\n    int pos = 1;\n    for (k in 1:K) {\n      int end = pos + J - 1;\n      x_r[k] = to_array_1d(x[pos:end]);\n      x_i[k] = to_array_1d(y[pos:end]);\n      pos += J;\n    }\n  }\n}\nThe integer J is set to the number of observations per group.3\nThe real data array x_r holds the predictors and the integer data array x_i holds the outcomes. The grouped data arrays are constructed by slicing the predictor vector x (and converting it to an array) and slicing the outcome array y.\nGiven the transformed data with groupings, the parameters are the same as the previous program. The model has the same priors for the hyperparameters mu and sigma, but moves the prior for beta and the likelihood to the mapped function.\nparameters {\n  array[K] vector[2] beta;\n  vector[2] mu;\n  vector&lt;lower=0&gt;[2] sigma;\n}\nmodel {\n  mu ~ normal(0, 2);\n  sigma ~ normal(0, 2);\n  target += sum(map_rect(bl_glm, append_row(mu, sigma), beta, x_r, x_i));\n                         \n}\nThe model as written here computes the priors for each group’s parameters along with the likelihood contribution for the group. An alternative mapping would leave the prior in the model block and only map the likelihood computation. In a serial setting this shouldn’t make much of a difference, but with parallelization, there is reduced communication (the prior’s parameters need not be transmitted) and also reduced parallelization with the version that leaves the prior in the model block.\n\n\n\n\nThe previous examples included rectangular data structures and single outputs. Despite the name, this is not technically required by map_rect.\n\n\nIf each group has a different number of observations, then the rectangular data structures for predictors and outcomes will need to be padded out to be rectangular. In addition, the size of the ragged structure will need to be passed as integer data. This holds for shards with varying numbers of parameters as well as varying numbers of data points.\n\n\n\nThe output of each mapped function is concatenated in order of inputs to produce the output of map_rect. When every shard returns a singleton (size one) array, the result is the same size as the number of shards and is easy to deal with downstream. If functions return longer arrays, they can still be structured using the to_matrix function if they are rectangular.\nIf the outputs are of varying sizes, then there will have to be some way to convert it back to a usable form based on the input, because there is no way to directly return sizes or a ragged structure.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Programming Techniques",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#item-response-models.section",
-    "href": "stan-users-guide/regression.html#item-response-models.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/parallelization.html#opencl",
+    "href": "stan-users-guide/parallelization.html#opencl",
+    "title": "Parallelization",
     "section": "",
-    "text": "Item-response theory (IRT) models the situation in which a number of students each answer one or more of a group of test questions. The model is based on parameters for the ability of the students, the difficulty of the questions, and in more articulated models, the discriminativeness of the questions and the probability of guessing correctly; see Gelman and Hill (2007, pps. 314–320) for a textbook introduction to hierarchical IRT models and Curtis (2010) for encodings of a range of IRT models in BUGS.\n\n\nThe data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.\ndata {\n  int&lt;lower=1&gt; J;                     // number of students\n  int&lt;lower=1&gt; K;                     // number of questions\n  int&lt;lower=1&gt; N;                     // number of observations\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // student for observation n\n  array[N] int&lt;lower=1, upper=K&gt; kk;  // question for observation n\n  array[N] int&lt;lower=0, upper=1&gt; y;   // correctness for observation n\n}\nThis declares a total of N student-question pairs in the data set, where each n in 1:N indexes a binary observation y[n] of the correctness of the answer of student jj[n] on question kk[n].\nThe prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.\n\n\n\nThe 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).\nThe model parameters are declared as follows.\nparameters {\n  real delta;            // mean student ability\n  array[J] real alpha;   // ability of student j - mean ability\n  array[K] real beta;    // difficulty of question k\n}\nThe parameter alpha[J] is the ability coefficient for student j and beta[k] is the difficulty coefficient for question k. The non-standard parameterization used here also includes an intercept term delta, which represents the average student’s response to the average question.4\nThe model itself is as follows.\nmodel {\n  alpha ~ std_normal();         // informative true prior\n  beta ~ std_normal();          // informative true prior\n  delta ~ normal(0.75, 1);      // informative true prior\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);\n  }\n}\nThis model uses the logit-parameterized Bernoulli distribution, where \\[\n\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right)\n=\n\\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\n\\]\nThe key to understanding it is the term inside the bernoulli_logit distribution, from which it follows that \\[\n\\Pr[y_n = 1] = \\operatorname{logit}^{-1}\\left(\\alpha_{jj[n]} - \\beta_{kk[n]}\n+ \\delta\\right).\n\\]\nThe model suffers from additive identifiability issues without the priors. For example, adding a term \\(\\xi\\) to each \\(\\alpha_j\\) and \\(\\beta_k\\) results in the same predictions. The use of priors for \\(\\alpha\\) and \\(\\beta\\) located at 0 identifies the parameters; see Gelman and Hill (2007) for a discussion of identifiability issues and alternative approaches to identification.\nFor testing purposes, the IRT 1PL model distributed with Stan uses informative priors that match the actual data generation process used to simulate the data in R (the simulation code is supplied in the same directory as the models). This is unrealistic for most practical applications, but allows Stan’s inferences to be validated. A simple sensitivity analysis with fatter priors shows that the posterior is fairly sensitive to the prior even with 400 students and 100 questions and only 25% missingness at random. For real applications, the priors should be fit hierarchically along with the other parameters, as described in the next section.\n\n\n\nThe simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.\nparameters {\n  real mu_beta;                // mean question difficulty\n  vector[J] alpha;             // ability for j - mean\n  vector[K] beta;              // difficulty for k\n  vector&lt;lower=0&gt;[K] gamma;    // discrimination of k\n  real&lt;lower=0&gt; sigma_beta;    // scale of difficulties\n  real&lt;lower=0&gt; sigma_gamma;   // scale of log discrimination\n}\nThe parameters should be clearer after the model definition.\nmodel {\n  alpha ~ std_normal();\n  beta ~ normal(0, sigma_beta);\n  gamma ~ lognormal(0, sigma_gamma);\n  mu_beta ~ cauchy(0, 5);\n  sigma_beta ~ cauchy(0, 5);\n  sigma_gamma ~ cauchy(0, 5);\n  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));\n}\nThe std_normal function is used here, defined by \\[\n\\texttt{std}\\mathtt{\\_}\\texttt{normal}(y)\n=\n\\textsf{normal}\\left(y \\mid 0, 1\\right).\n\\]\nThe distribution statement is also vectorized using elementwise multiplication; it is equivalent to\nfor (n in 1:N) {\n  y[n] ~ bernoulli_logit(gamma[kk[n]]\n                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));\n}\nThe 2PL model is similar to the 1PL model, with the additional parameter gamma[k] modeling how discriminative question k is. If gamma[k] is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter gamma[k] is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.\nThe model is parameterized here with student abilities alpha being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the problematic posteriors chapter for further discussion of identifiability. The difficulty and discrimination parameters beta and gamma then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as\nbeta ~ normal(0, 5);\ngamma ~ lognormal(0, 2);\nThe point is that the alpha determines the scale and location and beta and gamma are allowed to float.\nThe beta parameter is here given a non-centered parameterization, with parameter mu_beta serving as the mean beta location. An alternative would’ve been to take:\nbeta ~ normal(mu_beta, sigma_beta);\nand\ny[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));\nNon-centered parameterizations tend to be more efficient in hierarchical models; see the reparameterization section for more information on non-centered reparameterizations.\nThe intercept term mu_beta can’t itself be modeled hierarchically, so it is given a weakly informative \\(\\textsf{Cauchy}(0,5)\\) prior. Similarly, the scale terms, sigma_beta, and sigma_gamma, are given half-Cauchy priors. As mentioned earlier, the scale and location for alpha are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to \\((0,\\infty)\\) by their declarations.",
+    "text": "OpenCL (Open Computing Language) is a framework that enables writing programs that execute across heterogeneous platforms. An OpenCL program can be run on CPUs and GPUs. In order to run OpenCL programs, an OpenCL runtime be installed on the target system.\nStan’s OpenCL backend is currently supported in CmdStan and its wrappers. In order to use it, the model must be compiled with the STAN_OPENCL makefile flag. Setting this flag means that the Stan-to-C++ translator (stanc3) will be supplied the --use-opencl flag and that the OpenCL enabled backend (Stan Math functions) will be enabled.\nIn Stan, the following distributions can be automatically run in parallel on both CPUs and GPUs with OpenCL:\n\nbernoulli_lpmf\nbernoulli_logit_lpmf\nbernoulli_logit_glm_lpmf*\nbeta_lpdf\nbeta_proportion_lpdf\nbinomial_lpmf\ncategorical_logit_glm_lpmf*\ncauchy_lpdf\nchi_square_lpdf\ndouble_exponential_lpdf\nexp_mod_normal_lpdf\nexponential_lpdf\nfrechet_lpdf\ngamma_lpdf\ngumbel_lpdf\ninv_chi_square_lpdf\ninv_gamma_lpdf\nlogistic_lpdf\nlognormal_lpdf\nneg_binomial_lpmf\nneg_binomial_2_lpmf\nneg_binomial_2_log_lpmf\nneg_binomial_2_log_glm_lpmf*\nnormal_lpdf\nnormal_id_glm_lpdf*\nordered_logistic_glm_lpmf*\npareto_lpdf\npareto_type_2_lpdf\npoisson_lpmf\npoisson_log_lpmf\npoisson_log_glm_lpmf*\nrayleigh_lpdf\nscaled_inv_chi_square_lpdf\nskew_normal_lpdf\nstd_normal_lpdf\nstudent_t_lpdf\nuniform_lpdf\nweibull_lpdf\n\n* OpenCL is not used when the covariate argument to the GLM functions is a row_vector.",
+    "crumbs": [
+      "Stan Users Guide",
+      "Programming Techniques",
+      "Parallelization"
+    ]
+  },
+  {
+    "objectID": "stan-users-guide/parallelization.html#footnotes",
+    "href": "stan-users-guide/parallelization.html#footnotes",
+    "title": "Parallelization",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThe term “shard” is borrowed from databases, where it refers to a slice of the rows of a database. That is exactly what it is here if we think of rows of a dataframe. Stan’s shards are more general in that they need not correspond to rows of a dataframe.↩︎\nThis example is a simplified form of the model described in (Gelman and Hill 2007, sec. 14.2)↩︎\nThis makes the strong assumption that each group has the same number of observations!↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Programming Techniques",
+      "Parallelization"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#priors-for-identification.section",
-    "href": "stan-users-guide/regression.html#priors-for-identification.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html",
+    "href": "stan-users-guide/posterior-predictive-checks.html",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "One application of (hierarchical) priors is to identify the scale and/or location of a group of parameters. For example, in the IRT models discussed in the previous section, there is both a location and scale non-identifiability. With uniform priors, the posteriors will float in terms of both scale and location. See the collinearity section for a simple example of the problems this poses for estimation.\nThe non-identifiability is resolved by providing a standard normal (i.e., \\(\\textsf{normal}(0,1)\\)) prior on one group of coefficients, such as the student abilities. With a standard normal prior on the student abilities, the IRT model is identified in that the posterior will produce a group of estimates for student ability parameters that have a sample mean of close to zero and a sample variance of close to one. The difficulty and discrimination parameters for the questions should then be given a diffuse, or ideally a hierarchical prior, which will identify these parameters by scaling and locating relative to the student ability parameters.\n\n\n\nAnother case in which priors can help provide identifiability is in the case of collinearity in a linear regression. In linear regression, if two predictors are collinear (i.e, one is a linear function of the other), then their coefficients will have a correlation of 1 (or -1) in the posterior. This leads to non-identifiability. By placing normal priors on the coefficients, the maximum likelihood solution of two duplicated predictors (trivially collinear) will be half the value than would be obtained by only including one.\n\n\n\nIn a logistic regression, if a predictor is positive in cases of 1 outcomes and negative in cases of 0 outcomes, then the maximum likelihood estimate for the coefficient for that predictor diverges to infinity. This divergence can be controlled by providing a prior for the coefficient, which will “shrink” the estimate back toward zero and thus identify the model in the posterior.\nSimilar problems arise for sampling with improper flat priors. The sampler will try to draw large values. By providing a prior, the posterior will be concentrated around finite values, leading to well-behaved sampling.",
+    "text": "Posterior predictive checks are a way of measuring whether a model does a good job of capturing relevant aspects of the data, such as means, standard deviations, and quantiles (Rubin 1984; Andrew Gelman, Meng, and Stern 1996). Posterior predictive checking works by simulating new replicated data sets based on the fitted model parameters and then comparing statistics applied to the replicated data set with the same statistic applied to the original data set.\nPrior predictive checks evaluate the prior the same way. Specifically, they evaluate what data sets would be consistent with the prior. They will not be calibrated with actual data, but extreme values help diagnose priors that are either too strong, too weak, poorly shaped, or poorly located.\nPrior and posterior predictive checks are two cases of the general concept of predictive checks, just conditioning on different things (no data and the observed data, respectively). For hierarchical models, there are intermediate versions, as discussed in the section on hierarchical models and mixed replication.\n\n\nThe posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data \\(y\\).\nThe posterior predictive distribution for replications \\(y^{\\textrm{rep}}\\) of the original data set \\(y\\) given model parameters \\(\\theta\\) is defined by \\[\np(y^{\\textrm{rep}} \\mid y)\n= \\int p(y^{\\textrm{rep}} \\mid \\theta)\n       \\cdot p(\\theta \\mid y) \\, \\textrm{d}\\theta.\n\\]\nAs with other posterior predictive quantities, generating a replicated data set \\(y^{\\textrm{rep}}\\) from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters \\(\\theta = (\\alpha, \\beta, \\sigma).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  y ~ normal(alpha + beta * x, sigma);\n}\nTo generate a replicated data set y_rep for this simple model, the following generated quantities block suffices.\ngenerated quantities {\n  array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n}\nThe vectorized form of the normal random number generator is used with the original predictors x and the model parameters alpha, beta, and sigma. The replicated data variable y_rep is declared to be the same size as the original data y, but instead of a vector type, it is declared to be an array of reals to match the return type of the function normal_rng. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.\nThe posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in the chapter on posterior predictions in that the original predictors \\(x\\) are used. That is, the posterior predictions are for the original data.\n\n\n\nA standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.\nHere’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\ntransformed data {\n  real&lt;lower=0&gt; mean_y = mean(to_vector(y));\n  real&lt;lower=0&gt; sd_y = sd(to_vector(y));\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  y ~ poisson(lambda);\n  lambda ~ exponential(0.2);\n}\ngenerated quantities {\n  array[N] int&lt;lower=0&gt; y_rep = poisson_rng(rep_array(lambda, N));\n  real&lt;lower=0&gt; mean_y_rep = mean(to_vector(y_rep));\n  real&lt;lower=0&gt; sd_y_rep = sd(to_vector(y_rep));\n  int&lt;lower=0, upper=1&gt; mean_gte = (mean_y_rep &gt;= mean_y);\n  int&lt;lower=0, upper=1&gt; sd_gte = (sd_y_rep &gt;= sd_y);\n}\nThe generated quantities block creates a variable y_rep for the replicated data, variables mean_y_rep and sd_y_rep for the statistics of the replicated data, and indicator variables mean_gte and sd_gte for whether the replicated statistic is greater than or equal to the statistic applied to the original data.\nNow consider generating data \\(y \\sim \\textrm{Poisson}(5)\\). The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.\n\n\n\nPosterior predictive checks for Poisson data generating process and Poisson model.\n\n\nWith a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.\nNow consider generating over-dispersed data \\(y \\sim \\textrm{negative-binomial2}(5, 1).\\) This has the same mean as \\(\\textrm{Poisson}(5)\\), namely \\(5\\), but a standard deviation of \\(\\sqrt{5 + 5^2 /1} \\approx 5.5.\\) There is no way to fit this data with the Poisson model, because a variable distributed as \\(\\textrm{Poisson}(\\lambda)\\) has mean \\(\\lambda\\) and standard deviation \\(\\sqrt{\\lambda},\\) which is \\(\\sqrt{5}\\) for \\(\\textrm{Poisson}(5).\\) Here’s the resulting small multiples plot, again with original data in the upper left.\n\n\n\nPosterior predictive checks for negative binomial data generating process and Poisson model.\n\n\nThis time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.\n\n\n\nIf a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic \\(s(\\cdot)\\) in a replicated data set exceeds that in the original data, \\[\n\\Pr\\!\\left[ s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right]\n=\n\\int\n\\textrm{I}\\left( s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right)\n\\cdot p\\left( y^{\\textrm{rep}} \\mid y \\right)\n\\, \\textrm{d}{y^{\\textrm{rep}}}.\n\\] It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified (Bayarri and Berger 2000).\nNevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.\nTo calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mean_gt;\n  int&lt;lower=0, upper=1&gt; sd_gt;\n  {\n    array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n    mean_gt = mean(y_rep) &gt; mean(y);\n    sd_gt = sd(y_rep) &gt; sd(y);\n  }\n}\nThe indicator variable mean_gt will have value 1 if the mean of the simulated data y_rep is greater than or equal to the mean of he original data y. Because the values of y_rep are not needed for the posterior predictive checks, the program saves output space by using a local variable for y_rep. The statistics mean(u) and sd(y) could also be computed in the transformed data block and saved.\nFor the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.\n\n\n\nHistogram of means of replicated data sets; vertical red line at mean of original data.\n\n\nThe p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior \\(p\\)-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable mean_gt.\nThe standard deviation statistic tells a different story.\n\n\n\nScatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.\n\n\nHere, the original data has much higher standard deviation than any of the replicated data sets. The resulting \\(p\\)-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small \\(p\\)-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.\n\n\nAny statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.\nDespite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.\n\n\n\n\nPrior predictive checks generate data according to the prior in order to asses whether a prior is appropriate (Gabry et al. 2019). A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, \\[\ny^{\\textrm{sim}} \\sim p(y).\n\\] The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.\nThis is easy to carry out mechanically by simulating parameters \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta)\n\\] according to the priors, then simulating data \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}})\n\\] according to the data model given the simulated parameters. The result is a simulation from the joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta)\n\\] and thus \\[\ny^{\\textrm{sim}} \\sim p(y)\n\\] is a simulation from the prior predictive distribution.\n\n\nA prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.\nPrior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n}\ngenerated quantities {\n  real alpha = normal_rng(0, 1);\n  real beta = normal_rng(0, 1);\n  array[N] real y_sim = poisson_log_rng(alpha + beta * x);\n}\nRunning this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.\n\n\n\n\nSuppose we have a model for a football (aka soccer) league where there are \\(J\\) teams. Each team has a scoring rate \\(\\lambda_j\\) and in each game will be assumed to score \\(\\textrm{poisson}(\\lambda_j)\\) points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed \\[\n\\lambda_j \\sim \\textrm{gamma}(\\epsilon_1, \\epsilon_2).\n\\] This is not just a manufactured example; The BUGS Book recommends setting \\(\\epsilon = (0.5, 0.00001)\\), which corresponds to a Jeffreys prior for a Poisson rate parameter prior (Lunn et al. 2012, 85).\nSuppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.\ndata {\n  int&lt;lower=0&gt; J;\n  array[2] real&lt;lower=0&gt; epsilon;\n}\ngenerated quantities {\n  array[J] real&lt;lower=0&gt; lambda;\n  array[J, J] int y;\n  for (j in 1:J) lambda[j] = gamma_rng(epsilon[1], epsilon[2]);\n  for (i in 1:J) {\n    for (j in 1:J) {\n      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);\n    }\n  }\n}\nIn this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are \\(\\binom{J}{2} = \\frac{J \\cdot (J - 1)}{2}.\\)\nUsing the \\(\\textrm{gamma}(0.5, 0.00001)\\) reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, \\(y^{(1:20)}_{1, 2}\\).\n2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987\n7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66\nThat’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this \\(\\textrm{gamma}(0.5, 0.00001)\\) prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.\nThe posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values (A. Gelman 2006).\nJust as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.\nIn this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see (Gabry et al. 2019).\nPrior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.\n\n\n\nAndrew Gelman, Meng, and Stern (1996) discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.\nFor example, consider a simple varying intercept logistic regression, with intercepts \\(\\alpha_k\\) for \\(k \\in 1:K\\). Each data item \\(y_n \\in \\{ 0, 1 \\}\\) is assumed to correspond to group \\(kk_n \\in 1:K.\\) The data model is thus \\[\ny_n \\sim \\textrm{bernoulli}(\\textrm{logit}^{-1}(\\alpha_{kk[n]})).\n\\] The varying intercepts have a hierarchical normal prior, \\[\n\\alpha_k \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The hyperparameters are themselves given weakly informative priors, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 2)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\]\nLike in a posterior predictive check, the hyperparameters \\(\\mu\\) and \\(\\sigma\\) are drawn from the posterior, \\[\n\\mu^{(m)}, \\sigma^{(m)} \\sim p(\\mu, \\sigma \\mid y)\n\\] Like in a prior predictive check, replicated values of \\(\\alpha\\) are drawn from the hyperparameters, \\[\n\\alpha^{\\textrm{rep}(m)}_k \\sim \\textrm{normal}(\\alpha_k \\mid\n\\mu^{(m)}, \\sigma^{(m)}).\n\\] The data items are then each replicated using the replicated intercepts, \\[\ny^{\\textrm{rep}(m)}_n \\sim\n\\textrm{bernoulli}\n  (\\textrm{logit}^{-1}(\\alpha^{\\textrm{rep}(m)}_{kk[n]})).\n\\] Thus the \\(y^{\\textrm{rep}(m)}\\) can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original \\(K\\) groups.\nIn Stan, mixed predictive replications \\(y^{\\textrm{rep}(m)}\\) can be programmed directly.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;offset=mu, multiplier=sigma&gt;[K] alpha;\n}\nmodel {\n  mu ~ normal(0, 2);               // hyperprior\n  sigma ~ lognormal(0, 1);\n  alpha ~ normal(mu, sigma);       // hierarchical prior\n  y ~ bernoulli_logit(alpha[kk]);  // data model\n}\ngenerated quantities {\n  // alpha replicated;  mu and sigma not replicated\n  array[K] real alpha_rep\n    = normal_rng(rep_vector(mu, K), sigma);\n  array[N] int&lt;lower=0, upper=1&gt; y_rep\n    = bernoulli_logit_rng(alpha_rep[kk]);\n}\n\n\n\nFollowing Andrew Gelman, Meng, and Stern (1996), prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.\n\n\nFor example, posterior predictive replication may be formulated using distribution notation as follows. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] The heavily overloaded distribution notation is meant to indicate that both \\(y\\) and \\(y^{\\textrm{rep}}\\) are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities \\(p_{Y^{\\textrm{rep}} \\mid\n\\Theta}\\) and \\(p_{Y \\mid \\Theta}\\) are the same.\nThe joint density is \\[\np(\\theta, y, y^{\\textrm{rep}})\n= p(\\theta) \\cdot p(y \\mid \\theta) \\cdot p(y^{\\textrm{rep}} \\mid \\theta).\n\\] This again is assuming that the two distributions for \\(y\\) and \\(y^{\\textrm{rep}}\\) are identical.\nThe variable \\(y\\) is observed, with the predictive simulation \\(y^{\\textrm{rep}}\\) and parameter vector \\(\\theta\\) not observed. The posterior is \\(p(y^{\\textrm{rep}}, \\theta \\mid y)\\). Given draws from the posterior, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are retained.\n\n\n\nThe prior predictive model simply drops the data component of the posterior predictive model. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] This corresponds to the joint density \\[\np(\\theta, y^{\\textrm{rep}}) = p(\\theta) \\cdot p(y^{\\textrm{rep}} \\mid\n\\theta).\n\\]\nIt is typically straightforward to draw \\(\\theta\\) from the prior and \\(y^{\\textrm{rep}}\\) from the data model given \\(\\theta\\) efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.\n\n\n\nThe mixed replication corresponds to the model \\[\\begin{eqnarray*}\n\\phi & \\sim & p(\\phi)\n\\\\[2pt]\n\\alpha & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\alpha)\n\\\\[2pt]\n\\alpha^{\\textrm{rep}} & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\phi)\n\\end{eqnarray*}\\] The notation here is meant to indicate that \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\) have identical distributions, as do \\(y\\) and \\(y^{\\textrm{rep}}\\).\nThis corresponds to a joint model \\[\np(\\phi, \\alpha, \\alpha^{\\textrm{rep}}, y, y^{\\textrm{rep}})\n=\np(\\phi)\n\\cdot p(\\alpha \\mid \\phi)\n\\cdot p(y \\mid \\alpha)\n\\cdot p(\\alpha^{\\textrm{rep}} \\mid \\phi)\n\\cdot p(y^{\\textrm{rep}} \\mid \\alpha^{\\textrm{rep}}),\n\\] where \\(y\\) is the only observed variable, \\(\\alpha\\) contains the lower-level parameters and \\(\\phi\\) the hyperparameters. Note that \\(\\phi\\) is not replicated and instead appears in the distribution for both \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\).\nThe posterior is \\(p(\\phi, \\alpha, \\alpha^{\\textrm{rep}},\ny^{\\textrm{rep}} \\mid y)\\). From posterior draws, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are kept.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#multivariate-hierarchical-priors.section",
-    "href": "stan-users-guide/regression.html#multivariate-hierarchical-priors.section",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#simulating-from-the-posterior-predictive-distribution",
+    "href": "stan-users-guide/posterior-predictive-checks.html#simulating-from-the-posterior-predictive-distribution",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "In hierarchical regression models (and other situations), several individual-level variables may be assigned hierarchical priors. For example, a model with multiple varying intercepts and slopes within might assign them a multivariate prior.\nAs an example, the individuals might be people and the outcome income, with predictors such as education level and age, and the groups might be states or other geographic divisions. The effect of education level and age as well as an intercept might be allowed to vary by state. Furthermore, there might be state-level predictors, such as average state income and unemployment level.\n\n\nGelman and Hill (2007, chap. 13, Chapter 17) provide a discussion of a hierarchical model with \\(N\\) individuals organized into \\(J\\) groups. Each individual has a predictor row vector \\(x_n\\) of size \\(K\\); to unify the notation, they assume that \\(x_{n,1} = 1\\) is a fixed “intercept” predictor. To encode group membership, they assume individual \\(n\\) belongs to group \\(jj[n] \\in \\{ 1, \\dotsc, J \\}\\). Each individual \\(n\\) also has an observed outcome \\(y_n\\) taking on real values.\n\n\nThe model is a linear regression with slope and intercept coefficients varying by group, so that \\(\\beta_j\\) is the coefficient \\(K\\)-vector for group \\(j\\). The data model for individual \\(n\\) is then just \\[\ny_n \\sim \\textsf{normal}(x_n \\, \\beta_{jj[n]}, \\, \\sigma)\n\\quad\\text{for}\\quad n \\in \\{ 1, \\dotsc, N \\}.\n\\]\n\n\n\nGelman and Hill model the coefficient vectors \\(\\beta_j\\) as being drawn from a multivariate distribution with mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\), \\[\n\\beta_j \\sim \\textsf{multivariate normal}(\\mu_j, \\, \\Sigma)\n\\quad\\text{for}\\quad j \\in \\{ 1, \\dotsc, J \\}.\n\\]\nBelow, we discuss the full model of Gelman and Hill, which uses group-level predictors to model \\(\\mu\\); for now, we assume \\(\\mu\\) is a simple vector parameter.\n\n\n\nFor hierarchical modeling, the group-level mean vector \\(\\mu\\) and covariance matrix \\(\\Sigma\\) must themselves be given priors. The group-level mean vector can be given a reasonable weakly-informative prior for independent coefficients, such as \\[\n\\mu_j \\sim \\textsf{normal}(0,5).\n\\] If more is known about the expected coefficient values \\(\\beta_{j, k}\\), this information can be incorporated into the prior for \\(\\mu_j\\).\nFor the prior on the covariance matrix, Gelman and Hill suggest using a scaled inverse Wishart. That choice was motivated primarily by convenience as it is conjugate to the multivariate likelihood function and thus simplifies Gibbs sampling\nIn Stan, there is no restriction to conjugacy for multivariate priors, and we in fact recommend a slightly different approach. Like Gelman and Hill, we decompose our prior into a scale and a matrix, but are able to do so in a more natural way based on the actual variable scales and a correlation matrix. Specifically, we define \\[\n\\Sigma = \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau) \\times \\Omega \\times \\texttt{diag}\\mathtt{\\_}\\texttt{matrix}(\\tau),\n\\] where \\(\\Omega\\) is a correlation matrix and \\(\\tau\\) is the vector of coefficient scales. This mapping from scale vector \\(\\tau\\) and correlation matrix \\(\\Omega\\) can be inverted, using \\[\n\\tau_k = \\sqrt{\\Sigma_{k,k}}\n\\quad\\textsf{and}\\quad\n\\Omega_{i, j} = \\frac{\\Sigma_{i, j}}{\\tau_i \\, \\tau_j}.\n\\]\nThe components of the scale vector \\(\\tau\\) can be given any reasonable prior for scales, but we recommend something weakly informative like a half-Cauchy distribution with a small scale, such as \\[\n\\tau_k \\sim \\textsf{Cauchy}(0, 2.5)\n\\quad\\text{for}\\quad k \\in \\{ 1, \\dotsc, K \\}\n\\quad\\text{constrained\\ by}\\quad \\tau_k &gt; 0.\n\\] As for the prior means, if there is information about the scale of variation of coefficients across groups, it should be incorporated into the prior for \\(\\tau\\). For large numbers of exchangeable coefficients, the components of \\(\\tau\\) itself (perhaps excluding the intercept) may themselves be given a hierarchical prior.\nOur final recommendation is to give the correlation matrix \\(\\Omega\\) an LKJ prior with shape \\(\\eta \\geq 1\\),5\n\\[\n\\Omega \\sim \\textsf{LKJCorr}(\\eta).\n\\]\nThe LKJ correlation distribution is defined by \\[\n\\textsf{LKJCorr}\\left(\\Sigma \\mid \\eta\\right)\n\\propto\n\\operatorname{det}\\left(\\Sigma\\right)^{\\eta - 1}.\n\\]\nThe basic behavior of the LKJ correlation distribution is similar to that of a beta distribution. For \\(\\eta = 1\\), the result is a uniform distribution. Despite being the identity over correlation matrices, the marginal distribution over the entries in that matrix (i.e., the correlations) is not uniform between -1 and 1. Rather, it concentrates around zero as the dimensionality increases due to the complex constraints.\nFor \\(\\eta &gt; 1\\), the density increasingly concentrates mass around the unit matrix, i.e., favoring less correlation. For \\(\\eta &lt; 1\\), it increasingly concentrates mass in the other direction, i.e., favoring more correlation.\nThe LKJ prior may thus be used to control the expected amount of correlation among the parameters \\(\\beta_j\\). For a discussion of decomposing a covariance prior into a prior on correlation matrices and an independent prior on scales, see Barnard, McCulloch, and Meng (2000).\n\n\n\nTo complete Gelman and Hill’s model, suppose each group \\(j \\in \\{ 1, \\dotsc, J \\}\\) is supplied with an \\(L\\)-dimensional row-vector of group-level predictors \\(u_j\\). The prior mean for the \\(\\beta_j\\) can then itself be modeled as a regression, using an \\(L\\)-dimensional coefficient vector \\(\\gamma\\). The prior for the group-level coefficients then becomes \\[\n\\beta_j \\sim \\textsf{multivariate normal}(u_j \\, \\gamma, \\Sigma)\n\\]\nThe group-level coefficients \\(\\gamma\\) may themselves be given independent weakly informative priors, such as \\[\n\\gamma_l \\sim \\textsf{normal}(0,5).\n\\] As usual, information about the group-level means should be incorporated into this prior.\n\n\n\nThe Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.\ndata {\n  int&lt;lower=0&gt; N;              // num individuals\n  int&lt;lower=1&gt; K;              // num ind predictors\n  int&lt;lower=1&gt; J;              // num groups\n  int&lt;lower=1&gt; L;              // num group predictors\n  array[N] int&lt;lower=1, upper=J&gt; jj;  // group for individual\n  matrix[N, K] x;              // individual predictors\n  array[J] row_vector[L] u;    // group predictors\n  vector[N] y;                 // outcomes\n}\nparameters {\n  corr_matrix[K] Omega;        // prior correlation\n  vector&lt;lower=0&gt;[K] tau;      // prior scale\n  matrix[L, K] gamma;          // group coeffs\n  array[J] vector[K] beta;     // indiv coeffs by group\n  real&lt;lower=0&gt; sigma;         // prediction error scale\n}\nmodel {\n  tau ~ cauchy(0, 2.5);\n  Omega ~ lkj_corr(2);\n  to_vector(gamma) ~ normal(0, 5);\n  {\n    array[J] row_vector[K] u_gamma;\n    for (j in 1:J) {\n      u_gamma[j] = u[j] * gamma;\n    }\n    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));\n  }\n  for (n in 1:N) {\n    y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n  }\n}\nThe hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix Omega and scale vector tau are more natural to inspect in the output; to output Sigma, define it as a transformed parameter. The function quad_form_diag is defined so that quad_form_diag(Sigma, tau) is equivalent to diag_matrix(tau) * Sigma * diag_matrix(tau), where diag_matrix(tau) returns the matrix with tau on the diagonal and zeroes off diagonal; the version using quad_form_diag should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.\n\n\n\nThe code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:\nfor (n in 1:N) {\n  y[n] ~ normal(x[n] * beta[jj[n]], sigma);\n}\nwith the vectorized distribution statement:\n{\n  vector[N] x_beta_jj;\n  for (n in 1:N) {\n    x_beta_jj[n] = x[n] * beta[jj[n]];\n  }\n  y ~ normal(x_beta_jj, sigma);\n}\nThe outer brackets create a local scope in which to define the variable x_beta_jj, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.\nAlthough it is tempting to redeclare beta and include a revised model block distribution statement,\nparameters {\n  matrix[J, K] beta;\n// ...\n}\nmodel {\n  y ~ normal(rows_dot_product(x, beta[jj]), sigma);\n  // ...\n}\nthis fails because it breaks the vectorization for beta,6\nbeta ~ multi_normal(...);\nwhich requires beta to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because rows_dot_product cannot do much optimization in and of itself because there are no shared computations.\nThe code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.\n{\n  matrix[K, K] Sigma_beta;\n  Sigma_beta = quad_form_diag(Omega, tau);\n  for (j in 1:J) {\n    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);\n  }\n}\nIn this example, the covariance matrix Sigma_beta is defined as a local variable so as not to have to repeat the quadratic form computation \\(J\\) times. This vectorization can be combined with the Cholesky-factor optimization in the next section.\n\n\n\nThe multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.\ndata {\n  matrix[L, J] u;              // group predictors transposed\n  // ...\n}\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  matrix[K, L] gamma;\n  // ...\n}\ntransformed parameters {\n  matrix[K, J] beta;\n  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  // ...\n}\nThe data variable u was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with gamma and beta, to minimize the number of transposition operations. The new parameter L_Omega is the Cholesky factor of the original correlation matrix Omega, so that\nOmega = L_Omega * L_Omega'\nThe prior scale vector tau is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,\nSigma_beta\n  = quad_form_diag(Omega, tau)\n  = diag_pre_multiply(tau, L_Omega) * diag_pre_multiply(tau, L_Omega)'\nwhere the diagonal pre-multiply compound operation is defined by\ndiag_pre_multiply(a, b) = diag_matrix(a) * b\nThe new variable z is declared as a matrix, the entries of which are given independent standard normal priors; the to_vector operation turns the matrix into a vector so that it can be used as a vectorized argument to the univariate normal density. This results in every column of z being a \\(K\\)-variate normal random vector with the identity as covariance matrix. Therefore, multiplying z by the Cholesky factor of the covariance matrix and adding the mean (u * gamma)' produces a beta distributed as in the original model, where the variance is, letting \\(L = \\mathrm{diag}(\\tau)\\,\\Omega_L\\),\n\\[\n\\begin{aligned}\n\\mathbb{V}[\\beta] &= \\mathbb{E}\\big((L \\, z)(L \\, z)^\\top) \\\\\n&= \\mathbb{E}\\big((L \\, z \\, z^\\top \\, L^\\top) \\\\\n&= L \\, \\mathbb{E}(z \\, z^\\top) \\, L^\\top \\\\\n&= L \\, L^\\top =(\\mathrm{diag}(\\tau)\\,\\Omega_L)\\,(\\mathrm{diag}(\\tau)\\,\\Omega_L)^\\top \\\\\n&= \\mathrm{diag}(\\tau)\\,\\Omega\\,\\mathrm{diag}(\\tau) \\\\\n&= \\Sigma.\n\\end{aligned}\n\\] Where we have used the linearity of expectations (line 2 to 3), the definition of \\(\\Omega = \\Omega_L \\, \\Omega_L^\\top\\), and the fact that \\(\\mathbb{E}(z \\, z^\\top) = I\\) since \\(z \\sim \\mathcal{N}(0, I)\\).\nOmitting the remaining data declarations, which are the same as before with the exception of u, the optimized model is as follows.\nparameters {\n  matrix[K, J] z;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0, upper=pi() / 2&gt;[K] tau_unif;  // prior scale\n  matrix[K, L] gamma;                        // group coeffs\n  real&lt;lower=0&gt; sigma;                       // prediction error scale\n}\ntransformed parameters {\n  vector&lt;lower=0&gt;[K] tau = 2.5 * tan(tau_unif);\n  matrix[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;\n}\nmodel {\n  vector[N] mu;\n  for(n in 1:N) {\n    mu[n] = x[n, ] * beta[, jj[n]];\n  }\n  to_vector(z) ~ std_normal();\n  L_Omega ~ lkj_corr_cholesky(2);\n  to_vector(gamma) ~ normal(0, 5);\n  y ~ normal(mu, sigma);\n}\nThis model also reparameterizes the prior scale tau to avoid potential problems with the heavy tails of the Cauchy distribution. The statement tau_unif ~ uniform(0, pi() / 2) can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.",
+    "text": "The posterior predictive distribution is the distribution over new observations given previous observations. It’s predictive in the sense that it’s predicting behavior on new data that is not part of the training set. It’s posterior in that everything is conditioned on observed data \\(y\\).\nThe posterior predictive distribution for replications \\(y^{\\textrm{rep}}\\) of the original data set \\(y\\) given model parameters \\(\\theta\\) is defined by \\[\np(y^{\\textrm{rep}} \\mid y)\n= \\int p(y^{\\textrm{rep}} \\mid \\theta)\n       \\cdot p(\\theta \\mid y) \\, \\textrm{d}\\theta.\n\\]\nAs with other posterior predictive quantities, generating a replicated data set \\(y^{\\textrm{rep}}\\) from the posterior predictive distribution is straightforward using the generated quantities block. Consider a simple regression model with parameters \\(\\theta = (\\alpha, \\beta, \\sigma).\\)\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  alpha ~ normal(0, 2);\n  beta ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  y ~ normal(alpha + beta * x, sigma);\n}\nTo generate a replicated data set y_rep for this simple model, the following generated quantities block suffices.\ngenerated quantities {\n  array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n}\nThe vectorized form of the normal random number generator is used with the original predictors x and the model parameters alpha, beta, and sigma. The replicated data variable y_rep is declared to be the same size as the original data y, but instead of a vector type, it is declared to be an array of reals to match the return type of the function normal_rng. Because the vector and real array types have the same dimensions and layout, they can be plotted against one another and otherwise compared during downstream processing.\nThe posterior predictive sampling for posterior predictive checks is different from usual posterior predictive sampling discussed in the chapter on posterior predictions in that the original predictors \\(x\\) are used. That is, the posterior predictions are for the original data.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#prediction-forecasting-and-backcasting",
-    "href": "stan-users-guide/regression.html#prediction-forecasting-and-backcasting",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#plotting-multiples",
+    "href": "stan-users-guide/posterior-predictive-checks.html#plotting-multiples",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "Stan models can be used for “predicting” the values of arbitrary model unknowns. When predictions are about the future, they’re called “forecasts;” when they are predictions about the past, as in climate reconstruction or cosmology, they are sometimes called “backcasts” (or “aftcasts” or “hindcasts” or “antecasts,” depending on the author’s feelings about the opposite of “fore”).\n\n\nAs a simple example, the following linear regression provides the same setup for estimating the coefficients beta as in our very first example, using y for the N observations and x for the N predictor vectors. The model parameters and model for observations are exactly the same as before.\nTo make predictions, we need to be given the number of predictions, N_new, and their predictor matrix, x_new. The predictions themselves are modeled as a parameter y_new. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector y_new and prediction matrix x_new.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=0&gt; N;\n  matrix[N, K] x;\n  vector[N] y;\n\n  int&lt;lower=0&gt; N_new;\n  matrix[N_new, K] x_new;\n}\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n\n  vector[N_new] y_new;                  // predictions\n}\nmodel {\n  y ~ normal(x * beta, sigma);          // observed model\n\n  y_new ~ normal(x_new * beta, sigma);  // prediction model\n}\n\n\n\nWhere possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.\n// ...data as above...\n\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(x * beta, sigma);\n}\ngenerated quantities {\n  vector[N_new] y_new;\n  for (n in 1:N_new) {\n    y_new[n] = normal_rng(x_new[n] * beta, sigma);\n  }\n}\nNow the data are just as before, but the parameter y_new is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.\n\n\nIt is possible for values to overflow or underflow in generated quantities. The problem is that if the result is NaN, then any constraints placed on the variables will be violated. It is possible to check a value assigned by an RNG and reject it if it overflows, but this is both inefficient and leads to biased posterior estimates. Instead, the conditions causing overflow, such as trying to generate a negative binomial random variate with a mean of \\(2^{31}\\), must be intercepted and dealt with. This is typically done by reparameterizing or reimplementing the random number generator using real values rather than integers, which are upper-bounded by \\(2^{31} - 1\\) in Stan.",
+    "text": "A standard posterior predictive check would plot a histogram of each replicated data set along with the original data set and compare them by eye. For this purpose, only a few replications are needed. These should be taken by thinning a larger set of replications down to the size needed to ensure rough independence of the replications.\nHere’s a complete example where the model is a simple Poisson with a weakly informative exponential prior with a mean of 10 and standard deviation of 10.\ndata {\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=0&gt; y;\n}\ntransformed data {\n  real&lt;lower=0&gt; mean_y = mean(to_vector(y));\n  real&lt;lower=0&gt; sd_y = sd(to_vector(y));\n}\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nmodel {\n  y ~ poisson(lambda);\n  lambda ~ exponential(0.2);\n}\ngenerated quantities {\n  array[N] int&lt;lower=0&gt; y_rep = poisson_rng(rep_array(lambda, N));\n  real&lt;lower=0&gt; mean_y_rep = mean(to_vector(y_rep));\n  real&lt;lower=0&gt; sd_y_rep = sd(to_vector(y_rep));\n  int&lt;lower=0, upper=1&gt; mean_gte = (mean_y_rep &gt;= mean_y);\n  int&lt;lower=0, upper=1&gt; sd_gte = (sd_y_rep &gt;= sd_y);\n}\nThe generated quantities block creates a variable y_rep for the replicated data, variables mean_y_rep and sd_y_rep for the statistics of the replicated data, and indicator variables mean_gte and sd_gte for whether the replicated statistic is greater than or equal to the statistic applied to the original data.\nNow consider generating data \\(y \\sim \\textrm{Poisson}(5)\\). The resulting small multiples plot shows the original data plotted in the upper left and eight different posterior replications plotted in the remaining boxes.\n\n\n\nPosterior predictive checks for Poisson data generating process and Poisson model.\n\n\nWith a Poisson data-generating process and Poisson model, the posterior replications look similar to the original data. If it were easy to pick the original data out of the lineup, there would be a problem.\nNow consider generating over-dispersed data \\(y \\sim \\textrm{negative-binomial2}(5, 1).\\) This has the same mean as \\(\\textrm{Poisson}(5)\\), namely \\(5\\), but a standard deviation of \\(\\sqrt{5 + 5^2 /1} \\approx 5.5.\\) There is no way to fit this data with the Poisson model, because a variable distributed as \\(\\textrm{Poisson}(\\lambda)\\) has mean \\(\\lambda\\) and standard deviation \\(\\sqrt{\\lambda},\\) which is \\(\\sqrt{5}\\) for \\(\\textrm{Poisson}(5).\\) Here’s the resulting small multiples plot, again with original data in the upper left.\n\n\n\nPosterior predictive checks for negative binomial data generating process and Poisson model.\n\n\nThis time, the original data stands out in stark contrast to the replicated data sets, all of which are clearly more symmetric and lower variance than the original data. That is, the model’s not appropriately capturing the variance of the data.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#multivariate-outcomes",
-    "href": "stan-users-guide/regression.html#multivariate-outcomes",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#posterior-p-values",
+    "href": "stan-users-guide/posterior-predictive-checks.html#posterior-p-values",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "Most regressions are set up to model univariate observations (be they scalar, boolean, categorical, ordinal, or count). Even multinomial regressions are just repeated categorical regressions. In contrast, this section discusses regression when each observed value is multivariate. To relate multiple outcomes in a regression setting, their error terms are provided with covariance structure.\nThis section considers two cases, seemingly unrelated regressions for continuous multivariate quantities and multivariate probit regression for boolean multivariate quantities.\n\n\nThe first model considered is the “seemingly unrelated” regressions (SUR) of econometrics where several linear regressions share predictors and use a covariance error structure rather than independent errors (Zellner 1962; Greene 2011).\nThe model is easy to write down as a regression, \\[\\begin{align*}\ny_n        &=  x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nwhere \\(x_n\\) is a \\(J\\)-row-vector of predictors (\\(x\\) is an \\((N \\times\nJ)\\) matrix), \\(y_n\\) is a \\(K\\)-vector of observations, \\(\\beta\\) is a \\((K\n\\times J)\\) matrix of regression coefficients (vector \\(\\beta_k\\) holds coefficients for outcome \\(k\\)), and \\(\\Sigma\\) is covariance matrix governing the error. As usual, the intercept can be rolled into \\(x\\) as a column of ones.\nThe basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; J;\n  int&lt;lower=0&gt; N;\n  array[N] vector[J] x;\n  array[N] vector[K] y;\n}\nparameters {\n  matrix[K, J] beta;\n  cov_matrix[K] Sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n  }\n  y ~ multi_normal(mu, Sigma);\n}\nFor efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.\nFollowing the advice in the multivariate hierarchical priors section, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.\n// ...\nparameters {\n  matrix[K, J] beta;\n  cholesky_factor_corr[K] L_Omega;\n  vector&lt;lower=0&gt;[K] L_sigma;\n}\nmodel {\n  array[N] vector[K] mu;\n  matrix[K, K] L_Sigma;\n\n  for (n in 1:N) {\n    mu[n] = beta * x[n];\n\n  }\n\n  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);\n\n  to_vector(beta) ~ normal(0, 5);\n  L_Omega ~ lkj_corr_cholesky(4);\n  L_sigma ~ cauchy(0, 2.5);\n\n  y ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix beta to a vector.\nIf required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.\n\n\n\nThe multivariate probit model generates sequences of boolean variables by applying a step function to the output of a seemingly unrelated regression.\nThe observations \\(y_n\\) are \\(D\\)-vectors of boolean values (coded 0 for false, 1 for true). The values for the observations \\(y_n\\) are based on latent values \\(z_n\\) drawn from a seemingly unrelated regression model (see the previous section), \\[\\begin{align*}\nz_n        &=     x_n \\, \\beta + \\epsilon_n \\\\\n\\epsilon_n &\\sim  \\textsf{multivariate normal}(0, \\Sigma)\n\\end{align*}\\]\nThese are then put through the step function to produce a \\(K\\)-vector \\(z_n\\) of boolean values with elements defined by \\[\ny_{n, k} = \\operatorname{I}\\left(z_{n, k} &gt; 0\\right),\n\\] where \\(\\operatorname{I}()\\) is the indicator function taking the value 1 if its argument is true and 0 otherwise.\nUnlike in the seemingly unrelated regressions case, here the covariance matrix \\(\\Sigma\\) has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see Greene (2011)).\nMultivariate probit regression can be coded in Stan using the trick introduced by Albert and Chib (1993), where the underlying continuous value vectors \\(y_n\\) are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector \\(z\\) in two parts, based on whether the corresponding value of \\(y\\) is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.\nFirst, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in \\(y\\).\nfunctions {\n  int sum2d(array[,] int a) {\n    int s = 0;\n    for (i in 1:size(a)) {\n      s += sum(a[i]);\n    }\n    return s;\n  }\n}\nThe function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.\nThe data declaration block is much like for the seemingly unrelated regressions, but the observations y are now integers constrained to be 0 or 1.\ndata {\n  int&lt;lower=1&gt; K;\n  int&lt;lower=1&gt; D;\n  int&lt;lower=0&gt; N;\n  array[N, D] int&lt;lower=0, upper=1&gt; y;\n  array[N] vector[K] x;\n}\nAfter declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array y into positive and negative components, keeping track of indexes so that z can be easily reassembled in the transformed parameters block.\ntransformed data {\n  int&lt;lower=0&gt; N_pos;\n  array[sum2d(y)] int&lt;lower=1, upper=N&gt; n_pos;\n  array[size(n_pos)] int&lt;lower=1, upper=D&gt; d_pos;\n  int&lt;lower=0&gt; N_neg;\n  array[(N * D) - size(n_pos)] int&lt;lower=1, upper=N&gt; n_neg;\n  array[size(n_neg)] int&lt;lower=1, upper=D&gt; d_neg;\n\n  N_pos = size(n_pos);\n  N_neg = size(n_neg);\n  {\n    int i;\n    int j;\n    i = 1;\n    j = 1;\n    for (n in 1:N) {\n      for (d in 1:D) {\n        if (y[n, d] == 1) {\n          n_pos[i] = n;\n          d_pos[i] = d;\n          i += 1;\n        } else {\n          n_neg[j] = n;\n          d_neg[j] = d;\n          j += 1;\n        }\n      }\n    }\n  }\n}\nThe variables N_pos and N_neg are set to the number of true (1) and number of false (0) observations in y. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.\nThe parameters are declared as follows.\nparameters {\n  matrix[D, K] beta;\n  cholesky_factor_corr[D] L_Omega;\n  vector&lt;lower=0&gt;[N_pos] z_pos;\n  vector&lt;upper=0&gt;[N_neg] z_neg;\n}\nThese include the regression coefficients beta and the Cholesky factor of the correlation matrix, L_Omega. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).\nThe critical part of the parameter declaration is that the latent real value \\(z\\) is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct \\(z\\).\ntransformed parameters {\n  array[N] vector[D] z;\n  for (n in 1:N_pos) {\n    z[n_pos[n], d_pos[n]] = z_pos[n];\n  }\n  for (n in 1:N_neg) {\n    z[n_neg[n], d_neg[n]] = z_neg[n];\n  }\n}\nAt this point, the model is simple, pretty much recreating the seemingly unrelated regression.\nmodel {\n  L_Omega ~ lkj_corr_cholesky(4);\n  to_vector(beta) ~ normal(0, 5);\n  {\n    array[N] vector[D] beta_x;\n    for (n in 1:N) {\n      beta_x[n] = beta * x[n];\n    }\n    z ~ multi_normal_cholesky(beta_x, L_Omega);\n  }\n}\nThis simple form of model is made possible by the Albert and Chib-style constraints on z.\nFinally, the correlation matrix itself can be put back together in the generated quantities block if desired.\ngenerated quantities {\n  corr_matrix[D] Omega;\n  Omega = multiply_lower_tri_self_transpose(L_Omega);\n}\nThe same could be done for the seemingly unrelated regressions in the previous section.",
+    "text": "If a model captures the data well, summary statistics such as sample mean and standard deviation, should have similar values in the original and replicated data sets. This can be tested by means of a p-value-like statistic, which here is just the probability the test statistic \\(s(\\cdot)\\) in a replicated data set exceeds that in the original data, \\[\n\\Pr\\!\\left[ s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right]\n=\n\\int\n\\textrm{I}\\left( s(y^{\\textrm{rep}}) \\geq s(y) \\mid y \\right)\n\\cdot p\\left( y^{\\textrm{rep}} \\mid y \\right)\n\\, \\textrm{d}{y^{\\textrm{rep}}}.\n\\] It is important to note that ‘’p-values’’ is in quotes because these statistics are not classically calibrated, and thus will not in general have a uniform distribution even when the model is well specified (Bayarri and Berger 2000).\nNevertheless, values of this statistic very close to zero or one are cause for concern that the model is not fitting the data well. Unlike a visual test, this p-value-like test is easily automated for bulk model fitting.\nTo calculate event probabilities in Stan, it suffices to define indicator variables that take on value 1 if the event occurs and 0 if it does not. The posterior mean is then the event probability. For efficiency, indicator variables are defined in the generated quantities block.\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mean_gt;\n  int&lt;lower=0, upper=1&gt; sd_gt;\n  {\n    array[N] real y_rep = normal_rng(alpha + beta * x, sigma);\n    mean_gt = mean(y_rep) &gt; mean(y);\n    sd_gt = sd(y_rep) &gt; sd(y);\n  }\n}\nThe indicator variable mean_gt will have value 1 if the mean of the simulated data y_rep is greater than or equal to the mean of he original data y. Because the values of y_rep are not needed for the posterior predictive checks, the program saves output space by using a local variable for y_rep. The statistics mean(u) and sd(y) could also be computed in the transformed data block and saved.\nFor the example in the previous section, where over-dispersed data generated by a negative binomial distribution was fit with a simple Poisson model, the following plot illustrates the posterior p-value calculation for the mean statistic.\n\n\n\nHistogram of means of replicated data sets; vertical red line at mean of original data.\n\n\nThe p-value for the mean is just the percentage of replicated data sets whose statistic is greater than or equal that of the original data. Using a Poisson model for negative binomial data still fits the mean well, with a posterior \\(p\\)-value of 0.49. In Stan terms, it is extracted as the posterior mean of the indicator variable mean_gt.\nThe standard deviation statistic tells a different story.\n\n\n\nScatterplot of standard deviations of replicated data sets; the vertical red line is at standard deviation of original data.\n\n\nHere, the original data has much higher standard deviation than any of the replicated data sets. The resulting \\(p\\)-value estimated by Stan after a large number of iterations is exactly zero (the absolute error bounds are fine, but a lot of iterations are required to get good relative error bounds on small \\(p\\)-values by sampling). In other words, there were no posterior draws in which the replicated data set had a standard deviation greater than or equal to that of the original data set. Clearly, the model is not capturing the dispersion of the original data. The point of this exercise isn’t just to figure out that there’s a problem with a model, but to isolate where it is. Seeing that the data is over-dispersed compared to the Poisson model would be reason to fit a more general model like the negative binomial or a latent varying effects (aka random effects) model that can account for the over-dispersion.\n\n\nAny statistic may be used for the data, but these can be guided by the quantities of interest in the model itself. Popular choices in addition to mean and standard deviation are quantiles, such as the median, 5% or 95% quantiles, or even the maximum or minimum value to test extremes.\nDespite the range of choices, test statistics should ideally be ancillary, in the sense that they should be testing something other than the fit of a parameter. For example, a simple normal model of a data set will typically fit the mean and variance of the data quite well as long as the prior doesn’t dominate the posterior. In contrast, a Poisson model of the same data cannot capture both the mean and the variance of a data set if they are different, so they bear checking in the Poisson case. As we saw with the Poisson case, the posterior mean for the single rate parameter was located near the data mean, not the data variance. Other distributions such as the lognormal and gamma distribution, have means and variances that are functions of two or more parameters.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#applications-of-pseudorandom-number-generation",
-    "href": "stan-users-guide/regression.html#applications-of-pseudorandom-number-generation",
-    "title": "Regression Models",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#prior-predictive-checks",
+    "href": "stan-users-guide/posterior-predictive-checks.html#prior-predictive-checks",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "The main application of pseudorandom number generator (PRNGs) is for posterior inference, including prediction and posterior predictive checks. They can also be used for pure data simulation, which is like a posterior predictive check with no conditioning. See the function reference manual for a complete description of the syntax and usage of pseudorandom number generators.\n\n\nConsider predicting unobserved outcomes using linear regression. Given predictors \\(x_1, \\dotsc, x_N\\) and observed outcomes \\(y_1, \\dotsc, y_N\\), and assuming a standard linear regression with intercept \\(\\alpha\\), slope \\(\\beta\\), and error scale \\(\\sigma\\), along with improper uniform priors, the posterior over the parameters given \\(x\\) and \\(y\\) is \\[\np\\left(\\alpha, \\beta, \\sigma \\mid x, y \\right)\n\\propto\n\\prod_{n=1}^N\n  \\textsf{normal}\\left(y_n \\mid \\alpha + \\beta x_n, \\sigma\\right).\n\\]\nFor this model, the posterior predictive inference for a new outcome \\(\\tilde{y}_m\\) given a predictor \\(\\tilde{x}_m\\), conditioned on the observed data \\(x\\) and \\(y\\), is \\[\np\\left(\\tilde{y}_n \\mid \\tilde{x}_n, x, y\\right)\n= \\int_{(\\alpha,\\beta,\\sigma)}\n  \\textsf{normal}\\left(\\tilde{y}_n \\mid \\alpha + \\beta \\tilde{x}_n, \\sigma\\right)\n  \\times\n  p\\left(\\alpha, \\beta, \\sigma \\mid x, y\\right)\n  \\,\\textrm{d}(\\alpha,\\beta,\\sigma).\n\\]\nTo code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n  vector[N] x;\n  int&lt;lower=0&gt; N_tilde;\n  vector[N_tilde] x_tilde;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(alpha + beta * x, sigma);\n}\ngenerated quantities {\n  vector[N_tilde] y_tilde;\n  for (n in 1:N_tilde) {\n    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);\n  }\n}\nGiven observed predictors \\(x\\) and outcomes \\(y\\), y_tilde will be drawn according to \\(p\\left(\\tilde{y} \\mid \\tilde{x}, y, x\\right)\\). This means that, for example, the posterior mean for y_tilde is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).\n\n\n\nA good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors \\(x\\) are used in place of new predictors \\(\\tilde{x}\\) for unobserved outcomes. If the model fits the data well, the predictions for \\(\\tilde{y}\\) based on \\(x\\) should match the observed data \\(y\\).\nTo code posterior predictive checks in Stan requires only a slight modification of the prediction code to use \\(x\\) and \\(N\\) in place of \\(\\tilde{x}\\) and \\(\\tilde{N}\\),\ngenerated quantities {\n  vector[N] y_tilde;\n  for (n in 1:N) {\n    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);\n  }\n}\nGelman et al. (2013) recommend choosing several posterior draws \\(\\tilde{y}^{(1)}, \\dotsc, \\tilde{y}^{(M)}\\) and plotting each of them alongside the data \\(y\\) that was actually observed. If the model fits well, the simulated \\(\\tilde{y}\\) will look like the actual data \\(y\\).",
+    "text": "Prior predictive checks generate data according to the prior in order to asses whether a prior is appropriate (Gabry et al. 2019). A posterior predictive check generates replicated data according to the posterior predictive distribution. In contrast, the prior predictive check generates data according to the prior predictive distribution, \\[\ny^{\\textrm{sim}} \\sim p(y).\n\\] The prior predictive distribution is just like the posterior predictive distribution with no observed data, so that a prior predictive check is nothing more than the limiting case of a posterior predictive check with no data.\nThis is easy to carry out mechanically by simulating parameters \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta)\n\\] according to the priors, then simulating data \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}})\n\\] according to the data model given the simulated parameters. The result is a simulation from the joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta)\n\\] and thus \\[\ny^{\\textrm{sim}} \\sim p(y)\n\\] is a simulation from the prior predictive distribution.\n\n\nA prior predictive check is coded just like a posterior predictive check. If a posterior predictive check has already been coded and it’s possible to set the data to be empty, then no additional coding is necessary. The disadvantage to coding prior predictive checks as posterior predictive checks with no data is that Markov chain Monte Carlo will be used to sample the parameters, which is less efficient than taking independent draws using random number generation.\nPrior predictive checks can be coded entirely within the generated quantities block using random number generation. The resulting draws will be independent. Predictors must be read in from the actual data set—they do not have a generative model from which to be simulated. For a Poisson regression, prior predictive sampling can be encoded as the following complete Stan program.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] x;\n}\ngenerated quantities {\n  real alpha = normal_rng(0, 1);\n  real beta = normal_rng(0, 1);\n  array[N] real y_sim = poisson_log_rng(alpha + beta * x);\n}\nRunning this program using Stan’s fixed-parameter sampler yields draws from the prior. These may be plotted to consider their appropriateness.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/regression.html#footnotes",
-    "href": "stan-users-guide/regression.html#footnotes",
-    "title": "Regression Models",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nUnlike in Python and R, which are interpreted, Stan is translated to C++ and compiled, so loops and assignment statements are fast. Vectorized code is faster in Stan because (a) the expression tree used to compute derivatives can be simplified, leading to fewer virtual function calls, and (b) computations that would be repeated in the looping version, such as log(sigma) in the above model, will be computed once and reused.↩︎\nThe Phi_approx function is a rescaled version of the inverse logit function, so while the scale is roughly the same \\(\\Phi\\), the tails do not match.↩︎\nThe Bernoulli-logit distribution builds in the log link function, taking \\[\\texttt{bernoulli}\\mathtt{\\_}\\texttt{logit}\\left(y \\mid \\alpha\\right) = \\texttt{bernoulli}\\left(y \\mid \\operatorname{logit}^{-1}(\\alpha)\\right).\\]↩︎\nGelman and Hill (2007) treat the \\(\\delta\\) term equivalently as the location parameter in the distribution of student abilities.↩︎\nThe prior is named for Lewandowski, Kurowicka, and Joe, as it was derived by inverting the random correlation matrix generation strategy of Lewandowski, Kurowicka, and Joe (2009).↩︎\nThanks to Mike Lawrence for pointing this out in the GitHub issue for the manual.↩︎",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#example-of-prior-predictive-checks",
+    "href": "stan-users-guide/posterior-predictive-checks.html#example-of-prior-predictive-checks",
+    "title": "Posterior and Prior Predictive Checks",
+    "section": "",
+    "text": "Suppose we have a model for a football (aka soccer) league where there are \\(J\\) teams. Each team has a scoring rate \\(\\lambda_j\\) and in each game will be assumed to score \\(\\textrm{poisson}(\\lambda_j)\\) points. Yes, this model completely ignores defense. Suppose the modeler does not want to “put their thumb on the scale” and would rather “let the data speak for themselves” and so uses a prior with very wide tails, because it seems uninformative, such as the widely deployed \\[\n\\lambda_j \\sim \\textrm{gamma}(\\epsilon_1, \\epsilon_2).\n\\] This is not just a manufactured example; The BUGS Book recommends setting \\(\\epsilon = (0.5, 0.00001)\\), which corresponds to a Jeffreys prior for a Poisson rate parameter prior (Lunn et al. 2012, 85).\nSuppose the league plays a round-robin tournament wherein every team plays every other team. The following Stan model generates random team abilities and the results of such a round-robin tournament, which may be used to perform prior predictive checks.\ndata {\n  int&lt;lower=0&gt; J;\n  array[2] real&lt;lower=0&gt; epsilon;\n}\ngenerated quantities {\n  array[J] real&lt;lower=0&gt; lambda;\n  array[J, J] int y;\n  for (j in 1:J) lambda[j] = gamma_rng(epsilon[1], epsilon[2]);\n  for (i in 1:J) {\n    for (j in 1:J) {\n      y[i, j] = poisson_rng(lambda[i]) - poisson_rng(lambda[j]);\n    }\n  }\n}\nIn this simulation, teams play each other twice and play themselves once. This could be made more realistic by controlling the combinatorics to only generate a single result for each pair of teams, of which there are \\(\\binom{J}{2} = \\frac{J \\cdot (J - 1)}{2}.\\)\nUsing the \\(\\textrm{gamma}(0.5, 0.00001)\\) reference prior on team abilities, the following are the first 20 simulated point differences for the match between the first two teams, \\(y^{(1:20)}_{1, 2}\\).\n2597 -26000   5725  22496   1270   1072   4502  -2809   -302   4987\n7513   7527  -3268 -12374   3828   -158 -29889   2986  -1392     66\nThat’s some pretty highly scoring football games being simulated; all but one has a score differential greater than 100! In other words, this \\(\\textrm{gamma}(0.5, 0.00001)\\) prior is putting around 95% of its weight on score differentials above 100. Given that two teams combined rarely score 10 points, this prior is way out of line with prior knowledge about football matches; it is not only consistent with outcomes that have never occurred in the history of the sport, it puts most of the prior probability mass there.\nThe posterior predictive distribution can be strongly affected by the prior when there is not much observed data and substantial prior mass is concentrated around infeasible values (A. Gelman 2006).\nJust as with posterior predictive distributions, any statistics of the generated data may be evaluated. Here, the focus was on score difference between a single pair of teams, but it could’ve been on maximums, minimums, averages, variances, etc.\nIn this textbook example, the prior is univariate and directly related to the expected number of points scored, and could thus be directly inspected for consistency with prior knowledge about scoring rates in football. There will not be the same kind of direct connection when the prior and data model distributions are multivariate. In these more challenging situations, prior predictive checks are an easy way to get a handle on the implications of a prior in terms of what it says the data is going to look like; for a more complex application involving spatially heterogeneous air pollution concentration, see (Gabry et al. 2019).\nPrior predictive checks can also be compared with the data, but one should not expect them to be calibrated in the same way as posterior predictive checks. That would require guessing the posterior and encoding it in the prior. The goal is make sure the prior is not so wide that it will pull probability mass away from feasible values.",
     "crumbs": [
       "Stan Users Guide",
-      "Example Models",
-      "Regression Models"
+      "Posterior Inference & Model Checking",
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/simulation-based-calibration.html",
-    "href": "stan-users-guide/simulation-based-calibration.html",
-    "title": "Simulation-Based Calibration",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#mixed-replication",
+    "href": "stan-users-guide/posterior-predictive-checks.html#mixed-replication",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "A Bayesian posterior is calibrated if the posterior intervals have appropriate coverage. For example, 80% intervals are expected to contain the true parameter 80% of the time. If data is generated according to a model, Bayesian posterior inference with respect to that model is calibrated by construction. Simulation-based calibration (SBC) exploits this property of Bayesian inference to assess the soundness of a posterior sampler. Roughly, the way it works is by simulating parameters according to the prior, then simulating data conditioned on the simulated parameters, then testing posterior calibration of the inference algorithm over independently simulated data sets. This chapter follows Talts et al. (2018), which improves on the original approach developed by Cook, Gelman, and Rubin (2006).\n\n\nSuppose a Bayesian model is given in the form of a prior density \\(p(\\theta)\\) and sampling density \\(p(y \\mid \\theta).\\) Now consider a process that first simulates parameters from the prior, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta),\n\\] and then simulates data given the parameters, \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}}).\n\\] By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta).\n\\] From Bayes’s rule, it follows that for any observed (fixed) data \\(y\\), \\[\np(\\theta \\mid y) \\propto p(y, \\theta).\n\\] Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, \\[\n\\theta^{(1)}, \\ldots, \\theta^{(M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Because \\(\\theta^{\\textrm{sim}}\\) is also distributed as a draw from the posterior, the rank statistics of \\(\\theta^{\\textrm{sim}}\\) with respect to \\(\\theta^{(1)}, \\ldots \\theta^{(M)}\\) should be uniform.\nThis is one way to define calibration, because it follows that posterior intervals will have appropriate coverage (Dawid 1982; Gneiting, Balabdaoui, and Raftery 2007). If the rank of \\(\\theta^{\\textrm{sim}}\\) is uniform among the draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)},\\) then for any 90% interval selected, the probability the true value \\(\\theta^{\\textrm{sim}}\\) falls in it will also be 90%. The same goes for any other posterior interval.\n\n\n\nSuppose the Bayesian model to test has joint density \\[\np(y, \\theta) = p(y \\mid \\theta) \\cdot p(\\theta),\n\\] with data \\(y\\) and parameters \\(\\theta\\) (both are typically multivariate). Simulation-based calibration works by generating \\(N\\) simulated parameter and data pairs according to the joint density, \\[\n(y^{\\textrm{sim}(1)}, \\theta^{\\textrm{sim}(1)}),\n\\ldots, (y^{\\textrm{sim}(N)}, \\theta^{\\textrm{sim}(N)}),\n\\sim p(y, \\theta).\n\\] For each simulated data set \\(y^{\\textrm{sim}(n)}\\), use the algorithm to be tested to generate \\(M\\) posterior draws, which if everything is working properly, will be distributed marginally as \\[\n\\theta^{(n, 1)}, \\ldots, \\theta^{(n, M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}(n)}).\n\\] For a simulation \\(n\\) and parameter \\(k\\), the rank of the simulated parameter among the posterior draws is \\[\\begin{eqnarray*}\nr_{n, k}\n& = &\n\\textrm{rank}(\\theta_k^{\\textrm{sim}(n)},\n              (\\theta^{(n, 1)}, \\ldots, \\theta^{(n,M)}))\n\\\\[4pt]\n& = &\n\\sum_{m = 1}^M\n  \\textrm{I}[\\theta_k^{(n,m)} &lt; \\theta_k^{\\textrm{sim}(n)}].\n\\end{eqnarray*}\\] That is, the rank is the number of posterior draws \\(\\theta^{(n,m)}_k\\) that are less than the simulated draw \\(\\theta^{\\textrm{sim}(n)}_k.\\)\nIf the algorithm generates posterior draws according to the posterior, the ranks should be uniformly distributed from \\(0\\) to \\(M\\), so that the ranks plus one are uniformly distributed from \\(1\\) to \\(M + 1\\), \\[\nr_{n, k} + 1\n\\sim\n\\textrm{categorical}\\! \\left(\\frac{1}{M + 1}, \\ldots, \\frac{1}{M + 1}\\right).\n\\] Simulation-based calibration uses this expected behavior to test the calibration of each parameter of a model on simulated data. Talts et al. (2018) suggest plotting binned counts of \\(r_{1:N,\nk}\\) for different parameters \\(k\\); Cook, Gelman, and Rubin (2006) automate the process with a hypothesis test for uniformity.\n\n\n\nRunning simulation-based calibration in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.\n\n\nFor illustration, a very simple model will suffice. Suppose there are two parameters \\((\\mu, \\sigma)\\) with independent priors, \\[\n\\mu \\sim \\textrm{normal}(0, 1),\n\\] and \\[\n\\sigma \\sim \\textrm{lognormal}(0, 1).\n\\] The data \\(y = y_1, \\ldots, y_N\\) is drawn conditionally independently given the parameters, \\[\ny_n \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The joint prior density is thus \\[\np(\\mu, \\sigma)\n= \\textrm{normal}(\\mu \\mid 0, 1)\n  \\cdot \\textrm{lognormal}(\\sigma \\mid 0, 1),\n\\] and the sampling density is \\[\np(y \\mid \\mu, \\sigma)\n= \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, \\sigma).\n\\]\nFor example, suppose the following two parameter values are drawn from the prior in the first simulation, \\[\n(\\mu^{\\textrm{sim(1)}}, \\sigma^{\\textrm{sim(1)}}) = (1.01, 0.23).\n\\] Then data \\(y^{\\textrm{sim}(1)} \\sim p(y \\mid \\mu^{\\textrm{sim(1)}},\n\\sigma^{\\textrm{sim(1)}})\\) is drawn according to the sampling distribution. Next, \\(M = 4\\) draws are taken from the posterior \\(\\mu^{(1,m)}, \\sigma^{(1,m)} \\sim p(\\mu, \\sigma \\mid y^{\\textrm{sim}(1)})\\), \\[\n\\begin{array}{r|rr}\nm & \\mu^{(1,m)} & \\sigma^{(1,m)}\n\\\\ \\hline\n1 & 1.07 & 0.33\n\\\\\n2 & -0.32 & 0.14\n\\\\\n3 & -0.99 & 0.26\n\\\\\n4 & 1.51 & 0.31\n\\end{array}\n\\] Then the comparisons on which ranks are based look as follows, \\[\n\\begin{array}{r|cc}\nm & \\textrm{I}(\\mu^{(1,m)} &lt; \\mu^{\\textrm{sim}(1)})\n& \\textrm{I}(\\sigma^{(1,m)} &lt; \\sigma^{\\textrm{sim}(1)})\n\\\\ \\hline\n1 & 0 & 0\n\\\\\n2 & 1 & 1\n\\\\\n3 & 1 & 0\n\\\\\n4 & 0 & 0\n\\end{array}\n\\] The ranks are the column sums, \\(r_{1,1} = 2\\) and \\(r_{1,2} = 1\\). Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between \\(0\\) and \\(M\\), the number of posterior draws.\n\n\n\nTo code simulation-based calibration in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = normal_rng(mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nTo avoid confusion with the number of simulated data sets used for simulation-based calibration, J is used for the number of simulated data points.\nThe model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.\n\n\n\nThe entire simulation-based calibration process is as follows, where\n\np(theta) is the prior density\np(y | theta) is the sampling density\nK is the number of parameters\nN is the total number of simulated data sets and fits\nM is the number of posterior draws per simulated data set\n\nSBC(p(theta), p(y | theta), K, N, M)\n------------------------------------\nfor (n in 1:N) {\n    // simulate parameters and data\n    theta(sim(n)) ~ p(theta)\n    y(sim(n)) ~ p(y | theta(sim(n)))\n\n    // posterior draws given simulated data\n    for (m in 1:M) {\n        theta(n, m) ~ p(theta | y(sim(n)))\n    }\n    // calculate rank of sim among posterior draws\n    for (k in 1:K) {\n        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))\n    }\n}\n// test uniformity of each parameter\nfor (k in 1:K) {\n    test uniformity of rank(1:N, k)\n}\n\n\n\nThe draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws. Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of M, then thinning down to M draws.\n\n\n\n\nA simple, though not very highly powered, \\(\\chi^2\\)-squared test for uniformity can be formulated by binning the ranks \\(0:M\\) into \\(J\\) bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, Cook, Gelman, and Rubin (2006) transform the ranks using the inverse cumulative distribution function for the standard normal and then perform a test for normality. Talts et al. (2018) recommend visual inspection of the binned plots.\nThe bins don’t need to be exactly the same size. In general, if \\(b_j\\) is the number of ranks that fall into bin \\(j\\) and \\(e_j\\) is the number of ranks expected to fall into bin \\(j\\) (which will be proportional to its size under uniformity), the test statistic is \\[\nX^2 = \\sum_{j = 1}^J \\frac{(b_j - e_j)^2}{e_j}.\n\\] The terms are approximately square standard normal, so that under the null hypothesis of uniformity, \\[\nX^2 \\sim \\textrm{chiSquared}(J - 1),\n\\] with the corresponding \\(p\\)-value given by the complementary cumulative distribution function (CCDF) of \\(\\textrm{chiSquared}(J - 1)\\) applied to \\(X^2\\). Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., \\(e_j \\geq 5.\\)\n\n\nBecause there are \\(M + 1\\) possible ranks, with \\(J\\) bins, it is easiest to have \\(M + 1\\) be divisible by \\(J\\). For instance, if \\(J = 20\\) and \\(M = 999\\), then there are \\(1000\\) possible ranks and an expected count in each bin of \\(\\frac{M + 1}{J} = 50.\\)\nDistributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, \\[\n\\textrm{bin}(r_{n, m}, M, J)\n= 1 + \\left\\lfloor \\frac{r_{n, m}}{(M + 1) / J} \\right\\rfloor.\n\\] For example, with \\(M = 999\\) and \\(J = 20\\), \\((M + 1) / J = 50\\). The lowest rank checks out, \\[\n\\textrm{bin}(0, 999, 20) = 1 + \\lfloor 0 / 50 \\rfloor = 1,\n\\] as does the 50th rank, \\[\n\\textrm{bin}(49, 999, 20) = 1 + \\lfloor 49 / 50 \\rfloor = 1,\n\\] and the 51st is appropriately put in the second bin, \\[\n\\textrm{bin}(50, 999, 20) = 1 + \\lfloor 50 / 50 \\rfloor = 2.\n\\] The highest rank also checks out, with \\(\\textrm{bin}(1000, 999, 20) = 50.\\)\nTo summarize, the following pseudocode computes the \\(b_j\\) values for the \\(\\chi^2\\) test or for visualization in a histogram.\nInputs: M draws, J bins, N parameters, ranks r[n, m]\nb[1:J] = 0\nfor (m in 1:M) {\n  ++b[1 + floor(r[n, m] * J / (M + 1))]\n}\nwhere the ++b[n] notation is a common form of syntactic sugar for b[n] = b[n] + 1.\nIn general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.\n\n\n\n\nThis section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.\n\n\nConsider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 1)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1)\n\\\\[4pt]\ny_{1:10} & \\sim & \\textrm{normal}(\\mu, \\sigma).\n\\end{eqnarray*}\\] The Stan program for evaluating SBC for this model is\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = student_t_rng(4, mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; I_lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nAfter running this for enough iterations so that the effective sample size is larger than \\(M\\), then thinning to \\(M\\) draws (here \\(M = 999\\)), the ranks are computed and binned, and then plotted.\n\n\n\n\n\n\nFigure 1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.\n\n\n\n\n\n\nNow consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.\n\n\n\n\n\n\nFigure 2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.\n\n\n\n\n\n\nThe example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.\nThis can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means \\(y_j\\) and standard deviations \\(\\sigma_j\\) of differences in test score after the same intervention in \\(J = 8\\) different schools. Rubin (1981) applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by Talts et al. (2018); this section replicates their results.\nThe meta-analysis model has parameters for a population mean \\(\\mu\\) and standard deviation \\(\\tau &gt; 0\\) as well as the effect \\(\\theta_j\\) of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 5)\n\\\\[4pt]\n\\tau & \\sim & \\textrm{normal}_{+}(0, 5).\n\\end{eqnarray*}\\] School level effects are modeled as normal given the population parameters, \\[\n\\theta_j \\sim \\textrm{normal}(\\mu, \\tau).\n\\] The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, \\[\ny_j \\sim \\textrm{normal}(\\theta_j, \\sigma_j).\n\\]\nThis model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.\ntransformed data {\n  real mu_sim = normal_rng(0, 5);\n  real tau_sim = abs(normal_rng(0, 5));\n  int&lt;lower=0&gt; J = 8;\n  array[J] real theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);\n  array[J] real&lt;lower=0&gt; sigma = abs(normal_rng(rep_vector(0, J), 5));\n  array[J] real y = normal_rng(theta_sim, sigma);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; tau;\n  array[J] real theta;\n}\nmodel {\n  tau ~ normal(0, 5);\n  mu ~ normal(0, 5);\n  theta ~ normal(mu, tau);\n  y ~ normal(theta, sigma);\n}\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mu_lt_sim = mu &lt; mu_sim;\n  int&lt;lower=0, upper=1&gt; tau_lt_sim = tau &lt; tau_sim;\n  int&lt;lower=0, upper=1&gt; theta1_lt_sim = theta[1] &lt; theta_sim[1];\n}\nAs usual for simulation-based calibration, the transformed data encodes the data-generating process using random number generators. Here, the population parameters \\(\\mu\\) and \\(\\tau\\) are first simulated, then the school-level effects \\(\\theta\\), and then finally the observed data \\(\\sigma_j\\) and \\(y_j.\\) The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only \\(\\theta_1\\) because the schools are exchangeable in the simulation.\nWhen fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.\nSBC also diagnoses the problem. Here’s the rank plots for running \\(N =\n200\\) simulations with 1000 warmup iterations and \\(M = 999\\) draws per simulation used to compute the ranks.\n\n\n\n\n\n\n\n\n\n\n\n(a) \\(\\mu\\)\n\n\n\n\n\n\n\n\n\n\n\n(b) \\(\\tau\\)\n\n\n\n\n\n\n\n\n\n\n\n(c) \\({\\theta}_1\\)\n\n\n\n\n\n\n\nFigure 3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for \\(\\theta_1\\) (Figure 3 (c)).\n\n\n\nAlthough the population mean and standard deviation \\(\\mu\\) and \\(\\tau\\) appear well calibrated, \\(\\theta_1\\) tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as \\(\\tau\\) approaches zero and the \\(\\theta_j\\) become highly constrained. The chapter on reparameterization explains how to remedy this problem and fit this kind of hierarchical model with Stan.",
+    "text": "Andrew Gelman, Meng, and Stern (1996) discuss the case of mixed replication for hierarchical models in which the hyperparameters remain fixed, but varying effects are replicated. This is neither a purely prior nor purely posterior predictive check, but falls somewhere in between.\nFor example, consider a simple varying intercept logistic regression, with intercepts \\(\\alpha_k\\) for \\(k \\in 1:K\\). Each data item \\(y_n \\in \\{ 0, 1 \\}\\) is assumed to correspond to group \\(kk_n \\in 1:K.\\) The data model is thus \\[\ny_n \\sim \\textrm{bernoulli}(\\textrm{logit}^{-1}(\\alpha_{kk[n]})).\n\\] The varying intercepts have a hierarchical normal prior, \\[\n\\alpha_k \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The hyperparameters are themselves given weakly informative priors, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 2)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1).\n\\end{eqnarray*}\\]\nLike in a posterior predictive check, the hyperparameters \\(\\mu\\) and \\(\\sigma\\) are drawn from the posterior, \\[\n\\mu^{(m)}, \\sigma^{(m)} \\sim p(\\mu, \\sigma \\mid y)\n\\] Like in a prior predictive check, replicated values of \\(\\alpha\\) are drawn from the hyperparameters, \\[\n\\alpha^{\\textrm{rep}(m)}_k \\sim \\textrm{normal}(\\alpha_k \\mid\n\\mu^{(m)}, \\sigma^{(m)}).\n\\] The data items are then each replicated using the replicated intercepts, \\[\ny^{\\textrm{rep}(m)}_n \\sim\n\\textrm{bernoulli}\n  (\\textrm{logit}^{-1}(\\alpha^{\\textrm{rep}(m)}_{kk[n]})).\n\\] Thus the \\(y^{\\textrm{rep}(m)}\\) can be seen as a kind of posterior predictive replication of observations from new groups that were not among the original \\(K\\) groups.\nIn Stan, mixed predictive replications \\(y^{\\textrm{rep}(m)}\\) can be programmed directly.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n  vector&lt;offset=mu, multiplier=sigma&gt;[K] alpha;\n}\nmodel {\n  mu ~ normal(0, 2);               // hyperprior\n  sigma ~ lognormal(0, 1);\n  alpha ~ normal(mu, sigma);       // hierarchical prior\n  y ~ bernoulli_logit(alpha[kk]);  // data model\n}\ngenerated quantities {\n  // alpha replicated;  mu and sigma not replicated\n  array[K] real alpha_rep\n    = normal_rng(rep_vector(mu, K), sigma);\n  array[N] int&lt;lower=0, upper=1&gt; y_rep\n    = bernoulli_logit_rng(alpha_rep[kk]);\n}",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Simulation-Based Calibration"
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/simulation-based-calibration.html#bayes-is-calibrated-by-construction",
-    "href": "stan-users-guide/simulation-based-calibration.html#bayes-is-calibrated-by-construction",
-    "title": "Simulation-Based Calibration",
+    "objectID": "stan-users-guide/posterior-predictive-checks.html#joint-model-representation",
+    "href": "stan-users-guide/posterior-predictive-checks.html#joint-model-representation",
+    "title": "Posterior and Prior Predictive Checks",
     "section": "",
-    "text": "Suppose a Bayesian model is given in the form of a prior density \\(p(\\theta)\\) and sampling density \\(p(y \\mid \\theta).\\) Now consider a process that first simulates parameters from the prior, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta),\n\\] and then simulates data given the parameters, \\[\ny^{\\textrm{sim}} \\sim p(y \\mid \\theta^{\\textrm{sim}}).\n\\] By the definition of conditional densities, the simulated data and parameters constitute an independent draw from the model’s joint distribution, \\[\n(y^{\\textrm{sim}}, \\theta^{\\textrm{sim}}) \\sim p(y, \\theta).\n\\] From Bayes’s rule, it follows that for any observed (fixed) data \\(y\\), \\[\np(\\theta \\mid y) \\propto p(y, \\theta).\n\\] Therefore, the simulated parameters constitute a draw from the posterior for the simulated data, \\[\n\\theta^{\\textrm{sim}} \\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Now consider an algorithm that produces a sequence of draws from the posterior given this simulated data, \\[\n\\theta^{(1)}, \\ldots, \\theta^{(M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}}).\n\\] Because \\(\\theta^{\\textrm{sim}}\\) is also distributed as a draw from the posterior, the rank statistics of \\(\\theta^{\\textrm{sim}}\\) with respect to \\(\\theta^{(1)}, \\ldots \\theta^{(M)}\\) should be uniform.\nThis is one way to define calibration, because it follows that posterior intervals will have appropriate coverage (Dawid 1982; Gneiting, Balabdaoui, and Raftery 2007). If the rank of \\(\\theta^{\\textrm{sim}}\\) is uniform among the draws \\(\\theta^{(1)}, \\ldots, \\theta^{(M)},\\) then for any 90% interval selected, the probability the true value \\(\\theta^{\\textrm{sim}}\\) falls in it will also be 90%. The same goes for any other posterior interval.",
+    "text": "Following Andrew Gelman, Meng, and Stern (1996), prior, posterior, and mixed replications may all be defined as posteriors from joint models over parameters and observed and replicated data.\n\n\nFor example, posterior predictive replication may be formulated using distribution notation as follows. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] The heavily overloaded distribution notation is meant to indicate that both \\(y\\) and \\(y^{\\textrm{rep}}\\) are drawn from the same distribution, or more formally using capital letters to distinguish random variables, that the conditional densities \\(p_{Y^{\\textrm{rep}} \\mid\n\\Theta}\\) and \\(p_{Y \\mid \\Theta}\\) are the same.\nThe joint density is \\[\np(\\theta, y, y^{\\textrm{rep}})\n= p(\\theta) \\cdot p(y \\mid \\theta) \\cdot p(y^{\\textrm{rep}} \\mid \\theta).\n\\] This again is assuming that the two distributions for \\(y\\) and \\(y^{\\textrm{rep}}\\) are identical.\nThe variable \\(y\\) is observed, with the predictive simulation \\(y^{\\textrm{rep}}\\) and parameter vector \\(\\theta\\) not observed. The posterior is \\(p(y^{\\textrm{rep}}, \\theta \\mid y)\\). Given draws from the posterior, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are retained.\n\n\n\nThe prior predictive model simply drops the data component of the posterior predictive model. \\[\\begin{eqnarray*}\n\\theta & \\sim & p(\\theta)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\theta)\n\\end{eqnarray*}\\] This corresponds to the joint density \\[\np(\\theta, y^{\\textrm{rep}}) = p(\\theta) \\cdot p(y^{\\textrm{rep}} \\mid\n\\theta).\n\\]\nIt is typically straightforward to draw \\(\\theta\\) from the prior and \\(y^{\\textrm{rep}}\\) from the data model given \\(\\theta\\) efficiently. In cases where it is not, the model may be coded and executed just as the posterior predictive model, only with no data.\n\n\n\nThe mixed replication corresponds to the model \\[\\begin{eqnarray*}\n\\phi & \\sim & p(\\phi)\n\\\\[2pt]\n\\alpha & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny & \\sim & p(y \\mid \\alpha)\n\\\\[2pt]\n\\alpha^{\\textrm{rep}} & \\sim & p(\\alpha \\mid \\phi)\n\\\\[2pt]\ny^{\\textrm{rep}} & \\sim & p(y \\mid \\phi)\n\\end{eqnarray*}\\] The notation here is meant to indicate that \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\) have identical distributions, as do \\(y\\) and \\(y^{\\textrm{rep}}\\).\nThis corresponds to a joint model \\[\np(\\phi, \\alpha, \\alpha^{\\textrm{rep}}, y, y^{\\textrm{rep}})\n=\np(\\phi)\n\\cdot p(\\alpha \\mid \\phi)\n\\cdot p(y \\mid \\alpha)\n\\cdot p(\\alpha^{\\textrm{rep}} \\mid \\phi)\n\\cdot p(y^{\\textrm{rep}} \\mid \\alpha^{\\textrm{rep}}),\n\\] where \\(y\\) is the only observed variable, \\(\\alpha\\) contains the lower-level parameters and \\(\\phi\\) the hyperparameters. Note that \\(\\phi\\) is not replicated and instead appears in the distribution for both \\(\\alpha\\) and \\(\\alpha^{\\textrm{rep}}\\).\nThe posterior is \\(p(\\phi, \\alpha, \\alpha^{\\textrm{rep}},\ny^{\\textrm{rep}} \\mid y)\\). From posterior draws, the posterior predictive simulations \\(y^{\\textrm{rep}}\\) are kept.",
     "crumbs": [
       "Stan Users Guide",
       "Posterior Inference & Model Checking",
-      "Simulation-Based Calibration"
+      "Posterior and Prior Predictive Checks"
     ]
   },
   {
-    "objectID": "stan-users-guide/simulation-based-calibration.html#simulation-based-calibration-1",
-    "href": "stan-users-guide/simulation-based-calibration.html#simulation-based-calibration-1",
-    "title": "Simulation-Based Calibration",
+    "objectID": "stan-users-guide/problematic-posteriors.html",
+    "href": "stan-users-guide/problematic-posteriors.html",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "Suppose the Bayesian model to test has joint density \\[\np(y, \\theta) = p(y \\mid \\theta) \\cdot p(\\theta),\n\\] with data \\(y\\) and parameters \\(\\theta\\) (both are typically multivariate). Simulation-based calibration works by generating \\(N\\) simulated parameter and data pairs according to the joint density, \\[\n(y^{\\textrm{sim}(1)}, \\theta^{\\textrm{sim}(1)}),\n\\ldots, (y^{\\textrm{sim}(N)}, \\theta^{\\textrm{sim}(N)}),\n\\sim p(y, \\theta).\n\\] For each simulated data set \\(y^{\\textrm{sim}(n)}\\), use the algorithm to be tested to generate \\(M\\) posterior draws, which if everything is working properly, will be distributed marginally as \\[\n\\theta^{(n, 1)}, \\ldots, \\theta^{(n, M)}\n\\sim p(\\theta \\mid y^{\\textrm{sim}(n)}).\n\\] For a simulation \\(n\\) and parameter \\(k\\), the rank of the simulated parameter among the posterior draws is \\[\\begin{eqnarray*}\nr_{n, k}\n& = &\n\\textrm{rank}(\\theta_k^{\\textrm{sim}(n)},\n              (\\theta^{(n, 1)}, \\ldots, \\theta^{(n,M)}))\n\\\\[4pt]\n& = &\n\\sum_{m = 1}^M\n  \\textrm{I}[\\theta_k^{(n,m)} &lt; \\theta_k^{\\textrm{sim}(n)}].\n\\end{eqnarray*}\\] That is, the rank is the number of posterior draws \\(\\theta^{(n,m)}_k\\) that are less than the simulated draw \\(\\theta^{\\textrm{sim}(n)}_k.\\)\nIf the algorithm generates posterior draws according to the posterior, the ranks should be uniformly distributed from \\(0\\) to \\(M\\), so that the ranks plus one are uniformly distributed from \\(1\\) to \\(M + 1\\), \\[\nr_{n, k} + 1\n\\sim\n\\textrm{categorical}\\! \\left(\\frac{1}{M + 1}, \\ldots, \\frac{1}{M + 1}\\right).\n\\] Simulation-based calibration uses this expected behavior to test the calibration of each parameter of a model on simulated data. Talts et al. (2018) suggest plotting binned counts of \\(r_{1:N,\nk}\\) for different parameters \\(k\\); Cook, Gelman, and Rubin (2006) automate the process with a hypothesis test for uniformity.",
+    "text": "Mathematically speaking, with a proper posterior, one can do Bayesian inference and that’s that. There is not even a need to require a finite variance or even a finite mean—all that’s needed is a finite integral. Nevertheless, modeling is a tricky business and even experienced modelers sometimes code models that lead to improper priors. Furthermore, some posteriors are mathematically sound, but ill-behaved in practice. This chapter discusses issues in models that create problematic posterior inferences, either in general for Bayesian inference or in practice for Stan.\n\n\nThis section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.\n\n\n\n\nThe first example of collinearity is an artificial example involving redundant intercept parameters.1\nSuppose there are observations \\(y_n\\) for \\(n \\in \\{1,\\dotsc,N\\}\\), two intercept parameters \\(\\lambda_1\\) and \\(\\lambda_2\\), a scale parameter \\(\\sigma &gt; 0\\), and the data model \\[\ny_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2, \\sigma).\n\\]\nFor any constant \\(q\\), the sampling density for \\(y\\) does not change if we add \\(q\\) to \\(\\lambda_1\\) and subtract it from \\(\\lambda_2\\), i.e., \\[\np(y \\mid \\lambda_1, \\lambda_2,\\sigma)\n=\np(y \\mid \\lambda_1 + q, \\lambda_2 - q, \\sigma).\n\\]\nThe consequence is that an improper uniform prior \\(p(\\mu,\\sigma)\n\\propto 1\\) leads to an improper posterior. This impropriety arises because the neighborhoods around \\(\\lambda_1 + q, \\lambda_2 - q\\) have the same mass no matter what \\(q\\) is. Therefore, a sampler would need to spend as much time in the neighborhood of \\(\\lambda_1=1\\,000\\,000\\,000\\) and \\(\\lambda_2=-1\\,000\\,000\\,000\\) as it does in the neighborhood of \\(\\lambda_1=0\\) and \\(\\lambda_2=0\\), and so on for ever more far-ranging values.\nThe marginal posterior \\(p(\\lambda_1,\\lambda_2 \\mid y)\\) for this model is thus improper.2\nThe impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where \\(\\lambda_2 = \\lambda_1 + c\\) for some constant \\(c\\).\nContrast this model with a simple regression with a single intercept parameter \\(\\mu\\) and data model \\[\ny_n \\sim \\textsf{normal}(\\mu,\\sigma).\n\\] Even with an improper prior, the posterior is proper as long as there are at least two data points \\(y_n\\) with distinct values.\n\n\n\nConsider an item-response theory model for students \\(j \\in 1{:}J\\) with abilities \\(\\alpha_j\\) and test items \\(i \\in 1{:}I\\) with difficulties \\(\\beta_i\\). The observed data are an \\(I \\times J\\) array with entries \\(y_{i, j} \\in \\{ 0, 1 \\}\\) coded such that \\(y_{i, j} = 1\\) indicates that student \\(j\\) answered question \\(i\\) correctly. The sampling distribution for the data is \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\alpha_j - \\beta_i)).\n\\]\nFor any constant \\(c\\), the probability of \\(y\\) is unchanged by adding a constant \\(c\\) to all the abilities and subtracting it from all the difficulties, i.e., \\[\np(y \\mid \\alpha, \\beta)\n=\np(y \\mid \\alpha + c, \\beta - c).\n\\]\nThis leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.\n\n\n\nThe general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model \\[\ny_n \\sim \\textsf{normal}(x_n \\beta, \\sigma)\n\\] for an \\(N\\)-dimensional observation vector \\(y\\), an \\(N \\times K\\) predictor matrix \\(x\\), and a \\(K\\)-dimensional coefficient vector \\(\\beta\\).\nNow suppose that column \\(k\\) of the predictor matrix is a multiple of column \\(k'\\), i.e., there is some constant \\(c\\) such that \\(x_{n,k} = c\n\\, x_{n,k'}\\) for all \\(n\\). In this case, the coefficients \\(\\beta_k\\) and \\(\\beta_{k'}\\) can covary without changing the predictions, so that for any \\(d \\neq 0\\), \\[\np(y \\mid \\ldots, \\beta_k, \\dotsc, \\beta_{k'}, \\dotsc, \\sigma)\n=\np(y \\mid \\ldots, d \\beta_k, \\dotsc, \\frac{d}{c} \\, \\beta_{k'}, \\dotsc,\n\\sigma).\n\\]\nEven if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.\n\n\n\nConsider adding a discrimination parameter \\(\\delta_i\\) for each question in an IRT model, with data model \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\delta_i(\\alpha_j - \\beta_i))).\n\\] For any constant \\(c \\neq 0\\), multiplying \\(\\delta\\) by \\(c\\) and dividing \\(\\alpha\\) and \\(\\beta\\) by \\(c\\) produces the same likelihood, \\[\np(y \\mid \\delta,\\alpha,\\beta)\n= p(y \\mid c \\delta, \\frac{1}{c}\\alpha, \\frac{1}{c}\\beta).\n\\] If \\(c &lt; 0\\), this switches the signs of every component in \\(\\alpha\\), \\(\\beta\\), and \\(\\delta\\) without changing the density.\n\n\n\nIn order to parameterize a \\(K\\)-simplex (i.e., a \\(K\\)-vector with non-negative values that sum to one), only \\(K - 1\\) parameters are necessary because the \\(K\\)th is just one minus the sum of the first \\(K\n- 1\\) parameters, so that if \\(\\theta\\) is a \\(K\\)-simplex, \\[\n\\theta_K = 1 - \\sum_{k=1}^{K-1} \\theta_k.\n\\]\nThe softmax function maps a \\(K\\)-vector \\(\\alpha\\) of linear predictors to a \\(K\\)-simplex \\(\\theta = \\texttt{softmax}(\\alpha)\\) by defining \\[\n\\theta_k = \\frac{\\exp(\\alpha_k)}{\\sum_{k'=1}^K \\exp(\\alpha_{k'})}.\n\\]\nThe softmax function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters \\(\\alpha\\). In particular, adding or subtracting a constant from each \\(\\alpha_k\\) produces the same simplex \\(\\theta\\).\n\n\n\n\nAll of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.\n\n\nIn the case of the multiple intercepts, \\(\\lambda_1\\) and \\(\\lambda_2\\), the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter \\(\\mu\\) and sampling distribution \\(y_n \\sim \\textsf{normal}(\\mu, \\sigma)\\). The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix \\(x\\).\n\n\n\nThe IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability \\(\\alpha_1\\) can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.\nThis solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters \\(\\delta_i\\). To solve this problem, one of the discrimination parameters, say \\(\\delta_1\\), must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.\nThe many-to-one nature of \\(\\texttt{softmax}(\\alpha)\\) is typically mitigated by pinning a component of \\(\\alpha\\), for instance fixing \\(\\alpha_K = 0\\). The resulting mapping is one-to-one from \\(K-1\\) unconstrained parameters to a \\(K\\)-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a \\(K-1\\)-vector can be written as\nvector softmax_id(vector alpha) {\n  vector[num_elements(alpha) + 1] alphac1;\n  for (k in 1:num_elements(alpha)) {\n    alphac1[k] = alpha[k];\n  }\n  alphac1[num_elements(alphac1)] = 0;\n  return softmax(alphac1);\n}\n\n\n\nSo far, the models have been discussed as if the priors on the parameters were improper uniform priors.\nA more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.\nFor example, normal priors on the multiple intercepts, \\[\n\\lambda_1, \\lambda_2 \\sim \\textsf{normal}(0,\\tau),\n\\] with a constant scale \\(\\tau\\), ensure that the posterior mode is located at a point where \\(\\lambda_1 = \\lambda_2\\), because this minimizes \\(\\log \\textsf{normal}(\\lambda_1 \\mid 0,\\tau) + \\log\n\\textsf{normal}(\\lambda_2 \\mid 0,\\tau)\\).3\nThe following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.\nThe two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.\n\n\n\nTwo intercepts with improper prior\n\n\nAdding a standard normal prior for the intercepts results in a proper posterior.\n\n\n\nTwo intercepts with proper prior\n\n\nThe single intercept parameterization with no prior also has a proper posterior.\n\n\n\nSingle intercepts with improper prior\n\n\nThe addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.\nAn alternative strategy for identifying a \\(K\\)-simplex parameterization \\(\\theta = \\texttt{softmax}(\\alpha)\\) in terms of an unconstrained \\(K\\)-vector \\(\\alpha\\) is to place a prior on the components of \\(\\alpha\\) with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning \\(\\alpha_K =\n0\\), the prior-based approach models the \\(K\\) outcomes symmetrically rather than modeling \\(K-1\\) outcomes relative to the \\(K\\)-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.\n\n\n\nCare must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.\nIdeally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.\nWhen there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.\nPriors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters \\(\\alpha\\) to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.\n\n\n\n\n\nWhere collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.\n\n\nConsider a normal mixture model with two location parameters \\(\\mu_1\\) and \\(\\mu_2\\), a shared scale \\(\\sigma &gt; 0\\), a mixture ratio \\(\\theta \\in\n[0,1]\\), and data model \\[\np(y \\mid \\theta,\\mu_1,\\mu_2,\\sigma)\n= \\prod_{n=1}^N \\big( \\theta \\, \\textsf{normal}(y_n \\mid \\mu_1,\\sigma)\n                       + (1 - \\theta) \\, \\textsf{normal}(y_n \\mid \\mu_2,\\sigma) \\big).\n\\] The issue here is exchangeability of the mixture components, because \\[\np(\\theta,\\mu_1,\\mu_2,\\sigma \\mid y) = p\\big((1-\\theta),\\mu_2,\\mu_1,\\sigma \\mid y\\big).\n\\] The problem is exacerbated as the number of mixture components \\(K\\) grows, as in clustering models, leading to \\(K!\\) identical posterior maxima.\n\n\n\nThe analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the \\(\\hat{R}\\) convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for \\(\\mu_1\\) that is equal to that of \\(\\mu_2\\), and a posterior mean for \\(\\theta\\) that is always 1/2, no matter what the data are.\n\n\n\nIn some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for \\(\\theta\\) in the two component mixture model will always be 1/2.\n\n\n\nTheoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.\nBeing able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for \\(\\mu_1\\) to move to a new mode when sampled conditioned on the current values of \\(\\mu_2\\) and \\(\\theta\\). For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.\nEven with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.\n\n\n\nSeveral hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.\n\n\nOne common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining \\(\\mu_1 &lt; \\mu_2\\) in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering \\(\\mu_1 &gt; \\mu_2\\). In these cases, the posteriors are affected by the constraint and true posterior uncertainty in \\(\\mu_1\\) and \\(\\mu_2\\) is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use \\(M\\) posterior draws to estimate \\(\\Pr[\\mu_1 &gt; \\mu_2]\\), will fail, because the estimator \\[\n\\Pr[\\mu_1 &gt; \\mu_2]\n\\approx\n\\sum_{m=1}^M \\textrm{I}\\left(\\mu_1^{(m)} &gt; \\mu_2^{(m)}\\right)\n\\] will result in an estimate of 0 because the posterior respects the constraint in the model.\n\n\n\nAnother common approach is to run a single chain or to initialize the parameters near realistic values.4\nThis can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.\n\n\n\n\n\nIt is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of \\(K\\) normals might devolve to have \\(\\mu_i =\n\\mu_j\\) and \\(\\sigma_i = \\sigma_j\\) for \\(i \\neq j\\).\nThis will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw \\((m)\\), it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.\nIt may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.\nIn general, it is difficult to recover exactly the right \\(K\\) mixture components in a mixture model as \\(K\\) increases beyond one (yes, even a two-component mixture can have this problem).\n\n\n\nIn some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.\n\n\nOne such example is a binary mixture model with scales varying by component, \\(\\sigma_1\\) and \\(\\sigma_2\\) for locations \\(\\mu_1\\) and \\(\\mu_2\\). In this situation, the density grows without bound as \\(\\sigma_1 \\rightarrow 0\\) and \\(\\mu_1 \\rightarrow y_n\\) for some \\(n\\); that is, one of the mixture components concentrates all of its mass around a single data item \\(y_n\\).\n\n\n\nAnother example of unbounded densities arises with a posterior such as \\(\\textsf{beta}(\\phi \\mid 0.5,0.5)\\), which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as \\(\\phi \\rightarrow 0\\) and \\(\\phi \\rightarrow 1\\). Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior \\[\\begin{align*}\np(\\phi \\mid y)\n&\\propto\n   \\textsf{beta}(\\phi \\mid 0.5,0.5) \\times \\prod_{n=1}^N \\textsf{Bernoulli}(y_n \\mid \\phi) \\\\\n&=\n   \\textsf{beta}\\left(\\phi \\,\\middle|\\, 0.5 + \\sum_{n=1}^N y_n, 0.5 + N - \\sum_{n=1}^N y_n\\right).\n\\end{align*}\\]\nIf \\(N = 9\\) and each \\(y_n = 1\\), the posterior is \\(\\textsf{beta}(\\phi \\mid 9.5,0.5)\\). This posterior is unbounded as \\(\\phi\n\\rightarrow 1\\). Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.\n\n\nStan does not sample directly on the constrained \\((0,1)\\) space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values \\(\\phi\\) are logit-transformed to \\((-\\infty,\\infty)\\). The boundaries at 0 and 1 are pushed out to \\(-\\infty\\) and \\(\\infty\\) respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of \\((0,1)\\) is \\(\\log\n\\operatorname{logit}^{-1}(\\phi) + \\log \\operatorname{logit}(1 - \\phi)\\).\nThere are two problems that still arise, though. The first is that if the posterior mass for \\(\\phi\\) is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.\n\n\n\n\n\nIn some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.\n\n\nConsider a logistic regression model with \\(N\\) observed outcomes \\(y_n\n\\in \\{ 0, 1 \\}\\), an \\(N \\times K\\) matrix \\(x\\) of predictors, a \\(K\\)-dimensional coefficient vector \\(\\beta\\), and data model \\[\ny_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(x_n \\beta)).\n\\] Now suppose that column \\(k\\) of the predictor matrix is such that \\(x_{n,k} &gt; 0\\) if and only if \\(y_n = 1\\), a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as \\(\\beta_k \\rightarrow \\infty\\), because for cases with \\(y_n = 1\\), \\(x_n \\beta \\rightarrow \\infty\\) and hence \\(\\operatorname{logit}^{-1}(x_n \\beta) \\rightarrow 1\\).\nWith separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for \\(\\beta_k\\) is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for \\(\\beta\\), which ensures a proper posterior.\n\n\n\n\nSuppose your model includes a parameter \\(\\psi\\) that is defined on \\([0,1]\\) and is given a flat prior \\(\\textsf{uniform}(\\psi \\mid 0,1)\\). Now if the data don’t tell us anything about \\(\\psi\\), the posterior is also \\(\\textsf{uniform}(\\psi \\mid 0,1)\\).\nAlthough there is no maximum likelihood estimate for \\(\\psi\\), the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on \\([0,1]\\), the posterior mean for \\(\\psi\\) is well-defined with value \\(1/2\\). Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for \\(\\psi\\) at all points in \\([0,1]\\).\n\n\n\nWith an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the collinearity section and illustrated in the non-identifiable density plot.\n\n\nGibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.\nConsider what happens with initial values \\(\\lambda_1^{(0)}, \\lambda_2^{(0)}\\). Gibbs sampling proceeds in iteration \\(m\\) by drawing \\[\\begin{align*}\n\\lambda_1^{(m)} &\\sim p(\\lambda_1 \\mid \\lambda_2^{(m-1)}, \\sigma^{(m-1)},  y) \\\\\n\\lambda_2^{(m)} &\\sim p(\\lambda_2 \\mid \\lambda_1^{(m)},   \\sigma^{(m-1)},  y) \\\\\n\\sigma^{(m)}    &\\sim p(\\sigma    \\mid \\lambda_1^{(m)},   \\lambda_2^{(m)}, y).\n\\end{align*}\\]\nNow consider the draw for \\(\\lambda_1\\) (the draw for \\(\\lambda_2\\) is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next \\(\\lambda_1\\) can be drawn is highly constrained by the current values of \\(\\lambda_2\\) and \\(\\sigma\\). Gibbs will run quickly and provide seemingly reasonable inferences for \\(\\lambda_1 + \\lambda_2\\). But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.\n\n\n\nHamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for \\(\\lambda_1\\) and \\(\\lambda_2\\), the gradient of the log posterior is going to adjust for the correlation and the simulation will run \\(\\lambda_1\\) and \\(\\lambda_2\\) in opposite directions along the valley corresponding to the ridge in the posterior log density.\n\n\n\nStan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior (see Hoffman and Gelman 2014). NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.\nHere are results of Stan runs with default parameters fit to \\(N=100\\) data points generated from \\(y_n \\sim \\textsf{normal}(0,1)\\):\nTwo Scale Parameters, Improper Prior\nInference for Stan model: improper_stan\nWarmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total\nSampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total\n\n                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat\nlp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0\nn_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0\nlambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2\nlambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2\nsigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1\nmu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0\nTwo Scale Parameters, Weak Prior\nWarmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total\nSampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total\n\n                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat\nlp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0\nn_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0\nlambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nlambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nsigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0\nmu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0\nOne Scale Parameter, Improper Prior\nWarmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total\nSampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total\n\n                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat\nlp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0\nn_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0\nmu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0\nsigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0\nOn the top is the non-identified model with improper uniform priors and data model \\(y_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2,\n\\sigma)\\).\nIn the middle is the same data model as in top plus priors \\(\\lambda_k \\sim \\textsf{normal}(0,10)\\).\nOn the bottom is an identified model with an improper prior, with data model \\(y_n \\sim \\textsf{normal}(\\mu,\\sigma)\\). All models estimate \\(\\mu\\) at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value \\(\\mu=0\\) is within the 90% posterior intervals in all three models.\n\n\n\nTo illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the collinearity section.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real lambda1;\n  real lambda2;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  real mu;\n  mu = lambda1 + lambda2;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nThe second adds priors to the model block for lambda1 and lambda2 to the previous model.\nlambda1 ~ normal(0, 10);\nlambda2 ~ normal(0, 10);\nThe third involves a single location parameter, but no priors.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nAll three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.\n\nAs indicated by R_hat column, all parameters have converged other than \\(\\lambda_1\\) and \\(\\lambda_2\\) in the non-identified model.\nThe average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.\nThe effective sample size per second for \\(\\mu\\) is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for \\(\\sigma\\).\nIn the non-identified model, the 95% interval for \\(\\lambda_1\\) is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.\nIn all three models, the simulated value of \\(\\mu=0\\) and \\(\\sigma=1\\) are well within the posterior 90% intervals.\n\nThe first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Simulation-Based Calibration"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/simulation-based-calibration.html#sbc-in-stan",
-    "href": "stan-users-guide/simulation-based-calibration.html#sbc-in-stan",
-    "title": "Simulation-Based Calibration",
+    "objectID": "stan-users-guide/problematic-posteriors.html#collinearity.section",
+    "href": "stan-users-guide/problematic-posteriors.html#collinearity.section",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "Running simulation-based calibration in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.\n\n\nFor illustration, a very simple model will suffice. Suppose there are two parameters \\((\\mu, \\sigma)\\) with independent priors, \\[\n\\mu \\sim \\textrm{normal}(0, 1),\n\\] and \\[\n\\sigma \\sim \\textrm{lognormal}(0, 1).\n\\] The data \\(y = y_1, \\ldots, y_N\\) is drawn conditionally independently given the parameters, \\[\ny_n \\sim \\textrm{normal}(\\mu, \\sigma).\n\\] The joint prior density is thus \\[\np(\\mu, \\sigma)\n= \\textrm{normal}(\\mu \\mid 0, 1)\n  \\cdot \\textrm{lognormal}(\\sigma \\mid 0, 1),\n\\] and the sampling density is \\[\np(y \\mid \\mu, \\sigma)\n= \\prod_{n=1}^N \\textrm{normal}(y_n \\mid \\mu, \\sigma).\n\\]\nFor example, suppose the following two parameter values are drawn from the prior in the first simulation, \\[\n(\\mu^{\\textrm{sim(1)}}, \\sigma^{\\textrm{sim(1)}}) = (1.01, 0.23).\n\\] Then data \\(y^{\\textrm{sim}(1)} \\sim p(y \\mid \\mu^{\\textrm{sim(1)}},\n\\sigma^{\\textrm{sim(1)}})\\) is drawn according to the sampling distribution. Next, \\(M = 4\\) draws are taken from the posterior \\(\\mu^{(1,m)}, \\sigma^{(1,m)} \\sim p(\\mu, \\sigma \\mid y^{\\textrm{sim}(1)})\\), \\[\n\\begin{array}{r|rr}\nm & \\mu^{(1,m)} & \\sigma^{(1,m)}\n\\\\ \\hline\n1 & 1.07 & 0.33\n\\\\\n2 & -0.32 & 0.14\n\\\\\n3 & -0.99 & 0.26\n\\\\\n4 & 1.51 & 0.31\n\\end{array}\n\\] Then the comparisons on which ranks are based look as follows, \\[\n\\begin{array}{r|cc}\nm & \\textrm{I}(\\mu^{(1,m)} &lt; \\mu^{\\textrm{sim}(1)})\n& \\textrm{I}(\\sigma^{(1,m)} &lt; \\sigma^{\\textrm{sim}(1)})\n\\\\ \\hline\n1 & 0 & 0\n\\\\\n2 & 1 & 1\n\\\\\n3 & 1 & 0\n\\\\\n4 & 0 & 0\n\\end{array}\n\\] The ranks are the column sums, \\(r_{1,1} = 2\\) and \\(r_{1,2} = 1\\). Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between \\(0\\) and \\(M\\), the number of posterior draws.\n\n\n\nTo code simulation-based calibration in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = normal_rng(mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nTo avoid confusion with the number of simulated data sets used for simulation-based calibration, J is used for the number of simulated data points.\nThe model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.\n\n\n\nThe entire simulation-based calibration process is as follows, where\n\np(theta) is the prior density\np(y | theta) is the sampling density\nK is the number of parameters\nN is the total number of simulated data sets and fits\nM is the number of posterior draws per simulated data set\n\nSBC(p(theta), p(y | theta), K, N, M)\n------------------------------------\nfor (n in 1:N) {\n    // simulate parameters and data\n    theta(sim(n)) ~ p(theta)\n    y(sim(n)) ~ p(y | theta(sim(n)))\n\n    // posterior draws given simulated data\n    for (m in 1:M) {\n        theta(n, m) ~ p(theta | y(sim(n)))\n    }\n    // calculate rank of sim among posterior draws\n    for (k in 1:K) {\n        rank(n, k) = SUM_m I(theta[k](n,m) &lt; theta[k](sim(n)))\n    }\n}\n// test uniformity of each parameter\nfor (k in 1:K) {\n    test uniformity of rank(1:N, k)\n}\n\n\n\nThe draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws. Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of M, then thinning down to M draws.",
+    "text": "This section discusses problems related to the classical notion of identifiability, which lead to ridges in the posterior density and wreak havoc with both sampling and inference.\n\n\n\n\nThe first example of collinearity is an artificial example involving redundant intercept parameters.1\nSuppose there are observations \\(y_n\\) for \\(n \\in \\{1,\\dotsc,N\\}\\), two intercept parameters \\(\\lambda_1\\) and \\(\\lambda_2\\), a scale parameter \\(\\sigma &gt; 0\\), and the data model \\[\ny_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2, \\sigma).\n\\]\nFor any constant \\(q\\), the sampling density for \\(y\\) does not change if we add \\(q\\) to \\(\\lambda_1\\) and subtract it from \\(\\lambda_2\\), i.e., \\[\np(y \\mid \\lambda_1, \\lambda_2,\\sigma)\n=\np(y \\mid \\lambda_1 + q, \\lambda_2 - q, \\sigma).\n\\]\nThe consequence is that an improper uniform prior \\(p(\\mu,\\sigma)\n\\propto 1\\) leads to an improper posterior. This impropriety arises because the neighborhoods around \\(\\lambda_1 + q, \\lambda_2 - q\\) have the same mass no matter what \\(q\\) is. Therefore, a sampler would need to spend as much time in the neighborhood of \\(\\lambda_1=1\\,000\\,000\\,000\\) and \\(\\lambda_2=-1\\,000\\,000\\,000\\) as it does in the neighborhood of \\(\\lambda_1=0\\) and \\(\\lambda_2=0\\), and so on for ever more far-ranging values.\nThe marginal posterior \\(p(\\lambda_1,\\lambda_2 \\mid y)\\) for this model is thus improper.2\nThe impropriety shows up visually as a ridge in the posterior density, as illustrated in the left-hand plot. The ridge for this model is along the line where \\(\\lambda_2 = \\lambda_1 + c\\) for some constant \\(c\\).\nContrast this model with a simple regression with a single intercept parameter \\(\\mu\\) and data model \\[\ny_n \\sim \\textsf{normal}(\\mu,\\sigma).\n\\] Even with an improper prior, the posterior is proper as long as there are at least two data points \\(y_n\\) with distinct values.\n\n\n\nConsider an item-response theory model for students \\(j \\in 1{:}J\\) with abilities \\(\\alpha_j\\) and test items \\(i \\in 1{:}I\\) with difficulties \\(\\beta_i\\). The observed data are an \\(I \\times J\\) array with entries \\(y_{i, j} \\in \\{ 0, 1 \\}\\) coded such that \\(y_{i, j} = 1\\) indicates that student \\(j\\) answered question \\(i\\) correctly. The sampling distribution for the data is \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\alpha_j - \\beta_i)).\n\\]\nFor any constant \\(c\\), the probability of \\(y\\) is unchanged by adding a constant \\(c\\) to all the abilities and subtracting it from all the difficulties, i.e., \\[\np(y \\mid \\alpha, \\beta)\n=\np(y \\mid \\alpha + c, \\beta - c).\n\\]\nThis leads to a multivariate version of the ridge displayed by the regression with two intercepts discussed above.\n\n\n\nThe general form of the collinearity problem arises when predictors for a regression are collinear. For example, consider a linear regression data model \\[\ny_n \\sim \\textsf{normal}(x_n \\beta, \\sigma)\n\\] for an \\(N\\)-dimensional observation vector \\(y\\), an \\(N \\times K\\) predictor matrix \\(x\\), and a \\(K\\)-dimensional coefficient vector \\(\\beta\\).\nNow suppose that column \\(k\\) of the predictor matrix is a multiple of column \\(k'\\), i.e., there is some constant \\(c\\) such that \\(x_{n,k} = c\n\\, x_{n,k'}\\) for all \\(n\\). In this case, the coefficients \\(\\beta_k\\) and \\(\\beta_{k'}\\) can covary without changing the predictions, so that for any \\(d \\neq 0\\), \\[\np(y \\mid \\ldots, \\beta_k, \\dotsc, \\beta_{k'}, \\dotsc, \\sigma)\n=\np(y \\mid \\ldots, d \\beta_k, \\dotsc, \\frac{d}{c} \\, \\beta_{k'}, \\dotsc,\n\\sigma).\n\\]\nEven if columns of the predictor matrix are not exactly collinear as discussed above, they cause similar problems for inference if they are nearly collinear.\n\n\n\nConsider adding a discrimination parameter \\(\\delta_i\\) for each question in an IRT model, with data model \\[\ny_{i, j} \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(\\delta_i(\\alpha_j - \\beta_i))).\n\\] For any constant \\(c \\neq 0\\), multiplying \\(\\delta\\) by \\(c\\) and dividing \\(\\alpha\\) and \\(\\beta\\) by \\(c\\) produces the same likelihood, \\[\np(y \\mid \\delta,\\alpha,\\beta)\n= p(y \\mid c \\delta, \\frac{1}{c}\\alpha, \\frac{1}{c}\\beta).\n\\] If \\(c &lt; 0\\), this switches the signs of every component in \\(\\alpha\\), \\(\\beta\\), and \\(\\delta\\) without changing the density.\n\n\n\nIn order to parameterize a \\(K\\)-simplex (i.e., a \\(K\\)-vector with non-negative values that sum to one), only \\(K - 1\\) parameters are necessary because the \\(K\\)th is just one minus the sum of the first \\(K\n- 1\\) parameters, so that if \\(\\theta\\) is a \\(K\\)-simplex, \\[\n\\theta_K = 1 - \\sum_{k=1}^{K-1} \\theta_k.\n\\]\nThe softmax function maps a \\(K\\)-vector \\(\\alpha\\) of linear predictors to a \\(K\\)-simplex \\(\\theta = \\texttt{softmax}(\\alpha)\\) by defining \\[\n\\theta_k = \\frac{\\exp(\\alpha_k)}{\\sum_{k'=1}^K \\exp(\\alpha_{k'})}.\n\\]\nThe softmax function is many-to-one, which leads to a lack of identifiability of the unconstrained parameters \\(\\alpha\\). In particular, adding or subtracting a constant from each \\(\\alpha_k\\) produces the same simplex \\(\\theta\\).\n\n\n\n\nAll of the examples discussed in the previous section allow translation or scaling of parameters while leaving the data probability density invariant. These problems can be mitigated in several ways.\n\n\nIn the case of the multiple intercepts, \\(\\lambda_1\\) and \\(\\lambda_2\\), the simplest solution is to remove the redundant intercept, resulting in a model with a single intercept parameter \\(\\mu\\) and sampling distribution \\(y_n \\sim \\textsf{normal}(\\mu, \\sigma)\\). The same solution works for solving the problem with collinearity—just remove one of the columns of the predictor matrix \\(x\\).\n\n\n\nThe IRT model without a discrimination parameter can be fixed by pinning one of its parameters to a fixed value, typically 0. For example, the first student ability \\(\\alpha_1\\) can be fixed to 0. Now all other student ability parameters can be interpreted as being relative to student 1. Similarly, the difficulty parameters are interpretable relative to student 1’s ability to answer them.\nThis solution is not sufficient to deal with the multiplicative invariance introduced by the question discrimination parameters \\(\\delta_i\\). To solve this problem, one of the discrimination parameters, say \\(\\delta_1\\), must also be constrained. Because it’s a multiplicative and not an additive invariance, it must be constrained to a non-zero value, with 1 being a convenient choice. Now all of the discrimination parameters may be interpreted relative to item 1’s discrimination.\nThe many-to-one nature of \\(\\texttt{softmax}(\\alpha)\\) is typically mitigated by pinning a component of \\(\\alpha\\), for instance fixing \\(\\alpha_K = 0\\). The resulting mapping is one-to-one from \\(K-1\\) unconstrained parameters to a \\(K\\)-simplex. This is roughly how simplex-constrained parameters are defined in Stan; see the reference manual chapter on constrained parameter transforms for a precise definition. The Stan code for creating a simplex from a \\(K-1\\)-vector can be written as\nvector softmax_id(vector alpha) {\n  vector[num_elements(alpha) + 1] alphac1;\n  for (k in 1:num_elements(alpha)) {\n    alphac1[k] = alpha[k];\n  }\n  alphac1[num_elements(alphac1)] = 0;\n  return softmax(alphac1);\n}\n\n\n\nSo far, the models have been discussed as if the priors on the parameters were improper uniform priors.\nA more general Bayesian solution to these invariance problems is to impose proper priors on the parameters. This approach can be used to solve problems arising from either additive or multiplicative invariance.\nFor example, normal priors on the multiple intercepts, \\[\n\\lambda_1, \\lambda_2 \\sim \\textsf{normal}(0,\\tau),\n\\] with a constant scale \\(\\tau\\), ensure that the posterior mode is located at a point where \\(\\lambda_1 = \\lambda_2\\), because this minimizes \\(\\log \\textsf{normal}(\\lambda_1 \\mid 0,\\tau) + \\log\n\\textsf{normal}(\\lambda_2 \\mid 0,\\tau)\\).3\nThe following plots show the posteriors for two intercept parameterization without prior, two intercept parameterization with standard normal prior, and one intercept reparameterization without prior. For all three cases, the posterior is plotted for 100 data points drawn from a standard normal.\nThe two intercept parameterization leads to an improper prior with a ridge extending infinitely to the northwest and southeast.\n\n\n\nTwo intercepts with improper prior\n\n\nAdding a standard normal prior for the intercepts results in a proper posterior.\n\n\n\nTwo intercepts with proper prior\n\n\nThe single intercept parameterization with no prior also has a proper posterior.\n\n\n\nSingle intercepts with improper prior\n\n\nThe addition of a prior to the two intercepts model is shown in the second plot; the final plot shows the result of reparameterizing to a single intercept.\nAn alternative strategy for identifying a \\(K\\)-simplex parameterization \\(\\theta = \\texttt{softmax}(\\alpha)\\) in terms of an unconstrained \\(K\\)-vector \\(\\alpha\\) is to place a prior on the components of \\(\\alpha\\) with a fixed location (that is, specifically avoid hierarchical priors with varying location). Unlike the approaching of pinning \\(\\alpha_K =\n0\\), the prior-based approach models the \\(K\\) outcomes symmetrically rather than modeling \\(K-1\\) outcomes relative to the \\(K\\)-th. The pinned parameterization, on the other hand, is usually more efficient statistically because it does not have the extra degree of (prior constrained) wiggle room.\n\n\n\nCare must be used when adding a prior to resolve invariances. If the prior is taken to be too broad (i.e., too vague), the resolution is in theory only, and samplers will still struggle.\nIdeally, a realistic prior will be formulated based on substantive knowledge of the problem being modeled. Such a prior can be chosen to have the appropriate strength based on prior knowledge. A strongly informative prior makes sense if there is strong prior information.\nWhen there is not strong prior information, a weakly informative prior strikes the proper balance between controlling computational inference without dominating the data in the posterior. In most problems, the modeler will have at least some notion of the expected scale of the estimates and be able to choose a prior for identification purposes that does not dominate the data, but provides sufficient computational control on the posterior.\nPriors can also be used in the same way to control the additive invariance of the IRT model. A typical approach is to place a strong prior on student ability parameters \\(\\alpha\\) to control scale simply to control the additive invariance of the basic IRT model and the multiplicative invariance of the model extended with a item discrimination parameters; such a prior does not add any prior knowledge to the problem. Then a prior on item difficulty can be chosen that is either informative or weakly informative based on prior knowledge of the problem.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Simulation-Based Calibration"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/simulation-based-calibration.html#testing-uniformity",
-    "href": "stan-users-guide/simulation-based-calibration.html#testing-uniformity",
-    "title": "Simulation-Based Calibration",
+    "objectID": "stan-users-guide/problematic-posteriors.html#label-switching-problematic.section",
+    "href": "stan-users-guide/problematic-posteriors.html#label-switching-problematic.section",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "A simple, though not very highly powered, \\(\\chi^2\\)-squared test for uniformity can be formulated by binning the ranks \\(0:M\\) into \\(J\\) bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, Cook, Gelman, and Rubin (2006) transform the ranks using the inverse cumulative distribution function for the standard normal and then perform a test for normality. Talts et al. (2018) recommend visual inspection of the binned plots.\nThe bins don’t need to be exactly the same size. In general, if \\(b_j\\) is the number of ranks that fall into bin \\(j\\) and \\(e_j\\) is the number of ranks expected to fall into bin \\(j\\) (which will be proportional to its size under uniformity), the test statistic is \\[\nX^2 = \\sum_{j = 1}^J \\frac{(b_j - e_j)^2}{e_j}.\n\\] The terms are approximately square standard normal, so that under the null hypothesis of uniformity, \\[\nX^2 \\sim \\textrm{chiSquared}(J - 1),\n\\] with the corresponding \\(p\\)-value given by the complementary cumulative distribution function (CCDF) of \\(\\textrm{chiSquared}(J - 1)\\) applied to \\(X^2\\). Because this test relies on the binomial being approximately normal, the traditional advice is to make sure the expected count in each bin is at least five, i.e., \\(e_j \\geq 5.\\)\n\n\nBecause there are \\(M + 1\\) possible ranks, with \\(J\\) bins, it is easiest to have \\(M + 1\\) be divisible by \\(J\\). For instance, if \\(J = 20\\) and \\(M = 999\\), then there are \\(1000\\) possible ranks and an expected count in each bin of \\(\\frac{M + 1}{J} = 50.\\)\nDistributing the ranks into bins is another fiddly operation that can be done with integer arithmetic or the floor operation. Using floor, the following function determines the bin for a rank, \\[\n\\textrm{bin}(r_{n, m}, M, J)\n= 1 + \\left\\lfloor \\frac{r_{n, m}}{(M + 1) / J} \\right\\rfloor.\n\\] For example, with \\(M = 999\\) and \\(J = 20\\), \\((M + 1) / J = 50\\). The lowest rank checks out, \\[\n\\textrm{bin}(0, 999, 20) = 1 + \\lfloor 0 / 50 \\rfloor = 1,\n\\] as does the 50th rank, \\[\n\\textrm{bin}(49, 999, 20) = 1 + \\lfloor 49 / 50 \\rfloor = 1,\n\\] and the 51st is appropriately put in the second bin, \\[\n\\textrm{bin}(50, 999, 20) = 1 + \\lfloor 50 / 50 \\rfloor = 2.\n\\] The highest rank also checks out, with \\(\\textrm{bin}(1000, 999, 20) = 50.\\)\nTo summarize, the following pseudocode computes the \\(b_j\\) values for the \\(\\chi^2\\) test or for visualization in a histogram.\nInputs: M draws, J bins, N parameters, ranks r[n, m]\nb[1:J] = 0\nfor (m in 1:M) {\n  ++b[1 + floor(r[n, m] * J / (M + 1))]\n}\nwhere the ++b[n] notation is a common form of syntactic sugar for b[n] = b[n] + 1.\nIn general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.",
+    "text": "Where collinearity in regression models can lead to infinitely many posterior maxima, swapping components in a mixture model leads to finitely many posterior maxima.\n\n\nConsider a normal mixture model with two location parameters \\(\\mu_1\\) and \\(\\mu_2\\), a shared scale \\(\\sigma &gt; 0\\), a mixture ratio \\(\\theta \\in\n[0,1]\\), and data model \\[\np(y \\mid \\theta,\\mu_1,\\mu_2,\\sigma)\n= \\prod_{n=1}^N \\big( \\theta \\, \\textsf{normal}(y_n \\mid \\mu_1,\\sigma)\n                       + (1 - \\theta) \\, \\textsf{normal}(y_n \\mid \\mu_2,\\sigma) \\big).\n\\] The issue here is exchangeability of the mixture components, because \\[\np(\\theta,\\mu_1,\\mu_2,\\sigma \\mid y) = p\\big((1-\\theta),\\mu_2,\\mu_1,\\sigma \\mid y\\big).\n\\] The problem is exacerbated as the number of mixture components \\(K\\) grows, as in clustering models, leading to \\(K!\\) identical posterior maxima.\n\n\n\nThe analysis of posterior convergence and effective sample size is also difficult for mixture models. For example, the \\(\\hat{R}\\) convergence statistic reported by Stan and the computation of effective sample size are both compromised by label switching. The problem is that the posterior mean, a key ingredient in these computations, is affected by label switching, resulting in a posterior mean for \\(\\mu_1\\) that is equal to that of \\(\\mu_2\\), and a posterior mean for \\(\\theta\\) that is always 1/2, no matter what the data are.\n\n\n\nIn some sense, the index (or label) of a mixture component is irrelevant. Posterior predictive inferences can still be carried out without identifying mixture components. For example, the log probability of a new observation does not depend on the identities of the mixture components. The only sound Bayesian inferences in such models are those that are invariant to label switching. Posterior means for the parameters are meaningless because they are not invariant to label switching; for example, the posterior mean for \\(\\theta\\) in the two component mixture model will always be 1/2.\n\n\n\nTheoretically, this should not present a problem for inference because all of the integrals involved in posterior predictive inference will be well behaved. The problem in practice is computation.\nBeing able to carry out such invariant inferences in practice is an altogether different matter. It is almost always intractable to find even a single posterior mode, much less balance the exploration of the neighborhoods of multiple local maxima according to the probability masses. In Gibbs sampling, it is unlikely for \\(\\mu_1\\) to move to a new mode when sampled conditioned on the current values of \\(\\mu_2\\) and \\(\\theta\\). For HMC and NUTS, the problem is that the sampler gets stuck in one of the two “bowls” around the modes and cannot gather enough energy from random momentum assignment to move from one mode to another.\nEven with a proper posterior, all known sampling and inference techniques are notoriously ineffective when the number of modes grows super-exponentially as it does for mixture models with increasing numbers of components.\n\n\n\nSeveral hacks (i.e., “tricks”) have been suggested and employed to deal with the problems posed by label switching in practice.\n\n\nOne common strategy is to impose a constraint on the parameters that identifies the components. For instance, we might consider constraining \\(\\mu_1 &lt; \\mu_2\\) in the two-component normal mixture model discussed above. A problem that can arise from such an approach is when there is substantial probability mass for the opposite ordering \\(\\mu_1 &gt; \\mu_2\\). In these cases, the posteriors are affected by the constraint and true posterior uncertainty in \\(\\mu_1\\) and \\(\\mu_2\\) is not captured by the model with the constraint. In addition, standard approaches to posterior inference for event probabilities is compromised. For instance, attempting to use \\(M\\) posterior draws to estimate \\(\\Pr[\\mu_1 &gt; \\mu_2]\\), will fail, because the estimator \\[\n\\Pr[\\mu_1 &gt; \\mu_2]\n\\approx\n\\sum_{m=1}^M \\textrm{I}\\left(\\mu_1^{(m)} &gt; \\mu_2^{(m)}\\right)\n\\] will result in an estimate of 0 because the posterior respects the constraint in the model.\n\n\n\nAnother common approach is to run a single chain or to initialize the parameters near realistic values.4\nThis can work better than the hard constraint approach if reasonable initial values can be found and the labels do not switch within a Markov chain. The result is that all chains are glued to a neighborhood of a particular mode in the posterior.",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Simulation-Based Calibration"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/simulation-based-calibration.html#examples-of-simulation-based-calibration",
-    "href": "stan-users-guide/simulation-based-calibration.html#examples-of-simulation-based-calibration",
-    "title": "Simulation-Based Calibration",
+    "objectID": "stan-users-guide/problematic-posteriors.html#component-collapsing-in-mixture-models",
+    "href": "stan-users-guide/problematic-posteriors.html#component-collapsing-in-mixture-models",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "This section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.\n\n\nConsider the following simple model for a normal distribution with standard normal and lognormal priors on the location and scale parameters. \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 1)\n\\\\[4pt]\n\\sigma & \\sim & \\textrm{lognormal}(0, 1)\n\\\\[4pt]\ny_{1:10} & \\sim & \\textrm{normal}(\\mu, \\sigma).\n\\end{eqnarray*}\\] The Stan program for evaluating SBC for this model is\ntransformed data {\n  real mu_sim = normal_rng(0, 1);\n  real&lt;lower=0&gt; sigma_sim = lognormal_rng(0, 1);\n\n  int&lt;lower=0&gt; J = 10;\n  vector[J] y_sim;\n  for (j in 1:J) {\n    y_sim[j] = student_t_rng(4, mu_sim, sigma_sim);\n  }\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  mu ~ normal(0, 1);\n  sigma ~ lognormal(0, 1);\n\n  y_sim ~ normal(mu, sigma);\n}\ngenerated quantities {\n  array[2] int&lt;lower=0, upper=1&gt; I_lt_sim\n      = { mu &lt; mu_sim, sigma &lt; sigma_sim };\n}\nAfter running this for enough iterations so that the effective sample size is larger than \\(M\\), then thinning to \\(M\\) draws (here \\(M = 999\\)), the ranks are computed and binned, and then plotted.\n\n\n\n\n\n\nFigure 1: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location, standard lognormal prior on the scale. Both histograms appear uniform, which is consistent with inference being well calibrated.\n\n\n\n\n\n\nNow consider using a Student-t data generating process with a normal model. Compare the apparent uniformity of the well specified model with the ill-specified situation with Student-t generative process and normal model.\n\n\n\n\n\n\nFigure 2: Simulation based calibration plots for location and scale of a normal model with standard normal prior on the location standard lognormal prior on the scale with mismatched generative model using a Student-t data model with 4 degrees of freedom. The mean histogram appears uniform, but the scale parameter shows simulated values much smaller than fit values, clearly signaling the lack of calibration.\n\n\n\n\n\n\nThe example in the previous sections show hard-coded pathological behavior. The usual application of SBC is to diagnose problems with a sampler.\nThis can happen in Stan with well-specified models if the posterior geometry is too difficult (usually due to extreme stiffness that varies). A simple example is the eight schools problem, the data for which consists of sample means \\(y_j\\) and standard deviations \\(\\sigma_j\\) of differences in test score after the same intervention in \\(J = 8\\) different schools. Rubin (1981) applies a hierarchical model for a meta-analysis of the results, estimating the mean intervention effect and a varying effect for each school. With a standard parameterization and weak priors, this model has very challenging posterior geometry, as shown by Talts et al. (2018); this section replicates their results.\nThe meta-analysis model has parameters for a population mean \\(\\mu\\) and standard deviation \\(\\tau &gt; 0\\) as well as the effect \\(\\theta_j\\) of the treatment in each school. The model has weak normal and half-normal priors for the population-level parameters, \\[\\begin{eqnarray*}\n\\mu & \\sim & \\textrm{normal}(0, 5)\n\\\\[4pt]\n\\tau & \\sim & \\textrm{normal}_{+}(0, 5).\n\\end{eqnarray*}\\] School level effects are modeled as normal given the population parameters, \\[\n\\theta_j \\sim \\textrm{normal}(\\mu, \\tau).\n\\] The data is modeled as in a meta-analysis, given the school effect and sample standard deviation in the school, \\[\ny_j \\sim \\textrm{normal}(\\theta_j, \\sigma_j).\n\\]\nThis model can be coded in Stan with a data-generating process that simulates the parameters and then simulates data according to the parameters.\ntransformed data {\n  real mu_sim = normal_rng(0, 5);\n  real tau_sim = abs(normal_rng(0, 5));\n  int&lt;lower=0&gt; J = 8;\n  array[J] real theta_sim = normal_rng(rep_vector(mu_sim, J), tau_sim);\n  array[J] real&lt;lower=0&gt; sigma = abs(normal_rng(rep_vector(0, J), 5));\n  array[J] real y = normal_rng(theta_sim, sigma);\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; tau;\n  array[J] real theta;\n}\nmodel {\n  tau ~ normal(0, 5);\n  mu ~ normal(0, 5);\n  theta ~ normal(mu, tau);\n  y ~ normal(theta, sigma);\n}\ngenerated quantities {\n  int&lt;lower=0, upper=1&gt; mu_lt_sim = mu &lt; mu_sim;\n  int&lt;lower=0, upper=1&gt; tau_lt_sim = tau &lt; tau_sim;\n  int&lt;lower=0, upper=1&gt; theta1_lt_sim = theta[1] &lt; theta_sim[1];\n}\nAs usual for simulation-based calibration, the transformed data encodes the data-generating process using random number generators. Here, the population parameters \\(\\mu\\) and \\(\\tau\\) are first simulated, then the school-level effects \\(\\theta\\), and then finally the observed data \\(\\sigma_j\\) and \\(y_j.\\) The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only \\(\\theta_1\\) because the schools are exchangeable in the simulation.\nWhen fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.\nSBC also diagnoses the problem. Here’s the rank plots for running \\(N =\n200\\) simulations with 1000 warmup iterations and \\(M = 999\\) draws per simulation used to compute the ranks.\n\n\n\n\n\n\n\n\n\n\n\n(a) \\(\\mu\\)\n\n\n\n\n\n\n\n\n\n\n\n(b) \\(\\tau\\)\n\n\n\n\n\n\n\n\n\n\n\n(c) \\({\\theta}_1\\)\n\n\n\n\n\n\n\nFigure 3: Simulation based calibration plots for the eight-schools model with centered parameterization in Stan. The geometry is too difficult for the NUTS sampler to handle, as indicated by the plot for \\(\\theta_1\\) (Figure 3 (c)).\n\n\n\nAlthough the population mean and standard deviation \\(\\mu\\) and \\(\\tau\\) appear well calibrated, \\(\\theta_1\\) tells a very different story. The simulated values are much smaller than the values fit from the data. This is because Stan’s no-U-turn sampler is unable to sample with the model formulated in the centered parameterization—the posterior geometry has regions of extremely high curvature as \\(\\tau\\) approaches zero and the \\(\\theta_j\\) become highly constrained. The chapter on reparameterization explains how to remedy this problem and fit this kind of hierarchical model with Stan.",
+    "text": "It is possible for two mixture components in a mixture model to collapse to the same values during sampling or optimization. For example, a mixture of \\(K\\) normals might devolve to have \\(\\mu_i =\n\\mu_j\\) and \\(\\sigma_i = \\sigma_j\\) for \\(i \\neq j\\).\nThis will typically happen early in sampling due to initialization in MCMC or optimization or arise from random movement during MCMC. Once the parameters match for a given draw \\((m)\\), it can become hard to escape because there can be a trough of low-density mass between the current parameter values and the ones without collapsed components.\nIt may help to use a smaller step size during warmup, a stronger prior on each mixture component’s membership responsibility. A more extreme measure is to include additional mixture components to deal with the possibility that some of them may collapse.\nIn general, it is difficult to recover exactly the right \\(K\\) mixture components in a mixture model as \\(K\\) increases beyond one (yes, even a two-component mixture can have this problem).",
     "crumbs": [
       "Stan Users Guide",
-      "Posterior Inference & Model Checking",
-      "Simulation-Based Calibration"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html",
-    "href": "stan-users-guide/style-guide.html",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-densities",
+    "href": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-densities",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "This chapter describes the preferred style for laying out Stan models. These are not rules of the language, but simply recommendations for laying out programs in a text editor. Although these recommendations may seem arbitrary, they are similar to those of many teams for many programming languages. Like rules for typesetting text, the goal is to achieve readability without wasting white space either vertically or horizontally. This is the style used in the Stan documentation, and should align with the auto-formatting ability of stanc3.\n\n\nThe most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.\n\n\n\nLine lengths should not exceed 80 characters.1\nThis is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.\n\n\n\nThe recommended file extension for Stan model files is .stan. Files which contain only function definitions (intended for use with #include) should be given the .stanfunctions extension. A .stanfunctions file only includes the function definition and does not require the functions{} block wrapped around the function. A simple example of usage where the function is defined and saved in the file foo.stanfunctions:\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe function foo can be accessed in the Stan program by including the path to the foo.stanfunctions file as:\nfunctions {\n  #include foo.stanfunctions;\n}\n// ...body...\nFor Stan data dump files, the recommended extension is .R, or more informatively, .data.R. For JSON output, the recommended extension is .json.\n\n\n\nThe recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (_) used as a separator. Thus it is preferred to use sigma_y, rather than the run together sigmay, camel-case sigmaY, or capitalized camel-case SigmaY. An exception is often made for terms appearing in mathematical expressions with standard names, like A for a matrix.\nAnother exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an \\(M \\times N\\) matrix \\(a\\) would look as follows.\nfor (m in 1:M) {\n  for (n in 1:N) {\n     a[m, n] = ...\n  }\n}\n\n\n\nDeclaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.\nThe following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of mu in each iteration.\nmodel {\n  array[N] real mu;\n  for (n in 1:N) {\n    mu[n] = alpha * x[n] + beta;\n    y[n] ~ normal(mu[n],sigma);\n  }\n}\nBecause variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.\nmodel {\n  for (n in 1:N) {\n    real mu;\n    mu = alpha * x[n] + beta;\n    y[n] ~ normal(mu,sigma);\n  }\n}\nThe local variable can be eliminated altogether, as follows.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(alpha * x[n] + beta, sigma);\n  }\n}\nThere is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.\n\n\nIn the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.\nmodel {\n  vector[K] mu;\n  for (n in 1:N) {\n    for (k in 1:K) {\n      mu[k] = // ...\n    }\n    y[n] ~ multi_normal(mu,Sigma);\n}\nIn this case, the vector mu will be allocated outside of both loops, and used a total of N times.\n\n\n\n\n\n\nSingle-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThe use of loops and conditionals without brackets can be dangerous. For instance, consider this program.\nfor (n in 1:N)\n  z[n] ~ normal(nu,1);\n  y[n] ~ normal(mu,1);\nBecause Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.\nfor (n in 1:N) {\n  z[n] ~ normal(nu,1);\n}\ny[n] ~ normal(mu,1);\nTherefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the else branch contains exactly one conditional. Then, it is preferred to place the following if on the same line, as in the following.\nif (x) {\n  // ...\n} else if (y) {\n  // ...\n} else {\n  // ...\n}\n\n\n\nThe preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression a + b * c is preferred to the equivalent a + (b * c) or (a + (b * c)). The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.\nSimilarly, comparison operators can usually be written with minimal bracketing, with the form y[n] &gt; 0 || x[n] != 0 preferred to the bracketed form (y[n] &gt; 0) || (x[n] != 0).\n\n\n\nVertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThis also goes for parameters blocks, transformed data blocks, which should look as follows.\ntransformed parameters {\n  real sigma;\n  // ...\n}\nThe exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.\n\n\n\n\nWhile Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued x, one should use\nif (x != 0) { ...\nin place of\nif (x) { ...\nBeyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.\n\n\n\nFunctions are laid out the same way as in languages such as Java and C++. For example,\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.\nIf function names or argument lists are long, they can be written as\nmatrix\nfunction_to_do_some_hairy_algebra(matrix thingamabob,\n                                  vector doohickey2) {\n  // ...body...\n}\nThe function starts a new line, under the type. The arguments are aligned under each other.\nFunction documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the documenting functions section.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows correspond to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\n}\nThe open comment is /**, asterisks are aligned below the first asterisk of the open comment, and the end comment */ is also aligned on the asterisk. The tags @param and @return are used to label function arguments (i.e., parameters) and return values.\n\n\n\nStan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII 0x20), line feed (ASCII 0x0A), carriage return (0x0D), and tab (0x09). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.\n\n\nEach statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax\nreal mu, sigma;\n\n\n\nStan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.\n\n\n\nStan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.\n\n\n\nUse a space after ifs. For instance, use if (x &lt; y) {..., not if(x &lt; y){ ....\n\n\n\nThere should not be space between a function name and the arguments it applies to. For instance, use normal(0, 1), not normal (0,1).\n\n\n\nThere should be spaces around binary operators. For instance, use y[1] = x, not y[1]=x, use (x + y) * z not (x+y)*z.\nUnary operators are written without a space, such as in -x, !y.\n\n\n\nAnother exception to the above rule is when the assignment operator (=) is used inside a type constraint, such as\nreal&lt;lower=0&gt; x;\nSpaces should still be used in arithmetic and following commas, as in\nreal&lt;lower=0, upper=a * x + b&gt; x;\n\n\n\nSometimes expressions are too long to fit on a single line. In that case, the recommended form is to break before an operator,2 aligning the operator to a term above to indicate scoping. For example, use the following form\nvector[J] p_distance = Phi((distance_tolerance - overshot)\n                           ./ ((x + overshot) * sigma_distance))\n                       - Phi((-overshot)\n                             ./ ((x + overshot) * sigma_distance));\nHere, the elementwise division operator (./) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (Phi).\nFor functions with multiple arguments, break after a comma and line the next argument up underneath as follows.\ny[n] ~ normal(alpha + beta * x + gamma * y,\n              pow(tau,-0.5));\n\n\n\nCommas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.\nFor example,\nnormal(alpha * x[n] + beta, sigma);\nis preferred over\nnormal(alpha * x[n] + beta,sigma);\n\n\n\nWherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.\n\n\nNewlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point 0x0A). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point 0x0D) followed by a line-feed (LF) character.",
+    "text": "In some cases, the posterior density grows without bounds as parameters approach certain poles or boundaries. In such, there are no posterior modes and numerical stability issues can arise as sampled parameters approach constraint boundaries.\n\n\nOne such example is a binary mixture model with scales varying by component, \\(\\sigma_1\\) and \\(\\sigma_2\\) for locations \\(\\mu_1\\) and \\(\\mu_2\\). In this situation, the density grows without bound as \\(\\sigma_1 \\rightarrow 0\\) and \\(\\mu_1 \\rightarrow y_n\\) for some \\(n\\); that is, one of the mixture components concentrates all of its mass around a single data item \\(y_n\\).\n\n\n\nAnother example of unbounded densities arises with a posterior such as \\(\\textsf{beta}(\\phi \\mid 0.5,0.5)\\), which can arise if seemingly weak beta priors are used for groups that have no data. This density is unbounded as \\(\\phi \\rightarrow 0\\) and \\(\\phi \\rightarrow 1\\). Similarly, a Bernoulli data model coupled with a “weak” beta prior, leads to a posterior \\[\\begin{align*}\np(\\phi \\mid y)\n&\\propto\n   \\textsf{beta}(\\phi \\mid 0.5,0.5) \\times \\prod_{n=1}^N \\textsf{Bernoulli}(y_n \\mid \\phi) \\\\\n&=\n   \\textsf{beta}\\left(\\phi \\,\\middle|\\, 0.5 + \\sum_{n=1}^N y_n, 0.5 + N - \\sum_{n=1}^N y_n\\right).\n\\end{align*}\\]\nIf \\(N = 9\\) and each \\(y_n = 1\\), the posterior is \\(\\textsf{beta}(\\phi \\mid 9.5,0.5)\\). This posterior is unbounded as \\(\\phi\n\\rightarrow 1\\). Nevertheless, the posterior is proper, and although there is no posterior mode, the posterior mean is well-defined with a value of exactly 0.95.\n\n\nStan does not sample directly on the constrained \\((0,1)\\) space for this problem, so it doesn’t directly deal with unconstrained density values. Rather, the probability values \\(\\phi\\) are logit-transformed to \\((-\\infty,\\infty)\\). The boundaries at 0 and 1 are pushed out to \\(-\\infty\\) and \\(\\infty\\) respectively. The Jacobian adjustment that Stan automatically applies ensures the unconstrained density is proper. The adjustment for the particular case of \\((0,1)\\) is \\(\\log\n\\operatorname{logit}^{-1}(\\phi) + \\log \\operatorname{logit}(1 - \\phi)\\).\nThere are two problems that still arise, though. The first is that if the posterior mass for \\(\\phi\\) is near one of the boundaries, the logit-transformed parameter will have to sweep out long paths and thus can dominate the U-turn condition imposed by the no-U-turn sampler (NUTS). The second issue is that the inverse transform from the unconstrained space to the constrained space can underflow to 0 or overflow to 1, even when the unconstrained parameter is not infinite. Similar problems arise for the expectation terms in logistic regression, which is why the logit-scale parameterizations of the Bernoulli and binomial distributions are more stable.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#choose-a-consistent-style",
-    "href": "stan-users-guide/style-guide.html#choose-a-consistent-style",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-parameters",
+    "href": "stan-users-guide/problematic-posteriors.html#posteriors-with-unbounded-parameters",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "The most important point of style is consistency. Consistent coding style makes it easier to read not only a single program, but multiple programs. So when departing from this style guide, the number one recommendation is to do so consistently.",
+    "text": "In some cases, the posterior density will not grow without bound, but parameters will grow without bound with gradually increasing density values. Like the models discussed in the previous section that have densities that grow without bound, such models also have no posterior modes.\n\n\nConsider a logistic regression model with \\(N\\) observed outcomes \\(y_n\n\\in \\{ 0, 1 \\}\\), an \\(N \\times K\\) matrix \\(x\\) of predictors, a \\(K\\)-dimensional coefficient vector \\(\\beta\\), and data model \\[\ny_n \\sim \\textsf{Bernoulli}(\\operatorname{logit}^{-1}(x_n \\beta)).\n\\] Now suppose that column \\(k\\) of the predictor matrix is such that \\(x_{n,k} &gt; 0\\) if and only if \\(y_n = 1\\), a condition known as “separability.” In this case, predictive accuracy on the observed data continue to improve as \\(\\beta_k \\rightarrow \\infty\\), because for cases with \\(y_n = 1\\), \\(x_n \\beta \\rightarrow \\infty\\) and hence \\(\\operatorname{logit}^{-1}(x_n \\beta) \\rightarrow 1\\).\nWith separability, there is no maximum to the likelihood and hence no maximum likelihood estimate. From the Bayesian perspective, the posterior is improper and therefore the marginal posterior mean for \\(\\beta_k\\) is also not defined. The usual solution to this problem in Bayesian models is to include a proper prior for \\(\\beta\\), which ensures a proper posterior.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#line-length",
-    "href": "stan-users-guide/style-guide.html#line-length",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/problematic-posteriors.html#uniform-posteriors",
+    "href": "stan-users-guide/problematic-posteriors.html#uniform-posteriors",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "Line lengths should not exceed 80 characters.1\nThis is a typical recommendation for many programming language style guides because it makes it easier to lay out text edit windows side by side and to view the code on the web without wrapping, easier to view diffs from version control, etc. About the only thing that is sacrificed is laying out expressions on a single line.",
+    "text": "Suppose your model includes a parameter \\(\\psi\\) that is defined on \\([0,1]\\) and is given a flat prior \\(\\textsf{uniform}(\\psi \\mid 0,1)\\). Now if the data don’t tell us anything about \\(\\psi\\), the posterior is also \\(\\textsf{uniform}(\\psi \\mid 0,1)\\).\nAlthough there is no maximum likelihood estimate for \\(\\psi\\), the posterior is uniform over a closed interval and hence proper. In the case of a uniform posterior on \\([0,1]\\), the posterior mean for \\(\\psi\\) is well-defined with value \\(1/2\\). Although there is no posterior mode, posterior predictive inference may nevertheless do the right thing by simply integrating (i.e., averaging) over the predictions for \\(\\psi\\) at all points in \\([0,1]\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#file-extensions",
-    "href": "stan-users-guide/style-guide.html#file-extensions",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/problematic-posteriors.html#sampling-difficulties-with-problematic-priors",
+    "href": "stan-users-guide/problematic-posteriors.html#sampling-difficulties-with-problematic-priors",
+    "title": "Problematic Posteriors",
     "section": "",
-    "text": "The recommended file extension for Stan model files is .stan. Files which contain only function definitions (intended for use with #include) should be given the .stanfunctions extension. A .stanfunctions file only includes the function definition and does not require the functions{} block wrapped around the function. A simple example of usage where the function is defined and saved in the file foo.stanfunctions:\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe function foo can be accessed in the Stan program by including the path to the foo.stanfunctions file as:\nfunctions {\n  #include foo.stanfunctions;\n}\n// ...body...\nFor Stan data dump files, the recommended extension is .R, or more informatively, .data.R. For JSON output, the recommended extension is .json.",
+    "text": "With an improper posterior, it is theoretically impossible to properly explore the posterior. However, Gibbs sampling as performed by BUGS and JAGS, although still unable to properly sample from such an improper posterior, behaves differently in practice than the Hamiltonian Monte Carlo sampling performed by Stan when faced with an example such as the two intercept model discussed in the collinearity section and illustrated in the non-identifiable density plot.\n\n\nGibbs sampling, as performed by BUGS and JAGS, may appear to be efficient and well behaved for this unidentified model, but as discussed in the previous subsection, will not actually explore the posterior properly.\nConsider what happens with initial values \\(\\lambda_1^{(0)}, \\lambda_2^{(0)}\\). Gibbs sampling proceeds in iteration \\(m\\) by drawing \\[\\begin{align*}\n\\lambda_1^{(m)} &\\sim p(\\lambda_1 \\mid \\lambda_2^{(m-1)}, \\sigma^{(m-1)},  y) \\\\\n\\lambda_2^{(m)} &\\sim p(\\lambda_2 \\mid \\lambda_1^{(m)},   \\sigma^{(m-1)},  y) \\\\\n\\sigma^{(m)}    &\\sim p(\\sigma    \\mid \\lambda_1^{(m)},   \\lambda_2^{(m)}, y).\n\\end{align*}\\]\nNow consider the draw for \\(\\lambda_1\\) (the draw for \\(\\lambda_2\\) is symmetric), which is conjugate in this model and thus can be done efficiently. In this model, the range from which the next \\(\\lambda_1\\) can be drawn is highly constrained by the current values of \\(\\lambda_2\\) and \\(\\sigma\\). Gibbs will run quickly and provide seemingly reasonable inferences for \\(\\lambda_1 + \\lambda_2\\). But it will not explore the full range of the posterior; it will merely take a slow random walk from the initial values. This random walk behavior is typical of Gibbs sampling when posteriors are highly correlated and the primary reason to prefer Hamiltonian Monte Carlo to Gibbs sampling for models with parameters correlated in the posterior.\n\n\n\nHamiltonian Monte Carlo (HMC), as performed by Stan, is much more efficient at exploring posteriors in models where parameters are correlated in the posterior. In this particular example, the Hamiltonian dynamics (i.e., the motion of a fictitious particle given random momentum in the field defined by the negative log posterior) is going to run up and down along the valley defined by the potential energy (ridges in log posteriors correspond to valleys in potential energy). In practice, even with a random momentum for \\(\\lambda_1\\) and \\(\\lambda_2\\), the gradient of the log posterior is going to adjust for the correlation and the simulation will run \\(\\lambda_1\\) and \\(\\lambda_2\\) in opposite directions along the valley corresponding to the ridge in the posterior log density.\n\n\n\nStan’s default no-U-turn sampler (NUTS), is even more efficient at exploring the posterior (see Hoffman and Gelman 2014). NUTS simulates the motion of the fictitious particle representing the parameter values until it makes a U-turn, it will be defeated in most cases, as it will just move down the potential energy valley indefinitely without making a U-turn. What happens in practice is that the maximum number of leapfrog steps in the simulation will be hit in many of the iterations, causing a large number of log probability and gradient evaluations (1000 if the max tree depth is set to 10, as in the default). Thus sampling will appear to be slow. This is indicative of an improper posterior, not a bug in the NUTS algorithm or its implementation. It is simply not possible to sample from an improper posterior! Thus the behavior of HMC in general and NUTS in particular should be reassuring in that it will clearly fail in cases of improper posteriors, resulting in a clean diagnostic of sweeping out large paths in the posterior.\nHere are results of Stan runs with default parameters fit to \\(N=100\\) data points generated from \\(y_n \\sim \\textsf{normal}(0,1)\\):\nTwo Scale Parameters, Improper Prior\nInference for Stan model: improper_stan\nWarmup took (2.7, 2.6, 2.9, 2.9) seconds, 11 seconds total\nSampling took (3.4, 3.7, 3.6, 3.4) seconds, 14 seconds total\n\n                  Mean     MCSE   StdDev        5%       95%  N_Eff  N_Eff/s  R_hat\nlp__          -5.3e+01  7.0e-02  8.5e-01  -5.5e+01  -5.3e+01    150       11    1.0\nn_leapfrog__   1.4e+03  1.7e+01  9.2e+02   3.0e+00   2.0e+03   2987      212    1.0\nlambda1        1.3e+03  1.9e+03  2.7e+03  -2.3e+03   6.0e+03    2.1     0.15    5.2\nlambda2       -1.3e+03  1.9e+03  2.7e+03  -6.0e+03   2.3e+03    2.1     0.15    5.2\nsigma          1.0e+00  8.5e-03  6.2e-02   9.5e-01   1.2e+00     54      3.9    1.1\nmu             1.6e-01  1.9e-03  1.0e-01  -8.3e-03   3.3e-01   2966      211    1.0\nTwo Scale Parameters, Weak Prior\nWarmup took (0.40, 0.44, 0.40, 0.36) seconds, 1.6 seconds total\nSampling took (0.47, 0.40, 0.47, 0.39) seconds, 1.7 seconds total\n\n                 Mean     MCSE   StdDev        5%    95%  N_Eff  N_Eff/s  R_hat\nlp__              -54  4.9e-02  1.3e+00  -5.7e+01    -53    728      421    1.0\nn_leapfrog__      157  2.8e+00  1.5e+02   3.0e+00    511   3085     1784    1.0\nlambda1          0.31  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nlambda2         -0.14  2.8e-01  7.1e+00  -1.2e+01     12    638      369    1.0\nsigma             1.0  2.6e-03  8.0e-02   9.2e-01    1.2    939      543    1.0\nmu               0.16  1.8e-03  1.0e-01  -8.1e-03   0.33   3289     1902    1.0\nOne Scale Parameter, Improper Prior\nWarmup took (0.011, 0.012, 0.011, 0.011) seconds, 0.044 seconds total\nSampling took (0.017, 0.020, 0.020, 0.019) seconds, 0.077 seconds total\n\n                Mean     MCSE  StdDev        5%   50%   95%  N_Eff  N_Eff/s  R_hat\nlp__             -54  2.5e-02    0.91  -5.5e+01   -53   -53   1318    17198    1.0\nn_leapfrog__     3.2  2.7e-01     1.7   1.0e+00   3.0   7.0     39      507    1.0\nmu              0.17  2.1e-03    0.10  -3.8e-03  0.17  0.33   2408    31417    1.0\nsigma            1.0  1.6e-03   0.071   9.3e-01   1.0   1.2   2094    27321    1.0\nOn the top is the non-identified model with improper uniform priors and data model \\(y_n \\sim \\textsf{normal}(\\lambda_1 + \\lambda_2,\n\\sigma)\\).\nIn the middle is the same data model as in top plus priors \\(\\lambda_k \\sim \\textsf{normal}(0,10)\\).\nOn the bottom is an identified model with an improper prior, with data model \\(y_n \\sim \\textsf{normal}(\\mu,\\sigma)\\). All models estimate \\(\\mu\\) at roughly 0.16 with low Monte Carlo standard error, but a high posterior standard deviation of 0.1; the true value \\(\\mu=0\\) is within the 90% posterior intervals in all three models.\n\n\n\nTo illustrate the issues with sampling from non-identified and only weakly identified models, we fit three models with increasing degrees of identification of their parameters. The posteriors for these models is illustrated in the non-identifiable density plot. The first model is the unidentified model with two location parameters and no priors discussed in the collinearity section.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real lambda1;\n  real lambda2;\n  real&lt;lower=0&gt; sigma;\n}\ntransformed parameters {\n  real mu;\n  mu = lambda1 + lambda2;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nThe second adds priors to the model block for lambda1 and lambda2 to the previous model.\nlambda1 ~ normal(0, 10);\nlambda2 ~ normal(0, 10);\nThe third involves a single location parameter, but no priors.\ndata {\n  int N;\n  array[N] real y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma);\n}\nAll three of the example models were fit in Stan 2.1.0 with default parameters (1000 warmup iterations, 1000 sampling iterations, NUTS sampler with max tree depth of 10). The results are shown in the non-identified fits figure. The key statistics from these outputs are the following.\n\nAs indicated by R_hat column, all parameters have converged other than \\(\\lambda_1\\) and \\(\\lambda_2\\) in the non-identified model.\nThe average number of leapfrog steps is roughly 3 in the identified model, 150 in the model identified by a weak prior, and 1400 in the non-identified model.\nThe effective sample size per second for \\(\\mu\\) is roughly 31,000 in the identified model, 1,900 in the model identified with weakly informative priors, and 200 in the non-identified model; the results are similar for \\(\\sigma\\).\nIn the non-identified model, the 95% interval for \\(\\lambda_1\\) is (-2300,6000), whereas it is only (-12,12) in the model identified with weakly informative priors.\nIn all three models, the simulated value of \\(\\mu=0\\) and \\(\\sigma=1\\) are well within the posterior 90% intervals.\n\nThe first two points, lack of convergence and hitting the maximum number of leapfrog steps (equivalently maximum tree depth) are indicative of improper posteriors. Thus rather than covering up the problem with poor sampling as may be done with Gibbs samplers, Hamiltonian Monte Carlo tries to explore the posterior and its failure is a clear indication that something is amiss in the model.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#variable-naming",
-    "href": "stan-users-guide/style-guide.html#variable-naming",
-    "title": "Stan Program Style Guide",
-    "section": "",
-    "text": "The recommended variable naming is to follow C/C++ naming conventions, in which variables are lowercase, with the underscore character (_) used as a separator. Thus it is preferred to use sigma_y, rather than the run together sigmay, camel-case sigmaY, or capitalized camel-case SigmaY. An exception is often made for terms appearing in mathematical expressions with standard names, like A for a matrix.\nAnother exception to the lowercasing recommendation, which follows the C/C++ conventions, is for size constants, for which the recommended form is a single uppercase letter. The reason for this is that it allows the loop variables to match. So loops over the indices of an \\(M \\times N\\) matrix \\(a\\) would look as follows.\nfor (m in 1:M) {\n  for (n in 1:N) {\n     a[m, n] = ...\n  }\n}",
+    "objectID": "stan-users-guide/problematic-posteriors.html#footnotes",
+    "href": "stan-users-guide/problematic-posteriors.html#footnotes",
+    "title": "Problematic Posteriors",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nThis example was raised by Richard McElreath on the Stan users group in a query about the difference in behavior between Gibbs sampling as used in BUGS and JAGS and the Hamiltonian Monte Carlo (HMC) and no-U-turn samplers (NUTS) used by Stan.↩︎\nThe marginal posterior \\(p(\\sigma \\mid y)\\) for \\(\\sigma\\) is proper here as long as there are at least two distinct data points.↩︎\nA Laplace prior (or an L1 regularizer for penalized maximum likelihood estimation) is not sufficient to remove this additive invariance. It provides shrinkage, but does not in and of itself identify the parameters because adding a constant to \\(\\lambda_1\\) and subtracting it from \\(\\lambda_2\\) results in the same value for the prior density.↩︎\nTempering methods may be viewed as automated ways to carry out such a search for modes, though most MCMC tempering methods continue to search for modes on an ongoing basis; see (Swendsen and Wang 1986; Neal 1996).↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Problematic Posteriors"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#local-variable-scope",
-    "href": "stan-users-guide/style-guide.html#local-variable-scope",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/references.html",
+    "href": "stan-users-guide/references.html",
+    "title": "References",
     "section": "",
-    "text": "Declaring local variables in the block in which they are used aids in understanding programs because it cuts down on the amount of text scanning or memory required to reunite the declaration and definition.\nThe following Stan program corresponds to a direct translation of a BUGS model, which uses a different element of mu in each iteration.\nmodel {\n  array[N] real mu;\n  for (n in 1:N) {\n    mu[n] = alpha * x[n] + beta;\n    y[n] ~ normal(mu[n],sigma);\n  }\n}\nBecause variables can be reused in Stan and because they should be declared locally for clarity, this model should be recoded as follows.\nmodel {\n  for (n in 1:N) {\n    real mu;\n    mu = alpha * x[n] + beta;\n    y[n] ~ normal(mu,sigma);\n  }\n}\nThe local variable can be eliminated altogether, as follows.\nmodel {\n  for (n in 1:N) {\n    y[n] ~ normal(alpha * x[n] + beta, sigma);\n  }\n}\nThere is unlikely to be any measurable efficiency difference between the last two implementations, but both should be a bit more efficient than the BUGS translation.\n\n\nIn the case of local variables for compound structures, such as arrays, vectors, or matrices, if they are built up component by component rather than in large chunks, it can be more efficient to declare a local variable for the structure outside of the block in which it is used. This allows it to be allocated once and then reused.\nmodel {\n  vector[K] mu;\n  for (n in 1:N) {\n    for (k in 1:K) {\n      mu[k] = // ...\n    }\n    y[n] ~ multi_normal(mu,Sigma);\n}\nIn this case, the vector mu will be allocated outside of both loops, and used a total of N times.",
+    "text": "References\n\n\n\n\n Back to top"
+  },
+  {
+    "objectID": "stan-users-guide/reparameterization.html",
+    "href": "stan-users-guide/reparameterization.html",
+    "title": "Reparameterization and Change of Variables",
+    "section": "",
+    "text": "Stan supports a direct encoding of reparameterizations. Stan also supports changes of variables by directly incrementing the log probability accumulator with the log Jacobian of the transform.\n\n\nA Bayesian posterior is technically a probability measure, which is a parameterization-invariant, abstract mathematical object.1\nStan’s modeling language, on the other hand, defines a probability density, which is a non-unique, parameterization-dependent function in \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^{+}\\). In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.\nAs pointed out by Gelman (2004) in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not just that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.\n\n\n\nReparameterizations may be implemented directly using the transformed parameters block or just in the model block.\n\n\nThe beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.\n\n\nFor example, the Beta distribution is parameterized by two positive count parameters \\(\\alpha, \\beta &gt; 0\\). The following example illustrates a hierarchical Stan model with a vector of parameters theta are drawn i.i.d. for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n  // ...\n}\nmodel {\n  alpha ~ ...\n  beta ~ ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nIt is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter \\[\n\\phi = \\alpha / (\\alpha + \\beta)\n\\] and total count parameter \\[\n\\lambda = \\alpha + \\beta.\n\\] Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with \\(p(\\lambda) \\propto \\lambda^{-2.5}\\).\nparameters {\n  real&lt;lower=0, upper=1&gt; phi;\n  real&lt;lower=0.1&gt; lambda;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0&gt; alpha = lambda * phi;\n  real&lt;lower=0&gt; beta = lambda * (1 - phi);\n  // ...\n}\nmodel {\n  phi ~ beta(1, 1); // uniform on phi, could drop\n  lambda ~ pareto(0.1, 1.5);\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nThe new parameters, phi and lambda, are declared in the parameters block and the parameters for the Beta distribution, alpha and beta, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.\nmodel {\n  real alpha = lambda * phi;\n  real beta = lambda * (1 - phi);\n  // ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nWith vectorization, this could be expressed more compactly and efficiently as follows.\nmodel {\n  theta ~ beta(lambda * phi, lambda * (1 - phi));\n  // ...\n}\nIf the variables alpha and beta are of interest, they can be defined in the transformed parameter block and then used in the model.\n\n\n\nBecause the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, alpha and beta have the correct posterior distribution.\n\n\n\nThe same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are \\(K &gt; 0\\) dimensions being considered (\\(K=1\\) is trivial and \\(K=2\\) reduces to the beta distribution case). The traditional prior is\nparameters {\n  vector[K] alpha;\n  array[N] simplex[K] theta;\n  // ...\n}\nmodel {\n  alpha ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n}\nThis provides essentially \\(K\\) degrees of freedom, one for each dimension of alpha, and it is not obvious how to specify a reasonable prior for alpha.\nAn alternative coding is to use the mean, which is a simplex, and a total count.\nparameters {\n  simplex[K] phi;\n  real&lt;lower=0&gt; kappa;\n  array[N] simplex[K] theta;\n  // ...\n}\ntransformed parameters {\n  vector[K] alpha = kappa * phi;\n  // ...\n}\nmodel {\n  phi ~ // ...\n  kappa ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n  // ...\n}\nNow it is much easier to formulate priors, because phi is the expected value of theta and kappa (minus K) is the strength of the prior mean measured in number of prior observations.\n\n\n\n\nIf the variable \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\operatorname{logit}(u)\\) is distributed as \\(\\textsf{logistic}(0, 1)\\). This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in \\((0, 1)\\) to a logistically-distributed quantity.\nThings work the same way for the probit case: if \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\Phi^{-1}(u)\\) has a \\(\\textsf{normal}(0, 1)\\) distribution. The other way around, if \\(v\\) has a \\(\\textsf{normal}(0, 1)\\) distribution, then \\(\\Phi(v)\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution.\nIn order to use the probit and logistic as priors on variables constrained to \\((0, 1)\\), create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a \\((0, 1)\\)-constrained parameter theta and gives it a beta prior, then uses it as a parameter in a distribution (here using foo as a placeholder).\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nmodel {\n  theta ~ beta(a, b);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIf the variables a and b are one, then this imposes a uniform distribution theta. If a and b are both less than one, then the density on theta has a U shape, whereas if they are both greater than one, the density of theta has an inverted-U or more bell-like shape.\nRoughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,\nparameters {\n  real theta_raw;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0, upper=1&gt; theta = inv_logit(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ logistic(mu, sigma);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIn this model, an unconstrained parameter theta_raw gets a logistic prior, and then the transformed parameter theta is defined to be the inverse logit of theta_raw. In this parameterization, inv_logit(mu) is the mean of the implied prior on theta. The prior distribution on theta will be flat if sigma is one and mu is zero, and will be U-shaped if sigma is larger than one and bell shaped if sigma is less than one.\nWhen moving from a variable in \\((0, 1)\\) to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.\nparameters {\n  simplex[K] theta;\n  // ...\n}\nmodel {\n  theta ~ dirichlet(a);\n  // ...\n  y ~ foo(theta);\n}\nNow a is a vector with K rows, but it has the same shape properties as the pair a and b for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector \\([a b]^{\\top}\\). To formulate an unconstrained prior, the exact same strategy works as for the beta.\nparameters {\n  vector[K] theta_raw;\n  // ...\n}\ntransformed parameters {\n  simplex[K] theta = softmax(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by softmax(mu), but we have additional control of covariance through L_Sigma at the expense of having on the order of \\(K^2\\) parameters in the prior rather than order \\(K\\). If no covariance is desired, the number of parameters can be reduced back to \\(K\\) using a vectorized normal distribution as follows.\ntheta_raw ~ normal(mu, sigma);\nwhere either or both of mu and sigma can be vectors.\n\n\n\n\nChanges of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable \\(y &gt; 0\\) whose logarithm \\(\\log y\\) has a normal distribution. The distribution is being assigned to \\(\\log y\\).\nThe change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^N\\).\nThe probability must be scaled by a Jacobian adjustment equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.\nIn the case of log normals, if \\(y\\)’s logarithm is normal with mean \\(\\mu\\) and deviation \\(\\sigma\\), then the distribution of \\(y\\) is given by \\[\np(y)\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\left| \\frac{d}{dy} \\log y \\right|\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\frac{1}{y}.\n\\] Stan works on the log scale to prevent underflow, where \\[\n\\log p(y)\n=\n\\log \\textsf{normal}(\\log y \\mid \\mu, \\sigma) - \\log y.\n\\]\nIn Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.2\nparameters {\n  real&lt;lower=0&gt; y;\n  // ...\n}\nmodel {\n  log(y) ~ normal(mu, sigma);\n  target += -log(y);\n  // ...\n}\nIt is important, as always, to declare appropriate constraints on parameters; here y is constrained to be positive.\nIt would be slightly more efficient to define a local variable for the logarithm, as follows.\nmodel {\n  real log_y;\n  log_y = log(y);\n  log_y ~ normal(mu, sigma);\n  target += -log_y;\n  // ...\n}\nIf y were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.\n\n\nThis section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.\nIt does not matter whether the probability function is expressed using a distribution statement, such as\nlog(y) ~ normal(mu, sigma);\nor as an increment to the log probability function, as in\ntarget += normal_lpdf(log(y) | mu, sigma);\n\n\nLike the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.\nThe transform based approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y_inv;\n  y_inv = 1 / y;\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe change-of-variables approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y_inv;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y;\n  y = 1 / y_inv;  // change variables\n  jacobian += -2 * log(y_inv); // Jacobian adjustment\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is\n\\[\n\\log \\left| \\frac{d}{du} \\left( \\frac{1}{u} \\right) \\right|\n= \\log \\left| - u^{-2} \\right|\n= \\log u^{-2}\n=  -2 \\log u.\n\\]\n\n\n\n\nIn the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.\nparameters {\n  vector[K] u;      // multivariate parameter\n   // ...\n}\ntransformed parameters {\n  vector[K] v;     // transformed parameter\n  matrix[K, K] J;   // Jacobian matrix of transform\n  // ... compute v as a function of u ...\n  // ... compute J[m, n] = d.v[m] / d.u[n] ...\n  jacobian += log(abs(determinant(J)));\n  // ...\n}\nmodel {\n  v ~ // ...\n  // ...\n}\nIf the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.\nIn many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element v[k] of the transformed parameter vector only depends on elements u[1], …, u[k] of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.\ntransformed parameters {\n  // ...\n  vector[K] J_diag;  // diagonals of Jacobian matrix\n  // ...\n  // ... compute J[k, k] = d.v[k] / d.u[k] ...\n  jacobian += sum(log(J_diag));\n  // ...\n}\n\n\n\n\nStan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.\n\n\nFor example, suppose there is a vector parameter \\(\\alpha\\) with a vector \\(L\\) of lower bounds. The simplest way to deal with this if \\(L\\) is a constant is to shift a lower-bounded parameter.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L&gt;[N] alpha_raw;\n  // ...\n}\nThe above is equivalent to manually calculating the vector bounds by the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + alpha_raw;\n  // ...\n}\nThe Jacobian for adding a constant is one, so its log drops out of the log density.\nEven if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from \\((L, \\alpha_{\\textrm{raw}})\\) to \\((L + \\alpha_{\\textrm{raw}}, \\alpha_{\\textrm{raw}})\\) produces a Jacobian derivative matrix with a unit determinant.\nIt’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  vector&lt;lower=L + alpha_raw&gt;[N] alpha;\n  // ...\n}\nThis is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + exp(alpha_raw);\n  jacobian += sum(alpha_raw); // log Jacobian\n  // ...\n}\nmodel {\n  // ...\n}\nThe adjustment in the log Jacobian determinant of the transform mapping \\(\\alpha_{\\textrm{raw}}\\) to \\(\\alpha = L + \\exp(\\alpha_{\\textrm{raw}})\\). The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here \\(L\\) can even be a vector containing parameters that don’t depend on \\(\\alpha_{\\textrm{raw}}\\); if the bounds do depend on \\(\\alpha_{\\textrm{raw}}\\) then a revised Jacobian needs to be calculated taking into account the dependencies.\n\n\n\nSuppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to \\((0, 1)\\). This is easily accomplished as the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L, upper=U&gt;[N] alpha;\n  // ...\n}\nThe same may be accomplished by manually constructing the transform as follows.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + (U - L) .* alpha_raw;\n}\nThe expression U - L is multiplied by alpha_raw elementwise to produce a vector of variables in \\((0, U-L)\\), then adding \\(L\\) results in a variable ranging between \\((L, U)\\).\nIn this case, it is important that \\(L\\) and \\(U\\) are constants, otherwise a Jacobian would be required when multiplying by \\(U - L\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#parentheses-and-brackets",
-    "href": "stan-users-guide/style-guide.html#parentheses-and-brackets",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/reparameterization.html#theoretical-and-practical-background",
+    "href": "stan-users-guide/reparameterization.html#theoretical-and-practical-background",
+    "title": "Reparameterization and Change of Variables",
     "section": "",
-    "text": "Single-statement blocks can be rendered in several ways. The preferred style is fully bracketed with the statement appearing on its own line, as follows.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThe use of loops and conditionals without brackets can be dangerous. For instance, consider this program.\nfor (n in 1:N)\n  z[n] ~ normal(nu,1);\n  y[n] ~ normal(mu,1);\nBecause Stan ignores whitespace and the parser completes a statement as eagerly as possible (just as in C++), the previous program is equivalent to the following program.\nfor (n in 1:N) {\n  z[n] ~ normal(nu,1);\n}\ny[n] ~ normal(mu,1);\nTherefore, one should prefer to use braces. The only exception is when nesting if-else clauses, where the else branch contains exactly one conditional. Then, it is preferred to place the following if on the same line, as in the following.\nif (x) {\n  // ...\n} else if (y) {\n  // ...\n} else {\n  // ...\n}\n\n\n\nThe preferred style for operators minimizes parentheses. This reduces clutter in code that can actually make it harder to read expressions. For example, the expression a + b * c is preferred to the equivalent a + (b * c) or (a + (b * c)). The operator precedences and associativities follow those of pretty much every programming language including Fortran, C++, R, and Python; full details are provided in the reference manual.\nSimilarly, comparison operators can usually be written with minimal bracketing, with the form y[n] &gt; 0 || x[n] != 0 preferred to the bracketed form (y[n] &gt; 0) || (x[n] != 0).\n\n\n\nVertical space is valuable as it controls how much of a program you can see. The preferred Stan style is with the opening brace appearing at the end of a line.\nfor (n in 1:N) {\n  y[n] ~ normal(mu,1);\n}\nThis also goes for parameters blocks, transformed data blocks, which should look as follows.\ntransformed parameters {\n  real sigma;\n  // ...\n}\nThe exception to this rule is local blocks which only exist for scoping reasons. The opening brace of these blocks is not associated with any control flow or block structure, so it should appear on its own line.",
+    "text": "A Bayesian posterior is technically a probability measure, which is a parameterization-invariant, abstract mathematical object.1\nStan’s modeling language, on the other hand, defines a probability density, which is a non-unique, parameterization-dependent function in \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^{+}\\). In practice, this means a given model can be represented different ways in Stan, and different representations have different computational performances.\nAs pointed out by Gelman (2004) in a paper discussing the relation between parameterizations and Bayesian modeling, a change of parameterization often carries with it suggestions of how the model might change, because we tend to use certain natural classes of prior distributions. Thus, it’s not just that we have a fixed distribution that we want to sample from, with reparameterizations being computational aids. In addition, once we reparameterize and add prior information, the model itself typically changes, often in useful ways.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#conditionals",
-    "href": "stan-users-guide/style-guide.html#conditionals",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/reparameterization.html#reparameterizations",
+    "href": "stan-users-guide/reparameterization.html#reparameterizations",
+    "title": "Reparameterization and Change of Variables",
     "section": "",
-    "text": "While Stan supports the full C++-style conditional syntax, allowing real or integer values to act as conditions, real values should be avoided. For a real-valued x, one should use\nif (x != 0) { ...\nin place of\nif (x) { ...\nBeyond stylistic choices, one should be careful using real values in a conditional expression, as direct comparison can have unexpected results due to numerical accuracy.",
+    "text": "Reparameterizations may be implemented directly using the transformed parameters block or just in the model block.\n\n\nThe beta and Dirichlet distributions may both be reparameterized from a vector of counts to use a mean and total count.\n\n\nFor example, the Beta distribution is parameterized by two positive count parameters \\(\\alpha, \\beta &gt; 0\\). The following example illustrates a hierarchical Stan model with a vector of parameters theta are drawn i.i.d. for a Beta distribution whose parameters are themselves drawn from a hyperprior distribution.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; beta;\n  // ...\n}\nmodel {\n  alpha ~ ...\n  beta ~ ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nIt is often more natural to specify hyperpriors in terms of transformed parameters. In the case of the Beta, the obvious choice for reparameterization is in terms of a mean parameter \\[\n\\phi = \\alpha / (\\alpha + \\beta)\n\\] and total count parameter \\[\n\\lambda = \\alpha + \\beta.\n\\] Following @[GelmanEtAl:2013, Chapter 5] the mean gets a uniform prior and the count parameter a Pareto prior with \\(p(\\lambda) \\propto \\lambda^{-2.5}\\).\nparameters {\n  real&lt;lower=0, upper=1&gt; phi;\n  real&lt;lower=0.1&gt; lambda;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0&gt; alpha = lambda * phi;\n  real&lt;lower=0&gt; beta = lambda * (1 - phi);\n  // ...\n}\nmodel {\n  phi ~ beta(1, 1); // uniform on phi, could drop\n  lambda ~ pareto(0.1, 1.5);\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nThe new parameters, phi and lambda, are declared in the parameters block and the parameters for the Beta distribution, alpha and beta, are declared and defined in the transformed parameters block. And if their values are not of interest, they could instead be defined as local variables in the model as follows.\nmodel {\n  real alpha = lambda * phi;\n  real beta = lambda * (1 - phi);\n  // ...\n  for (n in 1:N) {\n    theta[n] ~ beta(alpha, beta);\n  }\n  // ...\n}\nWith vectorization, this could be expressed more compactly and efficiently as follows.\nmodel {\n  theta ~ beta(lambda * phi, lambda * (1 - phi));\n  // ...\n}\nIf the variables alpha and beta are of interest, they can be defined in the transformed parameter block and then used in the model.\n\n\n\nBecause the transformed parameters are being used, rather than given a distribution, there is no need to apply a Jacobian adjustment for the transform. For example, in the beta distribution example, alpha and beta have the correct posterior distribution.\n\n\n\nThe same thing can be done with a Dirichlet, replacing the mean for the Beta, which is a probability value, with a simplex. Assume there are \\(K &gt; 0\\) dimensions being considered (\\(K=1\\) is trivial and \\(K=2\\) reduces to the beta distribution case). The traditional prior is\nparameters {\n  vector[K] alpha;\n  array[N] simplex[K] theta;\n  // ...\n}\nmodel {\n  alpha ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n}\nThis provides essentially \\(K\\) degrees of freedom, one for each dimension of alpha, and it is not obvious how to specify a reasonable prior for alpha.\nAn alternative coding is to use the mean, which is a simplex, and a total count.\nparameters {\n  simplex[K] phi;\n  real&lt;lower=0&gt; kappa;\n  array[N] simplex[K] theta;\n  // ...\n}\ntransformed parameters {\n  vector[K] alpha = kappa * phi;\n  // ...\n}\nmodel {\n  phi ~ // ...\n  kappa ~ // ...\n  for (n in 1:N) {\n    theta[n] ~ dirichlet(alpha);\n  }\n  // ...\n}\nNow it is much easier to formulate priors, because phi is the expected value of theta and kappa (minus K) is the strength of the prior mean measured in number of prior observations.\n\n\n\n\nIf the variable \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\operatorname{logit}(u)\\) is distributed as \\(\\textsf{logistic}(0, 1)\\). This is because inverse logit is the cumulative distribution function (cdf) for the logistic distribution, so that the logit function itself is the inverse CDF and thus maps a uniform draw in \\((0, 1)\\) to a logistically-distributed quantity.\nThings work the same way for the probit case: if \\(u\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution, then \\(\\Phi^{-1}(u)\\) has a \\(\\textsf{normal}(0, 1)\\) distribution. The other way around, if \\(v\\) has a \\(\\textsf{normal}(0, 1)\\) distribution, then \\(\\Phi(v)\\) has a \\(\\textsf{uniform}(0, 1)\\) distribution.\nIn order to use the probit and logistic as priors on variables constrained to \\((0, 1)\\), create an unconstrained variable and transform it appropriately. For comparison, the following Stan program fragment declares a \\((0, 1)\\)-constrained parameter theta and gives it a beta prior, then uses it as a parameter in a distribution (here using foo as a placeholder).\nparameters {\n  real&lt;lower=0, upper=1&gt; theta;\n  // ...\n}\nmodel {\n  theta ~ beta(a, b);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIf the variables a and b are one, then this imposes a uniform distribution theta. If a and b are both less than one, then the density on theta has a U shape, whereas if they are both greater than one, the density of theta has an inverted-U or more bell-like shape.\nRoughly the same result can be achieved with unbounded parameters that are probit or inverse-logit-transformed. For example,\nparameters {\n  real theta_raw;\n  // ...\n}\ntransformed parameters {\n  real&lt;lower=0, upper=1&gt; theta = inv_logit(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ logistic(mu, sigma);\n  // ...\n  y ~ foo(theta);\n  // ...\n}\nIn this model, an unconstrained parameter theta_raw gets a logistic prior, and then the transformed parameter theta is defined to be the inverse logit of theta_raw. In this parameterization, inv_logit(mu) is the mean of the implied prior on theta. The prior distribution on theta will be flat if sigma is one and mu is zero, and will be U-shaped if sigma is larger than one and bell shaped if sigma is less than one.\nWhen moving from a variable in \\((0, 1)\\) to a simplex, the same trick may be performed using the softmax function, which is a multinomial generalization of the inverse logit function. First, consider a simplex parameter with a Dirichlet prior.\nparameters {\n  simplex[K] theta;\n  // ...\n}\nmodel {\n  theta ~ dirichlet(a);\n  // ...\n  y ~ foo(theta);\n}\nNow a is a vector with K rows, but it has the same shape properties as the pair a and b for a beta; the beta distribution is just the distribution of the first component of a Dirichlet with parameter vector \\([a b]^{\\top}\\). To formulate an unconstrained prior, the exact same strategy works as for the beta.\nparameters {\n  vector[K] theta_raw;\n  // ...\n}\ntransformed parameters {\n  simplex[K] theta = softmax(theta_raw);\n  // ...\n}\nmodel {\n  theta_raw ~ multi_normal_cholesky(mu, L_Sigma);\n}\nThe multivariate normal is used for convenience and efficiency with its Cholesky-factor parameterization. Now the mean is controlled by softmax(mu), but we have additional control of covariance through L_Sigma at the expense of having on the order of \\(K^2\\) parameters in the prior rather than order \\(K\\). If no covariance is desired, the number of parameters can be reduced back to \\(K\\) using a vectorized normal distribution as follows.\ntheta_raw ~ normal(mu, sigma);\nwhere either or both of mu and sigma can be vectors.",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#functions",
-    "href": "stan-users-guide/style-guide.html#functions",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/reparameterization.html#changes-of-variables",
+    "href": "stan-users-guide/reparameterization.html#changes-of-variables",
+    "title": "Reparameterization and Change of Variables",
     "section": "",
-    "text": "Functions are laid out the same way as in languages such as Java and C++. For example,\nreal foo(real x, real y) {\n  return sqrt(x * log(y));\n}\nThe return type is flush left, the parentheses for the arguments are adjacent to the arguments and function name, and there is a space after the comma for arguments after the first. The open curly brace for the body is on the same line as the function name, following the layout of loops and conditionals. The body itself is indented; here we use two spaces. The close curly brace appears on its own line.\nIf function names or argument lists are long, they can be written as\nmatrix\nfunction_to_do_some_hairy_algebra(matrix thingamabob,\n                                  vector doohickey2) {\n  // ...body...\n}\nThe function starts a new line, under the type. The arguments are aligned under each other.\nFunction documentation should follow the Javadoc and Doxygen styles. Here’s an example repeated from the documenting functions section.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows correspond to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\n}\nThe open comment is /**, asterisks are aligned below the first asterisk of the open comment, and the end comment */ is also aligned on the asterisk. The tags @param and @return are used to label function arguments (i.e., parameters) and return values.",
+    "text": "Changes of variables are applied when the transformation of a parameter is characterized by a distribution. The standard textbook example is the lognormal distribution, which is the distribution of a variable \\(y &gt; 0\\) whose logarithm \\(\\log y\\) has a normal distribution. The distribution is being assigned to \\(\\log y\\).\nThe change of variables requires an adjustment to the probability to account for the distortion caused by the transform. For this to work, univariate changes of variables must be monotonic and differentiable everywhere in their support. Multivariate changes of variables must be injective and differentiable everywhere in their support, and they must map \\(\\mathbb{R}^N \\rightarrow \\mathbb{R}^N\\).\nThe probability must be scaled by a Jacobian adjustment equal to the absolute determinant of the Jacobian of the transform. In the univariate case, the Jacobian adjustment is simply the absolute derivative of the transform.\nIn the case of log normals, if \\(y\\)’s logarithm is normal with mean \\(\\mu\\) and deviation \\(\\sigma\\), then the distribution of \\(y\\) is given by \\[\np(y)\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\left| \\frac{d}{dy} \\log y \\right|\n= \\textsf{normal}(\\log y \\mid \\mu, \\sigma) \\, \\frac{1}{y}.\n\\] Stan works on the log scale to prevent underflow, where \\[\n\\log p(y)\n=\n\\log \\textsf{normal}(\\log y \\mid \\mu, \\sigma) - \\log y.\n\\]\nIn Stan, the change of variables can be applied in the sampling statement. To adjust for the curvature, the log probability accumulator is incremented with the log absolute derivative of the transform. The lognormal distribution can thus be implemented directly in Stan as follows.2\nparameters {\n  real&lt;lower=0&gt; y;\n  // ...\n}\nmodel {\n  log(y) ~ normal(mu, sigma);\n  target += -log(y);\n  // ...\n}\nIt is important, as always, to declare appropriate constraints on parameters; here y is constrained to be positive.\nIt would be slightly more efficient to define a local variable for the logarithm, as follows.\nmodel {\n  real log_y;\n  log_y = log(y);\n  log_y ~ normal(mu, sigma);\n  target += -log_y;\n  // ...\n}\nIf y were declared as data instead of as a parameter, then the adjustment can be ignored because the data will be constant and Stan only requires the log probability up to a constant.\n\n\nThis section illustrates the difference between a change of variables and a simple variable transformation. A transformation samples a parameter, then transforms it, whereas a change of variables transforms a parameter, then samples it. Only the latter requires a Jacobian adjustment.\nIt does not matter whether the probability function is expressed using a distribution statement, such as\nlog(y) ~ normal(mu, sigma);\nor as an increment to the log probability function, as in\ntarget += normal_lpdf(log(y) | mu, sigma);\n\n\nLike the log normal, the inverse gamma distribution is a distribution of variables whose inverse has a gamma distribution. This section contrasts two approaches, first with a transform, then with a change of variables.\nThe transform based approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y_inv;\n  y_inv = 1 / y;\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe change-of-variables approach to defining y_inv to have an inverse gamma distribution can be coded as follows.\nparameters {\n  real&lt;lower=0&gt; y_inv;\n}\ntransformed parameters {\n  real&lt;lower=0&gt; y;\n  y = 1 / y_inv;  // change variables\n  jacobian += -2 * log(y_inv); // Jacobian adjustment\n}\nmodel {\n  y ~ gamma(2,4);\n}\nThe Jacobian adjustment is the log of the absolute derivative of the transform, which in this case is\n\\[\n\\log \\left| \\frac{d}{du} \\left( \\frac{1}{u} \\right) \\right|\n= \\log \\left| - u^{-2} \\right|\n= \\log u^{-2}\n=  -2 \\log u.\n\\]\n\n\n\n\nIn the case of a multivariate transform, the log of the absolute determinant of the Jacobian of the transform must be added to the log probability accumulator. In Stan, this can be coded as follows in the general case where the Jacobian is not a full matrix.\nparameters {\n  vector[K] u;      // multivariate parameter\n   // ...\n}\ntransformed parameters {\n  vector[K] v;     // transformed parameter\n  matrix[K, K] J;   // Jacobian matrix of transform\n  // ... compute v as a function of u ...\n  // ... compute J[m, n] = d.v[m] / d.u[n] ...\n  jacobian += log(abs(determinant(J)));\n  // ...\n}\nmodel {\n  v ~ // ...\n  // ...\n}\nIf the determinant of the Jacobian is known analytically, it will be more efficient to apply it directly than to call the determinant function, which is neither efficient nor particularly stable numerically.\nIn many cases, the Jacobian matrix will be triangular, so that only the diagonal elements will be required for the determinant calculation. Triangular Jacobians arise when each element v[k] of the transformed parameter vector only depends on elements u[1], …, u[k] of the parameter vector. For triangular matrices, the determinant is the product of the diagonal elements, so the transformed parameters block of the above model can be simplified and made more efficient by recoding as follows.\ntransformed parameters {\n  // ...\n  vector[K] J_diag;  // diagonals of Jacobian matrix\n  // ...\n  // ... compute J[k, k] = d.v[k] / d.u[k] ...\n  jacobian += sum(log(J_diag));\n  // ...\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#white-space",
-    "href": "stan-users-guide/style-guide.html#white-space",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/reparameterization.html#vectors-with-varying-bounds",
+    "href": "stan-users-guide/reparameterization.html#vectors-with-varying-bounds",
+    "title": "Reparameterization and Change of Variables",
     "section": "",
-    "text": "Stan allows spaces between elements of a program. The white space characters allowed in Stan programs include the space (ASCII 0x20), line feed (ASCII 0x0A), carriage return (0x0D), and tab (0x09). Stan treats all whitespace characters interchangeably, with any sequence of whitespace characters being syntactically equivalent to a single space character. Nevertheless, effective use of whitespace is the key to good program layout.\n\n\nEach statement of a program should appear on its own line. Declaring multiple variables of the same type can be accomplished in a single statement with the syntax\nreal mu, sigma;\n\n\n\nStan programs should not contain tab characters. Using tabs to layout a program is highly unportable because the number of spaces represented by a single tab character varies depending on which program is doing the rendering and how it is configured.\n\n\n\nStan has standardized on two space characters of indentation, which is the standard convention for C/C++ code.\n\n\n\nUse a space after ifs. For instance, use if (x &lt; y) {..., not if(x &lt; y){ ....\n\n\n\nThere should not be space between a function name and the arguments it applies to. For instance, use normal(0, 1), not normal (0,1).\n\n\n\nThere should be spaces around binary operators. For instance, use y[1] = x, not y[1]=x, use (x + y) * z not (x+y)*z.\nUnary operators are written without a space, such as in -x, !y.\n\n\n\nAnother exception to the above rule is when the assignment operator (=) is used inside a type constraint, such as\nreal&lt;lower=0&gt; x;\nSpaces should still be used in arithmetic and following commas, as in\nreal&lt;lower=0, upper=a * x + b&gt; x;\n\n\n\nSometimes expressions are too long to fit on a single line. In that case, the recommended form is to break before an operator,2 aligning the operator to a term above to indicate scoping. For example, use the following form\nvector[J] p_distance = Phi((distance_tolerance - overshot)\n                           ./ ((x + overshot) * sigma_distance))\n                       - Phi((-overshot)\n                             ./ ((x + overshot) * sigma_distance));\nHere, the elementwise division operator (./) is aligned to clearly signal the division is occurring inside the parethenesis, while the subtraction indicates it is between the function applications (Phi).\nFor functions with multiple arguments, break after a comma and line the next argument up underneath as follows.\ny[n] ~ normal(alpha + beta * x + gamma * y,\n              pow(tau,-0.5));\n\n\n\nCommas should always be followed by spaces, including in function arguments, sequence literals, between variable declarations, etc.\nFor example,\nnormal(alpha * x[n] + beta, sigma);\nis preferred over\nnormal(alpha * x[n] + beta,sigma);\n\n\n\nWherever possible, Stan programs should use a single line feed character to separate lines. All of the Stan developers (so far, at least) work on Unix-like operating systems and using a standard newline makes the programs easier for us to read and share.\n\n\nNewlines are signaled in Unix-like operating systems such as Linux and Mac OS X with a single line-feed (LF) character (ASCII code point 0x0A). Newlines are signaled in Windows using two characters, a carriage return (CR) character (ASCII code point 0x0D) followed by a line-feed (LF) character.",
+    "text": "Stan allows scalar and non-scalar upper and lower bounds to be declared in the constraints for a container data type. The transforms are calculated and their log Jacobians added to the log density accumulator; the Jacobian calculations are described in detail in the reference manual chapter on constrained parameter transforms.\n\n\nFor example, suppose there is a vector parameter \\(\\alpha\\) with a vector \\(L\\) of lower bounds. The simplest way to deal with this if \\(L\\) is a constant is to shift a lower-bounded parameter.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L&gt;[N] alpha_raw;\n  // ...\n}\nThe above is equivalent to manually calculating the vector bounds by the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + alpha_raw;\n  // ...\n}\nThe Jacobian for adding a constant is one, so its log drops out of the log density.\nEven if the lower bound is a parameter rather than data, there is no Jacobian required, because the transform from \\((L, \\alpha_{\\textrm{raw}})\\) to \\((L + \\alpha_{\\textrm{raw}}, \\alpha_{\\textrm{raw}})\\) produces a Jacobian derivative matrix with a unit determinant.\nIt’s also possible to implement the transform using an array or vector of parameters as bounds (with the requirement that the type of the variable must match the bound type) in the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0&gt;[N] alpha_raw;\n  vector&lt;lower=L + alpha_raw&gt;[N] alpha;\n  // ...\n}\nThis is equivalent to directly transforming an unconstrained parameter and accounting for the Jacobian.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  // ...\n}\nparameters {\n  vector[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + exp(alpha_raw);\n  jacobian += sum(alpha_raw); // log Jacobian\n  // ...\n}\nmodel {\n  // ...\n}\nThe adjustment in the log Jacobian determinant of the transform mapping \\(\\alpha_{\\textrm{raw}}\\) to \\(\\alpha = L + \\exp(\\alpha_{\\textrm{raw}})\\). The details are simple in this case because the Jacobian is diagonal; see the reference manual chapter on constrained parameter transforms for full details. Here \\(L\\) can even be a vector containing parameters that don’t depend on \\(\\alpha_{\\textrm{raw}}\\); if the bounds do depend on \\(\\alpha_{\\textrm{raw}}\\) then a revised Jacobian needs to be calculated taking into account the dependencies.\n\n\n\nSuppose there are lower and upper bounds that vary by parameter. These can be applied to shift and rescale a parameter constrained to \\((0, 1)\\). This is easily accomplished as the following.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=L, upper=U&gt;[N] alpha;\n  // ...\n}\nThe same may be accomplished by manually constructing the transform as follows.\ndata {\n  int N;\n  vector[N] L;  // lower bounds\n  vector[N] U;  // upper bounds\n  // ...\n}\nparameters {\n  vector&lt;lower=0, upper=1&gt;[N] alpha_raw;\n  // ...\n}\ntransformed parameters {\n  vector[N] alpha = L + (U - L) .* alpha_raw;\n}\nThe expression U - L is multiplied by alpha_raw elementwise to produce a vector of variables in \\((0, U-L)\\), then adding \\(L\\) results in a variable ranging between \\((L, U)\\).\nIn this case, it is important that \\(L\\) and \\(U\\) are constants, otherwise a Jacobian would be required when multiplying by \\(U - L\\).",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
     ]
   },
   {
-    "objectID": "stan-users-guide/style-guide.html#footnotes",
-    "href": "stan-users-guide/style-guide.html#footnotes",
-    "title": "Stan Program Style Guide",
+    "objectID": "stan-users-guide/reparameterization.html#footnotes",
+    "href": "stan-users-guide/reparameterization.html#footnotes",
+    "title": "Reparameterization and Change of Variables",
     "section": "Footnotes",
-    "text": "Footnotes\n\n\nEven 80 characters may be too many for rendering in print; for instance, in this manual, the number of code characters that fit on a line is about 65.↩︎\nThis is the usual convention in both typesetting and other programming languages. Neither R nor BUGS allows breaks before an operator because they allow newlines to signal the end of an expression or statement.↩︎",
+    "text": "Footnotes\n\n\nThis is in contrast to (penalized) maximum likelihood estimates, which are not parameterization invariant.↩︎\nThis example is for illustrative purposes only; the recommended way to implement the lognormal distribution in Stan is with the built-in lognormal probability function; see the functions reference manual for details.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Appendices",
-      "Stan Program Style Guide"
+      "Programming Techniques",
+      "Reparameterization and Change of Variables"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html",
-    "href": "stan-users-guide/time-series.html",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/sparse-ragged.html",
+    "href": "stan-users-guide/sparse-ragged.html",
+    "title": "Sparse and Ragged Data Structures",
     "section": "",
-    "text": "Times series data come arranged in temporal order. This chapter presents two kinds of time series models, regression-like models such as autoregressive and moving average models, and hidden Markov models.\nThe Gaussian processes chapter presents Gaussian processes, which may also be used for time-series (and spatial) data.\n\n\nA first-order autoregressive model (AR(1)) with normal noise takes each point \\(y_n\\) in a sequence \\(y\\) to be generated according to \\[\ny_n \\sim \\textsf{normal}(\\alpha + \\beta y_{n-1}, \\sigma).\n\\]\nThat is, the expected value of \\(y_n\\) is \\(\\alpha + \\beta y_{n-1}\\), with noise scaled as \\(\\sigma\\).\n\n\nWith improper flat priors on the regression coefficients \\(\\alpha\\) and \\(\\beta\\) and on the positively-constrained noise scale (\\(\\sigma\\)), the Stan program for the AR(1) model is as follows.1\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  for (n in 2:N) {\n    y[n] ~ normal(alpha + beta * y[n-1], sigma);\n  }\n}\nThe first observed data point, y[1], is not modeled here because there is nothing to condition on; instead, it acts to condition y[2]. This model also uses an improper prior for sigma, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in y over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of y is available.\n\n\nAlthough perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner\nmodel {\n  y[2:N] ~ normal(alpha + beta * y[1:(N - 1)], sigma);\n}\nThe left-hand side slicing operation pulls out the last \\(N-1\\) elements and the right-hand side version pulls out the first \\(N-1\\).\n\n\n\n\nProper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-\\(t\\) distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.\nTo enforce the estimation of a stationary AR(1) process, the slope coefficient beta may be constrained with bounds as follows.\nreal&lt;lower=-1, upper=1&gt; beta;\nIn practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of beta near zero.\n\n\n\nExtending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient gamma and the following model statement.\nfor (n in 3:N) {\n  y[n] ~ normal(alpha + beta*y[n-1] + gamma*y[n-2], sigma);\n}\n\n\n\nA general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] real y;\n}\nparameters {\n  real alpha;\n  array[K] real beta;\n  real sigma;\n}\nmodel {\n  for (n in (K+1):N) {\n    real mu = alpha;\n    for (k in 1:K) {\n      mu += beta[k] * y[n-k];\n    }\n    y[n] ~ normal(mu, sigma);\n  }\n}\n\n\n\nEconometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model (Engle 1982). Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.\nThe ARCH(1) model is typically presented as the following sequence of equations, where \\(r_t\\) is the observed return at time point \\(t\\) and \\(\\mu\\), \\(\\alpha_0\\), and \\(\\alpha_1\\) are unknown regression coefficient parameters.\n\\[\\begin{align*}\nr_t        &=  \\mu + a_t \\\\\na_t        &=  \\sigma_t \\epsilon_t \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\sigma^2_t &=  \\alpha_0 + \\alpha_1 a_{t-1}^2\n\\end{align*}\\]\nIn order to ensure the noise terms \\(\\sigma^2_t\\) are positive, the scale coefficients are constrained to be positive, \\(\\alpha_0, \\alpha_1\n&gt; 0\\). To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., \\(\\alpha_1 &lt; 1\\).2\nThe ARCH(1) model may be coded directly in Stan as follows.\ndata {\n  int&lt;lower=0&gt; T;                // number of time points\n  array[T] real r;               // return at time t\n}\nparameters {\n  real mu;                       // average return\n  real&lt;lower=0&gt; alpha0;          // noise intercept\n  real&lt;lower=0, upper=1&gt; alpha1; // noise slope\n}\nmodel {\n  for (t in 2:T) {\n    r[t] ~ normal(mu, sqrt(alpha0 + alpha1\n                                    * pow(r[t - 1] - mu,2)));\n  }\n}\nThe loop in the model is defined so that the return at time \\(t=1\\) is not modeled; the model in the next section shows how to model the return at \\(t=1\\). The model can be vectorized to be more efficient; the model in the next section provides an example.\n\n\n\n\nA set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.\n\n\nThe basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time \\(t-1\\) as a predictor of volatility at time \\(t\\), defining \\[\n\\sigma^2_t = \\alpha_0 + \\alpha_1 a^2_{t-1} + \\beta_1 \\sigma^2_{t-1}.\n\\]\nTo ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy \\(\\alpha_0, \\alpha_1,\n\\beta_1 &gt; 0\\) and the slopes \\(\\alpha_1 + \\beta_1 &lt; 1\\).\ndata {\n  int&lt;lower=0&gt; T;\n  array[T] real r;\n  real&lt;lower=0&gt; sigma1;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; alpha0;\n  real&lt;lower=0, upper=1&gt; alpha1;\n  real&lt;lower=0, upper=(1-alpha1)&gt; beta1;\n}\ntransformed parameters {\n  array[T] real&lt;lower=0&gt; sigma;\n  sigma[1] = sigma1;\n  for (t in 2:T) {\n    sigma[t] = sqrt(alpha0\n                     + alpha1 * pow(r[t - 1] - mu, 2)\n                     + beta1 * pow(sigma[t - 1], 2));\n  }\n}\nmodel {\n  r ~ normal(mu, sigma);\n}\nTo get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value sigma1 for the scale of the noise at \\(t = 1\\).\nThe constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on beta1 depends on the value of alpha1.\nA transformed parameter array of non-negative values sigma is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept alpha0, a slope alpha1 for the squared difference in return from the mean at the previous time, and a slope beta1 for the previous noise scale squared. Finally, the whole regression is inside the sqrt function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.\nWith the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because r and sigma are of length T, all of the data are modeled directly.\n\n\n\n\nA moving average model uses previous errors as predictors for future outcomes. For a moving average model of order \\(Q\\), \\(\\mbox{MA}(Q)\\), there is an overall mean parameter \\(\\mu\\) and regression coefficients \\(\\theta_q\\) for previous error terms. With \\(\\epsilon_t\\) being the noise at time \\(t\\), the model for outcome \\(y_t\\) is defined by \\[\ny_t = \\mu + \\theta_1 \\epsilon_{t-1} + \\dotsb + \\theta_Q \\epsilon_{t-Q}\n+ \\epsilon_t,\n\\] with the noise term \\(\\epsilon_t\\) for outcome \\(y_t\\) modeled as normal, \\[\n\\epsilon_t \\sim \\textsf{normal}(0,\\sigma).\n\\] In a proper Bayesian model, the parameters \\(\\mu\\), \\(\\theta\\), and \\(\\sigma\\) must all be given priors.\n\n\nAn \\(\\mbox{MA}(2)\\) model can be coded in Stan as follows.\ndata {\n  int&lt;lower=3&gt; T;          // number of observations\n  vector[T] y;             // observation at time T\n}\nparameters {\n  real mu;                 // mean\n  real&lt;lower=0&gt; sigma;     // error scale\n  vector[2] theta;         // lag coefficients\n}\ntransformed parameters {\n  vector[T] epsilon;       // error terms\n  epsilon[1] = y[1] - mu;\n  epsilon[2] = y[2] - mu - theta[1] * epsilon[1];\n  for (t in 3:T) {\n    epsilon[t] = ( y[t] - mu\n                    - theta[1] * epsilon[t - 1]\n                    - theta[2] * epsilon[t - 2] );\n  }\n}\nmodel {\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 3:T) {\n    y[t] ~ normal(mu\n                  + theta[1] * epsilon[t - 1]\n                  + theta[2] * epsilon[t - 2],\n                  sigma);\n  }\n}\nThe error terms \\(\\epsilon_t\\) are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to \\(y_n\\) for \\(n &gt; Q\\). In this example, the parameters are all given Cauchy (half-Cauchy for \\(\\sigma\\)) priors, although other priors can be used just as easily.\nThis model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the \\(\\epsilon_t\\) could also be sped up by using a dot product instead of a loop.\n\n\n\nA general \\(\\mbox{MA}(Q)\\) model with a vectorized distribution statement may be defined as follows.\ndata {\n  int&lt;lower=0&gt; Q;       // num previous noise terms\n  int&lt;lower=3&gt; T;       // num observations\n  vector[T] y;          // observation at time t\n}\nparameters {\n  real mu;              // mean\n  real&lt;lower=0&gt; sigma;  // error scale\n  vector[Q] theta;      // error coeff, lag -t\n}\ntransformed parameters {\n  vector[T] epsilon;    // error term at time t\n  for (t in 1:T) {\n    epsilon[t] = y[t] - mu;\n    for (q in 1:min(t - 1, Q)) {\n      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];\n    }\n  }\n}\nmodel {\n  vector[T] eta;\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 1:T) {\n    eta[t] = mu;\n    for (q in 1:min(t - 1, Q)) {\n      eta[t] = eta[t] + theta[q] * epsilon[t - q];\n    }\n  }\n  y ~ normal(eta, sigma);\n}\nHere all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).\n\n\n\n\nAutoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.\ndata {\n  int&lt;lower=1&gt; T;            // num observations\n  array[T] real y;                 // observed outputs\n}\nparameters {\n  real mu;                   // mean coeff\n  real phi;                  // autoregression coeff\n  real theta;                // moving avg coeff\n  real&lt;lower=0&gt; sigma;       // noise scale\n}\nmodel {\n  vector[T] nu;              // prediction for time t\n  vector[T] err;             // error for time t\n  nu[1] = mu + phi * mu;     // assume err[0] == 0\n  err[1] = y[1] - nu[1];\n  for (t in 2:T) {\n    nu[t] = mu + phi * y[t - 1] + theta * err[t - 1];\n    err[t] = y[t] - nu[t];\n  }\n  mu ~ normal(0, 10);        // priors\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err ~ normal(0, sigma);    // error model\n}\nThe data are declared in the same way as the other time-series regressions and the parameters are documented in the code.\nIn the model block, the local vector nu stores the predictions and err the errors. These are computed similarly to the errors in the moving average models described in the previous section.\nThe priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.\nOften in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring err as a transformed parameter, then defining it the same way as in the model above. The vector nu could still be a local variable, only now it will be in the transformed parameter block.\nWayne Folta suggested encoding the model without local vector variables as follows.\nmodel {\n  real err;\n  mu ~ normal(0, 10);\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err = y[1] - (mu + phi * mu);\n  err ~ normal(0, sigma);\n  for (t in 2:T) {\n    err = y[t] - (mu + phi * y[t - 1] + theta * err);\n    err ~ normal(0, sigma);\n  }\n}\nThis approach to ARMA models illustrates how local variables, such as err in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.\nBoth encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just err.\n\n\nMA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint3\nreal&lt;lower=-1, upper=1&gt; theta;\nWhen the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (theta, phi) outside the \\([-1,1]\\) interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).\nFurther, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.\nreal&lt;lower=-1, upper=1&gt; phi;\n\n\n\n\nStochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time (Kim, Shephard, and Chib 1998). The data consist of mean corrected (i.e., centered) returns \\(y_t\\) on an underlying asset at \\(T\\) equally spaced time points. Kim et al. formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter \\(h_t\\) for the log volatility, along with parameters \\(\\mu\\) for the mean log volatility, and \\(\\phi\\) for the persistence of the volatility term. The variable \\(\\epsilon_t\\) represents the white-noise shock (i.e., multiplicative error) on the asset return at time \\(t\\), whereas \\(\\delta_t\\) represents the shock on volatility at time \\(t\\). \\[\\begin{align*}\ny_t        &=    \\epsilon_t \\exp(h_t / 2) \\\\\nh_{t+1}    &=    \\mu + \\phi (h_t - \\mu) + \\delta_t \\sigma \\\\\nh_1        &\\sim \\textsf{normal}\\left( \\mu, \\frac{\\sigma}{\\sqrt{1 - \\phi^2}} \\right) \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\delta_t   &\\sim \\textsf{normal}(0,1)\n\\end{align*}\\]\nRearranging the first line, \\(\\epsilon_t = y_t \\exp(-h_t / 2)\\), allowing the distribution for \\(y_t\\) to be written as \\[\ny_t \\sim \\textsf{normal}(0,\\exp(h_t/2)).\n\\] The recurrence equation for \\(h_{t+1}\\) may be combined with the scaling of \\(\\delta_t\\) to yield the distribution \\[\nh_t \\sim \\mathsf{normal}(\\mu + \\phi(h_{t-1} - \\mu), \\sigma).\n\\] This formulation can be directly encoded, as shown in the following Stan model.\ndata {\n  int&lt;lower=0&gt; T;   // # time points (equally spaced)\n  vector[T] y;      // mean corrected return at time t\n}\nparameters {\n  real mu;                     // mean log volatility\n  real&lt;lower=-1, upper=1&gt; phi; // persistence of volatility\n  real&lt;lower=0&gt; sigma;         // white noise shock scale\n  vector[T] h;                 // log volatility at time t\n}\nmodel {\n  phi ~ uniform(-1, 1);\n  sigma ~ cauchy(0, 5);\n  mu ~ cauchy(0, 10);\n  h[1] ~ normal(mu, sigma / sqrt(1 - phi * phi));\n  for (t in 2:T) {\n    h[t] ~ normal(mu + phi * (h[t - 1] -  mu), sigma);\n  }\n  for (t in 1:T) {\n    y[t] ~ normal(0, exp(h[t] / 2));\n  }\n}\nCompared to the Kim et al. formulation, the Stan model adds priors for the parameters \\(\\phi\\), \\(\\sigma\\), and \\(\\mu\\). The shock terms \\(\\epsilon_t\\) and \\(\\delta_t\\) do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.\nThe posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with \\(\\mu = -1.02\\), \\(\\phi = 0.95\\), and \\(\\sigma = 0.25\\) leads to 95% posterior intervals for \\(\\mu\\) of \\((-1.23, -0.54)\\), for \\(\\phi\\) of \\((0.82, 0.98)\\), and for \\(\\sigma\\) of \\((0.16, 0.38)\\).\nThe NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in (Hoffman and Gelman 2014). Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of \\(T\\).\nIt is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return \\(y\\) is easily vectorized to\ny ~ normal(0, exp(h / 2));\nThis speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter h_std to be declared instead of h.\nparameters {\n  // ...\n  vector[T] h_std;  // std log volatility time t\n}\nThe original value of h is then defined in a transformed parameter block.\ntransformed parameters {\n  vector[T] h = h_std * sigma;  // now h ~ normal(0, sigma)\n  h[1] /= sqrt(1 - phi * phi);  // rescale h[1]\n  h += mu;\n  for (t in 2:T) {\n    h[t] += phi * (h[t - 1] - mu);\n  }\n}\nThe first assignment rescales h_std to have a \\(\\textsf{normal}(0,\\sigma)\\) distribution and temporarily assigns it to h. The second assignment rescales h[1] so that its prior differs from that of h[2] through h[T]. The next assignment supplies a mu offset, so that h[2] through h[T] are now distributed \\(\\textsf{normal}(\\mu,\\sigma)\\); note that this shift must be done after the rescaling of h[1]. The final loop adds in the moving average so that h[2] through h[T] are appropriately modeled relative to phi and mu.\nAs a final improvement, the distribution statements for h[1] to h[T] are replaced with a single vectorized standard normal distribution statement.\nmodel {\n  // ...\n  h_std ~ std_normal();\n}\nAlthough the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.\n\n\n\nA Hidden Markov model is a probabilistic model over \\(N\\) observations \\(y_{1:N}\\) and \\(N\\) hidden states \\(z_{1:N}\\). This models is defined by the conditional distributions \\(p(y_n \\mid z_n, \\phi)\\) and \\(p(z_n \\mid z_{n-1}, \\phi)\\). Here we make the dependency on additional model parameters \\(\\phi\\) explicit. (\\(\\phi\\) may be a vector of parameters.) The complete data likelihood is then \\[\np(y, z \\mid \\phi) = \\prod_n p(y_n \\mid z_n, \\phi) p(z_n \\mid z_{n - 1}, \\phi)\n\\] When \\(z_{1:N}\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(z\\) out.\nWhen each state \\(z\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can use Stan’s suite of HMM functions to marginalize out \\(z_{1:N}\\) and compute \\[\np(y_{1:N} \\mid \\phi) = \\int_{\\mathcal Z} p(y, z \\mid \\phi) \\text d z.\n\\] We start by defining the conditional observation distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid z_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n\\Gamma_{ij} = p(z_n = j \\mid z_{n - 1} = i, \\phi).\n\\] (This is a right-stochastic matrix.) Finally, we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n\\rho_k = p(z_0 = k \\mid \\phi).\n\\] It is common practice to set \\(\\rho\\) to be the stationary distribution of the HMM, that is \\(\\rho\\) is the first eigenvector of \\(\\Gamma\\) and solves \\(\\Gamma \\rho = \\rho\\).\nAs an example, consider a three-state model with \\(K=3\\). The observations are normally distributed conditional on the HMM states with \\[\n  y_n \\sim \\text{normal}(\\mu_k, \\sigma),\n\\] where \\(\\mu = (1, 5, 9)\\) and the standard deviation \\(\\sigma\\) is the same across all observations. The model is then\ndata {\n  int N;  // Number of observations\n  array[N] real y;\n}\n\nparameters {\n  // Rows of the transition matrix\n  array[3] simplex[3] gamma_arr;\n\n  // Initial state\n  simplex[3] rho;\n\n  // Parameters of measurement model\n  vector[3] mu;\n  real&lt;lower = 0.0&gt; sigma;\n}\n\ntransformed parameters {\n  // Build transition matrix\n  matrix[3, 3] gamma;\n  for (k in 1:3) gamma[k, ] = to_row_vector(gamma_arr[k]);\n\n  // Compute the log likelihoods in each possible state\n  matrix[3, N] log_omega;\n  for (n in 1:N) {\n    for (i in 1:3) {\n      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);\n    }\n  }\n}\n\nmodel {\n  // prior\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  \n  // no explicit prior on gamma_arr, meaning we default to a\n  // uniform prior over the simplexes.\n\n  // Increment target by log p(y | mu, sigma, Gamma, rho)\n  target += hmm_marginal(log_omega, gamma, rho);\n}\nThe last function hmm_marginal takes in all the ingredients of the HMM and computes the relevant log marginal distribution, \\(\\log p(y \\mid \\phi)\\).\nIf we desire draws from the posterior distribution of \\(z\\), we use the generated quantities block and draw, for each sample \\(\\phi\\), a sample from \\(p(z \\mid y, \\phi)\\). In effect, MCMC produces draws from \\(p(\\phi \\mid y)\\) and with the draws in generated quantities, we obtain draws from \\(p(\\phi \\mid y) p(z \\mid y, \\phi) = p(z, \\phi \\mid y)\\). It is also possible to compute the posterior probbability of each hidden state, that is \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). Averagging these probabilities over all MCMC draws, we obtain \\(\\text{Pr}(z_n = k \\mid y)\\).\ngenerated quantities {\n  array[N] int latent_states = hmm_latent_rng(log_omega, gamma, rho);\n  matrix[3, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);\n}\nhmm_hidden_state_prob returns the marginal probabilities of each state, \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). This function cannot be used to compute the joint probability \\(\\text{Pr}(z \\mid \\phi, y)\\), because such calculation requires accounting for the posterior correlation between the different components of \\(z\\). Therefore, hidden_probs should not be used to obtain posterior draws. Instead, users should rely on hmm_latent_rng.\ngenerated quantities {\n   array[N] int&lt;lower=1, upper=K&gt; z = hmm_latent_rng(...fill-in params here to match example...);\n}\nThe example in this section is derived from the more detailed case study by Ben Bales: https://mc-stan.org/users/documentation/case-studies/hmm-example.html.",
+    "text": "Stan does not directly support either sparse or ragged data structures, though both can be accommodated with some programming effort. The sparse matrices chapter introduces a special-purpose sparse matrix times dense vector multiplication, which should be used where applicable; this chapter covers more general data structures.\n\n\nCoding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the item-response model section. There are \\(J\\) students and \\(K\\) questions, and if every student answers every question, then it is practical to declare the data as a \\(J \\times K\\) array of answers.\ndata {\n  int&lt;lower=1&gt; J;\n  int&lt;lower=1&gt; K;\n  array[J, K] int&lt;lower=0, upper=1&gt; y;\n  // ...\nmodel {\n  for (j in 1:J) {\n    for (k in 1:K) {\n      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));\n    }\n  }\n  // ...\n}\nWhen not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.\nThe following missing data example shows an example with \\(J=3\\) and \\(K=4\\), with missing responses shown as NA, as in R.\n\\[\\begin{equation*}\ny\n=\n\\left[\n\\begin{array}{cccc}\n0 & 1 & \\mbox{NA} & 1\n\\\\\n0 & \\mbox{NA} & \\mbox{NA} & 1\n\\\\\n\\mbox{NA} & 0 & \\mbox{NA} & \\mbox{NA}\n\\end{array}\n\\right]\n\\end{equation*}\\]\nThere is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns \\(jj\\) and \\(kk\\) used for the indexes (following Gelman and Hill (2007)), the 2-D array \\(y\\) is recoded as a table. The number of rows in the table equals the number of defined array elements, here \\(y_{1,1} = 0\\), \\(y_{1,2} = 1\\), up to \\(y_{3,2} = 1\\). As the array becomes larger and sparser, the long form becomes the more economical encoding.\n\n\n\njj\nkk\ny\n\n\n\n\n1\n1\n0\n\n\n1\n2\n1\n\n\n1\n4\n1\n\n\n2\n1\n0\n\n\n2\n4\n1\n\n\n3\n2\n0\n\n\n\nLetting \\(N\\) be the number of \\(y\\) that are defined, here \\(N=6\\), the data and model can be formulated as follows.\ndata {\n  // ...\n  int&lt;lower=1&gt; N;\n  array[N] int&lt;lower=1, upper=J&gt; jj;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  // ...\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(delta[kk[n]]\n                           * (alpha[jj[n]] - beta[kk[n]]));\n  }\n  // ...\n}\nIn the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.\n\n\n\nRagged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.\nA general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.\nFor example, consider a data structure for three groups, each of which has a different number of observations.\n\n\n\n\n\n\n\\(y_1 =  \\left[1.3 \\ \\ 2.4 \\ \\ 0.9\\right]\\\\\\) \\(y_2 = \\left[-1.8 \\ \\ -0.1\\right]\\\\\\) \\(y_3 = \\left[12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7\\right]\\)\n\n\n\\(z = [1.3 \\ \\ 2.4 \\ \\ 0.9 \\ \\ -1.8 \\ \\ -0.1 \\ \\ 12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7]\\\\\\) \\(s  =  \\{ 3 \\ \\ 2 \\ \\ 4 \\}\\)\n\n\n\nOn the left is the definition of a ragged data structure \\(y\\) with three rows of different sizes (\\(y_1\\) is size 3, \\(y_2\\) size 2, and \\(y_3\\) size 4). On the right is an example of how to code the data in Stan, using a single vector \\(z\\) to hold all the values and a separate array of integers \\(s\\) to hold the group row sizes. In this example, \\(y_1 = z_{1:3}\\), \\(y_2 = z_{4:5}\\), and \\(y_3 = z_{6:9}\\).\nSuppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood \\[\\begin{equation*}\n\\sum_{n=1}^3 \\log \\textsf{normal}(y_n \\mid \\mu_n, \\sigma).\n\\end{equation*}\\] There’s no direct way to encode this in Stan.\nA full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.\nThe model can then be coded up using slicing operations as follows.\ndata {\n  int&lt;lower=0&gt; N;   // # observations\n  int&lt;lower=0&gt; K;   // # of groups\n  vector[N] y;      // observations\n  array[K] int s;   // group sizes\n  // ...\n}\nmodel {\n  int pos;\n  pos = 1;\n  for (k in 1:K) {\n    segment(y, pos, s[k]) ~ normal(mu[k], sigma);\n    pos = pos + s[k];\n  }\nThis coding allows for efficient vectorization, which is worth the copy cost entailed by the segment() vector slicing operation.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Sparse and Ragged Data Structures"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#autoregressive.section",
-    "href": "stan-users-guide/time-series.html#autoregressive.section",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/sparse-ragged.html#sparse-data-structures",
+    "href": "stan-users-guide/sparse-ragged.html#sparse-data-structures",
+    "title": "Sparse and Ragged Data Structures",
     "section": "",
-    "text": "A first-order autoregressive model (AR(1)) with normal noise takes each point \\(y_n\\) in a sequence \\(y\\) to be generated according to \\[\ny_n \\sim \\textsf{normal}(\\alpha + \\beta y_{n-1}, \\sigma).\n\\]\nThat is, the expected value of \\(y_n\\) is \\(\\alpha + \\beta y_{n-1}\\), with noise scaled as \\(\\sigma\\).\n\n\nWith improper flat priors on the regression coefficients \\(\\alpha\\) and \\(\\beta\\) and on the positively-constrained noise scale (\\(\\sigma\\)), the Stan program for the AR(1) model is as follows.1\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] y;\n}\nparameters {\n  real alpha;\n  real beta;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  for (n in 2:N) {\n    y[n] ~ normal(alpha + beta * y[n-1], sigma);\n  }\n}\nThe first observed data point, y[1], is not modeled here because there is nothing to condition on; instead, it acts to condition y[2]. This model also uses an improper prior for sigma, but there is no obstacle to adding an informative prior if information is available on the scale of the changes in y over time, or a weakly informative prior to help guide inference if rough knowledge of the scale of y is available.\n\n\nAlthough perhaps a bit more difficult to read, a much more efficient way to write the above model is by slicing the vectors, with the model above being replaced with the one-liner\nmodel {\n  y[2:N] ~ normal(alpha + beta * y[1:(N - 1)], sigma);\n}\nThe left-hand side slicing operation pulls out the last \\(N-1\\) elements and the right-hand side version pulls out the first \\(N-1\\).\n\n\n\n\nProper priors of a range of different families may be added for the regression coefficients and noise scale. The normal noise model can be changed to a Student-\\(t\\) distribution or any other distribution with unbounded support. The model could also be made hierarchical if multiple series of observations are available.\nTo enforce the estimation of a stationary AR(1) process, the slope coefficient beta may be constrained with bounds as follows.\nreal&lt;lower=-1, upper=1&gt; beta;\nIn practice, such a constraint is not recommended. If the data are not well fit by a stationary model it is best to know this. Stationary parameter estimates can be encouraged with a prior favoring values of beta near zero.\n\n\n\nExtending the order of the model is also straightforward. For example, an AR(2) model could be coded with the second-order coefficient gamma and the following model statement.\nfor (n in 3:N) {\n  y[n] ~ normal(alpha + beta*y[n-1] + gamma*y[n-2], sigma);\n}\n\n\n\nA general model where the order is itself given as data can be coded by putting the coefficients in an array and computing the linear predictor in a loop.\ndata {\n  int&lt;lower=0&gt; K;\n  int&lt;lower=0&gt; N;\n  array[N] real y;\n}\nparameters {\n  real alpha;\n  array[K] real beta;\n  real sigma;\n}\nmodel {\n  for (n in (K+1):N) {\n    real mu = alpha;\n    for (k in 1:K) {\n      mu += beta[k] * y[n-k];\n    }\n    y[n] ~ normal(mu, sigma);\n  }\n}\n\n\n\nEconometric and financial time-series models usually assume heteroscedasticity: they allow the scale of the noise terms defining the series to vary over time. The simplest such model is the autoregressive conditional heteroscedasticity (ARCH) model (Engle 1982). Unlike the autoregressive model AR(1), which modeled the mean of the series as varying over time but left the noise term fixed, the ARCH(1) model takes the scale of the noise terms to vary over time but leaves the mean term fixed. Models could be defined where both the mean and scale vary over time; the econometrics literature presents a wide range of time-series modeling choices.\nThe ARCH(1) model is typically presented as the following sequence of equations, where \\(r_t\\) is the observed return at time point \\(t\\) and \\(\\mu\\), \\(\\alpha_0\\), and \\(\\alpha_1\\) are unknown regression coefficient parameters.\n\\[\\begin{align*}\nr_t        &=  \\mu + a_t \\\\\na_t        &=  \\sigma_t \\epsilon_t \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\sigma^2_t &=  \\alpha_0 + \\alpha_1 a_{t-1}^2\n\\end{align*}\\]\nIn order to ensure the noise terms \\(\\sigma^2_t\\) are positive, the scale coefficients are constrained to be positive, \\(\\alpha_0, \\alpha_1\n&gt; 0\\). To ensure stationarity of the time series, the slope is constrained to be less than one, i.e., \\(\\alpha_1 &lt; 1\\).2\nThe ARCH(1) model may be coded directly in Stan as follows.\ndata {\n  int&lt;lower=0&gt; T;                // number of time points\n  array[T] real r;               // return at time t\n}\nparameters {\n  real mu;                       // average return\n  real&lt;lower=0&gt; alpha0;          // noise intercept\n  real&lt;lower=0, upper=1&gt; alpha1; // noise slope\n}\nmodel {\n  for (t in 2:T) {\n    r[t] ~ normal(mu, sqrt(alpha0 + alpha1\n                                    * pow(r[t - 1] - mu,2)));\n  }\n}\nThe loop in the model is defined so that the return at time \\(t=1\\) is not modeled; the model in the next section shows how to model the return at \\(t=1\\). The model can be vectorized to be more efficient; the model in the next section provides an example.",
+    "text": "Coding sparse data structures is as easy as moving from a matrix-like data structure to a database-like data structure. For example, consider the coding of sparse data for the IRT models discussed in the item-response model section. There are \\(J\\) students and \\(K\\) questions, and if every student answers every question, then it is practical to declare the data as a \\(J \\times K\\) array of answers.\ndata {\n  int&lt;lower=1&gt; J;\n  int&lt;lower=1&gt; K;\n  array[J, K] int&lt;lower=0, upper=1&gt; y;\n  // ...\nmodel {\n  for (j in 1:J) {\n    for (k in 1:K) {\n      y[j, k] ~ bernoulli_logit(delta[k] * (alpha[j] - beta[k]));\n    }\n  }\n  // ...\n}\nWhen not every student is given every question, the dense array coding will no longer work, because Stan does not support undefined values.\nThe following missing data example shows an example with \\(J=3\\) and \\(K=4\\), with missing responses shown as NA, as in R.\n\\[\\begin{equation*}\ny\n=\n\\left[\n\\begin{array}{cccc}\n0 & 1 & \\mbox{NA} & 1\n\\\\\n0 & \\mbox{NA} & \\mbox{NA} & 1\n\\\\\n\\mbox{NA} & 0 & \\mbox{NA} & \\mbox{NA}\n\\end{array}\n\\right]\n\\end{equation*}\\]\nThere is no support within Stan for R’s NA values, so this data structure cannot be used directly. Instead, it must be converted to a “long form” as in a database, with columns indicating the indices along with the value. With columns \\(jj\\) and \\(kk\\) used for the indexes (following Gelman and Hill (2007)), the 2-D array \\(y\\) is recoded as a table. The number of rows in the table equals the number of defined array elements, here \\(y_{1,1} = 0\\), \\(y_{1,2} = 1\\), up to \\(y_{3,2} = 1\\). As the array becomes larger and sparser, the long form becomes the more economical encoding.\n\n\n\njj\nkk\ny\n\n\n\n\n1\n1\n0\n\n\n1\n2\n1\n\n\n1\n4\n1\n\n\n2\n1\n0\n\n\n2\n4\n1\n\n\n3\n2\n0\n\n\n\nLetting \\(N\\) be the number of \\(y\\) that are defined, here \\(N=6\\), the data and model can be formulated as follows.\ndata {\n  // ...\n  int&lt;lower=1&gt; N;\n  array[N] int&lt;lower=1, upper=J&gt; jj;\n  array[N] int&lt;lower=1, upper=K&gt; kk;\n  array[N] int&lt;lower=0, upper=1&gt; y;\n  // ...\n}\nmodel {\n  for (n in 1:N) {\n    y[n] ~ bernoulli_logit(delta[kk[n]]\n                           * (alpha[jj[n]] - beta[kk[n]]));\n  }\n  // ...\n}\nIn the situation where there are no missing values, the two model formulations produce exactly the same log posterior density.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Sparse and Ragged Data Structures"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#modeling-temporal-heteroscedasticity",
-    "href": "stan-users-guide/time-series.html#modeling-temporal-heteroscedasticity",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/sparse-ragged.html#ragged-data-structs.section",
+    "href": "stan-users-guide/sparse-ragged.html#ragged-data-structs.section",
+    "title": "Sparse and Ragged Data Structures",
     "section": "",
-    "text": "A set of variables is homoscedastic if their variances are all the same; the variables are heteroscedastic if they do not all have the same variance. Heteroscedastic time-series models allow the noise term to vary over time.\n\n\nThe basic generalized autoregressive conditional heteroscedasticity (GARCH) model, GARCH(1,1), extends the ARCH(1) model by including the squared previous difference in return from the mean at time \\(t-1\\) as a predictor of volatility at time \\(t\\), defining \\[\n\\sigma^2_t = \\alpha_0 + \\alpha_1 a^2_{t-1} + \\beta_1 \\sigma^2_{t-1}.\n\\]\nTo ensure the scale term is positive and the resulting time series stationary, the coefficients must all satisfy \\(\\alpha_0, \\alpha_1,\n\\beta_1 &gt; 0\\) and the slopes \\(\\alpha_1 + \\beta_1 &lt; 1\\).\ndata {\n  int&lt;lower=0&gt; T;\n  array[T] real r;\n  real&lt;lower=0&gt; sigma1;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; alpha0;\n  real&lt;lower=0, upper=1&gt; alpha1;\n  real&lt;lower=0, upper=(1-alpha1)&gt; beta1;\n}\ntransformed parameters {\n  array[T] real&lt;lower=0&gt; sigma;\n  sigma[1] = sigma1;\n  for (t in 2:T) {\n    sigma[t] = sqrt(alpha0\n                     + alpha1 * pow(r[t - 1] - mu, 2)\n                     + beta1 * pow(sigma[t - 1], 2));\n  }\n}\nmodel {\n  r ~ normal(mu, sigma);\n}\nTo get the recursive definition of the volatility regression off the ground, the data declaration includes a non-negative value sigma1 for the scale of the noise at \\(t = 1\\).\nThe constraints are coded directly on the parameter declarations. This declaration is order-specific in that the constraint on beta1 depends on the value of alpha1.\nA transformed parameter array of non-negative values sigma is used to store the scale values at each time point. The definition of these values in the transformed parameters block is where the regression is now defined. There is an intercept alpha0, a slope alpha1 for the squared difference in return from the mean at the previous time, and a slope beta1 for the previous noise scale squared. Finally, the whole regression is inside the sqrt function because Stan requires scale (deviation) parameters (not variance parameters) for the normal distribution.\nWith the regression in the transformed parameters block, the model reduces a single vectorized distribution statement. Because r and sigma are of length T, all of the data are modeled directly.",
+    "text": "Ragged arrays are arrays that are not rectangular, but have different sized entries. This kind of structure crops up when there are different numbers of observations per entry.\nA general approach to dealing with ragged structure is to move to a full database-like data structure as discussed in the previous section. A more compact approach is possible with some indexing into a linear array.\nFor example, consider a data structure for three groups, each of which has a different number of observations.\n\n\n\n\n\n\n\\(y_1 =  \\left[1.3 \\ \\ 2.4 \\ \\ 0.9\\right]\\\\\\) \\(y_2 = \\left[-1.8 \\ \\ -0.1\\right]\\\\\\) \\(y_3 = \\left[12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7\\right]\\)\n\n\n\\(z = [1.3 \\ \\ 2.4 \\ \\ 0.9 \\ \\ -1.8 \\ \\ -0.1 \\ \\ 12.9 \\ \\ 18.7 \\ \\ 42.9 \\ \\ 4.7]\\\\\\) \\(s  =  \\{ 3 \\ \\ 2 \\ \\ 4 \\}\\)\n\n\n\nOn the left is the definition of a ragged data structure \\(y\\) with three rows of different sizes (\\(y_1\\) is size 3, \\(y_2\\) size 2, and \\(y_3\\) size 4). On the right is an example of how to code the data in Stan, using a single vector \\(z\\) to hold all the values and a separate array of integers \\(s\\) to hold the group row sizes. In this example, \\(y_1 = z_{1:3}\\), \\(y_2 = z_{4:5}\\), and \\(y_3 = z_{6:9}\\).\nSuppose the model is a simple varying intercept model, which, using vectorized notation, would yield a log-likelihood \\[\\begin{equation*}\n\\sum_{n=1}^3 \\log \\textsf{normal}(y_n \\mid \\mu_n, \\sigma).\n\\end{equation*}\\] There’s no direct way to encode this in Stan.\nA full database type structure could be used, as in the sparse example, but this is inefficient, wasting space for unnecessary indices and not allowing vector-based density operations. A better way to code this data is as a single list of values, with a separate data structure indicating the sizes of each subarray. This is indicated on the right of the example. This coding uses a single array for the values and a separate array for the sizes of each row.\nThe model can then be coded up using slicing operations as follows.\ndata {\n  int&lt;lower=0&gt; N;   // # observations\n  int&lt;lower=0&gt; K;   // # of groups\n  vector[N] y;      // observations\n  array[K] int s;   // group sizes\n  // ...\n}\nmodel {\n  int pos;\n  pos = 1;\n  for (k in 1:K) {\n    segment(y, pos, s[k]) ~ normal(mu[k], sigma);\n    pos = pos + s[k];\n  }\nThis coding allows for efficient vectorization, which is worth the copy cost entailed by the segment() vector slicing operation.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Sparse and Ragged Data Structures"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#moving-average-models",
-    "href": "stan-users-guide/time-series.html#moving-average-models",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/survival.html",
+    "href": "stan-users-guide/survival.html",
+    "title": "Survival Models",
     "section": "",
-    "text": "A moving average model uses previous errors as predictors for future outcomes. For a moving average model of order \\(Q\\), \\(\\mbox{MA}(Q)\\), there is an overall mean parameter \\(\\mu\\) and regression coefficients \\(\\theta_q\\) for previous error terms. With \\(\\epsilon_t\\) being the noise at time \\(t\\), the model for outcome \\(y_t\\) is defined by \\[\ny_t = \\mu + \\theta_1 \\epsilon_{t-1} + \\dotsb + \\theta_Q \\epsilon_{t-Q}\n+ \\epsilon_t,\n\\] with the noise term \\(\\epsilon_t\\) for outcome \\(y_t\\) modeled as normal, \\[\n\\epsilon_t \\sim \\textsf{normal}(0,\\sigma).\n\\] In a proper Bayesian model, the parameters \\(\\mu\\), \\(\\theta\\), and \\(\\sigma\\) must all be given priors.\n\n\nAn \\(\\mbox{MA}(2)\\) model can be coded in Stan as follows.\ndata {\n  int&lt;lower=3&gt; T;          // number of observations\n  vector[T] y;             // observation at time T\n}\nparameters {\n  real mu;                 // mean\n  real&lt;lower=0&gt; sigma;     // error scale\n  vector[2] theta;         // lag coefficients\n}\ntransformed parameters {\n  vector[T] epsilon;       // error terms\n  epsilon[1] = y[1] - mu;\n  epsilon[2] = y[2] - mu - theta[1] * epsilon[1];\n  for (t in 3:T) {\n    epsilon[t] = ( y[t] - mu\n                    - theta[1] * epsilon[t - 1]\n                    - theta[2] * epsilon[t - 2] );\n  }\n}\nmodel {\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 3:T) {\n    y[t] ~ normal(mu\n                  + theta[1] * epsilon[t - 1]\n                  + theta[2] * epsilon[t - 2],\n                  sigma);\n  }\n}\nThe error terms \\(\\epsilon_t\\) are defined as transformed parameters in terms of the observations and parameters. The definition of the distribution statement (which also defines the likelihood) follows the definition, which can only be applied to \\(y_n\\) for \\(n &gt; Q\\). In this example, the parameters are all given Cauchy (half-Cauchy for \\(\\sigma\\)) priors, although other priors can be used just as easily.\nThis model could be improved in terms of speed by vectorizing the distribution statement in the model block. Vectorizing the calculation of the \\(\\epsilon_t\\) could also be sped up by using a dot product instead of a loop.\n\n\n\nA general \\(\\mbox{MA}(Q)\\) model with a vectorized distribution statement may be defined as follows.\ndata {\n  int&lt;lower=0&gt; Q;       // num previous noise terms\n  int&lt;lower=3&gt; T;       // num observations\n  vector[T] y;          // observation at time t\n}\nparameters {\n  real mu;              // mean\n  real&lt;lower=0&gt; sigma;  // error scale\n  vector[Q] theta;      // error coeff, lag -t\n}\ntransformed parameters {\n  vector[T] epsilon;    // error term at time t\n  for (t in 1:T) {\n    epsilon[t] = y[t] - mu;\n    for (q in 1:min(t - 1, Q)) {\n      epsilon[t] = epsilon[t] - theta[q] * epsilon[t - q];\n    }\n  }\n}\nmodel {\n  vector[T] eta;\n  mu ~ cauchy(0, 2.5);\n  theta ~ cauchy(0, 2.5);\n  sigma ~ cauchy(0, 2.5);\n  for (t in 1:T) {\n    eta[t] = mu;\n    for (q in 1:min(t - 1, Q)) {\n      eta[t] = eta[t] + theta[q] * epsilon[t - q];\n    }\n  }\n  y ~ normal(eta, sigma);\n}\nHere all of the data are modeled, with missing terms just dropped from the regressions as in the calculation of the error terms. Both models converge quickly and mix well at convergence, with the vectorized model being faster (per iteration, not to converge—they compute the same model).",
+    "text": "Survival models apply to animals and plants as well as inanimate objects such as machine parts or electrical components. Survival models arise when there is an event of interest for a group of subjects, machine component, or other item that is\n\ncertain to occur after some amount of time,\nbut only measured for a fixed period of time, during which the event may not have occurred for all subjects.\n\nFor example, one might wish to estimate the the distribution of time to failure for solid state drives in a data center, but only measure drives for a two year period, after which some number will have failed and some will still be in service.\nSurvival models are often used comparatively, such as comparing time to death of patients diagnosed with stage one liver cancer under a new treatment and a standard treatment (pure controls are not allowed when there is an effective existing treatment for a serious condition). During a two year trial, some patients will die and others will survive.\nSurvival models may involve covariates, such as the factory at which a component is manufactured, the day on which it is manufactured, and the amount of usage it gets. A clinical trial might be adjusted for the sex and age of a cancer patient or the hospital at which treatment is received.\nSurvival models come in two main flavors, parametric and semi-parametric. In a parametric model, the survival time of a subject is modeled explicitly using a parametric probability distribution. There is a great deal of flexibility in how the parametric probability distribution is constructed. The sections below consider exponential and Weibull distributed survival times.\nRather than explicitly modeling a parametric survival probability, semi-parametric survival models instead model the relative effect on survival of covariates. The final sections of this chapter consider the proportional hazards survival model.\n\n\nThe exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if \\(T \\sim \\textrm{exponential}(\\lambda)\\) for some rate \\(\\lambda &gt; 0,\\) then \\[\\begin{equation*}\n\\Pr[T &gt; t] = \\Pr[T &gt; t + t' \\mid T &gt; t'].\n\\end{equation*}\\] If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.\nThe exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable \\(T\\) is \\[\\begin{equation*}\n\\mathbb{E}[T \\mid \\lambda] = \\frac{1}{\\lambda}.\n\\end{equation*}\\] The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) \\(\\beta = 1 / \\lambda\\).\nThe data for a survival model consists of two components. First, there is a vector \\(t \\in (0, \\infty)^N\\) of \\(N\\) observed failure times. Second, there is a censoring time \\(t^{\\textrm{cens}}\\) such that failure times greater than \\(t^{\\textrm{cens}}\\) are not observed. The censoring time assumption imposes a constraint which requires \\(t_n &lt; t^{\\textrm{cens}}\\) for all \\(n \\in 1{:}N.\\) For the censored subjects, the only thing required in the model is their total count, \\(N^\\textrm{cens}\\) (their covariates are also required for models with covariates).\nThe model for the observed failure times is exponential, so that for \\(n \\in 1{:}N,\\) \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda).\n\\end{equation*}\\]\nThe model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of \\[\\begin{equation*}\n\\Pr[T &gt; t^{\\textrm{cens}}] = 1 - F_T(t^{\\textrm{cens}}),\n\\end{equation*}\\] where \\(F_T\\) is the cumulative distribution function (cdf) of survival time \\(T\\) (\\(F_X(x) = \\Pr[X \\leq x]\\) is standard notation for the cdf of a random variable \\(X\\)). The function \\(1 - F_T(t)\\) is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood \\[\\begin{eqnarray*}\np(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\prod_{n=1}^{N^{\\textrm{cens}}}\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)\n\\\\\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)^{N^{\\textrm{cens}}}.\n\\end{eqnarray*}\\]\nOn the log scale, that’s \\[\\begin{eqnarray*}\n\\log p(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\sum_{n=1}^N \\log \\textrm{exponential}(t_n \\mid \\lambda)\n\\\\\n& & { } + N^{\\textrm{cens}} \\cdot \\log \\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda).\n\\end{eqnarray*}\\]\nThe model can be completed with a standard lognormal prior on \\(\\lambda,\\) \\[\\begin{equation*}\n\\lambda \\sim \\textrm{lognormal}(0, 1),\n\\end{equation*}\\] which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed \\(\\textrm{lognormal}(0, 1)\\). In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.\n\n\nThe data for a simple survival analysis without covariates can be coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n}\nIn this program, N is the number of uncensored observations and t contains the times of the uncensored observations. There are a further N_cens items that are right censored at time t_cens. Right censoring means that if the time to failure is greater than\nt_cens, it is only observed that the part survived until time t_cens. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.\nThere is a single rate parameter, the inverse of which is the expected time to failure.\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nThe exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.\nmodel {\n  t ~ exponential(lambda);\n  target += N_cens * exponential_lccdf(t_cens | lambda);\n\n  lambda ~ lognormal(0, 1);\n}\nThe likelihood for rate lambda is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector t as a having an exponential distribution with rate lambda. This data model could have been written as\nfor (n in 1:N) {\n  t[n] ~ exponential(lambda);\n}\nThe log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate lambda. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,\nfor (n in 1:N)\n  target += exponential_lccdf(t_cens | lambda);\nto define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.\n\n\n\n\nThe Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.\nThe Weibull distribution is parameterized by a shape \\(\\alpha &gt; 0\\) and scale \\(\\sigma &gt; 0.\\) For an outcome \\(t \\geq 0\\), the Weibull distribution’s probability density function is \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid \\alpha, \\sigma)\n= \\frac{\\alpha}{\\sigma}\n  \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n  \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^{\\alpha}\\right).\n\\end{equation*}\\] In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter \\(\\beta &gt; 0\\) with probability density function \\[\\begin{equation*}\n\\textrm{exponential}(t \\mid \\beta) =\n\\beta\n\\cdot\n\\exp(-\\beta \\cdot t).\n\\end{equation*}\\] When \\(\\alpha = 1,\\) the Weibull distribution reduces to an exponential distribution, \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid 1, \\sigma)\n=\n\\textrm{exponential}\\!\\left(t \\,\\bigg|\\, \\frac{1}{\\sigma}\\right).\n\\end{equation*}\\] In other words, the Weibull is a continuous expansion of the exponential distribution.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then the expected survival time is \\[\\begin{equation*}\n\\mathbb{E}[T] = \\sigma \\cdot \\Gamma\\!\\left(1 + \\frac{1}{\\alpha}\\right),\n\\end{equation*}\\] where the \\(\\Gamma\\) function is the continuous completion of the factorial function (i.e., \\(\\Gamma(1 + n) = n!\\ \\) for \\(n \\in\n\\mathbb{N}\\)). As \\(\\alpha \\rightarrow 0\\) for a fixed \\(\\sigma\\) or as \\(\\sigma \\rightarrow \\infty\\) for a fixed \\(\\alpha\\), the expected survival time goes to infinity.\nThere are three regimes of the Weibull distribution.\n\n\\(\\alpha &lt; 1.\\) A subject is more likely to fail early. When \\(\\alpha\n&lt; 1,\\) the Weibull density approaches infinity as \\(t \\rightarrow 0.\\)\n\\(\\alpha = 1.\\) The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When \\(\\alpha = 1,\\) the Weibull distribution approaches \\(\\sigma\\) as \\(t\n\\rightarrow 0.\\)\n\\(\\alpha &gt; 1.\\) Subjects are less likely to fail early. When \\(\\alpha &lt; 1,\\) the Weibull density approaches zero as \\(t \\rightarrow 0.\\)\n\nWith \\(\\alpha \\leq 1,\\) the mode is zero (\\(t = 0\\)), whereas with \\(\\alpha &gt; 1,\\) the mode is nonzero (\\(t &gt; 0\\)).\n\n\nWith Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.\nparameters {\n  real&lt;lower=0&gt; beta;\n}\nmodel {\n  t ~ exponential(beta);\n  target += N_cens * exponential_lccdf(t_cens | beta);\n\n  beta ~ lognormal(0, 1);\n}\nThe Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (alpha) and scale (sigma) parameters.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  t ~ weibull(alpha, sigma);\n  target += N_cens * weibull_lccdf(t_cens | alpha, sigma);\n\n  alpha ~ lognormal(0, 1);\n  sigma ~ lognormal(0, 1);\n}\nAs usual, if more is known about expected survival times, alpha and sigma should be given more informative priors.\n\n\n\n\nSuppose that for each of \\(n \\in 1{:}N\\) items observed, both censored and uncensored, there is a covariate (row) vector \\(x_n \\in\n\\mathbb{R}^K.\\) For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.\nSurvival with covariates replaces what is essentially a simple regression with only an intercept \\(\\lambda\\) with a generalized linear model with a log link, where the rate for item \\(n\\) is \\[\\begin{equation*}\n\\lambda_n = \\exp(x_n \\cdot \\beta),\n\\end{equation*}\\] where \\(\\beta \\in \\mathbb{R}^K\\) is a \\(K\\)-vector of regression coefficients. Thus \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda_n).\n\\end{equation*}\\] The censored items have probability \\[\\begin{equation*}\n\\Pr[n\\textrm{-th censored}] =\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid x^{\\textrm{cens}}_n\n\\cdot \\beta).\n\\end{equation*}\\]\nThe covariates form an \\(N \\times K\\) data matrix, \\(x \\in\n\\mathbb{R}^{N \\times K}\\). An intercept can be introduced by adding a column of 1 values to \\(x\\).\nA Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (t and t_cens) corresponds to the ordering of items in the covariate matrices (x and x_cens).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n  int&lt;lower=0&gt; K;\n  matrix[N, K] x;\n  matrix[N_cens, K] x_cens;\n}\nparameters {\n  vector[K] gamma;\n}\nmodel {\n  gamma ~ normal(0, 2);\n\n  t ~ exponential(exp(x * gamma));\n  target += exponential_lccdf(t_cens | exp(x_cens * gamma));\n}\nBoth the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.\n\n\n\nSuppose \\(T\\) is a random variable representing a survival time, with a smooth cumulative distribution function \\[\\begin{equation*}\nF_T(t) = \\Pr[T \\leq t],\n\\end{equation*}\\] so that its probability density function is \\[\\begin{equation*}\np_T(t) = \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t).\n\\end{equation*}\\]\nThe survival function \\(S(t)\\) is the probability of surviving until at least time \\(t\\), which is just the complementary cumulative distribution function (ccdf) of the survival random variable \\(T\\), \\[\\begin{equation*}\nS(t) = 1 - F_T(t).\n\\end{equation*}\\] The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).\nThe hazard function \\(h(t)\\) is the instantaneous risk of not surviving past time \\(t\\) assuming survival until time \\(t\\), which is given by \\[\\begin{equation*}\nh(t) = \\frac{p_T(t)}{S(t)} = \\frac{p_T(t)}{1 - F_T(t)}.\n\\end{equation*}\\] The cumulative hazard function \\(H(t)\\) is defined to be the accumulated hazard over time, \\[\\begin{equation*}\nH(t) = \\int_0^t h(u) \\, \\textrm{d}u.\n\\end{equation*}\\]\nThe hazard function and survival function are related through the differential equation \\[\\begin{eqnarray*}\nh(t) & = & -\\frac{\\textrm{d}}{\\textrm{d}t} \\log S(t).\n\\\\[4pt]\n& = & -\\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} S(t)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} -(1 - F_T(t))\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} (F_T(t) - 1)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t)\n\\\\[4pt]\n& = & \\frac{p_T(t)}{S(t)}.\n\\end{eqnarray*}\\]\nIf \\(T \\sim \\textrm{exponential}(\\beta)\\) has an exponential distribution, then its hazard function is constant, \\[\\begin{eqnarray*}\nh(t \\mid \\beta)\n& = & \\frac{p_T(t \\mid \\beta)}{S(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\textrm{exponential}(t \\mid \\beta)}{1 - \\textrm{exponentialCCDF}(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {1 - (1 - \\exp(-\\beta \\cdot t))}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {\\exp(-\\beta \\cdot t)}\n\\\\[4pt]\n& = & \\beta.\n\\end{eqnarray*}\\] The exponential distribution is the only distribution of survival times with a constant hazard function.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then its hazard function is \\[\\begin{eqnarray*}\nh(t \\mid \\alpha, \\sigma)\n& = & \\frac{p_T(t \\mid \\alpha, \\sigma)}{S(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = & \\frac{\\textrm{Weibull}(t \\mid \\alpha, \\sigma)}{1 - \\textrm{WeibullCCDF}(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = &\n\\frac{\\frac{\\alpha}{\\sigma} \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n      \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma} \\right)^\\alpha\\right)}\n     {1 - \\left(1 -\n       \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^\\alpha\n     \\right)\\right)}      \n\\\\[4pt]\n& = & \\frac{\\alpha}{\\sigma}\n  \\cdot\n  \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}.\n\\end{eqnarray*}\\]\nIf \\(\\alpha = 1\\) the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when \\(\\alpha = 1\\)). When \\(\\alpha &gt; 1,\\) the hazard grows as time passes, whereas when \\(\\alpha &lt; 1,\\) it decreases as time passes.\n\n\n\nThe exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. Cox (1972) introduced a semi-parametric survival model specified directly in terms of a hazard function \\(h(t)\\) rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.\nLet \\(x_n \\in \\mathbb{R}^K\\) be a (row) vector of covariates for subject \\(n\\) so that the full covariate data matrix is \\(x \\in \\mathbb{R}^{N \\times\nK}\\). In Cox’s model, the hazard function for subject \\(n\\) is defined conditionally in terms of their covariates \\(x_n\\) and the parameter vector \\(\\gamma \\in \\mathbb{R}^K\\) as \\[\\begin{equation*}\nh(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n \\cdot \\gamma),\n\\end{equation*}\\] where \\(h_0(t)\\) is a shared baseline hazard function and \\(x_n \\cdot\n\\gamma = \\sum_{k=1}^K x_{n, k} \\cdot \\beta_k\\) is a row vector-vector product.\nIn the semi-parametric, proportional hazards model, the baseline hazard function \\(h_0(t)\\) is not modeled. This is why it is called “semi-parametric.” Only the factor \\(\\exp(x_n \\cdot \\gamma),\\) which determines how individual \\(n\\) varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”\nCox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function \\(h_0(t)\\) is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function1.\n\n\nCox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function \\(h_0(t)\\) is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.\nThe hazard function \\(h(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n\n\\cdot \\beta)\\) for subject \\(n\\) represents the instantaneous probability that subject \\(n\\) fails at time \\(t\\) given that it has survived until time \\(t.\\) The probability that subject \\(n\\) is the first to fail among \\(N\\) subjects is thus proportional to subject \\(n\\)’s hazard function, \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail at time } t]\n\\propto h(t \\mid x_n, \\beta).\n\\end{equation*}\\] Normalizing yields \\[\\begin{eqnarray*}\n\\Pr[n \\textrm{ first to fail at time } t]\n& = &  \\frac{h(t \\mid x_n, \\beta)}\n            {\\sum_{n' = 1}^N h(t \\mid x_{n'}, \\beta)}\n\\\\[4pt]\n& = &  \\frac{h_0(t) \\cdot \\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N h_0(t) \\cdot \\exp(x_{n'} \\cdot \\beta)}\n\\\\[4pt]\n& = &  \\frac{\\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{eqnarray*}\\]\nSuppose there are \\(N\\) subjects with strictly ordered survival times \\(t_1 &lt;\nt_2 &lt; \\cdots &lt; t_N\\) and covariate (row) vectors \\(x_1, \\ldots, x_N\\). Let \\(t^{\\textrm{cens}}\\) be the (right) censoring time and let \\(N^{\\textrm{obs}}\\) be the largest value of \\(n\\) such that \\(t_n \\leq\nt^{\\textrm{cens}}\\). This means \\(N^{\\textrm{obs}}\\) is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.\nWith failure times sorted in decreasing order, the partial likelihood for each observed subject \\(n \\in 1{:}N^{\\textrm{obs}}\\) can be expressed as \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail among } n, n + 1, \\ldots N]\n= \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.\nThe partial likelihood, defined in this form by Breslow (1975), is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). \\[\\begin{equation*}\n\\Pr[\\textrm{observed failures ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n= \\prod_{n = 1}^{N^{\\textrm{obs}}}\n  \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] On the log scale, \\[\\begin{eqnarray*}\n\\log \\Pr[\\textrm{obs.\\ fail ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n& = &\n\\sum_{n = 1}^{N^{\\textrm{obs}}}\n  \\log \\left(\n          \\frac{\\exp(x_n \\cdot \\beta)}\n               {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}\n       \\right)\n\\\\[4pt]\n& = & x_n \\cdot \\beta - \\log \\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)\n\\\\\n& = & x_n \\cdot \\beta - \\textrm{logSumExp}_{n' = n}^N \\ x_{n'} \\cdot \\beta,\n\\end{eqnarray*}\\] where \\[\\begin{equation*}\n\\textrm{logSumExp}_{n = a}^b \\ x_n\n= \\log \\sum_{n = a}^b \\exp(x_n)\n\\end{equation*}\\] is implemented so as to preserve numerical precision.\nThis likelihood follows the same approach to ranking as that developed by Plackett (1975) for estimating the probability of the order of the first few finishers in a horse race.\nA simple normal prior on the components of \\(\\beta\\) completes the model, \\[\\begin{equation*}\n\\beta \\sim \\textrm{normal}(0, 2).\n\\end{equation*}\\] This should be scaled based on knowledge of the predictors.\n\n\n\nTo simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.\ndata {\n  int&lt;lower=0&gt; K;          // num covariates\n\n  int&lt;lower=0&gt; N;          // num uncensored obs\n  vector[N] t;             // event time (non-strict decreasing)\n  matrix[N, K] x;          // covariates for uncensored obs\n\n  int N_c;                 // num censored obs\n  real&lt;lower=t[N]&gt; t_c;    // censoring time\n  matrix[N_c, K] x_c;      // covariates for censored obs\n}\nThe parameters are just the coefficients.\nparameters {\n  vector[K] beta;          // slopes (no intercept)\n}\nThe prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.\nmodel {\n  beta ~ normal(0, 2);\n  ...\nThe log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\nThese vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.\n  real log_denom = log_sum_exp(log_theta_c);\nThen, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.\n  for (n in 1:N) {\n    log_denom = log_sum_exp(log_denom, log_theta[n]);\n    target += log_theta[n] - log_denom;   // log likelihood\n  }\nThe running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.\ntransformed data {\n  for (n in 2:N) {\n    if (!(t[n] &lt; t[n - 1])) {\n      reject(\"times must be strictly decreasing, but found\"\n             \"!(t[\", n, \"] &lt; t[, \", (n - 1), \"])\");\n    }   \n  }\n}\n\n\n\nTechnically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all \\(N!\\) permutations for a set of \\(N\\) subjects with tied survival times is not tractable. As an alternative, Efron (1977) introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.\nIn the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.\nfunctions {\n  int num_unique_starts(vector t) {\n    if (size(t) == 0) return 0;\n    int us = 1;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) us += 1;\n    }\n    return us;\n  }\nThis is then used to compute the value J to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.\n  array[] int unique_starts(vector t, int J) {\n    array[J + 1] int starts;\n    if (J == 0) return starts;\n    starts[1] = 1;\n    int pos = 2;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) {\n    starts[pos] = n;\n    pos += 1;\n      }\n    }\n    starts[J + 1] = size(t) + 1;\n    return starts;\n  }\n}\nThe data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.\ntransformed data {\n  int&lt;lower=0&gt; J = num_unique_starts(t);\n  array[J + 1] int&lt;lower=0&gt; starts = unique_starts(t, J);\n}\nFor each unique survival time j in 1:J, the subjects indexed from starts[j] to starts[j + 1] - 1 (inclusive) share the same survival time. The number of elements with survival time j is thus (starts[j + 1] - 1) - starts[j] + 1, or just starts[j + 1] - starts[j].\nThe parameters and prior are also the same—just a vector beta of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables J and starts.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\n  real log_denom_lhs = log_sum_exp(log_theta_c);\n  for (j in 1:J) {\n    int start = starts[j];\n    int end = starts[j + 1] - 1;\n    int len = end - start + 1;\n    real log_len = log(len);\n    real numerator = sum(log_theta[start:end]);\n    log_denom_lhs = log_sum_exp(log_denom_lhs,\n                                log_sum_exp(log_theta[start:end]));\n    vector[len] diff;\n    for (ell in 1:len) {\n      diff[ell] = log_diff_exp(log_denom_lhs,\n                               log(ell - 1) - log_len\n                               + log_sum_exp(log_theta[start:end]));\n    }\n    target += numerator - sum(diff);\n  }\nThe special function log_diff_exp is defined as\n\\[\\begin{equation*}\n\\textrm{logDiffExp}(u, v) = \\log(\\exp(u) - \\exp(v)).\n\\end{equation*}\\]\nBecause of how J and starts are constructed, the length len will always be strictly positive so that the log is well defined.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#autoregressive-moving-average-models",
-    "href": "stan-users-guide/time-series.html#autoregressive-moving-average-models",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/survival.html#exponential-survival-model",
+    "href": "stan-users-guide/survival.html#exponential-survival-model",
+    "title": "Survival Models",
     "section": "",
-    "text": "Autoregressive moving-average models (ARMA), combine the predictors of the autoregressive model and the moving average model. An ARMA(1,1) model, with a single state of history, can be encoded in Stan as follows.\ndata {\n  int&lt;lower=1&gt; T;            // num observations\n  array[T] real y;                 // observed outputs\n}\nparameters {\n  real mu;                   // mean coeff\n  real phi;                  // autoregression coeff\n  real theta;                // moving avg coeff\n  real&lt;lower=0&gt; sigma;       // noise scale\n}\nmodel {\n  vector[T] nu;              // prediction for time t\n  vector[T] err;             // error for time t\n  nu[1] = mu + phi * mu;     // assume err[0] == 0\n  err[1] = y[1] - nu[1];\n  for (t in 2:T) {\n    nu[t] = mu + phi * y[t - 1] + theta * err[t - 1];\n    err[t] = y[t] - nu[t];\n  }\n  mu ~ normal(0, 10);        // priors\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err ~ normal(0, sigma);    // error model\n}\nThe data are declared in the same way as the other time-series regressions and the parameters are documented in the code.\nIn the model block, the local vector nu stores the predictions and err the errors. These are computed similarly to the errors in the moving average models described in the previous section.\nThe priors are weakly informative for stationary processes. The data model only involves the error term, which is efficiently vectorized here.\nOften in models such as these, it is desirable to inspect the calculated error terms. This could easily be accomplished in Stan by declaring err as a transformed parameter, then defining it the same way as in the model above. The vector nu could still be a local variable, only now it will be in the transformed parameter block.\nWayne Folta suggested encoding the model without local vector variables as follows.\nmodel {\n  real err;\n  mu ~ normal(0, 10);\n  phi ~ normal(0, 2);\n  theta ~ normal(0, 2);\n  sigma ~ cauchy(0, 5);\n  err = y[1] - (mu + phi * mu);\n  err ~ normal(0, sigma);\n  for (t in 2:T) {\n    err = y[t] - (mu + phi * y[t - 1] + theta * err);\n    err ~ normal(0, sigma);\n  }\n}\nThis approach to ARMA models illustrates how local variables, such as err in this case, can be reused in Stan. Folta’s approach could be extended to higher order moving-average models by storing more than one error term as a local variable and reassigning them in the loop.\nBoth encodings are fast. The original encoding has the advantage of vectorizing the normal distribution, but it uses a bit more memory. A halfway point would be to vectorize just err.\n\n\nMA and ARMA models are not identifiable if the roots of the characteristic polynomial for the MA part lie inside the unit circle, so it’s necessary to add the following constraint3\nreal&lt;lower=-1, upper=1&gt; theta;\nWhen the model is run without the constraint, using synthetic data generated from the model, the simulation can sometimes find modes for (theta, phi) outside the \\([-1,1]\\) interval, which creates a multiple mode problem in the posterior and also causes the NUTS tree depth to get large (often above 10). Adding the constraint both improves the accuracy of the posterior and dramatically reduces the tree depth, which speeds up the simulation considerably (typically by much more than an order of magnitude).\nFurther, unless one thinks that the process is really non-stationary, it’s worth adding the following constraint to ensure stationarity.\nreal&lt;lower=-1, upper=1&gt; phi;",
+    "text": "The exponential distribution is commonly used in survival models where there is a constant risk of failure that does not go up the longer a subject survives. This is because the exponential distribution is memoryless in sense that if \\(T \\sim \\textrm{exponential}(\\lambda)\\) for some rate \\(\\lambda &gt; 0,\\) then \\[\\begin{equation*}\n\\Pr[T &gt; t] = \\Pr[T &gt; t + t' \\mid T &gt; t'].\n\\end{equation*}\\] If component survival times are distributed exponentially, it means the distribution of time to failure is the same no matter how long the item has already survived. This can be a reasonable assumption for electronic components, but is not a reasonable model for animal survival.\nThe exponential survival model has a single parameter for the rate, which assumes all subjects have the same distribution of failure time (this assumption is relaxed in the next section by introducing per-subject covariates). With the rate parameterization, the expected survival time for a component with survival time represented as the random variable \\(T\\) is \\[\\begin{equation*}\n\\mathbb{E}[T \\mid \\lambda] = \\frac{1}{\\lambda}.\n\\end{equation*}\\] The exponential distribution is sometimes parameterized in terms of a scale (i.e., inverse rate) \\(\\beta = 1 / \\lambda\\).\nThe data for a survival model consists of two components. First, there is a vector \\(t \\in (0, \\infty)^N\\) of \\(N\\) observed failure times. Second, there is a censoring time \\(t^{\\textrm{cens}}\\) such that failure times greater than \\(t^{\\textrm{cens}}\\) are not observed. The censoring time assumption imposes a constraint which requires \\(t_n &lt; t^{\\textrm{cens}}\\) for all \\(n \\in 1{:}N.\\) For the censored subjects, the only thing required in the model is their total count, \\(N^\\textrm{cens}\\) (their covariates are also required for models with covariates).\nThe model for the observed failure times is exponential, so that for \\(n \\in 1{:}N,\\) \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda).\n\\end{equation*}\\]\nThe model for the censored failure times is also exponential. All that is known of a censored item is that its failure time is greater than the censoring time, so each censored item contributes a factor to the likelihood of \\[\\begin{equation*}\n\\Pr[T &gt; t^{\\textrm{cens}}] = 1 - F_T(t^{\\textrm{cens}}),\n\\end{equation*}\\] where \\(F_T\\) is the cumulative distribution function (cdf) of survival time \\(T\\) (\\(F_X(x) = \\Pr[X \\leq x]\\) is standard notation for the cdf of a random variable \\(X\\)). The function \\(1 - F_T(t)\\) is the complementary cumulative distribution function (ccdf), and it is used directly to define the likelihood \\[\\begin{eqnarray*}\np(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\prod_{n=1}^{N^{\\textrm{cens}}}\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)\n\\\\\n& = &\n\\prod_{n=1}^N \\textrm{exponential}(t_n \\mid \\lambda)\n\\cdot\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda)^{N^{\\textrm{cens}}}.\n\\end{eqnarray*}\\]\nOn the log scale, that’s \\[\\begin{eqnarray*}\n\\log p(t, t^{\\textrm{cens}}, N^{\\textrm{cens}} \\mid \\lambda)\n& = &\n\\sum_{n=1}^N \\log \\textrm{exponential}(t_n \\mid \\lambda)\n\\\\\n& & { } + N^{\\textrm{cens}} \\cdot \\log \\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid \\lambda).\n\\end{eqnarray*}\\]\nThe model can be completed with a standard lognormal prior on \\(\\lambda,\\) \\[\\begin{equation*}\n\\lambda \\sim \\textrm{lognormal}(0, 1),\n\\end{equation*}\\] which is reasonable if failure times are in the range of 0.1 to 10 time units, because that’s roughly the 95% central interval for a variable distributed \\(\\textrm{lognormal}(0, 1)\\). In general, the range of the prior (and likelihood!) should be adjusted with prior knowledge of expected failure times.\n\n\nThe data for a simple survival analysis without covariates can be coded as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n}\nIn this program, N is the number of uncensored observations and t contains the times of the uncensored observations. There are a further N_cens items that are right censored at time t_cens. Right censoring means that if the time to failure is greater than\nt_cens, it is only observed that the part survived until time t_cens. In the case where there are no covariates, the model only needs the number of censored items because they all share the same censoring time.\nThere is a single rate parameter, the inverse of which is the expected time to failure.\nparameters {\n  real&lt;lower=0&gt; lambda;\n}\nThe exponential survival model and the prior are coded directly using vectorized distribution and ccdf statements. This both simplifies the code and makes it more computationally efficient by sharing computation across instances.\nmodel {\n  t ~ exponential(lambda);\n  target += N_cens * exponential_lccdf(t_cens | lambda);\n\n  lambda ~ lognormal(0, 1);\n}\nThe likelihood for rate lambda is just the density of exponential distribution for observed failure time. The Stan code is vectorized, modeling each entry of the vector t as a having an exponential distribution with rate lambda. This data model could have been written as\nfor (n in 1:N) {\n  t[n] ~ exponential(lambda);\n}\nThe log likelihood contribution given censored items is the number of censored items times the log complementary cumulative distribution function (lccdf) at the censoring time of the exponential distribution with rate lambda. The log likelihood terms arising from the censored events could have been added to the target log density one at a time,\nfor (n in 1:N)\n  target += exponential_lccdf(t_cens | lambda);\nto define the same log density, but it is much more efficient computationally to multiply by a constant than do a handful of sequential additions.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#stochastic-volatility-models",
-    "href": "stan-users-guide/time-series.html#stochastic-volatility-models",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/survival.html#weibull-survival-model",
+    "href": "stan-users-guide/survival.html#weibull-survival-model",
+    "title": "Survival Models",
     "section": "",
-    "text": "Stochastic volatility models treat the volatility (i.e., variance) of a return on an asset, such as an option to buy a security, as following a latent stochastic process in discrete time (Kim, Shephard, and Chib 1998). The data consist of mean corrected (i.e., centered) returns \\(y_t\\) on an underlying asset at \\(T\\) equally spaced time points. Kim et al. formulate a typical stochastic volatility model using the following regression-like equations, with a latent parameter \\(h_t\\) for the log volatility, along with parameters \\(\\mu\\) for the mean log volatility, and \\(\\phi\\) for the persistence of the volatility term. The variable \\(\\epsilon_t\\) represents the white-noise shock (i.e., multiplicative error) on the asset return at time \\(t\\), whereas \\(\\delta_t\\) represents the shock on volatility at time \\(t\\). \\[\\begin{align*}\ny_t        &=    \\epsilon_t \\exp(h_t / 2) \\\\\nh_{t+1}    &=    \\mu + \\phi (h_t - \\mu) + \\delta_t \\sigma \\\\\nh_1        &\\sim \\textsf{normal}\\left( \\mu, \\frac{\\sigma}{\\sqrt{1 - \\phi^2}} \\right) \\\\\n\\epsilon_t &\\sim \\textsf{normal}(0,1) \\\\\n\\delta_t   &\\sim \\textsf{normal}(0,1)\n\\end{align*}\\]\nRearranging the first line, \\(\\epsilon_t = y_t \\exp(-h_t / 2)\\), allowing the distribution for \\(y_t\\) to be written as \\[\ny_t \\sim \\textsf{normal}(0,\\exp(h_t/2)).\n\\] The recurrence equation for \\(h_{t+1}\\) may be combined with the scaling of \\(\\delta_t\\) to yield the distribution \\[\nh_t \\sim \\mathsf{normal}(\\mu + \\phi(h_{t-1} - \\mu), \\sigma).\n\\] This formulation can be directly encoded, as shown in the following Stan model.\ndata {\n  int&lt;lower=0&gt; T;   // # time points (equally spaced)\n  vector[T] y;      // mean corrected return at time t\n}\nparameters {\n  real mu;                     // mean log volatility\n  real&lt;lower=-1, upper=1&gt; phi; // persistence of volatility\n  real&lt;lower=0&gt; sigma;         // white noise shock scale\n  vector[T] h;                 // log volatility at time t\n}\nmodel {\n  phi ~ uniform(-1, 1);\n  sigma ~ cauchy(0, 5);\n  mu ~ cauchy(0, 10);\n  h[1] ~ normal(mu, sigma / sqrt(1 - phi * phi));\n  for (t in 2:T) {\n    h[t] ~ normal(mu + phi * (h[t - 1] -  mu), sigma);\n  }\n  for (t in 1:T) {\n    y[t] ~ normal(0, exp(h[t] / 2));\n  }\n}\nCompared to the Kim et al. formulation, the Stan model adds priors for the parameters \\(\\phi\\), \\(\\sigma\\), and \\(\\mu\\). The shock terms \\(\\epsilon_t\\) and \\(\\delta_t\\) do not appear explicitly in the model, although they could be calculated efficiently in a generated quantities block.\nThe posterior of a stochastic volatility model such as this one typically has high posterior variance. For example, simulating 500 data points from the above model with \\(\\mu = -1.02\\), \\(\\phi = 0.95\\), and \\(\\sigma = 0.25\\) leads to 95% posterior intervals for \\(\\mu\\) of \\((-1.23, -0.54)\\), for \\(\\phi\\) of \\((0.82, 0.98)\\), and for \\(\\sigma\\) of \\((0.16, 0.38)\\).\nThe NUTS draws show a high degree of autocorrelation, both for this model and the stochastic volatility model evaluated in (Hoffman and Gelman 2014). Using a non-diagonal mass matrix provides faster convergence and higher effective sample size than a diagonal mass matrix, but will not scale to large values of \\(T\\).\nIt is relatively straightforward to speed up the effective sample size per second generated by this model by one or more orders of magnitude. First, the distribution statements for return \\(y\\) is easily vectorized to\ny ~ normal(0, exp(h / 2));\nThis speeds up the iterations, but does not change the effective sample size because the underlying parameterization and log probability function have not changed. Mixing is improved by reparameterizing in terms of a standardized volatility, then rescaling. This requires a standardized parameter h_std to be declared instead of h.\nparameters {\n  // ...\n  vector[T] h_std;  // std log volatility time t\n}\nThe original value of h is then defined in a transformed parameter block.\ntransformed parameters {\n  vector[T] h = h_std * sigma;  // now h ~ normal(0, sigma)\n  h[1] /= sqrt(1 - phi * phi);  // rescale h[1]\n  h += mu;\n  for (t in 2:T) {\n    h[t] += phi * (h[t - 1] - mu);\n  }\n}\nThe first assignment rescales h_std to have a \\(\\textsf{normal}(0,\\sigma)\\) distribution and temporarily assigns it to h. The second assignment rescales h[1] so that its prior differs from that of h[2] through h[T]. The next assignment supplies a mu offset, so that h[2] through h[T] are now distributed \\(\\textsf{normal}(\\mu,\\sigma)\\); note that this shift must be done after the rescaling of h[1]. The final loop adds in the moving average so that h[2] through h[T] are appropriately modeled relative to phi and mu.\nAs a final improvement, the distribution statements for h[1] to h[T] are replaced with a single vectorized standard normal distribution statement.\nmodel {\n  // ...\n  h_std ~ std_normal();\n}\nAlthough the original model can take hundreds and sometimes thousands of iterations to converge, the reparameterized model reliably converges in tens of iterations. Mixing is also dramatically improved, which results in higher effective sample sizes per iteration. Finally, each iteration runs in roughly a quarter of the time of the original iterations.",
+    "text": "The Weibull distribution is a popular alternative to the exponential distribution in cases where there is a decreasing probability of survival as a subject gets older. The Weibull distribution models this by generalizing the exponential distribution to include a power-law trend.\nThe Weibull distribution is parameterized by a shape \\(\\alpha &gt; 0\\) and scale \\(\\sigma &gt; 0.\\) For an outcome \\(t \\geq 0\\), the Weibull distribution’s probability density function is \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid \\alpha, \\sigma)\n= \\frac{\\alpha}{\\sigma}\n  \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n  \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^{\\alpha}\\right).\n\\end{equation*}\\] In contrast, recall that the exponential distribution can be expressed using a rate (inverse scale) parameter \\(\\beta &gt; 0\\) with probability density function \\[\\begin{equation*}\n\\textrm{exponential}(t \\mid \\beta) =\n\\beta\n\\cdot\n\\exp(-\\beta \\cdot t).\n\\end{equation*}\\] When \\(\\alpha = 1,\\) the Weibull distribution reduces to an exponential distribution, \\[\\begin{equation*}\n\\textrm{Weibull}(t \\mid 1, \\sigma)\n=\n\\textrm{exponential}\\!\\left(t \\,\\bigg|\\, \\frac{1}{\\sigma}\\right).\n\\end{equation*}\\] In other words, the Weibull is a continuous expansion of the exponential distribution.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then the expected survival time is \\[\\begin{equation*}\n\\mathbb{E}[T] = \\sigma \\cdot \\Gamma\\!\\left(1 + \\frac{1}{\\alpha}\\right),\n\\end{equation*}\\] where the \\(\\Gamma\\) function is the continuous completion of the factorial function (i.e., \\(\\Gamma(1 + n) = n!\\ \\) for \\(n \\in\n\\mathbb{N}\\)). As \\(\\alpha \\rightarrow 0\\) for a fixed \\(\\sigma\\) or as \\(\\sigma \\rightarrow \\infty\\) for a fixed \\(\\alpha\\), the expected survival time goes to infinity.\nThere are three regimes of the Weibull distribution.\n\n\\(\\alpha &lt; 1.\\) A subject is more likely to fail early. When \\(\\alpha\n&lt; 1,\\) the Weibull density approaches infinity as \\(t \\rightarrow 0.\\)\n\\(\\alpha = 1.\\) The Weibull distribution reduces to the exponential distribution, with a constant rate of failure over time. When \\(\\alpha = 1,\\) the Weibull distribution approaches \\(\\sigma\\) as \\(t\n\\rightarrow 0.\\)\n\\(\\alpha &gt; 1.\\) Subjects are less likely to fail early. When \\(\\alpha &lt; 1,\\) the Weibull density approaches zero as \\(t \\rightarrow 0.\\)\n\nWith \\(\\alpha \\leq 1,\\) the mode is zero (\\(t = 0\\)), whereas with \\(\\alpha &gt; 1,\\) the mode is nonzero (\\(t &gt; 0\\)).\n\n\nWith Stan, one can just swap the exponential distribution for the Weibull distribution with the appropriate parameters and the model remains essentially the same. Recall the exponential model’s parameters and model block.\nparameters {\n  real&lt;lower=0&gt; beta;\n}\nmodel {\n  t ~ exponential(beta);\n  target += N_cens * exponential_lccdf(t_cens | beta);\n\n  beta ~ lognormal(0, 1);\n}\nThe Stan program for the Weibull model just swaps in the Weibull distribution and complementary cumulative distribution function with shape (alpha) and scale (sigma) parameters.\nparameters {\n  real&lt;lower=0&gt; alpha;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  t ~ weibull(alpha, sigma);\n  target += N_cens * weibull_lccdf(t_cens | alpha, sigma);\n\n  alpha ~ lognormal(0, 1);\n  sigma ~ lognormal(0, 1);\n}\nAs usual, if more is known about expected survival times, alpha and sigma should be given more informative priors.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#hmms.section",
-    "href": "stan-users-guide/time-series.html#hmms.section",
-    "title": "Time-Series Models",
+    "objectID": "stan-users-guide/survival.html#survival-with-covariates",
+    "href": "stan-users-guide/survival.html#survival-with-covariates",
+    "title": "Survival Models",
     "section": "",
-    "text": "A Hidden Markov model is a probabilistic model over \\(N\\) observations \\(y_{1:N}\\) and \\(N\\) hidden states \\(z_{1:N}\\). This models is defined by the conditional distributions \\(p(y_n \\mid z_n, \\phi)\\) and \\(p(z_n \\mid z_{n-1}, \\phi)\\). Here we make the dependency on additional model parameters \\(\\phi\\) explicit. (\\(\\phi\\) may be a vector of parameters.) The complete data likelihood is then \\[\np(y, z \\mid \\phi) = \\prod_n p(y_n \\mid z_n, \\phi) p(z_n \\mid z_{n - 1}, \\phi)\n\\] When \\(z_{1:N}\\) is continuous, the user can explicitly encode these distributions in Stan and use Markov chain Monte Carlo to integrate \\(z\\) out.\nWhen each state \\(z\\) takes a value over a discrete and finite set, say \\(\\{1, 2, ..., K\\}\\), we can use Stan’s suite of HMM functions to marginalize out \\(z_{1:N}\\) and compute \\[\np(y_{1:N} \\mid \\phi) = \\int_{\\mathcal Z} p(y, z \\mid \\phi) \\text d z.\n\\] We start by defining the conditional observation distribution, stored in a \\(K \\times N\\) matrix \\(\\omega\\) with \\[\n\\omega_{kn} = p(y_n \\mid z_n = k, \\phi).\n\\] Next, we introduce the \\(K \\times K\\) transition matrix, \\(\\Gamma\\), with \\[\n\\Gamma_{ij} = p(z_n = j \\mid z_{n - 1} = i, \\phi).\n\\] (This is a right-stochastic matrix.) Finally, we define the initial state \\(K\\)-vector \\(\\rho\\), with \\[\n\\rho_k = p(z_0 = k \\mid \\phi).\n\\] It is common practice to set \\(\\rho\\) to be the stationary distribution of the HMM, that is \\(\\rho\\) is the first eigenvector of \\(\\Gamma\\) and solves \\(\\Gamma \\rho = \\rho\\).\nAs an example, consider a three-state model with \\(K=3\\). The observations are normally distributed conditional on the HMM states with \\[\n  y_n \\sim \\text{normal}(\\mu_k, \\sigma),\n\\] where \\(\\mu = (1, 5, 9)\\) and the standard deviation \\(\\sigma\\) is the same across all observations. The model is then\ndata {\n  int N;  // Number of observations\n  array[N] real y;\n}\n\nparameters {\n  // Rows of the transition matrix\n  array[3] simplex[3] gamma_arr;\n\n  // Initial state\n  simplex[3] rho;\n\n  // Parameters of measurement model\n  vector[3] mu;\n  real&lt;lower = 0.0&gt; sigma;\n}\n\ntransformed parameters {\n  // Build transition matrix\n  matrix[3, 3] gamma;\n  for (k in 1:3) gamma[k, ] = to_row_vector(gamma_arr[k]);\n\n  // Compute the log likelihoods in each possible state\n  matrix[3, N] log_omega;\n  for (n in 1:N) {\n    for (i in 1:3) {\n      log_omega[i, n] = normal_lpdf(y[n] | mu[i], sigma);\n    }\n  }\n}\n\nmodel {\n  // prior\n  mu ~ normal(0, 1);\n  sigma ~ normal(0, 1);\n  \n  // no explicit prior on gamma_arr, meaning we default to a\n  // uniform prior over the simplexes.\n\n  // Increment target by log p(y | mu, sigma, Gamma, rho)\n  target += hmm_marginal(log_omega, gamma, rho);\n}\nThe last function hmm_marginal takes in all the ingredients of the HMM and computes the relevant log marginal distribution, \\(\\log p(y \\mid \\phi)\\).\nIf we desire draws from the posterior distribution of \\(z\\), we use the generated quantities block and draw, for each sample \\(\\phi\\), a sample from \\(p(z \\mid y, \\phi)\\). In effect, MCMC produces draws from \\(p(\\phi \\mid y)\\) and with the draws in generated quantities, we obtain draws from \\(p(\\phi \\mid y) p(z \\mid y, \\phi) = p(z, \\phi \\mid y)\\). It is also possible to compute the posterior probbability of each hidden state, that is \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). Averagging these probabilities over all MCMC draws, we obtain \\(\\text{Pr}(z_n = k \\mid y)\\).\ngenerated quantities {\n  array[N] int latent_states = hmm_latent_rng(log_omega, gamma, rho);\n  matrix[3, N] hidden_probs = hmm_hidden_state_prob(log_omega, gamma, rho);\n}\nhmm_hidden_state_prob returns the marginal probabilities of each state, \\(\\text{Pr}(z_n = k \\mid \\phi, y)\\). This function cannot be used to compute the joint probability \\(\\text{Pr}(z \\mid \\phi, y)\\), because such calculation requires accounting for the posterior correlation between the different components of \\(z\\). Therefore, hidden_probs should not be used to obtain posterior draws. Instead, users should rely on hmm_latent_rng.\ngenerated quantities {\n   array[N] int&lt;lower=1, upper=K&gt; z = hmm_latent_rng(...fill-in params here to match example...);\n}\nThe example in this section is derived from the more detailed case study by Ben Bales: https://mc-stan.org/users/documentation/case-studies/hmm-example.html.",
+    "text": "Suppose that for each of \\(n \\in 1{:}N\\) items observed, both censored and uncensored, there is a covariate (row) vector \\(x_n \\in\n\\mathbb{R}^K.\\) For example, a clinical trial may include the age (or a one-hot encoding of an age group) and the sex of a participant; an electronic component might include a one-hot encoding of the factory at which it was manufactured and a covariate for the load under which it has been run.\nSurvival with covariates replaces what is essentially a simple regression with only an intercept \\(\\lambda\\) with a generalized linear model with a log link, where the rate for item \\(n\\) is \\[\\begin{equation*}\n\\lambda_n = \\exp(x_n \\cdot \\beta),\n\\end{equation*}\\] where \\(\\beta \\in \\mathbb{R}^K\\) is a \\(K\\)-vector of regression coefficients. Thus \\[\\begin{equation*}\nt_n \\sim \\textrm{exponential}(\\lambda_n).\n\\end{equation*}\\] The censored items have probability \\[\\begin{equation*}\n\\Pr[n\\textrm{-th censored}] =\n\\textrm{exponentialCCDF}(t^{\\textrm{cens}} \\mid x^{\\textrm{cens}}_n\n\\cdot \\beta).\n\\end{equation*}\\]\nThe covariates form an \\(N \\times K\\) data matrix, \\(x \\in\n\\mathbb{R}^{N \\times K}\\). An intercept can be introduced by adding a column of 1 values to \\(x\\).\nA Stan program for the exponential survival model with covariates is as follows. It relies on the fact that the order of failure times (t and t_cens) corresponds to the ordering of items in the covariate matrices (x and x_cens).\ndata {\n  int&lt;lower=0&gt; N;\n  vector[N] t;\n  int&lt;lower=0&gt; N_cens;\n  real&lt;lower=0&gt; t_cens;\n  int&lt;lower=0&gt; K;\n  matrix[N, K] x;\n  matrix[N_cens, K] x_cens;\n}\nparameters {\n  vector[K] gamma;\n}\nmodel {\n  gamma ~ normal(0, 2);\n\n  t ~ exponential(exp(x * gamma));\n  target += exponential_lccdf(t_cens | exp(x_cens * gamma));\n}\nBoth the distribution statement for uncensored times and the log density increment statement for censored times are vectorized, one in terms of the exponential distribution and one in terms of the log complementary cumulative distribution function.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/time-series.html#footnotes",
-    "href": "stan-users-guide/time-series.html#footnotes",
-    "title": "Time-Series Models",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe intercept in this model is \\(\\alpha / (1 - \\beta)\\). An alternative parameterization in terms of an intercept \\(\\gamma\\) suggested Mark Scheuerell on GitHub is \\(y_n \\sim \\textsf{normal}\\left(\\gamma + \\beta \\cdot (y_{n-1} - \\gamma), \\sigma\\right)\\).↩︎\nIn practice, it can be useful to remove the constraint to test whether a non-stationary set of coefficients provides a better fit to the data. It can also be useful to add a trend term to the model, because an unfitted trend will manifest as non-stationarity.↩︎\nThis subsection is a lightly edited comment of Jonathan Gilligan’s on GitHub; see https://github.com/stan-dev/stan/issues/1617#issuecomment-160249142.↩︎",
+    "objectID": "stan-users-guide/survival.html#hazard-and-survival-functions",
+    "href": "stan-users-guide/survival.html#hazard-and-survival-functions",
+    "title": "Survival Models",
+    "section": "",
+    "text": "Suppose \\(T\\) is a random variable representing a survival time, with a smooth cumulative distribution function \\[\\begin{equation*}\nF_T(t) = \\Pr[T \\leq t],\n\\end{equation*}\\] so that its probability density function is \\[\\begin{equation*}\np_T(t) = \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t).\n\\end{equation*}\\]\nThe survival function \\(S(t)\\) is the probability of surviving until at least time \\(t\\), which is just the complementary cumulative distribution function (ccdf) of the survival random variable \\(T\\), \\[\\begin{equation*}\nS(t) = 1 - F_T(t).\n\\end{equation*}\\] The survival function appeared in the Stan model in the previous section as the likelihood for items that did not fail during the period of the experiment (i.e., the censored failure times for the items that survived through the trial period).\nThe hazard function \\(h(t)\\) is the instantaneous risk of not surviving past time \\(t\\) assuming survival until time \\(t\\), which is given by \\[\\begin{equation*}\nh(t) = \\frac{p_T(t)}{S(t)} = \\frac{p_T(t)}{1 - F_T(t)}.\n\\end{equation*}\\] The cumulative hazard function \\(H(t)\\) is defined to be the accumulated hazard over time, \\[\\begin{equation*}\nH(t) = \\int_0^t h(u) \\, \\textrm{d}u.\n\\end{equation*}\\]\nThe hazard function and survival function are related through the differential equation \\[\\begin{eqnarray*}\nh(t) & = & -\\frac{\\textrm{d}}{\\textrm{d}t} \\log S(t).\n\\\\[4pt]\n& = & -\\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} S(t)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} -(1 - F_T(t))\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} (F_T(t) - 1)\n\\\\[4pt]\n& = & \\frac{1}{S(t)} \\frac{\\textrm{d}}{\\textrm{d}t} F_T(t)\n\\\\[4pt]\n& = & \\frac{p_T(t)}{S(t)}.\n\\end{eqnarray*}\\]\nIf \\(T \\sim \\textrm{exponential}(\\beta)\\) has an exponential distribution, then its hazard function is constant, \\[\\begin{eqnarray*}\nh(t \\mid \\beta)\n& = & \\frac{p_T(t \\mid \\beta)}{S(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\textrm{exponential}(t \\mid \\beta)}{1 - \\textrm{exponentialCCDF}(t \\mid \\beta)}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {1 - (1 - \\exp(-\\beta \\cdot t))}\n\\\\[4pt]\n& = & \\frac{\\beta \\cdot \\exp(-\\beta \\cdot t)}\n           {\\exp(-\\beta \\cdot t)}\n\\\\[4pt]\n& = & \\beta.\n\\end{eqnarray*}\\] The exponential distribution is the only distribution of survival times with a constant hazard function.\nIf \\(T \\sim \\textrm{Weibull}(\\alpha, \\sigma),\\) then its hazard function is \\[\\begin{eqnarray*}\nh(t \\mid \\alpha, \\sigma)\n& = & \\frac{p_T(t \\mid \\alpha, \\sigma)}{S(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = & \\frac{\\textrm{Weibull}(t \\mid \\alpha, \\sigma)}{1 - \\textrm{WeibullCCDF}(t \\mid \\alpha, \\sigma)}\n\\\\[4pt]\n& = &\n\\frac{\\frac{\\alpha}{\\sigma} \\cdot \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}\n      \\cdot \\exp\\left(-\\left(\\frac{t}{\\sigma} \\right)^\\alpha\\right)}\n     {1 - \\left(1 -\n       \\exp\\left(-\\left(\\frac{t}{\\sigma}\\right)^\\alpha\n     \\right)\\right)}      \n\\\\[4pt]\n& = & \\frac{\\alpha}{\\sigma}\n  \\cdot\n  \\left( \\frac{t}{\\sigma} \\right)^{\\alpha - 1}.\n\\end{eqnarray*}\\]\nIf \\(\\alpha = 1\\) the hazard is constant over time (which also follows from the fact that the Weibull distribution reduces to the exponential distribution when \\(\\alpha = 1\\)). When \\(\\alpha &gt; 1,\\) the hazard grows as time passes, whereas when \\(\\alpha &lt; 1,\\) it decreases as time passes.",
     "crumbs": [
       "Stan Users Guide",
       "Example Models",
-      "Time-Series Models"
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html",
-    "href": "stan-users-guide/user-functions.html",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/survival.html#proportional-hazards-model",
+    "href": "stan-users-guide/survival.html#proportional-hazards-model",
+    "title": "Survival Models",
     "section": "",
-    "text": "This chapter explains functions from a user perspective with examples; see the language reference for a full specification. User-defined functions allow computations to be encapsulated into a single named unit and invoked elsewhere by name. Similarly, functions allow complex procedures to be broken down into more understandable components. Writing modular code using descriptively named functions is easier to understand than a monolithic program, even if the latter is heavily commented.1\n\n\nHere’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.\nfunctions {\n  real relative_diff(real x, real y) {\n    real abs_diff;\n    real avg_scale;\n    abs_diff = abs(x - y);\n    avg_scale = (abs(x) + abs(y)) / 2;\n    return abs_diff / avg_scale;\n  }\n}\n// ...\ngenerated quantities {\n  real rdiff;\n  rdiff = relative_diff(alpha, beta);\n}\nThe function is named relative_diff, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.\n\n\nAll functions are defined in their own block, which is labeled functions and must appear before all other program blocks. The user-defined functions block is optional.\n\n\n\nThe body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.\n\n\n\nReturn statements, such as the one on the last line of the definition of relative_diff above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.\nIf an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a fatal_error statement.\n\n\nRejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).\nThe most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.\nreal dbl_sqrt(real x) {\n  if (!(x &gt;= 0)) {\n    reject(\"dbl_sqrt(x): x must be positive; found x = \", x);\n  }\n  return 2 * sqrt(x);\n}\nThe negation of the positive test is important, because it also catches the case where x is a not-a-number value. If the condition had been coded as (x &lt; 0) it would not catch the not-a-number case, though it could be written as (x &lt; 0 || is_nan(x)). The positive infinite case is allowed through, but could also be checked with the is_inf(x) function. The square root function does not itself reject, but some downstream consumer of dbl_sqrt(-2) would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.\nThe effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.2\nIn generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.\n\n\n\n\nFunction argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the reference manual for full details.\nFor example, here’s a function to compute the entropy of a categorical distribution with simplex parameter theta.\nreal entropy(vector theta) {\n  return sum(theta .* log(theta));\n}\nAlthough theta must be a simplex, only the type vector is used.3\nUpper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.\n\n\n\nArray arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.\narray[] real baz(array[,] real x);\nThe notation [ ] is used for one-dimensional arrays (as in the return above), [ , ] for two-dimensional arrays, [ , , ] for three-dimensional arrays, and so on.\nFunctions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.\n\n\n\nA function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier data. The following is an example of a data-only function argument.\nreal foo(real y, data real mu) {\n  return -0.5 * (y - mu)^2;\n}\nThis qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.\nUse of the data qualifier must be consistent between the forward declaration and the definition of a functions.\nThis qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the ODE solver chapter for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the data prefix qualifier. This allows for compile-time type checking and increases overall program understandability.\n\n\n\n\nIn some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.\nfunctions {\n  void pretty_print_tri_lower(matrix x) {\n    if (rows(x) == 0) {\n      print(\"empty matrix\");\n      return;\n    }\n    print(\"rows=\", rows(x), \" cols=\", cols(x));\n    for (m in 1:rows(x)) {\n      for (n in 1:m) {\n        print(\"[\", m, \",\", n, \"]=\", x[m, n]);\n      }\n    }\n  }\n}\nThe special symbol void is used as the return type. This is not a type itself in that there are no values of type void; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.\nVoid functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ... code to set Sigma ...\n  pretty_print_tri_lower(Sigma);\n  // ...\n}\n\n\n\nFunctions whose names end in _lp are allowed to use sampling statements and target += statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.\nHere is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the reparameterization section for more information on efficient non-centered parameterizations\nfunctions {\n  vector center_lp(vector beta_raw, real mu, real sigma) {\n    beta_raw ~ std_normal();\n    sigma ~ cauchy(0, 5);\n    mu ~ cauchy(0, 2.5);\n    return sigma * beta_raw + mu;\n  }\n  // ...\n}\nparameters {\n  vector[K] beta_raw;\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // ...\n  beta = center_lp(beta_raw, mu_beta, sigma_beta);\n  // ...\n}\n\n\n\nFunctions whose names end in _jacobian can use the jacobian += statement. This can be used to implement a custom change of variables for arbitrary parameters.\nFor example, this function recreates the built-in &lt;upper=x&gt; transform on real numbers:\nreal my_upper_bound_jacobian(real x, real ub) {\n  jacobian += x;\n  return ub - exp(x);\n}\nIt can be used as a replacement for real&lt;lower=ub&gt; as follows:\nfunctions {\n  // my_upper_bound_jacobian as above\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  b ~ lognormal(0, 1);\n  // ...\n}\n\n\n\nA user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in _rng. Giving a function a name that ends in _rng allows it to access built-in functions and user-defined functions that end in _rng, which includes all the built-in PRNG functions. Only functions ending in _rng are able access the built-in PRNG functions. The use of functions ending in _rng must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in _rng.\nFor example, the following function generates an \\(N \\times K\\) data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.\nmatrix predictors_rng(int N, int K) {\n  matrix[N, K] x;\n  for (n in 1:N) {\n    x[n, 1] = 1.0;  // intercept\n    for (k in 2:K) {\n      x[n, k] = normal_rng(0, 1);\n    }\n  }\n  return x;\n}\nThe following function defines a simulator for regression outcomes based on a data matrix x, coefficients beta, and noise scale sigma.\nvector regression_rng(vector beta, matrix x, real sigma) {\n  vector[rows(x)] y;\n  vector[rows(x)] mu;\n  mu = x * beta;\n  for (n in 1:rows(x)) {\n    y[n] = normal_rng(mu[n], sigma);\n  }\n  return y;\n}\nThese might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n  // ...\n}\ngenerated quantities {\n  matrix[N_sim, K] x_sim;\n  vector[N_sim] y_sim;\n  x_sim = predictors_rng(N_sim, K);\n  y_sim = regression_rng(beta, x_sim, sigma);\n}\nA more sophisticated simulation might fit a multivariate normal to the predictors x and use the resulting parameters to generate multivariate normal draws for x_sim.\n\n\n\nProbability functions are distinguished in Stan by names ending in _lpdf for density functions and _lpmf for mass functions; in both cases, they must have real return types.\nSuppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.\nfunctions {\n  real unit_normal_lpdf(real y) {\n    return normal_lpdf(y | 0, 1);\n  }\n}\n// ...\nmodel {\n  alpha ~ unit_normal();\n  beta ~ unit_normal();\n  // ...\n}\nThe ability to use the unit_normal function as a density is keyed off its name ending in _lpdf (names ending in _lpmf for probability mass functions work the same way).\nIn general, if foo_lpdf is defined to consume \\(N + 1\\) arguments, then\ny ~ foo(theta1, ..., thetaN);\ncan be used as shorthand for\ntarget += foo_lpdf(y | theta1, ..., thetaN);\nAs with the built-in functions, the suffix _lpdf is dropped and the first argument moves to the left of the tilde symbol (~) in the distribution statement.\nFunctions ending in _lpmf (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (_lpdf) must be continuous (not an integer or integer array), whereas the first argument of a mass function (_lpmf) must be discrete (integer or integer array).\n\n\n\nAs described in the reference manual function overloading is permitted in Stan, beginning in version 2.29.\nThis means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.\n\n\nOverloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.\nFor example, consider a function triple which has the following three signatures\nreal triple(real x);\ncomplex triple(complex x);\narray[] real triple(array[] real);\nOne should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect\n// The function does what it says\ntriple(3.0) == 9.0\n// It is defined reasonably for different types\ntriple(to_complex(3.0)) == to_complex(triple(3.0))\n// A container version of this function works by element\ntriple({3.0, 4.0})[0] == triple({3.0, 4.0}[0])\nNote that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.\n\n\n\nStan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.\nConsider the following function signatures\nreal foo(int a, real b);\nreal foo(real a, real b);\nGiven these, the function call foo(1.5, 2.5) is unambiguous - it must resolve to the second signature. But, the function call foo(1, 1.5) could be valid for either under Stan’s promotion rules, which allow integers to be promoted to real numbers.\nTo resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call foo(1, 1.5) would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).\nFurthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures\nreal bar(int x, real y);\nreal bar(real x, int y);\nThese signatures do not have a unique minimum number of promotions for the call bar(1, 2). Both signatures require one int to real promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.\nPromotion from integers to complex numbers is considered to be two separate promotions, first from int to real, then from real to complex. This means that integer arguments will “prefer” a signature with real types over complex types.\nFor example, consider the function signatures\nreal pop(real x);\nreal pop(complex x);\nStan will select the first signature when pop is called with an integer argument such as pop(0).\n\n\n\n\nFunctions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.\nFor example, here’s some documentation for the prediction matrix generator.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows corresponding to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\nThe comment begins with /**, ends with */, and has an asterisk (*) on each line. It uses @param followed by the argument’s identifier to document a function argument. The tag @return is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with /* and ending with */ to the Stan parser.\nFor functions that raise exceptions, exceptions can be documented using @throws.4\nFor example,\n /** ...\n * @param theta\n * @throws If any of the entries of theta is negative.\n */\nreal entropy(vector theta) {\n  // ...\n}\nUsually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.\n\n\n\nFunctions may have a void or non-void return type and they may or may not have one of the special suffixes, _lpdf, _lpmf, _lp, or _rng.\n\n\nOnly functions declared to return void may be used as statements. These are also the only functions that use return statements with no arguments.\nOnly functions declared to return non-void values may be used as expressions. These functions require return statements with arguments of a type that matches the declared return type.\n\n\n\nOnly functions ending in _lpmf or _lpdf and with return type real may be used as probability functions in distribution statements.\nOnly functions ending in _lp may access the log probability accumulator through distribution statements or target += statements. Such functions may only be used in the transformed parameters or model blocks.\nOnly functions ending in _rng may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in _rng.\n\n\n\n\nStan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, \\(A^n\\), which is defined for a square matrix \\(A\\) and positive integer \\(n\\) by \\[\nA^n\n=\n\\begin{cases}\n\\textrm{I} & \\quad\\text{if } n = 0, \\text{ and} \\\\\nA \\, A^{n-1} & \\quad\\text{if } n &gt; 0.\n\\end{cases}\n\\]\nwhere \\(\\textrm{I}\\) is the identity matrix. This definition can be directly translated to a recursive function definition.\nmatrix matrix_pow(matrix a, int n) {\n  if (n == 0) {\n    return diag_matrix(rep_vector(1, rows(a)));\n  } else {\n    return a *  matrix_pow(a, n - 1);\n  }\n}\nIt would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.\nelse if (n == 1) {\n  return a;\n}\n\n\n\n\n\nTo generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then run through the inverse cumulative distribution function, \\(y = \\textrm{logit}(u)\\). If this were not already built in as logistic_rng(0, 1), it could be coded in Stan directly as\nreal standard_logistic_rng() {\n  real u = uniform_rng(0, 1);\n  real y = logit(u);\n  return y;\n}\nFollowing the same pattern, a standard normal RNG could be coded as\nreal standard_normal_rng() {\n  real u = uniform_rng(0, 1);\n  real y = inv_Phi(u);\n  return y;\n}\nthat is, \\(y = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}\\) is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function inv_Phi.\nIn order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate \\(\\textsf{normal}(\\mu, \\sigma)\\) variates, it is enough to generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then convert it to a standard normal variate, \\(z = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}(\\cdot)\\) is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, \\(y = \\mu +\n\\sigma \\times z\\). In code,\nreal my_normal_rng(real mu, real sigma) {\n  real u = uniform_rng(0, 1);\n  real z = inv_Phi(u);\n  real y = mu + sigma * z;\n  return y;\n}\nA robust version of this function would test that the arguments are finite and that sigma is non-negative, e.g.,\n  if (is_nan(mu) || is_inf(mu)) {\n    reject(\"my_normal_rng: mu must be finite; \",\n           \"found mu = \", mu);\n  }\n  if (is_nan(sigma) || is_inf(sigma) || sigma &lt; 0) {\n    reject(\"my_normal_rng: sigma must be finite and non-negative; \",\n           \"found sigma = \", sigma);\n  }\n\n\n\nOften truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.\n\n\nFor example, the following code generates a \\(\\textsf{Weibull}(\\alpha, \\sigma)\\) variate truncated below at a time \\(t\\),5\nreal weibull_lb_rng(real alpha, real sigma, real t) {\n  real p = weibull_cdf(t | alpha, sigma);   // cdf for lb\n  real u = uniform_rng(p, 1);               // unif in bounds\n  real y = sigma * (-log1m(u))^inv(alpha);  // inverse cdf\n  return y;\n}\n\n\n\nIf there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a \\(\\textsf{normal}(\\mu, \\sigma)\\) truncated to a region \\((a, b)\\), the following code suffices,\nreal normal_lub_rng(real mu, real sigma, real lb, real ub) {\n  real p_lb = normal_cdf(lb | mu, sigma);\n  real p_ub = normal_cdf(ub | mu, sigma);\n  real u = uniform_rng(p_lb, p_ub);\n  real y = mu + sigma * inv_Phi(u);\n  return y;\n}\nTo make this more robust, all variables should be tested for finiteness, sigma should be tested for positiveness, and lb and ub should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement\n  return mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),\n                                          normal_cdf(ub | mu, sigma)));\nfor readability. The names like p indicate probabilities, and p_lb and p_ub indicate the probabilities of the bounds. The variable u is clearly named as a uniform variate, and y is used to denote the variate being generated itself.",
+    "text": "The exponential model is parametric in that is specifies an explicit parametric form for the distribution of survival times. Cox (1972) introduced a semi-parametric survival model specified directly in terms of a hazard function \\(h(t)\\) rather than in terms of a distribution over survival times. Cox’s model is semi-parametric in that it does not model the full hazard function, instead modeling only the proportional differences in hazards among subjects.\nLet \\(x_n \\in \\mathbb{R}^K\\) be a (row) vector of covariates for subject \\(n\\) so that the full covariate data matrix is \\(x \\in \\mathbb{R}^{N \\times\nK}\\). In Cox’s model, the hazard function for subject \\(n\\) is defined conditionally in terms of their covariates \\(x_n\\) and the parameter vector \\(\\gamma \\in \\mathbb{R}^K\\) as \\[\\begin{equation*}\nh(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n \\cdot \\gamma),\n\\end{equation*}\\] where \\(h_0(t)\\) is a shared baseline hazard function and \\(x_n \\cdot\n\\gamma = \\sum_{k=1}^K x_{n, k} \\cdot \\beta_k\\) is a row vector-vector product.\nIn the semi-parametric, proportional hazards model, the baseline hazard function \\(h_0(t)\\) is not modeled. This is why it is called “semi-parametric.” Only the factor \\(\\exp(x_n \\cdot \\gamma),\\) which determines how individual \\(n\\) varies by a proportion from the baseline hazard, is modeled. This is why it’s called “proportional hazards.”\nCox’s proportional hazards model is not fully generative. There is no way to generate the times of failure because the baseline hazard function \\(h_0(t)\\) is unmodeled; if the baseline hazard were known, failure times could be generated. Cox’s proportional hazards model is generative for the ordering of failures conditional on a number of censored items. Proportional hazard models may also include parametric or non-parametric model for the baseline hazard function1.\n\n\nCox’s proportional specification of the hazard function is insufficient to generate random variates because the baseline hazard function \\(h_0(t)\\) is unknown. On the other hand, the proportional specification is sufficient to generate a partial likelihood that accounts for the order of the survival times.\nThe hazard function \\(h(t \\mid x_n, \\beta) = h_0(t) \\cdot \\exp(x_n\n\\cdot \\beta)\\) for subject \\(n\\) represents the instantaneous probability that subject \\(n\\) fails at time \\(t\\) given that it has survived until time \\(t.\\) The probability that subject \\(n\\) is the first to fail among \\(N\\) subjects is thus proportional to subject \\(n\\)’s hazard function, \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail at time } t]\n\\propto h(t \\mid x_n, \\beta).\n\\end{equation*}\\] Normalizing yields \\[\\begin{eqnarray*}\n\\Pr[n \\textrm{ first to fail at time } t]\n& = &  \\frac{h(t \\mid x_n, \\beta)}\n            {\\sum_{n' = 1}^N h(t \\mid x_{n'}, \\beta)}\n\\\\[4pt]\n& = &  \\frac{h_0(t) \\cdot \\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N h_0(t) \\cdot \\exp(x_{n'} \\cdot \\beta)}\n\\\\[4pt]\n& = &  \\frac{\\exp(x_n \\cdot \\beta)}\n            {\\sum_{n' = 1}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{eqnarray*}\\]\nSuppose there are \\(N\\) subjects with strictly ordered survival times \\(t_1 &lt;\nt_2 &lt; \\cdots &lt; t_N\\) and covariate (row) vectors \\(x_1, \\ldots, x_N\\). Let \\(t^{\\textrm{cens}}\\) be the (right) censoring time and let \\(N^{\\textrm{obs}}\\) be the largest value of \\(n\\) such that \\(t_n \\leq\nt^{\\textrm{cens}}\\). This means \\(N^{\\textrm{obs}}\\) is the number of subjects whose failure time was observed. The ordering is for convenient indexing and does not cause any loss of generality—survival times can simply be sorted into the necessary order.\nWith failure times sorted in decreasing order, the partial likelihood for each observed subject \\(n \\in 1{:}N^{\\textrm{obs}}\\) can be expressed as \\[\\begin{equation*}\n\\Pr[n \\textrm{ first to fail among } n, n + 1, \\ldots N]\n= \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] The group of items for comparison and hence the summation is over all items, including those with observed and censored failure times.\nThe partial likelihood, defined in this form by Breslow (1975), is just the product of the partial likelihoods for the observed subjects (i.e., excluding subjects whose failure time is censored). \\[\\begin{equation*}\n\\Pr[\\textrm{observed failures ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n= \\prod_{n = 1}^{N^{\\textrm{obs}}}\n  \\frac{\\exp(x_n \\cdot \\beta)}\n       {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}.\n\\end{equation*}\\] On the log scale, \\[\\begin{eqnarray*}\n\\log \\Pr[\\textrm{obs.\\ fail ordered } 1, \\ldots, N^{\\textrm{obs}} |\nx, \\beta]\n& = &\n\\sum_{n = 1}^{N^{\\textrm{obs}}}\n  \\log \\left(\n          \\frac{\\exp(x_n \\cdot \\beta)}\n               {\\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)}\n       \\right)\n\\\\[4pt]\n& = & x_n \\cdot \\beta - \\log \\sum_{n' = n}^N \\exp(x_{n'} \\cdot \\beta)\n\\\\\n& = & x_n \\cdot \\beta - \\textrm{logSumExp}_{n' = n}^N \\ x_{n'} \\cdot \\beta,\n\\end{eqnarray*}\\] where \\[\\begin{equation*}\n\\textrm{logSumExp}_{n = a}^b \\ x_n\n= \\log \\sum_{n = a}^b \\exp(x_n)\n\\end{equation*}\\] is implemented so as to preserve numerical precision.\nThis likelihood follows the same approach to ranking as that developed by Plackett (1975) for estimating the probability of the order of the first few finishers in a horse race.\nA simple normal prior on the components of \\(\\beta\\) completes the model, \\[\\begin{equation*}\n\\beta \\sim \\textrm{normal}(0, 2).\n\\end{equation*}\\] This should be scaled based on knowledge of the predictors.\n\n\n\nTo simplify the Stan program, the survival times for uncensored events are sorted into decreasing order (unlike in the mathematical presentation, where they were sorted into ascending order). The covariates for censored and uncensored observations are separated into two matrices.\ndata {\n  int&lt;lower=0&gt; K;          // num covariates\n\n  int&lt;lower=0&gt; N;          // num uncensored obs\n  vector[N] t;             // event time (non-strict decreasing)\n  matrix[N, K] x;          // covariates for uncensored obs\n\n  int N_c;                 // num censored obs\n  real&lt;lower=t[N]&gt; t_c;    // censoring time\n  matrix[N_c, K] x_c;      // covariates for censored obs\n}\nThe parameters are just the coefficients.\nparameters {\n  vector[K] beta;          // slopes (no intercept)\n}\nThe prior is a simple independent centered normal distribution on each element of the parameter vector, which is vectorized in the Stan code.\nmodel {\n  beta ~ normal(0, 2);\n  ...\nThe log likelihood is implemented so as to minimize duplicated effort. The first order of business is to calculate the linear predictors, which is done separately for the subjects whose event time is observed and those for which the event time is censored.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\nThese vectors are computed using efficient matrix-vector multiplies. The log of exponential values of the censored covariates times the coefficients is reused in the denominator of each factor, which on the log scale, starts with the log sum of exponentials of the censored items’ linear predictors.\n  real log_denom = log_sum_exp(log_theta_c);\nThen, for each observed survival time, going backwards from the latest to the earliest, the denominator can be incremented (which turns into a log sum of exponentials on the log scale), and then the target is updated with its likelihood contribution.\n  for (n in 1:N) {\n    log_denom = log_sum_exp(log_denom, log_theta[n]);\n    target += log_theta[n] - log_denom;   // log likelihood\n  }\nThe running log sum of exponentials is why the list is iterated in reverse order of survival times. It allows the log denominator to be accumulated one term at a time. The condition that the survival times are sorted into decreasing order is not checked. It could be checked very easily in the transformed data block by adding the following code.\ntransformed data {\n  for (n in 2:N) {\n    if (!(t[n] &lt; t[n - 1])) {\n      reject(\"times must be strictly decreasing, but found\"\n             \"!(t[\", n, \"] &lt; t[, \", (n - 1), \"])\");\n    }   \n  }\n}\n\n\n\nTechnically, for continuous survival times, the probability of two survival times being identical will be zero. Nevertheless, real data sets often round survival times, for instance to the nearest day or week in a multi-year clinical trial. The technically “correct” thing to do in the face of unknown survival times in a range would be to treat their order as unknown and infer it. But considering all \\(N!\\) permutations for a set of \\(N\\) subjects with tied survival times is not tractable. As an alternative, Efron (1977) introduced an approximate partial likelihood with better properties than a random permutation while not being quite as good as considering all permutations. Efron’s model averages the contributions as if they truly did occur simultaneously.\nIn the interest of completeness, here is the Stan code for an implementation of Efron’s estimator. It uses two user-defined functions. The first calculates how many different survival times occur in the data.\nfunctions {\n  int num_unique_starts(vector t) {\n    if (size(t) == 0) return 0;\n    int us = 1;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) us += 1;\n    }\n    return us;\n  }\nThis is then used to compute the value J to send into the function that computes the position in the array of failure times where each new failure time starts, plus an end point that goes one past the target. This is a standard way in Stan to code ragged arrays.\n  array[] int unique_starts(vector t, int J) {\n    array[J + 1] int starts;\n    if (J == 0) return starts;\n    starts[1] = 1;\n    int pos = 2;\n    for (n in 2:size(t)) {\n      if (t[n] != t[n - 1]) {\n    starts[pos] = n;\n    pos += 1;\n      }\n    }\n    starts[J + 1] = size(t) + 1;\n    return starts;\n  }\n}\nThe data format is exactly the same as for the model in the previous section, but in this case, the transformed data block is used to cache some precomputations required for the model, namely the ragged array grouping elements that share the same survival time.\ntransformed data {\n  int&lt;lower=0&gt; J = num_unique_starts(t);\n  array[J + 1] int&lt;lower=0&gt; starts = unique_starts(t, J);\n}\nFor each unique survival time j in 1:J, the subjects indexed from starts[j] to starts[j + 1] - 1 (inclusive) share the same survival time. The number of elements with survival time j is thus (starts[j + 1] - 1) - starts[j] + 1, or just starts[j + 1] - starts[j].\nThe parameters and prior are also the same—just a vector beta of coefficients with a centered normal prior. Although it starts with the same caching of results for later, and uses the same accumulator for the denominator, the overall partial likelihood is much more involved, and depends on the user-defined functions defining the transformed data variables J and starts.\n  vector[N] log_theta = x * beta;\n  vector[N_c] log_theta_c = x_c * beta;\n  real log_denom_lhs = log_sum_exp(log_theta_c);\n  for (j in 1:J) {\n    int start = starts[j];\n    int end = starts[j + 1] - 1;\n    int len = end - start + 1;\n    real log_len = log(len);\n    real numerator = sum(log_theta[start:end]);\n    log_denom_lhs = log_sum_exp(log_denom_lhs,\n                                log_sum_exp(log_theta[start:end]));\n    vector[len] diff;\n    for (ell in 1:len) {\n      diff[ell] = log_diff_exp(log_denom_lhs,\n                               log(ell - 1) - log_len\n                               + log_sum_exp(log_theta[start:end]));\n    }\n    target += numerator - sum(diff);\n  }\nThe special function log_diff_exp is defined as\n\\[\\begin{equation*}\n\\textrm{logDiffExp}(u, v) = \\log(\\exp(u) - \\exp(v)).\n\\end{equation*}\\]\nBecause of how J and starts are constructed, the length len will always be strictly positive so that the log is well defined.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Example Models",
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#basic-functions.section",
-    "href": "stan-users-guide/user-functions.html#basic-functions.section",
-    "title": "User-Defined Functions",
-    "section": "",
-    "text": "Here’s an example of a skeletal Stan program with a user-defined relative difference function employed in the generated quantities block to compute a relative differences between two parameters.\nfunctions {\n  real relative_diff(real x, real y) {\n    real abs_diff;\n    real avg_scale;\n    abs_diff = abs(x - y);\n    avg_scale = (abs(x) + abs(y)) / 2;\n    return abs_diff / avg_scale;\n  }\n}\n// ...\ngenerated quantities {\n  real rdiff;\n  rdiff = relative_diff(alpha, beta);\n}\nThe function is named relative_diff, and is declared to have two real-valued arguments and return a real-valued result. It is used the same way a built-in function would be used in the generated quantities block.\n\n\nAll functions are defined in their own block, which is labeled functions and must appear before all other program blocks. The user-defined functions block is optional.\n\n\n\nThe body (the part between the curly braces) contains ordinary Stan code, including local variables. The new function is used in the generated quantities block just as any of Stan’s built-in functions would be used.\n\n\n\nReturn statements, such as the one on the last line of the definition of relative_diff above, are only allowed in the bodies of function definitions. Return statements may appear anywhere in a function, but functions with non-void return types must end in a return statement.\n\n\n\nThe Stan reject statement provides a mechanism to report errors or problematic values encountered during program execution. It accepts any number of quoted string literals or Stan expressions as arguments. This statement is typically embedded in a conditional statement in order to detect bad or illegal outcomes of some processing step.\nIf an error is indicative of a problem from which it is not expected to be able to recover, Stan provides a fatal_error statement.\n\n\nRejection is used to flag errors that arise in inputs or in program state. It is far better to fail early with a localized informative error message than to run into problems much further downstream (as in rejecting a state or failing to compute a derivative).\nThe most common errors that are coded is to test that all of the arguments to a function are legal. The following function takes a square root of its input, so requires non-negative inputs; it is coded to guard against illegal inputs.\nreal dbl_sqrt(real x) {\n  if (!(x &gt;= 0)) {\n    reject(\"dbl_sqrt(x): x must be positive; found x = \", x);\n  }\n  return 2 * sqrt(x);\n}\nThe negation of the positive test is important, because it also catches the case where x is a not-a-number value. If the condition had been coded as (x &lt; 0) it would not catch the not-a-number case, though it could be written as (x &lt; 0 || is_nan(x)). The positive infinite case is allowed through, but could also be checked with the is_inf(x) function. The square root function does not itself reject, but some downstream consumer of dbl_sqrt(-2) would be likely to raise an error, at which point the origin of the illegal input requires detective work. Or even worse, as Matt Simpson pointed out in the GitHub comments, the function could go into an infinite loop if it starts with an infinite value and tries to reduce it by arithmetic, likely consuming all available memory and crashing an interface. Much better to catch errors early and report on their origin.\nThe effect of rejection depends on the program block in which the rejection is executed. In transformed data, rejections cause the program to fail to load. In transformed parameters or in the model block, rejections cause the current state to be rejected in the Metropolis sense.2\nIn generated quantities there is no way to recover and generate the remaining parameters, so rejections cause subsequent values to be reported as NaNs. Extra care should be taken in calling functions which may reject in the generated quantities block.\n\n\n\n\nFunction argument and return types for vector and matrix types are not declared with their sizes, unlike type declarations for variables. Function argument type declarations may not be declared with constraints, either lower or upper bounds or structured constraints like forming a simplex or correlation matrix, (as is also the case for local variables); see the table of types in the reference manual for full details.\nFor example, here’s a function to compute the entropy of a categorical distribution with simplex parameter theta.\nreal entropy(vector theta) {\n  return sum(theta .* log(theta));\n}\nAlthough theta must be a simplex, only the type vector is used.3\nUpper or lower bounds on values or constrained types are not allowed as return types or argument types in function declarations.\n\n\n\nArray arguments have their own syntax, which follows that used in this manual for function signatures. For example, a function that operates on a two-dimensional array to produce a one-dimensional array might be declared as follows.\narray[] real baz(array[,] real x);\nThe notation [ ] is used for one-dimensional arrays (as in the return above), [ , ] for two-dimensional arrays, [ , , ] for three-dimensional arrays, and so on.\nFunctions support arrays of any type, including matrix and vector types. As with other types, no constraints are allowed.\n\n\n\nA function argument which is a real-valued type or a container of a real-valued type, i.e., not an integer type or integer array type, can be qualified using the prefix qualifier data. The following is an example of a data-only function argument.\nreal foo(real y, data real mu) {\n  return -0.5 * (y - mu)^2;\n}\nThis qualifier restricts this argument to being invoked with expressions which consist only of data variables, transformed data variables, literals, and function calls. A data-only function argument cannot involve real variables declared in the parameters, transformed parameters, or model block. Attempts to invoke a function using an expression which contains parameter, transformed parameters, or model block variables as a data-only argument will result in an error message from the parser.\nUse of the data qualifier must be consistent between the forward declaration and the definition of a functions.\nThis qualifier should be used when writing functions that call the built-in ordinary differential equation (ODE) solvers, algebraic solvers, or map functions. These higher-order functions have strictly specified signatures where some arguments of are data only expressions. (See the ODE solver chapter for more usage details and the functions reference manual for full definitions.) When writing a function which calls the ODE or algebraic solver, arguments to that function which are passed into the call to the solver, either directly or indirectly, should have the data prefix qualifier. This allows for compile-time type checking and increases overall program understandability.",
+    "objectID": "stan-users-guide/survival.html#footnotes",
+    "href": "stan-users-guide/survival.html#footnotes",
+    "title": "Survival Models",
+    "section": "Footnotes",
+    "text": "Footnotes\n\n\nCox mentioned in his seminal paper that modeling the baseline hazard function would improve statistical efficiency, but he did not do it for computational reasons.↩︎",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Example Models",
+      "Survival Models"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#functions-as-statements",
-    "href": "stan-users-guide/user-functions.html#functions-as-statements",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/truncation-censoring.html",
+    "href": "stan-users-guide/truncation-censoring.html",
+    "title": "Truncated or Censored Data",
     "section": "",
-    "text": "In some cases, it makes sense to have functions that do not return a value. For example, a routine to print the lower-triangular portion of a matrix can be defined as follows.\nfunctions {\n  void pretty_print_tri_lower(matrix x) {\n    if (rows(x) == 0) {\n      print(\"empty matrix\");\n      return;\n    }\n    print(\"rows=\", rows(x), \" cols=\", cols(x));\n    for (m in 1:rows(x)) {\n      for (n in 1:m) {\n        print(\"[\", m, \",\", n, \"]=\", x[m, n]);\n      }\n    }\n  }\n}\nThe special symbol void is used as the return type. This is not a type itself in that there are no values of type void; it merely indicates the lack of a value. As such, return statements for void functions are not allowed to have arguments, as in the return statement in the body of the previous example.\nVoid functions applied to appropriately typed arguments may be used on their own as statements. For example, the pretty-print function defined above may be applied to a covariance matrix being defined in the transformed parameters block.\ntransformed parameters {\n  cov_matrix[K] Sigma;\n  // ... code to set Sigma ...\n  pretty_print_tri_lower(Sigma);\n  // ...\n}",
+    "text": "Data in which measurements have been truncated or censored can be coded in Stan following their respective probability models.\n\n\nTruncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.\n\n\n\nTruncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.\nTruncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are \\(y_n\\) with an upper truncation point of \\(U = 300\\) so that \\(y_n &lt; 300\\). In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  real U;\n  array[N] real&lt;upper=U&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma) T[ , U];\n}\nThe model declares an upper bound U as data and constrains the data for y to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.\nSee the Stan Reference Manual’s Statements chapter for how to use truncated distributions with the log probability increment statements (target += ...).\nThis model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.\n\n\nIf the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to \\(-\\infty\\). For instance, if variate y is sampled with the statement.\ny ~ normal(mu, sigma) T[L, U];\nthen if any value inside y is less than the value of L or greater than the value of U, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.\nTo avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if y is a parameter in the above model, the declaration should constrain it to fall between the values of L and U.\nparameters {\n  array[N] real&lt;lower=L, upper=U&gt; y;\n  // ...\n}\nIf in the above model, L or U is a parameter and y is data, then L and U must be appropriately constrained so that all data are in range and the value of L is less than that of U (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.\nparameters {\n  real&lt;upper=min(y)&gt; L;           // L &lt; y[n]\n  real&lt;lower=fmax(L, max(y))&gt; U;  // L &lt; U; y[n] &lt; U\nFor pairs of real numbers, the function fmax is used rather than max.\n\n\n\nIf the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real y;\n}\nparameters {\n  real&lt;upper=min(y)&gt; L;\n  real&lt;lower=max(y)&gt; U;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ // ...\n  U ~ // ...\n  y ~ normal(mu, sigma) T[L, U];\n}\nHere there is a lower truncation point L which is declared to be less than or equal to the minimum value of y. The upper truncation point U is declared to be larger than the maximum value of y. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With N declared as type int&lt;lower=1&gt;, there must be at least one data point. The constraint that L is less than U is enforced indirectly, based on the non-empty data.\nThe ellipses where the priors for the bounds L and U should go should be filled in with a an informative prior in order for this model to not concentrate L strongly around min(y) and U strongly around max(y).\n\n\n\n\nCensoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.\n\n\nOne way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  array[N_cens] real&lt;lower=U&gt; y_cens;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_cens ~ normal(mu, sigma);\n}\nBecause the censored data array y_cens is declared to be a parameter, it will be sampled along with the location and scale parameters mu and sigma. Because the censored data array y_cens is declared to have values of type real&lt;lower=U&gt;, all imputed values for censored data will be greater than U. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.\n\n\n\nAlthough it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of \\[\\begin{align*}\n\\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\int_U^{\\infty} \\textsf{normal}\\left(y_{\\mathrm{cens},m} \\mid \\mu,\\sigma \\right) \\,\\textsf{d}y_{\\mathrm{cens},m} \\\\\n  &= 1 - \\Phi\\left(\\frac{U - \\mu}{\\sigma}\\right),\n\\end{align*}\\]\nwhere \\(\\Phi()\\) is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that \\(y_{\\mathrm{cens},m}&gt;U\\). With \\(M\\) censored observations, the likelihood on the log scale is \\[\\begin{align*}\n\\log \\prod_{m=1}^M \\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\log \\left( 1 - \\Phi\\left(\\left(\\frac{U - \\mu}{\\sigma}\\right)\\right)^{M}\\right) \\\\\n  &= M \\times \\texttt{normal}\\mathtt{\\_}\\texttt{lccdf}\\left(U \\mid \\mu, \\sigma \\right),\n\\end{align*}\\]\nwhere normal_lccdf is the log of complementary CDF (Stan provides &lt;distr&gt;_lccdf for each distribution implemented in Stan).\nThe following right-censored model assumes that the censoring point is known, so it is declared as data.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lccdf(U | mu, sigma);\n}\nFor the observed values in y_obs, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.\nFor the left-censored data the CDF (normal_lcdf) has to be used instead of complementary CDF. If the censoring point variable (L) is unknown, its declaration should be moved from the data to the parameters block.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real&lt;upper=min(y_obs)&gt; L;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ normal(mu, sigma);\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lcdf(L | mu, sigma);\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Example Models",
+      "Truncated or Censored Data"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#functions-accessing-the-log-probability-accumulator",
-    "href": "stan-users-guide/user-functions.html#functions-accessing-the-log-probability-accumulator",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/truncation-censoring.html#truncation.section",
+    "href": "stan-users-guide/truncation-censoring.html#truncation.section",
+    "title": "Truncated or Censored Data",
     "section": "",
-    "text": "Functions whose names end in _lp are allowed to use sampling statements and target += statements; other functions are not. Because of this access, their use is restricted to the transformed parameters and model blocks.\nHere is an example of a function to assign standard normal priors to a vector of coefficients, along with a center and scale, and return the translated and scaled coefficients; see the reparameterization section for more information on efficient non-centered parameterizations\nfunctions {\n  vector center_lp(vector beta_raw, real mu, real sigma) {\n    beta_raw ~ std_normal();\n    sigma ~ cauchy(0, 5);\n    mu ~ cauchy(0, 2.5);\n    return sigma * beta_raw + mu;\n  }\n  // ...\n}\nparameters {\n  vector[K] beta_raw;\n  real mu_beta;\n  real&lt;lower=0&gt; sigma_beta;\n  // ...\n}\ntransformed parameters {\n  vector[K] beta;\n  // ...\n  beta = center_lp(beta_raw, mu_beta, sigma_beta);\n  // ...\n}",
+    "text": "Truncation in Stan is restricted to univariate distributions for which the corresponding log cumulative distribution function (CDF) and log complementary cumulative distribution (CCDF) functions are available. See the reference manual section on truncated distributions for more information on truncated distributions, CDFs, and CCDFs.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Example Models",
+      "Truncated or Censored Data"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#functions-implementing-change-of-variable-adjustments",
-    "href": "stan-users-guide/user-functions.html#functions-implementing-change-of-variable-adjustments",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/truncation-censoring.html#truncated-data.section",
+    "href": "stan-users-guide/truncation-censoring.html#truncated-data.section",
+    "title": "Truncated or Censored Data",
     "section": "",
-    "text": "Functions whose names end in _jacobian can use the jacobian += statement. This can be used to implement a custom change of variables for arbitrary parameters.\nFor example, this function recreates the built-in &lt;upper=x&gt; transform on real numbers:\nreal my_upper_bound_jacobian(real x, real ub) {\n  jacobian += x;\n  return ub - exp(x);\n}\nIt can be used as a replacement for real&lt;lower=ub&gt; as follows:\nfunctions {\n  // my_upper_bound_jacobian as above\n}\ndata {\n  real ub;\n}\nparameters {\n  real b_raw;\n}\ntransformed parameters {\n  real b = my_upper_bound_jacobian(b_raw, ub);\n}\nmodel {\n  b ~ lognormal(0, 1);\n  // ...\n}",
+    "text": "Truncated data are data for which measurements are only reported if they fall above a lower bound, below an upper bound, or between a lower and upper bound.\nTruncated data may be modeled in Stan using truncated distributions. For example, suppose the truncated data are \\(y_n\\) with an upper truncation point of \\(U = 300\\) so that \\(y_n &lt; 300\\). In Stan, this data can be modeled as following a truncated normal distribution for the observations as follows.\ndata {\n  int&lt;lower=0&gt; N;\n  real U;\n  array[N] real&lt;upper=U&gt; y;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y ~ normal(mu, sigma) T[ , U];\n}\nThe model declares an upper bound U as data and constrains the data for y to respect the constraint; this will be checked when the data are loaded into the model before sampling begins.\nSee the Stan Reference Manual’s Statements chapter for how to use truncated distributions with the log probability increment statements (target += ...).\nThis model implicitly uses an improper flat prior on the scale and location parameters; these could be given priors in the model using distribution statements.\n\n\nIf the sampled variate in a truncated distribution lies outside of the truncation range, the probability is zero, so the log probability will evaluate to \\(-\\infty\\). For instance, if variate y is sampled with the statement.\ny ~ normal(mu, sigma) T[L, U];\nthen if any value inside y is less than the value of L or greater than the value of U, the distribution statement produces a zero-probability estimate. For user-defined truncation, this zeroing outside of truncation bounds must be handled explicitly.\nTo avoid variables straying outside of truncation bounds, appropriate constraints are required. For example, if y is a parameter in the above model, the declaration should constrain it to fall between the values of L and U.\nparameters {\n  array[N] real&lt;lower=L, upper=U&gt; y;\n  // ...\n}\nIf in the above model, L or U is a parameter and y is data, then L and U must be appropriately constrained so that all data are in range and the value of L is less than that of U (if they are equal, the parameter range collapses to a single point and the Hamiltonian dynamics used by the sampler break down). The following declarations ensure the bounds are well behaved.\nparameters {\n  real&lt;upper=min(y)&gt; L;           // L &lt; y[n]\n  real&lt;lower=fmax(L, max(y))&gt; U;  // L &lt; U; y[n] &lt; U\nFor pairs of real numbers, the function fmax is used rather than max.\n\n\n\nIf the truncation points are unknown, they may be estimated as parameters. This can be done with a slight rearrangement of the variable declarations from the model in the previous section with known truncation points.\ndata {\n  int&lt;lower=1&gt; N;\n  array[N] real y;\n}\nparameters {\n  real&lt;upper=min(y)&gt; L;\n  real&lt;lower=max(y)&gt; U;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ // ...\n  U ~ // ...\n  y ~ normal(mu, sigma) T[L, U];\n}\nHere there is a lower truncation point L which is declared to be less than or equal to the minimum value of y. The upper truncation point U is declared to be larger than the maximum value of y. This declaration, although dependent on the data, only enforces the constraint that the data fall within the truncation bounds. With N declared as type int&lt;lower=1&gt;, there must be at least one data point. The constraint that L is less than U is enforced indirectly, based on the non-empty data.\nThe ellipses where the priors for the bounds L and U should go should be filled in with a an informative prior in order for this model to not concentrate L strongly around min(y) and U strongly around max(y).",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Example Models",
+      "Truncated or Censored Data"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#functions-acting-as-random-number-generators",
-    "href": "stan-users-guide/user-functions.html#functions-acting-as-random-number-generators",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/truncation-censoring.html#censored.section",
+    "href": "stan-users-guide/truncation-censoring.html#censored.section",
+    "title": "Truncated or Censored Data",
     "section": "",
-    "text": "A user-specified function can be declared to act as a (pseudo) random number generator (PRNG) by giving it a name that ends in _rng. Giving a function a name that ends in _rng allows it to access built-in functions and user-defined functions that end in _rng, which includes all the built-in PRNG functions. Only functions ending in _rng are able access the built-in PRNG functions. The use of functions ending in _rng must therefore be restricted to transformed data and generated quantities blocks like other PRNG functions; they may also be used in the bodies of other user-defined functions ending in _rng.\nFor example, the following function generates an \\(N \\times K\\) data matrix, the first column of which is filled with 1 values for the intercept and the remaining entries of which have values drawn from a standard normal PRNG.\nmatrix predictors_rng(int N, int K) {\n  matrix[N, K] x;\n  for (n in 1:N) {\n    x[n, 1] = 1.0;  // intercept\n    for (k in 2:K) {\n      x[n, k] = normal_rng(0, 1);\n    }\n  }\n  return x;\n}\nThe following function defines a simulator for regression outcomes based on a data matrix x, coefficients beta, and noise scale sigma.\nvector regression_rng(vector beta, matrix x, real sigma) {\n  vector[rows(x)] y;\n  vector[rows(x)] mu;\n  mu = x * beta;\n  for (n in 1:rows(x)) {\n    y[n] = normal_rng(mu[n], sigma);\n  }\n  return y;\n}\nThese might be used in a generated quantity block to simulate some fake data from a fitted regression model as follows.\nparameters {\n  vector[K] beta;\n  real&lt;lower=0&gt; sigma;\n  // ...\n}\ngenerated quantities {\n  matrix[N_sim, K] x_sim;\n  vector[N_sim] y_sim;\n  x_sim = predictors_rng(N_sim, K);\n  y_sim = regression_rng(beta, x_sim, sigma);\n}\nA more sophisticated simulation might fit a multivariate normal to the predictors x and use the resulting parameters to generate multivariate normal draws for x_sim.",
+    "text": "Censoring hides values from points that are too large, too small, or both. Unlike with truncated data, the number of data points that were censored is known. The textbook example is the household scale which does not report values above 300 pounds.\n\n\nOne way to model censored data is to treat the censored data as missing data that is constrained to fall in the censored range of values. Since Stan does not allow unknown values in its arrays or matrices, the censored values must be represented explicitly, as in the following right-censored case.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  array[N_cens] real&lt;lower=U&gt; y_cens;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  y_cens ~ normal(mu, sigma);\n}\nBecause the censored data array y_cens is declared to be a parameter, it will be sampled along with the location and scale parameters mu and sigma. Because the censored data array y_cens is declared to have values of type real&lt;lower=U&gt;, all imputed values for censored data will be greater than U. The imputed censored data affects the location and scale parameters through the last distribution statement in the model.\n\n\n\nAlthough it is wrong to ignore the censored values in estimating location and scale, it is not necessary to impute values. Instead, the values can be integrated out. Each censored data point has a probability of \\[\\begin{align*}\n\\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\int_U^{\\infty} \\textsf{normal}\\left(y_{\\mathrm{cens},m} \\mid \\mu,\\sigma \\right) \\,\\textsf{d}y_{\\mathrm{cens},m} \\\\\n  &= 1 - \\Phi\\left(\\frac{U - \\mu}{\\sigma}\\right),\n\\end{align*}\\]\nwhere \\(\\Phi()\\) is the standard normal cumulative distribution function. This probability is equivalent to the likelihood contribution of knowing that \\(y_{\\mathrm{cens},m}&gt;U\\). With \\(M\\) censored observations, the likelihood on the log scale is \\[\\begin{align*}\n\\log \\prod_{m=1}^M \\Pr[y_{\\mathrm{cens},m} &gt; U]\n  &= \\log \\left( 1 - \\Phi\\left(\\left(\\frac{U - \\mu}{\\sigma}\\right)\\right)^{M}\\right) \\\\\n  &= M \\times \\texttt{normal}\\mathtt{\\_}\\texttt{lccdf}\\left(U \\mid \\mu, \\sigma \\right),\n\\end{align*}\\]\nwhere normal_lccdf is the log of complementary CDF (Stan provides &lt;distr&gt;_lccdf for each distribution implemented in Stan).\nThe following right-censored model assumes that the censoring point is known, so it is declared as data.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n  real&lt;lower=max(y_obs)&gt; U;\n}\nparameters {\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lccdf(U | mu, sigma);\n}\nFor the observed values in y_obs, the normal model is used without truncation. The likelihood contribution from the integrated out censored values can not be coded with distribution statement, and the log probability is directly incremented using the calculated log cumulative normal probability of the censored observations.\nFor the left-censored data the CDF (normal_lcdf) has to be used instead of complementary CDF. If the censoring point variable (L) is unknown, its declaration should be moved from the data to the parameters block.\ndata {\n  int&lt;lower=0&gt; N_obs;\n  int&lt;lower=0&gt; N_cens;\n  array[N_obs] real y_obs;\n}\nparameters {\n  real&lt;upper=min(y_obs)&gt; L;\n  real mu;\n  real&lt;lower=0&gt; sigma;\n}\nmodel {\n  L ~ normal(mu, sigma);\n  y_obs ~ normal(mu, sigma);\n  target += N_cens * normal_lcdf(L | mu, sigma);\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Example Models",
+      "Truncated or Censored Data"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#user-defined-probability-functions",
-    "href": "stan-users-guide/user-functions.html#user-defined-probability-functions",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/using-stanc.html",
+    "href": "stan-users-guide/using-stanc.html",
+    "title": "Using the Stan Compiler",
     "section": "",
-    "text": "Probability functions are distinguished in Stan by names ending in _lpdf for density functions and _lpmf for mass functions; in both cases, they must have real return types.\nSuppose a model uses several standard normal distributions, for which there is not a specific overloaded density nor defaults in Stan. So rather than writing out the location of 0 and scale of 1 for all of them, a new density function may be defined and reused.\nfunctions {\n  real unit_normal_lpdf(real y) {\n    return normal_lpdf(y | 0, 1);\n  }\n}\n// ...\nmodel {\n  alpha ~ unit_normal();\n  beta ~ unit_normal();\n  // ...\n}\nThe ability to use the unit_normal function as a density is keyed off its name ending in _lpdf (names ending in _lpmf for probability mass functions work the same way).\nIn general, if foo_lpdf is defined to consume \\(N + 1\\) arguments, then\ny ~ foo(theta1, ..., thetaN);\ncan be used as shorthand for\ntarget += foo_lpdf(y | theta1, ..., thetaN);\nAs with the built-in functions, the suffix _lpdf is dropped and the first argument moves to the left of the tilde symbol (~) in the distribution statement.\nFunctions ending in _lpmf (for probability mass functions), behave exactly the same way. The difference is that the first argument of a density function (_lpdf) must be continuous (not an integer or integer array), whereas the first argument of a mass function (_lpmf) must be discrete (integer or integer array).",
+    "text": "Stan is used in most of our interfaces through the Stan compiler stanc. Since version 2.22, the Stan compiler has been implemented in OCaml and is referred to as stanc3. The binary name is still simply stanc, so this document uses both stanc and stanc3 interchangeably.\n\n\nThe stanc3 compiler has the following command-line syntax:\n&gt; stanc (options) &lt;model_file&gt;\nwhere &lt;model_file&gt; is either a path to a file ending in .stan or .stanfunctions (which automatically sets --standalone-functions), or '-' to read from standard input.\nThe stanc3 options are:\n\n--help / -? - Displays the complete list of stanc3 options, then exits.\n--version - Display stanc version number\n--info - Print information about the model, such as the type information for variables and the list of used distributions.\n--name=&lt;model_name&gt; - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.\n--o=&lt;file_name&gt; / -o=&lt;filename&gt; / --output=&lt;filename&gt; - Specify a path to an output file for generated C++ code (default = .hpp) or auto-formatting output (default: no file/print to stdout)\n--auto-format - Pretty prints the program to the console. See more on auto formatting.\n--allow-undefined - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.\n--canonicalize - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.\n--include_paths=&lt;dir1,...dirN&gt; - Takes a comma-separated list of directories that may contain a file in an #include directive.\n--max-line-length=&lt;number&gt; - Set the column number at which formatting with --auto-format attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.\n--print-canonical - Synonymous with --auto-format --canonicalize=deprecations,includes,parentheses,braces.\n--print-cpp - If set, output the generated C++ Stan model class to stdout.\n--filename-in-msg=&lt;name&gt; - Sets the filename used in compiler and runtime errors. If absent, the &lt;model_file&gt; argument is used.\n--standalone-functions - If set, only generate the code for the functions defined in the file. This is the default behavior for .stanfunctions files.\n--O0 (Default) Do not apply optimizations to the Stan code.\n--O1 Apply level 1 compiler optimizations (only basic optimizations).\n--Oexperimental WARNING: This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance! Allow the compiler to apply all optimizations to the Stan code.\n--O Synonym for --O1 as of Stan 2.37. In earlier versions this was a synonym for --Oexperimental.\n--use-opencl - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.\n--warn-pedantic - Emit warnings in Pedantic mode which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.\n--warn-uninitialized - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.\n--color - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are auto (the default), always, never. Can also be controlled by the STANC_COLOR environment variable.\n\nThe compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the --help option to see the full set.\n\n\n\nDuring model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).\n\n\nEven without the optional --warn-pedantic and --warn-uninitialized command line flags, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations\n\nA completely blank Stan program will produce the following warning message\nWarning in 'empty.stan', line 1, column 0 to line 2, column 0:\n    Empty model detected; this is a valid Stan model but likely unintended!\nThe use of any deprecated features will lead to warnings which will look as follows\n Warning in 'deprecated.stan', line 2, column 10 to column 17:\n     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with\n     an independent lognormal distribution on the scales, see:\n     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution\nA single Stan program can produce many warnings during compilation.\n\n\n\n\nErrors differ from warnings in their severity and format. In particular, errors are fatal and stop compilation, so at most one error is displayed per run of stanc.\nThere are five kinds of errors emitted by stanc3\n\nFile errors occur when the file passed to stanc is either missing or cannot be opened (i.e. has permissions issues). They look like\nError: file 'notfound.stan' not found or cannot be opened\nSyntactic errors occur whenever a program violates the Stan language’s syntax requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the #include directive fails, and “parsing” errors which result when the structure of the program is incorrect.\n\nThe lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of $ in a variable name will look like the following.\nSyntax error in 'char.stan', line 2, column 7 to column 8, lexing error:\n-------------------------------------------------\n  1:  data {\n  2:     int $ome_variable;\n             ^\n  3:  }\n-------------------------------------------------\nInvalid character found.\nWhen an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).\nSyntax error in './incl.stan', line 1, column 0, included from\n'./incl.stan', line 1, column 0, included from\n'incl.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include &lt;incl.stan&gt;\n         ^\n   -------------------------------------------------\nFile incl.stan recursively included itself.\nIt is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.\nSyntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:\n   -------------------------------------------------\n     1:  data {\n     2:     int&lt;lower=0&gt; N;\n     3:     vector x;\n                   ^\n     4:  }\n   -------------------------------------------------\nIll-formed type. Expected \"[\" expression \"]\" for vector size.\n\nSemantic errors (also known as type errors) occur when a program is structured correctly but features an error in the type rules imposed by the language. An example of this is assigning a real value to a variable defined as an integer.\nSemantic error in 'type.stan', line 2, column 3 to column 15:\n   -------------------------------------------------\n     1:  transformed data {\n     2:     int x = 1.5;\n            ^\n     3:  }\n   -------------------------------------------------\nIll-typed arguments supplied to assignment operator =:\nThe left hand side has type\n  int\nand the right hand side has type\n  real\nThe compiler will raise an error for use of any removed features for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.\nFinally, the compiler can raise an internal error. These are caused by bugs in the compiler, not your model, and we would appreciate it if you report them on the stanc3 repo with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.\n\n\n\n\n\nPedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.\nFor example, consider the following program.\ndata {\n  int N;\n  array[N] real x;\n}\nparameters {\n  real sigma;\n}\nmodel {\n  real mu;\n  x ~ normal(mu, sigma);\n}\nWhen pedantic mode is turned on, the compiler will produce the following warnings.\nWarning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:\n    The parameter sigma has no priors. This means either no prior is\n    provided, or the prior(s) depend on data variables. In the later case,\n    this may be a false positive.\nWarning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:\n    The variable mu may not have been assigned a value before its use.\nWarning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:\n    A normal distribution is given parameter sigma as a scale parameter\n    (argument 2), but sigma was not constrained to be strictly positive.\nHere are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):\n\nDistribution usages issues. Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.\nUnused parameter. A parameter is defined but doesn’t contribute to target.\nLarge or small constant in a distribution. Very large or very small constants are used as distribution arguments.\nControl flow depends on a parameter. Branching control flow (like if/else) depends on a parameter value .\nParameter has multiple tildes. A parameter is on the left-hand side of multiple tildes.\nParameter has zero or multiple priors. A parameter has zero or more than one prior distribution.\nVariable is used before assignment. A variable is used before being assigned a value.\nStrict or nonsensical parameter bounds. A parameter is given questionable bounds.\nNonlinear transformations. When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.\n\nSome important limitations of pedantic mode are listed at the end of this chapter.\n\n\nWhen an argument to a built-in distribution certainly does not match that distribution’s specification in the Stan Functions Reference, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g. simplex) is incompatible with the domain of the distribution. x\nFor example, consider the following program.\nparameters {\n  real unb_p;\n  real&lt;lower=0&gt; pos_p;\n}\nmodel {\n  1 ~ poisson(unb_p);\n  1 ~ poisson(pos_p);\n}\nThe parameter of poisson should be strictly positive, but unb_p is not constrained to be positive.\nPedantic mode produces the following warning.\nWarning in 'ex-dist-args.stan', line 6, column 14 to column 19:\n  A poisson distribution is given parameter unb_p as a rate parameter\n  (argument 1), but unb_p was not constrained to be strictly positive.\n\n\n\nPedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:\n\n\nAny use of uniform distribution generates a warning, except when the variate parameter’s declared upper and lower bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.\nFor example, consider the following program.\nparameters {\n  real a;\n  real&lt;lower=0, upper=1&gt; b;\n}\nmodel {\n  a ~ uniform(0, 1);\n  b ~ uniform(0, 1);\n}\na is assigned a uniform distribution that doesn’t match its constraints.\nPedantic mode produces the following warning.\nWarning in 'uniform-warn.stan', line 6, column 2 to column 20:\n  Parameter a is given a uniform distribution. The uniform distribution is\n  not recommended, for two reasons: (a) Except when there are logical or\n  physical constraints, it is very unusual for you to be sure that a\n  parameter will fall inside a specified range, and (b) The infinite gradient\n  induced by a uniform density can cause difficulties for Stan's sampling\n  algorithm. As a consequence, we recommend soft constraints rather than hard\n  constraints; for example, instead of giving an elasticity parameter a\n  uniform(0, 1) distribution, try normal(0.5, 0.5).\n\n\n\nGamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.\n\n\n\nAny use of the lkj_corr distribution generates a warning that suggests using the Cholesky variant instead. See the LKJ correlation distribution section of the Stan Functions Reference for details.\n\n\n\n\nA warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the target variable depends in any way on each of the parameters.\nFor example, consider the following program.\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(1, 1);\n}\na participates in the density function but b does not.\nPedantic mode produces the following warning.\nWarning in 'unused.stan', line 3, column 2 to column 9:\n    The parameter b was declared but was not used in the density calculation.\n\n\n\nWhen numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See the efficiency tuning section of the Stan User’s guide for a discussion of scaling parameters.\nFor example, consider the following program.\nparameters {\n  real x;\n  real y;\n}\nmodel {\n  x ~ normal(-100, 100);\n  y ~ normal(0, 1);\n}\nThe constants -100 and 100 suggest that x is not unit scaled.\nPedantic mode produces the following warning.\nWarning in 'constants-warn.stan', line 6, column 13 to column 17:\n    Argument -100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\nWarning in 'constants-warn.stan', line 6, column 19 to column 22:\n    Argument 100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\n\n\n\nControl flow statements, such as if, for and while should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.\nFor example, consider the following program.\nparameters {\n  real a;\n}\nmodel {\n  // x depends on parameter a\n  real x = a * a;\n\n  int m;\n\n  // the if-then-else depends on x which depends on a\n  if(x &gt; 0) {\n    //now m depends on x which depends on a\n    m = 1;\n  } else {\n    m = 2;\n  }\n\n  // for loop depends on m -&gt; x -&gt; a\n  for (i in 0:m) {\n    a ~ normal(i, 1);\n  }\n}\nThe if and for statements are control flow that depend (indirectly) on the value of the parameter m.\nPedantic mode produces the following warning.\nWarning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:\n  A control flow statement depends on parameter(s): a.\nWarning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:\n  A control flow statement depends on parameter(s): a.\n\n\n\nA warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent target += conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.\nPedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(0, 1);\n  a ~ normal(x, 1);\n\n  b ~ normal(1, 1);\n}\nPedantic mode produces the following warning.\nWarning in 'multi-tildes.stan', line 9, column 2 to column 19:\n  The parameter a is on the left-hand side of more than one tildes\n  statement.\n\n\n\nA warning is generated when a parameter appears to have greater than or less than one prior distribution factor.\nThis analysis depends on a factor graph representation of a Stan program. A factor F that depends on a parameter P is called a prior factor for P if there is no path in the factor graph from F to any data variable except through P.\nOne limitation of this approach is that the compiler cannot distinguish between modeled data variables and other convenient uses of data variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for int variables) are modeled data, which may cause extra warnings.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n  real c;\n  real d;\n}\nmodel\n{\n  a ~ normal(0, 1); // this is a prior\n  x ~ normal(a, 1); // this is not a prior, since data is involved\n\n  b ~ normal(x, 1); // this is also not a prior, since data is involved\n\n  // this is not a prior for c, since data is involved through b\n  // but it is a prior for b, since the data is only involved through b\n  c ~ normal(b, 1);\n\n  //these are multiple priors:\n  d ~ normal(0, 1);\n  1 ~ normal(d, 1);\n}\nOne prior is found for a and for b, while c only has a factor that touches a data variable and d has multiple priors.\nPedantic mode produces the following warning.\nWarning in 'priors.stan', line 7, column 2 to column 9:\n    The parameter c has no priors. This means either no prior is provided, or\n    the prior(s) depend on data variables. In the later case, this may be a\n    false positive.\nWarning in 'priors.stan', line 8, column 2 to column 9:\n    The parameter d has 2 priors.\n\n\n\n\nA warning is generated when any variable is used before it has been assigned a value.\nFor example, consider the following program.\ntransformed data {\n  real x;\n  if (1 &gt; 2) {\n    x = 1;\n  } else {\n    print(\"oops\");\n  }\n  print(x);\n}\nSince x is only assigned in one of the branches of the if statement, it might get to print(x) without having been assigned to.\nPedantic mode produces the following warning.\nWarning in 'uninit-warn.stan', line 7, column 8 to column 9:\n  The variable x may not have been assigned a value before its use.\n\n\n\nExcept when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds &lt;lower=.., upper=..&gt; except for &lt;lower=0, upper=1&gt; or &lt;lower=-1, upper=1&gt;.\nIn addition, a warning is generated when a parameter bound is found to have lower &gt;= upper.\nFor example, consider the following program.\nparameters {\n  real&lt;lower=0, upper=1&gt; a;\n  real&lt;lower=-1, upper=1&gt; b;\n  real&lt;lower=-2, upper=1012&gt; c;\n}\nmodel {\n  c ~ normal(b, a);\n}\nPedantic mode produces the following warning.\nWarning in 'hard-constraint.stan', line 4, column 2 to column 31:\n    Your Stan program has a parameter c with a lower and upper bound in its\n    declaration. These hard constraints are not recommended, for two reasons:\n    (a) Except when there are logical or physical constraints, it is very\n    unusual for you to be sure that a parameter will fall inside a specified\n    range, and (b) The infinite gradient induced by a hard constraint can\n    cause difficulties for Stan's sampling algorithm. As a consequence, we\n    recommend soft constraints rather than hard constraints; for example,\n    instead of constraining an elasticity parameter to fall between 0, and 1,\n    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.\n\n\n\nWhen a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the Changes of variables section.\nThis portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.\nFor example, consider the following program.\nparameters {\n  real y;\n}\nmodel {\n  log(y) ~ normal(0,1);\n}\nPedantic mode produces the following warning.\nWarning in 'jacobian.stan', line 5, column 2 to column 23:\n    Left-hand side of distribution statement (~) may contain a non-linear\n    transform of a parameter or local variable. If it does, you need to\n    include a target += statement with the log absolute determinant of the\n    Jacobian of the transform. You could also consider defining a transformed\n    parameter and using jacobian += in the transformed parameters block.\n\n\n\n\nConstant values are sometimes uncomputable\nPedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code normal(x, 1 - 2), the expression 1 - 2 will be evaluated to -1, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.\nContainer types\nCurrently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).\nData variables\nThe declaration information for data variables is currently not considered, so using data as incompatible arguments to distributions may not generate the appropriate warnings.\nControl flow dependent on parameters in nested functions\nIf a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.\n\n\n\n\n\nIn addition to compiling Stan programs, stanc3 features several flags which can be used to format Stan programs and update them to the most recent Stan syntax by removing any deprecation features which can be automatically replaced.\nThese flags work for both .stan model files and .stanfunctions function files. They can be combined with --o to redirect the formatted output to a new file.\n\n\nInvoking stanc --auto-format &lt;model_file&gt; will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the Stan Program Style Guide. This means spacing, indentation, and line length are all regularized. Some deprecated features, like the use of # for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.\nBy default, this will try to split lines at or before column 78. This number can be changed using --max-line-length.\n\n\n\nIn addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of if statements and for and while loops.\nThis can be done by using stanc --auto-format --canonicalize=... where ... is a comma-separated list of options. Currently these options are:\n\ndeprecations\nRemoves deprecated syntax such as replacing deprecated functions with their drop-in replacements.\nparentheses\nRemoves unnecessary extra parentheses, such as converting y = ((x-1)) to y = x - 1\nbraces\nPlaces braces around all blocks. For example, the following statement\nif (cond)\n  //result\nwill be formatted as\nif (cond) {\n  //result\n}\nand similarly for both kinds of loops containing a single statement.\nincludes\nThis will pretty-print code from other files included with #include as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same #include directives as the input program.\n\nInvoking stanc --print-canonical &lt;model_file&gt; is synonymous with running stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes\n\n\n\nThe formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:\n\nOddly placed comments\nIf your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string ^^^: to indicate they originally appeared higher in the program.\nWe hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.\nFailure to recreate strange #include structure\nPrinting without include inlining (--canonicalize=includes) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.\n\n\n\n\n\nThe stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.\nThis section introduces the available optimization options and describes their effect.\nTo print out a representation of the optimized Stan program, use the stanc3 command-line flag --debug-optimized-mir-pretty. To print an analogous representation of the Stan program prior to optimization, use the flag --debug-transformed-mir-pretty.\n\n\nTo turn optimizations on, the user specifies the desired optimization level. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.\nOptimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:\n\nO0 No optimizations are applied.\nO1 Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.\nOexperimental All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.\n\nO0 is the default setting.\nThe levels include these optimizations:\n\nO0 includes no optimizations.\nO1 includes:\n\nDead code elimination\nCopy propagation\nConstant propagation\nPartial evaluation\nFunction inlining\nMatrix memory layout optimization\n\nOexperimental includes optimizations specified by O1 and also:\n\nAutomatic-differentiation level optimization\nOne step loop unrolling\nExpression propagation\nLazy code motion\nStatic loop unrolling\n\n\nIn addition, Oexperimental will apply more repetitions of the optimizations, which may increase compile times.\n\n\n\n\n\nDead code is code that does not affect the behavior of the program. Code is not dead if it affects target, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.\nExample Stan program:\nmodel {\n  int i;\n  i = 5;\n  for (j in 1:10);\n  if (0) {\n    print(\"Dead code\");\n  } else {\n    print(\"Hi!\");\n  }\n}\nCompiler representation of program before dead code elimination (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  int i = 5;\n  for(j in 1:10) {\n    ;\n  }\n  if(0) {\n    FnPrint__(\"Dead code\");\n  } else {\n    FnPrint__(\"Hi!\");\n  }\n}\nCompiler representation of program after dead code elimination (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i;\n  FnPrint__(\"Hi!\");\n}\n\n\n\nConstant propagation replaces uses of a variable which is known to have a constant value C with that constant C. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).\nExample Stan program:\ntransformed data {\n  int n = 100;\n  int a[n];\n  for (i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before constant propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, n] a;\n  for(i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after constant propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, 100] a;\n  for(i in 1:100) {\n    a[i] = i;\n  }\n}\n\n\n\nCopy propagation is similar to expression propagation, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.\nExample Stan program:\nmodel {\n  int i = 1;\n  int j = i;\n  int k = i + j;\n}\nCompiler representation of program before copy propagation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n    int i = 1;\n    int j = i;\n    int k = (i + j);\n}\nCompiler representation of program after copy propagation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i = 1;\n  int j = i;\n  int k = (i + i);\n}\n\n\n\nPartial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.\nExample Stan program:\nmodel {\n  real a = 1 + 1;\n  real b = log(1 - a);\n  real c = a + b * 5;\n}\nCompiler representation of program before partial evaluation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real a = (1 + 1);\n  real b = log((1 - a));\n  real c = (a + (b * 5));\n}\nCompiler representation of program after partial evaluation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  real a = 2;\n  real b = log1m(a);\n  real c = fma(b, 5, a);\n}\n\n\n\nFunction inlining replaces each function call to each user-defined function f with the body of f. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return x + y;\n  }\n}\ntransformed data {\n  int a = 2;\n  int b = incr(a);\n}\nCompiler representation of program before function inlining (simplified from the output of --debug-transformed-mir-pretty):\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return (x + y);\n  }\n}\n\nprepare_data {\n  data int a = 2;\n  data int b = incr(a);\n}\nCompiler representation of program after function inlining (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int a;\n  a = 2;\n  data int b;\n  data int inline_sym1__;\n  data int inline_sym3__;\n  inline_sym3__ = 0;\n  for(inline_sym4__ in 1:1) {\n    int inline_sym2__;\n    inline_sym2__ = 1;\n    inline_sym3__ = 1;\n    inline_sym1__ = (a + inline_sym2__);\n    break;\n  }\n  b = inline_sym1__;\n}\nIn this code, the for loop and break is used to simulate the behavior of a return statement. The value to be returned is held in inline_sym1__. The flag variable inline_sym3__ indicates whether a return has occurred and is necessary to handle return statements nested inside loops within the function body.\n\n\n\nMatrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.\nThe first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.\nThe second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.\nMore general reading on AoS vs SoA can be found on Wikipedia\nThis optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:\n\nAny Stan Math Library functions the matrix is passed to must be able to support it.\nThe matrix should not be accessed/assigned elementwise in a loop.\n\nThe debug flag --debug-mem-patterns will list each variable and whether it is using the AoS representation or the SoA representation.\n\n\n\n\n\n\nStan variables can have two auto-differentiation (AD) levels: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.\nExample Stan program:\ndata {\n  real y;\n}\nmodel {\n  real x = y + 1;\n}\nCompiler representation of program before AD-level optimization (simplified from the output of --debug-transformed-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  real x = (y + 1);\n}\nCompiler representation of program after AD-level optimization (simplified from the output of --debug-optimized-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  data real x = (y + 1);\n}\n\n\n\nOne step loop unrolling is similar to static loop unrolling. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.\nExample Stan program:\ndata {\n  int n;\n}\ntransformed data {\n  int x = 0;\n  for (i in 1:n) {\n    x += i;\n  }\n}\nCompiler representation of program before one step static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  data int x = 0;\n  for(i in 1:n) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after one step static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  int x = 0;\n  if((n &gt;= 1)) {\n    x = (x + 1);\n    for(i in (1 + 1):n) {\n      x = (x + i);\n    }\n  }\n}\n\n\n\nConstant propagation replaces the uses of a variable which is known to have a constant value E with that constant E. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by lazy code motion to avoid unnecessarily recomputing expressions.\nExample Stan program:\ndata {\n  int m;\n}\ntransformed data {\n  int n = m+1;\n  int a[n];\n  for (i in 1:n-1) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before expression propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, n] a;\n  for(i in 1:(n - 1)) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after expression propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, (m + 1)] a;\n  for(i in 1:((m + 1) - 1)) {\n    a[i] = i;\n  }\n}\n\n\n\nLazy code motion rearranges the statements and expressions in a program with the goals of:\n\nAvoiding computing expressions more than once, and\nComputing expressions as late as possible (to minimize the strain on the working memory set).\n\nTo accomplish these goals, lazy code motion will perform optimizations such as:\n\nMoving a repeatedly calculated expression to its own variable (also referred to as common-subexpression elimination)\nMoving an expression outside of a loop if it does not need to be in the loop (also referred to as loop-invariant code motion)\n\nLazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.\nAs currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.\nExample Stan program:\nmodel {\n  real x;\n  real y;\n  real z;\n\n  for (i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program before lazy code motion (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real x;\n  real y;\n  real z;\n  for(i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program after lazy code motion (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  data real lcm_sym4__;\n  data real lcm_sym3__;\n  real x;\n  real y;\n  lcm_sym4__ = sqrt(10);\n  real z;\n  for(i in 1:10) {\n    x = lcm_sym4__;\n    y = sqrt(i);\n  }\n  z = lcm_sym4__;\n}\n\n\n\nStatic loop unrolling takes a loop with a predictable number of iterations X and replaces it by writing out the loop body X times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\ntransformed data {\n  int x = 0;\n  for (i in 1:4) {\n    x += i;\n  }\n}\nCompiler representation of program before static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int x = 0;\n  for(i in 1:4) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int x;\n  x = 0;\n  x = (x + 1);\n  x = (x + 2);\n  x = (x + 3);\n  x = (x + 4);\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Appendices",
+      "Using the Stan Compiler"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#overloading-functions",
-    "href": "stan-users-guide/user-functions.html#overloading-functions",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/using-stanc.html#stanc-args",
+    "href": "stan-users-guide/using-stanc.html#stanc-args",
+    "title": "Using the Stan Compiler",
     "section": "",
-    "text": "As described in the reference manual function overloading is permitted in Stan, beginning in version 2.29.\nThis means multiple functions can be defined with the same name as long as they accept different numbers or types of arguments. User-defined functions can also overload Stan library functions.\n\n\nOverloading is a powerful productivity tool in programming languages, but it can also lead to confusion. In particular, it can be unclear at first glance which version of a function is being called at any particular call site, especially with type promotion allowed between scalar types. Because of this, it is a programming best practice that overloaded functions maintain the same meaning across definitions.\nFor example, consider a function triple which has the following three signatures\nreal triple(real x);\ncomplex triple(complex x);\narray[] real triple(array[] real);\nOne should expect that all overloads of this function perform the same basic task. This should lead to definitions of these functions which would satisfy the following assumptions that someone reading the program would expect\n// The function does what it says\ntriple(3.0) == 9.0\n// It is defined reasonably for different types\ntriple(to_complex(3.0)) == to_complex(triple(3.0))\n// A container version of this function works by element\ntriple({3.0, 4.0})[0] == triple({3.0, 4.0}[0])\nNote that none of these properties are enforced by Stan, they are mentioned merely to warn against uses of overloading which cause confusion.\n\n\n\nStan resolves overloaded functions by the number and type of arguments passed to the function. This can be subtle when multiple signatures with the same number of arguments are present.\nConsider the following function signatures\nreal foo(int a, real b);\nreal foo(real a, real b);\nGiven these, the function call foo(1.5, 2.5) is unambiguous - it must resolve to the second signature. But, the function call foo(1, 1.5) could be valid for either under Stan’s promotion rules, which allow integers to be promoted to real numbers.\nTo resolve this, Stan selects the signature which requires the fewest number of promotions for a given function call. In the above case, this means the call foo(1, 1.5) would select the first signature, because it requires 0 promotions (the second signature would require 1 promotion).\nFurthermore, there must be only one such signature, e.g., the minimum number of promotions must be a unique minimum. This requirement forbids certain kinds of overloading. For example, consider the function signatures\nreal bar(int x, real y);\nreal bar(real x, int y);\nThese signatures do not have a unique minimum number of promotions for the call bar(1, 2). Both signatures require one int to real promotion, and so it cannot be determined which is correct. Stan will produce a compilation error in this case.\nPromotion from integers to complex numbers is considered to be two separate promotions, first from int to real, then from real to complex. This means that integer arguments will “prefer” a signature with real types over complex types.\nFor example, consider the function signatures\nreal pop(real x);\nreal pop(complex x);\nStan will select the first signature when pop is called with an integer argument such as pop(0).",
+    "text": "The stanc3 compiler has the following command-line syntax:\n&gt; stanc (options) &lt;model_file&gt;\nwhere &lt;model_file&gt; is either a path to a file ending in .stan or .stanfunctions (which automatically sets --standalone-functions), or '-' to read from standard input.\nThe stanc3 options are:\n\n--help / -? - Displays the complete list of stanc3 options, then exits.\n--version - Display stanc version number\n--info - Print information about the model, such as the type information for variables and the list of used distributions.\n--name=&lt;model_name&gt; - Specify the name of the class used for the implementation of the Stan model in the generated C++ code.\n--o=&lt;file_name&gt; / -o=&lt;filename&gt; / --output=&lt;filename&gt; - Specify a path to an output file for generated C++ code (default = .hpp) or auto-formatting output (default: no file/print to stdout)\n--auto-format - Pretty prints the program to the console. See more on auto formatting.\n--allow-undefined - Do not throw a parser error if there is a function in the Stan program that is declared but not defined in the functions block.\n--canonicalize - Make changes to the program before pretty-printing by specifying options in a comma separated list. Options are ‘deprecations’, ‘parentheses’, ‘braces’, ‘includes’, and ‘strip-comments’.\n--include_paths=&lt;dir1,...dirN&gt; - Takes a comma-separated list of directories that may contain a file in an #include directive.\n--max-line-length=&lt;number&gt; - Set the column number at which formatting with --auto-format attempts to split lines. The default value is 78, which results in most lines being shorter than 80 characters.\n--print-canonical - Synonymous with --auto-format --canonicalize=deprecations,includes,parentheses,braces.\n--print-cpp - If set, output the generated C++ Stan model class to stdout.\n--filename-in-msg=&lt;name&gt; - Sets the filename used in compiler and runtime errors. If absent, the &lt;model_file&gt; argument is used.\n--standalone-functions - If set, only generate the code for the functions defined in the file. This is the default behavior for .stanfunctions files.\n--O0 (Default) Do not apply optimizations to the Stan code.\n--O1 Apply level 1 compiler optimizations (only basic optimizations).\n--Oexperimental WARNING: This is currently an experimental feature whose components are not thoroughly tested and may not improve a programs performance! Allow the compiler to apply all optimizations to the Stan code.\n--O Synonym for --O1 as of Stan 2.37. In earlier versions this was a synonym for --Oexperimental.\n--use-opencl - If set, will use additional Stan OpenCL features enabled in the Stan-to-C++ compiler.\n--warn-pedantic - Emit warnings in Pedantic mode which warns of potential issues in the meaning of your program. Note: This may produce false positive warnings.\n--warn-uninitialized - Emit warnings about uninitialized variables to stderr. Currently an experimental feature.\n--color - Control whether errors and warnings are emitted with colored styling on terminals that support it. Valid values are auto (the default), always, never. Can also be controlled by the STANC_COLOR environment variable.\n\nThe compiler also provides a number of debug options which are primarily of interest to stanc3 developers; use the --help option to see the full set.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Appendices",
+      "Using the Stan Compiler"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#documenting-functions.section",
-    "href": "stan-users-guide/user-functions.html#documenting-functions.section",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/using-stanc.html#understanding-stanc3-errors-and-warnings",
+    "href": "stan-users-guide/using-stanc.html#understanding-stanc3-errors-and-warnings",
+    "title": "Using the Stan Compiler",
     "section": "",
-    "text": "Functions will ideally be documented at their interface level. The Stan style guide for function documentation follows the same format as used by the Doxygen (C++) and Javadoc (Java) automatic documentation systems. Such specifications indicate the variables and their types and the return value, prefaced with some descriptive text.\nFor example, here’s some documentation for the prediction matrix generator.\n/**\n * Return a data matrix of specified size with rows\n * corresponding to items and the first column filled\n * with the value 1 to represent the intercept and the\n * remaining columns randomly filled with unit-normal draws.\n *\n * @param N Number of rows corresponding to data items\n * @param K Number of predictors, counting the intercept, per\n *          item.\n * @return Simulated predictor matrix.\n */\nmatrix predictors_rng(int N, int K) {\n  // ...\nThe comment begins with /**, ends with */, and has an asterisk (*) on each line. It uses @param followed by the argument’s identifier to document a function argument. The tag @return is used to indicate the return value. Stan does not (yet) have an automatic documentation generator like Javadoc or Doxygen, so this just looks like a big comment starting with /* and ending with */ to the Stan parser.\nFor functions that raise exceptions, exceptions can be documented using @throws.4\nFor example,\n /** ...\n * @param theta\n * @throws If any of the entries of theta is negative.\n */\nreal entropy(vector theta) {\n  // ...\n}\nUsually an exception type would be provided, but these are not exposed as part of the Stan language, so there is no need to document them.",
+    "text": "During model compilation, stanc can produce a variety of errors (issues that prevent the model from being compiled) and warnings (non-fatal issues that should still be considered).\n\n\nEven without the optional --warn-pedantic and --warn-uninitialized command line flags, both of which enable additional warnings, stanc can still produce warnings about your program. In particular, warnings will be produced in two situations\n\nA completely blank Stan program will produce the following warning message\nWarning in 'empty.stan', line 1, column 0 to line 2, column 0:\n    Empty model detected; this is a valid Stan model but likely unintended!\nThe use of any deprecated features will lead to warnings which will look as follows\n Warning in 'deprecated.stan', line 2, column 10 to column 17:\n     lkj_cov is deprecated and will be removed in Stan 3.0. Use lkj_corr with\n     an independent lognormal distribution on the scales, see:\n     https://mc-stan.org/docs/reference-manual/deprecations.html#lkj_cov-distribution\nA single Stan program can produce many warnings during compilation.\n\n\n\n\nErrors differ from warnings in their severity and format. In particular, errors are fatal and stop compilation, so at most one error is displayed per run of stanc.\nThere are five kinds of errors emitted by stanc3\n\nFile errors occur when the file passed to stanc is either missing or cannot be opened (i.e. has permissions issues). They look like\nError: file 'notfound.stan' not found or cannot be opened\nSyntactic errors occur whenever a program violates the Stan language’s syntax requirements. There are three kinds of errors within syntax errors; “lexing” errors mean that the input was unable to be read properly on the character level, “include” errors which occur when the #include directive fails, and “parsing” errors which result when the structure of the program is incorrect.\n\nThe lexing errors occur due to the use of invalid characters in a program. For example, a lexing error due to the use of $ in a variable name will look like the following.\nSyntax error in 'char.stan', line 2, column 7 to column 8, lexing error:\n-------------------------------------------------\n  1:  data {\n  2:     int $ome_variable;\n             ^\n  3:  }\n-------------------------------------------------\nInvalid character found.\nWhen an include directive is used, it can lead to errors if the included file is not found, or if a file includes itself (including a recursive loop of includes, such as A -&gt; B -&gt; A).\nSyntax error in './incl.stan', line 1, column 0, included from\n'./incl.stan', line 1, column 0, included from\n'incl.stan', line 1, column 0, include error:\n   -------------------------------------------------\n     1:  #include &lt;incl.stan&gt;\n         ^\n   -------------------------------------------------\nFile incl.stan recursively included itself.\nIt is much more common to see parsing errors, which tend to have more in-depth explanations of the error found. For example, if a user forgets to put a size on a type like vector, as in the following, this raises a parsing (structural) error in the compiler.\nSyntax error in 'vec.stan', line 3, column 10 to column 11, parsing error:\n   -------------------------------------------------\n     1:  data {\n     2:     int&lt;lower=0&gt; N;\n     3:     vector x;\n                   ^\n     4:  }\n   -------------------------------------------------\nIll-formed type. Expected \"[\" expression \"]\" for vector size.\n\nSemantic errors (also known as type errors) occur when a program is structured correctly but features an error in the type rules imposed by the language. An example of this is assigning a real value to a variable defined as an integer.\nSemantic error in 'type.stan', line 2, column 3 to column 15:\n   -------------------------------------------------\n     1:  transformed data {\n     2:     int x = 1.5;\n            ^\n     3:  }\n   -------------------------------------------------\nIll-typed arguments supplied to assignment operator =:\nThe left hand side has type\n  int\nand the right hand side has type\n  real\nThe compiler will raise an error for use of any removed features for at least one version following their removal. The deprecation warnings mentioned above eventually turn into this kind of error to prompt the user to update their model. After the version of removal, these errors will be converted to one of the other types listed here, depending on the feature.\nFinally, the compiler can raise an internal error. These are caused by bugs in the compiler, not your model, and we would appreciate it if you report them on the stanc3 repo with the error message provided. These errors usually say something like “This should never happen,” and we apologize if they do.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Appendices",
+      "Using the Stan Compiler"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#summary-of-function-types",
-    "href": "stan-users-guide/user-functions.html#summary-of-function-types",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/using-stanc.html#pedantic-mode",
+    "href": "stan-users-guide/using-stanc.html#pedantic-mode",
+    "title": "Using the Stan Compiler",
     "section": "",
-    "text": "Functions may have a void or non-void return type and they may or may not have one of the special suffixes, _lpdf, _lpmf, _lp, or _rng.\n\n\nOnly functions declared to return void may be used as statements. These are also the only functions that use return statements with no arguments.\nOnly functions declared to return non-void values may be used as expressions. These functions require return statements with arguments of a type that matches the declared return type.\n\n\n\nOnly functions ending in _lpmf or _lpdf and with return type real may be used as probability functions in distribution statements.\nOnly functions ending in _lp may access the log probability accumulator through distribution statements or target += statements. Such functions may only be used in the transformed parameters or model blocks.\nOnly functions ending in _rng may access the built-in pseudo-random number generators. Such functions may only be used in the generated quantities block or transformed data block, or in the bodies of other user-defined functions ending in _rng.",
+    "text": "Pedantic mode is a compilation option built into Stanc3 that warns you about potential issues in your Stan program.\nFor example, consider the following program.\ndata {\n  int N;\n  array[N] real x;\n}\nparameters {\n  real sigma;\n}\nmodel {\n  real mu;\n  x ~ normal(mu, sigma);\n}\nWhen pedantic mode is turned on, the compiler will produce the following warnings.\nWarning in 'ped-mode-ex1.stan', line 6, column 2 to column 13:\n    The parameter sigma has no priors. This means either no prior is\n    provided, or the prior(s) depend on data variables. In the later case,\n    this may be a false positive.\nWarning in 'ped-mode-ex1.stan', line 10, column 13 to column 15:\n    The variable mu may not have been assigned a value before its use.\nWarning in 'ped-mode-ex1.stan', line 10, column 17 to column 22:\n    A normal distribution is given parameter sigma as a scale parameter\n    (argument 2), but sigma was not constrained to be strictly positive.\nHere are the kinds of issues that pedantic mode will find (which are described in more detail in following sections):\n\nDistribution usages issues. Distribution arguments don’t match the distribution specification, or some specific distribution is used in an inadvisable way.\nUnused parameter. A parameter is defined but doesn’t contribute to target.\nLarge or small constant in a distribution. Very large or very small constants are used as distribution arguments.\nControl flow depends on a parameter. Branching control flow (like if/else) depends on a parameter value .\nParameter has multiple tildes. A parameter is on the left-hand side of multiple tildes.\nParameter has zero or multiple priors. A parameter has zero or more than one prior distribution.\nVariable is used before assignment. A variable is used before being assigned a value.\nStrict or nonsensical parameter bounds. A parameter is given questionable bounds.\nNonlinear transformations. When the left-hand side of a tilde statement (or first argument of a log probability function) contains a nonlinear transform which may require a Jacobian change of variables adjustment.\n\nSome important limitations of pedantic mode are listed at the end of this chapter.\n\n\nWhen an argument to a built-in distribution certainly does not match that distribution’s specification in the Stan Functions Reference, a warning is thrown. This primarily checks if any distribution argument’s bounds at declaration, compile-time value, or subtype at declaration (e.g. simplex) is incompatible with the domain of the distribution. x\nFor example, consider the following program.\nparameters {\n  real unb_p;\n  real&lt;lower=0&gt; pos_p;\n}\nmodel {\n  1 ~ poisson(unb_p);\n  1 ~ poisson(pos_p);\n}\nThe parameter of poisson should be strictly positive, but unb_p is not constrained to be positive.\nPedantic mode produces the following warning.\nWarning in 'ex-dist-args.stan', line 6, column 14 to column 19:\n  A poisson distribution is given parameter unb_p as a rate parameter\n  (argument 1), but unb_p was not constrained to be strictly positive.\n\n\n\nPedantic mode checks for some specific uses of distributions that may indicate a statistical mistake:\n\n\nAny use of uniform distribution generates a warning, except when the variate parameter’s declared upper and lower bounds exactly match the uniform distribution bounds. In general, assigning a parameter a uniform distribution can create non-differentiable boundary conditions and is not recommended.\nFor example, consider the following program.\nparameters {\n  real a;\n  real&lt;lower=0, upper=1&gt; b;\n}\nmodel {\n  a ~ uniform(0, 1);\n  b ~ uniform(0, 1);\n}\na is assigned a uniform distribution that doesn’t match its constraints.\nPedantic mode produces the following warning.\nWarning in 'uniform-warn.stan', line 6, column 2 to column 20:\n  Parameter a is given a uniform distribution. The uniform distribution is\n  not recommended, for two reasons: (a) Except when there are logical or\n  physical constraints, it is very unusual for you to be sure that a\n  parameter will fall inside a specified range, and (b) The infinite gradient\n  induced by a uniform density can cause difficulties for Stan's sampling\n  algorithm. As a consequence, we recommend soft constraints rather than hard\n  constraints; for example, instead of giving an elasticity parameter a\n  uniform(0, 1) distribution, try normal(0.5, 0.5).\n\n\n\nGamma distributions are sometimes used as an attempt to assign an improper prior to a parameter. Pedantic mode gives a warning when the Gamma arguments indicate that this may be the case.\n\n\n\nAny use of the lkj_corr distribution generates a warning that suggests using the Cholesky variant instead. See the LKJ correlation distribution section of the Stan Functions Reference for details.\n\n\n\n\nA warning is generated when a parameter is declared but does not have any effect on the program. This is determined by checking whether the value of the target variable depends in any way on each of the parameters.\nFor example, consider the following program.\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(1, 1);\n}\na participates in the density function but b does not.\nPedantic mode produces the following warning.\nWarning in 'unused.stan', line 3, column 2 to column 9:\n    The parameter b was declared but was not used in the density calculation.\n\n\n\nWhen numbers with magnitude less than 0.1 or greater than 10 are used as arguments to a distribution, it indicates that some parameter is not scaled to unit value, so a warning is thrown. See the efficiency tuning section of the Stan User’s guide for a discussion of scaling parameters.\nFor example, consider the following program.\nparameters {\n  real x;\n  real y;\n}\nmodel {\n  x ~ normal(-100, 100);\n  y ~ normal(0, 1);\n}\nThe constants -100 and 100 suggest that x is not unit scaled.\nPedantic mode produces the following warning.\nWarning in 'constants-warn.stan', line 6, column 13 to column 17:\n    Argument -100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\nWarning in 'constants-warn.stan', line 6, column 19 to column 22:\n    Argument 100 suggests there may be parameters that are not unit scale;\n    consider rescaling with a multiplier, see:\n    https://mc-stan.org/docs/stan-users-guide/efficiency-tuning.html#standardizing-predictors\n\n\n\nControl flow statements, such as if, for and while should not depend on parameters or functions of parameters to determine their branching conditions. This is likely to introduce a discontinuity into the density function. Pedantic mode generates a warning when any branching condition may depend on a parameter value.\nFor example, consider the following program.\nparameters {\n  real a;\n}\nmodel {\n  // x depends on parameter a\n  real x = a * a;\n\n  int m;\n\n  // the if-then-else depends on x which depends on a\n  if(x &gt; 0) {\n    //now m depends on x which depends on a\n    m = 1;\n  } else {\n    m = 2;\n  }\n\n  // for loop depends on m -&gt; x -&gt; a\n  for (i in 0:m) {\n    a ~ normal(i, 1);\n  }\n}\nThe if and for statements are control flow that depend (indirectly) on the value of the parameter m.\nPedantic mode produces the following warning.\nWarning in 'param-dep-cf-warn.stan', line 11, column 2 to line 16, column 3:\n  A control flow statement depends on parameter(s): a.\nWarning in 'param-dep-cf-warn.stan', line 19, column 2 to line 21, column 3:\n  A control flow statement depends on parameter(s): a.\n\n\n\nA warning is generated when a parameter is found on the left-hand side of more than one ~ statements (or an equivalent target += conditional density statement). This pattern is not inherently an issue, but it is unusual and may indicate a mistake.\nPedantic mode only searches for repeated statements, it will not for example generate a warning when a ~ statement is executed repeatedly inside of a loop.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n}\nmodel {\n  a ~ normal(0, 1);\n  a ~ normal(x, 1);\n\n  b ~ normal(1, 1);\n}\nPedantic mode produces the following warning.\nWarning in 'multi-tildes.stan', line 9, column 2 to column 19:\n  The parameter a is on the left-hand side of more than one tildes\n  statement.\n\n\n\nA warning is generated when a parameter appears to have greater than or less than one prior distribution factor.\nThis analysis depends on a factor graph representation of a Stan program. A factor F that depends on a parameter P is called a prior factor for P if there is no path in the factor graph from F to any data variable except through P.\nOne limitation of this approach is that the compiler cannot distinguish between modeled data variables and other convenient uses of data variables such as data sizes or hyperparameters. This warning assumes that all data variables (except for int variables) are modeled data, which may cause extra warnings.\nFor example, consider the following program.\ndata {\n  real x;\n}\nparameters {\n  real a;\n  real b;\n  real c;\n  real d;\n}\nmodel\n{\n  a ~ normal(0, 1); // this is a prior\n  x ~ normal(a, 1); // this is not a prior, since data is involved\n\n  b ~ normal(x, 1); // this is also not a prior, since data is involved\n\n  // this is not a prior for c, since data is involved through b\n  // but it is a prior for b, since the data is only involved through b\n  c ~ normal(b, 1);\n\n  //these are multiple priors:\n  d ~ normal(0, 1);\n  1 ~ normal(d, 1);\n}\nOne prior is found for a and for b, while c only has a factor that touches a data variable and d has multiple priors.\nPedantic mode produces the following warning.\nWarning in 'priors.stan', line 7, column 2 to column 9:\n    The parameter c has no priors. This means either no prior is provided, or\n    the prior(s) depend on data variables. In the later case, this may be a\n    false positive.\nWarning in 'priors.stan', line 8, column 2 to column 9:\n    The parameter d has 2 priors.\n\n\n\n\nA warning is generated when any variable is used before it has been assigned a value.\nFor example, consider the following program.\ntransformed data {\n  real x;\n  if (1 &gt; 2) {\n    x = 1;\n  } else {\n    print(\"oops\");\n  }\n  print(x);\n}\nSince x is only assigned in one of the branches of the if statement, it might get to print(x) without having been assigned to.\nPedantic mode produces the following warning.\nWarning in 'uninit-warn.stan', line 7, column 8 to column 9:\n  The variable x may not have been assigned a value before its use.\n\n\n\nExcept when there are logical or physical constraints, it is very unusual for you to be sure that a parameter will fall inside a specified range. A warning is generated for all parameters declared with the bounds &lt;lower=.., upper=..&gt; except for &lt;lower=0, upper=1&gt; or &lt;lower=-1, upper=1&gt;.\nIn addition, a warning is generated when a parameter bound is found to have lower &gt;= upper.\nFor example, consider the following program.\nparameters {\n  real&lt;lower=0, upper=1&gt; a;\n  real&lt;lower=-1, upper=1&gt; b;\n  real&lt;lower=-2, upper=1012&gt; c;\n}\nmodel {\n  c ~ normal(b, a);\n}\nPedantic mode produces the following warning.\nWarning in 'hard-constraint.stan', line 4, column 2 to column 31:\n    Your Stan program has a parameter c with a lower and upper bound in its\n    declaration. These hard constraints are not recommended, for two reasons:\n    (a) Except when there are logical or physical constraints, it is very\n    unusual for you to be sure that a parameter will fall inside a specified\n    range, and (b) The infinite gradient induced by a hard constraint can\n    cause difficulties for Stan's sampling algorithm. As a consequence, we\n    recommend soft constraints rather than hard constraints; for example,\n    instead of constraining an elasticity parameter to fall between 0, and 1,\n    leave it unconstrained and give it a normal(0.5,0.5) prior distribution.\n\n\n\nWhen a parameter is transformed in a non-linear fashion, an adjustment must be applied to account for distortion caused by the transform. This is discussed in depth in the Changes of variables section.\nThis portion of pedantic mode tries to detect instances where such an adjustment would be necessary and remind the user.\nFor example, consider the following program.\nparameters {\n  real y;\n}\nmodel {\n  log(y) ~ normal(0,1);\n}\nPedantic mode produces the following warning.\nWarning in 'jacobian.stan', line 5, column 2 to column 23:\n    Left-hand side of distribution statement (~) may contain a non-linear\n    transform of a parameter or local variable. If it does, you need to\n    include a target += statement with the log absolute determinant of the\n    Jacobian of the transform. You could also consider defining a transformed\n    parameter and using jacobian += in the transformed parameters block.\n\n\n\n\nConstant values are sometimes uncomputable\nPedantic mode attempts to evaluate expressions down to literal values so that they can be used to generate warnings. For example, in the code normal(x, 1 - 2), the expression 1 - 2 will be evaluated to -1, which is not a valid variance argument so a warning is generated. However, this strategy is limited; it is often impossible to fully evaluate expressions in finite time.\nContainer types\nCurrently, indexed variables are not handled intelligently, so they are treated as monolithic variables. Each analysis treats indexed variables conservatively (erring toward generating fewer warnings).\nData variables\nThe declaration information for data variables is currently not considered, so using data as incompatible arguments to distributions may not generate the appropriate warnings.\nControl flow dependent on parameters in nested functions\nIf a parameter is passed as an argument to a user-defined function within another user-defined function, and then some control flow depends on that argument, the appropriate warning will not be thrown.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Appendices",
+      "Using the Stan Compiler"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#recursive-functions",
-    "href": "stan-users-guide/user-functions.html#recursive-functions",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/using-stanc.html#stanc-pretty-printing",
+    "href": "stan-users-guide/using-stanc.html#stanc-pretty-printing",
+    "title": "Using the Stan Compiler",
     "section": "",
-    "text": "Stan supports recursive function definitions, which can be useful for some applications. For instance, consider the matrix power operation, \\(A^n\\), which is defined for a square matrix \\(A\\) and positive integer \\(n\\) by \\[\nA^n\n=\n\\begin{cases}\n\\textrm{I} & \\quad\\text{if } n = 0, \\text{ and} \\\\\nA \\, A^{n-1} & \\quad\\text{if } n &gt; 0.\n\\end{cases}\n\\]\nwhere \\(\\textrm{I}\\) is the identity matrix. This definition can be directly translated to a recursive function definition.\nmatrix matrix_pow(matrix a, int n) {\n  if (n == 0) {\n    return diag_matrix(rep_vector(1, rows(a)));\n  } else {\n    return a *  matrix_pow(a, n - 1);\n  }\n}\nIt would be more efficient to not allow the recursion to go all the way to the base case, adding the following conditional clause.\nelse if (n == 1) {\n  return a;\n}",
+    "text": "In addition to compiling Stan programs, stanc3 features several flags which can be used to format Stan programs and update them to the most recent Stan syntax by removing any deprecation features which can be automatically replaced.\nThese flags work for both .stan model files and .stanfunctions function files. They can be combined with --o to redirect the formatted output to a new file.\n\n\nInvoking stanc --auto-format &lt;model_file&gt; will print a version of your model which has been re-formatted. The goal is to have this automatic formatting stay as close as possible to the Stan Program Style Guide. This means spacing, indentation, and line length are all regularized. Some deprecated features, like the use of # for line comments, are replaced, but the goal is mainly to preserve the program while formatting it.\nBy default, this will try to split lines at or before column 78. This number can be changed using --max-line-length.\n\n\n\nIn addition to automatic formatting, stanc can also “canonicalize” programs by updating deprecated syntax, removing unnecessary parenthesis, and adding braces around bodies of if statements and for and while loops.\nThis can be done by using stanc --auto-format --canonicalize=... where ... is a comma-separated list of options. Currently these options are:\n\ndeprecations\nRemoves deprecated syntax such as replacing deprecated functions with their drop-in replacements.\nparentheses\nRemoves unnecessary extra parentheses, such as converting y = ((x-1)) to y = x - 1\nbraces\nPlaces braces around all blocks. For example, the following statement\nif (cond)\n  //result\nwill be formatted as\nif (cond) {\n  //result\n}\nand similarly for both kinds of loops containing a single statement.\nincludes\nThis will pretty-print code from other files included with #include as part of the program. This was the default behavior prior to Stan 2.29. When not enabled, the pretty-printer output will include the same #include directives as the input program.\n\nInvoking stanc --print-canonical &lt;model_file&gt; is synonymous with running stanc --auto-format --canonicalize=deprecations,braces,parentheses,includes\n\n\n\nThe formatting and canonicalizing features of stanc3 are still under development. The following are some known issues one should be aware of before using either:\n\nOddly placed comments\nIf your Stan program features comments in unexpected places, such as inside an expression, they may be moved in the process of formatting. Moved comments are prefixed with the string ^^^: to indicate they originally appeared higher in the program.\nWe hope to improve this functionality in future versions. For now, this can usually be avoided by manually moving the comment outside of an expression, either by placing it on its own line or following a separator such as a comma or keyword.\nFailure to recreate strange #include structure\nPrinting without include inlining (--canonicalize=includes) can fail when includes were used in atypical locations, such as in the middle of statements. We recommend either printing with inlining enabled or reconsidering the use of includes in this way.",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Appendices",
+      "Using the Stan Compiler"
     ]
   },
   {
-    "objectID": "stan-users-guide/user-functions.html#truncated-random-number-generation",
-    "href": "stan-users-guide/user-functions.html#truncated-random-number-generation",
-    "title": "User-Defined Functions",
+    "objectID": "stan-users-guide/using-stanc.html#optimization",
+    "href": "stan-users-guide/using-stanc.html#optimization",
+    "title": "Using the Stan Compiler",
     "section": "",
-    "text": "To generate random numbers, it is often sufficient to invert their cumulative distribution functions. This is built into many of the random number generators. For example, to generate a standard logistic variate, first generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then run through the inverse cumulative distribution function, \\(y = \\textrm{logit}(u)\\). If this were not already built in as logistic_rng(0, 1), it could be coded in Stan directly as\nreal standard_logistic_rng() {\n  real u = uniform_rng(0, 1);\n  real y = logit(u);\n  return y;\n}\nFollowing the same pattern, a standard normal RNG could be coded as\nreal standard_normal_rng() {\n  real u = uniform_rng(0, 1);\n  real y = inv_Phi(u);\n  return y;\n}\nthat is, \\(y = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}\\) is the inverse cumulative distribution function for the standard normal distribution, implemented in the Stan function inv_Phi.\nIn order to generate non-standard variates of the location-scale variety, the variate is scaled by the scale parameter and shifted by the location parameter. For example, to generate \\(\\textsf{normal}(\\mu, \\sigma)\\) variates, it is enough to generate a uniform variate \\(u \\sim \\textsf{uniform}(0, 1)\\), then convert it to a standard normal variate, \\(z = \\Phi^{-1}(u)\\), where \\(\\Phi^{-1}(\\cdot)\\) is the inverse cumulative distribution function for the standard normal, and then, finally, scale and translate it, \\(y = \\mu +\n\\sigma \\times z\\). In code,\nreal my_normal_rng(real mu, real sigma) {\n  real u = uniform_rng(0, 1);\n  real z = inv_Phi(u);\n  real y = mu + sigma * z;\n  return y;\n}\nA robust version of this function would test that the arguments are finite and that sigma is non-negative, e.g.,\n  if (is_nan(mu) || is_inf(mu)) {\n    reject(\"my_normal_rng: mu must be finite; \",\n           \"found mu = \", mu);\n  }\n  if (is_nan(sigma) || is_inf(sigma) || sigma &lt; 0) {\n    reject(\"my_normal_rng: sigma must be finite and non-negative; \",\n           \"found sigma = \", sigma);\n  }\n\n\n\nOften truncated uniform variates are needed, as in survival analysis when a time of death is censored beyond the end of the observations. To generate a truncated random variate, the cumulative distribution is used to find the truncation point in the inverse CDF, a uniform variate is generated in range, and then the inverse CDF translates it back.\n\n\nFor example, the following code generates a \\(\\textsf{Weibull}(\\alpha, \\sigma)\\) variate truncated below at a time \\(t\\),5\nreal weibull_lb_rng(real alpha, real sigma, real t) {\n  real p = weibull_cdf(t | alpha, sigma);   // cdf for lb\n  real u = uniform_rng(p, 1);               // unif in bounds\n  real y = sigma * (-log1m(u))^inv(alpha);  // inverse cdf\n  return y;\n}\n\n\n\nIf there is a lower bound and upper bound, then the CDF trick is used twice to find a lower and upper bound. For example, to generate a \\(\\textsf{normal}(\\mu, \\sigma)\\) truncated to a region \\((a, b)\\), the following code suffices,\nreal normal_lub_rng(real mu, real sigma, real lb, real ub) {\n  real p_lb = normal_cdf(lb | mu, sigma);\n  real p_ub = normal_cdf(ub | mu, sigma);\n  real u = uniform_rng(p_lb, p_ub);\n  real y = mu + sigma * inv_Phi(u);\n  return y;\n}\nTo make this more robust, all variables should be tested for finiteness, sigma should be tested for positiveness, and lb and ub should be tested to ensure the upper bound is greater than the lower bound. While it may be tempting to compress lines, the variable names serve as a kind of chunking of operations and naming for readability; compare the multiple statement version above with the single statement\n  return mu + sigma * inv_Phi(uniform_rng(normal_cdf(lb | mu, sigma),\n                                          normal_cdf(ub | mu, sigma)));\nfor readability. The names like p indicate probabilities, and p_lb and p_ub indicate the probabilities of the bounds. The variable u is clearly named as a uniform variate, and y is used to denote the variate being generated itself.",
-    "crumbs": [
-      "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
-    ]
-  },
-  {
-    "objectID": "stan-users-guide/user-functions.html#footnotes",
-    "href": "stan-users-guide/user-functions.html#footnotes",
-    "title": "User-Defined Functions",
-    "section": "Footnotes",
-    "text": "Footnotes\n\n\nThe main problem with comments is that they can be misleading, either due to misunderstandings on the programmer’s part or because the program’s behavior is modified after the comment is written. The program always behaves the way the code is written, which is why refactoring complex code into understandable units is preferable to simply adding comments.↩︎\nJust because this makes it possible to code a rejection sampler does not make it a good idea. Rejections break differentiability and the smooth exploration of the posterior. In Hamiltonian Monte Carlo, it can cause the sampler to be reduced to a diffusive random walk.↩︎\nA range of built-in validation routines is coming to Stan soon! Alternatively, the reject statement can be used to check constraints on the simplex.↩︎\nAs of Stan 2.9.0, the only way a user-defined producer will raise an exception is if a function it calls (including distribution statements) raises an exception via the reject statement.↩︎\nThe original code and impetus for including this in the manual came from the Stan forums post; by user lcomm, who also explained truncation above and below.↩︎",
+    "text": "The stanc3 compiler can optimize the code of Stan model during compilation. The optimized model code behaves the same as unoptimized code, but it may be faster, more memory efficient, or more numerically stable.\nThis section introduces the available optimization options and describes their effect.\nTo print out a representation of the optimized Stan program, use the stanc3 command-line flag --debug-optimized-mir-pretty. To print an analogous representation of the Stan program prior to optimization, use the flag --debug-transformed-mir-pretty.\n\n\nTo turn optimizations on, the user specifies the desired optimization level. The level specifies the set of optimizations to use. The chosen optimizations are used in a specific order, with some of them applied repeatedly.\nOptimization levels are specified by the numbers 0 and 1 and the ‘experimental’ tag:\n\nO0 No optimizations are applied.\nO1 Optimizations that are simple, do not dramatically change the program, and are unlikely to noticeably slow down compile times are applied.\nOexperimental All optimizations are applied. Some of these are not thorougly tested and may not always improve a programs performance.\n\nO0 is the default setting.\nThe levels include these optimizations:\n\nO0 includes no optimizations.\nO1 includes:\n\nDead code elimination\nCopy propagation\nConstant propagation\nPartial evaluation\nFunction inlining\nMatrix memory layout optimization\n\nOexperimental includes optimizations specified by O1 and also:\n\nAutomatic-differentiation level optimization\nOne step loop unrolling\nExpression propagation\nLazy code motion\nStatic loop unrolling\n\n\nIn addition, Oexperimental will apply more repetitions of the optimizations, which may increase compile times.\n\n\n\n\n\nDead code is code that does not affect the behavior of the program. Code is not dead if it affects target, the value of any outside-observable variable like transformed parameters or generated quantities, or side effects such as print statements. Removing dead code can speed up a program by avoiding unnecessary computations.\nExample Stan program:\nmodel {\n  int i;\n  i = 5;\n  for (j in 1:10);\n  if (0) {\n    print(\"Dead code\");\n  } else {\n    print(\"Hi!\");\n  }\n}\nCompiler representation of program before dead code elimination (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  int i = 5;\n  for(j in 1:10) {\n    ;\n  }\n  if(0) {\n    FnPrint__(\"Dead code\");\n  } else {\n    FnPrint__(\"Hi!\");\n  }\n}\nCompiler representation of program after dead code elimination (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i;\n  FnPrint__(\"Hi!\");\n}\n\n\n\nConstant propagation replaces uses of a variable which is known to have a constant value C with that constant C. This removes the overhead of looking up the variable, and also makes many other optimizations possible (such as static loop unrolling and partial evaluation).\nExample Stan program:\ntransformed data {\n  int n = 100;\n  int a[n];\n  for (i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before constant propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, n] a;\n  for(i in 1:n) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after constant propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = 100;\n  data array[int, 100] a;\n  for(i in 1:100) {\n    a[i] = i;\n  }\n}\n\n\n\nCopy propagation is similar to expression propagation, but only propagates variables rather than arbitrary expressions. This can reduce the complexity of the code for other optimizations such as expression propagation.\nExample Stan program:\nmodel {\n  int i = 1;\n  int j = i;\n  int k = i + j;\n}\nCompiler representation of program before copy propagation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n    int i = 1;\n    int j = i;\n    int k = (i + j);\n}\nCompiler representation of program after copy propagation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  int i = 1;\n  int j = i;\n  int k = (i + i);\n}\n\n\n\nPartial evaluation searches for expressions that we can replace with a faster, simpler, more memory efficient, or more numerically stable expression with the same meaning.\nExample Stan program:\nmodel {\n  real a = 1 + 1;\n  real b = log(1 - a);\n  real c = a + b * 5;\n}\nCompiler representation of program before partial evaluation (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real a = (1 + 1);\n  real b = log((1 - a));\n  real c = (a + (b * 5));\n}\nCompiler representation of program after partial evaluation (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  real a = 2;\n  real b = log1m(a);\n  real c = fma(b, 5, a);\n}\n\n\n\nFunction inlining replaces each function call to each user-defined function f with the body of f. It does this by copying the function body to the call site and doing appropriately renaming the argument variables. This optimization can speed up a program by avoiding the overhead of a function call and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return x + y;\n  }\n}\ntransformed data {\n  int a = 2;\n  int b = incr(a);\n}\nCompiler representation of program before function inlining (simplified from the output of --debug-transformed-mir-pretty):\nfunctions {\n  int incr(int x) {\n    int y = 1;\n    return (x + y);\n  }\n}\n\nprepare_data {\n  data int a = 2;\n  data int b = incr(a);\n}\nCompiler representation of program after function inlining (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int a;\n  a = 2;\n  data int b;\n  data int inline_sym1__;\n  data int inline_sym3__;\n  inline_sym3__ = 0;\n  for(inline_sym4__ in 1:1) {\n    int inline_sym2__;\n    inline_sym2__ = 1;\n    inline_sym3__ = 1;\n    inline_sym1__ = (a + inline_sym2__);\n    break;\n  }\n  b = inline_sym1__;\n}\nIn this code, the for loop and break is used to simulate the behavior of a return statement. The value to be returned is held in inline_sym1__. The flag variable inline_sym3__ indicates whether a return has occurred and is necessary to handle return statements nested inside loops within the function body.\n\n\n\nMatrices and vector variables which require automatic-differentiation (AD) in Stan can be represented in two different forms.\nThe first (and default) representation is the “Array of Structs” (AoS) or “Matrix of vars” (matvar) layout. A “var” is the term used in the Stan implementation of autodiff for a single real. It is represented as a structure containing it’s value and its adjoint. The AoS representation constructs matrices and vectors by simply using those structures as the elements of the matrix internally. This is flexible and very general, but many operations want to deal with the values or the adjoints as blocks, requiring expensive memory access patterns.\nThe second representation is the “Struct of Arrays” (SoA) or “Var of matrices” (varmat) layout. Rather than a matrix containing tiny structures of one value and one adjoint each, this representation uses a single structure which contains separately a matrix of values and a matrix of adjoints. Some operations, like iterating over elements or assigning to specific indices, become more expensive, but many matrix operations like multiplications become much faster in this representation.\nMore general reading on AoS vs SoA can be found on Wikipedia\nThis optimization pass attempts to identify which matrix or vector variables in the Stan program are candidates for using the SoA representation. The conditions change over time, but broadly speaking:\n\nAny Stan Math Library functions the matrix is passed to must be able to support it.\nThe matrix should not be accessed/assigned elementwise in a loop.\n\nThe debug flag --debug-mem-patterns will list each variable and whether it is using the AoS representation or the SoA representation.\n\n\n\n\n\n\nStan variables can have two auto-differentiation (AD) levels: AD or non-AD. AD variables carry gradient information with them, which allows Stan to calculate the log-density gradient, but they also have more overhead than non-AD variables. It is therefore inefficient for a variable to be AD unnecessarily. AD-level optimization sets every variable to be a floating point type unless its gradient is necessary.\nExample Stan program:\ndata {\n  real y;\n}\nmodel {\n  real x = y + 1;\n}\nCompiler representation of program before AD-level optimization (simplified from the output of --debug-transformed-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  real x = (y + 1);\n}\nCompiler representation of program after AD-level optimization (simplified from the output of --debug-optimized-mir-pretty):\ninput_vars {\n  real y;\n}\n\nlog_prob {\n  data real x = (y + 1);\n}\n\n\n\nOne step loop unrolling is similar to static loop unrolling. However, this optimization only ‘unrolls’ the first loop iteration, and can therefore work even when the total number of iterations is not predictable. This can speed up a program by providing more opportunities for further optimizations such as partial evaluation and lazy code motion.\nExample Stan program:\ndata {\n  int n;\n}\ntransformed data {\n  int x = 0;\n  for (i in 1:n) {\n    x += i;\n  }\n}\nCompiler representation of program before one step static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  data int x = 0;\n  for(i in 1:n) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after one step static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int n = FnReadData__(\"n\")[1];\n  int x = 0;\n  if((n &gt;= 1)) {\n    x = (x + 1);\n    for(i in (1 + 1):n) {\n      x = (x + i);\n    }\n  }\n}\n\n\n\nConstant propagation replaces the uses of a variable which is known to have a constant value E with that constant E. This often results in recalculating the expression, but provides more opportunities for further optimizations such as partial evaluation. Expression propagation is always followed by lazy code motion to avoid unnecessarily recomputing expressions.\nExample Stan program:\ndata {\n  int m;\n}\ntransformed data {\n  int n = m+1;\n  int a[n];\n  for (i in 1:n-1) {\n    a[i] = i;\n  }\n}\nCompiler representation of program before expression propagation (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, n] a;\n  for(i in 1:(n - 1)) {\n    a[i] = i;\n  }\n}\nCompiler representation of program after expression propagation (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int m = FnReadData__(\"m\")[1];\n  data int n = (m + 1);\n  data array[int, (m + 1)] a;\n  for(i in 1:((m + 1) - 1)) {\n    a[i] = i;\n  }\n}\n\n\n\nLazy code motion rearranges the statements and expressions in a program with the goals of:\n\nAvoiding computing expressions more than once, and\nComputing expressions as late as possible (to minimize the strain on the working memory set).\n\nTo accomplish these goals, lazy code motion will perform optimizations such as:\n\nMoving a repeatedly calculated expression to its own variable (also referred to as common-subexpression elimination)\nMoving an expression outside of a loop if it does not need to be in the loop (also referred to as loop-invariant code motion)\n\nLazy code motion can make some programs significantly more efficient by avoiding redundant or early computations.\nAs currently implemented in the compiler, it may move items between blocks in a way that actually increases overall computation. Improving this is an ongoing project.\nExample Stan program:\nmodel {\n  real x;\n  real y;\n  real z;\n\n  for (i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program before lazy code motion (simplified from the output of --debug-transformed-mir-pretty):\nlog_prob {\n  real x;\n  real y;\n  real z;\n  for(i in 1:10) {\n    x = sqrt(10);\n    y = sqrt(i);\n  }\n  z = sqrt(10);\n}\nCompiler representation of program after lazy code motion (simplified from the output of --debug-optimized-mir-pretty):\nlog_prob {\n  data real lcm_sym4__;\n  data real lcm_sym3__;\n  real x;\n  real y;\n  lcm_sym4__ = sqrt(10);\n  real z;\n  for(i in 1:10) {\n    x = lcm_sym4__;\n    y = sqrt(i);\n  }\n  z = lcm_sym4__;\n}\n\n\n\nStatic loop unrolling takes a loop with a predictable number of iterations X and replaces it by writing out the loop body X times. The loop index in each repeat is replaced with the appropriate constant. This optimization can speed up a program by avoiding the overhead of a loop and providing more opportunities for further optimizations (such as partial evaluation).\nExample Stan program:\ntransformed data {\n  int x = 0;\n  for (i in 1:4) {\n    x += i;\n  }\n}\nCompiler representation of program before static loop unrolling (simplified from the output of --debug-transformed-mir-pretty):\nprepare_data {\n  data int x = 0;\n  for(i in 1:4) {\n    x = (x + i);\n  }\n}\nCompiler representation of program after static loop unrolling (simplified from the output of --debug-optimized-mir-pretty):\nprepare_data {\n  data int x;\n  x = 0;\n  x = (x + 1);\n  x = (x + 2);\n  x = (x + 3);\n  x = (x + 4);\n}",
     "crumbs": [
       "Stan Users Guide",
-      "Programming Techniques",
-      "User-Defined Functions"
+      "Appendices",
+      "Using the Stan Compiler"
     ]
   }
 ]
\ No newline at end of file
diff --git a/docs/stan-users-guide/algebraic-equations.html b/docs/stan-users-guide/algebraic-equations.html
index 85d905234..171d090df 100644
--- a/docs/stan-users-guide/algebraic-equations.html
+++ b/docs/stan-users-guide/algebraic-equations.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1296,8 +1308,35 @@ <h3 class="unnumbered anchored" data-anchor-id="maximum-number-of-steps">Maximum
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/algebraic-equations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/algebraic-equations.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/bootstrap.html b/docs/stan-users-guide/bootstrap.html
index 1140f65ac..041801f23 100644
--- a/docs/stan-users-guide/bootstrap.html
+++ b/docs/stan-users-guide/bootstrap.html
@@ -84,7 +84,7 @@
 <script src="../site_libs/quarto-search/fuse.min.js"></script>
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
-<link href="../stan-users-guide/using-stanc.html" rel="next">
+<link href="../stan-users-guide/multiple-imputation.html" rel="next">
 <link href="../stan-users-guide/decision-analysis.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link active">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1311,14 +1323,41 @@ <h2 class="anchored" data-anchor-id="bayesian-bootstrap-and-bagging">Bayesian bo
       </a>          
   </div>
   <div class="nav-page nav-page-next">
-      <a href="../stan-users-guide/using-stanc.html" class="pagination-link" aria-label="Using the Stan Compiler">
-        <span class="nav-page-text">Using the Stan Compiler</span> <i class="bi bi-arrow-right-short"></i>
+      <a href="../stan-users-guide/multiple-imputation.html" class="pagination-link" aria-label="Multiple imputation">
+        <span class="nav-page-text">Multiple imputation</span> <i class="bi bi-arrow-right-short"></i>
       </a>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/bootstrap.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/bootstrap.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/clustering.html b/docs/stan-users-guide/clustering.html
index bb3a43bfb..c69c45ec7 100644
--- a/docs/stan-users-guide/clustering.html
+++ b/docs/stan-users-guide/clustering.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1609,8 +1621,35 @@ <h4 class="unnumbered anchored" data-anchor-id="full-bayes-correlated-topic-mode
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/clustering.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/clustering.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/complex-numbers.html b/docs/stan-users-guide/complex-numbers.html
index 5ab464f1a..09e110b52 100644
--- a/docs/stan-users-guide/complex-numbers.html
+++ b/docs/stan-users-guide/complex-numbers.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1304,8 +1316,35 @@ <h3 class="unnumbered anchored" data-anchor-id="dependent-complex-error">Depende
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/complex-numbers.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/complex-numbers.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/copulas.html b/docs/stan-users-guide/copulas.html
index bbd72e525..b90c1957d 100644
--- a/docs/stan-users-guide/copulas.html
+++ b/docs/stan-users-guide/copulas.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1395,8 +1407,35 @@ <h2 class="anchored" data-anchor-id="further-readingviewing">Further reading/vie
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/copulas.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/copulas.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/cross-validation.html b/docs/stan-users-guide/cross-validation.html
index 78393efbe..c2c63070e 100644
--- a/docs/stan-users-guide/cross-validation.html
+++ b/docs/stan-users-guide/cross-validation.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1474,8 +1486,35 @@ <h3 class="unnumbered anchored" data-anchor-id="approximate-cross-validation">Ap
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/cross-validation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/cross-validation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/custom-probability.html b/docs/stan-users-guide/custom-probability.html
index 8e724a256..dcd0d5da8 100644
--- a/docs/stan-users-guide/custom-probability.html
+++ b/docs/stan-users-guide/custom-probability.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1280,8 +1292,35 @@ <h3 class="unnumbered anchored" data-anchor-id="bivariate-normal-cumulative-dist
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/custom-probability.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/custom-probability.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/dae.html b/docs/stan-users-guide/dae.html
index 665b22ee7..6c2b59173 100644
--- a/docs/stan-users-guide/dae.html
+++ b/docs/stan-users-guide/dae.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1312,8 +1324,35 @@ <h3 class="unnumbered anchored" data-anchor-id="maximum-number-of-steps">Maximum
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/dae.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/dae.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/decision-analysis.html b/docs/stan-users-guide/decision-analysis.html
index 918b21040..dd3672d7c 100644
--- a/docs/stan-users-guide/decision-analysis.html
+++ b/docs/stan-users-guide/decision-analysis.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1326,8 +1338,35 @@ <h2 class="anchored" data-anchor-id="continuous-choices">Continuous choices</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/decision-analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/decision-analysis.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/efficiency-tuning.html b/docs/stan-users-guide/efficiency-tuning.html
index df3e37a7e..f89182f93 100644
--- a/docs/stan-users-guide/efficiency-tuning.html
+++ b/docs/stan-users-guide/efficiency-tuning.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -2082,8 +2094,35 @@ <h2 class="anchored" data-anchor-id="using-map-reduce">Using map-reduce</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/efficiency-tuning.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/efficiency-tuning.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/finite-mixtures.html b/docs/stan-users-guide/finite-mixtures.html
index ea76dd5bb..b3a240090 100644
--- a/docs/stan-users-guide/finite-mixtures.html
+++ b/docs/stan-users-guide/finite-mixtures.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1613,8 +1625,35 @@ <h3 class="unnumbered anchored" data-anchor-id="comparison-to-model-averaging">C
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/finite-mixtures.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/finite-mixtures.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/floating-point.html b/docs/stan-users-guide/floating-point.html
index 4261cedd6..ceb78bfdf 100644
--- a/docs/stan-users-guide/floating-point.html
+++ b/docs/stan-users-guide/floating-point.html
@@ -31,7 +31,7 @@
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
 <link href="../stan-users-guide/matrices-arrays.html" rel="next">
-<link href="../stan-users-guide/survival.html" rel="prev">
+<link href="../stan-users-guide/wiener_diffusion_model.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
 <script src="../site_libs/quarto-html/popper.min.js"></script>
@@ -255,7 +255,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -373,6 +373,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -465,7 +471,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -497,6 +503,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1330,8 +1342,8 @@ <h2 class="anchored" data-anchor-id="comparing-floating-point-numbers">Comparing
 </script>
 <nav class="page-navigation">
   <div class="nav-page nav-page-previous">
-      <a href="../stan-users-guide/survival.html" class="pagination-link" aria-label="Survival Models">
-        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Survival Models</span>
+      <a href="../stan-users-guide/wiener_diffusion_model.html" class="pagination-link" aria-label="Wiener diffusion model">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Wiener diffusion model</span>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
@@ -1341,8 +1353,35 @@ <h2 class="anchored" data-anchor-id="comparing-floating-point-numbers">Comparing
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/floating-point.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/floating-point.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/for-bugs-users.html b/docs/stan-users-guide/for-bugs-users.html
index ff54338ba..52a424609 100644
--- a/docs/stan-users-guide/for-bugs-users.html
+++ b/docs/stan-users-guide/for-bugs-users.html
@@ -308,7 +308,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -426,6 +426,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -518,7 +524,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -550,6 +556,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1404,8 +1416,35 @@ <h2 class="anchored" data-anchor-id="the-stan-community">The Stan community</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/for-bugs-users.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/for-bugs-users.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/gaussian-processes.html b/docs/stan-users-guide/gaussian-processes.html
index 7b794b927..4f53b356e 100644
--- a/docs/stan-users-guide/gaussian-processes.html
+++ b/docs/stan-users-guide/gaussian-processes.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -625,8 +637,7 @@ <h1>Gaussian Processes</h1>
 <p>Gaussian processes are continuous stochastic processes and thus may be interpreted as providing a probability distribution over functions. A probability distribution over continuous functions may be viewed, roughly, as an uncountably infinite collection of random variables, one for each valid input. The generality of the supported functions makes Gaussian priors popular choices for priors in general multivariate (non-linear) regression problems.</p>
 <p>The defining feature of a Gaussian process is that the joint distribution of the function’s value at a finite number of input points is a multivariate normal distribution. This makes it tractable to both fit models from finite amounts of observed data and make predictions for finitely many new data points.</p>
 <p>Unlike a simple multivariate normal distribution, which is parameterized by a mean vector and covariance matrix, a Gaussian process is parameterized by a mean function and covariance function. The mean and covariance functions apply to vectors of inputs and return a mean vector and covariance matrix which provide the mean and covariance of the outputs corresponding to those input points in the functions drawn from the process.</p>
-<p>Gaussian processes can be encoded in Stan by implementing their mean and covariance functions or by using the specialized covariance functions outlined below, and plugging the result into the Gaussian model.<br>
-This form of model is straightforward and may be used for simulation, model fitting, or posterior predictive inference. A more efficient Stan implementation for the GP with a normally distributed outcome marginalizes over the latent Gaussian process, and applies a Cholesky-factor reparameterization of the Gaussian to compute the likelihood and the posterior predictive distribution analytically.</p>
+<p>Gaussian processes can be encoded in Stan by implementing their mean and covariance functions or by using the specialized covariance functions outlined below, and plugging the result into the Gaussian model. This form of model is straightforward and may be used for simulation, model fitting, or posterior predictive inference. A more efficient Stan implementation for the GP with a normally distributed outcome marginalizes over the latent Gaussian process, and applies a Cholesky-factor reparameterization of the Gaussian to compute the likelihood and the posterior predictive distribution analytically.</p>
 <p>After defining Gaussian processes, this chapter covers the basic implementations for simulation, hyperparameter estimation, and posterior predictive inference for univariate regressions, multivariate regressions, and multivariate logistic regressions. Gaussian processes are general, and by necessity this chapter only touches on some basic models. For more information, see <span class="citation" data-cites="RasmussenWilliams:2006">Rasmussen and Williams (<a href="#ref-RasmussenWilliams:2006" role="doc-biblioref">2006</a>)</span>.</p>
 <p>Note that fitting Gaussian processes as described below using exact inference by computing Cholesky of the covariance matrix scales cubicly with the size of data. Due to how Stan autodiff is implemented, Stan is also slower than Gaussian process specialized software. It is likely that Gaussian processes using exact inference by computing Cholesky of the covariance matrix with <span class="math inline">\(N&gt;1000\)</span> are too slow for practical purposes in Stan. There are many approximations to speed-up Gaussian process computation, from which the basis function approaches for 1-3 dimensional <span class="math inline">\(x\)</span> are easiest to implement in Stan (see, e.g., <span class="citation" data-cites="Riutort-Mayol:2023:HSGP">Riutort-Mayol et al. (<a href="#ref-Riutort-Mayol:2023:HSGP" role="doc-biblioref">2023</a>)</span>).</p>
 <section id="gaussian-process-regression" class="level2">
@@ -866,17 +877,128 @@ <h4 class="unnumbered anchored" data-anchor-id="poisson-gp">Poisson GP</h4>
 <span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
 <span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>}</span>
 <span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
-<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>  a ~ std_normal();</span>
-<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
-<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>  y ~ poisson_log(a + f);</span>
-<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb7-22"><a href="#cb7-22" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb7-23"><a href="#cb7-23" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-24"><a href="#cb7-24" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb7-25"><a href="#cb7-25" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb7-26"><a href="#cb7-26" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb7-27"><a href="#cb7-27" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb7-28"><a href="#cb7-28" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb7-29"><a href="#cb7-29" aria-hidden="true" tabindex="-1"></a>  a ~ std_normal();</span>
+<span id="cb7-30"><a href="#cb7-30" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb7-31"><a href="#cb7-31" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-32"><a href="#cb7-32" aria-hidden="true" tabindex="-1"></a>  y ~ poisson_log(a + f);</span>
+<span id="cb7-33"><a href="#cb7-33" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="poisson-gp-using-an-embedded-laplace-approximation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="poisson-gp-using-an-embedded-laplace-approximation">Poisson GP using an embedded Laplace approximation</h4>
+<p>For computational reasons, we may want to integrate out the latent variable <span class="math inline">\(f\)</span>, as was done in the normal output model. Unfortunately, exact marginalization over <span class="math inline">\(f\)</span> is not possible when the outcome model is not normal. Instead, we may perform <em>approximate</em> marginalization with an <em>embedded Laplace approximation</em> <span class="citation" data-cites="RasmussenWilliams:2006 Rue:2009 Margossian:2020">(<a href="#ref-RasmussenWilliams:2006" role="doc-biblioref">Rasmussen and Williams 2006</a>; <a href="#ref-Rue:2009" role="doc-biblioref">Rue, Martino, and Chopin 2009</a>; <a href="#ref-Margossian:2020" role="doc-biblioref">Margossian et al. 2020</a>)</span>. To do so, we first use the function <code>laplace_marginal</code> to approximate the marginal likelihood <span class="math inline">\(p(y \mid \rho, \alpha, a)\)</span> and sample the hyperparameters with Hamiltonian Monte Carlo sampling. Then, we recover the integrated out <span class="math inline">\(f\)</span> in the <code>generated quantities</code> block using <code>laplace_latent_rng</code>.</p>
+<p>The embedded Laplace approximation computes a Gaussian approximation of the conditional posterior, <span class="math display">\[
+  \hat p_\mathcal{L}(f \mid \rho, \alpha, a, y) \approx p(f \mid \rho, \alpha, a, y),
+\]</span> where <span class="math inline">\(\hat p_\mathcal{L}\)</span> is a Gaussian that matches the mode and curvature of <span class="math inline">\(p(f \mid \rho, \alpha, a, y)\)</span>. We then obtain an approximation of the marginal likelihood as follows: <span class="math display">\[
+  \hat p_\mathcal{L}(y \mid \rho, \alpha, a)
+    = \frac{p(f^* \mid \alpha, \rho) p(y \mid f^*, a)}{
+    \hat p_\mathcal{L}(f^* \mid \rho, \alpha, a, y)},
+\]</span> where <span class="math inline">\(f^*\)</span> is the mode of <span class="math inline">\(p(f \mid \rho, \alpha, a, y)\)</span>, obtained via numerical optimization.</p>
+<p>To use Stan’s embedded Laplace approximation, we must define the prior covariance function and the log likelihood function in the functions block.</p>
+<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// log likelihood function</span></span>
+<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ll_function(<span class="dt">vector</span> f, <span class="dt">real</span> a, <span class="dt">array</span>[] <span class="dt">int</span> y) {</span>
+<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">return</span> poisson_log_lpmf(y | a + f);</span>
+<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// covariance function</span></span>
+<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span> cov_function(<span class="dt">real</span> rho, <span class="dt">real</span> alpha, <span class="dt">array</span>[] <span class="dt">real</span> x, <span class="dt">int</span> N, <span class="dt">real</span> delta) {</span>
+<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> add_diag(K, delta)</span>
+<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The embedded Laplace relies on calculations of the log likelihood’s Hessian, <span class="math inline">\(\partial^2 \log p(y \mid f, a, \rho, \alpha) / \partial f^2\)</span>, and these calculations can be much faster when the Hessian is sparse. In particular, it is expected that the Hessian is block diagonal. In the <code>transformed data</code> block we can specify the block size of the Hessian.</p>
+<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> hessian_block_size = <span class="dv">1</span>;</span>
+<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For example, if <span class="math inline">\(y_i\)</span> depends only on <span class="math inline">\(f_i\)</span>, then the Hessian of the log likelihood is diagonal and the block size is 1. On the other hand, if the Hessian is not sparse, then we set the hessian block size to <span class="math inline">\(N\)</span>, where <span class="math inline">\(N\)</span> is the dimension of <span class="math inline">\(f\)</span>. Currently, Stan does not check the block size of the Hessian and so the user is responsible for correctly specifying the block size.</p>
+<p>Finally, we increment <code>target</code> in the model block with the approximation to <span class="math inline">\(\log p(y \mid \rho, \alpha, a)\)</span>.</p>
+<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>                             cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Notice that we do not need to construct <span class="math inline">\(f\)</span> explicitly, since it is marginalized out. Instead, we can recover the latent variables in <code>generated quantities</code>:</p>
+<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Users can set the control parameters of the embedded Laplace approximation, via <code>laplace_marginal_tol</code> and <code>laplace_latent_tol_rng</code>. When using these functions, the user must set <em>all</em> the control options and store them in a tuple. These control parameters mostly concern the numerical optimizer used to find the mode <span class="math inline">\(f^*\)</span> of <span class="math inline">\(p(f \mid \rho, \alpha, a)\)</span>.</p>
+<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">tuple</span>(<span class="dt">vector</span>[N], <span class="dt">real</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>) laplace_ops;</span>
+<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.1</span> = rep_vector(<span class="dv">0</span>, N);  <span class="co">// starting point for Laplace optimizer</span></span>
+<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.2</span> = <span class="fl">1.49e-8</span>;           <span class="co">// tolerance for optimizer</span></span>
+<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.3</span> = <span class="dv">500</span>;               <span class="co">// maximum number of steps for optimizer.</span></span>
+<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.4</span> = <span class="dv">1</span>;              <span class="co">// solver type being used.</span></span>
+<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.5</span> = <span class="dv">1000</span>;           <span class="co">// max number of steps for linesearch.</span></span>
+<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.6</span> = <span class="dv">1</span>;              <span class="co">// allow_fallback (1: TRUE, 0: FALSE)</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>If users want to depart from the defaults for only some of the control parameters, a tuple with the default values (as above) can be created with the helper callable <code>generate_laplace_options()</code>, and the specific control parameter can then be modified,</p>
+<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">tuple</span>(<span class="dt">vector</span>[N], <span class="dt">real</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>, <span class="dt">int</span>) laplace_ops =</span>
+<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>    generate_laplace_options(N);</span>
+<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>  laplace_ops<span class="fl">.2</span> = <span class="fl">1e-6</span>; <span class="co">// make tolerance of the optimizer less strict.</span></span>
+<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The tuple <code>laplace_ops</code> is then passed to <code>laplace_marginal_tol</code> and <code>laplace_rng_tol</code>.</p>
+<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal_tol(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>                                 cov_function, (rho, alpha, x, N, delta),</span>
+<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>                                 laplace_ops);</span>
+<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_rng(ll_function, (a, y), hessian_block_size,</span>
+<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta),</span>
+<span id="cb14-12"><a href="#cb14-12" aria-hidden="true" tabindex="-1"></a>                                   laplace_ops);</span>
+<span id="cb14-13"><a href="#cb14-13" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Stan also provides support for a limited menu of built-in functions, including the Poisson distribution with a log link and and prior mean <span class="math inline">\(m\)</span>. When using such a built-in function, the user does not need to specify a likelihood in the <code>functions</code> block. However, the user must strictly follow the signature of the likelihood: in this case, <span class="math inline">\(m\)</span> must be a vector of length <span class="math inline">\(N\)</span> (to allow for different offsets for each observation <span class="math inline">\(y_i\)</span>) and we must indicate which element of <span class="math inline">\(f\)</span> each component of <span class="math inline">\(y\)</span> matches using the variable <span class="math inline">\(y_\text{index}\)</span>. In our example, there is a simple pairing <span class="math inline">\((y_i, f_i)\)</span>, however we could imagine a scenario where multiple observations <span class="math inline">\((y_{j1}, y_{j2}, ...)\)</span> are observed for a single <span class="math inline">\(f_j\)</span>.</p>
+<div class="sourceCode" id="cb15"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[n_obs] <span class="dt">int</span> y_index;</span>
+<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:n_obs) y_index[i] = i - <span class="dv">1</span>;</span>
+<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-7"><a href="#cb15-7" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb15-8"><a href="#cb15-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-9"><a href="#cb15-9" aria-hidden="true" tabindex="-1"></a>transformed parameter {</span>
+<span id="cb15-10"><a href="#cb15-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] m = rep_vector(a, N);</span>
+<span id="cb15-11"><a href="#cb15-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-12"><a href="#cb15-12" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-13"><a href="#cb15-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb15-14"><a href="#cb15-14" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb15-15"><a href="#cb15-15" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal_poisson_log_lpmf(y | y_index, m,</span>
+<span id="cb15-16"><a href="#cb15-16" aria-hidden="true" tabindex="-1"></a>                                       cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb15-17"><a href="#cb15-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb15-18"><a href="#cb15-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb15-19"><a href="#cb15-19" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb15-20"><a href="#cb15-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_poisson_log_rng(y, y_index, m,</span>
+<span id="cb15-21"><a href="#cb15-21" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb15-22"><a href="#cb15-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>As before, we could specify the control parameters for the embedded Laplace approximation using <code>laplace_marginal_tol_poisson_log_lpmf</code> and <code>laplace_latent_tol_poisson_log_nrg</code>.</p>
+<p>Marginalization with a Laplace approximation can lead to faster inference, however it also introduces an approximation error. In practice, this error is negligible when using a Poisson likelihood and the approximation works well for log concave likelihoods <span class="citation" data-cites="Kuss:2005 Vanhatalo:2010 Cseke:2011 Vehtari:2016">(<a href="#ref-Kuss:2005" role="doc-biblioref">Kuss and Rasmussen 2005</a>; <a href="#ref-Vanhatalo:2010" role="doc-biblioref">Vanhatalo, Pietiläinen, and Vehtari 2010</a>; <a href="#ref-Cseke:2011" role="doc-biblioref">Cseke and Heskes 2011</a>; <a href="#ref-Vehtari:2016" role="doc-biblioref">Vehtari et al. 2016</a>)</span>. Still, users should exercise caution, especially when trying unconventional likelihoods.</p>
 </section>
-<section id="logistic-gaussian-process-regression" class="level4 unnumbered">
-<h4 class="unnumbered anchored" data-anchor-id="logistic-gaussian-process-regression">Logistic Gaussian process regression</h4>
+<section id="logistic-gp-regression" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="logistic-gp-regression">Logistic GP regression</h4>
 <p>For binary classification problems, the observed outputs <span class="math inline">\(z_n \in
 \{ 0,1 \}\)</span> are binary. These outputs are modeled using a Gaussian process with (unobserved) outputs <span class="math inline">\(y_n\)</span> through the logistic link, <span class="math display">\[
 z_n \sim \textsf{Bernoulli}(\operatorname{logit}^{-1}(y_n)),
@@ -884,16 +1006,51 @@ <h4 class="unnumbered anchored" data-anchor-id="logistic-gaussian-process-regres
 \Pr[z_n = 1] = \operatorname{logit}^{-1}(y_n).
 \]</span></p>
 <p>We can extend our latent variable GP Stan program to deal with classification problems. Below <code>a</code> is the bias term, which can help account for imbalanced classes in the training data:</p>
-<div class="sourceCode" id="cb8"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z;</span>
-<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
-<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(a + f);</span>
-<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb16"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z;</span>
+<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>  z ~ bernoulli_logit(a + f);</span>
+<span id="cb16-10"><a href="#cb16-10" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+<section id="logistic-gp-regression-with-an-embedded-laplace-approximation" class="level4 unnumbered">
+<h4 class="unnumbered anchored" data-anchor-id="logistic-gp-regression-with-an-embedded-laplace-approximation">Logistic GP regression with an embedded Laplace approximation</h4>
+<p>As with the Poisson GP, we cannot marginalize the latent variables exactly, however we can resort to an embedded Laplace approximation.</p>
+<div class="sourceCode" id="cb17"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a>  <span class="co">// log likelihood function</span></span>
+<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> ll_function(<span class="dt">vector</span> f, <span class="dt">real</span> a, <span class="dt">array</span>[] <span class="dt">int</span> z) {</span>
+<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">return</span> bernoulli_logit_lpmf(z | a + f);</span>
+<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// covariance function</span></span>
+<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span> cov_function(<span class="dt">real</span> rho, <span class="dt">real</span> alpha, <span class="dt">array</span>[] <span class="dt">real</span> x, <span class="dt">int</span> N, <span class="dt">real</span> delta) {</span>
+<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb17-10"><a href="#cb17-10" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> add_diag(K, delta)</span>
+<span id="cb17-11"><a href="#cb17-11" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb17-12"><a href="#cb17-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-13"><a href="#cb17-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-14"><a href="#cb17-14" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb17-15"><a href="#cb17-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-16"><a href="#cb17-16" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb17-17"><a href="#cb17-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> hessian_block_size = <span class="dv">1</span>;</span>
+<span id="cb17-18"><a href="#cb17-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-19"><a href="#cb17-19" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-20"><a href="#cb17-20" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb17-21"><a href="#cb17-21" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-22"><a href="#cb17-22" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb17-23"><a href="#cb17-23" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> laplace_marginal(ll_function, (a, z), hessian_block_size,</span>
+<span id="cb17-24"><a href="#cb17-24" aria-hidden="true" tabindex="-1"></a>                             cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb17-25"><a href="#cb17-25" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb17-26"><a href="#cb17-26" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb17-27"><a href="#cb17-27" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb17-28"><a href="#cb17-28" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f = laplace_latent_rng(ll_function, (a, z), hessian_block_size,</span>
+<span id="cb17-29"><a href="#cb17-29" aria-hidden="true" tabindex="-1"></a>                                   cov_function, (rho, alpha, x, N, delta));</span>
+<span id="cb17-30"><a href="#cb17-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>While marginalization with a Laplace approximation can lead to faster inference, it also introduces an approximation error. In practice, this error may not be negligible with a Bernoulli likelihood; for more discussion see, e.g. <span class="citation" data-cites="Vehtari:2016 Margossian:2020">(<a href="#ref-Vehtari:2016" role="doc-biblioref">Vehtari et al. 2016</a>; <a href="#ref-Margossian:2020" role="doc-biblioref">Margossian et al. 2020</a>)</span>.</p>
 </section>
 </section>
 <section id="automatic-relevance-determination" class="level3 unnumbered">
@@ -907,40 +1064,40 @@ <h3 class="unnumbered anchored" data-anchor-id="automatic-relevance-determinatio
 \]</span> The estimation of <span class="math inline">\(\rho\)</span> was termed “automatic relevance determination” by <span class="citation" data-cites="Neal:1996">Neal (<a href="#ref-Neal:1996" role="doc-biblioref">1996</a>)</span>, but this is misleading, because the magnitude of the scale of the posterior for each <span class="math inline">\(\rho_d\)</span> is dependent on the scaling of the input data along dimension <span class="math inline">\(d\)</span>. Moreover, the scale of the parameters <span class="math inline">\(\rho_d\)</span> measures non-linearity along the <span class="math inline">\(d\)</span>-th dimension, rather than “relevance” <span class="citation" data-cites="PiironenVehtari:2016">(<a href="#ref-PiironenVehtari:2016" role="doc-biblioref">Piironen and Vehtari 2016</a>)</span>.</p>
 <p>A priori, the closer <span class="math inline">\(\rho_d\)</span> is to zero, the more nonlinear the conditional mean in dimension <span class="math inline">\(d\)</span> is. A posteriori, the actual dependencies between <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span> play a role. With one covariate <span class="math inline">\(x_1\)</span> having a linear effect and another covariate <span class="math inline">\(x_2\)</span> having a nonlinear effect, it is possible that <span class="math inline">\(\rho_1 &gt; \rho_2\)</span> even if the predictive relevance of <span class="math inline">\(x_1\)</span> is higher <span class="citation" data-cites="RasmussenWilliams:2006">(<a href="#ref-RasmussenWilliams:2006" role="doc-biblioref">Rasmussen and Williams 2006, 80</a>)</span>. The collection of <span class="math inline">\(\rho_d\)</span> (or <span class="math inline">\(1/\rho_d\)</span>) parameters can also be modeled hierarchically.</p>
 <p>The implementation of automatic relevance determination is a straightforward extension of the one-dimensional case by modifying <code>rho</code> to be an array.</p>
-<div class="sourceCode" id="cb9"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
-<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
-<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] x;</span>
-<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
-<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
-<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
-<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
-<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
-<span id="cb9-13"><a href="#cb9-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb9-14"><a href="#cb9-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
-<span id="cb9-15"><a href="#cb9-15" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb9-16"><a href="#cb9-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb9-17"><a href="#cb9-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
-<span id="cb9-18"><a href="#cb9-18" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb9-19"><a href="#cb9-19" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
-<span id="cb9-20"><a href="#cb9-20" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
-<span id="cb9-21"><a href="#cb9-21" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb9-22"><a href="#cb9-22" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
-<span id="cb9-23"><a href="#cb9-23" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb9-24"><a href="#cb9-24" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
-<span id="cb9-25"><a href="#cb9-25" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
-<span id="cb9-26"><a href="#cb9-26" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb9-27"><a href="#cb9-27" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb9-28"><a href="#cb9-28" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
-<span id="cb9-29"><a href="#cb9-29" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb9-30"><a href="#cb9-30" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
-<span id="cb9-31"><a href="#cb9-31" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
-<span id="cb9-32"><a href="#cb9-32" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb9-33"><a href="#cb9-33" aria-hidden="true" tabindex="-1"></a>  y ~ normal(f, sigma);</span>
-<span id="cb9-34"><a href="#cb9-34" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb18"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] x;</span>
+<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb18-12"><a href="#cb18-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb18-13"><a href="#cb18-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb18-14"><a href="#cb18-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb18-15"><a href="#cb18-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb18-16"><a href="#cb18-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb18-17"><a href="#cb18-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb18-18"><a href="#cb18-18" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb18-19"><a href="#cb18-19" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb18-20"><a href="#cb18-20" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb18-21"><a href="#cb18-21" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb18-22"><a href="#cb18-22" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb18-23"><a href="#cb18-23" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb18-24"><a href="#cb18-24" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb18-25"><a href="#cb18-25" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb18-26"><a href="#cb18-26" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb18-27"><a href="#cb18-27" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb18-28"><a href="#cb18-28" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb18-29"><a href="#cb18-29" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb18-30"><a href="#cb18-30" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb18-31"><a href="#cb18-31" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb18-32"><a href="#cb18-32" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb18-33"><a href="#cb18-33" aria-hidden="true" tabindex="-1"></a>  y ~ normal(f, sigma);</span>
+<span id="cb18-34"><a href="#cb18-34" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="priors-gp.section" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="priors-gp.section">Priors for Gaussian process parameters</h3>
@@ -972,18 +1129,18 @@ <h4 class="unnumbered anchored" data-anchor-id="priors-for-length-scale">Priors
     p \in \mathbb{Z}
 \end{align*}\]</span></p>
 <p>which has an inverse gamma left tail if <span class="math inline">\(p \leq 0\)</span> and an inverse Gaussian right tail. This has not yet been implemented in Stan’s math library, but it is possible to implement as a user defined function:</p>
-<div class="sourceCode" id="cb10"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
-<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> generalized_inverse_gaussian_lpdf(<span class="dt">real</span> x, <span class="dt">int</span> p,</span>
-<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>                                        <span class="dt">real</span> a, <span class="dt">real</span> b) {</span>
-<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> p * <span class="fl">0.5</span> * log(a / b)</span>
-<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a>      - log(<span class="dv">2</span> * modified_bessel_second_kind(p, sqrt(a * b)))</span>
-<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>      + (p - <span class="dv">1</span>) * log(x)</span>
-<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a>      - (a * x + b / x) * <span class="fl">0.5</span>;</span>
-<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a> }</span>
-<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb10-12"><a href="#cb10-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb19"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> generalized_inverse_gaussian_lpdf(<span class="dt">real</span> x, <span class="dt">int</span> p,</span>
+<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>                                        <span class="dt">real</span> a, <span class="dt">real</span> b) {</span>
+<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> p * <span class="fl">0.5</span> * log(a / b)</span>
+<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a>      - log(<span class="dv">2</span> * modified_bessel_second_kind(p, sqrt(a * b)))</span>
+<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>      + (p - <span class="dv">1</span>) * log(x)</span>
+<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a>      - (a * x + b / x) * <span class="fl">0.5</span>;</span>
+<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a> }</span>
+<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>If we have high-frequency covariates in our fixed effects, we may wish to further regularize the GP away from high-frequency functions, which means we’ll need to penalize smaller length-scales. Luckily, we have a useful way of thinking about how length-scale affects the frequency of the functions supported by the GP. If we were to repeatedly draw from a zero-mean GP with a length-scale of <span class="math inline">\(\rho\)</span> in a fixed-domain <span class="math inline">\([0,T]\)</span>, we would get a distribution for the number of times each draw of the GP crossed the zero axis. The expectation of this random variable, the number of zero crossings, is <span class="math inline">\(T / \pi
 \rho\)</span>. You can see that as <span class="math inline">\(\rho\)</span> decreases, the expectation of the number of upcrossings increases as the GP is representing higher-frequency functions. Thus, this is a good statistic to keep in mind when setting a lower-bound for our prior on length-scale in the presence of high-frequency covariates. However, this statistic is only valid for one-dimensional inputs.</p>
 </section>
@@ -1004,59 +1161,59 @@ <h3 class="unnumbered anchored" data-anchor-id="predictive-inference-with-a-gaus
 p\left(\tilde{y}, y \mid \tilde{x},x\right).
 \]</span></p>
 <p>A direct implementation in Stan defines a model in terms of the joint distribution of the observed <span class="math inline">\(y\)</span> and unobserved <span class="math inline">\(\tilde{y}\)</span>.</p>
-<div class="sourceCode" id="cb11"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
-<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
-<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] y1;</span>
-<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
-<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
-<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
-<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
-<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N = N1 + N2;</span>
-<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
-<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
-<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a>    x[n1] = x1[n1];</span>
-<span id="cb11-14"><a href="#cb11-14" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb11-15"><a href="#cb11-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
-<span id="cb11-16"><a href="#cb11-16" aria-hidden="true" tabindex="-1"></a>    x[N1 + n2] = x2[n2];</span>
-<span id="cb11-17"><a href="#cb11-17" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb11-18"><a href="#cb11-18" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb11-19"><a href="#cb11-19" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb11-20"><a href="#cb11-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
-<span id="cb11-21"><a href="#cb11-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
-<span id="cb11-22"><a href="#cb11-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb11-23"><a href="#cb11-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
-<span id="cb11-24"><a href="#cb11-24" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb11-25"><a href="#cb11-25" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
-<span id="cb11-26"><a href="#cb11-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
-<span id="cb11-27"><a href="#cb11-27" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb11-28"><a href="#cb11-28" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
-<span id="cb11-29"><a href="#cb11-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
-<span id="cb11-30"><a href="#cb11-30" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb11-31"><a href="#cb11-31" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
-<span id="cb11-32"><a href="#cb11-32" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb11-33"><a href="#cb11-33" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
-<span id="cb11-34"><a href="#cb11-34" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb11-35"><a href="#cb11-35" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb11-36"><a href="#cb11-36" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
-<span id="cb11-37"><a href="#cb11-37" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
-<span id="cb11-38"><a href="#cb11-38" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb11-39"><a href="#cb11-39" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb11-40"><a href="#cb11-40" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb11-41"><a href="#cb11-41" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
-<span id="cb11-42"><a href="#cb11-42" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb11-43"><a href="#cb11-43" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
-<span id="cb11-44"><a href="#cb11-44" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
-<span id="cb11-45"><a href="#cb11-45" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb11-46"><a href="#cb11-46" aria-hidden="true" tabindex="-1"></a>  y1 ~ normal(f[<span class="dv">1</span>:N1], sigma);</span>
-<span id="cb11-47"><a href="#cb11-47" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb11-48"><a href="#cb11-48" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb11-49"><a href="#cb11-49" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] y2;</span>
-<span id="cb11-50"><a href="#cb11-50" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
-<span id="cb11-51"><a href="#cb11-51" aria-hidden="true" tabindex="-1"></a>    y2[n2] = normal_rng(f[N1 + n2], sigma);</span>
-<span id="cb11-52"><a href="#cb11-52" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb11-53"><a href="#cb11-53" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb20"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
+<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
+<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] y1;</span>
+<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
+<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
+<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb20-9"><a href="#cb20-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb20-10"><a href="#cb20-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N = N1 + N2;</span>
+<span id="cb20-11"><a href="#cb20-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb20-12"><a href="#cb20-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb20-13"><a href="#cb20-13" aria-hidden="true" tabindex="-1"></a>    x[n1] = x1[n1];</span>
+<span id="cb20-14"><a href="#cb20-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-15"><a href="#cb20-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb20-16"><a href="#cb20-16" aria-hidden="true" tabindex="-1"></a>    x[N1 + n2] = x2[n2];</span>
+<span id="cb20-17"><a href="#cb20-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-18"><a href="#cb20-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-19"><a href="#cb20-19" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb20-20"><a href="#cb20-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb20-21"><a href="#cb20-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb20-22"><a href="#cb20-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb20-23"><a href="#cb20-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb20-24"><a href="#cb20-24" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-25"><a href="#cb20-25" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb20-26"><a href="#cb20-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb20-27"><a href="#cb20-27" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb20-28"><a href="#cb20-28" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb20-29"><a href="#cb20-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb20-30"><a href="#cb20-30" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-31"><a href="#cb20-31" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb20-32"><a href="#cb20-32" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb20-33"><a href="#cb20-33" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb20-34"><a href="#cb20-34" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb20-35"><a href="#cb20-35" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-36"><a href="#cb20-36" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb20-37"><a href="#cb20-37" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb20-38"><a href="#cb20-38" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-39"><a href="#cb20-39" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-40"><a href="#cb20-40" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb20-41"><a href="#cb20-41" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb20-42"><a href="#cb20-42" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb20-43"><a href="#cb20-43" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb20-44"><a href="#cb20-44" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb20-45"><a href="#cb20-45" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb20-46"><a href="#cb20-46" aria-hidden="true" tabindex="-1"></a>  y1 ~ normal(f[<span class="dv">1</span>:N1], sigma);</span>
+<span id="cb20-47"><a href="#cb20-47" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb20-48"><a href="#cb20-48" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb20-49"><a href="#cb20-49" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] y2;</span>
+<span id="cb20-50"><a href="#cb20-50" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb20-51"><a href="#cb20-51" aria-hidden="true" tabindex="-1"></a>    y2[n2] = normal_rng(f[N1 + n2], sigma);</span>
+<span id="cb20-52"><a href="#cb20-52" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb20-53"><a href="#cb20-53" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The input vectors <code>x1</code> and <code>x2</code> are declared as data, as is the observed output vector <code>y1</code>. The unknown output vector <code>y2</code>, which corresponds to input vector <code>x2</code>, is declared in the generated quantities block and will be sampled when the model is executed.</p>
 <p>A transformed data block is used to combine the input vectors <code>x1</code> and <code>x2</code> into a single vector <code>x</code>.</p>
 <p>The model block declares and defines a local variable for the combined output vector <code>f</code>, which consists of the concatenation of the conditional mean for known outputs <code>y1</code> and unknown outputs <code>y2</code>. Thus the combined output vector <code>f</code> is aligned with the combined input vector <code>x</code>. All that is left is to define the univariate normal distribution statement for <code>y</code>.</p>
@@ -1065,59 +1222,59 @@ <h3 class="unnumbered anchored" data-anchor-id="predictive-inference-with-a-gaus
 <h4 class="unnumbered anchored" data-anchor-id="predictive-inference-in-non-gaussian-gps">Predictive inference in non-Gaussian GPs</h4>
 <p>We can do predictive inference in non-Gaussian GPs in much the same way as we do with Gaussian GPs.</p>
 <p>Consider the following full model for prediction using logistic Gaussian process regression.</p>
-<div class="sourceCode" id="cb12"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
-<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
-<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z1;</span>
-<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
-<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
-<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
-<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
-<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N = N1 + N2;</span>
-<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
-<span id="cb12-12"><a href="#cb12-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
-<span id="cb12-13"><a href="#cb12-13" aria-hidden="true" tabindex="-1"></a>    x[n1] = x1[n1];</span>
-<span id="cb12-14"><a href="#cb12-14" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb12-15"><a href="#cb12-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
-<span id="cb12-16"><a href="#cb12-16" aria-hidden="true" tabindex="-1"></a>    x[N1 + n2] = x2[n2];</span>
-<span id="cb12-17"><a href="#cb12-17" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb12-18"><a href="#cb12-18" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb12-19"><a href="#cb12-19" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb12-20"><a href="#cb12-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
-<span id="cb12-21"><a href="#cb12-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
-<span id="cb12-22"><a href="#cb12-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
-<span id="cb12-23"><a href="#cb12-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
-<span id="cb12-24"><a href="#cb12-24" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb12-25"><a href="#cb12-25" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
-<span id="cb12-26"><a href="#cb12-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
-<span id="cb12-27"><a href="#cb12-27" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb12-28"><a href="#cb12-28" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
-<span id="cb12-29"><a href="#cb12-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
-<span id="cb12-30"><a href="#cb12-30" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb12-31"><a href="#cb12-31" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
-<span id="cb12-32"><a href="#cb12-32" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb12-33"><a href="#cb12-33" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
-<span id="cb12-34"><a href="#cb12-34" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb12-35"><a href="#cb12-35" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb12-36"><a href="#cb12-36" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
-<span id="cb12-37"><a href="#cb12-37" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
-<span id="cb12-38"><a href="#cb12-38" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb12-39"><a href="#cb12-39" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb12-40"><a href="#cb12-40" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb12-41"><a href="#cb12-41" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
-<span id="cb12-42"><a href="#cb12-42" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb12-43"><a href="#cb12-43" aria-hidden="true" tabindex="-1"></a>  a ~ std_normal();</span>
-<span id="cb12-44"><a href="#cb12-44" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
-<span id="cb12-45"><a href="#cb12-45" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb12-46"><a href="#cb12-46" aria-hidden="true" tabindex="-1"></a>  z1 ~ bernoulli_logit(a + f[<span class="dv">1</span>:N1]);</span>
-<span id="cb12-47"><a href="#cb12-47" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb12-48"><a href="#cb12-48" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb12-49"><a href="#cb12-49" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">int</span> z2;</span>
-<span id="cb12-50"><a href="#cb12-50" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
-<span id="cb12-51"><a href="#cb12-51" aria-hidden="true" tabindex="-1"></a>    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);</span>
-<span id="cb12-52"><a href="#cb12-52" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb12-53"><a href="#cb12-53" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb21"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
+<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
+<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z1;</span>
+<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
+<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
+<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N = N1 + N2;</span>
+<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a>    x[n1] = x1[n1];</span>
+<span id="cb21-14"><a href="#cb21-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-15"><a href="#cb21-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb21-16"><a href="#cb21-16" aria-hidden="true" tabindex="-1"></a>    x[N1 + n2] = x2[n2];</span>
+<span id="cb21-17"><a href="#cb21-17" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-18"><a href="#cb21-18" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-19"><a href="#cb21-19" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb21-20"><a href="#cb21-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb21-21"><a href="#cb21-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb21-22"><a href="#cb21-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> a;</span>
+<span id="cb21-23"><a href="#cb21-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] eta;</span>
+<span id="cb21-24"><a href="#cb21-24" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-25"><a href="#cb21-25" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb21-26"><a href="#cb21-26" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] f;</span>
+<span id="cb21-27"><a href="#cb21-27" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb21-28"><a href="#cb21-28" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb21-29"><a href="#cb21-29" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, alpha, rho);</span>
+<span id="cb21-30"><a href="#cb21-30" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-31"><a href="#cb21-31" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb21-32"><a href="#cb21-32" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb21-33"><a href="#cb21-33" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb21-34"><a href="#cb21-34" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb21-35"><a href="#cb21-35" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-36"><a href="#cb21-36" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb21-37"><a href="#cb21-37" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta;</span>
+<span id="cb21-38"><a href="#cb21-38" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-39"><a href="#cb21-39" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-40"><a href="#cb21-40" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb21-41"><a href="#cb21-41" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb21-42"><a href="#cb21-42" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb21-43"><a href="#cb21-43" aria-hidden="true" tabindex="-1"></a>  a ~ std_normal();</span>
+<span id="cb21-44"><a href="#cb21-44" aria-hidden="true" tabindex="-1"></a>  eta ~ std_normal();</span>
+<span id="cb21-45"><a href="#cb21-45" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb21-46"><a href="#cb21-46" aria-hidden="true" tabindex="-1"></a>  z1 ~ bernoulli_logit(a + f[<span class="dv">1</span>:N1]);</span>
+<span id="cb21-47"><a href="#cb21-47" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb21-48"><a href="#cb21-48" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb21-49"><a href="#cb21-49" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">int</span> z2;</span>
+<span id="cb21-50"><a href="#cb21-50" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb21-51"><a href="#cb21-51" aria-hidden="true" tabindex="-1"></a>    z2[n2] = bernoulli_logit_rng(a + f[N1 + n2]);</span>
+<span id="cb21-52"><a href="#cb21-52" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb21-53"><a href="#cb21-53" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="analytical-form-of-joint-predictive-inference" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="analytical-form-of-joint-predictive-inference">Analytical form of joint predictive inference</h4>
@@ -1134,89 +1291,89 @@ <h4 class="unnumbered anchored" data-anchor-id="analytical-form-of-joint-predict
 \]</span></p>
 <p>There is no noise term including <span class="math inline">\(\sigma^2\)</span> because the indexes of elements in <span class="math inline">\(x\)</span> and <span class="math inline">\(\tilde{x}\)</span> are never the same.</p>
 <p>This Stan code below uses the analytic form of the posterior and provides sampling of the resulting multivariate normal through the Cholesky decomposition. The data declaration is the same as for the latent variable example, but we’ve defined a function called <code>gp_pred_rng</code> which will generate a draw from the posterior predictive mean conditioned on observed data <code>y1</code>. The code uses a Cholesky decomposition in triangular solves in order to cut down on the number of matrix-matrix multiplications when computing the conditional mean and the conditional covariance of <span class="math inline">\(p(\tilde{y})\)</span>.</p>
-<div class="sourceCode" id="cb13"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
-<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> gp_pred_rng(<span class="dt">array</span>[] <span class="dt">real</span> x2,</span>
-<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">vector</span> y1,</span>
-<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">array</span>[] <span class="dt">real</span> x1,</span>
-<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> alpha,</span>
-<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> rho,</span>
-<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> sigma,</span>
-<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> delta) {</span>
-<span id="cb13-9"><a href="#cb13-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N1 = rows(y1);</span>
-<span id="cb13-10"><a href="#cb13-10" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N2 = size(x2);</span>
-<span id="cb13-11"><a href="#cb13-11" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[N2] f2;</span>
-<span id="cb13-12"><a href="#cb13-12" aria-hidden="true" tabindex="-1"></a>    {</span>
-<span id="cb13-13"><a href="#cb13-13" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N1] L_K;</span>
-<span id="cb13-14"><a href="#cb13-14" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[N1] K_div_y1;</span>
-<span id="cb13-15"><a href="#cb13-15" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N2] k_x1_x2;</span>
-<span id="cb13-16"><a href="#cb13-16" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N2] v_pred;</span>
-<span id="cb13-17"><a href="#cb13-17" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[N2] f2_mu;</span>
-<span id="cb13-18"><a href="#cb13-18" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N2, N2] cov_f2;</span>
-<span id="cb13-19"><a href="#cb13-19" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N2, N2] diag_delta;</span>
-<span id="cb13-20"><a href="#cb13-20" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N1] K;</span>
-<span id="cb13-21"><a href="#cb13-21" aria-hidden="true" tabindex="-1"></a>      K = gp_exp_quad_cov(x1, alpha, rho);</span>
-<span id="cb13-22"><a href="#cb13-22" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
-<span id="cb13-23"><a href="#cb13-23" aria-hidden="true" tabindex="-1"></a>        K[n, n] = K[n, n] + square(sigma);</span>
-<span id="cb13-24"><a href="#cb13-24" aria-hidden="true" tabindex="-1"></a>      }</span>
-<span id="cb13-25"><a href="#cb13-25" aria-hidden="true" tabindex="-1"></a>      L_K = cholesky_decompose(K);</span>
-<span id="cb13-26"><a href="#cb13-26" aria-hidden="true" tabindex="-1"></a>      K_div_y1 = mdivide_left_tri_low(L_K, y1);</span>
-<span id="cb13-27"><a href="#cb13-27" aria-hidden="true" tabindex="-1"></a>      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';</span>
-<span id="cb13-28"><a href="#cb13-28" aria-hidden="true" tabindex="-1"></a>      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);</span>
-<span id="cb13-29"><a href="#cb13-29" aria-hidden="true" tabindex="-1"></a>      f2_mu = (k_x1_x2' * K_div_y1);</span>
-<span id="cb13-30"><a href="#cb13-30" aria-hidden="true" tabindex="-1"></a>      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);</span>
-<span id="cb13-31"><a href="#cb13-31" aria-hidden="true" tabindex="-1"></a>      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;</span>
-<span id="cb13-32"><a href="#cb13-32" aria-hidden="true" tabindex="-1"></a>      diag_delta = diag_matrix(rep_vector(delta, N2));</span>
-<span id="cb13-33"><a href="#cb13-33" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb13-34"><a href="#cb13-34" aria-hidden="true" tabindex="-1"></a>      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);</span>
-<span id="cb13-35"><a href="#cb13-35" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb13-36"><a href="#cb13-36" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> f2;</span>
-<span id="cb13-37"><a href="#cb13-37" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb13-38"><a href="#cb13-38" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb13-39"><a href="#cb13-39" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb13-40"><a href="#cb13-40" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
-<span id="cb13-41"><a href="#cb13-41" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
-<span id="cb13-42"><a href="#cb13-42" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] y1;</span>
-<span id="cb13-43"><a href="#cb13-43" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
-<span id="cb13-44"><a href="#cb13-44" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
-<span id="cb13-45"><a href="#cb13-45" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb13-46"><a href="#cb13-46" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
-<span id="cb13-47"><a href="#cb13-47" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] mu = rep_vector(<span class="dv">0</span>, N1);</span>
-<span id="cb13-48"><a href="#cb13-48" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
-<span id="cb13-49"><a href="#cb13-49" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb13-50"><a href="#cb13-50" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb13-51"><a href="#cb13-51" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
-<span id="cb13-52"><a href="#cb13-52" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
-<span id="cb13-53"><a href="#cb13-53" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb13-54"><a href="#cb13-54" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb13-55"><a href="#cb13-55" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb13-56"><a href="#cb13-56" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N1, N1] L_K;</span>
-<span id="cb13-57"><a href="#cb13-57" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb13-58"><a href="#cb13-58" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);</span>
-<span id="cb13-59"><a href="#cb13-59" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> sq_sigma = square(sigma);</span>
-<span id="cb13-60"><a href="#cb13-60" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb13-61"><a href="#cb13-61" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
-<span id="cb13-62"><a href="#cb13-62" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
-<span id="cb13-63"><a href="#cb13-63" aria-hidden="true" tabindex="-1"></a>      K[n1, n1] = K[n1, n1] + sq_sigma;</span>
-<span id="cb13-64"><a href="#cb13-64" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb13-65"><a href="#cb13-65" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb13-66"><a href="#cb13-66" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
-<span id="cb13-67"><a href="#cb13-67" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb13-68"><a href="#cb13-68" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb13-69"><a href="#cb13-69" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
-<span id="cb13-70"><a href="#cb13-70" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb13-71"><a href="#cb13-71" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
-<span id="cb13-72"><a href="#cb13-72" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb13-73"><a href="#cb13-73" aria-hidden="true" tabindex="-1"></a>  y1 ~ multi_normal_cholesky(mu, L_K);</span>
-<span id="cb13-74"><a href="#cb13-74" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb13-75"><a href="#cb13-75" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb13-76"><a href="#cb13-76" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] f2;</span>
-<span id="cb13-77"><a href="#cb13-77" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] y2;</span>
-<span id="cb13-78"><a href="#cb13-78" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb13-79"><a href="#cb13-79" aria-hidden="true" tabindex="-1"></a>  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);</span>
-<span id="cb13-80"><a href="#cb13-80" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
-<span id="cb13-81"><a href="#cb13-81" aria-hidden="true" tabindex="-1"></a>    y2[n2] = normal_rng(f2[n2], sigma);</span>
-<span id="cb13-82"><a href="#cb13-82" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb13-83"><a href="#cb13-83" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb22"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span> gp_pred_rng(<span class="dt">array</span>[] <span class="dt">real</span> x2,</span>
+<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">vector</span> y1,</span>
+<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">array</span>[] <span class="dt">real</span> x1,</span>
+<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> alpha,</span>
+<span id="cb22-6"><a href="#cb22-6" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> rho,</span>
+<span id="cb22-7"><a href="#cb22-7" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> sigma,</span>
+<span id="cb22-8"><a href="#cb22-8" aria-hidden="true" tabindex="-1"></a>                     <span class="dt">real</span> delta) {</span>
+<span id="cb22-9"><a href="#cb22-9" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N1 = rows(y1);</span>
+<span id="cb22-10"><a href="#cb22-10" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> N2 = size(x2);</span>
+<span id="cb22-11"><a href="#cb22-11" aria-hidden="true" tabindex="-1"></a>    <span class="dt">vector</span>[N2] f2;</span>
+<span id="cb22-12"><a href="#cb22-12" aria-hidden="true" tabindex="-1"></a>    {</span>
+<span id="cb22-13"><a href="#cb22-13" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N1] L_K;</span>
+<span id="cb22-14"><a href="#cb22-14" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[N1] K_div_y1;</span>
+<span id="cb22-15"><a href="#cb22-15" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N2] k_x1_x2;</span>
+<span id="cb22-16"><a href="#cb22-16" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N2] v_pred;</span>
+<span id="cb22-17"><a href="#cb22-17" aria-hidden="true" tabindex="-1"></a>      <span class="dt">vector</span>[N2] f2_mu;</span>
+<span id="cb22-18"><a href="#cb22-18" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N2, N2] cov_f2;</span>
+<span id="cb22-19"><a href="#cb22-19" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N2, N2] diag_delta;</span>
+<span id="cb22-20"><a href="#cb22-20" aria-hidden="true" tabindex="-1"></a>      <span class="dt">matrix</span>[N1, N1] K;</span>
+<span id="cb22-21"><a href="#cb22-21" aria-hidden="true" tabindex="-1"></a>      K = gp_exp_quad_cov(x1, alpha, rho);</span>
+<span id="cb22-22"><a href="#cb22-22" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb22-23"><a href="#cb22-23" aria-hidden="true" tabindex="-1"></a>        K[n, n] = K[n, n] + square(sigma);</span>
+<span id="cb22-24"><a href="#cb22-24" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb22-25"><a href="#cb22-25" aria-hidden="true" tabindex="-1"></a>      L_K = cholesky_decompose(K);</span>
+<span id="cb22-26"><a href="#cb22-26" aria-hidden="true" tabindex="-1"></a>      K_div_y1 = mdivide_left_tri_low(L_K, y1);</span>
+<span id="cb22-27"><a href="#cb22-27" aria-hidden="true" tabindex="-1"></a>      K_div_y1 = mdivide_right_tri_low(K_div_y1', L_K)';</span>
+<span id="cb22-28"><a href="#cb22-28" aria-hidden="true" tabindex="-1"></a>      k_x1_x2 = gp_exp_quad_cov(x1, x2, alpha, rho);</span>
+<span id="cb22-29"><a href="#cb22-29" aria-hidden="true" tabindex="-1"></a>      f2_mu = (k_x1_x2' * K_div_y1);</span>
+<span id="cb22-30"><a href="#cb22-30" aria-hidden="true" tabindex="-1"></a>      v_pred = mdivide_left_tri_low(L_K, k_x1_x2);</span>
+<span id="cb22-31"><a href="#cb22-31" aria-hidden="true" tabindex="-1"></a>      cov_f2 = gp_exp_quad_cov(x2, alpha, rho) - v_pred' * v_pred;</span>
+<span id="cb22-32"><a href="#cb22-32" aria-hidden="true" tabindex="-1"></a>      diag_delta = diag_matrix(rep_vector(delta, N2));</span>
+<span id="cb22-33"><a href="#cb22-33" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-34"><a href="#cb22-34" aria-hidden="true" tabindex="-1"></a>      f2 = multi_normal_rng(f2_mu, cov_f2 + diag_delta);</span>
+<span id="cb22-35"><a href="#cb22-35" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb22-36"><a href="#cb22-36" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> f2;</span>
+<span id="cb22-37"><a href="#cb22-37" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb22-38"><a href="#cb22-38" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-39"><a href="#cb22-39" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb22-40"><a href="#cb22-40" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N1;</span>
+<span id="cb22-41"><a href="#cb22-41" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N1] <span class="dt">real</span> x1;</span>
+<span id="cb22-42"><a href="#cb22-42" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] y1;</span>
+<span id="cb22-43"><a href="#cb22-43" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N2;</span>
+<span id="cb22-44"><a href="#cb22-44" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N2] <span class="dt">real</span> x2;</span>
+<span id="cb22-45"><a href="#cb22-45" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-46"><a href="#cb22-46" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb22-47"><a href="#cb22-47" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N1] mu = rep_vector(<span class="dv">0</span>, N1);</span>
+<span id="cb22-48"><a href="#cb22-48" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb22-49"><a href="#cb22-49" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-50"><a href="#cb22-50" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb22-51"><a href="#cb22-51" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb22-52"><a href="#cb22-52" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; alpha;</span>
+<span id="cb22-53"><a href="#cb22-53" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb22-54"><a href="#cb22-54" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-55"><a href="#cb22-55" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb22-56"><a href="#cb22-56" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N1, N1] L_K;</span>
+<span id="cb22-57"><a href="#cb22-57" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb22-58"><a href="#cb22-58" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N1, N1] K = gp_exp_quad_cov(x1, alpha, rho);</span>
+<span id="cb22-59"><a href="#cb22-59" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> sq_sigma = square(sigma);</span>
+<span id="cb22-60"><a href="#cb22-60" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-61"><a href="#cb22-61" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb22-62"><a href="#cb22-62" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n1 <span class="cf">in</span> <span class="dv">1</span>:N1) {</span>
+<span id="cb22-63"><a href="#cb22-63" aria-hidden="true" tabindex="-1"></a>      K[n1, n1] = K[n1, n1] + sq_sigma;</span>
+<span id="cb22-64"><a href="#cb22-64" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb22-65"><a href="#cb22-65" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-66"><a href="#cb22-66" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb22-67"><a href="#cb22-67" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb22-68"><a href="#cb22-68" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-69"><a href="#cb22-69" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb22-70"><a href="#cb22-70" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb22-71"><a href="#cb22-71" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb22-72"><a href="#cb22-72" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-73"><a href="#cb22-73" aria-hidden="true" tabindex="-1"></a>  y1 ~ multi_normal_cholesky(mu, L_K);</span>
+<span id="cb22-74"><a href="#cb22-74" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb22-75"><a href="#cb22-75" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb22-76"><a href="#cb22-76" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] f2;</span>
+<span id="cb22-77"><a href="#cb22-77" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N2] y2;</span>
+<span id="cb22-78"><a href="#cb22-78" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb22-79"><a href="#cb22-79" aria-hidden="true" tabindex="-1"></a>  f2 = gp_pred_rng(x2, y1, x1, alpha, rho, sigma, delta);</span>
+<span id="cb22-80"><a href="#cb22-80" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n2 <span class="cf">in</span> <span class="dv">1</span>:N2) {</span>
+<span id="cb22-81"><a href="#cb22-81" aria-hidden="true" tabindex="-1"></a>    y2[n2] = normal_rng(f2[n2], sigma);</span>
+<span id="cb22-82"><a href="#cb22-82" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb22-83"><a href="#cb22-83" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 </section>
 <section id="multiple-output-gaussian-processes" class="level3 unnumbered">
@@ -1252,50 +1409,50 @@ <h3 class="unnumbered anchored" data-anchor-id="multiple-output-gaussian-process
 L_{K(x \mid 1.0, \rho)} &amp;= \texttt{cholesky}\mathtt{\_}\texttt{decompose}\big(K(x \mid 1.0, \rho)\big)
 \end{align*}\]</span></p>
 <p>This can be implemented in Stan using a latent-variable GP formulation. We’ve used <span class="math inline">\(\textsf{LKJCorr}\)</span> for <span class="math inline">\(C(\phi)\)</span>, but any positive-definite matrix will do.</p>
-<div class="sourceCode" id="cb14"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb14-2"><a href="#cb14-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
-<span id="cb14-3"><a href="#cb14-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
-<span id="cb14-4"><a href="#cb14-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
-<span id="cb14-5"><a href="#cb14-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] y;</span>
-<span id="cb14-6"><a href="#cb14-6" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb14-7"><a href="#cb14-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
-<span id="cb14-8"><a href="#cb14-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
-<span id="cb14-9"><a href="#cb14-9" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb14-10"><a href="#cb14-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb14-11"><a href="#cb14-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
-<span id="cb14-12"><a href="#cb14-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[D] alpha;</span>
-<span id="cb14-13"><a href="#cb14-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb14-14"><a href="#cb14-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
-<span id="cb14-15"><a href="#cb14-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] eta;</span>
-<span id="cb14-16"><a href="#cb14-16" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb14-17"><a href="#cb14-17" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb14-18"><a href="#cb14-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] f;</span>
-<span id="cb14-19"><a href="#cb14-19" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb14-20"><a href="#cb14-20" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, <span class="fl">1.0</span>, rho);</span>
-<span id="cb14-21"><a href="#cb14-21" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
-<span id="cb14-22"><a href="#cb14-22" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb14-23"><a href="#cb14-23" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
-<span id="cb14-24"><a href="#cb14-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb14-25"><a href="#cb14-25" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
-<span id="cb14-26"><a href="#cb14-26" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb14-27"><a href="#cb14-27" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb14-28"><a href="#cb14-28" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
-<span id="cb14-29"><a href="#cb14-29" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta</span>
-<span id="cb14-30"><a href="#cb14-30" aria-hidden="true" tabindex="-1"></a>        * diag_pre_multiply(alpha, L_Omega)';</span>
-<span id="cb14-31"><a href="#cb14-31" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb14-32"><a href="#cb14-32" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb14-33"><a href="#cb14-33" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
-<span id="cb14-34"><a href="#cb14-34" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb14-35"><a href="#cb14-35" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
-<span id="cb14-36"><a href="#cb14-36" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">3</span>);</span>
-<span id="cb14-37"><a href="#cb14-37" aria-hidden="true" tabindex="-1"></a>  to_vector(eta) ~ std_normal();</span>
-<span id="cb14-38"><a href="#cb14-38" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb14-39"><a href="#cb14-39" aria-hidden="true" tabindex="-1"></a>  to_vector(y) ~ normal(to_vector(f), sigma);</span>
-<span id="cb14-40"><a href="#cb14-40" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb14-41"><a href="#cb14-41" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb14-42"><a href="#cb14-42" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, D] Omega;</span>
-<span id="cb14-43"><a href="#cb14-43" aria-hidden="true" tabindex="-1"></a>  Omega = L_Omega * L_Omega';</span>
-<span id="cb14-44"><a href="#cb14-44" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb23"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;</span>
+<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> x;</span>
+<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] y;</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta = <span class="fl">1e-9</span>;</span>
+<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; rho;</span>
+<span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[D] alpha;</span>
+<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
+<span id="cb23-15"><a href="#cb23-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] eta;</span>
+<span id="cb23-16"><a href="#cb23-16" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-17"><a href="#cb23-17" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb23-18"><a href="#cb23-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] f;</span>
+<span id="cb23-19"><a href="#cb23-19" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb23-20"><a href="#cb23-20" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] K = gp_exp_quad_cov(x, <span class="fl">1.0</span>, rho);</span>
+<span id="cb23-21"><a href="#cb23-21" aria-hidden="true" tabindex="-1"></a>    <span class="dt">matrix</span>[N, N] L_K;</span>
+<span id="cb23-22"><a href="#cb23-22" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-23"><a href="#cb23-23" aria-hidden="true" tabindex="-1"></a>    <span class="co">// diagonal elements</span></span>
+<span id="cb23-24"><a href="#cb23-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb23-25"><a href="#cb23-25" aria-hidden="true" tabindex="-1"></a>      K[n, n] = K[n, n] + delta;</span>
+<span id="cb23-26"><a href="#cb23-26" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb23-27"><a href="#cb23-27" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-28"><a href="#cb23-28" aria-hidden="true" tabindex="-1"></a>    L_K = cholesky_decompose(K);</span>
+<span id="cb23-29"><a href="#cb23-29" aria-hidden="true" tabindex="-1"></a>    f = L_K * eta</span>
+<span id="cb23-30"><a href="#cb23-30" aria-hidden="true" tabindex="-1"></a>        * diag_pre_multiply(alpha, L_Omega)';</span>
+<span id="cb23-31"><a href="#cb23-31" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb23-32"><a href="#cb23-32" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-33"><a href="#cb23-33" aria-hidden="true" tabindex="-1"></a>  rho ~ inv_gamma(<span class="dv">5</span>, <span class="dv">5</span>);</span>
+<span id="cb23-34"><a href="#cb23-34" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb23-35"><a href="#cb23-35" aria-hidden="true" tabindex="-1"></a>  sigma ~ std_normal();</span>
+<span id="cb23-36"><a href="#cb23-36" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">3</span>);</span>
+<span id="cb23-37"><a href="#cb23-37" aria-hidden="true" tabindex="-1"></a>  to_vector(eta) ~ std_normal();</span>
+<span id="cb23-38"><a href="#cb23-38" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb23-39"><a href="#cb23-39" aria-hidden="true" tabindex="-1"></a>  to_vector(y) ~ normal(to_vector(f), sigma);</span>
+<span id="cb23-40"><a href="#cb23-40" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb23-41"><a href="#cb23-41" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb23-42"><a href="#cb23-42" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, D] Omega;</span>
+<span id="cb23-43"><a href="#cb23-43" aria-hidden="true" tabindex="-1"></a>  Omega = L_Omega * L_Omega';</span>
+<span id="cb23-44"><a href="#cb23-44" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 
 
 
@@ -1305,6 +1462,15 @@ <h3 class="unnumbered anchored" data-anchor-id="multiple-output-gaussian-process
 
 
 <a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Cseke:2011" class="csl-entry" role="listitem">
+Cseke, Botond, and Tom Heskes. 2011. <span>“Approximate Marginals in Latent <span>Gaussian</span> Models.”</span> <em>Journal of Machine Learning Research</em> 12.
+</div>
+<div id="ref-Kuss:2005" class="csl-entry" role="listitem">
+Kuss, Malte, and Carl E Rasmussen. 2005. <span>“Assessing Approximate Inference for Binary <span>Gaussian</span> Process Classification.”</span> <em>Journal of Machine Learning Research</em> 6: 1679–1704.
+</div>
+<div id="ref-Margossian:2020" class="csl-entry" role="listitem">
+Margossian, Charles C, Aki Vehtari, Daniel Simpson, and Raj Agrawal. 2020. <span>“Hamiltonian Monte Carlo Using an Adjoint-Differentiated Laplace Approximation: Bayesian Inference for Latent Gaussian Models and Beyond.”</span> <em>Advances in Neural Information Processing Systems</em> 34.
+</div>
 <div id="ref-Neal:1996" class="csl-entry" role="listitem">
 Neal, Radford M. 1996. <em>Bayesian Learning for Neural Networks</em>. Lecture Notes in Statistics 118. New York: Springer.
 </div>
@@ -1320,6 +1486,15 @@ <h3 class="unnumbered anchored" data-anchor-id="multiple-output-gaussian-process
 <div id="ref-Riutort-Mayol:2023:HSGP" class="csl-entry" role="listitem">
 Riutort-Mayol, Gabriel, Paul-Christian Bürkner, Michael R Andersen, Arno Solin, and Aki Vehtari. 2023. <span>“Practical <span>Hilbert</span> Space Approximate <span>Bayesian</span> <span>Gaussian</span> Processes for Probabilistic Programming.”</span> <em>Statistics and Computing</em> 33 (1): 17.
 </div>
+<div id="ref-Rue:2009" class="csl-entry" role="listitem">
+Rue, Håvard, Sara Martino, and Nicolas Chopin. 2009. <span>“Approximate Bayesian Inference for Latent Gaussian Models by Using Integrated Nested Laplace Approximations.”</span> <em>Journal of the Royal Statistical Society: Series B (Statistical Methodology)</em> 71 (2): 319–92. <a href="https://doi.org/10.1111/j.1467-9868.2008.00700.x">https://doi.org/10.1111/j.1467-9868.2008.00700.x</a>.
+</div>
+<div id="ref-Vanhatalo:2010" class="csl-entry" role="listitem">
+Vanhatalo, Jarno, Ville Pietiläinen, and Aki Vehtari. 2010. <span>“Approximate Inference for Disease Mapping with Sparse <span>Gaussian</span> Processes.”</span> <em>Statistics in Medicine</em> 29 (15): 1580–1607.
+</div>
+<div id="ref-Vehtari:2016" class="csl-entry" role="listitem">
+Vehtari, Aki, Tommi Mononen, Ville Tolvanen, Tuomas Sivula, and Ole Winther. 2016. <span>“Bayesian Leave-One-Out Cross-Validation Approximations for <span>Gaussian</span> Latent Variable Models.”</span> <em>Journal of Machine Learning Research</em> 17 (103): 1–38. <a href="http://jmlr.org/papers/v17/14-540.html">http://jmlr.org/papers/v17/14-540.html</a>.
+</div>
 <div id="ref-zhang-gp:2004" class="csl-entry" role="listitem">
 Zhang, Hao. 2004. <span>“Inconsistent Estimation and Asymptotically Equal Interpolations in Model-Based Geostatistics.”</span> <em>Journal of the American Statistical Association</em> 99 (465): 250–61.
 </div>
@@ -1904,8 +2079,35 @@ <h3 class="unnumbered anchored" data-anchor-id="multiple-output-gaussian-process
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/gaussian-processes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/gaussian-processes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/hyperspherical-models.html b/docs/stan-users-guide/hyperspherical-models.html
index e0d154ccf..08c37a052 100644
--- a/docs/stan-users-guide/hyperspherical-models.html
+++ b/docs/stan-users-guide/hyperspherical-models.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1263,8 +1275,35 @@ <h2 class="anchored" data-anchor-id="circular-representations-of-days-and-years"
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/hyperspherical-models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/hyperspherical-models.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/img/Figure_DDM.pdf b/docs/stan-users-guide/img/Figure_DDM.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..381d4b538fde6f9c7ce7248a0d68a842f69082ce
GIT binary patch
literal 100560
zcmV)9K*hf$P((&8F)lR<CD7Ex*a|ivFd%PYY6?6&3NK7$ZfA68AT}{LIW!<3ARr(L
zFGgu>bY*fNFGg%(bY(<kV{c?-3O+sxb98cLVQmU{+N`}z(q%c0U^lN-_y)?2{~%V7
z<CX?mKo3fTE}4}k+P-i=fbWY(@8}+cS*Az$2|kjSmlp&`5GcR?<5&6huYW85i+@)8
zv;3;Rdig!;`qkU-wfe9B{-6K$U%#x(zkYeae=EPc?)mHg`vtcD{kI=~*56h`{Qu{_
z|M&mQLjUdg&0l^WJ^1(adyilLUiEirN)P@Q73c5#Xa2|E_PT;A_`mXh<#$zWUcjE7
z0kJ>-^`m5MzsuaZyxIT!@%Qn&&1zNuk6-`R*Y7Gzvq=8+M~PK_Hy!rxQRM#pN0~i-
zkJ@Bsnfv!2Wop^KXD`xjcP{u@idQ)Hy3KvS&r);!mbE@lmVf_QW|!Z4k2!9k3x1Sp
zW&SRu4arg;@Uv9Y-?f!n)&2X=GPD0~efLkH3x1Xo`Q4|@(JYjI|Jn0=Pr80<DP#QQ
z<*ARyl%H?!XTi&Ul)LZjXTcBrQS!dIpCv!=N6F9Y`%5=J^k>=o68}>6Lw}adclnp1
z7yc}HU+7<oe&~;~pV#`c><{|0Y`)#U6ut0g$@_}`QuITAmd!W)m!cQ`DEWW>y7KLc
zpZ}fjJtd6Q%WP`nd>{Yg{Fi3Trr6K_nkY(){d>&P{!-+!C?nQSnZ5lib6J%6`G8sG
zGDbqTy`ehqH*VAL_i9=s`}-;Jqt~BOKl=SC`Lp+*qCflpDf{y&K4pJC$EWOHUe_)A
z^Ld{4_2WhSWjy{|?=vF*<5PqE>;LTy{MTO=4HpLb>)*SrdjETi&0Q1hkNRT){ck^3
zu&DJ``g8K#g%;@||8&by>)*EQ;%ffWEk~_?+wu{A|8&by>)*C~r0GBYo=>fR-?Gc*
zfBZ#fwSV8X3;TckRcEz--?pzF|NM|gwSU{Ttw#TJyHxr2ZL+M@{g1aB)&70kzE1x0
z&m7hMecQh9{^RdCtNq)yqs{V9wfdvFziI9)n*VH@|Ng1;Z(8;>_#bZg*7-N>p7s16
zZunE_-?Yry{tvgCRX!TVqdWXr#rRf4Os)98e*N43b2LkD=>PqXXz~B}*T*(6ouJlc
zbsdAD&sn>xAET5Srp2gpv|qLU)+Lhm#*ir$xOMI<Ir}OHwqfRaKwaIT%qALeO;h%b
ztzEN5sb<xHYROX8oCz6YbCjNNcY8m^hLoBX>{bykXx(nXtY4<c8SA}-nPl&%+RuS)
zA-@93KE0{hO2XQ)#(=H23tdAuSf&j?z&6zeuv&{|>(qd~V%!7G>gx%2sROuY+4=J^
z%IH8blOo|PMGmZtUO*d~23$kWTQ<0wExN$D+t!4;uIjL^nNWsJG-IYP&lPN!=0EOx
z>}+PQ<~_GZ=-G-k8v|o)j(XYd#bMogH1Ng5>HWYqV$^MlmA)L76|cwG>S}uSZ`D;+
zu+~O%*6rVtnXPA2-FmfG*4Kxbm9SznF4!ouQcgze`P;GPfGx4f5Nx(716HM6wpTG5
z?f0D2d)OMQ!-h5uB!%54L%rVTHLF$482;UtN>;a<*nc-xY$%g$nJ_a(C4Z{}o7;v@
z%6YPG*6OBIZR?V?ZR)H2Gv8#l|Gb4_r7RiA_SXGtQ)x}stn)Bye79h|sJ$0sR>w@X
zZ0Z`5rJBDT-M$rAwJ#x=78}oPUvDqjmpO~T>?0^ypM6_qNuz~I%ik`koi>K(ZC83-
zlpFTCvp!LqcE-BByZr5*&8F#QKqa3;O>FU7vp+Ag)+kohI^%5oF1^<Z>-u5L>;1dz
z5wCIlt~QYrW}_0Q*)WRL*~dy^b4y`$U)K50VzaYH&kUJl6<E=h%;vC-rkD&Inm@Jq
zIw&Wle3*$<>N6$!{&sb~5U~;N*|%jfb%I8Sdt+iE6=sc(d@K_K%2N#MWj_5DX|TRq
z&2m;VX7lfEi%T}EHle#Q(|JN{PqwJcrV*3^-N8(b4p@~D?HUs%s31ENwXOwJVw{Yq
zQomX6NxHo^{bj3|l$q6=(Av&gX+4>VEjCr9WbHN(6Vgh{%9e>xF(#7Y-<U~IXZGkT
znEiW)*#y1MBsOph)_P3p`FmSWU?x67z)V|i5mXDIvh%mPn3&5Zx~$XTWzX1r8`$R0
z8p(FA^ttaEbIz<}tGdJNA1zobOfCDTJ?>>|_z2jbYPQjviC$y&`P4)<PuYU51EcJ)
zF$Tj~O9}C3y)To>OqgmmFzxH3n2n)kF?+84%b%}ZWP;OKU5UinGGgz4=96xdT>@+C
z!%S|%Y@{U{Rg;?w)>DKT+ssURa~h4>gfX$nPuW|PU}p1Hnq4K=Wn*g(7|Uk<*1B%%
zs<A4oCfizQV%Uq#`ufmLITQ8qvh5?X&$Joqd(UiRn8d8GG2<H+_D37P?WJF&4Dm$k
z7|GW3p=&0sdLWavCRh1<WM}Vex#*NL>3Mr%Glu~vSL7N}X7BWusKL}kBh{7KV?0g3
znb>3>-DKrVWMY%8d|P#EplVW+QdVosYosdbiLv=%W{ZildVgw7`Np|PnVBtg|746=
zwfR#-65HY?ph}iAwML!InbplX8$Vs9FXT^`DZiX5OdF>BsVVI?WX8J2*KSK!w}l91
zT3416K0uSlCO;GHxi}3NHofzgHlLMK5_{D-aHe54Mybyl^2*j<dwOT~%qy9RO+R39
z-Sx0)eRDRL28T6YM2wlLEy-+R!|KZQ+LVay`<(VlY|19HubyEq8Y+S5BsTx}DH>K5
zGsD7%v$U~$vd!eQwbJzVEN8;RZCW*LDa+ZSWRK1=tJypIVthl(rhgg1Yz-OMFoted
z%s3Hel|||>9VVHL*>fFrN(wvkv?^ndp2%bjJCyeQA8krdBpb+7+AL){yzf1j05}~n
zX2NectFgcuvk&m>I_^;GVfL`T2Rbx=I_$U0?Xb18Z`P*;8?n+GnkG2ebr|z}IfhnY
zavEx}Ucqeg@=KdKJImRp;G08z$`R|Zy{a>NVhLuA4MqvuEFV@~$1`U0KYMJ3SmC#=
z_joX~NSbL=Fy3}Drjx0L(vG@j$Ocwv!d}}^b&c8RII@+zCt90lEoT}UFdIl1Uwh&s
zhG3hi*?o1(x@?e4T{Yd-W@vU}n&1@J8n<iHKClhzbLMZC3C;;a=S`-p-ljj>%*o%1
zTTjM@=3{N^`~bCMPl9X`fkaX54zs1v#6MbleCsgNES#A_be^{PQ#M#FXE_nKlNaTz
zZ!sqtBWvQnVb1ns_EH+ZUG^4xGQL+s_HAadK0D7pV;wRw_>>wm^DL$y@ke0Hvm4V|
z=Z8a&>JKup$rzdDKeI)fh0WQRJD<w_^Yowcz<~tCrJ=KGR<})T(pyX#NoMcP*%*>f
zU;@Pcy<sOFy+Ld$m{moOo2@-7nTS7qlM?mWT<fwJq-HaIBvYm)R^U<e4B2<%V`*z!
z&nCJX)QtT&s{~9;=)i1d>|mq&tEu)~T!eXpIXX(BX<r}3ifQXJnD@-4on~*B2~9?+
znD9>~1BussLCq)FCN$+2Q>v6Kw|%4GY@XzZn7M3I$k><{ZB9rqwENK-#pE_}1!k65
z4OUHP^3%P`(O_ZA^g-C%E13<V$^Oc2Nqn!yY!IoB&4C@v6ir)_?3>bjG+POaeF`uM
zEfltft)+Z>pZ=NfW9Pz`_3Yo=X1Ty@+-ESR4qtK(lqmxCfhjDs5RiRCG_Z2wm{!4J
zfLI8l478VzrpeCig}QQEF^@ugPdAJiCDYZ91rA#n{1uHNzLM#a!+&em=1OKl(-~S=
zFtgf~5w;0b^S35?8*SJkbaZ;yLNzqDS{56Pg8>BNsM(cadX-n#R>>~2ATe+@HCvXm
z|H>|qeVs=|0tp*)hFtO-CbrP=Qxm0GOiYGC>_uirY>8$xdsuyf)?WQ*<<m3!80bG*
z@+oB|fXPh8%|73F^TxD!W%fqn0nDN>DaE+ipGTuNW~LoGYc?BeNPc71b7v){1LjU|
z{IqTjIcLm7rZed)e`@N<X51=TOt4vpk7YIe?J^UaJchmAn)R9XEMBfw&32cf$=Et+
z5;xm0bA#<l69`rFx4WLXX;bLdXLgA)dNz}9ySCCaXKeH((?wVKY|l^4fSK%u*(A*c
z!fX{(CR^AwHnc$Y`GIylX0;(3TKDBKSrSvum?Edx?(CXB?d%1@a(lp-t@jg$0p&d=
zGqLHvH4`IyV%1dknh{~y?lif{Z>_XseQT=sZ_CguXL?5Ax7&u{vRyjvFt%8)Y<HWw
zjGvGRr<m(Pd+PPIT#1Y=?23YUake0Y9hy3wm0V5O$*DZ9Jx#pE&2*?V!4`54hHAu5
zg~c)1YK(xe!yL$#Z}sZz^UUgOe0T`^Dq@0~v0{>wmthaxlTCZ`{?iiAU}&)wZhE2a
z5HV(~H0GzV9R>`{1SdVQWuoP?+6q_vw`=XJ@9cT{4*I-o9hW2aEVDjkTN-9_GV4=&
zF^sXT7ctGuUKu8FVReP{Fnh%nF!bBhh!vZpGI&#SK{0!U9WmIJ&;S(C!DCGH+dydM
z1&wdy4^7B*M<v!~Lu>O`X8Dcm#%$Y1bKVz${h}O(p#}wEGu%s=FiIv^VXzuoL@+7t
zfig?1CT6L0UL7dxGMGtDIx@7~jwl-JFSiPgKb%=AsDmI`r*AOVzY33{$;@oq!&z*j
z!t~6|ksB_%xEqcC)?P=J>nu+&lZ7i;kk+y%qx;lE{AHn`K1J-4zim#wYiIR{j?F9g
zw`W4)cGqkTAIz@%>=<n(y{&%ETFDlKA%X2FS<aTe?!Pq~y9P5^;-hKZp4qI*bcxlm
z*_|OVIdp)9mAhRkyTs~DlVDbzyDIisMlw_RU1FJTdcHcCpg9+;=QFD}i{X)|Y_26z
zlTM%(>$)*hn0s{=TH;0**#3c;s#~+ViJcD`J<;NQVt4A%IzZ}QVK4Jc#ElANGTm9s
zy3K)E0olrU`(YYhS7+&kiNa`05Nyk_X|^e#COEgV9<3TP!RbuGme?r5b^325ECSfB
zKAe7FaZnp<d>1UhnCzzEm*Z?^YNqB-tugVRVsQr>zFo>>-Zh&wOSU&><LJ{H6UWSO
zXR(#Ev5jn&!^YlOZkajkza8`bR_rf__1lJQpS1SEMk4FHi>rod5KN1Ei#D5z`Ls60
zd`!%cJ`aX9Ys;N8j^KiAglJ*=@CfmZ1<jWVi_O)I@xAt8BL)S-U8=WX1#2)<cgA65
zXB&NEAQUzPV~&}@;yWBLx$T<#a-r4giuq}59hX|{ifnhat%n^^3EFzbv$mFxYs<F_
zEmI3?GP8wzhPGO2cA3p~|E=kM*=04=@i2}6f|>mEp(>#UZZl+j#crWc7_hdy%@Su?
zAn|3=(6MU_HG9LSsu`nX%w*?2EQ^@|GXjnE<*p3NMu+Li@>On{@LYeb&g_V_zFJ*u
zD@O0qBU;cn;~~<w)&$I%VkHh2jcr^0f~_oPZzcv{_iWb$)i_MgZtZ60cFK8S31W(r
z_#V!_aJ<u)!p8JGfh|5gP@Cti)Y#)TWP1G>N`z`MIUlvHn;OV|Am^v)Lu}0C=9%Yl
z25)T6eTK5ttj%n-Bd%J%KV2p^51T!k5R;qdr(t%$RNdzRJ2u%i$>~GXLnk(--Ua`J
zXSQp-8}X%|jj6rxXpqC$i#^Py_Yu3Uj|Owo^2nDS%pPfl9Fohn>mfGvqAO;yjOOa9
ziM5)|j&deRtbPmA&_@n+D3BRWzL+9-W-=3-&Q>qk<QkYluyr3nOl%$&rW0(9b|if}
z7VPF^Y=ra=+aI=v>s+5zVzDK0zQf()GdFvsOjlRi4l@b1rfVWiLnq*{ZyjM3dLC%6
ze$dX!Y$b8oZvr<|7_(Q(iiHzxJlGPKRlSz1ZJZSc9b6M_IcupcB-xKSc0CLD*}X7i
zlJck~P1a>5GkpQ?aR9k7V74B%RyJYt)n$JlDM&G3)MbAktK>td2~9bw$-c~l=3z1a
z%~mm)d9<I{Hh}5*uDiR%9A;LIuccjzBr_T9bF)hffFY@8ICqWdYQ{>hdRT0(!IWJ!
zl(^XB-P?Dv50jQn%*q)$N@18>F#9SGOKD~1O7|CQws%M!L!3{8;(*@}W>r1P=}cz!
ztdHkFb4n&-9wxU1HMzN6Bkny1&%O`u$Oy}MoBe{*hAYz{!SPI-ZiAT49nYZLnQ&9e
zbn}qRW~%gc^kf;gI@7?8`oHrnDhjAef17MEqs&~X;--|T+B@qr1J1i==ETT$r|BQV
znbhnb`k$LWHT8e9V&-<~KU*A-8<Xl2YPNjc7%^JfF@I)uZ+CTt)mFNhk73i;t*P*5
z+1M~Xv+X?`6JE_;X=ip$wyKkD`i}i}-m4*d#mATuIUBCZB2|3>uwHAHGxf$Nz|?j$
zV&#LY>o7f9o%&Su&aEONn;j-Pow0)(Y?kILz_Mi_e`{Uc$Epq8S^4kXHP+SlZI?~@
zu!vih<9OHE4Cgdu7h$*ouvYu9HT)}E5*G`UT@zP;ZSgw$j-Wl<qH@&>WERU&O-&S^
zWx|ks#M3tfLX3ls-gq5rl>M>gs)?`%o@_6dSva+<U?y*T1+cH=mb0}hwkzhXZ<z4R
zZH*d?X|#Ozqse6^KAmyk7HpaLbhgWF*R&a50ZbMiSqWR)7gTW@>acCgfp7MdoIcnw
zdNzP+IC?2|jJiFwZK*uvuwzuurqGCXjY;uEV2?O+F;@FgC-*#S1xSmp0^x3Dq#f;l
zON}YPF-7$i)$A&{?5O5jOa5@+xaUaaGWULCODufBj7|K=eCuo{!shZ~X(Yg;dkY#H
zY|1%P_qn8bdZ~U!<x87R+4)f0eOi38Px80w#wI4TtZYKv<AV7Zn}TZg3jPbepC~ao
zJBP*1H!)K&H&%Tw9Jff>`CC&5efPh#*kpy7)({(;Em`k==upPa7P@uSTL@<u<*-zI
z-&AXo$(b2U;-DTT;ye8X_tjvlGd1~uVxJhM+jsxXd?;C;aN^*5z4a{F%E|Ng!<3pT
z04%n|Z3ORQsc)>XtI*FH3yTd~TUMi8>=zU>_M+!mij1Ld`+l;!O9tEY6*tik_UdQ;
zzW;Bk{#l@n^>TLJG4(F<-59pdON<pR3Dpl%8s<#0&FNkB4Wk-tn&AzKT|I6w6+fFW
zGjnI5g+b_h`zw){<aDHR>k@=$byvH5tc?laM+c?mkf^&Yu&Gkd;sjg6PXJ8x#_O5*
z9{6cW51R|bRxoIVF=ed!sXE6RaY|;)3@|@T={r^@{AcF&^OaI>3!NPZSZx;}!u@<D
zG(}?rnFzB*KiPCK#o0_8lfahvVTx&X$M!Y0nEx<|lT(;YC}%r6XJhR@^p*A?S0Xq7
z&rf~g$104uE%dj36D#Lvs<+z@-G1UUM;L6138Uv^YQ|}mV#3%r4hz*BB<JPNU5|3A
zIErE5`XSk^jz(cC{NoJa+~Q|-b(dtLa*pn&^CO!WCO4gpcKpwbS?B#QWz8_TVWtZ3
zBG!rxVq+ZeF;1q4-TI`S<3R1VSrdAGmLi|t+1fk(46^1zZ_FgCv;KKl^;4<p3;r1=
zvscbT*b*3e1D=R%mave;Gnl>7*$S<=sl-edyh+<WC!|48X8jzsNC<y3SL*bm-`&Dz
z0w(Nba0dJ72RmQR#`-Yj$^qxMao}#u)HFZ)-M0K^?@XQTe$HzJYci2@_N%#u3I20_
z_G@c(&H8MWb{V0j^5=7G`<NJF?P1xg`?mUcavZ#q*-Uq~W~|_tF+M%<5xvNlOisU}
zRg4%DryP=(@KiI_8aL=UIFAiR*e>%Nu!mQQu|A5sCX{eWGGw29b%frM%w6hG6~Zmi
zm@RbfnP6XEyo~KmK+ZJWXpPzQ?J$$Zn$7a@O!pgzy(yS0$PG<6G(!BRdvi9!0Pt%W
z>~iDVF88-)scd&w*frK7{I)r9MQ@og-|_EuR}Y&$v7X`uKHN8fox_oAvyqIK$%?NB
zpK^DIXtUfF{Bv(?*tUyIcB;E!?3!$}<AtHEWTR#=4EG30UwvaHyG6p&LxY*ZTBxq~
z?3#V#KAL^(`%6<pm$MZvSg_5f)co_vr{5+!ooOnaX0a3w)A8&d*#zgbjMB(U6{Z~!
zkfHMY!>s#u;l@P7|K(@}usYPee{9N}uN&AZS4ME2iM8FK>FbR0*x|1W?A<4jjk4Zm
ztiSGMGHa|Lz%l*pnm{~a8s>tPo$vMe@f5zXhJ)&vy1<1OCVoM&Dd)_@s2AjrG@;p=
z?p3u)C@MgEweAXJ0$guN%vN_lvUhbK(G~)qB5R1=UMivhxWb977QgG_eYZJKdO{Wj
z-h^p(87~0+h%<J-C{b5JGX{@;RSUh_q0WAsuEJ;m;tJ+oS2J$$0IhBJ3NkUbZk@xv
zNJh!b;XV@L#Nqy4<MZpO?kr{Nu-vY%d1kJ?v2nZZ3mU#p`~DVH#D6@S-ahUw?zgs=
ze4b{TaCIlO9X_u>o%?vMspQ0Rq31#k(X1DW`T%n@KNFg<;vQEz>Jj!1mA#eYdUbzp
zwxdr>m3rH;wYc*<x^vJPw~rQHgQe8`c=$ewI;<<-I8(g3Ut`JRZil9=;U-u{eiHb&
z;hDrGl{(smws^wWhD;{oRH(7%12ZXs_h2nQTIQSu_e>6}?vGuGz-KIuMxAZ8d$C~Z
zil?S)$YO&l7FS?3Ve9RrpS?qqw(c6KE5DP@sg3(MHfz0Kh-)jb#U>fBkLfp1n>|)b
zXy)2-PtEyheApu5EVp8TC_w(}E&73bxf5;=Ar7-C@1B{x?oTsP#NBW?xtJF&Q+9x3
z@B2I(A7tPf^*;Ib5wcl4?jxEd>MpjOyF&*KZX&q@><8crfUoQ9mzlJoD>jK%zsw}B
zxSwp?!U49S?uu&-q^QGfy14(SEjgc`cPYmkw^Y3)&7fGvMOXT*b7=9gH90Hy=hI}b
z-whtSUTE@uddbyuRt@fZxQ8?6xVNN9O83gOx7I&Un-4DTSSumh4?O#7pDJw~axdpK
zN`51F+IM%%HJ!c015K_aEiZ{r=7qS<a!PmQtQkg=TYJ3qcFQ%sN}Qy_L(!R>xqEgO
z0=mBJ9@hF{$~wo48RK?`%~OY>df}|xY2ZfI%txH~tG4CN-LWqZG`n;1b1rz?9xYuy
zzI(*}I*?nV4yUYJa&vVED81bWHw#{-dz){$U9#oL9iPqm?6-Aw=TLW5&$X4yESS07
z{fm>C@~Fqpz1GV_U0Sj2s|j0oPj8dzO)gWm=DJ^t+6nt&K_1+&828uXL=GxVZ!7o+
za#6rE8M$4=O3~x8pL5Nuqk^e#P-ecf#kD2BXJF!U-nONoE8`I=ZCq%wmzx1KU9y4p
z2FAHUu<uxyo+b3M=4PlvY&^ezpt%fkKMMyFzU!GYOSrsrf*O^YpP8_=q8<&2`wIze
zFq(&e>Mhq2K`))vFGTEo6<j_Weubm4^>g7ync_Nyette8%9Fjwk&H{Ner2PUPv6bj
zQ@@~r=R!ifpJ1ca`#fyop4%O=Z!>};4=|lqfZI3nxf-JF15Nw_aizJ9+j8!=IR>Z3
z7j83kfIUUtW(%!*!TiiN_nEwtSc&_>#fkoKmx>Ep<2)W|sO1B#AR}bsBR858wyj^A
z@e`<k!&DW1ZtpfngFBkGW&&5;>#xC}{MOC<K=pFk+eK@7g(mFuGO0dhrd6&RER6dx
z47mx*exZHI)!|&_1NF?`<w{yW_1okFX*mKa-(jI4)b|gR?!!xaJz@3L6M;YG2Ws~L
zmJBHbY`CBOatG!7@YT;Cc}laTQsODsoVG)icH!t}Ja*cc4i#TDJe>Es{p4uIOODZR
z>-JU(L%i>Am-x)sGS_Xy46oJBeXV{AnqJQUX}QaZB6lgLt*!ekX0DHO|7+)qP<EI-
zzMJL=Lgo6liC?dBX|xSguGAac;gbh6Iko&c8oT-l?bYh?`)Ik|S3i7iewnSu%^=`$
z+nQYU)7yGCLC<}@l6SX&Y|}ZExPuOy4eu}()VN3Ea2LP&$HKJ|nr2kEn8(@mPACOa
zolVJlpOUQ{|8;l(z&880eqAqi!5p$fju-i!cQh1F5jHeO2umUeaHYVn|H+4O*PpAh
z^P<|iU;k?}EI*iu3-Q~8sW}tSU~uuz;0`Z6ApSZ2YmIYnhbDZ1z4m(Rv?XHGM||za
zg_}F|+-@AVLvyeU_+8zQl=W;O@)o+JIR{+J?^PZfIiM-gKDKB`_vgf4;OC>oZ8P9@
zQ`!AbK2K?r+3!}GsMOncw$B(h6W1Z-iD1#VLx-<@50wVU=GOFkAeYtQv&uwesH@d?
zVHjW}<+#kfI&svWw~^k|$JYd9e0RXj$%d;+yCxh%Pn8;O55QTu#DSlGN_n8LFFO5L
z^ZwixX7OvVeb4)Iv%QP|JYBzmHWb}&!Y=)E$yV9rKi_dZ)}i3xKDp{wFC_E=!-ihk
zFVyDOSEb)K={LbPuYxNUD`9n)qv@Z^Ey=c}|6EqS3ctJN_kT66h&$6WhwV#E0l}rY
z;a>$e2WP@jeG+1>%VmPY0A51j2?U&eWyiPO_x-gQCd1*5XcoX_;@5w!X&O~@?BzqO
z=~(B`ddJVTsos+IP5KbyKyzE)NJ_smUBZB?sGlfo+_jjo8dfw&`*eW5DW6cpuZ^31
zUoJE)uKIOtE>~Qr+b@or?UXZ?(0ts&M;ETQpndXwbDS&LdZE~B2ksi`OGQKR>+WXr
zeA*uD|Di~?TWl_`4_D>OR`c8Vm7A59*-Q1`PMNo$jB_8~B<Q+u7e1!t7jgE5i!0{b
zhHr8oP+eYd%|vd&qqL*Lw|wEOBj-*&2L4KDlHW(Stz38}QAzw#eiON@wR0(-Z+%~A
ziyl9=@A3mpP8J5dw;yd2-sRlsms@bGb1}cqCb!@uTnF&#e8BB@^ZRbM--}=pF#Y`2
z114}*6FMnqc@v{W3=Wvp(s>DEv|QTw3b>xn?V32W2kPKk^^=}_6psebcr$tN-kAws
zer@*ptO@bHWEgGYg=quiVldF;+AWFyoJUX?sa@3-hmS-dDC;)f7rq5t?uqjyAm>m1
zb1n5YbarjArN+MfoZQ_4R-LyY0_r&IqkW*9cnIs7%?5?*Ucl8uV*_$7!VmW6*BhIJ
zt3_8QSX{L7$ty+X?T~s@2Wj_BNa#T8<KE#=o4;EG`$||R_MHd3%Vxy=Jw3X`F5~{Z
z;l2;lnTnjScO?9rOOLnCj(52S(TE%AKwIs*N7m^3Q*z?_3}k`4P;kE|vcvtr&93T%
zXTjC5>izjr)|U1AZmw4Xqi}zpp(^`65Em7j$Nre@1I1siEfwoFqupn)&Rd`2dIwsm
z)Yi?NPTJMGXlyQC`RG!9TqqpuacEfkg~xFed-#g&jbq~=u1o2;_h8VL9}nVE=D2KB
z?(pF^xySvrX}2E6g`0$$kj318tv?uE3B*&W_tP;$YjJ@dzsuQfN%3oS;@ijfA#Flz
zhqJHyb6fkW|2!|I0_`&z4tLGZadQoeI5mXTi7JVLjk~e<@yQ2n?@()ZSydK+MckL`
zcMJ3*K=If;{W26owgy%zNQw*R@PfOvTVrPr9_VUyz%jV3A$}{fCghUH@~}2(>H`-Z
z348<V7Bs_DJ^szDZ*8vA;D(x~J(bu}(+9UpigJ%ZwY{L$b=xqB>SH^~)uPO#YOKxN
z#LGhq;KnV1bKxFxpcTxX;y)j$Fd3}au|WLhE?hp-O!b+GTJzBv97bGbe<55@N&ULv
zyK15eSZ7SM<}RL$qD`ofH82yr>cNI07!HsUjN(iy6o$G3libj19;-~f%H8$;S}J@0
zTr8CNfl@pO(W<$p)bIu?KD?`q+s;Ow!faK%BcXnAjC+Z3gQzzAn(Xriu5cD}C~VMi
zbiO|&aTRi?IVGFNgInDrG~K6k#~WqbOnZ)f%|^I4ycIZlZ}a|~eIwY2<9e8fM?>6&
zSGzMEUZF&E0N9*9S4>?gx7?~HUQ}wGARz_<lCs^OTC0sV{Q7h<<?26?W3=8UF;l|d
z_MSP6_zawn{ah(b$l)7MatEbxR!?9GBF}_Vj6*^<5hAOk7TlWK%=5&}>bE3E^bUK3
z#BnJ2i5MJlGah&@F%!;4)(5iP;PJV_UlxcC;XmJ9ZprEIi4|ep9!+7ZedBW&CR)SB
zA{4;(a2CmklvaA&adW_3G~cqj$PDi3m2rjgLKC?TWy}lhJ$p>GvhVlm$8iH_f;30R
z57dOVv8Pr-tD+D>t+wzma=4qvRjc8K0BZvu2KOPqKkfzdiA}_CC~~-PV(<ko)|JKy
z5Bn>dgPeUkWM=f{!CT2A&c-1N8yfhf=la|@B{A(1*up!;Apy;RBfR;5(q?fuuk0g2
z%=9YPl>0oJ*?_P6{&Zd4%a9eggn(|oi+|$8^|0?7`|#}j5}0URwArBMrf){%9|T76
z!R;e=tO$Nu2p++T-ovZ{F3v7?I8+4eT%rKV&JWwe2l?=N2%lhMY#-fGdbu5%?)9-F
z)MmoPMIfT%<xz9<@Z`hcD4(A%&X|9>Lw~=JQ*?*i3Qo8y$BSqlYu^#t=gEmmkVxIq
z=R$M+JdW}qaYZ1O$ws)T>cRio6_9MO;dbxQ|MKIS5rbyC8a@)5jpm=>T&;YdvNy@C
zgWN4^>^9-ofEUk7xk6|-!Umung;%OW=^oz@)&Bl`nhfC)S8Jb=Q-^mR<v-cN-wrmy
zHxY<)^zg8Uad*I>OpkH6Rvwu(_9Lc;7F_qN8K^`LIQJuNk4zpbgUOl~B^Mi`9_LW1
zbiGjgTPEJaWo8L%s+3Xm{@T{|;t?BX`oS3VXqVeh`)0E42_)EozYn?4Y1H8kS76}i
z+XF~)8LM8X`$UlBMJHq_IbkL9pu4%PRUMiE{eeO-0rby2;c8es>QN-E=1)Vr5Ke&W
z?OJX>3Dr3~re*mI95<64<`rd!yLco?GaJJoX#|g}{z+=ai2~Mww*hzNSB^V+arJ1D
zns$3bvu)L5Y~EEA-4sV$wB(XYKk?(5uO+sk?Ay#J?S$yu{quACI2+?`pT)Q(r(c-E
zQEbZX`iwxJb?N=Jz1B($(b9Iky%dfafjvJpVFCytQ*p$*zcvk_c)?Xa&6p(WUQ*hp
z>qM#{lByYJ**L2w&JF5u=+=e!4*l{N0;wKtccyb#(?<oki86qjTUrK<1%h}DoX==J
z5c1qZw$@qdZKo3`0l0GILQ)cULtWVP3-@VdH@*FtCw^>Aw|v+txK3c3jCEVKue&<D
z+o8=r#lfJrmb)9ktsc#_b@@E^o<v#nsIPi^X(n!G)Z5&tFECN?aKH+>nl~?yxC6vg
zC@FXtZFD_uDqTB(q49vwc$Vi3rqg&ha9DN4tNK-M#>4d0Y@Hx;?9{0TV!0*F@DYdi
z<0Z^o_ei;H%HD#rdZb)D5^s%@;Cn!s{X!cqb(nsc4vBk1Yb#ZB<=zTU!)^DPZ2MWJ
z-@N)T;aCJeMj7|9)qTo1dApK~FvIxMVJGH?(hGOTJO;7pjpM>fIHg}6kRsGRAmNFC
zqA{~NS4o#oBU?DzE!jMT4o7@p?sTsIaYm^T2rc07052Y;Z+{@+Qh>LS>7hsP!OUHH
zAdx&gU~#&IbnZ26{~QUI&=Ly4CV&zVv>g)d>6ZrsjIFeerKJqe)ZD7>9t4JRE3RcY
z;*uxlk`cPY^aC!E`QX|zo-J38sO7NdLfpB1Q9Z)*8gZ2X$PHdEIV0$}!^IK`%yXt*
z!;r_-G?jTEcUw=C=a>W?m;UDRZrx-0jvqc2EK}~5n|_gzh#`6I)rqOyJ<u~ALLbOI
zJ77!CqK=C>S1&k#O<kws7tTpO+XeFZIc~KMXYQwW_r_*?NqzKToBd6$`mNZ>=<Wo~
zU`U4BfK{%g05xX|p2uuvvn!Dk8u#Jl!^_3X`{Q$$mjI5blyp3I5zpKx?DeMW;@-5w
zk?yQO4+YP8azs-Cn%P-AEcdFn%85@jOwuLllNE|PIeM$KH=qu;w%fQVD!!VSDKhV1
zY`jMa6JB!X3WyUXu;!jg$4`==1-|tXD@VuGvAD!RPRtA4j#RfilcD-@aWY*sK<dI1
zSv8Wi>2s2$z9sAU8Uc3%q;trPW|+oe(1}v#>636`(%Z^Vlqk;DOY8%8BGqv@0$A6l
zPHw9o+UBQuJh7Kkk9rkUI97K$auXZs+i1|)?xWirJTdLdJ=yGdSaDVgl~QjPR6J<_
zbm!0$czd3(_GL@XS&vvM9&n!EE<I`WLnPb=N(8ZiHV=Idt+{ms6`Th=z<e%XoiN<X
zW7~85*e=x5){XrQ9PCp&{bVm25$w$!7g9rk@=*iV%uyVE&D2M{dN}Z;Y~-C__-)@F
zh^KUeYT8j9BaHRJX#|aTD(uBq4+_z4YuY1^fQ?4V2zRSEi3D1U`?5|Ko(<JwA&gtr
z7JoTq$DN?~C)CnUME{ZxOq{z&CeUUIT>{+^{h)f>0lAIY@~($U-0gr4*O05n56B1R
z3?4aGJ(kV9T`*Hu^w9Ky=$wV0m)`Hx=Pa9G2%eH44?;VEyCbWBeM0q?6<^2Phos#~
zH_L&;kt-R&yZv^e?}U!~1d4ESI@4#YbJ)s9UOj$nJ^|;Y^9*>G{B$_=oO1Ffe1tIP
zmaLmjBiyaiMSKMJZCU-+&|6ILQO2iUvtaem+H&pWBu`EThtclh$zj~-f<5)VMa@J%
z(|Pnf!RSQw#s#nXJ^OLXa?j#a8y9qeGWLoL5?gOkHo6Y|tYVmH(ODUv2h+yc8+50F
z4mGI}G><$whbQHXIF~kRQk-}Eotg2S@{&+_#7OXT7j53Mt>YeX+<{{M1*&XcIL*a_
zeSe}T0DJm|b~fr+-0BHH)-DxdzC-gst&89F=m$z{a6uJW09|Qqc6O)<rvR1QuIX$i
z_i<8oh{~-&h=qQ#d&s9_`%IeoS*+NG9p}?srG6ev-CM<}=FBGu_z{8b)09luv8{fc
zKIJ02it~&&4k-Illj5g>o-W9pZM=2HdyU+8?7r>R51r}h0TsL@oTguACnh+iZS{b6
z^VTptRykDC<Gt?~Xy`6-pTZ2;-4DIG0l~p{JvHd~JcaZfiC0SfL3N*1L(2ZtusL8p
z`B<jhr%;cz56v4|PCfpa(+ECh3y_}>I_&BA)G?CJx+m#v4&&I`;YTv7M~XV|DzDgu
zHRyGj?dpL_d=ziy1E?w7E9m{6FmmcypiEBW{2=fi(HNjw<km3#c8gQ_Ji;37EKt}I
zw_Y4Dy$0*yr)Sp9aq{0(>6acA)cxsh4@~@=<!SdH&tUWY%yC<RL&Y2@N#m?f_BzA7
zf-b@{oLoDxI*j6Hr3a?FH*lZl+AH^nhsXNa2kK?LoEK{oBL)k&*2#FoX7BBkNsJYA
z$Epk6>nu(LfV%D5Cw5jRn`86VyWCm8X$o=ZlFTcVhpfY6M4aK)Rz7_Y@}wTRJ_BQ>
z?pYkPx7!7E@!&8|wV;d84er30mp%s8o}T52Be#7iep81=90F*akzmMv5?|Bc-g`Ov
zfO_JYq*BiMM1a6sgU1&3<VvmFE_6NW&+(=Xh}7{<%sNtTzidf3IjKr!ym0EbMV+Qb
zF+9@nJT!x8IaT5hJC#qrWcm#E$gZDfFw=0H8d`eiKqYckm7b?+gKM`0cz}+GzTs5a
zS#cU-ZE)droD5avz6Wlt;&9<G&LM!pQH=YN<~G*gwkTIPppw@8@F3m$R3Izo5`oD*
z)`LgD4JB}J<8qBNj5`-NsE0}>a6`Yf&K@Yw3Sl0{m|zaM$vFs7E+@fO&7+Lb{8Gz-
zsiOeN5^{2J)!WYQczZgGEJ#k(LqGU|Zo9SW^xzNBG)4x7pof3(m=K{WLd63>K=&<r
zObFo2Q;9*iKaT+cO-T=39tFZfNa&@#c`N`fN#-UA)WbG8%@dTteX4pguA!yhvMdKd
zbKeoVVt$*6v2NMbS`V7WBGe;Rv*FU}?U-3HTMtd4JO}~kp&|HI^B5V5mg+G5;|9^p
zyH9ySj;{;%IN8_iJIi<G_g^CYTJNFR4sNh8Rh`Cda;L5{9na=*%bLN{P7Na+tf70J
zA?J3xB<&;ScbofGJBo*kIMFitrOShqd*Xu+<%X@lohCg8fe_~&H3FK>xPxm99yfyc
z8hQ9QxQKu!tJ8gZYafmJ%W2wgTW&Twi<Ec{_n9-j^l}H0^iq0gdNU48!c(W|migUd
z5$bt2z2*k#z#`O>7_a61Cg9ejl7iOV&RZnWORaL?IquOUs>gu2)pm&)-cg*Bj$Y5V
z%iFlE+1fn(E%ccO!h;Zk9#+C{0@vG`iKEgFRk`&pX=g58kJo{NvPZ%P=Kbg4KE~=}
z+vUy<cN1=B#3U~v4IF)OzOTNkm=gh4-G1Iq-;!Sr&bCJ+;`;M}Ccn?yi72>0lNu}Y
z6NX|ta9cfz3^V?&Gnp+)>bSq}HXvT937^y^Z~$dqJN>{h*U6b^rue+Q!oE30)6)zn
zD^e+DLOZA})Fo)%`-6JeXQ!XeiZAuGhWAxao4?u*-ThWT0*^yy`wRj1%})(;rC`Px
z&tv9$5SFkn1{R@*p@7bm!B~WPWaWr39uDQqW&0+_g-28qq-`eZ<;<+H(|HTH%FzPY
zAiBREL}oZN2LLCldZgsgyjwSLl&T|WItLF4yNnwYS0BYOsg;0t5hn~`J7RK{JR5*=
zz0&Dt5)g{89Wluk)_n#o{fzc(b=@wo$q;m=Jln3o1kGy^!?WWZIe9{)2Iba&J2XxV
z2R<R=<N~z^53_!_2S*7y_N@Rr`BX6S^b5L=Jel*V(A!g&=%MSjHHp;eEd2<8ei5en
zZho$!9QLmxC+F1<wqLG2Lo1PI6e>f`M>{1i%#H2W5Kgyb=eNjdbWE;!Z`urwj`zt7
zI&)VMxMOzD+}s@mT;J!botpYLhM*ojU}~C&(8W~*zh9FlFauLij~=i}UUV3K99YJo
zQf|G(Ug(ikbtN3t?HG3``xG7g%;Eh&IETX3zj~U{T!A<vI+BYx_L6YTBX?H2lb#b4
zw_8!I>*r}jVt$)FxGpte5gB0n)#};#kJV~jd&l?WkCa@QkH!oQZvV~Gud(B~peDOd
z1v3KfLD_9N8E9wf$xt0<Su)*JZ@tZvz50!&y!bHLi=0Pztd&ab^w69~POg+^^K7`C
zQIt==h!b^wFXKa*>*w7_`t8pXoc~m?^F|U>&Z4q*&VyNgpA946y6yXplC$*anwY>x
zQT}+UPpnEp5;#rAL#XT)Wz&;sdN8=@lsJj~Qr$;R+A30SSyPZ}&^&*g9{=oUT~g~0
z>KQ_v=DzMkeJDp#O3fvB`g(eXTQZ!MQ@=OhG+kDwN8CDUfMoVcr#oi2w8$O-T^>ub
zdR!^Flv9tOchV3)svUy~EANiOZfCubOkM0S#*xH2x3@~F)kE{r>m+2-`Ea?^G%j5I
zEKM?V>uwUNs2@QoGcW2v_c)4|+#`)ic=s`)hV)pU1p66hZQ^$S?LuGk$Vyom8)r5T
z0C5I)?ed{JGCGT7n3vi7uG|>~NhhmG#F}*$CtI*WZGe(F?a&O?GKm_qj^<@bS}a4*
zEivTRt9TOgU5-K~J0pe6=BI?GwH<-Aos!eZsiyDCB=&o}U(Nd<z>wEONv;X2d&@c1
z*x)8Feax_3(sTwcJ?emcyPu-H>G}Q$4KOa%nC18`-NpuaMV!`}QiWg`<yhYw?B22j
zw+Xt|trzh-4ozPwDCrqazV141SvG>_q5add?LkqTu674(OFIJ7%$IpbOg$14w5M|>
zRQnN_`x<|f*oysFsyl-uL(M8)QA4gK&{(DSNJpSaV1c8N5anQ87=mSw3~4gC+<&G%
zH%o8(!z0c*&8`j)26bAR5osK1^Zp22Jfv<}a@L+foHM1XUtm9<`W+T7psyn^aq9k1
z0;wJvsr8`SbD%nnR02+Y=hf+2w_nPY{6o`;VRFlQABDg)p6nO*q<0S*L4TpPFdmHn
zhnjY&#vI>r*DXv!Jpz*eh_{ALm^`QZigHZCvzqLE<sFe2BT4-AJD<~ICLbt2;5zbW
zZ*8|%rZMYzH@Q;@Bt_7H>039snp%&zJn{B}@^PLQcCR)<SF2G&nvRnzp6v{{mc053
z-R+1=EqT~G3D4L|l)aOV0FAu|jev4a{Ti9moyr4o=KGN!T62#G%FI4Audz&m%@aby
z=}1QYi}?Ng^b2$8z`RvEJ!A2RqB?F-l-E->H`*rm;Lfw6cP6ueI|>+~8E*Udu^gC-
z2)D`BS)9<2Z=Lp<Yy3W4KqkTGCEDs4)oF?ctUV*xGw#Loi&STFnE1`N$X%%2qlT5s
z5Vx6jbgX9}rHMVbj{>G6cZMe>p&muv%<o)=unF~O^27zYtqo6??R4avA)I5WQSbdo
zxhPX&z{kE_ARTo;y**g*3a+2^<X%bbLGNeFO^~C*a>{ob!X@XxJfxTs@y49Wye*j>
zF@oN2iTe)jeZsx^vOZLM33|0XZW#2V$(w1J&&qv(e)KP0^8=+HIjQTe&cv!8Eog+1
zcIufCou)}Tac>`IdA9FXZ`oPrk(BCIMN-N6O*G7;iwvxX&iiDjx&dPoVB$o*f@O@Z
zxqV4JP(LksDjwiHhn(&-eAIZ3EpGD~imhqkxm4kiD-so??K$z3-6d!`xI)o)4wcmO
za5>#)T~egyrMizIpcz#e+NZXl=_xa@2sOrAW7B-Puj3Bb_wCs6Lv(s&>hTZ^y7C*v
zO|KloLvQ?vi343NK_h8{(~TEtZ9mRwCQfI>5pbG;!zPYjRikXn#nGLz+JW(M7G6<&
z%&^?d2jbN3Fu#z6JCGkVHNWKCl6n|DhmlIl;c{;y4_Q7Fa+N_G*1WU~-6Pt{RdL~I
z;b)7UJ5&=+@1eunr@7qg1eR_XivxP@hB;jyxcxK*D4$A?Md5UR28B3)K_gX%)5YZ(
zkDO6%{}tu+zziC=Jk)qQ<KJ3*%F7R|W>k>s(c;L5I2=wha3B;_+F>5^EI%q3LDQ3D
z`Of(nmnI*`wNPjK!-j`vcdD__E#DTLhxQD^w0x^;pI6W`fDe>RtB2+_zQtaE`<y`2
zt!h<I4IFfgTd#=woOr6(exG)G7I~-X<h)#eS?4vFuBS;k7!I177|Ul$OyvGy;yx$s
zxE2Jok9r?C^5`C_?s5XH`4R{i>#B3}@CY@j^;LrAatoF70AsZK`@!a_!};my;C9D>
zdh47s-q*;jxJ8UoU#Xf;oZFBs+vB_gBDW-E<34C;=|IJNJn%CDW{CFlVrIG?ECO!2
zZqoe5(j6BQOBytCR+wVO`-s7wkpsx7-&H+S6IQkc54^LR0Ft`~{ZSTQtTSFSk6RaE
z9I<u}4NC{=;hdeu{?@__45>GBUwvT(YJ9dcEjuWec@C5d*N8e-MnGBw7A)~@R%9)5
znj^a+T?}xOck0j#dI;TYt2xbhLW`J^?r!2Vg#03?WDOy?GVB(V3?GN4SH$uVc5?*{
zyBoUQ-2@tG*Ptg^lhbsQTEnjaM%hTA;B@6&UU}NM2-1fC%b|MfB{lt#Na+Q#2$jZ#
zy4rqt!7k#RF*6U9Tftf(5~cPFZFbEX^i!?uSQI3ZMFIDS<<C>_1T;&cYtVH&a3UBw
z9dVkyh`;Y1Cr<NgAOyd}vLv(LvURw}iPL?{wjv>bWsO_bwE9?@B(Ebaj~*3YzhuxR
z=ys12r}>>?25O9WhTL12ff|zXUbKKNB1IfJFC|p$p}Ty_E-j%3?2?m|xRWJb#ICWg
zxd=QNvwYXv1N`i2K)qBw<CxR-83fazR#cE{u1DL6Z*QlmuujOFq@n*Wk9!k~(@x*~
zbwbxfwHYoF>Xxn3JRDuiZ>A=lw-0!L1<h?-sEk;gGBFf%k^(&FzI~Xa)jl-$p%=0H
zM-F);poB`rkU!zI%7@B~<tZaCEy9S*op9f3x2{;8wtgBjm2@P{Ri2m}ZXY#I7V@F-
zxSAJQ79lXCXUQ-DpeixsdgUFFD7dVAXwDjf`tQ&OH1BglB_t4=0|`gpu|VlniNu34
zM#plKePIf^-$`?~nOk;w*1kj6eTd~(Bo6)5Af8svX?h4RLQ}fqiPL$zYcn7eu)H3#
zh)r2zh4QR0Nq3DE>V|w>+;y*#G#3M|t4F1I*#ox7w{mDsupNmn50zWC?f4!vqYf=D
zU+3>X1xLvlp3!%Kc%BB#H^VjL5g#;9a_JsD>4luxM6EkzD7SHkCe>S3JSV+Vy<No_
zGU2%W+I!~YP`PE*ZxbJy0-wvB>*(<{!<_@yI)h!^q6%6GNFP)#+|m>3t)LBALGya<
z(&J5kVCr;S*dtcTp*iEYOLv~2lAz(jeI!!-7zdbw8b420tA!<~J&TSH^$S~vuG;~#
zR5a<neI><N(4OLin?~g<PHc@`O1PjEble)$<Bc8~`$$qent9Uwp;{%Lcf_L!*Fai{
zcaw{$c#h2+@e%t-;%OFh#N&a*Zy#tvxs7%0XK^yBw=sf=2Te0iJUiTu;mzPBP#HC8
zd}4whcFv@Ed$4q@c;~4@3jV|Ku&JF3-ncUdJe?t-gk%cZZm+b@9j{^a+bf&ZHQEYu
zPH;<HBa4P^EI9TpiMNoqNuVMnI+h6%1b;B;L35{8JX_I9dW)KUpfLmz8jT#sn0O26
zvJ91-<DK!i1doQjv+M_Y%!32>Z49UNq<WuBLbtUk`RS}|u4NorZ?(s}*<s9O@h#>V
zgSc;{m^caJ0o)kR`5noeobEc}h2=JoG!J5$-P*Tov+p>m-m;C0kB3G!87PjFN$ZDl
z3oqz}_-7~V2`5*<66&R^@1&tB3JC>9jgirOHDnYRIo`VND2NBsK2&aNruCJi`*w`r
z`dRPof`)-{awsL-@e)WpcA)1xs7Z4OPUeE7AGku$)yQN5H2Wro8##3;e!)R6?8m3J
z#Z{M}X3T5SdD|Mtq4Ut4bhaHIdCfbQ(Bn(!^j3?H{Drr-QZ&TLjin7;8Y>fXCh6`a
zhWqo%iF;Z4rMF{xoOhb%Z;Or*`Le_^l3CrHN^MC)ITvDIW_Q!W5U4dqJfj~9u`gTE
zq_@^yThFp7B?y1nB%vf6z2%5!ZVM<!n@MlCCMSbQZ)1y7wV?Ahwq(>xZ)?JFPSABB
z=OGDiW1EwNq_;79k3DG|K0(hm_MtI0gj$}O?;%AU!IJOG*~#h*`xV%ew%omgu1h_J
z(4m^Jcj@^WQcMsLSaJ|^6(po@z8SrJBCSr)`{S90a;<qIKs+#RPVdxEpq`5==<O-j
z3s%zGRWtV_>8*C7QzyJ#P!CxiG#Vp1W?3VtT)K*0x|32{yqH{D36NRmg*EB9wi&c1
z?W_Sw*QLnMq_+#B<(&NqcQ3bEO*)zOL2q05nO%5$%e9w$4=K}#XF2L0=pN(JJ7C+L
zw}Re|X_@5@bj^Ng4C9H#n2$Ct+*9W_G}W)=YX}oj#B;Y_+AlQ`J|Aiye$BTtZoQ^S
zj`mP{FsTnE4dA8g(zW|bS4-KEf$xBO>jPzIeJPPW6Y7OaWj|DVDe_V+`Q?U%DFo0g
zYkfmx{0pVF`3+mI^MzXPXy?*K#I07xLx;bRhf0%|+Jax+8c2ZH9=bd*X3(Bg>SqK#
zP)2JH-F?g#>Yel-XqI*PrT~XPc)tYox%2P`^c-N{wBiQ;QV*Z<O*vhe5AAi`$@PPZ
z)LV8YCHndwsEUk-&biA=mq-13sO<GpA*;qqRohGX%Iib5>VDzpy9>Apg79O4Uy6bK
zmaVINSG&1<U)vw3r-b;XeboI@@(I2*Agbm=XKV5Yx_f!)axaB%N@k_;($%#;bhpv|
zK=+t$4LZjDdOw?+b+gu&uBQTSa6KdDx11hk0<!5t`wyY?fotvkP{MM(6^t$8?Z6x&
zZ@p%ZmABsIVTQjc5w!k5_ttj|a6fygvJ>~xrThEk_yK2aFZFmn-?YvmZw(Fe=B1x?
z?*mWX--l{_;h631!2Gd1e4sRuhpyhsAE;-t{iZVeGZ#pI(De`0lhS=t?PI=WNf_~6
zmJ|r{TV!r>Ak5KUDpTGX%-$~TrMmiCgV`9XKGduUeQOvZFPvS5zVxFLeBcRn_)R@y
z+5=@TdFbw2|3KZp<eTc$XUMS&^^ur_?ksgYw0e%vZ%WvQKhRYliAns<`V2V%JnL!?
z9iEc$n{w0ntzr66={H^NSvZK|*^53@Mt`4Sm;MfoEvWj|YEq@Wwdx*^$R$m}cx2}E
zUCH&*HJ?R_TnSU&dQA*|D+Zfv39G+!l=ZC`K-bcj66o@+VQzVboU945>qEJY@CUl~
zcx%{gzmK-}R(+@reO@pI;v<Ey4|VH9YcKT=RLgj4Ac*5jyYKTnRO+)%5$3b5w+8Hr
z{?IvFduv#=&A055_6Q7OefDRaB0oVNZw-6)M;#+cz$kAGDo+Ij=6^;^&PmueG}GkK
z#fWb|#~-M?)7&G<m$zO*`@CJSMP|J<5W!%+l*r%@U0voMsLFmvLFTS6-5%cVp|a(r
z+O@njsA~UyJy?m)JjY7ZXPqJjXnA29kANlyXnEEtqJY-&($VUp+Hb$)DGw#T=3B$^
zBdv$7S5YM#Xxndpi32?zy2m>oJwolHjS*0GJW`WLucNn@?(s-XaxV;hX)CWnM&Qv$
z(IiOd+%I&G)NeY*dOOQ@!CL{bSLC5)zdj2U@f+n2UBH*+rQ7{D9_kr4z9}(1)=Rg3
z54mC~>bKC;;%5=<rAzg#q4m<<vQ5jk&{X1%mG6a03Hnv*Tf>n0Hbyk{w}Pqf&)0nP
z5GBM>`&R9t@+^N^JW}La!`5e(D}iD9_ZO@LRe8R4sElu+sYLI1CK*(RzBSmtwwJO2
z)m|!dztnp0Ei|FJ#!E?q@z$`ayiA^Hu6}j!TepX*yhZoi^&LhvS^hSLbHsO9_ZKaH
zpsL?WrD7=RdTFcrE#jfOw3ia)W<HcF7yYHHJ`<X_qV-!HgYGJC4QsS-6IO{W+WK1q
zmmOca+w*>!UtoHmjz=tz&Y|xYuJNrIDES=5GvXw47~jg6lG0)5TW@#QfQORG?xCa2
zZ)Hq9_jY|B0%z8juCnB<!StB&QnqC4OSgP4Q%ZUVd4v-Q9(MZ!UEgc1lHy@MA`??l
zUs%T@o-q`4=}VV>FH=gS(elVlLXDPhWlV`RTJ~GR^b^g8l4anbt5<pZYZiZbYasGd
ze<+z@-WtyBzi-Ok_&a1=aQ%J}WPa%TrIi5BFO@OsLtB3ynCuVlJSY9bvrv-$VLs~=
z`5)$fD`0<ry;Rp9y0APH$v2fM<*k8WdM_pY%zUXy^Ox$}-!`n031UBEPAZ7~2+if8
z^6N`g$6G_2&y*(K=!<sgZ=GX2LqM$2@@>FESFJByt&Ts?U7uBoXrryXeYGX#5xPVh
zZR;(o9+2vx#F_Y_<b~+Z6chIp8!z=x6yLP9G2R-?0G5~bN4k2hu;)dP9%8<oHO=M`
zZRCeo@9)=GZ&NmtAH{+c5mH{dn>-5|Ng`yu{n8=N^de2feg>R85qrM0w*AO=&qP$;
zvTf+M`Ks87Du19-zRg!9-e~E6p!nB6bk1jR2sipjJED!&_YnkN;+JZBp?&%NZHlZ9
z@7O9XlwXusqy1T<NF320wSdeK{TXpmM~r90$sIAqOM4&k*3fm9myS8V_xw~n1>GO0
z%x~*eNhh!#wSb8JYrgd7`RRowY#wn*D0+E&pk3Yp)35cZ6+)-)FQpM$>z;2}lXd&0
zt!w=QRsC9^pd0|NhY|pKymXf@c?q4Pte37*^bgd7-h5MO^-Eql-F3fok5=Crw6yUD
zy4w04NsqDlrPVwAK3E3UZ^IGTO5bm-rWn<i_BO_T=^paZF=mm6+Qk1uALg3trK`<n
z%sqHdd#Q}?!&4~nL*BA`eV^JvwSDiAI^Ex=MXmV#yYf=fv;Kjuy&l-wGv?Tc?>xsw
zd<Fp%@!R}$Dy6&?ke24{0Z~8SvTGKVhni4mZw>fFygl2!?mtj%^`$!7_<m`G9&Pdm
zD!ufF*53OcsP@$!+Gf`Wc8?G9&~L{M;ONiXCmqH2+6SHUPi-yFCi+mB`qnFbHF>G7
z`BGCB_Dj`d`Fg3mk48cYk^WNMkJKa~#qKX{b+12A`Sw|X_O9&@RCT^|Ox^XN(<53v
zbbs$JK-YN05V1##>PuH`FWs%qx7zJlfWN7F^6<aX=~wapKv#Q~DguyxJFPg~y}z`L
z{tP-fDZ2iFu5YIm=xAfU^^X1Rv;v);@8Fvf(6_#HO?j!bRsKNtdPbeJ6l=Z|p|xMS
z9vMwq3VFWQ6&Y8)X&t+)hqki&Tb3xeZ&_0+A2E%o`0cm?9Yg0!r(eN(sFY`|A`WR8
zf1q04?#is?9UMX{*O$H$gD|9zB0v<<_E4gbw(-(h+oK*3hP0QLHV<R-P|sohP5b=z
zT!9iE`VUmTJy)QUJkKxHS;s@yoX?Ppr#$!r^-Q?mbobT&Ky|$xQ}UaB(e2?Y{y;_M
z18aYVoYWNinINR5*w0!;YKpzy8oFmO`lh36d+R01+j{7nx?f71?my67#`|b8%A+0-
zkF=IQ&{dxyCmv~Cf1tZQYZVbm+w%nzlC;hGQu4(-^h6u~hVL5c$pqzrcw+9CuF+m<
zpWl3IAc5aomP90PSt)a@m#+EDF>+I^^475T>Tg-XJ$*f}r||mrTyeUjz0`!r`~%(c
zy;6BLr(Iv#%YM7iO?9dd{dNm#Cj-wP=o~s<ic9Q%sg(Ls>E&BFxY<R%>7L(?E6vkg
z)NcjfGey*IJJz!-RQ&_hZ^xD9k^9^CI(35Hf7`Jq3j8zV#6a)k4^+Fnbj(rSdS@Mf
zfbu;_PrO`xq0ahJne%;GT$AOc8SUdQRQu|0S=Dd9m3Cs6d@ob}7kNgU3>7k8+VWQY
zU#afzWvY^)Vm|{$hKenJprYTZLB(9u{s*eM>O)OM)3;t}f2dY=d7ptq2>MWclzZTM
zo|c%T^%?Q*(J5YPdcn71P`l@B-!J`f#pFYK>)%E!bocT0jARaP4YPc^tvJOc{h_Ok
zZzC4EhrU!EKVs-E^Lv$oddQe>s_ol9%jsj(RT5RKZ!@+Iznr{Zx}?1nZ+&^`mi^K`
z_P1(K$yBk&AE>9L{iYLNl9z7n@3WCt<E2vGX-;s`(q77)gZ5G_&!7{Wv_3PP;G}i@
zfl7UZCc#PTw;c=B)?V7?GyjQA+Mbn)*rdI_bo4&n8pg8|44w1)JPf+t9Z0B*XQd)v
z#rJWvQ=RRld%hBVdZf?(Qd!SRMZStP{y=p-+5s6Wq`b7%N5qn{Le^Vt@AAF<2i5%+
zJ?WRf>8$TW$5zzxR?8KZ^-_K8j=DVX^?c}VrM<0@Ug|ANda3XCLzvR~ZN)-Y{nqa*
zVM^bDBuweMs1T;~Te<!}fBjqe^&h{=uYdj9fBO~x`+xk;YOwsOzy6)emGS3)|IeTQ
z#ZkqLP~*S+*RTKSpQbeZD)IlH|Nh_qGYkDMe*fq0{4c8gOIPzZ|HU)^>pb|s5lF+d
zh5s*#|BFBV!~b&r=b2TT7clLWp|U^!^`qnoqGl=!&Hm?)zjJZCo?t<w@+8Pn6v@B-
zC_yMl#k!6n_wPT-a9zOn2BOUU`;Rgeo3RtjBD-_J&r&yWa+bQ_XQ>Rsn}zc4Kg-;k
zTv_OXAEiE<dQs{FewHHnJh4);Q2zaA8G@jrLfk?Z{4B+7TCS^Sq5S*Lp67ew%-_RD
z|9pAscZJSn{d`?N3tsl4+<j+13x42_lK0L1Ect;yN`7A7U%L6BKg)j7VgD@qp+C#!
zyZlSh3xAfpFZ3@(KlDf0&ujfz_6Pl0Hs9`FieC7$<bB0|Df*#5%jTQ@OVJB|lzb%L
z|K|7Zk|qOUSmhocBG>r{l0lj22E+VWBvF*W7jO_a@JEr$coq)ucl{ZFgNMOstKF^3
zXz^3_N4uY*KU)5j{Mr7e=+7QJW0d0cKjR73C-VA_djCi%|35x8*uVbY-oStTMNpqO
zlfV9*&?Em_^8Jco`@{bBv%h!Ah)T20pRYU1`1dUTaJy0I-?e+h!#~_^RQh-A9%1tj
zw;PrIUAsqe{nPLFRQk8=x)l7UUvSp?w=KI+o&R*pQS0Bf>@xbFe#u$u-?fZO=|9^l
zHU4dj&EuT^)2&9Wf7`OJL;v{mMy-F_vM+1@^v9I7{$0zbMPGmZsdl}_f4^1!&s#l9
z*+1Ltr_TR*yS^^UKilr7&S!aj7S2DfuivY%t*QT)TX1gV(9-|=AJOFh@vo0EO4Q1R
zul><b$&uNCJCo)STE3F@fH;z}PB0lD8K8ho^@J-TwtVNDnkbGDFA1irJCjjPKyV_<
z<F-Ul#i^+=3yGGAk|^thDqfye+&(&waoDoqP1<AeRMy$Sts-8`Y+cFvbaH4%$iN=?
zR0t>UA?wBSzpe{~gK1*;WoT}K$#=AW@%(K4Le~hzlj1~&xD-S<1@XwqC;lO^<SWJ`
z;qN@nSwO#MPDr~QLBWV6KRA;1fOI{3Gz$+WenmvGow+!WCUD&DC4Wuc42R8={s!EH
z-W0#Fe?kcoMIOI2!m~-A*~`rhGD52@+4*Dl5+hjJir8z*)3oegCjeSlsh)45I2<F*
zo(%6!M6%va6t8sxW;!I+_l1kp280O9$c6-uZ$vC6Mws^d^R{udAi|f&IzO;`{zoEr
zt%xW}n0Vr_17$?mO`zW?C%{&lClnm9;~Rk!2#=Mq-+`OcXQE^6Na;%G_`%^bv8`6b
zmm^5L-zOK(YudA<8BA6cA*eC}4uRtvN6*BXB3fX=E%pisAmu@O4^(^s5xFG7b;iA)
z?d4{qrUETaoy9Y{oB)ADjBgyq5sWF~`NB3w&cs(LYh=OZcgQ`IQpAuVF1&|DBtBA)
zFqc5jqE2w5$bm<!_|7>Tp^Y-}I#4-m77t!~Vtx@Sev+__$V8d$lu+>=p^vyiBi{tm
zkux|#1(n{jA?aikASTc%BTI^39rxfrIh43?HKwq-Q(gXCE9Y;-=<$R^0R=HYj2=%K
z3&ekm1RVs}xrE(qV(Z`n8LUona}XG(M+zmv!&fi2Bw|j3g!4}ibaRX`BEz%wOFSEi
zXtXf0E&n1MO$5LrD!ktmCmhXcConZp;Z5>-!BJMph9?<ve|apKeS5PyQC7F7V_X-K
zn-#b{GsS^oqSv9Wh`dQic(Q{2+9%TKOejP`*q+~SX5Y1Q(neU89U)DDenH$LutXL_
z;=z*z5lGI~g!aV4p{EQBM&>;jnd4f(*437h?O`%Q6mXHpW`M^%iHQ6Jf+trRu$RaZ
zLmYV0i~&a`RRrRDx-+}^VG}_?Y?2+hOo;YQawy<-Z;*hbRRWr1YS~Hqw(VC$5Lq>U
zEuPSa_#r!@{1O0u5R{+bA0@(F6Y0I+B7v>cBJBW>Of0}TvQAvGh<QLbkIeoK+?=Wr
z&10uj#vzM{2L$fOSkXjzC;7xLVsS)tNSG&XAm&Doj1~lbZycx*T*EJU2lRXa#MYSm
z_Gpkmju;vep+gDp&cAxBj2f#BS9azpXtHr+Cl^PD#JMP~X2C^BaUju`vlZcL?)IBY
z_hB~CtbrcjDVSgWCTvBo`7BBZWtd5|l0Vk%vzPuy=11bZ56;gBLqT%sfP~idw%3R#
zzd(Y<0*Q~e12#gg{_?mI5l9=jIXfeOgh3712f5ja6*07B$>CSNi4_r9XJN#cgtg3q
zLWK9uu^GW1Mr4-+st-D`;v>T@@!b)x#CXVzGembMq8<?Ol3lmYsC!ANk?e%n?xY0p
zNDgge!R;Z?2+ly@_JBG$E%S&C^W;h(uzMu{E&&^M3VB?uI4mPHgG9DU0=t{+^^zv>
zV*cW4#Wf~^FnDUS1MMlq;qa)c2X^A~0J))<CA&wq@mLFyBayi71ch;^sW{<f#SAB=
zf<#g`qPmkS8n|a<2S7je>)pgo7&WV01i$#@krR5&g3^g$K!}9WPheCa=c2IPXBON9
znDp=nkq(}4?p*Q*c9OIw^nASpLNH-+hKP$eL@xkpF1YF;>;31ZmB(ms4adGj;2IGu
z8-1O-z(5mTHGkf?MC(vz$VTqqunl;k9h5BDJvAv|4Mt=QCX%}k7f}r|P%1V+MOv$;
z#ySD&f!wtN+L~JP=ZkA<z!qUgu@(yFVII(+_ADrVH_GD~=!N7Nt-x_-Hr`akD*$T0
zgQSDAFya<WJ&#VT_UO<Gg$Tq23THb4S!0(l*@;;&+tFxXIPC+?96Lgb(CzN;4h*Fl
z`vpgFn2ty=0rU&p1Sa6*I^e)Eak%Qd(64V3mY_x8VWPI<4(o82o((NrNrA2J6S5b$
zNGkx;$hJbzcKjECeV$JehTk7-L_}+BDqQBAt+}#8>bJJl^B>EFi-ZEeNQ8&+<yW=|
zQP3jAJ&?%;+ado6*6#PU^`t>w0UM#B_2k~of{mLg>PcKBE9|&HIE=i;u<e`Jo}ckV
z2*<)3L<`q*J!J-*oL2wx7zS%(m2+_?OFdsoc4u`mWT+>^s>zfT0x-!&WR`Qx(^>@C
z7tgX={e+@PeBVGJMK19SYNS*V&r>OWLSZW1$e)|xSsY?5o}ZrtQ@$TT%-a@_D<`L9
zB)fr$eMK-^5sr?8W!lLb0F=Hj<m={iBw|hzlzsMd6g6kfmWO|G1k4*JY&T5Kvq+EB
z;{;@P^$RF%U8p3)V9TS14KH4M%QJ_36z<9RulJ4|20*`yO{9cY<IQk+hKK{e8L3`@
zp7tK7?LLi8QUwq`p~;clSk#p*5Ra*FjA)pk3Bz>kENJSU4YY?@h}X@f8K1dCX^%u|
znBx;I5vUQIm@w>hBj*U+-jD7K6ygj3Gus!2J39%moxdG~fKk796XS!GrQ*fN7MJk!
zultG{WD{M0xC~d$bl;6phnb?BeSm`Q1)L`ZfbR{k@##?=KF@r{o}dopqRkNpC{=DY
z5eDmldRKSCN-jT#dz>{9xUiW#AVg&9X1}44@B-M|(FTB<d;q|iy~NvzcmhCTBn2FW
za02T2$Y%cB2D$p6NsG|e1Y7re*1!=dKY*So3AiFb9UB|N+m?NlLL#acmSI9pl#%tx
zqHC~BTj<*#XT(Xpn{8|xULRf5-y93lB>7O{*mX*>PZAS%YoyKqn#lE^lXx?s&CTvV
z*A|J|fSVKmz!4S*;njznS%s6}2-y8<HBi>dI{PC2KTu;kjOgl(&;dlnh!i0(Pl&D_
zYH{SR$?()1Ms6a)sC$A1*oagjKq0dQaK;83yE$%k10@n20Bs3!wj;DVfz_Ff>Nb2$
zrmitC>AL1`mCMwy8ZQhtCc*{4+K7R~i>aLo?U7@Z!-<Iq;vV&*?S!^0FYivBXEPKg
zs6Ea!-Xg{bvHsN^{8qNp{Sq~hlwtwLBus#McDj;3Cv$kfeVGvEAEsfn`NkAs{x#!O
z22`T_J6s|Aa9x_+hrC5ReBZ(H=?k0p8O?N3AIjDunJU{_eskJk*ZjGXP{WyoKorkJ
z=&a|W#swESt@M0fTyKef_j$|NfG<5TGSmW<xBw0#xt^a_5ZfPUvM!-1&vOL~9fXT5
zjG)7u`WQ@lW`@BK7?@Z@_CK&?>uYNCbCj74Nh7eocS+-e5d3$0kAB`R?LpQGhwDBj
z>A>UJN6&te#ViIX@*1$E*Vydx5!QH-7+m!{dP`kmGe>K@o-hRoj$VjX?{7EEYB@&O
zVIx$nO_CNpRo2M%3*7NIDN;8(z*A$g-KNbX+&iY_v0Z?R5CCWXuB;Kwd&W#3`83ND
z7oTrJ6PhJ{zIrp3YBsd}X0*e4OPa#C`uX@qto__y;a;?Vx?w5!4kbbS{UhJ;Cb_-B
z)qjU2Gy=@Jb;Mn5v3(Ep<d7~*S{7_<1RkLiiM4_IJP-NVRxmddkYT-LeWQd(mB@_l
zyP*5Xx|P}+CoKC6Z<(p8VO?R@xJ)fA;Tjsvq)LEwrAOT5SHcD=yB99ZQ;aB1xjn>w
z_z41)*x0^6XlQ?&<PeC|my&gU&Q`16e+K#vYCx$+3)1{X35MR;#`iTVcJY9oG2U0V
z))F>9neuf|j>KWNmNn1e4s#C}C{uD01)W>Zuo)Xg+bzViu68++l8xRO3fLwpS-?fa
zFkk1hIx~XCKI%YZ3nKP;=LiufVI~0mx-)RbZamw>(^YHsyY)Uao1iQ7liLva1?6Zm
zNADo&d0(LivHg`$iS6PMv?7ZW+sxB#CN!pUCK~(9)mOpp;h>U<>+o#V$6O)kxsMC#
zT6U6~C?r;g$?Xq%MLoJ+Od$5CD<Sz^D~!zEK=vwNHPlD;t@}};nXK9Gg;Ooe_tg&*
zzt`+4YCnR?_qWSoy?xk58wb|d5n<2U<gSmStr_uvo*Z2LHJsda0U3g$Tc14|Hpd(L
zr?57T*ahx-2A9K;7SNQ4>SJaq^HCkyG56$kuIWc6L*OC;Kd|?a4Gq7`46NmBQ~)>o
zTpr9VcC6TN_nXXad_-P5Hrz`;YIF4q;{n$Qi=^h8;Ks=~YID(YJ1_f8Y7V(_I85C=
zS~JO>^VdphNSLS347QFrtg$e$gKuKj9VAUO#AnPQCYP5z;9ZHQ;i}0x2A_QdXTc$G
zxnEs&D^R23Tr`n-%*B<+Wa#lMw9ox4Yrsvs{bOs!yXB6_KAweUE6gluM*A^+IUvcl
z?ZY|#k|vF3c76Gk<%j^7t*@=enqG2NJEE8mGRt}ZLAejUi%(!o?8s3D6vFIdi)+0<
z*UlEGdRnud-Bc64dQADD=?%=mI;ey)!RZ6X!R!J^Y?HvSczwHIpM&b{YQ^^YI5NqC
zO-Uia7KDrX18u0rfh#tlembgl-?a!R{rC^sj4$cuIy>njd;=tQeSq4Jmnk1iuw}CM
zY@fvbBu5-|#LmkO#SAFnb{Z#toO*OZ>VsOz-q_1JLs51PI5$YRm8|aN8)J5kM|{A~
zu2sOZS7kbFJPJo9tNXqb8w@j{o-I1|8<RESR=^^`8;2+rmj<?~a+%~{TXIXQB^w6)
zx8$B136%)?Y>4G{3db#TpB+OIxMnj8A>lA`u)$cReBVuC4}q}c?pgsHkxUDyxHPCI
zrr6oep1N~;PjYn3LaJVJhKUT@><q(5vg>ep>Uf7k=24zF!YSm20D5NcfbMYNn~-vh
z4-R%3LGwMVYUJ4jvb_!KVPJ9EGe~m<EOTb#p4pcDPl@-yaemuppER<^bN<u%*^sqa
z8<J*`4CQPYS`SCUHE?P0=&>yu5x;QE&aD~FStVNsa<+5Z?aub=#(H)~HQ{EPi7jFc
zdymtjL704xbGu@IxD<j%1F(h92XJxF&aU&0DanlxLgd3nswZ3sAws_IS?5S{&V<bK
zh`bV_K{!xV-!7MtE1N-+tp}Fa4(9O5*DyAk!=C}RiH#4uAB1pZT^&x}n>%okzB;z$
zcm5px#rNB1p0IFC)d-BwuKr%PA@Tdgz!6DvIlLOkYmaf<F4RKy2aXT!95g6sp}@$Y
z&DoV3sj$p~3se^crh%>=L%b0^WuBvp6>n|-|E#@BvL(5WW!cYB_ypn!|DO>=yQCU5
zfNW$nGD;*`IsM|=0C=t)-tf$Zi@fJ<ww#BmN~Mw@32ZQMRXgOs9V_6>*2HIO^QXk)
zf}>Jk_Tv!XE_NhABK7$BnWycHgtMjWuxI%+pdaR&EkH#S1H*HPqpw<Z47E#GdpZuY
z<DC`u+F4=smUM=-d>5X>uq?L*X<^4j)){-n$lw^avnv7fz?$#VNS%f=sO`D2I5q1C
z-~+pxl(<(CMhvqZ6ws7Y1BY6^FUz4><<$f0jQR?B>g`C(*Vr7Sb#(8SPN&>7;vh{v
z4g6u-(u_aPY4?avgEMrco1Mpbsz%^O@*Wr|i}eh+CH*|kqXo#!v`oYxE;AK2>}P@T
z*m&&Np<$~#s$3``>mad}D7d?w51u*RR`T?PLf<t3cWVqCw<V6;S=lz*KBj*$nA#zx
zzij$!Lg-#v@$Ocj-YuU7me6L2@EIsv80H)4uqBrUMCsdzA2-Dc16ykW&!VI2_;&h{
zE^M&1E%|#te9#dbjRo{Qa~;9Sp<yX0RTMk}@if@Vh$<J!w*-@CU<4{enD1xg+iaU5
zL6ZLIw#8{Ij@;Sb@Eog*&&^JL&kSH+JR^Rxhry*L2g7V#k_1V?Y(UUxFm0^~T&!+&
z5N=(CiAtLhu<Bof=02i=HR^a)Xq~`u{H<gN^}=4JAGk+qL*N;FhmPiFj_V8s1S{=A
zx;Hcv8`%t)%d_(W2iY1lip~|!y$a_^KsmoTT<*~@N52=}Z4ZM=oepG?MXCk{_El`c
z1x_p$V9PcU>TnHn78pP=z|0=q5A`t)S&t~Y#kBE4r)bo|+#``jhtu2{TJ<#>wr+KI
z28^G$nI<nsVE1gcnACU*3Z@;GvBfd7=9qEQW4hNwVH9vU&G~CeemmHfF9Yr##b#O-
zC+3EP*NNu|)fxz238r=>rE_R5F_VoOGwfJ@RrxZIt>mIJ>Sm$p``N7gkXMtZ%69Kw
zEiuulH}{nhC5vKxfoa+Dfdy_gZya-Ubf!RGAQ@vEV%=aQxqF&c`Uup^aLRGvGu(GE
zv=HJNqpIVQ5@xmwH(Q%Pp1|Rh;f^IYYdrN*c&{g)X#sUqup}xKceoYXD|2zqxXG=3
zYhZGxMQyLK34yVRucyPz7V2MSmU@%3d&zPb19tkci){}tHF=g&oWaVznEaad=R#H6
zOF*TkF^DRf!R!VyaIv_-&>s83j<cBz@)%hf2}g$ftQMly&RD$)Hn(C>`DSz3JhEbx
zY+4+{$ZQ+GUMx!lvv2&M<&AAGAFtUza3IP9tafIDy!RP*#m*|%M~lw+`qo``WyCO;
zF?R&h^iI^-aqfc6{Xb!6<Td6?afH1Kk{7J2R<?a3ATxnr-8QG9>&=KNe>XI59d<?~
z!G15(2e$#s_1+IMbaIS4L(qN3Gjwi)C3-df$jkiimM<Yt?oM{&XO3yn7K>jCCg0pb
z%)5|Tfi>&MwxtIo-5)Hw`Gt~qqdP1)Zu-D9JP)~+#{IIXj-h1GKIr2*$06R5<hxG~
zgKQ$pV0Ir9bi8{u$HOt)D+w9z%rT16kPWjOfpil&Gp=T94OF}<ah>vT^erpk=KU$-
zS+4OwqcgiNi5d2&bMj8L^Wg3^u*vBsdlYyGIYjwS@okmk@qp*02$mhw5I%Z%EG)$Y
z`yj>uYtK`ZxZK$peN`}r!#v8}6?J)F86`^QGyV#uFr?A#mf8Za`7=(Ehb(g;P1#@2
z)ME@en+`u}L$$jSeF5|Y*|5!4pf-?Zj%%Nx)0|pyV~^{MJXrzYM&07h$7G=j>oFHi
zvWA$G?L2Qet}{12(`<9ibf57|#$67z3@HK6%`N%ucBy$goE+CI%K>2>jz!ki#e3?u
z&jXX!DqD0ue6zyBH4#+&ycL=5g4Tfc_dGK>aFMVc*lgn*jqWr4ij^HE+>#%J!Se+1
z>ti5V^PVC1!mjaM9>?y11YB&)18a<{iyR?U@N^fqJurC7(3?nM*)0-;*mIw$-sdCi
z>T0gmdwTM31^F-tD1BtUKR<8JT&+2Uu{+uv*_f;KLOd=Yo6N8j10EsO&0MXIiRaS<
z%MpD3v>Uuqz``nnZDXY|HD)7(Vj^r`2X+SFEtuSS3Po;xw`^ScrWtJU-<M)9M!?e_
zm`~W;_P8Z<w-cK2z~;djz`-=X<N$T={FuW-u(BZ&a1r1Pj9q%9xX%bI7=37y3&cFx
zk1e#~5%<7@44B;aJ0r!76QbPX!w3*vd`dbO@TDj;Yp;b9goxdNa)PQRIiJ4Wf#@V7
zf{q}nC#Xk?yP7NiR!V??XAPT0i@VT5r7Wt$j&?6)qch20J@Gbd;@+^dn^<n0gYzlQ
zRCHJxaC6A0M6<r50X-wcm=<5)9|Cl#3q#vX_XgI)K}TtA=eW0e77E?M7RD+cx>1UU
zo-xGl;jFuQvu9{)M-4<vG<q~AP07=(LnodN(3|!i8SaWpGAC8j7lWd^SEIG;74&??
zL#=pb?0v6x?qD0md*#ye`KrxI9>IFGfC{1#F55$mcs|^63y!Xi%?{yim)ac34q^KI
z{><Q3K`Oq54m~}JsfN{Rt@r-dCdABgCeKTE=Qw-GS2NnX;7Pw!!y4`YDLV0cP~gh0
zCFog9HBsbo+NB%oc3GFmK~G;vigTIS{S%>Qrmr3)GxDvM(3&b*=cnG*Sd4S_y7#*D
zj7&u}^pWShbZ`ow{MGjxE5yPr1@ko0SzXGR9<YTlO8J+P^ulsV_vb5FE7s|U&4Pnm
z2+-=jYEUy)@J$};7Twy_(FygAcV6|q8!dST%<`h=gqy&SOGmR<p)(Q9TR|zjRAmwO
z%~jEbr;tw5Z~tK8!<QE-_aCQCJGGqA@9C-8q2t)~q9c9k#p3D~I#oXp(G|QlB~n@L
z)y_1q6nllV+O2mcHrb)M)LV9H^IUEQzD9taeoNe%a<q>RJq;!wg-_G$MlJ-L8AkUy
zv6z-3$Jk2R7mlTZeS44(@C;MBrS8i|u|*n$DL?h-1gzIMO2Fr0l`(@>GPsMjQEv^E
z+~=a%tu$y;=5U-lNIZY3J|Sqm-`z2Gm9bjEZ~1ezL9%CP36E=0B}$fe$sV^}TnyB%
zdDnXjQ_Fi`pZiw3$ZRWGO1VvQCblG+?i)CebgF`}SDle=srWKNvj-lC;!%ws&D`;U
zcD!z~(bl39bomwurWN0PCunXesBq1CFD26*R#&ulC;LZLoFaLIZqY`XJm|#FH7J2-
z3ST@8v~Sr?b~Xht?zwNhqYjII9HT~`Az*3H-fTv$&(%2k679Rz`wWX@RHA*2TLa55
zJukUeYS2v@B<PvjK25X9AS6nLc%y6uP~Fr{WwiJphHk5dGj+!Bqg{uhvZv|dD<ASi
z6U;L{XI6D2LBQ#2)zpGkQgw*7weH0li#1?wF{Tb8>4t@ot~77auBZWdosD&?9i$}?
z&3Dcngwu74PMsb;4xOmsi?-hG)h0snqB+m|WadSShumB-w+v$H2<#<oQ0w<<8{Kcw
zoW0sf=nj#+B&VJ6K~1c-VYFR4Yie^|6|5f>6f};3vD_9!A6T%pd4FF~))me9bFC`3
zcE$3pv4S>EwSldFD(I-8iso{E-v`l_xG1jlxh<VS0Nbp$Qp`3hnp@ba+@nTI4(m$3
z*E;X?ba<#Xa)Y8aD0)_0_ZBuneF|G%w|^Xp#+_&{J%@Z%l9~{-xEm7m{NL`ktaXd#
zfr|yOk96r7`HB@S*IcF(Z)>o;?^MXb4^=?ua1TD{8L6{sK#TX*dyLnnx0B?(@`u_M
zbhP-)`0lM_t>v=Ot<tlEk6Cx=O)Kzw-cOfcGm;<3Z`K`nq36@jM$-%<V`-CiDxQ3f
z&%JJi;6q!FZqJC!nn9ksE!qpc23#(tzMvQcH_pswq?hOK{e70@JM{(LpA$T-pO2l&
z{dvu-45590DyWmtN$BS_o{3RTU-&5Rb&De6uh@GcF>bYIW}a5`uB8;|ofJLa(}+)O
zV@z5^{&aU2;F&yzqVHQ9_xF>T5|&50aN*>E*Y)cA-s#S4H$ypOxKz$F&=MmpI@^_S
zI6S;{CKq%|T7r$@;mbC3wmV-o^b9Mv?0MZ=EnAX%PKi#ULg)A_B+%K`(G#$FfTyTz
z<f*y97ilPM59vz~8XV&ZO4~&O6U-*N-m<7;7>zj(q3tG#i6NJ`8k9^5<uCJ__tyv=
zID$c2nfGG&OJ-5uvJfN|2LF8DE`)6ydgQb36DDffIIY5p>ktUi`Q`9`pNTj)=}dI@
zS`CWlm+3#m6qqIT!L-Dlb#+4zEo~jG5^}KGr`|f4HKy`4)?4<>G)p&&DRA%hFjJdg
zgQ^ET)@?Ikika7?BS~+9P3{^^h_uK^RzrjjLeE5P&E3|F?xhg8p&EF}_s-6}Z!25<
zL<f3?k)l`AQWz7^A=L@F8)biDxjHS?P^I7-l#du#+}uZx2iK~pY?`YEECdVzM=ULX
zL(&t(m$(};^cZ5+;>Tjmx9qWzyGeemk$O-P(;Dp_#Rlz}M~Av&t>8o=0+{-To(X@d
zY4IWY(2W2IwC8be>@-W&lcnOWR?zdNhlEi0)IN0Lg%6#1-`m}4Cq?e`ey_IW@i4VX
zA0q-=fylrU`m;-PkShNVqhpL)mOWe5u$t>&Ku_PFg1Oz2Y+Y>N^aw~`?d&rv9LXb&
zppAV_VD<1G!O0vqwb8#W#c5ZU8Y><PH((AdfPH^zB|We!zr;Al7BUVTK)Lt?ox>iW
z-)RYX?K-TzyN3e_a>ONAV=;-kwu1biX~y>`SN5(ZY%52#_|fCX0qt&hWx4uiwd-l0
z7}p6@z>!07W5dl0)U~^`RnBs6mCwa$=C~Z%?^wl(5R)ES-4z7dsy)N$lKWn)ZF^rN
zv?u}luu4mJ2TkC>)nqCvq+cexdnz&`x>jS@OKMnh1m3d8HmtgS9(mldXJKKYJh{$W
zl;kv4LH==9XYMYdw&5wtxs^m#SaHX^#5Y*1wQ&<}mkyLFre0Bua~&g&=H}cr_Bo-8
zP%G$6F<US=Vb(y~*!LEQg|!+={oGg^&D4+XugR`sI!AMiTTZ7JQ?UT`!l4fsTHR+8
zIyqQSeB_9p)j0-N$+=>kkDzv8num_`feYt7V2x!bdfE*a4`}W&hcWZAP$Oz{$2Jsn
zAp8y;ei8=7_qL0+=-UIIvus80j4)WzHMQ!QJmxK0&)}`!_qppWTL(j~!b|VdvJNET
zep+~EzmynT(d^sG$Z0p!D-^Su$}KwNfDIih-?BscA<z~(gve<VgD6TM)}<UG0aN)@
zy9i1qjCQ4V;VC%Ow6(8K%W$+b<;D=xDt0_!{ff=)F+#BpDE7iR!)-xxN7D*A-9HJ+
zY^L@;lgIv0qCiB;4g&{mC$=l0$>ncEiAsRxHDh9&Yy?!joa{z&*5Yj#3okCw$p$sF
z)XyD^+D+Ume=NL72w<-!A{HyYLyu*-)U?Du->ab_W`biajeE5<?Tsfkf&(hg6w%dt
z*W0!m*GkbDi=B+`l1ZEGzG8@BZvagTXk9Vi<0M6(2=rqSFU8S`E}UU5#oJAEx<&}J
zatbNhXTD`yMw~@q<9GgG6^Q=<G|%XutOD`tChmN@bG9hHBBC>Ptup&ZXh}f(^v#Du
zXTdK7TIZ)3=0e{>LGJ5Q?;==5VsJaQ<1+hY*(KuU{@mF{A4rk&J8YTl#|6+9?a0=i
z-PV~&qUYmp$o4%n#jeVs+Giq0e?@j$pbS#QA+%faOvufOwpMSgl_LsKWsP*&8jfJ|
zM}z|O7+ZCi9Nvs?WD_E2eW{_z6!Z*cTQ8^ebc-IVq*w*@nnLIdc5CmJU4stF*|5r<
zuPrPIlud@UH<~K<d2tF;4K>1>L>JB?(CTq0xg^AclGRX>*HDsVw2UTqFcTft8H)#x
z!%DICI;+NZD=p}%0$Uib2a<D5FN_e{`+J>l19S0BaNxk>72${NqC(JKSpP{5VqUhi
zL@tt|D?vv?`~up-PA(knux$e^E#KQZD=?EkJf`)y6lYC6>#-T~z1rH^6yd-r-)hMk
zXZndfPAp5-UZ2qjiV&UJj%3-Ae6>*IgkOy&_hS&<+6Y7IqkBInoaRvkdb-!8RpIl{
z?okk;C_qS&Bi026n!;hm(cyDnX8^WJQ~o?c*FJ|1XIHS1+jIfRe6LQ(&=&0<c&=S^
zq2+pwf*VB)$%(}HQ1}_U6Vvn@@1=Iu0Kw9+?^||D!xI`dZjtcG)U8g}f)U)fkpM&F
z5K7)Yqaj5D#ZAH}%bQCl5t>t?bkVtr{Ot8!#*E<vy3xj%IwTzd&tiO81<jT6pyw53
zqj3NUoH~b~g<Y2jipZp1Ob-zcs9NP?^FXBh$k}xL6k7#}j47eL&yB7#a2u`ekPJQV
zKkF5o1yiIIK);dLTuIEta9T9UPTj&8_o88$yzwf5&f>Nd6afj-CiQKxf{J6Y)^4rD
zcSQUlaTd@r%e@$(N_w>|+=HR5;QWV5&Yt~Vta02T%E4UFj#irJh`pwQA&Fn|lzdQ(
zQJAOfE<@{D_kQsWHCkG^_dETCR}adENf2b(LH&hivbOGBmP=lWYQGOXpx8nyggE20
zKC_GH2Swjhw7M88G|OiJT4McCaQB+_edDpwFCD07AXCI?Y0!GcZq;xl8br^NtHsaI
znJ;Z^S&Qf1ZUy}e-Ui#}TBNYF-<L7(@9`Qkns2vb9f-5ZG>?M&46`Te&gmWy6}AhI
z(U=-U>usCmjX)dUqPr#=BsquKm-=USs)_c}PKurY$ATI#XI2@|)ZGw*$`xZO!pTj)
zg3x2BM&~ZKY&nCsBz@0o_S1%f0wMk;&>km%1+hoM=cmNAK4sfPR0CS(r{43?GA|Kk
z-J;vFgU2#Mx>k8yI<RacE4)@r?d&3Qv<{u)(+qPmML^?#_`t=tAxI?LrJF5J=sny#
z^bG8bmF%F#USPrPm-F_iV6Esj<6m;C0j3=D5)Q^=salNfTlB2wliCqizgyPSZ<C_a
zO2Dc-CR2_b=~h4(qS3lzZpFVEoeimod5oSYYOkX5DR;``WHjBumE|CI0peua)gnbF
zjxo>`N9u*;;C9XDalc<&)A>NB#H9PZ8XCTOw$a+HU>AJlfVKLn2fEnghL#+IAli&9
zWhJHNnuyTMnK!iWJDYR1#D+L@%@~<Doho_}XAv+ya3Q(@9W6#@q#VmM+RMK5EO574
zx3@Sggc3+6T4QJA&|bWopINcqt0^@tFs*XoX%SP?ZaWr3ht4yQtAe)14>~X$r*yva
zz1WhIKlB=_M2?`p0qpfd_Zi9cc3Qi$G+M-RI!W4i>z9sUO@5}c_@oxYUpIPs`M4B0
z2a^O3H=)9jH7jT6+R;N68*ZwAb=SSvF?FWiSq%+u=$V(2Uc_n57{@rG_iHtGf&}(v
zBgk3H=UdVOWuaw!ZfnrFMPm_PSPv40EK$JGZp#pxD_ZvW+!m6eMn5kHt;kgM8cMi3
zi%vP7BppB*L@W|n4qzYA^HGaLmqS{GBaGNB?IOLhqAZ<GOP1e>A{_fIOAxo3hRCp6
zw(1lDid5638*vWg68H37k3v*lO-5HN?hDR1)!;qU2coPgI+pK#9m+<t(Z18Yg+U<X
zlpE$PDVHO$6y_4{Nt}&9Yg+gEr=!>$b!_GM$Tkars@HH`41v4b6otx*UB4zm8*Pd@
zV~Zb)Rem?D!Epz#m*RLSI=hcxRjqFw?7U82=RqY}1-)z1s@~f4a#9Nq(8>oht>dS<
zePDymWeuLHdk#ap^n9S@<^<CyH`I(cDPM}KuhpA8+8a8NX_1)!j>8Y=Bxq@9*}U!(
zFG8e&bKU3N$#Mql*lmOnxGd(1@wws(iPwp;281;YMNmPsd9)ccBTOJ#`?zHZR=2X(
zCV&otHKBR!TecE%ZuB;nEaTJq&{TYXqsk7SDtp?;O}>iJkBW!HNaLKiYKc+xY^U3=
zd$qwgCptWMNK|_6qMvHd8!rZX9rq!YVilpT)h##(JOQ3*{L)?Q&7cZ?8~04z&1ra1
z)FN#*!TiZAmI;XxOfd^U2YMc8wC>n)twm$TG8$V?(1qJ1bcGXDMbe*0y&)IDdFw?s
z!01@{LkG_zn)<ExEIpH6!&$~gS70j(GfN35Vm`i^(1u(O^R3|+OWWzEf{Avu;5~w=
ztPT<Pfab6Rt29ci#b=7`h00zhhY6xNK7V#lljYSoW0u=y(23g#bj3zyCDb4rOFoLj
z*oFd5mo)~KZoSrna6lD@F|qC%tfIv63x=a)m4HsOpNB-h=;nkU8WLvEj)RQRe4mMs
zbAu)9ud19Amha<q<8kT4x521K*^A0E0Gir`r*n3}e!mXlU^tN$jIB55?A9zj^w*E+
z{4e6EjLz^$hjNl=)Wz?hS;ufgTanSZIBD}dCcUpu*`gCA(by@gO-BrZt;Z`w_1~{m
z<8ZHXDv4Z-sOv*O*+f{DM`e^-Ye!JQg_ZAZ4R&0OhAT9b%}UV}UL%!tA#Nw4(PLQ_
zZW}3)lfFfXp-|Cw_wOp}Lb95Ox<^taR*l+*!sw$PqD+~rC!MeQL$l<9p`+glGAq-+
zOh!82qJ?Fh(Qvh@U5>Iw!@(Ii*t8SO(TGF@7Mqt^jV&eU^3gAbE_^cBg9Wj%5_BYc
z-KYol3w63J>ggPytf%$5T8h0&XiRqPE-EWwm(7~HYL0WdW1y`&Y<nn!l;~>l;gSr%
zKk<8*2wc56Y*8&fL%0w^M+`>tIkL&swuXX^Vbc-@@GV=5cW)f|%B9k-fwuXnp*0^-
zYK!ytI@AFLkKJ7_9Nv{#j^&`{T>~wn-f9WRv1+kIgU+_>6GJ(924!o`Y1$tecbodY
z&#)!vBpPdTsAR#E)hKpODgoHZTD&L-$0^xN^v-ZUzh$!ocSLW$N5L}N%!I}Mp=SW^
zMqQ!|T02MQ&@tMrw`G>eqOm4~YOfI5<6OLk?!Rd44?=gbsj1x6Ohw4DW~Z^|S8-FV
zW({!IcAt38>N%fk{nW6m-;~f17Skb%XuG#{Y|*Z_gO`DK5S)6O!-z}*PO_nhrg>?W
zCMYzG7*|Od(CzarDK?4glH;ga{9G&M5!|40>k<vmXXqSe3!qAQ>ur@Ca?uEBf$mwn
zY)ZG@-q}Wl&OUDqLm{f7x0{nYU}dXeTq6`e^<@(jLT<N)V@%nkk$ZGfB_}Y8nd~q<
zDA{qGrg3j<PmZ6V5j34Et3vxsj#U?{2E;~e*2{6^4O9LdL6;<j9t+W73LDt^yU{rD
zsu~JxcFH9@kg5%cMn7XKQ$n`cDd+a7p(`hVXqvaI9ZwEva^5=_tt^MP2{E~<eY^-z
zeCQcrn$g(dtk}MU_lf4u#m)d^^V2cu!c^KG*&ym&ylG)H9LAA3bf7I_3DkU8YOFB3
zXH^UT6X3Gv!UO5T(rl>WfUN{Pi#V$@grx+LQxWYh=K;0E2%_HHP1`-&;VpRjlTA+N
zt^2~UDbrea>Uxc$V_Iu}Tzwk*fy-N)mC~hT2sY}HG*Du~o#qSA2y3%bD(eRw9M*`2
z^I>B_xQEm5_k^x(BPDu_J?Qx=Y<3C*x@n8wYyl1%*QGPr<a8#6Xju+4vO^a-5!`0^
zBo|(Tq1d%ap?8e5bT8F2R<>yO?M|pd%|wguK{TW=@jijB+X)cE#NrTj@mlEAM!0Al
zw_bLOy@qt3fWsyw{M)V^0^F682f^^0hHgeY(XeWp&S!^RYq)MO^n9(`%8q`Ep0B4=
zgYNv^IJah(L+~QJK!o~?_rkF|{aiH8>Kn)TY?q0~)(n(`8DLwVJLMIL^;@pflw$RB
zA)Pe}rh0!o*#`mE>5~C)vhopa`_hGBX0((KCAoxX`8?@)?a8iHMsxr<(Nw4RNm&9C
z^(yMdLX(!~HkjQ8Y4{Nlj&JE>;tN0=6OR+oh?8j59TY@kHQy9KZy6tVm_ZF^EJ_HD
zd#}d`nhss+TXaOnPT_;m@I5P4^t8pk)oR04aKwj&NG5_LqT%b+szSV+Moa~i!xi9)
zP{=0dzh!W^83tTvl8DA+((p=b)oJksOhcs^4{+#=g%xzLgDC1VdTXB1LofoApsQPZ
zpHD0vfs_Q^x7SR!)|S{T0FE#<pxWsZ^$eR(Lar?t=ZJL)R7VY>#h2)fv;+v0D7R1?
z)3he>M^EQ@zj?fYkth+34UtB6g5?#|`v^Ms63d^bC3&5_n(#quECRjCyCMLN?XRYF
zw?mtRQLG7@OQQ*rugqvfSTz(_k~eqVA4-Bmdo$O43p%}f<1caRJ=e%yL!1~wIsG#l
z8@df4O}ZMi#7SgBNt4qs(Qx&G$^}p~;#nJFn%F*~wci@%Cj3-1O!r3M15WJF90wUq
zVH4f1mcv;XI;+2&9J~09$c^R|v@jr$6v7)zG*)blqy^E%BjpMd4?x%r$ArypQxZ{N
zJ7|^Cg=lZLY~484h`MAEl-wa+L*BOr7ByZrf{CC?zNIu-D;i8$HIvh^ZVf}H2co^)
zXCN7()7o#nv|cru_bt1W`B5}ohya0AaD!%`o1-7m8DE){xN<hy+of5VE=2FZIHdw*
zFq}mjI-7<fYO*0LX59;4nOm-oWG29zGu{_6n>c%kj(Z&xHMc>xup6;|xjBqbi8kfw
zH;m>_tz(qr^u8Ii+lmd9b3{NFbQPPK&U>r30nP&5hRQj@)2N}gS-rIt5*c)O2Pde_
zAA*Di)th_d`Jt=XV2OK94T=hcHFsYzzmn7dWfRf)fux=^IAT~LtwFC|L5s0c!L;@a
zy@u65W`k~%)A=gOtyT*x!SuO-uGKoUdG0vqkVc$nL_k2-wAtyrvQHzNzf><9&WKIl
zp;=wCwOj3(JG^(DPyZHWfp2f8xqhwaiVogl9+H7?pYFNjxHPA2#=xv`1FQk@`WK8&
zqabuM1~{7g&`q8Q(Y9{|NO5=tHM)|5=7+tTTE;Egm@5Tyj9%=GGD?g3q8+$x^qQZ+
zn$2t*dghJUZfwwNy$X#DO6ZIZUSj*EaZpmsdY`4;W<g-qi`I1?c57c!w0M{69GKx*
zB^uutRI0_I$*29!$zwC){fxBU1W=s@h(<7Z=lr($NSWi~@0=@lA3frLxtqGJSI{qP
zwR@+{?NjS{OFmBK`ndzl0(&*N_>kC>((ZZ<@r`yVdsrjf<$1?f#J|N9ai73#v+E2)
zvnSal#^%1ax;6C~O}EaiMB1V;2X=gsa5)i;-D~JxK2W9z;_fT~xA%@z{Zr~J0=L;-
zXY}(a2r3Of(fhDSbtWGwlP4*K9_Z~I=}$yQvY~Q>6h5?<WVXQAu8t_Wyn}Q2cWW4m
zViffR574PyF45ScgpQKEh6pC<Xm;9!!=d=Sh2`xy_NQq!shth_I;KmM$S)LgC~#Dp
znTC9Moo15o81*j{xSH8s<Er>{B@?!Ba)io2$4$D~%rsV<FhNI8Ucqd)+E%)mUe|)|
z(vKQmL7cUAT+_R)O~>l@2?l9Y495BpN|HpQ^}<<(*<ItlERqIHZ|YS(wI0LQtg*4e
zv!#<EVNSnRY+gg9R3Vz82GS(>du7w_LLa7w)EyjV>Gz9z${OhHS#$1NDT~eR^uk)U
zeRc$3uS>Ug?nBe9wUNiiXpfJEaGx~1hS<=7l2*oPxn-H>?Iq0mx!NGliPLl`+rIV|
z!iBY?>RGb4OgLvlmBHL-y%#$R@#(D)HWqZ(<TWHW9XryV)fTB^L<B`w=$j3cV??k8
z)Tic~Rs~pU^O7sw-xnWF#t6^aPL!bc`4N6%9R<UloW|$TB#-Y`x*eJD$3DDhzeP!W
zJsbu*3v2(nM>x}Wf;vYJdKoIe#=^n8H!Z1}({#%cY2y!7KG)wUxH|O;$f2;3SCC_#
zQWS0t8UV#WI=|lmVZ3Yj3rO>pC3Sbu=tW(kO4is{OQ5q9+&yT$^;VxhIPdNO6+vjQ
z-@EN@uN;`}&-?IEr^g0R?;bm2^mA9|v-ajziXEqfpZg7j2Tzx_s(;HpPX3^y_#oLs
z?5CsoS-in}oQgnK@lmqhuMXMIn!IdSh#lR}T7$-;kz1h-e_-Fm?88&H8T5k=C;}@+
zL${@)my0cLG+r<H6`GHeJwo%KBZ~S$vXf$=`!^$Jv!H}$IlYzodU0F{qE3#5QM_eS
zyhRBTN6~!mbP<Iv+V-b{wWiqsBvQYZ+HLy9S`09#c|`YaF;Srfvb=M&JM@~Gi2pve
zoT0>h8jX|M&Ju8$$%Thy3t9;|bXq@jY%|@_*Sj^G)t0yR)d_>{<fCM}oenL~)S_&(
z4I(nhN6FSd&o=xWMDkDAvBPW7dYfkM><@O04dg(xw|Bc2+Gpqv7&9G;J*5Q=a|1n4
zd6|fY8HFNf^+B>lpmIkPv{ru&^}-?>)?oQiRya<Boo-&7l~1L}-35)lAwedems|PV
z?CHHePPTBChR*2SEgGBD3*Bvimiw(>D3HKtyl|~k1Lj-n%F&JD^cgsZH(|PU?mCOV
ztnea$?&!v$y3;xInKORxM+9Bxakl&#a`XtwzJ}9r>9P2HN|H=qP|hrzmP^lyx&%#^
zw!FQ)>-t=+;Xvke-1{XhMR(dR9VxY-e$D_L$yOR8mYZGK&V8t_Grn;g(6ZaOk^BvM
z>{g$b>g@Moo6UNs^<GTLSPlA4;=q^EKjs>r?NAzYZ!b6I2Og<4`<22yV&FaNlVyz~
zm7%O7{>E##4G-LnYqWQ`4MSJE*SbZU6tM*32x{OV-rNkRect=(qx%<LV^$jY$#*xo
zTsmBE`60@PMB>mhp8BLzLzQC?u;3f`=U?7!p<}qkRb%)a$<IlQ&f5x{hx<!#UFGwt
z2*T`9gMYb$Obv4db>m=IRF1=PD?nN}XucJc6;0CuRUf#jXRUyp&$0xax5{?L?x5rT
zp1IUtLRekUopR7y^vq_>FYh@Gp{%om=6efsX1CpIXxt1U2Xl0!h%SI~R^)WuvIIT`
z^&xQ}7kYbgoa#eo@r7yR+k@sB3z%BktpTqDe+i}DvZaUVTs;Xh^xW<Yns3<}A9G2!
z)>fNeDqM|#+b8jweE~I!k`X$b7mgfrRgLpa=nN~px)e8*FrrH@JfnBO_!tQ?A3iHp
z&%6UY%{}72%6+Om#}4Z2=us8y>g}C<vktmvzcdFulgaXy33=9mdiBad*N4s(4VL0#
zLJ48bOE28p7pFB`7zWG082eIu`m6$Xe7GS`GbARl1D1eHJ`U@>t4XR+r{(^B-lmOH
z^`ShRF*AklB9t6fPSdSmE18&6uR;3iK2nw)eMTu&%cTM7^uuAap|aEZ+O+IwLj*&s
zn_VRi>jzRVwMAK<#GoVk@)d;`<alT)Xumf`$Y~B55CcCh)^5?WnsEEQ?0dJvH|q|o
z_xGoB^{DjT>s&r9lwI!6R~u3Aw%|^~?SK-RjnFa1r3O(_8apbD@kr?p%B8m~!e;&o
zJV6Jvq`59Fmnv-9s2i>Wm8#V)?U%0QhEIvHnpNPA#$NMYtfyTsZI_NcE)7UO2{`VP
zp^#UC)}Unb4%+YCnxB-xTfteDxFJwhzmK~$x)^Qu8Q4#Z8e*VD-7WifS3+203<azN
zb)yB!o_Ej@wA8Yq-Wa|RxvM!|#DxSl26UrxAhTEJrMK)^TcvnA_r3@fi<gsc1;;MA
zEgoMvLWAu#xu5qA@<8-*>4ju7cbIOq2s2-Lt6kf@OrZD1w(|3ehWgr=p5fH#xHO(N
zWUL3-AZEQK^Q9#44H~d053a%JU?hZCZVQx%4}s1Ydm`R)kP%|dTlB)cQ{2|)TD`lY
z3G|j7N?PgkUTe&~FRekRk|;VYL058DlM&vdz>!w8;kdW372N0amff6Q2lXdIDm2ea
zOVFHdt@*N1>ZJkk8H|?@7BoVk@hODL=0GR({yZi&5=7kY9kO;%mJUJhO?4%?a(dxv
z^?}HHN3Fg*vm=Uq>AaNvm7w=#@p`=UHrA5b*Eii{JPYd2fT!AM@e;ytkz?U)?iMhU
zo#26M)_`GC%cFQ)I)kEj_Co2gJMBT+omxXpO+f}R9Q}YgvUOMjj<j!c-;%bg`vqs&
zF)__Bsui77UW&weyqR(zU?W)jQpvqKwfCh+yWD=EaxQh+f@X9&A6ihld?^}{L2FQ!
z3BOQ$(}L!p2*tm&BaHXb+47|U&42cbA!+8&*Cq-l?++u3?|LaZR6%FZjfm4r_sIK;
zqGB-Gd}x;Nj?;eWbT5-XXp!j4L-!n)wxFz-+e2G+FOxs%T(`kLwAZz6*>b6r7QQKR
zi!Ytms{IL9O8KGY#%wUyCL;TRBdsrp;?Adsf8fe(eZx%rsY74HAW)P_`h^ns_|Vg1
z$1jwSyN5Py_I=Z?1X>Se_bBKJItVp->1wyRKNMB(my)CT7s`J6L-$=<{DkYig@DMe
zeY*IE5{>!`9e1pAdt2Ib^V8%XD%F{9TEmg{58Be)ztC{x{Dbzg?w9_9h7aK%6o>ei
zPNmv<=-791`p~7d-9vFYdgu;^ogWl2%5<j^w+f*%mzU=Hx!n)k*HHVw!U_LFYcBOI
zTW6UMY_;rPC<mJl?VJlfba1Zr(%JG$CkHw&<v`@2i^GJ6o)=kvp?4?yL&+e~cX{a^
zFU<WeV1LkZcklk7bLN-U8fUhD&^_k^TVD_Bd6i%2XSMu;+6MMR_s;LNNT)a+x)c)i
zP|WLx;;sHoCD}dNOGnx-&6pK0UEy!?lNKfa{-%}b;H3<u4{hb?x+sU%^j3?%|9<G8
z-@H_rgTLwIeqXv;d+C8sqa(uK(=T-N`c26p(Z_tL1lPXlF$GI|Xi200LdRHd4V5^-
zm#*2$L(ikl{Lr3O`h|{FU&>a-3)k8YJgb@f(v^N`1mq3YLw8N-ZN^QY!52k8G>!4l
zoYVe=&RqIKOU?Zky7PJ|rx)#^NCefF&N5$$Ox#-o>0`=6>0|k!Gyl>IbGGtUuxfjs
zVeO~y6GYR~H&t*{d*STkEt+e8<|*PTM|tV_N}d*1=<d_Qw$$n0vaQbc3*B?R_j?u*
z>7^sDT3<TXcqy5i%1ieyFI~#c{7rY>Z<Fok<0Dm9dwRK&eWm%9%_DtRYnjE$FLdQT
zA6i>pFKwyhw+6<m{?J~^{DrQXUP|)y@leucwU>^%>qA#*<rlhIe``3#<M&LpBtiE*
zQz_ZrQvrf|Pm7`LHy(y;(tc=t1}5Mb-?P+`L}Pv{Xv($uO-Ts#)_ay{>8<x%?AtU@
zCdV~?p=<584anv8(w%mHYuGrke$k?YBITt!wU?rN{?>3zi|0cr|9U4sa(wEj?xD|o
zMB>n1-W%x81Oc4o+ejqAX|3rOx@&#u?Dgpj2us)f3!Uw0afGm2f6MOO)^}M_O7!m`
zlkBuontbT&^R3}5!oRIC<;)%Rt>G*=z9%WN^X9SN8kAlA8}>Tee(7TvkXg9Dy&S>(
zUm9V~^=*nN1@hXz(7nre=<fSza+Znv88ML*P2Y9_8K7HgFD)s}U+Bosh>4_VewrLf
z(fqwcA_I2I-wRkJ&@Sbzp^wtv8b*004m3sU_X?H-;H|#TK-||0N8N9&NPfPxZmCuM
zNl7iyo-tGTUA`Bv(9xesirf-?zV)7A@Y^sTA$%M4r7ewb!$6tm$9`$o-t0rye43mD
z6Z3loyGSsx@-K9(_SUe<dLM1Awf94J+s_MTrTC0OEEM;*Wxo&=P5BqPQ+sPb8}g+U
z-G-OW{LE9Ristdw&~|>>997X$-Wo<O{Vj{mMtdkZF3U@IZEp>$*T)<q%|uIY4SPNn
zkZ0m)G15#t;~-fk`eTccWumuV=zP05Stdq$>pfFXeQ)7RGUKfQ34{L9Im=rEK61a%
zJ=fa`v#<QpmDe+Bl4oMgm+rdLTLWn)zF!Y3#iyUMP~6Wvr974|EbSSf(G*S3JcX!e
zPA?@2Z@+Zqr^!(jeO3$$S<zD7XTS^ptpP2G_kPc7cqS^vUbna2)3lE@hFU{=MokHE
zd@o?vc6rNhYS~J@X-Tg^rkuC^fgy<TCtTwhGnJwDy;-&l-U?3Fe{6&8M}QAq^Osv7
zb?r-!$&%k^SYx%9BI+?-y7qUIZ&r%?d(hlU3;k^yK*!#14W*{?mMuGd51Jck(eyo0
zZKOqWersqmzZdH@&$ohS19&Gz4XlUGT)s_v=zONXofO^w*0A=|%eS<{zrWy(e);pY
zLudOQG&lJso=yhcZND{?-pfnbfhsSZeZ90P!uFwiw3m_#`K@7mZ$rc8)6KV1J$&2t
z&^_Ox<XC^3ap%?Ei=D~fX>t~dJN-iU{x&MN(vysrmb|}*c<4&yr3l^hhqjcy=PBsO
zPlrZcG=H08(3R&~17QLAp+uj(HIS70rK>#e2YJz7e!tr@ERcHQ`@jiE2*E?y%$zS>
z?b{gJq@HN|t+%FL-Wn=70Uz2*|2D>;YvlJKdK>MfBaQjiKxo}d*^<pKUGsaI+9aQt
z&)`J%iB*1~&;I--`^0*NOl8e`VQ$ax%tCRG{nD|&r>Tv^XnMv>G)B|6F}4vIP3x_p
zpP~GH2J)M~W$~VV@As_u(^~`5PxYZIm+{uH5JP_`d*g4DopIqAGs!<u*Gn_{&o7<r
z`*>=TfZ};zB{_ZiIr%4^iIVga{h6moKhf7)0sH&orF;Bx3!{(u(wS#^Yp^tA-*naa
zz3&IzbH8-=^}UAO<eykio0ETHJ%c7|#r<uMLHBNN4dppI*ock3Xr=nr+1k?t2#u!i
z1uS&r{L)c!`-QIj%u*<gmh|51nPQ$n7p2iM-m<%t{oXI}|KF7C6ZPq0WS^+dm`V1D
zmS0**Yi|u_0X$#&oEB}8PxR+SkbI)QH+w9bXJ{k$#CU%{+ITND(|8OEQcukE(j|Yx
zhpsu^``T?j-HY54>uGRuPptmZT%H4c=-T-$TiX6!zHZiud;UUa`d+?nltxqih2n4j
z(B7ZML1^?D?GzXPKElMk_@%qN@O=68y%d=v-nQNNRen))9qThkp{-dTvjEl7`m{Lm
zuaDCTR8lzK6iwQ<hPr!tY3q;E%0@!Fk6-BQ->cWX6>nEwx*xBVjU4@Wn}IBo`K8Z+
z<|c>1d>d?k%}*tf8)AJKjZjop-?C?{TQ5DYJpV%X{+ghm8~~4pp5gM#4ijD9J}c1P
z(s=1e-<|`aBY!nGbe3<Q715Q~OIIuTtzoCqexaj`@0LVxw0>FjqTdJ0!2G>%1eR3S
zTkA25@=I&^c3BZ!ZN9X%p5{YO$N%LX`sm}Oqx7fE$v-j5OK1B&JcX|GZ4!%)?^8SI
zF5i2kqU-y#s3cNQeDBRcbNXHhiGDj2084q=oHgRx&sifrjldG|d-*Gh1kgjtY4hFz
z(w}eH(bJv}B@w|}13nS&ovoVJFLamu(%nn@erf2BmiY^vHPwgaTI(-#*HIo?dfgAK
zXZf68`t8^OZ1w5;q@nnp`=Gu5cCE9T9KX=n_gn8Voy&ab9{r`qSXeLJdFJub`92z%
zDCVp$-Rl`O$waa0OG{qk7dpRvR-m=+@(bO&zqFmk$qywx$S-t#?=L_{dxjxINYmag
zJ;V7+NlN`zdsfx!rDt7{zpsW6X=%UEQJ$#^CDL!F7134eOG~Ryqmz!J?qBHmc3Oe9
z()wF(Ti;GA(B67}DYCx#rK8W6&N9+3bd9Ig$w@K#OBuA*OUE-tlagXS-zzyO=9~}B
zZKd&05~9Cl%UI(rdyL9wm>$af)*jm0?k_##$ohrO^vqR5Va&hKo!;A>J*T&E5LnJH
zeMbzmNFPH0Dbn&#q)1D9>6ttp^8hK*nqFFJt-nyEa{s2ae|xS#QHlPA&Tr2ZXqWQ)
zLwC>Zp`-Vw$;nC4=Pz`Q^wL#F{e|xF-q<GJ>K7#|#Vo(jIr{^1eVUx46zl0AB&Ar-
zTt!lfHQpLdt^1{|?(){#_I{e2rQ%*MMX&o8x>9=|ZMO872g}orK6K=#$)#XEexWNr
za}}DT<@tiqBrQF^R8ha*G*rj`gzp?mVv1+DAumOJnw-28t-SPne*LYXueRQ@=-<9&
zXX<Udbo8f>k(gqnw}usZ=zq|Sn)shIlbPb%b47H_^3trR*N3k8JyT_JQ>^?_X-_=x
zm`?ej-)=!-YyU!f+x?}u#IBdllwXQ|?0o4?^QEhQJFY-C@_;X$`FkCk(I(CN7rMV4
zSD+bTw1?u@TVFcs)8r(msO=ZJ>wIbJExq;j+<t-cyOZWDp`I7+o?kk9f1eiD<oVL8
z{PC0SI_g_??{B}A(^iCkeos@G3>EWfaZ*&w{?hUz(EbPA>wB8Yq^Vd>gOR3U&0px;
z-=;xkxwz{ubnlTLiX7xy?<~JmJ6C$2frJSAq5CoJf#Z2v<Vo|>;^>p+{nBF>d>aNC
zfzod`4$&vh1X{~?#GUA>?Y%RSI=nUX^zF7Hic9)KM`_=SSm<i|rStJ4hOX4VXDR6D
z>!rJV`)7$hi@Hq0it)XS%}q&<)=S4MFU4DbzI4s?(%ROyX^=@-vDz<mtohc^@g;fb
z+Vy=l@@u?wrnj3nNz=cSn1k}tot{QVtTaD89ktTjexWlzgC=UF`FkA;-KD&=^r!zL
zR$88!3bE3fUwV3JduwRVPB66h@AEL|cy}P7vpq9amt1>$>F(vFtG^>Uc`N!eQ<1k~
zJTnz}D@OZ;?(tX$B(9k0r6oT@ESW3jc&n{-esBLl_j-$xuVp>7=eMJ?R@~{WmJpWl
z(*3MEGK<CiSO=(<rt)5Gt<>MLyS-KxvZeWZ6$?Gxp+7YL(vawuzLN^NrQgQ&|NYm0
zOMm^3ztUg-{@?zuzx;pypZ=K-On>FS{*4$*|L1@IkDvcV=m-DLzy8aA9+T;><p2Nq
z=l}j6QRsj3``>Toe^c$RH~ODE@xR++|2L#Hj;-*&{`wRDXZ#N(e>5=v7k}(D)cNPX
zpNjvT+W!Yh#5u4z1`C({j~{=xD00R2lf)AZ9WN4p{ZRsG$j#c?i`>8eC_}J-b_Tr6
z{ritHnH6J27dCF^f}f>s>f|VO!Ov15${U5^??21jIIbvk!H-g(Ej=&w0Y6KTy&n0L
zC=`GHSq3$942WCkf}f=bs3mkg3dP@l-g&$ygnl~N!hXIyG~u@Cn8n-rS?;n=vHQk;
zmixd@!TZ{N7W}|Z!OyGv;}$>kN6}9%?4LzH^hePjQp=x`7yc;tGeqxG@<TsG|Jbs<
zMSsvAMdPLZQS!nc1@DXfqvVJFC>pQ#kCGRD3O-}q|KT?tqspmPJ*e3?@jvaqWa)*S
zsrSc!zRb%g$#ma8iv;o#B%8>bp8qIvnV0$bd{O2y8iE7=#UWF4eEho#yPST%pAtXs
z`cv%ZeSgaSxcg7hANT($`SU40MSniWr|chZ>z4ibJkR_3@gn|ckKgNkTI7FzYB+!V
ze|3TX{9~ff0I>i1H-u3B59XI2^N8Q)kNdq-mRA~e{_J{DMgz+Hhp+3E{@rzLfcl58
z>y`f9b#1)+hp+3E{@ryodX|6u1$(W3du0vA|MV-4TL1RS8r=IoePyrpZ?CM$;-7xa
zQS0Ab8JE_7_NvtQw^uAuUjL`B>b3sul{F{+<In50{_T}D$JT%P%3kZ=UD>Xf{;6wz
zYWuG*rm1oLqgVX5Po4kub#0#gXRrII^S{2XP1^tLbw73f*Vnb_`=7n;r_RTYcr1k9
z4T*1a;}{SBj~2@R`MLbH_J7z*==h%>o!;DaTor4DmPs;E38^;WTiC=0pc44(Suhdt
zuebnId;z*CcmnsRG8zcks1D^4R=ZLKMwOFXrPaKQd$5UGBQUM|<3(N$<+U#>3a*XH
zY*iwQ9hcsnGf@(?ch881bbsdHpS;<j-3*+Z`;*Und17yb`>WQ*%8?#O#y1((_uc{V
zWD^|8Qy`(+pINJu_B>V7zXd9YzmogW6Az%<-#G8DNEF|j><O8Qic*qiO9u{4`zv15
zu8ogqe>wB2nsT>W647eCh~<*PDE8y?SK081yII8M)g`UT%kaOesN~)!*~H3K<dxcr
z``e2H{)*PJCu~Ngzj40LCC(FNhdGght0L9qsVRViBob92(IsFh<YuXO1ej(XY~jLE
zQR2-euZ%u_<3{RKN*`w}@#jhmZXXvaRG-WAVM@^G^Q&*l-i-c8OcjUvm3WC6ml=&C
z{fbJUrxQhoUs3wNwrf22S&p8!S*wz2$E`0ZnpWb3Je>q^k#3J~F67@)=I6gtWr+?z
zG}ZakFqrlSDN$TSrZJfcH=6u~BhE^Kg6>a4Bo!z56<0OS42mMZLh~zs7$(t)_I}!Z
z>e`<812#CtuT9zR<2t14)2~|_DXxRGxkiwoRLBN<Cof~gttt5+fOi}x_zV|g+&i*7
zYL#7PPoat!|7iog9@0t0ZCX-Y-dWxku3m~-eYJA$b7zh<x%ugW-oS|vI3=?uXhDd-
zk_n~izxua`S6tPvPiy+?!6%M9DD(|df;mK_ixF{owi3|fWSKipIWD42^{OPtrqT{0
zpEcb#ksHckQxHC187e9cPnX7CUdoxi0xI7uLxVB~8ysikHQ0uKq(n<`T3;n@z2a%k
zjSPL^)y2`#-Y4TKw@KdUD2Q)tQ^81xj`n`#43qI;yG%Avq4!n61}5T*^gbyjDS5hv
z_g12$(3fC@SjBX@fm2;r;q%2=sp$1HHqGYUjS#7Amx(NEXzr7fP6bzOQIN=XO(oT(
z?P1Ht&T}o3kw$Ny2;8dF!*Sb#BwW_I+w^>aq_5JIYaRj5*IflSK5hY(I*IuqG@y~%
zZ{!TA<b`eb3DNIwl*A=4uD(fl!`xJ}=Y2(IpIkATN#-6GE{-cJ`C~o7A1eFGQo=1Q
z&(npRJ}Ex*<g5FI@N;X{9GhfH4r}44vPv*uMx9lO8LE;e)%ba1p3}X%m@-oOOjMLx
zy~SO2l=LaMfBblGD9ht0Gn+FM+<{#w4h>F%p1F#YevB=Y!!;+L2iO4>SI_L7;{DK<
z5s%<H*%NpGQGB*p*@gL0)o+|la&d_j8q6pS>q5Pno-YuMUcJc`lREKFm?o8F^@lBv
zCFz!)iVj8nGh<Zif}X(`@q7i-*DcY5_;o|bQ}405iZ_D#l@rNICavA0&c4;^Q;vM`
za}wCuOkZ+5&P1<JpLlI6E<W`wFs!b1OD?i6RP~z`*IrVW!=BhKFOQIZ=OB{dyLuWC
zbo4W7!;Gq)5DYdshOCGVd+umt^!b)3cy1v@$&3h<Ca&$N!%)%hviDFI-agpk!p32{
z&1UODX$DpDsd%DQMD&r{(bUs=xo{&8!9ZU;E1(Q!jvuRzrDt4yC~&??%jW~p718+L
zYE(&HYGWn&$f|tus_~3E{(1;=8nz>u){JJ`$R}uqTK%@hMtVSS#zahr6S@#pS#K;=
z7;tmuSQU8dwi5wPD+{K*d`cm{iL-Sk{N-7}fP^RO!6P3Qy7^T*r^G}T|DXwOLJ^`$
zd)bFkVw<osYSA8c+$YIFWGEGK&IVlS!#Mo+7H|*%$j@hODejYChtqO0QghtWRM;6X
z*BWpUvIQ*Jg|~qOJN;FW^XvDPcH@|lpL^OGr15n_sG64`M5gdlIcXVFPLoR7uWe~;
zU<5-@ZZ1zThfcoY2)LQ<*-sF}#{*yOLq76=ij1#Wsd)l6*^d=(Bk!v~#l=^$V9gdK
zJATG$w3SobI%|a{tFaX820-$B>P0NK%8|(D{&XQGynuLAs^t9qeBq6#54y)?w!-1*
zm%aA5p*=|6zzvOLRypeU3WisWZW963*^pZER2fL)D-vGDvgZLr13$UdmONV&jF_KF
z5QnEiMin3D68(6Vdg1c!yYnrGDvZ7|>A4x}xB=+CbG~eBv!9P$IpVg-4%KZrFyju#
zPG&xz>3GDm%#JZkH=CZY2Pyna^}NJh%oYP4gTOm^?{RUm->4x|9&zJm)tRbwK%d~I
zYGUQ5lzbNw%%hTL2HzJ>G9}e4x83dVG*v0q1%@SW0Q%>BMpCnBP{HrjEdT8}x6r`%
zZ*EWjWMSk2spQ*^>-IDB>ZfNly5gsOlYXRPiF!L0C7rcpNcOC{DBow2GgPvid74?E
zqTKnGB3T%^_Z2ONN_K<2<F?P*y$L3k$PUj$MdwSJS;YuR!o8r9JR@Er0~A>q_&jql
zn*h!GiqWglBKy4)kEE`fl>oUpJf9bhTjT~27b18{Z<b`JQS#Ukw$e+*)4id0U$;IR
ze8Oum^mXWJ#Zzfbt?z90m-_JW0^c&V&$xeI2sd!q>?t47xNkb<L4c+1xa?Ard9{uY
z`9LL5W=QkwKIq#Q(u~l{Mjr9&y>x;j4^AJF@F&w|G<#|B9pX+#4hF8HWReaCn;Lu!
zd)N4R(Zh0^X_=)7Tx=S!=Iz-`6r}AN%SEo<JP{Kj_6e;9?mc-zTp+nl)iH8>pZJjx
z#Lj50zT|L`dx5L3v7@B-D!%Qw%uL<phrQE|yPO(3xc2JzE|<?O(uHuehP7`m<31v3
zsWd8ecBTY5pdL1Iwln8l$l6<tK?^&~N`PZacz=<s$EO({+(dJDKGtm(N-0rgRor@2
zT5Iq7gcvP5T_sUx{F*H^%^0h!4_u@%mn~C$4rHl?DuZl5PoRiieIn5U{k*jqw|Sl?
z$&Mm6dO0LMP;<dq37052*(YLrsHH|hnNKiw2b^tL4TV{}#ZB&MBgR>q+KvFy^%Xe=
z8ofrp!b^<w3d{w&M8SMUUJX|x`Dz8(`dn<9KI27;UnA`$cs`91C2OrV+7Z^!0xC#d
z(_%6M9nJcV41>(AJOlQO3PZJopYika%{3naKet38)8_p$etmwEY$9sq{+c9a8b`9^
zFAT1%4$O|(x{rR=PrKMCpYL_HLen*07PCo#*sRqOG1_&AAsT96Qcw+NS(X|GX?#1x
z4^4H-GyK{1APV($+BS@(c7p6_Gs$_sD=u*?5;K~|t|NZiGMWpN+@9hLILWq(A>J-K
z6G&Ds?@X!z_qJUemJ-F`=*TUwBr3k|iY|R2eL~f^<t;vWuxe|Kwszd+YGeq(Jdv4g
zaHnJ+W7+EiNA`WhUEE$;_(pea@HXE(fH6YSK5Rre%Zfl}Q8Io19O3yy(gePFEoU7P
zlFBAm8zZiA3<O*Dk;Nu1Q<p!DR<6bx`(?JAswH4gq)Vt>LF{pv^G#=sEommVnsM+`
z=j}?eJ_DY?*Y2qI7H~FFC1@|s6aFDLudwqfpU_+98tk}LI{fFBwt~(%py;>R_99CH
ztPQ_7kwL>&eP>DGlM#VIV`(MHaJ2)DY0JJETeZl~>lJ7f?5QAuEGn(wjO%E%KvK}@
z-#jrvHl(&?`^zBq^ymR3@N;$=zziWi=%}u<P}g|jn_LLaH`MqUWP6zKEU&hi%(dik
z;Y?;SpXG{w5$rD|Hwkx(DvxPB^f}GF;?jgl+b6a>c`#D*iY?dM_NM_(ECHG^?-3<)
z#U!T(dHPb=_ReeSXa?~l(yvL&kR4VgJl#<9#{n8TFDM0S)%q@i;Zb4R)o*d6xH1vm
z`n(cv)uAjo-f2wou-U9}&)hts#<Px{+xF!B#00o~W^yAM$AB)&S}3l3y0Fb6XL&!P
zLUK9E8aTGSbpQnAtBz_U?QR_$xRK4N(ogsMA=F2UrXqK#?C?w|x!Y+%@#mI_XA2Ti
zAUUf_RL)e{J=rwy3{Wy<aeWVXrczt&HH~-%G1017@ow4!wOxUo>;79P$yb|Lj=i(u
zamL<m=ndO0Kt*0b<1z*6j2|zIbQ#Qko(NmXR(q!lk9eBh_;a-g46#BYn{6gk^|Q<+
z!@Vq9oB8%3>hXmX2N<sQxq=p_gEf7U$)X(JbUwgw;<D`rG~$z7=}45|#wbwv8mlx@
zM!=>paVB%=M>|~AmV3P=e~X3mHA5FFEd@J9$(AX5zLXVgG+XQ)o5{ZhLV22*%xlJ#
zCl2YntmYOFa`CO3y?h$J#}jU{a8}h(qA8?bKrhW^E+7<NI7iW9*0VPP6%)?tVKwi$
zn1Gx>)vO5j;jkzooI!4Fo2h_<Onc1Hwg5ySUYRU0B6?d<V_nmBwQ{?^JzuMzL&XU-
zLhy-~#Mi!kX$2g-4{h;z(sI<{6$mcFWW=w}trgoKOulQrbUN5I`;=wpNNB((bzHK;
zHk+n^Gubf-Uv8A#N@Q)vB=y2G=;}~|Otr(kbD8}qH$Ar|lJI!rGw(4i1|#;HLyhWo
z1#<8;+abx?b%)b7My`(0)9x#NUDc7;WX?1F=@+a6chBz%oYfE9`S=_ji?r(a11*{h
zH~9*Hd&R?y&Qi2Wn#pI2#+gacPWrsV9ML&=^>Rnj6f$_4br~cdcfc)&hQ*8+x4sqa
z&0R*aR3zZfmu;uvIYx5L2|8zXI~(=)I;!5+c~rk~LUM+k^7)Pp%J1Eb)zQ5_5{)y>
zKw~W$^;H7A>C4S0O}5kL^fipH!hxH-1Ps?CfOptC$&{Ha$RyzKEFx`$JFlh19Ul9_
zOy$N-fx}rzO29}O&K1CoGXrCD%H}D6*n4NdW4Ir+M#-}<vKa)Sq#1Crq_LeG+l@u)
zT~Xb-kRJn~>7H`(^YP+?`(Qu8&)oOu0gG%DN-j^13)JSe0-wD54g4aC8@A03?@)8s
zww+`}S`<8HyjefplL({lUaBqh>KU|KKrLah(5mluqwe0SEo0h<>1&W<#wL+7gtU0Z
zIporCrba&vLD^k#+*(%o;Tdo>Gx7Rz6Bb?f*6oOdbh~z&MB8*bbKkyY4IABTomjM>
z>RyJ8bcA8|fNR(R$?$V<B;6IgO@+x9TWGf@c{DVo--^rB!51hSt=#)*#X9hgyxg!g
zZHnIEYJiw~TXx=l%*-9v8EM((;;~!3TTgPzjBRtYXHP`QT_dTyBz*4#sIYDAO?vKB
zbllI_yKIB8ji?iwO{<Ce<3jrkzB=qBXPw>WBAZOJ<M|lQ84Y73=<!XRwN)pdMpeB3
zh>|6%18?>a{Nv>iTRsg-`_9W@tvD*_2)5PuIpR4wBi#K1&n(^TSa&vmZH#B~X+Ttb
zUZyy0e%*QvI2cPAu#`E@8H&d+Wx6$XS}SQXIYy9wk7pH?-KT*Lx&uy5*Q&<M9*@Jh
z;<Ck2>G5-^{_s8aFqk;L?6&CXDJCfVM!PSLQu0w}6Sk5xfuF1RLSd6QrVcw&+dWE_
zJ$ympJ3*+mcTx)YCcd*XD%b%@p6(a}oYTyp<Y8tgwbyCKPp5HP6vXh<W@JykosCYr
zVzRNUO7|DRLBH*zvKJYTI2DkMrH7A(!&Mm&u^y8j3jooWu1OZ(bhc$jahUD<bjY(8
zbzfFG%|;U7VyLNdK&t_BBZmOTVWZnOjtYAsg@6X9WTppj$g80!>uZ#(ip16tTd^kE
z@|4?JYScJpMweZ2*g)l`GT0M6LhBD-VFO38Mak;ksnKOG)lNy)^kXTYJUf_tw!aHh
z*;}<!67oq@hl|u^+WangU#I|V-WvDeaJwnK2nG+f$n;&d4*skl2QJ-=?%_*sbk~WE
zq}{6LPlvhn8V*JXu}eZj_Wm}D7I;Qry@}ZQMulBlE;>!=Iey&~axEXOaj$$PoMI++
z@Ao=g+2ys%XR{HEvnMTx+=~SPb_H8&k>nczJG*y|c5c$Wp*fcAzvbmq^VXqh(j;@*
zyIi=ot*F^<jc42qKNa~YHZRE{i=!DV%i$hMz$q6*a*%ABJy2e+Zq<^qWd&4>wMOOE
zqGZv+q)RfQI;%O`j9t=-A6G3H@NDzFeH7Q_FhOe^d#?mEUq>XgYl&Z@r>~!{ol4yK
zP-wLbgM6<uX0;22b0}Kt<^o1$D%-Ar?dn!vo%b@!ypgb+{P+Z=_1yfty^j>v-g1nQ
zIW@e_O1>96Z=dc&2s`iDToTS&U^&d_*J&n<lTlRGGj=LloMU+tacDbyZ8FoXvsQD@
z8);pnKx!>HmYL}2t9Q*XX9)KZgH5SvlSx-}vBQb?v7AQZJycS>ELqf38;cJ2G8Z{Q
z)dNn>Xw<Drza^I<sO3FYTU=<_aRHef+l&is#ZSw9tc+-(ZClq%j%`+wm6rA*<OXZ@
z6r$>ErP_tu>lnDSHpWkFnste$x^3$*(+%%L+J?%S!_R1C8;U^NXOevIh&oZ8v-h)n
zLkhTbDok3SHR3{@^|+RV?vz8c8vT|ZclWG?v6j`MrTY49Yx*>+8n3$=&{xN2-d%B%
zOYY%X9j6inH+H^n>+}mpv$K-2+&Kgd^)vk1&V7n6x0A|c+urW*ojnxPF@EeD-?saN
zR(e7aK;F7NU3f-dIs2CI9YR07pe%w>)7%RY$dq6%rsbAwjIA;u+-n}0bjBz1jvevi
zGaA=oUNt?%tt0lAZHlV*GG`FAQ{0t(J&JBR`xES&>|KMcWHaFz7|k?=@oU@ps4edC
z7`w^$e3QR3R$SYrOeIx-vEbUa2COMw4+}0rOXWx9N9VwqYz=sfUR!}8mR^&&jG!!2
z0&c!|&CMl6$=b7>m!vWMlO7!PnkNA1h+4pzZ4Fq8_e&h!i#c)O$`EW9TLEGjWrxRV
z8O1kqOWe>z&rv6rZN`JoS#62JPF`FFuQQt`&|40;tGRI1;_lf<q<1;xaMbF44clfM
z&3na9YG7vw-Zq!|aS8J16$yXsp4Q9G5Nvb%@Kq(+=u))Mkxt4{WzE>)S_+L@lI$pE
z?I!oLM;iSM!R9wT{I~>o-ng#|`;M#kj?cXUGFpj}BhP9c);X-CNp=<+*Uu}r<rTu{
zXH>EQX66+o=U&A_6Y{-JR;a#><(rA9qUU9Ec8e*JB^R3ga>*~=*;#VgO6acTVW+be
zb68wHV6x=e+Z=mSjJJ#KT(+&^@~yIH#?!U4f2NC!EK6`MKYv8^JZ0PRik(c@kbeVb
zYGnlE40hPG;JygJg{u?}zsvWM5JGRocDzn!w(V#n&l6ti?07|ZF>omQbW5u<>OAdh
zv&wlBGp^1?vS7=;{zV17Rhw1LFzW5G>AP7I2>U7{;AqW)YpTy5gt51oZGS5an@y~X
zEd$&1^DOkwo0$wv*)fgXF=7;RafO{R*tXo=M*!t>|8zf>ICGe-0FMdmqvht@(T@Al
zaOX4B&C~Snu}y!A?AlFR0iIRKif1`?v8@2lLNG<0Qyc9}T`M}<SzVnG*tgvFz4{s3
z%)22cKRE;Hla=A;N4y!y^RlxoM@s#`g`b72ykbKhG&lEeNY&54p(V2gPmAPaD@M`D
z$-73<WtJ6`Wt5((xf{ITPIfxA`m}ioDgoE3Tp9I38=i57PV@f$tW)Q*k9F4>bdUmt
zL)32bpV($St>B_k?!%r>cgRH;_oMs!m5asn?Cy(1(J5aY2I0PLy;}z@I<fUrF40|W
z25*RQ89ghoPB@rs65kDiW^L+gU~KVZIK4GHlrM9xTej40E0^={e9M;IZRyf`>oaKW
z-MVX^b(NqkcXLnd;Z-MkzTRy6VXfS&^=jt^XJ}5hXvfF1Dy7U0c2<;W+sk$DBB)K1
zco5Y3fpgmLPnYjK=sLr*ZJzc0x&GSAed*Gt5w~^_ZJO_1ltURmLp}?wv@f0Vfshwz
zSwYXLs&(^I%ROIuM(OScV&m3u7QRW#>ydN6^i1~E{OodDx8AxiMhez)ZUIkk--5^d
z>z8(&+Lk3#q+EK|bOqYoxfXa{%o!RA&elFrh~mdWt*NB@dnH3IX9^p;2U}Q>1>cnV
zxzJL`R?+_Nr%stGh0Szd`3ut9|0+6uC|)aVk3&RZ<<;{DdQ9MB0~D-Uk6;lU_&gR&
z<&tO6B3q>pOxN~>+A=Ouo{INX$+U8Yfzow;m4mYV-za-buW`$w#w?n{a)DxVby|T9
z>wCL>jWZV%Y=R%Rd%Eyxs3P&I82{ucd&Z<f4%(D&y>p<&&z+@R0-ioQ3DV|RzkcY_
z)6U)cecathIYG%>;%jbNEa_fXivdML=~ma;ImRB=Y+K!BXLoWCT;!XeYh-sfhnB|p
zR!lYc(wf<u;s%g!y_6*L>N8`cds|K0V9=G~l`~%8=TkUD+VRF1huD8F(qA4q=ukWu
zjZkv^P<b#HZPbrj?<ASCy|LMDD@>viRl}&Cr&#0=*0Y7>0X<V&GWt<z!lbO{(nWzb
z(YjskvTDimg_RFwgyY6?JVR&4^Oy*{@v&kW>4eaAzLBCOe88z5t|Pv9u>LO@@TEzP
z@iNjX0?)gP2@eZ}UGBn)`A4mtY<|#V|D_Kl58huzE1G$)HMB`t@Nww3+A~1Y-uP1F
zexHCCXi;3lZ?%i8Uw>utxYT{CMVh=rSFZEcyU928SJ9?$zhyV6nsl|%VxS^^dYvZK
z)#4=(5XqV9fo$&8OO;=zu<9t$?O4{G5l*nhyBX3O+M>n1B%qa)C^StfAJ~}JOZiv0
zG>egi188Y#owAIwN@IZz9VCB%k_+SB0DE$aCOC6z-a#ih2cXLjrv*{#*IoH(DR4C$
z&S{oB+r3xz0PKbB^r>~Dtxl~ihmkXJPpfLe^L?d!DT>|{jX)+~*2rJ7LKSP!f{wKA
zpw)aUD2OoYx%#!8h`MCAyGU_3T1Ih-43yY=Ip3=_4!Mgq?PF@1gxQ^mHY-O@%%ZHt
zSoCGi*n7QY2aA_X>YyDhGYPfpDq4~s{y@*yc818j6or4E?0hVz`(&0RY21q~G##N`
z%EC@Ja){h62ch@LHcEof*q%dwCN<8jdE7nL)ek%WANC%~jQ+U@xwBp?)n$jJ(ER!w
zYrSx7YcEtqTTs^Ig_|8|ekvd)up*N|s>2VRM95^@<NMMzL(;J9_(pklmFy8H0UP)l
zWULks)B(<HO-S<ZTM}7=rK0;<7z;<{GfiemAVkA1V(!k{ll)@4mp^vJ>A`3{=KC>g
zX_-vWi!`H`PE1KIHtx5u3BWssDsOS*7*(|sxh1$YbDg(dA~lD)7YB7gYi3!<kd|ud
zmreo<>7e@#m2A3t+nleo->Z=;q3JcGnl9~~<7^st%ox*7C+6<$DOOm!hf*TTDdzLQ
zXMK2HN+u81t{;lGAul)YeBtW$Zpi>}n}&dAy@#{3_%y<?1x{DtJuIK@&@)%`qSQ6r
z4O4WZF_Ad|iSHRb^N#F(wtey*^t|#jOCao;JdR9sav&nP*X82`g65i&kbBL(VJx~y
zE&=t8Iba*Ots$z~wvV7FnisStvZgM#QlN{~Au6TAeb}I*5m|=5ZR<V@p<SNPn?OEz
z$8ZZB%s$fj4SO3!^E04lXe!G)X_g+6s-0u6TXfUmDVaz7KL1N+!IJZJ^9_cYNVZ7*
z955Gbf3%&?_jrpQQ}j$;Xh!G!so)qxXKsQH_rHMRakY?CJ43vG`p|)GWTNEz(k6#-
zS_r9xOBV_h{GtB1voozTR&xFo>l>8I6W(Xs3?4C;YU_eOOy3WM&Vt8Lwgk-gYJ{IO
z-NWz$bLeWgZK&nar-q%-+}B(3;5yNP6b1Ea0v2Vy=g5OcFY)O>*~mkd+>f}k8fxhg
z*qFMIm{vXQaJLb`fgD9vOp*`x{H1JbaZHU{wl1PtkrbaLdI#&lZS~HTqT8O8U&f}E
zCi(o_!s*yeePQktPg{uIo9bYwbl<XLXU0nw>-F<wr?RR~CzE_U=*omT@s*8T?wt|X
zt9rwy3y8<Lx;Xm+`_f&yeNWY;Lvq&E_tKZMYh9O;!NtC()b@u?VmlBO-_<ug=<{?~
z<gI)6`2o1q%Lx#0*A{5)&Ay(A3s2C-W+#m>N54gx!j>wC16SxVF0$sLwSCuIba$Jf
zj?rIkS&}iUhFbf**<(-G40WuleDC&H4mMtP_vr+luXp%V<Zc6^o6QoaasfL%o#wm~
z3>DmrjWU>U4uw(E-496=$2d(}ECGXZaBp))7@WY3a|h;mi%(kXmD5-9mdAe!S~Kn)
zSpA<7j8!4w1DJa*iU4dZ3uLE^m>gIWn#D|r$qd*Bt1%8!>$uk+JU8o}9kw}eA-2y7
zC-2YiXG6GxvN^d~NdHUdt`xqIcCUX1=%LMtV?4R}MyD*O?H4*(F1K0E1Xx6m!5p2d
zdq5F)n{y17%xZkcg-%jFa8xtoTkdX@Y5HPEq2SKA=B>8K&Aaef?)aF{r{Xi_8Cq_a
zqNG&yy#0=J9Ft%ov6oWwzV(vMsAy4{d%iX39D_yaXJ}{pqTBlOTr(zPLPKniXDB8)
zASd!_@Dc+sAwr{S{d@kMZqcO(LN<wVtn5)RZ(LaIbxj8(We|cWb~uzT9Z2nSU#-U5
z8JN#YJ3o2(A(kj1@hUreG{PYC=@?)%Lh|DaDE^>b?>A*%7|S@o0fc)PquzQO{*G#~
z=1vcyMBhU9h@A*D%p1BE^s$~k^`dHJuP3<=i6}vpdWt*eAQ#2PA8$STNuDUHNws;q
z54M<XY1>(2Jm8#MeVlF7SfW7L2T>A?=!#P(!}=U;<_W3@tirx#{ggdcE=yIMZc6S!
z*{#vC6q|coN^FQ~m?=h-NkHIK4W}ctTecFbz=+cNthZhU673_D6`P2Ro4C*0KEgE}
zv^d!s>2^-Hf>DW#WMNxYz7H`rULi%UQ}_2%UPfw#YZPHW4_mBu<p2-L`Jo2YI&K9@
z?ZrqZ+dUv|m%0HZA`nIVXKF>abEKTb)Xzu+6lT=exra`|N9YC9m5HDQhcsSpPww+1
zy79Oy3R{iIO$tKOKBqPJy@i%Bz>A^;qIa<<WP3?hTVY<aD8zw~PP<}<S_@)4y~@z~
z@|ML!)2Ot2PWR4c!A*ouuiTR73JjLY#xO5k4VQkNL816gcf!F`!>#+6i;k?3d3`2_
z%^q^nw^XX@!HIH?%eG*hF{Es~H%GQ@%vvwi32Wg%xP<w&|0TPW?Z3I=6W@3sH|@FR
zmGAY5)Y16IV)Yhn4aJ*++kE@nz_HJ&8um<_zM_K;97ePYTFb3>6httJvb)FkOzZ)<
zFtl~<;U&<ilLYGBecpTElvpq9GY4#TIkQe5R+}-TZIY-kRB|t4-D*)<G7*Lq2P<rW
z=(v(&ksbY(Z<B?%xrKU^il{3Rl<B^rz{2Vtcfu5lEa0+W_Fbb5Y_p@w^`vfhjey!g
zcUJhp2aataXxrUcfW62USZR;ahhk=#I*CV>Lq~I6j8<HH2|9DK1E*uxx;+G)$>YUS
z`rIwDfy125V-A5bFq?Zv?2<D)^xS%gLO3O8KbsoRx&|b4%9DgBGLKuy+R)=L7SG;P
zLrFi&ZmL!`&bQc;?4glyx4N{a=pq^C3-V45b!`Y7o=QLyKg2mWPKSWInLYz2zyUj(
z9#UCIA9y|uIg6$=J`X!sywekr*|<f|)z5v?KT9e1Qm3nS(eNSyp23MD6m8LrTfs3R
z7$rC$U-OpTbGTD%tMRObFvHnQ+{e9dMfbD_wS7;)56zmWC`&?{ptwz#l*V*bGSN9b
zkUYXtp{ET^kEb<rQxM$J&0x~}QS_|Lm%h9++WQ#JES|SX1f_c}Tog=Mx4p*OX;>3V
zJtHe*<CeAfzSM6EDBq)eg~LSR$e>6+gQamE_ny`nzn%7UD>!|?6dumR<YyX%h|%2S
z35>JJzH}g>QP{I>(S)cx;uBFcyvM7-I}10HBGIjMX^|RY!M*2x=2&vYhRPmM&Z48*
z;b)2yXy%hryNaGQ-nd2E!qtRR#vaG;7^w}7T#i4xM@vA_LDgyC8NJ#qiw>sfxxaF6
zcNR96I8{yEmx<72hm7oC-&<H5k?2%*?)&@mrs&~_ruF-@NZP=5P66eBVpG%UoiY4!
zB7}Q&k#~Q*96|=CH8?xqJ19;RDogb1EnDc<`kGE#+^d}-8dHE!>lNb(2UzSK7IRIx
zHDGUa=w~ZNb`1KoI-!dXniU~qEJu9XV<NJaAcRM9>elAxf!WMcD;{fh-5O?R3#G^-
zFXk53ARN1@Z2C0LQY&EM-pP%P3fxb)n7g+M^4aOAHYPD5wDbR8+qvv&k|f7@-%oLt
zGmArI+&#7gNCMoA)H{j`fgB0~G?WG;sMl9~=HZ^Fs;dd82>~<RT~_gDM&2U)Vs7sF
z%(z#Se3N+55%UW}0g}L_a`3m9rN~&Ro=RLZykF2w;+nT894r#*16oE^)8u#rL@wAT
zxe?E0WeQ!*ou-`&#1MBt3BnZ>OF}o)ld7`Fo9V@f8Ql%9V<ucbmk)rTzdO}4hpxpk
zP$htLV|NF&W5#vQJBF<cx{=#p6<yDJ4QEl`o(iljRtbIDo^c{`SrT3D1DHkdAH$|i
zPdM$-N*>@${PColiY70rnpoLNPpovrbk<EcMWuz%g(a!HPH=#Dr5fTJ?4}vo09D;E
z8(*m-QFbPwy~g{2FSy-)?U}~{T?lZnl8T>s(c{@I1rd%h@{T>%{wQB`jP>`p$ondB
zw}OiZO3G6YqLwS}BGg1af%redA|lx@@(DEpB}_V0B1Jx7qm7KBZ0BO~DUZDzD-}AC
zwRkWmr%T=fEsp9;{{UO(scPsAP_e!})dK@ouE>?piThB!qAyX-pm+Q{x_1JSq^qHk
zd%<JW7CA`gtzL;9AFS&DXRU|@m=Ce9>B;rr4DiubG9OhW)4YVUbhYVzR~~eCq?xP8
z&DYUq^Lj4%B$J6%MI!a`RC;W9Mopu0A>fi(qJ}I-mG)21!@_NB?n*s9kE_(VKDH8O
zqEXkl+&wD$x<?1?6){kjv0dJYbp(&J-d&dk;!*dtWB17LVTC;;E%fu%^Hg&p&>^ml
z)kGIk>{@)E{mzw=U606>@Uul|dfaP*Hk4z}M%R|#nWyKS*-PvbHhl{#jTA9T9U{<J
zf6-msYdY)1jZDzbVHCFo&W0Sz4<eG6eZnf<5?z;la(sta5|EJ;`-H&7$dl1i+IOb;
z%tZTWj~f%O4NN@M30A}rUl5rt&8tT;Pc{?e)~Y7#J?AyI8zb#0HK!&0Nv<HgTx{gh
zo$eDVMgkYqI9{`n8!pNB;bmN9$LmSLQ}RZSDv15jV9*bVcNE1<EKn~45uGk}QP3s%
z4q`J`U96~7jEbqM+Q)ZnRmrO^rMzcr2tk3K5BDxpsv1PY19ZeU3M!A2|7ceP6?C}3
z&!ZhJ=ZkgZg>acm!yq3NALzJk=ggK?Q$7~;@(uJD??RtYr~(zegKr4Mga9`RV6t{E
z`&uCN%z!(-ax7{&YrLrastF2}loVt8<r0IIF8T?IKe6LT8BX*QVjSz*1E7-%PC5Jg
zE~~fF(I3x^34*CaPGy}DH<$YYqe!mBZh6%4TwM=q;Z=irg)XVQYh=jp#x#(35U(Am
zU1^W=bnj<Qxj~1yUPZ|sc`{S!qBxY1(Ey!<9$3rIyz|0DQ+FFW&xL5LK<Kk0I=4M=
zWif)zV!s0{6i4ZEuGzDXsyb&&maw34_NdbFgN7>yYSV3aBjFxsqh!S08_S_x0TCYf
zeAbIjJ|bSDC(Jk3EpzjASSxu2Ly|0<EaKqVdX8<QM9fGsW#XPl?5pIJE#?1wlWd6^
zd}8?=^J*Ljg^jD@u1j?X-7s1G1)uQ{Tn(<u_1jk`L+E;(r*|Gle55E*s#3M=8C?=p
z>e>924V#c^8-m38KJMB_WsN>z?S|u;^{%L@vm-m2Lwlo8maoGi0j_HM>gB|tyH-@8
z*u32J+3)O;bI{}3b>~Fk@6W54X*tw|D0R??)Jg|J@*mGNEidZ#f<<LKEoZQml2+8L
zy?213+IEE@V>O{{b9mWy8<8OJny%$Ri5Y2E7b5h^ZPGs@P{clphgP0gm1zQik<6-G
zZ%2iky=7QjO|veHyK4xN!QI{6ZEy?jZiBl!gy8P(?he5nf;$A40Kw%W&-1=}pX)pS
z&iv@<uI{?4R@Kxj>+Z_8D!f5v1+9K$B_K%dx=z)-g|0>fuNQ&^xA&{ph*hUj(h=&o
zz8P{MTty~@Muz}pEZGZoC1&<2-;WfVU9&5xya8zhUb{4FC=z28$ynfb%f6K53u#p4
zmb#w&Cb?kU|ExfB>i(4_5uXi`{!iO%k@^i7wE2SUvIbpatE&}gx9EWPBM&)0!ouoE
z0^y#)v)!W1c%3dzD?*FC{5N0)kr2Pc?+NbcYB;(}PtDqecs>S)KMY%JuoeugAsQ(S
zx__*KGoy|@s*BDQ;s>P2lz+^2zyeRlps`$U7vWb}MUHbDys$Sh>(pfgZj+qreJd_P
zR<Fj4=c2fOa>`UaDz1MW+$<SZR08@bkv|m@cv%Ms>2%Qpe^J$$&r=$BzW(;eu_oGa
zM{gmrUzy@|4?}xcakRF;c3>Ah>AqIVZIEFLcCuajc#7%t!rBDqSf;L7PaF#)C1e*S
zVy7^E7zx?E<>in}rJ)`+^_bsW<Ng}iBtz!Xvr`ZPt7xkh?*s?F`SonJ-kYyC4(Sc0
zv3Zvq!)0(uNG=TQyG=)Ukjg1VpPGL8b?N?cN0bvZwphN{;vf2{C?Z<qa>YfLu;YSE
z<Kb#om#7@gtCi!`B+WP*J51b}fn{@I&+A&*eJ(tv)2u*`K|+tP?9^mfKg=zB=S&aO
zyzC?v0T_<wi!I#<F!b%un?9i(ixB*r75YjAUo{L>QQn<2eu9;u1~rPMM;D3d#HWVK
zbniJ?k}~PGgIW%G{w!?dtsQ-SW><9O5ND*Yt|^ix1?P%A!DH%|%1h{VdX%znB06GW
zwEx~Z0<F<N>-5Q&Ay*T9taKq+G!A=@zlBGuNT_QI#(F#&f#7)9#vU<**@+R?&%tYw
zlV2R-_A`5CF%_S6eOvo|)RBfRTBEu>r&58xa_&PajbzNiSJ4aQ6Oz`l8-5{;C6w+a
zyy&W8%_G;zto?AHl2*l1eTv#m?Eo1Q#IiaHe{mABePGjdDVDg0aMj(G5G7t5MZL;6
zBmt*k*-KMMXFBk7lO_^NN^js|W6eS9C<HfB<V$lRg}{Zb#R7iu)>9Xu7mkmFL0DP8
zU~1*2$AtCju%{%=L%#CHGkxjJs82nK1vFR5{r$Gibl%}>jw`gHb8%&sz-`v+WBbW@
z$PEgg*?nu=)z2Qptnte~?)hi=^QOID&yw!2c$HW+y!fX!cr{LcW%NHeEx(=(s|la-
zLL%8{+<g`9zZ9P={|bS%lOtvumTz|!P5AQxvtusF54`H&_vTkGu|Xmbe>Wq2gymN@
zPf2lHBtwlm&$z5Z$-eQTO1gYpUB{V{R(jQ%QSkQilt@v+F;?Pcr>L<4E0kM4ATAF6
zo;0W|DAQzRt1R1eNJqIBN5+A@1;WhChnIT$;>}X`61lhelrzN}GC1q}mZOL(d1j-W
z$M5LbPK`Var%vlYAmudEWfc`k<L64AEcxxa7vc7Ligl~YA8cwn+9i(b)+~tt-*pEF
zp_CS9|4a=Cg6v(16oEq?)r?c`ZN7JcZjHEGIZqdHxFK2fanbC_p`jF+aQC3kF|Un5
zDLUr#pp1?Z+Lbp#Mu)9Jy$+Eq^WOnIOC*bKSsp+a|GvWs&mXb4ts8>!KDaY7q|T^~
zHshoV)a~%`E3`pS2*dqZhKLnavL+kwXr__`RRfU$`bD2C_Pu#`)5zY+EgpY|<YkZI
z$8vHn@w%Q2skd6(5mXq|GPw8PcJlnd@Li}&aB442crFMJ6&FFrDJx&9_5f33*lYcH
zKV>7DTUd@GxGAQM1w8BHr@G6hN=PjkCCPx`>A>{H(S{CCe+|%-^g;!SBb9gG!Mac6
zglA5rS(L~I?=lj113**w1%(xIxeIz;PYj6!A*<KzT{ll(mkL!hCnhO<m0WGGLA1GM
zBH*)`eUBD=PBQ%VQ@yolNxvhv8FRTi9Gd@R=OnvEy0$3o^=-rRY}8aUE(dEUAe2ko
zw(Y@szEx{Fgn%RWVL-2G#FFWDcD}Low-KrdF2E&1!bM`0u40sLH|urF48PFmImn_^
zP_~a&Cfp+K*aAIEUCoIl4vfE?!0&4L<G}}2e@oR^&%QqztviI5BQ(4yU4Yss7a1D&
z62?#5*6=W1`h1=^!Z{r>uglKpUcsk3x|mx5`rfjUlYg$Ma%>=!K(m>;qPL)Sl-Q6_
zr!G1)hh6Ek_!dk*EB`~^WUfP&%;6{?4&9!PA8D{yxuJ_m14x*DpnSEvqD(n~09<^i
zJuI9$IUSr|QJ*?~X_>#)M&0|8_+)&b<4m0g2aA)B3C7S=JU;@4mr!25a<8|7@h%^T
znQZ)V_Jb7=g#}lw!cmhW{8K>42pm))Jo&uxQF(;X>=CwyDpCVSt@~vqlvZ5=2;5sN
zJM`fTt2bMxYrT$>l#jF5#W+{yn*W^PmO&1+%0wFw(kd$HjoBiz9#M~xm4A;BL$5>z
zMgxvg#T!p{dJ=zBvHy5xG@yaCa$ltS1%m%evw12Fp*F;ZVujxERuIzSqQAp~-u!iz
z=%lO-CYxM#>rQFi%ws=h0KHNopmD5gC?SgiHaz3Xf$rWdp@RB$l+LbXy3eA;lfQI5
z_m`}U8Gtti1z1-~GakFBsAL}hV};;jH#N>btFcDh0mfJK*gx$M!_XIIBGHzpG_<Ak
z$1$7nL{3NwCysNZ1`mY}+xdJ#k{Khz1T%TQ%whsr506tYEX!j9k?KjmVTk?cu<*cA
zRsYGmDf!ZSF6Prdo+vbp(IGY!>Tefe2eqBsef#wt?NCg=Qs(`M(98R+tn=PCU*a=X
zT6Fn^-tX@+dQCk1onqODpywxcI`=!m5CwhTT0&=tLwetdgq`QCUTx3OBGh(_CJAiR
zBkOVz(9djgwTNXVfOPL*7_kOPH6O4Ot?1(^-!gr!IkLW<2_bl*FeTJNzv?f-iQOHK
z9~Z+hw$5mLCkg>n&#rJ#w$QjTaV*|SGwj}dnbuvy=e7?Wyv0qKqFbu7;keGP1;T_k
zpzLmq)31pz!0v>_+DG9O+=N`0gxp<mqj@oImTlM7iIrAN)<J`&#;{@Lz+bT97tN>G
ztrN6LBhilogpCqtJ5b<*fN)esz#XcS#I^4>qMaXqtcSP8FQrBR3gyKh0!(|ey&`Wa
zQ&1q`ph7P7RQMCJWjW7qSgQ(Au41%Sb>*bnS!tX`{`Sm;n7lpLXOdHX{I1!ils|-N
zHvy66j@l5zlXWBWt*>{3!@M)UE1!sV$`427TF@;!b0LvId~194i0mvAqN?gO5b!&O
z0V;u%q7<(A!@8{Iyk`~gbPHteESFgw7U#h=VrB5n1O8GpKj5F8G&6FYTOo$Ed@f2j
zff@MO%d|LN=%vV%CoOPV$g33L&Ln|9NAeudGg+?rE{o(pQ;jX_v>;;~m1FRsqZj@v
znjKx%0=<r7eIs-1VE8QPZ+DQ<)`$*uoM+I3e|_uLY3y+zuikf*hXGy+y|&OwCTY})
z?FL`a?y`+xcaQSqUDvijW^ryA(M&Un*~MB%kvCN0mPu1L$n)wGEH38;`2FZEkGYt^
zue3VJWnW4jUbCP4bLu&ggeZi+)A`v-Hi%~^!1f6Ki(@T)q(lEXygeaSfJ2^jx7*dj
zayNH;|9G7jr#^Mu4Zdt(q;N(`*E-*RZnKmTC7n_H2-qSFaucAwAjqef*O*bBOQ?!+
z0c_$sDZb3I(OQ@i$H!fRSV?3Wle2<(R!<4d<B$9=h<jpZy=x$qqvX|y)FtkKvKL7;
zy8Y2SJ#A`F`~3tVE%%UWM#<Xd&SEPVTVD%DCv4pGkJTQlrR^sLUUnHttk%udl8OZB
zB8sY20jff-_<#J9&9oYDr`2mBJIZi&76K%tAY*ARiPasBJ5eclzGgU-=47?C%2W9w
z2McD|DBrA+MLv$OmAq;bIm+pSfZlRf%Oz{br<GbzO}h$K5fojy6;W1!mqxTTquIN2
z;?J+U5o@H+gG<~&^jyov#2IM5?){!n>?0ILO$S5uP==6oVbk~lcozQsaY?Vw(YRTY
zYI#$L&-ugKJrxq8DN*caC#HF6JrMjH^{i4Kr-0noHj{*g`k`un{1ef`Sc1h~#?N+x
z_ZwCy;!P8ldKFE4XCTu`Y^}t;SHQQ5PYPJ(K~u7RQ9rW%;a=_hbn~NORL6_d=W}gP
zus8us=uDA!tEMx7!uBm>;=9yBdBf~jJ=GUtvh4OQD^Wx8CSfG%KMOn2wzxLGb9a=O
z%$aSY+g;wM-h25MCzE|s@*LgzQZqheI*mhk%Eo!?*wyb$Yd&QiJ*-lCH1SLGY2Ayo
zhX@{W#^o8*n1;#4S*XhfVianj)OeV?aV736-rvG3%7ru&FcRE+F5hpmtBXv<e#Iap
zkJVbT#s4HJ{-sfue%(R_GLk4-q^s&O4YpJ~-MhTC%p;h_<72f76s9*|Uc?s|BjhYD
zCS1wO9A~HK(X3E8qPDd~9nI3ozc+8~H#_u9->nH158H+J16jnpjo>I-iC^7G&zkGY
zgbds{@+#{DZQmDJo(;TCr_<$9?pV8)mU{_5$#z$L71`fjMZsqF_i0K_Po?S^oRXsp
zqyP|%Vs)?@gybw=hpJ@E$1cxTcR4HHSvxFL<XbUX^E|k(AX=hCt1ZH5R6WFsjwNjt
zOGC?HCppzg>+^uV($han>%p>i(^FfaJl36!iKq*-V<23BD)^|>fCKQ@4W6QuC)4^;
zS!&7F`cf#>SEkX9Ax|E36c_Rdt2{3ja(rusEMJYsh-D$?nDuu3nIX)@pJl&@X%WbL
zALGN*PhfqLxMWok1GMi|@}!p`&UHt{)IVNEo&hC!irDQ7`0^TkUx9CQxmsnWD%X|k
zOUPJ;%3_XT;#voVrjV{Q(rvyvf`KIORE_({P+6*1AT8(nHibH)=~cP!l7VCaEHI6s
zhOAK_1{Z&#wj{g~!0Y1bYosvsH$X`;xm<z>GmqZ)H-v0b05;<ON?;LJ)67x22<+#|
zc`3Og%-PiAb8XHWK3?b1tvHz%VfcwON|F+LB%h%D1o(bN;vB4qwFo_`GggYvUvL0g
z38)9!drDCtPd#GgDt0y&A<lXVUEBIC-X`($ol>(UwyZ%S_Yh%?L1N*>$SgKp_!3u<
zHzQc3$g+KK1M*B*fJaD&y?nTa=7bD=@XP+bMdJ|#GLuL#<gItV?${D<uEOu^{ORW`
zs}iGNw9`)HdGMqkM<iS8Sq}AnSFsqRRUQo%2mZDSCRLJ?V};p4fk^#)IEB8zI;IGh
znA%CoSAy{GXzq1S*)a&c)VJEb22m%xyH@;~VJa2ZZ7adPB=b<MJK$)KmOJnY+B~@y
zB3xpYJ5lu|1c_eE#-skmFilW77eCHRc=YHEQ3;PRH>lG61%$^_TYU~U`t)<q6Ej<V
z&YBSv_fG@Y+~kY{Aq@-UE@58ssDis#`JX+Pq*8xytN!d%5H!nvYmGb{MU_@*#AR2_
zz|%QVNaS!h9Voz0QDG{x$g_&Ng6X=)$a3fo|ILT)oNf_9qOM3)w3(eFf^0MjM18Uj
zR9MG_h=;;GBV1MW_<<$kKqs8ExozSz&DX0{M#L|^cG=mEZb)d<F6iOkaHNoR`f|>G
zLD>nwk`J#H!xhULJcQdy0?wwl9#W;F9iN_et(%Ho*5z4!$;H9%KVubSWZb%oqhO)&
z$4n6Qh=Kb#k>@h`An=TcC(^X9W;0kEw2Co-fq|yH(PrVZ?f-o_mGxt7FDa93`xDOs
zpPfnsTwB)XHIC9gI^qM<-^4qA`sj|YW7QQ5b7^3BRPt~^7(6E@roZz`!4Y+hJLkn^
zbgfq&AI!3Ckvppxl5(<3!v)6lCBa#@MVuppeV(0gr`ktdReO$31M^`>8(ZAZXW{7&
zJ5KX0_1gzBF$!ldob@&}JBIQ<e~xN&LT%)b<BY{8!BF`j^?TzI-r0Ui*3<L|$@DDZ
z1^QgLePMwP)hm<Qm0UGkXOcP-(f~}=#l(F!Mp^9RS$-<;p=9IoQLCJ>%V-WR(9`I9
zK0lk3u%#8ss>zP-LLur5Hp*8Ij)Ku<y?C^%%)tY|VM1kAd`v3WK&xn>xXy@>zf+rN
zVG!YR8|>jc`9W8Is4~p7y*r}sUEw8oAVn>a&5zT7K!zu}Z`7XQ8~!4VMm1m2Mbid_
zV|?JXpQF9P4<!c>KH(TbD-sAkkn7e<ULqB1YhJH5Z_D~K?8}Mqz;S<X8_dUh0{ztH
zhAv>vyKJrcwPK?zFNe?|CSpp0EmHtP?)dE0T*Zd#;eeQa!j{wqf8Gb3gK?E?3b38&
zrKrT(j6o+l7o*1d^-15dL$@NBgm^N|Q@ct=_N4Ui#^YM<Q0?IL*S+DNAcYH9R2%V4
z)78myCu`d-$y~uq;MWr(s_V@sktqzWYq(S5*kx>jW`ZcXH7xCUC&`z9UF1W#NxqEP
zh@hIrDgd3F%9kJ3o?#kX)aatKP^?`N#qf!52qzXwq8es%v?7rdG$!!I1{H$QGZ>R*
z!dh=#`i3AgaO@QI$}xuraPm<UrE2kP(>a#@8pGIiqNC4!>10c}hnKK&GV5`8@y8L*
zutS4Yhe<s-7F?&!V}&z=ucC8eIF?t@LdR>>YWd>G>teKtuC!+;cbkDRHdT7xKA(8x
zD*K#Uk-1i#glJM>eC{;j7YWbpSbsgXymv~e6xhn0TKwv4AWS>(e({~K+#@(rqEN4s
z5Y@-faBfml-+ty%2%U0a*^xdhgh$ue{&?B`p1WF7e*e#TG~v1jFeK9eckz#c2GpuR
zeMcNq<3yvq%;-V>r&`^1hs#1FWP1AC+3ylmPb&2U^}_ump{(lwfBX#NX%V%T1O-7W
z?B5gKh6;v#;&6n77(afZHN%S6NlbT2g4~H`A`0P`zM%`v$d`3-1W=`Dek0n!bh)%o
zHs)m^Y)9~-bloh^b3s=ZM+MC=te>Z>!AX`jV%H`=o##SFn)nSdjBl9kAlYqvj+&wR
zggmivx^gKMpUSP$LH`@9<+HE|-br0>pn{lut0w#x1T}oPmR<|vH4JtT3KyRIA|(An
z4vPkl#huWp=$;40t?(Xn;Z=Ozut`1Niic^<(k<WR=x25}7Nde^ZYnXG`wUNg?hxU=
zrLehjeuhs4&^J)}xuH@Q`5sXD6on(|Vtm5X^XNSN_d0|6=<LLQG@iV$4<9=+R7hXJ
zxzH<!Xi_FJYHNaKRZ<Kr6~FPSp4cY2Hwz_SkTjZ6CM_-$t5Dh5NGBRVtu!X8>9NHb
zXzW6GlC2BnK{)$oMJI<i!^OBtA+p<Xx#H7dVtC%2WoqIEB<xA(UVbPr@h9&tGFjyl
z2`Y-3VPhr)=pm?{Oly*s*X);;vJ}ONcS-HGkIy%blbu9{#1bkE!0$QIuMG<yiD&tL
z#2>6QFY}Q3j-88oH$yrRm{Ss?1IH#s0z_QrmMvrvwM!|;2UvoD5O6WQti*LB29W?8
zi1z+sO-(%qm82>7JZB(W>lj9in+4^UXW<{Yhw?{b@o?hCfLURVE4}^$g&Kx3kvwt@
z*>DFg7E!*zG>s0;#|tbOy=4;kerHYe5iL-h(6WiQ8b}Aqtxh@&;bAF?2E@Ox^E-V#
zs*~uT#RS)i3ne#_G&g+4I1laK$--jnwaPI=5P5WZh@`q(@Q@bh+h+uChD)g^OGDD*
zMBrcKB{gZ9{lz(vyJ^FDyLW5pIN=<X#yss=Ta?LMIC2q@L4M3V$q_+`oj6DJ&<h&z
zi=-ql_aJ|mOWTvpHp#7Cs{IJF4o&&1lUsT>#n|lykc|wD9nXoP+PaghvG&#aa~loI
z{SA#13zm*XK8$W~v^lEjuMkx#(`C^vDjokA+g{uurY0dRVSX&5WLTgYA#H!Scnm)}
zO|Frul3vTV3ZaEyc4`gy+r0Fe$ipl`4`|g^i?V`WoT%fc#*s4%Z<}}^1bL<M%22(|
zB%s_fo`l63<C7#Xfr4b()`AkGTm|c59sIQ~pAih^0XXmrUl@kHodcA%LOPs14Xux@
z;`cSot4>GR1R>o8HW!nFq;klHtMLfX=X-jmINCs%QZO#XRc*6QC`C4&rIvnA;pzIb
zq_xF5Lf+ZDDfrXM?mTF|*~|cUW?h3EE;&IYbngejwe=w7dVSVZK?8wjWiC0dcXK>x
ze1Aek`)t1qXx>25%o6GfM<&mq*s)t1j-0+wWaT@b9mX1I6yz%X7%7yB1u=HyCbTdj
zVk85V^^dR<{`>{+G`RCI?#iq9555tLZCTBcH-83I`9$Ybc$C0k-=3!l>rWP3V_O%*
zmpw;c0?Of+cncJl^8*IE;Mtw*{SmVu%)WtQO%;av7U|d}Kg;5w<`v3IKZoyB(`cvN
zAL<NkIE|YpGJ!L2QOsnK{%*7&VqTvEVSg7opbV$WYzZml`oJyio788E>uv6P(jm@+
zXy?jwPc%N%G)fd1l|5Lom|74nlUKZLMT(oz6>T=3<%4iF)?kghpZnEsjn&7~PFzVJ
z8YfV6SVz!)<X=)#%~yxKi=J<}D|f<F)=X98#3<wG@jd#+-Y$g~+{+G@nzQN>3MqDG
zPoF+-I_b#!rM4jS)Eq7ZL*oJ?)JvH5qLfJXQo7N_eT3!|u3~0}E%$H)*?~;zRElt$
z;}bN&oZuj-ZM-QM_M*DoyboI_(BcL#PejgWJ=&$VY}SQaYv#<DRBJLdrYbvT+H^-0
z&@UYFK4qkaxuL9b(C*sBjZAyevTb(~dw{EYO4k-MLXEdyme*TcW$^>dLMC@5cDYh|
zapK0)v?ok8j8TMki3A2iG`x2#{FiX(6y8~8F;5X6GPYn;o_S8&WfHNPFe4#*s4oVe
z-KS_+@{*4vA2iN#WfnD;<XQM<lM=jth$GTB><Zw1L0b{_d?Ri2Xz?1r-T&OCjwnwv
zBQ33!!X=xD90Q!oOq@1MqiD}dtsN7{+qd4QIARxR<%A!GQHl8}s|!lBRt&T%Nno>u
zKkG!z7SAHiMsRn_nHz{^h7z$S>lQZ_!mJRsr{$!b7MSYq+8?hkg5trijIAc>J`m?b
zO1Pk#7h4jVgJMfGi;hJ{liN30O2;*@;wrTlz_LLKu)g-trJti?G_*fz)f06-|5RWK
z)gPJIAr~bXRKRZcz=jAbjGOKoJvsBOPEwgjG#8MBE_{D0j)9$sM^&^c1IBMlA+@dw
z%QdxklmE?CYVr0PEQc*H+l8K-Coy0Q4gkSd{l^|o<|@?Ly7J(1@mQfqGmqMOZZ}(d
z2jS@^fau2CI%I8nL>lk>mJjJmIbku0ZqJr6)a1|tU*9cBV<UwyvM?qF_=&mu?A#Yr
zk2=9$=u{d?pS(?~;Xot=1rCa}esYZ`#phw)v*JskOq-fGWE<VIkH3W$5YPfoQhjtX
z{$yP~2EnFGwSc;)Ix6L@<kj8ri$iXgLv+{7DxGyn05U8UlH1V8NpMMPgJGmRS3iX-
zqJyKd4+>9meJVSd*TTz{o@NU|m#%iHttci1%sQP0VZ_;p;jq}Diz6!q_)u^@PZ#?a
z_;oOS)?b)mRr<)>#cCY&kv7v_m!bLLuJ%z49mj<>ciEsGnGVW1rpxVd*sHSa;~m#a
zXb$RiSM$mamh*7IdM{ahZ?o7D-vW*hB^#=D$O0U5e?l>Dvjkcrc#N_q_RnCw{2hdg
zXHd|ixiV(vy{~RJ0^me>>03k+j^g(3D{8eB?q(?~d5a^O)Eb?6!7#K{P7{&YW(`vV
zC@*1LEl+)1jqIu@zkn6h95*_K;Oqgd&zlXi{ECvPdA|(XpG8rJi>4ve#sTugpgQEV
znO{5eJV9aLG^6Ipkr+1e1=djHG`bP*X(`Q!h0tjSJB|SiZr}@*I-hJw+r7VWk<gR2
zyB6yPt=r5PhMC`4AtSRCz$qqTawK3YCdZ-=sFcyK9ev95fn(C;OSy+L@34%gO6Lw~
zaC0v|UO!JaMyL4J5;STXQGHNW+%&+=)=xaHB2OI)LlxI0OU&2DerC6_QoE|Ec}`c?
zoDT*LLTBmyQIBTJ$|sB!tgF0B*um*Zp}a71j;`o1F0)F?2~Hrque4aq4}o0_rDjU?
zCWBlnbK(EWkeD-JM=Dzu8N5t)L}%xHz8{f6jYX1h(OJZ~4pqJ&=tQZgO;j4Zi!<V_
zL&|7aei+fnQyqmw1wO{$!8%S2K;4n<tlv7$4g;FZPhvwcu}Zfp3a%uGmX)jM*(F&Z
zBlF~C`NK3U9(^L_#!QgE%Kyn<=eGHH>X_D$JeX8?;DCJVj-ob4+0-RUTioioEP0Av
zU^+PCQwR4Y2=9eSEb#k^0Vq0kDquZt*X|)$W6I>$o_ZY2VTLknrsy<<2M0xX6731B
zPz(iWDNhkgT3*0}8CZmqB;Cx6C+lHGHJ%boLZwbh6MsHTzq`_e^fyJGw_J)%c*i2%
zni%}njnFUrjd&b(fVedC<v>4F8u2X#j$&wcf_80w)X#x|IyA}yf<MqF%{UP9$kZEi
zgIT6Svtv2J_62^#um@D*KZi3K$||2%#}9v-EU1rd<8~E%j^VLz*=pod+HX^W#Xcyk
zm{ontcC}~OK>G%N#RYefVb-G7TP2R;6HMi_rQGr{*+6TldbaLS`JrFDvXvKM`{W|?
z+UH*AxU*n#Z#}G4i_3HHk*Sxu6}9oY%}bm4N!xzm1jrU2MD)2cH;S$u@e7p1sLu^w
zt*)e#m!Dusv8EA-wYpp7M_6ZsJ`ZWRI5XDr?#7R+ulm2BDm1_&^++)gWybY6R%bUO
zR?#+0d{!tpDwS`&6Zvf}p}fOj2#Eu_SK#jU;`7lDT5>aZKzIkj2OF&cYse3JyB@Nk
zUit;zcevz5r#dd)5%{pTzs0VnTKhsjy>NBG7A^8=<?Lnv$v^LkbjYaduRZ;aH`9Yc
z=Qqb(D<Zile=u(AjLsIFj@{hfzIg5YSo|CoN73*d|7TJ|jk@Nv!7o8U{G;a^pB3O?
z&t-OVo)AwS_-shp_L&`I<4bIMqe%bP);vhyDN&&iAJ4FQy3Vn(4wjHjc#@{S?qumM
zwK6MW(;~K=RqXlgMYq2DXoL&q8r04hlPCurMc~K=Hny{wN*YOj!7*##9ML#rBYPOb
zqr%tKSd6SHMlTNYT5;12yOZ@jiZ$&zIc0S_A~s#>TjX*oFBP?A!t&T6^~@d2C(qo(
z9G5?-(M6b&Gt$TG*0r3z09&w!*MvAEK?mNRLvL!vRn??^lZHYZRRub%?|-r@P&uj@
z&>!%%R@cq>bW&(X^F+@H!F0%OI3e<eUgr<fXi=FKWzicCt3sAFHw4ddkT*Yi;L@2~
zL)Rfn$BJvaf?#_V=jw6dSR_Tkf!N~VRTymt^6~~ZR75-NcoxEqD?wa_t^o|I%rXm9
z7?@s1^goxD3lLBoP6@+S`K<G8xr_;io?d(y2g*DS52CF7T|=L~rdYkn*V<nk)7c%H
z$5R~PYy{8Ee_SP$oS^!Mds>jziEN9}iX`s)mlvfQndw(gq#sepU#JqL%oH+W+umgk
z5P?3$^QnJJ;6x@J<P2zLHbPeOUg3TsuY&p7cMi1>FLzW$Py~AuOO58H;nR|d!4pW)
zGz;qjYoHZdFy7^^P3zQ!uRdMCPMraMR`y34UHxO5fucOlD%smx{-e74&b!^p1@>W*
z<dbP@iFAhoWSVODEz=n?4z<Ds^%H9v@<US#&;t#+&wX<{CKqmfCt5!m0&^VI(XzM5
zgC$=j22km(LROdkS0TlgI913u_L;pl*SOkW>f$Nk7a4O}Rz}R<x~k#~Bvo4EQ{ZeI
zOrru1>x)j$?XAl{*`5#{87h49lQ2NRVmH*%cqQa{?$m0O(!MAn>p^NW_G{u>OdoNP
zTA%?LaSW-AyVWj=!g<PqCRjblBPBR--mlp)`k(pnbV=N03y00?wB{-L*O{}(_XjzN
zSvqJvJ2fBt5-I9=!iAh>S4a7OYfp^8+4J@K+!ta`*v*8n#eGNZ!(UnW&aW_{L;6sZ
zd$q@ZlAa7+pF4d=HS0^sVd{p-7;VHUWZKt>>K*%a^$<G%Q5~I1``KU79WTQ|yX*F)
zQf4ze@MMUB=(RI!0|>yTKjYsdjnvG5O$3}Qh`HSWi1t|#8<FVLjkM|AT+_wSz><E-
z#&AYZpo4;yN$k3YL&Qt4cIx;CBno8*uXHVSKna^r*u-hM(>G9(CiZEcMNVT;U~}30
zHOQ2+{24tvJGJ@$v|f`)VZ!q*Vd!7HAx`_8h-Yx$Rx%a2U-4c$=Z<s!yqg$wb>7sG
zH>1oAlbe1!+r0Lep1z=zsJF~H4Zw-lBm*|`chZ8=HYA;0W`(998AUYMz;MbvA*Hdj
z#!0_*E*9iXnjqfN#q0UaS!!pO{&u_>Sz?b9iYwp|ueWZgfCa>KhEbZh#q!lOZJpWP
zh-+XELG@f-n*@JudJC_g&<TU`jE*B=A!IospOJTyR5}ZMq6K~T)F7Toy5JAx6|We^
zkgUnmW2gdNEBY6t;289Ais9w|pwA_PQx{AmVBno?@J19D%7r-z#-sw?ton$=yrOG=
z+vq0B)s<shHk>`jOiZT9hvq*h{u!bmXNik`#Le2w#oRnm9T`8<5FAB!8%<jV-Okgd
zh&%tuqw5C-&4q58MBI$@FJ$V7$H2p;6Nnw8&eHhXfMB*=XceWFl-St;FIbuQ@)>7)
zqh_`CcnEtoh)N9T!&~^&BNr$Sb%UzsrksveOg@I$ITa6v7{y$KVcq`xL*`=+)>rV&
zAJrn9_rfm~742|5E+}Pc>-C&74M+>eCSumy&{Rmf{z+8M;31Z~^F>qCAL&i96Y1^7
zA!Zs<xc{J=?(@PX9SIF)AgCpZCDrc_)vgslBxaUyxeez=CP7-diZqGJgWbb5=F_n(
zjB;mV|3K^`v=lu{3tK!7`qZ3Fuo6L>I|jU7+1{&L>eE3w_n2W0y}5eW$KapZ1${TE
zxHLQr1cQ)vnrz*hlP`ZVQSa)i3K=UPsAkR7upS#nHbuEXxO*(k1hv>%)&pG*w~3Zx
zpq9dnK4W>36*&fBiEu(;vVCG$B&<cU;^h*Gf`(a5l#d}e$1#G}wJ{@MXy&sJ%~MwV
zVZ=Qs*k6k7cpXU>=qj~9!ceUP2OilQdZS^hosFm-K@@v#=n))&K(jU1n8wPLGxC#D
za(tEVES+we1n#G1G~gS+ghi~9&Len_DPyL6&zBSl!Fpk(z^(AJ&%Mjm6E~(Bib_AJ
zy0cB081b#|y(U{B=BoEjR4VJy%A@%`x#1vh5DfAeH|0xg($SfMPokQu@KgGuq*dF_
zH7p#(S517y>@%D1;@(ngDaufhW7W;+J_Rcflq?kM!Ceyh6$)EPt9tZ@P5UJ*cWp`F
zxy4g_qK;Lawe%}qb_JegFfxNclM@WZXcY7cYv>80N|BgmZi1%GR@PFLxz^gJ!Oj^n
z(+K=)oL%G8Rqxj{m#ByxaOD*CuLrEcN8H_HXIYe@zsgUoOBMNG)#=GFfIjXvzd#W-
z&8SPI79Q&>r8q&Xe6mCa{Tafp1yfl)g8F<N7!;eU+*r*Z@Fuq}#JJwc@P$a(<DM{u
zFi<o3r_sH-g!~H19gIiH1y!%aA(5KrZBA4?QgKL#3E-B%M7dO<wBLi+_O`Ljnn~jD
zT)zge2F9<yHrCFXRD}?W2El!WO}akRo=yj!41Rj3lO!?-?<gvNWHjDs{(~AhCAxzM
zI+@Ss^Zu0q#fndiX|)n`N~&7rTw((`PRonC5M145RHr)MN<C9O#}9C3Oc&u}{4ADh
zD&le9-EJ-tzrS->G|WqDy!vw6Mv!Y6qYyWPA^SahWF@`$<yZWn)h*5o(Bt7_vz>);
zv7<<Q|1aT-Rh|b#Uj%#sbeO(9H*%s~V<nzQF`3w&D<U6@IkWxd2BX3(?h^R~&$)Sz
zikJfMHVWKf(b&i0RTbE}7+q5MPKsU7Yj&9TcMXXVWv;Et@mn@%mrp|SooA;xdPN5C
z0#vUlWEzrDE%OIE+7=xi2Tav>3G=zd=D+H6U*?A#O%^|!Wy5(EK1P_0=D6|j9=XI!
zndKe^@mBE8eL0or4zXHFwMqxxz#UHEzQ@`HkN~1>d?bcH)iClLrwpj2HDy+*4fc{r
z&q)ydEEJo9Xpk5swF&NQ+n7CU!)bcl)E;*i<5wy(m$oXRfZLN4jT7{_<e+*QLh|`l
z)eFWy?oj4y%~%B2$lsk#)N?tSXRAEPUKrQzL^VO^E3oQS9=lK`A<&{hOoikgLyo8n
z5mH!gdb+|%M54+~tv+DE0Mib1TNtygy9*+TE^T^l(-H*Y&&a9<*KHz_`tvpA>LnKo
ziJdnVEk{G<!jB>E)Vfx<21kmq4m#;Y;4F*;uP%zigjk-gDQ4V=<NijsyXYhr$(83v
z2`krKipn3@xAvSmO?8bUdSsT{+nFKGM+7;xf}xY%)Ko)p%;pH=nuhj3hI>c#WN}F5
zMv3f|3E+||LaDev3SZX4QXM1XB&^-xITUE;+5bwGhFhQKEL^$%nEqBos6V%XfjCK=
z+a%}0=jCLRr|^E!xD-xpQdr88Efe{Jm2AMsO>&hsn>axEhA^Vo(&OvMlsw{fD*3Q$
z{1%Y)2;H38J#a={iLjJYmNqe6c_t~@w-2+hD~o2_EwI8#wR!2je@d3WH&*8*I|2p@
zAH?4>N_b@En8T2lQBtzVyWJUsXC>@|Xf~?KZ(T`#Hce(2FD&fJ25jY7^@_#k!5EsE
z&u-n~jN(H`A_`AMin=5^AR4KZ^8flrJwT$UhKzc33KBq?4>|EMBO@plmNF*hx{W7N
z#4zmC<P}Czm~!7L&(*R{gqZS_<nG|rpFzp1q;nz3T+~^w0|P!6iZT|}h|P74CGY~l
z2SGeHmm7YFP^9hfCm@IZ6lcPDCKo&qM-%i#d9y9DOJvB}yro({7R*{tf7$k&0qG*)
zQZVjd1jbS7DBIDvFh(LRH86tTe5T~lt;6mP%AxlGt<^d0yxpm%`loEZ3VkNpZC%O1
zU~)RAQLR{vKagc3w<{*zJm{guoV7r|dPY9fLq^)@(2<i`rYj!@H3bbhpMvjt#VB4i
z_yEaJx&fyK<Kx;~UXEJeo=-$c9~LKW(TXS!<6aVE_pu+I0V(bFDJX6+1)z8nWBI27
zgm8k~4wOVxPX>xRYtiH^GcLYu&*pdj@O6SwF%*N>8zS8krHJYeIhRd_=Fgt^IX{Zx
zCISzBtY&3sip$>esov>0ETLfo>Whh2jrrjqM|FpvBF-r)GRerTbP^>I-h|fN)JO#%
zdv&y@Ql&GQ>^BQU)$jFG#g#fFWDa@vq&~@YP8I_SKN0Rk;8gRWFz3Sk8^xIOo{Kw$
z&?ogp?6bAE^P_FcMP{z^O95%Qhi#3S@9Wo8p`IBNLTPsfHVqDTQMZA$n~Nr0zZ~Ut
z>4dPE+m3z(L1nQSIwgX5+H9z^Z#C1BNQQk&a)j5Rn$$D}*o;{Gi|Gf+yf`3=5Q_ct
z70MxBQsQ#5<KEwXd%sL7`aL`xPhaN565jQ_zk__GVMgFYAYP_xz&y!j%d&f3-lp$i
zdpvRI6}Y1ARUZ#)V4oKC-#1{J&aH|Z-(T-}W(1!`7X|)k66?p|FEgT?zCWA4HV|Ar
zQhmt@9=i~H-%-?mmzcqQnf|i%ax;A?bsO`=^1i>3>*Ica{^kAD?|mTWqUkke`ZQ|@
zxN!1<55XJuFrMT0r0C~ERLJ@A`tX<|aDn#a_l=<M%f{u~pL@Ueo2_i6Y1plc$6wP_
z%?8g8iR~eAo}d2A>%V4FKK~Nh(z`kAN}Rsee!B`n{8dxo_BFHT+*6up!g=ix>^`n$
zDQq3R{rkIfO^Xj+k<pN|9P&fWVM4ocH$O)6G2zWtcMsuK=Hc2}-uR1m%?+xc_rpy8
zYM4yHy?qBhc2#Jzpul*+na$=m=iYl3BUPligO@|WxAFJ4SM#Q=*BZaK6v4N*o7TY|
z+AaN^d%>3>)_d9qZ^oUiju}Pooj+eXfo0{>e(yY&srnrghf54M{C9EHEasO>Bi@3G
z2EWc~=z`JgaJ6E@d<b_GyxqD4KYgTcPY~sgsd3NId(jY>F?AF?)8)Kj6x`~0Er}EK
ze%S_hKinyb!-`$A_k>XP@^W8Zh8Oe&WBw6#^!END_;x#dd3oP5(A}}Yyy8|Z=v^X5
zecbBzu9$K=InXuifx2g&AUIk)tBT^{J?e~Q!Ng{jA3VaaGx+6-bhfi0^K0AC^h4Z3
zl>W=@`{@k@^K#}`hwmOwh400%>QiBA%j=j7e$d4y{wu%dq4_u0nftissXsSaXQ-Sg
z{VeA6Z>ky3Tkg`q8(dIbZcjCxeFDEw9)EK@oTu8?94fGyKUyz|q|ts4KV*LUX-zzN
z(?-5`MXNYGyiP2*InYgb6J>3fo^#39!x}P2@RH>BuJ0^#!h&FGXW|TUGBvdQTeCN^
zLSP3l0semT@gXp(dN`N@7=ec7rU;Cp_I4msJCHMg<70p_0;962v%QOxv8gkF?O%et
zsfnebh`l>Nhv|dB%F4vR&JAGx_~K+^)kXL~ar$5fFf;!%T0r2x^8RxFM;_oqh>g9I
zii4rCDS-9wxWB?dAX6th0L#Bkad(iU3dj&-3it;HfWRp6p%;LU55OqF$_)6J3H6sr
zf|Uut_1`KhfcrmHmJfaY>ZT0%5YGs7vNu*S1p#zEgosH17*$Q(K>%HZ|JCfTgnu>r
zU)o<a{{OVUn*K`z02m$2%^aM}jQ~3TqdB9PsjH>2sj{Sq?q9k8F*Eajk#yLZSsAzi
z|8FG=7drzx;QuT+xw#qG0Bj$0*8i0MQaCw390A~D;rx)r$;R^0%gMzK;P_`W$44I*
zfQyOa1MA;$tengMb`~y%4~u?`U;%La!@<qQ@}K^HG5_WM4?g=}ZuSqp|DgXX@1K&H
z>7(<1SpOk$`~&{4;{P7{|6Tt-DSv_4K2W&+5y8gJ`GLpE%=TZief0i2`XAbdvTXlp
zbN!E6|1t321N*0+jfn-o#>CFRru#oG`L82&{=xcR`hS0-IvmW*49x#a|8J=1u(Pp!
z96TEv(_hnn)IZ$G%JP@O$@!r%8#mWKCDVsF*|}N%YR=C5;S4sGe<-Z%A9U6a<8uPI
zKlJ%<3JcQ*59eQVGP8eRv9bSk0UHPFUmniCSRd~B2aEZyynh_T$;!dN@xPk=jnV%E
z6{DP~ojJ&Y8NkfO%7pNr)`v$qSULaO`Z(_YRsozG%&dTaJ|CfFY4Z_#AK_&45pZIr
z#`Y$rf5YQH{}=A7F(7Ewg^iyASu#QX{-QFhASpLDX}AHVUNlxF)<pkC5<;002n?AN
zP*CA_nPEV#R2j&-p!>f2zU%#6d#lB)YAdtrzTtl5IrES^voe$dLMGM$Hk^kb6ax(h
zY79*E_O`JQTqw5`fJP<KKWd2}-X{9yl;DsqiY7%xKymX03>yk6xc@;_m<mY&6$R45
zs{=Av9K0WcrT;sN5J2c7nf4nlTI@IQ(m?OrSTKHJe+~i|;qMORz5K7}KyzFyvkX6a
z@O1+s|JuezviD1n|94m@aOfZdFn+W#=-q%7bR;L()ZRWI+}QgU@>Ww_AeAB>F|oeB
zzVJ6#Wwcm_<ij1<2N5L(Fk?83z^ec*2w!BR8CVaYH#rKaet0IA0Fw6y`E3O%eKZJ&
zUas{#1309J(W_8y5j3!gUtpN@y8Z!_P<ODE&#=9y52rR@{ek`O4EGv$H6p;}OB1wD
zNcUGzeJ^=T=Kh!qgb+-?C2X<Bya#YV>K-Ftgb;psKge4M==Kc~=BmRPO-OeX4y+&P
zVaZ*J3F0I|1c4;)qDKPjNe#w%%!)}8yEP0JDpchA%ZVa(q`yD&@CD=Tq1lFkgT7|3
zekdl^ucKRQI7xd&V%AR!<M<z2qTxdAkvy&)A_;>J1}J~eS2h9*#|7Jkbwz(t+g#Yf
zyym5P8IMT=Q1xP7z&rFKfx`##2s8PSflGZuh4L417Y&4XW4%(064U&&4F_EShJt2J
z6}^9X@`-csI4-8m8)^a70C~_s@e|m~pXd8A98pF|u8ECz`-jVCY!(&uuA1_Jd-Gkt
z;l!jJUw<%gbW4B#a4s|eQX&?tlo5{bJDo9H;M*e1)#X?ER7`?|-zB;e+^-d;IX$0x
zUsHJ8!QSbLQHN)6AhzE!4mgAy0iSNr-vpgs-NxSIc6-WS?CIVwLl?R>HvGcc{D$9i
zm_{H^F5Yp$2DI(5B)~l#qEQUqk(>Nql&XjZ(NE5QZz<7Z!4*J}r?Q4lLi8I23e^ev
zQYjOjz&f%8=l3%Baor5t-^p5E06z_)#Rk85er3`U+V&e_f`IHWBtjz#P4yE2kV)8<
zZY@2GhZ@}8P75p;7_{y&?nH#kf(gMvjpRT%|M?_KfCh+GDkUAf<ZJ>L31s3zNPy^_
zAb@fRB$e{jZD=GDg1a66vxb3_3<j6?HjoGb_xKV<jf{5vYWr3ITgwR%V0q2y!i4-b
z{`QI?1n@;;8ZE)&de95DmKQsfPMWjT4~Svbdg;2wltmAy>xUI96=;I*STsq<Jmbz{
z`J>p?e}ejz_cp0Z!70V82R7IGa^W$G5Pn^0%c*<s8<(?BFn<z;!^9wm=w`5*L45fH
z$qmA6nCS?C=&W?*_R+d|zci71OYxy`BqOeTM0=sDNrkVMxteq{17Skr<5W8HN!38^
zap(i;Q-_B&5&X_+9f#ulE$a%%WKB`}yALh3I!q`GE|phWhmlB=Pd~>UH&FrA;sr8G
zStPo8I%6lGW5ny=hCP9%S2#%fcZIS3Ba7mRbj({G;?q?HJ#CGKy&b6!<E%n_wZwJ7
zMBqc7$<S<J`K2@mYfVmXvc*gAFb~!`g2)REo`hV%p5hl=WO16P6S$WAfI~G9fxd>I
z+8N1EgWgV?vswhf)1N<KJr}j0WKAb&!6P@0x*etWPb2JC+X#mN5k#n{e!g<c!31@q
zj&`ki4@N_`bIlBn3^?E_a`}04xZe%JmK2~=idq(Gkx)=)Ejx>7ksAyTAB?<XM@sE4
zQ~w-o4~IW~TORP=nXXt93?LZ|@L^BMPf~+I{^l>_OlTT%@$3V9r3*25RdBk5J_F*^
zF)-oP|I82T^^Rr!4eo#^*(ZMiTFA6wS+;;o|A{t*(DNzayA89W%4aqnvZ1Hg9|mp|
zvqeOcmWo)Bc}3$~5MhV852iOPvp4B(3;HAQ1SFm}@M6)Q`8OV9?uu^~pRx;8RLQ?l
zGKg-XukO>pQcCEGQgD1TS#Mcj_o|Wd$RMI7HyzFX@i3I9Vnly>_$pffqH)yB&#9s6
z>h7v)U2Ks2^CRcw!ZPSh9$H+1zoLa3kGruYMZzXT+kQ$d*Z`Tx5!!mgNx25AUC=a@
zn|g9jp*E$dtfY1b4X#bl+fDjP={RNiE60_9dw2#-Fb0;2Gk-&Z{2t_(aqLx?0ztAZ
zCiI|{c3kUGn1chv5$k?2Jz^U-xp|`9#9hkGdnN+s`XV(gB<bE3s*F<sKJ-_UBmJn&
z8vZUq(~qf(oVNABPiTz?qLW;ps9!fOW?M{0KZ`EWXVRWm_VQ*Q&_#z?Q|`TOte2G*
zDxVZJ07leN^6bc4Sms9Z66L537-IALXT^`F!{tjp7md0q5Dqk=ARCKE)9X}3&ar2<
zUPsy-H$-{WkZEV_lS(mb!7Drm8iM8uZNX*j>rX0e;$<vkt#zy()dI2FcBsi$*n2*c
zJKxS#Sd4OA1l4|Uiwk_Ws|lMqQc*W<csE%hhpr^o7WF(jKAl`kJ_`}dUM8a|+{7z&
zPbiBST95}_jYKbbHjxW%?&eJ{Ta^jXYp-{#oljx`E^K=zX`8#E1FV}+iEdVfP0oK#
ztT0s(q+bMf^{E7Q51Awyd1PP|u{2MmHY2<_B#6NX`)%R!1^Ppyt2c8G<8LEj3!rnG
z-X>gAh&%siaU1qBI&<Urm~-lAm$pg9eBngv8`<G?(B``vU)tTeC^0n~f(xq^=}kS3
zonh0rDjUX^8d>@hy%v4WXr*0YIbtMzo|c5cFte2L0}@2(G~17hv7*S)P*Ik%>nZ1n
zS4gqa4QsqUsL-z`g=!wp9C%59RBNb2mQmu(e{*NICqtP4O+JgLP$HrKv<P)h#s*rU
zAG*vrOjUUIZaGur81=9DIZ^df!?$by_rzIcmOU*CZ(Tu@%DUj(Yh-v=_g1<~HD-~#
zDTXDN<?0i8uvd|pU`i8E%2Cw5by&SeUkGg{6wwrvNa@viB&&f)O<q#ZM3+a9&=NgD
zC_0&tpq}CjC81emT5BYmzBL!|#g7vJJ>$S4fs{fCmmeaz#Yi4Mv&Ik7Wgc}j#9#mE
zIwqQ(&L!({-%;Os!NW*b9@9!cdWhpr>q$FEX}C7n=?kYnjDECm+*2O+4xdqtMWb@L
z$RbV1+7-~ki~s(^iBA981sJ?xMn{(Tax&u;(x5@i+(j_KKL2D;8<%rxO2Ppe_B&7g
zez)CC(vrG<XuO7h{qPnjUH{o0yuFsbZuN9KIRLrbxXf9YyCcI~*jYG)M_Z%jn(s!_
zqn%9z9+t4GT{CfWavQhK2i60Y@wA+?Z4<Rgdo$D=pC7$O2f%LbV)*%vC5n_&j&Z=l
zD%<$^y8&aNQ-RsM3non`@Z8x-FsZ8PNu|N*Vf4^x<y>BN0A2r{Q@Ucsaa}yrMMZGu
zi8oKco0<S{W>-YxRddvRUDTvHdF0<YYug2>bPP-?r7^uF5Fd;-8gh5DRvs^s8SJXZ
z4-SD~0QaEGlZRAjZ5$vUG>;?52qAI)$?fG&tG_SW-=kfaY|wCVI(PP{waWc`=$nv?
zK(mVRHDkofn;_77kWiZ1-Agh(YQL8@u2m@@@&QxbeYx2grQ^@%#ZbA^Zp=f3aHn4^
zeAFK5ru_thk)1=5Us4`$AoU-3vQrGCp1!wQ!FwujD-S<*3E?R`f~-*5D~9_v<fU`B
zoB44^(oa802SfDbD;rZd4|hlTph4Z}%Nh0<P`o>>`6*3eh?#rWEwnoK_ljthFYn)U
z0rewS1m4e%PJT}mgrRJ8?cnvQ%<duJ^<_VJ6|ra!^#Y$51Ip1)oKYa|0)*{7F;gm&
z+R_C(_iSZsv1;ZG^kP)p%A9e(A=eRK*h55hH6mD|PDZ`7?sw{`$O>rP%~5?m&HEHw
zJJs&rV>vokGrQ7K5P3YCR3x?0IMR(K36oV1#$8tDAJ4k+`@%bJn!#}y^OEX~|3zuq
z_iO$hS#Oi?dgH^Y?a@37sXSz@_1sfj!dhE{ZIU3otQ1Kcc7}a~)bmbo7^x6s(=m&d
z8`ZJg02f^MNtT}%+C<U@j)i;TCX<s>Amw!MFq7kNCN?64v)FaFcO%{-kG(JwD<hRv
z&Cu!CnlNVOIG`CjSP`j)D|-`>iZI`Vsl5d+Wravd`>S{%Py6H)d`6Y8A9fVgM$M^&
z1o)RmF$;NOO>$o)OErs)sSYqx`&wt6F|fnP8imuCKWkwvti>K;;iAc~Mc#505T-gK
znkFef;N9f$_M-4s9ZG}<$Xxob1qJ9#ZaT3f1Xe^wJP7Os5ewQJvj$sMU7#(a?Nxs(
z+>)iVS^l}j$5otOQ*6ZWZrJMPd~aMDT_Z+x{66VMi=_`ULcYZ3Ny%wNPlX^y`wF-&
za%4Ux%<QgY*+9vET1~4YRmC=-*n(z27BI<aEWpm~5+hQ&#)d~nvcMh6(m<TEEr#00
zHp_t=y3$Jhax#KuH`kK1HR(3Q{KcOuk9$|A4+qj7Gn2m0Uf}plb3rj_ViVmyUWOyj
z3PtOETkw88DKD#Xi@OEbz!Z8rOo&nl4C{WoEfir^RIiUVT7iSKJ-|39OHo+ILryiE
zrjDw-U`jANfXeJ2dT;g(JjpQ1UpqY&V}9X(ZR(u#YIAiR1XiS!3^T+PV=b9Zb)xzX
zT1tAvkdMIa+*lUB6uuM-^y3{UFJRHut(RM}&qz*N%krI2`2ByJy;G1V0lQ^cb;`DF
z+qQYi)+yVzZQHhO+qP|U>hIgPrzc`2W?~-lAsLx38Iifx{&w8<^j@J|;Orj$q_x+A
z1d?ta?ldkc1=T~{skDo5+1vhdrz$dyVv?o#LEJEM`^0I5cgY?Fs<gS*+dl;<^+}Q1
zH04zz-A2XUv|UZCs<d0QCV!VW2v_X_5axm-HUGgQgv$ED4&gWmUh_t%8r%=P7OT~5
zTTS${Oo2cOjbX9fj`S$LxpZ>zDEnPrOUp@FW)4f53*`;wpqf0h27eID6qyD?td4s*
zsKvBvEJZDw?p<ms=)**v@pj7cYLe<<EH;{G2bqoP=o@1rNo_Cc<hMDkRodrNc67Wy
zFQ4(Jt(D1%O;_T?i<KJ3@aVD96Sd4erUW9YH&wV%iE5>YI7IU#Zez#KcRapQ+jccg
zMHG}9M$XNnHo>lB&(wxXDcp_g-T~C-oK{^B6!4Y1R0d$3Zhda=cQC|t-7*7bhnhjv
zrRQ(<fmvk~HCvD&VY7vu%o_+?LCco-?rCrn<i_oj2kh9z#y1xNoL|Zxvvd~pwfyyt
zOme{veNp;E@t2X#6usg_b1d8?ti<!CyQX-}r*jfUB1O}Jhi;6NL>a6}T`S1p39n6G
zl(*?R{T}nx#8l@^=5dxq`6k^*<yb5!c|C!Lx12sjEd^!CYp<4fLRVyIzy0ah`B_@C
zCwR81<z`BW!&!`>TX4)mPvQw{*V?Npot6d46dgpgzFa#SUh_<`l+deRq-e>II{V_E
zrA8y`#7Y$51I)3y`I#!{{>!eu6*n}h{}vtS8kPsA_6d~*>Vp(irwTAc_=z5X+~}iV
z!{>d(v5l%B#x;u}XdDmHy5&1eKe6u>o1WB#;!5L}B+TsE4t(FMw+H0S#Okj%=<sU>
z46m6DGadb!E|_koN~@x0(`IzoK@-upOy9y&3TLGokeuHF5pk$^%w=@D=(N-&pVDM^
z`&`)JDv5*=LFMf2K^xWCrnQJx<ko!T0+mH&_v7p6UI+tIuL2s9=T2%C@UbNTEihTa
zXJEpM;7qTAmXl^V*END-KvAf17n+%rDbEN^z2UP!xC5?9=tp?Kvf5_NmCyt*`aQ*>
zsZ_Z6nO~5&JtOx=^l`98a5>C_<SaM3sI^n5Z9i4TlJD}FOvtjZm1>9W9%*hSNGAoS
z+|=P@1Nb}Q=WoPbK7u8QhJfeQ$YM?@y7D7u-i!C-0CCxN655|6mZ1V2VFB-0rnY&g
zx)85&tng&icuV#|1Cn`!o_8}KA?P^>H>Pi)6}iwOnD)lBBcg)XKDn8mb+WulXsb5T
z;t<qll-*>!kxan*mY@9FdLIYWXUd%3_$_(KNrUyp4;UL`%UOy(?2QFE$*SsAwCqge
zSm(toXk?ND(P+PcZk*fPUu8xk0w5)mLEjzQO_>M8+7{svhcM%ipEa)I(W9E8^O@^2
z@@Pk5$A(z1NPZQQ3U*d1evt99@f}<$KpSd8NAIB5`GH0sJ0vJix8Kf78lAtcS0Y^x
zI$<I9IiV9Opa-H!lHojXZm%)cEizS&xpX>T5&90gI#l$nGN`Ux41kqq3YOkVXlJ1-
z{pX{lp>vf>FC)Zdw+`yXO#+IgqjM&Zd@{HKM!NZO?sYrO^%&+4BoP$Z?8NC=b-Rhc
zuw4*Xlaw%WN8ii~!0>i_wTm11jPT-s#_&1|2~2~nY1{R0`#AQVBBXeP3AozxG8F@5
z&XT~|i|byJ3^i`CPsa(*s}T?t|Ck=1>8~FpcyHhj1@tfBI;Wk3yy2WTU`fyNMiLXI
z<_H7+*-KU#lg@P99zodbI=*xGnc09WQ7P-$Ee-X?j-D-@oLY0p)D+RaxMi=Npk(M$
zfH-&3u~8GIA0(Ee)I1taIbsMY-bqmO$RR@3bxU?Q>Yx8BvscA(YxI0kBuCiMPt)bZ
z_%W+nmBzGNjpsT7Z^hTv_!Nf1NA!H%-}*Pg&%Fx6@bx_I(2h1^jPQvmz=)w7e9|UQ
znm)cMxcY@Jk!R>!L2=gNLb9u~!-Nn)kw;|(UAvyOHD#REep;gfF?3U@B2H#@$On2M
zEo%J0MlZH~D6;cd`Qjon7)A~XckC8xJ$RaGEw%?v8bM(CsN_bhN9OI?j`uNgcQUhs
z2aKN21ZcjPjI4+W+}wqs^4yU#yf0xkU33ep6>$re?fHs0$(KE<t#M~9W0hGUvTb%&
zeR?$8mD!k0sG{H$AX+8D_(sQOEpzJ5tCJC=&G_KD5t6Qc8L*pga6(+rT8Xwl-yx^f
zE@x9+g%!7JVmmyR6a6M*8X=r|zD$R!qS8l%X6#~=z28aW#X;;-+3&WUv|_V$G54*$
zAH^E*Ucz8L^|E|EuCq{RONAH_q4+o_?Ww%Nd|jPrE6*i09Tom$u@<Zlcfr=Gv89}D
zN~gr!ZFSjTz_PwC;x3!<NI#+#y%*hRDJw8df1`jz4kGlHwLxmWO2p@2OypayCzIJf
zNOcd>I&TEPNO*>>k60)j-{DjNeUzv*TGh>Cah1l6u`r%?{f(jJTip{@h%1*g-#6fu
zrb$bG5^GdpNYXUdS}<ko@Y6}<z*k+m&d~avMvot%P6iKm?`hD;<Eg)0U$_?auMkL%
zm56;=9Jvb6IOZuSi4tnz!?<^=zsMK;+-ZrxNj3WqzxB4DjGU<8j_$SLzRo05TrZH5
zgiCo2AB!5rS~=@?Y$X|@pq-2XWZv2+zy?o6kN0!agySKFHK`le*|yhW1w<WQRT`_J
z<Fky3oW2CrjXDtFbG9{YLD~LSqLm+I2N*-8O^|)aA5TnCToDB$rMkJK;&P6%B-{;F
zloe%jn{ltvJ3&OlaSy4p%Sf~s?Hks0JMTS?`6Na~tt{G;zgzjA!nM3?zh5pcj<GY!
z#d+)dx+AL0t+svH$36jLs)TJj<P6x2{Yg(xs2ra5Z+rVZ1qC9Xwj<Qn>0&_En9X7H
zUzU`#nh+4xVG2%gQJ!O;%}U9ta5TO;LaaPm>1*QV?itTl?tTJNLdur`WIH>wlTP+u
zN|nA&lXihV<V-nb6|IDO-3)hGCQaKnm(?FCGOL6&pB-n5m>^v>gLK(lR_f3RR-fco
z_n@wW#My*ixgQAH%98C0uS6}=Y9!&ID_Q)ob{M*WQ$w<2f7)}V#i5w7I%tC{NqC?Q
z99?3C+nVQ0wI>+Wrl@3{N0;_6b|u&Q74Z;UwRBOm-}GThnADvi&%kGtn<|LnZw$y{
z2G9*NZFi26>A4BCXSJi-t*_GLIu_Q~$Gly5KEV#kID}9mGW2Z1X8DWT^uS8v${B2*
zw+UCdGBeI_m3=}5>ZV5kmEpod{G|6!Za-ifDa-K|WtSjRj7DNybw|Xc(2u!<OUlom
ze_FY^ckMPfmF-6K67Du#y#0om+6~$Yu8y`XiV2t{?6tkfo-pzDlHl6G&WT7s@G->`
z=!zEz<_^5w$4na^%|q|MA_jraz7Vjm97gskA1dpq&fo780M$2f7=ESmpb;E6q)KkV
zZixJf^5vpEJZCN(I4&%9z8r*2Gol_*pAGXbitPj+#g~PO&>11r){XLRKqro^H=1^W
z1!kFUHLA6F{~U|Nsh#ylANjtpBSiJ6(LUqiN}v238;Sz0xekD;Q*N$=hd6s5?U%Zn
zXh08#K6wGzbd_CmwfcQzeOc0=GQuJ=At?`+P7V*Mrn|j3aIV8aKe~hdo<P!>q#FIt
zwMF+fG!_wnHrj~<w&8}4j3_Q)Y(RvKt(k{wNik(|z8`wkb43FGpSy|-CyaUS-KsuP
z*;di3f$IANwX2{*+$VT0NSR7XDeX*;z)9!6-Afj@y7HF}&mqyl-cz|X9bG{f!Y5K7
z)=lK?vXS&yQ)if~NmNHJcv*&4=a@F*HH*qO9C-4k?MtQ_df1XV628`3KywD%KXotp
zizkOj%iU)_28K|CJ6o@Y;&_Q<x6dY4H+nT`el`Yc|H30&J_6GIJn+!%z+GKC(V4^w
z2l$&8>Ne+}A|~umlmjMbT0x*BM*8eZ76y84i?s%iR;uH?ha3%=EyA}glPF9GWD*5J
z?7)wvdp*&)9iDPK2}Pl>ZhCRJm(O6JJg~#DZx0u(+I3b{OblJ_M0!mcRFAbx%~iNJ
z`IPQ`GS&v@oJRIn<{m^#m;{^{t?inY>T5leCrqK%KaD4+9-0k0gWrxvIi=^ez(Mt}
zX5=l&I@&Bn6>R3f%ESabXt$6i@;9(@THP`yOV^Uiu0*pvSo&@@`YCpmO~$TjPMo*?
zqp9+}J^QYhM>#G%ElkD5RNbhFelG<SWP<eqsFH(w^g0r35358y@6_1^!<Mnt{@yNf
zAoSvMqU9Wv2Gk{w>gI{P*}BNDQg+*D4k3<)u=DzixE`Le#d4~b8*UjJ(s=7wAT3fH
zR7flWSPblaLFUn$p`$^4(8g)Z+kyROU9r-p%uCp$#*9(lb<*tu+*B)%PFt48TV66*
zQ{!Ef3{<jm<0>rC<qWjrYV{F;GL{Gbv=u)*0{X31MVNS9n=`<{XO1q`TnC;FCD9<!
zsyN}5PKj2=m?9K9QJ>+MkpjjAfiS;&tJNk!=X^T9iLpaVa>vNt<d`yT3&s)xElu*|
z?D1D>lsl)JB4%SZMK;YObhExSGWp=|Oif*ngX7ftXCao0e4@yY8^~3+?t2=sM7>rg
zzd*2=a7d@`7YL5p+*kNqEO_@PEkWmnmY9*9iW{J$+ul&c)h8n-PmIf>Aytz=MH2`A
z!-L8tPuzo4-3{SFB`|BDqxA}1MK&1YZVAgXEB^Kl4c&zPMl|%JiF5IDx;Yz@7-8{Q
zh6Y8GL>uH|`7t>gNmtvG(!XwbJQkd87qVNnL7C6Zcy-ThGI#5KNvXI(*XdNeDF?&?
zy<u-YboDZ^Ki157m9Y~k$=-v55<Vkzg6BmK4iue3xlm~%?mO8%=*t9j<h?Hg@#6Rc
z1Z5(sg|`Y-+m|!swEL$-{RecC#o2Ohx`H55o!YQN&C}f^rE*iFjgj9<tDH-sa|*u!
zfxUU_&IA{rlW_XRm=2x118c3%-PC$oE}O@%Kd1-xdRRE{CtQgE0sv~0xAC0-LEz9e
z%idg?j(tx`*cVDsBs`zv?7Cvp^8M2~3<bBgENAwkeXshglB9vIkuUtJLmow}e)6a@
z&1Fni9)*)&?BiXl-yQp=BcL)W4bu$(0cYe-&=5M>(>Y5<E=h!dZskxUMN~5!^|D+<
z0qMDdmoh|a*zQqf?Qkr91bn)u>BDD8j&P{0cgcxeb><$-G)&4*`Hykj{FD`C8hJdq
zYWRY(Y&HNX+;5%Nna+Z?)#LUolegBCdjPM#?#f?U-*@y1CJ4|x)*Ur((enPyxuB-;
zyh5oFx$Z**B7Y|>h3gVt<9C+Sr)69Hm=wG$`@7+!=#P8=bn{CQ{=>Zguigj4f82@x
zX2$;|e*8zR5H@o7XK!ZfXk-6hs)ek+)vpqwEGVolEJY<GBO}4_yWLXX<W~dvuLA`C
zhXO+L>srt-ve5qu7i_;$0So=F0V3e=?>`0zE64AVLi)DiMrJ0azpLy_kiUw=f9<1T
z_|HxWM}14Pe*)GfmPYtA48I!`9gVD1e#HVvI<^0K7BdsW??p`Y?f+wNkkjeYInp`P
zxlsJqE&OupUq9r(JeB{=xcH9|^4|~mzo0l13;TbwZ^mCb&-{N2#aY?e8U8;*aa9nd
z1+ShG1bKJ}Korr)=sB4Ac^HL$bRg2&C<eWPBF*+fkT_Y2I0XqaKr`rASP+OXt|zyh
z6W*JylTDBBlgpnEFD~Ak9}b{;Omrx64*bi!b8vPIel>VvAo)3#cmQAie1H5{a1MCW
zKSCXXKjLG4#`1|^(10m#Hts0M$dSWiN`0o8X|#V}XYU<Bz$5+yk{AY(81du#f`b!(
z(Fp-dg62X#`*Q#|f&glOvo8i<m-_P?{G}%|VWyqvi3YF+hvQe%(1`zO;o={I3HKN1
zwe_V%zk+reFbee51{?+lhYlb9CIW8BWWtO}L_oZMentiEa04flmxbj11h7G0$It_Q
z^3Cteh4t0F`t#UpJ-ZVH5f21NI!zzLXYenOXdNrk|Bvk-Rdis;mtGFnew<u#09v+Q
zHAS%8n85xY^ovi1U9e9UOyAnv+n#~1qc19W=<f{-Xg~bTSUq-BP?taK{aSedUQRdy
z%%D?n00TJQ82;Kk@a#UdXz0g~%_A6}wAj${s*J$Cc3)phSzy2*Lis$vJbKkVoXGDe
zplpULtpSkzQ!tQX9<e{=awx(6g0_kykY73#P}qAgFJBJMfqgjGK25v3VaSX@1DiZ~
z>E%A^yTXS)6PdZDzU<(@@rm)E{9AzX(1Yrys<uefi0_&dpHh((J3DA${C}|RX!yVM
zqvbavHdlZyK|pf<X3P8R{8Yc5L}~p3ae#naz-M_f^vutjQ+9^MyBVAoH)Lp_+CO`~
zWY7R#?%!VZefkrCfG$JdR^J|Nf|%o!qAyXe-fW*XN%f3>fNoLX0Re%92L%B1_09hJ
z5?}BB%#efXt?DZKem^e-auEW|`-Bh4oausid^LGp_k{KVy_!(-0K?YF1H9}2w19>M
z;J4!K{jg2?bbQ*Jexn@y0KEDv7hIp6ziUo>NqzA(>anF#J=5{P)?}MvKRY3?gZBO`
zEMmWhRJY56rIu{?n6u33VS^>^%2xm25>5)?1p9HK_s+tf@X5asv3-fqtD<9@``-F~
zz2^b|2MXl=%<idy*!6iabKAOpR{8IXKg=YW+k*G+p>+rkk^tx-Lq?K}RH3^dfxSW=
zn4SU*eTvfai^E6rA7}xzK7Rs6gCoDyc!f&<272a_57JS*@evAf|0Tz+@wUNDdDeZ|
zAr2N~kOzO8ME*mF^<GQcc-1O;mfX@iUi%4bJH#Cz*6RZLH2#nn&zw}k$>j?1Bt+5F
z^YQ*^!g4*((LjCDok0Urh)^{aOy~2SwML2=Rn<xSLod0*q|jKV^z#h6W(AngQQ>`Z
z?QI;R1+IC^)#&?novn}v8pJ%ZZNp6eWHWKR9VAzUUacZ!s+gG+1vQv9tiea=21i{x
zlV+$#BmlM9*6;BTPUz|Tc4RQBGfoP%N|@LFr;2^fKzG6@?lnOoCkPS=A?CN{SyU>R
zD=>s@JId~zvZlyvEA!b)r!&w!g;(Oo{Iozby!HaKWhOj(j=}tE14wo^6q=1F0G^=A
zSo%`oqKmi%ahVvgF0;$6iX`%wZ4A`|&ijjawWzzo5a;jEAk(t4$NC1>BOGgr$_QpK
z2<sv;xro$kJ!WB@Jym`62V>1xbi`AXVhx%%^YN0ZC(TweUp@1K9miMhvFNrP^~9=`
zUj+DmP7(3QHb5m(0CIM82#9;RI!#djmktb~QXxIMz*>i3_wY@3Cj;76yx>u`Jd&)G
z%ATlInK=>1?0WtUVXl6M-q)CGY%yp>oL6arhi!H|N_OrDvD5jZq10gPpsw+k)7+DC
zk0Q#iI{Mie>>|y#^Fr3HgCpgQR8rr9Z?J-aTysFd*~s7Y8L#7%`U*+6R48dv#{N)I
zH;EGS@{g0CIFc6H`enSPykjPf9ahc066aDm`on0thBcydpmkd1Ow3jG-+DqO^K#(l
zt%PAS`_rzU2e!tSd2mI?YC}x~fkDd74pkyyotAE@2$LW!UOy`E$uw!Id(IXaW+H6p
zH#fH-T>h=rR9%WPUBW)Ad<*_<MzJyrD5?GtU4PbYWpE7j0--3Ay~X&W{tVH8Po)a4
zT<ZwlHwv|xODbxMxdlaw+(BZ4&6-^|2eV^A9yO3;ePRk4I(r)0Wi}=SGXH(-grtnC
z-=#`V3zAQcvyuEbAW6TKq=0I<Kr^T^WT;A;oRn)x90x&v<a$u_@>zhdyy;M1@Z6SN
zG6vs@mj`!}n4lFAZy(D#$Yisykuos7cQwTx!oZjG7}!{Tl{w7_l2M1IAA?IF2RIi^
z&zv-;o|V?z<X!2-+v2?&w7NPFWF2$kx<IeX!R|$WO{GigoKnsi$Wp2oZia(cJ=n#0
zmy~5<D$H({N<!9oE4K2PK6NE1A3Se=cnY<nEk}zU&X@&HY0V6zBoAg8Ld|(#Zz&7=
zAXpZE$2w^>THt8ISqBp&Yh+4vjxqS;NKf#z1San?{f4x9wJ(}B9QtHJ3xQ6HbZtZt
z?WP3mi8cv%Y@yn}iK6kIdskM}`uDxj57^Q+-(2GN&BYZ$Fj{uKh`CGp(HiPHS7Xxw
z!xzH3#8NTU7^g(ywC~!B1FsvWw5}z0qBUDbqh+w*OG93)lEHKs%!!xAFMYI?X$0JD
zqcMx%;uzQZj6#F*_ujdjL|k8wE@XNhI~&j&ke0Q6FylYBxJJXTOxN{V(_q8s1r4Cg
z_W2}K)zKC$4C12Uja<h{<%izBe$1zvfjlSyLe{e3B6h`Vmfl%YrOxQ>N^t37?N+m~
ztQp&-#rD?IIi(X928Xw_&DG2H_(}U0%+sm~bZ4D-P7pSj%u>LzlN8aaaz|#+YGFZx
z-6}YHX&bmc(($J~2!v=a{w7eNozn6k?FCxH@H>Jx4i53e%`dEiq4{Tshq#?KOK@?@
zE5y{4ri=<>pXFYZ{JXBNwO6)eUfU+Ak4B1ptK<*d_#O@ylWz`{U^|R9j0I)<&MXs%
zNlzIL*BjdR<tnh;E;Fcp!l$_d%(%}dOcH<Fna(8d$@bl`XD|X`7PKkLGAC4bHg7w$
zabCkm>f!3S<k+@QiEeUYISJ?9e0Q^45YQGMx`^(&&Mm}MB6kJIC^En<#uD!w4nsKU
zexr(BXKH1OzIeTro1wLT-8D9>T3*T(x~AevpVGhGPta`~<Tk2j5)ijk9-}1gqYm$k
zyI5Oj#xq2a{j3g?s%+R?^Me|r6g2*P<=%5S2n`wqg8Y4cZO%)H!ar!5^<%zUKBfP=
z2Zv+*m8u-nSdoAM^UhJS=*W^a`J!IM3pV39N*3P2b@oDN|9O1!Gk41DCml%>oK5T-
zl0KMVy>W3Kj~Qa-X58U43FWXhwW?KYXC;569F~HuI7Cvx0FeYr8*UR)OIMc?XP{rj
zY(L0{iuwYYuz94+%nsnewvc|qIqA5B$Ohthql(Cy2&sa2s~KL8*8EUb=@klQDlVSk
z;cRT?)g@nx)}-mq!nZ3`BPK?=0re4;!792sIVmlxltX#i-XNQNM?51M6ccz|{#oLf
z4#irH>h;ys7=gR_Ip!(ys=!e#hyA)TDnr*J{m!aq{Zep!_+?S=y~pWR-4$s0T?TS8
zzp^8!<&HaJS<YT=#D~bI+!BTx)mPk?^<~}T&}yHMXs8t0LFD_eZ&4y3Q&;~8l_VuC
zU&;J7why}Y4}V~U$oq_2!Sfv9i)vjHX@e6Q<3U)b#q8#8WS%l|Bm`%0`#8l^oIvkD
z@yhHw+I~5%fk1gbL2c=L*`46a$*7u0KZI8P@H7v*BDO;OZQ=2n4hv^rny0KOOE$&$
zRiev~P1SYG6s@6%*`Y}ce;I5|^PSp+gJcg0;`^J_3`@)riqEXj!MY;ap$1M6oV3sd
zijg~aAyS`XdqWgKAs(vIX@DhEIHnL4)WdOn3E)Vuix4RJpt{TBCF?y{d@bq1Z@_Uf
zOo41qq^wrXje>?k6U;<*$YF(9K>?9Oftaous$>GAPG-oCqn`>yxLzigTc)ex7sq;h
z;)admsNbFxvu2mB=Xj`!fJ6g_1^UrWIUkq!H6<&nBj3BEQqnSi^t(f8$tq7byA=J-
z3#CMVXW9rULsLVrn)Rf6dYcZKJYn_@8nIaK?bG@ph<c;Y@uXj!HkDTNW?a$Vpc+a(
z3SfE2bdx=cySKm4ucDcvx*Lvfv8|Z=BAK=+#yW|5(NFToL449t5#kQvlK0SIV_vYD
zWX_>e4${QBxvVuCR*kE4{|oy2k3NpLO%uOCLKNDPcIb;1;!FO$;qR-!EBt6XW9YR<
zW^>w(<F|3^>Kvgs$S3^U*;=t!Pol!i`b6eqO=3<6tBw|`8hj{(rHth{bERjO2A|TG
zaSVWHT}N$!waNw5Wgq;u0#S39M%t7GTY16Hy5j|xn@za0SkW;54-Q;UFC$x3$riue
zaL{>S+?ENfqz<!vBwR|xdcZ{}$Yhv(i&1jv>(N;v4J#$>CggZBQ**8*xgE0+iL$c3
z{Lk~*Mm%T!h9>0a#5NPUjAk!E<2F(GZ>M$#hIBdtW=(D0Dm><r=q0rW1bnT*__A$h
z0wZqtjFoDyM?KwRn@q#>iiaF@avq%Vhua&e^-4OaO&iE5mF=Xet+^^~)9qdLr3Ki8
zf3H_Dud%;oL<_y!&Qu$E@=IpoEm8^xco*lEm>)3n*zc&_2?_vEH4V+1__?R}XZ)Sh
za8$z-;ph3)Wso9RfjsN=Hcfxp4~{kjx7GHcgx=uh%X^0CxzDQc+^t)nd`%wj8Db5_
zFEtfLq<35k?&2kBn|hCf(|vEKOCPi?Y0I&+fuvQMpQCV7yHgJj!YRa(jpDX_Rl!_=
z-_GBYBFUQayPm_3on1bvx8a8*9nfLdBCz@sONxW_IvpNdPi0vise7I{_?*R;bJz?2
zDO#q!_sbD*i<A$MLix|;&lP21Pqyr5Xa)&aG%B`!rS{l&IdXFw`7$`iq6>W`tKj>?
zR2%1qICV9DUcF$Zn@1~n5jAyCX^@gcPY8<{VM<(|7Mu83S=U0(!tcx+0U{m5PKSnx
z$Uy^h-m??Z_va?-{!zSkKOJ}w1b2t$lYErinhbL7Xi56DBNiEVwi>d69oZ+g^|kDW
zgoVIS+Y=yVR!uY)DJOjy7nhJqIEPAe3E0sSxP$!JvV#f7dkeeb_77VsBh7Z+IA+a$
z_VEte8EBC#J8w>9hu=;Yw7MIHDc8vR?C-Km>!>nB7gBo3m1{j*cr1<YL!kM1p6$r}
zr!Edep0&U~=vzb$`%S&YE&eKcT}k-0Ja=z2Kx6aXmr5SU^AT%BgNDyrjbR9Qw^hX3
z(`zWn?QYAZ>*}B*2cS-rz6rzoj5^T*CXu1}46(yCbGT{`Z~NR`wo)-jTSNuDxy5{X
zc|}bXBA)7rh8Q_o&E936a=s{Z02{2+p_!g&JUV_q9DL07ZsqIxADB%sv!Zin@uqY4
zLhXZu+p^bO8n2*o+xE7~Xv&mc3>?;$?JR1+`WBuZdN~BB_{I)7_VrFIWnZVV!H!9d
zVVe~%Za{L4Y~VS;Pe&39nr83D=$Vv(^g*GDm~MjXZ7$=dKGMF&6P@;@7taq5hA*Bg
zA~Oua-*i@f{#U}!Zt@SGIOp|(2&D|hQYd_C;csSdG=nO~>tZ`?DOndX7Hi+C;+&m(
zpU0pcZ;i^la^Uj@4Mruek%V*l#zHSId=qeg)nYt&EtF+RygI(N5;cb^LS{AZwwc$2
z9*qF9&2<ofK)bv+5$4@7tBhj}mqPOTsdC`ME1+oL%^;7<Fb_1XO~(s=v*=i*S0jO4
zNK&Spltt39vz;eO0rDC9;pCqHH4VC5Ps8g?_gS*>Oo4h2lRBaG(i=NQ1JoA=f0HJe
zauoctY^<{cFa;#F>K1r_D@uC$t^!e9SfbZrOux@IwxuNX-^^2o=+8AQk;qSi1_&63
zHzoMV`nuWS+#VAUEO<_Y`mYR?2<?n8+Bdj1)$?#(_5}})ncU12<AUOsVhA3eXee1u
zW>R~|q_+!R7lZhR3tDB3!xp31KHjR+y3g@4PuQ(5ajc8<3r+2kvk0)32I;`@y_hpq
zkf{eE<$11|yp+dsRlM`JBSl92jp_WaFDYoJqh{JLA}Q(XVCr9rpEH-W3ZmE$Rt%1e
zIAOKg+zdP)ti!I$i93)_7L$7$;uAbCiC)jJ&d4Ib$i@!#Wmqutnn)uOZa##j{z$-C
z$KoZG*gT{iEMaOgoY}O7&68lZatK#FD5g9Vp@T&+I%cI@W>}&Gc3}U5tHBzxlb?n0
zQU@peAh(C;?&tU}NaMYk1A8+_dU@k>w%|R7(M>2$u{5Ns_&}zgEx>YcP35ss*eF}g
zy%c(}S~s;UCFz-sZJsXabX-3MjS`bi8#~_a3+&Sln3a~Sr-$*|7V(X1v=*BUmpdAN
z$wk5aTOL@6v(=1OIP^m4Ai>bmPB+ymo)8YwM#gO}Ym@@^m24^TD05HhAaP?8Ew#a5
z0Ux5nb3jj@L*cJ+8GJbtHO5sAxPx}a#4t>RqZEO*3G4YDDpj(*>S>t=RpY^+EPc6v
zk(bpk7C2YP<UsI3T9w4XK5IoX<${%gip5RUKyP6}$Nsjsp1G9oq0(ydaGBcDV*6h9
zXmT!2_y`fe&NXiJjFhs@oQT-Rkg2Q|-%dY$LIi>quNugQU12hcb5Kp3<;^rYC71Ig
z{Sz0;Hq@Re?8k{6PdC|~zqs!(VYzs^OtFPEHJO5icm67^#fiXHRKa*`SepDkRV7E*
zg$P`Oq!*Alc&4D@(Vo@4$FWY%o4DI{*~d*WFkN@fgGv`^x1SS>FcvlzzWF!Cq5#zU
zDO0IoqilURLtbY9*d0%MPp9-6Zf4&y*E9CXi=3S-l~wd+?H9~yTxnU)g(~FSv!L<H
z`d@c^X%YRy*u&i-*sI!WuCFRgLy>8myAIiGE;^awVBE0sXJX{$>U*42`DZ8{F8qvk
z+T!o-F(dxTxq~&?`t<7J0x9CILf0TgS(ldFb8H69QkUI4Xt~`HEci+!e0+Tp)rA2J
zmF0TX<6cO4ZjkP-x5U9aH8R+vuEX|}p62K06MGxZJJgJBOW_}odW^fWwxNN7Xr6}!
zgBHaQCan5z2C{?n>lLc&P}BKqWuQS_iEyI6q0Si*H4(IPssbzqBH?dDmO>%FhHfA_
zv50Gv@{X%+TilXB*kkyhaZwb<Y#CC$LHrMm2SZKWqTi{40GmQ##gX6kKrhZXEmhZ8
zN7-9vGM)FPdXkZgbM{K|=Qu5)W>x*H8v190%3*rTc+|+!U6>^jHIE96+LC=0ulF;-
zNAETH#H=`zjs!rkJB83kgr&9grzxcKEET!3hG2qNLR4e`%iUDr>_Y;Y!c|Bpv2}o8
z<(#N0kW*;cog$YcB}X!nhKla+DRz`ef%YtG_in7Rb<(c8Tv58U3*qqJ6&9q>5@pOG
zyYm5xw-Ix7;*SKkNWu|HgWkRy7(WyiM?z0n@`0>nyLUkig0<?w&%m`oB%h=mAHf^Y
zcWrw?!5!zzVThl_OQP1Qo1JjFZ@Z8O4f+af$raW%;~ixq%i}TCa{OX%gLX3>%4d@<
zh`T3)L<&fqPtW#<kA5s&uDDV1jhUr8&7@qORDszcdGm8?OIyt$Z(+d(MZ1n7+rxip
zxwXE^=q~XVmX9WGXfKquPD9yxCG<-vmv9kUjP=}#5v&H(9-kSAY&;CHHJnV6O}nva
zV3kE+oQbYZe<dv6vLD;1<#^*>Ou$CbVoD>!gEg9ffP;#r?@sW{voq6Aeds$yt<miz
zM&dL4vrcGL3J$J5%Vz)fH6hm_6XLo>t-WdJm8#3X`{Vqq=b=e19?Eu(1>VdbQT^ro
zdTx8z!rmBeb#0s4n}t}S;jNFC`V=u~7bz~~{A--<r-#`VLq*>g#dG5!&B%f@fNiBJ
ztkj0blEs%mSF=fxqqi)obl7Tp__#%Wwwn|+kA))%TANX!Xd6Gvn1ovw^HZr?V2qC(
zQSeCjD4<UHymM*tGYz}OH`!CicOO^;S1Smuqb?IJf(tU(YJ~Gj+`T$~T>Vcz0QdKA
zvPq<BCys3A=X}$;1FkkrkSOs&<{6Pvdw}!It3dE6FgmVS<&P45242Pj))!g2@}Xc8
zf_<8kIjB)5^@3Dp1@Cl)Tv>X((ncd8+9@!SH*_FBdYY04%?n>#NVAqke>I%VA}$R{
zkJI#fnhnW_<vdW;15n`faM~?&0}r9fy6e+kK`lqka6MjnzvV{y8tKjNlxc5|y8Q|4
zt&NG@(Tfn`$&qZTn@663^1=dilZT$lnw%t`4z&c#jai7@a1ZvEf7P}p`5~uaf(Y}k
z+_&u*9`-3Ek^NL+9SD-Fnn&2c+(Ap4H9T1cz*L2V-;TgWHNUZRwfmFXt&(t0edB$G
zgh`@|7G5P~)L5#KV2MW9&s9;i&TSVs&xc$E?jb}7Q^C6#DzYulTZv%#n<`WUD-$8n
zJCf6m!g=vW1qe#J`axY8>Yg33GJC7I`A>%PEwtRN#V(9OqYDIu?j1HbgjbE)g%>I=
z)iwRD651z?PKm4gmX$J_uuA`|Zj1MB=h?sWQs)rF4=rS!Fk6sl3aXvOi<bXx(wL5x
zcAiQ*N4*F*`XrUCg{4*33PWKbtTjZ>y6Ve;gJ`nOz~(~LXHXC-l|!hA+|zfwrWI;(
zT|YAc=uSMX<p|7TFKEhyP_P$@Q?h$uJF;&KDjHu4?do|8i)RD0o2`~RZbsw@dAr5$
z$XQ*IPm7DgYZXd9-h@j-gS|*|+~2S7k4KbTg_n>Hrj0u{M)9g(rbqrIlmsO@bU$Ws
z&jESSR>9$-V}9^BVu*mn)VDB2?NK<exs%{=y!vmpw$BvQSLg7ZE-K7y4S3f>95%+~
zhBv<(BYg+DJ&#|zJxLPoKS&jBm$1C1>u+C)r+r{%{`F<KXs2pKqYWF_87-7C{!~6X
z_mRGy+81Nr=7HSLOk?%H%}bZh%#<x%!E5kG{cWL=W9MeASP7yo`>ll4GKY2Yf|ybT
zKmS!0Uo;i0;A#zF;JUfs+1ABBXi2Zq4C}Z=$aoNZ5UFDcdR1yuEX7E?6S$IK)iY?^
z;A)ydF8+jY4H<o+jaBl`{ojDY|HP~Ro2&i5fkQDtAq6#sUvQ|v`v1TU>HjA;WZ?LX
zDF6TGhQEo{|DuNf!HfS7)X?3?-sb;;4F5-F{C`V;{|0gY&qw=T$dHBpe?dY<e0DY#
zmj4Eb|9g&>gN>Et|1&ad1W`)7LS?1i+9Cn^&Ct3?xj_Gq5G^zeVG;AZWI$kGqPUAV
zo4B}C%b)q%yc>^ckLkCcrf!=ljhRgkU5^`&ktQ3__7nHadS`j8SVZJMCx=J-2tcD2
z=iz@`U0>UN7kUhgr#@S>SzQfjQ6czAAn2gbFECm-Ai%vCP7wUhIcX$Fz*4t%0M|Fb
zPC<jtegkf90A5_&IUfu`eFFfHT(ni#+%XXO06~7-)W~~zHeK}9K7?=A{{TZ&O%;He
z(o*^nJQv?eDsa$d5J7-+{#K&Vr#n&2Tqu4xb12Y0uOBMlh(IFjYtrGVtE;Oi5XYxo
z|9<R3H4cDny;dTAv_n9DH@{1Oo(T|BKy3jZ$b)`E;8QMto^OQ<fvdf3esu8LQd__c
ze|+2+*lc(%c<5VKHhwu}pxi4^|4*aJyA(jM-J4|q*EZK5oNJq#9lzdKzG;3<&C&`i
z_<g8gc7V%1T72K;cu=%WuxUU5!AozfKib^6G2i@j_^TkM52m+dcKG}fDqsMfl{Yu%
z+?M)!Y^`7^V5>(iL7P1T+^p&d24rwfP9OrixT!r4iU0%r%`w_9N8d~;Cg6)$`#&pd
zxX^VCKid9TC0H$hz?at$3d*k=`+tMpVx|Fs!NBitZ^0w(0r{^1Vq>a^UT${ZyRCEi
z@NaCM{(bv>Tg$Wge*J^U<*(&Ay@OlW0XhZw)2pAI%m4EO{H@u4cm%TQXSD}_6$^y!
z|0ee+#xnX%_hxrbUkle8c@pQ^36Q<h<C8;;+hzn7)cX4U@7ttx^S-`(T|@HH2lAVZ
zm<;R*@YUfy;t%^9up2-p_opDh-iXkykC+=Q@J|-V3}1ISyaze|V~rtR$74-k*LODW
z#7;yX<m-)^#;n+|KLGU){AQ+D?8x)oz{ijD10U^=&*eAH(U0Km&y?8M!q5@i_&wJ5
zPcFY|ZN$yaE0HeGUd-l7Ag?vRp^r=yzfV;K=_u^w*?UhZCH8A8P(3_YiVqvYWghS|
z;2ICufECV9{<yAtzz)>$pJ3l?A)ekIJY)beI=Go1y%<fSi_ce6SMK{eIAC|J^q+18
z1Ouq1k1lObe?b6zXwb(YPwcw_V({A&FD)Ui*cm)fpw>2g`PTxd-MUVGUmoP(H;-&W
zPXOz!o$?-HK!BF-9+GXbX1joIA!K0NfFB`b5ZjJ#qWgP*R)QY_-gm(t8c6PS-fr(7
zK)&sZXT1G?rZGO1A^@A-_^Z0tKg!irWYEv8mp-Sjs}Vk9K02uIqhN*+96u=LRJg}V
zNXz!>BvMCHZ5)&wL#<|y$w>=7xXb5mQ4z`g8fh5ye?CYu7-xfeUbW+nw2b|pr+s?r
zP_LX6<7FRYdaIQEuLq5pjyKuQQtJG@*5taF(HAq{88rc2t^tGlrp2ScugoLcuCobS
zXCA_6yF3!Tc%PEJvAmz06iq4=#BnZ~9pjmn@rSuvUDFw6NG*ahaZ#ov0;WY>_XF6x
zzAeJ(O)mXZX#8-PZt;_QW_KKu=&sF&ZOXKAW(orKF@h*KVCFgYZh;eeHf75(*D8NP
z`iJNoc%u-58addp%_1^iwYWSnlJvGr#l2jbR&G79ZbYpi@gvjDq2@#toFsVaP$G3Q
z3hUP>)4_G7ZlLqm%o<WO&EJex&2AD!yYM;R&H(9&q*Iil5^&CV>J6!v{ibqK6myTg
zxD{+uZ&j4D7R|1Pv-?kzFJ+VNeJ9qWyiHd}%_V9%n^*TT;?CUC=jg6B&>~PB&mVce
zw!9qTkmzTAQ%e2lW<_xzd0$jb)Qm5qWD4RSe23IKa;pc7QnyVctHhSX3Kw3&PefCK
z$@|uhI%|KLtKk}Xe#CDc3#i!b;;_3hO+{OK`Qwd`o0DqIf=q&foL`A@#En;guWYkl
z(YE*xqDI*J$g55whSUwO!4YKOtGMwxeEL=K>^n^?o4^2r*dl5&)f>($kgMU|L2KWR
z$O_^MpcOR3O8=sChE&TsC$#Q3mPiYeB=?|+-b*hZc(6a1>tMq0_Ie}K9YOuD6k3b%
zwVKJ9q_d#XPBp2ntf;y&dsg4PbLd`A8%Z5eGr5S*^-@nNQ4|&rYMe}_EV;Y+Ti(t0
z!3yUUrmKK6?SAGz@B2}%pKR{Q?<Te1&532Y{!G8+-pip$oKdjrYxZ}{TwAdDbS@>&
zby9fIS!FW>jtdqEV25?;Af`lLkE8~03zbzDIcFGdDXP$8ye^5C67-OzXrb~vbD(s`
zE?Y~3dL`C6IZPPDpG2Y_dvp=~OUtp`EyIoWz+19Ovs~YNEW=TIO10IwGodB;8i0#r
z@A7T!tYNc;cC){H^J{>j%dnx9NbONtK|VWIKSqt0<-vRZgniXG^K5~PD!LL;fpNQ{
z(*oTbj*hRbCE35XxzaCku%|{UUz9dw)CT6z+!13n1tQ{D*$#iRc^~I>Ti&H~4#P$q
z<3?9wR{kAHZr1%WyFmQ3bzqH^xPCL+aU3EzQupc>#<liPn4OL!&^XORUk9IftG<M0
zG-*{neSjg8xy35;Q<)6=@|tMc2J8fTUHD{27M<j@%B6Y-0yx9QuW4_Hw<^~DOyUT`
zVfhxF%$EJwS_3h<gz~(feITSrreRt0Nu2qlz5cI=f^miV)H->4$~&RlM(f*q=Y`q1
zns$<a`t3`pHK7psw~kA`zP#d;CF?1QLNQwa7ceYaqIen}S-5shH-KTIJ7B6nix{M}
zVmcO;YxlkG3AeVw=!Im4Fsbls9{+D`eWlWUDZeG6TW~|5x!L*+MHRe`ASy@w03kub
z^b4h;a>ERwD8=%;(@gtA9$l!`8Ef_nVdOO)rTFjY<>CEew=KHoIUB3YniogfSXLD{
zA8=Isx%E3{ma2s5{)Jhm1&+_P`|Z1W!c_BF2~UTe=gB{V^{;NknFbAi4nPrhjKc`A
zS0ksZ>9xw;*lZ}K8v*Vd>9xLUAHNYL=7~b8>iI`INy&ZS2pSnPFrOB`jgPd=rL0RX
zdMVSY+gj4mhEuCIYRd64sOfZYfHH?_?+FcHZl5gyo|IWG3?_&y!raVQtt-MGE<*;n
zf$<Owt*#<(5bDPnnL{tD7H1YOU}KK2_ZM5?wG93N#k6~#p=nWZo>roxhI?6(()|wV
zn!H|f^x~P1&$)_Nv&Hj{2Tbiuz)|WECB1L?q3^lhud*FNz5*FWc~4g&wJ?F0!0xAK
z<*1XT6EzGef|n_%-&Dh?;BNCeMoSrm4FMgyd+CrfxaNHVPC{eVe{N%dr#KG28ChZ(
zDiOr!+x<vX@j@_vq@{xuCH2YUa94gm$yNZRT}=>)jqE+R+{%3FzA3tSnKkBlnRUsO
zzen|IxKpC05vpg}7VjyHL5Zun_6~waTWGHImOC|7)-+=7I_+keZyGgFst1!5GtEUF
z6}Ss+(XgJNETB8t#;V<W<*gIf2!Hn)q2uxJaj2N0FPzxxnmCaPw;@!CA|J=g%pCSS
zOnRK~*v6V`xpn_*!catl5kCez9!89Xf_B0N)9dh8_-YOZ+xjnEa!-NVqycr=LhJnC
zKals-td}{vv~e!#v12VSsuCPtG0p_|Crpp>UPk>Ba?as9X3TP40{U--_APWn32q!@
zx%Wn~yH+7e#x~d`lu9NMUgD&xyITULgR?6F<HO!FutR`k?LubWGG)r&H}|d{rSUlV
za8@aIm7XQzW=;QM=)#+|0x&yYl%DWar)d?CTy5hg%kj`^MPd65q}*CuyO@IF4Avm@
zez135-8I)Yc09R`Kiy;(Keglj(DGRI?jf9K%6+!;U%UjhnS=#@?6j-E+4Y<Q2lxp6
zE!PKU+AB$f5r0)H4;OSAoR3FU#fW^SWFhx|qZYPk|BOuS>m+u{k_-K9h<chp)5y$`
z`Hn0PX0HSv+!mdSt};{gMtynlUA>M&!-}9RQilQ~GA=h8x9IUX)3w!{35n%%k*lsr
zj;F@0iJr}qY@~?hS|q-yNSE7g{2d<GL9NdKfvR)v7tL`Oc3PGsl7EcJ2+TQx_3zRS
zt?yvg8Q1DZrFReKwt?)|L<A~41OiHAm8K0Rxg$3u{#}YmL6=xA-HVx})y_G-%srG#
z4J_*<Ywx}x1v4V-C<{sciiIjb>DEgZ!f~7@>v^x)2wNEs0Cx?4`IMZKBKL9ynJ619
z0dQXZ<*e1%@l$}t&bKtnGt+m4vnwEEX{XRfJ5~Kb!Q4^zRD-`4>-2f1__6jQFy;xX
zsHWy0%JlWzE~YBqpE+?2Zc*+f=gOidF%`A~9#QgE9B>_yc8^fO&muaBvL7<hSpH0E
zCP?9o!{9up%0Z5ST=(W19Suj%)-OAbGx(h-vUo2PcOE<pi|vTX0qk+pS%7v{ytg+Q
z6ivLg!G{!EaVqmZ*Av;BTB+|{hdhA7JBtK)EGmU~9_-EIsa<O#fhKU<@O)t4#hrJ;
zbok^z;I5CSN_LQCpomdsv?R9>V*|cyWO2Fd+S;3c46EInC>)vwsR^Q7aNSu;I^NQt
zc?QnB327A3y!IQ8Xfh7YT7p7tQVpIo+-T{1U0T1K7qCRS;xr-nsD9|fgb<r~9IDO6
zxtc?Ayre;V_kZVYo7dh`8<6)3-5i4y{rRg8;zAyv@nsNcTq8&7LAQYERF?HERow|Z
zQb}hw4dR`+Dz(iMb=j6$K#|&-Z@Qi6f3umsQuIf#@ewl(S_=1?7&B*_2-_CLYt$#b
zTXj1g7u!+Z1l^}WNont#N4*kzTxgy`xZEN>tgE7*_#7T1IGN6gWrhpdn^R9IZMll_
zi-e}~MbQTVZ>gd>5edm|rwsmg1m}{co#K-WQ@$cGgKI}${%yfrl&nZfjdBNkkUQ-S
z8Wjd2ijPIkuGOb&&H|-0RR|{7gU*t>>mh=bsFRjsV6Fw+KiJTBLf@0x0L(P5P+h#s
z6E9!vcDvAhT=rCrQm1YwWzU}XGhBiU4Q13qPd}nDY!m93P)0v4%6kH1Jj{OtJPue&
zm_X}~vK}M*<D^E*4rSEsikeP~W)nt4m7$Dy(de&rS44H&a;QW8X$Vbd2ap1-?W}ve
z=*z6Jxigkz2+9;W^yf(s&t~(v_LVFgg9xqt#ExO^c^(TQ1us;~@|I@Ia~0vE24X)}
zMQSF0IsQn)Y==jD#dOP~;@BSrnsDD3S5Uw5F<uhbn^vHN0e7l#<;QT~a+znIT`X7o
zan?REe_U2h)UacdvK2rA^ZG#c_y6;6Z;X8)hSi~W*<EA;O%vCS!DO`%fb;)2&L5c8
z>Zaby=Q5HC|G*yEeDUO-K)>7{*)SBzDpBMOT4A19t5BB@Y{>m>-FTy`_t)>NBit-=
zP9yF5BJHe)MFdrC7#oQ00DbW@+0ln_5>L)dpT7KCASGwn?PKq;cTdx&84K0YrGXq;
zvlEU>_mI?-TlqnuPpgg9orKdsJ}ezY&jCNeE&l2!pdig~uXf0?HF6GC#`_|gBrg{e
zzJ_n?eOKK*zy=W}wAQYvVT(G7MIp`ouAp@(AT4Rt8A56yxYlS&?iVxFTSqv(X6PQw
z8lkbNFrW_O2_VBkQaOWQYB2!8yCrs8y@Z5Q(I9RTXIyF{vSxx#!Pn+V$bEl>64xI|
z1O_HVN+M`Jpz+pw=~s8yl5@z&&@92wh_Y9pCh=$fS^;a$5iFx0WAeWJ-ztki&1QCb
zLuP*>OoehB7$4VtOdzW2{#~(J&~(Cywn2BcSq>!E#Y)&??Vb5>prVMhW6L9V!aYy?
z=7|>EYBZ&)X^FN{994%6rnd&);M7)AT_cA7KK<VN6QsIEO^EcFWCo;e9TFk;QirEr
zEh3sxY@-QZPa@w&erN>SwZqF?Q*B$v)lz@sQim$}5F|)@uh;q71PZvw3He6kAqnBW
zd(n#IIOLA@zQE4(x!WAIXN<8SR$wzq)ONiAHvG*QGBy}Iw%3HpO-|x7(6!=DjET^F
zlfyg+!|&8C3j*J2)e=*-;}G2`iZW!J3og8d$(IQ!f9V~rYqLoVLtDQgSesbu90bT{
zUI$?)9Hbw~i}(57l0cSS?VDojxC|c9jIeq6a7cu|Az-v5Cgh|vnxoE|HN;}~On}He
zUou)J^jIfI=5%M^EM6jf*(m?0z<Az`zKWcOn$HRmK8i|Onrvj|YfGQRV=6_M?iNsa
zQC#u!MGjhdwALjbVZBomR8qmSc^gVQ-;=t->egh_zwuW}sAV*W1a+nG9szkN%3aIE
zrq|`z*Qi?dh(JNAuYxvia#cmC=!?|<N^Calg6HrVniiSUvoXl3-)%Qu5LGFIhY`ND
zQ;4v9UdCw*@iFM?C%*`|=jAO;a;|w;wgw^gWO{y|y_T`gAb!ax>%6NUA_r0Tau>B6
z0@IcDjBtH<8?L-;%0*=JQ7{w4;F#K1O3KCPqpBOs9@FG*vV3qR897?fd@N{@ytM)3
zX9fPHVr*TWLOe>^r5T<30$mA-Y?#C9@NBh5aCNpP4T98ked1SYzx6{+y-T3^lcJWA
zx@BDwFu+C9VzX7d%F@F6EizGdIrznVv1jCa9&BT*FYm>f1dU<*odKpkyy}P>iUj`U
zlqHPa^Ct;Jhk{nQkvQso%(2W2zXIDeq3@+1RXxgV$8GB2B3o>>YakP09I+@Lqly+L
zjSVsHmDvJtt*JYHO!C&$lq`k}3^h#@INf{wUVc+x6GVnBD<T$a9)02nOPht%Z***b
zYHr0Fk{XQ=zu<wbgCZPe-B)RGKok`Vg=TJ`>yO|PS%FEN=QU9J-P$99d7(9$5>*bk
zX79kAHf95F{zoUZnymgeZ}N=M?2JiF30YJM_Ty?PoN}Zhon<?}^j{L-(V&v(VwH6~
zRw!sE^!k^@L630?Nig`=yRD$(#i#(MODWJ02I~;!J3etG-JY(kA&+f6w(Wz~pql6K
zb{1k>fbBsd#S7h72EX#ZUEjVSxvQCCQ)#s~I^ao9I1kLgOp#gF3K87k^vaZ>OB-fP
z1dikqr?(nyUC!5Ek0OHCPiV!;Z$ZE^7Wp5H)AHF<Q{%3XV0IT1YAZ8pxqEDYJ4J`9
z|3`K27+zVkZVe~hLC3aj+qT`YZ9D1MHafO#vxAOp+jidcv(K~lgT2o=-*>%#-dwqo
zHSbw-&Z@QMs99C_SYx<uG1(U9dL<7IWPMp8ZAzxXVA^0YW<M1#wsd>b37OY?sbbHV
zdu~y`-Xm4uoZ&4td$HNIAJV9erfLh?t<yg+9VAi2nqTErRuKs|%30FPDNwe-qOouU
z1B?w89Z4;Y^Tb3Hqg7d6yELt{Tu<Xz^saoY!`*bMPbK6}(DsnT9n(D>^~-hhFlUJ6
zEWOs;Zxs0WQsPND^8S}pKN||JJK1#;|Gtu(3mo}3dqt8u@PNLf7t!|g885-nt4(WY
z0#<m7gCYqVr4uPe0}`#3<{&F?qHVU90mDq}K~T2F06ReyR!T7C9uQP5rWf!9DEsmq
zI1OQprF8$2@6ESGeJ|0K<w0&w`dCx2$SXYiP=v3cfW*dAgSKa*g&elb!1a(Y{9SUK
zDl{@eJw6p@d`<S$hY=2wE*_=zB==x5?ft}P{)d8iKe0`GOqD{b^U%Q)hV|Vn%IMYb
zV|>f5v^#7KqTMsKy)q<GqAk87=gi;u4H~L@#3||5Plyf+1_1Y074yqe*zZs^Ivyyh
zgM|%=Izl2aluEA&Sq>!6t-vn{Us8>>(RFz%(GqZnQlrGpMV1;|3z!zr6Y=d(21BCh
zrC+m2B@n}@Du32Nv(M<7LyCn1*<F%i>YU?HdlqVDi^8<dIda{QusXSvOZ{?uNNde3
zCGvde()QJdUOYf^w}GQ4h@srr3_7259u_JJ%A3tOjt_z)G=BW<p4bke>b0q;x9U1`
zgZV03p&Om09d)Kg`ca0d`AjbvPc$co8&_QsVwKFB=g};0UTUgiNy<&TxnV)&7Ol*B
zcEL<wfm_j}SJ<e(%7x8bE#BwhMtp#o%EZIky*^=q2w>ebW3UzO{;)QcQx`B0mAgov
z+f)AQszoYm3=5_^0Rmk0OaWD*QY~O<L-;cMG$s5kX@3XIc#2IDn7)E&8?&o<01hM3
zrQocv_#5bt6VeJfqqYTKI{$*AcOoRe78^#(xtq0H>COrT{sM+fi@+IT1ji!7pHt8Q
z73=<aTzorL<j47baVPTdwd3X(&HGO93F{D4B-^Y72Nx~ZZAQ$R`&Bas@Fk0c<R#6w
zX0Xq#utZEjLZ2f{M>`I$Dm6#0s?7tIPNyr~x#Zw`WI1IRIre%+Fuv2zPJzp1<vgy{
zOC5_!!(VfMOS};DR2R3QPnC!k@;ps75D&#VRU&QWtvwGP1&uX~FKe=GP9b1GSPD$s
zu@ARHmp)eaO^*Mu<BvlC<`B@I4oI}$5|~Ph2ia;2BwdYSxQ2u~oL&%B9Yh(fF{jb#
zwbjKDOItqVeFG|FaQ;#f5p-e2{^<hO3t^qNMYWU$4OgWCqi<wMDfs@n@<)4XTRqoe
z-4nI#_soX{VZ(NJS`VI8GPR1SU26G$7Fk;1u6F|OoBi>{td#KD7ugD(b_c~fFM2fM
zwz(pp>kP24NJ@;{1kM(Z@?HK$S9qO!IGG$zk1&ZB_&ir|1vxzA);TCk5p3eS+@ttF
z9s4I<P#o2(*MT^HtgYKk%G_~9(T~KY41&fk)ZS75vE#<Q;(F~Ud5)xtI2c%ZN8UV0
zGVbA*i972diRfhCfxXBdSr}>a;53&ucGnZ|Y)1wyzK+ZX)Ku3r8~boF{n90&*JNF~
zYt2F7j>Qhvxl**~$X7pi4F{g6y}<RBCmi$MfGk3a$MI{mj3`2)7D+Il$d0|L9{o2K
zqJjxSmTRVE5lQv*DsTO~k~PQ#5K)P;Pm*$tsmerzv#2MIwZ+d&I8jM2i<wg#rMf-c
zWtt#m3~hbg(wt?I;gk_*)F;Dub@8(N^Dh^fWAFjDhpzH<qX;1~>B)jfNqRx5qXG$2
zZ~WXKII_PLlWwH+bnG|<(Xgwki3PcL3AqpCYdlxh%Cl3ggjPcZxSWG+&^+nX0p1aF
zS3x?asn@{n4?s9fPx;|2T;mVU;x%xK(e@TCRiLl&h9u4Qe?7~0%g7T#GOca{SyFqP
z&eWC+7Rq+y?U#hmzFJT++N;5E_eNIE*fLCA)9JW2Z(#}OJb#>1vhO|>w~Z)wtlYKD
z1m|X=x^9HHQ;vL_Yb!!17lE_?IoonKa0CWU60&@J^=WHJ$FK%@E-N#yWH6v-?T8H^
zG@XUQ=w<T@J%3-T$%x;VO7S5+iV<jJlc}1PGj=n5SJe|$4GsU~Z`R_D1Q?qx?JqB^
z3Qy`ftonR=o;4CyXf^x3K?8A*&i?J-<!wpLphk9+a?kC-+680`<#5nqrA~{do%}>d
zQ|Dt+!j_*x)MX&WtQB%BL4;^jMlVJ|@Gr*u)|Wclsk|>6#zD{@&DwFS)BL`2X^I34
z@M!^i%TI$W81Wr+w;_hsgy%&!9-O_0v3fUvs@POqz3uLVTH1Cc9L9aO95>=iE@H40
za0PyQlA{fh-FN$Nlya(2IqNN@-ETWey*rmySqCd9GAGHn^|FxYf|-t~9c1|CyUv>t
zGl#Kp5StiZNL(;;J>vPuy)ubckmMn4d7IA(PO`HgL)k|O8~rxQM6;f%XrqpZipgS>
zur>aZF2Q!Q%)^CM@9Cs<%8T!_(+lT$T1IyVg9@wi)_9oPg&r-*?b(nMEEC9pBNUD9
zgJeX6U&a$O!9;$qIouw_R(E(q-iJK0UVFxlSbDMDvCjD5qHmEb(nXEj7AC4hd?@`H
zrnBC<&HcAxS3Q1*V?JX3$r&S}V63P)%Uhz57GYvaEDZ0!{nV2x7Fl9mLR_Z@PoqBD
z^xRWDYa>s}*Kx_ztMg~nT_QC~p*IaN+dZRjk489>M15;=fLp)48wQLb8cUg1$z!Y*
zKyB3Bd92uNC#yvz;7}|Ygdx;2x0nTm^t*?lF46=k$==>1e0WYJ5OkLBi+1{@+6pxe
z!S_vkAccOK1YU`vB(<|n)VXRJ{(7%$#gzPAia%#Q^ykkzr(z(4B62ZQDJvvm#5FQS
z58x+hC}E4SZ{q!XFSh1#4V*K;bmLuI&+Etrq?_4@Es6^{;yJC`b}34l1TwwI=@b~I
zPmmNsHu1j)u7!#^7blT~jc;s5+`f=+vI7x#9d%&jI+=6$X<nJjugNp$m@qN_SYSd^
z1wE&a^(0Lm2m<CxIJ#*uWt~nA<FDpI24Ki&D`m$L8{Umv;Bl~jWqq*Sx`9WqbT)1+
zi3P~N1~KxoY3`m=0#Py(H5d(tvn84L?wZf>!N(#Z4LQgV6}E}4k-aSCCzLu6UuFJE
z<wLOYiR1_M8S5BpH!Ic;0K)5mDb)peRrZvE3sY-<siNX#V5nTaDi~7)StVYe!PogL
zzk!2ZX?|Qa=^ye|#GtcC+;saSBoeQQ<gj2L!hkbFnB>t#z)fd0kz6b(a9Kf`+=>`S
z{0Ieh_8QEsg&uRIc7p-k5EmnTl6x{b#FNKwbPKdu4lB~{{6a*-3jS%svO8X@T?}~=
zw)xes4K6Z5yQ|CkyFVm2BzV7+CU2-fEPAbiThHM<vrT)Qw6<qn*q$OZg8nRSV}>Je
z@xM6Hf9jGNEpF(2nUb|NP1xf_S^&7X?Pvy*8XG^D4Ot$M_FZjHZ-bB3n#_knQ5i9Y
z%jqGz>xFzJwByi>H6$ll=<EN4eeBnBnlL_lig7X4oa(k8?xvC%4=ZUymSY#52&Uz8
z>y4I%WlUPcp}bSKMwKyd+M!#$m?2Ll3~v$96EwcwWgLy~b%b03k%e8!EkZS>EgvTa
zFHnYNd|1vvKM=fvT(dFq9-s>;6mra`f-Pu))l4zC7-**$A>IVJ!8VSeO^1B$S4G9X
z-CyOjV<8tmbGO*#7;9Q-KuLDC7P9{Qg*v}wT?meTN?H#~F1AbKN8l%&0lVh>*7z}u
zh0GBMmD(G}ou&^b9|6%|$e%DQu?<_8kI3gF30Ehw%pI!=)f{fZ9oA@**!odD^F}q%
zAm5{Gb6q$v4lolXY%WF2-t+K@D>q^jLr=#YO+@T#vVKftBHVnZIt!@lmw6r+9?Bp)
z8tx=bu+m%~;OUf0$7c)~2f^DlC1fr00;TArr`oGCyd~IJC}VHG8h@|v3YfTUW|*rb
z3f!xzjj@Z97MVUn*)8lU7>=ev0rp*Dxau#YyqY&&wwdNTsF<fn((az8#gw>7`cy31
z&tHTC)7E<{|1crY8=vk_9BZ3ygZGg=tmiq4Hiao6cd@gO^<E+ClWy~1Uk5FGH(XlD
z(9P~lir{8?OdWbCcNdwe@Ssvt{RG4HN%&^^ZQ<<!V@5{CAFZ->J%aYIUD0oQjWXz>
zvAbG|cJVMkG(9hlnIN9HO?lH3XE2m%KgSGn`3iXe5bl<3+CmGw?H-nj2lMsR9-a>T
z<O0tK+R9$US{|a0hQo4(Pu2>2=BAX2nPPs`xN2NG0t~g>q?*Rq@R<_Rr%ol~dej7G
z6{lt~w1^t}#V!rB3XR-~KuzW;u^_^DO?=4qf{s6Qh;c#B2ul>K8@S@q5d`Ua42Y)V
z6SIQ|q*bvS!;S#W5nV6uh-;rdC75*v_{pHuD`py|pgi9o;@Unn+NCWE&Tb&$(u0Y4
ze@z%1<!rMj&2^GoL=bnmsK8wHx1fAz`lqB!DYwjGNLbNgJ2U9}eZA6Ry0jdf328@(
z`;>g@Ht`msFs}kJd8aoW`>f72EsUk{oyzEg)KeAX0u$FCrP1HnkS8x<gqDjS^7|Cs
z7S3FDVZjmrjcXKoP|b{X6w)p*Ar6V33nhccat263Sdv~2uaJqwItdjB6=w4s6JE56
zn#(vgx^RCOYsG^ix!R{cmoN?4%)E$Gciy*1I#ncfqeMy#op6Dp%P9kSDQul94_<WE
zwbcrvO?e6P6rlQgOC5$#zQ<Gu<(m_uo6wC4k}c-3p9%_*V>g{?v&6)+v<ah4s<~A#
zIZ302G_+z4G6v596<fkW4g@E*+erq-!Vk%m!R<+P#uaJWe|17Sq10Jz@0>an-t2V$
z*({0t)jZ!7Z3!{qtbND?eLJ#Yn=Tp1?v?cD1mhcl3RhewT1Q92<qlPOKAf8%^CD<$
z><0Q~z<@D!WxLW5GQ`g6152-T+bS!gz$IL23*jhk;pNHP_HOQ>Uqi)R93cP{GmjYJ
zcTX!Yb+<;xjTmd4Y<c)0akGj*cA`@#zE5-X2DVXIS4P)Z#dqe6X28^^z1$O8s_wCu
zpna7~5N-!eeNO#9qK_-%m-*t>DAc3{^7HG9Q3vc~L7TvOQC94I^Xjnk>KbhCi!>m`
zks8m<2jaVfYO|Q<SibPx)U^zD_)J@1TtR7F35kQ9(&Pkw&XZ@K!RI<|2OgjNq%p<n
zzSRj#99GqNbtE$oVO3`=FDUFP)Z96pVk)IXR)#_n^_GHv@Uh5DN=w#V%om~X6{0i+
zdx~RB!595yhm?7~ysEP6B62J~vBF})@MnDyBlwgPGr<|t%!LHxSk!S#tG|K63Jz~a
zm3NdH);rb0LgSQ$%hBp-c|t&K=JRP2G(y)UnM98JheLYk9(?^{KQlyOkK(0`IfO?s
zs)}>f0ets`0uY_69s^J0Iih(WQMh5zw>d(cHa<zBT0?GX3T{qZ-}A&ce7h=Wy;cIh
zbHOkseOVGw3O(Ef9<9fe;E*?XXL!#0?$0}!XaMSgVDB%K^~{<-U(l7IzXc*Cu$tZ_
zdvHw77ZlqO-lMbJ;4VmK)2@A~aY}iEP4X1MS<<{dI}xp>^@Z8)+LUVL9gJ|DYk`V=
z){uhdl$A|`xyjRVSB_HsaeSRbJg=<a9{&1u^SZCUEcE#DtHYl1tpo(OzoQnuWCMm+
zQ}o6rT)RJq>piPp%0m=F2v8)6K*H-N=Y3!&`zi6MM4XWcm!0Y&YZI7LvcR%Fvz26y
zEZPM@!AqdL;(5fvDpoR6-biFdi-qYs&c$Qj%-?Bu^#8Ia`Q51HKkIb}@QKMt3;&|s
z{Rg=YhJVU+FfjjLq22vAVjc1{iZq5arZko`Ry5W$)>eizHZ=A$e<kDn#=VpOBN^{^
z$CO{rH2);yF|hvAyW*FA2kl4u{;}_mWISe8CdU6Y8BY;dQBkwC01;JCyafbMh)3_h
z3h0n}NZ06_kFHTN8aohNUaMRS2ce)iKTw{MVtk@7Kl@GO<@@^UoBMo=^Y{8Owr9`G
z#yhW!pCg7klv$8MY+W;QqUqnNa796iO3W?Dpn!Y?f%$OZ0B9<~FoV9m(tURMse?+1
z84&+y4WI-G8Q!}*)5XdujS>K4=8F5RCx=%T3{6)I4fYG^r%z(<5Wzh^0A{-~Y)EBQ
zV2c9yd#V7mln|FdKReqEY>vDfqx=AE21ZOwOx$@#f>w0$)7^ljg)f2_VC!AEK*G}j
znm`5i>9yg07wsn55AV%3^!IOTYXj=G`pSnmAhwCjGXW844}#uZMH@~63I4)HD+_Z1
z^30Wj`whip3v%$S0V5aFR!avGGMEK-0~QrNEb4-nUl$d(gLUFVz@sQC5YOAs(d+Yr
zkM|u2DJAL4xr29(H?c2)H?SWbY&-^f_>B+2_ivFlK#)b16$HW81DBt>MGs_fQGJ9q
zx4ScRSU;V71ARHdgj7^v2J*f9{DuYt(cLx4T84`p`M{PlG0ym)+2QA3lC!d+BZ`T9
zQOm=`3Ift?-5ND<9_J{O(-!UJjT~u1ICHrVD+^N>O$L8<1~{$w{JA>~z<W7ehc$~;
zRE?h>6Ue{$v+f$KadzgRl*dGsw)AGE(F+iumkA>JK`_oIWB_OXO&$Oq$^`_FK3z^7
zK(F`J%W)*$uc%)F1Z;RgKU^ok9ymDGAnjk;2XMGS&Y<c(Z8noY0=;^CI9T}=q96tc
zKD<D^J9hA=k_VzE4tu|5JgjpgA;q&y$nXlHlj2fQ10y0M#{r0oi+}q3mNEbU`92MD
zaQ;;)j{uzXwN?Fv`L$GkoCj#*nIKZV%X@E1U{e+1^VO@Aoq06t7hi9%_s-%siXCtL
zhjGexl%e-pJVzLyt)#T`q<8P8ZYvxAaU6h*I&9W900%i2G~9OtGw5e3HBD5+v(39q
zEh$V?8X&_wgpAQ&eipdZ>rSl7SZ7dv=v}t3`|I(KmMpJiOhA}m47|FVJm<(Bh+l4S
zxZ65#D;}Odhwbmupg_`Ode61{`DbPEI1=Mi0|M#O>fiu;ks*Pzwav<;)ryu)zNaDn
z8~_wUipdQWoJj|85ZnXcg<VQ3`z3-091-zDjqTt8s=pQRofPo0OXp%tTe55Zef}w_
zLYLMW73g9U3mpG>BL7Wy*|wa%G^s0TbENhf3+^(}b91W+L!sGgj6E*}9pQ}pZi2iZ
zo6+yyP<Q+MMb+Ay3;VLc>yp)l_|^(KazlDOXE>5~t!t<eMVc0xG_>DTk)BE~#)zXk
z`up%&Xy1`h>|^V-?2y%D!qIL|81%_7*CA9EM%6=fDn$YkN^bgUH6NnkP(MQyMqOP&
zWPbnGeD`FN)JQE@p>F*s#ig7S84u)gJq8dSZ?D@2L+#W9tqv!d2tg%OtZOE1s*02B
zJ};vu|3S5%Uwtviw}MIx$u8$p!d7~$Mq!sF#?qzsIz2KbXpSZi@(R%<vvD3;hH?Sn
zQi$+SzsV09Zf#o-pJiAy6~|~~9HA|>imY3-z*SPqUhDHZT1~Kvg4j#f$D3q~jka?1
z#bw7K*oSrE3iBLzI%nzhb*B>y_34iB_m;svG!07>{5;foGsuvXu_)xO=4x!RJMyX=
z)x6si75lzr+!s=OBL9>6T%yC13*n&ojx1e_%Sphnn9ax9nd^K1yv!1JzD59m&1=mX
zk&l|Lcwmzp#`I!$%n)&=m^yk-YS9@t`qLuJPxf31@q%4}pxbe_TWt!3b5A4$M+uqB
z55u!-FHkZq-6RrjO!&9^#lSRyFXra=<V;2*CRSv63R`Hkgy8p`()V8~-M;~cpKGy^
zRi<MdtsKpFzSmW*+B&W4Z>tU;x4{}%I*<oIw{Xw6Lx0#>WY4HhZzVa5JNKt&Z3w4O
zc8$!Fs^63DQEjqRNoLX5E2>3tNgsW-bu3bl?0XzSfy(R{<6{04I2O5)zv5e<(4+;u
zG<JWm;Rfz3?MJ6F_w~uC*psX~BAgUZ2)UBat&s10^-<Nb!FuftyW6kTVv$S|L%+04
zP9eXzwsmEmK6tP`CFg;<E7Z>rMd`F|{IrtXuBiEnxnjL2#5v^6fy#-u+p)8^D^s`B
z$!sfT_IV(;yh09H4aR@Ps(cMlXr=y2Mrk%tmyp=e<X{>YQ0xFED6e5-G8q5W@DoL3
z>l|PMIs%3`pjR9Jok3j^MJ=oavxF+AuEr-f9Mo1@ZOO^E@W{6*$GMlUAAW^KKIFwy
zD4=FPVwRH-wGMl~u{8HtIr7s$1@&BMjno%auqRX9t#P2FDqDRQZOOefiJsbq_D-79
zF(wGc6XF8}QD|zoMntz`Iwby<x#T%0V7GK8*dqY>J)X5gT!&V=WnP~{-h#nf9~_h7
zfk=kIT!2UV7VY74T(YuUET1NKp{3z5bC12^mXuIj4zfV79oRlgeoc|lHiWe1W^)<6
zXx0^q=+sWNt~55{*O5}B_Bde)k_<&2vY|jQ%}wl~)fC%|c+*V_U2>#=@bSLc)C7(E
zx{`vFEmzZ8Th7}M&W*&P>fFyuBKLzqLf0yC`-npb9>D1MMiNxFjWGrs1-qk*)naM_
zT$7SmIEUmS)L-Z;yJuO4iRaGHv~D=7H1{)@8QG8&stu@%z#`rBmLO^BNlcL|_j8EM
zwq#DGZaoi~GQb~X4@I@mXRhD8+)FFGiKyw%UXuboNj+Liyu4v1rf>4~34<M~XG>De
zg;RRmkyEwVkNpIBVNYA>BSOSqQ9y8*)wtqbe3dc4^d9wlcs!iq$mzFKb0Yg>FMH$|
zyyZS~mhk#XRCxVD{t%t!NpeGpaO9qU$^8<D{FS@^YQ5ALn)<lEZk+AG$z0&Jsl)g%
zb(ov}(W2LhJ|k{xc+{G#XZZ&Zai&HfE>3-pOT&B+8_j-@_v1ysXEQYcOUVK5qTBuf
znu-KoG?D1qcvbWFY%wJ;InqT_Y$7Br17vRwHE)RHL@u62#nX{w`}KWNwk#80$vKhF
zb40r(*hL#5D`7+KyN{~)fTz^UF5@`QMM;B^T=mpKH9^}wL(*n`hs#wzExw_oGurP*
zOsXAAAdhr0K{|#L?|wSgl8Y|f9d)gGs2Fc@JX||Te>Y+wmzr71HxKC*eA(Rg<io4?
z<;$It8+|E|5qU06#>-h;JgyY+HI@d9C}7vGPo(suYB7V;a1~N^J@T{CQaODgG9nb<
zFF|6~KIfWHQ}smp+H9x`rzf7B)AM+tY8&Lk_E_x{r(EnRT=bxM<|$=b*z$I)YEnFP
zKiSct?R&r|B5~hY+{lV5+#Y88rc=TY&52-_;iXYP4Rx)p`(|?CuL&Izn6!d+An|z|
zLi>>~X)Xl4q^85~xhPfuSXRz$vb)~ndN^*5qb{>@W=8a^(Fn1v=4R5@z&XL6r}-v}
z*L*vXvxl|4XEX2XBCBmM_v-+*e;@DK)fV+i+3;nPeY};G)SC~4o61rS2b<xRQxtyk
z4>+f&xj+-~y7vg9@H%;d`fj@8Fv46rszPdY6@JSTEv}8HKr!5{UM-`&w_KS*X-XE1
zoi@K(6PFvXYXSDOdhQAXiB;dV?|LXm!#SiV^RZH=1)TyvBip6cOU#4nv(UW?+H#0=
zD8_JM0gB3KGoj$R7sJ~991E@8*|PRzlo*Q0Mf_(-f@tx8eQH53J<oe+z|OVhb~_pz
zxTvSBH<+x96{d%L5D`-Bm>MnWrcO`D$~sdfh+d33Ssa;#5MRM!_z#0`%gHlTja^?>
z{eieNzsmN1yt~x+3V6j_j%&bHrD!1nppa{+f*G#E7Ks>Avn4xBfc!SF4LUmFh22u&
zCF*hd#Tp<+v$Kia<<bOZZ9;rkj*4ZEGsp_$Xv@euW3=9@8(b42m(QXqr4GRGGoN9T
zO03gTmod0)C`7%Kj0KI&o)B4V;SU*cr_C&WIB|quRb-UL69QMFI_^=Rq!0niW`oU&
zEnTi*{}kmlGT&+BAKuQg!|9;K6pER3S+0uWWvi}LAxs%OAgUakUg@xY?6oBzngbVU
z14YW!bTDc*EwG=gIL<#(4R-$C-8${j??iQGw<??i*kKkSPVa<0R8IjG7IQ+viCz92
z=jhWqlkTJXiHFr{wzH)ZjhWu!n5uhcsS%<nNO+;BA{MwH&l@Z50{-}p&RB|5(*iQ*
z<GX-(na0M1-B2C#yUlo#c3oN3J^cL1)7vh}--%G|^J@3m=H_QuSo$v}Wph~;h?qs+
z7JyaE8XoE>SoDqLB*P%>L!6_FTDY*-;Twjtia*Jg7h6^kW(7D|rFa-}kz67_VG8Cy
z+Zyv)MiIbMkvgqBr;--l3?~N^FGG||4@%bdTVr+K`;=&^%#Hk@p=-z`g6>&QEwYLa
z(psE`##P{j99=f7Qfqz_jW^d(b<kGx%EM}C8a@=28XqQb-5NZ6$U}9ze+mx><NxNO
ze4=>Da+Ee49XNA39h`S43DnckTpj)FMGM|b<5op%cE#Hj-T!vjb6x(J9Garfuq}eX
zr<T{vHf$T;c0grCPTv&<L*j(Y26I*wZtG6RU>Q85x0oVAJR7{t*n<rzNclwAf$s6c
z#1Ec`qZ`i4OdLkDou;GfE>HT@&3omPKlGktZ)NZFN-R}?vT_Ll18vg2`Jh^Cf9u4b
zLtDZu@esg31(A%l3MF$jguk^Vx~$kUCjD+%#Xi)&4ZMb&6I7rvnNZcu2|79Dk;jv=
zc9@YImg`4y7%|wMZ;LYzGn@KXGddel=hrRQ;}>0SU_2{XQ-gARlcBkjD$faDDI2iS
zqU>rdvZOG)mBNd0m9sC^wOU6EKBkH?GEkNMQ5Es#imsYl8srXkhgp@vm6I#B=o>@R
zWjw-2){0M&dQNh(83Q--T)<IRr9Z>KuFvTB56>s<^V67)K(!h)%(HUS_9)retSbxZ
z%5%h%W-SfvBRjm#c?onDpNC!0DOyA$xD_Y_#N&ed%KQg!x1R{{fT6Mqk969SOvU%)
z3H8(19UqSpT+JyFyW_39UD+8bAG-6Q^&P*~efG9$4v{)*D{oqz+_J97H|Y40_>9+?
zLz<~ZpzIh_5}iR>JRct?<!8>@9mj7KF)+TlJAHijnVmuf>hVg(MWYBB)WN^)KHTg)
z2yy9{>th<bwTFDm$Z6-nYm?2R%#~pCywPUI=-5}7HhR<iDbi274v8su9hC<)A^-`=
znFq4vld4^jb~M<DBh47=?mVzqIfzIf-5hi@$SJwBS^Mas7k1AZmR+>dX<K1`f)R9G
z)U7gV<~2I3ZS-lCvE&oA((BlW-zd32R$tWC8ZmiuL8wFELJ&87WAbA-DB63ZiDCx9
z3dZCclkef<Y|prX@G`=HW4pBEwEN|PV7zI|3S&go0Ap)pdauS{IO3%d>|i4)hHnh0
z#;j}-W;(n9Q;n9*px&mQr}*Kldy=4%necr55c>!ypoZ%;lD@uv@N8gAZ2><U46$v9
z`jRN*s!rv}ZhRa|VQ;FnI`?Emb!)12t=#jQkI~6l(zIXGIlr1gQWD8DxGdi0SJjyW
zY#%LgwoV$RjBP;}=9)>9t+$@bAIvadLrcQOc)kAd-QOovg|@JX1U(%GNZ^H6r$T;8
z-@D9S;4hWRy6a1nq}V|?HbYoychkc<STNIoTAIeV^$m|ux!KT<C=0ywDmg{~==i|q
zLhi4|H6I)upge9dHU=xqPG$b6e!}B1gQ&&=hfd`oa`a}PDma%Pr)3Y?>Nt1*VpYRP
z;U*k*twz=17G9^{_96$0njo`M2|TtvDIr?rtW5OTuoY=#r--`6SAq&`MS&0Ne)9|)
zhAy#COkH1{kn#?2;icr{6Ms#mork%qSO}btIrKO=ep@ce7lpA`2vbvoCY0hY-1v+)
z%uz@T3^l2PdGjw>#$^hn_}7)-JlWS{T5WrAvK~*Kae_-Z#1`=(_V#DusLXcNvcn$2
z5>}Wh9#tPp7$k{?EHLO%Ba}&4%Xh)fpJe(7mv+HAoFlOC=xxa*mCvAv5%U$;Ga?to
zzXwWxUp8zWqlPYFlpheG1b{A}`A%BRh7g2-KJu2|b$LY0WO_TIJ7K5^6Y{L!WwAkK
z2j2jA)q@Yy5(CkLfdrb?;PqkUdbtP8CiA`3(11mmWIPFFozY;FJwAFW*W_4QPqich
z4`ZkaZs<Jcc{wy}47J2|>l_G{VC*|i(gEt+VdE=$B*A_Ps?~=pX<fmn|E%NjR~pae
zG&imLd2UMpF_Sd>m>q@wa1A!(n;v)JL<6z2Aegj6?~RG;AeK_}M8dt{f)=}=ghhaZ
z;)1PS;g3w1brjhXlinf8ZRK;w@&kEB%HANy!g~{L=g9rpneE9&y-n2Zx9f?rrt}^9
z*QH9iOGX+4M!}>v`qY7<<TThNm!^y;5!O(PX*$+RRD~y1Ht#&w=h4|li5s6ARY1-s
zeck@zo@~1*Vo`I@YcYI7oTL3)jpKmf<7EDTJDmzmsB5E=In|Ua!zg!imV2_Am0qnK
zy$UT(oLnZu%$>&*hQvZS(rm)=22}56xt{NC8$_o>=kwTFr)N=IcxhJW;=PaYow3pt
zSecwk6M0rG_mtHe4Lf2}0#!ar@P~)ac3<5^os&(&s~hL<AZh(FYzJ=2ubRFctHBiZ
zY;9`M<rif04jpZWAbfEWVG?sglf%V)ckp33OsC1G`dS&GJU(YpB6XWGRIxCk-u{E&
zVJy*}bycOt3pmdqI?AXg5c~}8J7bK0O7rkzL68uuMBHHHcoHlf9h@OUs5CGFS?b76
zES25T`BjY&w=t9moJpD5cu)$%WFK_#PGl#_WTNGYfD+tlBzoa!YfIiLKIof`nF2!<
z-I00Qs4N-$W6lA!cy}>qKSG)Oxue(yjcxvj^|tOn!qy!Tco5PdK;b<g?S3{R7enM?
z8vg2JEdSJz?h}N+OphyYhtwk`&)qgH!Mm(YBX~o4m;bTj6Qi3EY)5vg=3L@<leb?t
zA`9Y@2>T-C3UB(F@O(GEK<qXn9cvD$OZScA%7EcnVsPBnk{gAQyIBDbKRGO{^R}3M
zBklUbvP<kfV*Ap@PYH9xN-hMZ-SdYA!ZnWAq>hfpP1jS>iMh5leJZ`Yc{_$=JqGl#
zH(w^X;Ts>IRn+|jxA~C|K2+++U{Bl8ZKkL82?X&mL$O>2y|#M;!`qFt>-{L~ZemTe
z*iW;%q|bRiKfXgNghnTD?`z}H2Vf#G42bWBYMHD!w*`}V;~$4bmg?+D6>51{2}uvl
z0qwijUbM6TK2iDy)j#eks7xg+nOylj_h(N8kvRo<u$P5LA4;5iaZ=uEa9RGK7eV48
z&>^%X-VX*~ID}QtaxX7f&yYRCz@D8pgvB!0MSoP8*$0L8qS#{>E^<Iq{<IN(BVVhI
zW^C;^W26qRrrXoC-PxdI&pz-9byqae=7~#CP~%e&77Q!hK7^@jjJvaOOx0je-N0b9
zC+EBy&OU`9hbRN7yx_Qml;#w*SYO#|{Z>`uO@P^gmVu3?XP(eB{qm|$uiulNjxbf_
zG!A2-8<1VL(DGQ`V+?SdM4KAFC|4<M-R6rwNGtL*>fuYa3Z@e_f!;yt#)9?UHn4d`
zHtqPMZCq2%WKWD_Hb?EW6|uz8$&4YhPK(c%*GYxhbFcP>do!rGA=PI*h=z=HjWS%8
zN&aTQY;Y>u0d@`ApOcfqmWz=ODM?oECN_~ih{8i9S8Vv?Y|aN?)}XC-Gt>jur*VF&
zPf6^d4T(g^xkR~~?(Fj}<kOZKM@+#nD_n3_8pb$qF{UHUYqZRdYxeChvWas9L#5+Y
zcvZ3jO1UrDYc|07Gjk<VnC04R?nJXlFRuJod|M_Mb$h}g-|8*;HJ}`FO_uqYw`dGU
zwl0dH>1iZZE|~Rhu{A;`)J|%$WkcEo_hXL-F^Gb24wEqwyKGWy6?#EsI^-pdBpP;V
zg;n5w!uYY2eC1~mA(1ug8U^|6mfMg~1g4m0Y~YX(;^-`=A5`jkBGBXa7%Am25onqq
zcBZI<wtHbE`*mj^ky7fLlff`Lhe`h!utX54Lu_#Cw1kM_)#cYt20t=}+7TyMMTc`g
zE%!vD{cbGRJs<1Ks}Ny;t^laB<eOlK$=29FWvCZhxu0&Gp~gD6lQ?b$A1RsQV7!eJ
zVuWT2M{<4qcu&1f4^`Z8eRJdWN|`{&F`<Fl-A)c0X-`KkTYtnvSnXr4y4eBMzFyQ@
zI>Gr{KKpJF;pwyy*L>i{@#c^u-&)h_gae+<ot#cV9Abfz9HkX4FZ=RAdGD2WcCL%J
zMZcBZK;T8Mz*?XbxQdQJe~$`<z9aG(@%X!n!aeVw(qYXp!B|r6-P)1Cc#T*6eP!hH
zYaI7Xi#JnIL#NhpF7wtv<ZwS3332D0eW$#ez-pABha3{cr@)T2g=>_RmTJPokz-C0
zVjg$^y|_LhT5y?06`CMOCTS7-=9=uyjT<9qF;vOpS`jgAc%feO*H?Iatv5i<ASQ?f
zR2D|79hIx1LOS!vi2w{0v75%*)4GoFRL#;jEgAcwBWZq&e(oG1)f}I#BsQo0LZvBJ
z<HTMRK}5Jr1W~r#iJqKuX?6!pU(VWicMI_t_c81n2VF=_zU~##Z9|?aSXNwOe6124
z0i=Lp8+nq?JkN%ZRAKOm>SIVQ1Q(DF*&#K7`RU(tq!LTSe?r?x4%6fCWe`7%R+9xc
z>N8DVzNFz?ny8kR2H?RCG%A6CK+=jtBl_{teUacXEAeqG=jP6j!IM>a>SE6f)kB(~
zIjA|O%2Ne3K523EnZ6lE#z<pFqKvfyvD%+C_BF1AyyIMx2%)#~Wf4_W8zC=Nye?Eq
zu#$%&(&4S#w|8*H%>|Qogdu==?zWhAq&@sg2ST&VX2PmUIRlOpE!VFC)5?K*C1c8S
zSlbB~TE0J_k4{4)WLUN{zWLP=h(!Xj#R2Z#aT^yqT&-uK#GTe<UNoa<7|+;Dd0{EL
zq3WFk=#SiTf&&^G4m$zujzR{B*vM2_j4Hr?cO;XGGkZm$)skNNH`&79z3qP`&;BQo
z4l#Z?5mmmAY@vece@GYpQuk+LV#TGSWBf=LvatM1-Ty<f<Nu$!KjVMa>98@hGqpDO
zWgcK?XK3(8#SZ#kIl(_A2^s&U*zsXF@Lx(CxF3=u|AVEyrJ|Je7A<W1zKRP>!qVq{
z9Dm+`=o5cY%!YFD`Xvg&0*nw3Qh$N>*QW?D<x=cm;&E%EwTFP_=i8RFgBq@d5VxI^
zpBc*%+6!V&?(0I%&Z_DfX3LP{bBPU@GUGBQ_10nEu@*dAel|!kak5@DkVz*^MV4iX
z4R|R^Wo5y%dN+EL2x+Q}3~kDP4=hQsu~{$O++pRxVhF0HLCuZlY2Z<@yzc8G*k1um
zE0A19XGqUIR02(l9F9~{xKp7!0Wgh;E-MS&d=ehdaZ2*pS>1Q7x@}-<m^AAzus*xs
zyvfYFz)S78$v-XK^IP?n+mm)ms^Cy9(&1FAR5gvN$Vz*4)mCSIHSNOKeXXsPoNB!&
z32ms`x6)qlTK8Bm>3>Wuzw*FZEO>DuZkA!g6tUWB^6&|ox?BPTxeNp#z{Hi6>cdX{
z%o}<^M*O)7C<{!K4PXW~1r+hyO251duI)F$77chac>vL*K!B3$31y$|=1;T&Zjzc0
zkUe@;`VKMIY!0K{qMZ8Bfx;;5kiDj(hJvVLqyS;G>3l78x~zp`fU2ltS;E7;HpM_$
zR=v&YIF-%~sFke`jpvnPHuZw&L6Br@Ep~^Oz<S!Irm*84ZG0zG6B8S;qy_BrkEBUf
za-t<E2pRfTd>&S$cWXw$`R5W%H=Ya{VIO~HVmhp)o6e7koou)Cx3UVtp;RTb2FV_V
z)9|vj(QtY#6Wp}P#yB$w)?Csb6h-7;(e9lCI_Pdc$lZg@Q9k1#1V9)5ZeYMrV7Ia0
zOR=N&HY#8r22dCMv(g)>?p#gARKD73Af0vKUWcj|5`*0|TrggTgb(?S7ZS$ZGz5Oy
zMrbeM{Xc1*Y7QiNcrert8W=())>|F<OGD>?kOgNKmYI*2pjB^jnOLbc+uC;~y-$~Y
zU;YmFfA8h-A&KLt=iqAdVc_wr3;1x*;McV`{KXpho1@2{oIh}X^9GUsg&sS5hhMH6
zxNN^9aCHA`gO2W_n@}+|a4@mQrDOUHku-|dN>--&)&_>S^uHkaCyf3o^1EQrKa%yh
zEFT1f|80ieQt1~)5!-&psFUa?C4Yl`fvu|L62;PXe!Z3$Ts}O^r}O!35u3$30Cb+H
zOyiU8l~)&48O2r0pM)MeC$s4-_Sz|8clP#>XJ=Ox>^PDfFZ2b>x2*1NdNiIY<v+X5
zYLz<7(l)zeRV?dy=8X(IWv2GpO}T)dtDae|jpebPTxz&jp0iC!FlQQ#7SxAXYDW_n
z_TlXzywcug98M7sqSEcvNtf7<p=oWv3sB%J*oT*xELXMTtEe1+`D<E=14n*%s(`w~
ze{FkXh1#n!bJmP5uy`^fkr8;`yx{_x-tS=cobIT79`8Bnd-C8sgA%00rc5eh=2#wo
zdaGwcWquu4O}M<XWt42G#+6c1Nt?D+Nb#EPNW<Or7@Dfl04XScPm?g|cVdx8kl3~N
z2q0GjDTbs<qyt~kQ|~e;DyD4=sH$Vl@$KhBhX_O11xE%o^&KJ<AfTQQ$PQA|aIS6<
z1|ITIRD4~%K{@5r&_NT?Cv%0QEQSW=BrNkW_J`OAcG~Eo)1BY1sa1#gE-yMCC)r2G
zjcV-G*2Q|i4Y|FA;Jp*y$aLX<jmw`N@D0>G7OaYp`UY&@=NPT(8kb?v7TW;2+_!>n
zGg^q>5ty^EK61P5a>e!l8r%hLEG>9LJGye}a!6s171)-L_R;29oUak;Rep+TuM14~
z0lK*G6G>hM9xLAWn81t!c))Zq!SpZbMYx?4un9X5T>8=;CZcvFewna5BANaUEv<^_
z4qKcF7tL~0`@#X-EM|N5x%inFNY5)pefz>>nHWN<aud1KYgjz@uJZks8H%oHBtNaM
z^ZMsGO&^(L*l%2?7&1w)9yt#e_QgoJW!%oL^H8{D)E+raPp|Vnc71>5JnY#QOX8M6
zyLTP-A6Fpu%&1ws9b2JUf1SEy-b8BWcAITJKXpIIP`jVl3^7-;_l$^Ig)3bd<l5fm
z0TCWIynqP`EOh-lp8sfF^RK7(e~0Hk3vK-c&c779{sd>nUnV%e!}-_7?`Zx9cK?Xx
zf8hIHwmS5`e0KidM{~b{yVae*v@~mx_0Outcx$5+s_Jo~J>(A(M)Bbi2{AU?euHeC
z;~bXG2BIVl{(?h%z4=m0C#kOb0l2?`qXJY?%yR|w9buKrt1bGd?(A6$uGHeqtaL`;
zTkoO!mx6Im%clZQ<NM5j#gN-amlcd)9nP=Rs+LX-nYXXz_OupH5v>&4r`smU4*Gnl
zB^Att+XeKm1)eN?!;i7~hRu+Y>hCN`6QQS8dBlms`w!qsb&&FCYLxnjP5lkuhGZ3V
ztsrzwZ3ThD{1}muNrsUqp{Ic2!~%sivV!<w>YL8=93#MEAB&2fS`TTKT${S+BM0QK
zP`}D!fxC!){XG5!r3>z~%SXSbV%*TU5t%(dx`HGnSl@$o?8U{$_G}Pspr7RJ2bs0`
z2G~BIKPv<v%pD<uj+n*)!r13|2AzFAqsVQ}ADC*untnZaVg5(pF4E>$gRa}n+Y?xb
z7x?KcaNV7_+WFgYg`>9MTcSEwn|BGmRv1tDdFG=&2t5~A@-o2EqFe$t{Okl!T$4nf
zSmT0O;aKJPyt47ghfv%G(jO+GcN9T+@Vp|q!44gqi&+o*T*)`>bMwX{z&)*IM)!q<
zxLL^Wn&m>qBjmYRLThreL^RttygpzWy3AR8#VnJ$z9$w|={j{~p5btCooCD~!+zyD
zRXrXj?U8Y_vM)yCkx};%SlSo+wd>w>DquX0+B1Xx>^dGSvrzJh)3EmVd$V@ObN;qv
z4~3iOO^N;b^8Ex?{aJP|!hGG{6FN>SfpT+{`@o<#WO&BN2F_}GVesG4O!r4wx&IxU
z|4iWfGn{{80{&+>)6snh)cua<e~g#^i08lBWYPa|?C@VtXEs{e|4&cnWp>!MW0k5P
zfWvMk9zuC)$j^BOpV@9cOI=)4ME31~1se)9z&zeZ)taYS#t4MZ+B9=;U<Nv8dU20=
zbb3cJTnmM2UpDc`Xr?o*Hw0~+RjfMiYhuc*>!0X*YjSxZ@f3RAqsnY`xSHS$Y~HXD
z@r17~H%HBvMaA)ZU=5`Ea&Ntny>T2n4c_;kJaV>u540tYd1g#slOGcq!Pnh_k(@~M
z-g3aa+g8c$5G6yX!hSaD?1dcE^V-DQQ!%3HU64Vc-$I)K&lloxC>%j~b!sI9Vdm+;
zk+Hp}?KlbBl^^xa@833cBMHn5dyFcMnx^dtedbBdPZ+%3yg=^e|Djh8@97~1Jbkj>
z*h;#G1LY#&>c%6OeW*@XYPCMNt-IWB%F!{PN46_Y$&#i7H~3b`bm}s4J;Kx|-CQHa
zqefGrRISL~KMX6Y4nXuR6hQq`6#yo{IST;iw=Q)?fEfiU>q~$+0A?UofK>FX8fA#j
z-_XB;U;>ng^eKix`k@D!yIhbeg;Lxk()ah*@C)~&p)*nXSA{lyOYXNxbi)9{Mn<&g
z=Y;}rTLAvBgWBjCVnhC-iHVH0WL9kWi4)+w)*n00eUGmei17=vIl3&(2hHXidbhFE
zXET4tImRq4g4zP8J+m9BX@`Y*h4R_k_}2G^>3&!>ikX;DAczejrhck^kDbu1*bAtB
zk1MFJ;v?>Oeg<NQk-}4QBMm~5>7`5q@fgY>oZXn>JEYw3VeFwwv`|~X`@GInix9G3
zDI(=W%p<E+7=mIU6>F79y}Rjl-6P35ciT4#<}qntuT{xx7-m$^Pnbp-npyhG7*qPH
zl<9Ga5mbCA)odjlMJe=2GmYanj9i4p9xb$@7NOR0z&@ijwBx6<QC!fdBKW0;UNe=+
zUcSyI=Wm&-_|@)S+ijR)4Q?1UR|<H#7yG3qKF5Yo-CL4rDR>DzmMMj(uJo&HY=}}=
z>u5Pog1_08C4hjcTz&tRg!vpw-C3#%It+6{mmWjFJtr?cLO0f5&Q@^Tw2A6*w9dIr
zG`opWQuE$E_I6MFyq>i9-ZA`k&-}dJdg85Z{0^G(ilcbvJ+b)SZSWpM|IFim@;+hk
z&O$ss72k+<m^fWNz~js**l3)~A*0|XcTk^`^vSx9c3|1hX#uAy?CpT4t)unfshaE!
zG%?{&`R^UhzgOt|?}iV5SpxV=hx2PO;ZGgT$5i5<!-tR8zxOu(RqhXSfnS}=ALjye
ze`hhv>KYr`f2<)`JATYdK29cKYG9A6{@bua^N)1kzmJIobU%y(t&PEHemlj-;>2Gr
z_OWX5uc3;powb3ZzM&m1sg$XNqTyFu8v~<{e^hkTjI`ABj328E0(OSFAHIjUq#xnm
zH%+ar1a%z@aY+T)>1pYi=xAw~=@=N87#JvM=}Bm5Nj}O+SsVOMMHKCHZEOq;aA|~f
zE$j_H>Z>U9m0H-*!a`5i%Ia5@6irO+aX)_lQ8lW6R1KGrnvRZ|7MJw*Dw6%-?QpR(
zGy<o^WuOJ8{pSxZGZPa76Rr{NKVr=E^dB_n-}d5K{VT@uZ(?kJk9{=tugB7{v3`6A
zzkdFCzK_?xW`KW+u`$y#eB23tj<L|weta5#j<GTPB7Ofk#!AQfw_`tC8vkBD2F8zj
z>rclr(|;^R{XO<?j%8u{Yde_f*_b|_m47-P3oSG2U$4tT`$0MXYdMyG6Jw>L|LZka
zS!wD1dJR@qdZxd%mx=z@Ap1{sVPj(aSbqCUjG2M$ukB@HVPO7io!M9!82?s|jfvr}
zbz%Eh;rl~f9PD&Id^PRBIXQ7@6inR=f8A>|^48W4xW8KVt3MF4GP1_~4P?JI1aZ|l
z8R%FT7z7#kS()hg_yzg+1ZY`>1z8061R435nD~VRxN-mIBAU1#lMQ<ZT|0;0?r}CY
PR$2ydA|fFfVetP0cskg;

literal 0
HcmV?d00001

diff --git a/docs/stan-users-guide/img/Figure_DDM.png b/docs/stan-users-guide/img/Figure_DDM.png
new file mode 100644
index 0000000000000000000000000000000000000000..859185e8f65bfff0cc7dd3e07cfcbee6bc1c8155
GIT binary patch
literal 74686
zcmaI8cRbba|37X=CCUonWXsGbBOD=ngvj0wnc16z%(6$e6q3F7EPInZ%if#p?{!qK
zSG|7k&+m`ibZ|Yd@fi2V{c*pZ0rIk9*jGueA|WARONfgoAR(daA|atbFfqU<bM0Bk
zNJz+ZCc?t<rov*vmS&dLidMS%hC&9GdWH&OkMHyFa3dknhw1C+D2OxBx9Z-~(P`~s
zWVmYWpy2QST0zI7slFZZrM@er0h0VyLu2MP?#yQ-G&T|f*Q(}v%v9D@ug=EEmTTn-
z3PBO@Jqk#-EDQZTT^*fHq{65UsjJX6a)>%L18Jugk{%H<8K!|Ha>2meH9Q?I+yk=w
zxKIcTa)KJ#6@zBP0blzr<(H2RHTzftfh@fZPmcVAv{U2zE-md9ZGbxjQ&NFZ5?XBI
zE@b%B$oWZt7ZuwF1%gMr;rq7uudl!EynP$`?Z(qH9^5-guk~QKdYGQL@l*}<Gj)-0
z_&m<z&JQtgUHA=D)C(n_w{dq-y*&j`KiBfKrHLBK>)(!jO?(@38(EZ6(v;FV1ndYB
zx8ZXMBN-W_d*FLaBxFAmBsB0H*i0l6@DFTo3d-+K&~;N#e}9hxLHzJEfh8OX35p~k
zBBbPqyf%5|Exzbj{kDTqbpG6RFGwsp0d*i6@=bKq+gR3mH$K81Z$CPdhH*D#-@Xt3
zcHKY4i~bI#B;idotXtHJ*BduF_l7Ok?fCdeN7#z^_;&LaiPI~qDl3nNy$d&os^<&V
zDDly7p-9O8;|u&FrsMm5#zaa_Bos`@|M+qn*(=`__y789`m4A;Zzx$bss6`bB1S{g
zJ-YtCMn!z~6d!^~DymWW<bO^K84a^>hxmU_8W}B>9*X-Qu-x{||1oP^pLve|V@jAJ
zPdy=K_%D}9{?~qh&mfo`u>WI9xc;d~LL>Cw`?>!2W<j91(LMjilpwgGU@M<~Ni+Q)
z$AFj$1PkToPrhr;)FVqNXXhKNtA%yNGmlm%&ebD_y%!T@lKLQm6;@~{dhgk-Ow|FZ
z77Et>6u8*RFp-4h&KEwML%+Wtj%$c`Bu2UmVqkb9(T_6+soE13dJdRcp=$+A5MkOC
z^<Cau*xl(L{1dDL#dvME#4}{IOa>*tEK;$>t4NLgeXs#cGX-nmtw<Ami1kBVF_)x=
z52tb3Vae$0^`^)CPU|1|7L!)P0trObinqi+`z{x0WT_mAzq(49CK=$k=J-M_uC>9N
z(&fP4Y}}~L7n{;_=))DO7^Z@?W9=tnDu)~7%ah~!hwD+5c37?FX~NilpAm!+tVcC7
z9|!Sq%5zLKl<3!FrfnUp89mN}Lc_Vt{dvj5XcAR+14w2K-XB~J)z~eQXarAn+KNr0
zI}=S~>vOd7%+*6QtOLX_u5ZQRV^oHC{k>(L(2wAV<*`>Ty`kxG{bt!*%^{Q1>9%KV
zT3iBCe_fqeZe=@TM)F1htc>u{Rvezs!rQYAJ}Fvh%=Q><wL%vM=T#QXN7hXQTEyJF
zRP*hk4%d0!gy{dry$Wc636~~ac`t-8Hg4Z~6>aqS+$xlV$Lf@Wp8_WA<5hfb>`M-Q
z62;q-mMSu6$CnZ~Wuu>MR<&LV9YKB`*1@vdF7wdj#%L<>@cr97um4_$ygzWhYd7Pi
z?jrnPPFv^9x#PwQjJef3W(I97gCACT6Yf<H(~T0y9~rl`P^j-TlL|%DLO1POaYHn0
z!&JHjr>V49kw;HGP`L5_ItBV#Fmljlk^UUQ1{yJq<!wgtJyjn`dB@4|&qu^@>GP#j
zSiPgMtRV<3n=YMey&0A+Bf3#L4)ZFck!n5|!CUZddh>U{c|{GyEe%bnxPEC@Q|Je4
z1#6G;kckDS!jJbwpS;<T=t-KYc2M$xA?&6#OlY@%qM(H$Ss>`Ja<y8alCxoi(v?%=
zZ<nU$1lIEEN!P2(=(R(+wc8VY#^qj@txjisG*8awzg#qci$9$*|4ul%`rKJQ$(@&8
zz106}F0=Ktn^whvS{@t61J>m0V}H-JNiM}vhVGtw?d#i!eSuJ*B)gMDb|;<{+GUMs
zFCMln^+`1GB4wc&b~w*<^SKoo$%I#Ld8sf-%`L=x_h-jyH^0V`WHa7O`A=m0k_IfV
z+%<awAMAp>D|(XiJqRiW+6PI3uviW^dtR>}$J$Zpkn(-y<ZP`(YN{uW^YzG@)r^oF
z^%KLLIh}1ctJin_Iw~kc9l8Iefz;>APvMP(y-j6FZV%{l)B;@3?83ihD!LBf%neyg
z3v_51Gtu`5Qbd0C6YUllr|Nk$8-XYH`og6r=jQrf(I5Vu1)Np6pipxNFsv_dPeYG~
z9!tF`ZLwoH@nW>nTN1CX1`zm66t{(+Y~qY$5;1L$l)2Bo(Q=x*na5AXU2y07%xov~
z-)oNr(cXws_sZq*>s~cRI%p>KMwlcB2UN4oM<R8F$sY~BJ(jYl<JZa1%a@k)Z6V)n
z-ye2qT9wg=5IJMAn*P(ja0OAeql{VIldd@+{G0l9eWJrTf#=!xpf5M-SA;DGnyEO9
zsM|cwXN!n-Z*d5hySdbLCQ8+g_#8<!{AHWa2`rG}s>Ir{5suU(HnZJ=N=x+OaATQ6
zM=DCMG-G!UM3Q#PI-0hcL+ANnZBbc1X2`<{+)#cOrVOmjKWBv)K*ZA)t!Yl^zBIyg
zrqG^*+-N6IL0gMWrX0wilQ`Nz(<yRkW>kJ|z1xKx%R3lO!u3U~x|7cP%C5`TwdxVX
zp8vabU*dqpJa?gdeFYq_I1-$g!>d0_1^Qx`FqVT(;>8d(Qha>N`8PkIirsEcvvJze
zWHBw5E;?*dy%<uyv46)wd?(5Pj+K7D;sJO-s3*$mkZLNAtfHc*QIBjmAu}fB!n}UI
zLmtg%zbgh`={vOQrRT!Ls;yhdw6>3|I)kXpZ~mF9;f)}<tG^x7xQ3XR6<qM5cx@ux
zUA0hTE-TNqKLbZQvu9IGZ=o|q!$A|OoQ-QTK15|%w;mDAjP+$<IAZs2lQX`$HDJkU
zw@iw#V~BTs?>pq9d7>~eiv3%jU66>wH@TWMrH0|ByUE8E5>@u@PBU;%{?Bh;Rqf=q
z5&yAYAYR#`zr`Ze)OHp_MBN|B^?aLX^D9}X{bhM#wLco^$Z?7uQ0;$vUffz}ezvpQ
zLrIrNz0=Jnz(?P)sY*33CHVJfLg<){gi4{imx~W($ak$U^e~)a9n1<Mqn8eEkHV@a
z$72q`K6RG@&P<OZcqbAHU8^q8a$35Z7s(W0N6ugp{MU(!XaI2XMPeEEa-~YpI`ZiP
zeP{9!Vh&e(+2=;azQ-b!wYHfwDw+hLN4v}%jP`rR$;B3LoIWHjUuMOBkV|V2l3>)l
zXi(|pjGYJ_w;EE8;!e*FU9jpGR9QFa5*$;-jCu~*W<y*I4=uu`YEXA(ypbiOs8r0R
zxOh+z6ZmH^|JO(yx>U2`te1nSVs^$8Mam|k*xy45>5QAlOAg6vhBe@qXiCLk7`BFb
zGHK8sm^}~cdmphjTJXsGBYfcgpJodQeFKIog6lrK^!Em|VDgf=c1G0xocNX-=Qll>
z-x{5F3v7$zbtc9e^!V(zhtue>4$YVCHe*oVHc&P@7vVwJ)xXngO$L5&TJe$)F<$+B
zwDbMEl+%NFcd{v}uDJUrJGV8S@+Cyh6%Z}$v?cIJ^nVO%<v&E5-y1+JbmoGOe0=wx
z`Qg$+dA-<hA6y>mRhF&Qyy#++*_dc%{sU=gqXt3^PImq~JbG<shgaq+7QMm~n0D5R
z1S-y*-6=?kHJvXr`aiQPB!XbZ*I<iX&Ufdoj%u{+I!QiEOO4&bJv=ceQZ@fb=k*E$
zhY{L*AR|hlMPe(@qT^gnnX5t;VGqNnCELF&1dF+TFUH6Hbm9?ME3OmVVY+0lg@XC$
zWV!}ThpLA(j9gA|k}-&guI-tMlfie94z27CR{a(Xs{V(Ge>*&;0ih19nGxQ><;Qc1
zI@1DDah%&y3gNAR1QH6hkBS&~B~*<w-cV3PuI65A4JM-$_H~>~i*Z$WhtiYNOO8bJ
z_#XoS@X`kt4EOkvk;N@=HuQJke&ko<bfy&Ljxg&9Z#cp805(%`KjV@>AdB~6>h|Q?
zb*>luMS9n+OI`WftNej=2(x{>OuGQNG+mXF6&Ph*A9Gr>MHAR|d@B>{(Ur<$^YvrO
z#7i;g4GKcy2c3)|0xMDBG)}>yyPW0s{<O9)5x_Vco#Q61fluXqP~1+A=d<#dT!dJX
zZI|kam;<+-;xB=)sSZE^axA|(tSPYi!abZ!gAP|6q9prgc>?z#n#q{+xxhq1fGT_q
zXkG33d`$H9J4<(w)UCQH*WW#*3h7K(<|xwfIFb1PlFFl9X(&_2z5bUkU;b&Z&=l@#
zY<Q*Ig8a>h)BWOKpP6wyAEV8rTKh<1Wf(SO@cL0S3eIdsGSkuc^yz^O!E)bcs?FNf
z2<LsrT6Etu)IUcP=SDVaq@AEz^aTbBZ@ubz?j*m^RWz3#Ukz+RN`0=T#C$dE)wPJZ
zX!UA#H|OJ2&CR-=R87%yo!fs!AACMA5Mh;AvnOJpk(-zp%sk7~b!HRv2oR%`)y&j3
z;=~(9)>bx+G5;wGrO>IUH7&+@u4LF%A?<1pUt#;6@1Fy1->a7%ada!3AVO?CCJKeC
zPQU_o&Fl?U<dMf)EbUSWcF`))OcLSeXZ$<M8PPS*W3*}IjFiou3gG4bRogX*dB)~9
zEvnvE01JntqoE*6DJYy4n#Qz_l}IQQP!>_@uON%QWTSA!Emp)wGuYI+Z$kg%^_FBT
z`=bl`@PWSyHhdezGIy`EZXuSLddu;`P0Ia36M!|~LqRKd@%!wq;2gAnZ<4TXCMI{9
zk4h=FSdXc&i6u}M3Gwdu6AnV}Bdq?uqxkik!0P?EkQ33zt1GnaH}F-<SS0Q#=@1%Z
z^S^Jr9b%}u^F2uDs725#Jet;evqsBl201U1zG3kl+i3;HpZ1DRftXz~tVtX(ySF!g
z%;8BX=iI7Te*kvvt<$|%g18|q`W_grg;u)Z{G>xvAp#_kigNoM{b|bIaC{m6oE{R&
zeI{@Ox81p~BT6ZMQRKw4kxx<6Nyv;lEff+873@Xi`rAAN2X9Z^JJ56~T$tzul3VM}
z=c9B#SAXB%Iig>KHSGTpX`zq<G>|}DdVFpo@X&qct^MI>CRIvy+mA6wX3~P0jbs`M
z;bP>@t#n1OlMlMWVc$M`TOThYtJcJ66udg2E+_eOtzS}stLtxPve6=r)fc_ee!t`N
zD7p<@J&UmDj{c9Q1f#vdcM9fePrTUe&E(Hd6{V9#;w^Rp(^>z6o&)H?Iz?PpCOE+O
zp>l8@8?T+uIMEvJ6=In63wZaDWf?WcE@3T>z?MXXHD%evtw@fZ>m$pItpxE%A<60@
ze+Dxq1R#YsvLM_Ho{bJl1LGbee7@m~#JvbFkYRWoF6|?=NGC(4Xl^dz41m=Sble!E
ze&Epl$xV<O=;C6yemGI;h)9`wH{U*oMkek@njA%l>ye=XCyP+>OC~nE7xK9kPhg(|
z_g3b^p1yZg$Rl6I`u-!##`Mpyh)Y>7*LU}}(&hMtM2_1{DQGB^4pVJ0^o1(l#qCk}
zU7Xj^Cm&fKP6|1l?WZKzhHC={02n!f@c|#l!9SI}Dwc-hw^!_j?UC6x3@dsTcO|?r
zus?_f5%yO*+3c>2m1KSW_>`*SU`+%vhLV|?xxg_|YObECT4YqrO2l!-f%m|O932&r
z<MdPzL`AURl;y>&j6q-OtpYyzC(2A{k6}#*OJM?xrMkjKO@u*yEs$TElF|x_niYnl
z<&0#5i}N#fhYgL<BAA@-{SuRb=5q{nylv+i>%FC?Jl1pm*_u_#(X84m(NRh7>y$DT
z*d4cYoR;4CeXy9UB<_Zx!i}$HtCdJ%+_*!3jff@k+h@Oy2sr`bu8t<9X*6B^m6vR{
z5jOPzUCW&Z;&I_3Kjqv(r=r~@Ju>1d*<`A81xL-ghhNBPe=vobF6hyD`SAH`r*#TK
z%x1zox;EqAGp0wkUmLk=xAHZtxR{gUwJ$bl`)1vRMc!=a(X>*P{d%}%-Pw>9=RV<R
zjZ2w{cBTRipUhYKiXY#Df_5I1+uylnJmayu&{e3@fO53-KE&Yk$Mymds&mJ&s_SxE
z%=aG~98)BM60kj;OGE>ahF)awSR2THQYNTaE9|^@d(E_)7Nz1Ay-K&fN3QFEm!?3b
zruy8$xOabr4K$irDmgJMOV-%uchsdWhwiTors*{W=6bo8@MR%oKlu3K!rQ$!MabyD
zOj0tC2evs?W2F4OGnQjg;3EW=0q%ZwJc`E>zHa%9oRvt!3jUgmPnnZID3#DFKj3(-
z!V<BIF(`@|Y6mJYBDpm~jA>GMK3z1VUDgb}Iq@oltg_)%8$0d5Z4pt@kK)G21vl00
zc&W#6m(&Bm++Np=$B;7&h&=6rS6ybC-^|a32--Uwow--+^-FQ2aaobd3qg`m;&%7L
zrC7rFHp+zsE>3OwMo0La_pDqxx)uk`?_41uTnO~`YrQNS`|uq>P(CAg8AfV(x=P{D
zS559lmjtygY|;AAL5GUrxTfP#-z`HJDJOu{9&dU4L6J_tA>c4zCivH}0?TGrFU5Cb
z)~X)ScrjjTzB7Gs9#f3g=PQ?gPT@EMzq2{g_B28NJvVQw+=G{UY!w8EZRkgdf1tVp
zGXCE3Y}oOBEIO%)_HhS&Z*JmnH$jxa&gWcJs{S-7s6{1%4l8d%_Q-1mvtOHlgn~|F
zoRYQS8%fMz<YV?!6bUYV4OEuJ^k<0QcuBoAq%KrYmq;z_w^J9y?VNqzjZs2^yy8$R
zS>kDLO;roiL{#|qm>z*^&3om<cYX&^`3V4?`Qvn3<DxyikWl*YW-?ZmM*9&#P(B;o
zOQ=JOfs1~BB_I|{J=@Z^frt-IEA!|;<$L&CqJyiw`}2a`Nc6c0o3la-vLd~p$<GG$
zzhVw<1PY@{ZYNvrP>womq`S*E{r>X;y;p1FWvmaoI7<3nHMpOwKSUV4i?nyc>QIhN
zS<t8O=xC!2nU>q8G`B3!-RujD6z((&<CbziqG|RA7K#tAe>nZTP}jSaYuIr;A^!s<
zy_Z-$jX9h#Z*T0{XPPzw6-NMBr2u!>fhU)5MMQA}5q}ZkGzEeZcVYKt!Q)uFZ*cVr
z4*8nd!VdxWQ%7QMtCYmt?AouNo&iLl@ZrVaC+me+So>;T9m_Wkv(J3-Y`R{yt3xs`
zhO!x=7*$Fy+=d$y^4$1$TB*`0WTLE=dhgn-z)`ONY-AB8<c-m$3QMG^9Es36iBy!m
z-!h!5X_`^wtie*f^*B?#EEjMlEllUO!<r6l6dtEH-H+~+4FY$xEE}d-q9(Yf#CDjj
z|FvM{yP+U3;nKxvkBhUNvbwXbr22lbu`)|mY|;mjc(U9L=7KN3i`bmbMaV@55->`>
zfAVIou!AKculBgzJVKEs&q8(oSuZT+S@K$kw`U?+3^-hHGEiFhHCE}O>M14D&%Qe8
z`fJb>dbD&ooabHf8DomNiaC*yN~SAcp--=SoBo3X?$gM>e}aAXDb#7S;j?~lSCZWe
zE{ll~{tS%@>!EDbqF|>aGW?R<stw-5>W#@N^<>wbfc<U23EFS<O?osF!=BsEa4LET
zon+agVq9liUxQ?ES<i{DworH!uQfGf;|TnyuiR;8oD|q|w@pwhH5(1rt+UCq4;>)k
zFnay0*461`kuWc9wof4Lq2)%YoTOx-MeUIhBNrFffMcUW_5P50w3_h+8J~UenrHrg
zS`E}!Yqol~yXX;l2#?Vf@0`zmIDW|%qj0su%I@3i(A@pSWDm6%K;nX*X6KMvls6Qj
zRt~*vAzLh^5O8%&KTjY#va{(PKJ*f(>aD%1wmDjq;WBM+Rg*GFkhPok{7rPj<>ufA
z)|Z+KWr+kRWyvKMKdvAr5`vV0lqa+rKzFJf%f8$&0!JOlenWaBDr9kH{F_|grVAa1
zWMkaWcKn9kkyNsR(XgiaaY#<^%vKs7`}-&d_v9|Q%UXf-;%gGFcK)d}sTk>&X~$V?
zn`)3lhZ?-EaF7j?<yG$lj{D*H3_|9D$@sG8J7Qp%2fT1u7v-E1=lzxIXD0HaNewFp
zgYU&~@ccV_=?(N-L+@IQXCyhd_nZ0>k=5b(4?@=6UG_(MQ)e|xUPSF5)?G}lOL^xL
zX}cCYD>RT;ZKF#rUQ@cZ!7w(2b^HQXwXEnK*q7tLFGYwid_p7Q^!-06w+YNgXT{U!
ze!`#jM`4|uujBnlU=IO86q16Ot}Yg9R4AQLl(@V23NLstwTznE@cl+6z#w}eW|<0U
zoo0%OFF&y+z8tNvfo0jRk69o+^^P$gD=y$HcbIDplRh&n5r)0s6_VpyheiuY^2@;*
zc-PnRreRAf#RIY&E$2r)9!4V+weR!?1QH--OWP&`=@n<D@l!P}>Tu|gU=fo>d2c`D
zBS#W&ZGmFOvfLu8#0SG#E;&_#S?xtV0;{Dp-vb#7U+9fiC{I3+t)qJK8jv^uW=xhk
zOgX`#&9ANy2EbzG>vt+R!o2h6d`d2!$X|vR_N&#qvhtsv;qhF!S~nO)>M|;$tqktA
zE+*5peUP|JxJfQv*8LLhE_}Bcg$?ffv-)?}sOji}7Ub=;4w%Fv-VY|BXHEOqx9DFn
zmI;2wXS}-6k*k++4VUl761n^N(PC55%@qqj<&By{Er;D$SklW?wLbLfV(DZ7&9(Z>
zOxS+Q(PDD)phZ!h?Q;Ld{p21x3iluHYTQMK-Z5($ucgOZSdWhj!D8?0`&Q);xcKHw
z*pHiU2)GbtsrP(hw_TEFP%q8q*_(3NY>Q-2Wak>tD10YsGctxe)T?Q6_8~&ndbJX&
zs+<O~2y?bqeG}y7MAUj55qId$eM$E~WYGwjFx&gn^ymT(CELPPFJ|w_T11Weqr#_Z
zYJf$BpPo!>jaIA9CD@=?PdT}(R_x?auz#O<P=WqCP(xHuDq9DKwLKEQae17(0&B=X
z0E(5tEZA|yVv@_3?!>sPS6OTL$a)@crV=uOBoZCJ2aE<csWM5CGCj|S2=gtD*?&Uf
zIaw<*>`vez4(21RG>+kaue_#_*PY8W&Gq9`kkPynXFUmtvHN<-FgvHmb`#N?HydAH
z6@%Cl`_1u(F9zK1C0j=I@k*8~*SG`rJs1)F5PD@oDttFvQfPv5f31*&=wt`**bg?1
z65t@RekrMU*-X2VXZy_gDoT?O)I-&(hhM<OUNpSsd$!n`Wo@i#!DnA|gJ}VyW<oz$
z>+Rn-q7sGIbT2iCCA`9BF|funvSOt94XF&YPmB9rOVGvXZqLScZ*2TVcc>)S$*OI)
zbg$dVK|rEHJ^xsS97UOc-HK8{osng8s!+QhxwE+Kbl-cirtaw%D+j4+7AL35Ae+$~
z=oxM-1Ju;Pc1;EKtL!oW=!#w2ligQV$;*f~!u5|zUJSOIQ)ppfcaJ$4f5wSl_bL}%
z88zf#vtLm4OwQkxZpK2Sj?=3jMj0Lm9@Ep)B{TI`auw;!9!j(<CJyGVhuzl?HyC0{
z1|(P5%UQ4R@LQo_RHz4Q*hbCGi;9c6zmhP76@?_yb+1osQc*cqgCW`dIJ1gOtFIVY
zDw^q~LkNZ7RA>6&XzXas8hX~m0Waz>1J0u@PoM6{!L#u)%NUT`B+1=>=oZ*?#f^Tv
z0REn!NVzjr{!+h;g6~SM*;sMgWa(~mr5&B)+D2>otuYp6W*GoKVq`k@Cf^PBWa=7M
z`uEx&*#*rNT_eNUKfQ*3I(+r0?g8vFntyrg*$tT}k=ixg;j6su!g-wz2ZfB!+3>(D
z#PvY6x4VuYQr!Zpy@U12d}fBLeUiSwxKqjphF?hP%RA`O@bdnFz{wg1F}s0yw8JOl
zj^5a$IN1J92WwBE`48EyL=qCuuX#JB_q-5MxSCVW+;P-wF@}WGI25b1_Tv2Ho8NAU
zTO$^o0k_<y1Zt~#nFSY*Zld?QBcnt&)f&^O$=*U(Pf~A}^nUMXZtRT(Ju&x5%A+;w
z`_gf9j{)(^5^9Xd^VEU$C!1H+T}3Iqui$-@jD(Tr9cBiJYB*>1bmOc>!IH4rBjhmB
zET=tfR|z9;QdRfY@+WUQx*c5dEC_09jwyu!^|2Ia&-R|F<*Ajt@YSqr{YvIA5vHrH
z^?=j!XvQ0-J*RBy1xg`FWa2~1Bx5xs8SR-!L@NJ4Q@#*qnxJ{)KAl!1{aDgmb8wJ_
zMlup}l#$r0i};3*`I6aKuR@_gTW{CLue-B^%P^uSQmK5|BP{9$a_+Jq@O%The9L6R
zi{{b7i-Oq#m3TS=A$!?1RxOmaHu8!K_vF)am+%QIZx#2`zHmi3>{d@?ch!<WeDtPO
z6Lz&<p~T+@>X<tyc>Y7U5jCjW&h~++aqk1pmaGBVvP}~C!I@8ln)0g?Ha$@-<Sx>+
zCmVbbeD;k#Sj2>9fnTMc%T&xnl5oB7=ipHGj+}3gvbiB)OoHwJBaG^`C-LMT($iA4
z4a?BHG^UOLPm_ckz!I97a3gk)I=X%_S^Yq;_APq+b2<K(8>L1vA=L+QJ>4eJ)Mv*|
z0JJfg^fj^y`(YUD8hI52>`KhY{)z}tRKIscP%ffn5b=s$^=ahDs^;s!fy451aSeXp
z!R)g6WP+mw3UH>LfxNX<6iu+e_napEUyA#P3*&~1A)-AOfC!vm;k8|w87cwQ1d#(@
z*<pH-Lt-lOh;fe(v%mU>q(qy#Y9eW1F%75Oh^B}?DI7=ex@4w;US3=~(@43~#|d#-
zcqG1NKBL=qu52PF8)>mk)REzjEzd-pSRFH3Hv3;>FjV6zD6VNfyXB~r-1}UL@pRgG
zm3g$vAy30)y=0?ce|0#14|yZl?4HB4TUjE>ih^pDea`q})I!0ty$G|3^1ewTt<2{I
zsqx7}n}>cB-T65!-VHjwZr?~j>eBUH)pXr%B98N!8ZR{LFj($SBM$bnTzajWSg{(p
zb)Ik*=E~}?zH~+~H!8a0eyf5N)SRb#6%N1zW;{^FOI6v;;|x7$TB!&aZ~4r7-@HJ{
zNL+~8tdk25z;g=^r?gKRN#Qh}iMc|KIU;C=`})m=9FhZ~)Ih{LOrP!t*g1=~e&2*f
zwr1XUUmZl6!Vx@$@Vguo+pHeiRaF6(wYl!jqFt+@<up%VGL#+5X*SZ!FULs2=?20a
z$J(`NZmSv3(I~wOs;<S|R}}VT`&uo*WUCf_D*gTabJ`cjF;(iD-`W}TY|Bcl`$UP_
z3}ZAGHeB`xRVJO4qd+w8cqi=F`R-tSyhGDrGAdK`2M6Z#4gR|s`tR0vRtei+M7&8}
zUxiT@D{3yhG+3j%X)sBp<h%<Irz1+ewVVM^w*V#Tm7Tg*l+I`xrT%Egd)s3i?BR0*
za&8J`xg!_yayZQbJL2_Pcf54WZt@edAp8&U=N0%PMX-?AoE6ULd|mH5rWl!}>t8b?
zYyzE<`5Y{K*8%eqzYZ$#sp{42k>mZ<_{wfHGBzF0C9~13)FVSsW1011R-YfQe)H=p
zJ=h0e(!gnLror1n=VwHeS(3&61d712cU97wsA*@%`$fY@;>q9Cvy_>Mxn3xx+gH9_
z<=d>Lnyhko*zzR!4g=VHEAXlC_@wQC%))bE4Wg)p-;V6|Qv|$wZ`CGF7Oan#`H{CP
z_2j8jCu^n)&A;^ProJKLc9Z<HN8h0-x1uGOcNgzTfs5I7MDN`{8#<ipu3Fq0zCIoy
zLd2d53H3*5NTH!{GYG)Yu{-mXj1sY^n`A7TTqs#Rtr7pygLe2+CIMx-S`D}d+MY}n
z+%%47Wr&jH?HdxVTZw6^pkCpbea*2JK#0n5XBZn>77##vO#W%(al^LX8ebXd0%poy
zB1<_(1oh3BENYZmsrk6kPHA_t;Ivtp$-tB2!%a)0_7`kB0H;Q0B)hBKpb%i{;Xh0u
z&F6fxE_hGrBcz*2Wqmi8ZzFWKn@^5YE(^mpLnfZ~aKa{}CS_&Q$jsUD8;9;<XDn4l
zqJ7*z8mKwN>^CMDKKo*GTyrx?aXj8zUO1KwR!PJ@ZYH%fh>Kda*HhtKe;h*2zm^<8
z7&{r4e2`$%eLu)zdH4rc@5x5hW}t7cTA778O5qE9HBNMIlCHP+g*k7p<7-}*Zb=F9
ztU2B3DAa4Z>IATqA+YeBu7}gDz<d+at=kzgMluUYr1G?C=8jkM9-dHrQ|^T=-e3zM
z&@?mu{v}`slw)zPdiIXYCo8#xS$Wmf1f^EqCs$+vos26jJ7-`FHg==JC5E`AqBGl^
zGQa&n%^M?)h6TkT80zmiO;HTm;Ll~^)5q#Q%9h+y5-&MFlB+Ax7i0TC@a#9ROfg2w
zE&qPC<A$?ZTU*;fo#MUc&$5)cHtq9!zOK(?8D1%3;`1~?l=Y)G`fsGGb;Um%4scp9
zywbpQ!gvx*Hd4Iysw%rJf)2(zd>&Y5c!i3YZ_v{v$);8-4;*OpLHx5!rOa@{p6rsf
z>ilRj*AfqmZG)Bf(y{DA`XT&Aqe7$EmHBJLl^Z;~Pxz3`_ctcTCl_Cfo>+_(XBzgP
zbQLCN+ppBroE?luli_!MO+GNa;`k<`wByqVveK^fp3@zNiI>d<pfXYhfGrz@!?=?Y
z*pk{owmw*s`f<WuY-U;Q+Gu4oNz}b(?S-gcH)tror6+>WSmE8K-*o;8UXk1$#Tf7M
z45vn=#v>zFc`!@m;bHDchIDLrOBR0eLMJF`XrF?OhKmq_4CBc^&T7o6JQs0|1wtl7
z=|Nb2_!<T7Nkidb)}|kV{zWT8s6Fy##RhL4|3IT@28)i7*%Ri{Usx}N`>OGZSC7Dn
zG_6dWfsyeToR?$37SM=*cU&~HY`ftSJDS9sr3edLS(8b0%PJx+Vgw+Sqe$GY^b=ts
z^CxYw>0*<Jui2^t-2=&bMp;Jt1RgeCXJSjz&-pM{(?xpo%2`zlzjU&cc;_wf@_KU$
zif+m%Vj5ehlj#FOx>LGq>rv&=GN0h0NF(!J<&l&9yG6yGIz*#){3R_2akxY%nw38c
zx4-Wxp#e(rd0>!VyNrku^x_?_MFLHTAbIg_i<;hr%d}Q~1C_({dQn|r%XF;2-T?70
z$Oj4ahT02_+>!l;i*lc;J~HF26aIBH5Lf73;Xrm5%hge6bKhz0ln*W9>IIfn(|1S;
zPX|k7E8{gn>d~Wm*Zz_$u_>3H@CdnsaSw-6`}0Dbm*v=(OxLgE3Su}ADwbkyS8#%7
z%6~;k_8>2g=m^rV#=796UjeO^AU|lHmR2k;m&8KPVH@bBtu^5@xikcMo)MIA5_8ZR
zd(g-#`XkO1!Git#0)LHi`Zcl7n;Copd56{<Tq2#+F;x?b9xsoR6wjOwt+LnR(>4pQ
z(X$E%GO}TJB4T!G$o}fde3%%@L3=rx{<YWA|Fm48Z|iH_PA2%{>JGNPecpILfTpME
znUG&_?sJ<0vCA~@L_X%sXC0QT&GQ8j+jOGK_iTfpdYEhDu(pp+g`AG&Q-GjKq40x$
zP136g$M<r)_HzzX0ox2SEWe8+5)`rEJ7GP77a9kf)2x;?`_YM%FW@e*QY;lB-8_qT
zGL>`U5<Je2Hr(%Ff{qZ7w7A}TjfFS}G<ez6)%KEMj5%){*8@snw1BJd1Sx7yC`6|X
zXzh0TrG$=l<)n;y2wm*7!rlqH*~<^<=iTni-bMRy+bZZ+osKI62jr4?R~*+m=ZjH~
zfsZd|n-3mFtmOe`@z(_+vHPc1w-J3QTp46d`K577=FtKwiF?Y_Z1^&!$FkD!Y4>X&
z#Xo0vw(U)-xT{cG-8$2tg>ali;pjS>fTPTr+}~>zM5#O73Hov7E+H#hTK7^(!S>r%
z9X5Kg5OOXCEZZtj{v9pH+2H=EO#_WqZpYrjy29BQi^)gqPVAYOdW&sr;g0y+wyohf
zg2i+D!`?dsa?2dtw)Sg)I_+ZG1JaV#=2@G?q~AsXb_<hW<|N}C$yYEHINHs(k46hU
zOqpmY%)BPp%L1TbSdq-D_lvZNn8d>otzgg`E;M@6%3OvvUp{-vuH&gqD|<aKo&B%A
z8r1bF8{PwfWJZTq09DM<vl;knPkBs-W0Qt|kR>zvHt*jggq8|PInoonm<V^6!fFRX
z%QHLP0-MBTJuNNJyRPmpW%giwUoRW;>o5=i8$T8tGUfZlMg&gl!(1|}BH=pi&wnSj
z&;u%kzDMY}f%V!btHbICtfF4*M$+=pta8&JoH#?3y$ieIZib#o1^wAaY*GYJc-z(D
zH9`Q`T2_BJeI?E1(DTubzsmXMh3nCVAKpzUWC!IMA=<n*4%5+(*%XlRr#!NhDbG$^
z=EJ3``->TnM_aQ7sDKq9W{CLG2ZS4>Zu+^1|9^?B*K|Oiiv=1+<EP=pFGQhq>Et)O
zg#bzgfMeRj5@^ZZn_TlMw1FMNBjgAxe(2?CIKER+E1lb9r9gC|emihr5Lo;7oe#I^
z5C(S_9vR8W<NDm-?D*Erl`sq7(Qzp5b%s05?gz0n+O9Vv*T)C)hVtKuPuFTK5BE@M
zRytBM1f-a8CZoFj)6xU61s9kxdf%)Dcv8w+tZX&IQEk{kNAEiXuvNH;U)0iKbu*u6
zt8mVbf{FNHclh?KccHyGuVk7eB`kS8%%*2wa&wA2Q!CMviGt{l>FP;<t!3iJZblTQ
zBC0b-i>N^KC3JLbnw(%0{IFf+JrEcBm)3{VL`%QLjm)bpG?;qSaic%baqWj`l|6aI
z3`*F5+@WMkaA)|N1Dh0&KgJFQ64-<1kMAI~Sl`7{R9Ndyf4rxuvZs--X6KmyzSL*U
zhzVpT(XP91Xd-CQ=YeCIOikz;bw#tX9_YD}FOMo}Yw>K0V1L~?Ob)r!j{Z8jR|qj1
zTS)~%5CkM2px38R@=4oye`lm<Aw1BE8le<TWlk}<=6<e<S{Utmm`GCuQwQw}_raCH
z0q?@eN~ccg(Fj>edeZepp^JZQ0`s@BoZ8=WA*6c7%k%Y}<*0ZIO5kRX>@Dz+sjQYl
zX@I^OG?da=chZo@pb)YQpx~5E>?b;0nHbeqAdY0|l2#Sq{bNuO1+dA~T+4_apuaAh
z)B>}W)Dq3GD~{Tv*Aq$Vy5K!uU+7o#gH1W@8Pn$1doycZ=+Ph2)3r;199&U$`>jS{
zM$$^y!8(8c2dpB)_viBEzcvCq1ge4PGp)9CbRil|9#{!hkK+=rC>tDE^EEz3v~eG!
zffl<6YC#-_nj%qWtbSXmxd+!rC5aC}DeY8ZGJDgF%R*Y`fc5T+Kc+m#1lTheMKFST
zd3iVAEY^^<c{XO;^03{eCoB-cLVoWd92$u}lv89dRlPTSdbG@AgmdS|YU+G@K(X0!
zm~zge0~cq&Nj?H<B98S7h15TkBhX4D#-c3=gwNc9&cs`KEwH#8(sG9*g~Y3iKqyi6
zS6xK3&rUKNGJ@Q4G#G#o!WnAcM=7<(Wf~pz&Dm}>@Jxy##MpgQt>jUEb%nTOV2jr6
z(2H{ePrZM2Z6I~N#fp~?iH`@pI1vKw6SF$wjTU~Fa36GsTl<RjZdABraAtKT)#{OO
z8mWG5shB{ZDKlIQCpbDksmoO@`6ND&PJTjTAFL_^n+bdJ=8L=22Wvx=Vh@R@!oR&8
z*Ix`c=2Q1i9WksifEAs*;occ)w>l(IWYqP+a=I2l+9Jy5QI&LQ;>@S-pG1m1J_G`W
zk0^1g1=fajwyO(ambKrm;RYP@%+%hbmqzX=oC^_TQZ1F<l~-RV(R($*VZuCG$+aua
zxoGB%aV(kare0SoUgJ#8^9?OzsZ-6U&**_Z&Oc!dx_t{owWh^=z?`&AI!rUeq4nqY
zv}~3KGD1!Pya&1fd_?)vxte@&nh{wP@hL~ECJt(bpNy_a721k2?M@n5)N`YiqsMW+
zo3HIkd}}iL1{wT0F1J;E__PPniQ(wQxr9fKDpes6JCHu`|8W*8mKk)F+@reokwQlB
z2x|yJU_J2?>>Ta7P!2$7@qxr8vHK(5|ClW9HJ~tmzl9GvoA(|9nyC74$|e0!_p@&s
zp4_9mG?E|k#05YDs$xD`on6D}{Pd_lSF`H+$q2$;DDPSgYm#_kkJX_`vOAGb_0hRq
z#&TRtG^Ww=r=3h=g6FuZbprfsRl43Aa)PBU4hJfwbsE%Mrh#Wc<{@XB_re4(@H4e0
zsRlo#^`y&q5M?Sf>DPTZkM<n75TLS(uc7}b<+}{&ln52SLioih-R<j=JxTnApfoe;
z0mTy&2xivlS++~PLY!NlL}85^r5>qPUd6E_MF<n5_k?ytn;`T{=`zc8nD?M634vh!
z_!XY%&jG={7DEG+=x360FZ1oPXvRweoXn>!Vl|2eGGH2<Oj!*v%<<%}2O}|~?g35Q
z)1A5)v3p9mhtF!O#oA+rJ<51kLo!#7j{A}Fhozto8TAkTX+mTAhu&dpO!j&>?K*&6
z|DZ%A$?<qOqo^8DI!-{$VpudW>y0wM$buPu^u(ZvX1Ynt?o$sHV&<U$dJkbJ5s}%5
z{MvBAMOAtx4p2FrM4ji>g^rbU6nY#nFO<MSSf0}_lsH&AzwmK85xGGz<ic-{^ibo2
zQs$mcHip4?@%CHd8bAWN2#MThmB~smx2*y{I-`+F-yAqoyI7qsxN!T(VK;EN>=J&)
z>Htd~x^Q-qo`k^?`rM8;c2y5LcIzsO<WQTIS<Pp^3QzMV=~wwqKOFxzGPMvwFkQ^v
zYm|-Dx>ZfMY}Qzu0BXJUT&(mgZ;h!P9Lx8O^{JXFb=K^5z)tAr8B~|&v$WWOyy=g&
zmsKsG-J3+b5XHAKb7hs%`ML%nShqaXJ8Cd1qpr~}!{nxdBFz9JGqanvn%eQpF72$7
zgqWh7&4aOQ+8<eVZlh9YZyipg6kjQD?t5XR?5cATu71%Mo)%smD=Q*X8+PN>|1s0>
zx<$C<mb2DU#XIu^XGe9IhwLUvw(?;rJ;_&=?8St^Gn0I6#h3Wc`iSRF4v7W_2L|4(
zO(h;`r;5DzGWf}<bNWntKW5mS6TKst?5OU7qkBv8?D#bPe%?kgBAFK6DhIufdB9WJ
z<Aep*r_npkBPgofdgTJVH|C{vj4F~DTs~2pC)z7f>(mB*H^|t56gY_)pycbO1e0As
zES2J*k3lpU)O&f;=O52RKp})XVbO1aPF2S%BCVw<9u8g2K(A*9<8>QnTc2?zWxM$g
zCzw$4$DlS1yvQRv_SOZd8lkIz39)XGMhBT*PN(zj{T<L7=~(J}zb;xK`?`*FF8DV>
zf};`kCNryKKfi0U*c}z4<;);=OEE*1Yc+OBg7@iCc2Re@fy=%9F9CSFUvb`C!fy7f
z#xMFq1di!vWnF?GX0+j_>u(=96&_K<*c|XICd?{v)X2p@=$afZF;(GnQ0e3MePBLT
zuKx}|nr*t7(?ePsnx<~K*6G6^E{zj6e}ziwLeKL;O^@ZNOvQ{x7wU3Eq2v$D-X>Cv
zG3vJjmkfxnJ#^c$Y!*2ec?mTBxLEL&xZQg}dN8I`k;gg+p0l!zp(hIuzYk9qFs0WQ
z3lMFlf1oA!4r+mRBO{O6coDhVg;FO`sLFa?(k50;;OI;B!>>^|441$B^d<<_ca^&?
zRw66yRwV&Po6vQA6BH9spp{&4^4L3{q_wN(;?ZRH$3MpgkD`9<8L=Si5j;_8XPUI(
zGLSBly1cvx8<mKljkUpQTnGqlMoc0J=-d_p7*CqdL+4f0LHo>HxNC40O|2u6VRpfL
zveZ0Y^*%Fao>^(4)juiiCzOSJLw^iw1RXk#Ve)9hwrFN`1Hg_I8uwB10Lt|G9d3C<
zUu0Fl_>ztdeF)N^({6j|J{DcV3^`!W;?!~Sq~EeNOb!<4i!nS+^%(_JW1i@R*a(oJ
z9daY|Hh(>_5YNghpY)*xzT151+C+t-f}$eOa)Ej(b^zo0ZOCM$U57bQ@nzy|eGl!+
z!HjBwH#_eVGOLvf+wQEC^PQ_4ZF*e9>o&D_C-POvG>1^IKKrB;c7u$!{N%s82MJB|
z?A95)g^4DjI6Yr#d=G{B8X*&|OhyJmTBVR7OQyJpgP0+_73g>$JWdDfXT)qUU-w?R
zlU+h2s_Y>UhZO8B544<?_Khyiy*WEOoCZli%Xv91JW@nsV*48z$O3<?5Ml}kymafQ
zXIT^8PFRh+NBUeDYbTp^mdCviCg3%oFmMK0k;*_F^kJR8y(0Rux*rE!eWs!blB`)M
zNZ$bL3mXE0Mxn;p?r5#3X9Ajqp~Z#d(nO&0)(udHLA}OH82@=_61r_om=2q<JI3Hi
zrxfmxgg)+yVukg5uguT;Z#O|OB^8oSZ2NT0bour3#3QvlE&k=c_kPve&7|xGt+)}E
zi{FD-qr<4gB(1Frr4zWH#&McZ+AMV5H@R9Y`?nAtaQe?tWN(a&m6*n-gEjhk*eLQ@
z)n+<nuUcQ;TY`9?xE}I*Sd1p6wRN5xtTO`P9EUb`-$hu7=`hRZi{9M2bLRo0gSFAr
zI`|skuYG4n?YsY860jLFc#X>;i601H@uC$^-jL8#4*8!aWKK7BX8QE_<N%T{mOgTT
z=gy_A(7y%cnAee^Q)##L;4>N<HB|-gA|fqE-RQ-lzwf<)!%jd0L_X|BvQ0M+E%wy;
z*@>ujrQjpXlMdE8abVID8wrrLl3^{=+5#tV?d|aPr4hrIFhheOyt#Xb1plX(0__AK
z9u3E0lF8l)jr8u~kUzRzX|l2j#0b7_vbYBe@s<x(W4ME)Va8uB6~a&vaNXodWJLF$
zhJ$dDR;`(;&03D-Rgwp0^gvl6d4L`Aa;_eZnl1{cv@f&U<vC>*lPg0dp#4aG1P+XS
z&BnShnnCf&Sg{F}tk&rDH<#cE*B^2D(cilNAByaEz#qLPo%Bc*I61B9aBdWr>5#$n
z)@?D{2U!@}Tx426|1mK{;kFm1J`4ze0qO1$C}d`1xIX(*+XTd6UEljRgzjCg9)vVH
zGo|Ot`2RUKcqdAhN?tTD#v8^rhEEBx8>3k?$I$e+FteOA-2{^F@Y>P~$0(*t-^8xB
zn5tG?8OkvYULPs=oS>4U#$_X$A)Cma@cHLH%^@!NI4L^+Dd|N#4iZtHDjFp1?1_ki
z`smn=FL;reyQQD!4Kr3^#za^^Ll7D4m=<^x+vc$!hAxo6nW5xYx&cMm3h7TT3`Ag4
zf2rZWjuM<KHwwb#u^M0CDg{Lh5Ar0%I3!|`lKbQ+qy?UypKJozpmtm?Q%5*ptz(bh
zJe*px&s)YU&QS!M?xK>@<$WaoaIpRK4gJi28wwIiuxC1&<8IeO8ryH#TVIi7$0w@6
z>_FSyL>A<lAgAdP)0k}aGFhO#((A`xlBL_qva7KeEi^2Vy!1hOY7l;3p>-Ml$8rE4
z5B@&2p|2OckTb;8Jt&>s$|#?A8Iz@EPhnsZ_?TgX7rK!UOU6YnK00U#+4k54?_OR=
z!PNvq_U)%3dUbyVY6Aesb035HrZZIy9jnn|x>gft?@W%RWn~Gwu8f0xRK#!8c^e@e
zM6~>C(rt1A@aW=1D-)1%$33qimJU@y7<yDp_b*oMpYd^HL9e=l&1weP2Q^z-Q##aP
zxd#FTo{X4$mA(0Z8m#dE#9Zw571>1WEC#G%r93T7L=_Q7H6#It@}dZTra%n;4&3Qu
z=6u2Ok8Jj*TY{>KQ8}Bf%>Cl*u*2<el8)D9Fg7tx1yA$2R8UcQG={F&>)T;>o9cnH
zF#>&Gk%>y3yW4Iq{5jp$YG)kRl^;K4G%E-bk?lnP&&J%+)Byp_H$^s)S0PmdD<Zfa
z%YR|;b9zS0?{@(-^W=ec!@^h4u$KT@-5;(3FLyq#Aj4?QKzftm!#?I+v|st!wl@1S
z_cQ`@pw8W`acP+U>=6_qgpQEpgMh^DpbdOuSF6%?`8tPg-SG-zERT&*t`8>xoRALF
zWy~Zapk?;{$>V|R+;D9=hc$!@65zUr>$bEkmxHEk0enE1#HI4*O&{P%;!rq<0p$)k
zN2(=CALSe7?Ccb7UBAr+-KagWDAQCXkYNPb=!#7exdTdOBA<hOyx$ew5}Bqzf-Z!v
zt8Hh#LuQg^m_O|Q_qf4^fxPOQ0(6SsJAKT@OLM@zk0THkJFB`a><GKsyG<9n6ASHD
zm1FrFa!nAwGx#aWMH1}tq-sH&GZZynR_B53OE#~37Ggv!PLW3}=PC!$e-e|p5*#!s
z<JAS{BNe~IV(}poTOH`(Y7@xdrObn#=Yr<NbI|wB)~XqsJUBU{6|d6$j1kKL3m)A>
zQ{W=XR-AI(mFe3<lkFoelf|{){@!GCIBr?T47Aubaa))E0<OP+3)#rAQh%lZ1W-$_
z2Q>-h$>HY1htt40GSx~xWv=R}fs_{w^s4Q^4q*qcx_uIP?LPT*YuFw4`XY>UJux=l
zXx76+>64NK29U#U@PeXU1UB>HXn$C{>$|4M$@uhy5{8BFuc%7^68SSN-peraufK|m
zz(K$@Di=5y(YL@Y1%ee?%oWEara?eav@zyY>)F4$qPy6WyvZ_^mWi2AXuGVSIOu8W
z$e>+YjX*b(WO2bdpzO6i-ENTzxJ5HDt8{7fSH%FK^^M!+KPV24cB>b<93ps5A3-Z~
z3p>U65jg-8Vi?G<1}0d>g4JZTQ;|s>fRKHV>{`>&@|W)iq?tO{_d$rPI!OrB$}sAR
z>rRgZc|k@jg)cFo*aG$E@?KGfD$%Qaqx^4BrVjOFNh&ewB8gbEYma20rGbLp8at-X
z6<w}!&<@qC0Npyn=~}l!{gxXKe~^&)*zvaA!^~>olo4tRr=dYeHpq;7Q{Xm%J|PYK
z(4QIL6T-dQr56A5umIYW1|{sqT<yB4iAt98sPORcfV;NMpKRm@ZoejdC|hbd%>u|4
zG`;;1kVg|}H(X?k)VQEk=Hm}I@JfMKZ19YtTTEj7+`&{za0T&We*bTM(l4ulrd~Df
zwFCAA&gr%>%g~krXv=PC8o8{GmCzz(k@Au%2c<J$GFN_7v9`8e%Drw!r=6Wn)f>~Z
z$kAeRb;4-L6B`k*;r@s>chJ)%{>uzpK7qQ@lO?~zY?S?>=O=O?{^<FFfRhf%7)rur
ziKrw#gQErAdKz)LP^qOg<qY$sstEgftgm1_Lw9lkdu7b=6F2^KpZ{Oj1KLKQ{pa?o
zTgx2WF-ebSRmn}L$^%zVtifG~%5mAIwrhMq5?8be6m{&IEms@QXDKc&v8~f$B5VxL
z$4=xjQ3gH%9V*e;?q~nW58subutwh)a(-u@&0BPApROSEbD-_h@%4#!eq0h%3aGi$
zg#B*xgJOzl+&5{W%yPO{=1^@k4cglK_SeQB*ivo=vVY{$KaZKZ5WuM`?33=wCP+Y@
z3T)l_dSk9eMgN`y$jkBeB)w$ZRyygXbH#{`@j`xgdD`WM;@5(d-uQ4a@K0;O%|HRJ
zMu1(<WztVAyBRv({syYdp_>RXUOJNH^#WqnaWoZ01%<~r@y9?NJ_nloocoG${DX6T
zgFP`Uol!BV$kV$$g4&1@D`vp>Qu2X&7bN7HULQ#RO32W<2f0EVKvr|6+UuJCjV23X
z1~x<#H6H-Oa8PXovQ*CRJIkaD@(2QJS7Lnj$$;>X*LlylRzSN_ei)&vwe9%%x{teH
zXP6DFSbpoEFITk9kpFV_$?|kvU2Gq?l%dkRC}XWsaSIc>5jb}NlT_i=#uYuyX+YsG
zD~=nF1N<D+vl#K}8c{o_875Y-DFi0IN=4l>+-a=5Vy#*Mn7%a7ovSdD;kYzY@oSD$
zlXow#Nc_8j0rdHx9_WK+N7QlPVp3D09-3YA!O*oUH3d#9qr*ja2fO|S@qIlXAi$Qa
zbl9{gO%w#Aghs|aAhUK@t;n(6nBXasF}Yj_E;`t@r_eO1KetW%5&Z1rFg||h5YP(Z
z3!SkMj+j_hE!yr!X{QL21S!qPr3|2Up!!x?=!%aqAKyX+E*OE(@+S+pb#^}I&qY>K
z(D0EeK-^=*X9W*wT6MbrTikXTg8iQ&(_v<Pu39VXpw4QF)w5l?&PWPc&ktd|J55B|
zt>8WrgDRN~319N7xjBQrrec5#=I`(>R3(8L&9E;vPbU5W*E*L<_D04}Thm6iZX}$B
zwY2`}7Z6{xg!=fR{BV|jQHCiLUhjRcYb>o+cy$-&Rio#?ia6GACP#qcYqWYPHRyv#
zz^!+F*3X7LvlfDCwi>UcfeK3D-Q!mpiDFcjM>0YHZal6DkUsf?um>+TfX<YTTX66|
z9X6-TpJeesOCACF>q4SnU|#*{(e7kJ9niWIT22eZ^1GCd9**dTRJ1fQj>&d#4Qt@J
z)60#+678ZVfQBD^gHDbp?ee}jR0*W(C$FM@YwIo{4&*6XEdVmhgIO_V)3uWm)&L@O
zodE0|MvwN=CHeSznfW*^f;HA~+aIJIwp3G#tI&4acvqn<e3KJ1t4Q@T(ePFCh<Imk
z>&Ir`^@B^ZYJ!4%_4^j*Kj~3Vt{$t|WJ5*M#}wR?FbeHPkvz7u+i7f`vgzpX55B`g
z2@rS~;D_P{+|Tw^4s8K956<jmnJ^<sD2U6IRr$(G1=U|AghPp#FW(|kz>Ik5OES*;
ze;WiK7T+c)2b)7fLn8{*A%;(u&~)W^=J6fA@Y%0ftt#m$&SYjJg(D<nA0#3=k+{lN
zz*{<SIuPVNsF)s9**?wf4WE>Lu33!G-M^E`yzC27mxI8Ac7bGd@~`87#)2RmMD-3}
zre8J><8N^4sC-gl=zEkT`%(d)d%U_OslM^mH3a6j2Ewi&44pclt8@tEjd$ux72P;U
zmx-UKRK9VCJ5*Qw;a&o`LM0A>DF>PU&szlFK~9mGfb}H)(U|mv0y^?){>pGmM>O;5
zLA&(x7)m2X1lrM8udq&w-98>O%?rTzT4K!WmC%k534Zav$!;;hT^5_p3SRKjpDsh{
zK=BJl{C{5wwL_mGR0)L03<Ijbw`0J~i8es_!h(4+ud;I)h;?GFO;!z)bsT~Yt<gid
zw`@Wn^3(1v_Be#_M1j==Mr7GFzb0mP{bA(<-e-~Uc?Y1jSZBbqO}PEc!~f%TP_6Il
z5i-K=MvZqFFRZ#(_X9CjigA6g2Lrl}0kt#<H^Eijt$3kh2TCiWCs{6A9#P=q4q{k>
zZYk%wXEc~-Rl~4~w%g!Z79b;>elkhOXiA<CrsTgac0(Z{D92oT*?|~K#XzOmR2@Rh
z@x-EXnYPUWnQp*eFi;w*C<$)xG}qLE@Mp9JZYQ#%^&;-2@N9p<lc~bMCJ_gA_`#qp
z+#*zYfDj?Fp!ERknT-q|z{HF{w^bcHm*PLqjX{w;mxp1@fop>EzN8kRApQ&pBRtIo
zsN_0NfR=-amxJB5G>xa8;{IJW&@{pmAH=_<W?N9QmC8VCQM0e?a9|2<HHmnygG#|R
z9*_;qa}jPkZFG}vXu6tLf8w@MB%$X57=Ie5CzJ&p#9RGZ=iCN9W<*G3%q@Qmv~t`<
znjq-y<|1=rPw_NiP4NP2_4O=5QUR)?BC=3CdBNom5z_IIk@!4Ea}gY9x)+-l=RJ2w
z`RudDL-0oI9ON%Szqm1o#8_uPDxCg|_pL$o2vj<dA@}e|u!nb*i9xfn*miB?DX5$%
zfqFOssCla<2f^L19176r%25m;1EE_T2)U%at#^)nujm4j!*u7ImqXW%&<p_Lt|Qqb
z9qdLv1_pM!&dXbDa3AnWl=#q)#P~nnC;%x%`+r>hcRber`#z2%vXV{6zR121%HCyX
zy9k-t6_ve5R>@x3A-l^Sk-aktAq_&x9$BG&=cQ+__vicj?{#~=o;*F)c^>C+9OrQd
zMxgv=eb4)xstZrfT3*sfWr2tE!d_4-x_3SfOt(fNqp&x2YoE_7e~?>(@;%{!_>m{l
z8+$2DAB>h`7C<jQs2(9KThTye;;qz7u=qsN@GI$Qf3L(an|}5KIHxlf7EUgXYAyrD
z9P)p}OHdx!C`-YxAV8NKjAcbFYKYkY+L_&L`I5MVV7V3g2omJn4}u48U&#3V3cX_e
zDgqZAWSC7HQ)yq$GMO1YEpcTqh#{zX4Mki;iu@-Gj2cHwXidkKUIQ#hpoBeGB+ZGh
zJu89Jh5x2wCD`Qux5xb7gbdz3_l?i|vd5btjRyx?^U=4-dK;EuG8l56#4x<c&}qxG
zz(u&sH@yJ?W~H}RdD$?Yo<2|&94pq}xIcX42J@Oaj1~FNfZSlR6@T@qyYazwn1Yq;
zI+rz2`~Z6{w_~K-0oc58*xNd-SwOAGIsf*arepm7B=y9S;n0U>ir+76oJ2w_4DRdr
zgoJM6?>sbQ8MCFAldbVAMX#6X{rn9j<eT$0(;1xH5k0$>M>HgD5?<zt`#`oyJkWE5
zK(cg7!{~3ZE>q~-L@phLN9X9`a~T=_zVH8>1pgs>@QMjwU=3z|vxOa9%*Jq<Y{2XO
zTnURdto@Zg%A$Sj*JD!g8Xh?3E~0=iOhZc$y92SG`$7+C6%>L`+Qe=@S&^C6UDD_e
zFxuvJ)h}O}{RTb4FCH+SMGJHkDC?2g-$9)$$^&7Sq38K`7;FQ2GLRJK|D~Kjh8(6M
zIUB->zXYR~<{izU+$LCDU}K(16|&a&JqhaJ<xZ3vW$4=E%WQEwn5^>li-==&Kp@;9
zb9#H+De-I~jpXQMz{y}fUEdIwC?hOZl4$5@D&G6-wUj1;*C?LyxA8_|^wY|$D<1z`
z%W!YxG>BkI;D-;gfv^Xj+z|OIkb>BlUk;~&no$ZQvrs1rkbRP+r0;WpMB6~?&wMH;
zmhV>3o@UezHp*ZrOAyNgz96sB^3}tYWT&)Ima?q;CC3WXf8W9!PCh#Ut>tN+(7*42
z$V|>I((!E~t_cm&oDZ|G0zgz1%BRv98)XCdm}~@E94B}@2+~XUP`AWemrF{)G$Ap#
z4Ys*&rX)N@P@vbs)!q_b9jW!%U%yTEi%wlrGifvNjQA5lDLdDR5`zNg=_>Ke4xv*V
zg^R&a!7`qA{-3`h2t@%Z$8ajfQvA9pV54P!VC;tM+Nwaj$0bxsa`HFxTlWK)O6{=>
z)qx+H8Bg0<S<lh6f!$bxL5>tttIsnv4tleAwIZ`xv7hd3|82_{hWlTb&A$(B6HmdY
zgtWD3ID&tGqA4&TMv>Os`)lMSH|czQX5Ud42I3#oAKMY0-Gm+QON~ZUYV6-CXi2Ws
z;2O>CotBV14wz5v``;n|-}h$0q%F-QeS#2xOP`}h?iw+Cv6DkU8}dD2xH7!Et9ZYc
zX18U*dz5f=voDVQn}BFhg);5;yuY6S89(;zj%(`4{{L@{VwuDd$Fo!n$!S~{3?|uM
z|Fp9A9(E!c_dYo7{7l+O7j<e{qTMYs_h8Wn=4J)glJzgQ12!s{(u>vqKe1r3yx5mm
zk`3lFVr(a5-W_2!nCsO^in4sS`!S51$CpYTZw#TTSq)2@Or_$z#HwwFrsSdgh<gGU
zd&2)erlPupeL}y1QTw4Q#)Gb*4(o;-uL67Es~L(~Zc^JNMmB3_+)1cpi;cwzCuJVq
zdwpLbc}<mE3Jt7PJ=BS&#@&yGB3M4aFslKRiCK+L&dZu+kXeU0AbdkDVhuBh93cpv
zk%3bMLuf~t0#=Qa#X0+rn{On?EGhh4oz^oCIC<y*G+82_v98R?Pd$uM13MZ*dbAjC
zIn1TXBA(B8$H&JtT&6&etYps#Flzbyllg`~!z@lCZ2jZ|*uNb({J1-hzPwP){&REV
zM`%I9_h#FCNuC4q|E|TyLseZp0d$MS(x>*BB0hVo_754WPu8cauVXC!kSY;JM9tH+
zdX?_NDU*8)j;SZvcK!c_50OpcKqa6!QQL$qRaVTVvDD8|#c%!(-B&{iDTP1z*vHO;
zDhxO<`G0?7tNnkA*VpefZ(W1cg2W8h9=h9s^bO~5ag>EoEp=ZuuCyL|VHVTmT+Q*v
ziuEqNO*=Gk8wgnWc?1KF<U~>H2qdB{Pe#-dm!T><&N)n>E!Xcy)(QzToqmf47*oux
zk%U+Oy*vVVoHw%gZUvkG4&46)tBy-)KF3g)Klzw}MtakO=g;r`*6yVp@n@V7U+)02
zfSi`mTk+q&xSn*MT6^H{=S+$?zu853<3X>nJ~v@TU@@iKL;S>EH3*H!2S)Q7-;0rq
zrUS@?P&zaARsiFW1tLWGNLXm{54_cHQh}dhyl-tSciJtsz|i(TA10JYQ^Lm=ggSi9
zeuA>#v9SU|p~ngdy~w>4EoHk*2YQ?r7`Rxx4a}nu0`0M34l?+CrJehRFU>#s+I~@0
z&yhzB{VB6u|JbN3ybS!~J5!_9I`h~(FUw51b(q(lPEWwr!O-JB_Ajh3MvAN0{nqD8
zQpH9s>b~th)SU#jTODHzU++=%($Tp4#PSsKQ?X;l-0zk``hUR{g^t|L9t^3PpwL>l
z?H336KY0MNg~0Vz|C5j2_R6jA$_fn2O>{Aeqar{PmTR~oax~}84%tHqzgz>0(U9Ko
z+Cl7LWlJ8{eDBjMDGzaq@aRG=Q@Vd{_7Zo0UhPi{wHjKqRfR6?!j$n;J0=I@=G~UX
zb&PghcB}Pu-#h7sGm%!xV^hZm^H0Cxw!#o^>jez|5H(-}`flET*+2mx3Y-iE|MQhD
zfZLmVccmlQc5#$vJNyB5g*viD9|LUbcL15Z+67Fn4%qY4+Gy0b<e#2WFsVEn{e-!S
zN4O$aza>@=OAIcUKdgHS*cINl-0AVO$I3bx<JfTo9m@ePf3~Xjs+H%o+PwSoZF#WK
zAF*zBrsCI7@Fq){xgE&0aSciuds}ni%mXlN%$~K}Uz=HqNk~Hr;GLrOATDBra$`W#
ze}rz(RdN+IwRp&uh!C)fPfzCo&@FOV1LHf1OZO`Ydiv*hNtTXk7vJ&4IXdC4;@jT0
zUi&r@(r5duH>Sl@f^K1@XaTz%>vF&eC9|efGu;ceX(G5~NjsFXY<h)49;KO?=Rgue
zjUvh9_9?6L5;AV_u61|~KxYSDdX5xI+{r{-3}LA5(VbT=@V(*QoWhQJms%1}<nMrX
zkj0vabtkBzd;2h1FhLD4Rx!{%jeGVNjpp3L#pEkgTMVBem9luvSt>jAP5&ppZ{`yT
z%MM&|0QPQ<=_}Us8KYRiG;6H^e|{HcO)!7)KPsR81QcG``D)0%vy#IYPB8ks)`Xu&
zaCUFH^l}ziJA}unF91vwmZlN8RCbwfZ3(si_4ta`NTdy!p<Re0h-A^t&`rKEoE;80
zjeCgsQP23G?v4~y=BLtV@%7IwGtC1jNMyQs2VsXftg~(1d-n0nX>k)Ghs6pZFYR-G
zmEs^2`Dn%?*qJChG_eCyjoZcK;nayhmwX(ww~$#`4>#nQkY!FMntAPi^3dc7hFUN4
zigWckRuCJ_Xa-sfsOOuQaXhRkuSG8D@>xI321S1yJw1zia69aa=}r2>1Na0h)ib-5
zADT#>>KuZoX!Ehb^-d}9RDI)_fiGXLyty?*%J2lN3wdu{V*E;;4{tuym9&p>9l|T!
zFcE0_o_BF%iREwCj>AMSt7`uGtM0G_;X=*fueKXW#$W(5LEhuNY|pQRjgalo=|OAn
zgW&)UL0flUnWN79R3MLkyZzB1`iYz~p?!|t&g<nbMrNtJrMfUPSr0P}EjJ&p&j5v=
zh7DbL_gCDxj^(_Y!{=todM-q=tO8+UdAnKS#0g96fxIF~({h>7rSB!x*0mKIZ@F^+
zti@dWW4ud8KkE^LG;7iqyZiaH#X_+VfAv=@m1Ft;%4`v`*qL#K)kQ!7aKIOX{;pBG
z={Dn1-!J=n>*?kvywB%P<X=ODo!)!FTXN(p1Mxb8RW6+{3~_m5lO+Bi+66j8<g$4+
zeVl*Tm%}Ds!(R558)rHIU{eB1*SZ<0zdj0|_yBzT$|3pF)OLl1Od}f_Q&0x`fEym-
z2!?<~`q#q*52FMg6n^0woqlQZ;H?13n`_?KB6eqYfP>gDl`z_nNIs@%p?x?AIzy11
zU#9nR@!?w5e}R{MDS*{{D_tr6KA%YLoI`R@DnQ$${Fnje)^qeCD#KDcZ_cs&3i)$?
zL^{RtR=(KJ(gH2a?xNn%=4$|`UclUFVkw*<htTf`Gt68NRVdzH3_%5h&{mb~gd~FA
zW_+OiIknJ_8IvE%cK~ZdU32rkKQLlLufFjPRfL`;ow)Q;`An&UgTwmd`a{W|v?#3r
zEg96O&tRd_4@SeBB`iPEeAcq{@Ee0&s+a$yTvgV_g8xuS2tXgZ+Gl=$XB;I>z$)YR
zKTeZ!v4Z~CZAj%5@Qkbn<W-g@rMC7&k-W;AQdZW%CGHL;QLgj|!ShkB+^-Pg5nB!8
zoYLl_qSju;s~n_#E=1p2hIqd*vhn0-p=D|KO=g<pWsmiiBjaeF4>Q9um+h_|t$N71
zoWW+Yx(nVE+^<*#A3(p$<g0!o3`vM))P?1rp_3C%zn})2jmDkUlf4oScVnwaj&TDm
z23%i!Qpr{HsH}g@d3FojWd>}jXw5jZS`=PxO0-7VPrB`@aD*X{*qZj~TZ_8KDPzce
z6$LS@#t(s>C%Z5IBq%msq%cHO^dM`U>RwKH2ijN`-RCV2Remes`nAhud!c`No%oJo
zXbQve9g>Sz1slQYv?Hd|cSGtf2eTl~wKjqa*zBX{a5Xx_&PW-F{`&{X<Z_GNZP@s@
z6DzslOzF^bdBqQALqcA#_&>VN_)uRZDy;r9!`)``926|oD11!;m89Nr)VZpk(_!X@
z)Rt1yFp-{y&azgjAS;5@`oqgkf(t>vlmi{MHH_{K+)XOm!|mTwNjF$KfED-72kv}?
z5rsWQ1D@|80!RDeqQmPzr3vB}Zi%gImJ{K8*i|Rp2)-_k0aV`-cyVo`lUttb;#_9G
zKrDvceb&Z@dvV^KHJdeShrAWxT}>f&?aj*H3)3dX{k`K|OC#M#3ODz0AURxd&N?}U
z8c_IVpZF3c`a2hp`AhQnj8uht>Q890BgWMb;uu1XB96l~O|2pu@LvDr5vc^Zc=!Ui
zV%dRpR7bPSPGJ(i`o6NqT+NIB)Wi?Jy(BPDKTDI_642$=sH8h?lD;EG%~!2Rx%TlE
z0fA~>Jc3X4HGu7`X1ztPv`~o)BTf{dk=i6@xVN81+#D}2@@9QC-)0hoFl$!Q%Rz-T
zFWfq=Qtur!cDwbB{wcjEkN%Q}@?@2^4~pAjQtrfRz;0&X16k|=37bj&Wh9Me$8DXR
zI?NTe5$H2u^IDcZLC;+I?}2JB5p3Rz$*Y0>)&EW%nK|pHqMHhNFqbfvA2wxY0i0yL
zkkCa|<3Y#lJD_Zbm=MQfRaw9is2Yv2O`IHkGw|`?sic^;#C<Gx!XDU>#g!ad;4DMo
z4wTtNt$EmjCyspaj#1pu>V`h!qqfit4wf^KYVqir`q^OOZ@YDi$1GM}mCCaPCM9@^
z8Wpc$2i9Sm|3l~45dc*i(YbC0)>6FnVx`9ZHigCSq}7P8p^nFeRFq^LsKM_$IvSY`
zkB3A2$UgIy*XUNkT4^{N0wuXpR;?i1&=52E<n+;=UU&j};@Z?21NPqsUZjV+`CAea
zJrV%Yan9U%<+44hNXl{ZPNS_GJRSVq3Fu>43+lcFdckdeu!0(bV;C9~V{o6)_?-B;
zw{qLCVbR3>c>Zp{(M|1k!|J>BE^q@D$r`{6plA*5E8q5O^FDFokEe+~;`;;Hf%bvO
z8CEYA^ZZz^!{DO`{pURF)Ma(XhUuXvpEmqcj0mHDC7?V$J?xD4;ak$tn4K8%^8+uG
zgKNa?X>vIgnD#{uy!GT>*F^Kzp-i%p)Z);w6%hR9|3S=ut~+#@AlhoV%hrm64~1-8
zi0=na0+aa#f!c9h!ll7F_{>f(T#lbyknpB@&8L*VFT{l)nrm1OCnLz&<AF{?XpBH5
zzf9S@rD74bO(42*=hYfbqGI<s>bocvs%DFuY@X;xay<gM>L~)sPhWBLEX%+RwttL;
z)FLSCrHumN<g#?e4MIuZgG~`tmM#utq;~9p35UyU3iWtlv+vh1D)|%0bGif|^{AjL
zfQFb4EtifcZ4bKb%m*;QtHs6cO)9wwCmmZpH!FpxGu6T6+iz0u-CAm<`uHL#RefK5
zI8yu7H20qgHp{f~9H)w{MU#PzR}^X}71Z%V)>@sefk$$KrYAnx|1Qsw#p1@4W6rYJ
zzc*o=4hJfeq1#MGs961HZ>R0a*59I@EPE@H$k-IAM8&RTN;39=Su-uaNg*f-e(YGR
z;ZH?Et#c1fHUhGlhLt<EgCh}IWT~j+2-DEOu!!zkq71Pw5Av$=&o8F_%5Gu4ZICzZ
zJ}~{wmo;ju{?Wk`MFbzRUVB!!mWhITC9tpuM~tk@v!9~%*ThQ-XZlcl9&I)!<QeXZ
z*Ev*C9Qe=uB3vsdld-~ZzNYrQ@D|Wdv94mDt;RcS0{?WlF>@?h;=MkD%$<K#>m5Pj
zewV=yod)Uut7KmZHpzy@Z^RpxSN72N0}B(snS*(U@$qQXiXpw|3fo0Bx(LE4-BCAV
z_^B!axem8m@s}L-8tR_0S$Z`pMYK`<P|+;`LBL>01QowPyltq4s>K#k=dv_1<oJ*i
zCw#>nCQ$92wfZl2#&PnwdH(#_|MMHX+tDs<Fy3ZIWH{^pNMVdkCd)kVri_hw<EHI~
zJhq5T#r<@zlZ1hzwS<cBnR?hUC`MsIRKFAzgm60QGGZCqAw4KXYP0POhUHSYTuD)v
z;cnLZ@8!KuCP+AQ@u%tJzkk5UCP$InNh&KP)wNOo#Xl%<$}Zb49x{Cd<jH*?^(0iu
z@Zr@5r2Wuj+#g{foQ1|Z(zIm2J`q~H7+Wxndt=*`nvj=*G+y+UE^qub@>q0i3fr*~
zD*pIs(W5V-ONv}VhEoTNq_|v5s(sOW!;%{sg<6^EQpcN}tPYb%Wd=U9QJ}9r<9VyG
z;2RZERt2GW_w}oAT}oTg*2CGL#ERUN<ry2NFsSGoZ78eIY)0PHLZB<0OzTn~*5b!$
zMyM>A(Cm2e7LumG%XZpxwRvzjE!CCZEdK9>WyLAVf+(~m8;BM<|2YbTM8zt=zn5Tl
z$B%_JKcc)d*GC0@xW*c{!Gl^c>g3y%R@DmNPC{wbW1k=vq7`iz5q!M`p#>Ux2tG9(
z0xffk?sh77tLh=KXb&SBvpSqc6;uxd9PWhsD;4E2VS*+?!rB3h-H(}L{&R3$yFJP?
ze^g?njB>#I!CPfZZefZ+Jo1;T3D^0;Afv8P&TY*6K3fV=^4>ZGDqS0jm*P85Udz@s
zZ>?FrlQs~KE(xI!GsJdkP5FC`7PyW<fvQs`Ijb^%doL7dZ0v}$$HXAEZ{&C@b3F3?
zK^kzNnCj1~%$MbfK|qh?T3FkpS1}f*sL+Y2w89&5=D)!jX~|fNu^dyscX)baq0yl$
zmQv5<j*F`{_2)4A4<XSEJP77vlyZk7&#M>%EABjz6j^R-TJ8a2Q9K)Ao2{1p&!uK0
z$!vTD?%>?CR%yFHOHn-d8i3iA*p}GJsrOt;AJv|eoD>LHBT-WJNjB#w+<$(5OOl?T
zYa}^wg~!~?j?u}HFqfEc(2LM+qLMKvcJtle*V6<CI|0of)N~~B-=hUM#X_ttn)`qw
z_cVIL8DJr6^_AX=mc0))<+bk}uaQ<bV!rDl85}a(gOuBmRSMG4q22Y=RY`%q7bV@D
zO-Kx!H9Fi_N*@*6I;@Ne)nrJm#v#XKR$hW=14nkY%K}F05}}B$e5z-zl8Cu@3YymU
za8HpJAJy~-vu5{}tDvSYZL$eyHH3xNQ#FWS{1In)ao#UdM=Q|e#oJd<h*gjtV&wlv
z?yPo|J&_?R{i3X!ElHxdGrHYXpY`SXb|fQc1<k)#`Dsh*Pj*~CYW{b5+Pr`k-xE3P
zTmP#~bOgsg$8PjDxX#CCx_{u7>ch~QbL~GN_KS%0yUi36_)~t(etukGI)S-m+LX9T
zHQlQ9>B}g@Dr{5eJ#<RG*KH@piREw~)NYdWNBJSE2<-v?gSl{$7)5%4sy;SWFUycK
z)F0)F(>N?E87)k8JKizovn~|j$bRJ3E406CmWRV>Xl$D#xR~Xz4_EL9Z0P5^5Gzq;
z?$aD)C9TwkDAvphq-e=Na?`kH(e<yQ1h!!m;SLNGhWS@_mHOP86J22^fYsDzyqc_n
z!5JOYq>|FOSAgGQ_3tX92#)tBo1-!t9R9BYOW}Ton8gLS`Wwl{x6?xTMi91^j;p{h
zhY+oH5F=|!UP#%S4MJU_xh5Ai42`q;dt5|lSP^22GLFMiCnmm`H)eGYs#!AAHOm2!
z00lK2DSOs%VDU4POMem}oZrV_G<c#Vb`Czjc)92*qG4MwFn!QSVasA=hFW-pW)7GH
z^nInwV|kpL|CyQmlQ(aoAX$|oLo#=-%J$iZN26IH4!SsW6&B}eUPg>C*z&z@u0oKB
zpC{C!((Q134E&tdI~cuZuax=`3p7|%eQ+!-9(+*jKE`PNziP#fdFBN69{+phC|pmh
z?wz}(<#j~KXO=QO%v0+I6gkK4rnwJ>Y{Q11nR9}G-RVm%iHivx5-fZo_Mzx|rF=(}
zJ{2KvyjB;-*e2_4>iWyRcyy&&f^hnFz?5Y;$1~?DRqe-lY@e*6`Nx&IQ^>Mo<Jd@k
z%m(2zsiYpV*p^5PnTS0GF*^KR!G|EFwJ)s!QzILpQs$~3Q;{meyYMJ3K|rG_+m|U_
zsv_4{Q;V}g43};w5=Kk6!K}d?)k8E3_Q^*!j@z*(;F(aMiz_hhNOMw;Jy7rBAF$9z
z1MTdg_FKWUm*}f@f55-ORl@@H`X-sL2jIU`u*q@_i@Di4Mx4`ghvb4F3+ov%(sw`<
zlHy`Vp^DqSdYhkO@0x44Xxui2TB@g7<Hh(F@5x;?<Cuwe(?!hlLkx7xa~uP~PnkX?
zZ0a3V*+(lN)RPzf+5jVednwH}+9b#-$>C6~CM<;;!PmYTezgk0hl?<;bL~Q`zQ1`l
zlj(!oAb$dEAaWv>c{nf7lJgoz7rzkJZN$|057cGYVl&RwvlB$(Oq9)4Q5MH2{(I#F
zQeKhd?AJmIe#_BNLN<))HsEhWY=h%dC3E^KYN+skh{{``4dE<pM)I`#Z<+hRh=b5=
z#K2xFega%!-<?fA*zX8kVAJgqW}JYr01%=&!7O1kPWAW^gDN2*x21C1L!f!Fr4F9)
zpcy>ppIoV#?u$-310H4{h3yg(ysYxaC*yiIG+9XLo)3qmk-2V4x&nmmeHEPI(#+0j
z$y!exiw09gz1TBTwICi_p5G+}p(>~e`6oWmx|}X^xriFNE96nE`oK>qS_#zIX6Lw~
zpOO&{QmAKvzY>YCI~sY5Ae$)*!3>~(k+C$Z_Nza(<737?eE+(X$p!4fxovHd{sF~(
z0QxfvIU(nX{-~v;&vd+)TR&~kw(Kkvb>n9foryU3_HXFth8~+vFh%BQPk#m}C+Gon
zd;gZH+7#r8-Y{^gqEbD8sp2f&Dnv3R@9`SEbQctJhBWBIMZvYKldVHEzm;BA#VWgY
zPLFEk@9HamZpESD13VuFAeelaRz0~F!Iv+#R0ZQo0i*D1&(%r_8r#z5;7TM*zO%s@
zL|#|#{<K{-$t!!_Jv_rd1qffg8FTv6%Y&xEiDTGdNNI){RQ|rG3W4}*;HVWY^4Yw2
zQOp??>_#k1BRmJ0j%pfh*8;tuC<lo>!CEj8S4vYsC8BF`^Gz5f9@q!cBABX)MM!Ra
zuB7^f8JUgOTccJJ1J+)?*Zo~o6Yx&;5{KtBkJ}TxfH4So(^1xaXCSG>x1)<ZFJ-QL
z&~vm?j;!pWm?PsEgBTkzua)xp_%eFYTV{KgI~9XCun4QIa8kA+1Pe3b^W_H=(aRT9
zYp>cJ`yOUwKrka=>vPZNtpI>5>j-wanSI_6g}ceztXcX;T4?h_Nirlhu6O$HIxWuQ
zV3Kb)AK0ez{&olQ={#wiXX~~_wB2VY9;uuPLiU&k9#r?J4f5yGb1D$7@xeaqHfhJ#
zYg<zb`V7s@bJKxjrBxJ-t9F+AT;8-?Vkx_Si_P$s!mPqMVa)*O-9)|j$+PF=l78F!
zijY<o1vb2~i^fVT!L?!IEe;pfCmQ{hh3^+T9wmn1g6@5?Imr>CG3l8PCRhXVF6686
z4OCnqXz3bOc^N|AO%&_yXKra-vaxKFu%;{?uKSgem*RJdf-p~RHxR&+cLNbq$lq_W
zMjY2OM+V9@g%n}^^zCO0?fA?ZV!#bq$hJ}4kh%phepk%nvkdDgjH1ETkK*E1l`04n
z3|-(27o-xx-*N}1!V7oY4>y=SV+SM+Bb%)#XdZI-jlec4t0iryz-1~&I^#j?T{D$h
zB8Qf4nwz0{Yz)(F%m!n>2HW-hoVaDp1R12gA1t?)&W9zB2jN~q>wJft#E`ECSkJvr
zKoRQvo$C37SODFILlR#n-}M^mMJ~aldR+ORx7V4&6b(CwSdo7bZ;m;j{F_Llj`_b&
zG|2ENR=>LT_RqDMD;fce5QM1Rk>&sE2P^we@MS1|rDgiXc>4VIf&(ceb0tn7be2);
z2%@ge54t6yx#U@=;NN~BIEH1wkL?<9g5|o|j{hP`r?8b$$QWOTYePR-qkZIs3`tVy
zCjR0b!}3TDP1bF);JO>xGL|Aj#enrm!}AR@LWb+b70n$AlHM(I8K~=YSY4O%>%pmf
zvMm||Ti*g?=r`wBt%St|@CkW^Q%^Abi5sw1VQfOV%<f77q3Wy`pMN#~N2F<A#`Mi|
zjvh9DUs4xp?D&q^rSMN}Fk@U)31W1gVx<GX7tDsQ%txB}7XbSx$szcb1laWTMh6XC
zzS7d;cd>_l1BQh`KXs6=TEiQKU?FUh0k097IZx|oESMi1Z_pW^v-Rfk$Qwxte|0fP
zIlE_ys&f6}x6d(a%KcS{)6#)?(NPFnJgOsUC8S(I;9;edT`%m}9$Q$GPB)OMOif+c
zhg4+^rO!)jmP;Ug)kvs)REYXolVkL+1CEmc&INpEM)$Xn&d3Lbe*w`e+#y%W8KX|L
z&uoHwnus(>x64uO_Kf)b>`2Uun<IiNvX)`eTm#ma%-|<jGY9_3l+Bw_iut^2<uLD%
zjOTm|=O&GO;jsF#lQ;LmlV0s%3I7@)^5`3@5O1Ab&&@}h=&263B!B0ttp$e_v>oSB
zjY_pCn2e-@q{2+?%76&c3UQx`{dEY+h87bLc!{Y`i81TI^8I`(VFwph*evJMILw+J
zs7A~G9tJ}1N8SHj*cqJGig0<)bN}lH#5ll5sPq;30kC<LFU(b(Y8CBGIAZi)a}-fx
z;j{$kwPpiQtRH|14N;PM$89@WK0wqCFF>a_uUh|(@xCF;>DrD~>2*S0<=dte?^nps
zp5Njv$^hP}Y9PKbL{l}VyUY6h`dMS2k-_)LxJS*9@EJ@3XW5SOwn9i{RNy$FfUx_^
zuyDBfamXD>k$i$7S2xcNCr?8^J(D)L!LwKpXdupgpVcQRk+|UUf{=M9+1CG+kVk<@
z=v{YZDW<=ewMY_Ih2{Rq?qqnlMEqm)aLDI}lV-HzS0NxQoN%h?l_OIurtk#IIbY8y
z+hZHHQm9DW%B#Zo_Dq!3Q0{$~MWwIW*#g&a?pbVPY)7p3ZceC~ZIS`SIJygF!e?E2
z%bmj)14Qm{TSA;;d5wNS#A0D08nqV7#mjI|N8wafc@2jyi<f2eke`Fe`+F`rkr|nd
zKfD26Oik@h3a%H`tQ>sF__#J=@A0+|F)Tv0LG!;>meUQi*38$Gwln|L3~%LG_^+z(
zK{wpo{2eZP%zE$&f?nqc9j8)m`HPExjbI<hQoH$bY1Q@gt>t+lR~Fi`RyRX0GV=4t
zE%?VbSo3yB*Wfkmi2umLu&ow$HABkFO@)lJL*Qmz^32q<9Ljy=J^!w^OGy9#nvU%}
z?Mk4JoobEMscOl(mRG<nt-+6F?+H#MaY{$aEArso6Pnr$eexz7fY~D_vW;~JOPlJt
z!DuB~Cp6_!|LJR<rP&f6MF7-d(xb&Q43l^DrzQV=WHn?Mm`UL2aoW}F*9c;hn)uK1
za37m_9dt#X|6)WXs&QY*D}?AhsfWwfz)=SLU=a<D5Mimq!cV;Ev&Keu<m%Ka=@QV>
z+}kzCs`Ge!db9I>me0d}3kvvKgHifb-t|Benxvvtg2aVYnCex}i!a3H&2Z~)8nym1
zz!)xU+5)lX_G1@}K~HoxS%kH$8%H-hhK<{ol2AYkguHN?>f5q|P-)W1TjB3b?+@V;
z<BV79{#w3(cqkKK#NkePShiwXF%<Ue=YJ#eV>wYW^d)Ce?T61!OSr_f!0@@NamtN~
zet-%)sMMIoO?>3IAdqyvePtJv=kxtHQ{Q<3vx5LV@E(Q+so&H4d2re}_#eUBNk)n@
zGWzZKkz5V<dju>cUc(@KwAIunu7bKPOs1>ZvuI+KxijThTCTR7E&((g7(P@{7}`R4
znl7|M`S{1=%Uc~A?6_QF6ac-noa8$c-rZIdql`ss;?Yn*;WQHl>-FrPfmI6qDr1Ud
zH_v9RXJX}X0enPOJ!}kT&6b=QG@o?${*S)F0m|JesOsqS%^b(Z0nsf$+g%NN{3Gj@
z;59d%!qxis2hd{@SQQ^34#sw&&4FgFQ8gqs6S}|;Mq&r;uJk~ca-+$4F({f$WNqy5
zDpgu?X|z053^i_kIsPya4Kl8Z6^C<$m_IC8J)(_`4MnuGEr*5L@@<FNwxr5o6L}G~
z4;{)S(#&7FsWHP7f|HcS3?UTuQ;@En9mXL>@ROTEK=RMl(*Q(l+jgXxO`NjAv_gRP
z8iw%6?ZYAFi~YIrMK<W(kgZM-k}sRtt2ZyB;QRt6^<a#S2bdC#@sJ6n-23}Ne=CZ8
zHUWxoM&{+Kkq;R5Edq<wiL6knD^Yk&h*@7lvkRGoe@jzVNl4MqN*b|PIun1S^XV~@
zv_z3PJdu7u^LVcezrE>xZ2XI{`aIPS)U&LmcS$}cah_fNaV$jx{|zGcwrBxkvQ_AR
z{jk|LMwG;2>2Pjho`|<Bw(i;8urIBcq=q6%ZoM1(N4f!5SS@vaeulC^h$LC`dbaU#
za2k0NpTJ7R+<yh^txGoV@>MN+R5UvB23mkXe0L{Gr-SLisC>H<qcXqEOMhOs;$WzT
zy>RsEXX49xXk|I<CuU@y+gveroUOwDDGlS)@S(0OyLk$pzR}C9o8UR*6^UUZ%&QUC
zlXFlRW6^2R$dnMpsJ#k>j=syDXr6D_W!`o(4YsS4n}e@lX+Iu<sd=PU<5oi|x7#L7
zrRev+OS1?<KgdfMx`4UCADD2O(d>}ylCTO(6YWNreeZKe+9rlP2Wt*Io4@=4<BVB3
z*T0XYT`tmj-oZWl@@9G9W7{4A)q2ldpBZHTgmx>>*N6JO-DImMl?Ol(#1=b-+X_%5
zcilYbRxng-+QB(UwT_7__4@R0Li)sRK~cAkjKjCbfR>_jIBb@x0%6{_p>5nX>{P?;
z2%#N=19t+yAA^M`SQzro)`uKhDU#@B;1<#9zeIwe!akn+F{no-+FOQ-;`))Uw2mPT
zIR}4iU+mJ9jK%lDpcBW}^7Gvs_G^ybx!3R!+0)@F4ZDs6_q_rhD`bsWHC6k{d*Re&
zrt1|B8(Ul6=Cp*8UfT<TZyYb>DY!;$Q&IC(t45#*8D7qF%w*6$d!yA)-QvAVq-HwJ
zHnlHGtM>VneSl{6I8fSh<#5^4`Sd{%v0-_=@B|sI#$-=d4MiX5$)OUXARAlBPlqw{
zs1(&M#H*hz;GJ)Gw1Ar45BM_c9FtT7kKPq0*Wp+QTieNYdTXHk%i>qjF$?dz90rVh
zV8F`4!S}fw_jmd6S)qdF_gQ6?cUJ)zB<nwGDEk&hK^VPAth09i2T+gS!;cz&%d*}z
z1A5AtgYT(=U7GAcbKZdqKa%xw=YHW|tbZY>OAQr3aM~8uaIv{9#XqIW@$dhf7u(E_
zo(JFG<10>8G2K{^9bn2YR66na@s1-F4TogdUsjda+*KZ$p$}dOT>b@^&Y7fh_X6DE
zA{-puj-IUYXD)Td)ZUIws^pQ0&O?T{k8dIPR+MU0r=XQd{bWBnBZir2(NATC2nUHr
z9ed6)IBO4m`}Ci*cp<2S6<!a=(I!hw0g&@!-2@Y?BqP}Xoz&r`%gM^EKj~Bn-4{eD
zW$sGZ3*l)mbJ81#*)TOv$-+aA?V4XMeW`U+)<g8UjuOr-c0U7$C8%H$%yaHZ|N8zs
zgrY9qKV;%UfVfK@Q|aEBJL2;L1~0&b+a7S?i!26ztx;m17{aYnXJuuJ`+c*T8hbwc
zff*c?rdY*O7k$VGXN$1za<{%Ba*liro!UaW?GFdlMBT4`iqYjs^}j_6-inyOHG<_6
zA}`;+Zv`t^0=jJQIwYz(DDMX5#g+BJqJb|L^2Sud_H#JqjLCT`vjv8k1S4}V4jI$R
zhsU#k;G1Jfp5Zbbyv9HTo7-8^KJIQnv~upMwP9{yKO8odB?bGf{V;EuP;kb+LGf>n
z-dwX^uRVhQ3kQ4fY9-<xMt_$f4-7T1dXaq1A_$L0C4BTl5*nj(hUAA&@#<f00+&RQ
z*X^#&nL(lE0Ya%s>zVLtNzMObG$gOJsZF{2Zuu?4wbUSSSv323-i+T<LB%`8N}Zud
ze%z_q4Sy?T;uTCVHdi`_r5D+JE=?_8(20E6V?xPiTj8!xp^L`80xH~BoxfJ!b5;_x
znFQZRsr0wD1|?Q0SlBV?yCC>L-ss)dJj&F2<!uk)P!Lw30ocXmmFR&rdQOPS(L}ql
zto+2d)!dJ^mO{es6D(ki0jqnaZ=pH7569YfST{ys#*0IfV_YkWI;O<NA?$oGco*-U
zU(<gUiYHl-D`o#9;piABL`MJ418MdcsqH>7>pzrQegEU)^f8kNHSU<^DW+Sd!=~bu
z%Fos9)|Go*+|6Uj`i|wUXsrn6^wkZ`E1yK>Sf6y!MLE;0A#Ee0+`xX>4<Ic0co7*u
zs3fc^dtTyujvUht)>OKJd6QZNp1jWL13!vv?!*hqGf5~Ja6J&xP-xuezp^6oxmiPb
z!b?~n;zHlidzq7ZiVLqu*hjO-MmLmO$g+e8Z?5E9B)wio@outFJiQlhVDK!|_IL_w
zQ?hK<|HkIC_PcaCs~g5#tFoAP4l_i3pu?$Sx3{DH_yXXb+>sIl&b=no!`lkv@Vv)X
zDzt~gr^78q4e=H_eYeZ;(JxznH_20b)ffbaE?UMQZo=j`>GB7NOyZKVF~8T2*AHg1
z^w<lyY<0EdXe^q91m_fm!XDfChaZZSA_OfTn;>gH0b3r*gOQU%N+2k~5xqu&`8#AD
zO0!4&%L2e@jCZSDFz!XHDiHIJ$14`Ae2GP0g7Ci7=_-KEhWSGZPriF(38Q{vX689&
zq`h)S&jnh?!QTI&*y}L%*7cUI-^W0iPponj0>5*QE!D0@)_7ihbirB{*GX>+&iT27
zAs|x~NF$?>4y#9n_$Jfjqvpj*Ofvak)m4KW`q))y!<LT6=00mlQYSyh`e7?50jw9|
z82?N<y`vfPNIcs0+K<As72&1>a3b@>AXB*SQ^m-MuPfb4>{@LUDae|b37loy3ZsjA
zVNc#}Hj&%U5ZKsh@zsBHd+6FBB#&BK(5by@sWL@vWGV!K$&x~}OY9Wx02OYCO<9Ee
zq&k%T%8a~LDed-`K*TUJjWC8hW1!tQ9rJ6rakY!dH0Zj<H1iplaxM;*&{SU{of~-I
z);D>b$qO-Ulb{)#M@fo0^aVvLUM#ATJutC(?h!*g*m`A+!3vPDQt?Uv(;M7|L?>}q
zaJj9jtsPxv$kJg7wGUNoL^$71L<6C8?aV>G`IDEl^Zd5Xj${-xBcI{a1i4ooRq?;u
zNa*(-!C&?;3=28TVs;B6^f!+>X05iXIq-?Sibmg%xzzV?Ao*A=ov7I?!36l_EK;d5
zW3Jn*#dOJJZqEPTqf3ke(n**6zSn7W=T(v`>F2MTVv9wyb*Vg1wjr!u<*umv;21<^
zE4z@b|HP@h)-XMSSowVAXYZ9TjsPoN<fp?yqR95WX9<Z>xCY_O{+=ws7>iS=jb3aL
zm(^}tO8E?Zy)0NVFvm_gi9FDpba}wi8<XleAv&Z})^Em@KJN*4Kjbd2f4ghfz?5e|
z?n8nTIQN1F4p+vfr{Y!mp7$o{`kfH9&w@CjN;%?Ris{c+ufK{nAo=28NuQ9-?}bw&
zQu(HRr-`5J(VKI{i>qY*3GJH=^D*iwJ#?3D&M)4%*;d0Lo4r2s-)ATTSZ?OG=pNlE
ziTcjDN27JY^1@S;6-a=V&{~YnQV2NP{FXcZkt=bZi&cn#P@ce;G8Ul}$FZe7j)_(x
z%oiZN2Xc&$c3b=dVw}yV0ms%;h9`A3zCa<sapbMC8M+xidDMMrB$iS`mA&*Ir`-L>
zw7^9;uJW21-;ffe*A>}%HLJznHPdwEair5Ra5;z?RW4O^slA5vXyoe)d;%=RjMs8F
z$UVV&d9&(BY^>g#&-HF+a$wOVw>wabRl%SVykdHPVo&G%U<hRXiyFh9o&XWcVrazd
zAL?C&x{xRTYZ}sQ+B3J-y_MRmg>f)#;|#9Nbu3mI*>i&oR5$|Jx|OjCarnk6Qy7NI
z7%uQkJXu+hF<~P5XZz2e7*Q>eHIg8I9X{f0ISs4Gw()Cm4|A;;jxk1#*tSr;6(~+s
z(ALANsf>-^aH|_Uayg?z6r}FuYf{P7x2GxznwNWn;Z2uXgA2@yJARy3lX+D>oF!-A
z&4DwMmb4#0)>2~$0``6Gvp6Ph^4qa)h~S9ISuBqilX$;<9CTK)<t+k`Y1+^|U?W1e
zca?@on53>^LLCy8=5U^nw?!4*YW4pSH02*bd(Y*>LrU89<-aeOixL`$EzXkTY4d*{
zie+;N)VTTkqapX3pFX7USR__Kt>SCf@Tj~AYULfHV!gHHu?8!8!j9ubm~Pb%+)QeY
z_rjbnY@Hd@inS^+mRu0Dkb`r4Q}C%N4Y9g$`^PZs#%%j57@|XYz*Ws#=xB0nk^*lH
zwu@5N!|h?e^;PE{#s|SQ=q|JCXd+Y{&k+`|TY3wxlu(8eb?durV2Qk&s}}0QCf0@^
z4u4RaQ13;`b-j--4>Y#pi@=-H4+QW>icW`JzwPZuDfgQL6Pm92A5Olz)c4>Q>@MaN
zgumrE3UB`AVM?BWZuWH|?jL?|=l_I)vpCe3EWO%}I+=fyX?zum-T*Ca8~jAy7$qZR
z??A?-&G*sVc00kFv40RkpK1YwSvFbic}hDg-q|!{d>8b`_rcl+(ME8FDZZoc=?$r7
z7Fd~6^E!7e=Y)-gnhoJBqpZt~SOwNEc^03TP#uTgt6pJ!4v#1jQ-8v6Ey!G9_1dt3
zR*hT|3`sSm(em3un!CMGSRdOS7iVS%{Pxm8-ShM8TfCwZiL$Rx)K(B%?&6tDmFm8w
z?O3R*f}%nj-?c0(ztb@HsQB)^_n$3bgEwgN>y<IJDcy?vK|=~Sq}uOFQKu@8M=zsC
zDD4ff)T&dTB(0W8*L1hweVMqreX`zV>)0s3kZH0vP4laxGxG~S9^us2h^O60E&t3n
zZ3M^oK~eE&Dbp?YpEm-2>9HXCQ-?YafIXqpztJeSG<VW=hV&ficjKoLeK#L}VHNCO
z@P%n<$uUoPU!Gz<AR?P>Oi5LEXXeXqv7nsUxgFSpS4oC-fXy~%wmM~GlH?VfHG1qg
zhasJg9yQ44fJW}kLdi<adZF=FLPm7T-6EMqpI_kI<FL!}x?S}~c(7*1De!~Ad(vYu
zu1Gp1WtnYBVw!w0>^)t!?y=w1i}zhhNn#>HZRzT(r>vq25UbEM(@N7ajywsEK$jM|
zdaXm0z+=2tY0q>qZt!k%06T-+(i3{tO-O6ZM=CVj$O7YIan-GXEqamy%?BY!E~LKy
z{;MydSF^p@Eg8-YGuX1<8`rsBfSk7PQk$yEeW3goTH9YD&`%7t^@hmgR>s1NZ7`Tc
z&dOGL&Mgw@3lbILcL!PF7nI=?D74^`R*_Y}2|CW3ts%aiz=kP)!yN1<g^-?`+XY#W
zMDbg@UO(aT;zcP8pZS3lhs*`B$z?z|D1A`+@->lM?%tBj39R$uQJ{@pnnzMdV{A~C
z-qN_AF}hn$GlrK`ke<NovYC*~&hCPNNZ+}c>}^z%p$WIKWf*nGZ7|9#6gi(!%E=S{
zCgH}9_9Sbs*dykjS28c5Eww0)erKo#z92i(D@M#*-qtrX`zFEKD^1mXDYcKs^6-;B
zijWu4T*iR|Ou`+tj(5l}#PQwazDr+esVraop;EGdQ-S8P3Vbq%dXZZ-BZKdsVpU6Q
zdKsPm*gro-EJILK9GdLuLCcmRUpMgpP1rF(?GVM?29&5`;p){X%qTHtrb}KRX<{Gd
zDksx-hWrSx%CC3KX}Xt61CJD#)0R51meXH1PKB+atnwCeo;Vm>YIKHQS0gUq@`aYD
z=$9ELL)Er{_`dTzsPT;Oi*72O*vj$Th1<d*K`DAt5vqg*3u54|9+<w{vWD?<bn1Ul
zB$ckQQhct7cfg%09>>A?@G(bMq4{hiZnsm4g#^DlE}LHGFz4?glBlgTma51}mUoqE
za*aL2N-;aSH5iiHG`s0y@0XO>osbf3Y=0C3S;i&uY~F07{(T~DOWVxc6Lb0?N7gnw
zvN*Z-6bnF6X}+qV7YoZPO2Qzg#4+`&V#Dd|11P8yQ6(D+<)qlYbJVYg6*~Sm$TgA|
zT_98sre-2Y%PNWZLs_aZEYTDHBTaWtPB*V=;H+_i*Pt5Lj(eR(dHr1&l?5k~{SR+&
z;yPEfJ(8~k6-|`Iq4vw^uan~^3+xLa=**@jy`x`AwBD*@NFHhS240&C&TqjHA@=N+
zLy(@9nscYi|7`sEyLVMSTwQ1kcS}Bi^`6ew1`YL>U!dN|?|K2O%~vF+Q*q03wuxL`
z-SVAb;}K^{K!ZJ)H4u`7sM9Al3RF<i2UcNv&RvKSkY6q(7v~?(@2k_u9~&|QA2oMH
z(!WtHdp2m~r3`~(f^}>ikspnffo0it!9|xX@l0Xcs1h&ZvpB|s;lk<mR}30j7j>J}
z{mzq>tG3@FrHz>;$NjUB@+*lp;8$jGxRCd`BxO0IsYB8d<3EKTn?05}=#Na;5^`*T
zoJk)T$R9RPm~Xrgv|_|?P&ES6BGo~5IaXtuS4i}DYs)#JGBEZY=i4vDiZz7P^;L@W
zSV=Kc@L4hZ?r|eJreREYgv+D)qeI_{uMWvuo%SZvyr(Uehn+<@s9-$eLGh1oWp^vP
zY4~u!19EPA1I!2KWHLb!J=g2SeN7Ucx{Y?L@a{zp=Z|-|4hEU|90@eLm+HR)RK7u-
z+Dhg07H^3-l6#i}Ns0lLf`zDpVtL{D=i`A3b01k*fhijc3=P{<V&n-Hoe|@)ibJz$
zZ{56<Z-3iRwK7+X3)!JRnc;`Ac(5!RPdcAJ_J`GPMfb23CPV+jkl@(jnZ$mIag;vS
z&se5Q6qMh1IyZRr(g042u5DMH!&TubaJqSIS4^iMHBHa=a|KAmoFhj2t(?S7a&mGd
zR<Cz-l?Rf$$Z9olf5tzl9^C&EROj_ghWBr66vQcD*CwY(#_#y#KEvOhVU}C`%lR&N
z=2gXfIFGCgoLS>_iBQt_bnC~j!sd~J6_QCydj+g=Ed^ra1Bz-YjA7{xfG3Bi&y#-I
z8owELQJmQ41%@!nQ6uQWH9Nl7<!79WxWQ$~s$m?Tan_2Amu}rlO8X0lJpIbFW>i%1
zfY*~tmdr3V*rHq(b3zr(_tp&Q%}GuV<al%w(M95_RK>IrMKa2Pj#nXVcEa(kx_$>)
zCW6!2<W5=W;{Frvx5m^}^vL|sqyfIyJtZl3L29THA`?<$t33!S_OEZ3)2pi0KO%b>
zyLHeW?c1(K@a5)jZnrq}F%u;*e~tELU)a6ha0Xn5oziYIA0zhb&gm6>Z{G4(kJiP*
zhH*G#^8@s-2?ssK--_TI-~R63wLx<EW5W$g>v{WE1VH;72iLD<RgS#*M^+vZ=!b?b
zwv*^4JRZXRW84C9CSy8l!1MoTGCX*iPQi?RNv7I=hQ)ErH2JLK%VG|-PglPd=WhL0
zSdmhCPN9pDo8Q)GuZ;;C=4~>+B@vsF3;%0%8$X9FO*Urf%WQQmF!<!-3r|-O&&}D*
zw#P$QL07ntA27;$s86F*t3mx}{Lx-XX$8H0;m4hV(6Eo=gJlZ4POch=Rp4tg<oJNQ
zJ#wgz4c;15P+0)3q_8*ItPze34YoZ)T4K=x^2Q5;tqG3p6HfWUem%x>geU9vG}jD+
z=!q5&I#<utwp$i+!0%0cN@=4I<96a%%rj9Ep(jeZ6-7itVK@)IDaH=39v4`WXx)dm
z#bX(txL>(SXgnH29mLm4Y2PxZ3CGH|JrtxhdcKzA4LS%>7S(>Y)rPz;{*Iq4vHQCo
z(qMH%*NY=ea)(cCiR#RzX0Lt_z-el=zMuc)MV(Kb%;kUY@Q`8-U^{n$BCEtu8~9CH
zaO_kzKSiQ88$OGxwHwRL3o|&YL|?m%riC?K?2_uFHk0TP%+W2KsfOuDt9P^p+Q)+F
zFJtykE&4o90@tf+h}%mnO2iG3h*7}I;W(`L*pGrA(`q!eZfQU9rVvg>xpq@eDCwxH
zG{fp6{>?LSXv8nJAR{5xq}xKa)wVGp9G%VjWGu>{R1@E)ZKTA$@~b6M)p&g+_NFW2
zKu=bQv)viUx;#fnOqYvqI1m~7<ftc04wqXAzY94jhg-trrrhC{)g5~WF%*^5vaA?6
z>Hg{A<m%|yk1mO}4WEl-FCj@;?+is9BzQeBPge`m^hz8h3sZX+CXC8P64;HS>2H2a
znT%EhA`amY723UoVpVReJB-QG?sZgdw|n=N4BMaA>Rr<FEitg>qP&c*l&zK5g<`cj
z_};DhQc(9GMuzs^{Y_HH;X$kdoBwW0sFwWJq;uCFdCs(GCeP|c?XK9O=Ty4}btHAF
zY@H$|&A+vMs4pjp&1fZy^ijNWJ%rIx{@bYnRf7w|(H7HU7;U?Z{t03rxVe)yj`nsJ
z<e8+#mhqH&K920vM)%1|rOs{jiv+B<9E8myP0Bt9ajHkH3R78^NE7K(aPzPnSS`NC
zqcx39DLV!Fq6GW|^8VHC)3zyS()eF6!Ru8Q`^h)4o8-!JsC}ikInxmz%ZR;zuqkiu
zWwk6OB`^K)Tg{N0f+lPQ(h90IuY4$EtY}o254L|a^$C#611_`M0j<^f4dXAt<z&VD
zXZNboML@=jLE4~QQ;-K6BKWc;``pHBW8RTt8BwTiH9CyEO|+8UV@zIq6i;n65}b54
ztEUOMeKM&YviI1@cfIkp$ahIiag`|y48jl)rb8<(WZ17IKlZvPkDf(V+yJ9EhJNFT
zkuy-4I*@JULfuzBwlIsxZ-x(5araw1`gjRTb`rC{CvWh~NkemOOi_4obMGdwPHu+s
z<y3YaF(agGMq{_`8}4ag9-JH=IT@k46;+=^^DEZhEFT+~;gl}P1GcYERpk0CsS`L{
zQ)SGxzUOKzFjo_zOSzLnpS0Af|6pdL!%19_BqGq_=1U8H`H=Bx%0Y%=8~F5P?V2m&
z#W;((*#jmTNZ$XrzANFrnks#f%42nWmAv*$F`+|QzMSb5;YLc`k*FyfIBBmFMfI33
zGJogA-Nb-c&gnji-kRK<+_DHu^W?mIPL{~(>YIrgd{n7B_*K{b7|rgBtMAdw|0u3&
zK1SjcWj4xmb3Olav(%;R$Ics%h?Fy7e7IJ(mLWY_maF=y^xjm?F~(LHa$P7JvuYq5
z*f|(-o!PI(1<N3}x(^sYHI(@;Az}@}-#p{x&x-YMJ>zb3IA@w)BnA5`=58+&1(n32
ziv+1Kc-4@B&pJt6CNG&g#`A4%Q6#0Wh<(u`h5=!%l(57avI}e`tSmnT2Nbc#+^la+
z)1Uu>ah;5whnq~Cd%bYksiKjySqZZ$(Mh!pc18qdp{K+`zF6YU$A)Vvi)>^J)B}}@
z&@NudRr(^FH8zDbnS`~L)eAdzEJn&66)GU8@&ngDk?FHJeswsUuG>^tRAv}kpzFk(
zRB8&G97$CF^`uP1cCg5h^2MH|G-5>+rap7>YiyRqz2f%G$F~)Z^Z1e}#QbN&v}v_S
z&eFT695z8Tb$H89?BW60h*48=%IY_$psDw%Tu2VSNlgTDK6*t*fb;S6Z9ZN0cl&_q
zQ+R5{vogIpWUh}9Y|>rw^dfJ)dZ+33ia`64F0yr3JVVI(+UJCK>ZERTg0@*1XE452
z1x6J`XQPAPgZ^PheZYq9)J-yxZZbYedT9X?#<Y)|Pgjw2<_=H=37aFQuFTVNKy__)
zGK@6ELmwAW%gR{Ye#Z6q9=d~w*Csl24J39!6nO^VKk=uN-Y*Kv-Xh2Nce){P<(;C`
z>VPA*RS3QTmBTKDE9YuqPbfWIfeLpe*ekDiTHGyJhhu9gnR=e_yqG^g>@82IKU3z1
zjo|Nsi%mrx_la`hyeQKO?GreQlG78LmVt~9m|dPaPcHnlo(KI_*07++I`3Dx$Q^E_
zGgc5?9IpG6d$s`gcY3;^nTE9ao6VO~fw+94&!v}GCM<<cay0mEOL%^KAIqwUqeHR5
zTQl53H4~^CJTi{i$19Q3n056WmB(My=lnsRvoLuMcNcH##gDHaj8=7(EHz+Lzk|X`
zOiQ=hUxu{#!Mq1{7`^#pjG`Cvl&6akQ9xiguta>A;g+$B(ZXU2*vN?Ee1E-pj-yQe
z*iIl+<D$?U;npr-R@`8bqi5kQAyz>6Ogc6xaX*?k<M;vg?s<207hfmrrZ09Qy3S69
zVBIX#l7(T`^4Wb}8wOUlDB5RS*p=rMgfyJ%rf*xOojhN`k5rrCie}AGMyn7qOb+fI
zr5beTW{uD{EMLeLGGkUzTHlbMso&T=Cg{7F>f{4)mEVw4Juw)wdh9Q<qF)(`McV<f
zR~tO;F(;7)cKJ+ih%FvgR57y##qV@qZulxTvm$|X-Ht;8P_J%%wYC~U?lW9pt+)T<
zpJSAZ5c6SbI4Mypbw!qm2#+Hh_g1c|Go`a0pU#Z4Z9!YPG&g%^)Z%WCw})UYq_P<v
zLbkEQ@e7%iq!(5Em2V0@4qQ2($X0NcRHD+VtoWC%fvAC2uTE>C+xhY6e!5ycdqQ$;
zM`j^;y&@OJ5EM@>X&klV^I{Hb_+P^<iC8nbN3-<F6tQvzKUu4a5@kG9_0^aQ%!zwC
zeje-oip_T7UJ9xWOiY3>ksB&qNiSckpq}sxgzu^~+Mr&$DhkGHM;%#R)Nry}M&j}*
zUFFdM{*K~-K=((zuN&eEmt7PF1of~VKJ$P6^);*6$kqABIp=Q6I-;n$no*%2-$h@o
z$b3B?J6MEbb??LRhorDx&B9P_<h;Z;G<%L9l4!Gkad8M^8ht-Gs3sT~$nkaLGSa*4
z3a91RfViQOR_<^1#P89&zG^7$v>*8K`UpNpqw&ybAF=spS@jytg`?w$&BAj;HgOR}
zS%=OR#B*Hdv1O3n^6g8)uV2S59`YqdIkmi2hpE72F6?|VySHVcI^m#tIP6(p$vB6+
zyxme0OOc#H`G%gwiZ}%Qobk`xs}aAe>K)6UeV2=0)LB39#b|&3=*_;T(AQ8A*QIA;
zm}O<8q?`zfEHGIRE{`xFFP~Y6`6WrO&IaZrgSg_e(qm1aOsRLB6<3>o%R(TD%=(+`
zK9{)lS2a>;4L@bVXd;qDy%weS`<3tJqb^4qHMxA4w`~2e)D|7+xF2MzTzkYa<5zoR
zp(gaFgbt$;b+kI39sB=L_uuhY|Iz<A9xfSWuOzyxY$B9>5!u<YGs_ktg^<hMdnBWf
zy^732wo=KaNJd6xlp^2rY`@;`_viNe`}bd6m+LtmkMlT>^El^z?sx5-3(q+fo*S>6
zSv$!VKyA$-&`GS|S?^`y2^rI<)2c=>t8ttrqqI92JnzrsG6_6D>kxJ+7V;^)3_CFU
z*#@A(ePWrZ$ILx(A5a!tLwee5Bx(+gVs?VH&EoFl;an$jWrYz5d)O_N<$Bitm4NkK
zxuQYf#lO#17HvnLkQ}Mu3SGRxgGlkhy!D(~t6av0_o0P0Ve`0wqAb&%=mYptxwgB;
zG>;gDslv`5XL*FKtvfMdKyIS*+&@lFZsOq5CjweSg@+{+ocFd3Ig|?QD7R${s5j}R
z#UFP9KwF>IP|sG-sU}&d;MH~6X4OFGZ{R$9n@)T*^y1OAw0V5RvZ;m_`-7GmerCs9
zC#P<9-QdS}g4)TnG)4cqZ}R4V)QQQxTt3Tp?msf{KPL~vMX1|Om0Z$yW9srAbb}F-
z*{Vs|EPesRB9#(OVL006xYtxbwkDhFGg67=*tM6Dk6wbts#(vDZZ$7iT)~|#+@7aA
zDtrS6GRQ_R5*$84-uSgbi~JSwIx!C5?!N{58sgR$;d#FQaV%${2<_I^GlbINQ+xN7
zr<MJO$Ai7Xg<l&cJqwFoE2nvP)cerJbOfDOQL)wT`42%^9VIr&v2FbRQ>W)DJDGY7
zT_7;f+^TpsMu1V0>pQ7H-%3NKLh)4B|70?^9*OdTx@DJ-{Iy75AL)%MDFEZ$>PZ%T
zFmv#U2xxfcb<ye%BvH~TI45IzD6<nea<K~Ryl(VDWZqQx;TA!Z^mY<vbK?76!OEU$
zbxL1rrOHpsA8@gx9t`Ity1lfSQ&sBQefjn(lR(#2ykgPRmx)h|<`=iWF{M1!4qQcv
zo>*7qyuQ*AwljU>qYdM%AmqT2YJxZwuXh%4(BAKE!=D)WLSmxcv)tVIPBo(^zBE{H
zGzT8j5$O(&?BuaO&V<#XUz<CYM83bk?s}Lhn9WSKMb9OGJEYA>()O;80(;m|!gk22
zF&#j{;@Gwz0*76abxC;<e|6cR{ff7JFW&!XF=aU**qP{uKXb&Im0R8HWD*0*O2G^9
zp25`A_6BWPTwcOq$BLM5rSWzPEm?)-6W$L3aHA))4WOhI%!z$3Z$Tc}X#Mp5Y0{Hn
zP}ENej1?m9m{?7kQk=EaPWFwh;KW!}Y1vd$zbgFbMdTXx<ZhWP|M-E#D*0ru;4Afz
zrGF|ou+14_7c(h9k*RVlO%GIx-M@#b<usNte95{#m&V<XdDqZ=;&$%02RxrhXm8;3
zuBYz8ImncXvR&3ki+}hs^LYNF{fGlucGvGz(eaSnO2P9vUL`=-+M>V9p|7`TIN#85
zO<Q^-wqjkC@VuVN$h~OsqF0}#j#EQ1NTXJf{L@ok#gVz<*F&E#ueN6h2)x8TT48Wv
zB*^;R$BV>?DhhD^;SDrbVG+ec&aawv-xoCeh{q)x?|}NnJDdoDo0Lpb<>8dcLXvFL
z<stTj?RHWJi%#Z`-t%u-$iIEdg_-OvC!UuV+ZtNF{V5)MchngKCpG+j!AtFj+0Q`w
z-8R-gIz`!<q*G2|##roRI~#*M)65t^R#QZ25-P|!$#X~jBO_?;su<wY-zvq!+8zV2
zVGVE1h~+{Fb|fT&t6e)yn=Fq@A1}QnyT;Oz5;xgfX3^jdRc3<W-IMxbuSoaS<oL&H
zzM4bF8uRs}mlVq_hi{X8?Ne}8sw^pNw*Lm2Ga62esrjVzAxG1;b5{~(ar7zba2pN~
za(-WE<We*auO2c|Eyky!oVGYb6vKE+x6<`?>~r${-Ti<U#Iyt~9{r?^@pL%473&@p
z^h9DYxEcEwtFKzi+<Q*1k-@1tDe>Lq1Osm0?Uch47YH9m)IZ5)8jdW@?`WQ<q`+xu
z3f>=|)gBccJZ~&Uy2ID*vAdVZPBfpkBt59#H#z?|bUYF;ht~9P*QOSzTx1_M*9xf-
z4dNslEQra)^mv$kX<2#r`g|Sf^&eB1T}JMox&(A&;wN;!b1upE=Mkuw#7zVH?&EH#
zuT$8cjYP`$7I)+g$rZg$$x(M$7uMgpIw)+u9fl2h`L3NwCooNcIjJDpJu5ggh%-vh
z-L;iz5hFamc)EralUE1$gJ9-`3?TeUD_RlOkCPpnn-J#z*vn6U`4}4o(8RrN6^k(c
zct$bF8>)!xpVCN@0uuB6zTyo7dPOF-j1!dyDtbM<)MK}(^Y(q%c&)iS#itV3BSfM{
zhO0Gk6)8&v6__+Gn{IwNMeyqRe6&M{YvW~QQ5vf1#n-(Zgk6ti!zR#NmFTIlO-s2p
z!}sac<AFh4ncgdgainQUZSx^763vR%q*Ahk?_}#97q->Jr3JLhC6&b`37uz{8{5li
z6VSAkL!xAWIyh0>EHZiRZmOrT1HIXmg_~h{hnE_K`+HBW$m}$F8ZU`?PyDm17f|DF
zafy@;jMgzCCr*C~&Z9WG@x(sT%*&P{1<^@W@)SYD)It~92-qA;c&$26vX+du7B&n9
zV$BI|k>k?hD&VXWuj4K#oxFX^kbwMJ8*uuzQr<!xLaDQpO3d$S$&d$=7nAz|e0}KM
z)i4E8&r^+TkrxX*KD=a3448@!Ddi5svKNcb9};{;9hS=O)~x8;#ywM7GsfMnxmT%6
z-$fBOuBJ+GT#2($P9gRxAkuUP56SD^CpX+tG9W(n0FSqsvPy!K_xQB0I@c81L#xkH
z5qr7*QcQumG*5wr!qu~_;_2rOD6r&;<h;Ne&ZOgFnR-z4G2gh<Ck!g+Pb{N^5>!r=
zZHR>XJt@Ka!o-<=L-4K}FEcg!i75$B$ho7~dBa|gS!G=9S6;9?b@m-#h)Bl2X3!L$
zJkPKmj^s&y<8COBg8!@b?ZO{1FflWF<KQ%vr8ZF(@gUdS5o?ZcReVu`${Un=^wvZu
zf`G+MKsu6!kzdb#5wBSAMsquCt<B*W3+*h&>oVrgp-4@rfIEp#YB}=zG4JKdZb-5m
z*=`GeR-n|Hc(de^(viZ@Zu^g~wXNi$dd011atee-*}be<@F)&*g;P@NaWYS*dr;(;
zbmj<UJR-crr*+8c^S#1WmMpZm7@3j=;i}vWm$f9D@N7A>OUxjpMuSral}vq|N9Ih@
ziIwK$T3jiHM;$H}n0ZUnj^7}Zc&uaK5_+un@yP-avq-%DMAsyS8)+C{Goh*Wn8gN7
zjOJO|yC*rkD6&^&b8f5)u7sd^G*$#hC?p}ielDI;yEjx7y-2g_ESJnu74)*2!s3ll
zU-$dp1uG=o!ADV0&nU~M(c$|8gB6L{Aq-Uuv<|#VPGN8}V!x1Rl-}FtUYXQ##Z$Se
zA_TGXrJfY(^XiTG%!KQT`#W8fy0$DIl+IwasoHr_MDj#P=N|zL=S)oK=zin}89%X$
z6t{LHjo(|_6g5Dq)V}(a_6&!;DgKaf5H55G5L(lvq^F=R#NKBy(DMW0AvlJC3EIZ>
zkdG4+idXYU)-eGGlPRY*r)uNa2WVh8atL(@D`Nc&C}xrlAD3=ynP)p6oL@i6{=Fdx
zH{R^+9sO2Xxy2`)wWP$pF{5*{SVs}B?f_i+4RT7s3)&LZT2Zq=6nKk;VIy(Xp%5nx
z;m5(|mQTpohWIvCbRQJponCE*rn;jgs+<I_5&cgRHLttjkjIX5y4KBSx4H;0%VllO
zGVMXVE@-B{(UFpfr?tMxe0uZWapeLuwK1^~|4;b;Q$ox^q`=a0wo)6#dj*`IGl}L?
zKU(&Z5>jAXrp*W0a=ExRQv*WGhgpVBgaP8K#NF_u%K#j=di_`v1%%lrIUPpBN-=6R
zZpW4awefiH`<O3ei?#*QxGBA&=E8Z`Gg5&f596*bCDgCRuS+C}bN&?ZbsK<P8VI+*
zLxU23YNB3QiuzIM{_wOUgLlCMY}VZG<@bo0j4l<4j765q-V1d1^cO%0P{c?ruVtMh
zTr&(L*Zr{BEQ;ENdhRh&*|vm|Ln5cJGFd4`!J!^quib9t8h<&0)XH+|<9*sEMXGzR
zh0ojy|B^>|vrFCgpq_@R@3w;QqI)5oEcftGia`arK{x^vClW}lB_)&kq*+w!ym)hX
zNN{84r}_N|;<Az6i9<Q>t$u9sKm_r6K<kX2kQ=*Qz{96ji5NXD8P%?*z>9hEqx1=q
zQ7#&r3=<wc$;<fV2Kd(0ZQ09&Ko%Igt(=xxqJ76Q`Jvd2Pah5{h|z7I?aNl8F+7`-
z|CPcJ+o~TfVWD$a_NW!A1%H87V<&ph!PStGRD?s5&dsSVnf!}3&FTB<i~X5h>n12;
z_i-18EoR)jt8)|6FBp`Ip!%1TDMW0bbSg1oHa^GBioL8WL1cUtLAaf0>K$gA$)aBI
zc{FWp;>vhOg4okcXh!?eWvSHnUw4K!DIz{bVF#N~G*SYQwHIZh@3mzL*{Jvn9M|8e
z*t+?1SNNIz<v;*<_lrNeJ8cn~01PPLLoM*|812=4K`m0@gjRz-(w!>h2k?fkvcK^X
z(R)AWz*+{XxQn7U<h!zm7SD8EPG-}L6*_jdJMaByeY<0VSA@UU8vuCYg<-4c;+G2U
zMXSVA#%nxhAb!5XDx(j?EapUU6+wN1)ES!udPQ9)5-Y0Y%^l@SyduYS?r0W7%5W0X
zWLSaZN%iP5yFC5;7BuiRjXG6t9jr<wF|58?g6j*(js6#BODmJYDW@qZ4O*!E)HjD8
zFH>qvBi+2q!PuxUo1~!KCEzweudN66rF4}6`;~y<KuR(+aX4fA9sG<<7mqGxb~GKb
z^xXZj{_bt(_1jpd;tP}WHow|LfrA1eY)Hbqayk-l6x|#s%9J80PsS+z@Ss$E#C9l#
zPuakSnOTvBylnK@SZpxM2`++nh0P73DLu*W$xGG+m&t4iEH@Asvu1#32sS7&;Bc!7
zPWu}K4Z@3$mRwh66Am)O+K(7ck<FKFhGYj%i9uXKJn&_eIc^Z?-%2N8f=rZHbSgIL
zbIOCaK*IgdC{Nko=v+osQTM~M<++4PQQh;5mZqY|xgn{io#4h>EHw1#=3wzBZPZ)V
z7)i>Towix4hZbEE$1870j6b66t;|||1NvuG3UZbb(x+gXa8T+A(K}bNIDhjLz~*Ds
z%Cu%#Jlj4q)9eFeOUK3jqeHK%>Djo7`bP_2uRM9>Th8}>%Kg-z9WoMAN0^G`qZ1oG
zg)4IIfSZEOc|;%O{o6OD-m=-rY`m%5cH~P1!&L!QBy#Z>3mP5+!x-h{T4&L9GNi&$
zLJAD<)V##<CLLQUh8z-SXbITjEh9&<rm+;{O+YLwP-Lhie7$Aba1>I5y6oe+c!2P;
zU)$V+ot83`va1&ASf^V)>*KsLRi>nsq?t&trpfVrZhFf!nbXnvwo~Qf#0pr-SLAR-
zLVZe@KT)z|O>ZGAO?YRR{C3w5jwih(4+s$`ZM;1R5M<r7h;WIEp)>E#EQ#O$A~6B#
z1=-^B3Q;TU#*UQ=V{A)oNqG0&rUBio@rBgpESs07pYg<yD#aVcDy&TH64Mavi^=(4
zJ)A$h%Nz@U;OhwMA5#=i;94i^w~-}}&F4`Rpl~;)OXQ8)A5tU~AiZw)xWDAt-w=`K
zW){%Nr|H%3fb$f|W9+ez)NaaK+O0yRX_z|kVUwgCBDOecHs0PdXp0!JxzvO5wmyyV
zOV*(&+)<tgS_&m;l{EwzsQSBXj*{Ok!{%zw>nu%u1^}3_Udfx|bVP)8O4<^S={OY>
z+7L-499D|UxN6;b4-vt$QhggP8LiL@fHJuq?!&eWt#S0B7n1HV#;;Q_IM?QQau{IE
zrCKA{;vTH4vYUMn=J@ia2>_~cRIZi+l+ihTPGo{0tav58H3fUi^@?Bnxwe0f2Y@4B
z2<VXvm~$TYM%oO}xIl<=b<D{9_=(rSjg~0ruICy$a*jXfw7m`&zsR0xF0XY$*ORlP
zVr)ng7Fs{8RfBfx3|d($ncGdBZ>Ul`CAJRAxGtIt9UjVVqEK}8x}Y~~6|Y{~K`cjD
zTteWIcThe)Q-uH}rL(X;>M^5;Q86gv6JVXej}%HArD8dK!N@KzTYCZ^hH_SdL=iNY
z>ZFSG%W+tX*kbKYA@rzM+hyW+JuVkbQf?sqf1DkM=vN5T0z<HUkFz8B6us8_Cq^I_
z_f!?_`@|MneEU>i;K}(Q(8G@L3F)Q6?{}`fr4LSEw+J~*Pr3HOgRK1i`d>N$2+eKn
z5dZs$=F0xLY(f-L!Pnh+`ZT|*fTgHDl%l`^1-Xu`)K#3!6jF$6{rc}~WL@Fc_K_3K
z*|r|rV)9<5$X|JVPOb67470mcy!JPqy0&DD)mg62RQIU7cwZ>>4v|{(aR@UFt-F>a
ztCSv{FSL8^^}d7c*!=Vt0bCXxg;wr&>AKV0-2<+nT|ne5wS4QDE@t36px(Qcu|;a`
z$s^WnMAJ&o6#N8<zJW6nU7QAzr}KS4*DEjYuHS@^uMh|rNc$Zd2mJEUhs>+mo*k1h
z2ZK~5n%(t3^A*XNFG-?9PLJ8%vj)@!d(D`5_)_Cl0R<v&IBOGReFe>;F|ryDlXEe@
zn*d>0WRYP(e_0tLq>jKiD)`S|tjqLl-=23nm#t5TxrB*z2C|}-7eW`FusyDacI$rZ
zS+#G9>m2yU-d1c#2OTIp$eik6GU`j75i5tXif0itE;AZ8?^ZqPBak!&Ey7<0Zb&|t
z(%?z~C5oUKNOn3qyF1l9ZZu`K*zqNo?YJT7#Pq^a$zX>aX}{z;q1_AOCZ`L}-6}(R
zi9f3H6qqQORjmT#XCUY54sdL+vrz#~bG7v`ziEALC)#wUZ1c|We1^C3v4UBm{B8#+
zX|q0Yr}e6Nz|E;OUG`vNsew}QYq#|Ch%6ru^TGuwd;vN1i%OBWH-`jN5WvVuwMMV?
zkDkAiwLkXj8O5pL9kQKJIHT|d90(KOf3mxAuty-?jmztb6hJ{9();X#K|Z=V&T}qX
zg!zC4DkmVxKlEvSM@(y?A8}$8F(JKr(%@qw{?8Cz&Va(PfCQemdq+KRx#VT0YPTXS
z(E(Gzse8~|nf}`6`BJ!nA9T$IPS?%h*|IZ+tIf`447@ukK8den)MX$|<#VB#23Kc`
zvly%DRVtTyi6+m!e8PtRg>I-S;l)C|JkO|ah$`f3bu%~20l6Nk*iVgDr}83|MjoyS
z@YsI~WsUA);`V0nrZzkLuI`X*q|pVo5<=IVPs!umVDQIvm!6Rqq`VZn^(ge_+HfzU
z^r_e%{FlE`=tlx}Vf%R=6BPr@IkE>XHo_(`_mLs!8PsJmm{tMgYe^{@^pvNR9B06Z
z)1o&j6`NOT&$$!M$bc34{-Dybe5C$4UkD{_HLL2*2RZ8Y?z;4GMZTqoi8KICt-9j-
zho}YVCQZ7OeA#{Oa+6BQ!0d{7N{-_;d5s{^p;x8NwAtEhyy2Q%7r6`v2~>HBQ)dIx
zlU#2?B6`8X>_`T17k2}hPr}O6kR+`8u$i~h8ga+_<f+X}7f(Wi?A@qU3cJAP2CuHE
zyscp2%)<^=P4G7dzwncauVATpFn^G__eU?&@AGIR<|ZtRxy}y7NK6D86MUKJgD_1T
zS9+{CitLP9@8DI3nmi@DxR!byR!tm~W_ic!d{rkm7=Ml8lj_@Wx>9~*Y>Dq8-&Z(<
zLyzAbd|5cx#DOd>Y+^GbgXi<0d`mb7*As@iP^CR-^RpFDrMw@7>3`o{bVTN6qdw2q
zJ=tnVe;#?@`jq4}%8j;MU%Gsq`a7`XoIv_BTAz{OPdb?iO&};D@DZ8EfH>?OIm%s{
zMk<SW886-4Piwy}ndR~I^-Tn!x-suWO7snmEhrHoHb)<q-a$XU&*~tZa5W_cTRLkn
z@tcznN%k4>K+JBul(bnpE{w}^{ETvy5>Ho6nOz66n?n1ua7x9%Lys$H<qLU~Yl#s{
z{y2E8x(jYgDaK?h!~#b7haLq%WyIZ%v~ZBNqU#!ID?X*@u!o|Bin%IzF*UQrMTmVL
zjv7d1_KKtV45oV!B)#r4m6BkagKVsV9*YANGwiJL5~74^mpl9M<V2TPi&J8d2LA3^
zxmhv4Gvz)#?!1R>%Z-@jHWGEtD=UrU(P3DuR2S&jE6mPdgjJuLsW1$3_KM)VpU?7>
z7A|)$J9p9f^`BDANn9t!fFXe-yQK?V^TR6?NavyUVU;xP?!ec8u1rMwTV#_IS?)pU
z;s?{)c|Gm1unUoxC<^~nJJlSlcC|D|XuMg|jlufyqAj|FiaOxBV2%-AEGXKVJbILt
zr~&Z3`1n^np9vB%9A_<j1CHzt0^hVa%~ql`1`M2D_3@unJyDJ5IPsL|>vp%aY2s7q
z(UI>VXQd1b!il>;c&HH)nTKKSjW?LRt^d^~FWp|d>0}TlYR^HT<5H)#t@oF;cL8cQ
zq0|%)$*g?)?hiA@#<skbZid!P0Mg{Iv(71!zVs)+2G^!}0Ag(X=5=OPO)|p7v#lE^
zBQEgv9AkP2b*Ne_N=j|VI?RPVoFRmwfB-etI<Rx2IZm(y@Vi?=iI-x4`$s{vzo1Jj
zKliLj(mip<?Wx6FoRwA7qP%X_C^-=WRaT;`R9}R+MYqej&S=K7TW-oD&<leeh}p+v
zCRnTFn!!0EVZ_f9$T+*<)Xc!^c|_weR)56Nyf9Wk{Mt&!s-_D6Jm1SmyjHrqhCOEK
ztlg;tPufKVCT63I7ba~(O>u9B_%QFn#Tmji=YhP0Wa(?B1<|{|NhBl#M+R>>`)<zY
zT=ogVDFPk^5_!GKa3)DEW*4on^^-xk+x~c@N-HqO*3ftMF6W3|+<>x(82z;$q+*;n
zIE9A`Stx1q*re|v{68obwrQXt*OV)Ey2pu~gdijB<eRKi=!bC6JVLBZsFs-T%KPx8
zoP|!?v0Ck9|HB}sx<l(pyRK*z7UA|rQ#;%o(Wb6=7OMz~38rUV<JwGyO<G*CmI>V$
zFL>7ht!b9@J+I+-HCRZdxV3j}gC_Vzv2jWUV-C(r?J5!+Sw3z;a9m)cGc3D#<W%SN
z7FkSBJYnK$T4IGtXukNL-hm&ZKpg|ELerIQ%B-61D6}?>m(*g5;WszFv<@-olQ=o+
z4($U`;mgn!L-e+aO}Y;J4g=%opByJr=;nu)^C4FAM+dIuBe0vNHwO|BqR!Nfo6zP{
zbK83i4l2h>F3DHTYPaB$KfyU0eArIDwMUV@DwZua*2#<bHL3MfB9~4)knpmwYv7IG
zf5(j3^STNujllkJ-<oal5-%Yf%IdHl>@}3iB`MLa3<$2?4KUCh4U3l;@r8y+iErLp
zciIPvgTU#};M<0BPNZn}g=3?rl0*!}Gy-5igbLimB0HTyH-MP?s@B@wtG)=?RD|Q^
z9mw=`wqA?fd@a)ZI`7(V#@U}kg8}j^oGU(Egb#&*oWa<Z`;ufILc66NsxM{R*=d;a
zRmtW-XVVX23kVk1I2H7|cn|D~G8Y1#;z2f+2K+I?ks3V-*<#t9$^Q3c3j5lppb5Sq
zr@=>MBrn;BdlXoe;!AF5<#N3KkmkA$K?}Cc4AR_w6%_&4vtic(rbF!^PVS*ntKf}V
z)cg)|2z;C=b(yNXmGWmgywJWh1s-8=L5&7T11GqVKiWDn)vHE@&z5JFyz!0kTU(LZ
z<5|%7YR!QSj*mC35VveLxJzf046|rS#$0v!LPKLS`v(S@>$>x(?;Ae)8|;`vLA(E{
zt7R+0z9vc(cj!n!@>w8FiBRupzj{`Vth(+TLhTgtM$K;a-T05gM1dm86Z-BkZmwB!
zrD59fcB?GMMdqN5g{7p2RqxyQH02-{u>$TBJ(euWWGmSFE>^)2r7HcPHBZ2zo)acg
zdmBi%&Z8fO30ievZFKz5yR|3CePSzt6-f`hTM*LT4B+5E%fxM>n}L>;p3rJ{_#2SS
zhgA$5r%nZqgaXkeXQDJd8&2hQT41aj5DiGdi<J3h_}qh%^YZM|B5S)uyiPYf1-@<E
z*a{!NPib)<Wy?I7-_nej{-|l_Bp+WF(GmMaZ}cnE9Ign4ux?P7OHsgu+y13RsOjX*
zY`Lb^z~eb~jdcf07ZM5*bL+eH`o8@LefYu-yuJ?}eD5Sm*N6OuTF#63Qtb}fgq<e#
z7xyWB&*ob^#e85ll4f^&{9Qiyu6BERu=!E(CnP{d<8+s6efJzS)j+_rAjjFUM!-gK
z5vlGWtdNFKmwcU%wn(cfN9P=~3McB4*;~-)hGN%!7LSIu7oJIa`pO7dPWCOmD*itF
z<J$;d*Zhc85>IOTd>OINTrbHWJH^rOkL{ntQi^`xN9`f}%_Q|H1kQnXpbPI$9P^_n
zkx-#<cJ|h}>>9#0?h%tA+}P4DZAc&OLi5_>>$BIHZFF~=GX*GA#tME?@bF;|=#1nE
z&wL*m*yBD?V`ST<w@RxT?AZ4C({iQBxOS@?rG1a2yzS6-z5yle#CozqH!(dBac*h1
z@^$!=m9RJ{Hh)p5(FsmF25k+Y;lo>${T@I}sG_;Ku|btJaz^O}EdlB)fMk<F4VV87
z-wUgcBXFo`wb#kImh_@JF`V|IQBy{qNsdE0yKPdf`p4T?8p?ltUA82cadOCQ2l{{_
zR`l`H;ypGBS9-YaW7VD^j2xYmG8meTml3CwRT&dRr)iKx9EOqEfAs{;PAGc-?k!P<
zJnMTehMX2Y@*1YNNNxm1#n9bYpiOyQpIAX&5hJ;wv_=+!Ssp6A9gekOZsq6a++b*D
zXXAZ61>bc|u@<?p8R=LjRV>qu_I(9O)@OJ1Q1b2|_7)g_4bb^c@~w}7dafG)mI<f&
z&afKb#*$Mdxny{$p3+8|cJRq&AaQp(qoIVQ2S*M_lrLU-&RJ!D;p63Q$8<Rh${KR<
ze|PTy;2<Cx7e2>tfigk`K)c$RkJsQ>raEj$FXm?5W#)tMvH(GxQcQcUJhL0kz8L-@
z`;V@IY%|FI@60}ZsrjjUIG*ttx&KJSYCvfk_RbksORW@4Sow78EVqRno|`eZo}~JF
z&LnKR2JQ3}$kesmLOb&~Hvs0GXn)!O`runnaK9OnZ;d~oLu_dl-V=?PKS+NAy0}mc
zxbZFsWcr<^T5I^#_K44L*ie%4LAc*Ubgt0*3Fw~uF4>-y_Qj^OrwW^q^jx7)Mt?gU
z8x^-+|4$7_1zDL3sdd2=1w^>6q7DOs#Q8KUpgPN3*$pjg`EM-fWq4^=F4!1i@@mfq
zb~OI<Qh2D`>_!mal8a|1u>+XO`SVg+16ZsQ@rn~$nSGQX$PgW5njCdE@A(S+_;IHU
zbqd)^5-S!F4I>MY?^6D1&lE=rnN&}9QYHJh1=>FIAhFEj=`mHWBcqw$rD%Bs*FFKg
zJ8_=Fdz8zxMz%R8&zl_~d<waEo>viCh1$>-w}Z0bT?ZQTAa?mjYts*ILZuTw#*j4r
z2;WV+qukctV+(*NC@9_`j#@3p*Kxifzp9R$zO24+8GVMXiTst|e_fQySr`aAueyg6
z1@yQO`kw!G6e!--<gi%t`u9Nl$CU4V>3DMLd2X6$rtsJ2?j{+$nF%Q(JW>v|GtFy+
zS2PlWR7%zqo32B~;55soQSnjQ+W`iZsZHbvUa`N<eg5G0Om3Yb#sEnRJwHfhkrjX4
zGe<l)S$9Ty`rW=-EA~qzZzDTtcr5H>Zq{|xC>b<b3>#i%^sEji$0=?heu^}qiUIia
zsq@YQHxlLGbN*Dc)UNSjH^=EruC%O4^og+KjP8G5-4EW_`EKr%EpMNg2Arpy#G^&l
zwrnNFz;e^SXTiQn;Af8ejVwNYL1GxRQ!9>Fuktr6Qx5$X3-X`EUh(?81UsO<N%nx}
zNaP21{xSDwg3Q;6ic-!zcz7_7tR&x+_D1S39s`*T_?+}2Z@V9?DXOR{dAh*1hy?FU
z>Lix5@Gbq)M^p->XCN|ad`elztCvSCH<a{MC~7hRc=$t3pk?x~HZqHrmkcY;cdd#z
z&VXiN@r>EJg&AcL$#$p8NNvWD6=jXCP5D?s2s`Avh=y?4Bv`Gb4$w33XlZqH0Q<N$
z%~we}AZy9Mzg{$Xb@6<rq?pO6v8Z)LvGZ+vA&gIUTVAA?A4Zsvs#fZkf}QT1-Oh74
z^WhI2!H+G7op}b-=6`bmK$erO)?|cmfHnZ(EmD)U8AI|!nNyEKUfiY}oH-h`Yvs=F
zqIEt0A!UKRh^cnxujUB`glw83hI31kTxBaI#=7W{drvgM@ikzWMc)_Tuu*6QN0CBO
zG#z$`*;IuWcp|%D`)102b+0VRlICvW!Wz<3R|UGj!6%r%65~J78kw=iQ^gt60zWub
zo~QH}s;;IDMzfUcjw$@6xHVNLx-vr(AMa+WvPTG%=srxOGYe8O`0mm?FZEq*Z6C#6
z5g|jz*~uIN{SLL!*u@Xu+CCrZC2LwH{GTxkFXHU+Z}ZRxj4}Ij+d#t&sdR;QKK0@q
z0x@0rFB|H2#H+Qmn1&Fdv5NiJv!r%cCw>xmZ3ScW6^QI=TMjZ{EjCydV2!F@CTR2I
z*}BTaCVdo${_qA1;yDCok1KFtjiuw&tdHsRGV?;)RN|EzY{A^8(N`1*FO6$9AdO<X
zGaN@B;k+uwUVQd4_K}Op3A2<RQA^~tNA(_|7n%+$gr&1vrp-NpfePNNhvFb5*KI2o
zE7HYN<x71?q$n^VKZU1?NH<u}^H@}heSj9`5dq8FB+JJ&t2;gh$vxTCVi;j2N1FHz
z;xQyijzfXc$37kZr;v0cAf6m(8YxZq9Z`$NahOt(1mVe-a7w9Yc`Eq&>St*4Tq6%A
z$YvT?+%J)ogP)F_!qvtn#tVI#YA)x&6_E^ieU<h_d!p`YDB^G+xsoGvxyw+6rQs#i
zjORYpL|s1G5>ayRSh(_zdpPg&WPwW!M<};}1ea$dQ8&evnQ$1<VBky~C8OuDu`%H{
z3&AS*^58rBl!%nNJAte&ZE#}`cd_u-I||n$$LQp!Z{r%$blt?v7kx}cb;AC_)Nnv9
zE6jkXw|+Sk{cx)a*Nvtf3S4~|X>HvSJG|i;dMx~L6_t4b9dmjA>=2Qd>qrfR?DVh@
zu18dJ7)R!j{Pc}0or$KpVt53kYWfBiLDsN!b!W>4#sd-J@Rpd@l^>xH!X|@6sP_tA
zHTqC?UwfrMS7>C)CvpmBkmobN@znvm+p;f_xp*f6>kBdzfy4LD6MC>)GqGhgN_&(f
zvRm6pvzK&f4m9jcLt5edxyU0Fr0-u8XPb5@4OBdvfg?ZPRnApPO8xKmt{<wth4)Op
zWl_-;hzf#p4O$6IuH&fO<(Y=OC8;O5ihte>7K^%VVv4t4BSW*3@^HRcv5T}7pHv`H
zoji8YV^Hs@BYiawOR7_-5iTXINEvkx^sfFVJ1Kl9;Nt1x8Xh>}Xg(s0y^24`qukh4
zu?xKc1T?fn2DwLy$6LJ3kyj5O#M!N6765{g?`zszC-}HJ4xL~Zkz42-xT%}DMeS8V
znL)rb3HbU8eeW!|p885e1HqECUQgBWTU1qKERX#5zq`JQAJP(&o0}tsN?q%bLAKW)
z9tgV4bjh<ZKjDcDfD?gb6|@aw??zHh_FLjsl~)^6e2|<i?h(y8TjHmAFpxrk8UzK)
z2AmJj2oniC$b!0uxexS61?4L|w&$YM?{2bG;HB}hs01w>g4JuG@6H6IhZ^_9O{nJ3
zF*P>R1Qc8S8~imB3X=tN?B8=X^Dcw)8U%J2q1!;PJ@_hAlKMxew?J6(uwNx0^adOV
zZ-vLED7(g9;nZIi43bv0Z=W@`yn|Inu`R>Ve3MjGI*WS^^6~6IKN2DOR9pHUP&BIt
z?&UdWYw8(~*0aF6Z+36m8tSE0u4N()ZB<S<5oAakSvO$IqcUOwE!wH09K6?MG1+fV
zplg}S9x(5&AJ7l=X*UXeDafvoP+r_~wsktC#Q$N)%aFz>WdlPB&GTjtz4*qfl)A3}
z*qVNZ>j6DzbBS9$Qy=vI=3Kd1DuCjM12QyLnua|$&?}@HO4w<bAg1H?>su~Q_C#aL
z52I!aD0ya&9(h13U*>vjQX|fh8*<RL(Z_kW4vnHhQMnUPrs}z?)%Lzkm0CMhFs#c#
zH>Wwe1{l}QwG+@Dzi9CqBrd5*3Ts~jLuPaQ5**wF{B~4@97ihTvmD|ozM8D1SzOd7
z>g^3XQz5T;OJi`h7U0~6rYn(Ny|NXL|G1}1NFgrmK>gp)PSY2KjwJ<rb0hIElhh5Y
z{DsIFNTn2h_gDEKOO#jup<{XO5bg%ibJ0Azx#gIpqKvLfPALuYTxkMlIvk1R7N0wg
z8mNb*yF<eLIW5<X2EI{>8Sw=;ADtQ@<+MxbGU(}OyT5q%dq)F}-a34o0DR;pVk?hj
zi8$!rI_|P7RtJYKs**#5lsOc2WdhifvO@f@7fCzBwujePlI3i!|1F6?!aoYqD9Mj;
zS0?{t4LOMe<a*SP#_p+<Y<NS&6k*?x9#loF@U?Zg23p<48usR8_S05kmV^{C(B0a6
za`q!YS0#D*+EIp^m0qFooFx<fZ8QtU&J0^7F(}72pxK~FUaym@nIn$-h%gA?2{DO?
zNLTEC_S;YC^4kUk-ywU$u@*nt{b7i;a<m@6=AAfJaA_Xoiq6e8{bmjM3A1nmuUL94
z+4)x&71&P!3oZ-urboby*o8IiZ7t}*aoq_YcJ$*H;dxv`o<h7#3y7%N3i`{nG52>9
zn!{rI1s!$QV-jQGHq_}VJ?&m@>?_<hwn!$CtKHbtF#jvn*TV@0>hg|5?+rLO#)*{r
zAz-t=);fQ-yH<1Lk`;0u5r_0`Gl!v#zS;HJity8dIB~t#<gJ}Y7XBO1=*c#R)fM3u
zGyEpyg{(6!PFHR!7G#_q-C)huv}yKBE2QR0%#~nh6Kun{-jm0&J=U(u2YNY;oz7?+
zOHvvqK1-m*{6YM&yYs@2^iA9sqAX2H(eC-=g{5kiARvdUUdHnqg6___b^(3Kgl1J7
zqT(0$x!hR(#}z1AI?%xJ4;M9BOf5NF8*RXligiLIUHiw)W=|u-dM4NT<sUP!<wBto
z5^D+L-JwmU8>Ho4-qk716OR-OFb_OAZhJ2(9x9xo$Fy-vAR`2r2o9v;JfWsH40*IA
zYuI$u(HAS3{3kM{eQzl7=M^p$oLP7GNQ`a<O1D@aAWOck*H)>WKBV>PuF32sGf$`j
zts$EU8O5x6T;WDo6QHi-;w6$^T#nNEY6$0~f6n<BX_TX4EE)akZ-FToQ$}1)g*+^K
z4mEKL+`m}i1uGp<1Z8q1B73nglB*q=RH6M!E^~I1RGpz<jrhI9Mi=YZVuWZnK@3Sv
zc}2UK0~_f1ihwKdh#>u*+Qo#v;FMV9iWxcE7O$%c2iFmh|H^Bxy%dDIdg0cK5onUs
zHV%YbSdpg4@qfSzV37x)uE2BP?(M%1y@-bsPoLw@j<yIxxM6vBPJx8%)h2x&7K?^!
ztC^n4lsvkj2$68TMtFZJRxe&qH{tn~sk#KW1^#UmZXWy0@c@mF&`)?|h&k@PJE6qh
zeNI$5i-q^64{^^#hW+-7`z{}BtfZ{n-RM1g98qtof+Wd(vj^HWa}54U2GLI-7ZWCZ
zv5^|#eEWgm{PA$QbO9h2Usk_Y`>$uyWyDc3h)ZR$&~l|LPytP`_M#Z(@c^X}i1YXh
zE~Z_IE)p!pCWTW-;P>IggW84kq30g~lRtXNp(=_ICxAVmk9Fk;Zmhd&V)OyC`ki>U
zS7Si@k%7Q6oxT9{##gg1o*JZP%np46%FIgcu=HfRyl2E&l^~dn&vTgiy-HaPP*+}3
zx$$o`<6L$>Vy41(onQPlQ$dtw!gQLLU4L%u)|~(#K4<7g*gu4x?E`&r>CT9}nUIGc
zl+HAeuG*6am1;)bd$i!w9sZ`AEPQkK3u=}%ZHc+h6SnB|nA@bxJA{%-c~}yOx!o4J
z<juGj1>4?+lDOoXL6+c<NkS@_r#4$v590UOu3eV7*&+{_*{f&fDcP5yAV;M;61#Wm
za^K-ua1MzN8n6BCar!yZqzLfmX+Pgz&=_Cj0XCZMh-j#%OkMB3f)I}90?|!3m4;;8
zLljZ5yanQu`YWwEfz+hPPg2T6r<?vBC7<BUc(3Y2?YPHbh$8vcm<_{pl3@Gm+0j)_
z5*$O#nPPmRLt>MV)Q|OFLhu7?0qhk8FXt~qLG5s^SC+KzhRC3~Oq+C;Y6(eb5Uo)`
zi>u3jt%51w;)Tb`kA84){xsr%WLZouZ7#ENou->mQYP2L={Hw(+z@2Dk2%7*`GKnS
zu#kbyvj@@{EEYg!IgOIhRH(_=9=0UH2_@c)Q)E)MtY*_0Q*YMaCtfog1yn);0$t@a
zleqeUJk%%jhLSJ7S*6dV@VydDE7Atl4auJ)>9NGdAO8&<APT&WD#4aY<~swe-!D_P
zWkv~`GT<hE0JdXsO39OOn%x83iu!rU^|nu5BKjXL14<yrZr~n*cX;b;$)aXfLG7%Z
z#3IjwWfTM)TE#I?iKzmgL;L9z3I5*wQNm}iwRUse&!Al9T*I=<KkVU_yL0%+NldXi
zKl84#_GzcFhgdqgPw)ROabzpvN8}IaE<XB8-ZMZRU{Hz(9}-lZTw$V2#+vT3N=WH6
zXk~T;+`(CVp@QnJp2;Ils~a+Mm~~2i|D^;G9O{O)$k=46h2)ijb>i+ibb=V2WqH#!
zaZ&G+y^k&;Ov>MuquY$JU0%LS2u0yR|7F-kNoo!`g}%&-<mDhv`puC0Yx3q;&Dk9{
z1+O0ZlPtyDLBE82%T+e7OP2<3A#_I@AV?A{uS2~Hy#awCjz)HU#L7+FG%7w=;+j;H
zKARznuzH3^oJ+|g`3z9+w0lHeg-ELj0Yv2llGU%D<QY`NgwKFCkVrdeAy>f}>K&Cd
zu>Rd-Y5ntE$UxPFd;QyL5s3+cRG+a0=1s;Q@BY$_5aGsS-jra)2r6xVIfLj{#4?)I
zpuE~|!lA}44r!R|28o;3ALmMtU8EbtB`DK5^_@5=MORLrcs9LPZIrSR{~b<d$&Ns|
zQoT5~H-kCgHv*C;Rfd_3>z}7#sCr*TT>HBYVNi6BBCGS<z{kIz@`d}yyUY4;CrG|7
z*K(l*!a`zC)X^CbI>;NGftsAUuhr{2Hf2*wH=uHj9TSW*ED($LYL~7_bh2ybTD(xm
zD;9Hcf6Q4wsLaCQOoxab!Z~ZS^>6fbE;}4?Xe_nPz+e2!mJ6hC)lxXn7I|M-#fBgi
zPP={2hf&(c%V7S+B*<@HMIgND2sL}&_;bS@0FAw_iO_46^9*0porJvTOZy03@CuvO
zQ6fWxzVq6PHC@>gb`Qv7>(l9I1)0Z7&Y6zFNojxb?Hzw*tviQ~`#Ib=vh7P@9mhC&
zK?{Y#Ng|iM=StU(Vz*ssvR4zfcUDVWqKi^ZthEXa?;v?pi*v3GUzc1~eiUUr@>wA{
zZJ}(n@qX#uhWnR4q%B;^_x`>mX>&*1!~UBd>+|vZZxQ`?MCZPpm>kQO91`_YAQc_!
zjB7@&uR~?)UugK?Qh&XN!#Cn<pbNkJ5#g@-?sQ%D0!c?-rj$gXtidW<*fw@>`|XzZ
zw;Gqpw>-(?{YzRCV7zV(2hV3n;_SH#TDFDe&dHpAJLt;xQ-YfugFb~%*m`NseYTAJ
zhuU(%L%8Du#TX_}&0@$VeW8UFyxwc}ZmK1)rq*#(s9~VxArMM1je>UXaIS7|7gNnq
z^7B_W`nkM5f{U7>>2)exICiz)tWs@3AX~};RO@+=gysJ2ccee>00Pg#6llWRC&<Ek
z<w3C?J`HF?2@@7eT{Nz7w7YiX)TPKLR(oH!_Nq>gH4amGwujSr-sX3yeRptuXpd3F
z;N@=BTj3j@Y|Fjc^62D;g1>(Cn-MwgI!`U~GP454w5p6t96v-V9QP;|bj03kNwVJm
zDPfG<3{BX3`R7e>P(mu0z*I;aZmrsZNWe`oq3new%T;`fY68cZnXTBU5+3F2n{&@=
z>JEB)2!E5m53-CS`O)u{6h7b=)7_J@K$=wLT>?C5Ofau0$g!nUWMRE$f4xy86}+Nr
zOd`Y2TW7l+vLzQ4`amznIRvSQHt#o+-tBJ>wcc1Dlx(UxXyIYbe}}*6lXjtLL(R1j
zR(^!m-8)JgeL<m%pqQ=%XDN<Vk!gL*E|v-R>t%%AZo#$Eg#D+@5Taw5aZwhhW@OzM
ze%TE<Z2^cO>uQWBrDTD&cYmHLo$>4z*0kC_sl=pCJN^yT(4kgD_;rIa+MvwvLB1_<
z5P-q-Q!^``-#(qvHBqnFNOV-!D<FRNq~<%9;6uMA3J2AXK}(R7#i{=JfRgsZSatD3
z`@y2F?Q{Xp@4Gv9%dM9_0XJe75P|5>Vn2fTNZhdYKrrAsMz%9epWO7-QjfN8+U6aa
z5~ce63Sj$$iRVENcwu7X&<~@)L2(ygtVu__QVOo?i)0rxNq886syIT{8=sOwHz-NJ
zK=`S8b$L17?@j-1ar%jtrBVhDi`fdf12xB9iv1ZJf6O_e#NCVd^d8?P4*$BHbvg#X
z2N6wqle=@LHvq5{I%l$#KXRaCD21J{!EgU2(ZN8i`L({JE+GyM#j9n&o6GXt?)M`7
zF~~F~@UZxP7!#&n*GyDH^Avwc{49<}G(2-bOuC)P3(UXWS-|-{%)rNNBOoTWaeW?L
z``k5ywI&DW-*?yb37Fnb(q1pm;`uYQD!eCszFaB1P@ee#Z4jpe5y|c#<mR3Wq&!>$
z^0m>H$B5c<{ECYA@70VLUl#`FRt5bR%{um9pK_XI!-4cJ-qpeKi!w#;fNz%#z`ZFL
zG&+J#AbH{SeMKjYt2Uq!+h5~kyZ}#nx8Wt}uPOMkWQ7diRS(LTIFtPPbo$T-hwtmR
zazBn$CjgLC7s&k%x|&{C0s77_1ZU(v0?TqP_$x)*_DlahSphGYDNk?0oHVGsBbUmK
zZHCSXR`F1MknH^^Dxgs^9OwZnE@)d!yB(0jXo+hpi6A-8RLC{{>nA^d7f%Swjg2gQ
z{I>&X#T{d<0da2-Nz-5ucABWIbP{$PIoU7)F7w@;Xs9yw3iN`7=NeVpE82Gd|9C7V
z9FkUh0Rxhr#oSpSC{ez7^D$CB7V_S9yq%fdg(vU$GP7Eo0vm<N3H{}`!6q>5bZA23
zZkYoU(!XXl^dr$YJU(YQ4&K3$tzP~HweM>oGF>pl_i7Z&crQMlxVUwzBA0bdnG<pT
zKkSMSt+@a$S^sOyfdjujczI8e%np?3spY}D?QQ`>v@9sgQ29yKoOF3B2?HHfI#eV6
zifTw`?fAdreG-Oy=g6-sWAf-g2I%%f=QrX#Ip?B|ey`tl=Hn;)1_+K`lG79B<FBcP
zF5mn66!0$@Vr$lb-;bs1iNl!T03ai+!$$M!M-5D?tivE(rpV&mwdh&u4Y(u67sFso
zy~%vW%g&#$Tb~SA{=J%%#m?bBFL}-!2n)1GD`iI}CQM{zG1j2*&Lruj2`t*|AV`}4
zPN)~$<P3CIhdZ95|JN9~D?tQZWp}O4<JYU6;Uy<Tk(Q^%=d*&EXuv1nx7c2J<KSHj
z+8&}sQDt6I7pqG?UhRKdrB|d|Y4iO*s~GHP_SB!pY@>LIUIwkrJUCu}kKxGCRGP-O
zfc@6Ub~RnZfwt%#?tdPo{D7DR&+3EuzTftIBtYxlxo(65R0MtDuIva{8;!fQDD$5?
zfR)%#Wpg;%cu!9V{F-n!wrm`OJHqS%r`vyQ2rLBCcR5NpIw=3Wm##mWHw*9g>}G>f
z_tw%k<I|2KQz>M}&Y1U_kt4knp1y!#as$}xw~zl?6x<#x;qUOb6AIf8bWPU#e)D_+
zKv&D|7cc0_S3nr3YFs#^q)+Xa-Ti!Oe-6xf1C3jM{3*F2T8Oa)q?)Ziue{NI<ZSwi
zse<Q1*~29`Rir>NuL~9~=74B4v)(_K0qeQZs=uvq;B(9`gK8(XC9i=}a!u4hv-{`o
z*!Lo99kPyNXh*Ped4l)i5HaT(KXPtoiuoN5{dgeF9TY5&$CvvUS${qth6thu35!<L
z!poe%H>!c<;tLx<K3TOu`0z6Ce>NDjW+E8RUF!nHpSv^;lmQP~5Re>YL$Yj_r5e%8
zU^Fv;1y2GVZ=;7C|NG}+77_6|OuuIaNa~k{Qvmt5V|NfFq1UEl$EiG5-!=_}W>)y<
z{*HH$zs^1bUb%gt;<pRsZo;6+$?v$kxgjno{`PGLf~k!gZT14Aa4c`?m%#b?ZodXg
z;Shv{W5ImCM--FjpMS!j!aRVagWu?3m-Eu(lP7TOLKiXe{Qi?4zr^H%g|dop{u(KW
zFA0TR-kQK8^1%r_k%U2XBpHbPcrTIt$3_p}vV6kM>il^GHqI-Mf*VWT+1c?dl(N!*
zaysgBKK*Zph5VhaJ~F{W94~x-&q^5nQBlVu)fzClY!bI8AKue6ko`UENX#b(HhCyc
z^N)jRqeIAm2ZF(iE0jLC_3O=lzD^rNQw$nG;=k=59gT4$7xk3t<JbLnZ5_asBTpv#
zv#-WxHxqGWpn)fG-T2z-+cyF_KewIVPZ3TD{yOpS*6$(1L^cEGKR*@6`A-grw7?TJ
zoZl1u<A<Coun%ep4g48RSJY7v`<Y=;yZ*NSeQonIlmA-UG$CwAn)X8_Tp)Xw40GOY
zCrk712%^dPFd8GaWJABrCcqN^A%`QxDK?;o(q3c5`k%j^A|l6bvHkV2M1)Au8__va
zAbFY8>@57B!B*kLkk?Sk_&78EdLtG7m)Wv-uW;&9f8XxM7Xk9H_nhEXkoY~)Q;Gqw
zBrN~T(hnET23Hfc;6D1t!CT;Tx@P}(a4Bga06s}W^6$R=d*G4jk4H?pj^6vvp747X
zeoZ_Pe6OAw9{fK({{3G}H#}(aM5NSz9{=xj1z?dEQByek`$hh_H}Y56GhlwI&K*p@
zFZdsO12^3Y-&5t^W&Y2g|5*`eO*dH2?5C~B{vX#zTfp~3IzTG-fBykz9eASn+m!_W
zkL$-)!1vff+kfJV|2$zL5`K0+AHn;7Twe)^*;;jl{{)l&W1Lj@!60TXE&q37|4zpL
zn&SU4PCsHDw^Q5yo%Fvq`TzgvA0>K}Yf*^rxje$mqhAsWA&Dv+O7ncbZTAD<Qu}xe
z<Ns~iXkXZ$^4ARc3@R|rPxd^oLqaAmJG)sX;r$sm0Hk#x0XR6_PfJ{Xhx^}uVdjWp
zqtOa4KnbW2w7xLcAp@KXMb{0;IQ1?L7SBRWHwECcy^#Gv-#Zr-k8t2k*E@%o-Fo_G
zMgC{HFxL;gq!Pu~FbC;oH<0EVVa<Y0U0{t7Q1SFkg&fe3(A5u`bpSY-0e#36nf;v?
z;Bu${&*Y$eh5gun0+GK{++*yYS)qIr_uJ>y6p*~`VXB5R=LQ^1c5p_zblyPf5`aaQ
zjq~h?;I>qCX+r<EsauDFrL)O&f$GJ`L%bU!kj$00A1-73d^nxX6RK36<3Nu;1I4M(
z@o1I*dod}wz+!_a9U_Kkzcwh{z7go;fr4L0`Auk7oq^oNcX9mx`=?!pk)QfRhmbPH
z11|fUqT{Q(K^LPh*p61^U4+ENN`dbl3H9HX{`X(#QpCrsuOn#VyJ8ki3reyh5GJZX
z{L(mAW!{4OKR1v+i$&-xtWL_T_t5gbchAPgMj1%Y_qOg5KTj9gJk4iV^}n{0-?t5T
z1{sXoZS%G6^XK<%#l*z!28^((SKAM6+=hPgr~h*o`sZh7+!9ZoTG=d^Xou6;?j{Ns
z=ZT34bJZKDNkusHi2mQBqjdJi=-9@|jw~-PW22AJW1}Wya_-n+|Kp7QSTBtkx>Cs^
z=t)$#Ye1arK-ZML>jk>Bs%DQj)9Lxv#&?I);uG98QRGT>{-5`gAc3!StW$Ys=ggli
z`?rHlGHm{K&8q}QcElj=ULzB6(C#TF@vB_|_v>H62`}vM(feP0jvdEA^(dmL-4ssU
zu?@FZmsLmC_+;W;RsdHY9qj)(=%U|nOvcuPLaCzjW)Hie{kT`c_a>@!k<(@gi9-vk
z(X%g~Fz0<v{Ls=nZuX!GKQPS9X=8lhP#8Xbw~X^$MYYeHOP|{_j@_))_;md>aw6tD
zpY8g-L+dtC>p9m%Cw3uxS=n2Dn*^+na|WP6*pW{2R@45MPMPFv5p@a-`VKySLkTAH
z=2+6?9>41Mjg+<xz5APcZXaQ7V#JAO<{R!>H7+(fq&QqM4eo8Y7U8v-DOI+^Y`S}W
z_3c`W)7QXpGMQtflAcQ^s;rf}*^6{33iZ`<Z+OHN={>bEEjJphTr#@<fIe`j&#X>`
zrnCID)j$dTV_H53&Be9;AKx8ZwoEaae|}Umk~d_S@!R{m$xViVWpOs0uFvbOZu%S^
zUaBfEOlky`!pyrh>-k+(Lki_f<4z$AiY}YcvGsXx8<v=){TBPE_q#G=n)bF{-1PL^
z@x3|iS5$zE@-H*Cf;_<9z1jKsZ?n_Hvou(~mGqI|1Gs7IYhT{u?5s3oPk3BV62E%I
zde>{ST>k`OY5|EvyK8|w#x=E7PPe;;N-a~yYwD_e8+Olx^h$WvR#}CN7~&O#oj-b@
zREI-yY52tn^<%H{`o<zk_4yRjMM(%`MUEIf$hQi-YgDH7^fn`d?Vtxqz1M?-{HLtA
zUwq?y;lA|x{qF2H`m*Kkv&An)!Bq!@BT-A3#(jWme18S33+U_DujecJ?<AW%d+0JL
z?YH!5nP@315f_)xy3v2z?8erhf5uhq0JV%GiOGBleP;qOjvX>y&ak7$b62=z@i3eY
zcO0WObAG#=UKPrJ+GQwjq<JZsk7dpw`_rnx#^~jB0a<U)p>l!#>;R?Bx#+8_OlIX5
zCo8wM7Q+UL=Ti++SUi@*FAdq8Su5JVc=>ORQc2*)sx#vgsiF{c5-~lVe%ggeOQ*J0
z;Ys2C84-t}XFOHNz8=s0vLKGtr74MzBg+Ime}^uSO!`XhHJ9%cS-RGwFavg6<cEU;
z3+w92X6C}vOh+qZt@myWKeOz4mtFjL<>EkY`O<NBwOYlZe4Lk`F5VFr+W_IT(YSL_
zGo7Px!6Z)u^xUq}<ta=3cHFUP5ELbRG5NN&MCTz)Qb9|H9ZbZR5Lq9Nc@4nkCKgqg
zTl4}hql7qYRyg#6ES&ew(R6WSS*zX_MOh}0KF6WJWXlmjkSpRKKUC)N5%5VYlYXn|
zhSqQDuQXQ|=zc$XQfk{kP{v&B?vtd`4zy;XPDgLPxtMa)r&K3oKSjSFq_6$2yU}bG
zyjNg9=atV!xN9DNeem>Z^7S2a$m^?9eK?{w6E}Q&g@(`QVZ}J!T+>6R6RrkQzVnep
zu#%N>nv2VW)mBTgA3E`Dyiyz$O))3d=*&UvB$4=ozrMq8c(?DzvFlrlR}2Zf)5T6@
zc#CY~Ud?7B$*gf)7@_8?A1JrxGs32%vtR<t8@G@1bc+8TCVz|_{s$ssqq;NS28Hi2
zW_S!4Za~-FXekuZIYkL29EN!JIrr43PvTZc6TrR%4>NlCa`#iR2on!I&2!Pbv7-+j
zEVL)dxsXO3z_p?4>UFmn<c_1^F)^&`7g4S3)hlk5G74<@;BMPD`05_x!qD@$q0;te
z(Z{Z3^jW5)%W`vsye)oZU-I{Cl&g*G)kD6Zv+%6=G3&CH2u%y@`?&=f7eN9r-78?E
zJAtDYX_uTSes=|NP_J;Ie>(BD3u*UMOLXHm=jO;rb<AB`z8bWD^DFX~k|mQjmwKvf
z#3OhwGVwcT3~O?JCAJ@?wVO1{HI?Y>a~c<~33<YATus^CH@9#1%JQFwNxYBj_)89+
z1RqJh0aSj3RR)r}h%6hecyHtNO>0OF%^jy^K6S&{FHPcFnYE3e(RF4{Zm_L9+$Y~m
zHCG->uwBle=%Z78QIY88eg#E-5;rF{pM>17exQ2u^knVE^V61r<&}XK2Md_{EARD1
zLuKOmQ6CSbyw>~3^#9z5(WOYFHs!itZVXXf(%=a$!y09!Z6tLKbjnj%3lK?E+Yv(q
zXmReD6E6Vf9)p_7;N*GaSa>EKmx|-yY@DK;(4|PL{z6o)3eAleMs>fwAk2wFAIuF@
zlaFOS`7VC!ij>#MuVv4B&-T|nVC_6!{y^#e`FB-AbX(`NK2aF_suv*7lx>L|d0uV4
zlC9~8V&D)zlX4z97b+g)b3kJTnWHJ>e~~us6Q|RZH9L?EI5EM0duqR3;Fgd_5zl<a
z`>xFw1YS{(GBWY+E1*sBw^ux7i_1+6A9f|RN3<t93|x8MY!I@lMDVm<7{|0h+;gSB
zJ&xbB?V#Dxz`M?@r`n0X!v*l{S`eUwXBFi$dO@}4nT!%`#PDWyDSf<&cLynhXyjW)
zuI>3Xs63?lKW6gTnU#I*E^{n&Z}`avircBjO34d+aKY>FntbR<xfpzzq|yEc&Zc*?
zOd6f<a<f6TgG7H*18FR9LM?`to%(B~Xia*EsPsGT_emjVY**B+mMdt_G2!P<A0exR
zIwB<*O{GVuON;Z7y}V@*|B%Vf^#9l1TSrB?eqqBj7=#KUD2;#uqNJ##AR<yy(j28z
zL2^hD3n}SFM4F)oi4jb4XdD_;Qc8(Y`rVIu#B<j9egA&zUF-D^*2oOc%-qi%dtdvy
zt{wa(|E1niNtjH0zKXbyY+FjGhf5f1sHnrGq8_F6zrzWb(Ed<!YOep01=caPvlb^L
zjHYhi8t0X{IW`_kdIfvtB|Y1H5)vkXBCm4(rAf9Fu`gA2_z#EC`dapB<?c>Ox8{5J
z2Va-l>TI#yy!v4O@5N3;@)bVxJNG9C;U+@)tbct@uXcCi^{^oG20Eo-53V}?H_?Qy
zx)7_}x)?#UH$mTHYr{qF*6@qXFGKIqf}L4m*D07?H!SKyIra6)jA;2af7Oh<$$SZ=
zsg00r?3earrwbQ%=#T4`HExlOOP)LD@x=_t9Llfi(>O>P>Don;|078;KXOJe^e+?6
z>pw3zTI3(9ur>9RI7deXK(6!7pL66q*XR629WQ?wFWP^UtW#-pR1Tj>|9x3Yuf}t(
z|25teTh*IVz?(FBvDP<q>Ix}(ZJXrB0gilE4-#l*dD7;!vpL9ICKd?l7L%>Xnv6Mt
z{_a!lFE>0Ajv`Stz#UDHwaj^l^?#7h0yw-0JdijmhTIa%pq^;!r%QJcQ+Jzu&lnVG
zD+2M1^z`5F;TICKlAIq7-s(~nb{@M}E3LI>`3|m|x7i1oEXs6HXh_@mK+ADB-1FOz
zz0FWU+U5OMCCNyRF<dQu7hUOM+|^tsCe>%$5m36g^g2s)`(Py6SaI!}``DWUrL$M?
zrKPK-$({FJz5JQNU5u!MBGoe}_Dw9TeK94Rq5Iwx&HrgDu4<@;AbkDZ-c=|hJ5bW#
zRdS&i18j){&af&nr*=!)TQJ6*an1GYTGS&@27t=DkHQF;N++N%aRegUGx6JWUH9CC
z_DyDf^bKR7L{I=(Po<cL(3S4=?YgI76yu-s&QG`=jQ~O>MrtCoXiwrk*-^e-W8*=n
zO5eo?{KGXiTHnWbQ}`yD3mTMpCAadWjO}_jN4=t0X|sM}JfqoEqK<>Tdy|-0CACnE
z2rJ}mie6yZQ>CyNV1&NMG>daNYyll{pQ>iv9a&*=QHOURpRuI%1bcyID9c&<ZiM+B
z+pd~HM*{U@1QUPGv*d@CqUO$SSAp9_44-e_IE5SvZ|R{4P`HuzO^wTEj(Kja`afc}
zsOmb!;a{i82<b<3m<&xDn-4v!LUWCUHSebMKxq~iZ|ZD=Kk=F1X>4n1&(p2&)4w1U
zCH+xoY8xSFi0Li0ReudMmhboViU5sX!am6wQ49MD9aMYICHo#FCu?Faf|}eklrL0+
zG@Hk&28#X)WU_m+_Z*#w&BI6Ni5`K?4V!9c)Z65J&w-`uxQOnmWeQVUL^ZO{Upan*
z(5JsRj3S2U*}Nq#?Q+VI%hRi;Pm@sHA%_@bJl;xP_%Hc(Z%JJ2g^=B|sV{3cj&iHS
zRtbAQ=ZfU905b$ifV%DDWn>IlJ=ob9jiVHaVrcjM+5qT}A?2BlzC?%(=BS(R$d-_0
zUXH&;S_Cc^GDn1n@&-gV>z?2IRK^Fjz(}B2zK#+3Dg))(uQ|@@$e8XXquAT<*mJ2w
z^|4>X_F7@Mrz$reIr)b-TgOhb+*3%qvfJ8BN6Q~{Iss+2(M}{e<ic+D8muS*?o5$@
z9CvWg1*6u-_I)p;2KG9J3rdU}34e0FG~)tw=U(k5T51UhA&|E@Mh<;b50p<PNxG!J
z*OyBZAVIR_TBOVAeC#KNcaNAp>D2-m&h4aREob2Fjt7T?r`|nh_z|p+?J2XHj3f3d
z83%Lu#rE?YOHKFrg=*+dVwb44xJ?teEURSs#vXbe&;Wfmx68zHx2;pjiPQVoCsqVc
zt@f8(_7$&h#U(CH3UtXF7-ai7a!OjG;jvv)^^q4s4g5q416?dq{CzQ&Z7FjjSA!1<
zZaCM-LoI3ZGx(AY_$Y-QnZ{dO3ShU-Jf}R5fL^CBk*s`{TqFz-ui~8K7b|LmGE8ep
z;-#ew*OobCZ$+^ZXDMqDQ&*iZIz&^|nljgPkFBc2PVc*=XlmXT6SG8XgqGS(^LF<7
zG4oGt+6mIOCl13^A^a<&yZXZ0v)dJlNM~)OA{bP!7d3l7vFR_%fh12N!Z^Ku>cwFR
zM|zy{03fK<s(~*nD=Q~L(%x%2gb^-5sn8qPz=VF-^4!bRO=njhf6#qzr8h`!YjO>@
zYerRR_bY%yv;f=T0&Zh*wE)0GRIj|e1b%Eq$IE5eN8|PhkcwMpR34|K5eTPb+uQK{
zt9fNkH``uEX<#g%O?{6|#BPMgzm_t_(XP8SHSUxAK&Ete@2<8pae%;(89yb`{{{X+
znq|GEM}5*OlA=-6$%?ZJeEku_h$LklarO6TV>|a<ar?Ysmp%>(niSlEJ|GF4v?~bE
z1jH^bkgOKSq{WH9mz@O?nuALh;?XKDlT9%RgTr6Dy}6)zJabO7l;T3Is3}Ib(8Qr|
z62#3Ct6wKo_~;))jp!1<*Q(6r+mE4>HoC#?!B83x?YS@h-mvjZ)Gq~X*?m<_5IrRm
zm3pmHc1S>xJkJTcqKi~CvF{uvR(P<(W+$f(1J&Me5$Kam>59Q)ut7;U9T0yE;6cY;
zfA8WXaK$Z-y<&~ca15d`JKnT5`;kS&WC%f4KIi8P9M1U0|Bz1<awghWC-h4)?(0q)
z&&6prT%3v>ZVS&t3PDXeW*_ABTBR!LyjS;C&I;FO9HY93OCXDET6uC~(Rp8cp|<5o
zy(QXv;Y@tZ(WI0LmaTjJaGhy(+gqWl^~xTsC+j%w*;Sr9s`l{bQk&Ri{lyvPF5#V;
z^s0G>$^EygP}~w7QCVw-W5q#)CBhe$o3L4(|3|k{sl&inGw=im$>A(@m2JCrEh!e*
znU*Iztn4?TVT65=de~t}n`&qwdocz0XC2;U5`#Aaq0HG#)4LqKx?jv?;&MEkLAA{=
z-Yto8E1)A*O((AE&2bpOEN01UHkyF!e8Vg-qaQ_ph*r#|>qRlH6gP7LnmdOhhg7b1
z&VJPW>=jVbU)h@U+p4GVQ$I7!%_L?UMg82iPzyT&RH?4vyx~BaGdj?~JA1QSqKI%J
zB&p0j>wO5zSzS^5G?bB@eDAPvdO3j~aKs#W{+u7ES{qFP_GmEPbha7VAdS0CxFpzI
z8hlQd)4~VCY>m%JU>aKW_dO;3svDRIP)0msL(>P2gWsr786PZ{PR9R?NBJV_T9Y~4
zTjIPAI5fG0$w;?tE;eh#ht{`JWdzrCKI3p-93uf&%%!2b)JwCeKy=c0b}0ol5$`!K
zSV8~1)RKH?bA44L;zV22%uhI+eTeL`km_a+zLkf{aBMtLHfg=0b`$4qWiwT=(D^Lb
zj>J@P{*%K=IkgHv0JnUh<=Ue7YBO(&$Ey)(>vs0qVSa>$;)a6S)*z$~o;{k)piXa)
zxw}id-yzc~uiAA+7r&h3Dg=}sziRh3p6qF~LHY5|r$N~Fz>eg(?|_!NR8J3z&$bHq
zyBin~+r;KgK1yC*W2rBmyBz*rtgux{tYgNw_*#&i*T`DEu-w%psjo&7yu9^da=Q2o
zgD>AYFgzL#0|~?FZl-?M6`qti4s$&II$Lx7XjKi!Cwxe|5+$Duwa}7oNl#D)+#x7e
zTws3t3uItFM)GY)&+LWSSfx(hM5RIKxACLLct5g3Mo|Y{9|w6;sH(ukXF=S->+}>e
z)mSBOO5DQmEv*2UEpkh=!LLjQo2MU33jR6*&7aGf)4k6iz9IB{l*CVbns^YH7FD4C
z$!C-89<_kH#fW;I<($V!4wf#hF52*rL#NUk7FIJE^G8~1eA!&>sq?2Rar~Q_E5-a?
z4*jirUR`kXGc}c1J2jfG5-HIwR=A-g{<Rqh)<Jo48ikuWddA~U!2*{P#c@s|Ku&H@
z4Qc}3cHsyCSRO)U#{8Gt^lC%gL0k3Wtm6P|NFz=tpv%=eXz5m%pR0b(sC)UDF9jAQ
zQF%Mq#?vFxJ+;adyf|o`8q)SRuD5hq{0NsXhSrtiDa_{w?IRUT$Es%Jo%?O%^efc1
zw=Cq@rO3q{Z&~!0g>Njzl24RZdfN<O>c*sL5^Aoa8y${1Z?5PsF8EQl@LTlp2pM1o
zJ+10$q@xutg)Vy!8j1ZJHyfd}!Uj7U9jEvBW!DTZAC{pvB^&D^a>9wsY0n1>Og_C0
zW(?jX?UiR~e}Db$GSv#6YfnOTPht(e1G{3;9Mks^_!kan?Luic#0InXxSz49!#5t>
z`8-Kou}SM1WtR%uDl#WwIV-RU3azk>&AeBjDmSWQOVjr;6EqMB3!Y*QkqA>%7{XO~
zoEW*>SVpCds=ruClaTw&dz=vJ82!dWwmlX%Ro<5dV>7u5P^9)YXMNNjT#jfvwOE?a
z@M5}?XNy<N7@hKSWo6$WIrD0r(bT;8%>yLlLK}HnXh3yBBdA*K1XX|s$yxU*n4BCU
z_s?ro-D-@z9wr#BB1GL3sb0mC8XYIJsLzwCCyQ~!i3f^lpQ=QYEOw<@#&;Nf+CI-c
zESdc~S-x`AtaDHA;Io9<?nKwR<1BoiL6{l?hhLx8h{vZ8r0(F?p8XMBpFu^p#Ef<5
zL>r$yg{2R+M@yPul_?t^Ce^qkY7;xZW3uWEz;afgEO7C;o)*Bqozr=1Y}>Bp^6sF7
zO(?!Wb_ua^>@lMJaOH5aXuW+Ht5iCs9&Jr`bt&4b3-;AnhF~LG4N2j;Lw)Vb`&abD
zZm>EX4>l}GF|L`p<d#g6J)yexT8>BUkAP2dPj~h{)-=0!j|4<M=kV8lS41$MlctL_
z+iND^mpQ$!h=fNhf$2Ucx5V(FC@?i?;VUP1Zk3daKPSnZ`RTsV%^)sy8J6&snNl1P
zt<KnjA8lN~R7u8P=WnCE@B9sJ_>h%Jd;}%vb;^RyfSOxPI9xgLTQ2H^vKt@dw=Zy^
z9M_`v1pyH#o-OOOlpNH@9VpWl#ka?JWxhVlEPaP^V?T*KNPT4(Jk2m_flfSrd4e-(
z1{TS;iUe5lTM|Lg6P<1HoH>n<3F2tf+iS6BtYamn{mqpYXZlLB@HOqi!lw^2-Fakx
z^<ag$RCU9Qa>kL>*&3p)2!e>s(G8x6mo~M+q`NJ)6|9IzDt=_@;miS`a%xCd)VCb#
zO0!PcLL2XCp~}>sh)xw~Y~{Hh&FmU#kR<zpNoKxF-54Z3Mgv-?;d6S6i?2|kjsniZ
zo}ts85l3I&JKKq7N$vP3&<7wW1iwTi)4XYNjY#laQ{eUt`+dAQ7Rh_~EX{JCOFRuI
zMpB0n0#Jrmu{N7caq?q%Q~k|lpt8WcMI`G@z{wDVhF8yO>w(>L**42>K%ZZA-e4)S
zLkh8~S#xMIupW7VjWa4CoaCWIuio^*`qA<@<@<REdCcn3Yu>kyO-&VoRaT6B7;JsQ
zYfB+B-Q=ZEfuV4!F(;m_kEevOl=-OezB{T455#|bU@CghaiY1xWKskfO^-YVQ)3?X
z!;iC4b~m1<B{v_HHEH0x8>*^Ican4<L1^vTZ8XQ>Tbp0b;|Vu<&fD>Zbg8@B*XfWJ
z4ppl}tl}QuICkVWb19pSB%-iuh8$?tDvA_;iJ39=b!jc-L+~K}WW9YbCxZB*PqeAl
zKEzAR?ftp(;x4c;wnrOKw@5Fwh!iKPl>34QjN;rR`CmZqH@|Y(VWmQGB5{TP6TRqy
z<5(GRry6?X#~X&xLE=0bng$i-($!=qL~<W63ACEx+tgF0SHdGBKjGi<&kj_y6=w40
ztIPf5SpX*>$;&oom0{=<DPz#s_GWGPUS|3zl@q%lj@TbQft$X0)mf9kcWE@ysVvK`
zA@h|7K_6d3!RVMHX{;`-kE*&$TDZ9;x!iDKcZ3%b%tJ)qaFtk<iB3|Ha4RGq?+i_z
zpXS~Q&q2Dm=c2j!nd>$_v2F8hZ;os`kd`n<Z9<r-&_!asM&gtp?vS9W@}=2yt26KT
ziZf(yn(MYshRH~@q{GH!AKFh>nvS*M%Hkqt<Qj;<7%^9v%TxnE29e*>1iPYYfD<%b
zi~0Nc`xBumIKd!dtIU}3$Sdgw3ss|-JXhEC@{25g?@d=s<K|eL=A<$7m#^z4S6xt8
ze5*T~rFXLsjxeKm;iKla*T^RrFy10*x&a$&c#b98WZxG@DpxI8xKFU3eLH17O+>^o
ztgIlX?QWT9ff;mGkyShMAnFjqw+GG^Kc^FW_@#gb?_{r(ay$Gbd$W$SwFJwMz5VlF
ze1{0kQD!;BI@IXT8KkzG%ULqfP$6lAzEuwMy@=(aZF`~l3IP3khBKr@u_?vO8TmSS
zF%`<3U3}l8Z-QD;ln9mavgYHFbOoJft|}9y{2k<p@1Ry5+<#QKj=B6pJ%g$~UEfm=
zAP;^`c;>?**IVXyXAoJL;l3mZc1tVx((z<>lFrPKxO2rEOSxA9lr`z-aea+<WN(tp
zyss301~qH?i!<+6dED&+9<i}SO{Tag*2h2JS`&!W6JY1pHE$lQ?3-xzshjXD&<$b5
zzpX6!O|-h_PeH?l3a~xVsgtM6;JL(g;n9~0T&m275Tjl@zD*1#XP%;EybXPy54ubE
zSeQ~ytwFEYXTS@`gb>0*S`IL~*K=uq-h!NT(Qsv9WJJjBFW}B>d8L2L#JXdZNsDsk
zcNrpk+}Q=+1ID&t(+zFC0N@4>i3wlK3vAd2g#O@P47Mnv(MKXBEwSnLp0maNaSO)I
zI5Cdt?i^V-Bg5(G=^hDIT3gOXB+euxnQ<Jj7CiZK&08c~KETM=QzKK&_1OjL%|Y{c
zp;h9@E`Gy<C-SdRxKPEVFz8>Y7MH`fY$)8&`*pW9bcvmc1Qt#MDEQ@}TaH{eRM#7T
z$kXCp^mPjZNcM0h?H5RlHV759Vx%c4yx_J=FtnC+dwH82$Y?Y8w+BTTvn4KGm6Ma3
zoyc|_f)Z@{=WBW?z#GU0Tdny@O;yeX2Ve_NgkTGwB1dA_Uvrf+s>St7pCc;;f(^$u
zqsRNAmwn3^=fOu{d_f#wwAv$uC!zRJYd|`_^0{XTFOma{r^{XmgL{y<aF-;&|M~?h
z+~a8eQdQqxfhFef1|i>lWiNXn9*yUoC%fMDr8s8#F?{&wtW@bEw;cLP>h==t6Wyk}
z@pj|n2|eRz3ok5%c~_>repT5+fO*1Q+Hk^>%~Vw}yHb*WLVdFAg5;d2#0xRO)$zUT
z_V!;iQ9-su3s1wOOR)Rv0R&;22TM^w@$ai(p^+dl$SOQCc~*G<<FL6|mL}H)M+jH$
z2{#Mi_+dW7G?5tDY>BPq<iX-WuUL7%;M#Ceb1a+|LnG*KCmjb*`x)<<OmYcjlKp<I
zb{>2^3C&-31AQ(^`6(fA=oXmt#L3-r;_DV5Fug%5&1(!|8r+__cCkKd^m2>9M_ydj
zt6hs85y0q`)qbd#H~;(q$PwIbF7NUW+`+{uHHQ2++5|PQRa{UOsoO9>7lv8IzuCBU
zyHsxJZGr688a{1T>;6Zo3%o|lJ%v~IwN87jj+669DZXgz(lPUrJwj7u;`wZXmZo8C
zrLV+ZpTTpv!{5l`uNY-1+4gd49VqtvOld8k;*-#C*pmo2snswa{|m|-jMVB_B-VI#
z5pv29gH))~7&`%X!q~R37hW__LLkF}o%!tjZy3&xG6|--fz}BWqaX+h>2-ddUGIf-
zxz<yN0&mwzN?pXlulT61W?=4DFMZoIZPYa{DepvahvSG9_mjZWq1T6Bt>4?ra;DpH
zeSKux@yfH!9F4OD?QS45Dj(9tJ^G$Y+}FENRi&9KH%4D=(v4}*TjUz>81Jk)y42^1
zf85{<#8~mScQA<jz{qb%9Ej5*+4Fv7eEoaeKE8Q?NqVq-^Ky5xUytR-$ng{e2NGp$
z52#}G5#|0id|24eCfKvw2-RTOr`LA&7W?-YbeP1UTdKQvpTZuOU;J<ntzVU?H1nZ)
zx>Ip1N~;8re*nA^zUI@;X#B40DzJKgkxQJ^ro7*ol3TngK2UDQ+pkiJStTh5Kzy>n
z`I&+jNxaXp$2fM^WFq+v%ON{Bco0^5hwM{Km+UjOL<uL1TshDa2hrcAyOdgir#PUc
z8yJC@K2D$es@GTRS33wvv~OLfxcD|GvO_(vMk{f6WA|hPDiQyZIZZq4x9x*i{oxKf
z^Usw}W8Z>s$bSfYJ8RBg0{XoLO!vjak&SmW%g+Kn0xM=<2_<P#E-?i<E8T28v>%;T
zyDT9mE9E#66BWGQC<7K!NucAj-3Id%kY93fHwUWSoeqcVJZ(_cI+@)8nB%OO)v!3r
z-phq@l*EYuFHdV1U1s$dC62sgm&e5QiAD0(R5)H6xStlCPUW%S=!2`-bd4@hziX(U
zn(P;FRA4HMEP+k8>&lds0;<NAXe3eedtvLDQA41M&sn8G<UTynVrWYj^Rv7c@bGwe
zETc0JnbHRGVkeF^xUlfHNo_8ByRR*$C8o0FX2;@OH~EUJ)Ekl|MW%+`3<FU8Hmi<v
z{U0W?+l1*FT-KKxU7!hOozLocTD8TVwn$o<pP?ib`wi{@=ovgHMl4tX9~ix>y%LDD
zWhw_|ODzs~yQ&iPc;k0JNAjxRb4j~0tCG*UzYq0v6PJ_6*Hq0X)MeUH=r{w)F(kj9
zptoUs;L$DBTw*^NCqF6Rzit{yIo33et=>{?Gr*4pxvGBY@h)`V%>pCLJP+Mqj1EMF
zxKrVm+gHAI*ohKbyMueCZ?e0f*DM5ZRhf*|gP-2yb)!&F12B%1f71SS`8iLWom9DT
zO(U7coJhxHN4~zuj0ezpt3sH?oK-`$QMH-(&~_wY=v`ik3vbJfeJnp0luj4S<ZAZh
z>KGqLBdAC0m5{ospOWl%>KR|0S?PLL>Q=XeSYUi57<1>Gg}GkaO13-%yo9KaQ`6;p
zr2fLCa{pv`EYp4e@;ZEV<urF;&Bufz>eoz^7$f;boA-7`9UZFbxL(3}?(KN9@zSLT
zaJcoD>kqG4M^9|+HjSrAcm4PvlqpF39POdelODQ?9@a*rzs6tNfLPTyZS@Rds5fxF
zi|n(#dKZF@QP3Q@ni0Iqr*Al=KTiww&~`y}&L<Bm6`TOM3PXoUM!%sBFZ^Yt^@`5Q
z^%h1HjyO-!{1f)0ll#f-2eZ5EZWsApx&uK%BkU>`sMv(jJgG1OyC5#(Tjnt`(TW0-
z>#!v$pf=x4bO_>#RZ7)`2RYje6Ra33{cM`oX2GmgQoqKdcw|W@wSw@dhjpO;?3pg!
z+ePtPcPeEd9AzgZ7k6^Q&B$_iE;fck-TZ}|&tY&+DrqtOs6P>GJh4q~vX3=yRnzdY
z5~YYdnty`Qr&`o$|7zFGdYc!g>M2YpX>PU5=!C9b!Igdu;QP&kpo+Sfmd$SefJXj0
zS(7qUZY5v382kXU$Ed(b>LPHE#2PoxPcZg;Ikir39kY^gxh(AK)xe{qZ@%t^AKT^f
zwnoa*iuy%DDDjpywknR5B(!nF)x)vlwtpV2!VU{Mtpx}m&6gX9n~?gj<VP3VpYO6O
zuKTjEOC~=!_ddp(x^MbJZ{*N2+wle_tJ6GvFSuN1_8$^ZGd5K>D80~PCy!=|V?QTk
zbcp2!HQAfJ{p_8Esn;p4v`eZ~xiflI9;9sEp`!%|BRpXd4oW^HAjOQ|Z_D{{j$h%@
z&=^G^{l#~O(JcuK!mCeSO?HevAv|UgeiQLBkYzpB`}N&u3BGEZNPej0@G!>mtHj)o
zcb$I7WQ|X~oGS<#`@v=%R8?+jCPTIfcCl*p^dB_8&|H5efy*tjMq83NS-oo%ON;np
z>>3eBb$Tr8b@`8yHMim@5}u#p;$FZ`HqG{-?FZbx7qOhVKZnb~o<V(c%dk>c0+=?5
z=qZ!GPRnWX5&-iiEt-u)rf$atP#E=9E)@mFFiRZldfF4gD)oxS_KB2<i?eV&*IKCl
zZ>WPVdy0gTmNg2bHHo1lDD$SMW5Rx0-rJ9n4W=2~6*53_A`EiK+7uEJ2IRIVdLK9X
zA$yQ*B3pJ{f2N`+y~f9#>b;*;;}YdJ;hwGT^q?1x3l}F_L*riR;H(Qnp}KRiIffJT
zG5P*up)6=7O<isYyck#4X91pMfBrnZU~6?f<9BMRi4U24#J&Cco6A)gm8x7_U_*N7
z>Qs!!{lR@1KhRs{nun0lC)T~*FcJ>kg)=^phLZo9t-fN~HK45*Tx|=?`%$6bqYN0T
zHtwj&IwNF2?<=;Q=w#GaE3<a3q(mV>>_7)`R)PehO}{p2><88iYUa~dYG1@9m2v`q
z;7VRhoE||jSI6a<txo-z3d1h4^uVM9aSj<%ZXxwI8+AgIH232qa?MAi^$7C}eON<t
z{E1Vm4zDF0D;!)bnhzG2Qe2gA*uB8$6U`Ya8A4$&P<GS)I5BpZhUM151KK=7j`@r;
zzk+Fqffx+UI?@%dXlR(dFx8WAvKfzCs1Cg&&w(6UtF13a(5W|G?P6Hxd0ur%2#D34
zeq=;Hvd{axX|-jAhP)--mHhmAcTrKcv|4jExmYzZBHI^~tT#K^3n<hehsb!laeqOK
z-`VDcOMhJ<g}+!k7!`f+FQPC22F+Qj_!D_L!8G3wddiCLEczdhG7uhk(*wyL3`*I|
z-Lrk4`iFDRg!c?%(Q|Keu~y|uc?5egnHH6kQ(MpJm+Uq_5akncd-N&ybeWVSxa*MS
z_v8jy%_-(x-wW))c^$)Q%*Et0{nas}wwb;u&xL-a(LZx(BUpp+>-Lo*fByEzQ$K!y
zqemt!%|9hKKLg00QDiQ{x49Ja>AzXZL?jnL<iz${<n)89vGapH2=*SW`^^YP>qn*T
z-#=~_c~krXzyAG#0D90>t8$#Ucj4d9{JIF$t3{^&aapn}e83wFdVYlY-*5P95dM@D
z|NH$fJ^`u|{YPi({~j<|MAh{_E-U;0zVN?C=Ksb!%zl{Uk9*&#ZwBy?Fi^^(Ap9fD
zjW76Zqn><X1E*FbNC6>UELbC71Uc-U@+ZF`JD8aaH;hE&{t#=(R_zAA8OF{ErxE#B
z(>z^4B+?%n8|#(Lp|BG(W0uq@G%;qDa1;pr)ByM_!puQb(|rn{JUR-|a!77hG!Vg?
zKE4fh@BviY{KKoD$G{V_3`)!1SqOA5jxKWS{P&_=0LUQL@ur0)QefWs*cxGvTL98i
z;<aMih{Huox=ehZD{D_z+=KW7P#1x;`~t9+xDkoKQI(hmP+HrQmiDU!Rup$_tS)U5
z<vVtn<);MhaPo+tNRYvGKTDqF?`Yd!wx7yFt`i`%CMZL4feldzY7R&NVYxCOINtpX
z)So1s<#u}UfY&`bB6$VuLWKv;y8$H$qgUYW4?T`N6sx;gD~eq;ssB)*Wp|S--zn%0
zOHKSO2JAGgo7sgzXoyVyJ=Kp?8R)86`JZ6_f^^6(59{LLP__jv6UuV-mTdqb?_s8l
z+3(yAg1Z!0=^n?sci3zhA^wc)$>6GcrBEOB)&yMq2yw%boa#5g4!BR7JWscvM9Pw@
zLi3M3iTw)f<{1<gzh_K9*@1SV9D@8@`;+sXg;@1eX+qpr5G=?rQSK;v0OAD2_quiE
zslWd&&6&zttryg^tE4tZm(T+hPENFw<H@QM#gYei#&sMhafY5LV(vfd*04vLXZTZG
zgbXv~G3qQ7IOlGHwlw_lr`Soka`%N2i)KXfIqu1n^ph$9%M%)2_C<d@QK|g*I}-&w
zV(<Vm0XHu%@6m61K$p`8iD<XKDYzuxZ%IC@2i_FWa@C=<8Idu+*jls8K~#UPgWz;s
z0*%+QF)z(D5N5)_bh!)=t=vZ7hAV@shyrQ^3=8BS4I2M8-}6ak<I8KPux%U02$`M$
zovuNJqadi-8Ksn<SexEz*xmFu)uAO(qOFz!=L7B(8j`?yF$Yaa`nQ%H1w*(IalM$9
zGi_tXAG<e@+xipB5uJ1pHS4*nppO4k$ZK#&(mA6Bil9LWc62MiUV4!01iX+p6Oje5
z`dHd$z`ZF7k_R1dHZl)PgW7R9J`$vzVP9uVw+)c=e}Iq?saEoG5Paq@0*0j6Cg#vD
z+TrgA3w|6N?qd;a=+A=@{K$fVFQ5y3vThm{$l*cHYOtJ2;#BkXE=c}GFunk1qz4$_
ziktuV9_L^hG?tEC{3C3+%C0YWl{#h~$XOJkys8OfBR4_h3dRf$LcNr%9(5Gj<;4_#
z-J@l5M3rwrK_v!yP05v%z81>1A;N(8*7qUN+^QF}ZfpijXS|U&H3~5WnM5sjD2*f7
zlrcV~sQ<czNthXRoh{urM0V@2l7Mtm-9Bv)2Yp-~vB?a-r)CztC=E9(doF(`e2G$p
zx!$+Y`+F|Qva6CkX1KHBsHa7yKsXe_cxR>GcDTRSqy)MvWFcHYlLLO900lR6=V)yL
zfSA|(Drk>Gn0{?1Jq^nl^Da85ht})1#hp5@23=+wcrf1JRn}zeK4j6{AE`Sa8LE|?
zOmTO6DRsMd4eZZUd>12A>21%G@2rSvvO{rsw;X4Ke#9%t4~<50&}=eE#4&{_qG~Uw
zftWO6WQ$M{q2mj#$*sQndw|Jk_yn9I?B?IYR%hLzmVgp>Gk`pB5EM|pD<IY~9TNjf
zn%ZvzL(nYVHy#8AzeYi}4-$%)p>VF9QDN8M41h-BTDN5LJ+uNTsMB-X&dlUzuQWtA
z1k{PD%Fix?VXd;9TtI;%9-RzyWHHQx+TlDsq1tnh75phUN}EM;HOGd1f6QvW2wC$w
z5x4sM{MV)(<rNOV{JEC+eI_DcP=jv_k0MvSKB}5vhk2d@Ifb}s0j3X&s=%QPjZm1K
zZH@Z`9V;>WS3=p&1_WrC@NRm>t0CuZ!O}h|%y+m_5HnllsI7DI&@$zxPHLdIB!ET8
z@m$r}hq0!*GZhX4JjGb&>jR~szehSe4mQv?N!B}K?=Ra)wx2qmr(aD-K%A=Dchjrm
zWkXF4lq%r>z3Non`;1WM00g6Ahu$5TWcYLBG6mKTeqR_lKoB#C&jxVd@&gcNto^4<
zT1jd2#;csCF4C?4&>Snt$)ZYaQ}mqri?83)GvB$~jVQQjtQI%a?(MuoUB~`Ir>@=c
z%UxaiA%X#A3Bveixa6cUC~d!!%ki1Ooc}6qkBDb}-4=ES@^@iu3#RMeM*M_8mKR@Z
zqWuBbC<RF{>)~AeoSvZZw`LxuqdctfqhwDdcjl1S?mO~NviF3sG#t0Rew+URD;LIi
z)@y8<vqCWjfk1mvi=VOMAkVO9i90o%i)^%42Bi^d*EyI=;MvFv+=YT#wjF_7KnWyi
zf{ARX-$G}qj0bf<zDh^m0!#?$6rEl5I6vR8;zaK11TOLEdoI0e1MkpEm>$-=;j){b
zE*#_I`=a|ahB*!3R;+0S<Xg2z^pWiV>?itlaY>R|Cx9x_5k*?u1&yf3AFG^*FgwW?
zCcS1oYx#aBzHb69oRE_v6TwRF@8N=hJDyaBC?B`i@U`|unxqAX#qGnP*2b{g8ulnX
z=nu*c`wblo2OU)dfcUIpX*#pGA?VQog$;g$orDndwkDvG3u??HPDGBq>i|!$mruEA
zU@7#_+ehu}r3I&m0xIWyu8s<idWya2<~Ho$$Kf1}g}b5z;-OKoXo!6VD&g=FzWGtX
z6FjyQ7Z$*W74vOOtlN%F+&&IVfd875VI`);9<fRpNQ3#kVNaGi8{(2aCuVMC)tu<`
z;+Pk775df;Xdo|wWmoVc5}Kd0BY+#+DO5=~@9hL&*;h%xA37GmVhu1U=YWKLAGN5Z
zg+-q}>0@TH9_4kesTmOIuZo+^oDR*uTKsB+bpgFBwh3qGYW@4sEY?q;g%rP^Q@C9V
zj?->>)M+OSF)bKeV`K3o-o;j3o4U-lT`TYn_&G$u=L*K;m|&~0mQwzNOHz>OI!FK+
z+Y{_194-X_xn_VO1cDe_()xbg1-U}GuskUfXttN$vy~xbo28n_gKXPa9>2h2Q&Lmq
z$o{YSa-cyxYUlp`$d#AfW_K1BtkqpWu0W03-ngp82C^&@wbLKS2KXiB9PjS=9Vm4w
zz%oiW$9Lku?`5>Jpx{xbmVZEH=Tha(p1*h6ou!IA3g>~Xv1?53X8iwnJ_ULH|F^%_
zwhRC-4HZX6M}z&R(ea&&O_l#6f6Nvof!L?@%#46dPj2)E95S&71r0KA*aLS|6P4Zu
z{b!=BQAv)m@8bj_-{ZM~istWMKXvgOQaN+GX|r;Ce&^==Jq=-qFCGqnE>sO*S;r62
z8LQV6hYcJ)2t%AfS`{>!WvlmZKKbvVJc$U4!4FeF!NkPGfb6=mQr#pXl-QZmzaMp8
zl3jWo3K6_zYZ8;|5QtkDgpsRf*K1emibi~eoq`{)>QeDvZ^XnA4g?EiCorydW2j0+
zWCq|^+`KzdM7+A4;fkgWSy?p!5+Q!vrt{PPx<Wixn&)4oB_FK}dw}VGgf7HS=gxza
zB8CsgZczW%>tPlZ`Oj22L$k0ZLnkalxw@Mh<`Iw~9f83LJE}uBX3^*guf6AI7svnn
zx^#TSb>P0p!r)(!{Zen?-PEob<)S~&)SXcZ*e&!0@#%pi0F>mi$GyJvyym9*|9H7z
z5m>uS_vFReLwG8m$-=ig?4QmR|M*7WFxY;sNM^mSEd#t;VWX_j&Q2;+lINsCadTjH
z4uL<S6o1*uJwJHP%*Q+Z4IiY#My37G?H^a&S#q++$Vx61(^mia_dCzRBW?s?k>g9)
z2{Qiq(q~Ab!YoGk)6!yRbco6J(}O6QogVjJvkHC}&<Pu<U<mX2Kd-v;X?G#$5Zyz|
zo!|W9k*p2iRxuYseE!$<$si)&p;G)GsF-Y;C~PMxPv1%Z<ElF&AbT60KX51gABHM_
fd{zzSfNffI3FCn{j<vl+@K0G$O#v-$cKiPT<0s}b

literal 0
HcmV?d00001

diff --git a/docs/stan-users-guide/index.html b/docs/stan-users-guide/index.html
index 8f1c647cd..89ccd8743 100644
--- a/docs/stan-users-guide/index.html
+++ b/docs/stan-users-guide/index.html
@@ -74,7 +74,7 @@
 
 <link rel="stylesheet" href="../quarto-config/quarto_styles.css">
 <meta property="og:title" content="Stan User’s Guide – Stan Docs">
-<meta property="og:description" content="Version 2.38">
+<meta property="og:description" content="Version 2.39">
 <meta property="og:image" content="img/logo_tm.png">
 <meta property="og:site_name" content="Stan Docs">
 </head>
@@ -225,7 +225,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -343,6 +343,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -435,7 +441,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -467,6 +473,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -512,7 +524,7 @@
 <header id="title-block-header" class="quarto-title-block default">
 <div class="quarto-title">
 <h1 class="title">Stan User’s Guide</h1>
-<p class="subtitle lead">Version 2.38</p>
+<p class="subtitle lead">Version 2.39</p>
 </div>
 
 
@@ -538,7 +550,7 @@ <h1 class="title">Stan User’s Guide</h1>
 <li><p>The appendices provide an introduction to the stanc3 compiler used in the various interfaces to Stan, a style guide, and advice for users of BUGS and JAGS.</p></li>
 </ul>
 <p>We recommend working through this guide using the textbooks <em>Bayesian Data Analysis</em> and <em>Statistical Rethinking: A Bayesian Course with Examples in R and Stan</em> as references on the concepts, and using the <a href="https://mc-stan.org/docs/reference-manual/index.html"><em>Stan Reference Manual</em></a> when necessary to clarify programming issues.</p>
-<p><a href="https://mc-stan.org/docs/2_38/stan-users-guide-2_38.pdf">Download the pdf version of this manual</a>.</p>
+<p><a href="https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf">Download the pdf version of this manual</a>.</p>
 <section id="copyright-and-trademark" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="copyright-and-trademark">Copyright and trademark</h4>
 <ul>
@@ -1129,8 +1141,35 @@ <h4 class="unnumbered anchored" data-anchor-id="licensing">Licensing</h4>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/index.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/latent-discrete.html b/docs/stan-users-guide/latent-discrete.html
index 148e5785d..49bf88b9b 100644
--- a/docs/stan-users-guide/latent-discrete.html
+++ b/docs/stan-users-guide/latent-discrete.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1887,8 +1899,35 @@ <h3 class="unnumbered anchored" data-anchor-id="evaluating-the-conditional-expec
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/latent-discrete.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/latent-discrete.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/matrices-arrays.html b/docs/stan-users-guide/matrices-arrays.html
index a68551950..b638c4842 100644
--- a/docs/stan-users-guide/matrices-arrays.html
+++ b/docs/stan-users-guide/matrices-arrays.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1388,8 +1400,35 @@ <h2 class="anchored" data-anchor-id="aliasing-in-stan-containers">Aliasing in St
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/matrices-arrays.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/matrices-arrays.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/measurement-error.html b/docs/stan-users-guide/measurement-error.html
index 063fd2045..3fb2fee0e 100644
--- a/docs/stan-users-guide/measurement-error.html
+++ b/docs/stan-users-guide/measurement-error.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1421,8 +1433,35 @@ <h4 class="unnumbered anchored" data-anchor-id="extensions-and-alternatives">Ext
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/measurement-error.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/measurement-error.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/missing-data.html b/docs/stan-users-guide/missing-data.html
index edfdaf45f..38af5f613 100644
--- a/docs/stan-users-guide/missing-data.html
+++ b/docs/stan-users-guide/missing-data.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1369,8 +1381,35 @@ <h2 class="anchored" data-anchor-id="missing-multivariate-data">Missing multivar
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/missing-data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/missing-data.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/multi-indexing.html b/docs/stan-users-guide/multi-indexing.html
index 70ceb95af..ffaf5a741 100644
--- a/docs/stan-users-guide/multi-indexing.html
+++ b/docs/stan-users-guide/multi-indexing.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1414,8 +1426,35 @@ <h2 class="anchored" data-anchor-id="matrices-with-parameters-and-constants">Mat
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multi-indexing.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multi-indexing.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/multiple-imputation.html b/docs/stan-users-guide/multiple-imputation.html
new file mode 100644
index 000000000..5e4b37331
--- /dev/null
+++ b/docs/stan-users-guide/multiple-imputation.html
@@ -0,0 +1,1450 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Multiple Imputation</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/using-stanc.html" rel="next">
+<link href="../stan-users-guide/bootstrap.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Multiple Imputation">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/logo_tm.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multiple-imputation.html">Multiple imputation</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#multiple-imputation" id="toc-multiple-imputation" class="nav-link active" data-scroll-target="#multiple-imputation">Multiple imputation</a>
+  <ul class="collapse">
+  <li><a href="#motivation-and-outline" id="toc-motivation-and-outline" class="nav-link" data-scroll-target="#motivation-and-outline">Motivation and outline</a>
+  <ul class="collapse">
+  <li><a href="#pseudocode-for-outline" id="toc-pseudocode-for-outline" class="nav-link" data-scroll-target="#pseudocode-for-outline">Pseudocode for outline</a></li>
+  </ul></li>
+  <li><a href="#imputing-one-variable-in-stan" id="toc-imputing-one-variable-in-stan" class="nav-link" data-scroll-target="#imputing-one-variable-in-stan">Imputing one variable in Stan</a></li>
+  <li><a href="#imputing-two-or-more-variables" id="toc-imputing-two-or-more-variables" class="nav-link" data-scroll-target="#imputing-two-or-more-variables">Imputing two or more variables</a>
+  <ul class="collapse">
+  <li><a href="#imputing-two-variables-in-stan" id="toc-imputing-two-variables-in-stan" class="nav-link" data-scroll-target="#imputing-two-variables-in-stan">Imputing two variables in Stan</a></li>
+  </ul></li>
+  <li><a href="#combining-posterior-draws" id="toc-combining-posterior-draws" class="nav-link" data-scroll-target="#combining-posterior-draws">Combining posterior draws</a></li>
+  <li><a href="#cut-models" id="toc-cut-models" class="nav-link" data-scroll-target="#cut-models">Cut models</a></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multiple-imputation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/multiple-imputation.html">Multiple imputation</a></li></ol></nav></header>
+
+
+
+
+<section id="multiple-imputation" class="level1">
+<h1>Multiple imputation</h1>
+<p>Missing data is common in applied data analysis. Ignoring this missingness can distort posterior inferences and reduce their precision, and modeling it can improve inference quality. There are several ways to model missing data. Chapter 18 in <span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> offers an approximately Bayesian perspective, and the chapter <a href="https://mc-stan.org/docs/stan-users-guide/missing-data.html"><em>Missing Data and Partially Known Parameters</em></a> in the <em>Reference Manual</em> shows an approach that handles missing data as parameters.</p>
+<p>Another way to model missing data is through multiple imputation, which replaces missing values with sampled values to obtain different versions of a complete data set. Then, the model of interest can be fitted to each of these complete versions separately. Combining the resulting draws from these different fits produces a sample that accounts for the uncertainty in the missing values.</p>
+<section id="motivation-and-outline" class="level2">
+<h2 class="anchored" data-anchor-id="motivation-and-outline">Motivation and outline</h2>
+<p>Suppose that we have a data set <span class="math inline">\(x\)</span> with columns <span class="math inline">\(x_{\cdot, 1}, \ldots, x_{\cdot, K}\)</span> that make up covariates in a regression with quantities of interest <span class="math inline">\(\theta\)</span>. With the completely observed data, <span class="math inline">\(x^\text{comp}\)</span>, we could estimate the posterior distribution of <span class="math inline">\(\theta\)</span> as <span class="math inline">\(p(\theta \mid x^\text{comp})\)</span>. But with missing data, our matrix is split into <span class="math inline">\(x^{\text{obs}}\)</span> (the observed values of <span class="math inline">\(x\)</span>) and <span class="math inline">\(x^{\text{mis}}\)</span> (the missing values of <span class="math inline">\(x\)</span>).</p>
+<p>Fortunately, we can treat <span class="math inline">\(x^{\text{mis}}\)</span> as additional, nuisance parameters that we can estimate along with <span class="math inline">\(\theta\)</span> as <span class="math inline">\(p(\theta, x^{\text{mis}} \mid x^{\text{obs}})\)</span>. So, we can express the marginal distribution of <span class="math inline">\(\theta\)</span> given only the observed values <span class="math inline">\(x^{\text{obs}}\)</span> as <span class="math display">\[\begin{align*}
+p(\theta \mid x^{\text{obs}}) =&amp;
+\int
+    p(\theta, x^{\text{mis}} \mid x^{\text{obs}})
+    \mathrm{d}x^{\text{mis}} \\
+=&amp;
+\int
+    p(\theta \mid x^{\text{obs}}, x^{\text{mis}})
+    p(x^{\text{mis}} \mid x^{\text{obs}})
+    \mathrm{d}x^{\text{mis}} \\
+=&amp; \int
+    p(\theta \mid x^{\text{imp}})
+    p(x^{\text{mis}} \mid x^{\text{obs}})
+    \mathrm{d}x^{\text{mis}},
+\end{align*}\]</span> where <span class="math inline">\(x^{\text{imp}}\)</span> is a data set that includes imputed values of <span class="math inline">\(x^{\text{mis}}\)</span>.</p>
+<p>The equations above show that we do not need to describe <span class="math inline">\(p(\theta \mid x^{\text{obs}})\)</span> directly. Instead, we can first find a way to sample from <span class="math inline">\(x^{\text{mis}}\)</span> based on <span class="math inline">\(x^{\text{obs}}\)</span>, and then use these samples to fit the model <span class="math inline">\(p(\theta \mid x^\text{imp})\)</span>, treating <span class="math inline">\(x^\text{imp}\)</span> as if it was <span class="math inline">\(x^\text{comp}\)</span>.</p>
+<p>Note that the model for <span class="math inline">\(p(x^{\text{mis}} \mid x^{\text{obs}})\)</span> needs to be explicit. In a typical regression setting with covariates <span class="math inline">\(x\)</span> and outcome <span class="math inline">\(y\)</span>, this model for <span class="math inline">\(x\)</span> is not needed because inferences for the regression parameters are independent of the model of <span class="math inline">\(x\)</span> when <span class="math inline">\(x\)</span> is fully observed. But with missing data and multiple imputation, we need to generate values for <span class="math inline">\(x^{\text{mis}}\)</span> based on <span class="math inline">\(x^{\text{obs}}\)</span>, which in turn needs an explicit model for <span class="math inline">\(x^{\text{mis}}\)</span>.</p>
+<p>Thus, the general outline for multiple imputation with a number <span class="math inline">\(M\)</span> of imputations is:</p>
+<ol type="1">
+<li>Draw <span class="math inline">\(M\)</span> random samples <span class="math inline">\(x^{\text{mis}}_1, \ldots,
+x^{\text{mis}}_M\)</span> from the posterior predictive distribution <span class="math inline">\(p(x^{\text{mis}} \mid x^{\text{obs}})\)</span>.</li>
+<li>Use these samples to get <span class="math inline">\(M\)</span> different complete data sets <span class="math inline">\(x_1^\text{imp}, \ldots, x_M^\text{imp}\)</span>.</li>
+<li>For each of the <span class="math inline">\(M\)</span> imputed data sets, sample from the model of interest <span class="math inline">\(p(\theta \mid x_m^\text{imp})\)</span>, each time treating <span class="math inline">\(x_m^\text{imp}\)</span> as if it was <span class="math inline">\(x^\text{comp}\)</span>.</li>
+<li>Combine the draws for <span class="math inline">\(\theta\)</span> from all <span class="math inline">\(M\)</span> fits.</li>
+</ol>
+<section id="pseudocode-for-outline" class="level3">
+<h3 class="anchored" data-anchor-id="pseudocode-for-outline">Pseudocode for outline</h3>
+<p>Following the notation above, we can express the outline with the pseudocode below:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a>theta_estimates = []</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  x_mis &lt;- get_imputations(x_obs)</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  x_imp &lt;- add_imputations(x_obs, x_mis)</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>  imputation_fits &lt;- my_model.fit(x_imp).draws()</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a>  theta_estimates &lt;- theta_estimates.append(imputation_fits)</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Each of the functions in this pseudocode can encapsulate simple or complex procedures. <code>get_imputations()</code>, for example, may impute missing data by taking values directly from <code>x_obs</code>; or it may model a regression where an <span class="math inline">\(x_{\cdot, k}\)</span> with missing values is the outcome and the <span class="math inline">\(x_{\cdot, j}\)</span> without missing observations are predictors. Below we illustrate how to implement this second approach in Stan.</p>
+</section>
+</section>
+<section id="imputing-one-variable-in-stan" class="level2">
+<h2 class="anchored" data-anchor-id="imputing-one-variable-in-stan">Imputing one variable in Stan</h2>
+<p>Imagine our data set is composed of two numerical, continuous variables <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span>, and one binary variable <span class="math inline">\(z\)</span>. We want to estimate the parameters that govern the conditional association <span class="math inline">\(p(y \mid x, z)\)</span>, but several values of <span class="math inline">\(x\)</span> are missing. So, before we fit this model for <span class="math inline">\(y\)</span>, we impute the missing values in <span class="math inline">\(x\)</span>.</p>
+<p>We first find all the fully observed data points and identify their values as <span class="math inline">\(x^{\text{obs}}\)</span>, <span class="math inline">\(y^{\text{obs}}\)</span>, and <span class="math inline">\(z^{\text{obs}}\)</span>. Then we use these data to fit the model <span class="math display">\[
+x^{\text{obs}} \sim \text{normal}(\gamma_0 + \gamma_1 y^{\text{obs}} +
+    \gamma_2 z^{\text{obs}}, \lambda).
+\]</span></p>
+<p>This model can then give us samples for the missing values <span class="math inline">\(x^{\text{mis}}\)</span> conditional on the corresponding values of <span class="math inline">\(y\)</span> and <span class="math inline">\(z\)</span>, which we call <span class="math inline">\(y^{\text{aux}}\)</span> and <span class="math inline">\(z^{\text{aux}}\)</span>. The Stan code for this imputation is:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] x_obs;</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] y_obs;</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z_obs;  </span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_mis;   </span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_mis] y_aux;  </span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z_aux; </span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] gamma;</span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; lambda;</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>  gamma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>  lambda ~ exponential(<span class="dv">1</span>);</span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>  x_obs ~ normal(gamma[<span class="dv">1</span>] + gamma[<span class="dv">2</span>] * y_obs + gamma[<span class="dv">3</span>] * z_obs,</span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>          lambda);</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] x_imp = normal_rng(gamma[<span class="dv">1</span>] + gamma[<span class="dv">2</span>] * y_aux[n]</span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>                      + gamma[<span class="dv">3</span>] * z_aux[n], lambda);</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The <code>generated quantities</code> block automatically samples <span class="math inline">\(\gamma_0\)</span>, <span class="math inline">\(\gamma_1\)</span>, <span class="math inline">\(\gamma_2\)</span>, and <span class="math inline">\(\lambda\)</span> from their posterior distributions. So, the random draws of <code>x_imp</code> incorporate uncertainty from the estimated parameters and from the sampling variation in <span class="math inline">\(x^{\text{mis}}\)</span>.</p>
+<p>Multiple posterior draws of <span class="math inline">\(x^{\text{mis}}\)</span> give us multiple imputed data sets. With these data sets we can model <span class="math inline">\(p(y \mid \beta, \sigma, x, z)\)</span> as <span class="math display">\[
+y \sim \text{normal}(\beta_0 + \beta_1 x^\text{comp} + \beta_2 z,
+        \sigma),
+\]</span> where <span class="math inline">\(x^\text{comp}\)</span> contains observed and imputed values. The Stan code for this model is:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] z;</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] beta;</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  sigma ~ exponential(<span class="dv">1</span>);</span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>  y ~ normal(beta[<span class="dv">1</span>] + beta[<span class="dv">2</span>] * x + beta[<span class="dv">3</span>] * z, sigma);</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>With multiple imputation, the model for <span class="math inline">\(p(y \mid \beta, \sigma, x, z)\)</span> does not need to distinguish between observed and imputed values of <span class="math inline">\(x\)</span>. Instead, this model can treat all imputed data sets as complete because we can combine all the posterior draws from multiple fits.</p>
+</section>
+<section id="imputing-two-or-more-variables" class="level2">
+<h2 class="anchored" data-anchor-id="imputing-two-or-more-variables">Imputing two or more variables</h2>
+<p>A more general scenario involves an outcome variable <span class="math inline">\(y\)</span> and several explanatory variables <span class="math inline">\(x_1, \ldots, x_K\)</span>, each of which can have missing values that we want to impute.</p>
+<p>One solution is to do multiple imputation through chained equations, a procedure often called “MICE”<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a>. The MICE procedure in this scenario is:</p>
+<ol start="0" type="1">
+<li><strong>Initialize missing values in all <span class="math inline">\(x_i\)</span></strong>. For each variable <span class="math inline">\(x_i\)</span> with missing entries, fill its missing values with random samples from its observed values (or use another simple initialization rule).</li>
+<li><strong>Update each <span class="math inline">\(x_i\)</span> given <span class="math inline">\(y\)</span> and the other <span class="math inline">\(x\)</span>’s</strong>. For <span class="math inline">\(i=1, \ldots, K\)</span>, fit a model for the <em>observed</em> <span class="math inline">\(x_i\)</span> conditional on the current versions (with observed and imputed values) of <span class="math inline">\(y\)</span> and <span class="math inline">\(x_{-k} = x_1, \ldots x_{k-1}, x_{k+1}, \ldots,
+x_{K}\)</span>. Use this model to draw impuations from the predictive distribution of the <em>missing</em> <span class="math inline">\(x_i\)</span>. Completing this step for all <span class="math inline">\(i = 1, \ldots, K\)</span> constitutes a single imputation cycle.</li>
+<li><strong>Warmup period</strong>. Repeat the imputation cycle in step 1 several times as warmup to let the imputations stabilize.</li>
+<li><strong>Create M imputed datasets</strong>. After the warmup, record the current complete dataset as one imputed dataset. Then either <em>restart</em> from step 0 and repeat steps 1–3 until you have <span class="math inline">\(M\)</span> imputed datasets; or do a <em>single long run</em>, i.e., continue iterating steps 1–2 and save the imputed dataset at <span class="math inline">\(M\)</span> well-spaced iterations (e.g., every <span class="math inline">\(S\)</span> cycles) to obtain <span class="math inline">\(M\)</span> imputed datasets, without restarting from step 0 each time.</li>
+<li><strong>Fit the target model</strong> <span class="math inline">\(p(y \mid \beta, \sigma, x_1, \ldots,
+x_K)\)</span> separately to each of the <span class="math inline">\(M\)</span> imputed datasets and save the posterior draws for the parameters of interest.</li>
+<li><strong>Combine the draws</strong> (or other summaries) from all <span class="math inline">\(M\)</span> fits.</li>
+</ol>
+<p>Note that, as described here, the MICE procedure does not guarantee that the conditional distributions of all variables will be compatible. <em>Compatibility</em> means that there is a joint distribution for all the variables used in the imputation that can be decomposed as the conditional distributions we used. An incompatible imputation model will technically not sample from any well-defined probability distribution. But the consequences of this are not always serious. See section 6.4 in <span class="citation" data-cites="carpenter-etal:2023">Carpenter et al. (<a href="#ref-carpenter-etal:2023" role="doc-biblioref">2023</a>)</span> for a more detailed explanation of compatibility and of the related concept of <em>congeniality</em>.</p>
+<section id="imputing-two-variables-in-stan" class="level3">
+<h3 class="anchored" data-anchor-id="imputing-two-variables-in-stan">Imputing two variables in Stan</h3>
+<p>Imagine that we again want to use numerical variables <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span>, and dichotomic variable <span class="math inline">\(z\)</span> to estimate the parameters in <span class="math inline">\(p(y \mid \beta, \sigma, x, z)\)</span>. But now both <span class="math inline">\(x\)</span> and <span class="math inline">\(z\)</span> have missing values that we want to impute.</p>
+<p>To apply the MICE procedure in this example, we can reuse the models for <span class="math inline">\(x\)</span> and <span class="math inline">\(y\)</span> that we defined above. We also need a new model to impute <span class="math inline">\(z^{\text{mis}}\)</span>, so we use the logistic regression <span class="math display">\[
+z^{\text{obs}} \sim \text{Bernoulli}(
+  \text{logit}^{-1}(
+    \alpha_0 + \alpha_1 y^{\text{obs}} + \alpha_2 x^{\text{obs}})
+  ).
+\]</span> Here, <span class="math inline">\(z^{\text{obs}}\)</span> contains only the completely observed values from our original data set, while <span class="math inline">\(x^{\text{obs}}\)</span> and <span class="math inline">\(y^{\text{obs}}\)</span> contain all the values that correspond to <span class="math inline">\(z^{\text{obs}}\)</span>. Thus, <span class="math inline">\(x^{\text{obs}}\)</span> can include observed and imputed values, and <span class="math inline">\(y^{\text{obs}}\)</span> need not be the same as in the model for <span class="math inline">\(x^{\text{obs}}\)</span>.</p>
+<p>With this model we can sample values for <span class="math inline">\(z^{\text{mis}}\)</span> conditional on its corresponding values <span class="math inline">\(y^{\text{aux}}\)</span> and <span class="math inline">\(x^{\text{aux}}\)</span>.</p>
+<p>The Stan code to impute <span class="math inline">\(z^{\text{mis}}\)</span> is:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_obs;</span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_obs] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; z_obs;</span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] y_obs;</span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_obs] x_obs;</span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_mis;</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_mis] y_aux;</span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_mis] x_aux;</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[<span class="dv">3</span>] gamma;</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>  gamma ~ normal(<span class="dv">0</span>, <span class="dv">1</span>);</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>  z_obs ~ bernoulli_logit(gamma[<span class="dv">1</span>] + gamma[<span class="dv">2</span>] * y_obs </span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>                          + gamma[<span class="dv">3</span>] * x_obs);</span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N_mis] <span class="dt">int</span> z_imp = bernoulli_logit_rng(gamma[<span class="dv">1</span>]</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>                      + gamma[<span class="dv">2</span>] * y_aux</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>                      + gamma[<span class="dv">3</span>] * x_aux);</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>The pseudocode for this example of MICE, restarting the imputations after each imputation cycle, is shown below. Note that we do not need to initialize <span class="math inline">\(x^{\text{mis}}\)</span> in step 0 because we can impute it directly with the model for <span class="math inline">\(x^{\text{obs}}\)</span>.</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a>completed_datasets &lt;- []</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (m <span class="cf">in</span> <span class="dv">1</span>:M) {</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>  data_m &lt;- copy(data_orig)</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>  data_m.z[missing_idz] &lt;- random_sample(data_m.z[observed_idz])</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (cycle <span class="cf">in</span> <span class="dv">1</span>:n_warm) {</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>    stanmod_x &lt;- build_model(<span class="st">"model_for_x.stan"</span>)</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>    fit_x &lt;- stanmod_x.fit(</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>      obs_data=data_m[observed_idx],</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>      mis_data=data_m[missing_idx]</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>    )</span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>    data_m.x[missing_idx] &lt;- get_imputations(fit_x, <span class="st">"x_imp"</span>)</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>    stanmod_z &lt;- build_model(<span class="st">"model_for_z.stan"</span>)</span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>    fit_z &lt;- stanmod_z.fit(</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>      obs_data=data_m[observed_idz],</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>      mis_data=data_m[missing_idz]</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>    )</span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>    data_m.z[missing_idz] &lt;- get_imputations(fit_z, <span class="st">"z_imp"</span>)</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>  completed_datasets.append(data_m)</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>all_draws &lt;- []</span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (dataset <span class="cf">in</span> completed_datasets) {</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>  stanmod_y &lt;- build_model(<span class="st">"model_for_y.stan"</span>)</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>  fit_y &lt;- stanmod_y.fit(dataset)</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a>  all_draws.append(extract_draws(model_results))</span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb5-27"><a href="#cb5-27" aria-hidden="true" tabindex="-1"></a>all_results &lt;- combine_results(all_draws)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</section>
+</section>
+<section id="combining-posterior-draws" class="level2">
+<h2 class="anchored" data-anchor-id="combining-posterior-draws">Combining posterior draws</h2>
+<p>With Stan’s MCMC sampler, we can treat posterior draw chains from imputed data sets as if they were chains based on complete data. There is one important difference. Multiple imputation expresses uncertainty in the missing values as consistent differences in the estimates obtained from different imputated data sets. This means that chains obtained from the same imputed data set should converge, but chains obtained from different data sets do not have to. So, we need not worry if diagnostics<a href="#fn2" class="footnote-ref" id="fnref2" role="doc-noteref"><sup>2</sup></a> signal that the chains from different imputed data sets are not converging properly. See an example in <span class="citation" data-cites="burkner:2025">Bürkner (<a href="#ref-burkner:2025" role="doc-biblioref">2025</a>)</span>.</p>
+</section>
+<section id="cut-models" class="level2">
+<h2 class="anchored" data-anchor-id="cut-models">Cut models</h2>
+<p>A full Bayesian probability model includes a feedback flow of information between all parameters and all data. Cut models separate some parts of this feedback flow so that different subsets of data influence only some parameters in the model (see <span class="citation" data-cites="plummer:2015">Plummer (<a href="#ref-plummer:2015" role="doc-biblioref">2015</a>)</span>).</p>
+<p>Multiple imputation interrupts the flow of information from data to parameters. In our regression above, for example, the imputations influence the distribution of the parameters in <span class="math inline">\(p(y \mid x_1, \ldots, x_K)\)</span>, but these parameters do not influence the imputations. So, we could use multiple imputation to implement a cut model.</p>
+
+
+
+</section>
+</section>
+
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-burkner:2025" class="csl-entry" role="listitem">
+Bürkner, Paul. 2025. <span>“Handle Missing Values with Brms.”</span> <a href="https://cran.r-project.org/web/packages/brms/vignettes/brms_missings.html">https://cran.r-project.org/web/packages/brms/vignettes/brms_missings.html</a>.
+</div>
+<div id="ref-carpenter-etal:2023" class="csl-entry" role="listitem">
+Carpenter, James R., Jonathan W. Bartlett, Tim P. Morris, Angela M. Wood, Matteo Quartagno, and Michael G. Kenward. 2023. <em>Multiple Imputation and Its Application</em>. 2nd ed. John Wiley &amp; Sons, Ltd. <a href="https://doi.org/10.1002/9781119756118">https://doi.org/10.1002/9781119756118</a>.
+</div>
+<div id="ref-GelmanEtAl:2013" class="csl-entry" role="listitem">
+Gelman, Andrew, J. B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. <em>Bayesian Data Analysis</em>. Third Edition. London: Chapman &amp; Hall / CRC Press.
+</div>
+<div id="ref-plummer:2015" class="csl-entry" role="listitem">
+Plummer, Martyn. 2015. <span>“Cuts in <span>B</span>ayesian Graphical Models.”</span> <em>Statistics and Computing</em> 25: 37–43. <a href="https://doi.org/10.1007/s11222-014-9503-z">https://doi.org/10.1007/s11222-014-9503-z</a>.
+</div>
+<div id="ref-vanbuuren:2018" class="csl-entry" role="listitem">
+van Buuren, Stef. 2018. <em>Flexible Imputation of Missing Data</em>. 2nd ed. Chapman; Hall/CRC. <a href="https://doi.org/10.1201/9780429492259">https://doi.org/10.1201/9780429492259</a>.
+</div>
+</div></section><section id="footnotes" class="footnotes footnotes-end-of-document" role="doc-endnotes"><h2 class="anchored quarto-appendix-heading">Footnotes</h2>
+
+<ol>
+<li id="fn1"><p>Section 5 of chapter 4 in <span class="citation" data-cites="vanbuuren:2018">van Buuren (<a href="#ref-vanbuuren:2018" role="doc-biblioref">2018</a>)</span> details the MICE procedure in a frequentist context.<a href="#fnref1" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+<li id="fn2"><p>Such as <span class="math inline">\(\hat{R}\)</span>. See <a href="https://mc-stan.org/docs/reference-manual/analysis.html#split-r-hat-for-detecting-non-stationarity"><em>Split R-hat for detecting non-stationarity</em></a> in the <em>Reference Manual</em>.<a href="#fnref2" class="footnote-back" role="doc-backlink">↩︎</a></p></li>
+</ol>
+</section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/bootstrap.html" class="pagination-link" aria-label="The Bootstrap and Bagging">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">The Bootstrap and Bagging</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/using-stanc.html" class="pagination-link" aria-label="Using the Stan Compiler">
+        <span class="nav-page-text">Using the Stan Compiler</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/multiple-imputation.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/odes.html b/docs/stan-users-guide/odes.html
index 6b50a7f03..164f13882 100644
--- a/docs/stan-users-guide/odes.html
+++ b/docs/stan-users-guide/odes.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1497,8 +1509,35 @@ <h2 class="anchored" data-anchor-id="solving-a-system-of-linear-odes-using-a-mat
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/odes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/odes.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/one-dimensional-integrals.html b/docs/stan-users-guide/one-dimensional-integrals.html
index 19a3910ee..b48f60b63 100644
--- a/docs/stan-users-guide/one-dimensional-integrals.html
+++ b/docs/stan-users-guide/one-dimensional-integrals.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1368,8 +1380,35 @@ <h3 class="unnumbered anchored" data-anchor-id="integral-precision">Avoiding pre
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/one-dimensional-integrals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/one-dimensional-integrals.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/parallelization.html b/docs/stan-users-guide/parallelization.html
index 992015a2e..dc87548c5 100644
--- a/docs/stan-users-guide/parallelization.html
+++ b/docs/stan-users-guide/parallelization.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1590,8 +1602,35 @@ <h2 class="anchored" data-anchor-id="opencl">OpenCL</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/parallelization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/posterior-prediction.html b/docs/stan-users-guide/posterior-prediction.html
index 1c53ea2a6..b176170af 100644
--- a/docs/stan-users-guide/posterior-prediction.html
+++ b/docs/stan-users-guide/posterior-prediction.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1424,14 +1436,41 @@ <h2 class="anchored" data-anchor-id="stand-alone-generated-quantities-and-ongoin
       </a>          
   </div>
   <div class="nav-page nav-page-next">
-      <a href="../stan-users-guide/simulation-based-calibration.html" class="pagination-link" aria-label="Simulation-Based Calibration">
-        <span class="nav-page-text">Simulation-Based Calibration</span> <i class="bi bi-arrow-right-short"></i>
+      <a href="../stan-users-guide/simulation-based-calibration.html" class="pagination-link" aria-label="Simulation-Based Calibration Checking">
+        <span class="nav-page-text">Simulation-Based Calibration Checking</span> <i class="bi bi-arrow-right-short"></i>
       </a>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-prediction.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-prediction.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/posterior-predictive-checks.html b/docs/stan-users-guide/posterior-predictive-checks.html
index 36e0a391b..e76588b67 100644
--- a/docs/stan-users-guide/posterior-predictive-checks.html
+++ b/docs/stan-users-guide/posterior-predictive-checks.html
@@ -313,7 +313,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -431,6 +431,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -523,7 +529,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -555,6 +561,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1505,8 +1517,8 @@ <h3 class="unnumbered anchored" data-anchor-id="mixed-replication-for-hierarchic
 </script>
 <nav class="page-navigation">
   <div class="nav-page nav-page-previous">
-      <a href="../stan-users-guide/simulation-based-calibration.html" class="pagination-link" aria-label="Simulation-Based Calibration">
-        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Simulation-Based Calibration</span>
+      <a href="../stan-users-guide/simulation-based-calibration.html" class="pagination-link" aria-label="Simulation-Based Calibration Checking">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Simulation-Based Calibration Checking</span>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
@@ -1516,7 +1528,34 @@ <h3 class="unnumbered anchored" data-anchor-id="mixed-replication-for-hierarchic
   </div>
 </nav>
 </div> <!-- /content -->
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-predictive-checks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","descPosition":"bottom","loop":false,"openEffect":"zoom","selector":".lightbox"});
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/posterior-predictive-checks.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+<script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","descPosition":"bottom","loop":false,"openEffect":"zoom","selector":".lightbox"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {
diff --git a/docs/stan-users-guide/poststratification.html b/docs/stan-users-guide/poststratification.html
index f410ffbe1..179eaca67 100644
--- a/docs/stan-users-guide/poststratification.html
+++ b/docs/stan-users-guide/poststratification.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1465,8 +1477,35 @@ <h2 class="anchored" data-anchor-id="adding-group-level-predictors">Adding group
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/poststratification.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/poststratification.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/problematic-posteriors.html b/docs/stan-users-guide/problematic-posteriors.html
index 19076b1b9..651f38493 100644
--- a/docs/stan-users-guide/problematic-posteriors.html
+++ b/docs/stan-users-guide/problematic-posteriors.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1573,8 +1585,35 @@ <h3 class="unnumbered anchored" data-anchor-id="examples-fits-in-stan">Examples:
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/problematic-posteriors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/problematic-posteriors.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/proportionality-constants.html b/docs/stan-users-guide/proportionality-constants.html
index af4998be0..ece81d6b6 100644
--- a/docs/stan-users-guide/proportionality-constants.html
+++ b/docs/stan-users-guide/proportionality-constants.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1253,8 +1265,35 @@ <h2 class="anchored" data-anchor-id="limitations-on-using-_lupdf-and-_lupmf-func
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/proportionality-constants.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/proportionality-constants.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/references.html b/docs/stan-users-guide/references.html
index 9c8f97f9b..a4117cb99 100644
--- a/docs/stan-users-guide/references.html
+++ b/docs/stan-users-guide/references.html
@@ -787,8 +787,35 @@ <h1 class="unnumbered"><i style="font-size: 110%; color:var(--bs-body-color)">Re
 });
 </script>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/references.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/regression.html b/docs/stan-users-guide/regression.html
index 226edeb9d..379f4b6f1 100644
--- a/docs/stan-users-guide/regression.html
+++ b/docs/stan-users-guide/regression.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -957,47 +969,26 @@ <h3 class="unnumbered anchored" data-anchor-id="ordered-logistic-regression">Ord
 <span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
 <span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
 <span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; y;</span>
-<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">row_vector</span>[D] x;</span>
+<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, D] x;</span>
 <span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a>}</span>
 <span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
 <span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[D] beta;</span>
 <span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">ordered</span>[K - <span class="dv">1</span>] c;</span>
 <span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a>}</span>
 <span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a>    y[n] ~ ordered_logistic(x[n] * beta, c);</span>
-<span id="cb23-15"><a href="#cb23-15" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb23-16"><a href="#cb23-16" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a>  y ~ ordered_logistic(x * beta, c);</span>
+<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The vector of cutpoints <code>c</code> is declared as <code>ordered[K - 1]</code>, which guarantees that <code>c[k]</code> is less than <code>c[k + 1]</code>.</p>
 <p>If the cutpoints were assigned independent priors, the constraint effectively truncates the joint prior to support over points that satisfy the ordering constraint. Luckily, Stan does not need to compute the effect of the constraint on the normalizing term because the probability is needed only up to a proportion.</p>
+<p>The equivalent model can be written using <code>ordered_logistic_glm</code> distribution, which can provide more efficient computation in case of higher dimensional <code>beta</code>.</p>
+<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a>  y ~ ordered_logistic_glm(x,  beta, c);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <section id="ordered-probit" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="ordered-probit">Ordered probit</h4>
-<p>An ordered probit model could be coded in exactly the same way by swapping the cumulative logistic (<code>inv_logit</code>) for the cumulative normal (<code>Phi</code>).</p>
-<div class="sourceCode" id="cb24"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">2</span>&gt; K;</span>
-<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
-<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
-<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; y;</span>
-<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">row_vector</span>[D] x;</span>
-<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb24-8"><a href="#cb24-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb24-9"><a href="#cb24-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[D] beta;</span>
-<span id="cb24-10"><a href="#cb24-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">ordered</span>[K - <span class="dv">1</span>] c;</span>
-<span id="cb24-11"><a href="#cb24-11" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb24-12"><a href="#cb24-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb24-13"><a href="#cb24-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] theta;</span>
-<span id="cb24-14"><a href="#cb24-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb24-15"><a href="#cb24-15" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> eta;</span>
-<span id="cb24-16"><a href="#cb24-16" aria-hidden="true" tabindex="-1"></a>    eta = x[n] * beta;</span>
-<span id="cb24-17"><a href="#cb24-17" aria-hidden="true" tabindex="-1"></a>    theta[<span class="dv">1</span>] = <span class="dv">1</span> - Phi(eta - c[<span class="dv">1</span>]);</span>
-<span id="cb24-18"><a href="#cb24-18" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (k <span class="cf">in</span> <span class="dv">2</span>:(K - <span class="dv">1</span>)) {</span>
-<span id="cb24-19"><a href="#cb24-19" aria-hidden="true" tabindex="-1"></a>      theta[k] = Phi(eta - c[k - <span class="dv">1</span>]) - Phi(eta - c[k]);</span>
-<span id="cb24-20"><a href="#cb24-20" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb24-21"><a href="#cb24-21" aria-hidden="true" tabindex="-1"></a>    theta[K] = Phi(eta - c[K - <span class="dv">1</span>]);</span>
-<span id="cb24-22"><a href="#cb24-22" aria-hidden="true" tabindex="-1"></a>    y[n] ~ categorical(theta);</span>
-<span id="cb24-23"><a href="#cb24-23" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb24-24"><a href="#cb24-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
-<p>The logistic model could also be coded this way by replacing <code>Phi</code> with <code>inv_logit</code>, though the built-in encoding based on the softmax transform is more efficient and more numerically stable. A small efficiency gain could be achieved by computing the values <code>Phi(eta - c[k])</code> once and storing them for re-use.</p>
+<p>An ordered probit model can be coded in exactly the same way by using the built-in <code>ordered_probit</code> distribution.</p>
+<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  ordered_probit(x * beta, c);</span>
+<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>There is not yet an <code>ordered_probit_glm</code> distribution in Stan.</p>
 </section>
 </section>
 </section>
@@ -1007,52 +998,52 @@ <h2 class="anchored" data-anchor-id="hierarchical-regression">Hierarchical regre
 <p>Suppose each binary outcome <span class="math inline">\(y_n \in \{ 0, 1 \}\)</span> has an associated level, <span class="math inline">\(ll_n \in \{ 1, \dotsc, L \}\)</span>. Each outcome will also have an associated predictor vector <span class="math inline">\(x_n \in \mathbb{R}^D\)</span>. Each level <span class="math inline">\(l\)</span> gets its own coefficient vector <span class="math inline">\(\beta_l \in \mathbb{R}^D\)</span>. The hierarchical structure involves drawing the coefficients <span class="math inline">\(\beta_{l,d}
 \in \mathbb{R}\)</span> from a prior that is also estimated with the data. This hierarchically estimated prior determines the amount of pooling. If the data in each level are similar, strong pooling will be reflected in low hierarchical variance. If the data in the levels are dissimilar, weaker pooling will be reflected in higher hierarchical variance.</p>
 <p>The following model encodes a hierarchical logistic regression model with a hierarchical prior on the regression coefficients.</p>
-<div class="sourceCode" id="cb25"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
-<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
-<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; L;</span>
-<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
-<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=L&gt; ll;</span>
-<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">row_vector</span>[D] x;</span>
-<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb25-9"><a href="#cb25-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb25-10"><a href="#cb25-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span> mu;</span>
-<span id="cb25-11"><a href="#cb25-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb25-12"><a href="#cb25-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[L] <span class="dt">vector</span>[D] beta;</span>
-<span id="cb25-13"><a href="#cb25-13" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb25-14"><a href="#cb25-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb25-15"><a href="#cb25-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
-<span id="cb25-16"><a href="#cb25-16" aria-hidden="true" tabindex="-1"></a>    mu[d] ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
-<span id="cb25-17"><a href="#cb25-17" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (l <span class="cf">in</span> <span class="dv">1</span>:L) {</span>
-<span id="cb25-18"><a href="#cb25-18" aria-hidden="true" tabindex="-1"></a>      beta[l, d] ~ normal(mu[d], sigma[d]);</span>
-<span id="cb25-19"><a href="#cb25-19" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb25-20"><a href="#cb25-20" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb25-21"><a href="#cb25-21" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb25-22"><a href="#cb25-22" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));</span>
-<span id="cb25-23"><a href="#cb25-23" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb25-24"><a href="#cb25-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb26-2"><a href="#cb26-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb26-3"><a href="#cb26-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb26-4"><a href="#cb26-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; L;</span>
+<span id="cb26-5"><a href="#cb26-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb26-6"><a href="#cb26-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=L&gt; ll;</span>
+<span id="cb26-7"><a href="#cb26-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">row_vector</span>[D] x;</span>
+<span id="cb26-8"><a href="#cb26-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb26-9"><a href="#cb26-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb26-10"><a href="#cb26-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span> mu;</span>
+<span id="cb26-11"><a href="#cb26-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[D] <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb26-12"><a href="#cb26-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[L] <span class="dt">vector</span>[D] beta;</span>
+<span id="cb26-13"><a href="#cb26-13" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb26-14"><a href="#cb26-14" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb26-15"><a href="#cb26-15" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
+<span id="cb26-16"><a href="#cb26-16" aria-hidden="true" tabindex="-1"></a>    mu[d] ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
+<span id="cb26-17"><a href="#cb26-17" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (l <span class="cf">in</span> <span class="dv">1</span>:L) {</span>
+<span id="cb26-18"><a href="#cb26-18" aria-hidden="true" tabindex="-1"></a>      beta[l, d] ~ normal(mu[d], sigma[d]);</span>
+<span id="cb26-19"><a href="#cb26-19" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb26-20"><a href="#cb26-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb26-21"><a href="#cb26-21" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb26-22"><a href="#cb26-22" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli(inv_logit(x[n] * beta[ll[n]]));</span>
+<span id="cb26-23"><a href="#cb26-23" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb26-24"><a href="#cb26-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The standard deviation parameter <code>sigma</code> gets an implicit uniform prior on <span class="math inline">\((0,\infty)\)</span> because of its declaration with a lower-bound constraint of zero. Stan allows improper priors as long as the posterior is proper. Nevertheless, it is usually helpful to have informative or at least weakly informative priors for all parameters; see the <a href="#regression-priors.section">regression priors section</a> for recommendations on priors for regression coefficients and scales.</p>
 <section id="optimizing-the-model" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="optimizing-the-model">Optimizing the model</h4>
 <p>Where possible, vectorizing distribution statements leads to faster log probability and derivative evaluations. The speed boost is not because loops are eliminated, but because vectorization allows sharing subcomputations in the log probability and gradient calculations and because it reduces the size of the expression tree required for gradient calculations.</p>
 <p>The first optimization vectorizes the for-loop over <code>D</code> as</p>
-<div class="sourceCode" id="cb26"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a>mu ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
-<span id="cb26-2"><a href="#cb26-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (l <span class="cf">in</span> <span class="dv">1</span>:L) {</span>
-<span id="cb26-3"><a href="#cb26-3" aria-hidden="true" tabindex="-1"></a>  beta[l] ~ normal(mu, sigma);</span>
-<span id="cb26-4"><a href="#cb26-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a>mu ~ normal(<span class="dv">0</span>, <span class="dv">100</span>);</span>
+<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (l <span class="cf">in</span> <span class="dv">1</span>:L) {</span>
+<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a>  beta[l] ~ normal(mu, sigma);</span>
+<span id="cb27-4"><a href="#cb27-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The declaration of <code>beta</code> as an array of vectors means that the expression <code>beta[l]</code> denotes a vector. Although <code>beta</code> could have been declared as a matrix, an array of vectors (or a two-dimensional array) is more efficient for accessing rows; see the <a href="../stan-users-guide/matrices-arrays.html#indexing-efficiency.section">indexing efficiency section</a> for more information on the efficiency tradeoffs among arrays, vectors, and matrices.</p>
 <p>This model can be further sped up and at the same time made more arithmetically stable by replacing the application of inverse-logit inside the Bernoulli distribution with the logit-parameterized Bernoulli,<a href="#fn3" class="footnote-ref" id="fnref3" role="doc-noteref"><sup>3</sup></a></p>
-<div class="sourceCode" id="cb27"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);</span>
-<span id="cb27-3"><a href="#cb27-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(x[n] * beta[ll[n]]);</span>
+<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Unlike in R or BUGS, loops, array access and assignments are fast in Stan because they are translated directly to C++. In most cases, the cost of allocating and assigning to a container is more than made up for by the increased efficiency due to vectorizing the log probability and gradient calculations. Thus the following version is faster than the original formulation as a loop over a distribution statement.</p>
-<div class="sourceCode" id="cb28"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a>{</span>
-<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_beta_ll;</span>
-<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb28-4"><a href="#cb28-4" aria-hidden="true" tabindex="-1"></a>    x_beta_ll[n] = x[n] * beta[ll[n]];</span>
-<span id="cb28-5"><a href="#cb28-5" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb28-6"><a href="#cb28-6" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(x_beta_ll);</span>
-<span id="cb28-7"><a href="#cb28-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_beta_ll;</span>
+<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a>    x_beta_ll[n] = x[n] * beta[ll[n]];</span>
+<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(x_beta_ll);</span>
+<span id="cb29-7"><a href="#cb29-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The brackets introduce a new scope for the local variable <code>x_beta_ll</code>; alternatively, the variable may be declared at the top of the model block.</p>
 <p>In some cases, such as the above, the local variable assignment leads to models that are less readable. The recommended practice in such cases is to first develop and debug the more transparent version of the model and only work on optimizations when the simpler formulation has been debugged.</p>
 </section>
@@ -1080,14 +1071,14 @@ <h2 class="anchored" data-anchor-id="item-response-models.section">Item-response
 <section id="data-declaration-with-missingness" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="data-declaration-with-missingness">Data declaration with missingness</h3>
 <p>The data provided for an IRT model may be declared as follows to account for the fact that not every student is required to answer every question.</p>
-<div class="sourceCode" id="cb29"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;                     <span class="co">// number of students</span></span>
-<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;                     <span class="co">// number of questions</span></span>
-<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;                     <span class="co">// number of observations</span></span>
-<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;  <span class="co">// student for observation n</span></span>
-<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;  <span class="co">// question for observation n</span></span>
-<span id="cb29-7"><a href="#cb29-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;   <span class="co">// correctness for observation n</span></span>
-<span id="cb29-8"><a href="#cb29-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;                     <span class="co">// number of students</span></span>
+<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;                     <span class="co">// number of questions</span></span>
+<span id="cb30-4"><a href="#cb30-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; N;                     <span class="co">// number of observations</span></span>
+<span id="cb30-5"><a href="#cb30-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;  <span class="co">// student for observation n</span></span>
+<span id="cb30-6"><a href="#cb30-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=K&gt; kk;  <span class="co">// question for observation n</span></span>
+<span id="cb30-7"><a href="#cb30-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;   <span class="co">// correctness for observation n</span></span>
+<span id="cb30-8"><a href="#cb30-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>This declares a total of <code>N</code> student-question pairs in the data set, where each <code>n</code> in <code>1:N</code> indexes a binary observation <code>y[n]</code> of the correctness of the answer of student <code>jj[n]</code> on question <code>kk[n]</code>.</p>
 <p>The prior hyperparameters will be hard coded in the rest of this section for simplicity, though they could be coded as data in Stan for more flexibility.</p>
 </section>
@@ -1095,21 +1086,21 @@ <h3 class="unnumbered anchored" data-anchor-id="data-declaration-with-missingnes
 <h3 class="unnumbered anchored" data-anchor-id="pl-rasch-model">1PL (Rasch) model</h3>
 <p>The 1PL item-response model, also known as the Rasch model, has one parameter (1P) for questions and uses the logistic link function (L).</p>
 <p>The model parameters are declared as follows.</p>
-<div class="sourceCode" id="cb30"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb30-1"><a href="#cb30-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb30-2"><a href="#cb30-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta;            <span class="co">// mean student ability</span></span>
-<span id="cb30-3"><a href="#cb30-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> alpha;   <span class="co">// ability of student j - mean ability</span></span>
-<span id="cb30-4"><a href="#cb30-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">real</span> beta;    <span class="co">// difficulty of question k</span></span>
-<span id="cb30-5"><a href="#cb30-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> delta;            <span class="co">// mean student ability</span></span>
+<span id="cb31-3"><a href="#cb31-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">real</span> alpha;   <span class="co">// ability of student j - mean ability</span></span>
+<span id="cb31-4"><a href="#cb31-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[K] <span class="dt">real</span> beta;    <span class="co">// difficulty of question k</span></span>
+<span id="cb31-5"><a href="#cb31-5" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The parameter <code>alpha[J]</code> is the ability coefficient for student <code>j</code> and <code>beta[k]</code> is the difficulty coefficient for question <code>k</code>. The non-standard parameterization used here also includes an intercept term <code>delta</code>, which represents the average student’s response to the average question.<a href="#fn4" class="footnote-ref" id="fnref4" role="doc-noteref"><sup>4</sup></a></p>
 <p>The model itself is as follows.</p>
-<div class="sourceCode" id="cb31"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb31-1"><a href="#cb31-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb31-2"><a href="#cb31-2" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();         <span class="co">// informative true prior</span></span>
-<span id="cb31-3"><a href="#cb31-3" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();          <span class="co">// informative true prior</span></span>
-<span id="cb31-4"><a href="#cb31-4" aria-hidden="true" tabindex="-1"></a>  delta ~ normal(<span class="fl">0.75</span>, <span class="dv">1</span>);      <span class="co">// informative true prior</span></span>
-<span id="cb31-5"><a href="#cb31-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb31-6"><a href="#cb31-6" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);</span>
-<span id="cb31-7"><a href="#cb31-7" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb31-8"><a href="#cb31-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();         <span class="co">// informative true prior</span></span>
+<span id="cb32-3"><a href="#cb32-3" aria-hidden="true" tabindex="-1"></a>  beta ~ std_normal();          <span class="co">// informative true prior</span></span>
+<span id="cb32-4"><a href="#cb32-4" aria-hidden="true" tabindex="-1"></a>  delta ~ normal(<span class="fl">0.75</span>, <span class="dv">1</span>);      <span class="co">// informative true prior</span></span>
+<span id="cb32-5"><a href="#cb32-5" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb32-6"><a href="#cb32-6" aria-hidden="true" tabindex="-1"></a>    y[n] ~ bernoulli_logit(alpha[jj[n]] - beta[kk[n]] + delta);</span>
+<span id="cb32-7"><a href="#cb32-7" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb32-8"><a href="#cb32-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>This model uses the logit-parameterized Bernoulli distribution, where <span class="math display">\[
 \texttt{bernoulli}\mathtt{\_}\texttt{logit}\left(y \mid \alpha\right)
 =
@@ -1125,43 +1116,43 @@ <h3 class="unnumbered anchored" data-anchor-id="pl-rasch-model">1PL (Rasch) mode
 <section id="multilevel-2pl-model" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="multilevel-2pl-model">Multilevel 2PL model</h3>
 <p>The simple 1PL model described in the previous section is generalized in this section with the addition of a discrimination parameter to model how noisy a question is and by adding multilevel priors for the question difficulty and discrimination parameters. The model parameters are declared as follows.</p>
-<div class="sourceCode" id="cb32"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb32-1"><a href="#cb32-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb32-2"><a href="#cb32-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;                <span class="co">// mean question difficulty</span></span>
-<span id="cb32-3"><a href="#cb32-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[J] alpha;             <span class="co">// ability for j - mean</span></span>
-<span id="cb32-4"><a href="#cb32-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;              <span class="co">// difficulty for k</span></span>
-<span id="cb32-5"><a href="#cb32-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] gamma;    <span class="co">// discrimination of k</span></span>
-<span id="cb32-6"><a href="#cb32-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;    <span class="co">// scale of difficulties</span></span>
-<span id="cb32-7"><a href="#cb32-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_gamma;   <span class="co">// scale of log discrimination</span></span>
-<span id="cb32-8"><a href="#cb32-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb33-2"><a href="#cb33-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_beta;                <span class="co">// mean question difficulty</span></span>
+<span id="cb33-3"><a href="#cb33-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[J] alpha;             <span class="co">// ability for j - mean</span></span>
+<span id="cb33-4"><a href="#cb33-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;              <span class="co">// difficulty for k</span></span>
+<span id="cb33-5"><a href="#cb33-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] gamma;    <span class="co">// discrimination of k</span></span>
+<span id="cb33-6"><a href="#cb33-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_beta;    <span class="co">// scale of difficulties</span></span>
+<span id="cb33-7"><a href="#cb33-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_gamma;   <span class="co">// scale of log discrimination</span></span>
+<span id="cb33-8"><a href="#cb33-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The parameters should be clearer after the model definition.</p>
-<div class="sourceCode" id="cb33"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb33-1"><a href="#cb33-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb33-2"><a href="#cb33-2" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
-<span id="cb33-3"><a href="#cb33-3" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, sigma_beta);</span>
-<span id="cb33-4"><a href="#cb33-4" aria-hidden="true" tabindex="-1"></a>  gamma ~ lognormal(<span class="dv">0</span>, sigma_gamma);</span>
-<span id="cb33-5"><a href="#cb33-5" aria-hidden="true" tabindex="-1"></a>  mu_beta ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb33-6"><a href="#cb33-6" aria-hidden="true" tabindex="-1"></a>  sigma_beta ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb33-7"><a href="#cb33-7" aria-hidden="true" tabindex="-1"></a>  sigma_gamma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb33-8"><a href="#cb33-8" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));</span>
-<span id="cb33-9"><a href="#cb33-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb34-2"><a href="#cb34-2" aria-hidden="true" tabindex="-1"></a>  alpha ~ std_normal();</span>
+<span id="cb34-3"><a href="#cb34-3" aria-hidden="true" tabindex="-1"></a>  beta ~ normal(<span class="dv">0</span>, sigma_beta);</span>
+<span id="cb34-4"><a href="#cb34-4" aria-hidden="true" tabindex="-1"></a>  gamma ~ lognormal(<span class="dv">0</span>, sigma_gamma);</span>
+<span id="cb34-5"><a href="#cb34-5" aria-hidden="true" tabindex="-1"></a>  mu_beta ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb34-6"><a href="#cb34-6" aria-hidden="true" tabindex="-1"></a>  sigma_beta ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb34-7"><a href="#cb34-7" aria-hidden="true" tabindex="-1"></a>  sigma_gamma ~ cauchy(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb34-8"><a href="#cb34-8" aria-hidden="true" tabindex="-1"></a>  y ~ bernoulli_logit(gamma[kk] .* (alpha[jj] - (beta[kk] + mu_beta)));</span>
+<span id="cb34-9"><a href="#cb34-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The <code>std_normal</code> function is used here, defined by <span class="math display">\[
 \texttt{std}\mathtt{\_}\texttt{normal}(y)
 =
 \textsf{normal}\left(y \mid 0, 1\right).
 \]</span></p>
 <p>The distribution statement is also vectorized using elementwise multiplication; it is equivalent to</p>
-<div class="sourceCode" id="cb34"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb34-1"><a href="#cb34-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb34-2"><a href="#cb34-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(gamma[kk[n]]</span>
-<span id="cb34-3"><a href="#cb34-3" aria-hidden="true" tabindex="-1"></a>                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));</span>
-<span id="cb34-4"><a href="#cb34-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb35-2"><a href="#cb35-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ bernoulli_logit(gamma[kk[n]]</span>
+<span id="cb35-3"><a href="#cb35-3" aria-hidden="true" tabindex="-1"></a>                         * (alpha[jj[n]] - (beta[kk[n]] + mu_beta));</span>
+<span id="cb35-4"><a href="#cb35-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The 2PL model is similar to the 1PL model, with the additional parameter <code>gamma[k]</code> modeling how discriminative question <code>k</code> is. If <code>gamma[k]</code> is greater than 1, responses are more attenuated with less chance of getting a question right at random. The parameter <code>gamma[k]</code> is constrained to be positive, which prohibits there being questions that are easier for students of lesser ability; such questions are not unheard of, but they tend to be eliminated from most testing situations where an IRT model would be applied.</p>
 <p>The model is parameterized here with student abilities <code>alpha</code> being given a standard normal prior. This is to identify both the scale and the location of the parameters, both of which would be unidentified otherwise; see the <a href="../stan-users-guide/problematic-posteriors.html">problematic posteriors chapter</a> for further discussion of identifiability. The difficulty and discrimination parameters <code>beta</code> and <code>gamma</code> then have varying scales given hierarchically in this model. They could also be given weakly informative non-hierarchical priors, such as</p>
-<div class="sourceCode" id="cb35"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb35-1"><a href="#cb35-1" aria-hidden="true" tabindex="-1"></a>beta ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb35-2"><a href="#cb35-2" aria-hidden="true" tabindex="-1"></a>gamma ~ lognormal(<span class="dv">0</span>, <span class="dv">2</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a>beta ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb36-2"><a href="#cb36-2" aria-hidden="true" tabindex="-1"></a>gamma ~ lognormal(<span class="dv">0</span>, <span class="dv">2</span>);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The point is that the <code>alpha</code> determines the scale and location and <code>beta</code> and <code>gamma</code> are allowed to float.</p>
 <p>The <code>beta</code> parameter is here given a non-centered parameterization, with parameter <code>mu_beta</code> serving as the mean <code>beta</code> location. An alternative would’ve been to take:</p>
-<div class="sourceCode" id="cb36"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb36-1"><a href="#cb36-1" aria-hidden="true" tabindex="-1"></a>beta ~ normal(mu_beta, sigma_beta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a>beta ~ normal(mu_beta, sigma_beta);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>and</p>
-<div class="sourceCode" id="cb37"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb37-1"><a href="#cb37-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a>y[n] ~ bernoulli_logit(gamma[kk[n]] * (alpha[jj[n]] - beta[kk[n]]));</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Non-centered parameterizations tend to be more efficient in hierarchical models; see the <a href="../stan-users-guide/efficiency-tuning.html#reparameterization.section">reparameterization section</a> for more information on non-centered reparameterizations.</p>
 <p>The intercept term <code>mu_beta</code> can’t itself be modeled hierarchically, so it is given a weakly informative <span class="math inline">\(\textsf{Cauchy}(0,5)\)</span> prior. Similarly, the scale terms, <code>sigma_beta</code>, and <code>sigma_gamma</code>, are given half-Cauchy priors. As mentioned earlier, the scale and location for <code>alpha</code> are fixed to ensure identifiability. The truncation in the half-Cauchy prior is implicit; explicit truncation is not necessary because the log probability need only be calculated up to a proportion and the scale variables are constrained to <span class="math inline">\((0,\infty)\)</span> by their declarations.</p>
 </section>
@@ -1248,99 +1239,99 @@ <h4 class="unnumbered anchored" data-anchor-id="group-level-predictors-for-prior
 <section id="coding-the-model-in-stan" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="coding-the-model-in-stan">Coding the model in Stan</h4>
 <p>The Stan code for the full hierarchical model with multivariate priors on the group-level coefficients and group-level prior means follows its definition.</p>
-<div class="sourceCode" id="cb38"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb38-1"><a href="#cb38-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb38-2"><a href="#cb38-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;              <span class="co">// num individuals</span></span>
-<span id="cb38-3"><a href="#cb38-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;              <span class="co">// num ind predictors</span></span>
-<span id="cb38-4"><a href="#cb38-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;              <span class="co">// num groups</span></span>
-<span id="cb38-5"><a href="#cb38-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; L;              <span class="co">// num group predictors</span></span>
-<span id="cb38-6"><a href="#cb38-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;  <span class="co">// group for individual</span></span>
-<span id="cb38-7"><a href="#cb38-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;              <span class="co">// individual predictors</span></span>
-<span id="cb38-8"><a href="#cb38-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">row_vector</span>[L] u;    <span class="co">// group predictors</span></span>
-<span id="cb38-9"><a href="#cb38-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;                 <span class="co">// outcomes</span></span>
-<span id="cb38-10"><a href="#cb38-10" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb38-11"><a href="#cb38-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb38-12"><a href="#cb38-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[K] Omega;        <span class="co">// prior correlation</span></span>
-<span id="cb38-13"><a href="#cb38-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] tau;      <span class="co">// prior scale</span></span>
-<span id="cb38-14"><a href="#cb38-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[L, K] gamma;          <span class="co">// group coeffs</span></span>
-<span id="cb38-15"><a href="#cb38-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">vector</span>[K] beta;     <span class="co">// indiv coeffs by group</span></span>
-<span id="cb38-16"><a href="#cb38-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;         <span class="co">// prediction error scale</span></span>
-<span id="cb38-17"><a href="#cb38-17" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb38-18"><a href="#cb38-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb38-19"><a href="#cb38-19" aria-hidden="true" tabindex="-1"></a>  tau ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
-<span id="cb38-20"><a href="#cb38-20" aria-hidden="true" tabindex="-1"></a>  Omega ~ lkj_corr(<span class="dv">2</span>);</span>
-<span id="cb38-21"><a href="#cb38-21" aria-hidden="true" tabindex="-1"></a>  to_vector(gamma) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb38-22"><a href="#cb38-22" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb38-23"><a href="#cb38-23" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[J] <span class="dt">row_vector</span>[K] u_gamma;</span>
-<span id="cb38-24"><a href="#cb38-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
-<span id="cb38-25"><a href="#cb38-25" aria-hidden="true" tabindex="-1"></a>      u_gamma[j] = u[j] * gamma;</span>
-<span id="cb38-26"><a href="#cb38-26" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb38-27"><a href="#cb38-27" aria-hidden="true" tabindex="-1"></a>    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));</span>
-<span id="cb38-28"><a href="#cb38-28" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb38-29"><a href="#cb38-29" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb38-30"><a href="#cb38-30" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(x[n] * beta[jj[n]], sigma);</span>
-<span id="cb38-31"><a href="#cb38-31" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb38-32"><a href="#cb38-32" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb39-2"><a href="#cb39-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;              <span class="co">// num individuals</span></span>
+<span id="cb39-3"><a href="#cb39-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;              <span class="co">// num ind predictors</span></span>
+<span id="cb39-4"><a href="#cb39-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;              <span class="co">// num groups</span></span>
+<span id="cb39-5"><a href="#cb39-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; L;              <span class="co">// num group predictors</span></span>
+<span id="cb39-6"><a href="#cb39-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=J&gt; jj;  <span class="co">// group for individual</span></span>
+<span id="cb39-7"><a href="#cb39-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;              <span class="co">// individual predictors</span></span>
+<span id="cb39-8"><a href="#cb39-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">row_vector</span>[L] u;    <span class="co">// group predictors</span></span>
+<span id="cb39-9"><a href="#cb39-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;                 <span class="co">// outcomes</span></span>
+<span id="cb39-10"><a href="#cb39-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb39-11"><a href="#cb39-11" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb39-12"><a href="#cb39-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[K] Omega;        <span class="co">// prior correlation</span></span>
+<span id="cb39-13"><a href="#cb39-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] tau;      <span class="co">// prior scale</span></span>
+<span id="cb39-14"><a href="#cb39-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[L, K] gamma;          <span class="co">// group coeffs</span></span>
+<span id="cb39-15"><a href="#cb39-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[J] <span class="dt">vector</span>[K] beta;     <span class="co">// indiv coeffs by group</span></span>
+<span id="cb39-16"><a href="#cb39-16" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;         <span class="co">// prediction error scale</span></span>
+<span id="cb39-17"><a href="#cb39-17" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb39-18"><a href="#cb39-18" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb39-19"><a href="#cb39-19" aria-hidden="true" tabindex="-1"></a>  tau ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb39-20"><a href="#cb39-20" aria-hidden="true" tabindex="-1"></a>  Omega ~ lkj_corr(<span class="dv">2</span>);</span>
+<span id="cb39-21"><a href="#cb39-21" aria-hidden="true" tabindex="-1"></a>  to_vector(gamma) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb39-22"><a href="#cb39-22" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb39-23"><a href="#cb39-23" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[J] <span class="dt">row_vector</span>[K] u_gamma;</span>
+<span id="cb39-24"><a href="#cb39-24" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb39-25"><a href="#cb39-25" aria-hidden="true" tabindex="-1"></a>      u_gamma[j] = u[j] * gamma;</span>
+<span id="cb39-26"><a href="#cb39-26" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb39-27"><a href="#cb39-27" aria-hidden="true" tabindex="-1"></a>    beta ~ multi_normal(u_gamma, quad_form_diag(Omega, tau));</span>
+<span id="cb39-28"><a href="#cb39-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb39-29"><a href="#cb39-29" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb39-30"><a href="#cb39-30" aria-hidden="true" tabindex="-1"></a>    y[n] ~ normal(x[n] * beta[jj[n]], sigma);</span>
+<span id="cb39-31"><a href="#cb39-31" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb39-32"><a href="#cb39-32" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The hyperprior covariance matrix is defined implicitly through the quadratic form in the code because the correlation matrix <code>Omega</code> and scale vector <code>tau</code> are more natural to inspect in the output; to output <code>Sigma</code>, define it as a transformed parameter. The function <code>quad_form_diag</code> is defined so that <code>quad_form_diag(Sigma, tau)</code> is equivalent to <code>diag_matrix(tau) * Sigma * diag_matrix(tau)</code>, where <code>diag_matrix(tau)</code> returns the matrix with <code>tau</code> on the diagonal and zeroes off diagonal; the version using <code>quad_form_diag</code> should be faster. For details on these and other matrix arithmetic operators and functions, see the function reference manual.</p>
 </section>
 <section id="optimization-through-vectorization" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="optimization-through-vectorization">Optimization through vectorization</h4>
 <p>The code in the Stan program above can be sped up dramatically by replacing the the distribution statement inside the for loop:</p>
-<div class="sourceCode" id="cb39"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb39-1"><a href="#cb39-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb39-2"><a href="#cb39-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(x[n] * beta[jj[n]], sigma);</span>
-<span id="cb39-3"><a href="#cb39-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb40-2"><a href="#cb40-2" aria-hidden="true" tabindex="-1"></a>  y[n] ~ normal(x[n] * beta[jj[n]], sigma);</span>
+<span id="cb40-3"><a href="#cb40-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>with the vectorized distribution statement:</p>
-<div class="sourceCode" id="cb40"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb40-1"><a href="#cb40-1" aria-hidden="true" tabindex="-1"></a>{</span>
-<span id="cb40-2"><a href="#cb40-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_beta_jj;</span>
-<span id="cb40-3"><a href="#cb40-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb40-4"><a href="#cb40-4" aria-hidden="true" tabindex="-1"></a>    x_beta_jj[n] = x[n] * beta[jj[n]];</span>
-<span id="cb40-5"><a href="#cb40-5" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb40-6"><a href="#cb40-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x_beta_jj, sigma);</span>
-<span id="cb40-7"><a href="#cb40-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb41-2"><a href="#cb41-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x_beta_jj;</span>
+<span id="cb41-3"><a href="#cb41-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb41-4"><a href="#cb41-4" aria-hidden="true" tabindex="-1"></a>    x_beta_jj[n] = x[n] * beta[jj[n]];</span>
+<span id="cb41-5"><a href="#cb41-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb41-6"><a href="#cb41-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x_beta_jj, sigma);</span>
+<span id="cb41-7"><a href="#cb41-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The outer brackets create a local scope in which to define the variable <code>x_beta_jj</code>, which is then filled in a loop and used to define a vectorized distribution statement. The reason this is such a big win is that it allows us to take the log of sigma only once and it greatly reduces the size of the resulting expression graph by packing all of the work into a single distribution function.</p>
 <p>Although it is tempting to redeclare <code>beta</code> and include a revised model block distribution statement,</p>
-<div class="sourceCode" id="cb41"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb41-1"><a href="#cb41-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb41-2"><a href="#cb41-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[J, K] beta;</span>
-<span id="cb41-3"><a href="#cb41-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
-<span id="cb41-4"><a href="#cb41-4" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb41-5"><a href="#cb41-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb41-6"><a href="#cb41-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(rows_dot_product(x, beta[jj]), sigma);</span>
-<span id="cb41-7"><a href="#cb41-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb41-8"><a href="#cb41-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb42-2"><a href="#cb42-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[J, K] beta;</span>
+<span id="cb42-3"><a href="#cb42-3" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb42-4"><a href="#cb42-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb42-5"><a href="#cb42-5" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb42-6"><a href="#cb42-6" aria-hidden="true" tabindex="-1"></a>  y ~ normal(rows_dot_product(x, beta[jj]), sigma);</span>
+<span id="cb42-7"><a href="#cb42-7" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb42-8"><a href="#cb42-8" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>this fails because it breaks the vectorization for <code>beta</code>,<a href="#fn6" class="footnote-ref" id="fnref6" role="doc-noteref"><sup>6</sup></a></p>
-<div class="sourceCode" id="cb42"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb42-1"><a href="#cb42-1" aria-hidden="true" tabindex="-1"></a>beta ~ multi_normal(...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a>beta ~ multi_normal(...);</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>which requires <code>beta</code> to be an array of vectors. Both vectorizations are important, so the best solution is to just use the loop above, because <code>rows_dot_product</code> cannot do much optimization in and of itself because there are no shared computations.</p>
 <p>The code in the Stan program above also builds up an array of vectors for the outcomes and for the multivariate normal, which provides a major speedup by reducing the number of linear systems that need to be solved and differentiated.</p>
-<div class="sourceCode" id="cb43"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb43-1"><a href="#cb43-1" aria-hidden="true" tabindex="-1"></a>{</span>
-<span id="cb43-2"><a href="#cb43-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] Sigma_beta;</span>
-<span id="cb43-3"><a href="#cb43-3" aria-hidden="true" tabindex="-1"></a>  Sigma_beta = quad_form_diag(Omega, tau);</span>
-<span id="cb43-4"><a href="#cb43-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
-<span id="cb43-5"><a href="#cb43-5" aria-hidden="true" tabindex="-1"></a>    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);</span>
-<span id="cb43-6"><a href="#cb43-6" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb43-7"><a href="#cb43-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb44"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb44-1"><a href="#cb44-1" aria-hidden="true" tabindex="-1"></a>{</span>
+<span id="cb44-2"><a href="#cb44-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] Sigma_beta;</span>
+<span id="cb44-3"><a href="#cb44-3" aria-hidden="true" tabindex="-1"></a>  Sigma_beta = quad_form_diag(Omega, tau);</span>
+<span id="cb44-4"><a href="#cb44-4" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (j <span class="cf">in</span> <span class="dv">1</span>:J) {</span>
+<span id="cb44-5"><a href="#cb44-5" aria-hidden="true" tabindex="-1"></a>    beta[j] ~ multi_normal((u[j] * gamma)', Sigma_beta);</span>
+<span id="cb44-6"><a href="#cb44-6" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb44-7"><a href="#cb44-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>In this example, the covariance matrix <code>Sigma_beta</code> is defined as a local variable so as not to have to repeat the quadratic form computation <span class="math inline">\(J\)</span> times. This vectorization can be combined with the Cholesky-factor optimization in the next section.</p>
 </section>
 <section id="optimization-through-cholesky-factorization" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="optimization-through-cholesky-factorization">Optimization through Cholesky factorization</h4>
 <p>The multivariate normal density and LKJ prior on correlation matrices both require their matrix parameters to be factored. Vectorizing, as in the previous section, ensures this is only done once for each density. An even better solution, both in terms of efficiency and numerical stability, is to parameterize the model directly in terms of Cholesky factors of correlation matrices using the multivariate version of the non-centered parameterization. For the model in the previous section, the program fragment to replace the full matrix prior with an equivalent Cholesky factorized prior is as follows.</p>
-<div class="sourceCode" id="cb44"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb44-1"><a href="#cb44-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb44-2"><a href="#cb44-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[L, J] u;              <span class="co">// group predictors transposed</span></span>
-<span id="cb44-3"><a href="#cb44-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb44-4"><a href="#cb44-4" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb44-5"><a href="#cb44-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb44-6"><a href="#cb44-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] z;</span>
-<span id="cb44-7"><a href="#cb44-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
-<span id="cb44-8"><a href="#cb44-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, L] gamma;</span>
-<span id="cb44-9"><a href="#cb44-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb44-10"><a href="#cb44-10" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb44-11"><a href="#cb44-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
-<span id="cb44-12"><a href="#cb44-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
-<span id="cb44-13"><a href="#cb44-13" aria-hidden="true" tabindex="-1"></a>  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;</span>
-<span id="cb44-14"><a href="#cb44-14" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb44-15"><a href="#cb44-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb44-16"><a href="#cb44-16" aria-hidden="true" tabindex="-1"></a>  to_vector(z) ~ std_normal();</span>
-<span id="cb44-17"><a href="#cb44-17" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
-<span id="cb44-18"><a href="#cb44-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
-<span id="cb44-19"><a href="#cb44-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb45"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb45-1"><a href="#cb45-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb45-2"><a href="#cb45-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[L, J] u;              <span class="co">// group predictors transposed</span></span>
+<span id="cb45-3"><a href="#cb45-3" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb45-4"><a href="#cb45-4" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-5"><a href="#cb45-5" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb45-6"><a href="#cb45-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] z;</span>
+<span id="cb45-7"><a href="#cb45-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
+<span id="cb45-8"><a href="#cb45-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, L] gamma;</span>
+<span id="cb45-9"><a href="#cb45-9" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb45-10"><a href="#cb45-10" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-11"><a href="#cb45-11" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb45-12"><a href="#cb45-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
+<span id="cb45-13"><a href="#cb45-13" aria-hidden="true" tabindex="-1"></a>  beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;</span>
+<span id="cb45-14"><a href="#cb45-14" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb45-15"><a href="#cb45-15" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb45-16"><a href="#cb45-16" aria-hidden="true" tabindex="-1"></a>  to_vector(z) ~ std_normal();</span>
+<span id="cb45-17"><a href="#cb45-17" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
+<span id="cb45-18"><a href="#cb45-18" aria-hidden="true" tabindex="-1"></a>  <span class="co">// ...</span></span>
+<span id="cb45-19"><a href="#cb45-19" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The data variable <code>u</code> was originally an array of vectors, which is efficient for access; here it is redeclared as a matrix in order to use it in matrix arithmetic. Moreover, it is transposed, along with <code>gamma</code> and <code>beta</code>, to minimize the number of transposition operations. The new parameter <code>L_Omega</code> is the Cholesky factor of the original correlation matrix <code>Omega</code>, so that</p>
 <pre><code>Omega = L_Omega * L_Omega'</code></pre>
 <p>The prior scale vector <code>tau</code> is unchanged, and furthermore, pre-multiplying the Cholesky factor by the scale produces the Cholesky factor of the final covariance matrix,</p>
@@ -1361,27 +1352,27 @@ <h4 class="unnumbered anchored" data-anchor-id="optimization-through-cholesky-fa
 \end{aligned}
 \]</span> Where we have used the linearity of expectations (line 2 to 3), the definition of <span class="math inline">\(\Omega = \Omega_L \, \Omega_L^\top\)</span>, and the fact that <span class="math inline">\(\mathbb{E}(z \, z^\top) = I\)</span> since <span class="math inline">\(z \sim \mathcal{N}(0, I)\)</span>.</p>
 <p>Omitting the remaining data declarations, which are the same as before with the exception of <code>u</code>, the optimized model is as follows.</p>
-<div class="sourceCode" id="cb48"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb48-1"><a href="#cb48-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb48-2"><a href="#cb48-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] z;</span>
-<span id="cb48-3"><a href="#cb48-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
-<span id="cb48-4"><a href="#cb48-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=pi() / <span class="dv">2</span>&gt;[K] tau_unif;  <span class="co">// prior scale</span></span>
-<span id="cb48-5"><a href="#cb48-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, L] gamma;                        <span class="co">// group coeffs</span></span>
-<span id="cb48-6"><a href="#cb48-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;                       <span class="co">// prediction error scale</span></span>
-<span id="cb48-7"><a href="#cb48-7" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb48-8"><a href="#cb48-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
-<span id="cb48-9"><a href="#cb48-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] tau = <span class="fl">2.5</span> * tan(tau_unif);</span>
-<span id="cb48-10"><a href="#cb48-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;</span>
-<span id="cb48-11"><a href="#cb48-11" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb48-12"><a href="#cb48-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb48-13"><a href="#cb48-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu;</span>
-<span id="cb48-14"><a href="#cb48-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span>(n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb48-15"><a href="#cb48-15" aria-hidden="true" tabindex="-1"></a>    mu[n] = x[n, ] * beta[, jj[n]];</span>
-<span id="cb48-16"><a href="#cb48-16" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb48-17"><a href="#cb48-17" aria-hidden="true" tabindex="-1"></a>  to_vector(z) ~ std_normal();</span>
-<span id="cb48-18"><a href="#cb48-18" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
-<span id="cb48-19"><a href="#cb48-19" aria-hidden="true" tabindex="-1"></a>  to_vector(gamma) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb48-20"><a href="#cb48-20" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
-<span id="cb48-21"><a href="#cb48-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb49"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb49-1"><a href="#cb49-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb49-2"><a href="#cb49-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] z;</span>
+<span id="cb49-3"><a href="#cb49-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
+<span id="cb49-4"><a href="#cb49-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=pi() / <span class="dv">2</span>&gt;[K] tau_unif;  <span class="co">// prior scale</span></span>
+<span id="cb49-5"><a href="#cb49-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, L] gamma;                        <span class="co">// group coeffs</span></span>
+<span id="cb49-6"><a href="#cb49-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;                       <span class="co">// prediction error scale</span></span>
+<span id="cb49-7"><a href="#cb49-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb49-8"><a href="#cb49-8" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb49-9"><a href="#cb49-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] tau = <span class="fl">2.5</span> * tan(tau_unif);</span>
+<span id="cb49-10"><a href="#cb49-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta = gamma * u + diag_pre_multiply(tau, L_Omega) * z;</span>
+<span id="cb49-11"><a href="#cb49-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb49-12"><a href="#cb49-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb49-13"><a href="#cb49-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] mu;</span>
+<span id="cb49-14"><a href="#cb49-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span>(n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb49-15"><a href="#cb49-15" aria-hidden="true" tabindex="-1"></a>    mu[n] = x[n, ] * beta[, jj[n]];</span>
+<span id="cb49-16"><a href="#cb49-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb49-17"><a href="#cb49-17" aria-hidden="true" tabindex="-1"></a>  to_vector(z) ~ std_normal();</span>
+<span id="cb49-18"><a href="#cb49-18" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">2</span>);</span>
+<span id="cb49-19"><a href="#cb49-19" aria-hidden="true" tabindex="-1"></a>  to_vector(gamma) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb49-20"><a href="#cb49-20" aria-hidden="true" tabindex="-1"></a>  y ~ normal(mu, sigma);</span>
+<span id="cb49-21"><a href="#cb49-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>This model also reparameterizes the prior scale <code>tau</code> to avoid potential problems with the heavy tails of the Cauchy distribution. The statement <code>tau_unif ~ uniform(0, pi() / 2)</code> can be omitted from the model block because Stan increments the log posterior for parameters with uniform priors without it.</p>
 </section>
 </section>
@@ -1393,45 +1384,45 @@ <h2 class="anchored" data-anchor-id="prediction-forecasting-and-backcasting">Pre
 <h3 class="unnumbered anchored" data-anchor-id="programming-predictions">Programming predictions</h3>
 <p>As a simple example, the following linear regression provides the same setup for estimating the coefficients <code>beta</code> as in our very first example, using <code>y</code> for the <code>N</code> observations and <code>x</code> for the <code>N</code> predictor vectors. The model parameters and model for observations are exactly the same as before.</p>
 <p>To make predictions, we need to be given the number of predictions, <code>N_new</code>, and their predictor matrix, <code>x_new</code>. The predictions themselves are modeled as a parameter <code>y_new</code>. The model statement for the predictions is exactly the same as for the observations, with the new outcome vector <code>y_new</code> and prediction matrix <code>x_new</code>.</p>
-<div class="sourceCode" id="cb49"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb49-1"><a href="#cb49-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb49-2"><a href="#cb49-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
-<span id="cb49-3"><a href="#cb49-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
-<span id="cb49-4"><a href="#cb49-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
-<span id="cb49-5"><a href="#cb49-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
-<span id="cb49-6"><a href="#cb49-6" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb49-7"><a href="#cb49-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_new;</span>
-<span id="cb49-8"><a href="#cb49-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N_new, K] x_new;</span>
-<span id="cb49-9"><a href="#cb49-9" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb49-10"><a href="#cb49-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb49-11"><a href="#cb49-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
-<span id="cb49-12"><a href="#cb49-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb49-13"><a href="#cb49-13" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb49-14"><a href="#cb49-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_new] y_new;                  <span class="co">// predictions</span></span>
-<span id="cb49-15"><a href="#cb49-15" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb49-16"><a href="#cb49-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb49-17"><a href="#cb49-17" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, sigma);          <span class="co">// observed model</span></span>
-<span id="cb49-18"><a href="#cb49-18" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb49-19"><a href="#cb49-19" aria-hidden="true" tabindex="-1"></a>  y_new ~ normal(x_new * beta, sigma);  <span class="co">// prediction model</span></span>
-<span id="cb49-20"><a href="#cb49-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb50"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb50-1"><a href="#cb50-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb50-2"><a href="#cb50-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb50-3"><a href="#cb50-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb50-4"><a href="#cb50-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N, K] x;</span>
+<span id="cb50-5"><a href="#cb50-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb50-6"><a href="#cb50-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb50-7"><a href="#cb50-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_new;</span>
+<span id="cb50-8"><a href="#cb50-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[N_new, K] x_new;</span>
+<span id="cb50-9"><a href="#cb50-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb50-10"><a href="#cb50-10" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb50-11"><a href="#cb50-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb50-12"><a href="#cb50-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb50-13"><a href="#cb50-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb50-14"><a href="#cb50-14" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_new] y_new;                  <span class="co">// predictions</span></span>
+<span id="cb50-15"><a href="#cb50-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb50-16"><a href="#cb50-16" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb50-17"><a href="#cb50-17" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, sigma);          <span class="co">// observed model</span></span>
+<span id="cb50-18"><a href="#cb50-18" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb50-19"><a href="#cb50-19" aria-hidden="true" tabindex="-1"></a>  y_new ~ normal(x_new * beta, sigma);  <span class="co">// prediction model</span></span>
+<span id="cb50-20"><a href="#cb50-20" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 </section>
 <section id="predictions-as-generated-quantities" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="predictions-as-generated-quantities">Predictions as generated quantities</h3>
 <p>Where possible, the most efficient way to generate predictions is to use the generated quantities block. This provides proper Monte Carlo (not Markov chain Monte Carlo) inference, which can have a much higher effective sample size per iteration.</p>
-<div class="sourceCode" id="cb50"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb50-1"><a href="#cb50-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...data as above...</span></span>
-<span id="cb50-2"><a href="#cb50-2" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb50-3"><a href="#cb50-3" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb50-4"><a href="#cb50-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
-<span id="cb50-5"><a href="#cb50-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb50-6"><a href="#cb50-6" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb50-7"><a href="#cb50-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb50-8"><a href="#cb50-8" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, sigma);</span>
-<span id="cb50-9"><a href="#cb50-9" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb50-10"><a href="#cb50-10" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb50-11"><a href="#cb50-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_new] y_new;</span>
-<span id="cb50-12"><a href="#cb50-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_new) {</span>
-<span id="cb50-13"><a href="#cb50-13" aria-hidden="true" tabindex="-1"></a>    y_new[n] = normal_rng(x_new[n] * beta, sigma);</span>
-<span id="cb50-14"><a href="#cb50-14" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb50-15"><a href="#cb50-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...data as above...</span></span>
+<span id="cb51-2"><a href="#cb51-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb51-3"><a href="#cb51-3" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb51-4"><a href="#cb51-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[K] beta;</span>
+<span id="cb51-5"><a href="#cb51-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb51-6"><a href="#cb51-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb51-7"><a href="#cb51-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb51-8"><a href="#cb51-8" aria-hidden="true" tabindex="-1"></a>  y ~ normal(x * beta, sigma);</span>
+<span id="cb51-9"><a href="#cb51-9" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb51-10"><a href="#cb51-10" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb51-11"><a href="#cb51-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_new] y_new;</span>
+<span id="cb51-12"><a href="#cb51-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_new) {</span>
+<span id="cb51-13"><a href="#cb51-13" aria-hidden="true" tabindex="-1"></a>    y_new[n] = normal_rng(x_new[n] * beta, sigma);</span>
+<span id="cb51-14"><a href="#cb51-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb51-15"><a href="#cb51-15" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Now the data are just as before, but the parameter <code>y_new</code> is now declared as a generated quantity, and the prediction model is removed from the model and replaced by a pseudo-random draw from a normal distribution.</p>
 <section id="overflow-in-generated-quantities" class="level4 unnumbered">
 <h4 class="unnumbered anchored" data-anchor-id="overflow-in-generated-quantities">Overflow in generated quantities</h4>
@@ -1454,49 +1445,49 @@ <h3 class="unnumbered anchored" data-anchor-id="seemingly-unrelated-regressions"
 J)\)</span> matrix), <span class="math inline">\(y_n\)</span> is a <span class="math inline">\(K\)</span>-vector of observations, <span class="math inline">\(\beta\)</span> is a <span class="math inline">\((K
 \times J)\)</span> matrix of regression coefficients (vector <span class="math inline">\(\beta_k\)</span> holds coefficients for outcome <span class="math inline">\(k\)</span>), and <span class="math inline">\(\Sigma\)</span> is covariance matrix governing the error. As usual, the intercept can be rolled into <span class="math inline">\(x\)</span> as a column of ones.</p>
 <p>The basic Stan code is straightforward (though see below for more optimized code for use with LKJ priors on correlation).</p>
-<div class="sourceCode" id="cb51"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb51-1"><a href="#cb51-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb51-2"><a href="#cb51-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
-<span id="cb51-3"><a href="#cb51-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;</span>
-<span id="cb51-4"><a href="#cb51-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
-<span id="cb51-5"><a href="#cb51-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[J] x;</span>
-<span id="cb51-6"><a href="#cb51-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] y;</span>
-<span id="cb51-7"><a href="#cb51-7" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb51-8"><a href="#cb51-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb51-9"><a href="#cb51-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
-<span id="cb51-10"><a href="#cb51-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
-<span id="cb51-11"><a href="#cb51-11" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb51-12"><a href="#cb51-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb51-13"><a href="#cb51-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] mu;</span>
-<span id="cb51-14"><a href="#cb51-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb51-15"><a href="#cb51-15" aria-hidden="true" tabindex="-1"></a>    mu[n] = beta * x[n];</span>
-<span id="cb51-16"><a href="#cb51-16" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb51-17"><a href="#cb51-17" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, Sigma);</span>
-<span id="cb51-18"><a href="#cb51-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb52"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb52-1"><a href="#cb52-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb52-2"><a href="#cb52-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb52-3"><a href="#cb52-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; J;</span>
+<span id="cb52-4"><a href="#cb52-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb52-5"><a href="#cb52-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[J] x;</span>
+<span id="cb52-6"><a href="#cb52-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] y;</span>
+<span id="cb52-7"><a href="#cb52-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb52-8"><a href="#cb52-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb52-9"><a href="#cb52-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
+<span id="cb52-10"><a href="#cb52-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cov_matrix</span>[K] Sigma;</span>
+<span id="cb52-11"><a href="#cb52-11" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb52-12"><a href="#cb52-12" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb52-13"><a href="#cb52-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] mu;</span>
+<span id="cb52-14"><a href="#cb52-14" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb52-15"><a href="#cb52-15" aria-hidden="true" tabindex="-1"></a>    mu[n] = beta * x[n];</span>
+<span id="cb52-16"><a href="#cb52-16" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb52-17"><a href="#cb52-17" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal(mu, Sigma);</span>
+<span id="cb52-18"><a href="#cb52-18" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>For efficiency, the multivariate normal is vectorized by precomputing the array of mean vectors and sharing the same covariance matrix.</p>
 <p>Following the advice in the <a href="#multivariate-hierarchical-priors.section">multivariate hierarchical priors section</a>, we will place a weakly informative normal prior on the regression coefficients, an LKJ prior on the correlations and a half-Cauchy prior on standard deviations. The covariance structure is parameterized in terms of Cholesky factors for efficiency and arithmetic stability.</p>
-<div class="sourceCode" id="cb52"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb52-1"><a href="#cb52-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
-<span id="cb52-2"><a href="#cb52-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb52-3"><a href="#cb52-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
-<span id="cb52-4"><a href="#cb52-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
-<span id="cb52-5"><a href="#cb52-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] L_sigma;</span>
-<span id="cb52-6"><a href="#cb52-6" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb52-7"><a href="#cb52-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb52-8"><a href="#cb52-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] mu;</span>
-<span id="cb52-9"><a href="#cb52-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L_Sigma;</span>
-<span id="cb52-10"><a href="#cb52-10" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb52-11"><a href="#cb52-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb52-12"><a href="#cb52-12" aria-hidden="true" tabindex="-1"></a>    mu[n] = beta * x[n];</span>
-<span id="cb52-13"><a href="#cb52-13" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb52-14"><a href="#cb52-14" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb52-15"><a href="#cb52-15" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb52-16"><a href="#cb52-16" aria-hidden="true" tabindex="-1"></a>  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);</span>
-<span id="cb52-17"><a href="#cb52-17" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb52-18"><a href="#cb52-18" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb52-19"><a href="#cb52-19" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">4</span>);</span>
-<span id="cb52-20"><a href="#cb52-20" aria-hidden="true" tabindex="-1"></a>  L_sigma ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
-<span id="cb52-21"><a href="#cb52-21" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb52-22"><a href="#cb52-22" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal_cholesky(mu, L_Sigma);</span>
-<span id="cb52-23"><a href="#cb52-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb53"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb53-1"><a href="#cb53-1" aria-hidden="true" tabindex="-1"></a><span class="co">// ...</span></span>
+<span id="cb53-2"><a href="#cb53-2" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb53-3"><a href="#cb53-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, J] beta;</span>
+<span id="cb53-4"><a href="#cb53-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[K] L_Omega;</span>
+<span id="cb53-5"><a href="#cb53-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[K] L_sigma;</span>
+<span id="cb53-6"><a href="#cb53-6" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb53-7"><a href="#cb53-7" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb53-8"><a href="#cb53-8" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] mu;</span>
+<span id="cb53-9"><a href="#cb53-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[K, K] L_Sigma;</span>
+<span id="cb53-10"><a href="#cb53-10" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-11"><a href="#cb53-11" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb53-12"><a href="#cb53-12" aria-hidden="true" tabindex="-1"></a>    mu[n] = beta * x[n];</span>
+<span id="cb53-13"><a href="#cb53-13" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-14"><a href="#cb53-14" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb53-15"><a href="#cb53-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-16"><a href="#cb53-16" aria-hidden="true" tabindex="-1"></a>  L_Sigma = diag_pre_multiply(L_sigma, L_Omega);</span>
+<span id="cb53-17"><a href="#cb53-17" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-18"><a href="#cb53-18" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb53-19"><a href="#cb53-19" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">4</span>);</span>
+<span id="cb53-20"><a href="#cb53-20" aria-hidden="true" tabindex="-1"></a>  L_sigma ~ cauchy(<span class="dv">0</span>, <span class="fl">2.5</span>);</span>
+<span id="cb53-21"><a href="#cb53-21" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb53-22"><a href="#cb53-22" aria-hidden="true" tabindex="-1"></a>  y ~ multi_normal_cholesky(mu, L_Sigma);</span>
+<span id="cb53-23"><a href="#cb53-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The Cholesky factor of the covariance matrix is then reconstructed as a local variable and used in the model by scaling the Cholesky factor of the correlation matrices. The regression coefficients get a prior all at once by converting the matrix <code>beta</code> to a vector.</p>
 <p>If required, the full correlation or covariance matrices may be reconstructed from their Cholesky factors in the generated quantities block.</p>
 </section>
@@ -1513,92 +1504,92 @@ <h3 class="unnumbered anchored" data-anchor-id="multivariate-probit-regression">
 <p>Unlike in the seemingly unrelated regressions case, here the covariance matrix <span class="math inline">\(\Sigma\)</span> has unit standard deviations (i.e., it is a correlation matrix). As with ordinary probit and logistic regressions, letting the scale vary causes the model (which is defined only by a cutpoint at 0, not a scale) to be unidentified (see <span class="citation" data-cites="Greene:2011">Greene (<a href="#ref-Greene:2011" role="doc-biblioref">2011</a>)</span>).</p>
 <p>Multivariate probit regression can be coded in Stan using the trick introduced by <span class="citation" data-cites="AlbertChib:1993">Albert and Chib (<a href="#ref-AlbertChib:1993" role="doc-biblioref">1993</a>)</span>, where the underlying continuous value vectors <span class="math inline">\(y_n\)</span> are coded as truncated parameters. The key to coding the model in Stan is declaring the latent vector <span class="math inline">\(z\)</span> in two parts, based on whether the corresponding value of <span class="math inline">\(y\)</span> is 0 or 1. Otherwise, the model is identical to the seemingly unrelated regression model in the previous section.</p>
 <p>First, we introduce a sum function for two-dimensional arrays of integers; this is going to help us calculate how many total 1 values there are in <span class="math inline">\(y\)</span>.</p>
-<div class="sourceCode" id="cb53"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb53-1"><a href="#cb53-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
-<span id="cb53-2"><a href="#cb53-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> sum2d(<span class="dt">array</span>[,] <span class="dt">int</span> a) {</span>
-<span id="cb53-3"><a href="#cb53-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> s = <span class="dv">0</span>;</span>
-<span id="cb53-4"><a href="#cb53-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:size(a)) {</span>
-<span id="cb53-5"><a href="#cb53-5" aria-hidden="true" tabindex="-1"></a>      s += sum(a[i]);</span>
-<span id="cb53-6"><a href="#cb53-6" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb53-7"><a href="#cb53-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> s;</span>
-<span id="cb53-8"><a href="#cb53-8" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb53-9"><a href="#cb53-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb54"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb54-1"><a href="#cb54-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb54-2"><a href="#cb54-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> sum2d(<span class="dt">array</span>[,] <span class="dt">int</span> a) {</span>
+<span id="cb54-3"><a href="#cb54-3" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> s = <span class="dv">0</span>;</span>
+<span id="cb54-4"><a href="#cb54-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:size(a)) {</span>
+<span id="cb54-5"><a href="#cb54-5" aria-hidden="true" tabindex="-1"></a>      s += sum(a[i]);</span>
+<span id="cb54-6"><a href="#cb54-6" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb54-7"><a href="#cb54-7" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> s;</span>
+<span id="cb54-8"><a href="#cb54-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb54-9"><a href="#cb54-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The function is trivial, but it’s not a built-in for Stan and it’s easier to understand the rest of the model if it’s pulled into its own function so as not to create a distraction.</p>
 <p>The data declaration block is much like for the seemingly unrelated regressions, but the observations <code>y</code> are now integers constrained to be 0 or 1.</p>
-<div class="sourceCode" id="cb54"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb54-1"><a href="#cb54-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb54-2"><a href="#cb54-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
-<span id="cb54-3"><a href="#cb54-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
-<span id="cb54-4"><a href="#cb54-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
-<span id="cb54-5"><a href="#cb54-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N, D] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
-<span id="cb54-6"><a href="#cb54-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] x;</span>
-<span id="cb54-7"><a href="#cb54-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb55"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb55-1"><a href="#cb55-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb55-2"><a href="#cb55-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; K;</span>
+<span id="cb55-3"><a href="#cb55-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>&gt; D;</span>
+<span id="cb55-4"><a href="#cb55-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb55-5"><a href="#cb55-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N, D] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; y;</span>
+<span id="cb55-6"><a href="#cb55-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[K] x;</span>
+<span id="cb55-7"><a href="#cb55-7" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>After declaring the data, there is a rather involved transformed data block whose sole purpose is to sort the data array <code>y</code> into positive and negative components, keeping track of indexes so that <code>z</code> can be easily reassembled in the transformed parameters block.</p>
-<div class="sourceCode" id="cb55"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb55-1"><a href="#cb55-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
-<span id="cb55-2"><a href="#cb55-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_pos;</span>
-<span id="cb55-3"><a href="#cb55-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[sum2d(y)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; n_pos;</span>
-<span id="cb55-4"><a href="#cb55-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[size(n_pos)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=D&gt; d_pos;</span>
-<span id="cb55-5"><a href="#cb55-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_neg;</span>
-<span id="cb55-6"><a href="#cb55-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[(N * D) - size(n_pos)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; n_neg;</span>
-<span id="cb55-7"><a href="#cb55-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[size(n_neg)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=D&gt; d_neg;</span>
-<span id="cb55-8"><a href="#cb55-8" aria-hidden="true" tabindex="-1"></a></span>
-<span id="cb55-9"><a href="#cb55-9" aria-hidden="true" tabindex="-1"></a>  N_pos = size(n_pos);</span>
-<span id="cb55-10"><a href="#cb55-10" aria-hidden="true" tabindex="-1"></a>  N_neg = size(n_neg);</span>
-<span id="cb55-11"><a href="#cb55-11" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb55-12"><a href="#cb55-12" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> i;</span>
-<span id="cb55-13"><a href="#cb55-13" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> j;</span>
-<span id="cb55-14"><a href="#cb55-14" aria-hidden="true" tabindex="-1"></a>    i = <span class="dv">1</span>;</span>
-<span id="cb55-15"><a href="#cb55-15" aria-hidden="true" tabindex="-1"></a>    j = <span class="dv">1</span>;</span>
-<span id="cb55-16"><a href="#cb55-16" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb55-17"><a href="#cb55-17" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
-<span id="cb55-18"><a href="#cb55-18" aria-hidden="true" tabindex="-1"></a>        <span class="cf">if</span> (y[n, d] == <span class="dv">1</span>) {</span>
-<span id="cb55-19"><a href="#cb55-19" aria-hidden="true" tabindex="-1"></a>          n_pos[i] = n;</span>
-<span id="cb55-20"><a href="#cb55-20" aria-hidden="true" tabindex="-1"></a>          d_pos[i] = d;</span>
-<span id="cb55-21"><a href="#cb55-21" aria-hidden="true" tabindex="-1"></a>          i += <span class="dv">1</span>;</span>
-<span id="cb55-22"><a href="#cb55-22" aria-hidden="true" tabindex="-1"></a>        } <span class="cf">else</span> {</span>
-<span id="cb55-23"><a href="#cb55-23" aria-hidden="true" tabindex="-1"></a>          n_neg[j] = n;</span>
-<span id="cb55-24"><a href="#cb55-24" aria-hidden="true" tabindex="-1"></a>          d_neg[j] = d;</span>
-<span id="cb55-25"><a href="#cb55-25" aria-hidden="true" tabindex="-1"></a>          j += <span class="dv">1</span>;</span>
-<span id="cb55-26"><a href="#cb55-26" aria-hidden="true" tabindex="-1"></a>        }</span>
-<span id="cb55-27"><a href="#cb55-27" aria-hidden="true" tabindex="-1"></a>      }</span>
-<span id="cb55-28"><a href="#cb55-28" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb55-29"><a href="#cb55-29" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb55-30"><a href="#cb55-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb56"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb56-1"><a href="#cb56-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
+<span id="cb56-2"><a href="#cb56-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_pos;</span>
+<span id="cb56-3"><a href="#cb56-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[sum2d(y)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; n_pos;</span>
+<span id="cb56-4"><a href="#cb56-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[size(n_pos)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=D&gt; d_pos;</span>
+<span id="cb56-5"><a href="#cb56-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_neg;</span>
+<span id="cb56-6"><a href="#cb56-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[(N * D) - size(n_pos)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=N&gt; n_neg;</span>
+<span id="cb56-7"><a href="#cb56-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[size(n_neg)] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">1</span>, <span class="kw">upper</span>=D&gt; d_neg;</span>
+<span id="cb56-8"><a href="#cb56-8" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb56-9"><a href="#cb56-9" aria-hidden="true" tabindex="-1"></a>  N_pos = size(n_pos);</span>
+<span id="cb56-10"><a href="#cb56-10" aria-hidden="true" tabindex="-1"></a>  N_neg = size(n_neg);</span>
+<span id="cb56-11"><a href="#cb56-11" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb56-12"><a href="#cb56-12" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> i;</span>
+<span id="cb56-13"><a href="#cb56-13" aria-hidden="true" tabindex="-1"></a>    <span class="dt">int</span> j;</span>
+<span id="cb56-14"><a href="#cb56-14" aria-hidden="true" tabindex="-1"></a>    i = <span class="dv">1</span>;</span>
+<span id="cb56-15"><a href="#cb56-15" aria-hidden="true" tabindex="-1"></a>    j = <span class="dv">1</span>;</span>
+<span id="cb56-16"><a href="#cb56-16" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb56-17"><a href="#cb56-17" aria-hidden="true" tabindex="-1"></a>      <span class="cf">for</span> (d <span class="cf">in</span> <span class="dv">1</span>:D) {</span>
+<span id="cb56-18"><a href="#cb56-18" aria-hidden="true" tabindex="-1"></a>        <span class="cf">if</span> (y[n, d] == <span class="dv">1</span>) {</span>
+<span id="cb56-19"><a href="#cb56-19" aria-hidden="true" tabindex="-1"></a>          n_pos[i] = n;</span>
+<span id="cb56-20"><a href="#cb56-20" aria-hidden="true" tabindex="-1"></a>          d_pos[i] = d;</span>
+<span id="cb56-21"><a href="#cb56-21" aria-hidden="true" tabindex="-1"></a>          i += <span class="dv">1</span>;</span>
+<span id="cb56-22"><a href="#cb56-22" aria-hidden="true" tabindex="-1"></a>        } <span class="cf">else</span> {</span>
+<span id="cb56-23"><a href="#cb56-23" aria-hidden="true" tabindex="-1"></a>          n_neg[j] = n;</span>
+<span id="cb56-24"><a href="#cb56-24" aria-hidden="true" tabindex="-1"></a>          d_neg[j] = d;</span>
+<span id="cb56-25"><a href="#cb56-25" aria-hidden="true" tabindex="-1"></a>          j += <span class="dv">1</span>;</span>
+<span id="cb56-26"><a href="#cb56-26" aria-hidden="true" tabindex="-1"></a>        }</span>
+<span id="cb56-27"><a href="#cb56-27" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb56-28"><a href="#cb56-28" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb56-29"><a href="#cb56-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb56-30"><a href="#cb56-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The variables <code>N_pos</code> and <code>N_neg</code> are set to the number of true (1) and number of false (0) observations in <code>y</code>. The loop then fills in the sequence of indexes for the positive and negative values in four arrays.</p>
 <p>The parameters are declared as follows.</p>
-<div class="sourceCode" id="cb56"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb56-1"><a href="#cb56-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb56-2"><a href="#cb56-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, K] beta;</span>
-<span id="cb56-3"><a href="#cb56-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
-<span id="cb56-4"><a href="#cb56-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N_pos] z_pos;</span>
-<span id="cb56-5"><a href="#cb56-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt;[N_neg] z_neg;</span>
-<span id="cb56-6"><a href="#cb56-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb57"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb57-1"><a href="#cb57-1" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb57-2"><a href="#cb57-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">matrix</span>[D, K] beta;</span>
+<span id="cb57-3"><a href="#cb57-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">cholesky_factor_corr</span>[D] L_Omega;</span>
+<span id="cb57-4"><a href="#cb57-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt;[N_pos] z_pos;</span>
+<span id="cb57-5"><a href="#cb57-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>&lt;<span class="kw">upper</span>=<span class="dv">0</span>&gt;[N_neg] z_neg;</span>
+<span id="cb57-6"><a href="#cb57-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>These include the regression coefficients <code>beta</code> and the Cholesky factor of the correlation matrix, <code>L_Omega</code>. This time there is no scaling because the covariance matrix has unit scale (i.e., it is a correlation matrix; see above).</p>
 <p>The critical part of the parameter declaration is that the latent real value <span class="math inline">\(z\)</span> is broken into positive-constrained and negative-constrained components, whose size was conveniently calculated in the transformed data block. The transformed data block’s real work was to allow the transformed parameter block to reconstruct <span class="math inline">\(z\)</span>.</p>
-<div class="sourceCode" id="cb57"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb57-1"><a href="#cb57-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
-<span id="cb57-2"><a href="#cb57-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] z;</span>
-<span id="cb57-3"><a href="#cb57-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_pos) {</span>
-<span id="cb57-4"><a href="#cb57-4" aria-hidden="true" tabindex="-1"></a>    z[n_pos[n], d_pos[n]] = z_pos[n];</span>
-<span id="cb57-5"><a href="#cb57-5" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb57-6"><a href="#cb57-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_neg) {</span>
-<span id="cb57-7"><a href="#cb57-7" aria-hidden="true" tabindex="-1"></a>    z[n_neg[n], d_neg[n]] = z_neg[n];</span>
-<span id="cb57-8"><a href="#cb57-8" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb57-9"><a href="#cb57-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb58"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb58-1"><a href="#cb58-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span> {</span>
+<span id="cb58-2"><a href="#cb58-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">vector</span>[D] z;</span>
+<span id="cb58-3"><a href="#cb58-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_pos) {</span>
+<span id="cb58-4"><a href="#cb58-4" aria-hidden="true" tabindex="-1"></a>    z[n_pos[n], d_pos[n]] = z_pos[n];</span>
+<span id="cb58-5"><a href="#cb58-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb58-6"><a href="#cb58-6" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_neg) {</span>
+<span id="cb58-7"><a href="#cb58-7" aria-hidden="true" tabindex="-1"></a>    z[n_neg[n], d_neg[n]] = z_neg[n];</span>
+<span id="cb58-8"><a href="#cb58-8" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb58-9"><a href="#cb58-9" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>At this point, the model is simple, pretty much recreating the seemingly unrelated regression.</p>
-<div class="sourceCode" id="cb58"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb58-1"><a href="#cb58-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb58-2"><a href="#cb58-2" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">4</span>);</span>
-<span id="cb58-3"><a href="#cb58-3" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
-<span id="cb58-4"><a href="#cb58-4" aria-hidden="true" tabindex="-1"></a>  {</span>
-<span id="cb58-5"><a href="#cb58-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[N] <span class="dt">vector</span>[D] beta_x;</span>
-<span id="cb58-6"><a href="#cb58-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb58-7"><a href="#cb58-7" aria-hidden="true" tabindex="-1"></a>      beta_x[n] = beta * x[n];</span>
-<span id="cb58-8"><a href="#cb58-8" aria-hidden="true" tabindex="-1"></a>    }</span>
-<span id="cb58-9"><a href="#cb58-9" aria-hidden="true" tabindex="-1"></a>    z ~ multi_normal_cholesky(beta_x, L_Omega);</span>
-<span id="cb58-10"><a href="#cb58-10" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb58-11"><a href="#cb58-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb59"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb59-1"><a href="#cb59-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb59-2"><a href="#cb59-2" aria-hidden="true" tabindex="-1"></a>  L_Omega ~ lkj_corr_cholesky(<span class="dv">4</span>);</span>
+<span id="cb59-3"><a href="#cb59-3" aria-hidden="true" tabindex="-1"></a>  to_vector(beta) ~ normal(<span class="dv">0</span>, <span class="dv">5</span>);</span>
+<span id="cb59-4"><a href="#cb59-4" aria-hidden="true" tabindex="-1"></a>  {</span>
+<span id="cb59-5"><a href="#cb59-5" aria-hidden="true" tabindex="-1"></a>    <span class="dt">array</span>[N] <span class="dt">vector</span>[D] beta_x;</span>
+<span id="cb59-6"><a href="#cb59-6" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb59-7"><a href="#cb59-7" aria-hidden="true" tabindex="-1"></a>      beta_x[n] = beta * x[n];</span>
+<span id="cb59-8"><a href="#cb59-8" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb59-9"><a href="#cb59-9" aria-hidden="true" tabindex="-1"></a>    z ~ multi_normal_cholesky(beta_x, L_Omega);</span>
+<span id="cb59-10"><a href="#cb59-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb59-11"><a href="#cb59-11" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>This simple form of model is made possible by the Albert and Chib-style constraints on <code>z</code>.</p>
 <p>Finally, the correlation matrix itself can be put back together in the generated quantities block if desired.</p>
-<div class="sourceCode" id="cb59"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb59-1"><a href="#cb59-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb59-2"><a href="#cb59-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[D] Omega;</span>
-<span id="cb59-3"><a href="#cb59-3" aria-hidden="true" tabindex="-1"></a>  Omega = multiply_lower_tri_self_transpose(L_Omega);</span>
-<span id="cb59-4"><a href="#cb59-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb60"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb60-1"><a href="#cb60-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb60-2"><a href="#cb60-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">corr_matrix</span>[D] Omega;</span>
+<span id="cb60-3"><a href="#cb60-3" aria-hidden="true" tabindex="-1"></a>  Omega = multiply_lower_tri_self_transpose(L_Omega);</span>
+<span id="cb60-4"><a href="#cb60-4" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>The same could be done for the seemingly unrelated regressions in the previous section.</p>
 </section>
 </section>
@@ -1622,39 +1613,39 @@ <h3 class="unnumbered anchored" data-anchor-id="prediction">Prediction</h3>
   \,\textrm{d}(\alpha,\beta,\sigma).
 \]</span></p>
 <p>To code the posterior predictive inference in Stan, a standard linear regression is combined with a random number in the generated quantities block.</p>
-<div class="sourceCode" id="cb60"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb60-1"><a href="#cb60-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
-<span id="cb60-2"><a href="#cb60-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
-<span id="cb60-3"><a href="#cb60-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
-<span id="cb60-4"><a href="#cb60-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
-<span id="cb60-5"><a href="#cb60-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
-<span id="cb60-6"><a href="#cb60-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
-<span id="cb60-7"><a href="#cb60-7" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb60-8"><a href="#cb60-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
-<span id="cb60-9"><a href="#cb60-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
-<span id="cb60-10"><a href="#cb60-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
-<span id="cb60-11"><a href="#cb60-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
-<span id="cb60-12"><a href="#cb60-12" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb60-13"><a href="#cb60-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
-<span id="cb60-14"><a href="#cb60-14" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
-<span id="cb60-15"><a href="#cb60-15" aria-hidden="true" tabindex="-1"></a>}</span>
-<span id="cb60-16"><a href="#cb60-16" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb60-17"><a href="#cb60-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] y_tilde;</span>
-<span id="cb60-18"><a href="#cb60-18" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_tilde) {</span>
-<span id="cb60-19"><a href="#cb60-19" aria-hidden="true" tabindex="-1"></a>    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);</span>
-<span id="cb60-20"><a href="#cb60-20" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb60-21"><a href="#cb60-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb61"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb61-1"><a href="#cb61-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb61-2"><a href="#cb61-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N;</span>
+<span id="cb61-3"><a href="#cb61-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y;</span>
+<span id="cb61-4"><a href="#cb61-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] x;</span>
+<span id="cb61-5"><a href="#cb61-5" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N_tilde;</span>
+<span id="cb61-6"><a href="#cb61-6" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] x_tilde;</span>
+<span id="cb61-7"><a href="#cb61-7" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb61-8"><a href="#cb61-8" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb61-9"><a href="#cb61-9" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> alpha;</span>
+<span id="cb61-10"><a href="#cb61-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> beta;</span>
+<span id="cb61-11"><a href="#cb61-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma;</span>
+<span id="cb61-12"><a href="#cb61-12" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb61-13"><a href="#cb61-13" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb61-14"><a href="#cb61-14" aria-hidden="true" tabindex="-1"></a>  y ~ normal(alpha + beta * x, sigma);</span>
+<span id="cb61-15"><a href="#cb61-15" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb61-16"><a href="#cb61-16" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb61-17"><a href="#cb61-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N_tilde] y_tilde;</span>
+<span id="cb61-18"><a href="#cb61-18" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N_tilde) {</span>
+<span id="cb61-19"><a href="#cb61-19" aria-hidden="true" tabindex="-1"></a>    y_tilde[n] = normal_rng(alpha + beta * x_tilde[n], sigma);</span>
+<span id="cb61-20"><a href="#cb61-20" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb61-21"><a href="#cb61-21" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p>Given observed predictors <span class="math inline">\(x\)</span> and outcomes <span class="math inline">\(y\)</span>, <code>y_tilde</code> will be drawn according to <span class="math inline">\(p\left(\tilde{y} \mid \tilde{x}, y, x\right)\)</span>. This means that, for example, the posterior mean for <code>y_tilde</code> is the estimate of the outcome that minimizes expected square error (conditioned on the data and model).</p>
 </section>
 <section id="posterior-predictive-checks" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="posterior-predictive-checks">Posterior predictive checks</h3>
 <p>A good way to investigate the fit of a model to the data, a critical step in Bayesian data analysis, is to generate simulated data according to the parameters of the model. This is carried out with exactly the same procedure as before, only the observed data predictors <span class="math inline">\(x\)</span> are used in place of new predictors <span class="math inline">\(\tilde{x}\)</span> for unobserved outcomes. If the model fits the data well, the predictions for <span class="math inline">\(\tilde{y}\)</span> based on <span class="math inline">\(x\)</span> should match the observed data <span class="math inline">\(y\)</span>.</p>
 <p>To code posterior predictive checks in Stan requires only a slight modification of the prediction code to use <span class="math inline">\(x\)</span> and <span class="math inline">\(N\)</span> in place of <span class="math inline">\(\tilde{x}\)</span> and <span class="math inline">\(\tilde{N}\)</span>,</p>
-<div class="sourceCode" id="cb61"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb61-1"><a href="#cb61-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
-<span id="cb61-2"><a href="#cb61-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y_tilde;</span>
-<span id="cb61-3"><a href="#cb61-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
-<span id="cb61-4"><a href="#cb61-4" aria-hidden="true" tabindex="-1"></a>    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);</span>
-<span id="cb61-5"><a href="#cb61-5" aria-hidden="true" tabindex="-1"></a>  }</span>
-<span id="cb61-6"><a href="#cb61-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="sourceCode" id="cb62"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb62-1"><a href="#cb62-1" aria-hidden="true" tabindex="-1"></a><span class="kw">generated quantities</span> {</span>
+<span id="cb62-2"><a href="#cb62-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">vector</span>[N] y_tilde;</span>
+<span id="cb62-3"><a href="#cb62-3" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (n <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb62-4"><a href="#cb62-4" aria-hidden="true" tabindex="-1"></a>    y_tilde[n] = normal_rng(alpha + beta * x[n], sigma);</span>
+<span id="cb62-5"><a href="#cb62-5" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb62-6"><a href="#cb62-6" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
 <p><span class="citation" data-cites="GelmanEtAl:2013">Gelman et al. (<a href="#ref-GelmanEtAl:2013" role="doc-biblioref">2013</a>)</span> recommend choosing several posterior draws <span class="math inline">\(\tilde{y}^{(1)}, \dotsc, \tilde{y}^{(M)}\)</span> and plotting each of them alongside the data <span class="math inline">\(y\)</span> that was actually observed. If the model fits well, the simulated <span class="math inline">\(\tilde{y}\)</span> will look like the actual data <span class="math inline">\(y\)</span>.</p>
 
 
@@ -2280,8 +2271,35 @@ <h3 class="unnumbered anchored" data-anchor-id="posterior-predictive-checks">Pos
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/regression.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/regression.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/reparameterization.html b/docs/stan-users-guide/reparameterization.html
index ad073e45d..f88283870 100644
--- a/docs/stan-users-guide/reparameterization.html
+++ b/docs/stan-users-guide/reparameterization.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1583,8 +1595,35 @@ <h3 class="unnumbered anchored" data-anchor-id="varying-upper-and-lower-bounds">
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/reparameterization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/reparameterization.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/simulation-based-calibration.html b/docs/stan-users-guide/simulation-based-calibration.html
index 2f0477d4d..b3a86f5bf 100644
--- a/docs/stan-users-guide/simulation-based-calibration.html
+++ b/docs/stan-users-guide/simulation-based-calibration.html
@@ -7,7 +7,7 @@
 <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
 
 
-<title>Simulation-Based Calibration</title>
+<title>Simulation-Based Calibration Checking</title>
 <style>
 code{white-space: pre-wrap;}
 span.smallcaps{font-variant: small-caps;}
@@ -157,7 +157,7 @@
 </script>
 
 <link rel="stylesheet" href="../quarto-config/quarto_styles.css">
-<meta property="og:title" content="Simulation-Based Calibration">
+<meta property="og:title" content="Simulation-Based Calibration Checking">
 <meta property="og:description" content="">
 <meta property="og:image" content="./img/sbc-normal-normal.png">
 <meta property="og:site_name" content="Stan Docs">
@@ -289,7 +289,7 @@
       <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
         <i class="bi bi-layout-text-sidebar-reverse"></i>
       </button>
-        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration</a></li></ol></nav>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration Checking</a></li></ol></nav>
         <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
         </a>
     </div>
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link active">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -593,22 +605,22 @@
     <h2 id="toc-title">On this page</h2>
    
   <ul>
-  <li><a href="#simulation-based-calibration" id="toc-simulation-based-calibration" class="nav-link active" data-scroll-target="#simulation-based-calibration">Simulation-Based Calibration</a>
+  <li><a href="#simulation-based-calibration-checking" id="toc-simulation-based-calibration-checking" class="nav-link active" data-scroll-target="#simulation-based-calibration-checking">Simulation-Based Calibration Checking</a>
   <ul class="collapse">
   <li><a href="#bayes-is-calibrated-by-construction" id="toc-bayes-is-calibrated-by-construction" class="nav-link" data-scroll-target="#bayes-is-calibrated-by-construction">Bayes is calibrated by construction</a></li>
-  <li><a href="#simulation-based-calibration-1" id="toc-simulation-based-calibration-1" class="nav-link" data-scroll-target="#simulation-based-calibration-1">Simulation-based calibration</a></li>
+  <li><a href="#simulation-based-calibration-checking-1" id="toc-simulation-based-calibration-checking-1" class="nav-link" data-scroll-target="#simulation-based-calibration-checking-1">Simulation-based calibration checking</a></li>
   <li><a href="#sbc-in-stan" id="toc-sbc-in-stan" class="nav-link" data-scroll-target="#sbc-in-stan">SBC in Stan</a>
   <ul class="collapse">
   <li><a href="#example-model" id="toc-example-model" class="nav-link" data-scroll-target="#example-model">Example model</a></li>
-  <li><a href="#testing-a-stan-program-with-simulation-based-calibration" id="toc-testing-a-stan-program-with-simulation-based-calibration" class="nav-link" data-scroll-target="#testing-a-stan-program-with-simulation-based-calibration">Testing a Stan program with simulation-based calibration</a></li>
-  <li><a href="#pseudocode-for-simulation-based-calibration" id="toc-pseudocode-for-simulation-based-calibration" class="nav-link" data-scroll-target="#pseudocode-for-simulation-based-calibration">Pseudocode for simulation-based calibration</a></li>
+  <li><a href="#testing-a-stan-program-with-simulation-based-calibration-checking" id="toc-testing-a-stan-program-with-simulation-based-calibration-checking" class="nav-link" data-scroll-target="#testing-a-stan-program-with-simulation-based-calibration-checking">Testing a Stan program with simulation-based calibration checking</a></li>
+  <li><a href="#pseudocode-for-simulation-based-calibration-checking" id="toc-pseudocode-for-simulation-based-calibration-checking" class="nav-link" data-scroll-target="#pseudocode-for-simulation-based-calibration-checking">Pseudocode for simulation-based calibration checking</a></li>
   <li><a href="#the-importance-of-thinning" id="toc-the-importance-of-thinning" class="nav-link" data-scroll-target="#the-importance-of-thinning">The importance of thinning</a></li>
   </ul></li>
   <li><a href="#testing-uniformity" id="toc-testing-uniformity" class="nav-link" data-scroll-target="#testing-uniformity">Testing uniformity</a>
   <ul class="collapse">
   <li><a href="#indexing-to-simplify-arithmetic" id="toc-indexing-to-simplify-arithmetic" class="nav-link" data-scroll-target="#indexing-to-simplify-arithmetic">Indexing to simplify arithmetic</a></li>
   </ul></li>
-  <li><a href="#examples-of-simulation-based-calibration" id="toc-examples-of-simulation-based-calibration" class="nav-link" data-scroll-target="#examples-of-simulation-based-calibration">Examples of simulation-based calibration</a>
+  <li><a href="#examples-of-simulation-based-calibration-checking" id="toc-examples-of-simulation-based-calibration-checking" class="nav-link" data-scroll-target="#examples-of-simulation-based-calibration-checking">Examples of simulation-based calibration checking</a>
   <ul class="collapse">
   <li><a href="#when-things-go-right" id="toc-when-things-go-right" class="nav-link" data-scroll-target="#when-things-go-right">When things go right</a></li>
   <li><a href="#when-things-go-wrong" id="toc-when-things-go-wrong" class="nav-link" data-scroll-target="#when-things-go-wrong">When things go wrong</a></li>
@@ -619,14 +631,14 @@ <h2 id="toc-title">On this page</h2>
 <div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/simulation-based-calibration.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></nav>
     </div>
 <!-- main -->
-<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration</a></li></ol></nav></header>
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/posterior-prediction.html">Posterior Inference &amp; Model Checking</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/simulation-based-calibration.html">Simulation-Based Calibration Checking</a></li></ol></nav></header>
 
 
 
 
-<section id="simulation-based-calibration" class="level1">
-<h1>Simulation-Based Calibration</h1>
-<p>A Bayesian posterior is calibrated if the posterior intervals have appropriate coverage. For example, 80% intervals are expected to contain the true parameter 80% of the time. If data is generated according to a model, Bayesian posterior inference with respect to that model is calibrated by construction. Simulation-based calibration (SBC) exploits this property of Bayesian inference to assess the soundness of a posterior sampler. Roughly, the way it works is by simulating parameters according to the prior, then simulating data conditioned on the simulated parameters, then testing posterior calibration of the inference algorithm over independently simulated data sets. This chapter follows <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span>, which improves on the original approach developed by <span class="citation" data-cites="CookGelmanRubin:2006">Cook, Gelman, and Rubin (<a href="#ref-CookGelmanRubin:2006" role="doc-biblioref">2006</a>)</span>.</p>
+<section id="simulation-based-calibration-checking" class="level1">
+<h1>Simulation-Based Calibration Checking</h1>
+<p>A Bayesian posterior is calibrated if the posterior intervals have appropriate coverage. For example, 80% intervals are expected to contain the true parameter 80% of the time. If data is generated according to a model, Bayesian posterior inference with respect to that model is calibrated by construction. Simulation-based calibration checking (SBC) exploits this property of Bayesian inference to assess the soundness of a posterior sampler. Roughly, the way it works is by simulating parameters according to the prior, then simulating data conditioned on the simulated parameters, then testing posterior calibration of the inference algorithm over independently simulated data sets. This chapter follows <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span>, which improves on the original approach developed by <span class="citation" data-cites="CookGelmanRubin:2006">Cook, Gelman, and Rubin (<a href="#ref-CookGelmanRubin:2006" role="doc-biblioref">2006</a>)</span>. See also <span class="citation" data-cites="ModrakEtAl:2023">Modrák et al. (<a href="#ref-ModrakEtAl:2023" role="doc-biblioref">2023</a>)</span> for further improvements.</p>
 <section id="bayes-is-calibrated-by-construction" class="level2">
 <h2 class="anchored" data-anchor-id="bayes-is-calibrated-by-construction">Bayes is calibrated by construction</h2>
 <p>Suppose a Bayesian model is given in the form of a prior density <span class="math inline">\(p(\theta)\)</span> and sampling density <span class="math inline">\(p(y \mid \theta).\)</span> Now consider a process that first simulates parameters from the prior, <span class="math display">\[
@@ -645,11 +657,11 @@ <h2 class="anchored" data-anchor-id="bayes-is-calibrated-by-construction">Bayes
 \]</span> Because <span class="math inline">\(\theta^{\textrm{sim}}\)</span> is also distributed as a draw from the posterior, the rank statistics of <span class="math inline">\(\theta^{\textrm{sim}}\)</span> with respect to <span class="math inline">\(\theta^{(1)}, \ldots \theta^{(M)}\)</span> should be uniform.</p>
 <p>This is one way to define calibration, because it follows that posterior intervals will have appropriate coverage <span class="citation" data-cites="Dawid:1982 GneitingEtAl:2007">(<a href="#ref-Dawid:1982" role="doc-biblioref">Dawid 1982</a>; <a href="#ref-GneitingEtAl:2007" role="doc-biblioref">Gneiting, Balabdaoui, and Raftery 2007</a>)</span>. If the rank of <span class="math inline">\(\theta^{\textrm{sim}}\)</span> is uniform among the draws <span class="math inline">\(\theta^{(1)}, \ldots, \theta^{(M)},\)</span> then for any 90% interval selected, the probability the true value <span class="math inline">\(\theta^{\textrm{sim}}\)</span> falls in it will also be 90%. The same goes for any other posterior interval.</p>
 </section>
-<section id="simulation-based-calibration-1" class="level2">
-<h2 class="anchored" data-anchor-id="simulation-based-calibration-1">Simulation-based calibration</h2>
+<section id="simulation-based-calibration-checking-1" class="level2">
+<h2 class="anchored" data-anchor-id="simulation-based-calibration-checking-1">Simulation-based calibration checking</h2>
 <p>Suppose the Bayesian model to test has joint density <span class="math display">\[
 p(y, \theta) = p(y \mid \theta) \cdot p(\theta),
-\]</span> with data <span class="math inline">\(y\)</span> and parameters <span class="math inline">\(\theta\)</span> (both are typically multivariate). Simulation-based calibration works by generating <span class="math inline">\(N\)</span> simulated parameter and data pairs according to the joint density, <span class="math display">\[
+\]</span> with data <span class="math inline">\(y\)</span> and parameters <span class="math inline">\(\theta\)</span> (both are typically multivariate). Simulation-based calibration checking works by generating <span class="math inline">\(N\)</span> simulated parameter and data pairs according to the joint density, <span class="math display">\[
 (y^{\textrm{sim}(1)}, \theta^{\textrm{sim}(1)}),
 \ldots, (y^{\textrm{sim}(N)}, \theta^{\textrm{sim}(N)}),
 \sim p(y, \theta).
@@ -666,16 +678,16 @@ <h2 class="anchored" data-anchor-id="simulation-based-calibration-1">Simulation-
 \sum_{m = 1}^M
   \textrm{I}[\theta_k^{(n,m)} &lt; \theta_k^{\textrm{sim}(n)}].
 \end{eqnarray*}\]</span> That is, the rank is the number of posterior draws <span class="math inline">\(\theta^{(n,m)}_k\)</span> that are less than the simulated draw <span class="math inline">\(\theta^{\textrm{sim}(n)}_k.\)</span></p>
-<p>If the algorithm generates posterior draws according to the posterior, the ranks should be uniformly distributed from <span class="math inline">\(0\)</span> to <span class="math inline">\(M\)</span>, so that the ranks plus one are uniformly distributed from <span class="math inline">\(1\)</span> to <span class="math inline">\(M + 1\)</span>, <span class="math display">\[
+<p>If the algorithm generates posterior draws according to the posterior, the ranks should have uniform discrete distribution from <span class="math inline">\(0\)</span> to <span class="math inline">\(M\)</span>, so that the ranks plus one are uniformly distributed from <span class="math inline">\(1\)</span> to <span class="math inline">\(M + 1\)</span>, <span class="math display">\[
 r_{n, k} + 1
 \sim
 \textrm{categorical}\! \left(\frac{1}{M + 1}, \ldots, \frac{1}{M + 1}\right).
-\]</span> Simulation-based calibration uses this expected behavior to test the calibration of each parameter of a model on simulated data. <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span> suggest plotting binned counts of <span class="math inline">\(r_{1:N,
-k}\)</span> for different parameters <span class="math inline">\(k\)</span>; <span class="citation" data-cites="CookGelmanRubin:2006">Cook, Gelman, and Rubin (<a href="#ref-CookGelmanRubin:2006" role="doc-biblioref">2006</a>)</span> automate the process with a hypothesis test for uniformity.</p>
+\]</span> Simulation-based calibration checking uses this expected behavior to test the calibration of each parameter of a model on simulated data. <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span> suggest plotting binned counts of <span class="math inline">\(r_{1:N,
+k}\)</span> for different parameters <span class="math inline">\(k\)</span>; <span class="citation" data-cites="SailynojaEtAl:2022">Säilynoja, Bürkner, and Vehtari (<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">2022</a>)</span> provide a graphical test for discrete uniformity testing. Before uniformity testing the Markov chains should be thinned to remove autocorrelation as these uniformity tests assume independence <span class="citation" data-cites="SailynojaEtAl:2022">(<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">Säilynoja, Bürkner, and Vehtari 2022</a>)</span>.</p>
 </section>
 <section id="sbc-in-stan" class="level2">
 <h2 class="anchored" data-anchor-id="sbc-in-stan">SBC in Stan</h2>
-<p>Running simulation-based calibration in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.</p>
+<p>Running simulation-based calibration checking in Stan will test whether Stan’s sampling algorithm can sample from the posterior associated with data generated according to the model. The data simulation and posterior fitting and rank calculation can all be done within a single Stan program. Then Stan’s posterior sampler has to be run multiple times. Each run produces a rank for each parameter being assessed for uniformity. The total set of ranks can then be tested for uniformity.</p>
 <section id="example-model" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="example-model">Example model</h3>
 <p>For illustration, a very simple model will suffice. Suppose there are two parameters <span class="math inline">\((\mu, \sigma)\)</span> with independent priors, <span class="math display">\[
@@ -688,14 +700,14 @@ <h3 class="unnumbered anchored" data-anchor-id="example-model">Example model</h3
 p(\mu, \sigma)
 = \textrm{normal}(\mu \mid 0, 1)
   \cdot \textrm{lognormal}(\sigma \mid 0, 1),
-\]</span> and the sampling density is <span class="math display">\[
+\]</span> and the data model is <span class="math display">\[
 p(y \mid \mu, \sigma)
 = \prod_{n=1}^N \textrm{normal}(y_n \mid \mu, \sigma).
 \]</span></p>
 <p>For example, suppose the following two parameter values are drawn from the prior in the first simulation, <span class="math display">\[
 (\mu^{\textrm{sim(1)}}, \sigma^{\textrm{sim(1)}}) = (1.01, 0.23).
 \]</span> Then data <span class="math inline">\(y^{\textrm{sim}(1)} \sim p(y \mid \mu^{\textrm{sim(1)}},
-\sigma^{\textrm{sim(1)}})\)</span> is drawn according to the sampling distribution. Next, <span class="math inline">\(M = 4\)</span> draws are taken from the posterior <span class="math inline">\(\mu^{(1,m)}, \sigma^{(1,m)} \sim p(\mu, \sigma \mid y^{\textrm{sim}(1)})\)</span>, <span class="math display">\[
+\sigma^{\textrm{sim(1)}})\)</span> is drawn according to the data model. Next, <span class="math inline">\(M = 4\)</span> draws are taken from the posterior <span class="math inline">\(\mu^{(1,m)}, \sigma^{(1,m)} \sim p(\mu, \sigma \mid y^{\textrm{sim}(1)})\)</span>, <span class="math display">\[
 \begin{array}{r|rr}
 m &amp; \mu^{(1,m)} &amp; \sigma^{(1,m)}
 \\ \hline
@@ -722,9 +734,9 @@ <h3 class="unnumbered anchored" data-anchor-id="example-model">Example model</h3
 \end{array}
 \]</span> The ranks are the column sums, <span class="math inline">\(r_{1,1} = 2\)</span> and <span class="math inline">\(r_{1,2} = 1\)</span>. Because the simulated parameters are distributed according to the posterior, these ranks should be distributed uniformly between <span class="math inline">\(0\)</span> and <span class="math inline">\(M\)</span>, the number of posterior draws.</p>
 </section>
-<section id="testing-a-stan-program-with-simulation-based-calibration" class="level3 unnumbered">
-<h3 class="unnumbered anchored" data-anchor-id="testing-a-stan-program-with-simulation-based-calibration">Testing a Stan program with simulation-based calibration</h3>
-<p>To code simulation-based calibration in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.</p>
+<section id="testing-a-stan-program-with-simulation-based-calibration-checking" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="testing-a-stan-program-with-simulation-based-calibration-checking">Testing a Stan program with simulation-based calibration checking</h3>
+<p>To code simulation-based calibration checking in a Stan program, the transformed data block can be used to simulate parameters and data from the model. The parameters, transformed parameters, and model block then define the model over the simulated data. Then, in the generated quantities block, the program records an indicator for whether each parameter is less than the simulated value. As shown above, the rank is then the sum of the simulated indicator variables.</p>
 <div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span> {</span>
 <span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> mu_sim = normal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
 <span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sigma_sim = lognormal_rng(<span class="dv">0</span>, <span class="dv">1</span>);</span>
@@ -747,12 +759,12 @@ <h3 class="unnumbered anchored" data-anchor-id="testing-a-stan-program-with-simu
 <span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[<span class="dv">2</span>] <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; lt_sim</span>
 <span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>      = { mu &lt; mu_sim, sigma &lt; sigma_sim };</span>
 <span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
-<p>To avoid confusion with the number of simulated data sets used for simulation-based calibration, <code>J</code> is used for the number of simulated data points.</p>
+<p>To avoid confusion with the number of simulated data sets used for simulation-based calibration checking, <code>J</code> is used for the number of simulated data points.</p>
 <p>The model is implemented twice—once as a data generating process using random number generators in the transformed data block, then again in the parameters and model block. This duplication is a blessing and a curse. The curse is that it’s more work and twice the chance for errors. The blessing is that by implementing the model twice and comparing results, the chance of there being a mistake in the model is reduced.</p>
 </section>
-<section id="pseudocode-for-simulation-based-calibration" class="level3 unnumbered">
-<h3 class="unnumbered anchored" data-anchor-id="pseudocode-for-simulation-based-calibration">Pseudocode for simulation-based calibration</h3>
-<p>The entire simulation-based calibration process is as follows, where</p>
+<section id="pseudocode-for-simulation-based-calibration-checking" class="level3 unnumbered">
+<h3 class="unnumbered anchored" data-anchor-id="pseudocode-for-simulation-based-calibration-checking">Pseudocode for simulation-based calibration checking</h3>
+<p>The entire simulation-based calibration checking process is as follows, where</p>
 <ul>
 <li><code>p(theta)</code> is the prior density</li>
 <li><code>p(y | theta)</code> is the sampling density</li>
@@ -783,12 +795,12 @@ <h3 class="unnumbered anchored" data-anchor-id="pseudocode-for-simulation-based-
 </section>
 <section id="the-importance-of-thinning" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="the-importance-of-thinning">The importance of thinning</h3>
-<p>The draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws. Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of <code>M</code>, then thinning down to <code>M</code> draws.</p>
+<p>The draws from the posterior are assumed to be roughly independent. If they are not, artifacts may arise in the uniformity tests due to correlation in the posterior draws <span class="citation" data-cites="SailynojaEtAl:2022">(<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">Säilynoja, Bürkner, and Vehtari 2022</a>)</span>. Thus it is best to thin the posterior draws down to the point where the effective sample size is roughly the same as the number of thinned draws. This may require running the code a few times to judge the number of draws required to produce a target effective sample size. This operation that can be put into a loop that doubles the number of iterations until all parameters have an effective sample size of <code>M</code>, then thinning down to <code>M</code> draws.</p>
 </section>
 </section>
 <section id="testing-uniformity" class="level2">
 <h2 class="anchored" data-anchor-id="testing-uniformity">Testing uniformity</h2>
-<p>A simple, though not very highly powered, <span class="math inline">\(\chi^2\)</span>-squared test for uniformity can be formulated by binning the ranks <span class="math inline">\(0:M\)</span> into <span class="math inline">\(J\)</span> bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, <span class="citation" data-cites="CookGelmanRubin:2006">Cook, Gelman, and Rubin (<a href="#ref-CookGelmanRubin:2006" role="doc-biblioref">2006</a>)</span> transform the ranks using the inverse cumulative distribution function for the standard normal and then perform a test for normality. <span class="citation" data-cites="TaltsEtAl:2018">Talts et al. (<a href="#ref-TaltsEtAl:2018" role="doc-biblioref">2018</a>)</span> recommend visual inspection of the binned plots.</p>
+<p>A simple, though not very highly powered, <span class="math inline">\(\chi^2\)</span>-squared test for uniformity can be formulated by binning the ranks <span class="math inline">\(0:M\)</span> into <span class="math inline">\(J\)</span> bins and testing that the bins all have roughly the expected number of draws in them. Many other tests for uniformity are possible. For example, <span class="citation" data-cites="SailynojaEtAl:2022">Säilynoja, Bürkner, and Vehtari (<a href="#ref-SailynojaEtAl:2022" role="doc-biblioref">2022</a>)</span> use binomial model pointiwise for the empirical cumlative distribution function and adjust to obtain simulatenous envelope to be used as graphical uniformity test.</p>
 <p>The bins don’t need to be exactly the same size. In general, if <span class="math inline">\(b_j\)</span> is the number of ranks that fall into bin <span class="math inline">\(j\)</span> and <span class="math inline">\(e_j\)</span> is the number of ranks expected to fall into bin <span class="math inline">\(j\)</span> (which will be proportional to its size under uniformity), the test statistic is <span class="math display">\[
 X^2 = \sum_{j = 1}^J \frac{(b_j - e_j)^2}{e_j}.
 \]</span> The terms are approximately square standard normal, so that under the null hypothesis of uniformity, <span class="math display">\[
@@ -817,8 +829,8 @@ <h3 class="unnumbered anchored" data-anchor-id="indexing-to-simplify-arithmetic"
 <p>In general, a great deal of care must be taken in visualizing discrete data because it’s easy to introduce off-by-one errors and artifacts at the edges because of the way boundaries are computed by default. That’s why so much attention must be devoted to indexing and binning.</p>
 </section>
 </section>
-<section id="examples-of-simulation-based-calibration" class="level2">
-<h2 class="anchored" data-anchor-id="examples-of-simulation-based-calibration">Examples of simulation-based calibration</h2>
+<section id="examples-of-simulation-based-calibration-checking" class="level2">
+<h2 class="anchored" data-anchor-id="examples-of-simulation-based-calibration-checking">Examples of simulation-based calibration checking</h2>
 <p>This section will show what the results look like when the tests pass and then when they fail. The passing test will compare a normal model and normal data generating process, whereas the second will compare a normal model with a Student-t data generating process. The first will produce calibrated posteriors, the second will not.</p>
 <section id="when-things-go-right" class="level3 unnumbered">
 <h3 class="unnumbered anchored" data-anchor-id="when-things-go-right">When things go right</h3>
@@ -917,7 +929,7 @@ <h3 class="unnumbered anchored" data-anchor-id="when-stans-sampler-goes-wrong">W
 <span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; tau_lt_sim = tau &lt; tau_sim;</span>
 <span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span>&lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; theta1_lt_sim = theta[<span class="dv">1</span>] &lt; theta_sim[<span class="dv">1</span>];</span>
 <span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
-<p>As usual for simulation-based calibration, the transformed data encodes the data-generating process using random number generators. Here, the population parameters <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\tau\)</span> are first simulated, then the school-level effects <span class="math inline">\(\theta\)</span>, and then finally the observed data <span class="math inline">\(\sigma_j\)</span> and <span class="math inline">\(y_j.\)</span> The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only <span class="math inline">\(\theta_1\)</span> because the schools are exchangeable in the simulation.</p>
+<p>As usual for simulation-based calibration checking, the transformed data encodes the data-generating process using random number generators. Here, the population parameters <span class="math inline">\(\mu\)</span> and <span class="math inline">\(\tau\)</span> are first simulated, then the school-level effects <span class="math inline">\(\theta\)</span>, and then finally the observed data <span class="math inline">\(\sigma_j\)</span> and <span class="math inline">\(y_j.\)</span> The parameters and model are a direct encoding of the mathematical presentation using vectorized sampling statements. The generated quantities block includes indicators for parameter comparisons, saving only <span class="math inline">\(\theta_1\)</span> because the schools are exchangeable in the simulation.</p>
 <p>When fitting the model in Stan, multiple warning messages are provided that the sampler has diverged. The divergence warnings are in Stan’s sampler precisely to diagnose the sampler’s inability to follow the curvature in the posterior and provide independent confirmation that Stan’s sampler cannot fit this model as specified.</p>
 <p>SBC also diagnoses the problem. Here’s the rank plots for running <span class="math inline">\(N =
 200\)</span> simulations with 1000 warmup iterations and <span class="math inline">\(M = 999\)</span> draws per simulation used to compute the ranks.</p>
@@ -986,9 +998,15 @@ <h3 class="unnumbered anchored" data-anchor-id="when-stans-sampler-goes-wrong">W
 <div id="ref-GneitingEtAl:2007" class="csl-entry" role="listitem">
 Gneiting, Tilmann, Fadoua Balabdaoui, and Adrian E Raftery. 2007. <span>“Probabilistic Forecasts, Calibration and Sharpness.”</span> <em>Journal of the Royal Statistical Society: Series B (Statistical Methodology)</em> 69 (2): 243–68.
 </div>
+<div id="ref-ModrakEtAl:2023" class="csl-entry" role="listitem">
+Modrák, Martin, Angie H Moon, Shinyoung Kim, Paul Bürkner, Niko Huurre, Kateřina Faltejsková, Andrew Gelman, and Aki Vehtari. 2023. <span>“Simulation-Based Calibration Checking for <span>Bayesian</span> Computation: The Choice of Test Quantities Shapes Sensitivity.”</span> <em>Bayesian Analysis</em> 20 (2): 461.
+</div>
 <div id="ref-Rubin:1981" class="csl-entry" role="listitem">
 Rubin, Donald B. 1981. <span>“Estimation in Parallel Randomized Experiments.”</span> <em>Journal of Educational Statistics</em> 6: 377–401.
 </div>
+<div id="ref-SailynojaEtAl:2022" class="csl-entry" role="listitem">
+Säilynoja, Teemu, Paul-Christian Bürkner, and Aki Vehtari. 2022. <span>“Graphical Test for Discrete Uniformity and Its Applications in Goodness-of-Fit Evaluation and Multiple Sample Comparison.”</span> <em>Statistics and Computing</em> 32 (2): 32.
+</div>
 <div id="ref-TaltsEtAl:2018" class="csl-entry" role="listitem">
 Talts, Sean, Michael Betancourt, Daniel Simpson, Aki Vehtari, and Andrew Gelman. 2018. <span>“Validating <span>B</span>ayesian Inference Algorithms with Simulation-Based Calibration.”</span> <em>arXiv</em>, no. 1804.06788.
 </div>
@@ -1567,8 +1585,35 @@ <h3 class="unnumbered anchored" data-anchor-id="when-stans-sampler-goes-wrong">W
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/simulation-based-calibration.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/simulation-based-calibration.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/sparse-ragged.html b/docs/stan-users-guide/sparse-ragged.html
index 70f1f99ea..f7b72e969 100644
--- a/docs/stan-users-guide/sparse-ragged.html
+++ b/docs/stan-users-guide/sparse-ragged.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1325,8 +1337,35 @@ <h2 class="anchored" data-anchor-id="ragged-data-structs.section">Ragged data st
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/sparse-ragged.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/sparse-ragged.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/style-guide.html b/docs/stan-users-guide/style-guide.html
index ca18559fc..d2fc86cd0 100644
--- a/docs/stan-users-guide/style-guide.html
+++ b/docs/stan-users-guide/style-guide.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1421,8 +1433,35 @@ <h4 class="unnumbered anchored" data-anchor-id="platform-specificity-of-newlines
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/style-guide.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/style-guide.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/survival.html b/docs/stan-users-guide/survival.html
index e7b04bf70..314136c57 100644
--- a/docs/stan-users-guide/survival.html
+++ b/docs/stan-users-guide/survival.html
@@ -84,7 +84,7 @@
 <script src="../site_libs/quarto-search/fuse.min.js"></script>
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
-<link href="../stan-users-guide/floating-point.html" rel="next">
+<link href="../stan-users-guide/wiener_diffusion_model.html" rel="next">
 <link href="../stan-users-guide/dae.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link active">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1618,14 +1630,41 @@ <h3 class="unnumbered anchored" data-anchor-id="stan-model-for-tied-survival-tim
       </a>          
   </div>
   <div class="nav-page nav-page-next">
-      <a href="../stan-users-guide/floating-point.html" class="pagination-link" aria-label="Floating Point Arithmetic">
-        <span class="nav-page-text">Floating Point Arithmetic</span> <i class="bi bi-arrow-right-short"></i>
+      <a href="../stan-users-guide/wiener_diffusion_model.html" class="pagination-link" aria-label="Wiener diffusion model">
+        <span class="nav-page-text">Wiener diffusion model</span> <i class="bi bi-arrow-right-short"></i>
       </a>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/survival.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/survival.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/time-series.html b/docs/stan-users-guide/time-series.html
index 5a6c93f76..0b0120064 100644
--- a/docs/stan-users-guide/time-series.html
+++ b/docs/stan-users-guide/time-series.html
@@ -309,7 +309,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -427,6 +427,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -519,7 +525,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -551,6 +557,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1643,8 +1655,35 @@ <h2 class="anchored" data-anchor-id="hmms.section">Hidden Markov models</h2>
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/time-series.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/time-series.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/truncation-censoring.html b/docs/stan-users-guide/truncation-censoring.html
index 03906c6ae..fe4b0bdc9 100644
--- a/docs/stan-users-guide/truncation-censoring.html
+++ b/docs/stan-users-guide/truncation-censoring.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1309,8 +1321,35 @@ <h3 class="unnumbered anchored" data-anchor-id="integrating-out-censored-values"
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/truncation-censoring.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/truncation-censoring.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/user-functions.html b/docs/stan-users-guide/user-functions.html
index 769fbe52d..f5a2f7e71 100644
--- a/docs/stan-users-guide/user-functions.html
+++ b/docs/stan-users-guide/user-functions.html
@@ -289,7 +289,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -407,6 +407,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -499,7 +505,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -531,6 +537,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1596,8 +1608,35 @@ <h4 class="unnumbered anchored" data-anchor-id="truncating-above-and-below">Trun
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/user-functions.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/using-stanc.html b/docs/stan-users-guide/using-stanc.html
index a509cde5f..27987f747 100644
--- a/docs/stan-users-guide/using-stanc.html
+++ b/docs/stan-users-guide/using-stanc.html
@@ -65,7 +65,7 @@
 <script src="../site_libs/quarto-search/quarto-search.js"></script>
 <meta name="quarto:offset" content="../">
 <link href="../stan-users-guide/style-guide.html" rel="next">
-<link href="../stan-users-guide/bootstrap.html" rel="prev">
+<link href="../stan-users-guide/multiple-imputation.html" rel="prev">
 <link href="../img/logo_tm.png" rel="icon" type="image/png">
 <script src="../site_libs/quarto-html/quarto.js"></script>
 <script src="../site_libs/quarto-html/popper.min.js"></script>
@@ -260,7 +260,7 @@
 </li>
         <li class="sidebar-item sidebar-item-section">
       <span class="sidebar-item-text sidebar-link text-start">
- <span class="menu-text">Version 2.38</span></span>
+ <span class="menu-text">Version 2.39</span></span>
   </li>
         <li class="sidebar-item sidebar-item-section">
       <div class="sidebar-item-container"> 
@@ -378,6 +378,12 @@
   <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">Survival Models</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -470,7 +476,7 @@
           <li class="sidebar-item">
   <div class="sidebar-item-container"> 
   <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
- <span class="menu-text">Simulation-Based Calibration</span></a>
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
   </div>
 </li>
           <li class="sidebar-item">
@@ -502,6 +508,12 @@
   <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
  <span class="menu-text">The Bootstrap and Bagging</span></a>
   </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
 </li>
       </ul>
   </li>
@@ -1991,8 +2003,8 @@ <h4 class="unnumbered anchored" data-anchor-id="static-loop-unrolling">Static lo
 </script>
 <nav class="page-navigation">
   <div class="nav-page nav-page-previous">
-      <a href="../stan-users-guide/bootstrap.html" class="pagination-link" aria-label="The Bootstrap and Bagging">
-        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">The Bootstrap and Bagging</span>
+      <a href="../stan-users-guide/multiple-imputation.html" class="pagination-link" aria-label="Multiple imputation">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Multiple imputation</span>
       </a>          
   </div>
   <div class="nav-page nav-page-next">
@@ -2002,8 +2014,35 @@ <h4 class="unnumbered anchored" data-anchor-id="static-loop-unrolling">Static lo
   </div>
 </nav>
 </div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/using-stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
 
 
 
 
-<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/using-stanc.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer></body></html>
\ No newline at end of file
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/wiener_diffusion_model.html b/docs/stan-users-guide/wiener_diffusion_model.html
new file mode 100644
index 000000000..b8cbc39e1
--- /dev/null
+++ b/docs/stan-users-guide/wiener_diffusion_model.html
@@ -0,0 +1,1598 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.6.39">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Wiener Diffusion Model</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+/* CSS for citations */
+div.csl-bib-body { }
+div.csl-entry {
+  clear: both;
+  margin-bottom: 0em;
+}
+.hanging-indent div.csl-entry {
+  margin-left:2em;
+  text-indent:-2em;
+}
+div.csl-left-margin {
+  min-width:2em;
+  float:left;
+}
+div.csl-right-inline {
+  margin-left:2em;
+  padding-left:1em;
+}
+div.csl-indent {
+  margin-left: 2em;
+}</style>
+
+
+<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="../site_libs/quarto-nav/headroom.min.js"></script>
+<script src="../site_libs/clipboard/clipboard.min.js"></script>
+<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="../site_libs/quarto-search/fuse.min.js"></script>
+<script src="../site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="../">
+<link href="../stan-users-guide/floating-point.html" rel="next">
+<link href="../stan-users-guide/survival.html" rel="prev">
+<link href="../img/logo_tm.png" rel="icon" type="image/png">
+<script src="../site_libs/quarto-html/quarto.js"></script>
+<script src="../site_libs/quarto-html/popper.min.js"></script>
+<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="../site_libs/quarto-html/anchor.min.js"></script>
+<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-05fe91a66cf75bbbb8c9664867fe5124.css" rel="stylesheet" class="quarto-color-scheme" id="quarto-text-highlighting-styles">
+<link href="../site_libs/quarto-html/quarto-syntax-highlighting-dark-59fdbf2653e1aaac23acf64cc15a0d7d.css" rel="prefetch" class="quarto-color-scheme quarto-color-alternate" id="quarto-text-highlighting-styles">
+<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="../site_libs/bootstrap/bootstrap-c777a6582eb78ab223b9896ab55f446e.min.css" rel="stylesheet" append-hash="true" class="quarto-color-scheme" id="quarto-bootstrap" data-mode="light">
+<link href="../site_libs/bootstrap/bootstrap-dark-2df8a8547a8386440a781d5ef2c5f5b1.min.css" rel="prefetch" append-hash="true" class="quarto-color-scheme quarto-color-alternate" id="quarto-bootstrap" data-mode="dark">
+<script id="quarto-search-options" type="application/json">{
+  "location": "navbar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "end",
+  "type": "overlay",
+  "limit": 50,
+  "keyboard-shortcut": [
+    "f",
+    "/",
+    "s"
+  ],
+  "show-item-context": false,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-text-placeholder": "",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+<script type="text/javascript">
+const typesetMath = (el) => {
+  if (window.MathJax) {
+    // MathJax Typeset
+    window.MathJax.typeset([el]);
+  } else if (window.katex) {
+    // KaTeX Render
+    var mathElements = el.getElementsByClassName("math");
+    var macros = [];
+    for (var i = 0; i < mathElements.length; i++) {
+      var texText = mathElements[i].firstChild;
+      if (mathElements[i].tagName == "SPAN") {
+        window.katex.render(texText.data, mathElements[i], {
+          displayMode: mathElements[i].classList.contains('display'),
+          throwOnError: false,
+          macros: macros,
+          fleqn: false
+        });
+      }
+    }
+  }
+}
+window.Quarto = {
+  typesetMath
+};
+</script>
+
+<link rel="stylesheet" href="../quarto-config/quarto_styles.css">
+<meta property="og:title" content="Wiener Diffusion Model">
+<meta property="og:description" content="">
+<meta property="og:image" content="img/Figure_DDM.png">
+<meta property="og:site_name" content="Stan Docs">
+</head>
+
+<body class="nav-sidebar floating nav-fixed">
+
+<div id="quarto-search-results"></div>
+  <header id="quarto-header" class="headroom fixed-top">
+    <nav class="navbar navbar-expand-lg " data-bs-theme="dark">
+      <div class="navbar-container container-fluid">
+      <div class="navbar-brand-container mx-auto">
+    <a href="https://mc-stan.org/" class="navbar-brand navbar-brand-logo">
+    <img src="../img/logo_tm.png" alt="Stan logo" class="navbar-logo">
+    </a>
+  </div>
+            <div id="quarto-search" class="" title="Search"></div>
+          <button class="navbar-toggler" type="button" data-bs-toggle="collapse" data-bs-target="#navbarCollapse" aria-controls="navbarCollapse" role="menu" aria-expanded="false" aria-label="Toggle navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+  <span class="navbar-toggler-icon"></span>
+</button>
+          <div class="collapse navbar-collapse" id="navbarCollapse">
+            <ul class="navbar-nav navbar-nav-scroll me-auto">
+  <li class="nav-item">
+    <a class="nav-link" href="../index.html"> 
+<span class="menu-text">Documentation</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link active" href="../stan-users-guide/index.html" aria-current="page"> 
+<span class="menu-text">Stan Users Guide</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../reference-manual/index.html"> 
+<span class="menu-text">Reference Manual</span></a>
+  </li>  
+  <li class="nav-item">
+    <a class="nav-link" href="../functions-reference/index.html"> 
+<span class="menu-text">Functions Reference</span></a>
+  </li>  
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-interfaces" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Interfaces</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-interfaces">    
+        <li>
+    <a class="dropdown-item" href="../cmdstan-guide/index.html">
+ <span class="dropdown-text">CmdStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanpy" target="_blank">
+ <span class="dropdown-text">CmdStanPy</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/cmdstanr" target="_blank">
+ <span class="dropdown-text">CmdStanR</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://pystan.readthedocs.io/en/latest/" target="_blank">
+ <span class="dropdown-text">PyStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstan" target="_blank">
+ <span class="dropdown-text">RStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="http://stanjulia.github.io/Stan.jl/stable/INTRO/" target="_blank">
+ <span class="dropdown-text">Stan.jl</span></a>
+  </li>  
+    </ul>
+  </li>
+  <li class="nav-item dropdown ">
+    <a class="nav-link dropdown-toggle" href="#" id="nav-menu-other-packages" role="link" data-bs-toggle="dropdown" aria-expanded="false">
+ <span class="menu-text">Other Packages</span>
+    </a>
+    <ul class="dropdown-menu" aria-labelledby="nav-menu-other-packages">    
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/bayesplot/" target="_blank">
+ <span class="dropdown-text">bayesplot</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://roualdes.github.io/bridgestan/latest/" target="_blank">
+ <span class="dropdown-text">BridgeStan</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://paul-buerkner.github.io/brms/" target="_blank">
+ <span class="dropdown-text">brms</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/loo/" target="_blank">
+ <span class="dropdown-text">loo</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/posterior" target="_blank">
+ <span class="dropdown-text">posterior</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/projpred" target="_blank">
+ <span class="dropdown-text">projpred</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstanarm" target="_blank">
+ <span class="dropdown-text">rstanarm</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/rstantools" target="_blank">
+ <span class="dropdown-text">rstantools</span></a>
+  </li>  
+        <li>
+    <a class="dropdown-item" href="https://mc-stan.org/shinystan" target="_blank">
+ <span class="dropdown-text">shinystan</span></a>
+  </li>  
+    </ul>
+  </li>
+</ul>
+          </div> <!-- /navcollapse -->
+            <div class="quarto-navbar-tools tools-wide">
+    <a href="https://github.com/stan-dev" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-github"></i></a>
+    <a href="https://discourse.mc-stan.org" title="" class="quarto-navigation-tool px-1" aria-label=""><i class="bi bi-chat-text-fill"></i></a>
+  <a href="" class="quarto-color-scheme-toggle quarto-navigation-tool  px-1" onclick="window.quartoToggleColorScheme(); return false;" title="Toggle dark mode"><i class="bi"></i></a>
+  <a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
+  <div class="quarto-reader-toggle-btn">
+  <i class="bi"></i>
+  </div>
+</a>
+</div>
+      </div> <!-- /container-fluid -->
+    </nav>
+  <nav class="quarto-secondary-nav">
+    <div class="container-fluid d-flex">
+      <button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
+        <i class="bi bi-layout-text-sidebar-reverse"></i>
+      </button>
+        <nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/wiener_diffusion_model.html">Wiener diffusion model</a></li></ol></nav>
+        <a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">      
+        </a>
+    </div>
+  </nav>
+</header>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article page-navbar">
+<!-- sidebar -->
+  <nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation floating overflow-auto">
+    <div class="sidebar-menu-container"> 
+    <ul class="list-unstyled mt-1">
+        <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/index.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan User’s Guide</span></a>
+  </div>
+</li>
+        <li class="sidebar-item sidebar-item-section">
+      <span class="sidebar-item-text sidebar-link text-start">
+ <span class="menu-text">Version 2.39</span></span>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
+ <span class="menu-text">Example Models</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/regression.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Regression Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/time-series.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Time-Series Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/missing-data.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Missing Data and Partially Known Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/truncation-censoring.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Truncated or Censored Data</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/finite-mixtures.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Finite Mixtures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/measurement-error.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Measurement Error and Meta-Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/latent-discrete.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Latent Discrete Parameters</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/sparse-ragged.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Sparse and Ragged Data Structures</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/clustering.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Clustering Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/gaussian-processes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Gaussian Processes</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/copulas.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Copulas</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/hyperspherical-models.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Directions, Rotations, and Hyperspheres</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/algebraic-equations.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Solving Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/odes.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Ordinary Differential Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/one-dimensional-integrals.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Computing One Dimensional Integrals</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/complex-numbers.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Complex Numbers</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/dae.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Differential-Algebraic Equations</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/survival.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Survival Models</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/wiener_diffusion_model.html" class="sidebar-item-text sidebar-link active">
+ <span class="menu-text">Wiener diffusion model</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
+ <span class="menu-text">Programming Techniques</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/floating-point.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Floating Point Arithmetic</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/matrices-arrays.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Matrices, Vectors, Arrays, and Tuples</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multi-indexing.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple Indexing and Range Indexing</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/user-functions.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">User-Defined Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/custom-probability.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Custom Probability Functions</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/proportionality-constants.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Proportionality Constants</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/problematic-posteriors.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Problematic Posteriors</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/reparameterization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Reparameterization and Change of Variables</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/efficiency-tuning.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Efficiency Tuning</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/parallelization.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Parallelization</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
+ <span class="menu-text">Posterior Inference &amp; Model Checking</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-prediction.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior Predictive Sampling</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/simulation-based-calibration.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Simulation-Based Calibration Checking</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/posterior-predictive-checks.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Posterior and Prior Predictive Checks</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/cross-validation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Held-Out Evaluation and Cross-Validation</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/poststratification.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Poststratification</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/decision-analysis.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Decision Analysis</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/bootstrap.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">The Bootstrap and Bagging</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/multiple-imputation.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Multiple imputation</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+        <li class="sidebar-item sidebar-item-section">
+      <div class="sidebar-item-container"> 
+            <a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true">
+ <span class="menu-text">Appendices</span></a>
+          <a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-5" role="navigation" aria-expanded="true" aria-label="Toggle section">
+            <i class="bi bi-chevron-right ms-2"></i>
+          </a> 
+      </div>
+      <ul id="quarto-sidebar-section-5" class="collapse list-unstyled sidebar-section depth1 show">  
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/using-stanc.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Using the Stan Compiler</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/style-guide.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Stan Program Style Guide</span></a>
+  </div>
+</li>
+          <li class="sidebar-item">
+  <div class="sidebar-item-container"> 
+  <a href="../stan-users-guide/for-bugs-users.html" class="sidebar-item-text sidebar-link">
+ <span class="menu-text">Transitioning from BUGS</span></a>
+  </div>
+</li>
+      </ul>
+  </li>
+    </ul>
+    </div>
+</nav>
+<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
+<!-- margin-sidebar -->
+    <div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
+        <nav id="TOC" role="doc-toc" class="toc-active">
+    <h2 id="toc-title">On this page</h2>
+   
+  <ul>
+  <li><a href="#wiener-diffusion-model" id="toc-wiener-diffusion-model" class="nav-link active" data-scroll-target="#wiener-diffusion-model">Wiener diffusion model</a>
+  <ul class="collapse">
+  <li><a href="#function-call-example" id="toc-function-call-example" class="nav-link" data-scroll-target="#function-call-example">Function call example</a>
+  <ul class="collapse">
+  <li><a href="#the-data-block" id="toc-the-data-block" class="nav-link" data-scroll-target="#the-data-block">The data block</a></li>
+  <li><a href="#the-parameters-block" id="toc-the-parameters-block" class="nav-link" data-scroll-target="#the-parameters-block">The parameters block</a></li>
+  <li><a href="#the-model-block" id="toc-the-model-block" class="nav-link" data-scroll-target="#the-model-block">The model block</a></li>
+  </ul></li>
+  <li><a href="#wiener-truncation" id="toc-wiener-truncation" class="nav-link" data-scroll-target="#wiener-truncation">Truncated and censored data</a>
+  <ul class="collapse">
+  <li><a href="#modeling-truncated-data-with-the-diffusion-model" id="toc-modeling-truncated-data-with-the-diffusion-model" class="nav-link" data-scroll-target="#modeling-truncated-data-with-the-diffusion-model">Modeling truncated data with the diffusion model</a></li>
+  <li><a href="#modeling-censored-data-with-the-diffusion-model" id="toc-modeling-censored-data-with-the-diffusion-model" class="nav-link" data-scroll-target="#modeling-censored-data-with-the-diffusion-model">Modeling censored data with the diffusion model</a></li>
+  </ul></li>
+  </ul></li>
+  </ul>
+<div class="toc-actions"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/wiener_diffusion_model.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div><div class="quarto-alternate-formats"><h2>Other Formats</h2><ul><li><a href="wiener_diffusion_model.pdf"><i class="bi bi-file-pdf"></i>PDF</a></li></ul></div></nav>
+    </div>
+<!-- main -->
+<main class="content" id="quarto-document-content"><header id="title-block-header" class="quarto-title-block"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="../stan-users-guide/regression.html">Example Models</a></li><li class="breadcrumb-item"><a href="../stan-users-guide/wiener_diffusion_model.html">Wiener diffusion model</a></li></ol></nav></header>
+
+
+
+
+<section id="wiener-diffusion-model" class="level1">
+<h1>Wiener diffusion model</h1>
+<p>Diffusion models, sometimes also called Wiener diffusion models, are among the most frequently used model families in modeling two-alternative forced-choice tasks (see <span class="citation" data-cites="Wagenmakers2009">Wagenmakers (<a href="#ref-Wagenmakers2009" role="doc-biblioref">2009</a>)</span>, for a review). Diffusion models allow to model response times and responses jointly. The basic version of a diffusion model comprises four parameters: the boundary separation, <span class="math inline">\(a\)</span>, the relative starting point, <span class="math inline">\(w\)</span>, the drift rate, <span class="math inline">\(v\)</span>, and the non-decision time, <span class="math inline">\(t0\)</span> <span class="citation" data-cites="Ratcliff1978">(<a href="#ref-Ratcliff1978" role="doc-biblioref">Ratcliff 1978</a>)</span>. In the basic model, it is assumed that the four basic parameters are the same for the whole experiment. As this assumption is very strict and there are examples that suggest that the basic parameters can be different from trial to trial, so called inter-trial variabilities were introduced and the basic four- parameter model was extended to a seven-parameter model. In the seven-parameter extension of the diffusion model there are the following three parameters added: the inter-trial variability in relative starting point, <span class="math inline">\(s_w\)</span>, the inter-trial variability in drift rate, <span class="math inline">\(s_v\)</span>, and the inter-trial variability in non-decision time, <span class="math inline">\(s_{t0}\)</span> <span class="citation" data-cites="Ratcliff1998">Nicenboim, Schad, and Vasishth (<a href="#ref-nicenboim2025introduction" role="doc-biblioref">2025</a>)</span>.</p>
+<p>Data for the diffusion model is two-dimensional: There is one vector for the reaction times, <span class="math inline">\(y\)</span>, and one vector for the given responses, <span class="math inline">\(\text{resp}\)</span>. The reaction times shall be positive, continuous and in seconds, the responses shall be binary.</p>
+<p>As a diffusion model describes the decision process for a decision with exactly two choices, there exist reaction time distributions for each response alternative. This means that the probability density function (<span class="math inline">\(p\)</span>) splits into one part for one response alternative and one part for the other response alternative. In the following, we will refer to one alternative as the <em>upper response boundary</em> and to the other alternative as the <em>lower response boundary</em>. <span class="math inline">\(p\)</span> of the lower response boundary can be obtained when inserting <span class="math inline">\(-v\)</span> and <span class="math inline">\(1-w\)</span> to <span class="math inline">\(p\)</span> of the upper response boundary. Let’s call <span class="math inline">\(p\)</span> for the lower response boundary <span class="math inline">\(p_0\)</span> and <span class="math inline">\(p\)</span> for the upper response boundary <span class="math inline">\(p_1\)</span>. Then:</p>
+<p><span class="math display">\[
+p_0(a,t0,v,w,sv,sw,st0) = p_1(a,t0,-v,1-w,sv,sw,st0)
+\]</span></p>
+<p>Usually, a <span class="math inline">\(PDF\)</span> integrates to 1. In the case of the diffusion model, only the sum of both parts, <span class="math inline">\(p_0\)</span> and <span class="math inline">\(p_1\)</span>, integrates to 1. This is called <em>defective</em>.</p>
+<div id="fig-DDM" class="quarto-float quarto-figure quarto-figure-center anchored">
+<figure class="quarto-float quarto-float-fig figure">
+<div aria-describedby="fig-DDM-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+<img src="img/Figure_DDM.png" class="img-fluid figure-img">
+</div>
+<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-DDM-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
+Figure&nbsp;1: Figure 1: Realization of a Four-Parameter Diffusion Process Modeling the Binary Decision Process. Image from <span class="citation" data-cites="Henrich2024">Henrich et al. (<a href="#ref-Henrich2024" role="doc-biblioref">2024</a>)</span>, distributed under the Creative Commons Attribution 4.0 International License. <em>Note.</em> The parameters are the boundary separation a for two response alternatives, the relative starting point w, the drift rate v, and the non-decision time t0. The decision process is illustrated as a jagged line between the two boundaries. The predicted distributions of the reaction times are depicted as curved lines below and above the response boundaries (blue).
+</figcaption>
+</figure>
+</div>
+<p>In this model it is assumed that the decision process behaves like a random walk and we are interested in the first time that the random walk crosses one of the two decision boundaries. Hence, we are interested in the first-passage time of the decision process. The Stan function <code>wiener_lpdf()</code> returns the logarithm of the first-passage time density function for a diffusion model with up to seven parameters for upper boundary responses, <span class="math inline">\(\log(p_1)\)</span>. As can be seen above, it suffices to implement the density for only one response boundary, as the other can be obtained by mirroring the starting point and drift rate. Any combination of fixed and estimated parameters can be specified. In other words, with this implementation it is not only possible to estimate parameters of the fullseven- parameter model, but also to estimate restricted models such as the basic four- parameter model, or a five or six-parameter model, or even a one-parameter model when fixing the other six parameters.</p>
+<p>For example, it is possible to permit variability in just one or two parameters and to fix the other variabilities to 0, or even to estimate a three-parameter model when fixing more parameters (e.g., fixing the relative starting point at 0.5).</p>
+<p>It is assumed that the reaction time data that correspond to the upper response boundary <span class="math inline">\(y_\text{upper}\)</span> is distributed according to <code>wiener_lpdf()</code>:</p>
+<p><span class="math display">\[
+y_\text{lower} \sim \operatorname{wiener\_lpdf}(a, t0, w, v, s_v, s_w, s_{t0})
+\]</span> and the reaction time data that correspond to the lower response boundary <span class="math inline">\(y_\text{lower}\)</span> is distributed according to <code>wiener_lpdf()</code> with mirrored starting point and drift rate:</p>
+<p><span class="math display">\[
+y_\text{upper} \sim \operatorname{wiener\_lpdf}(a, t0, 1-w, -v, s_v, s_w, s_{t0})
+\]</span></p>
+<section id="function-call-example" class="level2">
+<h2 class="anchored" data-anchor-id="function-call-example">Function call example</h2>
+<p>The following example demonstrates a diffusion model call in Stan:</p>
+<div class="sourceCode" id="cb1"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="kw">data</span> {</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">int</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; N; <span class="co">// Number of trials</span></span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">real</span> rt; <span class="co">// response times (in seconds )</span></span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a>  <span class="dt">array</span>[N] <span class="dt">int</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; resp; <span class="co">// responses {0 ,1}</span></span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-6"><a href="#cb1-6" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed data</span>{</span>
+<span id="cb1-7"><a href="#cb1-7" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> min_rt = min(rt);</span>
+<span id="cb1-8"><a href="#cb1-8" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-9"><a href="#cb1-9" aria-hidden="true" tabindex="-1"></a><span class="kw">parameters</span> {</span>
+<span id="cb1-10"><a href="#cb1-10" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; a;                <span class="co">// boundary separation</span></span>
+<span id="cb1-11"><a href="#cb1-11" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> v;                          <span class="co">// drift</span></span>
+<span id="cb1-12"><a href="#cb1-12" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=<span class="dv">1</span>&gt; w;       <span class="co">// relative starting point</span></span>
+<span id="cb1-13"><a href="#cb1-13" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=min_rt&gt; t0; <span class="co">// non-decision time</span></span>
+<span id="cb1-14"><a href="#cb1-14" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-15"><a href="#cb1-15" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; sv;               <span class="co">// variability in drift</span></span>
+<span id="cb1-16"><a href="#cb1-16" aria-hidden="true" tabindex="-1"></a>  <span class="co">// variability in starting point</span></span>
+<span id="cb1-17"><a href="#cb1-17" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>, <span class="kw">upper</span>=fmin(<span class="dv">2</span> * w, <span class="dv">2</span> * (<span class="dv">1</span> - w))&gt; sw; </span>
+<span id="cb1-18"><a href="#cb1-18" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> &lt;<span class="kw">lower</span>=<span class="dv">0</span>&gt; st0;             <span class="co">// variability in non-decision time</span></span>
+<span id="cb1-19"><a href="#cb1-19" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-20"><a href="#cb1-20" aria-hidden="true" tabindex="-1"></a><span class="kw">transformed parameters</span>{</span>
+<span id="cb1-21"><a href="#cb1-21" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> one_minus_w = <span class="dv">1</span> - w;</span>
+<span id="cb1-22"><a href="#cb1-22" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> neg_v = -v;</span>
+<span id="cb1-23"><a href="#cb1-23" aria-hidden="true" tabindex="-1"></a>}</span>
+<span id="cb1-24"><a href="#cb1-24" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb1-25"><a href="#cb1-25" aria-hidden="true" tabindex="-1"></a>  <span class="co">// prior</span></span>
+<span id="cb1-26"><a href="#cb1-26" aria-hidden="true" tabindex="-1"></a>  a ~ normal(<span class="dv">1</span>, <span class="dv">1</span>);</span>
+<span id="cb1-27"><a href="#cb1-27" aria-hidden="true" tabindex="-1"></a>  w ~ normal(<span class="fl">0.5</span>, <span class="fl">0.1</span>);</span>
+<span id="cb1-28"><a href="#cb1-28" aria-hidden="true" tabindex="-1"></a>  v ~ normal(<span class="dv">2</span>, <span class="dv">3</span>);</span>
+<span id="cb1-29"><a href="#cb1-29" aria-hidden="true" tabindex="-1"></a>  t0 ~ normal(<span class="fl">0.435</span>, <span class="fl">0.12</span>);</span>
+<span id="cb1-30"><a href="#cb1-30" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-31"><a href="#cb1-31" aria-hidden="true" tabindex="-1"></a>  sv ~ normal(<span class="dv">1</span>, <span class="dv">3</span>);</span>
+<span id="cb1-32"><a href="#cb1-32" aria-hidden="true" tabindex="-1"></a>  st0 ~ normal(<span class="fl">0.183</span>, <span class="fl">0.09</span>);</span>
+<span id="cb1-33"><a href="#cb1-33" aria-hidden="true" tabindex="-1"></a>  sw ~ beta(<span class="dv">1</span>, <span class="dv">3</span>);</span>
+<span id="cb1-34"><a href="#cb1-34" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb1-35"><a href="#cb1-35" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood (diffusion model)</span></span>
+<span id="cb1-36"><a href="#cb1-36" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb1-37"><a href="#cb1-37" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb1-38"><a href="#cb1-38" aria-hidden="true" tabindex="-1"></a>      <span class="co">// upper boundary</span></span>
+<span id="cb1-39"><a href="#cb1-39" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_full_lpdf(rt[i] | a, t0, w, v,</span>
+<span id="cb1-40"><a href="#cb1-40" aria-hidden="true" tabindex="-1"></a>                                         sv, sw, st0);</span>
+<span id="cb1-41"><a href="#cb1-41" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> {</span>
+<span id="cb1-42"><a href="#cb1-42" aria-hidden="true" tabindex="-1"></a>      <span class="co">// lower boundary: mirror drift and starting point</span></span>
+<span id="cb1-43"><a href="#cb1-43" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_full_lpdf(rt[i] | a, t0, one_minus_w,</span>
+<span id="cb1-44"><a href="#cb1-44" aria-hidden="true" tabindex="-1"></a>                                         neg_v, sv, sw, st0);</span>
+<span id="cb1-45"><a href="#cb1-45" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb1-46"><a href="#cb1-46" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb1-47"><a href="#cb1-47" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<section id="the-data-block" class="level3">
+<h3 class="anchored" data-anchor-id="the-data-block">The data block</h3>
+<p>The data should consist of at least three variables:</p>
+<ol type="1">
+<li>The number of trials N,</li>
+<li>the response, coded as 0 = “lower bound” and 1 = “upper bound”, and</li>
+<li>the reaction times in seconds (not milliseconds).</li>
+</ol>
+<p>Note that two different ways of coding responses are commonly used: First, in <em>response coding</em>, the boundaries correspond to the two response alternatives. Second, in <em>accuracy coding</em>, the boundaries correspond to correct (upper bound) and wrong (lower bound) responses. This means, depending on the coding you choose, the bounds mentioned in the second variable above differ and the <em>response</em> variable will have a different form.</p>
+<p>Most often, an experimenter wants to find out whether an experimental manipulation influences the model parameters. As there exists psychological interpretations for each diffusion model parameter, the experimenter can draw conclusions from differing parameters. Therefore, usually an own diffusion model is being computed for each experimental group to enable a comparison of the parameters between the groups. This can be manipulation between different subjects, like an experimental group and a control group (so called between-subject manipulations). However, this can also be manipulations within the same subject by presenting stimuli from different experimental groups (so called within-subject manipulations). Depending on the experimental design, one would typically also provide the number of conditions and the condition associated with each trial as a vector. Then, one model for each condition will be computed. This means that the parameters also have to be defined for each condition.</p>
+<p>In a hierarchical setting, the data block would also specify the number of participants and the participant associated with each trial as a vector. It is also possible to hand over a precision value in the data block.</p>
+</section>
+<section id="the-parameters-block" class="level3">
+<h3 class="anchored" data-anchor-id="the-parameters-block">The parameters block</h3>
+<p>The model arguments of the <code>wiener_lpdf()</code> function that are not fixed to a certain value are defned as parameters in the parameters block. In this block, it is also possible to insert restrictions on the parameters. Note that the MCMC algorithm iteratively searches for the next parameter set. If the suggested sample falls outside the internally defined parameter ranges, the program will throw an error, which causes the algorithm to restart the current iteration. Since this slows down the sampling process, it is advisable to include the parameter ranges in the defnition of the parameters in the parameters block to improve the sampling process (see table below for the parameter ranges). In addition, the parameter space is further constrained by the following conditions:</p>
+<ol type="1">
+<li>The non-decision time <span class="math inline">\(t_0\)</span> has to be smaller or equal to the observed reaction time: <span class="math inline">\(t0 \leq y\)</span>.</li>
+<li>The varying relative starting point <span class="math inline">\(w\)</span> has to be in the interval (0,1) and thus,</li>
+</ol>
+<p><span class="math display">\[
+\begin{aligned}
+&amp;w + \frac{s_w}{2} &lt; 1 \text{, and} \\
+&amp;0 &lt; w-\frac{s_w}{2}
+\end{aligned}
+\]</span></p>
+<table class="caption-top table">
+<thead>
+<tr class="header">
+<th style="text-align: left;"><em>Parameter</em></th>
+<th style="text-align: left;"><em>Range</em></th>
+<th style="text-align: left;"></th>
+<th style="text-align: left;"><em>Parameter</em></th>
+<th style="text-align: left;"><em>Range</em></th>
+</tr>
+</thead>
+<tbody>
+<tr class="odd">
+<td style="text-align: left;"><span class="math inline">\(a\)</span></td>
+<td style="text-align: left;">(0, <span class="math inline">\(\infty\)</span>)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(y\)</span></td>
+<td style="text-align: left;">(0, <span class="math inline">\(\infty\)</span>)</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><span class="math inline">\(v\)</span></td>
+<td style="text-align: left;">(-<span class="math inline">\(\infty\)</span>, <span class="math inline">\(\infty\)</span>)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(s_v\)</span></td>
+<td style="text-align: left;">[0, <span class="math inline">\(\infty\)</span>)</td>
+</tr>
+<tr class="odd">
+<td style="text-align: left;"><span class="math inline">\(w\)</span></td>
+<td style="text-align: left;">(0,1)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(s_w\)</span></td>
+<td style="text-align: left;">[0,min(2<em>w, 2</em>(1-w)))</td>
+</tr>
+<tr class="even">
+<td style="text-align: left;"><span class="math inline">\(t_0\)</span></td>
+<td style="text-align: left;">[0,<span class="math inline">\(\infty\)</span>)</td>
+<td style="text-align: left;"></td>
+<td style="text-align: left;"><span class="math inline">\(s_{t0}\)</span></td>
+<td style="text-align: left;">[0,<span class="math inline">\(\infty\)</span>)</td>
+</tr>
+</tbody>
+</table>
+</section>
+<section id="the-model-block" class="level3">
+<h3 class="anchored" data-anchor-id="the-model-block">The model block</h3>
+<p>In the model block, the priors and likelihood are defined for the upper and the lower response boundary. Different kinds of priors can be specifed here. Generally, the regularization induced by mildly informative priors can help both statistically and computationally.</p>
+<p>In the second part of the model block, the data generating distribution is applied to all responses. The drift rate <span class="math inline">\(v\)</span> and relative starting point <span class="math inline">\(w\)</span> have to be mirrored for responses at the lower boundary.</p>
+<p>For more details regarding the application of the diffusion model in Stan, see <span class="citation" data-cites="Henrich2024">Henrich et al. (<a href="#ref-Henrich2024" role="doc-biblioref">2024</a>)</span>.</p>
+</section>
+</section>
+<section id="wiener-truncation" class="level2">
+<h2 class="anchored" data-anchor-id="wiener-truncation">Truncated and censored data</h2>
+<p>Truncation and censoring frequently occur in psychological data collection. For reaction time data, truncated and censored data regularly arise in psychological studies as a consequence of using response windows or deadlines. These are sometimes introduced in the analysis of data to exclude reaction times that appear too short or too long, but they are also sometimes already built into the study procedures to push participants to respond within a specifc temporal window.</p>
+<p>Depending on the implementation of the response window, two different types of data arise: <em>truncated</em> data or <em>censored</em> data. Since the effects of truncation or censoring on summary statistics such as mean, median, standard deviation, and skewness is regularly too large to ignore <span class="citation" data-cites="Ulrich1994">(<a href="#ref-Ulrich1994" role="doc-biblioref">Ulrich and Miller 1994</a>)</span>, data analysts are well advised to account for these effects.</p>
+<p>As described in the <a href="../stan-users-guide/truncation-censoring.html">Truncated or Censored Data</a> chapter, the cumulative distribution function (<span class="math inline">\(F\)</span>) and its complement (<span class="math inline">\(\text{CCDF}\)</span>) are needed to model truncated and censored data.</p>
+<p>As explained above, <span class="math inline">\(p\)</span> is defined <em>defectively</em>, meaning that only the sum of <span class="math inline">\(p\)</span>s for both response alternatives integrates to 1. For the same reason, <span class="math inline">\(F\)</span> and <span class="math inline">\(\text{CCDF}\)</span> are also implemented defectively. Analogously, only the sum of the <span class="math inline">\(F\)</span>s and <span class="math inline">\(\text{CCDF}\)</span>s for both response alternatives asymptotes above at 1.</p>
+<p>In the case of the diffusion model, <span class="math inline">\(F\)</span> asymptotes above at the probability <span class="math inline">\(PROB\)</span> to hit the corresponding response boundary: (for simplicity, we omit the inter-trial variabilities in the following)</p>
+<p><span class="math display">\[
+\begin{aligned}
+F_1(\infty\mid a,w,v) &amp;= \text{PROB}(a,w,v) \text{ and} \\
+F_0(\infty\mid a,w,v) &amp;= F_1(\infty\mid a,1-w,-v) = \text{PROB}(a,1-w,-v)
+\end{aligned}
+\]</span></p>
+<section id="modeling-truncated-data-with-the-diffusion-model" class="level3">
+<h3 class="anchored" data-anchor-id="modeling-truncated-data-with-the-diffusion-model">Modeling truncated data with the diffusion model</h3>
+<p>Data are called <em>truncated</em> when there is no information available for analysis from trials with values larger (or smaller) than a right (or left) reaction-time bound. In reaction time experiments, reaction time data are truncated if trials with reaction times outside the response window are excluded from the analysis. Not even a count of those omitted trials is kept.</p>
+<p>Let <span class="math inline">\(L\)</span> denote the left reaction-time bound and <span class="math inline">\(U\)</span> denote the right reaction-time bound of a response window.</p>
+<p>Then, the density of truncated data for both response boundaries 0 and 1, here denoted as <span class="math inline">\(\text{resp}\in\{0,1\}\)</span>, can be formulated as follows:</p>
+<p><span class="math display">\[
+\begin{aligned}
+&amp;p_{\text{resp}}(y \mid L&lt;X\leq U, a, w, v) = \\ &amp;\frac{p_{\text{resp}}(y \mid a, w, v)\cdot \mathbb{I}_{\{L&lt;y\leq U\}}}
+{\bigl(F_0(U \mid a, w, v)+F_1(U \mid a, w, v)\bigr) -
+\bigl(F_0(L\mid a, w, v)+F_1(L\mid a, w, v)\bigr)}
+\end{aligned}
+\]</span></p>
+<p>The density of left truncated data can be formulated as follows. <span class="math display">\[
+\begin{aligned}
+p_{\text{resp}}(y \mid L&lt;X, a, w, v) = \frac{p_{\text{resp}}(y \mid a, w, v)\cdot \mathbb{I}_{\{L&lt;y\}}}
+{1-\bigl(F_0(L \mid a, w, v)+F_1(L \mid a, w, v)\bigr)},
+\end{aligned}
+\]</span></p>
+<p>The density of right truncated data can be formulated as follows.</p>
+<p><span class="math display">\[
+\begin{aligned}
+p_{\text{resp}}(y \mid X\leq U, a, w, v) = \frac{p_{\text{resp}}(y \mid a, w, v)\cdot \mathbb{I}_{\{y\leq U\}}}{F_0(U \mid a, w, v)+F_1(U \mid a, w, v)}
+\end{aligned}
+\]</span></p>
+<p>As the functions are implemented defectively, a truncated diffusion model cannot be calculated with the truncation functor <span class="math inline">\(T[,]\)</span> as it would usually be done in Stan. This means the function call: <code>y ~ wiener(...)T[L,U]</code> does not work the way it is supposed to. When the truncation functor is called in Stan, Stan searches for a CDF implementation internally. In the case of the diffusion model, Stan would find the CDF, but is not aware of its defective implementation and calculates the computations as if it were a non-defective CDF. This causes misleading and incorrect results.</p>
+<p>To implement the truncated model, write out the function shown above on the log-scale with <code>left_bound = L</code> and <code>right_bound = U</code>, where <code>wiener_lcdf_unnorm()</code> calls the logarithmized CDF of the diffusion model at the response-1-boundary:</p>
+<div class="sourceCode" id="cb2"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> log_denom = log_diff_exp(</span>
+<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a>    log_sum_exp(</span>
+<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),</span>
+<span id="cb2-5"><a href="#cb2-5" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb2-6"><a href="#cb2-6" aria-hidden="true" tabindex="-1"></a>                                       sv, sw, st)),</span>
+<span id="cb2-7"><a href="#cb2-7" aria-hidden="true" tabindex="-1"></a>    log_sum_exp(</span>
+<span id="cb2-8"><a href="#cb2-8" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st),</span>
+<span id="cb2-9"><a href="#cb2-9" aria-hidden="true" tabindex="-1"></a>      wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb2-10"><a href="#cb2-10" aria-hidden="true" tabindex="-1"></a>                                      sv, sw, st)));</span>
+<span id="cb2-11"><a href="#cb2-11" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood</span></span>
+<span id="cb2-12"><a href="#cb2-12" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb2-13"><a href="#cb2-13" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb2-14"><a href="#cb2-14" aria-hidden="true" tabindex="-1"></a>      <span class="co">// response -1 boundary</span></span>
+<span id="cb2-15"><a href="#cb2-15" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_lpdf (rt[i] | a, t0, w, v, sv, sw, st);</span>
+<span id="cb2-16"><a href="#cb2-16" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> { </span>
+<span id="cb2-17"><a href="#cb2-17" aria-hidden="true" tabindex="-1"></a>      <span class="co">// response -0 boundary ( mirror v and w)</span></span>
+<span id="cb2-18"><a href="#cb2-18" aria-hidden="true" tabindex="-1"></a>      <span class="kw">target +=</span> wiener_lpdf (rt[i] | a, t0, one_minus_w, neg_v,</span>
+<span id="cb2-19"><a href="#cb2-19" aria-hidden="true" tabindex="-1"></a>                                     sv, sw, st);</span>
+<span id="cb2-20"><a href="#cb2-20" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb2-21"><a href="#cb2-21" aria-hidden="true" tabindex="-1"></a>  } <span class="co">// end for</span></span>
+<span id="cb2-22"><a href="#cb2-22" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> -N * log_denom;</span>
+<span id="cb2-23"><a href="#cb2-23" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For details of how to call a truncated model within the parallelization routine of <code>reduce_sum</code> or with truncation to only on side, see <span class="citation" data-cites="Henrich2026">Henrich and Klauer (<a href="#ref-Henrich2026" role="doc-biblioref">2026</a>)</span>.</p>
+</section>
+<section id="modeling-censored-data-with-the-diffusion-model" class="level3">
+<h3 class="anchored" data-anchor-id="modeling-censored-data-with-the-diffusion-model">Modeling censored data with the diffusion model</h3>
+<p>Data are <em>censored</em> when observations that are above or below a right or left boundary value are reported as occurrences of the event <span class="math inline">\((y &gt;
+U)\)</span>, for <span class="math inline">\(U\)</span> the right bound, or as occurrences of the event <span class="math inline">\((y \leq
+L)\)</span>, for <span class="math inline">\(L\)</span> the left bound, respectively. Like for truncated data, the range of the possible values is restricted, but the number of observations that fall outside the boundaries is kept, whereas in truncation, no count would be kept.</p>
+<p>For the censored model, we distinguish two cases. In the first case, the responses of the censored trials are known, but the reaction times are not known. In the second case, neither the responses nor the reaction times of the censored trials are known. Note that the second case differs from a truncated model in the fact that the number of censored trials is still known. Consider first the case where the response is known even for censored data.</p>
+<p>To model such data in Stan, the left and right reaction time bounds, <code>left_bound</code> and <code>right_bound</code>, respectively, are handed over in the data block, as well as a vector <code>censored</code> that tracks whether a trial is censored (= 1) or not (= 0), and counts of trials censored at the left reaction time bound and counts of trials censored at the right reaction time bound for each response in {0,1}. There are four such count variables: <code>N_cens_left_0</code>, <code>N_cens_left_1</code>, <code>N_cens_right_0</code>, <code>N_cens_right_1</code>:</p>
+<div class="sourceCode" id="cb3"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> {</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) {</span>
+<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (censored[i] == <span class="dv">0</span>) {</span>
+<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a>        y[i] ~ wiener(a, t0, w, v, sv, sw, st0);</span>
+<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (resp[i] == <span class="dv">0</span>) {</span>
+<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a>        y[i] ~ wiener(a, t0, one_minus_w, neg_v, sv, sw, st0);</span>
+<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb3-9"><a href="#cb3-9" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb3-10"><a href="#cb3-10" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb3-11"><a href="#cb3-11" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-12"><a href="#cb3-12" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood (response = 0)</span></span>
+<span id="cb3-13"><a href="#cb3-13" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_left_0 </span>
+<span id="cb3-14"><a href="#cb3-14" aria-hidden="true" tabindex="-1"></a>         * wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb3-15"><a href="#cb3-15" aria-hidden="true" tabindex="-1"></a>                                           sv, sw, st0);</span>
+<span id="cb3-16"><a href="#cb3-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-17"><a href="#cb3-17" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_right_0 </span>
+<span id="cb3-18"><a href="#cb3-18" aria-hidden="true" tabindex="-1"></a>         * wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w, neg_v,</span>
+<span id="cb3-19"><a href="#cb3-19" aria-hidden="true" tabindex="-1"></a>                                             sv, sw, st0);</span>
+<span id="cb3-20"><a href="#cb3-20" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-21"><a href="#cb3-21" aria-hidden="true" tabindex="-1"></a>  <span class="co">// likelihood (response = 1)</span></span>
+<span id="cb3-22"><a href="#cb3-22" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_left_1 </span>
+<span id="cb3-23"><a href="#cb3-23" aria-hidden="true" tabindex="-1"></a>         * wiener_lcdf_unnorm(left_bound | a, t0, w, v, sv, sw, st0);</span>
+<span id="cb3-24"><a href="#cb3-24" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb3-25"><a href="#cb3-25" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> N_cens_right_1 </span>
+<span id="cb3-26"><a href="#cb3-26" aria-hidden="true" tabindex="-1"></a>         * wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st0);</span>
+<span id="cb3-27"><a href="#cb3-27" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When data are censored at only one side, meaning that the reaction time constraint only exists for one of the two boundaries, omit the lines for the other side in the code. A both sided reaction time window would be, for example, when only reaction times are accepted that occur between 0.2 and 0.8 seconds. A one sided reaction time constraint would be, for example, when all reaction times below 0.8 seconds are accepted.</p>
+<p>When data consist of many conditions (as explained in the beginning), it is sometimes more convenient to loop over all trials instead of using count variables as described above, using the following notation and code. A vector containing the information whether a trial is censored or not, here <code>censored</code>, needs to be handed over in the data block. This vector splits the data into three bins: all trials <span class="math inline">\(i\)</span> with<code>censored[i]=0</code> are censored below the left reaction time bound, all trials <span class="math inline">\(i\)</span> with <code>censored[i]=1</code> fall between the reaction time bounds, and all trials <span class="math inline">\(i\)</span> with <code>censored[i]=2</code> are censored above the right reaction time bound. For non-censored trials, the log-PDF is computed, for left censored trials, the log-CDF is computed, and for right censored trials, the log-CCDF is computed:</p>
+<div class="sourceCode" id="cb4"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a><span class="kw">model</span> { </span>
+<span id="cb4-2"><a href="#cb4-2" aria-hidden="true" tabindex="-1"></a>  <span class="cf">for</span> (i <span class="cf">in</span> <span class="dv">1</span>:N) { </span>
+<span id="cb4-3"><a href="#cb4-3" aria-hidden="true" tabindex="-1"></a>    <span class="co">// right censored at right_bound</span></span>
+<span id="cb4-4"><a href="#cb4-4" aria-hidden="true" tabindex="-1"></a>    <span class="cf">if</span> (resp [i] == <span class="dv">1</span>) { </span>
+<span id="cb4-5"><a href="#cb4-5" aria-hidden="true" tabindex="-1"></a>      <span class="co">// upper response boundary</span></span>
+<span id="cb4-6"><a href="#cb4-6" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (censored[i] == <span class="dv">0</span>) {</span>
+<span id="cb4-7"><a href="#cb4-7" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lcdf_unnorm(left_bound | a, t0, w, v, </span>
+<span id="cb4-8"><a href="#cb4-8" aria-hidden="true" tabindex="-1"></a>                                                  sv, sw, st0);</span>
+<span id="cb4-9"><a href="#cb4-9" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">1</span>) {</span>
+<span id="cb4-10"><a href="#cb4-10" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lpdf(y[i] | a, t0, w, v, sv, sw, st0);</span>
+<span id="cb4-11"><a href="#cb4-11" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">2</span>) {</span>
+<span id="cb4-12"><a href="#cb4-12" aria-hidden="true" tabindex="-1"></a>         <span class="kw">target +=</span> wiener_lccdf_unnorm(right_bound | a, t0, w, v,</span>
+<span id="cb4-13"><a href="#cb4-13" aria-hidden="true" tabindex="-1"></a>                                                     sv, sw, st0);</span>
+<span id="cb4-14"><a href="#cb4-14" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb4-15"><a href="#cb4-15" aria-hidden="true" tabindex="-1"></a>    } <span class="cf">else</span> { </span>
+<span id="cb4-16"><a href="#cb4-16" aria-hidden="true" tabindex="-1"></a>      <span class="co">// lower response boundary (mirror drift and // starting point!)</span></span>
+<span id="cb4-17"><a href="#cb4-17" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (censored[i] == <span class="dv">0</span>) {</span>
+<span id="cb4-18"><a href="#cb4-18" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lcdf_unnorm(left_bound | a, t0, one_minus_w,</span>
+<span id="cb4-19"><a href="#cb4-19" aria-hidden="true" tabindex="-1"></a>                                                  neg_v, sv, sw, st0);</span>
+<span id="cb4-20"><a href="#cb4-20" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">1</span>) {</span>
+<span id="cb4-21"><a href="#cb4-21" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lpdf(y[i] | a, t0, one_minus_w, neg_v,</span>
+<span id="cb4-22"><a href="#cb4-22" aria-hidden="true" tabindex="-1"></a>                                     sv, sw, st0);</span>
+<span id="cb4-23"><a href="#cb4-23" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> <span class="cf">if</span> (censored[i] == <span class="dv">2</span>) {</span>
+<span id="cb4-24"><a href="#cb4-24" aria-hidden="true" tabindex="-1"></a>        <span class="kw">target +=</span> wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,</span>
+<span id="cb4-25"><a href="#cb4-25" aria-hidden="true" tabindex="-1"></a>                                                    neg_v, sv, sw, st0);</span>
+<span id="cb4-26"><a href="#cb4-26" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb4-27"><a href="#cb4-27" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb4-28"><a href="#cb4-28" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb4-29"><a href="#cb4-29" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>When the data are censored on only one side, omit the case that is not needed.</p>
+<p>Note that this block can be inserted in the defnition of the parallelization function, <code>partial_sum_wiener()</code>, as defined below.</p>
+<p>Sometimes also the response is missing (i.e., it is known that the reaction time in a trial fell outside the response window, but which response was given is unknown). One method that has been used to model such data has involved inferring the numbers of missing responses of either kind from the observed relative frequencies of the two responses. This approach has the problem that quite specifc assumptions on the missing data have to be made (namely, that the proportions of the two kinds of responses are the same for responses within and outside the response window).</p>
+<p>The following is a more principled approach that uses the cumulative distribution functions and their complements to provide the data-generating distribution of censored data. As before, let <span class="math inline">\(L\)</span> be the left reaction time bound, and <span class="math inline">\(U\)</span> the right reaction time bound, and consider decision times without inter-trial variabilities for the sake of simplicity. It follows that the likelihood contribution <span class="math inline">\(\textit{lik}_l\)</span> for a left-censored data point is given by</p>
+<p><span class="math display">\[
+\begin{aligned}
+\textit{lik}_l(a,w,v) = F_0(L\mid a,w,v) + F_1(L\mid a,w,v),
+\end{aligned}
+\]</span></p>
+<p>whereas the likelihood contribution <span class="math inline">\(lik_r\)</span> due to a right-censored data point is given by</p>
+<p><span class="math display">\[
+\begin{aligned}
+\textit{lik}_r(a,w,v) = \text{CCDF}_0(U\mid a,w,v) + \text{CCDF}_1(U\mid a,w,v).
+\end{aligned}
+\]</span></p>
+<p>See the following code for an example of Stan code implementing this second case of censoring. This model call deals with the problem of unknown responses by computing the probability of choosing the response-1 or response-0 boundary outside the response window. Here, the CDF and/or the CCDF are required, depending upon whether there is only left-censoring, right-censoring, or censoring both to the left and to the right. The following code shows the functions block for a model that is right-censored using the function <code>partial_sum_wiener()</code> to parallelize the execution of a single Stan chain across multiple cores:</p>
+<div class="sourceCode" id="cb5"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="kw">functions</span> {</span>
+<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a>  <span class="dt">real</span> partial_sum_wiener(<span class="dt">array</span>[] <span class="dt">real</span> rt_slice, <span class="dt">int</span> start,</span>
+<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">int</span> end, <span class="dt">real</span> a, <span class="dt">real</span> t0, <span class="dt">real</span> w,</span>
+<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">real</span> v, <span class="dt">real</span> sv, <span class="dt">real</span> sw, <span class="dt">real</span> st,</span>
+<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">array</span>[] <span class="dt">int</span> resp, <span class="dt">real</span> right_bound,</span>
+<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a>                          <span class="dt">array</span>[] <span class="dt">int</span> censored) {</span>
+<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>    <span class="dt">real</span> ans = <span class="dv">0</span>;</span>
+<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a>    <span class="cf">for</span> (i <span class="cf">in</span> start:end) {</span>
+<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a>      <span class="cf">if</span> (censored[i] == <span class="dv">1</span>) {</span>
+<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>        <span class="co">// not censored</span></span>
+<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a>        <span class="cf">if</span> (resp[i] == <span class="dv">1</span>) {</span>
+<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a>          <span class="co">// upper boundary</span></span>
+<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a>          ans += wiener_lpdf(rt_slice[i+<span class="dv">1</span>- start ] | a, t0, w, v,</span>
+<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a>                                                     sv, sw, st);</span>
+<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>        } <span class="cf">else</span> {</span>
+<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a>          <span class="co">// lower boundary(mirror v and w)</span></span>
+<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>          ans += wiener_lpdf(rt_slice[i+<span class="dv">1</span>- start ] | a, t0, one_minus_w,</span>
+<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a>                                                     neg_v, sv, sw, st);</span>
+<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a>        }</span>
+<span id="cb5-20"><a href="#cb5-20" aria-hidden="true" tabindex="-1"></a>      } <span class="cf">else</span> { </span>
+<span id="cb5-21"><a href="#cb5-21" aria-hidden="true" tabindex="-1"></a>        <span class="co">// censored</span></span>
+<span id="cb5-22"><a href="#cb5-22" aria-hidden="true" tabindex="-1"></a>        ans += log_sum_exp (</span>
+<span id="cb5-23"><a href="#cb5-23" aria-hidden="true" tabindex="-1"></a>          wiener_lccdf_unnorm(right_bound | a, t0, w, v, sv, sw, st),</span>
+<span id="cb5-24"><a href="#cb5-24" aria-hidden="true" tabindex="-1"></a>          wiener_lccdf_unnorm(right_bound | a, t0, one_minus_w,</span>
+<span id="cb5-25"><a href="#cb5-25" aria-hidden="true" tabindex="-1"></a>                                            neg_v, sv, sw, st);</span>
+<span id="cb5-26"><a href="#cb5-26" aria-hidden="true" tabindex="-1"></a>      }</span>
+<span id="cb5-27"><a href="#cb5-27" aria-hidden="true" tabindex="-1"></a>    }</span>
+<span id="cb5-28"><a href="#cb5-28" aria-hidden="true" tabindex="-1"></a>    <span class="cf">return</span> ans;</span>
+<span id="cb5-29"><a href="#cb5-29" aria-hidden="true" tabindex="-1"></a>  }</span>
+<span id="cb5-30"><a href="#cb5-30" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>Combine this block with the model block in the example above by using the function <code>reduce_sum()</code>.</p>
+<div class="sourceCode" id="cb6"><pre class="sourceCode stan code-with-copy"><code class="sourceCode stan"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a>  <span class="kw">target +=</span> reduce_sum(partial_sum_wiener, rt, <span class="dv">1</span>,</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a>    a, t0, w, v, sv, sw, st, resp, right_bound, censored);</span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>}</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<p>For more details, see <span class="citation" data-cites="Henrich2026">Henrich and Klauer (<a href="#ref-Henrich2026" role="doc-biblioref">2026</a>)</span>.</p>
+
+
+
+</section>
+</section>
+</section>
+
+<a onclick="window.scrollTo(0, 0); return false;" role="button" id="quarto-back-to-top"><i class="bi bi-arrow-up"></i> Back to top</a><div id="quarto-appendix" class="default"><section class="quarto-appendix-contents" role="doc-bibliography" id="quarto-bibliography"><h2 class="anchored quarto-appendix-heading">References</h2><div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list">
+<div id="ref-Henrich2024" class="csl-entry" role="listitem">
+Henrich, Franziska, Raphael Hartmann, Valentin Pratz, Andreas Voss, and Karl Christoph Klauer. 2024. <span>“The Seven-Parameter Diffusion Model: An Implementation in Stan for Bayesian Analyses.”</span> <em>Behavior Research Methods</em> 56 (4): 3102–16. <a href="https://doi.org/10.3758/s13428-023-02179-1">https://doi.org/10.3758/s13428-023-02179-1</a>.
+</div>
+<div id="ref-Henrich2026" class="csl-entry" role="listitem">
+Henrich, Franziska, and Karl Christoph Klauer. 2026. <span>“Modeling Truncated and Censored Data with the Diffusion Model in <span>S</span>tan.”</span> <em>Behavior Research Methods</em> 58 (42). <a href="https://doi.org/10.3758/s13428-025-02822-z">https://doi.org/10.3758/s13428-025-02822-z</a>.
+</div>
+<div id="ref-nicenboim2025introduction" class="csl-entry" role="listitem">
+Nicenboim, Bruno, Daniel J Schad, and Shravan Vasishth. 2025. <em>Introduction to <span>B</span>ayesian Data Analysis for Cognitive Science</em>. CRC Press.
+</div>
+<div id="ref-Ratcliff1978" class="csl-entry" role="listitem">
+Ratcliff, Roger. 1978. <span>“A Theory of Memory Retrieval.”</span> <em>Psychological Review</em> 85 (2): 59–108.
+</div>
+<div id="ref-Ratcliff1998" class="csl-entry" role="listitem">
+Ratcliff, Roger, and Jeffrey N. Rouder. 1998. <span>“Modelling Response Times for Two-Choice Decisions.”</span> <em>Psychological Science</em> 9 (5): 347–56.
+</div>
+<div id="ref-Ulrich1994" class="csl-entry" role="listitem">
+Ulrich, Rolf, and Jeff Miller. 1994. <span>“Effects of Truncation on Reaction Time Analysis.”</span> <em>Journal of Experimental Psychology: General</em> 123 (1): 34–80.
+</div>
+<div id="ref-Wagenmakers2009" class="csl-entry" role="listitem">
+Wagenmakers, Eric-Jan. 2009. <span>“Methodological and Empirical Developments for the <span>R</span>atcliff Diffusion Model of Response Times and Accuracy.”</span> <em>European Journal of Cognitive Psychology</em> 21 (5): 641–71. <a href="https://doi.org/10.1080/09541440802205067">https://doi.org/10.1080/09541440802205067</a>.
+</div>
+</div></section></div></main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const disableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'prefetch';
+    }
+  }
+  const enableStylesheet = (stylesheets) => {
+    for (let i=0; i < stylesheets.length; i++) {
+      const stylesheet = stylesheets[i];
+      stylesheet.rel = 'stylesheet';
+    }
+  }
+  const manageTransitions = (selector, allowTransitions) => {
+    const els = window.document.querySelectorAll(selector);
+    for (let i=0; i < els.length; i++) {
+      const el = els[i];
+      if (allowTransitions) {
+        el.classList.remove('notransition');
+      } else {
+        el.classList.add('notransition');
+      }
+    }
+  }
+  const toggleGiscusIfUsed = (isAlternate, darkModeDefault) => {
+    const baseTheme = document.querySelector('#giscus-base-theme')?.value ?? 'light';
+    const alternateTheme = document.querySelector('#giscus-alt-theme')?.value ?? 'dark';
+    let newTheme = '';
+    if(darkModeDefault) {
+      newTheme = isAlternate ? baseTheme : alternateTheme;
+    } else {
+      newTheme = isAlternate ? alternateTheme : baseTheme;
+    }
+    const changeGiscusTheme = () => {
+      // From: https://github.com/giscus/giscus/issues/336
+      const sendMessage = (message) => {
+        const iframe = document.querySelector('iframe.giscus-frame');
+        if (!iframe) return;
+        iframe.contentWindow.postMessage({ giscus: message }, 'https://giscus.app');
+      }
+      sendMessage({
+        setConfig: {
+          theme: newTheme
+        }
+      });
+    }
+    const isGiscussLoaded = window.document.querySelector('iframe.giscus-frame') !== null;
+    if (isGiscussLoaded) {
+      changeGiscusTheme();
+    }
+  }
+  const toggleColorMode = (alternate) => {
+    // Switch the stylesheets
+    const alternateStylesheets = window.document.querySelectorAll('link.quarto-color-scheme.quarto-color-alternate');
+    manageTransitions('#quarto-margin-sidebar .nav-link', false);
+    if (alternate) {
+      enableStylesheet(alternateStylesheets);
+      for (const sheetNode of alternateStylesheets) {
+        if (sheetNode.id === "quarto-bootstrap") {
+          toggleBodyColorMode(sheetNode);
+        }
+      }
+    } else {
+      disableStylesheet(alternateStylesheets);
+      toggleBodyColorPrimary();
+    }
+    manageTransitions('#quarto-margin-sidebar .nav-link', true);
+    // Switch the toggles
+    const toggles = window.document.querySelectorAll('.quarto-color-scheme-toggle');
+    for (let i=0; i < toggles.length; i++) {
+      const toggle = toggles[i];
+      if (toggle) {
+        if (alternate) {
+          toggle.classList.add("alternate");     
+        } else {
+          toggle.classList.remove("alternate");
+        }
+      }
+    }
+    // Hack to workaround the fact that safari doesn't
+    // properly recolor the scrollbar when toggling (#1455)
+    if (navigator.userAgent.indexOf('Safari') > 0 && navigator.userAgent.indexOf('Chrome') == -1) {
+      manageTransitions("body", false);
+      window.scrollTo(0, 1);
+      setTimeout(() => {
+        window.scrollTo(0, 0);
+        manageTransitions("body", true);
+      }, 40);  
+    }
+  }
+  const isFileUrl = () => { 
+    return window.location.protocol === 'file:';
+  }
+  const hasAlternateSentinel = () => {  
+    let styleSentinel = getColorSchemeSentinel();
+    if (styleSentinel !== null) {
+      return styleSentinel === "alternate";
+    } else {
+      return false;
+    }
+  }
+  const setStyleSentinel = (alternate) => {
+    const value = alternate ? "alternate" : "default";
+    if (!isFileUrl()) {
+      window.localStorage.setItem("quarto-color-scheme", value);
+    } else {
+      localAlternateSentinel = value;
+    }
+  }
+  const getColorSchemeSentinel = () => {
+    if (!isFileUrl()) {
+      const storageValue = window.localStorage.getItem("quarto-color-scheme");
+      return storageValue != null ? storageValue : localAlternateSentinel;
+    } else {
+      return localAlternateSentinel;
+    }
+  }
+  const darkModeDefault = false;
+  let localAlternateSentinel = darkModeDefault ? 'alternate' : 'default';
+  // Dark / light mode switch
+  window.quartoToggleColorScheme = () => {
+    // Read the current dark / light value 
+    let toAlternate = !hasAlternateSentinel();
+    toggleColorMode(toAlternate);
+    setStyleSentinel(toAlternate);
+    toggleGiscusIfUsed(toAlternate, darkModeDefault);
+  };
+  // Ensure there is a toggle, if there isn't float one in the top right
+  if (window.document.querySelector('.quarto-color-scheme-toggle') === null) {
+    const a = window.document.createElement('a');
+    a.classList.add('top-right');
+    a.classList.add('quarto-color-scheme-toggle');
+    a.href = "";
+    a.onclick = function() { try { window.quartoToggleColorScheme(); } catch {} return false; };
+    const i = window.document.createElement("i");
+    i.classList.add('bi');
+    a.appendChild(i);
+    window.document.body.appendChild(a);
+  }
+  // Switch to dark mode if need be
+  if (hasAlternateSentinel()) {
+    toggleColorMode(true);
+  } else {
+    toggleColorMode(false);
+  }
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const onCopySuccess = function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  }
+  const getTextToCopy = function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
+    text: getTextToCopy
+  });
+  clipboard.on('success', onCopySuccess);
+  if (window.document.getElementById('quarto-embedded-source-code-modal')) {
+    const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
+      text: getTextToCopy,
+      container: window.document.getElementById('quarto-embedded-source-code-modal')
+    });
+    clipboardModal.on('success', onCopySuccess);
+  }
+    var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
+    var mailtoRegex = new RegExp(/^mailto:/);
+      var filterRegex = new RegExp('/' + window.location.host + '/');
+    var isInternal = (href) => {
+        return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
+    }
+    // Inspect non-navigation links and adorn them if external
+ 	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
+    for (var i=0; i<links.length; i++) {
+      const link = links[i];
+      if (!isInternal(link.href)) {
+        // undo the damage that might have been done by quarto-nav.js in the case of
+        // links that we want to consider external
+        if (link.dataset.originalHref !== undefined) {
+          link.href = link.dataset.originalHref;
+        }
+      }
+    }
+  function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
+    const config = {
+      allowHTML: true,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start',
+    };
+    if (contentFn) {
+      config.content = contentFn;
+    }
+    if (onTriggerFn) {
+      config.onTrigger = onTriggerFn;
+    }
+    if (onUntriggerFn) {
+      config.onUntrigger = onUntriggerFn;
+    }
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      if (note) {
+        return note.innerHTML;
+      } else {
+        return "";
+      }
+    });
+  }
+  const xrefs = window.document.querySelectorAll('a.quarto-xref');
+  const processXRef = (id, note) => {
+    // Strip column container classes
+    const stripColumnClz = (el) => {
+      el.classList.remove("page-full", "page-columns");
+      if (el.children) {
+        for (const child of el.children) {
+          stripColumnClz(child);
+        }
+      }
+    }
+    stripColumnClz(note)
+    if (id === null || id.startsWith('sec-')) {
+      // Special case sections, only their first couple elements
+      const container = document.createElement("div");
+      if (note.children && note.children.length > 2) {
+        container.appendChild(note.children[0].cloneNode(true));
+        for (let i = 1; i < note.children.length; i++) {
+          const child = note.children[i];
+          if (child.tagName === "P" && child.innerText === "") {
+            continue;
+          } else {
+            container.appendChild(child.cloneNode(true));
+            break;
+          }
+        }
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(container);
+        }
+        return container.innerHTML
+      } else {
+        if (window.Quarto?.typesetMath) {
+          window.Quarto.typesetMath(note);
+        }
+        return note.innerHTML;
+      }
+    } else {
+      // Remove any anchor links if they are present
+      const anchorLink = note.querySelector('a.anchorjs-link');
+      if (anchorLink) {
+        anchorLink.remove();
+      }
+      if (window.Quarto?.typesetMath) {
+        window.Quarto.typesetMath(note);
+      }
+      if (note.classList.contains("callout")) {
+        return note.outerHTML;
+      } else {
+        return note.innerHTML;
+      }
+    }
+  }
+  for (var i=0; i<xrefs.length; i++) {
+    const xref = xrefs[i];
+    tippyHover(xref, undefined, function(instance) {
+      instance.disable();
+      let url = xref.getAttribute('href');
+      let hash = undefined; 
+      if (url.startsWith('#')) {
+        hash = url;
+      } else {
+        try { hash = new URL(url).hash; } catch {}
+      }
+      if (hash) {
+        const id = hash.replace(/^#\/?/, "");
+        const note = window.document.getElementById(id);
+        if (note !== null) {
+          try {
+            const html = processXRef(id, note.cloneNode(true));
+            instance.setContent(html);
+          } finally {
+            instance.enable();
+            instance.show();
+          }
+        } else {
+          // See if we can fetch this
+          fetch(url.split('#')[0])
+          .then(res => res.text())
+          .then(html => {
+            const parser = new DOMParser();
+            const htmlDoc = parser.parseFromString(html, "text/html");
+            const note = htmlDoc.getElementById(id);
+            if (note !== null) {
+              const html = processXRef(id, note);
+              instance.setContent(html);
+            } 
+          }).finally(() => {
+            instance.enable();
+            instance.show();
+          });
+        }
+      } else {
+        // See if we can fetch a full url (with no hash to target)
+        // This is a special case and we should probably do some content thinning / targeting
+        fetch(url)
+        .then(res => res.text())
+        .then(html => {
+          const parser = new DOMParser();
+          const htmlDoc = parser.parseFromString(html, "text/html");
+          const note = htmlDoc.querySelector('main.content');
+          if (note !== null) {
+            // This should only happen for chapter cross references
+            // (since there is no id in the URL)
+            // remove the first header
+            if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
+              note.children[0].remove();
+            }
+            const html = processXRef(null, note);
+            instance.setContent(html);
+          } 
+        }).finally(() => {
+          instance.enable();
+          instance.show();
+        });
+      }
+    }, function(instance) {
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            div.style.left = 0;
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+        // Handle positioning of the toggle
+    window.addEventListener(
+      "resize",
+      throttle(() => {
+        elRect = undefined;
+        if (selectedAnnoteEl) {
+          selectCodeLines(selectedAnnoteEl);
+        }
+      }, 10)
+    );
+    function throttle(fn, ms) {
+    let throttle = false;
+    let timer;
+      return (...args) => {
+        if(!throttle) { // first call gets through
+            fn.apply(this, args);
+            throttle = true;
+        } else { // all the others get throttled
+            if(timer) clearTimeout(timer); // cancel #2
+            timer = setTimeout(() => {
+              fn.apply(this, args);
+              timer = throttle = false;
+            }, ms);
+        }
+      };
+    }
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+<nav class="page-navigation">
+  <div class="nav-page nav-page-previous">
+      <a href="../stan-users-guide/survival.html" class="pagination-link" aria-label="Survival Models">
+        <i class="bi bi-arrow-left-short"></i> <span class="nav-page-text">Survival Models</span>
+      </a>          
+  </div>
+  <div class="nav-page nav-page-next">
+      <a href="../stan-users-guide/floating-point.html" class="pagination-link" aria-label="Floating Point Arithmetic">
+        <span class="nav-page-text">Floating Point Arithmetic</span> <i class="bi bi-arrow-right-short"></i>
+      </a>
+  </div>
+</nav>
+</div> <!-- /content -->
+<footer class="footer"><div class="nav-footer"><div class="nav-footer-center"><div class="toc-actions d-sm-block d-md-none"><ul><li><a href="https://github.com/stan-dev/docs/edit/master/src/stan-users-guide/wiener_diffusion_model.qmd" class="toc-action"><i class="bi bi-github"></i>Edit this page</a></li><li><a href="https://github.com/stan-dev/docs/issues/new" class="toc-action"><i class="bi empty"></i>Report an issue</a></li></ul></div></div></div></footer><script>
+// Workaround for Safari not scrolling to URL fragment identifiers
+// on fresh page loads. Uses MutationObserver to wait for Quarto's
+// post-load DOM rendering (sidebar, TOC, MathJax) to settle before
+// scrolling. Scrolls as soon as DOM changes stop for 80ms.
+window.addEventListener('load', function () {
+  if (!window.location.hash) return;
+  var el = document.querySelector(window.location.hash);
+  if (!el) return;
+
+  var timer;
+  var observer = new MutationObserver(function () {
+    clearTimeout(timer);
+    timer = setTimeout(function () {
+      observer.disconnect();
+      el.scrollIntoView();
+    }, 80);
+  });
+
+  observer.observe(document.body, { childList: true, subtree: true });
+
+  timer = setTimeout(function () {
+    observer.disconnect();
+    el.scrollIntoView();
+  }, 80);
+});
+</script>
+
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/stan-users-guide/wiener_diffusion_model.pdf b/docs/stan-users-guide/wiener_diffusion_model.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..fdec7b5f6f658697b7ba4465430774cfafd01b6b
GIT binary patch
literal 168306
zcmagEW2`Vxvu1g0+qP}nwr$(CZQHhOzsI(1pZ+?1XXYkzlm4?)yM9+CtMV*T1rae?
zMmkm~(&PJ^cPJ)C0tNzmBP%E#9(plL8y8b2dNCVA7gG^aV|x=*dKptYa~BH&CJuI1
zK0YXC7bjCgTPTms8|}$Bl6J)2FX~J1EMc%<Y=!e+z*vAinCIF1LiAKQzV|<RGE5k|
z$(gCkMF}RB3`-Z(WN8|DH8cHRZ$)&>zsl&~cYVFxKji)&9Dl#wudh4$bb9?2(ZU6b
zgGZ-s3ets)>EtA$3<uAT_cc|it9;;*y-QuG1*M{NrQxUKHnun{)c(%r*-`s@G2(Z`
zwNRFN-?sbjyL5jYKd%i^l02&5OGrTkrVgP9X>8#wmY=pH$Utmf*Lp`5T*&9#2}2-8
zWQ_jGCKCHo^QktJ!o;C{nIz6r$v5r*WWA5+WX7Q)T5OB2>tqMy(a3U=0g{?PCs@%I
zNl?k}#C?vcdXc{2N0NsyRl(2(T0*E|mtmR6NN*#EljOuIza;-i`Mt<J)P`pW8q%e4
zMeNs?Un$^=qmxOn!l24ktW<w$v@F6Tg}nVlgVNnAi^gLUn0@BHJVNaVUV*q(p{k3K
zek2J$@qHgKp<q214tPTNA;Mds2^4r^(<hdVFR<64jb4A`mKxN-s>l@l_0QW3%;VZ0
z&lwPEf@|Y2Vp_ri(M@|GNb$OZYy8A3csK4~wmE?aWkS;6`wHzM$<73vCqg(qNCAf<
zABwL+c)UeY;VMY;$^0cf+Z`gDTjN>Gzz)Jk@P<C*2r`<Cr#Oi~G%f+n$a?91VM<pz
z>|G3~PO47mwjZ3SR3zg0{+Dh$=7!;=;|0T8B9-a$F4{Su_kTo}fHW{Lydj{t{~%Q2
zjSrOPj^+b{&%Hr8xS9X__l@xcMh3E0-?Sj3^TM8lB%i=vke;i$0T)jMT|Muo9||po
z`M4~&d`J!sTLnAjmzmE$yg^90JZ*VFwFZlg+Y2)Rd3TcaG`+uIF%O!A<3Zeg%S?u!
zOWHZG4m{bS1H@ac+hUonq{$*6gSVfV>_xa8-*X~om|{S<uq)kuH#>|k0-(h+>3QYK
znc>$zh}EENW$`ZO(u#?TD-&mP#JLHonlntw&+5#isW);>V<okUnKA1Srt1=PO;+n|
zFujpFDI*FvRrw09^DA0r^yZdjlXL2d<rKj(R?p`MiHjGRd|OXtro5FF|2j2;FG4!z
zKtBg9^u~}{6e<JTErtj6qNV5aG4SNrPDt4E-6ZH{l7#z>n_AWYm$n9Hmk(9c41BDj
zFq(@wH|<KkKO`H`2bfx4O`|8KpF@q$Ee=7v5vB&)e!wDN(I>;UErj*j2?%<`m-d?C
zNo6ZBs56+Bmy%r+f@-l17%K_u>vOF%Rb{_+onId{E<YS?ebVu2Xge@VlFZeXeNm1P
z9m~0B9&#kuP6L=r&_#l8Rg|1JwYNAp>A2!1v*K~J=QlKxy!w1tg;iD=Q^2ru%~n-&
zbj)HYGacj}aCR>s8L2i7Olp9ckEczWlQ%>=srLjzsJ~vv#6?-o+&%jd>ye4pY^Z99
z@CutC-A2u5OcA}=jsguQm6Sk_h{6jNn7L8#q`h*YC1dT6Gg_C6owXB~HJ-?{mL$C9
zvgYCX25@$<GQg)RzMk<+<TDu87z-8iYcy*)A_gKL>S;*$?YdgHk&iY^s=WevdPY-l
z)%bG>%W=hUGa;QcG8?cBJ9yoX;C?*Q+Rn~pFI&HB?Pju1^W)|0*z*QQ5T?IK>~tcV
zyF<FgH#=hGCeLaqCCpoLJ$4Z^z~eTjP-LtR7wBqEd1CI`n?7Fo`owkYr#Lx;=WzBo
zFkV(PRz&f1FNnJDKA`C4@z&-P$-UQnl&A0_FG{3)pHbKyU29OdBiTR05`UokQQJ`k
zS+qK4Dp$0lPj+Hvxo}$Bmlly6R{4QAuD-I4isKwa{@E&_*Jk&4!8mx1Y4U}d>g`>H
zHq~K?b4+h$j_)q(YB#bV@}WvUyTGxb4m!nSiJvo-2?NYe_&(8nn3t&iXv_V2+pv8&
zn&4dF{Tnq*(qwGMY3BRf=C?C>hB0oo^uXv1kNGxn(ap?x2az)jds~;EjJY)t0J(p)
z;s4o!WSxVpG5JAL^JG1f?Y(hs;SuM+0Y$KiGl-U>cE%p5dA1i!-AtY#J3M=`BCCB)
z%In5#<x<Pvbd9n)LUETc@Ue}qqU4;ZdGc~2Yk|*uGoa=*weEd6FM8hLVt%7y@EMXP
z;5#L+aFJ7lYP@4uxG;aT<$BQTll}4~+6yZXvB=B^tZ2?@*d#h>m)e*{hL}d*Ase8I
zZ5fjGzBpAp>6aAeqxSi;Jm_}AJ1JL!wtvP^RbL6>*QdoG6*xK?(m6e542<jiqLIO6
zLwEF3);vxOH#X3F!F*6^9bTs^=hs^EaP6uCM5d;D<FNRlO*%|}fse3&3~qSny{ltY
zJpG<{5qn*=9;<4UROeO7D-_#7L}7(0@$<%av|FyZ5cZ7YY+uD#c`<6U?<w8egg?m;
zh^%~}WieR$z_U{|_?HKq;g~9ZD~1&?EFm1v=}Hw|^HuJH(svU&PNz7di^bm3*p<=B
z(aP3*C4oQlQ9D2u*9>ITo~PRH)sC>-1lh6o=|~h{=kRt#7Q$+-&40wpH`g>l9Nx6x
zl@`jDebwF)?UyaA<$vSW)Tvn$9*2>Zg0ARffT15g$(mA<c3sHCa_efss(;yl+rB0+
zTc$un`O2e#Q5`B+@N)e6YPVK%LodMBRnptC8=7KOL)b=dwohSU5mn<mrcJ-~a&z%<
z>&$)H=4I}Nq~?na*#(VQb@pTm(w<m!v^47Ykc!)m0ia5WeEyHOU8MQ-E)t;&ak}8l
zPKdalzZYGj_>f{|k#W2BR&V5z3)r=JcfiZ-U|G=0QCjcy<`Z*;>^%u~4UUsHajh7-
zB8-f|P*<i!V>vi|a<GmSh%M*7q`@NWQLE#H`?ZSZJ!A9MMV`#O@vIjMcII2=;Gh;>
zme(6|f{)0X4T{3=x>&H++;oTvn2m-@s^NlH_+q}@4z?Xu%`>l<$NrI5c$G)|%uX{!
z(iRrC#}U3W(t3sp$cxrSx_aft@0&Tz{a9d;-M*>fyGl_6?V>C~!(zA(;Ajz$c9&i~
zPgpYV6GJLvefRGU(J}2CkleL?F$)r!Z=}X^PTO2<=MroAh#Yq4czVMtexKct`p%~g
zc~G(wGt0flI(OP}#&}$htFX=Kd!g?!+oy``F`h!r{lcHu*9OY}<1)<s9}d*nl__c|
z-10d4<z_FV&`-L?`Q%kW4UhGbYUga^YUywp_>S7Oj&I$R7Bvr>FfJ2%XfMGES*eOy
z?AYV6A!4=Q$de~DBJ~x>1ep<#dZ(tLRtrD<E=M;?s^_b5kqXu7;C|M>r)g@LgB9F8
z%#i0YTV?Io4wZrDA(1yX30>!*J;a6u2w-g;*7ip4ayY}jzJf%;#Tj9o-HyPp9c&d$
zH%~PExD_bS&sE=9*Y??V)nFkiM4?^}-`HKFvr7%mKs@!;hD|ldlLA>eI4;3=R!iul
zWB1$XRurYJQ;dnBH1;tdEF`x@piSFFAU|c;@I`3eScSlUkjo{Tz&bE1z#$WHOU7-H
zb#BQkwLSQu0bBjA+Ny_o9`5rQhvOJ2)jpp9(q}Na3LWN(rn8ugi)Q8{ElRzidiAlL
zjg?`Et=lTpu_d`-S_tPAo{?t-xbWe31RJ`~spI{@8j2wH_^8!>?cBVE%fT9$R}mC3
zPtI(o-sNf8Ce_beN=TA1jTA%L{%U__WGLW5rV~lYOrL70yYIC1#rF}#Yu0&>)<HW8
zk1uP6m^I`>Je_fmob}&frJtoew4J9)fKZWM;)Hv1<m17SpibwNhs5D|d^e3ObZzDz
z{D0y9xO(?6MGXkWD0=?d*Z#h@{s7cpZ6u*g?M(iQqW@F<Z|-K|{Qm&_|Ijug0V4}D
z!+)VS69F3=Gy8vim<j%$HUT>uBlG_N?Ef9oV_iU1mUmY9a;1U;0)(ZQT_oJyB_W0x
z1`wDTm{S5;h;XIKfsmxqU4l!Jq{ayfBr08eiu>OB-g^K3wD#J~YWH$_-<sbxzH-hu
zb85oKfTZFaz$3VD!jKTqK&Ajxo}XJv00jyv2oNZQh9)hcL_39losu20gb^etaLAtU
z0Z>6eLWbT|1Sw$T;NgHRyt{!zL;;78n1+&=1PBB`AX0wO5#ka6D}sCq;sAIB1K4oD
z1d|-926%rE70l65%(DGy0k;kC0vcLcNdE3z0+PT%06~Kd0C*9mKo0{q5Mi9aGX@3~
zAg2ECNjpq+6)5B}2?_P}^#v2al@Q__(#{UR--Q(E0E{7!g6;!3fc#)#7Qj6Per3oc
zh9DVS1Bw3LW%uPM^bvr71~|404Ip45C-1{Jg%AK{P63c<bpry)L0-XYzQ70I->+-{
zhJuFv=-$*{>xC57?@SOdVBYRQ2EPj#%ma{@ae){V)=))03f}<<GLGm8CU7CgkAr;#
z6x<Vl!S1`A5d?H6ApnM8-q$=N7=SJ#g`kKEZ~MhiK2^b-r>q#nQ9HuHLBfQRzAxob
zqXGgL$8YJM&&+lV9P~8?^uv%*PS2mMA;j(F37IjLr?HDlKhZ&sh(5N?U<3h20+o`A
zl#Bo(&;bs?-4H)jcbE5&AGs;Mr(-h-C<c&kAsvQb03n081R4BE03{OOK>~z4goA*7
zneXMJMKmz>AwY`(;1J9yVvg@FG1x~g;$u5~K$hVRU`HI}F#x__zTT$Lgz3dOW_JGU
z-|xOqnUvLf>#K&}%n$v?Q&JE70ssIJEdv4~I1mV6Qc%DpjL-!CXpA9(epkTm?{sK0
zka3b9%5`Trf2vLM`!NQ7(z!eU{%FeJ#~09m_J6Za*aRF2FrE;9`JI2<r~cv(`>Vd~
zY5wlQmV0-0{KGr_$N%yfCV($*|Ih)3wd_&E0KFVy;0*p?TLZooYw<=9FK-_96lqZ)
zia|&-n8Rj)hKzy)8u|SwlyEP>9a%z(2I%}bp2qE8rLAujFh&sKLVmq;7&Ha;{YM#q
zfV&Ox5J<u@{DlZe#B3|}*50SX44$9o`IZd~I*#ZMA|s>$1Ry{r@<E(mFr;x12x1f~
zh)3?&TLFcF7&xGkfd*!9KpcXICH!=oTSx>To~K{7kkHZqAPRqnQ-B~ozQZYD5gvbR
ze~ZBz*nt8qAK6_QV1K87e~<(S{16x>%P~3L^+K#=MXn@Mmu&R|V;MESd!Lb|5d#~C
zz(p$fS|PhvOp<f1Idhp_<a>uM;dQv5Q+wr{(#`t83#{*!KcaCVw-xuC`i>GfoP9%h
zQ;{5IM%aXRLsShCt7eFvpccbTCvb!pC2RK2x6Own@jTkg&Wxkz(PbmMO59AU{k+Xp
zCEMt5lUqLKvKTLGhYK#k-r+yHJ+1K|53U;7<o`W0Z@8Fj$x9~rQd6pdg@K_{cxQGS
z3AOqTvAuHQ6;rI-!ZKAxA*yB3_YibXc%MA6CQ}Uv25UW38|!~C$zMvw{uV-i-dEF7
z*Q?vx5&P0F$|cr`Jr>Udy%(B{E|yf?NwP85=MSV=e20v4p=?75eN$nI$rK;S<D<ih
zQbk`vv=;@Qsk-3ktMjW~6OA_O9dtUYM&f+FyoC4PHn^l|I7#xKyR$VODE?tgu-@-O
zodrhX!NdFe$*hOqG)_9&brilEjXp26(K*tg0V>NB6)vGC8HBINfhd=@FE_xz!7p0&
zlv2Ys8=k!z`NU0B*xzNmobQiEd?c(72OP{-Z}A5bO$Pe1rWd8Eg1{yO2sq=KhTeYp
zD*Vud8vMvP-GN>!pf%DlU^cB5g%9||F+Kn~V2TgQ-nuO3STU_z0B5Zt%t7^I1SZ)q
zIx1taaFL9D#*G=clP#9w&05N%L=~1!a{z^(6}+21F)co2xi9NaK;jU2Jwb}ZVDav}
zOTCsot$gN{D65brkkbk8B5od2fs>2r3X`!Vm~6K%vwGJ{cxK~KlA2EDjlGW+DjU&W
zo&87`yHGi56y?`b^!D}EcC0juzl`O7-&zL$%7Ti@@m9BUVsf^$r;FKyYT3`Jh8V!&
zIf7d6I4RYmbn%;Ja8k}5$u*?6R+cx6B0zNV`?yQqD_*3p>#*JPc|>GWg&?6QJM%Ut
z%N_ww8OPm+%i*NiB7=@tX~lP(hdVd`oiiVo(L#4}lA5R3&Ag^R{pCQhZ?8~N0}~(Z
z!Am(6V}t6Loa;yL*7Nq_nw~v*vfDO=d?K`*2+wl3M4vvnn(Z;1ua@2+E@Xaf92G9U
zBMOf*r@#5wSg$KC*L=#W6BtoO%d*1mp_m)VidDfkBZ>SwzApQ?8n0UOy=~D|26CVh
zcCoSeFn!E`W}o_E8E~Y|cZZf$37vP=x~vhg=D){upu%aZ*5Y5+dc;uf6s=?;>1bs3
ztmBK*vV%{%M?LbLJxI7wW-=;p<yRf!6y-~@s}El|S5`A_{xex41+5{~681X3xSHKc
zyAI{gTPL9?*~P5zNUn?>U6ysZpNLuWY9;00JuIAEx2hDN)!Odfx|u~GxV0UarEcqu
z3AAp7$9vioG`U%w*<h%}$+`{d9aIkL8#PHW^2|mmWonztXoLE7NEQJT^xs402?_v3
zRBPiL$KHoQ<wN8&eNKKP6LlVIcOUmQx^`y+mB(?k%iJYlymg}Xi|Y11>GV5Htmx}l
z5t~~KMF-c6@}ZnY&9)ibln&>~h^qLG*^0TLx6-P%oG_BS$xKC}TUg5;19l;IS{y=0
z+K^{!uCB~K^pf$yEFs(I12^6tksH#JfHzNM47$UCX)shI$u9TdeR{P!k|IwAC0&G;
zD;LxMTmiWuVR6}@9lgsx%aHr?X}^|dn+&L5ovB?__v<}=n7OXWwWns{ZY+*g-sWHW
ziHhj$+sksTLoSsyMY80u-25aB@h&yvPj6L_a1^%h7+34p7eF`&gEn<ZA@}Y%msW?S
zBrR`Zpee*jZjYJ36`oB_R!jE-5!0wOZ7`C~+FOdm_h%;nU2tF$gGnccC<+zdqbE&V
z*y453Wt?<0#NJ+Yn-b2;;*j=y>u&0}<)WvlitV7CJVW!K_M#pkH{2TO@q^GGM?7D?
z=&y?ZgUqf&p;EqEVUomU?hS0`#{L^~qS1eJRS4NJqajK8zFhDQZB{2_?8TX3{r73m
z5TAc#O2p<e?thb!^t#_i)Sj_@X1s;{`2HIt*@R^e*wsMWxOugo76@BqT<I*x*_~}J
z=qwn@rKMi~$n&J(*~KCR365LarIE5byN}-J3+@R{e^tfaxeMQ_wHs!R&5KyCO~7jI
zYKZm96iv)7LqF_km1q2wWI$iyRBZOo6`86>;l|mDKee{?Q@Ppcee%p{<3?6`7*YR?
zU9x(^aa%OZRhj?ble>`5hZ2Y2+OCw!yZ*fIv9wiX_B^0x(Y6;@@j@ZBg39z6M|32{
zXw<{qT4}meYNWRbJ0ui{4$zalP!?FOqh*+M#5^7+JCw+ImD4+bTK`yhs9&oj&7k@A
zYU%nzbCdJy%r7|&ifR){Cws!%2PepS1Xq&M!&^Kn`gnjkzC$rE>K$3lW4+B8uKNXR
zB~0e35BUr#!s(QWhtgBcbO?t(s%KOjKm8pISpS_XFWo@mGpWN0(o2q0Y5b#C08{S6
z#R{&gdVFw4R<dBfjTe0)>k30M1Zc2G$(YP}yf4}p0pv+v#<1Uj?9XY-UvU;m#N4NG
zxx;yAKuEJ{{rIU@K|gAP@9+Bj@?oYp9B!}o0CPZj@dyfYFz?;Flu2uJK;e@<unO_g
z84l<*P|)5BIlU&eGmF3H$X3c0rT(9RUaYcvr89a0Y$M^VJy3LS3zQ}NZ1i`>agUy|
zG@s_{5(U;(Ax22UTvtH9<>XTR;zoOM)Wu?Isl-mpL?41USZ)&lXJunRBJ<9}txx<s
zo#Q(49mOy2x8l5?PSFv`K&#(&%loG7`9CINS>OiirO)`}t<GlKRDMWl38Hw^Z2L%w
zuY-_qVgcaR3noo>iVK-x4v4<XTz_wbnbaLL3y+jt1}CQ=^7)W)2FC{m7CgD@xNY}8
zBkpt0qi`ZCBjrtvu=%+9a7M;>1v6HFQet&C)>b@aL7o{?dkb#zYN7J3AJG!7uGu-r
z>{>s6)M$#G`YSOpK>QXF3t2)9Qa?pYRg0auZU9sJ250RlfU~H2xvSWht?+i{GEb2R
z;WY449~m+rQ|$>2lk_o2cUjD%Xv|HAaseC?*P%y#J{ps!9uzUY4WS879D9D~;!elh
zk@ihjP|Fy5l?S<JlJri?muGBr`S~sR79^kMy*~E8mbJ+(LTJaNS$Ap_eXt4AH6AZ=
zb}L#6C>iP>g2z%v#tYn>z8a<-xT4R^%o<`9R0Fa-P&!yXll+!q)Pi0SJjF*;NJJP5
z^wC^(=q1}SkbP9MeBjZ0%?$j@2?V>P_SC&u_fbat0FFY=L+wE{V0+{o+Ch81i))Q#
z`P7+RMEgW3wn8g7&A)yAx9!xz+?qYkc7<k!u;+1HxDti%zTf8(Ax3$%rWm6Q2w>Y2
zq?5{Yxou3?48wWK=$c!GWWy7XoT1UbHou_DY@?#Bt1A)4Z{DBQo>}isH@6Xm>h$t)
zy7)4bHPg8sc)t-#anD%N35bIy%d+p1?=rq2%oC+$6zaz9DofS{@p)@$o=Y<S#ND}r
z6338*1A-Yl-+z+G`h$3LcxaT=PmSj??jn^RhpRp6D74C{HdaT8V<=tI=hc4Y2b5?s
zRyrTXO46D$q79j<>!$iG$^)7E+Ss+3_vo!b?un2dy2YR@#U~m;BPWP8O(oqTiI99&
zEztFNUk2^A>$?ux7#BH$A(UEU;``lMvHXi!6cVvc`+PPwGjc4PHndl&TP&gVg_g}h
zpjhd$%_i90kMhvVnYY-=I`)10G}JK1DSA^~R8@7Pbz|7<v~kX|TXk`FW~NfQzBFkc
zi#qFcuNfQ|_&{&pN#`9kX(_EY5+uttTBisY2{O|SEd3UQqUv|Gc(ExO6-c<mOQfEY
zr>_sZ0n)pU^{u6pR9mKQZ4>rk9^|hyCM)SYEt`J9G?!eqy^xd$HT%>?;60v$o}Uk}
zB#nKt!xzWep>-9PA5I~;m6Y{6P~njaB|R)#h}@y8HUwUo@Y57#T{A}<IAvycSAtwW
zs$UEA)(i~-O)kvxVa|iGhQvwNF>aIt5~Yi*JmqX8OBVYU_-*HllBS}ib3(_SOjN|#
zY^l9#C{fAptv^)vS$ad>OLZjFm#tQbHl;=8eJ54etm%dQA;<Sz{-y23m1!IAHV?u#
z<eBc|UpPg%Ityp`4(e5wDk)>ROc6WqEW$4m$s4!2>uNnV#j2Fu#B>4Ndt1Is%n4L5
z>px`ZX;6BH65kc3<C`QZlu^Sh3Hn7jY8XMQ9s$*NwCcvo&h*WzBeRD@s)9|SO6s%4
zn4$v2Pr#lGv2amK{t`H*b&ylqWstNkN11&J-4@?CkIJntn!<?{Nh^|;j-5vVpLM&#
z3YOwcw_EfC^}{B&EG9WFfvs1}_p=qXaSNIAdK_RW7&{gpQRyWMGR??tA0bG%)Vx-*
z`n~izno=*Ba{Ghs9Pl;7!YN?#PEKGgn(T8r#B1^!{_-KJVseK`jr4CsAsIKp&1s8g
z_5TQPBmu3lSfl1)qe|f|ZbDa67q~XHLgT?uY4HBFF{@Ht5Lx&k<bv`9-;y$n^MdDg
zE?B9c3t|rWh{sW@@d&WIA@lge9F7~}Vvpl-T7}BnZ1vLUrqeiltBa>S6fv8V=i;a|
zj5$8j-c6Iu2+ewG!pj8<bSEv{iNAe?NfD2NENPI(pHuc0#ms+}9moUXvF{~!y-2P?
zhq%B2KCsU2@>2IA-Q?Ti%WCkIA4CMF@e050XG1|Ua1w3J-NPtzV?;9_OzK9*hO&S2
zFu&^M`j*qxZe=DSYR;>A%K9OjLkzCI1a%HP4{Od>x_$`Q@KKP38A=>6wIo!rmVP;z
z32~9vHmT`2n#;2-iCfdkriGx>{Q}>)b$Y$aPDBSoNu_~(x^!8v42yRzBOr}pC8E4)
z-6mngw#F5)G-VgkjVDZwvfYrPXiisiuu%(uPE}6r;ZXzH(-67%g}yHhxA;3EL;HAc
z8UNAhDSF(9_CD!FhCAg)Osjz%iKR+K@xpt)$J@2b*0vPT>-|I<I_v9EGjzzJd2llV
z)?6sr_^F^@M63;6PE<rJ)~vjZlT_Y2YnHVNDpyP_nnUr+;t87S7s-1y?zJ^xT0N0Q
zQ|57yWaT#Qr+~oqLSoNQ!OEZfvit)<aO7`T-YQ~3kN`A8(EFFnJkpW5+w^gW>*OO!
zhEJ4?r@JIuJzVJ~1){sW=_|!p@0sv&n*6#R4Owl>{QSyr`z*<Khj1)tcn#k(=N9S*
z@3sX;c2PK<l03Ue6m0AyRcl5z-+g}qX}|CC$r)&A54u9FYT&psI*>4Nv2u29$0=K1
zO84fOw|<72txpN+)<e%uLzH!tQiWRoY%=SDDXjb;N!c%t1l8Cl)$L+<X<X@~j_ujv
z^QKIJxM!HD&xQGC*|;u)<+z^2eFD)zpsV#Q0!@JA^M1HvJTAbq4$JuSI_2DjK5vHj
zjU~v0sTy|HsX&%BwJo&%Lm*jb;#N(0(e6&VueZmH7)@D7Z41+|nY%M<mf3Y)uLe1K
zSED9DZh6cPb|oWb_QcL0zI!aX_gwSlE;<rP0S15SnP4|^o?$0`06`W_Xz{G#Nn$|m
z=h21#HU4lmzlRTkQN#>rwVZ~cj0Muxi>db7oj-OcX*pMV52q7-51!}qjx@ucH=(Qb
zU?*#vQ!Tn{c~N(Mve28;l1HSb<QgniE6Vgi&u%As?!~8<9jeRp<gpc=rFk8^Uu1Mf
zQruC4ez?>vuhS)OUssErxNq(-HklvyA!`vWl5x38kEf<KNQ`dgZkl)4L+i^);$Ji5
zxtqFXzjL+tt9cm97W`SxXf^w`dO4-{uf%~GDLPvDc~Qnkb&ch=F2zBCTY4@w>cx5^
zOeyh-y+dn9HOqotg{9B-y4i?z^H9`FF8i5bTqo`zuEj=9aE{?l37G;^_#=0V%xayO
z-`R{fpvgcsXJ~}_5w>H=6o8570>cogL?)@*wHoFmMQ5V6kJtJplLzzPROan3rcOXz
ze`E=seCpERuy3X|9m83IX|)My>taXoteNv)54AIYUBxD2$7d!3L9`}00{o+oQ46n+
z;cipOMyzoOusj<v$EpNMEudNaOKK_=^uI5&ft{h!0E|o56+&0_JRkwPT_IU{F`+&E
zTa!b*8Ro<RU{^`^s(gM{4a$uww%>#rawH)~StF>zjR}A)-s*n8*Vbv5V@f+RPw<Of
z-{oq^M*P}LHf5JrS#x<qN$NXIV4~MNJG$b^!=Dt}K&ozV#v1!jr*I%2EHOM$B~z8g
z#niGY&dOB0EjH9ORV(``--!nyB$Fv`>5J<a^myGnwoOOB1J0#XCS{#mx--R{qHmD~
zKKB1yF77S~^QvWqn}_=2>MR`&gL$X^!ISDloqH6FI4wh|FE41EK29G8hrGqbqTdeV
zG`Csez;;+|kxM@|RCL;qkadwtuJExwlizJBY3lH_0eZr0ygFGM5>{T>uQy(Sg3`jO
z*TLj_dvr6dPCqI&0j@KSA^sH1`IXfjL<fD054q;8yLZ=hUuv@JMD^d@7t2_nz4asX
zdA+upFv+&x6gQ7x9wQ`qM7{-Ih`Op$T}to7?Q<HWQ4wpo0&tF)`XRHUa+5$^`EwG`
zEZE(2VKt<@Fh(x!2_l_six#@mOd7M)a&8kV2blX(n?uU@h#ornsJb79u;t8}ZcrBx
z^Qx`Y#7TEX<nhB8COHm!Cus~kgt`m5ab0#dneyHLHa92z+<Cvjk19EZ(W0{r>?0Qh
z$~q0eD-x?19bR{d*12=CFYr|TBLo}g#sO8~Bf|q_4$tnt;997v@R#LQpwdmp<304p
z#icP$xkbvWE?@sTxcl}Ux42Xt#|@Gnw%z>#$C$f}I*V^kcCE_@StOlweaT<2@DEbq
zyTC7rNkIv)B$DaN{t+%7`FTxRv^-lyJpM$FfL#0_Vq-gxAJjb6G*MrEJ}3cdZsRgG
zLg0sv<H94?@Cfxm7FAcRmLA}{apS^sV{`E5BW_y~_lx;&S^ZG%Ci^SD{i}|c7e?#Y
zs_FxD<=pw8?IQfgBHO1yy|EOOZ=Eu`xB2X^Fc^7)q!~LgXjWF^UzBf8S*)|z4N!Z|
z!=3z;=;W{a)_4~O=<PhHASjomwr{1;bckXoM;2O1RB9<K?d{&f=}q1GaF7VWeLNCK
ze>Bt|LN=dTZ}`2j?A3|RDhkj^H=WEr*8G(n%PoQfh`6;g|8y%Qu1dlG$Dn?xOc?a{
zP@V0Hxx}+yJ4hziA$Bud_n5446MBsIg1`+WTSFzSo8uia<2JZ|%?jUG_15h(DmF53
zF2A9tF9b{ULI%vXjj~%go|RzX275Du=E992$JpT(-)XjCUGss9K+(E;&0Nm_SH4Kf
z-*FFU#fS&g_*S%hc8t8*cj0ej0!_5H^KK%6pF)2BYHoXHP@frSZ?s_?747~NoC);A
zOTPzybL&cPo*)t&WL2WyR&<V(yhm9Dn3iJ;iJB7Qzb{o1;=3!}5jIhyiTfFTGHSVu
z(78&gG%c7z90IwAFp=f^LhpHe&f_W^i^{g`%js1$kBR!k0nf2JR=V!kQ(HYfdc7Cp
zJ7ZKg**?2i?bYgEaq!L55Uh6|GgMP>6enpOd}g|_Z&6{W^Hh~Qi`Miqm6maAIpPL!
zKNahmRnQI(-OrX?xFY9bzY<%#oew7)AN-`-PL?9j%*JJV&ypt7Kq0pl$Npp!u-y`<
z+*>)5u&zCQ*<nmu8*qQ|TfKyOT5(>Ifrq8ORUh|R0Vu=_=L=9P5C80YBGeUGgLc`Y
zw-1gZ>!AC)Tk1^Y%kN6ZH6jD3PZ8V4n{c;tRaC3$xz!d<5(nw%`yG8dHtT@x+9W^L
zJ~^uO(fyCCRCz=xr4(Q}WZ(^j*I=HW7VSwFx213o&e*Pcty7hcs8xdnv#Iy2&mE+-
zK>>rVvXHN;e5StMuQUy~eC^IdM6%Znc+bQ3D;jmO5aDGfX>1(qSEH6F<+dSzn3LZM
zL%gjKA`e=!S+YZU+Bb^|y^<+iIAXdf+a)_4oEs8xY5!iQQ<C28d}$kVkB;<#iKEpe
zea-=#H5^8U^u^6vQF?+WzmGD0Yd>8s(>!9ksUs%sNO8WtvESKccJsRg+g%|=^v@IY
zrceJflSHyfr$<0A%u*!0C*TJZS7Y%fYB2$#@0*UW=SoN1)KSe7P|EXQwEE_oiHkSh
z{n>=NRj|61Q{d@I?V307NxJcl=wA&8TZxO^8hv#hIMaSP>nj_<?j9}uwBc49%(J;$
z*=v>+JF_@Z*+sS%WvgT-)N|D-1v_bP=ZlJQp8`HBZl62(J^P64_jZz|&n~%_-H?=Y
zVu{CG2L7xwQnA69A3ug>r8tlsOJPmIbb6ZK$cUu>IK9wi>60^M&u9U3=D62hUO&bv
zAw9*w+i;Qu!7yQ^sCvo0Qtj^bJO$n1Iq}dDy;NDAyr;epsC17m+-TceA8CdB>_kh<
zugW^tirAu(=H~Xnl3h=-JMdW)LrZ+OUg43Q&i8&s6CJnx^A8Z(k&^*7F2WgiO0Xb+
z#>{<E4?rjcO#P}Kx3<gRiwe$_N-Qbw_Y{Y|_?*JfoE~HGy#woo)5PGrA)6Fgh)2wu
zfcmI+DO;cd+I(9j^Nn}O3^>PB@A_}|p~X0utXlJ2GeGbK#S09ip6*=!im7`l5uj%k
zG-)aIJZF;}H*s)QfzY)q@dl1pY-JZbYak)N{(08e1+oh~TE|0LN^hf;Hw!JZ>Pyjc
zB98!7b){AzZ-EAZkQ}=`KswJy&uxyIkVD;+6YI>q9n}HAdw`efkIwG{gOWKS46j{x
z{YRWaP+I|*MG~KIMs$JKD50q0jEzWR@_W+WisqbLM<BD3ugy>&yfnj!KY)Hw1>*ld
zCT9L`p5}i^nQWY_|Ca>VRz=qSkO86hOx*)MbrX0LKSUrj;W|VDr>$11b(5N;3_H??
zB1HJ_=Q9>ktr{<aY~J2%_dT@p>!mCEyoq-`((CwoHg8K>e_is^dr#EeT~k}fY71t5
zHLVRteqR2%)jm2Gcip#ZwoR6mhvTu0QZ9Wdt|nh{%uhwOpa8zxzuli)R9Ag^@<1s!
ztSZyNVXyk&n1dgeIlPe`qd0}XjbGjNX=H@>Y#T7UOlAv<Ij8tS6+Am`Do$1TRh{t~
zz%ntRrY7p(Q*6G-Ej{pf=ghO=rH!*~(Q35J{`Q{dIlttdFstvm^rreWc*kGyRL(8E
zj$5<BfJdud(=xuUAp6HtUz_d6asd0}r@2{Xsr$Yvs;%YBPJi8R&u86Y^dqbG(Fb><
z?Awj3Q=St?-0rZ$Cop{JVG|JSAq;>Rhd@Dg1TO<vAnKly47dTL08)YzU>PA393^0T
zRLO(DF+ikC2gynaKq5U1psH{|EpV_Ch(Xv(M)w_N$f&{CCGm;VWoA%<#~3b5480d-
z*mA~H1Y?c@AetdZu#3@<qkIlf6JxGGY--q{5~RRxxKkUy-n|W@zWcrXu71v;RRk*>
zhLW?(>EZ#@NZ-;DVcw@l@S1jEVLyqYjBD+KBHd0=qAC+P&)81T$ByE4*DRv+PP*gS
zmsuzJKhJVvpS@hi-4&Ue<Dv0kL0Jr}rnKHT<@;1NVWB=I{;+MTmp<hjPag5EN6x#7
zxY7^in|o*<<BJkf^LS@`;Cv)8$bE<xBq%K8OA^#((u}i%I^=(N>3(!YZa>SLx5Jz^
zNM8%AzXi(gLi1jFe2|_O-tU6sz0~zy+I*0nI9T5d^IK|kmj1KpTzZHfTl>6?IZAr3
z+f}GKY83=cWMzGe?P?QF^SPLngHE@n_ju9&W-I9Xf0%ui|3Q&pX65*Q6$w^Omj6MK
zVB_Fm{J$(ntU9R5Ki~dxL<Iy$Kvc1qxJB5dC0L~)3}CW`SVn{5Qthr1&_p@PL?uZ}
zKueeeI8ew)?ibI!Grqgtvu*F+v+KVvUv9qqKThB#EDUH0PJ-*gO9&1v0SyEaV1-52
zB!B>cq9B3<2u=hGAmMJIKZ(geGldjz7{K%odoNTJl$fzel|hS~OgbRAg+~`qh-jb?
zQj-u;Qvrei2ndoNdSMVLumY&pAWi@mP(UpRj^$vSiXZ_aMFw(n7P{&FH~>2ccmWM9
zt)#zpZh=YIs35@shX6W^8yMGN(~tmNz%d92n5eN|VvzP6W~|s0M5M>pS2U1rPY5Cf
zIVhfQ0DFv0OaqXYfTErPxB&edpyvU*g}qqFBw#?YIfi(CBjXU_P3)K;AO|3I3=pWd
z0Zxx0+yV*!I`#n#W$=RdkfA?}t8d1A@NZVEfQEwm{^6gKA8G`c-z`j-K!WWA0}eDW
zcOZ@-ok9R#S6o3BuvvJ3VO&4VAYEPr4*v#pj8mw#am;Tz92f<4CXfKfpP!Xna1c=8
zB3=+)gSvh$luuMJb`#c)V5p&4SSWGtgug0z)UY5S2jy|7AH8a5oCDamA7{6aL0laF
z*8TlR6sFMOZQjDHDu2y=kz@bqoC1pg4hWE>lq4{L9Y6(`5zTXT2V@$gPi@L?>6q%h
zJ@iNcARI?pfgi&-h3)9=HQ;Md&;rFg1^>Ojy7#kKo&V?&0Ko--uL|TCIp25Z9E{5k
z^SJGvC@{bcKnDTjFaSSZzrIa_hSPw6?xVl9zuuifSW{GDZ?W%w?BDjOO-w+*_oxVf
zfFL3xf&hkwmWlx+xBGwd6krDH`l<oHuPY(kL;y?vQKPaK`k>xFt-iPY5re?*<}|z@
zaE%H8Z+igkV3END9ry=->@)t|-}dLfs3(5_@BXXBw-=Y6+S5PMKm08Q>>1Rr^t^EO
zc@{XYu81691AqUPaX!QAx)i`O%D4QjSQia&z*F|+>i+PEW`uFW0=X~-77)((6+VdB
ze?%G7F>tH`?gM^43jjbu1PlHa4m3a=2Yp$196Y{jgAOI0=2NU3AchXmyG2Gw0Sr)}
zVo1kpG2D^C-=U5y&H+Zh#TW!65aI-mbO1VDzX9VQP(JE?qa*=Cd<rN==qW$=iG+C+
zDRAoj?D5iH_22eL!h{$VAU<YLfQYa^8|YeYI>av0+6Sf@zCj#Dd4k0U+`(R^o>G!n
zQp>rxJs@9%DO>x$KEKUbZ-={^Y0mnxX<<tcYbV3#{XcUz$gpB-dr1Bmr1qGVS}Ilk
zUg0)u0h7C{{myRv%;I$*wC}lF0zPi@lv2P#S;lwmSr}gIr%!i7<!dqO)TAwxb5diW
zM>5B>_^CYMX&UA;O$>+yp|?ASyp7?7Uw-e$M`C*trO|3d_?-S~ITnrdr%e;zlO^**
zp^y>d<;38`q{FyF!r6CY9Y3k+OD%VDUVZg?LcG%XB>$|=i?yQ~uCUtYqw*FRt$wzE
z<@O_B*olJ?2y4w`u9dEON!pQCNst<IdOd4NV@}y8(Y)dPen{3!`>IWF6-P&yS5>_?
zx456-+0xa<vBE&vmQg4~r5Bp8N*bN0o9e!p>L=r(Ut*Q((fwFXSJZuIchdNqSe_g?
ze+o{;b{%P^*KGr%ArA9PNyc{pYLJ6b^5Vily{a^6Lx+C!V3AZx7%&7kx`q12?)rKd
z(RY%BPVy9x<)qaP#B3|Ah&dNFi|&XD47&|}COs0$z^W5{E0Vn(@{&;V3dTuXFP}}M
zM-oQ#&3;@LUsU^*Q46QPzk9;mW%&2r$h-7#rTvi0o7xGC)-X}(k0`lX1X{n7^jtIE
zq3Bmir0gp>o~j#XP~+c>xd_W*=wR&Lrs}J@=QBCrG@NR1uT|r|Oy}y^qI-rr=F~34
zJ>-lxlXF;BL%#1NP1-nK_5;0fw0^9@s=L>l>!S&c()aeLQ-~UM^fN@6h3N2y&_K@S
z$TGb0cgV3);39r`cue4n?saAx(^ct{57`vj3GTDYR9V5u3{U8Xa`$V(;%WX7i7`7_
zPdyvX6A$~>sPV~njN|{J(pb7@ptW0BQ?|<=r8L`bIQDU}xD*%C082Hcq@!bSWTIc^
zVNs$8JSI#_$$A7{tM#`d`{%ovDog>A4%tcxs#giNfvG}8sCCLqdz2?~5)Q>|hQ_U4
z1P3Tsj1Gn^?l`7l@~`=N^Q4Lk*%I>&vTlOTw27Fif-v~iQ63-;|Hw>&PZrf$(T*dV
zc54SRx|eW5aMSiL%5WLj>MYJYR9t;5KYGGwY63$wvb1ap4!9rfUk%mQxOXh7<X?cU
zWccD`JBv4gUtRV}+oWW`?&qo{=U#T;s9qS-)PV6L@C`+!)3`Wrwj1D1TJu)a&qGP^
zVpSs6Uxo}+uyTxm=L&RhlGUMyOf+9~Gefb(q{ro(LClQzhs{Z1@vSm!$!OLEpzFe8
z%p|uH>b1)>M3>TSOTwM$l0qbuXc$kIPW3-}u%R`59!z|}RdfavkbG`0uMvUMaTr7|
zUNcP8(=>XRSqz)J5jCb%h^xoDrjTU*He4O~-nnM>u6Pk|IJlUuf`?t3@L`vaWWi!h
zzqR}rqOZ*%;_aGFT8EXzdo*R27*%}^Eas=+1$g(OFz`CsgWZ9)Zw!H(0o~)7j=eMA
zHtEcPk6{!ygRwXjky6*iS+_Gvh()z<pQ=<H`vnBDoNtHlq6Q1w$wi4emTlPh<<3^P
zVRWg$XNh;&E+nvJ?^cvK+0ErwOkWut-_y0#tvV5;9$vA`sVCE4^x(Tf+GDXugUHQL
z#;MDnSi)#ThK}^9;T~je;rYuXo%bUWp}z&0Lr3(;D1deq>x?1n3Eeq6Cy}(hu?a;K
zT_7Ff_1Le#C#tTI&`?=0DNTM?`BDk&d%QQ?IFS49nrA$lDi5wxJn;~CJ73LwIM;yh
zG1)T}R|>eXP9vqhWINw(={{De!ST4yqXmkb7YwuDy`C{kDt0knNIjAtdg07thQR*Q
zrK-%CR^Qvc@7BeAj~Z`+Z{n6`-$5h3%TM4UTKovu&vi#cUw-N(e(1flmQah?7bK_5
z2EUq2d2l`s=c4EBq4vGds9X-fAE?=mXi)Uh+Olnbt5WKnO{{py`t>@)uy>Z<s+&(n
z+EIIsm3)jnelY80Yp0#c7DWlPJx;B)XZI)yZHZOV0{SU<<aQPwF%1D#e0^^#Ophfv
zYF!9qd00JXP&|Ogwfo6X4Q;7T#)SRkEMInE&7FDEtmOlr_n9D%YUjRqBXasaJ^Ncc
zX9<*vp$*F;2?);`Nw(X%x=g|fxAZjYcAbHC-k4q2DRZ<{I8lvE$59?7t!9Kw1*40y
z4{xAvOiwg2tYvW;;YUMz154gMQDxx(@Miy)b;mX1vVz18>T{=##Fhf3hIFqT)r8*m
zR9WL20d657k?rkfX6f6j(16~m?ZwK!FI_J#PPPU86`Rc_wmvf>Bd3y2b>7u1m-aw1
zFBTdfa$EIX?ve$~R)^;M)7lb^xBWfoBl@nyStpP4zBVCC-!JpYW?=VLe0%(3-Q;(`
z<yqGoV)I)GdbYH-C#2(rH*ZtLQDw@H#IM>Oi5ELqHkkWk*YDinl$>Is646Z@@N{Th
zE-2gB^bDOUEu&Dw@-cY`wqZ;#JWlL)!K37JiTFdksg1nF1%vq{qStPD_b|Rh6*C@=
zJF<J4ZXrQvaHM==`5Wi7n%GRJdYG)Sa=Ge7_~U9?M{F2Qr*M2;h*O<VE%EX1`Ia6V
z_fUqnvNcyO-Rxbm*Mwc&W6}b>xs=7ZRh(cId_((_#+;M%02%W4m&_7d+y$E7vc%c0
zI?lNsUI>D$#2uQ6r{G_VA?NOv7@|@VbdBpUYlKLA2^yHU%hU?MiBK;QaN1E_ulHN-
zXPCrB>Q&&d%S@ya`GII<gS;mtEu}WNx!kDp8jF$=5~&gieI0c9G-jjhs3YePHK<6F
zYyppKZ}oq|dGBw$$SGXS`?E5(yo$|yZ*@`7IFQJYK)PAC)AGR9G*wNMNB0aWI+m{?
zFK8V()wwqJQmddlDvVF&t?){84Mf`oANrS%xzL$27QfJO>&<~doj=0Z4@x~BhV?lM
z8D&4FHN!3H(X^9b)~6g#xr@X{rz^u++F9y{v7~l~>X{$XIr}o~v)DJoRPTJG7d<s$
zo^WmjZ#{OFf3`C$`Shxx+Sqs34VGi-i8Wq|VE;*M5=lC>37RFvU~K3{f9N286h503
z-v!?hCc2m+ZoPBbGWT45%{tZ>i6lV35k4+9%EbFql;$_5bDrx{^26EmbkH;qA|P#K
zZ7x}Ae0sI`ResFk0mK@+8;b4JuAr|65q6b`+j_M!XRX<*ivKoUuDCtzquj(x#{~Xx
z;rj=e*lWvo1RTdgFG~`4%;BW;Se#<u)2lawuOdKaBAwbz(<<IiE>dXOsOYw#rqY<(
z3T!AGSxiY)RSgyXUN5$ixQe#4p}wbgnK5Lw2MAksiK`+%kGe5svJkQA8w%GEuvWyb
zX}lp38jPk^9eR?P@S^5z)d#$r=$|`fo9EWN<zdna;my80KgjIXvdHW^LC>ijX4D<5
z)ahF9AL_2H!KZ@;e9QREg0!RC89eu9J2BGUa#HS*(>Wn}xpu?@L0BdqpWUY^0l+l0
zwC|D@Uy@!4_RgcwOwvVOm)2K7OW_0y?KV5LgXq3EI}ttCyGD}-!rHDM8RHheYo`i!
z?tzQ6dHv=|w3vQ0)R>Sz@vM2uR%Gm(ypPTg{h+V?(RXBQCNqaq*J=MwqAVQGy}gNM
zk;=EqIuF%_3ItPqdCy9v>Z>05kH7Zz`Kdp~o>KL|#ym>Fo6c+~k2V|ic=3Ex<#?qZ
zc@q)xm)|bou7YOim<K<v$04lKzeGzEzT3XnR7HH)^Il;XB|Xro*@slR5<cZAtn3vk
z;ay6v3{`By9*;BZ-JTLPHG=pI!kF)#Z4pE@G{Iy*%Ts(Ht>;Ck@ci5DlM>`SN_<Lw
zbMgg=^^kg;o2O$&jI8)B&d5GrTkVD>@Vf%_5I_;VoL|q1Pz&m_DRkqc7&cB=Wxd$z
z$&2^oUf4G`@}83ag-ke{0jsiUqq|GH8p^u6hu6S6*H}ryO`O3W6)jXAO}ji=JC=2Q
zIZ&Hwckv~%Xb*8rbvw+%h~_%_aj7~hZ@Qy5-Z9R4#60HxR$kl1RwB8RF-Wc58sH&d
zYyBPrFD3Er#vH!%aw73=gaBde5H}yT4wSVAsTuU95;X8We9!_<F8N)nc%v*uZ<LOh
zyzaC_BI4iIlI+fHpr&<su2yVnf{h=6xl#orj~z1U#R;0nL=-V5jMXpVX*_)#^7J}L
z$0P3$7Y`Ja@f#GDwpNS!Xr`E8=IgZiRr<&Wpwa_uvCT!~_@MLZ1^#jJvp9KHZ5n=I
zwZ<=qEnXy9EItZ%jS%h1-EwQagURnYIjEwmQu#7++SzomYJ?eD`*<7V6QU8AIp;ey
zxw2ONoXdr|q_#wESHF1z%eSyY<cGbSNdD8d{4~SJp$cILjZnt&6yoS~pE~!K2{@hZ
zajLj_eR?u^^H~#}XB7FRw+#%s5qb4gc>2b@Y!X7OU^J6P<=2S%u>7DMQ9Iof-|I}z
zy^^)w_*IwS>N)s61@r!BQRS0|STbrhEq{+8S~N5hetYAehF8>x_vW)!l_T}-{@F><
z9<2^v(0<ru*${p<1<13~Lj(rv_2oib^1`Y$OE6gpFC3!IM~JG1rbV!XI<3Sy(zdgh
zDw!(A!LGO&59vjgw&0>Fl}T9WIa3K%$Ucms_y(wN*6)28+iZQzl}lm{F?gEMi)fPB
z+A|%dxiV5rong*b3d*&&%N4{Dl+tPZ#|u(j-amK~g6hs1w-Imgd$F}EEoJy&l`+b2
zsbzyqaTYpE$TYSsDL_8h#{uv8oQ!DAdnP<|W1>RjXo}gj#l5Xri2HUZbacw>X{nqT
znzRy6`20dk#d<cM(N8Y3Tl~HpDlk^uA!io39LxUoQJ2|wiJx=EVRwydS87;d;h2_7
zh`lmG4?*C|lB0$~GaRG9d&}&rI$5CRSF{@=IuT?>A9Q<7NjDce--#JR#n1@b^j7wo
zv#L`Z%Z|8abYjW{r_<?a<nv?~d0R!&jdHe}Hqe}u>~l^0et~^K9t}=Dd2}etidEQ3
z7M*<eB|Hlx32&ExpIUDJlzFs*rOkL@-x0Y)iq*j>Qv0Nw{#1$q9?RsCn|_^bgBsF}
zV+>!9J?W^h0PCv>LG(r81lc#l`CFXHcee=sVU+sz!S80xcL}SXT$XNQLSOxb!mv<`
z?d*}kYp=9bxn6KB{ARmpVN*fczmU*2SKi~Yc?uRQE|WQVx;q#$s2jW>Bh|zJ8@MYP
zkl11;z7QpUGWAw~il<l=QiHqGhF>!JM&&HY*xp4y+aZx01=>l@V<l&r4*rv7Bl#@*
zNaiegXCEiM#b}KXuE%@Cz>rTFq;(y3Js&&CT?M#@e!<K*MvSWxjlK=%^BEyszPs*Y
zQwUw}&8RAK{SUJ+cSt;Bv4q)~@QtiCm6Kz^mUPw~I~xs~hq{@;+MJ%_V|g=YrN~>Y
z!~E$wqrKhXv+~*eQiA9iGMIyV%JvmGeUl{xX^=5TRU@g3VeX6=6g^2jgdeBcd;<5V
zjwIKQd16*R|3&66F@k-xD@P=d3nz(wrmJZA(0STs`Fxdf2YYrV9UK4hT}Fosk-fB<
z>C~j+e-ZW$!J<UV+U@e%%eHOXwr$(C_Ofl;wr$(CZU6hch}(JhMBGNTs#g(NF=vh(
zU#5IaSIZH0BZAZ-=?5kaohzz(b!7MKb8L|FChfIf^>b4U&eUJ<pwdM-9OT9!jE9d$
zY$?QA6@vLbXDK&smT!z?%Igk-c;e~o>z3WX%^q0idB;8bkh7Dev5MZV|HJf}*V?x8
zVT!pAENHy40XLoB+C)Ye`?%W#`_=m`4b?^IC^Aj+H=$cC#i!F8jGHzAOpM%I{ZCV>
zM#j<+!Y^oNtpT21v*KTzyI50g&u?ySkRqO{bd6G!_30_SCl=5w^*Jp=);pacf^S5^
zCpV|j-54NH+3wf9o<)=wh8dpv%N)GZqeHzKx@^xG>HhwHad#2C!!79c6aj&0C%CKY
zo0=$ymibsPXweMe!fHmdke#I8Z&2MwTCU%#gN+)>gp&=8^{$AhNnl;mmEf@ui3*X~
zibedIdO_&KBJRy9yY71J@yi0?PZ2|=#nBvd<wy;N3BNR647K%3{%48;Y>Gvd$NoEm
zeK-@eRNdp9<?mf7biP{}$tG^DIjbpO6SRa{)eUp%=wFShM;WaX(WA@v;nqmhJgPA2
z%Z}B&zApryeK+KjbK*?85&$8d6hfbo*0$1L=8&#)ROBj}f{9{@(NTdc_tQmlkBMlC
z*P&s=wt<3G^P+0NE@9>ON?ekZ94Sbes(K@5*wJQ%I&-W&dvPkZ$$Or1#TmA4gd+;8
zEJ$IcDwx9#7lRb<qm~-PpNSq(grk&(ef_sE{wOTYgx;{^gV`$%AA*_$>or4PLF+?E
ze#yIjg12BFI*x*ZyRKIw5Wh=TL~Yf#yAgCh4xx{l^p)6>tE}y&yDBEuC*x`r_$9uE
z9Tq&4FJ|2k_s@n&6p*@K-W`#j16X=o@nhthv&;8d$$30!0&~OimKV0x_FBWf!h(%T
z4xPpJM@H#+b$?aR-4d*<pUgbaUMcNehI8~w>6cTl;3Bmd8@QDsSq-VZzA_Qnco^bp
zIhiDz_u|sQtBS!nliXbtB&^?ap4zG9c;jEqz{k*H%c3GeG@F4zf{SPFPw^~svNF#6
z=(|L1(H$j56EXv`Pia*P53j$<=M?&zk?WBOaXq5f-!=8iG~_=5aDF%P(Il6SWV^<L
zZWoTJ6ga<M+8?*EH%Hps+o$*EAeL!(8xo|xM9exwO3FADOf&rTG23IQ==-C2ZoQ-#
zS&#;?ZPbL7+woYk`4Z`Bw<vP;S45SM+U$>>w#m=;lA{-}a3sO%GAotr6XqC`aqDA$
ztMm#@@sT47pXi<hG$>zougrdDVAuJkdh7Wff{Nkl1fg{`WFkaxL5JE*aNdY}))r1`
zjN}7x|8<kiqSU%@WV^l=nm3$qb!dV`i5IiZiIh76U1#3}Le4<YamA{BmFY9_G8eJF
z$ud-q1e*~Y(_JjVOuDESrLrn{XDa2&Ga8gPn+Va)K#+W)gZR<YmAz<Q`Qk%cv^586
z;Ov%gX-IlqW<JvGNJgy}fU6&YgJwq3@1Prb2vs-SpZ5#vIBG{4@G=IhH#62rZ!Iig
zeZA@rCb73SC-=rKLy4zGbEs~gcm^wq3en9Td#h@5ll?l?6EQdEAoe1>*k6rm>`x0q
z&%y-}7T&n;IxxH(Q%j=;sKh!EBw4kNu|c?lm$hnnvJF9~iU@z4K}>4dBVW}YPVcr$
zBfJev4;T`sh%#Gwl~qvVs76C1nqa@yMAf@?+~B+)^Ax#<5h2V4?`NsVw!QBpLga6&
zQ4ws+ghcO2&N_=0#Ge!)C><Jx^kk@ecg4ybZQ>Wc7%sNa^0t?{F^Wts5fpoN+29b~
zH0u^$skqeE4Z2HdpEbKAt{+-g%k9Fe0<wFoK6+f|6c(f|Ac!AZ$+}>+A<-1ox=NO;
zjc(JKj+b|z%euyV2srvBm2HKk)z*u`U?FTZMbEn%DnNp1vd_Wi!!%}55UNx{sfau?
zc70|PYxCT{vH<8#y=~<PEaEO{%7sv{7fVue`d~YAZVfA&-ijO=c#BHr0(Dw!RyuD-
z<q3Iv#P7*j-BQj<N+RkMOF!R*%fdo@NOL_uZXQlXmEDDxkq&1}yEe!0s$phEjS@?P
zlbm{<vbpDheQ2xUaM3Y8d7Lpsz+)R)nWFb8oY*``@HpNAw%R&o3ma;3`OcOUXSWA^
z8z7FF;`1U}K1`8*f;?U(ZakhP2@f8nigrp_-ZBh!uEo<oF|!o@vRrmhHKEal5AKc?
z$(Vkr9AEfJ-%KBfvG4Fe9%QAndg110$Y*8AmaXD72B1<GpvkdwvsSJKQ<rafhG|>E
zy7)j$tNpq7H^BO;rDy|JXAA?^!v)W_A^u5AdYx`u&m}^}gW!iq9Y@foT9;}qM&g^u
zl?<zqN#g-m+X8wC6v{Ph@`W~D#XtXFP#Ww1AEn{5F>(AKl*~ruKNuOg>whq^t2mHK
zknyR|LEUDBa%H!mVMjc^00H*L?dqYJ{rU?4CSPo>*;W76x4VX%^0v)1vG>u{a#n|v
zUb@7KlN0pK%|i_*o;24NV+rdsyO*aSosU}ebl***N{?mcet&|RO*7xBsj-jT!b!IU
z59nvzC)=Z$BF=|<0}tD0z6BZfQmg5j)&yJQRLa^Zq7#&F=FgJzH4;*6mXj9w2InO#
zy**?J8iF<F#0HDawq8;ljWeiVLq~bY<d}~dxEG>&*AF|)Nu8yePF#ughb5Vu(C_{e
z56I$a53A2&Pvht8z}3))H}?&UFg-4HY85Nj=IqB$GY2~B_sDkg{fh&$bVohDjEY+3
zqJvVp?_y6T{;~J?LW4F~N%e20l=-kLn<A3bv6FW&r6yQ8G))Qv<d&gUpK)0^eQR)C
z6MI3xumC1hB+@Y?D!4_!II$2B?VM14sHV1CeTOLMxVN(M_x2OoHMh1thL{nB2Q+m#
zEGYNC{2{X-sJ$@Py*>tm)zgMXEvRfo@zq4>AqHM_GvBU0_SZw`!viG0qoh`ryP!vW
z!K~oF;9dz3b;Pt!5U2i^7<G^MOrs9C#<0}^H3Iu_B7!cU+@;Mihh6tu4rj2C?g%rP
zVS9RUwF~#-N+;}~4#f2D_MeghtuWt;3oIvn5C(6s<)uJmML7iQ1ljQ+IcJD~SrWoo
zf3nK(d*|SjkDz#rWWCMB9V-HJ;rYaJgPc1$m9v}=xRdYN6&6lKgL&C3Pn-$~bFol9
zx5|c2MJaNzL^KrU3Tbz6_+C8}hi!6{Ju=Cr?H-GUSGg_TS>!mMJQkR8$#CAe&(}^R
z$av*EZyt-#c;z(Txh<|Ai~haJ{^UL%IF(4{mBV=Tm<XCxBK65>Sbv_`qTBsmxM$r*
z>E`uZZoj?uI?K^~o!bw$(sc5Pj@?G6+#2OMJmdotoi(|G3k@mt_%C(x->iHl_W#lE
z8Sz;-=sEtkmCwq~!u)>}Q?xRuva(KR2{O8{WCs|a2%q7ZEyy|TxPj?ke*@Dr3{DV)
zqE3YlE@ELxL69O9<)jo*LC&X```_K~AFtI8x9sK_j!&Q5))(KL=}8lP>O5!>j=m)Y
z@vOgf_~Kw?6;{?1Fd+WIpaS>^0Ccqw*r9*F8GzlV_0g%ZgOlFv092r%qXsvZ`q;T;
z(SkuNJ@Eey6!05BU>HbX!2cl!0+RTJ3L6*$SnkWQCzn@)FAEkNr~}YdLEZodc5@ur
zpZvN+8v}0#MNUacIr>F`Rdx+D*n_7>EQ1>17~Hx;A<zezLkIOAwCDX5?<YHr8q7Bd
z3hL_W0_nF^7eF47*hl4?gNkti!|bo4k0OJH{NkXOhr0s%<VnZ>i)MZRJ^IsvRfz4N
zs}Btw#)iKKkB%4_d&e(mfR5k8KKG9~m&Jw<_yxN70>AnDy?~KZlYiYh`xW?+1`zpy
z1`@z0VPQr+`4eUTjj;!TE~~8}3VR&62ksX?lfy^%7ui4T&(UWGy7~+Lb%717ti}ow
za1Z>00SDFJH_u*$j~esFQ7|{l{HD_r7*tWPwWTkPje6I}$HEQ<)^FDtJ9nGtB2v&5
z=j(?WV^6$ve+n-T*Az#Acyj}|sQd}sp9$c%nWfL3$1biZD2NRb)DCR018<g}doJTW
zSEnbtpKJOB0_bawjCmG@2ZRdX7PQX?z(>7?1U6*Atq<t?yL~^4!mp10Cs@dyAN-#-
z5AnvuwFB++)jfj83w8t31hn5y1`YD<{qt=WSc--mEd2Hb^Xt+hm_ZqWnKI$`o%6QK
zi-MBGHYdj~j7g49Lko(Gikb)@DJcmA{F6Qc0R6iNc6O^SQ$z$w{@tnd#rj=oG|LCF
z_em6^)#rDzAhfRm1^n<W<7gGf{wKf>;<vZ_hw8}B=xvtz7j68vk-!BG<RCTkHucx9
zt>4x@XciA3rwNz63&2^y9RvRt$rARHMpFkJ`DXtmS62oboespL2q|akKSDl!{jMv!
zD$Wf|AZDKf{ONAeyA9hn1q%o^1QWl(KHn{>H}an+Jl?Jz{F=8<;K<WsIy7jyT)&m>
zz@WT5J{M9#T0jsZdVM^AKXPOU4i0&_^ty5KX|Hsoz`-E#<k-9*VYv(dXJG>XzPOe2
z@_(ZFAd!*(jgy|e!3=i-{*nXU_vzow=t=jj{;qzc))>&+p@ZD*<3JLA&K3U{Y&ujk
zR;Ko)?oT#8;vn3o`0O9FVJWry&Ttl`V<O#9zRXdU<TD4pniw43ereeG@!;OK_};U-
zlRn$R#_Y-N7EHwO@AQqgqRG<3l1B_%C^OO+#+!2W$7N6KL=0V+CcJk(%8%R5Ctn;7
zM8aN8@SMZsVb#6GWl$xfpcQ6q*9#z<j1M!#Vl_3D#1s#!7kj0dXT<2ji}V}CDsL2|
z%Xy<#8#00M`T0J-ndoJl>GrtFMGLE-<2<tH(bQb!5BZvY1dVD=s|R3F9)wnyP~5Mk
zM{W(+PQh<V&1A`(^m^ya(Ot}+6_sL2=i|L~j28kTq>~V!|5cnYIXJW?y~(j|E05RC
zxxm=y6x+4#K&Yjae>CEEv7O@-2Xm5bPBPD#nd;;kO3Y70a*FK57v($iam&*m>dzt?
zA2OH`9IS$WYnzZNnLgM5G0u^evo7VW=V@(oyzs4^(s?-%m&iUa8wxMKQk<r}mFn@~
zK{{)Hp~#ZpaTPKt=kT|4<H;UgmD}Jg)(!@+|8CeJ@z*qv3~5uqT3qjsA1BR~(83(Z
zD7)ds{8)#Z<}8$wEIAemeV*lb)}vCo^+7>$k&?Ury?lK11*gz8NG0RNM*MkQ56Kkz
zV`cS9$znQbZcAaPbb!%F4EfqC`}(KW>n~{3tu6;eZ5GbO*2QY?Z&U5IgX^x*p~l2z
z7re2JGi5Mr2k(*>>_0C*e@SEUAk}%+Z8$4$Pc)snZ*rAf>y`Y3W}mH2I*-mtSu>VL
z_5#?!rA$eB=zSawCbwsXhZQJfCT6dAE1*QFO&4}!=JjmP6VgpKkU?!l{lm4~hoV0^
ziX2b`wN}8hRN!{|UBjluZs!NLKd{q!okAMRsIp2yskpqcb8D3`Y_vJO;ElE~BG3d)
z<+^G1x|Y(hto?ztX16TdE&Rus#+AR{rFXC|*PznX@*sZsbEL4kMgdh5E@;cPdIwNs
ztNB4rWjV!wnAFAmY!MVt;tVddsAX?HjPTn8h$^OY1+WDZ2}=^tw@dKFxG9yY5#E|r
zN`u=#8^{w6z0*NYdj2OW=4ZiW<x5>XJ=xTsvU~vz+;S{_GZk6)eDE(@`;e`RARSET
zz=Q5&b6E{%8qLcN7g~m@ZMJwv;k|j>!Xd0*>WaP@Q5b=U063UZTgxLdrX$Na>EGN9
zpHU&ljT_+sA?WNR_8v)ndfARuBQ8a2CO;!cY^paBIVLM1KG|oCH{irHRfPlr9o|wK
zlS|eCC*=bfk;DR2p)f~?Q?}xUGSfpSS)Kj%Dn{|V2Q=}8qk02bTx9jhN|f$IQ7N(<
zWj>1W5D1-p-0|&nhnysfeQN_sl;Eh@q2-KZ?c%12lJo-)i$(|T=Wy=5l(PCl;0>|Y
z(NK{`HHA~;aU^e0OhQvBn&;MdW3H0psr7mZO(C9nX&k(BN-^3$jJ5sC>=UFbHyFB4
z+;uvqIjqbaC`$Flv}F)6o`xIHbj@TIsI{jBB$fwqR}0TR=PWspZ}R8jx|mCkKfYd-
zHGU+tj5pt@!9X(ac2ZwI*eO~20z;w@=UVyFG%Hcm-Y=9iT~0I8U|*b>TSFwsgj-5T
z&db^lyzAd`#@K#Sfp72U3tR=mHkz&!Ku+=(E@1~=OE<~iK;ojicZ%njbRW`tD#Vko
zf*W4<Ae7&{!w<WaX0WuE!%edsZ?0BC&uu+s=NS{cjPKTiu8cW}2NP3v6a$-MAf&n4
zA^3RB1@0}Yp&WFlp?>do!#?e_L~Ipj`0JjhXBcWy1aTzdJF|7|+4&MG5DMh$7PurR
zy2hw}T$+ARmnl4at;*MvX->PR<Q#eC0n#gCz$+xj6}V-4;aid8UdQhmgn-wyo9?rC
zpJl0|F+9z*A`PL31LLxmf#;ib)7F2{vY4Gtqvtg*rBEmPSfE{^NRNRobrs@DkEcB9
z-fCvsT+Vl{vVKk3C}oy5imk#2g}?R>eFO-a0|W{e6sEpP<itKJ(+CRI*Dq_u0?cFq
zqf0oAnp3EKXgVwrv^_*rJuU)mb=9uFNKA=^1S?Qj^=^6QG&Ous)Z0xA5DX>r3kKfr
zG#o<xIo|7C6IIJSM9bcEZhT}cN;`foHO$KwUgvvy^a9S9#iU+)%UjveMY|&%e)KDt
z;<%9<b9}W+Xki}p41Ub-f^=ZRLsGXe&ZK~6q4eGbQdhz;D;jzNKg$w?K;;!Y=lh$z
zA14x5xSDcnmzKnDT1}C=8lL6@jNOuh_}ZWH_^l38xChv~2lk89*V!Gyc)v$*gNFEb
z9u8=?swVE+oRVy9Wq$mjJk>S|xHwD>Tw@8-#t>X%S3=Asn|`BBqnZ?nn)?|pBZ&(g
zX-a9e)C6s=bb0nZLL~4H26atOehTGEWvSV)j=BOH&E20M9)&nFn|W)DrM3fhvJKHt
zCJM;WRug2dOL~Q-W4dK_E387B^Duo&x(Z12sb=uu0m`cAb72tr*CV?FT}th|IPy;9
zRG7*r#e$Z|Lg@)W{TspWeQpQnAa3;(j(b|1d1x2x_E>DqlorSRk&)6JSz4`|7Ot--
zs(RDsNWM&a*<4sfkiQ|~1<%8tt0{9d%sjrfgFtw6)a8f&eO+2YC4zDummN^sN{nzJ
zaOj;3;T(?%>l7@R<%%N~KtX%>7JYrm(teq!3avz=a&0iv<>i$AYFQ%7E)jtT7v-w=
zB~&F!j7`*|8G66%J)SwS`%iJTN@q~SC1Ci}3cJjVO)Oq}Dsf*GGhs8!kADG<i091s
zi<Z`7u3S+MH93_@#Gtk4E+<rI=_H`?`4Gzz8}~c7)8hQ5R!6OZ6Nh<@cs=ykBJs=a
zn{~1L9QBQA#Ob4FB(<Z9TRnE~gAPO_E0AJc;3$PU&Zg}aB~J4-m&F$vVQ$&|or~VX
zt~57}+oA=4J(kgujIOxj%~TMP@mFNrxYeJDF8-ZMS^gS8eC)Q%y&b(6tc=!|H2p^#
ztx#>DqHAR}37{oKemIGDh?g%6W-{D5*3bq2+E)a*W@g0wFg>fs?F7<}eR=f*f`Vy_
zhwiG`B<M~>ji($dt4nNb!*}!Yg=}jitl~dwplX&aZ%tHeMy3kVk<d=zZgFKDJUE<)
zEfaa=K=Re)HZ{a~!LGLH-X=U`_oyG(!o{BsX8bm>M2Ixxu3MiO<fTs&X~E^2P}Q=d
z(v8D*IQ_5w6*_7wlVfxYErleo1G^bzwn?G7>x;1XO8n4Mn<jOd?LXp4R{9#wdYZmP
zI4x}x=i)N66GR>dqvvl$=$@}1QNfXdf8ABDl&{$?GAH6fmaZ4Wiq54$2721-<34@q
zAqVL^>qsph`1|69f6fOUtKZWi(v6r7#gGIvi~2bx9Fn@uXlyAN`y%1UTv0jTZt9{O
zycn2l!p05P)5S=a!w#7TaG`~%KZtuUy~oT05lOiE5qvEr;dHv`diq|9WUoE_wyp&u
zUdc|jPOcv$GK8pWH;}L}=AGKl>LpGOu7bGqq%2d;0gTm<Dd_9aa<{_;J3Hd4%6;Oq
zUN+U7BAmJ)8z{NKg<8{yH9TEm)6(Dhe5e~Im?`0T#?m55Ax;81-1t~IwA3vb?8V)_
z4?Hfv40u5aY~?MCs|n4=SFY-O<^p8wA*Ram>vbtoBMG)j?`G9*{?s?>UNHGvD9g#g
z)DFkiBv~nY=p1NMIy;`{)r!{6Z#iJ@jW1U5iK5skf5aHND#+)IJgxG8#y(U|M?yT_
zFbJOC&N~%nvRr`cwrE@B6=t4Lb8^_#mNZotNTx2^m^j7s_}=mp>92oIxMNavh)44(
zQ3**Vh7DB(jXocK5EFpH<dt6NccofLo+uI<WpcW_UnF~2Q6u*!+4XyHGS$BI7sDF4
zs5b%o*|vwv+;mm9ZO<Rr)f5}|jHP@M^cIlkY7(isgjU4mke9C}CCUU^@%JYR+D4De
z?jJ8+z5sJlslmKI$hm8m!Gb#nb-hMe-i9J?T=M+8#vYuYJ~MMWdh^@o^QrPA+kftL
zIWoHpm1a&owFAWj>NTOT6z-z)p+^U!pt$itcK~TPmg&VoT)EKAuph62N>qc14Kb|1
z#(`Z^%35|$t^48*{NOmoxn6gb4kw$!HpM=xqUSzh!aKxW*O^Iw(5igTOa@L-3gr#O
z9_)}(wwFXWhpdJ2GPb6@M}cGf#+WPT5N%=2|F8s{zb_BW8jEfsjkt8nN-uiduL&nv
zbZjw4*NrfD#$*j@k47Qio5GK_l4AwLgKIC#r($Oz8nZO$+K(FU8~R9|FMFj5t5}Mz
zHji^of&*%MJfavG8HFu}#5a}*a=?)~gllbxLvQQXUL7YTvXu^I*lF?3N7r{|=rt;Q
z{`i|--J~uCw%rP98mFd`Ekeo@?5k@mCFA<*N^<nlvE&>I!?8BZn;-lP+>f!sL5y#R
znh^{JCG}^|X^0%)k_h{_jF2IUZZCvS%f7lV-w|$9%6l0}RiryYxwJ#s==C$gJ6p5T
zf!kQbdk#%Z(s<f4PO1uh4XU_A1L*t17eb$IC$^tmoT0rRFt>&&EidGb)qfE1Swhw0
zK*DD5k+}FV(UjaO&eC&+9`xLL{jqIert%bxeAJ}r@r-IR>iSXuM^Bd9ss)`noR<=>
zb5kV&Ht9s!Ix3^>2#}(I*isU}dELLkg=0u5mC!QMBBp);T>GlH0upSfb@aB<kO)Wc
zw}PFeB<w0g`=d1T4P{|r+=f>EhnJA~i8U65iK!uDv}pA!&#X$RlJKz>k}v;pMz`xE
zQQrH*CsBB#fYdrE+{x)i5}nntUVg$`RLT}x&AaYj3xh1hgbfZec9J?3XY(b@4M=W?
zc;gtN$1NI%fYE_cTJ;8o6uDT5Gbd(UGCM>zd()(Sh8DJhS#d;+8UVJ0E}Oib11S^>
zbMmLS@BV_6#o~F=V9rDbF8ou;*Lsh_5wQjEVSo^>BOYo13k5v0#g~zm@%{vqLoU11
z#F$N$Y&I2bm)UrVGbwJN(EL)?P@^K60BgJrVf?n>b2B1x2ED@J;1&d)XeJvk^$dOG
zy!9J1hUhdM-S%IsVpGXf(6Y;=I-O5@rl;=fDz6QIgn1@m{E^adlr{(I(}0&~im}8^
zC|u^b-`?C~C|f0F3h~KANrz)-@;bm-dC9?`D1^+6T`a{1i{UxhbM0sN<{M>B`pGEQ
z+HV_v@8s*vjl<PmvwiII-^aPCwyYz@?~Pi8duBRgX5rKy#*C4&v`qL7_qLo^G4=@S
zMF#eJbfpgs4!<Ig&#C2BsVDy@4M6S>BZJ}cfqcgWQgJKrM+rg`yo=Lk?aSba%QV5@
z7yTL?m`Br!6^--<lUOe+wpWUVtwG%*!x~*~yh0X}+@tp^rj$|z@_gdz7IeRMg@J6(
zJ(6pZ+f`iM>zh~}f=t_6$-(!e-UQhioLp{|xgy(+SL*t`mLrJ;p*nvR#Pf4EM|Cf8
zw=|2W`qtGKX!^h$hmoi1hqk|$nsB892m87V#U=Ut;}?hFNPk?#SR_0#6!5Wso&DL)
zv*?Ox)N7+vXIHE%WS-N<Yt|;Uy2ps#W>TEkx78YaL5rN@Voe7^Aa4+|nd5`f+b7;j
zLPgl65=UcZQ{fpH5KNdNWI>TAGA54_XdE|g9~wk>&0xe3%&RnKL(`e&hhWQ(VtUc$
zQ*5?`R1mgfFiWR8JBqdm!GG*6m6+-nF08ty<S7u}3(jaI`^&+Hk*XALT_pDC9EvCH
z4h=?<4_-(hLs8BFN?!r#PxG01m}1s53Ag7H1Q#w0KA?i+20TD}WZtp)UJmJre&y|3
zAzQNgf-YS?m_1G5d-5}MR#Il$`~v%t*^oEHIM=DS__KCISNjQt5)PRe*bB(r`=4C4
zMoex}!V(WQJgH2*EKB$VDdFMW4kesg>382Y-4jlcyEpcxrL2%^d5~C+Z{J#oceoN#
zdwN>;J+8^;R=ReKXbg*19huS$nJ{O50$3C#p8P?!(NEVrS0}sH0x~8?2f8kvbA9x#
zphz#7$`x`L^}Je`ex77KUZ>!XQyStVrY)ONfs6ddvSF1X;*xn!^#~Y)u~C>tB#$F>
z&9~gT!YKR*FC$|r^-p9<b-isxWXD%PPQ4oMIywM9sDnbA-;b5l7Lqs29|Avz^XEb-
zTtmG%tD@r0rEYz>sb95uY{nSHP<V*+i5*B!!vL7h;kEL-s!Mir<ZrNWmlsXoag2{K
z-_@2*!C`%=PB=x&oH0~^_M)B?8?`XZ>|B;iwGcH82Ko+rTU4AlN4{ZR%I3O!@QF$q
z{7WLk;AOkVu?@`dkM=HUT8!&km`qO;+>WC-7qAqN<)Bs9TsBZLU1QgqYX|Lq>Kgor
zusbkva4`(6lG_%)zKs}-2C}k{7V2DQ;j9gU^Q+c6-m3@904`JMGm_R7YGv)Z0tiRx
z#XhFI11Pp3^dsjmd&oW6aDKZ+_8%w~UB<d*brj4`B*>NvG_O06D_mSHnR1(S1p@e8
z)mVK_8lQOgL(5w-{AWYyDA;$XqvTl>pGGW47xJCqcVL6Kxv3m@m<3Q$<qe+_QW!(2
zywwUNCa!K){0Zfay9W0oyz%{8R~Lp<B;LAE$VA*KRI3?ofRCX+4m7#qOD@?F!um3?
zW<krbUFg2!<i;L34<jk&Zjnq>F1HaiC`zam{^al2gBCBXR4iau>v4FIETepR2tEiL
zm}51ah=%`dwjS1oaV|996l6W1GnqWND~DyIliIptHGIa^j+oQDYRH!l?-D*uxE#eI
z3B@~4!%FG1Pj^rn1efbkls1)WIcgMDLzspOWUEjYWD_HkH|d)K1NJOz$ti<SE;2KA
zP7Zf*Q!omx^tcil2z-x`ah?mYNS3%!*2g%$vz1pr8cCs+`Ri&tfyre)JOe5fO75Hx
z*10Gpru=ZP-pdq7!PGeE3a{*Z8>s7*VtU$-<8k6|cmEJB3eXn}bCdQI1~uQA5TXk6
z<)AR_*&AV|k3Wy+Y5YF4R31jKcSVZSPUS*rM404b*z2u<KWk)Vwp%L~0zD%#(s<m<
zWiRXF!s8HxoC2?R>D#nCqA@gx{>&h}`YhnoFDAN}Iq6Xh+B(}Fo*K|-@tAzZw|`X7
zD~v}fR8gR^rR(cdJ*w!p)y>Is_p=_jbsPe^?i<nwl8#W*GaBSwqcn6uxg?qNQd4^3
zH!U049v_Ay<JGSh6GqVbFg#R6xw^yk%Ci2n5IcTtm*~D~2SSN3%}k8H>K3rz*9Os`
z0z2-UBDnx@sVCZ?s<P1#6@?mql^p-Z2k6TKglx@Y5nF1GB%P{D8c=9{vNCIiq{~#N
zh;K{6vgeC_H{4v~t=?>mnnOyI0H`X8(Kw}0M}u<XT@VczA?CP9aH9Jd@1v2YeO)p0
zM_=0N67$wAT&6uKUqxa??~6u9p~022ESiLPlLV^juop9};NJ2Dl(Cw<_2n7rJ@MbM
zZx(zlBkgfe%%BT(p=48egZaBod=iifhGX(WG50nfN?MJ{KX!;ItrSv7HvEX(9Bw*j
z;6g2dMDU8fl@e~i`Ohd}B)twBek?FuQL(hiN~XB3t~TJ=5iB~9u~6!kR4eA?#cP@H
zB|GV50{8yW*yz=^B5k=(j>b7%=nBk=Ywfj?rhu#@=F-rT3v-N_3rWsSiA3XcgHwYC
zG5;*jJW|B7VpVh<&0hGIEx$XB3cU2Wm~2!h@AC_SS$oO6BXKxjk(UFuBl&>f#+k@6
ztEH%hO2LTBT*u?L(4uOHVa1H<3eI8joo>K1?8S9Nv>e-E&fmZ$B8eD4jzqxY7hbb+
z=ZD=~w8ZPC+`D!(ZL=l&1z#LhPjthpV57-rZb(2glL=Rl<0<G+F?+c>>nSC~Y!?cJ
zM9Xh9dR^I#|IN#o|F@pZK~MjGps@dN<xNibu1mGLP=ND(7CvG{T4>-RV_=RaV41tS
znwX&@h%ghO7P$A<*hZ^Nn|Pt9W&3t6F6<DuTwmT9?_R$crbm$oz56ykIh`z)-ImaU
zo0@I6Qypx%U85@_KOG)l6uwfQS9H099uIT8k^LtQ627SI&Gy*Us@Oz9Z=8{=0N#Ti
ziXW~^*U{JEt9S0M><|ai_)q4n9mN^3NkW4IIO(|*zXNCNmqWGu9&rk!I^0jw-a+V5
zL*IRZ6E#!1!8JKF#siEc$YK#b=h8{EZ`V#@Fjl@EJUNF~`kt%EW5p@I;^9LJPqL8Q
z$oJUt*hTuDh)=$>;^fiC{X5it!7;;TL?3Sn(8a6W)=u&hJQ#N=4^KYf{BtdaO55Gh
zLxas>3$C6KLyBWbYPL)jgwdZ`mTULP$4Qn}+4cqrK25p`m3n2);R$$oEdY|g5dd01
zbpY4^w`>61fBUqU0hW|#?Ct?p09ZlT0WvW28dRZx|6;0xVFOf%4Jk)L2V#a;x!;ki
zL{L4YFb)ql2#OA4V6spL)kU=aO&hjP@x+3_MMbtA=7$0BTm$_lBkuK$bD;jw!A8Z|
zuq-zL;s&^F48l$HIuU3DVgAEvg(*+>uVelfv)@bx*fPjvg*i``sIdg*#PUgI(RpoE
zse1W2sq?pGaTs2cYAHSf1Zt0jWte8z`zYcd;SOfl`vFE>a?*<+&{zUFMsz`8vPDEX
ztCD3T2}?DcyB}Nfh@2NOk~2bu9_9e_l;4eJ9ZFuEDn>!fDyCkIDKr6Exlx7Oub<)A
zD~6)?xO=Z;6`KzJQG>#sX-N(9ie-waoo%>^Ien;3l@Xs5NzI>H(?QxroXUti*DPtz
z)Lm5K-C8$x9cC8~0vMyEn=p%m>W)qWDKIPIk)=xh{(CvC_`pgnu<`iOanAy0bkDTC
zR>;SzJTN2WGa;Pj)rLY>$yemPN+n!lYglb>Pn^n5U)OCO^2ebn84O(QA^UGC_Gbib
zZ>0wK1l$!vRy+~!ilXc!!_06sN6BT|KDzhCF83kH@;+8Y!*BP@&nxNYZtD7P&&1Cw
z>*sFgm7kv3FL?Slp7M*|-1=|7@oy;OCtuLj@0{^38|myqQY*%J%3}2hpBuMut63qJ
zoRX))S#v=uklhgd$Y!AH8eU!G&lyQqPv_f5J;e`rO7gkte<|eu=32AR|BqhHi2rXB
z6wChx-yCeL|5q46bQ7p@(lsh8_4YOi@V_W}Hz_yh|1XOk8iuf#c|kHTC@4wXO`J_!
zT&fjl;V%EyYsPEl{kOTtZd!A8%S+Ge)@!ucPPF6HGpoT>9xDzJ8R+!r_y7TT%<3Wn
z$ld*|{og~6f$`jLk2a^LDJ?1lKLrdO9QF-P3kM9iAIk}f|1~d-1PNH?(E;H84%j7V
z*fn6t%?-edi#zX!A!uL-0Gfxk2AekyDjz7wkDC_tAkU_UzSfWM<NhDh8dXabpth`x
zeiYB`Zxz)auoh53fDHaNqOs?DQLQ{EemF}guzsIkDv-z^BJ3N|k?HH}>uFHu=e>Xd
z>>_mzfF1odB7U?ZzyJ^bD}dffP;)>X0YAvY0b`Iet^l4N#Y=(f{T+VvKX+yJfSUpM
zxUsM~@LcfFckXQbaw@=i*I)r(CRO*TfZ%(#D**28?!P!UcDK9!eQ|s<{90ONm00ix
zP$BF9SN*j3e_Im3(6+#5009KAe6fIZxbtFv`04Q1K+PY`@5mhR`6X1r0lcejZ!fs5
z4fNUCz*E83j@^Q{dIz~#H4qHR;9Oil1@>^$dLNYl2KifJbzYBum{iUFEMXn|uCC)k
z*Ejy^1Z0<DwE}`%-9RX+yloyR1i#150D^%3dAPg#6Zrthe*+K~TTS$OyZ6y!o5zQL
zYxivQ<NsqT&-VA*2qKTaj_2$iZgCg*3>2u(ASaLi_ZQ?xYvAY@bj{yp9{?*37(L)!
z?n#Vg?1%2%;eoymt}p5|-n|PTXSdfcmm0U-1U$Iy?Z@cHtZnO|p<+W*^2!hLhmDvF
z{2B1g=^+w`{T;*upo{xU5MX~)XwOf~0~X{r8+4Yhrvl!Kod2oT7_alGHmLh2hj(%}
zvLEv8R$X&WY$O1H`WJpHODt~m<$mz<SNf5U_Sf&~2j}=#@a}h7Y<zL}7;fSL>*qI*
zU#%|kcK3})k7qx2Yc+`17U0NFCYs-`x{`DZ_Uinjw~P||tqr&V9z4~L4dE&u<OOh@
z2Yk>5=eJ-&&ogir>I5j{Z;lX8UoRdq02v+J?5}>TmdWMUo4Gsp!#x~`r*_6~k0OF0
zRP$%Ij(30{06sL>)37)8eIfCmyHg)+A+EStJTTz4c6|A_La4p^F8;qf$RY1uIfUK-
zw%fZEy~Kb3tv|gaJ7O&kfj>gXAohX3Ldc-@oj*hm4*+chzXZG=g1<D7+#9?-zQ2Hc
zJC`qb2S(<xepMm>TfX>fdN;o+HPvL$FKt(TXK!nfe&c?+sPJRp#*rMqD3(;XC(1}G
zjv6FV$J6Z`lpMot7EdY3i+;E(7aq}(DFd467!5$5B$<qJ!M$%f@yFVx{x36rz4fTq
zE=mcqk1~DLDgie`rc5VW?B}WV0Y2+;-OT7qSs#pAfNnQ{A^kJr(SNQjquOtB2-{{K
z!)d#{l6-idQ+%;}pIwy9Diy_XE?b-vm{#ydxZB(_7-mVWLb7mCW+Vb<MBNVp*?fMi
zBIwPo{8efEahUG#Q+nrios;QqEJy6hwR2|+0}n8QDL7yjIQH*A5_`8~D=^oqenSU_
z>700@5rdmJ*s(1lv);71yfKpXx6Q?U+?rSKys>UYZ6Wca(l4OqMHO8nc<NE2bTf+@
z)+sao=uY247pz+}rfONfo2*&fCW&_AbH1Mg(i2IiDn}>cobxmoQ?K|>=cX#<o%nDo
z+Na&As$?%&T#w`ooTXgJCO`b0T$l1SUmLTOsN-x|+s}+Y_sE#1yWT{LM0LJ+;{D$C
zaf(NxpZyp1JAiIc91oiRP1Q`z_&P?WC=SYZM7=AwcE~7o*G#fTY)!0q=_CA1G%c8N
zVC$^A4%AWu*TnNHe*08N#qJP~-GgZ^+SbRPV0zM$TxSt%798yQMwBaVx(ae_pYw*c
z&3_m@%HB_2eHuBeVSMukK?c5>8?V!EKn>5a%gnkN959$IvNlVj@uCvB2JQo_?){jo
zFrg4yQ7gPm0i`RnM%Fd4ZP&R}T9_oI7ftj*dgaiI{n1hv6Nb0X7oq+b>X)U+R*bLB
zLe4CM1(kNXS#5Py&7Il1=Jta_?`FnC>X@3zO?<wOdP<q1sANd<bSib((<8w8exV;$
zIKL=E^$*kDSHa7HKjp^h*1r5+a>xC=SeE<m%zNI09Gb*A1-pUPK<Dg@6`Nnza>{%c
zg%6!g4nxp{V6gypc$Y3>YRt`OS|GPjc}=lvrt!9tDm}*AvUnLmFIlQKD$ffCN>AL1
ztu&ZVQk{#_q+!Bo6zYjrH<3|#uJv9yZj2Y+vR%6M#@16gj{0+&z3#mkEy4F7Toilv
z-?pw=Hd|;9$E$b$Mku;WJ6egfUgcHf^9zj=)C5@`ypJ!~H_dbJR><h$YY|l#k83(@
zu&t4pgt|JCg9p27gJLH~YNU!KX>&#$5Du+fF;;V6B97Iah<Ce>30{wtJxbSbY{YSH
zbaiGG=_6x{p4YiW;^*x{Tdbsw+quq@P{GmqH;-_x^~a)|3?zZ387BIA_@q0HWi*p1
zn~Ir344JHLR+-<b6xi3dB=dF<7ucJkX9u#FWS2EAwR=#&SvGzxM`OG-v5prKXBZCa
z_m~v6oTs*0h_PjqmxG)`Atf?R>)J2ktY@7KqhboiRq8X_l$~kc#0op@AK%?qX4e|p
zDFW*EZ{@bcB4idj7x{*Y$}^Vi=V%J0906Rw@EnPf8F*yjx^=xkhRvS9=|XK{(6-8%
zI8?5^kNRiax=NE*l3Bv!qVENKg}R0+<%cqUYebKb#vn_Jja!OpcwIqMj)p-(g2b6u
zN+p%XSwvBal?9jCj>mkuFzs{JoL9oA8$3$!kery2gA$Kzx|ex7o2=SbXWBScRXD#t
zsQB|6_slHSi8BL>b1sV<U+WJ$_YH(;mU9x`PP;EtKtm009>iIOjX;NB2)m}?gxG6Q
zGd1+u6&`GM6f;c#_s;a%-*rzvh!P7#q16riV_l@=esBa$jG36vOFyQ^I+jwlrI&q_
z={4=G8E7MEHJh~+c$w66x;VgD!*vgYhA{uOs{uT#uv{8W5?O_NSg_hwMm%1H4snCv
zAsE|SN8KVcOfWKsT~#m5E?vULp4=QPwZUr}8Ue?4c%P$bQ*oYEp`%9lSd-FifizFu
ztULSgEF|P!N3PrB`6d9SbtU2`_llA}wEoifJ{(lr4<lcLj-Y&GD3e;5K}=!~(6e&X
z%hHJ&hZe)j6gF(B<5Y6D`<$Sqj=_e4jo-g^${F79K7%Bqu^POzGr&`vgxroUvkaFC
zV)XBQCaL-$SU%Cx!HSal<#Twde4OSeg3+!eio`|r9bRo`J@?!e-@eY7^1RNuWywFF
z`ZV4vQ_~1FFztx<7R92(SKs&s!=o*>RQbxCnX71-Fn6Ezuq-r>87en`%Zi!jA&&{%
zhqY?jPEr=qo$g@O?Z5HXi)%)F_>9u=c=<V1PSY1n?srd~N=4Wasz#Gf;ALfvcpoJ{
zO?vHM&9~lpDwr`8lVBu_gG_`IW1*m(vcdE@DF|QB<6zq=(53VidQ2HomoK(03>krb
zq-DR(JETu=QIDT!`%snQ@QQIJ!arkrRrE0$oRV{n+%scV@DkAfD0Xb48%uEGAj^F;
ziQTseQ8Ko}E~8X2iSQC9SKr?eFdd#>6PO<Lor50%r05he^Ombn{`+(9=~J3ckdI`S
zaaZeGGj7!mEQKw;+bV*v^F`|mUw4^T1IyJleX*PjuT>Ux+(OE&$9IS+D$QaIK_7(p
z=GWhF{a`1M>juzGb@Nj@9}KUI*X$j^d8a<)xGLZ!s?R1a24JUO2hDBd9y-BC8f?2i
zy3$@tB8&#8*?763)8KqQsVPMkFr^528jV@mqZyf)JJw6=mZua2*b((MgQb&MBJ&+v
zAI@D1K6)&<7GGzj>5u*8#dr5P2@5ZVvPv5cip;#)YTBm9=giPicO@j2&qJ=hAvu|z
zuqAr2OtzCEns1f(p(0&rzYTDD+5odX2L!IkeOR)@UEFP5mPq+EB_pup2r;-%KeD}t
z*<f667?a*Rn%@C-To)0j^b!axl~tZGp5l($lrXvylY%a_UVad>NUxiBew}};kQ!Xk
zP0`tVM+#v?*i{jd{Flh23Y=lPd?_5ydAgDRmV>aJ`3P{|sKBS}njE#CC&)zEXbph#
z79eM<&W@i7JbtmQRgsmkC!A9WAxk@rKGvlc00!=fx~~@Uvs7=uGtG~65Q(usSWPv(
z@K|o3?{PU@_3^@qYj}t9AUR(iGli+R9r%Qjzv_hRl)QJ0`sX~dizw$Y3ylS6N-I$c
zX95Q2B~1=;9Q3BIz~p!&X0Bnyd4j?JREfoRv83zpaYSraOb%e5o6ZWXtMa3x*|2!>
ztsOqJ#D-Ia_oacz(cDI3|0eVi4Bk~F*lS5S)cbIM0Z;u#3kfWd+m7cG126uf3#QXA
z7Xo)<B2BWBEE7eHGP5<Ml^7fFbu*jGZO`7(@^eJ}!A$YUJXl>2<&x{(R?_*72F*KY
z_FYJ`nC5N3cvOpVXwDi8YKv;<wDDG3_xsBB^`ej^${nW}!B6ee049{!!s|$VF5cY|
zlH)ZU;%DF|f5)=!f!dI~Pw4gpv=~Ui0Mv~<Q1ja`%CuIF)QfHr)1^H7N2;a^WVDLT
zVFuJUX-#T}C;F;At&k$Et-yRIDd2W1W3?Dau;~di9a;+ah8Qz<f(Y9l#b?YfqepEg
z0T<g@-VEKZQCWHaf=8nYdqQY|Lb$>zA-ubCfcOF)BP4~+g=Llt+Lu#bIen#?@|%RF
z>Q%`P0dKjoCJ712VYeI}ocw&*+d=7BhN(b_n8CfXzu>NLK3Y~JwN|AQKG>7?7L5u6
z5yj6cchBa_J$I2(nkp0%{84w=)BOm+M$|>yIVjHx&ImT_gYfTZT_9#USC}5&)v3>4
z><)*p16=kr%`%rB7ZvZ`k8@mtOidNkB5!}9acnc{*f2(aF3JZ2Q#?!~0v;zUWlZ3W
zCt0u2g9%cT6{m9Q4kay@C5uTDqN*@PycqPi`fH;49XZtDfOLdrv_nXNwhq>PUi1~#
zxV%|wG6WThT>6V-h!=~6JjW^)jv<7$0b=KH&wQ^%k-}H16?tn5=K0EqF+;Imn__h{
z1r8u-n4O5o@7NxBR2;|SAT#b;(@N?$KE^8oNApUQaF8x_u7X$&TrSJ3^UIYQf6lrm
z=Fh9D$y#=7Qno@!5MDpXfdQj{j;6RrVpv^zx4k7Muyk>qSWH$c0l0wAlY+q+?H=m=
z0xlD&h)?X%tygdEN%X6O(M@BK>{2D(;8o_?^-2x-pvJshxQBPTh5&=Edcv)8*L2eE
zZ_=&?SVS<j#__?JPOw*hvt0ui7x9#=jF~HiLMb`x9zRE~{Rf(UtvIOGZcXH{+T93T
zy2s?^ysA$M16m!do@AUx@)7B1dJgzeZt*vN0Yzzs2lXSC?a>SHa^6?bWO=#Rh;@8Z
z-}{=LK{kkRq4f?eO?%WaEDCAv4@K=O0clB_u251d!SyC{a{t)rzIwu$bz{#E)=14Q
z#X$`iZvYt%lB!t*bE`oJ-fgkFnq?%M%0_Xsc+)aFk##e43chx4Lhgral=y)tB5-ga
zQW8PSLCyEREC2eV*4!gTh878iCY1d`bqSz_8%3;nXYkAcjH!nXqcs-8+O3?7#;k!P
zm`as+a6Ybw*dSE3gZmQo;F-iz9mAd+i(E*q%hm9wx_ir!ASDrL=hi3g#0Q>)ty68d
zwHQh@^HLq<c&bi2OkYjDq3P}B`X-Ek1N!}sXGjgr+ED3p$xKMSdL%;b<xX$?Iz%*+
zxF$2c-Xy-wg0M)o8>iR#=DPOI>*ax_<xVy7VMx%9KA(&ANt8b#r{tSaM<j#?p2e$@
z6Og+)2Lih@7asG}-m#{}SV1i)(K`)>*zmXK$k^cU*gli0x4B6#z&A>}v1URK%}xuT
z4F8tjSrPoLQ7biPI|<dBrYJ|ox#YrYoO+#<3XtCAy0MtTFt+y}hP8{c%|(EW;dK&*
z!a@3#y!=?`D-B}V)446NjnCu(%M4$TkAOr_2nDAlF(W6X(He8rswEbCU;;$${g%-_
zrN=r&vZOoz!{Q^tmxE$N1<vzf@?GpY+;U!o@L62i+H5DYP*?UW9$O{Cbiatoi{g%7
zAadBoqrD;d1nZlYsG9aChqtk;>m#`*ynbCa;|G7Wlv+lUNKj7-?+K8XqQbpgY-U4_
zeVwXxp9l=J<~n%uHcw5IioRGwL1L?M?@unDv3ap2JsX3p#{EvyB~i7?pK!wW4hj*L
zud8^?VLk>ugOpbR&;0!5Db95->$YIT-YoB*^S5%=S;TJ{72OYwBjjM}KJMbyBM`c>
z-chb^U*px+ExE`XJ_=@nSR8Z5DoMFm15^#exf7baEtXHtWD{o_n$JaTlJ|Cig6tp#
zD#o^zX~g5?J({ukZ?M(SsK$A$PVY8H1b0_Q(qKp}_h)|Pjyr$UwEILFpj7qLv~Anc
zz(FpOR=e%GHI`P^ACbxOtD$e^%YBo-7a?}02J$|f$<P?4KbhbfBWupMVMu?zU9yF-
zdx4Tcbt!07nuw!6#+}P8@GG(16Z>BWP&J}0c0HyaFLT7^x(BlmCJ>7YFsf-`(%BI6
z-<Yic*PDA1#wG8}&B<cPz){mhK{9+d9^|(Kwm@arvLoZL7SJb;v2<8S{l~@+rsr3E
zA*s;_@e3c>Iw>MxHvTFv4T_>-q0r0^b^{46lNFlPd*1+O+^;_&SQgo$DO2VC(drw#
z*THP$E%@xBR+lyS;Z2z}nVU6>EhURi#eP~VgHwr8qO<PcmsTJF84E6rDN)_PV}*it
zL2r0n8uFT;kOYT+yWb8zS&9yHxsn14Wv~rpzULEH*6Z!w9`@SNXWKb!3$A^M=wKnn
z1=txPQo7WOWALv~=>GW&nzxoEHl1E~tNSPU8RwB1gefZfMlq8654{Ry*z%?Y6M-|i
z#Mzx@d$;S&_mha=%`;kw%6l+~j8(xW<BWXH^z?)~B)G%nr26Wtdfq-8;BN8Jn#Tc)
zLwTWZ+UQ8$pAGW1G#V_HJvKAWYsqpO&maBpRh_Ro&YYFc4z0%%awYC1{&LGN`(vkZ
z?Z!BouF&Hqqce+9GG(0AZGKfXu_)7m4V{7#ReKydYZnN>gfQ{RjPgVuY-9;~wauM-
zizb`hOs;jm+V>{>eb?p;V!>oRZ)yA)gX^ilLQiiirUdTFN1fAFp?}%lSJkMee=-8?
zsd!%Gcg=%_DhlrK6n~tQ$(kU8hc3RvyR(*jg{L0&?O=)65v|Y4r0i9$WSEV~bhp|=
zZT(0NIle|ra&bq&Ia-4qh1J-pAyfyz&~;h9AlG1=s*ez~MX@%rf-17xpUZ~6;%ciy
zJwJ?a7T{60_)cMnzas!i%@#%-Zl+4P99Th{q2UDk6u8ys<U|JiYi<PEoM_LZo#)-X
zE1St)A(py_Nil-Xg$br{&HXJ@BI=7UA(JPJJguu3H3{Pbsvh)u9E{_<a`k+3q|xH6
zvr!9Hp8`isw7e5ljGE`fCxnARhwI7(6{(z#Xj)y)l(irt$HiTsk(erFcSLMP(pI(*
zH$?Agrn(sV{8Z?Pcq3@h5?5j>jUR<9N*GB5j;O<+F$}XFIb@Pa5j0eR^)Z}s`c_a9
z5WtQ%WLSDvxHLaS+Bsse?FueD_M~jD?iDirT%Ob0vC4>jp1XIc`!h-g>m0Xm4TLgP
zn_0pZldr<V#KQP-xFrfekwqlUzPyq;Le+h@l?~QC=I*gR<SX@KvUQ^`HORipv9#Y9
zrV)r26!79}DMM{j`0>447Ol!G^lZp@>b1A5sXb#<*)6YG3a#-fn-7Yb4%fMJSZOBt
zzdcEgu+o@&+j%u7uaN-kTI7s&BD|hAW^fw-hM@D78S(lks&89ou*b7udl4Za)Gw9L
zq^L9kR<=ZKA}&%RK9djkz|E%Hr$QJhiFdJjSOw!TliUk0i%R~0k6n@1D4716x?l(@
zDf=Zs3GA?Ewpn@Fd6w<1Q4%a+%C!zzB1Lj3Gnrn14X)V@D&i41vZcH%4otlI7acom
zh1GuQikQ3$MMHMTUUGKV@!0kMsQb#MxPm6!2ogL%a18`^cL@Xu9^9S97I$~o;4Z;~
z>*DSnWMOf4cUTrK?^|!x{c``n{W8<jJ$>fXsWYeQbU*VnfLm|BX6^v%FXS8j-<Dge
z&(OBdR9s;aA7d=XI}fj_^~SDhtwWYhXRE!0eqr@02`X|5?DdWjWwX!EApgqy^SIV1
ze=MzlbuH|la`D4QN5-B#LpEN*=QP7qCX(b-jkZm+{ycgdF~KaUqS>J(ost84DKurr
zDcX@x;aDdyEopcsn2Zw1Ib<N~J=K0|Xa?(7)HZWCg<3qbH5~HMtm3%ZFve(Iux^*{
zR<}R`YvqXF4WfkU`SahHunSxM4;SRV*z2OLTIDPR<Qi2({bNgN;rG|o!yRqyjY5!y
zCuWE2+=oR;vkos-Z;@3x?W&nwX5|4MC05DqS4zK|{V7OZdUXA>Qk7wcv+A8MI{|e&
zxD@U>2PrC+k*FY9u+_VAR}AEVWq6ON_{YaPO7<D6&;wcJ*H_#&Fq(}NDfL~!QBtU(
z(~~G7nbs9-Ffo{9>vof|U`kc`?L@dDw(c(4-qFCZ%f`Lxdi^;4A2|&fB&4j)!Ucph
z^21NlcXlJP@o9mBd$GfLM41c7ESL6<*V9;hN2aZTF5CyqOxG+M`<RLY3V$Q7>AH>9
zTEe1T%AD;A<XH)EuO@cQ2A`OHk&Tz9T?${|fDvU=6!rQ5hKM)_4e=A*v2V>|@Wx_X
z__v7Vx>=<!w8qBOw?V#Xx^&`S@Tu}oQVYzPDx@Uyn5T~oWX>!E@o6v1xYJ$adwjeU
zn^6_b90ENv-4xR>)v;N0{zZuzea#Clx`c55rg*<Sa#ds$_YEbNoi2=)rVpVuF7#W*
zO^_#wK)!z&?M8ZU=Z<R_3%{23&oHlU39rE-U1)W^GC$KwWGz~V`#I7E%adWl`#Wmk
z8dR4|ox0EagDB3kGeKmF*AzqZUz>z~vi6lORS~X<Mx@RUXr2{$<`hccSk|^9ENMT^
z<{HR_OXRzV4#=SxTrH{r_Ueed{BYHCw#+it3_Gu_TX{md&L6>QPCchG4l$K3)w>S4
z$iiGq*C3QT_1G71dntCM6sFU}eCyrd5fU;@#PadghpiDqvpQUGUT)#vp^)CSBffW$
zSv(AY=gm*-V*PCvV?m#)Wk$X;jG^ON%+$4>@mraAXq{;3>ISFz+sU|45^cJ7JU{cQ
zJn0znntb2$sgt!OsM`+=8%%t33+_O!Y)@@L1UUZt1?>noD5m46#6<sD?g}~W5~ISI
zIiHl5wD}OBqlo&`PUVjcf&`1k_{BH^#l=+r`chX1liy|26awL+RR@_}W{{?ko>a)-
zcYSz2<ynM9!1sfm_Ailj(S`A#gR|EtUcV+p4g0FAm))I68;9=X!^CWxDZux%QVtt&
z5BQg-U#wwD`<^zuCazT)XMN?Y`|U>=cjpQk>qy0=)-*Y{zChV-jM?AyLmd8r>!LZo
z!9*Je2`Nb>)WvhxV?K|<EC0S0r+O>c+!hHU)9frNGWIimj(-{FVp-2nwb%KAPs{^I
z-kO?VOLo|-@OI}lemZHJ@%{eM^;z&dGpDDMLxWd+Ybwg~!kCrj_H4xUGZ(_33mS{j
zgIr8ZP|g!N<#bV>HRc|}R!?+OAuBt%aYxS1&#W@XW5cN-NMNZF&PCm?En<Axq)7HN
zVmIS;`+NT~4`VUs<L}gB|K<Qx;UsZjn_H@gR!M3`5+c9Q{fv_u9wlngZ{)5IK7f9Q
ztb$WxJAe-(Y)UTU>Kux{OQp>y@uK_FVGj`P4Z@^JF|nh6cN?^K!$DNaVxtI?J0@v;
zZ-BqMK$5WSY70?=k0fEi9>K5h{Jp4>b@wpRO`9w)*VmUyfhDMhK**B~aWpA6P^ohc
zzi<AIP!gm^>6;==Q$O#@T%fHRYz*ToV-)LFotV3~Ydi0n`HdnGQ$VD}SS9;gMmJab
z0P7zM<L5Hco21~ri!E@0sap<`QIfmIc>~>`LJJ=?u&hKNNzkr+m*H=-c&;x!n+nJ5
z364s{CPjAWTBNjFSt?D`)W&Aa?KAx*KOCj+Q72J>tF=Io-j%iTnlgu>1sC`5A{T)c
z;yHVQ4{h3D7`#yO(M_u*?`&F>SgjE5JC2<8a(<GZqr0(-BF;`Q-Uo-R8!T)aH}kf?
z3Ga%o5dgmSEj?g01T`yZ)A49b2bu-H?u9=A6eLu%5eGTalJ@a+N|)tg-{cQuR=G8q
z0w`BL&<r!5@lNvg@RAI_`*uAzqqV54#h+e$VQCjEUs|>dk5=%#2E!6tNlg@P2)2+H
zG<eXbAV#jOFsy7#jW~}Z%(hQMBMVER|BUojUvqARm_EJ~zWJ;!R){AZGcQa}Se58V
z47=FTNte5=#2Z}hI26)@axvB~x2I@G{qOjNZILz4c}3w>OoD1uB{*|bX~%c9o1;+H
zp{N$t<gO^OySi+$J1i?8EBdKCZ9@YE;=77$2PT8^&HA3Sb>v<+oMM9zC(7EnF)7`H
zFHTIJy5+~qn);q+lpHLR_e60P-(B2xw!C3)rw-;Lmd6wVS39!Wu@dzE6`|p2jG1Hp
z>7~2tLxp{F6wpgBqo-NyANW9e9MpT7JT-qxbTQeI;kh5}sgaxXS<asBk7INSl77Ig
zA3-LGIc=$c`cA_dQ_g~Ar%^2=N109%3mDTIHnrYu9#7$Wg!>mIkF;7?ifPh7Iq@f!
zcm;v^VI>FqVE77d-Nx8!h!Ltp#PN3x(jP#Q7KR~6sH19(Of%{R-xQGn8}4~f4HN(N
zK#l7T@K@5@9i-bO!Lr(vk?w3QVjcR4xu|tr0+W44!I<P%LbvX4=m*0=$CjeDq)DR1
z+%Xi5`Wu&>=B8T=ap`c>2@IZurY+(}+;f`bs}m*e&Q+CK0Z++JJA!|tCULzBfVy~u
z>^O%4cLAaU;uKl?ODU_@LJI2Yjf9lQ(<v7VDW|%;;ptrLn{1}DkcI(8=#=D04&Bjc
z7j3ew-uj?O*RL!J&WI_5ue+Atcq@Dn8M@e+_8QD?DK{1?_&ctqUK_hZrf*v~!1Yw2
zdo}gH9TOF#X3y|;OS+3k<5}?F1D80i21*#O7R;CJXTKj*Eij}S^enIv%ig4ZD3cx#
zD<#8d@4Hogm=^C#%5pACaLBU%`j$Ox6gi8x#Q00^?g&)!TcPV$X!qt{M=W_YTUyC6
z%I`{z5$1Z#7<s7llA5XVX42OBfFblj@@DpB@#TSNPEj$Kpt^oNhV`&RHE4T{G3)}=
zQ!CF3ISi4`Doo_2OrmaA-}E6Hie%dVV@14tg**5j^OkQGsE^!!|CvdI7<TG}#fE%x
z@fCn>>m+5TjMC2{U^Dk!$rgF;rkshJVPVz0X38K23BS^!mc`r*%19j0ppkPuZb7z6
zRtJeJWhQ-g%tWjqptq&eR(wh+jxk@88F9E^6N?<-Tr>udNF(*YS6#Xw;9O6_v2=dm
zc9uf5E%W5q5obAK>k}RG=r>_Rvdcj}8Ip$)XA=KV{yy|&ZJ(Lo(t(F;H}uQWgN5}#
zT@)tce2Wjub?UE}FyTsRad54FSP?q=Q);ffXKoqlXX!FWEA;z)<MJ}L%s+<H3NEtu
z=|#-#GObimzQsS4U0)2H^19aaiI%2zs^bqbPBqMnEj)(H<Fom2|6Tl+ST02=>R0t#
zJagaujFkM|yiTPT-wLp!l6gUla!3s=kqe*vGe{G`llpvkg-iXj>zm3qmH9%K<Y)cT
zmI{H5Zt`Js{Uk&j52q~XU#=1Rxn~*XuKQLw*Q(SWyjb~>6Cq^6U+QqaDqH93Ll<2Q
z?e&rbGrp1{#rT1K@`n+OufMA#ima&#E!f6?&_N3MPk%_zlQy3j@cd5VX_qAUr|ntA
z<*I-m(bPsf#2F5TE3^5GIvAeP;V2iHfHk66fw?E&l~}6hr0I%t!f3eK(KT}_x!L74
z(ISVdX<g(&u=FMQtYgH2a67hXn=K8_5k`A-LgY`WA(Yrf(An8^xx-Xhgz5Q%8-kdS
zutB&PGH6a(-Jy1bi?Xx&z|*JDzRJrfeu<gU`fZ%N<njc(y<2b?)Kqns`0X7Ux5&>g
z**>;NI-Vexjo)^L`N~)$GFDZg{8XoC-#>uaO&#L$t^n60Wp~z`R`ATHeZtfFT3!j4
zi2c<|D4qw+{jLMU@yFFk%ij~%7_=3{i;5b{@CO~05Sx+u@K&4x3mZra8=4&MOLb9Y
za6sqQgGoJM^?BT2o=@Lz8d`@s17?9lS7`cI5;91qEPp~j7Ao`4Q3xG(z)$`ApgY6s
zwbcbr9aYnHb)+~LW7}Y^{6o@1qNQs#-BMnSt^$uH?j@b#;4R2Z%SzW%_B}>L6Qw+z
zbVguOC6MrWhn9PxvZlKGB6cz<r3yH0Hqlthi9F-VO?k#OcOi>98F$>;7HsOgLMGZ-
z;}@q*^2)Th2$})9AFZBNCWqALLQk8~vAZwnWdC@TG=n1duo@o+xKT=aRWI$WQM}9W
zHQZ_ruzD_3;MhEjIYeU5zgUM-MVpoSgTEQJf0v`GHxp)N5Edj4JWolaaI8T$Zler3
z{}IJyqC_LjU`)O!qW_p49`S<Zh9!94^KmDa;GIq=((5y0Be&kfGod=Ve<)5eujO5u
zx4^%J;xfl?_k=t*<ckXVtZSd@T+?4Zr}{{dE$LmKok-WR24ZY?Z_2lc4#jwYThS7r
zy7E|pN=lg+H--9M>Tz1b$JeRU3+gIf(J;-;>;8d?$m2^*=RNgXSrp-57kvu3CZeCs
z@f(|%9l-(~_q@jG4{_KLaIrMv$*^(3`_L}_Q|eRMM1X~mqZWj>8Ob$GeA$HCR_>1y
z!3AaUbEvZFdCcM}Ng7w-SZq!!(DIe+;tlfr|A3?Yf6P}jVCrRiCqApdTWU8^14^<3
zq%+k{QSITr@otnIF40hv^6fV%HMoFD3^%^Wu}~~JCpB6xwRX`O=^P*jo4Tl=rT&Ii
zq2GP0afdk&jvLLNwUWz2ZsrwoLSOn0y*?FB`Pe)a`<UP7hCm{2AKh1o!VLvAnYC<O
zn{sbq)=sRzr<gW|?bB_GG-s3V8Goy|OSX&IVZ}Z?-$x%4ip*M2<#b+oQl}$NZ40SW
zM)x0()f!Nh3A7nazBCUs`Hv{67}=s2SvvfHj|$?%#ibd=VML#SOZ*usshbxjhSAu3
zX6zD!obXs$`qXyFvgFa+%^o|be1)&6OoHq#t@&~46J9swX?K80Z`G8UIS7}(D87m&
zJ>0~bb@JIgz~O9&U~quuWth&+dIM?ydoVA`dyKnp*oHrK53nacE^rv`e+R^F3l0-&
z!|Mh0evJw~LUvcMCK>YBZrPqD`Eo&##e><?MQ%{EomevNfV?Gbc(r+#9B51QR8(j^
z9)R6@L87b(uOKZXZZF19i6-=q>H}|LI4>seukXJ3U+G8iJO{HLrsH>15k<cG#tI`H
zI=hwe9`<|CZ#oqePQ@Vm*v^gbOG*gy(BHNEikON~7Uqep`;#Z7*Dm1uMnn|3ELe8M
zt(dXC_cN;6W9G^_$NAv8&|FxN|H@;gb}CW9JLhI)|0j!gPQ#mGY5(VcO!po$AybLW
zJ~`}X*GWjZC2~)KX7$I}Ee4&?qHUXA22Y=xzfSAR_tQd+XZd~D3k`cugk)`$>Me0z
zgF}9((K%xqWUCz|;s1|)!t;N5y;MEy%~&)QjjhZ~oLN*{jGg~8mb7!U`ET<dD>)mB
zn3$cr4(po}2L~%N7auv>TgSu6p~oU`<m_nauJa!}H!JyDi_EIWqHN}5=i+E$<|HV{
zB5nt?b5yZ6GBG1({|{DFnMKLS+1bp|mYkjIze5RkXDJnDBj-2!e>Si3+p4zCZ;q00
zC9K&0bDM7(4pw9i_Wy|-<Q)Igvh$I1{#Pq3%%bFIXQE=}tiz)8Ly|?+%-vb<O(|&!
zd|Qb{68LuYA7&<Yre^;`qLOp7@$zvB3nM!@JDM5UAiFPBbjZ(`@x>bQRI5E8jK;i@
ztH*DrAZUwX$OJK`CsQwV+Kc+1M@gL@kOpfNOFbY7C#0t0!AXCn?RwePc)YFlV|b17
zQ-{_+|LyX86gqyIUp{Yqs(yXReZ4d2su=I`_2}_q=xRHAJbs13H@(6@WPafFw?mMx
z`(3a1uj68`r;if`zGOlV_qSeg2Cw?JZk93_FL*fn=g;OZL+f{~fea5_SAM&0YBu`a
zIXUlZMP6?UOl`WxR$AL3ug{nFmu(YPYOimu;DvK~*?Q@hTd4>&UmlXX^ItnJFsXDQ
z-?$}H^K|GvnVA)&%baD#){Xf=D^i2^u64qvdrH{}meMG!nBw=U1xQ`@isQ4*NAXRv
zGJ9bzLK{&gyQo*4U2ja*nTJ8A*y~?yeeM<C>?v4}t0fcv7?b>7A=loDt-jWDCeORO
zWpP7ExpwE~J!8H}ZRaQEJs8iwt~y(CsYEWMMy8ITvOFR;<`5{>tMZ+=U&@5rW9lEg
z7e)_*1A|E4gSfifg{S%xjHh~F=W5rNjtlb<XymK%&3|U{!2UCn_R;U9_3AV-_ci-m
zh$?!1?5$e4H`(j<b1GoGcya1AgH|RQ<~2U^qP$Z_O9@s??|?BnX(c^A`fKg6ONxOy
z^7+qsY-ca-t+c;Ep>&r83%SZ7PcXX2`up5JBG6Yl=rx_NnV;}b$%%pf0u%JwBj^5h
zTjJSF@J08WgW`2tB?xx>dW!bEV1SE;hwyZ~-d5@P>J>GihkX+FWbz(nC!+qAkWzng
zx>^mvJ*0Sb9{Y8?-gL5M^)kyORq3~Uh3WS{op<4($V-l5rQdqp*q+}TLuRAjdJ_uN
z?^*4?QZ?!8bG%fe-*Qu><g0U=>f^mlUBx^s4n#En;<MbT@Nzvx{9ozfW&5139Jbt4
zFma{W_*Br^WqHP!@e0g*eJQ)K)4t<+AhUe{Kk}UB+&;faDu?S0<4Z3XpiDJk!odQ8
zXRUz;^&fJT4_&^n<yR+8S-XQ*Se(V>>%+njH*0nt7~nz@ls~1_R!&xsL|8mf>eOJ+
z_yVR%xEqY8!)bOoW8{O<BQ_cBd5Y?-=8sbo+RLll#Z>CmOO^rV{jEln!C6QICQeB?
z_HJGxgP}<oxF_nEEEcP_JrxGU!vx12!Re;wrAt!*Heb6wj0$&@Ey?sQ)eibsf>89t
zc{GdBM7z1?5o)xf0LOFw9mXpkCytJh+{!*Sn?oM2*+vgU*%B!>870swqV+|IG7xmd
zr*ycRGniVtTb9AWZB<7#Rq0Bh@|-5ctW!9^)IE_iuM;4kdQS)>bpVOvCi-Y%6YB{%
z8!%e+<h&T@KZJF28_(H9CU+92;8YJS9$9(b+0Cn|*MKFRz^Cg8>6PXxtDcP-;DxaC
z<QiiR=0)0iWd+IO$K~2Rf2P`z&%h?SkLpFbO&V-Or}tTNUW<g~eWBcF-$GVt#5>mZ
zxysch@LyI_Wz<2Li+()nK-xr`bXE>YD!=lNnaiPyqaJm(;G<$yo8yE=o1=0V$GR!#
zwLQeRh>>9A4h*wav6Kf_T(eqxgA6-LJMc`@hvb@oM}&z*8C@i+oj96}EML=ciKc7}
z>{uYFd{Nyel(bs8K01J(1tBtH>&#`@+@S|H8lHMdM0c(A!oM<I2U|PZo=<ZoiVp#z
zRJr2vz)PKa5tceIt@=FwHt?CbV*&a2$!Po-)^>tf?eJX1{rd}`0IL+j8aWRck}-CY
zHF7ZkKTa9f*Go;Irkmnu{!1C(tBEO2LnY1-X?*0Tr77}Gi%>h_pn7_7b?N+Ei+KhC
znX$VBN``xx7*pT?V&c3Ru|JNgpdOnE!*DHYI&~%-)j0sy=eAL@+~wT)ZEUk1VnD^F
z((z)<qz|h-a<S-|$(W~v*wm%u0@#gJ@T$gH(WS~=TIWAoJ9uDb<!MRj^O@62liHR-
z=@CraYDK1}+@G;$WX48QB9z#u6SmgVekz&Y9WbQ9yuu)G9aG=9?NPv<az~F|I&Wbb
zVX<yG9a~IQl>EADP-|M+7<FsK?55e&qBqgv-W1`8YxPabIib^Jg=u+f-w7F$CW9Ok
zv_$C4%=+QeVxCpgzsv}Ile@>cAKZ$Nz^ac0CpByceOj_em*yQTiKB99R@|(o#|3FI
z`2aSrD%e<VwVr9<td?Fx{}Q|Lx!+!*ZKB=>>*GE-+g@NH6_HwECJI<3m|%Ha&dZ4~
zW+|Ahc0>`PlVk|ER?pC*S}-%Q#=su~yntJ6kN?)vms2lmTNp1!O*}nKEAV08Ue=vb
z)=f1K9D7XxP15=+YS^}kaPOpux=xD>3Q;eA9OI9GZ}P3FD?P~*8>uxOy-m%}fh`9~
z9EToW#vEbqX*G5Tl>z5Bsii%0KPcyumrMlZyVDm<$yQBw<xGteE4_7Gtn;jt6P4Ai
z$F1KurXy+}jQ~**^`3rK@{fZpKhx8hoL^0&wHhB)eqb$cQRm5|a5PG0Po&RJ@DYjV
z6|?LprqXp~(aRAPKbO&+a+?me4ldZrYJfsM0kkzzu!Thvhp0ucPg{7=@VPs#Q@L}K
zRj|C~`Rt30xFyS`2(L!L240@#cjFA{5%MC24y(5{Rz6PLX$WCl20x<H|1ymLjr)#9
zZ<mt7&!Aa%GQd&l`sy4b7DD=f!*ATL4|Ig;xEy`GVf78QUUd^Y4UIE5cZ}w}{WjWM
zHgsc<Kc;b$?Bx&vsIj;BSg{{L4-k!Dku}s-W1JV$JXzw1m4rGy2I55GB7-b&-9T>=
zmosdNDv)#sBeRiF`bIRYy2NP4-+nHX_Kuf?4bW@HTVC5uSdVdW!hfc8Y;;%CiOA3C
z$A1DEJFRJ~w*!~J6V=WTWl-}%T=1{G`9jr=90{#z^?QI<SF)M>0%MSpjZAT$QOx6-
zj=vmU#z=q|G)bgoMlyB1p|>3);sp%V+yqtLyR+6yXIQGRuDP-oCzgfI#I^#45!5Yv
z^_PFS#6S*s>yq1Pv@^|_%n1ouiKOujytXJq4KvFiJ}zUez<4=5u2nF=!qp^W<I5r;
z$a+?|2JP5s&n^{d5=netzPMtoSY&y8WIa+xaWHhx#OC<<U89L&AP_6y$FEcL`l5gR
zBHCWw$E>rAS?OZVIZcl$Z^2h{K1@&|-fo2*8x0|)^;VbB<ellAGGUkA9j+I9r-t~Q
z4WL$7U`oV<gq{5%iIC%(v6%c)%5c1%XjXeaOg!!ggS~bHKYdT6ULrGO0+J?>JNj~c
zlY7-205?#SUvH%jk_XP{$PxYooKHgV6BG?U^uKf%(U5k02hyjhNyKGV3NfUIi})mB
z-9mJa>+lN!9KPRFimBB%e%+u#+_p_Dt#ziTTZ!hPVZ7yVa&iaVK){J%0kpjAPER-3
zBZ!Y*=8vTT*+8V_F-O|{cyTu1?Rm&cmZe=NQh$tC@cuqQvq;Uki>LB8i4o*eunElB
z2~TSQsthlt;n^rivo=8g5f1EdJHi7}G||-MEv{xe#{L-^joo}-TwzE@`!n(~YWvf>
zdJy=zEN&uzWc-V~PQpp%j5p_7dfEq>6^ZFfQ)p$8w^!~<`i04UZ|d#I`U;?>8@OW5
zT8wVsh@M3E;VyuVvEg&Pk2q(I<5rq_kFmezrvlW5)VSi2$&-*)M;*2}F>ghlXacI)
z`8arV9sFmU7>dq>!^A#6Yw(^7Kx~#kTa&eZDH*c~N%a~;?MY;_Na5`4xhTfP*`$q2
z6ch9wZo3qcY~-1&v7kP}Tu;ts&&{QiQvW*29Pg7YCGi1?$SG9!ikP}-`cehENJ~yG
zRP00oAO(y7aC&L_1cLQp=DV{TQ2YZC?JGJnYX1?r!3p7Y_v^t153@VT?f2t<Sm_3u
zA-PfoJb+>n0c&m|YiVB-u#-7e@y9&{yXOX5G(Pk3AH^Jp41QKeBY+ipZMhn)usAa=
z*eN@9reJ8tTE=bq9ix7g`h;uhpOHm=ODvvcR;Qp-8SGDL*PynX7E6@_u(@jlyC-*|
zU0c%(r#bM3YF0wy0XU^N;h%mI3i&nx-wVW1rPf*G;%*SVdkk<D0yKUl(!Sqi)521j
zOrt1SJUht{IX!{zb@b{^tGve!d%r70%~5;U#f44jWjwEXmw{;nIpb^@v8!w_^|AhY
zZ#Op0XznSw30e-_tkdH^V{Ou`2D)XoM(4&Om82buc^%z0>PAE}okRwDI5d$5lVtE`
zC+o^Dj+6;j3CXt)PB@ox_TV~Vgj)40h77PltfKvE`iO^9>~4=vuE&pn-tM2E^u~c1
zwZ;jFsYV#9iSl08FSA%vh8DFrwx0QCpV<HvoO>m;BnYVEa5O@-W`B0;myk7{LDr`g
z=O-Rd=uhoHl2pn&#(H5h9YTN;`>~1_TXbzD9^1sYJ^{6ZkxR%8tJeSy>&^09I8J6E
zySs|hcQ&sO61K&Z>33iL^_a-rEQ?y~!N=I`V8{=<NKt_S;?~!ifXA4VLEy5K(;8j?
z><@8hJ_5WE$BJw^s|>ZKQ4OKa4_3Of0Cw35YCxH@J_+p!d)iM0e)fjdSX#mp77%i*
zNR}3TX%M+qi5=@g7joR6k#XR1pBAv7_hx+JHxYANGf~VdSYImTm{E~6F@!-Y9$ZIL
z0G<2sY?hxdEOC<9Wuwq(ZGIabJHc=dG8fx&T>zqyjaNZW@NK9&Dt#xlM{t9`2AtAq
zO{R37b#W23A#iAI6<Dt3Kq^L^Bj#%%*<acCoc%+R+w|z$`j6xi4b`*T=<l>7*;O9(
zjMHsBJNeV{`0y(BIm&sWHh-4_tEiL{74mS8TiI07yj*a#w<PfvUjowq>eRPJWMdEt
z{oC>}c02mVFM4EP#BwdyAA@YhzN$6eIuW4t9y00<H6qm4Fh2~5=(Ujj3>>SqY@*Af
z<^|Q0RuXdb9Ln{kEYkM6JlRn(epsv$&hj;MP$B|)<1u!pofL1TZUX^U#e0gCY?ryl
z4v0jZdZVUmT$x<5O+4n?2eU-jL*w@5waE>&F~F3*!8I-0biR`=lIs`s6PB>s<xpU5
z!(Fc=#c_L7d8p-QxhStF?##%?2JQiGx$$^mj8HD?v22dmjX;}V3nS>or^^`liB#mM
ziTRkzjyt0_;9Q^HUyu>JHLHr-pplt5k&g-a7j8I{*z}$)0b<sc(~M`w{Q0A|dNptF
znf6H}(5<mDjN}*6lILK3Q%aom7zl+)1;v`ZKEyk;N!!XbeZuv%^3WGz#x;M*8v8CV
z{rM)JJ|DrB!_<4gHeU(9rm1}b%UqlM(rjkB)*6+|tFPoQL4oPG>Ao;yRQh|44>7m9
zRj6Z<VTU7EYPJ~{x_*_DC*h5q+Im0+$@3XX?P=cauQMPp+<3Y%6*^vwjm}G8C8N0=
zHG<(iZ~<<fS8mGo6tM3)>S2Ur9GUb$I@xL8KAYA%cq16VBrK$qxnDWAxXo?q2_ujR
zwqoZaYpl#99QGyOFF6zkj_>TH?>iva0VjPz>*gyafQ^(@+Fc@g)Z7@ELYbcvGxtl3
zT)S0Q_umU}L%2r{#plFnxQo}|UxUq?tUHJ<i*jSiIxZ~Y`xc{w#*6PU+p6LaoMx|v
zQ%T23EklSC#+r`5a^L=80MteA9sP}H%$SIEr2bjqi!wVch3-NdPoU75M&S;4`yNiJ
zvp{wT)5L0A$iB+~NpXv@fibS%d<1Y$_Dq4nHyF!gO<~8nxoXC>R1TeetJ@qrUPW>m
z8q4S$NKM7o6!PLU8#Q($MmV0N7S+DflcPW?2Z-GV9##B$jnT*uY#YHvtW>p*lnR?H
z97GabY#To7GNx~kI6hY)lm|!$t7*bjDYZP<jGDve{kg-t7B<5s(k69oqw9GyKQcOL
zHp98MzI;c#6iJ@%??~r(@SYLd>x4R6x<^CqBDeqHV^v$*05L#4wz^0!@*4-isZ+fy
zs{hwAb6&TYL#siVYEMmIRZo&0S5ouh&g!yT9Y$h`{P-vVzl>3uE-RECL)Ox4Pv#!a
zy2(wNL_A9|Xv=^7GeAvf^iU3PR}r@_>Lf#LLP`yS=W1!uqdXHroa)3lAO1AvkqHZ&
z4P}`9+H)zFWgC<>NivnUl5Vdu>U<qP4)}ZHa$`yW;-Bq~ed@YqqDL;3^~(gZ6^!ea
zZ51b#EeRZRe4K1E-|HWCEjc9XOP@O-d&<oe;kxy!H{NqS6wx7$CrUiRwrcWT#b?uA
zV)FyI+PS%e%&WJi#~IBoR+e3uLl#+u>raJ<vGphPEzMmOn*Ju*JjKFX*UG<o4Gg7W
z@ex&xQmFa7uhF+o;kX^7>7H}tG~Nv=KV}uu&VEw8ovfsw%p{@a&#nz30;cRSbk~`E
z^|RJn9r}y5s`Yrj$_1Xkw@H<MstMOg*>+!mpzd>;;Tbi*uTh8Aa6L2VVBw)8>0xX5
z+g$QPWAlVo{y5H~nBtE8l9&IIYt|d@(?81tCl(zPyXH<}odm1h{Toq(m;+=<)#7ce
zFr>EghOepKNvUsSXkyr_v`f1fw}~N)wrudeFBmbS7S49qzy+p1I5&;iEHn@pyqCiz
z)k4x)##415I;yc53$VLO$`W2=^W^ro;`O9+jcyfcv?&wZtuaxZDxMB^jrn;gVHn5j
z#HJ4YMRPA`4KE#4{bAGAtdQi?Pd0|GhfU%ZXa0n~<)m~^NA|^jqj_I?m;K(?M3&LJ
zCtCh_*!vhr)4O+NZ27Ib&<ll-2sp~8a^F(u;aTP@xhtEg0C`ke_R;pPSG)op4N_A#
zQ~<?Rw*w#XGXt&ekY7+bb7|T51i}U;c-U0fc+`DZh(XejM;#SJpZl(cMbnUU@Moe-
zCarVd1Es-9+*X%)oloZYyEUrb>#EiXBk2TkTY03^TLLin`+3814z5Rk<6QjWWw+M}
z@3-G{`>hSHj$1~^=*ab5Us4ABJI1Dd8@O|mGP{bJ^69C!Isj-oI_d>_CwN~t?*8m+
zMMme_LVfGnWD%BcaCrWGzB@91>3p^O`=XT>bL;53#&*kmy^7y7B^!;yj5Lp-LbxYa
zH*u(Ui&j7g%-BSW)DyC0!x713<#0035BYxJ1!ef1w&Psrc#IWn;iA#zT@iz{A%$49
z7hl@OR8yurwFRfsr}$3pX7D)uf%M@mM45@wug+=_!e&<SRK`C<=pX|aM`aa}I}}i-
zc{bm>yG~ZNL&+xjq-lfqwHM}1#VKL(G5u4$gb2i$Z2%%ioY<Z6U=(TTI&{a?l=jLR
zk*Ja;Pirn9U~Y0Ma+7>Vu0g^3Zc5Q-YluF&h3N+5i_Nfa@fWm=vFf|UDTNP=L@L-f
zwHmZcHtt>IK}VS<{<gn6r5ES;mp9UA<LN7;GFH-v;dvW%b7jli#U<6Wz%ll$<eD-Q
zpPDB-#`SqBT5BK1YBs^zVpNQLK>?0LGd3Z<U{fd#Uwtaf*%mzoX<>W%rcmUrL6!ty
zsC9c{cH`g5>t8%HZ&k{UP+kaV>NO_C7qRd?Kjwr3Q)=P|IMO<1Ajp-?W&Lt^?LU`=
zX$rq90bs<9(x3b|G>&cx)>nJYKTw~T*&{#?5d{n;2cKAXN8j&*95!sn=Qu^6t4+`5
z=W^MnzLVDQ*6hEX?#f$2kzSm9@2ss|CZ2%Ft_1nTSNl#tXd1rpsS%_c+Vte^NZg(X
zk#lPTRBWWN2@73fb_}P=DRdR6MdM?a5GSNwsB(Y_c@Fq^<Ip4*nXid!!Zadlg(KUj
zlk8>KO>=!*g4Ytp^^TNqdYS&EF*%>58yXn2N1rN)PI9hT#6WINSrZ2{Cax~!?uo)X
z=+YR5(d2nK#hy#NoVfemU04&GIBqX9--7^UKsSv=E-=M&6Q6^GT{_$P<+-Z{My7CW
z&KA7z3Qs_G-)~_3@95N#$ZEt~N4Ic``W4(~Cp+L`Q<R+R-!`L@3}C}{NPeYh0e}tl
z{?ktb9~0^}iQZZ9p!8nt(yIHD44)n044)IItcrMQJr1*G%|yxE^Yw7aI<-KyC4Dxa
z_6z##<p~)CZoeUVW>^-dnB;SO{?&6|1=MwO@JIjat_6Gz|6GS3a5Sr$Xpvq+(mp4m
zaW((&IQnXFvl``Ol_ox>ZVgb;b!NtHT|wT>20P6V6*MN?3rx;FzkR&?(peiCbd=uO
z9fC*g`@NQKf4sjGZzrBJAChsV)+j?uFY48JhHUyl*Y4-Ix|>Lyn+w@L+z+`F*5PaV
z^nczp_eNCUb{%i|fcTU=#c<)*f-PMMSl+?HIb6GFe`$zuAjJ8pp>!<gPx@w6^;*2&
zLf+HY;IO@XS7~#P%+3a~&F<@zmK+)16vo=>-+a;EzCJEHV>COiS*fb_0919<I<%dk
z5Oyj;UvSm>{RA`DgwL4IKuxLMIe<)i7n(*A;cC%rsC!k}PQPRRz8KoS3cSdtcTSul
z*cr83q#FE3n0M&O9m(6CE`GL61&VUXp3Ak3&DVmhPSb>-I9&apqPb`H2jNKm3WjxF
zmG)KxlaX!Pu;R*1YE4OBtcT{!^nR*j{-uy6_uj=&IqGfI(%Sux9CQKqhD;*bds*=J
za_Hq}ae5Km=8UnD*)~mHwT4HLCB@dm#z73Y8irg($ESt49pZ^iXM|-8U5S%rF;HG^
z>kfl(YlfOPymw99A&*j~&|3-iy(LG*iAPKCMT#v(kZ}O5=*uzFQ-D_swk^h2^m@AU
zob-+}MGocuReQdcG(2V%SAUaps2~xU=?3Ag@Z%k3otEGkWd_v|b5nMkI3l_&Ma#z%
z#0?bx>f@8t^oD!}?g;g+@o&}5xxkg)_f7d{k=AI1rF}N7&99|Qe7U)9bVEcS)efqO
zJk22qji*W{i_2OrK8E}KS<~uuG}1PGiz{Exk$ER|v@@^fFpm9ZP~lIR`_V=aSmZ5V
z&U~iH@O2qO5Nef6Z$sk9vd<;sOX<Gk<~Gii-O>M1zKV*8B|>p8q<PX&#%;HPKr&Is
ztNqq>jH^>7tFTY}UwT%Hmn+_f*3J`B@RQLf|4=l+wrkjgO4uLI&59Mv4N360RMtaz
z)9&PIEei%`o70sOv-=7ulb&#=SN{utyiNgnNrjH1lglUCj#We+6#ZhaOuH7ny4?|y
z+=-_Xkc-Q{lAGF8WAYedTFWXdDlz@4bZDvrJoZCRyY{pkNykmmF7Zg7S*2;O)gWpK
z32OYEbfpIWeqs9uI=#Fd!^0=XU)`JpEqeVj|F-LLohB*@0?X`_0G;&%jZ|IchFeZ+
zO-1<(+nDq=x)yb*neVyu>H_W%23y2x&Nb6=sBY@^0HxwU!7jhN7F0(O6h!RK<8%($
z=62^m!sLwH^qR&$q-YUP6eJtgj4=NrrUk6;EZOrcKTf=HbcKn0Lkendoi-)YxA;SR
zeT6AT6FY`~N6S|FniZrdu1_qfbmax?i;1=88Wzk$1%X>R%OrJn9D$S=tl1j=%RE;t
zgnv#y%0GW_C~ApWc02>|A(RZ_*>bq}3rVa2P%6E8p2xZG)7uj@BcQzl5ij<sktjOb
zf~1zEyB{mbDNVO2c`~q@Dh)C(u}l5?RB?Q5&$*?71TvZcA2e2vt9$aE!4uD-w~kB8
z64ss})F(IWWx{}BwJIXUXeb+M3(}ZxzR$?~rta>?+wt{X8Ew(q<(-d~d>`o7jq<<F
z-D4!3^~rh(8Qsof2Z-z;THU`-L8XHOHs`_|5qF<2=BLO#Wa^8Qx9AqKn#7Y`HxJAO
zU6<0(aSXTwv+i4AI~IeovH$Fj@6?gS+t2M%AI2Bi7i{`btLG<uDt%Eh^rzlgwv0tk
zxY3nVRzev8RdG(Q*85RPu4zJ@nR`fLyH8CK2Gx$14QQ%Wb3`kfiHMwhjnDjqZBIz2
z^G8=ui<UJF5pP@H^va<L@(qXO4Bgw=Wlx^(Q$C#uZXkc)EMHP{5PBBcIV<j5v(&ad
zUf0u5ZzEO)K(#q-2y&}H3$KzbKjN92P)Ypi;{L9i8w&E(V{z>vt`m>Nd(3g;{Gy1-
z%_;Rsmg19rQFrz0NSMPyZh77HpQToVS8EPNk+cYu%jMuQKI~~j`^%{fbQU4<7rqzk
z7$>_hYVUxKJI~_=D&tux@6(}>Oo7`KTVjy7Qw{2a-fDc(E%56VZfhYQqVCwa8YvN9
z9^F6TCJX8N{eFXu{el4-jty6pHx7MaMax9Uxq3$F+_GEK6h063!q?ph!+*`<Vbwku
z$q2)H;2SUU)cMW&(H~}|<@0Vv_=ewho5wr&9p-XDzt`r57U0ah-MS}6Ld#{^sENfD
zlL>7xy0u#R19JUjJo0K-_+gsYFDtl!OY0h+DX3F@N7SpJURMD3msPu7W|;GcI%HR5
z-9ZSPzp*LYa-_lX?dC&YsQwbm+}?sl(K^o|PY-u>AOkLgkxOd9VeS~yNK?_Pu6ieO
zSKHpbA6Helz${1DsjU3_o}!jb19%=p^WI-RSjJUjW`TY0^u;!j;Ie$phCr+Lutoyv
zzRg$*+-|e21-~jwLN(`;UGa)T&HZW5p{d5QwS+tQr_bA>PCDz-GF=>cz5eB;xwOEQ
z#<W^ObxvuhvwtUDu2O$>Os0<kGbdBklZy?rhHmLHOKj$85=Y9k_e!r6UKJ?M=gQY9
zivi-9zi5qkMtCGEA%1y$2DtWKB0YJ{mj>wonX550cE{AOmmiWdlT#tGx9bY&puA`&
zvh{BVt@e1r#k+V8cOg*>dn=)Kj$}%D`t}n)Y<mhC3i?3an-!G%3c?64G8PHzQlLs*
zcj{k1pOt;FTPhlz=yL6R3@oBZMVij9KFTZLBD90ZY_QjMR}NEk>?@qux~8Z>8{c#h
zFOIysPu;GewxIPWLo8O<Gx#I5y*f5OorMQ>fdndg+Wd?Ml1YQ^(gSjH&eUlJZv*1d
zhR$tbj`x>1glTsUkIS`mBm-ZYyQd1I%T7~#o2?Mbk2}V!(>qH{5wBTNO88}Z^EhE~
zKfcl)bFblb!t0(zTvL(nE50}}-Eup-+SC>ecEYk&RzIj`gT8(^;yo$w$;)#8beu7P
z5OS>FJuhzZ(UG|q<8FU}C^NWRLW4$Ax8Z7pL1OPT-cuFP@?Kv0K=gQfZBr?;O9u`r
zqe*j@jo3a0=!XdjwGJ?EzqD_5V$7vSX4SZTd(M)L15L)dwY(MCBRUk-Ic<{yA3#+c
zx1gw|0qsimT{Lxlu<nVfqhRx%`I;lxqcS+<7HV6i@u9%pFE^haD*PD_Cv)f}?eyb-
zK@}6uF?YcmGIUZWcgqt|yKD|Tk*%;a-(~@1)_*)u_SK~&K5?+ngoX?`eDQjSJdH8g
zCa;n`Y`Hg_P9ekCQr>F@s-9hJ^0W-W8ZAMjgCUv08Sw)vew8Bl`bhmMW%sje;?@oU
zq+MeW+n<-#E?dh(_g`(qfMNC3P$wPe-&aAF;t!`Bgy?P)Qw~mOQM_Ei^AYqBfiZ2h
z50@i|w&zOLZdz)%0TL^z$8IP$l%ki5LNN9(;!OzZ^wN<i`xk+24LT;~7^DYQF5PU+
zm0x`s>Zx}UhOcV4Ess9pks^72JjUhhQx$rdsI;@Gn%xM^C=p6tRGviy$woTHeQ)f@
z1G_A<MjV+4jL!cVFSM$dSm_@VN7)!;3UJyab<D#V&)E_oD`O?7SgaL&-ecAko|U^z
zkI=sUM?fs83-J4@tMn7ry4hy+2U*ixu3?WkC;E?kX*?GzcKcCJC<Q)xJXZ9U9K)B;
z){luR(^jPAavh_qw=dfCUaz{0Nsj@_f}j5VZ7G?%KwuDVFOONCl+5V`s<l(*`#MB>
zMK}bNazAIBTq9K3<2um-ryZO-MW`Hb983AAGKYe6Rw+4GOPq&@%vR*QVtpngn|a!C
zcs=c|U$T5C7sIzb)54_0S3)$}I7syKe>lHe%o5g+)1w2e{H4)XwaO;{itr(!ZZ%iC
zJcj%E$(eokNaVLk!x{~nO7Hqz4qx;GYE~RfmribYlSs)CvB*JP(GZzpDp)02T#N^k
zqP!4yC1ZK~x_{yb^iApp-7C1?f%YK+pV)ze)Sa0Pdr}gNShu=y+0tu(8;-syvaV-4
za(EtrW3);tZ&dEH?SxOQ-8V?_rTfkvcO*PX-E+mksaub8y{>H-_3SX&A>>L7y9M2B
z9OYEVORll*{QhXj;iE+$r_>O}TyRruiMm+%pU+~IO&6?=&6z`L<(r}JtUr<=+<)y7
zIfK8Bkngvrb!@gbzJO-|Cl4ZFKp8TFvU~(g0XKU+IwO>mgI`gI2p@{}pH_IO*iNw)
zdj$r+zMt@?Y=88h0XXnhpHPa8-_S24K?O4*{?*VIXT8<2c%0^7f7{L|zpeh;nsXxP
z#R&{6%|fpMlDAiEC%HogCwbX(O6gYd$yfUmX;0)@U&ROJNoH4GRk#z~Kkr5-F_!xf
zA1^#V<r<m*$yR@Pwvc3;-PTgh0eS}{$p38b{&Kp$*y-D(&CO+*O!I3I_v?yRaSmHS
zR%m7zin1WJUdLS8ABm8OmtLLO7juFM#$J7OkRVSPk!&tI>Umwli3iqfx}N`gz1q&e
z3<9(f3S(u&?B#CiR(j{v@r+aktuR@iaBnX7<4wu5h*|LVBIJ>TX#5pFBwhbJ|MKFU
zfx3yQVV#?x66Qv_D+1|VN9XS4!3=l0xu^+qaO=*Oi*smJCr#(Nqbv@?)xEDEDE)!K
zdrxT_Cr8Q<G?IZwg-29Cy07%txkKq$hkC8g41fguoAm4EWeI$@g-K-o)PsLxk-Dr1
z1=O7(eOnCZ!f2VcoZCd#Fx)s;{wVE8P_ADJ{Sm(-L1RkCXGC?K*Va<0Mw0O!79hs@
z{$@&W(@InD=-0Q(Jw@@h&kFv>r`on6Nb`Jx#b%GF#5{>4dpKrcZSLQGauD^ATf)Va
zzx=csjQ9W!1ig3L(})%Dbhx!h$Z+)%ueX^u0j{ZiS^XlV+pv=$D{{G=QBBqK=VW=D
z_B8ll1$ONO<t5s-%^hf_Dltgw`vn|5eTo*%b2)@Hc7SD2|IN5|bbN^=2oMtCGWzi8
ztBON^TNgK4zI%_F$VA$=BYowilH(tLzb5=8Fo*D#(@dfrIUImP9?)S|_#ethZQP2T
zG-svYn<Lo$Gj+5PV&DVa(Xys2gg;-ch&O??j%-S>Y2P<H?usd6xrys7-k{uDn(=hN
z(rJ8w@*wX;`ep^g%xD*#rs_8S_T~Huk%68(P)Zw7cO~gWqsyb<ea;W9o6>4eFH34$
z^SR}fG8Wcf7xvxAtr<Kl@y#FmRcYzm@ok)iZ0Yp`{PQpUr`=QbQHy8YdF?e_2Fjn*
z><0U8=7HJdPA79^h{i2c6Bn3a_uEwCy7$s8z|pn9de1^#K`gtN#0@XebeI?svrC$9
zI(mecxHW^C^X?k5VxP@@1`w)Myx#uAx~T~#B1v#xJMS~f>(Q1F&D!@cYecgbHiP0e
zB#2OZ7_;3g!&_uP8c#7NOuiZ6|A(!fU*l)0BBc)RG805F7+uQS=*+q5)j_fh^App?
z#XA<#B0E3wO=W-3@56kli=J^kPx9Y_f`$&QA<k!9Hm}4;snb}<J=FeKm<=RYr-!MC
zAEPT9VvCAUr{`Awo{_?ZIx<w5@opv1>?dHh_T7ABPI%l{V$WgHh%4Q#*%w=B@%bgx
zdrjz&{B9UcpM{U|pt)}V3whkBSidiRXr43KgWOF=F=;;yjwJPE@5*-7<SCX+7?)Ay
zEG5(T_wVqp9%u6rWvfEd9He4o!$JavvQvLzqxdvfc>U$|O1AMPe+n<dUp?~KHlmxK
zmAKsx0=$3v+!$rT<g1=47e@}x;3f>ELzn&4s*2umnWN+k0Ri<4zp)ALH+TES2F9S<
zPN}KyV#|`81>G>GR%lOdi;uU~8v1^pRfuB1PkA{0Q2CGn(Rw2{{|g$8P~?=Gp*l*$
zi4@@TEN|e+LkBf2H`CE~W~9-vpBn<(rI4`o9ruyLv8ikJa()geCgj0)-eNIG4b9g!
z_kULyM_G+k1qmkRKRG(E>v4%5^<ryrB{Uv7B@}g19X}AhcY`Z*8jb;^4H{2JDDNL8
z+8fU-k18e6PJ(4mKDtrNJMewbTqsenEhk&rEAbGlO;w5W#8Kr{cjI@&COTTSn1-Ac
zEE5Z|osJ1XPTVye)REspk$9?tmhYVRr?K6=a%%3(WGA)aH0eq0b2gZTEN{iXtsuz2
z)dp~5_+V7FxWM6LAqsYTGaF{Ju8f&1J3TKl&&`5kU}(fADgrQ579z}(2@*VI@qnve
z9~Z`d5&KTeNws_OUK9d%vJD?>)=9%loP(^Eevwc{-}YL<#kkR*3}|2=|FgATqKxy8
zksSTxR?Zs%ryT{YQ^3nXn>YCytpmpO<1-t|&U>Jln^FAnnt&mf)h10n4M5iWu!;%G
z^IUn~5(Bxa3X@qm^+#WbFCW4fM%+Us<<k&KnP}HzRnfSaz6m(U3dv1fJD=NU%ryLz
zlR%CkCJ@j@S?=#6cAJ7@Ftid@xGKXl7SQ*bfuN@i=KAM6I2H;QY46BCYhK~BI4Y$t
z<O>vjzh88X>Yn=S%zK;VctztK%(|aak*PqrQaLN(4R>#&1p`m($X)&#EYYtwyq#Wh
zJl*-GsDKdm?`&(tXyq_a=(ba0;)lqr$Bkgm!H?TP`06JisKc1!)Rrm{o^t^ET_h@&
zj@>WrR&-c}ahl~nErMLc*+B@Ab4egNR~aT}oFIP~&2BVrDz}w#mfLf7`epW2f|5tS
z^o28?&01aMHmSo(fblE*U!Vc+#6Xab-yhm{UQeN}+(%j~M@XVzW*p;A@oUk_sP)?e
zL>Zv{cO0#EdX_e>uBAvN?~+%l4uLfH4$7|Kqm|#oLwxJJhD%JKav6_W6^o%H`?GpQ
zzNB)iIs;{U<20WhFayqJljIg_D}A$tcPZK;ucLX;X`#xATP1-x%){9>?Jk2}F)MB-
zou^=o?8C$NIPFwf#Rldy)V#$uxBhO|U{*nOKGm7+j&ppA17^qEUR1~`Mn~)Wo*p+L
zu7d}`kQ<T+4H?*Xms~aR(U-XzK11P0&El03rMO0n2_kI;p_la8Pr?9{_}tI@Cx;Jn
zYzb4XG>10m+lxz^QtahYk^C3*)r%oq2QKNSM*%Y4U+nfimeQ>-9Kdo|9gno~<P>(E
z$uCI%;pb`ZfLUR#Q7Vsuy@aQXN{d(^ePY%QgyYQ;bMS=2)w=3oDk;t!?9diN?m27M
z@}6wU@ESJmpc5cIKur5{yF&4fzpp)zxtAkLSD=<Zdds{BvAB)LK6ihK;XXWU=m~;&
zWY&>|XN^sXfYJ;eaqk11pC<i`#&I6{i?Vvmr~v2Rdb&*Gs!o$>at&@<(dxEDb~s2F
zXGjXG=rj3|wFy=i_W68ONZKm=(WLlHi?{)_4DWKe!K&Dp7d4G`)}59ri~!Eb*^QtC
z`Ox{Qgh>*Au*m*xIP8mMFuSql#rNdzxs_h1Jo)tFjm$b|2={JBrgDYZS?Fl$vKT#l
z?ox4_U)i<`_>-5fOa{<ye*}`qE|z2yoVgI>46@`?+rkh?R_ro;LcJfZ$~9-5wa+5d
z_S_exa#T;wDX|K&2En=e$A))qszS6J2#sy*H(jz&)pk1(Z7a9zH(4jY`uF~dh9>NB
zgEyLJHF)z9z_1JQRj4XC!#-BPh!8kPsvC{U-hSILh9g=4kGC{ThT2}dt8tQ)b_&^u
zbSTXby=8>&(=TS49`%WW=zq`3V*zCa??$j39UbPEC(qoy_@f!673Rf79s+FJ8jo%^
zFs3;r5km9A{4UZy=QD^6d=wSEXY!Q$v@xy{@x56j?)`iDpBQ?i@V(;F2SptojPIC#
z?olJn^>On}U?=tX{Pkq7$}d{4V?_M!w2s+o?3)=;YYUmz#>g0?n4?s$-<&x*P19o2
z!I01YKLC?JY`;7R0qCJ2_*U~68H$$bF#Y2O(agI~c|wk_3->tL*X%pXcjotBBK=zL
zq1g^@urO7f#%*$^t}`9a=5foK!P8C+BOR=vd!8ZZcDp3)Bj$IT`&K)Ohl@DTGWw;<
zgOq#XgAe6~t-qZnJqLjh=N>fzn$Ea`YYZMYg7_MF_&B(TfG4ZdeS2#kjrz-J+HYHK
zHad%xcn<fOGrjb32a)trdT4qx4ot#Rr|FjY-D45zc{jc02I;^e)RP#m<^3k$)})ex
z*4@rqB+*N)a^N}c(Il$JfVtIni5cEeoRf}T&$r9lxUJdRJp3*6nFqpy5P}|7!fyiC
z+nR}^(hpU+^)6{=E?$q<frGL~!UyL4=ixra>SNpG&JK4IZfC?KFCh&aeR008zN?rM
z0ax9A-cH|=Uk=W;M<n9<^MNM6&)bP8xImK{EAtbEVmokKJ&6o6{;o5bElTRRzwb66
zUa1M6)FyBMWnMe|z%kd!nP{f?yuHG{IYrac3@9s7DP}@Ds4dhbXx{sSde~>DpU#Re
z^|glgRZyG1+7I3RRzL!eLudO80r$;M4RfVn#u?9J=6ev9urCG{p@*S>&XmDegnDG<
zh%g=w<;-RKCdY+GR1>6aChFzPtg+L13%JVB0@xtBzaB(pI5Y<UC#!m-<j}lZH*l1y
zBWOAY4+*=B8x&U`#WAUsfOiom3}HKBa+W+BfO5Uk>1PrUim)9q$rjdq1}*)J_H1?C
zF0jcEbf!GpuD}G%YZ1e<;~hD9LZk-e)_*%RP7DV=A>-r%wFeKgez*rm2|D(z06Y0q
zF!J;Zx{o}W^QzF>Q<vzW>$WwC)afk!2!MVOruuGvuA&_FuOlbt)ep8`u02C5k!KVt
zL(WG#B`?g4?bi@aw`AwH$Z2#;u6b|T433WX$qYJkR}r{lcF)}09Ryt8=c}EX`ZtE4
z9z9@cnupNERRq6ZlP53(Q&5i{uu5KZ7=9dB#-UPfy~JMVkyUji9M$a@cPRT59sJDU
z{XjT}!qvZen$cW=I3qfei#Ya@aLprkR=bm)6BM^wQLXFeX+>gwn?1NLHDM7MVEfhT
z+4+yvYF&HB_vDY1T$zu?3=MAo&C{>3<GG+FyH5o(0`5WCZ8;feXX?pN9cNiG-BfSA
z&6B<Qji$W#FxiWoM|iB2O6~N}oJUTslxOp7xSdgyPrry0b$&17Lz(O6-AMZF&l8;g
zRIu|#5>(EjvUbjcS$>}lBjCF2`;L;c^yr$Hz(-O3c&bmVN<tDiO~*s1>=tFylWBS|
zxapKQiTzUDM^4%*Qg2yPkZaI9f1Mux>}Xw5>ksM~LY?Nm?nHekM^Z}7C3yOJdWKsv
zoR(9+H{di~R;NeYI%<Gq_DZKaX1KJ-9syk*OS5`hDY=wWkDzzb5I?FNg9$6|j>B$e
zy^%~^>@dcW#5%XPN~+aE^U~`iWYYO?xzsc+T>UIfGIQ&05~`>lK`ApY>OuE7ikI9Y
zjY)X-F`|a_Sf2#@8E0+ccK_`{U-QUHSs5E=HV*)C26yf9p*u1<i)5IW+5E2D83jov
zt4YL~brvUEutIHsk~!_r4AwG<8nce(WlCBsL(nZT<kzcs67yY-LMA&Sh0Nxsgr~I~
zfwi5I)5)o(@5?0id%R!G`yjxO*F;IK39Ea{In~(UCNF)=uwBx01};77fPK54qP^+)
z{s;{)F4dUj_%7YX26;uC)|yg<U>N0C-yH1TvIMsYy4I~1@jDJpUn(f+8BV_LI&N7u
zg6E<A)3fbCQJk)J2W(3_0@KWwc}Gk=5)-tib0$>#5t#cLf0NjX{aC6ygCs-EDqc}T
zt|rh}rT0ijph;kXqmdBhU|bl2WseMLGP&G;ram`IZ~Ma|&N|Jm4i5%(TAC4Q9BT9a
z2wXg*Zdr2Ho<W>5rK?|HKcD&?7A~N#BQSC5{!jv`9vZ3jpxbkxI*n8UPJQRq=~}m6
z%9Z>>(}`hn%X%M$z%-uh7x$!h4;n##p|&s{jR1$5cB#f3-*VS2OhP>ZlK_afhEA9~
zr~8U>Ov1C8?0w}Okr*RM{PjDZ(_<zdC_msj@@H>tw^yby>v=c1QwbzR(1Gb&H@TWx
zkGMSX_Ji_qo)>nnHbPgcQA3)JlPjL>47irO`U>6ch)XSb*gFZ&*h`eXla2t5y$6ke
za!&mknbV!h199g2ksn%fj|j@lJ~XefOoGi5Lc{4uM*fTV{rvO`bLqgmRXaUn@ra^2
zZc&uiQ#Lo+CimdZv!Zt<vw=Gb7@-+%`}naOn2QLv$<|q%(2#GP_L^(_K3za2!RIB~
z>KWB(iU+JcBiJ+U#q^6*XL6YM&9}&1sNJK6mCF#fnRaxnXCS4CJ-CkorXzQTCnljD
zMc&NsT!yd-^=R_M1-h*bPnYd<<eVX#W2sT^{YbeeQ)0l!zFi<4bwIs6Sn&$3pY`Nk
zN$o-JXUk2Hqr-B_cN@Yb=fFIqm=f{EoXWf{nH@2L-fxNf4(@%zz522~RC@_}wLNYa
z^rOj}X_?Q;eSm)SFJ1Ekr5`z|>#okksvj+AgpqdYnGv0)Njh<FA7^>C?^bWwS?7_I
z>Q+Tk$@xt*%%qD9tcT9~WT?6UV-sNFM7@G#jIOzTNj*?MEqN*);5~<&?lgSVc#bV@
z^BRh+Y2mq4;gKs66{PJs@s!;qXgat;(RU7&)bwyU-Dh1=r01o&k0PKMRT<i+wxH=L
zGqDIY##>|4e7djW4%qkY*zrSjdS&YI5DdEV8^ukp9K%Cz{E3MJT`fT)X@k>^7in!j
z&S@r2XT%Y3nt{V6j$c)yY|F*bowC}2@p2YkQGCp>+{_2!)b239kcB&tA2c<;<lK^a
z7(IuPO3UGLZzB&`J`-}4K^)e+v<%%N+R9aN;c4M#i=I1F6Hf1;!`r91-0K9EZWxOL
zdhUifT_3ppGzBQ1N{>b1bbkhgIDkPTRfp5X<r$BhQEvYg<@LY}8n`^vcst|YT71gO
z53FWXkm}Lm$cH!_PBU;I6jj<`9`h_eDi}f2lVthM`5BicAIP;(XZyp3hi7-HvCu8w
z7MzFn48yd1t81TE&@+G!luWCK<~6>>UV!_YK+~;iRZk5Zbc|cCi2Iy)s@Q&?c6%0i
zr|IOpTz^^THJGlaNjVq}nwuEQXG%=u{$b)iC+)Zv1hkKOA2{;p9;)tg0<HNH2pH?C
zbMx>BHL3Mgg647ymGb~&wEO$P=BmT_>FMBh$ANn5oHO3n$gQ|Vj8b2znopeDkS*Ke
zyaOV)BxU11XlUs`#e6*QGXiFa_VZ$9x*jY7Zn|#L{KnE97ZXbwG;&s$V#fQ3!Jd%=
z$f@5|Jya7`wgwNpvzq{ty9WJH7GJD0UNetd7hxQ+b`K3p2kPORoyPvw!VC<lH*;To
zVFhY@wlgg|D3^H-lnd90I#)(OS_Bp>@orXREpnP8yCPi-aFciH&<uJA-E6Bl&3HnK
zn3C>p;xvT(BB*2yA-OW_7L*Jhho)D=@(^}&1r56!y4~Fb8fn*{Cs~uzbdy@cuK`Bc
zNTJ|#<y>BQ+PDbPhX2c<dg~=M{gFuN1+oa0#)Z1tet5wy;+-)w50zWNS|Jjp_6u!x
z%^LJmt?XD7B$7n|_lV`sQ}6^dOQLJgbvtk(7&;wsn!Sj>?;a;k^J^dkzr?a6v){6H
zxW|dpeap5YA%JC#Th_GtSehiSBQ1{}6<@z(&?e}1j}xc)oni)RjCh9JTbO|wlJZ`(
zfG#3M96B#0RP3R<e9A5@p$6=dla#oVC0@j?v9GxZJQ=fm*V_a9>}o*0R6XOE)Aktz
z)1g*WkZZ0-+lg;)r>U?`$eg61|1Xbw6N}SM-~4q#*F?1$E)wdNt<yXlUCVE#CY`qr
zcz^}XZCt2~Se!C36m*gTJm|iCn55M{H20wwvHM33c_g5OO2v>r;kC+#%8cbHBQGt&
zh|HaE-)gt6Se~|i8Z(u2B+XTxm>g~&HBT1uq4Bty7g`n}Fr;V6Fae+{G30vX9g!%w
ztbAzB8iM-o&<8Z{b3!E~5Ss%DN8hnP=~aougEB_Pa+7^w3cBA(bGMmWc6rvmL)U$X
z<yRyQ{nQ|yR?ca92roiYy5ot{dAn;fAQZ5?9<zu|S!0FrtT0J;jTP#Kd|ljiuaYzu
z1Fow_rFq!{w#c_~Xil&li7yY8Tej`^9yFs4EiPZ@?>_}c$r+x}cY%1G2Fy3ZHRKT=
zG*5Er9zE%WoY_RJJ7p-hafc?=TUI<Ly;Hqi#Thc;xc%CD=HyVhWz}yJADRN6%bn}!
z@ioJp1J^o(UEQJzS_w!WR4&}o6Y8y?4Ov0+dhXKWO@CnObX?dYR?49{<G4$Ao}iMT
z;lh0+QvDbQn1UKVPgkpjC8#}%jt}(<TZgXO0kc#z>Arm>#aYmv;)I(<<t$EYja^E(
zpcQo78r0*B9vb^dQaqY@(*2=YC7yT0qY2kQT8Vd)i>Y{y%^dL&`$*zx7IVbofyHkh
zXhOM-b?s+yGOD*Rf{6!BGfzA_+>YVR;3ZHQHEDcef**Fyq<MR=bgX#isY43>!||}G
zoeSQ$GY34KA)$n13fgY3w9g%{VfEW9o7FYi3Uf|yOI#z1hHfl4_AQCGkhe*oA|*PO
z2@(W<FzG>ar&c^$(Mo!Ynth-#1QHsJ9LJb=3+b{9m7U|A@wfz!hP|`w2YSqd1NUtV
zr}d<IpG-oxwJG`OtZc4j99nO+$Gh2K%w_Q{<{5*yZ>5+x3F86W7|;0~$()?-I^u=p
zHjp$AVw&CBw`{ZTIH}&Ujf;<mMm8BJj+9C3hjI%q=!N)aC+rC)SHTkMrK|6xp(+Xq
z1xAgL(R?*z6c{<)y6z~52h%=OZfmCXm8AQ2jNtlN@9lzyfpKyuCEW26NIZ6+=RBxM
za|ll6f}|g~LeSO7WCAq%CWae1bt!(qK`-pbr?$mam!M|MYtnh!8pomY(4BO)9Upnk
zJD1SoOX&1gi;w(;x3^L>#L11N4P6>56LTi%?j?r%^U8^PS^A~7V|tu-n&)qejuH8?
z#4?gu-JD8oNkcgoVqa!=)58#`HAXz69}2NAThXMq)?QoBvMD79f7v9VBpkivh-YpK
zC`X$~Z?`5VgGp~=i&M3r^ES3*)Jtz`!f{T}bs^^=32$SYlZ2$VF?)|aX&gR5&o=g<
zF*bx+o|^9=MIFJC@5|ZA>J0l8*ps&0y@RexJ%-Ssny`22`5ID85D{2%5OWnIq;I|%
zy?r9BPSE?~nTB$$c_TnPFm6uo)KH+Fiz?{tDc1{D(%V%t_ay19cB4}#yj@TaSspYR
zBRXbTBdJ`vie9>tQd_*3Tw4i{S?7f{>AAKUv?uMX0ZG@T$j_v=3!~+n{Rww3w^~g)
znf5_%Tlkq>czesWmwXQ?(}-s|>L2JH<I+1|+nu+9-i~RR<qveterXKjiN%<YHZI&#
z=QlLfujOk96H&x-w_n;XH4#1^Y9D^hw=-_Nrb&+WP<t?`4<!xYrR&nQ`%70#*^z<o
zfP3o$WoUgVkv$XYg-c~WRC_7%QZ4!AhJ`5v&@5|xLuC95rMCGETd(tlTJLD*(niFs
zR>wn!zmSJYlb70pU)~x>fY=_oJTPX^o>b~*1U^tkYY*Lh%opmN^dD%Jb@`?Q^||x#
z2lO0Z-?ZWe|56X1@=ZBinGfxC-O2TXiqu<nCMEj%AE=6qht9dnOP5Fed#LR7QX#9x
zOI6!T`O51<wd#K1=erBI34-urf?tY({g$n(d{?`<d|%risHcSZrhU}?Qt}DDH6W_y
zLuYI92fBNC>2fcHZ%SsR@zT|`KXkXz{y_JbZw)%e{(3)~n{~6+m#(J*Zg4##=C_<4
zW&*P5L;DY*^nq*b{ZPVky%mft<L$s4B5%EBkCnIH<za@uDG{{(K=;;n3~)btsj?II
z(xv<R<@f<-Z7=nBKHs#?B5w^1^X8?Wb?*aD-rt96ec_nx?ZEu8Jba)uk%z9{%O9v`
zvi+tq`!g3vf6(<0)RWSEQ|)8EWl0$EU6vFG^IK$Yav;pnUn*1H8qD4<?WMZ<TZ7pc
zt3K4M34LoAA}^d>hQ9Qp6MWzab@)v^W7-2{FL~(hTmL}azvP?h)Mv=C3-ythgzhYL
zJhXa_&~HlEhd<C&ABjo)&iV{F0X*w!4;`M8@tbnf`K@94QRz2b?O8a8;@OKnR7QWF
zVVC|6jV-A9)@o9vy|wBdkH{rW!gyrn^j*pI(lwt&id+d(-g-?8ek%r>YzeEsbd>e2
z7(mz3mlEjmtzm9?hMcSkvg<>+j_?P%_IPX9ZNHDU_f~zV4t-uQ2I3=yun%?XLu)Vf
z4^+!|Yaoc@OS|v$JXGqlP7&s_uD1s4ivG|!TYGC*wavHelJ*D;Vtw{!ogzO$A8!qN
z_D3BfNx&#?4JuCs1m=H6OwLKzH#F1a(Zz^wKgS=aywltx%9poZLi@a3utjFQH4wpI
zzLdz|4_#g6AE?THM?vPUFWnyA?xC{frP{T;HK=O;emz);&pgLU)MuR{255O<8;^h{
z255QKDWZVZ^3u`jquOu3<S7p&zvf%R@*}N>u2)ed9BA8be~AM<9=gXnA3Z|tqm2<z
zc05v(NUx)}m+tXMO>!>`eQ7JNLPp@xN6{om=-e-KkJN8E#(F!;cEMW#u~+1wX1_iQ
z74aM84_&~Q<)z#GI3DU5H@+z`J=RONeh;}~DeAY-)Z%9m?WIfgt)cbO-m*>0x6oAL
zj+O6)N(uT^>s!N+`Zh*1^|ykl@6XqK^bjS)QTtZyq4F$$T0BzZTf^38mMei_`u7*C
z1XX#ycBqVRp{YdgcqSQChrTt~zqXgM0o7hAbHCJj@GUf<y2eXMgYnj|s=Q2|X|8^C
z@LRWss=P(_-1QwsHCg^PhI7PsS@#z$f1s-0N~K~b>UwFb`Yqz2yR?@Q<z_yVD;NEx
zt3DH&xT5u29fR&FZw+g-ZxdFDE!z591D73Ny4&-9nqOdgppHi@kj|m+7q0QG87TQ2
z#xvq1bQs^tn3B?A=v!}h*MNtT%I=|~&2ME)J@<Bf9|C9Am#(tpt-<t|@=~^B>r1zM
zFH=f-2YG}O2_AO)16|*1t&-wlKOz%TQD0ccBc3r7b?Hl&elJr>q|x%oOhS#8Z)Hr0
zHCpyt!}Jr)hmvLBp{rMU`)d||d21l@Q-3I#Vcr_f?Z0oz-uOFYTyXt<5oCVo`=ylt
z&o7lR>O)(99+>P8?>r~{!?RG5{$W1r6!{<Kek)*qf4x-KAG)wS6v;Q0D&?(#V0tel
z{mgu+N%NQL+}}2=k_lozV@@iF{RqwFq4MiXRmWRHo6nRc-sp>V>2IB5JVQXN(eiD;
zLRYOXU9FBk&|RNZifE&)ynVGL<`KF?8*S??s~(W*p~RW^qU43>&lD5)6dNz~P!!*^
zwK3ir%m9{`_D8yUuCV7tkRD>boi)wo5pCp$Snu!GSZ`A{lOM%`6cJKhx|=);8c8B#
zz5UW5&-5Zq#C`^xJP~`ow6^`och5vr-?DA!xB05ri7J1fQohYsCEjT1f1vo+KXlG#
za0oa0NIRm9*7p$vU*eZ)d!c>#{cVb@5AWD2E|g!CSfl+}qevXlAGLtY5&ao)Qb&wu
z#K|2o#!Gu2^48FGmY0q>zxVu9Jq6t#sLXHcRY@nX9<_jo{%gMU=lSV{CTt#YNho@G
zd!Sw30n@MbsTD$}?=PhhTI-%~S(A19rLAlI16BQ6pr9N8uZI!<dc1U(FL?=_qpX*%
zQuGhhgWh~oY4uB9I^A`@bdOfw8nm?W2fEt&9!Za}`K8r6{XSR*)^EcR*h=4Tt)>{&
zm-aTse(4_a(lKU{huXydLm%dv>!qvBXUsi#PkX71@556l@k8FSdwrkULA8DFkviSq
zr$w#!{k!r~(zE`7uDu@E+B4?Zi0?edMtlYV6Y<;pbt<L26_A$Z?Ez6g-?D2Km4}*8
zXm1VpM7%xQz3x9yZS|!(+xUKIgdT132P(bvht}TvAE@@#9@=Kt2X>DS^U!a{4&dm|
z+$SBy_u2=Y^G|Ip&nEg%nflf%eKmQhuK7|^7WPZkWchljypKjg3X%R&-H+5HA;s=5
zZFR3dQ2F**f%dNL4^(x&bWGj#q0=K;J#>HXFF@CL#1OGZi|R{PZ7<!e&bQj_S%AN(
zdh+nU(&<<6|3FuJmMQ{}emku=-MzoGjs6TeIVrmSfv#_-73gSVzV(j%?X&`&p6}qB
z641B4bWM4wv{n8<_j*R1v=nQ;6rr_Wx*i!#S_*l-*A*F8zG)r1tcSL;`&*VMxNlig
zDjzY8src=<0v$u=OQ&DKdZ?6Vts)L-8GoQ!-tNk*<sBSCE7zC45`!?Lk0L-6()LiI
zkhbyCTHB)@5Qemumo^V$^H9%W{!RP*_FRDy9{LYdzCBl<lRVEa)mg_w*PPFgi>EyJ
z1NBU}-*orY|3G!U9aHj~e$nmWEB-)5<^yYghMd$C`<Wo5rr6I~MQVz@-Ws}RG5V&X
zYkTV@$=iD9oVs61obEr+UB>%pGRmVK5RbH$KhRa5AtxSbU4NjvK5G>bN!#-U6Oy#e
z`cm@6JoH2x|Ay}x>d6G<fp}u>m#)!XYM<YHYaoH&Tb4v5Z&@jGte39&%rSCPtn${d
z_v&w1!aaRGu&40)_FQqgq`lOH$ovD{^1V`dHm6-*+RJ{s&`ou!5B+uvY9|BFALtx9
zUy4iYeyNoDQt9PeIk?$HzUiLdjw{X6UDR&{-!nzjZ#&krEL8mi)o;g@=8^l`_d0cg
z-hbP%Ckp&C<itSl;}2B3ymZV_-g;*pe}M8mNl(07eWA|!QknC8T3nOmr5WwxFI4;L
zZ&}rEzm;}kmwYc%{ug;hoD3B*U)u6k{a>l>?`5izp<+J+Muv(lf1sk@szJqE)cyym
zy6QtsMbo!lX@96zc6pzHL<ss&eUy9PdY+b;r1cr`?$Ie;YI?!9Vo<y1Y~L^aamD0A
zd+XmuEOhtr_Kai>Zw<42yRA6ICH<kRjc+3sx`)109zSB}F7tbpf_li9Z>sIvKg;Q3
z)KwBytZy^64!@kdU%I5d6mNZb>6ZP{KK8e2P{~xW#~-MtrTwN8Uy_$@?eDXZSL3Bp
z-f2#7($ZeaorCsLEzh76oU}eOo#3Q({DDe+geJjB>$e>X)z)6x<}?3^P1>H7irA#R
zzI60H-WtZU6AYd6`#cP~-W^D&jAx}HU&Z%vv{Rk!rF*^-eR`zN{!&@bN=3ekHU2<#
zJ=y^oE2O-%)knmVvO?BdZSV5E{Rh?k7Cq^gzv-;+M8{Ut@>a_gmi1D7?2fuT@b!G?
zZl%4gkzVR8OM0pA_d}S{`fbHRSN+!SD`86Cfh0`nyQmPR^jo?9KY#sO`Sl;a%CCR@
z+kg8N|NDRZ&uXyzs=xl7%a!rxfB(;)|HV<ojZovi{MWDl>7S-F{VMVQpa1^f|1%5y
zFMj{$?))#R{YzK#H~+;m|LZ*XzY$2ow1xjKivNp0{=@%r{^yxhn-?(cm7%gf|MjEf
z38H2y3(fxLkH2$qyq;h|r1B)lQ54C){wP5xNX5F2BKPk<%5Yu4_XeWO{ritH6`QdW
z%p$vU!Ov1RadMWr;Ag1}!<&Wj??21jn_OAwf*++mn|e{|1Adkw`8=^wvQYm0XBmQ^
zqe9$57yK;6ZCb9YXQBN2&z|Rd;>_Q}NB?|z>UV|CW&M0zKMP*=quhOGKMQ`~kCOMz
z{Ve%`KT3XH-(R}<p+C!h(qaEB`=LL}=DYk$(F=c;yf5@GML+aM+0SeJS@sA0SvKG9
zUy5G%v*dlne<}K*Kg;Hu{!7sdf0TSA-~Z<K?vf@0Vp!!KA0pTJ2$DgW=?266StL=E
zz!z{3H}FT1%Xk(J@OS+gfP;s@X{+6>%V_aa_D8#)qCZ;xl>FKLr|8cfJY$sN^*`eY
z)+h4%k9z+|DgQq{HQ2xY-`>D~{Y6lpIFrBrozNryTk`#iVf(}W_Orit$%sm`&Y!P4
z%lP*!|8Toe>EE?`#KS+_ZdCer?H*zC54RhY{$0CAa{bfq_f-0~?Yb2Fr(bZ^`nN5+
zP@VsD%Tep!w(K(cpMJ?%>)*AEOX)w`DmDIXi_PPl|I@8Tt$*9HuS5U%^G2<I+p;ff
z|MbU{wf<epr$t|X{;77o#(%$6{?A)IOW8l$?5EEEdAq(Y%0Jugr_N`2eHP9?udm;$
zu&t^8ms@ad<j~Up`ybKd|M9PnGD_6ShOhn6P|1<mfjg7t5n8^I_JBB&vQ97=AQ_;5
zP4$E;BDQ?zoSG<(5ibd*t2>iXPC#%X%j330P{paKF$;;7iIOPmgeqR1R@^>1k8#+t
z;Z53O@l@8?!L1@*%xqoB`gC$=N65e)`BVre?;-2O^S`bOg@b8g_+@Brg2{KZfARcm
z{X*9W#gpPhhqx3(I0f;@$tV6HvE(bpB;oHo%~?RdXHH1F9YMi}B|kWl_JDLfdo&9V
zCw@glvYoj&kS1{4?j?Us-VBG$ll}(Wgx(atv427d5=9=rG{Un<pV`aJ4KhNjE!p{F
z_Yxym+KSj~%hR;%Unc-sSgD?GqBtBQ&7KVJPDHZaP86?o0%kfS*7t>r)CPnI%gBZV
zk8eaQCPtX{`}4MOwIITm$2vc-d;UivcddvhN|<=!umfd8*iE3{DJQ^In<o?;vEv(o
z69|u$vEPB4(`TY%?MUfL==j0mGqJ5!#FryTyx%7m&uiMVq!~<B6(OiH0uF)W8%NK?
znj%_Y!Y%d+2q5J_dk<860TH<*!ga>IpY7#lq^1HbPMyUwx|{%kM2v47#u1Du;`zch
zN6y4oDr;oH=6A?Flv2cyA}+j#MI=5_k1&@&&!SFnqsW0rtoY739HEUe@j6gBY!(k*
zdt!bODt?l%jmSiq?vzmR9-)u8LL=V<(~&bcLIsuHvmxnZ6d)$hDkDpZUmf@0KRJ}R
za5bi|x>H^LTr1~q#OU#aL;(dcK#U$w8VkgKiv%46*tvw=ZDQ--0vW7Ma&r(Er$-7U
z!oycDw<KatgM{->4|H>kF(SjW^-DY(iD<MivMv8298CnkBPzV#6ek?bY9}x?QQ=MU
zdcjdv$%ZExa({U&nSFb+I#E`)r(;|flA9H{Ju}6DVxrffu86!zNO-b>{@N$f=}ag@
zLfD?)Z)V@MbJ9jwmK`BYfqp^UBd|mkMB>4d1rbQj)`a%N!=a}P3r6NW7@6Z*z}D54
zlI>wKLlkh4$7X=XJ&B0?1cE158nBnh5<?t#(u@H|CRGIDd%82b`C$`5L2Qy8xlD-m
zPI4&Vc5je?q*Vf%WNO(-`?l>@L=ague=VNShxj2oqWls7eh`$O;2$NzT@&fO;39#o
z)FSNwkW4JVIkHY%vWR&=IFHQ!4&0oo5zS+#RK_8ThzA7j$XL-tc_;bAFJf^-bV!&d
zZXo7HkBk-ses3J85nRJBc?a}-0mRmr`}Sy%K#mw15}`v0@6Nw^tc)6~4p(;ODQL2B
zWG5F#hs3!kt!BYRNO2(1m$Mb&YVP)%OZQ<m(X4?U;3=42{w8ciulX!W2xXW_wUR&9
z?X#EuN9IT3ybsRL2tz?~>41dR^|sfDD8E30#sZ0tw*xjpuKx145)nunxH&r`fP_H}
z*ax}Ui4`%lWy#@JzKInPS!ZFyn1r>=f<lD%&aoN6A4X)C1gZ}@vEn1cF7e$Fuf%xB
zj59=cC!!t@@seG)&!~Gzsgdl2*zTkR@JJ4AWWntr&<M^z;P!wzIW6;u4fEtmAh3HS
z04@O=b_#i1tvD<rG=oI8N&>r^?Ddi+@nZhsYQ;4sf-rb$vjgoZ#NqI$st0!B^8mS_
zm?gVMw((dCkt30~?gWK#sHr&NWyK69rh-ILH=??eD;l_GWCuV$_UqlmP8c<-Tm--P
z<&hJ5&4SX2VL*t4(obMiAm^g6-Deiu1eo;j2$2q+aPD042X>OQC-i*11VS)ja)yYD
zI7BZ1YA(3yA?y9;rj^HNa1F=4MBo|`EE|2DyTCvbUNwK-xJ2tvXUInG;IIvNq8*eh
z**!HWVGTxP4JMMi4;N7lGEgcuKt)=sr^Y$~>Ve#~1KOHe^XH3eYQPp@N3j+P=V2bu
zp!O^%emBbF8R&)N8Lhx^XExqc#47-5zk{TMvoPWoOg)cItoG>83WW&71qx?70$F31
zFxiP&Fx$~+U^wjq%^W*Ii_q=v?+y&58v6xDahQ%sFah)n+yo}z<T~KMGI6--ywI<2
z6PBPw;9;V+;|}X^m!1tRTuFhg?-Q~YxJWAi)X26%(02S6fqkA&5{BO&Y(zwBY${yl
zoUOUCL+ZD-)$<?Ag^Pp&z(|CL@#R;x2~p4@#XXS82HPS33D)lSwe_SyUI81SqV?q7
z&4P`aDe6gFB`fT>Ksb!N#<1<1*q)#9L<q;i8$=7&b3J7SoSatw@)!ndWR-JqCrdqF
zN_J;;GGwSH#Hz`Z6ap~GMr4+A%+p!~*%!~UTm6KhNqpZxAw@3n3~HoQ5zkX8enMd?
z-N>Jt;aMDFEuNpB1XI2rLCo70kSiyrWF)(RiG4*dTM>?qgk{>v8vvBPFXZdybR=R<
z6O?`Sa}+ga&6bCMas<p9Cu}!N&a+65)Z+wXcl8S>ZC$7&#9+&#h7B)Xd&@J2d=&1<
z_^<bl90owYi%q12R^!cZd4`Asz!|Atfu8mrsO>(DPErLBKB38x+*s6=Ef9~XaExe}
zpb5it>?~;No(;5zS%}xor5T^OL}`yiYMA2_EfJ^@oR~1|btC5p-QJJx3>4xF05jVc
zhC4e6u${jhgn&`McN61-m8Igv$QGCI^soDh8)Oq*fVd1-&UD|6QHPnLoPB_T?ggAD
z1%U4ju<_|p9X`)|#-5-K<)Y0I2Pjo;HW3EvfqGYW!b&bbhkKkg5xB6KJRn45>Sn*8
zk?;c8+tCJqn|uJknZ3l@iFg7)Vk89|g>VAu`N(Ge+y=S&ph=6+*92Sld)B}aDL;Uo
zDG9hDK^+?#!`qg9ltLn^7nWf{PLz@L$)anpOk3#NA7{i#y_;=p8(tq>)ZZKn(j@s%
z;@EXcvQH8dcWb210Gi14pObhqq0P<iKi3wC+JKuB0KgFz2;tR-oLPmF;0W0LYBf;S
z$~yZZ{y$J-JB;Y+jnDx^#fTIkFi(iC9%^yqugUP#97b*;!l-+K1=xsGB0wRt1#rd&
z8@oAfbps_59RO_!a<(J1JAu`ijp{ahO{T6fFzLGHZ<Wi`uo^E6HzvXbz}kp`#EYq&
z3hj|&mBWdN2;v^~qwR#YEHCd)oo6!?Ca68mG~Ob{2(kXv9sE|d)BO@Pkd$Hp$0SUE
zdUm>!KPPi|z<rqz<{zeEwE4yqVg5DaRR&a|{5xDB`*2;F-G{tIJbd55^63kk_ZiJ}
zQXk6JBAF`NT7GldVb}b*l2F5$gg_L}MCh#NqQ(UmIj!`3UtDjAe)oCH*?=!SFf!Bv
zmAC*7Be|ZRR}kADXtFM$DbI5S3>}1vEsUVUocb6{dS-^f5Ez(PMD{<hW$SBd^mCM%
z4M`)gzjsOFgAn|8dyjtJF6}|q3Ww`HCh5TA*+<WQlEo|rDe@YyrPtW(@)6c}kr-U{
zJbFuAVlzi;yq+)x365TfR_|{&%xXDC*kL17txb{^Jyq7o_6ywcI4M#$JHS(8vfZZ5
zB-}fu<*{9Wix2>3{;sSM&3nd7ANe%P6BnOvLKB)Le!hA$mTES%{bsbodP|zZxcd3{
zMy&nZU*TS~f4X5Q_zopO{QV=}@g}*w!qtC=B{Txex^=`|ZLxh1^yH8(Oj;IfYy=*m
z6N$Bf`#cZ%*;X(&6p&%PWqqTBNR`Np?z^D-$hwu<8z(IL3~!mKs$pGW*SJhAE#Vp(
z&7?|zb)`q#<yXQ6D!Ugh%u|dgPPskAe)tIjmDt$6Kxk-xoa7LQ)R&TVe$G~_-+u=B
z4r)NDM+?&YMhS-A*~a%ZD|YdKo-y86x7HFiKbi7%P>#f5x0W@};SO^T7${S65(S-G
z&#)OAMcXaJv#xeIl9G+y84B1YDp|lq#4umyvpO?^#y;vmWD6qpdFKcbC}Ac5{kk)7
z#%?^@#M4!4_Pg~yG@GC+^po2V`32=@GDq(q>Um$G2eJK?P>Jp05ws$U6Wh$wZ6-9P
zawZ!4%+*)H?%|-4iR<ue)yG^R=(&#z>RNV^n<yk!hso^^dPP0DUQ8hNs4F4)T`P>t
z-az&$U^Uc7_O1I-qM5AO?}bw>%=gs~6TjE&Dr!H1%J;X+VZD9WMjHp#*b!mR+vKi~
zqpcb7fSw#&{WYB2bpaWIqg$Un8aBrp`=_urj@Sk6dIp!nkrvRDi0WfzD)Uht*)jLz
zb*|}0CPUyN0za_#kqr&M%M7gLY*YX@`&=H(Eq1KfaQB<cZhS;uJ2u=)KWcOJ3*!OT
z2#ciVo8ZRDIBIjzayu{kOll6fayU%gJz6u#p7YmAYDk!;&kVMXIjpfTv4d}7*BvBH
zG{k4jAtslXJ>Xr5r{Su}ItHJ81ZTk^aJgSyb}LY$<6Jb6dd$U@$Ykj8EVR%4ENj3`
zy!~Tq#=GT?$v&QiW-H7rX-4}oeK{b>w(Y|?{*oq*XLf!0mF0*4n60m^$C_SpRy(4Y
z4>HSo071DAzKc&_Ozg-}1{A{VV~cCOKiAF{sCrtnp50UvzIsggq3I3G!8)jfGQsHs
z$HD9ZNNkh9uy}pDV4s8P?P|sL`ZzMlf=x*w!4`yz`U7pK#(^s~p?*55cHgxKDE;^k
z+Keyh=Q=y-BYXoSc71@_kC!PQOt59L_iUfU{v<~nb;Qog4#f;8;dUA)f1G-BLh6HB
z$==w@Izv%*4mdYRxRtE#<r`ymjz@gJ&#qO#vsYz0Z9ED`Cae3t6dMdPp`I-|_8XHm
z;#R;S!5fDt6_*CKs&bj+U|Vuat0fx-{kP<v8wr&N`fP~hb_&NWbDteU61ZkF3nAe!
za<IWzrF`E_Vh@3^<nCGl9Fa^5sJJw!C#Klh&Yrq+drxw7%tESOa)yZv+w2U(NV4m2
zdFptFL*`MQIKnC9h5&kI?||-b;hT_hj1LZW8bR|ttZL-h1hTyi>tSGV+A~OV1uS!B
z<DS`;{ZEPaz;S-tXP-2($8-ME`q_}RSsRjOkqqT*8CnlV!ZmPd@aVBE8xg;7%+9SD
z&RHc}2XeM^+wIQw>&AL^M>XMQn~5!A4SSE%qd}N_k8`_XfVdQbM+2~h&j)aE(9W*&
zjw#8F5JKd`Mye-V2q8kg?^)+aa?XUz^N73>qCq%NRo^a`kt>@)ldT7q*be6K$=5J8
zn!}#~wuy}oydQ*cWL+Ij-<vydk-j>%<#+xZ{l)j&XP&TdOw|aC&#wMnw;}QS#lR6s
zb2+>k$ZL;r+%D8Y_6Lp+?i@5IXraK!q0QNq8>z6&f(uj^1*U<n9z(nlJ!PJwixqEe
z|NpGLOR^=oj%C@;QTPPn3ICrFM7yLKHGphnH8M&hS~>mV+5mX29p3QFhKs!CZnm6<
zs!FAjAPH<Ra8*0xz#S{#%+|zbYV)VW<AS47VD{q>;4XF~K_d0|`I)EfjD)kL?67C~
zG@u{mn=L>^6a&L^iKDMtbquvjSbI7Sv*Vo=_S#ut^_FyowR{(z!>}y325Di(Mb;U6
z#mL|ox3en&^T3+#(@33$GpOyku{bsB2;c*|o0Pa$5=IQO9Td=%Qv-)uzAww6S>@FO
z>x}vedFt&*%-7f)q;+)fmrkeLG~yslJ`Ma~+tQ3b&uRCFPlGdbrJJ3{d8$U>M)DpQ
zDU0<CxF!8O&Z7m$%(P6zATBc%Htc7C@z{9m*r8#oJE~kLA?qNql_<Eooe!Ql-d6JT
zg+kvo0e5Q*9k(Tp+*#Q++dig$F__vRroU|ZY(nT>TJi2ypx!N?2A0rfiSQXHTo~pX
z>98f221M!Gh#xn_3Ikhf0nehN>-cv1kuGeowJrI3KYY*;9E}C^J#!tw$)RB>DOD6a
z1MxK2%7`i#$+rZPXJ7;>M40bq<lAhUAwiP<>9)mbERNjS-|!r(jL*$Ze$NbGUpymz
zvWLN?CI`c8U6KSz!E8X#XfSQ930$mhbr5b{g^5a=5wPlCgXTV>f;H-RR%o5Tar~`h
z2=&5VrXRRRYD3@|e20$aXO8O(1q3VYLb^9J6C2qKn9H;C0teX|G>Xm@&%FxgNkBQj
zIb811Fh{=^-)#?rN}Ud5kwvNo2KH5K!Uaw&7GTRZ5$bRaa~2ptF~H0o-4FFK4q1;V
zyT!EeLZ@id!rUW~Mu*ef8Cvx<8@6tBcLt1~xS1v|M_~7CwwTm-3JRtjm$Ah$v*ws_
z(_^~VL}3(gIL-NMN`5=omM;VD9>r!_7ANM0gx87Z3Dp`1UkRpmC8cv{E-{mh8#C-!
ze^vQ1kgepRGwNoc>igNO{E%0Zr^<HkUM(@vsW<nP5haUaeSvA&@__|zHE$eqbabXb
zUmzJ{9Ae#IB)NN<R{99k%W%qZ;WONKF|-ik8l$S?k`iXN3pZPvK%T(il;MsgH)}lg
zQh2W?pJ@SgRInr}6?eE5+beT%&bY~~eQRKHr$ueAu?c~(iLa-_%ogfjW|n%BvwO*M
z7z1|tv5RdFFEx3VQk=oczL@-)_UA%X+e<*Dr!j~sn!)S_GH|iD!O$N2!j7|<4DuLR
z8VN^+{HzwD*3MYH3O2W5Q2Aza**vmhlx$iY!^mtKzg{d$1ha4apyiEiFCVYjKX4$*
z1FUvtgS__{cg4;s*GG%a`TEvfc4fpcm@#(*)AUZ%*>UcI&HX=NXXG{JOL2s~3z8SC
zt5&vsBOo(@VBI#SqU+6wDt|XLZyk0<CBc3#(+9T!%k|z5GIVl`J44WY#xrzogC%-3
z{>aPx@0KqiQ0`84<7bX((H4te3nt&(Ld?66S%Ed{$hM^iB;6k@yZMEZccVKjId1yE
zG&~Qvmd5?Esg9v!&_3wnI>#a2lH|Kj4})wX%V2gN6Lh?LHpjy;+$#wg@60iZ(vS_a
z9D#HbIWw+iYYkMqD{-CjaP%!J;O6})<5{lpK%+CeFNqoUsB`j8we#TaHL%I)Cwmll
z2suRgPw{P)<MDv!r3jWC(-1y-cq}Z%1p6Sy0Bg@vl(^j48GThShr>L|-4%6tU>PM!
z<}>~ZrZA+@?UvdCu=z7il7}pFAx+s|(9~lLIhzhYYD2ZV5`6*m1lh37R-iVJW{zv0
zq0^jNabu6`j67KZ;6~lz&c|e-3hOZ!O|piVlkGfjIj%D|KGSS-&2*pfOvYUfwG1f%
z&&@6Q?RKeoJDeQXEXx659gaoT*2R13w$B5T*D70dK76yn!Zi_8{Ja&J?t<2U_V+w9
zIdGA%9@uQ-9F6WX{)&|yCft%Agu(L!@#|wCTJxSE_rk96T^`5ofdpJ^%mZtTtBV{V
zRq%8dw>>a;%g~!hVc9JbgxGVRsov)!?CNT+)_Z#LZw2`<2q=AIzCS;2&s?oJg|R!@
z9NCzw^+G%@Ae+pv6ayY1)y-V3kBR5g1j`Y8{<IssQozD0gKcA_F*Rl*gkmCWUk7#u
z;4PTkc?v~teYb2}`lcCd@!ywXFGj%AADB<r-1fL7bhi_l@xbQ68Nk6bzvKXQ@BEm<
zL$I<T6L1mW42)fRq`1!rEEs)glMBQ=*^e!>;t}`2gAAD5_d6rSjT55W<HHCLU3^M9
z7x1MhG;6Pg6NHG}fpUVXCOMzJ-GS&NBZ7_~swb#Nio2RC|5i$XfM*SxMT@)8LZvLK
z!;W?@Wur66Up?_QY~tRqw3}FNorCi!&Qx?*8gO&Ss6?~AqX9i5#F!Re;2#2XsS88f
zO!o%X#6d@CZRfbRdKL=Z!WPCVAG%SBhn_LS@8PVwdb4L}Yex-4OEh{kCr!!ItwSfC
z4$zzS9vSY6OEM=_(-(uHyH}&N>=pEU#Y3%lX6${hcJ5#s#e3z_^ZBaHN*=*_wSWqu
z5-!_Ajd(uXatn^Gj?E6?ZkO5|$qr%q{Qk_~RzWJhg$_MEim8UxYOVMF*CxcwawgA9
zcjq{J$yYPlyWmN`RKptX04X~0dr;uYt|jPMOf^yDaoVLD>vma}$U#qENs4os+5Hot
zXQr<nB{TA^m(ZFjTIZ+U)>w>l_PY1F^o&eJHT03^ymW91p#0VM8!N=ZEd}#5(pg=~
znI5o(FiQEClJvrIO84h0Su57*hs}b6TnNzWzG_f2R`5+8>lWSG)zJy{j(1-5y&Elg
z2F&uJ=Y*TUkV{9iSfMi!&09e!yi{co_svz&g{P2C({KM^;=`90D)%3!O*^%m(eLT0
z*`edu^`awv>c!&f7CKcw578C8HYHM7?$ypTuoQcRwA!tACN|ljxzt;BYV%xf2EIms
zo_<T*nsT&{4?PVgAB9iT>_#pGoEb*<I<c6RBFETD+82(cf_;0C5AY0Ax~1;RN3lg3
zgegDu=mf0SI7-0hVwEw2Rx-GYwoz{lmE7l|*{w8aQ|54-J4if#sXifSz2Dt2c9pSO
z!EgC<wL!9HXbF#NQ6);2cgY^NUR(^+u6fsc3scK`V4wR|yU1)ST1vT1b0)SVn(iAo
zkaVhou~(guZmIY(LbC@Rh~iO=AI;qHfp)xZveDL}6Lk3&38od_eJ5ycDyVSHdM_o@
z9adMgcPINtRh%Mugl^GBnmp*l&owB4XbN9E4YY6BPIfj0Fz&f;y`v6`e;lJmpCMpr
z(B5oDuFust`V#HC*82>LWK^PkjaviDFg-81S8C8r8YJkM+dfUR$si<3hIpfF1yJ48
zPGz+CAck(MhBI}>@S|OaqOzyy;wvBWL=((2K4(^SBtgLGYt__(R#J6{wzclX8jCex
zZZW0~BI$;Ok*+jv(XOZgd7X`Qs~w~z5Y2ba9fZ?$i%y*$J`SCz;fuE3?$stj@}fD<
z`()-ti-+7?F}Dn2>Im#5ZBXm?Y8%~e(VV^7N$3udy(Fid@j*?jwqdkgJ8NolUKOk#
z6%;g%fw9~cL?2kNwRwMEQPvgB`E#u*w|2$yuCanPPPKupe=6vxp^D~mf8PhumbfUc
z^tmmaLIB&Ww^GbDE1Fx_s@$VSOAhNwzSlbM^mKTrHgbcaHYj>lT=y0>LVXHbUblZ7
zipHI2FFl8RRg#(zw744*^!(rMx2$!G=7Ea^u#a@<8TpD8E!SM86K`v<yzf-V!VgtI
z=x`4{=ozWAYCwzk)_aWCrni&iz4C|J7Id`u%=qrDWUb}0(XG<6gpXNw=}jx}d)`l%
zU^9{*$Zysic%kRh&qmV>BV%ckb}F8Hj?cYrh2TS5k8aP1%$h--yDi!ay#`z^roNyU
z1UJsiXQY?s@BMw2<vaBS-k%dZt)Gvb%KdrGtPG)je=4Yx&`Id$HJ*u4PG9&a?{$kJ
z;;-0yA~9~YXJ(#O^sc28>75ik-_wXsY-3DXME-Pl7vPyZhNACV8~68<nG%*qx^Us-
zf!Fow``+o!Y&SzWWVlq$Gtd$vEjrtka5y}?btV^dOIm`B;^E6SbhbNRHuMZDx9oY{
zTP<6XdrpZ?qC)5REF{p`*3lENd4Q*=Y~-oAz!zyKZ4c>75E>lg2};{V0u#(8yWX;>
zV;GG&525WQiHRYXxEhp93gs{JoA=iU9XNtPTbcJ__)BI{-?9)S76$)(-!6o09D3xl
z?-M3!+BmJkit7*v()s1^f1imsIO$Au_gW2#=a=a}#1xn%^})2no^^FY4=rsStrBvu
z+Na(+m^G&IHP&19%rr|kiz#sL_ApbMV1ud$J=SeAVTzg8r6Wmif=%ukO^CF}NLE9H
z4?@pGZOz@*jP9inxS<+&$@k9Ay>Ba9{X_?PhLNII(^41{&>__cxf^AFV!1jk)ljA2
z8<dY2Slrx4j|bPPscf371}p>&0Y@w?fJ4#~#Fw}mGV~Z?*5b!v&A05ak-JHLtdV+9
z64M&(9>oUjnMa4ZW3AvsA_AEDhn@+4s%h~d`p}I43AE>NZ|pQn)sv;-u2#_VriX-3
z_|!gh;)M^LdEeXJY9~eR^nS0l<?%4JNgpEuTY<>H6Z*4DbdW0l52Is@Tb4ar)v%iD
zU_ejbpMtsFl5AaU;PePcVD0QPD;&uqj-ZWwPGI%$9>K{RH?`5fF2!kAml`V`3pZd6
zEP#D~Y9&3eE5F1z#}+aU96-7F1f9bkpx<c;dF?u^y}O44339|GSYt7Xy0(J+plQbU
zC|CBbCTuH5wfNEF#{unbcxAczXSM5TpBUE(RKSr#abv^H3)Ho{v{lY>Z<Wu*YUa2c
z+V5D!iV%|?THO@{+NwRn>5}_itZjQ=B(x|2`>;w&cLz=2z|~|bDx_Z~yL&1!Bf3^&
z*h^|yas=M8$2P3Geja(;vS(poqCC0ITa@H9Rzd!8S7+`nqPF2F%DI(9R#<Vzyu>$H
zthI3yZkG;}DyCjhjB_0$j^^gvHTF57i%=`*Ofg$9IAPX6+t~LOh=sKpO8wke8_m>@
z?yt$NV>(B3j9X5p7gMnS^}?YK7+T$D6FNCqP<-Twp4B-9SIN0zosXb)VVZ}I^nnZK
zJz$MxCwkfq7Y}IeF^4hpvQQ&xbH_FmbRhf=9exr9#rL+0w&>dfpR;U5?~E{5(lxc}
znLOq#TF>CE-}kxeEn5deuEI<2)3OdE;(l6qXTOvfThZ*>%E)Oq)GHLTn#wIY<bVww
zE8nt1`XSI3JA}w-6N4y9Al9WEA^}tRRJ#aDCX9BacHt>F)U>s)Ps?z$H08z+(<*j6
zVf~8D?J+{J4k-4*Im2y1b4Sw(I^91B%50|gK9k4(P@+IY%MJqvZ6~%Xp~>ZMM2Sj(
z<~3ttoNNSCy`1bua@OK)7z-~h(a8oiwA9ZXjM`1yD}OA!NeE!CCL$IqzC({?xzx18
zKi{jNB4&bPEscA%HSLWjHi82x&lJ(sd)M2x8`nzF8H=5a?~+NI?Y?4&VQ&CU3us+2
z-{T}jpa}G15iiBji7uRBF2&nTbh<_ev~mh5+GoCHTSlBkVdHoHU=@h}0W{C(psWJ%
z>n84eyK}ZEz9OPCcC9k|M`%ev`}EC+LubJ+1X|~(8RkOYLP759Q|}^JMPhI}w&ODU
zW!WX-=KkE-MjuF#^E+&r?Z*Ys7VXH^p54}&NuuZDZ^-sNG{vsUq1tC6Mt?<iTA&P4
z#UZp?@=VChindm7t(7ARQDu#E+8T~v^GAdN^cY)pm>k}WZ)6i9Xnm=n$rSVqW?L_(
z^>m9KtE5;3^_oKH40dbpmR*Am%Gt2Wp06z|36xESwKtk7_jz#&Qw=r3n?x7RBGBq_
zD7hrWgOb%ylGjj@WVDPXcQ6wj))|WjkHbo__ByM^b}KFDssdXWuLqKIO)rcP+WUK*
zZv%7jO>p4A;}zkD?V>`^UReK04q{%mv_vkFqANj1L;M2T!cHz6?XYbFEiK>MIx8@f
zKRl-OxD;niJ?pU<^1a&H+7#iyDc@?z8fW^6Jx(l3)?S~{2#OG$+Kyz|l6<vL<b+?1
zCii0y-P#C4>!W)=D4ga|1bVvHrB&hc(C$$Xq9{N}kt5ax2b#iR#?j$(UuOWeN>lzk
zLf1Zr4rf=ek=t|u$$YO)$j}z;A9$`^bfM*Xje;9R49SVa_)z#6yA#v&9Pg!e)&RlM
zvF}@UO8~_{I={mc8a8f`@X6GzPS=7F+_;edL*x)j-aextMFYi6!YIp|OD7SUQ=)Xy
zxr+Sk^<Kt|;RL$T#+W)J9Rbf`d|3s}mGYqH6=kDw012ErhoFUBmk5f;q+U!95f7+Z
z<zw?er2NR)bo~@t1&NF)p}o(It}}2Ot?rNvJ?}s36`TcAq!mEFk=R^G%*1e7G|5ig
z!Ws9XVVS(~DuK@8wiFZr3DYL^ZLxxiW3kq5t;Bak{2_4`&@#)t7@<mfwJqF(p{?Nj
zhf2<#{a&nb+#<@sT+ohIn&^nVrh*}fU-FcEP>fNSr|d36>st4I@eMUvTDkW-{e@Q#
z%7;l1WZFUfg=ez1?p>BkUW#hJ4?UpRLMwzg<Fr1ri|7YM-&C}^7%Mc(X8~Gb{ZVlD
zn)ZF;vC%IbsAnKk#A#{Jdd6<ma3vZ<&y%ag&(N7KZEabL=iY7w{S4j)+vi%Ou(aQo
zG4Jp38Zw%1w__cMv&l4%g8K}!C+p7X9uO6_3y{&68bs@Do8^r_8{eY4CL1I<huN3<
zXLqWJ_R>y@o&d*!8Zc*88PL?-5Q54TV=BVQO}~QBW2r{xF1KtsgSRAo&ujM6hJpei
z{wB~KCx8X9N5bc)#I-(U+eB0YTIQ$T^U*Rd5oX<@+p>elGDEsnd0aZMY$PkZR!r^e
zB6742o#WFCb23Fh<AC_U#kV0yB;2K&El=n@+&uIQ?2MJ{pvGQc!R?px_Nid4=r-eD
za;pKR9P<(m#$%~kjO|<Wtml*35m&!k*3@s4qSH#isyrrBjveV%Kp3LYx?^s|zZ#tl
zsfc-uo+xUsqVg$s%H(7;-NBXRAa()bWZTsuMJJ9i&=p7Oh2`LO&F68yUtH7qK&Ql{
z`@I?(zIwLN+O1$0eC2?(`l<)I*yM(m9D^X*j4Wj(rRJK5(9D@PwC_8cbGF2WICRYz
znK+#)dJ$(4Fg|c0x&a+6MrWiP%QV`{zVs|`w_3NiI4y({NGDojXXMacyqlj{vEHjG
zH7zi$a^YzaQ`2rc7DI>5Gmxu-w#E-SFdV0JzVp4<l9NC58mmN(puYj^^+Wd=$@O+x
zyR$S}#Bw@G+IZ`ij$uuHrnC5@7Q<gRdV2Y|6gdZz1P?c%!jUyAXXx6|Llzrus(^LZ
zz1T5zrruc%4R7d~my%w@Y0VhNIHLD!HFtsp_GTl<S<B~J(gJ0nWqfXH(78op5nosj
z5{4{Mz|n5Y5SuGn_W0ZulA=aGF9)s2RP`E4xI2qZIi4gPKp8|V5?KylAJOwsi$s@0
zT7@Hw*e&fMy|SV#olZ-Z--#j|`z=clx0;5?uv@n36atD=)1@164&)N|^j(iaR9;O+
zS1j%e&N$WJJ=6!HtSLH{?|vQ1Mzhhr)4hd3Amo%A<}E3gBe4|b67ETyjX-N!_xh)!
z*c^3i<@m@p3xKNEa9s?6yW13n%8OmUCPEu+iaKM9AB$CfH>|;N2d|gncqux&k6=};
zZyoHsPG9FiC0Yf&YtpLT+VpZ#3lGrB2Q#hXr@DP$gU)3Qo~nBeL%Z~RpylQS(<nF8
zj5sM@imR{Hn>^YZI+1CSnE#H$59lOlX=vHJ?h`LUq=9qY=iSM22JF~vgb}zb=8EyT
z;tGk^iLwTSH4Q~jL9}_a88jnIAX@vlWeHZdvezbn4uUnIdF@-a5^`?zHkT~p)B4a<
ze1D_L4xcJ}+Q&`4iqVgXhr~$ZoVaR<QT1%6+pl}I!8a#5Ja|Y{dhVj1YR?-l274X%
zA(mnlp{~^}I0!reo@xBjUF^-E3Vs{+Ox?|Ccu~|MZ8pLD$t;!$i4sgP3qS{Y9%!`g
z*mA8!W5zNXTTjr1+az>_6IDghpGdtS7r}Y!MK-|bSouQ-&m)@pt@kWFlU~DF#zt3Q
zD+@DA2`FMdzM0U5To3cD;TTKX>8FB;cD3L=f~l+y5%+-Rumh_!O030aitUBUUMGhM
zqB%Z)c2JY$)i`68+h)*-+X-~VMrI|{AR9|Qio@830#27T29|EU)`M_B6^Ajg?i#G3
z#PJJ;qhytUPP3nfM8D|fgdZ9bX3&m<jM03biI8)HCG4-NoD-Js<8<S3>BP6es7Kk0
z$}<3(+J&cccENtX4&q=qkrs@tH|XrvEIstskLmm`;;D?z@JWYql4#V$@1R-7a6?;>
z(YZKj^E@WKuTR;c6D85uDXUFK41%r4D@66*uT|r4uW~AhT#TseLqOR?Se8d+lv`^@
zP{D<j?`;isT#be+G?dLs(G*@Im31L*C!*0~Sr%>^DUp-DMTwzM(RTOmD(gbBnuxkb
zQYBW6+J?gDqaUJ7nXD(Bulhr?<bt82-wHA-)4xncI^Uv&Wu4J*wW?i?vPQ$f893Oq
z6U@<wL<AO_ms*W2CFt_eFNQ9BGT4I!v9S_#BzxVc2lfkfx-IJI9H6YH^}1S$y-H|I
zcI_@ID`A(-n!9R_bGl=otvhUcD1(&fYVqNc48T9}dzc7Zy*X@AEj~lI5JE=`M)Enb
z$<?-of{tO+5(e-sTZ?yZ9Qn$n(yoEF`Kh5bA5m(H^Y=Q`0R@lUT`wHom06DEpypiz
zEu-FQ3COW(u|$K;w(Ju_Ie7+UYtCuf9~yU?`o7PwCFmp?Yjdb%!Iae~c1|h**vVSF
zC<w<X*-Z4#a6i9gvjlfUZ@)*uGTY3A#r~mZ0PjX!q6}I)N9WKn+O4-`mdT>ACWLCQ
z5ZdEhyoT<-XzUL{cd@Cd+|^7)$g*aqvFBHDQ>|tVaM*UAc+ToMpKATou&m#d&=D5X
zA&Y3cw{>jMuD64ifp-v`dYi+DOae}_p@^n=X_h7^G>sTnNg2@X^DQYhiR+T%s9OA7
zE9Md0pmFOG4bNxj9A*okN_p#Tl^t@?2x)=tS-fmYx8B~_MupBkZw*5ss-m}>lRIE#
zt6^Lt6hHN46BI&jw}xX(*`$$sbW$ZJFpHV&Fg+;Qah#@cZ){JFpP>;nohz$C`%I2i
z7pn%uMr_v0apVnC{vAP=B!wOe(P0W3*!sKCIPt0)3T$@DB|MO-4TwfRV=Gfaw%IA?
z_Nk#OCxB?0x2zpc4ry}UI~c7jhqnnaxvG7<2vB_J8DW~y*x{_$zJ&LQ=Fi2>0A%yi
zG3mlo+8)^;>Rr5PVKf}ZkvVjrEn*4Od{}C%FuP|}3;z?~vgg7B>B7=%sN#UF1U!p4
zt22b91d&q_?Jnm5wZsUb-rP;wJ=@_ec>0q~PUo%r!m%mST6gMtjiO^(YkypQ8vB9E
zTbq^ArDO;;>XI~2V#1x~3(p8^vr{VT2OS*Nh=%iFV?nrw)A09%u5BYFdW=2j`6_I7
z3In=ni{ESk4jb2{Guh;HCWdHP4m7eu7djE#X89x+UW1|7wMn6OjI?ww)iYMMX!q?-
zs6x#|i|;`+q%iS5fv(#L5W~dc5Owid=+s8IXdSm+c8k4+bf193CMEpat{ei~m6Hd-
z@SBEiMm*85YMahyhh1y9ZZPzGt=r0uev6*3r&NRP{N6aXW|u?oBD_F^`i%F&u{-@-
zG|uW9$N6lRiN@9pl!F;yTc11S6^Zp*uG5ra^>ZPeH43JBe>~X-0oLi00dTVN5pDa@
zg<)p2ln*7jglPFZ>3QwRu2n{K06EcAr}s%&0uuEq>c&EomghE@-3Dp+5fP4W>0{yx
zKpYc~6VZs1Xw@AQL}NAI6hLnoA9t8R4QDJ$2#$NN#|WAZUFutOM8{6ygVFFkD^>Kg
z#lF>Q!&Pv^hlNNcf+V8h>(#14yqrc%1(d@T;EGVlCg;CpaJU%;TxgPr#$?j)N^I3>
z@dZpnr5O)!=!}IGbg+Xc>N9$4p3y@v0+gVuTYH~REFOWB1m3sTOt;pS*en2!Fg2js
z=@az~n@~coEg9#CbqQ2Q4Wh-D=#8`l2$U$dP#n{=Ch<p4=Xt+*yn&G@5seLzMs|Yb
z71a9(I`<OGpQj~xoxPgyL2E1my~?{H0FLdirggVNn}kuU37bo!36ihOXhc{w6j+is
zcikUKf<${W*L@2*y?f&?aqB(T$X-L77(zMyGa4JZ4Ixds8nnboWJ5`l(=gF+^@7R;
zP&DFM8)BN+KBBeX8s;YaR5VQYM&JWZ?9dzs8BJjm-L96ySr|I2znmPq_>IVo<`uLs
zAdnQo8%s1+Y>lJ^(ZwU>3KS1O*bT>o&2Cc?QD8f0mC}W1Z?|mSIM#@|WD%6yAznk?
zw+0q9UN(Y>ph~`_G+8SeOj$LP)3I(1L#GF#z1(LY8KTqLZ@si$HJbM=yOjA+G+l@Q
zfmLvWW}us+AJG|KnUuJ4Hrm^zS(z?G@4z^v0%b6qMH@Psh9YXRAuMLy3tyRAu8w3T
zz??JQ7c!eTdx?&F9TYXULAS6Qv46Qaj8KU-<>@z!=1;9-l;rfj8MNDq4V805Ko@is
zo0-mgtG5Bp0^NqnIl|MZp|)AQwG|Q>ba)3RsLdaOga*}{d*u0{tJz?Qdrl3C3WPOx
zUopRu)Bt4@(fNU-o-{aOSR$=KuU<imu~Nab_6)s-)j(#0Zj;mbD$1=^3oOC(xq+_L
zI<$H2IOvc@oM=QqK-aX{>AbQ}Bb>ifFB{H?P2ZteU9+`Y?U_5gcb!lF7G;5NZ>PC_
zt>}sl-eMk-fp4Ggx#YMsr)|c-tZ@UZ0rC15j83B<bTbAxn)}dAo(R#lZv{wkcm*}O
zl7i-ky_;IbE!&tY1#^sE?2R%?i~6D+xNY>BpTU~VY#Vy!joEH&&}+R4jSWiZj1FF6
z`=)VFQp|dvrQK#hVAhM)bsu(XUsAMqm+Ksu;aMdb-xyS?#iGfl{m#i_Gvob?wB7_z
zod$?TFnQ<vw)sez<Kyp~D|a6~;()oEx~*5xFKo4Yr_JqC>v>B)PUiZ#1Iz+@HM#hZ
z*pt%kdJXZ7b}4&UBi!YA$5+I^#T0R$z-_bZ3`4Ug*(Jv2zPGwH^%+gK&aFh+qA>?{
ze35WD5slqz=w3chrU>HhECILoj#T|q>MR1c*<NS#^C<`_4L{NQut;?#A1ae4DTN;B
z?H%b)L`Slra)cB<w3lSIz}T*iD7w6ZbNF{_7>Z&P^#l*lsa-D7*rJ4vlD&oqCh2H)
z+JwWQ_`QYY?Kt+QX*Q{y4f;B!OO(hj6muwWRGXQGe0iN_lJFSyFBG_%*<RzS_;e)`
zwsCTV%0S0Wy4lP$R-7<FM^9eCY`5B0x|v?rg6`6f8eT!1wRT+7yRA*f>h}o-X;cix
z`VdNzM5Fb>S%%qN<Gw7C225}2RX(*I!`G~_vBI;ZlObVFzgBEsL#0$9nxY2MB=~z}
z)9*qbriauW9A@eFi+ai$=<Qi^?prB~&F%EUTDE<51YoaAw|DMC)2+3U$H!=okA`rc
zG`xn`(1DUx#%Z}_ndj{#%=)?7AkT@@bSc}u_7=j0wWI1;vbRh)XG4|2+-SWQI}7pY
ztq?XAbl2oHBsU#9(w@~8sbfS0MOWyX4U}U<um#ko=9^XpSZec<E8X80A5O*y&)QCu
zp!fL^eqtR3!=9YR=g}mO?^n7VnefLxylB5gNqjvV20IID|GGyw(|3Y8M-O@#D!<0U
z!MryushZPt%Mxkh4^=+b-zc~`^$N(Lu##7hW1dnJZVejW0b#sr_zOt$mL+v}(C9^7
zqDt1-S4*I?6x=;%z4ca~KREC10Tn@Lu;07wZm%4e?$7)1QK!cSQ12c)WAt-Z=d<?a
zR*D^`grEBjga=QTwyJ;2Jx>0hqxc}%L+q!c`B}Wddz^|uSMgD@->(kY&zih!Sco0n
z&su}Vqmf&o4u4?Z#q7gVw;A+<4k!XEMnkuyqnC><Z!}&n`4yUvlRZN7pd*U<Lb8)$
zq5C%@XS1M$XF0u<`g(C(38GGpg;Bg^Q@lk95=YT|?{pD`F532|g0-gE03=esm)dRm
z#aaw7sCh*9ZZT1z1+u(zv^(^gnuz~Cww$5FeHx9E+RhSinaPERWeZvfIdob-bZj%-
z(bu~*oYj`M_SFf4?&PCnyPXa#(A1)Av<)IM$w$f7KhHM&9Ypd^*s;TF(0ZF@?(7eC
zjSb{Lv$uD<7usj&4j3~Xian(T4RZrMP<fe%h8cw-X!Sv|MWAv=6tq@<4fVnz8`fa?
zP*yljgPm?(oRv?d$lV2vz9B&-pO;(t-0bPSK2El9mWIyg-7OlM)C=8hfR_8MU?`Bl
zXuNQ(Qv>E(>&nrM;`A9fhc{umb?!QgzpU^gfbQtVp}NyK^qDh$??(h(=W(|D8gldq
z%D#rvap|%6d`gl`U{KC1oR&+^in;_%m$tmUz3cj1t>HlCblm$TEk$?QE*&YgpnlE(
z9m!T2BbJ+8+0K2auQR@J9MH1cxRLw~dhAx8m+I{IVw=r+r}bV;$yg2gPU66q(m&=J
zpY2c@bZ;*=<_8|BHT#vqJ!0TJ>yu@TBbA}7BmTx~xD5~7jcc@bxD7*ByVtr!n-sAG
z;|OZtA>P~!sD0l1>ZAJ?USn1o_{n!Sxm-G2aQPw1h(zMhGoJdSR6~_x5U}7I_~&2V
zZJ}eh#Z_bY9m&s0jLzE%oQL~Ma9!o|stCgDQG<WEgG>!`1$E<KSX7R~aw|YuIB32V
zlod_W0#zTls%Nc$ozJoaowv$%#_piw{+_wiUqV=2(4BJ7TlCCk%`fjc4WX>FgXVh+
zb7r^QYiQgIA_sGHq=+tna#rMY-LeEe2K6CvAQyUja-8ZzXYqw;<lBSh8Vi_O+N}Yv
z1b+#o-m;~K>0CVtGxXf<44QA*8Xt2>x7JphUn*RUfZHeWntcH^ijomJoEMH9b5)J=
zP3R0Oy}A@PlrW-8FFd1n!1x#mG9NxGRnNQwJ<UDhzRG>7J;x5}>*!Gx?CR~EeX|a_
zXTLNDJ(J1umI-;*fqM1ILDz@Q6%Cf+V?qgG%}X!b+ZU%bTo?w+z!>{deEO^ccYL@Z
zPctMYu>+QXOg;|ly{k#8QK#kpe%_{yQ}v-doG~+n?;?~OR!-BcU@MuJQ?Eh#>ONAI
z9eqYARm-IT>GZ>4w4t)o`r5SYXhQ@;tD9XV4(kU}FSSKkpTwXe`tlWp802_pDQLep
zM#yOn8W00NF4k_*vzl=Gz3h9p!#C>=tM~V(bM>h7-s@aGEtFmE&sQ5!@V4Mi!|i|)
znvKvg#-#>PQW`rdjqyn756Y#tEW&2~3Oqpvw4}K%Ete{6+Nc|@1C^@PF721D<%Um*
zv6@xjj>cZ|UaY5GFKw5OJuVGMKM6SQlcA7Tg4Up9^A6hY-I|}2!CS#umbf8MR=<zC
zHM$sW_Ziqvj2dE~Mcpm?cUMAKV+;kX1a+eY%AR-75wz5@qTU$35V@;4Uc`k2HU@N~
zav-x;=cTvoSzD!eJNLc_6^oaXZw1FLxh)=FIYNW&Ho2eo4)Q?sa_NO+Gk2J7wFom`
zdaGUAy-c9@#<ueFiiY~yn4aO(>9{nWHe{>^*&t@UCG({u@C_QUC=af|=wKv-SZ)iH
zh!26z7<(e#a*z>X&0F-sy;I!Q=UTnHqY3nu9ZFj1^j>Ssy)Ugnr;;c-EkRduSCbLm
zqrj0?wBfk7uoc|r^p@S6UI+CjLn<`SOH0t4Zms#UQR<}u@fnPl5Ee8-q46n%%H}{P
z^!_|1HWEbK?j5prP?ios?@e_jxpI2pYW0E0dq=IlJhLN;ed)ZE{gt5iX7PHw^fuO#
z+SfPTWIPM%&w!`eXz>!laFJu-ZSEE@lbzs!Yu12aQ_G`xTsniIclJW*u{-TS+nriN
zO-(@tF&zDXI<j?G0*<tAbKjD-tNR6K*)cKAFsc=uR9=e2dc2u(A7CR``%=lhI<@zu
zNW0vAp>i&D+Ja_uIv-k4x_l`bkwI%vmI=R5eA9yFpa{jkv?Gl7(%JH*0nLB*iy>*|
z(AOpkC+`m<i|=|VI#fYt(2a=GOZUk8i=tvM+I(o1@Q%}d>2xoXKWLHY%R~1Zm$sm+
znA<~Jb}y4Z>0GzLKeX4iZrO6Floq}za*HpW*Q)&qS4#Pz=EiI=*d`+Tfg`Oih~mzt
zhkxM8ZGFQ`{Ha4<#2`?VO8SKo`1sJ%W5+L)kh_OAZT5ZBt^`^SW%nrP3OWcidg*Gn
zxjz(D@0XIJ`4`H5`a}0!Tl|FUzJ-9uu6?@rhZ2qY3mtc?b9-CbbMw>WA1c+EZ(75V
z_7B?9+`rIp<otv7vhJ7ugN6^`9~6i9mrkYHdg$19a{ADvwcSH;I(q01hn*i3G0JqO
z61NJWGnbd<`nlZ?+}BY1z`_atLu)SeEn8=q4{WvUUnmEg5AB=_K6G%d_R`t%OD6|9
zFXcewp^L+Whn^Q%f1!6L`$Nef(RX?29xu%OE?|Gqa(D0kpmXMz)*5HFf6zVW16yAY
z>v@%5=x4S3gW3l6L-)?_wMeHp9=a3~^-#>~hvKdNO(oeq+Dk{;FU^=0FJ0kp@{<-N
z|Nf?x>ENXdr4Mc8>AEO~*7R13zyE&dpx?YynS;OS<bGeeT6^h%PopEk-_tL2^!iQ7
zA<@TtsRY-)=`jUMduU0c|3b%DZw-|=!I!St%0thi&HT`wR{DjGRbR?h#tYZl4?L@x
z{L+<vX$0gA)<btq>21bMpurbKKQxW;(45o$h0a|1Lrcy57rOI$DW@0hp-2SPm(DU@
zicH*F1L<SRL+NArp)>!|40E>fR<LS&pJDB%?-NAR(>GOcRD0p<<1LzNf95ITDo1(g
z`AVJ^SLp83!?x7v-?FXF_6yx}zV~|;5$UBPuUcO^*LW$Jo61Y~E-zim&iqYx-fxrb
z=i?()S9^N7l6|H5mdzu5S8JKY$}e=~J|9|JUN3E_<hKUKtNzel%KU|{nqErs_3=>B
zX0?}&y6Zz%Y2_EXT7PRe#^d))wIo6JJyR*!-ctdBdrym@?Kd8VY|?&ceFi4r7~iwh
zl0;*ED`?8K`Ata(_11fqXz8u@T<qI4P$tJUexYmaw++bU_R^hpe{0w{v3}8_gd*jo
zJGGaheE!yOOpE73DgSyWKXQEPsP3W9d_>~VUfvt%&jbOS<=aRk!D+4O7rJYG>Fo9C
z3kXZs{R^G#X>o+GTYt;$-PU(mQcCpiA(QO1Qks0|?DMVREW*F7G3Cr1^{wG7Ild<;
zvh(J#-x`!%{Tuc=+kWX|7?4@Gzr7s6{9hVj&h>4IDFyP{ztFwQc<Ao?X>yi{`x!Bj
z6iwfD0U4lMYA-D*&0pxq&xnbnXnvX;NzwehL?Q!r%ijxFCD1PAt)Y+7-x@}FCJr=3
z>-P$l1mLZ{&p_PQ3rF2=tw?^pwQi|Z{YgnJ(Vj6=`CYyju+Y(;Ns8POeZKXcVes29
zAR&Ak^`$M1Z^J;D=f{3&*WT<y*L<3s1QYXn1-nQvvGOl;toGKh%6cDdt+n?<ciYbk
zW~KOyLM#;bw`IQ&6;1gUx>I{=KpXO<72SrH&iu?%sEX$C*3fo-+8kBUQr;RyF8wWw
z&PIDEIWEgfcWrMCtJlXIBh5rhZw-4s6_97*X))4FJmVl)Ci-KGk!7N{U+8?hIaww~
zdh0z?PknFUOfuuG0SSZt(mBgp13q%U&^_1N3bU{L(v{aUYLaJS&6n=F(^~^+C%#_~
zE5)auvryd6Jf%FAFD&gDpwSdf&pd^wXihIB3U9x3<)_I}6@69=3t7=p-e<rI|E&Qn
ziT8fbYj`Fq#a_3!-qW;?HHKP4dqzzOaeOae*LHc!Z)({}zG+FXL8hFy{edBf@h4p4
z88elk_q|!R4BiS(*MDq-?MHwQUGtY)Aa(6akI9nXXINvkmm=yhUb^;olW$gv`+Lyb
zN(=pM8$ie2Zw;lU@|G<-eGi%&Y0>mOQEj9}bAD@RGrt$>HP5$#X9IXAMGdTn&Ro7t
zd+2<oznv7_|JJbf)62KC!@s}ajehy_wL@q79yB-kCZ0|P-EF@$l-|or*?}rAoqfHu
zDZ=)ld$gC53;C^Kd~ZX;=F`o$Qayaz_Ru}wqU2bAn{nsW-iw{d;AwIeiaY&6_x?62
zx6+f0mzKQ0hj{2p<)sMS^oO>TzUL|E$WMnxUNnE3W6+i7TLWPM`JqIgy)}@O`lYKp
z?+1C&Uw*&aGc1sL;`_h}NC?40+02|TUG3W#+oYao`>nU8UfvoiIRPKqO8+*-pljs!
zA$l9_r6Y~`)<9_8OWBgmFJ1F{n%X3vn9txu_K8(~q0j#OCi}#ChD>G6dtq+R@XSJS
zkNwiIzo)5<#AtfPOf*K*w=uR68BOc0p`W4reFpNIzh&{Be((3J_|sbh(oglFE0^)s
zun<FkD0|~?lbvzl88gX0QP)c|`p++&?fZCYlYruRU?n+y`Z@V0o{5t56aAT|NI%in
zTLJs~<E4B2atou6`O=wZdTX#WWZ!hv`MvK4-E+Tm_w~Jo-Q=HGPn(l}Vm*T<YsLL-
zk3sitZw=)+JJ^VgzG$WT*4f(A1PG0$?*%M$<owc6a{Gm@{LE4)jh6J@>zQJnK^LXb
zGTySgl>Oc>@&DhH>=X6rVq~AF&zMQ}iI!hlN^5ToX8}B4`kWSRl27#KMUZ@=zc+g<
zn`dYv_r!RAKiYUNHPd(u3sO(a^wK4N!-uXp-uv2ZKHZDl6YFVka!;)O(p;VcedyZx
zEnC|DUcPSDiF^J+XZl{gZj?q-{e|Lh|Ipr_#zAQG8SNAo|31RRz4)cOyzqSa^}Q6C
zBi^>%_*H&UbRFw6N1?4*AF}|}()zSG@~@B63RF@!-xN*Sw}!fVdTHyA)5=CdyN_S!
z?BA=`y%ld)Ub-Kzm5m(zc$<MNlli63f#xQM!F(HRf6Y%NksD%t8I4d>R^PH`tXnTV
zuRQ-k_x_ropd0{?ho0f`%MKG=-##nQ-qLvKNZ+0Vq9cDbICPe8pB2%S*GpF``K@86
z(te?%jPI62aI}6|^`hSg%fS4-a0HfA*IVl`jPgrs`F2?mU2VR!wVvifPsjh|9{T9x
zrK9wx&B;G8%1dYaK0Jl4^lcK0j_*@D=q}%Tq@wHlw5TLfP<-#rL38?E35k9?6aY(k
z+MG4w+s|1eK8?T<@q76ziUiO@$!YW60n(pu+0oOU4<!-7TLV53@13oh*DrLJ{L<Y^
z`+jNYkCyohoi){m=3477bk|WHT6*0NtY`V0U;6FX0c`c@`=p`xp8KG^|8}jjnjF8-
z+4o!TF`dhN=^p*1$5>b|-FfEm()m6bnJDJ0FWu`IHOWM=>Pt&r;}<%=eO92g?(z%W
zyT7!Z#>o#QJ;*O~eeW+oM|*}LL`c)#FFnKgOG!%oR(n>}>!oL1k-x8o5NT<@&{3YL
z3MJBSrxnpv>q|?kPotBLqV8Yl_;y->w$l1rZ(H9^E70D0ekroP`K6=Jm(DWMFLaHk
z)yYXQ`b!zK)=S4TMw60aKHn=jDdwCH&26ReP!giQWy@IOEqjd0XP6$!{MH`Y+U_qs
z<H-7j&h*SxLt)In(4F4fojs?waS&L}FMUT0v`8OA04dV)P^3spd+C`x9`gVx(wbgc
zYOTLerE>qKwSRlAKv9YQh0br!6=;|8`$Kon?V+Rhr^(4l(dRF8j`Y%1NBxEF@!r@b
z-|817E5$6o&^h}9bA6heq!jDvAS9(&&s;@PiZ$LEPObZ;t?u&H+xC8%oTcJkFGa8W
z7rIh=A8oeum<P+#jy`ncr^%&YK7OGqKXVnDq~-a7(IhQBzf@7b-!xRm|Ag-xN@9v<
zxFIh^eVUxS6s^4Ue183{p|7^yvgqHwWoPPbyma)ZkCB*Sq_>6@dgy=9jhgtMG?SU)
z+jB*9%<|H#sMm+C`8`u*a#O7QQfW^-@R&~dq2F#nVr&0Gd)xh`xWulP&Xiw@e(Zed
zPV=R!e><)~H}Zflo%wqmo6#oC`xm;u9ao?kVYG+h*jryZ>(k^UsHp81y6b#t>n*+Y
z_S}Af^ShJgETNtk?w(&ddw-u6*W~%qtNihk?mFsQcJFV$mD5&)e|}F>nG6;4X>n3i
z%>L5yBhdZ_-Rpar%A~1SPlJ)BV$EOZ+~1}_X1TcQFLdvbABr60TkkBtR6AFCpMiu3
z`=R?W?t$ZZTI5Od)8gop=Ka!R7knEA8G+JoHxAJ!&IDS^cf_6Os_nfqk~+LK^z`kv
zB8p4;Lq}=fi&*Gt`=#^oBZjWjzh^1v=<B7seEVmKK8w0c!iw>|jLl6+kJd}aEHA}d
zf4+3h_0rnbw`q_`S+UwLbgcQ-(D5aC>Du*uHu7t{bf&kPH%ZgKl$e9^(w&}0N31kI
zJsq{u+<u`mKZ7P}rTKdu3*Du>wDhO{BUW0TnF_JenqPW)X?tsE&rUG3_wVyC=y-P^
zp|d?RRhL|Qd+F}wrK`UqI(aMlGgFbbVmvbyc`HWyh3@fK2PCeT>7^w<LoAsq=6I{E
zb$)OELHBx#lCNbwwCA^@vsT>ct(Fj$@zVXQJ2H#K{a6R6mZtJvZLQSbvb()j7P6)J
zdld^k-Jw4;|I(1?mcEk;x~1R7_5c0Xe@lP;kH6Ai|Nh_pufP0%|DXPu4@`gMzy6IF
zOaJG8|Bs*lMd%0r&%gf5e;$+RujK##`RD)sA5rLk^ZVa#=6_S|uQ&RiJn_HVWB)g#
zHIA+DzyA6Y|7ZLUC4V$9{}+GkG}QU$zn_Z#o!b8gNyIs@IR*=t{f{4iw<vPO_LIaD
z4IM8MfBjJcX~@mm+Kb%3|0qMSfOZDF%>Db1GMN=)MHe=1=YpT5ZtCPHb-~Y4A<7$t
z;_pAp+&Hc%bit2OpDjHv^#MOik-Z-IlqeK`|5*k#bPR}F=z^c62&g4=JqpF&f8Ke#
zCxm`F*}{IlJT&39>6pda`dRL>PqF*PewO>dPr>`zeir<|Pr=Wt`{Ncr^heQ8F6^I0
zKlDe@A5zPok{A9c_%lTBQ}RPUMgQ2cy+wb}A4TJ({!#M69|iA={iEcE{wNx+_m7en
zehNNg-T&b?9;3>sRz0ZMH}OC1zhvo!ovHW7f4<DiD9Lo+KZ^wN5+s|*oSy$Ea+#O;
z`Fv64G8%#d|HUCwbbS1~3cH+szn>C6@A^~h=Y4<5{<!;3(I5B!Df#m$K1F{%$EWNc
zZ|j!*`8?12`tc(EXpi6PeOlyyerh;>{C{<U|NLX3&;YRi`Zt77{txDtAM=Rc=a2in
zQ<hg6b^h#nQAPvG{D-gWmHypzZGifRuj`fm-F0oe`-iXVmHypzHF}nR`~`che|u#O
z#{cvyj#~fr${O7JKYeAd^>44N$>N`W%~9*$T^X0wfA*@>__tRqQeOY3uj;k_?Ugkr
z{^QT<wf^muHOJO}`pRDG-(A_RnEt72ero%#FQ%z+{i9d>w@;n_^>uBY{b#THsq??S
zu1(ti>~%kN{@2&F>HD9(?x)Vjj(9AD-wlaxbK@8f|Bn{R|M|K6wf2A5OX&EYAD!OZ
zbzBu|g_cP&Q3<Iw;ak|m2cQ!8>{&1o@vpc5RD1!tDR=_+s4^M|*{BZX5>~rX1xA&V
zT&2~#jC-(&S|c#6`{PAk4&}8kD+;cS%4}64iyfEVoikAqwRg{mhID`C;GewNq1_Ce
zocoi{e0gGTg!`-3#>$Z%NX9oA*7x24@njPm$x|Sq+n-shllDAS)4v5Oh`*Bi(Gw4#
z+uu0vuSgW%o9qdhii%Q_XG;eTPWvlf)UJ(>Xn#5Ls+w}QTN2S~y@=(K!YKCR^H<sM
ziMv_E=G7&w$;<G+tElAOC)vcxRpgc0iu>D(1OAHEvL|dtrN42$&n3<iWrsPDgR3Id
z<*6xvgCr7FA<-pZDdc9Ucm$Yc9&F*lQc>c~Ca;V>f8$2#R7xLbE%E0{3~nD6Dpa4#
z^kGWS=<};@%HE9rNK6%n`jvQz8J8K2BmIg>pr;c>hhI_pz_x2V_*ssgw^^%_YR9cF
zDVkQ|ggl)DaFK40Z!YBDQRe5rQ)P(`Ks43))G(O#2PsipMW!*C3OAbkg(J>Nf`aZ(
zLnIX^`4v|+&kTwpze4jXe;6jwiS~Zled^kt_yaaL#jj1-?&CV7>(j4W8!4`Xw7Eu*
zp;X8QdnYer#jPp%Ab@uqC-@8(W86EkJZhC)W>2At82@Ply&lp@#cf(rUEW#V7p`84
zT79*0?{jC4HM#leg5JQ15I7~XCul*4zmf^1>c9H8h*w<IuTN|G>%k|EJSg-HQi3@|
zq>B-8dA1VJ<YbvUPdP54P4%iI$EMN_BcC<hH<264VN(!3Ul}SY4o{cHUS7(Xz5*)W
zEJK4b1{)k_<Tcoaf22f9aavy`ZoT4Z&W#Lx;nl^_(cUNHDz{1A=qQM9Y*WEVh>rGt
z<qVVYVY^H=P@(r#!3HMci}XGzCMkKkhWA#Yq|ldOgjmINx`9(&SmE=<S*hstGd9iU
z-Hi~bZI_8GYiRD1lTHO!ZBdZOc1<PKrR`zM#?EsslaWSmp9tKl)WdPxgCtzmy4&=8
zfuyg}m1`aW&(~cAH$H9wl{$&}AvB<o+Hd3xspN%i_X*MOZ<NF(Fs{Byc*EROv*&$9
zXP;a#nn~s!7cPz~EBRwR!5=F7%2L8DEzi@1oIWW&^yI7ih46E0)*PE;N)Bt`sIp2h
zU`Cx)i5aSrC)N0QW1iE!yO=Uk`b<=mTfN0yc9irfxPSb3a45^;C^MTg6x@MbDGm)z
zf}XjGlzxmYl*2VAp9k0h6<5#fo#Oq_ml2QPI@uF=08xClS=oj8QPpppO>%LG6&lPa
z4eLU^nw~EZjb6RU6_YygPnaf^W%Y+GjwR`qo{A1d{WD`!>w=!a81Z}s)7LH0gZOnr
z$W!mJx{5b~`jr#ON+zw{qt3q7>Qjz<@pBT`*-T$@JkCU~P@i~hD=t3uEikOEbxSU?
zFI4rL71v%;m&2aeE-#Oee&-;P;k$Yo5p?u3YQv1Go)8Q+Ifkr=4twrsWc2x#D0prm
zMahf^l_sw3sl!mw@3Qw$7v4VD;=;yZyUk|nLTLt7@~L>DRYdfW+tJk1dbw~T5WzrS
zJS(6KW{w}Lj-_W@eJF6gO3UX1(G}77-)dAzUTR|{`N*n#@~ZKSI{tbHa~ifInbwSE
z+sG$qhFbl$#zuNTaK=PTh!eUHRatK=RTyw{=2#VY>$VdCPAdzhy?ja`zKOGSCj8}D
z!GMG(>%k)*7P|RWJEz1%7yqCMZbA{FN_*LdQDU30GHTHtcHAe)L1ZWua?S=^>ccqv
z_ZDyv0Lag0Z7J@PV29IkGE#Hg(p1<PFxMJz5wZm=*@d@(1Uvmzk@M^KmUiQqk)M0o
z8l>@cL#UdUAVj9{Q#olFQ%;jg+OKVCY+wXKP;M?yF^5jR;t05z?%7We#K!|)?L$8D
zfQpQ-S*dvfHrbCAZzJ!kK*hyZvS7^?B|Cn`YP6M8+&XK8CabX&>jpsbeCkCkx5|;o
z=l*mdCcJ=nRI23s{Cwe!s1LfwWwyfM>6g9sxS>5r-oOowWL7!q_zH$sjcyYG)!C3*
z^Hdo~<0}$g#<J%DL<2v$)s{S46pWakN)U&qLPixI=Mw#RmU`jx?z{6Xh$@V}GU>S)
z>$m~vzH`28Y_p$_T{+^m$qv<RIWXf6$4+KGpXqqSv&@b$OgEdJum>spO!d6PUd$E)
z9)rL;dGB#?vfrp7Qyy{SXVsaibwHosrfOp4r<8mb63nBLX9nLFPBJCcEVte5@HAB^
z)&+(oZvgt|eMVBVX;8uM)hz$*Ik(Wj_it`b|72n00;%NNj_dX_^y;T)HM-)beUpBq
zVu^Y?7A2jvWk~j{x+vdgk~37YoOzmAprYLQmLgdgy7v_=he~#Xz2mme+Pw)TmdFmz
zL`COInpwpNNy5FLk~||`A_Ejz8TdSNF`EF*`-;)4(IWf36OW{>o0R~$IXs^ija%df
z5f>tON^h2As8RCR5w_Ax#nZi^cVD+Y8+^iRF!XikYQ<A&O|9>2^_Tka@dDp6wa>VJ
zUkEpF+3YDF(YS9q=0Sj^?zrqyl6kd`5BWePP-aN;>^|t*7t)N-%SImY>%DY>BM(j=
zlJF<fW;A<g@g3q$Mh*t9qhyi}2b&sv413r3dC|jin`xP)30!O%u;%UAOcbQ;8_Pwm
z-aHW#BK8Tb2JSt1LR=uZPSr7Te4qG{5yZ}DuD;}Ok$Zuwud$<~_bR^axXeu5=7+u0
zj=P*1JGl1h_b!*uEz*T>w1%~BFXKKUX{j_Sb#|r%IiMaka<(()UC7#7jX?`L%u0Y`
zOL%{gtjDJr9^6E8cs|x`7D_2mWmVjIRa$HB`-B)RJ6$DFXZ)HiG|d>RtPfnIF_$e<
zeGX))g(`z=Ku@5EUVS3b0{y(T8Mk?!C&`W?HhMWEK2USPSqYaYIoT&-e5j>HL77i5
zb_bkoSq+6*yv0rKX(Ps2o7#>5()ATN1{%Fazrst5^a{)cyF|f!MqUk9Bl&6t+4@{;
znm*%2i(ez{C3rrK5+!S`Hrf%^&;lw*UDIMR10Bu!jtqm$tvmzvj0!`wgrD*A^UXCM
z0zbDzBGcymGJbu2lWZbt<^Gx^W*SGb<Sz`atPaeM*}9K@)=#_GD4*|jwnEc2Uly}T
zf!M6o5;59!h#?wkU{X*GXIYjS25Edd#1Bn%$}{}g_8<!Nb=o$JrFMesX*0=rzbh_r
zED|%C$F3uO+cKI9l-!=;3^>WQiXq-EI}=D&FYipM0r$3D9F`Kr;poUMup}zJ?}{#c
zA$>yCxaBQAd9Z40jkb2&=4xaJ!90<fZE&Y#A7k0;14s6K#9iE8TKGnHZSXeVJb*Dm
z(mrfNIm?PbXHhbJ{~Y1@MA8Jlc`auh5|YX$R~sX)ats7p_L0RVE>o93jaIJ48vA9o
zoT?>YPoztzT|w+|ne$C&jV);=x0-SAQ|IkUvOWWz!PoAn_ZDzAQYC0F&J+G2H?Oes
zDxc6>=Njy|RXY6VmbQY<IiTpb+4dq!0;~<cIFUiaR()ql;gb=8L1Sqp$#As;j%mxj
z8e6r<&+8Rv73`@Xfh;Pm;Ed~NwLntP>EAptK{lkeW&6t@_VnlhB=B=~8o&%8KIo{f
zvryM~;hS6t&NtNf8Dx8y@GP&knas81ap6p6GN0v&e-Z32B{vCoiz<(4J@h%vz2ee@
zO4}#4Jb5ru^NKCk+xDjcPAmbMG4By2bHya52YLEZ*!IqA>SzY>B+{=*%a9#bCOq9x
z^Tz=iIxi>%YSsEKg5gnN+tqJzq_{E>-uk=}Z`GkJIo@eZ^03*ganIa5qQ<k1o!j>0
z{lo;geP(ha8pnVx%UUR|e7dmBB4>F&qe5~y${IMfy>$Qt<*SZrB<*e;9JrCqsnSpP
z`ytdvi>4xXsqFAfD7o8dLh<L8iDwHEQXo02N>t8N**)1b@C;BgWpRBEc&1WY?KO>f
z1~Jj9S@CY#1GQa&o$LNvD9KlwSdP83;&H~_Zs-l$E<i<ILE|z7>Wm*RjC2{yex3+h
z$yR%(3y*l3-S~602n?}8BAaa{RQ0pWCBwZeTbudzA?opk6bBft_PK%<r-L<plF6bR
z-*i5}apJP=2Q=c7T<J)Z;KnFW`5LP<Q%1n1FmWbx=|?+U)s}m`C4Y;B^ff~lDlG*&
zM#+{bd%lzvY&2W!9h=F&210q7napd(l_w7AysYLH5OVRYo4tG*zQ+@8vT#<_QKBiN
zUqCO-W-cHUUpPn6V%D=a0u>X^>R~nSxtM^QK-H`W_u;T8BAh{PZJVipgG_tO(Y63Y
zAzqm*F(P_fQDa@xcC~W5zdc{8pF_n7HA3)-m&Dh;eQ5<8ybo>hdD3#!;S~rj!(_y-
z&#e{PAWXh%zH~a+HT#rh=SXP4Cv{x1!#10yfHT=K314oM+)89^$0YT_GwAA2gG{x<
zy>prUDK|a0CX(=Y<1_CuEe0d@n?sH2b_H_qHQOP{+I5H1Hb$<F(bMiLeqGg(*ksN#
z{plC119#8w3Y^ss-1+z%9*eZ<_yaAP3pe=+fP2NmjLuTDN}9=Mi^iEr(N6lj!yM5$
zc=d8e(iAdynspf@A9uhlhla(B7`MI^?af_AvQ#AC&zEhd;W<Wf%?UbZb~_vO_d2TH
z*LhUGaYAy2o$~pP4a)D`jMdS-KN5{I%|K%<8ue8Iyy?r$Cr!4~=kzs<ufl<wyaWu_
zB!G9=JIR!pEXX9_@GK&2ggdXL#T_2|!c67HPJzQ&NlL&-8qO8KjWYvdbIRr^fY^Iy
zz+<={wMNOaFtQm0p`;mbv81t`9NUdW>RnOYx{x0Oq3ND-^7HZHgZp4V!Oz_H=mCpt
z6iO~njtkW0wgR8L`wjdeiW|1g4)0KN*S4KxMOqX*X1rNH-IEBT?_R1c^y(S3TR<&g
zvCyjTccbp!sx4#Mi0NyPW5y<tGlaBw#yRBDaHd8-4MEvmaok!~`QaIGH8b)0auXI^
z_tx!*gmk-hn?&1mJ9FQ@WepqMY@JxNpz2<RjdX-z_ke5I0m<-la3tLoy-kJ57h7nz
zCwVk9rQeFn)WH`h8?D^?X~jD5j=bEkHEoLC;c9@Gds}wie$31r*BNQq=Hjtiy<1Om
z%8YGuv}aF5$z3C<yd-?@1gNlW?M-^_RCL_W*t=|lv5lw`n@y{U`r|_T48A(-C1;)8
z=OUX-v*Y;~&KV73B<S%?owZdbpGH-@|A>+$s{?QL5d7oi5L-SCOZ(2tVXZhS=?J#f
z_&MS^IwRcu1J5ko?O1m<e{GCs@@YU+d|swFZGPQ)4LBG}8L*T&&KZiwFlD+mc3LZG
zGC4+&e~)JsmEEU-4!Q$QPS>i&%pQ-!x#F_LQR(q>ss8Xi_Ar<@zU;Q>=_w{C{6@Pk
zj#BbbXA`!PG=ZP1_(EZmIHnFeQrkUBmOXqy;yXd8wRchq_$I!yGb-2tNuKT)1Dw;$
zpyXj@D7DvV$4{qmTNK3b)MjK)zMYLuyJE7jtV;J6!9l<6qp}wnk2n>OjirZ=hQn1E
z5V0PU9}57{n660{-*mQRM{$_#`*g^&7j<7&I?YBB;9{t$azLvAb0dcU$6=$}H;xK>
zB87kkr(~uFaLB8nDC=vKtct|e5nHh)+VYg!T58ldW=5A?ao9lRrZU(QJwodbUtt4B
zu|>)1-l@@LFV#*-*7Rd3pgcR6e73&}RM}g#Qxft?RELYyX4?ENd0(giY~C98;c&Yt
zz6b^nw#f8dwhsQRAO|kpjPBt}Z*<p*jilYG=TC>Z^%@RF2(e2-L-zhQixzlBV7-ah
z`9_6ZTP`|H={bJg6ml&eu5qt?CY)j>cJKE(UD@Te%V)C@jI$>#h}??>0d@sjYmwv|
z0Xw^Qj&^R+y`edl?Z4&aQ}foLY0@Nf+Phr1wymhyZjEQ$4L=q6DK;<3B8#ILEX&~@
zO28=>L~@X9n>|oouWr?nvSkHSjI~DP)}mz5!lX+wqB^TN+l*b(iXT@k81QWKy?qqd
z<uE~O9DA<>G+##~v}=i9qo=Q*uboQV`A}%J41;{HGiJ35g>xuc>*fMRW-8mRfbHs5
zU!C_d%e;}Woc#C%rS;tWyuFVU*WPlBkvTQI&Pu))J8z%vL<l?Y*<2FNT3|WM=+|i`
zjFVAR)-!f0TbyHg6LDxed~Gt*t+Q5h&l_o7q(EvdIhL8|=&N_lFlPw&5ra*sX_HA;
zbg{#U_pzKt<2_VTyewJNQyYs8_c9kbLe&FK&S=!FO1~wSBB<p(R$E+X*>M4x9ovix
zZN*Q^eXNXVp>12&OO9<;l9iVBBIE{Z_7tM(Y^B<T-0K*)v^K_1ZJKq7rn+tGG1Cq2
zMB0YRn#0d%WgCh>+h>w|@Q6B5p0oF}d_xMjbSg|*pf%z`o%Ohugzl6>v>N@EA9wex
zg|U{^qNV!!ZEN~8s~WGn8qinAXWm_LlS}U5S{<hn1vhrSZ|n36N3*k%vfMcY4fQkp
z+RlB7FSnD*W!v8F@SQyr)G>bS8{fA3gjRY&5kTI$JzaQ4U^)Ai@Et-wy`U_DQPbQD
z5y+HaE~e#{Y>cfkA>3;onRLb{^Nt<y<1-r9VqP^p#jPXumu-rw_cCV?wNu=ceLae9
zI{Opso9tbKtz<Ld85qqph4E|K`KT@K@ff?w_k5GTGge&Nrc5POfU)4(wg#*zUJnZ{
zLQCaG<wxhhnQRSsj9y!TB9>m0xs0GJQvz<jc+JfvMakN;otLCB{gWOX^_nLD>4;ju
znQaYNiuX$#-itYL;mQzf7h3^h8D)pZY8k~hb4%RNM9)ztmu<#_&slAW!cJaX2Cp-l
zC(v6CxU0Ev)#C2iNThc;<#5#Mehu4Z9L;;hPikOi2;MfA`f&;J=oJZn?Vi@l&Jb*K
z`|wpI+UQcW(2-8cQDx28;#vxgTaxT3X6+{Tvqu{J48i6%J^Z)?c;2|L3;T|%_m0oK
z0y0{Ok|WP*9@aUmq)B!b8`sY(x8)VW=x0>20cPeECFfqnLlg47PgbbDjpdt(sG{d(
zb9Rd<k|h_K{c_1K-q~4l*-GfH<zc6@7IRo!K47xs+S?p^Q;fHZ?p(I5;_|JsXvWjE
zvwx<Gj4VrVE<b-n^*m+U@`{~I*pPn%XKG~x<P3J$wBWu7z=f+64!_Iyk`O{~#df?-
zXSVHVB+nCG>g;$$crkD&`gBXHGwM9;YqQFE6Em*PMzUbbzWzl8y;Yl4&M@ljvFW>6
z6A1e%Bj9Mwf@`YJAB3^DnQeb744X}?i!B4&^z$tA&zqSHP1!Mx-7#Vmb8&^8G1#`;
z-A4fBbN_TdmpF5ntpJY+?4#x8+|iEv(s1W9)XmfM@3BpPi|pD>TLGR`$%<z=cCoDh
z&q6Rool_g_OkFEF+gV+m5!kof_PzQU+swNmCqFp@>XVh>=SRF5$@8+aEk{cIz=fZM
zth{1F9yB-iZ%Eb8z@a6x1y76QWGhC|$;rD$(q)zvlx38js<|7y;7)crwfeMq2`T~C
zs$3cMLK~iOhEDVT{;X5yvX6Dw8FY{Wg+tVC^PkveJ+0uPQtrc^Pj|>g826+5`<08u
z^z817MA0c<9R}gPZoOLvEjqFFQ!de6Z3b_MaTz@;uueFbY!crMf@W>%YhY~gWH`Mw
zJCrYTu3NU$ZY!7b?|jRa-EHa8d+Rf3?cKU-pLLa>Eq8NI?BP`>dcNLl`(drztMzK<
z24`qaw`j-5vnr*`4t7?QY1_+n?;@y8lXwu+`hj!W?@yQSJ?J{avu&RB{ki_y%YEt6
zrxCYy5N(?8UX()_KSMqXt+X$l@_~>SX<0$fs;YJKQ_DSHdPeE)2V&#aa2CEv%j=PI
zzw}J@)%@&oTesf2Fh&a2a&7@nZ{LE){Ogx?o!XWqQ>0va)^r8h-MJQcUd$O93eMI(
zQHbKlLanK!`+FrrE@uiGy9ZlXkOkkA`nk|j$X3z*@25_gD}~K;U-=8t+y5#$ekfim
zZI44lVdd5H2zpH5V*?bdT90569r!#JOy!bi&>~x<5KPzhh1xPMQl5(URmrq+hJn&`
zewBl={og2iOs{dvqQ)$m!*YRQb9GvQ4eNWmeT_306l{VYw|lzqX{aLcsu=&|DSO7G
zLJr!LZ@qJ%#m}9kT>_pyI|<U}SigSg($mh}`hDEpNI5~tT;gkPS}f^aSBn8fL+Mu6
z+BwD^)@)ndWoLJC5M1P&plf7zH;0zS_*P6c_|lr$o8ktLZ@rWx^XfBWq<dRU+hEX@
z;*~RA;OA2~MB4Gj7>C$@FVbHgIp|P47>!VJ{ZM%@7j4v!Tkj;9v%Rs|ZYxZp5>>;f
zpQl&=T|lD0<Pg@gh2;S~Q(H3nQE9@Ytmo22fi}^)UGB1K$@7Jk4`qbo#&SGEXUFrH
z2)yyJVjAg$&~?6%q9uI5sUEH)zId?yFB$NqNsjR{(kcSayNd}A3x!?o!ixDvt(|Os
z&}09l4<!%YUqvgLd9O9JNm=l5=(pN4K-1p%QsjQ0fEZ{|T*Gg*i>zONW%9VxeXB*9
zyhB&6^VYk`H}zN1rf|PyH>sL*wb5drB7S<ECe_vAB@htFnd*UT?$t|`U#GC@DADa$
z)}0Yfu*JI>(i_^M#l0k;m6RwnO)4MQnAc1BSGY8bk%a?jX=<IajIv5&fesxce}Iw;
z<K6&!a*HN7b8Fr~Cpib8%MYgoQR~-T`DrO|H5|@qmOR_NSM~txh3)jIb)&6Ltu2R<
zGjUI=YQpn<rF<!h-V}{MCScacU$R0KYtVv@wC<qQd@Cr3FzdPcwVjB%WVgFWaX4B=
zaf%F-*n2tOt2GX}i#F|JYMO-EoryLpM^DV6ti@RLWzN`py=4cBmrUxQ9W65nwd*Qc
zk{|v+&)0T_$h;JVf1m7pET{WqmLzH1i!C%Ap<T+tPB(Ih+%5;9_sKR&g3#EWLw_bU
z&aHXeJ=WC^JO3Z{9?FdVxd^$lUMtmQho#W``W$P$aBXWZR7G1**5rko9cg|lASSRP
zlR>J(51mBFWZUEW(ltZUu<ZCod3KfT5hwv0_!?xa77x?`&TLIc^6y&`S%amb`&t+a
zN9Hq4W=SAK!!Ba(&fAmxV!M|=cE#zzXg%ipF>7g=OwfxoqnA!hNiH_-x3CGoJBBK6
zapV|PwG+7|xHWT~w_YMOhq@OBbwO)pS;&x<YU-Cx0u1S(`wo?Cx_aB3ue9H*kt?C;
zHKdv@?VaOn8h6YX(@rPm?(Hd7Si6T(BFib}^T20)cwR~-57w?9ink#zH}8Dm>h^BQ
z0C1azfM>mjv$Xg$!m$NTSK&P@pYG5zSM;LPHQfzUbfYnmIRT0989nok?0&X=@*ecO
z@-s^y?3z4|OmuP}BDvS);{<}{nv;-w&AwqQx=Ahp^^G}T8@R0@s@k@XpeULbv?j8q
zF1J#ki`5}2rNe#Lpra94hQ4j<J`15;p3s{>K6uA)3mwcp()kU08%6Uopl4_*%R6b7
z9+Ik^W3O9u)8Q$ZNBlnjOJ~87^L6tLhMGvWNc|iz7i@pDozM4piyl+-OkQY4=lrSQ
z7(-`nf)4k;fZ}nrkW@QEynp)8fo)`><onVlhjCg6sf0@x3KaaI{<yO<tut0~{uS#R
zl*<#|XWR@PF_&uVf<H{(4~5Qx$5FNf%=c=9pETXW@B?$`YPfBv<<h5yozUFZTk_yK
z(SZ~N^=bkZWxnUggGVp%=|I`YLzdi+xU(8+=@HnNx{#PwJ?(I}5y62RMOI9b5BL0~
zY-@2$ja#-ZqFRv@pCx(+>%ndH&XuCuo|Rw5rj{o8{M^Fn*iC(5?i5d3h~As(V5oH8
zvSVk)OBU<(^JJ&8s!u1Ad_3sNggWt+ja}}Y5!kDG!>0>~$GN&V`vUvYUAlcw)ulsn
z*4FpZm$PeKmy*H7zNggohfZQU5Eb9mH$CX{bXerAd-wSPxYf%E5OCKPXzk6uo{0-j
z(8gvbjW9>QMVZ2uDu@GD=rJy`=AyNI*IaaWo1u=;Uv62FF{_4J`@Pv?PuL7~tgC$Q
z_E-)!UUv8C1fH*V_*CR>1EQPF5~y+kJ3XD|yb}x++>DJfm~ak-QPbTINfgI8O<ODh
zgK}_hb4D1Pz>RYU=6Q=xTI-e5SMrv}e+ybO?j2bDpAn2zA>jj<doGFqY%B|8r;L~!
zSQDDXOo+)0*axdI4pZy6*B?AL>z*C9IdCDi&k85+&+lhLxPr1dxmig6OX#i?zL0jW
ze+KBG&52_?x%oz?EU4`lI$18aS<VDlM32E7ovV955qFz&43^Ale8+`OQa*50Gvr(D
zZj@>IVn?Ci&ba2Sw#d!9@LBHon9!%<Gv*muZkM8@RQ0_5j&vN8U?Q=XQuDs`lFq1T
zQJH(bHRv3JMd@c~XZxbt`tw{fCSyWFY>#IsCOIG{@@nuB127>%qiX$o{+({or3gYc
ziE^y$Q7~^@SnYL92P9<>f+%)4lrJ4f?Q>tP#@iW~&r3T$dHErhC?WAGJ9{+3AoS@N
zU^GJV;|nPMpk41bWnUP}IKTmfdl;kMdK><ZYO&@{528fhLidQB2sF$ax)$`Yo<8-W
zYGtn{xetjbL6v%nJLe!5#l|0RJ^M+XD62`edAkp`m~Cm>Sz|oloLqgJZPZwzK-mXT
z5{&4IQzygv9Bt+astByYzGnTDJytGDRh({0?m^kD(Xteqdt6Fvh-#QAMwCfF;8YE#
zBeYw#605+7()z5oUIr5FBa{`Jh>V-K&)Yu2H664#*&69~PPc+liHu}nTUNdgF*aTy
zMXpo#_fuX*YK3bQVLuOBtajx956bzW2Gu%l1xoG3NGIDpAa0ks0VN_3Mf_)KMYnUL
zoW<17NCXsS)Y!R)PQpj%1=E#@paq9CUT#nB^CY_QxGV}=jmb?4LeoB{HTS)RmNCGK
zq6DIMu_$DFNmpB8Ub85~fsjtSVuxA_Vm!Ud(E9S0#YEGnw0lnX&St?)gio*BlIIEx
zmdeI3FI^3nex5<0_)d4i!BoSo`<RQ4tdV(rCWp-)a?-a{s_Vgta*oTkV4X3fY`ix|
zwr$K>FVzWa;Xt^A`L+KgyOiy}x#APwcpx|Jx#pGc^@-Hc_{U=P7Htj1n}XYX``o~>
z&#D^sOq{-=gAN=<v<q6xt#=eeFp9Ff$M;O^0l6@=b?)IM(5aII>fL?bd*GB<FYGf1
zY<4-bP9IjAF{Evhs4!G=FJs+mQCcz)h7|`ZY=P*wl4Fq_{g!W&g}AwedX$Q&D-x9H
zzM{aw>K=E(6pJk2vS9XIqYZ4cqs#TAZg-7<+Cg_#_`wH`Z6Rpe-C2OW$QM{?kJ5)?
zW|=yPN0mcIb6t#9Tzm;SbFu@cW7oPp1f9v_#Z&s+EwX{boXuknfif_gdq?b&Gd%R%
zdWb?eC1^jJ8qm51By`G?geWqPTglqc<1iM_-c&<LKg({aRyNMJ*puv`k#V=Ww5RAI
z8RrY~P7ZZ#2ppbDKodX2IXF&-fV-JK11G=%JDVO-Sx6svJ`FjGrZheeJ6OEa6Oq}t
zMbFjGebYZnDfd#Rt9H@wA_AVli6azk(TrQcF(McxI3Qp1mfdr>Q*5j8tc5Vc*-YHW
zy>CVLv<S6*Pr(n(ny4sCLYtttO_-F%bX79ZIX#d(!cw894NZ@yHFHxC+|tcp()>~M
ztjw3byffPS7|tx7w@C!0doElQOj)<R#@lIF6G}ZJD`exAwfMf&Zwn~jqkM(KMB>Px
zNI!$6aUS=c))~K@_H-*aeZUkR&cx(r8ik0_+~f(2v&p`6Afi#&vu)9Ys665mQ8c{A
ztHC=9H<KdKt#oOT8e+k{=YHl`a>a(q9#PJsquSwTiW6w&lTy2io;BXMMccyFgj2>I
z$M6`b4UJrmKf6auK+!?fY2X>X+AWI?rs%o9a&LDQHkUY6P2HD?&}D~=>|ozpSR9e)
zRCezB`}3yg;fSX7`?N^fz;;dn<$z*S)9IZt{Bk0Mdv%d_f4m$*2B$SRJK#GgP82Fj
z^y@8K=-2w1PFvioogo@ifKcld;|T{?>>U<!O}RB-Z*=HqD@Jw<`m{Qsiw~L=A!95@
zeA{CpvX&r(M{?@c=I4Rg%u_2KYj)ilW@ihf$RjW27S<pfyQ*yZG|o~hVB+4%jg1Q2
zPq>)7w+iyv>8Lg)F(S0{|6kj=>}rxE$9dmRah5ZSLuK4OwggB5+>O*biVJ}p3Ia5g
z1|+E0SA6E-o~Np-38)DHGu>TQ@n=TfBK%@*?)l8PSCo8{c+nB_3qt{tz@>8Vx0t2K
zSgD>$Tr<31&`si+w<sJe66yn5Mpe_~cmzZ)*eAIW&t+u_UCo`QoeRVecR&fk6%|WD
zH`J4=vdEk1#fTZ*4X<M+TtAl&fS|uS)iZ~##WGMOfOKPb2eo6yb<aD7tqi)6+h7%4
z&wCALQQw{ltSwdvecGOJB6C?1UG4*zMerZPrcF;c?a@je;7t7Sq??K+FRGeY*-B5W
zbi{PlO*loRh0uj1sk}~bfOw@E;v4Ly8QK6<-7p(psUuN#CZWB?`++aG-G1$v#{yjl
zaIlh!pLx;a*)0VTjxq9%J=gvyUv!N1_qfRWDsi`hiwH`}QxBq+EAArHL_UG|Kf)p+
z*)Q@5H3B6}I#eP>K4GJcjG}DkV)7}Ey&NkQI+3+_Fej%=-U2O->P-ItTj!~2=nYV@
zzCG0g16HocmC%X%P`#orQO=-u{5-mM0+OVwp^<yRW7HNoNaw9yi5?%U>i}o1hy|Dr
zv9IaL_23Nf(N;1aRV35AgtK(D>3&xpba$kgtH{mQ(P#5|F8L&riB&}+_3~7DY<NaZ
zqjMqPl3AjLEJv00PtU``ZEWsJJw1=B)VV&k5@w=N*SOq0D*L)e2kjLxP?oV>-idVt
zkF?%hmj&Wc_qAj9$nar>JtQsk^VRcIb0W|ou8!427gFq6e4qW!m6BbL$d&N3MQD25
zYl1eEW6ws{mfxAD=bhP0>=QP93oDHjF-jdG&{%)bUEFIr>%@&r(9dBMw*}6I9Lo<P
zl9zqLD&7)Zmwj@4hgcGjkrexcz{SXu(No%Yruoc7`)H3F6R!<SJk<$S#1UT*nJ&$%
zM>0<~6Xe#aChR@uHMbih?I|^<CH+aRAiP{`<kFq)6Dmdm7t}ajvymGv$@k%9TxG}W
zNy1a|Mvp3p{n2314~cga#Z4?wF9Q*sE_PASCHW3wGgn=#s8o!KsjAw?cWhP3t1hLy
zXKM&Sfu0ZdE>o%+M8gAg#5W2mkCXpsR|FMwxWLb&9WCdJb>xL`nM=bU9~2+xxNYam
zmR3_f7WMKC^ce3#pHQd*6}^LR2*rc|Hws{~b}#!{Aoa|EJHB!(YB_7XsQs!53YC-;
zWBcV2gO)D(35q|l<474!^b=wn>)HdLlL}5b`};1dx6;uc&y5L!sYFg?oe?*e`vRj#
zuElP7)bU(h4{PC7gL;K7sl01s$nVB9karNT9jIMtkMnfzXHL06hq+!w$sTz!Q|Y2O
zl#$T@orE4(%g?;?!bDSd8#>R0Xskf!vm-jUJ#b|)g3e;U11uCr>2t2xvyZAeXG@l_
zpmFx7((!|aD+g-RZFnQ$9%!Rv#N8Xqp<MwH9{7CLi%vcwUZW?>H`gt5^L1D&c?Cm~
zESxOj;MsbPZKFiYNHS&Oo=EJg<drSu|9q2di5h%j`5g0V90-MttK+UqbqC!rS^Wi{
z@eo`MuFCb>S0_X0dYq?s9!GqnC{e0Xwd@&P5>@Ki{FM!xkZK!(#QHw&+DB!LK4I;K
z<C^uZsH(FgJDNj#qfnNw!y*B$YWwQt#G$)ZRH4|s-1XV-?2&WO<JxuSMB(qxtC(py
z)P*Q@(1_GZ2SV~6&owPC>i2?0Wjrlsu$7Wl)U3UCfTG%Vg&|`#p=@(_*>)R|An=;5
z<w1!VX;&8_^vZ40KO<1YK8lA{o>-M>0)Uass$6eJg`#St^9?g}+KX)u9G1QEm|gS*
zt_uOOKnDSDf3jDORwz+hK)dqu#dRIp4I>2+0(6_#Z$G`1oWN4|$fjoUHBnRkP%1wC
zy;!d%l#Wj%G6~|TAE}~1SFNC^cIxpW?la^XPAcnoTr!j!Gy^U5xzVLo^8f{B)Jxb`
zEVQ_GYINeI<vUOm-#rS{+Q^d~<6!%4)k@2;T^6_sw{K05P^%t5Jx1}&IgG9dEbn?)
zu+)(}E^zfO-!unkE;EH!m$O~<YJ{AN>3OcD?M{)yNF-9m(Ypsw>&gWbcJ$Q=*)u}v
zptE}w-V?D_N<Q?JbFzM`?uFX$%45=tIo3Byt@%&fT*Tg|$4^mEe^Dj+q#S!|fk0cm
zEl}ntv1ijKV0!)Wd7o<@;dl|Z9&g#sIe3Q<_}Ou5XccfZL+ZZrQJ}L*G=y?(+c^X3
za{M#32yv-aSh69G21r0|heB>n%jb#%zTbOrmMB+NhtBz;wy${oinU54b-&@Gg$GY<
zRgrQC0p48f(XxCzz>f_64z#x7B@%VS*Fa2%0FrFGh=Wt>Cc#x-s`9Ah*t?5z1r<c1
zMAh{z&Wj!^4XRJob%*JroMXqXc6E!NSn%2D+Lc(3Y;F@5(8JhgA$sz#OXN5iFzcOy
zdBh!ghn|?02Y(aa8*`j_522owGCl@2`|jUei2xDc8s2?|yVU|cH>+P!qg+>ER1W`g
zSJnWj4}-KsnvA7HTalQEO1|TAX(KRQx^%baI`!1uq1)&h`!!E?HbpNcYp<nABLO&$
z>KQP?NTHL!a`{MawI1j)FK?2#$OTukD{>fqE}&Qu%Ba+qLyr~XJ-0epr9i#k25ZZV
z3LNRjY;OyNF>)^#e{g%vH$6oH^cppsrYJmXvbnnVi|AKf5VTfrH&LU2QlR*lDo2gf
zTtn_s>LR$Q^F2UU*ABk58H}x_Sm<`mo#cp6QCX|kvX@o!u)`!U0@zj#J*JfnZ-TY*
zs1!wcL9O=RKqncGCbCn@0~|MDsr<EtyqZw!-YSX|NM6J4-P>?k=!F-GLLau2CO+<6
zXw(~~-}=2Bdlr30V27!~KbWlEdCY6t$9hL9_@YwX`z@&R$$hdSXazW@_xIpi&!Ol2
zpLPXLbEd4FKJ=aa`MT=uh4T(PX7{*{@YcTzoy+g_c%tm0(`CS1?3MNuJ5e)Ndp*wG
zJ6G%Go5K2X*!}FsRYB`J0}O0e_FO^2??uh3Tmlr~pG0kkqHgSr9oG00yV8|^0j=fn
z-duY`#2$1#zAr6_X<RgVM@1J5VOREgkDcd7!ONwrETR`(anPr@EmW_`0lKP}ibEaB
zGa<Hei^`*F4)i=w7Zw}iNpz`$oMhSGsh(xUTT;LlN^mu|0%m4?J1OAq{%KwB4ZybR
zH<x?^gq`j5Hl;Y0(A}y!f9U&eRV9ZOv03Flm+PE$+7Ap@*4d#;CG^~T9pK!TYq)jw
z6)A2j*EaIFokoFv+;9Rwm$!3(oL2%Ip5jKAKIb}Ao9lewJpUZNSC8~edUZtz#Y<Mp
zL!QmW#g|Hmc!L^^{j`IZTW2kFo4cng+5R0baNnd}yF#63@qxeBBGq@Dc~EtL!RO3+
z#*P=b-9Jrz7tl&0a}Ts^%Olh&x(SilD}w|I$HJXq3#+Y5WZel6Wk(#X!-|1lr5I@B
zd^_TqB>t*s`SFFJp2-`IH#yfkcIw4exM=nqt6;D$c)u6CI>r)y)UlItx~P-;qlkt@
zLJ=0Js@JZ0097vIxYqa1Y(t==sun-qL|PP3?7=-K_M#}0n5W4in*ut!6MPn01%X!m
zfmla-4^Tx6s=oFIc+8)OoX(kOr%Qz5FCo4lbj};2q^nGz)Lwh~E|eUF+VZ-+^ZH!Z
zq*GYT%_n|Qc5D@NE1+gV0nwbn`Di?HN<DARz_1XN@rvLtpgFs6w)LK$mUfDEw^8H4
z;0Ze2udXlHgjY@qsz&s=;|F^Fu2^t|0S&<t1>0ASBwf6`{xrh59aTTn2$TZ|iWI2Y
z5~|>~UcO4l{3>2>OAf0iy2-4Q^b6r_P<p6#XCX%AV;|}X;#q(h34^YXm{;7M_vp#7
z=Lt+5VeY8Z1l_7a#mjp^8<gNzd1Kca?Ucwj&xO-nH7|IpJYC7vpsRF}x80wj?O0Rk
zU_g^9-k8~5)K%z|vzxJ2L&ee+ta8=<ge}oc#$RO8yGtZ+=z)(BZ!A3w#HLcSy)ai$
z9h&7*@Z#B0C(H^^)%dXIq|WN=#M9YU&guNO)ACvm;~$jzW93_ODWM1l7NQdXF14o9
z$N(9Ws;b%dUfB}=O@b4aW04^62R=gw0aUFPRZJcAqa80$#8VyW8rdB$Q27Q}zd>&V
zrBwkBvF2((nXacGg5%Yx=Y2efvS!>YxblxjO^<I~675nr(;Az1n<WI=Fe`zDxI;&L
z6Sqp-iLw%%qW2PwEm0%@Dgh3yJCP*o>XGQH<oWDk!&em9_oY-H0zDtLXP7D-S_0iq
zt6u5fgACQxfN=O;)AF4|%}Z<(G)$hj;i$3D`N9){Em4#pw9383lbt39h@0?mEcbVl
zt10x!EN-XI!DzR7J*eEE=1WRuAUll?P_d{g%Z}AgQD>3H*&IrOL^t)?j$%21WAH5%
z)@}=j5!BF%5Z5SIT1a2Z6Sf&b>LR4f>7Pi%=RoJWJbi_hFN%jhoH`%T6Hqhf>6a2|
zs*ZvTvLlcZM8>;lc@$Ju);r!u{J`x*eY(q(K+h7pM9zkQZi#LL;Gn+rT>lW~L|?2*
z|N0$!eEzAt_kN;A8Wfp~s_tI#l1g5+Iz7BZo(pvQ`ZX+g_8o;zUw*J1a{}ire$12M
z`<=jAddZ~^x)QZU6%XiZszP+uo+VX@*f1bTz{H0kS0b)FP`nVOiU4fbEud!&UtVM#
z0hB=}1k}P*<{ix9;d#dH$Bk`s5IpKmfhhZS2n1~f7fO#Zd{NLfe8HS+y~pis_`5r%
zwVXrWRMChRbN!=19rJXazE|s9lm<|~lhv^KL6;Ykpmmd=;zeT9L^(3xcG}F@xDzwP
zV70V|$1#HD2Sx6->lL`nD^bXd`NJKwMiAiy0mLAP4=*704{}EGe&7pke?5+Qd{^dC
zv>*;A?!*cPy0&jmJD4zof*r(8pefFX`VDESI>(2&Qy!pCj9IR%&3Ejmk5{7L(C$P{
zZ#Wtv>pdI2(Y2TL9hviiif3_J0>{m<$kVv}_Qc0K(DAJL9^tCz$xOEqX}q8V4Rk!)
z<FX4iGR#9%R<8mH;bDPOf+s^JcB02!GiN*Os|hUBBzQ9Kox5o5gs()Y3AV$4NGiq&
z`*K*DpmVqa$60;ur#DcW8=lxJ7JCt>4JYci2w4r<CWvz)g4W2Pa1op`@;vXQO&Tc5
z*ju*^66mTDJP{G3=B3fe*jaS_%M_3#al0-WKY(uKo4K?R1nHq*zZ>%Ocf4bH<xN(>
z=%Eq;s0G{9xFknnxD|H^qq=qA5^nbn`h2mu-6Ux3X^FNf$rHUZu_mF#r+1txvBWz4
zeS}3!cOdckcuf=4n>|t5%{XtUC5W%&me$VuiXt9`J-l=KYD<GU#er?;2_JE=EsAi$
z?Fnxk6oGJ|Yrc2z_}IQVk;2QddpBPxkMkQ#f{H<#m%ZFP<e;`lVkazOk;nkm1`V|!
zvZI52raNOVO*lZUC)5DUk|P^R*kap<Q8s9-Kt!jCnJ{xsXPv*ND*48u9}D9|?7-V}
zQ|Ol3iI~?pfvNmQRV?Jmp<Y@kdT->DKv{T&t}jQhw!Efl25o+5adH(Gn$70j+Niqf
zqkC?fl{0I$u18NCnnF)ht$|cPb{qA8mS$PA7tdC&io2;6;naagNCS>4?;%!j>AVjo
zI$WD@sGps=xJ@X24TL|PY*O>vB#QaSG^hPq9&t=xbWnUu@Ytu@4eP8~1hwK$YKbPj
zpsR~&g70E0ugRX`(nb6I;)&ZM`^48dgDp7NV<DRmet5!q1U1MeVzuSPvIJoRv4_ta
zff;Ck!jF~x`-~Tz%~he!3;UwS;J>Rz$(M^Y?CNEqnZE)(HnKBFDqe7){kde5v$Dmm
zfE(&V=Z+lJz%LqZ#P{854n?)hX<n<fJnVF3tQA?5!TlhRs~AocXM@g5e~ZSRfC&9=
ze_f)C22{(XR@0zt4iq;aFcC0{_S$98f<bS$Bt_yWK%vJq6u++SL`yYqb=ix>O=O26
zR@S7v5Z^f7k~zDlWYT8f5pM7ERQG#;rj{g-QF_VWAFs>BWzQBJ>ohm?alOKGD`(DY
zjK@=`=*;F=`rLaXza9w#&~^KCw3!n)H>6!ObT2@+Cs+8=^L8P@rsUrfrJw_5K4L%K
z`>N!%Zn27(75x$&C5~CwZ5tRzMIW?XE!=2I1BxDuLcOi;nFpvvntZCbsd<Ddc_x)*
z1QUE9(?WhwjX-lnL_w@OF^_IfdA3fh$W`CpDU33_J>%1Bf6>MJExuR;kq5mQ#tlMe
zxg6-8Q9V|2UTZiX%t{FD4carzE5RR2I^7w(a_iWrIgW0(xA#VWN4j=jQ%muCrG#d0
z_nA?0`lwc$H%H0TBOr7JPp*Vj0!Pt2yHrPw&l@lB_Mh33GrMS0O}G-cp?Nv*0=J7p
zu4srVRCz?D=|^nUsD`Qs%{NvfvY}9ZNzc}(vH+dpnwYo*dE9ej9x0<*jRJL0QyUK{
zAmss$-8!Y0`em}HODRXVvZw@9{V>V9#d<ms4s|8WYN2~FB^zrloqkvOiP%7Mjoi9l
z(8Uwg)~V(pD+(lj`H_g4>IZ$5I7?Iufm-)dp-I>R?On-4R=72D>`+HKr8RG(JfXCH
z{ZRfapjoM9Q}V3b*9{b5QbciwMOo#fWdn99Nx6PoKY}9vD754a1f5jybXjxvxlXZS
z&)Q6qM}j1QG6uAUVKa+#E_Hw&xs@5GApO<a{fa>;kRYd)B}|h-F*;t7{tBKWAQcPb
z*@C4wwb1CALI)bG(@0F{6VaII?OHeUJUesA-;YZCK?%&6CnBeB41I&-lL^8vA)ge9
z+lgK%>@y}B9~K~4lMm%u<57!rdR`$?tu-_<KsT}`U2U@8JGDja;izWUG@ZmAc!fb%
z#34b|ik&oF38!{K{$dAFimBuQU?rRfAbEwmZ%v3-Sj<W-g#6^>Xtd}i4KPBc1N3~t
zUCP%xpic3gqR;!CYNuj^5bL}R(*c#n=px_Top7>$@QxBAt$DL(<$!HYWUWWd%A}ry
zf(*hu7NmYqu`r2sjj+up{Tzsr5O}fro{b8?DfC*vV2kQI;%Yruhf}Nax!Ht%BGUx8
z;Q<i&Y2gV@S~{R;LO4We;fu1T9F%)8W66MH2DJpA?#At>I$kYd4;|?f-6)!XK0(Wv
zxP9l{!C#*dl$^MI?6wL|<m-alcT8h~17V{~?*{y#D1^N;J^OmsNSMYsRMxysKW3i(
zxQge=4@pzB7d2Fy8C&X3ls0hd!=oFQQ!uG$p=yip2EF$Zop8R0@jMZ8nrMY0R!=CU
z-kzUA4Pwbq5Bh6@PTUs)kpvg)9okfR#uQ3$EJ2mt;ADNzJiuA09z8|d@4UJZVI5++
zKY4(&=uVyM{OvXFC%hmOO^C2W7et}N=Lp}GP|=#W=O~&G>Fe#i++{=Wv7u@opcWg#
z>@z<vFW>f$CNe636O%)EjR@Avp>@spKKlz9Lbc>oY{aI6+7iqX5)e||xoCahfRfml
zGgdvQBQQ(4`a0A+Zc~W}xt$u@HmJcYA?0Q9A>sAGEa~!&R!(7{Dh4`Jp%-)#I_m0W
z@uH{!3teNp(?v>MYuWkuW}a;gysa*kpPs0QKFVK50W;u2bBct0`*Ih~y2*A_zv%0N
zq6Q<hw|MQ*2`%Tl>!N92y2G3jq|xqkUfx){#iAM-i?nhNv^GpPjvFEpQ^rW~-R}u=
zZ_Z14W#|lL?1}sZ?LoTK)GQBsoGErCRES&6DVzgf0918kc+rhgM3*>Km{9nq-S7EX
zsxQ4_(g=cjVv?dgG)O^dY7W|D;a$|E56~43TzNL5$+ljR5b&JD<sj#iteugKm^j0@
zL!?Dp$UuN{xkIFb9oV@C=hG$Sy~2i42G4JJ3x4*b?*W1)upVs2>w$tU`YezsdWhYC
zLbPJ9ToFT=#6gXH30nJ%dx+;p!wMJ{VFgbdKlmkGvPbU>rdn9?+P2_JfG*@rL^=pL
zlDWj98TxeAS~&%yF`@6ct>)QmQcX+>vqVB=A{9!b1|@O(d$z4K7kO|Cn(8!_xe4vT
z1)mb|mSrGq(7-0F6yA*y9_>Ua6y_LTX}evkgd!o$nR;5SN=xdf=ks~;Oy^bQ{pR;!
z^@C3C2M=sT-eua&?Q&~vy+@!woKRfq9w_qO`a;eUIPwVVA&%G;9JU;bEZY=X({e}r
zf#MD4OwBx-(TRhvw5=d4Oj94mYkG%QI4KcB(F8NSMy3gr{tD`7QA1Z|(knuWCMskJ
zV_>U4jL;IzW<gp0y<cH;W&stKR;<c!_yHx!4pFW}o@LT9!mnYD+#cu}!I~u3py%%g
zOiJ938IkFU`v=9ut>=}$pJ+JibIPR9#Qj6kL>6iAj6mtzu2rH%4ctU4ly)oZ4))%H
zjcl!6kQ(ZFpi+J9Y9w~8>V;S+5*oZ>Jwk|}yWIWhY4>uMtUlkM&edFVU_mRx|L&3<
zs(FNpMx<W69S?nBVbaZ5UvJR)KoKX@*l{g~g*shxZ|T@?_c^XdP2~0Mj2+y0P=%ae
z7uEGnR|MKVvb&Ekw9K?`O3CG-7_eQsaPOoH4J|F8(UL|e`ctwTvO&TkhBMqCfE$}*
z&q7uFlTJTs74gh`VNPMeMF<@o62{gLwg*MAM$f!Qba)}qiKGeFkP$%8O{u*$fhd<)
zkRIU^b?<MMV>>b(x(a(IcHXL?brDuY4|LEj-0hd!2uG;26|k22?VtsUWPimj%iU$+
z3~t>Ti_j<-4b0u^+3!e^m^o9sE%5-i8bLxC>am1^PDD+(SP35rRT~Jmz-VLJ5;b%V
zI2ldV11;2_GFLii_CV`H<9QPFLE{Ca@R6a%WU@Tjd1bHH^gQp$8Z~z^Vx#*xC`4@c
zn|fb4g+amBhtjG&E*PT)^8{a@hDh$Bc?6;+q{vo8JV7bb5jw*6Tf|=xH6is^`g;}U
z`MaA_BmDt55vv|5m&}{6uY=K3mtbj6kULcBZIyVoK$h+zv}Pxj)zqd_C~j;?lwbte
zw3Jm|G>>3c;sSak+(4lMbAX+UmW6W&jdn;2HEuX|8!QtNdiLy`SQmkl<3?TYqIm?&
z^`Z8PmQxra4-Qpfw44HT_=B<}S<NduX69K^NG92Qr?B9CeULWl;gl|vrzlq-@zNt)
zv(&G#2|dP<<=M8_IwXF|?L&OF19BW^Cm|6~%FRN7LUz*FXiE!Ps3s-DGJ{Y82#vrq
zA+aK0iXdzPy27SdSYB{bmCgyFb5IDl$`XxtXeS^0LB^oxP3X#z2t{Kc(LpxwUc%*0
zuP&)Vp(R&Kh;TSELp;QpSG!pG?i5O1*dhtSb663`S#*y;*kpWFbXx>>u}Oyt_}7an
zbUoDJ@tWL=ydLGZWEW~U4xox7poq|8p%C|SXlTm)t#O+_4av{NBCfta=POWt!D0z!
z7bj0kDk8sf9Dt>JNX=!QA9IvBwcT^zz~9&DlkE>GXPNt2LojDSiq#7)P0DzcaSBm2
zA)i#|dj$(0M@LYg<)X(pT6#2Xm3Oj8y66+J5K>%ncbdK?%Jl9~G$ar;8FD7BYrG`1
zu>ETqxhgXE^D1&O6kAuK23>@VXAfoOg;XeI*h9T2TY!yiz!$|ZwLn=xJrrV=2T)ZV
zE5e9HjXe=6pkh=}Ubm2|K-7daDOU;fp_;FX=basS1ys0bsiWpM56kOgiqO>l-WhZX
zO;M_>hF-jtP)yhvMrhkF`bSWcCL+&mXeV@1tp{~$gj~U*F8~JkP{ZaMK?XItHguWU
zBgdZ41z#0Ye_V%qv0GAP3Yv8Xpc6JCti^0NwCEtge(Uy6$+I1oqsvG*u58>LNF>Xh
zRhsuEdUn>QxVLMGO><|pgc#Z4({$6eW?&c4+*c(CmXkt;<Hj8FYs4qqvYpPOvmAP?
zpd~wewvi)|!jr4I(ZWh}(_lwtX$0^VN}+J3alTpzB`u^v*^)eN65At-1KKUhBLq`u
zAr*?=1wji7iY_Q?#)s-XqSSnu2<^%_tniT*N}<ST;7eG@{(vr2O_Xz(2@ZmUe!u$6
zYZz!a%HQrG@4x6DfvO4DJEKq7qJhM{2{m$WfD4@hW{`A_WlqJx)hsnf8cP`up`=Yo
z`-tJLDq5NM=Uc_ya?59wFaa(&CeWQD@w8|jG2GI06_P;ZCx|SZw}q%U@4x7%!OdSZ
zkHEIU`dcBP0=hVyc$8!3SYnhy$xG$gXwK9@sZCGxxD1cbT|>5K(L4fnV^=ee_tNIV
zYZSpcx*@DBhL1j_=dm2R=zyoMt~{{~;t`^0;!Nr=skSJkaxY4Hd6JCtZ*G@60efx-
zDL1X}1PqSQ<6kYI&z0zW{IH{j>t_gs5?AgLhI>IP?x-G1?@6?Ec*rX22&>T0G@$3m
z4R9n#>qs5oHW&pQ(m%u_ZINXF71dU5(|t6CP-ycY(;g)+YstM;*k-*Kux4{Fm9S*a
z%2Z3rWwrJmfx<zB|IT7n6Wvl%<+^sgISni7x8QE-j^!6r>nypA$SE1*siDB8t!Tyo
zXiA{jj@Yb8)y(sHWxt11hc7}a*CGZILx{l2h~`0t&Kc~SraGlO4HZ9&p=Y1D@0B4{
zK`$LOhJF$}ZWxRQC8ul7X)1F}sc5h7C7m1W%|<)*A&HvZZleh11!^6?kR!Btw|m1E
z<QlnF3r#A}Nl96kI7^%jjZo5@l+R(ACc2@Ru*yE6<ZI+6=ruyPHwniEQ;pV3U38Rd
zPl9TvlQeA!?7R=2MV%p@3V3&)(!-1~1VV2lzC~j|6RSaQD>p08KF-3u<jb<91Ueh6
zj;<cQ<wZ9PlkP0jMAtyl1T>UpjE)fyOyp$OnipVdIH+$v6l@HEYw~$rEz&G6VQ=WT
zUPE*37^7tb!ito;Op8Z@qcv{$Gz$ko7n**I&CrmsM^YX`pdghILHG1U5*3sgD5cs;
z06lFcNZeEhIL_nqqL6k-)%1`DHf>OzbuA@2l!3|!AObwD^=}AD@P>11tmW_3=}x6s
zp($(9;+|UJ3i|Vb9`k%_h1<`_Ng3_*JPaSI9i}2(zu#j7&BfF_!So`uv?f6fK@$=I
z>SuV-(jO0bu|MV!C|6M!d}OW&bRrz1a8Iy*Otk8z?T3<`k*EY^wY4@&w0BzRkcOii
zD^Ti~eOsILo$vXB2ba#bbiF9MtW9c7R(Knxh4k2kUE5}?b6p}JN;0cQ-~|np2$kE!
zha)=h)<LU`go~c#r0Sr3th_PXh^kTg(W8df@Zu?ICe1!(ymG__MKmD@6g35=imX07
z4i*~Erhub=yD%DG)X+0gBrCi#V~ax1L~mU#w)pn2`Bf1Xbt|CQV6qJ?1apnOclpPA
ztf~2;w)}>l5AwaX+2Is($D`c9B(lJ^MCd%!7C<L74}68x2r<HwraQJq5VQe}QhF8E
zgkGZI3hw)KgZXDt6P)*cUA_t+>QdK`iIo;l;PLP&RoOYVNNP=J=qi}6Thl*>5ZX6n
zLeaKYWnct*K^IwiUl&)Nt&4ww*;Jo*EG~qe!&w^Mv(Y_IN0_1JVY>T6567j?0#(Z(
zO(Ar#4Vlp9;nO;UhXE?dXO@Z*Y)zwU1SKk6iT|0GwhN>MndRYefi8Ce)T~<=Z6msT
zkT@bOBf55`U4z_g&|!yX_G%3cGNTAjloK|S6;GCq5yMldE!*fAoP7u|T|Aff2xq%#
zktmuug|l~fs_ESAnqv_rkhg=$ZHcbssiw8VIW)o{%TrA$js_@?y-OiH!8Pn|+1aq#
zR9NjSv9_WB0dx^Ez{au=Z8JPU6og$;;vKs;dL~lT$n6nNaLY>CBR2sYLF7@@rab}`
zrUX@GDE>+fuuATL{VtTB%x)t~sfvWyEa)t5eC_0kn<*3`lkU8wH{1lO)IV}3Pg)+R
zgyI&+d|M+gVXEhev^uVf3@8E0E_pM{DIgEwNxZV(>7IvBWYf(R1TZs6xKBUXlS8Sh
zQ(kVBXblZIp`CyRv(@MrAvqJ1P4J@DJ+XJ*`Rj3+vzEk_q~&l8^mq<c(kHdOLo24Z
zde}$n5u;_q(0#E8AA=eEFhqiq*<f^xn9hORq2g}%gjddF=Hph62Iren2b@FCCV4g{
zh?Ohq2SAM`Bd9v1GMS-)%w_<Iaz`xC(0ViHo30sA29vB?m$f~j2Euqz%t??>I{u(0
z-U)G~JFkr!xZQ>yJ+v7%ZXl1DXV`*&4=P39E;gnGcpSP~Jr5d!f?Fz5!#~Cm&1N<V
zp(IwcpoB8dhMtZ#hvYuSjWvO;8~cbaVM|i&u93%(&7-o8xjA+}`;9tib>Fl+QRKN%
z2NmU{tI<^co_23C-4Kuo@HhzWn`XCFz^z3VeS|1--%_{ssM1;2uF<jiQux9}thhT0
z;Og$2@_qL}>Aa)O<J^b1SvY#c`7ppKxDT}QxqDhM`dV+)>QFS1gB}{Zpo=WJ3m*hV
z$$j%Yuw6%Tdw)NtrdVQhGp@c<#)rI#UpgbIMKKiE;@!sZTwmq{t5*SvULz6<O3Pn1
z+R@&L+7R7jec2{9tWxxRls#<k1bl}E9U+0f*$DRIk)jdhx1gIkbQ$GA<=wb}+YR_$
z;(1Gkdtak`_7^3k>sxhvi5my>@kH{@xWR@P{5ZV_rPVxHpW>TPB^u&FyGmAH+xqd5
zW?lplJ!cc!r6TUC_+z=ejGjZ!j^5<p?)|)DUDJn;CbN<o)|IobR#?wq=086h==<}1
z*-+=d@1C}yKsuoT(S@1d>}~_J{EnKmLc(0Qp#xCs9;84#I>%O;Tj^3;6hLf<l~{mX
z&8XrjQZoy+)fH}Pj_vgBzF&CBiFFg(1mVgsLpCfXg2>$!6?0}&nIgiY=`(`T9?Ju?
zHHR+wPF&iK5}mCtdp2jVMP=P?>fGZd+gh<qQdVvrG+in54XP-6Q&rd<s*a^roZ%Dt
zdZ3h_b@iBxr4vWcN$E9fus41nZ2<WFlaCBPa67-}d?_#5RF%SHv!oVOPAn9X^`5H_
zHm)ua_#0W;+-E#nmn+GkL+u>^TO{*00-?qI)90Bh4=~kB5wz#6h1gi#KkXI`-ZuiF
z2%0c<hTH<E%BHZrg9pFW(y|thGDkxJP!&aptrEHAO-*3)C?4Uu8E8SY*@HM?c7ZP1
zF-p`Z5)*sqE!NlAqY4gi>m7&MJZqwDIAa~f`uu$_!>M`a<%?^8c82;~mumh^uy5|^
zEN<y%ktXOC-GtK9_8x(!>JNQIdT1lDLf}LzLX_lys->D@W?!z%n&>7??o=L^(4;RF
zZT6hQ9&{LyJXVmCHw`1iH-WY>VhvS%**W?ptp;4d?F7`3Oz5o~r3dqlDG+y8eYczv
zI)Wy((Fb)0U|Ei%%e{D7D{{RXR?nj~DVqW8sr5+_Rzh+)rJ*;iOZ4<T=&|>__TsG6
zHOC^B>d$LNnsB2HDy|9NWzZxxRixBb>T4<`=VfSg_$m>7_qMtdI2W?K5W)}&F^?$d
z*nC3xX*{eFP*MJ?OI<bQq$b}*D20$U(BrvwkFe%eMVE-~o6=coVlj}tt&d<wQ@2f*
z2yAd=i-PB}rR(i)YpNJ+>K*7|PLO{_U=9>DVOdxG9Xk8GS+q!6?xiHZ47p=}wLI0D
z$aP56Ds*Bt#Z}99u+>8ddP{W7GkGP7MmHA!@--5|`#l+5B6v%r$7iuw(<Z{)F`Xsk
z#5W-_yDNIUSonXGCbBw+KyuOA$vyE}%*Y6DJivYTq!r9=(1kRQk`KWf+0>FfPRLs$
z_@$um<2~w{mI1P$&-N(MekV3%cM~s+VmCl#!MqQAj$GR36@d#@5hz;wfKPZCn`m0S
z^!%(!-iU(g#U>v8yocRTAQdg_J>DaVSeplwAnMdack>_~<TDGj3@lb+S}k{WnnVx)
zmDWp*H;E<@bc9kOZoS8aMW+<uu|2~^q)Wu`U1?i%limasA(^1h^8`m^Z<!iG>y8cw
zpfrG2BvPG#FM7PZT0Ql+@*+qR2|U*lU+(@3&wc7en>qBXFr|3#JMwd&%Qx2c%*1wc
z-djy(C!mK+&;IP*@_EnC?k7gFX`kzX7LiyaP_#X~D|DINM{{-2K+gj&LRT~Z7OHv!
zNfcSjBapn+qoK`Y3-m0JUVqYQTArx!ar4O6HIG1#qdG;hYqzThAdS3-CuDbyJg>Fi
z>}~T!R~5wszwh#7gc`N}h_cLEhXi_zk0LT1GUz4HO?O98?1K6$bdq{kA<&iXJ%&3)
ztH%;Yuc2Nptx)n$fTNcdV8J&;8KTB5pd<)ZKa?CUJJGXz3q?Sn2I_<pC{Xj-eL{`>
z5n7Plz8;`mOfT4B(d`qImMWqJJ>{m>g-%Rq7ZK<=GqyM}w#=@Ik<hb*i!AhvE2#y#
zI=N35(-?WZ#u6&-UAab&&`IVEDT(=l=e6nr;S9W}k@SItG~xwQQMZ?l(ZhQON|CD2
zb8lj{Rl1P^Z!`j|5(Ve<37F`01bJ3qt^2i~ySNiPF459cc`l7lpbE!b!lLIf={7U{
z0p7;0LO1t8{Hv?F2s(8Rsa4#vH_)>T)ahhIYdHld4B~*5D02aYY2wqR&MDE2;+~W(
zZp#ADvzIyb5oP2%2bJhR#G49Llth)X<i)VtqXLwir#N@Vpo*0PX?8-jLrnnp2(+Rs
zGStaAo)yLm!5wKs?3stv?SmM$o*dbUA)v}o^4Z|yvDd*{4DET)F^2Q-_~a5j&f;{E
zWUKFC=YjwNNpj7&<I+v4`eatUy{(1Hqa3a?oL6hg%ME1?^9uI)sGM}SZfUYmb?4w7
z*o_3&hhiEOdL*TBgA_tH1QRqEF4Y~d3~D<#K#K(i+LTR=9PJij30-VvA}+Q(XhWe=
zPsU<7#6QBQ5pnX0EP{4O)FLjfTL=M)z=rczZCaiSR*4>t?O}O@i3%#;(zKZ~IG@PY
zH%H6bJnWh*&msucSd1W$AY?K`tSoti<1m}hy5oK$3<_)1*`s%)8h!V5-}*TdRSr|a
zBUW>4sYD_4e)m{3q!ZeF;ft6v=-K&a_a$M(JA(!d>^Ud=j+N-_PJNVBc0u}@`A2HG
z;qnIvCR|t>PoC^-l0|$-+m})Vips9Gnth|$0w*#i+=RVGqN`5dmD*k{=e6XgGWNNZ
z0idSNJVSA<ys$0Vdv2q8sRl~KP-N;7PmB%`t7`?!9;`x*wmBTN-ncWUQ_{Gw`oz4@
zBxQ*k@)qJ_nA&{(nRSba;Q>;YHC*L0LFhTYB<!6hL*}aMYp72=2Uaa45>S13Z02-{
zY_<>AsAzfI*{BwSGdxQkV8WY0cB9Unzdv6*yAmee+Bp=qbOE(@A0ZcfmI<T`p38a$
zqy_}gqU(&nT^&76QoAqcQlqW?A%%)q?YVL&I!KQU3zGq9P?SuVK$-Ex6>n{hwpf)#
z2{`7%6vNB%T(q#!WUYlELxTuh2bJ>YTF;sR&4l{+u}2<Yh`Xh#`7dMPw)GE+&O_k~
zbn4ThJbdPx1Tz~U6KdIm>myXHbEj+r%PTt<)P%0RVzE@yxGB)C(mfz^FPcI;FB(Lk
zWkPxPzPe{Zk>ug$rN=ufW7_=mxg4NrjZTly5=)Yv$l02v{N|D8YV;QTP<i-@E!Wh`
zMCghefFFp}t$O7iA3lB{EE9ge^Grj%%BOk^jhq#F*+QSESe}2<#EXY$_LMw7?KwN>
zr7(dR-J^I15H-q?R8BO#FLjqZ>ZWxOui-J|;$a~RO7d!|8@Jsn*y@F%bJ5Q>z(T_r
zfmHG)O0$*^X`;ltxz%v_!(~?D9illz(_CA=@1lk1&8GWio(OuR`H5!9pLaSt=yi?G
zW}xSTJF7d=AL~ZGg=*KBYMM~<2<Ofh|Bh~fA|Q-xeMZL^uP-|3m%~+=wVbV0#K0s;
z(ncQEq(sgFvqs4yY=peI-O=Z{7Pa}_TFZDuJyEG=Nou7g2;)aXk3W6yHYobV41MOU
zzyLkVaH)Q4FQo{yJ-(VAzwgPRZBorOq>s9CD9jyxKH7V#j?}3}KDSpPC=7UwB@0pl
zg`}!w>)AIRB2+mkR>Np62D(sp)XHtK_JbmfsI~jKSwB!4&`VT=vEB<KfIPLJUY4dn
zMD4Vt*o*_93Ca!Ouy(6<go;En#ByGV7Szau+SNq{g=y>CPn+{a7x58p?z|;JpiMmv
zhpJuIi=xBiYkJP>n0AjI`7uo6;m{$X3v_X~01+eeudW0((VsyVWrn{4h<uB`oi<0H
zi$de6J{KiV1XFnVLGQ8$DCrFrliT(gaHCqGZ|0Us2;9>*q}lQ<RIjBSU(($Y3(X;*
zwM=z9dva`{PXF$-*N7=(q^L5UN{YraB*S8NN7^f%A%Rl!9f_uCd0fcOO$+jvCC7G=
z-%vB?5oalQg3wk`9jG@;E0mg2(3O^c!4sstohr+Bf~#^=w%7N<&PzSOj$8bmi3S84
zbUnAp9q49ppDv-Br>E6B;Jl3)YC(MqX311NYpf;g&m>`yK|yvjP`EnUdyX4A62;Kd
zp19vPi#!UI9*CF>i|-!g9>}anfL!%ffkvgTBq`d?10W-!1I%%|E`yE-CykBrxjKqM
zVTbEv{RT%tC**3Oc4@IfAt&o1_T~M6E>U}@ywsIvLvvoc5(%ILCyozdG<MrEg6;}^
z3PqrxISIubLb>PGAfGMk7R+<dpaFswSacs!-nojsLd9!mx2|7~0Bc@f*lq1D4DOTf
zKab_e260H}o^fNOjYgT6f{8t6(5Lx#yKeRbaKQKwxO189x^lm+)=Q#OEu0Yav8Uz4
zmYO$Wutbf3P%_=1y^WD)gZQs!GoxRw&`pMUN=ahpaW^SSy`mNmmk<r2COq=lVv$sY
z<qTm-vlg!sdZ3g|OesG1JVH@l2NsXjY965y_eXSYULpfvBP-zRbdS}SAW!p+Y4xjg
z3hE8IP)7?^mQY1=St_5{FV+6w(W1P@esTs=VMGaD#UtZGCX2#_?Ms#gJ)X?g)5#8Z
zLJ0xsRcc{n7d5vSvU2H%le66fA7p#4XXnH=uDgFv7V8fUBFOYfl_wq<%vsXT$g4ax
zu8ew7taBtwkZ223lCmhKtXm=^aH01jeU`CuXv0p{3id(>RH6hE(ujbxL@}ZJro2EA
z&0j)qo?76djBTny(DI_FAeo@&xwM@B-0~=fdYjBZnf738vv6*U^n$S7)nvWqaZOz;
zKov2$=;nh2oit%`lyo||Y$=}fSeca~$9|`uLEHqjRaibWVlsfHEyN^yHUgyzrhuYO
zsssF&kDs28|NilO{LN2)`1tA1@1Kqx_QC)Ebbd(&$Ou9L{LilddL+@Qp1=J5&-Vwv
zdKN8CIE(jG`RA_(`qf|m-3PVpYNc`i{r5W1Kl;hlKJ{23UymEuFAnSf`)B>L9Ps%l
zAD@KE?mz$GPhbB=&=>sAAK(1*&+ka|jUQ?E!n8P}%e^i9|LcGM!=LW8{f*D-oyAbp
z>iio5JBRqopMUyKe|;XLH~jth`JX=S5FhdLKYsk(@Ba1ofB*B}o>9*S-|qS5&nUKF
z`}mZ)g^zj|_0wPdoG1I{K;K^T=e?BA_gen&gA3-bt9M+SzwLTS9?Wyw`2hEiuh)m%
z5xSE9bFa638Kq*yb4(5RujiAxW4=8Sw&@-7-@d;c-<;>$+o8++e6RBlKYaMm!rF&Q
zqxWyS8x^gFwm&}0qwH+nkaNKIGGbH=(&hZ;KmN=A{{3gQ-~F$D{+B=g{`0uRzboHg
z;rBoM#WVLS_<S$n-@DLHeBt#UyilpC&wu|q@0ee^%;(oG^F8*Dt}<uu*T{T7)nMlA
zuPlTRZWmdNM139MPJDO0KN#zmj*~+@jjwp0Ui?=+&}DHy>|HnWFF)VE{ilyVeEi@6
zc#NGxkSJl7rpvZ%-m-1mwr$(CZQinN-m-1m_EbmAqC2J|=HGprjL67E=9A~eIoT_T
zLy6t6^8`}#@^W8Yh2--EU>FHI{r>sn`+gj~xp`?F?CIQMSaYl4^DdL7JZtm&kxP4=
z8tfkRfIl!x;u|ZSQ-*W#9&<u5r)Rb-4E{&EH-vvnJlEBjt=&F6^BVUWrT6*xbMZjN
zu$n#I>3hIg>3cn@{9c^i`ZX?z9dwPsbL;mpyzuQh`x5st{q&G?3C{*Mz-UJEt(^6-
z<1P`r#SYT#_Fmi7&wC8_cE|d9m2OviBFkj<X0;-iL6sPO!tlLpML6})PI_=lB{w>{
zNyxW7*n|5JWo3|=cf;Mw6f%$Vnd0}O=fr=`2xVex?BwidVqo)MCp$w+C>BNndV>Et
zaC6g%T39=qIMRt)8#tQ?n;6*{o6t#{*qS+;6EL!}v-9#o{ohO7*Q+(hVh`Ejx?j|W
z;De>0n=v2`4d|7CC#wUxVBg~t6xPZ{@_&996U8Nysl#wX1XYL@T7(s}@zBriUqtkj
z-^=Lwi+|4V-_$(c-QM4v9zO7I((HDwL@G^GDto&8beeSg+9z0g9WRc)?sl>iznQ+$
zA0Ol;%X$g-<!<ffN^g$uTv@ETj%w}vx_-zJT@op(q!!~<sSmRq?bKk;OLwh&It9Od
z$}Di{Om(DlN(dML;@FbX?kW{oc{aR$B@3m`>8v4XZHwfE&%=2(^aW2<6Uz>HWa2Be
z+!?#coN}zU<q{3n!>Nk5D!M}^@BF&^%g@V>rQK=ox#UULRI?;v*>c5go%m*;7mlcr
zdB!&u*;?QL*YaB_vJ{N;@*AvWQdl`r19Wg>6j{HedV{-VA-r?6@I6PCC_UU*U{te?
z7-$eb#Dm#7JPiUSylCC%*ri}uT3fy^YX^0rRogpi536eWM*D%uDwcssdtVkN|4*92
zk^JdhmUZMTEH9N(-4F!M6KBi0IkhzXjrxMs;00kCC*?``louw*a?r`&eUXcEz1wUM
zcP}M-5ph!TnAo2k?0F7H8!x?!M|nS;pSj;x29wa~$VoYW`MEXXvM2`p@`0>TJD9w1
z?9RBl)v?9ikDZF?GcXrG*e5g?T^Rj^KhBq}0b-SvQL~qjvM42-^WP5pUXuY>zM4~2
zb_xyTN>!4esmu_O*U}~=-53>FRa5>NC$aM75~JMdC#VWhl8L8A4elC&|C^0y2zD<I
zrrnHVu)%r2o7GW)X9+pmM|FG7$vrFHGAbRT*&08Uvo*W1R`ZZ`B_#>Wx3o#%VE~|h
z8Gs2C7FD*I&=c90j$qkVb)of<s|-Q)+knBz?E_B^$Io?ETWKh+ZXN)xR_yzkE6!t)
z(F-1OKtB$$X>RS!HzyaB>=ir!mkJ_AA0xD4G_85+?mxm?$%zReTIl+~*;XwTVrUQA
z7QA41Ku@*L!BzkLwFdifbpR^^`z)-sz#J(HPjXZZgIp$Mwy=OC?yBC84{*7T&bj3U
z^&cV}KRHUGA@_dteT=|a3+e-I0os#N>vT+%e`ca*Js)<267w_BqgX`kl~~qUhB=OD
zh<q1e(Q_h5j4cgq0pJ{ALS;*;iWoy>jyBz*fpu7^TKH#gWlOq%2dN@nYt}0T?k;ZL
zJ|C{46N`p{u|^8jRrNOEiR}dCn^nYW8dxD0<*wQ)7|<B@4?_hw=+`dXbAEl3sOpQY
z8c1^h4n4{+`pD7mjU^oZHib+sh?9q*Wc<D~6rkbhm1zme@BgLaUp_|I46U)I8J*J3
zwC8E|woE^MxaY|{DX4!=W75w{lgCbnu_sHKr5r?N51GnvT?VL+PGQeRE*J%}l9-1@
zjD)nkXqrV-{O>ZU%3ww!^K8smbAYg9R1dbhbc3pB6-rZbiX$-?38z=ZgH5fX8^m5`
zi6Y%N&w2v7WFBO=g-9v{ElI)sFIrDd|8#D<(coXu2qDr&F;tCrGmamJQd`wt_R6((
zc3<{0@aBJ%c5as2?-@#E&&p+vnA?^*1(UWCAmMk_)yc8;6P!+9t>NiS$l5;2T<~i1
zyItXuvJQPBu_;Y73Bjqgn4*E<dAr!4260>J8&V1}61u?uB-}(s9;XiiZJFj5Mj9rv
zZg-W#k`+hC@)u@mpf(HIsR*<k$nxc*49Cyck}Es|5wA@<*+8G4-7^Zj*iB6L%nbXa
ztr}<zsA!gg^~8v8v7Kq`Os`^fx<6!Ze39WkRd1Aw1aC%kwvNhN70T6c-I;VQ<mb2X
zxR^!N!vJtcNxN)0jEFVv+r*GRNsfI!dL&PARM4dt>|q5ptobO;0c_|9f;%n9se(yg
zS0fRZBxBZzclB#n-xci-5W?Q$OVu1cr->_BT`9V&JV2SI(pp)B!|_C(*k`zL$_1h;
zT%-$h30i9&it!bp=#d=NP*>d1Lfu1cQuT`w9<=0!J>hO2g?)rPiL~O+pjeGJsaQ-3
zNZF{>(9T3Ipx&6&lSO2cW1)Ez8Zx#AOmlRgP46n!>NMTfJ1>wF#}(E)>08{ky1JpI
zAB?uaSS_+o19W{HY<<!Os;<^XMKzbB*NLl98zt=9{T+gKR-_rEnM4L|%#9UMs?49l
zIzj5*Q?h0YSnF?+nUth3E%k(z8<SQ|XWaENR^C+5Mc!6{{yx<&<zAzKc+E3MHDzvE
z(jZ(qyIoNWKTZZ-x$s;GHEc)j2Qbtp%i|D%WS^u*6wMF);R29)1-xv_&L}T%f+GHL
zF1Mc@A(+l>W&jxB48_QK{9`^>05Efn6iDu7u-+yhR9O57!!QqbwOvmM!MJ;n$&n9&
z5yF9kV&h1Z&|ZfhodZl=CKvSQ6cL=?&OX2E97fZWY;P{|k%$vt<S01F%r!!fJ;{iQ
z($X;cX~7@3Q4?k-S162!g)%@oCO7|kH%EG@F#y6&)aZ#Q>0rG1C>Zr@eCT976>}F;
z2Hz}4XSv^}!s2xLO>SfR?hWSXU}ie>Z&x`kcE&OjF-4}X{Ij}z(#VfGSW{43J&yAX
zh8W2+QJ@PsIUwiIGxm^clgTvE*=5McGKR9;(CY+c2A>(OkfalLuKfIHgSY_2ni<?|
z%&`qRgg+sv?|3?&zV%G&EkL-A(Ii>BpGuZcM@Z#Dm&Qj^ff6tHf<4Un9w)HpigT;$
zI%bxrg%wx8a=qNz2pqN=S#>h2Wj)w=0qwTQUk|6bCem&O1;~sw&g|3(to&sRxG@+x
z0_@-#f>ZKkro@jn#rYbi)*kD@PetZ=N|z{{oVmmH4*pmH)(t9PA{#3^c~;*tz`#X0
zaMekCV>CuY5Ui;^18A5RZDE+<sXR2*9gz?$cxWkctCH1=XcQ@xmC|r*^D1&8^mNZ}
zyF(A!!Fwv&C=^$nR92QY>h7F2YFmH!pgd<(xlx_2g+2D>A-sU7d&6m}RfltN1Z-Y0
zvA|%unOKmG{$b6Gl0h5nPSXzv&it`GXK)-&oThGvp;eZvops5xU4PNZBp(cKEvo7i
z>$|aWQH-GlsM;MthX|f!w-2<f>D=Xnq0I<R$9X9X>UGQ><a`&mhsnST+gz9IH0-dU
zx652&YCLH-e494H=lSM%Pdt|!l>z0J=_F?QGxc-tAI6p}xZZLjNa!}vMQOTd-Db@+
zma4R&L!=-dl~-?TF`e7$&yUU!QwgM```)n8=ojzLqboTG<CY33F8WeCuM)jK9L3MP
zH>~8{nLELsWGhmw*D^hlm3CXqOr2PmXt*}yB(3K;#~D#MG<fl{k@p7q)~3l{NGX0P
zj+d09bPffJPw0|QyFw48J4@#zQN`Dl)5kpRKAtRDNwA(y{zW#rZq)-Dkg?BJ^s)1+
zYW9on@oG74>iE<3Tof$Z*&&MiHz)avoUQ!jYVLY2$NNJ{W-Wsvdk~&br83Cm{U{r4
z40DdY%OgA>>JKRT-Dml?=zCV$$}rf4o9LFq8&w~x;0bz}zg4%SramL79WGZLGY{xK
zF-uVrbNlBq>Hx1jHN7_puV1Y{!18}QV*VG3{q@BE|0$M%gPr-mQ|z#ol-(gKOwWtj
zE4*+@IUB;Ee7;V8D~%(Sr#NE<&r{vU4?Jphm`IEwGhz-=Mi>vHL=wJUVwca(BH2RT
zI(M;Od-rEfY)|n@Yu1b&UTdCTV^$*3czMg{MJYd&m3}8MaTUzQ!O@JD1tIt6)2kO&
zC6dZs6BGL~l#MnPrFN4?zB4*yb^ppB?vS&f5GUotZ_|V!!PK(O=Ovv<HW?D7LALC^
zDco6Xae(xy_Mf0fPnl6dr!Xtd8#8H}M7(^&Bn8bB-ov9kS&lHbiW23D+&hctKxZm|
zmwU>D$sOvpFfN~+7|Gs}MpoRv-%X(Aia4IoqA;F~3RA@_M60uv<>=0$9FeC3{FtW6
zIS}*RX$@t`gDO~<ri(%Hkb=SQ3%u7WVo*FG%Q4G>EKSWDIZ_bP>=WpgFZn8fXqni#
z3oq$}{-sP3)-i@Rz2#{zK*+aMbgvJkpZD2&0mRJGpjZjyI*uW`A`k^cICMSNAlFS`
z#(Pl|1qAXu{x|U*MUU+<zFPUrKmyH0=;O72^llK27&K0Ki*Rd2SuRiQ)GE?)!S0{s
zG+f}oQm%>^!$8hw8)0MqmCt!2K9XdQ2VNe`^3uuF5@AK%s2+4GZQ$#Xk^6ajjsClV
z)Ept*spo|Xj;S{<YQwY(0qXs7->TVIdAYSkRDkMx;EcP1^HZ+d4ki_@R1_WBh2oTt
zSYkal310QnM~e873fK|FyAG`p*NtQF>z%mwkwgUOC5uD|URvU!l0+U*C2b)pw-@>$
zcK+Lulv7>L7AzFn1fFPtW=kMm02J-r^wu^Wl=F9!u9EW)%rsjf*^G1qgi8?pVymOo
zPWRP(5|fgyI;#H)G1JLy2#k!0`<Mdgf;odYBKTA=Br|@2pFWPz;S$1uKi~>jFih#c
zR<A;CqEfd?NjrKHZ-T?lu-7}T+tMLqK}^yW8L`TFs=reF?FJ3l!LT&fp`Zd#g@Q}_
z2$I9ACY+NY;Ih(VAuKDX+-t<BDfOff{Ap|qcy259!m?ILJ7*7wqOWyC(elSpUvtpZ
zXkzad^*{bJC6jocDXIfye+>1TkiJr$nW}5U{7jt3lu;<sTwJu27@-Tin_R8W;W~^_
zd0oP=(VOe2&6t&bYfr@H)kjdp`~PXsvj?6<bHEn+yaKlZ*WwlMFlfJ<d@U-j4&x?*
zM)=}Jq}T)OxsYmGg{?bYvT1tOb(d1K1J$tX+^Qalu@f2ACm+p0LYyZb?jS+$1&mu#
z?lvz2p%RawTCuFf3@&O}fnAi^{}2T7iy|`f`2#pRHmDN^av-_c47$lhY&rk-ftR19
z-XHTrnv)X+D=7$i*fZPnj=l%b{}z3kNotN!HqRBJ@Mv{2i+lc6kYZr9_kXZPzKoX+
zl-a-&rW3o<{u=76pTz4nPfyVjYfl+)_oE$$gr}R4XYYosxV@MIH3D-=5U`mJN8y}N
z!QSh~#G%-G2n606M8bg(8QJAObyR*~Jf!RZr1)I&x$=HZ-txqgOe`2+Gk}U#-6IBh
z69R7TYpM{#KP{?EGxh|DU3<|?c#?uMSub)0wQGI9foIx=zSwFv2)*5fVM@OB_&pdv
zG6fSEr4u@948nkpM;MsOKWlZ%=o#)M#F@3Rc3bVQdqJyYK9ZPdxd#zlx2dvaLF%~a
z)5O?sXP@^agBJnA2g^C>kB1EgP>S$^<B{+nz>zOC1N%Y@baN{z>g@F&4<rYhO0IKe
zLUAVj=$-GwL+aJn28!FfQ8xVeQED7qHm{s`GVk@D&HdgGp3G~VJ9ELm)45<;Gf6yu
zM_oz0l$`DiSyEbOU|pG;lMEXjL;8R!0aOD>DkU76a7L+~85Y^T1rohEH?5fbp_NX$
zI(uY*@EV6iOg)b3_RQt&)3I$_wir~b-OV-$cl!|;$=9wniPyRz8NmbwTSW;8(jna&
zsWfsDFqnibI4aZ-u1dYvU7(m2$}sogUcn+Bh=)7Wo*9jK4kRr)n$g5qQ9h@;O$=do
zW#OxJkKg2L$H612kVGp+TODlSWfev+K#L7RGW`@jwFZ%>!O|ES5qE{lqf>owHuc0f
z+eRn6XPbhiIQlEkra!Xmo9z}R;p!Eq>AI)n+hn#=i3{^z2t`}o`bN92crkk2nVy)+
zr2(o9)*O}Cb57YmOSkH5Dm5tHEH6se7ra_j0FrJc`L<e&Z6Y>23;gLOD?*JmQp4F+
zTtRzD)l%&Txp3Y%0d|d)CUwT9CXV^p6O;PanWYrKrcU4)^h>1xN90*q_Ih?4Sxm?s
z+|$w`KKo9lg6V4=bH%*EZYh`bN7vfWY*Gl2nL<V~Uqkz=PX++v=3|FbH>C{+tRkTn
zIUThrjC(A-px)Z`SGz$oPPdl6e*lGAXnj~pI*)gU;lWL+x!kAVgW%k0s;c8FT5roZ
zJ(|H13f)kuHIyS}?B&U<;B+}p+BaV<bkj^+9ZPXcAF>FlF21}JYnw;&#sy~;_i5#d
z5l1d;buIt}U5AHYGlT`?tW8L|{YKd%)MzfM?Ol6V8m1R8-K;0p)W~bshMlN3__CDq
z`i4C!ePAE%m!-FCsJQ(U&%WZCWbvNKt2v6BI?Wux$9Q6X(2I^{YWco^@i9{B{$C1U
zrT<STfRTlX{=ZYel;%Y2B`e(4jM^*sdV+8_46`Q#KN3MF%(X3jlvZnpDNEGh4?1x|
zq41bvb7kd8K=J!}gS4VIo`UYT`eWg|pC=^QS8Xc(kEB~~=?>q^PfeD~TW!$i%Eggg
zuPoS)xA8H)ly>#YMAY*c(d#g0_n%&x^+8Lv*3Y+|&D0(IJ)IKO%CQ4}r(v7Srxm*i
zxn6U1XHPjB&o@`C=&>{jq*l%>(HcKL6{tW(S*oL@4ZfbQme0>C#P=)`yDcmioPS(p
z$t-yUTT#~r`BpOTI&RKt_;c&!osK-HtQ9I^Oi)(2O9IZws9E^Ged$i$SHqZTVxJ3G
zfSrM@6H#l+Lv@s(1C3VV-)D&{F3F9U-I9yQ;Ca5&r6R)n=B&DCtDhfT3QZD)pcv|W
zHYhfq{7A3{C!y+u;d~TMY>q20)fznX*f$SxDF?O1EuTUx73Z+>g14DIoLf7|^L*la
zdmeHIYuR$NdL|oK@NFm&6@*)?EQdv&YIVmGxlZ)M3b56w&I!IZ$<gxA3BDfGKT+4N
zRAUuwhPwURpfLxPU7SwU>vD~QWMU0C`?7mmNmQ)MNx&aAum~C}rSo*QVZ^w8cA3$%
zAF$k1(B$C|l@S3-3a^ZA=ck1X#5BOUaI5-i7n>CC=wau_D-ga|6{n7w(vRZh$I5yZ
z+W%PQR8*?#rB6hQlIu-QO)P~yoMDfXYH7e{N<ynh@jgiuKL1EvFoVr~(J_k6y#J88
z<mYDmcr~(|e`!)b>k55VCPuQ>+4ha=12(|Mukj9+GYo#VI>v_9t()gbaBkqVlU>5x
znm9L?3Kw<j=mZ9ZIr9NdpXotWV#*a=RuCBp=WQa|rgQ7{eW)d9(r@3JQ$U8#D`Cxf
z5PWRYvJh!utda$hx5U4VVrtPdwEzsO+;?Su5SdVTW}57#-;#>@NI)|a^U3^c+C=o}
znP_7CUE|xq)`(+ob(%!-yP(Q0wkXAkQk0fnrZC0vK5U)A6_x>exFkwG!R&oFPA{|#
zD4OnuC98#s428g)&W@~*V@KCMeqvHuM+4L`K-s+V-FGr^GKaL(hvGC6n67LELwdEl
zdrYLTYKx4JhPLOq2apt0Vuo^oUIzP2{Sa<xcZpIlq=3rD>dxO2CKP4k(L#NSI--2*
z-ISE7rgIp=bh-xcO|YqT3s=|iC~(JPL8Mz$FXiV;VByIQKNHUAblx6JLynVpnUnTw
z3Q>kF&W6SjxG<L!UkiApuq+j}Ej5*My4_ed3QBEGFtS*gf?<>t|HwL)%T5Q)y1}MN
zowtk#88-G=SGeYHl%1)<zzY}`L}+rpC{{%m_H3Up@t~mIo1-yuZ_bC5wi*Zzvb^w4
zxG_ZOi2@20N_t9`Yg(5hW-c30gKCkR7oah;;Xhkz)`b@X-*-}HTl9Gs!J9~2+7hd;
zb()d0EzLx1ip2LR$_D=huOo!s>8T(I(yn?c%DZ7rbsc``22QT0Z+klH&U-~%-hU0F
z2tTvZ=}JIi@Tu1TN|dH>5dGN?z+{S6L(ASQGy4s(zE#_rTLqkHEob34Pu#2YE+TDq
zPS{8k^GguPlK&FZ6>Fy8*{6*0r8DlFSZ5;UB(u;FLrlk@kpf{x642=bYldH<#}=N7
z$ja1K4^{2JK`7Sly*G<-QqqZr6Kan~(C!kBqIIE4ZG<_7XSJn=+$_x#K53OS&)c;{
zw$YMN=_BU3v7}(@$5TSn4xt0<@IcH@m4^$*oGl9^2Hh9|XoBMy8M~8lRNZB~>OmH(
z5z3W7P6-pdYVLAveR2jg%|XRfPr$Opv1*2$Tg}*O%Uwe7i7x&Ym`mz-I*<)dlel5{
zQ1XoE3*o?@nQo}&)RkA!pC3RmjS>G9RbQE3j(AFIwigSWl^l*@`}6bg)T*<DDKBu-
zq?D;R1ie4Us?r^OHIKJ)>$~A=mm5aHs+S;E2~~c2NwluK^sxEKefc04_gPG?%4J&<
z#LpZh1S@rNY>X@Hy7%X@XDoVi2tJfYz!TKzX@)MKf($y{td48lakn2^PeY7~eOq&t
zbmYdQB@;|irFUqS!X*qM{Lnf~*ztY-*@#WVgHBKcsQD=R0x;*D(QMV8n86gZWva^?
z{>eacp;5}G+6M}BJS%Fk#_r6*K2K1k8TFEdCPy4c2Py;lm;#AG4AWA#i8Asl_FQ)A
zQ1M};Jt<Qx8)s2p{os;`L9|w_MEBxfOvqmo?a6A5Z^gbnSx6IXfdNoJG}^?$^x6*G
z35(%6sd==bCrRNvF=K9XpT#-x<}FW~wlp_Z)rls4H`s&Q@{d)n@y@UUbv!1;p-#>D
zei(v4yeJ*a6Esx>G{bCCo~~$V?(F5;vtTvd-f*W#sszGR*gnsM&Jj;@m$gSIDwTX5
zYSm>Gx<%`*BrRh;Q(-KQ6Fuqb#=*AkK+p8Pg_q0cNW^QaN4&w*mkt12kv)mH)C|G*
zdX^l2%!S!M1}cO(&5phE{_F&SyQVQuU5%4o9^}xCo;CE9#C)U!xSrI+-b;-7MOqIV
zkn3%fOrVYB(_Qs0S=c8@XyXujLLz3Fd%?9h{xLpZRWgkrFAQtFGcUfd9W9n%lGj<Q
z2Lk=OcmAE7%hje!{TSfPkLoz|N>m8H-xQEl7j={uQmtkEKRJ;5iy$J7^NA<CDq-rA
z-oLANdLd^s%lE|zmg!1d-5ttDN`kXjc|OA=fzdnYgGDQC8fWv%oPuPC3`|2+!|@+d
zkMDhjNFID!tnvphbdpy+)Hib_?G=EaM+m|TJ!(tyxqMAiyY|*`XZ98{{kZ>Dj2^<$
zA^zx062U${v{chMpU)rIeiX{$dm=*fK1Oa_^n96537~}=bw**p=L8W8JBstLDOGbP
zFe82HXe#j+vO?IN>~V8$Wf4Suz3I+mQ|faN!xef>ET#7|x;TodPvL>g#o?iBZrPaD
z9jIOe9c3?6XSv7?KqQ$Wy5V!C4VW4LO|C;K?a1MD%6z>&eivU-1EK2BV9Jww3ODls
zeFQ=_T#Z#HeQ!1~p~pZtd9b3VCk3dk%UD`7*R!ZxS?!_zmJIcGB*p~l81Y-or47ZO
zj-SBzh94B(Rf1xn;!7pBkpfy*?}TM2ol*KS)CzfJ3B~swTF&az9km;aw*@8{h+rO^
z=uQ6GeP@@Qsmk*);E;`CY8tDEUdVGi*rhnrO(ej@EQT|Rs43H#K$&}km^2cj2S#V$
zhRAX)V4vMkz8kdYBwbeUU3G9e(H;f}gl$(S>0)=KTX7&+G@6rGmnSUu-8VC^*L7=7
z)_u|qM}WV1x2E_Y&jYP40<E5P26Ae#`P1S|Wz?&y^O9}FNPK@LdI42+Qa%0`PGtIz
zIFXr|@xOCoHCAUVaXZ6{+9Nzn9CZ(%0T3`R@HAZRq&>Z6HSHZ3`qoWn<nVq`R`uK^
ztSfk8$CzPv-sbn}Wa(Yr8quy_x6emF?{|b>Z=Ub>m(9nKS|et4!?dYar`-s}$=9=B
zocH#&?N7ToZsPKDtl#asp-><BwuMbkx2B98i_glsjs@uV^b>MkSPL+94L|$#F0v53
zeLHmRkI%=-b6hJXapxdJzcn2zrPUV@G5Fa)8S>>l@O5;c)O{ubNia1nbZcx?XSZM%
z{d@sHk>2+1289>2fH#K0$3(5izzl~;x5y=Cv$(uKK77@yF6QnZU&LOoPiWM0VVB8q
ze=B|q2Z4`~XyW8k1VI#uRYJ}1De!r&THCJ1+K5*YPEoXLV($IFeF=8<xc|mg-sugI
zo+@9Stcf@`u|9MGs^-TEVl|FL@`)2!`w4`IbY8{F?=~8Gs+whR^W=vc)Z~x4gPbou
zW(2TNI~~JR#`iempiZ4Q7XKB{<4Be}GvWmTt!LW|=LwHyaYA5V!YwWS@%z(9<g&}R
zPC@RvFLI2_&cN44aYrOJ1>`&>aoSPr80niv40o=1Hx(#vAT{l!ar;qb+#d}W`x*H`
zHcn%IKfpXa?Q|1<*XcdZxmVGxx~f?$KkS~8VjG5?H###%5y`y9w0Fbb8Ts<}5>-;<
z`k9s+^)qsX0)yIIJwik}tg7?3E>5mKMYR;Lxe)!|VY=he@jk)Uw1DcX2&C;Ky;lxu
z(OWv09z8*oF!mcw0@b8hb1GX1B%8yg%F^}0HfH4_Y%hm5BGR2XcL>O!b2xHj>l&mA
z4)5`PMoIEfYX*n$-t-&KIRS6PbILzw+_+*?*KjWBU~W|Lt~7}_&<e!vp0RaPxehCj
z(TLrsiSu{L*Mt;w)#&L$1HJ{LyryzlKyX5JcT5AmFvd=9Xv<0T_2ahQmP~L{YT4+j
z{xu@n1s>8-QD#|ORbSmG7iB&@P$D~_rS>}+hdAUW!0+F_6J)QHhmQNEHb(n;%Ul-v
zk<<nU(JVNk!@_?Rc)Xwqq2`c8u#t75iGb$DL%;@i3#CW+nrM+@hlJM;wGwjUNj@Wl
zlJumSCMFZ`as)hC0++WyU!b#iB#2GM;}`FDD-u%bFZ9}|_%H`OwwmuoUs@T<#nE^9
z(ZSAt>Al@ZTDVb1+tfI8;ad`|;73I0Hfde&OA`oqcQg9b6*%tpdrE!a?&F{D@ahTL
zYMY|25E_1<(<7APm-b^ZLbKGVMM8@2Xg1WtXxd^SwdL+TpZQ46Qk4$=1QN)R3H@&V
zh#cA;dx*)M9DoaBB0~#%zB(8{%8rbX+$u)}LU=kfz8~Cb#@V{UyHE`GEN_k!n%f#f
zF=IF0-`6L|lq!n|(S{c0OUdD^DA!s$OIT_6Q=uP3C0=2Ttg8LTc=RG$cEV9<7^lTR
zM=lkcbe{!duz!0!F!DBCl4XiIEAuHu?{Ru%b<%5DOr}MC8yUwD&`5*Pbfq{DHSs#*
zslr~jzVq))>=8m@AO=_}@n#@tu2;y-BhB0cX%rz<6^NF|PjoB`tc-csW4O!Zy98{G
zTU>Xxp^{OxE(WG*>Izbsv+80~O4y<`{Ed8KhmmND+DkmD`-(^JWv2W~1+@9R0dF%Y
z+48kZI)3xEf3uXEi>e?3@?gmYLt|i|`sCqCr4b~M7!f$vgs*Cdx0T0QAUa(+`f!bn
z{Pxgv0y+RvvFR14JZmCb-QVhfkfhu3^gh|bV}<gvLR~-W@iqNcJ~n#yOfYS8h{I=K
zc=7Q7E4lC1FRzanREHMqB$Xy-kRe0h=A%Kp3IFjCLSEX48-`X(<@G+rNr9tP8Aw~`
znwr<+Zb$J0XmwP><Uwh!%VlbH_qSLLJORg{31xT8p1?E`4w-cJqNcbZX|c)ZuQ^IK
z8V9BWK|7RSZl|O;2}HuLb`y#m3CED}d;nkTa)du^tN5^GLwCt7s5xN_s@bWijlNIs
zj=5n3$dR4B@DrMu$|bUllUulO@l!_sZZR00`jpVUcKOFoVGQ>9eVj+k)+|{l4umsB
zYf%{e9sAu3GuLW$GKgs6hdnT9Uai6~RO79cck@4noFlV7EH@{?%e&({!;PVh;LiH_
z#(2xr<cTd^UvW;BoxxB;f&Fr&AN@_EC1>728aZc+yc{gB(JaN`OG5I$(#(+}CmO{v
zhf#S;u4w*MvtQGzeu0$jE@T^y5@L$|C9XyNcep<r$H_BeBS~^sve-@Q#2zCo9gN(c
zye#(jbATfXmlMaYF1)deM=qLuY{N&9qIjk}x*t}rfB^&1<vuUftRpY+-s0-VLNQff
zzR+FT*(f|-42OspFLV766>I)%QQ0*4w?WV}nRTf5H>7vMHCf{6yF?~yx}(}K!kw9Z
ze-Rg^Fha8@86-Ad#Y)R$n)k|xH_rt1;6AkU8`?yH{TDmCzB-!@j_Y+iz|6~F8o@`p
zBP>)YPx8V}tlz)h{?imx=;m|Tj$#S>=z=`LY6^qA_o3e1eUHF2MvvXIB6-!FM%9VN
zUgDwBP_5yWrka3_l{-69O$GU&i3w`IF@fs|dNe2evI<WK*^HFC;rfDkBb(vF7rW->
zgo8Ji+LFoBw6sYO?#e*mqKj5jqVl5WD!1A3q-#pUk2ZgB=WZ3=ET3X2By}9<iqq@b
ze1)6(0Q9*CP`wpV*0Osow6cgBdB7@tRdTMrcd|X2cw^5qy)Zq#y*!r$_jm=Nb!kYt
zu?{I9Yh>TD>pDGydkl@?8(M$7DWBYGRNE|_`shq0Tt+(iKt;jHdmxV+OuA{h-L5-8
zv3u$AE9kjOh3%;>cx-6!PH{M6y${z+`K>a*!C{H^H1`L}rYUG&b+_dv^_R1UxZ8*N
zjK_^V@V2#l5DBWBS)aI;80=0;Frh}FB6~45Rw*{z3-xnaI5N(-{&PPP!)@?{J#xZj
zw5oHB!gu*cZ^LA|Lg<yk#np|OI_ZoK0=Lz`>`9kfW$5C2hG*7O%LKX&G9%LWHJmE>
z3JubMi~6#j^pZ0bX0STYKgI)BM#W}y*W4Yg#%SmSB4nC4+6l4{^Oq!zYo8Xaxsu=|
z$M$frk2w#MTzjQH$}OCWD~N(zX`29cj5zs$nsKx~Z#*nL;FI&laYXl60sl8{(&=~M
z7V+K<oD~$#G;xJ-$F8DDCn~U635g!8tv*{eWn<bZ)r8M{Od~K-NXV}|EkQMx8~A!<
z_XDmgdqMZV@Z5i-{xPz#v;23S8~*jM5w|0De^5OF5AXE=K;pu<gWXPvZo%M7>X7Dn
z*ZutHGX@HPr6nudz(?B_6id_Ak46#D{dAN^P5vyBo_^}-`hGI~ywmu7JG{Td-K6RH
zs-Tw$JId;9-yBuCT;G=(`zF)g<oR(GKi*|JzfnD$C`o4Vy_1!^C-{+5#!UC#tgsz?
z1;`nIa6^K}(in{^<^6EFur$=LO)tIlRJ=QUeSGB6`FY}xv7`xRNmnFhsdaT_P((ZA
zY?|Z6TLlE)bK6;6N;7h6NQ^vL`<U7?&j(M3ckacqKU6?J)EOl+zA_fgrmXob)Y}$b
zpwL+eYzF#rM1g?ljqE2L>tcq#u58+)F<21Wu#6=!h&@b<F18oh-|uwAHrEyPiu%N3
zn7bHuUe-<16>ZMtX(CXbDyJJa#9;1;n-Lehxn<NC=6o;SE*iRz^UUtE+aV%NmN=CD
zDTja(l_;HQp<5GOBW2RqVOAN4<g1n}7E{ykCK^^H_R(E{$QN>}XhRV%(Y>KgArWy3
zNCc=zEGOSE6wnFg3T(-QgaNUsjEMV~pY21NB>+qoE&tr5U}XJN8DTN%R68IgGT$()
z8yAG(U&G?Lky#%gKR!)0sBDe~o1kn+(o1vc%Mv1;^>Zzv0t>~C$5iGZC5lN>UgZxd
z3W$&PA#VH=5)p)LV?qijz%`7T7$|Z-PG&47I)mnSK5gk@X6rD7kvLr4pzg~Axel~1
zuGIiKd|y&i*6GI&Vs6z!|7gzHj-p8!lB{fOR)I_21CjyF*^(t45|X~!fZ;S{2F5&<
zR>W|^<HkuWXiy&xqJ_?usvh3Yw%JuBmGw^nV~7nQ;V?O2)|HaV1@7F?rm+Q}{UwKw
zbS;6}8P@vI)}DAO;-(TEwemnfdCo>El46Lb9jAu|Zr+4JooG&F-W_f{T1tt9{ZdoC
z!l9kwvf8Ynv`NQ4>Zv`7`p1GYx~2HM7AxHbmOPY(AwSj}{y6D2>H;TkRn%MV?zx31
ziH0nadvmweB`j@#2hIoOx}OR;6yDSGDi?=xIpuAy$d=MflrzHxmQP~dik9PW;@Z)#
zX(KvUj0orJXr&MEG%LLETdmL0<$gE+=p4mb2OS5YT4n$%kGLNW9rXc*=iBDA{;iV1
z$8*aIR6u+XSu=Z<Hg`5~jrEVbzadnQ!lgzLqD2sjghnYiz`ShU7gQvEvEoTj;O<AR
z4{CU7-YL!VvQ0tgX_Xa~AQZ883Bp9ggJ-mFABmu*jCcS)K{p|U8KGRlAgjJscvE6G
z;f%VNedNEJa8!t_bV$mRv7Tvd(A`W+pw%4Yi`uEF0g!*3GpbJ0x_~P#huNEz=Bpq=
zXrttp^M9+k1Z|4cZP@_WbbJTyna-=NAjD!yi1It*a`Pz{pRvYL*z{f*LYVc8wk3Q_
z@pc6qQ+O+!^eqpq{}L<WiYp79=LKmV1Rid#l%6+ntAKM5Mzz&Bj%6vSq^q9B*ge<m
zH0)>9&C6OSsEp1rFgiUo0sI}ijn=hylg4oyr9^4&5wuAAd}@6^+Sz~$au%$ZM~o0i
zN6@`wa26!0HTI|ex@-GW-R0|3i<$`SFRhmec@#n>;6Q|6D3Bm`@1zqLoana90w!0j
zK=jWl#XAF4y~P%7BU!lI3ue0wy-6XRpiDs@e;uQsB(rP1J+pH&zB9NbRWYr5$MuZ{
z3fr^7#rK)-_#o<~x$vFx-~L@mIUX-5)~m<NToz@lVz%F5?emDhXO18M{%2E|<G}j@
z-Z)vjOPZ>rFt%*=I;%C3bsAB8-#tX2x1jvg5_Y&X7{o{y11M?CzXS0(S(|>@?{n|Q
z^xRtglT>m(@D5ghNy-GF5&sNn8ak-e#8LBYfoFRi#1WW*Li>&d&^7kU*?tJ?!%~7)
z|0jytR&K6nk^E9H0%0R{EpxPbwE@pB;UQV~GA!~*ARur%0-`~y8sTA5J1niCUOSHC
z6j)P)X{o;@f9exf_NlQHmyhObC<ybhInj4u$NtS#zegWe-%co0LIHivYuPYq_M1Rj
z0Zz5<QC^U|A~xGR76&1ZpOQN$SQPkJMq(KPA+Uuq(9WU_h$31|e^)kBIyEF}GIFL1
z{haDFnmx72%ivzDQ@ewYVxF@OP@ne(5|vMBn4Jensx&$j=)_c!qch5kZB%%cQRgh^
zIPORq_FVCp!=B|O!pE{Og%8$qfmca)tG;ICIn`&Lra%}hvu3yl3>x#DswYjj{PW7C
zex;wBEA=v^!3!-m`OAg0S#`m$<62n9XY&r{%NNzm>TcEjWT?L_;mIfC&I?(#=|txy
z0IU0*=p5uF(b4aCmRq-j9Ul(wprI;fvAwW%j}`iPT0F5N`PV-uSl8n%F=i2!Avgy^
z8inyme}#x;R;*h1u)mr%3s?}vK!PtA4$yh(8z~g-cyUW=&jcNKk(cxQMDudcLW!0}
zCmuBzZacSpfmh6AyUFAaKra|V;iA0bye3Fw)@tfLP~K%C#Sl10D(n>sH3u`pNh#dc
zws|QsWBS(gfug!fJ5GA=U2^h^L+3capxq_oA<^WK<+3<pNfz#$PT(!s;1na{N?OS!
zD({lRaoA28Gea7)b#`(_5K5GUi7)S`+8Y{tzUKI3*`LuH*)$lJS!BIj6mrV*Bz%`h
zWj(GF2o5+eRU#A0PA{98)P0AnR>xb`o)-e9glK|Z1QfDBR}6OOxe;7~HS3?!Y~;D*
z#4TvE{TE|9lTW{Yc)d9Y8N||`xUj+%ik%nPuzb7T_1;=8$vZxS_KHXiD?2~LgBlPl
zT#$6Z*P7&(9U?$!y3&7Cq%Y8X7o`cxin)fY@O7(eeO6gMz13xFUn^)#-*2Hb{S|`I
zqOr07Yd1>)In+@bNj+Q{S>mnJd_!2kdmH*}Q>?8Hcfk+u^zhP!3q0T9I@s_9h+His
z;Dk3+OV@(120w96IqDXI8oh*7C}1JlT{kPew169w4?p1BS2NLRfge3UvhzFYo>z}#
zc+DCZFK5itaF?dU(zCx;?IH=jp*Oz0jsSmW{Bvzk&J~7iU_jijtdT6#2wz|wV@MF_
z3dnp;v)I-EbfY>4b8dhZB?e_-Zn&IK2Iy($lS+M{A*2H?!()D))rIxEwrFA77*_Vg
zk^@1;xL3#VL23(EI5H2%b4-Pn(B?$roXtE*$x6SQBl1K<vZKInAG+2e@2+sDMr&r>
zUiWCqRUgW(ZQgmEm{|Vl@}KSWlF*}+I&$!?ZSU^otdcUOm#{HxdTfqjc6@>>+<~hE
z0FzePj!yW#?}n+rkLVa%gevK9%Yz?;5f{~GuDVpo8RcT`bA20%rBwIubSD}B>zu5)
znib10zjP?sSXq;Lt7rCKeKIIClF(iiH3idozAF_#$0a`(M4|b@RI1QAyYy=AK8x#|
z!F2oO9xMRhAKZ+2+>9@62(E@SWMrJ4Oy3td`Z8~ez36V*2bkA!`MTBP_ElmcnwpXW
zvfW%Qd4@J+^JPCMYTH)AC+JqR>(I??JSq09kFQoYSeV=f;z>Y8kiokl??x<WJ2QQ1
z4Ub~$v_+n?F^*_Wn2prg>_e4l5;_ke!AO(QRV}KA9+hoaj=LwQC#kjCZ_vB-D$K?W
zc!-|9y7(n`<^sK&@Y)4GZggCkqEOM`Wh<Uw$%Rpi)&|wJ;FfhZl78QZ{Qz+F`}{9a
z54QhI9%uTW^7zuPN$wYudrs6I@ezd3cKr8*{HK80W%jk=gGe>J#_&vO-#-3=kCAC5
z(`e*AVKHI3G)kz}<LUVrTczGaX)u2?NP&D-rVZY>zfk%5a(#SnEZnd3ehizZxUfzv
zSVV{hJWaqwJKB11et!sfIeC80Eb!Hp^>e7y^()BLYDcarglKb~=Qp?b@4VjCz+jcK
z$sWb0`NmJL3wds8T;V@DxqEC1NK>d}IhIV-H}9o{eCEIOeGxVD@Lj{iEIlrQSTc3=
z2Kg7QKk(Dw)xxmwuo2Q#*)jgHXw~O4$@x+S?rI5lZ-D^7xs-0_<i*c=R5{RBFyX7X
zgkKLM#?<(@#B(#$z9Jr%BAJac@b%E_q5B<fr2QDE?=3YrshS+5?x1IN-`BHzy1lL3
z+n{MPPI93tBE7BUHgR2%k7E@DN)%S5P0N;L$mB?3RO;}tQfds7wXGMQH~@a$%1d*u
zFnWS+!)vY!y$&`luALR0HO7c6Ot8wxayaj%ur{BpXC}Jvt(3~1CD$~ceQAzF$cUt0
zg%I9@gjfj<&z=EVPLof*As_KfTR13S$p!*lQ;6?EAl?(_<H%T$c8ilbW;7Q^NHLCx
zUTbLzp-C5jl&tB;X(FlP9%9D$D_zMrULZS}S9@8wj4m6_{R#FcWMoNDR}2qs?=Bw~
zT#&|4Mx7o}@_TnzO4aJO$1)BFM_|CIR*he1*1y;RZu!nx8p$F|w;Md{sl$lsH#$gE
zzCV-GQagtHKy)G1p&-pCUWVz5j6%nQQAsMkv(V0>T}pDPlE4sH8U?fD=>kGxc+Q^I
z144Sh2anhTB6+Xk2CsLmm-WjIGFcRdgwdgZ{ub&_REX?zlxsw78gIVgFjs{^zWv<K
z&j_4P#$lzzo;(RW4vj5?#O+09>8C_yDl+|EZ2s0>e8jaxYx2$tIoO<}bR*LM<wS4}
zttx!BS`az4He#wUkQC_v;FTo)>&GX!E$(pQl~oz;-Udu~v*;KZuwHQr(URb{p**pm
z%MTz};*N?~5<;mTrn49{8(Xly8fU^j_oaS*@{Dp#RZ{FFuo{_6MNJrUMW0UckR)rl
z$QWrR9zBK=Z%>UHqTr-6f#cy7f$HM%Y<8xTptroL+Gr+^vS|}S)GUD|J`q;W-de_^
zFJ%Aaue{6-ZskZC%Dqf$l;Q^JhLKvr>98blGce+Jdc#&VTdUwX(?ug<0;RtEXVc>p
zC?r!d^FX8!SrA@)y=T%8DcrFYayf9win6)Y98G|GW*tSUzvFtJ<=8!}Iox4t`2<x-
zt#yT7R}1$M46+pMU|;32wXud6zUr8>BiQW<A55&OCV)*T>Z+LuPfB8X|1%Q7=dcbJ
zZn({Zo)(pACt9lXmOs;IY=CM9RH1}hJ8#xzdqJ{OF8&D!GCJATJr0UL*JB9)NQPYF
zQX2AIUW!1)^4TGc0&Q}6M%d(EP<|^I&b#4sqm|LchGEzT3;LZ-46-08^$-#&H5_Ls
ziH_83;|i_}j*Lp7u*3yhg{#SWL>#Uy&L>%~-k5L|OoyHdV^X>~w1tm`&UGiOrJ!8a
z%wTtHP3T_;_z+!Hp>Q-$f&vw^Ms;7X@S1-_O|VaE%})lAYF{0pk^>4(vA}}bvkAP2
z1nma7{Q*HdCRfeP^nQ+6+w9g9IdeKns)fDUs**BN4*RH>?vRKg?p5_p6<=bYU%LsK
z0_{s^E>24IuWg%8!!cSEKZbJ@HW>g`N^a6y-tBo09hrJ&oy4}LxmFhZrIIVDrmDGK
z@S!#b8u|)~lN?Ir(+8dR<WBUoqu;Tx({4*oESqX$(tgH64k>$&uw-b0OCeySG3IU6
z9!1TT(Q(U!*li6rf2Rw#S-9G+FTJo)^d_}cz?FfShR%HjY<9M6S}Q(Cj@MOP%Mc#p
z;jBywa%@U~P|_20!zzPfJnfRCNA!e*qKa4I0c_;i4myl{Wu!#_6-cF3Jc8^O9S9EI
zfdM=;aTnW^FADLoTN7=wD8W_sg?J3&g^(ks0gRRnW-@-*s@z6AyT%z<uNiWU4^Bh?
zY648hy!M2am#EOqCdu~TwqTu`qn4_P@~gBYX5A&7vl#d2RYn$$#v5XXq=gt1sir#C
z9E<W6#^09Y)5;0X)D7Z|7$voqT2+8`fYQs2(&q(}QkE*;TB1V;gwSTgqM9D}28@jP
z8X6fW?bsgO#p4$u`5A2F`oJU&uioZ#<)ANFw1)4$59d`C@Kzprr36{UM5u&Za0e8Z
zU+=d_jqxR{NMj4Gkbd9Ov*ail;+!;J*pv4&dgL*DzMTapFQk-WwMF*5`OL4w9!1Q!
zW*Oyzp+}<=j+#ANM4eyU{8ag_R6d;>_(x*ErpV?PYq^2I%Hu6W0+g6JTooKO=**_^
zY7yIzagGGpQ<Z}I_yV_*6Hm^fLokvSBv}F+;8{sR5~*IzqOI;Ht&Hy=M(XEq10iG5
z^p1=pDxnJcfyd!B)-9ji9a?3l{;+IR8b{W&r`wX>-)5)s-PdjO8F>JCt!Yr4?@3g&
zGoz~P_QsPw&}VB~X9aB1TKXFUS_*57-pMqCe{y=$DOZY$3O9U@%C5N2O|1Gj3&E@y
zUUNc0CSmGHQlIoUavw$Ge52K@QaNvu|MFG8RZ+t#h+z8(ns1Fz_j-qt6%qK5a((}u
zM{X#b$eBnlW0hW5=-!j{(9>REUUkY0aOn9SG&VB#NO5$}`7H|(ObVVw_hi*ke<`lH
zLZYhn3`>a4sQ?l+x#$R*y+$1ZAG&Af8ztDR=KhMe$!^P*>B)j2ierVGUJUV?l7EiL
z@(B2MB(Psqtl73f(TwCSMK@$)mhKq-4#@4A-HG<)@}az~3D8?@j+5$f+4(C(PuLB!
z4%|G)1lY4$c-b*+XmpVuY2mR8lUkKg_BQXeeOUMPdCj8HKR3gkZuKuN!sUAhUaxlN
zu%#Kb&*EG?EY7uhif9GgC_HMZ#2ej^oIr}?AmtQ4dMNwCiKU|?y4~~OpNWOIh=JT4
zaPvxQQ^`XPbEU+|!a=i+%PC$xJ4KE8*<nEEvH%eSzVOw2h0;g^srK}K9sRgqWxf5s
zV@S6DOjc%L{U5UO6qdyA81n8~?GQYfEJzCp;ap#z+uzRUaq6}}MI|!{p6Q1V2D2Z6
zb}YF<@;r2|Ix$h|5<8>*v;I`1<kx!`|6O_X<ac(q^7!WS(=B>&uV;3zSDW^ml*asg
zV50g_Z&~u}&2Cb(qK|U%RN<watL`5;nt3_u_<Y8k6s&}t!q?u8DQ^CGZTbAlbC|Pe
zjQlko%B-%)*hOIcm-BG@+evl&<I{tK!~RWz(uBaLEUxh^38K~R(b`|F1QL~qFQCBX
zJ7#%6h4M?XE|A|<)>UsF+??0XzIK0&pqCLQklZQb&(Hmi-H)d4+cJfGb$p1ixcwAy
zlt5Mn4nfVI1`I>eb?skcy-1pA%ootU?6+nH%g-f(SuQogz+|<-SXbFeOjA6mrC^G&
zMnZUCjH}R6sMQg;UEx}~%W`a;E#iUK67+fJDR-P1)CKK!LP&5Ulo>t>tiMJ~>%`c=
zc?<<8i0^={M{yDO-{>{k`Bsl)IHL%d6SZyzOaweLx*u&>-cpJ>+@P;SEv6~l+w)}&
zYisSL6lsJ}uf?47_Nl76ki1j5EJ?{gK!;eF=M+rv2(nYDj&+)MWkhJtfg{Z4q9@B7
zE!nx+tUET8ueI>F@=2H!R!fz>z0KFOscb8G7SSB4#wt|j2O2Ip-GYgC97FT&_TvO}
z2#kDplr3rKNvVJoq1h4pN2^Y(L6owP2HrR`%?6tJT<Uii^p?;?s6=8C+b+jw%eiza
zld^%x>QXgG<MJjx>y-3Q>3Vtnip)88(BW<YZH?+8He0~PC_^#%N!tD(shBp5&IWf{
zP{p!c<RHmEvXEIt3QN?N!AW9MxN9$$nX<K|63`mT96qf@4MqAUCIW0m{vs)2lU^=O
z(#Y4H@ivIV^dw&EX1H_y&QgXbo{mUe@{$uf+X_(j8>dz%^`*JU`xQY4eRtjB0xqhn
z=0AyKKJ03^jwZ=15nJIFUVgTg1?t*^VHxxGO^wC_N>x1GwV3Tn_%((v0)y6@W@KR&
zAb>||C+vZJAN-E`UkF~q%0g-iN6s2Hynk2D9w>;F^)SnBNXjC%Hj-xupz5fpiTi&u
zrMr%MW6z<(ra&_xa&~8=^PpSd8LCGL3lA&yFr~tzkfG<Gc^8ZhV%~|c*4#QMH1==k
zUckqrRMo_h&;A_#M_oF9ZBMysdLj9#aA3ad3MZA6D;EsHnR0}wnUAlmqf}rCw<6nj
zMWOI;Ht8Pk?3zE*)`NkIc9XJ&SG8XF>%hbb1*Bt=lMaPUk2tO3T(=65I~5f;{z-t*
z+mX%_l9kK)_(_a5Q&&7%8Rn8z@vZ<^U#yW-Z>1>%zbo%{V7O;x4KCM}ekwYNjo!;7
z)xW25ASBc@SCmyk(m;B1-fHrtD~^UW<f-aF(al|>rlmLsya+q5Hi$nJj4Qg&_o!v4
zzNq}hAFxu+{*KT0Z%?s=W?$o=i`iL8UTi6JK~j}xAZ>GEDh`z1(~XTPV!@IIdFWDd
zLd0<F{tJ_2E$p4J!d>Hn6vurY5eq2nJl_rMz7*Dy{G;W=X_-&skSZzxf`-Ff5^kVz
z5?%&JtXr>*KSK&Wp+se$8f<eorGj&hW!V0fkCou?%z@+3g-Tb#8TUDXLRMVbKScz$
z`C0-a7JbQxv*D>$3yntW^ig3ujug8&#p#{5Cy`_c0Z!yv^;uz{cy>*^l0<Lpry4Iz
z!)T9Twf2!>r&T`>7@xO0->?L-DmW#@(xXTmv{)x|kG_k1EE!SZxR>$=*UR3q(SRDt
zX0>$KOz-z`1Tm?!bIi3Egz~pOk$Le~dqa1cOlwnlC~n*54ArUM&IZ>TyZ#@_{wYeg
zV2joTGgsQSZRbkcwr$(CZQHhO+qP|^)_tn8@7b-|u9v^fr~hHbm~HeJ@kQ@gjQWxw
zlGM*qgj_RR#~XPKI|Ef+R|HanroP$YDnfbduP(sT&~G`AQI^v40!V=k#%!D^s%!?`
z=tN@_{?g0A3_ES{w~PyH;+YgjQ}|8!E}K?9LaOZq<!Z6y(;v?0>({WGd?ZU+q_pnG
z>*+bsk<-_Po_KNrXHxRsR}Q%hP#UydhH)>f{J0WYFW1Dl9n|Psusy$F&Eq}g%>+?D
zZihGL@Pl36#@($;SBaBZ7eFw0?TX5-GSMDvUlYY`QwHqCZx=D5kL!>}Nv^d?`N^TJ
zvc?@nXyAs<iV)=)c&qhEF)qGk9C^i7q{1AoQ3a_3X2{9SA4?$L(ZT?&(#91tx80)6
zXu{LO%kcY_A2qG+CI;Bd9duAJ1YUi|n;He>)Foa_K#4@qvrWOx^zzNgGpVQHSc?<+
zb+n8;)ub8udUsf8(Bp<UTa3mTCzm5HFVpPY?Io0ZkgCS5HaD@78V?p#^4OqX_j7{q
z9cUgmEA}a69cXG$QM7drXZq0>{5z`AtgSmWI{sP59;6eRf!m!k>*;fuC<ffNEMBs6
zIn+Uaji(u$oeSQ%K(@_SDPPaR8v@Q9#FtJn&(%f@7UX4bXCf?xmQ!x+T|J_h7O4wG
zYF6JS0^*EvNffL-dl8&f?^sIli3@a9=bHxf2%KL*x&bkIOTyN*0(YY37pZ>H5rkY%
zwu`m4dvpnI%eIK{+6tf&yz7-<FUQJrOd)gP)5?}bBM`O{ZBL_<$vmuQWjl)VsAxmG
z#{}J*CDEXLycF@7uV<}jmK6-nHPz!yct`Y?gL`~+&<FE{MQ(ywD+h_Y=Fj-u4$oCY
zg051c%!0H0fUl)QK)$<~%H;U3Ze$t={nVd+_YW{?zD1mst4Q?rn^oWJDDqoaYW79$
zw!Hao0Q_&y;+FrJN0rf!XarF;_1;fwvx!}UJkAG)Kn~)lSKfGJsxY2hQt$P*R0Bwl
zc+s$1(8JvrV>FtFHR7XRS!7#3t^zmfRyXU4T<uMU$*z!ULby2HV<BS~Kd^tUE$7`f
z+;;!Me~ONs5|Y!P1biv)5dMtau%S5|R9)M1kSHJx0)5-(UhD6x=C715OB{yK^f-Pg
zJU4z5Y<a940{%9UVrN>+w%H%w$)(O_-NE|IETiI*@+emV?ICizS~K|!`p7K#zI!+R
zHN!{4wPAN)Vi>e+r-`>2B{NIf?649W1v13yPYeB=PsnAz6j@WxY-Rea+Mio9r&|qG
zE>`3s(RLn6+IthDOWBTS@Sx2tC$^7FBV0AfsUUuefTeZyX~F5><s%TO1ss5VADk(a
zCYNskqqdIFRS>gX14J4hFLz2|jUsrHmucyGsP!Ih#NDF4_0(?4*E(e;vbd7Xj*{|J
z8(jQQTc^$5nQO;2GrES+0p0>+S&$_+%@d8YOS+Ed8m%mX=mh&o)S@~|*U5pubO@V7
z@ZP(qS7UXp#>S2Mfv8EQbA5Ubm%jMLqfPhedudv6I9GGs?C5{gT+#Mh4tg6>Il1@T
zO#>evnp@#2iqday-js6U`)5LPO=olk!uMZvB+LJVCo(eqhmPFVkc=hz3*Yld^#M1u
zcHD|Q4G5uf0Gfeyz-Pje#dh`c3q+ebS$k7Nme^Nk#+n|pS#x??@nHDgP$pdBYjGd<
zGkJVm`*f-J@zwrARZ#zC{nv-H`Rl{H@cxp9yO)|D{WJxL4}PX%SdFA863P#ChEA84
z?({fSQ%zOB`L@U5HJ{}9d*A*2r?jn%SxuywD;G)COKO2!Us)?h*Oyb`9NQ-c7S-y*
zSm1;2qeunjx|K)L9L0NYs=Hd+kmT9v+z{rOzV=BI=yfS1*GtNKYw>oLmyVamRVVI`
zoV@geE3Z0AajVlj3s%!vf>~uM;bQq1N@Wwnq&??pe;7X`ik^X0-1(_d#y}E`pS78Q
z)nX3t5DVTd%Ha929%7jiIia5rH;e|Gx;9SaZUp^0Vgp1_z(6~A^2Qd`^B|lD{AUJ|
zyPG87T(K_3B0XSA?i|wHM6Vt3M^7`o0yEZyQr)YB)4HGeflHwq@CR#e^80o63Gihw
zST8m<c(RE}Ka$c%70A?s5e3m92lr*DKBo}ij1NEy*vM*Kw!T7^Kfyu$n+l?k+{rgU
zFW^xs^ZPE~w(vu}f{d9dOT5ia<WuKD0wag9<Xc{s(SE+Wc(EU0@`v9zr$f(QCUv*5
zsb>YX*J+aygjXz#y+en92p;7GJo&q*tldNu+&~&R2|wsvFO@f0KdU+$#;ST$RVop;
zNn!Xm=d@nu?gV!jH<?(1kRdAJ$~<-$cVQROUv8YJMbQTp5qA;JmJGvC&$4e)>IT+&
z81AL$$gjx7KLvxFB}<GfuYw++kPZ=eQcj`+kS+MzQsMQk6^2CS%WRi5&4&b7BugNk
z=z8WK_UWVLc#+?q$f@y^kxE9=ccVGKMeHAHYe&z43j>%rr{uA$LWK*$HIVHG*o0Ju
z|1P7Kt_^HFe)3xiuMmtcd<DmSK4Q(uhAu40HlKLpT*TlMRDu@|O<Yhm8gMNkW0TjE
zHKTI&#=_KS?*dY8?a~*~%@Kj224fhraN%f$g>4|g;Z!G_8=a1iJ5miDK+`|auodbs
z%{WwsnlO(It3K;6`cnHmbuS9+XYZ<<UlKd`w6w|N+%fveDh=BA*dB680TMyAXDr>S
zbmkHEV*HUwiKnpN<Hu+e9ox#(NjLC^{aEAdk(-?S6J9%Q)#Tn|W_EE2&^4T(<Vr@`
zR}b;chut2#vNejr;Z4jB*BKy)W!+Ekd~51aNKX{gcAef6l?yLwXd7-N+zJIwd|3rw
zb{N!0(|g@7Z@K1IxROyF+fQmYS;O7b10d|dbmmv|X|5)8jw5wM%h86Ggs`JQOT+~P
zRPCHM1z6tVC)uvW*&077d3ki0VqF({+W`J78VO$#xnUED_`pc7*-D<}X&9Ya40tl=
zWAe+A<eWx|QK91lc_y5&Bif<)5G(u##TG+g2Ul~2umRM=zP+s6Z|z@P3Fq_iZ1K@f
zlyLsgB&vei#6izij03-sv*X-EFXSIhjRsKNWGflm3o%2zKrPN^rb2|1r5Rg!ngSEO
zC?IFpY<I51ppQ;iS*=8QUX|>L2vmNt8sh5a(fYYDFQ?3B;$&VHV3di|8glmLd&4dE
z#xS9l4IArfXqa+33nY;z=l)37JGQ3y;&90{2lj;p6`W3IIV&dTFlm)(43btYYCWl=
zKOr0i4qVG@Z}vB<Dl%<P+l;Gy9YZAugHMeWffA9IG*(s08G)}_>-7s?O(Ev9th0ks
zj;7)b$>tguN2In#4Wp-jp0`$7hEFS#8|ZOmdpD$Z;|(?PFsK3`qxp`tUrF+d;S!Pq
z0~mawe7F3e@L*@Eshw2?gc*dY@3ypfGpmON1f=F6WQKrQ5fQR@R)J~Z*%hI9dV!P-
z+m=qH^gov8sf}b3Bt!rSX87G^!t!DrjRhgi&d?aIdZ69utE^6q-1aLx%lIghquC_+
z>INmY(GiODiPcbA_6~$2givU()jN*IyUr_tpYA#$W}hfPL%NEU;m}P8y29!guq!a=
zZP)Pim~^OW^mIBG^X#b_G~8=v0;4-v4MC~~yfYvu$<ue$;1w~bOl3da5=#RSMTWIS
zL&00wHXyCHI&rM^6~pkY%oOtGMTZBaD@l7R_Qo7McOLaEml+vzwU70VX=Mt1{wh-i
zwzA*l6`<yof;;xd^OGCfI2uvn?_Q2Jp>Dqz#<_TojNGMQU>*9v3=e_AiN4rv?q09N
zFZs^?GT6NHj#&Q8YH~CU?&ml>lsCgAx7BX{1bZo-@Q{NH*cqI|MXlCRq;HKalPXmO
zd#TqH)w38Qzhwy@<7N@)lP^s~u&I`9Kh)i0wk{or3>B?RUM$@B(4r}M#rfQcmg=sQ
zU8okjXZm~#kxFo4K{7qLc?(y*YX3{d3y>r=<EA`voGgLhU`o<CbQ#RS(78b9=XH;q
z0Hq*a^N5rH4AUFHSuL~S4n9OeJnXv5kQH;QpC+k>c+D9~I`BLIisj5<p;#5XTl`bc
z_lyt#_w=Yvvqi(T?S27~O6l&*b1cyFn)<DqQ=4NgbFkwrgK?FTz8^dzZQZ?6w?FTy
zoj*sqdn_h-tyF2*+99AHqiLG19FeWJIK#A=Mq}Zw-{sBWQG2P*PZc=V?{c&HqhnfZ
zW}q0AW>$L}r{uW2P_j+9&SmDy1dY;$^4ZBy)>J@cjkF&c7<<|!ZP}p~>oEGs2@ay!
zqU^4{>(}HfYYJ1YLi6!CXrRo$sU&tDIsjF;B8Ha#{POXxkKZW5c5R}6(_2xRMV;ul
z)F6@rc0_Xt;>KJQE@08^`8~{a^&zQ2oP0-Lck;Zz>Kc8%^BFjwT7H0E+y8r<*{~Tu
z${9bfeGEl)8?wol_8KT*){zQ<20@fK3XDV2+&3Ej1}2=Nlq^Sz8Qjh}OhTc^ZKD!Y
zThg|P264FitW7GG&N4Xck2bP&@bsv``|&yH>I+1K)V}y%)DQdrq<)y#S^rO%Z&_p7
zhKLoh=ak|DWGZghm+wyBU)rzRn4uLv^t`^gl$-e0SFK@;P~B=GK@^iS5@XYXrSs>q
z(xAD8m$iBPN!A*7!MEM})3)Q|@Auc~o2roEqiyA*ZE1_|Z-?5~B{%2qRsj8swd`It
zxDS&G#Q6cJPB{F*%j%`ej*lnYC213K+^X|Br&fvf?_0z7SL8yb#O!!k3*G6o2cV9v
z>;$Cc!o=h)SvqmNg14k^agv8DZ`^85E_IsIYDwn?*YEwY?j~<<SEwU429+_QD5t$l
zPaG(u=%WHV8O4@m2W)t2JmQS`cIhn3wJa+^k|_X5Q#hpBM^dvNTE~RPJ6nyB&tS4q
zr+)jNM@ac*an#b==E!=&vN%E_NQ_G5K3}lq=3LPRvm?2rIXl)!y+_fRKi=oJF(~E2
zKVxF1uY)^+B$oQhFdF3g4$v^xDg9Jhe0+Uh6}hVgLPT6{v=N)P%k<;qQGXLdYCxJg
zH#uq9cm3=*bfg_V+lax`2V(4dQjXRu-xQI!YdcdLSbyVEr*#Sdcm<aaw3sJn{c1cf
z9qoMJ`L@hasu8c$$&}1`^yCaNm$-P#7f1V|BL^<+;CAd#qUUIX9q0ZiczdEp#kWP>
zY_p9By^<OiWDE0>=FR!M1^4x<>@d4Z8S8RB_`DcRI#WAV2<ePRoqR`Fg*la65gti;
zC|cna^qjWaYB}s2eK90?COlNkv%{qyi&~8>x$)S;kV}MH{#cQ(zLEl(^Sxywj1pdO
zQ%WdlEcWv>p&DZ!JMosot0&?wPY=Agv6^tIJGd?^ATV?1p*PncHHB^XNNkR)?J6bM
zP>Z_v+K4~4DF~g%G<3f_4X0Jwr><I#ceE0P_mIZF3;+v?D}jvt@SFG+rl=n)CauKU
zX0Y$$EwMO~!r5f8-^<!$lj-m2Pd^I1#a}x@fu!CM^_AkI8h+1{9lI95{bjOjYg^ri
z=#&h9;3u!Hahk^q;*l@DQ{@HlM&HK^=7}0FoDW6#wR+MY;JQ2GF@GnT11p;5<OTFX
zKm9EQ`WkhS))={<KnZ%iw^%je@trH<H*@=%$72M##S=lxi$SwS2E`RsEqB-f%SCr{
zO~2_3y&scw6hbB-&9poZB2>{yE~8t2`YLL!iPuswFUC1yxHWGk;M;vAZCR*EK>MmN
zvc*NbTb9{eo(iR;vx8)o3$rz=S55byul|hiX^Lpt$O*x=*P5TJ1pRIX%qFXv4%|5M
zFdI#^6)LoI-JxrHR8A<UGLfPr$~YwhFPmdJ!WucIv5{lQOvSCp<dg+Eyeol|7?Yd{
zZLv%#QzWQnnke|qi`5((G&b##Ao(UN-l9{W&cDmgHIl9#;duf-!j@y@#ee3xwK(G_
z&YdEgQZn3_+r$PFC5Br;DARh0*{sws`Xo1(8PHfHF6J1am!qxsnq2yloc)76+JcR_
zwGr^gkN`?K@Xnk(Ul5;TSy2Q|#&T61S`<PCMi;gNxW>_7pN9vhDqxyu;K6RP(m_}q
zio4fbNmcAn2}qfe`a!oTBXlwDzUIiPUO}~wZDQL7TL`yx>Wk_^zk4^9VJ-L41pE0K
z8-a7v=7KvS{ahyI81;QPK9sLI8Uj(jE-fr55YhOU6c2>+NsV~`r#iGHwx&QHAwX?!
zt*ro+KXC>D>&X*@9Imy9?@tG4WNI>VItzWD2xRrKCMe&zp+XARdKdt4r~wi=Jq7EN
zd(gk2+Wtb*e_nH(i+^ooHcR~OOLq`jG|a(i7@qGSJ6C8SK_KPH{0v6uyiWhveDU|_
zU{+KoWM{?-^xpjWmn<c>wjh*S+>q|y%cIGTB1f|{FL0zoyb@H%v0!bFzzI2ciO`DA
z%eD{$ri=cU+?EPo;}6E!Xi!50i8$iI!i8C2L&B{gCCso})SZC+lpa&^Ku@L%qAl=L
z_gZQM@>RDPJE<_s^}l}MxB8WzmR8(haFk#pLc{~MvyF_ZPU6L2*s%|J@S<090izEU
zCLGu&*@Z2pfE4^PmXA1CfxJh#Uh|H-@guM4w78?I3YZXGx?>%a4Ft*bOjd(n^EWtL
z8GY1J9&eCx`M$jXTlPvIIbRCeadK_FfN7=>rH-w4=}*h+nUuAc-^LEQu?>)9No8!}
zrkYta3?`F)9Ff%Ca{Lu#$|O$Cxb0jQhMHH_thi|Y1Z0#+k#bhvifW<qTc$b@ATNQc
zrRd$w8%y)0dwyfv1=E$GZRNw-K+m$t(npkb)tvk+jTJh8L}<T)lZzgR?-;_Ij~VJj
zBfI28se*22>jdR+emXvLEm9J`l&S^mUYbt!_r#-(gOPT0<4E1>z<R`40ka$-T8<s=
zo#7e3C+n!^rp<QijH>X=m>}eA|5+y#gP<-TK5}R`o|MQYi&?%4L5y=?9aMp78o%1;
z=Ap9xBiNGEEOg@Zy%noQ>#&m3uHpKb`o1&kl<Z2w%S44sLliG_rVnANJYR+)tK;x-
zu5Z09c0#q%WMDP24nOZ|10><{KPfVxtJ3Pd3ac57$!WE3VaVwVQA~ard(Tc$*@ZBV
zzA!jZd#Tt3o`3G46g8VfVe+W`o)5Vo$SJyK9~Kf^e8ur765^s#Kfy)dp?@EVC+OIP
zjkkC1Ma$FOc7)@>VKWB|SxrWl71vGq9oa0XU5vA~0mDR;5G6TmNr|li%&lU$s(Crk
z`S*`cmUv8<u?UeuWjrQfmdO!O>ENDfA<j;PPR%_JbGZ&(F&Rb2lXo?a$#;0Jh3av)
zs0;-WYoCTV+H6k?u)uo?ZJ=F}ACGV5*wU23d*%7SFs2U6n=)ialD|IIi+$?_i`D#x
z)0c)6Z0)AV@?}4(t(6+IBHqyHD4Hl*8>{hO;qd{l4M@YsJ)Ow8gqzOlu{{|mF=SDy
zi7*cF^dLprWMU|HV&4LkTve-&q(g>$!w8})cb&&lr^T$-U_O@KI7lf%qCp8lqJVj@
z@S=*N-j3A?hqwj2A}25kK-&`@xV?lVh)?4^g$vu8grt*sg@XseB_t9Wg|iYkii>1C
zouC(WAv%hh)`L(q{9O>}>tk_@8bsW;*S&rvqvWuf<+Y_oHI#+fRb_HhqEL#IR1-*w
zly=!W;8(`pi~hx(Ah^o_Zz`ESvVCCBd$;puJU;ST0Bkq?DBE2x4PV{HukY*9U%+2>
za*F?3qOtvNp274?|KS<DtvMaD$pYW=O7Zl^WITO84j9#wd=B)CW_Tw8_u$z^=Ks-X
z<i8<=#_P`-4)dtt2H}LKzC5tv#vd%{;-kDK4)DJh=g+fmPZT=3vp&8a={{e>(jt|4
zf$2Isht={Mi%tLv$R(+Zx3U_Ev1mh)xxa4je_`tr6{DowgXBXDeTVx~epYfddSjR=
zK+`&WG{CH-zvfF%1NCEw{Bz;nh-qk>*W64M7E@1w*zj(rG)ZE)-7S+0qbyK1?uvD6
zH4Zbj%#hKUz-2aSp>7~%fn^`^Ldm{^cPost8E>R8coab&CRhI&uTkMQieq>k`|@Fo
zO$=RnOGg6%ms0C>=^sbeVzVcRh^ZbOCrYvpmA>|XWX}*Dtv#OqU=(0)@JA0TUO6W1
z*iCUCD0--o3~6XgSz_ko3@9QD$y-YsmC<KW6wRL%`6BX!b1JqI@?pjm<AA50etA-A
zkOUN})HF(X3^rEhMbS`CqPwBll>ZGMu~P>z+mh@x$k7Q{Q&HH@qZfL`P-^K1{FJAD
z`chKHripu!uGJ<}y)T7M@OdbSyZqL&!vFqcqxeoxPP`Ha>)RP%6mg>aephlLrtKBf
z%2uHJ{5MHcx|CCvh@yc_3kInYQUhl`RT?F0kt93ZJDQ{<IW$_<3o(F=Hgs5wNENc4
zV22!C;I*I7acG93dnk9h4X&VRRSQa3vJNxC!T16PM9Y|4b4D61jvEK8iR|XoSsp1s
zjDm!w+$N8QzL5vp87&J>cL<}1kNQ-BP#J(f42Ksz8Q1pMuY}!%6Y9`c(jVDcPnc3j
z-D*f9u&7)!zacAFH*7YeyYWOx|3tP)>6iH>Fo4`&Oh-6YW<*r6Xgs?7pgu;4qL4kv
zN&=}dr=ZA<<2g<1fjQV~x_~}U=@2>sOwZ-vd-1#K1m<I5&9}K^AL<6ck1SUBnURIz
zM4z!D_&zb~2#!%_2*6i{bkUSLNHw&{Zis&_#%N>=1yza$7>uaP{w&HEkAPCbHl53U
z<XFfLpp~AY6dD$jAeS+b=S<>1QNE#_bzK^imgzqNAE3>$%!Lx^K{Y3l7v|&1CVCtD
z2jyI*lpkxJO|lB%JPx!lE-gS)Fm_CD*$O92?W^Ss-xq`ywKm@$EC<M|(JCkupy(5f
zXeopNK}xDpv1tw1TE|z&LYDjZpgvJ@Yre<j{<|D2jlP&85)(;T-}KUG=R+h6CNRA~
z9hp;VVuM^Ba~$C-t#RSL+Ww=LGZH7>lt=y7$0xm@KT%V;F73yd>HEdJ`2KSh?&*ug
zT^L(6V8x;8eOky~7_~S{j5C<$#XWPUbQW)aTcuKi-(qy3)$>bdwaz50ppBR$(_jqr
z*k%-jVdmOu&|vzcFIvlgwv6({-8-kb=C$$(V8OcRRL=jJw6tAshtlCmPP`>&*hk%`
z_a>xp&fc6;Vx_1e{MM}!FiIB_y&wgi#Z9`A5f}G@g>Rzr+!43^tC|OqJSn{>h4a-M
zg|B!szOWB3v>G#CU4&o|I<H_I73L)Dp-Kz9vr^A1X=guyU-z%=10QhnA26WzY@kS^
z7o4=CaP!Ek6=gH8qO16qh(N`yw&?mmj;eMS-FJlTj6j?#<k$EZfd{yF?!mLMzVm1X
zBDN2)cC0wsv_9Ae$thkXpYV5-cm2{d-=Z&?liA~4m|KDxVlTT`8Uhj0&SiL__3sl(
zKe>{}`2IG67KR@TLj;*rwJYRbg_JC&^wniY<s_A_$?73*1L!Qm1F+GxKBHcAtT7(K
z5s~)V8_Oea|3Hw9HkXJt9c(114EQ&K{?WD~x=kzdc^d*8ApGrGK|4#1w(H=Ek&Co>
zHaD<+`|a&28;aEU$B9mx0&DG>4h|BSr$rmYev^HiWa+?ox9$9OayR<(A{TK_=kP}k
zS_}z}wP&ZdvIrx~=0P420B6dAO%kksd^nwN>NURQ620kEOBKWX(C|tjeP*kxNgW=9
z?l6_pHm~|i{;@~WC;HzYMCvTsePtQi$r%Tj*vrphO~FzKlqyC*7ntl6cc5!z9#fEU
zc-kh>Hux2DJEI<Mt+CYw7h*4(+*?Srbr&HVTkx5I;g-R6hw*XE<Uh4JH8)Kp5de@1
zi^1j#Ay>m6PUeCP$#(C3As7CUnT1CeBdXplniC}J7#PayxMX}<jF!Yw8iZe7DP}#H
zw^duq)tL96UGMYjEuIU<MF29LR(k{^5Fs;Y^Vy`Zn4<CZ7pCtsm_{gE@c{z6GppMs
z09N^2@F_&4AI*I$jQJ{FQ2Vc-kf11YwSR=!;i^Sn@Bkt-f^kmaAU6n7ZUU`g71(2P
z1{ZF?SYuG!e|!Bh`IEW#P4K_S+u|3StrlAoJ`hRP$(l9d9gDjC$jwl`NhXv5<zTCx
z5PelN_U_Sz?Fme{ZTrjWRCW^WC-S!8J~R4&_Py(8Hx+1Uf_B?WNTu9GEB`*<pgeT|
z3)K@<8z5!tC+^h;-8@_=^BC4y4^5O`W|K>pgKZkR!rH@-BtAbV$1_S&?P}z7)Mm#p
zMeMYQwsw!LC){izl&>#h1cU$bd&ZFem;u{zjvk4-+x`@j@655dEvuYw@@u`!S#jJx
zf9ksxLaC^$;p(55u`iQzTsP<8xlgOLhQ1$PqDS(ZpSm{odT|=jkc0l)GUDViRDYdf
zVLLw*UmSKB-)Jni>bxiK&hmbNT(tw#|KHky{eOed8Cm|rskK^tGKQ!VX8ZJCer1(d
zQ4Zf=B)~XGLnr@j0JYZkb?3yv%r6~bLO~H_>gJ0S{h>?p&cWeboX(Hdmcb=&D{<NH
z^~3${$BX9om;K{+>+5mtZ{M)GqLuyRr?dUc8+;1V-{6KW-=EKEjW#u`tY#l>!4lW1
zSvlU^6~Zi8g5(WMN$*Yx*`gIIXRDI;(?qG7$fA$yLK&$!()&Lg@~K!@Z_YX)oo3f)
zE(wBoH+?isdJiidu`QXYB%>H*V0G4Vt_Ls)DD6@D&S6ha)rVx_#tEAr(+mkwrS^3b
z;2EPFZ(P5p-?_@x#wwZBn1V=M7StN;(Bc#=%4gg@;}mYr9eQQCYbsml0#l^?6;P7_
z1dn;Wy>!lDa8IoG-_bVjtF~a#ygzRGN4b(LJR`?R{Xq$6cv<DGI$cZzeGiA0?X<nU
z4UVmjmVO)ZBQmwT8PA>5)qrRA?91PU(Uen!9;W#f_>#(~zzNXhv##e{y^kXP>zJwf
zz4ufaPls_9bXFtpE?nLJAemRgbf{bp25kr)TUt{ClPLZ7*3&^t{%+NwwoX+)T@<t<
zGy0G14lpurd+nzNE9enr^d#f;Q+4VdbwSF;Jh7z3O_eaawLhOJYApr4xmtrn7bM3v
zD1#N6AZ%u|<w*c>_yvNa&N;;+hgk6UW8-}S-r;YRf0o|jDZSKcJ;Ym4pDL()kt!|(
zXCl1NXEAo4JWi1apXTz>N>s88!Z_Fr9Z~P&_KcLQbQn+eSBWP|pkY{$`;26D;+OLt
zz2ctNWjklh>D=wLhTG0KLFFfayJtgfp+Tz@)3?~^MypJScLWo0nUNMBORE^N1<vij
zZpVcNl{LnM>bWhoUiLO_u#7ZDH>h)aJi8`tA~YRL*h>vCgKclg3qY|?jY_9FP{kLk
z2O*ZG+==JfubQwnL7=NA)lII)-4=3o47Xz15OhMdmeb9Z;~^|(5p?!xCGa+ZEWeW0
zLQFjS7@ymywt3)A_WadDEQv>0EUh$Nn<yiQpJ62&mVlNpyW>jf+GRVe+9eP1t$BUf
zi;SM%54|e-c2VZZ+PE7rL-CbM*>_UDQ``^iaIHJ-Tz7k*5XwLSh`&Z+kZ1j+^O2fF
zwntE$X`sy+iT_R@gAvomoHPzfVctiJ7DtP2V6IEv2tguCXVemN(vGXer7LO8_REY`
zV8Yg1%H&|oL1;GEW}IboHK@3OwD__axCJ3BCw`2_#nH_*I|>rDw2@Cd1qb4tV|7S#
zym*<Z{oO|{#zUk{K_IuoGEanPVZ=r2&%@2v5zX@UYOMTxfqjqxJ9uU5Y44sUOGNJ3
z-Fa-u2A<Z?-SpHXYkF2(pNULmsmTU9t{`(NUcl>hv?N}&QW+R%4<la>rw}I~GpxuH
z^EHpQw>U~aBGFYmrE%m413Z%=q&-)lZ_m9b9T1}ht-SAn7dg64rXTQW^Vigj*;`pZ
z-E+zCh0`(|Xbse><c6+s&9|P}GpbA*b^bSA|96=LHG1Z=GDdQ{*yQyP%qyB3`<<Mq
zpC|hzirg`Uf(8zu=Ws%QfhN+4IYpYq><b<%e*>ey7ZVGp4XvI*?C1x=ikfse$Z14G
zC^_7KsMFjF<-MXGt4jSE!fKkuJEM+n4b;O6!lw@ArwC+25QGmj&*Gj`V9oSbo=56Z
zJ5fL9$|BiE*x`Z*ny@WCvj<#{`5d&*Em@hTkck(RzHYeAt785W(HF4SLd1x2jePq<
zZfZk8vgWj?m4J$n7+N}kq;I9sJn3oRC~&Gl&jD`K9qa<BePj#)^NYf)ptBz&yHIZx
zLbCbpjRgU>)pX`!^a^XWKWF^`;-KiF$!2+VsPj<{*j5757{A0em*%+XK=?@d##*zu
zE#GF~Hc~H~-pNcU5_}o(I{`pb8nDdgj^+l!(7rP3G_1;8T7i692yRoW+>S*{J(T4c
zSNu^VhpuC5MOIH)iDM}72>Pu(LnV06`LMDKl{SOJ!lKZ?oa@o~6K<TX*1WC6Ld#N3
zl)exE|Gb9Oi8xE=yum|9Mu8t4LJ&lT8=I*Ta)XlSFGii<u$<5)HcVoSvRh7~6Gr8!
z{M}75oVdL}T7q!Q-(#Pua!hjAP{_Jug94FAlWKQ0=9=uICvd@@X#LswMXx$Ts5&!%
zkso5NDQ$dxAtaMcYO9dt+Wv-d-OM>%)?g(f;eImd{zy~2w1Km2lQAu`QBwa&dR+yh
z%Wtz2LN!mE;R>D5`7N{qBa0__#lsx1Co9?c50}D>mCM&aM{(HfjPnl)n&@uR5f~h*
zHce+9m8QTF3!(V9NA5&H^s<Iy16h&`g0v^!##_Ur#DXKBqQ?F;F5#%!_MYMRlFc@g
zq$!%NwRWR?Uh_kebR|CZIF;VpB1(9>20>a8;_G|PPtYf<&ytk6yh+)RRzRuK7Zt4%
z+!h0;1DXm?Q1-cRWYXhn`4omxO+K&8-LIoRQ(P;B>^~zP&CfarL%gcM<>9i%PJ?XU
zO64ZcPsN&RbY_&iYwY5&nAi>KPC4D~ML;f>9TLhdA>U(vlYoi;E(zHMrGCefoB{^E
z&!IH@<r%OOpL6ncfw16qg0Kfuiumc^gm=GmzCS9Wor#n=IA30Chc_v;AfFXr{fyVc
zrm&ZIl35Q4S6v9T((y1^+A%f&>JNoAF~i(=1bo#G#X&jEAhd$($gZUCq=-F{=f1nN
zo=@M;HMQN#)npNOzU%C}xMub1h3L?j_DV}|?QzOVfYP_HA(Z4nd*d+egz#!DhoHOE
zpcH)7qkqR<#?{+3=N;rSFdj}~)B@%tH?tvtrR5EgmX8QR*!LtaC$dWc>*XGQ!U%a=
zmF!gAe!Q~>hV+l9Da58u?g1lEkgUf~<J;owt272fvry6A&kN8q9%x5<egYdJr0w(h
zy?6W@lwRmyg8>F>r_E2>zpLhAxayC?oU-{RRSORlI<Bo6>a(&tARXYZz<Y;KkdYVN
z1h~<z=E-aC7u+^p%KiVXju=?~C-#Sl<v$#3wHKr<1Nq=V_d0tGVkt=8u1oL*3Yx*C
zF)ajUM{pDgY8o;Gf4rS{NkSnZV>>wy*$hjCRmNNlUgV#h#gQyED3GJSwU>uzove$G
zJ*1kRjf!%eRbf)HZz}iYoQJM*7e{~WyuP<mld-i;yDx1Gym^johJgRBxqCnlaUBv;
z;T;NT6J+KL*AJ(CbA#cMJ?m9gpV2)vbbTrWC{fi~Im<|Xmvq4mMsp+}IaU$R?`wH!
zY(rs&R*))>Q>oW8=|V(4zDrOi9PD<C+Fva!^(Jeybe?x@+Y_>?AHBzcUE||T<nv9w
z=D|la1%Mx@F)_-s85LMH{Xsg05lAuvv(7`_AQS{mPZPm-szK?HgCz)z{TSej<kW}<
z%HIIzWUq*h;}&Xu1iSc4baYsjO{5$4(H3rC^pLO7^Y<6z*Tg3Fd~l)6xFcs!c{e#@
zvszRRZeih3Wx-J*)JIL`{1Xm#qMi4@t=9i!Jo1<IKYT~DlXN3D_>hNpA1O}kCDo-E
zf02pvM*QkJt)0;}M$@8L>PYpY;Cwzw9bsve<ld&U(>-UhFSD&(-fOHu89vYd!MdRK
z78@I#madP<NTRMb8|31`+e$zbUn;2$u_l?T4Zc)#GE+Xk?wh|J?>`S6x3tEYL$jW)
zJ>Drl?}y!DQDxf@ox8W)NEGg=csHy4+Ai1SNR!6ILsncXT426tvXvd29kz=baJrnk
zO{Vux_3S~(gVndZ+r!O)VTIICt0Zx%&dr@K#>ZakIoeH8qq3wa!*An+t1G|5*Vv;m
zr5Cp6rPt#gTezc?<A1FGV8~?<rUVch5F?G%@S`RLp7=8xQf(3PkC}>%k*Vf!&9$MM
zMLI$%?O}_NhMB`VVIn$Bu#5TUNeMJF;+L@ym!)(W0~+vn)Zu&J%ac9IlZg%61zkl{
zpy5}L@;3{RAS+=dh(JFp-RX(l<;gLDz=a4cPsuD3E9qGCmvs^J#|JUOz*PYB2NLuL
z3iX<~p~C?OfJp{~B0JCD#W4aYm5$ZqFFwdDDHdKrH>n{unLQ==qe>oh9!5hqNOm>p
zySUL`sdP2<cWNS1lDm^7iJORm{QVft3rGoqv@pp7GirU}D*!uddg3zxk!PCp#kMfX
z-zX5I?nG&Bp?Y~PY`re(Qj=Zu|2BE7<<?sFsSf%PA7B<cByN+1gb=8b%fmL}i==mA
z(=NGZnw}nAtFqaN6{Ln<rN=E5Uun8w@cpdGl?nbA5P2%j`rnF>?SDBi|I7IQDnjYl
zOnUf`+b>AG)(Vw-uoJ-uLO@Uo<wcduZAn|;Eb4_3{7~$NTT2u27(R))&nVNcnEGyt
z7Yo~tc&nea9`=j)EY3v@@5|ek`J*DbYzTBR>(!{nfZr1hxc(BhrNP>dit&@T!)x1r
z*>mc#+3Zi%EhloZxwW<*4!;%Gm!qD;gbNo;CtbXt(r;C-7jJFRA04cxIaaNJ1X!sn
z+ajO%bn2VAPp#qKEj1RVv`5E<Y=78yLiZ(gn=v3@TsX-`=Zp@iw?aDRTUgL=tjaxI
zj5OO#(aaWW4x$V;&bPc)P(w;>*owK<{U_v#%OI85DT6Ecl=CV1A^vruB64{OIYdK-
z5#kYpsN!(NtVYDCnVL7>SV`F^eng}@L}Uky4}G&C_CCZk;sxcxf-pPm%ml#<u4Kr{
zHHu{La!7*sqMRWy2U$ILV&fbov4P+q?C>Wc{Cme>#PD3IsCc6sVzDH$y0`>+Sqb8C
z0i&Ry)p)hgcUoc>z)M!7H}ZtXU?Jr2yC8ylZ{anxST2ZoH(rE~@FkDo1G8;&#9ACu
z9!Ucgh3fAjt821jOJ4J~hs1~r^T$#G?~SkYFnh}Z*C_$e3O~9OVG?p!rekcBnS@ye
z$iI(j{)l#FN-a(WTSl^)J{;wX!i`%@)7u)ZJNqmFnMNE>+4GaZf1d`EKze=ud9X_S
zx03pw7z}2H|Lj7ElNN*Y@WQV@kTi-CcakZH;33R#^vu@bi^8cfT;edzR>-FDy|vYx
zPV)XRqJ`~`dZb&(n*%I?;JjL<5|~59_05HvuC5_N={3<X2SQ8H7-?U-P0DydnS#r`
z$~571YdAWSwWq6t^_EmvqK^h;KA23uT-Syq79D%{=ktkTX_40NZ7@ryFAL$zl<<UH
zF&nM@+(TB{uBILtH#^$WHT9|tpIb>kC0(!IaLq<w2q5k^#|B2G9T_&>#$(Da68)5z
zva~9n_eX&1AGz8hcAUCXuE!;(_mgj25k^^_8*U1D#QrpZ#dQqT`jY%|@=|i*t9jL>
z_X~@#K8Q`Bc;wOglYC({Y0P5&Y#pHXPNXYZZV6TG!1bg^lnG;slxsSCQ1A^Y*tSpy
zZc=3`H+*F({)|GZp70082x1-!Qemlecot)GmMHrqNX$%68u^6NO~CoNknSZ{p;OEs
z<?2Y^OH~1|7BkQ|OJ|WfAN~j9=^~aI;l0uHuBRh_&z^^(z8MQ63%eH2+Pv=Xw`jkq
z+eH6mecAq}Aq?|>rsZfSZCh0GAqU@jN3>h9BRl(Of(uw&@NfJB1iS$XIm_pw7#Yyq
zKQF4J=_rZs&Zl%NukABao=aZ7i5GhH&0Lwo1#dg~PWKTt5Is3wc*cv*K3ohUey__h
z5}USI4lNxC<5zxtRA*;<by_z2;oMzbb7N9xNw>OtmD^EpaO&Q!ITGTgo=sf2Wt}R%
z_2Da(357UpFJ~3oA~Ky>ug+cCU&)-1J$#W_Z;xqr*0o3O6J?Ib;z@)NCXek_VR5oY
zON!^-h)61}dLZj7PAbL5TzPY_BCS6KKj(vt-F0Rr8>dc$izpL3BZwQt9AFgGDEL9C
zB;!E11F+42I%199)`7xuxeH4E^hX+W!z`%-&Q!$|5s&%pNo@x*gF`~{kUUVZHKGEW
zMpX8t8-Tc+@ZA;6p$+Tgm9|UHEnZy#2KA^Jt;(&673JHct7V5Tx08Mp<k##EGji+W
zJxRrwqJZllH()&t8{^+VF`~aEX@SfTBSUA28`X(j-l3|Ivt{KTJMJZGA?Z-}f}GU|
zOOG~<@YDocMWDoO??q8JaFQCa%Gq5(n_jMA%*@)m6ghvj5LO?~nA+>PF{aF`i~M*V
zY)XPZ0*u@WU|$<UddlwR#@M5$YTv(t=*UBG{I^PB_+Od}BP;v=31ex;Z(0QXKh5P|
zrEDpn!g<a?A)r>{U%O0p@bF2AV5ud-KM%CliW|{!#PzY3PJP`!R+Itf{tIhNSYhKI
z1zBX9P*)5>^`z~h|1)yl6yJT$mvnlnTF^7@x97VvaZ>2@+p%-ZC)Chu$jckP!y3!T
z1vEHA!iyRVZ>@#$%qYXARFD`~g6r3ta&Nx?bww-H@@C@2IxXioKUgzvd0B2+oWquv
zhJV*E)dIEcq{N7gl3x~nHZK{t%qTXw%m@eai6*P1yg9F1ApJt1(HOYrH35x-B-&oA
zcvj}*)PxL{GUH!37JrF|hm8`WHR~2R?2>bB%#<$wL_(h3tn=E~pm{u%ptW~tnxs@R
zk;5Av7(D&oUhe;Nj{5)d<IrB<w++Nc2KXoT0}NA(5vr2^GLTlz3{`1ouHV?~ucs^F
zo56j*BITUx6K8ncx%CCo*G5P<6G6nPL}H38xJYQAKW#(eqqVtE>LJ8uqnp0>o?#&c
zz4tKaO2ZLv{b0MC&aQ6z)F$slN71%^kfxq0jw2H053el_0)(5W3V)-EoKY=XB$NOW
z?7t>?pkD`4?-cj+MP72ML9{5UHn5#G-<IEwY~_LdMZ`2pmmRfp6p%!hBI;wgfaI4b
zO^I*H(m?e~_#IgXE!Ebz-&8}Jd7#Kz2r`zLM{O2JE?cui$zS>RtDMIcLsG^L^%`;N
zj<tkEDLNgWMnWrxr(+pHDc^=$S{A+9q+J?xpcXXn%$7sYFBW4eW(N{uy5~szTAX2d
z5OH$z;l6xebFJj@i~m*J@!<-YV>@DYX!3pwCu&&e^xxKl`Ttw&ip~a3?smp>g5p9V
zwl+>m|DNdu|9>UqveW;^HxR7~rG#XPnuE>=lyQ&(kH*Nxq;hFk>8G~qd|9X_TuoOY
zCx3*1{|ADY01z0$b^uQRoOGW;2zd??P?&OWbEBiWqvibmY~`?Kd1J+fyQAe3?#F6Y
z2HfKOJm)$qaeRvRwb!-Bv-cAn_hjGy-i_<F`I2RHOt~@8lt#5^s8eOf1ZXXsPplq!
zWPvLMV^HRxu?N?zBF<OyjG+lB8?x^f*li31i6N@hvg^|8%$QB*4bt@(hdAp%qYg@U
z7#uei{^P^<bJ2IJBXyX{pt$m|*r8b}8wC*^`|ATzSR+LtC9TLrARzE`$|)O-aS|xK
zF0IWdSS6X-(BSA^N)5k}HqKo%|5(L%$|*Q2JehR3h?qVCm{be1->9re0wu9b$*8E@
z1ekGOoDro^nLH8WMwdc5ex$7EUMhr3r^0AST;b%}l{J)!3dP}XqY5fna%Is}l{$1`
zf<!uTGl#XjTpR}xgWNxyN8;bCm!HChKsnBGw<`H^7b3UW<vC)`l_eCkf@xVQKwx}G
za+#q4S$U6KmZ!77jR+t!syhTkpJ0v4il+tL6G=0sdPqfIe+|<Lrv+g%DA<s*UX2<e
zQNLV`qhx4FjRo(p{S`E7zlN)hdyUs!(>-;x3RiLR^_ceo{Phf3F2Fg!)*i+u8ysJ0
z$Ti22oA(*f9mQ=MQ<Fu8zD5ikG}f)rB^!ySW8l&sY0m(?&-56`U`?tIo_P@-14X?;
zP1qb#V!|n*pdva%pkhsrk|B|l$aRj)Lw5aFAg6M-*<Xe*K`t9t8mhbQ%4XDgve`#N
zk~q`#RZiQjwYVfv1dulz$TXlKjal>+E{>%YgWUFo(DXeA$;8Y(RS#h6h;awlH4+S9
zD~N?UYAZ*VuV3rnyWm5O5>R$(kz`H~E=<T$2hADUmJK)t^K7hVF)t^G5nxc(o@HWq
zvQO#=p^!_ivQQ64zmmPFH`WU{AZ204s^*5Y%R6!Pn6dFYX-#9Jv^GD=L(-=l_vloS
zeO<mDYJ7ZezO+@!*Vn7v*|j2YWu20b#Jqe>(>#x=f(A`0p<6h4eL2Znvw|c^i0O~u
zJY>_^jKZ77<!DGurITiMzh3;;pETK2EcA!ap8~w~1we++Jq&#FpB!RL^Y9&<_!DGA
zVIbt69T&fy;M@5EtWi&E?`|gnYag&s=b=E>q4hkznqn8K<qs|g%D0mb=R2GyGu!Fa
zHFmG%ta)`)eZR$1dgN5WAGkQWjF}3(X!MBHU4AC=c!=(w5i4dS8n-(1nVX+Erz1$~
z){JPD@b;?it_%(>4I6TN?VY)K1gsS_lEi{$u{hV;MMg~AzE?c@tUQEWYE)~!`!@39
zg!=Oj)uU8rtW<RN!#Z`KSue0wuKn%M#kZ^C{`&(9*17yt5rNlePN&jsJ3W#q0$A3x
zSm*0{JDV~8_-?kPBnXa|K)?;G9m|zEq%6Av#gnIYpzl-r8^g6!F!uwVg5oWFCdGm<
z%68DBILUZ=tnY9gDYFDV0|T(S#WNlPN=%mJ;7FDBXnV<hptfm1>=>aqQBgizA9(D}
zFCRsI6C@G?$Z3)8@G8I*4_FD_2xkGhF`yYIGOgRE^WYvk$|q%RR2+S{My8=;5*iPv
zWd%sehOwkEuI3~j?B-?%GK|0zfF}AG%=PX_8A*V9l7y)W?P+r*nfsNhzX?el)wyHk
zPjIy<uZej5(*Qkx0_TbS#5}Na(x9x`oy?70w)w-?<@C!6gjyFrdH+lw*|UN51*0)>
zKvg*jlFdEOeMmI!h^b^&OfgGjQ^L|k!okp7^7!t!JvGioE!bU*!xQKP+r?pz`ptg8
z1%4wx4GDK?jyfW-TcNenZ#xC(Tlh@LKf<8ndqmP4GXB65aJ<)^<*1fg>z0odPv}$q
zY`UMmm~z}N59crep*jNxZ)}D0Jd9!891FAxLIua&v4fQogg|YRv;{NWKb9$p@FEfM
zWA2bavNWC!e{K;X+#HP&AC*k$#4wCjV#3+m?ygi}KrcSj@YC^VTA=~evG#@tOzM+D
z2PJ~ARL-l>n6CtASrFwULoi!lO7kNEe(h8%F=0v+Zr&gj^t9B}^q|V{`Myk=?J@de
z_gP+J$A-kWU1<mWc%k{Bm2*nWg9F@_LJS`8yzakQ+i_8a9ttDKj`pmv;HgqR*RL}_
z2(mIGK=22P{~Wvd6TRf0P(K8(et9$`@}sr`p#6<ipkr)4D0tCFva0s4yJ7Qxqru7}
zyLj1LU;D!`7K1vsDdtGrZDGz11~dis4I|k6W+$Z~s$(4|DeJPJoI`aaxy8E+68Jr+
z?+Wr=eWzWCRDSC^%QO;~&Y-sn%OPAXO*?T<&3jzpZw1^iT5EpyTzr!Vq%w_)2IF&D
zh(_%t)2}>B?l$nuQhVjbWDWQ1EBz}A+VHJa+<SV2bJ9fCS>cT{1wRN9d7JlccAqOa
zr^Wm`<G_YB^D_Z~CP&^!>H`GK1P-L@>khQ_$vV`RMWW?;0VadW#_HuHeF=1IS8xWJ
zc>&Lf=obTN+&ld0|6*oc&O;YFu|Xr9R`WU~L6<=05`d|(<Wsh&v@HX-Oil387Lj0y
z;BvS^yJzPwI>~Goy|^49;@q7GU?Wt?Y9&98SYD$QdBr|G69d{46UJqayOcn0UUYsw
zY!C2G=+9B<VT)UA0Z<|WK}|@#LiFi5J_^drlF@iEj=*=<CvBWkJp3($L~lxwa!N!U
zLD_rd;{0Hy8VipVf&lm9@{}l1F)s%Q>#%K&aJo~=d+N*`<cRm64l0AFD|S6qQF?$a
zuFSBUv!Bv-QJ9cJv@BZ7k%R$O?loDPY4)9>p-o~OEy+@p$QcKmH-5XWsHyB^`Y>Mr
z=bUp*QVLZ-B7ZM@Kc8tyD<of%z-Y*jUQ0(#Tl(DYp7}tczN+ju!Ltj5Ikq@giYRkR
zo-;$x^&ABD+4jTKW+C_1^_%5fNJ-q%?8gFW!wMOKU8sR@iLa-Wo>x7*?7nv1k*Pv<
z<Z+JI+|_CKC-?aR4eR;jz@(yFNLQIJB`FmRZ85l`DGFU7dC@L7q{DOQaR$+0^Lhfg
zbsoe*Ahl1v575d9sbZj%!$oKF6R?a|@NXI|%MBJT3T-`8b63379vs<#Sbh{12Es%5
zvUujshL)CRN`Ut)Zo8!!TECW*_reGJn<#HGkHy}D&&ENGj_f3uY53-1TSZxaGbDVr
zs4QlI0}-SPijd=TX%UK2*$rV-_P(o9>EH43<u(&FGuV#eteQwlsg>K4A;{UX8xvnf
zG@t#|Ofs<aBIsFZAO`>mN=NfHy|Z}yk&Th&N-8%;th2^>-Ug#>cE?QPzFogf78MGt
zL0q`*qtgU)!fe{~LlqmWPcb=lC@H=J?lh(?CO4Wu%eO%LD+hZ`byP8e!iR~LlZ8mR
zTP`;hJv!Qr-s~xPAZJ?bIHc$l5`x;{x&77n+QyLt4K76>WaB4U!JdHYzhM3U62)$7
zYdB4cUw2*Xkl79C-;}LC0D_l<+(|X0(@V#C54uq`LxUsT*2C;wpjf{)hL?+zg*_>l
z;cWbC2a^9p!YE}ByU>i~yBw&y<CCS*Qb2feYjUF*_)M>J1ZG5yyc+pM5gL}a$A#49
zE0Y6TqadRtryxr(Ugf1+496;=$F7}o&b=~Z5YDY^u<3gHt+I4>Ozj2D7s6_Ba9zHA
zIjpqr=Qy7b^k7QAFIO>9D&OXM@efzu5S&ifLbKl;1f6#Dq8EFb&61;j>jdxAzzy=&
z71>d~I)_d>dY@O{jn_4DV&+v%oAs5~<L*P|qrt~G#smW%&fHX1Frr5#hfo&slY5yp
z+PiLb&l@54Jn8CQmzMrF<^De%n=t+dQ(~#7w+9cegsQZW&|kMF7V-xG6A>8vzy10D
zF<|}wcU!>9#=!J{ZjFmIz}%2VTXYB{Ei(Y=2<s$Gq#@uvZ2jjv8T35tZ@b4q3^i#R
zCd3?Wm0Psbur|>%cRa(9@QL$*!NVaal7fP!BJR6)HP`vv=(e_8cN@RPe_TX4PrpjP
zPoso_l2Hf{!>v?nG|-Uu5ZUy<(wuR19DGydCYOi5_<p?CfbRVE)Ddh~gYk$@z%=*u
zxraVI)S72UKfxrJ>FSTX1K^9F5O+GGE(Qm21A~7<lwZ*;+Ubi%!GfI81caQ(JdnP_
zU&BX-c!d<?V|iIKPH~Se>Zi7e=(otYJPy|G56S4jA!A^l?;l`WGueE;isn;ZF<c%;
zH)ve#AE6?kA0O)-A)Hz@c~)$e)jF5?;|G-SIuZE&U21T5Xn%5C_;Xw?xmh;H2e||W
zuet;RkNWlTVwyE?uaQj9Fpw`y$-ZEo?_TX79_?Ba!8$yHHiA7p*J#%F>UM1G<@DON
zCGc2d`{?0184Rzk2DrGcx&{XWO=GaoXkRh0kT3o}l$}#_X3@5-V_OyTk8RsV#Yu&K
zY}-yMw(X>1+fK!{ZS&N9JG-_0a8GOVea*Sv#^}9&oqhVJCOXG-kCXBW*TnHKU9*0d
z@8q}%JXW=<CL0GI&Bin<)xz{9*~ZX_p!J1*o{y8Ei;}s2lpx6m)1Z(M6LT=O1OlZ5
z3R#5mpt%tOa|*Vk95Dr2V))pgsu3=8cAD@cK!8+$;t0hl!eh|e2(&IvQ;d!x1w}UE
za1hRjye?8xzKUcC?HmT$h`%miMar6@8C4;ebyuVn?J5|0SK1w2AeeDi@&z4Kj2Kl6
z0-Z#R3ROW2MOlokE-{?MD1vKNXx6S^Tih=kavxUkDY$Fsx4KwSI7YEc-K7e74)qpY
zw-Nh>WtdvdpbS`X@WRix;4G0Po#&>n_Vc1YX6=`k-39NfS1q0h*ZC*6vWj6jo$@hL
z*Ig3NJge(9&&WRb?;nO8gU!QFIENDaXS^+Tw%b_g*%z)oA)k}i|3HCPxl27Ry;-0Z
z4+cEBrvt$((vx=JzqZV_1fdf&b(W&K>{3~&449ZTd0U-fSOH$6HmYXE3c>hm{kOcN
z;-;y-x@;CinaZUGbv;_7W3scrK_+HP6KlAr%{9Nst&3)>AXk6E*GkZMOIS%;SavmB
z=T#nlFPJV$VI@zr*!UAw&*DfVd;h-PTsAh{ATOMaA0}h9x{*~g^bv?P>ez0MWjDL1
zg*KP+2d<By5s~|3>p(LRg3+dYnUI<@16-l?F|3t?MVO3@77SZsQ@X8gx;$qy6Bh>y
z6I+>0Bmy0M7q-$AZqFqj(sOBBbw-sMgxZ{Q^!Blc5)4+NFg;uT#%p`nyrYg{^)`Id
zafTzmslC{k398K#NP!rgU+1yewrydiMc{b%rl5i9WR`o@s9wm~UVPs;E;Ew_nDusM
z-VsA>I}T%uwE8pKF&$Cw0Z$BHm~)w)AwSmjggq@Vh=Q2>eDjjvB*6)rKS%erPP(m!
zmMk<8ZlI-TtLM$|p~_m4+Fx6J&m@|`wiD?$*o>)qeTvrEaa`ys#H^&o&nzZZimMR)
zR17;;0;lof&7v-&mQ0@Gz6f2?7$v?~#C%>bWX*W(Vq-v0a3jKK0AB}9lLyt-@5mJO
zy9t2KJd=qZ2xnjG=m6u=b6^hcyO87#(e7?XKt3Nq7Y**f$_k4)RY^Rp7MKfn$d26n
z=_K5dfoQlSq|T|O=CKTXN#880w;Uu99<|6<ZsaRBmvl;XFf0!f?bN-2QBV^P^x3%u
zTfzontW~|R?s=rlGrrY=4RWWRd|~my@r<S+*wD}+=CVrlymNW^sj~vwZz2JQdoaXa
z+EHdE3INRDFcfr+sum@s6*}6SVROo6%@>6cS9D8)RsLUCMU>j_L_o~gR514iZ8{6-
z%j7Z($pCG77UIu|oLO^P52UNr1sKaL4d5#?EUeP(>&!ZJbc~kX5)|Zu`@A~<2ZZw+
zU1+PGmP8)TYGVqT0iobl#St_qEF|#c<fc}f<{YD1rnm@mmFZn;{r2x8NHwuva7gkZ
zv0zWdnPY{EL?}aMlQ2HVg%`JHE{w}Mr9}D)@jB<-;mOGR&HML;_*xqOVU`C1CC_~p
z@)kWutZ-eubsB1^|2PDFuV})BsR3#OuH9BpOd-1Yd`^_q0tX>7t%>oYQubdA5mB5d
zv{BNue5G;&Z?qK_#krlf*A*(L^?ho0G}mEUbDA7~^+F3pl&C7mxFssNvMc36>BeKi
zwFjw5MUhRDRam9$GG%IZ1M8hTcSY0vqEB<U4A+)J;F3wT2LE}5I@I7S$x!@aa<O=a
zVCMV-^A#%+dMoGEyE8<+gfsnFJR*e2E<DDA$iq$ucJ;9hklhe7Ai&TtAOO>logEXC
z9b~)yZGs(M%!zC!^ap9<B$b2$&uVh5Dd+XcPF!5UPH5oA!v%g`E@tDJMMslL`C`S%
z0h^lk)#Z9P1{e#XqbVw-$WGs^8P!$#N+)d7*bL8P9<2LxhiGjS#*_$1w2;-x?q@C6
z_?8tdOpI=eD>iEt>)cBhP?AX?QLmjw-#u%yhr(*yx}EF@V*>{%8>=AZ-mVVwhSs8(
z?<LX~zm1A|pK9V;LM6lw690)34=kIOlc{xm2e}U6r=kS$X>yR4rThg;LPiu&eO9S(
z1-7ij1~kUeY<RZP(GDALu1a8Cj`NIh0Uttlb*ufSoX-S<WX#kmEc<L@v`2GX5Q9B3
z7n}TymjT|IrAVP1RN{(tm?JP>g^z>HTOem*2ejr+;#Bm`A!b{3DL7KJ_$S5}Ssz5w
z2hHYak-6NW=Ok7=g#}82>>6VyXw?OxfxZ-PHbox76lY|`b^hRzhF)TkqVxh)BAY?k
zGelaiwZr-*_B0fJKs6#;MKUg8?Zs*6NGozPA*ZT)sUJ~<6wnDJLUzT(+#i8bYBucj
zS>7w{cs0$TWw54mZDRsp$*t~+P#zocN?@K+CG0R$2q~641gu#x`}EkKMX!D+j4&HP
z&gHhVZrrLf;T?^zLtwvb)WhQ%GKAniSuXA&8&iD#0)d~csdu?rzvu$J@{z1JZn0EE
z^|?(<Ip_Ritqn51W>W9r#5t8#zcbh6)TGgkns>skqe9@qWzghr4~tui9KA`&V@ycK
z;gx4rKSzZPzvh=REYUYQil*6nCny6iV?Z6nsSW$=8t_C^+`)5w*zoap&U=C%@1D3%
zC5+oystgN!ePhkGB_v}81$}Y!yU#AsrPt@A#$J{>JO2yI+9;zZ><lX9p1j<p&`^&*
zsu6iKw?QAzQ9JOOiV3n<wj7yPepNJ5A_Jd^3d3>QWAqjcNxnPR0)wP-Npd?fesx23
z=|ro=VRj)p2!z7yb$ggoS`t#1CP@##F7-qZdgep-LwI`rfUWjK7I0{<0hIT3I~WBO
z1(jm}xve;9RpRMns;?-?YXo!Fgm>{^IU`&%61(2ic>t90^qYq_@;|W^*qr?YSQWlK
z9dYInuq^ltv&vaJ#oRuDnWIZ<rBchDCs^SuoUc85oSroTQ_f_B*yWUi21l#oWpOIt
zcbmxyt$8znO2Se<y(9s>Mc7zG<b{SRzM)%jIdE`JxQaegSxlGV(?EFs$15FkJ9DBV
zu#EveQCFt><!rv5#NwmrMO=h6{E|PTs>*fq2YBb%ok7LpZ;EX^D#X}JJzu}6q6h=i
z&0|vG#rF?fn6JZ3M_YG4IO;J<MgSG`TA2Fg&+#vX31e7fQ{bpyDx&wWEB%dG!Ds(o
z<0H#(0Mw)xx9qMQ9M+{iR-zm6n~C2nZH|~Sq!J&FUy-19$b>`GtMa3NoyofaYQ(L;
z{frQvxLtaS;p-{B>L=P(>>ZHp+b(+5?m;Y-U~*^AiKI08aS=$dFKxaV&6|AuTzl8)
zW6LdvHA!bM>s?RHeu*BHvC@QJ&Ffa%S&QFy_u{C-b}CEN`4)+ZQNC07idtXrqWvsI
zSzHO9S6eVWG(>DR%%JyCmpD6s(Kmb<`Se)wdC%7>WzaMKcQsn?#+kofc+;2Hw^HEg
z<;YU0%Yj~ZDws#~7t47nZr|cud2MGI1_V2MrED6}AHA+s?0IbM=_?C!DO%IAXACBA
z+(w~%8Rfxw-upfH23eMIP`z+it=c6!4XSWOIAlj-iUG4jwe~eIblG3Z2c{_+*T&r;
z0|QzM#a$iN1O4<wNIKph2l-vm&<H!751{pdKUWm$JtIWDc$^MkL4A*Xb+KRThLpS^
zSM&b1!P0P*uXzzZ81Nm51aY403%d@I#tcv@ZEQ2VHe^X5KI+4Cx?HC!{rcOz1%jD}
z-nx@58_YrqaqO=_4&l-k$@fQv^F=#9)EX|NqlwOum`G^42NYxIE#;3%3HPekP_6bu
zRFdZhM=Gr+dM6rqeZA5}F`RpxIPKW>pWHS_ZnFU%f8%X!vN~bpxKD0nBpR$jTp%NF
zSk!U?#|cpl+(|D7v*%1KJc(JkJ_&HN4UvZa>@V7aTkc}MFS*D1jf_u|5f)6{iL1^`
z^{H!^t;lB!fbbg&;A!|iMZ~H@jUE_iI-zPIPsd16-mo%V&`+r)lh1>@9lkt7Twb$m
z{#>j2{rJIHUvV=ILX@A6z<|8kRob`);IfO^fQes(0+Msyp=Q~z)@uh;`FaT1?D9{#
zrj#X=lQSgd^#wiJ5$E=D*){v{a@$4?NC+XSGXxxWYFeH|Y-}bY3Kt90H7Q-8pAV|o
zZ@IiT>)YwRvp}D)<ohB|)sSS~@0tW8WQ{Nk_g}x%BFS$M5_}tL(LSQzZIJF`0tjtp
zcTH1<5L}Cl_mVKFcbOO=Fp5%+G-i`$i-m<}<4H27A+kU3=8~^L7nk`~&T44VWxvF>
z@EQCYNk|`sK%$Iikj<$xQqw)|UY6TI@yn5JgZh3i^O)ZlevbL9J=}=#P>_&N<Lsz=
ziv<4p?!k8frLJvZFvnp?ETS|W8kz^aVw#(aUesx4#@kxZk`|$xek*vKSSnFJBh00i
zI;Re!<kytTCr7ImAm0=!)$57NhGihKn@v5ff~*7IH$d5MlBtQp9Aua^rQJIjU2%_E
z395|_My4!(b9-+X^>6RIJw852o=?aO(#9`1Ez7Q;n(*l-E-#<8v<xOrg%Kvi8B_FD
zGf)nScnj{?6b=$l8iY$SCSetoiM|H9bd2#ceN!rpNSC8p*4)k{JN)YdqB$G>21E?a
zA(>$z03%c!XcxojwB1GrzY~l8bnCd*p_nlWdYfo=SgI;Tqjh8TVDykZ?ySzCltn9;
zIw-wE;`inzV}}@_Pw%YSaRp=`gsz8a=d3UKw`()uz#(hYc?ZRajr%#r4;B}bo3e6;
zPG+eF5Ttp@IA1Oo+f8o1qzjQ|2!xzTfR;1zT$C|&k8}0gZ3iqU-<S7D^Ahr|+8G}o
za<SaE@YcW?`B9eF+~x_k*{t@+2`$XJmF9i_%BTlkO}CaEnW@5Cbg*`JR5)k^)xJ8l
z@~7G4ithdJ&IfQRS4J7#jC`@r1*%jx`_pTjH!3@OjrD;1373<dwfLF*aq=|sx!>Tt
zhCo4NnpFFtTn|U+i^P+7ufwIx{>{VM6L5I;jh2$(S6m+=&Cb`Z@4E^?ln-0dMX{a|
zW`eAI9k$Sr2nO9vD=H!9%i-|3WG-p8bB2Azk~-|u{f?sb%i5M(1L;EwpA$H;mi-SY
z5ZD(mnL)}>B}9lGchdckbj~S7k#rI;8cXmj#j$C203c4<Oe46d7sT6yi6tmj>>DtN
z-qBp#>hwHP%8oTgmDNIT^DU=!(+sceoa5UB?lLO;rI&k&N8sL977=a3XY4b+0bbDD
z#MHU-Y5z_sR5nL*gb=+qX~pglwArAN^9#FEX25+~?XsJV3*nmR7sjCQ^_E1gd(R26
zcpX>T-et5`+%N-R!QRrm+hCkUdVbxB5u47ln&{aFZNPGkXM?pcO)Hp$1+LD3-xz^!
zLMW=?gM?KerCu5%U8M4A)F)5dS3__19ll6hMG74T_MqI)zd86GK)5K7P2`f%<_X|e
zX|lt<F@FJ1Jv2Q6ZGKob^7w+?Gyq5Jb_GngTL@gL3P;v-A{d;SAOG%x4>%p%WW5bo
z57C+^N_dBDL%$5^<4w3xu86T^it(e|KC-%+_gpHSq(h>=82OL$>eO;jXFx#PG~Ix8
z#)}q4e}FJ70pIs%XgVbyec5MmUNLm0Tpbc51J&=?5nFTeosHJJwz(3LL8ZT(TeN^!
zIEhR3yOo?lIlisO1s3-kd@`9q!qzwo$#Yb{i{1!pxV4?K+8i9vQeapRdS_7?-NYkv
za?lIkxwK6E-?GI)o`suz&FOJP3tzovy%*pgK|R`(?DAMup-lRN$<wsKIrbTDHTj}*
zvV**GhSB`Je!183WlAfu13MoLFv$4YAaQ^#7-h$v0Ms{zcj{aKlT5Nqd1yw0`O{c<
zgFD(&-Yvdj0PNFnGcg09<w7HiT5`_k4Xao-^oVk*E}&b{Z4!?s43YAi#|kBKnO02F
zb&?0JCh`@oQ~8M}DmOWiaG!k9;ZLfdQSIYSd@Iq_y*qY&_flui@aP`79SN=dH!n-i
z39$!uc=SE|*7T(QCPicYX3ggNp2EaIbl^ZR{70}TW%P+y=|u0bk4+hx(~Voba~gW7
zf5ez}t<ot0!ML?N&zC42o*U)c_)_~LIXI!lM!S2|7mn=p{w|2$TZ)cyfE|A2ZaaCh
z{`Cp8KCe0t=~zG3$<M!6kAYvZr-w7utKdo_T^WkBZmcblSLzL#(~1f7NSP6mAaoLK
zbB<|NY#=_ah;$}|q_mh{Lba2<%Q)BK=TXo}1d2C4z@{(u>HDq}o)NHxp@i)fx5Jhm
zj%Z_P%fb1S<{Fj<rAlZ^8=yA~fok)FruBGr(}8+EiS}SV_l7vqR25cs+ubQ<<6UH$
zM*d>G*~FCsEo3VrrIy-`?-{<&<-(B82^m?=RT@$fPuZm5oq;xNl&dqF@R+uZH_MG4
zWjU<9P0lEcEvLoIL*l(1WCM4_{_V1>I{CH2pQlC!__3$Q;vcqJO!W2&YqQY?m(-7#
zHbE}U9IzGXTxs}IdzS|aX<3Q+^T$9CbE#@xSnpO^JsX5G&tup#(q3>32_IO&vccvn
z7)O3I-{6dxHD&$PG5^imw%(Q{{(18GH4-z6E$;U2`d84ifA!D$$j%YmBHL>uw~Pc}
zrRE0S_)3*%Xlj<{mdF~dQGFbQvIL#*K@|tqHd0M@_nKB`p~^0trl4J`>o!kpH4WQh
z0enpy?7PWDpS%G1`BNDge?8>KA@U^#Xe7%CY%#*O2d9Y}F}yy(RxR4Wm3~Phfwc|&
zP&F0n!ZBLqx_lM=3mRr=#1O62d&lZ|R@**4s5a@4VW$Ehn5j4OJ7_it0Wv`#bVHxR
zUE{9v5p|q)&i76)`Cy*QXG5?(V<9gMreN#G*7G|ZJKU5P_CuWI51FKfxFU+L-$Sx|
zoOV$Vm_<9;U3f7XjZ{9r4N~!(A8J-i=r*z!o%>Eloi~24YKDs1Q`hUgfXik}?B)83
zeXhdtta7xPl>VqiUin9PszPX0$^#Z65AY5>RIiyK!z1^p;Zr>YXSQ@_2<Y66;e7Ek
z+u0!{9Rf_LjnrUaE&csIj=iO2@Z|y;NokoW1=)GmL1tnZk7ef|qJ;9RJm465c))B9
z#1JWxRzq3jeV{XrZ!=AlQ_>L<Bn;d9F*6eMmBJ*Lhmw^Nr&7xENVBl|=b(p?plnlM
zduwh`oK8GOAbO+m5$>NgSAJ1XGAcm6WzlH%=P5&WyUTQwe^Sex6iOux<|oHD_xwU+
zCs`Bg65Gz(s&B|YAM*eu%#q!F76)-KhY1VQT<JehDtDFe-wdO%;+A&U-od55dKC2@
zcsx7T;4x!dhCC8z#CLf=Opp#976|OGWTaj(in+KN*AX4>-K=n0ww-sQ%*J82j`_sb
zH&#W|u&m;4Cb$BVY+LGtDtT@67U?v#2}j4PosuM|Xj-BOu}DY<*wk|br~`XR?S(vB
z9LB=jclLxRkK4qLyH?*Yrdv~OVHgk*?xS3AzXw<VA{^U(-&P|She!LF=C{{9_RS-{
zes{bdIi6#7Lidm~IcRYgNwpI^)E+oVtnTFxLr$=`vArQb|C$4`?V;BV#yTC)31Wmr
zm3?fTOq`3|5}+&>cM|V@MjxkLKmX0@Q_1yiG+BC61F0G5MhgM6>^ILzQn`n0raFF+
z-PoPx@RL|~x<WL*C>h?|W|H{$Js2eE%jU^@Wh!zGvVDBrQIZ2746drXQ4$mQ0F{pN
z&&M#mESyEGM<To+j!ZGUK}EXstBls)62{_{=!1!0@a6mX+u4?u)oC;KVNo$a39<<8
zBdRi$?Pe$ZD25daFdMive@F*kOttKVNXVe23KcFMy<@2SkTim!)}4_^D|IYDyXUCm
zt`!xJ?#bi`?o)|Mp=>lE*>z6PE>{NYEZ_&@z444-Au17RKcZs;%_qCprlX3ao;XvK
z$&(a_a&78K5;G|(zfU9e(e}lS>VL7j!>hCum`6mZFmr_vR9K0P#k{xpP@3@F4Op``
ztv>{fUdy_n;!!SCq3(s|Lf6UHj$rv)uVaEvhvhmz+5}jUFGWUa@0lo<`R2ny9FjuI
z{G&K#e5cY-AChv&?@U|SY3^59yFS53;>Eu!(gro%{9=OL^jcj1a|p!G24vo1RI#bJ
zTe(ps8>s96?4Zq=9FcA;&v;<-eCF0%&EG?disVRHTzF_OnW0T(XXv+n=i>{}kZOdo
z3Ej=MQuIyF*h`-%xV$u#F6fJk8#y><y@NnCyRk4g^?LK`ph^OP{^~xkzWutOkQk)^
zQy~50kB*sn95IEIEj0$-<Y-!TxQeK4>`Yu8oRP%EL}(U`;;(XO`Do#+hE9ih@DFYP
z0pYs0pjj3%VLG404T_`u(_ZF#woEmF6tK0hYz3*ro8v*JaY-p<74;oASb~!PA@q0}
z`06W+*s~`QLE5yRkPp89=j43PSXa*{Oq^ud!4-siT~|-C{FSfoS2;Ug!6W-FUCC_c
zrq?rJbs2syJKGQZX13dmIFP<RFpe`9hd}qFU+EOA452a7;j+qsY7J)Jq|%e2Z_e#`
z#FeVhWqiUyBJ>d}f99d#Cu&X`pTbs4HJdjHEt_lbYF*cH<%5N~9+R{FB)pFGtChDo
ziaP3n+EkTyyl}r*wgP^}|0pa%M|Xp?0xiC>eyYr9%-F07nQFt1x)vq79$Zb?C`I~O
zq~^BSomS!zu9S{|9;j8{;gg7<(MimZ(ctJ$_})#Dl*wyB0V7)%&$j}c9|}ilFfB|a
z*=yntIhLmni?p*;F!8An{@7XBbGdN5Uv0nyp+7j5V$dL5#{5?zi)+634p)t^tv2cB
ze?d>lxQRxVRm&dKA&oln90xqqluJ#8C<N^$QQM&wQGR9bDnfNJj8l!YTKAGc3sFrG
z88L;K%lA@|)#l82#}k)-Rw|hx;x5Fi=cRV*>;VzHPE$VVU-^rSb3Xn?uaN$U%FpQ{
za*xmC^c_7~I9XmZ?#t7=q$l6ul9`i3n+cyvrS3o5u>{G*a%CkaDA^DRe>3ax24{^K
z$#bT6XOa?8ZWT?a+;7U;%S{{N=vnGCnP`eZB1(&)CnDU)@tnZ!e(j*Se~>Q&8XbUS
z(Kl&s-35zIm-6*|4_*fv@qNK>OZwN(saeRe6|bIRJqBOkWx4?pU!N>WS#lV1R_2$v
z8mZDq-=I5g;tI6%=!lz+0Jk;)WZ!u?Q~}{4jyR)seClCKP4w$drgU_gvj(z4f;l@b
zHth<ebde&_MMq-?2}vhLt7VcfbEBP!uC#~8?$!Od&XK(KaWgIMLRd$Vt>N&_pUk?G
zZOgMq&S#2*FqS=CE#0ptcNiizRu(sTBTWGP-)L$DP?7arTJy(?^^$wcM`~d;yHJ#H
zORL&i$A<s|<TK|H=6$Q`TAKoZMCGi=8~tM0Q_oLZx+?}Mh_7ssYmB#ODH??>k^~Lf
z#~1p5FKJI|a++#|8l(G9-TXkD-{MH3U!pgF?hoB`NO<bS_pllyA5g02Dw;ISDT)ds
zG?9~&$zS3!@PCEUB~TC4Qo2cgnp$g&N#kwhLNm^UO3fZiuq~T&d%4D`T~*yh{+3sw
z2>^-55|Q_%`IUM5dMHcW(%>R5Ew3z>8`QC?!F5%`_@@HG+_>laLo}tU$Ad(Kme<Yo
zkZ6=4n84ZB$NN`CZgvimvu78Qh)-4IldN8?BWC#Zi8=l83iorj$R-`{_c0OLzp>8o
zxb1cP`+k}ug|YDQD4lR%ixS2t^6=b5()v)sR~Qv;Qkqjnn6vUxn7_TBaI?);qXW;y
za{`?8!E<1UkY8>p)Y+%K&p=96@pAUQ2V~H~yz##%-2Z=2hX3S^|C7(-n6~skU%$Kk
z(6>MR9-~3g?U{BUR6y|1;Mo4#NB)=F`Tr`TbFi}ghjyQ=1?&D(t=mq@&NGkPz!(K3
z5iat?_=gq*Fi~EpIq<?J|Gm3NKC{_k?c&wUCec}cE~TVV$O=A5rPiMcY%jdN-Jg|s
z?@0{wFJkH6=V^L|d(*7z%rSY}E5|E8ecxlkoZOxsqSz6<ZWa-&)J__578-sJwO~5e
z{`Ru8$-!{vt1U-YjCTwOY0U^C+QT{$@Ricn3wytM`tWU>gKydSvUlOa&<{4t8-gQh
zLBXw85@*4k#Ssa_Eh4rILYWjO{()cV=nqx;y&!@)+s3{F9{}I)A+J!*%$7T%SNd?d
zRh%Y178)%q%86C%<Q&x01bh~gy{c4GdKu{{>DZ`<<Pf=RK9jRxSZ6_c01r8Tk^Wtr
zWI>pHqlkJK1MF%#dcD2Ts1<4ln#utVti!(Ok{p5-qpXG)`IX86AXp3_>8P<ZC=Phu
zM#hnRz4#p&E%h5oIsyqcdJQ8XU3fKYu&5MRV#+7T3}EA6AQQ_;$UzbLo-afzBxNZe
z6=g}JUPsZgRvn!~JQo#BDFRH0sQ=ra%1npDr8AKL$Y^K)>J+j*9&-RK)5*c4hcxL~
zXQ|<$K}lm43Wr#E&HA^;8Ccp>M(vNcw;^<s)q3A9(n27rLj(@`cZk`Sw2j~!&I#uz
z<EUNjHj3q;fnfah&xJXa9&X>77n}R_U5#PCWujMOzg$5>_uqm8OrVoOrYNjwIO7Or
z5-f@=DyD3#p&ltX6TgiU(&!}qYGUK<8${w9_j-UIBWEz1_vIb*Mg)unyaqkrmb0Fk
zm2AsRSAQi9feQ9#j;v32@CfnbIrma``wK=8{jIrwevUOEG+Sck2lMb}+QBOl)kkDQ
zoP)MSfWyV-d!aXciF*`{L2eRwS}>m8842>k`iH<)=hNHZu^y|PSFFMn@ESBGOOVVW
zxcwaMBP;z)hwX7b#+lO?FDruP^$bDeCmcCwz^)yg!n~ZaW8+_%j|b}fO|s>BoiSuo
z7l?Vm(}DEcfDK0;*gH83#EbJVmjKz^rd+@APsc*v>_u{Y6Hj7LSLuhgFG?$9vt!$m
zfYRm6IECVwnPtSLgNnM{+Cb)(nSY~}*AZ!sPPbyiMKk~zpGLx}O`|vE+j;X*sPK>q
z$IRwSIX!qUs8hROaI63n;km)LlBp@9x=FNyw~{3{jM5K1$n_GmWh=Bm%*TUn#H&jE
z=>;d$yV~7nYIyPXlmvG2@sl@k*whqZj9W95E*dd|e`?CbA;TMKjEq#Os(OIWtvtFu
zmulP!jgjH{`SgzDj|@%WAdZdU;EatS;!I5;B2J<0uE@pOwT3M_5TxoJH|)L?U<)kz
z;CflSw=~d@8L-pIt#dOQM%Xe;hQ2maLmv@%z7~8?<)Pab|2fZGLcUzyL1$}FMYuOC
zw-n!cagpXFMq6S#@~-4T)bH!CN19H}D(eeK&5rhe>VM73aQO4cgse`wt7&eyuH(-2
zh?bkpTt!Y|5ZSlMV6qKs4iY6-Z#fhd2SUZ1m!fjI8U02BW`qFw695|#__@?0Ou6|5
z^^S<2UitF%*+Jhoa<yN(ei~iyegV~~l*KSZZqW~S&vteg2OCZo1TDx+ilsQ5Fi*Jy
z9Sr$XfLx@XE{)CQlet#?l99ib*UYU-Y-;td>i1W~liLmlz9)`p-^G-Zhz1Fo%wD?h
zy(uw=5b*;&1KKmea53E{Gor1yPCMM=FETxe{LtSt>_%<-5^8EXp_#)2;R<|piIM}m
zVUoKyf?$2uSI@8zpH~!kpWlc4`PXVyC>qH^K?T9}eR>p|)!`OOBu`A#mxY29qH37)
zQd;q}gKMw$-eyc#JF`IN(i8X#K%}IuIMK8x*Oybk349ENo48DH&w*k7PK*8fqcSpK
zSCp4MMO`G(5$Y~cKF!<Bz5eju0u0|CQheDeL~FqF;cn6W$4HOC@f;uh(klDCXzY_r
zEj!58>RMOTRZx45tvmGW{Z^tR<B#>rp^dcJ#BEI~k|qD9{QP@%o@@!@WlVhmoc5HC
z=9?$H=Qf7~!F^%$bJW<xYkNKhg)4&3k$au3ot&^H{&m4o?UW`L-YW5jkLVftp8Hn6
zj%U*64PHt`J>S6~g2J=awQ-~iX>&y1_5F@r)M(FE7|KXyKV~H20JJ*<Ir9Ca^a%mh
zbXKc4D&~BPsXlJD#u>hTvLKjmFxcJzl@ElC{{3-2qE2fLkNWts3iij%a77S9XUO{-
zVvD2NW30dFB&$fg=WhRsS63Z4Otv%q+8l{k{Ry+MS;f~{=zr45{MP2NW6Mv?*|NW$
zAY$-#6kS7v28Ok`qa)Wm43ijwHP%Jj@acZw<jUhNVPW3jCtW7j+d;D7Oi?Fb(o$x+
zf<@-1I}<8+D(gE4@loX_iVeZspy83p#d-ORPMmi|KW`s3n7kM>?|_agHtS1W=q++*
zS|&jQp9{o=EFB1^!FAv3oE_7#v3kJ{WnK@8(>n}!JGTJeeBZxb@EQ8VTU8wDuM(YH
zpj-QA@btTy9fLej<q8c2F|}_zkgv>PO^G%ly6M~ZS!<z(Qd4rtgwY&j<9@+7CZk&x
zfw(A{RB`ag=RtIh8BhrUf!hKlOmk%=6zY7q8uUW~0VfPOk<h!s+p@d+P1l_}63>=b
z`hj+FC)2>pm!Z^Q^!(u8A*?HtT+cdHVP(5RftfB={Y3>G=VK<ft>uhy!e}fqaf+;3
zRt#)-lS(4*p+kDn0)y4=^i11JbTFxKb@<xeome@Hdv@O2hOYV8g8T&Ui9g;B)rDOH
z&@GEilx%4JS&9gtls?lpVuu%{ZV^Fa1~VNH6yg=Eehgr7r*mf`NX6u^#v<fHnSIIO
ze_*VCW}*qgclUJ1Bc#G1t3(l6^lhp#C0|?AHAzFC`oATCz1B2aKa0d*JQyq-60>yL
zJ%1+_V^pc|ZA9jBGEJ+LQP2Kk;pKHHNLT^Lfzr*m)=Cn*G7;5a*?lS3{%yJmnRJkd
z80HDTc@k5$e)2vb*@J49|E6?6Gj^GWGKXYphtEfm<MFXw($$)F7E0X3FR?f3SzQv}
zWYYByTNvpfEuMS`RV_@QZOm)1%-Vr!7*Gic&K0QVdaq5t_5ur9F{NTiz)+6N*l3kD
z6cfS%$+#&^@l3EW%d$~_UiN=V5|S^6b5>!Q7Lw4!k?-@vHojmG22bl#4iV^!_&qgM
zE+OjSd1SjiDX}%nTeXk1ojS#UzMq;i?0evmAf*zKb&!}^_pSMX5Hp0s1zAb{lXtL|
zDp`(DGSu3~XDI`R2?@&&%ALzOdB3ia-Ov5{eM`|C%%_cT^j^x{pXX_6^M%gk;@<5=
zfMA`jyZr;rLqz1ym)Y&owQ>AoR>ZCUh$eJdD}v>wcpO2tNE3M2kgcoIw)QP;@6N2I
zC-S>2B}X`k%9jdo@`#{v(U6TeH?(JGcsuY~r|)sUT-IDhZt6jaI@ZgcQQmysZ9-t(
ztF%m+ZILs9pCYt%gFHUo!U7MyWajvD*GG4n3%BYIZsq(2T$8^*zc)4|ri-zjerszZ
zh=2Q18MEcwk6>CL3NkLTvryDNy$&s!iPkthyAqN{YEx1Bm8bm<WNXwH&mq293@7AW
zf#2U^?u}MQ5Nf1mqxb-7>Q6J#tdT`7l(hB5bK~N;QiX-_xH6qY6%0g2u@8HcB<b!I
zzZP8@Ri#!DZFj_}3&pV4Zl!H8W<<A;q8g1jP>MtfylRRX7K%cKRMYSj4(tpP(~iPB
zh@8d_+*t4hlAAz1ddTYhUf|B4wP9odVg-tJY14w0t-VxcuayPZL&$4;gheR71S9be
z%6i<nqEpoYwv1d$T2&P!q>Nm5h;x(3E)L1^MT=Egd50fJ=Y|Xir(};76kT=xN2w_j
zjAH1GgDL?THzLE<tR20P!MSG5FFEGBKK9YqtnVcepc^vMe1o`q#7G6zxBiW~TdjFw
zp5q781&4hGlZB3oBmxvMzX720IYGs>`S4#dA*;t%BHMLj=1|(YKR_=T@>wc-iOFN}
z!dJ%2)clSZs%Zk(c<zUmVVN>8Mho#!xAeS;{tXxTX{%9Tg*w=*f4~>L1e!>_%G-{v
zBkweN#<C!8L&d^FsjLvpskP#QhHa@hxaH|3qyoH?RelAJI@jAnZB@`3n`<xQH?~bO
zu>K2-ndN7r*oNn^J14V$x)o0pmG&v0Op3XqSdcS|O`E$=B&xTV>^E=7Llu|95NDoM
z(**I?F0X-I&KR15ALLCtVx|de0~H$mf6&m<$WF|}u3S%CU!yhg=YqO4ld9CPV5SQN
zlug*6;ik)xb(-a!w5DMl%!w$;rlv=$i&K&sd56AEguHXrF0OCfm=0K2rLV#Yoga`U
zhh^{Vy7&gu3hXuT@JU?IfCU9F>2;Cl-)c{1WXHjUZnB;7rJ$YR{Y|yu3YLJQpXVr}
zv8N^mzhgm6=e``1=QXcK0*H<|0WX3RCu*dvcGE5&B*{_V_q_}R7nU`e)ti#Di!tN)
zLY5<itrjwz%~@9mPaJ#GN@3tP(=)*`NxlK{TgKIm{cbItUaei|TVZ!vdBye%tBiDn
zuC%SII(~06r&BFX8Dhu|9kHm(M;hZQKj&2%CNL2b-uNHAiVD6Ro=iSV5V4&-xM}zG
zw;|wWTJs2T2@W5x^JCuSELb&l<;k0#94%u8U@tQ%#s0Yz;bbl3^R>+CG&p=V+sue;
zkky1JQMEGYeDZm0poQCZLoCMKM@813hMuIlZAbp{6!)*vAPe|ZP6ibsL+v-S<`Z=~
zQChd0m@Q2sp<c&dZjwj`T%L(H>=%%_i67>BT};UN?Ti`Z^Gua-l*`2hnLGSBsWXQo
zhvN2Vz9ucawo@UlT;#Ukt+wZpxttzwPcC%G-M!b<R9G+HQVJInFb17c!K^ynSd<kl
zFQ1#=S~(OD7mZ>C<#VJ;U`(=sD+Dcd7R6ItJ+n!cdT6MQe4B&3uCdO!vMvADY-4xA
zz2$R4#bth8pC%LC$Zd(5{Ud{JxH_~zib|;M>VjhI%2W{=^3AT2PvBNqZ@CS+<8|fV
zkNJ{Ju1A#*iX?Q3&-h8*`nMNbRRZy#zx~e&S?v#8tID;CZ94u><N_MK3B3mR1Q2bD
zTU(~}OE&FXN<5vLm#=Uei(zEJqXwBy{g21?`+Gp9p_QCBa5v(n-5id(fh@|=aWAQo
z+p!qmkSUXZQ1wyS26+*0d=UkbB}k?w@3PpAzA%Q>(!7!;k6+PdId$f2)%sCo_6bKD
zbhRx<1Lid|e}y_^F34TN!C0%F^NAiQe1GId^CQbyITd?pFwsA&taV6dzpz{<X}SCp
z#Cl1^8N<#fGBcxd5%XOqBdYSmxq$?0X~v!K&mDeG#K^C+Cy3`Q<(z)TtTlJF2mkB{
zt=ruj@cYr56SDyT)3b_R16Bm#6V_7&F~%H_RSG4$um`Mu-|h_e+5gAlaf)d%xyhxb
zo8_|p?E@0uOD)7&Js!Nw&#tj2CQ3B^aOzbxgs}X!-LA|Tm=QM2cxnm{3Sfb^S&rnM
zMcE!-dHyMTL5SNE9g)77nIk{7<Xu4?Vr=uW%CEP?O%Upm0!<8Qy(|!fkcd^tMj=@{
zIIMWru0W{kP|iYy4|lj&dWUl1zq|B73YY?>#uE5IVMClI^Pu7t^$>>LoSJ}gmsztA
z2!0yjsjV%iKl3k5qTU=S;(O>8$=s(n7U^h8YfZLl*`0KVMoNR{fG~s&!1^|AKuab=
zvo7PGx73Jr?`I{Sd%K8Cy|jukSTd<Lf_@9u?)3HO<E(y8PjV4_q1V4jy!?%258c;%
zpZnZZ0)Vqe`eKnRxoV}N(r5VH8J<O3HDg@skk*6ad5&V6-_e_3>%=a|SzvY!QPuba
zgHG5TBa28#fOchhdRl6K{3Z$88Edk0b_<M$BOxSt<w7mW^|T~x*oFWFj|!B=HQtyV
zk}Lig$1?GPt%2K1ongsuQOx%F*?OWZC5ycz<zjOTL!eThEADWjV0afo8C;CeamXgE
zz`|9)#EgsUvqj^zHZ376Kil*Rpx^1{sBe35VIzY7SRh&>3!R^9nr78u&({#1L_;UZ
zY@r<eky}c@3tDUV<i7t0W0-v2S%FV^$Ju0eW->HYWU+{^Ul!}Dp!Dl=6H6D9d}7!6
zX_m`^%+J3RU9t}F7nOIO$#xjlt4+-G3^X3Ri2DTU;*#X_)`!JX5AzPG^J)^Cs|l90
zNkhlBs~ea;5_jky&g)S)5A=AGE8`aX)FmjMkl9Up965HryxTuQ=#468r_$HU=8q(z
zkqdEaPIBSMGbY*0{D!#5R<V#)JG#LZ_P~(S#+`~;q}WZ<W~}GYSQ8;})VXaoJmS(7
z*BO2OU{yrBNJPai0YIV!kmFH(^h&6v%X~vff#kr$WJf9@i0H=dq=nWDhW%)zz~DO$
zeFMDf%1rNV@@<SdQ4JL;X@4pozK{~sW`GdSxQLdiqfBgC#T}KwOu93L&z2SZ7_7;j
z=N(t{jaDs!-X`=nBH|L1B!7sQ(Qw;e6~|=Hb_70-AX#ep(2zIa+Fv7K!<0-`y47zP
zXB)#Jj>P^#7eE;G2CQYI)#wb-4<(^Ua(_vq@2(eEXY_4jU4rE5T<y57e%+DeAdwoo
z9n`J#iL~Y65rS@RqmC^t;`jkMmv&|2an9MJRDGl?!kIGiJZ8&i3<AVUS|fxI3QUuZ
zJyj!Qw%(kC&9L*eiS1(k5;<vj7#lNfT!ZncRoZ^)<{7p*W?QT1#pv23{^>OOB}dzY
zhYH@P?8`%f)86Ht!v=*V{S7?r7U?^vi&41stoyC=KM0Wvb0%%n^`jq6&i&<drHPe}
zOgd4SN<&`kk~{{S7-lL^tKzd6s!81B0esUZ$!pFuBBm~e<f=l|4Mj_P_HNv`)LD<l
zlQD0M^schf_jOB9Zqh$5N!HkL<(tNX3j#5;*H|j=&d#5g@RMN1HbfYCTch2MS#k-+
zy8#RuDmMazYqwm()-5aL1&gnTj%Q1JkW+6D^*K!SDEzde>QB6-+HZ(eBs3D4dKlm`
zjYpc3)RTfyqT^iWGbkK5i;+pmnM+}F3F-P0>x~oYW)AtW4#%;*3rfebA<{`b3wYkZ
z?^D{hDEE8jS)XRe@0yo2js_BXa&~zzxYTw{on-w?sE(PKLqGsTHv*A{bY92X<rA;e
z!!POV!a%H?ty-VVoOR)x@x$Mg`NIu5n~1qYXBS1Vx&;NJ3MJDtMTu$eyTLVZrH5`C
zO~JHie8*r{Z&_d`ax}vjZ6W9flO{AU?FH?D8U*M*MR$wJ4h&)IY<`pqL7jAvXmKoo
zAgKIkrDN(^zHlCOf@JM^G`2LCpu7u5DKu0<W7IP=2xt8;@}6Hja7fE!v#7}|Nnnw{
zOx^74_@c>V+C5@4j^qJ$_X)4^r++)myo$4W?s7eT-gVC3o5B**UkWwD=V3fwGnG_7
zOT`KlRnJE?GY}($pgyzQz$|yD@W1kWN5&?fcd(p)RrC8JZJPJT3Ui{FCfHwZ@n-i}
zS%2Fen+%9|9wwahc|T$BN=(8s<b#Nmj&qnV!t!vBXS79i4)(@aIHmT#FQRcN`?>yA
zv;5pIu6~;P7Mg^<Z$UFOWDA}JZZE~pB)BRqiwTSA3~lj`E2)VM9xV+1y2S&vFwdV_
zlLg)UKu4;8&07LW8hUnmznA2k!~)zvk=vF6iA1++`z2i#gmnf~es3M^>r0hlS5Z{<
ziM((g?E`vi3cNs_{ctD^zYr^jCac4^9yn~iGqDomAry+)IbcUE*(m(4nECqxex6zf
z6~c3@*9=xCy>z%orI9sf+&G}YcAC^<zoAy^DY`*5^jQvm;G}XJ)jXklxd=#;e&9eA
zM(AM7u06zjxrvy<BXx$nX4zK~RY!i*k=~Wow2fhQzAhia@m%FD)1G|xDhT+-Qw<`E
zg8S+#zLK-;JU^n7z!q7j=zX<pPj`&<OQJpg`W7X9daFUpM!p_=-AH|21-%-&M_jFe
z=fQ&jLZECZhPY!JEyHDi@caaQtj27;VVZo0jm=YX3fhU~ymbC>zJu5-Njy#a+CQDs
zzn*W*ij{oa%kS!J!PC?S-~JKccjebny?DlkHV`~oOU{y<Q#pfYIMG)dgD9!t%y?Wf
z8S4EhOD#R|06}t9+&o~&n4;Npb<(bMUEg*p_RWPKeG<u6cewF%-{`8YNo%d|{_Jfq
z5YE$%#j{Gp<FU+ZrxLWbHb{rGVpU5arH2AId4!!4z>hPSZ;|OnZ`7{#D(<k}`>{%7
z3m3e>a;x+-mRG0v`ua-0qwdoJ^=?UM*7L*6b46T@3ui+QbkzbW4&e@29u`*<pA9Au
zj2GG?%Qdrl;R*)s4B^5#Y5f8zb845W84@v+c=sDL1_?3#zlkLO3qc}bVeM?<$S7fL
z=xicpVr&OAVU#nmHFGv6{!ciz)hNAz*}pJjpb?6>K_0k)5%Dm`FvxaNXtXKQzl|xv
z?|<2Ja1iWUXgIPU@W@ho_mqf$5Sl-RpddF-P-Xw^68|er`F~Yv*tog=Bk)-L|CAb&
zb^($sr95@5f+u`_N;*=iq$xv8$hvyQkbfaLueB13f3zNpH8%OVqNzJdi)M<`tr>os
zQyGiZ9|85|!r}x<%?`N^R|EvdiboIA8;Mw$KF$noSMoVo07~xH@9)p8>wgCs9ETUn
zkO@1H!)}#iQVxYqm&q;&KoXRS!PmC<o%C&iBJ%<}sL1k>#r$O<_x(^exEVb;=ocE3
zZyLhlB{G6EqT-{Nf9i*vr$`0^&WFqA%LA)`kgd=I*Ah03kn_UzfCB&`^b1iC$C4@%
zE0Xw$Q4MPj2NR9xB+|(p*h(G20r;e^3c?}b^)zWQ4kJCvEzqV5AmOq8A`+%L=|${V
z%Sr9QJJrANDFhOyH!ue_kOwxXrHSSqI+(&kW{KHqi%7spru9c0T^#|<9EP;ARO^^@
zlGrH&(X)pq_6}f%ypcZXj!f;*>s!aNtc6)LCgY17xX)cLe;;*HU`TOgdPwUEW2Lg%
z5r!oW9Xied&}JiF@5xuoQAHfiJ}};RweDFZdXm@$-8Vazn{A?VT092UMyq+r$N(&`
zl;TuJ04H+T&fitx`d}aX@6{p>rXo-x+xEDrr=ti5+y@>&1{UN9p(8T*c=1scAY?t7
z2aZ-yXJD1Mp6C)8UcCIMDG(kl<U*_<ep3n`Efzo=0fW3tmL3j+N`@vjUbI_c=M={+
zyfZ>-^4kpCZ1@7-DTF(9%&hshkLiV^Igu4^rC$~uP-JxW_HV2pn_=Hz9lA+A9--Fj
z=(6FVp`W+v*Gm6L0mW4IK_*@1$5c{MIctfYQP0Oq+DJ$43O_!fE5XZ-Uap@$(!Tc&
zs;Pkw=6WIM9d=h?pU@z*X4$fN<ETTMLM07pIqyL3U=MzcENgJ(?_pj4Q~3~vg6`G1
z9jnEb9d+a~R}24PF|2fzDS=4~7hc+k=N%x%xF&-oWJY4~Zw=}Ixl<^tf`*Yfms-*7
znDvdk$RfRvA?wJAK%HHAHo3#czxOQNI7V+>_D~(F;Eghc%pKt?<#9GgyYIrr=>)1P
znWH3{u_;NQ4VaZv4Hd82dUf$9iMO;bTth_^N>1fkvQVUgCtSK-Rbd<FXGw&EyjJHv
zhd{N|cd*_v7Zwiv1`~0%Zbhf=!&c;&8XhsEC1m{TA>`!|07&7Il2|A~5vb=ne233(
zkSAzU18$6ZyJ{lMbAW;7{$mG`;k8YV95c)ri|n5QjU$l7Zi%>3XsGPKqMKd9q*Cg+
zssTv>=Ri2LEu+3udwl$7f^^X9y421Ggdx&0O06IHkeBvr!gxrOwYI5$CWs5m&M%JD
zr}KtU7L2P94C>bDiYf3s5St6_+0SLyub&bPG29?%(j3D<N4g-zubDm_YBN7uY(|<L
zJX8c?TcOsK5{Go*n9SnE3pMbfJ&l^7)Ns`#a7t((O(dtZoAhcx1}_ZJ*v}RtbgfVG
zU9Jb}?D@Qr1Z0(BRnAw9OuP?lb+h!|`wWl!Oh-vCU3zu((C6ki7Pms4Q?GU?!^<K9
z<$){l*7^oHv7osfBM4bARYsXIa?NN6*s_ZdGl3$QbsmVG>;e@=$O_~t-ZYsiDidEE
z=x9t(ECz40D(qAYb70XqMe3I&6gJwSjK93G6nZp~e{2OCk1~dKW52+BckjRc82udb
z<}YD@R8JjJ)e`wl4BBD);Duv9$)08(cB=RuW3$-G##5+VO>$JNRWvc+7V-W!+w#L|
zK>1E)<TmKDwQ@U7Q{jpl>=q=k;{Y1>yz|${9xtbItg^Kw54+XVE25wOPrFY^?$7sc
zkZsZ3=R^`X9Ffs9N9ah#x~@BI5<z}eA3rrc(T<wxkQeN)93LS)GC+y-vB?c90Fzt(
zvRj>qpv+IHve<t48R5;70$nDib{zxp1K^P1?<e$2=jSd>9oj47<!U1%wV1Tiudq{k
zZ7)o33;{4oG()!l#@;;G)-NnprQPk(P*(Mif7bei<S_D0NUYz1yd=n`i4*NB>yQi0
z<bTn4pSm+@(Q;owT8MWAV+fts<ZWJZHHo&vi*twvd`YoIj*ZXlj*Q!?V}e;fM3JRi
zs474n52J9=2^rLuWmj=Bs%ibm)d-+}_tVs?>e0j*dKOMAi3M|?CUxy}oCnGJ{EiAp
zO%G^7sr@CBSE?qfBBdmzx+Y=CtPsLq`%qa6+ckn)KOwGd+$|vl79<f%EgLXp!mN>u
z@^An4)xRN=M+@R<fqGjZm(m5M--jD4(Kv+)fv8|4TqJO>7N0MaD~=l#9r9q;DkPm=
zUNkO;?;rnfhKG^5-Bnf3!9BoQ5uW1CqlOSFoS>w(ZoNzmxHqe_lq+%MeWk`IplApo
z^h-J=Wz?1aneBp!<@QOl$XT&<c|~*C!kMsHmEuOXC^?2#y%R}*Cw02TmYu;I6{+j)
zJhohT5EF*<#AeAYMOj4GT6L`>OTFbJG>B(t6WH=rqJI>0@2h`q5^gsT)sp_T89;r}
zJqRA*G`O(OM^92boQD@EH<?eStDpz$V&T(k6VKKJARb%58L2olO{)AM)u}RZVNc_1
zQDwk7E*g50=IuETp(b$EE~I)uS2v+9E09$rY>QlkOU=thS}sT0P_~Nm8l5x}S4T?u
zRs5%{riNQ8_>+Ad8dKn#_1R?@mft#TT3ZTu^YHPaDa-eVI`4R|;CRo#gC<%6S*y|w
z8#65isUmJv52$1UCz`Ndn5^r4b=P9}(#F&JVm$is<BkQ!<wR6!^}0I~s?_iHOUR5h
z{qUir-M%CJ2-ySUN(^~rvN+5kHtLUyGlYfaE=7%>r&9GvRYq+(BF<N|ximLZ>hA@X
zVllEA)G0XSpL`0@#ly-C<rcCrRqs1u`ZeqFGO0Ar7}6n~H$KwR#6*mvAPlHl`}WJ~
zMmhawpKUmi2Sw8x>wcd%m&lc{+`scDHkQ<8Ul<0${uI*Ndtaj7Ie+ec*?t5%m4`0S
z&Oj%8Os`h-<<512VBfg}OOa)uhsZ`BnmiUG#_-f1sWRewA-x0EI^2A;wN0UJ1upzX
z8%s@*w?(6&3gHZB%zRy{OXo~T-ga;+m({HvLwR}Sjw+VomNL3!mv07+G*45wEDJ~@
z=yp!wi#2tBp3O3@@8Z+aj}R0lyrlT?qWkJixmpQLV;Q)C8Zc5*5?@`5=@4)nt93SG
zsCFz0Nm7uXprh~8N3^PQXLwvrH3D(remhrtShoI<U&gRp+Mg4NtE|Vn0RAPBwX@6Z
zhHjbvLy-nchT^06!@t}tHY@!OweyoT90GUtUb$7m(9-&*(Sv7f;>M)0arL5Xc-sE_
z=h1B$_mD7$hul#;TEW!n%^~;3Ax#Rf96}o>y^YR<qvC%kd#51Lf^OZiY}YQ^wr$(q
zW!tuG+qP}nwr#8T?U(*f-_xh#zOR`1l94Op%Q?r`yHMs(chET)uie;lBpLc6TgVr!
zlX*LR^vLwKj<RsFqN+>P{gA%CL$&*c!aJh2%Lr@;6uB-i75=2mL+6r90!OF3vkJ#n
zxc~~IQj~h);}iGE2`Uf_Q<_(4%tcsuKh*T`7uK9P(;w&dp=Fir$nBqhWv|$|G?CW#
zc^Ev`Q-Xab*PnI$I>x7Kiv>tf-aUl(C5rKQ?D)|sIr+N^|B5!QI&x4X1egW<N`{om
zyc(1gp7qaq+qwicd1bZrQ*ajFI*%Q8tqVC|dLE7Jq8_NMMqq1sMI}Ki&&%^$5HW7O
zVW4(S$}8IM7TGV2V)I&NmdOD~ln_@Wom3$`&D=Rk5G}mf^4bQ`!OY-Jiv`;LH6>P)
zp8g=T$_~IM!eJ$}^qPw=&9F$Gx}4dtnNbe4z}xLIo5eyMF3UxP<3s%_mtu^|GV_XO
z*hHXl>N!KcRZ^Gj@3)A_9_e*kGd>v<dlmq@==i2c;_7}|&?@JxB_T_4a&>9CgrY8-
z?$$Y9s0S=qT(Hu9uL(d2Tqg=w`-Hkso2c;=P1E_ckyPjKs)zh77Q?1ZwS>eCJr?E!
z19V-oG6Mb9Zcv7L;NWh7yGb6+D35CAr0B;eIQo;h#XsvE_+<jL`q+$(l$u&?4{|H`
zzmE4Tri(RTrGN{>wrBbad#(x--J_hgHtB;3(h^S8-4?cQ8C-YiNz3q2B}d|P2gMZS
zIuyM^bHfMCa7&ftQ7<=Ogt8-#bXnP&650F^Lc@AoC5{Knt>hOAu6bbHHae0V!@jE;
zSdN`7*BeS+0GHFqCLke#_V|Y{TiOqr?{;3y{zyPPraNQ3`R|X!n$t)36qv3yfH7s+
z!bEO!yCd;>!N8R~f-;7|%rfgA$)<uavxI}CB0NJ=K3HaevuBG04~I4d#B9a8lfluL
zYgeJ4J2kVOWh3%I9g-3YDoBo>{)kCan&}d|$W3L55PQ+XL%i_V-<83N&j+@aaj3B0
z(X=dpl8KYJj)ot2-mlaLUe4{jL|0wI59_(nACJ#vG@rU-COYeOP`p5^j(Crhuq%^M
z)7c|iq5{xP!`Fjxik{aPHhz;!bBDS|GbuKoDheAe4g<(e%sr9~VLn+`dmvw`&zFw)
zk?SQPw=?|XP7j)Gp6-tu#EIXydJ?B@S|ad#$lyl}JI0}3pwYxgtaXlm(`EL;4jZ`Z
ztXUaTm;uqRnVp?gmSiuB-bl|K@H05k<E$O#jk~rRFC(Q7AVVC+Ra#$sKz~7wGzlc?
zLkw9kFufZXxXZpYE$g$btXi0xTfi|6F(`Z@w)_71^4OA;<8bUyGfZpcxWgB}SA_qO
zxd`yi?h2M%u~9X5v5yeZgIl<0ci0~rZ)n#*K)KMt9@RQ?>!8>FVxB7fbz5Lt&xOXg
z&Xp;LDe~!W!LkAc9Wf}*SxZkHpY3CMj;@AH;gkY4=g{NPj?9+fYU+&eTjUD4>5koY
zd#7Xx&!hCEqE%Y(!AQrIQhTixJAta=Cm7&#QtVO9r&GQnR{3;-!eA|SoyPxNxK6pn
z8PfKJ>J&$-w?rct+al5koi&(WUDLug>Ezh5vISj!JGSlP_BhW|hT`i~kOjAbv-U^%
z5nCr?>sZj0+D@b)lPyoET>PzuF|TLini@h;kI|I=j`K!qWd#(0O)pt^j1;3ZOs=iU
zggBhDehwf7^D=}OnfJNJr#51<^n7haGRR$EY?UyXwnNPnayxdf?p`Cq^Y%EcEOEAc
z!D{ZX8YtuUZKx;HUjSxpSu;~cL8|APcTLb`Io@N2M}IQZj(UA|BwQ$YMNPhpp~&G*
zznxP%32FONbERz6Utx+?{dfC*8_=?FSR~D=tP<f51*`x@cz#>izsJ1Gdcw*X6Sj0I
zoS@0=In4{m^<{UCT+%J0u^{v;X~S~`_uURiqL7$H4qSUwS)a&rUrgG6LJF1SGvc5Y
zWidQeH7=*ddZp|u!Z*%n@#u0Zcjsi-n;!q68`!0{p;N=Z9R+5Tgm9Ig%boN+X2(-k
z<d={f_5)CGz*QZ6M9C132CX?#-e|n5po3s$yH6N2l>5Bsi9*P^v@bt+QPA`!7GZ1-
ziK@4KB6&RqPNwIa+%Y?lm}M5O1aXN}8rgO)z*;f}%}IY97@N^`1|~l9H9E&NrJm<l
zbjj58JU0?#e@?>@X(oG2UIkmxToxmBc7TMs?M6DKHJeg@_A-w+2^HeIe0XT3U6RHm
z#4H+^fYneiqre{CwqHHHRS4)0c23a<umTTNv;T@r|CJrKnW`Ut-DL+6>a2i*NBx7c
zm;!nx4K`3hO(Aa(kYuivY+<>W48P}szTzp|6y<ReT-?}-Twf4W6_ge?R$QSb#&N-#
zX%4(|RCZz3ua)WWa526Grt1O^HwxRQrtMUg(?vJ@2tGN>_<j0w?6T!qhsLrgtqOgz
zkaTKf0g4HH54EhD1q~Anx-XcH&jc4s!@b42_M-8Pa9-A&qGC9@=zWi9zI0WPdsRO>
z|K>c0d--ZwE)rqn;CuAL!mtK?3Q{W9+}gB+*je!rQQ_Odb2rF~4or(q%>_N`JMmQc
zoqrW5nafX@A`W$>nVC|yX}d4Qu+Li&gxM<P+;y#e**)?KVHsTgm2VPXPi(r$Yt?Vu
z@BVfG@Kp;&SoQXk1`W<*dJ8kqoXZHT<zu)+{PWZ$Oh}rU`$pO$w&OWwUg3yC3g3^W
zklET)Y6L>(uPS`l9GH-=nXgdMr;IWAHx^`-rxdH^lb7IXc_lRS6!(InPB-WtyDNlR
zaBZWuc1R1}L$)c7;+Uu4@hw=W`?e*HMrfP<;7_zxz=YPlC;O3cRtIeCVn@oalHc+0
z(5<j%0`eVp{HIZN<?Os{%V6nzqH(lo%wCE|UKxJkk{`5BQaQ4Mjeyg~-}H>Oxb9Vg
zk=$(&9s-PXa(Wf@@y=+nPS3j&3kk&(2U~q83JpJ-Bq=qU!Do#;>R2m#_RGn7u<~kt
z>XZ^PyaX6*>r0DrVwWXsC6>M296xnfnNgYk8Fy!d*KuE7Xfs*a_e9_NobTvdei_zB
zx-?f16tlz@>)&2`n?AMEdxRy2n~XM|H>?O+D*E(m(gjmzHdVc52Q^xT;R<Qm4UEM}
zuE*`u-JkXx80eq-hcrC+>}Tcxd^)gSLG!Gv>*reMerm_yR1s-(EE*UbctJ7`vaIa0
z=D7&z-A*bgtNeVO<R{C{_9>(4m(>v716OHSYNKDaX&HU9|1Q(}4~W_Scd*F#U$VRZ
zyT!Xsj+uG`dHxOG2nB#Z6V&&AugysRA2|8{<Ghfanc+VHY0>JC+DfAxd<;h*hzjgE
zE18C;M;1CBDtVeT7S{fhP*B|}Ngad@#8vSK;`sWA10Dmo{}8E0DD|3_D5()rXPDI<
zm(?#V$t#*#*1L_aF*u)|wgZGuFE2f43CJchQ|alqpTE7gzaso=>xe-9eeQ+5y@ay&
zU7rUs`wc^B2*uU^0KSpFA&wy2_-?9wQe^kOs1J?TFfl+)6AEU|O`h06Rk441(S>Ib
zg82ESUA~twI5}oUetEoS?=`eWXJZ^8KP7_v?C?tv<&Yc0JVB5kKofJTj!;V0ux_wW
zaZ4kVN$BjIU48dehI@>Ri==B3Xzg8fJH}je2%cS%KZ1iqsg@w;;}_!E=*x2}mv&V!
z^W+bA#Lt05vYc$&xQR&xE5hQyLV5{gq#@Ku#f>FJ4tj!4wTt~9#z4w7^P^G@Y>nNd
z-NXgm(wG@&skJFo#s`ZpTdd%cKFAN07oy|y_w6C~7ltE?7lQ8W%C(C(S{J5elF1!a
z$c=3f-J}DwrpEF>*fzU{d22yIXm`#&FE;$9Ep=2dlbHqclR|<k1qlt2w11bI46gx#
z8Ci)#`f4>mq{PEHhO`ZE8FJP6xy7i7Q4u2{*$g1-lid8(MZ3h^5U9dbhAa+H1`_W`
zy~3P_JO)zN_^pV)!T}DP4UiGVtq59>Y!NoYc!vHGCWYZ22@@lV2qQBHQzMF95hVph
zh!Z9RuVKA42dL{mM&iUvZ6OIm^8VS%Ey{S3KckI8eSyZ?plPy1mO(NX`6iYSO=KIs
zyjEK)UFXaCdO8vY=L(&MLxJ<nO<*|AH*7jgRVr3GEcLlPI!W47K5;V4cx~!Ge|l<O
z#47zg$h!REqxICTY5S74F2$XppR;KK^WZSn0;0K#7(Pxyv8YZ8VRN3(mXt+n=q)0L
z?nA#%rupTcy-A;RA7wSyvfPaJ29vj;XGR4Ql3_75`f(vOfsI@_$v59bl`MnFef{#e
zz@on71SFqLJ4pl$O@A?|i4>{)ja0Rl&>#1!kd41SX+Zz&2H(B<ZGfMlG0Ar0dqc5a
zNN>Ve=UDg13yWSYVf<{2QZcu3&gh^>STCI`DN}EzrDf14s9w&Xl#-I1<o7hqEDtFs
zpOvmPS**ZHgkuEXka4E5GPD+R>>h$7^(8-Y9U$+Z-A=OwnCNikxo3%MV{fde>ZlS>
zoN^8VV*XZ9(9hf-gbR<q439V{U_6BB3BCtFybX^kf6vNMC?&aM{zq{3K!y5=2<wel
zcs5_Rkf23#1?fA_4SP6*_Btd^*H)H;Up2fUf1~e(RVO6EqzhfMg#Y7kEN$rQO%quS
zQ>&3TM)<Y^B~>l>7u?W{-Fh6}cc-3<gA|FM&tA<%g!qoMS(28tCAv~SmXDTXSzGZ|
zpk^LMFa~$>$|P^^^A>;jgS*+TdUs>_VUw(1<$f62bhp!`{}Xf%WH~kXE-A|ADWWan
zGI^52o%8J(#h#Pw_WIbP(l0R-G^e2(RNgTA9+VEcfGa6P+Rp^VKD2|_E`2S_mAMYN
z%edy6A?<o!x;~vtQ5g<yb(i+aSQ1N0s}rg1=pXgDg*ljJ?4*T)cs9~FOSS}*bG-!!
zC9!%&Hq4r)Om7n^55Lwfl1Gyw=GY}S9Jr!&=3I{Pxf&g^cN4UW-<1^oB@hHSLphC4
z4_+;Nv;9y2C(n9x#DF6);2Lo^PHRvwZg#wY*espBiS;85$|`1NH=Rm=_kdkC<m3c+
z^Mb*o=BDx%({`mHMDa$s^T_g^pynkGatW8y+zf{|!1`Ov@$NwK3glT86Hfwqu0YIj
zkFa57%6Vgh`hleIDSiH`7Hf#1_!A!JjLYkGd`r0!AqP4i&l=qx6Sqdmp`M{SD?B*T
zCp!>T#_ThJ$9UvMAHd53QM-8{h=2eO5Y)*NJGFW{w{|74$!c1b&g8u^wU=WZZaB`|
zBXQE`I3#3AwWPVc=I{dBhV8)%c3Lw6`*tRnHjd{_2<LK0!1N3{8`O@>33ULg3uhZ`
z6BVZ>Dm?rw>Xz5)<^t%fVEgvN>e0&a1ajY-$r{p=iCq~53uh<)ih_pePYPi6udKC4
z&G%2&5q7h%TBhRa_hP3V5;`O@-X{JP&}Lp`>@s-Aj9tNk8&M3}y*B0AbZSEHAibub
zi|{qewSJu5KA~{EGGfNhv#t-W+a<pkS00d&_64F{GIP2$!2n8+xKekft7li&Ni&tu
zE3L2H`}fbZe@k+^ly4`9sk(NVx7S+PaYEw*v&Wt2L5ei!;ko;T6b-aoq71Ay9x>aO
zUyh&>>WtW;dIdbkH#_`--AQ=uv>t4hOd;5V#q_XaYC2QkglTk@QV1xj<{oY2<mL)L
zl~q5Q1!8T$&XaR*<X@S+M&QxUSt%QuUjGvOQ8Zo6_Va9O-&pV^l-VL!-{+*S<IUnV
zNLp7LUwH04M2|$@I<`EIuJ2g*EDZMdSFBT4kb0IgnT=7`M({jV_cu6o8nm6ylbQEu
zz$#>z4zJ4wKe>IKvmFR4UZ-5?=)I!{+?t5s-Z&K=ZMsO*^=DE0bAWHb60v<x@3o>k
zUX8`jTMm2pH!zlwdl`Z+KfI#Z8=9G^t;|l8r2HmAW|cp-W``B&LL$dvQ0oR~B#c=U
zysoMbZ4)&-^&0u&(#qr+$eWv)nVSrR)OK?~*$j1Tn!Uf0W1kPV3#qF18!B|S9)MQa
zbhs|1>6OO5!hw(CI@}>HST_Bt4K1)tq&bWAk9+t|8T|OA$Vbt!ZL)EeFqhU=4ErK=
z`JL2u=^}QUP|)<R279C*qp809a1VOk2Y&(B1iIb4d>$bH6$07Qm7G`6Qk9beZ-B)v
z>F8TyJU^>7A2%W+V2@A#SV>5@YZMj<8A~n<n}oQ38PYTb&b;=94D2f*qW)l56<7iX
z0t@m&-=OP%TD-+w&8Lb1fu_eP<;fbxgMAZDNFP2dCy*ERK=&aDza?%2%_|K)AA&!I
z4n@bS`*n4i6%pqn{SZ3{j;H3WzB{JwQKcHtwL59gsk3S9dsu!2YhPZXm7Rm=PRtf*
zG3#hCwx9{$03`t{lqa)Y8KBnQYxk7WR;Ugt0uc&jrlV8EuaZZf`z^@5ZcJBC#!-8s
z1DN(<z=QV!B&9ady)R*b^J{2|KR!-rGEzlTO&K_^Ys+v-P0x&#QJ9?&4+pahqu}?R
z)m7xSJ3c28q-Y+jr?1B^z-3fuMpnYUd`fCPp-AgUsenhW5D|1!u(O*<jfFS*edHZ|
zZQAC#zgA~!0TLq=1oFI2<=(nqU3(b1x)RZ+M{lW^Kt!||vl~wCi*#^SI{0Gn>kimu
zQ9Z(>N2w;(EbRTwqKd`W4dN?+@zzWcEwB)0!-#|#!341FmyfS+>gW<Mc$^zG<X}J$
z8d#HXrKk@K$dCEDk2a*M-o+`O2SC5GRW!7?bAzTOv9w|zQd-w*OYyS1zB_MZiC6_t
z42{-%^|M|&*jGq%t4?!ixqh9nUg{7mIh3ZjZJxh@(pct_rokep;j#0NhwrLR3f}c+
z3sHbG8Pp;ofSb<Y2S&scRvK*y+$0M%?^nDrW8IbhMcgjY>!=RmVuiN~;WC*zkP=M|
zBbaxzqsjssF4(d6_%~V2{t|D-p-CJ7FE(|+mSg#WS89_R=dL}f-}z1YzRto9+8%Cn
z!Jy0)Wrb`TDmUzjsIK24lZsvZZ*OQT6=f-r0pcZ1m%@cv!bM#hz6W{#OR~{z?_&zx
zZJz8gzo<1%%{;s<s9E-$=x<s|aZ3q31V2~I5{{2|=U~l`G0-)|>_P_d6pYhh>1i@3
ztxf(4kiw<3%9oKdZGzEo*9pS!rIKqWX(%b{8A-#c&=k_bG__^n_$gW4Vl<Zd^kCYO
zhbIE21;#>*lhk`I4*c26lU2YQ79K4jC9PYUp7CNyof3|HYYpcbDiElrHvc2sDyAKf
z+dtrJP~m%a7G6lhLg52Y9Km8J-uaU$@T)(TR^Nl;C&A8t{w^nW^?YjIn1M$HCEgT2
zi~L-48S%n!Ir44klU0Fj`^|l0MN&&2XLoqW9pAN~JU8N`ih^VEuGr`{)Mc20L5k!2
z1+Olq(b>K+w|nx?|I5!=arm2u6Z1<2rjPd$hSIX5?ilPbCN>CfmKGhfA|lQ{3oIjY
zbWLKnkj~^R7Hg>A<NF*CEL}3CCutBVB`JAHTs<CrFX=34<@Ak-jbv&N!#;Se_gw0<
zmm-5p)#~GPT}cI;i*2b^F5}m53&&)!A}*@3>Q2SHd1tn`HB<8%kJI3&<Clf_yiLds
zDf@AQ>jcPa+*H|Hnwt9iY}@<37f6-cb$e=4NO^1IFUuZ!wD%5J=;m^IDUNI!vuspa
zdg^y~EP#>W<!LRT*dtg_HCt}YuIce+edku(>Bb%WC)NOqf@W?S^(6%7t_4N1dTGj+
z2CVE8=bdH07uN0C*X*<V2oDWBX)@@c4;ruE6<Vi~PPS&?<$IL&jm(V9-44A${sER=
zB?V<wa>5bw@Nn**#2RBh$u;M88?0;g@9x$_Q!;bP;hZ}J^lm7-(*Q+hPf9M3TB`>f
zG5v&<b4rb(Zo;S_JaL)G<QMTH5ke%oYu;vVa$%xTa>X=`9lT5WefU?C#7h4z)yn>v
z6_wFj5*eZgYnBi%h1nc7vlcgQy=fuG(Le6gjDO=WG9h~lgSJ($Ak{1A@r>m=g_Gfj
z7Ot(8_zpS*TtvB?nBYQ@J<_&tAQc(NTm~edIz+a_#d;^&ZGU;2V!2Xg`ibbGKZC|^
zIOvL2kQcpE$uHKt%sh#nPBM9oo6d4(Ld#9fEpKgjG#wB{$UqWpNvfrVQMCb<50Zn@
z^Lrizd)gwsS+Yg&)V2OZwsH&KWn)uyLRc92Yj!4;ExCZVtc@`@(P_fcmF|$H?7BHK
z#z%!;)vuKg7jUYRug}A;3QSU7=Tm6?_I`t#NKt|84*Wt(kfoEx%es5H$93I;ZO<!h
z6R@n;P~p2%ca3az*=n&d@Y8!OFJX6!-(&*uY6&xI>auDo(Azr^o<jO~U$T^F;F{P9
z*PCq!c<!L-qMDe)qG3qX#6*Na_b?{yectua9`)Da;S&CRTYUEjuo^-xl&K9#Zd5bI
zt6`Q@9ek?Zm5o2G6VEDfETy<4;16i#OJ!PWJCg~jiC`?`!u%oXdLN6}#ucb)7d`D?
z+d^)K{n3WXRU2F^%#_05+rew~SNLvLWgZZms(e*S8})bHb@%yJWEgA<zTcXoj8`%7
zDD2gnyKu%*R@UaF=2auLBNYFd8mg@L_qQY~*DBc_y7=fAwUfz*YqsNyHumsqy*-?S
ziZ)-5?@Pc;wvNyqZFfRk%jJdzQHjbDg>uOy5xVyuXYBr*pM#m!L4DHM0|yfbv|@Wz
z%Gg*A$<}Z&xO@eGBnibux||-@?=?ivi_KTC9rs@*NT)1U91eoeR8{I&v)Qsn>nh4p
z&PJFnCC`FOY{$z*9jR3t-)*)yk$ZCfr1a$MM9ThQhL|XlJ2;8;*rN-W&5nRe)inQU
zt<l{&6bDYf#0O;i#qgU1#ofHTY9G@d$h~b@1`}bmQ+gsFrVHNi67S=ey?!;<J!XJg
z*N9N5j`#w5JA+~N$c1RNC_{$}HVx8*qK{nNblh*G)kRpn$=hK(#suhUNVQnpYO`D5
zr5Q6JWVP8S)OcC5WSQ;!^k0GPWs-Y1Q~|0j7ib&g2lubWLQ3NY_UBCrV*b^5fWK(X
za%C2zv1BK*ND=d`1!I&|$!Q^|HKhr|#h*F~r)GJx-&ow%h-rH<opzydbo-|Z*^t_d
z{0591fOHE3EOp;UtWKKB{3k8N-HAR+Kix-Nk{peAg?xe2i=l*Zl`VAmfqUi(d>XqE
zt$7E`S8HFST_YsS#EK|r=P~qR5m6}~D_)!ON=sI@_)-BMb4Lf%pGg%SH4*1jj4Wu4
zse5r{jpHWy`*UXw05)R&cApww&4@OoyvNUlK+OM)J$sDe1E2*&M{RdlEV4|le$N@N
z6R`y?CQ-o0XW4J_M!rEoaAtrRF)87K5))rVc(S+*geeMFsdgI(8KjEDF~uX}gOz%-
z=G1_5B<Pjaky2P&AuGE9ymnY5??mw35rm6`4T;YTzLfZmn2ecEur>00i(j|f%OVHL
z_u|T3xs!*)Sr5a^B((JhK*sMKK%O;2>NNS>)8o%=k83?iM{jX~`QPv{gXzh`)J})4
z-1d|wQtcvk6HvQDUg3dgX6W>2ywRito$ulSU-Er&XXe!8SGfF;Pr$s%%$j?l*s8TJ
zf&XUlvP`>uuP=^I=e+ieK1F;vXW3^wL-50rTXfHUM@b43e4CMKYN%_7{R4IK2J9J6
z*@qAGvK`ixE$X@@F(ajLa6B`iA}3*BLn2%~DX~i*rkQ>ue(5A6ElvNs6zxCw2>(Mu
z_g^%w|I1;J<+J*7ap36=&ibu&3;@7238C@dRsWC1!vE8Zj)94h?LRX*og_<nWVJt|
zu8{hGh9H7f(+4-f`8`nn`3n`g8w2Dj@i76C)qeie2oM_50b}z0cUyLk2g!;8{N#D)
zh`N5{EiE{o4b9ck^K{uUgJ_rR*WmXx-A*M~&76ylrZ<OMzuAs6wl~-b5HQ5z2{3ZI
z^^tUo>8wMy2N9G}oQ=EfB2h?cf}li<L1PJo5D10F2GaO&3S%>rKxnUr<>c?%y}4s*
zNFW-mQ+%$WM3RU^*GKLV>b-?J_n3&-{BgqM^-*rYVT8B{X-V+}bO&9Pr;qK`9y3NP
z>Xivi;o+A`?+=uX`DJTR2}sDhVUM#|0uiWIizrDC$tjA!bLmfdnW$HD0Wdv^514(J
zJ0K732HEupcft6PlVB*QGgg2iCOrjP4XV%$5wNK8<C#+aO&J_SrHytT^7;#oCVDg{
z{2Mn@)OEpZ8Uh{peoPwd2zc;*!H4|y%bGnD1pS%{MaR=3kiiKf33^_n4a9G|!?x~x
zL{z-qmKP^cQ`4H4`c&jc)BLj<&R~TipT2d!*-UPZ=`-TuDZ*W1SHulrz{0!=qG`rR
zG~<F{2z;&l1LAMuB<~XZ@_lL2v{*vC#b!;>1px$2`a#?*u^TqyLpb93F3{O5V%M<I
zE+#q(4e}zVCr`SxhLf~|w{qrlPIKX^?DW01fMSPr&h_jtm5BA2BKXTwhk&fG=1XlX
z5dJZ{;G^b$4o2ox!7GK;^5nio)fh4S1x-~B%WT_YH~vDGs@umaeh01MCa21`ZeYFC
zN9ris&}_{l$=5vU4&Zj&a{ifVl(ZjWr4;3paj^JH3N@A|x?dmGM|BsWF1=@m{o_Ux
zA$$=WN}eTgA%D8o=!T;tpQDv5cDn)_tx|UqIk1c|&GGhp2ExSZRBoH`V>O`7-3NDx
z-3iSsd5^nuC{!EUr|^_o>%a^&elmB~MXBBu5~WzB;(Jj1k=@e0KGm9ZanKm#d(V8a
zCUPS(fShQT#_J9Vg?#mRF()dS9M9uj{lgrsOuk1DVU(C+Na;LSa*$=nR4i=`@APC5
z=ifn3&-8=ieR`6amJ}hApob`@KW-9l#cQkzIwY=8j!T6zzM0ZW_lU6gKrYS_Y4}%f
zaGS2)8xZ^)OO|}g$U9j%V%sxC#Ij6R4Dr5iyet4K)B0I8+Wt;_SI_rVLX@J%7z<I&
zS29LN40hGH0IGahO`7kJgbdx(+r*T~+G=yxSHSUNsF+Va&y?ua=mN)D5NjKD(_{Tp
zneD2|97)faWznFv64{KrPn7Mjn$_LW945-v_pQHeoTLtI0xZL~3O1*v(~(%FkS{H<
zZo=vH5woPSeOO6KZSZ4|8m_P+zq)dH|BPqGjeCbSmW<ukVuP31KQrrFAi!ak)!r3}
zMJC~F2RljBD9KIeL&W^qT213AY`~sQzes-Eb8%O~CMQjQ&UV)JBrk60z*(7cso?o_
zcIW1fK1N1hg6w)n$RI1$%Rx9w;gdXn2o8R(_;^lbB1JyY<+h(vrh3lIR_~yW2^Drc
zvm<Y^8iVs1-H5o^2fk?@<$X?tw7l1z_4?cxnO-}PH>I^Lv)g9+j>B!MmMab#eo<w8
z&N#t|DSNsrRRr@({70d(vL^G!Uh233`fLh_m#Fr>E%s$@`$o1lQ$%Y1I>tX|{2#wB
z`G*#hbRj}Tk&Zaw+j}0#zD!q|C**DB<p-aXnLbi2h<_MK_iQTxF?5$3H?;AaD%fDi
z*=lWh<HW&(yE>xy7l@wlFTj6y7XL?o_5X0v{};Qmr5*qvC|C#>d<ZBt6aYLP5Y2yA
z_&*+b{!b5iMkZ#)|8mHydOIPjpytk*5F%7MWw^+J??gpFg!%v;<2(HE2Rc?tBFq~W
zN+^-oh=Y{aAZ`lYNHHK=gqOsE`7X$%_?BE$g$^d_5L-fF7q8cMZ7#p+X!8l6eH-!q
zska5f=J-6x8sBoYx#@g$^?ltTBO)Y%Oqc$L*{0EAXKG#VAt$kCF`Y<wSr}eFz-~nf
z^FzQ$ACpDS4MQ*_1TBW4Q4*D=B=Yrz^?t~F?!A8C03o1qllFL>3V=FJe6tkIa;RyV
zFY(HdmnJeAaw3cfBQZPRI$+Zn3N%zTw#VEKq0U~nY3f8JD=sRk*{7zhRIsgUj8sxg
zAr`+qzPccGDb&oI<gjWNRkfZmU?Z*9N?5DT0=~suf6q!^K7+APVx2T@#OTnm%8@Z`
zuJ@oyHLa%+H(-Jh#k$<9YN!5bzL4f@mPs8xNpEygU9p*~zjU>l)mybU;p$v3TTZv;
zWMzmqw&ujKtlOlSB_Z2<e3Hmg$<2y(wi++-lx#6N_YJ@NyO9y!=02Fwtkqg8)4+9Y
z#cFEY=;YEoZzT1>r~Zoo*=2$Z9d2a65HCqhq>4}(?qq;mAMBPshVO#x2??uD{DQaw
z5ry-YK{zVNfLKA;i*}RRGS!=hhQ~lwSbxYfem(v-VK&)y@;1)mA?AQ&_+3o3UqvA_
z@G`)T8r|C^p&?8LFnO1{ZAwH%r5K<BWz}Vi|8d9dqB?oRMNQ36_q!9w?p2jV8@JJ>
z63A>Bk@;~s$gP6?`rv~L7;pJw2H-+OYV<rpprf{XVKy(iRav%`sT4t;1yr=TGihe*
z1{eMNlLB39a4&HGv;VA;pp8E1lgU|&Sx|A=7GWYQl+uc1PrhBKJKTr&dtvE(rqd2p
zi01*9ZWXFJSf{DByFE7FM%TB161jRK;)Gf>)`tyE@vi(w7qh?HO*C*9v`}AZEIK!h
z{37qH-~SpJ4xn3I{9`i4QULRu`EmV2$&2VvtllV-;<GJAbP<&pA4NzdUt^d0?O8jo
zco4j-TKjnZfWRZm>IM}IqX~#sK>@gvnXYQP?(pGg6^*aQN#E(!=3JZI8)zbcQNfol
zt}B7aokTD8uvw8RJLl&nDANDiNUYNH4<+kY&_SF-Klc;6$JJ@6K6Mt)i#1u}pBbh6
z!J|^x)7h_J-baDI6`|n)wbD9S3E{y9hZXuV#(?~6<3KXOs>plYFkLbpzylD0Ct@*Y
zL4<xbi6WeT3vtTL^!8*L^otzQ!X3c(<?n3d!B%BN!!m~(D<UT;btZFty%E<s#rK)j
z&YriukR6d>i{om%yLdnUr05Z?J4xQb-ymyCLtzKR8yT{p(gY+5!~;GoKC*1*m3>&(
zq}p6sTo@L_Sp$*AfRpNd2s^E^92z|zMVGHwD#H-qA{(mU#Q0ker+1md-?celzAQki
z)k-gCjq&X3?NF0XA0Eg&zjP4td+T=D#Ci(ZdtGG9pvSForP_dGow|PTUkf;25_;gA
zPte&?8cA|{HFEVxvTJHUOLljIgP+?5BC~lH>jY;{g=^X-mp;UQJ6{{)b`(!;&gQ`g
zm$?|5GXXkhx#_&ltiOhYjLd*7rBCTRMM@@?$bl()*gItS7VW3ic81@#thDnLlKDE<
zcEN9j%HzVa#7V;{<c8V8R1k~v6LxvU3Pb2kQNrj?`G^XcZ2&9kLxc_@Fe*a<8FW{S
zZp3F=HfK0+Rdsjc;B*6IqAjyieF`BOz_m<5%KDk`>XAu)1m@VC+*&Xn61WFg4Xn4_
z?rrEj<&N6wbhF?xWpa|S!dHA`v3LMvM<k-yc_5Wi%`p6np;M4aH>}(jjZMqwv?j9x
zB3XlZX3c3wjX8_vxF#K+c-G$3jFeLMZmG)Vne2iCIA2A6&GOo7^tkx52y<OiPDd`i
zy@=kvayI-uEp%a`;99UdOOM?0mhdUEJ=SJ@Lm7oW=2i(T#H)Jo>H#yxf4ew4ygFBI
zn)a2Q?u<tp&9pPIPmU>i{PTGq3kiTz2Kx=)Y0{(>b)u2Drqg1;+OoRbjrXR~Mn5R|
zN4z8Gr4@<o0M`+CdACfB@PhP`L0Fh8@ssb6|LynzFX8+&Jk$XQ^!q{hIH1{;ij<xP
zm+ud?nw_+dw?nB^J`Zz@Fu@Z1qQE;R_JDDo#1V-$y@ZcMH~9u;s$(&i+LEyh;IejW
zuW&KCcN@6RoiTZypI!Gsp>Qcutihx<{Pq|_U7-^YAmeT3te3@C`VbA=PNT(@XupC5
zY26sc!wC2tywAId&r3DGGOc&J$v;~MaRQUdLFy2EiqU4p9<jmpMX7izW^Obo2l7<~
z;**xlgNn9cBKAh{c{mt|A>9BY{p!K4=&w)|Sjhs}e_*rl0?PRg<Rga8D^`R|Q@VZ}
z_GhyDD)ZLpAvF>_=8bFU>6KkuVr{_=&B&vj((PmWkbo_J&k$$%6N(IMvVO^kD*%#B
zEFGO)T3DWgG7skQ(V~8+A!0$wtqZ_LDsBqeTCy<dYbAPq%Z3+G<?5|@u66gv@7q^R
zY7MXjlMUl=5jQr|sMZospmGUnn~I(w%#<y+X)`{kG}Bv^bKOSotr3|tQMsQ)yXN&Y
zdR*1M+i;q2Y58mlSh}W1h$*uI)^&H!ZalHlMrKr;=IzXH`DTL8?J!rjnFz-ALfVi=
z%4It?mo>Vk0bTAor5?_mUegm{rUD_M%BP4k7|HiB*C$-f{)u1I9|H)b<?w1Adh>yg
zzA8H|SIOjj9etl*^!4cUOH4!TL{KB1_xqIqo)l-vd2&E;3B&M=6PDUfRZ%imA9?Zo
zU?ac_&jG)`%d)cgdVW9j@@ks#<kMR=P<u2mMR>U*3NRnV0@<^v0_T+@)TwxjG507R
z0V9YXsoN-=AFQSQl!<mQl638HYi3K}k2_!SjGV-q(V!jx)+40WdV3dQJUP>6RnVho
z#60Z!b@-7Tq@17q7oR#=MrqmTadtuG79B!cgI^bJmD!wCnb7uvx=M>Xa>Jo?Aw@Jf
z;C1)IPKz$L2F?s(->SfQFUw!z^hAxoO0#ZMXJBR^rm-z6QHxX-wY`<gyEG{vn$<1J
zkuQ>BBdQsCD}KNwAS!eJn=2T@fAqrq4|lu&cc0Ck9>UTe`Lu)mx;_y=0OSQ=<o~@k
z+y7%H|6e+CIynOiB`0e-S$qaowtr@F2S+D-24)tH|3OmDLeKhN7M|ALbifiz@XoD0
zYb9|F*mmxfX8%JFkuqIxG@iuL0u~4!W~H!307whqe*ONzQ(6JY>}L~}H<YacaVDi=
zY?95LoV~1)M<q;fue(nKZTv?T`r3fNS-*g$=%3O(xG9*eCawlz=MPkKkbGH;;SkG(
z#4&U*rsAQ`-aHznU~>d$M3H@4NnRMy#b+M0ax(9PV4+V^J$-?^VhE5-au9uC{=@(R
zw2CZhNP!Hwos`HebYz8MpJ150*(r=NxdywM8E_@eL^X1-SbHOsj-S~jH1l$QBw?7K
z7;^xHXr6<g<hcqWwQ`{n!Jm=0FkZBgd?rBB4?U<fE5az@)#|y1FnS3nC%80q47UTl
zBL3F<dS)FUBg=ULNa1QzV|1%P{#C$I*f<jX2IJl}S2FAi{_?~*e3oziE&t?2$hEK^
z3w*fz8O50UGmb8D8vHw_nC8B|{N&)iupIH1ziUT)S#$k@KzQU)m|}kpb5hUTor2Zu
zpuUciQADEH!Nfx;Yx|Mf{*YZEeH}K-83Bw0s0x<ld$K6V@%Xz9$_~8mBrw!%Ye`My
zHNb*Ei^!5b><#s4Kg@tSASd%sijF-RQ6N0!aD(=GA%K8{fDr;?zVEpK^%FdRR2LMW
zxeeS%DaZI{`+Wu%aqXGvOJhO_xS@xE`R640G0JxllNA+Ca?27n2eL#+(p->pYTgkq
zKw}OSO!lqi(bxw-$v^;_(gz`wfCxbd9P2^kAL6_T0x`G2NlM7e>BQj@jzYcRK$bZj
z<>^sD`}&_CZO0Ip$38KTc*NK(#B?Q<6(K3<2Nq8@8F{X(;6@!m(hyn(kpz2l3UP_G
zsCxuv1PN^~!99^HlKAN$up_Q%$+5)qgXid}2QkCJ7&98o-J{E}0}x0kG5lH6XXFP2
zP9A`$5eI3Ys7ogT@jrraqNrDdno4Ad#>TJdGj{_^qZBAW3{V!LLP&IF7TD)sM%}`v
z!U$C-S}-Io^dAQmO9tzfkThI#Hn7B=*T->*&0mB98r?TS38^>|XWFCWuu4Qb=hq9>
za>5s=b3*$7036>&_q>D5ij;zbK4szJ=O=74z+fKtx9utS`8lkq$&n>p6nQ>b`~<E2
z8%6e8C0p89>A{KsabckDt(&bj3&yC+Ez-!g?7Em+y1YK$)uSC%n)ploe4j|R3|a#w
zkQ%Z-khx&rH6rifHZP(;X&WM!^L1`{_Wk5jyiu(@stzV<OC|+S8bm=On)O)YMCv6n
zO>?YYj2+k+6;Qt<g!6tR6^#yxAc=T*Se9y_*2_8v&R8Z-_(b@TyuzE{(p;$%o*+jS
zfX85x<x-6BfVf3F9kUxuMI_c4Q!F0KL9f80xVkO7Y7O#;_{xHvGwlyNWO%+ylJo2b
z7%#Lbk68jepW>#;{E;h128|nM&;KFvjIF?&OOJiTB#M?J@Nl{Q^dQrXu3S5tqsF(|
zOIPu@Iq&>UjVJrK-xcre`n_-C@{OA;7AeA$sciaedpP-V(7C9s|DCL#M`QZ@<-ppb
zC2e~CFgkCo)6K)<>H0F5o7?T>j$3)kR(R>IuXuib{`A@pi91K;wIR-{+)X0*IO);A
zZ^f|OIKTb+{Psaq6Hr5*O+tnwavC{5{q}Hv^ieVbtJJNr)aomQr0~7bM6anj%Nt{I
zcBjBnfx2wF{EJOCJUePuh$;WWTn9z~TAoD?+XKjzz%<OYlPZ3XaPdYX$&FP&8UL%P
zL3pfm9@Fx+vPYiNJWzl#0%hTurD0AmxfaGM?yxqZl=8wjp0I?balw$YWd}TtlItd>
zT3YCyZDU)|4#<Igd31ko)uT5XpQS=4I?chDc9D9&d~lc*&UX3D?ahUuD_d((G!BcJ
z`*`eHI!o)*?e_w`dMr8LmLRtc-l51GBb+TO_EVW7qxE;m<XohvrML?(1gm>32*G@j
zG>Ei{VCWr5pfftJlPuu0l+XM1hQg`WeU%Nfz&K>^Y5fx=$nWuPUeANStJ3uF97Suj
z@uLFv)IuJgkMjC^Z|iCSHD@y^iAI!mo_W)Y`Stj)vGlao$4geu#lKuFpG1bVCoI<6
zAJjz6nWe_2Qi7$rB(bBT@??{mg(^j}977lPDO<bYKDsNH0pzsfea5v5jpWq0=O{|*
zp4Y)Jbpm@LuAR_ey#)6_K|fw?Y)O-#T`zJ>Lm?$`(|G)Svw<b?N7SPxwTA)Ru6U#E
zIN{F9p_8g}+oA)yA52lu@dU$(gW+%T$;LS@4E?mTdU|tB`Xt9CGsNtPPt;D{GHL#F
z4}ib>Sj6Hc;y@GO&_jLqskshMY!-A8lXOs(8SZ{>kH@d>82s7?Dm><I9NDMhx8|j!
z&}4~<tp9m!b0#Z~vSDah&Czi_g_JL6)5^<n_&k9`?%tJC`cro|`{`*}6>6hg;v`~X
z4{}iUyAx}UcIvOUXS3{YoF@b%BUKa|cnQ({-cizdI85&KO@_$i4KJLSrmVKdOMV2W
zuyg$V<+D}q0A2FV_)~3KHr?~p!M*eL<FL_hoBG!DucC_DqCH2pj+$c9#-{vqF)L*j
zMDcw*Z_e0tdaj5lF0ifxzAK#9@IF0?<FDvo)KF8Vtv!ti%tmRcNyC!Jl*09Ix-n{Q
z-qLvQN?b5ftS5}T^6{a3ifS=rSnU3;8EPpTr$)fK<KB>D(m?s2m8Sbxv-Cd#%7mIk
z4E4~06#BuBRD%f8QqX}-OeCu?Fbg1|Zk52CBwKZ-9NQR7xST<{QMRLz?|Wxf9_|k&
z3>h>ru`}_ODGzSy44vEL5fPLrC<YUPEk{L+%Mc8f1k>MUt4L4wPM1`U*xtF^o?cEI
zf0vG{e(WB+dD5fl_$c$YQvS7#A%o~RPcx_?dq8b7sG++!*H-+ELQQ2mgvT>_$~T_a
z#Da=rX!%&*4v_-HnpiE4FE(jxB1waW>r83xTDG2sKdgMmC#~21*!QR6dp~=5enyKk
zbm00h<;kQXFpvUt81s1W;p_o5a^cpPDpAW_;3|ySv6b47m=Z<%47Lw8S?z&8yp(I~
z=)&HvM_O}m0!Q(ZYFF%03Z}a)XI|w#nV9%c#CC)<Ra2l%j`nIs?WNuenAMEpJn5Ld
z$VlyxtFQ^$pvnJGKk3h;({Dq%l=7TX$r0jiQA|+pGwwceHP0{{1TX%8agK)RH6{m(
zV6>e4Vo=NUFw)Db(xtE&2Ru&Jodo20&?<C+F7Gg_tHIxV&<bm@9Q%?jcZQd@Y<7kR
zu<TY#wLMqVsqho_!CLGH{d<hYc-^91a0%`$B;k_du&&WzUtp7!wrl>RQKi-3725Ad
z(0i-C3j}2XP4hHS=55i}Jbj#?Bg0S7>$c`INE^M~4(pR%v)9zgJb%C7Rqa{Y!mM9x
z=@{>SnXN-oM?A~KYY3^{gWla+zWHajdXQxaj)r<{Hn<pucK*ioQd8?h%m*b7b!BmW
z;$#gst-ZpmqO3&1XEp_OC$l0jj*Pvx=aXw;JVAoQfs@|z;rzs8O(m*4Pa9>kjZ#+1
z`CO&Z@q1Wy5d3>sTUWE$#iDYJQ=@`sc!UV#Ni)W)$~t4+Vmg_JyD@d(TD4xx?`9Z;
z33Sskg%!XlS1*{jrxl&UCY;P%>B{vZiUJU5s>WERG>GQ8!ns7CliiK>@|mzCeylda
zTfJ3v+yP7TpN7V_eqE~y-fy!{N%W=F$@s~;c*P*0)Ke-x<<k4let|Zf=b}_1?KIA}
z%4ejjxNyqanz4r1Ws7kx;*N^dE7~pE%`5clOPvo^WViw!au%hKZ&Uq7^2AlY@)NwK
zrB;2Bq4LE_-Zgr1KJz4fsF|1QEZSq!w0U<A{ok#nU9U1eFN6V10_{HVN3$cxfqTyD
z#PLxlc-<Q$yd_u6+*di?-qgKkvyo-co7p^nJ_D1qmb<M<C^}3kW!{cT?Ubs9_nuYw
zMp|2zHOAZLsvjC_7K?R@oA5Q2!p{2W7Va_HWm`3Ps}Vr3imJv45IHO5m1v9(|GNfb
zOad57GvqDmRyp2A9ugX$D!&8DKBTgX^(SPt#HSAEg6NX#b3t&Aa_FlB+ciDR*w&7Y
zgR*Js&$dhNlvAu)n&WG`98R4ddHD5a;NzHTxtsb{1JF?G!F@b9u?rty5ynR$Cix#e
zAw<qRv$5-s(e>^w+DwN>dAA%d8^@W1=bCAcU6bw&6YoW~^~Kf^6e+G$hU?1?Jwv(8
z2Ka_;@Xc*-EG=kP8LoL7GR<qFjflCfEAx{tnTuVcpXsdp@!v;+S)j7Lf*d)%Y1&Ko
zcwNLJ*TPx`2^~XBEbm0nn$y4&SP>t8PB|=2wU?UV<MkPyq4k%h#%12ER&URg?h0Pp
zJ9E$>YWnF|M)K(+Geg`K*1Y@W8o<9cn0vymo{qaOm1s!7tH<bQH&kP`9HQ3O7H{12
z>7F7!1Ak^*UZ!3N3)x^Fw#VH5ec933-D~UV%Fe>r9TBCo--!rOFbmXZEO2F0d%MnC
z_Uy$IyB!DaIfncdw|g7B)}Kj!TTUE7v$GcHeu%~8aP<$Cd!xugM}b*}g-f^H+GMNV
zT~#|mWK)&zO?#cJdlTrzS}7KJZuP#^GX?)=6U4C7tEYq1ZgP6a<-KoWTwhb9{liIP
zHpZv-!`jRBdH?1008dPk+y?UXmz{U!0%3YqH1gx@icVV$K~>cJ4(5?e@C*D0&vf$d
z_Htq=#S?rA7aQu}(r@ZwWrqh>t**D`Ruq{ZQMZt!=e35dyXQN=SX#B7EIlt4jVRmV
zWtT2g{32X4jyhSWD7<u-(BWb-3NV#D{Z^5x*>wmjTosB?sZzTmu<`a>VMdBj?L#t#
z@GM^TTmGy;fKg!pY@mc|0hU>92m?%%q7Vg42)S>SfgP?AEX6ZF)U-8!8^t!j4Z378
z*i0EgX)i95K)>@~zNjT)+XUy5|Fc`GJTFZ2z?7uDEiv8VC{b2JNjNZgN_|)}k40mq
z+lJZCSTW+jwwUt1@`?$LJU_~i6IA2Zv;Uq?YafY95Lf0C63{sV=c3KPQj}r<dQnES
z-Gv-ddlvYd5PF?Gh$a*`w6r|BO3k0}Xmw}0eu1qyJ}7mKt+PXh>_;&vAl)2(WU;Xn
z#-l={KoX!7KBGCz*FF$Tn7BXNpdzFr{sP2<;92tH(r@^1DJ4?L9wVO%B?jkg(kcup
zX1qdViX^EbBFRldRnnTEvyYfh4R8a&s_(+?kbFhR(QzYiZqTB07I83s4<LDa3gS6?
z18lHBX2Es`K$j!Nm*U*=RsKb+DP&#1x1xV6A=MzJk(uaYB5M-Ra1DccNgdko0}w@l
zK|$!=!~_2ggE8|Hvd}{rH0?o(rmAuSj)bWSaQyAV;Dt+e$r;Swi<5_|rV#a4f+P+O
z7vhDKJuWE2&UwNm18Z}O7N9=jw`|157e&MFHTO#cdYob=jBDO++PiIgyI(2M#75*T
z(nL*x=4-l=uo(r6<%9!uv#Y}h1fw?C_t*;^!$SxrG<Ttl_F{(S43{k>2q*1SF;dfj
zC(5-6u`3|*vPdHIf5ESm)FjT@(}n{h#X;l=kn*-RKo^UlN!gDA@9kRrp`WnUF43ge
zuvTeuDI_e!gOfcOl!OA-4;*a)v30cxZcfM6y-uj34A(Hw-t8yHJt{oESdxUepAcKb
z)^VI7mq?yI{iN;J&;CMd)PXo`qHuVEjbf${o=nR|Y((ZJZw2EqY!Ba`GeKtbqFTd@
zbI<@I<sno}V|IfPvZ?bII1u9<iT-)Mit=*t8L>nSdj=cV_4&Cuj;;OuIylapH=-RL
zKYu!jP}%)r&|PXhO@v=e`p}L&ja<hQldCPcy*#`?adnI$TNOK16IUOJ<crBazPTys
zG5#TBr`g=Hq{!^Lx=IxURrpTu*>83=v}^2TgJ^9pWV_u@-e_+5Zdh2{ty%P$9{6|}
z(T}-A9>$A$e>#*73aFOpcW-1Uh-lg(_yQ+w+{G74lRkZBGkB*9i}sbvz%9w1%!Z(4
zgRy*Sl7R9zkH)5Px~AJ*>-oUA&7v0Id++&hap33<AgWmDhFM*P0Fz#1rakq#{H|QM
zWy#MZ`}dQ+KlrB@J2LR#%IN!Jcd$5Z_JXc`zo{LrJAi`bS#DsENy(`|@}0F*y366J
zH2mB!bfYgp=VM(G1Y06(51{mMgg24WHi_!XuC?>^d~iD}_xlCZ{vEXbZuQ{M^!65(
zWnN-rIV1c{lEqLE%jCA#!X79VODP*i1X&YY@kI3`q+W<*=|pGA8)z8G<*v4#!F2*~
zbG7O~9{4bmCIb~bNa6HzJ&9(!(+rWV;0<%LW_|NUw;yxu%bY$J!<LqF;x-6_bUy;u
z>ieYxbu_5}z8N<b2|k>|OTPJJpZNp{K36LCmEWackji!s*!6zidCK<%VEI>;Y*EZx
z_BcAha)+GYb{I=qWu;3&(wV1J-E<m@fYpr%(c#S%>Ft#egj6RRaP&4`@~Tu4-ovtj
zoM1y)K@dkU+2%e}arlb2IDkEilCR6-=VkL$>L6saFapUVlTtV7=i>(fc?nDfTL|Wt
zS>y?a;AB!UgZFJin;}!1gZT9t(bw*S%C(&k{&eDe_iwjbr_oL1aN$3I(cPwUyY9>7
zX0NyMLand&W&~y1MKondILz;n)Q2H~{p>%WR65ynXbLXKrq%kGZ!$OW?{$g~y~Y-;
zM#}xQFW|O7sm=c$6f!a}{J)|?bwy(nItgJdZgys4Hhm5Q1~z&Ib~aWPLt`U-Lw0=*
zCS!UNBV!{rdfxy4FLdY>ZEc;P{?`(UbP_fuw)p?6aY>O*$=t&jpOJx)PECsep9!B)
z=YI+!8UM@BAIp-2m5e@m=>6+AjCJ!8x02!Wf}-Q}9~xSf#cFp|XBifyrVT}lGS6jY
zCYpKzbLutqx`sNve$jyiRb)gE=b6M;8z1)vK3yA!#77*&_!AQ{Br?-_O!x!iuAqXN
zBJrS}5H<OEeQ2xj>N=K@eb}y;*87qO;!WF;LjvhO*0iu|dI=H+gVW6lkQ^=7KBMF(
z)0QT+@o&>@46X_gCF}DdMf8T75FO=zTuLXEaPH@y+9jW+W89s7kFHwH=Y<@X(ZM$e
zEak%FmnQI(vK5v8Yqu{6_piEUe(Mr<e{%j+^{3muA2hlu-k<xjTRkNH`0YtePgmuC
z&VAWFb(Q?5oR|ITq4wrCjTlxJ2=E;Vk?C;`*`Idfr%6oHWy1zz1L^cf0drb|3#IrD
zFWtc^o@ABwAZJeVVMG4#B)h%?t~(^c6V=uo+`5C${gFmYzjT2~-Vxs&JjWlz%xQO&
z>Hb{E*j6CZcbqAvPuqYyJX!DA1C~2ot~+?ulO(tcjr|Utv|!!-xb9B-?n3r=M`Cw~
zcNg-{JE&a2;Qdh|>_Mi@(qFnS?y~0odis0smG8%^@6Yeux8MHfzmuE3FA4klUix+M
zD<Q2l>#P1A{k3P;mhfw(ub4JQS31U)UHCQa`ujE6S6ZdkykDg)@VfieA<Jw0d-~>f
zuFyWWrS3)Pn|rQ%^&=+zmtEc2^@siNrW`CwC5lTDi%KerQq#DAW89`@mRzc;uKsRZ
E0CFexxc~qF

literal 0
HcmV?d00001

diff --git a/src/functions-reference/functions_index.qmd b/src/functions-reference/functions_index.qmd
index 6862dc8e5..cc9e2b6b5 100644
--- a/src/functions-reference/functions_index.qmd
+++ b/src/functions-reference/functions_index.qmd
@@ -4164,6 +4164,11 @@ pagetitle: Alphabetical Index
  - <div class='index-container'>[`(matrix A) : real`](matrix_operations.qmd#index-entry-932fc0bd0a556886a853004f8ad6bc3432760b62) <span class='detail'>(matrix_operations.html)</span></div>
 
 
+<a id='trace_dot' href='#trace_dot' class='anchored unlink'>**trace_dot**:</a>
+
+ - <div class='index-container'>[`(matrix A, matrix B) : real`](matrix_operations.qmd#index-entry-b067f1dc33e4e60522d490bc635d4305d0e9e7fb) <span class='detail'>(matrix_operations.html)</span></div>
+
+
 <a id='trace_gen_quad_form' href='#trace_gen_quad_form' class='anchored unlink'>**trace_gen_quad_form**:</a>
 
  - <div class='index-container'>[`(matrix D ,matrix A, matrix B) : real`](matrix_operations.qmd#index-entry-725c7c07b916a247f645ce441212b4a82778b096) <span class='detail'>(matrix_operations.html)</span></div>
diff --git a/src/index.qmd b/src/index.qmd
index 94832b434..50aa0f62e 100644
--- a/src/index.qmd
+++ b/src/index.qmd
@@ -28,6 +28,7 @@ can be found in the table below:
 
 | Version | Stan Reference Manual                                                                                                  | Stan Users Guide                                                                                                       | CmdStan Guide                                                                                                    | Stan Functions Reference                                                                                                     |
 |---------|------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------------|
+| 2.39    | [html](https://mc-stan.org/docs/2_39/reference-manual/) [pdf](https://mc-stan.org/docs/2_39/reference-manual-2_39.pdf) | [html](https://mc-stan.org/docs/2_39/stan-users-guide/) [pdf](https://mc-stan.org/docs/2_39/stan-users-guide-2_39.pdf) | [html](https://mc-stan.org/docs/2_39/cmdstan-guide/) [pdf](https://mc-stan.org/docs/2_39/cmdstan-guide-2_39.pdf) | [html](https://mc-stan.org/docs/2_39/functions-reference/) [pdf](https://mc-stan.org/docs/2_39/functions-reference-2_39.pdf) |
 | 2.38    | [html](https://mc-stan.org/docs/2_38/reference-manual/) [pdf](https://mc-stan.org/docs/2_38/reference-manual-2_38.pdf) | [html](https://mc-stan.org/docs/2_38/stan-users-guide/) [pdf](https://mc-stan.org/docs/2_38/stan-users-guide-2_38.pdf) | [html](https://mc-stan.org/docs/2_38/cmdstan-guide/) [pdf](https://mc-stan.org/docs/2_38/cmdstan-guide-2_38.pdf) | [html](https://mc-stan.org/docs/2_38/functions-reference/) [pdf](https://mc-stan.org/docs/2_38/functions-reference-2_38.pdf) |
 | 2.37    | [html](https://mc-stan.org/docs/2_37/reference-manual/) [pdf](https://mc-stan.org/docs/2_37/reference-manual-2_37.pdf) | [html](https://mc-stan.org/docs/2_37/stan-users-guide/) [pdf](https://mc-stan.org/docs/2_37/stan-users-guide-2_37.pdf) | [html](https://mc-stan.org/docs/2_37/cmdstan-guide/) [pdf](https://mc-stan.org/docs/2_37/cmdstan-guide-2_37.pdf) | [html](https://mc-stan.org/docs/2_37/functions-reference/) [pdf](https://mc-stan.org/docs/2_37/functions-reference-2_37.pdf) |
 | 2.36    | [html](https://mc-stan.org/docs/2_36/reference-manual/) [pdf](https://mc-stan.org/docs/2_36/reference-manual-2_36.pdf) | [html](https://mc-stan.org/docs/2_36/stan-users-guide/) [pdf](https://mc-stan.org/docs/2_36/stan-users-guide-2_36.pdf) | [html](https://mc-stan.org/docs/2_36/cmdstan-guide/) [pdf](https://mc-stan.org/docs/2_36/cmdstan-guide-2_36.pdf) | [html](https://mc-stan.org/docs/2_36/functions-reference/) [pdf](https://mc-stan.org/docs/2_36/functions-reference-2_36.pdf) |